diff --git a/.gitattributes b/.gitattributes index 36b7b67156eef078087473a4048f6a8a66623e1f..df6dd94690fcad795569e89b682d36f81da4d755 100644 --- a/.gitattributes +++ b/.gitattributes @@ -67,3 +67,21 @@ triton/gpu0/A7X6T6PKJCRMT3VM6R2EQWTJL4JVXK2EY6S4UGCLO42GUJM54JWQ/chunk_bwd_kerne triton/gpu0/AEQGKJYL6DLKXOG3TU4HKZV7AVJHMVX2MVTUSYTRLB5UQBUBVLEA/l2norm_bwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text triton/gpu0/AKSFD27IP2GUNEHHZEPXU4XCHBVIYLE7I6AGS43DEHS44EMEKUNA/recompute_w_u_fwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text triton/gpu0/AOEYMJ4TTFDXRPXMBW52VLMJ7OBEVHHGC7Z5TZAFKQFJAGELKGPA/l2norm_bwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text +triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text +triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text +triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.cubin filter=lfs diff=lfs merge=lfs -text +triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.cubin filter=lfs diff=lfs merge=lfs -text +triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text +triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text +triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text +triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text +triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text +triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text +triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text +triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.cubin filter=lfs diff=lfs merge=lfs -text +triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.cubin filter=lfs diff=lfs merge=lfs -text +triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text +triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text +triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text +triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.cubin filter=lfs diff=lfs merge=lfs -text +triton/gpu0/COEH6RZWJ4F4NLI6BWD6G2VBWJCNYUFXAFKSMEBEK2VRZ3DVOIQA/chunk_fwd_kernel_o.cubin filter=lfs diff=lfs merge=lfs -text diff --git a/triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.cubin b/triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..f0c87b49c11949a0803e6250aba0a6d0f32f274e Binary files /dev/null and b/triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.cubin differ diff --git a/triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.json b/triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..70d6ec95026181c86cb1b8a4eecdb63ffcad8e05 --- /dev/null +++ b/triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"hash": "03d8bc7170f11d56ce55d76a8ebd7ff419244e019db745f811b3f1d8a9e30b37", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 1, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 32, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_fwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.llir b/triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..4bd998a916deea1dcc376f2130c5dd7440fb9787 --- /dev/null +++ b/triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.llir @@ -0,0 +1,633 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 +@.str.2 = private unnamed_addr constant [17 x i8] c"__CUDA_PREC_SQRT\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_fwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, float %3, i32 %4, ptr addrspace(1) readnone captures(none) %5, ptr addrspace(1) readnone captures(none) %6) local_unnamed_addr #0 !dbg !5 { + %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %9 = zext nneg i32 %8 to i64, !dbg !14 + %10 = shl nuw nsw i64 %9, 3, !dbg !15 + %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !16 + %12 = and i32 %11, 16, !dbg !16 + %.lobit = lshr exact i32 %12, 4, !dbg !16 + %13 = or disjoint i32 %.lobit, 2, !dbg !16 + %14 = or disjoint i32 %.lobit, 4, !dbg !16 + %15 = or disjoint i32 %.lobit, 6, !dbg !16 + %16 = zext nneg i32 %.lobit to i64, !dbg !17 + %17 = zext nneg i32 %13 to i64, !dbg !17 + %18 = zext nneg i32 %14 to i64, !dbg !17 + %19 = zext nneg i32 %15 to i64, !dbg !17 + %20 = or disjoint i64 %10, %16, !dbg !17 + %21 = or disjoint i64 %10, %17, !dbg !17 + %22 = or disjoint i64 %10, %18, !dbg !17 + %23 = or disjoint i64 %10, %19, !dbg !17 + %24 = sext i32 %4 to i64, !dbg !18 + %25 = icmp slt i64 %20, %24, !dbg !18 + %26 = icmp slt i64 %21, %24, !dbg !18 + %27 = icmp slt i64 %22, %24, !dbg !18 + %28 = icmp slt i64 %23, %24, !dbg !18 + %29 = shl nuw nsw i32 %11, 3, !dbg !19 + %30 = and i32 %29, 120, !dbg !19 + %31 = shl nuw nsw i64 %20, 7, !dbg !20 + %32 = shl nuw nsw i64 %21, 7, !dbg !20 + %33 = shl nuw nsw i64 %22, 7, !dbg !20 + %34 = shl nuw nsw i64 %23, 7, !dbg !20 + %35 = getelementptr bfloat, ptr addrspace(1) %0, i64 %31, !dbg !21 + %36 = getelementptr bfloat, ptr addrspace(1) %0, i64 %32, !dbg !21 + %37 = getelementptr bfloat, ptr addrspace(1) %0, i64 %33, !dbg !21 + %38 = getelementptr bfloat, ptr addrspace(1) %0, i64 %34, !dbg !21 + %39 = zext nneg i32 %30 to i64, !dbg !22 + %40 = getelementptr bfloat, ptr addrspace(1) %35, i64 %39, !dbg !22 + %41 = getelementptr bfloat, ptr addrspace(1) %36, i64 %39, !dbg !22 + %42 = getelementptr bfloat, ptr addrspace(1) %37, i64 %39, !dbg !22 + %43 = getelementptr bfloat, ptr addrspace(1) %38, i64 %39, !dbg !22 + %44 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %40, i1 %25) #6, !dbg !23 + %45 = extractvalue { i32, i32, i32, i32 } %44, 0, !dbg !23 + %46 = bitcast i32 %45 to <2 x bfloat>, !dbg !23 + %47 = extractvalue { i32, i32, i32, i32 } %44, 1, !dbg !23 + %48 = bitcast i32 %47 to <2 x bfloat>, !dbg !23 + %49 = extractvalue { i32, i32, i32, i32 } %44, 2, !dbg !23 + %50 = bitcast i32 %49 to <2 x bfloat>, !dbg !23 + %51 = extractvalue { i32, i32, i32, i32 } %44, 3, !dbg !23 + %52 = bitcast i32 %51 to <2 x bfloat>, !dbg !23 + %53 = extractelement <2 x bfloat> %46, i64 0, !dbg !23 + %54 = extractelement <2 x bfloat> %46, i64 1, !dbg !23 + %55 = extractelement <2 x bfloat> %48, i64 0, !dbg !23 + %56 = extractelement <2 x bfloat> %48, i64 1, !dbg !23 + %57 = extractelement <2 x bfloat> %50, i64 0, !dbg !23 + %58 = extractelement <2 x bfloat> %50, i64 1, !dbg !23 + %59 = extractelement <2 x bfloat> %52, i64 0, !dbg !23 + %60 = extractelement <2 x bfloat> %52, i64 1, !dbg !23 + %61 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %41, i1 %26) #6, !dbg !23 + %62 = extractvalue { i32, i32, i32, i32 } %61, 0, !dbg !23 + %63 = bitcast i32 %62 to <2 x bfloat>, !dbg !23 + %64 = extractvalue { i32, i32, i32, i32 } %61, 1, !dbg !23 + %65 = bitcast i32 %64 to <2 x bfloat>, !dbg !23 + %66 = extractvalue { i32, i32, i32, i32 } %61, 2, !dbg !23 + %67 = bitcast i32 %66 to <2 x bfloat>, !dbg !23 + %68 = extractvalue { i32, i32, i32, i32 } %61, 3, !dbg !23 + %69 = bitcast i32 %68 to <2 x bfloat>, !dbg !23 + %70 = extractelement <2 x bfloat> %63, i64 0, !dbg !23 + %71 = extractelement <2 x bfloat> %63, i64 1, !dbg !23 + %72 = extractelement <2 x bfloat> %65, i64 0, !dbg !23 + %73 = extractelement <2 x bfloat> %65, i64 1, !dbg !23 + %74 = extractelement <2 x bfloat> %67, i64 0, !dbg !23 + %75 = extractelement <2 x bfloat> %67, i64 1, !dbg !23 + %76 = extractelement <2 x bfloat> %69, i64 0, !dbg !23 + %77 = extractelement <2 x bfloat> %69, i64 1, !dbg !23 + %78 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %42, i1 %27) #6, !dbg !23 + %79 = extractvalue { i32, i32, i32, i32 } %78, 0, !dbg !23 + %80 = bitcast i32 %79 to <2 x bfloat>, !dbg !23 + %81 = extractvalue { i32, i32, i32, i32 } %78, 1, !dbg !23 + %82 = bitcast i32 %81 to <2 x bfloat>, !dbg !23 + %83 = extractvalue { i32, i32, i32, i32 } %78, 2, !dbg !23 + %84 = bitcast i32 %83 to <2 x bfloat>, !dbg !23 + %85 = extractvalue { i32, i32, i32, i32 } %78, 3, !dbg !23 + %86 = bitcast i32 %85 to <2 x bfloat>, !dbg !23 + %87 = extractelement <2 x bfloat> %80, i64 0, !dbg !23 + %88 = extractelement <2 x bfloat> %80, i64 1, !dbg !23 + %89 = extractelement <2 x bfloat> %82, i64 0, !dbg !23 + %90 = extractelement <2 x bfloat> %82, i64 1, !dbg !23 + %91 = extractelement <2 x bfloat> %84, i64 0, !dbg !23 + %92 = extractelement <2 x bfloat> %84, i64 1, !dbg !23 + %93 = extractelement <2 x bfloat> %86, i64 0, !dbg !23 + %94 = extractelement <2 x bfloat> %86, i64 1, !dbg !23 + %95 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %43, i1 %28) #6, !dbg !23 + %96 = extractvalue { i32, i32, i32, i32 } %95, 0, !dbg !23 + %97 = bitcast i32 %96 to <2 x bfloat>, !dbg !23 + %98 = extractvalue { i32, i32, i32, i32 } %95, 1, !dbg !23 + %99 = bitcast i32 %98 to <2 x bfloat>, !dbg !23 + %100 = extractvalue { i32, i32, i32, i32 } %95, 2, !dbg !23 + %101 = bitcast i32 %100 to <2 x bfloat>, !dbg !23 + %102 = extractvalue { i32, i32, i32, i32 } %95, 3, !dbg !23 + %103 = bitcast i32 %102 to <2 x bfloat>, !dbg !23 + %104 = extractelement <2 x bfloat> %97, i64 0, !dbg !23 + %105 = extractelement <2 x bfloat> %97, i64 1, !dbg !23 + %106 = extractelement <2 x bfloat> %99, i64 0, !dbg !23 + %107 = extractelement <2 x bfloat> %99, i64 1, !dbg !23 + %108 = extractelement <2 x bfloat> %101, i64 0, !dbg !23 + %109 = extractelement <2 x bfloat> %101, i64 1, !dbg !23 + %110 = extractelement <2 x bfloat> %103, i64 0, !dbg !23 + %111 = extractelement <2 x bfloat> %103, i64 1, !dbg !23 + %112 = fpext bfloat %53 to float, !dbg !24 + %113 = fpext bfloat %54 to float, !dbg !24 + %114 = fpext bfloat %55 to float, !dbg !24 + %115 = fpext bfloat %56 to float, !dbg !24 + %116 = fpext bfloat %57 to float, !dbg !24 + %117 = fpext bfloat %58 to float, !dbg !24 + %118 = fpext bfloat %59 to float, !dbg !24 + %119 = fpext bfloat %60 to float, !dbg !24 + %120 = fpext bfloat %70 to float, !dbg !24 + %121 = fpext bfloat %71 to float, !dbg !24 + %122 = fpext bfloat %72 to float, !dbg !24 + %123 = fpext bfloat %73 to float, !dbg !24 + %124 = fpext bfloat %74 to float, !dbg !24 + %125 = fpext bfloat %75 to float, !dbg !24 + %126 = fpext bfloat %76 to float, !dbg !24 + %127 = fpext bfloat %77 to float, !dbg !24 + %128 = fpext bfloat %87 to float, !dbg !24 + %129 = fpext bfloat %88 to float, !dbg !24 + %130 = fpext bfloat %89 to float, !dbg !24 + %131 = fpext bfloat %90 to float, !dbg !24 + %132 = fpext bfloat %91 to float, !dbg !24 + %133 = fpext bfloat %92 to float, !dbg !24 + %134 = fpext bfloat %93 to float, !dbg !24 + %135 = fpext bfloat %94 to float, !dbg !24 + %136 = fpext bfloat %104 to float, !dbg !24 + %137 = fpext bfloat %105 to float, !dbg !24 + %138 = fpext bfloat %106 to float, !dbg !24 + %139 = fpext bfloat %107 to float, !dbg !24 + %140 = fpext bfloat %108 to float, !dbg !24 + %141 = fpext bfloat %109 to float, !dbg !24 + %142 = fpext bfloat %110 to float, !dbg !24 + %143 = fpext bfloat %111 to float, !dbg !24 + %144 = fmul float %112, %112, !dbg !25 + %145 = fmul float %113, %113, !dbg !25 + %146 = fmul float %114, %114, !dbg !25 + %147 = fmul float %115, %115, !dbg !25 + %148 = fmul float %116, %116, !dbg !25 + %149 = fmul float %117, %117, !dbg !25 + %150 = fmul float %118, %118, !dbg !25 + %151 = fmul float %119, %119, !dbg !25 + %152 = fmul float %120, %120, !dbg !25 + %153 = fmul float %121, %121, !dbg !25 + %154 = fmul float %122, %122, !dbg !25 + %155 = fmul float %123, %123, !dbg !25 + %156 = fmul float %124, %124, !dbg !25 + %157 = fmul float %125, %125, !dbg !25 + %158 = fmul float %126, %126, !dbg !25 + %159 = fmul float %127, %127, !dbg !25 + %160 = fmul float %128, %128, !dbg !25 + %161 = fmul float %129, %129, !dbg !25 + %162 = fmul float %130, %130, !dbg !25 + %163 = fmul float %131, %131, !dbg !25 + %164 = fmul float %132, %132, !dbg !25 + %165 = fmul float %133, %133, !dbg !25 + %166 = fmul float %134, %134, !dbg !25 + %167 = fmul float %135, %135, !dbg !25 + %168 = fmul float %136, %136, !dbg !25 + %169 = fmul float %137, %137, !dbg !25 + %170 = fmul float %138, %138, !dbg !25 + %171 = fmul float %139, %139, !dbg !25 + %172 = fmul float %140, %140, !dbg !25 + %173 = fmul float %141, %141, !dbg !25 + %174 = fmul float %142, %142, !dbg !25 + %175 = fmul float %143, %143, !dbg !25 + %176 = fadd float %144, %145, !dbg !26 + %177 = fadd float %146, %176, !dbg !26 + %178 = fadd float %147, %177, !dbg !26 + %179 = fadd float %148, %178, !dbg !26 + %180 = fadd float %149, %179, !dbg !26 + %181 = fadd float %150, %180, !dbg !26 + %182 = fadd float %151, %181, !dbg !26 + %183 = fadd float %152, %153, !dbg !26 + %184 = fadd float %154, %183, !dbg !26 + %185 = fadd float %155, %184, !dbg !26 + %186 = fadd float %156, %185, !dbg !26 + %187 = fadd float %157, %186, !dbg !26 + %188 = fadd float %158, %187, !dbg !26 + %189 = fadd float %159, %188, !dbg !26 + %190 = fadd float %160, %161, !dbg !26 + %191 = fadd float %162, %190, !dbg !26 + %192 = fadd float %163, %191, !dbg !26 + %193 = fadd float %164, %192, !dbg !26 + %194 = fadd float %165, %193, !dbg !26 + %195 = fadd float %166, %194, !dbg !26 + %196 = fadd float %167, %195, !dbg !26 + %197 = fadd float %168, %169, !dbg !26 + %198 = fadd float %170, %197, !dbg !26 + %199 = fadd float %171, %198, !dbg !26 + %200 = fadd float %172, %199, !dbg !26 + %201 = fadd float %173, %200, !dbg !26 + %202 = fadd float %174, %201, !dbg !26 + %203 = fadd float %175, %202, !dbg !26 + %204 = bitcast float %182 to i32, !dbg !29 + %205 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %204, i32 8, i32 31), !dbg !29 + %206 = bitcast i32 %205 to float, !dbg !29 + %207 = fadd float %182, %206, !dbg !26 + %208 = bitcast float %207 to i32, !dbg !29 + %209 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %208, i32 4, i32 31), !dbg !29 + %210 = bitcast i32 %209 to float, !dbg !29 + %211 = fadd float %207, %210, !dbg !26 + %212 = bitcast float %211 to i32, !dbg !29 + %213 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %212, i32 2, i32 31), !dbg !29 + %214 = bitcast i32 %213 to float, !dbg !29 + %215 = fadd float %211, %214, !dbg !26 + %216 = bitcast float %215 to i32, !dbg !29 + %217 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %216, i32 1, i32 31), !dbg !29 + %218 = bitcast i32 %217 to float, !dbg !29 + %219 = fadd float %215, %218, !dbg !26 + %220 = bitcast float %189 to i32, !dbg !29 + %221 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %220, i32 8, i32 31), !dbg !29 + %222 = bitcast i32 %221 to float, !dbg !29 + %223 = fadd float %189, %222, !dbg !26 + %224 = bitcast float %223 to i32, !dbg !29 + %225 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %224, i32 4, i32 31), !dbg !29 + %226 = bitcast i32 %225 to float, !dbg !29 + %227 = fadd float %223, %226, !dbg !26 + %228 = bitcast float %227 to i32, !dbg !29 + %229 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %228, i32 2, i32 31), !dbg !29 + %230 = bitcast i32 %229 to float, !dbg !29 + %231 = fadd float %227, %230, !dbg !26 + %232 = bitcast float %231 to i32, !dbg !29 + %233 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %232, i32 1, i32 31), !dbg !29 + %234 = bitcast i32 %233 to float, !dbg !29 + %235 = fadd float %231, %234, !dbg !26 + %236 = bitcast float %196 to i32, !dbg !29 + %237 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %236, i32 8, i32 31), !dbg !29 + %238 = bitcast i32 %237 to float, !dbg !29 + %239 = fadd float %196, %238, !dbg !26 + %240 = bitcast float %239 to i32, !dbg !29 + %241 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %240, i32 4, i32 31), !dbg !29 + %242 = bitcast i32 %241 to float, !dbg !29 + %243 = fadd float %239, %242, !dbg !26 + %244 = bitcast float %243 to i32, !dbg !29 + %245 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %244, i32 2, i32 31), !dbg !29 + %246 = bitcast i32 %245 to float, !dbg !29 + %247 = fadd float %243, %246, !dbg !26 + %248 = bitcast float %247 to i32, !dbg !29 + %249 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %248, i32 1, i32 31), !dbg !29 + %250 = bitcast i32 %249 to float, !dbg !29 + %251 = fadd float %247, %250, !dbg !26 + %252 = bitcast float %203 to i32, !dbg !29 + %253 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %252, i32 8, i32 31), !dbg !29 + %254 = bitcast i32 %253 to float, !dbg !29 + %255 = fadd float %203, %254, !dbg !26 + %256 = bitcast float %255 to i32, !dbg !29 + %257 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %256, i32 4, i32 31), !dbg !29 + %258 = bitcast i32 %257 to float, !dbg !29 + %259 = fadd float %255, %258, !dbg !26 + %260 = bitcast float %259 to i32, !dbg !29 + %261 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %260, i32 2, i32 31), !dbg !29 + %262 = bitcast i32 %261 to float, !dbg !29 + %263 = fadd float %259, %262, !dbg !26 + %264 = bitcast float %263 to i32, !dbg !29 + %265 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %264, i32 1, i32 31), !dbg !29 + %266 = bitcast i32 %265 to float, !dbg !29 + %267 = fadd float %263, %266, !dbg !26 + %268 = fadd float %3, %219, !dbg !32 + %269 = fadd float %3, %235, !dbg !32 + %270 = fadd float %3, %251, !dbg !32 + %271 = fadd float %3, %267, !dbg !32 + %272 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i = icmp eq i32 %272, 0, !dbg !33 + %273 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i = icmp eq i32 %273, 0, !dbg !33 + br i1 %.not.i, label %279, label %274, !dbg !33 + +274: ; preds = %7 + br i1 %.not1.i, label %277, label %275, !dbg !33 + +275: ; preds = %274 + %276 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %268) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +277: ; preds = %274 + %278 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %268) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +279: ; preds = %7 + br i1 %.not1.i, label %282, label %280, !dbg !33 + +280: ; preds = %279 + %281 = tail call float @llvm.nvvm.sqrt.rn.f(float %268) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +282: ; preds = %279 + %283 = tail call float @llvm.nvvm.sqrt.approx.f(float %268) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +__nv_sqrtf.exit: ; preds = %275, %277, %280, %282 + %.0.i = phi float [ %276, %275 ], [ %278, %277 ], [ %281, %280 ], [ %283, %282 ], !dbg !33 + %284 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i1 = icmp eq i32 %284, 0, !dbg !33 + %285 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i4 = icmp eq i32 %285, 0, !dbg !33 + br i1 %.not.i1, label %291, label %286, !dbg !33 + +286: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %289, label %287, !dbg !33 + +287: ; preds = %286 + %288 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %269) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +289: ; preds = %286 + %290 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %269) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +291: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %294, label %292, !dbg !33 + +292: ; preds = %291 + %293 = tail call float @llvm.nvvm.sqrt.rn.f(float %269) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +294: ; preds = %291 + %295 = tail call float @llvm.nvvm.sqrt.approx.f(float %269) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +__nv_sqrtf.exit5: ; preds = %287, %289, %292, %294 + %.0.i3 = phi float [ %288, %287 ], [ %290, %289 ], [ %293, %292 ], [ %295, %294 ], !dbg !33 + %296 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i6 = icmp eq i32 %296, 0, !dbg !33 + %297 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i9 = icmp eq i32 %297, 0, !dbg !33 + br i1 %.not.i6, label %303, label %298, !dbg !33 + +298: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %301, label %299, !dbg !33 + +299: ; preds = %298 + %300 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %270) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +301: ; preds = %298 + %302 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %270) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +303: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %306, label %304, !dbg !33 + +304: ; preds = %303 + %305 = tail call float @llvm.nvvm.sqrt.rn.f(float %270) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +306: ; preds = %303 + %307 = tail call float @llvm.nvvm.sqrt.approx.f(float %270) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +__nv_sqrtf.exit10: ; preds = %299, %301, %304, %306 + %.0.i8 = phi float [ %300, %299 ], [ %302, %301 ], [ %305, %304 ], [ %307, %306 ], !dbg !33 + %308 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i11 = icmp eq i32 %308, 0, !dbg !33 + %309 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i14 = icmp eq i32 %309, 0, !dbg !33 + br i1 %.not.i11, label %315, label %310, !dbg !33 + +310: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %313, label %311, !dbg !33 + +311: ; preds = %310 + %312 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %271) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +313: ; preds = %310 + %314 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %271) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +315: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %318, label %316, !dbg !33 + +316: ; preds = %315 + %317 = tail call float @llvm.nvvm.sqrt.rn.f(float %271) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +318: ; preds = %315 + %319 = tail call float @llvm.nvvm.sqrt.approx.f(float %271) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +__nv_sqrtf.exit15: ; preds = %311, %313, %316, %318 + %.0.i13 = phi float [ %312, %311 ], [ %314, %313 ], [ %317, %316 ], [ %319, %318 ], !dbg !33 + %320 = and i32 %11, 7, !dbg !16 + %321 = zext nneg i32 %320 to i64, !dbg !17 + %322 = or disjoint i64 %10, %321, !dbg !17 + %323 = getelementptr float, ptr addrspace(1) %2, i64 %322, !dbg !34 + %324 = getelementptr bfloat, ptr addrspace(1) %1, i64 %34, !dbg !35 + %325 = getelementptr bfloat, ptr addrspace(1) %324, i64 %39, !dbg !36 + %326 = getelementptr bfloat, ptr addrspace(1) %1, i64 %33, !dbg !35 + %327 = getelementptr bfloat, ptr addrspace(1) %326, i64 %39, !dbg !36 + %328 = getelementptr bfloat, ptr addrspace(1) %1, i64 %32, !dbg !35 + %329 = getelementptr bfloat, ptr addrspace(1) %328, i64 %39, !dbg !36 + %330 = getelementptr bfloat, ptr addrspace(1) %1, i64 %31, !dbg !35 + %331 = getelementptr bfloat, ptr addrspace(1) %330, i64 %39, !dbg !36 + %332 = icmp slt i64 %322, %24, !dbg !18 + %333 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i), !dbg !37 + %334 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i3), !dbg !37 + %335 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i8), !dbg !37 + %336 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i13), !dbg !37 + %337 = fmul float %333, %112, !dbg !38 + %338 = fmul float %333, %113, !dbg !38 + %339 = fmul float %333, %114, !dbg !38 + %340 = fmul float %333, %115, !dbg !38 + %341 = fmul float %333, %116, !dbg !38 + %342 = fmul float %333, %117, !dbg !38 + %343 = fmul float %333, %118, !dbg !38 + %344 = fmul float %333, %119, !dbg !38 + %345 = fmul float %334, %120, !dbg !38 + %346 = fmul float %334, %121, !dbg !38 + %347 = fmul float %334, %122, !dbg !38 + %348 = fmul float %334, %123, !dbg !38 + %349 = fmul float %334, %124, !dbg !38 + %350 = fmul float %334, %125, !dbg !38 + %351 = fmul float %334, %126, !dbg !38 + %352 = fmul float %334, %127, !dbg !38 + %353 = fmul float %335, %128, !dbg !38 + %354 = fmul float %335, %129, !dbg !38 + %355 = fmul float %335, %130, !dbg !38 + %356 = fmul float %335, %131, !dbg !38 + %357 = fmul float %335, %132, !dbg !38 + %358 = fmul float %335, %133, !dbg !38 + %359 = fmul float %335, %134, !dbg !38 + %360 = fmul float %335, %135, !dbg !38 + %361 = fmul float %336, %136, !dbg !38 + %362 = fmul float %336, %137, !dbg !38 + %363 = fmul float %336, %138, !dbg !38 + %364 = fmul float %336, %139, !dbg !38 + %365 = fmul float %336, %140, !dbg !38 + %366 = fmul float %336, %141, !dbg !38 + %367 = fmul float %336, %142, !dbg !38 + %368 = fmul float %336, %143, !dbg !38 + %369 = fptrunc float %337 to bfloat, !dbg !39 + %370 = fptrunc float %338 to bfloat, !dbg !39 + %371 = fptrunc float %339 to bfloat, !dbg !39 + %372 = fptrunc float %340 to bfloat, !dbg !39 + %373 = fptrunc float %341 to bfloat, !dbg !39 + %374 = fptrunc float %342 to bfloat, !dbg !39 + %375 = fptrunc float %343 to bfloat, !dbg !39 + %376 = fptrunc float %344 to bfloat, !dbg !39 + %377 = fptrunc float %345 to bfloat, !dbg !39 + %378 = fptrunc float %346 to bfloat, !dbg !39 + %379 = fptrunc float %347 to bfloat, !dbg !39 + %380 = fptrunc float %348 to bfloat, !dbg !39 + %381 = fptrunc float %349 to bfloat, !dbg !39 + %382 = fptrunc float %350 to bfloat, !dbg !39 + %383 = fptrunc float %351 to bfloat, !dbg !39 + %384 = fptrunc float %352 to bfloat, !dbg !39 + %385 = fptrunc float %353 to bfloat, !dbg !39 + %386 = fptrunc float %354 to bfloat, !dbg !39 + %387 = fptrunc float %355 to bfloat, !dbg !39 + %388 = fptrunc float %356 to bfloat, !dbg !39 + %389 = fptrunc float %357 to bfloat, !dbg !39 + %390 = fptrunc float %358 to bfloat, !dbg !39 + %391 = fptrunc float %359 to bfloat, !dbg !39 + %392 = fptrunc float %360 to bfloat, !dbg !39 + %393 = fptrunc float %361 to bfloat, !dbg !39 + %394 = fptrunc float %362 to bfloat, !dbg !39 + %395 = fptrunc float %363 to bfloat, !dbg !39 + %396 = fptrunc float %364 to bfloat, !dbg !39 + %397 = fptrunc float %365 to bfloat, !dbg !39 + %398 = fptrunc float %366 to bfloat, !dbg !39 + %399 = fptrunc float %367 to bfloat, !dbg !39 + %400 = fptrunc float %368 to bfloat, !dbg !39 + %401 = insertelement <2 x bfloat> poison, bfloat %369, i64 0, !dbg !40 + %402 = insertelement <2 x bfloat> %401, bfloat %370, i64 1, !dbg !40 + %403 = bitcast <2 x bfloat> %402 to i32, !dbg !40 + %404 = insertelement <2 x bfloat> poison, bfloat %371, i64 0, !dbg !40 + %405 = insertelement <2 x bfloat> %404, bfloat %372, i64 1, !dbg !40 + %406 = bitcast <2 x bfloat> %405 to i32, !dbg !40 + %407 = insertelement <2 x bfloat> poison, bfloat %373, i64 0, !dbg !40 + %408 = insertelement <2 x bfloat> %407, bfloat %374, i64 1, !dbg !40 + %409 = bitcast <2 x bfloat> %408 to i32, !dbg !40 + %410 = insertelement <2 x bfloat> poison, bfloat %375, i64 0, !dbg !40 + %411 = insertelement <2 x bfloat> %410, bfloat %376, i64 1, !dbg !40 + %412 = bitcast <2 x bfloat> %411 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %403, i32 %406, i32 %409, i32 %412, ptr addrspace(1) %331, i1 %25) #6, !dbg !40 + %413 = insertelement <2 x bfloat> poison, bfloat %377, i64 0, !dbg !40 + %414 = insertelement <2 x bfloat> %413, bfloat %378, i64 1, !dbg !40 + %415 = bitcast <2 x bfloat> %414 to i32, !dbg !40 + %416 = insertelement <2 x bfloat> poison, bfloat %379, i64 0, !dbg !40 + %417 = insertelement <2 x bfloat> %416, bfloat %380, i64 1, !dbg !40 + %418 = bitcast <2 x bfloat> %417 to i32, !dbg !40 + %419 = insertelement <2 x bfloat> poison, bfloat %381, i64 0, !dbg !40 + %420 = insertelement <2 x bfloat> %419, bfloat %382, i64 1, !dbg !40 + %421 = bitcast <2 x bfloat> %420 to i32, !dbg !40 + %422 = insertelement <2 x bfloat> poison, bfloat %383, i64 0, !dbg !40 + %423 = insertelement <2 x bfloat> %422, bfloat %384, i64 1, !dbg !40 + %424 = bitcast <2 x bfloat> %423 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %415, i32 %418, i32 %421, i32 %424, ptr addrspace(1) %329, i1 %26) #6, !dbg !40 + %425 = insertelement <2 x bfloat> poison, bfloat %385, i64 0, !dbg !40 + %426 = insertelement <2 x bfloat> %425, bfloat %386, i64 1, !dbg !40 + %427 = bitcast <2 x bfloat> %426 to i32, !dbg !40 + %428 = insertelement <2 x bfloat> poison, bfloat %387, i64 0, !dbg !40 + %429 = insertelement <2 x bfloat> %428, bfloat %388, i64 1, !dbg !40 + %430 = bitcast <2 x bfloat> %429 to i32, !dbg !40 + %431 = insertelement <2 x bfloat> poison, bfloat %389, i64 0, !dbg !40 + %432 = insertelement <2 x bfloat> %431, bfloat %390, i64 1, !dbg !40 + %433 = bitcast <2 x bfloat> %432 to i32, !dbg !40 + %434 = insertelement <2 x bfloat> poison, bfloat %391, i64 0, !dbg !40 + %435 = insertelement <2 x bfloat> %434, bfloat %392, i64 1, !dbg !40 + %436 = bitcast <2 x bfloat> %435 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %427, i32 %430, i32 %433, i32 %436, ptr addrspace(1) %327, i1 %27) #6, !dbg !40 + %437 = insertelement <2 x bfloat> poison, bfloat %393, i64 0, !dbg !40 + %438 = insertelement <2 x bfloat> %437, bfloat %394, i64 1, !dbg !40 + %439 = bitcast <2 x bfloat> %438 to i32, !dbg !40 + %440 = insertelement <2 x bfloat> poison, bfloat %395, i64 0, !dbg !40 + %441 = insertelement <2 x bfloat> %440, bfloat %396, i64 1, !dbg !40 + %442 = bitcast <2 x bfloat> %441 to i32, !dbg !40 + %443 = insertelement <2 x bfloat> poison, bfloat %397, i64 0, !dbg !40 + %444 = insertelement <2 x bfloat> %443, bfloat %398, i64 1, !dbg !40 + %445 = bitcast <2 x bfloat> %444 to i32, !dbg !40 + %446 = insertelement <2 x bfloat> poison, bfloat %399, i64 0, !dbg !40 + %447 = insertelement <2 x bfloat> %446, bfloat %400, i64 1, !dbg !40 + %448 = bitcast <2 x bfloat> %447 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %439, i32 %442, i32 %445, i32 %448, ptr addrspace(1) %325, i1 %28) #6, !dbg !40 + %449 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %12, !dbg !41 + %450 = bitcast float %333 to i32, !dbg !41 + %451 = bitcast float %334 to i32, !dbg !41 + %452 = bitcast float %335 to i32, !dbg !41 + %453 = bitcast float %336 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %449, i32 %450, i32 %451, i32 %452, i32 %453) #6, !dbg !41 + tail call void @llvm.nvvm.bar.warp.sync(i32 -1), !dbg !41 + %454 = shl nuw nsw i32 %11, 1, !dbg !41 + %455 = and i32 %454, 12, !dbg !41 + %456 = shl nuw nsw i32 %11, 4, !dbg !41 + %457 = and i32 %456, 16, !dbg !41 + %458 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %457, !dbg !41 + %459 = getelementptr inbounds nuw i8, ptr addrspace(3) %458, i32 %455, !dbg !41 + %460 = load i32, ptr addrspace(3) %459, align 4, !dbg !41 + %461 = and i32 %11, 24, !dbg !41 + %462 = icmp eq i32 %461, 0, !dbg !41 + %463 = and i1 %462, %332, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %460, ptr addrspace(1) %323, i1 %463) #6, !dbg !41 + ret void, !dbg !42 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.div.full(float, float) #3 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.bar.warp.sync(i32) #4 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #5 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.f(float) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="32" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) } +attributes #4 = { convergent nocallback nounwind } +attributes #5 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #6 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "l2norm_fwd_kernel", linkageName: "l2norm_fwd_kernel", scope: !1, file: !1, line: 69, type: !6, scopeLine: 69, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !10, !11, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 70, column: 24, scope: !5) +!14 = !DILocation(line: 70, column: 30, scope: !5) +!15 = !DILocation(line: 71, column: 16, scope: !5) +!16 = !DILocation(line: 71, column: 34, scope: !5) +!17 = !DILocation(line: 71, column: 21, scope: !5) +!18 = !DILocation(line: 73, column: 16, scope: !5) +!19 = !DILocation(line: 74, column: 30, scope: !5) +!20 = !DILocation(line: 75, column: 29, scope: !5) +!21 = !DILocation(line: 75, column: 14, scope: !5) +!22 = !DILocation(line: 75, column: 33, scope: !5) +!23 = !DILocation(line: 79, column: 18, scope: !5) +!24 = !DILocation(line: 79, column: 47, scope: !5) +!25 = !DILocation(line: 80, column: 38, scope: !5) +!26 = !DILocation(line: 263, column: 15, scope: !27, inlinedAt: !29) +!27 = distinct !DILexicalBlockFile(scope: !5, file: !28, discriminator: 0) +!28 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!29 = !DILocation(line: 293, column: 36, scope: !27, inlinedAt: !30) +!30 = !DILocation(line: 80, column: 43, scope: !31) +!31 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!32 = !DILocation(line: 80, column: 48, scope: !5) +!33 = !DILocation(line: 80, column: 25, scope: !5) +!34 = !DILocation(line: 77, column: 20, scope: !5) +!35 = !DILocation(line: 76, column: 14, scope: !5) +!36 = !DILocation(line: 76, column: 33, scope: !5) +!37 = !DILocation(line: 80, column: 17, scope: !5) +!38 = !DILocation(line: 81, column: 16, scope: !5) +!39 = !DILocation(line: 83, column: 25, scope: !5) +!40 = !DILocation(line: 83, column: 18, scope: !5) +!41 = !DILocation(line: 84, column: 21, scope: !5) +!42 = !DILocation(line: 84, column: 4, scope: !5) diff --git a/triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.ptx b/triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..3fc2508145d10f5bed3aa6f423050f6d381ce762 --- /dev/null +++ b/triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.ptx @@ -0,0 +1,642 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_fwd_kernel // -- Begin function l2norm_fwd_kernel +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; +.global .align 1 .b8 _$_str_$_2[17] = {95, 95, 67, 85, 68, 65, 95, 80, 82, 69, 67, 95, 83, 81, 82, 84}; + // @l2norm_fwd_kernel +.visible .entry l2norm_fwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_2, + .param .f32 l2norm_fwd_kernel_param_3, + .param .u32 l2norm_fwd_kernel_param_4, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_6 +) +.reqntid 32 +{ + .reg .pred %p<8>; + .reg .b16 %rs<33>; + .reg .b32 %r<196>; + .reg .b64 %rd<39>; + .loc 1 69 0 // l2norm.py:69:0 +$L__func_begin0: + .loc 1 69 0 // l2norm.py:69:0 + +// %bb.0: // %__nv_sqrtf.exit + ld.param.b64 %rd10, [l2norm_fwd_kernel_param_0]; + ld.param.b64 %rd11, [l2norm_fwd_kernel_param_1]; +$L__tmp0: + .loc 1 70 24 // l2norm.py:70:24 + mov.u32 %r40, %ctaid.x; + ld.param.b64 %rd12, [l2norm_fwd_kernel_param_2]; + .loc 1 71 16 // l2norm.py:71:16 + mul.wide.u32 %rd13, %r40, 8; + ld.param.b32 %r41, [l2norm_fwd_kernel_param_3]; + ld.param.s32 %rd14, [l2norm_fwd_kernel_param_4]; + .loc 1 71 34 // l2norm.py:71:34 + mov.u32 %r42, %tid.x; + and.b32 %r43, %r42, 16; + bfe.u32 %r44, %r42, 4, 1; + or.b32 %r45, %r44, 2; + or.b32 %r46, %r44, 4; + or.b32 %r47, %r44, 6; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd15, %r44; + cvt.u64.u32 %rd16, %r45; + cvt.u64.u32 %rd17, %r46; + cvt.u64.u32 %rd18, %r47; + or.b64 %rd19, %rd13, %rd15; + or.b64 %rd20, %rd13, %rd16; + or.b64 %rd21, %rd13, %rd17; + or.b64 %rd22, %rd13, %rd18; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p1, %rd19, %rd14; + setp.lt.s64 %p2, %rd20, %rd14; + setp.lt.s64 %p3, %rd21, %rd14; + setp.lt.s64 %p4, %rd22, %rd14; + .loc 1 74 30 // l2norm.py:74:30 + shl.b32 %r48, %r42, 3; + and.b32 %r49, %r48, 120; + .loc 1 75 14 // l2norm.py:75:14 + shl.b64 %rd23, %rd19, 8; + add.s64 %rd24, %rd10, %rd23; + shl.b64 %rd25, %rd20, 8; + add.s64 %rd26, %rd10, %rd25; + shl.b64 %rd27, %rd21, 8; + add.s64 %rd28, %rd10, %rd27; + shl.b64 %rd29, %rd22, 8; + add.s64 %rd30, %rd10, %rd29; + .loc 1 75 33 // l2norm.py:75:33 + mul.wide.u32 %rd31, %r49, 2; + add.s64 %rd1, %rd24, %rd31; + add.s64 %rd2, %rd26, %rd31; + add.s64 %rd3, %rd28, %rd31; + add.s64 %rd4, %rd30, %rd31; + mov.b32 %r5, 0; + .loc 1 79 18 // l2norm.py:79:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + mov.b32 {%rs5, %rs6}, %r3; + mov.b32 {%rs7, %rs8}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs9, %rs10}, %r6; + mov.b32 {%rs11, %rs12}, %r7; + mov.b32 {%rs13, %rs14}, %r8; + mov.b32 {%rs15, %rs16}, %r9; + // begin inline asm + mov.u32 %r10, %r5; + mov.u32 %r11, %r5; + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + @%p3 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs17, %rs18}, %r10; + mov.b32 {%rs19, %rs20}, %r11; + mov.b32 {%rs21, %rs22}, %r12; + mov.b32 {%rs23, %rs24}, %r13; + // begin inline asm + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + @%p4 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ]; + // end inline asm + mov.b32 {%rs25, %rs26}, %r14; + mov.b32 {%rs27, %rs28}, %r15; + mov.b32 {%rs29, %rs30}, %r16; + mov.b32 {%rs31, %rs32}, %r17; + .loc 1 79 47 // l2norm.py:79:47 + cvt.f32.bf16 %r50, %rs1; + cvt.f32.bf16 %r51, %rs2; + cvt.f32.bf16 %r52, %rs3; + cvt.f32.bf16 %r53, %rs4; + cvt.f32.bf16 %r54, %rs5; + cvt.f32.bf16 %r55, %rs6; + cvt.f32.bf16 %r56, %rs7; + cvt.f32.bf16 %r57, %rs8; + cvt.f32.bf16 %r58, %rs9; + cvt.f32.bf16 %r59, %rs10; + cvt.f32.bf16 %r60, %rs11; + cvt.f32.bf16 %r61, %rs12; + cvt.f32.bf16 %r62, %rs13; + cvt.f32.bf16 %r63, %rs14; + cvt.f32.bf16 %r64, %rs15; + cvt.f32.bf16 %r65, %rs16; + cvt.f32.bf16 %r66, %rs17; + cvt.f32.bf16 %r67, %rs18; + cvt.f32.bf16 %r68, %rs19; + cvt.f32.bf16 %r69, %rs20; + cvt.f32.bf16 %r70, %rs21; + cvt.f32.bf16 %r71, %rs22; + cvt.f32.bf16 %r72, %rs23; + cvt.f32.bf16 %r73, %rs24; + cvt.f32.bf16 %r74, %rs25; + cvt.f32.bf16 %r75, %rs26; + cvt.f32.bf16 %r76, %rs27; + cvt.f32.bf16 %r77, %rs28; + cvt.f32.bf16 %r78, %rs29; + cvt.f32.bf16 %r79, %rs30; + cvt.f32.bf16 %r80, %rs31; + cvt.f32.bf16 %r81, %rs32; + .loc 1 80 38 // l2norm.py:80:38 + mul.f32 %r82, %r51, %r51; + mul.f32 %r83, %r59, %r59; + mul.f32 %r84, %r67, %r67; + mul.f32 %r85, %r75, %r75; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + fma.rn.f32.bf16 %r86, %rs1, %rs1, %r82; + fma.rn.f32.bf16 %r87, %rs3, %rs3, %r86; + fma.rn.f32.bf16 %r88, %rs4, %rs4, %r87; + fma.rn.f32.bf16 %r89, %rs5, %rs5, %r88; + fma.rn.f32.bf16 %r90, %rs6, %rs6, %r89; + fma.rn.f32.bf16 %r91, %rs7, %rs7, %r90; + fma.rn.f32.bf16 %r92, %rs8, %rs8, %r91; + fma.rn.f32.bf16 %r93, %rs9, %rs9, %r83; + fma.rn.f32.bf16 %r94, %rs11, %rs11, %r93; + fma.rn.f32.bf16 %r95, %rs12, %rs12, %r94; + fma.rn.f32.bf16 %r96, %rs13, %rs13, %r95; + fma.rn.f32.bf16 %r97, %rs14, %rs14, %r96; + fma.rn.f32.bf16 %r98, %rs15, %rs15, %r97; + fma.rn.f32.bf16 %r99, %rs16, %rs16, %r98; + fma.rn.f32.bf16 %r100, %rs17, %rs17, %r84; + fma.rn.f32.bf16 %r101, %rs19, %rs19, %r100; + fma.rn.f32.bf16 %r102, %rs20, %rs20, %r101; + fma.rn.f32.bf16 %r103, %rs21, %rs21, %r102; + fma.rn.f32.bf16 %r104, %rs22, %rs22, %r103; + fma.rn.f32.bf16 %r105, %rs23, %rs23, %r104; + fma.rn.f32.bf16 %r106, %rs24, %rs24, %r105; + fma.rn.f32.bf16 %r107, %rs25, %rs25, %r85; + fma.rn.f32.bf16 %r108, %rs27, %rs27, %r107; + fma.rn.f32.bf16 %r109, %rs28, %rs28, %r108; + fma.rn.f32.bf16 %r110, %rs29, %rs29, %r109; + fma.rn.f32.bf16 %r111, %rs30, %rs30, %r110; + fma.rn.f32.bf16 %r112, %rs31, %rs31, %r111; + fma.rn.f32.bf16 %r113, %rs32, %rs32, %r112; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r114, %r92, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r115, %r92, %r114; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r116, %r115, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r117, %r115, %r116; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r118, %r117, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r119, %r117, %r118; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r120, %r119, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r121, %r119, %r120; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r122, %r99, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r123, %r99, %r122; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r124, %r123, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r125, %r123, %r124; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r126, %r125, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r127, %r125, %r126; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r128, %r127, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r129, %r127, %r128; +$L__tmp18: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r130, %r106, 8, 31, -1; +$L__tmp19: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r131, %r106, %r130; +$L__tmp20: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r132, %r131, 4, 31, -1; +$L__tmp21: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r133, %r131, %r132; +$L__tmp22: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r134, %r133, 2, 31, -1; +$L__tmp23: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r135, %r133, %r134; +$L__tmp24: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r136, %r135, 1, 31, -1; +$L__tmp25: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r137, %r135, %r136; +$L__tmp26: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r138, %r113, 8, 31, -1; +$L__tmp27: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r139, %r113, %r138; +$L__tmp28: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r140, %r139, 4, 31, -1; +$L__tmp29: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r141, %r139, %r140; +$L__tmp30: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r142, %r141, 2, 31, -1; +$L__tmp31: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r143, %r141, %r142; +$L__tmp32: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r144, %r143, 1, 31, -1; +$L__tmp33: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r145, %r143, %r144; +$L__tmp34: + .loc 1 80 48 // l2norm.py:80:48 + add.f32 %r146, %r41, %r121; + add.f32 %r147, %r41, %r129; + add.f32 %r148, %r41, %r137; + add.f32 %r149, %r41, %r145; + .loc 1 80 25 // l2norm.py:80:25 + sqrt.approx.ftz.f32 %r150, %r146; + sqrt.approx.ftz.f32 %r151, %r147; + sqrt.approx.ftz.f32 %r152, %r148; + sqrt.approx.ftz.f32 %r153, %r149; + .loc 1 71 34 // l2norm.py:71:34 + and.b32 %r154, %r42, 7; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd32, %r154; + or.b64 %rd33, %rd13, %rd32; + .loc 1 77 20 // l2norm.py:77:20 + shl.b64 %rd34, %rd33, 2; + add.s64 %rd9, %rd12, %rd34; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd35, %rd11, %rd29; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd8, %rd35, %rd31; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd36, %rd11, %rd27; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd7, %rd36, %rd31; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd37, %rd11, %rd25; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd6, %rd37, %rd31; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd38, %rd11, %rd23; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd5, %rd38, %rd31; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p6, %rd33, %rd14; + mov.b32 %r155, 0f3F800000; + .loc 1 80 17 // l2norm.py:80:17 + div.full.f32 %r35, %r155, %r150; + div.full.f32 %r36, %r155, %r151; + div.full.f32 %r37, %r155, %r152; + div.full.f32 %r38, %r155, %r153; + .loc 1 81 16 // l2norm.py:81:16 + mul.f32 %r156, %r35, %r50; + mul.f32 %r157, %r35, %r51; + mul.f32 %r158, %r35, %r52; + mul.f32 %r159, %r35, %r53; + mul.f32 %r160, %r35, %r54; + mul.f32 %r161, %r35, %r55; + mul.f32 %r162, %r35, %r56; + mul.f32 %r163, %r35, %r57; + mul.f32 %r164, %r36, %r58; + mul.f32 %r165, %r36, %r59; + mul.f32 %r166, %r36, %r60; + mul.f32 %r167, %r36, %r61; + mul.f32 %r168, %r36, %r62; + mul.f32 %r169, %r36, %r63; + mul.f32 %r170, %r36, %r64; + mul.f32 %r171, %r36, %r65; + mul.f32 %r172, %r37, %r66; + mul.f32 %r173, %r37, %r67; + mul.f32 %r174, %r37, %r68; + mul.f32 %r175, %r37, %r69; + mul.f32 %r176, %r37, %r70; + mul.f32 %r177, %r37, %r71; + mul.f32 %r178, %r37, %r72; + mul.f32 %r179, %r37, %r73; + mul.f32 %r180, %r38, %r74; + mul.f32 %r181, %r38, %r75; + mul.f32 %r182, %r38, %r76; + mul.f32 %r183, %r38, %r77; + mul.f32 %r184, %r38, %r78; + mul.f32 %r185, %r38, %r79; + mul.f32 %r186, %r38, %r80; + mul.f32 %r187, %r38, %r81; + .loc 1 83 18 // l2norm.py:83:18 + cvt.rn.bf16x2.f32 %r18, %r157, %r156; + cvt.rn.bf16x2.f32 %r19, %r159, %r158; + cvt.rn.bf16x2.f32 %r20, %r161, %r160; + cvt.rn.bf16x2.f32 %r21, %r163, %r162; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd5 + 0 ], { %r18, %r19, %r20, %r21 }; + // end inline asm + cvt.rn.bf16x2.f32 %r22, %r165, %r164; + cvt.rn.bf16x2.f32 %r23, %r167, %r166; + cvt.rn.bf16x2.f32 %r24, %r169, %r168; + cvt.rn.bf16x2.f32 %r25, %r171, %r170; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd6 + 0 ], { %r22, %r23, %r24, %r25 }; + // end inline asm + cvt.rn.bf16x2.f32 %r26, %r173, %r172; + cvt.rn.bf16x2.f32 %r27, %r175, %r174; + cvt.rn.bf16x2.f32 %r28, %r177, %r176; + cvt.rn.bf16x2.f32 %r29, %r179, %r178; + // begin inline asm + @%p3 st.global.v4.b32 [ %rd7 + 0 ], { %r26, %r27, %r28, %r29 }; + // end inline asm + cvt.rn.bf16x2.f32 %r30, %r181, %r180; + cvt.rn.bf16x2.f32 %r31, %r183, %r182; + cvt.rn.bf16x2.f32 %r32, %r185, %r184; + cvt.rn.bf16x2.f32 %r33, %r187, %r186; + // begin inline asm + @%p4 st.global.v4.b32 [ %rd8 + 0 ], { %r30, %r31, %r32, %r33 }; + // end inline asm + .loc 1 84 21 // l2norm.py:84:21 + mov.b32 %r188, global_smem; + add.s32 %r34, %r188, %r43; + // begin inline asm + st.shared.v4.b32 [ %r34 + 0 ], { %r35, %r36, %r37, %r38 }; + // end inline asm + bar.warp.sync -1; + shl.b32 %r189, %r42, 1; + and.b32 %r190, %r189, 12; + shl.b32 %r191, %r42, 4; + and.b32 %r192, %r191, 16; + add.s32 %r193, %r188, %r192; + add.s32 %r194, %r193, %r190; + ld.shared.b32 %r39, [%r194]; + and.b32 %r195, %r42, 24; + setp.eq.b32 %p7, %r195, 0; + and.pred %p5, %p7, %p6; + // begin inline asm + @%p5 st.global.b32 [ %rd9 + 0 ], { %r39 }; + // end inline asm + .loc 1 84 4 // l2norm.py:84:4 + ret; +$L__tmp35: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp34 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 80 // DW_AT_call_line +.b8 43 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp34 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.source b/triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..c7c5a735c81381ccb0efd0d2a8f8d86240068e11 --- /dev/null +++ b/triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.source @@ -0,0 +1,172 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc38 = loc(unknown) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc45 = loc("x"(#loc)) +#loc46 = loc("y"(#loc)) +#loc47 = loc("rstd"(#loc)) +#loc48 = loc("eps"(#loc)) +#loc49 = loc("T"(#loc)) +#loc81 = loc("input"(#loc36)) +#loc82 = loc("a"(#loc41)) +#loc83 = loc("b"(#loc41)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc50) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc51) + %o_t = arith.constant 8 : i32 loc(#loc52) + %o_t_1 = arith.constant 8 : i64 loc(#loc52) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc52) + %o_t_3 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32> loc(#loc53) + %o_t_4 = arith.extsi %o_t_3 : tensor<8xi32> to tensor<8xi64> loc(#loc54) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<8xi64> loc(#loc54) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<8xi64> loc(#loc54) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc55) + %m_t = arith.extsi %T : i32 to i64 loc(#loc56) + %m_t_7 = tt.splat %m_t : i64 -> tensor<8xi64> loc(#loc56) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<8xi64> loc(#loc56) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<8xi1> -> tensor<8x1xi1> loc(#loc57) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc58) + %m_x_10 = arith.constant 128 : i32 loc(#loc59) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc59) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc59) + %m_x_13 = tt.broadcast %m_x : tensor<8x1xi1> -> tensor<8x128xi1> loc(#loc60) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<8x128xi1> loc(#loc60) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<8x128xi1> loc(#loc60) + %p_x = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc61) + %p_x_16 = arith.constant 128 : i32 loc(#loc62) + %p_x_17 = arith.constant 128 : i64 loc(#loc62) + %p_x_18 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc62) + %p_x_19 = arith.muli %p_x, %p_x_18 : tensor<8x1xi64> loc(#loc62) + %p_x_20 = tt.splat %x : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc63) + %p_x_21 = tt.addptr %p_x_20, %p_x_19 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc63) + %p_x_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc64) + %p_x_23 = tt.broadcast %p_x_21 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc65) + %p_x_24 = tt.broadcast %p_x_22 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc65) + %p_x_25 = tt.addptr %p_x_23, %p_x_24 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc65) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc66) + %p_y_26 = arith.constant 128 : i32 loc(#loc67) + %p_y_27 = arith.constant 128 : i64 loc(#loc67) + %p_y_28 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc67) + %p_y_29 = arith.muli %p_y, %p_y_28 : tensor<8x1xi64> loc(#loc67) + %p_y_30 = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc68) + %p_y_31 = tt.addptr %p_y_30, %p_y_29 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc68) + %p_y_32 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc69) + %p_y_33 = tt.broadcast %p_y_31 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc70) + %p_y_34 = tt.broadcast %p_y_32 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc70) + %p_y_35 = tt.addptr %p_y_33, %p_y_34 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc70) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr> loc(#loc71) + %p_rstd_36 = tt.addptr %p_rstd, %o_t_6 : tensor<8x!tt.ptr>, tensor<8xi64> loc(#loc71) + %b_x = arith.constant 0.000000e+00 : f32 loc(#loc72) + %b_x_37 = arith.constant dense<0.000000e+00> : tensor<8x128xf32> loc(#loc72) + %b_x_38 = arith.truncf %b_x_37 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc72) + %b_x_39 = tt.load %p_x_25, %m_x_15, %b_x_38 : tensor<8x128x!tt.ptr> loc(#loc72) + %b_x_40 = arith.extf %b_x_39 : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc73) + %b_rstd = arith.mulf %b_x_40, %b_x_40 : tensor<8x128xf32> loc(#loc74) + %b_rstd_41 = tt.call @triton.language.standard.sum__fp32S8_128S_c1_cFalse_cNone(%b_rstd) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc75) + %b_rstd_42 = tt.splat %eps : f32 -> tensor<8xf32> loc(#loc76) + %b_rstd_43 = arith.addf %b_rstd_41, %b_rstd_42 : tensor<8xf32> loc(#loc76) + %b_rstd_44 = math.sqrt %b_rstd_43 : tensor<8xf32> loc(#loc77) + %b_rstd_45 = arith.constant 1 : i32 loc(#loc78) + %b_rstd_46 = arith.constant 1.000000e+00 : f32 loc(#loc78) + %b_rstd_47 = arith.constant dense<1.000000e+00> : tensor<8xf32> loc(#loc78) + %b_rstd_48 = arith.divf %b_rstd_47, %b_rstd_44 : tensor<8xf32> loc(#loc78) + %b_y = tt.expand_dims %b_rstd_48 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc79) + %b_y_49 = tt.broadcast %b_y : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc80) + %b_y_50 = arith.mulf %b_x_40, %b_y_49 : tensor<8x128xf32> loc(#loc80) + %0 = arith.truncf %b_y_50 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc32) + tt.store %p_y_35, %0, %m_x_15 : tensor<8x128x!tt.ptr> loc(#loc33) + tt.store %p_rstd_36, %b_rstd_48, %m_t_8 : tensor<8x!tt.ptr> loc(#loc34) + tt.return loc(#loc35) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S8_128S_c1_cFalse_cNone(%input: tensor<8x128xf32> loc("input"(#loc36))) -> tensor<8xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc37) + tt.reduce.return %2 : f32 loc(#loc37) + }) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc37) + tt.return %0 : tensor<8xf32> loc(#loc39) + ^bb1: // no predecessors + %1 = ub.poison : tensor<8xf32> loc(#loc40) + tt.return %1 : tensor<8xf32> loc(#loc40) + } loc(#loc36) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc41)), %b: f32 loc("b"(#loc41))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc42) + tt.return %0 : f32 loc(#loc43) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc44) + tt.return %1 : f32 loc(#loc44) + } loc(#loc41) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:18) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:29) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:37) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc50 = loc("i_t"(#loc1)) +#loc51 = loc("i_t"(#loc2)) +#loc52 = loc("o_t"(#loc3)) +#loc53 = loc("o_t"(#loc4)) +#loc54 = loc("o_t"(#loc5)) +#loc55 = loc("o_d"(#loc6)) +#loc56 = loc("m_t"(#loc7)) +#loc57 = loc("m_x"(#loc8)) +#loc58 = loc("m_x"(#loc9)) +#loc59 = loc("m_x"(#loc10)) +#loc60 = loc("m_x"(#loc11)) +#loc61 = loc("p_x"(#loc12)) +#loc62 = loc("p_x"(#loc13)) +#loc63 = loc("p_x"(#loc14)) +#loc64 = loc("p_x"(#loc15)) +#loc65 = loc("p_x"(#loc16)) +#loc66 = loc("p_y"(#loc17)) +#loc67 = loc("p_y"(#loc18)) +#loc68 = loc("p_y"(#loc19)) +#loc69 = loc("p_y"(#loc20)) +#loc70 = loc("p_y"(#loc21)) +#loc71 = loc("p_rstd"(#loc22)) +#loc72 = loc("b_x"(#loc23)) +#loc73 = loc("b_x"(#loc24)) +#loc74 = loc("b_rstd"(#loc25)) +#loc75 = loc("b_rstd"(#loc26)) +#loc76 = loc("b_rstd"(#loc27)) +#loc77 = loc("b_rstd"(#loc28)) +#loc78 = loc("b_rstd"(#loc29)) +#loc79 = loc("b_y"(#loc30)) +#loc80 = loc("b_y"(#loc31)) diff --git a/triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.ttgir b/triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..1c4d2f1ae1e45290e174028273575be07fb7713b --- /dev/null +++ b/triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.ttgir @@ -0,0 +1,135 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [1, 1], order = [1, 0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [1], order = [0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc1 = loc(unknown) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc34 = loc("x"(#loc)) +#loc35 = loc("y"(#loc)) +#loc36 = loc("rstd"(#loc)) +#loc37 = loc("eps"(#loc)) +#loc38 = loc("T"(#loc)) +#loc59 = loc("b_rstd"(#loc23)) +#loc66 = loc(callsite(#loc1 at #loc59)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 1 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<1.000000e+00> : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<8x128xbf16, #blocked> loc(#loc1) + %c8_i64 = arith.constant 8 : i64 loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<8x1xi64, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc39) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc40) + %o_t = arith.muli %i_t_3, %c8_i64 : i64 loc(#loc41) + %o_t_4 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc42) + %o_t_5 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32, #blocked1> loc(#loc42) + %o_t_6 = arith.extsi %o_t_4 : tensor<8xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_7 = arith.extsi %o_t_5 : tensor<8xi32, #blocked1> to tensor<8xi64, #blocked1> loc(#loc43) + %o_t_8 = tt.splat %o_t : i64 -> tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_9 = tt.splat %o_t : i64 -> tensor<8xi64, #blocked1> loc(#loc43) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<8xi64, #blocked1> loc(#loc43) + %m_t = arith.extsi %T : i32 to i64 loc(#loc44) + %m_t_12 = tt.splat %m_t : i64 -> tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_13 = tt.splat %m_t : i64 -> tensor<8xi64, #blocked1> loc(#loc44) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<8xi64, #blocked1> loc(#loc44) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<8xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xi1, #blocked> loc(#loc45) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc46) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc46) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_1 : tensor<1x128xi32, #blocked> loc(#loc47) + %m_x_19 = tt.broadcast %m_x : tensor<8x1xi1, #blocked> -> tensor<8x128xi1, #blocked> loc(#loc48) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<8x128xi1, #blocked> loc(#loc48) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<8x128xi1, #blocked> loc(#loc48) + %p_x = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xi64, #blocked> loc(#loc49) + %p_x_22 = arith.muli %p_x, %cst_2 : tensor<8x1xi64, #blocked> loc(#loc50) + %p_x_23 = tt.splat %x : !tt.ptr -> tensor<8x1x!tt.ptr, #blocked> loc(#loc51) + %p_x_24 = tt.addptr %p_x_23, %p_x_22 : tensor<8x1x!tt.ptr, #blocked>, tensor<8x1xi64, #blocked> loc(#loc51) + %p_x_25 = tt.broadcast %p_x_24 : tensor<8x1x!tt.ptr, #blocked> -> tensor<8x128x!tt.ptr, #blocked> loc(#loc52) + %p_x_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<8x128xi32, #blocked> loc(#loc52) + %p_x_27 = tt.addptr %p_x_25, %p_x_26 : tensor<8x128x!tt.ptr, #blocked>, tensor<8x128xi32, #blocked> loc(#loc52) + %p_y = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr, #blocked> loc(#loc53) + %p_y_28 = tt.addptr %p_y, %p_x_22 : tensor<8x1x!tt.ptr, #blocked>, tensor<8x1xi64, #blocked> loc(#loc53) + %p_y_29 = tt.broadcast %p_y_28 : tensor<8x1x!tt.ptr, #blocked> -> tensor<8x128x!tt.ptr, #blocked> loc(#loc54) + %p_y_30 = tt.addptr %p_y_29, %p_x_26 : tensor<8x128x!tt.ptr, #blocked>, tensor<8x128xi32, #blocked> loc(#loc54) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr, #blocked1> loc(#loc55) + %p_rstd_31 = tt.addptr %p_rstd, %o_t_11 : tensor<8x!tt.ptr, #blocked1>, tensor<8xi64, #blocked1> loc(#loc55) + %b_x = tt.load %p_x_27, %m_x_21, %cst_0 : tensor<8x128x!tt.ptr, #blocked> loc(#loc56) + %b_x_32 = arith.extf %b_x : tensor<8x128xbf16, #blocked> to tensor<8x128xf32, #blocked> loc(#loc57) + %b_rstd = arith.mulf %b_x_32, %b_x_32 : tensor<8x128xf32, #blocked> loc(#loc58) + %b_rstd_33 = "tt.reduce"(%b_rstd) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_40: f32 loc(callsite(#loc1 at #loc59)), %b_rstd_41: f32 loc(callsite(#loc1 at #loc59))): + %b_rstd_42 = arith.addf %b_rstd_40, %b_rstd_41 : f32 loc(#loc67) + tt.reduce.return %b_rstd_42 : f32 loc(#loc65) + }) : (tensor<8x128xf32, #blocked>) -> tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc65) + %b_rstd_34 = tt.splat %eps : f32 -> tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_36 = math.sqrt %b_rstd_35 : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc61) + %b_rstd_37 = arith.divf %cst, %b_rstd_36 : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc62) + %b_y = tt.expand_dims %b_rstd_37 {axis = 1 : i32} : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xf32, #blocked> loc(#loc63) + %b_y_38 = tt.broadcast %b_y : tensor<8x1xf32, #blocked> -> tensor<8x128xf32, #blocked> loc(#loc64) + %b_y_39 = arith.mulf %b_x_32, %b_y_38 : tensor<8x128xf32, #blocked> loc(#loc64) + %0 = arith.truncf %b_y_39 : tensor<8x128xf32, #blocked> to tensor<8x128xbf16, #blocked> loc(#loc30) + tt.store %p_y_30, %0, %m_x_21 : tensor<8x128x!tt.ptr, #blocked> loc(#loc31) + %1 = ttg.convert_layout %b_rstd_37 : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8xf32, #blocked1> loc(#loc32) + tt.store %p_rstd_31, %1, %m_t_15 : tensor<8x!tt.ptr, #blocked1> loc(#loc32) + tt.return loc(#loc33) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc39 = loc("i_t"(#loc2)) +#loc40 = loc("i_t"(#loc3)) +#loc41 = loc("o_t"(#loc4)) +#loc42 = loc("o_t"(#loc5)) +#loc43 = loc("o_t"(#loc6)) +#loc44 = loc("m_t"(#loc7)) +#loc45 = loc("m_x"(#loc8)) +#loc46 = loc("m_x"(#loc9)) +#loc47 = loc("m_x"(#loc10)) +#loc48 = loc("m_x"(#loc11)) +#loc49 = loc("p_x"(#loc12)) +#loc50 = loc("p_x"(#loc13)) +#loc51 = loc("p_x"(#loc14)) +#loc52 = loc("p_x"(#loc15)) +#loc53 = loc("p_y"(#loc16)) +#loc54 = loc("p_y"(#loc17)) +#loc55 = loc("p_rstd"(#loc18)) +#loc56 = loc("b_x"(#loc19)) +#loc57 = loc("b_x"(#loc20)) +#loc58 = loc("b_rstd"(#loc21)) +#loc60 = loc("b_rstd"(#loc25)) +#loc61 = loc("b_rstd"(#loc26)) +#loc62 = loc("b_rstd"(#loc27)) +#loc63 = loc("b_y"(#loc28)) +#loc64 = loc("b_y"(#loc29)) +#loc65 = loc(callsite(#loc22 at #loc59)) +#loc67 = loc(callsite(#loc24 at #loc65)) diff --git a/triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.ttir b/triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..8bbf17da02489a0d195ed0bd020180f9e4ac2496 --- /dev/null +++ b/triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.ttir @@ -0,0 +1,128 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc3 = loc(unknown) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc35 = loc("x"(#loc)) +#loc36 = loc("y"(#loc)) +#loc37 = loc("rstd"(#loc)) +#loc38 = loc("eps"(#loc)) +#loc39 = loc("T"(#loc)) +#loc62 = loc("b_rstd"(#loc25)) +#loc68 = loc(callsite(#loc3 at #loc62)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %b_x = arith.constant dense<0.000000e+00> : tensor<8x128xbf16> loc(#loc40) + %b_rstd = arith.constant dense<1.000000e+00> : tensor<8xf32> loc(#loc41) + %cst = arith.constant dense<128> : tensor<8x1xi64> loc(#loc3) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc42) + %o_t = arith.constant 8 : i64 loc(#loc43) + %i_t = tt.get_program_id x : i32 loc(#loc44) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc45) + %o_t_1 = arith.muli %i_t_0, %o_t : i64 loc(#loc43) + %o_t_2 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32> loc(#loc46) + %o_t_3 = arith.extsi %o_t_2 : tensor<8xi32> to tensor<8xi64> loc(#loc47) + %o_t_4 = tt.splat %o_t_1 : i64 -> tensor<8xi64> loc(#loc47) + %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<8xi64> loc(#loc47) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc48) + %m_t = arith.extsi %T : i32 to i64 loc(#loc49) + %m_t_6 = tt.splat %m_t : i64 -> tensor<8xi64> loc(#loc49) + %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<8xi64> loc(#loc49) + %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<8xi1> -> tensor<8x1xi1> loc(#loc50) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc51) + %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc42) + %m_x_11 = tt.broadcast %m_x_8 : tensor<8x1xi1> -> tensor<8x128xi1> loc(#loc52) + %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<8x128xi1> loc(#loc52) + %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<8x128xi1> loc(#loc52) + %p_x = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc53) + %p_x_14 = arith.muli %p_x, %cst : tensor<8x1xi64> loc(#loc54) + %p_x_15 = tt.splat %x : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc55) + %p_x_16 = tt.addptr %p_x_15, %p_x_14 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc55) + %p_x_17 = tt.broadcast %p_x_16 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc56) + %p_x_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc56) + %p_x_19 = tt.addptr %p_x_17, %p_x_18 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc56) + %p_y = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc57) + %p_y_20 = tt.addptr %p_y, %p_x_14 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc57) + %p_y_21 = tt.broadcast %p_y_20 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc58) + %p_y_22 = tt.addptr %p_y_21, %p_x_18 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr> loc(#loc59) + %p_rstd_23 = tt.addptr %p_rstd, %o_t_5 : tensor<8x!tt.ptr>, tensor<8xi64> loc(#loc59) + %b_x_24 = tt.load %p_x_19, %m_x_13, %b_x : tensor<8x128x!tt.ptr> loc(#loc40) + %b_x_25 = arith.extf %b_x_24 : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc60) + %b_rstd_26 = arith.mulf %b_x_25, %b_x_25 : tensor<8x128xf32> loc(#loc61) + %b_rstd_27 = "tt.reduce"(%b_rstd_26) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_34: f32 loc(callsite(#loc3 at #loc62)), %b_rstd_35: f32 loc(callsite(#loc3 at #loc62))): + %b_rstd_36 = arith.addf %b_rstd_34, %b_rstd_35 : f32 loc(#loc69) + tt.reduce.return %b_rstd_36 : f32 loc(#loc67) + }) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc67) + %b_rstd_28 = tt.splat %eps : f32 -> tensor<8xf32> loc(#loc63) + %b_rstd_29 = arith.addf %b_rstd_27, %b_rstd_28 : tensor<8xf32> loc(#loc63) + %b_rstd_30 = math.sqrt %b_rstd_29 : tensor<8xf32> loc(#loc64) + %b_rstd_31 = arith.divf %b_rstd, %b_rstd_30 : tensor<8xf32> loc(#loc41) + %b_y = tt.expand_dims %b_rstd_31 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc65) + %b_y_32 = tt.broadcast %b_y : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc66) + %b_y_33 = arith.mulf %b_x_25, %b_y_32 : tensor<8x128xf32> loc(#loc66) + %0 = arith.truncf %b_y_33 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc31) + tt.store %p_y_22, %0, %m_x_13 : tensor<8x128x!tt.ptr> loc(#loc32) + tt.store %p_rstd_23, %b_rstd_31, %m_t_7 : tensor<8x!tt.ptr> loc(#loc33) + tt.return loc(#loc34) + } loc(#loc) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc40 = loc("b_x"(#loc1)) +#loc41 = loc("b_rstd"(#loc2)) +#loc42 = loc("m_x"(#loc4)) +#loc43 = loc("o_t"(#loc5)) +#loc44 = loc("i_t"(#loc6)) +#loc45 = loc("i_t"(#loc7)) +#loc46 = loc("o_t"(#loc8)) +#loc47 = loc("o_t"(#loc9)) +#loc48 = loc("o_d"(#loc10)) +#loc49 = loc("m_t"(#loc11)) +#loc50 = loc("m_x"(#loc12)) +#loc51 = loc("m_x"(#loc13)) +#loc52 = loc("m_x"(#loc14)) +#loc53 = loc("p_x"(#loc15)) +#loc54 = loc("p_x"(#loc16)) +#loc55 = loc("p_x"(#loc17)) +#loc56 = loc("p_x"(#loc18)) +#loc57 = loc("p_y"(#loc19)) +#loc58 = loc("p_y"(#loc20)) +#loc59 = loc("p_rstd"(#loc21)) +#loc60 = loc("b_x"(#loc22)) +#loc61 = loc("b_rstd"(#loc23)) +#loc63 = loc("b_rstd"(#loc27)) +#loc64 = loc("b_rstd"(#loc28)) +#loc65 = loc("b_y"(#loc29)) +#loc66 = loc("b_y"(#loc30)) +#loc67 = loc(callsite(#loc24 at #loc62)) +#loc69 = loc(callsite(#loc26 at #loc67)) diff --git a/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/__grp__l2norm_fwd_kernel.json b/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/__grp__l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..2e608ef27c8b9ef5e3427e1b7e32c532884d73a8 --- /dev/null +++ b/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/__grp__l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_fwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.source", "l2norm_fwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.ttir", "l2norm_fwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.ttgir", "l2norm_fwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.llir", "l2norm_fwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.ptx", "l2norm_fwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.cubin", "l2norm_fwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.cubin b/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..659e774d5c1cdcc61dbedcb70476363832000050 Binary files /dev/null and b/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.cubin differ diff --git a/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.json b/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..2507dba44e4a933bdf3ef2da2ee49fc902e2304e --- /dev/null +++ b/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"hash": "0456a746cb418c500e0da68ae7c294d8470e84db5fdaded2d9aaa4dd5f0fbbb8", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 2, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 64, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_fwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.llir b/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..267204aaecd30d546f721969de45a97dabd6d352 --- /dev/null +++ b/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.llir @@ -0,0 +1,630 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 +@.str.2 = private unnamed_addr constant [17 x i8] c"__CUDA_PREC_SQRT\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_fwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, float %3, i32 %4, ptr addrspace(1) readnone captures(none) %5, ptr addrspace(1) readnone captures(none) %6) local_unnamed_addr #0 !dbg !5 { + %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %9 = zext nneg i32 %8 to i64, !dbg !14 + %10 = shl nuw nsw i64 %9, 4, !dbg !15 + %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !16 + %12 = and i32 %11, 48, !dbg !16 + %13 = lshr exact i32 %12, 4, !dbg !16 + %14 = or disjoint i32 %13, 4, !dbg !16 + %15 = or disjoint i32 %13, 8, !dbg !16 + %16 = or disjoint i32 %13, 12, !dbg !16 + %17 = and i32 %11, 15, !dbg !16 + %18 = zext nneg i32 %13 to i64, !dbg !17 + %19 = zext nneg i32 %14 to i64, !dbg !17 + %20 = zext nneg i32 %15 to i64, !dbg !17 + %21 = zext nneg i32 %16 to i64, !dbg !17 + %22 = or disjoint i64 %10, %18, !dbg !17 + %23 = or disjoint i64 %10, %19, !dbg !17 + %24 = or disjoint i64 %10, %20, !dbg !17 + %25 = or disjoint i64 %10, %21, !dbg !17 + %26 = sext i32 %4 to i64, !dbg !18 + %27 = icmp slt i64 %22, %26, !dbg !18 + %28 = icmp slt i64 %23, %26, !dbg !18 + %29 = icmp slt i64 %24, %26, !dbg !18 + %30 = icmp slt i64 %25, %26, !dbg !18 + %31 = shl nuw nsw i32 %17, 3, !dbg !19 + %32 = shl nuw nsw i64 %22, 7, !dbg !20 + %33 = shl nuw nsw i64 %23, 7, !dbg !20 + %34 = shl nuw nsw i64 %24, 7, !dbg !20 + %35 = shl nuw nsw i64 %25, 7, !dbg !20 + %36 = getelementptr bfloat, ptr addrspace(1) %0, i64 %32, !dbg !21 + %37 = getelementptr bfloat, ptr addrspace(1) %0, i64 %33, !dbg !21 + %38 = getelementptr bfloat, ptr addrspace(1) %0, i64 %34, !dbg !21 + %39 = getelementptr bfloat, ptr addrspace(1) %0, i64 %35, !dbg !21 + %40 = zext nneg i32 %31 to i64, !dbg !22 + %41 = getelementptr bfloat, ptr addrspace(1) %36, i64 %40, !dbg !22 + %42 = getelementptr bfloat, ptr addrspace(1) %37, i64 %40, !dbg !22 + %43 = getelementptr bfloat, ptr addrspace(1) %38, i64 %40, !dbg !22 + %44 = getelementptr bfloat, ptr addrspace(1) %39, i64 %40, !dbg !22 + %45 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %41, i1 %27) #6, !dbg !23 + %46 = extractvalue { i32, i32, i32, i32 } %45, 0, !dbg !23 + %47 = bitcast i32 %46 to <2 x bfloat>, !dbg !23 + %48 = extractvalue { i32, i32, i32, i32 } %45, 1, !dbg !23 + %49 = bitcast i32 %48 to <2 x bfloat>, !dbg !23 + %50 = extractvalue { i32, i32, i32, i32 } %45, 2, !dbg !23 + %51 = bitcast i32 %50 to <2 x bfloat>, !dbg !23 + %52 = extractvalue { i32, i32, i32, i32 } %45, 3, !dbg !23 + %53 = bitcast i32 %52 to <2 x bfloat>, !dbg !23 + %54 = extractelement <2 x bfloat> %47, i64 0, !dbg !23 + %55 = extractelement <2 x bfloat> %47, i64 1, !dbg !23 + %56 = extractelement <2 x bfloat> %49, i64 0, !dbg !23 + %57 = extractelement <2 x bfloat> %49, i64 1, !dbg !23 + %58 = extractelement <2 x bfloat> %51, i64 0, !dbg !23 + %59 = extractelement <2 x bfloat> %51, i64 1, !dbg !23 + %60 = extractelement <2 x bfloat> %53, i64 0, !dbg !23 + %61 = extractelement <2 x bfloat> %53, i64 1, !dbg !23 + %62 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %42, i1 %28) #6, !dbg !23 + %63 = extractvalue { i32, i32, i32, i32 } %62, 0, !dbg !23 + %64 = bitcast i32 %63 to <2 x bfloat>, !dbg !23 + %65 = extractvalue { i32, i32, i32, i32 } %62, 1, !dbg !23 + %66 = bitcast i32 %65 to <2 x bfloat>, !dbg !23 + %67 = extractvalue { i32, i32, i32, i32 } %62, 2, !dbg !23 + %68 = bitcast i32 %67 to <2 x bfloat>, !dbg !23 + %69 = extractvalue { i32, i32, i32, i32 } %62, 3, !dbg !23 + %70 = bitcast i32 %69 to <2 x bfloat>, !dbg !23 + %71 = extractelement <2 x bfloat> %64, i64 0, !dbg !23 + %72 = extractelement <2 x bfloat> %64, i64 1, !dbg !23 + %73 = extractelement <2 x bfloat> %66, i64 0, !dbg !23 + %74 = extractelement <2 x bfloat> %66, i64 1, !dbg !23 + %75 = extractelement <2 x bfloat> %68, i64 0, !dbg !23 + %76 = extractelement <2 x bfloat> %68, i64 1, !dbg !23 + %77 = extractelement <2 x bfloat> %70, i64 0, !dbg !23 + %78 = extractelement <2 x bfloat> %70, i64 1, !dbg !23 + %79 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %43, i1 %29) #6, !dbg !23 + %80 = extractvalue { i32, i32, i32, i32 } %79, 0, !dbg !23 + %81 = bitcast i32 %80 to <2 x bfloat>, !dbg !23 + %82 = extractvalue { i32, i32, i32, i32 } %79, 1, !dbg !23 + %83 = bitcast i32 %82 to <2 x bfloat>, !dbg !23 + %84 = extractvalue { i32, i32, i32, i32 } %79, 2, !dbg !23 + %85 = bitcast i32 %84 to <2 x bfloat>, !dbg !23 + %86 = extractvalue { i32, i32, i32, i32 } %79, 3, !dbg !23 + %87 = bitcast i32 %86 to <2 x bfloat>, !dbg !23 + %88 = extractelement <2 x bfloat> %81, i64 0, !dbg !23 + %89 = extractelement <2 x bfloat> %81, i64 1, !dbg !23 + %90 = extractelement <2 x bfloat> %83, i64 0, !dbg !23 + %91 = extractelement <2 x bfloat> %83, i64 1, !dbg !23 + %92 = extractelement <2 x bfloat> %85, i64 0, !dbg !23 + %93 = extractelement <2 x bfloat> %85, i64 1, !dbg !23 + %94 = extractelement <2 x bfloat> %87, i64 0, !dbg !23 + %95 = extractelement <2 x bfloat> %87, i64 1, !dbg !23 + %96 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %44, i1 %30) #6, !dbg !23 + %97 = extractvalue { i32, i32, i32, i32 } %96, 0, !dbg !23 + %98 = bitcast i32 %97 to <2 x bfloat>, !dbg !23 + %99 = extractvalue { i32, i32, i32, i32 } %96, 1, !dbg !23 + %100 = bitcast i32 %99 to <2 x bfloat>, !dbg !23 + %101 = extractvalue { i32, i32, i32, i32 } %96, 2, !dbg !23 + %102 = bitcast i32 %101 to <2 x bfloat>, !dbg !23 + %103 = extractvalue { i32, i32, i32, i32 } %96, 3, !dbg !23 + %104 = bitcast i32 %103 to <2 x bfloat>, !dbg !23 + %105 = extractelement <2 x bfloat> %98, i64 0, !dbg !23 + %106 = extractelement <2 x bfloat> %98, i64 1, !dbg !23 + %107 = extractelement <2 x bfloat> %100, i64 0, !dbg !23 + %108 = extractelement <2 x bfloat> %100, i64 1, !dbg !23 + %109 = extractelement <2 x bfloat> %102, i64 0, !dbg !23 + %110 = extractelement <2 x bfloat> %102, i64 1, !dbg !23 + %111 = extractelement <2 x bfloat> %104, i64 0, !dbg !23 + %112 = extractelement <2 x bfloat> %104, i64 1, !dbg !23 + %113 = fpext bfloat %54 to float, !dbg !24 + %114 = fpext bfloat %55 to float, !dbg !24 + %115 = fpext bfloat %56 to float, !dbg !24 + %116 = fpext bfloat %57 to float, !dbg !24 + %117 = fpext bfloat %58 to float, !dbg !24 + %118 = fpext bfloat %59 to float, !dbg !24 + %119 = fpext bfloat %60 to float, !dbg !24 + %120 = fpext bfloat %61 to float, !dbg !24 + %121 = fpext bfloat %71 to float, !dbg !24 + %122 = fpext bfloat %72 to float, !dbg !24 + %123 = fpext bfloat %73 to float, !dbg !24 + %124 = fpext bfloat %74 to float, !dbg !24 + %125 = fpext bfloat %75 to float, !dbg !24 + %126 = fpext bfloat %76 to float, !dbg !24 + %127 = fpext bfloat %77 to float, !dbg !24 + %128 = fpext bfloat %78 to float, !dbg !24 + %129 = fpext bfloat %88 to float, !dbg !24 + %130 = fpext bfloat %89 to float, !dbg !24 + %131 = fpext bfloat %90 to float, !dbg !24 + %132 = fpext bfloat %91 to float, !dbg !24 + %133 = fpext bfloat %92 to float, !dbg !24 + %134 = fpext bfloat %93 to float, !dbg !24 + %135 = fpext bfloat %94 to float, !dbg !24 + %136 = fpext bfloat %95 to float, !dbg !24 + %137 = fpext bfloat %105 to float, !dbg !24 + %138 = fpext bfloat %106 to float, !dbg !24 + %139 = fpext bfloat %107 to float, !dbg !24 + %140 = fpext bfloat %108 to float, !dbg !24 + %141 = fpext bfloat %109 to float, !dbg !24 + %142 = fpext bfloat %110 to float, !dbg !24 + %143 = fpext bfloat %111 to float, !dbg !24 + %144 = fpext bfloat %112 to float, !dbg !24 + %145 = fmul float %113, %113, !dbg !25 + %146 = fmul float %114, %114, !dbg !25 + %147 = fmul float %115, %115, !dbg !25 + %148 = fmul float %116, %116, !dbg !25 + %149 = fmul float %117, %117, !dbg !25 + %150 = fmul float %118, %118, !dbg !25 + %151 = fmul float %119, %119, !dbg !25 + %152 = fmul float %120, %120, !dbg !25 + %153 = fmul float %121, %121, !dbg !25 + %154 = fmul float %122, %122, !dbg !25 + %155 = fmul float %123, %123, !dbg !25 + %156 = fmul float %124, %124, !dbg !25 + %157 = fmul float %125, %125, !dbg !25 + %158 = fmul float %126, %126, !dbg !25 + %159 = fmul float %127, %127, !dbg !25 + %160 = fmul float %128, %128, !dbg !25 + %161 = fmul float %129, %129, !dbg !25 + %162 = fmul float %130, %130, !dbg !25 + %163 = fmul float %131, %131, !dbg !25 + %164 = fmul float %132, %132, !dbg !25 + %165 = fmul float %133, %133, !dbg !25 + %166 = fmul float %134, %134, !dbg !25 + %167 = fmul float %135, %135, !dbg !25 + %168 = fmul float %136, %136, !dbg !25 + %169 = fmul float %137, %137, !dbg !25 + %170 = fmul float %138, %138, !dbg !25 + %171 = fmul float %139, %139, !dbg !25 + %172 = fmul float %140, %140, !dbg !25 + %173 = fmul float %141, %141, !dbg !25 + %174 = fmul float %142, %142, !dbg !25 + %175 = fmul float %143, %143, !dbg !25 + %176 = fmul float %144, %144, !dbg !25 + %177 = fadd float %145, %146, !dbg !26 + %178 = fadd float %147, %177, !dbg !26 + %179 = fadd float %148, %178, !dbg !26 + %180 = fadd float %149, %179, !dbg !26 + %181 = fadd float %150, %180, !dbg !26 + %182 = fadd float %151, %181, !dbg !26 + %183 = fadd float %152, %182, !dbg !26 + %184 = fadd float %153, %154, !dbg !26 + %185 = fadd float %155, %184, !dbg !26 + %186 = fadd float %156, %185, !dbg !26 + %187 = fadd float %157, %186, !dbg !26 + %188 = fadd float %158, %187, !dbg !26 + %189 = fadd float %159, %188, !dbg !26 + %190 = fadd float %160, %189, !dbg !26 + %191 = fadd float %161, %162, !dbg !26 + %192 = fadd float %163, %191, !dbg !26 + %193 = fadd float %164, %192, !dbg !26 + %194 = fadd float %165, %193, !dbg !26 + %195 = fadd float %166, %194, !dbg !26 + %196 = fadd float %167, %195, !dbg !26 + %197 = fadd float %168, %196, !dbg !26 + %198 = fadd float %169, %170, !dbg !26 + %199 = fadd float %171, %198, !dbg !26 + %200 = fadd float %172, %199, !dbg !26 + %201 = fadd float %173, %200, !dbg !26 + %202 = fadd float %174, %201, !dbg !26 + %203 = fadd float %175, %202, !dbg !26 + %204 = fadd float %176, %203, !dbg !26 + %205 = bitcast float %183 to i32, !dbg !29 + %206 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %205, i32 8, i32 31), !dbg !29 + %207 = bitcast i32 %206 to float, !dbg !29 + %208 = fadd float %183, %207, !dbg !26 + %209 = bitcast float %208 to i32, !dbg !29 + %210 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %209, i32 4, i32 31), !dbg !29 + %211 = bitcast i32 %210 to float, !dbg !29 + %212 = fadd float %208, %211, !dbg !26 + %213 = bitcast float %212 to i32, !dbg !29 + %214 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %213, i32 2, i32 31), !dbg !29 + %215 = bitcast i32 %214 to float, !dbg !29 + %216 = fadd float %212, %215, !dbg !26 + %217 = bitcast float %216 to i32, !dbg !29 + %218 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %217, i32 1, i32 31), !dbg !29 + %219 = bitcast i32 %218 to float, !dbg !29 + %220 = fadd float %216, %219, !dbg !26 + %221 = bitcast float %190 to i32, !dbg !29 + %222 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %221, i32 8, i32 31), !dbg !29 + %223 = bitcast i32 %222 to float, !dbg !29 + %224 = fadd float %190, %223, !dbg !26 + %225 = bitcast float %224 to i32, !dbg !29 + %226 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %225, i32 4, i32 31), !dbg !29 + %227 = bitcast i32 %226 to float, !dbg !29 + %228 = fadd float %224, %227, !dbg !26 + %229 = bitcast float %228 to i32, !dbg !29 + %230 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %229, i32 2, i32 31), !dbg !29 + %231 = bitcast i32 %230 to float, !dbg !29 + %232 = fadd float %228, %231, !dbg !26 + %233 = bitcast float %232 to i32, !dbg !29 + %234 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %233, i32 1, i32 31), !dbg !29 + %235 = bitcast i32 %234 to float, !dbg !29 + %236 = fadd float %232, %235, !dbg !26 + %237 = bitcast float %197 to i32, !dbg !29 + %238 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %237, i32 8, i32 31), !dbg !29 + %239 = bitcast i32 %238 to float, !dbg !29 + %240 = fadd float %197, %239, !dbg !26 + %241 = bitcast float %240 to i32, !dbg !29 + %242 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %241, i32 4, i32 31), !dbg !29 + %243 = bitcast i32 %242 to float, !dbg !29 + %244 = fadd float %240, %243, !dbg !26 + %245 = bitcast float %244 to i32, !dbg !29 + %246 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %245, i32 2, i32 31), !dbg !29 + %247 = bitcast i32 %246 to float, !dbg !29 + %248 = fadd float %244, %247, !dbg !26 + %249 = bitcast float %248 to i32, !dbg !29 + %250 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %249, i32 1, i32 31), !dbg !29 + %251 = bitcast i32 %250 to float, !dbg !29 + %252 = fadd float %248, %251, !dbg !26 + %253 = bitcast float %204 to i32, !dbg !29 + %254 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %253, i32 8, i32 31), !dbg !29 + %255 = bitcast i32 %254 to float, !dbg !29 + %256 = fadd float %204, %255, !dbg !26 + %257 = bitcast float %256 to i32, !dbg !29 + %258 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %257, i32 4, i32 31), !dbg !29 + %259 = bitcast i32 %258 to float, !dbg !29 + %260 = fadd float %256, %259, !dbg !26 + %261 = bitcast float %260 to i32, !dbg !29 + %262 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %261, i32 2, i32 31), !dbg !29 + %263 = bitcast i32 %262 to float, !dbg !29 + %264 = fadd float %260, %263, !dbg !26 + %265 = bitcast float %264 to i32, !dbg !29 + %266 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %265, i32 1, i32 31), !dbg !29 + %267 = bitcast i32 %266 to float, !dbg !29 + %268 = fadd float %264, %267, !dbg !26 + %269 = fadd float %3, %220, !dbg !32 + %270 = fadd float %3, %236, !dbg !32 + %271 = fadd float %3, %252, !dbg !32 + %272 = fadd float %3, %268, !dbg !32 + %273 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i = icmp eq i32 %273, 0, !dbg !33 + %274 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i = icmp eq i32 %274, 0, !dbg !33 + br i1 %.not.i, label %280, label %275, !dbg !33 + +275: ; preds = %7 + br i1 %.not1.i, label %278, label %276, !dbg !33 + +276: ; preds = %275 + %277 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %269) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +278: ; preds = %275 + %279 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %269) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +280: ; preds = %7 + br i1 %.not1.i, label %283, label %281, !dbg !33 + +281: ; preds = %280 + %282 = tail call float @llvm.nvvm.sqrt.rn.f(float %269) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +283: ; preds = %280 + %284 = tail call float @llvm.nvvm.sqrt.approx.f(float %269) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +__nv_sqrtf.exit: ; preds = %276, %278, %281, %283 + %.0.i = phi float [ %277, %276 ], [ %279, %278 ], [ %282, %281 ], [ %284, %283 ], !dbg !33 + %285 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i1 = icmp eq i32 %285, 0, !dbg !33 + %286 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i4 = icmp eq i32 %286, 0, !dbg !33 + br i1 %.not.i1, label %292, label %287, !dbg !33 + +287: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %290, label %288, !dbg !33 + +288: ; preds = %287 + %289 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %270) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +290: ; preds = %287 + %291 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %270) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +292: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %295, label %293, !dbg !33 + +293: ; preds = %292 + %294 = tail call float @llvm.nvvm.sqrt.rn.f(float %270) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +295: ; preds = %292 + %296 = tail call float @llvm.nvvm.sqrt.approx.f(float %270) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +__nv_sqrtf.exit5: ; preds = %288, %290, %293, %295 + %.0.i3 = phi float [ %289, %288 ], [ %291, %290 ], [ %294, %293 ], [ %296, %295 ], !dbg !33 + %297 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i6 = icmp eq i32 %297, 0, !dbg !33 + %298 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i9 = icmp eq i32 %298, 0, !dbg !33 + br i1 %.not.i6, label %304, label %299, !dbg !33 + +299: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %302, label %300, !dbg !33 + +300: ; preds = %299 + %301 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %271) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +302: ; preds = %299 + %303 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %271) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +304: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %307, label %305, !dbg !33 + +305: ; preds = %304 + %306 = tail call float @llvm.nvvm.sqrt.rn.f(float %271) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +307: ; preds = %304 + %308 = tail call float @llvm.nvvm.sqrt.approx.f(float %271) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +__nv_sqrtf.exit10: ; preds = %300, %302, %305, %307 + %.0.i8 = phi float [ %301, %300 ], [ %303, %302 ], [ %306, %305 ], [ %308, %307 ], !dbg !33 + %309 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i11 = icmp eq i32 %309, 0, !dbg !33 + %310 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i14 = icmp eq i32 %310, 0, !dbg !33 + br i1 %.not.i11, label %316, label %311, !dbg !33 + +311: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %314, label %312, !dbg !33 + +312: ; preds = %311 + %313 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %272) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +314: ; preds = %311 + %315 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %272) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +316: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %319, label %317, !dbg !33 + +317: ; preds = %316 + %318 = tail call float @llvm.nvvm.sqrt.rn.f(float %272) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +319: ; preds = %316 + %320 = tail call float @llvm.nvvm.sqrt.approx.f(float %272) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +__nv_sqrtf.exit15: ; preds = %312, %314, %317, %319 + %.0.i13 = phi float [ %313, %312 ], [ %315, %314 ], [ %318, %317 ], [ %320, %319 ], !dbg !33 + %321 = zext nneg i32 %17 to i64, !dbg !17 + %322 = or disjoint i64 %10, %321, !dbg !17 + %323 = getelementptr float, ptr addrspace(1) %2, i64 %322, !dbg !34 + %324 = getelementptr bfloat, ptr addrspace(1) %1, i64 %35, !dbg !35 + %325 = getelementptr bfloat, ptr addrspace(1) %324, i64 %40, !dbg !36 + %326 = getelementptr bfloat, ptr addrspace(1) %1, i64 %34, !dbg !35 + %327 = getelementptr bfloat, ptr addrspace(1) %326, i64 %40, !dbg !36 + %328 = getelementptr bfloat, ptr addrspace(1) %1, i64 %33, !dbg !35 + %329 = getelementptr bfloat, ptr addrspace(1) %328, i64 %40, !dbg !36 + %330 = getelementptr bfloat, ptr addrspace(1) %1, i64 %32, !dbg !35 + %331 = getelementptr bfloat, ptr addrspace(1) %330, i64 %40, !dbg !36 + %332 = icmp slt i64 %322, %26, !dbg !18 + %333 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i), !dbg !37 + %334 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i3), !dbg !37 + %335 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i8), !dbg !37 + %336 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i13), !dbg !37 + %337 = fmul float %333, %113, !dbg !38 + %338 = fmul float %333, %114, !dbg !38 + %339 = fmul float %333, %115, !dbg !38 + %340 = fmul float %333, %116, !dbg !38 + %341 = fmul float %333, %117, !dbg !38 + %342 = fmul float %333, %118, !dbg !38 + %343 = fmul float %333, %119, !dbg !38 + %344 = fmul float %333, %120, !dbg !38 + %345 = fmul float %334, %121, !dbg !38 + %346 = fmul float %334, %122, !dbg !38 + %347 = fmul float %334, %123, !dbg !38 + %348 = fmul float %334, %124, !dbg !38 + %349 = fmul float %334, %125, !dbg !38 + %350 = fmul float %334, %126, !dbg !38 + %351 = fmul float %334, %127, !dbg !38 + %352 = fmul float %334, %128, !dbg !38 + %353 = fmul float %335, %129, !dbg !38 + %354 = fmul float %335, %130, !dbg !38 + %355 = fmul float %335, %131, !dbg !38 + %356 = fmul float %335, %132, !dbg !38 + %357 = fmul float %335, %133, !dbg !38 + %358 = fmul float %335, %134, !dbg !38 + %359 = fmul float %335, %135, !dbg !38 + %360 = fmul float %335, %136, !dbg !38 + %361 = fmul float %336, %137, !dbg !38 + %362 = fmul float %336, %138, !dbg !38 + %363 = fmul float %336, %139, !dbg !38 + %364 = fmul float %336, %140, !dbg !38 + %365 = fmul float %336, %141, !dbg !38 + %366 = fmul float %336, %142, !dbg !38 + %367 = fmul float %336, %143, !dbg !38 + %368 = fmul float %336, %144, !dbg !38 + %369 = fptrunc float %337 to bfloat, !dbg !39 + %370 = fptrunc float %338 to bfloat, !dbg !39 + %371 = fptrunc float %339 to bfloat, !dbg !39 + %372 = fptrunc float %340 to bfloat, !dbg !39 + %373 = fptrunc float %341 to bfloat, !dbg !39 + %374 = fptrunc float %342 to bfloat, !dbg !39 + %375 = fptrunc float %343 to bfloat, !dbg !39 + %376 = fptrunc float %344 to bfloat, !dbg !39 + %377 = fptrunc float %345 to bfloat, !dbg !39 + %378 = fptrunc float %346 to bfloat, !dbg !39 + %379 = fptrunc float %347 to bfloat, !dbg !39 + %380 = fptrunc float %348 to bfloat, !dbg !39 + %381 = fptrunc float %349 to bfloat, !dbg !39 + %382 = fptrunc float %350 to bfloat, !dbg !39 + %383 = fptrunc float %351 to bfloat, !dbg !39 + %384 = fptrunc float %352 to bfloat, !dbg !39 + %385 = fptrunc float %353 to bfloat, !dbg !39 + %386 = fptrunc float %354 to bfloat, !dbg !39 + %387 = fptrunc float %355 to bfloat, !dbg !39 + %388 = fptrunc float %356 to bfloat, !dbg !39 + %389 = fptrunc float %357 to bfloat, !dbg !39 + %390 = fptrunc float %358 to bfloat, !dbg !39 + %391 = fptrunc float %359 to bfloat, !dbg !39 + %392 = fptrunc float %360 to bfloat, !dbg !39 + %393 = fptrunc float %361 to bfloat, !dbg !39 + %394 = fptrunc float %362 to bfloat, !dbg !39 + %395 = fptrunc float %363 to bfloat, !dbg !39 + %396 = fptrunc float %364 to bfloat, !dbg !39 + %397 = fptrunc float %365 to bfloat, !dbg !39 + %398 = fptrunc float %366 to bfloat, !dbg !39 + %399 = fptrunc float %367 to bfloat, !dbg !39 + %400 = fptrunc float %368 to bfloat, !dbg !39 + %401 = insertelement <2 x bfloat> poison, bfloat %369, i64 0, !dbg !40 + %402 = insertelement <2 x bfloat> %401, bfloat %370, i64 1, !dbg !40 + %403 = bitcast <2 x bfloat> %402 to i32, !dbg !40 + %404 = insertelement <2 x bfloat> poison, bfloat %371, i64 0, !dbg !40 + %405 = insertelement <2 x bfloat> %404, bfloat %372, i64 1, !dbg !40 + %406 = bitcast <2 x bfloat> %405 to i32, !dbg !40 + %407 = insertelement <2 x bfloat> poison, bfloat %373, i64 0, !dbg !40 + %408 = insertelement <2 x bfloat> %407, bfloat %374, i64 1, !dbg !40 + %409 = bitcast <2 x bfloat> %408 to i32, !dbg !40 + %410 = insertelement <2 x bfloat> poison, bfloat %375, i64 0, !dbg !40 + %411 = insertelement <2 x bfloat> %410, bfloat %376, i64 1, !dbg !40 + %412 = bitcast <2 x bfloat> %411 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %403, i32 %406, i32 %409, i32 %412, ptr addrspace(1) %331, i1 %27) #6, !dbg !40 + %413 = insertelement <2 x bfloat> poison, bfloat %377, i64 0, !dbg !40 + %414 = insertelement <2 x bfloat> %413, bfloat %378, i64 1, !dbg !40 + %415 = bitcast <2 x bfloat> %414 to i32, !dbg !40 + %416 = insertelement <2 x bfloat> poison, bfloat %379, i64 0, !dbg !40 + %417 = insertelement <2 x bfloat> %416, bfloat %380, i64 1, !dbg !40 + %418 = bitcast <2 x bfloat> %417 to i32, !dbg !40 + %419 = insertelement <2 x bfloat> poison, bfloat %381, i64 0, !dbg !40 + %420 = insertelement <2 x bfloat> %419, bfloat %382, i64 1, !dbg !40 + %421 = bitcast <2 x bfloat> %420 to i32, !dbg !40 + %422 = insertelement <2 x bfloat> poison, bfloat %383, i64 0, !dbg !40 + %423 = insertelement <2 x bfloat> %422, bfloat %384, i64 1, !dbg !40 + %424 = bitcast <2 x bfloat> %423 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %415, i32 %418, i32 %421, i32 %424, ptr addrspace(1) %329, i1 %28) #6, !dbg !40 + %425 = insertelement <2 x bfloat> poison, bfloat %385, i64 0, !dbg !40 + %426 = insertelement <2 x bfloat> %425, bfloat %386, i64 1, !dbg !40 + %427 = bitcast <2 x bfloat> %426 to i32, !dbg !40 + %428 = insertelement <2 x bfloat> poison, bfloat %387, i64 0, !dbg !40 + %429 = insertelement <2 x bfloat> %428, bfloat %388, i64 1, !dbg !40 + %430 = bitcast <2 x bfloat> %429 to i32, !dbg !40 + %431 = insertelement <2 x bfloat> poison, bfloat %389, i64 0, !dbg !40 + %432 = insertelement <2 x bfloat> %431, bfloat %390, i64 1, !dbg !40 + %433 = bitcast <2 x bfloat> %432 to i32, !dbg !40 + %434 = insertelement <2 x bfloat> poison, bfloat %391, i64 0, !dbg !40 + %435 = insertelement <2 x bfloat> %434, bfloat %392, i64 1, !dbg !40 + %436 = bitcast <2 x bfloat> %435 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %427, i32 %430, i32 %433, i32 %436, ptr addrspace(1) %327, i1 %29) #6, !dbg !40 + %437 = insertelement <2 x bfloat> poison, bfloat %393, i64 0, !dbg !40 + %438 = insertelement <2 x bfloat> %437, bfloat %394, i64 1, !dbg !40 + %439 = bitcast <2 x bfloat> %438 to i32, !dbg !40 + %440 = insertelement <2 x bfloat> poison, bfloat %395, i64 0, !dbg !40 + %441 = insertelement <2 x bfloat> %440, bfloat %396, i64 1, !dbg !40 + %442 = bitcast <2 x bfloat> %441 to i32, !dbg !40 + %443 = insertelement <2 x bfloat> poison, bfloat %397, i64 0, !dbg !40 + %444 = insertelement <2 x bfloat> %443, bfloat %398, i64 1, !dbg !40 + %445 = bitcast <2 x bfloat> %444 to i32, !dbg !40 + %446 = insertelement <2 x bfloat> poison, bfloat %399, i64 0, !dbg !40 + %447 = insertelement <2 x bfloat> %446, bfloat %400, i64 1, !dbg !40 + %448 = bitcast <2 x bfloat> %447 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %439, i32 %442, i32 %445, i32 %448, ptr addrspace(1) %325, i1 %30) #6, !dbg !40 + %449 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %12, !dbg !41 + %450 = bitcast float %333 to i32, !dbg !41 + %451 = bitcast float %334 to i32, !dbg !41 + %452 = bitcast float %335 to i32, !dbg !41 + %453 = bitcast float %336 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %449, i32 %450, i32 %451, i32 %452, i32 %453) #6, !dbg !41 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !41 + %454 = shl nuw nsw i32 %11, 4, !dbg !41 + %455 = and i32 %454, 48, !dbg !41 + %456 = and i32 %11, 12, !dbg !41 + %457 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %455, !dbg !41 + %458 = getelementptr inbounds nuw i8, ptr addrspace(3) %457, i32 %456, !dbg !41 + %459 = load i32, ptr addrspace(3) %458, align 4, !dbg !41 + %460 = icmp eq i32 %12, 0, !dbg !41 + %461 = and i1 %460, %332, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %459, ptr addrspace(1) %323, i1 %461) #6, !dbg !41 + ret void, !dbg !42 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.div.full(float, float) #3 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #4 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #5 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.f(float) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="64" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) } +attributes #4 = { convergent nocallback nounwind } +attributes #5 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #6 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "l2norm_fwd_kernel", linkageName: "l2norm_fwd_kernel", scope: !1, file: !1, line: 69, type: !6, scopeLine: 69, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !10, !11, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 70, column: 24, scope: !5) +!14 = !DILocation(line: 70, column: 30, scope: !5) +!15 = !DILocation(line: 71, column: 16, scope: !5) +!16 = !DILocation(line: 71, column: 34, scope: !5) +!17 = !DILocation(line: 71, column: 21, scope: !5) +!18 = !DILocation(line: 73, column: 16, scope: !5) +!19 = !DILocation(line: 74, column: 30, scope: !5) +!20 = !DILocation(line: 75, column: 29, scope: !5) +!21 = !DILocation(line: 75, column: 14, scope: !5) +!22 = !DILocation(line: 75, column: 33, scope: !5) +!23 = !DILocation(line: 79, column: 18, scope: !5) +!24 = !DILocation(line: 79, column: 47, scope: !5) +!25 = !DILocation(line: 80, column: 38, scope: !5) +!26 = !DILocation(line: 263, column: 15, scope: !27, inlinedAt: !29) +!27 = distinct !DILexicalBlockFile(scope: !5, file: !28, discriminator: 0) +!28 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!29 = !DILocation(line: 293, column: 36, scope: !27, inlinedAt: !30) +!30 = !DILocation(line: 80, column: 43, scope: !31) +!31 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!32 = !DILocation(line: 80, column: 48, scope: !5) +!33 = !DILocation(line: 80, column: 25, scope: !5) +!34 = !DILocation(line: 77, column: 20, scope: !5) +!35 = !DILocation(line: 76, column: 14, scope: !5) +!36 = !DILocation(line: 76, column: 33, scope: !5) +!37 = !DILocation(line: 80, column: 17, scope: !5) +!38 = !DILocation(line: 81, column: 16, scope: !5) +!39 = !DILocation(line: 83, column: 25, scope: !5) +!40 = !DILocation(line: 83, column: 18, scope: !5) +!41 = !DILocation(line: 84, column: 21, scope: !5) +!42 = !DILocation(line: 84, column: 4, scope: !5) diff --git a/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.ptx b/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..7a390d646d0671a8fe4ddf2bf7e2b0d27e516f39 --- /dev/null +++ b/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.ptx @@ -0,0 +1,638 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_fwd_kernel // -- Begin function l2norm_fwd_kernel +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; +.global .align 1 .b8 _$_str_$_2[17] = {95, 95, 67, 85, 68, 65, 95, 80, 82, 69, 67, 95, 83, 81, 82, 84}; + // @l2norm_fwd_kernel +.visible .entry l2norm_fwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_2, + .param .f32 l2norm_fwd_kernel_param_3, + .param .u32 l2norm_fwd_kernel_param_4, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_6 +) +.reqntid 64 +{ + .reg .pred %p<8>; + .reg .b16 %rs<33>; + .reg .b32 %r<193>; + .reg .b64 %rd<39>; + .loc 1 69 0 // l2norm.py:69:0 +$L__func_begin0: + .loc 1 69 0 // l2norm.py:69:0 + +// %bb.0: // %__nv_sqrtf.exit + ld.param.b64 %rd10, [l2norm_fwd_kernel_param_0]; + ld.param.b64 %rd11, [l2norm_fwd_kernel_param_1]; +$L__tmp0: + .loc 1 70 24 // l2norm.py:70:24 + mov.u32 %r40, %ctaid.x; + ld.param.b64 %rd12, [l2norm_fwd_kernel_param_2]; + .loc 1 71 16 // l2norm.py:71:16 + mul.wide.u32 %rd13, %r40, 16; + ld.param.b32 %r41, [l2norm_fwd_kernel_param_3]; + ld.param.s32 %rd14, [l2norm_fwd_kernel_param_4]; + .loc 1 71 34 // l2norm.py:71:34 + mov.u32 %r42, %tid.x; + and.b32 %r43, %r42, 48; + bfe.u32 %r44, %r42, 4, 2; + or.b32 %r45, %r44, 4; + or.b32 %r46, %r44, 8; + or.b32 %r47, %r44, 12; + and.b32 %r48, %r42, 15; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd15, %r44; + cvt.u64.u32 %rd16, %r45; + cvt.u64.u32 %rd17, %r46; + cvt.u64.u32 %rd18, %r47; + or.b64 %rd19, %rd13, %rd15; + or.b64 %rd20, %rd13, %rd16; + or.b64 %rd21, %rd13, %rd17; + or.b64 %rd22, %rd13, %rd18; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p1, %rd19, %rd14; + setp.lt.s64 %p2, %rd20, %rd14; + setp.lt.s64 %p3, %rd21, %rd14; + setp.lt.s64 %p4, %rd22, %rd14; + .loc 1 74 30 // l2norm.py:74:30 + shl.b32 %r49, %r48, 3; + .loc 1 75 14 // l2norm.py:75:14 + shl.b64 %rd23, %rd19, 8; + add.s64 %rd24, %rd10, %rd23; + shl.b64 %rd25, %rd20, 8; + add.s64 %rd26, %rd10, %rd25; + shl.b64 %rd27, %rd21, 8; + add.s64 %rd28, %rd10, %rd27; + shl.b64 %rd29, %rd22, 8; + add.s64 %rd30, %rd10, %rd29; + .loc 1 75 33 // l2norm.py:75:33 + mul.wide.u32 %rd31, %r49, 2; + add.s64 %rd1, %rd24, %rd31; + add.s64 %rd2, %rd26, %rd31; + add.s64 %rd3, %rd28, %rd31; + add.s64 %rd4, %rd30, %rd31; + mov.b32 %r5, 0; + .loc 1 79 18 // l2norm.py:79:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + mov.b32 {%rs5, %rs6}, %r3; + mov.b32 {%rs7, %rs8}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs9, %rs10}, %r6; + mov.b32 {%rs11, %rs12}, %r7; + mov.b32 {%rs13, %rs14}, %r8; + mov.b32 {%rs15, %rs16}, %r9; + // begin inline asm + mov.u32 %r10, %r5; + mov.u32 %r11, %r5; + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + @%p3 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs17, %rs18}, %r10; + mov.b32 {%rs19, %rs20}, %r11; + mov.b32 {%rs21, %rs22}, %r12; + mov.b32 {%rs23, %rs24}, %r13; + // begin inline asm + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + @%p4 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ]; + // end inline asm + mov.b32 {%rs25, %rs26}, %r14; + mov.b32 {%rs27, %rs28}, %r15; + mov.b32 {%rs29, %rs30}, %r16; + mov.b32 {%rs31, %rs32}, %r17; + .loc 1 79 47 // l2norm.py:79:47 + cvt.f32.bf16 %r50, %rs1; + cvt.f32.bf16 %r51, %rs2; + cvt.f32.bf16 %r52, %rs3; + cvt.f32.bf16 %r53, %rs4; + cvt.f32.bf16 %r54, %rs5; + cvt.f32.bf16 %r55, %rs6; + cvt.f32.bf16 %r56, %rs7; + cvt.f32.bf16 %r57, %rs8; + cvt.f32.bf16 %r58, %rs9; + cvt.f32.bf16 %r59, %rs10; + cvt.f32.bf16 %r60, %rs11; + cvt.f32.bf16 %r61, %rs12; + cvt.f32.bf16 %r62, %rs13; + cvt.f32.bf16 %r63, %rs14; + cvt.f32.bf16 %r64, %rs15; + cvt.f32.bf16 %r65, %rs16; + cvt.f32.bf16 %r66, %rs17; + cvt.f32.bf16 %r67, %rs18; + cvt.f32.bf16 %r68, %rs19; + cvt.f32.bf16 %r69, %rs20; + cvt.f32.bf16 %r70, %rs21; + cvt.f32.bf16 %r71, %rs22; + cvt.f32.bf16 %r72, %rs23; + cvt.f32.bf16 %r73, %rs24; + cvt.f32.bf16 %r74, %rs25; + cvt.f32.bf16 %r75, %rs26; + cvt.f32.bf16 %r76, %rs27; + cvt.f32.bf16 %r77, %rs28; + cvt.f32.bf16 %r78, %rs29; + cvt.f32.bf16 %r79, %rs30; + cvt.f32.bf16 %r80, %rs31; + cvt.f32.bf16 %r81, %rs32; + .loc 1 80 38 // l2norm.py:80:38 + mul.f32 %r82, %r51, %r51; + mul.f32 %r83, %r59, %r59; + mul.f32 %r84, %r67, %r67; + mul.f32 %r85, %r75, %r75; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + fma.rn.f32.bf16 %r86, %rs1, %rs1, %r82; + fma.rn.f32.bf16 %r87, %rs3, %rs3, %r86; + fma.rn.f32.bf16 %r88, %rs4, %rs4, %r87; + fma.rn.f32.bf16 %r89, %rs5, %rs5, %r88; + fma.rn.f32.bf16 %r90, %rs6, %rs6, %r89; + fma.rn.f32.bf16 %r91, %rs7, %rs7, %r90; + fma.rn.f32.bf16 %r92, %rs8, %rs8, %r91; + fma.rn.f32.bf16 %r93, %rs9, %rs9, %r83; + fma.rn.f32.bf16 %r94, %rs11, %rs11, %r93; + fma.rn.f32.bf16 %r95, %rs12, %rs12, %r94; + fma.rn.f32.bf16 %r96, %rs13, %rs13, %r95; + fma.rn.f32.bf16 %r97, %rs14, %rs14, %r96; + fma.rn.f32.bf16 %r98, %rs15, %rs15, %r97; + fma.rn.f32.bf16 %r99, %rs16, %rs16, %r98; + fma.rn.f32.bf16 %r100, %rs17, %rs17, %r84; + fma.rn.f32.bf16 %r101, %rs19, %rs19, %r100; + fma.rn.f32.bf16 %r102, %rs20, %rs20, %r101; + fma.rn.f32.bf16 %r103, %rs21, %rs21, %r102; + fma.rn.f32.bf16 %r104, %rs22, %rs22, %r103; + fma.rn.f32.bf16 %r105, %rs23, %rs23, %r104; + fma.rn.f32.bf16 %r106, %rs24, %rs24, %r105; + fma.rn.f32.bf16 %r107, %rs25, %rs25, %r85; + fma.rn.f32.bf16 %r108, %rs27, %rs27, %r107; + fma.rn.f32.bf16 %r109, %rs28, %rs28, %r108; + fma.rn.f32.bf16 %r110, %rs29, %rs29, %r109; + fma.rn.f32.bf16 %r111, %rs30, %rs30, %r110; + fma.rn.f32.bf16 %r112, %rs31, %rs31, %r111; + fma.rn.f32.bf16 %r113, %rs32, %rs32, %r112; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r114, %r92, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r115, %r92, %r114; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r116, %r115, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r117, %r115, %r116; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r118, %r117, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r119, %r117, %r118; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r120, %r119, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r121, %r119, %r120; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r122, %r99, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r123, %r99, %r122; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r124, %r123, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r125, %r123, %r124; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r126, %r125, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r127, %r125, %r126; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r128, %r127, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r129, %r127, %r128; +$L__tmp18: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r130, %r106, 8, 31, -1; +$L__tmp19: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r131, %r106, %r130; +$L__tmp20: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r132, %r131, 4, 31, -1; +$L__tmp21: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r133, %r131, %r132; +$L__tmp22: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r134, %r133, 2, 31, -1; +$L__tmp23: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r135, %r133, %r134; +$L__tmp24: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r136, %r135, 1, 31, -1; +$L__tmp25: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r137, %r135, %r136; +$L__tmp26: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r138, %r113, 8, 31, -1; +$L__tmp27: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r139, %r113, %r138; +$L__tmp28: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r140, %r139, 4, 31, -1; +$L__tmp29: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r141, %r139, %r140; +$L__tmp30: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r142, %r141, 2, 31, -1; +$L__tmp31: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r143, %r141, %r142; +$L__tmp32: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r144, %r143, 1, 31, -1; +$L__tmp33: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r145, %r143, %r144; +$L__tmp34: + .loc 1 80 48 // l2norm.py:80:48 + add.f32 %r146, %r41, %r121; + add.f32 %r147, %r41, %r129; + add.f32 %r148, %r41, %r137; + add.f32 %r149, %r41, %r145; + .loc 1 80 25 // l2norm.py:80:25 + sqrt.approx.ftz.f32 %r150, %r146; + sqrt.approx.ftz.f32 %r151, %r147; + sqrt.approx.ftz.f32 %r152, %r148; + sqrt.approx.ftz.f32 %r153, %r149; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd32, %r48; + or.b64 %rd33, %rd13, %rd32; + .loc 1 77 20 // l2norm.py:77:20 + shl.b64 %rd34, %rd33, 2; + add.s64 %rd9, %rd12, %rd34; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd35, %rd11, %rd29; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd8, %rd35, %rd31; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd36, %rd11, %rd27; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd7, %rd36, %rd31; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd37, %rd11, %rd25; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd6, %rd37, %rd31; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd38, %rd11, %rd23; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd5, %rd38, %rd31; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p6, %rd33, %rd14; + mov.b32 %r154, 0f3F800000; + .loc 1 80 17 // l2norm.py:80:17 + div.full.f32 %r35, %r154, %r150; + div.full.f32 %r36, %r154, %r151; + div.full.f32 %r37, %r154, %r152; + div.full.f32 %r38, %r154, %r153; + .loc 1 81 16 // l2norm.py:81:16 + mul.f32 %r155, %r35, %r50; + mul.f32 %r156, %r35, %r51; + mul.f32 %r157, %r35, %r52; + mul.f32 %r158, %r35, %r53; + mul.f32 %r159, %r35, %r54; + mul.f32 %r160, %r35, %r55; + mul.f32 %r161, %r35, %r56; + mul.f32 %r162, %r35, %r57; + mul.f32 %r163, %r36, %r58; + mul.f32 %r164, %r36, %r59; + mul.f32 %r165, %r36, %r60; + mul.f32 %r166, %r36, %r61; + mul.f32 %r167, %r36, %r62; + mul.f32 %r168, %r36, %r63; + mul.f32 %r169, %r36, %r64; + mul.f32 %r170, %r36, %r65; + mul.f32 %r171, %r37, %r66; + mul.f32 %r172, %r37, %r67; + mul.f32 %r173, %r37, %r68; + mul.f32 %r174, %r37, %r69; + mul.f32 %r175, %r37, %r70; + mul.f32 %r176, %r37, %r71; + mul.f32 %r177, %r37, %r72; + mul.f32 %r178, %r37, %r73; + mul.f32 %r179, %r38, %r74; + mul.f32 %r180, %r38, %r75; + mul.f32 %r181, %r38, %r76; + mul.f32 %r182, %r38, %r77; + mul.f32 %r183, %r38, %r78; + mul.f32 %r184, %r38, %r79; + mul.f32 %r185, %r38, %r80; + mul.f32 %r186, %r38, %r81; + .loc 1 83 18 // l2norm.py:83:18 + cvt.rn.bf16x2.f32 %r18, %r156, %r155; + cvt.rn.bf16x2.f32 %r19, %r158, %r157; + cvt.rn.bf16x2.f32 %r20, %r160, %r159; + cvt.rn.bf16x2.f32 %r21, %r162, %r161; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd5 + 0 ], { %r18, %r19, %r20, %r21 }; + // end inline asm + cvt.rn.bf16x2.f32 %r22, %r164, %r163; + cvt.rn.bf16x2.f32 %r23, %r166, %r165; + cvt.rn.bf16x2.f32 %r24, %r168, %r167; + cvt.rn.bf16x2.f32 %r25, %r170, %r169; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd6 + 0 ], { %r22, %r23, %r24, %r25 }; + // end inline asm + cvt.rn.bf16x2.f32 %r26, %r172, %r171; + cvt.rn.bf16x2.f32 %r27, %r174, %r173; + cvt.rn.bf16x2.f32 %r28, %r176, %r175; + cvt.rn.bf16x2.f32 %r29, %r178, %r177; + // begin inline asm + @%p3 st.global.v4.b32 [ %rd7 + 0 ], { %r26, %r27, %r28, %r29 }; + // end inline asm + cvt.rn.bf16x2.f32 %r30, %r180, %r179; + cvt.rn.bf16x2.f32 %r31, %r182, %r181; + cvt.rn.bf16x2.f32 %r32, %r184, %r183; + cvt.rn.bf16x2.f32 %r33, %r186, %r185; + // begin inline asm + @%p4 st.global.v4.b32 [ %rd8 + 0 ], { %r30, %r31, %r32, %r33 }; + // end inline asm + .loc 1 84 21 // l2norm.py:84:21 + mov.b32 %r187, global_smem; + add.s32 %r34, %r187, %r43; + // begin inline asm + st.shared.v4.b32 [ %r34 + 0 ], { %r35, %r36, %r37, %r38 }; + // end inline asm + bar.sync 0; + shl.b32 %r188, %r42, 4; + and.b32 %r189, %r188, 48; + and.b32 %r190, %r42, 12; + add.s32 %r191, %r187, %r189; + add.s32 %r192, %r191, %r190; + ld.shared.b32 %r39, [%r192]; + setp.eq.b32 %p7, %r43, 0; + and.pred %p5, %p7, %p6; + // begin inline asm + @%p5 st.global.b32 [ %rd9 + 0 ], { %r39 }; + // end inline asm + .loc 1 84 4 // l2norm.py:84:4 + ret; +$L__tmp35: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp34 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 80 // DW_AT_call_line +.b8 43 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp34 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.source b/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..897f77c236e43b6c12f4512723699931dc7dd6b5 --- /dev/null +++ b/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.source @@ -0,0 +1,172 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc38 = loc(unknown) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc45 = loc("x"(#loc)) +#loc46 = loc("y"(#loc)) +#loc47 = loc("rstd"(#loc)) +#loc48 = loc("eps"(#loc)) +#loc49 = loc("T"(#loc)) +#loc81 = loc("input"(#loc36)) +#loc82 = loc("a"(#loc41)) +#loc83 = loc("b"(#loc41)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc50) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc51) + %o_t = arith.constant 16 : i32 loc(#loc52) + %o_t_1 = arith.constant 16 : i64 loc(#loc52) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc52) + %o_t_3 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32> loc(#loc53) + %o_t_4 = arith.extsi %o_t_3 : tensor<16xi32> to tensor<16xi64> loc(#loc54) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<16xi64> loc(#loc54) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<16xi64> loc(#loc54) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc55) + %m_t = arith.extsi %T : i32 to i64 loc(#loc56) + %m_t_7 = tt.splat %m_t : i64 -> tensor<16xi64> loc(#loc56) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<16xi64> loc(#loc56) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc57) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc58) + %m_x_10 = arith.constant 128 : i32 loc(#loc59) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc59) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc59) + %m_x_13 = tt.broadcast %m_x : tensor<16x1xi1> -> tensor<16x128xi1> loc(#loc60) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<16x128xi1> loc(#loc60) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<16x128xi1> loc(#loc60) + %p_x = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc61) + %p_x_16 = arith.constant 128 : i32 loc(#loc62) + %p_x_17 = arith.constant 128 : i64 loc(#loc62) + %p_x_18 = arith.constant dense<128> : tensor<16x1xi64> loc(#loc62) + %p_x_19 = arith.muli %p_x, %p_x_18 : tensor<16x1xi64> loc(#loc62) + %p_x_20 = tt.splat %x : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc63) + %p_x_21 = tt.addptr %p_x_20, %p_x_19 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc63) + %p_x_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc64) + %p_x_23 = tt.broadcast %p_x_21 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc65) + %p_x_24 = tt.broadcast %p_x_22 : tensor<1x128xi32> -> tensor<16x128xi32> loc(#loc65) + %p_x_25 = tt.addptr %p_x_23, %p_x_24 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc65) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc66) + %p_y_26 = arith.constant 128 : i32 loc(#loc67) + %p_y_27 = arith.constant 128 : i64 loc(#loc67) + %p_y_28 = arith.constant dense<128> : tensor<16x1xi64> loc(#loc67) + %p_y_29 = arith.muli %p_y, %p_y_28 : tensor<16x1xi64> loc(#loc67) + %p_y_30 = tt.splat %y : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc68) + %p_y_31 = tt.addptr %p_y_30, %p_y_29 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc68) + %p_y_32 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc69) + %p_y_33 = tt.broadcast %p_y_31 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc70) + %p_y_34 = tt.broadcast %p_y_32 : tensor<1x128xi32> -> tensor<16x128xi32> loc(#loc70) + %p_y_35 = tt.addptr %p_y_33, %p_y_34 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc70) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc71) + %p_rstd_36 = tt.addptr %p_rstd, %o_t_6 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc71) + %b_x = arith.constant 0.000000e+00 : f32 loc(#loc72) + %b_x_37 = arith.constant dense<0.000000e+00> : tensor<16x128xf32> loc(#loc72) + %b_x_38 = arith.truncf %b_x_37 : tensor<16x128xf32> to tensor<16x128xbf16> loc(#loc72) + %b_x_39 = tt.load %p_x_25, %m_x_15, %b_x_38 : tensor<16x128x!tt.ptr> loc(#loc72) + %b_x_40 = arith.extf %b_x_39 : tensor<16x128xbf16> to tensor<16x128xf32> loc(#loc73) + %b_rstd = arith.mulf %b_x_40, %b_x_40 : tensor<16x128xf32> loc(#loc74) + %b_rstd_41 = tt.call @triton.language.standard.sum__fp32S16_128S_c1_cFalse_cNone(%b_rstd) : (tensor<16x128xf32>) -> tensor<16xf32> loc(#loc75) + %b_rstd_42 = tt.splat %eps : f32 -> tensor<16xf32> loc(#loc76) + %b_rstd_43 = arith.addf %b_rstd_41, %b_rstd_42 : tensor<16xf32> loc(#loc76) + %b_rstd_44 = math.sqrt %b_rstd_43 : tensor<16xf32> loc(#loc77) + %b_rstd_45 = arith.constant 1 : i32 loc(#loc78) + %b_rstd_46 = arith.constant 1.000000e+00 : f32 loc(#loc78) + %b_rstd_47 = arith.constant dense<1.000000e+00> : tensor<16xf32> loc(#loc78) + %b_rstd_48 = arith.divf %b_rstd_47, %b_rstd_44 : tensor<16xf32> loc(#loc78) + %b_y = tt.expand_dims %b_rstd_48 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc79) + %b_y_49 = tt.broadcast %b_y : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc80) + %b_y_50 = arith.mulf %b_x_40, %b_y_49 : tensor<16x128xf32> loc(#loc80) + %0 = arith.truncf %b_y_50 : tensor<16x128xf32> to tensor<16x128xbf16> loc(#loc32) + tt.store %p_y_35, %0, %m_x_15 : tensor<16x128x!tt.ptr> loc(#loc33) + tt.store %p_rstd_36, %b_rstd_48, %m_t_8 : tensor<16x!tt.ptr> loc(#loc34) + tt.return loc(#loc35) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S16_128S_c1_cFalse_cNone(%input: tensor<16x128xf32> loc("input"(#loc36))) -> tensor<16xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc37) + tt.reduce.return %2 : f32 loc(#loc37) + }) : (tensor<16x128xf32>) -> tensor<16xf32> loc(#loc37) + tt.return %0 : tensor<16xf32> loc(#loc39) + ^bb1: // no predecessors + %1 = ub.poison : tensor<16xf32> loc(#loc40) + tt.return %1 : tensor<16xf32> loc(#loc40) + } loc(#loc36) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc41)), %b: f32 loc("b"(#loc41))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc42) + tt.return %0 : f32 loc(#loc43) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc44) + tt.return %1 : f32 loc(#loc44) + } loc(#loc41) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:18) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:29) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:37) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc50 = loc("i_t"(#loc1)) +#loc51 = loc("i_t"(#loc2)) +#loc52 = loc("o_t"(#loc3)) +#loc53 = loc("o_t"(#loc4)) +#loc54 = loc("o_t"(#loc5)) +#loc55 = loc("o_d"(#loc6)) +#loc56 = loc("m_t"(#loc7)) +#loc57 = loc("m_x"(#loc8)) +#loc58 = loc("m_x"(#loc9)) +#loc59 = loc("m_x"(#loc10)) +#loc60 = loc("m_x"(#loc11)) +#loc61 = loc("p_x"(#loc12)) +#loc62 = loc("p_x"(#loc13)) +#loc63 = loc("p_x"(#loc14)) +#loc64 = loc("p_x"(#loc15)) +#loc65 = loc("p_x"(#loc16)) +#loc66 = loc("p_y"(#loc17)) +#loc67 = loc("p_y"(#loc18)) +#loc68 = loc("p_y"(#loc19)) +#loc69 = loc("p_y"(#loc20)) +#loc70 = loc("p_y"(#loc21)) +#loc71 = loc("p_rstd"(#loc22)) +#loc72 = loc("b_x"(#loc23)) +#loc73 = loc("b_x"(#loc24)) +#loc74 = loc("b_rstd"(#loc25)) +#loc75 = loc("b_rstd"(#loc26)) +#loc76 = loc("b_rstd"(#loc27)) +#loc77 = loc("b_rstd"(#loc28)) +#loc78 = loc("b_rstd"(#loc29)) +#loc79 = loc("b_y"(#loc30)) +#loc80 = loc("b_y"(#loc31)) diff --git a/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.ttgir b/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..30f930034894455e5eb41d8476c85b40e99f4bbd --- /dev/null +++ b/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.ttgir @@ -0,0 +1,135 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [2, 1], order = [1, 0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [2], order = [0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc1 = loc(unknown) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc34 = loc("x"(#loc)) +#loc35 = loc("y"(#loc)) +#loc36 = loc("rstd"(#loc)) +#loc37 = loc("eps"(#loc)) +#loc38 = loc("T"(#loc)) +#loc59 = loc("b_rstd"(#loc23)) +#loc66 = loc(callsite(#loc1 at #loc59)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 2 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<1.000000e+00> : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<16x128xbf16, #blocked> loc(#loc1) + %c16_i64 = arith.constant 16 : i64 loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<16x1xi64, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc39) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc40) + %o_t = arith.muli %i_t_3, %c16_i64 : i64 loc(#loc41) + %o_t_4 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc42) + %o_t_5 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32, #blocked1> loc(#loc42) + %o_t_6 = arith.extsi %o_t_4 : tensor<16xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_7 = arith.extsi %o_t_5 : tensor<16xi32, #blocked1> to tensor<16xi64, #blocked1> loc(#loc43) + %o_t_8 = tt.splat %o_t : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_9 = tt.splat %o_t : i64 -> tensor<16xi64, #blocked1> loc(#loc43) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<16xi64, #blocked1> loc(#loc43) + %m_t = arith.extsi %T : i32 to i64 loc(#loc44) + %m_t_12 = tt.splat %m_t : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_13 = tt.splat %m_t : i64 -> tensor<16xi64, #blocked1> loc(#loc44) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<16xi64, #blocked1> loc(#loc44) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi1, #blocked> loc(#loc45) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc46) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc46) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_1 : tensor<1x128xi32, #blocked> loc(#loc47) + %m_x_19 = tt.broadcast %m_x : tensor<16x1xi1, #blocked> -> tensor<16x128xi1, #blocked> loc(#loc48) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<16x128xi1, #blocked> loc(#loc48) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<16x128xi1, #blocked> loc(#loc48) + %p_x = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi64, #blocked> loc(#loc49) + %p_x_22 = arith.muli %p_x, %cst_2 : tensor<16x1xi64, #blocked> loc(#loc50) + %p_x_23 = tt.splat %x : !tt.ptr -> tensor<16x1x!tt.ptr, #blocked> loc(#loc51) + %p_x_24 = tt.addptr %p_x_23, %p_x_22 : tensor<16x1x!tt.ptr, #blocked>, tensor<16x1xi64, #blocked> loc(#loc51) + %p_x_25 = tt.broadcast %p_x_24 : tensor<16x1x!tt.ptr, #blocked> -> tensor<16x128x!tt.ptr, #blocked> loc(#loc52) + %p_x_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<16x128xi32, #blocked> loc(#loc52) + %p_x_27 = tt.addptr %p_x_25, %p_x_26 : tensor<16x128x!tt.ptr, #blocked>, tensor<16x128xi32, #blocked> loc(#loc52) + %p_y = tt.splat %y : !tt.ptr -> tensor<16x1x!tt.ptr, #blocked> loc(#loc53) + %p_y_28 = tt.addptr %p_y, %p_x_22 : tensor<16x1x!tt.ptr, #blocked>, tensor<16x1xi64, #blocked> loc(#loc53) + %p_y_29 = tt.broadcast %p_y_28 : tensor<16x1x!tt.ptr, #blocked> -> tensor<16x128x!tt.ptr, #blocked> loc(#loc54) + %p_y_30 = tt.addptr %p_y_29, %p_x_26 : tensor<16x128x!tt.ptr, #blocked>, tensor<16x128xi32, #blocked> loc(#loc54) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<16x!tt.ptr, #blocked1> loc(#loc55) + %p_rstd_31 = tt.addptr %p_rstd, %o_t_11 : tensor<16x!tt.ptr, #blocked1>, tensor<16xi64, #blocked1> loc(#loc55) + %b_x = tt.load %p_x_27, %m_x_21, %cst_0 : tensor<16x128x!tt.ptr, #blocked> loc(#loc56) + %b_x_32 = arith.extf %b_x : tensor<16x128xbf16, #blocked> to tensor<16x128xf32, #blocked> loc(#loc57) + %b_rstd = arith.mulf %b_x_32, %b_x_32 : tensor<16x128xf32, #blocked> loc(#loc58) + %b_rstd_33 = "tt.reduce"(%b_rstd) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_40: f32 loc(callsite(#loc1 at #loc59)), %b_rstd_41: f32 loc(callsite(#loc1 at #loc59))): + %b_rstd_42 = arith.addf %b_rstd_40, %b_rstd_41 : f32 loc(#loc67) + tt.reduce.return %b_rstd_42 : f32 loc(#loc65) + }) : (tensor<16x128xf32, #blocked>) -> tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc65) + %b_rstd_34 = tt.splat %eps : f32 -> tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_36 = math.sqrt %b_rstd_35 : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc61) + %b_rstd_37 = arith.divf %cst, %b_rstd_36 : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc62) + %b_y = tt.expand_dims %b_rstd_37 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xf32, #blocked> loc(#loc63) + %b_y_38 = tt.broadcast %b_y : tensor<16x1xf32, #blocked> -> tensor<16x128xf32, #blocked> loc(#loc64) + %b_y_39 = arith.mulf %b_x_32, %b_y_38 : tensor<16x128xf32, #blocked> loc(#loc64) + %0 = arith.truncf %b_y_39 : tensor<16x128xf32, #blocked> to tensor<16x128xbf16, #blocked> loc(#loc30) + tt.store %p_y_30, %0, %m_x_21 : tensor<16x128x!tt.ptr, #blocked> loc(#loc31) + %1 = ttg.convert_layout %b_rstd_37 : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16xf32, #blocked1> loc(#loc32) + tt.store %p_rstd_31, %1, %m_t_15 : tensor<16x!tt.ptr, #blocked1> loc(#loc32) + tt.return loc(#loc33) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc39 = loc("i_t"(#loc2)) +#loc40 = loc("i_t"(#loc3)) +#loc41 = loc("o_t"(#loc4)) +#loc42 = loc("o_t"(#loc5)) +#loc43 = loc("o_t"(#loc6)) +#loc44 = loc("m_t"(#loc7)) +#loc45 = loc("m_x"(#loc8)) +#loc46 = loc("m_x"(#loc9)) +#loc47 = loc("m_x"(#loc10)) +#loc48 = loc("m_x"(#loc11)) +#loc49 = loc("p_x"(#loc12)) +#loc50 = loc("p_x"(#loc13)) +#loc51 = loc("p_x"(#loc14)) +#loc52 = loc("p_x"(#loc15)) +#loc53 = loc("p_y"(#loc16)) +#loc54 = loc("p_y"(#loc17)) +#loc55 = loc("p_rstd"(#loc18)) +#loc56 = loc("b_x"(#loc19)) +#loc57 = loc("b_x"(#loc20)) +#loc58 = loc("b_rstd"(#loc21)) +#loc60 = loc("b_rstd"(#loc25)) +#loc61 = loc("b_rstd"(#loc26)) +#loc62 = loc("b_rstd"(#loc27)) +#loc63 = loc("b_y"(#loc28)) +#loc64 = loc("b_y"(#loc29)) +#loc65 = loc(callsite(#loc22 at #loc59)) +#loc67 = loc(callsite(#loc24 at #loc65)) diff --git a/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.ttir b/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..d7d00d4871c5899dd9e1884d5ca3c3a665bff374 --- /dev/null +++ b/triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.ttir @@ -0,0 +1,128 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc3 = loc(unknown) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc35 = loc("x"(#loc)) +#loc36 = loc("y"(#loc)) +#loc37 = loc("rstd"(#loc)) +#loc38 = loc("eps"(#loc)) +#loc39 = loc("T"(#loc)) +#loc62 = loc("b_rstd"(#loc25)) +#loc68 = loc(callsite(#loc3 at #loc62)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %b_x = arith.constant dense<0.000000e+00> : tensor<16x128xbf16> loc(#loc40) + %b_rstd = arith.constant dense<1.000000e+00> : tensor<16xf32> loc(#loc41) + %cst = arith.constant dense<128> : tensor<16x1xi64> loc(#loc3) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc42) + %o_t = arith.constant 16 : i64 loc(#loc43) + %i_t = tt.get_program_id x : i32 loc(#loc44) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc45) + %o_t_1 = arith.muli %i_t_0, %o_t : i64 loc(#loc43) + %o_t_2 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32> loc(#loc46) + %o_t_3 = arith.extsi %o_t_2 : tensor<16xi32> to tensor<16xi64> loc(#loc47) + %o_t_4 = tt.splat %o_t_1 : i64 -> tensor<16xi64> loc(#loc47) + %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<16xi64> loc(#loc47) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc48) + %m_t = arith.extsi %T : i32 to i64 loc(#loc49) + %m_t_6 = tt.splat %m_t : i64 -> tensor<16xi64> loc(#loc49) + %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<16xi64> loc(#loc49) + %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc50) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc51) + %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc42) + %m_x_11 = tt.broadcast %m_x_8 : tensor<16x1xi1> -> tensor<16x128xi1> loc(#loc52) + %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<16x128xi1> loc(#loc52) + %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<16x128xi1> loc(#loc52) + %p_x = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc53) + %p_x_14 = arith.muli %p_x, %cst : tensor<16x1xi64> loc(#loc54) + %p_x_15 = tt.splat %x : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc55) + %p_x_16 = tt.addptr %p_x_15, %p_x_14 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc55) + %p_x_17 = tt.broadcast %p_x_16 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc56) + %p_x_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<16x128xi32> loc(#loc56) + %p_x_19 = tt.addptr %p_x_17, %p_x_18 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc56) + %p_y = tt.splat %y : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc57) + %p_y_20 = tt.addptr %p_y, %p_x_14 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc57) + %p_y_21 = tt.broadcast %p_y_20 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc58) + %p_y_22 = tt.addptr %p_y_21, %p_x_18 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc59) + %p_rstd_23 = tt.addptr %p_rstd, %o_t_5 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc59) + %b_x_24 = tt.load %p_x_19, %m_x_13, %b_x : tensor<16x128x!tt.ptr> loc(#loc40) + %b_x_25 = arith.extf %b_x_24 : tensor<16x128xbf16> to tensor<16x128xf32> loc(#loc60) + %b_rstd_26 = arith.mulf %b_x_25, %b_x_25 : tensor<16x128xf32> loc(#loc61) + %b_rstd_27 = "tt.reduce"(%b_rstd_26) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_34: f32 loc(callsite(#loc3 at #loc62)), %b_rstd_35: f32 loc(callsite(#loc3 at #loc62))): + %b_rstd_36 = arith.addf %b_rstd_34, %b_rstd_35 : f32 loc(#loc69) + tt.reduce.return %b_rstd_36 : f32 loc(#loc67) + }) : (tensor<16x128xf32>) -> tensor<16xf32> loc(#loc67) + %b_rstd_28 = tt.splat %eps : f32 -> tensor<16xf32> loc(#loc63) + %b_rstd_29 = arith.addf %b_rstd_27, %b_rstd_28 : tensor<16xf32> loc(#loc63) + %b_rstd_30 = math.sqrt %b_rstd_29 : tensor<16xf32> loc(#loc64) + %b_rstd_31 = arith.divf %b_rstd, %b_rstd_30 : tensor<16xf32> loc(#loc41) + %b_y = tt.expand_dims %b_rstd_31 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc65) + %b_y_32 = tt.broadcast %b_y : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc66) + %b_y_33 = arith.mulf %b_x_25, %b_y_32 : tensor<16x128xf32> loc(#loc66) + %0 = arith.truncf %b_y_33 : tensor<16x128xf32> to tensor<16x128xbf16> loc(#loc31) + tt.store %p_y_22, %0, %m_x_13 : tensor<16x128x!tt.ptr> loc(#loc32) + tt.store %p_rstd_23, %b_rstd_31, %m_t_7 : tensor<16x!tt.ptr> loc(#loc33) + tt.return loc(#loc34) + } loc(#loc) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc40 = loc("b_x"(#loc1)) +#loc41 = loc("b_rstd"(#loc2)) +#loc42 = loc("m_x"(#loc4)) +#loc43 = loc("o_t"(#loc5)) +#loc44 = loc("i_t"(#loc6)) +#loc45 = loc("i_t"(#loc7)) +#loc46 = loc("o_t"(#loc8)) +#loc47 = loc("o_t"(#loc9)) +#loc48 = loc("o_d"(#loc10)) +#loc49 = loc("m_t"(#loc11)) +#loc50 = loc("m_x"(#loc12)) +#loc51 = loc("m_x"(#loc13)) +#loc52 = loc("m_x"(#loc14)) +#loc53 = loc("p_x"(#loc15)) +#loc54 = loc("p_x"(#loc16)) +#loc55 = loc("p_x"(#loc17)) +#loc56 = loc("p_x"(#loc18)) +#loc57 = loc("p_y"(#loc19)) +#loc58 = loc("p_y"(#loc20)) +#loc59 = loc("p_rstd"(#loc21)) +#loc60 = loc("b_x"(#loc22)) +#loc61 = loc("b_rstd"(#loc23)) +#loc63 = loc("b_rstd"(#loc27)) +#loc64 = loc("b_rstd"(#loc28)) +#loc65 = loc("b_y"(#loc29)) +#loc66 = loc("b_y"(#loc30)) +#loc67 = loc(callsite(#loc24 at #loc62)) +#loc69 = loc(callsite(#loc26 at #loc67)) diff --git a/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/__grp__l2norm_fwd_kernel.json b/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/__grp__l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..e180d9b3ad12785ce6ac95672daac1d154553b4b --- /dev/null +++ b/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/__grp__l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_fwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.source", "l2norm_fwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.ttir", "l2norm_fwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.ttgir", "l2norm_fwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.llir", "l2norm_fwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.ptx", "l2norm_fwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.cubin", "l2norm_fwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.cubin b/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..da161322c15171f8c62bc812d13ce80efa02139d --- /dev/null +++ b/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.cubin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:01f06397f0b30f222dafe7d0ff24bfb233bf0b57b9c1dc7883c733a1b6747267 +size 160768 diff --git a/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.json b/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..4ca564b324071193aff052f1ef64a09791e2547b --- /dev/null +++ b/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"hash": "0519090fa6f56020d1219a049b0ca11692d737f10874ce64023e475aaac58357", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 2, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 512, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_fwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.llir b/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..6b00d4f55664f17046aa8f1a8a24783f89b7a438 --- /dev/null +++ b/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.llir @@ -0,0 +1,4181 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 +@.str.2 = private unnamed_addr constant [17 x i8] c"__CUDA_PREC_SQRT\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_fwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, float %3, i32 %4, ptr addrspace(1) readnone captures(none) %5, ptr addrspace(1) readnone captures(none) %6) local_unnamed_addr #0 !dbg !5 { + %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %9 = zext nneg i32 %8 to i64, !dbg !14 + %10 = shl nuw nsw i64 %9, 7, !dbg !15 + %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !16 + %12 = lshr i32 %11, 4, !dbg !16 + %13 = and i32 %12, 3, !dbg !16 + %14 = or disjoint i32 %13, 4, !dbg !16 + %15 = insertelement <2 x i32> poison, i32 %13, i64 0, !dbg !16 + %16 = insertelement <4 x i32> poison, i32 %13, i64 0, !dbg !16 + %17 = insertelement <8 x i32> poison, i32 %12, i64 0, !dbg !16 + %18 = insertelement <8 x i32> %17, i32 %13, i64 1, !dbg !16 + %19 = insertelement <32 x i32> poison, i32 %14, i64 30, !dbg !17 + %20 = insertelement <32 x i32> %19, i32 %13, i64 31, !dbg !17 + %21 = shufflevector <8 x i32> %18, <8 x i32> poison, <32 x i32> , !dbg !17 + %22 = or <32 x i32> %21, , !dbg !17 + %23 = shufflevector <32 x i32> %22, <32 x i32> %20, <32 x i32> , !dbg !17 + %24 = shufflevector <8 x i32> %18, <8 x i32> poison, <32 x i32> , !dbg !17 + %25 = or <32 x i32> %24, , !dbg !17 + %26 = shufflevector <32 x i32> %23, <32 x i32> %25, <32 x i32> , !dbg !17 + %27 = shufflevector <4 x i32> %16, <4 x i32> poison, <32 x i32> , !dbg !17 + %28 = or disjoint <32 x i32> %27, , !dbg !17 + %29 = shufflevector <32 x i32> %26, <32 x i32> %28, <32 x i32> , !dbg !17 + %30 = shufflevector <2 x i32> %15, <2 x i32> poison, <32 x i32> , !dbg !17 + %31 = or disjoint <32 x i32> %30, , !dbg !17 + %32 = shufflevector <32 x i32> %29, <32 x i32> %31, <32 x i32> , !dbg !17 + %33 = zext <32 x i32> %32 to <32 x i64>, !dbg !17 + %34 = insertelement <32 x i64> poison, i64 %10, i64 0, !dbg !17 + %35 = shufflevector <32 x i64> %34, <32 x i64> poison, <32 x i32> zeroinitializer, !dbg !17 + %36 = or disjoint <32 x i64> %35, %33, !dbg !17 + %37 = insertelement <32 x i32> poison, i32 %4, i64 0, !dbg !18 + %38 = shufflevector <32 x i32> %37, <32 x i32> poison, <32 x i32> zeroinitializer, !dbg !18 + %39 = sext <32 x i32> %38 to <32 x i64>, !dbg !18 + %40 = icmp slt <32 x i64> %36, %39, !dbg !18 + %41 = shl nuw nsw i32 %11, 3, !dbg !19 + %42 = and i32 %41, 120, !dbg !19 + %43 = extractelement <32 x i64> %36, i64 31, !dbg !20 + %44 = shl nuw nsw i64 %43, 7, !dbg !20 + %45 = extractelement <32 x i64> %36, i64 30, !dbg !20 + %46 = shl nuw nsw i64 %45, 7, !dbg !20 + %47 = extractelement <32 x i64> %36, i64 29, !dbg !20 + %48 = shl nuw nsw i64 %47, 7, !dbg !20 + %49 = extractelement <32 x i64> %36, i64 28, !dbg !20 + %50 = shl nuw nsw i64 %49, 7, !dbg !20 + %51 = extractelement <32 x i64> %36, i64 27, !dbg !20 + %52 = shl nuw nsw i64 %51, 7, !dbg !20 + %53 = extractelement <32 x i64> %36, i64 26, !dbg !20 + %54 = shl nuw nsw i64 %53, 7, !dbg !20 + %55 = extractelement <32 x i64> %36, i64 25, !dbg !20 + %56 = shl nuw nsw i64 %55, 7, !dbg !20 + %57 = extractelement <32 x i64> %36, i64 24, !dbg !20 + %58 = shl nuw nsw i64 %57, 7, !dbg !20 + %59 = extractelement <32 x i64> %36, i64 23, !dbg !20 + %60 = shl nuw nsw i64 %59, 7, !dbg !20 + %61 = extractelement <32 x i64> %36, i64 22, !dbg !20 + %62 = shl nuw nsw i64 %61, 7, !dbg !20 + %63 = extractelement <32 x i64> %36, i64 21, !dbg !20 + %64 = shl nuw nsw i64 %63, 7, !dbg !20 + %65 = extractelement <32 x i64> %36, i64 20, !dbg !20 + %66 = shl nuw nsw i64 %65, 7, !dbg !20 + %67 = extractelement <32 x i64> %36, i64 19, !dbg !20 + %68 = shl nuw nsw i64 %67, 7, !dbg !20 + %69 = extractelement <32 x i64> %36, i64 18, !dbg !20 + %70 = shl nuw nsw i64 %69, 7, !dbg !20 + %71 = extractelement <32 x i64> %36, i64 17, !dbg !20 + %72 = shl nuw nsw i64 %71, 7, !dbg !20 + %73 = extractelement <32 x i64> %36, i64 16, !dbg !20 + %74 = shl nuw nsw i64 %73, 7, !dbg !20 + %75 = extractelement <32 x i64> %36, i64 15, !dbg !20 + %76 = shl nuw nsw i64 %75, 7, !dbg !20 + %77 = extractelement <32 x i64> %36, i64 14, !dbg !20 + %78 = shl nuw nsw i64 %77, 7, !dbg !20 + %79 = extractelement <32 x i64> %36, i64 13, !dbg !20 + %80 = shl nuw nsw i64 %79, 7, !dbg !20 + %81 = extractelement <32 x i64> %36, i64 12, !dbg !20 + %82 = shl nuw nsw i64 %81, 7, !dbg !20 + %83 = extractelement <32 x i64> %36, i64 11, !dbg !20 + %84 = shl nuw nsw i64 %83, 7, !dbg !20 + %85 = extractelement <32 x i64> %36, i64 10, !dbg !20 + %86 = shl nuw nsw i64 %85, 7, !dbg !20 + %87 = extractelement <32 x i64> %36, i64 9, !dbg !20 + %88 = shl nuw nsw i64 %87, 7, !dbg !20 + %89 = extractelement <32 x i64> %36, i64 8, !dbg !20 + %90 = shl nuw nsw i64 %89, 7, !dbg !20 + %91 = extractelement <32 x i64> %36, i64 7, !dbg !20 + %92 = shl nuw nsw i64 %91, 7, !dbg !20 + %93 = extractelement <32 x i64> %36, i64 6, !dbg !20 + %94 = shl nuw nsw i64 %93, 7, !dbg !20 + %95 = extractelement <32 x i64> %36, i64 5, !dbg !20 + %96 = shl nuw nsw i64 %95, 7, !dbg !20 + %97 = extractelement <32 x i64> %36, i64 4, !dbg !20 + %98 = shl nuw nsw i64 %97, 7, !dbg !20 + %99 = extractelement <32 x i64> %36, i64 3, !dbg !20 + %100 = shl nuw nsw i64 %99, 7, !dbg !20 + %101 = extractelement <32 x i64> %36, i64 2, !dbg !20 + %102 = shl nuw nsw i64 %101, 7, !dbg !20 + %103 = extractelement <32 x i64> %36, i64 1, !dbg !20 + %104 = shl nuw nsw i64 %103, 7, !dbg !20 + %105 = extractelement <32 x i64> %36, i64 0, !dbg !20 + %106 = shl nuw nsw i64 %105, 7, !dbg !20 + %107 = getelementptr bfloat, ptr addrspace(1) %0, i64 %44, !dbg !21 + %108 = getelementptr bfloat, ptr addrspace(1) %0, i64 %46, !dbg !21 + %109 = getelementptr bfloat, ptr addrspace(1) %0, i64 %48, !dbg !21 + %110 = getelementptr bfloat, ptr addrspace(1) %0, i64 %50, !dbg !21 + %111 = getelementptr bfloat, ptr addrspace(1) %0, i64 %52, !dbg !21 + %112 = getelementptr bfloat, ptr addrspace(1) %0, i64 %54, !dbg !21 + %113 = getelementptr bfloat, ptr addrspace(1) %0, i64 %56, !dbg !21 + %114 = getelementptr bfloat, ptr addrspace(1) %0, i64 %58, !dbg !21 + %115 = getelementptr bfloat, ptr addrspace(1) %0, i64 %60, !dbg !21 + %116 = getelementptr bfloat, ptr addrspace(1) %0, i64 %62, !dbg !21 + %117 = getelementptr bfloat, ptr addrspace(1) %0, i64 %64, !dbg !21 + %118 = getelementptr bfloat, ptr addrspace(1) %0, i64 %66, !dbg !21 + %119 = getelementptr bfloat, ptr addrspace(1) %0, i64 %68, !dbg !21 + %120 = getelementptr bfloat, ptr addrspace(1) %0, i64 %70, !dbg !21 + %121 = getelementptr bfloat, ptr addrspace(1) %0, i64 %72, !dbg !21 + %122 = getelementptr bfloat, ptr addrspace(1) %0, i64 %74, !dbg !21 + %123 = getelementptr bfloat, ptr addrspace(1) %0, i64 %76, !dbg !21 + %124 = getelementptr bfloat, ptr addrspace(1) %0, i64 %78, !dbg !21 + %125 = getelementptr bfloat, ptr addrspace(1) %0, i64 %80, !dbg !21 + %126 = getelementptr bfloat, ptr addrspace(1) %0, i64 %82, !dbg !21 + %127 = getelementptr bfloat, ptr addrspace(1) %0, i64 %84, !dbg !21 + %128 = getelementptr bfloat, ptr addrspace(1) %0, i64 %86, !dbg !21 + %129 = getelementptr bfloat, ptr addrspace(1) %0, i64 %88, !dbg !21 + %130 = getelementptr bfloat, ptr addrspace(1) %0, i64 %90, !dbg !21 + %131 = getelementptr bfloat, ptr addrspace(1) %0, i64 %92, !dbg !21 + %132 = getelementptr bfloat, ptr addrspace(1) %0, i64 %94, !dbg !21 + %133 = getelementptr bfloat, ptr addrspace(1) %0, i64 %96, !dbg !21 + %134 = getelementptr bfloat, ptr addrspace(1) %0, i64 %98, !dbg !21 + %135 = getelementptr bfloat, ptr addrspace(1) %0, i64 %100, !dbg !21 + %136 = getelementptr bfloat, ptr addrspace(1) %0, i64 %102, !dbg !21 + %137 = getelementptr bfloat, ptr addrspace(1) %0, i64 %104, !dbg !21 + %138 = getelementptr bfloat, ptr addrspace(1) %0, i64 %106, !dbg !21 + %139 = zext nneg i32 %42 to i64, !dbg !22 + %140 = getelementptr bfloat, ptr addrspace(1) %107, i64 %139, !dbg !22 + %141 = getelementptr bfloat, ptr addrspace(1) %108, i64 %139, !dbg !22 + %142 = getelementptr bfloat, ptr addrspace(1) %109, i64 %139, !dbg !22 + %143 = getelementptr bfloat, ptr addrspace(1) %110, i64 %139, !dbg !22 + %144 = getelementptr bfloat, ptr addrspace(1) %111, i64 %139, !dbg !22 + %145 = getelementptr bfloat, ptr addrspace(1) %112, i64 %139, !dbg !22 + %146 = getelementptr bfloat, ptr addrspace(1) %113, i64 %139, !dbg !22 + %147 = getelementptr bfloat, ptr addrspace(1) %114, i64 %139, !dbg !22 + %148 = getelementptr bfloat, ptr addrspace(1) %115, i64 %139, !dbg !22 + %149 = getelementptr bfloat, ptr addrspace(1) %116, i64 %139, !dbg !22 + %150 = getelementptr bfloat, ptr addrspace(1) %117, i64 %139, !dbg !22 + %151 = getelementptr bfloat, ptr addrspace(1) %118, i64 %139, !dbg !22 + %152 = getelementptr bfloat, ptr addrspace(1) %119, i64 %139, !dbg !22 + %153 = getelementptr bfloat, ptr addrspace(1) %120, i64 %139, !dbg !22 + %154 = getelementptr bfloat, ptr addrspace(1) %121, i64 %139, !dbg !22 + %155 = getelementptr bfloat, ptr addrspace(1) %122, i64 %139, !dbg !22 + %156 = getelementptr bfloat, ptr addrspace(1) %123, i64 %139, !dbg !22 + %157 = getelementptr bfloat, ptr addrspace(1) %124, i64 %139, !dbg !22 + %158 = getelementptr bfloat, ptr addrspace(1) %125, i64 %139, !dbg !22 + %159 = getelementptr bfloat, ptr addrspace(1) %126, i64 %139, !dbg !22 + %160 = getelementptr bfloat, ptr addrspace(1) %127, i64 %139, !dbg !22 + %161 = getelementptr bfloat, ptr addrspace(1) %128, i64 %139, !dbg !22 + %162 = getelementptr bfloat, ptr addrspace(1) %129, i64 %139, !dbg !22 + %163 = getelementptr bfloat, ptr addrspace(1) %130, i64 %139, !dbg !22 + %164 = getelementptr bfloat, ptr addrspace(1) %131, i64 %139, !dbg !22 + %165 = getelementptr bfloat, ptr addrspace(1) %132, i64 %139, !dbg !22 + %166 = getelementptr bfloat, ptr addrspace(1) %133, i64 %139, !dbg !22 + %167 = getelementptr bfloat, ptr addrspace(1) %134, i64 %139, !dbg !22 + %168 = getelementptr bfloat, ptr addrspace(1) %135, i64 %139, !dbg !22 + %169 = getelementptr bfloat, ptr addrspace(1) %136, i64 %139, !dbg !22 + %170 = getelementptr bfloat, ptr addrspace(1) %137, i64 %139, !dbg !22 + %171 = getelementptr bfloat, ptr addrspace(1) %138, i64 %139, !dbg !22 + %172 = extractelement <32 x i1> %40, i64 31, !dbg !23 + %173 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %140, i1 %172) #6, !dbg !23 + %174 = extractvalue { i32, i32, i32, i32 } %173, 0, !dbg !23 + %175 = bitcast i32 %174 to <2 x bfloat>, !dbg !23 + %176 = extractvalue { i32, i32, i32, i32 } %173, 1, !dbg !23 + %177 = bitcast i32 %176 to <2 x bfloat>, !dbg !23 + %178 = extractvalue { i32, i32, i32, i32 } %173, 2, !dbg !23 + %179 = bitcast i32 %178 to <2 x bfloat>, !dbg !23 + %180 = extractvalue { i32, i32, i32, i32 } %173, 3, !dbg !23 + %181 = bitcast i32 %180 to <2 x bfloat>, !dbg !23 + %182 = extractelement <2 x bfloat> %175, i64 0, !dbg !23 + %183 = extractelement <2 x bfloat> %175, i64 1, !dbg !23 + %184 = extractelement <2 x bfloat> %177, i64 0, !dbg !23 + %185 = extractelement <2 x bfloat> %177, i64 1, !dbg !23 + %186 = extractelement <2 x bfloat> %179, i64 0, !dbg !23 + %187 = extractelement <2 x bfloat> %179, i64 1, !dbg !23 + %188 = extractelement <2 x bfloat> %181, i64 0, !dbg !23 + %189 = extractelement <2 x bfloat> %181, i64 1, !dbg !23 + %190 = extractelement <32 x i1> %40, i64 30, !dbg !23 + %191 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %141, i1 %190) #6, !dbg !23 + %192 = extractvalue { i32, i32, i32, i32 } %191, 0, !dbg !23 + %193 = bitcast i32 %192 to <2 x bfloat>, !dbg !23 + %194 = extractvalue { i32, i32, i32, i32 } %191, 1, !dbg !23 + %195 = bitcast i32 %194 to <2 x bfloat>, !dbg !23 + %196 = extractvalue { i32, i32, i32, i32 } %191, 2, !dbg !23 + %197 = bitcast i32 %196 to <2 x bfloat>, !dbg !23 + %198 = extractvalue { i32, i32, i32, i32 } %191, 3, !dbg !23 + %199 = bitcast i32 %198 to <2 x bfloat>, !dbg !23 + %200 = extractelement <2 x bfloat> %193, i64 0, !dbg !23 + %201 = extractelement <2 x bfloat> %193, i64 1, !dbg !23 + %202 = extractelement <2 x bfloat> %195, i64 0, !dbg !23 + %203 = extractelement <2 x bfloat> %195, i64 1, !dbg !23 + %204 = extractelement <2 x bfloat> %197, i64 0, !dbg !23 + %205 = extractelement <2 x bfloat> %197, i64 1, !dbg !23 + %206 = extractelement <2 x bfloat> %199, i64 0, !dbg !23 + %207 = extractelement <2 x bfloat> %199, i64 1, !dbg !23 + %208 = extractelement <32 x i1> %40, i64 29, !dbg !23 + %209 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %142, i1 %208) #6, !dbg !23 + %210 = extractvalue { i32, i32, i32, i32 } %209, 0, !dbg !23 + %211 = bitcast i32 %210 to <2 x bfloat>, !dbg !23 + %212 = extractvalue { i32, i32, i32, i32 } %209, 1, !dbg !23 + %213 = bitcast i32 %212 to <2 x bfloat>, !dbg !23 + %214 = extractvalue { i32, i32, i32, i32 } %209, 2, !dbg !23 + %215 = bitcast i32 %214 to <2 x bfloat>, !dbg !23 + %216 = extractvalue { i32, i32, i32, i32 } %209, 3, !dbg !23 + %217 = bitcast i32 %216 to <2 x bfloat>, !dbg !23 + %218 = extractelement <2 x bfloat> %211, i64 0, !dbg !23 + %219 = extractelement <2 x bfloat> %211, i64 1, !dbg !23 + %220 = extractelement <2 x bfloat> %213, i64 0, !dbg !23 + %221 = extractelement <2 x bfloat> %213, i64 1, !dbg !23 + %222 = extractelement <2 x bfloat> %215, i64 0, !dbg !23 + %223 = extractelement <2 x bfloat> %215, i64 1, !dbg !23 + %224 = extractelement <2 x bfloat> %217, i64 0, !dbg !23 + %225 = extractelement <2 x bfloat> %217, i64 1, !dbg !23 + %226 = extractelement <32 x i1> %40, i64 28, !dbg !23 + %227 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %143, i1 %226) #6, !dbg !23 + %228 = extractvalue { i32, i32, i32, i32 } %227, 0, !dbg !23 + %229 = bitcast i32 %228 to <2 x bfloat>, !dbg !23 + %230 = extractvalue { i32, i32, i32, i32 } %227, 1, !dbg !23 + %231 = bitcast i32 %230 to <2 x bfloat>, !dbg !23 + %232 = extractvalue { i32, i32, i32, i32 } %227, 2, !dbg !23 + %233 = bitcast i32 %232 to <2 x bfloat>, !dbg !23 + %234 = extractvalue { i32, i32, i32, i32 } %227, 3, !dbg !23 + %235 = bitcast i32 %234 to <2 x bfloat>, !dbg !23 + %236 = extractelement <2 x bfloat> %229, i64 0, !dbg !23 + %237 = extractelement <2 x bfloat> %229, i64 1, !dbg !23 + %238 = extractelement <2 x bfloat> %231, i64 0, !dbg !23 + %239 = extractelement <2 x bfloat> %231, i64 1, !dbg !23 + %240 = extractelement <2 x bfloat> %233, i64 0, !dbg !23 + %241 = extractelement <2 x bfloat> %233, i64 1, !dbg !23 + %242 = extractelement <2 x bfloat> %235, i64 0, !dbg !23 + %243 = extractelement <2 x bfloat> %235, i64 1, !dbg !23 + %244 = extractelement <32 x i1> %40, i64 27, !dbg !23 + %245 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %144, i1 %244) #6, !dbg !23 + %246 = extractvalue { i32, i32, i32, i32 } %245, 0, !dbg !23 + %247 = bitcast i32 %246 to <2 x bfloat>, !dbg !23 + %248 = extractvalue { i32, i32, i32, i32 } %245, 1, !dbg !23 + %249 = bitcast i32 %248 to <2 x bfloat>, !dbg !23 + %250 = extractvalue { i32, i32, i32, i32 } %245, 2, !dbg !23 + %251 = bitcast i32 %250 to <2 x bfloat>, !dbg !23 + %252 = extractvalue { i32, i32, i32, i32 } %245, 3, !dbg !23 + %253 = bitcast i32 %252 to <2 x bfloat>, !dbg !23 + %254 = extractelement <2 x bfloat> %247, i64 0, !dbg !23 + %255 = extractelement <2 x bfloat> %247, i64 1, !dbg !23 + %256 = extractelement <2 x bfloat> %249, i64 0, !dbg !23 + %257 = extractelement <2 x bfloat> %249, i64 1, !dbg !23 + %258 = extractelement <2 x bfloat> %251, i64 0, !dbg !23 + %259 = extractelement <2 x bfloat> %251, i64 1, !dbg !23 + %260 = extractelement <2 x bfloat> %253, i64 0, !dbg !23 + %261 = extractelement <2 x bfloat> %253, i64 1, !dbg !23 + %262 = extractelement <32 x i1> %40, i64 26, !dbg !23 + %263 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %145, i1 %262) #6, !dbg !23 + %264 = extractvalue { i32, i32, i32, i32 } %263, 0, !dbg !23 + %265 = bitcast i32 %264 to <2 x bfloat>, !dbg !23 + %266 = extractvalue { i32, i32, i32, i32 } %263, 1, !dbg !23 + %267 = bitcast i32 %266 to <2 x bfloat>, !dbg !23 + %268 = extractvalue { i32, i32, i32, i32 } %263, 2, !dbg !23 + %269 = bitcast i32 %268 to <2 x bfloat>, !dbg !23 + %270 = extractvalue { i32, i32, i32, i32 } %263, 3, !dbg !23 + %271 = bitcast i32 %270 to <2 x bfloat>, !dbg !23 + %272 = extractelement <2 x bfloat> %265, i64 0, !dbg !23 + %273 = extractelement <2 x bfloat> %265, i64 1, !dbg !23 + %274 = extractelement <2 x bfloat> %267, i64 0, !dbg !23 + %275 = extractelement <2 x bfloat> %267, i64 1, !dbg !23 + %276 = extractelement <2 x bfloat> %269, i64 0, !dbg !23 + %277 = extractelement <2 x bfloat> %269, i64 1, !dbg !23 + %278 = extractelement <2 x bfloat> %271, i64 0, !dbg !23 + %279 = extractelement <2 x bfloat> %271, i64 1, !dbg !23 + %280 = extractelement <32 x i1> %40, i64 25, !dbg !23 + %281 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %146, i1 %280) #6, !dbg !23 + %282 = extractvalue { i32, i32, i32, i32 } %281, 0, !dbg !23 + %283 = bitcast i32 %282 to <2 x bfloat>, !dbg !23 + %284 = extractvalue { i32, i32, i32, i32 } %281, 1, !dbg !23 + %285 = bitcast i32 %284 to <2 x bfloat>, !dbg !23 + %286 = extractvalue { i32, i32, i32, i32 } %281, 2, !dbg !23 + %287 = bitcast i32 %286 to <2 x bfloat>, !dbg !23 + %288 = extractvalue { i32, i32, i32, i32 } %281, 3, !dbg !23 + %289 = bitcast i32 %288 to <2 x bfloat>, !dbg !23 + %290 = extractelement <2 x bfloat> %283, i64 0, !dbg !23 + %291 = extractelement <2 x bfloat> %283, i64 1, !dbg !23 + %292 = extractelement <2 x bfloat> %285, i64 0, !dbg !23 + %293 = extractelement <2 x bfloat> %285, i64 1, !dbg !23 + %294 = extractelement <2 x bfloat> %287, i64 0, !dbg !23 + %295 = extractelement <2 x bfloat> %287, i64 1, !dbg !23 + %296 = extractelement <2 x bfloat> %289, i64 0, !dbg !23 + %297 = extractelement <2 x bfloat> %289, i64 1, !dbg !23 + %298 = extractelement <32 x i1> %40, i64 24, !dbg !23 + %299 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %147, i1 %298) #6, !dbg !23 + %300 = extractvalue { i32, i32, i32, i32 } %299, 0, !dbg !23 + %301 = bitcast i32 %300 to <2 x bfloat>, !dbg !23 + %302 = extractvalue { i32, i32, i32, i32 } %299, 1, !dbg !23 + %303 = bitcast i32 %302 to <2 x bfloat>, !dbg !23 + %304 = extractvalue { i32, i32, i32, i32 } %299, 2, !dbg !23 + %305 = bitcast i32 %304 to <2 x bfloat>, !dbg !23 + %306 = extractvalue { i32, i32, i32, i32 } %299, 3, !dbg !23 + %307 = bitcast i32 %306 to <2 x bfloat>, !dbg !23 + %308 = extractelement <2 x bfloat> %301, i64 0, !dbg !23 + %309 = extractelement <2 x bfloat> %301, i64 1, !dbg !23 + %310 = extractelement <2 x bfloat> %303, i64 0, !dbg !23 + %311 = extractelement <2 x bfloat> %303, i64 1, !dbg !23 + %312 = extractelement <2 x bfloat> %305, i64 0, !dbg !23 + %313 = extractelement <2 x bfloat> %305, i64 1, !dbg !23 + %314 = extractelement <2 x bfloat> %307, i64 0, !dbg !23 + %315 = extractelement <2 x bfloat> %307, i64 1, !dbg !23 + %316 = extractelement <32 x i1> %40, i64 23, !dbg !23 + %317 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %148, i1 %316) #6, !dbg !23 + %318 = extractvalue { i32, i32, i32, i32 } %317, 0, !dbg !23 + %319 = bitcast i32 %318 to <2 x bfloat>, !dbg !23 + %320 = extractvalue { i32, i32, i32, i32 } %317, 1, !dbg !23 + %321 = bitcast i32 %320 to <2 x bfloat>, !dbg !23 + %322 = extractvalue { i32, i32, i32, i32 } %317, 2, !dbg !23 + %323 = bitcast i32 %322 to <2 x bfloat>, !dbg !23 + %324 = extractvalue { i32, i32, i32, i32 } %317, 3, !dbg !23 + %325 = bitcast i32 %324 to <2 x bfloat>, !dbg !23 + %326 = extractelement <2 x bfloat> %319, i64 0, !dbg !23 + %327 = extractelement <2 x bfloat> %319, i64 1, !dbg !23 + %328 = extractelement <2 x bfloat> %321, i64 0, !dbg !23 + %329 = extractelement <2 x bfloat> %321, i64 1, !dbg !23 + %330 = extractelement <2 x bfloat> %323, i64 0, !dbg !23 + %331 = extractelement <2 x bfloat> %323, i64 1, !dbg !23 + %332 = extractelement <2 x bfloat> %325, i64 0, !dbg !23 + %333 = extractelement <2 x bfloat> %325, i64 1, !dbg !23 + %334 = extractelement <32 x i1> %40, i64 22, !dbg !23 + %335 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %149, i1 %334) #6, !dbg !23 + %336 = extractvalue { i32, i32, i32, i32 } %335, 0, !dbg !23 + %337 = bitcast i32 %336 to <2 x bfloat>, !dbg !23 + %338 = extractvalue { i32, i32, i32, i32 } %335, 1, !dbg !23 + %339 = bitcast i32 %338 to <2 x bfloat>, !dbg !23 + %340 = extractvalue { i32, i32, i32, i32 } %335, 2, !dbg !23 + %341 = bitcast i32 %340 to <2 x bfloat>, !dbg !23 + %342 = extractvalue { i32, i32, i32, i32 } %335, 3, !dbg !23 + %343 = bitcast i32 %342 to <2 x bfloat>, !dbg !23 + %344 = extractelement <2 x bfloat> %337, i64 0, !dbg !23 + %345 = extractelement <2 x bfloat> %337, i64 1, !dbg !23 + %346 = extractelement <2 x bfloat> %339, i64 0, !dbg !23 + %347 = extractelement <2 x bfloat> %339, i64 1, !dbg !23 + %348 = extractelement <2 x bfloat> %341, i64 0, !dbg !23 + %349 = extractelement <2 x bfloat> %341, i64 1, !dbg !23 + %350 = extractelement <2 x bfloat> %343, i64 0, !dbg !23 + %351 = extractelement <2 x bfloat> %343, i64 1, !dbg !23 + %352 = extractelement <32 x i1> %40, i64 21, !dbg !23 + %353 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %150, i1 %352) #6, !dbg !23 + %354 = extractvalue { i32, i32, i32, i32 } %353, 0, !dbg !23 + %355 = bitcast i32 %354 to <2 x bfloat>, !dbg !23 + %356 = extractvalue { i32, i32, i32, i32 } %353, 1, !dbg !23 + %357 = bitcast i32 %356 to <2 x bfloat>, !dbg !23 + %358 = extractvalue { i32, i32, i32, i32 } %353, 2, !dbg !23 + %359 = bitcast i32 %358 to <2 x bfloat>, !dbg !23 + %360 = extractvalue { i32, i32, i32, i32 } %353, 3, !dbg !23 + %361 = bitcast i32 %360 to <2 x bfloat>, !dbg !23 + %362 = extractelement <2 x bfloat> %355, i64 0, !dbg !23 + %363 = extractelement <2 x bfloat> %355, i64 1, !dbg !23 + %364 = extractelement <2 x bfloat> %357, i64 0, !dbg !23 + %365 = extractelement <2 x bfloat> %357, i64 1, !dbg !23 + %366 = extractelement <2 x bfloat> %359, i64 0, !dbg !23 + %367 = extractelement <2 x bfloat> %359, i64 1, !dbg !23 + %368 = extractelement <2 x bfloat> %361, i64 0, !dbg !23 + %369 = extractelement <2 x bfloat> %361, i64 1, !dbg !23 + %370 = extractelement <32 x i1> %40, i64 20, !dbg !23 + %371 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %151, i1 %370) #6, !dbg !23 + %372 = extractvalue { i32, i32, i32, i32 } %371, 0, !dbg !23 + %373 = bitcast i32 %372 to <2 x bfloat>, !dbg !23 + %374 = extractvalue { i32, i32, i32, i32 } %371, 1, !dbg !23 + %375 = bitcast i32 %374 to <2 x bfloat>, !dbg !23 + %376 = extractvalue { i32, i32, i32, i32 } %371, 2, !dbg !23 + %377 = bitcast i32 %376 to <2 x bfloat>, !dbg !23 + %378 = extractvalue { i32, i32, i32, i32 } %371, 3, !dbg !23 + %379 = bitcast i32 %378 to <2 x bfloat>, !dbg !23 + %380 = extractelement <2 x bfloat> %373, i64 0, !dbg !23 + %381 = extractelement <2 x bfloat> %373, i64 1, !dbg !23 + %382 = extractelement <2 x bfloat> %375, i64 0, !dbg !23 + %383 = extractelement <2 x bfloat> %375, i64 1, !dbg !23 + %384 = extractelement <2 x bfloat> %377, i64 0, !dbg !23 + %385 = extractelement <2 x bfloat> %377, i64 1, !dbg !23 + %386 = extractelement <2 x bfloat> %379, i64 0, !dbg !23 + %387 = extractelement <2 x bfloat> %379, i64 1, !dbg !23 + %388 = extractelement <32 x i1> %40, i64 19, !dbg !23 + %389 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %152, i1 %388) #6, !dbg !23 + %390 = extractvalue { i32, i32, i32, i32 } %389, 0, !dbg !23 + %391 = bitcast i32 %390 to <2 x bfloat>, !dbg !23 + %392 = extractvalue { i32, i32, i32, i32 } %389, 1, !dbg !23 + %393 = bitcast i32 %392 to <2 x bfloat>, !dbg !23 + %394 = extractvalue { i32, i32, i32, i32 } %389, 2, !dbg !23 + %395 = bitcast i32 %394 to <2 x bfloat>, !dbg !23 + %396 = extractvalue { i32, i32, i32, i32 } %389, 3, !dbg !23 + %397 = bitcast i32 %396 to <2 x bfloat>, !dbg !23 + %398 = extractelement <2 x bfloat> %391, i64 0, !dbg !23 + %399 = extractelement <2 x bfloat> %391, i64 1, !dbg !23 + %400 = extractelement <2 x bfloat> %393, i64 0, !dbg !23 + %401 = extractelement <2 x bfloat> %393, i64 1, !dbg !23 + %402 = extractelement <2 x bfloat> %395, i64 0, !dbg !23 + %403 = extractelement <2 x bfloat> %395, i64 1, !dbg !23 + %404 = extractelement <2 x bfloat> %397, i64 0, !dbg !23 + %405 = extractelement <2 x bfloat> %397, i64 1, !dbg !23 + %406 = extractelement <32 x i1> %40, i64 18, !dbg !23 + %407 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %153, i1 %406) #6, !dbg !23 + %408 = extractvalue { i32, i32, i32, i32 } %407, 0, !dbg !23 + %409 = bitcast i32 %408 to <2 x bfloat>, !dbg !23 + %410 = extractvalue { i32, i32, i32, i32 } %407, 1, !dbg !23 + %411 = bitcast i32 %410 to <2 x bfloat>, !dbg !23 + %412 = extractvalue { i32, i32, i32, i32 } %407, 2, !dbg !23 + %413 = bitcast i32 %412 to <2 x bfloat>, !dbg !23 + %414 = extractvalue { i32, i32, i32, i32 } %407, 3, !dbg !23 + %415 = bitcast i32 %414 to <2 x bfloat>, !dbg !23 + %416 = extractelement <2 x bfloat> %409, i64 0, !dbg !23 + %417 = extractelement <2 x bfloat> %409, i64 1, !dbg !23 + %418 = extractelement <2 x bfloat> %411, i64 0, !dbg !23 + %419 = extractelement <2 x bfloat> %411, i64 1, !dbg !23 + %420 = extractelement <2 x bfloat> %413, i64 0, !dbg !23 + %421 = extractelement <2 x bfloat> %413, i64 1, !dbg !23 + %422 = extractelement <2 x bfloat> %415, i64 0, !dbg !23 + %423 = extractelement <2 x bfloat> %415, i64 1, !dbg !23 + %424 = extractelement <32 x i1> %40, i64 17, !dbg !23 + %425 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %154, i1 %424) #6, !dbg !23 + %426 = extractvalue { i32, i32, i32, i32 } %425, 0, !dbg !23 + %427 = bitcast i32 %426 to <2 x bfloat>, !dbg !23 + %428 = extractvalue { i32, i32, i32, i32 } %425, 1, !dbg !23 + %429 = bitcast i32 %428 to <2 x bfloat>, !dbg !23 + %430 = extractvalue { i32, i32, i32, i32 } %425, 2, !dbg !23 + %431 = bitcast i32 %430 to <2 x bfloat>, !dbg !23 + %432 = extractvalue { i32, i32, i32, i32 } %425, 3, !dbg !23 + %433 = bitcast i32 %432 to <2 x bfloat>, !dbg !23 + %434 = extractelement <2 x bfloat> %427, i64 0, !dbg !23 + %435 = extractelement <2 x bfloat> %427, i64 1, !dbg !23 + %436 = extractelement <2 x bfloat> %429, i64 0, !dbg !23 + %437 = extractelement <2 x bfloat> %429, i64 1, !dbg !23 + %438 = extractelement <2 x bfloat> %431, i64 0, !dbg !23 + %439 = extractelement <2 x bfloat> %431, i64 1, !dbg !23 + %440 = extractelement <2 x bfloat> %433, i64 0, !dbg !23 + %441 = extractelement <2 x bfloat> %433, i64 1, !dbg !23 + %442 = extractelement <32 x i1> %40, i64 16, !dbg !23 + %443 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %155, i1 %442) #6, !dbg !23 + %444 = extractvalue { i32, i32, i32, i32 } %443, 0, !dbg !23 + %445 = bitcast i32 %444 to <2 x bfloat>, !dbg !23 + %446 = extractvalue { i32, i32, i32, i32 } %443, 1, !dbg !23 + %447 = bitcast i32 %446 to <2 x bfloat>, !dbg !23 + %448 = extractvalue { i32, i32, i32, i32 } %443, 2, !dbg !23 + %449 = bitcast i32 %448 to <2 x bfloat>, !dbg !23 + %450 = extractvalue { i32, i32, i32, i32 } %443, 3, !dbg !23 + %451 = bitcast i32 %450 to <2 x bfloat>, !dbg !23 + %452 = extractelement <2 x bfloat> %445, i64 0, !dbg !23 + %453 = extractelement <2 x bfloat> %445, i64 1, !dbg !23 + %454 = extractelement <2 x bfloat> %447, i64 0, !dbg !23 + %455 = extractelement <2 x bfloat> %447, i64 1, !dbg !23 + %456 = extractelement <2 x bfloat> %449, i64 0, !dbg !23 + %457 = extractelement <2 x bfloat> %449, i64 1, !dbg !23 + %458 = extractelement <2 x bfloat> %451, i64 0, !dbg !23 + %459 = extractelement <2 x bfloat> %451, i64 1, !dbg !23 + %460 = extractelement <32 x i1> %40, i64 15, !dbg !23 + %461 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %156, i1 %460) #6, !dbg !23 + %462 = extractvalue { i32, i32, i32, i32 } %461, 0, !dbg !23 + %463 = bitcast i32 %462 to <2 x bfloat>, !dbg !23 + %464 = extractvalue { i32, i32, i32, i32 } %461, 1, !dbg !23 + %465 = bitcast i32 %464 to <2 x bfloat>, !dbg !23 + %466 = extractvalue { i32, i32, i32, i32 } %461, 2, !dbg !23 + %467 = bitcast i32 %466 to <2 x bfloat>, !dbg !23 + %468 = extractvalue { i32, i32, i32, i32 } %461, 3, !dbg !23 + %469 = bitcast i32 %468 to <2 x bfloat>, !dbg !23 + %470 = extractelement <2 x bfloat> %463, i64 0, !dbg !23 + %471 = extractelement <2 x bfloat> %463, i64 1, !dbg !23 + %472 = extractelement <2 x bfloat> %465, i64 0, !dbg !23 + %473 = extractelement <2 x bfloat> %465, i64 1, !dbg !23 + %474 = extractelement <2 x bfloat> %467, i64 0, !dbg !23 + %475 = extractelement <2 x bfloat> %467, i64 1, !dbg !23 + %476 = extractelement <2 x bfloat> %469, i64 0, !dbg !23 + %477 = extractelement <2 x bfloat> %469, i64 1, !dbg !23 + %478 = extractelement <32 x i1> %40, i64 14, !dbg !23 + %479 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %157, i1 %478) #6, !dbg !23 + %480 = extractvalue { i32, i32, i32, i32 } %479, 0, !dbg !23 + %481 = bitcast i32 %480 to <2 x bfloat>, !dbg !23 + %482 = extractvalue { i32, i32, i32, i32 } %479, 1, !dbg !23 + %483 = bitcast i32 %482 to <2 x bfloat>, !dbg !23 + %484 = extractvalue { i32, i32, i32, i32 } %479, 2, !dbg !23 + %485 = bitcast i32 %484 to <2 x bfloat>, !dbg !23 + %486 = extractvalue { i32, i32, i32, i32 } %479, 3, !dbg !23 + %487 = bitcast i32 %486 to <2 x bfloat>, !dbg !23 + %488 = extractelement <2 x bfloat> %481, i64 0, !dbg !23 + %489 = extractelement <2 x bfloat> %481, i64 1, !dbg !23 + %490 = extractelement <2 x bfloat> %483, i64 0, !dbg !23 + %491 = extractelement <2 x bfloat> %483, i64 1, !dbg !23 + %492 = extractelement <2 x bfloat> %485, i64 0, !dbg !23 + %493 = extractelement <2 x bfloat> %485, i64 1, !dbg !23 + %494 = extractelement <2 x bfloat> %487, i64 0, !dbg !23 + %495 = extractelement <2 x bfloat> %487, i64 1, !dbg !23 + %496 = extractelement <32 x i1> %40, i64 13, !dbg !23 + %497 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %158, i1 %496) #6, !dbg !23 + %498 = extractvalue { i32, i32, i32, i32 } %497, 0, !dbg !23 + %499 = bitcast i32 %498 to <2 x bfloat>, !dbg !23 + %500 = extractvalue { i32, i32, i32, i32 } %497, 1, !dbg !23 + %501 = bitcast i32 %500 to <2 x bfloat>, !dbg !23 + %502 = extractvalue { i32, i32, i32, i32 } %497, 2, !dbg !23 + %503 = bitcast i32 %502 to <2 x bfloat>, !dbg !23 + %504 = extractvalue { i32, i32, i32, i32 } %497, 3, !dbg !23 + %505 = bitcast i32 %504 to <2 x bfloat>, !dbg !23 + %506 = extractelement <2 x bfloat> %499, i64 0, !dbg !23 + %507 = extractelement <2 x bfloat> %499, i64 1, !dbg !23 + %508 = extractelement <2 x bfloat> %501, i64 0, !dbg !23 + %509 = extractelement <2 x bfloat> %501, i64 1, !dbg !23 + %510 = extractelement <2 x bfloat> %503, i64 0, !dbg !23 + %511 = extractelement <2 x bfloat> %503, i64 1, !dbg !23 + %512 = extractelement <2 x bfloat> %505, i64 0, !dbg !23 + %513 = extractelement <2 x bfloat> %505, i64 1, !dbg !23 + %514 = extractelement <32 x i1> %40, i64 12, !dbg !23 + %515 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %159, i1 %514) #6, !dbg !23 + %516 = extractvalue { i32, i32, i32, i32 } %515, 0, !dbg !23 + %517 = bitcast i32 %516 to <2 x bfloat>, !dbg !23 + %518 = extractvalue { i32, i32, i32, i32 } %515, 1, !dbg !23 + %519 = bitcast i32 %518 to <2 x bfloat>, !dbg !23 + %520 = extractvalue { i32, i32, i32, i32 } %515, 2, !dbg !23 + %521 = bitcast i32 %520 to <2 x bfloat>, !dbg !23 + %522 = extractvalue { i32, i32, i32, i32 } %515, 3, !dbg !23 + %523 = bitcast i32 %522 to <2 x bfloat>, !dbg !23 + %524 = extractelement <2 x bfloat> %517, i64 0, !dbg !23 + %525 = extractelement <2 x bfloat> %517, i64 1, !dbg !23 + %526 = extractelement <2 x bfloat> %519, i64 0, !dbg !23 + %527 = extractelement <2 x bfloat> %519, i64 1, !dbg !23 + %528 = extractelement <2 x bfloat> %521, i64 0, !dbg !23 + %529 = extractelement <2 x bfloat> %521, i64 1, !dbg !23 + %530 = extractelement <2 x bfloat> %523, i64 0, !dbg !23 + %531 = extractelement <2 x bfloat> %523, i64 1, !dbg !23 + %532 = extractelement <32 x i1> %40, i64 11, !dbg !23 + %533 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %160, i1 %532) #6, !dbg !23 + %534 = extractvalue { i32, i32, i32, i32 } %533, 0, !dbg !23 + %535 = bitcast i32 %534 to <2 x bfloat>, !dbg !23 + %536 = extractvalue { i32, i32, i32, i32 } %533, 1, !dbg !23 + %537 = bitcast i32 %536 to <2 x bfloat>, !dbg !23 + %538 = extractvalue { i32, i32, i32, i32 } %533, 2, !dbg !23 + %539 = bitcast i32 %538 to <2 x bfloat>, !dbg !23 + %540 = extractvalue { i32, i32, i32, i32 } %533, 3, !dbg !23 + %541 = bitcast i32 %540 to <2 x bfloat>, !dbg !23 + %542 = extractelement <2 x bfloat> %535, i64 0, !dbg !23 + %543 = extractelement <2 x bfloat> %535, i64 1, !dbg !23 + %544 = extractelement <2 x bfloat> %537, i64 0, !dbg !23 + %545 = extractelement <2 x bfloat> %537, i64 1, !dbg !23 + %546 = extractelement <2 x bfloat> %539, i64 0, !dbg !23 + %547 = extractelement <2 x bfloat> %539, i64 1, !dbg !23 + %548 = extractelement <2 x bfloat> %541, i64 0, !dbg !23 + %549 = extractelement <2 x bfloat> %541, i64 1, !dbg !23 + %550 = extractelement <32 x i1> %40, i64 10, !dbg !23 + %551 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %161, i1 %550) #6, !dbg !23 + %552 = extractvalue { i32, i32, i32, i32 } %551, 0, !dbg !23 + %553 = bitcast i32 %552 to <2 x bfloat>, !dbg !23 + %554 = extractvalue { i32, i32, i32, i32 } %551, 1, !dbg !23 + %555 = bitcast i32 %554 to <2 x bfloat>, !dbg !23 + %556 = extractvalue { i32, i32, i32, i32 } %551, 2, !dbg !23 + %557 = bitcast i32 %556 to <2 x bfloat>, !dbg !23 + %558 = extractvalue { i32, i32, i32, i32 } %551, 3, !dbg !23 + %559 = bitcast i32 %558 to <2 x bfloat>, !dbg !23 + %560 = extractelement <2 x bfloat> %553, i64 0, !dbg !23 + %561 = extractelement <2 x bfloat> %553, i64 1, !dbg !23 + %562 = extractelement <2 x bfloat> %555, i64 0, !dbg !23 + %563 = extractelement <2 x bfloat> %555, i64 1, !dbg !23 + %564 = extractelement <2 x bfloat> %557, i64 0, !dbg !23 + %565 = extractelement <2 x bfloat> %557, i64 1, !dbg !23 + %566 = extractelement <2 x bfloat> %559, i64 0, !dbg !23 + %567 = extractelement <2 x bfloat> %559, i64 1, !dbg !23 + %568 = extractelement <32 x i1> %40, i64 9, !dbg !23 + %569 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %162, i1 %568) #6, !dbg !23 + %570 = extractvalue { i32, i32, i32, i32 } %569, 0, !dbg !23 + %571 = bitcast i32 %570 to <2 x bfloat>, !dbg !23 + %572 = extractvalue { i32, i32, i32, i32 } %569, 1, !dbg !23 + %573 = bitcast i32 %572 to <2 x bfloat>, !dbg !23 + %574 = extractvalue { i32, i32, i32, i32 } %569, 2, !dbg !23 + %575 = bitcast i32 %574 to <2 x bfloat>, !dbg !23 + %576 = extractvalue { i32, i32, i32, i32 } %569, 3, !dbg !23 + %577 = bitcast i32 %576 to <2 x bfloat>, !dbg !23 + %578 = extractelement <2 x bfloat> %571, i64 0, !dbg !23 + %579 = extractelement <2 x bfloat> %571, i64 1, !dbg !23 + %580 = extractelement <2 x bfloat> %573, i64 0, !dbg !23 + %581 = extractelement <2 x bfloat> %573, i64 1, !dbg !23 + %582 = extractelement <2 x bfloat> %575, i64 0, !dbg !23 + %583 = extractelement <2 x bfloat> %575, i64 1, !dbg !23 + %584 = extractelement <2 x bfloat> %577, i64 0, !dbg !23 + %585 = extractelement <2 x bfloat> %577, i64 1, !dbg !23 + %586 = extractelement <32 x i1> %40, i64 8, !dbg !23 + %587 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %163, i1 %586) #6, !dbg !23 + %588 = extractvalue { i32, i32, i32, i32 } %587, 0, !dbg !23 + %589 = bitcast i32 %588 to <2 x bfloat>, !dbg !23 + %590 = extractvalue { i32, i32, i32, i32 } %587, 1, !dbg !23 + %591 = bitcast i32 %590 to <2 x bfloat>, !dbg !23 + %592 = extractvalue { i32, i32, i32, i32 } %587, 2, !dbg !23 + %593 = bitcast i32 %592 to <2 x bfloat>, !dbg !23 + %594 = extractvalue { i32, i32, i32, i32 } %587, 3, !dbg !23 + %595 = bitcast i32 %594 to <2 x bfloat>, !dbg !23 + %596 = extractelement <2 x bfloat> %589, i64 0, !dbg !23 + %597 = extractelement <2 x bfloat> %589, i64 1, !dbg !23 + %598 = extractelement <2 x bfloat> %591, i64 0, !dbg !23 + %599 = extractelement <2 x bfloat> %591, i64 1, !dbg !23 + %600 = extractelement <2 x bfloat> %593, i64 0, !dbg !23 + %601 = extractelement <2 x bfloat> %593, i64 1, !dbg !23 + %602 = extractelement <2 x bfloat> %595, i64 0, !dbg !23 + %603 = extractelement <2 x bfloat> %595, i64 1, !dbg !23 + %604 = extractelement <32 x i1> %40, i64 7, !dbg !23 + %605 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %164, i1 %604) #6, !dbg !23 + %606 = extractvalue { i32, i32, i32, i32 } %605, 0, !dbg !23 + %607 = bitcast i32 %606 to <2 x bfloat>, !dbg !23 + %608 = extractvalue { i32, i32, i32, i32 } %605, 1, !dbg !23 + %609 = bitcast i32 %608 to <2 x bfloat>, !dbg !23 + %610 = extractvalue { i32, i32, i32, i32 } %605, 2, !dbg !23 + %611 = bitcast i32 %610 to <2 x bfloat>, !dbg !23 + %612 = extractvalue { i32, i32, i32, i32 } %605, 3, !dbg !23 + %613 = bitcast i32 %612 to <2 x bfloat>, !dbg !23 + %614 = extractelement <2 x bfloat> %607, i64 0, !dbg !23 + %615 = extractelement <2 x bfloat> %607, i64 1, !dbg !23 + %616 = extractelement <2 x bfloat> %609, i64 0, !dbg !23 + %617 = extractelement <2 x bfloat> %609, i64 1, !dbg !23 + %618 = extractelement <2 x bfloat> %611, i64 0, !dbg !23 + %619 = extractelement <2 x bfloat> %611, i64 1, !dbg !23 + %620 = extractelement <2 x bfloat> %613, i64 0, !dbg !23 + %621 = extractelement <2 x bfloat> %613, i64 1, !dbg !23 + %622 = extractelement <32 x i1> %40, i64 6, !dbg !23 + %623 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %165, i1 %622) #6, !dbg !23 + %624 = extractvalue { i32, i32, i32, i32 } %623, 0, !dbg !23 + %625 = bitcast i32 %624 to <2 x bfloat>, !dbg !23 + %626 = extractvalue { i32, i32, i32, i32 } %623, 1, !dbg !23 + %627 = bitcast i32 %626 to <2 x bfloat>, !dbg !23 + %628 = extractvalue { i32, i32, i32, i32 } %623, 2, !dbg !23 + %629 = bitcast i32 %628 to <2 x bfloat>, !dbg !23 + %630 = extractvalue { i32, i32, i32, i32 } %623, 3, !dbg !23 + %631 = bitcast i32 %630 to <2 x bfloat>, !dbg !23 + %632 = extractelement <2 x bfloat> %625, i64 0, !dbg !23 + %633 = extractelement <2 x bfloat> %625, i64 1, !dbg !23 + %634 = extractelement <2 x bfloat> %627, i64 0, !dbg !23 + %635 = extractelement <2 x bfloat> %627, i64 1, !dbg !23 + %636 = extractelement <2 x bfloat> %629, i64 0, !dbg !23 + %637 = extractelement <2 x bfloat> %629, i64 1, !dbg !23 + %638 = extractelement <2 x bfloat> %631, i64 0, !dbg !23 + %639 = extractelement <2 x bfloat> %631, i64 1, !dbg !23 + %640 = extractelement <32 x i1> %40, i64 5, !dbg !23 + %641 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %166, i1 %640) #6, !dbg !23 + %642 = extractvalue { i32, i32, i32, i32 } %641, 0, !dbg !23 + %643 = bitcast i32 %642 to <2 x bfloat>, !dbg !23 + %644 = extractvalue { i32, i32, i32, i32 } %641, 1, !dbg !23 + %645 = bitcast i32 %644 to <2 x bfloat>, !dbg !23 + %646 = extractvalue { i32, i32, i32, i32 } %641, 2, !dbg !23 + %647 = bitcast i32 %646 to <2 x bfloat>, !dbg !23 + %648 = extractvalue { i32, i32, i32, i32 } %641, 3, !dbg !23 + %649 = bitcast i32 %648 to <2 x bfloat>, !dbg !23 + %650 = extractelement <2 x bfloat> %643, i64 0, !dbg !23 + %651 = extractelement <2 x bfloat> %643, i64 1, !dbg !23 + %652 = extractelement <2 x bfloat> %645, i64 0, !dbg !23 + %653 = extractelement <2 x bfloat> %645, i64 1, !dbg !23 + %654 = extractelement <2 x bfloat> %647, i64 0, !dbg !23 + %655 = extractelement <2 x bfloat> %647, i64 1, !dbg !23 + %656 = extractelement <2 x bfloat> %649, i64 0, !dbg !23 + %657 = extractelement <2 x bfloat> %649, i64 1, !dbg !23 + %658 = extractelement <32 x i1> %40, i64 4, !dbg !23 + %659 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %167, i1 %658) #6, !dbg !23 + %660 = extractvalue { i32, i32, i32, i32 } %659, 0, !dbg !23 + %661 = bitcast i32 %660 to <2 x bfloat>, !dbg !23 + %662 = extractvalue { i32, i32, i32, i32 } %659, 1, !dbg !23 + %663 = bitcast i32 %662 to <2 x bfloat>, !dbg !23 + %664 = extractvalue { i32, i32, i32, i32 } %659, 2, !dbg !23 + %665 = bitcast i32 %664 to <2 x bfloat>, !dbg !23 + %666 = extractvalue { i32, i32, i32, i32 } %659, 3, !dbg !23 + %667 = bitcast i32 %666 to <2 x bfloat>, !dbg !23 + %668 = extractelement <2 x bfloat> %661, i64 0, !dbg !23 + %669 = extractelement <2 x bfloat> %661, i64 1, !dbg !23 + %670 = extractelement <2 x bfloat> %663, i64 0, !dbg !23 + %671 = extractelement <2 x bfloat> %663, i64 1, !dbg !23 + %672 = extractelement <2 x bfloat> %665, i64 0, !dbg !23 + %673 = extractelement <2 x bfloat> %665, i64 1, !dbg !23 + %674 = extractelement <2 x bfloat> %667, i64 0, !dbg !23 + %675 = extractelement <2 x bfloat> %667, i64 1, !dbg !23 + %676 = extractelement <32 x i1> %40, i64 3, !dbg !23 + %677 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %168, i1 %676) #6, !dbg !23 + %678 = extractvalue { i32, i32, i32, i32 } %677, 0, !dbg !23 + %679 = bitcast i32 %678 to <2 x bfloat>, !dbg !23 + %680 = extractvalue { i32, i32, i32, i32 } %677, 1, !dbg !23 + %681 = bitcast i32 %680 to <2 x bfloat>, !dbg !23 + %682 = extractvalue { i32, i32, i32, i32 } %677, 2, !dbg !23 + %683 = bitcast i32 %682 to <2 x bfloat>, !dbg !23 + %684 = extractvalue { i32, i32, i32, i32 } %677, 3, !dbg !23 + %685 = bitcast i32 %684 to <2 x bfloat>, !dbg !23 + %686 = extractelement <2 x bfloat> %679, i64 0, !dbg !23 + %687 = extractelement <2 x bfloat> %679, i64 1, !dbg !23 + %688 = extractelement <2 x bfloat> %681, i64 0, !dbg !23 + %689 = extractelement <2 x bfloat> %681, i64 1, !dbg !23 + %690 = extractelement <2 x bfloat> %683, i64 0, !dbg !23 + %691 = extractelement <2 x bfloat> %683, i64 1, !dbg !23 + %692 = extractelement <2 x bfloat> %685, i64 0, !dbg !23 + %693 = extractelement <2 x bfloat> %685, i64 1, !dbg !23 + %694 = extractelement <32 x i1> %40, i64 2, !dbg !23 + %695 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %169, i1 %694) #6, !dbg !23 + %696 = extractvalue { i32, i32, i32, i32 } %695, 0, !dbg !23 + %697 = bitcast i32 %696 to <2 x bfloat>, !dbg !23 + %698 = extractvalue { i32, i32, i32, i32 } %695, 1, !dbg !23 + %699 = bitcast i32 %698 to <2 x bfloat>, !dbg !23 + %700 = extractvalue { i32, i32, i32, i32 } %695, 2, !dbg !23 + %701 = bitcast i32 %700 to <2 x bfloat>, !dbg !23 + %702 = extractvalue { i32, i32, i32, i32 } %695, 3, !dbg !23 + %703 = bitcast i32 %702 to <2 x bfloat>, !dbg !23 + %704 = extractelement <2 x bfloat> %697, i64 0, !dbg !23 + %705 = extractelement <2 x bfloat> %697, i64 1, !dbg !23 + %706 = extractelement <2 x bfloat> %699, i64 0, !dbg !23 + %707 = extractelement <2 x bfloat> %699, i64 1, !dbg !23 + %708 = extractelement <2 x bfloat> %701, i64 0, !dbg !23 + %709 = extractelement <2 x bfloat> %701, i64 1, !dbg !23 + %710 = extractelement <2 x bfloat> %703, i64 0, !dbg !23 + %711 = extractelement <2 x bfloat> %703, i64 1, !dbg !23 + %712 = extractelement <32 x i1> %40, i64 1, !dbg !23 + %713 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %170, i1 %712) #6, !dbg !23 + %714 = extractvalue { i32, i32, i32, i32 } %713, 0, !dbg !23 + %715 = bitcast i32 %714 to <2 x bfloat>, !dbg !23 + %716 = extractvalue { i32, i32, i32, i32 } %713, 1, !dbg !23 + %717 = bitcast i32 %716 to <2 x bfloat>, !dbg !23 + %718 = extractvalue { i32, i32, i32, i32 } %713, 2, !dbg !23 + %719 = bitcast i32 %718 to <2 x bfloat>, !dbg !23 + %720 = extractvalue { i32, i32, i32, i32 } %713, 3, !dbg !23 + %721 = bitcast i32 %720 to <2 x bfloat>, !dbg !23 + %722 = extractelement <2 x bfloat> %715, i64 0, !dbg !23 + %723 = extractelement <2 x bfloat> %715, i64 1, !dbg !23 + %724 = extractelement <2 x bfloat> %717, i64 0, !dbg !23 + %725 = extractelement <2 x bfloat> %717, i64 1, !dbg !23 + %726 = extractelement <2 x bfloat> %719, i64 0, !dbg !23 + %727 = extractelement <2 x bfloat> %719, i64 1, !dbg !23 + %728 = extractelement <2 x bfloat> %721, i64 0, !dbg !23 + %729 = extractelement <2 x bfloat> %721, i64 1, !dbg !23 + %730 = extractelement <32 x i1> %40, i64 0, !dbg !23 + %731 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %171, i1 %730) #6, !dbg !23 + %732 = extractvalue { i32, i32, i32, i32 } %731, 0, !dbg !23 + %733 = bitcast i32 %732 to <2 x bfloat>, !dbg !23 + %734 = extractvalue { i32, i32, i32, i32 } %731, 1, !dbg !23 + %735 = bitcast i32 %734 to <2 x bfloat>, !dbg !23 + %736 = extractvalue { i32, i32, i32, i32 } %731, 2, !dbg !23 + %737 = bitcast i32 %736 to <2 x bfloat>, !dbg !23 + %738 = extractvalue { i32, i32, i32, i32 } %731, 3, !dbg !23 + %739 = bitcast i32 %738 to <2 x bfloat>, !dbg !23 + %740 = extractelement <2 x bfloat> %733, i64 0, !dbg !23 + %741 = extractelement <2 x bfloat> %733, i64 1, !dbg !23 + %742 = extractelement <2 x bfloat> %735, i64 0, !dbg !23 + %743 = extractelement <2 x bfloat> %735, i64 1, !dbg !23 + %744 = extractelement <2 x bfloat> %737, i64 0, !dbg !23 + %745 = extractelement <2 x bfloat> %737, i64 1, !dbg !23 + %746 = extractelement <2 x bfloat> %739, i64 0, !dbg !23 + %747 = extractelement <2 x bfloat> %739, i64 1, !dbg !23 + %748 = fpext bfloat %182 to float, !dbg !24 + %749 = fpext bfloat %183 to float, !dbg !24 + %750 = fpext bfloat %184 to float, !dbg !24 + %751 = fpext bfloat %185 to float, !dbg !24 + %752 = fpext bfloat %186 to float, !dbg !24 + %753 = fpext bfloat %187 to float, !dbg !24 + %754 = fpext bfloat %188 to float, !dbg !24 + %755 = fpext bfloat %189 to float, !dbg !24 + %756 = fpext bfloat %200 to float, !dbg !24 + %757 = fpext bfloat %201 to float, !dbg !24 + %758 = fpext bfloat %202 to float, !dbg !24 + %759 = fpext bfloat %203 to float, !dbg !24 + %760 = fpext bfloat %204 to float, !dbg !24 + %761 = fpext bfloat %205 to float, !dbg !24 + %762 = fpext bfloat %206 to float, !dbg !24 + %763 = fpext bfloat %207 to float, !dbg !24 + %764 = fpext bfloat %218 to float, !dbg !24 + %765 = fpext bfloat %219 to float, !dbg !24 + %766 = fpext bfloat %220 to float, !dbg !24 + %767 = fpext bfloat %221 to float, !dbg !24 + %768 = fpext bfloat %222 to float, !dbg !24 + %769 = fpext bfloat %223 to float, !dbg !24 + %770 = fpext bfloat %224 to float, !dbg !24 + %771 = fpext bfloat %225 to float, !dbg !24 + %772 = fpext bfloat %236 to float, !dbg !24 + %773 = fpext bfloat %237 to float, !dbg !24 + %774 = fpext bfloat %238 to float, !dbg !24 + %775 = fpext bfloat %239 to float, !dbg !24 + %776 = fpext bfloat %240 to float, !dbg !24 + %777 = fpext bfloat %241 to float, !dbg !24 + %778 = fpext bfloat %242 to float, !dbg !24 + %779 = fpext bfloat %243 to float, !dbg !24 + %780 = fpext bfloat %254 to float, !dbg !24 + %781 = fpext bfloat %255 to float, !dbg !24 + %782 = fpext bfloat %256 to float, !dbg !24 + %783 = fpext bfloat %257 to float, !dbg !24 + %784 = fpext bfloat %258 to float, !dbg !24 + %785 = fpext bfloat %259 to float, !dbg !24 + %786 = fpext bfloat %260 to float, !dbg !24 + %787 = fpext bfloat %261 to float, !dbg !24 + %788 = fpext bfloat %272 to float, !dbg !24 + %789 = fpext bfloat %273 to float, !dbg !24 + %790 = fpext bfloat %274 to float, !dbg !24 + %791 = fpext bfloat %275 to float, !dbg !24 + %792 = fpext bfloat %276 to float, !dbg !24 + %793 = fpext bfloat %277 to float, !dbg !24 + %794 = fpext bfloat %278 to float, !dbg !24 + %795 = fpext bfloat %279 to float, !dbg !24 + %796 = fpext bfloat %290 to float, !dbg !24 + %797 = fpext bfloat %291 to float, !dbg !24 + %798 = fpext bfloat %292 to float, !dbg !24 + %799 = fpext bfloat %293 to float, !dbg !24 + %800 = fpext bfloat %294 to float, !dbg !24 + %801 = fpext bfloat %295 to float, !dbg !24 + %802 = fpext bfloat %296 to float, !dbg !24 + %803 = fpext bfloat %297 to float, !dbg !24 + %804 = fpext bfloat %308 to float, !dbg !24 + %805 = fpext bfloat %309 to float, !dbg !24 + %806 = fpext bfloat %310 to float, !dbg !24 + %807 = fpext bfloat %311 to float, !dbg !24 + %808 = fpext bfloat %312 to float, !dbg !24 + %809 = fpext bfloat %313 to float, !dbg !24 + %810 = fpext bfloat %314 to float, !dbg !24 + %811 = fpext bfloat %315 to float, !dbg !24 + %812 = fpext bfloat %326 to float, !dbg !24 + %813 = fpext bfloat %327 to float, !dbg !24 + %814 = fpext bfloat %328 to float, !dbg !24 + %815 = fpext bfloat %329 to float, !dbg !24 + %816 = fpext bfloat %330 to float, !dbg !24 + %817 = fpext bfloat %331 to float, !dbg !24 + %818 = fpext bfloat %332 to float, !dbg !24 + %819 = fpext bfloat %333 to float, !dbg !24 + %820 = fpext bfloat %344 to float, !dbg !24 + %821 = fpext bfloat %345 to float, !dbg !24 + %822 = fpext bfloat %346 to float, !dbg !24 + %823 = fpext bfloat %347 to float, !dbg !24 + %824 = fpext bfloat %348 to float, !dbg !24 + %825 = fpext bfloat %349 to float, !dbg !24 + %826 = fpext bfloat %350 to float, !dbg !24 + %827 = fpext bfloat %351 to float, !dbg !24 + %828 = fpext bfloat %362 to float, !dbg !24 + %829 = fpext bfloat %363 to float, !dbg !24 + %830 = fpext bfloat %364 to float, !dbg !24 + %831 = fpext bfloat %365 to float, !dbg !24 + %832 = fpext bfloat %366 to float, !dbg !24 + %833 = fpext bfloat %367 to float, !dbg !24 + %834 = fpext bfloat %368 to float, !dbg !24 + %835 = fpext bfloat %369 to float, !dbg !24 + %836 = fpext bfloat %380 to float, !dbg !24 + %837 = fpext bfloat %381 to float, !dbg !24 + %838 = fpext bfloat %382 to float, !dbg !24 + %839 = fpext bfloat %383 to float, !dbg !24 + %840 = fpext bfloat %384 to float, !dbg !24 + %841 = fpext bfloat %385 to float, !dbg !24 + %842 = fpext bfloat %386 to float, !dbg !24 + %843 = fpext bfloat %387 to float, !dbg !24 + %844 = fpext bfloat %398 to float, !dbg !24 + %845 = fpext bfloat %399 to float, !dbg !24 + %846 = fpext bfloat %400 to float, !dbg !24 + %847 = fpext bfloat %401 to float, !dbg !24 + %848 = fpext bfloat %402 to float, !dbg !24 + %849 = fpext bfloat %403 to float, !dbg !24 + %850 = fpext bfloat %404 to float, !dbg !24 + %851 = fpext bfloat %405 to float, !dbg !24 + %852 = fpext bfloat %416 to float, !dbg !24 + %853 = fpext bfloat %417 to float, !dbg !24 + %854 = fpext bfloat %418 to float, !dbg !24 + %855 = fpext bfloat %419 to float, !dbg !24 + %856 = fpext bfloat %420 to float, !dbg !24 + %857 = fpext bfloat %421 to float, !dbg !24 + %858 = fpext bfloat %422 to float, !dbg !24 + %859 = fpext bfloat %423 to float, !dbg !24 + %860 = fpext bfloat %434 to float, !dbg !24 + %861 = fpext bfloat %435 to float, !dbg !24 + %862 = fpext bfloat %436 to float, !dbg !24 + %863 = fpext bfloat %437 to float, !dbg !24 + %864 = fpext bfloat %438 to float, !dbg !24 + %865 = fpext bfloat %439 to float, !dbg !24 + %866 = fpext bfloat %440 to float, !dbg !24 + %867 = fpext bfloat %441 to float, !dbg !24 + %868 = fpext bfloat %452 to float, !dbg !24 + %869 = fpext bfloat %453 to float, !dbg !24 + %870 = fpext bfloat %454 to float, !dbg !24 + %871 = fpext bfloat %455 to float, !dbg !24 + %872 = fpext bfloat %456 to float, !dbg !24 + %873 = fpext bfloat %457 to float, !dbg !24 + %874 = fpext bfloat %458 to float, !dbg !24 + %875 = fpext bfloat %459 to float, !dbg !24 + %876 = fpext bfloat %470 to float, !dbg !24 + %877 = fpext bfloat %471 to float, !dbg !24 + %878 = fpext bfloat %472 to float, !dbg !24 + %879 = fpext bfloat %473 to float, !dbg !24 + %880 = fpext bfloat %474 to float, !dbg !24 + %881 = fpext bfloat %475 to float, !dbg !24 + %882 = fpext bfloat %476 to float, !dbg !24 + %883 = fpext bfloat %477 to float, !dbg !24 + %884 = fpext bfloat %488 to float, !dbg !24 + %885 = fpext bfloat %489 to float, !dbg !24 + %886 = fpext bfloat %490 to float, !dbg !24 + %887 = fpext bfloat %491 to float, !dbg !24 + %888 = fpext bfloat %492 to float, !dbg !24 + %889 = fpext bfloat %493 to float, !dbg !24 + %890 = fpext bfloat %494 to float, !dbg !24 + %891 = fpext bfloat %495 to float, !dbg !24 + %892 = fpext bfloat %506 to float, !dbg !24 + %893 = fpext bfloat %507 to float, !dbg !24 + %894 = fpext bfloat %508 to float, !dbg !24 + %895 = fpext bfloat %509 to float, !dbg !24 + %896 = fpext bfloat %510 to float, !dbg !24 + %897 = fpext bfloat %511 to float, !dbg !24 + %898 = fpext bfloat %512 to float, !dbg !24 + %899 = fpext bfloat %513 to float, !dbg !24 + %900 = fpext bfloat %524 to float, !dbg !24 + %901 = fpext bfloat %525 to float, !dbg !24 + %902 = fpext bfloat %526 to float, !dbg !24 + %903 = fpext bfloat %527 to float, !dbg !24 + %904 = fpext bfloat %528 to float, !dbg !24 + %905 = fpext bfloat %529 to float, !dbg !24 + %906 = fpext bfloat %530 to float, !dbg !24 + %907 = fpext bfloat %531 to float, !dbg !24 + %908 = fpext bfloat %542 to float, !dbg !24 + %909 = fpext bfloat %543 to float, !dbg !24 + %910 = fpext bfloat %544 to float, !dbg !24 + %911 = fpext bfloat %545 to float, !dbg !24 + %912 = fpext bfloat %546 to float, !dbg !24 + %913 = fpext bfloat %547 to float, !dbg !24 + %914 = fpext bfloat %548 to float, !dbg !24 + %915 = fpext bfloat %549 to float, !dbg !24 + %916 = fpext bfloat %560 to float, !dbg !24 + %917 = fpext bfloat %561 to float, !dbg !24 + %918 = fpext bfloat %562 to float, !dbg !24 + %919 = fpext bfloat %563 to float, !dbg !24 + %920 = fpext bfloat %564 to float, !dbg !24 + %921 = fpext bfloat %565 to float, !dbg !24 + %922 = fpext bfloat %566 to float, !dbg !24 + %923 = fpext bfloat %567 to float, !dbg !24 + %924 = fpext bfloat %578 to float, !dbg !24 + %925 = fpext bfloat %579 to float, !dbg !24 + %926 = fpext bfloat %580 to float, !dbg !24 + %927 = fpext bfloat %581 to float, !dbg !24 + %928 = fpext bfloat %582 to float, !dbg !24 + %929 = fpext bfloat %583 to float, !dbg !24 + %930 = fpext bfloat %584 to float, !dbg !24 + %931 = fpext bfloat %585 to float, !dbg !24 + %932 = fpext bfloat %596 to float, !dbg !24 + %933 = fpext bfloat %597 to float, !dbg !24 + %934 = fpext bfloat %598 to float, !dbg !24 + %935 = fpext bfloat %599 to float, !dbg !24 + %936 = fpext bfloat %600 to float, !dbg !24 + %937 = fpext bfloat %601 to float, !dbg !24 + %938 = fpext bfloat %602 to float, !dbg !24 + %939 = fpext bfloat %603 to float, !dbg !24 + %940 = fpext bfloat %614 to float, !dbg !24 + %941 = fpext bfloat %615 to float, !dbg !24 + %942 = fpext bfloat %616 to float, !dbg !24 + %943 = fpext bfloat %617 to float, !dbg !24 + %944 = fpext bfloat %618 to float, !dbg !24 + %945 = fpext bfloat %619 to float, !dbg !24 + %946 = fpext bfloat %620 to float, !dbg !24 + %947 = fpext bfloat %621 to float, !dbg !24 + %948 = fpext bfloat %632 to float, !dbg !24 + %949 = fpext bfloat %633 to float, !dbg !24 + %950 = fpext bfloat %634 to float, !dbg !24 + %951 = fpext bfloat %635 to float, !dbg !24 + %952 = fpext bfloat %636 to float, !dbg !24 + %953 = fpext bfloat %637 to float, !dbg !24 + %954 = fpext bfloat %638 to float, !dbg !24 + %955 = fpext bfloat %639 to float, !dbg !24 + %956 = fpext bfloat %650 to float, !dbg !24 + %957 = fpext bfloat %651 to float, !dbg !24 + %958 = fpext bfloat %652 to float, !dbg !24 + %959 = fpext bfloat %653 to float, !dbg !24 + %960 = fpext bfloat %654 to float, !dbg !24 + %961 = fpext bfloat %655 to float, !dbg !24 + %962 = fpext bfloat %656 to float, !dbg !24 + %963 = fpext bfloat %657 to float, !dbg !24 + %964 = fpext bfloat %668 to float, !dbg !24 + %965 = fpext bfloat %669 to float, !dbg !24 + %966 = fpext bfloat %670 to float, !dbg !24 + %967 = fpext bfloat %671 to float, !dbg !24 + %968 = fpext bfloat %672 to float, !dbg !24 + %969 = fpext bfloat %673 to float, !dbg !24 + %970 = fpext bfloat %674 to float, !dbg !24 + %971 = fpext bfloat %675 to float, !dbg !24 + %972 = fpext bfloat %686 to float, !dbg !24 + %973 = fpext bfloat %687 to float, !dbg !24 + %974 = fpext bfloat %688 to float, !dbg !24 + %975 = fpext bfloat %689 to float, !dbg !24 + %976 = fpext bfloat %690 to float, !dbg !24 + %977 = fpext bfloat %691 to float, !dbg !24 + %978 = fpext bfloat %692 to float, !dbg !24 + %979 = fpext bfloat %693 to float, !dbg !24 + %980 = fpext bfloat %704 to float, !dbg !24 + %981 = fpext bfloat %705 to float, !dbg !24 + %982 = fpext bfloat %706 to float, !dbg !24 + %983 = fpext bfloat %707 to float, !dbg !24 + %984 = fpext bfloat %708 to float, !dbg !24 + %985 = fpext bfloat %709 to float, !dbg !24 + %986 = fpext bfloat %710 to float, !dbg !24 + %987 = fpext bfloat %711 to float, !dbg !24 + %988 = fpext bfloat %722 to float, !dbg !24 + %989 = fpext bfloat %723 to float, !dbg !24 + %990 = fpext bfloat %724 to float, !dbg !24 + %991 = fpext bfloat %725 to float, !dbg !24 + %992 = fpext bfloat %726 to float, !dbg !24 + %993 = fpext bfloat %727 to float, !dbg !24 + %994 = fpext bfloat %728 to float, !dbg !24 + %995 = fpext bfloat %729 to float, !dbg !24 + %996 = fpext bfloat %740 to float, !dbg !24 + %997 = fpext bfloat %741 to float, !dbg !24 + %998 = fpext bfloat %742 to float, !dbg !24 + %999 = fpext bfloat %743 to float, !dbg !24 + %1000 = fpext bfloat %744 to float, !dbg !24 + %1001 = fpext bfloat %745 to float, !dbg !24 + %1002 = fpext bfloat %746 to float, !dbg !24 + %1003 = fpext bfloat %747 to float, !dbg !24 + %1004 = fmul float %748, %748, !dbg !25 + %1005 = fmul float %749, %749, !dbg !25 + %1006 = fmul float %750, %750, !dbg !25 + %1007 = fmul float %751, %751, !dbg !25 + %1008 = fmul float %752, %752, !dbg !25 + %1009 = fmul float %753, %753, !dbg !25 + %1010 = fmul float %754, %754, !dbg !25 + %1011 = fmul float %755, %755, !dbg !25 + %1012 = fmul float %756, %756, !dbg !25 + %1013 = fmul float %757, %757, !dbg !25 + %1014 = fmul float %758, %758, !dbg !25 + %1015 = fmul float %759, %759, !dbg !25 + %1016 = fmul float %760, %760, !dbg !25 + %1017 = fmul float %761, %761, !dbg !25 + %1018 = fmul float %762, %762, !dbg !25 + %1019 = fmul float %763, %763, !dbg !25 + %1020 = fmul float %764, %764, !dbg !25 + %1021 = fmul float %765, %765, !dbg !25 + %1022 = fmul float %766, %766, !dbg !25 + %1023 = fmul float %767, %767, !dbg !25 + %1024 = fmul float %768, %768, !dbg !25 + %1025 = fmul float %769, %769, !dbg !25 + %1026 = fmul float %770, %770, !dbg !25 + %1027 = fmul float %771, %771, !dbg !25 + %1028 = fmul float %772, %772, !dbg !25 + %1029 = fmul float %773, %773, !dbg !25 + %1030 = fmul float %774, %774, !dbg !25 + %1031 = fmul float %775, %775, !dbg !25 + %1032 = fmul float %776, %776, !dbg !25 + %1033 = fmul float %777, %777, !dbg !25 + %1034 = fmul float %778, %778, !dbg !25 + %1035 = fmul float %779, %779, !dbg !25 + %1036 = fmul float %780, %780, !dbg !25 + %1037 = fmul float %781, %781, !dbg !25 + %1038 = fmul float %782, %782, !dbg !25 + %1039 = fmul float %783, %783, !dbg !25 + %1040 = fmul float %784, %784, !dbg !25 + %1041 = fmul float %785, %785, !dbg !25 + %1042 = fmul float %786, %786, !dbg !25 + %1043 = fmul float %787, %787, !dbg !25 + %1044 = fmul float %788, %788, !dbg !25 + %1045 = fmul float %789, %789, !dbg !25 + %1046 = fmul float %790, %790, !dbg !25 + %1047 = fmul float %791, %791, !dbg !25 + %1048 = fmul float %792, %792, !dbg !25 + %1049 = fmul float %793, %793, !dbg !25 + %1050 = fmul float %794, %794, !dbg !25 + %1051 = fmul float %795, %795, !dbg !25 + %1052 = fmul float %796, %796, !dbg !25 + %1053 = fmul float %797, %797, !dbg !25 + %1054 = fmul float %798, %798, !dbg !25 + %1055 = fmul float %799, %799, !dbg !25 + %1056 = fmul float %800, %800, !dbg !25 + %1057 = fmul float %801, %801, !dbg !25 + %1058 = fmul float %802, %802, !dbg !25 + %1059 = fmul float %803, %803, !dbg !25 + %1060 = fmul float %804, %804, !dbg !25 + %1061 = fmul float %805, %805, !dbg !25 + %1062 = fmul float %806, %806, !dbg !25 + %1063 = fmul float %807, %807, !dbg !25 + %1064 = fmul float %808, %808, !dbg !25 + %1065 = fmul float %809, %809, !dbg !25 + %1066 = fmul float %810, %810, !dbg !25 + %1067 = fmul float %811, %811, !dbg !25 + %1068 = fmul float %812, %812, !dbg !25 + %1069 = fmul float %813, %813, !dbg !25 + %1070 = fmul float %814, %814, !dbg !25 + %1071 = fmul float %815, %815, !dbg !25 + %1072 = fmul float %816, %816, !dbg !25 + %1073 = fmul float %817, %817, !dbg !25 + %1074 = fmul float %818, %818, !dbg !25 + %1075 = fmul float %819, %819, !dbg !25 + %1076 = fmul float %820, %820, !dbg !25 + %1077 = fmul float %821, %821, !dbg !25 + %1078 = fmul float %822, %822, !dbg !25 + %1079 = fmul float %823, %823, !dbg !25 + %1080 = fmul float %824, %824, !dbg !25 + %1081 = fmul float %825, %825, !dbg !25 + %1082 = fmul float %826, %826, !dbg !25 + %1083 = fmul float %827, %827, !dbg !25 + %1084 = fmul float %828, %828, !dbg !25 + %1085 = fmul float %829, %829, !dbg !25 + %1086 = fmul float %830, %830, !dbg !25 + %1087 = fmul float %831, %831, !dbg !25 + %1088 = fmul float %832, %832, !dbg !25 + %1089 = fmul float %833, %833, !dbg !25 + %1090 = fmul float %834, %834, !dbg !25 + %1091 = fmul float %835, %835, !dbg !25 + %1092 = fmul float %836, %836, !dbg !25 + %1093 = fmul float %837, %837, !dbg !25 + %1094 = fmul float %838, %838, !dbg !25 + %1095 = fmul float %839, %839, !dbg !25 + %1096 = fmul float %840, %840, !dbg !25 + %1097 = fmul float %841, %841, !dbg !25 + %1098 = fmul float %842, %842, !dbg !25 + %1099 = fmul float %843, %843, !dbg !25 + %1100 = fmul float %844, %844, !dbg !25 + %1101 = fmul float %845, %845, !dbg !25 + %1102 = fmul float %846, %846, !dbg !25 + %1103 = fmul float %847, %847, !dbg !25 + %1104 = fmul float %848, %848, !dbg !25 + %1105 = fmul float %849, %849, !dbg !25 + %1106 = fmul float %850, %850, !dbg !25 + %1107 = fmul float %851, %851, !dbg !25 + %1108 = fmul float %852, %852, !dbg !25 + %1109 = fmul float %853, %853, !dbg !25 + %1110 = fmul float %854, %854, !dbg !25 + %1111 = fmul float %855, %855, !dbg !25 + %1112 = fmul float %856, %856, !dbg !25 + %1113 = fmul float %857, %857, !dbg !25 + %1114 = fmul float %858, %858, !dbg !25 + %1115 = fmul float %859, %859, !dbg !25 + %1116 = fmul float %860, %860, !dbg !25 + %1117 = fmul float %861, %861, !dbg !25 + %1118 = fmul float %862, %862, !dbg !25 + %1119 = fmul float %863, %863, !dbg !25 + %1120 = fmul float %864, %864, !dbg !25 + %1121 = fmul float %865, %865, !dbg !25 + %1122 = fmul float %866, %866, !dbg !25 + %1123 = fmul float %867, %867, !dbg !25 + %1124 = fmul float %868, %868, !dbg !25 + %1125 = fmul float %869, %869, !dbg !25 + %1126 = fmul float %870, %870, !dbg !25 + %1127 = fmul float %871, %871, !dbg !25 + %1128 = fmul float %872, %872, !dbg !25 + %1129 = fmul float %873, %873, !dbg !25 + %1130 = fmul float %874, %874, !dbg !25 + %1131 = fmul float %875, %875, !dbg !25 + %1132 = fmul float %876, %876, !dbg !25 + %1133 = fmul float %877, %877, !dbg !25 + %1134 = fmul float %878, %878, !dbg !25 + %1135 = fmul float %879, %879, !dbg !25 + %1136 = fmul float %880, %880, !dbg !25 + %1137 = fmul float %881, %881, !dbg !25 + %1138 = fmul float %882, %882, !dbg !25 + %1139 = fmul float %883, %883, !dbg !25 + %1140 = fmul float %884, %884, !dbg !25 + %1141 = fmul float %885, %885, !dbg !25 + %1142 = fmul float %886, %886, !dbg !25 + %1143 = fmul float %887, %887, !dbg !25 + %1144 = fmul float %888, %888, !dbg !25 + %1145 = fmul float %889, %889, !dbg !25 + %1146 = fmul float %890, %890, !dbg !25 + %1147 = fmul float %891, %891, !dbg !25 + %1148 = fmul float %892, %892, !dbg !25 + %1149 = fmul float %893, %893, !dbg !25 + %1150 = fmul float %894, %894, !dbg !25 + %1151 = fmul float %895, %895, !dbg !25 + %1152 = fmul float %896, %896, !dbg !25 + %1153 = fmul float %897, %897, !dbg !25 + %1154 = fmul float %898, %898, !dbg !25 + %1155 = fmul float %899, %899, !dbg !25 + %1156 = fmul float %900, %900, !dbg !25 + %1157 = fmul float %901, %901, !dbg !25 + %1158 = fmul float %902, %902, !dbg !25 + %1159 = fmul float %903, %903, !dbg !25 + %1160 = fmul float %904, %904, !dbg !25 + %1161 = fmul float %905, %905, !dbg !25 + %1162 = fmul float %906, %906, !dbg !25 + %1163 = fmul float %907, %907, !dbg !25 + %1164 = fmul float %908, %908, !dbg !25 + %1165 = fmul float %909, %909, !dbg !25 + %1166 = fmul float %910, %910, !dbg !25 + %1167 = fmul float %911, %911, !dbg !25 + %1168 = fmul float %912, %912, !dbg !25 + %1169 = fmul float %913, %913, !dbg !25 + %1170 = fmul float %914, %914, !dbg !25 + %1171 = fmul float %915, %915, !dbg !25 + %1172 = fmul float %916, %916, !dbg !25 + %1173 = fmul float %917, %917, !dbg !25 + %1174 = fmul float %918, %918, !dbg !25 + %1175 = fmul float %919, %919, !dbg !25 + %1176 = fmul float %920, %920, !dbg !25 + %1177 = fmul float %921, %921, !dbg !25 + %1178 = fmul float %922, %922, !dbg !25 + %1179 = fmul float %923, %923, !dbg !25 + %1180 = fmul float %924, %924, !dbg !25 + %1181 = fmul float %925, %925, !dbg !25 + %1182 = fmul float %926, %926, !dbg !25 + %1183 = fmul float %927, %927, !dbg !25 + %1184 = fmul float %928, %928, !dbg !25 + %1185 = fmul float %929, %929, !dbg !25 + %1186 = fmul float %930, %930, !dbg !25 + %1187 = fmul float %931, %931, !dbg !25 + %1188 = fmul float %932, %932, !dbg !25 + %1189 = fmul float %933, %933, !dbg !25 + %1190 = fmul float %934, %934, !dbg !25 + %1191 = fmul float %935, %935, !dbg !25 + %1192 = fmul float %936, %936, !dbg !25 + %1193 = fmul float %937, %937, !dbg !25 + %1194 = fmul float %938, %938, !dbg !25 + %1195 = fmul float %939, %939, !dbg !25 + %1196 = fmul float %940, %940, !dbg !25 + %1197 = fmul float %941, %941, !dbg !25 + %1198 = fmul float %942, %942, !dbg !25 + %1199 = fmul float %943, %943, !dbg !25 + %1200 = fmul float %944, %944, !dbg !25 + %1201 = fmul float %945, %945, !dbg !25 + %1202 = fmul float %946, %946, !dbg !25 + %1203 = fmul float %947, %947, !dbg !25 + %1204 = fmul float %948, %948, !dbg !25 + %1205 = fmul float %949, %949, !dbg !25 + %1206 = fmul float %950, %950, !dbg !25 + %1207 = fmul float %951, %951, !dbg !25 + %1208 = fmul float %952, %952, !dbg !25 + %1209 = fmul float %953, %953, !dbg !25 + %1210 = fmul float %954, %954, !dbg !25 + %1211 = fmul float %955, %955, !dbg !25 + %1212 = fmul float %956, %956, !dbg !25 + %1213 = fmul float %957, %957, !dbg !25 + %1214 = fmul float %958, %958, !dbg !25 + %1215 = fmul float %959, %959, !dbg !25 + %1216 = fmul float %960, %960, !dbg !25 + %1217 = fmul float %961, %961, !dbg !25 + %1218 = fmul float %962, %962, !dbg !25 + %1219 = fmul float %963, %963, !dbg !25 + %1220 = fmul float %964, %964, !dbg !25 + %1221 = fmul float %965, %965, !dbg !25 + %1222 = fmul float %966, %966, !dbg !25 + %1223 = fmul float %967, %967, !dbg !25 + %1224 = fmul float %968, %968, !dbg !25 + %1225 = fmul float %969, %969, !dbg !25 + %1226 = fmul float %970, %970, !dbg !25 + %1227 = fmul float %971, %971, !dbg !25 + %1228 = fmul float %972, %972, !dbg !25 + %1229 = fmul float %973, %973, !dbg !25 + %1230 = fmul float %974, %974, !dbg !25 + %1231 = fmul float %975, %975, !dbg !25 + %1232 = fmul float %976, %976, !dbg !25 + %1233 = fmul float %977, %977, !dbg !25 + %1234 = fmul float %978, %978, !dbg !25 + %1235 = fmul float %979, %979, !dbg !25 + %1236 = fmul float %980, %980, !dbg !25 + %1237 = fmul float %981, %981, !dbg !25 + %1238 = fmul float %982, %982, !dbg !25 + %1239 = fmul float %983, %983, !dbg !25 + %1240 = fmul float %984, %984, !dbg !25 + %1241 = fmul float %985, %985, !dbg !25 + %1242 = fmul float %986, %986, !dbg !25 + %1243 = fmul float %987, %987, !dbg !25 + %1244 = fmul float %988, %988, !dbg !25 + %1245 = fmul float %989, %989, !dbg !25 + %1246 = fmul float %990, %990, !dbg !25 + %1247 = fmul float %991, %991, !dbg !25 + %1248 = fmul float %992, %992, !dbg !25 + %1249 = fmul float %993, %993, !dbg !25 + %1250 = fmul float %994, %994, !dbg !25 + %1251 = fmul float %995, %995, !dbg !25 + %1252 = fmul float %996, %996, !dbg !25 + %1253 = fmul float %997, %997, !dbg !25 + %1254 = fmul float %998, %998, !dbg !25 + %1255 = fmul float %999, %999, !dbg !25 + %1256 = fmul float %1000, %1000, !dbg !25 + %1257 = fmul float %1001, %1001, !dbg !25 + %1258 = fmul float %1002, %1002, !dbg !25 + %1259 = fmul float %1003, %1003, !dbg !25 + %1260 = fadd float %1004, %1005, !dbg !26 + %1261 = fadd float %1006, %1260, !dbg !26 + %1262 = fadd float %1007, %1261, !dbg !26 + %1263 = fadd float %1008, %1262, !dbg !26 + %1264 = fadd float %1009, %1263, !dbg !26 + %1265 = fadd float %1010, %1264, !dbg !26 + %1266 = fadd float %1011, %1265, !dbg !26 + %1267 = fadd float %1012, %1013, !dbg !26 + %1268 = fadd float %1014, %1267, !dbg !26 + %1269 = fadd float %1015, %1268, !dbg !26 + %1270 = fadd float %1016, %1269, !dbg !26 + %1271 = fadd float %1017, %1270, !dbg !26 + %1272 = fadd float %1018, %1271, !dbg !26 + %1273 = fadd float %1019, %1272, !dbg !26 + %1274 = fadd float %1020, %1021, !dbg !26 + %1275 = fadd float %1022, %1274, !dbg !26 + %1276 = fadd float %1023, %1275, !dbg !26 + %1277 = fadd float %1024, %1276, !dbg !26 + %1278 = fadd float %1025, %1277, !dbg !26 + %1279 = fadd float %1026, %1278, !dbg !26 + %1280 = fadd float %1027, %1279, !dbg !26 + %1281 = fadd float %1028, %1029, !dbg !26 + %1282 = fadd float %1030, %1281, !dbg !26 + %1283 = fadd float %1031, %1282, !dbg !26 + %1284 = fadd float %1032, %1283, !dbg !26 + %1285 = fadd float %1033, %1284, !dbg !26 + %1286 = fadd float %1034, %1285, !dbg !26 + %1287 = fadd float %1035, %1286, !dbg !26 + %1288 = fadd float %1036, %1037, !dbg !26 + %1289 = fadd float %1038, %1288, !dbg !26 + %1290 = fadd float %1039, %1289, !dbg !26 + %1291 = fadd float %1040, %1290, !dbg !26 + %1292 = fadd float %1041, %1291, !dbg !26 + %1293 = fadd float %1042, %1292, !dbg !26 + %1294 = fadd float %1043, %1293, !dbg !26 + %1295 = fadd float %1044, %1045, !dbg !26 + %1296 = fadd float %1046, %1295, !dbg !26 + %1297 = fadd float %1047, %1296, !dbg !26 + %1298 = fadd float %1048, %1297, !dbg !26 + %1299 = fadd float %1049, %1298, !dbg !26 + %1300 = fadd float %1050, %1299, !dbg !26 + %1301 = fadd float %1051, %1300, !dbg !26 + %1302 = fadd float %1052, %1053, !dbg !26 + %1303 = fadd float %1054, %1302, !dbg !26 + %1304 = fadd float %1055, %1303, !dbg !26 + %1305 = fadd float %1056, %1304, !dbg !26 + %1306 = fadd float %1057, %1305, !dbg !26 + %1307 = fadd float %1058, %1306, !dbg !26 + %1308 = fadd float %1059, %1307, !dbg !26 + %1309 = fadd float %1060, %1061, !dbg !26 + %1310 = fadd float %1062, %1309, !dbg !26 + %1311 = fadd float %1063, %1310, !dbg !26 + %1312 = fadd float %1064, %1311, !dbg !26 + %1313 = fadd float %1065, %1312, !dbg !26 + %1314 = fadd float %1066, %1313, !dbg !26 + %1315 = fadd float %1067, %1314, !dbg !26 + %1316 = fadd float %1068, %1069, !dbg !26 + %1317 = fadd float %1070, %1316, !dbg !26 + %1318 = fadd float %1071, %1317, !dbg !26 + %1319 = fadd float %1072, %1318, !dbg !26 + %1320 = fadd float %1073, %1319, !dbg !26 + %1321 = fadd float %1074, %1320, !dbg !26 + %1322 = fadd float %1075, %1321, !dbg !26 + %1323 = fadd float %1076, %1077, !dbg !26 + %1324 = fadd float %1078, %1323, !dbg !26 + %1325 = fadd float %1079, %1324, !dbg !26 + %1326 = fadd float %1080, %1325, !dbg !26 + %1327 = fadd float %1081, %1326, !dbg !26 + %1328 = fadd float %1082, %1327, !dbg !26 + %1329 = fadd float %1083, %1328, !dbg !26 + %1330 = fadd float %1084, %1085, !dbg !26 + %1331 = fadd float %1086, %1330, !dbg !26 + %1332 = fadd float %1087, %1331, !dbg !26 + %1333 = fadd float %1088, %1332, !dbg !26 + %1334 = fadd float %1089, %1333, !dbg !26 + %1335 = fadd float %1090, %1334, !dbg !26 + %1336 = fadd float %1091, %1335, !dbg !26 + %1337 = fadd float %1092, %1093, !dbg !26 + %1338 = fadd float %1094, %1337, !dbg !26 + %1339 = fadd float %1095, %1338, !dbg !26 + %1340 = fadd float %1096, %1339, !dbg !26 + %1341 = fadd float %1097, %1340, !dbg !26 + %1342 = fadd float %1098, %1341, !dbg !26 + %1343 = fadd float %1099, %1342, !dbg !26 + %1344 = fadd float %1100, %1101, !dbg !26 + %1345 = fadd float %1102, %1344, !dbg !26 + %1346 = fadd float %1103, %1345, !dbg !26 + %1347 = fadd float %1104, %1346, !dbg !26 + %1348 = fadd float %1105, %1347, !dbg !26 + %1349 = fadd float %1106, %1348, !dbg !26 + %1350 = fadd float %1107, %1349, !dbg !26 + %1351 = fadd float %1108, %1109, !dbg !26 + %1352 = fadd float %1110, %1351, !dbg !26 + %1353 = fadd float %1111, %1352, !dbg !26 + %1354 = fadd float %1112, %1353, !dbg !26 + %1355 = fadd float %1113, %1354, !dbg !26 + %1356 = fadd float %1114, %1355, !dbg !26 + %1357 = fadd float %1115, %1356, !dbg !26 + %1358 = fadd float %1116, %1117, !dbg !26 + %1359 = fadd float %1118, %1358, !dbg !26 + %1360 = fadd float %1119, %1359, !dbg !26 + %1361 = fadd float %1120, %1360, !dbg !26 + %1362 = fadd float %1121, %1361, !dbg !26 + %1363 = fadd float %1122, %1362, !dbg !26 + %1364 = fadd float %1123, %1363, !dbg !26 + %1365 = fadd float %1124, %1125, !dbg !26 + %1366 = fadd float %1126, %1365, !dbg !26 + %1367 = fadd float %1127, %1366, !dbg !26 + %1368 = fadd float %1128, %1367, !dbg !26 + %1369 = fadd float %1129, %1368, !dbg !26 + %1370 = fadd float %1130, %1369, !dbg !26 + %1371 = fadd float %1131, %1370, !dbg !26 + %1372 = fadd float %1132, %1133, !dbg !26 + %1373 = fadd float %1134, %1372, !dbg !26 + %1374 = fadd float %1135, %1373, !dbg !26 + %1375 = fadd float %1136, %1374, !dbg !26 + %1376 = fadd float %1137, %1375, !dbg !26 + %1377 = fadd float %1138, %1376, !dbg !26 + %1378 = fadd float %1139, %1377, !dbg !26 + %1379 = fadd float %1140, %1141, !dbg !26 + %1380 = fadd float %1142, %1379, !dbg !26 + %1381 = fadd float %1143, %1380, !dbg !26 + %1382 = fadd float %1144, %1381, !dbg !26 + %1383 = fadd float %1145, %1382, !dbg !26 + %1384 = fadd float %1146, %1383, !dbg !26 + %1385 = fadd float %1147, %1384, !dbg !26 + %1386 = fadd float %1148, %1149, !dbg !26 + %1387 = fadd float %1150, %1386, !dbg !26 + %1388 = fadd float %1151, %1387, !dbg !26 + %1389 = fadd float %1152, %1388, !dbg !26 + %1390 = fadd float %1153, %1389, !dbg !26 + %1391 = fadd float %1154, %1390, !dbg !26 + %1392 = fadd float %1155, %1391, !dbg !26 + %1393 = fadd float %1156, %1157, !dbg !26 + %1394 = fadd float %1158, %1393, !dbg !26 + %1395 = fadd float %1159, %1394, !dbg !26 + %1396 = fadd float %1160, %1395, !dbg !26 + %1397 = fadd float %1161, %1396, !dbg !26 + %1398 = fadd float %1162, %1397, !dbg !26 + %1399 = fadd float %1163, %1398, !dbg !26 + %1400 = fadd float %1164, %1165, !dbg !26 + %1401 = fadd float %1166, %1400, !dbg !26 + %1402 = fadd float %1167, %1401, !dbg !26 + %1403 = fadd float %1168, %1402, !dbg !26 + %1404 = fadd float %1169, %1403, !dbg !26 + %1405 = fadd float %1170, %1404, !dbg !26 + %1406 = fadd float %1171, %1405, !dbg !26 + %1407 = fadd float %1172, %1173, !dbg !26 + %1408 = fadd float %1174, %1407, !dbg !26 + %1409 = fadd float %1175, %1408, !dbg !26 + %1410 = fadd float %1176, %1409, !dbg !26 + %1411 = fadd float %1177, %1410, !dbg !26 + %1412 = fadd float %1178, %1411, !dbg !26 + %1413 = fadd float %1179, %1412, !dbg !26 + %1414 = fadd float %1180, %1181, !dbg !26 + %1415 = fadd float %1182, %1414, !dbg !26 + %1416 = fadd float %1183, %1415, !dbg !26 + %1417 = fadd float %1184, %1416, !dbg !26 + %1418 = fadd float %1185, %1417, !dbg !26 + %1419 = fadd float %1186, %1418, !dbg !26 + %1420 = fadd float %1187, %1419, !dbg !26 + %1421 = fadd float %1188, %1189, !dbg !26 + %1422 = fadd float %1190, %1421, !dbg !26 + %1423 = fadd float %1191, %1422, !dbg !26 + %1424 = fadd float %1192, %1423, !dbg !26 + %1425 = fadd float %1193, %1424, !dbg !26 + %1426 = fadd float %1194, %1425, !dbg !26 + %1427 = fadd float %1195, %1426, !dbg !26 + %1428 = fadd float %1196, %1197, !dbg !26 + %1429 = fadd float %1198, %1428, !dbg !26 + %1430 = fadd float %1199, %1429, !dbg !26 + %1431 = fadd float %1200, %1430, !dbg !26 + %1432 = fadd float %1201, %1431, !dbg !26 + %1433 = fadd float %1202, %1432, !dbg !26 + %1434 = fadd float %1203, %1433, !dbg !26 + %1435 = fadd float %1204, %1205, !dbg !26 + %1436 = fadd float %1206, %1435, !dbg !26 + %1437 = fadd float %1207, %1436, !dbg !26 + %1438 = fadd float %1208, %1437, !dbg !26 + %1439 = fadd float %1209, %1438, !dbg !26 + %1440 = fadd float %1210, %1439, !dbg !26 + %1441 = fadd float %1211, %1440, !dbg !26 + %1442 = fadd float %1212, %1213, !dbg !26 + %1443 = fadd float %1214, %1442, !dbg !26 + %1444 = fadd float %1215, %1443, !dbg !26 + %1445 = fadd float %1216, %1444, !dbg !26 + %1446 = fadd float %1217, %1445, !dbg !26 + %1447 = fadd float %1218, %1446, !dbg !26 + %1448 = fadd float %1219, %1447, !dbg !26 + %1449 = fadd float %1220, %1221, !dbg !26 + %1450 = fadd float %1222, %1449, !dbg !26 + %1451 = fadd float %1223, %1450, !dbg !26 + %1452 = fadd float %1224, %1451, !dbg !26 + %1453 = fadd float %1225, %1452, !dbg !26 + %1454 = fadd float %1226, %1453, !dbg !26 + %1455 = fadd float %1227, %1454, !dbg !26 + %1456 = fadd float %1228, %1229, !dbg !26 + %1457 = fadd float %1230, %1456, !dbg !26 + %1458 = fadd float %1231, %1457, !dbg !26 + %1459 = fadd float %1232, %1458, !dbg !26 + %1460 = fadd float %1233, %1459, !dbg !26 + %1461 = fadd float %1234, %1460, !dbg !26 + %1462 = fadd float %1235, %1461, !dbg !26 + %1463 = fadd float %1236, %1237, !dbg !26 + %1464 = fadd float %1238, %1463, !dbg !26 + %1465 = fadd float %1239, %1464, !dbg !26 + %1466 = fadd float %1240, %1465, !dbg !26 + %1467 = fadd float %1241, %1466, !dbg !26 + %1468 = fadd float %1242, %1467, !dbg !26 + %1469 = fadd float %1243, %1468, !dbg !26 + %1470 = fadd float %1244, %1245, !dbg !26 + %1471 = fadd float %1246, %1470, !dbg !26 + %1472 = fadd float %1247, %1471, !dbg !26 + %1473 = fadd float %1248, %1472, !dbg !26 + %1474 = fadd float %1249, %1473, !dbg !26 + %1475 = fadd float %1250, %1474, !dbg !26 + %1476 = fadd float %1251, %1475, !dbg !26 + %1477 = fadd float %1252, %1253, !dbg !26 + %1478 = fadd float %1254, %1477, !dbg !26 + %1479 = fadd float %1255, %1478, !dbg !26 + %1480 = fadd float %1256, %1479, !dbg !26 + %1481 = fadd float %1257, %1480, !dbg !26 + %1482 = fadd float %1258, %1481, !dbg !26 + %1483 = fadd float %1259, %1482, !dbg !26 + %1484 = bitcast float %1266 to i32, !dbg !29 + %1485 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1484, i32 8, i32 31), !dbg !29 + %1486 = bitcast i32 %1485 to float, !dbg !29 + %1487 = fadd float %1266, %1486, !dbg !26 + %1488 = bitcast float %1487 to i32, !dbg !29 + %1489 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1488, i32 4, i32 31), !dbg !29 + %1490 = bitcast i32 %1489 to float, !dbg !29 + %1491 = fadd float %1487, %1490, !dbg !26 + %1492 = bitcast float %1491 to i32, !dbg !29 + %1493 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1492, i32 2, i32 31), !dbg !29 + %1494 = bitcast i32 %1493 to float, !dbg !29 + %1495 = fadd float %1491, %1494, !dbg !26 + %1496 = bitcast float %1495 to i32, !dbg !29 + %1497 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1496, i32 1, i32 31), !dbg !29 + %1498 = bitcast i32 %1497 to float, !dbg !29 + %1499 = fadd float %1495, %1498, !dbg !26 + %1500 = bitcast float %1273 to i32, !dbg !29 + %1501 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1500, i32 8, i32 31), !dbg !29 + %1502 = bitcast i32 %1501 to float, !dbg !29 + %1503 = fadd float %1273, %1502, !dbg !26 + %1504 = bitcast float %1503 to i32, !dbg !29 + %1505 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1504, i32 4, i32 31), !dbg !29 + %1506 = bitcast i32 %1505 to float, !dbg !29 + %1507 = fadd float %1503, %1506, !dbg !26 + %1508 = bitcast float %1507 to i32, !dbg !29 + %1509 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1508, i32 2, i32 31), !dbg !29 + %1510 = bitcast i32 %1509 to float, !dbg !29 + %1511 = fadd float %1507, %1510, !dbg !26 + %1512 = bitcast float %1511 to i32, !dbg !29 + %1513 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1512, i32 1, i32 31), !dbg !29 + %1514 = bitcast i32 %1513 to float, !dbg !29 + %1515 = fadd float %1511, %1514, !dbg !26 + %1516 = bitcast float %1280 to i32, !dbg !29 + %1517 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1516, i32 8, i32 31), !dbg !29 + %1518 = bitcast i32 %1517 to float, !dbg !29 + %1519 = fadd float %1280, %1518, !dbg !26 + %1520 = bitcast float %1519 to i32, !dbg !29 + %1521 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1520, i32 4, i32 31), !dbg !29 + %1522 = bitcast i32 %1521 to float, !dbg !29 + %1523 = fadd float %1519, %1522, !dbg !26 + %1524 = bitcast float %1523 to i32, !dbg !29 + %1525 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1524, i32 2, i32 31), !dbg !29 + %1526 = bitcast i32 %1525 to float, !dbg !29 + %1527 = fadd float %1523, %1526, !dbg !26 + %1528 = bitcast float %1527 to i32, !dbg !29 + %1529 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1528, i32 1, i32 31), !dbg !29 + %1530 = bitcast i32 %1529 to float, !dbg !29 + %1531 = fadd float %1527, %1530, !dbg !26 + %1532 = bitcast float %1287 to i32, !dbg !29 + %1533 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1532, i32 8, i32 31), !dbg !29 + %1534 = bitcast i32 %1533 to float, !dbg !29 + %1535 = fadd float %1287, %1534, !dbg !26 + %1536 = bitcast float %1535 to i32, !dbg !29 + %1537 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1536, i32 4, i32 31), !dbg !29 + %1538 = bitcast i32 %1537 to float, !dbg !29 + %1539 = fadd float %1535, %1538, !dbg !26 + %1540 = bitcast float %1539 to i32, !dbg !29 + %1541 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1540, i32 2, i32 31), !dbg !29 + %1542 = bitcast i32 %1541 to float, !dbg !29 + %1543 = fadd float %1539, %1542, !dbg !26 + %1544 = bitcast float %1543 to i32, !dbg !29 + %1545 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1544, i32 1, i32 31), !dbg !29 + %1546 = bitcast i32 %1545 to float, !dbg !29 + %1547 = fadd float %1543, %1546, !dbg !26 + %1548 = bitcast float %1294 to i32, !dbg !29 + %1549 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1548, i32 8, i32 31), !dbg !29 + %1550 = bitcast i32 %1549 to float, !dbg !29 + %1551 = fadd float %1294, %1550, !dbg !26 + %1552 = bitcast float %1551 to i32, !dbg !29 + %1553 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1552, i32 4, i32 31), !dbg !29 + %1554 = bitcast i32 %1553 to float, !dbg !29 + %1555 = fadd float %1551, %1554, !dbg !26 + %1556 = bitcast float %1555 to i32, !dbg !29 + %1557 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1556, i32 2, i32 31), !dbg !29 + %1558 = bitcast i32 %1557 to float, !dbg !29 + %1559 = fadd float %1555, %1558, !dbg !26 + %1560 = bitcast float %1559 to i32, !dbg !29 + %1561 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1560, i32 1, i32 31), !dbg !29 + %1562 = bitcast i32 %1561 to float, !dbg !29 + %1563 = fadd float %1559, %1562, !dbg !26 + %1564 = bitcast float %1301 to i32, !dbg !29 + %1565 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1564, i32 8, i32 31), !dbg !29 + %1566 = bitcast i32 %1565 to float, !dbg !29 + %1567 = fadd float %1301, %1566, !dbg !26 + %1568 = bitcast float %1567 to i32, !dbg !29 + %1569 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1568, i32 4, i32 31), !dbg !29 + %1570 = bitcast i32 %1569 to float, !dbg !29 + %1571 = fadd float %1567, %1570, !dbg !26 + %1572 = bitcast float %1571 to i32, !dbg !29 + %1573 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1572, i32 2, i32 31), !dbg !29 + %1574 = bitcast i32 %1573 to float, !dbg !29 + %1575 = fadd float %1571, %1574, !dbg !26 + %1576 = bitcast float %1575 to i32, !dbg !29 + %1577 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1576, i32 1, i32 31), !dbg !29 + %1578 = bitcast i32 %1577 to float, !dbg !29 + %1579 = fadd float %1575, %1578, !dbg !26 + %1580 = bitcast float %1308 to i32, !dbg !29 + %1581 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1580, i32 8, i32 31), !dbg !29 + %1582 = bitcast i32 %1581 to float, !dbg !29 + %1583 = fadd float %1308, %1582, !dbg !26 + %1584 = bitcast float %1583 to i32, !dbg !29 + %1585 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1584, i32 4, i32 31), !dbg !29 + %1586 = bitcast i32 %1585 to float, !dbg !29 + %1587 = fadd float %1583, %1586, !dbg !26 + %1588 = bitcast float %1587 to i32, !dbg !29 + %1589 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1588, i32 2, i32 31), !dbg !29 + %1590 = bitcast i32 %1589 to float, !dbg !29 + %1591 = fadd float %1587, %1590, !dbg !26 + %1592 = bitcast float %1591 to i32, !dbg !29 + %1593 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1592, i32 1, i32 31), !dbg !29 + %1594 = bitcast i32 %1593 to float, !dbg !29 + %1595 = fadd float %1591, %1594, !dbg !26 + %1596 = bitcast float %1315 to i32, !dbg !29 + %1597 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1596, i32 8, i32 31), !dbg !29 + %1598 = bitcast i32 %1597 to float, !dbg !29 + %1599 = fadd float %1315, %1598, !dbg !26 + %1600 = bitcast float %1599 to i32, !dbg !29 + %1601 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1600, i32 4, i32 31), !dbg !29 + %1602 = bitcast i32 %1601 to float, !dbg !29 + %1603 = fadd float %1599, %1602, !dbg !26 + %1604 = bitcast float %1603 to i32, !dbg !29 + %1605 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1604, i32 2, i32 31), !dbg !29 + %1606 = bitcast i32 %1605 to float, !dbg !29 + %1607 = fadd float %1603, %1606, !dbg !26 + %1608 = bitcast float %1607 to i32, !dbg !29 + %1609 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1608, i32 1, i32 31), !dbg !29 + %1610 = bitcast i32 %1609 to float, !dbg !29 + %1611 = fadd float %1607, %1610, !dbg !26 + %1612 = bitcast float %1322 to i32, !dbg !29 + %1613 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1612, i32 8, i32 31), !dbg !29 + %1614 = bitcast i32 %1613 to float, !dbg !29 + %1615 = fadd float %1322, %1614, !dbg !26 + %1616 = bitcast float %1615 to i32, !dbg !29 + %1617 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1616, i32 4, i32 31), !dbg !29 + %1618 = bitcast i32 %1617 to float, !dbg !29 + %1619 = fadd float %1615, %1618, !dbg !26 + %1620 = bitcast float %1619 to i32, !dbg !29 + %1621 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1620, i32 2, i32 31), !dbg !29 + %1622 = bitcast i32 %1621 to float, !dbg !29 + %1623 = fadd float %1619, %1622, !dbg !26 + %1624 = bitcast float %1623 to i32, !dbg !29 + %1625 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1624, i32 1, i32 31), !dbg !29 + %1626 = bitcast i32 %1625 to float, !dbg !29 + %1627 = fadd float %1623, %1626, !dbg !26 + %1628 = bitcast float %1329 to i32, !dbg !29 + %1629 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1628, i32 8, i32 31), !dbg !29 + %1630 = bitcast i32 %1629 to float, !dbg !29 + %1631 = fadd float %1329, %1630, !dbg !26 + %1632 = bitcast float %1631 to i32, !dbg !29 + %1633 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1632, i32 4, i32 31), !dbg !29 + %1634 = bitcast i32 %1633 to float, !dbg !29 + %1635 = fadd float %1631, %1634, !dbg !26 + %1636 = bitcast float %1635 to i32, !dbg !29 + %1637 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1636, i32 2, i32 31), !dbg !29 + %1638 = bitcast i32 %1637 to float, !dbg !29 + %1639 = fadd float %1635, %1638, !dbg !26 + %1640 = bitcast float %1639 to i32, !dbg !29 + %1641 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1640, i32 1, i32 31), !dbg !29 + %1642 = bitcast i32 %1641 to float, !dbg !29 + %1643 = fadd float %1639, %1642, !dbg !26 + %1644 = bitcast float %1336 to i32, !dbg !29 + %1645 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1644, i32 8, i32 31), !dbg !29 + %1646 = bitcast i32 %1645 to float, !dbg !29 + %1647 = fadd float %1336, %1646, !dbg !26 + %1648 = bitcast float %1647 to i32, !dbg !29 + %1649 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1648, i32 4, i32 31), !dbg !29 + %1650 = bitcast i32 %1649 to float, !dbg !29 + %1651 = fadd float %1647, %1650, !dbg !26 + %1652 = bitcast float %1651 to i32, !dbg !29 + %1653 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1652, i32 2, i32 31), !dbg !29 + %1654 = bitcast i32 %1653 to float, !dbg !29 + %1655 = fadd float %1651, %1654, !dbg !26 + %1656 = bitcast float %1655 to i32, !dbg !29 + %1657 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1656, i32 1, i32 31), !dbg !29 + %1658 = bitcast i32 %1657 to float, !dbg !29 + %1659 = fadd float %1655, %1658, !dbg !26 + %1660 = bitcast float %1343 to i32, !dbg !29 + %1661 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1660, i32 8, i32 31), !dbg !29 + %1662 = bitcast i32 %1661 to float, !dbg !29 + %1663 = fadd float %1343, %1662, !dbg !26 + %1664 = bitcast float %1663 to i32, !dbg !29 + %1665 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1664, i32 4, i32 31), !dbg !29 + %1666 = bitcast i32 %1665 to float, !dbg !29 + %1667 = fadd float %1663, %1666, !dbg !26 + %1668 = bitcast float %1667 to i32, !dbg !29 + %1669 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1668, i32 2, i32 31), !dbg !29 + %1670 = bitcast i32 %1669 to float, !dbg !29 + %1671 = fadd float %1667, %1670, !dbg !26 + %1672 = bitcast float %1671 to i32, !dbg !29 + %1673 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1672, i32 1, i32 31), !dbg !29 + %1674 = bitcast i32 %1673 to float, !dbg !29 + %1675 = fadd float %1671, %1674, !dbg !26 + %1676 = bitcast float %1350 to i32, !dbg !29 + %1677 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1676, i32 8, i32 31), !dbg !29 + %1678 = bitcast i32 %1677 to float, !dbg !29 + %1679 = fadd float %1350, %1678, !dbg !26 + %1680 = bitcast float %1679 to i32, !dbg !29 + %1681 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1680, i32 4, i32 31), !dbg !29 + %1682 = bitcast i32 %1681 to float, !dbg !29 + %1683 = fadd float %1679, %1682, !dbg !26 + %1684 = bitcast float %1683 to i32, !dbg !29 + %1685 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1684, i32 2, i32 31), !dbg !29 + %1686 = bitcast i32 %1685 to float, !dbg !29 + %1687 = fadd float %1683, %1686, !dbg !26 + %1688 = bitcast float %1687 to i32, !dbg !29 + %1689 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1688, i32 1, i32 31), !dbg !29 + %1690 = bitcast i32 %1689 to float, !dbg !29 + %1691 = fadd float %1687, %1690, !dbg !26 + %1692 = bitcast float %1357 to i32, !dbg !29 + %1693 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1692, i32 8, i32 31), !dbg !29 + %1694 = bitcast i32 %1693 to float, !dbg !29 + %1695 = fadd float %1357, %1694, !dbg !26 + %1696 = bitcast float %1695 to i32, !dbg !29 + %1697 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1696, i32 4, i32 31), !dbg !29 + %1698 = bitcast i32 %1697 to float, !dbg !29 + %1699 = fadd float %1695, %1698, !dbg !26 + %1700 = bitcast float %1699 to i32, !dbg !29 + %1701 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1700, i32 2, i32 31), !dbg !29 + %1702 = bitcast i32 %1701 to float, !dbg !29 + %1703 = fadd float %1699, %1702, !dbg !26 + %1704 = bitcast float %1703 to i32, !dbg !29 + %1705 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1704, i32 1, i32 31), !dbg !29 + %1706 = bitcast i32 %1705 to float, !dbg !29 + %1707 = fadd float %1703, %1706, !dbg !26 + %1708 = bitcast float %1364 to i32, !dbg !29 + %1709 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1708, i32 8, i32 31), !dbg !29 + %1710 = bitcast i32 %1709 to float, !dbg !29 + %1711 = fadd float %1364, %1710, !dbg !26 + %1712 = bitcast float %1711 to i32, !dbg !29 + %1713 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1712, i32 4, i32 31), !dbg !29 + %1714 = bitcast i32 %1713 to float, !dbg !29 + %1715 = fadd float %1711, %1714, !dbg !26 + %1716 = bitcast float %1715 to i32, !dbg !29 + %1717 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1716, i32 2, i32 31), !dbg !29 + %1718 = bitcast i32 %1717 to float, !dbg !29 + %1719 = fadd float %1715, %1718, !dbg !26 + %1720 = bitcast float %1719 to i32, !dbg !29 + %1721 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1720, i32 1, i32 31), !dbg !29 + %1722 = bitcast i32 %1721 to float, !dbg !29 + %1723 = fadd float %1719, %1722, !dbg !26 + %1724 = bitcast float %1371 to i32, !dbg !29 + %1725 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1724, i32 8, i32 31), !dbg !29 + %1726 = bitcast i32 %1725 to float, !dbg !29 + %1727 = fadd float %1371, %1726, !dbg !26 + %1728 = bitcast float %1727 to i32, !dbg !29 + %1729 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1728, i32 4, i32 31), !dbg !29 + %1730 = bitcast i32 %1729 to float, !dbg !29 + %1731 = fadd float %1727, %1730, !dbg !26 + %1732 = bitcast float %1731 to i32, !dbg !29 + %1733 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1732, i32 2, i32 31), !dbg !29 + %1734 = bitcast i32 %1733 to float, !dbg !29 + %1735 = fadd float %1731, %1734, !dbg !26 + %1736 = bitcast float %1735 to i32, !dbg !29 + %1737 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1736, i32 1, i32 31), !dbg !29 + %1738 = bitcast i32 %1737 to float, !dbg !29 + %1739 = fadd float %1735, %1738, !dbg !26 + %1740 = bitcast float %1378 to i32, !dbg !29 + %1741 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1740, i32 8, i32 31), !dbg !29 + %1742 = bitcast i32 %1741 to float, !dbg !29 + %1743 = fadd float %1378, %1742, !dbg !26 + %1744 = bitcast float %1743 to i32, !dbg !29 + %1745 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1744, i32 4, i32 31), !dbg !29 + %1746 = bitcast i32 %1745 to float, !dbg !29 + %1747 = fadd float %1743, %1746, !dbg !26 + %1748 = bitcast float %1747 to i32, !dbg !29 + %1749 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1748, i32 2, i32 31), !dbg !29 + %1750 = bitcast i32 %1749 to float, !dbg !29 + %1751 = fadd float %1747, %1750, !dbg !26 + %1752 = bitcast float %1751 to i32, !dbg !29 + %1753 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1752, i32 1, i32 31), !dbg !29 + %1754 = bitcast i32 %1753 to float, !dbg !29 + %1755 = fadd float %1751, %1754, !dbg !26 + %1756 = bitcast float %1385 to i32, !dbg !29 + %1757 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1756, i32 8, i32 31), !dbg !29 + %1758 = bitcast i32 %1757 to float, !dbg !29 + %1759 = fadd float %1385, %1758, !dbg !26 + %1760 = bitcast float %1759 to i32, !dbg !29 + %1761 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1760, i32 4, i32 31), !dbg !29 + %1762 = bitcast i32 %1761 to float, !dbg !29 + %1763 = fadd float %1759, %1762, !dbg !26 + %1764 = bitcast float %1763 to i32, !dbg !29 + %1765 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1764, i32 2, i32 31), !dbg !29 + %1766 = bitcast i32 %1765 to float, !dbg !29 + %1767 = fadd float %1763, %1766, !dbg !26 + %1768 = bitcast float %1767 to i32, !dbg !29 + %1769 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1768, i32 1, i32 31), !dbg !29 + %1770 = bitcast i32 %1769 to float, !dbg !29 + %1771 = fadd float %1767, %1770, !dbg !26 + %1772 = bitcast float %1392 to i32, !dbg !29 + %1773 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1772, i32 8, i32 31), !dbg !29 + %1774 = bitcast i32 %1773 to float, !dbg !29 + %1775 = fadd float %1392, %1774, !dbg !26 + %1776 = bitcast float %1775 to i32, !dbg !29 + %1777 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1776, i32 4, i32 31), !dbg !29 + %1778 = bitcast i32 %1777 to float, !dbg !29 + %1779 = fadd float %1775, %1778, !dbg !26 + %1780 = bitcast float %1779 to i32, !dbg !29 + %1781 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1780, i32 2, i32 31), !dbg !29 + %1782 = bitcast i32 %1781 to float, !dbg !29 + %1783 = fadd float %1779, %1782, !dbg !26 + %1784 = bitcast float %1783 to i32, !dbg !29 + %1785 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1784, i32 1, i32 31), !dbg !29 + %1786 = bitcast i32 %1785 to float, !dbg !29 + %1787 = fadd float %1783, %1786, !dbg !26 + %1788 = bitcast float %1399 to i32, !dbg !29 + %1789 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1788, i32 8, i32 31), !dbg !29 + %1790 = bitcast i32 %1789 to float, !dbg !29 + %1791 = fadd float %1399, %1790, !dbg !26 + %1792 = bitcast float %1791 to i32, !dbg !29 + %1793 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1792, i32 4, i32 31), !dbg !29 + %1794 = bitcast i32 %1793 to float, !dbg !29 + %1795 = fadd float %1791, %1794, !dbg !26 + %1796 = bitcast float %1795 to i32, !dbg !29 + %1797 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1796, i32 2, i32 31), !dbg !29 + %1798 = bitcast i32 %1797 to float, !dbg !29 + %1799 = fadd float %1795, %1798, !dbg !26 + %1800 = bitcast float %1799 to i32, !dbg !29 + %1801 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1800, i32 1, i32 31), !dbg !29 + %1802 = bitcast i32 %1801 to float, !dbg !29 + %1803 = fadd float %1799, %1802, !dbg !26 + %1804 = bitcast float %1406 to i32, !dbg !29 + %1805 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1804, i32 8, i32 31), !dbg !29 + %1806 = bitcast i32 %1805 to float, !dbg !29 + %1807 = fadd float %1406, %1806, !dbg !26 + %1808 = bitcast float %1807 to i32, !dbg !29 + %1809 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1808, i32 4, i32 31), !dbg !29 + %1810 = bitcast i32 %1809 to float, !dbg !29 + %1811 = fadd float %1807, %1810, !dbg !26 + %1812 = bitcast float %1811 to i32, !dbg !29 + %1813 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1812, i32 2, i32 31), !dbg !29 + %1814 = bitcast i32 %1813 to float, !dbg !29 + %1815 = fadd float %1811, %1814, !dbg !26 + %1816 = bitcast float %1815 to i32, !dbg !29 + %1817 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1816, i32 1, i32 31), !dbg !29 + %1818 = bitcast i32 %1817 to float, !dbg !29 + %1819 = fadd float %1815, %1818, !dbg !26 + %1820 = bitcast float %1413 to i32, !dbg !29 + %1821 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1820, i32 8, i32 31), !dbg !29 + %1822 = bitcast i32 %1821 to float, !dbg !29 + %1823 = fadd float %1413, %1822, !dbg !26 + %1824 = bitcast float %1823 to i32, !dbg !29 + %1825 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1824, i32 4, i32 31), !dbg !29 + %1826 = bitcast i32 %1825 to float, !dbg !29 + %1827 = fadd float %1823, %1826, !dbg !26 + %1828 = bitcast float %1827 to i32, !dbg !29 + %1829 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1828, i32 2, i32 31), !dbg !29 + %1830 = bitcast i32 %1829 to float, !dbg !29 + %1831 = fadd float %1827, %1830, !dbg !26 + %1832 = bitcast float %1831 to i32, !dbg !29 + %1833 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1832, i32 1, i32 31), !dbg !29 + %1834 = bitcast i32 %1833 to float, !dbg !29 + %1835 = fadd float %1831, %1834, !dbg !26 + %1836 = bitcast float %1420 to i32, !dbg !29 + %1837 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1836, i32 8, i32 31), !dbg !29 + %1838 = bitcast i32 %1837 to float, !dbg !29 + %1839 = fadd float %1420, %1838, !dbg !26 + %1840 = bitcast float %1839 to i32, !dbg !29 + %1841 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1840, i32 4, i32 31), !dbg !29 + %1842 = bitcast i32 %1841 to float, !dbg !29 + %1843 = fadd float %1839, %1842, !dbg !26 + %1844 = bitcast float %1843 to i32, !dbg !29 + %1845 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1844, i32 2, i32 31), !dbg !29 + %1846 = bitcast i32 %1845 to float, !dbg !29 + %1847 = fadd float %1843, %1846, !dbg !26 + %1848 = bitcast float %1847 to i32, !dbg !29 + %1849 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1848, i32 1, i32 31), !dbg !29 + %1850 = bitcast i32 %1849 to float, !dbg !29 + %1851 = fadd float %1847, %1850, !dbg !26 + %1852 = bitcast float %1427 to i32, !dbg !29 + %1853 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1852, i32 8, i32 31), !dbg !29 + %1854 = bitcast i32 %1853 to float, !dbg !29 + %1855 = fadd float %1427, %1854, !dbg !26 + %1856 = bitcast float %1855 to i32, !dbg !29 + %1857 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1856, i32 4, i32 31), !dbg !29 + %1858 = bitcast i32 %1857 to float, !dbg !29 + %1859 = fadd float %1855, %1858, !dbg !26 + %1860 = bitcast float %1859 to i32, !dbg !29 + %1861 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1860, i32 2, i32 31), !dbg !29 + %1862 = bitcast i32 %1861 to float, !dbg !29 + %1863 = fadd float %1859, %1862, !dbg !26 + %1864 = bitcast float %1863 to i32, !dbg !29 + %1865 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1864, i32 1, i32 31), !dbg !29 + %1866 = bitcast i32 %1865 to float, !dbg !29 + %1867 = fadd float %1863, %1866, !dbg !26 + %1868 = bitcast float %1434 to i32, !dbg !29 + %1869 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1868, i32 8, i32 31), !dbg !29 + %1870 = bitcast i32 %1869 to float, !dbg !29 + %1871 = fadd float %1434, %1870, !dbg !26 + %1872 = bitcast float %1871 to i32, !dbg !29 + %1873 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1872, i32 4, i32 31), !dbg !29 + %1874 = bitcast i32 %1873 to float, !dbg !29 + %1875 = fadd float %1871, %1874, !dbg !26 + %1876 = bitcast float %1875 to i32, !dbg !29 + %1877 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1876, i32 2, i32 31), !dbg !29 + %1878 = bitcast i32 %1877 to float, !dbg !29 + %1879 = fadd float %1875, %1878, !dbg !26 + %1880 = bitcast float %1879 to i32, !dbg !29 + %1881 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1880, i32 1, i32 31), !dbg !29 + %1882 = bitcast i32 %1881 to float, !dbg !29 + %1883 = fadd float %1879, %1882, !dbg !26 + %1884 = bitcast float %1441 to i32, !dbg !29 + %1885 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1884, i32 8, i32 31), !dbg !29 + %1886 = bitcast i32 %1885 to float, !dbg !29 + %1887 = fadd float %1441, %1886, !dbg !26 + %1888 = bitcast float %1887 to i32, !dbg !29 + %1889 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1888, i32 4, i32 31), !dbg !29 + %1890 = bitcast i32 %1889 to float, !dbg !29 + %1891 = fadd float %1887, %1890, !dbg !26 + %1892 = bitcast float %1891 to i32, !dbg !29 + %1893 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1892, i32 2, i32 31), !dbg !29 + %1894 = bitcast i32 %1893 to float, !dbg !29 + %1895 = fadd float %1891, %1894, !dbg !26 + %1896 = bitcast float %1895 to i32, !dbg !29 + %1897 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1896, i32 1, i32 31), !dbg !29 + %1898 = bitcast i32 %1897 to float, !dbg !29 + %1899 = fadd float %1895, %1898, !dbg !26 + %1900 = bitcast float %1448 to i32, !dbg !29 + %1901 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1900, i32 8, i32 31), !dbg !29 + %1902 = bitcast i32 %1901 to float, !dbg !29 + %1903 = fadd float %1448, %1902, !dbg !26 + %1904 = bitcast float %1903 to i32, !dbg !29 + %1905 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1904, i32 4, i32 31), !dbg !29 + %1906 = bitcast i32 %1905 to float, !dbg !29 + %1907 = fadd float %1903, %1906, !dbg !26 + %1908 = bitcast float %1907 to i32, !dbg !29 + %1909 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1908, i32 2, i32 31), !dbg !29 + %1910 = bitcast i32 %1909 to float, !dbg !29 + %1911 = fadd float %1907, %1910, !dbg !26 + %1912 = bitcast float %1911 to i32, !dbg !29 + %1913 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1912, i32 1, i32 31), !dbg !29 + %1914 = bitcast i32 %1913 to float, !dbg !29 + %1915 = fadd float %1911, %1914, !dbg !26 + %1916 = bitcast float %1455 to i32, !dbg !29 + %1917 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1916, i32 8, i32 31), !dbg !29 + %1918 = bitcast i32 %1917 to float, !dbg !29 + %1919 = fadd float %1455, %1918, !dbg !26 + %1920 = bitcast float %1919 to i32, !dbg !29 + %1921 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1920, i32 4, i32 31), !dbg !29 + %1922 = bitcast i32 %1921 to float, !dbg !29 + %1923 = fadd float %1919, %1922, !dbg !26 + %1924 = bitcast float %1923 to i32, !dbg !29 + %1925 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1924, i32 2, i32 31), !dbg !29 + %1926 = bitcast i32 %1925 to float, !dbg !29 + %1927 = fadd float %1923, %1926, !dbg !26 + %1928 = bitcast float %1927 to i32, !dbg !29 + %1929 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1928, i32 1, i32 31), !dbg !29 + %1930 = bitcast i32 %1929 to float, !dbg !29 + %1931 = fadd float %1927, %1930, !dbg !26 + %1932 = bitcast float %1462 to i32, !dbg !29 + %1933 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1932, i32 8, i32 31), !dbg !29 + %1934 = bitcast i32 %1933 to float, !dbg !29 + %1935 = fadd float %1462, %1934, !dbg !26 + %1936 = bitcast float %1935 to i32, !dbg !29 + %1937 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1936, i32 4, i32 31), !dbg !29 + %1938 = bitcast i32 %1937 to float, !dbg !29 + %1939 = fadd float %1935, %1938, !dbg !26 + %1940 = bitcast float %1939 to i32, !dbg !29 + %1941 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1940, i32 2, i32 31), !dbg !29 + %1942 = bitcast i32 %1941 to float, !dbg !29 + %1943 = fadd float %1939, %1942, !dbg !26 + %1944 = bitcast float %1943 to i32, !dbg !29 + %1945 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1944, i32 1, i32 31), !dbg !29 + %1946 = bitcast i32 %1945 to float, !dbg !29 + %1947 = fadd float %1943, %1946, !dbg !26 + %1948 = bitcast float %1469 to i32, !dbg !29 + %1949 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1948, i32 8, i32 31), !dbg !29 + %1950 = bitcast i32 %1949 to float, !dbg !29 + %1951 = fadd float %1469, %1950, !dbg !26 + %1952 = bitcast float %1951 to i32, !dbg !29 + %1953 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1952, i32 4, i32 31), !dbg !29 + %1954 = bitcast i32 %1953 to float, !dbg !29 + %1955 = fadd float %1951, %1954, !dbg !26 + %1956 = bitcast float %1955 to i32, !dbg !29 + %1957 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1956, i32 2, i32 31), !dbg !29 + %1958 = bitcast i32 %1957 to float, !dbg !29 + %1959 = fadd float %1955, %1958, !dbg !26 + %1960 = bitcast float %1959 to i32, !dbg !29 + %1961 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1960, i32 1, i32 31), !dbg !29 + %1962 = bitcast i32 %1961 to float, !dbg !29 + %1963 = fadd float %1959, %1962, !dbg !26 + %1964 = bitcast float %1476 to i32, !dbg !29 + %1965 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1964, i32 8, i32 31), !dbg !29 + %1966 = bitcast i32 %1965 to float, !dbg !29 + %1967 = fadd float %1476, %1966, !dbg !26 + %1968 = bitcast float %1967 to i32, !dbg !29 + %1969 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1968, i32 4, i32 31), !dbg !29 + %1970 = bitcast i32 %1969 to float, !dbg !29 + %1971 = fadd float %1967, %1970, !dbg !26 + %1972 = bitcast float %1971 to i32, !dbg !29 + %1973 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1972, i32 2, i32 31), !dbg !29 + %1974 = bitcast i32 %1973 to float, !dbg !29 + %1975 = fadd float %1971, %1974, !dbg !26 + %1976 = bitcast float %1975 to i32, !dbg !29 + %1977 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1976, i32 1, i32 31), !dbg !29 + %1978 = bitcast i32 %1977 to float, !dbg !29 + %1979 = fadd float %1975, %1978, !dbg !26 + %1980 = bitcast float %1483 to i32, !dbg !29 + %1981 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1980, i32 8, i32 31), !dbg !29 + %1982 = bitcast i32 %1981 to float, !dbg !29 + %1983 = fadd float %1483, %1982, !dbg !26 + %1984 = bitcast float %1983 to i32, !dbg !29 + %1985 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1984, i32 4, i32 31), !dbg !29 + %1986 = bitcast i32 %1985 to float, !dbg !29 + %1987 = fadd float %1983, %1986, !dbg !26 + %1988 = bitcast float %1987 to i32, !dbg !29 + %1989 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1988, i32 2, i32 31), !dbg !29 + %1990 = bitcast i32 %1989 to float, !dbg !29 + %1991 = fadd float %1987, %1990, !dbg !26 + %1992 = bitcast float %1991 to i32, !dbg !29 + %1993 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1992, i32 1, i32 31), !dbg !29 + %1994 = bitcast i32 %1993 to float, !dbg !29 + %1995 = fadd float %1991, %1994, !dbg !26 + %1996 = fadd float %3, %1499, !dbg !32 + %1997 = fadd float %3, %1515, !dbg !32 + %1998 = fadd float %3, %1531, !dbg !32 + %1999 = fadd float %3, %1547, !dbg !32 + %2000 = fadd float %3, %1563, !dbg !32 + %2001 = fadd float %3, %1579, !dbg !32 + %2002 = fadd float %3, %1595, !dbg !32 + %2003 = fadd float %3, %1611, !dbg !32 + %2004 = fadd float %3, %1627, !dbg !32 + %2005 = fadd float %3, %1643, !dbg !32 + %2006 = fadd float %3, %1659, !dbg !32 + %2007 = fadd float %3, %1675, !dbg !32 + %2008 = fadd float %3, %1691, !dbg !32 + %2009 = fadd float %3, %1707, !dbg !32 + %2010 = fadd float %3, %1723, !dbg !32 + %2011 = fadd float %3, %1739, !dbg !32 + %2012 = fadd float %3, %1755, !dbg !32 + %2013 = fadd float %3, %1771, !dbg !32 + %2014 = fadd float %3, %1787, !dbg !32 + %2015 = fadd float %3, %1803, !dbg !32 + %2016 = fadd float %3, %1819, !dbg !32 + %2017 = fadd float %3, %1835, !dbg !32 + %2018 = fadd float %3, %1851, !dbg !32 + %2019 = fadd float %3, %1867, !dbg !32 + %2020 = fadd float %3, %1883, !dbg !32 + %2021 = fadd float %3, %1899, !dbg !32 + %2022 = fadd float %3, %1915, !dbg !32 + %2023 = fadd float %3, %1931, !dbg !32 + %2024 = fadd float %3, %1947, !dbg !32 + %2025 = fadd float %3, %1963, !dbg !32 + %2026 = fadd float %3, %1979, !dbg !32 + %2027 = fadd float %3, %1995, !dbg !32 + %2028 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i = icmp eq i32 %2028, 0, !dbg !33 + %2029 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i = icmp eq i32 %2029, 0, !dbg !33 + br i1 %.not.i, label %2035, label %2030, !dbg !33 + +2030: ; preds = %7 + br i1 %.not1.i, label %2033, label %2031, !dbg !33 + +2031: ; preds = %2030 + %2032 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1996) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +2033: ; preds = %2030 + %2034 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1996) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +2035: ; preds = %7 + br i1 %.not1.i, label %2038, label %2036, !dbg !33 + +2036: ; preds = %2035 + %2037 = tail call float @llvm.nvvm.sqrt.rn.f(float %1996) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +2038: ; preds = %2035 + %2039 = tail call float @llvm.nvvm.sqrt.approx.f(float %1996) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +__nv_sqrtf.exit: ; preds = %2031, %2033, %2036, %2038 + %.0.i = phi float [ %2032, %2031 ], [ %2034, %2033 ], [ %2037, %2036 ], [ %2039, %2038 ], !dbg !33 + %2040 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i1 = icmp eq i32 %2040, 0, !dbg !33 + %2041 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i4 = icmp eq i32 %2041, 0, !dbg !33 + br i1 %.not.i1, label %2047, label %2042, !dbg !33 + +2042: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %2045, label %2043, !dbg !33 + +2043: ; preds = %2042 + %2044 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1997) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +2045: ; preds = %2042 + %2046 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1997) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +2047: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %2050, label %2048, !dbg !33 + +2048: ; preds = %2047 + %2049 = tail call float @llvm.nvvm.sqrt.rn.f(float %1997) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +2050: ; preds = %2047 + %2051 = tail call float @llvm.nvvm.sqrt.approx.f(float %1997) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +__nv_sqrtf.exit5: ; preds = %2043, %2045, %2048, %2050 + %.0.i3 = phi float [ %2044, %2043 ], [ %2046, %2045 ], [ %2049, %2048 ], [ %2051, %2050 ], !dbg !33 + %2052 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i6 = icmp eq i32 %2052, 0, !dbg !33 + %2053 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i9 = icmp eq i32 %2053, 0, !dbg !33 + br i1 %.not.i6, label %2059, label %2054, !dbg !33 + +2054: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %2057, label %2055, !dbg !33 + +2055: ; preds = %2054 + %2056 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1998) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +2057: ; preds = %2054 + %2058 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1998) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +2059: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %2062, label %2060, !dbg !33 + +2060: ; preds = %2059 + %2061 = tail call float @llvm.nvvm.sqrt.rn.f(float %1998) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +2062: ; preds = %2059 + %2063 = tail call float @llvm.nvvm.sqrt.approx.f(float %1998) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +__nv_sqrtf.exit10: ; preds = %2055, %2057, %2060, %2062 + %.0.i8 = phi float [ %2056, %2055 ], [ %2058, %2057 ], [ %2061, %2060 ], [ %2063, %2062 ], !dbg !33 + %2064 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i11 = icmp eq i32 %2064, 0, !dbg !33 + %2065 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i14 = icmp eq i32 %2065, 0, !dbg !33 + br i1 %.not.i11, label %2071, label %2066, !dbg !33 + +2066: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %2069, label %2067, !dbg !33 + +2067: ; preds = %2066 + %2068 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1999) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +2069: ; preds = %2066 + %2070 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1999) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +2071: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %2074, label %2072, !dbg !33 + +2072: ; preds = %2071 + %2073 = tail call float @llvm.nvvm.sqrt.rn.f(float %1999) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +2074: ; preds = %2071 + %2075 = tail call float @llvm.nvvm.sqrt.approx.f(float %1999) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +__nv_sqrtf.exit15: ; preds = %2067, %2069, %2072, %2074 + %.0.i13 = phi float [ %2068, %2067 ], [ %2070, %2069 ], [ %2073, %2072 ], [ %2075, %2074 ], !dbg !33 + %2076 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i16 = icmp eq i32 %2076, 0, !dbg !33 + %2077 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i19 = icmp eq i32 %2077, 0, !dbg !33 + br i1 %.not.i16, label %2083, label %2078, !dbg !33 + +2078: ; preds = %__nv_sqrtf.exit15 + br i1 %.not1.i19, label %2081, label %2079, !dbg !33 + +2079: ; preds = %2078 + %2080 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2000) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +2081: ; preds = %2078 + %2082 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2000) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +2083: ; preds = %__nv_sqrtf.exit15 + br i1 %.not1.i19, label %2086, label %2084, !dbg !33 + +2084: ; preds = %2083 + %2085 = tail call float @llvm.nvvm.sqrt.rn.f(float %2000) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +2086: ; preds = %2083 + %2087 = tail call float @llvm.nvvm.sqrt.approx.f(float %2000) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +__nv_sqrtf.exit20: ; preds = %2079, %2081, %2084, %2086 + %.0.i18 = phi float [ %2080, %2079 ], [ %2082, %2081 ], [ %2085, %2084 ], [ %2087, %2086 ], !dbg !33 + %2088 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i21 = icmp eq i32 %2088, 0, !dbg !33 + %2089 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i24 = icmp eq i32 %2089, 0, !dbg !33 + br i1 %.not.i21, label %2095, label %2090, !dbg !33 + +2090: ; preds = %__nv_sqrtf.exit20 + br i1 %.not1.i24, label %2093, label %2091, !dbg !33 + +2091: ; preds = %2090 + %2092 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2001) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +2093: ; preds = %2090 + %2094 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2001) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +2095: ; preds = %__nv_sqrtf.exit20 + br i1 %.not1.i24, label %2098, label %2096, !dbg !33 + +2096: ; preds = %2095 + %2097 = tail call float @llvm.nvvm.sqrt.rn.f(float %2001) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +2098: ; preds = %2095 + %2099 = tail call float @llvm.nvvm.sqrt.approx.f(float %2001) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +__nv_sqrtf.exit25: ; preds = %2091, %2093, %2096, %2098 + %.0.i23 = phi float [ %2092, %2091 ], [ %2094, %2093 ], [ %2097, %2096 ], [ %2099, %2098 ], !dbg !33 + %2100 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i26 = icmp eq i32 %2100, 0, !dbg !33 + %2101 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i29 = icmp eq i32 %2101, 0, !dbg !33 + br i1 %.not.i26, label %2107, label %2102, !dbg !33 + +2102: ; preds = %__nv_sqrtf.exit25 + br i1 %.not1.i29, label %2105, label %2103, !dbg !33 + +2103: ; preds = %2102 + %2104 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2002) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +2105: ; preds = %2102 + %2106 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2002) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +2107: ; preds = %__nv_sqrtf.exit25 + br i1 %.not1.i29, label %2110, label %2108, !dbg !33 + +2108: ; preds = %2107 + %2109 = tail call float @llvm.nvvm.sqrt.rn.f(float %2002) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +2110: ; preds = %2107 + %2111 = tail call float @llvm.nvvm.sqrt.approx.f(float %2002) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +__nv_sqrtf.exit30: ; preds = %2103, %2105, %2108, %2110 + %.0.i28 = phi float [ %2104, %2103 ], [ %2106, %2105 ], [ %2109, %2108 ], [ %2111, %2110 ], !dbg !33 + %2112 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i31 = icmp eq i32 %2112, 0, !dbg !33 + %2113 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i34 = icmp eq i32 %2113, 0, !dbg !33 + br i1 %.not.i31, label %2119, label %2114, !dbg !33 + +2114: ; preds = %__nv_sqrtf.exit30 + br i1 %.not1.i34, label %2117, label %2115, !dbg !33 + +2115: ; preds = %2114 + %2116 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2003) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +2117: ; preds = %2114 + %2118 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2003) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +2119: ; preds = %__nv_sqrtf.exit30 + br i1 %.not1.i34, label %2122, label %2120, !dbg !33 + +2120: ; preds = %2119 + %2121 = tail call float @llvm.nvvm.sqrt.rn.f(float %2003) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +2122: ; preds = %2119 + %2123 = tail call float @llvm.nvvm.sqrt.approx.f(float %2003) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +__nv_sqrtf.exit35: ; preds = %2115, %2117, %2120, %2122 + %.0.i33 = phi float [ %2116, %2115 ], [ %2118, %2117 ], [ %2121, %2120 ], [ %2123, %2122 ], !dbg !33 + %2124 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i36 = icmp eq i32 %2124, 0, !dbg !33 + %2125 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i39 = icmp eq i32 %2125, 0, !dbg !33 + br i1 %.not.i36, label %2131, label %2126, !dbg !33 + +2126: ; preds = %__nv_sqrtf.exit35 + br i1 %.not1.i39, label %2129, label %2127, !dbg !33 + +2127: ; preds = %2126 + %2128 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2004) #6, !dbg !33 + br label %__nv_sqrtf.exit40, !dbg !33 + +2129: ; preds = %2126 + %2130 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2004) #6, !dbg !33 + br label %__nv_sqrtf.exit40, !dbg !33 + +2131: ; preds = %__nv_sqrtf.exit35 + br i1 %.not1.i39, label %2134, label %2132, !dbg !33 + +2132: ; preds = %2131 + %2133 = tail call float @llvm.nvvm.sqrt.rn.f(float %2004) #6, !dbg !33 + br label %__nv_sqrtf.exit40, !dbg !33 + +2134: ; preds = %2131 + %2135 = tail call float @llvm.nvvm.sqrt.approx.f(float %2004) #6, !dbg !33 + br label %__nv_sqrtf.exit40, !dbg !33 + +__nv_sqrtf.exit40: ; preds = %2127, %2129, %2132, %2134 + %.0.i38 = phi float [ %2128, %2127 ], [ %2130, %2129 ], [ %2133, %2132 ], [ %2135, %2134 ], !dbg !33 + %2136 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i41 = icmp eq i32 %2136, 0, !dbg !33 + %2137 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i44 = icmp eq i32 %2137, 0, !dbg !33 + br i1 %.not.i41, label %2143, label %2138, !dbg !33 + +2138: ; preds = %__nv_sqrtf.exit40 + br i1 %.not1.i44, label %2141, label %2139, !dbg !33 + +2139: ; preds = %2138 + %2140 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2005) #6, !dbg !33 + br label %__nv_sqrtf.exit45, !dbg !33 + +2141: ; preds = %2138 + %2142 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2005) #6, !dbg !33 + br label %__nv_sqrtf.exit45, !dbg !33 + +2143: ; preds = %__nv_sqrtf.exit40 + br i1 %.not1.i44, label %2146, label %2144, !dbg !33 + +2144: ; preds = %2143 + %2145 = tail call float @llvm.nvvm.sqrt.rn.f(float %2005) #6, !dbg !33 + br label %__nv_sqrtf.exit45, !dbg !33 + +2146: ; preds = %2143 + %2147 = tail call float @llvm.nvvm.sqrt.approx.f(float %2005) #6, !dbg !33 + br label %__nv_sqrtf.exit45, !dbg !33 + +__nv_sqrtf.exit45: ; preds = %2139, %2141, %2144, %2146 + %.0.i43 = phi float [ %2140, %2139 ], [ %2142, %2141 ], [ %2145, %2144 ], [ %2147, %2146 ], !dbg !33 + %2148 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i46 = icmp eq i32 %2148, 0, !dbg !33 + %2149 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i49 = icmp eq i32 %2149, 0, !dbg !33 + br i1 %.not.i46, label %2155, label %2150, !dbg !33 + +2150: ; preds = %__nv_sqrtf.exit45 + br i1 %.not1.i49, label %2153, label %2151, !dbg !33 + +2151: ; preds = %2150 + %2152 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2006) #6, !dbg !33 + br label %__nv_sqrtf.exit50, !dbg !33 + +2153: ; preds = %2150 + %2154 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2006) #6, !dbg !33 + br label %__nv_sqrtf.exit50, !dbg !33 + +2155: ; preds = %__nv_sqrtf.exit45 + br i1 %.not1.i49, label %2158, label %2156, !dbg !33 + +2156: ; preds = %2155 + %2157 = tail call float @llvm.nvvm.sqrt.rn.f(float %2006) #6, !dbg !33 + br label %__nv_sqrtf.exit50, !dbg !33 + +2158: ; preds = %2155 + %2159 = tail call float @llvm.nvvm.sqrt.approx.f(float %2006) #6, !dbg !33 + br label %__nv_sqrtf.exit50, !dbg !33 + +__nv_sqrtf.exit50: ; preds = %2151, %2153, %2156, %2158 + %.0.i48 = phi float [ %2152, %2151 ], [ %2154, %2153 ], [ %2157, %2156 ], [ %2159, %2158 ], !dbg !33 + %2160 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i51 = icmp eq i32 %2160, 0, !dbg !33 + %2161 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i54 = icmp eq i32 %2161, 0, !dbg !33 + br i1 %.not.i51, label %2167, label %2162, !dbg !33 + +2162: ; preds = %__nv_sqrtf.exit50 + br i1 %.not1.i54, label %2165, label %2163, !dbg !33 + +2163: ; preds = %2162 + %2164 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2007) #6, !dbg !33 + br label %__nv_sqrtf.exit55, !dbg !33 + +2165: ; preds = %2162 + %2166 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2007) #6, !dbg !33 + br label %__nv_sqrtf.exit55, !dbg !33 + +2167: ; preds = %__nv_sqrtf.exit50 + br i1 %.not1.i54, label %2170, label %2168, !dbg !33 + +2168: ; preds = %2167 + %2169 = tail call float @llvm.nvvm.sqrt.rn.f(float %2007) #6, !dbg !33 + br label %__nv_sqrtf.exit55, !dbg !33 + +2170: ; preds = %2167 + %2171 = tail call float @llvm.nvvm.sqrt.approx.f(float %2007) #6, !dbg !33 + br label %__nv_sqrtf.exit55, !dbg !33 + +__nv_sqrtf.exit55: ; preds = %2163, %2165, %2168, %2170 + %.0.i53 = phi float [ %2164, %2163 ], [ %2166, %2165 ], [ %2169, %2168 ], [ %2171, %2170 ], !dbg !33 + %2172 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i56 = icmp eq i32 %2172, 0, !dbg !33 + %2173 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i59 = icmp eq i32 %2173, 0, !dbg !33 + br i1 %.not.i56, label %2179, label %2174, !dbg !33 + +2174: ; preds = %__nv_sqrtf.exit55 + br i1 %.not1.i59, label %2177, label %2175, !dbg !33 + +2175: ; preds = %2174 + %2176 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2008) #6, !dbg !33 + br label %__nv_sqrtf.exit60, !dbg !33 + +2177: ; preds = %2174 + %2178 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2008) #6, !dbg !33 + br label %__nv_sqrtf.exit60, !dbg !33 + +2179: ; preds = %__nv_sqrtf.exit55 + br i1 %.not1.i59, label %2182, label %2180, !dbg !33 + +2180: ; preds = %2179 + %2181 = tail call float @llvm.nvvm.sqrt.rn.f(float %2008) #6, !dbg !33 + br label %__nv_sqrtf.exit60, !dbg !33 + +2182: ; preds = %2179 + %2183 = tail call float @llvm.nvvm.sqrt.approx.f(float %2008) #6, !dbg !33 + br label %__nv_sqrtf.exit60, !dbg !33 + +__nv_sqrtf.exit60: ; preds = %2175, %2177, %2180, %2182 + %.0.i58 = phi float [ %2176, %2175 ], [ %2178, %2177 ], [ %2181, %2180 ], [ %2183, %2182 ], !dbg !33 + %2184 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i61 = icmp eq i32 %2184, 0, !dbg !33 + %2185 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i64 = icmp eq i32 %2185, 0, !dbg !33 + br i1 %.not.i61, label %2191, label %2186, !dbg !33 + +2186: ; preds = %__nv_sqrtf.exit60 + br i1 %.not1.i64, label %2189, label %2187, !dbg !33 + +2187: ; preds = %2186 + %2188 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2009) #6, !dbg !33 + br label %__nv_sqrtf.exit65, !dbg !33 + +2189: ; preds = %2186 + %2190 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2009) #6, !dbg !33 + br label %__nv_sqrtf.exit65, !dbg !33 + +2191: ; preds = %__nv_sqrtf.exit60 + br i1 %.not1.i64, label %2194, label %2192, !dbg !33 + +2192: ; preds = %2191 + %2193 = tail call float @llvm.nvvm.sqrt.rn.f(float %2009) #6, !dbg !33 + br label %__nv_sqrtf.exit65, !dbg !33 + +2194: ; preds = %2191 + %2195 = tail call float @llvm.nvvm.sqrt.approx.f(float %2009) #6, !dbg !33 + br label %__nv_sqrtf.exit65, !dbg !33 + +__nv_sqrtf.exit65: ; preds = %2187, %2189, %2192, %2194 + %.0.i63 = phi float [ %2188, %2187 ], [ %2190, %2189 ], [ %2193, %2192 ], [ %2195, %2194 ], !dbg !33 + %2196 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i66 = icmp eq i32 %2196, 0, !dbg !33 + %2197 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i69 = icmp eq i32 %2197, 0, !dbg !33 + br i1 %.not.i66, label %2203, label %2198, !dbg !33 + +2198: ; preds = %__nv_sqrtf.exit65 + br i1 %.not1.i69, label %2201, label %2199, !dbg !33 + +2199: ; preds = %2198 + %2200 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2010) #6, !dbg !33 + br label %__nv_sqrtf.exit70, !dbg !33 + +2201: ; preds = %2198 + %2202 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2010) #6, !dbg !33 + br label %__nv_sqrtf.exit70, !dbg !33 + +2203: ; preds = %__nv_sqrtf.exit65 + br i1 %.not1.i69, label %2206, label %2204, !dbg !33 + +2204: ; preds = %2203 + %2205 = tail call float @llvm.nvvm.sqrt.rn.f(float %2010) #6, !dbg !33 + br label %__nv_sqrtf.exit70, !dbg !33 + +2206: ; preds = %2203 + %2207 = tail call float @llvm.nvvm.sqrt.approx.f(float %2010) #6, !dbg !33 + br label %__nv_sqrtf.exit70, !dbg !33 + +__nv_sqrtf.exit70: ; preds = %2199, %2201, %2204, %2206 + %.0.i68 = phi float [ %2200, %2199 ], [ %2202, %2201 ], [ %2205, %2204 ], [ %2207, %2206 ], !dbg !33 + %2208 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i71 = icmp eq i32 %2208, 0, !dbg !33 + %2209 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i74 = icmp eq i32 %2209, 0, !dbg !33 + br i1 %.not.i71, label %2215, label %2210, !dbg !33 + +2210: ; preds = %__nv_sqrtf.exit70 + br i1 %.not1.i74, label %2213, label %2211, !dbg !33 + +2211: ; preds = %2210 + %2212 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2011) #6, !dbg !33 + br label %__nv_sqrtf.exit75, !dbg !33 + +2213: ; preds = %2210 + %2214 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2011) #6, !dbg !33 + br label %__nv_sqrtf.exit75, !dbg !33 + +2215: ; preds = %__nv_sqrtf.exit70 + br i1 %.not1.i74, label %2218, label %2216, !dbg !33 + +2216: ; preds = %2215 + %2217 = tail call float @llvm.nvvm.sqrt.rn.f(float %2011) #6, !dbg !33 + br label %__nv_sqrtf.exit75, !dbg !33 + +2218: ; preds = %2215 + %2219 = tail call float @llvm.nvvm.sqrt.approx.f(float %2011) #6, !dbg !33 + br label %__nv_sqrtf.exit75, !dbg !33 + +__nv_sqrtf.exit75: ; preds = %2211, %2213, %2216, %2218 + %.0.i73 = phi float [ %2212, %2211 ], [ %2214, %2213 ], [ %2217, %2216 ], [ %2219, %2218 ], !dbg !33 + %2220 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i76 = icmp eq i32 %2220, 0, !dbg !33 + %2221 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i79 = icmp eq i32 %2221, 0, !dbg !33 + br i1 %.not.i76, label %2227, label %2222, !dbg !33 + +2222: ; preds = %__nv_sqrtf.exit75 + br i1 %.not1.i79, label %2225, label %2223, !dbg !33 + +2223: ; preds = %2222 + %2224 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2012) #6, !dbg !33 + br label %__nv_sqrtf.exit80, !dbg !33 + +2225: ; preds = %2222 + %2226 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2012) #6, !dbg !33 + br label %__nv_sqrtf.exit80, !dbg !33 + +2227: ; preds = %__nv_sqrtf.exit75 + br i1 %.not1.i79, label %2230, label %2228, !dbg !33 + +2228: ; preds = %2227 + %2229 = tail call float @llvm.nvvm.sqrt.rn.f(float %2012) #6, !dbg !33 + br label %__nv_sqrtf.exit80, !dbg !33 + +2230: ; preds = %2227 + %2231 = tail call float @llvm.nvvm.sqrt.approx.f(float %2012) #6, !dbg !33 + br label %__nv_sqrtf.exit80, !dbg !33 + +__nv_sqrtf.exit80: ; preds = %2223, %2225, %2228, %2230 + %.0.i78 = phi float [ %2224, %2223 ], [ %2226, %2225 ], [ %2229, %2228 ], [ %2231, %2230 ], !dbg !33 + %2232 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i81 = icmp eq i32 %2232, 0, !dbg !33 + %2233 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i84 = icmp eq i32 %2233, 0, !dbg !33 + br i1 %.not.i81, label %2239, label %2234, !dbg !33 + +2234: ; preds = %__nv_sqrtf.exit80 + br i1 %.not1.i84, label %2237, label %2235, !dbg !33 + +2235: ; preds = %2234 + %2236 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2013) #6, !dbg !33 + br label %__nv_sqrtf.exit85, !dbg !33 + +2237: ; preds = %2234 + %2238 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2013) #6, !dbg !33 + br label %__nv_sqrtf.exit85, !dbg !33 + +2239: ; preds = %__nv_sqrtf.exit80 + br i1 %.not1.i84, label %2242, label %2240, !dbg !33 + +2240: ; preds = %2239 + %2241 = tail call float @llvm.nvvm.sqrt.rn.f(float %2013) #6, !dbg !33 + br label %__nv_sqrtf.exit85, !dbg !33 + +2242: ; preds = %2239 + %2243 = tail call float @llvm.nvvm.sqrt.approx.f(float %2013) #6, !dbg !33 + br label %__nv_sqrtf.exit85, !dbg !33 + +__nv_sqrtf.exit85: ; preds = %2235, %2237, %2240, %2242 + %.0.i83 = phi float [ %2236, %2235 ], [ %2238, %2237 ], [ %2241, %2240 ], [ %2243, %2242 ], !dbg !33 + %2244 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i86 = icmp eq i32 %2244, 0, !dbg !33 + %2245 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i89 = icmp eq i32 %2245, 0, !dbg !33 + br i1 %.not.i86, label %2251, label %2246, !dbg !33 + +2246: ; preds = %__nv_sqrtf.exit85 + br i1 %.not1.i89, label %2249, label %2247, !dbg !33 + +2247: ; preds = %2246 + %2248 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2014) #6, !dbg !33 + br label %__nv_sqrtf.exit90, !dbg !33 + +2249: ; preds = %2246 + %2250 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2014) #6, !dbg !33 + br label %__nv_sqrtf.exit90, !dbg !33 + +2251: ; preds = %__nv_sqrtf.exit85 + br i1 %.not1.i89, label %2254, label %2252, !dbg !33 + +2252: ; preds = %2251 + %2253 = tail call float @llvm.nvvm.sqrt.rn.f(float %2014) #6, !dbg !33 + br label %__nv_sqrtf.exit90, !dbg !33 + +2254: ; preds = %2251 + %2255 = tail call float @llvm.nvvm.sqrt.approx.f(float %2014) #6, !dbg !33 + br label %__nv_sqrtf.exit90, !dbg !33 + +__nv_sqrtf.exit90: ; preds = %2247, %2249, %2252, %2254 + %.0.i88 = phi float [ %2248, %2247 ], [ %2250, %2249 ], [ %2253, %2252 ], [ %2255, %2254 ], !dbg !33 + %2256 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i91 = icmp eq i32 %2256, 0, !dbg !33 + %2257 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i94 = icmp eq i32 %2257, 0, !dbg !33 + br i1 %.not.i91, label %2263, label %2258, !dbg !33 + +2258: ; preds = %__nv_sqrtf.exit90 + br i1 %.not1.i94, label %2261, label %2259, !dbg !33 + +2259: ; preds = %2258 + %2260 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2015) #6, !dbg !33 + br label %__nv_sqrtf.exit95, !dbg !33 + +2261: ; preds = %2258 + %2262 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2015) #6, !dbg !33 + br label %__nv_sqrtf.exit95, !dbg !33 + +2263: ; preds = %__nv_sqrtf.exit90 + br i1 %.not1.i94, label %2266, label %2264, !dbg !33 + +2264: ; preds = %2263 + %2265 = tail call float @llvm.nvvm.sqrt.rn.f(float %2015) #6, !dbg !33 + br label %__nv_sqrtf.exit95, !dbg !33 + +2266: ; preds = %2263 + %2267 = tail call float @llvm.nvvm.sqrt.approx.f(float %2015) #6, !dbg !33 + br label %__nv_sqrtf.exit95, !dbg !33 + +__nv_sqrtf.exit95: ; preds = %2259, %2261, %2264, %2266 + %.0.i93 = phi float [ %2260, %2259 ], [ %2262, %2261 ], [ %2265, %2264 ], [ %2267, %2266 ], !dbg !33 + %2268 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i96 = icmp eq i32 %2268, 0, !dbg !33 + %2269 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i99 = icmp eq i32 %2269, 0, !dbg !33 + br i1 %.not.i96, label %2275, label %2270, !dbg !33 + +2270: ; preds = %__nv_sqrtf.exit95 + br i1 %.not1.i99, label %2273, label %2271, !dbg !33 + +2271: ; preds = %2270 + %2272 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2016) #6, !dbg !33 + br label %__nv_sqrtf.exit100, !dbg !33 + +2273: ; preds = %2270 + %2274 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2016) #6, !dbg !33 + br label %__nv_sqrtf.exit100, !dbg !33 + +2275: ; preds = %__nv_sqrtf.exit95 + br i1 %.not1.i99, label %2278, label %2276, !dbg !33 + +2276: ; preds = %2275 + %2277 = tail call float @llvm.nvvm.sqrt.rn.f(float %2016) #6, !dbg !33 + br label %__nv_sqrtf.exit100, !dbg !33 + +2278: ; preds = %2275 + %2279 = tail call float @llvm.nvvm.sqrt.approx.f(float %2016) #6, !dbg !33 + br label %__nv_sqrtf.exit100, !dbg !33 + +__nv_sqrtf.exit100: ; preds = %2271, %2273, %2276, %2278 + %.0.i98 = phi float [ %2272, %2271 ], [ %2274, %2273 ], [ %2277, %2276 ], [ %2279, %2278 ], !dbg !33 + %2280 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i101 = icmp eq i32 %2280, 0, !dbg !33 + %2281 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i104 = icmp eq i32 %2281, 0, !dbg !33 + br i1 %.not.i101, label %2287, label %2282, !dbg !33 + +2282: ; preds = %__nv_sqrtf.exit100 + br i1 %.not1.i104, label %2285, label %2283, !dbg !33 + +2283: ; preds = %2282 + %2284 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2017) #6, !dbg !33 + br label %__nv_sqrtf.exit105, !dbg !33 + +2285: ; preds = %2282 + %2286 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2017) #6, !dbg !33 + br label %__nv_sqrtf.exit105, !dbg !33 + +2287: ; preds = %__nv_sqrtf.exit100 + br i1 %.not1.i104, label %2290, label %2288, !dbg !33 + +2288: ; preds = %2287 + %2289 = tail call float @llvm.nvvm.sqrt.rn.f(float %2017) #6, !dbg !33 + br label %__nv_sqrtf.exit105, !dbg !33 + +2290: ; preds = %2287 + %2291 = tail call float @llvm.nvvm.sqrt.approx.f(float %2017) #6, !dbg !33 + br label %__nv_sqrtf.exit105, !dbg !33 + +__nv_sqrtf.exit105: ; preds = %2283, %2285, %2288, %2290 + %.0.i103 = phi float [ %2284, %2283 ], [ %2286, %2285 ], [ %2289, %2288 ], [ %2291, %2290 ], !dbg !33 + %2292 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i106 = icmp eq i32 %2292, 0, !dbg !33 + %2293 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i109 = icmp eq i32 %2293, 0, !dbg !33 + br i1 %.not.i106, label %2299, label %2294, !dbg !33 + +2294: ; preds = %__nv_sqrtf.exit105 + br i1 %.not1.i109, label %2297, label %2295, !dbg !33 + +2295: ; preds = %2294 + %2296 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2018) #6, !dbg !33 + br label %__nv_sqrtf.exit110, !dbg !33 + +2297: ; preds = %2294 + %2298 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2018) #6, !dbg !33 + br label %__nv_sqrtf.exit110, !dbg !33 + +2299: ; preds = %__nv_sqrtf.exit105 + br i1 %.not1.i109, label %2302, label %2300, !dbg !33 + +2300: ; preds = %2299 + %2301 = tail call float @llvm.nvvm.sqrt.rn.f(float %2018) #6, !dbg !33 + br label %__nv_sqrtf.exit110, !dbg !33 + +2302: ; preds = %2299 + %2303 = tail call float @llvm.nvvm.sqrt.approx.f(float %2018) #6, !dbg !33 + br label %__nv_sqrtf.exit110, !dbg !33 + +__nv_sqrtf.exit110: ; preds = %2295, %2297, %2300, %2302 + %.0.i108 = phi float [ %2296, %2295 ], [ %2298, %2297 ], [ %2301, %2300 ], [ %2303, %2302 ], !dbg !33 + %2304 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i111 = icmp eq i32 %2304, 0, !dbg !33 + %2305 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i114 = icmp eq i32 %2305, 0, !dbg !33 + br i1 %.not.i111, label %2311, label %2306, !dbg !33 + +2306: ; preds = %__nv_sqrtf.exit110 + br i1 %.not1.i114, label %2309, label %2307, !dbg !33 + +2307: ; preds = %2306 + %2308 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2019) #6, !dbg !33 + br label %__nv_sqrtf.exit115, !dbg !33 + +2309: ; preds = %2306 + %2310 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2019) #6, !dbg !33 + br label %__nv_sqrtf.exit115, !dbg !33 + +2311: ; preds = %__nv_sqrtf.exit110 + br i1 %.not1.i114, label %2314, label %2312, !dbg !33 + +2312: ; preds = %2311 + %2313 = tail call float @llvm.nvvm.sqrt.rn.f(float %2019) #6, !dbg !33 + br label %__nv_sqrtf.exit115, !dbg !33 + +2314: ; preds = %2311 + %2315 = tail call float @llvm.nvvm.sqrt.approx.f(float %2019) #6, !dbg !33 + br label %__nv_sqrtf.exit115, !dbg !33 + +__nv_sqrtf.exit115: ; preds = %2307, %2309, %2312, %2314 + %.0.i113 = phi float [ %2308, %2307 ], [ %2310, %2309 ], [ %2313, %2312 ], [ %2315, %2314 ], !dbg !33 + %2316 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i116 = icmp eq i32 %2316, 0, !dbg !33 + %2317 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i119 = icmp eq i32 %2317, 0, !dbg !33 + br i1 %.not.i116, label %2323, label %2318, !dbg !33 + +2318: ; preds = %__nv_sqrtf.exit115 + br i1 %.not1.i119, label %2321, label %2319, !dbg !33 + +2319: ; preds = %2318 + %2320 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2020) #6, !dbg !33 + br label %__nv_sqrtf.exit120, !dbg !33 + +2321: ; preds = %2318 + %2322 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2020) #6, !dbg !33 + br label %__nv_sqrtf.exit120, !dbg !33 + +2323: ; preds = %__nv_sqrtf.exit115 + br i1 %.not1.i119, label %2326, label %2324, !dbg !33 + +2324: ; preds = %2323 + %2325 = tail call float @llvm.nvvm.sqrt.rn.f(float %2020) #6, !dbg !33 + br label %__nv_sqrtf.exit120, !dbg !33 + +2326: ; preds = %2323 + %2327 = tail call float @llvm.nvvm.sqrt.approx.f(float %2020) #6, !dbg !33 + br label %__nv_sqrtf.exit120, !dbg !33 + +__nv_sqrtf.exit120: ; preds = %2319, %2321, %2324, %2326 + %.0.i118 = phi float [ %2320, %2319 ], [ %2322, %2321 ], [ %2325, %2324 ], [ %2327, %2326 ], !dbg !33 + %2328 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i121 = icmp eq i32 %2328, 0, !dbg !33 + %2329 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i124 = icmp eq i32 %2329, 0, !dbg !33 + br i1 %.not.i121, label %2335, label %2330, !dbg !33 + +2330: ; preds = %__nv_sqrtf.exit120 + br i1 %.not1.i124, label %2333, label %2331, !dbg !33 + +2331: ; preds = %2330 + %2332 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2021) #6, !dbg !33 + br label %__nv_sqrtf.exit125, !dbg !33 + +2333: ; preds = %2330 + %2334 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2021) #6, !dbg !33 + br label %__nv_sqrtf.exit125, !dbg !33 + +2335: ; preds = %__nv_sqrtf.exit120 + br i1 %.not1.i124, label %2338, label %2336, !dbg !33 + +2336: ; preds = %2335 + %2337 = tail call float @llvm.nvvm.sqrt.rn.f(float %2021) #6, !dbg !33 + br label %__nv_sqrtf.exit125, !dbg !33 + +2338: ; preds = %2335 + %2339 = tail call float @llvm.nvvm.sqrt.approx.f(float %2021) #6, !dbg !33 + br label %__nv_sqrtf.exit125, !dbg !33 + +__nv_sqrtf.exit125: ; preds = %2331, %2333, %2336, %2338 + %.0.i123 = phi float [ %2332, %2331 ], [ %2334, %2333 ], [ %2337, %2336 ], [ %2339, %2338 ], !dbg !33 + %2340 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i126 = icmp eq i32 %2340, 0, !dbg !33 + %2341 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i129 = icmp eq i32 %2341, 0, !dbg !33 + br i1 %.not.i126, label %2347, label %2342, !dbg !33 + +2342: ; preds = %__nv_sqrtf.exit125 + br i1 %.not1.i129, label %2345, label %2343, !dbg !33 + +2343: ; preds = %2342 + %2344 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2022) #6, !dbg !33 + br label %__nv_sqrtf.exit130, !dbg !33 + +2345: ; preds = %2342 + %2346 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2022) #6, !dbg !33 + br label %__nv_sqrtf.exit130, !dbg !33 + +2347: ; preds = %__nv_sqrtf.exit125 + br i1 %.not1.i129, label %2350, label %2348, !dbg !33 + +2348: ; preds = %2347 + %2349 = tail call float @llvm.nvvm.sqrt.rn.f(float %2022) #6, !dbg !33 + br label %__nv_sqrtf.exit130, !dbg !33 + +2350: ; preds = %2347 + %2351 = tail call float @llvm.nvvm.sqrt.approx.f(float %2022) #6, !dbg !33 + br label %__nv_sqrtf.exit130, !dbg !33 + +__nv_sqrtf.exit130: ; preds = %2343, %2345, %2348, %2350 + %.0.i128 = phi float [ %2344, %2343 ], [ %2346, %2345 ], [ %2349, %2348 ], [ %2351, %2350 ], !dbg !33 + %2352 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i131 = icmp eq i32 %2352, 0, !dbg !33 + %2353 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i134 = icmp eq i32 %2353, 0, !dbg !33 + br i1 %.not.i131, label %2359, label %2354, !dbg !33 + +2354: ; preds = %__nv_sqrtf.exit130 + br i1 %.not1.i134, label %2357, label %2355, !dbg !33 + +2355: ; preds = %2354 + %2356 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2023) #6, !dbg !33 + br label %__nv_sqrtf.exit135, !dbg !33 + +2357: ; preds = %2354 + %2358 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2023) #6, !dbg !33 + br label %__nv_sqrtf.exit135, !dbg !33 + +2359: ; preds = %__nv_sqrtf.exit130 + br i1 %.not1.i134, label %2362, label %2360, !dbg !33 + +2360: ; preds = %2359 + %2361 = tail call float @llvm.nvvm.sqrt.rn.f(float %2023) #6, !dbg !33 + br label %__nv_sqrtf.exit135, !dbg !33 + +2362: ; preds = %2359 + %2363 = tail call float @llvm.nvvm.sqrt.approx.f(float %2023) #6, !dbg !33 + br label %__nv_sqrtf.exit135, !dbg !33 + +__nv_sqrtf.exit135: ; preds = %2355, %2357, %2360, %2362 + %.0.i133 = phi float [ %2356, %2355 ], [ %2358, %2357 ], [ %2361, %2360 ], [ %2363, %2362 ], !dbg !33 + %2364 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i136 = icmp eq i32 %2364, 0, !dbg !33 + %2365 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i139 = icmp eq i32 %2365, 0, !dbg !33 + br i1 %.not.i136, label %2371, label %2366, !dbg !33 + +2366: ; preds = %__nv_sqrtf.exit135 + br i1 %.not1.i139, label %2369, label %2367, !dbg !33 + +2367: ; preds = %2366 + %2368 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2024) #6, !dbg !33 + br label %__nv_sqrtf.exit140, !dbg !33 + +2369: ; preds = %2366 + %2370 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2024) #6, !dbg !33 + br label %__nv_sqrtf.exit140, !dbg !33 + +2371: ; preds = %__nv_sqrtf.exit135 + br i1 %.not1.i139, label %2374, label %2372, !dbg !33 + +2372: ; preds = %2371 + %2373 = tail call float @llvm.nvvm.sqrt.rn.f(float %2024) #6, !dbg !33 + br label %__nv_sqrtf.exit140, !dbg !33 + +2374: ; preds = %2371 + %2375 = tail call float @llvm.nvvm.sqrt.approx.f(float %2024) #6, !dbg !33 + br label %__nv_sqrtf.exit140, !dbg !33 + +__nv_sqrtf.exit140: ; preds = %2367, %2369, %2372, %2374 + %.0.i138 = phi float [ %2368, %2367 ], [ %2370, %2369 ], [ %2373, %2372 ], [ %2375, %2374 ], !dbg !33 + %2376 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i141 = icmp eq i32 %2376, 0, !dbg !33 + %2377 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i144 = icmp eq i32 %2377, 0, !dbg !33 + br i1 %.not.i141, label %2383, label %2378, !dbg !33 + +2378: ; preds = %__nv_sqrtf.exit140 + br i1 %.not1.i144, label %2381, label %2379, !dbg !33 + +2379: ; preds = %2378 + %2380 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2025) #6, !dbg !33 + br label %__nv_sqrtf.exit145, !dbg !33 + +2381: ; preds = %2378 + %2382 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2025) #6, !dbg !33 + br label %__nv_sqrtf.exit145, !dbg !33 + +2383: ; preds = %__nv_sqrtf.exit140 + br i1 %.not1.i144, label %2386, label %2384, !dbg !33 + +2384: ; preds = %2383 + %2385 = tail call float @llvm.nvvm.sqrt.rn.f(float %2025) #6, !dbg !33 + br label %__nv_sqrtf.exit145, !dbg !33 + +2386: ; preds = %2383 + %2387 = tail call float @llvm.nvvm.sqrt.approx.f(float %2025) #6, !dbg !33 + br label %__nv_sqrtf.exit145, !dbg !33 + +__nv_sqrtf.exit145: ; preds = %2379, %2381, %2384, %2386 + %.0.i143 = phi float [ %2380, %2379 ], [ %2382, %2381 ], [ %2385, %2384 ], [ %2387, %2386 ], !dbg !33 + %2388 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i146 = icmp eq i32 %2388, 0, !dbg !33 + %2389 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i149 = icmp eq i32 %2389, 0, !dbg !33 + br i1 %.not.i146, label %2395, label %2390, !dbg !33 + +2390: ; preds = %__nv_sqrtf.exit145 + br i1 %.not1.i149, label %2393, label %2391, !dbg !33 + +2391: ; preds = %2390 + %2392 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2026) #6, !dbg !33 + br label %__nv_sqrtf.exit150, !dbg !33 + +2393: ; preds = %2390 + %2394 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2026) #6, !dbg !33 + br label %__nv_sqrtf.exit150, !dbg !33 + +2395: ; preds = %__nv_sqrtf.exit145 + br i1 %.not1.i149, label %2398, label %2396, !dbg !33 + +2396: ; preds = %2395 + %2397 = tail call float @llvm.nvvm.sqrt.rn.f(float %2026) #6, !dbg !33 + br label %__nv_sqrtf.exit150, !dbg !33 + +2398: ; preds = %2395 + %2399 = tail call float @llvm.nvvm.sqrt.approx.f(float %2026) #6, !dbg !33 + br label %__nv_sqrtf.exit150, !dbg !33 + +__nv_sqrtf.exit150: ; preds = %2391, %2393, %2396, %2398 + %.0.i148 = phi float [ %2392, %2391 ], [ %2394, %2393 ], [ %2397, %2396 ], [ %2399, %2398 ], !dbg !33 + %2400 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i151 = icmp eq i32 %2400, 0, !dbg !33 + %2401 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i154 = icmp eq i32 %2401, 0, !dbg !33 + br i1 %.not.i151, label %2407, label %2402, !dbg !33 + +2402: ; preds = %__nv_sqrtf.exit150 + br i1 %.not1.i154, label %2405, label %2403, !dbg !33 + +2403: ; preds = %2402 + %2404 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2027) #6, !dbg !33 + br label %__nv_sqrtf.exit155, !dbg !33 + +2405: ; preds = %2402 + %2406 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2027) #6, !dbg !33 + br label %__nv_sqrtf.exit155, !dbg !33 + +2407: ; preds = %__nv_sqrtf.exit150 + br i1 %.not1.i154, label %2410, label %2408, !dbg !33 + +2408: ; preds = %2407 + %2409 = tail call float @llvm.nvvm.sqrt.rn.f(float %2027) #6, !dbg !33 + br label %__nv_sqrtf.exit155, !dbg !33 + +2410: ; preds = %2407 + %2411 = tail call float @llvm.nvvm.sqrt.approx.f(float %2027) #6, !dbg !33 + br label %__nv_sqrtf.exit155, !dbg !33 + +__nv_sqrtf.exit155: ; preds = %2403, %2405, %2408, %2410 + %.0.i153 = phi float [ %2404, %2403 ], [ %2406, %2405 ], [ %2409, %2408 ], [ %2411, %2410 ], !dbg !33 + %2412 = sext i32 %4 to i64, !dbg !18 + %2413 = shl nuw nsw i32 %11, 1, !dbg !16 + %2414 = and i32 %2413, 126, !dbg !16 + %2415 = or disjoint i32 %2414, 1, !dbg !16 + %2416 = zext nneg i32 %2415 to i64, !dbg !17 + %2417 = or disjoint i64 %10, %2416, !dbg !17 + %2418 = getelementptr float, ptr addrspace(1) %2, i64 %2417, !dbg !34 + %2419 = zext nneg i32 %2414 to i64, !dbg !17 + %2420 = or disjoint i64 %10, %2419, !dbg !17 + %2421 = getelementptr float, ptr addrspace(1) %2, i64 %2420, !dbg !34 + %2422 = getelementptr bfloat, ptr addrspace(1) %1, i64 %106, !dbg !35 + %2423 = getelementptr bfloat, ptr addrspace(1) %2422, i64 %139, !dbg !36 + %2424 = getelementptr bfloat, ptr addrspace(1) %1, i64 %104, !dbg !35 + %2425 = getelementptr bfloat, ptr addrspace(1) %2424, i64 %139, !dbg !36 + %2426 = getelementptr bfloat, ptr addrspace(1) %1, i64 %102, !dbg !35 + %2427 = getelementptr bfloat, ptr addrspace(1) %2426, i64 %139, !dbg !36 + %2428 = getelementptr bfloat, ptr addrspace(1) %1, i64 %100, !dbg !35 + %2429 = getelementptr bfloat, ptr addrspace(1) %2428, i64 %139, !dbg !36 + %2430 = getelementptr bfloat, ptr addrspace(1) %1, i64 %98, !dbg !35 + %2431 = getelementptr bfloat, ptr addrspace(1) %2430, i64 %139, !dbg !36 + %2432 = getelementptr bfloat, ptr addrspace(1) %1, i64 %96, !dbg !35 + %2433 = getelementptr bfloat, ptr addrspace(1) %2432, i64 %139, !dbg !36 + %2434 = getelementptr bfloat, ptr addrspace(1) %1, i64 %94, !dbg !35 + %2435 = getelementptr bfloat, ptr addrspace(1) %2434, i64 %139, !dbg !36 + %2436 = getelementptr bfloat, ptr addrspace(1) %1, i64 %92, !dbg !35 + %2437 = getelementptr bfloat, ptr addrspace(1) %2436, i64 %139, !dbg !36 + %2438 = getelementptr bfloat, ptr addrspace(1) %1, i64 %90, !dbg !35 + %2439 = getelementptr bfloat, ptr addrspace(1) %2438, i64 %139, !dbg !36 + %2440 = getelementptr bfloat, ptr addrspace(1) %1, i64 %88, !dbg !35 + %2441 = getelementptr bfloat, ptr addrspace(1) %2440, i64 %139, !dbg !36 + %2442 = getelementptr bfloat, ptr addrspace(1) %1, i64 %86, !dbg !35 + %2443 = getelementptr bfloat, ptr addrspace(1) %2442, i64 %139, !dbg !36 + %2444 = getelementptr bfloat, ptr addrspace(1) %1, i64 %84, !dbg !35 + %2445 = getelementptr bfloat, ptr addrspace(1) %2444, i64 %139, !dbg !36 + %2446 = getelementptr bfloat, ptr addrspace(1) %1, i64 %82, !dbg !35 + %2447 = getelementptr bfloat, ptr addrspace(1) %2446, i64 %139, !dbg !36 + %2448 = getelementptr bfloat, ptr addrspace(1) %1, i64 %80, !dbg !35 + %2449 = getelementptr bfloat, ptr addrspace(1) %2448, i64 %139, !dbg !36 + %2450 = getelementptr bfloat, ptr addrspace(1) %1, i64 %78, !dbg !35 + %2451 = getelementptr bfloat, ptr addrspace(1) %2450, i64 %139, !dbg !36 + %2452 = getelementptr bfloat, ptr addrspace(1) %1, i64 %76, !dbg !35 + %2453 = getelementptr bfloat, ptr addrspace(1) %2452, i64 %139, !dbg !36 + %2454 = getelementptr bfloat, ptr addrspace(1) %1, i64 %74, !dbg !35 + %2455 = getelementptr bfloat, ptr addrspace(1) %2454, i64 %139, !dbg !36 + %2456 = getelementptr bfloat, ptr addrspace(1) %1, i64 %72, !dbg !35 + %2457 = getelementptr bfloat, ptr addrspace(1) %2456, i64 %139, !dbg !36 + %2458 = getelementptr bfloat, ptr addrspace(1) %1, i64 %70, !dbg !35 + %2459 = getelementptr bfloat, ptr addrspace(1) %2458, i64 %139, !dbg !36 + %2460 = getelementptr bfloat, ptr addrspace(1) %1, i64 %68, !dbg !35 + %2461 = getelementptr bfloat, ptr addrspace(1) %2460, i64 %139, !dbg !36 + %2462 = getelementptr bfloat, ptr addrspace(1) %1, i64 %66, !dbg !35 + %2463 = getelementptr bfloat, ptr addrspace(1) %2462, i64 %139, !dbg !36 + %2464 = getelementptr bfloat, ptr addrspace(1) %1, i64 %64, !dbg !35 + %2465 = getelementptr bfloat, ptr addrspace(1) %2464, i64 %139, !dbg !36 + %2466 = getelementptr bfloat, ptr addrspace(1) %1, i64 %62, !dbg !35 + %2467 = getelementptr bfloat, ptr addrspace(1) %2466, i64 %139, !dbg !36 + %2468 = getelementptr bfloat, ptr addrspace(1) %1, i64 %60, !dbg !35 + %2469 = getelementptr bfloat, ptr addrspace(1) %2468, i64 %139, !dbg !36 + %2470 = getelementptr bfloat, ptr addrspace(1) %1, i64 %58, !dbg !35 + %2471 = getelementptr bfloat, ptr addrspace(1) %2470, i64 %139, !dbg !36 + %2472 = getelementptr bfloat, ptr addrspace(1) %1, i64 %56, !dbg !35 + %2473 = getelementptr bfloat, ptr addrspace(1) %2472, i64 %139, !dbg !36 + %2474 = getelementptr bfloat, ptr addrspace(1) %1, i64 %54, !dbg !35 + %2475 = getelementptr bfloat, ptr addrspace(1) %2474, i64 %139, !dbg !36 + %2476 = getelementptr bfloat, ptr addrspace(1) %1, i64 %52, !dbg !35 + %2477 = getelementptr bfloat, ptr addrspace(1) %2476, i64 %139, !dbg !36 + %2478 = getelementptr bfloat, ptr addrspace(1) %1, i64 %50, !dbg !35 + %2479 = getelementptr bfloat, ptr addrspace(1) %2478, i64 %139, !dbg !36 + %2480 = getelementptr bfloat, ptr addrspace(1) %1, i64 %48, !dbg !35 + %2481 = getelementptr bfloat, ptr addrspace(1) %2480, i64 %139, !dbg !36 + %2482 = getelementptr bfloat, ptr addrspace(1) %1, i64 %46, !dbg !35 + %2483 = getelementptr bfloat, ptr addrspace(1) %2482, i64 %139, !dbg !36 + %2484 = getelementptr bfloat, ptr addrspace(1) %1, i64 %44, !dbg !35 + %2485 = getelementptr bfloat, ptr addrspace(1) %2484, i64 %139, !dbg !36 + %2486 = icmp slt i64 %2417, %2412, !dbg !18 + %2487 = icmp slt i64 %2420, %2412, !dbg !18 + %2488 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i), !dbg !37 + %2489 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i3), !dbg !37 + %2490 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i8), !dbg !37 + %2491 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i13), !dbg !37 + %2492 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i18), !dbg !37 + %2493 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i23), !dbg !37 + %2494 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i28), !dbg !37 + %2495 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i33), !dbg !37 + %2496 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i38), !dbg !37 + %2497 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i43), !dbg !37 + %2498 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i48), !dbg !37 + %2499 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i53), !dbg !37 + %2500 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i58), !dbg !37 + %2501 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i63), !dbg !37 + %2502 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i68), !dbg !37 + %2503 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i73), !dbg !37 + %2504 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i78), !dbg !37 + %2505 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i83), !dbg !37 + %2506 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i88), !dbg !37 + %2507 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i93), !dbg !37 + %2508 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i98), !dbg !37 + %2509 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i103), !dbg !37 + %2510 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i108), !dbg !37 + %2511 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i113), !dbg !37 + %2512 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i118), !dbg !37 + %2513 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i123), !dbg !37 + %2514 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i128), !dbg !37 + %2515 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i133), !dbg !37 + %2516 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i138), !dbg !37 + %2517 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i143), !dbg !37 + %2518 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i148), !dbg !37 + %2519 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i153), !dbg !37 + %2520 = fmul float %2488, %748, !dbg !38 + %2521 = fmul float %2488, %749, !dbg !38 + %2522 = fmul float %2488, %750, !dbg !38 + %2523 = fmul float %2488, %751, !dbg !38 + %2524 = fmul float %2488, %752, !dbg !38 + %2525 = fmul float %2488, %753, !dbg !38 + %2526 = fmul float %2488, %754, !dbg !38 + %2527 = fmul float %2488, %755, !dbg !38 + %2528 = fmul float %2489, %756, !dbg !38 + %2529 = fmul float %2489, %757, !dbg !38 + %2530 = fmul float %2489, %758, !dbg !38 + %2531 = fmul float %2489, %759, !dbg !38 + %2532 = fmul float %2489, %760, !dbg !38 + %2533 = fmul float %2489, %761, !dbg !38 + %2534 = fmul float %2489, %762, !dbg !38 + %2535 = fmul float %2489, %763, !dbg !38 + %2536 = fmul float %2490, %764, !dbg !38 + %2537 = fmul float %2490, %765, !dbg !38 + %2538 = fmul float %2490, %766, !dbg !38 + %2539 = fmul float %2490, %767, !dbg !38 + %2540 = fmul float %2490, %768, !dbg !38 + %2541 = fmul float %2490, %769, !dbg !38 + %2542 = fmul float %2490, %770, !dbg !38 + %2543 = fmul float %2490, %771, !dbg !38 + %2544 = fmul float %2491, %772, !dbg !38 + %2545 = fmul float %2491, %773, !dbg !38 + %2546 = fmul float %2491, %774, !dbg !38 + %2547 = fmul float %2491, %775, !dbg !38 + %2548 = fmul float %2491, %776, !dbg !38 + %2549 = fmul float %2491, %777, !dbg !38 + %2550 = fmul float %2491, %778, !dbg !38 + %2551 = fmul float %2491, %779, !dbg !38 + %2552 = fmul float %2492, %780, !dbg !38 + %2553 = fmul float %2492, %781, !dbg !38 + %2554 = fmul float %2492, %782, !dbg !38 + %2555 = fmul float %2492, %783, !dbg !38 + %2556 = fmul float %2492, %784, !dbg !38 + %2557 = fmul float %2492, %785, !dbg !38 + %2558 = fmul float %2492, %786, !dbg !38 + %2559 = fmul float %2492, %787, !dbg !38 + %2560 = fmul float %2493, %788, !dbg !38 + %2561 = fmul float %2493, %789, !dbg !38 + %2562 = fmul float %2493, %790, !dbg !38 + %2563 = fmul float %2493, %791, !dbg !38 + %2564 = fmul float %2493, %792, !dbg !38 + %2565 = fmul float %2493, %793, !dbg !38 + %2566 = fmul float %2493, %794, !dbg !38 + %2567 = fmul float %2493, %795, !dbg !38 + %2568 = fmul float %2494, %796, !dbg !38 + %2569 = fmul float %2494, %797, !dbg !38 + %2570 = fmul float %2494, %798, !dbg !38 + %2571 = fmul float %2494, %799, !dbg !38 + %2572 = fmul float %2494, %800, !dbg !38 + %2573 = fmul float %2494, %801, !dbg !38 + %2574 = fmul float %2494, %802, !dbg !38 + %2575 = fmul float %2494, %803, !dbg !38 + %2576 = fmul float %2495, %804, !dbg !38 + %2577 = fmul float %2495, %805, !dbg !38 + %2578 = fmul float %2495, %806, !dbg !38 + %2579 = fmul float %2495, %807, !dbg !38 + %2580 = fmul float %2495, %808, !dbg !38 + %2581 = fmul float %2495, %809, !dbg !38 + %2582 = fmul float %2495, %810, !dbg !38 + %2583 = fmul float %2495, %811, !dbg !38 + %2584 = fmul float %2496, %812, !dbg !38 + %2585 = fmul float %2496, %813, !dbg !38 + %2586 = fmul float %2496, %814, !dbg !38 + %2587 = fmul float %2496, %815, !dbg !38 + %2588 = fmul float %2496, %816, !dbg !38 + %2589 = fmul float %2496, %817, !dbg !38 + %2590 = fmul float %2496, %818, !dbg !38 + %2591 = fmul float %2496, %819, !dbg !38 + %2592 = fmul float %2497, %820, !dbg !38 + %2593 = fmul float %2497, %821, !dbg !38 + %2594 = fmul float %2497, %822, !dbg !38 + %2595 = fmul float %2497, %823, !dbg !38 + %2596 = fmul float %2497, %824, !dbg !38 + %2597 = fmul float %2497, %825, !dbg !38 + %2598 = fmul float %2497, %826, !dbg !38 + %2599 = fmul float %2497, %827, !dbg !38 + %2600 = fmul float %2498, %828, !dbg !38 + %2601 = fmul float %2498, %829, !dbg !38 + %2602 = fmul float %2498, %830, !dbg !38 + %2603 = fmul float %2498, %831, !dbg !38 + %2604 = fmul float %2498, %832, !dbg !38 + %2605 = fmul float %2498, %833, !dbg !38 + %2606 = fmul float %2498, %834, !dbg !38 + %2607 = fmul float %2498, %835, !dbg !38 + %2608 = fmul float %2499, %836, !dbg !38 + %2609 = fmul float %2499, %837, !dbg !38 + %2610 = fmul float %2499, %838, !dbg !38 + %2611 = fmul float %2499, %839, !dbg !38 + %2612 = fmul float %2499, %840, !dbg !38 + %2613 = fmul float %2499, %841, !dbg !38 + %2614 = fmul float %2499, %842, !dbg !38 + %2615 = fmul float %2499, %843, !dbg !38 + %2616 = fmul float %2500, %844, !dbg !38 + %2617 = fmul float %2500, %845, !dbg !38 + %2618 = fmul float %2500, %846, !dbg !38 + %2619 = fmul float %2500, %847, !dbg !38 + %2620 = fmul float %2500, %848, !dbg !38 + %2621 = fmul float %2500, %849, !dbg !38 + %2622 = fmul float %2500, %850, !dbg !38 + %2623 = fmul float %2500, %851, !dbg !38 + %2624 = fmul float %2501, %852, !dbg !38 + %2625 = fmul float %2501, %853, !dbg !38 + %2626 = fmul float %2501, %854, !dbg !38 + %2627 = fmul float %2501, %855, !dbg !38 + %2628 = fmul float %2501, %856, !dbg !38 + %2629 = fmul float %2501, %857, !dbg !38 + %2630 = fmul float %2501, %858, !dbg !38 + %2631 = fmul float %2501, %859, !dbg !38 + %2632 = fmul float %2502, %860, !dbg !38 + %2633 = fmul float %2502, %861, !dbg !38 + %2634 = fmul float %2502, %862, !dbg !38 + %2635 = fmul float %2502, %863, !dbg !38 + %2636 = fmul float %2502, %864, !dbg !38 + %2637 = fmul float %2502, %865, !dbg !38 + %2638 = fmul float %2502, %866, !dbg !38 + %2639 = fmul float %2502, %867, !dbg !38 + %2640 = fmul float %2503, %868, !dbg !38 + %2641 = fmul float %2503, %869, !dbg !38 + %2642 = fmul float %2503, %870, !dbg !38 + %2643 = fmul float %2503, %871, !dbg !38 + %2644 = fmul float %2503, %872, !dbg !38 + %2645 = fmul float %2503, %873, !dbg !38 + %2646 = fmul float %2503, %874, !dbg !38 + %2647 = fmul float %2503, %875, !dbg !38 + %2648 = fmul float %2504, %876, !dbg !38 + %2649 = fmul float %2504, %877, !dbg !38 + %2650 = fmul float %2504, %878, !dbg !38 + %2651 = fmul float %2504, %879, !dbg !38 + %2652 = fmul float %2504, %880, !dbg !38 + %2653 = fmul float %2504, %881, !dbg !38 + %2654 = fmul float %2504, %882, !dbg !38 + %2655 = fmul float %2504, %883, !dbg !38 + %2656 = fmul float %2505, %884, !dbg !38 + %2657 = fmul float %2505, %885, !dbg !38 + %2658 = fmul float %2505, %886, !dbg !38 + %2659 = fmul float %2505, %887, !dbg !38 + %2660 = fmul float %2505, %888, !dbg !38 + %2661 = fmul float %2505, %889, !dbg !38 + %2662 = fmul float %2505, %890, !dbg !38 + %2663 = fmul float %2505, %891, !dbg !38 + %2664 = fmul float %2506, %892, !dbg !38 + %2665 = fmul float %2506, %893, !dbg !38 + %2666 = fmul float %2506, %894, !dbg !38 + %2667 = fmul float %2506, %895, !dbg !38 + %2668 = fmul float %2506, %896, !dbg !38 + %2669 = fmul float %2506, %897, !dbg !38 + %2670 = fmul float %2506, %898, !dbg !38 + %2671 = fmul float %2506, %899, !dbg !38 + %2672 = fmul float %2507, %900, !dbg !38 + %2673 = fmul float %2507, %901, !dbg !38 + %2674 = fmul float %2507, %902, !dbg !38 + %2675 = fmul float %2507, %903, !dbg !38 + %2676 = fmul float %2507, %904, !dbg !38 + %2677 = fmul float %2507, %905, !dbg !38 + %2678 = fmul float %2507, %906, !dbg !38 + %2679 = fmul float %2507, %907, !dbg !38 + %2680 = fmul float %2508, %908, !dbg !38 + %2681 = fmul float %2508, %909, !dbg !38 + %2682 = fmul float %2508, %910, !dbg !38 + %2683 = fmul float %2508, %911, !dbg !38 + %2684 = fmul float %2508, %912, !dbg !38 + %2685 = fmul float %2508, %913, !dbg !38 + %2686 = fmul float %2508, %914, !dbg !38 + %2687 = fmul float %2508, %915, !dbg !38 + %2688 = fmul float %2509, %916, !dbg !38 + %2689 = fmul float %2509, %917, !dbg !38 + %2690 = fmul float %2509, %918, !dbg !38 + %2691 = fmul float %2509, %919, !dbg !38 + %2692 = fmul float %2509, %920, !dbg !38 + %2693 = fmul float %2509, %921, !dbg !38 + %2694 = fmul float %2509, %922, !dbg !38 + %2695 = fmul float %2509, %923, !dbg !38 + %2696 = fmul float %2510, %924, !dbg !38 + %2697 = fmul float %2510, %925, !dbg !38 + %2698 = fmul float %2510, %926, !dbg !38 + %2699 = fmul float %2510, %927, !dbg !38 + %2700 = fmul float %2510, %928, !dbg !38 + %2701 = fmul float %2510, %929, !dbg !38 + %2702 = fmul float %2510, %930, !dbg !38 + %2703 = fmul float %2510, %931, !dbg !38 + %2704 = fmul float %2511, %932, !dbg !38 + %2705 = fmul float %2511, %933, !dbg !38 + %2706 = fmul float %2511, %934, !dbg !38 + %2707 = fmul float %2511, %935, !dbg !38 + %2708 = fmul float %2511, %936, !dbg !38 + %2709 = fmul float %2511, %937, !dbg !38 + %2710 = fmul float %2511, %938, !dbg !38 + %2711 = fmul float %2511, %939, !dbg !38 + %2712 = fmul float %2512, %940, !dbg !38 + %2713 = fmul float %2512, %941, !dbg !38 + %2714 = fmul float %2512, %942, !dbg !38 + %2715 = fmul float %2512, %943, !dbg !38 + %2716 = fmul float %2512, %944, !dbg !38 + %2717 = fmul float %2512, %945, !dbg !38 + %2718 = fmul float %2512, %946, !dbg !38 + %2719 = fmul float %2512, %947, !dbg !38 + %2720 = fmul float %2513, %948, !dbg !38 + %2721 = fmul float %2513, %949, !dbg !38 + %2722 = fmul float %2513, %950, !dbg !38 + %2723 = fmul float %2513, %951, !dbg !38 + %2724 = fmul float %2513, %952, !dbg !38 + %2725 = fmul float %2513, %953, !dbg !38 + %2726 = fmul float %2513, %954, !dbg !38 + %2727 = fmul float %2513, %955, !dbg !38 + %2728 = fmul float %2514, %956, !dbg !38 + %2729 = fmul float %2514, %957, !dbg !38 + %2730 = fmul float %2514, %958, !dbg !38 + %2731 = fmul float %2514, %959, !dbg !38 + %2732 = fmul float %2514, %960, !dbg !38 + %2733 = fmul float %2514, %961, !dbg !38 + %2734 = fmul float %2514, %962, !dbg !38 + %2735 = fmul float %2514, %963, !dbg !38 + %2736 = fmul float %2515, %964, !dbg !38 + %2737 = fmul float %2515, %965, !dbg !38 + %2738 = fmul float %2515, %966, !dbg !38 + %2739 = fmul float %2515, %967, !dbg !38 + %2740 = fmul float %2515, %968, !dbg !38 + %2741 = fmul float %2515, %969, !dbg !38 + %2742 = fmul float %2515, %970, !dbg !38 + %2743 = fmul float %2515, %971, !dbg !38 + %2744 = fmul float %2516, %972, !dbg !38 + %2745 = fmul float %2516, %973, !dbg !38 + %2746 = fmul float %2516, %974, !dbg !38 + %2747 = fmul float %2516, %975, !dbg !38 + %2748 = fmul float %2516, %976, !dbg !38 + %2749 = fmul float %2516, %977, !dbg !38 + %2750 = fmul float %2516, %978, !dbg !38 + %2751 = fmul float %2516, %979, !dbg !38 + %2752 = fmul float %2517, %980, !dbg !38 + %2753 = fmul float %2517, %981, !dbg !38 + %2754 = fmul float %2517, %982, !dbg !38 + %2755 = fmul float %2517, %983, !dbg !38 + %2756 = fmul float %2517, %984, !dbg !38 + %2757 = fmul float %2517, %985, !dbg !38 + %2758 = fmul float %2517, %986, !dbg !38 + %2759 = fmul float %2517, %987, !dbg !38 + %2760 = fmul float %2518, %988, !dbg !38 + %2761 = fmul float %2518, %989, !dbg !38 + %2762 = fmul float %2518, %990, !dbg !38 + %2763 = fmul float %2518, %991, !dbg !38 + %2764 = fmul float %2518, %992, !dbg !38 + %2765 = fmul float %2518, %993, !dbg !38 + %2766 = fmul float %2518, %994, !dbg !38 + %2767 = fmul float %2518, %995, !dbg !38 + %2768 = fmul float %2519, %996, !dbg !38 + %2769 = fmul float %2519, %997, !dbg !38 + %2770 = fmul float %2519, %998, !dbg !38 + %2771 = fmul float %2519, %999, !dbg !38 + %2772 = fmul float %2519, %1000, !dbg !38 + %2773 = fmul float %2519, %1001, !dbg !38 + %2774 = fmul float %2519, %1002, !dbg !38 + %2775 = fmul float %2519, %1003, !dbg !38 + %2776 = fptrunc float %2520 to bfloat, !dbg !39 + %2777 = fptrunc float %2521 to bfloat, !dbg !39 + %2778 = fptrunc float %2522 to bfloat, !dbg !39 + %2779 = fptrunc float %2523 to bfloat, !dbg !39 + %2780 = fptrunc float %2524 to bfloat, !dbg !39 + %2781 = fptrunc float %2525 to bfloat, !dbg !39 + %2782 = fptrunc float %2526 to bfloat, !dbg !39 + %2783 = fptrunc float %2527 to bfloat, !dbg !39 + %2784 = fptrunc float %2528 to bfloat, !dbg !39 + %2785 = fptrunc float %2529 to bfloat, !dbg !39 + %2786 = fptrunc float %2530 to bfloat, !dbg !39 + %2787 = fptrunc float %2531 to bfloat, !dbg !39 + %2788 = fptrunc float %2532 to bfloat, !dbg !39 + %2789 = fptrunc float %2533 to bfloat, !dbg !39 + %2790 = fptrunc float %2534 to bfloat, !dbg !39 + %2791 = fptrunc float %2535 to bfloat, !dbg !39 + %2792 = fptrunc float %2536 to bfloat, !dbg !39 + %2793 = fptrunc float %2537 to bfloat, !dbg !39 + %2794 = fptrunc float %2538 to bfloat, !dbg !39 + %2795 = fptrunc float %2539 to bfloat, !dbg !39 + %2796 = fptrunc float %2540 to bfloat, !dbg !39 + %2797 = fptrunc float %2541 to bfloat, !dbg !39 + %2798 = fptrunc float %2542 to bfloat, !dbg !39 + %2799 = fptrunc float %2543 to bfloat, !dbg !39 + %2800 = fptrunc float %2544 to bfloat, !dbg !39 + %2801 = fptrunc float %2545 to bfloat, !dbg !39 + %2802 = fptrunc float %2546 to bfloat, !dbg !39 + %2803 = fptrunc float %2547 to bfloat, !dbg !39 + %2804 = fptrunc float %2548 to bfloat, !dbg !39 + %2805 = fptrunc float %2549 to bfloat, !dbg !39 + %2806 = fptrunc float %2550 to bfloat, !dbg !39 + %2807 = fptrunc float %2551 to bfloat, !dbg !39 + %2808 = fptrunc float %2552 to bfloat, !dbg !39 + %2809 = fptrunc float %2553 to bfloat, !dbg !39 + %2810 = fptrunc float %2554 to bfloat, !dbg !39 + %2811 = fptrunc float %2555 to bfloat, !dbg !39 + %2812 = fptrunc float %2556 to bfloat, !dbg !39 + %2813 = fptrunc float %2557 to bfloat, !dbg !39 + %2814 = fptrunc float %2558 to bfloat, !dbg !39 + %2815 = fptrunc float %2559 to bfloat, !dbg !39 + %2816 = fptrunc float %2560 to bfloat, !dbg !39 + %2817 = fptrunc float %2561 to bfloat, !dbg !39 + %2818 = fptrunc float %2562 to bfloat, !dbg !39 + %2819 = fptrunc float %2563 to bfloat, !dbg !39 + %2820 = fptrunc float %2564 to bfloat, !dbg !39 + %2821 = fptrunc float %2565 to bfloat, !dbg !39 + %2822 = fptrunc float %2566 to bfloat, !dbg !39 + %2823 = fptrunc float %2567 to bfloat, !dbg !39 + %2824 = fptrunc float %2568 to bfloat, !dbg !39 + %2825 = fptrunc float %2569 to bfloat, !dbg !39 + %2826 = fptrunc float %2570 to bfloat, !dbg !39 + %2827 = fptrunc float %2571 to bfloat, !dbg !39 + %2828 = fptrunc float %2572 to bfloat, !dbg !39 + %2829 = fptrunc float %2573 to bfloat, !dbg !39 + %2830 = fptrunc float %2574 to bfloat, !dbg !39 + %2831 = fptrunc float %2575 to bfloat, !dbg !39 + %2832 = fptrunc float %2576 to bfloat, !dbg !39 + %2833 = fptrunc float %2577 to bfloat, !dbg !39 + %2834 = fptrunc float %2578 to bfloat, !dbg !39 + %2835 = fptrunc float %2579 to bfloat, !dbg !39 + %2836 = fptrunc float %2580 to bfloat, !dbg !39 + %2837 = fptrunc float %2581 to bfloat, !dbg !39 + %2838 = fptrunc float %2582 to bfloat, !dbg !39 + %2839 = fptrunc float %2583 to bfloat, !dbg !39 + %2840 = fptrunc float %2584 to bfloat, !dbg !39 + %2841 = fptrunc float %2585 to bfloat, !dbg !39 + %2842 = fptrunc float %2586 to bfloat, !dbg !39 + %2843 = fptrunc float %2587 to bfloat, !dbg !39 + %2844 = fptrunc float %2588 to bfloat, !dbg !39 + %2845 = fptrunc float %2589 to bfloat, !dbg !39 + %2846 = fptrunc float %2590 to bfloat, !dbg !39 + %2847 = fptrunc float %2591 to bfloat, !dbg !39 + %2848 = fptrunc float %2592 to bfloat, !dbg !39 + %2849 = fptrunc float %2593 to bfloat, !dbg !39 + %2850 = fptrunc float %2594 to bfloat, !dbg !39 + %2851 = fptrunc float %2595 to bfloat, !dbg !39 + %2852 = fptrunc float %2596 to bfloat, !dbg !39 + %2853 = fptrunc float %2597 to bfloat, !dbg !39 + %2854 = fptrunc float %2598 to bfloat, !dbg !39 + %2855 = fptrunc float %2599 to bfloat, !dbg !39 + %2856 = fptrunc float %2600 to bfloat, !dbg !39 + %2857 = fptrunc float %2601 to bfloat, !dbg !39 + %2858 = fptrunc float %2602 to bfloat, !dbg !39 + %2859 = fptrunc float %2603 to bfloat, !dbg !39 + %2860 = fptrunc float %2604 to bfloat, !dbg !39 + %2861 = fptrunc float %2605 to bfloat, !dbg !39 + %2862 = fptrunc float %2606 to bfloat, !dbg !39 + %2863 = fptrunc float %2607 to bfloat, !dbg !39 + %2864 = fptrunc float %2608 to bfloat, !dbg !39 + %2865 = fptrunc float %2609 to bfloat, !dbg !39 + %2866 = fptrunc float %2610 to bfloat, !dbg !39 + %2867 = fptrunc float %2611 to bfloat, !dbg !39 + %2868 = fptrunc float %2612 to bfloat, !dbg !39 + %2869 = fptrunc float %2613 to bfloat, !dbg !39 + %2870 = fptrunc float %2614 to bfloat, !dbg !39 + %2871 = fptrunc float %2615 to bfloat, !dbg !39 + %2872 = fptrunc float %2616 to bfloat, !dbg !39 + %2873 = fptrunc float %2617 to bfloat, !dbg !39 + %2874 = fptrunc float %2618 to bfloat, !dbg !39 + %2875 = fptrunc float %2619 to bfloat, !dbg !39 + %2876 = fptrunc float %2620 to bfloat, !dbg !39 + %2877 = fptrunc float %2621 to bfloat, !dbg !39 + %2878 = fptrunc float %2622 to bfloat, !dbg !39 + %2879 = fptrunc float %2623 to bfloat, !dbg !39 + %2880 = fptrunc float %2624 to bfloat, !dbg !39 + %2881 = fptrunc float %2625 to bfloat, !dbg !39 + %2882 = fptrunc float %2626 to bfloat, !dbg !39 + %2883 = fptrunc float %2627 to bfloat, !dbg !39 + %2884 = fptrunc float %2628 to bfloat, !dbg !39 + %2885 = fptrunc float %2629 to bfloat, !dbg !39 + %2886 = fptrunc float %2630 to bfloat, !dbg !39 + %2887 = fptrunc float %2631 to bfloat, !dbg !39 + %2888 = fptrunc float %2632 to bfloat, !dbg !39 + %2889 = fptrunc float %2633 to bfloat, !dbg !39 + %2890 = fptrunc float %2634 to bfloat, !dbg !39 + %2891 = fptrunc float %2635 to bfloat, !dbg !39 + %2892 = fptrunc float %2636 to bfloat, !dbg !39 + %2893 = fptrunc float %2637 to bfloat, !dbg !39 + %2894 = fptrunc float %2638 to bfloat, !dbg !39 + %2895 = fptrunc float %2639 to bfloat, !dbg !39 + %2896 = fptrunc float %2640 to bfloat, !dbg !39 + %2897 = fptrunc float %2641 to bfloat, !dbg !39 + %2898 = fptrunc float %2642 to bfloat, !dbg !39 + %2899 = fptrunc float %2643 to bfloat, !dbg !39 + %2900 = fptrunc float %2644 to bfloat, !dbg !39 + %2901 = fptrunc float %2645 to bfloat, !dbg !39 + %2902 = fptrunc float %2646 to bfloat, !dbg !39 + %2903 = fptrunc float %2647 to bfloat, !dbg !39 + %2904 = fptrunc float %2648 to bfloat, !dbg !39 + %2905 = fptrunc float %2649 to bfloat, !dbg !39 + %2906 = fptrunc float %2650 to bfloat, !dbg !39 + %2907 = fptrunc float %2651 to bfloat, !dbg !39 + %2908 = fptrunc float %2652 to bfloat, !dbg !39 + %2909 = fptrunc float %2653 to bfloat, !dbg !39 + %2910 = fptrunc float %2654 to bfloat, !dbg !39 + %2911 = fptrunc float %2655 to bfloat, !dbg !39 + %2912 = fptrunc float %2656 to bfloat, !dbg !39 + %2913 = fptrunc float %2657 to bfloat, !dbg !39 + %2914 = fptrunc float %2658 to bfloat, !dbg !39 + %2915 = fptrunc float %2659 to bfloat, !dbg !39 + %2916 = fptrunc float %2660 to bfloat, !dbg !39 + %2917 = fptrunc float %2661 to bfloat, !dbg !39 + %2918 = fptrunc float %2662 to bfloat, !dbg !39 + %2919 = fptrunc float %2663 to bfloat, !dbg !39 + %2920 = fptrunc float %2664 to bfloat, !dbg !39 + %2921 = fptrunc float %2665 to bfloat, !dbg !39 + %2922 = fptrunc float %2666 to bfloat, !dbg !39 + %2923 = fptrunc float %2667 to bfloat, !dbg !39 + %2924 = fptrunc float %2668 to bfloat, !dbg !39 + %2925 = fptrunc float %2669 to bfloat, !dbg !39 + %2926 = fptrunc float %2670 to bfloat, !dbg !39 + %2927 = fptrunc float %2671 to bfloat, !dbg !39 + %2928 = fptrunc float %2672 to bfloat, !dbg !39 + %2929 = fptrunc float %2673 to bfloat, !dbg !39 + %2930 = fptrunc float %2674 to bfloat, !dbg !39 + %2931 = fptrunc float %2675 to bfloat, !dbg !39 + %2932 = fptrunc float %2676 to bfloat, !dbg !39 + %2933 = fptrunc float %2677 to bfloat, !dbg !39 + %2934 = fptrunc float %2678 to bfloat, !dbg !39 + %2935 = fptrunc float %2679 to bfloat, !dbg !39 + %2936 = fptrunc float %2680 to bfloat, !dbg !39 + %2937 = fptrunc float %2681 to bfloat, !dbg !39 + %2938 = fptrunc float %2682 to bfloat, !dbg !39 + %2939 = fptrunc float %2683 to bfloat, !dbg !39 + %2940 = fptrunc float %2684 to bfloat, !dbg !39 + %2941 = fptrunc float %2685 to bfloat, !dbg !39 + %2942 = fptrunc float %2686 to bfloat, !dbg !39 + %2943 = fptrunc float %2687 to bfloat, !dbg !39 + %2944 = fptrunc float %2688 to bfloat, !dbg !39 + %2945 = fptrunc float %2689 to bfloat, !dbg !39 + %2946 = fptrunc float %2690 to bfloat, !dbg !39 + %2947 = fptrunc float %2691 to bfloat, !dbg !39 + %2948 = fptrunc float %2692 to bfloat, !dbg !39 + %2949 = fptrunc float %2693 to bfloat, !dbg !39 + %2950 = fptrunc float %2694 to bfloat, !dbg !39 + %2951 = fptrunc float %2695 to bfloat, !dbg !39 + %2952 = fptrunc float %2696 to bfloat, !dbg !39 + %2953 = fptrunc float %2697 to bfloat, !dbg !39 + %2954 = fptrunc float %2698 to bfloat, !dbg !39 + %2955 = fptrunc float %2699 to bfloat, !dbg !39 + %2956 = fptrunc float %2700 to bfloat, !dbg !39 + %2957 = fptrunc float %2701 to bfloat, !dbg !39 + %2958 = fptrunc float %2702 to bfloat, !dbg !39 + %2959 = fptrunc float %2703 to bfloat, !dbg !39 + %2960 = fptrunc float %2704 to bfloat, !dbg !39 + %2961 = fptrunc float %2705 to bfloat, !dbg !39 + %2962 = fptrunc float %2706 to bfloat, !dbg !39 + %2963 = fptrunc float %2707 to bfloat, !dbg !39 + %2964 = fptrunc float %2708 to bfloat, !dbg !39 + %2965 = fptrunc float %2709 to bfloat, !dbg !39 + %2966 = fptrunc float %2710 to bfloat, !dbg !39 + %2967 = fptrunc float %2711 to bfloat, !dbg !39 + %2968 = fptrunc float %2712 to bfloat, !dbg !39 + %2969 = fptrunc float %2713 to bfloat, !dbg !39 + %2970 = fptrunc float %2714 to bfloat, !dbg !39 + %2971 = fptrunc float %2715 to bfloat, !dbg !39 + %2972 = fptrunc float %2716 to bfloat, !dbg !39 + %2973 = fptrunc float %2717 to bfloat, !dbg !39 + %2974 = fptrunc float %2718 to bfloat, !dbg !39 + %2975 = fptrunc float %2719 to bfloat, !dbg !39 + %2976 = fptrunc float %2720 to bfloat, !dbg !39 + %2977 = fptrunc float %2721 to bfloat, !dbg !39 + %2978 = fptrunc float %2722 to bfloat, !dbg !39 + %2979 = fptrunc float %2723 to bfloat, !dbg !39 + %2980 = fptrunc float %2724 to bfloat, !dbg !39 + %2981 = fptrunc float %2725 to bfloat, !dbg !39 + %2982 = fptrunc float %2726 to bfloat, !dbg !39 + %2983 = fptrunc float %2727 to bfloat, !dbg !39 + %2984 = fptrunc float %2728 to bfloat, !dbg !39 + %2985 = fptrunc float %2729 to bfloat, !dbg !39 + %2986 = fptrunc float %2730 to bfloat, !dbg !39 + %2987 = fptrunc float %2731 to bfloat, !dbg !39 + %2988 = fptrunc float %2732 to bfloat, !dbg !39 + %2989 = fptrunc float %2733 to bfloat, !dbg !39 + %2990 = fptrunc float %2734 to bfloat, !dbg !39 + %2991 = fptrunc float %2735 to bfloat, !dbg !39 + %2992 = fptrunc float %2736 to bfloat, !dbg !39 + %2993 = fptrunc float %2737 to bfloat, !dbg !39 + %2994 = fptrunc float %2738 to bfloat, !dbg !39 + %2995 = fptrunc float %2739 to bfloat, !dbg !39 + %2996 = fptrunc float %2740 to bfloat, !dbg !39 + %2997 = fptrunc float %2741 to bfloat, !dbg !39 + %2998 = fptrunc float %2742 to bfloat, !dbg !39 + %2999 = fptrunc float %2743 to bfloat, !dbg !39 + %3000 = fptrunc float %2744 to bfloat, !dbg !39 + %3001 = fptrunc float %2745 to bfloat, !dbg !39 + %3002 = fptrunc float %2746 to bfloat, !dbg !39 + %3003 = fptrunc float %2747 to bfloat, !dbg !39 + %3004 = fptrunc float %2748 to bfloat, !dbg !39 + %3005 = fptrunc float %2749 to bfloat, !dbg !39 + %3006 = fptrunc float %2750 to bfloat, !dbg !39 + %3007 = fptrunc float %2751 to bfloat, !dbg !39 + %3008 = fptrunc float %2752 to bfloat, !dbg !39 + %3009 = fptrunc float %2753 to bfloat, !dbg !39 + %3010 = fptrunc float %2754 to bfloat, !dbg !39 + %3011 = fptrunc float %2755 to bfloat, !dbg !39 + %3012 = fptrunc float %2756 to bfloat, !dbg !39 + %3013 = fptrunc float %2757 to bfloat, !dbg !39 + %3014 = fptrunc float %2758 to bfloat, !dbg !39 + %3015 = fptrunc float %2759 to bfloat, !dbg !39 + %3016 = fptrunc float %2760 to bfloat, !dbg !39 + %3017 = fptrunc float %2761 to bfloat, !dbg !39 + %3018 = fptrunc float %2762 to bfloat, !dbg !39 + %3019 = fptrunc float %2763 to bfloat, !dbg !39 + %3020 = fptrunc float %2764 to bfloat, !dbg !39 + %3021 = fptrunc float %2765 to bfloat, !dbg !39 + %3022 = fptrunc float %2766 to bfloat, !dbg !39 + %3023 = fptrunc float %2767 to bfloat, !dbg !39 + %3024 = fptrunc float %2768 to bfloat, !dbg !39 + %3025 = fptrunc float %2769 to bfloat, !dbg !39 + %3026 = fptrunc float %2770 to bfloat, !dbg !39 + %3027 = fptrunc float %2771 to bfloat, !dbg !39 + %3028 = fptrunc float %2772 to bfloat, !dbg !39 + %3029 = fptrunc float %2773 to bfloat, !dbg !39 + %3030 = fptrunc float %2774 to bfloat, !dbg !39 + %3031 = fptrunc float %2775 to bfloat, !dbg !39 + %3032 = insertelement <2 x bfloat> poison, bfloat %2776, i64 0, !dbg !40 + %3033 = insertelement <2 x bfloat> %3032, bfloat %2777, i64 1, !dbg !40 + %3034 = bitcast <2 x bfloat> %3033 to i32, !dbg !40 + %3035 = insertelement <2 x bfloat> poison, bfloat %2778, i64 0, !dbg !40 + %3036 = insertelement <2 x bfloat> %3035, bfloat %2779, i64 1, !dbg !40 + %3037 = bitcast <2 x bfloat> %3036 to i32, !dbg !40 + %3038 = insertelement <2 x bfloat> poison, bfloat %2780, i64 0, !dbg !40 + %3039 = insertelement <2 x bfloat> %3038, bfloat %2781, i64 1, !dbg !40 + %3040 = bitcast <2 x bfloat> %3039 to i32, !dbg !40 + %3041 = insertelement <2 x bfloat> poison, bfloat %2782, i64 0, !dbg !40 + %3042 = insertelement <2 x bfloat> %3041, bfloat %2783, i64 1, !dbg !40 + %3043 = bitcast <2 x bfloat> %3042 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3034, i32 %3037, i32 %3040, i32 %3043, ptr addrspace(1) %2485, i1 %172) #6, !dbg !40 + %3044 = insertelement <2 x bfloat> poison, bfloat %2784, i64 0, !dbg !40 + %3045 = insertelement <2 x bfloat> %3044, bfloat %2785, i64 1, !dbg !40 + %3046 = bitcast <2 x bfloat> %3045 to i32, !dbg !40 + %3047 = insertelement <2 x bfloat> poison, bfloat %2786, i64 0, !dbg !40 + %3048 = insertelement <2 x bfloat> %3047, bfloat %2787, i64 1, !dbg !40 + %3049 = bitcast <2 x bfloat> %3048 to i32, !dbg !40 + %3050 = insertelement <2 x bfloat> poison, bfloat %2788, i64 0, !dbg !40 + %3051 = insertelement <2 x bfloat> %3050, bfloat %2789, i64 1, !dbg !40 + %3052 = bitcast <2 x bfloat> %3051 to i32, !dbg !40 + %3053 = insertelement <2 x bfloat> poison, bfloat %2790, i64 0, !dbg !40 + %3054 = insertelement <2 x bfloat> %3053, bfloat %2791, i64 1, !dbg !40 + %3055 = bitcast <2 x bfloat> %3054 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3046, i32 %3049, i32 %3052, i32 %3055, ptr addrspace(1) %2483, i1 %190) #6, !dbg !40 + %3056 = insertelement <2 x bfloat> poison, bfloat %2792, i64 0, !dbg !40 + %3057 = insertelement <2 x bfloat> %3056, bfloat %2793, i64 1, !dbg !40 + %3058 = bitcast <2 x bfloat> %3057 to i32, !dbg !40 + %3059 = insertelement <2 x bfloat> poison, bfloat %2794, i64 0, !dbg !40 + %3060 = insertelement <2 x bfloat> %3059, bfloat %2795, i64 1, !dbg !40 + %3061 = bitcast <2 x bfloat> %3060 to i32, !dbg !40 + %3062 = insertelement <2 x bfloat> poison, bfloat %2796, i64 0, !dbg !40 + %3063 = insertelement <2 x bfloat> %3062, bfloat %2797, i64 1, !dbg !40 + %3064 = bitcast <2 x bfloat> %3063 to i32, !dbg !40 + %3065 = insertelement <2 x bfloat> poison, bfloat %2798, i64 0, !dbg !40 + %3066 = insertelement <2 x bfloat> %3065, bfloat %2799, i64 1, !dbg !40 + %3067 = bitcast <2 x bfloat> %3066 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3058, i32 %3061, i32 %3064, i32 %3067, ptr addrspace(1) %2481, i1 %208) #6, !dbg !40 + %3068 = insertelement <2 x bfloat> poison, bfloat %2800, i64 0, !dbg !40 + %3069 = insertelement <2 x bfloat> %3068, bfloat %2801, i64 1, !dbg !40 + %3070 = bitcast <2 x bfloat> %3069 to i32, !dbg !40 + %3071 = insertelement <2 x bfloat> poison, bfloat %2802, i64 0, !dbg !40 + %3072 = insertelement <2 x bfloat> %3071, bfloat %2803, i64 1, !dbg !40 + %3073 = bitcast <2 x bfloat> %3072 to i32, !dbg !40 + %3074 = insertelement <2 x bfloat> poison, bfloat %2804, i64 0, !dbg !40 + %3075 = insertelement <2 x bfloat> %3074, bfloat %2805, i64 1, !dbg !40 + %3076 = bitcast <2 x bfloat> %3075 to i32, !dbg !40 + %3077 = insertelement <2 x bfloat> poison, bfloat %2806, i64 0, !dbg !40 + %3078 = insertelement <2 x bfloat> %3077, bfloat %2807, i64 1, !dbg !40 + %3079 = bitcast <2 x bfloat> %3078 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3070, i32 %3073, i32 %3076, i32 %3079, ptr addrspace(1) %2479, i1 %226) #6, !dbg !40 + %3080 = insertelement <2 x bfloat> poison, bfloat %2808, i64 0, !dbg !40 + %3081 = insertelement <2 x bfloat> %3080, bfloat %2809, i64 1, !dbg !40 + %3082 = bitcast <2 x bfloat> %3081 to i32, !dbg !40 + %3083 = insertelement <2 x bfloat> poison, bfloat %2810, i64 0, !dbg !40 + %3084 = insertelement <2 x bfloat> %3083, bfloat %2811, i64 1, !dbg !40 + %3085 = bitcast <2 x bfloat> %3084 to i32, !dbg !40 + %3086 = insertelement <2 x bfloat> poison, bfloat %2812, i64 0, !dbg !40 + %3087 = insertelement <2 x bfloat> %3086, bfloat %2813, i64 1, !dbg !40 + %3088 = bitcast <2 x bfloat> %3087 to i32, !dbg !40 + %3089 = insertelement <2 x bfloat> poison, bfloat %2814, i64 0, !dbg !40 + %3090 = insertelement <2 x bfloat> %3089, bfloat %2815, i64 1, !dbg !40 + %3091 = bitcast <2 x bfloat> %3090 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3082, i32 %3085, i32 %3088, i32 %3091, ptr addrspace(1) %2477, i1 %244) #6, !dbg !40 + %3092 = insertelement <2 x bfloat> poison, bfloat %2816, i64 0, !dbg !40 + %3093 = insertelement <2 x bfloat> %3092, bfloat %2817, i64 1, !dbg !40 + %3094 = bitcast <2 x bfloat> %3093 to i32, !dbg !40 + %3095 = insertelement <2 x bfloat> poison, bfloat %2818, i64 0, !dbg !40 + %3096 = insertelement <2 x bfloat> %3095, bfloat %2819, i64 1, !dbg !40 + %3097 = bitcast <2 x bfloat> %3096 to i32, !dbg !40 + %3098 = insertelement <2 x bfloat> poison, bfloat %2820, i64 0, !dbg !40 + %3099 = insertelement <2 x bfloat> %3098, bfloat %2821, i64 1, !dbg !40 + %3100 = bitcast <2 x bfloat> %3099 to i32, !dbg !40 + %3101 = insertelement <2 x bfloat> poison, bfloat %2822, i64 0, !dbg !40 + %3102 = insertelement <2 x bfloat> %3101, bfloat %2823, i64 1, !dbg !40 + %3103 = bitcast <2 x bfloat> %3102 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3094, i32 %3097, i32 %3100, i32 %3103, ptr addrspace(1) %2475, i1 %262) #6, !dbg !40 + %3104 = insertelement <2 x bfloat> poison, bfloat %2824, i64 0, !dbg !40 + %3105 = insertelement <2 x bfloat> %3104, bfloat %2825, i64 1, !dbg !40 + %3106 = bitcast <2 x bfloat> %3105 to i32, !dbg !40 + %3107 = insertelement <2 x bfloat> poison, bfloat %2826, i64 0, !dbg !40 + %3108 = insertelement <2 x bfloat> %3107, bfloat %2827, i64 1, !dbg !40 + %3109 = bitcast <2 x bfloat> %3108 to i32, !dbg !40 + %3110 = insertelement <2 x bfloat> poison, bfloat %2828, i64 0, !dbg !40 + %3111 = insertelement <2 x bfloat> %3110, bfloat %2829, i64 1, !dbg !40 + %3112 = bitcast <2 x bfloat> %3111 to i32, !dbg !40 + %3113 = insertelement <2 x bfloat> poison, bfloat %2830, i64 0, !dbg !40 + %3114 = insertelement <2 x bfloat> %3113, bfloat %2831, i64 1, !dbg !40 + %3115 = bitcast <2 x bfloat> %3114 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3106, i32 %3109, i32 %3112, i32 %3115, ptr addrspace(1) %2473, i1 %280) #6, !dbg !40 + %3116 = insertelement <2 x bfloat> poison, bfloat %2832, i64 0, !dbg !40 + %3117 = insertelement <2 x bfloat> %3116, bfloat %2833, i64 1, !dbg !40 + %3118 = bitcast <2 x bfloat> %3117 to i32, !dbg !40 + %3119 = insertelement <2 x bfloat> poison, bfloat %2834, i64 0, !dbg !40 + %3120 = insertelement <2 x bfloat> %3119, bfloat %2835, i64 1, !dbg !40 + %3121 = bitcast <2 x bfloat> %3120 to i32, !dbg !40 + %3122 = insertelement <2 x bfloat> poison, bfloat %2836, i64 0, !dbg !40 + %3123 = insertelement <2 x bfloat> %3122, bfloat %2837, i64 1, !dbg !40 + %3124 = bitcast <2 x bfloat> %3123 to i32, !dbg !40 + %3125 = insertelement <2 x bfloat> poison, bfloat %2838, i64 0, !dbg !40 + %3126 = insertelement <2 x bfloat> %3125, bfloat %2839, i64 1, !dbg !40 + %3127 = bitcast <2 x bfloat> %3126 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3118, i32 %3121, i32 %3124, i32 %3127, ptr addrspace(1) %2471, i1 %298) #6, !dbg !40 + %3128 = insertelement <2 x bfloat> poison, bfloat %2840, i64 0, !dbg !40 + %3129 = insertelement <2 x bfloat> %3128, bfloat %2841, i64 1, !dbg !40 + %3130 = bitcast <2 x bfloat> %3129 to i32, !dbg !40 + %3131 = insertelement <2 x bfloat> poison, bfloat %2842, i64 0, !dbg !40 + %3132 = insertelement <2 x bfloat> %3131, bfloat %2843, i64 1, !dbg !40 + %3133 = bitcast <2 x bfloat> %3132 to i32, !dbg !40 + %3134 = insertelement <2 x bfloat> poison, bfloat %2844, i64 0, !dbg !40 + %3135 = insertelement <2 x bfloat> %3134, bfloat %2845, i64 1, !dbg !40 + %3136 = bitcast <2 x bfloat> %3135 to i32, !dbg !40 + %3137 = insertelement <2 x bfloat> poison, bfloat %2846, i64 0, !dbg !40 + %3138 = insertelement <2 x bfloat> %3137, bfloat %2847, i64 1, !dbg !40 + %3139 = bitcast <2 x bfloat> %3138 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3130, i32 %3133, i32 %3136, i32 %3139, ptr addrspace(1) %2469, i1 %316) #6, !dbg !40 + %3140 = insertelement <2 x bfloat> poison, bfloat %2848, i64 0, !dbg !40 + %3141 = insertelement <2 x bfloat> %3140, bfloat %2849, i64 1, !dbg !40 + %3142 = bitcast <2 x bfloat> %3141 to i32, !dbg !40 + %3143 = insertelement <2 x bfloat> poison, bfloat %2850, i64 0, !dbg !40 + %3144 = insertelement <2 x bfloat> %3143, bfloat %2851, i64 1, !dbg !40 + %3145 = bitcast <2 x bfloat> %3144 to i32, !dbg !40 + %3146 = insertelement <2 x bfloat> poison, bfloat %2852, i64 0, !dbg !40 + %3147 = insertelement <2 x bfloat> %3146, bfloat %2853, i64 1, !dbg !40 + %3148 = bitcast <2 x bfloat> %3147 to i32, !dbg !40 + %3149 = insertelement <2 x bfloat> poison, bfloat %2854, i64 0, !dbg !40 + %3150 = insertelement <2 x bfloat> %3149, bfloat %2855, i64 1, !dbg !40 + %3151 = bitcast <2 x bfloat> %3150 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3142, i32 %3145, i32 %3148, i32 %3151, ptr addrspace(1) %2467, i1 %334) #6, !dbg !40 + %3152 = insertelement <2 x bfloat> poison, bfloat %2856, i64 0, !dbg !40 + %3153 = insertelement <2 x bfloat> %3152, bfloat %2857, i64 1, !dbg !40 + %3154 = bitcast <2 x bfloat> %3153 to i32, !dbg !40 + %3155 = insertelement <2 x bfloat> poison, bfloat %2858, i64 0, !dbg !40 + %3156 = insertelement <2 x bfloat> %3155, bfloat %2859, i64 1, !dbg !40 + %3157 = bitcast <2 x bfloat> %3156 to i32, !dbg !40 + %3158 = insertelement <2 x bfloat> poison, bfloat %2860, i64 0, !dbg !40 + %3159 = insertelement <2 x bfloat> %3158, bfloat %2861, i64 1, !dbg !40 + %3160 = bitcast <2 x bfloat> %3159 to i32, !dbg !40 + %3161 = insertelement <2 x bfloat> poison, bfloat %2862, i64 0, !dbg !40 + %3162 = insertelement <2 x bfloat> %3161, bfloat %2863, i64 1, !dbg !40 + %3163 = bitcast <2 x bfloat> %3162 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3154, i32 %3157, i32 %3160, i32 %3163, ptr addrspace(1) %2465, i1 %352) #6, !dbg !40 + %3164 = insertelement <2 x bfloat> poison, bfloat %2864, i64 0, !dbg !40 + %3165 = insertelement <2 x bfloat> %3164, bfloat %2865, i64 1, !dbg !40 + %3166 = bitcast <2 x bfloat> %3165 to i32, !dbg !40 + %3167 = insertelement <2 x bfloat> poison, bfloat %2866, i64 0, !dbg !40 + %3168 = insertelement <2 x bfloat> %3167, bfloat %2867, i64 1, !dbg !40 + %3169 = bitcast <2 x bfloat> %3168 to i32, !dbg !40 + %3170 = insertelement <2 x bfloat> poison, bfloat %2868, i64 0, !dbg !40 + %3171 = insertelement <2 x bfloat> %3170, bfloat %2869, i64 1, !dbg !40 + %3172 = bitcast <2 x bfloat> %3171 to i32, !dbg !40 + %3173 = insertelement <2 x bfloat> poison, bfloat %2870, i64 0, !dbg !40 + %3174 = insertelement <2 x bfloat> %3173, bfloat %2871, i64 1, !dbg !40 + %3175 = bitcast <2 x bfloat> %3174 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3166, i32 %3169, i32 %3172, i32 %3175, ptr addrspace(1) %2463, i1 %370) #6, !dbg !40 + %3176 = insertelement <2 x bfloat> poison, bfloat %2872, i64 0, !dbg !40 + %3177 = insertelement <2 x bfloat> %3176, bfloat %2873, i64 1, !dbg !40 + %3178 = bitcast <2 x bfloat> %3177 to i32, !dbg !40 + %3179 = insertelement <2 x bfloat> poison, bfloat %2874, i64 0, !dbg !40 + %3180 = insertelement <2 x bfloat> %3179, bfloat %2875, i64 1, !dbg !40 + %3181 = bitcast <2 x bfloat> %3180 to i32, !dbg !40 + %3182 = insertelement <2 x bfloat> poison, bfloat %2876, i64 0, !dbg !40 + %3183 = insertelement <2 x bfloat> %3182, bfloat %2877, i64 1, !dbg !40 + %3184 = bitcast <2 x bfloat> %3183 to i32, !dbg !40 + %3185 = insertelement <2 x bfloat> poison, bfloat %2878, i64 0, !dbg !40 + %3186 = insertelement <2 x bfloat> %3185, bfloat %2879, i64 1, !dbg !40 + %3187 = bitcast <2 x bfloat> %3186 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3178, i32 %3181, i32 %3184, i32 %3187, ptr addrspace(1) %2461, i1 %388) #6, !dbg !40 + %3188 = insertelement <2 x bfloat> poison, bfloat %2880, i64 0, !dbg !40 + %3189 = insertelement <2 x bfloat> %3188, bfloat %2881, i64 1, !dbg !40 + %3190 = bitcast <2 x bfloat> %3189 to i32, !dbg !40 + %3191 = insertelement <2 x bfloat> poison, bfloat %2882, i64 0, !dbg !40 + %3192 = insertelement <2 x bfloat> %3191, bfloat %2883, i64 1, !dbg !40 + %3193 = bitcast <2 x bfloat> %3192 to i32, !dbg !40 + %3194 = insertelement <2 x bfloat> poison, bfloat %2884, i64 0, !dbg !40 + %3195 = insertelement <2 x bfloat> %3194, bfloat %2885, i64 1, !dbg !40 + %3196 = bitcast <2 x bfloat> %3195 to i32, !dbg !40 + %3197 = insertelement <2 x bfloat> poison, bfloat %2886, i64 0, !dbg !40 + %3198 = insertelement <2 x bfloat> %3197, bfloat %2887, i64 1, !dbg !40 + %3199 = bitcast <2 x bfloat> %3198 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3190, i32 %3193, i32 %3196, i32 %3199, ptr addrspace(1) %2459, i1 %406) #6, !dbg !40 + %3200 = insertelement <2 x bfloat> poison, bfloat %2888, i64 0, !dbg !40 + %3201 = insertelement <2 x bfloat> %3200, bfloat %2889, i64 1, !dbg !40 + %3202 = bitcast <2 x bfloat> %3201 to i32, !dbg !40 + %3203 = insertelement <2 x bfloat> poison, bfloat %2890, i64 0, !dbg !40 + %3204 = insertelement <2 x bfloat> %3203, bfloat %2891, i64 1, !dbg !40 + %3205 = bitcast <2 x bfloat> %3204 to i32, !dbg !40 + %3206 = insertelement <2 x bfloat> poison, bfloat %2892, i64 0, !dbg !40 + %3207 = insertelement <2 x bfloat> %3206, bfloat %2893, i64 1, !dbg !40 + %3208 = bitcast <2 x bfloat> %3207 to i32, !dbg !40 + %3209 = insertelement <2 x bfloat> poison, bfloat %2894, i64 0, !dbg !40 + %3210 = insertelement <2 x bfloat> %3209, bfloat %2895, i64 1, !dbg !40 + %3211 = bitcast <2 x bfloat> %3210 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3202, i32 %3205, i32 %3208, i32 %3211, ptr addrspace(1) %2457, i1 %424) #6, !dbg !40 + %3212 = insertelement <2 x bfloat> poison, bfloat %2896, i64 0, !dbg !40 + %3213 = insertelement <2 x bfloat> %3212, bfloat %2897, i64 1, !dbg !40 + %3214 = bitcast <2 x bfloat> %3213 to i32, !dbg !40 + %3215 = insertelement <2 x bfloat> poison, bfloat %2898, i64 0, !dbg !40 + %3216 = insertelement <2 x bfloat> %3215, bfloat %2899, i64 1, !dbg !40 + %3217 = bitcast <2 x bfloat> %3216 to i32, !dbg !40 + %3218 = insertelement <2 x bfloat> poison, bfloat %2900, i64 0, !dbg !40 + %3219 = insertelement <2 x bfloat> %3218, bfloat %2901, i64 1, !dbg !40 + %3220 = bitcast <2 x bfloat> %3219 to i32, !dbg !40 + %3221 = insertelement <2 x bfloat> poison, bfloat %2902, i64 0, !dbg !40 + %3222 = insertelement <2 x bfloat> %3221, bfloat %2903, i64 1, !dbg !40 + %3223 = bitcast <2 x bfloat> %3222 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3214, i32 %3217, i32 %3220, i32 %3223, ptr addrspace(1) %2455, i1 %442) #6, !dbg !40 + %3224 = insertelement <2 x bfloat> poison, bfloat %2904, i64 0, !dbg !40 + %3225 = insertelement <2 x bfloat> %3224, bfloat %2905, i64 1, !dbg !40 + %3226 = bitcast <2 x bfloat> %3225 to i32, !dbg !40 + %3227 = insertelement <2 x bfloat> poison, bfloat %2906, i64 0, !dbg !40 + %3228 = insertelement <2 x bfloat> %3227, bfloat %2907, i64 1, !dbg !40 + %3229 = bitcast <2 x bfloat> %3228 to i32, !dbg !40 + %3230 = insertelement <2 x bfloat> poison, bfloat %2908, i64 0, !dbg !40 + %3231 = insertelement <2 x bfloat> %3230, bfloat %2909, i64 1, !dbg !40 + %3232 = bitcast <2 x bfloat> %3231 to i32, !dbg !40 + %3233 = insertelement <2 x bfloat> poison, bfloat %2910, i64 0, !dbg !40 + %3234 = insertelement <2 x bfloat> %3233, bfloat %2911, i64 1, !dbg !40 + %3235 = bitcast <2 x bfloat> %3234 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3226, i32 %3229, i32 %3232, i32 %3235, ptr addrspace(1) %2453, i1 %460) #6, !dbg !40 + %3236 = insertelement <2 x bfloat> poison, bfloat %2912, i64 0, !dbg !40 + %3237 = insertelement <2 x bfloat> %3236, bfloat %2913, i64 1, !dbg !40 + %3238 = bitcast <2 x bfloat> %3237 to i32, !dbg !40 + %3239 = insertelement <2 x bfloat> poison, bfloat %2914, i64 0, !dbg !40 + %3240 = insertelement <2 x bfloat> %3239, bfloat %2915, i64 1, !dbg !40 + %3241 = bitcast <2 x bfloat> %3240 to i32, !dbg !40 + %3242 = insertelement <2 x bfloat> poison, bfloat %2916, i64 0, !dbg !40 + %3243 = insertelement <2 x bfloat> %3242, bfloat %2917, i64 1, !dbg !40 + %3244 = bitcast <2 x bfloat> %3243 to i32, !dbg !40 + %3245 = insertelement <2 x bfloat> poison, bfloat %2918, i64 0, !dbg !40 + %3246 = insertelement <2 x bfloat> %3245, bfloat %2919, i64 1, !dbg !40 + %3247 = bitcast <2 x bfloat> %3246 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3238, i32 %3241, i32 %3244, i32 %3247, ptr addrspace(1) %2451, i1 %478) #6, !dbg !40 + %3248 = insertelement <2 x bfloat> poison, bfloat %2920, i64 0, !dbg !40 + %3249 = insertelement <2 x bfloat> %3248, bfloat %2921, i64 1, !dbg !40 + %3250 = bitcast <2 x bfloat> %3249 to i32, !dbg !40 + %3251 = insertelement <2 x bfloat> poison, bfloat %2922, i64 0, !dbg !40 + %3252 = insertelement <2 x bfloat> %3251, bfloat %2923, i64 1, !dbg !40 + %3253 = bitcast <2 x bfloat> %3252 to i32, !dbg !40 + %3254 = insertelement <2 x bfloat> poison, bfloat %2924, i64 0, !dbg !40 + %3255 = insertelement <2 x bfloat> %3254, bfloat %2925, i64 1, !dbg !40 + %3256 = bitcast <2 x bfloat> %3255 to i32, !dbg !40 + %3257 = insertelement <2 x bfloat> poison, bfloat %2926, i64 0, !dbg !40 + %3258 = insertelement <2 x bfloat> %3257, bfloat %2927, i64 1, !dbg !40 + %3259 = bitcast <2 x bfloat> %3258 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3250, i32 %3253, i32 %3256, i32 %3259, ptr addrspace(1) %2449, i1 %496) #6, !dbg !40 + %3260 = insertelement <2 x bfloat> poison, bfloat %2928, i64 0, !dbg !40 + %3261 = insertelement <2 x bfloat> %3260, bfloat %2929, i64 1, !dbg !40 + %3262 = bitcast <2 x bfloat> %3261 to i32, !dbg !40 + %3263 = insertelement <2 x bfloat> poison, bfloat %2930, i64 0, !dbg !40 + %3264 = insertelement <2 x bfloat> %3263, bfloat %2931, i64 1, !dbg !40 + %3265 = bitcast <2 x bfloat> %3264 to i32, !dbg !40 + %3266 = insertelement <2 x bfloat> poison, bfloat %2932, i64 0, !dbg !40 + %3267 = insertelement <2 x bfloat> %3266, bfloat %2933, i64 1, !dbg !40 + %3268 = bitcast <2 x bfloat> %3267 to i32, !dbg !40 + %3269 = insertelement <2 x bfloat> poison, bfloat %2934, i64 0, !dbg !40 + %3270 = insertelement <2 x bfloat> %3269, bfloat %2935, i64 1, !dbg !40 + %3271 = bitcast <2 x bfloat> %3270 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3262, i32 %3265, i32 %3268, i32 %3271, ptr addrspace(1) %2447, i1 %514) #6, !dbg !40 + %3272 = insertelement <2 x bfloat> poison, bfloat %2936, i64 0, !dbg !40 + %3273 = insertelement <2 x bfloat> %3272, bfloat %2937, i64 1, !dbg !40 + %3274 = bitcast <2 x bfloat> %3273 to i32, !dbg !40 + %3275 = insertelement <2 x bfloat> poison, bfloat %2938, i64 0, !dbg !40 + %3276 = insertelement <2 x bfloat> %3275, bfloat %2939, i64 1, !dbg !40 + %3277 = bitcast <2 x bfloat> %3276 to i32, !dbg !40 + %3278 = insertelement <2 x bfloat> poison, bfloat %2940, i64 0, !dbg !40 + %3279 = insertelement <2 x bfloat> %3278, bfloat %2941, i64 1, !dbg !40 + %3280 = bitcast <2 x bfloat> %3279 to i32, !dbg !40 + %3281 = insertelement <2 x bfloat> poison, bfloat %2942, i64 0, !dbg !40 + %3282 = insertelement <2 x bfloat> %3281, bfloat %2943, i64 1, !dbg !40 + %3283 = bitcast <2 x bfloat> %3282 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3274, i32 %3277, i32 %3280, i32 %3283, ptr addrspace(1) %2445, i1 %532) #6, !dbg !40 + %3284 = insertelement <2 x bfloat> poison, bfloat %2944, i64 0, !dbg !40 + %3285 = insertelement <2 x bfloat> %3284, bfloat %2945, i64 1, !dbg !40 + %3286 = bitcast <2 x bfloat> %3285 to i32, !dbg !40 + %3287 = insertelement <2 x bfloat> poison, bfloat %2946, i64 0, !dbg !40 + %3288 = insertelement <2 x bfloat> %3287, bfloat %2947, i64 1, !dbg !40 + %3289 = bitcast <2 x bfloat> %3288 to i32, !dbg !40 + %3290 = insertelement <2 x bfloat> poison, bfloat %2948, i64 0, !dbg !40 + %3291 = insertelement <2 x bfloat> %3290, bfloat %2949, i64 1, !dbg !40 + %3292 = bitcast <2 x bfloat> %3291 to i32, !dbg !40 + %3293 = insertelement <2 x bfloat> poison, bfloat %2950, i64 0, !dbg !40 + %3294 = insertelement <2 x bfloat> %3293, bfloat %2951, i64 1, !dbg !40 + %3295 = bitcast <2 x bfloat> %3294 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3286, i32 %3289, i32 %3292, i32 %3295, ptr addrspace(1) %2443, i1 %550) #6, !dbg !40 + %3296 = insertelement <2 x bfloat> poison, bfloat %2952, i64 0, !dbg !40 + %3297 = insertelement <2 x bfloat> %3296, bfloat %2953, i64 1, !dbg !40 + %3298 = bitcast <2 x bfloat> %3297 to i32, !dbg !40 + %3299 = insertelement <2 x bfloat> poison, bfloat %2954, i64 0, !dbg !40 + %3300 = insertelement <2 x bfloat> %3299, bfloat %2955, i64 1, !dbg !40 + %3301 = bitcast <2 x bfloat> %3300 to i32, !dbg !40 + %3302 = insertelement <2 x bfloat> poison, bfloat %2956, i64 0, !dbg !40 + %3303 = insertelement <2 x bfloat> %3302, bfloat %2957, i64 1, !dbg !40 + %3304 = bitcast <2 x bfloat> %3303 to i32, !dbg !40 + %3305 = insertelement <2 x bfloat> poison, bfloat %2958, i64 0, !dbg !40 + %3306 = insertelement <2 x bfloat> %3305, bfloat %2959, i64 1, !dbg !40 + %3307 = bitcast <2 x bfloat> %3306 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3298, i32 %3301, i32 %3304, i32 %3307, ptr addrspace(1) %2441, i1 %568) #6, !dbg !40 + %3308 = insertelement <2 x bfloat> poison, bfloat %2960, i64 0, !dbg !40 + %3309 = insertelement <2 x bfloat> %3308, bfloat %2961, i64 1, !dbg !40 + %3310 = bitcast <2 x bfloat> %3309 to i32, !dbg !40 + %3311 = insertelement <2 x bfloat> poison, bfloat %2962, i64 0, !dbg !40 + %3312 = insertelement <2 x bfloat> %3311, bfloat %2963, i64 1, !dbg !40 + %3313 = bitcast <2 x bfloat> %3312 to i32, !dbg !40 + %3314 = insertelement <2 x bfloat> poison, bfloat %2964, i64 0, !dbg !40 + %3315 = insertelement <2 x bfloat> %3314, bfloat %2965, i64 1, !dbg !40 + %3316 = bitcast <2 x bfloat> %3315 to i32, !dbg !40 + %3317 = insertelement <2 x bfloat> poison, bfloat %2966, i64 0, !dbg !40 + %3318 = insertelement <2 x bfloat> %3317, bfloat %2967, i64 1, !dbg !40 + %3319 = bitcast <2 x bfloat> %3318 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3310, i32 %3313, i32 %3316, i32 %3319, ptr addrspace(1) %2439, i1 %586) #6, !dbg !40 + %3320 = insertelement <2 x bfloat> poison, bfloat %2968, i64 0, !dbg !40 + %3321 = insertelement <2 x bfloat> %3320, bfloat %2969, i64 1, !dbg !40 + %3322 = bitcast <2 x bfloat> %3321 to i32, !dbg !40 + %3323 = insertelement <2 x bfloat> poison, bfloat %2970, i64 0, !dbg !40 + %3324 = insertelement <2 x bfloat> %3323, bfloat %2971, i64 1, !dbg !40 + %3325 = bitcast <2 x bfloat> %3324 to i32, !dbg !40 + %3326 = insertelement <2 x bfloat> poison, bfloat %2972, i64 0, !dbg !40 + %3327 = insertelement <2 x bfloat> %3326, bfloat %2973, i64 1, !dbg !40 + %3328 = bitcast <2 x bfloat> %3327 to i32, !dbg !40 + %3329 = insertelement <2 x bfloat> poison, bfloat %2974, i64 0, !dbg !40 + %3330 = insertelement <2 x bfloat> %3329, bfloat %2975, i64 1, !dbg !40 + %3331 = bitcast <2 x bfloat> %3330 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3322, i32 %3325, i32 %3328, i32 %3331, ptr addrspace(1) %2437, i1 %604) #6, !dbg !40 + %3332 = insertelement <2 x bfloat> poison, bfloat %2976, i64 0, !dbg !40 + %3333 = insertelement <2 x bfloat> %3332, bfloat %2977, i64 1, !dbg !40 + %3334 = bitcast <2 x bfloat> %3333 to i32, !dbg !40 + %3335 = insertelement <2 x bfloat> poison, bfloat %2978, i64 0, !dbg !40 + %3336 = insertelement <2 x bfloat> %3335, bfloat %2979, i64 1, !dbg !40 + %3337 = bitcast <2 x bfloat> %3336 to i32, !dbg !40 + %3338 = insertelement <2 x bfloat> poison, bfloat %2980, i64 0, !dbg !40 + %3339 = insertelement <2 x bfloat> %3338, bfloat %2981, i64 1, !dbg !40 + %3340 = bitcast <2 x bfloat> %3339 to i32, !dbg !40 + %3341 = insertelement <2 x bfloat> poison, bfloat %2982, i64 0, !dbg !40 + %3342 = insertelement <2 x bfloat> %3341, bfloat %2983, i64 1, !dbg !40 + %3343 = bitcast <2 x bfloat> %3342 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3334, i32 %3337, i32 %3340, i32 %3343, ptr addrspace(1) %2435, i1 %622) #6, !dbg !40 + %3344 = insertelement <2 x bfloat> poison, bfloat %2984, i64 0, !dbg !40 + %3345 = insertelement <2 x bfloat> %3344, bfloat %2985, i64 1, !dbg !40 + %3346 = bitcast <2 x bfloat> %3345 to i32, !dbg !40 + %3347 = insertelement <2 x bfloat> poison, bfloat %2986, i64 0, !dbg !40 + %3348 = insertelement <2 x bfloat> %3347, bfloat %2987, i64 1, !dbg !40 + %3349 = bitcast <2 x bfloat> %3348 to i32, !dbg !40 + %3350 = insertelement <2 x bfloat> poison, bfloat %2988, i64 0, !dbg !40 + %3351 = insertelement <2 x bfloat> %3350, bfloat %2989, i64 1, !dbg !40 + %3352 = bitcast <2 x bfloat> %3351 to i32, !dbg !40 + %3353 = insertelement <2 x bfloat> poison, bfloat %2990, i64 0, !dbg !40 + %3354 = insertelement <2 x bfloat> %3353, bfloat %2991, i64 1, !dbg !40 + %3355 = bitcast <2 x bfloat> %3354 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3346, i32 %3349, i32 %3352, i32 %3355, ptr addrspace(1) %2433, i1 %640) #6, !dbg !40 + %3356 = insertelement <2 x bfloat> poison, bfloat %2992, i64 0, !dbg !40 + %3357 = insertelement <2 x bfloat> %3356, bfloat %2993, i64 1, !dbg !40 + %3358 = bitcast <2 x bfloat> %3357 to i32, !dbg !40 + %3359 = insertelement <2 x bfloat> poison, bfloat %2994, i64 0, !dbg !40 + %3360 = insertelement <2 x bfloat> %3359, bfloat %2995, i64 1, !dbg !40 + %3361 = bitcast <2 x bfloat> %3360 to i32, !dbg !40 + %3362 = insertelement <2 x bfloat> poison, bfloat %2996, i64 0, !dbg !40 + %3363 = insertelement <2 x bfloat> %3362, bfloat %2997, i64 1, !dbg !40 + %3364 = bitcast <2 x bfloat> %3363 to i32, !dbg !40 + %3365 = insertelement <2 x bfloat> poison, bfloat %2998, i64 0, !dbg !40 + %3366 = insertelement <2 x bfloat> %3365, bfloat %2999, i64 1, !dbg !40 + %3367 = bitcast <2 x bfloat> %3366 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3358, i32 %3361, i32 %3364, i32 %3367, ptr addrspace(1) %2431, i1 %658) #6, !dbg !40 + %3368 = insertelement <2 x bfloat> poison, bfloat %3000, i64 0, !dbg !40 + %3369 = insertelement <2 x bfloat> %3368, bfloat %3001, i64 1, !dbg !40 + %3370 = bitcast <2 x bfloat> %3369 to i32, !dbg !40 + %3371 = insertelement <2 x bfloat> poison, bfloat %3002, i64 0, !dbg !40 + %3372 = insertelement <2 x bfloat> %3371, bfloat %3003, i64 1, !dbg !40 + %3373 = bitcast <2 x bfloat> %3372 to i32, !dbg !40 + %3374 = insertelement <2 x bfloat> poison, bfloat %3004, i64 0, !dbg !40 + %3375 = insertelement <2 x bfloat> %3374, bfloat %3005, i64 1, !dbg !40 + %3376 = bitcast <2 x bfloat> %3375 to i32, !dbg !40 + %3377 = insertelement <2 x bfloat> poison, bfloat %3006, i64 0, !dbg !40 + %3378 = insertelement <2 x bfloat> %3377, bfloat %3007, i64 1, !dbg !40 + %3379 = bitcast <2 x bfloat> %3378 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3370, i32 %3373, i32 %3376, i32 %3379, ptr addrspace(1) %2429, i1 %676) #6, !dbg !40 + %3380 = insertelement <2 x bfloat> poison, bfloat %3008, i64 0, !dbg !40 + %3381 = insertelement <2 x bfloat> %3380, bfloat %3009, i64 1, !dbg !40 + %3382 = bitcast <2 x bfloat> %3381 to i32, !dbg !40 + %3383 = insertelement <2 x bfloat> poison, bfloat %3010, i64 0, !dbg !40 + %3384 = insertelement <2 x bfloat> %3383, bfloat %3011, i64 1, !dbg !40 + %3385 = bitcast <2 x bfloat> %3384 to i32, !dbg !40 + %3386 = insertelement <2 x bfloat> poison, bfloat %3012, i64 0, !dbg !40 + %3387 = insertelement <2 x bfloat> %3386, bfloat %3013, i64 1, !dbg !40 + %3388 = bitcast <2 x bfloat> %3387 to i32, !dbg !40 + %3389 = insertelement <2 x bfloat> poison, bfloat %3014, i64 0, !dbg !40 + %3390 = insertelement <2 x bfloat> %3389, bfloat %3015, i64 1, !dbg !40 + %3391 = bitcast <2 x bfloat> %3390 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3382, i32 %3385, i32 %3388, i32 %3391, ptr addrspace(1) %2427, i1 %694) #6, !dbg !40 + %3392 = insertelement <2 x bfloat> poison, bfloat %3016, i64 0, !dbg !40 + %3393 = insertelement <2 x bfloat> %3392, bfloat %3017, i64 1, !dbg !40 + %3394 = bitcast <2 x bfloat> %3393 to i32, !dbg !40 + %3395 = insertelement <2 x bfloat> poison, bfloat %3018, i64 0, !dbg !40 + %3396 = insertelement <2 x bfloat> %3395, bfloat %3019, i64 1, !dbg !40 + %3397 = bitcast <2 x bfloat> %3396 to i32, !dbg !40 + %3398 = insertelement <2 x bfloat> poison, bfloat %3020, i64 0, !dbg !40 + %3399 = insertelement <2 x bfloat> %3398, bfloat %3021, i64 1, !dbg !40 + %3400 = bitcast <2 x bfloat> %3399 to i32, !dbg !40 + %3401 = insertelement <2 x bfloat> poison, bfloat %3022, i64 0, !dbg !40 + %3402 = insertelement <2 x bfloat> %3401, bfloat %3023, i64 1, !dbg !40 + %3403 = bitcast <2 x bfloat> %3402 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3394, i32 %3397, i32 %3400, i32 %3403, ptr addrspace(1) %2425, i1 %712) #6, !dbg !40 + %3404 = insertelement <2 x bfloat> poison, bfloat %3024, i64 0, !dbg !40 + %3405 = insertelement <2 x bfloat> %3404, bfloat %3025, i64 1, !dbg !40 + %3406 = bitcast <2 x bfloat> %3405 to i32, !dbg !40 + %3407 = insertelement <2 x bfloat> poison, bfloat %3026, i64 0, !dbg !40 + %3408 = insertelement <2 x bfloat> %3407, bfloat %3027, i64 1, !dbg !40 + %3409 = bitcast <2 x bfloat> %3408 to i32, !dbg !40 + %3410 = insertelement <2 x bfloat> poison, bfloat %3028, i64 0, !dbg !40 + %3411 = insertelement <2 x bfloat> %3410, bfloat %3029, i64 1, !dbg !40 + %3412 = bitcast <2 x bfloat> %3411 to i32, !dbg !40 + %3413 = insertelement <2 x bfloat> poison, bfloat %3030, i64 0, !dbg !40 + %3414 = insertelement <2 x bfloat> %3413, bfloat %3031, i64 1, !dbg !40 + %3415 = bitcast <2 x bfloat> %3414 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3406, i32 %3409, i32 %3412, i32 %3415, ptr addrspace(1) %2423, i1 %730) #6, !dbg !40 + %3416 = and i32 %11, 16, !dbg !41 + %3417 = icmp eq i32 %3416, 0, !dbg !41 + %3418 = select i1 %3417, i32 0, i32 320, !dbg !41 + %3419 = and i32 %11, 32, !dbg !41 + %3420 = shl nuw nsw i32 %3419, 1, !dbg !41 + %3421 = xor i32 %3418, %3420, !dbg !41 + %3422 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %3421, !dbg !41 + %3423 = bitcast float %2488 to i32, !dbg !41 + %3424 = bitcast float %2489 to i32, !dbg !41 + %3425 = bitcast float %2490 to i32, !dbg !41 + %3426 = bitcast float %2491 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %3422, i32 %3423, i32 %3424, i32 %3425, i32 %3426) #6, !dbg !41 + %3427 = getelementptr inbounds nuw i8, ptr addrspace(3) %3422, i32 16, !dbg !41 + %3428 = bitcast float %2492 to i32, !dbg !41 + %3429 = bitcast float %2493 to i32, !dbg !41 + %3430 = bitcast float %2494 to i32, !dbg !41 + %3431 = bitcast float %2495 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %3427, i32 %3428, i32 %3429, i32 %3430, i32 %3431) #6, !dbg !41 + %3432 = getelementptr inbounds nuw i8, ptr addrspace(3) %3422, i32 32, !dbg !41 + %3433 = bitcast float %2496 to i32, !dbg !41 + %3434 = bitcast float %2497 to i32, !dbg !41 + %3435 = bitcast float %2498 to i32, !dbg !41 + %3436 = bitcast float %2499 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %3432, i32 %3433, i32 %3434, i32 %3435, i32 %3436) #6, !dbg !41 + %3437 = getelementptr inbounds nuw i8, ptr addrspace(3) %3422, i32 48, !dbg !41 + %3438 = bitcast float %2500 to i32, !dbg !41 + %3439 = bitcast float %2501 to i32, !dbg !41 + %3440 = bitcast float %2502 to i32, !dbg !41 + %3441 = bitcast float %2503 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %3437, i32 %3438, i32 %3439, i32 %3440, i32 %3441) #6, !dbg !41 + %3442 = getelementptr inbounds nuw i8, ptr addrspace(3) %3422, i32 128, !dbg !41 + %3443 = bitcast float %2504 to i32, !dbg !41 + %3444 = bitcast float %2505 to i32, !dbg !41 + %3445 = bitcast float %2506 to i32, !dbg !41 + %3446 = bitcast float %2507 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %3442, i32 %3443, i32 %3444, i32 %3445, i32 %3446) #6, !dbg !41 + %3447 = getelementptr inbounds nuw i8, ptr addrspace(3) %3422, i32 144, !dbg !41 + %3448 = bitcast float %2508 to i32, !dbg !41 + %3449 = bitcast float %2509 to i32, !dbg !41 + %3450 = bitcast float %2510 to i32, !dbg !41 + %3451 = bitcast float %2511 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %3447, i32 %3448, i32 %3449, i32 %3450, i32 %3451) #6, !dbg !41 + %3452 = getelementptr inbounds nuw i8, ptr addrspace(3) %3422, i32 160, !dbg !41 + %3453 = bitcast float %2512 to i32, !dbg !41 + %3454 = bitcast float %2513 to i32, !dbg !41 + %3455 = bitcast float %2514 to i32, !dbg !41 + %3456 = bitcast float %2515 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %3452, i32 %3453, i32 %3454, i32 %3455, i32 %3456) #6, !dbg !41 + %3457 = getelementptr inbounds nuw i8, ptr addrspace(3) %3422, i32 176, !dbg !41 + %3458 = bitcast float %2516 to i32, !dbg !41 + %3459 = bitcast float %2517 to i32, !dbg !41 + %3460 = bitcast float %2518 to i32, !dbg !41 + %3461 = bitcast float %2519 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %3457, i32 %3458, i32 %3459, i32 %3460, i32 %3461) #6, !dbg !41 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !41 + %3462 = and i32 %2413, 60, !dbg !41 + %3463 = shl nuw nsw i32 %11, 6, !dbg !41 + %3464 = and i32 %3463, 64, !dbg !41 + %3465 = shl nuw nsw i32 %3419, 2, !dbg !41 + %3466 = or disjoint i32 %3464, %3465, !dbg !41 + %3467 = or disjoint i32 %3466, %3462, !dbg !41 + %3468 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %3467, !dbg !41 + %3469 = load i32, ptr addrspace(3) %3468, align 4, !dbg !41 + %3470 = xor i32 %3467, 320, !dbg !41 + %3471 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %3470, !dbg !41 + %3472 = load i32, ptr addrspace(3) %3471, align 4, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %3469, ptr addrspace(1) %2421, i1 %2487) #6, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %3472, ptr addrspace(1) %2418, i1 %2486) #6, !dbg !41 + ret void, !dbg !42 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.div.full(float, float) #3 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #4 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #5 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.f(float) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="64" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) } +attributes #4 = { convergent nocallback nounwind } +attributes #5 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #6 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "l2norm_fwd_kernel", linkageName: "l2norm_fwd_kernel", scope: !1, file: !1, line: 69, type: !6, scopeLine: 69, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !10, !11, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 70, column: 24, scope: !5) +!14 = !DILocation(line: 70, column: 30, scope: !5) +!15 = !DILocation(line: 71, column: 16, scope: !5) +!16 = !DILocation(line: 71, column: 34, scope: !5) +!17 = !DILocation(line: 71, column: 21, scope: !5) +!18 = !DILocation(line: 73, column: 16, scope: !5) +!19 = !DILocation(line: 74, column: 30, scope: !5) +!20 = !DILocation(line: 75, column: 29, scope: !5) +!21 = !DILocation(line: 75, column: 14, scope: !5) +!22 = !DILocation(line: 75, column: 33, scope: !5) +!23 = !DILocation(line: 79, column: 18, scope: !5) +!24 = !DILocation(line: 79, column: 47, scope: !5) +!25 = !DILocation(line: 80, column: 38, scope: !5) +!26 = !DILocation(line: 263, column: 15, scope: !27, inlinedAt: !29) +!27 = distinct !DILexicalBlockFile(scope: !5, file: !28, discriminator: 0) +!28 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!29 = !DILocation(line: 293, column: 36, scope: !27, inlinedAt: !30) +!30 = !DILocation(line: 80, column: 43, scope: !31) +!31 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!32 = !DILocation(line: 80, column: 48, scope: !5) +!33 = !DILocation(line: 80, column: 25, scope: !5) +!34 = !DILocation(line: 77, column: 20, scope: !5) +!35 = !DILocation(line: 76, column: 14, scope: !5) +!36 = !DILocation(line: 76, column: 33, scope: !5) +!37 = !DILocation(line: 80, column: 17, scope: !5) +!38 = !DILocation(line: 81, column: 16, scope: !5) +!39 = !DILocation(line: 83, column: 25, scope: !5) +!40 = !DILocation(line: 83, column: 18, scope: !5) +!41 = !DILocation(line: 84, column: 21, scope: !5) +!42 = !DILocation(line: 84, column: 4, scope: !5) diff --git a/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.ptx b/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..390dc6b6b24319ae356015ccef9e075b97a6f91d --- /dev/null +++ b/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.ptx @@ -0,0 +1,3024 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_fwd_kernel // -- Begin function l2norm_fwd_kernel +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; +.global .align 1 .b8 _$_str_$_2[17] = {95, 95, 67, 85, 68, 65, 95, 80, 82, 69, 67, 95, 83, 81, 82, 84}; + // @l2norm_fwd_kernel +.visible .entry l2norm_fwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_2, + .param .f32 l2norm_fwd_kernel_param_3, + .param .u32 l2norm_fwd_kernel_param_4, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_6 +) +.reqntid 64 +{ + .reg .pred %p<50>; + .reg .b16 %rs<310>; + .reg .b32 %r<1476>; + .reg .b64 %rd<238>; + .loc 1 69 0 // l2norm.py:69:0 +$L__func_begin0: + .loc 1 69 0 // l2norm.py:69:0 + +// %bb.0: // %__nv_sqrtf.exit + ld.param.b64 %rd67, [l2norm_fwd_kernel_param_0]; + ld.param.b64 %rd68, [l2norm_fwd_kernel_param_1]; +$L__tmp0: + .loc 1 70 24 // l2norm.py:70:24 + mov.u32 %r300, %ctaid.x; + ld.param.b64 %rd69, [l2norm_fwd_kernel_param_2]; + .loc 1 71 16 // l2norm.py:71:16 + mul.wide.u32 %rd70, %r300, 128; + ld.param.b32 %r301, [l2norm_fwd_kernel_param_3]; + ld.param.s32 %rd71, [l2norm_fwd_kernel_param_4]; + .loc 1 71 34 // l2norm.py:71:34 + mov.u32 %r302, %tid.x; + shr.u32 %r303, %r302, 4; + bfe.u32 %r304, %r302, 4, 2; + or.b32 %r305, %r304, 4; + .loc 1 71 21 // l2norm.py:71:21 + or.b32 %r306, %r304, 120; + or.b32 %r307, %r303, 124; + or.b32 %r308, %r304, 112; + or.b32 %r309, %r304, 116; + or.b32 %r310, %r304, 104; + or.b32 %r311, %r304, 108; + or.b32 %r312, %r304, 96; + or.b32 %r313, %r304, 100; + or.b32 %r314, %r304, 88; + or.b32 %r315, %r304, 92; + or.b32 %r316, %r304, 80; + or.b32 %r317, %r304, 84; + or.b32 %r318, %r304, 72; + or.b32 %r319, %r304, 76; + or.b32 %r320, %r304, 64; + or.b32 %r321, %r304, 68; + or.b32 %r322, %r304, 32; + or.b32 %r323, %r304, 36; + or.b32 %r324, %r304, 40; + or.b32 %r325, %r304, 44; + or.b32 %r326, %r304, 48; + or.b32 %r327, %r304, 52; + or.b32 %r328, %r304, 56; + or.b32 %r329, %r303, 60; + or.b32 %r330, %r304, 16; + or.b32 %r331, %r304, 20; + or.b32 %r332, %r304, 24; + or.b32 %r333, %r304, 28; + or.b32 %r334, %r304, 8; + or.b32 %r335, %r304, 12; + cvt.u64.u32 %rd72, %r321; + cvt.u64.u32 %rd73, %r320; + cvt.u64.u32 %rd74, %r319; + cvt.u64.u32 %rd75, %r318; + cvt.u64.u32 %rd76, %r317; + cvt.u64.u32 %rd77, %r316; + cvt.u64.u32 %rd78, %r315; + cvt.u64.u32 %rd79, %r314; + cvt.u64.u32 %rd80, %r313; + cvt.u64.u32 %rd81, %r312; + cvt.u64.u32 %rd82, %r311; + cvt.u64.u32 %rd83, %r310; + cvt.u64.u32 %rd84, %r309; + cvt.u64.u32 %rd85, %r308; + cvt.u64.u32 %rd86, %r307; + cvt.u64.u32 %rd87, %r306; + cvt.u64.u32 %rd88, %r329; + cvt.u64.u32 %rd89, %r328; + cvt.u64.u32 %rd90, %r327; + cvt.u64.u32 %rd91, %r326; + cvt.u64.u32 %rd92, %r325; + cvt.u64.u32 %rd93, %r324; + cvt.u64.u32 %rd94, %r323; + cvt.u64.u32 %rd95, %r322; + cvt.u64.u32 %rd96, %r333; + cvt.u64.u32 %rd97, %r332; + cvt.u64.u32 %rd98, %r331; + cvt.u64.u32 %rd99, %r330; + cvt.u64.u32 %rd100, %r335; + cvt.u64.u32 %rd101, %r334; + cvt.u64.u32 %rd102, %r305; + cvt.u64.u32 %rd103, %r304; + or.b64 %rd104, %rd70, %rd103; + or.b64 %rd105, %rd70, %rd102; + or.b64 %rd106, %rd70, %rd101; + or.b64 %rd107, %rd70, %rd100; + or.b64 %rd108, %rd70, %rd99; + or.b64 %rd109, %rd70, %rd98; + or.b64 %rd110, %rd70, %rd97; + or.b64 %rd111, %rd70, %rd96; + or.b64 %rd112, %rd70, %rd95; + or.b64 %rd113, %rd70, %rd94; + or.b64 %rd114, %rd70, %rd93; + or.b64 %rd115, %rd70, %rd92; + or.b64 %rd116, %rd70, %rd91; + or.b64 %rd117, %rd70, %rd90; + or.b64 %rd118, %rd70, %rd89; + or.b64 %rd119, %rd70, %rd88; + or.b64 %rd120, %rd70, %rd87; + or.b64 %rd121, %rd70, %rd86; + or.b64 %rd122, %rd70, %rd85; + or.b64 %rd123, %rd70, %rd84; + or.b64 %rd124, %rd70, %rd83; + or.b64 %rd125, %rd70, %rd82; + or.b64 %rd126, %rd70, %rd81; + or.b64 %rd127, %rd70, %rd80; + or.b64 %rd128, %rd70, %rd79; + or.b64 %rd129, %rd70, %rd78; + or.b64 %rd130, %rd70, %rd77; + or.b64 %rd131, %rd70, %rd76; + or.b64 %rd132, %rd70, %rd75; + or.b64 %rd133, %rd70, %rd74; + or.b64 %rd134, %rd70, %rd73; + or.b64 %rd135, %rd70, %rd72; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p35, %rd135, %rd71; + selp.b16 %rs1, 1, 0, %p35; + shl.b16 %rs2, %rs1, 2; + setp.lt.s64 %p36, %rd134, %rd71; + selp.b16 %rs3, -1, 0, %p36; + shl.b16 %rs4, %rs3, 3; + or.b16 %rs5, %rs4, %rs2; + setp.lt.s64 %p37, %rd133, %rd71; + selp.b16 %rs6, 1, 0, %p37; + setp.lt.s64 %p38, %rd132, %rd71; + selp.b16 %rs7, -1, 0, %p38; + shl.b16 %rs8, %rs7, 1; + or.b16 %rs9, %rs6, %rs8; + and.b16 %rs10, %rs9, 3; + or.b16 %rs11, %rs10, %rs5; + shl.b16 %rs12, %rs11, 12; + setp.lt.s64 %p39, %rd131, %rd71; + selp.b16 %rs13, 1, 0, %p39; + shl.b16 %rs14, %rs13, 2; + setp.lt.s64 %p40, %rd130, %rd71; + selp.b16 %rs15, -1, 0, %p40; + shl.b16 %rs16, %rs15, 3; + or.b16 %rs17, %rs16, %rs14; + setp.lt.s64 %p41, %rd129, %rd71; + selp.b16 %rs18, 1, 0, %p41; + setp.lt.s64 %p42, %rd128, %rd71; + selp.b16 %rs19, -1, 0, %p42; + shl.b16 %rs20, %rs19, 1; + or.b16 %rs21, %rs18, %rs20; + and.b16 %rs22, %rs21, 3; + or.b16 %rs23, %rs22, %rs17; + and.b16 %rs24, %rs23, 15; + shl.b16 %rs25, %rs24, 8; + or.b16 %rs26, %rs12, %rs25; + setp.lt.s64 %p43, %rd127, %rd71; + selp.b16 %rs27, 1, 0, %p43; + shl.b16 %rs28, %rs27, 2; + setp.lt.s64 %p44, %rd126, %rd71; + selp.b16 %rs29, -1, 0, %p44; + shl.b16 %rs30, %rs29, 3; + or.b16 %rs31, %rs30, %rs28; + setp.lt.s64 %p45, %rd125, %rd71; + selp.b16 %rs32, 1, 0, %p45; + setp.lt.s64 %p46, %rd124, %rd71; + selp.b16 %rs33, -1, 0, %p46; + shl.b16 %rs34, %rs33, 1; + or.b16 %rs35, %rs32, %rs34; + and.b16 %rs36, %rs35, 3; + or.b16 %rs37, %rs36, %rs31; + shl.b16 %rs38, %rs37, 4; + setp.lt.s64 %p47, %rd123, %rd71; + selp.b16 %rs39, 1, 0, %p47; + shl.b16 %rs40, %rs39, 2; + setp.lt.s64 %p48, %rd122, %rd71; + selp.b16 %rs41, -1, 0, %p48; + shl.b16 %rs42, %rs41, 3; + or.b16 %rs43, %rs42, %rs40; + setp.lt.s64 %p32, %rd121, %rd71; + selp.b16 %rs44, 1, 0, %p32; + setp.lt.s64 %p49, %rd120, %rd71; + selp.b16 %rs45, -1, 0, %p49; + shl.b16 %rs46, %rs45, 1; + or.b16 %rs47, %rs44, %rs46; + and.b16 %rs48, %rs47, 3; + or.b16 %rs49, %rs48, %rs43; + and.b16 %rs50, %rs49, 15; + or.b16 %rs51, %rs50, %rs38; + and.b16 %rs52, %rs51, 255; + or.b16 %rs53, %rs52, %rs26; + cvt.u32.u16 %r336, %rs53; + setp.lt.s64 %p16, %rd119, %rd71; + setp.lt.s64 %p15, %rd118, %rd71; + setp.lt.s64 %p14, %rd117, %rd71; + setp.lt.s64 %p13, %rd116, %rd71; + setp.lt.s64 %p12, %rd115, %rd71; + setp.lt.s64 %p11, %rd114, %rd71; + setp.lt.s64 %p10, %rd113, %rd71; + setp.lt.s64 %p9, %rd112, %rd71; + setp.lt.s64 %p1, %rd111, %rd71; + setp.lt.s64 %p8, %rd110, %rd71; + setp.lt.s64 %p7, %rd109, %rd71; + setp.lt.s64 %p6, %rd108, %rd71; + setp.lt.s64 %p5, %rd107, %rd71; + setp.lt.s64 %p4, %rd106, %rd71; + setp.lt.s64 %p3, %rd105, %rd71; + setp.lt.s64 %p2, %rd104, %rd71; + .loc 1 74 30 // l2norm.py:74:30 + shl.b32 %r337, %r302, 3; + and.b32 %r338, %r337, 120; + .loc 1 75 14 // l2norm.py:75:14 + shl.b64 %rd136, %rd104, 8; + add.s64 %rd137, %rd67, %rd136; + shl.b64 %rd138, %rd105, 8; + add.s64 %rd139, %rd67, %rd138; + shl.b64 %rd140, %rd106, 8; + add.s64 %rd141, %rd67, %rd140; + shl.b64 %rd142, %rd107, 8; + add.s64 %rd143, %rd67, %rd142; + shl.b64 %rd144, %rd108, 8; + add.s64 %rd145, %rd67, %rd144; + shl.b64 %rd146, %rd109, 8; + add.s64 %rd147, %rd67, %rd146; + shl.b64 %rd148, %rd110, 8; + add.s64 %rd149, %rd67, %rd148; + shl.b64 %rd150, %rd111, 8; + add.s64 %rd151, %rd67, %rd150; + shl.b64 %rd152, %rd112, 8; + add.s64 %rd153, %rd67, %rd152; + shl.b64 %rd154, %rd113, 8; + add.s64 %rd155, %rd67, %rd154; + shl.b64 %rd156, %rd114, 8; + add.s64 %rd157, %rd67, %rd156; + shl.b64 %rd158, %rd115, 8; + add.s64 %rd159, %rd67, %rd158; + shl.b64 %rd160, %rd116, 8; + add.s64 %rd161, %rd67, %rd160; + shl.b64 %rd162, %rd117, 8; + add.s64 %rd163, %rd67, %rd162; + shl.b64 %rd164, %rd118, 8; + add.s64 %rd165, %rd67, %rd164; + shl.b64 %rd166, %rd119, 8; + add.s64 %rd167, %rd67, %rd166; + shl.b64 %rd168, %rd134, 8; + add.s64 %rd169, %rd67, %rd168; + shl.b64 %rd170, %rd135, 8; + add.s64 %rd171, %rd67, %rd170; + shl.b64 %rd172, %rd132, 8; + add.s64 %rd173, %rd67, %rd172; + shl.b64 %rd174, %rd133, 8; + add.s64 %rd175, %rd67, %rd174; + shl.b64 %rd176, %rd130, 8; + add.s64 %rd177, %rd67, %rd176; + shl.b64 %rd178, %rd131, 8; + add.s64 %rd179, %rd67, %rd178; + shl.b64 %rd180, %rd128, 8; + add.s64 %rd181, %rd67, %rd180; + shl.b64 %rd182, %rd129, 8; + add.s64 %rd183, %rd67, %rd182; + shl.b64 %rd184, %rd126, 8; + add.s64 %rd185, %rd67, %rd184; + shl.b64 %rd186, %rd127, 8; + add.s64 %rd187, %rd67, %rd186; + shl.b64 %rd188, %rd124, 8; + add.s64 %rd189, %rd67, %rd188; + shl.b64 %rd190, %rd125, 8; + add.s64 %rd191, %rd67, %rd190; + shl.b64 %rd192, %rd122, 8; + add.s64 %rd193, %rd67, %rd192; + shl.b64 %rd194, %rd123, 8; + add.s64 %rd195, %rd67, %rd194; + shl.b64 %rd196, %rd120, 8; + add.s64 %rd197, %rd67, %rd196; + shl.b64 %rd198, %rd121, 8; + add.s64 %rd199, %rd67, %rd198; + .loc 1 75 33 // l2norm.py:75:33 + mul.wide.u32 %rd200, %r338, 2; + add.s64 %rd1, %rd137, %rd200; + add.s64 %rd2, %rd139, %rd200; + add.s64 %rd3, %rd141, %rd200; + add.s64 %rd4, %rd143, %rd200; + add.s64 %rd5, %rd145, %rd200; + add.s64 %rd6, %rd147, %rd200; + add.s64 %rd7, %rd149, %rd200; + add.s64 %rd8, %rd151, %rd200; + add.s64 %rd9, %rd153, %rd200; + add.s64 %rd10, %rd155, %rd200; + add.s64 %rd11, %rd157, %rd200; + add.s64 %rd12, %rd159, %rd200; + add.s64 %rd13, %rd161, %rd200; + add.s64 %rd14, %rd163, %rd200; + add.s64 %rd15, %rd165, %rd200; + add.s64 %rd16, %rd167, %rd200; + add.s64 %rd17, %rd169, %rd200; + add.s64 %rd18, %rd171, %rd200; + add.s64 %rd19, %rd173, %rd200; + add.s64 %rd20, %rd175, %rd200; + add.s64 %rd21, %rd177, %rd200; + add.s64 %rd22, %rd179, %rd200; + add.s64 %rd23, %rd181, %rd200; + add.s64 %rd24, %rd183, %rd200; + add.s64 %rd25, %rd185, %rd200; + add.s64 %rd26, %rd187, %rd200; + add.s64 %rd27, %rd189, %rd200; + add.s64 %rd28, %rd191, %rd200; + add.s64 %rd29, %rd193, %rd200; + add.s64 %rd30, %rd195, %rd200; + add.s64 %rd31, %rd197, %rd200; + add.s64 %rd32, %rd199, %rd200; + mov.b32 %r5, 0; + .loc 1 79 18 // l2norm.py:79:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p2 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs54, %rs55}, %r1; + mov.b32 {%rs56, %rs57}, %r2; + mov.b32 {%rs58, %rs59}, %r3; + mov.b32 {%rs60, %rs61}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p3 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs62, %rs63}, %r6; + mov.b32 {%rs64, %rs65}, %r7; + mov.b32 {%rs66, %rs67}, %r8; + mov.b32 {%rs68, %rs69}, %r9; + // begin inline asm + mov.u32 %r10, %r5; + mov.u32 %r11, %r5; + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + @%p4 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs70, %rs71}, %r10; + mov.b32 {%rs72, %rs73}, %r11; + mov.b32 {%rs74, %rs75}, %r12; + mov.b32 {%rs76, %rs77}, %r13; + // begin inline asm + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + @%p5 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ]; + // end inline asm + mov.b32 {%rs78, %rs79}, %r14; + mov.b32 {%rs80, %rs81}, %r15; + mov.b32 {%rs82, %rs83}, %r16; + mov.b32 {%rs84, %rs85}, %r17; + // begin inline asm + mov.u32 %r18, %r5; + mov.u32 %r19, %r5; + mov.u32 %r20, %r5; + mov.u32 %r21, %r5; + @%p6 ld.global.v4.b32 { %r18, %r19, %r20, %r21 }, [ %rd5 + 0 ]; + // end inline asm + mov.b32 {%rs86, %rs87}, %r18; + mov.b32 {%rs88, %rs89}, %r19; + mov.b32 {%rs90, %rs91}, %r20; + mov.b32 {%rs92, %rs93}, %r21; + // begin inline asm + mov.u32 %r22, %r5; + mov.u32 %r23, %r5; + mov.u32 %r24, %r5; + mov.u32 %r25, %r5; + @%p7 ld.global.v4.b32 { %r22, %r23, %r24, %r25 }, [ %rd6 + 0 ]; + // end inline asm + mov.b32 {%rs94, %rs95}, %r22; + mov.b32 {%rs96, %rs97}, %r23; + mov.b32 {%rs98, %rs99}, %r24; + mov.b32 {%rs100, %rs101}, %r25; + // begin inline asm + mov.u32 %r26, %r5; + mov.u32 %r27, %r5; + mov.u32 %r28, %r5; + mov.u32 %r29, %r5; + @%p8 ld.global.v4.b32 { %r26, %r27, %r28, %r29 }, [ %rd7 + 0 ]; + // end inline asm + mov.b32 {%rs102, %rs103}, %r26; + mov.b32 {%rs104, %rs105}, %r27; + mov.b32 {%rs106, %rs107}, %r28; + mov.b32 {%rs108, %rs109}, %r29; + // begin inline asm + mov.u32 %r30, %r5; + mov.u32 %r31, %r5; + mov.u32 %r32, %r5; + mov.u32 %r33, %r5; + @%p1 ld.global.v4.b32 { %r30, %r31, %r32, %r33 }, [ %rd8 + 0 ]; + // end inline asm + mov.b32 {%rs110, %rs111}, %r30; + mov.b32 {%rs112, %rs113}, %r31; + mov.b32 {%rs114, %rs115}, %r32; + mov.b32 {%rs116, %rs117}, %r33; + // begin inline asm + mov.u32 %r34, %r5; + mov.u32 %r35, %r5; + mov.u32 %r36, %r5; + mov.u32 %r37, %r5; + @%p9 ld.global.v4.b32 { %r34, %r35, %r36, %r37 }, [ %rd9 + 0 ]; + // end inline asm + mov.b32 {%rs118, %rs119}, %r34; + mov.b32 {%rs120, %rs121}, %r35; + mov.b32 {%rs122, %rs123}, %r36; + mov.b32 {%rs124, %rs125}, %r37; + // begin inline asm + mov.u32 %r38, %r5; + mov.u32 %r39, %r5; + mov.u32 %r40, %r5; + mov.u32 %r41, %r5; + @%p10 ld.global.v4.b32 { %r38, %r39, %r40, %r41 }, [ %rd10 + 0 ]; + // end inline asm + mov.b32 {%rs126, %rs127}, %r38; + mov.b32 {%rs128, %rs129}, %r39; + mov.b32 {%rs130, %rs131}, %r40; + mov.b32 {%rs132, %rs133}, %r41; + // begin inline asm + mov.u32 %r42, %r5; + mov.u32 %r43, %r5; + mov.u32 %r44, %r5; + mov.u32 %r45, %r5; + @%p11 ld.global.v4.b32 { %r42, %r43, %r44, %r45 }, [ %rd11 + 0 ]; + // end inline asm + mov.b32 {%rs134, %rs135}, %r42; + mov.b32 {%rs136, %rs137}, %r43; + mov.b32 {%rs138, %rs139}, %r44; + mov.b32 {%rs140, %rs141}, %r45; + // begin inline asm + mov.u32 %r46, %r5; + mov.u32 %r47, %r5; + mov.u32 %r48, %r5; + mov.u32 %r49, %r5; + @%p12 ld.global.v4.b32 { %r46, %r47, %r48, %r49 }, [ %rd12 + 0 ]; + // end inline asm + mov.b32 {%rs142, %rs143}, %r46; + mov.b32 {%rs144, %rs145}, %r47; + mov.b32 {%rs146, %rs147}, %r48; + mov.b32 {%rs148, %rs149}, %r49; + // begin inline asm + mov.u32 %r50, %r5; + mov.u32 %r51, %r5; + mov.u32 %r52, %r5; + mov.u32 %r53, %r5; + @%p13 ld.global.v4.b32 { %r50, %r51, %r52, %r53 }, [ %rd13 + 0 ]; + // end inline asm + mov.b32 {%rs150, %rs151}, %r50; + mov.b32 {%rs152, %rs153}, %r51; + mov.b32 {%rs154, %rs155}, %r52; + mov.b32 {%rs156, %rs157}, %r53; + // begin inline asm + mov.u32 %r54, %r5; + mov.u32 %r55, %r5; + mov.u32 %r56, %r5; + mov.u32 %r57, %r5; + @%p14 ld.global.v4.b32 { %r54, %r55, %r56, %r57 }, [ %rd14 + 0 ]; + // end inline asm + mov.b32 {%rs158, %rs159}, %r54; + mov.b32 {%rs160, %rs161}, %r55; + mov.b32 {%rs162, %rs163}, %r56; + mov.b32 {%rs164, %rs165}, %r57; + // begin inline asm + mov.u32 %r58, %r5; + mov.u32 %r59, %r5; + mov.u32 %r60, %r5; + mov.u32 %r61, %r5; + @%p15 ld.global.v4.b32 { %r58, %r59, %r60, %r61 }, [ %rd15 + 0 ]; + // end inline asm + mov.b32 {%rs166, %rs167}, %r58; + mov.b32 {%rs168, %rs169}, %r59; + mov.b32 {%rs170, %rs171}, %r60; + mov.b32 {%rs172, %rs173}, %r61; + // begin inline asm + mov.u32 %r62, %r5; + mov.u32 %r63, %r5; + mov.u32 %r64, %r5; + mov.u32 %r65, %r5; + @%p16 ld.global.v4.b32 { %r62, %r63, %r64, %r65 }, [ %rd16 + 0 ]; + // end inline asm + mov.b32 {%rs174, %rs175}, %r62; + mov.b32 {%rs176, %rs177}, %r63; + mov.b32 {%rs178, %rs179}, %r64; + mov.b32 {%rs180, %rs181}, %r65; + shr.u32 %r339, %r336, 15; + and.b32 %r340, %r339, 1; + setp.ne.b32 %p17, %r340, 0; + // begin inline asm + mov.u32 %r66, %r5; + mov.u32 %r67, %r5; + mov.u32 %r68, %r5; + mov.u32 %r69, %r5; + @%p17 ld.global.v4.b32 { %r66, %r67, %r68, %r69 }, [ %rd17 + 0 ]; + // end inline asm + mov.b32 {%rs182, %rs183}, %r66; + mov.b32 {%rs184, %rs185}, %r67; + mov.b32 {%rs186, %rs187}, %r68; + mov.b32 {%rs188, %rs189}, %r69; + shr.u32 %r341, %r336, 14; + and.b32 %r342, %r341, 1; + setp.ne.b32 %p18, %r342, 0; + // begin inline asm + mov.u32 %r70, %r5; + mov.u32 %r71, %r5; + mov.u32 %r72, %r5; + mov.u32 %r73, %r5; + @%p18 ld.global.v4.b32 { %r70, %r71, %r72, %r73 }, [ %rd18 + 0 ]; + // end inline asm + mov.b32 {%rs190, %rs191}, %r70; + mov.b32 {%rs192, %rs193}, %r71; + mov.b32 {%rs194, %rs195}, %r72; + mov.b32 {%rs196, %rs197}, %r73; + shr.u32 %r343, %r336, 13; + and.b32 %r344, %r343, 1; + setp.ne.b32 %p19, %r344, 0; + // begin inline asm + mov.u32 %r74, %r5; + mov.u32 %r75, %r5; + mov.u32 %r76, %r5; + mov.u32 %r77, %r5; + @%p19 ld.global.v4.b32 { %r74, %r75, %r76, %r77 }, [ %rd19 + 0 ]; + // end inline asm + mov.b32 {%rs198, %rs199}, %r74; + mov.b32 {%rs200, %rs201}, %r75; + mov.b32 {%rs202, %rs203}, %r76; + mov.b32 {%rs204, %rs205}, %r77; + shr.u32 %r345, %r336, 12; + and.b32 %r346, %r345, 1; + setp.ne.b32 %p20, %r346, 0; + // begin inline asm + mov.u32 %r78, %r5; + mov.u32 %r79, %r5; + mov.u32 %r80, %r5; + mov.u32 %r81, %r5; + @%p20 ld.global.v4.b32 { %r78, %r79, %r80, %r81 }, [ %rd20 + 0 ]; + // end inline asm + mov.b32 {%rs206, %rs207}, %r78; + mov.b32 {%rs208, %rs209}, %r79; + mov.b32 {%rs210, %rs211}, %r80; + mov.b32 {%rs212, %rs213}, %r81; + shr.u32 %r347, %r336, 11; + and.b32 %r348, %r347, 1; + setp.ne.b32 %p21, %r348, 0; + // begin inline asm + mov.u32 %r82, %r5; + mov.u32 %r83, %r5; + mov.u32 %r84, %r5; + mov.u32 %r85, %r5; + @%p21 ld.global.v4.b32 { %r82, %r83, %r84, %r85 }, [ %rd21 + 0 ]; + // end inline asm + mov.b32 {%rs214, %rs215}, %r82; + mov.b32 {%rs216, %rs217}, %r83; + mov.b32 {%rs218, %rs219}, %r84; + mov.b32 {%rs220, %rs221}, %r85; + shr.u32 %r349, %r336, 10; + and.b32 %r350, %r349, 1; + setp.ne.b32 %p22, %r350, 0; + // begin inline asm + mov.u32 %r86, %r5; + mov.u32 %r87, %r5; + mov.u32 %r88, %r5; + mov.u32 %r89, %r5; + @%p22 ld.global.v4.b32 { %r86, %r87, %r88, %r89 }, [ %rd22 + 0 ]; + // end inline asm + mov.b32 {%rs222, %rs223}, %r86; + mov.b32 {%rs224, %rs225}, %r87; + mov.b32 {%rs226, %rs227}, %r88; + mov.b32 {%rs228, %rs229}, %r89; + shr.u32 %r351, %r336, 9; + and.b32 %r352, %r351, 1; + setp.ne.b32 %p23, %r352, 0; + // begin inline asm + mov.u32 %r90, %r5; + mov.u32 %r91, %r5; + mov.u32 %r92, %r5; + mov.u32 %r93, %r5; + @%p23 ld.global.v4.b32 { %r90, %r91, %r92, %r93 }, [ %rd23 + 0 ]; + // end inline asm + mov.b32 {%rs230, %rs231}, %r90; + mov.b32 {%rs232, %rs233}, %r91; + mov.b32 {%rs234, %rs235}, %r92; + mov.b32 {%rs236, %rs237}, %r93; + shr.u32 %r353, %r336, 8; + and.b32 %r354, %r353, 1; + setp.ne.b32 %p24, %r354, 0; + // begin inline asm + mov.u32 %r94, %r5; + mov.u32 %r95, %r5; + mov.u32 %r96, %r5; + mov.u32 %r97, %r5; + @%p24 ld.global.v4.b32 { %r94, %r95, %r96, %r97 }, [ %rd24 + 0 ]; + // end inline asm + mov.b32 {%rs238, %rs239}, %r94; + mov.b32 {%rs240, %rs241}, %r95; + mov.b32 {%rs242, %rs243}, %r96; + mov.b32 {%rs244, %rs245}, %r97; + shr.u32 %r355, %r336, 7; + and.b32 %r356, %r355, 1; + setp.ne.b32 %p25, %r356, 0; + // begin inline asm + mov.u32 %r98, %r5; + mov.u32 %r99, %r5; + mov.u32 %r100, %r5; + mov.u32 %r101, %r5; + @%p25 ld.global.v4.b32 { %r98, %r99, %r100, %r101 }, [ %rd25 + 0 ]; + // end inline asm + mov.b32 {%rs246, %rs247}, %r98; + mov.b32 {%rs248, %rs249}, %r99; + mov.b32 {%rs250, %rs251}, %r100; + mov.b32 {%rs252, %rs253}, %r101; + shr.u32 %r357, %r336, 6; + and.b32 %r358, %r357, 1; + setp.ne.b32 %p26, %r358, 0; + // begin inline asm + mov.u32 %r102, %r5; + mov.u32 %r103, %r5; + mov.u32 %r104, %r5; + mov.u32 %r105, %r5; + @%p26 ld.global.v4.b32 { %r102, %r103, %r104, %r105 }, [ %rd26 + 0 ]; + // end inline asm + mov.b32 {%rs254, %rs255}, %r102; + mov.b32 {%rs256, %rs257}, %r103; + mov.b32 {%rs258, %rs259}, %r104; + mov.b32 {%rs260, %rs261}, %r105; + shr.u32 %r359, %r336, 5; + and.b32 %r360, %r359, 1; + setp.ne.b32 %p27, %r360, 0; + // begin inline asm + mov.u32 %r106, %r5; + mov.u32 %r107, %r5; + mov.u32 %r108, %r5; + mov.u32 %r109, %r5; + @%p27 ld.global.v4.b32 { %r106, %r107, %r108, %r109 }, [ %rd27 + 0 ]; + // end inline asm + mov.b32 {%rs262, %rs263}, %r106; + mov.b32 {%rs264, %rs265}, %r107; + mov.b32 {%rs266, %rs267}, %r108; + mov.b32 {%rs268, %rs269}, %r109; + shr.u32 %r361, %r336, 4; + and.b32 %r362, %r361, 1; + setp.ne.b32 %p28, %r362, 0; + // begin inline asm + mov.u32 %r110, %r5; + mov.u32 %r111, %r5; + mov.u32 %r112, %r5; + mov.u32 %r113, %r5; + @%p28 ld.global.v4.b32 { %r110, %r111, %r112, %r113 }, [ %rd28 + 0 ]; + // end inline asm + mov.b32 {%rs270, %rs271}, %r110; + mov.b32 {%rs272, %rs273}, %r111; + mov.b32 {%rs274, %rs275}, %r112; + mov.b32 {%rs276, %rs277}, %r113; + shr.u32 %r363, %r336, 3; + and.b32 %r364, %r363, 1; + setp.ne.b32 %p29, %r364, 0; + // begin inline asm + mov.u32 %r114, %r5; + mov.u32 %r115, %r5; + mov.u32 %r116, %r5; + mov.u32 %r117, %r5; + @%p29 ld.global.v4.b32 { %r114, %r115, %r116, %r117 }, [ %rd29 + 0 ]; + // end inline asm + mov.b32 {%rs278, %rs279}, %r114; + mov.b32 {%rs280, %rs281}, %r115; + mov.b32 {%rs282, %rs283}, %r116; + mov.b32 {%rs284, %rs285}, %r117; + shr.u32 %r365, %r336, 2; + and.b32 %r366, %r365, 1; + setp.ne.b32 %p30, %r366, 0; + // begin inline asm + mov.u32 %r118, %r5; + mov.u32 %r119, %r5; + mov.u32 %r120, %r5; + mov.u32 %r121, %r5; + @%p30 ld.global.v4.b32 { %r118, %r119, %r120, %r121 }, [ %rd30 + 0 ]; + // end inline asm + mov.b32 {%rs286, %rs287}, %r118; + mov.b32 {%rs288, %rs289}, %r119; + mov.b32 {%rs290, %rs291}, %r120; + mov.b32 {%rs292, %rs293}, %r121; + shr.u32 %r367, %r336, 1; + and.b32 %r368, %r367, 1; + setp.ne.b32 %p31, %r368, 0; + // begin inline asm + mov.u32 %r122, %r5; + mov.u32 %r123, %r5; + mov.u32 %r124, %r5; + mov.u32 %r125, %r5; + @%p31 ld.global.v4.b32 { %r122, %r123, %r124, %r125 }, [ %rd31 + 0 ]; + // end inline asm + mov.b32 {%rs294, %rs295}, %r122; + mov.b32 {%rs296, %rs297}, %r123; + mov.b32 {%rs298, %rs299}, %r124; + mov.b32 {%rs300, %rs301}, %r125; + // begin inline asm + mov.u32 %r126, %r5; + mov.u32 %r127, %r5; + mov.u32 %r128, %r5; + mov.u32 %r129, %r5; + @%p32 ld.global.v4.b32 { %r126, %r127, %r128, %r129 }, [ %rd32 + 0 ]; + // end inline asm + mov.b32 {%rs302, %rs303}, %r126; + mov.b32 {%rs304, %rs305}, %r127; + mov.b32 {%rs306, %rs307}, %r128; + mov.b32 {%rs308, %rs309}, %r129; + .loc 1 79 47 // l2norm.py:79:47 + cvt.f32.bf16 %r369, %rs54; + cvt.f32.bf16 %r370, %rs55; + cvt.f32.bf16 %r371, %rs56; + cvt.f32.bf16 %r372, %rs57; + cvt.f32.bf16 %r373, %rs58; + cvt.f32.bf16 %r374, %rs59; + cvt.f32.bf16 %r375, %rs60; + cvt.f32.bf16 %r376, %rs61; + cvt.f32.bf16 %r377, %rs62; + cvt.f32.bf16 %r378, %rs63; + cvt.f32.bf16 %r379, %rs64; + cvt.f32.bf16 %r380, %rs65; + cvt.f32.bf16 %r381, %rs66; + cvt.f32.bf16 %r382, %rs67; + cvt.f32.bf16 %r383, %rs68; + cvt.f32.bf16 %r384, %rs69; + cvt.f32.bf16 %r385, %rs70; + cvt.f32.bf16 %r386, %rs71; + cvt.f32.bf16 %r387, %rs72; + cvt.f32.bf16 %r388, %rs73; + cvt.f32.bf16 %r389, %rs74; + cvt.f32.bf16 %r390, %rs75; + cvt.f32.bf16 %r391, %rs76; + cvt.f32.bf16 %r392, %rs77; + cvt.f32.bf16 %r393, %rs78; + cvt.f32.bf16 %r394, %rs79; + cvt.f32.bf16 %r395, %rs80; + cvt.f32.bf16 %r396, %rs81; + cvt.f32.bf16 %r397, %rs82; + cvt.f32.bf16 %r398, %rs83; + cvt.f32.bf16 %r399, %rs84; + cvt.f32.bf16 %r400, %rs85; + cvt.f32.bf16 %r401, %rs86; + cvt.f32.bf16 %r402, %rs87; + cvt.f32.bf16 %r403, %rs88; + cvt.f32.bf16 %r404, %rs89; + cvt.f32.bf16 %r405, %rs90; + cvt.f32.bf16 %r406, %rs91; + cvt.f32.bf16 %r407, %rs92; + cvt.f32.bf16 %r408, %rs93; + cvt.f32.bf16 %r409, %rs94; + cvt.f32.bf16 %r410, %rs95; + cvt.f32.bf16 %r411, %rs96; + cvt.f32.bf16 %r412, %rs97; + cvt.f32.bf16 %r413, %rs98; + cvt.f32.bf16 %r414, %rs99; + cvt.f32.bf16 %r415, %rs100; + cvt.f32.bf16 %r416, %rs101; + cvt.f32.bf16 %r417, %rs102; + cvt.f32.bf16 %r418, %rs103; + cvt.f32.bf16 %r419, %rs104; + cvt.f32.bf16 %r420, %rs105; + cvt.f32.bf16 %r421, %rs106; + cvt.f32.bf16 %r422, %rs107; + cvt.f32.bf16 %r423, %rs108; + cvt.f32.bf16 %r424, %rs109; + cvt.f32.bf16 %r425, %rs110; + cvt.f32.bf16 %r426, %rs111; + cvt.f32.bf16 %r427, %rs112; + cvt.f32.bf16 %r428, %rs113; + cvt.f32.bf16 %r429, %rs114; + cvt.f32.bf16 %r430, %rs115; + cvt.f32.bf16 %r431, %rs116; + cvt.f32.bf16 %r432, %rs117; + cvt.f32.bf16 %r433, %rs118; + cvt.f32.bf16 %r434, %rs119; + cvt.f32.bf16 %r435, %rs120; + cvt.f32.bf16 %r436, %rs121; + cvt.f32.bf16 %r437, %rs122; + cvt.f32.bf16 %r438, %rs123; + cvt.f32.bf16 %r439, %rs124; + cvt.f32.bf16 %r440, %rs125; + cvt.f32.bf16 %r441, %rs126; + cvt.f32.bf16 %r442, %rs127; + cvt.f32.bf16 %r443, %rs128; + cvt.f32.bf16 %r444, %rs129; + cvt.f32.bf16 %r445, %rs130; + cvt.f32.bf16 %r446, %rs131; + cvt.f32.bf16 %r447, %rs132; + cvt.f32.bf16 %r448, %rs133; + cvt.f32.bf16 %r449, %rs134; + cvt.f32.bf16 %r450, %rs135; + cvt.f32.bf16 %r451, %rs136; + cvt.f32.bf16 %r452, %rs137; + cvt.f32.bf16 %r453, %rs138; + cvt.f32.bf16 %r454, %rs139; + cvt.f32.bf16 %r455, %rs140; + cvt.f32.bf16 %r456, %rs141; + cvt.f32.bf16 %r457, %rs142; + cvt.f32.bf16 %r458, %rs143; + cvt.f32.bf16 %r459, %rs144; + cvt.f32.bf16 %r460, %rs145; + cvt.f32.bf16 %r461, %rs146; + cvt.f32.bf16 %r462, %rs147; + cvt.f32.bf16 %r463, %rs148; + cvt.f32.bf16 %r464, %rs149; + cvt.f32.bf16 %r465, %rs150; + cvt.f32.bf16 %r466, %rs151; + cvt.f32.bf16 %r467, %rs152; + cvt.f32.bf16 %r468, %rs153; + cvt.f32.bf16 %r469, %rs154; + cvt.f32.bf16 %r470, %rs155; + cvt.f32.bf16 %r471, %rs156; + cvt.f32.bf16 %r472, %rs157; + cvt.f32.bf16 %r473, %rs158; + cvt.f32.bf16 %r474, %rs159; + cvt.f32.bf16 %r475, %rs160; + cvt.f32.bf16 %r476, %rs161; + cvt.f32.bf16 %r477, %rs162; + cvt.f32.bf16 %r478, %rs163; + cvt.f32.bf16 %r479, %rs164; + cvt.f32.bf16 %r480, %rs165; + cvt.f32.bf16 %r481, %rs166; + cvt.f32.bf16 %r482, %rs167; + cvt.f32.bf16 %r483, %rs168; + cvt.f32.bf16 %r484, %rs169; + cvt.f32.bf16 %r485, %rs170; + cvt.f32.bf16 %r486, %rs171; + cvt.f32.bf16 %r487, %rs172; + cvt.f32.bf16 %r488, %rs173; + cvt.f32.bf16 %r489, %rs174; + cvt.f32.bf16 %r490, %rs175; + cvt.f32.bf16 %r491, %rs176; + cvt.f32.bf16 %r492, %rs177; + cvt.f32.bf16 %r493, %rs178; + cvt.f32.bf16 %r494, %rs179; + cvt.f32.bf16 %r495, %rs180; + cvt.f32.bf16 %r496, %rs181; + cvt.f32.bf16 %r497, %rs182; + cvt.f32.bf16 %r498, %rs183; + cvt.f32.bf16 %r499, %rs184; + cvt.f32.bf16 %r500, %rs185; + cvt.f32.bf16 %r501, %rs186; + cvt.f32.bf16 %r502, %rs187; + cvt.f32.bf16 %r503, %rs188; + cvt.f32.bf16 %r504, %rs189; + cvt.f32.bf16 %r505, %rs190; + cvt.f32.bf16 %r506, %rs191; + cvt.f32.bf16 %r507, %rs192; + cvt.f32.bf16 %r508, %rs193; + cvt.f32.bf16 %r509, %rs194; + cvt.f32.bf16 %r510, %rs195; + cvt.f32.bf16 %r511, %rs196; + cvt.f32.bf16 %r512, %rs197; + cvt.f32.bf16 %r513, %rs198; + cvt.f32.bf16 %r514, %rs199; + cvt.f32.bf16 %r515, %rs200; + cvt.f32.bf16 %r516, %rs201; + cvt.f32.bf16 %r517, %rs202; + cvt.f32.bf16 %r518, %rs203; + cvt.f32.bf16 %r519, %rs204; + cvt.f32.bf16 %r520, %rs205; + cvt.f32.bf16 %r521, %rs206; + cvt.f32.bf16 %r522, %rs207; + cvt.f32.bf16 %r523, %rs208; + cvt.f32.bf16 %r524, %rs209; + cvt.f32.bf16 %r525, %rs210; + cvt.f32.bf16 %r526, %rs211; + cvt.f32.bf16 %r527, %rs212; + cvt.f32.bf16 %r528, %rs213; + cvt.f32.bf16 %r529, %rs214; + cvt.f32.bf16 %r530, %rs215; + cvt.f32.bf16 %r531, %rs216; + cvt.f32.bf16 %r532, %rs217; + cvt.f32.bf16 %r533, %rs218; + cvt.f32.bf16 %r534, %rs219; + cvt.f32.bf16 %r535, %rs220; + cvt.f32.bf16 %r536, %rs221; + cvt.f32.bf16 %r537, %rs222; + cvt.f32.bf16 %r538, %rs223; + cvt.f32.bf16 %r539, %rs224; + cvt.f32.bf16 %r540, %rs225; + cvt.f32.bf16 %r541, %rs226; + cvt.f32.bf16 %r542, %rs227; + cvt.f32.bf16 %r543, %rs228; + cvt.f32.bf16 %r544, %rs229; + cvt.f32.bf16 %r545, %rs230; + cvt.f32.bf16 %r546, %rs231; + cvt.f32.bf16 %r547, %rs232; + cvt.f32.bf16 %r548, %rs233; + cvt.f32.bf16 %r549, %rs234; + cvt.f32.bf16 %r550, %rs235; + cvt.f32.bf16 %r551, %rs236; + cvt.f32.bf16 %r552, %rs237; + cvt.f32.bf16 %r553, %rs238; + cvt.f32.bf16 %r554, %rs239; + cvt.f32.bf16 %r555, %rs240; + cvt.f32.bf16 %r556, %rs241; + cvt.f32.bf16 %r557, %rs242; + cvt.f32.bf16 %r558, %rs243; + cvt.f32.bf16 %r559, %rs244; + cvt.f32.bf16 %r560, %rs245; + cvt.f32.bf16 %r561, %rs246; + cvt.f32.bf16 %r562, %rs247; + cvt.f32.bf16 %r563, %rs248; + cvt.f32.bf16 %r564, %rs249; + cvt.f32.bf16 %r565, %rs250; + cvt.f32.bf16 %r566, %rs251; + cvt.f32.bf16 %r567, %rs252; + cvt.f32.bf16 %r568, %rs253; + cvt.f32.bf16 %r569, %rs254; + cvt.f32.bf16 %r570, %rs255; + cvt.f32.bf16 %r571, %rs256; + cvt.f32.bf16 %r572, %rs257; + cvt.f32.bf16 %r573, %rs258; + cvt.f32.bf16 %r574, %rs259; + cvt.f32.bf16 %r575, %rs260; + cvt.f32.bf16 %r576, %rs261; + cvt.f32.bf16 %r577, %rs262; + cvt.f32.bf16 %r578, %rs263; + cvt.f32.bf16 %r579, %rs264; + cvt.f32.bf16 %r580, %rs265; + cvt.f32.bf16 %r581, %rs266; + cvt.f32.bf16 %r582, %rs267; + cvt.f32.bf16 %r583, %rs268; + cvt.f32.bf16 %r584, %rs269; + cvt.f32.bf16 %r585, %rs270; + cvt.f32.bf16 %r586, %rs271; + cvt.f32.bf16 %r587, %rs272; + cvt.f32.bf16 %r588, %rs273; + cvt.f32.bf16 %r589, %rs274; + cvt.f32.bf16 %r590, %rs275; + cvt.f32.bf16 %r591, %rs276; + cvt.f32.bf16 %r592, %rs277; + cvt.f32.bf16 %r593, %rs278; + cvt.f32.bf16 %r594, %rs279; + cvt.f32.bf16 %r595, %rs280; + cvt.f32.bf16 %r596, %rs281; + cvt.f32.bf16 %r597, %rs282; + cvt.f32.bf16 %r598, %rs283; + cvt.f32.bf16 %r599, %rs284; + cvt.f32.bf16 %r600, %rs285; + cvt.f32.bf16 %r601, %rs286; + cvt.f32.bf16 %r602, %rs287; + cvt.f32.bf16 %r603, %rs288; + cvt.f32.bf16 %r604, %rs289; + cvt.f32.bf16 %r605, %rs290; + cvt.f32.bf16 %r606, %rs291; + cvt.f32.bf16 %r607, %rs292; + cvt.f32.bf16 %r608, %rs293; + cvt.f32.bf16 %r609, %rs294; + cvt.f32.bf16 %r610, %rs295; + cvt.f32.bf16 %r611, %rs296; + cvt.f32.bf16 %r612, %rs297; + cvt.f32.bf16 %r613, %rs298; + cvt.f32.bf16 %r614, %rs299; + cvt.f32.bf16 %r615, %rs300; + cvt.f32.bf16 %r616, %rs301; + cvt.f32.bf16 %r617, %rs302; + cvt.f32.bf16 %r618, %rs303; + cvt.f32.bf16 %r619, %rs304; + cvt.f32.bf16 %r620, %rs305; + cvt.f32.bf16 %r621, %rs306; + cvt.f32.bf16 %r622, %rs307; + cvt.f32.bf16 %r623, %rs308; + cvt.f32.bf16 %r624, %rs309; + .loc 1 80 38 // l2norm.py:80:38 + mul.f32 %r625, %r370, %r370; + mul.f32 %r626, %r378, %r378; + mul.f32 %r627, %r386, %r386; + mul.f32 %r628, %r394, %r394; + mul.f32 %r629, %r402, %r402; + mul.f32 %r630, %r410, %r410; + mul.f32 %r631, %r418, %r418; + mul.f32 %r632, %r426, %r426; + mul.f32 %r633, %r434, %r434; + mul.f32 %r634, %r442, %r442; + mul.f32 %r635, %r450, %r450; + mul.f32 %r636, %r458, %r458; + mul.f32 %r637, %r466, %r466; + mul.f32 %r638, %r474, %r474; + mul.f32 %r639, %r482, %r482; + mul.f32 %r640, %r490, %r490; + mul.f32 %r641, %r498, %r498; + mul.f32 %r642, %r506, %r506; + mul.f32 %r643, %r514, %r514; + mul.f32 %r644, %r522, %r522; + mul.f32 %r645, %r530, %r530; + mul.f32 %r646, %r538, %r538; + mul.f32 %r647, %r546, %r546; + mul.f32 %r648, %r554, %r554; + mul.f32 %r649, %r562, %r562; + mul.f32 %r650, %r570, %r570; + mul.f32 %r651, %r578, %r578; + mul.f32 %r652, %r586, %r586; + mul.f32 %r653, %r594, %r594; + mul.f32 %r654, %r602, %r602; + mul.f32 %r655, %r610, %r610; + mul.f32 %r656, %r618, %r618; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + fma.rn.f32.bf16 %r657, %rs54, %rs54, %r625; + fma.rn.f32.bf16 %r658, %rs56, %rs56, %r657; + fma.rn.f32.bf16 %r659, %rs57, %rs57, %r658; + fma.rn.f32.bf16 %r660, %rs58, %rs58, %r659; + fma.rn.f32.bf16 %r661, %rs59, %rs59, %r660; + fma.rn.f32.bf16 %r662, %rs60, %rs60, %r661; + fma.rn.f32.bf16 %r663, %rs61, %rs61, %r662; + fma.rn.f32.bf16 %r664, %rs62, %rs62, %r626; + fma.rn.f32.bf16 %r665, %rs64, %rs64, %r664; + fma.rn.f32.bf16 %r666, %rs65, %rs65, %r665; + fma.rn.f32.bf16 %r667, %rs66, %rs66, %r666; + fma.rn.f32.bf16 %r668, %rs67, %rs67, %r667; + fma.rn.f32.bf16 %r669, %rs68, %rs68, %r668; + fma.rn.f32.bf16 %r670, %rs69, %rs69, %r669; + fma.rn.f32.bf16 %r671, %rs70, %rs70, %r627; + fma.rn.f32.bf16 %r672, %rs72, %rs72, %r671; + fma.rn.f32.bf16 %r673, %rs73, %rs73, %r672; + fma.rn.f32.bf16 %r674, %rs74, %rs74, %r673; + fma.rn.f32.bf16 %r675, %rs75, %rs75, %r674; + fma.rn.f32.bf16 %r676, %rs76, %rs76, %r675; + fma.rn.f32.bf16 %r677, %rs77, %rs77, %r676; + fma.rn.f32.bf16 %r678, %rs78, %rs78, %r628; + fma.rn.f32.bf16 %r679, %rs80, %rs80, %r678; + fma.rn.f32.bf16 %r680, %rs81, %rs81, %r679; + fma.rn.f32.bf16 %r681, %rs82, %rs82, %r680; + fma.rn.f32.bf16 %r682, %rs83, %rs83, %r681; + fma.rn.f32.bf16 %r683, %rs84, %rs84, %r682; + fma.rn.f32.bf16 %r684, %rs85, %rs85, %r683; + fma.rn.f32.bf16 %r685, %rs86, %rs86, %r629; + fma.rn.f32.bf16 %r686, %rs88, %rs88, %r685; + fma.rn.f32.bf16 %r687, %rs89, %rs89, %r686; + fma.rn.f32.bf16 %r688, %rs90, %rs90, %r687; + fma.rn.f32.bf16 %r689, %rs91, %rs91, %r688; + fma.rn.f32.bf16 %r690, %rs92, %rs92, %r689; + fma.rn.f32.bf16 %r691, %rs93, %rs93, %r690; + fma.rn.f32.bf16 %r692, %rs94, %rs94, %r630; + fma.rn.f32.bf16 %r693, %rs96, %rs96, %r692; + fma.rn.f32.bf16 %r694, %rs97, %rs97, %r693; + fma.rn.f32.bf16 %r695, %rs98, %rs98, %r694; + fma.rn.f32.bf16 %r696, %rs99, %rs99, %r695; + fma.rn.f32.bf16 %r697, %rs100, %rs100, %r696; + fma.rn.f32.bf16 %r698, %rs101, %rs101, %r697; + fma.rn.f32.bf16 %r699, %rs102, %rs102, %r631; + fma.rn.f32.bf16 %r700, %rs104, %rs104, %r699; + fma.rn.f32.bf16 %r701, %rs105, %rs105, %r700; + fma.rn.f32.bf16 %r702, %rs106, %rs106, %r701; + fma.rn.f32.bf16 %r703, %rs107, %rs107, %r702; + fma.rn.f32.bf16 %r704, %rs108, %rs108, %r703; + fma.rn.f32.bf16 %r705, %rs109, %rs109, %r704; + fma.rn.f32.bf16 %r706, %rs110, %rs110, %r632; + fma.rn.f32.bf16 %r707, %rs112, %rs112, %r706; + fma.rn.f32.bf16 %r708, %rs113, %rs113, %r707; + fma.rn.f32.bf16 %r709, %rs114, %rs114, %r708; + fma.rn.f32.bf16 %r710, %rs115, %rs115, %r709; + fma.rn.f32.bf16 %r711, %rs116, %rs116, %r710; + fma.rn.f32.bf16 %r712, %rs117, %rs117, %r711; + fma.rn.f32.bf16 %r713, %rs118, %rs118, %r633; + fma.rn.f32.bf16 %r714, %rs120, %rs120, %r713; + fma.rn.f32.bf16 %r715, %rs121, %rs121, %r714; + fma.rn.f32.bf16 %r716, %rs122, %rs122, %r715; + fma.rn.f32.bf16 %r717, %rs123, %rs123, %r716; + fma.rn.f32.bf16 %r718, %rs124, %rs124, %r717; + fma.rn.f32.bf16 %r719, %rs125, %rs125, %r718; + fma.rn.f32.bf16 %r720, %rs126, %rs126, %r634; + fma.rn.f32.bf16 %r721, %rs128, %rs128, %r720; + fma.rn.f32.bf16 %r722, %rs129, %rs129, %r721; + fma.rn.f32.bf16 %r723, %rs130, %rs130, %r722; + fma.rn.f32.bf16 %r724, %rs131, %rs131, %r723; + fma.rn.f32.bf16 %r725, %rs132, %rs132, %r724; + fma.rn.f32.bf16 %r726, %rs133, %rs133, %r725; + fma.rn.f32.bf16 %r727, %rs134, %rs134, %r635; + fma.rn.f32.bf16 %r728, %rs136, %rs136, %r727; + fma.rn.f32.bf16 %r729, %rs137, %rs137, %r728; + fma.rn.f32.bf16 %r730, %rs138, %rs138, %r729; + fma.rn.f32.bf16 %r731, %rs139, %rs139, %r730; + fma.rn.f32.bf16 %r732, %rs140, %rs140, %r731; + fma.rn.f32.bf16 %r733, %rs141, %rs141, %r732; + fma.rn.f32.bf16 %r734, %rs142, %rs142, %r636; + fma.rn.f32.bf16 %r735, %rs144, %rs144, %r734; + fma.rn.f32.bf16 %r736, %rs145, %rs145, %r735; + fma.rn.f32.bf16 %r737, %rs146, %rs146, %r736; + fma.rn.f32.bf16 %r738, %rs147, %rs147, %r737; + fma.rn.f32.bf16 %r739, %rs148, %rs148, %r738; + fma.rn.f32.bf16 %r740, %rs149, %rs149, %r739; + fma.rn.f32.bf16 %r741, %rs150, %rs150, %r637; + fma.rn.f32.bf16 %r742, %rs152, %rs152, %r741; + fma.rn.f32.bf16 %r743, %rs153, %rs153, %r742; + fma.rn.f32.bf16 %r744, %rs154, %rs154, %r743; + fma.rn.f32.bf16 %r745, %rs155, %rs155, %r744; + fma.rn.f32.bf16 %r746, %rs156, %rs156, %r745; + fma.rn.f32.bf16 %r747, %rs157, %rs157, %r746; + fma.rn.f32.bf16 %r748, %rs158, %rs158, %r638; + fma.rn.f32.bf16 %r749, %rs160, %rs160, %r748; + fma.rn.f32.bf16 %r750, %rs161, %rs161, %r749; + fma.rn.f32.bf16 %r751, %rs162, %rs162, %r750; + fma.rn.f32.bf16 %r752, %rs163, %rs163, %r751; + fma.rn.f32.bf16 %r753, %rs164, %rs164, %r752; + fma.rn.f32.bf16 %r754, %rs165, %rs165, %r753; + fma.rn.f32.bf16 %r755, %rs166, %rs166, %r639; + fma.rn.f32.bf16 %r756, %rs168, %rs168, %r755; + fma.rn.f32.bf16 %r757, %rs169, %rs169, %r756; + fma.rn.f32.bf16 %r758, %rs170, %rs170, %r757; + fma.rn.f32.bf16 %r759, %rs171, %rs171, %r758; + fma.rn.f32.bf16 %r760, %rs172, %rs172, %r759; + fma.rn.f32.bf16 %r761, %rs173, %rs173, %r760; + fma.rn.f32.bf16 %r762, %rs174, %rs174, %r640; + fma.rn.f32.bf16 %r763, %rs176, %rs176, %r762; + fma.rn.f32.bf16 %r764, %rs177, %rs177, %r763; + fma.rn.f32.bf16 %r765, %rs178, %rs178, %r764; + fma.rn.f32.bf16 %r766, %rs179, %rs179, %r765; + fma.rn.f32.bf16 %r767, %rs180, %rs180, %r766; + fma.rn.f32.bf16 %r768, %rs181, %rs181, %r767; + fma.rn.f32.bf16 %r769, %rs182, %rs182, %r641; + fma.rn.f32.bf16 %r770, %rs184, %rs184, %r769; + fma.rn.f32.bf16 %r771, %rs185, %rs185, %r770; + fma.rn.f32.bf16 %r772, %rs186, %rs186, %r771; + fma.rn.f32.bf16 %r773, %rs187, %rs187, %r772; + fma.rn.f32.bf16 %r774, %rs188, %rs188, %r773; + fma.rn.f32.bf16 %r775, %rs189, %rs189, %r774; + fma.rn.f32.bf16 %r776, %rs190, %rs190, %r642; + fma.rn.f32.bf16 %r777, %rs192, %rs192, %r776; + fma.rn.f32.bf16 %r778, %rs193, %rs193, %r777; + fma.rn.f32.bf16 %r779, %rs194, %rs194, %r778; + fma.rn.f32.bf16 %r780, %rs195, %rs195, %r779; + fma.rn.f32.bf16 %r781, %rs196, %rs196, %r780; + fma.rn.f32.bf16 %r782, %rs197, %rs197, %r781; + fma.rn.f32.bf16 %r783, %rs198, %rs198, %r643; + fma.rn.f32.bf16 %r784, %rs200, %rs200, %r783; + fma.rn.f32.bf16 %r785, %rs201, %rs201, %r784; + fma.rn.f32.bf16 %r786, %rs202, %rs202, %r785; + fma.rn.f32.bf16 %r787, %rs203, %rs203, %r786; + fma.rn.f32.bf16 %r788, %rs204, %rs204, %r787; + fma.rn.f32.bf16 %r789, %rs205, %rs205, %r788; + fma.rn.f32.bf16 %r790, %rs206, %rs206, %r644; + fma.rn.f32.bf16 %r791, %rs208, %rs208, %r790; + fma.rn.f32.bf16 %r792, %rs209, %rs209, %r791; + fma.rn.f32.bf16 %r793, %rs210, %rs210, %r792; + fma.rn.f32.bf16 %r794, %rs211, %rs211, %r793; + fma.rn.f32.bf16 %r795, %rs212, %rs212, %r794; + fma.rn.f32.bf16 %r796, %rs213, %rs213, %r795; + fma.rn.f32.bf16 %r797, %rs214, %rs214, %r645; + fma.rn.f32.bf16 %r798, %rs216, %rs216, %r797; + fma.rn.f32.bf16 %r799, %rs217, %rs217, %r798; + fma.rn.f32.bf16 %r800, %rs218, %rs218, %r799; + fma.rn.f32.bf16 %r801, %rs219, %rs219, %r800; + fma.rn.f32.bf16 %r802, %rs220, %rs220, %r801; + fma.rn.f32.bf16 %r803, %rs221, %rs221, %r802; + fma.rn.f32.bf16 %r804, %rs222, %rs222, %r646; + fma.rn.f32.bf16 %r805, %rs224, %rs224, %r804; + fma.rn.f32.bf16 %r806, %rs225, %rs225, %r805; + fma.rn.f32.bf16 %r807, %rs226, %rs226, %r806; + fma.rn.f32.bf16 %r808, %rs227, %rs227, %r807; + fma.rn.f32.bf16 %r809, %rs228, %rs228, %r808; + fma.rn.f32.bf16 %r810, %rs229, %rs229, %r809; + fma.rn.f32.bf16 %r811, %rs230, %rs230, %r647; + fma.rn.f32.bf16 %r812, %rs232, %rs232, %r811; + fma.rn.f32.bf16 %r813, %rs233, %rs233, %r812; + fma.rn.f32.bf16 %r814, %rs234, %rs234, %r813; + fma.rn.f32.bf16 %r815, %rs235, %rs235, %r814; + fma.rn.f32.bf16 %r816, %rs236, %rs236, %r815; + fma.rn.f32.bf16 %r817, %rs237, %rs237, %r816; + fma.rn.f32.bf16 %r818, %rs238, %rs238, %r648; + fma.rn.f32.bf16 %r819, %rs240, %rs240, %r818; + fma.rn.f32.bf16 %r820, %rs241, %rs241, %r819; + fma.rn.f32.bf16 %r821, %rs242, %rs242, %r820; + fma.rn.f32.bf16 %r822, %rs243, %rs243, %r821; + fma.rn.f32.bf16 %r823, %rs244, %rs244, %r822; + fma.rn.f32.bf16 %r824, %rs245, %rs245, %r823; + fma.rn.f32.bf16 %r825, %rs246, %rs246, %r649; + fma.rn.f32.bf16 %r826, %rs248, %rs248, %r825; + fma.rn.f32.bf16 %r827, %rs249, %rs249, %r826; + fma.rn.f32.bf16 %r828, %rs250, %rs250, %r827; + fma.rn.f32.bf16 %r829, %rs251, %rs251, %r828; + fma.rn.f32.bf16 %r830, %rs252, %rs252, %r829; + fma.rn.f32.bf16 %r831, %rs253, %rs253, %r830; + fma.rn.f32.bf16 %r832, %rs254, %rs254, %r650; + fma.rn.f32.bf16 %r833, %rs256, %rs256, %r832; + fma.rn.f32.bf16 %r834, %rs257, %rs257, %r833; + fma.rn.f32.bf16 %r835, %rs258, %rs258, %r834; + fma.rn.f32.bf16 %r836, %rs259, %rs259, %r835; + fma.rn.f32.bf16 %r837, %rs260, %rs260, %r836; + fma.rn.f32.bf16 %r838, %rs261, %rs261, %r837; + fma.rn.f32.bf16 %r839, %rs262, %rs262, %r651; + fma.rn.f32.bf16 %r840, %rs264, %rs264, %r839; + fma.rn.f32.bf16 %r841, %rs265, %rs265, %r840; + fma.rn.f32.bf16 %r842, %rs266, %rs266, %r841; + fma.rn.f32.bf16 %r843, %rs267, %rs267, %r842; + fma.rn.f32.bf16 %r844, %rs268, %rs268, %r843; + fma.rn.f32.bf16 %r845, %rs269, %rs269, %r844; + fma.rn.f32.bf16 %r846, %rs270, %rs270, %r652; + fma.rn.f32.bf16 %r847, %rs272, %rs272, %r846; + fma.rn.f32.bf16 %r848, %rs273, %rs273, %r847; + fma.rn.f32.bf16 %r849, %rs274, %rs274, %r848; + fma.rn.f32.bf16 %r850, %rs275, %rs275, %r849; + fma.rn.f32.bf16 %r851, %rs276, %rs276, %r850; + fma.rn.f32.bf16 %r852, %rs277, %rs277, %r851; + fma.rn.f32.bf16 %r853, %rs278, %rs278, %r653; + fma.rn.f32.bf16 %r854, %rs280, %rs280, %r853; + fma.rn.f32.bf16 %r855, %rs281, %rs281, %r854; + fma.rn.f32.bf16 %r856, %rs282, %rs282, %r855; + fma.rn.f32.bf16 %r857, %rs283, %rs283, %r856; + fma.rn.f32.bf16 %r858, %rs284, %rs284, %r857; + fma.rn.f32.bf16 %r859, %rs285, %rs285, %r858; + fma.rn.f32.bf16 %r860, %rs286, %rs286, %r654; + fma.rn.f32.bf16 %r861, %rs288, %rs288, %r860; + fma.rn.f32.bf16 %r862, %rs289, %rs289, %r861; + fma.rn.f32.bf16 %r863, %rs290, %rs290, %r862; + fma.rn.f32.bf16 %r864, %rs291, %rs291, %r863; + fma.rn.f32.bf16 %r865, %rs292, %rs292, %r864; + fma.rn.f32.bf16 %r866, %rs293, %rs293, %r865; + fma.rn.f32.bf16 %r867, %rs294, %rs294, %r655; + fma.rn.f32.bf16 %r868, %rs296, %rs296, %r867; + fma.rn.f32.bf16 %r869, %rs297, %rs297, %r868; + fma.rn.f32.bf16 %r870, %rs298, %rs298, %r869; + fma.rn.f32.bf16 %r871, %rs299, %rs299, %r870; + fma.rn.f32.bf16 %r872, %rs300, %rs300, %r871; + fma.rn.f32.bf16 %r873, %rs301, %rs301, %r872; + fma.rn.f32.bf16 %r874, %rs302, %rs302, %r656; + fma.rn.f32.bf16 %r875, %rs304, %rs304, %r874; + fma.rn.f32.bf16 %r876, %rs305, %rs305, %r875; + fma.rn.f32.bf16 %r877, %rs306, %rs306, %r876; + fma.rn.f32.bf16 %r878, %rs307, %rs307, %r877; + fma.rn.f32.bf16 %r879, %rs308, %rs308, %r878; + fma.rn.f32.bf16 %r880, %rs309, %rs309, %r879; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r881, %r663, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r882, %r663, %r881; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r883, %r882, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r884, %r882, %r883; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r885, %r884, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r886, %r884, %r885; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r887, %r886, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r888, %r886, %r887; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r889, %r670, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r890, %r670, %r889; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r891, %r890, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r892, %r890, %r891; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r893, %r892, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r894, %r892, %r893; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r895, %r894, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r896, %r894, %r895; +$L__tmp18: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r897, %r677, 8, 31, -1; +$L__tmp19: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r898, %r677, %r897; +$L__tmp20: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r899, %r898, 4, 31, -1; +$L__tmp21: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r900, %r898, %r899; +$L__tmp22: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r901, %r900, 2, 31, -1; +$L__tmp23: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r902, %r900, %r901; +$L__tmp24: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r903, %r902, 1, 31, -1; +$L__tmp25: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r904, %r902, %r903; +$L__tmp26: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r905, %r684, 8, 31, -1; +$L__tmp27: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r906, %r684, %r905; +$L__tmp28: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r907, %r906, 4, 31, -1; +$L__tmp29: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r908, %r906, %r907; +$L__tmp30: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r909, %r908, 2, 31, -1; +$L__tmp31: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r910, %r908, %r909; +$L__tmp32: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r911, %r910, 1, 31, -1; +$L__tmp33: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r912, %r910, %r911; +$L__tmp34: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r913, %r691, 8, 31, -1; +$L__tmp35: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r914, %r691, %r913; +$L__tmp36: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r915, %r914, 4, 31, -1; +$L__tmp37: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r916, %r914, %r915; +$L__tmp38: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r917, %r916, 2, 31, -1; +$L__tmp39: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r918, %r916, %r917; +$L__tmp40: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r919, %r918, 1, 31, -1; +$L__tmp41: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r920, %r918, %r919; +$L__tmp42: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r921, %r698, 8, 31, -1; +$L__tmp43: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r922, %r698, %r921; +$L__tmp44: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r923, %r922, 4, 31, -1; +$L__tmp45: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r924, %r922, %r923; +$L__tmp46: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r925, %r924, 2, 31, -1; +$L__tmp47: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r926, %r924, %r925; +$L__tmp48: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r927, %r926, 1, 31, -1; +$L__tmp49: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r928, %r926, %r927; +$L__tmp50: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r929, %r705, 8, 31, -1; +$L__tmp51: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r930, %r705, %r929; +$L__tmp52: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r931, %r930, 4, 31, -1; +$L__tmp53: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r932, %r930, %r931; +$L__tmp54: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r933, %r932, 2, 31, -1; +$L__tmp55: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r934, %r932, %r933; +$L__tmp56: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r935, %r934, 1, 31, -1; +$L__tmp57: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r936, %r934, %r935; +$L__tmp58: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r937, %r712, 8, 31, -1; +$L__tmp59: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r938, %r712, %r937; +$L__tmp60: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r939, %r938, 4, 31, -1; +$L__tmp61: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r940, %r938, %r939; +$L__tmp62: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r941, %r940, 2, 31, -1; +$L__tmp63: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r942, %r940, %r941; +$L__tmp64: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r943, %r942, 1, 31, -1; +$L__tmp65: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r944, %r942, %r943; +$L__tmp66: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r945, %r719, 8, 31, -1; +$L__tmp67: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r946, %r719, %r945; +$L__tmp68: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r947, %r946, 4, 31, -1; +$L__tmp69: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r948, %r946, %r947; +$L__tmp70: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r949, %r948, 2, 31, -1; +$L__tmp71: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r950, %r948, %r949; +$L__tmp72: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r951, %r950, 1, 31, -1; +$L__tmp73: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r952, %r950, %r951; +$L__tmp74: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r953, %r726, 8, 31, -1; +$L__tmp75: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r954, %r726, %r953; +$L__tmp76: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r955, %r954, 4, 31, -1; +$L__tmp77: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r956, %r954, %r955; +$L__tmp78: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r957, %r956, 2, 31, -1; +$L__tmp79: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r958, %r956, %r957; +$L__tmp80: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r959, %r958, 1, 31, -1; +$L__tmp81: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r960, %r958, %r959; +$L__tmp82: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r961, %r733, 8, 31, -1; +$L__tmp83: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r962, %r733, %r961; +$L__tmp84: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r963, %r962, 4, 31, -1; +$L__tmp85: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r964, %r962, %r963; +$L__tmp86: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r965, %r964, 2, 31, -1; +$L__tmp87: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r966, %r964, %r965; +$L__tmp88: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r967, %r966, 1, 31, -1; +$L__tmp89: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r968, %r966, %r967; +$L__tmp90: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r969, %r740, 8, 31, -1; +$L__tmp91: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r970, %r740, %r969; +$L__tmp92: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r971, %r970, 4, 31, -1; +$L__tmp93: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r972, %r970, %r971; +$L__tmp94: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r973, %r972, 2, 31, -1; +$L__tmp95: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r974, %r972, %r973; +$L__tmp96: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r975, %r974, 1, 31, -1; +$L__tmp97: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r976, %r974, %r975; +$L__tmp98: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r977, %r747, 8, 31, -1; +$L__tmp99: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r978, %r747, %r977; +$L__tmp100: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r979, %r978, 4, 31, -1; +$L__tmp101: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r980, %r978, %r979; +$L__tmp102: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r981, %r980, 2, 31, -1; +$L__tmp103: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r982, %r980, %r981; +$L__tmp104: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r983, %r982, 1, 31, -1; +$L__tmp105: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r984, %r982, %r983; +$L__tmp106: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r985, %r754, 8, 31, -1; +$L__tmp107: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r986, %r754, %r985; +$L__tmp108: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r987, %r986, 4, 31, -1; +$L__tmp109: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r988, %r986, %r987; +$L__tmp110: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r989, %r988, 2, 31, -1; +$L__tmp111: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r990, %r988, %r989; +$L__tmp112: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r991, %r990, 1, 31, -1; +$L__tmp113: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r992, %r990, %r991; +$L__tmp114: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r993, %r761, 8, 31, -1; +$L__tmp115: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r994, %r761, %r993; +$L__tmp116: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r995, %r994, 4, 31, -1; +$L__tmp117: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r996, %r994, %r995; +$L__tmp118: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r997, %r996, 2, 31, -1; +$L__tmp119: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r998, %r996, %r997; +$L__tmp120: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r999, %r998, 1, 31, -1; +$L__tmp121: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1000, %r998, %r999; +$L__tmp122: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1001, %r768, 8, 31, -1; +$L__tmp123: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1002, %r768, %r1001; +$L__tmp124: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1003, %r1002, 4, 31, -1; +$L__tmp125: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1004, %r1002, %r1003; +$L__tmp126: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1005, %r1004, 2, 31, -1; +$L__tmp127: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1006, %r1004, %r1005; +$L__tmp128: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1007, %r1006, 1, 31, -1; +$L__tmp129: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1008, %r1006, %r1007; +$L__tmp130: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1009, %r775, 8, 31, -1; +$L__tmp131: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1010, %r775, %r1009; +$L__tmp132: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1011, %r1010, 4, 31, -1; +$L__tmp133: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1012, %r1010, %r1011; +$L__tmp134: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1013, %r1012, 2, 31, -1; +$L__tmp135: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1014, %r1012, %r1013; +$L__tmp136: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1015, %r1014, 1, 31, -1; +$L__tmp137: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1016, %r1014, %r1015; +$L__tmp138: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1017, %r782, 8, 31, -1; +$L__tmp139: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1018, %r782, %r1017; +$L__tmp140: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1019, %r1018, 4, 31, -1; +$L__tmp141: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1020, %r1018, %r1019; +$L__tmp142: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1021, %r1020, 2, 31, -1; +$L__tmp143: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1022, %r1020, %r1021; +$L__tmp144: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1023, %r1022, 1, 31, -1; +$L__tmp145: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1024, %r1022, %r1023; +$L__tmp146: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1025, %r789, 8, 31, -1; +$L__tmp147: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1026, %r789, %r1025; +$L__tmp148: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1027, %r1026, 4, 31, -1; +$L__tmp149: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1028, %r1026, %r1027; +$L__tmp150: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1029, %r1028, 2, 31, -1; +$L__tmp151: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1030, %r1028, %r1029; +$L__tmp152: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1031, %r1030, 1, 31, -1; +$L__tmp153: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1032, %r1030, %r1031; +$L__tmp154: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1033, %r796, 8, 31, -1; +$L__tmp155: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1034, %r796, %r1033; +$L__tmp156: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1035, %r1034, 4, 31, -1; +$L__tmp157: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1036, %r1034, %r1035; +$L__tmp158: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1037, %r1036, 2, 31, -1; +$L__tmp159: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1038, %r1036, %r1037; +$L__tmp160: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1039, %r1038, 1, 31, -1; +$L__tmp161: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1040, %r1038, %r1039; +$L__tmp162: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1041, %r803, 8, 31, -1; +$L__tmp163: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1042, %r803, %r1041; +$L__tmp164: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1043, %r1042, 4, 31, -1; +$L__tmp165: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1044, %r1042, %r1043; +$L__tmp166: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1045, %r1044, 2, 31, -1; +$L__tmp167: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1046, %r1044, %r1045; +$L__tmp168: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1047, %r1046, 1, 31, -1; +$L__tmp169: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1048, %r1046, %r1047; +$L__tmp170: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1049, %r810, 8, 31, -1; +$L__tmp171: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1050, %r810, %r1049; +$L__tmp172: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1051, %r1050, 4, 31, -1; +$L__tmp173: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1052, %r1050, %r1051; +$L__tmp174: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1053, %r1052, 2, 31, -1; +$L__tmp175: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1054, %r1052, %r1053; +$L__tmp176: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1055, %r1054, 1, 31, -1; +$L__tmp177: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1056, %r1054, %r1055; +$L__tmp178: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1057, %r817, 8, 31, -1; +$L__tmp179: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1058, %r817, %r1057; +$L__tmp180: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1059, %r1058, 4, 31, -1; +$L__tmp181: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1060, %r1058, %r1059; +$L__tmp182: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1061, %r1060, 2, 31, -1; +$L__tmp183: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1062, %r1060, %r1061; +$L__tmp184: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1063, %r1062, 1, 31, -1; +$L__tmp185: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1064, %r1062, %r1063; +$L__tmp186: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1065, %r824, 8, 31, -1; +$L__tmp187: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1066, %r824, %r1065; +$L__tmp188: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1067, %r1066, 4, 31, -1; +$L__tmp189: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1068, %r1066, %r1067; +$L__tmp190: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1069, %r1068, 2, 31, -1; +$L__tmp191: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1070, %r1068, %r1069; +$L__tmp192: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1071, %r1070, 1, 31, -1; +$L__tmp193: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1072, %r1070, %r1071; +$L__tmp194: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1073, %r831, 8, 31, -1; +$L__tmp195: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1074, %r831, %r1073; +$L__tmp196: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1075, %r1074, 4, 31, -1; +$L__tmp197: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1076, %r1074, %r1075; +$L__tmp198: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1077, %r1076, 2, 31, -1; +$L__tmp199: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1078, %r1076, %r1077; +$L__tmp200: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1079, %r1078, 1, 31, -1; +$L__tmp201: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1080, %r1078, %r1079; +$L__tmp202: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1081, %r838, 8, 31, -1; +$L__tmp203: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1082, %r838, %r1081; +$L__tmp204: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1083, %r1082, 4, 31, -1; +$L__tmp205: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1084, %r1082, %r1083; +$L__tmp206: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1085, %r1084, 2, 31, -1; +$L__tmp207: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1086, %r1084, %r1085; +$L__tmp208: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1087, %r1086, 1, 31, -1; +$L__tmp209: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1088, %r1086, %r1087; +$L__tmp210: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1089, %r845, 8, 31, -1; +$L__tmp211: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1090, %r845, %r1089; +$L__tmp212: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1091, %r1090, 4, 31, -1; +$L__tmp213: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1092, %r1090, %r1091; +$L__tmp214: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1093, %r1092, 2, 31, -1; +$L__tmp215: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1094, %r1092, %r1093; +$L__tmp216: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1095, %r1094, 1, 31, -1; +$L__tmp217: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1096, %r1094, %r1095; +$L__tmp218: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1097, %r852, 8, 31, -1; +$L__tmp219: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1098, %r852, %r1097; +$L__tmp220: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1099, %r1098, 4, 31, -1; +$L__tmp221: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1100, %r1098, %r1099; +$L__tmp222: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1101, %r1100, 2, 31, -1; +$L__tmp223: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1102, %r1100, %r1101; +$L__tmp224: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1103, %r1102, 1, 31, -1; +$L__tmp225: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1104, %r1102, %r1103; +$L__tmp226: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1105, %r859, 8, 31, -1; +$L__tmp227: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1106, %r859, %r1105; +$L__tmp228: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1107, %r1106, 4, 31, -1; +$L__tmp229: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1108, %r1106, %r1107; +$L__tmp230: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1109, %r1108, 2, 31, -1; +$L__tmp231: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1110, %r1108, %r1109; +$L__tmp232: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1111, %r1110, 1, 31, -1; +$L__tmp233: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1112, %r1110, %r1111; +$L__tmp234: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1113, %r866, 8, 31, -1; +$L__tmp235: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1114, %r866, %r1113; +$L__tmp236: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1115, %r1114, 4, 31, -1; +$L__tmp237: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1116, %r1114, %r1115; +$L__tmp238: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1117, %r1116, 2, 31, -1; +$L__tmp239: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1118, %r1116, %r1117; +$L__tmp240: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1119, %r1118, 1, 31, -1; +$L__tmp241: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1120, %r1118, %r1119; +$L__tmp242: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1121, %r873, 8, 31, -1; +$L__tmp243: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1122, %r873, %r1121; +$L__tmp244: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1123, %r1122, 4, 31, -1; +$L__tmp245: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1124, %r1122, %r1123; +$L__tmp246: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1125, %r1124, 2, 31, -1; +$L__tmp247: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1126, %r1124, %r1125; +$L__tmp248: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1127, %r1126, 1, 31, -1; +$L__tmp249: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1128, %r1126, %r1127; +$L__tmp250: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1129, %r880, 8, 31, -1; +$L__tmp251: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1130, %r880, %r1129; +$L__tmp252: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1131, %r1130, 4, 31, -1; +$L__tmp253: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1132, %r1130, %r1131; +$L__tmp254: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1133, %r1132, 2, 31, -1; +$L__tmp255: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1134, %r1132, %r1133; +$L__tmp256: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1135, %r1134, 1, 31, -1; +$L__tmp257: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1136, %r1134, %r1135; +$L__tmp258: + .loc 1 80 48 // l2norm.py:80:48 + add.f32 %r1137, %r301, %r888; + add.f32 %r1138, %r301, %r896; + add.f32 %r1139, %r301, %r904; + add.f32 %r1140, %r301, %r912; + add.f32 %r1141, %r301, %r920; + add.f32 %r1142, %r301, %r928; + add.f32 %r1143, %r301, %r936; + add.f32 %r1144, %r301, %r944; + add.f32 %r1145, %r301, %r952; + add.f32 %r1146, %r301, %r960; + add.f32 %r1147, %r301, %r968; + add.f32 %r1148, %r301, %r976; + add.f32 %r1149, %r301, %r984; + add.f32 %r1150, %r301, %r992; + add.f32 %r1151, %r301, %r1000; + add.f32 %r1152, %r301, %r1008; + add.f32 %r1153, %r301, %r1016; + add.f32 %r1154, %r301, %r1024; + add.f32 %r1155, %r301, %r1032; + add.f32 %r1156, %r301, %r1040; + add.f32 %r1157, %r301, %r1048; + add.f32 %r1158, %r301, %r1056; + add.f32 %r1159, %r301, %r1064; + add.f32 %r1160, %r301, %r1072; + add.f32 %r1161, %r301, %r1080; + add.f32 %r1162, %r301, %r1088; + add.f32 %r1163, %r301, %r1096; + add.f32 %r1164, %r301, %r1104; + add.f32 %r1165, %r301, %r1112; + add.f32 %r1166, %r301, %r1120; + add.f32 %r1167, %r301, %r1128; + add.f32 %r1168, %r301, %r1136; + .loc 1 80 25 // l2norm.py:80:25 + sqrt.approx.ftz.f32 %r1169, %r1137; + sqrt.approx.ftz.f32 %r1170, %r1138; + sqrt.approx.ftz.f32 %r1171, %r1139; + sqrt.approx.ftz.f32 %r1172, %r1140; + sqrt.approx.ftz.f32 %r1173, %r1141; + sqrt.approx.ftz.f32 %r1174, %r1142; + sqrt.approx.ftz.f32 %r1175, %r1143; + sqrt.approx.ftz.f32 %r1176, %r1144; + sqrt.approx.ftz.f32 %r1177, %r1145; + sqrt.approx.ftz.f32 %r1178, %r1146; + sqrt.approx.ftz.f32 %r1179, %r1147; + sqrt.approx.ftz.f32 %r1180, %r1148; + sqrt.approx.ftz.f32 %r1181, %r1149; + sqrt.approx.ftz.f32 %r1182, %r1150; + sqrt.approx.ftz.f32 %r1183, %r1151; + sqrt.approx.ftz.f32 %r1184, %r1152; + sqrt.approx.ftz.f32 %r1185, %r1153; + sqrt.approx.ftz.f32 %r1186, %r1154; + sqrt.approx.ftz.f32 %r1187, %r1155; + sqrt.approx.ftz.f32 %r1188, %r1156; + sqrt.approx.ftz.f32 %r1189, %r1157; + sqrt.approx.ftz.f32 %r1190, %r1158; + sqrt.approx.ftz.f32 %r1191, %r1159; + sqrt.approx.ftz.f32 %r1192, %r1160; + sqrt.approx.ftz.f32 %r1193, %r1161; + sqrt.approx.ftz.f32 %r1194, %r1162; + sqrt.approx.ftz.f32 %r1195, %r1163; + sqrt.approx.ftz.f32 %r1196, %r1164; + sqrt.approx.ftz.f32 %r1197, %r1165; + sqrt.approx.ftz.f32 %r1198, %r1166; + sqrt.approx.ftz.f32 %r1199, %r1167; + sqrt.approx.ftz.f32 %r1200, %r1168; + .loc 1 71 34 // l2norm.py:71:34 + shl.b32 %r1201, %r302, 1; + and.b32 %r1202, %r1201, 126; + or.b32 %r1203, %r1202, 1; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd201, %r1203; + or.b64 %rd202, %rd70, %rd201; + cvt.u64.u32 %rd203, %r1202; + or.b64 %rd204, %rd70, %rd203; + .loc 1 77 20 // l2norm.py:77:20 + shl.b64 %rd205, %rd204, 2; + add.s64 %rd65, %rd69, %rd205; + add.s64 %rd66, %rd65, 4; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd206, %rd68, %rd198; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd64, %rd206, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd207, %rd68, %rd196; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd63, %rd207, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd208, %rd68, %rd194; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd62, %rd208, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd209, %rd68, %rd192; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd61, %rd209, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd210, %rd68, %rd190; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd60, %rd210, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd211, %rd68, %rd188; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd59, %rd211, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd212, %rd68, %rd186; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd58, %rd212, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd213, %rd68, %rd184; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd57, %rd213, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd214, %rd68, %rd182; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd56, %rd214, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd215, %rd68, %rd180; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd55, %rd215, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd216, %rd68, %rd178; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd54, %rd216, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd217, %rd68, %rd176; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd53, %rd217, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd218, %rd68, %rd174; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd52, %rd218, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd219, %rd68, %rd172; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd51, %rd219, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd220, %rd68, %rd170; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd50, %rd220, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd221, %rd68, %rd168; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd49, %rd221, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd222, %rd68, %rd166; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd48, %rd222, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd223, %rd68, %rd164; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd47, %rd223, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd224, %rd68, %rd162; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd46, %rd224, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd225, %rd68, %rd160; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd45, %rd225, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd226, %rd68, %rd158; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd44, %rd226, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd227, %rd68, %rd156; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd43, %rd227, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd228, %rd68, %rd154; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd42, %rd228, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd229, %rd68, %rd152; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd41, %rd229, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd230, %rd68, %rd150; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd40, %rd230, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd231, %rd68, %rd148; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd39, %rd231, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd232, %rd68, %rd146; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd38, %rd232, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd233, %rd68, %rd144; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd37, %rd233, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd234, %rd68, %rd142; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd36, %rd234, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd235, %rd68, %rd140; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd35, %rd235, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd236, %rd68, %rd138; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd34, %rd236, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd237, %rd68, %rd136; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd33, %rd237, %rd200; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p34, %rd202, %rd71; + setp.lt.s64 %p33, %rd204, %rd71; + mov.b32 %r1204, 0f3F800000; + .loc 1 80 17 // l2norm.py:80:17 + div.full.f32 %r259, %r1204, %r1169; + div.full.f32 %r260, %r1204, %r1170; + div.full.f32 %r261, %r1204, %r1171; + div.full.f32 %r262, %r1204, %r1172; + div.full.f32 %r264, %r1204, %r1173; + div.full.f32 %r265, %r1204, %r1174; + div.full.f32 %r266, %r1204, %r1175; + div.full.f32 %r267, %r1204, %r1176; + div.full.f32 %r269, %r1204, %r1177; + div.full.f32 %r270, %r1204, %r1178; + div.full.f32 %r271, %r1204, %r1179; + div.full.f32 %r272, %r1204, %r1180; + div.full.f32 %r274, %r1204, %r1181; + div.full.f32 %r275, %r1204, %r1182; + div.full.f32 %r276, %r1204, %r1183; + div.full.f32 %r277, %r1204, %r1184; + div.full.f32 %r279, %r1204, %r1185; + div.full.f32 %r280, %r1204, %r1186; + div.full.f32 %r281, %r1204, %r1187; + div.full.f32 %r282, %r1204, %r1188; + div.full.f32 %r284, %r1204, %r1189; + div.full.f32 %r285, %r1204, %r1190; + div.full.f32 %r286, %r1204, %r1191; + div.full.f32 %r287, %r1204, %r1192; + div.full.f32 %r289, %r1204, %r1193; + div.full.f32 %r290, %r1204, %r1194; + div.full.f32 %r291, %r1204, %r1195; + div.full.f32 %r292, %r1204, %r1196; + div.full.f32 %r294, %r1204, %r1197; + div.full.f32 %r295, %r1204, %r1198; + div.full.f32 %r296, %r1204, %r1199; + div.full.f32 %r297, %r1204, %r1200; + .loc 1 81 16 // l2norm.py:81:16 + mul.f32 %r1205, %r259, %r369; + mul.f32 %r1206, %r259, %r370; + mul.f32 %r1207, %r259, %r371; + mul.f32 %r1208, %r259, %r372; + mul.f32 %r1209, %r259, %r373; + mul.f32 %r1210, %r259, %r374; + mul.f32 %r1211, %r259, %r375; + mul.f32 %r1212, %r259, %r376; + mul.f32 %r1213, %r260, %r377; + mul.f32 %r1214, %r260, %r378; + mul.f32 %r1215, %r260, %r379; + mul.f32 %r1216, %r260, %r380; + mul.f32 %r1217, %r260, %r381; + mul.f32 %r1218, %r260, %r382; + mul.f32 %r1219, %r260, %r383; + mul.f32 %r1220, %r260, %r384; + mul.f32 %r1221, %r261, %r385; + mul.f32 %r1222, %r261, %r386; + mul.f32 %r1223, %r261, %r387; + mul.f32 %r1224, %r261, %r388; + mul.f32 %r1225, %r261, %r389; + mul.f32 %r1226, %r261, %r390; + mul.f32 %r1227, %r261, %r391; + mul.f32 %r1228, %r261, %r392; + mul.f32 %r1229, %r262, %r393; + mul.f32 %r1230, %r262, %r394; + mul.f32 %r1231, %r262, %r395; + mul.f32 %r1232, %r262, %r396; + mul.f32 %r1233, %r262, %r397; + mul.f32 %r1234, %r262, %r398; + mul.f32 %r1235, %r262, %r399; + mul.f32 %r1236, %r262, %r400; + mul.f32 %r1237, %r264, %r401; + mul.f32 %r1238, %r264, %r402; + mul.f32 %r1239, %r264, %r403; + mul.f32 %r1240, %r264, %r404; + mul.f32 %r1241, %r264, %r405; + mul.f32 %r1242, %r264, %r406; + mul.f32 %r1243, %r264, %r407; + mul.f32 %r1244, %r264, %r408; + mul.f32 %r1245, %r265, %r409; + mul.f32 %r1246, %r265, %r410; + mul.f32 %r1247, %r265, %r411; + mul.f32 %r1248, %r265, %r412; + mul.f32 %r1249, %r265, %r413; + mul.f32 %r1250, %r265, %r414; + mul.f32 %r1251, %r265, %r415; + mul.f32 %r1252, %r265, %r416; + mul.f32 %r1253, %r266, %r417; + mul.f32 %r1254, %r266, %r418; + mul.f32 %r1255, %r266, %r419; + mul.f32 %r1256, %r266, %r420; + mul.f32 %r1257, %r266, %r421; + mul.f32 %r1258, %r266, %r422; + mul.f32 %r1259, %r266, %r423; + mul.f32 %r1260, %r266, %r424; + mul.f32 %r1261, %r267, %r425; + mul.f32 %r1262, %r267, %r426; + mul.f32 %r1263, %r267, %r427; + mul.f32 %r1264, %r267, %r428; + mul.f32 %r1265, %r267, %r429; + mul.f32 %r1266, %r267, %r430; + mul.f32 %r1267, %r267, %r431; + mul.f32 %r1268, %r267, %r432; + mul.f32 %r1269, %r269, %r433; + mul.f32 %r1270, %r269, %r434; + mul.f32 %r1271, %r269, %r435; + mul.f32 %r1272, %r269, %r436; + mul.f32 %r1273, %r269, %r437; + mul.f32 %r1274, %r269, %r438; + mul.f32 %r1275, %r269, %r439; + mul.f32 %r1276, %r269, %r440; + mul.f32 %r1277, %r270, %r441; + mul.f32 %r1278, %r270, %r442; + mul.f32 %r1279, %r270, %r443; + mul.f32 %r1280, %r270, %r444; + mul.f32 %r1281, %r270, %r445; + mul.f32 %r1282, %r270, %r446; + mul.f32 %r1283, %r270, %r447; + mul.f32 %r1284, %r270, %r448; + mul.f32 %r1285, %r271, %r449; + mul.f32 %r1286, %r271, %r450; + mul.f32 %r1287, %r271, %r451; + mul.f32 %r1288, %r271, %r452; + mul.f32 %r1289, %r271, %r453; + mul.f32 %r1290, %r271, %r454; + mul.f32 %r1291, %r271, %r455; + mul.f32 %r1292, %r271, %r456; + mul.f32 %r1293, %r272, %r457; + mul.f32 %r1294, %r272, %r458; + mul.f32 %r1295, %r272, %r459; + mul.f32 %r1296, %r272, %r460; + mul.f32 %r1297, %r272, %r461; + mul.f32 %r1298, %r272, %r462; + mul.f32 %r1299, %r272, %r463; + mul.f32 %r1300, %r272, %r464; + mul.f32 %r1301, %r274, %r465; + mul.f32 %r1302, %r274, %r466; + mul.f32 %r1303, %r274, %r467; + mul.f32 %r1304, %r274, %r468; + mul.f32 %r1305, %r274, %r469; + mul.f32 %r1306, %r274, %r470; + mul.f32 %r1307, %r274, %r471; + mul.f32 %r1308, %r274, %r472; + mul.f32 %r1309, %r275, %r473; + mul.f32 %r1310, %r275, %r474; + mul.f32 %r1311, %r275, %r475; + mul.f32 %r1312, %r275, %r476; + mul.f32 %r1313, %r275, %r477; + mul.f32 %r1314, %r275, %r478; + mul.f32 %r1315, %r275, %r479; + mul.f32 %r1316, %r275, %r480; + mul.f32 %r1317, %r276, %r481; + mul.f32 %r1318, %r276, %r482; + mul.f32 %r1319, %r276, %r483; + mul.f32 %r1320, %r276, %r484; + mul.f32 %r1321, %r276, %r485; + mul.f32 %r1322, %r276, %r486; + mul.f32 %r1323, %r276, %r487; + mul.f32 %r1324, %r276, %r488; + mul.f32 %r1325, %r277, %r489; + mul.f32 %r1326, %r277, %r490; + mul.f32 %r1327, %r277, %r491; + mul.f32 %r1328, %r277, %r492; + mul.f32 %r1329, %r277, %r493; + mul.f32 %r1330, %r277, %r494; + mul.f32 %r1331, %r277, %r495; + mul.f32 %r1332, %r277, %r496; + mul.f32 %r1333, %r279, %r497; + mul.f32 %r1334, %r279, %r498; + mul.f32 %r1335, %r279, %r499; + mul.f32 %r1336, %r279, %r500; + mul.f32 %r1337, %r279, %r501; + mul.f32 %r1338, %r279, %r502; + mul.f32 %r1339, %r279, %r503; + mul.f32 %r1340, %r279, %r504; + mul.f32 %r1341, %r280, %r505; + mul.f32 %r1342, %r280, %r506; + mul.f32 %r1343, %r280, %r507; + mul.f32 %r1344, %r280, %r508; + mul.f32 %r1345, %r280, %r509; + mul.f32 %r1346, %r280, %r510; + mul.f32 %r1347, %r280, %r511; + mul.f32 %r1348, %r280, %r512; + mul.f32 %r1349, %r281, %r513; + mul.f32 %r1350, %r281, %r514; + mul.f32 %r1351, %r281, %r515; + mul.f32 %r1352, %r281, %r516; + mul.f32 %r1353, %r281, %r517; + mul.f32 %r1354, %r281, %r518; + mul.f32 %r1355, %r281, %r519; + mul.f32 %r1356, %r281, %r520; + mul.f32 %r1357, %r282, %r521; + mul.f32 %r1358, %r282, %r522; + mul.f32 %r1359, %r282, %r523; + mul.f32 %r1360, %r282, %r524; + mul.f32 %r1361, %r282, %r525; + mul.f32 %r1362, %r282, %r526; + mul.f32 %r1363, %r282, %r527; + mul.f32 %r1364, %r282, %r528; + mul.f32 %r1365, %r284, %r529; + mul.f32 %r1366, %r284, %r530; + mul.f32 %r1367, %r284, %r531; + mul.f32 %r1368, %r284, %r532; + mul.f32 %r1369, %r284, %r533; + mul.f32 %r1370, %r284, %r534; + mul.f32 %r1371, %r284, %r535; + mul.f32 %r1372, %r284, %r536; + mul.f32 %r1373, %r285, %r537; + mul.f32 %r1374, %r285, %r538; + mul.f32 %r1375, %r285, %r539; + mul.f32 %r1376, %r285, %r540; + mul.f32 %r1377, %r285, %r541; + mul.f32 %r1378, %r285, %r542; + mul.f32 %r1379, %r285, %r543; + mul.f32 %r1380, %r285, %r544; + mul.f32 %r1381, %r286, %r545; + mul.f32 %r1382, %r286, %r546; + mul.f32 %r1383, %r286, %r547; + mul.f32 %r1384, %r286, %r548; + mul.f32 %r1385, %r286, %r549; + mul.f32 %r1386, %r286, %r550; + mul.f32 %r1387, %r286, %r551; + mul.f32 %r1388, %r286, %r552; + mul.f32 %r1389, %r287, %r553; + mul.f32 %r1390, %r287, %r554; + mul.f32 %r1391, %r287, %r555; + mul.f32 %r1392, %r287, %r556; + mul.f32 %r1393, %r287, %r557; + mul.f32 %r1394, %r287, %r558; + mul.f32 %r1395, %r287, %r559; + mul.f32 %r1396, %r287, %r560; + mul.f32 %r1397, %r289, %r561; + mul.f32 %r1398, %r289, %r562; + mul.f32 %r1399, %r289, %r563; + mul.f32 %r1400, %r289, %r564; + mul.f32 %r1401, %r289, %r565; + mul.f32 %r1402, %r289, %r566; + mul.f32 %r1403, %r289, %r567; + mul.f32 %r1404, %r289, %r568; + mul.f32 %r1405, %r290, %r569; + mul.f32 %r1406, %r290, %r570; + mul.f32 %r1407, %r290, %r571; + mul.f32 %r1408, %r290, %r572; + mul.f32 %r1409, %r290, %r573; + mul.f32 %r1410, %r290, %r574; + mul.f32 %r1411, %r290, %r575; + mul.f32 %r1412, %r290, %r576; + mul.f32 %r1413, %r291, %r577; + mul.f32 %r1414, %r291, %r578; + mul.f32 %r1415, %r291, %r579; + mul.f32 %r1416, %r291, %r580; + mul.f32 %r1417, %r291, %r581; + mul.f32 %r1418, %r291, %r582; + mul.f32 %r1419, %r291, %r583; + mul.f32 %r1420, %r291, %r584; + mul.f32 %r1421, %r292, %r585; + mul.f32 %r1422, %r292, %r586; + mul.f32 %r1423, %r292, %r587; + mul.f32 %r1424, %r292, %r588; + mul.f32 %r1425, %r292, %r589; + mul.f32 %r1426, %r292, %r590; + mul.f32 %r1427, %r292, %r591; + mul.f32 %r1428, %r292, %r592; + mul.f32 %r1429, %r294, %r593; + mul.f32 %r1430, %r294, %r594; + mul.f32 %r1431, %r294, %r595; + mul.f32 %r1432, %r294, %r596; + mul.f32 %r1433, %r294, %r597; + mul.f32 %r1434, %r294, %r598; + mul.f32 %r1435, %r294, %r599; + mul.f32 %r1436, %r294, %r600; + mul.f32 %r1437, %r295, %r601; + mul.f32 %r1438, %r295, %r602; + mul.f32 %r1439, %r295, %r603; + mul.f32 %r1440, %r295, %r604; + mul.f32 %r1441, %r295, %r605; + mul.f32 %r1442, %r295, %r606; + mul.f32 %r1443, %r295, %r607; + mul.f32 %r1444, %r295, %r608; + mul.f32 %r1445, %r296, %r609; + mul.f32 %r1446, %r296, %r610; + mul.f32 %r1447, %r296, %r611; + mul.f32 %r1448, %r296, %r612; + mul.f32 %r1449, %r296, %r613; + mul.f32 %r1450, %r296, %r614; + mul.f32 %r1451, %r296, %r615; + mul.f32 %r1452, %r296, %r616; + mul.f32 %r1453, %r297, %r617; + mul.f32 %r1454, %r297, %r618; + mul.f32 %r1455, %r297, %r619; + mul.f32 %r1456, %r297, %r620; + mul.f32 %r1457, %r297, %r621; + mul.f32 %r1458, %r297, %r622; + mul.f32 %r1459, %r297, %r623; + mul.f32 %r1460, %r297, %r624; + .loc 1 83 18 // l2norm.py:83:18 + cvt.rn.bf16x2.f32 %r130, %r1206, %r1205; + cvt.rn.bf16x2.f32 %r131, %r1208, %r1207; + cvt.rn.bf16x2.f32 %r132, %r1210, %r1209; + cvt.rn.bf16x2.f32 %r133, %r1212, %r1211; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd33 + 0 ], { %r130, %r131, %r132, %r133 }; + // end inline asm + cvt.rn.bf16x2.f32 %r134, %r1214, %r1213; + cvt.rn.bf16x2.f32 %r135, %r1216, %r1215; + cvt.rn.bf16x2.f32 %r136, %r1218, %r1217; + cvt.rn.bf16x2.f32 %r137, %r1220, %r1219; + // begin inline asm + @%p3 st.global.v4.b32 [ %rd34 + 0 ], { %r134, %r135, %r136, %r137 }; + // end inline asm + cvt.rn.bf16x2.f32 %r138, %r1222, %r1221; + cvt.rn.bf16x2.f32 %r139, %r1224, %r1223; + cvt.rn.bf16x2.f32 %r140, %r1226, %r1225; + cvt.rn.bf16x2.f32 %r141, %r1228, %r1227; + // begin inline asm + @%p4 st.global.v4.b32 [ %rd35 + 0 ], { %r138, %r139, %r140, %r141 }; + // end inline asm + cvt.rn.bf16x2.f32 %r142, %r1230, %r1229; + cvt.rn.bf16x2.f32 %r143, %r1232, %r1231; + cvt.rn.bf16x2.f32 %r144, %r1234, %r1233; + cvt.rn.bf16x2.f32 %r145, %r1236, %r1235; + // begin inline asm + @%p5 st.global.v4.b32 [ %rd36 + 0 ], { %r142, %r143, %r144, %r145 }; + // end inline asm + cvt.rn.bf16x2.f32 %r146, %r1238, %r1237; + cvt.rn.bf16x2.f32 %r147, %r1240, %r1239; + cvt.rn.bf16x2.f32 %r148, %r1242, %r1241; + cvt.rn.bf16x2.f32 %r149, %r1244, %r1243; + // begin inline asm + @%p6 st.global.v4.b32 [ %rd37 + 0 ], { %r146, %r147, %r148, %r149 }; + // end inline asm + cvt.rn.bf16x2.f32 %r150, %r1246, %r1245; + cvt.rn.bf16x2.f32 %r151, %r1248, %r1247; + cvt.rn.bf16x2.f32 %r152, %r1250, %r1249; + cvt.rn.bf16x2.f32 %r153, %r1252, %r1251; + // begin inline asm + @%p7 st.global.v4.b32 [ %rd38 + 0 ], { %r150, %r151, %r152, %r153 }; + // end inline asm + cvt.rn.bf16x2.f32 %r154, %r1254, %r1253; + cvt.rn.bf16x2.f32 %r155, %r1256, %r1255; + cvt.rn.bf16x2.f32 %r156, %r1258, %r1257; + cvt.rn.bf16x2.f32 %r157, %r1260, %r1259; + // begin inline asm + @%p8 st.global.v4.b32 [ %rd39 + 0 ], { %r154, %r155, %r156, %r157 }; + // end inline asm + cvt.rn.bf16x2.f32 %r158, %r1262, %r1261; + cvt.rn.bf16x2.f32 %r159, %r1264, %r1263; + cvt.rn.bf16x2.f32 %r160, %r1266, %r1265; + cvt.rn.bf16x2.f32 %r161, %r1268, %r1267; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd40 + 0 ], { %r158, %r159, %r160, %r161 }; + // end inline asm + cvt.rn.bf16x2.f32 %r162, %r1270, %r1269; + cvt.rn.bf16x2.f32 %r163, %r1272, %r1271; + cvt.rn.bf16x2.f32 %r164, %r1274, %r1273; + cvt.rn.bf16x2.f32 %r165, %r1276, %r1275; + // begin inline asm + @%p9 st.global.v4.b32 [ %rd41 + 0 ], { %r162, %r163, %r164, %r165 }; + // end inline asm + cvt.rn.bf16x2.f32 %r166, %r1278, %r1277; + cvt.rn.bf16x2.f32 %r167, %r1280, %r1279; + cvt.rn.bf16x2.f32 %r168, %r1282, %r1281; + cvt.rn.bf16x2.f32 %r169, %r1284, %r1283; + // begin inline asm + @%p10 st.global.v4.b32 [ %rd42 + 0 ], { %r166, %r167, %r168, %r169 }; + // end inline asm + cvt.rn.bf16x2.f32 %r170, %r1286, %r1285; + cvt.rn.bf16x2.f32 %r171, %r1288, %r1287; + cvt.rn.bf16x2.f32 %r172, %r1290, %r1289; + cvt.rn.bf16x2.f32 %r173, %r1292, %r1291; + // begin inline asm + @%p11 st.global.v4.b32 [ %rd43 + 0 ], { %r170, %r171, %r172, %r173 }; + // end inline asm + cvt.rn.bf16x2.f32 %r174, %r1294, %r1293; + cvt.rn.bf16x2.f32 %r175, %r1296, %r1295; + cvt.rn.bf16x2.f32 %r176, %r1298, %r1297; + cvt.rn.bf16x2.f32 %r177, %r1300, %r1299; + // begin inline asm + @%p12 st.global.v4.b32 [ %rd44 + 0 ], { %r174, %r175, %r176, %r177 }; + // end inline asm + cvt.rn.bf16x2.f32 %r178, %r1302, %r1301; + cvt.rn.bf16x2.f32 %r179, %r1304, %r1303; + cvt.rn.bf16x2.f32 %r180, %r1306, %r1305; + cvt.rn.bf16x2.f32 %r181, %r1308, %r1307; + // begin inline asm + @%p13 st.global.v4.b32 [ %rd45 + 0 ], { %r178, %r179, %r180, %r181 }; + // end inline asm + cvt.rn.bf16x2.f32 %r182, %r1310, %r1309; + cvt.rn.bf16x2.f32 %r183, %r1312, %r1311; + cvt.rn.bf16x2.f32 %r184, %r1314, %r1313; + cvt.rn.bf16x2.f32 %r185, %r1316, %r1315; + // begin inline asm + @%p14 st.global.v4.b32 [ %rd46 + 0 ], { %r182, %r183, %r184, %r185 }; + // end inline asm + cvt.rn.bf16x2.f32 %r186, %r1318, %r1317; + cvt.rn.bf16x2.f32 %r187, %r1320, %r1319; + cvt.rn.bf16x2.f32 %r188, %r1322, %r1321; + cvt.rn.bf16x2.f32 %r189, %r1324, %r1323; + // begin inline asm + @%p15 st.global.v4.b32 [ %rd47 + 0 ], { %r186, %r187, %r188, %r189 }; + // end inline asm + cvt.rn.bf16x2.f32 %r190, %r1326, %r1325; + cvt.rn.bf16x2.f32 %r191, %r1328, %r1327; + cvt.rn.bf16x2.f32 %r192, %r1330, %r1329; + cvt.rn.bf16x2.f32 %r193, %r1332, %r1331; + // begin inline asm + @%p16 st.global.v4.b32 [ %rd48 + 0 ], { %r190, %r191, %r192, %r193 }; + // end inline asm + cvt.rn.bf16x2.f32 %r194, %r1334, %r1333; + cvt.rn.bf16x2.f32 %r195, %r1336, %r1335; + cvt.rn.bf16x2.f32 %r196, %r1338, %r1337; + cvt.rn.bf16x2.f32 %r197, %r1340, %r1339; + // begin inline asm + @%p17 st.global.v4.b32 [ %rd49 + 0 ], { %r194, %r195, %r196, %r197 }; + // end inline asm + cvt.rn.bf16x2.f32 %r198, %r1342, %r1341; + cvt.rn.bf16x2.f32 %r199, %r1344, %r1343; + cvt.rn.bf16x2.f32 %r200, %r1346, %r1345; + cvt.rn.bf16x2.f32 %r201, %r1348, %r1347; + // begin inline asm + @%p18 st.global.v4.b32 [ %rd50 + 0 ], { %r198, %r199, %r200, %r201 }; + // end inline asm + cvt.rn.bf16x2.f32 %r202, %r1350, %r1349; + cvt.rn.bf16x2.f32 %r203, %r1352, %r1351; + cvt.rn.bf16x2.f32 %r204, %r1354, %r1353; + cvt.rn.bf16x2.f32 %r205, %r1356, %r1355; + // begin inline asm + @%p19 st.global.v4.b32 [ %rd51 + 0 ], { %r202, %r203, %r204, %r205 }; + // end inline asm + cvt.rn.bf16x2.f32 %r206, %r1358, %r1357; + cvt.rn.bf16x2.f32 %r207, %r1360, %r1359; + cvt.rn.bf16x2.f32 %r208, %r1362, %r1361; + cvt.rn.bf16x2.f32 %r209, %r1364, %r1363; + // begin inline asm + @%p20 st.global.v4.b32 [ %rd52 + 0 ], { %r206, %r207, %r208, %r209 }; + // end inline asm + cvt.rn.bf16x2.f32 %r210, %r1366, %r1365; + cvt.rn.bf16x2.f32 %r211, %r1368, %r1367; + cvt.rn.bf16x2.f32 %r212, %r1370, %r1369; + cvt.rn.bf16x2.f32 %r213, %r1372, %r1371; + // begin inline asm + @%p21 st.global.v4.b32 [ %rd53 + 0 ], { %r210, %r211, %r212, %r213 }; + // end inline asm + cvt.rn.bf16x2.f32 %r214, %r1374, %r1373; + cvt.rn.bf16x2.f32 %r215, %r1376, %r1375; + cvt.rn.bf16x2.f32 %r216, %r1378, %r1377; + cvt.rn.bf16x2.f32 %r217, %r1380, %r1379; + // begin inline asm + @%p22 st.global.v4.b32 [ %rd54 + 0 ], { %r214, %r215, %r216, %r217 }; + // end inline asm + cvt.rn.bf16x2.f32 %r218, %r1382, %r1381; + cvt.rn.bf16x2.f32 %r219, %r1384, %r1383; + cvt.rn.bf16x2.f32 %r220, %r1386, %r1385; + cvt.rn.bf16x2.f32 %r221, %r1388, %r1387; + // begin inline asm + @%p23 st.global.v4.b32 [ %rd55 + 0 ], { %r218, %r219, %r220, %r221 }; + // end inline asm + cvt.rn.bf16x2.f32 %r222, %r1390, %r1389; + cvt.rn.bf16x2.f32 %r223, %r1392, %r1391; + cvt.rn.bf16x2.f32 %r224, %r1394, %r1393; + cvt.rn.bf16x2.f32 %r225, %r1396, %r1395; + // begin inline asm + @%p24 st.global.v4.b32 [ %rd56 + 0 ], { %r222, %r223, %r224, %r225 }; + // end inline asm + cvt.rn.bf16x2.f32 %r226, %r1398, %r1397; + cvt.rn.bf16x2.f32 %r227, %r1400, %r1399; + cvt.rn.bf16x2.f32 %r228, %r1402, %r1401; + cvt.rn.bf16x2.f32 %r229, %r1404, %r1403; + // begin inline asm + @%p25 st.global.v4.b32 [ %rd57 + 0 ], { %r226, %r227, %r228, %r229 }; + // end inline asm + cvt.rn.bf16x2.f32 %r230, %r1406, %r1405; + cvt.rn.bf16x2.f32 %r231, %r1408, %r1407; + cvt.rn.bf16x2.f32 %r232, %r1410, %r1409; + cvt.rn.bf16x2.f32 %r233, %r1412, %r1411; + // begin inline asm + @%p26 st.global.v4.b32 [ %rd58 + 0 ], { %r230, %r231, %r232, %r233 }; + // end inline asm + cvt.rn.bf16x2.f32 %r234, %r1414, %r1413; + cvt.rn.bf16x2.f32 %r235, %r1416, %r1415; + cvt.rn.bf16x2.f32 %r236, %r1418, %r1417; + cvt.rn.bf16x2.f32 %r237, %r1420, %r1419; + // begin inline asm + @%p27 st.global.v4.b32 [ %rd59 + 0 ], { %r234, %r235, %r236, %r237 }; + // end inline asm + cvt.rn.bf16x2.f32 %r238, %r1422, %r1421; + cvt.rn.bf16x2.f32 %r239, %r1424, %r1423; + cvt.rn.bf16x2.f32 %r240, %r1426, %r1425; + cvt.rn.bf16x2.f32 %r241, %r1428, %r1427; + // begin inline asm + @%p28 st.global.v4.b32 [ %rd60 + 0 ], { %r238, %r239, %r240, %r241 }; + // end inline asm + cvt.rn.bf16x2.f32 %r242, %r1430, %r1429; + cvt.rn.bf16x2.f32 %r243, %r1432, %r1431; + cvt.rn.bf16x2.f32 %r244, %r1434, %r1433; + cvt.rn.bf16x2.f32 %r245, %r1436, %r1435; + // begin inline asm + @%p29 st.global.v4.b32 [ %rd61 + 0 ], { %r242, %r243, %r244, %r245 }; + // end inline asm + cvt.rn.bf16x2.f32 %r246, %r1438, %r1437; + cvt.rn.bf16x2.f32 %r247, %r1440, %r1439; + cvt.rn.bf16x2.f32 %r248, %r1442, %r1441; + cvt.rn.bf16x2.f32 %r249, %r1444, %r1443; + // begin inline asm + @%p30 st.global.v4.b32 [ %rd62 + 0 ], { %r246, %r247, %r248, %r249 }; + // end inline asm + cvt.rn.bf16x2.f32 %r250, %r1446, %r1445; + cvt.rn.bf16x2.f32 %r251, %r1448, %r1447; + cvt.rn.bf16x2.f32 %r252, %r1450, %r1449; + cvt.rn.bf16x2.f32 %r253, %r1452, %r1451; + // begin inline asm + @%p31 st.global.v4.b32 [ %rd63 + 0 ], { %r250, %r251, %r252, %r253 }; + // end inline asm + cvt.rn.bf16x2.f32 %r254, %r1454, %r1453; + cvt.rn.bf16x2.f32 %r255, %r1456, %r1455; + cvt.rn.bf16x2.f32 %r256, %r1458, %r1457; + cvt.rn.bf16x2.f32 %r257, %r1460, %r1459; + // begin inline asm + @%p32 st.global.v4.b32 [ %rd64 + 0 ], { %r254, %r255, %r256, %r257 }; + // end inline asm + .loc 1 84 21 // l2norm.py:84:21 + bfe.s32 %r1461, %r302, 4, 1; + and.b32 %r1462, %r1461, 320; + and.b32 %r1463, %r302, 32; + shl.b32 %r1464, %r1463, 1; + xor.b32 %r1465, %r1462, %r1464; + mov.b32 %r1466, global_smem; + add.s32 %r258, %r1466, %r1465; + // begin inline asm + st.shared.v4.b32 [ %r258 + 0 ], { %r259, %r260, %r261, %r262 }; + // end inline asm + add.s32 %r263, %r258, 16; + // begin inline asm + st.shared.v4.b32 [ %r263 + 0 ], { %r264, %r265, %r266, %r267 }; + // end inline asm + add.s32 %r268, %r258, 32; + // begin inline asm + st.shared.v4.b32 [ %r268 + 0 ], { %r269, %r270, %r271, %r272 }; + // end inline asm + add.s32 %r273, %r258, 48; + // begin inline asm + st.shared.v4.b32 [ %r273 + 0 ], { %r274, %r275, %r276, %r277 }; + // end inline asm + add.s32 %r278, %r258, 128; + // begin inline asm + st.shared.v4.b32 [ %r278 + 0 ], { %r279, %r280, %r281, %r282 }; + // end inline asm + add.s32 %r283, %r258, 144; + // begin inline asm + st.shared.v4.b32 [ %r283 + 0 ], { %r284, %r285, %r286, %r287 }; + // end inline asm + add.s32 %r288, %r258, 160; + // begin inline asm + st.shared.v4.b32 [ %r288 + 0 ], { %r289, %r290, %r291, %r292 }; + // end inline asm + add.s32 %r293, %r258, 176; + // begin inline asm + st.shared.v4.b32 [ %r293 + 0 ], { %r294, %r295, %r296, %r297 }; + // end inline asm + bar.sync 0; + and.b32 %r1467, %r1201, 60; + shl.b32 %r1468, %r302, 6; + and.b32 %r1469, %r1468, 64; + shl.b32 %r1470, %r1463, 2; + or.b32 %r1471, %r1469, %r1470; + or.b32 %r1472, %r1471, %r1467; + add.s32 %r1473, %r1466, %r1472; + ld.shared.b32 %r298, [%r1473]; + xor.b32 %r1474, %r1472, 64; + add.s32 %r1475, %r1466, %r1474; + ld.shared.b32 %r299, [%r1475+256]; + // begin inline asm + @%p33 st.global.b32 [ %rd65 + 0 ], { %r298 }; + // end inline asm + // begin inline asm + @%p34 st.global.b32 [ %rd66 + 0 ], { %r299 }; + // end inline asm + .loc 1 84 4 // l2norm.py:84:4 + ret; +$L__tmp259: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp258 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 80 // DW_AT_call_line +.b8 43 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp258 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.source b/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..65999ed34486c21231e2e10dd8e7d7c057966940 --- /dev/null +++ b/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.source @@ -0,0 +1,172 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc38 = loc(unknown) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc45 = loc("x"(#loc)) +#loc46 = loc("y"(#loc)) +#loc47 = loc("rstd"(#loc)) +#loc48 = loc("eps"(#loc)) +#loc49 = loc("T"(#loc)) +#loc81 = loc("input"(#loc36)) +#loc82 = loc("a"(#loc41)) +#loc83 = loc("b"(#loc41)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc50) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc51) + %o_t = arith.constant 128 : i32 loc(#loc52) + %o_t_1 = arith.constant 128 : i64 loc(#loc52) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc52) + %o_t_3 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc53) + %o_t_4 = arith.extsi %o_t_3 : tensor<128xi32> to tensor<128xi64> loc(#loc54) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<128xi64> loc(#loc54) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<128xi64> loc(#loc54) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc55) + %m_t = arith.extsi %T : i32 to i64 loc(#loc56) + %m_t_7 = tt.splat %m_t : i64 -> tensor<128xi64> loc(#loc56) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<128xi64> loc(#loc56) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<128xi1> -> tensor<128x1xi1> loc(#loc57) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc58) + %m_x_10 = arith.constant 128 : i32 loc(#loc59) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc59) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc59) + %m_x_13 = tt.broadcast %m_x : tensor<128x1xi1> -> tensor<128x128xi1> loc(#loc60) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<128x128xi1> loc(#loc60) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<128x128xi1> loc(#loc60) + %p_x = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc61) + %p_x_16 = arith.constant 128 : i32 loc(#loc62) + %p_x_17 = arith.constant 128 : i64 loc(#loc62) + %p_x_18 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc62) + %p_x_19 = arith.muli %p_x, %p_x_18 : tensor<128x1xi64> loc(#loc62) + %p_x_20 = tt.splat %x : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc63) + %p_x_21 = tt.addptr %p_x_20, %p_x_19 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc63) + %p_x_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc64) + %p_x_23 = tt.broadcast %p_x_21 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc65) + %p_x_24 = tt.broadcast %p_x_22 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc65) + %p_x_25 = tt.addptr %p_x_23, %p_x_24 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc65) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc66) + %p_y_26 = arith.constant 128 : i32 loc(#loc67) + %p_y_27 = arith.constant 128 : i64 loc(#loc67) + %p_y_28 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc67) + %p_y_29 = arith.muli %p_y, %p_y_28 : tensor<128x1xi64> loc(#loc67) + %p_y_30 = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc68) + %p_y_31 = tt.addptr %p_y_30, %p_y_29 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc68) + %p_y_32 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc69) + %p_y_33 = tt.broadcast %p_y_31 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc70) + %p_y_34 = tt.broadcast %p_y_32 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc70) + %p_y_35 = tt.addptr %p_y_33, %p_y_34 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc70) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr> loc(#loc71) + %p_rstd_36 = tt.addptr %p_rstd, %o_t_6 : tensor<128x!tt.ptr>, tensor<128xi64> loc(#loc71) + %b_x = arith.constant 0.000000e+00 : f32 loc(#loc72) + %b_x_37 = arith.constant dense<0.000000e+00> : tensor<128x128xf32> loc(#loc72) + %b_x_38 = arith.truncf %b_x_37 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc72) + %b_x_39 = tt.load %p_x_25, %m_x_15, %b_x_38 : tensor<128x128x!tt.ptr> loc(#loc72) + %b_x_40 = arith.extf %b_x_39 : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc73) + %b_rstd = arith.mulf %b_x_40, %b_x_40 : tensor<128x128xf32> loc(#loc74) + %b_rstd_41 = tt.call @triton.language.standard.sum__fp32S128_128S_c1_cFalse_cNone(%b_rstd) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc75) + %b_rstd_42 = tt.splat %eps : f32 -> tensor<128xf32> loc(#loc76) + %b_rstd_43 = arith.addf %b_rstd_41, %b_rstd_42 : tensor<128xf32> loc(#loc76) + %b_rstd_44 = math.sqrt %b_rstd_43 : tensor<128xf32> loc(#loc77) + %b_rstd_45 = arith.constant 1 : i32 loc(#loc78) + %b_rstd_46 = arith.constant 1.000000e+00 : f32 loc(#loc78) + %b_rstd_47 = arith.constant dense<1.000000e+00> : tensor<128xf32> loc(#loc78) + %b_rstd_48 = arith.divf %b_rstd_47, %b_rstd_44 : tensor<128xf32> loc(#loc78) + %b_y = tt.expand_dims %b_rstd_48 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc79) + %b_y_49 = tt.broadcast %b_y : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc80) + %b_y_50 = arith.mulf %b_x_40, %b_y_49 : tensor<128x128xf32> loc(#loc80) + %0 = arith.truncf %b_y_50 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc32) + tt.store %p_y_35, %0, %m_x_15 : tensor<128x128x!tt.ptr> loc(#loc33) + tt.store %p_rstd_36, %b_rstd_48, %m_t_8 : tensor<128x!tt.ptr> loc(#loc34) + tt.return loc(#loc35) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S128_128S_c1_cFalse_cNone(%input: tensor<128x128xf32> loc("input"(#loc36))) -> tensor<128xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc37) + tt.reduce.return %2 : f32 loc(#loc37) + }) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc37) + tt.return %0 : tensor<128xf32> loc(#loc39) + ^bb1: // no predecessors + %1 = ub.poison : tensor<128xf32> loc(#loc40) + tt.return %1 : tensor<128xf32> loc(#loc40) + } loc(#loc36) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc41)), %b: f32 loc("b"(#loc41))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc42) + tt.return %0 : f32 loc(#loc43) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc44) + tt.return %1 : f32 loc(#loc44) + } loc(#loc41) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:18) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:29) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:37) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc50 = loc("i_t"(#loc1)) +#loc51 = loc("i_t"(#loc2)) +#loc52 = loc("o_t"(#loc3)) +#loc53 = loc("o_t"(#loc4)) +#loc54 = loc("o_t"(#loc5)) +#loc55 = loc("o_d"(#loc6)) +#loc56 = loc("m_t"(#loc7)) +#loc57 = loc("m_x"(#loc8)) +#loc58 = loc("m_x"(#loc9)) +#loc59 = loc("m_x"(#loc10)) +#loc60 = loc("m_x"(#loc11)) +#loc61 = loc("p_x"(#loc12)) +#loc62 = loc("p_x"(#loc13)) +#loc63 = loc("p_x"(#loc14)) +#loc64 = loc("p_x"(#loc15)) +#loc65 = loc("p_x"(#loc16)) +#loc66 = loc("p_y"(#loc17)) +#loc67 = loc("p_y"(#loc18)) +#loc68 = loc("p_y"(#loc19)) +#loc69 = loc("p_y"(#loc20)) +#loc70 = loc("p_y"(#loc21)) +#loc71 = loc("p_rstd"(#loc22)) +#loc72 = loc("b_x"(#loc23)) +#loc73 = loc("b_x"(#loc24)) +#loc74 = loc("b_rstd"(#loc25)) +#loc75 = loc("b_rstd"(#loc26)) +#loc76 = loc("b_rstd"(#loc27)) +#loc77 = loc("b_rstd"(#loc28)) +#loc78 = loc("b_rstd"(#loc29)) +#loc79 = loc("b_y"(#loc30)) +#loc80 = loc("b_y"(#loc31)) diff --git a/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.ttgir b/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..bf833fcff3eaf2220658ed40775f48f108ef9417 --- /dev/null +++ b/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.ttgir @@ -0,0 +1,135 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [2, 1], order = [1, 0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [2], threadsPerWarp = [32], warpsPerCTA = [2], order = [0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc1 = loc(unknown) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc34 = loc("x"(#loc)) +#loc35 = loc("y"(#loc)) +#loc36 = loc("rstd"(#loc)) +#loc37 = loc("eps"(#loc)) +#loc38 = loc("T"(#loc)) +#loc59 = loc("b_rstd"(#loc23)) +#loc66 = loc(callsite(#loc1 at #loc59)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 2 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<1.000000e+00> : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<128x128xbf16, #blocked> loc(#loc1) + %c128_i64 = arith.constant 128 : i64 loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<128x1xi64, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc39) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc40) + %o_t = arith.muli %i_t_3, %c128_i64 : i64 loc(#loc41) + %o_t_4 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc42) + %o_t_5 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #blocked1> loc(#loc42) + %o_t_6 = arith.extsi %o_t_4 : tensor<128xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_7 = arith.extsi %o_t_5 : tensor<128xi32, #blocked1> to tensor<128xi64, #blocked1> loc(#loc43) + %o_t_8 = tt.splat %o_t : i64 -> tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_9 = tt.splat %o_t : i64 -> tensor<128xi64, #blocked1> loc(#loc43) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<128xi64, #blocked1> loc(#loc43) + %m_t = arith.extsi %T : i32 to i64 loc(#loc44) + %m_t_12 = tt.splat %m_t : i64 -> tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_13 = tt.splat %m_t : i64 -> tensor<128xi64, #blocked1> loc(#loc44) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<128xi64, #blocked1> loc(#loc44) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<128xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128x1xi1, #blocked> loc(#loc45) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc46) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc46) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_1 : tensor<1x128xi32, #blocked> loc(#loc47) + %m_x_19 = tt.broadcast %m_x : tensor<128x1xi1, #blocked> -> tensor<128x128xi1, #blocked> loc(#loc48) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<128x128xi1, #blocked> loc(#loc48) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<128x128xi1, #blocked> loc(#loc48) + %p_x = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128x1xi64, #blocked> loc(#loc49) + %p_x_22 = arith.muli %p_x, %cst_2 : tensor<128x1xi64, #blocked> loc(#loc50) + %p_x_23 = tt.splat %x : !tt.ptr -> tensor<128x1x!tt.ptr, #blocked> loc(#loc51) + %p_x_24 = tt.addptr %p_x_23, %p_x_22 : tensor<128x1x!tt.ptr, #blocked>, tensor<128x1xi64, #blocked> loc(#loc51) + %p_x_25 = tt.broadcast %p_x_24 : tensor<128x1x!tt.ptr, #blocked> -> tensor<128x128x!tt.ptr, #blocked> loc(#loc52) + %p_x_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<128x128xi32, #blocked> loc(#loc52) + %p_x_27 = tt.addptr %p_x_25, %p_x_26 : tensor<128x128x!tt.ptr, #blocked>, tensor<128x128xi32, #blocked> loc(#loc52) + %p_y = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr, #blocked> loc(#loc53) + %p_y_28 = tt.addptr %p_y, %p_x_22 : tensor<128x1x!tt.ptr, #blocked>, tensor<128x1xi64, #blocked> loc(#loc53) + %p_y_29 = tt.broadcast %p_y_28 : tensor<128x1x!tt.ptr, #blocked> -> tensor<128x128x!tt.ptr, #blocked> loc(#loc54) + %p_y_30 = tt.addptr %p_y_29, %p_x_26 : tensor<128x128x!tt.ptr, #blocked>, tensor<128x128xi32, #blocked> loc(#loc54) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr, #blocked1> loc(#loc55) + %p_rstd_31 = tt.addptr %p_rstd, %o_t_11 : tensor<128x!tt.ptr, #blocked1>, tensor<128xi64, #blocked1> loc(#loc55) + %b_x = tt.load %p_x_27, %m_x_21, %cst_0 : tensor<128x128x!tt.ptr, #blocked> loc(#loc56) + %b_x_32 = arith.extf %b_x : tensor<128x128xbf16, #blocked> to tensor<128x128xf32, #blocked> loc(#loc57) + %b_rstd = arith.mulf %b_x_32, %b_x_32 : tensor<128x128xf32, #blocked> loc(#loc58) + %b_rstd_33 = "tt.reduce"(%b_rstd) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_40: f32 loc(callsite(#loc1 at #loc59)), %b_rstd_41: f32 loc(callsite(#loc1 at #loc59))): + %b_rstd_42 = arith.addf %b_rstd_40, %b_rstd_41 : f32 loc(#loc67) + tt.reduce.return %b_rstd_42 : f32 loc(#loc65) + }) : (tensor<128x128xf32, #blocked>) -> tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc65) + %b_rstd_34 = tt.splat %eps : f32 -> tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_36 = math.sqrt %b_rstd_35 : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc61) + %b_rstd_37 = arith.divf %cst, %b_rstd_36 : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc62) + %b_y = tt.expand_dims %b_rstd_37 {axis = 1 : i32} : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128x1xf32, #blocked> loc(#loc63) + %b_y_38 = tt.broadcast %b_y : tensor<128x1xf32, #blocked> -> tensor<128x128xf32, #blocked> loc(#loc64) + %b_y_39 = arith.mulf %b_x_32, %b_y_38 : tensor<128x128xf32, #blocked> loc(#loc64) + %0 = arith.truncf %b_y_39 : tensor<128x128xf32, #blocked> to tensor<128x128xbf16, #blocked> loc(#loc30) + tt.store %p_y_30, %0, %m_x_21 : tensor<128x128x!tt.ptr, #blocked> loc(#loc31) + %1 = ttg.convert_layout %b_rstd_37 : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128xf32, #blocked1> loc(#loc32) + tt.store %p_rstd_31, %1, %m_t_15 : tensor<128x!tt.ptr, #blocked1> loc(#loc32) + tt.return loc(#loc33) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc39 = loc("i_t"(#loc2)) +#loc40 = loc("i_t"(#loc3)) +#loc41 = loc("o_t"(#loc4)) +#loc42 = loc("o_t"(#loc5)) +#loc43 = loc("o_t"(#loc6)) +#loc44 = loc("m_t"(#loc7)) +#loc45 = loc("m_x"(#loc8)) +#loc46 = loc("m_x"(#loc9)) +#loc47 = loc("m_x"(#loc10)) +#loc48 = loc("m_x"(#loc11)) +#loc49 = loc("p_x"(#loc12)) +#loc50 = loc("p_x"(#loc13)) +#loc51 = loc("p_x"(#loc14)) +#loc52 = loc("p_x"(#loc15)) +#loc53 = loc("p_y"(#loc16)) +#loc54 = loc("p_y"(#loc17)) +#loc55 = loc("p_rstd"(#loc18)) +#loc56 = loc("b_x"(#loc19)) +#loc57 = loc("b_x"(#loc20)) +#loc58 = loc("b_rstd"(#loc21)) +#loc60 = loc("b_rstd"(#loc25)) +#loc61 = loc("b_rstd"(#loc26)) +#loc62 = loc("b_rstd"(#loc27)) +#loc63 = loc("b_y"(#loc28)) +#loc64 = loc("b_y"(#loc29)) +#loc65 = loc(callsite(#loc22 at #loc59)) +#loc67 = loc(callsite(#loc24 at #loc65)) diff --git a/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.ttir b/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..8f0dfe3371851abdc6b1c20ee94003c63d3c42a8 --- /dev/null +++ b/triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.ttir @@ -0,0 +1,125 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc3 = loc(unknown) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc34 = loc("x"(#loc)) +#loc35 = loc("y"(#loc)) +#loc36 = loc("rstd"(#loc)) +#loc37 = loc("eps"(#loc)) +#loc38 = loc("T"(#loc)) +#loc60 = loc("b_rstd"(#loc24)) +#loc66 = loc(callsite(#loc3 at #loc60)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %b_x = arith.constant dense<0.000000e+00> : tensor<128x128xbf16> loc(#loc39) + %b_rstd = arith.constant dense<1.000000e+00> : tensor<128xf32> loc(#loc40) + %cst = arith.constant dense<128> : tensor<128x1xi64> loc(#loc3) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc41) + %c128_i64 = arith.constant 128 : i64 loc(#loc3) + %i_t = tt.get_program_id x : i32 loc(#loc42) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc43) + %o_t = arith.muli %i_t_0, %c128_i64 : i64 loc(#loc44) + %o_t_1 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc45) + %o_t_2 = arith.extsi %o_t_1 : tensor<128xi32> to tensor<128xi64> loc(#loc46) + %o_t_3 = tt.splat %o_t : i64 -> tensor<128xi64> loc(#loc46) + %o_t_4 = arith.addi %o_t_3, %o_t_2 : tensor<128xi64> loc(#loc46) + %m_t = arith.extsi %T : i32 to i64 loc(#loc47) + %m_t_5 = tt.splat %m_t : i64 -> tensor<128xi64> loc(#loc47) + %m_t_6 = arith.cmpi slt, %o_t_4, %m_t_5 : tensor<128xi64> loc(#loc47) + %m_x_7 = tt.expand_dims %m_t_6 {axis = 1 : i32} : tensor<128xi1> -> tensor<128x1xi1> loc(#loc48) + %m_x_8 = tt.expand_dims %o_t_1 {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc49) + %m_x_9 = arith.cmpi slt, %m_x_8, %m_x : tensor<1x128xi32> loc(#loc41) + %m_x_10 = tt.broadcast %m_x_7 : tensor<128x1xi1> -> tensor<128x128xi1> loc(#loc50) + %m_x_11 = tt.broadcast %m_x_9 : tensor<1x128xi1> -> tensor<128x128xi1> loc(#loc50) + %m_x_12 = arith.andi %m_x_10, %m_x_11 : tensor<128x128xi1> loc(#loc50) + %p_x = tt.expand_dims %o_t_4 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc51) + %p_x_13 = arith.muli %p_x, %cst : tensor<128x1xi64> loc(#loc52) + %p_x_14 = tt.splat %x : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc53) + %p_x_15 = tt.addptr %p_x_14, %p_x_13 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc53) + %p_x_16 = tt.broadcast %p_x_15 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc54) + %p_x_17 = tt.broadcast %m_x_8 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc54) + %p_x_18 = tt.addptr %p_x_16, %p_x_17 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc54) + %p_y = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc55) + %p_y_19 = tt.addptr %p_y, %p_x_13 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc55) + %p_y_20 = tt.broadcast %p_y_19 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc56) + %p_y_21 = tt.addptr %p_y_20, %p_x_17 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc56) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr> loc(#loc57) + %p_rstd_22 = tt.addptr %p_rstd, %o_t_4 : tensor<128x!tt.ptr>, tensor<128xi64> loc(#loc57) + %b_x_23 = tt.load %p_x_18, %m_x_12, %b_x : tensor<128x128x!tt.ptr> loc(#loc39) + %b_x_24 = arith.extf %b_x_23 : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc58) + %b_rstd_25 = arith.mulf %b_x_24, %b_x_24 : tensor<128x128xf32> loc(#loc59) + %b_rstd_26 = "tt.reduce"(%b_rstd_25) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_33: f32 loc(callsite(#loc3 at #loc60)), %b_rstd_34: f32 loc(callsite(#loc3 at #loc60))): + %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : f32 loc(#loc67) + tt.reduce.return %b_rstd_35 : f32 loc(#loc65) + }) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc65) + %b_rstd_27 = tt.splat %eps : f32 -> tensor<128xf32> loc(#loc61) + %b_rstd_28 = arith.addf %b_rstd_26, %b_rstd_27 : tensor<128xf32> loc(#loc61) + %b_rstd_29 = math.sqrt %b_rstd_28 : tensor<128xf32> loc(#loc62) + %b_rstd_30 = arith.divf %b_rstd, %b_rstd_29 : tensor<128xf32> loc(#loc40) + %b_y = tt.expand_dims %b_rstd_30 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc63) + %b_y_31 = tt.broadcast %b_y : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc64) + %b_y_32 = arith.mulf %b_x_24, %b_y_31 : tensor<128x128xf32> loc(#loc64) + %0 = arith.truncf %b_y_32 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc30) + tt.store %p_y_21, %0, %m_x_12 : tensor<128x128x!tt.ptr> loc(#loc31) + tt.store %p_rstd_22, %b_rstd_30, %m_t_6 : tensor<128x!tt.ptr> loc(#loc32) + tt.return loc(#loc33) + } loc(#loc) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc39 = loc("b_x"(#loc1)) +#loc40 = loc("b_rstd"(#loc2)) +#loc41 = loc("m_x"(#loc4)) +#loc42 = loc("i_t"(#loc5)) +#loc43 = loc("i_t"(#loc6)) +#loc44 = loc("o_t"(#loc7)) +#loc45 = loc("o_t"(#loc8)) +#loc46 = loc("o_t"(#loc9)) +#loc47 = loc("m_t"(#loc10)) +#loc48 = loc("m_x"(#loc11)) +#loc49 = loc("m_x"(#loc12)) +#loc50 = loc("m_x"(#loc13)) +#loc51 = loc("p_x"(#loc14)) +#loc52 = loc("p_x"(#loc15)) +#loc53 = loc("p_x"(#loc16)) +#loc54 = loc("p_x"(#loc17)) +#loc55 = loc("p_y"(#loc18)) +#loc56 = loc("p_y"(#loc19)) +#loc57 = loc("p_rstd"(#loc20)) +#loc58 = loc("b_x"(#loc21)) +#loc59 = loc("b_rstd"(#loc22)) +#loc61 = loc("b_rstd"(#loc26)) +#loc62 = loc("b_rstd"(#loc27)) +#loc63 = loc("b_y"(#loc28)) +#loc64 = loc("b_y"(#loc29)) +#loc65 = loc(callsite(#loc23 at #loc60)) +#loc67 = loc(callsite(#loc25 at #loc65)) diff --git a/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/__grp__l2norm_fwd_kernel.json b/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/__grp__l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..ebfcd303d38ed6de7b2e08609aca53837b1e6247 --- /dev/null +++ b/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/__grp__l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_fwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.source", "l2norm_fwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.ttir", "l2norm_fwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.ttgir", "l2norm_fwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.llir", "l2norm_fwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.ptx", "l2norm_fwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.cubin", "l2norm_fwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.cubin b/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..406d90e7df79f722999ea5b13a6aea3e9b984bb6 Binary files /dev/null and b/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.cubin differ diff --git a/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.json b/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..a8a1fb9b617d7d0fa6989c3a287284fb6fd84709 --- /dev/null +++ b/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"hash": "051f4635e9b9fc0df40e7c28edf1ccc5f232cb866100a32f16d01654261767ee", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 2, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 128, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_fwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.llir b/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..627c0508dc94ce9f7595b0f6ff9a2f82a7e68083 --- /dev/null +++ b/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.llir @@ -0,0 +1,1143 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 +@.str.2 = private unnamed_addr constant [17 x i8] c"__CUDA_PREC_SQRT\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_fwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, float %3, i32 %4, ptr addrspace(1) readnone captures(none) %5, ptr addrspace(1) readnone captures(none) %6) local_unnamed_addr #0 !dbg !5 { + %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %9 = zext nneg i32 %8 to i64, !dbg !14 + %10 = shl nuw nsw i64 %9, 5, !dbg !15 + %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !16 + %12 = and i32 %11, 48, !dbg !16 + %13 = lshr exact i32 %12, 4, !dbg !16 + %14 = or disjoint i32 %13, 4, !dbg !16 + %15 = or disjoint i32 %13, 8, !dbg !16 + %16 = or disjoint i32 %13, 12, !dbg !16 + %17 = or disjoint i32 %13, 16, !dbg !16 + %18 = or disjoint i32 %13, 20, !dbg !16 + %19 = or disjoint i32 %13, 24, !dbg !16 + %20 = or disjoint i32 %13, 28, !dbg !16 + %21 = zext nneg i32 %13 to i64, !dbg !17 + %22 = zext nneg i32 %14 to i64, !dbg !17 + %23 = zext nneg i32 %15 to i64, !dbg !17 + %24 = zext nneg i32 %16 to i64, !dbg !17 + %25 = zext nneg i32 %17 to i64, !dbg !17 + %26 = zext nneg i32 %18 to i64, !dbg !17 + %27 = zext nneg i32 %19 to i64, !dbg !17 + %28 = zext nneg i32 %20 to i64, !dbg !17 + %29 = or disjoint i64 %10, %21, !dbg !17 + %30 = or disjoint i64 %10, %22, !dbg !17 + %31 = or disjoint i64 %10, %23, !dbg !17 + %32 = or disjoint i64 %10, %24, !dbg !17 + %33 = or disjoint i64 %10, %25, !dbg !17 + %34 = or disjoint i64 %10, %26, !dbg !17 + %35 = or disjoint i64 %10, %27, !dbg !17 + %36 = or disjoint i64 %10, %28, !dbg !17 + %37 = sext i32 %4 to i64, !dbg !18 + %38 = icmp slt i64 %29, %37, !dbg !18 + %39 = icmp slt i64 %30, %37, !dbg !18 + %40 = icmp slt i64 %31, %37, !dbg !18 + %41 = icmp slt i64 %32, %37, !dbg !18 + %42 = icmp slt i64 %33, %37, !dbg !18 + %43 = icmp slt i64 %34, %37, !dbg !18 + %44 = icmp slt i64 %35, %37, !dbg !18 + %45 = icmp slt i64 %36, %37, !dbg !18 + %46 = shl nuw nsw i32 %11, 3, !dbg !19 + %47 = and i32 %46, 120, !dbg !19 + %48 = shl nuw nsw i64 %29, 7, !dbg !20 + %49 = shl nuw nsw i64 %30, 7, !dbg !20 + %50 = shl nuw nsw i64 %31, 7, !dbg !20 + %51 = shl nuw nsw i64 %32, 7, !dbg !20 + %52 = shl nuw nsw i64 %33, 7, !dbg !20 + %53 = shl nuw nsw i64 %34, 7, !dbg !20 + %54 = shl nuw nsw i64 %35, 7, !dbg !20 + %55 = shl nuw nsw i64 %36, 7, !dbg !20 + %56 = getelementptr bfloat, ptr addrspace(1) %0, i64 %48, !dbg !21 + %57 = getelementptr bfloat, ptr addrspace(1) %0, i64 %49, !dbg !21 + %58 = getelementptr bfloat, ptr addrspace(1) %0, i64 %50, !dbg !21 + %59 = getelementptr bfloat, ptr addrspace(1) %0, i64 %51, !dbg !21 + %60 = getelementptr bfloat, ptr addrspace(1) %0, i64 %52, !dbg !21 + %61 = getelementptr bfloat, ptr addrspace(1) %0, i64 %53, !dbg !21 + %62 = getelementptr bfloat, ptr addrspace(1) %0, i64 %54, !dbg !21 + %63 = getelementptr bfloat, ptr addrspace(1) %0, i64 %55, !dbg !21 + %64 = zext nneg i32 %47 to i64, !dbg !22 + %65 = getelementptr bfloat, ptr addrspace(1) %56, i64 %64, !dbg !22 + %66 = getelementptr bfloat, ptr addrspace(1) %57, i64 %64, !dbg !22 + %67 = getelementptr bfloat, ptr addrspace(1) %58, i64 %64, !dbg !22 + %68 = getelementptr bfloat, ptr addrspace(1) %59, i64 %64, !dbg !22 + %69 = getelementptr bfloat, ptr addrspace(1) %60, i64 %64, !dbg !22 + %70 = getelementptr bfloat, ptr addrspace(1) %61, i64 %64, !dbg !22 + %71 = getelementptr bfloat, ptr addrspace(1) %62, i64 %64, !dbg !22 + %72 = getelementptr bfloat, ptr addrspace(1) %63, i64 %64, !dbg !22 + %73 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %65, i1 %38) #6, !dbg !23 + %74 = extractvalue { i32, i32, i32, i32 } %73, 0, !dbg !23 + %75 = bitcast i32 %74 to <2 x bfloat>, !dbg !23 + %76 = extractvalue { i32, i32, i32, i32 } %73, 1, !dbg !23 + %77 = bitcast i32 %76 to <2 x bfloat>, !dbg !23 + %78 = extractvalue { i32, i32, i32, i32 } %73, 2, !dbg !23 + %79 = bitcast i32 %78 to <2 x bfloat>, !dbg !23 + %80 = extractvalue { i32, i32, i32, i32 } %73, 3, !dbg !23 + %81 = bitcast i32 %80 to <2 x bfloat>, !dbg !23 + %82 = extractelement <2 x bfloat> %75, i64 0, !dbg !23 + %83 = extractelement <2 x bfloat> %75, i64 1, !dbg !23 + %84 = extractelement <2 x bfloat> %77, i64 0, !dbg !23 + %85 = extractelement <2 x bfloat> %77, i64 1, !dbg !23 + %86 = extractelement <2 x bfloat> %79, i64 0, !dbg !23 + %87 = extractelement <2 x bfloat> %79, i64 1, !dbg !23 + %88 = extractelement <2 x bfloat> %81, i64 0, !dbg !23 + %89 = extractelement <2 x bfloat> %81, i64 1, !dbg !23 + %90 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %66, i1 %39) #6, !dbg !23 + %91 = extractvalue { i32, i32, i32, i32 } %90, 0, !dbg !23 + %92 = bitcast i32 %91 to <2 x bfloat>, !dbg !23 + %93 = extractvalue { i32, i32, i32, i32 } %90, 1, !dbg !23 + %94 = bitcast i32 %93 to <2 x bfloat>, !dbg !23 + %95 = extractvalue { i32, i32, i32, i32 } %90, 2, !dbg !23 + %96 = bitcast i32 %95 to <2 x bfloat>, !dbg !23 + %97 = extractvalue { i32, i32, i32, i32 } %90, 3, !dbg !23 + %98 = bitcast i32 %97 to <2 x bfloat>, !dbg !23 + %99 = extractelement <2 x bfloat> %92, i64 0, !dbg !23 + %100 = extractelement <2 x bfloat> %92, i64 1, !dbg !23 + %101 = extractelement <2 x bfloat> %94, i64 0, !dbg !23 + %102 = extractelement <2 x bfloat> %94, i64 1, !dbg !23 + %103 = extractelement <2 x bfloat> %96, i64 0, !dbg !23 + %104 = extractelement <2 x bfloat> %96, i64 1, !dbg !23 + %105 = extractelement <2 x bfloat> %98, i64 0, !dbg !23 + %106 = extractelement <2 x bfloat> %98, i64 1, !dbg !23 + %107 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %67, i1 %40) #6, !dbg !23 + %108 = extractvalue { i32, i32, i32, i32 } %107, 0, !dbg !23 + %109 = bitcast i32 %108 to <2 x bfloat>, !dbg !23 + %110 = extractvalue { i32, i32, i32, i32 } %107, 1, !dbg !23 + %111 = bitcast i32 %110 to <2 x bfloat>, !dbg !23 + %112 = extractvalue { i32, i32, i32, i32 } %107, 2, !dbg !23 + %113 = bitcast i32 %112 to <2 x bfloat>, !dbg !23 + %114 = extractvalue { i32, i32, i32, i32 } %107, 3, !dbg !23 + %115 = bitcast i32 %114 to <2 x bfloat>, !dbg !23 + %116 = extractelement <2 x bfloat> %109, i64 0, !dbg !23 + %117 = extractelement <2 x bfloat> %109, i64 1, !dbg !23 + %118 = extractelement <2 x bfloat> %111, i64 0, !dbg !23 + %119 = extractelement <2 x bfloat> %111, i64 1, !dbg !23 + %120 = extractelement <2 x bfloat> %113, i64 0, !dbg !23 + %121 = extractelement <2 x bfloat> %113, i64 1, !dbg !23 + %122 = extractelement <2 x bfloat> %115, i64 0, !dbg !23 + %123 = extractelement <2 x bfloat> %115, i64 1, !dbg !23 + %124 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %68, i1 %41) #6, !dbg !23 + %125 = extractvalue { i32, i32, i32, i32 } %124, 0, !dbg !23 + %126 = bitcast i32 %125 to <2 x bfloat>, !dbg !23 + %127 = extractvalue { i32, i32, i32, i32 } %124, 1, !dbg !23 + %128 = bitcast i32 %127 to <2 x bfloat>, !dbg !23 + %129 = extractvalue { i32, i32, i32, i32 } %124, 2, !dbg !23 + %130 = bitcast i32 %129 to <2 x bfloat>, !dbg !23 + %131 = extractvalue { i32, i32, i32, i32 } %124, 3, !dbg !23 + %132 = bitcast i32 %131 to <2 x bfloat>, !dbg !23 + %133 = extractelement <2 x bfloat> %126, i64 0, !dbg !23 + %134 = extractelement <2 x bfloat> %126, i64 1, !dbg !23 + %135 = extractelement <2 x bfloat> %128, i64 0, !dbg !23 + %136 = extractelement <2 x bfloat> %128, i64 1, !dbg !23 + %137 = extractelement <2 x bfloat> %130, i64 0, !dbg !23 + %138 = extractelement <2 x bfloat> %130, i64 1, !dbg !23 + %139 = extractelement <2 x bfloat> %132, i64 0, !dbg !23 + %140 = extractelement <2 x bfloat> %132, i64 1, !dbg !23 + %141 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %69, i1 %42) #6, !dbg !23 + %142 = extractvalue { i32, i32, i32, i32 } %141, 0, !dbg !23 + %143 = bitcast i32 %142 to <2 x bfloat>, !dbg !23 + %144 = extractvalue { i32, i32, i32, i32 } %141, 1, !dbg !23 + %145 = bitcast i32 %144 to <2 x bfloat>, !dbg !23 + %146 = extractvalue { i32, i32, i32, i32 } %141, 2, !dbg !23 + %147 = bitcast i32 %146 to <2 x bfloat>, !dbg !23 + %148 = extractvalue { i32, i32, i32, i32 } %141, 3, !dbg !23 + %149 = bitcast i32 %148 to <2 x bfloat>, !dbg !23 + %150 = extractelement <2 x bfloat> %143, i64 0, !dbg !23 + %151 = extractelement <2 x bfloat> %143, i64 1, !dbg !23 + %152 = extractelement <2 x bfloat> %145, i64 0, !dbg !23 + %153 = extractelement <2 x bfloat> %145, i64 1, !dbg !23 + %154 = extractelement <2 x bfloat> %147, i64 0, !dbg !23 + %155 = extractelement <2 x bfloat> %147, i64 1, !dbg !23 + %156 = extractelement <2 x bfloat> %149, i64 0, !dbg !23 + %157 = extractelement <2 x bfloat> %149, i64 1, !dbg !23 + %158 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %70, i1 %43) #6, !dbg !23 + %159 = extractvalue { i32, i32, i32, i32 } %158, 0, !dbg !23 + %160 = bitcast i32 %159 to <2 x bfloat>, !dbg !23 + %161 = extractvalue { i32, i32, i32, i32 } %158, 1, !dbg !23 + %162 = bitcast i32 %161 to <2 x bfloat>, !dbg !23 + %163 = extractvalue { i32, i32, i32, i32 } %158, 2, !dbg !23 + %164 = bitcast i32 %163 to <2 x bfloat>, !dbg !23 + %165 = extractvalue { i32, i32, i32, i32 } %158, 3, !dbg !23 + %166 = bitcast i32 %165 to <2 x bfloat>, !dbg !23 + %167 = extractelement <2 x bfloat> %160, i64 0, !dbg !23 + %168 = extractelement <2 x bfloat> %160, i64 1, !dbg !23 + %169 = extractelement <2 x bfloat> %162, i64 0, !dbg !23 + %170 = extractelement <2 x bfloat> %162, i64 1, !dbg !23 + %171 = extractelement <2 x bfloat> %164, i64 0, !dbg !23 + %172 = extractelement <2 x bfloat> %164, i64 1, !dbg !23 + %173 = extractelement <2 x bfloat> %166, i64 0, !dbg !23 + %174 = extractelement <2 x bfloat> %166, i64 1, !dbg !23 + %175 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %71, i1 %44) #6, !dbg !23 + %176 = extractvalue { i32, i32, i32, i32 } %175, 0, !dbg !23 + %177 = bitcast i32 %176 to <2 x bfloat>, !dbg !23 + %178 = extractvalue { i32, i32, i32, i32 } %175, 1, !dbg !23 + %179 = bitcast i32 %178 to <2 x bfloat>, !dbg !23 + %180 = extractvalue { i32, i32, i32, i32 } %175, 2, !dbg !23 + %181 = bitcast i32 %180 to <2 x bfloat>, !dbg !23 + %182 = extractvalue { i32, i32, i32, i32 } %175, 3, !dbg !23 + %183 = bitcast i32 %182 to <2 x bfloat>, !dbg !23 + %184 = extractelement <2 x bfloat> %177, i64 0, !dbg !23 + %185 = extractelement <2 x bfloat> %177, i64 1, !dbg !23 + %186 = extractelement <2 x bfloat> %179, i64 0, !dbg !23 + %187 = extractelement <2 x bfloat> %179, i64 1, !dbg !23 + %188 = extractelement <2 x bfloat> %181, i64 0, !dbg !23 + %189 = extractelement <2 x bfloat> %181, i64 1, !dbg !23 + %190 = extractelement <2 x bfloat> %183, i64 0, !dbg !23 + %191 = extractelement <2 x bfloat> %183, i64 1, !dbg !23 + %192 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %72, i1 %45) #6, !dbg !23 + %193 = extractvalue { i32, i32, i32, i32 } %192, 0, !dbg !23 + %194 = bitcast i32 %193 to <2 x bfloat>, !dbg !23 + %195 = extractvalue { i32, i32, i32, i32 } %192, 1, !dbg !23 + %196 = bitcast i32 %195 to <2 x bfloat>, !dbg !23 + %197 = extractvalue { i32, i32, i32, i32 } %192, 2, !dbg !23 + %198 = bitcast i32 %197 to <2 x bfloat>, !dbg !23 + %199 = extractvalue { i32, i32, i32, i32 } %192, 3, !dbg !23 + %200 = bitcast i32 %199 to <2 x bfloat>, !dbg !23 + %201 = extractelement <2 x bfloat> %194, i64 0, !dbg !23 + %202 = extractelement <2 x bfloat> %194, i64 1, !dbg !23 + %203 = extractelement <2 x bfloat> %196, i64 0, !dbg !23 + %204 = extractelement <2 x bfloat> %196, i64 1, !dbg !23 + %205 = extractelement <2 x bfloat> %198, i64 0, !dbg !23 + %206 = extractelement <2 x bfloat> %198, i64 1, !dbg !23 + %207 = extractelement <2 x bfloat> %200, i64 0, !dbg !23 + %208 = extractelement <2 x bfloat> %200, i64 1, !dbg !23 + %209 = fpext bfloat %82 to float, !dbg !24 + %210 = fpext bfloat %83 to float, !dbg !24 + %211 = fpext bfloat %84 to float, !dbg !24 + %212 = fpext bfloat %85 to float, !dbg !24 + %213 = fpext bfloat %86 to float, !dbg !24 + %214 = fpext bfloat %87 to float, !dbg !24 + %215 = fpext bfloat %88 to float, !dbg !24 + %216 = fpext bfloat %89 to float, !dbg !24 + %217 = fpext bfloat %99 to float, !dbg !24 + %218 = fpext bfloat %100 to float, !dbg !24 + %219 = fpext bfloat %101 to float, !dbg !24 + %220 = fpext bfloat %102 to float, !dbg !24 + %221 = fpext bfloat %103 to float, !dbg !24 + %222 = fpext bfloat %104 to float, !dbg !24 + %223 = fpext bfloat %105 to float, !dbg !24 + %224 = fpext bfloat %106 to float, !dbg !24 + %225 = fpext bfloat %116 to float, !dbg !24 + %226 = fpext bfloat %117 to float, !dbg !24 + %227 = fpext bfloat %118 to float, !dbg !24 + %228 = fpext bfloat %119 to float, !dbg !24 + %229 = fpext bfloat %120 to float, !dbg !24 + %230 = fpext bfloat %121 to float, !dbg !24 + %231 = fpext bfloat %122 to float, !dbg !24 + %232 = fpext bfloat %123 to float, !dbg !24 + %233 = fpext bfloat %133 to float, !dbg !24 + %234 = fpext bfloat %134 to float, !dbg !24 + %235 = fpext bfloat %135 to float, !dbg !24 + %236 = fpext bfloat %136 to float, !dbg !24 + %237 = fpext bfloat %137 to float, !dbg !24 + %238 = fpext bfloat %138 to float, !dbg !24 + %239 = fpext bfloat %139 to float, !dbg !24 + %240 = fpext bfloat %140 to float, !dbg !24 + %241 = fpext bfloat %150 to float, !dbg !24 + %242 = fpext bfloat %151 to float, !dbg !24 + %243 = fpext bfloat %152 to float, !dbg !24 + %244 = fpext bfloat %153 to float, !dbg !24 + %245 = fpext bfloat %154 to float, !dbg !24 + %246 = fpext bfloat %155 to float, !dbg !24 + %247 = fpext bfloat %156 to float, !dbg !24 + %248 = fpext bfloat %157 to float, !dbg !24 + %249 = fpext bfloat %167 to float, !dbg !24 + %250 = fpext bfloat %168 to float, !dbg !24 + %251 = fpext bfloat %169 to float, !dbg !24 + %252 = fpext bfloat %170 to float, !dbg !24 + %253 = fpext bfloat %171 to float, !dbg !24 + %254 = fpext bfloat %172 to float, !dbg !24 + %255 = fpext bfloat %173 to float, !dbg !24 + %256 = fpext bfloat %174 to float, !dbg !24 + %257 = fpext bfloat %184 to float, !dbg !24 + %258 = fpext bfloat %185 to float, !dbg !24 + %259 = fpext bfloat %186 to float, !dbg !24 + %260 = fpext bfloat %187 to float, !dbg !24 + %261 = fpext bfloat %188 to float, !dbg !24 + %262 = fpext bfloat %189 to float, !dbg !24 + %263 = fpext bfloat %190 to float, !dbg !24 + %264 = fpext bfloat %191 to float, !dbg !24 + %265 = fpext bfloat %201 to float, !dbg !24 + %266 = fpext bfloat %202 to float, !dbg !24 + %267 = fpext bfloat %203 to float, !dbg !24 + %268 = fpext bfloat %204 to float, !dbg !24 + %269 = fpext bfloat %205 to float, !dbg !24 + %270 = fpext bfloat %206 to float, !dbg !24 + %271 = fpext bfloat %207 to float, !dbg !24 + %272 = fpext bfloat %208 to float, !dbg !24 + %273 = fmul float %209, %209, !dbg !25 + %274 = fmul float %210, %210, !dbg !25 + %275 = fmul float %211, %211, !dbg !25 + %276 = fmul float %212, %212, !dbg !25 + %277 = fmul float %213, %213, !dbg !25 + %278 = fmul float %214, %214, !dbg !25 + %279 = fmul float %215, %215, !dbg !25 + %280 = fmul float %216, %216, !dbg !25 + %281 = fmul float %217, %217, !dbg !25 + %282 = fmul float %218, %218, !dbg !25 + %283 = fmul float %219, %219, !dbg !25 + %284 = fmul float %220, %220, !dbg !25 + %285 = fmul float %221, %221, !dbg !25 + %286 = fmul float %222, %222, !dbg !25 + %287 = fmul float %223, %223, !dbg !25 + %288 = fmul float %224, %224, !dbg !25 + %289 = fmul float %225, %225, !dbg !25 + %290 = fmul float %226, %226, !dbg !25 + %291 = fmul float %227, %227, !dbg !25 + %292 = fmul float %228, %228, !dbg !25 + %293 = fmul float %229, %229, !dbg !25 + %294 = fmul float %230, %230, !dbg !25 + %295 = fmul float %231, %231, !dbg !25 + %296 = fmul float %232, %232, !dbg !25 + %297 = fmul float %233, %233, !dbg !25 + %298 = fmul float %234, %234, !dbg !25 + %299 = fmul float %235, %235, !dbg !25 + %300 = fmul float %236, %236, !dbg !25 + %301 = fmul float %237, %237, !dbg !25 + %302 = fmul float %238, %238, !dbg !25 + %303 = fmul float %239, %239, !dbg !25 + %304 = fmul float %240, %240, !dbg !25 + %305 = fmul float %241, %241, !dbg !25 + %306 = fmul float %242, %242, !dbg !25 + %307 = fmul float %243, %243, !dbg !25 + %308 = fmul float %244, %244, !dbg !25 + %309 = fmul float %245, %245, !dbg !25 + %310 = fmul float %246, %246, !dbg !25 + %311 = fmul float %247, %247, !dbg !25 + %312 = fmul float %248, %248, !dbg !25 + %313 = fmul float %249, %249, !dbg !25 + %314 = fmul float %250, %250, !dbg !25 + %315 = fmul float %251, %251, !dbg !25 + %316 = fmul float %252, %252, !dbg !25 + %317 = fmul float %253, %253, !dbg !25 + %318 = fmul float %254, %254, !dbg !25 + %319 = fmul float %255, %255, !dbg !25 + %320 = fmul float %256, %256, !dbg !25 + %321 = fmul float %257, %257, !dbg !25 + %322 = fmul float %258, %258, !dbg !25 + %323 = fmul float %259, %259, !dbg !25 + %324 = fmul float %260, %260, !dbg !25 + %325 = fmul float %261, %261, !dbg !25 + %326 = fmul float %262, %262, !dbg !25 + %327 = fmul float %263, %263, !dbg !25 + %328 = fmul float %264, %264, !dbg !25 + %329 = fmul float %265, %265, !dbg !25 + %330 = fmul float %266, %266, !dbg !25 + %331 = fmul float %267, %267, !dbg !25 + %332 = fmul float %268, %268, !dbg !25 + %333 = fmul float %269, %269, !dbg !25 + %334 = fmul float %270, %270, !dbg !25 + %335 = fmul float %271, %271, !dbg !25 + %336 = fmul float %272, %272, !dbg !25 + %337 = fadd float %273, %274, !dbg !26 + %338 = fadd float %275, %337, !dbg !26 + %339 = fadd float %276, %338, !dbg !26 + %340 = fadd float %277, %339, !dbg !26 + %341 = fadd float %278, %340, !dbg !26 + %342 = fadd float %279, %341, !dbg !26 + %343 = fadd float %280, %342, !dbg !26 + %344 = fadd float %281, %282, !dbg !26 + %345 = fadd float %283, %344, !dbg !26 + %346 = fadd float %284, %345, !dbg !26 + %347 = fadd float %285, %346, !dbg !26 + %348 = fadd float %286, %347, !dbg !26 + %349 = fadd float %287, %348, !dbg !26 + %350 = fadd float %288, %349, !dbg !26 + %351 = fadd float %289, %290, !dbg !26 + %352 = fadd float %291, %351, !dbg !26 + %353 = fadd float %292, %352, !dbg !26 + %354 = fadd float %293, %353, !dbg !26 + %355 = fadd float %294, %354, !dbg !26 + %356 = fadd float %295, %355, !dbg !26 + %357 = fadd float %296, %356, !dbg !26 + %358 = fadd float %297, %298, !dbg !26 + %359 = fadd float %299, %358, !dbg !26 + %360 = fadd float %300, %359, !dbg !26 + %361 = fadd float %301, %360, !dbg !26 + %362 = fadd float %302, %361, !dbg !26 + %363 = fadd float %303, %362, !dbg !26 + %364 = fadd float %304, %363, !dbg !26 + %365 = fadd float %305, %306, !dbg !26 + %366 = fadd float %307, %365, !dbg !26 + %367 = fadd float %308, %366, !dbg !26 + %368 = fadd float %309, %367, !dbg !26 + %369 = fadd float %310, %368, !dbg !26 + %370 = fadd float %311, %369, !dbg !26 + %371 = fadd float %312, %370, !dbg !26 + %372 = fadd float %313, %314, !dbg !26 + %373 = fadd float %315, %372, !dbg !26 + %374 = fadd float %316, %373, !dbg !26 + %375 = fadd float %317, %374, !dbg !26 + %376 = fadd float %318, %375, !dbg !26 + %377 = fadd float %319, %376, !dbg !26 + %378 = fadd float %320, %377, !dbg !26 + %379 = fadd float %321, %322, !dbg !26 + %380 = fadd float %323, %379, !dbg !26 + %381 = fadd float %324, %380, !dbg !26 + %382 = fadd float %325, %381, !dbg !26 + %383 = fadd float %326, %382, !dbg !26 + %384 = fadd float %327, %383, !dbg !26 + %385 = fadd float %328, %384, !dbg !26 + %386 = fadd float %329, %330, !dbg !26 + %387 = fadd float %331, %386, !dbg !26 + %388 = fadd float %332, %387, !dbg !26 + %389 = fadd float %333, %388, !dbg !26 + %390 = fadd float %334, %389, !dbg !26 + %391 = fadd float %335, %390, !dbg !26 + %392 = fadd float %336, %391, !dbg !26 + %393 = bitcast float %343 to i32, !dbg !29 + %394 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %393, i32 8, i32 31), !dbg !29 + %395 = bitcast i32 %394 to float, !dbg !29 + %396 = fadd float %343, %395, !dbg !26 + %397 = bitcast float %396 to i32, !dbg !29 + %398 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %397, i32 4, i32 31), !dbg !29 + %399 = bitcast i32 %398 to float, !dbg !29 + %400 = fadd float %396, %399, !dbg !26 + %401 = bitcast float %400 to i32, !dbg !29 + %402 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %401, i32 2, i32 31), !dbg !29 + %403 = bitcast i32 %402 to float, !dbg !29 + %404 = fadd float %400, %403, !dbg !26 + %405 = bitcast float %404 to i32, !dbg !29 + %406 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %405, i32 1, i32 31), !dbg !29 + %407 = bitcast i32 %406 to float, !dbg !29 + %408 = fadd float %404, %407, !dbg !26 + %409 = bitcast float %350 to i32, !dbg !29 + %410 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %409, i32 8, i32 31), !dbg !29 + %411 = bitcast i32 %410 to float, !dbg !29 + %412 = fadd float %350, %411, !dbg !26 + %413 = bitcast float %412 to i32, !dbg !29 + %414 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %413, i32 4, i32 31), !dbg !29 + %415 = bitcast i32 %414 to float, !dbg !29 + %416 = fadd float %412, %415, !dbg !26 + %417 = bitcast float %416 to i32, !dbg !29 + %418 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %417, i32 2, i32 31), !dbg !29 + %419 = bitcast i32 %418 to float, !dbg !29 + %420 = fadd float %416, %419, !dbg !26 + %421 = bitcast float %420 to i32, !dbg !29 + %422 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %421, i32 1, i32 31), !dbg !29 + %423 = bitcast i32 %422 to float, !dbg !29 + %424 = fadd float %420, %423, !dbg !26 + %425 = bitcast float %357 to i32, !dbg !29 + %426 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %425, i32 8, i32 31), !dbg !29 + %427 = bitcast i32 %426 to float, !dbg !29 + %428 = fadd float %357, %427, !dbg !26 + %429 = bitcast float %428 to i32, !dbg !29 + %430 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %429, i32 4, i32 31), !dbg !29 + %431 = bitcast i32 %430 to float, !dbg !29 + %432 = fadd float %428, %431, !dbg !26 + %433 = bitcast float %432 to i32, !dbg !29 + %434 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %433, i32 2, i32 31), !dbg !29 + %435 = bitcast i32 %434 to float, !dbg !29 + %436 = fadd float %432, %435, !dbg !26 + %437 = bitcast float %436 to i32, !dbg !29 + %438 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %437, i32 1, i32 31), !dbg !29 + %439 = bitcast i32 %438 to float, !dbg !29 + %440 = fadd float %436, %439, !dbg !26 + %441 = bitcast float %364 to i32, !dbg !29 + %442 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %441, i32 8, i32 31), !dbg !29 + %443 = bitcast i32 %442 to float, !dbg !29 + %444 = fadd float %364, %443, !dbg !26 + %445 = bitcast float %444 to i32, !dbg !29 + %446 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %445, i32 4, i32 31), !dbg !29 + %447 = bitcast i32 %446 to float, !dbg !29 + %448 = fadd float %444, %447, !dbg !26 + %449 = bitcast float %448 to i32, !dbg !29 + %450 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %449, i32 2, i32 31), !dbg !29 + %451 = bitcast i32 %450 to float, !dbg !29 + %452 = fadd float %448, %451, !dbg !26 + %453 = bitcast float %452 to i32, !dbg !29 + %454 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %453, i32 1, i32 31), !dbg !29 + %455 = bitcast i32 %454 to float, !dbg !29 + %456 = fadd float %452, %455, !dbg !26 + %457 = bitcast float %371 to i32, !dbg !29 + %458 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %457, i32 8, i32 31), !dbg !29 + %459 = bitcast i32 %458 to float, !dbg !29 + %460 = fadd float %371, %459, !dbg !26 + %461 = bitcast float %460 to i32, !dbg !29 + %462 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %461, i32 4, i32 31), !dbg !29 + %463 = bitcast i32 %462 to float, !dbg !29 + %464 = fadd float %460, %463, !dbg !26 + %465 = bitcast float %464 to i32, !dbg !29 + %466 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %465, i32 2, i32 31), !dbg !29 + %467 = bitcast i32 %466 to float, !dbg !29 + %468 = fadd float %464, %467, !dbg !26 + %469 = bitcast float %468 to i32, !dbg !29 + %470 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %469, i32 1, i32 31), !dbg !29 + %471 = bitcast i32 %470 to float, !dbg !29 + %472 = fadd float %468, %471, !dbg !26 + %473 = bitcast float %378 to i32, !dbg !29 + %474 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %473, i32 8, i32 31), !dbg !29 + %475 = bitcast i32 %474 to float, !dbg !29 + %476 = fadd float %378, %475, !dbg !26 + %477 = bitcast float %476 to i32, !dbg !29 + %478 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %477, i32 4, i32 31), !dbg !29 + %479 = bitcast i32 %478 to float, !dbg !29 + %480 = fadd float %476, %479, !dbg !26 + %481 = bitcast float %480 to i32, !dbg !29 + %482 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %481, i32 2, i32 31), !dbg !29 + %483 = bitcast i32 %482 to float, !dbg !29 + %484 = fadd float %480, %483, !dbg !26 + %485 = bitcast float %484 to i32, !dbg !29 + %486 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %485, i32 1, i32 31), !dbg !29 + %487 = bitcast i32 %486 to float, !dbg !29 + %488 = fadd float %484, %487, !dbg !26 + %489 = bitcast float %385 to i32, !dbg !29 + %490 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %489, i32 8, i32 31), !dbg !29 + %491 = bitcast i32 %490 to float, !dbg !29 + %492 = fadd float %385, %491, !dbg !26 + %493 = bitcast float %492 to i32, !dbg !29 + %494 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %493, i32 4, i32 31), !dbg !29 + %495 = bitcast i32 %494 to float, !dbg !29 + %496 = fadd float %492, %495, !dbg !26 + %497 = bitcast float %496 to i32, !dbg !29 + %498 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %497, i32 2, i32 31), !dbg !29 + %499 = bitcast i32 %498 to float, !dbg !29 + %500 = fadd float %496, %499, !dbg !26 + %501 = bitcast float %500 to i32, !dbg !29 + %502 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %501, i32 1, i32 31), !dbg !29 + %503 = bitcast i32 %502 to float, !dbg !29 + %504 = fadd float %500, %503, !dbg !26 + %505 = bitcast float %392 to i32, !dbg !29 + %506 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %505, i32 8, i32 31), !dbg !29 + %507 = bitcast i32 %506 to float, !dbg !29 + %508 = fadd float %392, %507, !dbg !26 + %509 = bitcast float %508 to i32, !dbg !29 + %510 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %509, i32 4, i32 31), !dbg !29 + %511 = bitcast i32 %510 to float, !dbg !29 + %512 = fadd float %508, %511, !dbg !26 + %513 = bitcast float %512 to i32, !dbg !29 + %514 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %513, i32 2, i32 31), !dbg !29 + %515 = bitcast i32 %514 to float, !dbg !29 + %516 = fadd float %512, %515, !dbg !26 + %517 = bitcast float %516 to i32, !dbg !29 + %518 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %517, i32 1, i32 31), !dbg !29 + %519 = bitcast i32 %518 to float, !dbg !29 + %520 = fadd float %516, %519, !dbg !26 + %521 = fadd float %3, %408, !dbg !32 + %522 = fadd float %3, %424, !dbg !32 + %523 = fadd float %3, %440, !dbg !32 + %524 = fadd float %3, %456, !dbg !32 + %525 = fadd float %3, %472, !dbg !32 + %526 = fadd float %3, %488, !dbg !32 + %527 = fadd float %3, %504, !dbg !32 + %528 = fadd float %3, %520, !dbg !32 + %529 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i = icmp eq i32 %529, 0, !dbg !33 + %530 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i = icmp eq i32 %530, 0, !dbg !33 + br i1 %.not.i, label %536, label %531, !dbg !33 + +531: ; preds = %7 + br i1 %.not1.i, label %534, label %532, !dbg !33 + +532: ; preds = %531 + %533 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %521) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +534: ; preds = %531 + %535 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %521) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +536: ; preds = %7 + br i1 %.not1.i, label %539, label %537, !dbg !33 + +537: ; preds = %536 + %538 = tail call float @llvm.nvvm.sqrt.rn.f(float %521) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +539: ; preds = %536 + %540 = tail call float @llvm.nvvm.sqrt.approx.f(float %521) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +__nv_sqrtf.exit: ; preds = %532, %534, %537, %539 + %.0.i = phi float [ %533, %532 ], [ %535, %534 ], [ %538, %537 ], [ %540, %539 ], !dbg !33 + %541 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i1 = icmp eq i32 %541, 0, !dbg !33 + %542 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i4 = icmp eq i32 %542, 0, !dbg !33 + br i1 %.not.i1, label %548, label %543, !dbg !33 + +543: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %546, label %544, !dbg !33 + +544: ; preds = %543 + %545 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %522) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +546: ; preds = %543 + %547 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %522) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +548: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %551, label %549, !dbg !33 + +549: ; preds = %548 + %550 = tail call float @llvm.nvvm.sqrt.rn.f(float %522) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +551: ; preds = %548 + %552 = tail call float @llvm.nvvm.sqrt.approx.f(float %522) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +__nv_sqrtf.exit5: ; preds = %544, %546, %549, %551 + %.0.i3 = phi float [ %545, %544 ], [ %547, %546 ], [ %550, %549 ], [ %552, %551 ], !dbg !33 + %553 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i6 = icmp eq i32 %553, 0, !dbg !33 + %554 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i9 = icmp eq i32 %554, 0, !dbg !33 + br i1 %.not.i6, label %560, label %555, !dbg !33 + +555: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %558, label %556, !dbg !33 + +556: ; preds = %555 + %557 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %523) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +558: ; preds = %555 + %559 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %523) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +560: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %563, label %561, !dbg !33 + +561: ; preds = %560 + %562 = tail call float @llvm.nvvm.sqrt.rn.f(float %523) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +563: ; preds = %560 + %564 = tail call float @llvm.nvvm.sqrt.approx.f(float %523) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +__nv_sqrtf.exit10: ; preds = %556, %558, %561, %563 + %.0.i8 = phi float [ %557, %556 ], [ %559, %558 ], [ %562, %561 ], [ %564, %563 ], !dbg !33 + %565 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i11 = icmp eq i32 %565, 0, !dbg !33 + %566 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i14 = icmp eq i32 %566, 0, !dbg !33 + br i1 %.not.i11, label %572, label %567, !dbg !33 + +567: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %570, label %568, !dbg !33 + +568: ; preds = %567 + %569 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %524) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +570: ; preds = %567 + %571 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %524) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +572: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %575, label %573, !dbg !33 + +573: ; preds = %572 + %574 = tail call float @llvm.nvvm.sqrt.rn.f(float %524) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +575: ; preds = %572 + %576 = tail call float @llvm.nvvm.sqrt.approx.f(float %524) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +__nv_sqrtf.exit15: ; preds = %568, %570, %573, %575 + %.0.i13 = phi float [ %569, %568 ], [ %571, %570 ], [ %574, %573 ], [ %576, %575 ], !dbg !33 + %577 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i16 = icmp eq i32 %577, 0, !dbg !33 + %578 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i19 = icmp eq i32 %578, 0, !dbg !33 + br i1 %.not.i16, label %584, label %579, !dbg !33 + +579: ; preds = %__nv_sqrtf.exit15 + br i1 %.not1.i19, label %582, label %580, !dbg !33 + +580: ; preds = %579 + %581 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %525) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +582: ; preds = %579 + %583 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %525) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +584: ; preds = %__nv_sqrtf.exit15 + br i1 %.not1.i19, label %587, label %585, !dbg !33 + +585: ; preds = %584 + %586 = tail call float @llvm.nvvm.sqrt.rn.f(float %525) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +587: ; preds = %584 + %588 = tail call float @llvm.nvvm.sqrt.approx.f(float %525) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +__nv_sqrtf.exit20: ; preds = %580, %582, %585, %587 + %.0.i18 = phi float [ %581, %580 ], [ %583, %582 ], [ %586, %585 ], [ %588, %587 ], !dbg !33 + %589 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i21 = icmp eq i32 %589, 0, !dbg !33 + %590 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i24 = icmp eq i32 %590, 0, !dbg !33 + br i1 %.not.i21, label %596, label %591, !dbg !33 + +591: ; preds = %__nv_sqrtf.exit20 + br i1 %.not1.i24, label %594, label %592, !dbg !33 + +592: ; preds = %591 + %593 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %526) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +594: ; preds = %591 + %595 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %526) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +596: ; preds = %__nv_sqrtf.exit20 + br i1 %.not1.i24, label %599, label %597, !dbg !33 + +597: ; preds = %596 + %598 = tail call float @llvm.nvvm.sqrt.rn.f(float %526) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +599: ; preds = %596 + %600 = tail call float @llvm.nvvm.sqrt.approx.f(float %526) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +__nv_sqrtf.exit25: ; preds = %592, %594, %597, %599 + %.0.i23 = phi float [ %593, %592 ], [ %595, %594 ], [ %598, %597 ], [ %600, %599 ], !dbg !33 + %601 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i26 = icmp eq i32 %601, 0, !dbg !33 + %602 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i29 = icmp eq i32 %602, 0, !dbg !33 + br i1 %.not.i26, label %608, label %603, !dbg !33 + +603: ; preds = %__nv_sqrtf.exit25 + br i1 %.not1.i29, label %606, label %604, !dbg !33 + +604: ; preds = %603 + %605 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %527) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +606: ; preds = %603 + %607 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %527) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +608: ; preds = %__nv_sqrtf.exit25 + br i1 %.not1.i29, label %611, label %609, !dbg !33 + +609: ; preds = %608 + %610 = tail call float @llvm.nvvm.sqrt.rn.f(float %527) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +611: ; preds = %608 + %612 = tail call float @llvm.nvvm.sqrt.approx.f(float %527) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +__nv_sqrtf.exit30: ; preds = %604, %606, %609, %611 + %.0.i28 = phi float [ %605, %604 ], [ %607, %606 ], [ %610, %609 ], [ %612, %611 ], !dbg !33 + %613 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i31 = icmp eq i32 %613, 0, !dbg !33 + %614 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i34 = icmp eq i32 %614, 0, !dbg !33 + br i1 %.not.i31, label %620, label %615, !dbg !33 + +615: ; preds = %__nv_sqrtf.exit30 + br i1 %.not1.i34, label %618, label %616, !dbg !33 + +616: ; preds = %615 + %617 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %528) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +618: ; preds = %615 + %619 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %528) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +620: ; preds = %__nv_sqrtf.exit30 + br i1 %.not1.i34, label %623, label %621, !dbg !33 + +621: ; preds = %620 + %622 = tail call float @llvm.nvvm.sqrt.rn.f(float %528) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +623: ; preds = %620 + %624 = tail call float @llvm.nvvm.sqrt.approx.f(float %528) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +__nv_sqrtf.exit35: ; preds = %616, %618, %621, %623 + %.0.i33 = phi float [ %617, %616 ], [ %619, %618 ], [ %622, %621 ], [ %624, %623 ], !dbg !33 + %625 = and i32 %11, 31, !dbg !16 + %626 = zext nneg i32 %625 to i64, !dbg !17 + %627 = or disjoint i64 %10, %626, !dbg !17 + %628 = getelementptr float, ptr addrspace(1) %2, i64 %627, !dbg !34 + %629 = getelementptr bfloat, ptr addrspace(1) %1, i64 %55, !dbg !35 + %630 = getelementptr bfloat, ptr addrspace(1) %629, i64 %64, !dbg !36 + %631 = getelementptr bfloat, ptr addrspace(1) %1, i64 %54, !dbg !35 + %632 = getelementptr bfloat, ptr addrspace(1) %631, i64 %64, !dbg !36 + %633 = getelementptr bfloat, ptr addrspace(1) %1, i64 %53, !dbg !35 + %634 = getelementptr bfloat, ptr addrspace(1) %633, i64 %64, !dbg !36 + %635 = getelementptr bfloat, ptr addrspace(1) %1, i64 %52, !dbg !35 + %636 = getelementptr bfloat, ptr addrspace(1) %635, i64 %64, !dbg !36 + %637 = getelementptr bfloat, ptr addrspace(1) %1, i64 %51, !dbg !35 + %638 = getelementptr bfloat, ptr addrspace(1) %637, i64 %64, !dbg !36 + %639 = getelementptr bfloat, ptr addrspace(1) %1, i64 %50, !dbg !35 + %640 = getelementptr bfloat, ptr addrspace(1) %639, i64 %64, !dbg !36 + %641 = getelementptr bfloat, ptr addrspace(1) %1, i64 %49, !dbg !35 + %642 = getelementptr bfloat, ptr addrspace(1) %641, i64 %64, !dbg !36 + %643 = getelementptr bfloat, ptr addrspace(1) %1, i64 %48, !dbg !35 + %644 = getelementptr bfloat, ptr addrspace(1) %643, i64 %64, !dbg !36 + %645 = icmp slt i64 %627, %37, !dbg !18 + %646 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i), !dbg !37 + %647 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i3), !dbg !37 + %648 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i8), !dbg !37 + %649 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i13), !dbg !37 + %650 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i18), !dbg !37 + %651 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i23), !dbg !37 + %652 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i28), !dbg !37 + %653 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i33), !dbg !37 + %654 = fmul float %646, %209, !dbg !38 + %655 = fmul float %646, %210, !dbg !38 + %656 = fmul float %646, %211, !dbg !38 + %657 = fmul float %646, %212, !dbg !38 + %658 = fmul float %646, %213, !dbg !38 + %659 = fmul float %646, %214, !dbg !38 + %660 = fmul float %646, %215, !dbg !38 + %661 = fmul float %646, %216, !dbg !38 + %662 = fmul float %647, %217, !dbg !38 + %663 = fmul float %647, %218, !dbg !38 + %664 = fmul float %647, %219, !dbg !38 + %665 = fmul float %647, %220, !dbg !38 + %666 = fmul float %647, %221, !dbg !38 + %667 = fmul float %647, %222, !dbg !38 + %668 = fmul float %647, %223, !dbg !38 + %669 = fmul float %647, %224, !dbg !38 + %670 = fmul float %648, %225, !dbg !38 + %671 = fmul float %648, %226, !dbg !38 + %672 = fmul float %648, %227, !dbg !38 + %673 = fmul float %648, %228, !dbg !38 + %674 = fmul float %648, %229, !dbg !38 + %675 = fmul float %648, %230, !dbg !38 + %676 = fmul float %648, %231, !dbg !38 + %677 = fmul float %648, %232, !dbg !38 + %678 = fmul float %649, %233, !dbg !38 + %679 = fmul float %649, %234, !dbg !38 + %680 = fmul float %649, %235, !dbg !38 + %681 = fmul float %649, %236, !dbg !38 + %682 = fmul float %649, %237, !dbg !38 + %683 = fmul float %649, %238, !dbg !38 + %684 = fmul float %649, %239, !dbg !38 + %685 = fmul float %649, %240, !dbg !38 + %686 = fmul float %650, %241, !dbg !38 + %687 = fmul float %650, %242, !dbg !38 + %688 = fmul float %650, %243, !dbg !38 + %689 = fmul float %650, %244, !dbg !38 + %690 = fmul float %650, %245, !dbg !38 + %691 = fmul float %650, %246, !dbg !38 + %692 = fmul float %650, %247, !dbg !38 + %693 = fmul float %650, %248, !dbg !38 + %694 = fmul float %651, %249, !dbg !38 + %695 = fmul float %651, %250, !dbg !38 + %696 = fmul float %651, %251, !dbg !38 + %697 = fmul float %651, %252, !dbg !38 + %698 = fmul float %651, %253, !dbg !38 + %699 = fmul float %651, %254, !dbg !38 + %700 = fmul float %651, %255, !dbg !38 + %701 = fmul float %651, %256, !dbg !38 + %702 = fmul float %652, %257, !dbg !38 + %703 = fmul float %652, %258, !dbg !38 + %704 = fmul float %652, %259, !dbg !38 + %705 = fmul float %652, %260, !dbg !38 + %706 = fmul float %652, %261, !dbg !38 + %707 = fmul float %652, %262, !dbg !38 + %708 = fmul float %652, %263, !dbg !38 + %709 = fmul float %652, %264, !dbg !38 + %710 = fmul float %653, %265, !dbg !38 + %711 = fmul float %653, %266, !dbg !38 + %712 = fmul float %653, %267, !dbg !38 + %713 = fmul float %653, %268, !dbg !38 + %714 = fmul float %653, %269, !dbg !38 + %715 = fmul float %653, %270, !dbg !38 + %716 = fmul float %653, %271, !dbg !38 + %717 = fmul float %653, %272, !dbg !38 + %718 = fptrunc float %654 to bfloat, !dbg !39 + %719 = fptrunc float %655 to bfloat, !dbg !39 + %720 = fptrunc float %656 to bfloat, !dbg !39 + %721 = fptrunc float %657 to bfloat, !dbg !39 + %722 = fptrunc float %658 to bfloat, !dbg !39 + %723 = fptrunc float %659 to bfloat, !dbg !39 + %724 = fptrunc float %660 to bfloat, !dbg !39 + %725 = fptrunc float %661 to bfloat, !dbg !39 + %726 = fptrunc float %662 to bfloat, !dbg !39 + %727 = fptrunc float %663 to bfloat, !dbg !39 + %728 = fptrunc float %664 to bfloat, !dbg !39 + %729 = fptrunc float %665 to bfloat, !dbg !39 + %730 = fptrunc float %666 to bfloat, !dbg !39 + %731 = fptrunc float %667 to bfloat, !dbg !39 + %732 = fptrunc float %668 to bfloat, !dbg !39 + %733 = fptrunc float %669 to bfloat, !dbg !39 + %734 = fptrunc float %670 to bfloat, !dbg !39 + %735 = fptrunc float %671 to bfloat, !dbg !39 + %736 = fptrunc float %672 to bfloat, !dbg !39 + %737 = fptrunc float %673 to bfloat, !dbg !39 + %738 = fptrunc float %674 to bfloat, !dbg !39 + %739 = fptrunc float %675 to bfloat, !dbg !39 + %740 = fptrunc float %676 to bfloat, !dbg !39 + %741 = fptrunc float %677 to bfloat, !dbg !39 + %742 = fptrunc float %678 to bfloat, !dbg !39 + %743 = fptrunc float %679 to bfloat, !dbg !39 + %744 = fptrunc float %680 to bfloat, !dbg !39 + %745 = fptrunc float %681 to bfloat, !dbg !39 + %746 = fptrunc float %682 to bfloat, !dbg !39 + %747 = fptrunc float %683 to bfloat, !dbg !39 + %748 = fptrunc float %684 to bfloat, !dbg !39 + %749 = fptrunc float %685 to bfloat, !dbg !39 + %750 = fptrunc float %686 to bfloat, !dbg !39 + %751 = fptrunc float %687 to bfloat, !dbg !39 + %752 = fptrunc float %688 to bfloat, !dbg !39 + %753 = fptrunc float %689 to bfloat, !dbg !39 + %754 = fptrunc float %690 to bfloat, !dbg !39 + %755 = fptrunc float %691 to bfloat, !dbg !39 + %756 = fptrunc float %692 to bfloat, !dbg !39 + %757 = fptrunc float %693 to bfloat, !dbg !39 + %758 = fptrunc float %694 to bfloat, !dbg !39 + %759 = fptrunc float %695 to bfloat, !dbg !39 + %760 = fptrunc float %696 to bfloat, !dbg !39 + %761 = fptrunc float %697 to bfloat, !dbg !39 + %762 = fptrunc float %698 to bfloat, !dbg !39 + %763 = fptrunc float %699 to bfloat, !dbg !39 + %764 = fptrunc float %700 to bfloat, !dbg !39 + %765 = fptrunc float %701 to bfloat, !dbg !39 + %766 = fptrunc float %702 to bfloat, !dbg !39 + %767 = fptrunc float %703 to bfloat, !dbg !39 + %768 = fptrunc float %704 to bfloat, !dbg !39 + %769 = fptrunc float %705 to bfloat, !dbg !39 + %770 = fptrunc float %706 to bfloat, !dbg !39 + %771 = fptrunc float %707 to bfloat, !dbg !39 + %772 = fptrunc float %708 to bfloat, !dbg !39 + %773 = fptrunc float %709 to bfloat, !dbg !39 + %774 = fptrunc float %710 to bfloat, !dbg !39 + %775 = fptrunc float %711 to bfloat, !dbg !39 + %776 = fptrunc float %712 to bfloat, !dbg !39 + %777 = fptrunc float %713 to bfloat, !dbg !39 + %778 = fptrunc float %714 to bfloat, !dbg !39 + %779 = fptrunc float %715 to bfloat, !dbg !39 + %780 = fptrunc float %716 to bfloat, !dbg !39 + %781 = fptrunc float %717 to bfloat, !dbg !39 + %782 = insertelement <2 x bfloat> poison, bfloat %718, i64 0, !dbg !40 + %783 = insertelement <2 x bfloat> %782, bfloat %719, i64 1, !dbg !40 + %784 = bitcast <2 x bfloat> %783 to i32, !dbg !40 + %785 = insertelement <2 x bfloat> poison, bfloat %720, i64 0, !dbg !40 + %786 = insertelement <2 x bfloat> %785, bfloat %721, i64 1, !dbg !40 + %787 = bitcast <2 x bfloat> %786 to i32, !dbg !40 + %788 = insertelement <2 x bfloat> poison, bfloat %722, i64 0, !dbg !40 + %789 = insertelement <2 x bfloat> %788, bfloat %723, i64 1, !dbg !40 + %790 = bitcast <2 x bfloat> %789 to i32, !dbg !40 + %791 = insertelement <2 x bfloat> poison, bfloat %724, i64 0, !dbg !40 + %792 = insertelement <2 x bfloat> %791, bfloat %725, i64 1, !dbg !40 + %793 = bitcast <2 x bfloat> %792 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %784, i32 %787, i32 %790, i32 %793, ptr addrspace(1) %644, i1 %38) #6, !dbg !40 + %794 = insertelement <2 x bfloat> poison, bfloat %726, i64 0, !dbg !40 + %795 = insertelement <2 x bfloat> %794, bfloat %727, i64 1, !dbg !40 + %796 = bitcast <2 x bfloat> %795 to i32, !dbg !40 + %797 = insertelement <2 x bfloat> poison, bfloat %728, i64 0, !dbg !40 + %798 = insertelement <2 x bfloat> %797, bfloat %729, i64 1, !dbg !40 + %799 = bitcast <2 x bfloat> %798 to i32, !dbg !40 + %800 = insertelement <2 x bfloat> poison, bfloat %730, i64 0, !dbg !40 + %801 = insertelement <2 x bfloat> %800, bfloat %731, i64 1, !dbg !40 + %802 = bitcast <2 x bfloat> %801 to i32, !dbg !40 + %803 = insertelement <2 x bfloat> poison, bfloat %732, i64 0, !dbg !40 + %804 = insertelement <2 x bfloat> %803, bfloat %733, i64 1, !dbg !40 + %805 = bitcast <2 x bfloat> %804 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %796, i32 %799, i32 %802, i32 %805, ptr addrspace(1) %642, i1 %39) #6, !dbg !40 + %806 = insertelement <2 x bfloat> poison, bfloat %734, i64 0, !dbg !40 + %807 = insertelement <2 x bfloat> %806, bfloat %735, i64 1, !dbg !40 + %808 = bitcast <2 x bfloat> %807 to i32, !dbg !40 + %809 = insertelement <2 x bfloat> poison, bfloat %736, i64 0, !dbg !40 + %810 = insertelement <2 x bfloat> %809, bfloat %737, i64 1, !dbg !40 + %811 = bitcast <2 x bfloat> %810 to i32, !dbg !40 + %812 = insertelement <2 x bfloat> poison, bfloat %738, i64 0, !dbg !40 + %813 = insertelement <2 x bfloat> %812, bfloat %739, i64 1, !dbg !40 + %814 = bitcast <2 x bfloat> %813 to i32, !dbg !40 + %815 = insertelement <2 x bfloat> poison, bfloat %740, i64 0, !dbg !40 + %816 = insertelement <2 x bfloat> %815, bfloat %741, i64 1, !dbg !40 + %817 = bitcast <2 x bfloat> %816 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %808, i32 %811, i32 %814, i32 %817, ptr addrspace(1) %640, i1 %40) #6, !dbg !40 + %818 = insertelement <2 x bfloat> poison, bfloat %742, i64 0, !dbg !40 + %819 = insertelement <2 x bfloat> %818, bfloat %743, i64 1, !dbg !40 + %820 = bitcast <2 x bfloat> %819 to i32, !dbg !40 + %821 = insertelement <2 x bfloat> poison, bfloat %744, i64 0, !dbg !40 + %822 = insertelement <2 x bfloat> %821, bfloat %745, i64 1, !dbg !40 + %823 = bitcast <2 x bfloat> %822 to i32, !dbg !40 + %824 = insertelement <2 x bfloat> poison, bfloat %746, i64 0, !dbg !40 + %825 = insertelement <2 x bfloat> %824, bfloat %747, i64 1, !dbg !40 + %826 = bitcast <2 x bfloat> %825 to i32, !dbg !40 + %827 = insertelement <2 x bfloat> poison, bfloat %748, i64 0, !dbg !40 + %828 = insertelement <2 x bfloat> %827, bfloat %749, i64 1, !dbg !40 + %829 = bitcast <2 x bfloat> %828 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %820, i32 %823, i32 %826, i32 %829, ptr addrspace(1) %638, i1 %41) #6, !dbg !40 + %830 = insertelement <2 x bfloat> poison, bfloat %750, i64 0, !dbg !40 + %831 = insertelement <2 x bfloat> %830, bfloat %751, i64 1, !dbg !40 + %832 = bitcast <2 x bfloat> %831 to i32, !dbg !40 + %833 = insertelement <2 x bfloat> poison, bfloat %752, i64 0, !dbg !40 + %834 = insertelement <2 x bfloat> %833, bfloat %753, i64 1, !dbg !40 + %835 = bitcast <2 x bfloat> %834 to i32, !dbg !40 + %836 = insertelement <2 x bfloat> poison, bfloat %754, i64 0, !dbg !40 + %837 = insertelement <2 x bfloat> %836, bfloat %755, i64 1, !dbg !40 + %838 = bitcast <2 x bfloat> %837 to i32, !dbg !40 + %839 = insertelement <2 x bfloat> poison, bfloat %756, i64 0, !dbg !40 + %840 = insertelement <2 x bfloat> %839, bfloat %757, i64 1, !dbg !40 + %841 = bitcast <2 x bfloat> %840 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %832, i32 %835, i32 %838, i32 %841, ptr addrspace(1) %636, i1 %42) #6, !dbg !40 + %842 = insertelement <2 x bfloat> poison, bfloat %758, i64 0, !dbg !40 + %843 = insertelement <2 x bfloat> %842, bfloat %759, i64 1, !dbg !40 + %844 = bitcast <2 x bfloat> %843 to i32, !dbg !40 + %845 = insertelement <2 x bfloat> poison, bfloat %760, i64 0, !dbg !40 + %846 = insertelement <2 x bfloat> %845, bfloat %761, i64 1, !dbg !40 + %847 = bitcast <2 x bfloat> %846 to i32, !dbg !40 + %848 = insertelement <2 x bfloat> poison, bfloat %762, i64 0, !dbg !40 + %849 = insertelement <2 x bfloat> %848, bfloat %763, i64 1, !dbg !40 + %850 = bitcast <2 x bfloat> %849 to i32, !dbg !40 + %851 = insertelement <2 x bfloat> poison, bfloat %764, i64 0, !dbg !40 + %852 = insertelement <2 x bfloat> %851, bfloat %765, i64 1, !dbg !40 + %853 = bitcast <2 x bfloat> %852 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %844, i32 %847, i32 %850, i32 %853, ptr addrspace(1) %634, i1 %43) #6, !dbg !40 + %854 = insertelement <2 x bfloat> poison, bfloat %766, i64 0, !dbg !40 + %855 = insertelement <2 x bfloat> %854, bfloat %767, i64 1, !dbg !40 + %856 = bitcast <2 x bfloat> %855 to i32, !dbg !40 + %857 = insertelement <2 x bfloat> poison, bfloat %768, i64 0, !dbg !40 + %858 = insertelement <2 x bfloat> %857, bfloat %769, i64 1, !dbg !40 + %859 = bitcast <2 x bfloat> %858 to i32, !dbg !40 + %860 = insertelement <2 x bfloat> poison, bfloat %770, i64 0, !dbg !40 + %861 = insertelement <2 x bfloat> %860, bfloat %771, i64 1, !dbg !40 + %862 = bitcast <2 x bfloat> %861 to i32, !dbg !40 + %863 = insertelement <2 x bfloat> poison, bfloat %772, i64 0, !dbg !40 + %864 = insertelement <2 x bfloat> %863, bfloat %773, i64 1, !dbg !40 + %865 = bitcast <2 x bfloat> %864 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %856, i32 %859, i32 %862, i32 %865, ptr addrspace(1) %632, i1 %44) #6, !dbg !40 + %866 = insertelement <2 x bfloat> poison, bfloat %774, i64 0, !dbg !40 + %867 = insertelement <2 x bfloat> %866, bfloat %775, i64 1, !dbg !40 + %868 = bitcast <2 x bfloat> %867 to i32, !dbg !40 + %869 = insertelement <2 x bfloat> poison, bfloat %776, i64 0, !dbg !40 + %870 = insertelement <2 x bfloat> %869, bfloat %777, i64 1, !dbg !40 + %871 = bitcast <2 x bfloat> %870 to i32, !dbg !40 + %872 = insertelement <2 x bfloat> poison, bfloat %778, i64 0, !dbg !40 + %873 = insertelement <2 x bfloat> %872, bfloat %779, i64 1, !dbg !40 + %874 = bitcast <2 x bfloat> %873 to i32, !dbg !40 + %875 = insertelement <2 x bfloat> poison, bfloat %780, i64 0, !dbg !40 + %876 = insertelement <2 x bfloat> %875, bfloat %781, i64 1, !dbg !40 + %877 = bitcast <2 x bfloat> %876 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %868, i32 %871, i32 %874, i32 %877, ptr addrspace(1) %630, i1 %45) #6, !dbg !40 + %878 = shl nuw nsw i32 %12, 1, !dbg !41 + %879 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %878, !dbg !41 + %880 = bitcast float %646 to i32, !dbg !41 + %881 = bitcast float %647 to i32, !dbg !41 + %882 = bitcast float %648 to i32, !dbg !41 + %883 = bitcast float %649 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %879, i32 %880, i32 %881, i32 %882, i32 %883) #6, !dbg !41 + %884 = getelementptr inbounds nuw i8, ptr addrspace(3) %879, i32 16, !dbg !41 + %885 = bitcast float %650 to i32, !dbg !41 + %886 = bitcast float %651 to i32, !dbg !41 + %887 = bitcast float %652 to i32, !dbg !41 + %888 = bitcast float %653 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %884, i32 %885, i32 %886, i32 %887, i32 %888) #6, !dbg !41 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !41 + %889 = shl nuw nsw i32 %11, 5, !dbg !41 + %890 = and i32 %889, 96, !dbg !41 + %891 = and i32 %11, 28, !dbg !41 + %892 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %890, !dbg !41 + %893 = getelementptr inbounds nuw i8, ptr addrspace(3) %892, i32 %891, !dbg !41 + %894 = load i32, ptr addrspace(3) %893, align 4, !dbg !41 + %895 = and i32 %11, 32, !dbg !41 + %896 = icmp eq i32 %895, 0, !dbg !41 + %897 = and i1 %896, %645, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %894, ptr addrspace(1) %628, i1 %897) #6, !dbg !41 + ret void, !dbg !42 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.div.full(float, float) #3 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #4 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #5 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.f(float) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="64" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) } +attributes #4 = { convergent nocallback nounwind } +attributes #5 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #6 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "l2norm_fwd_kernel", linkageName: "l2norm_fwd_kernel", scope: !1, file: !1, line: 69, type: !6, scopeLine: 69, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !10, !11, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 70, column: 24, scope: !5) +!14 = !DILocation(line: 70, column: 30, scope: !5) +!15 = !DILocation(line: 71, column: 16, scope: !5) +!16 = !DILocation(line: 71, column: 34, scope: !5) +!17 = !DILocation(line: 71, column: 21, scope: !5) +!18 = !DILocation(line: 73, column: 16, scope: !5) +!19 = !DILocation(line: 74, column: 30, scope: !5) +!20 = !DILocation(line: 75, column: 29, scope: !5) +!21 = !DILocation(line: 75, column: 14, scope: !5) +!22 = !DILocation(line: 75, column: 33, scope: !5) +!23 = !DILocation(line: 79, column: 18, scope: !5) +!24 = !DILocation(line: 79, column: 47, scope: !5) +!25 = !DILocation(line: 80, column: 38, scope: !5) +!26 = !DILocation(line: 263, column: 15, scope: !27, inlinedAt: !29) +!27 = distinct !DILexicalBlockFile(scope: !5, file: !28, discriminator: 0) +!28 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!29 = !DILocation(line: 293, column: 36, scope: !27, inlinedAt: !30) +!30 = !DILocation(line: 80, column: 43, scope: !31) +!31 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!32 = !DILocation(line: 80, column: 48, scope: !5) +!33 = !DILocation(line: 80, column: 25, scope: !5) +!34 = !DILocation(line: 77, column: 20, scope: !5) +!35 = !DILocation(line: 76, column: 14, scope: !5) +!36 = !DILocation(line: 76, column: 33, scope: !5) +!37 = !DILocation(line: 80, column: 17, scope: !5) +!38 = !DILocation(line: 81, column: 16, scope: !5) +!39 = !DILocation(line: 83, column: 25, scope: !5) +!40 = !DILocation(line: 83, column: 18, scope: !5) +!41 = !DILocation(line: 84, column: 21, scope: !5) +!42 = !DILocation(line: 84, column: 4, scope: !5) diff --git a/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.ptx b/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..bf9f670baccbc9ea04a140b39905f33c68561db6 --- /dev/null +++ b/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.ptx @@ -0,0 +1,966 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_fwd_kernel // -- Begin function l2norm_fwd_kernel +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; +.global .align 1 .b8 _$_str_$_2[17] = {95, 95, 67, 85, 68, 65, 95, 80, 82, 69, 67, 95, 83, 81, 82, 84}; + // @l2norm_fwd_kernel +.visible .entry l2norm_fwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_2, + .param .f32 l2norm_fwd_kernel_param_3, + .param .u32 l2norm_fwd_kernel_param_4, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_6 +) +.reqntid 64 +{ + .reg .pred %p<12>; + .reg .b16 %rs<65>; + .reg .b32 %r<373>; + .reg .b64 %rd<67>; + .loc 1 69 0 // l2norm.py:69:0 +$L__func_begin0: + .loc 1 69 0 // l2norm.py:69:0 + +// %bb.0: // %__nv_sqrtf.exit + ld.param.b64 %rd18, [l2norm_fwd_kernel_param_0]; + ld.param.b64 %rd19, [l2norm_fwd_kernel_param_1]; +$L__tmp0: + .loc 1 70 24 // l2norm.py:70:24 + mov.u32 %r77, %ctaid.x; + ld.param.b64 %rd20, [l2norm_fwd_kernel_param_2]; + .loc 1 71 16 // l2norm.py:71:16 + mul.wide.u32 %rd21, %r77, 32; + ld.param.b32 %r78, [l2norm_fwd_kernel_param_3]; + ld.param.s32 %rd22, [l2norm_fwd_kernel_param_4]; + .loc 1 71 34 // l2norm.py:71:34 + mov.u32 %r79, %tid.x; + and.b32 %r80, %r79, 48; + bfe.u32 %r81, %r79, 4, 2; + or.b32 %r82, %r81, 4; + or.b32 %r83, %r81, 8; + or.b32 %r84, %r81, 12; + or.b32 %r85, %r81, 16; + or.b32 %r86, %r81, 20; + or.b32 %r87, %r81, 24; + or.b32 %r88, %r81, 28; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd23, %r81; + cvt.u64.u32 %rd24, %r82; + cvt.u64.u32 %rd25, %r83; + cvt.u64.u32 %rd26, %r84; + cvt.u64.u32 %rd27, %r85; + cvt.u64.u32 %rd28, %r86; + cvt.u64.u32 %rd29, %r87; + cvt.u64.u32 %rd30, %r88; + or.b64 %rd31, %rd21, %rd23; + or.b64 %rd32, %rd21, %rd24; + or.b64 %rd33, %rd21, %rd25; + or.b64 %rd34, %rd21, %rd26; + or.b64 %rd35, %rd21, %rd27; + or.b64 %rd36, %rd21, %rd28; + or.b64 %rd37, %rd21, %rd29; + or.b64 %rd38, %rd21, %rd30; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p1, %rd31, %rd22; + setp.lt.s64 %p2, %rd32, %rd22; + setp.lt.s64 %p3, %rd33, %rd22; + setp.lt.s64 %p4, %rd34, %rd22; + setp.lt.s64 %p5, %rd35, %rd22; + setp.lt.s64 %p6, %rd36, %rd22; + setp.lt.s64 %p7, %rd37, %rd22; + setp.lt.s64 %p8, %rd38, %rd22; + .loc 1 74 30 // l2norm.py:74:30 + shl.b32 %r89, %r79, 3; + and.b32 %r90, %r89, 120; + .loc 1 75 14 // l2norm.py:75:14 + shl.b64 %rd39, %rd31, 8; + add.s64 %rd40, %rd18, %rd39; + shl.b64 %rd41, %rd32, 8; + add.s64 %rd42, %rd18, %rd41; + shl.b64 %rd43, %rd33, 8; + add.s64 %rd44, %rd18, %rd43; + shl.b64 %rd45, %rd34, 8; + add.s64 %rd46, %rd18, %rd45; + shl.b64 %rd47, %rd35, 8; + add.s64 %rd48, %rd18, %rd47; + shl.b64 %rd49, %rd36, 8; + add.s64 %rd50, %rd18, %rd49; + shl.b64 %rd51, %rd37, 8; + add.s64 %rd52, %rd18, %rd51; + shl.b64 %rd53, %rd38, 8; + add.s64 %rd54, %rd18, %rd53; + .loc 1 75 33 // l2norm.py:75:33 + mul.wide.u32 %rd55, %r90, 2; + add.s64 %rd1, %rd40, %rd55; + add.s64 %rd2, %rd42, %rd55; + add.s64 %rd3, %rd44, %rd55; + add.s64 %rd4, %rd46, %rd55; + add.s64 %rd5, %rd48, %rd55; + add.s64 %rd6, %rd50, %rd55; + add.s64 %rd7, %rd52, %rd55; + add.s64 %rd8, %rd54, %rd55; + mov.b32 %r5, 0; + .loc 1 79 18 // l2norm.py:79:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + mov.b32 {%rs5, %rs6}, %r3; + mov.b32 {%rs7, %rs8}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs9, %rs10}, %r6; + mov.b32 {%rs11, %rs12}, %r7; + mov.b32 {%rs13, %rs14}, %r8; + mov.b32 {%rs15, %rs16}, %r9; + // begin inline asm + mov.u32 %r10, %r5; + mov.u32 %r11, %r5; + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + @%p3 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs17, %rs18}, %r10; + mov.b32 {%rs19, %rs20}, %r11; + mov.b32 {%rs21, %rs22}, %r12; + mov.b32 {%rs23, %rs24}, %r13; + // begin inline asm + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + @%p4 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ]; + // end inline asm + mov.b32 {%rs25, %rs26}, %r14; + mov.b32 {%rs27, %rs28}, %r15; + mov.b32 {%rs29, %rs30}, %r16; + mov.b32 {%rs31, %rs32}, %r17; + // begin inline asm + mov.u32 %r18, %r5; + mov.u32 %r19, %r5; + mov.u32 %r20, %r5; + mov.u32 %r21, %r5; + @%p5 ld.global.v4.b32 { %r18, %r19, %r20, %r21 }, [ %rd5 + 0 ]; + // end inline asm + mov.b32 {%rs33, %rs34}, %r18; + mov.b32 {%rs35, %rs36}, %r19; + mov.b32 {%rs37, %rs38}, %r20; + mov.b32 {%rs39, %rs40}, %r21; + // begin inline asm + mov.u32 %r22, %r5; + mov.u32 %r23, %r5; + mov.u32 %r24, %r5; + mov.u32 %r25, %r5; + @%p6 ld.global.v4.b32 { %r22, %r23, %r24, %r25 }, [ %rd6 + 0 ]; + // end inline asm + mov.b32 {%rs41, %rs42}, %r22; + mov.b32 {%rs43, %rs44}, %r23; + mov.b32 {%rs45, %rs46}, %r24; + mov.b32 {%rs47, %rs48}, %r25; + // begin inline asm + mov.u32 %r26, %r5; + mov.u32 %r27, %r5; + mov.u32 %r28, %r5; + mov.u32 %r29, %r5; + @%p7 ld.global.v4.b32 { %r26, %r27, %r28, %r29 }, [ %rd7 + 0 ]; + // end inline asm + mov.b32 {%rs49, %rs50}, %r26; + mov.b32 {%rs51, %rs52}, %r27; + mov.b32 {%rs53, %rs54}, %r28; + mov.b32 {%rs55, %rs56}, %r29; + // begin inline asm + mov.u32 %r30, %r5; + mov.u32 %r31, %r5; + mov.u32 %r32, %r5; + mov.u32 %r33, %r5; + @%p8 ld.global.v4.b32 { %r30, %r31, %r32, %r33 }, [ %rd8 + 0 ]; + // end inline asm + mov.b32 {%rs57, %rs58}, %r30; + mov.b32 {%rs59, %rs60}, %r31; + mov.b32 {%rs61, %rs62}, %r32; + mov.b32 {%rs63, %rs64}, %r33; + .loc 1 79 47 // l2norm.py:79:47 + cvt.f32.bf16 %r91, %rs1; + cvt.f32.bf16 %r92, %rs2; + cvt.f32.bf16 %r93, %rs3; + cvt.f32.bf16 %r94, %rs4; + cvt.f32.bf16 %r95, %rs5; + cvt.f32.bf16 %r96, %rs6; + cvt.f32.bf16 %r97, %rs7; + cvt.f32.bf16 %r98, %rs8; + cvt.f32.bf16 %r99, %rs9; + cvt.f32.bf16 %r100, %rs10; + cvt.f32.bf16 %r101, %rs11; + cvt.f32.bf16 %r102, %rs12; + cvt.f32.bf16 %r103, %rs13; + cvt.f32.bf16 %r104, %rs14; + cvt.f32.bf16 %r105, %rs15; + cvt.f32.bf16 %r106, %rs16; + cvt.f32.bf16 %r107, %rs17; + cvt.f32.bf16 %r108, %rs18; + cvt.f32.bf16 %r109, %rs19; + cvt.f32.bf16 %r110, %rs20; + cvt.f32.bf16 %r111, %rs21; + cvt.f32.bf16 %r112, %rs22; + cvt.f32.bf16 %r113, %rs23; + cvt.f32.bf16 %r114, %rs24; + cvt.f32.bf16 %r115, %rs25; + cvt.f32.bf16 %r116, %rs26; + cvt.f32.bf16 %r117, %rs27; + cvt.f32.bf16 %r118, %rs28; + cvt.f32.bf16 %r119, %rs29; + cvt.f32.bf16 %r120, %rs30; + cvt.f32.bf16 %r121, %rs31; + cvt.f32.bf16 %r122, %rs32; + cvt.f32.bf16 %r123, %rs33; + cvt.f32.bf16 %r124, %rs34; + cvt.f32.bf16 %r125, %rs35; + cvt.f32.bf16 %r126, %rs36; + cvt.f32.bf16 %r127, %rs37; + cvt.f32.bf16 %r128, %rs38; + cvt.f32.bf16 %r129, %rs39; + cvt.f32.bf16 %r130, %rs40; + cvt.f32.bf16 %r131, %rs41; + cvt.f32.bf16 %r132, %rs42; + cvt.f32.bf16 %r133, %rs43; + cvt.f32.bf16 %r134, %rs44; + cvt.f32.bf16 %r135, %rs45; + cvt.f32.bf16 %r136, %rs46; + cvt.f32.bf16 %r137, %rs47; + cvt.f32.bf16 %r138, %rs48; + cvt.f32.bf16 %r139, %rs49; + cvt.f32.bf16 %r140, %rs50; + cvt.f32.bf16 %r141, %rs51; + cvt.f32.bf16 %r142, %rs52; + cvt.f32.bf16 %r143, %rs53; + cvt.f32.bf16 %r144, %rs54; + cvt.f32.bf16 %r145, %rs55; + cvt.f32.bf16 %r146, %rs56; + cvt.f32.bf16 %r147, %rs57; + cvt.f32.bf16 %r148, %rs58; + cvt.f32.bf16 %r149, %rs59; + cvt.f32.bf16 %r150, %rs60; + cvt.f32.bf16 %r151, %rs61; + cvt.f32.bf16 %r152, %rs62; + cvt.f32.bf16 %r153, %rs63; + cvt.f32.bf16 %r154, %rs64; + .loc 1 80 38 // l2norm.py:80:38 + mul.f32 %r155, %r92, %r92; + mul.f32 %r156, %r100, %r100; + mul.f32 %r157, %r108, %r108; + mul.f32 %r158, %r116, %r116; + mul.f32 %r159, %r124, %r124; + mul.f32 %r160, %r132, %r132; + mul.f32 %r161, %r140, %r140; + mul.f32 %r162, %r148, %r148; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + fma.rn.f32.bf16 %r163, %rs1, %rs1, %r155; + fma.rn.f32.bf16 %r164, %rs3, %rs3, %r163; + fma.rn.f32.bf16 %r165, %rs4, %rs4, %r164; + fma.rn.f32.bf16 %r166, %rs5, %rs5, %r165; + fma.rn.f32.bf16 %r167, %rs6, %rs6, %r166; + fma.rn.f32.bf16 %r168, %rs7, %rs7, %r167; + fma.rn.f32.bf16 %r169, %rs8, %rs8, %r168; + fma.rn.f32.bf16 %r170, %rs9, %rs9, %r156; + fma.rn.f32.bf16 %r171, %rs11, %rs11, %r170; + fma.rn.f32.bf16 %r172, %rs12, %rs12, %r171; + fma.rn.f32.bf16 %r173, %rs13, %rs13, %r172; + fma.rn.f32.bf16 %r174, %rs14, %rs14, %r173; + fma.rn.f32.bf16 %r175, %rs15, %rs15, %r174; + fma.rn.f32.bf16 %r176, %rs16, %rs16, %r175; + fma.rn.f32.bf16 %r177, %rs17, %rs17, %r157; + fma.rn.f32.bf16 %r178, %rs19, %rs19, %r177; + fma.rn.f32.bf16 %r179, %rs20, %rs20, %r178; + fma.rn.f32.bf16 %r180, %rs21, %rs21, %r179; + fma.rn.f32.bf16 %r181, %rs22, %rs22, %r180; + fma.rn.f32.bf16 %r182, %rs23, %rs23, %r181; + fma.rn.f32.bf16 %r183, %rs24, %rs24, %r182; + fma.rn.f32.bf16 %r184, %rs25, %rs25, %r158; + fma.rn.f32.bf16 %r185, %rs27, %rs27, %r184; + fma.rn.f32.bf16 %r186, %rs28, %rs28, %r185; + fma.rn.f32.bf16 %r187, %rs29, %rs29, %r186; + fma.rn.f32.bf16 %r188, %rs30, %rs30, %r187; + fma.rn.f32.bf16 %r189, %rs31, %rs31, %r188; + fma.rn.f32.bf16 %r190, %rs32, %rs32, %r189; + fma.rn.f32.bf16 %r191, %rs33, %rs33, %r159; + fma.rn.f32.bf16 %r192, %rs35, %rs35, %r191; + fma.rn.f32.bf16 %r193, %rs36, %rs36, %r192; + fma.rn.f32.bf16 %r194, %rs37, %rs37, %r193; + fma.rn.f32.bf16 %r195, %rs38, %rs38, %r194; + fma.rn.f32.bf16 %r196, %rs39, %rs39, %r195; + fma.rn.f32.bf16 %r197, %rs40, %rs40, %r196; + fma.rn.f32.bf16 %r198, %rs41, %rs41, %r160; + fma.rn.f32.bf16 %r199, %rs43, %rs43, %r198; + fma.rn.f32.bf16 %r200, %rs44, %rs44, %r199; + fma.rn.f32.bf16 %r201, %rs45, %rs45, %r200; + fma.rn.f32.bf16 %r202, %rs46, %rs46, %r201; + fma.rn.f32.bf16 %r203, %rs47, %rs47, %r202; + fma.rn.f32.bf16 %r204, %rs48, %rs48, %r203; + fma.rn.f32.bf16 %r205, %rs49, %rs49, %r161; + fma.rn.f32.bf16 %r206, %rs51, %rs51, %r205; + fma.rn.f32.bf16 %r207, %rs52, %rs52, %r206; + fma.rn.f32.bf16 %r208, %rs53, %rs53, %r207; + fma.rn.f32.bf16 %r209, %rs54, %rs54, %r208; + fma.rn.f32.bf16 %r210, %rs55, %rs55, %r209; + fma.rn.f32.bf16 %r211, %rs56, %rs56, %r210; + fma.rn.f32.bf16 %r212, %rs57, %rs57, %r162; + fma.rn.f32.bf16 %r213, %rs59, %rs59, %r212; + fma.rn.f32.bf16 %r214, %rs60, %rs60, %r213; + fma.rn.f32.bf16 %r215, %rs61, %rs61, %r214; + fma.rn.f32.bf16 %r216, %rs62, %rs62, %r215; + fma.rn.f32.bf16 %r217, %rs63, %rs63, %r216; + fma.rn.f32.bf16 %r218, %rs64, %rs64, %r217; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r219, %r169, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r220, %r169, %r219; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r221, %r220, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r222, %r220, %r221; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r223, %r222, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r224, %r222, %r223; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r225, %r224, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r226, %r224, %r225; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r227, %r176, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r228, %r176, %r227; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r229, %r228, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r230, %r228, %r229; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r231, %r230, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r232, %r230, %r231; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r233, %r232, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r234, %r232, %r233; +$L__tmp18: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r235, %r183, 8, 31, -1; +$L__tmp19: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r236, %r183, %r235; +$L__tmp20: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r237, %r236, 4, 31, -1; +$L__tmp21: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r238, %r236, %r237; +$L__tmp22: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r239, %r238, 2, 31, -1; +$L__tmp23: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r240, %r238, %r239; +$L__tmp24: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r241, %r240, 1, 31, -1; +$L__tmp25: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r242, %r240, %r241; +$L__tmp26: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r243, %r190, 8, 31, -1; +$L__tmp27: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r244, %r190, %r243; +$L__tmp28: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r245, %r244, 4, 31, -1; +$L__tmp29: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r246, %r244, %r245; +$L__tmp30: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r247, %r246, 2, 31, -1; +$L__tmp31: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r248, %r246, %r247; +$L__tmp32: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r249, %r248, 1, 31, -1; +$L__tmp33: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r250, %r248, %r249; +$L__tmp34: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r251, %r197, 8, 31, -1; +$L__tmp35: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r252, %r197, %r251; +$L__tmp36: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r253, %r252, 4, 31, -1; +$L__tmp37: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r254, %r252, %r253; +$L__tmp38: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r255, %r254, 2, 31, -1; +$L__tmp39: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r256, %r254, %r255; +$L__tmp40: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r257, %r256, 1, 31, -1; +$L__tmp41: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r258, %r256, %r257; +$L__tmp42: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r259, %r204, 8, 31, -1; +$L__tmp43: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r260, %r204, %r259; +$L__tmp44: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r261, %r260, 4, 31, -1; +$L__tmp45: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r262, %r260, %r261; +$L__tmp46: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r263, %r262, 2, 31, -1; +$L__tmp47: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r264, %r262, %r263; +$L__tmp48: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r265, %r264, 1, 31, -1; +$L__tmp49: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r266, %r264, %r265; +$L__tmp50: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r267, %r211, 8, 31, -1; +$L__tmp51: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r268, %r211, %r267; +$L__tmp52: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r269, %r268, 4, 31, -1; +$L__tmp53: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r270, %r268, %r269; +$L__tmp54: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r271, %r270, 2, 31, -1; +$L__tmp55: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r272, %r270, %r271; +$L__tmp56: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r273, %r272, 1, 31, -1; +$L__tmp57: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r274, %r272, %r273; +$L__tmp58: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r275, %r218, 8, 31, -1; +$L__tmp59: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r276, %r218, %r275; +$L__tmp60: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r277, %r276, 4, 31, -1; +$L__tmp61: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r278, %r276, %r277; +$L__tmp62: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r279, %r278, 2, 31, -1; +$L__tmp63: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r280, %r278, %r279; +$L__tmp64: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r281, %r280, 1, 31, -1; +$L__tmp65: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r282, %r280, %r281; +$L__tmp66: + .loc 1 80 48 // l2norm.py:80:48 + add.f32 %r283, %r78, %r226; + add.f32 %r284, %r78, %r234; + add.f32 %r285, %r78, %r242; + add.f32 %r286, %r78, %r250; + add.f32 %r287, %r78, %r258; + add.f32 %r288, %r78, %r266; + add.f32 %r289, %r78, %r274; + add.f32 %r290, %r78, %r282; + .loc 1 80 25 // l2norm.py:80:25 + sqrt.approx.ftz.f32 %r291, %r283; + sqrt.approx.ftz.f32 %r292, %r284; + sqrt.approx.ftz.f32 %r293, %r285; + sqrt.approx.ftz.f32 %r294, %r286; + sqrt.approx.ftz.f32 %r295, %r287; + sqrt.approx.ftz.f32 %r296, %r288; + sqrt.approx.ftz.f32 %r297, %r289; + sqrt.approx.ftz.f32 %r298, %r290; + .loc 1 71 34 // l2norm.py:71:34 + and.b32 %r299, %r79, 31; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd56, %r299; + or.b64 %rd57, %rd21, %rd56; + .loc 1 77 20 // l2norm.py:77:20 + shl.b64 %rd58, %rd57, 2; + add.s64 %rd17, %rd20, %rd58; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd59, %rd19, %rd53; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd16, %rd59, %rd55; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd60, %rd19, %rd51; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd15, %rd60, %rd55; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd61, %rd19, %rd49; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd14, %rd61, %rd55; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd62, %rd19, %rd47; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd13, %rd62, %rd55; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd63, %rd19, %rd45; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd12, %rd63, %rd55; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd64, %rd19, %rd43; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd11, %rd64, %rd55; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd65, %rd19, %rd41; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd10, %rd65, %rd55; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd66, %rd19, %rd39; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd9, %rd66, %rd55; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p10, %rd57, %rd22; + mov.b32 %r300, 0f3F800000; + .loc 1 80 17 // l2norm.py:80:17 + div.full.f32 %r67, %r300, %r291; + div.full.f32 %r68, %r300, %r292; + div.full.f32 %r69, %r300, %r293; + div.full.f32 %r70, %r300, %r294; + div.full.f32 %r72, %r300, %r295; + div.full.f32 %r73, %r300, %r296; + div.full.f32 %r74, %r300, %r297; + div.full.f32 %r75, %r300, %r298; + .loc 1 81 16 // l2norm.py:81:16 + mul.f32 %r301, %r67, %r91; + mul.f32 %r302, %r67, %r92; + mul.f32 %r303, %r67, %r93; + mul.f32 %r304, %r67, %r94; + mul.f32 %r305, %r67, %r95; + mul.f32 %r306, %r67, %r96; + mul.f32 %r307, %r67, %r97; + mul.f32 %r308, %r67, %r98; + mul.f32 %r309, %r68, %r99; + mul.f32 %r310, %r68, %r100; + mul.f32 %r311, %r68, %r101; + mul.f32 %r312, %r68, %r102; + mul.f32 %r313, %r68, %r103; + mul.f32 %r314, %r68, %r104; + mul.f32 %r315, %r68, %r105; + mul.f32 %r316, %r68, %r106; + mul.f32 %r317, %r69, %r107; + mul.f32 %r318, %r69, %r108; + mul.f32 %r319, %r69, %r109; + mul.f32 %r320, %r69, %r110; + mul.f32 %r321, %r69, %r111; + mul.f32 %r322, %r69, %r112; + mul.f32 %r323, %r69, %r113; + mul.f32 %r324, %r69, %r114; + mul.f32 %r325, %r70, %r115; + mul.f32 %r326, %r70, %r116; + mul.f32 %r327, %r70, %r117; + mul.f32 %r328, %r70, %r118; + mul.f32 %r329, %r70, %r119; + mul.f32 %r330, %r70, %r120; + mul.f32 %r331, %r70, %r121; + mul.f32 %r332, %r70, %r122; + mul.f32 %r333, %r72, %r123; + mul.f32 %r334, %r72, %r124; + mul.f32 %r335, %r72, %r125; + mul.f32 %r336, %r72, %r126; + mul.f32 %r337, %r72, %r127; + mul.f32 %r338, %r72, %r128; + mul.f32 %r339, %r72, %r129; + mul.f32 %r340, %r72, %r130; + mul.f32 %r341, %r73, %r131; + mul.f32 %r342, %r73, %r132; + mul.f32 %r343, %r73, %r133; + mul.f32 %r344, %r73, %r134; + mul.f32 %r345, %r73, %r135; + mul.f32 %r346, %r73, %r136; + mul.f32 %r347, %r73, %r137; + mul.f32 %r348, %r73, %r138; + mul.f32 %r349, %r74, %r139; + mul.f32 %r350, %r74, %r140; + mul.f32 %r351, %r74, %r141; + mul.f32 %r352, %r74, %r142; + mul.f32 %r353, %r74, %r143; + mul.f32 %r354, %r74, %r144; + mul.f32 %r355, %r74, %r145; + mul.f32 %r356, %r74, %r146; + mul.f32 %r357, %r75, %r147; + mul.f32 %r358, %r75, %r148; + mul.f32 %r359, %r75, %r149; + mul.f32 %r360, %r75, %r150; + mul.f32 %r361, %r75, %r151; + mul.f32 %r362, %r75, %r152; + mul.f32 %r363, %r75, %r153; + mul.f32 %r364, %r75, %r154; + .loc 1 83 18 // l2norm.py:83:18 + cvt.rn.bf16x2.f32 %r34, %r302, %r301; + cvt.rn.bf16x2.f32 %r35, %r304, %r303; + cvt.rn.bf16x2.f32 %r36, %r306, %r305; + cvt.rn.bf16x2.f32 %r37, %r308, %r307; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd9 + 0 ], { %r34, %r35, %r36, %r37 }; + // end inline asm + cvt.rn.bf16x2.f32 %r38, %r310, %r309; + cvt.rn.bf16x2.f32 %r39, %r312, %r311; + cvt.rn.bf16x2.f32 %r40, %r314, %r313; + cvt.rn.bf16x2.f32 %r41, %r316, %r315; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd10 + 0 ], { %r38, %r39, %r40, %r41 }; + // end inline asm + cvt.rn.bf16x2.f32 %r42, %r318, %r317; + cvt.rn.bf16x2.f32 %r43, %r320, %r319; + cvt.rn.bf16x2.f32 %r44, %r322, %r321; + cvt.rn.bf16x2.f32 %r45, %r324, %r323; + // begin inline asm + @%p3 st.global.v4.b32 [ %rd11 + 0 ], { %r42, %r43, %r44, %r45 }; + // end inline asm + cvt.rn.bf16x2.f32 %r46, %r326, %r325; + cvt.rn.bf16x2.f32 %r47, %r328, %r327; + cvt.rn.bf16x2.f32 %r48, %r330, %r329; + cvt.rn.bf16x2.f32 %r49, %r332, %r331; + // begin inline asm + @%p4 st.global.v4.b32 [ %rd12 + 0 ], { %r46, %r47, %r48, %r49 }; + // end inline asm + cvt.rn.bf16x2.f32 %r50, %r334, %r333; + cvt.rn.bf16x2.f32 %r51, %r336, %r335; + cvt.rn.bf16x2.f32 %r52, %r338, %r337; + cvt.rn.bf16x2.f32 %r53, %r340, %r339; + // begin inline asm + @%p5 st.global.v4.b32 [ %rd13 + 0 ], { %r50, %r51, %r52, %r53 }; + // end inline asm + cvt.rn.bf16x2.f32 %r54, %r342, %r341; + cvt.rn.bf16x2.f32 %r55, %r344, %r343; + cvt.rn.bf16x2.f32 %r56, %r346, %r345; + cvt.rn.bf16x2.f32 %r57, %r348, %r347; + // begin inline asm + @%p6 st.global.v4.b32 [ %rd14 + 0 ], { %r54, %r55, %r56, %r57 }; + // end inline asm + cvt.rn.bf16x2.f32 %r58, %r350, %r349; + cvt.rn.bf16x2.f32 %r59, %r352, %r351; + cvt.rn.bf16x2.f32 %r60, %r354, %r353; + cvt.rn.bf16x2.f32 %r61, %r356, %r355; + // begin inline asm + @%p7 st.global.v4.b32 [ %rd15 + 0 ], { %r58, %r59, %r60, %r61 }; + // end inline asm + cvt.rn.bf16x2.f32 %r62, %r358, %r357; + cvt.rn.bf16x2.f32 %r63, %r360, %r359; + cvt.rn.bf16x2.f32 %r64, %r362, %r361; + cvt.rn.bf16x2.f32 %r65, %r364, %r363; + // begin inline asm + @%p8 st.global.v4.b32 [ %rd16 + 0 ], { %r62, %r63, %r64, %r65 }; + // end inline asm + .loc 1 84 21 // l2norm.py:84:21 + shl.b32 %r365, %r80, 1; + mov.b32 %r366, global_smem; + add.s32 %r66, %r366, %r365; + // begin inline asm + st.shared.v4.b32 [ %r66 + 0 ], { %r67, %r68, %r69, %r70 }; + // end inline asm + add.s32 %r71, %r66, 16; + // begin inline asm + st.shared.v4.b32 [ %r71 + 0 ], { %r72, %r73, %r74, %r75 }; + // end inline asm + bar.sync 0; + shl.b32 %r367, %r79, 5; + and.b32 %r368, %r367, 96; + and.b32 %r369, %r79, 28; + add.s32 %r370, %r366, %r368; + add.s32 %r371, %r370, %r369; + ld.shared.b32 %r76, [%r371]; + and.b32 %r372, %r79, 32; + setp.eq.b32 %p11, %r372, 0; + and.pred %p9, %p11, %p10; + // begin inline asm + @%p9 st.global.b32 [ %rd17 + 0 ], { %r76 }; + // end inline asm + .loc 1 84 4 // l2norm.py:84:4 + ret; +$L__tmp67: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp66 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 80 // DW_AT_call_line +.b8 43 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp66 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.source b/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..d9fa83aac0600b1e5d35d95ed8aaad5c5636e577 --- /dev/null +++ b/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.source @@ -0,0 +1,172 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc38 = loc(unknown) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc45 = loc("x"(#loc)) +#loc46 = loc("y"(#loc)) +#loc47 = loc("rstd"(#loc)) +#loc48 = loc("eps"(#loc)) +#loc49 = loc("T"(#loc)) +#loc81 = loc("input"(#loc36)) +#loc82 = loc("a"(#loc41)) +#loc83 = loc("b"(#loc41)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc50) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc51) + %o_t = arith.constant 32 : i32 loc(#loc52) + %o_t_1 = arith.constant 32 : i64 loc(#loc52) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc52) + %o_t_3 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32> loc(#loc53) + %o_t_4 = arith.extsi %o_t_3 : tensor<32xi32> to tensor<32xi64> loc(#loc54) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<32xi64> loc(#loc54) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<32xi64> loc(#loc54) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc55) + %m_t = arith.extsi %T : i32 to i64 loc(#loc56) + %m_t_7 = tt.splat %m_t : i64 -> tensor<32xi64> loc(#loc56) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<32xi64> loc(#loc56) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<32xi1> -> tensor<32x1xi1> loc(#loc57) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc58) + %m_x_10 = arith.constant 128 : i32 loc(#loc59) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc59) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc59) + %m_x_13 = tt.broadcast %m_x : tensor<32x1xi1> -> tensor<32x128xi1> loc(#loc60) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<32x128xi1> loc(#loc60) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<32x128xi1> loc(#loc60) + %p_x = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<32xi64> -> tensor<32x1xi64> loc(#loc61) + %p_x_16 = arith.constant 128 : i32 loc(#loc62) + %p_x_17 = arith.constant 128 : i64 loc(#loc62) + %p_x_18 = arith.constant dense<128> : tensor<32x1xi64> loc(#loc62) + %p_x_19 = arith.muli %p_x, %p_x_18 : tensor<32x1xi64> loc(#loc62) + %p_x_20 = tt.splat %x : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc63) + %p_x_21 = tt.addptr %p_x_20, %p_x_19 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc63) + %p_x_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc64) + %p_x_23 = tt.broadcast %p_x_21 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc65) + %p_x_24 = tt.broadcast %p_x_22 : tensor<1x128xi32> -> tensor<32x128xi32> loc(#loc65) + %p_x_25 = tt.addptr %p_x_23, %p_x_24 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc65) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<32xi64> -> tensor<32x1xi64> loc(#loc66) + %p_y_26 = arith.constant 128 : i32 loc(#loc67) + %p_y_27 = arith.constant 128 : i64 loc(#loc67) + %p_y_28 = arith.constant dense<128> : tensor<32x1xi64> loc(#loc67) + %p_y_29 = arith.muli %p_y, %p_y_28 : tensor<32x1xi64> loc(#loc67) + %p_y_30 = tt.splat %y : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc68) + %p_y_31 = tt.addptr %p_y_30, %p_y_29 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc68) + %p_y_32 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc69) + %p_y_33 = tt.broadcast %p_y_31 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc70) + %p_y_34 = tt.broadcast %p_y_32 : tensor<1x128xi32> -> tensor<32x128xi32> loc(#loc70) + %p_y_35 = tt.addptr %p_y_33, %p_y_34 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc70) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<32x!tt.ptr> loc(#loc71) + %p_rstd_36 = tt.addptr %p_rstd, %o_t_6 : tensor<32x!tt.ptr>, tensor<32xi64> loc(#loc71) + %b_x = arith.constant 0.000000e+00 : f32 loc(#loc72) + %b_x_37 = arith.constant dense<0.000000e+00> : tensor<32x128xf32> loc(#loc72) + %b_x_38 = arith.truncf %b_x_37 : tensor<32x128xf32> to tensor<32x128xbf16> loc(#loc72) + %b_x_39 = tt.load %p_x_25, %m_x_15, %b_x_38 : tensor<32x128x!tt.ptr> loc(#loc72) + %b_x_40 = arith.extf %b_x_39 : tensor<32x128xbf16> to tensor<32x128xf32> loc(#loc73) + %b_rstd = arith.mulf %b_x_40, %b_x_40 : tensor<32x128xf32> loc(#loc74) + %b_rstd_41 = tt.call @triton.language.standard.sum__fp32S32_128S_c1_cFalse_cNone(%b_rstd) : (tensor<32x128xf32>) -> tensor<32xf32> loc(#loc75) + %b_rstd_42 = tt.splat %eps : f32 -> tensor<32xf32> loc(#loc76) + %b_rstd_43 = arith.addf %b_rstd_41, %b_rstd_42 : tensor<32xf32> loc(#loc76) + %b_rstd_44 = math.sqrt %b_rstd_43 : tensor<32xf32> loc(#loc77) + %b_rstd_45 = arith.constant 1 : i32 loc(#loc78) + %b_rstd_46 = arith.constant 1.000000e+00 : f32 loc(#loc78) + %b_rstd_47 = arith.constant dense<1.000000e+00> : tensor<32xf32> loc(#loc78) + %b_rstd_48 = arith.divf %b_rstd_47, %b_rstd_44 : tensor<32xf32> loc(#loc78) + %b_y = tt.expand_dims %b_rstd_48 {axis = 1 : i32} : tensor<32xf32> -> tensor<32x1xf32> loc(#loc79) + %b_y_49 = tt.broadcast %b_y : tensor<32x1xf32> -> tensor<32x128xf32> loc(#loc80) + %b_y_50 = arith.mulf %b_x_40, %b_y_49 : tensor<32x128xf32> loc(#loc80) + %0 = arith.truncf %b_y_50 : tensor<32x128xf32> to tensor<32x128xbf16> loc(#loc32) + tt.store %p_y_35, %0, %m_x_15 : tensor<32x128x!tt.ptr> loc(#loc33) + tt.store %p_rstd_36, %b_rstd_48, %m_t_8 : tensor<32x!tt.ptr> loc(#loc34) + tt.return loc(#loc35) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S32_128S_c1_cFalse_cNone(%input: tensor<32x128xf32> loc("input"(#loc36))) -> tensor<32xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc37) + tt.reduce.return %2 : f32 loc(#loc37) + }) : (tensor<32x128xf32>) -> tensor<32xf32> loc(#loc37) + tt.return %0 : tensor<32xf32> loc(#loc39) + ^bb1: // no predecessors + %1 = ub.poison : tensor<32xf32> loc(#loc40) + tt.return %1 : tensor<32xf32> loc(#loc40) + } loc(#loc36) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc41)), %b: f32 loc("b"(#loc41))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc42) + tt.return %0 : f32 loc(#loc43) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc44) + tt.return %1 : f32 loc(#loc44) + } loc(#loc41) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:18) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:29) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:37) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc50 = loc("i_t"(#loc1)) +#loc51 = loc("i_t"(#loc2)) +#loc52 = loc("o_t"(#loc3)) +#loc53 = loc("o_t"(#loc4)) +#loc54 = loc("o_t"(#loc5)) +#loc55 = loc("o_d"(#loc6)) +#loc56 = loc("m_t"(#loc7)) +#loc57 = loc("m_x"(#loc8)) +#loc58 = loc("m_x"(#loc9)) +#loc59 = loc("m_x"(#loc10)) +#loc60 = loc("m_x"(#loc11)) +#loc61 = loc("p_x"(#loc12)) +#loc62 = loc("p_x"(#loc13)) +#loc63 = loc("p_x"(#loc14)) +#loc64 = loc("p_x"(#loc15)) +#loc65 = loc("p_x"(#loc16)) +#loc66 = loc("p_y"(#loc17)) +#loc67 = loc("p_y"(#loc18)) +#loc68 = loc("p_y"(#loc19)) +#loc69 = loc("p_y"(#loc20)) +#loc70 = loc("p_y"(#loc21)) +#loc71 = loc("p_rstd"(#loc22)) +#loc72 = loc("b_x"(#loc23)) +#loc73 = loc("b_x"(#loc24)) +#loc74 = loc("b_rstd"(#loc25)) +#loc75 = loc("b_rstd"(#loc26)) +#loc76 = loc("b_rstd"(#loc27)) +#loc77 = loc("b_rstd"(#loc28)) +#loc78 = loc("b_rstd"(#loc29)) +#loc79 = loc("b_y"(#loc30)) +#loc80 = loc("b_y"(#loc31)) diff --git a/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.ttgir b/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..b137a7870e377a1a99f2fc1d8c75cb728717eea2 --- /dev/null +++ b/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.ttgir @@ -0,0 +1,135 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [2, 1], order = [1, 0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [2], order = [0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc1 = loc(unknown) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc34 = loc("x"(#loc)) +#loc35 = loc("y"(#loc)) +#loc36 = loc("rstd"(#loc)) +#loc37 = loc("eps"(#loc)) +#loc38 = loc("T"(#loc)) +#loc59 = loc("b_rstd"(#loc23)) +#loc66 = loc(callsite(#loc1 at #loc59)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 2 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<1.000000e+00> : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<32x128xbf16, #blocked> loc(#loc1) + %c32_i64 = arith.constant 32 : i64 loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<32x1xi64, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc39) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc40) + %o_t = arith.muli %i_t_3, %c32_i64 : i64 loc(#loc41) + %o_t_4 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc42) + %o_t_5 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32, #blocked1> loc(#loc42) + %o_t_6 = arith.extsi %o_t_4 : tensor<32xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_7 = arith.extsi %o_t_5 : tensor<32xi32, #blocked1> to tensor<32xi64, #blocked1> loc(#loc43) + %o_t_8 = tt.splat %o_t : i64 -> tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_9 = tt.splat %o_t : i64 -> tensor<32xi64, #blocked1> loc(#loc43) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<32xi64, #blocked1> loc(#loc43) + %m_t = arith.extsi %T : i32 to i64 loc(#loc44) + %m_t_12 = tt.splat %m_t : i64 -> tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_13 = tt.splat %m_t : i64 -> tensor<32xi64, #blocked1> loc(#loc44) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<32xi64, #blocked1> loc(#loc44) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<32xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32x1xi1, #blocked> loc(#loc45) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc46) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc46) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_1 : tensor<1x128xi32, #blocked> loc(#loc47) + %m_x_19 = tt.broadcast %m_x : tensor<32x1xi1, #blocked> -> tensor<32x128xi1, #blocked> loc(#loc48) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<32x128xi1, #blocked> loc(#loc48) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<32x128xi1, #blocked> loc(#loc48) + %p_x = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32x1xi64, #blocked> loc(#loc49) + %p_x_22 = arith.muli %p_x, %cst_2 : tensor<32x1xi64, #blocked> loc(#loc50) + %p_x_23 = tt.splat %x : !tt.ptr -> tensor<32x1x!tt.ptr, #blocked> loc(#loc51) + %p_x_24 = tt.addptr %p_x_23, %p_x_22 : tensor<32x1x!tt.ptr, #blocked>, tensor<32x1xi64, #blocked> loc(#loc51) + %p_x_25 = tt.broadcast %p_x_24 : tensor<32x1x!tt.ptr, #blocked> -> tensor<32x128x!tt.ptr, #blocked> loc(#loc52) + %p_x_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<32x128xi32, #blocked> loc(#loc52) + %p_x_27 = tt.addptr %p_x_25, %p_x_26 : tensor<32x128x!tt.ptr, #blocked>, tensor<32x128xi32, #blocked> loc(#loc52) + %p_y = tt.splat %y : !tt.ptr -> tensor<32x1x!tt.ptr, #blocked> loc(#loc53) + %p_y_28 = tt.addptr %p_y, %p_x_22 : tensor<32x1x!tt.ptr, #blocked>, tensor<32x1xi64, #blocked> loc(#loc53) + %p_y_29 = tt.broadcast %p_y_28 : tensor<32x1x!tt.ptr, #blocked> -> tensor<32x128x!tt.ptr, #blocked> loc(#loc54) + %p_y_30 = tt.addptr %p_y_29, %p_x_26 : tensor<32x128x!tt.ptr, #blocked>, tensor<32x128xi32, #blocked> loc(#loc54) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<32x!tt.ptr, #blocked1> loc(#loc55) + %p_rstd_31 = tt.addptr %p_rstd, %o_t_11 : tensor<32x!tt.ptr, #blocked1>, tensor<32xi64, #blocked1> loc(#loc55) + %b_x = tt.load %p_x_27, %m_x_21, %cst_0 : tensor<32x128x!tt.ptr, #blocked> loc(#loc56) + %b_x_32 = arith.extf %b_x : tensor<32x128xbf16, #blocked> to tensor<32x128xf32, #blocked> loc(#loc57) + %b_rstd = arith.mulf %b_x_32, %b_x_32 : tensor<32x128xf32, #blocked> loc(#loc58) + %b_rstd_33 = "tt.reduce"(%b_rstd) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_40: f32 loc(callsite(#loc1 at #loc59)), %b_rstd_41: f32 loc(callsite(#loc1 at #loc59))): + %b_rstd_42 = arith.addf %b_rstd_40, %b_rstd_41 : f32 loc(#loc67) + tt.reduce.return %b_rstd_42 : f32 loc(#loc65) + }) : (tensor<32x128xf32, #blocked>) -> tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc65) + %b_rstd_34 = tt.splat %eps : f32 -> tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_36 = math.sqrt %b_rstd_35 : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc61) + %b_rstd_37 = arith.divf %cst, %b_rstd_36 : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc62) + %b_y = tt.expand_dims %b_rstd_37 {axis = 1 : i32} : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32x1xf32, #blocked> loc(#loc63) + %b_y_38 = tt.broadcast %b_y : tensor<32x1xf32, #blocked> -> tensor<32x128xf32, #blocked> loc(#loc64) + %b_y_39 = arith.mulf %b_x_32, %b_y_38 : tensor<32x128xf32, #blocked> loc(#loc64) + %0 = arith.truncf %b_y_39 : tensor<32x128xf32, #blocked> to tensor<32x128xbf16, #blocked> loc(#loc30) + tt.store %p_y_30, %0, %m_x_21 : tensor<32x128x!tt.ptr, #blocked> loc(#loc31) + %1 = ttg.convert_layout %b_rstd_37 : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32xf32, #blocked1> loc(#loc32) + tt.store %p_rstd_31, %1, %m_t_15 : tensor<32x!tt.ptr, #blocked1> loc(#loc32) + tt.return loc(#loc33) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc39 = loc("i_t"(#loc2)) +#loc40 = loc("i_t"(#loc3)) +#loc41 = loc("o_t"(#loc4)) +#loc42 = loc("o_t"(#loc5)) +#loc43 = loc("o_t"(#loc6)) +#loc44 = loc("m_t"(#loc7)) +#loc45 = loc("m_x"(#loc8)) +#loc46 = loc("m_x"(#loc9)) +#loc47 = loc("m_x"(#loc10)) +#loc48 = loc("m_x"(#loc11)) +#loc49 = loc("p_x"(#loc12)) +#loc50 = loc("p_x"(#loc13)) +#loc51 = loc("p_x"(#loc14)) +#loc52 = loc("p_x"(#loc15)) +#loc53 = loc("p_y"(#loc16)) +#loc54 = loc("p_y"(#loc17)) +#loc55 = loc("p_rstd"(#loc18)) +#loc56 = loc("b_x"(#loc19)) +#loc57 = loc("b_x"(#loc20)) +#loc58 = loc("b_rstd"(#loc21)) +#loc60 = loc("b_rstd"(#loc25)) +#loc61 = loc("b_rstd"(#loc26)) +#loc62 = loc("b_rstd"(#loc27)) +#loc63 = loc("b_y"(#loc28)) +#loc64 = loc("b_y"(#loc29)) +#loc65 = loc(callsite(#loc22 at #loc59)) +#loc67 = loc(callsite(#loc24 at #loc65)) diff --git a/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.ttir b/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..3688281bceaf7f8abdf8568104090be5e0170141 --- /dev/null +++ b/triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.ttir @@ -0,0 +1,128 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc3 = loc(unknown) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc35 = loc("x"(#loc)) +#loc36 = loc("y"(#loc)) +#loc37 = loc("rstd"(#loc)) +#loc38 = loc("eps"(#loc)) +#loc39 = loc("T"(#loc)) +#loc62 = loc("b_rstd"(#loc25)) +#loc68 = loc(callsite(#loc3 at #loc62)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %b_x = arith.constant dense<0.000000e+00> : tensor<32x128xbf16> loc(#loc40) + %b_rstd = arith.constant dense<1.000000e+00> : tensor<32xf32> loc(#loc41) + %cst = arith.constant dense<128> : tensor<32x1xi64> loc(#loc3) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc42) + %o_t = arith.constant 32 : i64 loc(#loc43) + %i_t = tt.get_program_id x : i32 loc(#loc44) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc45) + %o_t_1 = arith.muli %i_t_0, %o_t : i64 loc(#loc43) + %o_t_2 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32> loc(#loc46) + %o_t_3 = arith.extsi %o_t_2 : tensor<32xi32> to tensor<32xi64> loc(#loc47) + %o_t_4 = tt.splat %o_t_1 : i64 -> tensor<32xi64> loc(#loc47) + %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<32xi64> loc(#loc47) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc48) + %m_t = arith.extsi %T : i32 to i64 loc(#loc49) + %m_t_6 = tt.splat %m_t : i64 -> tensor<32xi64> loc(#loc49) + %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<32xi64> loc(#loc49) + %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<32xi1> -> tensor<32x1xi1> loc(#loc50) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc51) + %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc42) + %m_x_11 = tt.broadcast %m_x_8 : tensor<32x1xi1> -> tensor<32x128xi1> loc(#loc52) + %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<32x128xi1> loc(#loc52) + %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<32x128xi1> loc(#loc52) + %p_x = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<32xi64> -> tensor<32x1xi64> loc(#loc53) + %p_x_14 = arith.muli %p_x, %cst : tensor<32x1xi64> loc(#loc54) + %p_x_15 = tt.splat %x : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc55) + %p_x_16 = tt.addptr %p_x_15, %p_x_14 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc55) + %p_x_17 = tt.broadcast %p_x_16 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc56) + %p_x_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<32x128xi32> loc(#loc56) + %p_x_19 = tt.addptr %p_x_17, %p_x_18 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc56) + %p_y = tt.splat %y : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc57) + %p_y_20 = tt.addptr %p_y, %p_x_14 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc57) + %p_y_21 = tt.broadcast %p_y_20 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc58) + %p_y_22 = tt.addptr %p_y_21, %p_x_18 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<32x!tt.ptr> loc(#loc59) + %p_rstd_23 = tt.addptr %p_rstd, %o_t_5 : tensor<32x!tt.ptr>, tensor<32xi64> loc(#loc59) + %b_x_24 = tt.load %p_x_19, %m_x_13, %b_x : tensor<32x128x!tt.ptr> loc(#loc40) + %b_x_25 = arith.extf %b_x_24 : tensor<32x128xbf16> to tensor<32x128xf32> loc(#loc60) + %b_rstd_26 = arith.mulf %b_x_25, %b_x_25 : tensor<32x128xf32> loc(#loc61) + %b_rstd_27 = "tt.reduce"(%b_rstd_26) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_34: f32 loc(callsite(#loc3 at #loc62)), %b_rstd_35: f32 loc(callsite(#loc3 at #loc62))): + %b_rstd_36 = arith.addf %b_rstd_34, %b_rstd_35 : f32 loc(#loc69) + tt.reduce.return %b_rstd_36 : f32 loc(#loc67) + }) : (tensor<32x128xf32>) -> tensor<32xf32> loc(#loc67) + %b_rstd_28 = tt.splat %eps : f32 -> tensor<32xf32> loc(#loc63) + %b_rstd_29 = arith.addf %b_rstd_27, %b_rstd_28 : tensor<32xf32> loc(#loc63) + %b_rstd_30 = math.sqrt %b_rstd_29 : tensor<32xf32> loc(#loc64) + %b_rstd_31 = arith.divf %b_rstd, %b_rstd_30 : tensor<32xf32> loc(#loc41) + %b_y = tt.expand_dims %b_rstd_31 {axis = 1 : i32} : tensor<32xf32> -> tensor<32x1xf32> loc(#loc65) + %b_y_32 = tt.broadcast %b_y : tensor<32x1xf32> -> tensor<32x128xf32> loc(#loc66) + %b_y_33 = arith.mulf %b_x_25, %b_y_32 : tensor<32x128xf32> loc(#loc66) + %0 = arith.truncf %b_y_33 : tensor<32x128xf32> to tensor<32x128xbf16> loc(#loc31) + tt.store %p_y_22, %0, %m_x_13 : tensor<32x128x!tt.ptr> loc(#loc32) + tt.store %p_rstd_23, %b_rstd_31, %m_t_7 : tensor<32x!tt.ptr> loc(#loc33) + tt.return loc(#loc34) + } loc(#loc) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc40 = loc("b_x"(#loc1)) +#loc41 = loc("b_rstd"(#loc2)) +#loc42 = loc("m_x"(#loc4)) +#loc43 = loc("o_t"(#loc5)) +#loc44 = loc("i_t"(#loc6)) +#loc45 = loc("i_t"(#loc7)) +#loc46 = loc("o_t"(#loc8)) +#loc47 = loc("o_t"(#loc9)) +#loc48 = loc("o_d"(#loc10)) +#loc49 = loc("m_t"(#loc11)) +#loc50 = loc("m_x"(#loc12)) +#loc51 = loc("m_x"(#loc13)) +#loc52 = loc("m_x"(#loc14)) +#loc53 = loc("p_x"(#loc15)) +#loc54 = loc("p_x"(#loc16)) +#loc55 = loc("p_x"(#loc17)) +#loc56 = loc("p_x"(#loc18)) +#loc57 = loc("p_y"(#loc19)) +#loc58 = loc("p_y"(#loc20)) +#loc59 = loc("p_rstd"(#loc21)) +#loc60 = loc("b_x"(#loc22)) +#loc61 = loc("b_rstd"(#loc23)) +#loc63 = loc("b_rstd"(#loc27)) +#loc64 = loc("b_rstd"(#loc28)) +#loc65 = loc("b_y"(#loc29)) +#loc66 = loc("b_y"(#loc30)) +#loc67 = loc(callsite(#loc24 at #loc62)) +#loc69 = loc(callsite(#loc26 at #loc67)) diff --git a/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/__grp__l2norm_bwd_kernel.json b/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/__grp__l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..ff5f426202d3bb9caa314af472605144fc9c5b7d --- /dev/null +++ b/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/__grp__l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_bwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.source", "l2norm_bwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.ttir", "l2norm_bwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.ttgir", "l2norm_bwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.llir", "l2norm_bwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.ptx", "l2norm_bwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.cubin", "l2norm_bwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.cubin b/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..7d983c42ab235cca77589520bc84b6224077598e Binary files /dev/null and b/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.cubin differ diff --git a/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.json b/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..fb44e0d492fbc3126662a6fa4147d09ad6b0640c --- /dev/null +++ b/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"hash": "0572d64dc42619f5afd10564c68284709c7c470c8541f052c22241f27d7deab7", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 8, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 0, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_bwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.llir b/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..34ac243461b951de8e152baed95400f872c9ccdc --- /dev/null +++ b/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.llir @@ -0,0 +1,1269 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_bwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, ptr addrspace(1) %3, float %4, i32 %5, ptr addrspace(1) readnone captures(none) %6, ptr addrspace(1) readnone captures(none) %7) local_unnamed_addr #0 !dbg !4 { + %9 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !12 + %10 = zext nneg i32 %9 to i64, !dbg !13 + %11 = shl nuw nsw i64 %10, 7, !dbg !14 + %12 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !15 + %13 = lshr i32 %12, 4, !dbg !15 + %14 = and i32 %13, 15, !dbg !15 + %15 = or disjoint i32 %14, 16, !dbg !15 + %16 = insertelement <2 x i32> poison, i32 %13, i64 0, !dbg !15 + %17 = insertelement <2 x i32> %16, i32 %14, i64 1, !dbg !15 + %18 = or <2 x i32> %17, , !dbg !15 + %19 = insertelement <4 x i32> poison, i32 %13, i64 0, !dbg !15 + %20 = insertelement <4 x i32> %19, i32 %14, i64 1, !dbg !15 + %21 = shufflevector <2 x i32> %18, <2 x i32> poison, <8 x i32> , !dbg !16 + %22 = insertelement <8 x i32> %21, i32 %15, i64 6, !dbg !16 + %23 = insertelement <8 x i32> %22, i32 %14, i64 7, !dbg !16 + %24 = shufflevector <4 x i32> %20, <4 x i32> poison, <8 x i32> , !dbg !16 + %25 = or <8 x i32> %24, , !dbg !16 + %26 = shufflevector <8 x i32> %25, <8 x i32> %23, <8 x i32> , !dbg !16 + %27 = zext <8 x i32> %26 to <8 x i64>, !dbg !16 + %28 = insertelement <8 x i64> poison, i64 %11, i64 0, !dbg !16 + %29 = shufflevector <8 x i64> %28, <8 x i64> poison, <8 x i32> zeroinitializer, !dbg !16 + %30 = or disjoint <8 x i64> %29, %27, !dbg !16 + %31 = insertelement <8 x i32> poison, i32 %5, i64 0, !dbg !17 + %32 = shufflevector <8 x i32> %31, <8 x i32> poison, <8 x i32> zeroinitializer, !dbg !17 + %33 = sext <8 x i32> %32 to <8 x i64>, !dbg !17 + %34 = icmp slt <8 x i64> %30, %33, !dbg !17 + %35 = shl nuw nsw i32 %12, 3, !dbg !18 + %36 = and i32 %35, 120, !dbg !18 + %37 = extractelement <8 x i64> %30, i64 7, !dbg !19 + %38 = shl nuw nsw i64 %37, 7, !dbg !19 + %39 = extractelement <8 x i64> %30, i64 6, !dbg !19 + %40 = shl nuw nsw i64 %39, 7, !dbg !19 + %41 = extractelement <8 x i64> %30, i64 5, !dbg !19 + %42 = shl nuw nsw i64 %41, 7, !dbg !19 + %43 = extractelement <8 x i64> %30, i64 4, !dbg !19 + %44 = shl nuw nsw i64 %43, 7, !dbg !19 + %45 = extractelement <8 x i64> %30, i64 3, !dbg !19 + %46 = shl nuw nsw i64 %45, 7, !dbg !19 + %47 = extractelement <8 x i64> %30, i64 2, !dbg !19 + %48 = shl nuw nsw i64 %47, 7, !dbg !19 + %49 = extractelement <8 x i64> %30, i64 1, !dbg !19 + %50 = shl nuw nsw i64 %49, 7, !dbg !19 + %51 = extractelement <8 x i64> %30, i64 0, !dbg !19 + %52 = shl nuw nsw i64 %51, 7, !dbg !19 + %53 = getelementptr bfloat, ptr addrspace(1) %0, i64 %38, !dbg !20 + %54 = getelementptr bfloat, ptr addrspace(1) %0, i64 %40, !dbg !20 + %55 = getelementptr bfloat, ptr addrspace(1) %0, i64 %42, !dbg !20 + %56 = getelementptr bfloat, ptr addrspace(1) %0, i64 %44, !dbg !20 + %57 = getelementptr bfloat, ptr addrspace(1) %0, i64 %46, !dbg !20 + %58 = getelementptr bfloat, ptr addrspace(1) %0, i64 %48, !dbg !20 + %59 = getelementptr bfloat, ptr addrspace(1) %0, i64 %50, !dbg !20 + %60 = getelementptr bfloat, ptr addrspace(1) %0, i64 %52, !dbg !20 + %61 = zext nneg i32 %36 to i64, !dbg !21 + %62 = getelementptr bfloat, ptr addrspace(1) %53, i64 %61, !dbg !21 + %63 = getelementptr bfloat, ptr addrspace(1) %54, i64 %61, !dbg !21 + %64 = getelementptr bfloat, ptr addrspace(1) %55, i64 %61, !dbg !21 + %65 = getelementptr bfloat, ptr addrspace(1) %56, i64 %61, !dbg !21 + %66 = getelementptr bfloat, ptr addrspace(1) %57, i64 %61, !dbg !21 + %67 = getelementptr bfloat, ptr addrspace(1) %58, i64 %61, !dbg !21 + %68 = getelementptr bfloat, ptr addrspace(1) %59, i64 %61, !dbg !21 + %69 = getelementptr bfloat, ptr addrspace(1) %60, i64 %61, !dbg !21 + %70 = getelementptr float, ptr addrspace(1) %1, i64 %37, !dbg !22 + %71 = getelementptr float, ptr addrspace(1) %1, i64 %39, !dbg !22 + %72 = getelementptr float, ptr addrspace(1) %1, i64 %41, !dbg !22 + %73 = getelementptr float, ptr addrspace(1) %1, i64 %43, !dbg !22 + %74 = getelementptr float, ptr addrspace(1) %1, i64 %45, !dbg !22 + %75 = getelementptr float, ptr addrspace(1) %1, i64 %47, !dbg !22 + %76 = getelementptr float, ptr addrspace(1) %1, i64 %49, !dbg !22 + %77 = getelementptr float, ptr addrspace(1) %1, i64 %51, !dbg !22 + %78 = getelementptr bfloat, ptr addrspace(1) %2, i64 %38, !dbg !23 + %79 = getelementptr bfloat, ptr addrspace(1) %2, i64 %40, !dbg !23 + %80 = getelementptr bfloat, ptr addrspace(1) %2, i64 %42, !dbg !23 + %81 = getelementptr bfloat, ptr addrspace(1) %2, i64 %44, !dbg !23 + %82 = getelementptr bfloat, ptr addrspace(1) %2, i64 %46, !dbg !23 + %83 = getelementptr bfloat, ptr addrspace(1) %2, i64 %48, !dbg !23 + %84 = getelementptr bfloat, ptr addrspace(1) %2, i64 %50, !dbg !23 + %85 = getelementptr bfloat, ptr addrspace(1) %2, i64 %52, !dbg !23 + %86 = getelementptr bfloat, ptr addrspace(1) %78, i64 %61, !dbg !24 + %87 = getelementptr bfloat, ptr addrspace(1) %79, i64 %61, !dbg !24 + %88 = getelementptr bfloat, ptr addrspace(1) %80, i64 %61, !dbg !24 + %89 = getelementptr bfloat, ptr addrspace(1) %81, i64 %61, !dbg !24 + %90 = getelementptr bfloat, ptr addrspace(1) %82, i64 %61, !dbg !24 + %91 = getelementptr bfloat, ptr addrspace(1) %83, i64 %61, !dbg !24 + %92 = getelementptr bfloat, ptr addrspace(1) %84, i64 %61, !dbg !24 + %93 = getelementptr bfloat, ptr addrspace(1) %85, i64 %61, !dbg !24 + %94 = getelementptr bfloat, ptr addrspace(1) %3, i64 %38, !dbg !25 + %95 = getelementptr bfloat, ptr addrspace(1) %3, i64 %40, !dbg !25 + %96 = getelementptr bfloat, ptr addrspace(1) %3, i64 %42, !dbg !25 + %97 = getelementptr bfloat, ptr addrspace(1) %3, i64 %44, !dbg !25 + %98 = getelementptr bfloat, ptr addrspace(1) %3, i64 %46, !dbg !25 + %99 = getelementptr bfloat, ptr addrspace(1) %3, i64 %48, !dbg !25 + %100 = getelementptr bfloat, ptr addrspace(1) %3, i64 %50, !dbg !25 + %101 = getelementptr bfloat, ptr addrspace(1) %3, i64 %52, !dbg !25 + %102 = getelementptr bfloat, ptr addrspace(1) %94, i64 %61, !dbg !26 + %103 = getelementptr bfloat, ptr addrspace(1) %95, i64 %61, !dbg !26 + %104 = getelementptr bfloat, ptr addrspace(1) %96, i64 %61, !dbg !26 + %105 = getelementptr bfloat, ptr addrspace(1) %97, i64 %61, !dbg !26 + %106 = getelementptr bfloat, ptr addrspace(1) %98, i64 %61, !dbg !26 + %107 = getelementptr bfloat, ptr addrspace(1) %99, i64 %61, !dbg !26 + %108 = getelementptr bfloat, ptr addrspace(1) %100, i64 %61, !dbg !26 + %109 = getelementptr bfloat, ptr addrspace(1) %101, i64 %61, !dbg !26 + %110 = extractelement <8 x i1> %34, i64 7, !dbg !27 + %111 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %62, i1 %110) #3, !dbg !27 + %112 = extractvalue { i32, i32, i32, i32 } %111, 0, !dbg !27 + %113 = bitcast i32 %112 to <2 x bfloat>, !dbg !27 + %114 = extractvalue { i32, i32, i32, i32 } %111, 1, !dbg !27 + %115 = bitcast i32 %114 to <2 x bfloat>, !dbg !27 + %116 = extractvalue { i32, i32, i32, i32 } %111, 2, !dbg !27 + %117 = bitcast i32 %116 to <2 x bfloat>, !dbg !27 + %118 = extractvalue { i32, i32, i32, i32 } %111, 3, !dbg !27 + %119 = bitcast i32 %118 to <2 x bfloat>, !dbg !27 + %120 = extractelement <2 x bfloat> %113, i64 0, !dbg !27 + %121 = extractelement <2 x bfloat> %113, i64 1, !dbg !27 + %122 = extractelement <2 x bfloat> %115, i64 0, !dbg !27 + %123 = extractelement <2 x bfloat> %115, i64 1, !dbg !27 + %124 = extractelement <2 x bfloat> %117, i64 0, !dbg !27 + %125 = extractelement <2 x bfloat> %117, i64 1, !dbg !27 + %126 = extractelement <2 x bfloat> %119, i64 0, !dbg !27 + %127 = extractelement <2 x bfloat> %119, i64 1, !dbg !27 + %128 = extractelement <8 x i1> %34, i64 6, !dbg !27 + %129 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %63, i1 %128) #3, !dbg !27 + %130 = extractvalue { i32, i32, i32, i32 } %129, 0, !dbg !27 + %131 = bitcast i32 %130 to <2 x bfloat>, !dbg !27 + %132 = extractvalue { i32, i32, i32, i32 } %129, 1, !dbg !27 + %133 = bitcast i32 %132 to <2 x bfloat>, !dbg !27 + %134 = extractvalue { i32, i32, i32, i32 } %129, 2, !dbg !27 + %135 = bitcast i32 %134 to <2 x bfloat>, !dbg !27 + %136 = extractvalue { i32, i32, i32, i32 } %129, 3, !dbg !27 + %137 = bitcast i32 %136 to <2 x bfloat>, !dbg !27 + %138 = extractelement <2 x bfloat> %131, i64 0, !dbg !27 + %139 = extractelement <2 x bfloat> %131, i64 1, !dbg !27 + %140 = extractelement <2 x bfloat> %133, i64 0, !dbg !27 + %141 = extractelement <2 x bfloat> %133, i64 1, !dbg !27 + %142 = extractelement <2 x bfloat> %135, i64 0, !dbg !27 + %143 = extractelement <2 x bfloat> %135, i64 1, !dbg !27 + %144 = extractelement <2 x bfloat> %137, i64 0, !dbg !27 + %145 = extractelement <2 x bfloat> %137, i64 1, !dbg !27 + %146 = extractelement <8 x i1> %34, i64 5, !dbg !27 + %147 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %64, i1 %146) #3, !dbg !27 + %148 = extractvalue { i32, i32, i32, i32 } %147, 0, !dbg !27 + %149 = bitcast i32 %148 to <2 x bfloat>, !dbg !27 + %150 = extractvalue { i32, i32, i32, i32 } %147, 1, !dbg !27 + %151 = bitcast i32 %150 to <2 x bfloat>, !dbg !27 + %152 = extractvalue { i32, i32, i32, i32 } %147, 2, !dbg !27 + %153 = bitcast i32 %152 to <2 x bfloat>, !dbg !27 + %154 = extractvalue { i32, i32, i32, i32 } %147, 3, !dbg !27 + %155 = bitcast i32 %154 to <2 x bfloat>, !dbg !27 + %156 = extractelement <2 x bfloat> %149, i64 0, !dbg !27 + %157 = extractelement <2 x bfloat> %149, i64 1, !dbg !27 + %158 = extractelement <2 x bfloat> %151, i64 0, !dbg !27 + %159 = extractelement <2 x bfloat> %151, i64 1, !dbg !27 + %160 = extractelement <2 x bfloat> %153, i64 0, !dbg !27 + %161 = extractelement <2 x bfloat> %153, i64 1, !dbg !27 + %162 = extractelement <2 x bfloat> %155, i64 0, !dbg !27 + %163 = extractelement <2 x bfloat> %155, i64 1, !dbg !27 + %164 = extractelement <8 x i1> %34, i64 4, !dbg !27 + %165 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %65, i1 %164) #3, !dbg !27 + %166 = extractvalue { i32, i32, i32, i32 } %165, 0, !dbg !27 + %167 = bitcast i32 %166 to <2 x bfloat>, !dbg !27 + %168 = extractvalue { i32, i32, i32, i32 } %165, 1, !dbg !27 + %169 = bitcast i32 %168 to <2 x bfloat>, !dbg !27 + %170 = extractvalue { i32, i32, i32, i32 } %165, 2, !dbg !27 + %171 = bitcast i32 %170 to <2 x bfloat>, !dbg !27 + %172 = extractvalue { i32, i32, i32, i32 } %165, 3, !dbg !27 + %173 = bitcast i32 %172 to <2 x bfloat>, !dbg !27 + %174 = extractelement <2 x bfloat> %167, i64 0, !dbg !27 + %175 = extractelement <2 x bfloat> %167, i64 1, !dbg !27 + %176 = extractelement <2 x bfloat> %169, i64 0, !dbg !27 + %177 = extractelement <2 x bfloat> %169, i64 1, !dbg !27 + %178 = extractelement <2 x bfloat> %171, i64 0, !dbg !27 + %179 = extractelement <2 x bfloat> %171, i64 1, !dbg !27 + %180 = extractelement <2 x bfloat> %173, i64 0, !dbg !27 + %181 = extractelement <2 x bfloat> %173, i64 1, !dbg !27 + %182 = extractelement <8 x i1> %34, i64 3, !dbg !27 + %183 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %66, i1 %182) #3, !dbg !27 + %184 = extractvalue { i32, i32, i32, i32 } %183, 0, !dbg !27 + %185 = bitcast i32 %184 to <2 x bfloat>, !dbg !27 + %186 = extractvalue { i32, i32, i32, i32 } %183, 1, !dbg !27 + %187 = bitcast i32 %186 to <2 x bfloat>, !dbg !27 + %188 = extractvalue { i32, i32, i32, i32 } %183, 2, !dbg !27 + %189 = bitcast i32 %188 to <2 x bfloat>, !dbg !27 + %190 = extractvalue { i32, i32, i32, i32 } %183, 3, !dbg !27 + %191 = bitcast i32 %190 to <2 x bfloat>, !dbg !27 + %192 = extractelement <2 x bfloat> %185, i64 0, !dbg !27 + %193 = extractelement <2 x bfloat> %185, i64 1, !dbg !27 + %194 = extractelement <2 x bfloat> %187, i64 0, !dbg !27 + %195 = extractelement <2 x bfloat> %187, i64 1, !dbg !27 + %196 = extractelement <2 x bfloat> %189, i64 0, !dbg !27 + %197 = extractelement <2 x bfloat> %189, i64 1, !dbg !27 + %198 = extractelement <2 x bfloat> %191, i64 0, !dbg !27 + %199 = extractelement <2 x bfloat> %191, i64 1, !dbg !27 + %200 = extractelement <8 x i1> %34, i64 2, !dbg !27 + %201 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %67, i1 %200) #3, !dbg !27 + %202 = extractvalue { i32, i32, i32, i32 } %201, 0, !dbg !27 + %203 = bitcast i32 %202 to <2 x bfloat>, !dbg !27 + %204 = extractvalue { i32, i32, i32, i32 } %201, 1, !dbg !27 + %205 = bitcast i32 %204 to <2 x bfloat>, !dbg !27 + %206 = extractvalue { i32, i32, i32, i32 } %201, 2, !dbg !27 + %207 = bitcast i32 %206 to <2 x bfloat>, !dbg !27 + %208 = extractvalue { i32, i32, i32, i32 } %201, 3, !dbg !27 + %209 = bitcast i32 %208 to <2 x bfloat>, !dbg !27 + %210 = extractelement <2 x bfloat> %203, i64 0, !dbg !27 + %211 = extractelement <2 x bfloat> %203, i64 1, !dbg !27 + %212 = extractelement <2 x bfloat> %205, i64 0, !dbg !27 + %213 = extractelement <2 x bfloat> %205, i64 1, !dbg !27 + %214 = extractelement <2 x bfloat> %207, i64 0, !dbg !27 + %215 = extractelement <2 x bfloat> %207, i64 1, !dbg !27 + %216 = extractelement <2 x bfloat> %209, i64 0, !dbg !27 + %217 = extractelement <2 x bfloat> %209, i64 1, !dbg !27 + %218 = extractelement <8 x i1> %34, i64 1, !dbg !27 + %219 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %68, i1 %218) #3, !dbg !27 + %220 = extractvalue { i32, i32, i32, i32 } %219, 0, !dbg !27 + %221 = bitcast i32 %220 to <2 x bfloat>, !dbg !27 + %222 = extractvalue { i32, i32, i32, i32 } %219, 1, !dbg !27 + %223 = bitcast i32 %222 to <2 x bfloat>, !dbg !27 + %224 = extractvalue { i32, i32, i32, i32 } %219, 2, !dbg !27 + %225 = bitcast i32 %224 to <2 x bfloat>, !dbg !27 + %226 = extractvalue { i32, i32, i32, i32 } %219, 3, !dbg !27 + %227 = bitcast i32 %226 to <2 x bfloat>, !dbg !27 + %228 = extractelement <2 x bfloat> %221, i64 0, !dbg !27 + %229 = extractelement <2 x bfloat> %221, i64 1, !dbg !27 + %230 = extractelement <2 x bfloat> %223, i64 0, !dbg !27 + %231 = extractelement <2 x bfloat> %223, i64 1, !dbg !27 + %232 = extractelement <2 x bfloat> %225, i64 0, !dbg !27 + %233 = extractelement <2 x bfloat> %225, i64 1, !dbg !27 + %234 = extractelement <2 x bfloat> %227, i64 0, !dbg !27 + %235 = extractelement <2 x bfloat> %227, i64 1, !dbg !27 + %236 = extractelement <8 x i1> %34, i64 0, !dbg !27 + %237 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %69, i1 %236) #3, !dbg !27 + %238 = extractvalue { i32, i32, i32, i32 } %237, 0, !dbg !27 + %239 = bitcast i32 %238 to <2 x bfloat>, !dbg !27 + %240 = extractvalue { i32, i32, i32, i32 } %237, 1, !dbg !27 + %241 = bitcast i32 %240 to <2 x bfloat>, !dbg !27 + %242 = extractvalue { i32, i32, i32, i32 } %237, 2, !dbg !27 + %243 = bitcast i32 %242 to <2 x bfloat>, !dbg !27 + %244 = extractvalue { i32, i32, i32, i32 } %237, 3, !dbg !27 + %245 = bitcast i32 %244 to <2 x bfloat>, !dbg !27 + %246 = extractelement <2 x bfloat> %239, i64 0, !dbg !27 + %247 = extractelement <2 x bfloat> %239, i64 1, !dbg !27 + %248 = extractelement <2 x bfloat> %241, i64 0, !dbg !27 + %249 = extractelement <2 x bfloat> %241, i64 1, !dbg !27 + %250 = extractelement <2 x bfloat> %243, i64 0, !dbg !27 + %251 = extractelement <2 x bfloat> %243, i64 1, !dbg !27 + %252 = extractelement <2 x bfloat> %245, i64 0, !dbg !27 + %253 = extractelement <2 x bfloat> %245, i64 1, !dbg !27 + %254 = fpext bfloat %120 to float, !dbg !28 + %255 = fpext bfloat %121 to float, !dbg !28 + %256 = fpext bfloat %122 to float, !dbg !28 + %257 = fpext bfloat %123 to float, !dbg !28 + %258 = fpext bfloat %124 to float, !dbg !28 + %259 = fpext bfloat %125 to float, !dbg !28 + %260 = fpext bfloat %126 to float, !dbg !28 + %261 = fpext bfloat %127 to float, !dbg !28 + %262 = fpext bfloat %138 to float, !dbg !28 + %263 = fpext bfloat %139 to float, !dbg !28 + %264 = fpext bfloat %140 to float, !dbg !28 + %265 = fpext bfloat %141 to float, !dbg !28 + %266 = fpext bfloat %142 to float, !dbg !28 + %267 = fpext bfloat %143 to float, !dbg !28 + %268 = fpext bfloat %144 to float, !dbg !28 + %269 = fpext bfloat %145 to float, !dbg !28 + %270 = fpext bfloat %156 to float, !dbg !28 + %271 = fpext bfloat %157 to float, !dbg !28 + %272 = fpext bfloat %158 to float, !dbg !28 + %273 = fpext bfloat %159 to float, !dbg !28 + %274 = fpext bfloat %160 to float, !dbg !28 + %275 = fpext bfloat %161 to float, !dbg !28 + %276 = fpext bfloat %162 to float, !dbg !28 + %277 = fpext bfloat %163 to float, !dbg !28 + %278 = fpext bfloat %174 to float, !dbg !28 + %279 = fpext bfloat %175 to float, !dbg !28 + %280 = fpext bfloat %176 to float, !dbg !28 + %281 = fpext bfloat %177 to float, !dbg !28 + %282 = fpext bfloat %178 to float, !dbg !28 + %283 = fpext bfloat %179 to float, !dbg !28 + %284 = fpext bfloat %180 to float, !dbg !28 + %285 = fpext bfloat %181 to float, !dbg !28 + %286 = fpext bfloat %192 to float, !dbg !28 + %287 = fpext bfloat %193 to float, !dbg !28 + %288 = fpext bfloat %194 to float, !dbg !28 + %289 = fpext bfloat %195 to float, !dbg !28 + %290 = fpext bfloat %196 to float, !dbg !28 + %291 = fpext bfloat %197 to float, !dbg !28 + %292 = fpext bfloat %198 to float, !dbg !28 + %293 = fpext bfloat %199 to float, !dbg !28 + %294 = fpext bfloat %210 to float, !dbg !28 + %295 = fpext bfloat %211 to float, !dbg !28 + %296 = fpext bfloat %212 to float, !dbg !28 + %297 = fpext bfloat %213 to float, !dbg !28 + %298 = fpext bfloat %214 to float, !dbg !28 + %299 = fpext bfloat %215 to float, !dbg !28 + %300 = fpext bfloat %216 to float, !dbg !28 + %301 = fpext bfloat %217 to float, !dbg !28 + %302 = fpext bfloat %228 to float, !dbg !28 + %303 = fpext bfloat %229 to float, !dbg !28 + %304 = fpext bfloat %230 to float, !dbg !28 + %305 = fpext bfloat %231 to float, !dbg !28 + %306 = fpext bfloat %232 to float, !dbg !28 + %307 = fpext bfloat %233 to float, !dbg !28 + %308 = fpext bfloat %234 to float, !dbg !28 + %309 = fpext bfloat %235 to float, !dbg !28 + %310 = fpext bfloat %246 to float, !dbg !28 + %311 = fpext bfloat %247 to float, !dbg !28 + %312 = fpext bfloat %248 to float, !dbg !28 + %313 = fpext bfloat %249 to float, !dbg !28 + %314 = fpext bfloat %250 to float, !dbg !28 + %315 = fpext bfloat %251 to float, !dbg !28 + %316 = fpext bfloat %252 to float, !dbg !28 + %317 = fpext bfloat %253 to float, !dbg !28 + %318 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %70, i1 %110) #3, !dbg !29 + %319 = bitcast i32 %318 to float, !dbg !29 + %320 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %71, i1 %128) #3, !dbg !29 + %321 = bitcast i32 %320 to float, !dbg !29 + %322 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %72, i1 %146) #3, !dbg !29 + %323 = bitcast i32 %322 to float, !dbg !29 + %324 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %73, i1 %164) #3, !dbg !29 + %325 = bitcast i32 %324 to float, !dbg !29 + %326 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %74, i1 %182) #3, !dbg !29 + %327 = bitcast i32 %326 to float, !dbg !29 + %328 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %75, i1 %200) #3, !dbg !29 + %329 = bitcast i32 %328 to float, !dbg !29 + %330 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %76, i1 %218) #3, !dbg !29 + %331 = bitcast i32 %330 to float, !dbg !29 + %332 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %77, i1 %236) #3, !dbg !29 + %333 = bitcast i32 %332 to float, !dbg !29 + %334 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %86, i1 %110) #3, !dbg !30 + %335 = extractvalue { i32, i32, i32, i32 } %334, 0, !dbg !30 + %336 = bitcast i32 %335 to <2 x bfloat>, !dbg !30 + %337 = extractvalue { i32, i32, i32, i32 } %334, 1, !dbg !30 + %338 = bitcast i32 %337 to <2 x bfloat>, !dbg !30 + %339 = extractvalue { i32, i32, i32, i32 } %334, 2, !dbg !30 + %340 = bitcast i32 %339 to <2 x bfloat>, !dbg !30 + %341 = extractvalue { i32, i32, i32, i32 } %334, 3, !dbg !30 + %342 = bitcast i32 %341 to <2 x bfloat>, !dbg !30 + %343 = extractelement <2 x bfloat> %336, i64 0, !dbg !30 + %344 = extractelement <2 x bfloat> %336, i64 1, !dbg !30 + %345 = extractelement <2 x bfloat> %338, i64 0, !dbg !30 + %346 = extractelement <2 x bfloat> %338, i64 1, !dbg !30 + %347 = extractelement <2 x bfloat> %340, i64 0, !dbg !30 + %348 = extractelement <2 x bfloat> %340, i64 1, !dbg !30 + %349 = extractelement <2 x bfloat> %342, i64 0, !dbg !30 + %350 = extractelement <2 x bfloat> %342, i64 1, !dbg !30 + %351 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %87, i1 %128) #3, !dbg !30 + %352 = extractvalue { i32, i32, i32, i32 } %351, 0, !dbg !30 + %353 = bitcast i32 %352 to <2 x bfloat>, !dbg !30 + %354 = extractvalue { i32, i32, i32, i32 } %351, 1, !dbg !30 + %355 = bitcast i32 %354 to <2 x bfloat>, !dbg !30 + %356 = extractvalue { i32, i32, i32, i32 } %351, 2, !dbg !30 + %357 = bitcast i32 %356 to <2 x bfloat>, !dbg !30 + %358 = extractvalue { i32, i32, i32, i32 } %351, 3, !dbg !30 + %359 = bitcast i32 %358 to <2 x bfloat>, !dbg !30 + %360 = extractelement <2 x bfloat> %353, i64 0, !dbg !30 + %361 = extractelement <2 x bfloat> %353, i64 1, !dbg !30 + %362 = extractelement <2 x bfloat> %355, i64 0, !dbg !30 + %363 = extractelement <2 x bfloat> %355, i64 1, !dbg !30 + %364 = extractelement <2 x bfloat> %357, i64 0, !dbg !30 + %365 = extractelement <2 x bfloat> %357, i64 1, !dbg !30 + %366 = extractelement <2 x bfloat> %359, i64 0, !dbg !30 + %367 = extractelement <2 x bfloat> %359, i64 1, !dbg !30 + %368 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %88, i1 %146) #3, !dbg !30 + %369 = extractvalue { i32, i32, i32, i32 } %368, 0, !dbg !30 + %370 = bitcast i32 %369 to <2 x bfloat>, !dbg !30 + %371 = extractvalue { i32, i32, i32, i32 } %368, 1, !dbg !30 + %372 = bitcast i32 %371 to <2 x bfloat>, !dbg !30 + %373 = extractvalue { i32, i32, i32, i32 } %368, 2, !dbg !30 + %374 = bitcast i32 %373 to <2 x bfloat>, !dbg !30 + %375 = extractvalue { i32, i32, i32, i32 } %368, 3, !dbg !30 + %376 = bitcast i32 %375 to <2 x bfloat>, !dbg !30 + %377 = extractelement <2 x bfloat> %370, i64 0, !dbg !30 + %378 = extractelement <2 x bfloat> %370, i64 1, !dbg !30 + %379 = extractelement <2 x bfloat> %372, i64 0, !dbg !30 + %380 = extractelement <2 x bfloat> %372, i64 1, !dbg !30 + %381 = extractelement <2 x bfloat> %374, i64 0, !dbg !30 + %382 = extractelement <2 x bfloat> %374, i64 1, !dbg !30 + %383 = extractelement <2 x bfloat> %376, i64 0, !dbg !30 + %384 = extractelement <2 x bfloat> %376, i64 1, !dbg !30 + %385 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %89, i1 %164) #3, !dbg !30 + %386 = extractvalue { i32, i32, i32, i32 } %385, 0, !dbg !30 + %387 = bitcast i32 %386 to <2 x bfloat>, !dbg !30 + %388 = extractvalue { i32, i32, i32, i32 } %385, 1, !dbg !30 + %389 = bitcast i32 %388 to <2 x bfloat>, !dbg !30 + %390 = extractvalue { i32, i32, i32, i32 } %385, 2, !dbg !30 + %391 = bitcast i32 %390 to <2 x bfloat>, !dbg !30 + %392 = extractvalue { i32, i32, i32, i32 } %385, 3, !dbg !30 + %393 = bitcast i32 %392 to <2 x bfloat>, !dbg !30 + %394 = extractelement <2 x bfloat> %387, i64 0, !dbg !30 + %395 = extractelement <2 x bfloat> %387, i64 1, !dbg !30 + %396 = extractelement <2 x bfloat> %389, i64 0, !dbg !30 + %397 = extractelement <2 x bfloat> %389, i64 1, !dbg !30 + %398 = extractelement <2 x bfloat> %391, i64 0, !dbg !30 + %399 = extractelement <2 x bfloat> %391, i64 1, !dbg !30 + %400 = extractelement <2 x bfloat> %393, i64 0, !dbg !30 + %401 = extractelement <2 x bfloat> %393, i64 1, !dbg !30 + %402 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %90, i1 %182) #3, !dbg !30 + %403 = extractvalue { i32, i32, i32, i32 } %402, 0, !dbg !30 + %404 = bitcast i32 %403 to <2 x bfloat>, !dbg !30 + %405 = extractvalue { i32, i32, i32, i32 } %402, 1, !dbg !30 + %406 = bitcast i32 %405 to <2 x bfloat>, !dbg !30 + %407 = extractvalue { i32, i32, i32, i32 } %402, 2, !dbg !30 + %408 = bitcast i32 %407 to <2 x bfloat>, !dbg !30 + %409 = extractvalue { i32, i32, i32, i32 } %402, 3, !dbg !30 + %410 = bitcast i32 %409 to <2 x bfloat>, !dbg !30 + %411 = extractelement <2 x bfloat> %404, i64 0, !dbg !30 + %412 = extractelement <2 x bfloat> %404, i64 1, !dbg !30 + %413 = extractelement <2 x bfloat> %406, i64 0, !dbg !30 + %414 = extractelement <2 x bfloat> %406, i64 1, !dbg !30 + %415 = extractelement <2 x bfloat> %408, i64 0, !dbg !30 + %416 = extractelement <2 x bfloat> %408, i64 1, !dbg !30 + %417 = extractelement <2 x bfloat> %410, i64 0, !dbg !30 + %418 = extractelement <2 x bfloat> %410, i64 1, !dbg !30 + %419 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %91, i1 %200) #3, !dbg !30 + %420 = extractvalue { i32, i32, i32, i32 } %419, 0, !dbg !30 + %421 = bitcast i32 %420 to <2 x bfloat>, !dbg !30 + %422 = extractvalue { i32, i32, i32, i32 } %419, 1, !dbg !30 + %423 = bitcast i32 %422 to <2 x bfloat>, !dbg !30 + %424 = extractvalue { i32, i32, i32, i32 } %419, 2, !dbg !30 + %425 = bitcast i32 %424 to <2 x bfloat>, !dbg !30 + %426 = extractvalue { i32, i32, i32, i32 } %419, 3, !dbg !30 + %427 = bitcast i32 %426 to <2 x bfloat>, !dbg !30 + %428 = extractelement <2 x bfloat> %421, i64 0, !dbg !30 + %429 = extractelement <2 x bfloat> %421, i64 1, !dbg !30 + %430 = extractelement <2 x bfloat> %423, i64 0, !dbg !30 + %431 = extractelement <2 x bfloat> %423, i64 1, !dbg !30 + %432 = extractelement <2 x bfloat> %425, i64 0, !dbg !30 + %433 = extractelement <2 x bfloat> %425, i64 1, !dbg !30 + %434 = extractelement <2 x bfloat> %427, i64 0, !dbg !30 + %435 = extractelement <2 x bfloat> %427, i64 1, !dbg !30 + %436 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %92, i1 %218) #3, !dbg !30 + %437 = extractvalue { i32, i32, i32, i32 } %436, 0, !dbg !30 + %438 = bitcast i32 %437 to <2 x bfloat>, !dbg !30 + %439 = extractvalue { i32, i32, i32, i32 } %436, 1, !dbg !30 + %440 = bitcast i32 %439 to <2 x bfloat>, !dbg !30 + %441 = extractvalue { i32, i32, i32, i32 } %436, 2, !dbg !30 + %442 = bitcast i32 %441 to <2 x bfloat>, !dbg !30 + %443 = extractvalue { i32, i32, i32, i32 } %436, 3, !dbg !30 + %444 = bitcast i32 %443 to <2 x bfloat>, !dbg !30 + %445 = extractelement <2 x bfloat> %438, i64 0, !dbg !30 + %446 = extractelement <2 x bfloat> %438, i64 1, !dbg !30 + %447 = extractelement <2 x bfloat> %440, i64 0, !dbg !30 + %448 = extractelement <2 x bfloat> %440, i64 1, !dbg !30 + %449 = extractelement <2 x bfloat> %442, i64 0, !dbg !30 + %450 = extractelement <2 x bfloat> %442, i64 1, !dbg !30 + %451 = extractelement <2 x bfloat> %444, i64 0, !dbg !30 + %452 = extractelement <2 x bfloat> %444, i64 1, !dbg !30 + %453 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %93, i1 %236) #3, !dbg !30 + %454 = extractvalue { i32, i32, i32, i32 } %453, 0, !dbg !30 + %455 = bitcast i32 %454 to <2 x bfloat>, !dbg !30 + %456 = extractvalue { i32, i32, i32, i32 } %453, 1, !dbg !30 + %457 = bitcast i32 %456 to <2 x bfloat>, !dbg !30 + %458 = extractvalue { i32, i32, i32, i32 } %453, 2, !dbg !30 + %459 = bitcast i32 %458 to <2 x bfloat>, !dbg !30 + %460 = extractvalue { i32, i32, i32, i32 } %453, 3, !dbg !30 + %461 = bitcast i32 %460 to <2 x bfloat>, !dbg !30 + %462 = extractelement <2 x bfloat> %455, i64 0, !dbg !30 + %463 = extractelement <2 x bfloat> %455, i64 1, !dbg !30 + %464 = extractelement <2 x bfloat> %457, i64 0, !dbg !30 + %465 = extractelement <2 x bfloat> %457, i64 1, !dbg !30 + %466 = extractelement <2 x bfloat> %459, i64 0, !dbg !30 + %467 = extractelement <2 x bfloat> %459, i64 1, !dbg !30 + %468 = extractelement <2 x bfloat> %461, i64 0, !dbg !30 + %469 = extractelement <2 x bfloat> %461, i64 1, !dbg !30 + %470 = fpext bfloat %343 to float, !dbg !31 + %471 = fpext bfloat %344 to float, !dbg !31 + %472 = fpext bfloat %345 to float, !dbg !31 + %473 = fpext bfloat %346 to float, !dbg !31 + %474 = fpext bfloat %347 to float, !dbg !31 + %475 = fpext bfloat %348 to float, !dbg !31 + %476 = fpext bfloat %349 to float, !dbg !31 + %477 = fpext bfloat %350 to float, !dbg !31 + %478 = fpext bfloat %360 to float, !dbg !31 + %479 = fpext bfloat %361 to float, !dbg !31 + %480 = fpext bfloat %362 to float, !dbg !31 + %481 = fpext bfloat %363 to float, !dbg !31 + %482 = fpext bfloat %364 to float, !dbg !31 + %483 = fpext bfloat %365 to float, !dbg !31 + %484 = fpext bfloat %366 to float, !dbg !31 + %485 = fpext bfloat %367 to float, !dbg !31 + %486 = fpext bfloat %377 to float, !dbg !31 + %487 = fpext bfloat %378 to float, !dbg !31 + %488 = fpext bfloat %379 to float, !dbg !31 + %489 = fpext bfloat %380 to float, !dbg !31 + %490 = fpext bfloat %381 to float, !dbg !31 + %491 = fpext bfloat %382 to float, !dbg !31 + %492 = fpext bfloat %383 to float, !dbg !31 + %493 = fpext bfloat %384 to float, !dbg !31 + %494 = fpext bfloat %394 to float, !dbg !31 + %495 = fpext bfloat %395 to float, !dbg !31 + %496 = fpext bfloat %396 to float, !dbg !31 + %497 = fpext bfloat %397 to float, !dbg !31 + %498 = fpext bfloat %398 to float, !dbg !31 + %499 = fpext bfloat %399 to float, !dbg !31 + %500 = fpext bfloat %400 to float, !dbg !31 + %501 = fpext bfloat %401 to float, !dbg !31 + %502 = fpext bfloat %411 to float, !dbg !31 + %503 = fpext bfloat %412 to float, !dbg !31 + %504 = fpext bfloat %413 to float, !dbg !31 + %505 = fpext bfloat %414 to float, !dbg !31 + %506 = fpext bfloat %415 to float, !dbg !31 + %507 = fpext bfloat %416 to float, !dbg !31 + %508 = fpext bfloat %417 to float, !dbg !31 + %509 = fpext bfloat %418 to float, !dbg !31 + %510 = fpext bfloat %428 to float, !dbg !31 + %511 = fpext bfloat %429 to float, !dbg !31 + %512 = fpext bfloat %430 to float, !dbg !31 + %513 = fpext bfloat %431 to float, !dbg !31 + %514 = fpext bfloat %432 to float, !dbg !31 + %515 = fpext bfloat %433 to float, !dbg !31 + %516 = fpext bfloat %434 to float, !dbg !31 + %517 = fpext bfloat %435 to float, !dbg !31 + %518 = fpext bfloat %445 to float, !dbg !31 + %519 = fpext bfloat %446 to float, !dbg !31 + %520 = fpext bfloat %447 to float, !dbg !31 + %521 = fpext bfloat %448 to float, !dbg !31 + %522 = fpext bfloat %449 to float, !dbg !31 + %523 = fpext bfloat %450 to float, !dbg !31 + %524 = fpext bfloat %451 to float, !dbg !31 + %525 = fpext bfloat %452 to float, !dbg !31 + %526 = fpext bfloat %462 to float, !dbg !31 + %527 = fpext bfloat %463 to float, !dbg !31 + %528 = fpext bfloat %464 to float, !dbg !31 + %529 = fpext bfloat %465 to float, !dbg !31 + %530 = fpext bfloat %466 to float, !dbg !31 + %531 = fpext bfloat %467 to float, !dbg !31 + %532 = fpext bfloat %468 to float, !dbg !31 + %533 = fpext bfloat %469 to float, !dbg !31 + %534 = fmul float %319, %470, !dbg !32 + %535 = fmul float %319, %471, !dbg !32 + %536 = fmul float %319, %472, !dbg !32 + %537 = fmul float %319, %473, !dbg !32 + %538 = fmul float %319, %474, !dbg !32 + %539 = fmul float %319, %475, !dbg !32 + %540 = fmul float %319, %476, !dbg !32 + %541 = fmul float %319, %477, !dbg !32 + %542 = fmul float %321, %478, !dbg !32 + %543 = fmul float %321, %479, !dbg !32 + %544 = fmul float %321, %480, !dbg !32 + %545 = fmul float %321, %481, !dbg !32 + %546 = fmul float %321, %482, !dbg !32 + %547 = fmul float %321, %483, !dbg !32 + %548 = fmul float %321, %484, !dbg !32 + %549 = fmul float %321, %485, !dbg !32 + %550 = fmul float %323, %486, !dbg !32 + %551 = fmul float %323, %487, !dbg !32 + %552 = fmul float %323, %488, !dbg !32 + %553 = fmul float %323, %489, !dbg !32 + %554 = fmul float %323, %490, !dbg !32 + %555 = fmul float %323, %491, !dbg !32 + %556 = fmul float %323, %492, !dbg !32 + %557 = fmul float %323, %493, !dbg !32 + %558 = fmul float %325, %494, !dbg !32 + %559 = fmul float %325, %495, !dbg !32 + %560 = fmul float %325, %496, !dbg !32 + %561 = fmul float %325, %497, !dbg !32 + %562 = fmul float %325, %498, !dbg !32 + %563 = fmul float %325, %499, !dbg !32 + %564 = fmul float %325, %500, !dbg !32 + %565 = fmul float %325, %501, !dbg !32 + %566 = fmul float %327, %502, !dbg !32 + %567 = fmul float %327, %503, !dbg !32 + %568 = fmul float %327, %504, !dbg !32 + %569 = fmul float %327, %505, !dbg !32 + %570 = fmul float %327, %506, !dbg !32 + %571 = fmul float %327, %507, !dbg !32 + %572 = fmul float %327, %508, !dbg !32 + %573 = fmul float %327, %509, !dbg !32 + %574 = fmul float %329, %510, !dbg !32 + %575 = fmul float %329, %511, !dbg !32 + %576 = fmul float %329, %512, !dbg !32 + %577 = fmul float %329, %513, !dbg !32 + %578 = fmul float %329, %514, !dbg !32 + %579 = fmul float %329, %515, !dbg !32 + %580 = fmul float %329, %516, !dbg !32 + %581 = fmul float %329, %517, !dbg !32 + %582 = fmul float %331, %518, !dbg !32 + %583 = fmul float %331, %519, !dbg !32 + %584 = fmul float %331, %520, !dbg !32 + %585 = fmul float %331, %521, !dbg !32 + %586 = fmul float %331, %522, !dbg !32 + %587 = fmul float %331, %523, !dbg !32 + %588 = fmul float %331, %524, !dbg !32 + %589 = fmul float %331, %525, !dbg !32 + %590 = fmul float %333, %526, !dbg !32 + %591 = fmul float %333, %527, !dbg !32 + %592 = fmul float %333, %528, !dbg !32 + %593 = fmul float %333, %529, !dbg !32 + %594 = fmul float %333, %530, !dbg !32 + %595 = fmul float %333, %531, !dbg !32 + %596 = fmul float %333, %532, !dbg !32 + %597 = fmul float %333, %533, !dbg !32 + %598 = fmul float %254, %470, !dbg !33 + %599 = fmul float %255, %471, !dbg !33 + %600 = fmul float %256, %472, !dbg !33 + %601 = fmul float %257, %473, !dbg !33 + %602 = fmul float %258, %474, !dbg !33 + %603 = fmul float %259, %475, !dbg !33 + %604 = fmul float %260, %476, !dbg !33 + %605 = fmul float %261, %477, !dbg !33 + %606 = fmul float %262, %478, !dbg !33 + %607 = fmul float %263, %479, !dbg !33 + %608 = fmul float %264, %480, !dbg !33 + %609 = fmul float %265, %481, !dbg !33 + %610 = fmul float %266, %482, !dbg !33 + %611 = fmul float %267, %483, !dbg !33 + %612 = fmul float %268, %484, !dbg !33 + %613 = fmul float %269, %485, !dbg !33 + %614 = fmul float %270, %486, !dbg !33 + %615 = fmul float %271, %487, !dbg !33 + %616 = fmul float %272, %488, !dbg !33 + %617 = fmul float %273, %489, !dbg !33 + %618 = fmul float %274, %490, !dbg !33 + %619 = fmul float %275, %491, !dbg !33 + %620 = fmul float %276, %492, !dbg !33 + %621 = fmul float %277, %493, !dbg !33 + %622 = fmul float %278, %494, !dbg !33 + %623 = fmul float %279, %495, !dbg !33 + %624 = fmul float %280, %496, !dbg !33 + %625 = fmul float %281, %497, !dbg !33 + %626 = fmul float %282, %498, !dbg !33 + %627 = fmul float %283, %499, !dbg !33 + %628 = fmul float %284, %500, !dbg !33 + %629 = fmul float %285, %501, !dbg !33 + %630 = fmul float %286, %502, !dbg !33 + %631 = fmul float %287, %503, !dbg !33 + %632 = fmul float %288, %504, !dbg !33 + %633 = fmul float %289, %505, !dbg !33 + %634 = fmul float %290, %506, !dbg !33 + %635 = fmul float %291, %507, !dbg !33 + %636 = fmul float %292, %508, !dbg !33 + %637 = fmul float %293, %509, !dbg !33 + %638 = fmul float %294, %510, !dbg !33 + %639 = fmul float %295, %511, !dbg !33 + %640 = fmul float %296, %512, !dbg !33 + %641 = fmul float %297, %513, !dbg !33 + %642 = fmul float %298, %514, !dbg !33 + %643 = fmul float %299, %515, !dbg !33 + %644 = fmul float %300, %516, !dbg !33 + %645 = fmul float %301, %517, !dbg !33 + %646 = fmul float %302, %518, !dbg !33 + %647 = fmul float %303, %519, !dbg !33 + %648 = fmul float %304, %520, !dbg !33 + %649 = fmul float %305, %521, !dbg !33 + %650 = fmul float %306, %522, !dbg !33 + %651 = fmul float %307, %523, !dbg !33 + %652 = fmul float %308, %524, !dbg !33 + %653 = fmul float %309, %525, !dbg !33 + %654 = fmul float %310, %526, !dbg !33 + %655 = fmul float %311, %527, !dbg !33 + %656 = fmul float %312, %528, !dbg !33 + %657 = fmul float %313, %529, !dbg !33 + %658 = fmul float %314, %530, !dbg !33 + %659 = fmul float %315, %531, !dbg !33 + %660 = fmul float %316, %532, !dbg !33 + %661 = fmul float %317, %533, !dbg !33 + %662 = fadd float %598, %599, !dbg !34 + %663 = fadd float %600, %662, !dbg !34 + %664 = fadd float %601, %663, !dbg !34 + %665 = fadd float %602, %664, !dbg !34 + %666 = fadd float %603, %665, !dbg !34 + %667 = fadd float %604, %666, !dbg !34 + %668 = fadd float %605, %667, !dbg !34 + %669 = fadd float %606, %607, !dbg !34 + %670 = fadd float %608, %669, !dbg !34 + %671 = fadd float %609, %670, !dbg !34 + %672 = fadd float %610, %671, !dbg !34 + %673 = fadd float %611, %672, !dbg !34 + %674 = fadd float %612, %673, !dbg !34 + %675 = fadd float %613, %674, !dbg !34 + %676 = fadd float %614, %615, !dbg !34 + %677 = fadd float %616, %676, !dbg !34 + %678 = fadd float %617, %677, !dbg !34 + %679 = fadd float %618, %678, !dbg !34 + %680 = fadd float %619, %679, !dbg !34 + %681 = fadd float %620, %680, !dbg !34 + %682 = fadd float %621, %681, !dbg !34 + %683 = fadd float %622, %623, !dbg !34 + %684 = fadd float %624, %683, !dbg !34 + %685 = fadd float %625, %684, !dbg !34 + %686 = fadd float %626, %685, !dbg !34 + %687 = fadd float %627, %686, !dbg !34 + %688 = fadd float %628, %687, !dbg !34 + %689 = fadd float %629, %688, !dbg !34 + %690 = fadd float %630, %631, !dbg !34 + %691 = fadd float %632, %690, !dbg !34 + %692 = fadd float %633, %691, !dbg !34 + %693 = fadd float %634, %692, !dbg !34 + %694 = fadd float %635, %693, !dbg !34 + %695 = fadd float %636, %694, !dbg !34 + %696 = fadd float %637, %695, !dbg !34 + %697 = fadd float %638, %639, !dbg !34 + %698 = fadd float %640, %697, !dbg !34 + %699 = fadd float %641, %698, !dbg !34 + %700 = fadd float %642, %699, !dbg !34 + %701 = fadd float %643, %700, !dbg !34 + %702 = fadd float %644, %701, !dbg !34 + %703 = fadd float %645, %702, !dbg !34 + %704 = fadd float %646, %647, !dbg !34 + %705 = fadd float %648, %704, !dbg !34 + %706 = fadd float %649, %705, !dbg !34 + %707 = fadd float %650, %706, !dbg !34 + %708 = fadd float %651, %707, !dbg !34 + %709 = fadd float %652, %708, !dbg !34 + %710 = fadd float %653, %709, !dbg !34 + %711 = fadd float %654, %655, !dbg !34 + %712 = fadd float %656, %711, !dbg !34 + %713 = fadd float %657, %712, !dbg !34 + %714 = fadd float %658, %713, !dbg !34 + %715 = fadd float %659, %714, !dbg !34 + %716 = fadd float %660, %715, !dbg !34 + %717 = fadd float %661, %716, !dbg !34 + %718 = bitcast float %668 to i32, !dbg !37 + %719 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %718, i32 8, i32 31), !dbg !37 + %720 = bitcast i32 %719 to float, !dbg !37 + %721 = fadd float %668, %720, !dbg !34 + %722 = bitcast float %721 to i32, !dbg !37 + %723 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %722, i32 4, i32 31), !dbg !37 + %724 = bitcast i32 %723 to float, !dbg !37 + %725 = fadd float %721, %724, !dbg !34 + %726 = bitcast float %725 to i32, !dbg !37 + %727 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %726, i32 2, i32 31), !dbg !37 + %728 = bitcast i32 %727 to float, !dbg !37 + %729 = fadd float %725, %728, !dbg !34 + %730 = bitcast float %729 to i32, !dbg !37 + %731 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %730, i32 1, i32 31), !dbg !37 + %732 = bitcast i32 %731 to float, !dbg !37 + %733 = fadd float %729, %732, !dbg !34 + %734 = bitcast float %675 to i32, !dbg !37 + %735 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %734, i32 8, i32 31), !dbg !37 + %736 = bitcast i32 %735 to float, !dbg !37 + %737 = fadd float %675, %736, !dbg !34 + %738 = bitcast float %737 to i32, !dbg !37 + %739 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %738, i32 4, i32 31), !dbg !37 + %740 = bitcast i32 %739 to float, !dbg !37 + %741 = fadd float %737, %740, !dbg !34 + %742 = bitcast float %741 to i32, !dbg !37 + %743 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %742, i32 2, i32 31), !dbg !37 + %744 = bitcast i32 %743 to float, !dbg !37 + %745 = fadd float %741, %744, !dbg !34 + %746 = bitcast float %745 to i32, !dbg !37 + %747 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %746, i32 1, i32 31), !dbg !37 + %748 = bitcast i32 %747 to float, !dbg !37 + %749 = fadd float %745, %748, !dbg !34 + %750 = bitcast float %682 to i32, !dbg !37 + %751 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %750, i32 8, i32 31), !dbg !37 + %752 = bitcast i32 %751 to float, !dbg !37 + %753 = fadd float %682, %752, !dbg !34 + %754 = bitcast float %753 to i32, !dbg !37 + %755 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %754, i32 4, i32 31), !dbg !37 + %756 = bitcast i32 %755 to float, !dbg !37 + %757 = fadd float %753, %756, !dbg !34 + %758 = bitcast float %757 to i32, !dbg !37 + %759 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %758, i32 2, i32 31), !dbg !37 + %760 = bitcast i32 %759 to float, !dbg !37 + %761 = fadd float %757, %760, !dbg !34 + %762 = bitcast float %761 to i32, !dbg !37 + %763 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %762, i32 1, i32 31), !dbg !37 + %764 = bitcast i32 %763 to float, !dbg !37 + %765 = fadd float %761, %764, !dbg !34 + %766 = bitcast float %689 to i32, !dbg !37 + %767 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %766, i32 8, i32 31), !dbg !37 + %768 = bitcast i32 %767 to float, !dbg !37 + %769 = fadd float %689, %768, !dbg !34 + %770 = bitcast float %769 to i32, !dbg !37 + %771 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %770, i32 4, i32 31), !dbg !37 + %772 = bitcast i32 %771 to float, !dbg !37 + %773 = fadd float %769, %772, !dbg !34 + %774 = bitcast float %773 to i32, !dbg !37 + %775 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %774, i32 2, i32 31), !dbg !37 + %776 = bitcast i32 %775 to float, !dbg !37 + %777 = fadd float %773, %776, !dbg !34 + %778 = bitcast float %777 to i32, !dbg !37 + %779 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %778, i32 1, i32 31), !dbg !37 + %780 = bitcast i32 %779 to float, !dbg !37 + %781 = fadd float %777, %780, !dbg !34 + %782 = bitcast float %696 to i32, !dbg !37 + %783 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %782, i32 8, i32 31), !dbg !37 + %784 = bitcast i32 %783 to float, !dbg !37 + %785 = fadd float %696, %784, !dbg !34 + %786 = bitcast float %785 to i32, !dbg !37 + %787 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %786, i32 4, i32 31), !dbg !37 + %788 = bitcast i32 %787 to float, !dbg !37 + %789 = fadd float %785, %788, !dbg !34 + %790 = bitcast float %789 to i32, !dbg !37 + %791 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %790, i32 2, i32 31), !dbg !37 + %792 = bitcast i32 %791 to float, !dbg !37 + %793 = fadd float %789, %792, !dbg !34 + %794 = bitcast float %793 to i32, !dbg !37 + %795 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %794, i32 1, i32 31), !dbg !37 + %796 = bitcast i32 %795 to float, !dbg !37 + %797 = fadd float %793, %796, !dbg !34 + %798 = bitcast float %703 to i32, !dbg !37 + %799 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %798, i32 8, i32 31), !dbg !37 + %800 = bitcast i32 %799 to float, !dbg !37 + %801 = fadd float %703, %800, !dbg !34 + %802 = bitcast float %801 to i32, !dbg !37 + %803 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %802, i32 4, i32 31), !dbg !37 + %804 = bitcast i32 %803 to float, !dbg !37 + %805 = fadd float %801, %804, !dbg !34 + %806 = bitcast float %805 to i32, !dbg !37 + %807 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %806, i32 2, i32 31), !dbg !37 + %808 = bitcast i32 %807 to float, !dbg !37 + %809 = fadd float %805, %808, !dbg !34 + %810 = bitcast float %809 to i32, !dbg !37 + %811 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %810, i32 1, i32 31), !dbg !37 + %812 = bitcast i32 %811 to float, !dbg !37 + %813 = fadd float %809, %812, !dbg !34 + %814 = bitcast float %710 to i32, !dbg !37 + %815 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %814, i32 8, i32 31), !dbg !37 + %816 = bitcast i32 %815 to float, !dbg !37 + %817 = fadd float %710, %816, !dbg !34 + %818 = bitcast float %817 to i32, !dbg !37 + %819 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %818, i32 4, i32 31), !dbg !37 + %820 = bitcast i32 %819 to float, !dbg !37 + %821 = fadd float %817, %820, !dbg !34 + %822 = bitcast float %821 to i32, !dbg !37 + %823 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %822, i32 2, i32 31), !dbg !37 + %824 = bitcast i32 %823 to float, !dbg !37 + %825 = fadd float %821, %824, !dbg !34 + %826 = bitcast float %825 to i32, !dbg !37 + %827 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %826, i32 1, i32 31), !dbg !37 + %828 = bitcast i32 %827 to float, !dbg !37 + %829 = fadd float %825, %828, !dbg !34 + %830 = bitcast float %717 to i32, !dbg !37 + %831 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %830, i32 8, i32 31), !dbg !37 + %832 = bitcast i32 %831 to float, !dbg !37 + %833 = fadd float %717, %832, !dbg !34 + %834 = bitcast float %833 to i32, !dbg !37 + %835 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %834, i32 4, i32 31), !dbg !37 + %836 = bitcast i32 %835 to float, !dbg !37 + %837 = fadd float %833, %836, !dbg !34 + %838 = bitcast float %837 to i32, !dbg !37 + %839 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %838, i32 2, i32 31), !dbg !37 + %840 = bitcast i32 %839 to float, !dbg !37 + %841 = fadd float %837, %840, !dbg !34 + %842 = bitcast float %841 to i32, !dbg !37 + %843 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %842, i32 1, i32 31), !dbg !37 + %844 = bitcast i32 %843 to float, !dbg !37 + %845 = fadd float %841, %844, !dbg !34 + %846 = fmul float %733, %254, !dbg !40 + %847 = fmul float %733, %255, !dbg !40 + %848 = fmul float %733, %256, !dbg !40 + %849 = fmul float %733, %257, !dbg !40 + %850 = fmul float %733, %258, !dbg !40 + %851 = fmul float %733, %259, !dbg !40 + %852 = fmul float %733, %260, !dbg !40 + %853 = fmul float %733, %261, !dbg !40 + %854 = fmul float %749, %262, !dbg !40 + %855 = fmul float %749, %263, !dbg !40 + %856 = fmul float %749, %264, !dbg !40 + %857 = fmul float %749, %265, !dbg !40 + %858 = fmul float %749, %266, !dbg !40 + %859 = fmul float %749, %267, !dbg !40 + %860 = fmul float %749, %268, !dbg !40 + %861 = fmul float %749, %269, !dbg !40 + %862 = fmul float %765, %270, !dbg !40 + %863 = fmul float %765, %271, !dbg !40 + %864 = fmul float %765, %272, !dbg !40 + %865 = fmul float %765, %273, !dbg !40 + %866 = fmul float %765, %274, !dbg !40 + %867 = fmul float %765, %275, !dbg !40 + %868 = fmul float %765, %276, !dbg !40 + %869 = fmul float %765, %277, !dbg !40 + %870 = fmul float %781, %278, !dbg !40 + %871 = fmul float %781, %279, !dbg !40 + %872 = fmul float %781, %280, !dbg !40 + %873 = fmul float %781, %281, !dbg !40 + %874 = fmul float %781, %282, !dbg !40 + %875 = fmul float %781, %283, !dbg !40 + %876 = fmul float %781, %284, !dbg !40 + %877 = fmul float %781, %285, !dbg !40 + %878 = fmul float %797, %286, !dbg !40 + %879 = fmul float %797, %287, !dbg !40 + %880 = fmul float %797, %288, !dbg !40 + %881 = fmul float %797, %289, !dbg !40 + %882 = fmul float %797, %290, !dbg !40 + %883 = fmul float %797, %291, !dbg !40 + %884 = fmul float %797, %292, !dbg !40 + %885 = fmul float %797, %293, !dbg !40 + %886 = fmul float %813, %294, !dbg !40 + %887 = fmul float %813, %295, !dbg !40 + %888 = fmul float %813, %296, !dbg !40 + %889 = fmul float %813, %297, !dbg !40 + %890 = fmul float %813, %298, !dbg !40 + %891 = fmul float %813, %299, !dbg !40 + %892 = fmul float %813, %300, !dbg !40 + %893 = fmul float %813, %301, !dbg !40 + %894 = fmul float %829, %302, !dbg !40 + %895 = fmul float %829, %303, !dbg !40 + %896 = fmul float %829, %304, !dbg !40 + %897 = fmul float %829, %305, !dbg !40 + %898 = fmul float %829, %306, !dbg !40 + %899 = fmul float %829, %307, !dbg !40 + %900 = fmul float %829, %308, !dbg !40 + %901 = fmul float %829, %309, !dbg !40 + %902 = fmul float %845, %310, !dbg !40 + %903 = fmul float %845, %311, !dbg !40 + %904 = fmul float %845, %312, !dbg !40 + %905 = fmul float %845, %313, !dbg !40 + %906 = fmul float %845, %314, !dbg !40 + %907 = fmul float %845, %315, !dbg !40 + %908 = fmul float %845, %316, !dbg !40 + %909 = fmul float %845, %317, !dbg !40 + %910 = fmul float %846, %319, !dbg !41 + %911 = fmul float %847, %319, !dbg !41 + %912 = fmul float %848, %319, !dbg !41 + %913 = fmul float %849, %319, !dbg !41 + %914 = fmul float %850, %319, !dbg !41 + %915 = fmul float %851, %319, !dbg !41 + %916 = fmul float %852, %319, !dbg !41 + %917 = fmul float %853, %319, !dbg !41 + %918 = fmul float %854, %321, !dbg !41 + %919 = fmul float %855, %321, !dbg !41 + %920 = fmul float %856, %321, !dbg !41 + %921 = fmul float %857, %321, !dbg !41 + %922 = fmul float %858, %321, !dbg !41 + %923 = fmul float %859, %321, !dbg !41 + %924 = fmul float %860, %321, !dbg !41 + %925 = fmul float %861, %321, !dbg !41 + %926 = fmul float %862, %323, !dbg !41 + %927 = fmul float %863, %323, !dbg !41 + %928 = fmul float %864, %323, !dbg !41 + %929 = fmul float %865, %323, !dbg !41 + %930 = fmul float %866, %323, !dbg !41 + %931 = fmul float %867, %323, !dbg !41 + %932 = fmul float %868, %323, !dbg !41 + %933 = fmul float %869, %323, !dbg !41 + %934 = fmul float %870, %325, !dbg !41 + %935 = fmul float %871, %325, !dbg !41 + %936 = fmul float %872, %325, !dbg !41 + %937 = fmul float %873, %325, !dbg !41 + %938 = fmul float %874, %325, !dbg !41 + %939 = fmul float %875, %325, !dbg !41 + %940 = fmul float %876, %325, !dbg !41 + %941 = fmul float %877, %325, !dbg !41 + %942 = fmul float %878, %327, !dbg !41 + %943 = fmul float %879, %327, !dbg !41 + %944 = fmul float %880, %327, !dbg !41 + %945 = fmul float %881, %327, !dbg !41 + %946 = fmul float %882, %327, !dbg !41 + %947 = fmul float %883, %327, !dbg !41 + %948 = fmul float %884, %327, !dbg !41 + %949 = fmul float %885, %327, !dbg !41 + %950 = fmul float %886, %329, !dbg !41 + %951 = fmul float %887, %329, !dbg !41 + %952 = fmul float %888, %329, !dbg !41 + %953 = fmul float %889, %329, !dbg !41 + %954 = fmul float %890, %329, !dbg !41 + %955 = fmul float %891, %329, !dbg !41 + %956 = fmul float %892, %329, !dbg !41 + %957 = fmul float %893, %329, !dbg !41 + %958 = fmul float %894, %331, !dbg !41 + %959 = fmul float %895, %331, !dbg !41 + %960 = fmul float %896, %331, !dbg !41 + %961 = fmul float %897, %331, !dbg !41 + %962 = fmul float %898, %331, !dbg !41 + %963 = fmul float %899, %331, !dbg !41 + %964 = fmul float %900, %331, !dbg !41 + %965 = fmul float %901, %331, !dbg !41 + %966 = fmul float %902, %333, !dbg !41 + %967 = fmul float %903, %333, !dbg !41 + %968 = fmul float %904, %333, !dbg !41 + %969 = fmul float %905, %333, !dbg !41 + %970 = fmul float %906, %333, !dbg !41 + %971 = fmul float %907, %333, !dbg !41 + %972 = fmul float %908, %333, !dbg !41 + %973 = fmul float %909, %333, !dbg !41 + %974 = fsub float %534, %910, !dbg !42 + %975 = fsub float %535, %911, !dbg !42 + %976 = fsub float %536, %912, !dbg !42 + %977 = fsub float %537, %913, !dbg !42 + %978 = fsub float %538, %914, !dbg !42 + %979 = fsub float %539, %915, !dbg !42 + %980 = fsub float %540, %916, !dbg !42 + %981 = fsub float %541, %917, !dbg !42 + %982 = fsub float %542, %918, !dbg !42 + %983 = fsub float %543, %919, !dbg !42 + %984 = fsub float %544, %920, !dbg !42 + %985 = fsub float %545, %921, !dbg !42 + %986 = fsub float %546, %922, !dbg !42 + %987 = fsub float %547, %923, !dbg !42 + %988 = fsub float %548, %924, !dbg !42 + %989 = fsub float %549, %925, !dbg !42 + %990 = fsub float %550, %926, !dbg !42 + %991 = fsub float %551, %927, !dbg !42 + %992 = fsub float %552, %928, !dbg !42 + %993 = fsub float %553, %929, !dbg !42 + %994 = fsub float %554, %930, !dbg !42 + %995 = fsub float %555, %931, !dbg !42 + %996 = fsub float %556, %932, !dbg !42 + %997 = fsub float %557, %933, !dbg !42 + %998 = fsub float %558, %934, !dbg !42 + %999 = fsub float %559, %935, !dbg !42 + %1000 = fsub float %560, %936, !dbg !42 + %1001 = fsub float %561, %937, !dbg !42 + %1002 = fsub float %562, %938, !dbg !42 + %1003 = fsub float %563, %939, !dbg !42 + %1004 = fsub float %564, %940, !dbg !42 + %1005 = fsub float %565, %941, !dbg !42 + %1006 = fsub float %566, %942, !dbg !42 + %1007 = fsub float %567, %943, !dbg !42 + %1008 = fsub float %568, %944, !dbg !42 + %1009 = fsub float %569, %945, !dbg !42 + %1010 = fsub float %570, %946, !dbg !42 + %1011 = fsub float %571, %947, !dbg !42 + %1012 = fsub float %572, %948, !dbg !42 + %1013 = fsub float %573, %949, !dbg !42 + %1014 = fsub float %574, %950, !dbg !42 + %1015 = fsub float %575, %951, !dbg !42 + %1016 = fsub float %576, %952, !dbg !42 + %1017 = fsub float %577, %953, !dbg !42 + %1018 = fsub float %578, %954, !dbg !42 + %1019 = fsub float %579, %955, !dbg !42 + %1020 = fsub float %580, %956, !dbg !42 + %1021 = fsub float %581, %957, !dbg !42 + %1022 = insertelement <2 x float> poison, float %582, i64 0, !dbg !42 + %1023 = insertelement <2 x float> %1022, float %583, i64 1, !dbg !42 + %1024 = insertelement <2 x float> poison, float %958, i64 0, !dbg !42 + %1025 = insertelement <2 x float> %1024, float %959, i64 1, !dbg !42 + %1026 = fsub <2 x float> %1023, %1025, !dbg !42 + %1027 = insertelement <2 x float> poison, float %584, i64 0, !dbg !42 + %1028 = insertelement <2 x float> %1027, float %585, i64 1, !dbg !42 + %1029 = insertelement <2 x float> poison, float %960, i64 0, !dbg !42 + %1030 = insertelement <2 x float> %1029, float %961, i64 1, !dbg !42 + %1031 = fsub <2 x float> %1028, %1030, !dbg !42 + %1032 = insertelement <2 x float> poison, float %586, i64 0, !dbg !42 + %1033 = insertelement <2 x float> %1032, float %587, i64 1, !dbg !42 + %1034 = insertelement <2 x float> poison, float %962, i64 0, !dbg !42 + %1035 = insertelement <2 x float> %1034, float %963, i64 1, !dbg !42 + %1036 = fsub <2 x float> %1033, %1035, !dbg !42 + %1037 = insertelement <2 x float> poison, float %588, i64 0, !dbg !42 + %1038 = insertelement <2 x float> %1037, float %589, i64 1, !dbg !42 + %1039 = insertelement <2 x float> poison, float %964, i64 0, !dbg !42 + %1040 = insertelement <2 x float> %1039, float %965, i64 1, !dbg !42 + %1041 = fsub <2 x float> %1038, %1040, !dbg !42 + %1042 = insertelement <2 x float> poison, float %590, i64 0, !dbg !42 + %1043 = insertelement <2 x float> %1042, float %591, i64 1, !dbg !42 + %1044 = insertelement <2 x float> poison, float %966, i64 0, !dbg !42 + %1045 = insertelement <2 x float> %1044, float %967, i64 1, !dbg !42 + %1046 = fsub <2 x float> %1043, %1045, !dbg !42 + %1047 = insertelement <2 x float> poison, float %592, i64 0, !dbg !42 + %1048 = insertelement <2 x float> %1047, float %593, i64 1, !dbg !42 + %1049 = insertelement <2 x float> poison, float %968, i64 0, !dbg !42 + %1050 = insertelement <2 x float> %1049, float %969, i64 1, !dbg !42 + %1051 = fsub <2 x float> %1048, %1050, !dbg !42 + %1052 = insertelement <2 x float> poison, float %594, i64 0, !dbg !42 + %1053 = insertelement <2 x float> %1052, float %595, i64 1, !dbg !42 + %1054 = insertelement <2 x float> poison, float %970, i64 0, !dbg !42 + %1055 = insertelement <2 x float> %1054, float %971, i64 1, !dbg !42 + %1056 = fsub <2 x float> %1053, %1055, !dbg !42 + %1057 = insertelement <2 x float> poison, float %596, i64 0, !dbg !42 + %1058 = insertelement <2 x float> %1057, float %597, i64 1, !dbg !42 + %1059 = insertelement <2 x float> poison, float %972, i64 0, !dbg !42 + %1060 = insertelement <2 x float> %1059, float %973, i64 1, !dbg !42 + %1061 = fsub <2 x float> %1058, %1060, !dbg !42 + %1062 = fptrunc float %974 to bfloat, !dbg !43 + %1063 = fptrunc float %975 to bfloat, !dbg !43 + %1064 = fptrunc float %976 to bfloat, !dbg !43 + %1065 = fptrunc float %977 to bfloat, !dbg !43 + %1066 = fptrunc float %978 to bfloat, !dbg !43 + %1067 = fptrunc float %979 to bfloat, !dbg !43 + %1068 = fptrunc float %980 to bfloat, !dbg !43 + %1069 = fptrunc float %981 to bfloat, !dbg !43 + %1070 = fptrunc float %982 to bfloat, !dbg !43 + %1071 = fptrunc float %983 to bfloat, !dbg !43 + %1072 = fptrunc float %984 to bfloat, !dbg !43 + %1073 = fptrunc float %985 to bfloat, !dbg !43 + %1074 = fptrunc float %986 to bfloat, !dbg !43 + %1075 = fptrunc float %987 to bfloat, !dbg !43 + %1076 = fptrunc float %988 to bfloat, !dbg !43 + %1077 = fptrunc float %989 to bfloat, !dbg !43 + %1078 = fptrunc float %990 to bfloat, !dbg !43 + %1079 = fptrunc float %991 to bfloat, !dbg !43 + %1080 = fptrunc float %992 to bfloat, !dbg !43 + %1081 = fptrunc float %993 to bfloat, !dbg !43 + %1082 = fptrunc float %994 to bfloat, !dbg !43 + %1083 = fptrunc float %995 to bfloat, !dbg !43 + %1084 = fptrunc float %996 to bfloat, !dbg !43 + %1085 = fptrunc float %997 to bfloat, !dbg !43 + %1086 = fptrunc float %998 to bfloat, !dbg !43 + %1087 = fptrunc float %999 to bfloat, !dbg !43 + %1088 = fptrunc float %1000 to bfloat, !dbg !43 + %1089 = fptrunc float %1001 to bfloat, !dbg !43 + %1090 = fptrunc float %1002 to bfloat, !dbg !43 + %1091 = fptrunc float %1003 to bfloat, !dbg !43 + %1092 = fptrunc float %1004 to bfloat, !dbg !43 + %1093 = fptrunc float %1005 to bfloat, !dbg !43 + %1094 = fptrunc float %1006 to bfloat, !dbg !43 + %1095 = fptrunc float %1007 to bfloat, !dbg !43 + %1096 = fptrunc float %1008 to bfloat, !dbg !43 + %1097 = fptrunc float %1009 to bfloat, !dbg !43 + %1098 = fptrunc float %1010 to bfloat, !dbg !43 + %1099 = fptrunc float %1011 to bfloat, !dbg !43 + %1100 = fptrunc float %1012 to bfloat, !dbg !43 + %1101 = fptrunc float %1013 to bfloat, !dbg !43 + %1102 = fptrunc float %1014 to bfloat, !dbg !43 + %1103 = fptrunc float %1015 to bfloat, !dbg !43 + %1104 = fptrunc float %1016 to bfloat, !dbg !43 + %1105 = fptrunc float %1017 to bfloat, !dbg !43 + %1106 = fptrunc float %1018 to bfloat, !dbg !43 + %1107 = fptrunc float %1019 to bfloat, !dbg !43 + %1108 = fptrunc float %1020 to bfloat, !dbg !43 + %1109 = fptrunc float %1021 to bfloat, !dbg !43 + %1110 = fptrunc <2 x float> %1026 to <2 x bfloat>, !dbg !43 + %1111 = fptrunc <2 x float> %1031 to <2 x bfloat>, !dbg !43 + %1112 = fptrunc <2 x float> %1036 to <2 x bfloat>, !dbg !43 + %1113 = fptrunc <2 x float> %1041 to <2 x bfloat>, !dbg !43 + %1114 = fptrunc <2 x float> %1046 to <2 x bfloat>, !dbg !43 + %1115 = fptrunc <2 x float> %1051 to <2 x bfloat>, !dbg !43 + %1116 = fptrunc <2 x float> %1056 to <2 x bfloat>, !dbg !43 + %1117 = fptrunc <2 x float> %1061 to <2 x bfloat>, !dbg !43 + %1118 = insertelement <2 x bfloat> poison, bfloat %1062, i64 0, !dbg !44 + %1119 = insertelement <2 x bfloat> %1118, bfloat %1063, i64 1, !dbg !44 + %1120 = bitcast <2 x bfloat> %1119 to i32, !dbg !44 + %1121 = insertelement <2 x bfloat> poison, bfloat %1064, i64 0, !dbg !44 + %1122 = insertelement <2 x bfloat> %1121, bfloat %1065, i64 1, !dbg !44 + %1123 = bitcast <2 x bfloat> %1122 to i32, !dbg !44 + %1124 = insertelement <2 x bfloat> poison, bfloat %1066, i64 0, !dbg !44 + %1125 = insertelement <2 x bfloat> %1124, bfloat %1067, i64 1, !dbg !44 + %1126 = bitcast <2 x bfloat> %1125 to i32, !dbg !44 + %1127 = insertelement <2 x bfloat> poison, bfloat %1068, i64 0, !dbg !44 + %1128 = insertelement <2 x bfloat> %1127, bfloat %1069, i64 1, !dbg !44 + %1129 = bitcast <2 x bfloat> %1128 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1120, i32 %1123, i32 %1126, i32 %1129, ptr addrspace(1) %102, i1 %110) #3, !dbg !44 + %1130 = insertelement <2 x bfloat> poison, bfloat %1070, i64 0, !dbg !44 + %1131 = insertelement <2 x bfloat> %1130, bfloat %1071, i64 1, !dbg !44 + %1132 = bitcast <2 x bfloat> %1131 to i32, !dbg !44 + %1133 = insertelement <2 x bfloat> poison, bfloat %1072, i64 0, !dbg !44 + %1134 = insertelement <2 x bfloat> %1133, bfloat %1073, i64 1, !dbg !44 + %1135 = bitcast <2 x bfloat> %1134 to i32, !dbg !44 + %1136 = insertelement <2 x bfloat> poison, bfloat %1074, i64 0, !dbg !44 + %1137 = insertelement <2 x bfloat> %1136, bfloat %1075, i64 1, !dbg !44 + %1138 = bitcast <2 x bfloat> %1137 to i32, !dbg !44 + %1139 = insertelement <2 x bfloat> poison, bfloat %1076, i64 0, !dbg !44 + %1140 = insertelement <2 x bfloat> %1139, bfloat %1077, i64 1, !dbg !44 + %1141 = bitcast <2 x bfloat> %1140 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1132, i32 %1135, i32 %1138, i32 %1141, ptr addrspace(1) %103, i1 %128) #3, !dbg !44 + %1142 = insertelement <2 x bfloat> poison, bfloat %1078, i64 0, !dbg !44 + %1143 = insertelement <2 x bfloat> %1142, bfloat %1079, i64 1, !dbg !44 + %1144 = bitcast <2 x bfloat> %1143 to i32, !dbg !44 + %1145 = insertelement <2 x bfloat> poison, bfloat %1080, i64 0, !dbg !44 + %1146 = insertelement <2 x bfloat> %1145, bfloat %1081, i64 1, !dbg !44 + %1147 = bitcast <2 x bfloat> %1146 to i32, !dbg !44 + %1148 = insertelement <2 x bfloat> poison, bfloat %1082, i64 0, !dbg !44 + %1149 = insertelement <2 x bfloat> %1148, bfloat %1083, i64 1, !dbg !44 + %1150 = bitcast <2 x bfloat> %1149 to i32, !dbg !44 + %1151 = insertelement <2 x bfloat> poison, bfloat %1084, i64 0, !dbg !44 + %1152 = insertelement <2 x bfloat> %1151, bfloat %1085, i64 1, !dbg !44 + %1153 = bitcast <2 x bfloat> %1152 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1144, i32 %1147, i32 %1150, i32 %1153, ptr addrspace(1) %104, i1 %146) #3, !dbg !44 + %1154 = insertelement <2 x bfloat> poison, bfloat %1086, i64 0, !dbg !44 + %1155 = insertelement <2 x bfloat> %1154, bfloat %1087, i64 1, !dbg !44 + %1156 = bitcast <2 x bfloat> %1155 to i32, !dbg !44 + %1157 = insertelement <2 x bfloat> poison, bfloat %1088, i64 0, !dbg !44 + %1158 = insertelement <2 x bfloat> %1157, bfloat %1089, i64 1, !dbg !44 + %1159 = bitcast <2 x bfloat> %1158 to i32, !dbg !44 + %1160 = insertelement <2 x bfloat> poison, bfloat %1090, i64 0, !dbg !44 + %1161 = insertelement <2 x bfloat> %1160, bfloat %1091, i64 1, !dbg !44 + %1162 = bitcast <2 x bfloat> %1161 to i32, !dbg !44 + %1163 = insertelement <2 x bfloat> poison, bfloat %1092, i64 0, !dbg !44 + %1164 = insertelement <2 x bfloat> %1163, bfloat %1093, i64 1, !dbg !44 + %1165 = bitcast <2 x bfloat> %1164 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1156, i32 %1159, i32 %1162, i32 %1165, ptr addrspace(1) %105, i1 %164) #3, !dbg !44 + %1166 = insertelement <2 x bfloat> poison, bfloat %1094, i64 0, !dbg !44 + %1167 = insertelement <2 x bfloat> %1166, bfloat %1095, i64 1, !dbg !44 + %1168 = bitcast <2 x bfloat> %1167 to i32, !dbg !44 + %1169 = insertelement <2 x bfloat> poison, bfloat %1096, i64 0, !dbg !44 + %1170 = insertelement <2 x bfloat> %1169, bfloat %1097, i64 1, !dbg !44 + %1171 = bitcast <2 x bfloat> %1170 to i32, !dbg !44 + %1172 = insertelement <2 x bfloat> poison, bfloat %1098, i64 0, !dbg !44 + %1173 = insertelement <2 x bfloat> %1172, bfloat %1099, i64 1, !dbg !44 + %1174 = bitcast <2 x bfloat> %1173 to i32, !dbg !44 + %1175 = insertelement <2 x bfloat> poison, bfloat %1100, i64 0, !dbg !44 + %1176 = insertelement <2 x bfloat> %1175, bfloat %1101, i64 1, !dbg !44 + %1177 = bitcast <2 x bfloat> %1176 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1168, i32 %1171, i32 %1174, i32 %1177, ptr addrspace(1) %106, i1 %182) #3, !dbg !44 + %1178 = insertelement <2 x bfloat> poison, bfloat %1102, i64 0, !dbg !44 + %1179 = insertelement <2 x bfloat> %1178, bfloat %1103, i64 1, !dbg !44 + %1180 = bitcast <2 x bfloat> %1179 to i32, !dbg !44 + %1181 = insertelement <2 x bfloat> poison, bfloat %1104, i64 0, !dbg !44 + %1182 = insertelement <2 x bfloat> %1181, bfloat %1105, i64 1, !dbg !44 + %1183 = bitcast <2 x bfloat> %1182 to i32, !dbg !44 + %1184 = insertelement <2 x bfloat> poison, bfloat %1106, i64 0, !dbg !44 + %1185 = insertelement <2 x bfloat> %1184, bfloat %1107, i64 1, !dbg !44 + %1186 = bitcast <2 x bfloat> %1185 to i32, !dbg !44 + %1187 = insertelement <2 x bfloat> poison, bfloat %1108, i64 0, !dbg !44 + %1188 = insertelement <2 x bfloat> %1187, bfloat %1109, i64 1, !dbg !44 + %1189 = bitcast <2 x bfloat> %1188 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1180, i32 %1183, i32 %1186, i32 %1189, ptr addrspace(1) %107, i1 %200) #3, !dbg !44 + %1190 = bitcast <2 x bfloat> %1110 to i32, !dbg !44 + %1191 = bitcast <2 x bfloat> %1111 to i32, !dbg !44 + %1192 = bitcast <2 x bfloat> %1112 to i32, !dbg !44 + %1193 = bitcast <2 x bfloat> %1113 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1190, i32 %1191, i32 %1192, i32 %1193, ptr addrspace(1) %108, i1 %218) #3, !dbg !44 + %1194 = bitcast <2 x bfloat> %1114 to i32, !dbg !44 + %1195 = bitcast <2 x bfloat> %1115 to i32, !dbg !44 + %1196 = bitcast <2 x bfloat> %1116 to i32, !dbg !44 + %1197 = bitcast <2 x bfloat> %1117 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1194, i32 %1195, i32 %1196, i32 %1197, ptr addrspace(1) %109, i1 %236) #3, !dbg !44 + ret void, !dbg !45 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +attributes #0 = { nounwind "nvvm.reqntid"="256" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = distinct !DISubprogram(name: "l2norm_bwd_kernel", linkageName: "l2norm_bwd_kernel", scope: !1, file: !1, line: 93, type: !5, scopeLine: 93, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!5 = !DISubroutineType(cc: DW_CC_normal, types: !6) +!6 = !{null, !7, !9, !7, !7, !10, !11, !7, !7} +!7 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !8, size: 64, dwarfAddressSpace: 1) +!8 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!9 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !10, size: 64, dwarfAddressSpace: 1) +!10 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!11 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!12 = !DILocation(line: 94, column: 24, scope: !4) +!13 = !DILocation(line: 94, column: 30, scope: !4) +!14 = !DILocation(line: 95, column: 16, scope: !4) +!15 = !DILocation(line: 95, column: 34, scope: !4) +!16 = !DILocation(line: 95, column: 21, scope: !4) +!17 = !DILocation(line: 97, column: 16, scope: !4) +!18 = !DILocation(line: 98, column: 30, scope: !4) +!19 = !DILocation(line: 99, column: 29, scope: !4) +!20 = !DILocation(line: 99, column: 14, scope: !4) +!21 = !DILocation(line: 99, column: 33, scope: !4) +!22 = !DILocation(line: 100, column: 20, scope: !4) +!23 = !DILocation(line: 101, column: 16, scope: !4) +!24 = !DILocation(line: 101, column: 35, scope: !4) +!25 = !DILocation(line: 102, column: 16, scope: !4) +!26 = !DILocation(line: 102, column: 35, scope: !4) +!27 = !DILocation(line: 104, column: 18, scope: !4) +!28 = !DILocation(line: 104, column: 47, scope: !4) +!29 = !DILocation(line: 105, column: 21, scope: !4) +!30 = !DILocation(line: 106, column: 19, scope: !4) +!31 = !DILocation(line: 106, column: 49, scope: !4) +!32 = !DILocation(line: 107, column: 18, scope: !4) +!33 = !DILocation(line: 107, column: 50, scope: !4) +!34 = !DILocation(line: 263, column: 15, scope: !35, inlinedAt: !37) +!35 = distinct !DILexicalBlockFile(scope: !4, file: !36, discriminator: 0) +!36 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!37 = !DILocation(line: 293, column: 36, scope: !35, inlinedAt: !38) +!38 = !DILocation(line: 107, column: 55, scope: !39) +!39 = distinct !DILexicalBlockFile(scope: !4, file: !1, discriminator: 0) +!40 = !DILocation(line: 107, column: 69, scope: !4) +!41 = !DILocation(line: 107, column: 75, scope: !4) +!42 = !DILocation(line: 107, column: 36, scope: !4) +!43 = !DILocation(line: 108, column: 27, scope: !4) +!44 = !DILocation(line: 108, column: 19, scope: !4) +!45 = !DILocation(line: 108, column: 4, scope: !4) diff --git a/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.ptx b/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..4c3848dc801d6b9b5bb800ef00d70ae1b18234b4 --- /dev/null +++ b/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.ptx @@ -0,0 +1,1322 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_bwd_kernel // -- Begin function l2norm_bwd_kernel + // @l2norm_bwd_kernel +.visible .entry l2norm_bwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_2, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_3, + .param .f32 l2norm_bwd_kernel_param_4, + .param .u32 l2norm_bwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_6, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_7 +) +.reqntid 256 +{ + .reg .pred %p<9>; + .reg .b16 %rs<129>; + .reg .b32 %r<631>; + .reg .b64 %rd<120>; + .loc 1 93 0 // l2norm.py:93:0 +$L__func_begin0: + .loc 1 93 0 // l2norm.py:93:0 + +// %bb.0: + ld.param.b64 %rd33, [l2norm_bwd_kernel_param_0]; + ld.param.b64 %rd34, [l2norm_bwd_kernel_param_1]; +$L__tmp0: + .loc 1 94 24 // l2norm.py:94:24 + mov.u32 %r106, %ctaid.x; + ld.param.b64 %rd35, [l2norm_bwd_kernel_param_2]; + .loc 1 95 16 // l2norm.py:95:16 + mul.wide.u32 %rd36, %r106, 128; + ld.param.b64 %rd37, [l2norm_bwd_kernel_param_3]; + .loc 1 95 34 // l2norm.py:95:34 + mov.u32 %r107, %tid.x; + shr.u32 %r108, %r107, 4; + ld.param.s32 %rd38, [l2norm_bwd_kernel_param_5]; + bfe.u32 %r109, %r107, 4, 4; + or.b32 %r110, %r109, 16; + or.b32 %r111, %r109, 32; + or.b32 %r112, %r108, 48; + .loc 1 95 21 // l2norm.py:95:21 + or.b32 %r113, %r109, 64; + or.b32 %r114, %r109, 80; + or.b32 %r115, %r109, 96; + or.b32 %r116, %r108, 112; + cvt.u64.u32 %rd39, %r116; + cvt.u64.u32 %rd40, %r115; + cvt.u64.u32 %rd41, %r114; + cvt.u64.u32 %rd42, %r113; + cvt.u64.u32 %rd43, %r112; + cvt.u64.u32 %rd44, %r111; + cvt.u64.u32 %rd45, %r110; + cvt.u64.u32 %rd46, %r109; + or.b64 %rd47, %rd36, %rd46; + or.b64 %rd48, %rd36, %rd45; + or.b64 %rd49, %rd36, %rd44; + or.b64 %rd50, %rd36, %rd43; + or.b64 %rd51, %rd36, %rd42; + or.b64 %rd52, %rd36, %rd41; + or.b64 %rd53, %rd36, %rd40; + or.b64 %rd54, %rd36, %rd39; + .loc 1 97 16 // l2norm.py:97:16 + setp.lt.s64 %p8, %rd54, %rd38; + setp.lt.s64 %p7, %rd53, %rd38; + setp.lt.s64 %p6, %rd52, %rd38; + setp.lt.s64 %p5, %rd51, %rd38; + setp.lt.s64 %p4, %rd50, %rd38; + setp.lt.s64 %p3, %rd49, %rd38; + setp.lt.s64 %p2, %rd48, %rd38; + setp.lt.s64 %p1, %rd47, %rd38; + .loc 1 98 30 // l2norm.py:98:30 + shl.b32 %r117, %r107, 3; + and.b32 %r118, %r117, 120; + .loc 1 99 14 // l2norm.py:99:14 + shl.b64 %rd55, %rd47, 8; + add.s64 %rd56, %rd33, %rd55; + shl.b64 %rd57, %rd48, 8; + add.s64 %rd58, %rd33, %rd57; + shl.b64 %rd59, %rd49, 8; + add.s64 %rd60, %rd33, %rd59; + shl.b64 %rd61, %rd50, 8; + add.s64 %rd62, %rd33, %rd61; + shl.b64 %rd63, %rd51, 8; + add.s64 %rd64, %rd33, %rd63; + shl.b64 %rd65, %rd52, 8; + add.s64 %rd66, %rd33, %rd65; + shl.b64 %rd67, %rd53, 8; + add.s64 %rd68, %rd33, %rd67; + shl.b64 %rd69, %rd54, 8; + add.s64 %rd70, %rd33, %rd69; + .loc 1 99 33 // l2norm.py:99:33 + mul.wide.u32 %rd71, %r118, 2; + add.s64 %rd1, %rd56, %rd71; + add.s64 %rd2, %rd58, %rd71; + add.s64 %rd3, %rd60, %rd71; + add.s64 %rd4, %rd62, %rd71; + add.s64 %rd5, %rd64, %rd71; + add.s64 %rd6, %rd66, %rd71; + add.s64 %rd7, %rd68, %rd71; + add.s64 %rd8, %rd70, %rd71; + .loc 1 100 20 // l2norm.py:100:20 + shl.b64 %rd72, %rd47, 2; + add.s64 %rd9, %rd34, %rd72; + shl.b64 %rd73, %rd48, 2; + add.s64 %rd10, %rd34, %rd73; + shl.b64 %rd74, %rd49, 2; + add.s64 %rd11, %rd34, %rd74; + shl.b64 %rd75, %rd50, 2; + add.s64 %rd12, %rd34, %rd75; + shl.b64 %rd76, %rd51, 2; + add.s64 %rd13, %rd34, %rd76; + shl.b64 %rd77, %rd52, 2; + add.s64 %rd14, %rd34, %rd77; + shl.b64 %rd78, %rd53, 2; + add.s64 %rd15, %rd34, %rd78; + shl.b64 %rd79, %rd54, 2; + add.s64 %rd16, %rd34, %rd79; + .loc 1 101 16 // l2norm.py:101:16 + add.s64 %rd80, %rd35, %rd55; + add.s64 %rd81, %rd35, %rd57; + add.s64 %rd82, %rd35, %rd59; + add.s64 %rd83, %rd35, %rd61; + add.s64 %rd84, %rd35, %rd63; + add.s64 %rd85, %rd35, %rd65; + add.s64 %rd86, %rd35, %rd67; + add.s64 %rd87, %rd35, %rd69; + .loc 1 101 35 // l2norm.py:101:35 + add.s64 %rd17, %rd80, %rd71; + add.s64 %rd18, %rd81, %rd71; + add.s64 %rd19, %rd82, %rd71; + add.s64 %rd20, %rd83, %rd71; + add.s64 %rd21, %rd84, %rd71; + add.s64 %rd22, %rd85, %rd71; + add.s64 %rd23, %rd86, %rd71; + add.s64 %rd24, %rd87, %rd71; + .loc 1 102 16 // l2norm.py:102:16 + add.s64 %rd88, %rd37, %rd55; + add.s64 %rd89, %rd37, %rd57; + add.s64 %rd90, %rd37, %rd59; + add.s64 %rd91, %rd37, %rd61; + add.s64 %rd92, %rd37, %rd63; + add.s64 %rd93, %rd37, %rd65; + add.s64 %rd94, %rd37, %rd67; + add.s64 %rd95, %rd37, %rd69; + .loc 1 102 35 // l2norm.py:102:35 + add.s64 %rd25, %rd88, %rd71; + add.s64 %rd26, %rd89, %rd71; + add.s64 %rd27, %rd90, %rd71; + add.s64 %rd28, %rd91, %rd71; + add.s64 %rd29, %rd92, %rd71; + add.s64 %rd30, %rd93, %rd71; + add.s64 %rd31, %rd94, %rd71; + add.s64 %rd32, %rd95, %rd71; + mov.b32 %r5, 0; + .loc 1 104 18 // l2norm.py:104:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + mov.b32 {%rs5, %rs6}, %r3; + mov.b32 {%rs7, %rs8}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs9, %rs10}, %r6; + mov.b32 {%rs11, %rs12}, %r7; + mov.b32 {%rs13, %rs14}, %r8; + mov.b32 {%rs15, %rs16}, %r9; + // begin inline asm + mov.u32 %r10, %r5; + mov.u32 %r11, %r5; + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + @%p3 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs17, %rs18}, %r10; + mov.b32 {%rs19, %rs20}, %r11; + mov.b32 {%rs21, %rs22}, %r12; + mov.b32 {%rs23, %rs24}, %r13; + // begin inline asm + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + @%p4 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ]; + // end inline asm + mov.b32 {%rs25, %rs26}, %r14; + mov.b32 {%rs27, %rs28}, %r15; + mov.b32 {%rs29, %rs30}, %r16; + mov.b32 {%rs31, %rs32}, %r17; + // begin inline asm + mov.u32 %r18, %r5; + mov.u32 %r19, %r5; + mov.u32 %r20, %r5; + mov.u32 %r21, %r5; + @%p5 ld.global.v4.b32 { %r18, %r19, %r20, %r21 }, [ %rd5 + 0 ]; + // end inline asm + mov.b32 {%rs33, %rs34}, %r18; + mov.b32 {%rs35, %rs36}, %r19; + mov.b32 {%rs37, %rs38}, %r20; + mov.b32 {%rs39, %rs40}, %r21; + // begin inline asm + mov.u32 %r22, %r5; + mov.u32 %r23, %r5; + mov.u32 %r24, %r5; + mov.u32 %r25, %r5; + @%p6 ld.global.v4.b32 { %r22, %r23, %r24, %r25 }, [ %rd6 + 0 ]; + // end inline asm + mov.b32 {%rs41, %rs42}, %r22; + mov.b32 {%rs43, %rs44}, %r23; + mov.b32 {%rs45, %rs46}, %r24; + mov.b32 {%rs47, %rs48}, %r25; + // begin inline asm + mov.u32 %r26, %r5; + mov.u32 %r27, %r5; + mov.u32 %r28, %r5; + mov.u32 %r29, %r5; + @%p7 ld.global.v4.b32 { %r26, %r27, %r28, %r29 }, [ %rd7 + 0 ]; + // end inline asm + mov.b32 {%rs49, %rs50}, %r26; + mov.b32 {%rs51, %rs52}, %r27; + mov.b32 {%rs53, %rs54}, %r28; + mov.b32 {%rs55, %rs56}, %r29; + // begin inline asm + mov.u32 %r30, %r5; + mov.u32 %r31, %r5; + mov.u32 %r32, %r5; + mov.u32 %r33, %r5; + @%p8 ld.global.v4.b32 { %r30, %r31, %r32, %r33 }, [ %rd8 + 0 ]; + // end inline asm + mov.b32 {%rs57, %rs58}, %r30; + mov.b32 {%rs59, %rs60}, %r31; + mov.b32 {%rs61, %rs62}, %r32; + mov.b32 {%rs63, %rs64}, %r33; + .loc 1 104 47 // l2norm.py:104:47 + cvt.f32.bf16 %r119, %rs1; + cvt.f32.bf16 %r120, %rs2; + cvt.f32.bf16 %r121, %rs3; + cvt.f32.bf16 %r122, %rs4; + cvt.f32.bf16 %r123, %rs5; + cvt.f32.bf16 %r124, %rs6; + cvt.f32.bf16 %r125, %rs7; + cvt.f32.bf16 %r126, %rs8; + cvt.f32.bf16 %r127, %rs9; + cvt.f32.bf16 %r128, %rs10; + cvt.f32.bf16 %r129, %rs11; + cvt.f32.bf16 %r130, %rs12; + cvt.f32.bf16 %r131, %rs13; + cvt.f32.bf16 %r132, %rs14; + cvt.f32.bf16 %r133, %rs15; + cvt.f32.bf16 %r134, %rs16; + cvt.f32.bf16 %r135, %rs17; + cvt.f32.bf16 %r136, %rs18; + cvt.f32.bf16 %r137, %rs19; + cvt.f32.bf16 %r138, %rs20; + cvt.f32.bf16 %r139, %rs21; + cvt.f32.bf16 %r140, %rs22; + cvt.f32.bf16 %r141, %rs23; + cvt.f32.bf16 %r142, %rs24; + cvt.f32.bf16 %r143, %rs25; + cvt.f32.bf16 %r144, %rs26; + cvt.f32.bf16 %r145, %rs27; + cvt.f32.bf16 %r146, %rs28; + cvt.f32.bf16 %r147, %rs29; + cvt.f32.bf16 %r148, %rs30; + cvt.f32.bf16 %r149, %rs31; + cvt.f32.bf16 %r150, %rs32; + cvt.f32.bf16 %r151, %rs33; + cvt.f32.bf16 %r152, %rs34; + cvt.f32.bf16 %r153, %rs35; + cvt.f32.bf16 %r154, %rs36; + cvt.f32.bf16 %r155, %rs37; + cvt.f32.bf16 %r156, %rs38; + cvt.f32.bf16 %r157, %rs39; + cvt.f32.bf16 %r158, %rs40; + cvt.f32.bf16 %r159, %rs41; + cvt.f32.bf16 %r160, %rs42; + cvt.f32.bf16 %r161, %rs43; + cvt.f32.bf16 %r162, %rs44; + cvt.f32.bf16 %r163, %rs45; + cvt.f32.bf16 %r164, %rs46; + cvt.f32.bf16 %r165, %rs47; + cvt.f32.bf16 %r166, %rs48; + cvt.f32.bf16 %r167, %rs49; + cvt.f32.bf16 %r168, %rs50; + cvt.f32.bf16 %r169, %rs51; + cvt.f32.bf16 %r170, %rs52; + cvt.f32.bf16 %r171, %rs53; + cvt.f32.bf16 %r172, %rs54; + cvt.f32.bf16 %r173, %rs55; + cvt.f32.bf16 %r174, %rs56; + cvt.f32.bf16 %r175, %rs57; + cvt.f32.bf16 %r176, %rs58; + cvt.f32.bf16 %r177, %rs59; + cvt.f32.bf16 %r178, %rs60; + cvt.f32.bf16 %r179, %rs61; + cvt.f32.bf16 %r180, %rs62; + cvt.f32.bf16 %r181, %rs63; + cvt.f32.bf16 %r182, %rs64; + .loc 1 105 21 // l2norm.py:105:21 + // begin inline asm + mov.u32 %r34, %r5; + @%p1 ld.global.b32 { %r34 }, [ %rd9 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r35, %r5; + @%p2 ld.global.b32 { %r35 }, [ %rd10 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r36, %r5; + @%p3 ld.global.b32 { %r36 }, [ %rd11 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r37, %r5; + @%p4 ld.global.b32 { %r37 }, [ %rd12 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r38, %r5; + @%p5 ld.global.b32 { %r38 }, [ %rd13 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r39, %r5; + @%p6 ld.global.b32 { %r39 }, [ %rd14 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r40, %r5; + @%p7 ld.global.b32 { %r40 }, [ %rd15 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r41, %r5; + @%p8 ld.global.b32 { %r41 }, [ %rd16 + 0 ]; + // end inline asm + .loc 1 106 19 // l2norm.py:106:19 + // begin inline asm + mov.u32 %r42, %r5; + mov.u32 %r43, %r5; + mov.u32 %r44, %r5; + mov.u32 %r45, %r5; + @%p1 ld.global.v4.b32 { %r42, %r43, %r44, %r45 }, [ %rd17 + 0 ]; + // end inline asm + mov.b32 {%rs65, %rs66}, %r42; + mov.b32 {%rs67, %rs68}, %r43; + mov.b32 {%rs69, %rs70}, %r44; + mov.b32 {%rs71, %rs72}, %r45; + // begin inline asm + mov.u32 %r46, %r5; + mov.u32 %r47, %r5; + mov.u32 %r48, %r5; + mov.u32 %r49, %r5; + @%p2 ld.global.v4.b32 { %r46, %r47, %r48, %r49 }, [ %rd18 + 0 ]; + // end inline asm + mov.b32 {%rs73, %rs74}, %r46; + mov.b32 {%rs75, %rs76}, %r47; + mov.b32 {%rs77, %rs78}, %r48; + mov.b32 {%rs79, %rs80}, %r49; + // begin inline asm + mov.u32 %r50, %r5; + mov.u32 %r51, %r5; + mov.u32 %r52, %r5; + mov.u32 %r53, %r5; + @%p3 ld.global.v4.b32 { %r50, %r51, %r52, %r53 }, [ %rd19 + 0 ]; + // end inline asm + mov.b32 {%rs81, %rs82}, %r50; + mov.b32 {%rs83, %rs84}, %r51; + mov.b32 {%rs85, %rs86}, %r52; + mov.b32 {%rs87, %rs88}, %r53; + // begin inline asm + mov.u32 %r54, %r5; + mov.u32 %r55, %r5; + mov.u32 %r56, %r5; + mov.u32 %r57, %r5; + @%p4 ld.global.v4.b32 { %r54, %r55, %r56, %r57 }, [ %rd20 + 0 ]; + // end inline asm + mov.b32 {%rs89, %rs90}, %r54; + mov.b32 {%rs91, %rs92}, %r55; + mov.b32 {%rs93, %rs94}, %r56; + mov.b32 {%rs95, %rs96}, %r57; + // begin inline asm + mov.u32 %r58, %r5; + mov.u32 %r59, %r5; + mov.u32 %r60, %r5; + mov.u32 %r61, %r5; + @%p5 ld.global.v4.b32 { %r58, %r59, %r60, %r61 }, [ %rd21 + 0 ]; + // end inline asm + mov.b32 {%rs97, %rs98}, %r58; + mov.b32 {%rs99, %rs100}, %r59; + mov.b32 {%rs101, %rs102}, %r60; + mov.b32 {%rs103, %rs104}, %r61; + // begin inline asm + mov.u32 %r62, %r5; + mov.u32 %r63, %r5; + mov.u32 %r64, %r5; + mov.u32 %r65, %r5; + @%p6 ld.global.v4.b32 { %r62, %r63, %r64, %r65 }, [ %rd22 + 0 ]; + // end inline asm + mov.b32 {%rs105, %rs106}, %r62; + mov.b32 {%rs107, %rs108}, %r63; + mov.b32 {%rs109, %rs110}, %r64; + mov.b32 {%rs111, %rs112}, %r65; + // begin inline asm + mov.u32 %r66, %r5; + mov.u32 %r67, %r5; + mov.u32 %r68, %r5; + mov.u32 %r69, %r5; + @%p7 ld.global.v4.b32 { %r66, %r67, %r68, %r69 }, [ %rd23 + 0 ]; + // end inline asm + mov.b32 {%rs113, %rs114}, %r66; + mov.b32 {%rs115, %rs116}, %r67; + mov.b32 {%rs117, %rs118}, %r68; + mov.b32 {%rs119, %rs120}, %r69; + // begin inline asm + mov.u32 %r70, %r5; + mov.u32 %r71, %r5; + mov.u32 %r72, %r5; + mov.u32 %r73, %r5; + @%p8 ld.global.v4.b32 { %r70, %r71, %r72, %r73 }, [ %rd24 + 0 ]; + // end inline asm + mov.b32 {%rs121, %rs122}, %r70; + mov.b32 {%rs123, %rs124}, %r71; + mov.b32 {%rs125, %rs126}, %r72; + mov.b32 {%rs127, %rs128}, %r73; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r183, %rs65; + cvt.f32.bf16 %r184, %rs66; + cvt.f32.bf16 %r185, %rs67; + cvt.f32.bf16 %r186, %rs68; + cvt.f32.bf16 %r187, %rs69; + cvt.f32.bf16 %r188, %rs70; + cvt.f32.bf16 %r189, %rs71; + cvt.f32.bf16 %r190, %rs72; + cvt.f32.bf16 %r191, %rs73; + cvt.f32.bf16 %r192, %rs74; + cvt.f32.bf16 %r193, %rs75; + cvt.f32.bf16 %r194, %rs76; + cvt.f32.bf16 %r195, %rs77; + cvt.f32.bf16 %r196, %rs78; + cvt.f32.bf16 %r197, %rs79; + cvt.f32.bf16 %r198, %rs80; + cvt.f32.bf16 %r199, %rs81; + cvt.f32.bf16 %r200, %rs82; + cvt.f32.bf16 %r201, %rs83; + cvt.f32.bf16 %r202, %rs84; + cvt.f32.bf16 %r203, %rs85; + cvt.f32.bf16 %r204, %rs86; + cvt.f32.bf16 %r205, %rs87; + cvt.f32.bf16 %r206, %rs88; + cvt.f32.bf16 %r207, %rs89; + cvt.f32.bf16 %r208, %rs90; + cvt.f32.bf16 %r209, %rs91; + cvt.f32.bf16 %r210, %rs92; + cvt.f32.bf16 %r211, %rs93; + cvt.f32.bf16 %r212, %rs94; + cvt.f32.bf16 %r213, %rs95; + cvt.f32.bf16 %r214, %rs96; + cvt.f32.bf16 %r215, %rs97; + cvt.f32.bf16 %r216, %rs98; + cvt.f32.bf16 %r217, %rs99; + cvt.f32.bf16 %r218, %rs100; + cvt.f32.bf16 %r219, %rs101; + cvt.f32.bf16 %r220, %rs102; + cvt.f32.bf16 %r221, %rs103; + cvt.f32.bf16 %r222, %rs104; + cvt.f32.bf16 %r223, %rs105; + cvt.f32.bf16 %r224, %rs106; + cvt.f32.bf16 %r225, %rs107; + cvt.f32.bf16 %r226, %rs108; + cvt.f32.bf16 %r227, %rs109; + cvt.f32.bf16 %r228, %rs110; + cvt.f32.bf16 %r229, %rs111; + cvt.f32.bf16 %r230, %rs112; + cvt.f32.bf16 %r231, %rs113; + cvt.f32.bf16 %r232, %rs114; + cvt.f32.bf16 %r233, %rs115; + cvt.f32.bf16 %r234, %rs116; + cvt.f32.bf16 %r235, %rs117; + cvt.f32.bf16 %r236, %rs118; + cvt.f32.bf16 %r237, %rs119; + cvt.f32.bf16 %r238, %rs120; + cvt.f32.bf16 %r239, %rs121; + cvt.f32.bf16 %r240, %rs122; + cvt.f32.bf16 %r241, %rs123; + cvt.f32.bf16 %r242, %rs124; + cvt.f32.bf16 %r243, %rs125; + cvt.f32.bf16 %r244, %rs126; + cvt.f32.bf16 %r245, %rs127; + cvt.f32.bf16 %r246, %rs128; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r247, %r40, %r231; + mul.f32 %r248, %r40, %r232; + mul.f32 %r249, %r40, %r233; + mul.f32 %r250, %r40, %r234; + mul.f32 %r251, %r40, %r235; + mul.f32 %r252, %r40, %r236; + mul.f32 %r253, %r40, %r237; + mul.f32 %r254, %r40, %r238; + mul.f32 %r255, %r41, %r239; + mul.f32 %r256, %r41, %r240; + mul.f32 %r257, %r41, %r241; + mul.f32 %r258, %r41, %r242; + mul.f32 %r259, %r41, %r243; + mul.f32 %r260, %r41, %r244; + mul.f32 %r261, %r41, %r245; + mul.f32 %r262, %r41, %r246; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r263, %r120, %r184; + mul.f32 %r264, %r128, %r192; + mul.f32 %r265, %r136, %r200; + mul.f32 %r266, %r144, %r208; + mul.f32 %r267, %r152, %r216; + mul.f32 %r268, %r160, %r224; + mul.f32 %r269, %r168, %r232; + mul.f32 %r270, %r176, %r240; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r271, %rs1, %rs65, %r263; + fma.rn.f32.bf16 %r272, %rs3, %rs67, %r271; + fma.rn.f32.bf16 %r273, %rs4, %rs68, %r272; + fma.rn.f32.bf16 %r274, %rs5, %rs69, %r273; + fma.rn.f32.bf16 %r275, %rs6, %rs70, %r274; + fma.rn.f32.bf16 %r276, %rs7, %rs71, %r275; + fma.rn.f32.bf16 %r277, %rs8, %rs72, %r276; + fma.rn.f32.bf16 %r278, %rs9, %rs73, %r264; + fma.rn.f32.bf16 %r279, %rs11, %rs75, %r278; + fma.rn.f32.bf16 %r280, %rs12, %rs76, %r279; + fma.rn.f32.bf16 %r281, %rs13, %rs77, %r280; + fma.rn.f32.bf16 %r282, %rs14, %rs78, %r281; + fma.rn.f32.bf16 %r283, %rs15, %rs79, %r282; + fma.rn.f32.bf16 %r284, %rs16, %rs80, %r283; + fma.rn.f32.bf16 %r285, %rs17, %rs81, %r265; + fma.rn.f32.bf16 %r286, %rs19, %rs83, %r285; + fma.rn.f32.bf16 %r287, %rs20, %rs84, %r286; + fma.rn.f32.bf16 %r288, %rs21, %rs85, %r287; + fma.rn.f32.bf16 %r289, %rs22, %rs86, %r288; + fma.rn.f32.bf16 %r290, %rs23, %rs87, %r289; + fma.rn.f32.bf16 %r291, %rs24, %rs88, %r290; + fma.rn.f32.bf16 %r292, %rs25, %rs89, %r266; + fma.rn.f32.bf16 %r293, %rs27, %rs91, %r292; + fma.rn.f32.bf16 %r294, %rs28, %rs92, %r293; + fma.rn.f32.bf16 %r295, %rs29, %rs93, %r294; + fma.rn.f32.bf16 %r296, %rs30, %rs94, %r295; + fma.rn.f32.bf16 %r297, %rs31, %rs95, %r296; + fma.rn.f32.bf16 %r298, %rs32, %rs96, %r297; + fma.rn.f32.bf16 %r299, %rs33, %rs97, %r267; + fma.rn.f32.bf16 %r300, %rs35, %rs99, %r299; + fma.rn.f32.bf16 %r301, %rs36, %rs100, %r300; + fma.rn.f32.bf16 %r302, %rs37, %rs101, %r301; + fma.rn.f32.bf16 %r303, %rs38, %rs102, %r302; + fma.rn.f32.bf16 %r304, %rs39, %rs103, %r303; + fma.rn.f32.bf16 %r305, %rs40, %rs104, %r304; + fma.rn.f32.bf16 %r306, %rs41, %rs105, %r268; + fma.rn.f32.bf16 %r307, %rs43, %rs107, %r306; + fma.rn.f32.bf16 %r308, %rs44, %rs108, %r307; + fma.rn.f32.bf16 %r309, %rs45, %rs109, %r308; + fma.rn.f32.bf16 %r310, %rs46, %rs110, %r309; + fma.rn.f32.bf16 %r311, %rs47, %rs111, %r310; + fma.rn.f32.bf16 %r312, %rs48, %rs112, %r311; + fma.rn.f32.bf16 %r313, %rs49, %rs113, %r269; + fma.rn.f32.bf16 %r314, %rs51, %rs115, %r313; + fma.rn.f32.bf16 %r315, %rs52, %rs116, %r314; + fma.rn.f32.bf16 %r316, %rs53, %rs117, %r315; + fma.rn.f32.bf16 %r317, %rs54, %rs118, %r316; + fma.rn.f32.bf16 %r318, %rs55, %rs119, %r317; + fma.rn.f32.bf16 %r319, %rs56, %rs120, %r318; + fma.rn.f32.bf16 %r320, %rs57, %rs121, %r270; + fma.rn.f32.bf16 %r321, %rs59, %rs123, %r320; + fma.rn.f32.bf16 %r322, %rs60, %rs124, %r321; + fma.rn.f32.bf16 %r323, %rs61, %rs125, %r322; + fma.rn.f32.bf16 %r324, %rs62, %rs126, %r323; + fma.rn.f32.bf16 %r325, %rs63, %rs127, %r324; + fma.rn.f32.bf16 %r326, %rs64, %rs128, %r325; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r327, %r277, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r328, %r277, %r327; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r329, %r328, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r330, %r328, %r329; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r331, %r330, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r332, %r330, %r331; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r333, %r332, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r334, %r332, %r333; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r335, %r284, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r336, %r284, %r335; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r337, %r336, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r338, %r336, %r337; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r339, %r338, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r340, %r338, %r339; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r341, %r340, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r342, %r340, %r341; +$L__tmp18: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r343, %r291, 8, 31, -1; +$L__tmp19: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r344, %r291, %r343; +$L__tmp20: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r345, %r344, 4, 31, -1; +$L__tmp21: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r346, %r344, %r345; +$L__tmp22: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r347, %r346, 2, 31, -1; +$L__tmp23: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r348, %r346, %r347; +$L__tmp24: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r349, %r348, 1, 31, -1; +$L__tmp25: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r350, %r348, %r349; +$L__tmp26: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r351, %r298, 8, 31, -1; +$L__tmp27: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r352, %r298, %r351; +$L__tmp28: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r353, %r352, 4, 31, -1; +$L__tmp29: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r354, %r352, %r353; +$L__tmp30: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r355, %r354, 2, 31, -1; +$L__tmp31: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r356, %r354, %r355; +$L__tmp32: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r357, %r356, 1, 31, -1; +$L__tmp33: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r358, %r356, %r357; +$L__tmp34: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r359, %r305, 8, 31, -1; +$L__tmp35: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r360, %r305, %r359; +$L__tmp36: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r361, %r360, 4, 31, -1; +$L__tmp37: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r362, %r360, %r361; +$L__tmp38: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r363, %r362, 2, 31, -1; +$L__tmp39: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r364, %r362, %r363; +$L__tmp40: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r365, %r364, 1, 31, -1; +$L__tmp41: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r366, %r364, %r365; +$L__tmp42: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r367, %r312, 8, 31, -1; +$L__tmp43: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r368, %r312, %r367; +$L__tmp44: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r369, %r368, 4, 31, -1; +$L__tmp45: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r370, %r368, %r369; +$L__tmp46: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r371, %r370, 2, 31, -1; +$L__tmp47: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r372, %r370, %r371; +$L__tmp48: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r373, %r372, 1, 31, -1; +$L__tmp49: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r374, %r372, %r373; +$L__tmp50: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r375, %r319, 8, 31, -1; +$L__tmp51: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r376, %r319, %r375; +$L__tmp52: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r377, %r376, 4, 31, -1; +$L__tmp53: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r378, %r376, %r377; +$L__tmp54: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r379, %r378, 2, 31, -1; +$L__tmp55: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r380, %r378, %r379; +$L__tmp56: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r381, %r380, 1, 31, -1; +$L__tmp57: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r382, %r380, %r381; +$L__tmp58: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r383, %r326, 8, 31, -1; +$L__tmp59: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r384, %r326, %r383; +$L__tmp60: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r385, %r384, 4, 31, -1; +$L__tmp61: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r386, %r384, %r385; +$L__tmp62: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r387, %r386, 2, 31, -1; +$L__tmp63: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r388, %r386, %r387; +$L__tmp64: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r389, %r388, 1, 31, -1; +$L__tmp65: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r390, %r388, %r389; +$L__tmp66: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r391, %r334, %r119; + mul.f32 %r392, %r334, %r120; + mul.f32 %r393, %r334, %r121; + mul.f32 %r394, %r334, %r122; + mul.f32 %r395, %r334, %r123; + mul.f32 %r396, %r334, %r124; + mul.f32 %r397, %r334, %r125; + mul.f32 %r398, %r334, %r126; + mul.f32 %r399, %r342, %r127; + mul.f32 %r400, %r342, %r128; + mul.f32 %r401, %r342, %r129; + mul.f32 %r402, %r342, %r130; + mul.f32 %r403, %r342, %r131; + mul.f32 %r404, %r342, %r132; + mul.f32 %r405, %r342, %r133; + mul.f32 %r406, %r342, %r134; + mul.f32 %r407, %r350, %r135; + mul.f32 %r408, %r350, %r136; + mul.f32 %r409, %r350, %r137; + mul.f32 %r410, %r350, %r138; + mul.f32 %r411, %r350, %r139; + mul.f32 %r412, %r350, %r140; + mul.f32 %r413, %r350, %r141; + mul.f32 %r414, %r350, %r142; + mul.f32 %r415, %r358, %r143; + mul.f32 %r416, %r358, %r144; + mul.f32 %r417, %r358, %r145; + mul.f32 %r418, %r358, %r146; + mul.f32 %r419, %r358, %r147; + mul.f32 %r420, %r358, %r148; + mul.f32 %r421, %r358, %r149; + mul.f32 %r422, %r358, %r150; + mul.f32 %r423, %r366, %r151; + mul.f32 %r424, %r366, %r152; + mul.f32 %r425, %r366, %r153; + mul.f32 %r426, %r366, %r154; + mul.f32 %r427, %r366, %r155; + mul.f32 %r428, %r366, %r156; + mul.f32 %r429, %r366, %r157; + mul.f32 %r430, %r366, %r158; + mul.f32 %r431, %r374, %r159; + mul.f32 %r432, %r374, %r160; + mul.f32 %r433, %r374, %r161; + mul.f32 %r434, %r374, %r162; + mul.f32 %r435, %r374, %r163; + mul.f32 %r436, %r374, %r164; + mul.f32 %r437, %r374, %r165; + mul.f32 %r438, %r374, %r166; + mul.f32 %r439, %r382, %r167; + mul.f32 %r440, %r382, %r168; + mul.f32 %r441, %r382, %r169; + mul.f32 %r442, %r382, %r170; + mul.f32 %r443, %r382, %r171; + mul.f32 %r444, %r382, %r172; + mul.f32 %r445, %r382, %r173; + mul.f32 %r446, %r382, %r174; + mul.f32 %r447, %r390, %r175; + mul.f32 %r448, %r390, %r176; + mul.f32 %r449, %r390, %r177; + mul.f32 %r450, %r390, %r178; + mul.f32 %r451, %r390, %r179; + mul.f32 %r452, %r390, %r180; + mul.f32 %r453, %r390, %r181; + mul.f32 %r454, %r390, %r182; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r455, %r391, %r34; + mul.f32 %r456, %r392, %r34; + mul.f32 %r457, %r393, %r34; + mul.f32 %r458, %r394, %r34; + mul.f32 %r459, %r395, %r34; + mul.f32 %r460, %r396, %r34; + mul.f32 %r461, %r397, %r34; + mul.f32 %r462, %r398, %r34; + mul.f32 %r463, %r399, %r35; + mul.f32 %r464, %r400, %r35; + mul.f32 %r465, %r401, %r35; + mul.f32 %r466, %r402, %r35; + mul.f32 %r467, %r403, %r35; + mul.f32 %r468, %r404, %r35; + mul.f32 %r469, %r405, %r35; + mul.f32 %r470, %r406, %r35; + mul.f32 %r471, %r407, %r36; + mul.f32 %r472, %r408, %r36; + mul.f32 %r473, %r409, %r36; + mul.f32 %r474, %r410, %r36; + mul.f32 %r475, %r411, %r36; + mul.f32 %r476, %r412, %r36; + mul.f32 %r477, %r413, %r36; + mul.f32 %r478, %r414, %r36; + mul.f32 %r479, %r415, %r37; + mul.f32 %r480, %r416, %r37; + mul.f32 %r481, %r417, %r37; + mul.f32 %r482, %r418, %r37; + mul.f32 %r483, %r419, %r37; + mul.f32 %r484, %r420, %r37; + mul.f32 %r485, %r421, %r37; + mul.f32 %r486, %r422, %r37; + mul.f32 %r487, %r423, %r38; + mul.f32 %r488, %r424, %r38; + mul.f32 %r489, %r425, %r38; + mul.f32 %r490, %r426, %r38; + mul.f32 %r491, %r427, %r38; + mul.f32 %r492, %r428, %r38; + mul.f32 %r493, %r429, %r38; + mul.f32 %r494, %r430, %r38; + mul.f32 %r495, %r431, %r39; + mul.f32 %r496, %r432, %r39; + mul.f32 %r497, %r433, %r39; + mul.f32 %r498, %r434, %r39; + mul.f32 %r499, %r435, %r39; + mul.f32 %r500, %r436, %r39; + mul.f32 %r501, %r437, %r39; + mul.f32 %r502, %r438, %r39; + mul.f32 %r503, %r439, %r40; + mul.f32 %r504, %r440, %r40; + mul.f32 %r505, %r441, %r40; + mul.f32 %r506, %r442, %r40; + mul.f32 %r507, %r443, %r40; + mul.f32 %r508, %r444, %r40; + mul.f32 %r509, %r445, %r40; + mul.f32 %r510, %r446, %r40; + mul.f32 %r511, %r447, %r41; + mul.f32 %r512, %r448, %r41; + mul.f32 %r513, %r449, %r41; + mul.f32 %r514, %r450, %r41; + mul.f32 %r515, %r451, %r41; + mul.f32 %r516, %r452, %r41; + mul.f32 %r517, %r453, %r41; + mul.f32 %r518, %r454, %r41; + .loc 1 107 36 // l2norm.py:107:36 + neg.f32 %r519, %r455; + fma.rn.f32 %r520, %r34, %r183, %r519; + neg.f32 %r521, %r456; + fma.rn.f32 %r522, %r34, %r184, %r521; + neg.f32 %r523, %r457; + fma.rn.f32 %r524, %r34, %r185, %r523; + neg.f32 %r525, %r458; + fma.rn.f32 %r526, %r34, %r186, %r525; + neg.f32 %r527, %r459; + fma.rn.f32 %r528, %r34, %r187, %r527; + neg.f32 %r529, %r460; + fma.rn.f32 %r530, %r34, %r188, %r529; + neg.f32 %r531, %r461; + fma.rn.f32 %r532, %r34, %r189, %r531; + neg.f32 %r533, %r462; + fma.rn.f32 %r534, %r34, %r190, %r533; + neg.f32 %r535, %r463; + fma.rn.f32 %r536, %r35, %r191, %r535; + neg.f32 %r537, %r464; + fma.rn.f32 %r538, %r35, %r192, %r537; + neg.f32 %r539, %r465; + fma.rn.f32 %r540, %r35, %r193, %r539; + neg.f32 %r541, %r466; + fma.rn.f32 %r542, %r35, %r194, %r541; + neg.f32 %r543, %r467; + fma.rn.f32 %r544, %r35, %r195, %r543; + neg.f32 %r545, %r468; + fma.rn.f32 %r546, %r35, %r196, %r545; + neg.f32 %r547, %r469; + fma.rn.f32 %r548, %r35, %r197, %r547; + neg.f32 %r549, %r470; + fma.rn.f32 %r550, %r35, %r198, %r549; + neg.f32 %r551, %r471; + fma.rn.f32 %r552, %r36, %r199, %r551; + neg.f32 %r553, %r472; + fma.rn.f32 %r554, %r36, %r200, %r553; + neg.f32 %r555, %r473; + fma.rn.f32 %r556, %r36, %r201, %r555; + neg.f32 %r557, %r474; + fma.rn.f32 %r558, %r36, %r202, %r557; + neg.f32 %r559, %r475; + fma.rn.f32 %r560, %r36, %r203, %r559; + neg.f32 %r561, %r476; + fma.rn.f32 %r562, %r36, %r204, %r561; + neg.f32 %r563, %r477; + fma.rn.f32 %r564, %r36, %r205, %r563; + neg.f32 %r565, %r478; + fma.rn.f32 %r566, %r36, %r206, %r565; + neg.f32 %r567, %r479; + fma.rn.f32 %r568, %r37, %r207, %r567; + neg.f32 %r569, %r480; + fma.rn.f32 %r570, %r37, %r208, %r569; + neg.f32 %r571, %r481; + fma.rn.f32 %r572, %r37, %r209, %r571; + neg.f32 %r573, %r482; + fma.rn.f32 %r574, %r37, %r210, %r573; + neg.f32 %r575, %r483; + fma.rn.f32 %r576, %r37, %r211, %r575; + neg.f32 %r577, %r484; + fma.rn.f32 %r578, %r37, %r212, %r577; + neg.f32 %r579, %r485; + fma.rn.f32 %r580, %r37, %r213, %r579; + neg.f32 %r581, %r486; + fma.rn.f32 %r582, %r37, %r214, %r581; + neg.f32 %r583, %r487; + fma.rn.f32 %r584, %r38, %r215, %r583; + neg.f32 %r585, %r488; + fma.rn.f32 %r586, %r38, %r216, %r585; + neg.f32 %r587, %r489; + fma.rn.f32 %r588, %r38, %r217, %r587; + neg.f32 %r589, %r490; + fma.rn.f32 %r590, %r38, %r218, %r589; + neg.f32 %r591, %r491; + fma.rn.f32 %r592, %r38, %r219, %r591; + neg.f32 %r593, %r492; + fma.rn.f32 %r594, %r38, %r220, %r593; + neg.f32 %r595, %r493; + fma.rn.f32 %r596, %r38, %r221, %r595; + neg.f32 %r597, %r494; + fma.rn.f32 %r598, %r38, %r222, %r597; + neg.f32 %r599, %r495; + fma.rn.f32 %r600, %r39, %r223, %r599; + neg.f32 %r601, %r496; + fma.rn.f32 %r602, %r39, %r224, %r601; + neg.f32 %r603, %r497; + fma.rn.f32 %r604, %r39, %r225, %r603; + neg.f32 %r605, %r498; + fma.rn.f32 %r606, %r39, %r226, %r605; + neg.f32 %r607, %r499; + fma.rn.f32 %r608, %r39, %r227, %r607; + neg.f32 %r609, %r500; + fma.rn.f32 %r610, %r39, %r228, %r609; + neg.f32 %r611, %r501; + fma.rn.f32 %r612, %r39, %r229, %r611; + neg.f32 %r613, %r502; + fma.rn.f32 %r614, %r39, %r230, %r613; + mov.b64 %rd96, {%r247, %r248}; + mov.b64 %rd97, {%r503, %r504}; + sub.f32x2 %rd98, %rd96, %rd97; + mov.b64 %rd99, {%r249, %r250}; + mov.b64 %rd100, {%r505, %r506}; + sub.f32x2 %rd101, %rd99, %rd100; + mov.b64 %rd102, {%r251, %r252}; + mov.b64 %rd103, {%r507, %r508}; + sub.f32x2 %rd104, %rd102, %rd103; + mov.b64 %rd105, {%r253, %r254}; + mov.b64 %rd106, {%r509, %r510}; + sub.f32x2 %rd107, %rd105, %rd106; + mov.b64 %rd108, {%r255, %r256}; + mov.b64 %rd109, {%r511, %r512}; + sub.f32x2 %rd110, %rd108, %rd109; + mov.b64 %rd111, {%r257, %r258}; + mov.b64 %rd112, {%r513, %r514}; + sub.f32x2 %rd113, %rd111, %rd112; + mov.b64 %rd114, {%r259, %r260}; + mov.b64 %rd115, {%r515, %r516}; + sub.f32x2 %rd116, %rd114, %rd115; + mov.b64 %rd117, {%r261, %r262}; + mov.b64 %rd118, {%r517, %r518}; + sub.f32x2 %rd119, %rd117, %rd118; + .loc 1 108 27 // l2norm.py:108:27 + mov.b64 {%r615, %r616}, %rd98; + cvt.rn.bf16x2.f32 %r98, %r616, %r615; + mov.b64 {%r617, %r618}, %rd101; + cvt.rn.bf16x2.f32 %r99, %r618, %r617; + mov.b64 {%r619, %r620}, %rd104; + cvt.rn.bf16x2.f32 %r100, %r620, %r619; + mov.b64 {%r621, %r622}, %rd107; + cvt.rn.bf16x2.f32 %r101, %r622, %r621; + mov.b64 {%r623, %r624}, %rd110; + cvt.rn.bf16x2.f32 %r102, %r624, %r623; + mov.b64 {%r625, %r626}, %rd113; + cvt.rn.bf16x2.f32 %r103, %r626, %r625; + mov.b64 {%r627, %r628}, %rd116; + cvt.rn.bf16x2.f32 %r104, %r628, %r627; + mov.b64 {%r629, %r630}, %rd119; + cvt.rn.bf16x2.f32 %r105, %r630, %r629; + .loc 1 108 19 // l2norm.py:108:19 + cvt.rn.bf16x2.f32 %r74, %r522, %r520; + cvt.rn.bf16x2.f32 %r75, %r526, %r524; + cvt.rn.bf16x2.f32 %r76, %r530, %r528; + cvt.rn.bf16x2.f32 %r77, %r534, %r532; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd25 + 0 ], { %r74, %r75, %r76, %r77 }; + // end inline asm + cvt.rn.bf16x2.f32 %r78, %r538, %r536; + cvt.rn.bf16x2.f32 %r79, %r542, %r540; + cvt.rn.bf16x2.f32 %r80, %r546, %r544; + cvt.rn.bf16x2.f32 %r81, %r550, %r548; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd26 + 0 ], { %r78, %r79, %r80, %r81 }; + // end inline asm + cvt.rn.bf16x2.f32 %r82, %r554, %r552; + cvt.rn.bf16x2.f32 %r83, %r558, %r556; + cvt.rn.bf16x2.f32 %r84, %r562, %r560; + cvt.rn.bf16x2.f32 %r85, %r566, %r564; + // begin inline asm + @%p3 st.global.v4.b32 [ %rd27 + 0 ], { %r82, %r83, %r84, %r85 }; + // end inline asm + cvt.rn.bf16x2.f32 %r86, %r570, %r568; + cvt.rn.bf16x2.f32 %r87, %r574, %r572; + cvt.rn.bf16x2.f32 %r88, %r578, %r576; + cvt.rn.bf16x2.f32 %r89, %r582, %r580; + // begin inline asm + @%p4 st.global.v4.b32 [ %rd28 + 0 ], { %r86, %r87, %r88, %r89 }; + // end inline asm + cvt.rn.bf16x2.f32 %r90, %r586, %r584; + cvt.rn.bf16x2.f32 %r91, %r590, %r588; + cvt.rn.bf16x2.f32 %r92, %r594, %r592; + cvt.rn.bf16x2.f32 %r93, %r598, %r596; + // begin inline asm + @%p5 st.global.v4.b32 [ %rd29 + 0 ], { %r90, %r91, %r92, %r93 }; + // end inline asm + cvt.rn.bf16x2.f32 %r94, %r602, %r600; + cvt.rn.bf16x2.f32 %r95, %r606, %r604; + cvt.rn.bf16x2.f32 %r96, %r610, %r608; + cvt.rn.bf16x2.f32 %r97, %r614, %r612; + // begin inline asm + @%p6 st.global.v4.b32 [ %rd30 + 0 ], { %r94, %r95, %r96, %r97 }; + // end inline asm + // begin inline asm + @%p7 st.global.v4.b32 [ %rd31 + 0 ], { %r98, %r99, %r100, %r101 }; + // end inline asm + // begin inline asm + @%p8 st.global.v4.b32 [ %rd32 + 0 ], { %r102, %r103, %r104, %r105 }; + // end inline asm + .loc 1 108 4 // l2norm.py:108:4 + ret; +$L__tmp67: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 98 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp66 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 107 // DW_AT_call_line +.b8 55 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp66 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.source b/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..35265513bc03aa2afd3347f28178370dc92e754e --- /dev/null +++ b/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.source @@ -0,0 +1,210 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc47 = loc(unknown) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc54 = loc("y"(#loc)) +#loc55 = loc("rstd"(#loc)) +#loc56 = loc("dy"(#loc)) +#loc57 = loc("dx"(#loc)) +#loc58 = loc("eps"(#loc)) +#loc59 = loc("T"(#loc)) +#loc101 = loc("input"(#loc45)) +#loc102 = loc("a"(#loc50)) +#loc103 = loc("b"(#loc50)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc60) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc61) + %o_t = arith.constant 128 : i32 loc(#loc62) + %o_t_1 = arith.constant 128 : i64 loc(#loc62) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc62) + %o_t_3 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc63) + %o_t_4 = arith.extsi %o_t_3 : tensor<128xi32> to tensor<128xi64> loc(#loc64) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<128xi64> loc(#loc64) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<128xi64> loc(#loc64) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc65) + %m_t = arith.extsi %T : i32 to i64 loc(#loc66) + %m_t_7 = tt.splat %m_t : i64 -> tensor<128xi64> loc(#loc66) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<128xi64> loc(#loc66) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<128xi1> -> tensor<128x1xi1> loc(#loc67) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc68) + %m_x_10 = arith.constant 128 : i32 loc(#loc69) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc69) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc69) + %m_x_13 = tt.broadcast %m_x : tensor<128x1xi1> -> tensor<128x128xi1> loc(#loc70) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<128x128xi1> loc(#loc70) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<128x128xi1> loc(#loc70) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc71) + %p_y_16 = arith.constant 128 : i32 loc(#loc72) + %p_y_17 = arith.constant 128 : i64 loc(#loc72) + %p_y_18 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc72) + %p_y_19 = arith.muli %p_y, %p_y_18 : tensor<128x1xi64> loc(#loc72) + %p_y_20 = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc73) + %p_y_21 = tt.addptr %p_y_20, %p_y_19 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc73) + %p_y_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc74) + %p_y_23 = tt.broadcast %p_y_21 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc75) + %p_y_24 = tt.broadcast %p_y_22 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc75) + %p_y_25 = tt.addptr %p_y_23, %p_y_24 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc75) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr> loc(#loc76) + %p_rstd_26 = tt.addptr %p_rstd, %o_t_6 : tensor<128x!tt.ptr>, tensor<128xi64> loc(#loc76) + %p_dy = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc77) + %p_dy_27 = arith.constant 128 : i32 loc(#loc78) + %p_dy_28 = arith.constant 128 : i64 loc(#loc78) + %p_dy_29 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc78) + %p_dy_30 = arith.muli %p_dy, %p_dy_29 : tensor<128x1xi64> loc(#loc78) + %p_dy_31 = tt.splat %dy : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc79) + %p_dy_32 = tt.addptr %p_dy_31, %p_dy_30 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc79) + %p_dy_33 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc80) + %p_dy_34 = tt.broadcast %p_dy_32 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc81) + %p_dy_35 = tt.broadcast %p_dy_33 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc81) + %p_dy_36 = tt.addptr %p_dy_34, %p_dy_35 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc81) + %p_dx = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc82) + %p_dx_37 = arith.constant 128 : i32 loc(#loc83) + %p_dx_38 = arith.constant 128 : i64 loc(#loc83) + %p_dx_39 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc83) + %p_dx_40 = arith.muli %p_dx, %p_dx_39 : tensor<128x1xi64> loc(#loc83) + %p_dx_41 = tt.splat %dx : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc84) + %p_dx_42 = tt.addptr %p_dx_41, %p_dx_40 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc84) + %p_dx_43 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc85) + %p_dx_44 = tt.broadcast %p_dx_42 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc86) + %p_dx_45 = tt.broadcast %p_dx_43 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc86) + %p_dx_46 = tt.addptr %p_dx_44, %p_dx_45 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc86) + %b_y = arith.constant 0.000000e+00 : f32 loc(#loc87) + %b_y_47 = arith.constant dense<0.000000e+00> : tensor<128x128xf32> loc(#loc87) + %b_y_48 = arith.truncf %b_y_47 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc87) + %b_y_49 = tt.load %p_y_25, %m_x_15, %b_y_48 : tensor<128x128x!tt.ptr> loc(#loc87) + %b_y_50 = arith.extf %b_y_49 : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc88) + %b_rstd = arith.constant 0.000000e+00 : f32 loc(#loc89) + %b_rstd_51 = arith.constant dense<0.000000e+00> : tensor<128xf32> loc(#loc89) + %b_rstd_52 = tt.load %p_rstd_26, %m_t_8, %b_rstd_51 : tensor<128x!tt.ptr> loc(#loc89) + %b_dy = arith.constant 0.000000e+00 : f32 loc(#loc90) + %b_dy_53 = arith.constant dense<0.000000e+00> : tensor<128x128xf32> loc(#loc90) + %b_dy_54 = arith.truncf %b_dy_53 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc90) + %b_dy_55 = tt.load %p_dy_36, %m_x_15, %b_dy_54 : tensor<128x128x!tt.ptr> loc(#loc90) + %b_dy_56 = arith.extf %b_dy_55 : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc91) + %b_dx = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc92) + %b_dx_57 = tt.broadcast %b_dx : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc93) + %b_dx_58 = arith.mulf %b_dy_56, %b_dx_57 : tensor<128x128xf32> loc(#loc93) + %b_dx_59 = arith.mulf %b_dy_56, %b_y_50 : tensor<128x128xf32> loc(#loc94) + %b_dx_60 = tt.call @triton.language.standard.sum__fp32S128_128S_c1_cFalse_cNone(%b_dx_59) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc95) + %b_dx_61 = tt.expand_dims %b_dx_60 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc96) + %b_dx_62 = tt.broadcast %b_dx_61 : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc97) + %b_dx_63 = arith.mulf %b_dx_62, %b_y_50 : tensor<128x128xf32> loc(#loc97) + %b_dx_64 = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc98) + %b_dx_65 = tt.broadcast %b_dx_64 : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc99) + %b_dx_66 = arith.mulf %b_dx_63, %b_dx_65 : tensor<128x128xf32> loc(#loc99) + %b_dx_67 = arith.subf %b_dx_58, %b_dx_66 : tensor<128x128xf32> loc(#loc100) + %0 = arith.truncf %b_dx_67 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc42) + tt.store %p_dx_46, %0, %m_x_15 : tensor<128x128x!tt.ptr> loc(#loc43) + tt.return loc(#loc44) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S128_128S_c1_cFalse_cNone(%input: tensor<128x128xf32> loc("input"(#loc45))) -> tensor<128xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc46) + tt.reduce.return %2 : f32 loc(#loc46) + }) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc46) + tt.return %0 : tensor<128xf32> loc(#loc48) + ^bb1: // no predecessors + %1 = ub.poison : tensor<128xf32> loc(#loc49) + tt.return %1 : tensor<128xf32> loc(#loc49) + } loc(#loc45) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc50)), %b: f32 loc("b"(#loc50))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc51) + tt.return %0 : f32 loc(#loc52) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc53) + tt.return %1 : f32 loc(#loc53) + } loc(#loc50) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:31) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:39) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:20) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:31) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:39) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:82) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc60 = loc("i_t"(#loc1)) +#loc61 = loc("i_t"(#loc2)) +#loc62 = loc("o_t"(#loc3)) +#loc63 = loc("o_t"(#loc4)) +#loc64 = loc("o_t"(#loc5)) +#loc65 = loc("o_d"(#loc6)) +#loc66 = loc("m_t"(#loc7)) +#loc67 = loc("m_x"(#loc8)) +#loc68 = loc("m_x"(#loc9)) +#loc69 = loc("m_x"(#loc10)) +#loc70 = loc("m_x"(#loc11)) +#loc71 = loc("p_y"(#loc12)) +#loc72 = loc("p_y"(#loc13)) +#loc73 = loc("p_y"(#loc14)) +#loc74 = loc("p_y"(#loc15)) +#loc75 = loc("p_y"(#loc16)) +#loc76 = loc("p_rstd"(#loc17)) +#loc77 = loc("p_dy"(#loc18)) +#loc78 = loc("p_dy"(#loc19)) +#loc79 = loc("p_dy"(#loc20)) +#loc80 = loc("p_dy"(#loc21)) +#loc81 = loc("p_dy"(#loc22)) +#loc82 = loc("p_dx"(#loc23)) +#loc83 = loc("p_dx"(#loc24)) +#loc84 = loc("p_dx"(#loc25)) +#loc85 = loc("p_dx"(#loc26)) +#loc86 = loc("p_dx"(#loc27)) +#loc87 = loc("b_y"(#loc28)) +#loc88 = loc("b_y"(#loc29)) +#loc89 = loc("b_rstd"(#loc30)) +#loc90 = loc("b_dy"(#loc31)) +#loc91 = loc("b_dy"(#loc32)) +#loc92 = loc("b_dx"(#loc33)) +#loc93 = loc("b_dx"(#loc34)) +#loc94 = loc("b_dx"(#loc35)) +#loc95 = loc("b_dx"(#loc36)) +#loc96 = loc("b_dx"(#loc37)) +#loc97 = loc("b_dx"(#loc38)) +#loc98 = loc("b_dx"(#loc39)) +#loc99 = loc("b_dx"(#loc40)) +#loc100 = loc("b_dx"(#loc41)) diff --git a/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.ttgir b/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..cf4819b5fa01984c7a6220e72e56e103945fa67d --- /dev/null +++ b/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.ttgir @@ -0,0 +1,146 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [8, 1], order = [1, 0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc39 = loc("y"(#loc)) +#loc40 = loc("rstd"(#loc)) +#loc41 = loc("dy"(#loc)) +#loc42 = loc("dx"(#loc)) +#loc43 = loc("eps"(#loc)) +#loc44 = loc("T"(#loc)) +#loc72 = loc("b_dx"(#loc30)) +#loc78 = loc(callsite(#loc1 at #loc72)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 8 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %c128_i64 = arith.constant 128 : i64 loc(#loc1) + %cst = arith.constant dense<0.000000e+00> : tensor<128x128xbf16, #blocked> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<128x1xi64, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc45) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc46) + %o_t = arith.muli %i_t_3, %c128_i64 : i64 loc(#loc47) + %o_t_4 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc48) + %o_t_5 = arith.extsi %o_t_4 : tensor<128xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %o_t_6 = tt.splat %o_t : i64 -> tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %o_t_7 = arith.addi %o_t_6, %o_t_5 : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %m_t = arith.extsi %T : i32 to i64 loc(#loc50) + %m_t_8 = tt.splat %m_t : i64 -> tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc50) + %m_t_9 = arith.cmpi slt, %o_t_7, %m_t_8 : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc50) + %m_x = tt.expand_dims %m_t_9 {axis = 1 : i32} : tensor<128xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128x1xi1, #blocked> loc(#loc51) + %m_x_10 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc52) + %m_x_11 = tt.expand_dims %m_x_10 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc52) + %m_x_12 = arith.cmpi slt, %m_x_11, %cst_2 : tensor<1x128xi32, #blocked> loc(#loc53) + %m_x_13 = tt.broadcast %m_x : tensor<128x1xi1, #blocked> -> tensor<128x128xi1, #blocked> loc(#loc54) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1, #blocked> -> tensor<128x128xi1, #blocked> loc(#loc54) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<128x128xi1, #blocked> loc(#loc54) + %p_y = tt.expand_dims %o_t_7 {axis = 1 : i32} : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128x1xi64, #blocked> loc(#loc55) + %p_y_16 = arith.muli %p_y, %cst_1 : tensor<128x1xi64, #blocked> loc(#loc56) + %p_y_17 = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr, #blocked> loc(#loc57) + %p_y_18 = tt.addptr %p_y_17, %p_y_16 : tensor<128x1x!tt.ptr, #blocked>, tensor<128x1xi64, #blocked> loc(#loc57) + %p_y_19 = tt.broadcast %p_y_18 : tensor<128x1x!tt.ptr, #blocked> -> tensor<128x128x!tt.ptr, #blocked> loc(#loc58) + %p_y_20 = tt.broadcast %m_x_11 : tensor<1x128xi32, #blocked> -> tensor<128x128xi32, #blocked> loc(#loc58) + %p_y_21 = tt.addptr %p_y_19, %p_y_20 : tensor<128x128x!tt.ptr, #blocked>, tensor<128x128xi32, #blocked> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc59) + %p_rstd_22 = tt.addptr %p_rstd, %o_t_7 : tensor<128x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>>, tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc59) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<128x1x!tt.ptr, #blocked> loc(#loc60) + %p_dy_23 = tt.addptr %p_dy, %p_y_16 : tensor<128x1x!tt.ptr, #blocked>, tensor<128x1xi64, #blocked> loc(#loc60) + %p_dy_24 = tt.broadcast %p_dy_23 : tensor<128x1x!tt.ptr, #blocked> -> tensor<128x128x!tt.ptr, #blocked> loc(#loc61) + %p_dy_25 = tt.addptr %p_dy_24, %p_y_20 : tensor<128x128x!tt.ptr, #blocked>, tensor<128x128xi32, #blocked> loc(#loc61) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<128x1x!tt.ptr, #blocked> loc(#loc62) + %p_dx_26 = tt.addptr %p_dx, %p_y_16 : tensor<128x1x!tt.ptr, #blocked>, tensor<128x1xi64, #blocked> loc(#loc62) + %p_dx_27 = tt.broadcast %p_dx_26 : tensor<128x1x!tt.ptr, #blocked> -> tensor<128x128x!tt.ptr, #blocked> loc(#loc63) + %p_dx_28 = tt.addptr %p_dx_27, %p_y_20 : tensor<128x128x!tt.ptr, #blocked>, tensor<128x128xi32, #blocked> loc(#loc63) + %b_y = tt.load %p_y_21, %m_x_15, %cst : tensor<128x128x!tt.ptr, #blocked> loc(#loc64) + %b_y_29 = arith.extf %b_y : tensor<128x128xbf16, #blocked> to tensor<128x128xf32, #blocked> loc(#loc65) + %b_rstd = tt.load %p_rstd_22, %m_t_9, %cst_0 : tensor<128x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc66) + %b_dy = tt.load %p_dy_25, %m_x_15, %cst : tensor<128x128x!tt.ptr, #blocked> loc(#loc67) + %b_dy_30 = arith.extf %b_dy : tensor<128x128xbf16, #blocked> to tensor<128x128xf32, #blocked> loc(#loc68) + %b_dx = tt.expand_dims %b_rstd {axis = 1 : i32} : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128x1xf32, #blocked> loc(#loc69) + %b_dx_31 = tt.broadcast %b_dx : tensor<128x1xf32, #blocked> -> tensor<128x128xf32, #blocked> loc(#loc70) + %b_dx_32 = arith.mulf %b_dy_30, %b_dx_31 : tensor<128x128xf32, #blocked> loc(#loc70) + %b_dx_33 = arith.mulf %b_dy_30, %b_y_29 : tensor<128x128xf32, #blocked> loc(#loc71) + %b_dx_34 = "tt.reduce"(%b_dx_33) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_40: f32 loc(callsite(#loc1 at #loc72)), %b_dx_41: f32 loc(callsite(#loc1 at #loc72))): + %b_dx_42 = arith.addf %b_dx_40, %b_dx_41 : f32 loc(#loc79) + tt.reduce.return %b_dx_42 : f32 loc(#loc77) + }) : (tensor<128x128xf32, #blocked>) -> tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc77) + %b_dx_35 = tt.expand_dims %b_dx_34 {axis = 1 : i32} : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128x1xf32, #blocked> loc(#loc73) + %b_dx_36 = tt.broadcast %b_dx_35 : tensor<128x1xf32, #blocked> -> tensor<128x128xf32, #blocked> loc(#loc74) + %b_dx_37 = arith.mulf %b_dx_36, %b_y_29 : tensor<128x128xf32, #blocked> loc(#loc74) + %b_dx_38 = arith.mulf %b_dx_37, %b_dx_31 : tensor<128x128xf32, #blocked> loc(#loc75) + %b_dx_39 = arith.subf %b_dx_32, %b_dx_38 : tensor<128x128xf32, #blocked> loc(#loc76) + %0 = arith.truncf %b_dx_39 : tensor<128x128xf32, #blocked> to tensor<128x128xbf16, #blocked> loc(#loc36) + tt.store %p_dx_28, %0, %m_x_15 : tensor<128x128x!tt.ptr, #blocked> loc(#loc37) + tt.return loc(#loc38) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc45 = loc("i_t"(#loc2)) +#loc46 = loc("i_t"(#loc3)) +#loc47 = loc("o_t"(#loc4)) +#loc48 = loc("o_t"(#loc5)) +#loc49 = loc("o_t"(#loc6)) +#loc50 = loc("m_t"(#loc7)) +#loc51 = loc("m_x"(#loc8)) +#loc52 = loc("m_x"(#loc9)) +#loc53 = loc("m_x"(#loc10)) +#loc54 = loc("m_x"(#loc11)) +#loc55 = loc("p_y"(#loc12)) +#loc56 = loc("p_y"(#loc13)) +#loc57 = loc("p_y"(#loc14)) +#loc58 = loc("p_y"(#loc15)) +#loc59 = loc("p_rstd"(#loc16)) +#loc60 = loc("p_dy"(#loc17)) +#loc61 = loc("p_dy"(#loc18)) +#loc62 = loc("p_dx"(#loc19)) +#loc63 = loc("p_dx"(#loc20)) +#loc64 = loc("b_y"(#loc21)) +#loc65 = loc("b_y"(#loc22)) +#loc66 = loc("b_rstd"(#loc23)) +#loc67 = loc("b_dy"(#loc24)) +#loc68 = loc("b_dy"(#loc25)) +#loc69 = loc("b_dx"(#loc26)) +#loc70 = loc("b_dx"(#loc27)) +#loc71 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc32)) +#loc74 = loc("b_dx"(#loc33)) +#loc75 = loc("b_dx"(#loc34)) +#loc76 = loc("b_dx"(#loc35)) +#loc77 = loc(callsite(#loc29 at #loc72)) +#loc79 = loc(callsite(#loc31 at #loc77)) diff --git a/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.ttir b/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..f242766466ced3ec80bd72ef6238b6accc66dbe7 --- /dev/null +++ b/triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.ttir @@ -0,0 +1,144 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc39 = loc("y"(#loc)) +#loc40 = loc("rstd"(#loc)) +#loc41 = loc("dy"(#loc)) +#loc42 = loc("dx"(#loc)) +#loc43 = loc("eps"(#loc)) +#loc44 = loc("T"(#loc)) +#loc72 = loc("b_dx"(#loc30)) +#loc78 = loc(callsite(#loc1 at #loc72)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<0.000000e+00> : tensor<128x128xbf16> loc(#loc1) + %b_rstd = arith.constant dense<0.000000e+00> : tensor<128xf32> loc(#loc45) + %cst_0 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc1) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc46) + %c128_i64 = arith.constant 128 : i64 loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc47) + %i_t_1 = arith.extsi %i_t : i32 to i64 loc(#loc48) + %o_t = arith.muli %i_t_1, %c128_i64 : i64 loc(#loc49) + %o_t_2 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc50) + %o_t_3 = arith.extsi %o_t_2 : tensor<128xi32> to tensor<128xi64> loc(#loc51) + %o_t_4 = tt.splat %o_t : i64 -> tensor<128xi64> loc(#loc51) + %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<128xi64> loc(#loc51) + %m_t = arith.extsi %T : i32 to i64 loc(#loc52) + %m_t_6 = tt.splat %m_t : i64 -> tensor<128xi64> loc(#loc52) + %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<128xi64> loc(#loc52) + %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<128xi1> -> tensor<128x1xi1> loc(#loc53) + %m_x_9 = tt.expand_dims %o_t_2 {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc54) + %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc46) + %m_x_11 = tt.broadcast %m_x_8 : tensor<128x1xi1> -> tensor<128x128xi1> loc(#loc55) + %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<128x128xi1> loc(#loc55) + %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<128x128xi1> loc(#loc55) + %p_y = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc56) + %p_y_14 = arith.muli %p_y, %cst_0 : tensor<128x1xi64> loc(#loc57) + %p_y_15 = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc58) + %p_y_16 = tt.addptr %p_y_15, %p_y_14 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc58) + %p_y_17 = tt.broadcast %p_y_16 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc59) + %p_y_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc59) + %p_y_19 = tt.addptr %p_y_17, %p_y_18 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc59) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr> loc(#loc60) + %p_rstd_20 = tt.addptr %p_rstd, %o_t_5 : tensor<128x!tt.ptr>, tensor<128xi64> loc(#loc60) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc61) + %p_dy_21 = tt.addptr %p_dy, %p_y_14 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc61) + %p_dy_22 = tt.broadcast %p_dy_21 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc62) + %p_dy_23 = tt.addptr %p_dy_22, %p_y_18 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc62) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc63) + %p_dx_24 = tt.addptr %p_dx, %p_y_14 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc63) + %p_dx_25 = tt.broadcast %p_dx_24 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc64) + %p_dx_26 = tt.addptr %p_dx_25, %p_y_18 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc64) + %b_y = tt.load %p_y_19, %m_x_13, %cst : tensor<128x128x!tt.ptr> loc(#loc65) + %b_y_27 = arith.extf %b_y : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc66) + %b_rstd_28 = tt.load %p_rstd_20, %m_t_7, %b_rstd : tensor<128x!tt.ptr> loc(#loc45) + %b_dy = tt.load %p_dy_23, %m_x_13, %cst : tensor<128x128x!tt.ptr> loc(#loc67) + %b_dy_29 = arith.extf %b_dy : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc68) + %b_dx = tt.expand_dims %b_rstd_28 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc69) + %b_dx_30 = tt.broadcast %b_dx : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc70) + %b_dx_31 = arith.mulf %b_dy_29, %b_dx_30 : tensor<128x128xf32> loc(#loc70) + %b_dx_32 = arith.mulf %b_dy_29, %b_y_27 : tensor<128x128xf32> loc(#loc71) + %b_dx_33 = "tt.reduce"(%b_dx_32) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_39: f32 loc(callsite(#loc1 at #loc72)), %b_dx_40: f32 loc(callsite(#loc1 at #loc72))): + %b_dx_41 = arith.addf %b_dx_39, %b_dx_40 : f32 loc(#loc79) + tt.reduce.return %b_dx_41 : f32 loc(#loc77) + }) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc77) + %b_dx_34 = tt.expand_dims %b_dx_33 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc73) + %b_dx_35 = tt.broadcast %b_dx_34 : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc74) + %b_dx_36 = arith.mulf %b_dx_35, %b_y_27 : tensor<128x128xf32> loc(#loc74) + %b_dx_37 = arith.mulf %b_dx_36, %b_dx_30 : tensor<128x128xf32> loc(#loc75) + %b_dx_38 = arith.subf %b_dx_31, %b_dx_37 : tensor<128x128xf32> loc(#loc76) + %0 = arith.truncf %b_dx_38 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc36) + tt.store %p_dx_26, %0, %m_x_13 : tensor<128x128x!tt.ptr> loc(#loc37) + tt.return loc(#loc38) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc45 = loc("b_rstd"(#loc2)) +#loc46 = loc("m_x"(#loc3)) +#loc47 = loc("i_t"(#loc4)) +#loc48 = loc("i_t"(#loc5)) +#loc49 = loc("o_t"(#loc6)) +#loc50 = loc("o_t"(#loc7)) +#loc51 = loc("o_t"(#loc8)) +#loc52 = loc("m_t"(#loc9)) +#loc53 = loc("m_x"(#loc10)) +#loc54 = loc("m_x"(#loc11)) +#loc55 = loc("m_x"(#loc12)) +#loc56 = loc("p_y"(#loc13)) +#loc57 = loc("p_y"(#loc14)) +#loc58 = loc("p_y"(#loc15)) +#loc59 = loc("p_y"(#loc16)) +#loc60 = loc("p_rstd"(#loc17)) +#loc61 = loc("p_dy"(#loc18)) +#loc62 = loc("p_dy"(#loc19)) +#loc63 = loc("p_dx"(#loc20)) +#loc64 = loc("p_dx"(#loc21)) +#loc65 = loc("b_y"(#loc22)) +#loc66 = loc("b_y"(#loc23)) +#loc67 = loc("b_dy"(#loc24)) +#loc68 = loc("b_dy"(#loc25)) +#loc69 = loc("b_dx"(#loc26)) +#loc70 = loc("b_dx"(#loc27)) +#loc71 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc32)) +#loc74 = loc("b_dx"(#loc33)) +#loc75 = loc("b_dx"(#loc34)) +#loc76 = loc("b_dx"(#loc35)) +#loc77 = loc(callsite(#loc29 at #loc72)) +#loc79 = loc(callsite(#loc31 at #loc77)) diff --git a/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/__grp__l2norm_bwd_kernel.json b/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/__grp__l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..4e35946e40f3f9d028e6aaf51b9c27f9ace0351c --- /dev/null +++ b/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/__grp__l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_bwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.source", "l2norm_bwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.ttir", "l2norm_bwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.ttgir", "l2norm_bwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.llir", "l2norm_bwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.ptx", "l2norm_bwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.cubin", "l2norm_bwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.cubin b/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..d144bd686bcb0724cb3f120aa07fb76de9d9095e --- /dev/null +++ b/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.cubin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:891a986d506009060f5cc64e4ab10f2b7bbcaa1fbdb0463144c7091e004da2f5 +size 241800 diff --git a/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.json b/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..e1679f3979d52203a4f147199b12f2c06f7d3872 --- /dev/null +++ b/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"hash": "05ec39cf6851b972d756ce8d1d615f5fcd72533ca11f5424780a0f8b299ed2dc", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 2, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 512, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_bwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.llir b/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..a77fd8a23fa93b6b096cc3aab501ebbd80e2193c --- /dev/null +++ b/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.llir @@ -0,0 +1,4790 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_bwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, ptr addrspace(1) %3, float %4, i32 %5, ptr addrspace(1) readnone captures(none) %6, ptr addrspace(1) readnone captures(none) %7) local_unnamed_addr #0 !dbg !4 { + %9 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !12 + %10 = zext nneg i32 %9 to i64, !dbg !13 + %11 = shl nuw nsw i64 %10, 7, !dbg !14 + %12 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !15 + %13 = lshr i32 %12, 4, !dbg !15 + %14 = and i32 %13, 3, !dbg !15 + %15 = or disjoint i32 %14, 4, !dbg !15 + %16 = shl nuw nsw i32 %12, 1, !dbg !15 + %17 = zext nneg i32 %14 to i64, !dbg !16 + %18 = zext nneg i32 %15 to i64, !dbg !16 + %19 = or disjoint i64 %11, %17, !dbg !16 + %20 = or disjoint i64 %11, %18, !dbg !16 + %21 = sext i32 %5 to i64, !dbg !17 + %22 = icmp slt i64 %19, %21, !dbg !17 + %23 = icmp slt i64 %20, %21, !dbg !17 + %24 = insertelement <16 x i32> poison, i32 %14, i64 0, !dbg !15 + %25 = insertelement <16 x i32> %24, i32 %13, i64 1, !dbg !15 + %26 = insertelement <8 x i32> poison, i32 %14, i64 0, !dbg !15 + %27 = insertelement <4 x i32> poison, i32 %14, i64 0, !dbg !15 + %28 = or disjoint i32 %14, 120, !dbg !15 + %29 = or i32 %13, 124, !dbg !15 + %30 = and i32 %16, 126, !dbg !15 + %31 = or disjoint i32 %30, 1, !dbg !15 + %32 = insertelement <32 x i32> poison, i32 %31, i64 0, !dbg !16 + %33 = insertelement <32 x i32> %32, i32 %30, i64 1, !dbg !16 + %34 = insertelement <32 x i32> %33, i32 %29, i64 2, !dbg !16 + %35 = insertelement <32 x i32> %34, i32 %28, i64 3, !dbg !16 + %36 = shufflevector <16 x i32> %25, <16 x i32> poison, <32 x i32> , !dbg !16 + %37 = or <32 x i32> %36, , !dbg !16 + %38 = shufflevector <32 x i32> %35, <32 x i32> %37, <32 x i32> , !dbg !16 + %39 = shufflevector <8 x i32> %26, <8 x i32> poison, <32 x i32> , !dbg !16 + %40 = or disjoint <32 x i32> %39, , !dbg !16 + %41 = shufflevector <32 x i32> %38, <32 x i32> %40, <32 x i32> , !dbg !16 + %42 = shufflevector <4 x i32> %27, <4 x i32> poison, <32 x i32> , !dbg !16 + %43 = or disjoint <32 x i32> %42, , !dbg !16 + %44 = shufflevector <32 x i32> %41, <32 x i32> %43, <32 x i32> , !dbg !16 + %45 = zext <32 x i32> %44 to <32 x i64>, !dbg !16 + %46 = insertelement <32 x i64> poison, i64 %11, i64 0, !dbg !16 + %47 = shufflevector <32 x i64> %46, <32 x i64> poison, <32 x i32> zeroinitializer, !dbg !16 + %48 = or disjoint <32 x i64> %47, %45, !dbg !16 + %49 = insertelement <32 x i32> poison, i32 %5, i64 0, !dbg !17 + %50 = shufflevector <32 x i32> %49, <32 x i32> poison, <32 x i32> zeroinitializer, !dbg !17 + %51 = sext <32 x i32> %50 to <32 x i64>, !dbg !17 + %52 = icmp slt <32 x i64> %48, %51, !dbg !17 + %53 = shl nuw nsw i32 %12, 3, !dbg !18 + %54 = and i32 %53, 120, !dbg !18 + %55 = shl nuw nsw i64 %19, 7, !dbg !19 + %56 = shl nuw nsw i64 %20, 7, !dbg !19 + %57 = extractelement <32 x i64> %48, i64 31, !dbg !19 + %58 = shl nuw nsw i64 %57, 7, !dbg !19 + %59 = extractelement <32 x i64> %48, i64 30, !dbg !19 + %60 = shl nuw nsw i64 %59, 7, !dbg !19 + %61 = extractelement <32 x i64> %48, i64 29, !dbg !19 + %62 = shl nuw nsw i64 %61, 7, !dbg !19 + %63 = extractelement <32 x i64> %48, i64 28, !dbg !19 + %64 = shl nuw nsw i64 %63, 7, !dbg !19 + %65 = extractelement <32 x i64> %48, i64 27, !dbg !19 + %66 = shl nuw nsw i64 %65, 7, !dbg !19 + %67 = extractelement <32 x i64> %48, i64 26, !dbg !19 + %68 = shl nuw nsw i64 %67, 7, !dbg !19 + %69 = extractelement <32 x i64> %48, i64 25, !dbg !19 + %70 = shl nuw nsw i64 %69, 7, !dbg !19 + %71 = extractelement <32 x i64> %48, i64 24, !dbg !19 + %72 = shl nuw nsw i64 %71, 7, !dbg !19 + %73 = extractelement <32 x i64> %48, i64 23, !dbg !19 + %74 = shl nuw nsw i64 %73, 7, !dbg !19 + %75 = extractelement <32 x i64> %48, i64 22, !dbg !19 + %76 = shl nuw nsw i64 %75, 7, !dbg !19 + %77 = extractelement <32 x i64> %48, i64 21, !dbg !19 + %78 = shl nuw nsw i64 %77, 7, !dbg !19 + %79 = extractelement <32 x i64> %48, i64 20, !dbg !19 + %80 = shl nuw nsw i64 %79, 7, !dbg !19 + %81 = extractelement <32 x i64> %48, i64 19, !dbg !19 + %82 = shl nuw nsw i64 %81, 7, !dbg !19 + %83 = extractelement <32 x i64> %48, i64 18, !dbg !19 + %84 = shl nuw nsw i64 %83, 7, !dbg !19 + %85 = extractelement <32 x i64> %48, i64 17, !dbg !19 + %86 = shl nuw nsw i64 %85, 7, !dbg !19 + %87 = extractelement <32 x i64> %48, i64 16, !dbg !19 + %88 = shl nuw nsw i64 %87, 7, !dbg !19 + %89 = extractelement <32 x i64> %48, i64 15, !dbg !19 + %90 = shl nuw nsw i64 %89, 7, !dbg !19 + %91 = extractelement <32 x i64> %48, i64 14, !dbg !19 + %92 = shl nuw nsw i64 %91, 7, !dbg !19 + %93 = extractelement <32 x i64> %48, i64 13, !dbg !19 + %94 = shl nuw nsw i64 %93, 7, !dbg !19 + %95 = extractelement <32 x i64> %48, i64 12, !dbg !19 + %96 = shl nuw nsw i64 %95, 7, !dbg !19 + %97 = extractelement <32 x i64> %48, i64 11, !dbg !19 + %98 = shl nuw nsw i64 %97, 7, !dbg !19 + %99 = extractelement <32 x i64> %48, i64 10, !dbg !19 + %100 = shl nuw nsw i64 %99, 7, !dbg !19 + %101 = extractelement <32 x i64> %48, i64 9, !dbg !19 + %102 = shl nuw nsw i64 %101, 7, !dbg !19 + %103 = extractelement <32 x i64> %48, i64 8, !dbg !19 + %104 = shl nuw nsw i64 %103, 7, !dbg !19 + %105 = extractelement <32 x i64> %48, i64 7, !dbg !19 + %106 = shl nuw nsw i64 %105, 7, !dbg !19 + %107 = extractelement <32 x i64> %48, i64 6, !dbg !19 + %108 = shl nuw nsw i64 %107, 7, !dbg !19 + %109 = extractelement <32 x i64> %48, i64 5, !dbg !19 + %110 = shl nuw nsw i64 %109, 7, !dbg !19 + %111 = extractelement <32 x i64> %48, i64 4, !dbg !19 + %112 = shl nuw nsw i64 %111, 7, !dbg !19 + %113 = extractelement <32 x i64> %48, i64 3, !dbg !19 + %114 = shl nuw nsw i64 %113, 7, !dbg !19 + %115 = extractelement <32 x i64> %48, i64 2, !dbg !19 + %116 = shl nuw nsw i64 %115, 7, !dbg !19 + %117 = getelementptr bfloat, ptr addrspace(1) %0, i64 %55, !dbg !20 + %118 = getelementptr bfloat, ptr addrspace(1) %0, i64 %56, !dbg !20 + %119 = getelementptr bfloat, ptr addrspace(1) %0, i64 %58, !dbg !20 + %120 = getelementptr bfloat, ptr addrspace(1) %0, i64 %60, !dbg !20 + %121 = getelementptr bfloat, ptr addrspace(1) %0, i64 %62, !dbg !20 + %122 = getelementptr bfloat, ptr addrspace(1) %0, i64 %64, !dbg !20 + %123 = getelementptr bfloat, ptr addrspace(1) %0, i64 %66, !dbg !20 + %124 = getelementptr bfloat, ptr addrspace(1) %0, i64 %68, !dbg !20 + %125 = getelementptr bfloat, ptr addrspace(1) %0, i64 %70, !dbg !20 + %126 = getelementptr bfloat, ptr addrspace(1) %0, i64 %72, !dbg !20 + %127 = getelementptr bfloat, ptr addrspace(1) %0, i64 %74, !dbg !20 + %128 = getelementptr bfloat, ptr addrspace(1) %0, i64 %76, !dbg !20 + %129 = getelementptr bfloat, ptr addrspace(1) %0, i64 %78, !dbg !20 + %130 = getelementptr bfloat, ptr addrspace(1) %0, i64 %80, !dbg !20 + %131 = getelementptr bfloat, ptr addrspace(1) %0, i64 %82, !dbg !20 + %132 = getelementptr bfloat, ptr addrspace(1) %0, i64 %84, !dbg !20 + %133 = getelementptr bfloat, ptr addrspace(1) %0, i64 %86, !dbg !20 + %134 = getelementptr bfloat, ptr addrspace(1) %0, i64 %88, !dbg !20 + %135 = getelementptr bfloat, ptr addrspace(1) %0, i64 %90, !dbg !20 + %136 = getelementptr bfloat, ptr addrspace(1) %0, i64 %92, !dbg !20 + %137 = getelementptr bfloat, ptr addrspace(1) %0, i64 %94, !dbg !20 + %138 = getelementptr bfloat, ptr addrspace(1) %0, i64 %96, !dbg !20 + %139 = getelementptr bfloat, ptr addrspace(1) %0, i64 %98, !dbg !20 + %140 = getelementptr bfloat, ptr addrspace(1) %0, i64 %100, !dbg !20 + %141 = getelementptr bfloat, ptr addrspace(1) %0, i64 %102, !dbg !20 + %142 = getelementptr bfloat, ptr addrspace(1) %0, i64 %104, !dbg !20 + %143 = getelementptr bfloat, ptr addrspace(1) %0, i64 %106, !dbg !20 + %144 = getelementptr bfloat, ptr addrspace(1) %0, i64 %108, !dbg !20 + %145 = getelementptr bfloat, ptr addrspace(1) %0, i64 %110, !dbg !20 + %146 = getelementptr bfloat, ptr addrspace(1) %0, i64 %112, !dbg !20 + %147 = getelementptr bfloat, ptr addrspace(1) %0, i64 %114, !dbg !20 + %148 = getelementptr bfloat, ptr addrspace(1) %0, i64 %116, !dbg !20 + %149 = zext nneg i32 %54 to i64, !dbg !21 + %150 = getelementptr bfloat, ptr addrspace(1) %117, i64 %149, !dbg !21 + %151 = getelementptr bfloat, ptr addrspace(1) %118, i64 %149, !dbg !21 + %152 = getelementptr bfloat, ptr addrspace(1) %119, i64 %149, !dbg !21 + %153 = getelementptr bfloat, ptr addrspace(1) %120, i64 %149, !dbg !21 + %154 = getelementptr bfloat, ptr addrspace(1) %121, i64 %149, !dbg !21 + %155 = getelementptr bfloat, ptr addrspace(1) %122, i64 %149, !dbg !21 + %156 = getelementptr bfloat, ptr addrspace(1) %123, i64 %149, !dbg !21 + %157 = getelementptr bfloat, ptr addrspace(1) %124, i64 %149, !dbg !21 + %158 = getelementptr bfloat, ptr addrspace(1) %125, i64 %149, !dbg !21 + %159 = getelementptr bfloat, ptr addrspace(1) %126, i64 %149, !dbg !21 + %160 = getelementptr bfloat, ptr addrspace(1) %127, i64 %149, !dbg !21 + %161 = getelementptr bfloat, ptr addrspace(1) %128, i64 %149, !dbg !21 + %162 = getelementptr bfloat, ptr addrspace(1) %129, i64 %149, !dbg !21 + %163 = getelementptr bfloat, ptr addrspace(1) %130, i64 %149, !dbg !21 + %164 = getelementptr bfloat, ptr addrspace(1) %131, i64 %149, !dbg !21 + %165 = getelementptr bfloat, ptr addrspace(1) %132, i64 %149, !dbg !21 + %166 = getelementptr bfloat, ptr addrspace(1) %133, i64 %149, !dbg !21 + %167 = getelementptr bfloat, ptr addrspace(1) %134, i64 %149, !dbg !21 + %168 = getelementptr bfloat, ptr addrspace(1) %135, i64 %149, !dbg !21 + %169 = getelementptr bfloat, ptr addrspace(1) %136, i64 %149, !dbg !21 + %170 = getelementptr bfloat, ptr addrspace(1) %137, i64 %149, !dbg !21 + %171 = getelementptr bfloat, ptr addrspace(1) %138, i64 %149, !dbg !21 + %172 = getelementptr bfloat, ptr addrspace(1) %139, i64 %149, !dbg !21 + %173 = getelementptr bfloat, ptr addrspace(1) %140, i64 %149, !dbg !21 + %174 = getelementptr bfloat, ptr addrspace(1) %141, i64 %149, !dbg !21 + %175 = getelementptr bfloat, ptr addrspace(1) %142, i64 %149, !dbg !21 + %176 = getelementptr bfloat, ptr addrspace(1) %143, i64 %149, !dbg !21 + %177 = getelementptr bfloat, ptr addrspace(1) %144, i64 %149, !dbg !21 + %178 = getelementptr bfloat, ptr addrspace(1) %145, i64 %149, !dbg !21 + %179 = getelementptr bfloat, ptr addrspace(1) %146, i64 %149, !dbg !21 + %180 = getelementptr bfloat, ptr addrspace(1) %147, i64 %149, !dbg !21 + %181 = getelementptr bfloat, ptr addrspace(1) %148, i64 %149, !dbg !21 + %182 = extractelement <32 x i64> %48, i64 1, !dbg !22 + %183 = getelementptr float, ptr addrspace(1) %1, i64 %182, !dbg !22 + %184 = extractelement <32 x i64> %48, i64 0, !dbg !22 + %185 = getelementptr float, ptr addrspace(1) %1, i64 %184, !dbg !22 + %186 = getelementptr bfloat, ptr addrspace(1) %2, i64 %55, !dbg !23 + %187 = getelementptr bfloat, ptr addrspace(1) %2, i64 %56, !dbg !23 + %188 = getelementptr bfloat, ptr addrspace(1) %2, i64 %58, !dbg !23 + %189 = getelementptr bfloat, ptr addrspace(1) %2, i64 %60, !dbg !23 + %190 = getelementptr bfloat, ptr addrspace(1) %2, i64 %62, !dbg !23 + %191 = getelementptr bfloat, ptr addrspace(1) %2, i64 %64, !dbg !23 + %192 = getelementptr bfloat, ptr addrspace(1) %2, i64 %66, !dbg !23 + %193 = getelementptr bfloat, ptr addrspace(1) %2, i64 %68, !dbg !23 + %194 = getelementptr bfloat, ptr addrspace(1) %2, i64 %70, !dbg !23 + %195 = getelementptr bfloat, ptr addrspace(1) %2, i64 %72, !dbg !23 + %196 = getelementptr bfloat, ptr addrspace(1) %2, i64 %74, !dbg !23 + %197 = getelementptr bfloat, ptr addrspace(1) %2, i64 %76, !dbg !23 + %198 = getelementptr bfloat, ptr addrspace(1) %2, i64 %78, !dbg !23 + %199 = getelementptr bfloat, ptr addrspace(1) %2, i64 %80, !dbg !23 + %200 = getelementptr bfloat, ptr addrspace(1) %2, i64 %82, !dbg !23 + %201 = getelementptr bfloat, ptr addrspace(1) %2, i64 %84, !dbg !23 + %202 = getelementptr bfloat, ptr addrspace(1) %2, i64 %86, !dbg !23 + %203 = getelementptr bfloat, ptr addrspace(1) %2, i64 %88, !dbg !23 + %204 = getelementptr bfloat, ptr addrspace(1) %2, i64 %90, !dbg !23 + %205 = getelementptr bfloat, ptr addrspace(1) %2, i64 %92, !dbg !23 + %206 = getelementptr bfloat, ptr addrspace(1) %2, i64 %94, !dbg !23 + %207 = getelementptr bfloat, ptr addrspace(1) %2, i64 %96, !dbg !23 + %208 = getelementptr bfloat, ptr addrspace(1) %2, i64 %98, !dbg !23 + %209 = getelementptr bfloat, ptr addrspace(1) %2, i64 %100, !dbg !23 + %210 = getelementptr bfloat, ptr addrspace(1) %2, i64 %102, !dbg !23 + %211 = getelementptr bfloat, ptr addrspace(1) %2, i64 %104, !dbg !23 + %212 = getelementptr bfloat, ptr addrspace(1) %2, i64 %106, !dbg !23 + %213 = getelementptr bfloat, ptr addrspace(1) %2, i64 %108, !dbg !23 + %214 = getelementptr bfloat, ptr addrspace(1) %2, i64 %110, !dbg !23 + %215 = getelementptr bfloat, ptr addrspace(1) %2, i64 %112, !dbg !23 + %216 = getelementptr bfloat, ptr addrspace(1) %2, i64 %114, !dbg !23 + %217 = getelementptr bfloat, ptr addrspace(1) %2, i64 %116, !dbg !23 + %218 = getelementptr bfloat, ptr addrspace(1) %186, i64 %149, !dbg !24 + %219 = getelementptr bfloat, ptr addrspace(1) %187, i64 %149, !dbg !24 + %220 = getelementptr bfloat, ptr addrspace(1) %188, i64 %149, !dbg !24 + %221 = getelementptr bfloat, ptr addrspace(1) %189, i64 %149, !dbg !24 + %222 = getelementptr bfloat, ptr addrspace(1) %190, i64 %149, !dbg !24 + %223 = getelementptr bfloat, ptr addrspace(1) %191, i64 %149, !dbg !24 + %224 = getelementptr bfloat, ptr addrspace(1) %192, i64 %149, !dbg !24 + %225 = getelementptr bfloat, ptr addrspace(1) %193, i64 %149, !dbg !24 + %226 = getelementptr bfloat, ptr addrspace(1) %194, i64 %149, !dbg !24 + %227 = getelementptr bfloat, ptr addrspace(1) %195, i64 %149, !dbg !24 + %228 = getelementptr bfloat, ptr addrspace(1) %196, i64 %149, !dbg !24 + %229 = getelementptr bfloat, ptr addrspace(1) %197, i64 %149, !dbg !24 + %230 = getelementptr bfloat, ptr addrspace(1) %198, i64 %149, !dbg !24 + %231 = getelementptr bfloat, ptr addrspace(1) %199, i64 %149, !dbg !24 + %232 = getelementptr bfloat, ptr addrspace(1) %200, i64 %149, !dbg !24 + %233 = getelementptr bfloat, ptr addrspace(1) %201, i64 %149, !dbg !24 + %234 = getelementptr bfloat, ptr addrspace(1) %202, i64 %149, !dbg !24 + %235 = getelementptr bfloat, ptr addrspace(1) %203, i64 %149, !dbg !24 + %236 = getelementptr bfloat, ptr addrspace(1) %204, i64 %149, !dbg !24 + %237 = getelementptr bfloat, ptr addrspace(1) %205, i64 %149, !dbg !24 + %238 = getelementptr bfloat, ptr addrspace(1) %206, i64 %149, !dbg !24 + %239 = getelementptr bfloat, ptr addrspace(1) %207, i64 %149, !dbg !24 + %240 = getelementptr bfloat, ptr addrspace(1) %208, i64 %149, !dbg !24 + %241 = getelementptr bfloat, ptr addrspace(1) %209, i64 %149, !dbg !24 + %242 = getelementptr bfloat, ptr addrspace(1) %210, i64 %149, !dbg !24 + %243 = getelementptr bfloat, ptr addrspace(1) %211, i64 %149, !dbg !24 + %244 = getelementptr bfloat, ptr addrspace(1) %212, i64 %149, !dbg !24 + %245 = getelementptr bfloat, ptr addrspace(1) %213, i64 %149, !dbg !24 + %246 = getelementptr bfloat, ptr addrspace(1) %214, i64 %149, !dbg !24 + %247 = getelementptr bfloat, ptr addrspace(1) %215, i64 %149, !dbg !24 + %248 = getelementptr bfloat, ptr addrspace(1) %216, i64 %149, !dbg !24 + %249 = getelementptr bfloat, ptr addrspace(1) %217, i64 %149, !dbg !24 + %250 = getelementptr bfloat, ptr addrspace(1) %3, i64 %55, !dbg !25 + %251 = getelementptr bfloat, ptr addrspace(1) %3, i64 %56, !dbg !25 + %252 = getelementptr bfloat, ptr addrspace(1) %3, i64 %58, !dbg !25 + %253 = getelementptr bfloat, ptr addrspace(1) %3, i64 %60, !dbg !25 + %254 = getelementptr bfloat, ptr addrspace(1) %3, i64 %62, !dbg !25 + %255 = getelementptr bfloat, ptr addrspace(1) %3, i64 %64, !dbg !25 + %256 = getelementptr bfloat, ptr addrspace(1) %3, i64 %66, !dbg !25 + %257 = getelementptr bfloat, ptr addrspace(1) %3, i64 %68, !dbg !25 + %258 = getelementptr bfloat, ptr addrspace(1) %3, i64 %70, !dbg !25 + %259 = getelementptr bfloat, ptr addrspace(1) %3, i64 %72, !dbg !25 + %260 = getelementptr bfloat, ptr addrspace(1) %3, i64 %74, !dbg !25 + %261 = getelementptr bfloat, ptr addrspace(1) %3, i64 %76, !dbg !25 + %262 = getelementptr bfloat, ptr addrspace(1) %3, i64 %78, !dbg !25 + %263 = getelementptr bfloat, ptr addrspace(1) %3, i64 %80, !dbg !25 + %264 = getelementptr bfloat, ptr addrspace(1) %3, i64 %82, !dbg !25 + %265 = getelementptr bfloat, ptr addrspace(1) %3, i64 %84, !dbg !25 + %266 = getelementptr bfloat, ptr addrspace(1) %3, i64 %86, !dbg !25 + %267 = getelementptr bfloat, ptr addrspace(1) %3, i64 %88, !dbg !25 + %268 = getelementptr bfloat, ptr addrspace(1) %3, i64 %90, !dbg !25 + %269 = getelementptr bfloat, ptr addrspace(1) %3, i64 %92, !dbg !25 + %270 = getelementptr bfloat, ptr addrspace(1) %3, i64 %94, !dbg !25 + %271 = getelementptr bfloat, ptr addrspace(1) %3, i64 %96, !dbg !25 + %272 = getelementptr bfloat, ptr addrspace(1) %3, i64 %98, !dbg !25 + %273 = getelementptr bfloat, ptr addrspace(1) %3, i64 %100, !dbg !25 + %274 = getelementptr bfloat, ptr addrspace(1) %3, i64 %102, !dbg !25 + %275 = getelementptr bfloat, ptr addrspace(1) %3, i64 %104, !dbg !25 + %276 = getelementptr bfloat, ptr addrspace(1) %3, i64 %106, !dbg !25 + %277 = getelementptr bfloat, ptr addrspace(1) %3, i64 %108, !dbg !25 + %278 = getelementptr bfloat, ptr addrspace(1) %3, i64 %110, !dbg !25 + %279 = getelementptr bfloat, ptr addrspace(1) %3, i64 %112, !dbg !25 + %280 = getelementptr bfloat, ptr addrspace(1) %3, i64 %114, !dbg !25 + %281 = getelementptr bfloat, ptr addrspace(1) %3, i64 %116, !dbg !25 + %282 = getelementptr bfloat, ptr addrspace(1) %250, i64 %149, !dbg !26 + %283 = getelementptr bfloat, ptr addrspace(1) %251, i64 %149, !dbg !26 + %284 = getelementptr bfloat, ptr addrspace(1) %252, i64 %149, !dbg !26 + %285 = getelementptr bfloat, ptr addrspace(1) %253, i64 %149, !dbg !26 + %286 = getelementptr bfloat, ptr addrspace(1) %254, i64 %149, !dbg !26 + %287 = getelementptr bfloat, ptr addrspace(1) %255, i64 %149, !dbg !26 + %288 = getelementptr bfloat, ptr addrspace(1) %256, i64 %149, !dbg !26 + %289 = getelementptr bfloat, ptr addrspace(1) %257, i64 %149, !dbg !26 + %290 = getelementptr bfloat, ptr addrspace(1) %258, i64 %149, !dbg !26 + %291 = getelementptr bfloat, ptr addrspace(1) %259, i64 %149, !dbg !26 + %292 = getelementptr bfloat, ptr addrspace(1) %260, i64 %149, !dbg !26 + %293 = getelementptr bfloat, ptr addrspace(1) %261, i64 %149, !dbg !26 + %294 = getelementptr bfloat, ptr addrspace(1) %262, i64 %149, !dbg !26 + %295 = getelementptr bfloat, ptr addrspace(1) %263, i64 %149, !dbg !26 + %296 = getelementptr bfloat, ptr addrspace(1) %264, i64 %149, !dbg !26 + %297 = getelementptr bfloat, ptr addrspace(1) %265, i64 %149, !dbg !26 + %298 = getelementptr bfloat, ptr addrspace(1) %266, i64 %149, !dbg !26 + %299 = getelementptr bfloat, ptr addrspace(1) %267, i64 %149, !dbg !26 + %300 = getelementptr bfloat, ptr addrspace(1) %268, i64 %149, !dbg !26 + %301 = getelementptr bfloat, ptr addrspace(1) %269, i64 %149, !dbg !26 + %302 = getelementptr bfloat, ptr addrspace(1) %270, i64 %149, !dbg !26 + %303 = getelementptr bfloat, ptr addrspace(1) %271, i64 %149, !dbg !26 + %304 = getelementptr bfloat, ptr addrspace(1) %272, i64 %149, !dbg !26 + %305 = getelementptr bfloat, ptr addrspace(1) %273, i64 %149, !dbg !26 + %306 = getelementptr bfloat, ptr addrspace(1) %274, i64 %149, !dbg !26 + %307 = getelementptr bfloat, ptr addrspace(1) %275, i64 %149, !dbg !26 + %308 = getelementptr bfloat, ptr addrspace(1) %276, i64 %149, !dbg !26 + %309 = getelementptr bfloat, ptr addrspace(1) %277, i64 %149, !dbg !26 + %310 = getelementptr bfloat, ptr addrspace(1) %278, i64 %149, !dbg !26 + %311 = getelementptr bfloat, ptr addrspace(1) %279, i64 %149, !dbg !26 + %312 = getelementptr bfloat, ptr addrspace(1) %280, i64 %149, !dbg !26 + %313 = getelementptr bfloat, ptr addrspace(1) %281, i64 %149, !dbg !26 + %314 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %150, i1 %22) #4, !dbg !27 + %315 = extractvalue { i32, i32, i32, i32 } %314, 0, !dbg !27 + %316 = bitcast i32 %315 to <2 x bfloat>, !dbg !27 + %317 = extractvalue { i32, i32, i32, i32 } %314, 1, !dbg !27 + %318 = bitcast i32 %317 to <2 x bfloat>, !dbg !27 + %319 = extractvalue { i32, i32, i32, i32 } %314, 2, !dbg !27 + %320 = bitcast i32 %319 to <2 x bfloat>, !dbg !27 + %321 = extractvalue { i32, i32, i32, i32 } %314, 3, !dbg !27 + %322 = bitcast i32 %321 to <2 x bfloat>, !dbg !27 + %323 = extractelement <2 x bfloat> %316, i64 0, !dbg !27 + %324 = extractelement <2 x bfloat> %316, i64 1, !dbg !27 + %325 = extractelement <2 x bfloat> %318, i64 0, !dbg !27 + %326 = extractelement <2 x bfloat> %318, i64 1, !dbg !27 + %327 = extractelement <2 x bfloat> %320, i64 0, !dbg !27 + %328 = extractelement <2 x bfloat> %320, i64 1, !dbg !27 + %329 = extractelement <2 x bfloat> %322, i64 0, !dbg !27 + %330 = extractelement <2 x bfloat> %322, i64 1, !dbg !27 + %331 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %151, i1 %23) #4, !dbg !27 + %332 = extractvalue { i32, i32, i32, i32 } %331, 0, !dbg !27 + %333 = bitcast i32 %332 to <2 x bfloat>, !dbg !27 + %334 = extractvalue { i32, i32, i32, i32 } %331, 1, !dbg !27 + %335 = bitcast i32 %334 to <2 x bfloat>, !dbg !27 + %336 = extractvalue { i32, i32, i32, i32 } %331, 2, !dbg !27 + %337 = bitcast i32 %336 to <2 x bfloat>, !dbg !27 + %338 = extractvalue { i32, i32, i32, i32 } %331, 3, !dbg !27 + %339 = bitcast i32 %338 to <2 x bfloat>, !dbg !27 + %340 = extractelement <2 x bfloat> %333, i64 0, !dbg !27 + %341 = extractelement <2 x bfloat> %333, i64 1, !dbg !27 + %342 = extractelement <2 x bfloat> %335, i64 0, !dbg !27 + %343 = extractelement <2 x bfloat> %335, i64 1, !dbg !27 + %344 = extractelement <2 x bfloat> %337, i64 0, !dbg !27 + %345 = extractelement <2 x bfloat> %337, i64 1, !dbg !27 + %346 = extractelement <2 x bfloat> %339, i64 0, !dbg !27 + %347 = extractelement <2 x bfloat> %339, i64 1, !dbg !27 + %348 = extractelement <32 x i1> %52, i64 31, !dbg !27 + %349 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %152, i1 %348) #4, !dbg !27 + %350 = extractvalue { i32, i32, i32, i32 } %349, 0, !dbg !27 + %351 = bitcast i32 %350 to <2 x bfloat>, !dbg !27 + %352 = extractvalue { i32, i32, i32, i32 } %349, 1, !dbg !27 + %353 = bitcast i32 %352 to <2 x bfloat>, !dbg !27 + %354 = extractvalue { i32, i32, i32, i32 } %349, 2, !dbg !27 + %355 = bitcast i32 %354 to <2 x bfloat>, !dbg !27 + %356 = extractvalue { i32, i32, i32, i32 } %349, 3, !dbg !27 + %357 = bitcast i32 %356 to <2 x bfloat>, !dbg !27 + %358 = extractelement <2 x bfloat> %351, i64 0, !dbg !27 + %359 = extractelement <2 x bfloat> %351, i64 1, !dbg !27 + %360 = extractelement <2 x bfloat> %353, i64 0, !dbg !27 + %361 = extractelement <2 x bfloat> %353, i64 1, !dbg !27 + %362 = extractelement <2 x bfloat> %355, i64 0, !dbg !27 + %363 = extractelement <2 x bfloat> %355, i64 1, !dbg !27 + %364 = extractelement <2 x bfloat> %357, i64 0, !dbg !27 + %365 = extractelement <2 x bfloat> %357, i64 1, !dbg !27 + %366 = extractelement <32 x i1> %52, i64 30, !dbg !27 + %367 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %153, i1 %366) #4, !dbg !27 + %368 = extractvalue { i32, i32, i32, i32 } %367, 0, !dbg !27 + %369 = bitcast i32 %368 to <2 x bfloat>, !dbg !27 + %370 = extractvalue { i32, i32, i32, i32 } %367, 1, !dbg !27 + %371 = bitcast i32 %370 to <2 x bfloat>, !dbg !27 + %372 = extractvalue { i32, i32, i32, i32 } %367, 2, !dbg !27 + %373 = bitcast i32 %372 to <2 x bfloat>, !dbg !27 + %374 = extractvalue { i32, i32, i32, i32 } %367, 3, !dbg !27 + %375 = bitcast i32 %374 to <2 x bfloat>, !dbg !27 + %376 = extractelement <2 x bfloat> %369, i64 0, !dbg !27 + %377 = extractelement <2 x bfloat> %369, i64 1, !dbg !27 + %378 = extractelement <2 x bfloat> %371, i64 0, !dbg !27 + %379 = extractelement <2 x bfloat> %371, i64 1, !dbg !27 + %380 = extractelement <2 x bfloat> %373, i64 0, !dbg !27 + %381 = extractelement <2 x bfloat> %373, i64 1, !dbg !27 + %382 = extractelement <2 x bfloat> %375, i64 0, !dbg !27 + %383 = extractelement <2 x bfloat> %375, i64 1, !dbg !27 + %384 = extractelement <32 x i1> %52, i64 29, !dbg !27 + %385 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %154, i1 %384) #4, !dbg !27 + %386 = extractvalue { i32, i32, i32, i32 } %385, 0, !dbg !27 + %387 = bitcast i32 %386 to <2 x bfloat>, !dbg !27 + %388 = extractvalue { i32, i32, i32, i32 } %385, 1, !dbg !27 + %389 = bitcast i32 %388 to <2 x bfloat>, !dbg !27 + %390 = extractvalue { i32, i32, i32, i32 } %385, 2, !dbg !27 + %391 = bitcast i32 %390 to <2 x bfloat>, !dbg !27 + %392 = extractvalue { i32, i32, i32, i32 } %385, 3, !dbg !27 + %393 = bitcast i32 %392 to <2 x bfloat>, !dbg !27 + %394 = extractelement <2 x bfloat> %387, i64 0, !dbg !27 + %395 = extractelement <2 x bfloat> %387, i64 1, !dbg !27 + %396 = extractelement <2 x bfloat> %389, i64 0, !dbg !27 + %397 = extractelement <2 x bfloat> %389, i64 1, !dbg !27 + %398 = extractelement <2 x bfloat> %391, i64 0, !dbg !27 + %399 = extractelement <2 x bfloat> %391, i64 1, !dbg !27 + %400 = extractelement <2 x bfloat> %393, i64 0, !dbg !27 + %401 = extractelement <2 x bfloat> %393, i64 1, !dbg !27 + %402 = extractelement <32 x i1> %52, i64 28, !dbg !27 + %403 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %155, i1 %402) #4, !dbg !27 + %404 = extractvalue { i32, i32, i32, i32 } %403, 0, !dbg !27 + %405 = bitcast i32 %404 to <2 x bfloat>, !dbg !27 + %406 = extractvalue { i32, i32, i32, i32 } %403, 1, !dbg !27 + %407 = bitcast i32 %406 to <2 x bfloat>, !dbg !27 + %408 = extractvalue { i32, i32, i32, i32 } %403, 2, !dbg !27 + %409 = bitcast i32 %408 to <2 x bfloat>, !dbg !27 + %410 = extractvalue { i32, i32, i32, i32 } %403, 3, !dbg !27 + %411 = bitcast i32 %410 to <2 x bfloat>, !dbg !27 + %412 = extractelement <2 x bfloat> %405, i64 0, !dbg !27 + %413 = extractelement <2 x bfloat> %405, i64 1, !dbg !27 + %414 = extractelement <2 x bfloat> %407, i64 0, !dbg !27 + %415 = extractelement <2 x bfloat> %407, i64 1, !dbg !27 + %416 = extractelement <2 x bfloat> %409, i64 0, !dbg !27 + %417 = extractelement <2 x bfloat> %409, i64 1, !dbg !27 + %418 = extractelement <2 x bfloat> %411, i64 0, !dbg !27 + %419 = extractelement <2 x bfloat> %411, i64 1, !dbg !27 + %420 = extractelement <32 x i1> %52, i64 27, !dbg !27 + %421 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %156, i1 %420) #4, !dbg !27 + %422 = extractvalue { i32, i32, i32, i32 } %421, 0, !dbg !27 + %423 = bitcast i32 %422 to <2 x bfloat>, !dbg !27 + %424 = extractvalue { i32, i32, i32, i32 } %421, 1, !dbg !27 + %425 = bitcast i32 %424 to <2 x bfloat>, !dbg !27 + %426 = extractvalue { i32, i32, i32, i32 } %421, 2, !dbg !27 + %427 = bitcast i32 %426 to <2 x bfloat>, !dbg !27 + %428 = extractvalue { i32, i32, i32, i32 } %421, 3, !dbg !27 + %429 = bitcast i32 %428 to <2 x bfloat>, !dbg !27 + %430 = extractelement <2 x bfloat> %423, i64 0, !dbg !27 + %431 = extractelement <2 x bfloat> %423, i64 1, !dbg !27 + %432 = extractelement <2 x bfloat> %425, i64 0, !dbg !27 + %433 = extractelement <2 x bfloat> %425, i64 1, !dbg !27 + %434 = extractelement <2 x bfloat> %427, i64 0, !dbg !27 + %435 = extractelement <2 x bfloat> %427, i64 1, !dbg !27 + %436 = extractelement <2 x bfloat> %429, i64 0, !dbg !27 + %437 = extractelement <2 x bfloat> %429, i64 1, !dbg !27 + %438 = extractelement <32 x i1> %52, i64 26, !dbg !27 + %439 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %157, i1 %438) #4, !dbg !27 + %440 = extractvalue { i32, i32, i32, i32 } %439, 0, !dbg !27 + %441 = bitcast i32 %440 to <2 x bfloat>, !dbg !27 + %442 = extractvalue { i32, i32, i32, i32 } %439, 1, !dbg !27 + %443 = bitcast i32 %442 to <2 x bfloat>, !dbg !27 + %444 = extractvalue { i32, i32, i32, i32 } %439, 2, !dbg !27 + %445 = bitcast i32 %444 to <2 x bfloat>, !dbg !27 + %446 = extractvalue { i32, i32, i32, i32 } %439, 3, !dbg !27 + %447 = bitcast i32 %446 to <2 x bfloat>, !dbg !27 + %448 = extractelement <2 x bfloat> %441, i64 0, !dbg !27 + %449 = extractelement <2 x bfloat> %441, i64 1, !dbg !27 + %450 = extractelement <2 x bfloat> %443, i64 0, !dbg !27 + %451 = extractelement <2 x bfloat> %443, i64 1, !dbg !27 + %452 = extractelement <2 x bfloat> %445, i64 0, !dbg !27 + %453 = extractelement <2 x bfloat> %445, i64 1, !dbg !27 + %454 = extractelement <2 x bfloat> %447, i64 0, !dbg !27 + %455 = extractelement <2 x bfloat> %447, i64 1, !dbg !27 + %456 = extractelement <32 x i1> %52, i64 25, !dbg !27 + %457 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %158, i1 %456) #4, !dbg !27 + %458 = extractvalue { i32, i32, i32, i32 } %457, 0, !dbg !27 + %459 = bitcast i32 %458 to <2 x bfloat>, !dbg !27 + %460 = extractvalue { i32, i32, i32, i32 } %457, 1, !dbg !27 + %461 = bitcast i32 %460 to <2 x bfloat>, !dbg !27 + %462 = extractvalue { i32, i32, i32, i32 } %457, 2, !dbg !27 + %463 = bitcast i32 %462 to <2 x bfloat>, !dbg !27 + %464 = extractvalue { i32, i32, i32, i32 } %457, 3, !dbg !27 + %465 = bitcast i32 %464 to <2 x bfloat>, !dbg !27 + %466 = extractelement <2 x bfloat> %459, i64 0, !dbg !27 + %467 = extractelement <2 x bfloat> %459, i64 1, !dbg !27 + %468 = extractelement <2 x bfloat> %461, i64 0, !dbg !27 + %469 = extractelement <2 x bfloat> %461, i64 1, !dbg !27 + %470 = extractelement <2 x bfloat> %463, i64 0, !dbg !27 + %471 = extractelement <2 x bfloat> %463, i64 1, !dbg !27 + %472 = extractelement <2 x bfloat> %465, i64 0, !dbg !27 + %473 = extractelement <2 x bfloat> %465, i64 1, !dbg !27 + %474 = extractelement <32 x i1> %52, i64 24, !dbg !27 + %475 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %159, i1 %474) #4, !dbg !27 + %476 = extractvalue { i32, i32, i32, i32 } %475, 0, !dbg !27 + %477 = bitcast i32 %476 to <2 x bfloat>, !dbg !27 + %478 = extractvalue { i32, i32, i32, i32 } %475, 1, !dbg !27 + %479 = bitcast i32 %478 to <2 x bfloat>, !dbg !27 + %480 = extractvalue { i32, i32, i32, i32 } %475, 2, !dbg !27 + %481 = bitcast i32 %480 to <2 x bfloat>, !dbg !27 + %482 = extractvalue { i32, i32, i32, i32 } %475, 3, !dbg !27 + %483 = bitcast i32 %482 to <2 x bfloat>, !dbg !27 + %484 = extractelement <2 x bfloat> %477, i64 0, !dbg !27 + %485 = extractelement <2 x bfloat> %477, i64 1, !dbg !27 + %486 = extractelement <2 x bfloat> %479, i64 0, !dbg !27 + %487 = extractelement <2 x bfloat> %479, i64 1, !dbg !27 + %488 = extractelement <2 x bfloat> %481, i64 0, !dbg !27 + %489 = extractelement <2 x bfloat> %481, i64 1, !dbg !27 + %490 = extractelement <2 x bfloat> %483, i64 0, !dbg !27 + %491 = extractelement <2 x bfloat> %483, i64 1, !dbg !27 + %492 = extractelement <32 x i1> %52, i64 23, !dbg !27 + %493 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %160, i1 %492) #4, !dbg !27 + %494 = extractvalue { i32, i32, i32, i32 } %493, 0, !dbg !27 + %495 = bitcast i32 %494 to <2 x bfloat>, !dbg !27 + %496 = extractvalue { i32, i32, i32, i32 } %493, 1, !dbg !27 + %497 = bitcast i32 %496 to <2 x bfloat>, !dbg !27 + %498 = extractvalue { i32, i32, i32, i32 } %493, 2, !dbg !27 + %499 = bitcast i32 %498 to <2 x bfloat>, !dbg !27 + %500 = extractvalue { i32, i32, i32, i32 } %493, 3, !dbg !27 + %501 = bitcast i32 %500 to <2 x bfloat>, !dbg !27 + %502 = extractelement <2 x bfloat> %495, i64 0, !dbg !27 + %503 = extractelement <2 x bfloat> %495, i64 1, !dbg !27 + %504 = extractelement <2 x bfloat> %497, i64 0, !dbg !27 + %505 = extractelement <2 x bfloat> %497, i64 1, !dbg !27 + %506 = extractelement <2 x bfloat> %499, i64 0, !dbg !27 + %507 = extractelement <2 x bfloat> %499, i64 1, !dbg !27 + %508 = extractelement <2 x bfloat> %501, i64 0, !dbg !27 + %509 = extractelement <2 x bfloat> %501, i64 1, !dbg !27 + %510 = extractelement <32 x i1> %52, i64 22, !dbg !27 + %511 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %161, i1 %510) #4, !dbg !27 + %512 = extractvalue { i32, i32, i32, i32 } %511, 0, !dbg !27 + %513 = bitcast i32 %512 to <2 x bfloat>, !dbg !27 + %514 = extractvalue { i32, i32, i32, i32 } %511, 1, !dbg !27 + %515 = bitcast i32 %514 to <2 x bfloat>, !dbg !27 + %516 = extractvalue { i32, i32, i32, i32 } %511, 2, !dbg !27 + %517 = bitcast i32 %516 to <2 x bfloat>, !dbg !27 + %518 = extractvalue { i32, i32, i32, i32 } %511, 3, !dbg !27 + %519 = bitcast i32 %518 to <2 x bfloat>, !dbg !27 + %520 = extractelement <2 x bfloat> %513, i64 0, !dbg !27 + %521 = extractelement <2 x bfloat> %513, i64 1, !dbg !27 + %522 = extractelement <2 x bfloat> %515, i64 0, !dbg !27 + %523 = extractelement <2 x bfloat> %515, i64 1, !dbg !27 + %524 = extractelement <2 x bfloat> %517, i64 0, !dbg !27 + %525 = extractelement <2 x bfloat> %517, i64 1, !dbg !27 + %526 = extractelement <2 x bfloat> %519, i64 0, !dbg !27 + %527 = extractelement <2 x bfloat> %519, i64 1, !dbg !27 + %528 = extractelement <32 x i1> %52, i64 21, !dbg !27 + %529 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %162, i1 %528) #4, !dbg !27 + %530 = extractvalue { i32, i32, i32, i32 } %529, 0, !dbg !27 + %531 = bitcast i32 %530 to <2 x bfloat>, !dbg !27 + %532 = extractvalue { i32, i32, i32, i32 } %529, 1, !dbg !27 + %533 = bitcast i32 %532 to <2 x bfloat>, !dbg !27 + %534 = extractvalue { i32, i32, i32, i32 } %529, 2, !dbg !27 + %535 = bitcast i32 %534 to <2 x bfloat>, !dbg !27 + %536 = extractvalue { i32, i32, i32, i32 } %529, 3, !dbg !27 + %537 = bitcast i32 %536 to <2 x bfloat>, !dbg !27 + %538 = extractelement <2 x bfloat> %531, i64 0, !dbg !27 + %539 = extractelement <2 x bfloat> %531, i64 1, !dbg !27 + %540 = extractelement <2 x bfloat> %533, i64 0, !dbg !27 + %541 = extractelement <2 x bfloat> %533, i64 1, !dbg !27 + %542 = extractelement <2 x bfloat> %535, i64 0, !dbg !27 + %543 = extractelement <2 x bfloat> %535, i64 1, !dbg !27 + %544 = extractelement <2 x bfloat> %537, i64 0, !dbg !27 + %545 = extractelement <2 x bfloat> %537, i64 1, !dbg !27 + %546 = extractelement <32 x i1> %52, i64 20, !dbg !27 + %547 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %163, i1 %546) #4, !dbg !27 + %548 = extractvalue { i32, i32, i32, i32 } %547, 0, !dbg !27 + %549 = bitcast i32 %548 to <2 x bfloat>, !dbg !27 + %550 = extractvalue { i32, i32, i32, i32 } %547, 1, !dbg !27 + %551 = bitcast i32 %550 to <2 x bfloat>, !dbg !27 + %552 = extractvalue { i32, i32, i32, i32 } %547, 2, !dbg !27 + %553 = bitcast i32 %552 to <2 x bfloat>, !dbg !27 + %554 = extractvalue { i32, i32, i32, i32 } %547, 3, !dbg !27 + %555 = bitcast i32 %554 to <2 x bfloat>, !dbg !27 + %556 = extractelement <2 x bfloat> %549, i64 0, !dbg !27 + %557 = extractelement <2 x bfloat> %549, i64 1, !dbg !27 + %558 = extractelement <2 x bfloat> %551, i64 0, !dbg !27 + %559 = extractelement <2 x bfloat> %551, i64 1, !dbg !27 + %560 = extractelement <2 x bfloat> %553, i64 0, !dbg !27 + %561 = extractelement <2 x bfloat> %553, i64 1, !dbg !27 + %562 = extractelement <2 x bfloat> %555, i64 0, !dbg !27 + %563 = extractelement <2 x bfloat> %555, i64 1, !dbg !27 + %564 = extractelement <32 x i1> %52, i64 19, !dbg !27 + %565 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %164, i1 %564) #4, !dbg !27 + %566 = extractvalue { i32, i32, i32, i32 } %565, 0, !dbg !27 + %567 = bitcast i32 %566 to <2 x bfloat>, !dbg !27 + %568 = extractvalue { i32, i32, i32, i32 } %565, 1, !dbg !27 + %569 = bitcast i32 %568 to <2 x bfloat>, !dbg !27 + %570 = extractvalue { i32, i32, i32, i32 } %565, 2, !dbg !27 + %571 = bitcast i32 %570 to <2 x bfloat>, !dbg !27 + %572 = extractvalue { i32, i32, i32, i32 } %565, 3, !dbg !27 + %573 = bitcast i32 %572 to <2 x bfloat>, !dbg !27 + %574 = extractelement <2 x bfloat> %567, i64 0, !dbg !27 + %575 = extractelement <2 x bfloat> %567, i64 1, !dbg !27 + %576 = extractelement <2 x bfloat> %569, i64 0, !dbg !27 + %577 = extractelement <2 x bfloat> %569, i64 1, !dbg !27 + %578 = extractelement <2 x bfloat> %571, i64 0, !dbg !27 + %579 = extractelement <2 x bfloat> %571, i64 1, !dbg !27 + %580 = extractelement <2 x bfloat> %573, i64 0, !dbg !27 + %581 = extractelement <2 x bfloat> %573, i64 1, !dbg !27 + %582 = extractelement <32 x i1> %52, i64 18, !dbg !27 + %583 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %165, i1 %582) #4, !dbg !27 + %584 = extractvalue { i32, i32, i32, i32 } %583, 0, !dbg !27 + %585 = bitcast i32 %584 to <2 x bfloat>, !dbg !27 + %586 = extractvalue { i32, i32, i32, i32 } %583, 1, !dbg !27 + %587 = bitcast i32 %586 to <2 x bfloat>, !dbg !27 + %588 = extractvalue { i32, i32, i32, i32 } %583, 2, !dbg !27 + %589 = bitcast i32 %588 to <2 x bfloat>, !dbg !27 + %590 = extractvalue { i32, i32, i32, i32 } %583, 3, !dbg !27 + %591 = bitcast i32 %590 to <2 x bfloat>, !dbg !27 + %592 = extractelement <2 x bfloat> %585, i64 0, !dbg !27 + %593 = extractelement <2 x bfloat> %585, i64 1, !dbg !27 + %594 = extractelement <2 x bfloat> %587, i64 0, !dbg !27 + %595 = extractelement <2 x bfloat> %587, i64 1, !dbg !27 + %596 = extractelement <2 x bfloat> %589, i64 0, !dbg !27 + %597 = extractelement <2 x bfloat> %589, i64 1, !dbg !27 + %598 = extractelement <2 x bfloat> %591, i64 0, !dbg !27 + %599 = extractelement <2 x bfloat> %591, i64 1, !dbg !27 + %600 = extractelement <32 x i1> %52, i64 17, !dbg !27 + %601 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %166, i1 %600) #4, !dbg !27 + %602 = extractvalue { i32, i32, i32, i32 } %601, 0, !dbg !27 + %603 = bitcast i32 %602 to <2 x bfloat>, !dbg !27 + %604 = extractvalue { i32, i32, i32, i32 } %601, 1, !dbg !27 + %605 = bitcast i32 %604 to <2 x bfloat>, !dbg !27 + %606 = extractvalue { i32, i32, i32, i32 } %601, 2, !dbg !27 + %607 = bitcast i32 %606 to <2 x bfloat>, !dbg !27 + %608 = extractvalue { i32, i32, i32, i32 } %601, 3, !dbg !27 + %609 = bitcast i32 %608 to <2 x bfloat>, !dbg !27 + %610 = extractelement <2 x bfloat> %603, i64 0, !dbg !27 + %611 = extractelement <2 x bfloat> %603, i64 1, !dbg !27 + %612 = extractelement <2 x bfloat> %605, i64 0, !dbg !27 + %613 = extractelement <2 x bfloat> %605, i64 1, !dbg !27 + %614 = extractelement <2 x bfloat> %607, i64 0, !dbg !27 + %615 = extractelement <2 x bfloat> %607, i64 1, !dbg !27 + %616 = extractelement <2 x bfloat> %609, i64 0, !dbg !27 + %617 = extractelement <2 x bfloat> %609, i64 1, !dbg !27 + %618 = extractelement <32 x i1> %52, i64 16, !dbg !27 + %619 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %167, i1 %618) #4, !dbg !27 + %620 = extractvalue { i32, i32, i32, i32 } %619, 0, !dbg !27 + %621 = bitcast i32 %620 to <2 x bfloat>, !dbg !27 + %622 = extractvalue { i32, i32, i32, i32 } %619, 1, !dbg !27 + %623 = bitcast i32 %622 to <2 x bfloat>, !dbg !27 + %624 = extractvalue { i32, i32, i32, i32 } %619, 2, !dbg !27 + %625 = bitcast i32 %624 to <2 x bfloat>, !dbg !27 + %626 = extractvalue { i32, i32, i32, i32 } %619, 3, !dbg !27 + %627 = bitcast i32 %626 to <2 x bfloat>, !dbg !27 + %628 = extractelement <2 x bfloat> %621, i64 0, !dbg !27 + %629 = extractelement <2 x bfloat> %621, i64 1, !dbg !27 + %630 = extractelement <2 x bfloat> %623, i64 0, !dbg !27 + %631 = extractelement <2 x bfloat> %623, i64 1, !dbg !27 + %632 = extractelement <2 x bfloat> %625, i64 0, !dbg !27 + %633 = extractelement <2 x bfloat> %625, i64 1, !dbg !27 + %634 = extractelement <2 x bfloat> %627, i64 0, !dbg !27 + %635 = extractelement <2 x bfloat> %627, i64 1, !dbg !27 + %636 = extractelement <32 x i1> %52, i64 15, !dbg !27 + %637 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %168, i1 %636) #4, !dbg !27 + %638 = extractvalue { i32, i32, i32, i32 } %637, 0, !dbg !27 + %639 = bitcast i32 %638 to <2 x bfloat>, !dbg !27 + %640 = extractvalue { i32, i32, i32, i32 } %637, 1, !dbg !27 + %641 = bitcast i32 %640 to <2 x bfloat>, !dbg !27 + %642 = extractvalue { i32, i32, i32, i32 } %637, 2, !dbg !27 + %643 = bitcast i32 %642 to <2 x bfloat>, !dbg !27 + %644 = extractvalue { i32, i32, i32, i32 } %637, 3, !dbg !27 + %645 = bitcast i32 %644 to <2 x bfloat>, !dbg !27 + %646 = extractelement <2 x bfloat> %639, i64 0, !dbg !27 + %647 = extractelement <2 x bfloat> %639, i64 1, !dbg !27 + %648 = extractelement <2 x bfloat> %641, i64 0, !dbg !27 + %649 = extractelement <2 x bfloat> %641, i64 1, !dbg !27 + %650 = extractelement <2 x bfloat> %643, i64 0, !dbg !27 + %651 = extractelement <2 x bfloat> %643, i64 1, !dbg !27 + %652 = extractelement <2 x bfloat> %645, i64 0, !dbg !27 + %653 = extractelement <2 x bfloat> %645, i64 1, !dbg !27 + %654 = extractelement <32 x i1> %52, i64 14, !dbg !27 + %655 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %169, i1 %654) #4, !dbg !27 + %656 = extractvalue { i32, i32, i32, i32 } %655, 0, !dbg !27 + %657 = bitcast i32 %656 to <2 x bfloat>, !dbg !27 + %658 = extractvalue { i32, i32, i32, i32 } %655, 1, !dbg !27 + %659 = bitcast i32 %658 to <2 x bfloat>, !dbg !27 + %660 = extractvalue { i32, i32, i32, i32 } %655, 2, !dbg !27 + %661 = bitcast i32 %660 to <2 x bfloat>, !dbg !27 + %662 = extractvalue { i32, i32, i32, i32 } %655, 3, !dbg !27 + %663 = bitcast i32 %662 to <2 x bfloat>, !dbg !27 + %664 = extractelement <2 x bfloat> %657, i64 0, !dbg !27 + %665 = extractelement <2 x bfloat> %657, i64 1, !dbg !27 + %666 = extractelement <2 x bfloat> %659, i64 0, !dbg !27 + %667 = extractelement <2 x bfloat> %659, i64 1, !dbg !27 + %668 = extractelement <2 x bfloat> %661, i64 0, !dbg !27 + %669 = extractelement <2 x bfloat> %661, i64 1, !dbg !27 + %670 = extractelement <2 x bfloat> %663, i64 0, !dbg !27 + %671 = extractelement <2 x bfloat> %663, i64 1, !dbg !27 + %672 = extractelement <32 x i1> %52, i64 13, !dbg !27 + %673 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %170, i1 %672) #4, !dbg !27 + %674 = extractvalue { i32, i32, i32, i32 } %673, 0, !dbg !27 + %675 = bitcast i32 %674 to <2 x bfloat>, !dbg !27 + %676 = extractvalue { i32, i32, i32, i32 } %673, 1, !dbg !27 + %677 = bitcast i32 %676 to <2 x bfloat>, !dbg !27 + %678 = extractvalue { i32, i32, i32, i32 } %673, 2, !dbg !27 + %679 = bitcast i32 %678 to <2 x bfloat>, !dbg !27 + %680 = extractvalue { i32, i32, i32, i32 } %673, 3, !dbg !27 + %681 = bitcast i32 %680 to <2 x bfloat>, !dbg !27 + %682 = extractelement <2 x bfloat> %675, i64 0, !dbg !27 + %683 = extractelement <2 x bfloat> %675, i64 1, !dbg !27 + %684 = extractelement <2 x bfloat> %677, i64 0, !dbg !27 + %685 = extractelement <2 x bfloat> %677, i64 1, !dbg !27 + %686 = extractelement <2 x bfloat> %679, i64 0, !dbg !27 + %687 = extractelement <2 x bfloat> %679, i64 1, !dbg !27 + %688 = extractelement <2 x bfloat> %681, i64 0, !dbg !27 + %689 = extractelement <2 x bfloat> %681, i64 1, !dbg !27 + %690 = extractelement <32 x i1> %52, i64 12, !dbg !27 + %691 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %171, i1 %690) #4, !dbg !27 + %692 = extractvalue { i32, i32, i32, i32 } %691, 0, !dbg !27 + %693 = bitcast i32 %692 to <2 x bfloat>, !dbg !27 + %694 = extractvalue { i32, i32, i32, i32 } %691, 1, !dbg !27 + %695 = bitcast i32 %694 to <2 x bfloat>, !dbg !27 + %696 = extractvalue { i32, i32, i32, i32 } %691, 2, !dbg !27 + %697 = bitcast i32 %696 to <2 x bfloat>, !dbg !27 + %698 = extractvalue { i32, i32, i32, i32 } %691, 3, !dbg !27 + %699 = bitcast i32 %698 to <2 x bfloat>, !dbg !27 + %700 = extractelement <2 x bfloat> %693, i64 0, !dbg !27 + %701 = extractelement <2 x bfloat> %693, i64 1, !dbg !27 + %702 = extractelement <2 x bfloat> %695, i64 0, !dbg !27 + %703 = extractelement <2 x bfloat> %695, i64 1, !dbg !27 + %704 = extractelement <2 x bfloat> %697, i64 0, !dbg !27 + %705 = extractelement <2 x bfloat> %697, i64 1, !dbg !27 + %706 = extractelement <2 x bfloat> %699, i64 0, !dbg !27 + %707 = extractelement <2 x bfloat> %699, i64 1, !dbg !27 + %708 = extractelement <32 x i1> %52, i64 11, !dbg !27 + %709 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %172, i1 %708) #4, !dbg !27 + %710 = extractvalue { i32, i32, i32, i32 } %709, 0, !dbg !27 + %711 = bitcast i32 %710 to <2 x bfloat>, !dbg !27 + %712 = extractvalue { i32, i32, i32, i32 } %709, 1, !dbg !27 + %713 = bitcast i32 %712 to <2 x bfloat>, !dbg !27 + %714 = extractvalue { i32, i32, i32, i32 } %709, 2, !dbg !27 + %715 = bitcast i32 %714 to <2 x bfloat>, !dbg !27 + %716 = extractvalue { i32, i32, i32, i32 } %709, 3, !dbg !27 + %717 = bitcast i32 %716 to <2 x bfloat>, !dbg !27 + %718 = extractelement <2 x bfloat> %711, i64 0, !dbg !27 + %719 = extractelement <2 x bfloat> %711, i64 1, !dbg !27 + %720 = extractelement <2 x bfloat> %713, i64 0, !dbg !27 + %721 = extractelement <2 x bfloat> %713, i64 1, !dbg !27 + %722 = extractelement <2 x bfloat> %715, i64 0, !dbg !27 + %723 = extractelement <2 x bfloat> %715, i64 1, !dbg !27 + %724 = extractelement <2 x bfloat> %717, i64 0, !dbg !27 + %725 = extractelement <2 x bfloat> %717, i64 1, !dbg !27 + %726 = extractelement <32 x i1> %52, i64 10, !dbg !27 + %727 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %173, i1 %726) #4, !dbg !27 + %728 = extractvalue { i32, i32, i32, i32 } %727, 0, !dbg !27 + %729 = bitcast i32 %728 to <2 x bfloat>, !dbg !27 + %730 = extractvalue { i32, i32, i32, i32 } %727, 1, !dbg !27 + %731 = bitcast i32 %730 to <2 x bfloat>, !dbg !27 + %732 = extractvalue { i32, i32, i32, i32 } %727, 2, !dbg !27 + %733 = bitcast i32 %732 to <2 x bfloat>, !dbg !27 + %734 = extractvalue { i32, i32, i32, i32 } %727, 3, !dbg !27 + %735 = bitcast i32 %734 to <2 x bfloat>, !dbg !27 + %736 = extractelement <2 x bfloat> %729, i64 0, !dbg !27 + %737 = extractelement <2 x bfloat> %729, i64 1, !dbg !27 + %738 = extractelement <2 x bfloat> %731, i64 0, !dbg !27 + %739 = extractelement <2 x bfloat> %731, i64 1, !dbg !27 + %740 = extractelement <2 x bfloat> %733, i64 0, !dbg !27 + %741 = extractelement <2 x bfloat> %733, i64 1, !dbg !27 + %742 = extractelement <2 x bfloat> %735, i64 0, !dbg !27 + %743 = extractelement <2 x bfloat> %735, i64 1, !dbg !27 + %744 = extractelement <32 x i1> %52, i64 9, !dbg !27 + %745 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %174, i1 %744) #4, !dbg !27 + %746 = extractvalue { i32, i32, i32, i32 } %745, 0, !dbg !27 + %747 = bitcast i32 %746 to <2 x bfloat>, !dbg !27 + %748 = extractvalue { i32, i32, i32, i32 } %745, 1, !dbg !27 + %749 = bitcast i32 %748 to <2 x bfloat>, !dbg !27 + %750 = extractvalue { i32, i32, i32, i32 } %745, 2, !dbg !27 + %751 = bitcast i32 %750 to <2 x bfloat>, !dbg !27 + %752 = extractvalue { i32, i32, i32, i32 } %745, 3, !dbg !27 + %753 = bitcast i32 %752 to <2 x bfloat>, !dbg !27 + %754 = extractelement <2 x bfloat> %747, i64 0, !dbg !27 + %755 = extractelement <2 x bfloat> %747, i64 1, !dbg !27 + %756 = extractelement <2 x bfloat> %749, i64 0, !dbg !27 + %757 = extractelement <2 x bfloat> %749, i64 1, !dbg !27 + %758 = extractelement <2 x bfloat> %751, i64 0, !dbg !27 + %759 = extractelement <2 x bfloat> %751, i64 1, !dbg !27 + %760 = extractelement <2 x bfloat> %753, i64 0, !dbg !27 + %761 = extractelement <2 x bfloat> %753, i64 1, !dbg !27 + %762 = extractelement <32 x i1> %52, i64 8, !dbg !27 + %763 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %175, i1 %762) #4, !dbg !27 + %764 = extractvalue { i32, i32, i32, i32 } %763, 0, !dbg !27 + %765 = bitcast i32 %764 to <2 x bfloat>, !dbg !27 + %766 = extractvalue { i32, i32, i32, i32 } %763, 1, !dbg !27 + %767 = bitcast i32 %766 to <2 x bfloat>, !dbg !27 + %768 = extractvalue { i32, i32, i32, i32 } %763, 2, !dbg !27 + %769 = bitcast i32 %768 to <2 x bfloat>, !dbg !27 + %770 = extractvalue { i32, i32, i32, i32 } %763, 3, !dbg !27 + %771 = bitcast i32 %770 to <2 x bfloat>, !dbg !27 + %772 = extractelement <2 x bfloat> %765, i64 0, !dbg !27 + %773 = extractelement <2 x bfloat> %765, i64 1, !dbg !27 + %774 = extractelement <2 x bfloat> %767, i64 0, !dbg !27 + %775 = extractelement <2 x bfloat> %767, i64 1, !dbg !27 + %776 = extractelement <2 x bfloat> %769, i64 0, !dbg !27 + %777 = extractelement <2 x bfloat> %769, i64 1, !dbg !27 + %778 = extractelement <2 x bfloat> %771, i64 0, !dbg !27 + %779 = extractelement <2 x bfloat> %771, i64 1, !dbg !27 + %780 = extractelement <32 x i1> %52, i64 7, !dbg !27 + %781 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %176, i1 %780) #4, !dbg !27 + %782 = extractvalue { i32, i32, i32, i32 } %781, 0, !dbg !27 + %783 = bitcast i32 %782 to <2 x bfloat>, !dbg !27 + %784 = extractvalue { i32, i32, i32, i32 } %781, 1, !dbg !27 + %785 = bitcast i32 %784 to <2 x bfloat>, !dbg !27 + %786 = extractvalue { i32, i32, i32, i32 } %781, 2, !dbg !27 + %787 = bitcast i32 %786 to <2 x bfloat>, !dbg !27 + %788 = extractvalue { i32, i32, i32, i32 } %781, 3, !dbg !27 + %789 = bitcast i32 %788 to <2 x bfloat>, !dbg !27 + %790 = extractelement <2 x bfloat> %783, i64 0, !dbg !27 + %791 = extractelement <2 x bfloat> %783, i64 1, !dbg !27 + %792 = extractelement <2 x bfloat> %785, i64 0, !dbg !27 + %793 = extractelement <2 x bfloat> %785, i64 1, !dbg !27 + %794 = extractelement <2 x bfloat> %787, i64 0, !dbg !27 + %795 = extractelement <2 x bfloat> %787, i64 1, !dbg !27 + %796 = extractelement <2 x bfloat> %789, i64 0, !dbg !27 + %797 = extractelement <2 x bfloat> %789, i64 1, !dbg !27 + %798 = extractelement <32 x i1> %52, i64 6, !dbg !27 + %799 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %177, i1 %798) #4, !dbg !27 + %800 = extractvalue { i32, i32, i32, i32 } %799, 0, !dbg !27 + %801 = bitcast i32 %800 to <2 x bfloat>, !dbg !27 + %802 = extractvalue { i32, i32, i32, i32 } %799, 1, !dbg !27 + %803 = bitcast i32 %802 to <2 x bfloat>, !dbg !27 + %804 = extractvalue { i32, i32, i32, i32 } %799, 2, !dbg !27 + %805 = bitcast i32 %804 to <2 x bfloat>, !dbg !27 + %806 = extractvalue { i32, i32, i32, i32 } %799, 3, !dbg !27 + %807 = bitcast i32 %806 to <2 x bfloat>, !dbg !27 + %808 = extractelement <2 x bfloat> %801, i64 0, !dbg !27 + %809 = extractelement <2 x bfloat> %801, i64 1, !dbg !27 + %810 = extractelement <2 x bfloat> %803, i64 0, !dbg !27 + %811 = extractelement <2 x bfloat> %803, i64 1, !dbg !27 + %812 = extractelement <2 x bfloat> %805, i64 0, !dbg !27 + %813 = extractelement <2 x bfloat> %805, i64 1, !dbg !27 + %814 = extractelement <2 x bfloat> %807, i64 0, !dbg !27 + %815 = extractelement <2 x bfloat> %807, i64 1, !dbg !27 + %816 = extractelement <32 x i1> %52, i64 5, !dbg !27 + %817 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %178, i1 %816) #4, !dbg !27 + %818 = extractvalue { i32, i32, i32, i32 } %817, 0, !dbg !27 + %819 = bitcast i32 %818 to <2 x bfloat>, !dbg !27 + %820 = extractvalue { i32, i32, i32, i32 } %817, 1, !dbg !27 + %821 = bitcast i32 %820 to <2 x bfloat>, !dbg !27 + %822 = extractvalue { i32, i32, i32, i32 } %817, 2, !dbg !27 + %823 = bitcast i32 %822 to <2 x bfloat>, !dbg !27 + %824 = extractvalue { i32, i32, i32, i32 } %817, 3, !dbg !27 + %825 = bitcast i32 %824 to <2 x bfloat>, !dbg !27 + %826 = extractelement <2 x bfloat> %819, i64 0, !dbg !27 + %827 = extractelement <2 x bfloat> %819, i64 1, !dbg !27 + %828 = extractelement <2 x bfloat> %821, i64 0, !dbg !27 + %829 = extractelement <2 x bfloat> %821, i64 1, !dbg !27 + %830 = extractelement <2 x bfloat> %823, i64 0, !dbg !27 + %831 = extractelement <2 x bfloat> %823, i64 1, !dbg !27 + %832 = extractelement <2 x bfloat> %825, i64 0, !dbg !27 + %833 = extractelement <2 x bfloat> %825, i64 1, !dbg !27 + %834 = extractelement <32 x i1> %52, i64 4, !dbg !27 + %835 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %179, i1 %834) #4, !dbg !27 + %836 = extractvalue { i32, i32, i32, i32 } %835, 0, !dbg !27 + %837 = bitcast i32 %836 to <2 x bfloat>, !dbg !27 + %838 = extractvalue { i32, i32, i32, i32 } %835, 1, !dbg !27 + %839 = bitcast i32 %838 to <2 x bfloat>, !dbg !27 + %840 = extractvalue { i32, i32, i32, i32 } %835, 2, !dbg !27 + %841 = bitcast i32 %840 to <2 x bfloat>, !dbg !27 + %842 = extractvalue { i32, i32, i32, i32 } %835, 3, !dbg !27 + %843 = bitcast i32 %842 to <2 x bfloat>, !dbg !27 + %844 = extractelement <2 x bfloat> %837, i64 0, !dbg !27 + %845 = extractelement <2 x bfloat> %837, i64 1, !dbg !27 + %846 = extractelement <2 x bfloat> %839, i64 0, !dbg !27 + %847 = extractelement <2 x bfloat> %839, i64 1, !dbg !27 + %848 = extractelement <2 x bfloat> %841, i64 0, !dbg !27 + %849 = extractelement <2 x bfloat> %841, i64 1, !dbg !27 + %850 = extractelement <2 x bfloat> %843, i64 0, !dbg !27 + %851 = extractelement <2 x bfloat> %843, i64 1, !dbg !27 + %852 = extractelement <32 x i1> %52, i64 3, !dbg !27 + %853 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %180, i1 %852) #4, !dbg !27 + %854 = extractvalue { i32, i32, i32, i32 } %853, 0, !dbg !27 + %855 = bitcast i32 %854 to <2 x bfloat>, !dbg !27 + %856 = extractvalue { i32, i32, i32, i32 } %853, 1, !dbg !27 + %857 = bitcast i32 %856 to <2 x bfloat>, !dbg !27 + %858 = extractvalue { i32, i32, i32, i32 } %853, 2, !dbg !27 + %859 = bitcast i32 %858 to <2 x bfloat>, !dbg !27 + %860 = extractvalue { i32, i32, i32, i32 } %853, 3, !dbg !27 + %861 = bitcast i32 %860 to <2 x bfloat>, !dbg !27 + %862 = extractelement <2 x bfloat> %855, i64 0, !dbg !27 + %863 = extractelement <2 x bfloat> %855, i64 1, !dbg !27 + %864 = extractelement <2 x bfloat> %857, i64 0, !dbg !27 + %865 = extractelement <2 x bfloat> %857, i64 1, !dbg !27 + %866 = extractelement <2 x bfloat> %859, i64 0, !dbg !27 + %867 = extractelement <2 x bfloat> %859, i64 1, !dbg !27 + %868 = extractelement <2 x bfloat> %861, i64 0, !dbg !27 + %869 = extractelement <2 x bfloat> %861, i64 1, !dbg !27 + %870 = extractelement <32 x i1> %52, i64 2, !dbg !27 + %871 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %181, i1 %870) #4, !dbg !27 + %872 = extractvalue { i32, i32, i32, i32 } %871, 0, !dbg !27 + %873 = bitcast i32 %872 to <2 x bfloat>, !dbg !27 + %874 = extractvalue { i32, i32, i32, i32 } %871, 1, !dbg !27 + %875 = bitcast i32 %874 to <2 x bfloat>, !dbg !27 + %876 = extractvalue { i32, i32, i32, i32 } %871, 2, !dbg !27 + %877 = bitcast i32 %876 to <2 x bfloat>, !dbg !27 + %878 = extractvalue { i32, i32, i32, i32 } %871, 3, !dbg !27 + %879 = bitcast i32 %878 to <2 x bfloat>, !dbg !27 + %880 = extractelement <2 x bfloat> %873, i64 0, !dbg !27 + %881 = extractelement <2 x bfloat> %873, i64 1, !dbg !27 + %882 = extractelement <2 x bfloat> %875, i64 0, !dbg !27 + %883 = extractelement <2 x bfloat> %875, i64 1, !dbg !27 + %884 = extractelement <2 x bfloat> %877, i64 0, !dbg !27 + %885 = extractelement <2 x bfloat> %877, i64 1, !dbg !27 + %886 = extractelement <2 x bfloat> %879, i64 0, !dbg !27 + %887 = extractelement <2 x bfloat> %879, i64 1, !dbg !27 + %888 = fpext bfloat %323 to float, !dbg !28 + %889 = fpext bfloat %324 to float, !dbg !28 + %890 = fpext bfloat %325 to float, !dbg !28 + %891 = fpext bfloat %326 to float, !dbg !28 + %892 = fpext bfloat %327 to float, !dbg !28 + %893 = fpext bfloat %328 to float, !dbg !28 + %894 = fpext bfloat %329 to float, !dbg !28 + %895 = fpext bfloat %330 to float, !dbg !28 + %896 = fpext bfloat %340 to float, !dbg !28 + %897 = fpext bfloat %341 to float, !dbg !28 + %898 = fpext bfloat %342 to float, !dbg !28 + %899 = fpext bfloat %343 to float, !dbg !28 + %900 = fpext bfloat %344 to float, !dbg !28 + %901 = fpext bfloat %345 to float, !dbg !28 + %902 = fpext bfloat %346 to float, !dbg !28 + %903 = fpext bfloat %347 to float, !dbg !28 + %904 = fpext bfloat %358 to float, !dbg !28 + %905 = fpext bfloat %359 to float, !dbg !28 + %906 = fpext bfloat %360 to float, !dbg !28 + %907 = fpext bfloat %361 to float, !dbg !28 + %908 = fpext bfloat %362 to float, !dbg !28 + %909 = fpext bfloat %363 to float, !dbg !28 + %910 = fpext bfloat %364 to float, !dbg !28 + %911 = fpext bfloat %365 to float, !dbg !28 + %912 = fpext bfloat %376 to float, !dbg !28 + %913 = fpext bfloat %377 to float, !dbg !28 + %914 = fpext bfloat %378 to float, !dbg !28 + %915 = fpext bfloat %379 to float, !dbg !28 + %916 = fpext bfloat %380 to float, !dbg !28 + %917 = fpext bfloat %381 to float, !dbg !28 + %918 = fpext bfloat %382 to float, !dbg !28 + %919 = fpext bfloat %383 to float, !dbg !28 + %920 = fpext bfloat %394 to float, !dbg !28 + %921 = fpext bfloat %395 to float, !dbg !28 + %922 = fpext bfloat %396 to float, !dbg !28 + %923 = fpext bfloat %397 to float, !dbg !28 + %924 = fpext bfloat %398 to float, !dbg !28 + %925 = fpext bfloat %399 to float, !dbg !28 + %926 = fpext bfloat %400 to float, !dbg !28 + %927 = fpext bfloat %401 to float, !dbg !28 + %928 = fpext bfloat %412 to float, !dbg !28 + %929 = fpext bfloat %413 to float, !dbg !28 + %930 = fpext bfloat %414 to float, !dbg !28 + %931 = fpext bfloat %415 to float, !dbg !28 + %932 = fpext bfloat %416 to float, !dbg !28 + %933 = fpext bfloat %417 to float, !dbg !28 + %934 = fpext bfloat %418 to float, !dbg !28 + %935 = fpext bfloat %419 to float, !dbg !28 + %936 = fpext bfloat %430 to float, !dbg !28 + %937 = fpext bfloat %431 to float, !dbg !28 + %938 = fpext bfloat %432 to float, !dbg !28 + %939 = fpext bfloat %433 to float, !dbg !28 + %940 = fpext bfloat %434 to float, !dbg !28 + %941 = fpext bfloat %435 to float, !dbg !28 + %942 = fpext bfloat %436 to float, !dbg !28 + %943 = fpext bfloat %437 to float, !dbg !28 + %944 = fpext bfloat %448 to float, !dbg !28 + %945 = fpext bfloat %449 to float, !dbg !28 + %946 = fpext bfloat %450 to float, !dbg !28 + %947 = fpext bfloat %451 to float, !dbg !28 + %948 = fpext bfloat %452 to float, !dbg !28 + %949 = fpext bfloat %453 to float, !dbg !28 + %950 = fpext bfloat %454 to float, !dbg !28 + %951 = fpext bfloat %455 to float, !dbg !28 + %952 = fpext bfloat %466 to float, !dbg !28 + %953 = fpext bfloat %467 to float, !dbg !28 + %954 = fpext bfloat %468 to float, !dbg !28 + %955 = fpext bfloat %469 to float, !dbg !28 + %956 = fpext bfloat %470 to float, !dbg !28 + %957 = fpext bfloat %471 to float, !dbg !28 + %958 = fpext bfloat %472 to float, !dbg !28 + %959 = fpext bfloat %473 to float, !dbg !28 + %960 = fpext bfloat %484 to float, !dbg !28 + %961 = fpext bfloat %485 to float, !dbg !28 + %962 = fpext bfloat %486 to float, !dbg !28 + %963 = fpext bfloat %487 to float, !dbg !28 + %964 = fpext bfloat %488 to float, !dbg !28 + %965 = fpext bfloat %489 to float, !dbg !28 + %966 = fpext bfloat %490 to float, !dbg !28 + %967 = fpext bfloat %491 to float, !dbg !28 + %968 = fpext bfloat %502 to float, !dbg !28 + %969 = fpext bfloat %503 to float, !dbg !28 + %970 = fpext bfloat %504 to float, !dbg !28 + %971 = fpext bfloat %505 to float, !dbg !28 + %972 = fpext bfloat %506 to float, !dbg !28 + %973 = fpext bfloat %507 to float, !dbg !28 + %974 = fpext bfloat %508 to float, !dbg !28 + %975 = fpext bfloat %509 to float, !dbg !28 + %976 = fpext bfloat %520 to float, !dbg !28 + %977 = fpext bfloat %521 to float, !dbg !28 + %978 = fpext bfloat %522 to float, !dbg !28 + %979 = fpext bfloat %523 to float, !dbg !28 + %980 = fpext bfloat %524 to float, !dbg !28 + %981 = fpext bfloat %525 to float, !dbg !28 + %982 = fpext bfloat %526 to float, !dbg !28 + %983 = fpext bfloat %527 to float, !dbg !28 + %984 = fpext bfloat %538 to float, !dbg !28 + %985 = fpext bfloat %539 to float, !dbg !28 + %986 = fpext bfloat %540 to float, !dbg !28 + %987 = fpext bfloat %541 to float, !dbg !28 + %988 = fpext bfloat %542 to float, !dbg !28 + %989 = fpext bfloat %543 to float, !dbg !28 + %990 = fpext bfloat %544 to float, !dbg !28 + %991 = fpext bfloat %545 to float, !dbg !28 + %992 = fpext bfloat %556 to float, !dbg !28 + %993 = fpext bfloat %557 to float, !dbg !28 + %994 = fpext bfloat %558 to float, !dbg !28 + %995 = fpext bfloat %559 to float, !dbg !28 + %996 = fpext bfloat %560 to float, !dbg !28 + %997 = fpext bfloat %561 to float, !dbg !28 + %998 = fpext bfloat %562 to float, !dbg !28 + %999 = fpext bfloat %563 to float, !dbg !28 + %1000 = fpext bfloat %574 to float, !dbg !28 + %1001 = fpext bfloat %575 to float, !dbg !28 + %1002 = fpext bfloat %576 to float, !dbg !28 + %1003 = fpext bfloat %577 to float, !dbg !28 + %1004 = fpext bfloat %578 to float, !dbg !28 + %1005 = fpext bfloat %579 to float, !dbg !28 + %1006 = fpext bfloat %580 to float, !dbg !28 + %1007 = fpext bfloat %581 to float, !dbg !28 + %1008 = fpext bfloat %592 to float, !dbg !28 + %1009 = fpext bfloat %593 to float, !dbg !28 + %1010 = fpext bfloat %594 to float, !dbg !28 + %1011 = fpext bfloat %595 to float, !dbg !28 + %1012 = fpext bfloat %596 to float, !dbg !28 + %1013 = fpext bfloat %597 to float, !dbg !28 + %1014 = fpext bfloat %598 to float, !dbg !28 + %1015 = fpext bfloat %599 to float, !dbg !28 + %1016 = fpext bfloat %610 to float, !dbg !28 + %1017 = fpext bfloat %611 to float, !dbg !28 + %1018 = fpext bfloat %612 to float, !dbg !28 + %1019 = fpext bfloat %613 to float, !dbg !28 + %1020 = fpext bfloat %614 to float, !dbg !28 + %1021 = fpext bfloat %615 to float, !dbg !28 + %1022 = fpext bfloat %616 to float, !dbg !28 + %1023 = fpext bfloat %617 to float, !dbg !28 + %1024 = fpext bfloat %628 to float, !dbg !28 + %1025 = fpext bfloat %629 to float, !dbg !28 + %1026 = fpext bfloat %630 to float, !dbg !28 + %1027 = fpext bfloat %631 to float, !dbg !28 + %1028 = fpext bfloat %632 to float, !dbg !28 + %1029 = fpext bfloat %633 to float, !dbg !28 + %1030 = fpext bfloat %634 to float, !dbg !28 + %1031 = fpext bfloat %635 to float, !dbg !28 + %1032 = fpext bfloat %646 to float, !dbg !28 + %1033 = fpext bfloat %647 to float, !dbg !28 + %1034 = fpext bfloat %648 to float, !dbg !28 + %1035 = fpext bfloat %649 to float, !dbg !28 + %1036 = fpext bfloat %650 to float, !dbg !28 + %1037 = fpext bfloat %651 to float, !dbg !28 + %1038 = fpext bfloat %652 to float, !dbg !28 + %1039 = fpext bfloat %653 to float, !dbg !28 + %1040 = fpext bfloat %664 to float, !dbg !28 + %1041 = fpext bfloat %665 to float, !dbg !28 + %1042 = fpext bfloat %666 to float, !dbg !28 + %1043 = fpext bfloat %667 to float, !dbg !28 + %1044 = fpext bfloat %668 to float, !dbg !28 + %1045 = fpext bfloat %669 to float, !dbg !28 + %1046 = fpext bfloat %670 to float, !dbg !28 + %1047 = fpext bfloat %671 to float, !dbg !28 + %1048 = fpext bfloat %682 to float, !dbg !28 + %1049 = fpext bfloat %683 to float, !dbg !28 + %1050 = fpext bfloat %684 to float, !dbg !28 + %1051 = fpext bfloat %685 to float, !dbg !28 + %1052 = fpext bfloat %686 to float, !dbg !28 + %1053 = fpext bfloat %687 to float, !dbg !28 + %1054 = fpext bfloat %688 to float, !dbg !28 + %1055 = fpext bfloat %689 to float, !dbg !28 + %1056 = fpext bfloat %700 to float, !dbg !28 + %1057 = fpext bfloat %701 to float, !dbg !28 + %1058 = fpext bfloat %702 to float, !dbg !28 + %1059 = fpext bfloat %703 to float, !dbg !28 + %1060 = fpext bfloat %704 to float, !dbg !28 + %1061 = fpext bfloat %705 to float, !dbg !28 + %1062 = fpext bfloat %706 to float, !dbg !28 + %1063 = fpext bfloat %707 to float, !dbg !28 + %1064 = fpext bfloat %718 to float, !dbg !28 + %1065 = fpext bfloat %719 to float, !dbg !28 + %1066 = fpext bfloat %720 to float, !dbg !28 + %1067 = fpext bfloat %721 to float, !dbg !28 + %1068 = fpext bfloat %722 to float, !dbg !28 + %1069 = fpext bfloat %723 to float, !dbg !28 + %1070 = fpext bfloat %724 to float, !dbg !28 + %1071 = fpext bfloat %725 to float, !dbg !28 + %1072 = fpext bfloat %736 to float, !dbg !28 + %1073 = fpext bfloat %737 to float, !dbg !28 + %1074 = fpext bfloat %738 to float, !dbg !28 + %1075 = fpext bfloat %739 to float, !dbg !28 + %1076 = fpext bfloat %740 to float, !dbg !28 + %1077 = fpext bfloat %741 to float, !dbg !28 + %1078 = fpext bfloat %742 to float, !dbg !28 + %1079 = fpext bfloat %743 to float, !dbg !28 + %1080 = fpext bfloat %754 to float, !dbg !28 + %1081 = fpext bfloat %755 to float, !dbg !28 + %1082 = fpext bfloat %756 to float, !dbg !28 + %1083 = fpext bfloat %757 to float, !dbg !28 + %1084 = fpext bfloat %758 to float, !dbg !28 + %1085 = fpext bfloat %759 to float, !dbg !28 + %1086 = fpext bfloat %760 to float, !dbg !28 + %1087 = fpext bfloat %761 to float, !dbg !28 + %1088 = fpext bfloat %772 to float, !dbg !28 + %1089 = fpext bfloat %773 to float, !dbg !28 + %1090 = fpext bfloat %774 to float, !dbg !28 + %1091 = fpext bfloat %775 to float, !dbg !28 + %1092 = fpext bfloat %776 to float, !dbg !28 + %1093 = fpext bfloat %777 to float, !dbg !28 + %1094 = fpext bfloat %778 to float, !dbg !28 + %1095 = fpext bfloat %779 to float, !dbg !28 + %1096 = fpext bfloat %790 to float, !dbg !28 + %1097 = fpext bfloat %791 to float, !dbg !28 + %1098 = fpext bfloat %792 to float, !dbg !28 + %1099 = fpext bfloat %793 to float, !dbg !28 + %1100 = fpext bfloat %794 to float, !dbg !28 + %1101 = fpext bfloat %795 to float, !dbg !28 + %1102 = fpext bfloat %796 to float, !dbg !28 + %1103 = fpext bfloat %797 to float, !dbg !28 + %1104 = fpext bfloat %808 to float, !dbg !28 + %1105 = fpext bfloat %809 to float, !dbg !28 + %1106 = fpext bfloat %810 to float, !dbg !28 + %1107 = fpext bfloat %811 to float, !dbg !28 + %1108 = fpext bfloat %812 to float, !dbg !28 + %1109 = fpext bfloat %813 to float, !dbg !28 + %1110 = fpext bfloat %814 to float, !dbg !28 + %1111 = fpext bfloat %815 to float, !dbg !28 + %1112 = fpext bfloat %826 to float, !dbg !28 + %1113 = fpext bfloat %827 to float, !dbg !28 + %1114 = fpext bfloat %828 to float, !dbg !28 + %1115 = fpext bfloat %829 to float, !dbg !28 + %1116 = fpext bfloat %830 to float, !dbg !28 + %1117 = fpext bfloat %831 to float, !dbg !28 + %1118 = fpext bfloat %832 to float, !dbg !28 + %1119 = fpext bfloat %833 to float, !dbg !28 + %1120 = fpext bfloat %844 to float, !dbg !28 + %1121 = fpext bfloat %845 to float, !dbg !28 + %1122 = fpext bfloat %846 to float, !dbg !28 + %1123 = fpext bfloat %847 to float, !dbg !28 + %1124 = fpext bfloat %848 to float, !dbg !28 + %1125 = fpext bfloat %849 to float, !dbg !28 + %1126 = fpext bfloat %850 to float, !dbg !28 + %1127 = fpext bfloat %851 to float, !dbg !28 + %1128 = fpext bfloat %862 to float, !dbg !28 + %1129 = fpext bfloat %863 to float, !dbg !28 + %1130 = fpext bfloat %864 to float, !dbg !28 + %1131 = fpext bfloat %865 to float, !dbg !28 + %1132 = fpext bfloat %866 to float, !dbg !28 + %1133 = fpext bfloat %867 to float, !dbg !28 + %1134 = fpext bfloat %868 to float, !dbg !28 + %1135 = fpext bfloat %869 to float, !dbg !28 + %1136 = fpext bfloat %880 to float, !dbg !28 + %1137 = fpext bfloat %881 to float, !dbg !28 + %1138 = fpext bfloat %882 to float, !dbg !28 + %1139 = fpext bfloat %883 to float, !dbg !28 + %1140 = fpext bfloat %884 to float, !dbg !28 + %1141 = fpext bfloat %885 to float, !dbg !28 + %1142 = fpext bfloat %886 to float, !dbg !28 + %1143 = fpext bfloat %887 to float, !dbg !28 + %1144 = extractelement <32 x i1> %52, i64 1, !dbg !29 + %1145 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %183, i1 %1144) #4, !dbg !29 + %1146 = extractelement <32 x i1> %52, i64 0, !dbg !29 + %1147 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %185, i1 %1146) #4, !dbg !29 + %1148 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %218, i1 %22) #4, !dbg !30 + %1149 = extractvalue { i32, i32, i32, i32 } %1148, 0, !dbg !30 + %1150 = bitcast i32 %1149 to <2 x bfloat>, !dbg !30 + %1151 = extractvalue { i32, i32, i32, i32 } %1148, 1, !dbg !30 + %1152 = bitcast i32 %1151 to <2 x bfloat>, !dbg !30 + %1153 = extractvalue { i32, i32, i32, i32 } %1148, 2, !dbg !30 + %1154 = bitcast i32 %1153 to <2 x bfloat>, !dbg !30 + %1155 = extractvalue { i32, i32, i32, i32 } %1148, 3, !dbg !30 + %1156 = bitcast i32 %1155 to <2 x bfloat>, !dbg !30 + %1157 = extractelement <2 x bfloat> %1150, i64 0, !dbg !30 + %1158 = extractelement <2 x bfloat> %1150, i64 1, !dbg !30 + %1159 = extractelement <2 x bfloat> %1152, i64 0, !dbg !30 + %1160 = extractelement <2 x bfloat> %1152, i64 1, !dbg !30 + %1161 = extractelement <2 x bfloat> %1154, i64 0, !dbg !30 + %1162 = extractelement <2 x bfloat> %1154, i64 1, !dbg !30 + %1163 = extractelement <2 x bfloat> %1156, i64 0, !dbg !30 + %1164 = extractelement <2 x bfloat> %1156, i64 1, !dbg !30 + %1165 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %219, i1 %23) #4, !dbg !30 + %1166 = extractvalue { i32, i32, i32, i32 } %1165, 0, !dbg !30 + %1167 = bitcast i32 %1166 to <2 x bfloat>, !dbg !30 + %1168 = extractvalue { i32, i32, i32, i32 } %1165, 1, !dbg !30 + %1169 = bitcast i32 %1168 to <2 x bfloat>, !dbg !30 + %1170 = extractvalue { i32, i32, i32, i32 } %1165, 2, !dbg !30 + %1171 = bitcast i32 %1170 to <2 x bfloat>, !dbg !30 + %1172 = extractvalue { i32, i32, i32, i32 } %1165, 3, !dbg !30 + %1173 = bitcast i32 %1172 to <2 x bfloat>, !dbg !30 + %1174 = extractelement <2 x bfloat> %1167, i64 0, !dbg !30 + %1175 = extractelement <2 x bfloat> %1167, i64 1, !dbg !30 + %1176 = extractelement <2 x bfloat> %1169, i64 0, !dbg !30 + %1177 = extractelement <2 x bfloat> %1169, i64 1, !dbg !30 + %1178 = extractelement <2 x bfloat> %1171, i64 0, !dbg !30 + %1179 = extractelement <2 x bfloat> %1171, i64 1, !dbg !30 + %1180 = extractelement <2 x bfloat> %1173, i64 0, !dbg !30 + %1181 = extractelement <2 x bfloat> %1173, i64 1, !dbg !30 + %1182 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %220, i1 %348) #4, !dbg !30 + %1183 = extractvalue { i32, i32, i32, i32 } %1182, 0, !dbg !30 + %1184 = bitcast i32 %1183 to <2 x bfloat>, !dbg !30 + %1185 = extractvalue { i32, i32, i32, i32 } %1182, 1, !dbg !30 + %1186 = bitcast i32 %1185 to <2 x bfloat>, !dbg !30 + %1187 = extractvalue { i32, i32, i32, i32 } %1182, 2, !dbg !30 + %1188 = bitcast i32 %1187 to <2 x bfloat>, !dbg !30 + %1189 = extractvalue { i32, i32, i32, i32 } %1182, 3, !dbg !30 + %1190 = bitcast i32 %1189 to <2 x bfloat>, !dbg !30 + %1191 = extractelement <2 x bfloat> %1184, i64 0, !dbg !30 + %1192 = extractelement <2 x bfloat> %1184, i64 1, !dbg !30 + %1193 = extractelement <2 x bfloat> %1186, i64 0, !dbg !30 + %1194 = extractelement <2 x bfloat> %1186, i64 1, !dbg !30 + %1195 = extractelement <2 x bfloat> %1188, i64 0, !dbg !30 + %1196 = extractelement <2 x bfloat> %1188, i64 1, !dbg !30 + %1197 = extractelement <2 x bfloat> %1190, i64 0, !dbg !30 + %1198 = extractelement <2 x bfloat> %1190, i64 1, !dbg !30 + %1199 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %221, i1 %366) #4, !dbg !30 + %1200 = extractvalue { i32, i32, i32, i32 } %1199, 0, !dbg !30 + %1201 = bitcast i32 %1200 to <2 x bfloat>, !dbg !30 + %1202 = extractvalue { i32, i32, i32, i32 } %1199, 1, !dbg !30 + %1203 = bitcast i32 %1202 to <2 x bfloat>, !dbg !30 + %1204 = extractvalue { i32, i32, i32, i32 } %1199, 2, !dbg !30 + %1205 = bitcast i32 %1204 to <2 x bfloat>, !dbg !30 + %1206 = extractvalue { i32, i32, i32, i32 } %1199, 3, !dbg !30 + %1207 = bitcast i32 %1206 to <2 x bfloat>, !dbg !30 + %1208 = extractelement <2 x bfloat> %1201, i64 0, !dbg !30 + %1209 = extractelement <2 x bfloat> %1201, i64 1, !dbg !30 + %1210 = extractelement <2 x bfloat> %1203, i64 0, !dbg !30 + %1211 = extractelement <2 x bfloat> %1203, i64 1, !dbg !30 + %1212 = extractelement <2 x bfloat> %1205, i64 0, !dbg !30 + %1213 = extractelement <2 x bfloat> %1205, i64 1, !dbg !30 + %1214 = extractelement <2 x bfloat> %1207, i64 0, !dbg !30 + %1215 = extractelement <2 x bfloat> %1207, i64 1, !dbg !30 + %1216 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %222, i1 %384) #4, !dbg !30 + %1217 = extractvalue { i32, i32, i32, i32 } %1216, 0, !dbg !30 + %1218 = bitcast i32 %1217 to <2 x bfloat>, !dbg !30 + %1219 = extractvalue { i32, i32, i32, i32 } %1216, 1, !dbg !30 + %1220 = bitcast i32 %1219 to <2 x bfloat>, !dbg !30 + %1221 = extractvalue { i32, i32, i32, i32 } %1216, 2, !dbg !30 + %1222 = bitcast i32 %1221 to <2 x bfloat>, !dbg !30 + %1223 = extractvalue { i32, i32, i32, i32 } %1216, 3, !dbg !30 + %1224 = bitcast i32 %1223 to <2 x bfloat>, !dbg !30 + %1225 = extractelement <2 x bfloat> %1218, i64 0, !dbg !30 + %1226 = extractelement <2 x bfloat> %1218, i64 1, !dbg !30 + %1227 = extractelement <2 x bfloat> %1220, i64 0, !dbg !30 + %1228 = extractelement <2 x bfloat> %1220, i64 1, !dbg !30 + %1229 = extractelement <2 x bfloat> %1222, i64 0, !dbg !30 + %1230 = extractelement <2 x bfloat> %1222, i64 1, !dbg !30 + %1231 = extractelement <2 x bfloat> %1224, i64 0, !dbg !30 + %1232 = extractelement <2 x bfloat> %1224, i64 1, !dbg !30 + %1233 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %223, i1 %402) #4, !dbg !30 + %1234 = extractvalue { i32, i32, i32, i32 } %1233, 0, !dbg !30 + %1235 = bitcast i32 %1234 to <2 x bfloat>, !dbg !30 + %1236 = extractvalue { i32, i32, i32, i32 } %1233, 1, !dbg !30 + %1237 = bitcast i32 %1236 to <2 x bfloat>, !dbg !30 + %1238 = extractvalue { i32, i32, i32, i32 } %1233, 2, !dbg !30 + %1239 = bitcast i32 %1238 to <2 x bfloat>, !dbg !30 + %1240 = extractvalue { i32, i32, i32, i32 } %1233, 3, !dbg !30 + %1241 = bitcast i32 %1240 to <2 x bfloat>, !dbg !30 + %1242 = extractelement <2 x bfloat> %1235, i64 0, !dbg !30 + %1243 = extractelement <2 x bfloat> %1235, i64 1, !dbg !30 + %1244 = extractelement <2 x bfloat> %1237, i64 0, !dbg !30 + %1245 = extractelement <2 x bfloat> %1237, i64 1, !dbg !30 + %1246 = extractelement <2 x bfloat> %1239, i64 0, !dbg !30 + %1247 = extractelement <2 x bfloat> %1239, i64 1, !dbg !30 + %1248 = extractelement <2 x bfloat> %1241, i64 0, !dbg !30 + %1249 = extractelement <2 x bfloat> %1241, i64 1, !dbg !30 + %1250 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %224, i1 %420) #4, !dbg !30 + %1251 = extractvalue { i32, i32, i32, i32 } %1250, 0, !dbg !30 + %1252 = bitcast i32 %1251 to <2 x bfloat>, !dbg !30 + %1253 = extractvalue { i32, i32, i32, i32 } %1250, 1, !dbg !30 + %1254 = bitcast i32 %1253 to <2 x bfloat>, !dbg !30 + %1255 = extractvalue { i32, i32, i32, i32 } %1250, 2, !dbg !30 + %1256 = bitcast i32 %1255 to <2 x bfloat>, !dbg !30 + %1257 = extractvalue { i32, i32, i32, i32 } %1250, 3, !dbg !30 + %1258 = bitcast i32 %1257 to <2 x bfloat>, !dbg !30 + %1259 = extractelement <2 x bfloat> %1252, i64 0, !dbg !30 + %1260 = extractelement <2 x bfloat> %1252, i64 1, !dbg !30 + %1261 = extractelement <2 x bfloat> %1254, i64 0, !dbg !30 + %1262 = extractelement <2 x bfloat> %1254, i64 1, !dbg !30 + %1263 = extractelement <2 x bfloat> %1256, i64 0, !dbg !30 + %1264 = extractelement <2 x bfloat> %1256, i64 1, !dbg !30 + %1265 = extractelement <2 x bfloat> %1258, i64 0, !dbg !30 + %1266 = extractelement <2 x bfloat> %1258, i64 1, !dbg !30 + %1267 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %225, i1 %438) #4, !dbg !30 + %1268 = extractvalue { i32, i32, i32, i32 } %1267, 0, !dbg !30 + %1269 = bitcast i32 %1268 to <2 x bfloat>, !dbg !30 + %1270 = extractvalue { i32, i32, i32, i32 } %1267, 1, !dbg !30 + %1271 = bitcast i32 %1270 to <2 x bfloat>, !dbg !30 + %1272 = extractvalue { i32, i32, i32, i32 } %1267, 2, !dbg !30 + %1273 = bitcast i32 %1272 to <2 x bfloat>, !dbg !30 + %1274 = extractvalue { i32, i32, i32, i32 } %1267, 3, !dbg !30 + %1275 = bitcast i32 %1274 to <2 x bfloat>, !dbg !30 + %1276 = extractelement <2 x bfloat> %1269, i64 0, !dbg !30 + %1277 = extractelement <2 x bfloat> %1269, i64 1, !dbg !30 + %1278 = extractelement <2 x bfloat> %1271, i64 0, !dbg !30 + %1279 = extractelement <2 x bfloat> %1271, i64 1, !dbg !30 + %1280 = extractelement <2 x bfloat> %1273, i64 0, !dbg !30 + %1281 = extractelement <2 x bfloat> %1273, i64 1, !dbg !30 + %1282 = extractelement <2 x bfloat> %1275, i64 0, !dbg !30 + %1283 = extractelement <2 x bfloat> %1275, i64 1, !dbg !30 + %1284 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %226, i1 %456) #4, !dbg !30 + %1285 = extractvalue { i32, i32, i32, i32 } %1284, 0, !dbg !30 + %1286 = bitcast i32 %1285 to <2 x bfloat>, !dbg !30 + %1287 = extractvalue { i32, i32, i32, i32 } %1284, 1, !dbg !30 + %1288 = bitcast i32 %1287 to <2 x bfloat>, !dbg !30 + %1289 = extractvalue { i32, i32, i32, i32 } %1284, 2, !dbg !30 + %1290 = bitcast i32 %1289 to <2 x bfloat>, !dbg !30 + %1291 = extractvalue { i32, i32, i32, i32 } %1284, 3, !dbg !30 + %1292 = bitcast i32 %1291 to <2 x bfloat>, !dbg !30 + %1293 = extractelement <2 x bfloat> %1286, i64 0, !dbg !30 + %1294 = extractelement <2 x bfloat> %1286, i64 1, !dbg !30 + %1295 = extractelement <2 x bfloat> %1288, i64 0, !dbg !30 + %1296 = extractelement <2 x bfloat> %1288, i64 1, !dbg !30 + %1297 = extractelement <2 x bfloat> %1290, i64 0, !dbg !30 + %1298 = extractelement <2 x bfloat> %1290, i64 1, !dbg !30 + %1299 = extractelement <2 x bfloat> %1292, i64 0, !dbg !30 + %1300 = extractelement <2 x bfloat> %1292, i64 1, !dbg !30 + %1301 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %227, i1 %474) #4, !dbg !30 + %1302 = extractvalue { i32, i32, i32, i32 } %1301, 0, !dbg !30 + %1303 = bitcast i32 %1302 to <2 x bfloat>, !dbg !30 + %1304 = extractvalue { i32, i32, i32, i32 } %1301, 1, !dbg !30 + %1305 = bitcast i32 %1304 to <2 x bfloat>, !dbg !30 + %1306 = extractvalue { i32, i32, i32, i32 } %1301, 2, !dbg !30 + %1307 = bitcast i32 %1306 to <2 x bfloat>, !dbg !30 + %1308 = extractvalue { i32, i32, i32, i32 } %1301, 3, !dbg !30 + %1309 = bitcast i32 %1308 to <2 x bfloat>, !dbg !30 + %1310 = extractelement <2 x bfloat> %1303, i64 0, !dbg !30 + %1311 = extractelement <2 x bfloat> %1303, i64 1, !dbg !30 + %1312 = extractelement <2 x bfloat> %1305, i64 0, !dbg !30 + %1313 = extractelement <2 x bfloat> %1305, i64 1, !dbg !30 + %1314 = extractelement <2 x bfloat> %1307, i64 0, !dbg !30 + %1315 = extractelement <2 x bfloat> %1307, i64 1, !dbg !30 + %1316 = extractelement <2 x bfloat> %1309, i64 0, !dbg !30 + %1317 = extractelement <2 x bfloat> %1309, i64 1, !dbg !30 + %1318 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %228, i1 %492) #4, !dbg !30 + %1319 = extractvalue { i32, i32, i32, i32 } %1318, 0, !dbg !30 + %1320 = bitcast i32 %1319 to <2 x bfloat>, !dbg !30 + %1321 = extractvalue { i32, i32, i32, i32 } %1318, 1, !dbg !30 + %1322 = bitcast i32 %1321 to <2 x bfloat>, !dbg !30 + %1323 = extractvalue { i32, i32, i32, i32 } %1318, 2, !dbg !30 + %1324 = bitcast i32 %1323 to <2 x bfloat>, !dbg !30 + %1325 = extractvalue { i32, i32, i32, i32 } %1318, 3, !dbg !30 + %1326 = bitcast i32 %1325 to <2 x bfloat>, !dbg !30 + %1327 = extractelement <2 x bfloat> %1320, i64 0, !dbg !30 + %1328 = extractelement <2 x bfloat> %1320, i64 1, !dbg !30 + %1329 = extractelement <2 x bfloat> %1322, i64 0, !dbg !30 + %1330 = extractelement <2 x bfloat> %1322, i64 1, !dbg !30 + %1331 = extractelement <2 x bfloat> %1324, i64 0, !dbg !30 + %1332 = extractelement <2 x bfloat> %1324, i64 1, !dbg !30 + %1333 = extractelement <2 x bfloat> %1326, i64 0, !dbg !30 + %1334 = extractelement <2 x bfloat> %1326, i64 1, !dbg !30 + %1335 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %229, i1 %510) #4, !dbg !30 + %1336 = extractvalue { i32, i32, i32, i32 } %1335, 0, !dbg !30 + %1337 = bitcast i32 %1336 to <2 x bfloat>, !dbg !30 + %1338 = extractvalue { i32, i32, i32, i32 } %1335, 1, !dbg !30 + %1339 = bitcast i32 %1338 to <2 x bfloat>, !dbg !30 + %1340 = extractvalue { i32, i32, i32, i32 } %1335, 2, !dbg !30 + %1341 = bitcast i32 %1340 to <2 x bfloat>, !dbg !30 + %1342 = extractvalue { i32, i32, i32, i32 } %1335, 3, !dbg !30 + %1343 = bitcast i32 %1342 to <2 x bfloat>, !dbg !30 + %1344 = extractelement <2 x bfloat> %1337, i64 0, !dbg !30 + %1345 = extractelement <2 x bfloat> %1337, i64 1, !dbg !30 + %1346 = extractelement <2 x bfloat> %1339, i64 0, !dbg !30 + %1347 = extractelement <2 x bfloat> %1339, i64 1, !dbg !30 + %1348 = extractelement <2 x bfloat> %1341, i64 0, !dbg !30 + %1349 = extractelement <2 x bfloat> %1341, i64 1, !dbg !30 + %1350 = extractelement <2 x bfloat> %1343, i64 0, !dbg !30 + %1351 = extractelement <2 x bfloat> %1343, i64 1, !dbg !30 + %1352 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %230, i1 %528) #4, !dbg !30 + %1353 = extractvalue { i32, i32, i32, i32 } %1352, 0, !dbg !30 + %1354 = bitcast i32 %1353 to <2 x bfloat>, !dbg !30 + %1355 = extractvalue { i32, i32, i32, i32 } %1352, 1, !dbg !30 + %1356 = bitcast i32 %1355 to <2 x bfloat>, !dbg !30 + %1357 = extractvalue { i32, i32, i32, i32 } %1352, 2, !dbg !30 + %1358 = bitcast i32 %1357 to <2 x bfloat>, !dbg !30 + %1359 = extractvalue { i32, i32, i32, i32 } %1352, 3, !dbg !30 + %1360 = bitcast i32 %1359 to <2 x bfloat>, !dbg !30 + %1361 = extractelement <2 x bfloat> %1354, i64 0, !dbg !30 + %1362 = extractelement <2 x bfloat> %1354, i64 1, !dbg !30 + %1363 = extractelement <2 x bfloat> %1356, i64 0, !dbg !30 + %1364 = extractelement <2 x bfloat> %1356, i64 1, !dbg !30 + %1365 = extractelement <2 x bfloat> %1358, i64 0, !dbg !30 + %1366 = extractelement <2 x bfloat> %1358, i64 1, !dbg !30 + %1367 = extractelement <2 x bfloat> %1360, i64 0, !dbg !30 + %1368 = extractelement <2 x bfloat> %1360, i64 1, !dbg !30 + %1369 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %231, i1 %546) #4, !dbg !30 + %1370 = extractvalue { i32, i32, i32, i32 } %1369, 0, !dbg !30 + %1371 = bitcast i32 %1370 to <2 x bfloat>, !dbg !30 + %1372 = extractvalue { i32, i32, i32, i32 } %1369, 1, !dbg !30 + %1373 = bitcast i32 %1372 to <2 x bfloat>, !dbg !30 + %1374 = extractvalue { i32, i32, i32, i32 } %1369, 2, !dbg !30 + %1375 = bitcast i32 %1374 to <2 x bfloat>, !dbg !30 + %1376 = extractvalue { i32, i32, i32, i32 } %1369, 3, !dbg !30 + %1377 = bitcast i32 %1376 to <2 x bfloat>, !dbg !30 + %1378 = extractelement <2 x bfloat> %1371, i64 0, !dbg !30 + %1379 = extractelement <2 x bfloat> %1371, i64 1, !dbg !30 + %1380 = extractelement <2 x bfloat> %1373, i64 0, !dbg !30 + %1381 = extractelement <2 x bfloat> %1373, i64 1, !dbg !30 + %1382 = extractelement <2 x bfloat> %1375, i64 0, !dbg !30 + %1383 = extractelement <2 x bfloat> %1375, i64 1, !dbg !30 + %1384 = extractelement <2 x bfloat> %1377, i64 0, !dbg !30 + %1385 = extractelement <2 x bfloat> %1377, i64 1, !dbg !30 + %1386 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %232, i1 %564) #4, !dbg !30 + %1387 = extractvalue { i32, i32, i32, i32 } %1386, 0, !dbg !30 + %1388 = bitcast i32 %1387 to <2 x bfloat>, !dbg !30 + %1389 = extractvalue { i32, i32, i32, i32 } %1386, 1, !dbg !30 + %1390 = bitcast i32 %1389 to <2 x bfloat>, !dbg !30 + %1391 = extractvalue { i32, i32, i32, i32 } %1386, 2, !dbg !30 + %1392 = bitcast i32 %1391 to <2 x bfloat>, !dbg !30 + %1393 = extractvalue { i32, i32, i32, i32 } %1386, 3, !dbg !30 + %1394 = bitcast i32 %1393 to <2 x bfloat>, !dbg !30 + %1395 = extractelement <2 x bfloat> %1388, i64 0, !dbg !30 + %1396 = extractelement <2 x bfloat> %1388, i64 1, !dbg !30 + %1397 = extractelement <2 x bfloat> %1390, i64 0, !dbg !30 + %1398 = extractelement <2 x bfloat> %1390, i64 1, !dbg !30 + %1399 = extractelement <2 x bfloat> %1392, i64 0, !dbg !30 + %1400 = extractelement <2 x bfloat> %1392, i64 1, !dbg !30 + %1401 = extractelement <2 x bfloat> %1394, i64 0, !dbg !30 + %1402 = extractelement <2 x bfloat> %1394, i64 1, !dbg !30 + %1403 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %233, i1 %582) #4, !dbg !30 + %1404 = extractvalue { i32, i32, i32, i32 } %1403, 0, !dbg !30 + %1405 = bitcast i32 %1404 to <2 x bfloat>, !dbg !30 + %1406 = extractvalue { i32, i32, i32, i32 } %1403, 1, !dbg !30 + %1407 = bitcast i32 %1406 to <2 x bfloat>, !dbg !30 + %1408 = extractvalue { i32, i32, i32, i32 } %1403, 2, !dbg !30 + %1409 = bitcast i32 %1408 to <2 x bfloat>, !dbg !30 + %1410 = extractvalue { i32, i32, i32, i32 } %1403, 3, !dbg !30 + %1411 = bitcast i32 %1410 to <2 x bfloat>, !dbg !30 + %1412 = extractelement <2 x bfloat> %1405, i64 0, !dbg !30 + %1413 = extractelement <2 x bfloat> %1405, i64 1, !dbg !30 + %1414 = extractelement <2 x bfloat> %1407, i64 0, !dbg !30 + %1415 = extractelement <2 x bfloat> %1407, i64 1, !dbg !30 + %1416 = extractelement <2 x bfloat> %1409, i64 0, !dbg !30 + %1417 = extractelement <2 x bfloat> %1409, i64 1, !dbg !30 + %1418 = extractelement <2 x bfloat> %1411, i64 0, !dbg !30 + %1419 = extractelement <2 x bfloat> %1411, i64 1, !dbg !30 + %1420 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %234, i1 %600) #4, !dbg !30 + %1421 = extractvalue { i32, i32, i32, i32 } %1420, 0, !dbg !30 + %1422 = bitcast i32 %1421 to <2 x bfloat>, !dbg !30 + %1423 = extractvalue { i32, i32, i32, i32 } %1420, 1, !dbg !30 + %1424 = bitcast i32 %1423 to <2 x bfloat>, !dbg !30 + %1425 = extractvalue { i32, i32, i32, i32 } %1420, 2, !dbg !30 + %1426 = bitcast i32 %1425 to <2 x bfloat>, !dbg !30 + %1427 = extractvalue { i32, i32, i32, i32 } %1420, 3, !dbg !30 + %1428 = bitcast i32 %1427 to <2 x bfloat>, !dbg !30 + %1429 = extractelement <2 x bfloat> %1422, i64 0, !dbg !30 + %1430 = extractelement <2 x bfloat> %1422, i64 1, !dbg !30 + %1431 = extractelement <2 x bfloat> %1424, i64 0, !dbg !30 + %1432 = extractelement <2 x bfloat> %1424, i64 1, !dbg !30 + %1433 = extractelement <2 x bfloat> %1426, i64 0, !dbg !30 + %1434 = extractelement <2 x bfloat> %1426, i64 1, !dbg !30 + %1435 = extractelement <2 x bfloat> %1428, i64 0, !dbg !30 + %1436 = extractelement <2 x bfloat> %1428, i64 1, !dbg !30 + %1437 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %235, i1 %618) #4, !dbg !30 + %1438 = extractvalue { i32, i32, i32, i32 } %1437, 0, !dbg !30 + %1439 = bitcast i32 %1438 to <2 x bfloat>, !dbg !30 + %1440 = extractvalue { i32, i32, i32, i32 } %1437, 1, !dbg !30 + %1441 = bitcast i32 %1440 to <2 x bfloat>, !dbg !30 + %1442 = extractvalue { i32, i32, i32, i32 } %1437, 2, !dbg !30 + %1443 = bitcast i32 %1442 to <2 x bfloat>, !dbg !30 + %1444 = extractvalue { i32, i32, i32, i32 } %1437, 3, !dbg !30 + %1445 = bitcast i32 %1444 to <2 x bfloat>, !dbg !30 + %1446 = extractelement <2 x bfloat> %1439, i64 0, !dbg !30 + %1447 = extractelement <2 x bfloat> %1439, i64 1, !dbg !30 + %1448 = extractelement <2 x bfloat> %1441, i64 0, !dbg !30 + %1449 = extractelement <2 x bfloat> %1441, i64 1, !dbg !30 + %1450 = extractelement <2 x bfloat> %1443, i64 0, !dbg !30 + %1451 = extractelement <2 x bfloat> %1443, i64 1, !dbg !30 + %1452 = extractelement <2 x bfloat> %1445, i64 0, !dbg !30 + %1453 = extractelement <2 x bfloat> %1445, i64 1, !dbg !30 + %1454 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %236, i1 %636) #4, !dbg !30 + %1455 = extractvalue { i32, i32, i32, i32 } %1454, 0, !dbg !30 + %1456 = bitcast i32 %1455 to <2 x bfloat>, !dbg !30 + %1457 = extractvalue { i32, i32, i32, i32 } %1454, 1, !dbg !30 + %1458 = bitcast i32 %1457 to <2 x bfloat>, !dbg !30 + %1459 = extractvalue { i32, i32, i32, i32 } %1454, 2, !dbg !30 + %1460 = bitcast i32 %1459 to <2 x bfloat>, !dbg !30 + %1461 = extractvalue { i32, i32, i32, i32 } %1454, 3, !dbg !30 + %1462 = bitcast i32 %1461 to <2 x bfloat>, !dbg !30 + %1463 = extractelement <2 x bfloat> %1456, i64 0, !dbg !30 + %1464 = extractelement <2 x bfloat> %1456, i64 1, !dbg !30 + %1465 = extractelement <2 x bfloat> %1458, i64 0, !dbg !30 + %1466 = extractelement <2 x bfloat> %1458, i64 1, !dbg !30 + %1467 = extractelement <2 x bfloat> %1460, i64 0, !dbg !30 + %1468 = extractelement <2 x bfloat> %1460, i64 1, !dbg !30 + %1469 = extractelement <2 x bfloat> %1462, i64 0, !dbg !30 + %1470 = extractelement <2 x bfloat> %1462, i64 1, !dbg !30 + %1471 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %237, i1 %654) #4, !dbg !30 + %1472 = extractvalue { i32, i32, i32, i32 } %1471, 0, !dbg !30 + %1473 = bitcast i32 %1472 to <2 x bfloat>, !dbg !30 + %1474 = extractvalue { i32, i32, i32, i32 } %1471, 1, !dbg !30 + %1475 = bitcast i32 %1474 to <2 x bfloat>, !dbg !30 + %1476 = extractvalue { i32, i32, i32, i32 } %1471, 2, !dbg !30 + %1477 = bitcast i32 %1476 to <2 x bfloat>, !dbg !30 + %1478 = extractvalue { i32, i32, i32, i32 } %1471, 3, !dbg !30 + %1479 = bitcast i32 %1478 to <2 x bfloat>, !dbg !30 + %1480 = extractelement <2 x bfloat> %1473, i64 0, !dbg !30 + %1481 = extractelement <2 x bfloat> %1473, i64 1, !dbg !30 + %1482 = extractelement <2 x bfloat> %1475, i64 0, !dbg !30 + %1483 = extractelement <2 x bfloat> %1475, i64 1, !dbg !30 + %1484 = extractelement <2 x bfloat> %1477, i64 0, !dbg !30 + %1485 = extractelement <2 x bfloat> %1477, i64 1, !dbg !30 + %1486 = extractelement <2 x bfloat> %1479, i64 0, !dbg !30 + %1487 = extractelement <2 x bfloat> %1479, i64 1, !dbg !30 + %1488 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %238, i1 %672) #4, !dbg !30 + %1489 = extractvalue { i32, i32, i32, i32 } %1488, 0, !dbg !30 + %1490 = bitcast i32 %1489 to <2 x bfloat>, !dbg !30 + %1491 = extractvalue { i32, i32, i32, i32 } %1488, 1, !dbg !30 + %1492 = bitcast i32 %1491 to <2 x bfloat>, !dbg !30 + %1493 = extractvalue { i32, i32, i32, i32 } %1488, 2, !dbg !30 + %1494 = bitcast i32 %1493 to <2 x bfloat>, !dbg !30 + %1495 = extractvalue { i32, i32, i32, i32 } %1488, 3, !dbg !30 + %1496 = bitcast i32 %1495 to <2 x bfloat>, !dbg !30 + %1497 = extractelement <2 x bfloat> %1490, i64 0, !dbg !30 + %1498 = extractelement <2 x bfloat> %1490, i64 1, !dbg !30 + %1499 = extractelement <2 x bfloat> %1492, i64 0, !dbg !30 + %1500 = extractelement <2 x bfloat> %1492, i64 1, !dbg !30 + %1501 = extractelement <2 x bfloat> %1494, i64 0, !dbg !30 + %1502 = extractelement <2 x bfloat> %1494, i64 1, !dbg !30 + %1503 = extractelement <2 x bfloat> %1496, i64 0, !dbg !30 + %1504 = extractelement <2 x bfloat> %1496, i64 1, !dbg !30 + %1505 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %239, i1 %690) #4, !dbg !30 + %1506 = extractvalue { i32, i32, i32, i32 } %1505, 0, !dbg !30 + %1507 = bitcast i32 %1506 to <2 x bfloat>, !dbg !30 + %1508 = extractvalue { i32, i32, i32, i32 } %1505, 1, !dbg !30 + %1509 = bitcast i32 %1508 to <2 x bfloat>, !dbg !30 + %1510 = extractvalue { i32, i32, i32, i32 } %1505, 2, !dbg !30 + %1511 = bitcast i32 %1510 to <2 x bfloat>, !dbg !30 + %1512 = extractvalue { i32, i32, i32, i32 } %1505, 3, !dbg !30 + %1513 = bitcast i32 %1512 to <2 x bfloat>, !dbg !30 + %1514 = extractelement <2 x bfloat> %1507, i64 0, !dbg !30 + %1515 = extractelement <2 x bfloat> %1507, i64 1, !dbg !30 + %1516 = extractelement <2 x bfloat> %1509, i64 0, !dbg !30 + %1517 = extractelement <2 x bfloat> %1509, i64 1, !dbg !30 + %1518 = extractelement <2 x bfloat> %1511, i64 0, !dbg !30 + %1519 = extractelement <2 x bfloat> %1511, i64 1, !dbg !30 + %1520 = extractelement <2 x bfloat> %1513, i64 0, !dbg !30 + %1521 = extractelement <2 x bfloat> %1513, i64 1, !dbg !30 + %1522 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %240, i1 %708) #4, !dbg !30 + %1523 = extractvalue { i32, i32, i32, i32 } %1522, 0, !dbg !30 + %1524 = bitcast i32 %1523 to <2 x bfloat>, !dbg !30 + %1525 = extractvalue { i32, i32, i32, i32 } %1522, 1, !dbg !30 + %1526 = bitcast i32 %1525 to <2 x bfloat>, !dbg !30 + %1527 = extractvalue { i32, i32, i32, i32 } %1522, 2, !dbg !30 + %1528 = bitcast i32 %1527 to <2 x bfloat>, !dbg !30 + %1529 = extractvalue { i32, i32, i32, i32 } %1522, 3, !dbg !30 + %1530 = bitcast i32 %1529 to <2 x bfloat>, !dbg !30 + %1531 = extractelement <2 x bfloat> %1524, i64 0, !dbg !30 + %1532 = extractelement <2 x bfloat> %1524, i64 1, !dbg !30 + %1533 = extractelement <2 x bfloat> %1526, i64 0, !dbg !30 + %1534 = extractelement <2 x bfloat> %1526, i64 1, !dbg !30 + %1535 = extractelement <2 x bfloat> %1528, i64 0, !dbg !30 + %1536 = extractelement <2 x bfloat> %1528, i64 1, !dbg !30 + %1537 = extractelement <2 x bfloat> %1530, i64 0, !dbg !30 + %1538 = extractelement <2 x bfloat> %1530, i64 1, !dbg !30 + %1539 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %241, i1 %726) #4, !dbg !30 + %1540 = extractvalue { i32, i32, i32, i32 } %1539, 0, !dbg !30 + %1541 = bitcast i32 %1540 to <2 x bfloat>, !dbg !30 + %1542 = extractvalue { i32, i32, i32, i32 } %1539, 1, !dbg !30 + %1543 = bitcast i32 %1542 to <2 x bfloat>, !dbg !30 + %1544 = extractvalue { i32, i32, i32, i32 } %1539, 2, !dbg !30 + %1545 = bitcast i32 %1544 to <2 x bfloat>, !dbg !30 + %1546 = extractvalue { i32, i32, i32, i32 } %1539, 3, !dbg !30 + %1547 = bitcast i32 %1546 to <2 x bfloat>, !dbg !30 + %1548 = extractelement <2 x bfloat> %1541, i64 0, !dbg !30 + %1549 = extractelement <2 x bfloat> %1541, i64 1, !dbg !30 + %1550 = extractelement <2 x bfloat> %1543, i64 0, !dbg !30 + %1551 = extractelement <2 x bfloat> %1543, i64 1, !dbg !30 + %1552 = extractelement <2 x bfloat> %1545, i64 0, !dbg !30 + %1553 = extractelement <2 x bfloat> %1545, i64 1, !dbg !30 + %1554 = extractelement <2 x bfloat> %1547, i64 0, !dbg !30 + %1555 = extractelement <2 x bfloat> %1547, i64 1, !dbg !30 + %1556 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %242, i1 %744) #4, !dbg !30 + %1557 = extractvalue { i32, i32, i32, i32 } %1556, 0, !dbg !30 + %1558 = bitcast i32 %1557 to <2 x bfloat>, !dbg !30 + %1559 = extractvalue { i32, i32, i32, i32 } %1556, 1, !dbg !30 + %1560 = bitcast i32 %1559 to <2 x bfloat>, !dbg !30 + %1561 = extractvalue { i32, i32, i32, i32 } %1556, 2, !dbg !30 + %1562 = bitcast i32 %1561 to <2 x bfloat>, !dbg !30 + %1563 = extractvalue { i32, i32, i32, i32 } %1556, 3, !dbg !30 + %1564 = bitcast i32 %1563 to <2 x bfloat>, !dbg !30 + %1565 = extractelement <2 x bfloat> %1558, i64 0, !dbg !30 + %1566 = extractelement <2 x bfloat> %1558, i64 1, !dbg !30 + %1567 = extractelement <2 x bfloat> %1560, i64 0, !dbg !30 + %1568 = extractelement <2 x bfloat> %1560, i64 1, !dbg !30 + %1569 = extractelement <2 x bfloat> %1562, i64 0, !dbg !30 + %1570 = extractelement <2 x bfloat> %1562, i64 1, !dbg !30 + %1571 = extractelement <2 x bfloat> %1564, i64 0, !dbg !30 + %1572 = extractelement <2 x bfloat> %1564, i64 1, !dbg !30 + %1573 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %243, i1 %762) #4, !dbg !30 + %1574 = extractvalue { i32, i32, i32, i32 } %1573, 0, !dbg !30 + %1575 = bitcast i32 %1574 to <2 x bfloat>, !dbg !30 + %1576 = extractvalue { i32, i32, i32, i32 } %1573, 1, !dbg !30 + %1577 = bitcast i32 %1576 to <2 x bfloat>, !dbg !30 + %1578 = extractvalue { i32, i32, i32, i32 } %1573, 2, !dbg !30 + %1579 = bitcast i32 %1578 to <2 x bfloat>, !dbg !30 + %1580 = extractvalue { i32, i32, i32, i32 } %1573, 3, !dbg !30 + %1581 = bitcast i32 %1580 to <2 x bfloat>, !dbg !30 + %1582 = extractelement <2 x bfloat> %1575, i64 0, !dbg !30 + %1583 = extractelement <2 x bfloat> %1575, i64 1, !dbg !30 + %1584 = extractelement <2 x bfloat> %1577, i64 0, !dbg !30 + %1585 = extractelement <2 x bfloat> %1577, i64 1, !dbg !30 + %1586 = extractelement <2 x bfloat> %1579, i64 0, !dbg !30 + %1587 = extractelement <2 x bfloat> %1579, i64 1, !dbg !30 + %1588 = extractelement <2 x bfloat> %1581, i64 0, !dbg !30 + %1589 = extractelement <2 x bfloat> %1581, i64 1, !dbg !30 + %1590 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %244, i1 %780) #4, !dbg !30 + %1591 = extractvalue { i32, i32, i32, i32 } %1590, 0, !dbg !30 + %1592 = bitcast i32 %1591 to <2 x bfloat>, !dbg !30 + %1593 = extractvalue { i32, i32, i32, i32 } %1590, 1, !dbg !30 + %1594 = bitcast i32 %1593 to <2 x bfloat>, !dbg !30 + %1595 = extractvalue { i32, i32, i32, i32 } %1590, 2, !dbg !30 + %1596 = bitcast i32 %1595 to <2 x bfloat>, !dbg !30 + %1597 = extractvalue { i32, i32, i32, i32 } %1590, 3, !dbg !30 + %1598 = bitcast i32 %1597 to <2 x bfloat>, !dbg !30 + %1599 = extractelement <2 x bfloat> %1592, i64 0, !dbg !30 + %1600 = extractelement <2 x bfloat> %1592, i64 1, !dbg !30 + %1601 = extractelement <2 x bfloat> %1594, i64 0, !dbg !30 + %1602 = extractelement <2 x bfloat> %1594, i64 1, !dbg !30 + %1603 = extractelement <2 x bfloat> %1596, i64 0, !dbg !30 + %1604 = extractelement <2 x bfloat> %1596, i64 1, !dbg !30 + %1605 = extractelement <2 x bfloat> %1598, i64 0, !dbg !30 + %1606 = extractelement <2 x bfloat> %1598, i64 1, !dbg !30 + %1607 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %245, i1 %798) #4, !dbg !30 + %1608 = extractvalue { i32, i32, i32, i32 } %1607, 0, !dbg !30 + %1609 = bitcast i32 %1608 to <2 x bfloat>, !dbg !30 + %1610 = extractvalue { i32, i32, i32, i32 } %1607, 1, !dbg !30 + %1611 = bitcast i32 %1610 to <2 x bfloat>, !dbg !30 + %1612 = extractvalue { i32, i32, i32, i32 } %1607, 2, !dbg !30 + %1613 = bitcast i32 %1612 to <2 x bfloat>, !dbg !30 + %1614 = extractvalue { i32, i32, i32, i32 } %1607, 3, !dbg !30 + %1615 = bitcast i32 %1614 to <2 x bfloat>, !dbg !30 + %1616 = extractelement <2 x bfloat> %1609, i64 0, !dbg !30 + %1617 = extractelement <2 x bfloat> %1609, i64 1, !dbg !30 + %1618 = extractelement <2 x bfloat> %1611, i64 0, !dbg !30 + %1619 = extractelement <2 x bfloat> %1611, i64 1, !dbg !30 + %1620 = extractelement <2 x bfloat> %1613, i64 0, !dbg !30 + %1621 = extractelement <2 x bfloat> %1613, i64 1, !dbg !30 + %1622 = extractelement <2 x bfloat> %1615, i64 0, !dbg !30 + %1623 = extractelement <2 x bfloat> %1615, i64 1, !dbg !30 + %1624 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %246, i1 %816) #4, !dbg !30 + %1625 = extractvalue { i32, i32, i32, i32 } %1624, 0, !dbg !30 + %1626 = bitcast i32 %1625 to <2 x bfloat>, !dbg !30 + %1627 = extractvalue { i32, i32, i32, i32 } %1624, 1, !dbg !30 + %1628 = bitcast i32 %1627 to <2 x bfloat>, !dbg !30 + %1629 = extractvalue { i32, i32, i32, i32 } %1624, 2, !dbg !30 + %1630 = bitcast i32 %1629 to <2 x bfloat>, !dbg !30 + %1631 = extractvalue { i32, i32, i32, i32 } %1624, 3, !dbg !30 + %1632 = bitcast i32 %1631 to <2 x bfloat>, !dbg !30 + %1633 = extractelement <2 x bfloat> %1626, i64 0, !dbg !30 + %1634 = extractelement <2 x bfloat> %1626, i64 1, !dbg !30 + %1635 = extractelement <2 x bfloat> %1628, i64 0, !dbg !30 + %1636 = extractelement <2 x bfloat> %1628, i64 1, !dbg !30 + %1637 = extractelement <2 x bfloat> %1630, i64 0, !dbg !30 + %1638 = extractelement <2 x bfloat> %1630, i64 1, !dbg !30 + %1639 = extractelement <2 x bfloat> %1632, i64 0, !dbg !30 + %1640 = extractelement <2 x bfloat> %1632, i64 1, !dbg !30 + %1641 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %247, i1 %834) #4, !dbg !30 + %1642 = extractvalue { i32, i32, i32, i32 } %1641, 0, !dbg !30 + %1643 = bitcast i32 %1642 to <2 x bfloat>, !dbg !30 + %1644 = extractvalue { i32, i32, i32, i32 } %1641, 1, !dbg !30 + %1645 = bitcast i32 %1644 to <2 x bfloat>, !dbg !30 + %1646 = extractvalue { i32, i32, i32, i32 } %1641, 2, !dbg !30 + %1647 = bitcast i32 %1646 to <2 x bfloat>, !dbg !30 + %1648 = extractvalue { i32, i32, i32, i32 } %1641, 3, !dbg !30 + %1649 = bitcast i32 %1648 to <2 x bfloat>, !dbg !30 + %1650 = extractelement <2 x bfloat> %1643, i64 0, !dbg !30 + %1651 = extractelement <2 x bfloat> %1643, i64 1, !dbg !30 + %1652 = extractelement <2 x bfloat> %1645, i64 0, !dbg !30 + %1653 = extractelement <2 x bfloat> %1645, i64 1, !dbg !30 + %1654 = extractelement <2 x bfloat> %1647, i64 0, !dbg !30 + %1655 = extractelement <2 x bfloat> %1647, i64 1, !dbg !30 + %1656 = extractelement <2 x bfloat> %1649, i64 0, !dbg !30 + %1657 = extractelement <2 x bfloat> %1649, i64 1, !dbg !30 + %1658 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %248, i1 %852) #4, !dbg !30 + %1659 = extractvalue { i32, i32, i32, i32 } %1658, 0, !dbg !30 + %1660 = bitcast i32 %1659 to <2 x bfloat>, !dbg !30 + %1661 = extractvalue { i32, i32, i32, i32 } %1658, 1, !dbg !30 + %1662 = bitcast i32 %1661 to <2 x bfloat>, !dbg !30 + %1663 = extractvalue { i32, i32, i32, i32 } %1658, 2, !dbg !30 + %1664 = bitcast i32 %1663 to <2 x bfloat>, !dbg !30 + %1665 = extractvalue { i32, i32, i32, i32 } %1658, 3, !dbg !30 + %1666 = bitcast i32 %1665 to <2 x bfloat>, !dbg !30 + %1667 = extractelement <2 x bfloat> %1660, i64 0, !dbg !30 + %1668 = extractelement <2 x bfloat> %1660, i64 1, !dbg !30 + %1669 = extractelement <2 x bfloat> %1662, i64 0, !dbg !30 + %1670 = extractelement <2 x bfloat> %1662, i64 1, !dbg !30 + %1671 = extractelement <2 x bfloat> %1664, i64 0, !dbg !30 + %1672 = extractelement <2 x bfloat> %1664, i64 1, !dbg !30 + %1673 = extractelement <2 x bfloat> %1666, i64 0, !dbg !30 + %1674 = extractelement <2 x bfloat> %1666, i64 1, !dbg !30 + %1675 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %249, i1 %870) #4, !dbg !30 + %1676 = extractvalue { i32, i32, i32, i32 } %1675, 0, !dbg !30 + %1677 = bitcast i32 %1676 to <2 x bfloat>, !dbg !30 + %1678 = extractvalue { i32, i32, i32, i32 } %1675, 1, !dbg !30 + %1679 = bitcast i32 %1678 to <2 x bfloat>, !dbg !30 + %1680 = extractvalue { i32, i32, i32, i32 } %1675, 2, !dbg !30 + %1681 = bitcast i32 %1680 to <2 x bfloat>, !dbg !30 + %1682 = extractvalue { i32, i32, i32, i32 } %1675, 3, !dbg !30 + %1683 = bitcast i32 %1682 to <2 x bfloat>, !dbg !30 + %1684 = extractelement <2 x bfloat> %1677, i64 0, !dbg !30 + %1685 = extractelement <2 x bfloat> %1677, i64 1, !dbg !30 + %1686 = extractelement <2 x bfloat> %1679, i64 0, !dbg !30 + %1687 = extractelement <2 x bfloat> %1679, i64 1, !dbg !30 + %1688 = extractelement <2 x bfloat> %1681, i64 0, !dbg !30 + %1689 = extractelement <2 x bfloat> %1681, i64 1, !dbg !30 + %1690 = extractelement <2 x bfloat> %1683, i64 0, !dbg !30 + %1691 = extractelement <2 x bfloat> %1683, i64 1, !dbg !30 + %1692 = fpext bfloat %1157 to float, !dbg !31 + %1693 = fpext bfloat %1158 to float, !dbg !31 + %1694 = fpext bfloat %1159 to float, !dbg !31 + %1695 = fpext bfloat %1160 to float, !dbg !31 + %1696 = fpext bfloat %1161 to float, !dbg !31 + %1697 = fpext bfloat %1162 to float, !dbg !31 + %1698 = fpext bfloat %1163 to float, !dbg !31 + %1699 = fpext bfloat %1164 to float, !dbg !31 + %1700 = fpext bfloat %1174 to float, !dbg !31 + %1701 = fpext bfloat %1175 to float, !dbg !31 + %1702 = fpext bfloat %1176 to float, !dbg !31 + %1703 = fpext bfloat %1177 to float, !dbg !31 + %1704 = fpext bfloat %1178 to float, !dbg !31 + %1705 = fpext bfloat %1179 to float, !dbg !31 + %1706 = fpext bfloat %1180 to float, !dbg !31 + %1707 = fpext bfloat %1181 to float, !dbg !31 + %1708 = fpext bfloat %1191 to float, !dbg !31 + %1709 = fpext bfloat %1192 to float, !dbg !31 + %1710 = fpext bfloat %1193 to float, !dbg !31 + %1711 = fpext bfloat %1194 to float, !dbg !31 + %1712 = fpext bfloat %1195 to float, !dbg !31 + %1713 = fpext bfloat %1196 to float, !dbg !31 + %1714 = fpext bfloat %1197 to float, !dbg !31 + %1715 = fpext bfloat %1198 to float, !dbg !31 + %1716 = fpext bfloat %1208 to float, !dbg !31 + %1717 = fpext bfloat %1209 to float, !dbg !31 + %1718 = fpext bfloat %1210 to float, !dbg !31 + %1719 = fpext bfloat %1211 to float, !dbg !31 + %1720 = fpext bfloat %1212 to float, !dbg !31 + %1721 = fpext bfloat %1213 to float, !dbg !31 + %1722 = fpext bfloat %1214 to float, !dbg !31 + %1723 = fpext bfloat %1215 to float, !dbg !31 + %1724 = fpext bfloat %1225 to float, !dbg !31 + %1725 = fpext bfloat %1226 to float, !dbg !31 + %1726 = fpext bfloat %1227 to float, !dbg !31 + %1727 = fpext bfloat %1228 to float, !dbg !31 + %1728 = fpext bfloat %1229 to float, !dbg !31 + %1729 = fpext bfloat %1230 to float, !dbg !31 + %1730 = fpext bfloat %1231 to float, !dbg !31 + %1731 = fpext bfloat %1232 to float, !dbg !31 + %1732 = fpext bfloat %1242 to float, !dbg !31 + %1733 = fpext bfloat %1243 to float, !dbg !31 + %1734 = fpext bfloat %1244 to float, !dbg !31 + %1735 = fpext bfloat %1245 to float, !dbg !31 + %1736 = fpext bfloat %1246 to float, !dbg !31 + %1737 = fpext bfloat %1247 to float, !dbg !31 + %1738 = fpext bfloat %1248 to float, !dbg !31 + %1739 = fpext bfloat %1249 to float, !dbg !31 + %1740 = fpext bfloat %1259 to float, !dbg !31 + %1741 = fpext bfloat %1260 to float, !dbg !31 + %1742 = fpext bfloat %1261 to float, !dbg !31 + %1743 = fpext bfloat %1262 to float, !dbg !31 + %1744 = fpext bfloat %1263 to float, !dbg !31 + %1745 = fpext bfloat %1264 to float, !dbg !31 + %1746 = fpext bfloat %1265 to float, !dbg !31 + %1747 = fpext bfloat %1266 to float, !dbg !31 + %1748 = fpext bfloat %1276 to float, !dbg !31 + %1749 = fpext bfloat %1277 to float, !dbg !31 + %1750 = fpext bfloat %1278 to float, !dbg !31 + %1751 = fpext bfloat %1279 to float, !dbg !31 + %1752 = fpext bfloat %1280 to float, !dbg !31 + %1753 = fpext bfloat %1281 to float, !dbg !31 + %1754 = fpext bfloat %1282 to float, !dbg !31 + %1755 = fpext bfloat %1283 to float, !dbg !31 + %1756 = fpext bfloat %1293 to float, !dbg !31 + %1757 = fpext bfloat %1294 to float, !dbg !31 + %1758 = fpext bfloat %1295 to float, !dbg !31 + %1759 = fpext bfloat %1296 to float, !dbg !31 + %1760 = fpext bfloat %1297 to float, !dbg !31 + %1761 = fpext bfloat %1298 to float, !dbg !31 + %1762 = fpext bfloat %1299 to float, !dbg !31 + %1763 = fpext bfloat %1300 to float, !dbg !31 + %1764 = fpext bfloat %1310 to float, !dbg !31 + %1765 = fpext bfloat %1311 to float, !dbg !31 + %1766 = fpext bfloat %1312 to float, !dbg !31 + %1767 = fpext bfloat %1313 to float, !dbg !31 + %1768 = fpext bfloat %1314 to float, !dbg !31 + %1769 = fpext bfloat %1315 to float, !dbg !31 + %1770 = fpext bfloat %1316 to float, !dbg !31 + %1771 = fpext bfloat %1317 to float, !dbg !31 + %1772 = fpext bfloat %1327 to float, !dbg !31 + %1773 = fpext bfloat %1328 to float, !dbg !31 + %1774 = fpext bfloat %1329 to float, !dbg !31 + %1775 = fpext bfloat %1330 to float, !dbg !31 + %1776 = fpext bfloat %1331 to float, !dbg !31 + %1777 = fpext bfloat %1332 to float, !dbg !31 + %1778 = fpext bfloat %1333 to float, !dbg !31 + %1779 = fpext bfloat %1334 to float, !dbg !31 + %1780 = fpext bfloat %1344 to float, !dbg !31 + %1781 = fpext bfloat %1345 to float, !dbg !31 + %1782 = fpext bfloat %1346 to float, !dbg !31 + %1783 = fpext bfloat %1347 to float, !dbg !31 + %1784 = fpext bfloat %1348 to float, !dbg !31 + %1785 = fpext bfloat %1349 to float, !dbg !31 + %1786 = fpext bfloat %1350 to float, !dbg !31 + %1787 = fpext bfloat %1351 to float, !dbg !31 + %1788 = fpext bfloat %1361 to float, !dbg !31 + %1789 = fpext bfloat %1362 to float, !dbg !31 + %1790 = fpext bfloat %1363 to float, !dbg !31 + %1791 = fpext bfloat %1364 to float, !dbg !31 + %1792 = fpext bfloat %1365 to float, !dbg !31 + %1793 = fpext bfloat %1366 to float, !dbg !31 + %1794 = fpext bfloat %1367 to float, !dbg !31 + %1795 = fpext bfloat %1368 to float, !dbg !31 + %1796 = fpext bfloat %1378 to float, !dbg !31 + %1797 = fpext bfloat %1379 to float, !dbg !31 + %1798 = fpext bfloat %1380 to float, !dbg !31 + %1799 = fpext bfloat %1381 to float, !dbg !31 + %1800 = fpext bfloat %1382 to float, !dbg !31 + %1801 = fpext bfloat %1383 to float, !dbg !31 + %1802 = fpext bfloat %1384 to float, !dbg !31 + %1803 = fpext bfloat %1385 to float, !dbg !31 + %1804 = fpext bfloat %1395 to float, !dbg !31 + %1805 = fpext bfloat %1396 to float, !dbg !31 + %1806 = fpext bfloat %1397 to float, !dbg !31 + %1807 = fpext bfloat %1398 to float, !dbg !31 + %1808 = fpext bfloat %1399 to float, !dbg !31 + %1809 = fpext bfloat %1400 to float, !dbg !31 + %1810 = fpext bfloat %1401 to float, !dbg !31 + %1811 = fpext bfloat %1402 to float, !dbg !31 + %1812 = fpext bfloat %1412 to float, !dbg !31 + %1813 = fpext bfloat %1413 to float, !dbg !31 + %1814 = fpext bfloat %1414 to float, !dbg !31 + %1815 = fpext bfloat %1415 to float, !dbg !31 + %1816 = fpext bfloat %1416 to float, !dbg !31 + %1817 = fpext bfloat %1417 to float, !dbg !31 + %1818 = fpext bfloat %1418 to float, !dbg !31 + %1819 = fpext bfloat %1419 to float, !dbg !31 + %1820 = fpext bfloat %1429 to float, !dbg !31 + %1821 = fpext bfloat %1430 to float, !dbg !31 + %1822 = fpext bfloat %1431 to float, !dbg !31 + %1823 = fpext bfloat %1432 to float, !dbg !31 + %1824 = fpext bfloat %1433 to float, !dbg !31 + %1825 = fpext bfloat %1434 to float, !dbg !31 + %1826 = fpext bfloat %1435 to float, !dbg !31 + %1827 = fpext bfloat %1436 to float, !dbg !31 + %1828 = fpext bfloat %1446 to float, !dbg !31 + %1829 = fpext bfloat %1447 to float, !dbg !31 + %1830 = fpext bfloat %1448 to float, !dbg !31 + %1831 = fpext bfloat %1449 to float, !dbg !31 + %1832 = fpext bfloat %1450 to float, !dbg !31 + %1833 = fpext bfloat %1451 to float, !dbg !31 + %1834 = fpext bfloat %1452 to float, !dbg !31 + %1835 = fpext bfloat %1453 to float, !dbg !31 + %1836 = fpext bfloat %1463 to float, !dbg !31 + %1837 = fpext bfloat %1464 to float, !dbg !31 + %1838 = fpext bfloat %1465 to float, !dbg !31 + %1839 = fpext bfloat %1466 to float, !dbg !31 + %1840 = fpext bfloat %1467 to float, !dbg !31 + %1841 = fpext bfloat %1468 to float, !dbg !31 + %1842 = fpext bfloat %1469 to float, !dbg !31 + %1843 = fpext bfloat %1470 to float, !dbg !31 + %1844 = fpext bfloat %1480 to float, !dbg !31 + %1845 = fpext bfloat %1481 to float, !dbg !31 + %1846 = fpext bfloat %1482 to float, !dbg !31 + %1847 = fpext bfloat %1483 to float, !dbg !31 + %1848 = fpext bfloat %1484 to float, !dbg !31 + %1849 = fpext bfloat %1485 to float, !dbg !31 + %1850 = fpext bfloat %1486 to float, !dbg !31 + %1851 = fpext bfloat %1487 to float, !dbg !31 + %1852 = fpext bfloat %1497 to float, !dbg !31 + %1853 = fpext bfloat %1498 to float, !dbg !31 + %1854 = fpext bfloat %1499 to float, !dbg !31 + %1855 = fpext bfloat %1500 to float, !dbg !31 + %1856 = fpext bfloat %1501 to float, !dbg !31 + %1857 = fpext bfloat %1502 to float, !dbg !31 + %1858 = fpext bfloat %1503 to float, !dbg !31 + %1859 = fpext bfloat %1504 to float, !dbg !31 + %1860 = fpext bfloat %1514 to float, !dbg !31 + %1861 = fpext bfloat %1515 to float, !dbg !31 + %1862 = fpext bfloat %1516 to float, !dbg !31 + %1863 = fpext bfloat %1517 to float, !dbg !31 + %1864 = fpext bfloat %1518 to float, !dbg !31 + %1865 = fpext bfloat %1519 to float, !dbg !31 + %1866 = fpext bfloat %1520 to float, !dbg !31 + %1867 = fpext bfloat %1521 to float, !dbg !31 + %1868 = fpext bfloat %1531 to float, !dbg !31 + %1869 = fpext bfloat %1532 to float, !dbg !31 + %1870 = fpext bfloat %1533 to float, !dbg !31 + %1871 = fpext bfloat %1534 to float, !dbg !31 + %1872 = fpext bfloat %1535 to float, !dbg !31 + %1873 = fpext bfloat %1536 to float, !dbg !31 + %1874 = fpext bfloat %1537 to float, !dbg !31 + %1875 = fpext bfloat %1538 to float, !dbg !31 + %1876 = fpext bfloat %1548 to float, !dbg !31 + %1877 = fpext bfloat %1549 to float, !dbg !31 + %1878 = fpext bfloat %1550 to float, !dbg !31 + %1879 = fpext bfloat %1551 to float, !dbg !31 + %1880 = fpext bfloat %1552 to float, !dbg !31 + %1881 = fpext bfloat %1553 to float, !dbg !31 + %1882 = fpext bfloat %1554 to float, !dbg !31 + %1883 = fpext bfloat %1555 to float, !dbg !31 + %1884 = fpext bfloat %1565 to float, !dbg !31 + %1885 = fpext bfloat %1566 to float, !dbg !31 + %1886 = fpext bfloat %1567 to float, !dbg !31 + %1887 = fpext bfloat %1568 to float, !dbg !31 + %1888 = fpext bfloat %1569 to float, !dbg !31 + %1889 = fpext bfloat %1570 to float, !dbg !31 + %1890 = fpext bfloat %1571 to float, !dbg !31 + %1891 = fpext bfloat %1572 to float, !dbg !31 + %1892 = fpext bfloat %1582 to float, !dbg !31 + %1893 = fpext bfloat %1583 to float, !dbg !31 + %1894 = fpext bfloat %1584 to float, !dbg !31 + %1895 = fpext bfloat %1585 to float, !dbg !31 + %1896 = fpext bfloat %1586 to float, !dbg !31 + %1897 = fpext bfloat %1587 to float, !dbg !31 + %1898 = fpext bfloat %1588 to float, !dbg !31 + %1899 = fpext bfloat %1589 to float, !dbg !31 + %1900 = fpext bfloat %1599 to float, !dbg !31 + %1901 = fpext bfloat %1600 to float, !dbg !31 + %1902 = fpext bfloat %1601 to float, !dbg !31 + %1903 = fpext bfloat %1602 to float, !dbg !31 + %1904 = fpext bfloat %1603 to float, !dbg !31 + %1905 = fpext bfloat %1604 to float, !dbg !31 + %1906 = fpext bfloat %1605 to float, !dbg !31 + %1907 = fpext bfloat %1606 to float, !dbg !31 + %1908 = fpext bfloat %1616 to float, !dbg !31 + %1909 = fpext bfloat %1617 to float, !dbg !31 + %1910 = fpext bfloat %1618 to float, !dbg !31 + %1911 = fpext bfloat %1619 to float, !dbg !31 + %1912 = fpext bfloat %1620 to float, !dbg !31 + %1913 = fpext bfloat %1621 to float, !dbg !31 + %1914 = fpext bfloat %1622 to float, !dbg !31 + %1915 = fpext bfloat %1623 to float, !dbg !31 + %1916 = fpext bfloat %1633 to float, !dbg !31 + %1917 = fpext bfloat %1634 to float, !dbg !31 + %1918 = fpext bfloat %1635 to float, !dbg !31 + %1919 = fpext bfloat %1636 to float, !dbg !31 + %1920 = fpext bfloat %1637 to float, !dbg !31 + %1921 = fpext bfloat %1638 to float, !dbg !31 + %1922 = fpext bfloat %1639 to float, !dbg !31 + %1923 = fpext bfloat %1640 to float, !dbg !31 + %1924 = fpext bfloat %1650 to float, !dbg !31 + %1925 = fpext bfloat %1651 to float, !dbg !31 + %1926 = fpext bfloat %1652 to float, !dbg !31 + %1927 = fpext bfloat %1653 to float, !dbg !31 + %1928 = fpext bfloat %1654 to float, !dbg !31 + %1929 = fpext bfloat %1655 to float, !dbg !31 + %1930 = fpext bfloat %1656 to float, !dbg !31 + %1931 = fpext bfloat %1657 to float, !dbg !31 + %1932 = fpext bfloat %1667 to float, !dbg !31 + %1933 = fpext bfloat %1668 to float, !dbg !31 + %1934 = fpext bfloat %1669 to float, !dbg !31 + %1935 = fpext bfloat %1670 to float, !dbg !31 + %1936 = fpext bfloat %1671 to float, !dbg !31 + %1937 = fpext bfloat %1672 to float, !dbg !31 + %1938 = fpext bfloat %1673 to float, !dbg !31 + %1939 = fpext bfloat %1674 to float, !dbg !31 + %1940 = fpext bfloat %1684 to float, !dbg !31 + %1941 = fpext bfloat %1685 to float, !dbg !31 + %1942 = fpext bfloat %1686 to float, !dbg !31 + %1943 = fpext bfloat %1687 to float, !dbg !31 + %1944 = fpext bfloat %1688 to float, !dbg !31 + %1945 = fpext bfloat %1689 to float, !dbg !31 + %1946 = fpext bfloat %1690 to float, !dbg !31 + %1947 = fpext bfloat %1691 to float, !dbg !31 + %1948 = and i32 %16, 60, !dbg !32 + %1949 = shl nuw nsw i32 %12, 6, !dbg !32 + %1950 = and i32 %1949, 64, !dbg !32 + %1951 = and i32 %12, 32, !dbg !32 + %1952 = shl nuw nsw i32 %1951, 2, !dbg !32 + %1953 = or disjoint i32 %1950, %1952, !dbg !32 + %1954 = or disjoint i32 %1953, %1948, !dbg !32 + %1955 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %1954, !dbg !32 + %1956 = insertelement <1 x i32> poison, i32 %1145, i64 0, !dbg !32 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) %1955, <1 x i32> %1956) #4, !dbg !32 + %1957 = xor i32 %1954, 320, !dbg !32 + %1958 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %1957, !dbg !32 + %1959 = insertelement <1 x i32> poison, i32 %1147, i64 0, !dbg !32 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) %1958, <1 x i32> %1959) #4, !dbg !32 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !32 + %1960 = and i32 %12, 16, !dbg !32 + %1961 = icmp eq i32 %1960, 0, !dbg !32 + %1962 = select i1 %1961, i32 0, i32 320, !dbg !32 + %1963 = shl nuw nsw i32 %1951, 1, !dbg !32 + %1964 = xor i32 %1962, %1963, !dbg !32 + %1965 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %1964, !dbg !32 + %1966 = load float, ptr addrspace(3) %1965, align 16, !dbg !32 + %1967 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 4, !dbg !32 + %1968 = load float, ptr addrspace(3) %1967, align 4, !dbg !32 + %1969 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 8, !dbg !32 + %1970 = load float, ptr addrspace(3) %1969, align 8, !dbg !32 + %1971 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 12, !dbg !32 + %1972 = load float, ptr addrspace(3) %1971, align 4, !dbg !32 + %1973 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 16, !dbg !32 + %1974 = load float, ptr addrspace(3) %1973, align 16, !dbg !32 + %1975 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 20, !dbg !32 + %1976 = load float, ptr addrspace(3) %1975, align 4, !dbg !32 + %1977 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 24, !dbg !32 + %1978 = load float, ptr addrspace(3) %1977, align 8, !dbg !32 + %1979 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 28, !dbg !32 + %1980 = load float, ptr addrspace(3) %1979, align 4, !dbg !32 + %1981 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 32, !dbg !32 + %1982 = load float, ptr addrspace(3) %1981, align 16, !dbg !32 + %1983 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 36, !dbg !32 + %1984 = load float, ptr addrspace(3) %1983, align 4, !dbg !32 + %1985 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 40, !dbg !32 + %1986 = load float, ptr addrspace(3) %1985, align 8, !dbg !32 + %1987 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 44, !dbg !32 + %1988 = load float, ptr addrspace(3) %1987, align 4, !dbg !32 + %1989 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 48, !dbg !32 + %1990 = load float, ptr addrspace(3) %1989, align 16, !dbg !32 + %1991 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 52, !dbg !32 + %1992 = load float, ptr addrspace(3) %1991, align 4, !dbg !32 + %1993 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 56, !dbg !32 + %1994 = load float, ptr addrspace(3) %1993, align 8, !dbg !32 + %1995 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 60, !dbg !32 + %1996 = load float, ptr addrspace(3) %1995, align 4, !dbg !32 + %1997 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 128, !dbg !32 + %1998 = load float, ptr addrspace(3) %1997, align 16, !dbg !32 + %1999 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 132, !dbg !32 + %2000 = load float, ptr addrspace(3) %1999, align 4, !dbg !32 + %2001 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 136, !dbg !32 + %2002 = load float, ptr addrspace(3) %2001, align 8, !dbg !32 + %2003 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 140, !dbg !32 + %2004 = load float, ptr addrspace(3) %2003, align 4, !dbg !32 + %2005 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 144, !dbg !32 + %2006 = load float, ptr addrspace(3) %2005, align 16, !dbg !32 + %2007 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 148, !dbg !32 + %2008 = load float, ptr addrspace(3) %2007, align 4, !dbg !32 + %2009 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 152, !dbg !32 + %2010 = load float, ptr addrspace(3) %2009, align 8, !dbg !32 + %2011 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 156, !dbg !32 + %2012 = load float, ptr addrspace(3) %2011, align 4, !dbg !32 + %2013 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 160, !dbg !32 + %2014 = load float, ptr addrspace(3) %2013, align 16, !dbg !32 + %2015 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 164, !dbg !32 + %2016 = load float, ptr addrspace(3) %2015, align 4, !dbg !32 + %2017 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 168, !dbg !32 + %2018 = load float, ptr addrspace(3) %2017, align 8, !dbg !32 + %2019 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 172, !dbg !32 + %2020 = load float, ptr addrspace(3) %2019, align 4, !dbg !32 + %2021 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 176, !dbg !32 + %2022 = load float, ptr addrspace(3) %2021, align 16, !dbg !32 + %2023 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 180, !dbg !32 + %2024 = load float, ptr addrspace(3) %2023, align 4, !dbg !32 + %2025 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 184, !dbg !32 + %2026 = load float, ptr addrspace(3) %2025, align 8, !dbg !32 + %2027 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 188, !dbg !32 + %2028 = load float, ptr addrspace(3) %2027, align 4, !dbg !32 + %2029 = fmul float %1966, %1692, !dbg !33 + %2030 = fmul float %1966, %1693, !dbg !33 + %2031 = fmul float %1966, %1694, !dbg !33 + %2032 = fmul float %1966, %1695, !dbg !33 + %2033 = fmul float %1966, %1696, !dbg !33 + %2034 = fmul float %1966, %1697, !dbg !33 + %2035 = fmul float %1966, %1698, !dbg !33 + %2036 = fmul float %1966, %1699, !dbg !33 + %2037 = fmul float %1968, %1700, !dbg !33 + %2038 = fmul float %1968, %1701, !dbg !33 + %2039 = fmul float %1968, %1702, !dbg !33 + %2040 = fmul float %1968, %1703, !dbg !33 + %2041 = fmul float %1968, %1704, !dbg !33 + %2042 = fmul float %1968, %1705, !dbg !33 + %2043 = fmul float %1968, %1706, !dbg !33 + %2044 = fmul float %1968, %1707, !dbg !33 + %2045 = fmul float %1970, %1708, !dbg !33 + %2046 = fmul float %1970, %1709, !dbg !33 + %2047 = fmul float %1970, %1710, !dbg !33 + %2048 = fmul float %1970, %1711, !dbg !33 + %2049 = fmul float %1970, %1712, !dbg !33 + %2050 = fmul float %1970, %1713, !dbg !33 + %2051 = fmul float %1970, %1714, !dbg !33 + %2052 = fmul float %1970, %1715, !dbg !33 + %2053 = fmul float %1972, %1716, !dbg !33 + %2054 = fmul float %1972, %1717, !dbg !33 + %2055 = fmul float %1972, %1718, !dbg !33 + %2056 = fmul float %1972, %1719, !dbg !33 + %2057 = fmul float %1972, %1720, !dbg !33 + %2058 = fmul float %1972, %1721, !dbg !33 + %2059 = fmul float %1972, %1722, !dbg !33 + %2060 = fmul float %1972, %1723, !dbg !33 + %2061 = fmul float %1974, %1724, !dbg !33 + %2062 = fmul float %1974, %1725, !dbg !33 + %2063 = fmul float %1974, %1726, !dbg !33 + %2064 = fmul float %1974, %1727, !dbg !33 + %2065 = fmul float %1974, %1728, !dbg !33 + %2066 = fmul float %1974, %1729, !dbg !33 + %2067 = fmul float %1974, %1730, !dbg !33 + %2068 = fmul float %1974, %1731, !dbg !33 + %2069 = fmul float %1976, %1732, !dbg !33 + %2070 = fmul float %1976, %1733, !dbg !33 + %2071 = fmul float %1976, %1734, !dbg !33 + %2072 = fmul float %1976, %1735, !dbg !33 + %2073 = fmul float %1976, %1736, !dbg !33 + %2074 = fmul float %1976, %1737, !dbg !33 + %2075 = fmul float %1976, %1738, !dbg !33 + %2076 = fmul float %1976, %1739, !dbg !33 + %2077 = fmul float %1978, %1740, !dbg !33 + %2078 = fmul float %1978, %1741, !dbg !33 + %2079 = fmul float %1978, %1742, !dbg !33 + %2080 = fmul float %1978, %1743, !dbg !33 + %2081 = fmul float %1978, %1744, !dbg !33 + %2082 = fmul float %1978, %1745, !dbg !33 + %2083 = fmul float %1978, %1746, !dbg !33 + %2084 = fmul float %1978, %1747, !dbg !33 + %2085 = fmul float %1980, %1748, !dbg !33 + %2086 = fmul float %1980, %1749, !dbg !33 + %2087 = fmul float %1980, %1750, !dbg !33 + %2088 = fmul float %1980, %1751, !dbg !33 + %2089 = fmul float %1980, %1752, !dbg !33 + %2090 = fmul float %1980, %1753, !dbg !33 + %2091 = fmul float %1980, %1754, !dbg !33 + %2092 = fmul float %1980, %1755, !dbg !33 + %2093 = fmul float %1982, %1756, !dbg !33 + %2094 = fmul float %1982, %1757, !dbg !33 + %2095 = fmul float %1982, %1758, !dbg !33 + %2096 = fmul float %1982, %1759, !dbg !33 + %2097 = fmul float %1982, %1760, !dbg !33 + %2098 = fmul float %1982, %1761, !dbg !33 + %2099 = fmul float %1982, %1762, !dbg !33 + %2100 = fmul float %1982, %1763, !dbg !33 + %2101 = fmul float %1984, %1764, !dbg !33 + %2102 = fmul float %1984, %1765, !dbg !33 + %2103 = fmul float %1984, %1766, !dbg !33 + %2104 = fmul float %1984, %1767, !dbg !33 + %2105 = fmul float %1984, %1768, !dbg !33 + %2106 = fmul float %1984, %1769, !dbg !33 + %2107 = fmul float %1984, %1770, !dbg !33 + %2108 = fmul float %1984, %1771, !dbg !33 + %2109 = fmul float %1986, %1772, !dbg !33 + %2110 = fmul float %1986, %1773, !dbg !33 + %2111 = fmul float %1986, %1774, !dbg !33 + %2112 = fmul float %1986, %1775, !dbg !33 + %2113 = fmul float %1986, %1776, !dbg !33 + %2114 = fmul float %1986, %1777, !dbg !33 + %2115 = fmul float %1986, %1778, !dbg !33 + %2116 = fmul float %1986, %1779, !dbg !33 + %2117 = fmul float %1988, %1780, !dbg !33 + %2118 = fmul float %1988, %1781, !dbg !33 + %2119 = fmul float %1988, %1782, !dbg !33 + %2120 = fmul float %1988, %1783, !dbg !33 + %2121 = fmul float %1988, %1784, !dbg !33 + %2122 = fmul float %1988, %1785, !dbg !33 + %2123 = fmul float %1988, %1786, !dbg !33 + %2124 = fmul float %1988, %1787, !dbg !33 + %2125 = fmul float %1990, %1788, !dbg !33 + %2126 = fmul float %1990, %1789, !dbg !33 + %2127 = fmul float %1990, %1790, !dbg !33 + %2128 = fmul float %1990, %1791, !dbg !33 + %2129 = fmul float %1990, %1792, !dbg !33 + %2130 = fmul float %1990, %1793, !dbg !33 + %2131 = fmul float %1990, %1794, !dbg !33 + %2132 = fmul float %1990, %1795, !dbg !33 + %2133 = fmul float %1992, %1796, !dbg !33 + %2134 = fmul float %1992, %1797, !dbg !33 + %2135 = fmul float %1992, %1798, !dbg !33 + %2136 = fmul float %1992, %1799, !dbg !33 + %2137 = fmul float %1992, %1800, !dbg !33 + %2138 = fmul float %1992, %1801, !dbg !33 + %2139 = fmul float %1992, %1802, !dbg !33 + %2140 = fmul float %1992, %1803, !dbg !33 + %2141 = fmul float %1994, %1804, !dbg !33 + %2142 = fmul float %1994, %1805, !dbg !33 + %2143 = fmul float %1994, %1806, !dbg !33 + %2144 = fmul float %1994, %1807, !dbg !33 + %2145 = fmul float %1994, %1808, !dbg !33 + %2146 = fmul float %1994, %1809, !dbg !33 + %2147 = fmul float %1994, %1810, !dbg !33 + %2148 = fmul float %1994, %1811, !dbg !33 + %2149 = fmul float %1996, %1812, !dbg !33 + %2150 = fmul float %1996, %1813, !dbg !33 + %2151 = fmul float %1996, %1814, !dbg !33 + %2152 = fmul float %1996, %1815, !dbg !33 + %2153 = fmul float %1996, %1816, !dbg !33 + %2154 = fmul float %1996, %1817, !dbg !33 + %2155 = fmul float %1996, %1818, !dbg !33 + %2156 = fmul float %1996, %1819, !dbg !33 + %2157 = fmul float %1998, %1820, !dbg !33 + %2158 = fmul float %1998, %1821, !dbg !33 + %2159 = fmul float %1998, %1822, !dbg !33 + %2160 = fmul float %1998, %1823, !dbg !33 + %2161 = fmul float %1998, %1824, !dbg !33 + %2162 = fmul float %1998, %1825, !dbg !33 + %2163 = fmul float %1998, %1826, !dbg !33 + %2164 = fmul float %1998, %1827, !dbg !33 + %2165 = fmul float %2000, %1828, !dbg !33 + %2166 = fmul float %2000, %1829, !dbg !33 + %2167 = fmul float %2000, %1830, !dbg !33 + %2168 = fmul float %2000, %1831, !dbg !33 + %2169 = fmul float %2000, %1832, !dbg !33 + %2170 = fmul float %2000, %1833, !dbg !33 + %2171 = fmul float %2000, %1834, !dbg !33 + %2172 = fmul float %2000, %1835, !dbg !33 + %2173 = fmul float %2002, %1836, !dbg !33 + %2174 = fmul float %2002, %1837, !dbg !33 + %2175 = fmul float %2002, %1838, !dbg !33 + %2176 = fmul float %2002, %1839, !dbg !33 + %2177 = fmul float %2002, %1840, !dbg !33 + %2178 = fmul float %2002, %1841, !dbg !33 + %2179 = fmul float %2002, %1842, !dbg !33 + %2180 = fmul float %2002, %1843, !dbg !33 + %2181 = fmul float %2004, %1844, !dbg !33 + %2182 = fmul float %2004, %1845, !dbg !33 + %2183 = fmul float %2004, %1846, !dbg !33 + %2184 = fmul float %2004, %1847, !dbg !33 + %2185 = fmul float %2004, %1848, !dbg !33 + %2186 = fmul float %2004, %1849, !dbg !33 + %2187 = fmul float %2004, %1850, !dbg !33 + %2188 = fmul float %2004, %1851, !dbg !33 + %2189 = fmul float %2006, %1852, !dbg !33 + %2190 = fmul float %2006, %1853, !dbg !33 + %2191 = fmul float %2006, %1854, !dbg !33 + %2192 = fmul float %2006, %1855, !dbg !33 + %2193 = fmul float %2006, %1856, !dbg !33 + %2194 = fmul float %2006, %1857, !dbg !33 + %2195 = fmul float %2006, %1858, !dbg !33 + %2196 = fmul float %2006, %1859, !dbg !33 + %2197 = fmul float %2008, %1860, !dbg !33 + %2198 = fmul float %2008, %1861, !dbg !33 + %2199 = fmul float %2008, %1862, !dbg !33 + %2200 = fmul float %2008, %1863, !dbg !33 + %2201 = fmul float %2008, %1864, !dbg !33 + %2202 = fmul float %2008, %1865, !dbg !33 + %2203 = fmul float %2008, %1866, !dbg !33 + %2204 = fmul float %2008, %1867, !dbg !33 + %2205 = fmul float %2010, %1868, !dbg !33 + %2206 = fmul float %2010, %1869, !dbg !33 + %2207 = fmul float %2010, %1870, !dbg !33 + %2208 = fmul float %2010, %1871, !dbg !33 + %2209 = fmul float %2010, %1872, !dbg !33 + %2210 = fmul float %2010, %1873, !dbg !33 + %2211 = fmul float %2010, %1874, !dbg !33 + %2212 = fmul float %2010, %1875, !dbg !33 + %2213 = fmul float %2012, %1876, !dbg !33 + %2214 = fmul float %2012, %1877, !dbg !33 + %2215 = fmul float %2012, %1878, !dbg !33 + %2216 = fmul float %2012, %1879, !dbg !33 + %2217 = fmul float %2012, %1880, !dbg !33 + %2218 = fmul float %2012, %1881, !dbg !33 + %2219 = fmul float %2012, %1882, !dbg !33 + %2220 = fmul float %2012, %1883, !dbg !33 + %2221 = fmul float %2014, %1884, !dbg !33 + %2222 = fmul float %2014, %1885, !dbg !33 + %2223 = fmul float %2014, %1886, !dbg !33 + %2224 = fmul float %2014, %1887, !dbg !33 + %2225 = fmul float %2014, %1888, !dbg !33 + %2226 = fmul float %2014, %1889, !dbg !33 + %2227 = fmul float %2014, %1890, !dbg !33 + %2228 = fmul float %2014, %1891, !dbg !33 + %2229 = fmul float %2016, %1892, !dbg !33 + %2230 = fmul float %2016, %1893, !dbg !33 + %2231 = fmul float %2016, %1894, !dbg !33 + %2232 = fmul float %2016, %1895, !dbg !33 + %2233 = fmul float %2016, %1896, !dbg !33 + %2234 = fmul float %2016, %1897, !dbg !33 + %2235 = fmul float %2016, %1898, !dbg !33 + %2236 = fmul float %2016, %1899, !dbg !33 + %2237 = fmul float %2018, %1900, !dbg !33 + %2238 = fmul float %2018, %1901, !dbg !33 + %2239 = fmul float %2018, %1902, !dbg !33 + %2240 = fmul float %2018, %1903, !dbg !33 + %2241 = fmul float %2018, %1904, !dbg !33 + %2242 = fmul float %2018, %1905, !dbg !33 + %2243 = fmul float %2018, %1906, !dbg !33 + %2244 = fmul float %2018, %1907, !dbg !33 + %2245 = fmul float %2020, %1908, !dbg !33 + %2246 = fmul float %2020, %1909, !dbg !33 + %2247 = fmul float %2020, %1910, !dbg !33 + %2248 = fmul float %2020, %1911, !dbg !33 + %2249 = fmul float %2020, %1912, !dbg !33 + %2250 = fmul float %2020, %1913, !dbg !33 + %2251 = fmul float %2020, %1914, !dbg !33 + %2252 = fmul float %2020, %1915, !dbg !33 + %2253 = fmul float %2022, %1916, !dbg !33 + %2254 = fmul float %2022, %1917, !dbg !33 + %2255 = fmul float %2022, %1918, !dbg !33 + %2256 = fmul float %2022, %1919, !dbg !33 + %2257 = fmul float %2022, %1920, !dbg !33 + %2258 = fmul float %2022, %1921, !dbg !33 + %2259 = fmul float %2022, %1922, !dbg !33 + %2260 = fmul float %2022, %1923, !dbg !33 + %2261 = fmul float %2024, %1924, !dbg !33 + %2262 = fmul float %2024, %1925, !dbg !33 + %2263 = fmul float %2024, %1926, !dbg !33 + %2264 = fmul float %2024, %1927, !dbg !33 + %2265 = fmul float %2024, %1928, !dbg !33 + %2266 = fmul float %2024, %1929, !dbg !33 + %2267 = fmul float %2024, %1930, !dbg !33 + %2268 = fmul float %2024, %1931, !dbg !33 + %2269 = fmul float %2026, %1932, !dbg !33 + %2270 = fmul float %2026, %1933, !dbg !33 + %2271 = fmul float %2026, %1934, !dbg !33 + %2272 = fmul float %2026, %1935, !dbg !33 + %2273 = fmul float %2026, %1936, !dbg !33 + %2274 = fmul float %2026, %1937, !dbg !33 + %2275 = fmul float %2026, %1938, !dbg !33 + %2276 = fmul float %2026, %1939, !dbg !33 + %2277 = fmul float %2028, %1940, !dbg !33 + %2278 = fmul float %2028, %1941, !dbg !33 + %2279 = fmul float %2028, %1942, !dbg !33 + %2280 = fmul float %2028, %1943, !dbg !33 + %2281 = fmul float %2028, %1944, !dbg !33 + %2282 = fmul float %2028, %1945, !dbg !33 + %2283 = fmul float %2028, %1946, !dbg !33 + %2284 = fmul float %2028, %1947, !dbg !33 + %2285 = fmul float %888, %1692, !dbg !34 + %2286 = fmul float %889, %1693, !dbg !34 + %2287 = fmul float %890, %1694, !dbg !34 + %2288 = fmul float %891, %1695, !dbg !34 + %2289 = fmul float %892, %1696, !dbg !34 + %2290 = fmul float %893, %1697, !dbg !34 + %2291 = fmul float %894, %1698, !dbg !34 + %2292 = fmul float %895, %1699, !dbg !34 + %2293 = fmul float %896, %1700, !dbg !34 + %2294 = fmul float %897, %1701, !dbg !34 + %2295 = fmul float %898, %1702, !dbg !34 + %2296 = fmul float %899, %1703, !dbg !34 + %2297 = fmul float %900, %1704, !dbg !34 + %2298 = fmul float %901, %1705, !dbg !34 + %2299 = fmul float %902, %1706, !dbg !34 + %2300 = fmul float %903, %1707, !dbg !34 + %2301 = fmul float %904, %1708, !dbg !34 + %2302 = fmul float %905, %1709, !dbg !34 + %2303 = fmul float %906, %1710, !dbg !34 + %2304 = fmul float %907, %1711, !dbg !34 + %2305 = fmul float %908, %1712, !dbg !34 + %2306 = fmul float %909, %1713, !dbg !34 + %2307 = fmul float %910, %1714, !dbg !34 + %2308 = fmul float %911, %1715, !dbg !34 + %2309 = fmul float %912, %1716, !dbg !34 + %2310 = fmul float %913, %1717, !dbg !34 + %2311 = fmul float %914, %1718, !dbg !34 + %2312 = fmul float %915, %1719, !dbg !34 + %2313 = fmul float %916, %1720, !dbg !34 + %2314 = fmul float %917, %1721, !dbg !34 + %2315 = fmul float %918, %1722, !dbg !34 + %2316 = fmul float %919, %1723, !dbg !34 + %2317 = fmul float %920, %1724, !dbg !34 + %2318 = fmul float %921, %1725, !dbg !34 + %2319 = fmul float %922, %1726, !dbg !34 + %2320 = fmul float %923, %1727, !dbg !34 + %2321 = fmul float %924, %1728, !dbg !34 + %2322 = fmul float %925, %1729, !dbg !34 + %2323 = fmul float %926, %1730, !dbg !34 + %2324 = fmul float %927, %1731, !dbg !34 + %2325 = fmul float %928, %1732, !dbg !34 + %2326 = fmul float %929, %1733, !dbg !34 + %2327 = fmul float %930, %1734, !dbg !34 + %2328 = fmul float %931, %1735, !dbg !34 + %2329 = fmul float %932, %1736, !dbg !34 + %2330 = fmul float %933, %1737, !dbg !34 + %2331 = fmul float %934, %1738, !dbg !34 + %2332 = fmul float %935, %1739, !dbg !34 + %2333 = fmul float %936, %1740, !dbg !34 + %2334 = fmul float %937, %1741, !dbg !34 + %2335 = fmul float %938, %1742, !dbg !34 + %2336 = fmul float %939, %1743, !dbg !34 + %2337 = fmul float %940, %1744, !dbg !34 + %2338 = fmul float %941, %1745, !dbg !34 + %2339 = fmul float %942, %1746, !dbg !34 + %2340 = fmul float %943, %1747, !dbg !34 + %2341 = fmul float %944, %1748, !dbg !34 + %2342 = fmul float %945, %1749, !dbg !34 + %2343 = fmul float %946, %1750, !dbg !34 + %2344 = fmul float %947, %1751, !dbg !34 + %2345 = fmul float %948, %1752, !dbg !34 + %2346 = fmul float %949, %1753, !dbg !34 + %2347 = fmul float %950, %1754, !dbg !34 + %2348 = fmul float %951, %1755, !dbg !34 + %2349 = fmul float %952, %1756, !dbg !34 + %2350 = fmul float %953, %1757, !dbg !34 + %2351 = fmul float %954, %1758, !dbg !34 + %2352 = fmul float %955, %1759, !dbg !34 + %2353 = fmul float %956, %1760, !dbg !34 + %2354 = fmul float %957, %1761, !dbg !34 + %2355 = fmul float %958, %1762, !dbg !34 + %2356 = fmul float %959, %1763, !dbg !34 + %2357 = fmul float %960, %1764, !dbg !34 + %2358 = fmul float %961, %1765, !dbg !34 + %2359 = fmul float %962, %1766, !dbg !34 + %2360 = fmul float %963, %1767, !dbg !34 + %2361 = fmul float %964, %1768, !dbg !34 + %2362 = fmul float %965, %1769, !dbg !34 + %2363 = fmul float %966, %1770, !dbg !34 + %2364 = fmul float %967, %1771, !dbg !34 + %2365 = fmul float %968, %1772, !dbg !34 + %2366 = fmul float %969, %1773, !dbg !34 + %2367 = fmul float %970, %1774, !dbg !34 + %2368 = fmul float %971, %1775, !dbg !34 + %2369 = fmul float %972, %1776, !dbg !34 + %2370 = fmul float %973, %1777, !dbg !34 + %2371 = fmul float %974, %1778, !dbg !34 + %2372 = fmul float %975, %1779, !dbg !34 + %2373 = fmul float %976, %1780, !dbg !34 + %2374 = fmul float %977, %1781, !dbg !34 + %2375 = fmul float %978, %1782, !dbg !34 + %2376 = fmul float %979, %1783, !dbg !34 + %2377 = fmul float %980, %1784, !dbg !34 + %2378 = fmul float %981, %1785, !dbg !34 + %2379 = fmul float %982, %1786, !dbg !34 + %2380 = fmul float %983, %1787, !dbg !34 + %2381 = fmul float %984, %1788, !dbg !34 + %2382 = fmul float %985, %1789, !dbg !34 + %2383 = fmul float %986, %1790, !dbg !34 + %2384 = fmul float %987, %1791, !dbg !34 + %2385 = fmul float %988, %1792, !dbg !34 + %2386 = fmul float %989, %1793, !dbg !34 + %2387 = fmul float %990, %1794, !dbg !34 + %2388 = fmul float %991, %1795, !dbg !34 + %2389 = fmul float %992, %1796, !dbg !34 + %2390 = fmul float %993, %1797, !dbg !34 + %2391 = fmul float %994, %1798, !dbg !34 + %2392 = fmul float %995, %1799, !dbg !34 + %2393 = fmul float %996, %1800, !dbg !34 + %2394 = fmul float %997, %1801, !dbg !34 + %2395 = fmul float %998, %1802, !dbg !34 + %2396 = fmul float %999, %1803, !dbg !34 + %2397 = fmul float %1000, %1804, !dbg !34 + %2398 = fmul float %1001, %1805, !dbg !34 + %2399 = fmul float %1002, %1806, !dbg !34 + %2400 = fmul float %1003, %1807, !dbg !34 + %2401 = fmul float %1004, %1808, !dbg !34 + %2402 = fmul float %1005, %1809, !dbg !34 + %2403 = fmul float %1006, %1810, !dbg !34 + %2404 = fmul float %1007, %1811, !dbg !34 + %2405 = fmul float %1008, %1812, !dbg !34 + %2406 = fmul float %1009, %1813, !dbg !34 + %2407 = fmul float %1010, %1814, !dbg !34 + %2408 = fmul float %1011, %1815, !dbg !34 + %2409 = fmul float %1012, %1816, !dbg !34 + %2410 = fmul float %1013, %1817, !dbg !34 + %2411 = fmul float %1014, %1818, !dbg !34 + %2412 = fmul float %1015, %1819, !dbg !34 + %2413 = fmul float %1016, %1820, !dbg !34 + %2414 = fmul float %1017, %1821, !dbg !34 + %2415 = fmul float %1018, %1822, !dbg !34 + %2416 = fmul float %1019, %1823, !dbg !34 + %2417 = fmul float %1020, %1824, !dbg !34 + %2418 = fmul float %1021, %1825, !dbg !34 + %2419 = fmul float %1022, %1826, !dbg !34 + %2420 = fmul float %1023, %1827, !dbg !34 + %2421 = fmul float %1024, %1828, !dbg !34 + %2422 = fmul float %1025, %1829, !dbg !34 + %2423 = fmul float %1026, %1830, !dbg !34 + %2424 = fmul float %1027, %1831, !dbg !34 + %2425 = fmul float %1028, %1832, !dbg !34 + %2426 = fmul float %1029, %1833, !dbg !34 + %2427 = fmul float %1030, %1834, !dbg !34 + %2428 = fmul float %1031, %1835, !dbg !34 + %2429 = fmul float %1032, %1836, !dbg !34 + %2430 = fmul float %1033, %1837, !dbg !34 + %2431 = fmul float %1034, %1838, !dbg !34 + %2432 = fmul float %1035, %1839, !dbg !34 + %2433 = fmul float %1036, %1840, !dbg !34 + %2434 = fmul float %1037, %1841, !dbg !34 + %2435 = fmul float %1038, %1842, !dbg !34 + %2436 = fmul float %1039, %1843, !dbg !34 + %2437 = fmul float %1040, %1844, !dbg !34 + %2438 = fmul float %1041, %1845, !dbg !34 + %2439 = fmul float %1042, %1846, !dbg !34 + %2440 = fmul float %1043, %1847, !dbg !34 + %2441 = fmul float %1044, %1848, !dbg !34 + %2442 = fmul float %1045, %1849, !dbg !34 + %2443 = fmul float %1046, %1850, !dbg !34 + %2444 = fmul float %1047, %1851, !dbg !34 + %2445 = fmul float %1048, %1852, !dbg !34 + %2446 = fmul float %1049, %1853, !dbg !34 + %2447 = fmul float %1050, %1854, !dbg !34 + %2448 = fmul float %1051, %1855, !dbg !34 + %2449 = fmul float %1052, %1856, !dbg !34 + %2450 = fmul float %1053, %1857, !dbg !34 + %2451 = fmul float %1054, %1858, !dbg !34 + %2452 = fmul float %1055, %1859, !dbg !34 + %2453 = fmul float %1056, %1860, !dbg !34 + %2454 = fmul float %1057, %1861, !dbg !34 + %2455 = fmul float %1058, %1862, !dbg !34 + %2456 = fmul float %1059, %1863, !dbg !34 + %2457 = fmul float %1060, %1864, !dbg !34 + %2458 = fmul float %1061, %1865, !dbg !34 + %2459 = fmul float %1062, %1866, !dbg !34 + %2460 = fmul float %1063, %1867, !dbg !34 + %2461 = fmul float %1064, %1868, !dbg !34 + %2462 = fmul float %1065, %1869, !dbg !34 + %2463 = fmul float %1066, %1870, !dbg !34 + %2464 = fmul float %1067, %1871, !dbg !34 + %2465 = fmul float %1068, %1872, !dbg !34 + %2466 = fmul float %1069, %1873, !dbg !34 + %2467 = fmul float %1070, %1874, !dbg !34 + %2468 = fmul float %1071, %1875, !dbg !34 + %2469 = fmul float %1072, %1876, !dbg !34 + %2470 = fmul float %1073, %1877, !dbg !34 + %2471 = fmul float %1074, %1878, !dbg !34 + %2472 = fmul float %1075, %1879, !dbg !34 + %2473 = fmul float %1076, %1880, !dbg !34 + %2474 = fmul float %1077, %1881, !dbg !34 + %2475 = fmul float %1078, %1882, !dbg !34 + %2476 = fmul float %1079, %1883, !dbg !34 + %2477 = fmul float %1080, %1884, !dbg !34 + %2478 = fmul float %1081, %1885, !dbg !34 + %2479 = fmul float %1082, %1886, !dbg !34 + %2480 = fmul float %1083, %1887, !dbg !34 + %2481 = fmul float %1084, %1888, !dbg !34 + %2482 = fmul float %1085, %1889, !dbg !34 + %2483 = fmul float %1086, %1890, !dbg !34 + %2484 = fmul float %1087, %1891, !dbg !34 + %2485 = fmul float %1088, %1892, !dbg !34 + %2486 = fmul float %1089, %1893, !dbg !34 + %2487 = fmul float %1090, %1894, !dbg !34 + %2488 = fmul float %1091, %1895, !dbg !34 + %2489 = fmul float %1092, %1896, !dbg !34 + %2490 = fmul float %1093, %1897, !dbg !34 + %2491 = fmul float %1094, %1898, !dbg !34 + %2492 = fmul float %1095, %1899, !dbg !34 + %2493 = fmul float %1096, %1900, !dbg !34 + %2494 = fmul float %1097, %1901, !dbg !34 + %2495 = fmul float %1098, %1902, !dbg !34 + %2496 = fmul float %1099, %1903, !dbg !34 + %2497 = fmul float %1100, %1904, !dbg !34 + %2498 = fmul float %1101, %1905, !dbg !34 + %2499 = fmul float %1102, %1906, !dbg !34 + %2500 = fmul float %1103, %1907, !dbg !34 + %2501 = fmul float %1104, %1908, !dbg !34 + %2502 = fmul float %1105, %1909, !dbg !34 + %2503 = fmul float %1106, %1910, !dbg !34 + %2504 = fmul float %1107, %1911, !dbg !34 + %2505 = fmul float %1108, %1912, !dbg !34 + %2506 = fmul float %1109, %1913, !dbg !34 + %2507 = fmul float %1110, %1914, !dbg !34 + %2508 = fmul float %1111, %1915, !dbg !34 + %2509 = fmul float %1112, %1916, !dbg !34 + %2510 = fmul float %1113, %1917, !dbg !34 + %2511 = fmul float %1114, %1918, !dbg !34 + %2512 = fmul float %1115, %1919, !dbg !34 + %2513 = fmul float %1116, %1920, !dbg !34 + %2514 = fmul float %1117, %1921, !dbg !34 + %2515 = fmul float %1118, %1922, !dbg !34 + %2516 = fmul float %1119, %1923, !dbg !34 + %2517 = fmul float %1120, %1924, !dbg !34 + %2518 = fmul float %1121, %1925, !dbg !34 + %2519 = fmul float %1122, %1926, !dbg !34 + %2520 = fmul float %1123, %1927, !dbg !34 + %2521 = fmul float %1124, %1928, !dbg !34 + %2522 = fmul float %1125, %1929, !dbg !34 + %2523 = fmul float %1126, %1930, !dbg !34 + %2524 = fmul float %1127, %1931, !dbg !34 + %2525 = fmul float %1128, %1932, !dbg !34 + %2526 = fmul float %1129, %1933, !dbg !34 + %2527 = fmul float %1130, %1934, !dbg !34 + %2528 = fmul float %1131, %1935, !dbg !34 + %2529 = fmul float %1132, %1936, !dbg !34 + %2530 = fmul float %1133, %1937, !dbg !34 + %2531 = fmul float %1134, %1938, !dbg !34 + %2532 = fmul float %1135, %1939, !dbg !34 + %2533 = fmul float %1136, %1940, !dbg !34 + %2534 = fmul float %1137, %1941, !dbg !34 + %2535 = fmul float %1138, %1942, !dbg !34 + %2536 = fmul float %1139, %1943, !dbg !34 + %2537 = fmul float %1140, %1944, !dbg !34 + %2538 = fmul float %1141, %1945, !dbg !34 + %2539 = fmul float %1142, %1946, !dbg !34 + %2540 = fmul float %1143, %1947, !dbg !34 + %2541 = fadd float %2285, %2286, !dbg !35 + %2542 = fadd float %2287, %2541, !dbg !35 + %2543 = fadd float %2288, %2542, !dbg !35 + %2544 = fadd float %2289, %2543, !dbg !35 + %2545 = fadd float %2290, %2544, !dbg !35 + %2546 = fadd float %2291, %2545, !dbg !35 + %2547 = fadd float %2292, %2546, !dbg !35 + %2548 = fadd float %2293, %2294, !dbg !35 + %2549 = fadd float %2295, %2548, !dbg !35 + %2550 = fadd float %2296, %2549, !dbg !35 + %2551 = fadd float %2297, %2550, !dbg !35 + %2552 = fadd float %2298, %2551, !dbg !35 + %2553 = fadd float %2299, %2552, !dbg !35 + %2554 = fadd float %2300, %2553, !dbg !35 + %2555 = fadd float %2301, %2302, !dbg !35 + %2556 = fadd float %2303, %2555, !dbg !35 + %2557 = fadd float %2304, %2556, !dbg !35 + %2558 = fadd float %2305, %2557, !dbg !35 + %2559 = fadd float %2306, %2558, !dbg !35 + %2560 = fadd float %2307, %2559, !dbg !35 + %2561 = fadd float %2308, %2560, !dbg !35 + %2562 = fadd float %2309, %2310, !dbg !35 + %2563 = fadd float %2311, %2562, !dbg !35 + %2564 = fadd float %2312, %2563, !dbg !35 + %2565 = fadd float %2313, %2564, !dbg !35 + %2566 = fadd float %2314, %2565, !dbg !35 + %2567 = fadd float %2315, %2566, !dbg !35 + %2568 = fadd float %2316, %2567, !dbg !35 + %2569 = fadd float %2317, %2318, !dbg !35 + %2570 = fadd float %2319, %2569, !dbg !35 + %2571 = fadd float %2320, %2570, !dbg !35 + %2572 = fadd float %2321, %2571, !dbg !35 + %2573 = fadd float %2322, %2572, !dbg !35 + %2574 = fadd float %2323, %2573, !dbg !35 + %2575 = fadd float %2324, %2574, !dbg !35 + %2576 = fadd float %2325, %2326, !dbg !35 + %2577 = fadd float %2327, %2576, !dbg !35 + %2578 = fadd float %2328, %2577, !dbg !35 + %2579 = fadd float %2329, %2578, !dbg !35 + %2580 = fadd float %2330, %2579, !dbg !35 + %2581 = fadd float %2331, %2580, !dbg !35 + %2582 = fadd float %2332, %2581, !dbg !35 + %2583 = fadd float %2333, %2334, !dbg !35 + %2584 = fadd float %2335, %2583, !dbg !35 + %2585 = fadd float %2336, %2584, !dbg !35 + %2586 = fadd float %2337, %2585, !dbg !35 + %2587 = fadd float %2338, %2586, !dbg !35 + %2588 = fadd float %2339, %2587, !dbg !35 + %2589 = fadd float %2340, %2588, !dbg !35 + %2590 = fadd float %2341, %2342, !dbg !35 + %2591 = fadd float %2343, %2590, !dbg !35 + %2592 = fadd float %2344, %2591, !dbg !35 + %2593 = fadd float %2345, %2592, !dbg !35 + %2594 = fadd float %2346, %2593, !dbg !35 + %2595 = fadd float %2347, %2594, !dbg !35 + %2596 = fadd float %2348, %2595, !dbg !35 + %2597 = fadd float %2349, %2350, !dbg !35 + %2598 = fadd float %2351, %2597, !dbg !35 + %2599 = fadd float %2352, %2598, !dbg !35 + %2600 = fadd float %2353, %2599, !dbg !35 + %2601 = fadd float %2354, %2600, !dbg !35 + %2602 = fadd float %2355, %2601, !dbg !35 + %2603 = fadd float %2356, %2602, !dbg !35 + %2604 = fadd float %2357, %2358, !dbg !35 + %2605 = fadd float %2359, %2604, !dbg !35 + %2606 = fadd float %2360, %2605, !dbg !35 + %2607 = fadd float %2361, %2606, !dbg !35 + %2608 = fadd float %2362, %2607, !dbg !35 + %2609 = fadd float %2363, %2608, !dbg !35 + %2610 = fadd float %2364, %2609, !dbg !35 + %2611 = fadd float %2365, %2366, !dbg !35 + %2612 = fadd float %2367, %2611, !dbg !35 + %2613 = fadd float %2368, %2612, !dbg !35 + %2614 = fadd float %2369, %2613, !dbg !35 + %2615 = fadd float %2370, %2614, !dbg !35 + %2616 = fadd float %2371, %2615, !dbg !35 + %2617 = fadd float %2372, %2616, !dbg !35 + %2618 = fadd float %2373, %2374, !dbg !35 + %2619 = fadd float %2375, %2618, !dbg !35 + %2620 = fadd float %2376, %2619, !dbg !35 + %2621 = fadd float %2377, %2620, !dbg !35 + %2622 = fadd float %2378, %2621, !dbg !35 + %2623 = fadd float %2379, %2622, !dbg !35 + %2624 = fadd float %2380, %2623, !dbg !35 + %2625 = fadd float %2381, %2382, !dbg !35 + %2626 = fadd float %2383, %2625, !dbg !35 + %2627 = fadd float %2384, %2626, !dbg !35 + %2628 = fadd float %2385, %2627, !dbg !35 + %2629 = fadd float %2386, %2628, !dbg !35 + %2630 = fadd float %2387, %2629, !dbg !35 + %2631 = fadd float %2388, %2630, !dbg !35 + %2632 = fadd float %2389, %2390, !dbg !35 + %2633 = fadd float %2391, %2632, !dbg !35 + %2634 = fadd float %2392, %2633, !dbg !35 + %2635 = fadd float %2393, %2634, !dbg !35 + %2636 = fadd float %2394, %2635, !dbg !35 + %2637 = fadd float %2395, %2636, !dbg !35 + %2638 = fadd float %2396, %2637, !dbg !35 + %2639 = fadd float %2397, %2398, !dbg !35 + %2640 = fadd float %2399, %2639, !dbg !35 + %2641 = fadd float %2400, %2640, !dbg !35 + %2642 = fadd float %2401, %2641, !dbg !35 + %2643 = fadd float %2402, %2642, !dbg !35 + %2644 = fadd float %2403, %2643, !dbg !35 + %2645 = fadd float %2404, %2644, !dbg !35 + %2646 = fadd float %2405, %2406, !dbg !35 + %2647 = fadd float %2407, %2646, !dbg !35 + %2648 = fadd float %2408, %2647, !dbg !35 + %2649 = fadd float %2409, %2648, !dbg !35 + %2650 = fadd float %2410, %2649, !dbg !35 + %2651 = fadd float %2411, %2650, !dbg !35 + %2652 = fadd float %2412, %2651, !dbg !35 + %2653 = fadd float %2413, %2414, !dbg !35 + %2654 = fadd float %2415, %2653, !dbg !35 + %2655 = fadd float %2416, %2654, !dbg !35 + %2656 = fadd float %2417, %2655, !dbg !35 + %2657 = fadd float %2418, %2656, !dbg !35 + %2658 = fadd float %2419, %2657, !dbg !35 + %2659 = fadd float %2420, %2658, !dbg !35 + %2660 = fadd float %2421, %2422, !dbg !35 + %2661 = fadd float %2423, %2660, !dbg !35 + %2662 = fadd float %2424, %2661, !dbg !35 + %2663 = fadd float %2425, %2662, !dbg !35 + %2664 = fadd float %2426, %2663, !dbg !35 + %2665 = fadd float %2427, %2664, !dbg !35 + %2666 = fadd float %2428, %2665, !dbg !35 + %2667 = fadd float %2429, %2430, !dbg !35 + %2668 = fadd float %2431, %2667, !dbg !35 + %2669 = fadd float %2432, %2668, !dbg !35 + %2670 = fadd float %2433, %2669, !dbg !35 + %2671 = fadd float %2434, %2670, !dbg !35 + %2672 = fadd float %2435, %2671, !dbg !35 + %2673 = fadd float %2436, %2672, !dbg !35 + %2674 = fadd float %2437, %2438, !dbg !35 + %2675 = fadd float %2439, %2674, !dbg !35 + %2676 = fadd float %2440, %2675, !dbg !35 + %2677 = fadd float %2441, %2676, !dbg !35 + %2678 = fadd float %2442, %2677, !dbg !35 + %2679 = fadd float %2443, %2678, !dbg !35 + %2680 = fadd float %2444, %2679, !dbg !35 + %2681 = fadd float %2445, %2446, !dbg !35 + %2682 = fadd float %2447, %2681, !dbg !35 + %2683 = fadd float %2448, %2682, !dbg !35 + %2684 = fadd float %2449, %2683, !dbg !35 + %2685 = fadd float %2450, %2684, !dbg !35 + %2686 = fadd float %2451, %2685, !dbg !35 + %2687 = fadd float %2452, %2686, !dbg !35 + %2688 = fadd float %2453, %2454, !dbg !35 + %2689 = fadd float %2455, %2688, !dbg !35 + %2690 = fadd float %2456, %2689, !dbg !35 + %2691 = fadd float %2457, %2690, !dbg !35 + %2692 = fadd float %2458, %2691, !dbg !35 + %2693 = fadd float %2459, %2692, !dbg !35 + %2694 = fadd float %2460, %2693, !dbg !35 + %2695 = fadd float %2461, %2462, !dbg !35 + %2696 = fadd float %2463, %2695, !dbg !35 + %2697 = fadd float %2464, %2696, !dbg !35 + %2698 = fadd float %2465, %2697, !dbg !35 + %2699 = fadd float %2466, %2698, !dbg !35 + %2700 = fadd float %2467, %2699, !dbg !35 + %2701 = fadd float %2468, %2700, !dbg !35 + %2702 = fadd float %2469, %2470, !dbg !35 + %2703 = fadd float %2471, %2702, !dbg !35 + %2704 = fadd float %2472, %2703, !dbg !35 + %2705 = fadd float %2473, %2704, !dbg !35 + %2706 = fadd float %2474, %2705, !dbg !35 + %2707 = fadd float %2475, %2706, !dbg !35 + %2708 = fadd float %2476, %2707, !dbg !35 + %2709 = fadd float %2477, %2478, !dbg !35 + %2710 = fadd float %2479, %2709, !dbg !35 + %2711 = fadd float %2480, %2710, !dbg !35 + %2712 = fadd float %2481, %2711, !dbg !35 + %2713 = fadd float %2482, %2712, !dbg !35 + %2714 = fadd float %2483, %2713, !dbg !35 + %2715 = fadd float %2484, %2714, !dbg !35 + %2716 = fadd float %2485, %2486, !dbg !35 + %2717 = fadd float %2487, %2716, !dbg !35 + %2718 = fadd float %2488, %2717, !dbg !35 + %2719 = fadd float %2489, %2718, !dbg !35 + %2720 = fadd float %2490, %2719, !dbg !35 + %2721 = fadd float %2491, %2720, !dbg !35 + %2722 = fadd float %2492, %2721, !dbg !35 + %2723 = fadd float %2493, %2494, !dbg !35 + %2724 = fadd float %2495, %2723, !dbg !35 + %2725 = fadd float %2496, %2724, !dbg !35 + %2726 = fadd float %2497, %2725, !dbg !35 + %2727 = fadd float %2498, %2726, !dbg !35 + %2728 = fadd float %2499, %2727, !dbg !35 + %2729 = fadd float %2500, %2728, !dbg !35 + %2730 = fadd float %2501, %2502, !dbg !35 + %2731 = fadd float %2503, %2730, !dbg !35 + %2732 = fadd float %2504, %2731, !dbg !35 + %2733 = fadd float %2505, %2732, !dbg !35 + %2734 = fadd float %2506, %2733, !dbg !35 + %2735 = fadd float %2507, %2734, !dbg !35 + %2736 = fadd float %2508, %2735, !dbg !35 + %2737 = fadd float %2509, %2510, !dbg !35 + %2738 = fadd float %2511, %2737, !dbg !35 + %2739 = fadd float %2512, %2738, !dbg !35 + %2740 = fadd float %2513, %2739, !dbg !35 + %2741 = fadd float %2514, %2740, !dbg !35 + %2742 = fadd float %2515, %2741, !dbg !35 + %2743 = fadd float %2516, %2742, !dbg !35 + %2744 = fadd float %2517, %2518, !dbg !35 + %2745 = fadd float %2519, %2744, !dbg !35 + %2746 = fadd float %2520, %2745, !dbg !35 + %2747 = fadd float %2521, %2746, !dbg !35 + %2748 = fadd float %2522, %2747, !dbg !35 + %2749 = fadd float %2523, %2748, !dbg !35 + %2750 = fadd float %2524, %2749, !dbg !35 + %2751 = fadd float %2525, %2526, !dbg !35 + %2752 = fadd float %2527, %2751, !dbg !35 + %2753 = fadd float %2528, %2752, !dbg !35 + %2754 = fadd float %2529, %2753, !dbg !35 + %2755 = fadd float %2530, %2754, !dbg !35 + %2756 = fadd float %2531, %2755, !dbg !35 + %2757 = fadd float %2532, %2756, !dbg !35 + %2758 = fadd float %2533, %2534, !dbg !35 + %2759 = fadd float %2535, %2758, !dbg !35 + %2760 = fadd float %2536, %2759, !dbg !35 + %2761 = fadd float %2537, %2760, !dbg !35 + %2762 = fadd float %2538, %2761, !dbg !35 + %2763 = fadd float %2539, %2762, !dbg !35 + %2764 = fadd float %2540, %2763, !dbg !35 + %2765 = bitcast float %2547 to i32, !dbg !38 + %2766 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2765, i32 8, i32 31), !dbg !38 + %2767 = bitcast i32 %2766 to float, !dbg !38 + %2768 = fadd float %2547, %2767, !dbg !35 + %2769 = bitcast float %2768 to i32, !dbg !38 + %2770 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2769, i32 4, i32 31), !dbg !38 + %2771 = bitcast i32 %2770 to float, !dbg !38 + %2772 = fadd float %2768, %2771, !dbg !35 + %2773 = bitcast float %2772 to i32, !dbg !38 + %2774 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2773, i32 2, i32 31), !dbg !38 + %2775 = bitcast i32 %2774 to float, !dbg !38 + %2776 = fadd float %2772, %2775, !dbg !35 + %2777 = bitcast float %2776 to i32, !dbg !38 + %2778 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2777, i32 1, i32 31), !dbg !38 + %2779 = bitcast i32 %2778 to float, !dbg !38 + %2780 = fadd float %2776, %2779, !dbg !35 + %2781 = bitcast float %2554 to i32, !dbg !38 + %2782 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2781, i32 8, i32 31), !dbg !38 + %2783 = bitcast i32 %2782 to float, !dbg !38 + %2784 = fadd float %2554, %2783, !dbg !35 + %2785 = bitcast float %2784 to i32, !dbg !38 + %2786 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2785, i32 4, i32 31), !dbg !38 + %2787 = bitcast i32 %2786 to float, !dbg !38 + %2788 = fadd float %2784, %2787, !dbg !35 + %2789 = bitcast float %2788 to i32, !dbg !38 + %2790 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2789, i32 2, i32 31), !dbg !38 + %2791 = bitcast i32 %2790 to float, !dbg !38 + %2792 = fadd float %2788, %2791, !dbg !35 + %2793 = bitcast float %2792 to i32, !dbg !38 + %2794 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2793, i32 1, i32 31), !dbg !38 + %2795 = bitcast i32 %2794 to float, !dbg !38 + %2796 = fadd float %2792, %2795, !dbg !35 + %2797 = bitcast float %2561 to i32, !dbg !38 + %2798 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2797, i32 8, i32 31), !dbg !38 + %2799 = bitcast i32 %2798 to float, !dbg !38 + %2800 = fadd float %2561, %2799, !dbg !35 + %2801 = bitcast float %2800 to i32, !dbg !38 + %2802 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2801, i32 4, i32 31), !dbg !38 + %2803 = bitcast i32 %2802 to float, !dbg !38 + %2804 = fadd float %2800, %2803, !dbg !35 + %2805 = bitcast float %2804 to i32, !dbg !38 + %2806 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2805, i32 2, i32 31), !dbg !38 + %2807 = bitcast i32 %2806 to float, !dbg !38 + %2808 = fadd float %2804, %2807, !dbg !35 + %2809 = bitcast float %2808 to i32, !dbg !38 + %2810 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2809, i32 1, i32 31), !dbg !38 + %2811 = bitcast i32 %2810 to float, !dbg !38 + %2812 = fadd float %2808, %2811, !dbg !35 + %2813 = bitcast float %2568 to i32, !dbg !38 + %2814 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2813, i32 8, i32 31), !dbg !38 + %2815 = bitcast i32 %2814 to float, !dbg !38 + %2816 = fadd float %2568, %2815, !dbg !35 + %2817 = bitcast float %2816 to i32, !dbg !38 + %2818 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2817, i32 4, i32 31), !dbg !38 + %2819 = bitcast i32 %2818 to float, !dbg !38 + %2820 = fadd float %2816, %2819, !dbg !35 + %2821 = bitcast float %2820 to i32, !dbg !38 + %2822 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2821, i32 2, i32 31), !dbg !38 + %2823 = bitcast i32 %2822 to float, !dbg !38 + %2824 = fadd float %2820, %2823, !dbg !35 + %2825 = bitcast float %2824 to i32, !dbg !38 + %2826 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2825, i32 1, i32 31), !dbg !38 + %2827 = bitcast i32 %2826 to float, !dbg !38 + %2828 = fadd float %2824, %2827, !dbg !35 + %2829 = bitcast float %2575 to i32, !dbg !38 + %2830 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2829, i32 8, i32 31), !dbg !38 + %2831 = bitcast i32 %2830 to float, !dbg !38 + %2832 = fadd float %2575, %2831, !dbg !35 + %2833 = bitcast float %2832 to i32, !dbg !38 + %2834 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2833, i32 4, i32 31), !dbg !38 + %2835 = bitcast i32 %2834 to float, !dbg !38 + %2836 = fadd float %2832, %2835, !dbg !35 + %2837 = bitcast float %2836 to i32, !dbg !38 + %2838 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2837, i32 2, i32 31), !dbg !38 + %2839 = bitcast i32 %2838 to float, !dbg !38 + %2840 = fadd float %2836, %2839, !dbg !35 + %2841 = bitcast float %2840 to i32, !dbg !38 + %2842 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2841, i32 1, i32 31), !dbg !38 + %2843 = bitcast i32 %2842 to float, !dbg !38 + %2844 = fadd float %2840, %2843, !dbg !35 + %2845 = bitcast float %2582 to i32, !dbg !38 + %2846 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2845, i32 8, i32 31), !dbg !38 + %2847 = bitcast i32 %2846 to float, !dbg !38 + %2848 = fadd float %2582, %2847, !dbg !35 + %2849 = bitcast float %2848 to i32, !dbg !38 + %2850 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2849, i32 4, i32 31), !dbg !38 + %2851 = bitcast i32 %2850 to float, !dbg !38 + %2852 = fadd float %2848, %2851, !dbg !35 + %2853 = bitcast float %2852 to i32, !dbg !38 + %2854 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2853, i32 2, i32 31), !dbg !38 + %2855 = bitcast i32 %2854 to float, !dbg !38 + %2856 = fadd float %2852, %2855, !dbg !35 + %2857 = bitcast float %2856 to i32, !dbg !38 + %2858 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2857, i32 1, i32 31), !dbg !38 + %2859 = bitcast i32 %2858 to float, !dbg !38 + %2860 = fadd float %2856, %2859, !dbg !35 + %2861 = bitcast float %2589 to i32, !dbg !38 + %2862 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2861, i32 8, i32 31), !dbg !38 + %2863 = bitcast i32 %2862 to float, !dbg !38 + %2864 = fadd float %2589, %2863, !dbg !35 + %2865 = bitcast float %2864 to i32, !dbg !38 + %2866 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2865, i32 4, i32 31), !dbg !38 + %2867 = bitcast i32 %2866 to float, !dbg !38 + %2868 = fadd float %2864, %2867, !dbg !35 + %2869 = bitcast float %2868 to i32, !dbg !38 + %2870 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2869, i32 2, i32 31), !dbg !38 + %2871 = bitcast i32 %2870 to float, !dbg !38 + %2872 = fadd float %2868, %2871, !dbg !35 + %2873 = bitcast float %2872 to i32, !dbg !38 + %2874 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2873, i32 1, i32 31), !dbg !38 + %2875 = bitcast i32 %2874 to float, !dbg !38 + %2876 = fadd float %2872, %2875, !dbg !35 + %2877 = bitcast float %2596 to i32, !dbg !38 + %2878 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2877, i32 8, i32 31), !dbg !38 + %2879 = bitcast i32 %2878 to float, !dbg !38 + %2880 = fadd float %2596, %2879, !dbg !35 + %2881 = bitcast float %2880 to i32, !dbg !38 + %2882 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2881, i32 4, i32 31), !dbg !38 + %2883 = bitcast i32 %2882 to float, !dbg !38 + %2884 = fadd float %2880, %2883, !dbg !35 + %2885 = bitcast float %2884 to i32, !dbg !38 + %2886 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2885, i32 2, i32 31), !dbg !38 + %2887 = bitcast i32 %2886 to float, !dbg !38 + %2888 = fadd float %2884, %2887, !dbg !35 + %2889 = bitcast float %2888 to i32, !dbg !38 + %2890 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2889, i32 1, i32 31), !dbg !38 + %2891 = bitcast i32 %2890 to float, !dbg !38 + %2892 = fadd float %2888, %2891, !dbg !35 + %2893 = bitcast float %2603 to i32, !dbg !38 + %2894 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2893, i32 8, i32 31), !dbg !38 + %2895 = bitcast i32 %2894 to float, !dbg !38 + %2896 = fadd float %2603, %2895, !dbg !35 + %2897 = bitcast float %2896 to i32, !dbg !38 + %2898 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2897, i32 4, i32 31), !dbg !38 + %2899 = bitcast i32 %2898 to float, !dbg !38 + %2900 = fadd float %2896, %2899, !dbg !35 + %2901 = bitcast float %2900 to i32, !dbg !38 + %2902 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2901, i32 2, i32 31), !dbg !38 + %2903 = bitcast i32 %2902 to float, !dbg !38 + %2904 = fadd float %2900, %2903, !dbg !35 + %2905 = bitcast float %2904 to i32, !dbg !38 + %2906 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2905, i32 1, i32 31), !dbg !38 + %2907 = bitcast i32 %2906 to float, !dbg !38 + %2908 = fadd float %2904, %2907, !dbg !35 + %2909 = bitcast float %2610 to i32, !dbg !38 + %2910 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2909, i32 8, i32 31), !dbg !38 + %2911 = bitcast i32 %2910 to float, !dbg !38 + %2912 = fadd float %2610, %2911, !dbg !35 + %2913 = bitcast float %2912 to i32, !dbg !38 + %2914 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2913, i32 4, i32 31), !dbg !38 + %2915 = bitcast i32 %2914 to float, !dbg !38 + %2916 = fadd float %2912, %2915, !dbg !35 + %2917 = bitcast float %2916 to i32, !dbg !38 + %2918 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2917, i32 2, i32 31), !dbg !38 + %2919 = bitcast i32 %2918 to float, !dbg !38 + %2920 = fadd float %2916, %2919, !dbg !35 + %2921 = bitcast float %2920 to i32, !dbg !38 + %2922 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2921, i32 1, i32 31), !dbg !38 + %2923 = bitcast i32 %2922 to float, !dbg !38 + %2924 = fadd float %2920, %2923, !dbg !35 + %2925 = bitcast float %2617 to i32, !dbg !38 + %2926 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2925, i32 8, i32 31), !dbg !38 + %2927 = bitcast i32 %2926 to float, !dbg !38 + %2928 = fadd float %2617, %2927, !dbg !35 + %2929 = bitcast float %2928 to i32, !dbg !38 + %2930 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2929, i32 4, i32 31), !dbg !38 + %2931 = bitcast i32 %2930 to float, !dbg !38 + %2932 = fadd float %2928, %2931, !dbg !35 + %2933 = bitcast float %2932 to i32, !dbg !38 + %2934 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2933, i32 2, i32 31), !dbg !38 + %2935 = bitcast i32 %2934 to float, !dbg !38 + %2936 = fadd float %2932, %2935, !dbg !35 + %2937 = bitcast float %2936 to i32, !dbg !38 + %2938 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2937, i32 1, i32 31), !dbg !38 + %2939 = bitcast i32 %2938 to float, !dbg !38 + %2940 = fadd float %2936, %2939, !dbg !35 + %2941 = bitcast float %2624 to i32, !dbg !38 + %2942 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2941, i32 8, i32 31), !dbg !38 + %2943 = bitcast i32 %2942 to float, !dbg !38 + %2944 = fadd float %2624, %2943, !dbg !35 + %2945 = bitcast float %2944 to i32, !dbg !38 + %2946 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2945, i32 4, i32 31), !dbg !38 + %2947 = bitcast i32 %2946 to float, !dbg !38 + %2948 = fadd float %2944, %2947, !dbg !35 + %2949 = bitcast float %2948 to i32, !dbg !38 + %2950 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2949, i32 2, i32 31), !dbg !38 + %2951 = bitcast i32 %2950 to float, !dbg !38 + %2952 = fadd float %2948, %2951, !dbg !35 + %2953 = bitcast float %2952 to i32, !dbg !38 + %2954 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2953, i32 1, i32 31), !dbg !38 + %2955 = bitcast i32 %2954 to float, !dbg !38 + %2956 = fadd float %2952, %2955, !dbg !35 + %2957 = bitcast float %2631 to i32, !dbg !38 + %2958 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2957, i32 8, i32 31), !dbg !38 + %2959 = bitcast i32 %2958 to float, !dbg !38 + %2960 = fadd float %2631, %2959, !dbg !35 + %2961 = bitcast float %2960 to i32, !dbg !38 + %2962 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2961, i32 4, i32 31), !dbg !38 + %2963 = bitcast i32 %2962 to float, !dbg !38 + %2964 = fadd float %2960, %2963, !dbg !35 + %2965 = bitcast float %2964 to i32, !dbg !38 + %2966 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2965, i32 2, i32 31), !dbg !38 + %2967 = bitcast i32 %2966 to float, !dbg !38 + %2968 = fadd float %2964, %2967, !dbg !35 + %2969 = bitcast float %2968 to i32, !dbg !38 + %2970 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2969, i32 1, i32 31), !dbg !38 + %2971 = bitcast i32 %2970 to float, !dbg !38 + %2972 = fadd float %2968, %2971, !dbg !35 + %2973 = bitcast float %2638 to i32, !dbg !38 + %2974 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2973, i32 8, i32 31), !dbg !38 + %2975 = bitcast i32 %2974 to float, !dbg !38 + %2976 = fadd float %2638, %2975, !dbg !35 + %2977 = bitcast float %2976 to i32, !dbg !38 + %2978 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2977, i32 4, i32 31), !dbg !38 + %2979 = bitcast i32 %2978 to float, !dbg !38 + %2980 = fadd float %2976, %2979, !dbg !35 + %2981 = bitcast float %2980 to i32, !dbg !38 + %2982 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2981, i32 2, i32 31), !dbg !38 + %2983 = bitcast i32 %2982 to float, !dbg !38 + %2984 = fadd float %2980, %2983, !dbg !35 + %2985 = bitcast float %2984 to i32, !dbg !38 + %2986 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2985, i32 1, i32 31), !dbg !38 + %2987 = bitcast i32 %2986 to float, !dbg !38 + %2988 = fadd float %2984, %2987, !dbg !35 + %2989 = bitcast float %2645 to i32, !dbg !38 + %2990 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2989, i32 8, i32 31), !dbg !38 + %2991 = bitcast i32 %2990 to float, !dbg !38 + %2992 = fadd float %2645, %2991, !dbg !35 + %2993 = bitcast float %2992 to i32, !dbg !38 + %2994 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2993, i32 4, i32 31), !dbg !38 + %2995 = bitcast i32 %2994 to float, !dbg !38 + %2996 = fadd float %2992, %2995, !dbg !35 + %2997 = bitcast float %2996 to i32, !dbg !38 + %2998 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2997, i32 2, i32 31), !dbg !38 + %2999 = bitcast i32 %2998 to float, !dbg !38 + %3000 = fadd float %2996, %2999, !dbg !35 + %3001 = bitcast float %3000 to i32, !dbg !38 + %3002 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3001, i32 1, i32 31), !dbg !38 + %3003 = bitcast i32 %3002 to float, !dbg !38 + %3004 = fadd float %3000, %3003, !dbg !35 + %3005 = bitcast float %2652 to i32, !dbg !38 + %3006 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3005, i32 8, i32 31), !dbg !38 + %3007 = bitcast i32 %3006 to float, !dbg !38 + %3008 = fadd float %2652, %3007, !dbg !35 + %3009 = bitcast float %3008 to i32, !dbg !38 + %3010 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3009, i32 4, i32 31), !dbg !38 + %3011 = bitcast i32 %3010 to float, !dbg !38 + %3012 = fadd float %3008, %3011, !dbg !35 + %3013 = bitcast float %3012 to i32, !dbg !38 + %3014 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3013, i32 2, i32 31), !dbg !38 + %3015 = bitcast i32 %3014 to float, !dbg !38 + %3016 = fadd float %3012, %3015, !dbg !35 + %3017 = bitcast float %3016 to i32, !dbg !38 + %3018 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3017, i32 1, i32 31), !dbg !38 + %3019 = bitcast i32 %3018 to float, !dbg !38 + %3020 = fadd float %3016, %3019, !dbg !35 + %3021 = bitcast float %2659 to i32, !dbg !38 + %3022 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3021, i32 8, i32 31), !dbg !38 + %3023 = bitcast i32 %3022 to float, !dbg !38 + %3024 = fadd float %2659, %3023, !dbg !35 + %3025 = bitcast float %3024 to i32, !dbg !38 + %3026 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3025, i32 4, i32 31), !dbg !38 + %3027 = bitcast i32 %3026 to float, !dbg !38 + %3028 = fadd float %3024, %3027, !dbg !35 + %3029 = bitcast float %3028 to i32, !dbg !38 + %3030 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3029, i32 2, i32 31), !dbg !38 + %3031 = bitcast i32 %3030 to float, !dbg !38 + %3032 = fadd float %3028, %3031, !dbg !35 + %3033 = bitcast float %3032 to i32, !dbg !38 + %3034 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3033, i32 1, i32 31), !dbg !38 + %3035 = bitcast i32 %3034 to float, !dbg !38 + %3036 = fadd float %3032, %3035, !dbg !35 + %3037 = bitcast float %2666 to i32, !dbg !38 + %3038 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3037, i32 8, i32 31), !dbg !38 + %3039 = bitcast i32 %3038 to float, !dbg !38 + %3040 = fadd float %2666, %3039, !dbg !35 + %3041 = bitcast float %3040 to i32, !dbg !38 + %3042 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3041, i32 4, i32 31), !dbg !38 + %3043 = bitcast i32 %3042 to float, !dbg !38 + %3044 = fadd float %3040, %3043, !dbg !35 + %3045 = bitcast float %3044 to i32, !dbg !38 + %3046 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3045, i32 2, i32 31), !dbg !38 + %3047 = bitcast i32 %3046 to float, !dbg !38 + %3048 = fadd float %3044, %3047, !dbg !35 + %3049 = bitcast float %3048 to i32, !dbg !38 + %3050 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3049, i32 1, i32 31), !dbg !38 + %3051 = bitcast i32 %3050 to float, !dbg !38 + %3052 = fadd float %3048, %3051, !dbg !35 + %3053 = bitcast float %2673 to i32, !dbg !38 + %3054 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3053, i32 8, i32 31), !dbg !38 + %3055 = bitcast i32 %3054 to float, !dbg !38 + %3056 = fadd float %2673, %3055, !dbg !35 + %3057 = bitcast float %3056 to i32, !dbg !38 + %3058 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3057, i32 4, i32 31), !dbg !38 + %3059 = bitcast i32 %3058 to float, !dbg !38 + %3060 = fadd float %3056, %3059, !dbg !35 + %3061 = bitcast float %3060 to i32, !dbg !38 + %3062 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3061, i32 2, i32 31), !dbg !38 + %3063 = bitcast i32 %3062 to float, !dbg !38 + %3064 = fadd float %3060, %3063, !dbg !35 + %3065 = bitcast float %3064 to i32, !dbg !38 + %3066 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3065, i32 1, i32 31), !dbg !38 + %3067 = bitcast i32 %3066 to float, !dbg !38 + %3068 = fadd float %3064, %3067, !dbg !35 + %3069 = bitcast float %2680 to i32, !dbg !38 + %3070 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3069, i32 8, i32 31), !dbg !38 + %3071 = bitcast i32 %3070 to float, !dbg !38 + %3072 = fadd float %2680, %3071, !dbg !35 + %3073 = bitcast float %3072 to i32, !dbg !38 + %3074 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3073, i32 4, i32 31), !dbg !38 + %3075 = bitcast i32 %3074 to float, !dbg !38 + %3076 = fadd float %3072, %3075, !dbg !35 + %3077 = bitcast float %3076 to i32, !dbg !38 + %3078 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3077, i32 2, i32 31), !dbg !38 + %3079 = bitcast i32 %3078 to float, !dbg !38 + %3080 = fadd float %3076, %3079, !dbg !35 + %3081 = bitcast float %3080 to i32, !dbg !38 + %3082 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3081, i32 1, i32 31), !dbg !38 + %3083 = bitcast i32 %3082 to float, !dbg !38 + %3084 = fadd float %3080, %3083, !dbg !35 + %3085 = bitcast float %2687 to i32, !dbg !38 + %3086 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3085, i32 8, i32 31), !dbg !38 + %3087 = bitcast i32 %3086 to float, !dbg !38 + %3088 = fadd float %2687, %3087, !dbg !35 + %3089 = bitcast float %3088 to i32, !dbg !38 + %3090 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3089, i32 4, i32 31), !dbg !38 + %3091 = bitcast i32 %3090 to float, !dbg !38 + %3092 = fadd float %3088, %3091, !dbg !35 + %3093 = bitcast float %3092 to i32, !dbg !38 + %3094 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3093, i32 2, i32 31), !dbg !38 + %3095 = bitcast i32 %3094 to float, !dbg !38 + %3096 = fadd float %3092, %3095, !dbg !35 + %3097 = bitcast float %3096 to i32, !dbg !38 + %3098 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3097, i32 1, i32 31), !dbg !38 + %3099 = bitcast i32 %3098 to float, !dbg !38 + %3100 = fadd float %3096, %3099, !dbg !35 + %3101 = bitcast float %2694 to i32, !dbg !38 + %3102 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3101, i32 8, i32 31), !dbg !38 + %3103 = bitcast i32 %3102 to float, !dbg !38 + %3104 = fadd float %2694, %3103, !dbg !35 + %3105 = bitcast float %3104 to i32, !dbg !38 + %3106 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3105, i32 4, i32 31), !dbg !38 + %3107 = bitcast i32 %3106 to float, !dbg !38 + %3108 = fadd float %3104, %3107, !dbg !35 + %3109 = bitcast float %3108 to i32, !dbg !38 + %3110 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3109, i32 2, i32 31), !dbg !38 + %3111 = bitcast i32 %3110 to float, !dbg !38 + %3112 = fadd float %3108, %3111, !dbg !35 + %3113 = bitcast float %3112 to i32, !dbg !38 + %3114 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3113, i32 1, i32 31), !dbg !38 + %3115 = bitcast i32 %3114 to float, !dbg !38 + %3116 = fadd float %3112, %3115, !dbg !35 + %3117 = bitcast float %2701 to i32, !dbg !38 + %3118 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3117, i32 8, i32 31), !dbg !38 + %3119 = bitcast i32 %3118 to float, !dbg !38 + %3120 = fadd float %2701, %3119, !dbg !35 + %3121 = bitcast float %3120 to i32, !dbg !38 + %3122 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3121, i32 4, i32 31), !dbg !38 + %3123 = bitcast i32 %3122 to float, !dbg !38 + %3124 = fadd float %3120, %3123, !dbg !35 + %3125 = bitcast float %3124 to i32, !dbg !38 + %3126 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3125, i32 2, i32 31), !dbg !38 + %3127 = bitcast i32 %3126 to float, !dbg !38 + %3128 = fadd float %3124, %3127, !dbg !35 + %3129 = bitcast float %3128 to i32, !dbg !38 + %3130 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3129, i32 1, i32 31), !dbg !38 + %3131 = bitcast i32 %3130 to float, !dbg !38 + %3132 = fadd float %3128, %3131, !dbg !35 + %3133 = bitcast float %2708 to i32, !dbg !38 + %3134 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3133, i32 8, i32 31), !dbg !38 + %3135 = bitcast i32 %3134 to float, !dbg !38 + %3136 = fadd float %2708, %3135, !dbg !35 + %3137 = bitcast float %3136 to i32, !dbg !38 + %3138 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3137, i32 4, i32 31), !dbg !38 + %3139 = bitcast i32 %3138 to float, !dbg !38 + %3140 = fadd float %3136, %3139, !dbg !35 + %3141 = bitcast float %3140 to i32, !dbg !38 + %3142 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3141, i32 2, i32 31), !dbg !38 + %3143 = bitcast i32 %3142 to float, !dbg !38 + %3144 = fadd float %3140, %3143, !dbg !35 + %3145 = bitcast float %3144 to i32, !dbg !38 + %3146 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3145, i32 1, i32 31), !dbg !38 + %3147 = bitcast i32 %3146 to float, !dbg !38 + %3148 = fadd float %3144, %3147, !dbg !35 + %3149 = bitcast float %2715 to i32, !dbg !38 + %3150 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3149, i32 8, i32 31), !dbg !38 + %3151 = bitcast i32 %3150 to float, !dbg !38 + %3152 = fadd float %2715, %3151, !dbg !35 + %3153 = bitcast float %3152 to i32, !dbg !38 + %3154 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3153, i32 4, i32 31), !dbg !38 + %3155 = bitcast i32 %3154 to float, !dbg !38 + %3156 = fadd float %3152, %3155, !dbg !35 + %3157 = bitcast float %3156 to i32, !dbg !38 + %3158 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3157, i32 2, i32 31), !dbg !38 + %3159 = bitcast i32 %3158 to float, !dbg !38 + %3160 = fadd float %3156, %3159, !dbg !35 + %3161 = bitcast float %3160 to i32, !dbg !38 + %3162 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3161, i32 1, i32 31), !dbg !38 + %3163 = bitcast i32 %3162 to float, !dbg !38 + %3164 = fadd float %3160, %3163, !dbg !35 + %3165 = bitcast float %2722 to i32, !dbg !38 + %3166 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3165, i32 8, i32 31), !dbg !38 + %3167 = bitcast i32 %3166 to float, !dbg !38 + %3168 = fadd float %2722, %3167, !dbg !35 + %3169 = bitcast float %3168 to i32, !dbg !38 + %3170 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3169, i32 4, i32 31), !dbg !38 + %3171 = bitcast i32 %3170 to float, !dbg !38 + %3172 = fadd float %3168, %3171, !dbg !35 + %3173 = bitcast float %3172 to i32, !dbg !38 + %3174 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3173, i32 2, i32 31), !dbg !38 + %3175 = bitcast i32 %3174 to float, !dbg !38 + %3176 = fadd float %3172, %3175, !dbg !35 + %3177 = bitcast float %3176 to i32, !dbg !38 + %3178 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3177, i32 1, i32 31), !dbg !38 + %3179 = bitcast i32 %3178 to float, !dbg !38 + %3180 = fadd float %3176, %3179, !dbg !35 + %3181 = bitcast float %2729 to i32, !dbg !38 + %3182 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3181, i32 8, i32 31), !dbg !38 + %3183 = bitcast i32 %3182 to float, !dbg !38 + %3184 = fadd float %2729, %3183, !dbg !35 + %3185 = bitcast float %3184 to i32, !dbg !38 + %3186 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3185, i32 4, i32 31), !dbg !38 + %3187 = bitcast i32 %3186 to float, !dbg !38 + %3188 = fadd float %3184, %3187, !dbg !35 + %3189 = bitcast float %3188 to i32, !dbg !38 + %3190 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3189, i32 2, i32 31), !dbg !38 + %3191 = bitcast i32 %3190 to float, !dbg !38 + %3192 = fadd float %3188, %3191, !dbg !35 + %3193 = bitcast float %3192 to i32, !dbg !38 + %3194 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3193, i32 1, i32 31), !dbg !38 + %3195 = bitcast i32 %3194 to float, !dbg !38 + %3196 = fadd float %3192, %3195, !dbg !35 + %3197 = bitcast float %2736 to i32, !dbg !38 + %3198 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3197, i32 8, i32 31), !dbg !38 + %3199 = bitcast i32 %3198 to float, !dbg !38 + %3200 = fadd float %2736, %3199, !dbg !35 + %3201 = bitcast float %3200 to i32, !dbg !38 + %3202 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3201, i32 4, i32 31), !dbg !38 + %3203 = bitcast i32 %3202 to float, !dbg !38 + %3204 = fadd float %3200, %3203, !dbg !35 + %3205 = bitcast float %3204 to i32, !dbg !38 + %3206 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3205, i32 2, i32 31), !dbg !38 + %3207 = bitcast i32 %3206 to float, !dbg !38 + %3208 = fadd float %3204, %3207, !dbg !35 + %3209 = bitcast float %3208 to i32, !dbg !38 + %3210 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3209, i32 1, i32 31), !dbg !38 + %3211 = bitcast i32 %3210 to float, !dbg !38 + %3212 = fadd float %3208, %3211, !dbg !35 + %3213 = bitcast float %2743 to i32, !dbg !38 + %3214 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3213, i32 8, i32 31), !dbg !38 + %3215 = bitcast i32 %3214 to float, !dbg !38 + %3216 = fadd float %2743, %3215, !dbg !35 + %3217 = bitcast float %3216 to i32, !dbg !38 + %3218 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3217, i32 4, i32 31), !dbg !38 + %3219 = bitcast i32 %3218 to float, !dbg !38 + %3220 = fadd float %3216, %3219, !dbg !35 + %3221 = bitcast float %3220 to i32, !dbg !38 + %3222 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3221, i32 2, i32 31), !dbg !38 + %3223 = bitcast i32 %3222 to float, !dbg !38 + %3224 = fadd float %3220, %3223, !dbg !35 + %3225 = bitcast float %3224 to i32, !dbg !38 + %3226 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3225, i32 1, i32 31), !dbg !38 + %3227 = bitcast i32 %3226 to float, !dbg !38 + %3228 = fadd float %3224, %3227, !dbg !35 + %3229 = bitcast float %2750 to i32, !dbg !38 + %3230 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3229, i32 8, i32 31), !dbg !38 + %3231 = bitcast i32 %3230 to float, !dbg !38 + %3232 = fadd float %2750, %3231, !dbg !35 + %3233 = bitcast float %3232 to i32, !dbg !38 + %3234 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3233, i32 4, i32 31), !dbg !38 + %3235 = bitcast i32 %3234 to float, !dbg !38 + %3236 = fadd float %3232, %3235, !dbg !35 + %3237 = bitcast float %3236 to i32, !dbg !38 + %3238 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3237, i32 2, i32 31), !dbg !38 + %3239 = bitcast i32 %3238 to float, !dbg !38 + %3240 = fadd float %3236, %3239, !dbg !35 + %3241 = bitcast float %3240 to i32, !dbg !38 + %3242 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3241, i32 1, i32 31), !dbg !38 + %3243 = bitcast i32 %3242 to float, !dbg !38 + %3244 = fadd float %3240, %3243, !dbg !35 + %3245 = bitcast float %2757 to i32, !dbg !38 + %3246 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3245, i32 8, i32 31), !dbg !38 + %3247 = bitcast i32 %3246 to float, !dbg !38 + %3248 = fadd float %2757, %3247, !dbg !35 + %3249 = bitcast float %3248 to i32, !dbg !38 + %3250 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3249, i32 4, i32 31), !dbg !38 + %3251 = bitcast i32 %3250 to float, !dbg !38 + %3252 = fadd float %3248, %3251, !dbg !35 + %3253 = bitcast float %3252 to i32, !dbg !38 + %3254 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3253, i32 2, i32 31), !dbg !38 + %3255 = bitcast i32 %3254 to float, !dbg !38 + %3256 = fadd float %3252, %3255, !dbg !35 + %3257 = bitcast float %3256 to i32, !dbg !38 + %3258 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3257, i32 1, i32 31), !dbg !38 + %3259 = bitcast i32 %3258 to float, !dbg !38 + %3260 = fadd float %3256, %3259, !dbg !35 + %3261 = bitcast float %2764 to i32, !dbg !38 + %3262 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3261, i32 8, i32 31), !dbg !38 + %3263 = bitcast i32 %3262 to float, !dbg !38 + %3264 = fadd float %2764, %3263, !dbg !35 + %3265 = bitcast float %3264 to i32, !dbg !38 + %3266 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3265, i32 4, i32 31), !dbg !38 + %3267 = bitcast i32 %3266 to float, !dbg !38 + %3268 = fadd float %3264, %3267, !dbg !35 + %3269 = bitcast float %3268 to i32, !dbg !38 + %3270 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3269, i32 2, i32 31), !dbg !38 + %3271 = bitcast i32 %3270 to float, !dbg !38 + %3272 = fadd float %3268, %3271, !dbg !35 + %3273 = bitcast float %3272 to i32, !dbg !38 + %3274 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3273, i32 1, i32 31), !dbg !38 + %3275 = bitcast i32 %3274 to float, !dbg !38 + %3276 = fadd float %3272, %3275, !dbg !35 + %3277 = fmul float %2780, %888, !dbg !41 + %3278 = fmul float %2780, %889, !dbg !41 + %3279 = fmul float %2780, %890, !dbg !41 + %3280 = fmul float %2780, %891, !dbg !41 + %3281 = fmul float %2780, %892, !dbg !41 + %3282 = fmul float %2780, %893, !dbg !41 + %3283 = fmul float %2780, %894, !dbg !41 + %3284 = fmul float %2780, %895, !dbg !41 + %3285 = fmul float %2796, %896, !dbg !41 + %3286 = fmul float %2796, %897, !dbg !41 + %3287 = fmul float %2796, %898, !dbg !41 + %3288 = fmul float %2796, %899, !dbg !41 + %3289 = fmul float %2796, %900, !dbg !41 + %3290 = fmul float %2796, %901, !dbg !41 + %3291 = fmul float %2796, %902, !dbg !41 + %3292 = fmul float %2796, %903, !dbg !41 + %3293 = fmul float %2812, %904, !dbg !41 + %3294 = fmul float %2812, %905, !dbg !41 + %3295 = fmul float %2812, %906, !dbg !41 + %3296 = fmul float %2812, %907, !dbg !41 + %3297 = fmul float %2812, %908, !dbg !41 + %3298 = fmul float %2812, %909, !dbg !41 + %3299 = fmul float %2812, %910, !dbg !41 + %3300 = fmul float %2812, %911, !dbg !41 + %3301 = fmul float %2828, %912, !dbg !41 + %3302 = fmul float %2828, %913, !dbg !41 + %3303 = fmul float %2828, %914, !dbg !41 + %3304 = fmul float %2828, %915, !dbg !41 + %3305 = fmul float %2828, %916, !dbg !41 + %3306 = fmul float %2828, %917, !dbg !41 + %3307 = fmul float %2828, %918, !dbg !41 + %3308 = fmul float %2828, %919, !dbg !41 + %3309 = fmul float %2844, %920, !dbg !41 + %3310 = fmul float %2844, %921, !dbg !41 + %3311 = fmul float %2844, %922, !dbg !41 + %3312 = fmul float %2844, %923, !dbg !41 + %3313 = fmul float %2844, %924, !dbg !41 + %3314 = fmul float %2844, %925, !dbg !41 + %3315 = fmul float %2844, %926, !dbg !41 + %3316 = fmul float %2844, %927, !dbg !41 + %3317 = fmul float %2860, %928, !dbg !41 + %3318 = fmul float %2860, %929, !dbg !41 + %3319 = fmul float %2860, %930, !dbg !41 + %3320 = fmul float %2860, %931, !dbg !41 + %3321 = fmul float %2860, %932, !dbg !41 + %3322 = fmul float %2860, %933, !dbg !41 + %3323 = fmul float %2860, %934, !dbg !41 + %3324 = fmul float %2860, %935, !dbg !41 + %3325 = fmul float %2876, %936, !dbg !41 + %3326 = fmul float %2876, %937, !dbg !41 + %3327 = fmul float %2876, %938, !dbg !41 + %3328 = fmul float %2876, %939, !dbg !41 + %3329 = fmul float %2876, %940, !dbg !41 + %3330 = fmul float %2876, %941, !dbg !41 + %3331 = fmul float %2876, %942, !dbg !41 + %3332 = fmul float %2876, %943, !dbg !41 + %3333 = fmul float %2892, %944, !dbg !41 + %3334 = fmul float %2892, %945, !dbg !41 + %3335 = fmul float %2892, %946, !dbg !41 + %3336 = fmul float %2892, %947, !dbg !41 + %3337 = fmul float %2892, %948, !dbg !41 + %3338 = fmul float %2892, %949, !dbg !41 + %3339 = fmul float %2892, %950, !dbg !41 + %3340 = fmul float %2892, %951, !dbg !41 + %3341 = fmul float %2908, %952, !dbg !41 + %3342 = fmul float %2908, %953, !dbg !41 + %3343 = fmul float %2908, %954, !dbg !41 + %3344 = fmul float %2908, %955, !dbg !41 + %3345 = fmul float %2908, %956, !dbg !41 + %3346 = fmul float %2908, %957, !dbg !41 + %3347 = fmul float %2908, %958, !dbg !41 + %3348 = fmul float %2908, %959, !dbg !41 + %3349 = fmul float %2924, %960, !dbg !41 + %3350 = fmul float %2924, %961, !dbg !41 + %3351 = fmul float %2924, %962, !dbg !41 + %3352 = fmul float %2924, %963, !dbg !41 + %3353 = fmul float %2924, %964, !dbg !41 + %3354 = fmul float %2924, %965, !dbg !41 + %3355 = fmul float %2924, %966, !dbg !41 + %3356 = fmul float %2924, %967, !dbg !41 + %3357 = fmul float %2940, %968, !dbg !41 + %3358 = fmul float %2940, %969, !dbg !41 + %3359 = fmul float %2940, %970, !dbg !41 + %3360 = fmul float %2940, %971, !dbg !41 + %3361 = fmul float %2940, %972, !dbg !41 + %3362 = fmul float %2940, %973, !dbg !41 + %3363 = fmul float %2940, %974, !dbg !41 + %3364 = fmul float %2940, %975, !dbg !41 + %3365 = fmul float %2956, %976, !dbg !41 + %3366 = fmul float %2956, %977, !dbg !41 + %3367 = fmul float %2956, %978, !dbg !41 + %3368 = fmul float %2956, %979, !dbg !41 + %3369 = fmul float %2956, %980, !dbg !41 + %3370 = fmul float %2956, %981, !dbg !41 + %3371 = fmul float %2956, %982, !dbg !41 + %3372 = fmul float %2956, %983, !dbg !41 + %3373 = fmul float %2972, %984, !dbg !41 + %3374 = fmul float %2972, %985, !dbg !41 + %3375 = fmul float %2972, %986, !dbg !41 + %3376 = fmul float %2972, %987, !dbg !41 + %3377 = fmul float %2972, %988, !dbg !41 + %3378 = fmul float %2972, %989, !dbg !41 + %3379 = fmul float %2972, %990, !dbg !41 + %3380 = fmul float %2972, %991, !dbg !41 + %3381 = fmul float %2988, %992, !dbg !41 + %3382 = fmul float %2988, %993, !dbg !41 + %3383 = fmul float %2988, %994, !dbg !41 + %3384 = fmul float %2988, %995, !dbg !41 + %3385 = fmul float %2988, %996, !dbg !41 + %3386 = fmul float %2988, %997, !dbg !41 + %3387 = fmul float %2988, %998, !dbg !41 + %3388 = fmul float %2988, %999, !dbg !41 + %3389 = fmul float %3004, %1000, !dbg !41 + %3390 = fmul float %3004, %1001, !dbg !41 + %3391 = fmul float %3004, %1002, !dbg !41 + %3392 = fmul float %3004, %1003, !dbg !41 + %3393 = fmul float %3004, %1004, !dbg !41 + %3394 = fmul float %3004, %1005, !dbg !41 + %3395 = fmul float %3004, %1006, !dbg !41 + %3396 = fmul float %3004, %1007, !dbg !41 + %3397 = fmul float %3020, %1008, !dbg !41 + %3398 = fmul float %3020, %1009, !dbg !41 + %3399 = fmul float %3020, %1010, !dbg !41 + %3400 = fmul float %3020, %1011, !dbg !41 + %3401 = fmul float %3020, %1012, !dbg !41 + %3402 = fmul float %3020, %1013, !dbg !41 + %3403 = fmul float %3020, %1014, !dbg !41 + %3404 = fmul float %3020, %1015, !dbg !41 + %3405 = fmul float %3036, %1016, !dbg !41 + %3406 = fmul float %3036, %1017, !dbg !41 + %3407 = fmul float %3036, %1018, !dbg !41 + %3408 = fmul float %3036, %1019, !dbg !41 + %3409 = fmul float %3036, %1020, !dbg !41 + %3410 = fmul float %3036, %1021, !dbg !41 + %3411 = fmul float %3036, %1022, !dbg !41 + %3412 = fmul float %3036, %1023, !dbg !41 + %3413 = fmul float %3052, %1024, !dbg !41 + %3414 = fmul float %3052, %1025, !dbg !41 + %3415 = fmul float %3052, %1026, !dbg !41 + %3416 = fmul float %3052, %1027, !dbg !41 + %3417 = fmul float %3052, %1028, !dbg !41 + %3418 = fmul float %3052, %1029, !dbg !41 + %3419 = fmul float %3052, %1030, !dbg !41 + %3420 = fmul float %3052, %1031, !dbg !41 + %3421 = fmul float %3068, %1032, !dbg !41 + %3422 = fmul float %3068, %1033, !dbg !41 + %3423 = fmul float %3068, %1034, !dbg !41 + %3424 = fmul float %3068, %1035, !dbg !41 + %3425 = fmul float %3068, %1036, !dbg !41 + %3426 = fmul float %3068, %1037, !dbg !41 + %3427 = fmul float %3068, %1038, !dbg !41 + %3428 = fmul float %3068, %1039, !dbg !41 + %3429 = fmul float %3084, %1040, !dbg !41 + %3430 = fmul float %3084, %1041, !dbg !41 + %3431 = fmul float %3084, %1042, !dbg !41 + %3432 = fmul float %3084, %1043, !dbg !41 + %3433 = fmul float %3084, %1044, !dbg !41 + %3434 = fmul float %3084, %1045, !dbg !41 + %3435 = fmul float %3084, %1046, !dbg !41 + %3436 = fmul float %3084, %1047, !dbg !41 + %3437 = fmul float %3100, %1048, !dbg !41 + %3438 = fmul float %3100, %1049, !dbg !41 + %3439 = fmul float %3100, %1050, !dbg !41 + %3440 = fmul float %3100, %1051, !dbg !41 + %3441 = fmul float %3100, %1052, !dbg !41 + %3442 = fmul float %3100, %1053, !dbg !41 + %3443 = fmul float %3100, %1054, !dbg !41 + %3444 = fmul float %3100, %1055, !dbg !41 + %3445 = fmul float %3116, %1056, !dbg !41 + %3446 = fmul float %3116, %1057, !dbg !41 + %3447 = fmul float %3116, %1058, !dbg !41 + %3448 = fmul float %3116, %1059, !dbg !41 + %3449 = fmul float %3116, %1060, !dbg !41 + %3450 = fmul float %3116, %1061, !dbg !41 + %3451 = fmul float %3116, %1062, !dbg !41 + %3452 = fmul float %3116, %1063, !dbg !41 + %3453 = fmul float %3132, %1064, !dbg !41 + %3454 = fmul float %3132, %1065, !dbg !41 + %3455 = fmul float %3132, %1066, !dbg !41 + %3456 = fmul float %3132, %1067, !dbg !41 + %3457 = fmul float %3132, %1068, !dbg !41 + %3458 = fmul float %3132, %1069, !dbg !41 + %3459 = fmul float %3132, %1070, !dbg !41 + %3460 = fmul float %3132, %1071, !dbg !41 + %3461 = fmul float %3148, %1072, !dbg !41 + %3462 = fmul float %3148, %1073, !dbg !41 + %3463 = fmul float %3148, %1074, !dbg !41 + %3464 = fmul float %3148, %1075, !dbg !41 + %3465 = fmul float %3148, %1076, !dbg !41 + %3466 = fmul float %3148, %1077, !dbg !41 + %3467 = fmul float %3148, %1078, !dbg !41 + %3468 = fmul float %3148, %1079, !dbg !41 + %3469 = fmul float %3164, %1080, !dbg !41 + %3470 = fmul float %3164, %1081, !dbg !41 + %3471 = fmul float %3164, %1082, !dbg !41 + %3472 = fmul float %3164, %1083, !dbg !41 + %3473 = fmul float %3164, %1084, !dbg !41 + %3474 = fmul float %3164, %1085, !dbg !41 + %3475 = fmul float %3164, %1086, !dbg !41 + %3476 = fmul float %3164, %1087, !dbg !41 + %3477 = fmul float %3180, %1088, !dbg !41 + %3478 = fmul float %3180, %1089, !dbg !41 + %3479 = fmul float %3180, %1090, !dbg !41 + %3480 = fmul float %3180, %1091, !dbg !41 + %3481 = fmul float %3180, %1092, !dbg !41 + %3482 = fmul float %3180, %1093, !dbg !41 + %3483 = fmul float %3180, %1094, !dbg !41 + %3484 = fmul float %3180, %1095, !dbg !41 + %3485 = fmul float %3196, %1096, !dbg !41 + %3486 = fmul float %3196, %1097, !dbg !41 + %3487 = fmul float %3196, %1098, !dbg !41 + %3488 = fmul float %3196, %1099, !dbg !41 + %3489 = fmul float %3196, %1100, !dbg !41 + %3490 = fmul float %3196, %1101, !dbg !41 + %3491 = fmul float %3196, %1102, !dbg !41 + %3492 = fmul float %3196, %1103, !dbg !41 + %3493 = fmul float %3212, %1104, !dbg !41 + %3494 = fmul float %3212, %1105, !dbg !41 + %3495 = fmul float %3212, %1106, !dbg !41 + %3496 = fmul float %3212, %1107, !dbg !41 + %3497 = fmul float %3212, %1108, !dbg !41 + %3498 = fmul float %3212, %1109, !dbg !41 + %3499 = fmul float %3212, %1110, !dbg !41 + %3500 = fmul float %3212, %1111, !dbg !41 + %3501 = fmul float %3228, %1112, !dbg !41 + %3502 = fmul float %3228, %1113, !dbg !41 + %3503 = fmul float %3228, %1114, !dbg !41 + %3504 = fmul float %3228, %1115, !dbg !41 + %3505 = fmul float %3228, %1116, !dbg !41 + %3506 = fmul float %3228, %1117, !dbg !41 + %3507 = fmul float %3228, %1118, !dbg !41 + %3508 = fmul float %3228, %1119, !dbg !41 + %3509 = fmul float %3244, %1120, !dbg !41 + %3510 = fmul float %3244, %1121, !dbg !41 + %3511 = fmul float %3244, %1122, !dbg !41 + %3512 = fmul float %3244, %1123, !dbg !41 + %3513 = fmul float %3244, %1124, !dbg !41 + %3514 = fmul float %3244, %1125, !dbg !41 + %3515 = fmul float %3244, %1126, !dbg !41 + %3516 = fmul float %3244, %1127, !dbg !41 + %3517 = fmul float %3260, %1128, !dbg !41 + %3518 = fmul float %3260, %1129, !dbg !41 + %3519 = fmul float %3260, %1130, !dbg !41 + %3520 = fmul float %3260, %1131, !dbg !41 + %3521 = fmul float %3260, %1132, !dbg !41 + %3522 = fmul float %3260, %1133, !dbg !41 + %3523 = fmul float %3260, %1134, !dbg !41 + %3524 = fmul float %3260, %1135, !dbg !41 + %3525 = fmul float %3276, %1136, !dbg !41 + %3526 = fmul float %3276, %1137, !dbg !41 + %3527 = fmul float %3276, %1138, !dbg !41 + %3528 = fmul float %3276, %1139, !dbg !41 + %3529 = fmul float %3276, %1140, !dbg !41 + %3530 = fmul float %3276, %1141, !dbg !41 + %3531 = fmul float %3276, %1142, !dbg !41 + %3532 = fmul float %3276, %1143, !dbg !41 + %3533 = fmul float %1966, %3277, !dbg !42 + %3534 = fmul float %1966, %3278, !dbg !42 + %3535 = fmul float %1966, %3279, !dbg !42 + %3536 = fmul float %1966, %3280, !dbg !42 + %3537 = fmul float %1966, %3281, !dbg !42 + %3538 = fmul float %1966, %3282, !dbg !42 + %3539 = fmul float %1966, %3283, !dbg !42 + %3540 = fmul float %1966, %3284, !dbg !42 + %3541 = fmul float %1968, %3285, !dbg !42 + %3542 = fmul float %1968, %3286, !dbg !42 + %3543 = fmul float %1968, %3287, !dbg !42 + %3544 = fmul float %1968, %3288, !dbg !42 + %3545 = fmul float %1968, %3289, !dbg !42 + %3546 = fmul float %1968, %3290, !dbg !42 + %3547 = fmul float %1968, %3291, !dbg !42 + %3548 = fmul float %1968, %3292, !dbg !42 + %3549 = fmul float %1970, %3293, !dbg !42 + %3550 = fmul float %1970, %3294, !dbg !42 + %3551 = fmul float %1970, %3295, !dbg !42 + %3552 = fmul float %1970, %3296, !dbg !42 + %3553 = fmul float %1970, %3297, !dbg !42 + %3554 = fmul float %1970, %3298, !dbg !42 + %3555 = fmul float %1970, %3299, !dbg !42 + %3556 = fmul float %1970, %3300, !dbg !42 + %3557 = fmul float %1972, %3301, !dbg !42 + %3558 = fmul float %1972, %3302, !dbg !42 + %3559 = fmul float %1972, %3303, !dbg !42 + %3560 = fmul float %1972, %3304, !dbg !42 + %3561 = fmul float %1972, %3305, !dbg !42 + %3562 = fmul float %1972, %3306, !dbg !42 + %3563 = fmul float %1972, %3307, !dbg !42 + %3564 = fmul float %1972, %3308, !dbg !42 + %3565 = fmul float %1974, %3309, !dbg !42 + %3566 = fmul float %1974, %3310, !dbg !42 + %3567 = fmul float %1974, %3311, !dbg !42 + %3568 = fmul float %1974, %3312, !dbg !42 + %3569 = fmul float %1974, %3313, !dbg !42 + %3570 = fmul float %1974, %3314, !dbg !42 + %3571 = fmul float %1974, %3315, !dbg !42 + %3572 = fmul float %1974, %3316, !dbg !42 + %3573 = fmul float %1976, %3317, !dbg !42 + %3574 = fmul float %1976, %3318, !dbg !42 + %3575 = fmul float %1976, %3319, !dbg !42 + %3576 = fmul float %1976, %3320, !dbg !42 + %3577 = fmul float %1976, %3321, !dbg !42 + %3578 = fmul float %1976, %3322, !dbg !42 + %3579 = fmul float %1976, %3323, !dbg !42 + %3580 = fmul float %1976, %3324, !dbg !42 + %3581 = fmul float %1978, %3325, !dbg !42 + %3582 = fmul float %1978, %3326, !dbg !42 + %3583 = fmul float %1978, %3327, !dbg !42 + %3584 = fmul float %1978, %3328, !dbg !42 + %3585 = fmul float %1978, %3329, !dbg !42 + %3586 = fmul float %1978, %3330, !dbg !42 + %3587 = fmul float %1978, %3331, !dbg !42 + %3588 = fmul float %1978, %3332, !dbg !42 + %3589 = fmul float %1980, %3333, !dbg !42 + %3590 = fmul float %1980, %3334, !dbg !42 + %3591 = fmul float %1980, %3335, !dbg !42 + %3592 = fmul float %1980, %3336, !dbg !42 + %3593 = fmul float %1980, %3337, !dbg !42 + %3594 = fmul float %1980, %3338, !dbg !42 + %3595 = fmul float %1980, %3339, !dbg !42 + %3596 = fmul float %1980, %3340, !dbg !42 + %3597 = fmul float %1982, %3341, !dbg !42 + %3598 = fmul float %1982, %3342, !dbg !42 + %3599 = fmul float %1982, %3343, !dbg !42 + %3600 = fmul float %1982, %3344, !dbg !42 + %3601 = fmul float %1982, %3345, !dbg !42 + %3602 = fmul float %1982, %3346, !dbg !42 + %3603 = fmul float %1982, %3347, !dbg !42 + %3604 = fmul float %1982, %3348, !dbg !42 + %3605 = fmul float %1984, %3349, !dbg !42 + %3606 = fmul float %1984, %3350, !dbg !42 + %3607 = fmul float %1984, %3351, !dbg !42 + %3608 = fmul float %1984, %3352, !dbg !42 + %3609 = fmul float %1984, %3353, !dbg !42 + %3610 = fmul float %1984, %3354, !dbg !42 + %3611 = fmul float %1984, %3355, !dbg !42 + %3612 = fmul float %1984, %3356, !dbg !42 + %3613 = fmul float %1986, %3357, !dbg !42 + %3614 = fmul float %1986, %3358, !dbg !42 + %3615 = fmul float %1986, %3359, !dbg !42 + %3616 = fmul float %1986, %3360, !dbg !42 + %3617 = fmul float %1986, %3361, !dbg !42 + %3618 = fmul float %1986, %3362, !dbg !42 + %3619 = fmul float %1986, %3363, !dbg !42 + %3620 = fmul float %1986, %3364, !dbg !42 + %3621 = fmul float %1988, %3365, !dbg !42 + %3622 = fmul float %1988, %3366, !dbg !42 + %3623 = fmul float %1988, %3367, !dbg !42 + %3624 = fmul float %1988, %3368, !dbg !42 + %3625 = fmul float %1988, %3369, !dbg !42 + %3626 = fmul float %1988, %3370, !dbg !42 + %3627 = fmul float %1988, %3371, !dbg !42 + %3628 = fmul float %1988, %3372, !dbg !42 + %3629 = fmul float %1990, %3373, !dbg !42 + %3630 = fmul float %1990, %3374, !dbg !42 + %3631 = fmul float %1990, %3375, !dbg !42 + %3632 = fmul float %1990, %3376, !dbg !42 + %3633 = fmul float %1990, %3377, !dbg !42 + %3634 = fmul float %1990, %3378, !dbg !42 + %3635 = fmul float %1990, %3379, !dbg !42 + %3636 = fmul float %1990, %3380, !dbg !42 + %3637 = fmul float %1992, %3381, !dbg !42 + %3638 = fmul float %1992, %3382, !dbg !42 + %3639 = fmul float %1992, %3383, !dbg !42 + %3640 = fmul float %1992, %3384, !dbg !42 + %3641 = fmul float %1992, %3385, !dbg !42 + %3642 = fmul float %1992, %3386, !dbg !42 + %3643 = fmul float %1992, %3387, !dbg !42 + %3644 = fmul float %1992, %3388, !dbg !42 + %3645 = fmul float %1994, %3389, !dbg !42 + %3646 = fmul float %1994, %3390, !dbg !42 + %3647 = fmul float %1994, %3391, !dbg !42 + %3648 = fmul float %1994, %3392, !dbg !42 + %3649 = fmul float %1994, %3393, !dbg !42 + %3650 = fmul float %1994, %3394, !dbg !42 + %3651 = fmul float %1994, %3395, !dbg !42 + %3652 = fmul float %1994, %3396, !dbg !42 + %3653 = fmul float %1996, %3397, !dbg !42 + %3654 = fmul float %1996, %3398, !dbg !42 + %3655 = fmul float %1996, %3399, !dbg !42 + %3656 = fmul float %1996, %3400, !dbg !42 + %3657 = fmul float %1996, %3401, !dbg !42 + %3658 = fmul float %1996, %3402, !dbg !42 + %3659 = fmul float %1996, %3403, !dbg !42 + %3660 = fmul float %1996, %3404, !dbg !42 + %3661 = fmul float %1998, %3405, !dbg !42 + %3662 = fmul float %1998, %3406, !dbg !42 + %3663 = fmul float %1998, %3407, !dbg !42 + %3664 = fmul float %1998, %3408, !dbg !42 + %3665 = fmul float %1998, %3409, !dbg !42 + %3666 = fmul float %1998, %3410, !dbg !42 + %3667 = fmul float %1998, %3411, !dbg !42 + %3668 = fmul float %1998, %3412, !dbg !42 + %3669 = fmul float %2000, %3413, !dbg !42 + %3670 = fmul float %2000, %3414, !dbg !42 + %3671 = fmul float %2000, %3415, !dbg !42 + %3672 = fmul float %2000, %3416, !dbg !42 + %3673 = fmul float %2000, %3417, !dbg !42 + %3674 = fmul float %2000, %3418, !dbg !42 + %3675 = fmul float %2000, %3419, !dbg !42 + %3676 = fmul float %2000, %3420, !dbg !42 + %3677 = fmul float %2002, %3421, !dbg !42 + %3678 = fmul float %2002, %3422, !dbg !42 + %3679 = fmul float %2002, %3423, !dbg !42 + %3680 = fmul float %2002, %3424, !dbg !42 + %3681 = fmul float %2002, %3425, !dbg !42 + %3682 = fmul float %2002, %3426, !dbg !42 + %3683 = fmul float %2002, %3427, !dbg !42 + %3684 = fmul float %2002, %3428, !dbg !42 + %3685 = fmul float %2004, %3429, !dbg !42 + %3686 = fmul float %2004, %3430, !dbg !42 + %3687 = fmul float %2004, %3431, !dbg !42 + %3688 = fmul float %2004, %3432, !dbg !42 + %3689 = fmul float %2004, %3433, !dbg !42 + %3690 = fmul float %2004, %3434, !dbg !42 + %3691 = fmul float %2004, %3435, !dbg !42 + %3692 = fmul float %2004, %3436, !dbg !42 + %3693 = fmul float %2006, %3437, !dbg !42 + %3694 = fmul float %2006, %3438, !dbg !42 + %3695 = fmul float %2006, %3439, !dbg !42 + %3696 = fmul float %2006, %3440, !dbg !42 + %3697 = fmul float %2006, %3441, !dbg !42 + %3698 = fmul float %2006, %3442, !dbg !42 + %3699 = fmul float %2006, %3443, !dbg !42 + %3700 = fmul float %2006, %3444, !dbg !42 + %3701 = fmul float %2008, %3445, !dbg !42 + %3702 = fmul float %2008, %3446, !dbg !42 + %3703 = fmul float %2008, %3447, !dbg !42 + %3704 = fmul float %2008, %3448, !dbg !42 + %3705 = fmul float %2008, %3449, !dbg !42 + %3706 = fmul float %2008, %3450, !dbg !42 + %3707 = fmul float %2008, %3451, !dbg !42 + %3708 = fmul float %2008, %3452, !dbg !42 + %3709 = fmul float %2010, %3453, !dbg !42 + %3710 = fmul float %2010, %3454, !dbg !42 + %3711 = fmul float %2010, %3455, !dbg !42 + %3712 = fmul float %2010, %3456, !dbg !42 + %3713 = fmul float %2010, %3457, !dbg !42 + %3714 = fmul float %2010, %3458, !dbg !42 + %3715 = fmul float %2010, %3459, !dbg !42 + %3716 = fmul float %2010, %3460, !dbg !42 + %3717 = fmul float %2012, %3461, !dbg !42 + %3718 = fmul float %2012, %3462, !dbg !42 + %3719 = fmul float %2012, %3463, !dbg !42 + %3720 = fmul float %2012, %3464, !dbg !42 + %3721 = fmul float %2012, %3465, !dbg !42 + %3722 = fmul float %2012, %3466, !dbg !42 + %3723 = fmul float %2012, %3467, !dbg !42 + %3724 = fmul float %2012, %3468, !dbg !42 + %3725 = fmul float %2014, %3469, !dbg !42 + %3726 = fmul float %2014, %3470, !dbg !42 + %3727 = fmul float %2014, %3471, !dbg !42 + %3728 = fmul float %2014, %3472, !dbg !42 + %3729 = fmul float %2014, %3473, !dbg !42 + %3730 = fmul float %2014, %3474, !dbg !42 + %3731 = fmul float %2014, %3475, !dbg !42 + %3732 = fmul float %2014, %3476, !dbg !42 + %3733 = fmul float %2016, %3477, !dbg !42 + %3734 = fmul float %2016, %3478, !dbg !42 + %3735 = fmul float %2016, %3479, !dbg !42 + %3736 = fmul float %2016, %3480, !dbg !42 + %3737 = fmul float %2016, %3481, !dbg !42 + %3738 = fmul float %2016, %3482, !dbg !42 + %3739 = fmul float %2016, %3483, !dbg !42 + %3740 = fmul float %2016, %3484, !dbg !42 + %3741 = fmul float %2018, %3485, !dbg !42 + %3742 = fmul float %2018, %3486, !dbg !42 + %3743 = fmul float %2018, %3487, !dbg !42 + %3744 = fmul float %2018, %3488, !dbg !42 + %3745 = fmul float %2018, %3489, !dbg !42 + %3746 = fmul float %2018, %3490, !dbg !42 + %3747 = fmul float %2018, %3491, !dbg !42 + %3748 = fmul float %2018, %3492, !dbg !42 + %3749 = fmul float %2020, %3493, !dbg !42 + %3750 = fmul float %2020, %3494, !dbg !42 + %3751 = fmul float %2020, %3495, !dbg !42 + %3752 = fmul float %2020, %3496, !dbg !42 + %3753 = fmul float %2020, %3497, !dbg !42 + %3754 = fmul float %2020, %3498, !dbg !42 + %3755 = fmul float %2020, %3499, !dbg !42 + %3756 = fmul float %2020, %3500, !dbg !42 + %3757 = fmul float %2022, %3501, !dbg !42 + %3758 = fmul float %2022, %3502, !dbg !42 + %3759 = fmul float %2022, %3503, !dbg !42 + %3760 = fmul float %2022, %3504, !dbg !42 + %3761 = fmul float %2022, %3505, !dbg !42 + %3762 = fmul float %2022, %3506, !dbg !42 + %3763 = fmul float %2022, %3507, !dbg !42 + %3764 = fmul float %2022, %3508, !dbg !42 + %3765 = fmul float %2024, %3509, !dbg !42 + %3766 = fmul float %2024, %3510, !dbg !42 + %3767 = fmul float %2024, %3511, !dbg !42 + %3768 = fmul float %2024, %3512, !dbg !42 + %3769 = fmul float %2024, %3513, !dbg !42 + %3770 = fmul float %2024, %3514, !dbg !42 + %3771 = fmul float %2024, %3515, !dbg !42 + %3772 = fmul float %2024, %3516, !dbg !42 + %3773 = fmul float %2026, %3517, !dbg !42 + %3774 = fmul float %2026, %3518, !dbg !42 + %3775 = fmul float %2026, %3519, !dbg !42 + %3776 = fmul float %2026, %3520, !dbg !42 + %3777 = fmul float %2026, %3521, !dbg !42 + %3778 = fmul float %2026, %3522, !dbg !42 + %3779 = fmul float %2026, %3523, !dbg !42 + %3780 = fmul float %2026, %3524, !dbg !42 + %3781 = fmul float %2028, %3525, !dbg !42 + %3782 = fmul float %2028, %3526, !dbg !42 + %3783 = fmul float %2028, %3527, !dbg !42 + %3784 = fmul float %2028, %3528, !dbg !42 + %3785 = fmul float %2028, %3529, !dbg !42 + %3786 = fmul float %2028, %3530, !dbg !42 + %3787 = fmul float %2028, %3531, !dbg !42 + %3788 = fmul float %2028, %3532, !dbg !42 + %3789 = fsub float %2029, %3533, !dbg !43 + %3790 = fsub float %2030, %3534, !dbg !43 + %3791 = fsub float %2031, %3535, !dbg !43 + %3792 = fsub float %2032, %3536, !dbg !43 + %3793 = fsub float %2033, %3537, !dbg !43 + %3794 = fsub float %2034, %3538, !dbg !43 + %3795 = fsub float %2035, %3539, !dbg !43 + %3796 = fsub float %2036, %3540, !dbg !43 + %3797 = fsub float %2037, %3541, !dbg !43 + %3798 = fsub float %2038, %3542, !dbg !43 + %3799 = fsub float %2039, %3543, !dbg !43 + %3800 = fsub float %2040, %3544, !dbg !43 + %3801 = fsub float %2041, %3545, !dbg !43 + %3802 = fsub float %2042, %3546, !dbg !43 + %3803 = fsub float %2043, %3547, !dbg !43 + %3804 = fsub float %2044, %3548, !dbg !43 + %3805 = insertelement <2 x float> poison, float %2045, i64 0, !dbg !43 + %3806 = insertelement <2 x float> %3805, float %2046, i64 1, !dbg !43 + %3807 = insertelement <2 x float> poison, float %3549, i64 0, !dbg !43 + %3808 = insertelement <2 x float> %3807, float %3550, i64 1, !dbg !43 + %3809 = fsub <2 x float> %3806, %3808, !dbg !43 + %3810 = insertelement <2 x float> poison, float %2047, i64 0, !dbg !43 + %3811 = insertelement <2 x float> %3810, float %2048, i64 1, !dbg !43 + %3812 = insertelement <2 x float> poison, float %3551, i64 0, !dbg !43 + %3813 = insertelement <2 x float> %3812, float %3552, i64 1, !dbg !43 + %3814 = fsub <2 x float> %3811, %3813, !dbg !43 + %3815 = insertelement <2 x float> poison, float %2049, i64 0, !dbg !43 + %3816 = insertelement <2 x float> %3815, float %2050, i64 1, !dbg !43 + %3817 = insertelement <2 x float> poison, float %3553, i64 0, !dbg !43 + %3818 = insertelement <2 x float> %3817, float %3554, i64 1, !dbg !43 + %3819 = fsub <2 x float> %3816, %3818, !dbg !43 + %3820 = insertelement <2 x float> poison, float %2051, i64 0, !dbg !43 + %3821 = insertelement <2 x float> %3820, float %2052, i64 1, !dbg !43 + %3822 = insertelement <2 x float> poison, float %3555, i64 0, !dbg !43 + %3823 = insertelement <2 x float> %3822, float %3556, i64 1, !dbg !43 + %3824 = fsub <2 x float> %3821, %3823, !dbg !43 + %3825 = insertelement <2 x float> poison, float %2053, i64 0, !dbg !43 + %3826 = insertelement <2 x float> %3825, float %2054, i64 1, !dbg !43 + %3827 = insertelement <2 x float> poison, float %3557, i64 0, !dbg !43 + %3828 = insertelement <2 x float> %3827, float %3558, i64 1, !dbg !43 + %3829 = fsub <2 x float> %3826, %3828, !dbg !43 + %3830 = insertelement <2 x float> poison, float %2055, i64 0, !dbg !43 + %3831 = insertelement <2 x float> %3830, float %2056, i64 1, !dbg !43 + %3832 = insertelement <2 x float> poison, float %3559, i64 0, !dbg !43 + %3833 = insertelement <2 x float> %3832, float %3560, i64 1, !dbg !43 + %3834 = fsub <2 x float> %3831, %3833, !dbg !43 + %3835 = insertelement <2 x float> poison, float %2057, i64 0, !dbg !43 + %3836 = insertelement <2 x float> %3835, float %2058, i64 1, !dbg !43 + %3837 = insertelement <2 x float> poison, float %3561, i64 0, !dbg !43 + %3838 = insertelement <2 x float> %3837, float %3562, i64 1, !dbg !43 + %3839 = fsub <2 x float> %3836, %3838, !dbg !43 + %3840 = insertelement <2 x float> poison, float %2059, i64 0, !dbg !43 + %3841 = insertelement <2 x float> %3840, float %2060, i64 1, !dbg !43 + %3842 = insertelement <2 x float> poison, float %3563, i64 0, !dbg !43 + %3843 = insertelement <2 x float> %3842, float %3564, i64 1, !dbg !43 + %3844 = fsub <2 x float> %3841, %3843, !dbg !43 + %3845 = insertelement <2 x float> poison, float %2061, i64 0, !dbg !43 + %3846 = insertelement <2 x float> %3845, float %2062, i64 1, !dbg !43 + %3847 = insertelement <2 x float> poison, float %3565, i64 0, !dbg !43 + %3848 = insertelement <2 x float> %3847, float %3566, i64 1, !dbg !43 + %3849 = fsub <2 x float> %3846, %3848, !dbg !43 + %3850 = insertelement <2 x float> poison, float %2063, i64 0, !dbg !43 + %3851 = insertelement <2 x float> %3850, float %2064, i64 1, !dbg !43 + %3852 = insertelement <2 x float> poison, float %3567, i64 0, !dbg !43 + %3853 = insertelement <2 x float> %3852, float %3568, i64 1, !dbg !43 + %3854 = fsub <2 x float> %3851, %3853, !dbg !43 + %3855 = insertelement <2 x float> poison, float %2065, i64 0, !dbg !43 + %3856 = insertelement <2 x float> %3855, float %2066, i64 1, !dbg !43 + %3857 = insertelement <2 x float> poison, float %3569, i64 0, !dbg !43 + %3858 = insertelement <2 x float> %3857, float %3570, i64 1, !dbg !43 + %3859 = fsub <2 x float> %3856, %3858, !dbg !43 + %3860 = insertelement <2 x float> poison, float %2067, i64 0, !dbg !43 + %3861 = insertelement <2 x float> %3860, float %2068, i64 1, !dbg !43 + %3862 = insertelement <2 x float> poison, float %3571, i64 0, !dbg !43 + %3863 = insertelement <2 x float> %3862, float %3572, i64 1, !dbg !43 + %3864 = fsub <2 x float> %3861, %3863, !dbg !43 + %3865 = insertelement <2 x float> poison, float %2069, i64 0, !dbg !43 + %3866 = insertelement <2 x float> %3865, float %2070, i64 1, !dbg !43 + %3867 = insertelement <2 x float> poison, float %3573, i64 0, !dbg !43 + %3868 = insertelement <2 x float> %3867, float %3574, i64 1, !dbg !43 + %3869 = fsub <2 x float> %3866, %3868, !dbg !43 + %3870 = insertelement <2 x float> poison, float %2071, i64 0, !dbg !43 + %3871 = insertelement <2 x float> %3870, float %2072, i64 1, !dbg !43 + %3872 = insertelement <2 x float> poison, float %3575, i64 0, !dbg !43 + %3873 = insertelement <2 x float> %3872, float %3576, i64 1, !dbg !43 + %3874 = fsub <2 x float> %3871, %3873, !dbg !43 + %3875 = insertelement <2 x float> poison, float %2073, i64 0, !dbg !43 + %3876 = insertelement <2 x float> %3875, float %2074, i64 1, !dbg !43 + %3877 = insertelement <2 x float> poison, float %3577, i64 0, !dbg !43 + %3878 = insertelement <2 x float> %3877, float %3578, i64 1, !dbg !43 + %3879 = fsub <2 x float> %3876, %3878, !dbg !43 + %3880 = insertelement <2 x float> poison, float %2075, i64 0, !dbg !43 + %3881 = insertelement <2 x float> %3880, float %2076, i64 1, !dbg !43 + %3882 = insertelement <2 x float> poison, float %3579, i64 0, !dbg !43 + %3883 = insertelement <2 x float> %3882, float %3580, i64 1, !dbg !43 + %3884 = fsub <2 x float> %3881, %3883, !dbg !43 + %3885 = insertelement <2 x float> poison, float %2077, i64 0, !dbg !43 + %3886 = insertelement <2 x float> %3885, float %2078, i64 1, !dbg !43 + %3887 = insertelement <2 x float> poison, float %3581, i64 0, !dbg !43 + %3888 = insertelement <2 x float> %3887, float %3582, i64 1, !dbg !43 + %3889 = fsub <2 x float> %3886, %3888, !dbg !43 + %3890 = insertelement <2 x float> poison, float %2079, i64 0, !dbg !43 + %3891 = insertelement <2 x float> %3890, float %2080, i64 1, !dbg !43 + %3892 = insertelement <2 x float> poison, float %3583, i64 0, !dbg !43 + %3893 = insertelement <2 x float> %3892, float %3584, i64 1, !dbg !43 + %3894 = fsub <2 x float> %3891, %3893, !dbg !43 + %3895 = insertelement <2 x float> poison, float %2081, i64 0, !dbg !43 + %3896 = insertelement <2 x float> %3895, float %2082, i64 1, !dbg !43 + %3897 = insertelement <2 x float> poison, float %3585, i64 0, !dbg !43 + %3898 = insertelement <2 x float> %3897, float %3586, i64 1, !dbg !43 + %3899 = fsub <2 x float> %3896, %3898, !dbg !43 + %3900 = insertelement <2 x float> poison, float %2083, i64 0, !dbg !43 + %3901 = insertelement <2 x float> %3900, float %2084, i64 1, !dbg !43 + %3902 = insertelement <2 x float> poison, float %3587, i64 0, !dbg !43 + %3903 = insertelement <2 x float> %3902, float %3588, i64 1, !dbg !43 + %3904 = fsub <2 x float> %3901, %3903, !dbg !43 + %3905 = insertelement <2 x float> poison, float %2085, i64 0, !dbg !43 + %3906 = insertelement <2 x float> %3905, float %2086, i64 1, !dbg !43 + %3907 = insertelement <2 x float> poison, float %3589, i64 0, !dbg !43 + %3908 = insertelement <2 x float> %3907, float %3590, i64 1, !dbg !43 + %3909 = fsub <2 x float> %3906, %3908, !dbg !43 + %3910 = insertelement <2 x float> poison, float %2087, i64 0, !dbg !43 + %3911 = insertelement <2 x float> %3910, float %2088, i64 1, !dbg !43 + %3912 = insertelement <2 x float> poison, float %3591, i64 0, !dbg !43 + %3913 = insertelement <2 x float> %3912, float %3592, i64 1, !dbg !43 + %3914 = fsub <2 x float> %3911, %3913, !dbg !43 + %3915 = insertelement <2 x float> poison, float %2089, i64 0, !dbg !43 + %3916 = insertelement <2 x float> %3915, float %2090, i64 1, !dbg !43 + %3917 = insertelement <2 x float> poison, float %3593, i64 0, !dbg !43 + %3918 = insertelement <2 x float> %3917, float %3594, i64 1, !dbg !43 + %3919 = fsub <2 x float> %3916, %3918, !dbg !43 + %3920 = insertelement <2 x float> poison, float %2091, i64 0, !dbg !43 + %3921 = insertelement <2 x float> %3920, float %2092, i64 1, !dbg !43 + %3922 = insertelement <2 x float> poison, float %3595, i64 0, !dbg !43 + %3923 = insertelement <2 x float> %3922, float %3596, i64 1, !dbg !43 + %3924 = fsub <2 x float> %3921, %3923, !dbg !43 + %3925 = insertelement <2 x float> poison, float %2093, i64 0, !dbg !43 + %3926 = insertelement <2 x float> %3925, float %2094, i64 1, !dbg !43 + %3927 = insertelement <2 x float> poison, float %3597, i64 0, !dbg !43 + %3928 = insertelement <2 x float> %3927, float %3598, i64 1, !dbg !43 + %3929 = fsub <2 x float> %3926, %3928, !dbg !43 + %3930 = insertelement <2 x float> poison, float %2095, i64 0, !dbg !43 + %3931 = insertelement <2 x float> %3930, float %2096, i64 1, !dbg !43 + %3932 = insertelement <2 x float> poison, float %3599, i64 0, !dbg !43 + %3933 = insertelement <2 x float> %3932, float %3600, i64 1, !dbg !43 + %3934 = fsub <2 x float> %3931, %3933, !dbg !43 + %3935 = insertelement <2 x float> poison, float %2097, i64 0, !dbg !43 + %3936 = insertelement <2 x float> %3935, float %2098, i64 1, !dbg !43 + %3937 = insertelement <2 x float> poison, float %3601, i64 0, !dbg !43 + %3938 = insertelement <2 x float> %3937, float %3602, i64 1, !dbg !43 + %3939 = fsub <2 x float> %3936, %3938, !dbg !43 + %3940 = insertelement <2 x float> poison, float %2099, i64 0, !dbg !43 + %3941 = insertelement <2 x float> %3940, float %2100, i64 1, !dbg !43 + %3942 = insertelement <2 x float> poison, float %3603, i64 0, !dbg !43 + %3943 = insertelement <2 x float> %3942, float %3604, i64 1, !dbg !43 + %3944 = fsub <2 x float> %3941, %3943, !dbg !43 + %3945 = insertelement <2 x float> poison, float %2101, i64 0, !dbg !43 + %3946 = insertelement <2 x float> %3945, float %2102, i64 1, !dbg !43 + %3947 = insertelement <2 x float> poison, float %3605, i64 0, !dbg !43 + %3948 = insertelement <2 x float> %3947, float %3606, i64 1, !dbg !43 + %3949 = fsub <2 x float> %3946, %3948, !dbg !43 + %3950 = insertelement <2 x float> poison, float %2103, i64 0, !dbg !43 + %3951 = insertelement <2 x float> %3950, float %2104, i64 1, !dbg !43 + %3952 = insertelement <2 x float> poison, float %3607, i64 0, !dbg !43 + %3953 = insertelement <2 x float> %3952, float %3608, i64 1, !dbg !43 + %3954 = fsub <2 x float> %3951, %3953, !dbg !43 + %3955 = insertelement <2 x float> poison, float %2105, i64 0, !dbg !43 + %3956 = insertelement <2 x float> %3955, float %2106, i64 1, !dbg !43 + %3957 = insertelement <2 x float> poison, float %3609, i64 0, !dbg !43 + %3958 = insertelement <2 x float> %3957, float %3610, i64 1, !dbg !43 + %3959 = fsub <2 x float> %3956, %3958, !dbg !43 + %3960 = insertelement <2 x float> poison, float %2107, i64 0, !dbg !43 + %3961 = insertelement <2 x float> %3960, float %2108, i64 1, !dbg !43 + %3962 = insertelement <2 x float> poison, float %3611, i64 0, !dbg !43 + %3963 = insertelement <2 x float> %3962, float %3612, i64 1, !dbg !43 + %3964 = fsub <2 x float> %3961, %3963, !dbg !43 + %3965 = insertelement <2 x float> poison, float %2109, i64 0, !dbg !43 + %3966 = insertelement <2 x float> %3965, float %2110, i64 1, !dbg !43 + %3967 = insertelement <2 x float> poison, float %3613, i64 0, !dbg !43 + %3968 = insertelement <2 x float> %3967, float %3614, i64 1, !dbg !43 + %3969 = fsub <2 x float> %3966, %3968, !dbg !43 + %3970 = insertelement <2 x float> poison, float %2111, i64 0, !dbg !43 + %3971 = insertelement <2 x float> %3970, float %2112, i64 1, !dbg !43 + %3972 = insertelement <2 x float> poison, float %3615, i64 0, !dbg !43 + %3973 = insertelement <2 x float> %3972, float %3616, i64 1, !dbg !43 + %3974 = fsub <2 x float> %3971, %3973, !dbg !43 + %3975 = insertelement <2 x float> poison, float %2113, i64 0, !dbg !43 + %3976 = insertelement <2 x float> %3975, float %2114, i64 1, !dbg !43 + %3977 = insertelement <2 x float> poison, float %3617, i64 0, !dbg !43 + %3978 = insertelement <2 x float> %3977, float %3618, i64 1, !dbg !43 + %3979 = fsub <2 x float> %3976, %3978, !dbg !43 + %3980 = insertelement <2 x float> poison, float %2115, i64 0, !dbg !43 + %3981 = insertelement <2 x float> %3980, float %2116, i64 1, !dbg !43 + %3982 = insertelement <2 x float> poison, float %3619, i64 0, !dbg !43 + %3983 = insertelement <2 x float> %3982, float %3620, i64 1, !dbg !43 + %3984 = fsub <2 x float> %3981, %3983, !dbg !43 + %3985 = insertelement <2 x float> poison, float %2117, i64 0, !dbg !43 + %3986 = insertelement <2 x float> %3985, float %2118, i64 1, !dbg !43 + %3987 = insertelement <2 x float> poison, float %3621, i64 0, !dbg !43 + %3988 = insertelement <2 x float> %3987, float %3622, i64 1, !dbg !43 + %3989 = fsub <2 x float> %3986, %3988, !dbg !43 + %3990 = insertelement <2 x float> poison, float %2119, i64 0, !dbg !43 + %3991 = insertelement <2 x float> %3990, float %2120, i64 1, !dbg !43 + %3992 = insertelement <2 x float> poison, float %3623, i64 0, !dbg !43 + %3993 = insertelement <2 x float> %3992, float %3624, i64 1, !dbg !43 + %3994 = fsub <2 x float> %3991, %3993, !dbg !43 + %3995 = insertelement <2 x float> poison, float %2121, i64 0, !dbg !43 + %3996 = insertelement <2 x float> %3995, float %2122, i64 1, !dbg !43 + %3997 = insertelement <2 x float> poison, float %3625, i64 0, !dbg !43 + %3998 = insertelement <2 x float> %3997, float %3626, i64 1, !dbg !43 + %3999 = fsub <2 x float> %3996, %3998, !dbg !43 + %4000 = insertelement <2 x float> poison, float %2123, i64 0, !dbg !43 + %4001 = insertelement <2 x float> %4000, float %2124, i64 1, !dbg !43 + %4002 = insertelement <2 x float> poison, float %3627, i64 0, !dbg !43 + %4003 = insertelement <2 x float> %4002, float %3628, i64 1, !dbg !43 + %4004 = fsub <2 x float> %4001, %4003, !dbg !43 + %4005 = insertelement <2 x float> poison, float %2125, i64 0, !dbg !43 + %4006 = insertelement <2 x float> %4005, float %2126, i64 1, !dbg !43 + %4007 = insertelement <2 x float> poison, float %3629, i64 0, !dbg !43 + %4008 = insertelement <2 x float> %4007, float %3630, i64 1, !dbg !43 + %4009 = fsub <2 x float> %4006, %4008, !dbg !43 + %4010 = insertelement <2 x float> poison, float %2127, i64 0, !dbg !43 + %4011 = insertelement <2 x float> %4010, float %2128, i64 1, !dbg !43 + %4012 = insertelement <2 x float> poison, float %3631, i64 0, !dbg !43 + %4013 = insertelement <2 x float> %4012, float %3632, i64 1, !dbg !43 + %4014 = fsub <2 x float> %4011, %4013, !dbg !43 + %4015 = insertelement <2 x float> poison, float %2129, i64 0, !dbg !43 + %4016 = insertelement <2 x float> %4015, float %2130, i64 1, !dbg !43 + %4017 = insertelement <2 x float> poison, float %3633, i64 0, !dbg !43 + %4018 = insertelement <2 x float> %4017, float %3634, i64 1, !dbg !43 + %4019 = fsub <2 x float> %4016, %4018, !dbg !43 + %4020 = insertelement <2 x float> poison, float %2131, i64 0, !dbg !43 + %4021 = insertelement <2 x float> %4020, float %2132, i64 1, !dbg !43 + %4022 = insertelement <2 x float> poison, float %3635, i64 0, !dbg !43 + %4023 = insertelement <2 x float> %4022, float %3636, i64 1, !dbg !43 + %4024 = fsub <2 x float> %4021, %4023, !dbg !43 + %4025 = insertelement <2 x float> poison, float %2133, i64 0, !dbg !43 + %4026 = insertelement <2 x float> %4025, float %2134, i64 1, !dbg !43 + %4027 = insertelement <2 x float> poison, float %3637, i64 0, !dbg !43 + %4028 = insertelement <2 x float> %4027, float %3638, i64 1, !dbg !43 + %4029 = fsub <2 x float> %4026, %4028, !dbg !43 + %4030 = insertelement <2 x float> poison, float %2135, i64 0, !dbg !43 + %4031 = insertelement <2 x float> %4030, float %2136, i64 1, !dbg !43 + %4032 = insertelement <2 x float> poison, float %3639, i64 0, !dbg !43 + %4033 = insertelement <2 x float> %4032, float %3640, i64 1, !dbg !43 + %4034 = fsub <2 x float> %4031, %4033, !dbg !43 + %4035 = insertelement <2 x float> poison, float %2137, i64 0, !dbg !43 + %4036 = insertelement <2 x float> %4035, float %2138, i64 1, !dbg !43 + %4037 = insertelement <2 x float> poison, float %3641, i64 0, !dbg !43 + %4038 = insertelement <2 x float> %4037, float %3642, i64 1, !dbg !43 + %4039 = fsub <2 x float> %4036, %4038, !dbg !43 + %4040 = insertelement <2 x float> poison, float %2139, i64 0, !dbg !43 + %4041 = insertelement <2 x float> %4040, float %2140, i64 1, !dbg !43 + %4042 = insertelement <2 x float> poison, float %3643, i64 0, !dbg !43 + %4043 = insertelement <2 x float> %4042, float %3644, i64 1, !dbg !43 + %4044 = fsub <2 x float> %4041, %4043, !dbg !43 + %4045 = insertelement <2 x float> poison, float %2141, i64 0, !dbg !43 + %4046 = insertelement <2 x float> %4045, float %2142, i64 1, !dbg !43 + %4047 = insertelement <2 x float> poison, float %3645, i64 0, !dbg !43 + %4048 = insertelement <2 x float> %4047, float %3646, i64 1, !dbg !43 + %4049 = fsub <2 x float> %4046, %4048, !dbg !43 + %4050 = insertelement <2 x float> poison, float %2143, i64 0, !dbg !43 + %4051 = insertelement <2 x float> %4050, float %2144, i64 1, !dbg !43 + %4052 = insertelement <2 x float> poison, float %3647, i64 0, !dbg !43 + %4053 = insertelement <2 x float> %4052, float %3648, i64 1, !dbg !43 + %4054 = fsub <2 x float> %4051, %4053, !dbg !43 + %4055 = insertelement <2 x float> poison, float %2145, i64 0, !dbg !43 + %4056 = insertelement <2 x float> %4055, float %2146, i64 1, !dbg !43 + %4057 = insertelement <2 x float> poison, float %3649, i64 0, !dbg !43 + %4058 = insertelement <2 x float> %4057, float %3650, i64 1, !dbg !43 + %4059 = fsub <2 x float> %4056, %4058, !dbg !43 + %4060 = insertelement <2 x float> poison, float %2147, i64 0, !dbg !43 + %4061 = insertelement <2 x float> %4060, float %2148, i64 1, !dbg !43 + %4062 = insertelement <2 x float> poison, float %3651, i64 0, !dbg !43 + %4063 = insertelement <2 x float> %4062, float %3652, i64 1, !dbg !43 + %4064 = fsub <2 x float> %4061, %4063, !dbg !43 + %4065 = insertelement <2 x float> poison, float %2149, i64 0, !dbg !43 + %4066 = insertelement <2 x float> %4065, float %2150, i64 1, !dbg !43 + %4067 = insertelement <2 x float> poison, float %3653, i64 0, !dbg !43 + %4068 = insertelement <2 x float> %4067, float %3654, i64 1, !dbg !43 + %4069 = fsub <2 x float> %4066, %4068, !dbg !43 + %4070 = insertelement <2 x float> poison, float %2151, i64 0, !dbg !43 + %4071 = insertelement <2 x float> %4070, float %2152, i64 1, !dbg !43 + %4072 = insertelement <2 x float> poison, float %3655, i64 0, !dbg !43 + %4073 = insertelement <2 x float> %4072, float %3656, i64 1, !dbg !43 + %4074 = fsub <2 x float> %4071, %4073, !dbg !43 + %4075 = insertelement <2 x float> poison, float %2153, i64 0, !dbg !43 + %4076 = insertelement <2 x float> %4075, float %2154, i64 1, !dbg !43 + %4077 = insertelement <2 x float> poison, float %3657, i64 0, !dbg !43 + %4078 = insertelement <2 x float> %4077, float %3658, i64 1, !dbg !43 + %4079 = fsub <2 x float> %4076, %4078, !dbg !43 + %4080 = insertelement <2 x float> poison, float %2155, i64 0, !dbg !43 + %4081 = insertelement <2 x float> %4080, float %2156, i64 1, !dbg !43 + %4082 = insertelement <2 x float> poison, float %3659, i64 0, !dbg !43 + %4083 = insertelement <2 x float> %4082, float %3660, i64 1, !dbg !43 + %4084 = fsub <2 x float> %4081, %4083, !dbg !43 + %4085 = insertelement <2 x float> poison, float %2157, i64 0, !dbg !43 + %4086 = insertelement <2 x float> %4085, float %2158, i64 1, !dbg !43 + %4087 = insertelement <2 x float> poison, float %3661, i64 0, !dbg !43 + %4088 = insertelement <2 x float> %4087, float %3662, i64 1, !dbg !43 + %4089 = fsub <2 x float> %4086, %4088, !dbg !43 + %4090 = insertelement <2 x float> poison, float %2159, i64 0, !dbg !43 + %4091 = insertelement <2 x float> %4090, float %2160, i64 1, !dbg !43 + %4092 = insertelement <2 x float> poison, float %3663, i64 0, !dbg !43 + %4093 = insertelement <2 x float> %4092, float %3664, i64 1, !dbg !43 + %4094 = fsub <2 x float> %4091, %4093, !dbg !43 + %4095 = insertelement <2 x float> poison, float %2161, i64 0, !dbg !43 + %4096 = insertelement <2 x float> %4095, float %2162, i64 1, !dbg !43 + %4097 = insertelement <2 x float> poison, float %3665, i64 0, !dbg !43 + %4098 = insertelement <2 x float> %4097, float %3666, i64 1, !dbg !43 + %4099 = fsub <2 x float> %4096, %4098, !dbg !43 + %4100 = insertelement <2 x float> poison, float %2163, i64 0, !dbg !43 + %4101 = insertelement <2 x float> %4100, float %2164, i64 1, !dbg !43 + %4102 = insertelement <2 x float> poison, float %3667, i64 0, !dbg !43 + %4103 = insertelement <2 x float> %4102, float %3668, i64 1, !dbg !43 + %4104 = fsub <2 x float> %4101, %4103, !dbg !43 + %4105 = insertelement <2 x float> poison, float %2165, i64 0, !dbg !43 + %4106 = insertelement <2 x float> %4105, float %2166, i64 1, !dbg !43 + %4107 = insertelement <2 x float> poison, float %3669, i64 0, !dbg !43 + %4108 = insertelement <2 x float> %4107, float %3670, i64 1, !dbg !43 + %4109 = fsub <2 x float> %4106, %4108, !dbg !43 + %4110 = insertelement <2 x float> poison, float %2167, i64 0, !dbg !43 + %4111 = insertelement <2 x float> %4110, float %2168, i64 1, !dbg !43 + %4112 = insertelement <2 x float> poison, float %3671, i64 0, !dbg !43 + %4113 = insertelement <2 x float> %4112, float %3672, i64 1, !dbg !43 + %4114 = fsub <2 x float> %4111, %4113, !dbg !43 + %4115 = insertelement <2 x float> poison, float %2169, i64 0, !dbg !43 + %4116 = insertelement <2 x float> %4115, float %2170, i64 1, !dbg !43 + %4117 = insertelement <2 x float> poison, float %3673, i64 0, !dbg !43 + %4118 = insertelement <2 x float> %4117, float %3674, i64 1, !dbg !43 + %4119 = fsub <2 x float> %4116, %4118, !dbg !43 + %4120 = insertelement <2 x float> poison, float %2171, i64 0, !dbg !43 + %4121 = insertelement <2 x float> %4120, float %2172, i64 1, !dbg !43 + %4122 = insertelement <2 x float> poison, float %3675, i64 0, !dbg !43 + %4123 = insertelement <2 x float> %4122, float %3676, i64 1, !dbg !43 + %4124 = fsub <2 x float> %4121, %4123, !dbg !43 + %4125 = insertelement <2 x float> poison, float %2173, i64 0, !dbg !43 + %4126 = insertelement <2 x float> %4125, float %2174, i64 1, !dbg !43 + %4127 = insertelement <2 x float> poison, float %3677, i64 0, !dbg !43 + %4128 = insertelement <2 x float> %4127, float %3678, i64 1, !dbg !43 + %4129 = fsub <2 x float> %4126, %4128, !dbg !43 + %4130 = insertelement <2 x float> poison, float %2175, i64 0, !dbg !43 + %4131 = insertelement <2 x float> %4130, float %2176, i64 1, !dbg !43 + %4132 = insertelement <2 x float> poison, float %3679, i64 0, !dbg !43 + %4133 = insertelement <2 x float> %4132, float %3680, i64 1, !dbg !43 + %4134 = fsub <2 x float> %4131, %4133, !dbg !43 + %4135 = insertelement <2 x float> poison, float %2177, i64 0, !dbg !43 + %4136 = insertelement <2 x float> %4135, float %2178, i64 1, !dbg !43 + %4137 = insertelement <2 x float> poison, float %3681, i64 0, !dbg !43 + %4138 = insertelement <2 x float> %4137, float %3682, i64 1, !dbg !43 + %4139 = fsub <2 x float> %4136, %4138, !dbg !43 + %4140 = insertelement <2 x float> poison, float %2179, i64 0, !dbg !43 + %4141 = insertelement <2 x float> %4140, float %2180, i64 1, !dbg !43 + %4142 = insertelement <2 x float> poison, float %3683, i64 0, !dbg !43 + %4143 = insertelement <2 x float> %4142, float %3684, i64 1, !dbg !43 + %4144 = fsub <2 x float> %4141, %4143, !dbg !43 + %4145 = insertelement <2 x float> poison, float %2181, i64 0, !dbg !43 + %4146 = insertelement <2 x float> %4145, float %2182, i64 1, !dbg !43 + %4147 = insertelement <2 x float> poison, float %3685, i64 0, !dbg !43 + %4148 = insertelement <2 x float> %4147, float %3686, i64 1, !dbg !43 + %4149 = fsub <2 x float> %4146, %4148, !dbg !43 + %4150 = insertelement <2 x float> poison, float %2183, i64 0, !dbg !43 + %4151 = insertelement <2 x float> %4150, float %2184, i64 1, !dbg !43 + %4152 = insertelement <2 x float> poison, float %3687, i64 0, !dbg !43 + %4153 = insertelement <2 x float> %4152, float %3688, i64 1, !dbg !43 + %4154 = fsub <2 x float> %4151, %4153, !dbg !43 + %4155 = insertelement <2 x float> poison, float %2185, i64 0, !dbg !43 + %4156 = insertelement <2 x float> %4155, float %2186, i64 1, !dbg !43 + %4157 = insertelement <2 x float> poison, float %3689, i64 0, !dbg !43 + %4158 = insertelement <2 x float> %4157, float %3690, i64 1, !dbg !43 + %4159 = fsub <2 x float> %4156, %4158, !dbg !43 + %4160 = insertelement <2 x float> poison, float %2187, i64 0, !dbg !43 + %4161 = insertelement <2 x float> %4160, float %2188, i64 1, !dbg !43 + %4162 = insertelement <2 x float> poison, float %3691, i64 0, !dbg !43 + %4163 = insertelement <2 x float> %4162, float %3692, i64 1, !dbg !43 + %4164 = fsub <2 x float> %4161, %4163, !dbg !43 + %4165 = insertelement <2 x float> poison, float %2189, i64 0, !dbg !43 + %4166 = insertelement <2 x float> %4165, float %2190, i64 1, !dbg !43 + %4167 = insertelement <2 x float> poison, float %3693, i64 0, !dbg !43 + %4168 = insertelement <2 x float> %4167, float %3694, i64 1, !dbg !43 + %4169 = fsub <2 x float> %4166, %4168, !dbg !43 + %4170 = insertelement <2 x float> poison, float %2191, i64 0, !dbg !43 + %4171 = insertelement <2 x float> %4170, float %2192, i64 1, !dbg !43 + %4172 = insertelement <2 x float> poison, float %3695, i64 0, !dbg !43 + %4173 = insertelement <2 x float> %4172, float %3696, i64 1, !dbg !43 + %4174 = fsub <2 x float> %4171, %4173, !dbg !43 + %4175 = insertelement <2 x float> poison, float %2193, i64 0, !dbg !43 + %4176 = insertelement <2 x float> %4175, float %2194, i64 1, !dbg !43 + %4177 = insertelement <2 x float> poison, float %3697, i64 0, !dbg !43 + %4178 = insertelement <2 x float> %4177, float %3698, i64 1, !dbg !43 + %4179 = fsub <2 x float> %4176, %4178, !dbg !43 + %4180 = insertelement <2 x float> poison, float %2195, i64 0, !dbg !43 + %4181 = insertelement <2 x float> %4180, float %2196, i64 1, !dbg !43 + %4182 = insertelement <2 x float> poison, float %3699, i64 0, !dbg !43 + %4183 = insertelement <2 x float> %4182, float %3700, i64 1, !dbg !43 + %4184 = fsub <2 x float> %4181, %4183, !dbg !43 + %4185 = insertelement <2 x float> poison, float %2197, i64 0, !dbg !43 + %4186 = insertelement <2 x float> %4185, float %2198, i64 1, !dbg !43 + %4187 = insertelement <2 x float> poison, float %3701, i64 0, !dbg !43 + %4188 = insertelement <2 x float> %4187, float %3702, i64 1, !dbg !43 + %4189 = fsub <2 x float> %4186, %4188, !dbg !43 + %4190 = insertelement <2 x float> poison, float %2199, i64 0, !dbg !43 + %4191 = insertelement <2 x float> %4190, float %2200, i64 1, !dbg !43 + %4192 = insertelement <2 x float> poison, float %3703, i64 0, !dbg !43 + %4193 = insertelement <2 x float> %4192, float %3704, i64 1, !dbg !43 + %4194 = fsub <2 x float> %4191, %4193, !dbg !43 + %4195 = insertelement <2 x float> poison, float %2201, i64 0, !dbg !43 + %4196 = insertelement <2 x float> %4195, float %2202, i64 1, !dbg !43 + %4197 = insertelement <2 x float> poison, float %3705, i64 0, !dbg !43 + %4198 = insertelement <2 x float> %4197, float %3706, i64 1, !dbg !43 + %4199 = fsub <2 x float> %4196, %4198, !dbg !43 + %4200 = insertelement <2 x float> poison, float %2203, i64 0, !dbg !43 + %4201 = insertelement <2 x float> %4200, float %2204, i64 1, !dbg !43 + %4202 = insertelement <2 x float> poison, float %3707, i64 0, !dbg !43 + %4203 = insertelement <2 x float> %4202, float %3708, i64 1, !dbg !43 + %4204 = fsub <2 x float> %4201, %4203, !dbg !43 + %4205 = insertelement <2 x float> poison, float %2205, i64 0, !dbg !43 + %4206 = insertelement <2 x float> %4205, float %2206, i64 1, !dbg !43 + %4207 = insertelement <2 x float> poison, float %3709, i64 0, !dbg !43 + %4208 = insertelement <2 x float> %4207, float %3710, i64 1, !dbg !43 + %4209 = fsub <2 x float> %4206, %4208, !dbg !43 + %4210 = insertelement <2 x float> poison, float %2207, i64 0, !dbg !43 + %4211 = insertelement <2 x float> %4210, float %2208, i64 1, !dbg !43 + %4212 = insertelement <2 x float> poison, float %3711, i64 0, !dbg !43 + %4213 = insertelement <2 x float> %4212, float %3712, i64 1, !dbg !43 + %4214 = fsub <2 x float> %4211, %4213, !dbg !43 + %4215 = insertelement <2 x float> poison, float %2209, i64 0, !dbg !43 + %4216 = insertelement <2 x float> %4215, float %2210, i64 1, !dbg !43 + %4217 = insertelement <2 x float> poison, float %3713, i64 0, !dbg !43 + %4218 = insertelement <2 x float> %4217, float %3714, i64 1, !dbg !43 + %4219 = fsub <2 x float> %4216, %4218, !dbg !43 + %4220 = insertelement <2 x float> poison, float %2211, i64 0, !dbg !43 + %4221 = insertelement <2 x float> %4220, float %2212, i64 1, !dbg !43 + %4222 = insertelement <2 x float> poison, float %3715, i64 0, !dbg !43 + %4223 = insertelement <2 x float> %4222, float %3716, i64 1, !dbg !43 + %4224 = fsub <2 x float> %4221, %4223, !dbg !43 + %4225 = insertelement <2 x float> poison, float %2213, i64 0, !dbg !43 + %4226 = insertelement <2 x float> %4225, float %2214, i64 1, !dbg !43 + %4227 = insertelement <2 x float> poison, float %3717, i64 0, !dbg !43 + %4228 = insertelement <2 x float> %4227, float %3718, i64 1, !dbg !43 + %4229 = fsub <2 x float> %4226, %4228, !dbg !43 + %4230 = insertelement <2 x float> poison, float %2215, i64 0, !dbg !43 + %4231 = insertelement <2 x float> %4230, float %2216, i64 1, !dbg !43 + %4232 = insertelement <2 x float> poison, float %3719, i64 0, !dbg !43 + %4233 = insertelement <2 x float> %4232, float %3720, i64 1, !dbg !43 + %4234 = fsub <2 x float> %4231, %4233, !dbg !43 + %4235 = insertelement <2 x float> poison, float %2217, i64 0, !dbg !43 + %4236 = insertelement <2 x float> %4235, float %2218, i64 1, !dbg !43 + %4237 = insertelement <2 x float> poison, float %3721, i64 0, !dbg !43 + %4238 = insertelement <2 x float> %4237, float %3722, i64 1, !dbg !43 + %4239 = fsub <2 x float> %4236, %4238, !dbg !43 + %4240 = insertelement <2 x float> poison, float %2219, i64 0, !dbg !43 + %4241 = insertelement <2 x float> %4240, float %2220, i64 1, !dbg !43 + %4242 = insertelement <2 x float> poison, float %3723, i64 0, !dbg !43 + %4243 = insertelement <2 x float> %4242, float %3724, i64 1, !dbg !43 + %4244 = fsub <2 x float> %4241, %4243, !dbg !43 + %4245 = insertelement <2 x float> poison, float %2221, i64 0, !dbg !43 + %4246 = insertelement <2 x float> %4245, float %2222, i64 1, !dbg !43 + %4247 = insertelement <2 x float> poison, float %3725, i64 0, !dbg !43 + %4248 = insertelement <2 x float> %4247, float %3726, i64 1, !dbg !43 + %4249 = fsub <2 x float> %4246, %4248, !dbg !43 + %4250 = insertelement <2 x float> poison, float %2223, i64 0, !dbg !43 + %4251 = insertelement <2 x float> %4250, float %2224, i64 1, !dbg !43 + %4252 = insertelement <2 x float> poison, float %3727, i64 0, !dbg !43 + %4253 = insertelement <2 x float> %4252, float %3728, i64 1, !dbg !43 + %4254 = fsub <2 x float> %4251, %4253, !dbg !43 + %4255 = insertelement <2 x float> poison, float %2225, i64 0, !dbg !43 + %4256 = insertelement <2 x float> %4255, float %2226, i64 1, !dbg !43 + %4257 = insertelement <2 x float> poison, float %3729, i64 0, !dbg !43 + %4258 = insertelement <2 x float> %4257, float %3730, i64 1, !dbg !43 + %4259 = fsub <2 x float> %4256, %4258, !dbg !43 + %4260 = insertelement <2 x float> poison, float %2227, i64 0, !dbg !43 + %4261 = insertelement <2 x float> %4260, float %2228, i64 1, !dbg !43 + %4262 = insertelement <2 x float> poison, float %3731, i64 0, !dbg !43 + %4263 = insertelement <2 x float> %4262, float %3732, i64 1, !dbg !43 + %4264 = fsub <2 x float> %4261, %4263, !dbg !43 + %4265 = insertelement <2 x float> poison, float %2229, i64 0, !dbg !43 + %4266 = insertelement <2 x float> %4265, float %2230, i64 1, !dbg !43 + %4267 = insertelement <2 x float> poison, float %3733, i64 0, !dbg !43 + %4268 = insertelement <2 x float> %4267, float %3734, i64 1, !dbg !43 + %4269 = fsub <2 x float> %4266, %4268, !dbg !43 + %4270 = insertelement <2 x float> poison, float %2231, i64 0, !dbg !43 + %4271 = insertelement <2 x float> %4270, float %2232, i64 1, !dbg !43 + %4272 = insertelement <2 x float> poison, float %3735, i64 0, !dbg !43 + %4273 = insertelement <2 x float> %4272, float %3736, i64 1, !dbg !43 + %4274 = fsub <2 x float> %4271, %4273, !dbg !43 + %4275 = insertelement <2 x float> poison, float %2233, i64 0, !dbg !43 + %4276 = insertelement <2 x float> %4275, float %2234, i64 1, !dbg !43 + %4277 = insertelement <2 x float> poison, float %3737, i64 0, !dbg !43 + %4278 = insertelement <2 x float> %4277, float %3738, i64 1, !dbg !43 + %4279 = fsub <2 x float> %4276, %4278, !dbg !43 + %4280 = insertelement <2 x float> poison, float %2235, i64 0, !dbg !43 + %4281 = insertelement <2 x float> %4280, float %2236, i64 1, !dbg !43 + %4282 = insertelement <2 x float> poison, float %3739, i64 0, !dbg !43 + %4283 = insertelement <2 x float> %4282, float %3740, i64 1, !dbg !43 + %4284 = fsub <2 x float> %4281, %4283, !dbg !43 + %4285 = insertelement <2 x float> poison, float %2237, i64 0, !dbg !43 + %4286 = insertelement <2 x float> %4285, float %2238, i64 1, !dbg !43 + %4287 = insertelement <2 x float> poison, float %3741, i64 0, !dbg !43 + %4288 = insertelement <2 x float> %4287, float %3742, i64 1, !dbg !43 + %4289 = fsub <2 x float> %4286, %4288, !dbg !43 + %4290 = insertelement <2 x float> poison, float %2239, i64 0, !dbg !43 + %4291 = insertelement <2 x float> %4290, float %2240, i64 1, !dbg !43 + %4292 = insertelement <2 x float> poison, float %3743, i64 0, !dbg !43 + %4293 = insertelement <2 x float> %4292, float %3744, i64 1, !dbg !43 + %4294 = fsub <2 x float> %4291, %4293, !dbg !43 + %4295 = insertelement <2 x float> poison, float %2241, i64 0, !dbg !43 + %4296 = insertelement <2 x float> %4295, float %2242, i64 1, !dbg !43 + %4297 = insertelement <2 x float> poison, float %3745, i64 0, !dbg !43 + %4298 = insertelement <2 x float> %4297, float %3746, i64 1, !dbg !43 + %4299 = fsub <2 x float> %4296, %4298, !dbg !43 + %4300 = insertelement <2 x float> poison, float %2243, i64 0, !dbg !43 + %4301 = insertelement <2 x float> %4300, float %2244, i64 1, !dbg !43 + %4302 = insertelement <2 x float> poison, float %3747, i64 0, !dbg !43 + %4303 = insertelement <2 x float> %4302, float %3748, i64 1, !dbg !43 + %4304 = fsub <2 x float> %4301, %4303, !dbg !43 + %4305 = insertelement <2 x float> poison, float %2245, i64 0, !dbg !43 + %4306 = insertelement <2 x float> %4305, float %2246, i64 1, !dbg !43 + %4307 = insertelement <2 x float> poison, float %3749, i64 0, !dbg !43 + %4308 = insertelement <2 x float> %4307, float %3750, i64 1, !dbg !43 + %4309 = fsub <2 x float> %4306, %4308, !dbg !43 + %4310 = insertelement <2 x float> poison, float %2247, i64 0, !dbg !43 + %4311 = insertelement <2 x float> %4310, float %2248, i64 1, !dbg !43 + %4312 = insertelement <2 x float> poison, float %3751, i64 0, !dbg !43 + %4313 = insertelement <2 x float> %4312, float %3752, i64 1, !dbg !43 + %4314 = fsub <2 x float> %4311, %4313, !dbg !43 + %4315 = insertelement <2 x float> poison, float %2249, i64 0, !dbg !43 + %4316 = insertelement <2 x float> %4315, float %2250, i64 1, !dbg !43 + %4317 = insertelement <2 x float> poison, float %3753, i64 0, !dbg !43 + %4318 = insertelement <2 x float> %4317, float %3754, i64 1, !dbg !43 + %4319 = fsub <2 x float> %4316, %4318, !dbg !43 + %4320 = insertelement <2 x float> poison, float %2251, i64 0, !dbg !43 + %4321 = insertelement <2 x float> %4320, float %2252, i64 1, !dbg !43 + %4322 = insertelement <2 x float> poison, float %3755, i64 0, !dbg !43 + %4323 = insertelement <2 x float> %4322, float %3756, i64 1, !dbg !43 + %4324 = fsub <2 x float> %4321, %4323, !dbg !43 + %4325 = insertelement <2 x float> poison, float %2253, i64 0, !dbg !43 + %4326 = insertelement <2 x float> %4325, float %2254, i64 1, !dbg !43 + %4327 = insertelement <2 x float> poison, float %3757, i64 0, !dbg !43 + %4328 = insertelement <2 x float> %4327, float %3758, i64 1, !dbg !43 + %4329 = fsub <2 x float> %4326, %4328, !dbg !43 + %4330 = insertelement <2 x float> poison, float %2255, i64 0, !dbg !43 + %4331 = insertelement <2 x float> %4330, float %2256, i64 1, !dbg !43 + %4332 = insertelement <2 x float> poison, float %3759, i64 0, !dbg !43 + %4333 = insertelement <2 x float> %4332, float %3760, i64 1, !dbg !43 + %4334 = fsub <2 x float> %4331, %4333, !dbg !43 + %4335 = insertelement <2 x float> poison, float %2257, i64 0, !dbg !43 + %4336 = insertelement <2 x float> %4335, float %2258, i64 1, !dbg !43 + %4337 = insertelement <2 x float> poison, float %3761, i64 0, !dbg !43 + %4338 = insertelement <2 x float> %4337, float %3762, i64 1, !dbg !43 + %4339 = fsub <2 x float> %4336, %4338, !dbg !43 + %4340 = insertelement <2 x float> poison, float %2259, i64 0, !dbg !43 + %4341 = insertelement <2 x float> %4340, float %2260, i64 1, !dbg !43 + %4342 = insertelement <2 x float> poison, float %3763, i64 0, !dbg !43 + %4343 = insertelement <2 x float> %4342, float %3764, i64 1, !dbg !43 + %4344 = fsub <2 x float> %4341, %4343, !dbg !43 + %4345 = insertelement <2 x float> poison, float %2261, i64 0, !dbg !43 + %4346 = insertelement <2 x float> %4345, float %2262, i64 1, !dbg !43 + %4347 = insertelement <2 x float> poison, float %3765, i64 0, !dbg !43 + %4348 = insertelement <2 x float> %4347, float %3766, i64 1, !dbg !43 + %4349 = fsub <2 x float> %4346, %4348, !dbg !43 + %4350 = insertelement <2 x float> poison, float %2263, i64 0, !dbg !43 + %4351 = insertelement <2 x float> %4350, float %2264, i64 1, !dbg !43 + %4352 = insertelement <2 x float> poison, float %3767, i64 0, !dbg !43 + %4353 = insertelement <2 x float> %4352, float %3768, i64 1, !dbg !43 + %4354 = fsub <2 x float> %4351, %4353, !dbg !43 + %4355 = insertelement <2 x float> poison, float %2265, i64 0, !dbg !43 + %4356 = insertelement <2 x float> %4355, float %2266, i64 1, !dbg !43 + %4357 = insertelement <2 x float> poison, float %3769, i64 0, !dbg !43 + %4358 = insertelement <2 x float> %4357, float %3770, i64 1, !dbg !43 + %4359 = fsub <2 x float> %4356, %4358, !dbg !43 + %4360 = insertelement <2 x float> poison, float %2267, i64 0, !dbg !43 + %4361 = insertelement <2 x float> %4360, float %2268, i64 1, !dbg !43 + %4362 = insertelement <2 x float> poison, float %3771, i64 0, !dbg !43 + %4363 = insertelement <2 x float> %4362, float %3772, i64 1, !dbg !43 + %4364 = fsub <2 x float> %4361, %4363, !dbg !43 + %4365 = insertelement <2 x float> poison, float %2269, i64 0, !dbg !43 + %4366 = insertelement <2 x float> %4365, float %2270, i64 1, !dbg !43 + %4367 = insertelement <2 x float> poison, float %3773, i64 0, !dbg !43 + %4368 = insertelement <2 x float> %4367, float %3774, i64 1, !dbg !43 + %4369 = fsub <2 x float> %4366, %4368, !dbg !43 + %4370 = insertelement <2 x float> poison, float %2271, i64 0, !dbg !43 + %4371 = insertelement <2 x float> %4370, float %2272, i64 1, !dbg !43 + %4372 = insertelement <2 x float> poison, float %3775, i64 0, !dbg !43 + %4373 = insertelement <2 x float> %4372, float %3776, i64 1, !dbg !43 + %4374 = fsub <2 x float> %4371, %4373, !dbg !43 + %4375 = insertelement <2 x float> poison, float %2273, i64 0, !dbg !43 + %4376 = insertelement <2 x float> %4375, float %2274, i64 1, !dbg !43 + %4377 = insertelement <2 x float> poison, float %3777, i64 0, !dbg !43 + %4378 = insertelement <2 x float> %4377, float %3778, i64 1, !dbg !43 + %4379 = fsub <2 x float> %4376, %4378, !dbg !43 + %4380 = insertelement <2 x float> poison, float %2275, i64 0, !dbg !43 + %4381 = insertelement <2 x float> %4380, float %2276, i64 1, !dbg !43 + %4382 = insertelement <2 x float> poison, float %3779, i64 0, !dbg !43 + %4383 = insertelement <2 x float> %4382, float %3780, i64 1, !dbg !43 + %4384 = fsub <2 x float> %4381, %4383, !dbg !43 + %4385 = insertelement <2 x float> poison, float %2277, i64 0, !dbg !43 + %4386 = insertelement <2 x float> %4385, float %2278, i64 1, !dbg !43 + %4387 = insertelement <2 x float> poison, float %3781, i64 0, !dbg !43 + %4388 = insertelement <2 x float> %4387, float %3782, i64 1, !dbg !43 + %4389 = fsub <2 x float> %4386, %4388, !dbg !43 + %4390 = insertelement <2 x float> poison, float %2279, i64 0, !dbg !43 + %4391 = insertelement <2 x float> %4390, float %2280, i64 1, !dbg !43 + %4392 = insertelement <2 x float> poison, float %3783, i64 0, !dbg !43 + %4393 = insertelement <2 x float> %4392, float %3784, i64 1, !dbg !43 + %4394 = fsub <2 x float> %4391, %4393, !dbg !43 + %4395 = insertelement <2 x float> poison, float %2281, i64 0, !dbg !43 + %4396 = insertelement <2 x float> %4395, float %2282, i64 1, !dbg !43 + %4397 = insertelement <2 x float> poison, float %3785, i64 0, !dbg !43 + %4398 = insertelement <2 x float> %4397, float %3786, i64 1, !dbg !43 + %4399 = fsub <2 x float> %4396, %4398, !dbg !43 + %4400 = insertelement <2 x float> poison, float %2283, i64 0, !dbg !43 + %4401 = insertelement <2 x float> %4400, float %2284, i64 1, !dbg !43 + %4402 = insertelement <2 x float> poison, float %3787, i64 0, !dbg !43 + %4403 = insertelement <2 x float> %4402, float %3788, i64 1, !dbg !43 + %4404 = fsub <2 x float> %4401, %4403, !dbg !43 + %4405 = fptrunc float %3789 to bfloat, !dbg !44 + %4406 = fptrunc float %3790 to bfloat, !dbg !44 + %4407 = fptrunc float %3791 to bfloat, !dbg !44 + %4408 = fptrunc float %3792 to bfloat, !dbg !44 + %4409 = fptrunc float %3793 to bfloat, !dbg !44 + %4410 = fptrunc float %3794 to bfloat, !dbg !44 + %4411 = fptrunc float %3795 to bfloat, !dbg !44 + %4412 = fptrunc float %3796 to bfloat, !dbg !44 + %4413 = fptrunc float %3797 to bfloat, !dbg !44 + %4414 = fptrunc float %3798 to bfloat, !dbg !44 + %4415 = fptrunc float %3799 to bfloat, !dbg !44 + %4416 = fptrunc float %3800 to bfloat, !dbg !44 + %4417 = fptrunc float %3801 to bfloat, !dbg !44 + %4418 = fptrunc float %3802 to bfloat, !dbg !44 + %4419 = fptrunc float %3803 to bfloat, !dbg !44 + %4420 = fptrunc float %3804 to bfloat, !dbg !44 + %4421 = fptrunc <2 x float> %3809 to <2 x bfloat>, !dbg !44 + %4422 = fptrunc <2 x float> %3814 to <2 x bfloat>, !dbg !44 + %4423 = fptrunc <2 x float> %3819 to <2 x bfloat>, !dbg !44 + %4424 = fptrunc <2 x float> %3824 to <2 x bfloat>, !dbg !44 + %4425 = fptrunc <2 x float> %3829 to <2 x bfloat>, !dbg !44 + %4426 = fptrunc <2 x float> %3834 to <2 x bfloat>, !dbg !44 + %4427 = fptrunc <2 x float> %3839 to <2 x bfloat>, !dbg !44 + %4428 = fptrunc <2 x float> %3844 to <2 x bfloat>, !dbg !44 + %4429 = fptrunc <2 x float> %3849 to <2 x bfloat>, !dbg !44 + %4430 = fptrunc <2 x float> %3854 to <2 x bfloat>, !dbg !44 + %4431 = fptrunc <2 x float> %3859 to <2 x bfloat>, !dbg !44 + %4432 = fptrunc <2 x float> %3864 to <2 x bfloat>, !dbg !44 + %4433 = fptrunc <2 x float> %3869 to <2 x bfloat>, !dbg !44 + %4434 = fptrunc <2 x float> %3874 to <2 x bfloat>, !dbg !44 + %4435 = fptrunc <2 x float> %3879 to <2 x bfloat>, !dbg !44 + %4436 = fptrunc <2 x float> %3884 to <2 x bfloat>, !dbg !44 + %4437 = fptrunc <2 x float> %3889 to <2 x bfloat>, !dbg !44 + %4438 = fptrunc <2 x float> %3894 to <2 x bfloat>, !dbg !44 + %4439 = fptrunc <2 x float> %3899 to <2 x bfloat>, !dbg !44 + %4440 = fptrunc <2 x float> %3904 to <2 x bfloat>, !dbg !44 + %4441 = fptrunc <2 x float> %3909 to <2 x bfloat>, !dbg !44 + %4442 = fptrunc <2 x float> %3914 to <2 x bfloat>, !dbg !44 + %4443 = fptrunc <2 x float> %3919 to <2 x bfloat>, !dbg !44 + %4444 = fptrunc <2 x float> %3924 to <2 x bfloat>, !dbg !44 + %4445 = fptrunc <2 x float> %3929 to <2 x bfloat>, !dbg !44 + %4446 = fptrunc <2 x float> %3934 to <2 x bfloat>, !dbg !44 + %4447 = fptrunc <2 x float> %3939 to <2 x bfloat>, !dbg !44 + %4448 = fptrunc <2 x float> %3944 to <2 x bfloat>, !dbg !44 + %4449 = fptrunc <2 x float> %3949 to <2 x bfloat>, !dbg !44 + %4450 = fptrunc <2 x float> %3954 to <2 x bfloat>, !dbg !44 + %4451 = fptrunc <2 x float> %3959 to <2 x bfloat>, !dbg !44 + %4452 = fptrunc <2 x float> %3964 to <2 x bfloat>, !dbg !44 + %4453 = fptrunc <2 x float> %3969 to <2 x bfloat>, !dbg !44 + %4454 = fptrunc <2 x float> %3974 to <2 x bfloat>, !dbg !44 + %4455 = fptrunc <2 x float> %3979 to <2 x bfloat>, !dbg !44 + %4456 = fptrunc <2 x float> %3984 to <2 x bfloat>, !dbg !44 + %4457 = fptrunc <2 x float> %3989 to <2 x bfloat>, !dbg !44 + %4458 = fptrunc <2 x float> %3994 to <2 x bfloat>, !dbg !44 + %4459 = fptrunc <2 x float> %3999 to <2 x bfloat>, !dbg !44 + %4460 = fptrunc <2 x float> %4004 to <2 x bfloat>, !dbg !44 + %4461 = fptrunc <2 x float> %4009 to <2 x bfloat>, !dbg !44 + %4462 = fptrunc <2 x float> %4014 to <2 x bfloat>, !dbg !44 + %4463 = fptrunc <2 x float> %4019 to <2 x bfloat>, !dbg !44 + %4464 = fptrunc <2 x float> %4024 to <2 x bfloat>, !dbg !44 + %4465 = fptrunc <2 x float> %4029 to <2 x bfloat>, !dbg !44 + %4466 = fptrunc <2 x float> %4034 to <2 x bfloat>, !dbg !44 + %4467 = fptrunc <2 x float> %4039 to <2 x bfloat>, !dbg !44 + %4468 = fptrunc <2 x float> %4044 to <2 x bfloat>, !dbg !44 + %4469 = fptrunc <2 x float> %4049 to <2 x bfloat>, !dbg !44 + %4470 = fptrunc <2 x float> %4054 to <2 x bfloat>, !dbg !44 + %4471 = fptrunc <2 x float> %4059 to <2 x bfloat>, !dbg !44 + %4472 = fptrunc <2 x float> %4064 to <2 x bfloat>, !dbg !44 + %4473 = fptrunc <2 x float> %4069 to <2 x bfloat>, !dbg !44 + %4474 = fptrunc <2 x float> %4074 to <2 x bfloat>, !dbg !44 + %4475 = fptrunc <2 x float> %4079 to <2 x bfloat>, !dbg !44 + %4476 = fptrunc <2 x float> %4084 to <2 x bfloat>, !dbg !44 + %4477 = fptrunc <2 x float> %4089 to <2 x bfloat>, !dbg !44 + %4478 = fptrunc <2 x float> %4094 to <2 x bfloat>, !dbg !44 + %4479 = fptrunc <2 x float> %4099 to <2 x bfloat>, !dbg !44 + %4480 = fptrunc <2 x float> %4104 to <2 x bfloat>, !dbg !44 + %4481 = fptrunc <2 x float> %4109 to <2 x bfloat>, !dbg !44 + %4482 = fptrunc <2 x float> %4114 to <2 x bfloat>, !dbg !44 + %4483 = fptrunc <2 x float> %4119 to <2 x bfloat>, !dbg !44 + %4484 = fptrunc <2 x float> %4124 to <2 x bfloat>, !dbg !44 + %4485 = fptrunc <2 x float> %4129 to <2 x bfloat>, !dbg !44 + %4486 = fptrunc <2 x float> %4134 to <2 x bfloat>, !dbg !44 + %4487 = fptrunc <2 x float> %4139 to <2 x bfloat>, !dbg !44 + %4488 = fptrunc <2 x float> %4144 to <2 x bfloat>, !dbg !44 + %4489 = fptrunc <2 x float> %4149 to <2 x bfloat>, !dbg !44 + %4490 = fptrunc <2 x float> %4154 to <2 x bfloat>, !dbg !44 + %4491 = fptrunc <2 x float> %4159 to <2 x bfloat>, !dbg !44 + %4492 = fptrunc <2 x float> %4164 to <2 x bfloat>, !dbg !44 + %4493 = fptrunc <2 x float> %4169 to <2 x bfloat>, !dbg !44 + %4494 = fptrunc <2 x float> %4174 to <2 x bfloat>, !dbg !44 + %4495 = fptrunc <2 x float> %4179 to <2 x bfloat>, !dbg !44 + %4496 = fptrunc <2 x float> %4184 to <2 x bfloat>, !dbg !44 + %4497 = fptrunc <2 x float> %4189 to <2 x bfloat>, !dbg !44 + %4498 = fptrunc <2 x float> %4194 to <2 x bfloat>, !dbg !44 + %4499 = fptrunc <2 x float> %4199 to <2 x bfloat>, !dbg !44 + %4500 = fptrunc <2 x float> %4204 to <2 x bfloat>, !dbg !44 + %4501 = fptrunc <2 x float> %4209 to <2 x bfloat>, !dbg !44 + %4502 = fptrunc <2 x float> %4214 to <2 x bfloat>, !dbg !44 + %4503 = fptrunc <2 x float> %4219 to <2 x bfloat>, !dbg !44 + %4504 = fptrunc <2 x float> %4224 to <2 x bfloat>, !dbg !44 + %4505 = fptrunc <2 x float> %4229 to <2 x bfloat>, !dbg !44 + %4506 = fptrunc <2 x float> %4234 to <2 x bfloat>, !dbg !44 + %4507 = fptrunc <2 x float> %4239 to <2 x bfloat>, !dbg !44 + %4508 = fptrunc <2 x float> %4244 to <2 x bfloat>, !dbg !44 + %4509 = fptrunc <2 x float> %4249 to <2 x bfloat>, !dbg !44 + %4510 = fptrunc <2 x float> %4254 to <2 x bfloat>, !dbg !44 + %4511 = fptrunc <2 x float> %4259 to <2 x bfloat>, !dbg !44 + %4512 = fptrunc <2 x float> %4264 to <2 x bfloat>, !dbg !44 + %4513 = fptrunc <2 x float> %4269 to <2 x bfloat>, !dbg !44 + %4514 = fptrunc <2 x float> %4274 to <2 x bfloat>, !dbg !44 + %4515 = fptrunc <2 x float> %4279 to <2 x bfloat>, !dbg !44 + %4516 = fptrunc <2 x float> %4284 to <2 x bfloat>, !dbg !44 + %4517 = fptrunc <2 x float> %4289 to <2 x bfloat>, !dbg !44 + %4518 = fptrunc <2 x float> %4294 to <2 x bfloat>, !dbg !44 + %4519 = fptrunc <2 x float> %4299 to <2 x bfloat>, !dbg !44 + %4520 = fptrunc <2 x float> %4304 to <2 x bfloat>, !dbg !44 + %4521 = fptrunc <2 x float> %4309 to <2 x bfloat>, !dbg !44 + %4522 = fptrunc <2 x float> %4314 to <2 x bfloat>, !dbg !44 + %4523 = fptrunc <2 x float> %4319 to <2 x bfloat>, !dbg !44 + %4524 = fptrunc <2 x float> %4324 to <2 x bfloat>, !dbg !44 + %4525 = fptrunc <2 x float> %4329 to <2 x bfloat>, !dbg !44 + %4526 = fptrunc <2 x float> %4334 to <2 x bfloat>, !dbg !44 + %4527 = fptrunc <2 x float> %4339 to <2 x bfloat>, !dbg !44 + %4528 = fptrunc <2 x float> %4344 to <2 x bfloat>, !dbg !44 + %4529 = fptrunc <2 x float> %4349 to <2 x bfloat>, !dbg !44 + %4530 = fptrunc <2 x float> %4354 to <2 x bfloat>, !dbg !44 + %4531 = fptrunc <2 x float> %4359 to <2 x bfloat>, !dbg !44 + %4532 = fptrunc <2 x float> %4364 to <2 x bfloat>, !dbg !44 + %4533 = fptrunc <2 x float> %4369 to <2 x bfloat>, !dbg !44 + %4534 = fptrunc <2 x float> %4374 to <2 x bfloat>, !dbg !44 + %4535 = fptrunc <2 x float> %4379 to <2 x bfloat>, !dbg !44 + %4536 = fptrunc <2 x float> %4384 to <2 x bfloat>, !dbg !44 + %4537 = fptrunc <2 x float> %4389 to <2 x bfloat>, !dbg !44 + %4538 = fptrunc <2 x float> %4394 to <2 x bfloat>, !dbg !44 + %4539 = fptrunc <2 x float> %4399 to <2 x bfloat>, !dbg !44 + %4540 = fptrunc <2 x float> %4404 to <2 x bfloat>, !dbg !44 + %4541 = insertelement <2 x bfloat> poison, bfloat %4405, i64 0, !dbg !45 + %4542 = insertelement <2 x bfloat> %4541, bfloat %4406, i64 1, !dbg !45 + %4543 = bitcast <2 x bfloat> %4542 to i32, !dbg !45 + %4544 = insertelement <2 x bfloat> poison, bfloat %4407, i64 0, !dbg !45 + %4545 = insertelement <2 x bfloat> %4544, bfloat %4408, i64 1, !dbg !45 + %4546 = bitcast <2 x bfloat> %4545 to i32, !dbg !45 + %4547 = insertelement <2 x bfloat> poison, bfloat %4409, i64 0, !dbg !45 + %4548 = insertelement <2 x bfloat> %4547, bfloat %4410, i64 1, !dbg !45 + %4549 = bitcast <2 x bfloat> %4548 to i32, !dbg !45 + %4550 = insertelement <2 x bfloat> poison, bfloat %4411, i64 0, !dbg !45 + %4551 = insertelement <2 x bfloat> %4550, bfloat %4412, i64 1, !dbg !45 + %4552 = bitcast <2 x bfloat> %4551 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4543, i32 %4546, i32 %4549, i32 %4552, ptr addrspace(1) %282, i1 %22) #4, !dbg !45 + %4553 = insertelement <2 x bfloat> poison, bfloat %4413, i64 0, !dbg !45 + %4554 = insertelement <2 x bfloat> %4553, bfloat %4414, i64 1, !dbg !45 + %4555 = bitcast <2 x bfloat> %4554 to i32, !dbg !45 + %4556 = insertelement <2 x bfloat> poison, bfloat %4415, i64 0, !dbg !45 + %4557 = insertelement <2 x bfloat> %4556, bfloat %4416, i64 1, !dbg !45 + %4558 = bitcast <2 x bfloat> %4557 to i32, !dbg !45 + %4559 = insertelement <2 x bfloat> poison, bfloat %4417, i64 0, !dbg !45 + %4560 = insertelement <2 x bfloat> %4559, bfloat %4418, i64 1, !dbg !45 + %4561 = bitcast <2 x bfloat> %4560 to i32, !dbg !45 + %4562 = insertelement <2 x bfloat> poison, bfloat %4419, i64 0, !dbg !45 + %4563 = insertelement <2 x bfloat> %4562, bfloat %4420, i64 1, !dbg !45 + %4564 = bitcast <2 x bfloat> %4563 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4555, i32 %4558, i32 %4561, i32 %4564, ptr addrspace(1) %283, i1 %23) #4, !dbg !45 + %4565 = bitcast <2 x bfloat> %4421 to i32, !dbg !45 + %4566 = bitcast <2 x bfloat> %4422 to i32, !dbg !45 + %4567 = bitcast <2 x bfloat> %4423 to i32, !dbg !45 + %4568 = bitcast <2 x bfloat> %4424 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4565, i32 %4566, i32 %4567, i32 %4568, ptr addrspace(1) %284, i1 %348) #4, !dbg !45 + %4569 = bitcast <2 x bfloat> %4425 to i32, !dbg !45 + %4570 = bitcast <2 x bfloat> %4426 to i32, !dbg !45 + %4571 = bitcast <2 x bfloat> %4427 to i32, !dbg !45 + %4572 = bitcast <2 x bfloat> %4428 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4569, i32 %4570, i32 %4571, i32 %4572, ptr addrspace(1) %285, i1 %366) #4, !dbg !45 + %4573 = bitcast <2 x bfloat> %4429 to i32, !dbg !45 + %4574 = bitcast <2 x bfloat> %4430 to i32, !dbg !45 + %4575 = bitcast <2 x bfloat> %4431 to i32, !dbg !45 + %4576 = bitcast <2 x bfloat> %4432 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4573, i32 %4574, i32 %4575, i32 %4576, ptr addrspace(1) %286, i1 %384) #4, !dbg !45 + %4577 = bitcast <2 x bfloat> %4433 to i32, !dbg !45 + %4578 = bitcast <2 x bfloat> %4434 to i32, !dbg !45 + %4579 = bitcast <2 x bfloat> %4435 to i32, !dbg !45 + %4580 = bitcast <2 x bfloat> %4436 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4577, i32 %4578, i32 %4579, i32 %4580, ptr addrspace(1) %287, i1 %402) #4, !dbg !45 + %4581 = bitcast <2 x bfloat> %4437 to i32, !dbg !45 + %4582 = bitcast <2 x bfloat> %4438 to i32, !dbg !45 + %4583 = bitcast <2 x bfloat> %4439 to i32, !dbg !45 + %4584 = bitcast <2 x bfloat> %4440 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4581, i32 %4582, i32 %4583, i32 %4584, ptr addrspace(1) %288, i1 %420) #4, !dbg !45 + %4585 = bitcast <2 x bfloat> %4441 to i32, !dbg !45 + %4586 = bitcast <2 x bfloat> %4442 to i32, !dbg !45 + %4587 = bitcast <2 x bfloat> %4443 to i32, !dbg !45 + %4588 = bitcast <2 x bfloat> %4444 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4585, i32 %4586, i32 %4587, i32 %4588, ptr addrspace(1) %289, i1 %438) #4, !dbg !45 + %4589 = bitcast <2 x bfloat> %4445 to i32, !dbg !45 + %4590 = bitcast <2 x bfloat> %4446 to i32, !dbg !45 + %4591 = bitcast <2 x bfloat> %4447 to i32, !dbg !45 + %4592 = bitcast <2 x bfloat> %4448 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4589, i32 %4590, i32 %4591, i32 %4592, ptr addrspace(1) %290, i1 %456) #4, !dbg !45 + %4593 = bitcast <2 x bfloat> %4449 to i32, !dbg !45 + %4594 = bitcast <2 x bfloat> %4450 to i32, !dbg !45 + %4595 = bitcast <2 x bfloat> %4451 to i32, !dbg !45 + %4596 = bitcast <2 x bfloat> %4452 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4593, i32 %4594, i32 %4595, i32 %4596, ptr addrspace(1) %291, i1 %474) #4, !dbg !45 + %4597 = bitcast <2 x bfloat> %4453 to i32, !dbg !45 + %4598 = bitcast <2 x bfloat> %4454 to i32, !dbg !45 + %4599 = bitcast <2 x bfloat> %4455 to i32, !dbg !45 + %4600 = bitcast <2 x bfloat> %4456 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4597, i32 %4598, i32 %4599, i32 %4600, ptr addrspace(1) %292, i1 %492) #4, !dbg !45 + %4601 = bitcast <2 x bfloat> %4457 to i32, !dbg !45 + %4602 = bitcast <2 x bfloat> %4458 to i32, !dbg !45 + %4603 = bitcast <2 x bfloat> %4459 to i32, !dbg !45 + %4604 = bitcast <2 x bfloat> %4460 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4601, i32 %4602, i32 %4603, i32 %4604, ptr addrspace(1) %293, i1 %510) #4, !dbg !45 + %4605 = bitcast <2 x bfloat> %4461 to i32, !dbg !45 + %4606 = bitcast <2 x bfloat> %4462 to i32, !dbg !45 + %4607 = bitcast <2 x bfloat> %4463 to i32, !dbg !45 + %4608 = bitcast <2 x bfloat> %4464 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4605, i32 %4606, i32 %4607, i32 %4608, ptr addrspace(1) %294, i1 %528) #4, !dbg !45 + %4609 = bitcast <2 x bfloat> %4465 to i32, !dbg !45 + %4610 = bitcast <2 x bfloat> %4466 to i32, !dbg !45 + %4611 = bitcast <2 x bfloat> %4467 to i32, !dbg !45 + %4612 = bitcast <2 x bfloat> %4468 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4609, i32 %4610, i32 %4611, i32 %4612, ptr addrspace(1) %295, i1 %546) #4, !dbg !45 + %4613 = bitcast <2 x bfloat> %4469 to i32, !dbg !45 + %4614 = bitcast <2 x bfloat> %4470 to i32, !dbg !45 + %4615 = bitcast <2 x bfloat> %4471 to i32, !dbg !45 + %4616 = bitcast <2 x bfloat> %4472 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4613, i32 %4614, i32 %4615, i32 %4616, ptr addrspace(1) %296, i1 %564) #4, !dbg !45 + %4617 = bitcast <2 x bfloat> %4473 to i32, !dbg !45 + %4618 = bitcast <2 x bfloat> %4474 to i32, !dbg !45 + %4619 = bitcast <2 x bfloat> %4475 to i32, !dbg !45 + %4620 = bitcast <2 x bfloat> %4476 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4617, i32 %4618, i32 %4619, i32 %4620, ptr addrspace(1) %297, i1 %582) #4, !dbg !45 + %4621 = bitcast <2 x bfloat> %4477 to i32, !dbg !45 + %4622 = bitcast <2 x bfloat> %4478 to i32, !dbg !45 + %4623 = bitcast <2 x bfloat> %4479 to i32, !dbg !45 + %4624 = bitcast <2 x bfloat> %4480 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4621, i32 %4622, i32 %4623, i32 %4624, ptr addrspace(1) %298, i1 %600) #4, !dbg !45 + %4625 = bitcast <2 x bfloat> %4481 to i32, !dbg !45 + %4626 = bitcast <2 x bfloat> %4482 to i32, !dbg !45 + %4627 = bitcast <2 x bfloat> %4483 to i32, !dbg !45 + %4628 = bitcast <2 x bfloat> %4484 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4625, i32 %4626, i32 %4627, i32 %4628, ptr addrspace(1) %299, i1 %618) #4, !dbg !45 + %4629 = bitcast <2 x bfloat> %4485 to i32, !dbg !45 + %4630 = bitcast <2 x bfloat> %4486 to i32, !dbg !45 + %4631 = bitcast <2 x bfloat> %4487 to i32, !dbg !45 + %4632 = bitcast <2 x bfloat> %4488 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4629, i32 %4630, i32 %4631, i32 %4632, ptr addrspace(1) %300, i1 %636) #4, !dbg !45 + %4633 = bitcast <2 x bfloat> %4489 to i32, !dbg !45 + %4634 = bitcast <2 x bfloat> %4490 to i32, !dbg !45 + %4635 = bitcast <2 x bfloat> %4491 to i32, !dbg !45 + %4636 = bitcast <2 x bfloat> %4492 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4633, i32 %4634, i32 %4635, i32 %4636, ptr addrspace(1) %301, i1 %654) #4, !dbg !45 + %4637 = bitcast <2 x bfloat> %4493 to i32, !dbg !45 + %4638 = bitcast <2 x bfloat> %4494 to i32, !dbg !45 + %4639 = bitcast <2 x bfloat> %4495 to i32, !dbg !45 + %4640 = bitcast <2 x bfloat> %4496 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4637, i32 %4638, i32 %4639, i32 %4640, ptr addrspace(1) %302, i1 %672) #4, !dbg !45 + %4641 = bitcast <2 x bfloat> %4497 to i32, !dbg !45 + %4642 = bitcast <2 x bfloat> %4498 to i32, !dbg !45 + %4643 = bitcast <2 x bfloat> %4499 to i32, !dbg !45 + %4644 = bitcast <2 x bfloat> %4500 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4641, i32 %4642, i32 %4643, i32 %4644, ptr addrspace(1) %303, i1 %690) #4, !dbg !45 + %4645 = bitcast <2 x bfloat> %4501 to i32, !dbg !45 + %4646 = bitcast <2 x bfloat> %4502 to i32, !dbg !45 + %4647 = bitcast <2 x bfloat> %4503 to i32, !dbg !45 + %4648 = bitcast <2 x bfloat> %4504 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4645, i32 %4646, i32 %4647, i32 %4648, ptr addrspace(1) %304, i1 %708) #4, !dbg !45 + %4649 = bitcast <2 x bfloat> %4505 to i32, !dbg !45 + %4650 = bitcast <2 x bfloat> %4506 to i32, !dbg !45 + %4651 = bitcast <2 x bfloat> %4507 to i32, !dbg !45 + %4652 = bitcast <2 x bfloat> %4508 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4649, i32 %4650, i32 %4651, i32 %4652, ptr addrspace(1) %305, i1 %726) #4, !dbg !45 + %4653 = bitcast <2 x bfloat> %4509 to i32, !dbg !45 + %4654 = bitcast <2 x bfloat> %4510 to i32, !dbg !45 + %4655 = bitcast <2 x bfloat> %4511 to i32, !dbg !45 + %4656 = bitcast <2 x bfloat> %4512 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4653, i32 %4654, i32 %4655, i32 %4656, ptr addrspace(1) %306, i1 %744) #4, !dbg !45 + %4657 = bitcast <2 x bfloat> %4513 to i32, !dbg !45 + %4658 = bitcast <2 x bfloat> %4514 to i32, !dbg !45 + %4659 = bitcast <2 x bfloat> %4515 to i32, !dbg !45 + %4660 = bitcast <2 x bfloat> %4516 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4657, i32 %4658, i32 %4659, i32 %4660, ptr addrspace(1) %307, i1 %762) #4, !dbg !45 + %4661 = bitcast <2 x bfloat> %4517 to i32, !dbg !45 + %4662 = bitcast <2 x bfloat> %4518 to i32, !dbg !45 + %4663 = bitcast <2 x bfloat> %4519 to i32, !dbg !45 + %4664 = bitcast <2 x bfloat> %4520 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4661, i32 %4662, i32 %4663, i32 %4664, ptr addrspace(1) %308, i1 %780) #4, !dbg !45 + %4665 = bitcast <2 x bfloat> %4521 to i32, !dbg !45 + %4666 = bitcast <2 x bfloat> %4522 to i32, !dbg !45 + %4667 = bitcast <2 x bfloat> %4523 to i32, !dbg !45 + %4668 = bitcast <2 x bfloat> %4524 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4665, i32 %4666, i32 %4667, i32 %4668, ptr addrspace(1) %309, i1 %798) #4, !dbg !45 + %4669 = bitcast <2 x bfloat> %4525 to i32, !dbg !45 + %4670 = bitcast <2 x bfloat> %4526 to i32, !dbg !45 + %4671 = bitcast <2 x bfloat> %4527 to i32, !dbg !45 + %4672 = bitcast <2 x bfloat> %4528 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4669, i32 %4670, i32 %4671, i32 %4672, ptr addrspace(1) %310, i1 %816) #4, !dbg !45 + %4673 = bitcast <2 x bfloat> %4529 to i32, !dbg !45 + %4674 = bitcast <2 x bfloat> %4530 to i32, !dbg !45 + %4675 = bitcast <2 x bfloat> %4531 to i32, !dbg !45 + %4676 = bitcast <2 x bfloat> %4532 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4673, i32 %4674, i32 %4675, i32 %4676, ptr addrspace(1) %311, i1 %834) #4, !dbg !45 + %4677 = bitcast <2 x bfloat> %4533 to i32, !dbg !45 + %4678 = bitcast <2 x bfloat> %4534 to i32, !dbg !45 + %4679 = bitcast <2 x bfloat> %4535 to i32, !dbg !45 + %4680 = bitcast <2 x bfloat> %4536 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4677, i32 %4678, i32 %4679, i32 %4680, ptr addrspace(1) %312, i1 %852) #4, !dbg !45 + %4681 = bitcast <2 x bfloat> %4537 to i32, !dbg !45 + %4682 = bitcast <2 x bfloat> %4538 to i32, !dbg !45 + %4683 = bitcast <2 x bfloat> %4539 to i32, !dbg !45 + %4684 = bitcast <2 x bfloat> %4540 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4681, i32 %4682, i32 %4683, i32 %4684, ptr addrspace(1) %313, i1 %870) #4, !dbg !45 + ret void, !dbg !46 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #2 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="64" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind } +attributes #3 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #4 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = distinct !DISubprogram(name: "l2norm_bwd_kernel", linkageName: "l2norm_bwd_kernel", scope: !1, file: !1, line: 93, type: !5, scopeLine: 93, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!5 = !DISubroutineType(cc: DW_CC_normal, types: !6) +!6 = !{null, !7, !9, !7, !7, !10, !11, !7, !7} +!7 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !8, size: 64, dwarfAddressSpace: 1) +!8 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!9 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !10, size: 64, dwarfAddressSpace: 1) +!10 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!11 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!12 = !DILocation(line: 94, column: 24, scope: !4) +!13 = !DILocation(line: 94, column: 30, scope: !4) +!14 = !DILocation(line: 95, column: 16, scope: !4) +!15 = !DILocation(line: 95, column: 34, scope: !4) +!16 = !DILocation(line: 95, column: 21, scope: !4) +!17 = !DILocation(line: 97, column: 16, scope: !4) +!18 = !DILocation(line: 98, column: 30, scope: !4) +!19 = !DILocation(line: 99, column: 29, scope: !4) +!20 = !DILocation(line: 99, column: 14, scope: !4) +!21 = !DILocation(line: 99, column: 33, scope: !4) +!22 = !DILocation(line: 100, column: 20, scope: !4) +!23 = !DILocation(line: 101, column: 16, scope: !4) +!24 = !DILocation(line: 101, column: 35, scope: !4) +!25 = !DILocation(line: 102, column: 16, scope: !4) +!26 = !DILocation(line: 102, column: 35, scope: !4) +!27 = !DILocation(line: 104, column: 18, scope: !4) +!28 = !DILocation(line: 104, column: 47, scope: !4) +!29 = !DILocation(line: 105, column: 21, scope: !4) +!30 = !DILocation(line: 106, column: 19, scope: !4) +!31 = !DILocation(line: 106, column: 49, scope: !4) +!32 = !DILocation(line: 107, column: 25, scope: !4) +!33 = !DILocation(line: 107, column: 18, scope: !4) +!34 = !DILocation(line: 107, column: 50, scope: !4) +!35 = !DILocation(line: 263, column: 15, scope: !36, inlinedAt: !38) +!36 = distinct !DILexicalBlockFile(scope: !4, file: !37, discriminator: 0) +!37 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!38 = !DILocation(line: 293, column: 36, scope: !36, inlinedAt: !39) +!39 = !DILocation(line: 107, column: 55, scope: !40) +!40 = distinct !DILexicalBlockFile(scope: !4, file: !1, discriminator: 0) +!41 = !DILocation(line: 107, column: 69, scope: !4) +!42 = !DILocation(line: 107, column: 75, scope: !4) +!43 = !DILocation(line: 107, column: 36, scope: !4) +!44 = !DILocation(line: 108, column: 27, scope: !4) +!45 = !DILocation(line: 108, column: 19, scope: !4) +!46 = !DILocation(line: 108, column: 4, scope: !4) diff --git a/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.ptx b/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..11d855bd123cb50f0f26a193fefd6b54cd056c8a --- /dev/null +++ b/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.ptx @@ -0,0 +1,4535 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_bwd_kernel // -- Begin function l2norm_bwd_kernel +.extern .shared .align 16 .b8 global_smem[]; + // @l2norm_bwd_kernel +.visible .entry l2norm_bwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_2, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_3, + .param .f32 l2norm_bwd_kernel_param_4, + .param .u32 l2norm_bwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_6, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_7 +) +.reqntid 64 +{ + .reg .pred %p<50>; + .reg .b16 %rs<566>; + .reg .b32 %r<2556>; + .reg .b64 %rd<664>; + .loc 1 93 0 // l2norm.py:93:0 +$L__func_begin0: + .loc 1 93 0 // l2norm.py:93:0 + +// %bb.0: + ld.param.b64 %rd99, [l2norm_bwd_kernel_param_0]; + ld.param.b64 %rd100, [l2norm_bwd_kernel_param_1]; +$L__tmp0: + .loc 1 94 24 // l2norm.py:94:24 + mov.u32 %r390, %ctaid.x; + ld.param.b64 %rd101, [l2norm_bwd_kernel_param_2]; + .loc 1 95 16 // l2norm.py:95:16 + mul.wide.u32 %rd102, %r390, 128; + ld.param.b64 %rd103, [l2norm_bwd_kernel_param_3]; + .loc 1 95 34 // l2norm.py:95:34 + mov.u32 %r391, %tid.x; + shr.u32 %r392, %r391, 4; + ld.param.s32 %rd104, [l2norm_bwd_kernel_param_5]; + bfe.u32 %r393, %r391, 4, 2; + or.b32 %r394, %r393, 4; + shl.b32 %r395, %r391, 1; + .loc 1 95 21 // l2norm.py:95:21 + cvt.u64.u32 %rd105, %r393; + cvt.u64.u32 %rd106, %r394; + or.b64 %rd107, %rd102, %rd105; + or.b64 %rd108, %rd102, %rd106; + .loc 1 97 16 // l2norm.py:97:16 + setp.lt.s64 %p1, %rd107, %rd104; + setp.lt.s64 %p2, %rd108, %rd104; + .loc 1 95 34 // l2norm.py:95:34 + or.b32 %r396, %r393, 120; + or.b32 %r397, %r392, 124; + and.b32 %r398, %r395, 126; + or.b32 %r399, %r398, 1; + .loc 1 95 21 // l2norm.py:95:21 + or.b32 %r400, %r393, 8; + or.b32 %r401, %r393, 12; + or.b32 %r402, %r393, 16; + or.b32 %r403, %r393, 20; + or.b32 %r404, %r393, 24; + or.b32 %r405, %r393, 28; + or.b32 %r406, %r393, 32; + or.b32 %r407, %r393, 36; + or.b32 %r408, %r393, 40; + or.b32 %r409, %r393, 44; + or.b32 %r410, %r393, 48; + or.b32 %r411, %r393, 52; + or.b32 %r412, %r393, 56; + or.b32 %r413, %r392, 60; + or.b32 %r414, %r393, 64; + or.b32 %r415, %r393, 68; + or.b32 %r416, %r393, 96; + or.b32 %r417, %r393, 100; + or.b32 %r418, %r393, 88; + or.b32 %r419, %r393, 92; + or.b32 %r420, %r393, 80; + or.b32 %r421, %r393, 84; + or.b32 %r422, %r393, 72; + or.b32 %r423, %r393, 76; + or.b32 %r424, %r393, 112; + or.b32 %r425, %r393, 116; + or.b32 %r426, %r393, 104; + or.b32 %r427, %r393, 108; + cvt.u64.u32 %rd109, %r423; + cvt.u64.u32 %rd110, %r422; + cvt.u64.u32 %rd111, %r421; + cvt.u64.u32 %rd112, %r420; + cvt.u64.u32 %rd113, %r419; + cvt.u64.u32 %rd114, %r418; + cvt.u64.u32 %rd115, %r417; + cvt.u64.u32 %rd116, %r416; + cvt.u64.u32 %rd117, %r427; + cvt.u64.u32 %rd118, %r426; + cvt.u64.u32 %rd119, %r425; + cvt.u64.u32 %rd120, %r424; + cvt.u64.u32 %rd121, %r397; + cvt.u64.u32 %rd122, %r396; + cvt.u64.u32 %rd123, %r398; + cvt.u64.u32 %rd124, %r399; + cvt.u64.u32 %rd125, %r415; + cvt.u64.u32 %rd126, %r414; + cvt.u64.u32 %rd127, %r413; + cvt.u64.u32 %rd128, %r412; + cvt.u64.u32 %rd129, %r411; + cvt.u64.u32 %rd130, %r410; + cvt.u64.u32 %rd131, %r409; + cvt.u64.u32 %rd132, %r408; + cvt.u64.u32 %rd133, %r407; + cvt.u64.u32 %rd134, %r406; + cvt.u64.u32 %rd135, %r405; + cvt.u64.u32 %rd136, %r404; + cvt.u64.u32 %rd137, %r403; + cvt.u64.u32 %rd138, %r402; + cvt.u64.u32 %rd139, %r401; + cvt.u64.u32 %rd140, %r400; + or.b64 %rd141, %rd102, %rd140; + or.b64 %rd142, %rd102, %rd139; + or.b64 %rd143, %rd102, %rd138; + or.b64 %rd144, %rd102, %rd137; + or.b64 %rd145, %rd102, %rd136; + or.b64 %rd146, %rd102, %rd135; + or.b64 %rd147, %rd102, %rd134; + or.b64 %rd148, %rd102, %rd133; + or.b64 %rd149, %rd102, %rd132; + or.b64 %rd150, %rd102, %rd131; + or.b64 %rd151, %rd102, %rd130; + or.b64 %rd152, %rd102, %rd129; + or.b64 %rd153, %rd102, %rd128; + or.b64 %rd154, %rd102, %rd127; + or.b64 %rd155, %rd102, %rd126; + or.b64 %rd156, %rd102, %rd125; + or.b64 %rd157, %rd102, %rd124; + or.b64 %rd158, %rd102, %rd123; + or.b64 %rd159, %rd102, %rd122; + or.b64 %rd160, %rd102, %rd121; + or.b64 %rd161, %rd102, %rd120; + or.b64 %rd162, %rd102, %rd119; + or.b64 %rd163, %rd102, %rd118; + or.b64 %rd164, %rd102, %rd117; + or.b64 %rd165, %rd102, %rd116; + or.b64 %rd166, %rd102, %rd115; + or.b64 %rd167, %rd102, %rd114; + or.b64 %rd168, %rd102, %rd113; + or.b64 %rd169, %rd102, %rd112; + or.b64 %rd170, %rd102, %rd111; + or.b64 %rd171, %rd102, %rd110; + or.b64 %rd172, %rd102, %rd109; + .loc 1 97 16 // l2norm.py:97:16 + setp.lt.s64 %p35, %rd172, %rd104; + selp.b16 %rs1, 1, 0, %p35; + shl.b16 %rs2, %rs1, 2; + setp.lt.s64 %p36, %rd171, %rd104; + selp.b16 %rs3, -1, 0, %p36; + shl.b16 %rs4, %rs3, 3; + or.b16 %rs5, %rs4, %rs2; + setp.lt.s64 %p37, %rd170, %rd104; + selp.b16 %rs6, 1, 0, %p37; + setp.lt.s64 %p38, %rd169, %rd104; + selp.b16 %rs7, -1, 0, %p38; + shl.b16 %rs8, %rs7, 1; + or.b16 %rs9, %rs6, %rs8; + and.b16 %rs10, %rs9, 3; + or.b16 %rs11, %rs10, %rs5; + shl.b16 %rs12, %rs11, 12; + setp.lt.s64 %p39, %rd168, %rd104; + selp.b16 %rs13, 1, 0, %p39; + shl.b16 %rs14, %rs13, 2; + setp.lt.s64 %p40, %rd167, %rd104; + selp.b16 %rs15, -1, 0, %p40; + shl.b16 %rs16, %rs15, 3; + or.b16 %rs17, %rs16, %rs14; + setp.lt.s64 %p41, %rd166, %rd104; + selp.b16 %rs18, 1, 0, %p41; + setp.lt.s64 %p42, %rd165, %rd104; + selp.b16 %rs19, -1, 0, %p42; + shl.b16 %rs20, %rs19, 1; + or.b16 %rs21, %rs18, %rs20; + and.b16 %rs22, %rs21, 3; + or.b16 %rs23, %rs22, %rs17; + and.b16 %rs24, %rs23, 15; + shl.b16 %rs25, %rs24, 8; + or.b16 %rs26, %rs12, %rs25; + setp.lt.s64 %p43, %rd164, %rd104; + selp.b16 %rs27, 1, 0, %p43; + shl.b16 %rs28, %rs27, 2; + setp.lt.s64 %p44, %rd163, %rd104; + selp.b16 %rs29, -1, 0, %p44; + shl.b16 %rs30, %rs29, 3; + or.b16 %rs31, %rs30, %rs28; + setp.lt.s64 %p45, %rd162, %rd104; + selp.b16 %rs32, 1, 0, %p45; + setp.lt.s64 %p46, %rd161, %rd104; + selp.b16 %rs33, -1, 0, %p46; + shl.b16 %rs34, %rs33, 1; + or.b16 %rs35, %rs32, %rs34; + and.b16 %rs36, %rs35, 3; + or.b16 %rs37, %rs36, %rs31; + shl.b16 %rs38, %rs37, 4; + setp.lt.s64 %p47, %rd160, %rd104; + selp.b16 %rs39, 1, 0, %p47; + shl.b16 %rs40, %rs39, 2; + setp.lt.s64 %p48, %rd159, %rd104; + selp.b16 %rs41, -1, 0, %p48; + shl.b16 %rs42, %rs41, 3; + or.b16 %rs43, %rs42, %rs40; + setp.lt.s64 %p49, %rd158, %rd104; + selp.b16 %rs44, -1, 0, %p49; + shl.b16 %rs45, %rs44, 1; + setp.lt.s64 %p5, %rd157, %rd104; + selp.b16 %rs46, 1, 0, %p5; + or.b16 %rs47, %rs46, %rs45; + and.b16 %rs48, %rs47, 3; + or.b16 %rs49, %rs48, %rs43; + and.b16 %rs50, %rs49, 15; + or.b16 %rs51, %rs50, %rs38; + and.b16 %rs52, %rs51, 255; + or.b16 %rs53, %rs52, %rs26; + cvt.u32.u16 %r428, %rs53; + setp.lt.s64 %p20, %rd156, %rd104; + setp.lt.s64 %p19, %rd155, %rd104; + setp.lt.s64 %p18, %rd154, %rd104; + setp.lt.s64 %p17, %rd153, %rd104; + setp.lt.s64 %p16, %rd152, %rd104; + setp.lt.s64 %p15, %rd151, %rd104; + setp.lt.s64 %p14, %rd150, %rd104; + setp.lt.s64 %p13, %rd149, %rd104; + setp.lt.s64 %p3, %rd148, %rd104; + setp.lt.s64 %p12, %rd147, %rd104; + setp.lt.s64 %p11, %rd146, %rd104; + setp.lt.s64 %p10, %rd145, %rd104; + setp.lt.s64 %p9, %rd144, %rd104; + setp.lt.s64 %p8, %rd143, %rd104; + setp.lt.s64 %p7, %rd142, %rd104; + setp.lt.s64 %p6, %rd141, %rd104; + .loc 1 98 30 // l2norm.py:98:30 + shl.b32 %r429, %r391, 3; + and.b32 %r430, %r429, 120; + .loc 1 99 14 // l2norm.py:99:14 + shl.b64 %rd173, %rd107, 8; + add.s64 %rd174, %rd99, %rd173; + shl.b64 %rd175, %rd108, 8; + add.s64 %rd176, %rd99, %rd175; + shl.b64 %rd177, %rd141, 8; + add.s64 %rd178, %rd99, %rd177; + shl.b64 %rd179, %rd142, 8; + add.s64 %rd180, %rd99, %rd179; + shl.b64 %rd181, %rd143, 8; + add.s64 %rd182, %rd99, %rd181; + shl.b64 %rd183, %rd144, 8; + add.s64 %rd184, %rd99, %rd183; + shl.b64 %rd185, %rd145, 8; + add.s64 %rd186, %rd99, %rd185; + shl.b64 %rd187, %rd146, 8; + add.s64 %rd188, %rd99, %rd187; + shl.b64 %rd189, %rd147, 8; + add.s64 %rd190, %rd99, %rd189; + shl.b64 %rd191, %rd148, 8; + add.s64 %rd192, %rd99, %rd191; + shl.b64 %rd193, %rd149, 8; + add.s64 %rd194, %rd99, %rd193; + shl.b64 %rd195, %rd150, 8; + add.s64 %rd196, %rd99, %rd195; + shl.b64 %rd197, %rd151, 8; + add.s64 %rd198, %rd99, %rd197; + shl.b64 %rd199, %rd152, 8; + add.s64 %rd200, %rd99, %rd199; + shl.b64 %rd201, %rd153, 8; + add.s64 %rd202, %rd99, %rd201; + shl.b64 %rd203, %rd154, 8; + add.s64 %rd204, %rd99, %rd203; + shl.b64 %rd205, %rd155, 8; + add.s64 %rd206, %rd99, %rd205; + shl.b64 %rd207, %rd156, 8; + add.s64 %rd208, %rd99, %rd207; + shl.b64 %rd209, %rd171, 8; + add.s64 %rd210, %rd99, %rd209; + shl.b64 %rd211, %rd172, 8; + add.s64 %rd212, %rd99, %rd211; + shl.b64 %rd213, %rd169, 8; + add.s64 %rd214, %rd99, %rd213; + shl.b64 %rd215, %rd170, 8; + add.s64 %rd216, %rd99, %rd215; + shl.b64 %rd217, %rd167, 8; + add.s64 %rd218, %rd99, %rd217; + shl.b64 %rd219, %rd168, 8; + add.s64 %rd220, %rd99, %rd219; + shl.b64 %rd221, %rd165, 8; + add.s64 %rd222, %rd99, %rd221; + shl.b64 %rd223, %rd166, 8; + add.s64 %rd224, %rd99, %rd223; + shl.b64 %rd225, %rd163, 8; + add.s64 %rd226, %rd99, %rd225; + shl.b64 %rd227, %rd164, 8; + add.s64 %rd228, %rd99, %rd227; + shl.b64 %rd229, %rd161, 8; + add.s64 %rd230, %rd99, %rd229; + shl.b64 %rd231, %rd162, 8; + add.s64 %rd232, %rd99, %rd231; + shl.b64 %rd233, %rd159, 8; + add.s64 %rd234, %rd99, %rd233; + shl.b64 %rd235, %rd160, 8; + add.s64 %rd236, %rd99, %rd235; + .loc 1 99 33 // l2norm.py:99:33 + mul.wide.u32 %rd237, %r430, 2; + add.s64 %rd1, %rd174, %rd237; + add.s64 %rd2, %rd176, %rd237; + add.s64 %rd3, %rd178, %rd237; + add.s64 %rd4, %rd180, %rd237; + add.s64 %rd5, %rd182, %rd237; + add.s64 %rd6, %rd184, %rd237; + add.s64 %rd7, %rd186, %rd237; + add.s64 %rd8, %rd188, %rd237; + add.s64 %rd9, %rd190, %rd237; + add.s64 %rd10, %rd192, %rd237; + add.s64 %rd11, %rd194, %rd237; + add.s64 %rd12, %rd196, %rd237; + add.s64 %rd13, %rd198, %rd237; + add.s64 %rd14, %rd200, %rd237; + add.s64 %rd15, %rd202, %rd237; + add.s64 %rd16, %rd204, %rd237; + add.s64 %rd17, %rd206, %rd237; + add.s64 %rd18, %rd208, %rd237; + add.s64 %rd19, %rd210, %rd237; + add.s64 %rd20, %rd212, %rd237; + add.s64 %rd21, %rd214, %rd237; + add.s64 %rd22, %rd216, %rd237; + add.s64 %rd23, %rd218, %rd237; + add.s64 %rd24, %rd220, %rd237; + add.s64 %rd25, %rd222, %rd237; + add.s64 %rd26, %rd224, %rd237; + add.s64 %rd27, %rd226, %rd237; + add.s64 %rd28, %rd228, %rd237; + add.s64 %rd29, %rd230, %rd237; + add.s64 %rd30, %rd232, %rd237; + add.s64 %rd31, %rd234, %rd237; + add.s64 %rd32, %rd236, %rd237; + .loc 1 100 20 // l2norm.py:100:20 + shl.b64 %rd238, %rd158, 2; + add.s64 %rd33, %rd100, %rd238; + shl.b64 %rd239, %rd157, 2; + add.s64 %rd34, %rd100, %rd239; + .loc 1 101 16 // l2norm.py:101:16 + add.s64 %rd240, %rd101, %rd173; + add.s64 %rd241, %rd101, %rd175; + add.s64 %rd242, %rd101, %rd177; + add.s64 %rd243, %rd101, %rd179; + add.s64 %rd244, %rd101, %rd181; + add.s64 %rd245, %rd101, %rd183; + add.s64 %rd246, %rd101, %rd185; + add.s64 %rd247, %rd101, %rd187; + add.s64 %rd248, %rd101, %rd189; + add.s64 %rd249, %rd101, %rd191; + add.s64 %rd250, %rd101, %rd193; + add.s64 %rd251, %rd101, %rd195; + add.s64 %rd252, %rd101, %rd197; + add.s64 %rd253, %rd101, %rd199; + add.s64 %rd254, %rd101, %rd201; + add.s64 %rd255, %rd101, %rd203; + add.s64 %rd256, %rd101, %rd205; + add.s64 %rd257, %rd101, %rd207; + add.s64 %rd258, %rd101, %rd209; + add.s64 %rd259, %rd101, %rd211; + add.s64 %rd260, %rd101, %rd213; + add.s64 %rd261, %rd101, %rd215; + add.s64 %rd262, %rd101, %rd217; + add.s64 %rd263, %rd101, %rd219; + add.s64 %rd264, %rd101, %rd221; + add.s64 %rd265, %rd101, %rd223; + add.s64 %rd266, %rd101, %rd225; + add.s64 %rd267, %rd101, %rd227; + add.s64 %rd268, %rd101, %rd229; + add.s64 %rd269, %rd101, %rd231; + add.s64 %rd270, %rd101, %rd233; + add.s64 %rd271, %rd101, %rd235; + .loc 1 101 35 // l2norm.py:101:35 + add.s64 %rd35, %rd240, %rd237; + add.s64 %rd36, %rd241, %rd237; + add.s64 %rd37, %rd242, %rd237; + add.s64 %rd38, %rd243, %rd237; + add.s64 %rd39, %rd244, %rd237; + add.s64 %rd40, %rd245, %rd237; + add.s64 %rd41, %rd246, %rd237; + add.s64 %rd42, %rd247, %rd237; + add.s64 %rd43, %rd248, %rd237; + add.s64 %rd44, %rd249, %rd237; + add.s64 %rd45, %rd250, %rd237; + add.s64 %rd46, %rd251, %rd237; + add.s64 %rd47, %rd252, %rd237; + add.s64 %rd48, %rd253, %rd237; + add.s64 %rd49, %rd254, %rd237; + add.s64 %rd50, %rd255, %rd237; + add.s64 %rd51, %rd256, %rd237; + add.s64 %rd52, %rd257, %rd237; + add.s64 %rd53, %rd258, %rd237; + add.s64 %rd54, %rd259, %rd237; + add.s64 %rd55, %rd260, %rd237; + add.s64 %rd56, %rd261, %rd237; + add.s64 %rd57, %rd262, %rd237; + add.s64 %rd58, %rd263, %rd237; + add.s64 %rd59, %rd264, %rd237; + add.s64 %rd60, %rd265, %rd237; + add.s64 %rd61, %rd266, %rd237; + add.s64 %rd62, %rd267, %rd237; + add.s64 %rd63, %rd268, %rd237; + add.s64 %rd64, %rd269, %rd237; + add.s64 %rd65, %rd270, %rd237; + add.s64 %rd66, %rd271, %rd237; + .loc 1 102 16 // l2norm.py:102:16 + add.s64 %rd272, %rd103, %rd173; + add.s64 %rd273, %rd103, %rd175; + add.s64 %rd274, %rd103, %rd177; + add.s64 %rd275, %rd103, %rd179; + add.s64 %rd276, %rd103, %rd181; + add.s64 %rd277, %rd103, %rd183; + add.s64 %rd278, %rd103, %rd185; + add.s64 %rd279, %rd103, %rd187; + add.s64 %rd280, %rd103, %rd189; + add.s64 %rd281, %rd103, %rd191; + add.s64 %rd282, %rd103, %rd193; + add.s64 %rd283, %rd103, %rd195; + add.s64 %rd284, %rd103, %rd197; + add.s64 %rd285, %rd103, %rd199; + add.s64 %rd286, %rd103, %rd201; + add.s64 %rd287, %rd103, %rd203; + add.s64 %rd288, %rd103, %rd205; + add.s64 %rd289, %rd103, %rd207; + add.s64 %rd290, %rd103, %rd209; + add.s64 %rd291, %rd103, %rd211; + add.s64 %rd292, %rd103, %rd213; + add.s64 %rd293, %rd103, %rd215; + add.s64 %rd294, %rd103, %rd217; + add.s64 %rd295, %rd103, %rd219; + add.s64 %rd296, %rd103, %rd221; + add.s64 %rd297, %rd103, %rd223; + add.s64 %rd298, %rd103, %rd225; + add.s64 %rd299, %rd103, %rd227; + add.s64 %rd300, %rd103, %rd229; + add.s64 %rd301, %rd103, %rd231; + add.s64 %rd302, %rd103, %rd233; + add.s64 %rd303, %rd103, %rd235; + .loc 1 102 35 // l2norm.py:102:35 + add.s64 %rd67, %rd272, %rd237; + add.s64 %rd68, %rd273, %rd237; + add.s64 %rd69, %rd274, %rd237; + add.s64 %rd70, %rd275, %rd237; + add.s64 %rd71, %rd276, %rd237; + add.s64 %rd72, %rd277, %rd237; + add.s64 %rd73, %rd278, %rd237; + add.s64 %rd74, %rd279, %rd237; + add.s64 %rd75, %rd280, %rd237; + add.s64 %rd76, %rd281, %rd237; + add.s64 %rd77, %rd282, %rd237; + add.s64 %rd78, %rd283, %rd237; + add.s64 %rd79, %rd284, %rd237; + add.s64 %rd80, %rd285, %rd237; + add.s64 %rd81, %rd286, %rd237; + add.s64 %rd82, %rd287, %rd237; + add.s64 %rd83, %rd288, %rd237; + add.s64 %rd84, %rd289, %rd237; + add.s64 %rd85, %rd290, %rd237; + add.s64 %rd86, %rd291, %rd237; + add.s64 %rd87, %rd292, %rd237; + add.s64 %rd88, %rd293, %rd237; + add.s64 %rd89, %rd294, %rd237; + add.s64 %rd90, %rd295, %rd237; + add.s64 %rd91, %rd296, %rd237; + add.s64 %rd92, %rd297, %rd237; + add.s64 %rd93, %rd298, %rd237; + add.s64 %rd94, %rd299, %rd237; + add.s64 %rd95, %rd300, %rd237; + add.s64 %rd96, %rd301, %rd237; + add.s64 %rd97, %rd302, %rd237; + add.s64 %rd98, %rd303, %rd237; + mov.b32 %r5, 0; + .loc 1 104 18 // l2norm.py:104:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs54, %rs55}, %r1; + mov.b32 {%rs56, %rs57}, %r2; + mov.b32 {%rs58, %rs59}, %r3; + mov.b32 {%rs60, %rs61}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs62, %rs63}, %r6; + mov.b32 {%rs64, %rs65}, %r7; + mov.b32 {%rs66, %rs67}, %r8; + mov.b32 {%rs68, %rs69}, %r9; + // begin inline asm + mov.u32 %r10, %r5; + mov.u32 %r11, %r5; + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + @%p6 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs70, %rs71}, %r10; + mov.b32 {%rs72, %rs73}, %r11; + mov.b32 {%rs74, %rs75}, %r12; + mov.b32 {%rs76, %rs77}, %r13; + // begin inline asm + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + @%p7 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ]; + // end inline asm + mov.b32 {%rs78, %rs79}, %r14; + mov.b32 {%rs80, %rs81}, %r15; + mov.b32 {%rs82, %rs83}, %r16; + mov.b32 {%rs84, %rs85}, %r17; + // begin inline asm + mov.u32 %r18, %r5; + mov.u32 %r19, %r5; + mov.u32 %r20, %r5; + mov.u32 %r21, %r5; + @%p8 ld.global.v4.b32 { %r18, %r19, %r20, %r21 }, [ %rd5 + 0 ]; + // end inline asm + mov.b32 {%rs86, %rs87}, %r18; + mov.b32 {%rs88, %rs89}, %r19; + mov.b32 {%rs90, %rs91}, %r20; + mov.b32 {%rs92, %rs93}, %r21; + // begin inline asm + mov.u32 %r22, %r5; + mov.u32 %r23, %r5; + mov.u32 %r24, %r5; + mov.u32 %r25, %r5; + @%p9 ld.global.v4.b32 { %r22, %r23, %r24, %r25 }, [ %rd6 + 0 ]; + // end inline asm + mov.b32 {%rs94, %rs95}, %r22; + mov.b32 {%rs96, %rs97}, %r23; + mov.b32 {%rs98, %rs99}, %r24; + mov.b32 {%rs100, %rs101}, %r25; + // begin inline asm + mov.u32 %r26, %r5; + mov.u32 %r27, %r5; + mov.u32 %r28, %r5; + mov.u32 %r29, %r5; + @%p10 ld.global.v4.b32 { %r26, %r27, %r28, %r29 }, [ %rd7 + 0 ]; + // end inline asm + mov.b32 {%rs102, %rs103}, %r26; + mov.b32 {%rs104, %rs105}, %r27; + mov.b32 {%rs106, %rs107}, %r28; + mov.b32 {%rs108, %rs109}, %r29; + // begin inline asm + mov.u32 %r30, %r5; + mov.u32 %r31, %r5; + mov.u32 %r32, %r5; + mov.u32 %r33, %r5; + @%p11 ld.global.v4.b32 { %r30, %r31, %r32, %r33 }, [ %rd8 + 0 ]; + // end inline asm + mov.b32 {%rs110, %rs111}, %r30; + mov.b32 {%rs112, %rs113}, %r31; + mov.b32 {%rs114, %rs115}, %r32; + mov.b32 {%rs116, %rs117}, %r33; + // begin inline asm + mov.u32 %r34, %r5; + mov.u32 %r35, %r5; + mov.u32 %r36, %r5; + mov.u32 %r37, %r5; + @%p12 ld.global.v4.b32 { %r34, %r35, %r36, %r37 }, [ %rd9 + 0 ]; + // end inline asm + mov.b32 {%rs118, %rs119}, %r34; + mov.b32 {%rs120, %rs121}, %r35; + mov.b32 {%rs122, %rs123}, %r36; + mov.b32 {%rs124, %rs125}, %r37; + // begin inline asm + mov.u32 %r38, %r5; + mov.u32 %r39, %r5; + mov.u32 %r40, %r5; + mov.u32 %r41, %r5; + @%p3 ld.global.v4.b32 { %r38, %r39, %r40, %r41 }, [ %rd10 + 0 ]; + // end inline asm + mov.b32 {%rs126, %rs127}, %r38; + mov.b32 {%rs128, %rs129}, %r39; + mov.b32 {%rs130, %rs131}, %r40; + mov.b32 {%rs132, %rs133}, %r41; + // begin inline asm + mov.u32 %r42, %r5; + mov.u32 %r43, %r5; + mov.u32 %r44, %r5; + mov.u32 %r45, %r5; + @%p13 ld.global.v4.b32 { %r42, %r43, %r44, %r45 }, [ %rd11 + 0 ]; + // end inline asm + mov.b32 {%rs134, %rs135}, %r42; + mov.b32 {%rs136, %rs137}, %r43; + mov.b32 {%rs138, %rs139}, %r44; + mov.b32 {%rs140, %rs141}, %r45; + // begin inline asm + mov.u32 %r46, %r5; + mov.u32 %r47, %r5; + mov.u32 %r48, %r5; + mov.u32 %r49, %r5; + @%p14 ld.global.v4.b32 { %r46, %r47, %r48, %r49 }, [ %rd12 + 0 ]; + // end inline asm + mov.b32 {%rs142, %rs143}, %r46; + mov.b32 {%rs144, %rs145}, %r47; + mov.b32 {%rs146, %rs147}, %r48; + mov.b32 {%rs148, %rs149}, %r49; + // begin inline asm + mov.u32 %r50, %r5; + mov.u32 %r51, %r5; + mov.u32 %r52, %r5; + mov.u32 %r53, %r5; + @%p15 ld.global.v4.b32 { %r50, %r51, %r52, %r53 }, [ %rd13 + 0 ]; + // end inline asm + mov.b32 {%rs150, %rs151}, %r50; + mov.b32 {%rs152, %rs153}, %r51; + mov.b32 {%rs154, %rs155}, %r52; + mov.b32 {%rs156, %rs157}, %r53; + // begin inline asm + mov.u32 %r54, %r5; + mov.u32 %r55, %r5; + mov.u32 %r56, %r5; + mov.u32 %r57, %r5; + @%p16 ld.global.v4.b32 { %r54, %r55, %r56, %r57 }, [ %rd14 + 0 ]; + // end inline asm + mov.b32 {%rs158, %rs159}, %r54; + mov.b32 {%rs160, %rs161}, %r55; + mov.b32 {%rs162, %rs163}, %r56; + mov.b32 {%rs164, %rs165}, %r57; + // begin inline asm + mov.u32 %r58, %r5; + mov.u32 %r59, %r5; + mov.u32 %r60, %r5; + mov.u32 %r61, %r5; + @%p17 ld.global.v4.b32 { %r58, %r59, %r60, %r61 }, [ %rd15 + 0 ]; + // end inline asm + mov.b32 {%rs166, %rs167}, %r58; + mov.b32 {%rs168, %rs169}, %r59; + mov.b32 {%rs170, %rs171}, %r60; + mov.b32 {%rs172, %rs173}, %r61; + // begin inline asm + mov.u32 %r62, %r5; + mov.u32 %r63, %r5; + mov.u32 %r64, %r5; + mov.u32 %r65, %r5; + @%p18 ld.global.v4.b32 { %r62, %r63, %r64, %r65 }, [ %rd16 + 0 ]; + // end inline asm + mov.b32 {%rs174, %rs175}, %r62; + mov.b32 {%rs176, %rs177}, %r63; + mov.b32 {%rs178, %rs179}, %r64; + mov.b32 {%rs180, %rs181}, %r65; + // begin inline asm + mov.u32 %r66, %r5; + mov.u32 %r67, %r5; + mov.u32 %r68, %r5; + mov.u32 %r69, %r5; + @%p19 ld.global.v4.b32 { %r66, %r67, %r68, %r69 }, [ %rd17 + 0 ]; + // end inline asm + mov.b32 {%rs182, %rs183}, %r66; + mov.b32 {%rs184, %rs185}, %r67; + mov.b32 {%rs186, %rs187}, %r68; + mov.b32 {%rs188, %rs189}, %r69; + // begin inline asm + mov.u32 %r70, %r5; + mov.u32 %r71, %r5; + mov.u32 %r72, %r5; + mov.u32 %r73, %r5; + @%p20 ld.global.v4.b32 { %r70, %r71, %r72, %r73 }, [ %rd18 + 0 ]; + // end inline asm + mov.b32 {%rs190, %rs191}, %r70; + mov.b32 {%rs192, %rs193}, %r71; + mov.b32 {%rs194, %rs195}, %r72; + mov.b32 {%rs196, %rs197}, %r73; + shr.u32 %r431, %r428, 15; + and.b32 %r432, %r431, 1; + setp.ne.b32 %p21, %r432, 0; + // begin inline asm + mov.u32 %r74, %r5; + mov.u32 %r75, %r5; + mov.u32 %r76, %r5; + mov.u32 %r77, %r5; + @%p21 ld.global.v4.b32 { %r74, %r75, %r76, %r77 }, [ %rd19 + 0 ]; + // end inline asm + mov.b32 {%rs198, %rs199}, %r74; + mov.b32 {%rs200, %rs201}, %r75; + mov.b32 {%rs202, %rs203}, %r76; + mov.b32 {%rs204, %rs205}, %r77; + shr.u32 %r433, %r428, 14; + and.b32 %r434, %r433, 1; + setp.ne.b32 %p22, %r434, 0; + // begin inline asm + mov.u32 %r78, %r5; + mov.u32 %r79, %r5; + mov.u32 %r80, %r5; + mov.u32 %r81, %r5; + @%p22 ld.global.v4.b32 { %r78, %r79, %r80, %r81 }, [ %rd20 + 0 ]; + // end inline asm + mov.b32 {%rs206, %rs207}, %r78; + mov.b32 {%rs208, %rs209}, %r79; + mov.b32 {%rs210, %rs211}, %r80; + mov.b32 {%rs212, %rs213}, %r81; + shr.u32 %r435, %r428, 13; + and.b32 %r436, %r435, 1; + setp.ne.b32 %p23, %r436, 0; + // begin inline asm + mov.u32 %r82, %r5; + mov.u32 %r83, %r5; + mov.u32 %r84, %r5; + mov.u32 %r85, %r5; + @%p23 ld.global.v4.b32 { %r82, %r83, %r84, %r85 }, [ %rd21 + 0 ]; + // end inline asm + mov.b32 {%rs214, %rs215}, %r82; + mov.b32 {%rs216, %rs217}, %r83; + mov.b32 {%rs218, %rs219}, %r84; + mov.b32 {%rs220, %rs221}, %r85; + shr.u32 %r437, %r428, 12; + and.b32 %r438, %r437, 1; + setp.ne.b32 %p24, %r438, 0; + // begin inline asm + mov.u32 %r86, %r5; + mov.u32 %r87, %r5; + mov.u32 %r88, %r5; + mov.u32 %r89, %r5; + @%p24 ld.global.v4.b32 { %r86, %r87, %r88, %r89 }, [ %rd22 + 0 ]; + // end inline asm + mov.b32 {%rs222, %rs223}, %r86; + mov.b32 {%rs224, %rs225}, %r87; + mov.b32 {%rs226, %rs227}, %r88; + mov.b32 {%rs228, %rs229}, %r89; + shr.u32 %r439, %r428, 11; + and.b32 %r440, %r439, 1; + setp.ne.b32 %p25, %r440, 0; + // begin inline asm + mov.u32 %r90, %r5; + mov.u32 %r91, %r5; + mov.u32 %r92, %r5; + mov.u32 %r93, %r5; + @%p25 ld.global.v4.b32 { %r90, %r91, %r92, %r93 }, [ %rd23 + 0 ]; + // end inline asm + mov.b32 {%rs230, %rs231}, %r90; + mov.b32 {%rs232, %rs233}, %r91; + mov.b32 {%rs234, %rs235}, %r92; + mov.b32 {%rs236, %rs237}, %r93; + shr.u32 %r441, %r428, 10; + and.b32 %r442, %r441, 1; + setp.ne.b32 %p26, %r442, 0; + // begin inline asm + mov.u32 %r94, %r5; + mov.u32 %r95, %r5; + mov.u32 %r96, %r5; + mov.u32 %r97, %r5; + @%p26 ld.global.v4.b32 { %r94, %r95, %r96, %r97 }, [ %rd24 + 0 ]; + // end inline asm + mov.b32 {%rs238, %rs239}, %r94; + mov.b32 {%rs240, %rs241}, %r95; + mov.b32 {%rs242, %rs243}, %r96; + mov.b32 {%rs244, %rs245}, %r97; + shr.u32 %r443, %r428, 9; + and.b32 %r444, %r443, 1; + setp.ne.b32 %p27, %r444, 0; + // begin inline asm + mov.u32 %r98, %r5; + mov.u32 %r99, %r5; + mov.u32 %r100, %r5; + mov.u32 %r101, %r5; + @%p27 ld.global.v4.b32 { %r98, %r99, %r100, %r101 }, [ %rd25 + 0 ]; + // end inline asm + mov.b32 {%rs246, %rs247}, %r98; + mov.b32 {%rs248, %rs249}, %r99; + mov.b32 {%rs250, %rs251}, %r100; + mov.b32 {%rs252, %rs253}, %r101; + shr.u32 %r445, %r428, 8; + and.b32 %r446, %r445, 1; + setp.ne.b32 %p28, %r446, 0; + // begin inline asm + mov.u32 %r102, %r5; + mov.u32 %r103, %r5; + mov.u32 %r104, %r5; + mov.u32 %r105, %r5; + @%p28 ld.global.v4.b32 { %r102, %r103, %r104, %r105 }, [ %rd26 + 0 ]; + // end inline asm + mov.b32 {%rs254, %rs255}, %r102; + mov.b32 {%rs256, %rs257}, %r103; + mov.b32 {%rs258, %rs259}, %r104; + mov.b32 {%rs260, %rs261}, %r105; + shr.u32 %r447, %r428, 7; + and.b32 %r448, %r447, 1; + setp.ne.b32 %p29, %r448, 0; + // begin inline asm + mov.u32 %r106, %r5; + mov.u32 %r107, %r5; + mov.u32 %r108, %r5; + mov.u32 %r109, %r5; + @%p29 ld.global.v4.b32 { %r106, %r107, %r108, %r109 }, [ %rd27 + 0 ]; + // end inline asm + mov.b32 {%rs262, %rs263}, %r106; + mov.b32 {%rs264, %rs265}, %r107; + mov.b32 {%rs266, %rs267}, %r108; + mov.b32 {%rs268, %rs269}, %r109; + shr.u32 %r449, %r428, 6; + and.b32 %r450, %r449, 1; + setp.ne.b32 %p30, %r450, 0; + // begin inline asm + mov.u32 %r110, %r5; + mov.u32 %r111, %r5; + mov.u32 %r112, %r5; + mov.u32 %r113, %r5; + @%p30 ld.global.v4.b32 { %r110, %r111, %r112, %r113 }, [ %rd28 + 0 ]; + // end inline asm + mov.b32 {%rs270, %rs271}, %r110; + mov.b32 {%rs272, %rs273}, %r111; + mov.b32 {%rs274, %rs275}, %r112; + mov.b32 {%rs276, %rs277}, %r113; + shr.u32 %r451, %r428, 5; + and.b32 %r452, %r451, 1; + setp.ne.b32 %p31, %r452, 0; + // begin inline asm + mov.u32 %r114, %r5; + mov.u32 %r115, %r5; + mov.u32 %r116, %r5; + mov.u32 %r117, %r5; + @%p31 ld.global.v4.b32 { %r114, %r115, %r116, %r117 }, [ %rd29 + 0 ]; + // end inline asm + mov.b32 {%rs278, %rs279}, %r114; + mov.b32 {%rs280, %rs281}, %r115; + mov.b32 {%rs282, %rs283}, %r116; + mov.b32 {%rs284, %rs285}, %r117; + shr.u32 %r453, %r428, 4; + and.b32 %r454, %r453, 1; + setp.ne.b32 %p32, %r454, 0; + // begin inline asm + mov.u32 %r118, %r5; + mov.u32 %r119, %r5; + mov.u32 %r120, %r5; + mov.u32 %r121, %r5; + @%p32 ld.global.v4.b32 { %r118, %r119, %r120, %r121 }, [ %rd30 + 0 ]; + // end inline asm + mov.b32 {%rs286, %rs287}, %r118; + mov.b32 {%rs288, %rs289}, %r119; + mov.b32 {%rs290, %rs291}, %r120; + mov.b32 {%rs292, %rs293}, %r121; + shr.u32 %r455, %r428, 3; + and.b32 %r456, %r455, 1; + setp.ne.b32 %p33, %r456, 0; + // begin inline asm + mov.u32 %r122, %r5; + mov.u32 %r123, %r5; + mov.u32 %r124, %r5; + mov.u32 %r125, %r5; + @%p33 ld.global.v4.b32 { %r122, %r123, %r124, %r125 }, [ %rd31 + 0 ]; + // end inline asm + mov.b32 {%rs294, %rs295}, %r122; + mov.b32 {%rs296, %rs297}, %r123; + mov.b32 {%rs298, %rs299}, %r124; + mov.b32 {%rs300, %rs301}, %r125; + shr.u32 %r457, %r428, 2; + and.b32 %r458, %r457, 1; + setp.ne.b32 %p34, %r458, 0; + // begin inline asm + mov.u32 %r126, %r5; + mov.u32 %r127, %r5; + mov.u32 %r128, %r5; + mov.u32 %r129, %r5; + @%p34 ld.global.v4.b32 { %r126, %r127, %r128, %r129 }, [ %rd32 + 0 ]; + // end inline asm + mov.b32 {%rs302, %rs303}, %r126; + mov.b32 {%rs304, %rs305}, %r127; + mov.b32 {%rs306, %rs307}, %r128; + mov.b32 {%rs308, %rs309}, %r129; + .loc 1 104 47 // l2norm.py:104:47 + cvt.f32.bf16 %r459, %rs54; + cvt.f32.bf16 %r460, %rs55; + cvt.f32.bf16 %r461, %rs56; + cvt.f32.bf16 %r462, %rs57; + cvt.f32.bf16 %r463, %rs58; + cvt.f32.bf16 %r464, %rs59; + cvt.f32.bf16 %r465, %rs60; + cvt.f32.bf16 %r466, %rs61; + cvt.f32.bf16 %r467, %rs62; + cvt.f32.bf16 %r468, %rs63; + cvt.f32.bf16 %r469, %rs64; + cvt.f32.bf16 %r470, %rs65; + cvt.f32.bf16 %r471, %rs66; + cvt.f32.bf16 %r472, %rs67; + cvt.f32.bf16 %r473, %rs68; + cvt.f32.bf16 %r474, %rs69; + cvt.f32.bf16 %r475, %rs70; + cvt.f32.bf16 %r476, %rs71; + cvt.f32.bf16 %r477, %rs72; + cvt.f32.bf16 %r478, %rs73; + cvt.f32.bf16 %r479, %rs74; + cvt.f32.bf16 %r480, %rs75; + cvt.f32.bf16 %r481, %rs76; + cvt.f32.bf16 %r482, %rs77; + cvt.f32.bf16 %r483, %rs78; + cvt.f32.bf16 %r484, %rs79; + cvt.f32.bf16 %r485, %rs80; + cvt.f32.bf16 %r486, %rs81; + cvt.f32.bf16 %r487, %rs82; + cvt.f32.bf16 %r488, %rs83; + cvt.f32.bf16 %r489, %rs84; + cvt.f32.bf16 %r490, %rs85; + cvt.f32.bf16 %r491, %rs86; + cvt.f32.bf16 %r492, %rs87; + cvt.f32.bf16 %r493, %rs88; + cvt.f32.bf16 %r494, %rs89; + cvt.f32.bf16 %r495, %rs90; + cvt.f32.bf16 %r496, %rs91; + cvt.f32.bf16 %r497, %rs92; + cvt.f32.bf16 %r498, %rs93; + cvt.f32.bf16 %r499, %rs94; + cvt.f32.bf16 %r500, %rs95; + cvt.f32.bf16 %r501, %rs96; + cvt.f32.bf16 %r502, %rs97; + cvt.f32.bf16 %r503, %rs98; + cvt.f32.bf16 %r504, %rs99; + cvt.f32.bf16 %r505, %rs100; + cvt.f32.bf16 %r506, %rs101; + cvt.f32.bf16 %r507, %rs102; + cvt.f32.bf16 %r508, %rs103; + cvt.f32.bf16 %r509, %rs104; + cvt.f32.bf16 %r510, %rs105; + cvt.f32.bf16 %r511, %rs106; + cvt.f32.bf16 %r512, %rs107; + cvt.f32.bf16 %r513, %rs108; + cvt.f32.bf16 %r514, %rs109; + cvt.f32.bf16 %r515, %rs110; + cvt.f32.bf16 %r516, %rs111; + cvt.f32.bf16 %r517, %rs112; + cvt.f32.bf16 %r518, %rs113; + cvt.f32.bf16 %r519, %rs114; + cvt.f32.bf16 %r520, %rs115; + cvt.f32.bf16 %r521, %rs116; + cvt.f32.bf16 %r522, %rs117; + cvt.f32.bf16 %r523, %rs118; + cvt.f32.bf16 %r524, %rs119; + cvt.f32.bf16 %r525, %rs120; + cvt.f32.bf16 %r526, %rs121; + cvt.f32.bf16 %r527, %rs122; + cvt.f32.bf16 %r528, %rs123; + cvt.f32.bf16 %r529, %rs124; + cvt.f32.bf16 %r530, %rs125; + cvt.f32.bf16 %r531, %rs126; + cvt.f32.bf16 %r532, %rs127; + cvt.f32.bf16 %r533, %rs128; + cvt.f32.bf16 %r534, %rs129; + cvt.f32.bf16 %r535, %rs130; + cvt.f32.bf16 %r536, %rs131; + cvt.f32.bf16 %r537, %rs132; + cvt.f32.bf16 %r538, %rs133; + cvt.f32.bf16 %r539, %rs134; + cvt.f32.bf16 %r540, %rs135; + cvt.f32.bf16 %r541, %rs136; + cvt.f32.bf16 %r542, %rs137; + cvt.f32.bf16 %r543, %rs138; + cvt.f32.bf16 %r544, %rs139; + cvt.f32.bf16 %r545, %rs140; + cvt.f32.bf16 %r546, %rs141; + cvt.f32.bf16 %r547, %rs142; + cvt.f32.bf16 %r548, %rs143; + cvt.f32.bf16 %r549, %rs144; + cvt.f32.bf16 %r550, %rs145; + cvt.f32.bf16 %r551, %rs146; + cvt.f32.bf16 %r552, %rs147; + cvt.f32.bf16 %r553, %rs148; + cvt.f32.bf16 %r554, %rs149; + cvt.f32.bf16 %r555, %rs150; + cvt.f32.bf16 %r556, %rs151; + cvt.f32.bf16 %r557, %rs152; + cvt.f32.bf16 %r558, %rs153; + cvt.f32.bf16 %r559, %rs154; + cvt.f32.bf16 %r560, %rs155; + cvt.f32.bf16 %r561, %rs156; + cvt.f32.bf16 %r562, %rs157; + cvt.f32.bf16 %r563, %rs158; + cvt.f32.bf16 %r564, %rs159; + cvt.f32.bf16 %r565, %rs160; + cvt.f32.bf16 %r566, %rs161; + cvt.f32.bf16 %r567, %rs162; + cvt.f32.bf16 %r568, %rs163; + cvt.f32.bf16 %r569, %rs164; + cvt.f32.bf16 %r570, %rs165; + cvt.f32.bf16 %r571, %rs166; + cvt.f32.bf16 %r572, %rs167; + cvt.f32.bf16 %r573, %rs168; + cvt.f32.bf16 %r574, %rs169; + cvt.f32.bf16 %r575, %rs170; + cvt.f32.bf16 %r576, %rs171; + cvt.f32.bf16 %r577, %rs172; + cvt.f32.bf16 %r578, %rs173; + cvt.f32.bf16 %r579, %rs174; + cvt.f32.bf16 %r580, %rs175; + cvt.f32.bf16 %r581, %rs176; + cvt.f32.bf16 %r582, %rs177; + cvt.f32.bf16 %r583, %rs178; + cvt.f32.bf16 %r584, %rs179; + cvt.f32.bf16 %r585, %rs180; + cvt.f32.bf16 %r586, %rs181; + cvt.f32.bf16 %r587, %rs182; + cvt.f32.bf16 %r588, %rs183; + cvt.f32.bf16 %r589, %rs184; + cvt.f32.bf16 %r590, %rs185; + cvt.f32.bf16 %r591, %rs186; + cvt.f32.bf16 %r592, %rs187; + cvt.f32.bf16 %r593, %rs188; + cvt.f32.bf16 %r594, %rs189; + cvt.f32.bf16 %r595, %rs190; + cvt.f32.bf16 %r596, %rs191; + cvt.f32.bf16 %r597, %rs192; + cvt.f32.bf16 %r598, %rs193; + cvt.f32.bf16 %r599, %rs194; + cvt.f32.bf16 %r600, %rs195; + cvt.f32.bf16 %r601, %rs196; + cvt.f32.bf16 %r602, %rs197; + cvt.f32.bf16 %r603, %rs198; + cvt.f32.bf16 %r604, %rs199; + cvt.f32.bf16 %r605, %rs200; + cvt.f32.bf16 %r606, %rs201; + cvt.f32.bf16 %r607, %rs202; + cvt.f32.bf16 %r608, %rs203; + cvt.f32.bf16 %r609, %rs204; + cvt.f32.bf16 %r610, %rs205; + cvt.f32.bf16 %r611, %rs206; + cvt.f32.bf16 %r612, %rs207; + cvt.f32.bf16 %r613, %rs208; + cvt.f32.bf16 %r614, %rs209; + cvt.f32.bf16 %r615, %rs210; + cvt.f32.bf16 %r616, %rs211; + cvt.f32.bf16 %r617, %rs212; + cvt.f32.bf16 %r618, %rs213; + cvt.f32.bf16 %r619, %rs214; + cvt.f32.bf16 %r620, %rs215; + cvt.f32.bf16 %r621, %rs216; + cvt.f32.bf16 %r622, %rs217; + cvt.f32.bf16 %r623, %rs218; + cvt.f32.bf16 %r624, %rs219; + cvt.f32.bf16 %r625, %rs220; + cvt.f32.bf16 %r626, %rs221; + cvt.f32.bf16 %r627, %rs222; + cvt.f32.bf16 %r628, %rs223; + cvt.f32.bf16 %r629, %rs224; + cvt.f32.bf16 %r630, %rs225; + cvt.f32.bf16 %r631, %rs226; + cvt.f32.bf16 %r632, %rs227; + cvt.f32.bf16 %r633, %rs228; + cvt.f32.bf16 %r634, %rs229; + cvt.f32.bf16 %r635, %rs230; + cvt.f32.bf16 %r636, %rs231; + cvt.f32.bf16 %r637, %rs232; + cvt.f32.bf16 %r638, %rs233; + cvt.f32.bf16 %r639, %rs234; + cvt.f32.bf16 %r640, %rs235; + cvt.f32.bf16 %r641, %rs236; + cvt.f32.bf16 %r642, %rs237; + cvt.f32.bf16 %r643, %rs238; + cvt.f32.bf16 %r644, %rs239; + cvt.f32.bf16 %r645, %rs240; + cvt.f32.bf16 %r646, %rs241; + cvt.f32.bf16 %r647, %rs242; + cvt.f32.bf16 %r648, %rs243; + cvt.f32.bf16 %r649, %rs244; + cvt.f32.bf16 %r650, %rs245; + cvt.f32.bf16 %r651, %rs246; + cvt.f32.bf16 %r652, %rs247; + cvt.f32.bf16 %r653, %rs248; + cvt.f32.bf16 %r654, %rs249; + cvt.f32.bf16 %r655, %rs250; + cvt.f32.bf16 %r656, %rs251; + cvt.f32.bf16 %r657, %rs252; + cvt.f32.bf16 %r658, %rs253; + cvt.f32.bf16 %r659, %rs254; + cvt.f32.bf16 %r660, %rs255; + cvt.f32.bf16 %r661, %rs256; + cvt.f32.bf16 %r662, %rs257; + cvt.f32.bf16 %r663, %rs258; + cvt.f32.bf16 %r664, %rs259; + cvt.f32.bf16 %r665, %rs260; + cvt.f32.bf16 %r666, %rs261; + cvt.f32.bf16 %r667, %rs262; + cvt.f32.bf16 %r668, %rs263; + cvt.f32.bf16 %r669, %rs264; + cvt.f32.bf16 %r670, %rs265; + cvt.f32.bf16 %r671, %rs266; + cvt.f32.bf16 %r672, %rs267; + cvt.f32.bf16 %r673, %rs268; + cvt.f32.bf16 %r674, %rs269; + cvt.f32.bf16 %r675, %rs270; + cvt.f32.bf16 %r676, %rs271; + cvt.f32.bf16 %r677, %rs272; + cvt.f32.bf16 %r678, %rs273; + cvt.f32.bf16 %r679, %rs274; + cvt.f32.bf16 %r680, %rs275; + cvt.f32.bf16 %r681, %rs276; + cvt.f32.bf16 %r682, %rs277; + cvt.f32.bf16 %r683, %rs278; + cvt.f32.bf16 %r684, %rs279; + cvt.f32.bf16 %r685, %rs280; + cvt.f32.bf16 %r686, %rs281; + cvt.f32.bf16 %r687, %rs282; + cvt.f32.bf16 %r688, %rs283; + cvt.f32.bf16 %r689, %rs284; + cvt.f32.bf16 %r690, %rs285; + cvt.f32.bf16 %r691, %rs286; + cvt.f32.bf16 %r692, %rs287; + cvt.f32.bf16 %r693, %rs288; + cvt.f32.bf16 %r694, %rs289; + cvt.f32.bf16 %r695, %rs290; + cvt.f32.bf16 %r696, %rs291; + cvt.f32.bf16 %r697, %rs292; + cvt.f32.bf16 %r698, %rs293; + cvt.f32.bf16 %r699, %rs294; + cvt.f32.bf16 %r700, %rs295; + cvt.f32.bf16 %r701, %rs296; + cvt.f32.bf16 %r702, %rs297; + cvt.f32.bf16 %r703, %rs298; + cvt.f32.bf16 %r704, %rs299; + cvt.f32.bf16 %r705, %rs300; + cvt.f32.bf16 %r706, %rs301; + cvt.f32.bf16 %r707, %rs302; + cvt.f32.bf16 %r708, %rs303; + cvt.f32.bf16 %r709, %rs304; + cvt.f32.bf16 %r710, %rs305; + cvt.f32.bf16 %r711, %rs306; + cvt.f32.bf16 %r712, %rs307; + cvt.f32.bf16 %r713, %rs308; + cvt.f32.bf16 %r714, %rs309; + .loc 1 105 21 // l2norm.py:105:21 + shr.u32 %r715, %r428, 1; + and.b32 %r716, %r715, 1; + setp.ne.b32 %p4, %r716, 0; + // begin inline asm + mov.u32 %r259, %r5; + @%p4 ld.global.b32 { %r259 }, [ %rd33 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r261, %r5; + @%p5 ld.global.b32 { %r261 }, [ %rd34 + 0 ]; + // end inline asm + .loc 1 106 19 // l2norm.py:106:19 + // begin inline asm + mov.u32 %r130, %r5; + mov.u32 %r131, %r5; + mov.u32 %r132, %r5; + mov.u32 %r133, %r5; + @%p1 ld.global.v4.b32 { %r130, %r131, %r132, %r133 }, [ %rd35 + 0 ]; + // end inline asm + mov.b32 {%rs310, %rs311}, %r130; + mov.b32 {%rs312, %rs313}, %r131; + mov.b32 {%rs314, %rs315}, %r132; + mov.b32 {%rs316, %rs317}, %r133; + // begin inline asm + mov.u32 %r134, %r5; + mov.u32 %r135, %r5; + mov.u32 %r136, %r5; + mov.u32 %r137, %r5; + @%p2 ld.global.v4.b32 { %r134, %r135, %r136, %r137 }, [ %rd36 + 0 ]; + // end inline asm + mov.b32 {%rs318, %rs319}, %r134; + mov.b32 {%rs320, %rs321}, %r135; + mov.b32 {%rs322, %rs323}, %r136; + mov.b32 {%rs324, %rs325}, %r137; + // begin inline asm + mov.u32 %r138, %r5; + mov.u32 %r139, %r5; + mov.u32 %r140, %r5; + mov.u32 %r141, %r5; + @%p6 ld.global.v4.b32 { %r138, %r139, %r140, %r141 }, [ %rd37 + 0 ]; + // end inline asm + mov.b32 {%rs326, %rs327}, %r138; + mov.b32 {%rs328, %rs329}, %r139; + mov.b32 {%rs330, %rs331}, %r140; + mov.b32 {%rs332, %rs333}, %r141; + // begin inline asm + mov.u32 %r142, %r5; + mov.u32 %r143, %r5; + mov.u32 %r144, %r5; + mov.u32 %r145, %r5; + @%p7 ld.global.v4.b32 { %r142, %r143, %r144, %r145 }, [ %rd38 + 0 ]; + // end inline asm + mov.b32 {%rs334, %rs335}, %r142; + mov.b32 {%rs336, %rs337}, %r143; + mov.b32 {%rs338, %rs339}, %r144; + mov.b32 {%rs340, %rs341}, %r145; + // begin inline asm + mov.u32 %r146, %r5; + mov.u32 %r147, %r5; + mov.u32 %r148, %r5; + mov.u32 %r149, %r5; + @%p8 ld.global.v4.b32 { %r146, %r147, %r148, %r149 }, [ %rd39 + 0 ]; + // end inline asm + mov.b32 {%rs342, %rs343}, %r146; + mov.b32 {%rs344, %rs345}, %r147; + mov.b32 {%rs346, %rs347}, %r148; + mov.b32 {%rs348, %rs349}, %r149; + // begin inline asm + mov.u32 %r150, %r5; + mov.u32 %r151, %r5; + mov.u32 %r152, %r5; + mov.u32 %r153, %r5; + @%p9 ld.global.v4.b32 { %r150, %r151, %r152, %r153 }, [ %rd40 + 0 ]; + // end inline asm + mov.b32 {%rs350, %rs351}, %r150; + mov.b32 {%rs352, %rs353}, %r151; + mov.b32 {%rs354, %rs355}, %r152; + mov.b32 {%rs356, %rs357}, %r153; + // begin inline asm + mov.u32 %r154, %r5; + mov.u32 %r155, %r5; + mov.u32 %r156, %r5; + mov.u32 %r157, %r5; + @%p10 ld.global.v4.b32 { %r154, %r155, %r156, %r157 }, [ %rd41 + 0 ]; + // end inline asm + mov.b32 {%rs358, %rs359}, %r154; + mov.b32 {%rs360, %rs361}, %r155; + mov.b32 {%rs362, %rs363}, %r156; + mov.b32 {%rs364, %rs365}, %r157; + // begin inline asm + mov.u32 %r158, %r5; + mov.u32 %r159, %r5; + mov.u32 %r160, %r5; + mov.u32 %r161, %r5; + @%p11 ld.global.v4.b32 { %r158, %r159, %r160, %r161 }, [ %rd42 + 0 ]; + // end inline asm + mov.b32 {%rs366, %rs367}, %r158; + mov.b32 {%rs368, %rs369}, %r159; + mov.b32 {%rs370, %rs371}, %r160; + mov.b32 {%rs372, %rs373}, %r161; + // begin inline asm + mov.u32 %r162, %r5; + mov.u32 %r163, %r5; + mov.u32 %r164, %r5; + mov.u32 %r165, %r5; + @%p12 ld.global.v4.b32 { %r162, %r163, %r164, %r165 }, [ %rd43 + 0 ]; + // end inline asm + mov.b32 {%rs374, %rs375}, %r162; + mov.b32 {%rs376, %rs377}, %r163; + mov.b32 {%rs378, %rs379}, %r164; + mov.b32 {%rs380, %rs381}, %r165; + // begin inline asm + mov.u32 %r166, %r5; + mov.u32 %r167, %r5; + mov.u32 %r168, %r5; + mov.u32 %r169, %r5; + @%p3 ld.global.v4.b32 { %r166, %r167, %r168, %r169 }, [ %rd44 + 0 ]; + // end inline asm + mov.b32 {%rs382, %rs383}, %r166; + mov.b32 {%rs384, %rs385}, %r167; + mov.b32 {%rs386, %rs387}, %r168; + mov.b32 {%rs388, %rs389}, %r169; + // begin inline asm + mov.u32 %r170, %r5; + mov.u32 %r171, %r5; + mov.u32 %r172, %r5; + mov.u32 %r173, %r5; + @%p13 ld.global.v4.b32 { %r170, %r171, %r172, %r173 }, [ %rd45 + 0 ]; + // end inline asm + mov.b32 {%rs390, %rs391}, %r170; + mov.b32 {%rs392, %rs393}, %r171; + mov.b32 {%rs394, %rs395}, %r172; + mov.b32 {%rs396, %rs397}, %r173; + // begin inline asm + mov.u32 %r174, %r5; + mov.u32 %r175, %r5; + mov.u32 %r176, %r5; + mov.u32 %r177, %r5; + @%p14 ld.global.v4.b32 { %r174, %r175, %r176, %r177 }, [ %rd46 + 0 ]; + // end inline asm + mov.b32 {%rs398, %rs399}, %r174; + mov.b32 {%rs400, %rs401}, %r175; + mov.b32 {%rs402, %rs403}, %r176; + mov.b32 {%rs404, %rs405}, %r177; + // begin inline asm + mov.u32 %r178, %r5; + mov.u32 %r179, %r5; + mov.u32 %r180, %r5; + mov.u32 %r181, %r5; + @%p15 ld.global.v4.b32 { %r178, %r179, %r180, %r181 }, [ %rd47 + 0 ]; + // end inline asm + mov.b32 {%rs406, %rs407}, %r178; + mov.b32 {%rs408, %rs409}, %r179; + mov.b32 {%rs410, %rs411}, %r180; + mov.b32 {%rs412, %rs413}, %r181; + // begin inline asm + mov.u32 %r182, %r5; + mov.u32 %r183, %r5; + mov.u32 %r184, %r5; + mov.u32 %r185, %r5; + @%p16 ld.global.v4.b32 { %r182, %r183, %r184, %r185 }, [ %rd48 + 0 ]; + // end inline asm + mov.b32 {%rs414, %rs415}, %r182; + mov.b32 {%rs416, %rs417}, %r183; + mov.b32 {%rs418, %rs419}, %r184; + mov.b32 {%rs420, %rs421}, %r185; + // begin inline asm + mov.u32 %r186, %r5; + mov.u32 %r187, %r5; + mov.u32 %r188, %r5; + mov.u32 %r189, %r5; + @%p17 ld.global.v4.b32 { %r186, %r187, %r188, %r189 }, [ %rd49 + 0 ]; + // end inline asm + mov.b32 {%rs422, %rs423}, %r186; + mov.b32 {%rs424, %rs425}, %r187; + mov.b32 {%rs426, %rs427}, %r188; + mov.b32 {%rs428, %rs429}, %r189; + // begin inline asm + mov.u32 %r190, %r5; + mov.u32 %r191, %r5; + mov.u32 %r192, %r5; + mov.u32 %r193, %r5; + @%p18 ld.global.v4.b32 { %r190, %r191, %r192, %r193 }, [ %rd50 + 0 ]; + // end inline asm + mov.b32 {%rs430, %rs431}, %r190; + mov.b32 {%rs432, %rs433}, %r191; + mov.b32 {%rs434, %rs435}, %r192; + mov.b32 {%rs436, %rs437}, %r193; + // begin inline asm + mov.u32 %r194, %r5; + mov.u32 %r195, %r5; + mov.u32 %r196, %r5; + mov.u32 %r197, %r5; + @%p19 ld.global.v4.b32 { %r194, %r195, %r196, %r197 }, [ %rd51 + 0 ]; + // end inline asm + mov.b32 {%rs438, %rs439}, %r194; + mov.b32 {%rs440, %rs441}, %r195; + mov.b32 {%rs442, %rs443}, %r196; + mov.b32 {%rs444, %rs445}, %r197; + // begin inline asm + mov.u32 %r198, %r5; + mov.u32 %r199, %r5; + mov.u32 %r200, %r5; + mov.u32 %r201, %r5; + @%p20 ld.global.v4.b32 { %r198, %r199, %r200, %r201 }, [ %rd52 + 0 ]; + // end inline asm + mov.b32 {%rs446, %rs447}, %r198; + mov.b32 {%rs448, %rs449}, %r199; + mov.b32 {%rs450, %rs451}, %r200; + mov.b32 {%rs452, %rs453}, %r201; + // begin inline asm + mov.u32 %r202, %r5; + mov.u32 %r203, %r5; + mov.u32 %r204, %r5; + mov.u32 %r205, %r5; + @%p21 ld.global.v4.b32 { %r202, %r203, %r204, %r205 }, [ %rd53 + 0 ]; + // end inline asm + mov.b32 {%rs454, %rs455}, %r202; + mov.b32 {%rs456, %rs457}, %r203; + mov.b32 {%rs458, %rs459}, %r204; + mov.b32 {%rs460, %rs461}, %r205; + // begin inline asm + mov.u32 %r206, %r5; + mov.u32 %r207, %r5; + mov.u32 %r208, %r5; + mov.u32 %r209, %r5; + @%p22 ld.global.v4.b32 { %r206, %r207, %r208, %r209 }, [ %rd54 + 0 ]; + // end inline asm + mov.b32 {%rs462, %rs463}, %r206; + mov.b32 {%rs464, %rs465}, %r207; + mov.b32 {%rs466, %rs467}, %r208; + mov.b32 {%rs468, %rs469}, %r209; + // begin inline asm + mov.u32 %r210, %r5; + mov.u32 %r211, %r5; + mov.u32 %r212, %r5; + mov.u32 %r213, %r5; + @%p23 ld.global.v4.b32 { %r210, %r211, %r212, %r213 }, [ %rd55 + 0 ]; + // end inline asm + mov.b32 {%rs470, %rs471}, %r210; + mov.b32 {%rs472, %rs473}, %r211; + mov.b32 {%rs474, %rs475}, %r212; + mov.b32 {%rs476, %rs477}, %r213; + // begin inline asm + mov.u32 %r214, %r5; + mov.u32 %r215, %r5; + mov.u32 %r216, %r5; + mov.u32 %r217, %r5; + @%p24 ld.global.v4.b32 { %r214, %r215, %r216, %r217 }, [ %rd56 + 0 ]; + // end inline asm + mov.b32 {%rs478, %rs479}, %r214; + mov.b32 {%rs480, %rs481}, %r215; + mov.b32 {%rs482, %rs483}, %r216; + mov.b32 {%rs484, %rs485}, %r217; + // begin inline asm + mov.u32 %r218, %r5; + mov.u32 %r219, %r5; + mov.u32 %r220, %r5; + mov.u32 %r221, %r5; + @%p25 ld.global.v4.b32 { %r218, %r219, %r220, %r221 }, [ %rd57 + 0 ]; + // end inline asm + mov.b32 {%rs486, %rs487}, %r218; + mov.b32 {%rs488, %rs489}, %r219; + mov.b32 {%rs490, %rs491}, %r220; + mov.b32 {%rs492, %rs493}, %r221; + // begin inline asm + mov.u32 %r222, %r5; + mov.u32 %r223, %r5; + mov.u32 %r224, %r5; + mov.u32 %r225, %r5; + @%p26 ld.global.v4.b32 { %r222, %r223, %r224, %r225 }, [ %rd58 + 0 ]; + // end inline asm + mov.b32 {%rs494, %rs495}, %r222; + mov.b32 {%rs496, %rs497}, %r223; + mov.b32 {%rs498, %rs499}, %r224; + mov.b32 {%rs500, %rs501}, %r225; + // begin inline asm + mov.u32 %r226, %r5; + mov.u32 %r227, %r5; + mov.u32 %r228, %r5; + mov.u32 %r229, %r5; + @%p27 ld.global.v4.b32 { %r226, %r227, %r228, %r229 }, [ %rd59 + 0 ]; + // end inline asm + mov.b32 {%rs502, %rs503}, %r226; + mov.b32 {%rs504, %rs505}, %r227; + mov.b32 {%rs506, %rs507}, %r228; + mov.b32 {%rs508, %rs509}, %r229; + // begin inline asm + mov.u32 %r230, %r5; + mov.u32 %r231, %r5; + mov.u32 %r232, %r5; + mov.u32 %r233, %r5; + @%p28 ld.global.v4.b32 { %r230, %r231, %r232, %r233 }, [ %rd60 + 0 ]; + // end inline asm + mov.b32 {%rs510, %rs511}, %r230; + mov.b32 {%rs512, %rs513}, %r231; + mov.b32 {%rs514, %rs515}, %r232; + mov.b32 {%rs516, %rs517}, %r233; + // begin inline asm + mov.u32 %r234, %r5; + mov.u32 %r235, %r5; + mov.u32 %r236, %r5; + mov.u32 %r237, %r5; + @%p29 ld.global.v4.b32 { %r234, %r235, %r236, %r237 }, [ %rd61 + 0 ]; + // end inline asm + mov.b32 {%rs518, %rs519}, %r234; + mov.b32 {%rs520, %rs521}, %r235; + mov.b32 {%rs522, %rs523}, %r236; + mov.b32 {%rs524, %rs525}, %r237; + // begin inline asm + mov.u32 %r238, %r5; + mov.u32 %r239, %r5; + mov.u32 %r240, %r5; + mov.u32 %r241, %r5; + @%p30 ld.global.v4.b32 { %r238, %r239, %r240, %r241 }, [ %rd62 + 0 ]; + // end inline asm + mov.b32 {%rs526, %rs527}, %r238; + mov.b32 {%rs528, %rs529}, %r239; + mov.b32 {%rs530, %rs531}, %r240; + mov.b32 {%rs532, %rs533}, %r241; + // begin inline asm + mov.u32 %r242, %r5; + mov.u32 %r243, %r5; + mov.u32 %r244, %r5; + mov.u32 %r245, %r5; + @%p31 ld.global.v4.b32 { %r242, %r243, %r244, %r245 }, [ %rd63 + 0 ]; + // end inline asm + mov.b32 {%rs534, %rs535}, %r242; + mov.b32 {%rs536, %rs537}, %r243; + mov.b32 {%rs538, %rs539}, %r244; + mov.b32 {%rs540, %rs541}, %r245; + // begin inline asm + mov.u32 %r246, %r5; + mov.u32 %r247, %r5; + mov.u32 %r248, %r5; + mov.u32 %r249, %r5; + @%p32 ld.global.v4.b32 { %r246, %r247, %r248, %r249 }, [ %rd64 + 0 ]; + // end inline asm + mov.b32 {%rs542, %rs543}, %r246; + mov.b32 {%rs544, %rs545}, %r247; + mov.b32 {%rs546, %rs547}, %r248; + mov.b32 {%rs548, %rs549}, %r249; + // begin inline asm + mov.u32 %r250, %r5; + mov.u32 %r251, %r5; + mov.u32 %r252, %r5; + mov.u32 %r253, %r5; + @%p33 ld.global.v4.b32 { %r250, %r251, %r252, %r253 }, [ %rd65 + 0 ]; + // end inline asm + mov.b32 {%rs550, %rs551}, %r250; + mov.b32 {%rs552, %rs553}, %r251; + mov.b32 {%rs554, %rs555}, %r252; + mov.b32 {%rs556, %rs557}, %r253; + // begin inline asm + mov.u32 %r254, %r5; + mov.u32 %r255, %r5; + mov.u32 %r256, %r5; + mov.u32 %r257, %r5; + @%p34 ld.global.v4.b32 { %r254, %r255, %r256, %r257 }, [ %rd66 + 0 ]; + // end inline asm + mov.b32 {%rs558, %rs559}, %r254; + mov.b32 {%rs560, %rs561}, %r255; + mov.b32 {%rs562, %rs563}, %r256; + mov.b32 {%rs564, %rs565}, %r257; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r717, %rs310; + cvt.f32.bf16 %r718, %rs311; + cvt.f32.bf16 %r719, %rs312; + cvt.f32.bf16 %r720, %rs313; + cvt.f32.bf16 %r721, %rs314; + cvt.f32.bf16 %r722, %rs315; + cvt.f32.bf16 %r723, %rs316; + cvt.f32.bf16 %r724, %rs317; + cvt.f32.bf16 %r725, %rs318; + cvt.f32.bf16 %r726, %rs319; + cvt.f32.bf16 %r727, %rs320; + cvt.f32.bf16 %r728, %rs321; + cvt.f32.bf16 %r729, %rs322; + cvt.f32.bf16 %r730, %rs323; + cvt.f32.bf16 %r731, %rs324; + cvt.f32.bf16 %r732, %rs325; + cvt.f32.bf16 %r733, %rs326; + cvt.f32.bf16 %r734, %rs327; + cvt.f32.bf16 %r735, %rs328; + cvt.f32.bf16 %r736, %rs329; + cvt.f32.bf16 %r737, %rs330; + cvt.f32.bf16 %r738, %rs331; + cvt.f32.bf16 %r739, %rs332; + cvt.f32.bf16 %r740, %rs333; + cvt.f32.bf16 %r741, %rs334; + cvt.f32.bf16 %r742, %rs335; + cvt.f32.bf16 %r743, %rs336; + cvt.f32.bf16 %r744, %rs337; + cvt.f32.bf16 %r745, %rs338; + cvt.f32.bf16 %r746, %rs339; + cvt.f32.bf16 %r747, %rs340; + cvt.f32.bf16 %r748, %rs341; + cvt.f32.bf16 %r749, %rs342; + cvt.f32.bf16 %r750, %rs343; + cvt.f32.bf16 %r751, %rs344; + cvt.f32.bf16 %r752, %rs345; + cvt.f32.bf16 %r753, %rs346; + cvt.f32.bf16 %r754, %rs347; + cvt.f32.bf16 %r755, %rs348; + cvt.f32.bf16 %r756, %rs349; + cvt.f32.bf16 %r757, %rs350; + cvt.f32.bf16 %r758, %rs351; + cvt.f32.bf16 %r759, %rs352; + cvt.f32.bf16 %r760, %rs353; + cvt.f32.bf16 %r761, %rs354; + cvt.f32.bf16 %r762, %rs355; + cvt.f32.bf16 %r763, %rs356; + cvt.f32.bf16 %r764, %rs357; + cvt.f32.bf16 %r765, %rs358; + cvt.f32.bf16 %r766, %rs359; + cvt.f32.bf16 %r767, %rs360; + cvt.f32.bf16 %r768, %rs361; + cvt.f32.bf16 %r769, %rs362; + cvt.f32.bf16 %r770, %rs363; + cvt.f32.bf16 %r771, %rs364; + cvt.f32.bf16 %r772, %rs365; + cvt.f32.bf16 %r773, %rs366; + cvt.f32.bf16 %r774, %rs367; + cvt.f32.bf16 %r775, %rs368; + cvt.f32.bf16 %r776, %rs369; + cvt.f32.bf16 %r777, %rs370; + cvt.f32.bf16 %r778, %rs371; + cvt.f32.bf16 %r779, %rs372; + cvt.f32.bf16 %r780, %rs373; + cvt.f32.bf16 %r781, %rs374; + cvt.f32.bf16 %r782, %rs375; + cvt.f32.bf16 %r783, %rs376; + cvt.f32.bf16 %r784, %rs377; + cvt.f32.bf16 %r785, %rs378; + cvt.f32.bf16 %r786, %rs379; + cvt.f32.bf16 %r787, %rs380; + cvt.f32.bf16 %r788, %rs381; + cvt.f32.bf16 %r789, %rs382; + cvt.f32.bf16 %r790, %rs383; + cvt.f32.bf16 %r791, %rs384; + cvt.f32.bf16 %r792, %rs385; + cvt.f32.bf16 %r793, %rs386; + cvt.f32.bf16 %r794, %rs387; + cvt.f32.bf16 %r795, %rs388; + cvt.f32.bf16 %r796, %rs389; + cvt.f32.bf16 %r797, %rs390; + cvt.f32.bf16 %r798, %rs391; + cvt.f32.bf16 %r799, %rs392; + cvt.f32.bf16 %r800, %rs393; + cvt.f32.bf16 %r801, %rs394; + cvt.f32.bf16 %r802, %rs395; + cvt.f32.bf16 %r803, %rs396; + cvt.f32.bf16 %r804, %rs397; + cvt.f32.bf16 %r805, %rs398; + cvt.f32.bf16 %r806, %rs399; + cvt.f32.bf16 %r807, %rs400; + cvt.f32.bf16 %r808, %rs401; + cvt.f32.bf16 %r809, %rs402; + cvt.f32.bf16 %r810, %rs403; + cvt.f32.bf16 %r811, %rs404; + cvt.f32.bf16 %r812, %rs405; + cvt.f32.bf16 %r813, %rs406; + cvt.f32.bf16 %r814, %rs407; + cvt.f32.bf16 %r815, %rs408; + cvt.f32.bf16 %r816, %rs409; + cvt.f32.bf16 %r817, %rs410; + cvt.f32.bf16 %r818, %rs411; + cvt.f32.bf16 %r819, %rs412; + cvt.f32.bf16 %r820, %rs413; + cvt.f32.bf16 %r821, %rs414; + cvt.f32.bf16 %r822, %rs415; + cvt.f32.bf16 %r823, %rs416; + cvt.f32.bf16 %r824, %rs417; + cvt.f32.bf16 %r825, %rs418; + cvt.f32.bf16 %r826, %rs419; + cvt.f32.bf16 %r827, %rs420; + cvt.f32.bf16 %r828, %rs421; + cvt.f32.bf16 %r829, %rs422; + cvt.f32.bf16 %r830, %rs423; + cvt.f32.bf16 %r831, %rs424; + cvt.f32.bf16 %r832, %rs425; + cvt.f32.bf16 %r833, %rs426; + cvt.f32.bf16 %r834, %rs427; + cvt.f32.bf16 %r835, %rs428; + cvt.f32.bf16 %r836, %rs429; + cvt.f32.bf16 %r837, %rs430; + cvt.f32.bf16 %r838, %rs431; + cvt.f32.bf16 %r839, %rs432; + cvt.f32.bf16 %r840, %rs433; + cvt.f32.bf16 %r841, %rs434; + cvt.f32.bf16 %r842, %rs435; + cvt.f32.bf16 %r843, %rs436; + cvt.f32.bf16 %r844, %rs437; + cvt.f32.bf16 %r845, %rs438; + cvt.f32.bf16 %r846, %rs439; + cvt.f32.bf16 %r847, %rs440; + cvt.f32.bf16 %r848, %rs441; + cvt.f32.bf16 %r849, %rs442; + cvt.f32.bf16 %r850, %rs443; + cvt.f32.bf16 %r851, %rs444; + cvt.f32.bf16 %r852, %rs445; + cvt.f32.bf16 %r853, %rs446; + cvt.f32.bf16 %r854, %rs447; + cvt.f32.bf16 %r855, %rs448; + cvt.f32.bf16 %r856, %rs449; + cvt.f32.bf16 %r857, %rs450; + cvt.f32.bf16 %r858, %rs451; + cvt.f32.bf16 %r859, %rs452; + cvt.f32.bf16 %r860, %rs453; + cvt.f32.bf16 %r861, %rs454; + cvt.f32.bf16 %r862, %rs455; + cvt.f32.bf16 %r863, %rs456; + cvt.f32.bf16 %r864, %rs457; + cvt.f32.bf16 %r865, %rs458; + cvt.f32.bf16 %r866, %rs459; + cvt.f32.bf16 %r867, %rs460; + cvt.f32.bf16 %r868, %rs461; + cvt.f32.bf16 %r869, %rs462; + cvt.f32.bf16 %r870, %rs463; + cvt.f32.bf16 %r871, %rs464; + cvt.f32.bf16 %r872, %rs465; + cvt.f32.bf16 %r873, %rs466; + cvt.f32.bf16 %r874, %rs467; + cvt.f32.bf16 %r875, %rs468; + cvt.f32.bf16 %r876, %rs469; + cvt.f32.bf16 %r877, %rs470; + cvt.f32.bf16 %r878, %rs471; + cvt.f32.bf16 %r879, %rs472; + cvt.f32.bf16 %r880, %rs473; + cvt.f32.bf16 %r881, %rs474; + cvt.f32.bf16 %r882, %rs475; + cvt.f32.bf16 %r883, %rs476; + cvt.f32.bf16 %r884, %rs477; + cvt.f32.bf16 %r885, %rs478; + cvt.f32.bf16 %r886, %rs479; + cvt.f32.bf16 %r887, %rs480; + cvt.f32.bf16 %r888, %rs481; + cvt.f32.bf16 %r889, %rs482; + cvt.f32.bf16 %r890, %rs483; + cvt.f32.bf16 %r891, %rs484; + cvt.f32.bf16 %r892, %rs485; + cvt.f32.bf16 %r893, %rs486; + cvt.f32.bf16 %r894, %rs487; + cvt.f32.bf16 %r895, %rs488; + cvt.f32.bf16 %r896, %rs489; + cvt.f32.bf16 %r897, %rs490; + cvt.f32.bf16 %r898, %rs491; + cvt.f32.bf16 %r899, %rs492; + cvt.f32.bf16 %r900, %rs493; + cvt.f32.bf16 %r901, %rs494; + cvt.f32.bf16 %r902, %rs495; + cvt.f32.bf16 %r903, %rs496; + cvt.f32.bf16 %r904, %rs497; + cvt.f32.bf16 %r905, %rs498; + cvt.f32.bf16 %r906, %rs499; + cvt.f32.bf16 %r907, %rs500; + cvt.f32.bf16 %r908, %rs501; + cvt.f32.bf16 %r909, %rs502; + cvt.f32.bf16 %r910, %rs503; + cvt.f32.bf16 %r911, %rs504; + cvt.f32.bf16 %r912, %rs505; + cvt.f32.bf16 %r913, %rs506; + cvt.f32.bf16 %r914, %rs507; + cvt.f32.bf16 %r915, %rs508; + cvt.f32.bf16 %r916, %rs509; + cvt.f32.bf16 %r917, %rs510; + cvt.f32.bf16 %r918, %rs511; + cvt.f32.bf16 %r919, %rs512; + cvt.f32.bf16 %r920, %rs513; + cvt.f32.bf16 %r921, %rs514; + cvt.f32.bf16 %r922, %rs515; + cvt.f32.bf16 %r923, %rs516; + cvt.f32.bf16 %r924, %rs517; + cvt.f32.bf16 %r925, %rs518; + cvt.f32.bf16 %r926, %rs519; + cvt.f32.bf16 %r927, %rs520; + cvt.f32.bf16 %r928, %rs521; + cvt.f32.bf16 %r929, %rs522; + cvt.f32.bf16 %r930, %rs523; + cvt.f32.bf16 %r931, %rs524; + cvt.f32.bf16 %r932, %rs525; + cvt.f32.bf16 %r933, %rs526; + cvt.f32.bf16 %r934, %rs527; + cvt.f32.bf16 %r935, %rs528; + cvt.f32.bf16 %r936, %rs529; + cvt.f32.bf16 %r937, %rs530; + cvt.f32.bf16 %r938, %rs531; + cvt.f32.bf16 %r939, %rs532; + cvt.f32.bf16 %r940, %rs533; + cvt.f32.bf16 %r941, %rs534; + cvt.f32.bf16 %r942, %rs535; + cvt.f32.bf16 %r943, %rs536; + cvt.f32.bf16 %r944, %rs537; + cvt.f32.bf16 %r945, %rs538; + cvt.f32.bf16 %r946, %rs539; + cvt.f32.bf16 %r947, %rs540; + cvt.f32.bf16 %r948, %rs541; + cvt.f32.bf16 %r949, %rs542; + cvt.f32.bf16 %r950, %rs543; + cvt.f32.bf16 %r951, %rs544; + cvt.f32.bf16 %r952, %rs545; + cvt.f32.bf16 %r953, %rs546; + cvt.f32.bf16 %r954, %rs547; + cvt.f32.bf16 %r955, %rs548; + cvt.f32.bf16 %r956, %rs549; + cvt.f32.bf16 %r957, %rs550; + cvt.f32.bf16 %r958, %rs551; + cvt.f32.bf16 %r959, %rs552; + cvt.f32.bf16 %r960, %rs553; + cvt.f32.bf16 %r961, %rs554; + cvt.f32.bf16 %r962, %rs555; + cvt.f32.bf16 %r963, %rs556; + cvt.f32.bf16 %r964, %rs557; + cvt.f32.bf16 %r965, %rs558; + cvt.f32.bf16 %r966, %rs559; + cvt.f32.bf16 %r967, %rs560; + cvt.f32.bf16 %r968, %rs561; + cvt.f32.bf16 %r969, %rs562; + cvt.f32.bf16 %r970, %rs563; + cvt.f32.bf16 %r971, %rs564; + cvt.f32.bf16 %r972, %rs565; + .loc 1 107 25 // l2norm.py:107:25 + and.b32 %r973, %r395, 60; + shl.b32 %r974, %r391, 6; + and.b32 %r975, %r974, 64; + and.b32 %r976, %r391, 32; + shl.b32 %r977, %r976, 2; + or.b32 %r978, %r975, %r977; + or.b32 %r979, %r978, %r973; + mov.b32 %r980, global_smem; + add.s32 %r258, %r980, %r979; + // begin inline asm + st.shared.b32 [ %r258 + 0 ], %r259; + // end inline asm + xor.b32 %r981, %r979, 64; + add.s32 %r982, %r980, %r981; + add.s32 %r260, %r982, 256; + // begin inline asm + st.shared.b32 [ %r260 + 0 ], %r261; + // end inline asm + bar.sync 0; + bfe.s32 %r983, %r391, 4, 1; + and.b32 %r984, %r983, 320; + shl.b32 %r985, %r976, 1; + xor.b32 %r986, %r984, %r985; + add.s32 %r987, %r980, %r986; + ld.shared.v4.b32 {%r988, %r989, %r990, %r991}, [%r987]; + ld.shared.v4.b32 {%r992, %r993, %r994, %r995}, [%r987+16]; + ld.shared.v4.b32 {%r996, %r997, %r998, %r999}, [%r987+32]; + ld.shared.v4.b32 {%r1000, %r1001, %r1002, %r1003}, [%r987+48]; + ld.shared.v4.b32 {%r1004, %r1005, %r1006, %r1007}, [%r987+128]; + ld.shared.v4.b32 {%r1008, %r1009, %r1010, %r1011}, [%r987+144]; + ld.shared.v4.b32 {%r1012, %r1013, %r1014, %r1015}, [%r987+160]; + ld.shared.v4.b32 {%r1016, %r1017, %r1018, %r1019}, [%r987+176]; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r1020, %r990, %r733; + mul.f32 %r1021, %r990, %r734; + mul.f32 %r1022, %r990, %r735; + mul.f32 %r1023, %r990, %r736; + mul.f32 %r1024, %r990, %r737; + mul.f32 %r1025, %r990, %r738; + mul.f32 %r1026, %r990, %r739; + mul.f32 %r1027, %r990, %r740; + mul.f32 %r1028, %r991, %r741; + mul.f32 %r1029, %r991, %r742; + mul.f32 %r1030, %r991, %r743; + mul.f32 %r1031, %r991, %r744; + mul.f32 %r1032, %r991, %r745; + mul.f32 %r1033, %r991, %r746; + mul.f32 %r1034, %r991, %r747; + mul.f32 %r1035, %r991, %r748; + mul.f32 %r1036, %r992, %r749; + mul.f32 %r1037, %r992, %r750; + mul.f32 %r1038, %r992, %r751; + mul.f32 %r1039, %r992, %r752; + mul.f32 %r1040, %r992, %r753; + mul.f32 %r1041, %r992, %r754; + mul.f32 %r1042, %r992, %r755; + mul.f32 %r1043, %r992, %r756; + mul.f32 %r1044, %r993, %r757; + mul.f32 %r1045, %r993, %r758; + mul.f32 %r1046, %r993, %r759; + mul.f32 %r1047, %r993, %r760; + mul.f32 %r1048, %r993, %r761; + mul.f32 %r1049, %r993, %r762; + mul.f32 %r1050, %r993, %r763; + mul.f32 %r1051, %r993, %r764; + mul.f32 %r1052, %r994, %r765; + mul.f32 %r1053, %r994, %r766; + mul.f32 %r1054, %r994, %r767; + mul.f32 %r1055, %r994, %r768; + mul.f32 %r1056, %r994, %r769; + mul.f32 %r1057, %r994, %r770; + mul.f32 %r1058, %r994, %r771; + mul.f32 %r1059, %r994, %r772; + mul.f32 %r1060, %r995, %r773; + mul.f32 %r1061, %r995, %r774; + mul.f32 %r1062, %r995, %r775; + mul.f32 %r1063, %r995, %r776; + mul.f32 %r1064, %r995, %r777; + mul.f32 %r1065, %r995, %r778; + mul.f32 %r1066, %r995, %r779; + mul.f32 %r1067, %r995, %r780; + mul.f32 %r1068, %r996, %r781; + mul.f32 %r1069, %r996, %r782; + mul.f32 %r1070, %r996, %r783; + mul.f32 %r1071, %r996, %r784; + mul.f32 %r1072, %r996, %r785; + mul.f32 %r1073, %r996, %r786; + mul.f32 %r1074, %r996, %r787; + mul.f32 %r1075, %r996, %r788; + mul.f32 %r1076, %r997, %r789; + mul.f32 %r1077, %r997, %r790; + mul.f32 %r1078, %r997, %r791; + mul.f32 %r1079, %r997, %r792; + mul.f32 %r1080, %r997, %r793; + mul.f32 %r1081, %r997, %r794; + mul.f32 %r1082, %r997, %r795; + mul.f32 %r1083, %r997, %r796; + mul.f32 %r1084, %r998, %r797; + mul.f32 %r1085, %r998, %r798; + mul.f32 %r1086, %r998, %r799; + mul.f32 %r1087, %r998, %r800; + mul.f32 %r1088, %r998, %r801; + mul.f32 %r1089, %r998, %r802; + mul.f32 %r1090, %r998, %r803; + mul.f32 %r1091, %r998, %r804; + mul.f32 %r1092, %r999, %r805; + mul.f32 %r1093, %r999, %r806; + mul.f32 %r1094, %r999, %r807; + mul.f32 %r1095, %r999, %r808; + mul.f32 %r1096, %r999, %r809; + mul.f32 %r1097, %r999, %r810; + mul.f32 %r1098, %r999, %r811; + mul.f32 %r1099, %r999, %r812; + mul.f32 %r1100, %r1000, %r813; + mul.f32 %r1101, %r1000, %r814; + mul.f32 %r1102, %r1000, %r815; + mul.f32 %r1103, %r1000, %r816; + mul.f32 %r1104, %r1000, %r817; + mul.f32 %r1105, %r1000, %r818; + mul.f32 %r1106, %r1000, %r819; + mul.f32 %r1107, %r1000, %r820; + mul.f32 %r1108, %r1001, %r821; + mul.f32 %r1109, %r1001, %r822; + mul.f32 %r1110, %r1001, %r823; + mul.f32 %r1111, %r1001, %r824; + mul.f32 %r1112, %r1001, %r825; + mul.f32 %r1113, %r1001, %r826; + mul.f32 %r1114, %r1001, %r827; + mul.f32 %r1115, %r1001, %r828; + mul.f32 %r1116, %r1002, %r829; + mul.f32 %r1117, %r1002, %r830; + mul.f32 %r1118, %r1002, %r831; + mul.f32 %r1119, %r1002, %r832; + mul.f32 %r1120, %r1002, %r833; + mul.f32 %r1121, %r1002, %r834; + mul.f32 %r1122, %r1002, %r835; + mul.f32 %r1123, %r1002, %r836; + mul.f32 %r1124, %r1003, %r837; + mul.f32 %r1125, %r1003, %r838; + mul.f32 %r1126, %r1003, %r839; + mul.f32 %r1127, %r1003, %r840; + mul.f32 %r1128, %r1003, %r841; + mul.f32 %r1129, %r1003, %r842; + mul.f32 %r1130, %r1003, %r843; + mul.f32 %r1131, %r1003, %r844; + mul.f32 %r1132, %r1004, %r845; + mul.f32 %r1133, %r1004, %r846; + mul.f32 %r1134, %r1004, %r847; + mul.f32 %r1135, %r1004, %r848; + mul.f32 %r1136, %r1004, %r849; + mul.f32 %r1137, %r1004, %r850; + mul.f32 %r1138, %r1004, %r851; + mul.f32 %r1139, %r1004, %r852; + mul.f32 %r1140, %r1005, %r853; + mul.f32 %r1141, %r1005, %r854; + mul.f32 %r1142, %r1005, %r855; + mul.f32 %r1143, %r1005, %r856; + mul.f32 %r1144, %r1005, %r857; + mul.f32 %r1145, %r1005, %r858; + mul.f32 %r1146, %r1005, %r859; + mul.f32 %r1147, %r1005, %r860; + mul.f32 %r1148, %r1006, %r861; + mul.f32 %r1149, %r1006, %r862; + mul.f32 %r1150, %r1006, %r863; + mul.f32 %r1151, %r1006, %r864; + mul.f32 %r1152, %r1006, %r865; + mul.f32 %r1153, %r1006, %r866; + mul.f32 %r1154, %r1006, %r867; + mul.f32 %r1155, %r1006, %r868; + mul.f32 %r1156, %r1007, %r869; + mul.f32 %r1157, %r1007, %r870; + mul.f32 %r1158, %r1007, %r871; + mul.f32 %r1159, %r1007, %r872; + mul.f32 %r1160, %r1007, %r873; + mul.f32 %r1161, %r1007, %r874; + mul.f32 %r1162, %r1007, %r875; + mul.f32 %r1163, %r1007, %r876; + mul.f32 %r1164, %r1008, %r877; + mul.f32 %r1165, %r1008, %r878; + mul.f32 %r1166, %r1008, %r879; + mul.f32 %r1167, %r1008, %r880; + mul.f32 %r1168, %r1008, %r881; + mul.f32 %r1169, %r1008, %r882; + mul.f32 %r1170, %r1008, %r883; + mul.f32 %r1171, %r1008, %r884; + mul.f32 %r1172, %r1009, %r885; + mul.f32 %r1173, %r1009, %r886; + mul.f32 %r1174, %r1009, %r887; + mul.f32 %r1175, %r1009, %r888; + mul.f32 %r1176, %r1009, %r889; + mul.f32 %r1177, %r1009, %r890; + mul.f32 %r1178, %r1009, %r891; + mul.f32 %r1179, %r1009, %r892; + mul.f32 %r1180, %r1010, %r893; + mul.f32 %r1181, %r1010, %r894; + mul.f32 %r1182, %r1010, %r895; + mul.f32 %r1183, %r1010, %r896; + mul.f32 %r1184, %r1010, %r897; + mul.f32 %r1185, %r1010, %r898; + mul.f32 %r1186, %r1010, %r899; + mul.f32 %r1187, %r1010, %r900; + mul.f32 %r1188, %r1011, %r901; + mul.f32 %r1189, %r1011, %r902; + mul.f32 %r1190, %r1011, %r903; + mul.f32 %r1191, %r1011, %r904; + mul.f32 %r1192, %r1011, %r905; + mul.f32 %r1193, %r1011, %r906; + mul.f32 %r1194, %r1011, %r907; + mul.f32 %r1195, %r1011, %r908; + mul.f32 %r1196, %r1012, %r909; + mul.f32 %r1197, %r1012, %r910; + mul.f32 %r1198, %r1012, %r911; + mul.f32 %r1199, %r1012, %r912; + mul.f32 %r1200, %r1012, %r913; + mul.f32 %r1201, %r1012, %r914; + mul.f32 %r1202, %r1012, %r915; + mul.f32 %r1203, %r1012, %r916; + mul.f32 %r1204, %r1013, %r917; + mul.f32 %r1205, %r1013, %r918; + mul.f32 %r1206, %r1013, %r919; + mul.f32 %r1207, %r1013, %r920; + mul.f32 %r1208, %r1013, %r921; + mul.f32 %r1209, %r1013, %r922; + mul.f32 %r1210, %r1013, %r923; + mul.f32 %r1211, %r1013, %r924; + mul.f32 %r1212, %r1014, %r925; + mul.f32 %r1213, %r1014, %r926; + mul.f32 %r1214, %r1014, %r927; + mul.f32 %r1215, %r1014, %r928; + mul.f32 %r1216, %r1014, %r929; + mul.f32 %r1217, %r1014, %r930; + mul.f32 %r1218, %r1014, %r931; + mul.f32 %r1219, %r1014, %r932; + mul.f32 %r1220, %r1015, %r933; + mul.f32 %r1221, %r1015, %r934; + mul.f32 %r1222, %r1015, %r935; + mul.f32 %r1223, %r1015, %r936; + mul.f32 %r1224, %r1015, %r937; + mul.f32 %r1225, %r1015, %r938; + mul.f32 %r1226, %r1015, %r939; + mul.f32 %r1227, %r1015, %r940; + mul.f32 %r1228, %r1016, %r941; + mul.f32 %r1229, %r1016, %r942; + mul.f32 %r1230, %r1016, %r943; + mul.f32 %r1231, %r1016, %r944; + mul.f32 %r1232, %r1016, %r945; + mul.f32 %r1233, %r1016, %r946; + mul.f32 %r1234, %r1016, %r947; + mul.f32 %r1235, %r1016, %r948; + mul.f32 %r1236, %r1017, %r949; + mul.f32 %r1237, %r1017, %r950; + mul.f32 %r1238, %r1017, %r951; + mul.f32 %r1239, %r1017, %r952; + mul.f32 %r1240, %r1017, %r953; + mul.f32 %r1241, %r1017, %r954; + mul.f32 %r1242, %r1017, %r955; + mul.f32 %r1243, %r1017, %r956; + mul.f32 %r1244, %r1018, %r957; + mul.f32 %r1245, %r1018, %r958; + mul.f32 %r1246, %r1018, %r959; + mul.f32 %r1247, %r1018, %r960; + mul.f32 %r1248, %r1018, %r961; + mul.f32 %r1249, %r1018, %r962; + mul.f32 %r1250, %r1018, %r963; + mul.f32 %r1251, %r1018, %r964; + mul.f32 %r1252, %r1019, %r965; + mul.f32 %r1253, %r1019, %r966; + mul.f32 %r1254, %r1019, %r967; + mul.f32 %r1255, %r1019, %r968; + mul.f32 %r1256, %r1019, %r969; + mul.f32 %r1257, %r1019, %r970; + mul.f32 %r1258, %r1019, %r971; + mul.f32 %r1259, %r1019, %r972; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r1260, %r460, %r718; + mul.f32 %r1261, %r468, %r726; + mul.f32 %r1262, %r476, %r734; + mul.f32 %r1263, %r484, %r742; + mul.f32 %r1264, %r492, %r750; + mul.f32 %r1265, %r500, %r758; + mul.f32 %r1266, %r508, %r766; + mul.f32 %r1267, %r516, %r774; + mul.f32 %r1268, %r524, %r782; + mul.f32 %r1269, %r532, %r790; + mul.f32 %r1270, %r540, %r798; + mul.f32 %r1271, %r548, %r806; + mul.f32 %r1272, %r556, %r814; + mul.f32 %r1273, %r564, %r822; + mul.f32 %r1274, %r572, %r830; + mul.f32 %r1275, %r580, %r838; + mul.f32 %r1276, %r588, %r846; + mul.f32 %r1277, %r596, %r854; + mul.f32 %r1278, %r604, %r862; + mul.f32 %r1279, %r612, %r870; + mul.f32 %r1280, %r620, %r878; + mul.f32 %r1281, %r628, %r886; + mul.f32 %r1282, %r636, %r894; + mul.f32 %r1283, %r644, %r902; + mul.f32 %r1284, %r652, %r910; + mul.f32 %r1285, %r660, %r918; + mul.f32 %r1286, %r668, %r926; + mul.f32 %r1287, %r676, %r934; + mul.f32 %r1288, %r684, %r942; + mul.f32 %r1289, %r692, %r950; + mul.f32 %r1290, %r700, %r958; + mul.f32 %r1291, %r708, %r966; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1292, %rs54, %rs310, %r1260; + fma.rn.f32.bf16 %r1293, %rs56, %rs312, %r1292; + fma.rn.f32.bf16 %r1294, %rs57, %rs313, %r1293; + fma.rn.f32.bf16 %r1295, %rs58, %rs314, %r1294; + fma.rn.f32.bf16 %r1296, %rs59, %rs315, %r1295; + fma.rn.f32.bf16 %r1297, %rs60, %rs316, %r1296; + fma.rn.f32.bf16 %r1298, %rs61, %rs317, %r1297; + fma.rn.f32.bf16 %r1299, %rs62, %rs318, %r1261; + fma.rn.f32.bf16 %r1300, %rs64, %rs320, %r1299; + fma.rn.f32.bf16 %r1301, %rs65, %rs321, %r1300; + fma.rn.f32.bf16 %r1302, %rs66, %rs322, %r1301; + fma.rn.f32.bf16 %r1303, %rs67, %rs323, %r1302; + fma.rn.f32.bf16 %r1304, %rs68, %rs324, %r1303; + fma.rn.f32.bf16 %r1305, %rs69, %rs325, %r1304; + fma.rn.f32.bf16 %r1306, %rs70, %rs326, %r1262; + fma.rn.f32.bf16 %r1307, %rs72, %rs328, %r1306; + fma.rn.f32.bf16 %r1308, %rs73, %rs329, %r1307; + fma.rn.f32.bf16 %r1309, %rs74, %rs330, %r1308; + fma.rn.f32.bf16 %r1310, %rs75, %rs331, %r1309; + fma.rn.f32.bf16 %r1311, %rs76, %rs332, %r1310; + fma.rn.f32.bf16 %r1312, %rs77, %rs333, %r1311; + fma.rn.f32.bf16 %r1313, %rs78, %rs334, %r1263; + fma.rn.f32.bf16 %r1314, %rs80, %rs336, %r1313; + fma.rn.f32.bf16 %r1315, %rs81, %rs337, %r1314; + fma.rn.f32.bf16 %r1316, %rs82, %rs338, %r1315; + fma.rn.f32.bf16 %r1317, %rs83, %rs339, %r1316; + fma.rn.f32.bf16 %r1318, %rs84, %rs340, %r1317; + fma.rn.f32.bf16 %r1319, %rs85, %rs341, %r1318; + fma.rn.f32.bf16 %r1320, %rs86, %rs342, %r1264; + fma.rn.f32.bf16 %r1321, %rs88, %rs344, %r1320; + fma.rn.f32.bf16 %r1322, %rs89, %rs345, %r1321; + fma.rn.f32.bf16 %r1323, %rs90, %rs346, %r1322; + fma.rn.f32.bf16 %r1324, %rs91, %rs347, %r1323; + fma.rn.f32.bf16 %r1325, %rs92, %rs348, %r1324; + fma.rn.f32.bf16 %r1326, %rs93, %rs349, %r1325; + fma.rn.f32.bf16 %r1327, %rs94, %rs350, %r1265; + fma.rn.f32.bf16 %r1328, %rs96, %rs352, %r1327; + fma.rn.f32.bf16 %r1329, %rs97, %rs353, %r1328; + fma.rn.f32.bf16 %r1330, %rs98, %rs354, %r1329; + fma.rn.f32.bf16 %r1331, %rs99, %rs355, %r1330; + fma.rn.f32.bf16 %r1332, %rs100, %rs356, %r1331; + fma.rn.f32.bf16 %r1333, %rs101, %rs357, %r1332; + fma.rn.f32.bf16 %r1334, %rs102, %rs358, %r1266; + fma.rn.f32.bf16 %r1335, %rs104, %rs360, %r1334; + fma.rn.f32.bf16 %r1336, %rs105, %rs361, %r1335; + fma.rn.f32.bf16 %r1337, %rs106, %rs362, %r1336; + fma.rn.f32.bf16 %r1338, %rs107, %rs363, %r1337; + fma.rn.f32.bf16 %r1339, %rs108, %rs364, %r1338; + fma.rn.f32.bf16 %r1340, %rs109, %rs365, %r1339; + fma.rn.f32.bf16 %r1341, %rs110, %rs366, %r1267; + fma.rn.f32.bf16 %r1342, %rs112, %rs368, %r1341; + fma.rn.f32.bf16 %r1343, %rs113, %rs369, %r1342; + fma.rn.f32.bf16 %r1344, %rs114, %rs370, %r1343; + fma.rn.f32.bf16 %r1345, %rs115, %rs371, %r1344; + fma.rn.f32.bf16 %r1346, %rs116, %rs372, %r1345; + fma.rn.f32.bf16 %r1347, %rs117, %rs373, %r1346; + fma.rn.f32.bf16 %r1348, %rs118, %rs374, %r1268; + fma.rn.f32.bf16 %r1349, %rs120, %rs376, %r1348; + fma.rn.f32.bf16 %r1350, %rs121, %rs377, %r1349; + fma.rn.f32.bf16 %r1351, %rs122, %rs378, %r1350; + fma.rn.f32.bf16 %r1352, %rs123, %rs379, %r1351; + fma.rn.f32.bf16 %r1353, %rs124, %rs380, %r1352; + fma.rn.f32.bf16 %r1354, %rs125, %rs381, %r1353; + fma.rn.f32.bf16 %r1355, %rs126, %rs382, %r1269; + fma.rn.f32.bf16 %r1356, %rs128, %rs384, %r1355; + fma.rn.f32.bf16 %r1357, %rs129, %rs385, %r1356; + fma.rn.f32.bf16 %r1358, %rs130, %rs386, %r1357; + fma.rn.f32.bf16 %r1359, %rs131, %rs387, %r1358; + fma.rn.f32.bf16 %r1360, %rs132, %rs388, %r1359; + fma.rn.f32.bf16 %r1361, %rs133, %rs389, %r1360; + fma.rn.f32.bf16 %r1362, %rs134, %rs390, %r1270; + fma.rn.f32.bf16 %r1363, %rs136, %rs392, %r1362; + fma.rn.f32.bf16 %r1364, %rs137, %rs393, %r1363; + fma.rn.f32.bf16 %r1365, %rs138, %rs394, %r1364; + fma.rn.f32.bf16 %r1366, %rs139, %rs395, %r1365; + fma.rn.f32.bf16 %r1367, %rs140, %rs396, %r1366; + fma.rn.f32.bf16 %r1368, %rs141, %rs397, %r1367; + fma.rn.f32.bf16 %r1369, %rs142, %rs398, %r1271; + fma.rn.f32.bf16 %r1370, %rs144, %rs400, %r1369; + fma.rn.f32.bf16 %r1371, %rs145, %rs401, %r1370; + fma.rn.f32.bf16 %r1372, %rs146, %rs402, %r1371; + fma.rn.f32.bf16 %r1373, %rs147, %rs403, %r1372; + fma.rn.f32.bf16 %r1374, %rs148, %rs404, %r1373; + fma.rn.f32.bf16 %r1375, %rs149, %rs405, %r1374; + fma.rn.f32.bf16 %r1376, %rs150, %rs406, %r1272; + fma.rn.f32.bf16 %r1377, %rs152, %rs408, %r1376; + fma.rn.f32.bf16 %r1378, %rs153, %rs409, %r1377; + fma.rn.f32.bf16 %r1379, %rs154, %rs410, %r1378; + fma.rn.f32.bf16 %r1380, %rs155, %rs411, %r1379; + fma.rn.f32.bf16 %r1381, %rs156, %rs412, %r1380; + fma.rn.f32.bf16 %r1382, %rs157, %rs413, %r1381; + fma.rn.f32.bf16 %r1383, %rs158, %rs414, %r1273; + fma.rn.f32.bf16 %r1384, %rs160, %rs416, %r1383; + fma.rn.f32.bf16 %r1385, %rs161, %rs417, %r1384; + fma.rn.f32.bf16 %r1386, %rs162, %rs418, %r1385; + fma.rn.f32.bf16 %r1387, %rs163, %rs419, %r1386; + fma.rn.f32.bf16 %r1388, %rs164, %rs420, %r1387; + fma.rn.f32.bf16 %r1389, %rs165, %rs421, %r1388; + fma.rn.f32.bf16 %r1390, %rs166, %rs422, %r1274; + fma.rn.f32.bf16 %r1391, %rs168, %rs424, %r1390; + fma.rn.f32.bf16 %r1392, %rs169, %rs425, %r1391; + fma.rn.f32.bf16 %r1393, %rs170, %rs426, %r1392; + fma.rn.f32.bf16 %r1394, %rs171, %rs427, %r1393; + fma.rn.f32.bf16 %r1395, %rs172, %rs428, %r1394; + fma.rn.f32.bf16 %r1396, %rs173, %rs429, %r1395; + fma.rn.f32.bf16 %r1397, %rs174, %rs430, %r1275; + fma.rn.f32.bf16 %r1398, %rs176, %rs432, %r1397; + fma.rn.f32.bf16 %r1399, %rs177, %rs433, %r1398; + fma.rn.f32.bf16 %r1400, %rs178, %rs434, %r1399; + fma.rn.f32.bf16 %r1401, %rs179, %rs435, %r1400; + fma.rn.f32.bf16 %r1402, %rs180, %rs436, %r1401; + fma.rn.f32.bf16 %r1403, %rs181, %rs437, %r1402; + fma.rn.f32.bf16 %r1404, %rs182, %rs438, %r1276; + fma.rn.f32.bf16 %r1405, %rs184, %rs440, %r1404; + fma.rn.f32.bf16 %r1406, %rs185, %rs441, %r1405; + fma.rn.f32.bf16 %r1407, %rs186, %rs442, %r1406; + fma.rn.f32.bf16 %r1408, %rs187, %rs443, %r1407; + fma.rn.f32.bf16 %r1409, %rs188, %rs444, %r1408; + fma.rn.f32.bf16 %r1410, %rs189, %rs445, %r1409; + fma.rn.f32.bf16 %r1411, %rs190, %rs446, %r1277; + fma.rn.f32.bf16 %r1412, %rs192, %rs448, %r1411; + fma.rn.f32.bf16 %r1413, %rs193, %rs449, %r1412; + fma.rn.f32.bf16 %r1414, %rs194, %rs450, %r1413; + fma.rn.f32.bf16 %r1415, %rs195, %rs451, %r1414; + fma.rn.f32.bf16 %r1416, %rs196, %rs452, %r1415; + fma.rn.f32.bf16 %r1417, %rs197, %rs453, %r1416; + fma.rn.f32.bf16 %r1418, %rs198, %rs454, %r1278; + fma.rn.f32.bf16 %r1419, %rs200, %rs456, %r1418; + fma.rn.f32.bf16 %r1420, %rs201, %rs457, %r1419; + fma.rn.f32.bf16 %r1421, %rs202, %rs458, %r1420; + fma.rn.f32.bf16 %r1422, %rs203, %rs459, %r1421; + fma.rn.f32.bf16 %r1423, %rs204, %rs460, %r1422; + fma.rn.f32.bf16 %r1424, %rs205, %rs461, %r1423; + fma.rn.f32.bf16 %r1425, %rs206, %rs462, %r1279; + fma.rn.f32.bf16 %r1426, %rs208, %rs464, %r1425; + fma.rn.f32.bf16 %r1427, %rs209, %rs465, %r1426; + fma.rn.f32.bf16 %r1428, %rs210, %rs466, %r1427; + fma.rn.f32.bf16 %r1429, %rs211, %rs467, %r1428; + fma.rn.f32.bf16 %r1430, %rs212, %rs468, %r1429; + fma.rn.f32.bf16 %r1431, %rs213, %rs469, %r1430; + fma.rn.f32.bf16 %r1432, %rs214, %rs470, %r1280; + fma.rn.f32.bf16 %r1433, %rs216, %rs472, %r1432; + fma.rn.f32.bf16 %r1434, %rs217, %rs473, %r1433; + fma.rn.f32.bf16 %r1435, %rs218, %rs474, %r1434; + fma.rn.f32.bf16 %r1436, %rs219, %rs475, %r1435; + fma.rn.f32.bf16 %r1437, %rs220, %rs476, %r1436; + fma.rn.f32.bf16 %r1438, %rs221, %rs477, %r1437; + fma.rn.f32.bf16 %r1439, %rs222, %rs478, %r1281; + fma.rn.f32.bf16 %r1440, %rs224, %rs480, %r1439; + fma.rn.f32.bf16 %r1441, %rs225, %rs481, %r1440; + fma.rn.f32.bf16 %r1442, %rs226, %rs482, %r1441; + fma.rn.f32.bf16 %r1443, %rs227, %rs483, %r1442; + fma.rn.f32.bf16 %r1444, %rs228, %rs484, %r1443; + fma.rn.f32.bf16 %r1445, %rs229, %rs485, %r1444; + fma.rn.f32.bf16 %r1446, %rs230, %rs486, %r1282; + fma.rn.f32.bf16 %r1447, %rs232, %rs488, %r1446; + fma.rn.f32.bf16 %r1448, %rs233, %rs489, %r1447; + fma.rn.f32.bf16 %r1449, %rs234, %rs490, %r1448; + fma.rn.f32.bf16 %r1450, %rs235, %rs491, %r1449; + fma.rn.f32.bf16 %r1451, %rs236, %rs492, %r1450; + fma.rn.f32.bf16 %r1452, %rs237, %rs493, %r1451; + fma.rn.f32.bf16 %r1453, %rs238, %rs494, %r1283; + fma.rn.f32.bf16 %r1454, %rs240, %rs496, %r1453; + fma.rn.f32.bf16 %r1455, %rs241, %rs497, %r1454; + fma.rn.f32.bf16 %r1456, %rs242, %rs498, %r1455; + fma.rn.f32.bf16 %r1457, %rs243, %rs499, %r1456; + fma.rn.f32.bf16 %r1458, %rs244, %rs500, %r1457; + fma.rn.f32.bf16 %r1459, %rs245, %rs501, %r1458; + fma.rn.f32.bf16 %r1460, %rs246, %rs502, %r1284; + fma.rn.f32.bf16 %r1461, %rs248, %rs504, %r1460; + fma.rn.f32.bf16 %r1462, %rs249, %rs505, %r1461; + fma.rn.f32.bf16 %r1463, %rs250, %rs506, %r1462; + fma.rn.f32.bf16 %r1464, %rs251, %rs507, %r1463; + fma.rn.f32.bf16 %r1465, %rs252, %rs508, %r1464; + fma.rn.f32.bf16 %r1466, %rs253, %rs509, %r1465; + fma.rn.f32.bf16 %r1467, %rs254, %rs510, %r1285; + fma.rn.f32.bf16 %r1468, %rs256, %rs512, %r1467; + fma.rn.f32.bf16 %r1469, %rs257, %rs513, %r1468; + fma.rn.f32.bf16 %r1470, %rs258, %rs514, %r1469; + fma.rn.f32.bf16 %r1471, %rs259, %rs515, %r1470; + fma.rn.f32.bf16 %r1472, %rs260, %rs516, %r1471; + fma.rn.f32.bf16 %r1473, %rs261, %rs517, %r1472; + fma.rn.f32.bf16 %r1474, %rs262, %rs518, %r1286; + fma.rn.f32.bf16 %r1475, %rs264, %rs520, %r1474; + fma.rn.f32.bf16 %r1476, %rs265, %rs521, %r1475; + fma.rn.f32.bf16 %r1477, %rs266, %rs522, %r1476; + fma.rn.f32.bf16 %r1478, %rs267, %rs523, %r1477; + fma.rn.f32.bf16 %r1479, %rs268, %rs524, %r1478; + fma.rn.f32.bf16 %r1480, %rs269, %rs525, %r1479; + fma.rn.f32.bf16 %r1481, %rs270, %rs526, %r1287; + fma.rn.f32.bf16 %r1482, %rs272, %rs528, %r1481; + fma.rn.f32.bf16 %r1483, %rs273, %rs529, %r1482; + fma.rn.f32.bf16 %r1484, %rs274, %rs530, %r1483; + fma.rn.f32.bf16 %r1485, %rs275, %rs531, %r1484; + fma.rn.f32.bf16 %r1486, %rs276, %rs532, %r1485; + fma.rn.f32.bf16 %r1487, %rs277, %rs533, %r1486; + fma.rn.f32.bf16 %r1488, %rs278, %rs534, %r1288; + fma.rn.f32.bf16 %r1489, %rs280, %rs536, %r1488; + fma.rn.f32.bf16 %r1490, %rs281, %rs537, %r1489; + fma.rn.f32.bf16 %r1491, %rs282, %rs538, %r1490; + fma.rn.f32.bf16 %r1492, %rs283, %rs539, %r1491; + fma.rn.f32.bf16 %r1493, %rs284, %rs540, %r1492; + fma.rn.f32.bf16 %r1494, %rs285, %rs541, %r1493; + fma.rn.f32.bf16 %r1495, %rs286, %rs542, %r1289; + fma.rn.f32.bf16 %r1496, %rs288, %rs544, %r1495; + fma.rn.f32.bf16 %r1497, %rs289, %rs545, %r1496; + fma.rn.f32.bf16 %r1498, %rs290, %rs546, %r1497; + fma.rn.f32.bf16 %r1499, %rs291, %rs547, %r1498; + fma.rn.f32.bf16 %r1500, %rs292, %rs548, %r1499; + fma.rn.f32.bf16 %r1501, %rs293, %rs549, %r1500; + fma.rn.f32.bf16 %r1502, %rs294, %rs550, %r1290; + fma.rn.f32.bf16 %r1503, %rs296, %rs552, %r1502; + fma.rn.f32.bf16 %r1504, %rs297, %rs553, %r1503; + fma.rn.f32.bf16 %r1505, %rs298, %rs554, %r1504; + fma.rn.f32.bf16 %r1506, %rs299, %rs555, %r1505; + fma.rn.f32.bf16 %r1507, %rs300, %rs556, %r1506; + fma.rn.f32.bf16 %r1508, %rs301, %rs557, %r1507; + fma.rn.f32.bf16 %r1509, %rs302, %rs558, %r1291; + fma.rn.f32.bf16 %r1510, %rs304, %rs560, %r1509; + fma.rn.f32.bf16 %r1511, %rs305, %rs561, %r1510; + fma.rn.f32.bf16 %r1512, %rs306, %rs562, %r1511; + fma.rn.f32.bf16 %r1513, %rs307, %rs563, %r1512; + fma.rn.f32.bf16 %r1514, %rs308, %rs564, %r1513; + fma.rn.f32.bf16 %r1515, %rs309, %rs565, %r1514; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1516, %r1298, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1517, %r1298, %r1516; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1518, %r1517, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1519, %r1517, %r1518; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1520, %r1519, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1521, %r1519, %r1520; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1522, %r1521, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1523, %r1521, %r1522; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1524, %r1305, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1525, %r1305, %r1524; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1526, %r1525, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1527, %r1525, %r1526; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1528, %r1527, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1529, %r1527, %r1528; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1530, %r1529, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1531, %r1529, %r1530; +$L__tmp18: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1532, %r1312, 8, 31, -1; +$L__tmp19: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1533, %r1312, %r1532; +$L__tmp20: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1534, %r1533, 4, 31, -1; +$L__tmp21: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1535, %r1533, %r1534; +$L__tmp22: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1536, %r1535, 2, 31, -1; +$L__tmp23: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1537, %r1535, %r1536; +$L__tmp24: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1538, %r1537, 1, 31, -1; +$L__tmp25: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1539, %r1537, %r1538; +$L__tmp26: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1540, %r1319, 8, 31, -1; +$L__tmp27: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1541, %r1319, %r1540; +$L__tmp28: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1542, %r1541, 4, 31, -1; +$L__tmp29: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1543, %r1541, %r1542; +$L__tmp30: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1544, %r1543, 2, 31, -1; +$L__tmp31: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1545, %r1543, %r1544; +$L__tmp32: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1546, %r1545, 1, 31, -1; +$L__tmp33: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1547, %r1545, %r1546; +$L__tmp34: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1548, %r1326, 8, 31, -1; +$L__tmp35: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1549, %r1326, %r1548; +$L__tmp36: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1550, %r1549, 4, 31, -1; +$L__tmp37: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1551, %r1549, %r1550; +$L__tmp38: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1552, %r1551, 2, 31, -1; +$L__tmp39: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1553, %r1551, %r1552; +$L__tmp40: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1554, %r1553, 1, 31, -1; +$L__tmp41: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1555, %r1553, %r1554; +$L__tmp42: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1556, %r1333, 8, 31, -1; +$L__tmp43: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1557, %r1333, %r1556; +$L__tmp44: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1558, %r1557, 4, 31, -1; +$L__tmp45: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1559, %r1557, %r1558; +$L__tmp46: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1560, %r1559, 2, 31, -1; +$L__tmp47: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1561, %r1559, %r1560; +$L__tmp48: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1562, %r1561, 1, 31, -1; +$L__tmp49: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1563, %r1561, %r1562; +$L__tmp50: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1564, %r1340, 8, 31, -1; +$L__tmp51: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1565, %r1340, %r1564; +$L__tmp52: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1566, %r1565, 4, 31, -1; +$L__tmp53: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1567, %r1565, %r1566; +$L__tmp54: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1568, %r1567, 2, 31, -1; +$L__tmp55: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1569, %r1567, %r1568; +$L__tmp56: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1570, %r1569, 1, 31, -1; +$L__tmp57: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1571, %r1569, %r1570; +$L__tmp58: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1572, %r1347, 8, 31, -1; +$L__tmp59: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1573, %r1347, %r1572; +$L__tmp60: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1574, %r1573, 4, 31, -1; +$L__tmp61: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1575, %r1573, %r1574; +$L__tmp62: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1576, %r1575, 2, 31, -1; +$L__tmp63: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1577, %r1575, %r1576; +$L__tmp64: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1578, %r1577, 1, 31, -1; +$L__tmp65: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1579, %r1577, %r1578; +$L__tmp66: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1580, %r1354, 8, 31, -1; +$L__tmp67: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1581, %r1354, %r1580; +$L__tmp68: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1582, %r1581, 4, 31, -1; +$L__tmp69: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1583, %r1581, %r1582; +$L__tmp70: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1584, %r1583, 2, 31, -1; +$L__tmp71: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1585, %r1583, %r1584; +$L__tmp72: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1586, %r1585, 1, 31, -1; +$L__tmp73: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1587, %r1585, %r1586; +$L__tmp74: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1588, %r1361, 8, 31, -1; +$L__tmp75: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1589, %r1361, %r1588; +$L__tmp76: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1590, %r1589, 4, 31, -1; +$L__tmp77: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1591, %r1589, %r1590; +$L__tmp78: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1592, %r1591, 2, 31, -1; +$L__tmp79: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1593, %r1591, %r1592; +$L__tmp80: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1594, %r1593, 1, 31, -1; +$L__tmp81: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1595, %r1593, %r1594; +$L__tmp82: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1596, %r1368, 8, 31, -1; +$L__tmp83: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1597, %r1368, %r1596; +$L__tmp84: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1598, %r1597, 4, 31, -1; +$L__tmp85: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1599, %r1597, %r1598; +$L__tmp86: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1600, %r1599, 2, 31, -1; +$L__tmp87: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1601, %r1599, %r1600; +$L__tmp88: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1602, %r1601, 1, 31, -1; +$L__tmp89: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1603, %r1601, %r1602; +$L__tmp90: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1604, %r1375, 8, 31, -1; +$L__tmp91: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1605, %r1375, %r1604; +$L__tmp92: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1606, %r1605, 4, 31, -1; +$L__tmp93: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1607, %r1605, %r1606; +$L__tmp94: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1608, %r1607, 2, 31, -1; +$L__tmp95: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1609, %r1607, %r1608; +$L__tmp96: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1610, %r1609, 1, 31, -1; +$L__tmp97: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1611, %r1609, %r1610; +$L__tmp98: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1612, %r1382, 8, 31, -1; +$L__tmp99: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1613, %r1382, %r1612; +$L__tmp100: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1614, %r1613, 4, 31, -1; +$L__tmp101: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1615, %r1613, %r1614; +$L__tmp102: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1616, %r1615, 2, 31, -1; +$L__tmp103: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1617, %r1615, %r1616; +$L__tmp104: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1618, %r1617, 1, 31, -1; +$L__tmp105: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1619, %r1617, %r1618; +$L__tmp106: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1620, %r1389, 8, 31, -1; +$L__tmp107: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1621, %r1389, %r1620; +$L__tmp108: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1622, %r1621, 4, 31, -1; +$L__tmp109: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1623, %r1621, %r1622; +$L__tmp110: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1624, %r1623, 2, 31, -1; +$L__tmp111: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1625, %r1623, %r1624; +$L__tmp112: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1626, %r1625, 1, 31, -1; +$L__tmp113: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1627, %r1625, %r1626; +$L__tmp114: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1628, %r1396, 8, 31, -1; +$L__tmp115: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1629, %r1396, %r1628; +$L__tmp116: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1630, %r1629, 4, 31, -1; +$L__tmp117: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1631, %r1629, %r1630; +$L__tmp118: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1632, %r1631, 2, 31, -1; +$L__tmp119: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1633, %r1631, %r1632; +$L__tmp120: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1634, %r1633, 1, 31, -1; +$L__tmp121: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1635, %r1633, %r1634; +$L__tmp122: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1636, %r1403, 8, 31, -1; +$L__tmp123: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1637, %r1403, %r1636; +$L__tmp124: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1638, %r1637, 4, 31, -1; +$L__tmp125: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1639, %r1637, %r1638; +$L__tmp126: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1640, %r1639, 2, 31, -1; +$L__tmp127: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1641, %r1639, %r1640; +$L__tmp128: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1642, %r1641, 1, 31, -1; +$L__tmp129: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1643, %r1641, %r1642; +$L__tmp130: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1644, %r1410, 8, 31, -1; +$L__tmp131: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1645, %r1410, %r1644; +$L__tmp132: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1646, %r1645, 4, 31, -1; +$L__tmp133: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1647, %r1645, %r1646; +$L__tmp134: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1648, %r1647, 2, 31, -1; +$L__tmp135: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1649, %r1647, %r1648; +$L__tmp136: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1650, %r1649, 1, 31, -1; +$L__tmp137: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1651, %r1649, %r1650; +$L__tmp138: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1652, %r1417, 8, 31, -1; +$L__tmp139: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1653, %r1417, %r1652; +$L__tmp140: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1654, %r1653, 4, 31, -1; +$L__tmp141: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1655, %r1653, %r1654; +$L__tmp142: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1656, %r1655, 2, 31, -1; +$L__tmp143: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1657, %r1655, %r1656; +$L__tmp144: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1658, %r1657, 1, 31, -1; +$L__tmp145: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1659, %r1657, %r1658; +$L__tmp146: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1660, %r1424, 8, 31, -1; +$L__tmp147: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1661, %r1424, %r1660; +$L__tmp148: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1662, %r1661, 4, 31, -1; +$L__tmp149: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1663, %r1661, %r1662; +$L__tmp150: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1664, %r1663, 2, 31, -1; +$L__tmp151: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1665, %r1663, %r1664; +$L__tmp152: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1666, %r1665, 1, 31, -1; +$L__tmp153: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1667, %r1665, %r1666; +$L__tmp154: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1668, %r1431, 8, 31, -1; +$L__tmp155: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1669, %r1431, %r1668; +$L__tmp156: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1670, %r1669, 4, 31, -1; +$L__tmp157: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1671, %r1669, %r1670; +$L__tmp158: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1672, %r1671, 2, 31, -1; +$L__tmp159: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1673, %r1671, %r1672; +$L__tmp160: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1674, %r1673, 1, 31, -1; +$L__tmp161: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1675, %r1673, %r1674; +$L__tmp162: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1676, %r1438, 8, 31, -1; +$L__tmp163: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1677, %r1438, %r1676; +$L__tmp164: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1678, %r1677, 4, 31, -1; +$L__tmp165: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1679, %r1677, %r1678; +$L__tmp166: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1680, %r1679, 2, 31, -1; +$L__tmp167: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1681, %r1679, %r1680; +$L__tmp168: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1682, %r1681, 1, 31, -1; +$L__tmp169: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1683, %r1681, %r1682; +$L__tmp170: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1684, %r1445, 8, 31, -1; +$L__tmp171: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1685, %r1445, %r1684; +$L__tmp172: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1686, %r1685, 4, 31, -1; +$L__tmp173: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1687, %r1685, %r1686; +$L__tmp174: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1688, %r1687, 2, 31, -1; +$L__tmp175: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1689, %r1687, %r1688; +$L__tmp176: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1690, %r1689, 1, 31, -1; +$L__tmp177: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1691, %r1689, %r1690; +$L__tmp178: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1692, %r1452, 8, 31, -1; +$L__tmp179: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1693, %r1452, %r1692; +$L__tmp180: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1694, %r1693, 4, 31, -1; +$L__tmp181: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1695, %r1693, %r1694; +$L__tmp182: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1696, %r1695, 2, 31, -1; +$L__tmp183: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1697, %r1695, %r1696; +$L__tmp184: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1698, %r1697, 1, 31, -1; +$L__tmp185: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1699, %r1697, %r1698; +$L__tmp186: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1700, %r1459, 8, 31, -1; +$L__tmp187: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1701, %r1459, %r1700; +$L__tmp188: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1702, %r1701, 4, 31, -1; +$L__tmp189: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1703, %r1701, %r1702; +$L__tmp190: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1704, %r1703, 2, 31, -1; +$L__tmp191: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1705, %r1703, %r1704; +$L__tmp192: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1706, %r1705, 1, 31, -1; +$L__tmp193: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1707, %r1705, %r1706; +$L__tmp194: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1708, %r1466, 8, 31, -1; +$L__tmp195: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1709, %r1466, %r1708; +$L__tmp196: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1710, %r1709, 4, 31, -1; +$L__tmp197: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1711, %r1709, %r1710; +$L__tmp198: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1712, %r1711, 2, 31, -1; +$L__tmp199: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1713, %r1711, %r1712; +$L__tmp200: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1714, %r1713, 1, 31, -1; +$L__tmp201: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1715, %r1713, %r1714; +$L__tmp202: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1716, %r1473, 8, 31, -1; +$L__tmp203: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1717, %r1473, %r1716; +$L__tmp204: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1718, %r1717, 4, 31, -1; +$L__tmp205: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1719, %r1717, %r1718; +$L__tmp206: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1720, %r1719, 2, 31, -1; +$L__tmp207: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1721, %r1719, %r1720; +$L__tmp208: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1722, %r1721, 1, 31, -1; +$L__tmp209: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1723, %r1721, %r1722; +$L__tmp210: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1724, %r1480, 8, 31, -1; +$L__tmp211: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1725, %r1480, %r1724; +$L__tmp212: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1726, %r1725, 4, 31, -1; +$L__tmp213: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1727, %r1725, %r1726; +$L__tmp214: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1728, %r1727, 2, 31, -1; +$L__tmp215: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1729, %r1727, %r1728; +$L__tmp216: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1730, %r1729, 1, 31, -1; +$L__tmp217: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1731, %r1729, %r1730; +$L__tmp218: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1732, %r1487, 8, 31, -1; +$L__tmp219: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1733, %r1487, %r1732; +$L__tmp220: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1734, %r1733, 4, 31, -1; +$L__tmp221: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1735, %r1733, %r1734; +$L__tmp222: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1736, %r1735, 2, 31, -1; +$L__tmp223: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1737, %r1735, %r1736; +$L__tmp224: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1738, %r1737, 1, 31, -1; +$L__tmp225: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1739, %r1737, %r1738; +$L__tmp226: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1740, %r1494, 8, 31, -1; +$L__tmp227: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1741, %r1494, %r1740; +$L__tmp228: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1742, %r1741, 4, 31, -1; +$L__tmp229: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1743, %r1741, %r1742; +$L__tmp230: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1744, %r1743, 2, 31, -1; +$L__tmp231: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1745, %r1743, %r1744; +$L__tmp232: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1746, %r1745, 1, 31, -1; +$L__tmp233: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1747, %r1745, %r1746; +$L__tmp234: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1748, %r1501, 8, 31, -1; +$L__tmp235: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1749, %r1501, %r1748; +$L__tmp236: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1750, %r1749, 4, 31, -1; +$L__tmp237: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1751, %r1749, %r1750; +$L__tmp238: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1752, %r1751, 2, 31, -1; +$L__tmp239: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1753, %r1751, %r1752; +$L__tmp240: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1754, %r1753, 1, 31, -1; +$L__tmp241: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1755, %r1753, %r1754; +$L__tmp242: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1756, %r1508, 8, 31, -1; +$L__tmp243: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1757, %r1508, %r1756; +$L__tmp244: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1758, %r1757, 4, 31, -1; +$L__tmp245: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1759, %r1757, %r1758; +$L__tmp246: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1760, %r1759, 2, 31, -1; +$L__tmp247: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1761, %r1759, %r1760; +$L__tmp248: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1762, %r1761, 1, 31, -1; +$L__tmp249: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1763, %r1761, %r1762; +$L__tmp250: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1764, %r1515, 8, 31, -1; +$L__tmp251: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1765, %r1515, %r1764; +$L__tmp252: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1766, %r1765, 4, 31, -1; +$L__tmp253: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1767, %r1765, %r1766; +$L__tmp254: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1768, %r1767, 2, 31, -1; +$L__tmp255: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1769, %r1767, %r1768; +$L__tmp256: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1770, %r1769, 1, 31, -1; +$L__tmp257: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1771, %r1769, %r1770; +$L__tmp258: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r1772, %r1523, %r459; + mul.f32 %r1773, %r1523, %r460; + mul.f32 %r1774, %r1523, %r461; + mul.f32 %r1775, %r1523, %r462; + mul.f32 %r1776, %r1523, %r463; + mul.f32 %r1777, %r1523, %r464; + mul.f32 %r1778, %r1523, %r465; + mul.f32 %r1779, %r1523, %r466; + mul.f32 %r1780, %r1531, %r467; + mul.f32 %r1781, %r1531, %r468; + mul.f32 %r1782, %r1531, %r469; + mul.f32 %r1783, %r1531, %r470; + mul.f32 %r1784, %r1531, %r471; + mul.f32 %r1785, %r1531, %r472; + mul.f32 %r1786, %r1531, %r473; + mul.f32 %r1787, %r1531, %r474; + mul.f32 %r1788, %r1539, %r475; + mul.f32 %r1789, %r1539, %r476; + mul.f32 %r1790, %r1539, %r477; + mul.f32 %r1791, %r1539, %r478; + mul.f32 %r1792, %r1539, %r479; + mul.f32 %r1793, %r1539, %r480; + mul.f32 %r1794, %r1539, %r481; + mul.f32 %r1795, %r1539, %r482; + mul.f32 %r1796, %r1547, %r483; + mul.f32 %r1797, %r1547, %r484; + mul.f32 %r1798, %r1547, %r485; + mul.f32 %r1799, %r1547, %r486; + mul.f32 %r1800, %r1547, %r487; + mul.f32 %r1801, %r1547, %r488; + mul.f32 %r1802, %r1547, %r489; + mul.f32 %r1803, %r1547, %r490; + mul.f32 %r1804, %r1555, %r491; + mul.f32 %r1805, %r1555, %r492; + mul.f32 %r1806, %r1555, %r493; + mul.f32 %r1807, %r1555, %r494; + mul.f32 %r1808, %r1555, %r495; + mul.f32 %r1809, %r1555, %r496; + mul.f32 %r1810, %r1555, %r497; + mul.f32 %r1811, %r1555, %r498; + mul.f32 %r1812, %r1563, %r499; + mul.f32 %r1813, %r1563, %r500; + mul.f32 %r1814, %r1563, %r501; + mul.f32 %r1815, %r1563, %r502; + mul.f32 %r1816, %r1563, %r503; + mul.f32 %r1817, %r1563, %r504; + mul.f32 %r1818, %r1563, %r505; + mul.f32 %r1819, %r1563, %r506; + mul.f32 %r1820, %r1571, %r507; + mul.f32 %r1821, %r1571, %r508; + mul.f32 %r1822, %r1571, %r509; + mul.f32 %r1823, %r1571, %r510; + mul.f32 %r1824, %r1571, %r511; + mul.f32 %r1825, %r1571, %r512; + mul.f32 %r1826, %r1571, %r513; + mul.f32 %r1827, %r1571, %r514; + mul.f32 %r1828, %r1579, %r515; + mul.f32 %r1829, %r1579, %r516; + mul.f32 %r1830, %r1579, %r517; + mul.f32 %r1831, %r1579, %r518; + mul.f32 %r1832, %r1579, %r519; + mul.f32 %r1833, %r1579, %r520; + mul.f32 %r1834, %r1579, %r521; + mul.f32 %r1835, %r1579, %r522; + mul.f32 %r1836, %r1587, %r523; + mul.f32 %r1837, %r1587, %r524; + mul.f32 %r1838, %r1587, %r525; + mul.f32 %r1839, %r1587, %r526; + mul.f32 %r1840, %r1587, %r527; + mul.f32 %r1841, %r1587, %r528; + mul.f32 %r1842, %r1587, %r529; + mul.f32 %r1843, %r1587, %r530; + mul.f32 %r1844, %r1595, %r531; + mul.f32 %r1845, %r1595, %r532; + mul.f32 %r1846, %r1595, %r533; + mul.f32 %r1847, %r1595, %r534; + mul.f32 %r1848, %r1595, %r535; + mul.f32 %r1849, %r1595, %r536; + mul.f32 %r1850, %r1595, %r537; + mul.f32 %r1851, %r1595, %r538; + mul.f32 %r1852, %r1603, %r539; + mul.f32 %r1853, %r1603, %r540; + mul.f32 %r1854, %r1603, %r541; + mul.f32 %r1855, %r1603, %r542; + mul.f32 %r1856, %r1603, %r543; + mul.f32 %r1857, %r1603, %r544; + mul.f32 %r1858, %r1603, %r545; + mul.f32 %r1859, %r1603, %r546; + mul.f32 %r1860, %r1611, %r547; + mul.f32 %r1861, %r1611, %r548; + mul.f32 %r1862, %r1611, %r549; + mul.f32 %r1863, %r1611, %r550; + mul.f32 %r1864, %r1611, %r551; + mul.f32 %r1865, %r1611, %r552; + mul.f32 %r1866, %r1611, %r553; + mul.f32 %r1867, %r1611, %r554; + mul.f32 %r1868, %r1619, %r555; + mul.f32 %r1869, %r1619, %r556; + mul.f32 %r1870, %r1619, %r557; + mul.f32 %r1871, %r1619, %r558; + mul.f32 %r1872, %r1619, %r559; + mul.f32 %r1873, %r1619, %r560; + mul.f32 %r1874, %r1619, %r561; + mul.f32 %r1875, %r1619, %r562; + mul.f32 %r1876, %r1627, %r563; + mul.f32 %r1877, %r1627, %r564; + mul.f32 %r1878, %r1627, %r565; + mul.f32 %r1879, %r1627, %r566; + mul.f32 %r1880, %r1627, %r567; + mul.f32 %r1881, %r1627, %r568; + mul.f32 %r1882, %r1627, %r569; + mul.f32 %r1883, %r1627, %r570; + mul.f32 %r1884, %r1635, %r571; + mul.f32 %r1885, %r1635, %r572; + mul.f32 %r1886, %r1635, %r573; + mul.f32 %r1887, %r1635, %r574; + mul.f32 %r1888, %r1635, %r575; + mul.f32 %r1889, %r1635, %r576; + mul.f32 %r1890, %r1635, %r577; + mul.f32 %r1891, %r1635, %r578; + mul.f32 %r1892, %r1643, %r579; + mul.f32 %r1893, %r1643, %r580; + mul.f32 %r1894, %r1643, %r581; + mul.f32 %r1895, %r1643, %r582; + mul.f32 %r1896, %r1643, %r583; + mul.f32 %r1897, %r1643, %r584; + mul.f32 %r1898, %r1643, %r585; + mul.f32 %r1899, %r1643, %r586; + mul.f32 %r1900, %r1651, %r587; + mul.f32 %r1901, %r1651, %r588; + mul.f32 %r1902, %r1651, %r589; + mul.f32 %r1903, %r1651, %r590; + mul.f32 %r1904, %r1651, %r591; + mul.f32 %r1905, %r1651, %r592; + mul.f32 %r1906, %r1651, %r593; + mul.f32 %r1907, %r1651, %r594; + mul.f32 %r1908, %r1659, %r595; + mul.f32 %r1909, %r1659, %r596; + mul.f32 %r1910, %r1659, %r597; + mul.f32 %r1911, %r1659, %r598; + mul.f32 %r1912, %r1659, %r599; + mul.f32 %r1913, %r1659, %r600; + mul.f32 %r1914, %r1659, %r601; + mul.f32 %r1915, %r1659, %r602; + mul.f32 %r1916, %r1667, %r603; + mul.f32 %r1917, %r1667, %r604; + mul.f32 %r1918, %r1667, %r605; + mul.f32 %r1919, %r1667, %r606; + mul.f32 %r1920, %r1667, %r607; + mul.f32 %r1921, %r1667, %r608; + mul.f32 %r1922, %r1667, %r609; + mul.f32 %r1923, %r1667, %r610; + mul.f32 %r1924, %r1675, %r611; + mul.f32 %r1925, %r1675, %r612; + mul.f32 %r1926, %r1675, %r613; + mul.f32 %r1927, %r1675, %r614; + mul.f32 %r1928, %r1675, %r615; + mul.f32 %r1929, %r1675, %r616; + mul.f32 %r1930, %r1675, %r617; + mul.f32 %r1931, %r1675, %r618; + mul.f32 %r1932, %r1683, %r619; + mul.f32 %r1933, %r1683, %r620; + mul.f32 %r1934, %r1683, %r621; + mul.f32 %r1935, %r1683, %r622; + mul.f32 %r1936, %r1683, %r623; + mul.f32 %r1937, %r1683, %r624; + mul.f32 %r1938, %r1683, %r625; + mul.f32 %r1939, %r1683, %r626; + mul.f32 %r1940, %r1691, %r627; + mul.f32 %r1941, %r1691, %r628; + mul.f32 %r1942, %r1691, %r629; + mul.f32 %r1943, %r1691, %r630; + mul.f32 %r1944, %r1691, %r631; + mul.f32 %r1945, %r1691, %r632; + mul.f32 %r1946, %r1691, %r633; + mul.f32 %r1947, %r1691, %r634; + mul.f32 %r1948, %r1699, %r635; + mul.f32 %r1949, %r1699, %r636; + mul.f32 %r1950, %r1699, %r637; + mul.f32 %r1951, %r1699, %r638; + mul.f32 %r1952, %r1699, %r639; + mul.f32 %r1953, %r1699, %r640; + mul.f32 %r1954, %r1699, %r641; + mul.f32 %r1955, %r1699, %r642; + mul.f32 %r1956, %r1707, %r643; + mul.f32 %r1957, %r1707, %r644; + mul.f32 %r1958, %r1707, %r645; + mul.f32 %r1959, %r1707, %r646; + mul.f32 %r1960, %r1707, %r647; + mul.f32 %r1961, %r1707, %r648; + mul.f32 %r1962, %r1707, %r649; + mul.f32 %r1963, %r1707, %r650; + mul.f32 %r1964, %r1715, %r651; + mul.f32 %r1965, %r1715, %r652; + mul.f32 %r1966, %r1715, %r653; + mul.f32 %r1967, %r1715, %r654; + mul.f32 %r1968, %r1715, %r655; + mul.f32 %r1969, %r1715, %r656; + mul.f32 %r1970, %r1715, %r657; + mul.f32 %r1971, %r1715, %r658; + mul.f32 %r1972, %r1723, %r659; + mul.f32 %r1973, %r1723, %r660; + mul.f32 %r1974, %r1723, %r661; + mul.f32 %r1975, %r1723, %r662; + mul.f32 %r1976, %r1723, %r663; + mul.f32 %r1977, %r1723, %r664; + mul.f32 %r1978, %r1723, %r665; + mul.f32 %r1979, %r1723, %r666; + mul.f32 %r1980, %r1731, %r667; + mul.f32 %r1981, %r1731, %r668; + mul.f32 %r1982, %r1731, %r669; + mul.f32 %r1983, %r1731, %r670; + mul.f32 %r1984, %r1731, %r671; + mul.f32 %r1985, %r1731, %r672; + mul.f32 %r1986, %r1731, %r673; + mul.f32 %r1987, %r1731, %r674; + mul.f32 %r1988, %r1739, %r675; + mul.f32 %r1989, %r1739, %r676; + mul.f32 %r1990, %r1739, %r677; + mul.f32 %r1991, %r1739, %r678; + mul.f32 %r1992, %r1739, %r679; + mul.f32 %r1993, %r1739, %r680; + mul.f32 %r1994, %r1739, %r681; + mul.f32 %r1995, %r1739, %r682; + mul.f32 %r1996, %r1747, %r683; + mul.f32 %r1997, %r1747, %r684; + mul.f32 %r1998, %r1747, %r685; + mul.f32 %r1999, %r1747, %r686; + mul.f32 %r2000, %r1747, %r687; + mul.f32 %r2001, %r1747, %r688; + mul.f32 %r2002, %r1747, %r689; + mul.f32 %r2003, %r1747, %r690; + mul.f32 %r2004, %r1755, %r691; + mul.f32 %r2005, %r1755, %r692; + mul.f32 %r2006, %r1755, %r693; + mul.f32 %r2007, %r1755, %r694; + mul.f32 %r2008, %r1755, %r695; + mul.f32 %r2009, %r1755, %r696; + mul.f32 %r2010, %r1755, %r697; + mul.f32 %r2011, %r1755, %r698; + mul.f32 %r2012, %r1763, %r699; + mul.f32 %r2013, %r1763, %r700; + mul.f32 %r2014, %r1763, %r701; + mul.f32 %r2015, %r1763, %r702; + mul.f32 %r2016, %r1763, %r703; + mul.f32 %r2017, %r1763, %r704; + mul.f32 %r2018, %r1763, %r705; + mul.f32 %r2019, %r1763, %r706; + mul.f32 %r2020, %r1771, %r707; + mul.f32 %r2021, %r1771, %r708; + mul.f32 %r2022, %r1771, %r709; + mul.f32 %r2023, %r1771, %r710; + mul.f32 %r2024, %r1771, %r711; + mul.f32 %r2025, %r1771, %r712; + mul.f32 %r2026, %r1771, %r713; + mul.f32 %r2027, %r1771, %r714; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r2028, %r988, %r1772; + mul.f32 %r2029, %r988, %r1773; + mul.f32 %r2030, %r988, %r1774; + mul.f32 %r2031, %r988, %r1775; + mul.f32 %r2032, %r988, %r1776; + mul.f32 %r2033, %r988, %r1777; + mul.f32 %r2034, %r988, %r1778; + mul.f32 %r2035, %r988, %r1779; + mul.f32 %r2036, %r989, %r1780; + mul.f32 %r2037, %r989, %r1781; + mul.f32 %r2038, %r989, %r1782; + mul.f32 %r2039, %r989, %r1783; + mul.f32 %r2040, %r989, %r1784; + mul.f32 %r2041, %r989, %r1785; + mul.f32 %r2042, %r989, %r1786; + mul.f32 %r2043, %r989, %r1787; + mul.f32 %r2044, %r990, %r1788; + mul.f32 %r2045, %r990, %r1789; + mul.f32 %r2046, %r990, %r1790; + mul.f32 %r2047, %r990, %r1791; + mul.f32 %r2048, %r990, %r1792; + mul.f32 %r2049, %r990, %r1793; + mul.f32 %r2050, %r990, %r1794; + mul.f32 %r2051, %r990, %r1795; + mul.f32 %r2052, %r991, %r1796; + mul.f32 %r2053, %r991, %r1797; + mul.f32 %r2054, %r991, %r1798; + mul.f32 %r2055, %r991, %r1799; + mul.f32 %r2056, %r991, %r1800; + mul.f32 %r2057, %r991, %r1801; + mul.f32 %r2058, %r991, %r1802; + mul.f32 %r2059, %r991, %r1803; + mul.f32 %r2060, %r992, %r1804; + mul.f32 %r2061, %r992, %r1805; + mul.f32 %r2062, %r992, %r1806; + mul.f32 %r2063, %r992, %r1807; + mul.f32 %r2064, %r992, %r1808; + mul.f32 %r2065, %r992, %r1809; + mul.f32 %r2066, %r992, %r1810; + mul.f32 %r2067, %r992, %r1811; + mul.f32 %r2068, %r993, %r1812; + mul.f32 %r2069, %r993, %r1813; + mul.f32 %r2070, %r993, %r1814; + mul.f32 %r2071, %r993, %r1815; + mul.f32 %r2072, %r993, %r1816; + mul.f32 %r2073, %r993, %r1817; + mul.f32 %r2074, %r993, %r1818; + mul.f32 %r2075, %r993, %r1819; + mul.f32 %r2076, %r994, %r1820; + mul.f32 %r2077, %r994, %r1821; + mul.f32 %r2078, %r994, %r1822; + mul.f32 %r2079, %r994, %r1823; + mul.f32 %r2080, %r994, %r1824; + mul.f32 %r2081, %r994, %r1825; + mul.f32 %r2082, %r994, %r1826; + mul.f32 %r2083, %r994, %r1827; + mul.f32 %r2084, %r995, %r1828; + mul.f32 %r2085, %r995, %r1829; + mul.f32 %r2086, %r995, %r1830; + mul.f32 %r2087, %r995, %r1831; + mul.f32 %r2088, %r995, %r1832; + mul.f32 %r2089, %r995, %r1833; + mul.f32 %r2090, %r995, %r1834; + mul.f32 %r2091, %r995, %r1835; + mul.f32 %r2092, %r996, %r1836; + mul.f32 %r2093, %r996, %r1837; + mul.f32 %r2094, %r996, %r1838; + mul.f32 %r2095, %r996, %r1839; + mul.f32 %r2096, %r996, %r1840; + mul.f32 %r2097, %r996, %r1841; + mul.f32 %r2098, %r996, %r1842; + mul.f32 %r2099, %r996, %r1843; + mul.f32 %r2100, %r997, %r1844; + mul.f32 %r2101, %r997, %r1845; + mul.f32 %r2102, %r997, %r1846; + mul.f32 %r2103, %r997, %r1847; + mul.f32 %r2104, %r997, %r1848; + mul.f32 %r2105, %r997, %r1849; + mul.f32 %r2106, %r997, %r1850; + mul.f32 %r2107, %r997, %r1851; + mul.f32 %r2108, %r998, %r1852; + mul.f32 %r2109, %r998, %r1853; + mul.f32 %r2110, %r998, %r1854; + mul.f32 %r2111, %r998, %r1855; + mul.f32 %r2112, %r998, %r1856; + mul.f32 %r2113, %r998, %r1857; + mul.f32 %r2114, %r998, %r1858; + mul.f32 %r2115, %r998, %r1859; + mul.f32 %r2116, %r999, %r1860; + mul.f32 %r2117, %r999, %r1861; + mul.f32 %r2118, %r999, %r1862; + mul.f32 %r2119, %r999, %r1863; + mul.f32 %r2120, %r999, %r1864; + mul.f32 %r2121, %r999, %r1865; + mul.f32 %r2122, %r999, %r1866; + mul.f32 %r2123, %r999, %r1867; + mul.f32 %r2124, %r1000, %r1868; + mul.f32 %r2125, %r1000, %r1869; + mul.f32 %r2126, %r1000, %r1870; + mul.f32 %r2127, %r1000, %r1871; + mul.f32 %r2128, %r1000, %r1872; + mul.f32 %r2129, %r1000, %r1873; + mul.f32 %r2130, %r1000, %r1874; + mul.f32 %r2131, %r1000, %r1875; + mul.f32 %r2132, %r1001, %r1876; + mul.f32 %r2133, %r1001, %r1877; + mul.f32 %r2134, %r1001, %r1878; + mul.f32 %r2135, %r1001, %r1879; + mul.f32 %r2136, %r1001, %r1880; + mul.f32 %r2137, %r1001, %r1881; + mul.f32 %r2138, %r1001, %r1882; + mul.f32 %r2139, %r1001, %r1883; + mul.f32 %r2140, %r1002, %r1884; + mul.f32 %r2141, %r1002, %r1885; + mul.f32 %r2142, %r1002, %r1886; + mul.f32 %r2143, %r1002, %r1887; + mul.f32 %r2144, %r1002, %r1888; + mul.f32 %r2145, %r1002, %r1889; + mul.f32 %r2146, %r1002, %r1890; + mul.f32 %r2147, %r1002, %r1891; + mul.f32 %r2148, %r1003, %r1892; + mul.f32 %r2149, %r1003, %r1893; + mul.f32 %r2150, %r1003, %r1894; + mul.f32 %r2151, %r1003, %r1895; + mul.f32 %r2152, %r1003, %r1896; + mul.f32 %r2153, %r1003, %r1897; + mul.f32 %r2154, %r1003, %r1898; + mul.f32 %r2155, %r1003, %r1899; + mul.f32 %r2156, %r1004, %r1900; + mul.f32 %r2157, %r1004, %r1901; + mul.f32 %r2158, %r1004, %r1902; + mul.f32 %r2159, %r1004, %r1903; + mul.f32 %r2160, %r1004, %r1904; + mul.f32 %r2161, %r1004, %r1905; + mul.f32 %r2162, %r1004, %r1906; + mul.f32 %r2163, %r1004, %r1907; + mul.f32 %r2164, %r1005, %r1908; + mul.f32 %r2165, %r1005, %r1909; + mul.f32 %r2166, %r1005, %r1910; + mul.f32 %r2167, %r1005, %r1911; + mul.f32 %r2168, %r1005, %r1912; + mul.f32 %r2169, %r1005, %r1913; + mul.f32 %r2170, %r1005, %r1914; + mul.f32 %r2171, %r1005, %r1915; + mul.f32 %r2172, %r1006, %r1916; + mul.f32 %r2173, %r1006, %r1917; + mul.f32 %r2174, %r1006, %r1918; + mul.f32 %r2175, %r1006, %r1919; + mul.f32 %r2176, %r1006, %r1920; + mul.f32 %r2177, %r1006, %r1921; + mul.f32 %r2178, %r1006, %r1922; + mul.f32 %r2179, %r1006, %r1923; + mul.f32 %r2180, %r1007, %r1924; + mul.f32 %r2181, %r1007, %r1925; + mul.f32 %r2182, %r1007, %r1926; + mul.f32 %r2183, %r1007, %r1927; + mul.f32 %r2184, %r1007, %r1928; + mul.f32 %r2185, %r1007, %r1929; + mul.f32 %r2186, %r1007, %r1930; + mul.f32 %r2187, %r1007, %r1931; + mul.f32 %r2188, %r1008, %r1932; + mul.f32 %r2189, %r1008, %r1933; + mul.f32 %r2190, %r1008, %r1934; + mul.f32 %r2191, %r1008, %r1935; + mul.f32 %r2192, %r1008, %r1936; + mul.f32 %r2193, %r1008, %r1937; + mul.f32 %r2194, %r1008, %r1938; + mul.f32 %r2195, %r1008, %r1939; + mul.f32 %r2196, %r1009, %r1940; + mul.f32 %r2197, %r1009, %r1941; + mul.f32 %r2198, %r1009, %r1942; + mul.f32 %r2199, %r1009, %r1943; + mul.f32 %r2200, %r1009, %r1944; + mul.f32 %r2201, %r1009, %r1945; + mul.f32 %r2202, %r1009, %r1946; + mul.f32 %r2203, %r1009, %r1947; + mul.f32 %r2204, %r1010, %r1948; + mul.f32 %r2205, %r1010, %r1949; + mul.f32 %r2206, %r1010, %r1950; + mul.f32 %r2207, %r1010, %r1951; + mul.f32 %r2208, %r1010, %r1952; + mul.f32 %r2209, %r1010, %r1953; + mul.f32 %r2210, %r1010, %r1954; + mul.f32 %r2211, %r1010, %r1955; + mul.f32 %r2212, %r1011, %r1956; + mul.f32 %r2213, %r1011, %r1957; + mul.f32 %r2214, %r1011, %r1958; + mul.f32 %r2215, %r1011, %r1959; + mul.f32 %r2216, %r1011, %r1960; + mul.f32 %r2217, %r1011, %r1961; + mul.f32 %r2218, %r1011, %r1962; + mul.f32 %r2219, %r1011, %r1963; + mul.f32 %r2220, %r1012, %r1964; + mul.f32 %r2221, %r1012, %r1965; + mul.f32 %r2222, %r1012, %r1966; + mul.f32 %r2223, %r1012, %r1967; + mul.f32 %r2224, %r1012, %r1968; + mul.f32 %r2225, %r1012, %r1969; + mul.f32 %r2226, %r1012, %r1970; + mul.f32 %r2227, %r1012, %r1971; + mul.f32 %r2228, %r1013, %r1972; + mul.f32 %r2229, %r1013, %r1973; + mul.f32 %r2230, %r1013, %r1974; + mul.f32 %r2231, %r1013, %r1975; + mul.f32 %r2232, %r1013, %r1976; + mul.f32 %r2233, %r1013, %r1977; + mul.f32 %r2234, %r1013, %r1978; + mul.f32 %r2235, %r1013, %r1979; + mul.f32 %r2236, %r1014, %r1980; + mul.f32 %r2237, %r1014, %r1981; + mul.f32 %r2238, %r1014, %r1982; + mul.f32 %r2239, %r1014, %r1983; + mul.f32 %r2240, %r1014, %r1984; + mul.f32 %r2241, %r1014, %r1985; + mul.f32 %r2242, %r1014, %r1986; + mul.f32 %r2243, %r1014, %r1987; + mul.f32 %r2244, %r1015, %r1988; + mul.f32 %r2245, %r1015, %r1989; + mul.f32 %r2246, %r1015, %r1990; + mul.f32 %r2247, %r1015, %r1991; + mul.f32 %r2248, %r1015, %r1992; + mul.f32 %r2249, %r1015, %r1993; + mul.f32 %r2250, %r1015, %r1994; + mul.f32 %r2251, %r1015, %r1995; + mul.f32 %r2252, %r1016, %r1996; + mul.f32 %r2253, %r1016, %r1997; + mul.f32 %r2254, %r1016, %r1998; + mul.f32 %r2255, %r1016, %r1999; + mul.f32 %r2256, %r1016, %r2000; + mul.f32 %r2257, %r1016, %r2001; + mul.f32 %r2258, %r1016, %r2002; + mul.f32 %r2259, %r1016, %r2003; + mul.f32 %r2260, %r1017, %r2004; + mul.f32 %r2261, %r1017, %r2005; + mul.f32 %r2262, %r1017, %r2006; + mul.f32 %r2263, %r1017, %r2007; + mul.f32 %r2264, %r1017, %r2008; + mul.f32 %r2265, %r1017, %r2009; + mul.f32 %r2266, %r1017, %r2010; + mul.f32 %r2267, %r1017, %r2011; + mul.f32 %r2268, %r1018, %r2012; + mul.f32 %r2269, %r1018, %r2013; + mul.f32 %r2270, %r1018, %r2014; + mul.f32 %r2271, %r1018, %r2015; + mul.f32 %r2272, %r1018, %r2016; + mul.f32 %r2273, %r1018, %r2017; + mul.f32 %r2274, %r1018, %r2018; + mul.f32 %r2275, %r1018, %r2019; + mul.f32 %r2276, %r1019, %r2020; + mul.f32 %r2277, %r1019, %r2021; + mul.f32 %r2278, %r1019, %r2022; + mul.f32 %r2279, %r1019, %r2023; + mul.f32 %r2280, %r1019, %r2024; + mul.f32 %r2281, %r1019, %r2025; + mul.f32 %r2282, %r1019, %r2026; + mul.f32 %r2283, %r1019, %r2027; + .loc 1 107 36 // l2norm.py:107:36 + neg.f32 %r2284, %r2028; + fma.rn.f32 %r2285, %r988, %r717, %r2284; + neg.f32 %r2286, %r2029; + fma.rn.f32 %r2287, %r988, %r718, %r2286; + neg.f32 %r2288, %r2030; + fma.rn.f32 %r2289, %r988, %r719, %r2288; + neg.f32 %r2290, %r2031; + fma.rn.f32 %r2291, %r988, %r720, %r2290; + neg.f32 %r2292, %r2032; + fma.rn.f32 %r2293, %r988, %r721, %r2292; + neg.f32 %r2294, %r2033; + fma.rn.f32 %r2295, %r988, %r722, %r2294; + neg.f32 %r2296, %r2034; + fma.rn.f32 %r2297, %r988, %r723, %r2296; + neg.f32 %r2298, %r2035; + fma.rn.f32 %r2299, %r988, %r724, %r2298; + neg.f32 %r2300, %r2036; + fma.rn.f32 %r2301, %r989, %r725, %r2300; + neg.f32 %r2302, %r2037; + fma.rn.f32 %r2303, %r989, %r726, %r2302; + neg.f32 %r2304, %r2038; + fma.rn.f32 %r2305, %r989, %r727, %r2304; + neg.f32 %r2306, %r2039; + fma.rn.f32 %r2307, %r989, %r728, %r2306; + neg.f32 %r2308, %r2040; + fma.rn.f32 %r2309, %r989, %r729, %r2308; + neg.f32 %r2310, %r2041; + fma.rn.f32 %r2311, %r989, %r730, %r2310; + neg.f32 %r2312, %r2042; + fma.rn.f32 %r2313, %r989, %r731, %r2312; + neg.f32 %r2314, %r2043; + fma.rn.f32 %r2315, %r989, %r732, %r2314; + mov.b64 %rd304, {%r1020, %r1021}; + mov.b64 %rd305, {%r2044, %r2045}; + sub.f32x2 %rd306, %rd304, %rd305; + mov.b64 %rd307, {%r1022, %r1023}; + mov.b64 %rd308, {%r2046, %r2047}; + sub.f32x2 %rd309, %rd307, %rd308; + mov.b64 %rd310, {%r1024, %r1025}; + mov.b64 %rd311, {%r2048, %r2049}; + sub.f32x2 %rd312, %rd310, %rd311; + mov.b64 %rd313, {%r1026, %r1027}; + mov.b64 %rd314, {%r2050, %r2051}; + sub.f32x2 %rd315, %rd313, %rd314; + mov.b64 %rd316, {%r1028, %r1029}; + mov.b64 %rd317, {%r2052, %r2053}; + sub.f32x2 %rd318, %rd316, %rd317; + mov.b64 %rd319, {%r1030, %r1031}; + mov.b64 %rd320, {%r2054, %r2055}; + sub.f32x2 %rd321, %rd319, %rd320; + mov.b64 %rd322, {%r1032, %r1033}; + mov.b64 %rd323, {%r2056, %r2057}; + sub.f32x2 %rd324, %rd322, %rd323; + mov.b64 %rd325, {%r1034, %r1035}; + mov.b64 %rd326, {%r2058, %r2059}; + sub.f32x2 %rd327, %rd325, %rd326; + mov.b64 %rd328, {%r1036, %r1037}; + mov.b64 %rd329, {%r2060, %r2061}; + sub.f32x2 %rd330, %rd328, %rd329; + mov.b64 %rd331, {%r1038, %r1039}; + mov.b64 %rd332, {%r2062, %r2063}; + sub.f32x2 %rd333, %rd331, %rd332; + mov.b64 %rd334, {%r1040, %r1041}; + mov.b64 %rd335, {%r2064, %r2065}; + sub.f32x2 %rd336, %rd334, %rd335; + mov.b64 %rd337, {%r1042, %r1043}; + mov.b64 %rd338, {%r2066, %r2067}; + sub.f32x2 %rd339, %rd337, %rd338; + mov.b64 %rd340, {%r1044, %r1045}; + mov.b64 %rd341, {%r2068, %r2069}; + sub.f32x2 %rd342, %rd340, %rd341; + mov.b64 %rd343, {%r1046, %r1047}; + mov.b64 %rd344, {%r2070, %r2071}; + sub.f32x2 %rd345, %rd343, %rd344; + mov.b64 %rd346, {%r1048, %r1049}; + mov.b64 %rd347, {%r2072, %r2073}; + sub.f32x2 %rd348, %rd346, %rd347; + mov.b64 %rd349, {%r1050, %r1051}; + mov.b64 %rd350, {%r2074, %r2075}; + sub.f32x2 %rd351, %rd349, %rd350; + mov.b64 %rd352, {%r1052, %r1053}; + mov.b64 %rd353, {%r2076, %r2077}; + sub.f32x2 %rd354, %rd352, %rd353; + mov.b64 %rd355, {%r1054, %r1055}; + mov.b64 %rd356, {%r2078, %r2079}; + sub.f32x2 %rd357, %rd355, %rd356; + mov.b64 %rd358, {%r1056, %r1057}; + mov.b64 %rd359, {%r2080, %r2081}; + sub.f32x2 %rd360, %rd358, %rd359; + mov.b64 %rd361, {%r1058, %r1059}; + mov.b64 %rd362, {%r2082, %r2083}; + sub.f32x2 %rd363, %rd361, %rd362; + mov.b64 %rd364, {%r1060, %r1061}; + mov.b64 %rd365, {%r2084, %r2085}; + sub.f32x2 %rd366, %rd364, %rd365; + mov.b64 %rd367, {%r1062, %r1063}; + mov.b64 %rd368, {%r2086, %r2087}; + sub.f32x2 %rd369, %rd367, %rd368; + mov.b64 %rd370, {%r1064, %r1065}; + mov.b64 %rd371, {%r2088, %r2089}; + sub.f32x2 %rd372, %rd370, %rd371; + mov.b64 %rd373, {%r1066, %r1067}; + mov.b64 %rd374, {%r2090, %r2091}; + sub.f32x2 %rd375, %rd373, %rd374; + mov.b64 %rd376, {%r1068, %r1069}; + mov.b64 %rd377, {%r2092, %r2093}; + sub.f32x2 %rd378, %rd376, %rd377; + mov.b64 %rd379, {%r1070, %r1071}; + mov.b64 %rd380, {%r2094, %r2095}; + sub.f32x2 %rd381, %rd379, %rd380; + mov.b64 %rd382, {%r1072, %r1073}; + mov.b64 %rd383, {%r2096, %r2097}; + sub.f32x2 %rd384, %rd382, %rd383; + mov.b64 %rd385, {%r1074, %r1075}; + mov.b64 %rd386, {%r2098, %r2099}; + sub.f32x2 %rd387, %rd385, %rd386; + mov.b64 %rd388, {%r1076, %r1077}; + mov.b64 %rd389, {%r2100, %r2101}; + sub.f32x2 %rd390, %rd388, %rd389; + mov.b64 %rd391, {%r1078, %r1079}; + mov.b64 %rd392, {%r2102, %r2103}; + sub.f32x2 %rd393, %rd391, %rd392; + mov.b64 %rd394, {%r1080, %r1081}; + mov.b64 %rd395, {%r2104, %r2105}; + sub.f32x2 %rd396, %rd394, %rd395; + mov.b64 %rd397, {%r1082, %r1083}; + mov.b64 %rd398, {%r2106, %r2107}; + sub.f32x2 %rd399, %rd397, %rd398; + mov.b64 %rd400, {%r1084, %r1085}; + mov.b64 %rd401, {%r2108, %r2109}; + sub.f32x2 %rd402, %rd400, %rd401; + mov.b64 %rd403, {%r1086, %r1087}; + mov.b64 %rd404, {%r2110, %r2111}; + sub.f32x2 %rd405, %rd403, %rd404; + mov.b64 %rd406, {%r1088, %r1089}; + mov.b64 %rd407, {%r2112, %r2113}; + sub.f32x2 %rd408, %rd406, %rd407; + mov.b64 %rd409, {%r1090, %r1091}; + mov.b64 %rd410, {%r2114, %r2115}; + sub.f32x2 %rd411, %rd409, %rd410; + mov.b64 %rd412, {%r1092, %r1093}; + mov.b64 %rd413, {%r2116, %r2117}; + sub.f32x2 %rd414, %rd412, %rd413; + mov.b64 %rd415, {%r1094, %r1095}; + mov.b64 %rd416, {%r2118, %r2119}; + sub.f32x2 %rd417, %rd415, %rd416; + mov.b64 %rd418, {%r1096, %r1097}; + mov.b64 %rd419, {%r2120, %r2121}; + sub.f32x2 %rd420, %rd418, %rd419; + mov.b64 %rd421, {%r1098, %r1099}; + mov.b64 %rd422, {%r2122, %r2123}; + sub.f32x2 %rd423, %rd421, %rd422; + mov.b64 %rd424, {%r1100, %r1101}; + mov.b64 %rd425, {%r2124, %r2125}; + sub.f32x2 %rd426, %rd424, %rd425; + mov.b64 %rd427, {%r1102, %r1103}; + mov.b64 %rd428, {%r2126, %r2127}; + sub.f32x2 %rd429, %rd427, %rd428; + mov.b64 %rd430, {%r1104, %r1105}; + mov.b64 %rd431, {%r2128, %r2129}; + sub.f32x2 %rd432, %rd430, %rd431; + mov.b64 %rd433, {%r1106, %r1107}; + mov.b64 %rd434, {%r2130, %r2131}; + sub.f32x2 %rd435, %rd433, %rd434; + mov.b64 %rd436, {%r1108, %r1109}; + mov.b64 %rd437, {%r2132, %r2133}; + sub.f32x2 %rd438, %rd436, %rd437; + mov.b64 %rd439, {%r1110, %r1111}; + mov.b64 %rd440, {%r2134, %r2135}; + sub.f32x2 %rd441, %rd439, %rd440; + mov.b64 %rd442, {%r1112, %r1113}; + mov.b64 %rd443, {%r2136, %r2137}; + sub.f32x2 %rd444, %rd442, %rd443; + mov.b64 %rd445, {%r1114, %r1115}; + mov.b64 %rd446, {%r2138, %r2139}; + sub.f32x2 %rd447, %rd445, %rd446; + mov.b64 %rd448, {%r1116, %r1117}; + mov.b64 %rd449, {%r2140, %r2141}; + sub.f32x2 %rd450, %rd448, %rd449; + mov.b64 %rd451, {%r1118, %r1119}; + mov.b64 %rd452, {%r2142, %r2143}; + sub.f32x2 %rd453, %rd451, %rd452; + mov.b64 %rd454, {%r1120, %r1121}; + mov.b64 %rd455, {%r2144, %r2145}; + sub.f32x2 %rd456, %rd454, %rd455; + mov.b64 %rd457, {%r1122, %r1123}; + mov.b64 %rd458, {%r2146, %r2147}; + sub.f32x2 %rd459, %rd457, %rd458; + mov.b64 %rd460, {%r1124, %r1125}; + mov.b64 %rd461, {%r2148, %r2149}; + sub.f32x2 %rd462, %rd460, %rd461; + mov.b64 %rd463, {%r1126, %r1127}; + mov.b64 %rd464, {%r2150, %r2151}; + sub.f32x2 %rd465, %rd463, %rd464; + mov.b64 %rd466, {%r1128, %r1129}; + mov.b64 %rd467, {%r2152, %r2153}; + sub.f32x2 %rd468, %rd466, %rd467; + mov.b64 %rd469, {%r1130, %r1131}; + mov.b64 %rd470, {%r2154, %r2155}; + sub.f32x2 %rd471, %rd469, %rd470; + mov.b64 %rd472, {%r1132, %r1133}; + mov.b64 %rd473, {%r2156, %r2157}; + sub.f32x2 %rd474, %rd472, %rd473; + mov.b64 %rd475, {%r1134, %r1135}; + mov.b64 %rd476, {%r2158, %r2159}; + sub.f32x2 %rd477, %rd475, %rd476; + mov.b64 %rd478, {%r1136, %r1137}; + mov.b64 %rd479, {%r2160, %r2161}; + sub.f32x2 %rd480, %rd478, %rd479; + mov.b64 %rd481, {%r1138, %r1139}; + mov.b64 %rd482, {%r2162, %r2163}; + sub.f32x2 %rd483, %rd481, %rd482; + mov.b64 %rd484, {%r1140, %r1141}; + mov.b64 %rd485, {%r2164, %r2165}; + sub.f32x2 %rd486, %rd484, %rd485; + mov.b64 %rd487, {%r1142, %r1143}; + mov.b64 %rd488, {%r2166, %r2167}; + sub.f32x2 %rd489, %rd487, %rd488; + mov.b64 %rd490, {%r1144, %r1145}; + mov.b64 %rd491, {%r2168, %r2169}; + sub.f32x2 %rd492, %rd490, %rd491; + mov.b64 %rd493, {%r1146, %r1147}; + mov.b64 %rd494, {%r2170, %r2171}; + sub.f32x2 %rd495, %rd493, %rd494; + mov.b64 %rd496, {%r1148, %r1149}; + mov.b64 %rd497, {%r2172, %r2173}; + sub.f32x2 %rd498, %rd496, %rd497; + mov.b64 %rd499, {%r1150, %r1151}; + mov.b64 %rd500, {%r2174, %r2175}; + sub.f32x2 %rd501, %rd499, %rd500; + mov.b64 %rd502, {%r1152, %r1153}; + mov.b64 %rd503, {%r2176, %r2177}; + sub.f32x2 %rd504, %rd502, %rd503; + mov.b64 %rd505, {%r1154, %r1155}; + mov.b64 %rd506, {%r2178, %r2179}; + sub.f32x2 %rd507, %rd505, %rd506; + mov.b64 %rd508, {%r1156, %r1157}; + mov.b64 %rd509, {%r2180, %r2181}; + sub.f32x2 %rd510, %rd508, %rd509; + mov.b64 %rd511, {%r1158, %r1159}; + mov.b64 %rd512, {%r2182, %r2183}; + sub.f32x2 %rd513, %rd511, %rd512; + mov.b64 %rd514, {%r1160, %r1161}; + mov.b64 %rd515, {%r2184, %r2185}; + sub.f32x2 %rd516, %rd514, %rd515; + mov.b64 %rd517, {%r1162, %r1163}; + mov.b64 %rd518, {%r2186, %r2187}; + sub.f32x2 %rd519, %rd517, %rd518; + mov.b64 %rd520, {%r1164, %r1165}; + mov.b64 %rd521, {%r2188, %r2189}; + sub.f32x2 %rd522, %rd520, %rd521; + mov.b64 %rd523, {%r1166, %r1167}; + mov.b64 %rd524, {%r2190, %r2191}; + sub.f32x2 %rd525, %rd523, %rd524; + mov.b64 %rd526, {%r1168, %r1169}; + mov.b64 %rd527, {%r2192, %r2193}; + sub.f32x2 %rd528, %rd526, %rd527; + mov.b64 %rd529, {%r1170, %r1171}; + mov.b64 %rd530, {%r2194, %r2195}; + sub.f32x2 %rd531, %rd529, %rd530; + mov.b64 %rd532, {%r1172, %r1173}; + mov.b64 %rd533, {%r2196, %r2197}; + sub.f32x2 %rd534, %rd532, %rd533; + mov.b64 %rd535, {%r1174, %r1175}; + mov.b64 %rd536, {%r2198, %r2199}; + sub.f32x2 %rd537, %rd535, %rd536; + mov.b64 %rd538, {%r1176, %r1177}; + mov.b64 %rd539, {%r2200, %r2201}; + sub.f32x2 %rd540, %rd538, %rd539; + mov.b64 %rd541, {%r1178, %r1179}; + mov.b64 %rd542, {%r2202, %r2203}; + sub.f32x2 %rd543, %rd541, %rd542; + mov.b64 %rd544, {%r1180, %r1181}; + mov.b64 %rd545, {%r2204, %r2205}; + sub.f32x2 %rd546, %rd544, %rd545; + mov.b64 %rd547, {%r1182, %r1183}; + mov.b64 %rd548, {%r2206, %r2207}; + sub.f32x2 %rd549, %rd547, %rd548; + mov.b64 %rd550, {%r1184, %r1185}; + mov.b64 %rd551, {%r2208, %r2209}; + sub.f32x2 %rd552, %rd550, %rd551; + mov.b64 %rd553, {%r1186, %r1187}; + mov.b64 %rd554, {%r2210, %r2211}; + sub.f32x2 %rd555, %rd553, %rd554; + mov.b64 %rd556, {%r1188, %r1189}; + mov.b64 %rd557, {%r2212, %r2213}; + sub.f32x2 %rd558, %rd556, %rd557; + mov.b64 %rd559, {%r1190, %r1191}; + mov.b64 %rd560, {%r2214, %r2215}; + sub.f32x2 %rd561, %rd559, %rd560; + mov.b64 %rd562, {%r1192, %r1193}; + mov.b64 %rd563, {%r2216, %r2217}; + sub.f32x2 %rd564, %rd562, %rd563; + mov.b64 %rd565, {%r1194, %r1195}; + mov.b64 %rd566, {%r2218, %r2219}; + sub.f32x2 %rd567, %rd565, %rd566; + mov.b64 %rd568, {%r1196, %r1197}; + mov.b64 %rd569, {%r2220, %r2221}; + sub.f32x2 %rd570, %rd568, %rd569; + mov.b64 %rd571, {%r1198, %r1199}; + mov.b64 %rd572, {%r2222, %r2223}; + sub.f32x2 %rd573, %rd571, %rd572; + mov.b64 %rd574, {%r1200, %r1201}; + mov.b64 %rd575, {%r2224, %r2225}; + sub.f32x2 %rd576, %rd574, %rd575; + mov.b64 %rd577, {%r1202, %r1203}; + mov.b64 %rd578, {%r2226, %r2227}; + sub.f32x2 %rd579, %rd577, %rd578; + mov.b64 %rd580, {%r1204, %r1205}; + mov.b64 %rd581, {%r2228, %r2229}; + sub.f32x2 %rd582, %rd580, %rd581; + mov.b64 %rd583, {%r1206, %r1207}; + mov.b64 %rd584, {%r2230, %r2231}; + sub.f32x2 %rd585, %rd583, %rd584; + mov.b64 %rd586, {%r1208, %r1209}; + mov.b64 %rd587, {%r2232, %r2233}; + sub.f32x2 %rd588, %rd586, %rd587; + mov.b64 %rd589, {%r1210, %r1211}; + mov.b64 %rd590, {%r2234, %r2235}; + sub.f32x2 %rd591, %rd589, %rd590; + mov.b64 %rd592, {%r1212, %r1213}; + mov.b64 %rd593, {%r2236, %r2237}; + sub.f32x2 %rd594, %rd592, %rd593; + mov.b64 %rd595, {%r1214, %r1215}; + mov.b64 %rd596, {%r2238, %r2239}; + sub.f32x2 %rd597, %rd595, %rd596; + mov.b64 %rd598, {%r1216, %r1217}; + mov.b64 %rd599, {%r2240, %r2241}; + sub.f32x2 %rd600, %rd598, %rd599; + mov.b64 %rd601, {%r1218, %r1219}; + mov.b64 %rd602, {%r2242, %r2243}; + sub.f32x2 %rd603, %rd601, %rd602; + mov.b64 %rd604, {%r1220, %r1221}; + mov.b64 %rd605, {%r2244, %r2245}; + sub.f32x2 %rd606, %rd604, %rd605; + mov.b64 %rd607, {%r1222, %r1223}; + mov.b64 %rd608, {%r2246, %r2247}; + sub.f32x2 %rd609, %rd607, %rd608; + mov.b64 %rd610, {%r1224, %r1225}; + mov.b64 %rd611, {%r2248, %r2249}; + sub.f32x2 %rd612, %rd610, %rd611; + mov.b64 %rd613, {%r1226, %r1227}; + mov.b64 %rd614, {%r2250, %r2251}; + sub.f32x2 %rd615, %rd613, %rd614; + mov.b64 %rd616, {%r1228, %r1229}; + mov.b64 %rd617, {%r2252, %r2253}; + sub.f32x2 %rd618, %rd616, %rd617; + mov.b64 %rd619, {%r1230, %r1231}; + mov.b64 %rd620, {%r2254, %r2255}; + sub.f32x2 %rd621, %rd619, %rd620; + mov.b64 %rd622, {%r1232, %r1233}; + mov.b64 %rd623, {%r2256, %r2257}; + sub.f32x2 %rd624, %rd622, %rd623; + mov.b64 %rd625, {%r1234, %r1235}; + mov.b64 %rd626, {%r2258, %r2259}; + sub.f32x2 %rd627, %rd625, %rd626; + mov.b64 %rd628, {%r1236, %r1237}; + mov.b64 %rd629, {%r2260, %r2261}; + sub.f32x2 %rd630, %rd628, %rd629; + mov.b64 %rd631, {%r1238, %r1239}; + mov.b64 %rd632, {%r2262, %r2263}; + sub.f32x2 %rd633, %rd631, %rd632; + mov.b64 %rd634, {%r1240, %r1241}; + mov.b64 %rd635, {%r2264, %r2265}; + sub.f32x2 %rd636, %rd634, %rd635; + mov.b64 %rd637, {%r1242, %r1243}; + mov.b64 %rd638, {%r2266, %r2267}; + sub.f32x2 %rd639, %rd637, %rd638; + mov.b64 %rd640, {%r1244, %r1245}; + mov.b64 %rd641, {%r2268, %r2269}; + sub.f32x2 %rd642, %rd640, %rd641; + mov.b64 %rd643, {%r1246, %r1247}; + mov.b64 %rd644, {%r2270, %r2271}; + sub.f32x2 %rd645, %rd643, %rd644; + mov.b64 %rd646, {%r1248, %r1249}; + mov.b64 %rd647, {%r2272, %r2273}; + sub.f32x2 %rd648, %rd646, %rd647; + mov.b64 %rd649, {%r1250, %r1251}; + mov.b64 %rd650, {%r2274, %r2275}; + sub.f32x2 %rd651, %rd649, %rd650; + mov.b64 %rd652, {%r1252, %r1253}; + mov.b64 %rd653, {%r2276, %r2277}; + sub.f32x2 %rd654, %rd652, %rd653; + mov.b64 %rd655, {%r1254, %r1255}; + mov.b64 %rd656, {%r2278, %r2279}; + sub.f32x2 %rd657, %rd655, %rd656; + mov.b64 %rd658, {%r1256, %r1257}; + mov.b64 %rd659, {%r2280, %r2281}; + sub.f32x2 %rd660, %rd658, %rd659; + mov.b64 %rd661, {%r1258, %r1259}; + mov.b64 %rd662, {%r2282, %r2283}; + sub.f32x2 %rd663, %rd661, %rd662; + .loc 1 108 27 // l2norm.py:108:27 + mov.b64 {%r2316, %r2317}, %rd306; + cvt.rn.bf16x2.f32 %r270, %r2317, %r2316; + mov.b64 {%r2318, %r2319}, %rd309; + cvt.rn.bf16x2.f32 %r271, %r2319, %r2318; + mov.b64 {%r2320, %r2321}, %rd312; + cvt.rn.bf16x2.f32 %r272, %r2321, %r2320; + mov.b64 {%r2322, %r2323}, %rd315; + cvt.rn.bf16x2.f32 %r273, %r2323, %r2322; + mov.b64 {%r2324, %r2325}, %rd318; + cvt.rn.bf16x2.f32 %r274, %r2325, %r2324; + mov.b64 {%r2326, %r2327}, %rd321; + cvt.rn.bf16x2.f32 %r275, %r2327, %r2326; + mov.b64 {%r2328, %r2329}, %rd324; + cvt.rn.bf16x2.f32 %r276, %r2329, %r2328; + mov.b64 {%r2330, %r2331}, %rd327; + cvt.rn.bf16x2.f32 %r277, %r2331, %r2330; + mov.b64 {%r2332, %r2333}, %rd330; + cvt.rn.bf16x2.f32 %r278, %r2333, %r2332; + mov.b64 {%r2334, %r2335}, %rd333; + cvt.rn.bf16x2.f32 %r279, %r2335, %r2334; + mov.b64 {%r2336, %r2337}, %rd336; + cvt.rn.bf16x2.f32 %r280, %r2337, %r2336; + mov.b64 {%r2338, %r2339}, %rd339; + cvt.rn.bf16x2.f32 %r281, %r2339, %r2338; + mov.b64 {%r2340, %r2341}, %rd342; + cvt.rn.bf16x2.f32 %r282, %r2341, %r2340; + mov.b64 {%r2342, %r2343}, %rd345; + cvt.rn.bf16x2.f32 %r283, %r2343, %r2342; + mov.b64 {%r2344, %r2345}, %rd348; + cvt.rn.bf16x2.f32 %r284, %r2345, %r2344; + mov.b64 {%r2346, %r2347}, %rd351; + cvt.rn.bf16x2.f32 %r285, %r2347, %r2346; + mov.b64 {%r2348, %r2349}, %rd354; + cvt.rn.bf16x2.f32 %r286, %r2349, %r2348; + mov.b64 {%r2350, %r2351}, %rd357; + cvt.rn.bf16x2.f32 %r287, %r2351, %r2350; + mov.b64 {%r2352, %r2353}, %rd360; + cvt.rn.bf16x2.f32 %r288, %r2353, %r2352; + mov.b64 {%r2354, %r2355}, %rd363; + cvt.rn.bf16x2.f32 %r289, %r2355, %r2354; + mov.b64 {%r2356, %r2357}, %rd366; + cvt.rn.bf16x2.f32 %r290, %r2357, %r2356; + mov.b64 {%r2358, %r2359}, %rd369; + cvt.rn.bf16x2.f32 %r291, %r2359, %r2358; + mov.b64 {%r2360, %r2361}, %rd372; + cvt.rn.bf16x2.f32 %r292, %r2361, %r2360; + mov.b64 {%r2362, %r2363}, %rd375; + cvt.rn.bf16x2.f32 %r293, %r2363, %r2362; + mov.b64 {%r2364, %r2365}, %rd378; + cvt.rn.bf16x2.f32 %r294, %r2365, %r2364; + mov.b64 {%r2366, %r2367}, %rd381; + cvt.rn.bf16x2.f32 %r295, %r2367, %r2366; + mov.b64 {%r2368, %r2369}, %rd384; + cvt.rn.bf16x2.f32 %r296, %r2369, %r2368; + mov.b64 {%r2370, %r2371}, %rd387; + cvt.rn.bf16x2.f32 %r297, %r2371, %r2370; + mov.b64 {%r2372, %r2373}, %rd390; + cvt.rn.bf16x2.f32 %r298, %r2373, %r2372; + mov.b64 {%r2374, %r2375}, %rd393; + cvt.rn.bf16x2.f32 %r299, %r2375, %r2374; + mov.b64 {%r2376, %r2377}, %rd396; + cvt.rn.bf16x2.f32 %r300, %r2377, %r2376; + mov.b64 {%r2378, %r2379}, %rd399; + cvt.rn.bf16x2.f32 %r301, %r2379, %r2378; + mov.b64 {%r2380, %r2381}, %rd402; + cvt.rn.bf16x2.f32 %r302, %r2381, %r2380; + mov.b64 {%r2382, %r2383}, %rd405; + cvt.rn.bf16x2.f32 %r303, %r2383, %r2382; + mov.b64 {%r2384, %r2385}, %rd408; + cvt.rn.bf16x2.f32 %r304, %r2385, %r2384; + mov.b64 {%r2386, %r2387}, %rd411; + cvt.rn.bf16x2.f32 %r305, %r2387, %r2386; + mov.b64 {%r2388, %r2389}, %rd414; + cvt.rn.bf16x2.f32 %r306, %r2389, %r2388; + mov.b64 {%r2390, %r2391}, %rd417; + cvt.rn.bf16x2.f32 %r307, %r2391, %r2390; + mov.b64 {%r2392, %r2393}, %rd420; + cvt.rn.bf16x2.f32 %r308, %r2393, %r2392; + mov.b64 {%r2394, %r2395}, %rd423; + cvt.rn.bf16x2.f32 %r309, %r2395, %r2394; + mov.b64 {%r2396, %r2397}, %rd426; + cvt.rn.bf16x2.f32 %r310, %r2397, %r2396; + mov.b64 {%r2398, %r2399}, %rd429; + cvt.rn.bf16x2.f32 %r311, %r2399, %r2398; + mov.b64 {%r2400, %r2401}, %rd432; + cvt.rn.bf16x2.f32 %r312, %r2401, %r2400; + mov.b64 {%r2402, %r2403}, %rd435; + cvt.rn.bf16x2.f32 %r313, %r2403, %r2402; + mov.b64 {%r2404, %r2405}, %rd438; + cvt.rn.bf16x2.f32 %r314, %r2405, %r2404; + mov.b64 {%r2406, %r2407}, %rd441; + cvt.rn.bf16x2.f32 %r315, %r2407, %r2406; + mov.b64 {%r2408, %r2409}, %rd444; + cvt.rn.bf16x2.f32 %r316, %r2409, %r2408; + mov.b64 {%r2410, %r2411}, %rd447; + cvt.rn.bf16x2.f32 %r317, %r2411, %r2410; + mov.b64 {%r2412, %r2413}, %rd450; + cvt.rn.bf16x2.f32 %r318, %r2413, %r2412; + mov.b64 {%r2414, %r2415}, %rd453; + cvt.rn.bf16x2.f32 %r319, %r2415, %r2414; + mov.b64 {%r2416, %r2417}, %rd456; + cvt.rn.bf16x2.f32 %r320, %r2417, %r2416; + mov.b64 {%r2418, %r2419}, %rd459; + cvt.rn.bf16x2.f32 %r321, %r2419, %r2418; + mov.b64 {%r2420, %r2421}, %rd462; + cvt.rn.bf16x2.f32 %r322, %r2421, %r2420; + mov.b64 {%r2422, %r2423}, %rd465; + cvt.rn.bf16x2.f32 %r323, %r2423, %r2422; + mov.b64 {%r2424, %r2425}, %rd468; + cvt.rn.bf16x2.f32 %r324, %r2425, %r2424; + mov.b64 {%r2426, %r2427}, %rd471; + cvt.rn.bf16x2.f32 %r325, %r2427, %r2426; + mov.b64 {%r2428, %r2429}, %rd474; + cvt.rn.bf16x2.f32 %r326, %r2429, %r2428; + mov.b64 {%r2430, %r2431}, %rd477; + cvt.rn.bf16x2.f32 %r327, %r2431, %r2430; + mov.b64 {%r2432, %r2433}, %rd480; + cvt.rn.bf16x2.f32 %r328, %r2433, %r2432; + mov.b64 {%r2434, %r2435}, %rd483; + cvt.rn.bf16x2.f32 %r329, %r2435, %r2434; + mov.b64 {%r2436, %r2437}, %rd486; + cvt.rn.bf16x2.f32 %r330, %r2437, %r2436; + mov.b64 {%r2438, %r2439}, %rd489; + cvt.rn.bf16x2.f32 %r331, %r2439, %r2438; + mov.b64 {%r2440, %r2441}, %rd492; + cvt.rn.bf16x2.f32 %r332, %r2441, %r2440; + mov.b64 {%r2442, %r2443}, %rd495; + cvt.rn.bf16x2.f32 %r333, %r2443, %r2442; + mov.b64 {%r2444, %r2445}, %rd498; + cvt.rn.bf16x2.f32 %r334, %r2445, %r2444; + mov.b64 {%r2446, %r2447}, %rd501; + cvt.rn.bf16x2.f32 %r335, %r2447, %r2446; + mov.b64 {%r2448, %r2449}, %rd504; + cvt.rn.bf16x2.f32 %r336, %r2449, %r2448; + mov.b64 {%r2450, %r2451}, %rd507; + cvt.rn.bf16x2.f32 %r337, %r2451, %r2450; + mov.b64 {%r2452, %r2453}, %rd510; + cvt.rn.bf16x2.f32 %r338, %r2453, %r2452; + mov.b64 {%r2454, %r2455}, %rd513; + cvt.rn.bf16x2.f32 %r339, %r2455, %r2454; + mov.b64 {%r2456, %r2457}, %rd516; + cvt.rn.bf16x2.f32 %r340, %r2457, %r2456; + mov.b64 {%r2458, %r2459}, %rd519; + cvt.rn.bf16x2.f32 %r341, %r2459, %r2458; + mov.b64 {%r2460, %r2461}, %rd522; + cvt.rn.bf16x2.f32 %r342, %r2461, %r2460; + mov.b64 {%r2462, %r2463}, %rd525; + cvt.rn.bf16x2.f32 %r343, %r2463, %r2462; + mov.b64 {%r2464, %r2465}, %rd528; + cvt.rn.bf16x2.f32 %r344, %r2465, %r2464; + mov.b64 {%r2466, %r2467}, %rd531; + cvt.rn.bf16x2.f32 %r345, %r2467, %r2466; + mov.b64 {%r2468, %r2469}, %rd534; + cvt.rn.bf16x2.f32 %r346, %r2469, %r2468; + mov.b64 {%r2470, %r2471}, %rd537; + cvt.rn.bf16x2.f32 %r347, %r2471, %r2470; + mov.b64 {%r2472, %r2473}, %rd540; + cvt.rn.bf16x2.f32 %r348, %r2473, %r2472; + mov.b64 {%r2474, %r2475}, %rd543; + cvt.rn.bf16x2.f32 %r349, %r2475, %r2474; + mov.b64 {%r2476, %r2477}, %rd546; + cvt.rn.bf16x2.f32 %r350, %r2477, %r2476; + mov.b64 {%r2478, %r2479}, %rd549; + cvt.rn.bf16x2.f32 %r351, %r2479, %r2478; + mov.b64 {%r2480, %r2481}, %rd552; + cvt.rn.bf16x2.f32 %r352, %r2481, %r2480; + mov.b64 {%r2482, %r2483}, %rd555; + cvt.rn.bf16x2.f32 %r353, %r2483, %r2482; + mov.b64 {%r2484, %r2485}, %rd558; + cvt.rn.bf16x2.f32 %r354, %r2485, %r2484; + mov.b64 {%r2486, %r2487}, %rd561; + cvt.rn.bf16x2.f32 %r355, %r2487, %r2486; + mov.b64 {%r2488, %r2489}, %rd564; + cvt.rn.bf16x2.f32 %r356, %r2489, %r2488; + mov.b64 {%r2490, %r2491}, %rd567; + cvt.rn.bf16x2.f32 %r357, %r2491, %r2490; + mov.b64 {%r2492, %r2493}, %rd570; + cvt.rn.bf16x2.f32 %r358, %r2493, %r2492; + mov.b64 {%r2494, %r2495}, %rd573; + cvt.rn.bf16x2.f32 %r359, %r2495, %r2494; + mov.b64 {%r2496, %r2497}, %rd576; + cvt.rn.bf16x2.f32 %r360, %r2497, %r2496; + mov.b64 {%r2498, %r2499}, %rd579; + cvt.rn.bf16x2.f32 %r361, %r2499, %r2498; + mov.b64 {%r2500, %r2501}, %rd582; + cvt.rn.bf16x2.f32 %r362, %r2501, %r2500; + mov.b64 {%r2502, %r2503}, %rd585; + cvt.rn.bf16x2.f32 %r363, %r2503, %r2502; + mov.b64 {%r2504, %r2505}, %rd588; + cvt.rn.bf16x2.f32 %r364, %r2505, %r2504; + mov.b64 {%r2506, %r2507}, %rd591; + cvt.rn.bf16x2.f32 %r365, %r2507, %r2506; + mov.b64 {%r2508, %r2509}, %rd594; + cvt.rn.bf16x2.f32 %r366, %r2509, %r2508; + mov.b64 {%r2510, %r2511}, %rd597; + cvt.rn.bf16x2.f32 %r367, %r2511, %r2510; + mov.b64 {%r2512, %r2513}, %rd600; + cvt.rn.bf16x2.f32 %r368, %r2513, %r2512; + mov.b64 {%r2514, %r2515}, %rd603; + cvt.rn.bf16x2.f32 %r369, %r2515, %r2514; + mov.b64 {%r2516, %r2517}, %rd606; + cvt.rn.bf16x2.f32 %r370, %r2517, %r2516; + mov.b64 {%r2518, %r2519}, %rd609; + cvt.rn.bf16x2.f32 %r371, %r2519, %r2518; + mov.b64 {%r2520, %r2521}, %rd612; + cvt.rn.bf16x2.f32 %r372, %r2521, %r2520; + mov.b64 {%r2522, %r2523}, %rd615; + cvt.rn.bf16x2.f32 %r373, %r2523, %r2522; + mov.b64 {%r2524, %r2525}, %rd618; + cvt.rn.bf16x2.f32 %r374, %r2525, %r2524; + mov.b64 {%r2526, %r2527}, %rd621; + cvt.rn.bf16x2.f32 %r375, %r2527, %r2526; + mov.b64 {%r2528, %r2529}, %rd624; + cvt.rn.bf16x2.f32 %r376, %r2529, %r2528; + mov.b64 {%r2530, %r2531}, %rd627; + cvt.rn.bf16x2.f32 %r377, %r2531, %r2530; + mov.b64 {%r2532, %r2533}, %rd630; + cvt.rn.bf16x2.f32 %r378, %r2533, %r2532; + mov.b64 {%r2534, %r2535}, %rd633; + cvt.rn.bf16x2.f32 %r379, %r2535, %r2534; + mov.b64 {%r2536, %r2537}, %rd636; + cvt.rn.bf16x2.f32 %r380, %r2537, %r2536; + mov.b64 {%r2538, %r2539}, %rd639; + cvt.rn.bf16x2.f32 %r381, %r2539, %r2538; + mov.b64 {%r2540, %r2541}, %rd642; + cvt.rn.bf16x2.f32 %r382, %r2541, %r2540; + mov.b64 {%r2542, %r2543}, %rd645; + cvt.rn.bf16x2.f32 %r383, %r2543, %r2542; + mov.b64 {%r2544, %r2545}, %rd648; + cvt.rn.bf16x2.f32 %r384, %r2545, %r2544; + mov.b64 {%r2546, %r2547}, %rd651; + cvt.rn.bf16x2.f32 %r385, %r2547, %r2546; + mov.b64 {%r2548, %r2549}, %rd654; + cvt.rn.bf16x2.f32 %r386, %r2549, %r2548; + mov.b64 {%r2550, %r2551}, %rd657; + cvt.rn.bf16x2.f32 %r387, %r2551, %r2550; + mov.b64 {%r2552, %r2553}, %rd660; + cvt.rn.bf16x2.f32 %r388, %r2553, %r2552; + mov.b64 {%r2554, %r2555}, %rd663; + cvt.rn.bf16x2.f32 %r389, %r2555, %r2554; + .loc 1 108 19 // l2norm.py:108:19 + cvt.rn.bf16x2.f32 %r262, %r2287, %r2285; + cvt.rn.bf16x2.f32 %r263, %r2291, %r2289; + cvt.rn.bf16x2.f32 %r264, %r2295, %r2293; + cvt.rn.bf16x2.f32 %r265, %r2299, %r2297; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd67 + 0 ], { %r262, %r263, %r264, %r265 }; + // end inline asm + cvt.rn.bf16x2.f32 %r266, %r2303, %r2301; + cvt.rn.bf16x2.f32 %r267, %r2307, %r2305; + cvt.rn.bf16x2.f32 %r268, %r2311, %r2309; + cvt.rn.bf16x2.f32 %r269, %r2315, %r2313; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd68 + 0 ], { %r266, %r267, %r268, %r269 }; + // end inline asm + // begin inline asm + @%p6 st.global.v4.b32 [ %rd69 + 0 ], { %r270, %r271, %r272, %r273 }; + // end inline asm + // begin inline asm + @%p7 st.global.v4.b32 [ %rd70 + 0 ], { %r274, %r275, %r276, %r277 }; + // end inline asm + // begin inline asm + @%p8 st.global.v4.b32 [ %rd71 + 0 ], { %r278, %r279, %r280, %r281 }; + // end inline asm + // begin inline asm + @%p9 st.global.v4.b32 [ %rd72 + 0 ], { %r282, %r283, %r284, %r285 }; + // end inline asm + // begin inline asm + @%p10 st.global.v4.b32 [ %rd73 + 0 ], { %r286, %r287, %r288, %r289 }; + // end inline asm + // begin inline asm + @%p11 st.global.v4.b32 [ %rd74 + 0 ], { %r290, %r291, %r292, %r293 }; + // end inline asm + // begin inline asm + @%p12 st.global.v4.b32 [ %rd75 + 0 ], { %r294, %r295, %r296, %r297 }; + // end inline asm + // begin inline asm + @%p3 st.global.v4.b32 [ %rd76 + 0 ], { %r298, %r299, %r300, %r301 }; + // end inline asm + // begin inline asm + @%p13 st.global.v4.b32 [ %rd77 + 0 ], { %r302, %r303, %r304, %r305 }; + // end inline asm + // begin inline asm + @%p14 st.global.v4.b32 [ %rd78 + 0 ], { %r306, %r307, %r308, %r309 }; + // end inline asm + // begin inline asm + @%p15 st.global.v4.b32 [ %rd79 + 0 ], { %r310, %r311, %r312, %r313 }; + // end inline asm + // begin inline asm + @%p16 st.global.v4.b32 [ %rd80 + 0 ], { %r314, %r315, %r316, %r317 }; + // end inline asm + // begin inline asm + @%p17 st.global.v4.b32 [ %rd81 + 0 ], { %r318, %r319, %r320, %r321 }; + // end inline asm + // begin inline asm + @%p18 st.global.v4.b32 [ %rd82 + 0 ], { %r322, %r323, %r324, %r325 }; + // end inline asm + // begin inline asm + @%p19 st.global.v4.b32 [ %rd83 + 0 ], { %r326, %r327, %r328, %r329 }; + // end inline asm + // begin inline asm + @%p20 st.global.v4.b32 [ %rd84 + 0 ], { %r330, %r331, %r332, %r333 }; + // end inline asm + // begin inline asm + @%p21 st.global.v4.b32 [ %rd85 + 0 ], { %r334, %r335, %r336, %r337 }; + // end inline asm + // begin inline asm + @%p22 st.global.v4.b32 [ %rd86 + 0 ], { %r338, %r339, %r340, %r341 }; + // end inline asm + // begin inline asm + @%p23 st.global.v4.b32 [ %rd87 + 0 ], { %r342, %r343, %r344, %r345 }; + // end inline asm + // begin inline asm + @%p24 st.global.v4.b32 [ %rd88 + 0 ], { %r346, %r347, %r348, %r349 }; + // end inline asm + // begin inline asm + @%p25 st.global.v4.b32 [ %rd89 + 0 ], { %r350, %r351, %r352, %r353 }; + // end inline asm + // begin inline asm + @%p26 st.global.v4.b32 [ %rd90 + 0 ], { %r354, %r355, %r356, %r357 }; + // end inline asm + // begin inline asm + @%p27 st.global.v4.b32 [ %rd91 + 0 ], { %r358, %r359, %r360, %r361 }; + // end inline asm + // begin inline asm + @%p28 st.global.v4.b32 [ %rd92 + 0 ], { %r362, %r363, %r364, %r365 }; + // end inline asm + // begin inline asm + @%p29 st.global.v4.b32 [ %rd93 + 0 ], { %r366, %r367, %r368, %r369 }; + // end inline asm + // begin inline asm + @%p30 st.global.v4.b32 [ %rd94 + 0 ], { %r370, %r371, %r372, %r373 }; + // end inline asm + // begin inline asm + @%p31 st.global.v4.b32 [ %rd95 + 0 ], { %r374, %r375, %r376, %r377 }; + // end inline asm + // begin inline asm + @%p32 st.global.v4.b32 [ %rd96 + 0 ], { %r378, %r379, %r380, %r381 }; + // end inline asm + // begin inline asm + @%p33 st.global.v4.b32 [ %rd97 + 0 ], { %r382, %r383, %r384, %r385 }; + // end inline asm + // begin inline asm + @%p34 st.global.v4.b32 [ %rd98 + 0 ], { %r386, %r387, %r388, %r389 }; + // end inline asm + .loc 1 108 4 // l2norm.py:108:4 + ret; +$L__tmp259: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 98 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp258 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 107 // DW_AT_call_line +.b8 55 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp258 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.source b/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..35265513bc03aa2afd3347f28178370dc92e754e --- /dev/null +++ b/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.source @@ -0,0 +1,210 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc47 = loc(unknown) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc54 = loc("y"(#loc)) +#loc55 = loc("rstd"(#loc)) +#loc56 = loc("dy"(#loc)) +#loc57 = loc("dx"(#loc)) +#loc58 = loc("eps"(#loc)) +#loc59 = loc("T"(#loc)) +#loc101 = loc("input"(#loc45)) +#loc102 = loc("a"(#loc50)) +#loc103 = loc("b"(#loc50)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc60) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc61) + %o_t = arith.constant 128 : i32 loc(#loc62) + %o_t_1 = arith.constant 128 : i64 loc(#loc62) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc62) + %o_t_3 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc63) + %o_t_4 = arith.extsi %o_t_3 : tensor<128xi32> to tensor<128xi64> loc(#loc64) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<128xi64> loc(#loc64) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<128xi64> loc(#loc64) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc65) + %m_t = arith.extsi %T : i32 to i64 loc(#loc66) + %m_t_7 = tt.splat %m_t : i64 -> tensor<128xi64> loc(#loc66) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<128xi64> loc(#loc66) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<128xi1> -> tensor<128x1xi1> loc(#loc67) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc68) + %m_x_10 = arith.constant 128 : i32 loc(#loc69) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc69) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc69) + %m_x_13 = tt.broadcast %m_x : tensor<128x1xi1> -> tensor<128x128xi1> loc(#loc70) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<128x128xi1> loc(#loc70) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<128x128xi1> loc(#loc70) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc71) + %p_y_16 = arith.constant 128 : i32 loc(#loc72) + %p_y_17 = arith.constant 128 : i64 loc(#loc72) + %p_y_18 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc72) + %p_y_19 = arith.muli %p_y, %p_y_18 : tensor<128x1xi64> loc(#loc72) + %p_y_20 = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc73) + %p_y_21 = tt.addptr %p_y_20, %p_y_19 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc73) + %p_y_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc74) + %p_y_23 = tt.broadcast %p_y_21 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc75) + %p_y_24 = tt.broadcast %p_y_22 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc75) + %p_y_25 = tt.addptr %p_y_23, %p_y_24 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc75) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr> loc(#loc76) + %p_rstd_26 = tt.addptr %p_rstd, %o_t_6 : tensor<128x!tt.ptr>, tensor<128xi64> loc(#loc76) + %p_dy = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc77) + %p_dy_27 = arith.constant 128 : i32 loc(#loc78) + %p_dy_28 = arith.constant 128 : i64 loc(#loc78) + %p_dy_29 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc78) + %p_dy_30 = arith.muli %p_dy, %p_dy_29 : tensor<128x1xi64> loc(#loc78) + %p_dy_31 = tt.splat %dy : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc79) + %p_dy_32 = tt.addptr %p_dy_31, %p_dy_30 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc79) + %p_dy_33 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc80) + %p_dy_34 = tt.broadcast %p_dy_32 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc81) + %p_dy_35 = tt.broadcast %p_dy_33 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc81) + %p_dy_36 = tt.addptr %p_dy_34, %p_dy_35 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc81) + %p_dx = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc82) + %p_dx_37 = arith.constant 128 : i32 loc(#loc83) + %p_dx_38 = arith.constant 128 : i64 loc(#loc83) + %p_dx_39 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc83) + %p_dx_40 = arith.muli %p_dx, %p_dx_39 : tensor<128x1xi64> loc(#loc83) + %p_dx_41 = tt.splat %dx : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc84) + %p_dx_42 = tt.addptr %p_dx_41, %p_dx_40 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc84) + %p_dx_43 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc85) + %p_dx_44 = tt.broadcast %p_dx_42 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc86) + %p_dx_45 = tt.broadcast %p_dx_43 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc86) + %p_dx_46 = tt.addptr %p_dx_44, %p_dx_45 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc86) + %b_y = arith.constant 0.000000e+00 : f32 loc(#loc87) + %b_y_47 = arith.constant dense<0.000000e+00> : tensor<128x128xf32> loc(#loc87) + %b_y_48 = arith.truncf %b_y_47 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc87) + %b_y_49 = tt.load %p_y_25, %m_x_15, %b_y_48 : tensor<128x128x!tt.ptr> loc(#loc87) + %b_y_50 = arith.extf %b_y_49 : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc88) + %b_rstd = arith.constant 0.000000e+00 : f32 loc(#loc89) + %b_rstd_51 = arith.constant dense<0.000000e+00> : tensor<128xf32> loc(#loc89) + %b_rstd_52 = tt.load %p_rstd_26, %m_t_8, %b_rstd_51 : tensor<128x!tt.ptr> loc(#loc89) + %b_dy = arith.constant 0.000000e+00 : f32 loc(#loc90) + %b_dy_53 = arith.constant dense<0.000000e+00> : tensor<128x128xf32> loc(#loc90) + %b_dy_54 = arith.truncf %b_dy_53 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc90) + %b_dy_55 = tt.load %p_dy_36, %m_x_15, %b_dy_54 : tensor<128x128x!tt.ptr> loc(#loc90) + %b_dy_56 = arith.extf %b_dy_55 : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc91) + %b_dx = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc92) + %b_dx_57 = tt.broadcast %b_dx : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc93) + %b_dx_58 = arith.mulf %b_dy_56, %b_dx_57 : tensor<128x128xf32> loc(#loc93) + %b_dx_59 = arith.mulf %b_dy_56, %b_y_50 : tensor<128x128xf32> loc(#loc94) + %b_dx_60 = tt.call @triton.language.standard.sum__fp32S128_128S_c1_cFalse_cNone(%b_dx_59) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc95) + %b_dx_61 = tt.expand_dims %b_dx_60 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc96) + %b_dx_62 = tt.broadcast %b_dx_61 : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc97) + %b_dx_63 = arith.mulf %b_dx_62, %b_y_50 : tensor<128x128xf32> loc(#loc97) + %b_dx_64 = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc98) + %b_dx_65 = tt.broadcast %b_dx_64 : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc99) + %b_dx_66 = arith.mulf %b_dx_63, %b_dx_65 : tensor<128x128xf32> loc(#loc99) + %b_dx_67 = arith.subf %b_dx_58, %b_dx_66 : tensor<128x128xf32> loc(#loc100) + %0 = arith.truncf %b_dx_67 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc42) + tt.store %p_dx_46, %0, %m_x_15 : tensor<128x128x!tt.ptr> loc(#loc43) + tt.return loc(#loc44) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S128_128S_c1_cFalse_cNone(%input: tensor<128x128xf32> loc("input"(#loc45))) -> tensor<128xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc46) + tt.reduce.return %2 : f32 loc(#loc46) + }) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc46) + tt.return %0 : tensor<128xf32> loc(#loc48) + ^bb1: // no predecessors + %1 = ub.poison : tensor<128xf32> loc(#loc49) + tt.return %1 : tensor<128xf32> loc(#loc49) + } loc(#loc45) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc50)), %b: f32 loc("b"(#loc50))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc51) + tt.return %0 : f32 loc(#loc52) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc53) + tt.return %1 : f32 loc(#loc53) + } loc(#loc50) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:31) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:39) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:20) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:31) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:39) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:82) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc60 = loc("i_t"(#loc1)) +#loc61 = loc("i_t"(#loc2)) +#loc62 = loc("o_t"(#loc3)) +#loc63 = loc("o_t"(#loc4)) +#loc64 = loc("o_t"(#loc5)) +#loc65 = loc("o_d"(#loc6)) +#loc66 = loc("m_t"(#loc7)) +#loc67 = loc("m_x"(#loc8)) +#loc68 = loc("m_x"(#loc9)) +#loc69 = loc("m_x"(#loc10)) +#loc70 = loc("m_x"(#loc11)) +#loc71 = loc("p_y"(#loc12)) +#loc72 = loc("p_y"(#loc13)) +#loc73 = loc("p_y"(#loc14)) +#loc74 = loc("p_y"(#loc15)) +#loc75 = loc("p_y"(#loc16)) +#loc76 = loc("p_rstd"(#loc17)) +#loc77 = loc("p_dy"(#loc18)) +#loc78 = loc("p_dy"(#loc19)) +#loc79 = loc("p_dy"(#loc20)) +#loc80 = loc("p_dy"(#loc21)) +#loc81 = loc("p_dy"(#loc22)) +#loc82 = loc("p_dx"(#loc23)) +#loc83 = loc("p_dx"(#loc24)) +#loc84 = loc("p_dx"(#loc25)) +#loc85 = loc("p_dx"(#loc26)) +#loc86 = loc("p_dx"(#loc27)) +#loc87 = loc("b_y"(#loc28)) +#loc88 = loc("b_y"(#loc29)) +#loc89 = loc("b_rstd"(#loc30)) +#loc90 = loc("b_dy"(#loc31)) +#loc91 = loc("b_dy"(#loc32)) +#loc92 = loc("b_dx"(#loc33)) +#loc93 = loc("b_dx"(#loc34)) +#loc94 = loc("b_dx"(#loc35)) +#loc95 = loc("b_dx"(#loc36)) +#loc96 = loc("b_dx"(#loc37)) +#loc97 = loc("b_dx"(#loc38)) +#loc98 = loc("b_dx"(#loc39)) +#loc99 = loc("b_dx"(#loc40)) +#loc100 = loc("b_dx"(#loc41)) diff --git a/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.ttgir b/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..bfa2ef918972bb7aa416f60a993d58f4b42040eb --- /dev/null +++ b/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.ttgir @@ -0,0 +1,154 @@ +#blocked = #ttg.blocked<{sizePerThread = [2], threadsPerWarp = [32], warpsPerCTA = [2], order = [0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [2, 1], order = [1, 0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc39 = loc("y"(#loc)) +#loc40 = loc("rstd"(#loc)) +#loc41 = loc("dy"(#loc)) +#loc42 = loc("dx"(#loc)) +#loc43 = loc("eps"(#loc)) +#loc44 = loc("T"(#loc)) +#loc72 = loc("b_dx"(#loc30)) +#loc78 = loc(callsite(#loc1 at #loc72)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 2 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %c128_i64 = arith.constant 128 : i64 loc(#loc1) + %cst = arith.constant dense<0.000000e+00> : tensor<128xf32, #blocked> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<128x128xbf16, #blocked1> loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<128x1xi64, #blocked1> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<1x128xi32, #blocked1> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc45) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc46) + %o_t = arith.muli %i_t_3, %c128_i64 : i64 loc(#loc47) + %o_t_4 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc48) + %o_t_5 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #blocked> loc(#loc48) + %o_t_6 = arith.extsi %o_t_4 : tensor<128xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> to tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc49) + %o_t_7 = arith.extsi %o_t_5 : tensor<128xi32, #blocked> to tensor<128xi64, #blocked> loc(#loc49) + %o_t_8 = tt.splat %o_t : i64 -> tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc49) + %o_t_9 = tt.splat %o_t : i64 -> tensor<128xi64, #blocked> loc(#loc49) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc49) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<128xi64, #blocked> loc(#loc49) + %m_t = arith.extsi %T : i32 to i64 loc(#loc50) + %m_t_12 = tt.splat %m_t : i64 -> tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc50) + %m_t_13 = tt.splat %m_t : i64 -> tensor<128xi64, #blocked> loc(#loc50) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc50) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<128xi64, #blocked> loc(#loc50) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<128xi1, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<128x1xi1, #blocked1> loc(#loc51) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc52) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x128xi32, #blocked1> loc(#loc52) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_2 : tensor<1x128xi32, #blocked1> loc(#loc53) + %m_x_19 = tt.broadcast %m_x : tensor<128x1xi1, #blocked1> -> tensor<128x128xi1, #blocked1> loc(#loc54) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked1> -> tensor<128x128xi1, #blocked1> loc(#loc54) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<128x128xi1, #blocked1> loc(#loc54) + %p_y = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<128x1xi64, #blocked1> loc(#loc55) + %p_y_22 = arith.muli %p_y, %cst_1 : tensor<128x1xi64, #blocked1> loc(#loc56) + %p_y_23 = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr, #blocked1> loc(#loc57) + %p_y_24 = tt.addptr %p_y_23, %p_y_22 : tensor<128x1x!tt.ptr, #blocked1>, tensor<128x1xi64, #blocked1> loc(#loc57) + %p_y_25 = tt.broadcast %p_y_24 : tensor<128x1x!tt.ptr, #blocked1> -> tensor<128x128x!tt.ptr, #blocked1> loc(#loc58) + %p_y_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked1> -> tensor<128x128xi32, #blocked1> loc(#loc58) + %p_y_27 = tt.addptr %p_y_25, %p_y_26 : tensor<128x128x!tt.ptr, #blocked1>, tensor<128x128xi32, #blocked1> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr, #blocked> loc(#loc59) + %p_rstd_28 = tt.addptr %p_rstd, %o_t_11 : tensor<128x!tt.ptr, #blocked>, tensor<128xi64, #blocked> loc(#loc59) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<128x1x!tt.ptr, #blocked1> loc(#loc60) + %p_dy_29 = tt.addptr %p_dy, %p_y_22 : tensor<128x1x!tt.ptr, #blocked1>, tensor<128x1xi64, #blocked1> loc(#loc60) + %p_dy_30 = tt.broadcast %p_dy_29 : tensor<128x1x!tt.ptr, #blocked1> -> tensor<128x128x!tt.ptr, #blocked1> loc(#loc61) + %p_dy_31 = tt.addptr %p_dy_30, %p_y_26 : tensor<128x128x!tt.ptr, #blocked1>, tensor<128x128xi32, #blocked1> loc(#loc61) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<128x1x!tt.ptr, #blocked1> loc(#loc62) + %p_dx_32 = tt.addptr %p_dx, %p_y_22 : tensor<128x1x!tt.ptr, #blocked1>, tensor<128x1xi64, #blocked1> loc(#loc62) + %p_dx_33 = tt.broadcast %p_dx_32 : tensor<128x1x!tt.ptr, #blocked1> -> tensor<128x128x!tt.ptr, #blocked1> loc(#loc63) + %p_dx_34 = tt.addptr %p_dx_33, %p_y_26 : tensor<128x128x!tt.ptr, #blocked1>, tensor<128x128xi32, #blocked1> loc(#loc63) + %b_y = tt.load %p_y_27, %m_x_21, %cst_0 : tensor<128x128x!tt.ptr, #blocked1> loc(#loc64) + %b_y_35 = arith.extf %b_y : tensor<128x128xbf16, #blocked1> to tensor<128x128xf32, #blocked1> loc(#loc65) + %b_rstd = tt.load %p_rstd_28, %m_t_15, %cst : tensor<128x!tt.ptr, #blocked> loc(#loc66) + %b_dy = tt.load %p_dy_31, %m_x_21, %cst_0 : tensor<128x128x!tt.ptr, #blocked1> loc(#loc67) + %b_dy_36 = arith.extf %b_dy : tensor<128x128xbf16, #blocked1> to tensor<128x128xf32, #blocked1> loc(#loc68) + %b_dx = ttg.convert_layout %b_rstd : tensor<128xf32, #blocked> -> tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc69) + %b_dx_37 = tt.expand_dims %b_dx {axis = 1 : i32} : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<128x1xf32, #blocked1> loc(#loc69) + %b_dx_38 = tt.broadcast %b_dx_37 : tensor<128x1xf32, #blocked1> -> tensor<128x128xf32, #blocked1> loc(#loc70) + %b_dx_39 = arith.mulf %b_dy_36, %b_dx_38 : tensor<128x128xf32, #blocked1> loc(#loc70) + %b_dx_40 = arith.mulf %b_dy_36, %b_y_35 : tensor<128x128xf32, #blocked1> loc(#loc71) + %b_dx_41 = "tt.reduce"(%b_dx_40) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_47: f32 loc(callsite(#loc1 at #loc72)), %b_dx_48: f32 loc(callsite(#loc1 at #loc72))): + %b_dx_49 = arith.addf %b_dx_47, %b_dx_48 : f32 loc(#loc79) + tt.reduce.return %b_dx_49 : f32 loc(#loc77) + }) : (tensor<128x128xf32, #blocked1>) -> tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc77) + %b_dx_42 = tt.expand_dims %b_dx_41 {axis = 1 : i32} : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<128x1xf32, #blocked1> loc(#loc73) + %b_dx_43 = tt.broadcast %b_dx_42 : tensor<128x1xf32, #blocked1> -> tensor<128x128xf32, #blocked1> loc(#loc74) + %b_dx_44 = arith.mulf %b_dx_43, %b_y_35 : tensor<128x128xf32, #blocked1> loc(#loc74) + %b_dx_45 = arith.mulf %b_dx_44, %b_dx_38 : tensor<128x128xf32, #blocked1> loc(#loc75) + %b_dx_46 = arith.subf %b_dx_39, %b_dx_45 : tensor<128x128xf32, #blocked1> loc(#loc76) + %0 = arith.truncf %b_dx_46 : tensor<128x128xf32, #blocked1> to tensor<128x128xbf16, #blocked1> loc(#loc36) + tt.store %p_dx_34, %0, %m_x_21 : tensor<128x128x!tt.ptr, #blocked1> loc(#loc37) + tt.return loc(#loc38) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc45 = loc("i_t"(#loc2)) +#loc46 = loc("i_t"(#loc3)) +#loc47 = loc("o_t"(#loc4)) +#loc48 = loc("o_t"(#loc5)) +#loc49 = loc("o_t"(#loc6)) +#loc50 = loc("m_t"(#loc7)) +#loc51 = loc("m_x"(#loc8)) +#loc52 = loc("m_x"(#loc9)) +#loc53 = loc("m_x"(#loc10)) +#loc54 = loc("m_x"(#loc11)) +#loc55 = loc("p_y"(#loc12)) +#loc56 = loc("p_y"(#loc13)) +#loc57 = loc("p_y"(#loc14)) +#loc58 = loc("p_y"(#loc15)) +#loc59 = loc("p_rstd"(#loc16)) +#loc60 = loc("p_dy"(#loc17)) +#loc61 = loc("p_dy"(#loc18)) +#loc62 = loc("p_dx"(#loc19)) +#loc63 = loc("p_dx"(#loc20)) +#loc64 = loc("b_y"(#loc21)) +#loc65 = loc("b_y"(#loc22)) +#loc66 = loc("b_rstd"(#loc23)) +#loc67 = loc("b_dy"(#loc24)) +#loc68 = loc("b_dy"(#loc25)) +#loc69 = loc("b_dx"(#loc26)) +#loc70 = loc("b_dx"(#loc27)) +#loc71 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc32)) +#loc74 = loc("b_dx"(#loc33)) +#loc75 = loc("b_dx"(#loc34)) +#loc76 = loc("b_dx"(#loc35)) +#loc77 = loc(callsite(#loc29 at #loc72)) +#loc79 = loc(callsite(#loc31 at #loc77)) diff --git a/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.ttir b/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..f242766466ced3ec80bd72ef6238b6accc66dbe7 --- /dev/null +++ b/triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.ttir @@ -0,0 +1,144 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc39 = loc("y"(#loc)) +#loc40 = loc("rstd"(#loc)) +#loc41 = loc("dy"(#loc)) +#loc42 = loc("dx"(#loc)) +#loc43 = loc("eps"(#loc)) +#loc44 = loc("T"(#loc)) +#loc72 = loc("b_dx"(#loc30)) +#loc78 = loc(callsite(#loc1 at #loc72)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<0.000000e+00> : tensor<128x128xbf16> loc(#loc1) + %b_rstd = arith.constant dense<0.000000e+00> : tensor<128xf32> loc(#loc45) + %cst_0 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc1) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc46) + %c128_i64 = arith.constant 128 : i64 loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc47) + %i_t_1 = arith.extsi %i_t : i32 to i64 loc(#loc48) + %o_t = arith.muli %i_t_1, %c128_i64 : i64 loc(#loc49) + %o_t_2 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc50) + %o_t_3 = arith.extsi %o_t_2 : tensor<128xi32> to tensor<128xi64> loc(#loc51) + %o_t_4 = tt.splat %o_t : i64 -> tensor<128xi64> loc(#loc51) + %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<128xi64> loc(#loc51) + %m_t = arith.extsi %T : i32 to i64 loc(#loc52) + %m_t_6 = tt.splat %m_t : i64 -> tensor<128xi64> loc(#loc52) + %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<128xi64> loc(#loc52) + %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<128xi1> -> tensor<128x1xi1> loc(#loc53) + %m_x_9 = tt.expand_dims %o_t_2 {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc54) + %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc46) + %m_x_11 = tt.broadcast %m_x_8 : tensor<128x1xi1> -> tensor<128x128xi1> loc(#loc55) + %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<128x128xi1> loc(#loc55) + %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<128x128xi1> loc(#loc55) + %p_y = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc56) + %p_y_14 = arith.muli %p_y, %cst_0 : tensor<128x1xi64> loc(#loc57) + %p_y_15 = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc58) + %p_y_16 = tt.addptr %p_y_15, %p_y_14 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc58) + %p_y_17 = tt.broadcast %p_y_16 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc59) + %p_y_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc59) + %p_y_19 = tt.addptr %p_y_17, %p_y_18 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc59) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr> loc(#loc60) + %p_rstd_20 = tt.addptr %p_rstd, %o_t_5 : tensor<128x!tt.ptr>, tensor<128xi64> loc(#loc60) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc61) + %p_dy_21 = tt.addptr %p_dy, %p_y_14 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc61) + %p_dy_22 = tt.broadcast %p_dy_21 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc62) + %p_dy_23 = tt.addptr %p_dy_22, %p_y_18 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc62) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc63) + %p_dx_24 = tt.addptr %p_dx, %p_y_14 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc63) + %p_dx_25 = tt.broadcast %p_dx_24 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc64) + %p_dx_26 = tt.addptr %p_dx_25, %p_y_18 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc64) + %b_y = tt.load %p_y_19, %m_x_13, %cst : tensor<128x128x!tt.ptr> loc(#loc65) + %b_y_27 = arith.extf %b_y : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc66) + %b_rstd_28 = tt.load %p_rstd_20, %m_t_7, %b_rstd : tensor<128x!tt.ptr> loc(#loc45) + %b_dy = tt.load %p_dy_23, %m_x_13, %cst : tensor<128x128x!tt.ptr> loc(#loc67) + %b_dy_29 = arith.extf %b_dy : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc68) + %b_dx = tt.expand_dims %b_rstd_28 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc69) + %b_dx_30 = tt.broadcast %b_dx : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc70) + %b_dx_31 = arith.mulf %b_dy_29, %b_dx_30 : tensor<128x128xf32> loc(#loc70) + %b_dx_32 = arith.mulf %b_dy_29, %b_y_27 : tensor<128x128xf32> loc(#loc71) + %b_dx_33 = "tt.reduce"(%b_dx_32) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_39: f32 loc(callsite(#loc1 at #loc72)), %b_dx_40: f32 loc(callsite(#loc1 at #loc72))): + %b_dx_41 = arith.addf %b_dx_39, %b_dx_40 : f32 loc(#loc79) + tt.reduce.return %b_dx_41 : f32 loc(#loc77) + }) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc77) + %b_dx_34 = tt.expand_dims %b_dx_33 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc73) + %b_dx_35 = tt.broadcast %b_dx_34 : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc74) + %b_dx_36 = arith.mulf %b_dx_35, %b_y_27 : tensor<128x128xf32> loc(#loc74) + %b_dx_37 = arith.mulf %b_dx_36, %b_dx_30 : tensor<128x128xf32> loc(#loc75) + %b_dx_38 = arith.subf %b_dx_31, %b_dx_37 : tensor<128x128xf32> loc(#loc76) + %0 = arith.truncf %b_dx_38 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc36) + tt.store %p_dx_26, %0, %m_x_13 : tensor<128x128x!tt.ptr> loc(#loc37) + tt.return loc(#loc38) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc45 = loc("b_rstd"(#loc2)) +#loc46 = loc("m_x"(#loc3)) +#loc47 = loc("i_t"(#loc4)) +#loc48 = loc("i_t"(#loc5)) +#loc49 = loc("o_t"(#loc6)) +#loc50 = loc("o_t"(#loc7)) +#loc51 = loc("o_t"(#loc8)) +#loc52 = loc("m_t"(#loc9)) +#loc53 = loc("m_x"(#loc10)) +#loc54 = loc("m_x"(#loc11)) +#loc55 = loc("m_x"(#loc12)) +#loc56 = loc("p_y"(#loc13)) +#loc57 = loc("p_y"(#loc14)) +#loc58 = loc("p_y"(#loc15)) +#loc59 = loc("p_y"(#loc16)) +#loc60 = loc("p_rstd"(#loc17)) +#loc61 = loc("p_dy"(#loc18)) +#loc62 = loc("p_dy"(#loc19)) +#loc63 = loc("p_dx"(#loc20)) +#loc64 = loc("p_dx"(#loc21)) +#loc65 = loc("b_y"(#loc22)) +#loc66 = loc("b_y"(#loc23)) +#loc67 = loc("b_dy"(#loc24)) +#loc68 = loc("b_dy"(#loc25)) +#loc69 = loc("b_dx"(#loc26)) +#loc70 = loc("b_dx"(#loc27)) +#loc71 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc32)) +#loc74 = loc("b_dx"(#loc33)) +#loc75 = loc("b_dx"(#loc34)) +#loc76 = loc("b_dx"(#loc35)) +#loc77 = loc(callsite(#loc29 at #loc72)) +#loc79 = loc(callsite(#loc31 at #loc77)) diff --git a/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/__grp__l2norm_fwd_kernel.json b/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/__grp__l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..0f6fc35d54d0921fc4d5a20e9bd5d76c4b13f115 --- /dev/null +++ b/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/__grp__l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_fwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.source", "l2norm_fwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.ttir", "l2norm_fwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.ttgir", "l2norm_fwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.llir", "l2norm_fwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.ptx", "l2norm_fwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.cubin", "l2norm_fwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.cubin b/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..df770b40e5433c2f720e7380afd037f43acfba22 Binary files /dev/null and b/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.cubin differ diff --git a/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.json b/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..f0f8f5541c59da575caab00084282ff0beefa917 --- /dev/null +++ b/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"hash": "061ad091fa17036a05e3d16897b59532e2860bc3301b331f6264abbfc8d5a412", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 8, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 32, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_fwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.llir b/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..27d00ac4ca681e0bc167613b3a65471d04884850 --- /dev/null +++ b/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.llir @@ -0,0 +1,219 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 +@.str.2 = private unnamed_addr constant [17 x i8] c"__CUDA_PREC_SQRT\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_fwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, float %3, i32 %4, ptr addrspace(1) readnone captures(none) %5, ptr addrspace(1) readnone captures(none) %6) local_unnamed_addr #0 !dbg !5 { + %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %9 = zext nneg i32 %8 to i64, !dbg !14 + %10 = shl nuw nsw i64 %9, 3, !dbg !15 + %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !16 + %12 = and i32 %11, 224, !dbg !16 + %13 = lshr exact i32 %12, 5, !dbg !16 + %14 = zext nneg i32 %13 to i64, !dbg !17 + %15 = or disjoint i64 %10, %14, !dbg !17 + %16 = sext i32 %4 to i64, !dbg !18 + %17 = icmp slt i64 %15, %16, !dbg !18 + %18 = shl nuw nsw i32 %11, 2, !dbg !19 + %19 = and i32 %18, 124, !dbg !19 + %20 = shl nuw nsw i64 %15, 7, !dbg !20 + %21 = getelementptr bfloat, ptr addrspace(1) %0, i64 %20, !dbg !21 + %22 = zext nneg i32 %19 to i64, !dbg !22 + %23 = getelementptr bfloat, ptr addrspace(1) %21, i64 %22, !dbg !22 + %24 = tail call { i32, i32 } asm sideeffect "mov.u32 $0, $2;\0A\09mov.u32 $1, $3;\0A\09@$5 ld.global.v2.b32 { $0, $1 }, [ $4 + 0 ];", "=r,=r,r,r,l,b"(i32 0, i32 0, ptr addrspace(1) %23, i1 %17) #6, !dbg !23 + %25 = extractvalue { i32, i32 } %24, 0, !dbg !23 + %26 = bitcast i32 %25 to <2 x bfloat>, !dbg !23 + %27 = extractvalue { i32, i32 } %24, 1, !dbg !23 + %28 = bitcast i32 %27 to <2 x bfloat>, !dbg !23 + %29 = extractelement <2 x bfloat> %26, i64 0, !dbg !23 + %30 = extractelement <2 x bfloat> %26, i64 1, !dbg !23 + %31 = extractelement <2 x bfloat> %28, i64 0, !dbg !23 + %32 = extractelement <2 x bfloat> %28, i64 1, !dbg !23 + %33 = fpext bfloat %29 to float, !dbg !24 + %34 = fpext bfloat %30 to float, !dbg !24 + %35 = fpext bfloat %31 to float, !dbg !24 + %36 = fpext bfloat %32 to float, !dbg !24 + %37 = fmul float %33, %33, !dbg !25 + %38 = fmul float %34, %34, !dbg !25 + %39 = fmul float %35, %35, !dbg !25 + %40 = fmul float %36, %36, !dbg !25 + %41 = fadd float %37, %38, !dbg !26 + %42 = fadd float %39, %41, !dbg !26 + %43 = fadd float %40, %42, !dbg !26 + %44 = bitcast float %43 to i32, !dbg !29 + %45 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %44, i32 16, i32 31), !dbg !29 + %46 = bitcast i32 %45 to float, !dbg !29 + %47 = fadd float %43, %46, !dbg !26 + %48 = bitcast float %47 to i32, !dbg !29 + %49 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %48, i32 8, i32 31), !dbg !29 + %50 = bitcast i32 %49 to float, !dbg !29 + %51 = fadd float %47, %50, !dbg !26 + %52 = bitcast float %51 to i32, !dbg !29 + %53 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %52, i32 4, i32 31), !dbg !29 + %54 = bitcast i32 %53 to float, !dbg !29 + %55 = fadd float %51, %54, !dbg !26 + %56 = bitcast float %55 to i32, !dbg !29 + %57 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %56, i32 2, i32 31), !dbg !29 + %58 = bitcast i32 %57 to float, !dbg !29 + %59 = fadd float %55, %58, !dbg !26 + %60 = bitcast float %59 to i32, !dbg !29 + %61 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %60, i32 1, i32 31), !dbg !29 + %62 = bitcast i32 %61 to float, !dbg !29 + %63 = fadd float %59, %62, !dbg !26 + %64 = fadd float %3, %63, !dbg !32 + %65 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i = icmp eq i32 %65, 0, !dbg !33 + %66 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i = icmp eq i32 %66, 0, !dbg !33 + br i1 %.not.i, label %72, label %67, !dbg !33 + +67: ; preds = %7 + br i1 %.not1.i, label %70, label %68, !dbg !33 + +68: ; preds = %67 + %69 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %64) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +70: ; preds = %67 + %71 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %64) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +72: ; preds = %7 + br i1 %.not1.i, label %75, label %73, !dbg !33 + +73: ; preds = %72 + %74 = tail call float @llvm.nvvm.sqrt.rn.f(float %64) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +75: ; preds = %72 + %76 = tail call float @llvm.nvvm.sqrt.approx.f(float %64) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +__nv_sqrtf.exit: ; preds = %68, %70, %73, %75 + %.0.i = phi float [ %69, %68 ], [ %71, %70 ], [ %74, %73 ], [ %76, %75 ], !dbg !33 + %77 = and i32 %11, 7, !dbg !16 + %78 = zext nneg i32 %77 to i64, !dbg !17 + %79 = or disjoint i64 %10, %78, !dbg !17 + %80 = getelementptr float, ptr addrspace(1) %2, i64 %79, !dbg !34 + %81 = getelementptr bfloat, ptr addrspace(1) %1, i64 %20, !dbg !35 + %82 = getelementptr bfloat, ptr addrspace(1) %81, i64 %22, !dbg !36 + %83 = icmp slt i64 %79, %16, !dbg !18 + %84 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i), !dbg !37 + %85 = fmul float %84, %33, !dbg !38 + %86 = fmul float %84, %34, !dbg !38 + %87 = fmul float %84, %35, !dbg !38 + %88 = fmul float %84, %36, !dbg !38 + %89 = fptrunc float %85 to bfloat, !dbg !39 + %90 = fptrunc float %86 to bfloat, !dbg !39 + %91 = fptrunc float %87 to bfloat, !dbg !39 + %92 = fptrunc float %88 to bfloat, !dbg !39 + %93 = insertelement <2 x bfloat> poison, bfloat %89, i64 0, !dbg !40 + %94 = insertelement <2 x bfloat> %93, bfloat %90, i64 1, !dbg !40 + %95 = bitcast <2 x bfloat> %94 to i32, !dbg !40 + %96 = insertelement <2 x bfloat> poison, bfloat %91, i64 0, !dbg !40 + %97 = insertelement <2 x bfloat> %96, bfloat %92, i64 1, !dbg !40 + %98 = bitcast <2 x bfloat> %97 to i32, !dbg !40 + tail call void asm sideeffect "@$3 st.global.v2.b32 [ $2 + 0 ], { $0, $1 };", "r,r,l,b"(i32 %95, i32 %98, ptr addrspace(1) %82, i1 %17) #6, !dbg !40 + %99 = lshr exact i32 %12, 3, !dbg !41 + %100 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %99, !dbg !41 + %101 = bitcast float %84 to <1 x i32>, !dbg !41 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) %100, <1 x i32> %101) #6, !dbg !41 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !41 + %102 = shl nuw nsw i32 %77, 2, !dbg !41 + %103 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %102, !dbg !41 + %104 = load i32, ptr addrspace(3) %103, align 4, !dbg !41 + %105 = and i32 %11, 248, !dbg !41 + %106 = icmp eq i32 %105, 0, !dbg !41 + %107 = and i1 %106, %83, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %104, ptr addrspace(1) %80, i1 %107) #6, !dbg !41 + ret void, !dbg !42 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.div.full(float, float) #3 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #4 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #5 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.f(float) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="256" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) } +attributes #4 = { convergent nocallback nounwind } +attributes #5 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #6 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "l2norm_fwd_kernel", linkageName: "l2norm_fwd_kernel", scope: !1, file: !1, line: 69, type: !6, scopeLine: 69, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !10, !11, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 70, column: 24, scope: !5) +!14 = !DILocation(line: 70, column: 30, scope: !5) +!15 = !DILocation(line: 71, column: 16, scope: !5) +!16 = !DILocation(line: 71, column: 34, scope: !5) +!17 = !DILocation(line: 71, column: 21, scope: !5) +!18 = !DILocation(line: 73, column: 16, scope: !5) +!19 = !DILocation(line: 74, column: 30, scope: !5) +!20 = !DILocation(line: 75, column: 29, scope: !5) +!21 = !DILocation(line: 75, column: 14, scope: !5) +!22 = !DILocation(line: 75, column: 33, scope: !5) +!23 = !DILocation(line: 79, column: 18, scope: !5) +!24 = !DILocation(line: 79, column: 47, scope: !5) +!25 = !DILocation(line: 80, column: 38, scope: !5) +!26 = !DILocation(line: 263, column: 15, scope: !27, inlinedAt: !29) +!27 = distinct !DILexicalBlockFile(scope: !5, file: !28, discriminator: 0) +!28 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!29 = !DILocation(line: 293, column: 36, scope: !27, inlinedAt: !30) +!30 = !DILocation(line: 80, column: 43, scope: !31) +!31 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!32 = !DILocation(line: 80, column: 48, scope: !5) +!33 = !DILocation(line: 80, column: 25, scope: !5) +!34 = !DILocation(line: 77, column: 20, scope: !5) +!35 = !DILocation(line: 76, column: 14, scope: !5) +!36 = !DILocation(line: 76, column: 33, scope: !5) +!37 = !DILocation(line: 80, column: 17, scope: !5) +!38 = !DILocation(line: 81, column: 16, scope: !5) +!39 = !DILocation(line: 83, column: 25, scope: !5) +!40 = !DILocation(line: 83, column: 18, scope: !5) +!41 = !DILocation(line: 84, column: 21, scope: !5) +!42 = !DILocation(line: 84, column: 4, scope: !5) diff --git a/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.ptx b/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..c16ed71e04eef53fd16c76a87317b64d00c6dc4f --- /dev/null +++ b/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.ptx @@ -0,0 +1,387 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_fwd_kernel // -- Begin function l2norm_fwd_kernel +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; +.global .align 1 .b8 _$_str_$_2[17] = {95, 95, 67, 85, 68, 65, 95, 80, 82, 69, 67, 95, 83, 81, 82, 84}; + // @l2norm_fwd_kernel +.visible .entry l2norm_fwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_2, + .param .f32 l2norm_fwd_kernel_param_3, + .param .u32 l2norm_fwd_kernel_param_4, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_6 +) +.reqntid 256 +{ + .reg .pred %p<5>; + .reg .b16 %rs<5>; + .reg .b32 %r<47>; + .reg .b64 %rd<18>; + .loc 1 69 0 // l2norm.py:69:0 +$L__func_begin0: + .loc 1 69 0 // l2norm.py:69:0 + +// %bb.0: // %__nv_sqrtf.exit + ld.param.b64 %rd4, [l2norm_fwd_kernel_param_0]; + ld.param.b64 %rd5, [l2norm_fwd_kernel_param_1]; +$L__tmp0: + .loc 1 70 24 // l2norm.py:70:24 + mov.u32 %r9, %ctaid.x; + ld.param.b64 %rd6, [l2norm_fwd_kernel_param_2]; + .loc 1 71 16 // l2norm.py:71:16 + mul.wide.u32 %rd7, %r9, 8; + ld.param.b32 %r10, [l2norm_fwd_kernel_param_3]; + ld.param.s32 %rd8, [l2norm_fwd_kernel_param_4]; + .loc 1 71 34 // l2norm.py:71:34 + mov.u32 %r11, %tid.x; + and.b32 %r12, %r11, 224; + bfe.u32 %r13, %r11, 5, 3; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd9, %r13; + or.b64 %rd10, %rd7, %rd9; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p1, %rd10, %rd8; + .loc 1 74 30 // l2norm.py:74:30 + shl.b32 %r14, %r11, 2; + and.b32 %r15, %r14, 124; + .loc 1 75 14 // l2norm.py:75:14 + shl.b64 %rd11, %rd10, 8; + add.s64 %rd12, %rd4, %rd11; + .loc 1 75 33 // l2norm.py:75:33 + mul.wide.u32 %rd13, %r15, 2; + add.s64 %rd1, %rd12, %rd13; + mov.b32 %r3, 0; + .loc 1 79 18 // l2norm.py:79:18 + // begin inline asm + mov.u32 %r1, %r3; + mov.u32 %r2, %r3; + @%p1 ld.global.v2.b32 { %r1, %r2 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + .loc 1 79 47 // l2norm.py:79:47 + cvt.f32.bf16 %r16, %rs1; + cvt.f32.bf16 %r17, %rs2; + cvt.f32.bf16 %r18, %rs3; + cvt.f32.bf16 %r19, %rs4; + .loc 1 80 38 // l2norm.py:80:38 + mul.f32 %r20, %r17, %r17; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + fma.rn.f32.bf16 %r21, %rs1, %rs1, %r20; + fma.rn.f32.bf16 %r22, %rs3, %rs3, %r21; + fma.rn.f32.bf16 %r23, %rs4, %rs4, %r22; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r24, %r23, 16, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r25, %r23, %r24; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r26, %r25, 8, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r27, %r25, %r26; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r28, %r27, 4, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r29, %r27, %r28; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r30, %r29, 2, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r31, %r29, %r30; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r32, %r31, 1, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r33, %r31, %r32; +$L__tmp12: + .loc 1 80 48 // l2norm.py:80:48 + add.f32 %r34, %r10, %r33; + .loc 1 80 25 // l2norm.py:80:25 + sqrt.approx.ftz.f32 %r35, %r34; + .loc 1 71 34 // l2norm.py:71:34 + and.b32 %r36, %r11, 7; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd14, %r36; + or.b64 %rd15, %rd7, %rd14; + .loc 1 77 20 // l2norm.py:77:20 + shl.b64 %rd16, %rd15, 2; + add.s64 %rd3, %rd6, %rd16; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd17, %rd5, %rd11; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd2, %rd17, %rd13; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p3, %rd15, %rd8; + mov.b32 %r37, 0f3F800000; + .loc 1 80 17 // l2norm.py:80:17 + div.full.f32 %r7, %r37, %r35; + .loc 1 81 16 // l2norm.py:81:16 + mul.f32 %r38, %r7, %r16; + mul.f32 %r39, %r7, %r17; + mul.f32 %r40, %r7, %r18; + mul.f32 %r41, %r7, %r19; + .loc 1 83 18 // l2norm.py:83:18 + cvt.rn.bf16x2.f32 %r4, %r39, %r38; + cvt.rn.bf16x2.f32 %r5, %r41, %r40; + // begin inline asm + @%p1 st.global.v2.b32 [ %rd2 + 0 ], { %r4, %r5 }; + // end inline asm + .loc 1 84 21 // l2norm.py:84:21 + shr.u32 %r42, %r12, 3; + mov.b32 %r43, global_smem; + add.s32 %r6, %r43, %r42; + // begin inline asm + st.shared.b32 [ %r6 + 0 ], %r7; + // end inline asm + bar.sync 0; + shl.b32 %r44, %r36, 2; + add.s32 %r45, %r43, %r44; + ld.shared.b32 %r8, [%r45]; + and.b32 %r46, %r11, 248; + setp.eq.b32 %p4, %r46, 0; + and.pred %p2, %p4, %p3; + // begin inline asm + @%p2 st.global.b32 [ %rd3 + 0 ], { %r8 }; + // end inline asm + .loc 1 84 4 // l2norm.py:84:4 + ret; +$L__tmp13: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp12 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 80 // DW_AT_call_line +.b8 43 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp12 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.source b/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..c7c5a735c81381ccb0efd0d2a8f8d86240068e11 --- /dev/null +++ b/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.source @@ -0,0 +1,172 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc38 = loc(unknown) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc45 = loc("x"(#loc)) +#loc46 = loc("y"(#loc)) +#loc47 = loc("rstd"(#loc)) +#loc48 = loc("eps"(#loc)) +#loc49 = loc("T"(#loc)) +#loc81 = loc("input"(#loc36)) +#loc82 = loc("a"(#loc41)) +#loc83 = loc("b"(#loc41)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc50) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc51) + %o_t = arith.constant 8 : i32 loc(#loc52) + %o_t_1 = arith.constant 8 : i64 loc(#loc52) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc52) + %o_t_3 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32> loc(#loc53) + %o_t_4 = arith.extsi %o_t_3 : tensor<8xi32> to tensor<8xi64> loc(#loc54) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<8xi64> loc(#loc54) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<8xi64> loc(#loc54) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc55) + %m_t = arith.extsi %T : i32 to i64 loc(#loc56) + %m_t_7 = tt.splat %m_t : i64 -> tensor<8xi64> loc(#loc56) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<8xi64> loc(#loc56) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<8xi1> -> tensor<8x1xi1> loc(#loc57) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc58) + %m_x_10 = arith.constant 128 : i32 loc(#loc59) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc59) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc59) + %m_x_13 = tt.broadcast %m_x : tensor<8x1xi1> -> tensor<8x128xi1> loc(#loc60) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<8x128xi1> loc(#loc60) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<8x128xi1> loc(#loc60) + %p_x = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc61) + %p_x_16 = arith.constant 128 : i32 loc(#loc62) + %p_x_17 = arith.constant 128 : i64 loc(#loc62) + %p_x_18 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc62) + %p_x_19 = arith.muli %p_x, %p_x_18 : tensor<8x1xi64> loc(#loc62) + %p_x_20 = tt.splat %x : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc63) + %p_x_21 = tt.addptr %p_x_20, %p_x_19 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc63) + %p_x_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc64) + %p_x_23 = tt.broadcast %p_x_21 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc65) + %p_x_24 = tt.broadcast %p_x_22 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc65) + %p_x_25 = tt.addptr %p_x_23, %p_x_24 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc65) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc66) + %p_y_26 = arith.constant 128 : i32 loc(#loc67) + %p_y_27 = arith.constant 128 : i64 loc(#loc67) + %p_y_28 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc67) + %p_y_29 = arith.muli %p_y, %p_y_28 : tensor<8x1xi64> loc(#loc67) + %p_y_30 = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc68) + %p_y_31 = tt.addptr %p_y_30, %p_y_29 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc68) + %p_y_32 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc69) + %p_y_33 = tt.broadcast %p_y_31 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc70) + %p_y_34 = tt.broadcast %p_y_32 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc70) + %p_y_35 = tt.addptr %p_y_33, %p_y_34 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc70) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr> loc(#loc71) + %p_rstd_36 = tt.addptr %p_rstd, %o_t_6 : tensor<8x!tt.ptr>, tensor<8xi64> loc(#loc71) + %b_x = arith.constant 0.000000e+00 : f32 loc(#loc72) + %b_x_37 = arith.constant dense<0.000000e+00> : tensor<8x128xf32> loc(#loc72) + %b_x_38 = arith.truncf %b_x_37 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc72) + %b_x_39 = tt.load %p_x_25, %m_x_15, %b_x_38 : tensor<8x128x!tt.ptr> loc(#loc72) + %b_x_40 = arith.extf %b_x_39 : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc73) + %b_rstd = arith.mulf %b_x_40, %b_x_40 : tensor<8x128xf32> loc(#loc74) + %b_rstd_41 = tt.call @triton.language.standard.sum__fp32S8_128S_c1_cFalse_cNone(%b_rstd) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc75) + %b_rstd_42 = tt.splat %eps : f32 -> tensor<8xf32> loc(#loc76) + %b_rstd_43 = arith.addf %b_rstd_41, %b_rstd_42 : tensor<8xf32> loc(#loc76) + %b_rstd_44 = math.sqrt %b_rstd_43 : tensor<8xf32> loc(#loc77) + %b_rstd_45 = arith.constant 1 : i32 loc(#loc78) + %b_rstd_46 = arith.constant 1.000000e+00 : f32 loc(#loc78) + %b_rstd_47 = arith.constant dense<1.000000e+00> : tensor<8xf32> loc(#loc78) + %b_rstd_48 = arith.divf %b_rstd_47, %b_rstd_44 : tensor<8xf32> loc(#loc78) + %b_y = tt.expand_dims %b_rstd_48 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc79) + %b_y_49 = tt.broadcast %b_y : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc80) + %b_y_50 = arith.mulf %b_x_40, %b_y_49 : tensor<8x128xf32> loc(#loc80) + %0 = arith.truncf %b_y_50 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc32) + tt.store %p_y_35, %0, %m_x_15 : tensor<8x128x!tt.ptr> loc(#loc33) + tt.store %p_rstd_36, %b_rstd_48, %m_t_8 : tensor<8x!tt.ptr> loc(#loc34) + tt.return loc(#loc35) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S8_128S_c1_cFalse_cNone(%input: tensor<8x128xf32> loc("input"(#loc36))) -> tensor<8xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc37) + tt.reduce.return %2 : f32 loc(#loc37) + }) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc37) + tt.return %0 : tensor<8xf32> loc(#loc39) + ^bb1: // no predecessors + %1 = ub.poison : tensor<8xf32> loc(#loc40) + tt.return %1 : tensor<8xf32> loc(#loc40) + } loc(#loc36) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc41)), %b: f32 loc("b"(#loc41))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc42) + tt.return %0 : f32 loc(#loc43) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc44) + tt.return %1 : f32 loc(#loc44) + } loc(#loc41) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:18) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:29) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:37) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc50 = loc("i_t"(#loc1)) +#loc51 = loc("i_t"(#loc2)) +#loc52 = loc("o_t"(#loc3)) +#loc53 = loc("o_t"(#loc4)) +#loc54 = loc("o_t"(#loc5)) +#loc55 = loc("o_d"(#loc6)) +#loc56 = loc("m_t"(#loc7)) +#loc57 = loc("m_x"(#loc8)) +#loc58 = loc("m_x"(#loc9)) +#loc59 = loc("m_x"(#loc10)) +#loc60 = loc("m_x"(#loc11)) +#loc61 = loc("p_x"(#loc12)) +#loc62 = loc("p_x"(#loc13)) +#loc63 = loc("p_x"(#loc14)) +#loc64 = loc("p_x"(#loc15)) +#loc65 = loc("p_x"(#loc16)) +#loc66 = loc("p_y"(#loc17)) +#loc67 = loc("p_y"(#loc18)) +#loc68 = loc("p_y"(#loc19)) +#loc69 = loc("p_y"(#loc20)) +#loc70 = loc("p_y"(#loc21)) +#loc71 = loc("p_rstd"(#loc22)) +#loc72 = loc("b_x"(#loc23)) +#loc73 = loc("b_x"(#loc24)) +#loc74 = loc("b_rstd"(#loc25)) +#loc75 = loc("b_rstd"(#loc26)) +#loc76 = loc("b_rstd"(#loc27)) +#loc77 = loc("b_rstd"(#loc28)) +#loc78 = loc("b_rstd"(#loc29)) +#loc79 = loc("b_y"(#loc30)) +#loc80 = loc("b_y"(#loc31)) diff --git a/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.ttgir b/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..99e651af993cad92497ad25ea6ed3004b53b5810 --- /dev/null +++ b/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.ttgir @@ -0,0 +1,135 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 4], threadsPerWarp = [1, 32], warpsPerCTA = [8, 1], order = [1, 0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [8], order = [0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc1 = loc(unknown) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc34 = loc("x"(#loc)) +#loc35 = loc("y"(#loc)) +#loc36 = loc("rstd"(#loc)) +#loc37 = loc("eps"(#loc)) +#loc38 = loc("T"(#loc)) +#loc59 = loc("b_rstd"(#loc23)) +#loc66 = loc(callsite(#loc1 at #loc59)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 8 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<1.000000e+00> : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<8x128xbf16, #blocked> loc(#loc1) + %c8_i64 = arith.constant 8 : i64 loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<8x1xi64, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc39) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc40) + %o_t = arith.muli %i_t_3, %c8_i64 : i64 loc(#loc41) + %o_t_4 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc42) + %o_t_5 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32, #blocked1> loc(#loc42) + %o_t_6 = arith.extsi %o_t_4 : tensor<8xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_7 = arith.extsi %o_t_5 : tensor<8xi32, #blocked1> to tensor<8xi64, #blocked1> loc(#loc43) + %o_t_8 = tt.splat %o_t : i64 -> tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_9 = tt.splat %o_t : i64 -> tensor<8xi64, #blocked1> loc(#loc43) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<8xi64, #blocked1> loc(#loc43) + %m_t = arith.extsi %T : i32 to i64 loc(#loc44) + %m_t_12 = tt.splat %m_t : i64 -> tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_13 = tt.splat %m_t : i64 -> tensor<8xi64, #blocked1> loc(#loc44) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<8xi64, #blocked1> loc(#loc44) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<8xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xi1, #blocked> loc(#loc45) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc46) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc46) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_1 : tensor<1x128xi32, #blocked> loc(#loc47) + %m_x_19 = tt.broadcast %m_x : tensor<8x1xi1, #blocked> -> tensor<8x128xi1, #blocked> loc(#loc48) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<8x128xi1, #blocked> loc(#loc48) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<8x128xi1, #blocked> loc(#loc48) + %p_x = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xi64, #blocked> loc(#loc49) + %p_x_22 = arith.muli %p_x, %cst_2 : tensor<8x1xi64, #blocked> loc(#loc50) + %p_x_23 = tt.splat %x : !tt.ptr -> tensor<8x1x!tt.ptr, #blocked> loc(#loc51) + %p_x_24 = tt.addptr %p_x_23, %p_x_22 : tensor<8x1x!tt.ptr, #blocked>, tensor<8x1xi64, #blocked> loc(#loc51) + %p_x_25 = tt.broadcast %p_x_24 : tensor<8x1x!tt.ptr, #blocked> -> tensor<8x128x!tt.ptr, #blocked> loc(#loc52) + %p_x_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<8x128xi32, #blocked> loc(#loc52) + %p_x_27 = tt.addptr %p_x_25, %p_x_26 : tensor<8x128x!tt.ptr, #blocked>, tensor<8x128xi32, #blocked> loc(#loc52) + %p_y = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr, #blocked> loc(#loc53) + %p_y_28 = tt.addptr %p_y, %p_x_22 : tensor<8x1x!tt.ptr, #blocked>, tensor<8x1xi64, #blocked> loc(#loc53) + %p_y_29 = tt.broadcast %p_y_28 : tensor<8x1x!tt.ptr, #blocked> -> tensor<8x128x!tt.ptr, #blocked> loc(#loc54) + %p_y_30 = tt.addptr %p_y_29, %p_x_26 : tensor<8x128x!tt.ptr, #blocked>, tensor<8x128xi32, #blocked> loc(#loc54) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr, #blocked1> loc(#loc55) + %p_rstd_31 = tt.addptr %p_rstd, %o_t_11 : tensor<8x!tt.ptr, #blocked1>, tensor<8xi64, #blocked1> loc(#loc55) + %b_x = tt.load %p_x_27, %m_x_21, %cst_0 : tensor<8x128x!tt.ptr, #blocked> loc(#loc56) + %b_x_32 = arith.extf %b_x : tensor<8x128xbf16, #blocked> to tensor<8x128xf32, #blocked> loc(#loc57) + %b_rstd = arith.mulf %b_x_32, %b_x_32 : tensor<8x128xf32, #blocked> loc(#loc58) + %b_rstd_33 = "tt.reduce"(%b_rstd) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_40: f32 loc(callsite(#loc1 at #loc59)), %b_rstd_41: f32 loc(callsite(#loc1 at #loc59))): + %b_rstd_42 = arith.addf %b_rstd_40, %b_rstd_41 : f32 loc(#loc67) + tt.reduce.return %b_rstd_42 : f32 loc(#loc65) + }) : (tensor<8x128xf32, #blocked>) -> tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc65) + %b_rstd_34 = tt.splat %eps : f32 -> tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_36 = math.sqrt %b_rstd_35 : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc61) + %b_rstd_37 = arith.divf %cst, %b_rstd_36 : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc62) + %b_y = tt.expand_dims %b_rstd_37 {axis = 1 : i32} : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xf32, #blocked> loc(#loc63) + %b_y_38 = tt.broadcast %b_y : tensor<8x1xf32, #blocked> -> tensor<8x128xf32, #blocked> loc(#loc64) + %b_y_39 = arith.mulf %b_x_32, %b_y_38 : tensor<8x128xf32, #blocked> loc(#loc64) + %0 = arith.truncf %b_y_39 : tensor<8x128xf32, #blocked> to tensor<8x128xbf16, #blocked> loc(#loc30) + tt.store %p_y_30, %0, %m_x_21 : tensor<8x128x!tt.ptr, #blocked> loc(#loc31) + %1 = ttg.convert_layout %b_rstd_37 : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8xf32, #blocked1> loc(#loc32) + tt.store %p_rstd_31, %1, %m_t_15 : tensor<8x!tt.ptr, #blocked1> loc(#loc32) + tt.return loc(#loc33) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc39 = loc("i_t"(#loc2)) +#loc40 = loc("i_t"(#loc3)) +#loc41 = loc("o_t"(#loc4)) +#loc42 = loc("o_t"(#loc5)) +#loc43 = loc("o_t"(#loc6)) +#loc44 = loc("m_t"(#loc7)) +#loc45 = loc("m_x"(#loc8)) +#loc46 = loc("m_x"(#loc9)) +#loc47 = loc("m_x"(#loc10)) +#loc48 = loc("m_x"(#loc11)) +#loc49 = loc("p_x"(#loc12)) +#loc50 = loc("p_x"(#loc13)) +#loc51 = loc("p_x"(#loc14)) +#loc52 = loc("p_x"(#loc15)) +#loc53 = loc("p_y"(#loc16)) +#loc54 = loc("p_y"(#loc17)) +#loc55 = loc("p_rstd"(#loc18)) +#loc56 = loc("b_x"(#loc19)) +#loc57 = loc("b_x"(#loc20)) +#loc58 = loc("b_rstd"(#loc21)) +#loc60 = loc("b_rstd"(#loc25)) +#loc61 = loc("b_rstd"(#loc26)) +#loc62 = loc("b_rstd"(#loc27)) +#loc63 = loc("b_y"(#loc28)) +#loc64 = loc("b_y"(#loc29)) +#loc65 = loc(callsite(#loc22 at #loc59)) +#loc67 = loc(callsite(#loc24 at #loc65)) diff --git a/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.ttir b/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..8bbf17da02489a0d195ed0bd020180f9e4ac2496 --- /dev/null +++ b/triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.ttir @@ -0,0 +1,128 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc3 = loc(unknown) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc35 = loc("x"(#loc)) +#loc36 = loc("y"(#loc)) +#loc37 = loc("rstd"(#loc)) +#loc38 = loc("eps"(#loc)) +#loc39 = loc("T"(#loc)) +#loc62 = loc("b_rstd"(#loc25)) +#loc68 = loc(callsite(#loc3 at #loc62)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %b_x = arith.constant dense<0.000000e+00> : tensor<8x128xbf16> loc(#loc40) + %b_rstd = arith.constant dense<1.000000e+00> : tensor<8xf32> loc(#loc41) + %cst = arith.constant dense<128> : tensor<8x1xi64> loc(#loc3) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc42) + %o_t = arith.constant 8 : i64 loc(#loc43) + %i_t = tt.get_program_id x : i32 loc(#loc44) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc45) + %o_t_1 = arith.muli %i_t_0, %o_t : i64 loc(#loc43) + %o_t_2 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32> loc(#loc46) + %o_t_3 = arith.extsi %o_t_2 : tensor<8xi32> to tensor<8xi64> loc(#loc47) + %o_t_4 = tt.splat %o_t_1 : i64 -> tensor<8xi64> loc(#loc47) + %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<8xi64> loc(#loc47) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc48) + %m_t = arith.extsi %T : i32 to i64 loc(#loc49) + %m_t_6 = tt.splat %m_t : i64 -> tensor<8xi64> loc(#loc49) + %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<8xi64> loc(#loc49) + %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<8xi1> -> tensor<8x1xi1> loc(#loc50) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc51) + %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc42) + %m_x_11 = tt.broadcast %m_x_8 : tensor<8x1xi1> -> tensor<8x128xi1> loc(#loc52) + %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<8x128xi1> loc(#loc52) + %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<8x128xi1> loc(#loc52) + %p_x = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc53) + %p_x_14 = arith.muli %p_x, %cst : tensor<8x1xi64> loc(#loc54) + %p_x_15 = tt.splat %x : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc55) + %p_x_16 = tt.addptr %p_x_15, %p_x_14 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc55) + %p_x_17 = tt.broadcast %p_x_16 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc56) + %p_x_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc56) + %p_x_19 = tt.addptr %p_x_17, %p_x_18 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc56) + %p_y = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc57) + %p_y_20 = tt.addptr %p_y, %p_x_14 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc57) + %p_y_21 = tt.broadcast %p_y_20 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc58) + %p_y_22 = tt.addptr %p_y_21, %p_x_18 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr> loc(#loc59) + %p_rstd_23 = tt.addptr %p_rstd, %o_t_5 : tensor<8x!tt.ptr>, tensor<8xi64> loc(#loc59) + %b_x_24 = tt.load %p_x_19, %m_x_13, %b_x : tensor<8x128x!tt.ptr> loc(#loc40) + %b_x_25 = arith.extf %b_x_24 : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc60) + %b_rstd_26 = arith.mulf %b_x_25, %b_x_25 : tensor<8x128xf32> loc(#loc61) + %b_rstd_27 = "tt.reduce"(%b_rstd_26) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_34: f32 loc(callsite(#loc3 at #loc62)), %b_rstd_35: f32 loc(callsite(#loc3 at #loc62))): + %b_rstd_36 = arith.addf %b_rstd_34, %b_rstd_35 : f32 loc(#loc69) + tt.reduce.return %b_rstd_36 : f32 loc(#loc67) + }) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc67) + %b_rstd_28 = tt.splat %eps : f32 -> tensor<8xf32> loc(#loc63) + %b_rstd_29 = arith.addf %b_rstd_27, %b_rstd_28 : tensor<8xf32> loc(#loc63) + %b_rstd_30 = math.sqrt %b_rstd_29 : tensor<8xf32> loc(#loc64) + %b_rstd_31 = arith.divf %b_rstd, %b_rstd_30 : tensor<8xf32> loc(#loc41) + %b_y = tt.expand_dims %b_rstd_31 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc65) + %b_y_32 = tt.broadcast %b_y : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc66) + %b_y_33 = arith.mulf %b_x_25, %b_y_32 : tensor<8x128xf32> loc(#loc66) + %0 = arith.truncf %b_y_33 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc31) + tt.store %p_y_22, %0, %m_x_13 : tensor<8x128x!tt.ptr> loc(#loc32) + tt.store %p_rstd_23, %b_rstd_31, %m_t_7 : tensor<8x!tt.ptr> loc(#loc33) + tt.return loc(#loc34) + } loc(#loc) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc40 = loc("b_x"(#loc1)) +#loc41 = loc("b_rstd"(#loc2)) +#loc42 = loc("m_x"(#loc4)) +#loc43 = loc("o_t"(#loc5)) +#loc44 = loc("i_t"(#loc6)) +#loc45 = loc("i_t"(#loc7)) +#loc46 = loc("o_t"(#loc8)) +#loc47 = loc("o_t"(#loc9)) +#loc48 = loc("o_d"(#loc10)) +#loc49 = loc("m_t"(#loc11)) +#loc50 = loc("m_x"(#loc12)) +#loc51 = loc("m_x"(#loc13)) +#loc52 = loc("m_x"(#loc14)) +#loc53 = loc("p_x"(#loc15)) +#loc54 = loc("p_x"(#loc16)) +#loc55 = loc("p_x"(#loc17)) +#loc56 = loc("p_x"(#loc18)) +#loc57 = loc("p_y"(#loc19)) +#loc58 = loc("p_y"(#loc20)) +#loc59 = loc("p_rstd"(#loc21)) +#loc60 = loc("b_x"(#loc22)) +#loc61 = loc("b_rstd"(#loc23)) +#loc63 = loc("b_rstd"(#loc27)) +#loc64 = loc("b_rstd"(#loc28)) +#loc65 = loc("b_y"(#loc29)) +#loc66 = loc("b_y"(#loc30)) +#loc67 = loc(callsite(#loc24 at #loc62)) +#loc69 = loc(callsite(#loc26 at #loc67)) diff --git a/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/__grp__chunk_gated_delta_rule_fwd_kkt_solve_kernel.json b/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/__grp__chunk_gated_delta_rule_fwd_kkt_solve_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..e1bbad988e05c7a3e09812bc8a512b98e53c2532 --- /dev/null +++ b/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/__grp__chunk_gated_delta_rule_fwd_kkt_solve_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"chunk_gated_delta_rule_fwd_kkt_solve_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.source", "chunk_gated_delta_rule_fwd_kkt_solve_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.ttir", "chunk_gated_delta_rule_fwd_kkt_solve_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.ttgir", "chunk_gated_delta_rule_fwd_kkt_solve_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.llir", "chunk_gated_delta_rule_fwd_kkt_solve_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.ptx", "chunk_gated_delta_rule_fwd_kkt_solve_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.cubin", "chunk_gated_delta_rule_fwd_kkt_solve_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.cubin b/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..96f81dae4b2dcbcb2c0eb6a92fd1740c74428070 --- /dev/null +++ b/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.cubin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9bb766e55a38988733fa5624f5ad76c205d5a2425aa8f68d3aa02460f5caa9bd +size 210080 diff --git a/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.json b/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..05c8fcf31389f76089745a49fd786ff70a46f0ec --- /dev/null +++ b/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.json @@ -0,0 +1 @@ +{"hash": "06294eeeb36b73c59509c32e4cc6d7a7538db59f6d66486b9f92f9f566ebb85f", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 2, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 7168, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "chunk_gated_delta_rule_fwd_kkt_solve_kernel"} \ No newline at end of file diff --git a/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.llir b/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..bb8d9ae7d2b51d75029acacb59244ce0b6152a20 --- /dev/null +++ b/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.llir @@ -0,0 +1,5113 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @chunk_gated_delta_rule_fwd_kkt_solve_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, ptr addrspace(1) %3, i32 %4, ptr addrspace(1) readnone captures(none) %5, ptr addrspace(1) readnone captures(none) %6) local_unnamed_addr #0 !dbg !5 { + %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %9 = zext nneg i32 %8 to i64, !dbg !14 + %10 = sext i32 %4 to i64, !dbg !15 + %11 = shl nuw nsw i64 %9, 6, !dbg !16 + %.not = icmp slt i64 %11, %10, !dbg !17 + br i1 %.not, label %12, label %common.ret, !dbg !17 + +common.ret: ; preds = %7, %._crit_edge542 + ret void, !dbg !18 + +12: ; preds = %7 + %13 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.y(), !dbg !19 + %14 = zext nneg i32 %13 to i64, !dbg !20 + %15 = and i64 %14, 65504, !dbg !15 + %16 = and i64 %14, 31, !dbg !21 + %17 = or disjoint i64 %11, 16, !dbg !22 + %18 = or disjoint i64 %11, 32, !dbg !23 + %19 = or disjoint i64 %11, 48, !dbg !24 + %20 = mul nuw nsw i64 %15, %10, !dbg !25 + %21 = or disjoint i64 %20, %16, !dbg !26 + %.idx = shl nuw nsw i64 %21, 8, !dbg !27 + %22 = getelementptr i8, ptr addrspace(1) %0, i64 %.idx, !dbg !27 + %23 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !28 + %24 = and i32 %23, 56, !dbg !28 + %25 = lshr exact i32 %24, 3, !dbg !28 + %26 = or disjoint i32 %25, 8, !dbg !28 + %27 = and i32 %23, 28, !dbg !28 + %28 = and i32 %23, 3, !dbg !28 + %29 = and i32 %23, 32, !dbg !28 + %30 = lshr exact i32 %29, 2, !dbg !28 + %31 = zext nneg i32 %25 to i64, !dbg !29 + %32 = zext nneg i32 %26 to i64, !dbg !29 + %33 = or disjoint i64 %11, %31, !dbg !29 + %34 = or disjoint i64 %11, %32, !dbg !29 + %35 = icmp samesign ult i64 %33, %10, !dbg !30 + %36 = icmp samesign ult i64 %34, %10, !dbg !30 + %37 = or disjoint i64 %17, %31, !dbg !31 + %38 = or disjoint i64 %17, %32, !dbg !31 + %39 = icmp samesign ult i64 %37, %10, !dbg !32 + %40 = icmp samesign ult i64 %38, %10, !dbg !32 + %41 = or disjoint i64 %18, %31, !dbg !33 + %42 = or disjoint i64 %18, %32, !dbg !33 + %43 = icmp samesign ult i64 %41, %10, !dbg !34 + %44 = icmp samesign ult i64 %42, %10, !dbg !34 + %45 = or disjoint i64 %19, %31, !dbg !35 + %46 = or disjoint i64 %19, %32, !dbg !35 + %47 = icmp samesign ult i64 %45, %10, !dbg !36 + %48 = icmp samesign ult i64 %46, %10, !dbg !36 + %49 = shl nuw nsw i32 %28, 1, !dbg !28 + %50 = or disjoint i32 %49, %30, !dbg !28 + %51 = or disjoint i32 %50, 1, !dbg !28 + %52 = or disjoint i32 %49, 1, !dbg !28 + %53 = or disjoint i32 %49, 8, !dbg !28 + %54 = or disjoint i32 %49, 9, !dbg !28 + %55 = insertelement <4 x i32> poison, i32 %49, i64 0, !dbg !29 + %56 = insertelement <4 x i32> %55, i32 %52, i64 1, !dbg !29 + %57 = insertelement <4 x i32> %56, i32 %53, i64 2, !dbg !29 + %58 = insertelement <4 x i32> %57, i32 %54, i64 3, !dbg !29 + %.mask = and <4 x i32> %58, splat (i32 255), !dbg !33 + %59 = zext nneg <4 x i32> %.mask to <4 x i64>, !dbg !33 + %60 = extractelement <4 x i64> %59, i64 0, !dbg !29 + %61 = or disjoint i64 %11, %60, !dbg !29 + %62 = extractelement <4 x i64> %59, i64 1, !dbg !29 + %63 = or disjoint i64 %11, %62, !dbg !29 + %64 = extractelement <4 x i64> %59, i64 2, !dbg !29 + %65 = or disjoint i64 %11, %64, !dbg !29 + %66 = extractelement <4 x i64> %59, i64 3, !dbg !29 + %67 = or disjoint i64 %11, %66, !dbg !29 + %68 = icmp samesign ult i64 %61, %10, !dbg !30 + %69 = icmp samesign ult i64 %63, %10, !dbg !30 + %70 = icmp samesign ult i64 %65, %10, !dbg !30 + %71 = icmp samesign ult i64 %67, %10, !dbg !30 + %72 = insertelement <2 x i32> poison, i32 %50, i64 0, !dbg !29 + %73 = insertelement <2 x i32> %72, i32 %51, i64 1, !dbg !29 + %74 = insertelement <2 x i64> poison, i64 %11, i64 0, !dbg !29 + %75 = shufflevector <2 x i64> %74, <2 x i64> poison, <2 x i32> zeroinitializer, !dbg !29 + %76 = zext nneg <2 x i32> %73 to <2 x i64>, !dbg !29 + %77 = or disjoint <2 x i64> %75, %76, !dbg !29 + %78 = insertelement <2 x i64> poison, i64 %10, i64 0, !dbg !30 + %79 = shufflevector <2 x i64> %78, <2 x i64> poison, <2 x i32> zeroinitializer, !dbg !30 + %80 = icmp ult <2 x i64> %77, %79, !dbg !30 + %81 = shufflevector <2 x i1> %80, <2 x i1> poison, <4 x i32> , !dbg !30 + %82 = lshr exact i32 %27, 2, !dbg !28 + %83 = or disjoint i32 %82, 8, !dbg !28 + %84 = insertelement <2 x i32> poison, i32 %82, i64 0, !dbg !29 + %85 = insertelement <2 x i32> %84, i32 %83, i64 1, !dbg !29 + %86 = zext nneg <2 x i32> %85 to <2 x i64>, !dbg !29 + %87 = or disjoint <2 x i64> %75, %86, !dbg !29 + %88 = shl <2 x i64> %79, splat (i64 32), !dbg !30 + %89 = ashr exact <2 x i64> %88, splat (i64 32), !dbg !30 + %90 = icmp ult <2 x i64> %87, %89, !dbg !30 + %91 = shufflevector <2 x i1> %90, <2 x i1> poison, <4 x i32> , !dbg !30 + %92 = extractelement <2 x i64> %86, i64 1, !dbg !35 + %93 = or disjoint i64 %17, %92, !dbg !31 + %94 = extractelement <2 x i64> %86, i64 0, !dbg !35 + %95 = or disjoint i64 %17, %94, !dbg !31 + %96 = insertelement <4 x i64> poison, i64 %17, i64 0, !dbg !31 + %97 = shufflevector <4 x i64> %96, <4 x i64> poison, <4 x i32> zeroinitializer, !dbg !31 + %98 = or disjoint <4 x i64> %97, %59, !dbg !31 + %99 = icmp samesign ult i64 %93, %10, !dbg !32 + %100 = icmp samesign ult i64 %95, %10, !dbg !32 + %101 = insertelement <4 x i32> poison, i32 %4, i64 0, !dbg !32 + %102 = shufflevector <4 x i32> %101, <4 x i32> poison, <4 x i32> zeroinitializer, !dbg !32 + %103 = sext <4 x i32> %102 to <4 x i64>, !dbg !32 + %104 = icmp ult <4 x i64> %98, %103, !dbg !32 + %105 = or disjoint i64 %18, %92, !dbg !33 + %106 = or disjoint i64 %18, %94, !dbg !33 + %107 = insertelement <4 x i64> poison, i64 %18, i64 0, !dbg !33 + %108 = shufflevector <4 x i64> %107, <4 x i64> poison, <4 x i32> zeroinitializer, !dbg !33 + %109 = or disjoint <4 x i64> %108, %59, !dbg !33 + %110 = icmp samesign ult i64 %105, %10, !dbg !34 + %111 = icmp samesign ult i64 %106, %10, !dbg !34 + %112 = sext <4 x i32> %102 to <4 x i64>, !dbg !34 + %113 = icmp ult <4 x i64> %109, %112, !dbg !34 + %114 = or disjoint i64 %19, %92, !dbg !35 + %115 = or disjoint i64 %19, %94, !dbg !35 + %116 = insertelement <4 x i64> poison, i64 %19, i64 0, !dbg !35 + %117 = shufflevector <4 x i64> %116, <4 x i64> poison, <4 x i32> zeroinitializer, !dbg !35 + %.mask755 = and <4 x i32> %58, splat (i32 255), !dbg !35 + %118 = zext nneg <4 x i32> %.mask755 to <4 x i64>, !dbg !35 + %119 = or disjoint <4 x i64> %117, %118, !dbg !35 + %120 = icmp samesign ult i64 %114, %10, !dbg !36 + %121 = icmp samesign ult i64 %115, %10, !dbg !36 + %122 = sext <4 x i32> %102 to <4 x i64>, !dbg !36 + %123 = icmp ult <4 x i64> %119, %122, !dbg !36 + %124 = getelementptr bfloat, ptr addrspace(1) %2, i64 %21, !dbg !37 + %125 = extractelement <2 x i64> %87, i64 0, !dbg !38 + %126 = shl nuw nsw i64 %125, 5, !dbg !38 + %127 = extractelement <2 x i64> %87, i64 1, !dbg !38 + %128 = shl nuw nsw i64 %127, 5, !dbg !38 + %129 = getelementptr bfloat, ptr addrspace(1) %124, i64 %126, !dbg !39 + %130 = getelementptr bfloat, ptr addrspace(1) %124, i64 %128, !dbg !39 + %131 = shl nuw nsw i64 %95, 5, !dbg !40 + %132 = shl nuw nsw i64 %93, 5, !dbg !40 + %133 = getelementptr bfloat, ptr addrspace(1) %124, i64 %131, !dbg !41 + %134 = getelementptr bfloat, ptr addrspace(1) %124, i64 %132, !dbg !41 + %135 = shl nuw nsw i64 %106, 5, !dbg !42 + %136 = shl nuw nsw i64 %105, 5, !dbg !42 + %137 = getelementptr bfloat, ptr addrspace(1) %124, i64 %135, !dbg !43 + %138 = getelementptr bfloat, ptr addrspace(1) %124, i64 %136, !dbg !43 + %139 = shl nuw nsw i64 %115, 5, !dbg !44 + %140 = shl nuw nsw i64 %114, 5, !dbg !44 + %141 = getelementptr bfloat, ptr addrspace(1) %124, i64 %139, !dbg !45 + %142 = getelementptr bfloat, ptr addrspace(1) %124, i64 %140, !dbg !45 + %143 = extractelement <2 x i1> %90, i64 0, !dbg !46 + %144 = tail call i16 asm sideeffect "mov.u16 $0, $1;\0A\09@$3 ld.global.b16 { $0 }, [ $2 + 0 ];", "=c,c,l,b"(i16 0, ptr addrspace(1) %129, i1 %143) #2, !dbg !47 + %145 = extractelement <2 x i1> %90, i64 1, !dbg !46 + %146 = tail call i16 asm sideeffect "mov.u16 $0, $1;\0A\09@$3 ld.global.b16 { $0 }, [ $2 + 0 ];", "=c,c,l,b"(i16 0, ptr addrspace(1) %130, i1 %145) #2, !dbg !47 + %147 = tail call i16 asm sideeffect "mov.u16 $0, $1;\0A\09@$3 ld.global.b16 { $0 }, [ $2 + 0 ];", "=c,c,l,b"(i16 0, ptr addrspace(1) %133, i1 %100) #2, !dbg !48 + %148 = tail call i16 asm sideeffect "mov.u16 $0, $1;\0A\09@$3 ld.global.b16 { $0 }, [ $2 + 0 ];", "=c,c,l,b"(i16 0, ptr addrspace(1) %134, i1 %99) #2, !dbg !48 + %149 = tail call i16 asm sideeffect "mov.u16 $0, $1;\0A\09@$3 ld.global.b16 { $0 }, [ $2 + 0 ];", "=c,c,l,b"(i16 0, ptr addrspace(1) %137, i1 %111) #2, !dbg !49 + %150 = tail call i16 asm sideeffect "mov.u16 $0, $1;\0A\09@$3 ld.global.b16 { $0 }, [ $2 + 0 ];", "=c,c,l,b"(i16 0, ptr addrspace(1) %138, i1 %110) #2, !dbg !49 + %151 = tail call i16 asm sideeffect "mov.u16 $0, $1;\0A\09@$3 ld.global.b16 { $0 }, [ $2 + 0 ];", "=c,c,l,b"(i16 0, ptr addrspace(1) %141, i1 %121) #2, !dbg !50 + %152 = tail call i16 asm sideeffect "mov.u16 $0, $1;\0A\09@$3 ld.global.b16 { $0 }, [ $2 + 0 ];", "=c,c,l,b"(i16 0, ptr addrspace(1) %142, i1 %120) #2, !dbg !50 + %153 = getelementptr float, ptr addrspace(1) %1, i64 %21, !dbg !51 + %154 = getelementptr float, ptr addrspace(1) %153, i64 %126, !dbg !52 + %155 = getelementptr float, ptr addrspace(1) %153, i64 %128, !dbg !52 + %156 = extractelement <2 x i64> %77, i64 0, !dbg !52 + %.idx8 = shl nuw nsw i64 %156, 7, !dbg !52 + %157 = getelementptr i8, ptr addrspace(1) %153, i64 %.idx8, !dbg !52 + %158 = extractelement <2 x i64> %77, i64 1, !dbg !52 + %.idx9 = shl nuw nsw i64 %158, 7, !dbg !52 + %159 = getelementptr i8, ptr addrspace(1) %153, i64 %.idx9, !dbg !52 + %.idx10 = shl nuw nsw i64 %61, 7, !dbg !52 + %160 = getelementptr i8, ptr addrspace(1) %153, i64 %.idx10, !dbg !52 + %.idx11 = shl nuw nsw i64 %63, 7, !dbg !52 + %161 = getelementptr i8, ptr addrspace(1) %153, i64 %.idx11, !dbg !52 + %.idx12 = shl nuw nsw i64 %65, 7, !dbg !52 + %162 = getelementptr i8, ptr addrspace(1) %153, i64 %.idx12, !dbg !52 + %.idx13 = shl nuw nsw i64 %67, 7, !dbg !52 + %163 = getelementptr i8, ptr addrspace(1) %153, i64 %.idx13, !dbg !52 + %164 = getelementptr float, ptr addrspace(1) %153, i64 %131, !dbg !53 + %165 = getelementptr float, ptr addrspace(1) %153, i64 %132, !dbg !53 + %166 = extractelement <4 x i64> %98, i64 0, !dbg !53 + %.idx14 = shl nuw nsw i64 %166, 7, !dbg !53 + %167 = getelementptr i8, ptr addrspace(1) %153, i64 %.idx14, !dbg !53 + %168 = extractelement <4 x i64> %98, i64 1, !dbg !53 + %.idx15 = shl nuw nsw i64 %168, 7, !dbg !53 + %169 = getelementptr i8, ptr addrspace(1) %153, i64 %.idx15, !dbg !53 + %170 = extractelement <4 x i64> %98, i64 2, !dbg !53 + %.idx16 = shl nuw nsw i64 %170, 7, !dbg !53 + %171 = getelementptr i8, ptr addrspace(1) %153, i64 %.idx16, !dbg !53 + %172 = extractelement <4 x i64> %98, i64 3, !dbg !53 + %.idx17 = shl nuw nsw i64 %172, 7, !dbg !53 + %173 = getelementptr i8, ptr addrspace(1) %153, i64 %.idx17, !dbg !53 + %174 = getelementptr float, ptr addrspace(1) %153, i64 %135, !dbg !54 + %175 = getelementptr float, ptr addrspace(1) %153, i64 %136, !dbg !54 + %176 = extractelement <4 x i64> %109, i64 0, !dbg !54 + %.idx18 = shl nuw nsw i64 %176, 7, !dbg !54 + %177 = getelementptr i8, ptr addrspace(1) %153, i64 %.idx18, !dbg !54 + %178 = extractelement <4 x i64> %109, i64 1, !dbg !54 + %.idx19 = shl nuw nsw i64 %178, 7, !dbg !54 + %179 = getelementptr i8, ptr addrspace(1) %153, i64 %.idx19, !dbg !54 + %180 = extractelement <4 x i64> %109, i64 2, !dbg !54 + %.idx20 = shl nuw nsw i64 %180, 7, !dbg !54 + %181 = getelementptr i8, ptr addrspace(1) %153, i64 %.idx20, !dbg !54 + %182 = extractelement <4 x i64> %109, i64 3, !dbg !54 + %.idx21 = shl nuw nsw i64 %182, 7, !dbg !54 + %183 = getelementptr i8, ptr addrspace(1) %153, i64 %.idx21, !dbg !54 + %184 = getelementptr float, ptr addrspace(1) %153, i64 %139, !dbg !55 + %185 = getelementptr float, ptr addrspace(1) %153, i64 %140, !dbg !55 + %186 = extractelement <4 x i64> %119, i64 0, !dbg !55 + %.idx22 = shl nuw nsw i64 %186, 7, !dbg !55 + %187 = getelementptr i8, ptr addrspace(1) %153, i64 %.idx22, !dbg !55 + %188 = extractelement <4 x i64> %119, i64 1, !dbg !55 + %.idx23 = shl nuw nsw i64 %188, 7, !dbg !55 + %189 = getelementptr i8, ptr addrspace(1) %153, i64 %.idx23, !dbg !55 + %190 = extractelement <4 x i64> %119, i64 2, !dbg !55 + %.idx24 = shl nuw nsw i64 %190, 7, !dbg !55 + %191 = getelementptr i8, ptr addrspace(1) %153, i64 %.idx24, !dbg !55 + %192 = extractelement <4 x i64> %119, i64 3, !dbg !55 + %.idx25 = shl nuw nsw i64 %192, 7, !dbg !55 + %193 = getelementptr i8, ptr addrspace(1) %153, i64 %.idx25, !dbg !55 + %194 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %154, i1 %143) #2, !dbg !46 + %195 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %155, i1 %145) #2, !dbg !46 + %196 = extractelement <2 x i1> %80, i64 0, !dbg !46 + %197 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %157, i1 %196) #2, !dbg !46 + %198 = extractelement <2 x i1> %80, i64 1, !dbg !46 + %199 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %159, i1 %198) #2, !dbg !46 + %200 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %160, i1 %68) #2, !dbg !46 + %201 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %161, i1 %69) #2, !dbg !46 + %202 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %162, i1 %70) #2, !dbg !46 + %203 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %163, i1 %71) #2, !dbg !46 + %204 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %164, i1 %100) #2, !dbg !56 + %205 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %165, i1 %99) #2, !dbg !56 + %206 = extractelement <4 x i1> %104, i64 0, !dbg !56 + %207 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %167, i1 %206) #2, !dbg !56 + %208 = extractelement <4 x i1> %104, i64 1, !dbg !56 + %209 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %169, i1 %208) #2, !dbg !56 + %210 = extractelement <4 x i1> %104, i64 2, !dbg !56 + %211 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %171, i1 %210) #2, !dbg !56 + %212 = extractelement <4 x i1> %104, i64 3, !dbg !56 + %213 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %173, i1 %212) #2, !dbg !56 + %214 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %174, i1 %111) #2, !dbg !57 + %215 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %175, i1 %110) #2, !dbg !57 + %216 = extractelement <4 x i1> %113, i64 0, !dbg !57 + %217 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %177, i1 %216) #2, !dbg !57 + %218 = extractelement <4 x i1> %113, i64 1, !dbg !57 + %219 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %179, i1 %218) #2, !dbg !57 + %220 = extractelement <4 x i1> %113, i64 2, !dbg !57 + %221 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %181, i1 %220) #2, !dbg !57 + %222 = extractelement <4 x i1> %113, i64 3, !dbg !57 + %223 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %183, i1 %222) #2, !dbg !57 + %224 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %184, i1 %121) #2, !dbg !58 + %225 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %185, i1 %120) #2, !dbg !58 + %226 = extractelement <4 x i1> %123, i64 0, !dbg !58 + %227 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %187, i1 %226) #2, !dbg !58 + %228 = extractelement <4 x i1> %123, i64 1, !dbg !58 + %229 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %189, i1 %228) #2, !dbg !58 + %230 = extractelement <4 x i1> %123, i64 2, !dbg !58 + %231 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %191, i1 %230) #2, !dbg !58 + %232 = extractelement <4 x i1> %123, i64 3, !dbg !58 + %233 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %193, i1 %232) #2, !dbg !58 + %234 = and i32 %23, 7, !dbg !59 + %235 = shl nuw nsw i32 %234, 3, !dbg !59 + %.idx26 = shl nuw nsw i64 %33, 13, !dbg !60 + %236 = getelementptr i8, ptr addrspace(1) %22, i64 %.idx26, !dbg !60 + %.idx27 = shl nuw nsw i64 %34, 13, !dbg !60 + %237 = getelementptr i8, ptr addrspace(1) %22, i64 %.idx27, !dbg !60 + %238 = icmp samesign ult i64 %17, %10, !dbg !61 + %239 = zext nneg i32 %235 to i64, !dbg !62 + %240 = getelementptr bfloat, ptr addrspace(1) %236, i64 %239, !dbg !62 + %241 = getelementptr bfloat, ptr addrspace(1) %237, i64 %239, !dbg !62 + %242 = shl nuw nsw i32 %23, 4, !dbg !63 + %243 = and i32 %242, 1008, !dbg !63 + %244 = shl nuw nsw i32 %24, 1, !dbg !63 + %245 = xor i32 %243, %244, !dbg !63 + %246 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %245, !dbg !63 + %247 = select i1 %35, i32 16, i32 0, !dbg !63 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %246, ptr addrspace(1) %240, i32 %247) #2, !dbg !63 + %248 = or disjoint i32 %245, 1024, !dbg !63 + %249 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %248, !dbg !63 + %250 = select i1 %36, i32 16, i32 0, !dbg !63 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %249, ptr addrspace(1) %241, i32 %250) #2, !dbg !63 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !63 + %251 = or disjoint i32 %235, 64, !dbg !64 + %252 = zext nneg i32 %251 to i64, !dbg !62 + %253 = getelementptr bfloat, ptr addrspace(1) %236, i64 %252, !dbg !62 + %254 = getelementptr bfloat, ptr addrspace(1) %237, i64 %252, !dbg !62 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !63 + %255 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 2048), i32 %245, !dbg !63 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %255, ptr addrspace(1) %253, i32 %247) #2, !dbg !63 + %256 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 2048), i32 %248, !dbg !63 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %256, ptr addrspace(1) %254, i32 %250) #2, !dbg !63 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !63 + %invariant.op = or disjoint i32 %235, 128, !dbg !65 + %257 = shl nuw nsw i32 %234, 7 + %258 = shl nuw nsw i32 %234, 4 + %259 = shl nuw nsw i32 %23, 1 + %260 = and i32 %259, 48 + %261 = or disjoint i32 %257, %258 + %262 = xor i32 %261, %260 + %263 = xor i32 %262, 64 + %264 = shl nuw nsw i32 %29, 5 + %265 = or disjoint i32 %257, %264 + %266 = xor i32 %258, %260 + %267 = or disjoint i32 %265, %266 + %268 = xor i32 %267, 64 + %269 = shl nuw nsw i32 %23, 7 + %270 = and i32 %269, 1920 + %271 = and i32 %23, 16 + %272 = xor i32 %258, %271 + %273 = or disjoint i32 %272, %270 + %274 = xor i32 %273, 32 + %275 = xor i32 %273, 64 + %276 = xor i32 %273, 96 + %.idx203 = shl nuw nsw i64 %37, 13 + %277 = getelementptr i8, ptr addrspace(1) %22, i64 %.idx203 + %.idx204 = shl nuw nsw i64 %38, 13 + %278 = getelementptr i8, ptr addrspace(1) %22, i64 %.idx204 + %279 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 4096), i32 %245 + %280 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 4096), i32 %248 + %281 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 6144), i32 %245 + %282 = or disjoint i32 %266, %257 + %283 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 6144), i32 %282 + %284 = xor i32 %282, 64 + %285 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 6144), i32 %284 + %286 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 4096), i32 %273 + %287 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 4096), i32 %274 + %288 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 4096), i32 %275 + %289 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 4096), i32 %276 + %290 = icmp samesign ult i64 %18, %10 + %.idx219 = shl nuw nsw i64 %41, 13 + %291 = getelementptr i8, ptr addrspace(1) %22, i64 %.idx219 + %.idx220 = shl nuw nsw i64 %42, 13 + %292 = getelementptr i8, ptr addrspace(1) %22, i64 %.idx220 + %293 = icmp samesign ult i64 %19, %10 + %.idx235 = shl nuw nsw i64 %45, 13 + %294 = getelementptr i8, ptr addrspace(1) %22, i64 %.idx235 + %.idx236 = shl nuw nsw i64 %46, 13 + %295 = getelementptr i8, ptr addrspace(1) %22, i64 %.idx236 + %296 = shl nuw nsw i32 %23, 3, !dbg !65 + %297 = and i32 %296, 56, !dbg !65 + %298 = zext nneg i32 %297 to i64, !dbg !65 + %299 = zext nneg i32 %invariant.op to i64, !dbg !65 + br label %300, !dbg !65 + +300: ; preds = %12, %1007 + %indvars.iv593 = phi i64 [ 0, %12 ], [ 64, %1007 ] + %indvars.iv591 = phi i64 [ %298, %12 ], [ %indvars.iv.next592, %1007 ] + %301 = phi i1 [ true, %12 ], [ false, %1007 ] + %302 = phi i32 [ -1, %12 ], [ %373, %1007 ] + %303 = phi i32 [ 1, %12 ], [ %441, %1007 ] + %304 = phi float [ 0.000000e+00, %12 ], [ %1060, %1007 ] + %305 = phi float [ 0.000000e+00, %12 ], [ %1061, %1007 ] + %306 = phi float [ 0.000000e+00, %12 ], [ %1062, %1007 ] + %307 = phi float [ 0.000000e+00, %12 ], [ %1063, %1007 ] + %308 = phi float [ 0.000000e+00, %12 ], [ %1064, %1007 ] + %309 = phi float [ 0.000000e+00, %12 ], [ %1065, %1007 ] + %310 = phi float [ 0.000000e+00, %12 ], [ %1066, %1007 ] + %311 = phi float [ 0.000000e+00, %12 ], [ %1067, %1007 ] + %312 = phi float [ 0.000000e+00, %12 ], [ %1052, %1007 ] + %313 = phi float [ 0.000000e+00, %12 ], [ %1053, %1007 ] + %314 = phi float [ 0.000000e+00, %12 ], [ %1054, %1007 ] + %315 = phi float [ 0.000000e+00, %12 ], [ %1055, %1007 ] + %316 = phi float [ 0.000000e+00, %12 ], [ %1056, %1007 ] + %317 = phi float [ 0.000000e+00, %12 ], [ %1057, %1007 ] + %318 = phi float [ 0.000000e+00, %12 ], [ %1058, %1007 ] + %319 = phi float [ 0.000000e+00, %12 ], [ %1059, %1007 ] + %320 = phi float [ 0.000000e+00, %12 ], [ %1044, %1007 ] + %321 = phi float [ 0.000000e+00, %12 ], [ %1045, %1007 ] + %322 = phi float [ 0.000000e+00, %12 ], [ %1046, %1007 ] + %323 = phi float [ 0.000000e+00, %12 ], [ %1047, %1007 ] + %324 = phi float [ 0.000000e+00, %12 ], [ %1048, %1007 ] + %325 = phi float [ 0.000000e+00, %12 ], [ %1049, %1007 ] + %326 = phi float [ 0.000000e+00, %12 ], [ %1050, %1007 ] + %327 = phi float [ 0.000000e+00, %12 ], [ %1051, %1007 ] + %328 = phi float [ 0.000000e+00, %12 ], [ %1036, %1007 ] + %329 = phi float [ 0.000000e+00, %12 ], [ %1037, %1007 ] + %330 = phi float [ 0.000000e+00, %12 ], [ %1038, %1007 ] + %331 = phi float [ 0.000000e+00, %12 ], [ %1039, %1007 ] + %332 = phi float [ 0.000000e+00, %12 ], [ %1040, %1007 ] + %333 = phi float [ 0.000000e+00, %12 ], [ %1041, %1007 ] + %334 = phi float [ 0.000000e+00, %12 ], [ %1042, %1007 ] + %335 = phi float [ 0.000000e+00, %12 ], [ %1043, %1007 ] + %336 = phi float [ 0.000000e+00, %12 ], [ %1028, %1007 ] + %337 = phi float [ 0.000000e+00, %12 ], [ %1029, %1007 ] + %338 = phi float [ 0.000000e+00, %12 ], [ %1030, %1007 ] + %339 = phi float [ 0.000000e+00, %12 ], [ %1031, %1007 ] + %340 = phi float [ 0.000000e+00, %12 ], [ %1032, %1007 ] + %341 = phi float [ 0.000000e+00, %12 ], [ %1033, %1007 ] + %342 = phi float [ 0.000000e+00, %12 ], [ %1034, %1007 ] + %343 = phi float [ 0.000000e+00, %12 ], [ %1035, %1007 ] + %344 = phi float [ 0.000000e+00, %12 ], [ %1020, %1007 ] + %345 = phi float [ 0.000000e+00, %12 ], [ %1021, %1007 ] + %346 = phi float [ 0.000000e+00, %12 ], [ %1022, %1007 ] + %347 = phi float [ 0.000000e+00, %12 ], [ %1023, %1007 ] + %348 = phi float [ 0.000000e+00, %12 ], [ %1024, %1007 ] + %349 = phi float [ 0.000000e+00, %12 ], [ %1025, %1007 ] + %350 = phi float [ 0.000000e+00, %12 ], [ %1026, %1007 ] + %351 = phi float [ 0.000000e+00, %12 ], [ %1027, %1007 ] + %352 = phi float [ 0.000000e+00, %12 ], [ %1016, %1007 ] + %353 = phi float [ 0.000000e+00, %12 ], [ %1017, %1007 ] + %354 = phi float [ 0.000000e+00, %12 ], [ %1018, %1007 ] + %355 = phi float [ 0.000000e+00, %12 ], [ %1019, %1007 ] + %356 = phi float [ 0.000000e+00, %12 ], [ %1012, %1007 ] + %357 = phi float [ 0.000000e+00, %12 ], [ %1013, %1007 ] + %358 = phi float [ 0.000000e+00, %12 ], [ %1014, %1007 ] + %359 = phi float [ 0.000000e+00, %12 ], [ %1015, %1007 ] + %360 = phi float [ 0.000000e+00, %12 ], [ %1008, %1007 ] + %361 = phi float [ 0.000000e+00, %12 ], [ %1009, %1007 ] + %362 = phi float [ 0.000000e+00, %12 ], [ %1010, %1007 ] + %363 = phi float [ 0.000000e+00, %12 ], [ %1011, %1007 ] + %364 = phi float [ 0.000000e+00, %12 ], [ %435, %1007 ] + %365 = phi float [ 0.000000e+00, %12 ], [ %436, %1007 ] + %366 = phi float [ 0.000000e+00, %12 ], [ %437, %1007 ] + %367 = phi float [ 0.000000e+00, %12 ], [ %438, %1007 ] + %368 = phi <4 x float> [ zeroinitializer, %12 ], [ %1068, %1007 ] + %369 = phi <4 x float> [ zeroinitializer, %12 ], [ %1069, %1007 ] + %370 = phi <4 x float> [ zeroinitializer, %12 ], [ %1070, %1007 ] + %371 = add i32 %302, 1, !dbg !65 + %372 = icmp sgt i32 %371, 1, !dbg !65 + %373 = select i1 %372, i32 0, i32 %371, !dbg !65 + tail call void @llvm.nvvm.cp.async.wait.group(i32 1), !dbg !63 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !63 + %.idx201 = shl i32 %373, 11, !dbg !63 + %374 = getelementptr i8, ptr addrspace(3) @global_smem, i32 %.idx201, !dbg !63 + %375 = getelementptr inbounds nuw i8, ptr addrspace(3) %374, i32 %262, !dbg !66 + %376 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %375), !dbg !66 + %377 = getelementptr inbounds nuw i8, ptr addrspace(3) %375, i32 1024, !dbg !66 + %378 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %377), !dbg !66 + %379 = getelementptr inbounds nuw i8, ptr addrspace(3) %374, i32 %263, !dbg !66 + %380 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %379), !dbg !66 + %381 = getelementptr inbounds nuw i8, ptr addrspace(3) %379, i32 1024, !dbg !66 + %382 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %381), !dbg !66 + %383 = getelementptr inbounds nuw i8, ptr addrspace(3) %374, i32 %267, !dbg !66 + %384 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %383), !dbg !66 + %385 = extractvalue { i32, i32, i32, i32 } %384, 0, !dbg !66 + %386 = extractvalue { i32, i32, i32, i32 } %384, 1, !dbg !66 + %387 = extractvalue { i32, i32, i32, i32 } %384, 2, !dbg !66 + %388 = extractvalue { i32, i32, i32, i32 } %384, 3, !dbg !66 + %389 = getelementptr inbounds nuw i8, ptr addrspace(3) %374, i32 %268, !dbg !66 + %390 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %389), !dbg !66 + %391 = extractvalue { i32, i32, i32, i32 } %390, 0, !dbg !66 + %392 = extractvalue { i32, i32, i32, i32 } %390, 1, !dbg !66 + %393 = extractvalue { i32, i32, i32, i32 } %390, 2, !dbg !66 + %394 = extractvalue { i32, i32, i32, i32 } %390, 3, !dbg !66 + %395 = getelementptr inbounds nuw i8, ptr addrspace(3) %374, i32 %273, !dbg !63 + %396 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %395), !dbg !63 + %397 = extractvalue { i32, i32, i32, i32 } %396, 0, !dbg !63 + %398 = extractvalue { i32, i32, i32, i32 } %396, 1, !dbg !63 + %399 = extractvalue { i32, i32, i32, i32 } %396, 2, !dbg !63 + %400 = extractvalue { i32, i32, i32, i32 } %396, 3, !dbg !63 + %401 = getelementptr inbounds nuw i8, ptr addrspace(3) %374, i32 %274, !dbg !63 + %402 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %401), !dbg !63 + %403 = extractvalue { i32, i32, i32, i32 } %402, 0, !dbg !63 + %404 = extractvalue { i32, i32, i32, i32 } %402, 1, !dbg !63 + %405 = extractvalue { i32, i32, i32, i32 } %402, 2, !dbg !63 + %406 = extractvalue { i32, i32, i32, i32 } %402, 3, !dbg !63 + %407 = getelementptr inbounds nuw i8, ptr addrspace(3) %374, i32 %275, !dbg !63 + %408 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %407), !dbg !63 + %409 = extractvalue { i32, i32, i32, i32 } %408, 0, !dbg !63 + %410 = extractvalue { i32, i32, i32, i32 } %408, 1, !dbg !63 + %411 = extractvalue { i32, i32, i32, i32 } %408, 2, !dbg !63 + %412 = extractvalue { i32, i32, i32, i32 } %408, 3, !dbg !63 + %413 = getelementptr inbounds nuw i8, ptr addrspace(3) %374, i32 %276, !dbg !63 + %414 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %413), !dbg !63 + %415 = extractvalue { i32, i32, i32, i32 } %414, 0, !dbg !63 + %416 = extractvalue { i32, i32, i32, i32 } %414, 1, !dbg !63 + %417 = extractvalue { i32, i32, i32, i32 } %414, 2, !dbg !63 + %418 = extractvalue { i32, i32, i32, i32 } %414, 3, !dbg !63 + %419 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %364, float %365, float %366, float %367, i32 %397, i32 %398, i32 %399, i32 %400, i32 %385, i32 %386) #2, !dbg !66 + %420 = extractvalue { float, float, float, float } %419, 0, !dbg !66 + %421 = extractvalue { float, float, float, float } %419, 1, !dbg !66 + %422 = extractvalue { float, float, float, float } %419, 2, !dbg !66 + %423 = extractvalue { float, float, float, float } %419, 3, !dbg !66 + %424 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %420, float %421, float %422, float %423, i32 %403, i32 %404, i32 %405, i32 %406, i32 %387, i32 %388) #2, !dbg !66 + %425 = extractvalue { float, float, float, float } %424, 0, !dbg !66 + %426 = extractvalue { float, float, float, float } %424, 1, !dbg !66 + %427 = extractvalue { float, float, float, float } %424, 2, !dbg !66 + %428 = extractvalue { float, float, float, float } %424, 3, !dbg !66 + %429 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %425, float %426, float %427, float %428, i32 %409, i32 %410, i32 %411, i32 %412, i32 %391, i32 %392) #2, !dbg !66 + %430 = extractvalue { float, float, float, float } %429, 0, !dbg !66 + %431 = extractvalue { float, float, float, float } %429, 1, !dbg !66 + %432 = extractvalue { float, float, float, float } %429, 2, !dbg !66 + %433 = extractvalue { float, float, float, float } %429, 3, !dbg !66 + %434 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %430, float %431, float %432, float %433, i32 %415, i32 %416, i32 %417, i32 %418, i32 %393, i32 %394) #2, !dbg !66 + %435 = extractvalue { float, float, float, float } %434, 0, !dbg !66 + %436 = extractvalue { float, float, float, float } %434, 1, !dbg !66 + %437 = extractvalue { float, float, float, float } %434, 2, !dbg !66 + %438 = extractvalue { float, float, float, float } %434, 3, !dbg !66 + %439 = add i32 %303, 1, !dbg !65 + %440 = icmp sgt i32 %439, 1, !dbg !65 + %441 = select i1 %440, i32 0, i32 %439, !dbg !65 + %442 = or disjoint i64 %indvars.iv593, %299 + %443 = getelementptr bfloat, ptr addrspace(1) %236, i64 %442, !dbg !62 + %444 = getelementptr bfloat, ptr addrspace(1) %237, i64 %442, !dbg !62 + %.idx202 = shl i32 %441, 11, !dbg !63 + %445 = getelementptr i8, ptr addrspace(3) @global_smem, i32 %.idx202, !dbg !63 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !63 + %446 = getelementptr inbounds nuw i8, ptr addrspace(3) %445, i32 %245, !dbg !63 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %446, ptr addrspace(1) %443, i32 0) #2, !dbg !63 + %447 = getelementptr inbounds nuw i8, ptr addrspace(3) %445, i32 %248, !dbg !63 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %447, ptr addrspace(1) %444, i32 0) #2, !dbg !63 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !63 + br i1 %238, label %448, label %1007, !dbg !67 + +448: ; preds = %300 + %449 = extractvalue { i32, i32, i32, i32 } %382, 3, !dbg !66 + %450 = extractvalue { i32, i32, i32, i32 } %382, 2, !dbg !66 + %451 = extractvalue { i32, i32, i32, i32 } %382, 1, !dbg !66 + %452 = extractvalue { i32, i32, i32, i32 } %382, 0, !dbg !66 + %453 = extractvalue { i32, i32, i32, i32 } %380, 3, !dbg !66 + %454 = extractvalue { i32, i32, i32, i32 } %380, 2, !dbg !66 + %455 = extractvalue { i32, i32, i32, i32 } %380, 1, !dbg !66 + %456 = extractvalue { i32, i32, i32, i32 } %380, 0, !dbg !66 + %457 = extractvalue { i32, i32, i32, i32 } %378, 3, !dbg !66 + %458 = extractvalue { i32, i32, i32, i32 } %378, 2, !dbg !66 + %459 = extractvalue { i32, i32, i32, i32 } %378, 1, !dbg !66 + %460 = extractvalue { i32, i32, i32, i32 } %378, 0, !dbg !66 + %461 = extractvalue { i32, i32, i32, i32 } %376, 3, !dbg !66 + %462 = extractvalue { i32, i32, i32, i32 } %376, 2, !dbg !66 + %463 = extractvalue { i32, i32, i32, i32 } %376, 1, !dbg !66 + %464 = extractvalue { i32, i32, i32, i32 } %376, 0, !dbg !66 + %465 = getelementptr bfloat, ptr addrspace(1) %277, i64 %indvars.iv591, !dbg !68 + %466 = getelementptr bfloat, ptr addrspace(1) %278, i64 %indvars.iv591, !dbg !68 + %467 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %465, i1 %39) #2, !dbg !69 + %468 = extractvalue { i32, i32, i32, i32 } %467, 0, !dbg !69 + %469 = extractvalue { i32, i32, i32, i32 } %467, 1, !dbg !69 + %470 = extractvalue { i32, i32, i32, i32 } %467, 2, !dbg !69 + %471 = extractvalue { i32, i32, i32, i32 } %467, 3, !dbg !69 + %472 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %466, i1 %40) #2, !dbg !69 + %473 = extractvalue { i32, i32, i32, i32 } %472, 0, !dbg !69 + %474 = extractvalue { i32, i32, i32, i32 } %472, 1, !dbg !69 + %475 = extractvalue { i32, i32, i32, i32 } %472, 2, !dbg !69 + %476 = extractvalue { i32, i32, i32, i32 } %472, 3, !dbg !69 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %279, i32 %468, i32 %469, i32 %470, i32 %471) #2, !dbg !69 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %280, i32 %473, i32 %474, i32 %475, i32 %476) #2, !dbg !69 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %281, i32 %468, i32 %469, i32 %470, i32 %471) #2, !dbg !70 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !70 + %477 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %283), !dbg !70 + %478 = extractvalue { i32, i32, i32, i32 } %477, 0, !dbg !70 + %479 = extractvalue { i32, i32, i32, i32 } %477, 1, !dbg !70 + %480 = extractvalue { i32, i32, i32, i32 } %477, 2, !dbg !70 + %481 = extractvalue { i32, i32, i32, i32 } %477, 3, !dbg !70 + %482 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %285), !dbg !70 + %483 = extractvalue { i32, i32, i32, i32 } %482, 0, !dbg !70 + %484 = extractvalue { i32, i32, i32, i32 } %482, 1, !dbg !70 + %485 = extractvalue { i32, i32, i32, i32 } %482, 2, !dbg !70 + %486 = extractvalue { i32, i32, i32, i32 } %482, 3, !dbg !70 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !70 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %281, i32 %473, i32 %474, i32 %475, i32 %476) #2, !dbg !70 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !70 + %487 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %283), !dbg !70 + %488 = extractvalue { i32, i32, i32, i32 } %487, 0, !dbg !70 + %489 = extractvalue { i32, i32, i32, i32 } %487, 1, !dbg !70 + %490 = extractvalue { i32, i32, i32, i32 } %487, 2, !dbg !70 + %491 = extractvalue { i32, i32, i32, i32 } %487, 3, !dbg !70 + %492 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %285), !dbg !70 + %493 = extractvalue { i32, i32, i32, i32 } %492, 0, !dbg !70 + %494 = extractvalue { i32, i32, i32, i32 } %492, 1, !dbg !70 + %495 = extractvalue { i32, i32, i32, i32 } %492, 2, !dbg !70 + %496 = extractvalue { i32, i32, i32, i32 } %492, 3, !dbg !70 + %497 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %286), !dbg !69 + %498 = extractvalue { i32, i32, i32, i32 } %497, 0, !dbg !69 + %499 = extractvalue { i32, i32, i32, i32 } %497, 1, !dbg !69 + %500 = extractvalue { i32, i32, i32, i32 } %497, 2, !dbg !69 + %501 = extractvalue { i32, i32, i32, i32 } %497, 3, !dbg !69 + %502 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %287), !dbg !69 + %503 = extractvalue { i32, i32, i32, i32 } %502, 0, !dbg !69 + %504 = extractvalue { i32, i32, i32, i32 } %502, 1, !dbg !69 + %505 = extractvalue { i32, i32, i32, i32 } %502, 2, !dbg !69 + %506 = extractvalue { i32, i32, i32, i32 } %502, 3, !dbg !69 + %507 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %288), !dbg !69 + %508 = extractvalue { i32, i32, i32, i32 } %507, 0, !dbg !69 + %509 = extractvalue { i32, i32, i32, i32 } %507, 1, !dbg !69 + %510 = extractvalue { i32, i32, i32, i32 } %507, 2, !dbg !69 + %511 = extractvalue { i32, i32, i32, i32 } %507, 3, !dbg !69 + %512 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %289), !dbg !69 + %513 = extractvalue { i32, i32, i32, i32 } %512, 0, !dbg !69 + %514 = extractvalue { i32, i32, i32, i32 } %512, 1, !dbg !69 + %515 = extractvalue { i32, i32, i32, i32 } %512, 2, !dbg !69 + %516 = extractvalue { i32, i32, i32, i32 } %512, 3, !dbg !69 + %517 = extractelement <4 x float> %370, i64 0, !dbg !71 + %518 = extractelement <4 x float> %370, i64 1, !dbg !71 + %519 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %517, float %518, float %360, float %361, i32 %498, i32 %499, i32 %500, i32 %501, i32 %478, i32 %479) #2, !dbg !71 + %520 = extractvalue { float, float, float, float } %519, 0, !dbg !71 + %521 = extractvalue { float, float, float, float } %519, 1, !dbg !71 + %522 = extractvalue { float, float, float, float } %519, 2, !dbg !71 + %523 = extractvalue { float, float, float, float } %519, 3, !dbg !71 + %524 = extractelement <4 x float> %370, i64 2, !dbg !71 + %525 = extractelement <4 x float> %370, i64 3, !dbg !71 + %526 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %362, float %363, float %524, float %525, i32 %498, i32 %499, i32 %500, i32 %501, i32 %488, i32 %489) #2, !dbg !71 + %527 = extractvalue { float, float, float, float } %526, 0, !dbg !71 + %528 = extractvalue { float, float, float, float } %526, 1, !dbg !71 + %529 = extractvalue { float, float, float, float } %526, 2, !dbg !71 + %530 = extractvalue { float, float, float, float } %526, 3, !dbg !71 + %531 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %520, float %521, float %522, float %523, i32 %503, i32 %504, i32 %505, i32 %506, i32 %480, i32 %481) #2, !dbg !71 + %532 = extractvalue { float, float, float, float } %531, 0, !dbg !71 + %533 = extractvalue { float, float, float, float } %531, 1, !dbg !71 + %534 = extractvalue { float, float, float, float } %531, 2, !dbg !71 + %535 = extractvalue { float, float, float, float } %531, 3, !dbg !71 + %536 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %527, float %528, float %529, float %530, i32 %503, i32 %504, i32 %505, i32 %506, i32 %490, i32 %491) #2, !dbg !71 + %537 = extractvalue { float, float, float, float } %536, 0, !dbg !71 + %538 = extractvalue { float, float, float, float } %536, 1, !dbg !71 + %539 = extractvalue { float, float, float, float } %536, 2, !dbg !71 + %540 = extractvalue { float, float, float, float } %536, 3, !dbg !71 + %541 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %532, float %533, float %534, float %535, i32 %508, i32 %509, i32 %510, i32 %511, i32 %483, i32 %484) #2, !dbg !71 + %542 = extractvalue { float, float, float, float } %541, 0, !dbg !71 + %543 = extractvalue { float, float, float, float } %541, 1, !dbg !71 + %544 = extractvalue { float, float, float, float } %541, 2, !dbg !71 + %545 = extractvalue { float, float, float, float } %541, 3, !dbg !71 + %546 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %537, float %538, float %539, float %540, i32 %508, i32 %509, i32 %510, i32 %511, i32 %493, i32 %494) #2, !dbg !71 + %547 = extractvalue { float, float, float, float } %546, 0, !dbg !71 + %548 = extractvalue { float, float, float, float } %546, 1, !dbg !71 + %549 = extractvalue { float, float, float, float } %546, 2, !dbg !71 + %550 = extractvalue { float, float, float, float } %546, 3, !dbg !71 + %551 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %542, float %543, float %544, float %545, i32 %513, i32 %514, i32 %515, i32 %516, i32 %485, i32 %486) #2, !dbg !71 + %552 = extractvalue { float, float, float, float } %551, 0, !dbg !71 + %553 = extractvalue { float, float, float, float } %551, 1, !dbg !71 + %554 = extractvalue { float, float, float, float } %551, 2, !dbg !71 + %555 = extractvalue { float, float, float, float } %551, 3, !dbg !71 + %556 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %547, float %548, float %549, float %550, i32 %513, i32 %514, i32 %515, i32 %516, i32 %495, i32 %496) #2, !dbg !71 + %557 = extractvalue { float, float, float, float } %556, 0, !dbg !71 + %558 = extractvalue { float, float, float, float } %556, 1, !dbg !71 + %559 = extractvalue { float, float, float, float } %556, 2, !dbg !71 + %560 = extractvalue { float, float, float, float } %556, 3, !dbg !71 + %561 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %344, float %345, float %346, float %347, i32 %498, i32 %499, i32 %500, i32 %501, i32 %464, i32 %463) #2, !dbg !72 + %562 = extractvalue { float, float, float, float } %561, 0, !dbg !72 + %563 = extractvalue { float, float, float, float } %561, 1, !dbg !72 + %564 = extractvalue { float, float, float, float } %561, 2, !dbg !72 + %565 = extractvalue { float, float, float, float } %561, 3, !dbg !72 + %566 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %348, float %349, float %350, float %351, i32 %498, i32 %499, i32 %500, i32 %501, i32 %460, i32 %459) #2, !dbg !72 + %567 = extractvalue { float, float, float, float } %566, 0, !dbg !72 + %568 = extractvalue { float, float, float, float } %566, 1, !dbg !72 + %569 = extractvalue { float, float, float, float } %566, 2, !dbg !72 + %570 = extractvalue { float, float, float, float } %566, 3, !dbg !72 + %571 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %562, float %563, float %564, float %565, i32 %503, i32 %504, i32 %505, i32 %506, i32 %462, i32 %461) #2, !dbg !72 + %572 = extractvalue { float, float, float, float } %571, 0, !dbg !72 + %573 = extractvalue { float, float, float, float } %571, 1, !dbg !72 + %574 = extractvalue { float, float, float, float } %571, 2, !dbg !72 + %575 = extractvalue { float, float, float, float } %571, 3, !dbg !72 + %576 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %567, float %568, float %569, float %570, i32 %503, i32 %504, i32 %505, i32 %506, i32 %458, i32 %457) #2, !dbg !72 + %577 = extractvalue { float, float, float, float } %576, 0, !dbg !72 + %578 = extractvalue { float, float, float, float } %576, 1, !dbg !72 + %579 = extractvalue { float, float, float, float } %576, 2, !dbg !72 + %580 = extractvalue { float, float, float, float } %576, 3, !dbg !72 + %581 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %572, float %573, float %574, float %575, i32 %508, i32 %509, i32 %510, i32 %511, i32 %456, i32 %455) #2, !dbg !72 + %582 = extractvalue { float, float, float, float } %581, 0, !dbg !72 + %583 = extractvalue { float, float, float, float } %581, 1, !dbg !72 + %584 = extractvalue { float, float, float, float } %581, 2, !dbg !72 + %585 = extractvalue { float, float, float, float } %581, 3, !dbg !72 + %586 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %577, float %578, float %579, float %580, i32 %508, i32 %509, i32 %510, i32 %511, i32 %452, i32 %451) #2, !dbg !72 + %587 = extractvalue { float, float, float, float } %586, 0, !dbg !72 + %588 = extractvalue { float, float, float, float } %586, 1, !dbg !72 + %589 = extractvalue { float, float, float, float } %586, 2, !dbg !72 + %590 = extractvalue { float, float, float, float } %586, 3, !dbg !72 + %591 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %582, float %583, float %584, float %585, i32 %513, i32 %514, i32 %515, i32 %516, i32 %454, i32 %453) #2, !dbg !72 + %592 = extractvalue { float, float, float, float } %591, 0, !dbg !72 + %593 = extractvalue { float, float, float, float } %591, 1, !dbg !72 + %594 = extractvalue { float, float, float, float } %591, 2, !dbg !72 + %595 = extractvalue { float, float, float, float } %591, 3, !dbg !72 + %596 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %587, float %588, float %589, float %590, i32 %513, i32 %514, i32 %515, i32 %516, i32 %450, i32 %449) #2, !dbg !72 + %597 = extractvalue { float, float, float, float } %596, 0, !dbg !72 + %598 = extractvalue { float, float, float, float } %596, 1, !dbg !72 + %599 = extractvalue { float, float, float, float } %596, 2, !dbg !72 + %600 = extractvalue { float, float, float, float } %596, 3, !dbg !72 + %601 = insertelement <4 x float> poison, float %552, i64 0, !dbg !71 + %602 = insertelement <4 x float> %601, float %553, i64 1, !dbg !71 + %603 = insertelement <4 x float> %602, float %559, i64 2, !dbg !71 + %604 = insertelement <4 x float> %603, float %560, i64 3, !dbg !71 + br i1 %290, label %605, label %1007, !dbg !73 + +605: ; preds = %448 + %606 = getelementptr bfloat, ptr addrspace(1) %291, i64 %indvars.iv591, !dbg !74 + %607 = getelementptr bfloat, ptr addrspace(1) %292, i64 %indvars.iv591, !dbg !74 + %608 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %606, i1 %43) #2, !dbg !75 + %609 = extractvalue { i32, i32, i32, i32 } %608, 0, !dbg !75 + %610 = extractvalue { i32, i32, i32, i32 } %608, 1, !dbg !75 + %611 = extractvalue { i32, i32, i32, i32 } %608, 2, !dbg !75 + %612 = extractvalue { i32, i32, i32, i32 } %608, 3, !dbg !75 + %613 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %607, i1 %44) #2, !dbg !75 + %614 = extractvalue { i32, i32, i32, i32 } %613, 0, !dbg !75 + %615 = extractvalue { i32, i32, i32, i32 } %613, 1, !dbg !75 + %616 = extractvalue { i32, i32, i32, i32 } %613, 2, !dbg !75 + %617 = extractvalue { i32, i32, i32, i32 } %613, 3, !dbg !75 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !75 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %279, i32 %609, i32 %610, i32 %611, i32 %612) #2, !dbg !75 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %280, i32 %614, i32 %615, i32 %616, i32 %617) #2, !dbg !75 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %281, i32 %609, i32 %610, i32 %611, i32 %612) #2, !dbg !76 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !76 + %618 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %283), !dbg !76 + %619 = extractvalue { i32, i32, i32, i32 } %618, 0, !dbg !76 + %620 = extractvalue { i32, i32, i32, i32 } %618, 1, !dbg !76 + %621 = extractvalue { i32, i32, i32, i32 } %618, 2, !dbg !76 + %622 = extractvalue { i32, i32, i32, i32 } %618, 3, !dbg !76 + %623 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %285), !dbg !76 + %624 = extractvalue { i32, i32, i32, i32 } %623, 0, !dbg !76 + %625 = extractvalue { i32, i32, i32, i32 } %623, 1, !dbg !76 + %626 = extractvalue { i32, i32, i32, i32 } %623, 2, !dbg !76 + %627 = extractvalue { i32, i32, i32, i32 } %623, 3, !dbg !76 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !76 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %281, i32 %614, i32 %615, i32 %616, i32 %617) #2, !dbg !76 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !76 + %628 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %283), !dbg !76 + %629 = extractvalue { i32, i32, i32, i32 } %628, 0, !dbg !76 + %630 = extractvalue { i32, i32, i32, i32 } %628, 1, !dbg !76 + %631 = extractvalue { i32, i32, i32, i32 } %628, 2, !dbg !76 + %632 = extractvalue { i32, i32, i32, i32 } %628, 3, !dbg !76 + %633 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %285), !dbg !76 + %634 = extractvalue { i32, i32, i32, i32 } %633, 0, !dbg !76 + %635 = extractvalue { i32, i32, i32, i32 } %633, 1, !dbg !76 + %636 = extractvalue { i32, i32, i32, i32 } %633, 2, !dbg !76 + %637 = extractvalue { i32, i32, i32, i32 } %633, 3, !dbg !76 + %638 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %286), !dbg !75 + %639 = extractvalue { i32, i32, i32, i32 } %638, 0, !dbg !75 + %640 = extractvalue { i32, i32, i32, i32 } %638, 1, !dbg !75 + %641 = extractvalue { i32, i32, i32, i32 } %638, 2, !dbg !75 + %642 = extractvalue { i32, i32, i32, i32 } %638, 3, !dbg !75 + %643 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %287), !dbg !75 + %644 = extractvalue { i32, i32, i32, i32 } %643, 0, !dbg !75 + %645 = extractvalue { i32, i32, i32, i32 } %643, 1, !dbg !75 + %646 = extractvalue { i32, i32, i32, i32 } %643, 2, !dbg !75 + %647 = extractvalue { i32, i32, i32, i32 } %643, 3, !dbg !75 + %648 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %288), !dbg !75 + %649 = extractvalue { i32, i32, i32, i32 } %648, 0, !dbg !75 + %650 = extractvalue { i32, i32, i32, i32 } %648, 1, !dbg !75 + %651 = extractvalue { i32, i32, i32, i32 } %648, 2, !dbg !75 + %652 = extractvalue { i32, i32, i32, i32 } %648, 3, !dbg !75 + %653 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %289), !dbg !75 + %654 = extractvalue { i32, i32, i32, i32 } %653, 0, !dbg !75 + %655 = extractvalue { i32, i32, i32, i32 } %653, 1, !dbg !75 + %656 = extractvalue { i32, i32, i32, i32 } %653, 2, !dbg !75 + %657 = extractvalue { i32, i32, i32, i32 } %653, 3, !dbg !75 + %658 = extractelement <4 x float> %369, i64 0, !dbg !77 + %659 = extractelement <4 x float> %369, i64 1, !dbg !77 + %660 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %658, float %659, float %356, float %357, i32 %639, i32 %640, i32 %641, i32 %642, i32 %619, i32 %620) #2, !dbg !77 + %661 = extractvalue { float, float, float, float } %660, 0, !dbg !77 + %662 = extractvalue { float, float, float, float } %660, 1, !dbg !77 + %663 = extractvalue { float, float, float, float } %660, 2, !dbg !77 + %664 = extractvalue { float, float, float, float } %660, 3, !dbg !77 + %665 = extractelement <4 x float> %369, i64 2, !dbg !77 + %666 = extractelement <4 x float> %369, i64 3, !dbg !77 + %667 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %358, float %359, float %665, float %666, i32 %639, i32 %640, i32 %641, i32 %642, i32 %629, i32 %630) #2, !dbg !77 + %668 = extractvalue { float, float, float, float } %667, 0, !dbg !77 + %669 = extractvalue { float, float, float, float } %667, 1, !dbg !77 + %670 = extractvalue { float, float, float, float } %667, 2, !dbg !77 + %671 = extractvalue { float, float, float, float } %667, 3, !dbg !77 + %672 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %661, float %662, float %663, float %664, i32 %644, i32 %645, i32 %646, i32 %647, i32 %621, i32 %622) #2, !dbg !77 + %673 = extractvalue { float, float, float, float } %672, 0, !dbg !77 + %674 = extractvalue { float, float, float, float } %672, 1, !dbg !77 + %675 = extractvalue { float, float, float, float } %672, 2, !dbg !77 + %676 = extractvalue { float, float, float, float } %672, 3, !dbg !77 + %677 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %668, float %669, float %670, float %671, i32 %644, i32 %645, i32 %646, i32 %647, i32 %631, i32 %632) #2, !dbg !77 + %678 = extractvalue { float, float, float, float } %677, 0, !dbg !77 + %679 = extractvalue { float, float, float, float } %677, 1, !dbg !77 + %680 = extractvalue { float, float, float, float } %677, 2, !dbg !77 + %681 = extractvalue { float, float, float, float } %677, 3, !dbg !77 + %682 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %673, float %674, float %675, float %676, i32 %649, i32 %650, i32 %651, i32 %652, i32 %624, i32 %625) #2, !dbg !77 + %683 = extractvalue { float, float, float, float } %682, 0, !dbg !77 + %684 = extractvalue { float, float, float, float } %682, 1, !dbg !77 + %685 = extractvalue { float, float, float, float } %682, 2, !dbg !77 + %686 = extractvalue { float, float, float, float } %682, 3, !dbg !77 + %687 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %678, float %679, float %680, float %681, i32 %649, i32 %650, i32 %651, i32 %652, i32 %634, i32 %635) #2, !dbg !77 + %688 = extractvalue { float, float, float, float } %687, 0, !dbg !77 + %689 = extractvalue { float, float, float, float } %687, 1, !dbg !77 + %690 = extractvalue { float, float, float, float } %687, 2, !dbg !77 + %691 = extractvalue { float, float, float, float } %687, 3, !dbg !77 + %692 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %683, float %684, float %685, float %686, i32 %654, i32 %655, i32 %656, i32 %657, i32 %626, i32 %627) #2, !dbg !77 + %693 = extractvalue { float, float, float, float } %692, 0, !dbg !77 + %694 = extractvalue { float, float, float, float } %692, 1, !dbg !77 + %695 = extractvalue { float, float, float, float } %692, 2, !dbg !77 + %696 = extractvalue { float, float, float, float } %692, 3, !dbg !77 + %697 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %688, float %689, float %690, float %691, i32 %654, i32 %655, i32 %656, i32 %657, i32 %636, i32 %637) #2, !dbg !77 + %698 = extractvalue { float, float, float, float } %697, 0, !dbg !77 + %699 = extractvalue { float, float, float, float } %697, 1, !dbg !77 + %700 = extractvalue { float, float, float, float } %697, 2, !dbg !77 + %701 = extractvalue { float, float, float, float } %697, 3, !dbg !77 + %702 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %336, float %337, float %338, float %339, i32 %639, i32 %640, i32 %641, i32 %642, i32 %464, i32 %463) #2, !dbg !78 + %703 = extractvalue { float, float, float, float } %702, 0, !dbg !78 + %704 = extractvalue { float, float, float, float } %702, 1, !dbg !78 + %705 = extractvalue { float, float, float, float } %702, 2, !dbg !78 + %706 = extractvalue { float, float, float, float } %702, 3, !dbg !78 + %707 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %340, float %341, float %342, float %343, i32 %639, i32 %640, i32 %641, i32 %642, i32 %460, i32 %459) #2, !dbg !78 + %708 = extractvalue { float, float, float, float } %707, 0, !dbg !78 + %709 = extractvalue { float, float, float, float } %707, 1, !dbg !78 + %710 = extractvalue { float, float, float, float } %707, 2, !dbg !78 + %711 = extractvalue { float, float, float, float } %707, 3, !dbg !78 + %712 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %703, float %704, float %705, float %706, i32 %644, i32 %645, i32 %646, i32 %647, i32 %462, i32 %461) #2, !dbg !78 + %713 = extractvalue { float, float, float, float } %712, 0, !dbg !78 + %714 = extractvalue { float, float, float, float } %712, 1, !dbg !78 + %715 = extractvalue { float, float, float, float } %712, 2, !dbg !78 + %716 = extractvalue { float, float, float, float } %712, 3, !dbg !78 + %717 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %708, float %709, float %710, float %711, i32 %644, i32 %645, i32 %646, i32 %647, i32 %458, i32 %457) #2, !dbg !78 + %718 = extractvalue { float, float, float, float } %717, 0, !dbg !78 + %719 = extractvalue { float, float, float, float } %717, 1, !dbg !78 + %720 = extractvalue { float, float, float, float } %717, 2, !dbg !78 + %721 = extractvalue { float, float, float, float } %717, 3, !dbg !78 + %722 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %713, float %714, float %715, float %716, i32 %649, i32 %650, i32 %651, i32 %652, i32 %456, i32 %455) #2, !dbg !78 + %723 = extractvalue { float, float, float, float } %722, 0, !dbg !78 + %724 = extractvalue { float, float, float, float } %722, 1, !dbg !78 + %725 = extractvalue { float, float, float, float } %722, 2, !dbg !78 + %726 = extractvalue { float, float, float, float } %722, 3, !dbg !78 + %727 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %718, float %719, float %720, float %721, i32 %649, i32 %650, i32 %651, i32 %652, i32 %452, i32 %451) #2, !dbg !78 + %728 = extractvalue { float, float, float, float } %727, 0, !dbg !78 + %729 = extractvalue { float, float, float, float } %727, 1, !dbg !78 + %730 = extractvalue { float, float, float, float } %727, 2, !dbg !78 + %731 = extractvalue { float, float, float, float } %727, 3, !dbg !78 + %732 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %723, float %724, float %725, float %726, i32 %654, i32 %655, i32 %656, i32 %657, i32 %454, i32 %453) #2, !dbg !78 + %733 = extractvalue { float, float, float, float } %732, 0, !dbg !78 + %734 = extractvalue { float, float, float, float } %732, 1, !dbg !78 + %735 = extractvalue { float, float, float, float } %732, 2, !dbg !78 + %736 = extractvalue { float, float, float, float } %732, 3, !dbg !78 + %737 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %728, float %729, float %730, float %731, i32 %654, i32 %655, i32 %656, i32 %657, i32 %450, i32 %449) #2, !dbg !78 + %738 = extractvalue { float, float, float, float } %737, 0, !dbg !78 + %739 = extractvalue { float, float, float, float } %737, 1, !dbg !78 + %740 = extractvalue { float, float, float, float } %737, 2, !dbg !78 + %741 = extractvalue { float, float, float, float } %737, 3, !dbg !78 + %742 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %328, float %329, float %330, float %331, i32 %639, i32 %640, i32 %641, i32 %642, i32 %478, i32 %479) #2, !dbg !79 + %743 = extractvalue { float, float, float, float } %742, 0, !dbg !79 + %744 = extractvalue { float, float, float, float } %742, 1, !dbg !79 + %745 = extractvalue { float, float, float, float } %742, 2, !dbg !79 + %746 = extractvalue { float, float, float, float } %742, 3, !dbg !79 + %747 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %332, float %333, float %334, float %335, i32 %639, i32 %640, i32 %641, i32 %642, i32 %488, i32 %489) #2, !dbg !79 + %748 = extractvalue { float, float, float, float } %747, 0, !dbg !79 + %749 = extractvalue { float, float, float, float } %747, 1, !dbg !79 + %750 = extractvalue { float, float, float, float } %747, 2, !dbg !79 + %751 = extractvalue { float, float, float, float } %747, 3, !dbg !79 + %752 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %743, float %744, float %745, float %746, i32 %644, i32 %645, i32 %646, i32 %647, i32 %480, i32 %481) #2, !dbg !79 + %753 = extractvalue { float, float, float, float } %752, 0, !dbg !79 + %754 = extractvalue { float, float, float, float } %752, 1, !dbg !79 + %755 = extractvalue { float, float, float, float } %752, 2, !dbg !79 + %756 = extractvalue { float, float, float, float } %752, 3, !dbg !79 + %757 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %748, float %749, float %750, float %751, i32 %644, i32 %645, i32 %646, i32 %647, i32 %490, i32 %491) #2, !dbg !79 + %758 = extractvalue { float, float, float, float } %757, 0, !dbg !79 + %759 = extractvalue { float, float, float, float } %757, 1, !dbg !79 + %760 = extractvalue { float, float, float, float } %757, 2, !dbg !79 + %761 = extractvalue { float, float, float, float } %757, 3, !dbg !79 + %762 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %753, float %754, float %755, float %756, i32 %649, i32 %650, i32 %651, i32 %652, i32 %483, i32 %484) #2, !dbg !79 + %763 = extractvalue { float, float, float, float } %762, 0, !dbg !79 + %764 = extractvalue { float, float, float, float } %762, 1, !dbg !79 + %765 = extractvalue { float, float, float, float } %762, 2, !dbg !79 + %766 = extractvalue { float, float, float, float } %762, 3, !dbg !79 + %767 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %758, float %759, float %760, float %761, i32 %649, i32 %650, i32 %651, i32 %652, i32 %493, i32 %494) #2, !dbg !79 + %768 = extractvalue { float, float, float, float } %767, 0, !dbg !79 + %769 = extractvalue { float, float, float, float } %767, 1, !dbg !79 + %770 = extractvalue { float, float, float, float } %767, 2, !dbg !79 + %771 = extractvalue { float, float, float, float } %767, 3, !dbg !79 + %772 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %763, float %764, float %765, float %766, i32 %654, i32 %655, i32 %656, i32 %657, i32 %485, i32 %486) #2, !dbg !79 + %773 = extractvalue { float, float, float, float } %772, 0, !dbg !79 + %774 = extractvalue { float, float, float, float } %772, 1, !dbg !79 + %775 = extractvalue { float, float, float, float } %772, 2, !dbg !79 + %776 = extractvalue { float, float, float, float } %772, 3, !dbg !79 + %777 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %768, float %769, float %770, float %771, i32 %654, i32 %655, i32 %656, i32 %657, i32 %495, i32 %496) #2, !dbg !79 + %778 = extractvalue { float, float, float, float } %777, 0, !dbg !79 + %779 = extractvalue { float, float, float, float } %777, 1, !dbg !79 + %780 = extractvalue { float, float, float, float } %777, 2, !dbg !79 + %781 = extractvalue { float, float, float, float } %777, 3, !dbg !79 + %782 = insertelement <4 x float> poison, float %693, i64 0, !dbg !73 + %783 = insertelement <4 x float> %782, float %694, i64 1, !dbg !73 + %784 = insertelement <4 x float> %783, float %700, i64 2, !dbg !73 + %785 = insertelement <4 x float> %784, float %701, i64 3, !dbg !73 + br i1 %293, label %786, label %1007, !dbg !80 + +786: ; preds = %605 + %787 = getelementptr bfloat, ptr addrspace(1) %294, i64 %indvars.iv591, !dbg !81 + %788 = getelementptr bfloat, ptr addrspace(1) %295, i64 %indvars.iv591, !dbg !81 + %789 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %787, i1 %47) #2, !dbg !82 + %790 = extractvalue { i32, i32, i32, i32 } %789, 0, !dbg !82 + %791 = extractvalue { i32, i32, i32, i32 } %789, 1, !dbg !82 + %792 = extractvalue { i32, i32, i32, i32 } %789, 2, !dbg !82 + %793 = extractvalue { i32, i32, i32, i32 } %789, 3, !dbg !82 + %794 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %788, i1 %48) #2, !dbg !82 + %795 = extractvalue { i32, i32, i32, i32 } %794, 0, !dbg !82 + %796 = extractvalue { i32, i32, i32, i32 } %794, 1, !dbg !82 + %797 = extractvalue { i32, i32, i32, i32 } %794, 2, !dbg !82 + %798 = extractvalue { i32, i32, i32, i32 } %794, 3, !dbg !82 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !82 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %279, i32 %790, i32 %791, i32 %792, i32 %793) #2, !dbg !82 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %280, i32 %795, i32 %796, i32 %797, i32 %798) #2, !dbg !82 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %281, i32 %790, i32 %791, i32 %792, i32 %793) #2, !dbg !83 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !83 + %799 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %283), !dbg !83 + %800 = extractvalue { i32, i32, i32, i32 } %799, 0, !dbg !83 + %801 = extractvalue { i32, i32, i32, i32 } %799, 1, !dbg !83 + %802 = extractvalue { i32, i32, i32, i32 } %799, 2, !dbg !83 + %803 = extractvalue { i32, i32, i32, i32 } %799, 3, !dbg !83 + %804 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %285), !dbg !83 + %805 = extractvalue { i32, i32, i32, i32 } %804, 0, !dbg !83 + %806 = extractvalue { i32, i32, i32, i32 } %804, 1, !dbg !83 + %807 = extractvalue { i32, i32, i32, i32 } %804, 2, !dbg !83 + %808 = extractvalue { i32, i32, i32, i32 } %804, 3, !dbg !83 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !83 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %281, i32 %795, i32 %796, i32 %797, i32 %798) #2, !dbg !83 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !83 + %809 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %283), !dbg !83 + %810 = extractvalue { i32, i32, i32, i32 } %809, 0, !dbg !83 + %811 = extractvalue { i32, i32, i32, i32 } %809, 1, !dbg !83 + %812 = extractvalue { i32, i32, i32, i32 } %809, 2, !dbg !83 + %813 = extractvalue { i32, i32, i32, i32 } %809, 3, !dbg !83 + %814 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %285), !dbg !83 + %815 = extractvalue { i32, i32, i32, i32 } %814, 0, !dbg !83 + %816 = extractvalue { i32, i32, i32, i32 } %814, 1, !dbg !83 + %817 = extractvalue { i32, i32, i32, i32 } %814, 2, !dbg !83 + %818 = extractvalue { i32, i32, i32, i32 } %814, 3, !dbg !83 + %819 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %286), !dbg !82 + %820 = extractvalue { i32, i32, i32, i32 } %819, 0, !dbg !82 + %821 = extractvalue { i32, i32, i32, i32 } %819, 1, !dbg !82 + %822 = extractvalue { i32, i32, i32, i32 } %819, 2, !dbg !82 + %823 = extractvalue { i32, i32, i32, i32 } %819, 3, !dbg !82 + %824 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %287), !dbg !82 + %825 = extractvalue { i32, i32, i32, i32 } %824, 0, !dbg !82 + %826 = extractvalue { i32, i32, i32, i32 } %824, 1, !dbg !82 + %827 = extractvalue { i32, i32, i32, i32 } %824, 2, !dbg !82 + %828 = extractvalue { i32, i32, i32, i32 } %824, 3, !dbg !82 + %829 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %288), !dbg !82 + %830 = extractvalue { i32, i32, i32, i32 } %829, 0, !dbg !82 + %831 = extractvalue { i32, i32, i32, i32 } %829, 1, !dbg !82 + %832 = extractvalue { i32, i32, i32, i32 } %829, 2, !dbg !82 + %833 = extractvalue { i32, i32, i32, i32 } %829, 3, !dbg !82 + %834 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %289), !dbg !82 + %835 = extractvalue { i32, i32, i32, i32 } %834, 0, !dbg !82 + %836 = extractvalue { i32, i32, i32, i32 } %834, 1, !dbg !82 + %837 = extractvalue { i32, i32, i32, i32 } %834, 2, !dbg !82 + %838 = extractvalue { i32, i32, i32, i32 } %834, 3, !dbg !82 + %839 = extractelement <4 x float> %368, i64 0, !dbg !84 + %840 = extractelement <4 x float> %368, i64 1, !dbg !84 + %841 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %839, float %840, float %352, float %353, i32 %820, i32 %821, i32 %822, i32 %823, i32 %800, i32 %801) #2, !dbg !84 + %842 = extractvalue { float, float, float, float } %841, 0, !dbg !84 + %843 = extractvalue { float, float, float, float } %841, 1, !dbg !84 + %844 = extractvalue { float, float, float, float } %841, 2, !dbg !84 + %845 = extractvalue { float, float, float, float } %841, 3, !dbg !84 + %846 = extractelement <4 x float> %368, i64 2, !dbg !84 + %847 = extractelement <4 x float> %368, i64 3, !dbg !84 + %848 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %354, float %355, float %846, float %847, i32 %820, i32 %821, i32 %822, i32 %823, i32 %810, i32 %811) #2, !dbg !84 + %849 = extractvalue { float, float, float, float } %848, 0, !dbg !84 + %850 = extractvalue { float, float, float, float } %848, 1, !dbg !84 + %851 = extractvalue { float, float, float, float } %848, 2, !dbg !84 + %852 = extractvalue { float, float, float, float } %848, 3, !dbg !84 + %853 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %842, float %843, float %844, float %845, i32 %825, i32 %826, i32 %827, i32 %828, i32 %802, i32 %803) #2, !dbg !84 + %854 = extractvalue { float, float, float, float } %853, 0, !dbg !84 + %855 = extractvalue { float, float, float, float } %853, 1, !dbg !84 + %856 = extractvalue { float, float, float, float } %853, 2, !dbg !84 + %857 = extractvalue { float, float, float, float } %853, 3, !dbg !84 + %858 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %849, float %850, float %851, float %852, i32 %825, i32 %826, i32 %827, i32 %828, i32 %812, i32 %813) #2, !dbg !84 + %859 = extractvalue { float, float, float, float } %858, 0, !dbg !84 + %860 = extractvalue { float, float, float, float } %858, 1, !dbg !84 + %861 = extractvalue { float, float, float, float } %858, 2, !dbg !84 + %862 = extractvalue { float, float, float, float } %858, 3, !dbg !84 + %863 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %854, float %855, float %856, float %857, i32 %830, i32 %831, i32 %832, i32 %833, i32 %805, i32 %806) #2, !dbg !84 + %864 = extractvalue { float, float, float, float } %863, 0, !dbg !84 + %865 = extractvalue { float, float, float, float } %863, 1, !dbg !84 + %866 = extractvalue { float, float, float, float } %863, 2, !dbg !84 + %867 = extractvalue { float, float, float, float } %863, 3, !dbg !84 + %868 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %859, float %860, float %861, float %862, i32 %830, i32 %831, i32 %832, i32 %833, i32 %815, i32 %816) #2, !dbg !84 + %869 = extractvalue { float, float, float, float } %868, 0, !dbg !84 + %870 = extractvalue { float, float, float, float } %868, 1, !dbg !84 + %871 = extractvalue { float, float, float, float } %868, 2, !dbg !84 + %872 = extractvalue { float, float, float, float } %868, 3, !dbg !84 + %873 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %864, float %865, float %866, float %867, i32 %835, i32 %836, i32 %837, i32 %838, i32 %807, i32 %808) #2, !dbg !84 + %874 = extractvalue { float, float, float, float } %873, 0, !dbg !84 + %875 = extractvalue { float, float, float, float } %873, 1, !dbg !84 + %876 = extractvalue { float, float, float, float } %873, 2, !dbg !84 + %877 = extractvalue { float, float, float, float } %873, 3, !dbg !84 + %878 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %869, float %870, float %871, float %872, i32 %835, i32 %836, i32 %837, i32 %838, i32 %817, i32 %818) #2, !dbg !84 + %879 = extractvalue { float, float, float, float } %878, 0, !dbg !84 + %880 = extractvalue { float, float, float, float } %878, 1, !dbg !84 + %881 = extractvalue { float, float, float, float } %878, 2, !dbg !84 + %882 = extractvalue { float, float, float, float } %878, 3, !dbg !84 + %883 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %320, float %321, float %322, float %323, i32 %820, i32 %821, i32 %822, i32 %823, i32 %464, i32 %463) #2, !dbg !85 + %884 = extractvalue { float, float, float, float } %883, 0, !dbg !85 + %885 = extractvalue { float, float, float, float } %883, 1, !dbg !85 + %886 = extractvalue { float, float, float, float } %883, 2, !dbg !85 + %887 = extractvalue { float, float, float, float } %883, 3, !dbg !85 + %888 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %324, float %325, float %326, float %327, i32 %820, i32 %821, i32 %822, i32 %823, i32 %460, i32 %459) #2, !dbg !85 + %889 = extractvalue { float, float, float, float } %888, 0, !dbg !85 + %890 = extractvalue { float, float, float, float } %888, 1, !dbg !85 + %891 = extractvalue { float, float, float, float } %888, 2, !dbg !85 + %892 = extractvalue { float, float, float, float } %888, 3, !dbg !85 + %893 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %884, float %885, float %886, float %887, i32 %825, i32 %826, i32 %827, i32 %828, i32 %462, i32 %461) #2, !dbg !85 + %894 = extractvalue { float, float, float, float } %893, 0, !dbg !85 + %895 = extractvalue { float, float, float, float } %893, 1, !dbg !85 + %896 = extractvalue { float, float, float, float } %893, 2, !dbg !85 + %897 = extractvalue { float, float, float, float } %893, 3, !dbg !85 + %898 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %889, float %890, float %891, float %892, i32 %825, i32 %826, i32 %827, i32 %828, i32 %458, i32 %457) #2, !dbg !85 + %899 = extractvalue { float, float, float, float } %898, 0, !dbg !85 + %900 = extractvalue { float, float, float, float } %898, 1, !dbg !85 + %901 = extractvalue { float, float, float, float } %898, 2, !dbg !85 + %902 = extractvalue { float, float, float, float } %898, 3, !dbg !85 + %903 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %894, float %895, float %896, float %897, i32 %830, i32 %831, i32 %832, i32 %833, i32 %456, i32 %455) #2, !dbg !85 + %904 = extractvalue { float, float, float, float } %903, 0, !dbg !85 + %905 = extractvalue { float, float, float, float } %903, 1, !dbg !85 + %906 = extractvalue { float, float, float, float } %903, 2, !dbg !85 + %907 = extractvalue { float, float, float, float } %903, 3, !dbg !85 + %908 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %899, float %900, float %901, float %902, i32 %830, i32 %831, i32 %832, i32 %833, i32 %452, i32 %451) #2, !dbg !85 + %909 = extractvalue { float, float, float, float } %908, 0, !dbg !85 + %910 = extractvalue { float, float, float, float } %908, 1, !dbg !85 + %911 = extractvalue { float, float, float, float } %908, 2, !dbg !85 + %912 = extractvalue { float, float, float, float } %908, 3, !dbg !85 + %913 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %904, float %905, float %906, float %907, i32 %835, i32 %836, i32 %837, i32 %838, i32 %454, i32 %453) #2, !dbg !85 + %914 = extractvalue { float, float, float, float } %913, 0, !dbg !85 + %915 = extractvalue { float, float, float, float } %913, 1, !dbg !85 + %916 = extractvalue { float, float, float, float } %913, 2, !dbg !85 + %917 = extractvalue { float, float, float, float } %913, 3, !dbg !85 + %918 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %909, float %910, float %911, float %912, i32 %835, i32 %836, i32 %837, i32 %838, i32 %450, i32 %449) #2, !dbg !85 + %919 = extractvalue { float, float, float, float } %918, 0, !dbg !85 + %920 = extractvalue { float, float, float, float } %918, 1, !dbg !85 + %921 = extractvalue { float, float, float, float } %918, 2, !dbg !85 + %922 = extractvalue { float, float, float, float } %918, 3, !dbg !85 + %923 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %312, float %313, float %314, float %315, i32 %820, i32 %821, i32 %822, i32 %823, i32 %478, i32 %479) #2, !dbg !86 + %924 = extractvalue { float, float, float, float } %923, 0, !dbg !86 + %925 = extractvalue { float, float, float, float } %923, 1, !dbg !86 + %926 = extractvalue { float, float, float, float } %923, 2, !dbg !86 + %927 = extractvalue { float, float, float, float } %923, 3, !dbg !86 + %928 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %316, float %317, float %318, float %319, i32 %820, i32 %821, i32 %822, i32 %823, i32 %488, i32 %489) #2, !dbg !86 + %929 = extractvalue { float, float, float, float } %928, 0, !dbg !86 + %930 = extractvalue { float, float, float, float } %928, 1, !dbg !86 + %931 = extractvalue { float, float, float, float } %928, 2, !dbg !86 + %932 = extractvalue { float, float, float, float } %928, 3, !dbg !86 + %933 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %924, float %925, float %926, float %927, i32 %825, i32 %826, i32 %827, i32 %828, i32 %480, i32 %481) #2, !dbg !86 + %934 = extractvalue { float, float, float, float } %933, 0, !dbg !86 + %935 = extractvalue { float, float, float, float } %933, 1, !dbg !86 + %936 = extractvalue { float, float, float, float } %933, 2, !dbg !86 + %937 = extractvalue { float, float, float, float } %933, 3, !dbg !86 + %938 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %929, float %930, float %931, float %932, i32 %825, i32 %826, i32 %827, i32 %828, i32 %490, i32 %491) #2, !dbg !86 + %939 = extractvalue { float, float, float, float } %938, 0, !dbg !86 + %940 = extractvalue { float, float, float, float } %938, 1, !dbg !86 + %941 = extractvalue { float, float, float, float } %938, 2, !dbg !86 + %942 = extractvalue { float, float, float, float } %938, 3, !dbg !86 + %943 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %934, float %935, float %936, float %937, i32 %830, i32 %831, i32 %832, i32 %833, i32 %483, i32 %484) #2, !dbg !86 + %944 = extractvalue { float, float, float, float } %943, 0, !dbg !86 + %945 = extractvalue { float, float, float, float } %943, 1, !dbg !86 + %946 = extractvalue { float, float, float, float } %943, 2, !dbg !86 + %947 = extractvalue { float, float, float, float } %943, 3, !dbg !86 + %948 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %939, float %940, float %941, float %942, i32 %830, i32 %831, i32 %832, i32 %833, i32 %493, i32 %494) #2, !dbg !86 + %949 = extractvalue { float, float, float, float } %948, 0, !dbg !86 + %950 = extractvalue { float, float, float, float } %948, 1, !dbg !86 + %951 = extractvalue { float, float, float, float } %948, 2, !dbg !86 + %952 = extractvalue { float, float, float, float } %948, 3, !dbg !86 + %953 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %944, float %945, float %946, float %947, i32 %835, i32 %836, i32 %837, i32 %838, i32 %485, i32 %486) #2, !dbg !86 + %954 = extractvalue { float, float, float, float } %953, 0, !dbg !86 + %955 = extractvalue { float, float, float, float } %953, 1, !dbg !86 + %956 = extractvalue { float, float, float, float } %953, 2, !dbg !86 + %957 = extractvalue { float, float, float, float } %953, 3, !dbg !86 + %958 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %949, float %950, float %951, float %952, i32 %835, i32 %836, i32 %837, i32 %838, i32 %495, i32 %496) #2, !dbg !86 + %959 = extractvalue { float, float, float, float } %958, 0, !dbg !86 + %960 = extractvalue { float, float, float, float } %958, 1, !dbg !86 + %961 = extractvalue { float, float, float, float } %958, 2, !dbg !86 + %962 = extractvalue { float, float, float, float } %958, 3, !dbg !86 + %963 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %304, float %305, float %306, float %307, i32 %820, i32 %821, i32 %822, i32 %823, i32 %619, i32 %620) #2, !dbg !87 + %964 = extractvalue { float, float, float, float } %963, 0, !dbg !87 + %965 = extractvalue { float, float, float, float } %963, 1, !dbg !87 + %966 = extractvalue { float, float, float, float } %963, 2, !dbg !87 + %967 = extractvalue { float, float, float, float } %963, 3, !dbg !87 + %968 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %308, float %309, float %310, float %311, i32 %820, i32 %821, i32 %822, i32 %823, i32 %629, i32 %630) #2, !dbg !87 + %969 = extractvalue { float, float, float, float } %968, 0, !dbg !87 + %970 = extractvalue { float, float, float, float } %968, 1, !dbg !87 + %971 = extractvalue { float, float, float, float } %968, 2, !dbg !87 + %972 = extractvalue { float, float, float, float } %968, 3, !dbg !87 + %973 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %964, float %965, float %966, float %967, i32 %825, i32 %826, i32 %827, i32 %828, i32 %621, i32 %622) #2, !dbg !87 + %974 = extractvalue { float, float, float, float } %973, 0, !dbg !87 + %975 = extractvalue { float, float, float, float } %973, 1, !dbg !87 + %976 = extractvalue { float, float, float, float } %973, 2, !dbg !87 + %977 = extractvalue { float, float, float, float } %973, 3, !dbg !87 + %978 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %969, float %970, float %971, float %972, i32 %825, i32 %826, i32 %827, i32 %828, i32 %631, i32 %632) #2, !dbg !87 + %979 = extractvalue { float, float, float, float } %978, 0, !dbg !87 + %980 = extractvalue { float, float, float, float } %978, 1, !dbg !87 + %981 = extractvalue { float, float, float, float } %978, 2, !dbg !87 + %982 = extractvalue { float, float, float, float } %978, 3, !dbg !87 + %983 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %974, float %975, float %976, float %977, i32 %830, i32 %831, i32 %832, i32 %833, i32 %624, i32 %625) #2, !dbg !87 + %984 = extractvalue { float, float, float, float } %983, 0, !dbg !87 + %985 = extractvalue { float, float, float, float } %983, 1, !dbg !87 + %986 = extractvalue { float, float, float, float } %983, 2, !dbg !87 + %987 = extractvalue { float, float, float, float } %983, 3, !dbg !87 + %988 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %979, float %980, float %981, float %982, i32 %830, i32 %831, i32 %832, i32 %833, i32 %634, i32 %635) #2, !dbg !87 + %989 = extractvalue { float, float, float, float } %988, 0, !dbg !87 + %990 = extractvalue { float, float, float, float } %988, 1, !dbg !87 + %991 = extractvalue { float, float, float, float } %988, 2, !dbg !87 + %992 = extractvalue { float, float, float, float } %988, 3, !dbg !87 + %993 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %984, float %985, float %986, float %987, i32 %835, i32 %836, i32 %837, i32 %838, i32 %626, i32 %627) #2, !dbg !87 + %994 = extractvalue { float, float, float, float } %993, 0, !dbg !87 + %995 = extractvalue { float, float, float, float } %993, 1, !dbg !87 + %996 = extractvalue { float, float, float, float } %993, 2, !dbg !87 + %997 = extractvalue { float, float, float, float } %993, 3, !dbg !87 + %998 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %989, float %990, float %991, float %992, i32 %835, i32 %836, i32 %837, i32 %838, i32 %636, i32 %637) #2, !dbg !87 + %999 = extractvalue { float, float, float, float } %998, 0, !dbg !87 + %1000 = extractvalue { float, float, float, float } %998, 1, !dbg !87 + %1001 = extractvalue { float, float, float, float } %998, 2, !dbg !87 + %1002 = extractvalue { float, float, float, float } %998, 3, !dbg !87 + %1003 = insertelement <4 x float> poison, float %874, i64 0, !dbg !73 + %1004 = insertelement <4 x float> %1003, float %875, i64 1, !dbg !73 + %1005 = insertelement <4 x float> %1004, float %881, i64 2, !dbg !73 + %1006 = insertelement <4 x float> %1005, float %882, i64 3, !dbg !73 + br label %1007, !dbg !80 + +1007: ; preds = %786, %605, %448, %300 + %1008 = phi float [ %360, %300 ], [ %554, %448 ], [ %554, %605 ], [ %554, %786 ], !dbg !71 + %1009 = phi float [ %361, %300 ], [ %555, %448 ], [ %555, %605 ], [ %555, %786 ], !dbg !71 + %1010 = phi float [ %362, %300 ], [ %557, %448 ], [ %557, %605 ], [ %557, %786 ], !dbg !71 + %1011 = phi float [ %363, %300 ], [ %558, %448 ], [ %558, %605 ], [ %558, %786 ], !dbg !71 + %1012 = phi float [ %356, %300 ], [ %356, %448 ], [ %695, %605 ], [ %695, %786 ], !dbg !73 + %1013 = phi float [ %357, %300 ], [ %357, %448 ], [ %696, %605 ], [ %696, %786 ], !dbg !73 + %1014 = phi float [ %358, %300 ], [ %358, %448 ], [ %698, %605 ], [ %698, %786 ], !dbg !73 + %1015 = phi float [ %359, %300 ], [ %359, %448 ], [ %699, %605 ], [ %699, %786 ], !dbg !73 + %1016 = phi float [ %352, %300 ], [ %352, %448 ], [ %352, %605 ], [ %876, %786 ], !dbg !73 + %1017 = phi float [ %353, %300 ], [ %353, %448 ], [ %353, %605 ], [ %877, %786 ], !dbg !73 + %1018 = phi float [ %354, %300 ], [ %354, %448 ], [ %354, %605 ], [ %879, %786 ], !dbg !73 + %1019 = phi float [ %355, %300 ], [ %355, %448 ], [ %355, %605 ], [ %880, %786 ], !dbg !73 + %1020 = phi float [ %344, %300 ], [ %592, %448 ], [ %592, %605 ], [ %592, %786 ], !dbg !72 + %1021 = phi float [ %345, %300 ], [ %593, %448 ], [ %593, %605 ], [ %593, %786 ], !dbg !72 + %1022 = phi float [ %346, %300 ], [ %594, %448 ], [ %594, %605 ], [ %594, %786 ], !dbg !72 + %1023 = phi float [ %347, %300 ], [ %595, %448 ], [ %595, %605 ], [ %595, %786 ], !dbg !72 + %1024 = phi float [ %348, %300 ], [ %597, %448 ], [ %597, %605 ], [ %597, %786 ], !dbg !72 + %1025 = phi float [ %349, %300 ], [ %598, %448 ], [ %598, %605 ], [ %598, %786 ], !dbg !72 + %1026 = phi float [ %350, %300 ], [ %599, %448 ], [ %599, %605 ], [ %599, %786 ], !dbg !72 + %1027 = phi float [ %351, %300 ], [ %600, %448 ], [ %600, %605 ], [ %600, %786 ], !dbg !72 + %1028 = phi float [ %336, %300 ], [ %336, %448 ], [ %733, %605 ], [ %733, %786 ], !dbg !73 + %1029 = phi float [ %337, %300 ], [ %337, %448 ], [ %734, %605 ], [ %734, %786 ], !dbg !73 + %1030 = phi float [ %338, %300 ], [ %338, %448 ], [ %735, %605 ], [ %735, %786 ], !dbg !73 + %1031 = phi float [ %339, %300 ], [ %339, %448 ], [ %736, %605 ], [ %736, %786 ], !dbg !73 + %1032 = phi float [ %340, %300 ], [ %340, %448 ], [ %738, %605 ], [ %738, %786 ], !dbg !73 + %1033 = phi float [ %341, %300 ], [ %341, %448 ], [ %739, %605 ], [ %739, %786 ], !dbg !73 + %1034 = phi float [ %342, %300 ], [ %342, %448 ], [ %740, %605 ], [ %740, %786 ], !dbg !73 + %1035 = phi float [ %343, %300 ], [ %343, %448 ], [ %741, %605 ], [ %741, %786 ], !dbg !73 + %1036 = phi float [ %328, %300 ], [ %328, %448 ], [ %773, %605 ], [ %773, %786 ], !dbg !73 + %1037 = phi float [ %329, %300 ], [ %329, %448 ], [ %774, %605 ], [ %774, %786 ], !dbg !73 + %1038 = phi float [ %330, %300 ], [ %330, %448 ], [ %775, %605 ], [ %775, %786 ], !dbg !73 + %1039 = phi float [ %331, %300 ], [ %331, %448 ], [ %776, %605 ], [ %776, %786 ], !dbg !73 + %1040 = phi float [ %332, %300 ], [ %332, %448 ], [ %778, %605 ], [ %778, %786 ], !dbg !73 + %1041 = phi float [ %333, %300 ], [ %333, %448 ], [ %779, %605 ], [ %779, %786 ], !dbg !73 + %1042 = phi float [ %334, %300 ], [ %334, %448 ], [ %780, %605 ], [ %780, %786 ], !dbg !73 + %1043 = phi float [ %335, %300 ], [ %335, %448 ], [ %781, %605 ], [ %781, %786 ], !dbg !73 + %1044 = phi float [ %320, %300 ], [ %320, %448 ], [ %320, %605 ], [ %914, %786 ], !dbg !73 + %1045 = phi float [ %321, %300 ], [ %321, %448 ], [ %321, %605 ], [ %915, %786 ], !dbg !73 + %1046 = phi float [ %322, %300 ], [ %322, %448 ], [ %322, %605 ], [ %916, %786 ], !dbg !73 + %1047 = phi float [ %323, %300 ], [ %323, %448 ], [ %323, %605 ], [ %917, %786 ], !dbg !73 + %1048 = phi float [ %324, %300 ], [ %324, %448 ], [ %324, %605 ], [ %919, %786 ], !dbg !73 + %1049 = phi float [ %325, %300 ], [ %325, %448 ], [ %325, %605 ], [ %920, %786 ], !dbg !73 + %1050 = phi float [ %326, %300 ], [ %326, %448 ], [ %326, %605 ], [ %921, %786 ], !dbg !73 + %1051 = phi float [ %327, %300 ], [ %327, %448 ], [ %327, %605 ], [ %922, %786 ], !dbg !73 + %1052 = phi float [ %312, %300 ], [ %312, %448 ], [ %312, %605 ], [ %954, %786 ], !dbg !73 + %1053 = phi float [ %313, %300 ], [ %313, %448 ], [ %313, %605 ], [ %955, %786 ], !dbg !73 + %1054 = phi float [ %314, %300 ], [ %314, %448 ], [ %314, %605 ], [ %956, %786 ], !dbg !73 + %1055 = phi float [ %315, %300 ], [ %315, %448 ], [ %315, %605 ], [ %957, %786 ], !dbg !73 + %1056 = phi float [ %316, %300 ], [ %316, %448 ], [ %316, %605 ], [ %959, %786 ], !dbg !73 + %1057 = phi float [ %317, %300 ], [ %317, %448 ], [ %317, %605 ], [ %960, %786 ], !dbg !73 + %1058 = phi float [ %318, %300 ], [ %318, %448 ], [ %318, %605 ], [ %961, %786 ], !dbg !73 + %1059 = phi float [ %319, %300 ], [ %319, %448 ], [ %319, %605 ], [ %962, %786 ], !dbg !73 + %1060 = phi float [ %304, %300 ], [ %304, %448 ], [ %304, %605 ], [ %994, %786 ], !dbg !73 + %1061 = phi float [ %305, %300 ], [ %305, %448 ], [ %305, %605 ], [ %995, %786 ], !dbg !73 + %1062 = phi float [ %306, %300 ], [ %306, %448 ], [ %306, %605 ], [ %996, %786 ], !dbg !73 + %1063 = phi float [ %307, %300 ], [ %307, %448 ], [ %307, %605 ], [ %997, %786 ], !dbg !73 + %1064 = phi float [ %308, %300 ], [ %308, %448 ], [ %308, %605 ], [ %999, %786 ], !dbg !73 + %1065 = phi float [ %309, %300 ], [ %309, %448 ], [ %309, %605 ], [ %1000, %786 ], !dbg !73 + %1066 = phi float [ %310, %300 ], [ %310, %448 ], [ %310, %605 ], [ %1001, %786 ], !dbg !73 + %1067 = phi float [ %311, %300 ], [ %311, %448 ], [ %311, %605 ], [ %1002, %786 ], !dbg !73 + %1068 = phi <4 x float> [ %368, %300 ], [ %368, %448 ], [ %368, %605 ], [ %1006, %786 ], !dbg !73 + %1069 = phi <4 x float> [ %369, %300 ], [ %369, %448 ], [ %785, %605 ], [ %785, %786 ], !dbg !73 + %1070 = phi <4 x float> [ %370, %300 ], [ %604, %448 ], [ %604, %605 ], [ %604, %786 ], !dbg !71 + %indvars.iv.next592 = add nuw nsw i64 %indvars.iv591, 64, !dbg !65 + br i1 %301, label %300, label %1071, !dbg !65 + +1071: ; preds = %1007 + %.idx7 = shl nuw nsw i64 %21, 7, !dbg !88 + %1072 = getelementptr i8, ptr addrspace(1) %3, i64 %.idx7, !dbg !88 + %1073 = lshr i32 %23, 2, !dbg !28 + %1074 = and i32 %1073, 15, !dbg !28 + %1075 = shl nuw nsw i32 %28, 2, !dbg !28 + %1076 = zext nneg i32 %1074 to i64, !dbg !29 + %1077 = or disjoint i64 %11, %1076, !dbg !29 + %1078 = icmp samesign ult i64 %1077, %10, !dbg !30 + %1079 = or disjoint i64 %17, %1076, !dbg !31 + %1080 = icmp samesign ult i64 %1079, %10, !dbg !32 + %1081 = or disjoint i64 %18, %1076, !dbg !33 + %1082 = icmp samesign ult i64 %1081, %10, !dbg !34 + %1083 = or disjoint i64 %19, %1076, !dbg !35 + %1084 = icmp samesign ult i64 %1083, %10, !dbg !36 + %1085 = insertelement <2 x i16> poison, i16 %144, i64 0, !dbg !47 + %1086 = insertelement <2 x i16> %1085, i16 %146, i64 1, !dbg !47 + %1087 = bitcast <2 x i16> %1086 to <2 x bfloat>, !dbg !47 + %1088 = shufflevector <2 x bfloat> %1087, <2 x bfloat> poison, <4 x i32> , !dbg !89 + %1089 = fpext <4 x bfloat> %1088 to <4 x float>, !dbg !89 + %1090 = insertelement <2 x i16> poison, i16 %147, i64 0, !dbg !48 + %1091 = insertelement <2 x i16> %1090, i16 %148, i64 1, !dbg !48 + %1092 = bitcast <2 x i16> %1091 to <2 x bfloat>, !dbg !48 + %1093 = shufflevector <2 x bfloat> %1092, <2 x bfloat> poison, <4 x i32> , !dbg !90 + %1094 = fpext <4 x bfloat> %1093 to <4 x float>, !dbg !90 + %1095 = insertelement <2 x i16> poison, i16 %149, i64 0, !dbg !49 + %1096 = insertelement <2 x i16> %1095, i16 %150, i64 1, !dbg !49 + %1097 = bitcast <2 x i16> %1096 to <2 x bfloat>, !dbg !49 + %1098 = shufflevector <2 x bfloat> %1097, <2 x bfloat> poison, <4 x i32> , !dbg !91 + %1099 = fpext <4 x bfloat> %1098 to <4 x float>, !dbg !91 + %1100 = insertelement <2 x i16> poison, i16 %151, i64 0, !dbg !50 + %1101 = insertelement <2 x i16> %1100, i16 %152, i64 1, !dbg !50 + %1102 = bitcast <2 x i16> %1101 to <2 x bfloat>, !dbg !50 + %1103 = shufflevector <2 x bfloat> %1102, <2 x bfloat> poison, <4 x i32> , !dbg !92 + %1104 = fpext <4 x bfloat> %1103 to <4 x float>, !dbg !92 + %1105 = bitcast i32 %194 to float, !dbg !46 + %1106 = bitcast i32 %195 to float, !dbg !46 + %1107 = bitcast i32 %197 to float, !dbg !46 + %1108 = bitcast i32 %199 to float, !dbg !46 + %1109 = bitcast i32 %200 to float, !dbg !46 + %1110 = bitcast i32 %201 to float, !dbg !46 + %1111 = bitcast i32 %202 to float, !dbg !46 + %1112 = bitcast i32 %203 to float, !dbg !46 + %1113 = bitcast i32 %204 to float, !dbg !56 + %1114 = bitcast i32 %205 to float, !dbg !56 + %1115 = bitcast i32 %207 to float, !dbg !56 + %1116 = bitcast i32 %209 to float, !dbg !56 + %1117 = bitcast i32 %211 to float, !dbg !56 + %1118 = bitcast i32 %213 to float, !dbg !56 + %1119 = bitcast i32 %214 to float, !dbg !57 + %1120 = bitcast i32 %215 to float, !dbg !57 + %1121 = bitcast i32 %217 to float, !dbg !57 + %1122 = bitcast i32 %219 to float, !dbg !57 + %1123 = bitcast i32 %221 to float, !dbg !57 + %1124 = bitcast i32 %223 to float, !dbg !57 + %1125 = bitcast i32 %224 to float, !dbg !58 + %1126 = bitcast i32 %225 to float, !dbg !58 + %1127 = bitcast i32 %227 to float, !dbg !58 + %1128 = bitcast i32 %229 to float, !dbg !58 + %1129 = bitcast i32 %231 to float, !dbg !58 + %1130 = bitcast i32 %233 to float, !dbg !58 + tail call void @llvm.nvvm.cp.async.wait.group(i32 0), !dbg !65 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !65 + %1131 = shufflevector <2 x i32> %85, <2 x i32> poison, <4 x i32> , !dbg !93 + %1132 = shufflevector <2 x i32> %73, <2 x i32> poison, <4 x i32> , !dbg !93 + %1133 = icmp samesign ugt <4 x i32> %1131, %1132, !dbg !93 + %1134 = insertelement <4 x i32> poison, i32 %82, i64 0, !dbg !93 + %1135 = insertelement <4 x i32> %1134, i32 %83, i64 1, !dbg !93 + %1136 = shufflevector <4 x i32> %1135, <4 x i32> poison, <4 x i32> , !dbg !93 + %1137 = insertelement <4 x i32> %56, i32 %54, i64 3, !dbg !93 + %1138 = shufflevector <4 x i32> %1137, <4 x i32> poison, <4 x i32> , !dbg !93 + %1139 = icmp samesign ugt <4 x i32> %1136, %1138, !dbg !93 + %1140 = icmp eq i32 %82, %49, !dbg !94 + %1141 = icmp eq i32 %82, %52, !dbg !94 + %1142 = icmp eq i32 %83, %54, !dbg !94 + %1143 = icmp eq i32 %82, %50, !dbg !94 + %1144 = icmp eq i32 %82, %51, !dbg !94 + %1145 = icmp eq i32 %83, %50, !dbg !94 + %1146 = icmp eq i32 %83, %51, !dbg !94 + %1147 = and <4 x i1> %1133, %91, !dbg !95 + %1148 = and <4 x i1> %81, %1147, !dbg !96 + %1149 = fsub float %1105, %1107, !dbg !97 + %1150 = fsub float %1105, %1108, !dbg !97 + %1151 = fsub float %1106, %1107, !dbg !97 + %1152 = fsub float %1106, %1108, !dbg !97 + %1153 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !98 + %.not.i = icmp eq i32 %1153, 0, !dbg !98 + br i1 %.not.i, label %1156, label %1154, !dbg !98 + +1154: ; preds = %1071 + %1155 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1149), !dbg !98 + br label %__nv_exp2f.exit, !dbg !98 + +1156: ; preds = %1071 + %1157 = tail call float @llvm.nvvm.ex2.approx.f32(float %1149), !dbg !98 + br label %__nv_exp2f.exit, !dbg !98 + +__nv_exp2f.exit: ; preds = %1154, %1156 + %.0.i = phi float [ %1155, %1154 ], [ %1157, %1156 ], !dbg !98 + %1158 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !98 + %.not.i251 = icmp eq i32 %1158, 0, !dbg !98 + br i1 %.not.i251, label %1161, label %1159, !dbg !98 + +1159: ; preds = %__nv_exp2f.exit + %1160 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1150), !dbg !98 + br label %__nv_exp2f.exit253, !dbg !98 + +1161: ; preds = %__nv_exp2f.exit + %1162 = tail call float @llvm.nvvm.ex2.approx.f32(float %1150), !dbg !98 + br label %__nv_exp2f.exit253, !dbg !98 + +__nv_exp2f.exit253: ; preds = %1159, %1161 + %.0.i252 = phi float [ %1160, %1159 ], [ %1162, %1161 ], !dbg !98 + %1163 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !98 + %.not.i254 = icmp eq i32 %1163, 0, !dbg !98 + br i1 %.not.i254, label %1166, label %1164, !dbg !98 + +1164: ; preds = %__nv_exp2f.exit253 + %1165 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1151), !dbg !98 + br label %__nv_exp2f.exit256, !dbg !98 + +1166: ; preds = %__nv_exp2f.exit253 + %1167 = tail call float @llvm.nvvm.ex2.approx.f32(float %1151), !dbg !98 + br label %__nv_exp2f.exit256, !dbg !98 + +__nv_exp2f.exit256: ; preds = %1164, %1166 + %.0.i255 = phi float [ %1165, %1164 ], [ %1167, %1166 ], !dbg !98 + %1168 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !98 + %.not.i257 = icmp eq i32 %1168, 0, !dbg !98 + br i1 %.not.i257, label %1171, label %1169, !dbg !98 + +1169: ; preds = %__nv_exp2f.exit256 + %1170 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1152), !dbg !98 + br label %__nv_exp2f.exit259, !dbg !98 + +1171: ; preds = %__nv_exp2f.exit256 + %1172 = tail call float @llvm.nvvm.ex2.approx.f32(float %1152), !dbg !98 + br label %__nv_exp2f.exit259, !dbg !98 + +__nv_exp2f.exit259: ; preds = %1169, %1171 + %.0.i258 = phi float [ %1170, %1169 ], [ %1172, %1171 ], !dbg !98 + %1173 = insertelement <4 x float> poison, float %.0.i, i64 0, !dbg !103 + %1174 = insertelement <4 x float> %1173, float %.0.i252, i64 1, !dbg !103 + %1175 = insertelement <4 x float> %1174, float %.0.i255, i64 2, !dbg !103 + %1176 = insertelement <4 x float> %1175, float %.0.i258, i64 3, !dbg !103 + %1177 = select <4 x i1> %1148, <4 x float> %1176, <4 x float> zeroinitializer, !dbg !103 + %1178 = insertelement <4 x float> poison, float %435, i64 0, !dbg !104 + %1179 = insertelement <4 x float> %1178, float %436, i64 1, !dbg !104 + %1180 = insertelement <4 x float> %1179, float %437, i64 2, !dbg !104 + %1181 = insertelement <4 x float> %1180, float %438, i64 3, !dbg !104 + %1182 = fmul <4 x float> %1181, %1177, !dbg !104 + %1183 = shufflevector <4 x i1> %1139, <4 x i1> poison, <4 x i32> , !dbg !105 + %1184 = insertelement <4 x i1> poison, i1 %100, i64 0, !dbg !105 + %1185 = insertelement <4 x i1> %1184, i1 %99, i64 1, !dbg !105 + %1186 = shufflevector <4 x i1> %1185, <4 x i1> poison, <4 x i32> , !dbg !105 + %1187 = and <4 x i1> %1183, %1186, !dbg !105 + %1188 = and i1 %99, %206, !dbg !106 + %1189 = and i1 %99, %208, !dbg !106 + %1190 = and <4 x i1> %104, %1187, !dbg !106 + %1191 = fsub float %1113, %1115, !dbg !107 + %1192 = fsub float %1113, %1116, !dbg !107 + %1193 = fsub float %1114, %1115, !dbg !107 + %1194 = fsub float %1114, %1116, !dbg !107 + %1195 = fsub float %1114, %1117, !dbg !107 + %1196 = fsub float %1114, %1118, !dbg !107 + %1197 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !108 + %.not.i260 = icmp eq i32 %1197, 0, !dbg !108 + br i1 %.not.i260, label %1200, label %1198, !dbg !108 + +1198: ; preds = %__nv_exp2f.exit259 + %1199 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1191), !dbg !108 + br label %__nv_exp2f.exit262, !dbg !108 + +1200: ; preds = %__nv_exp2f.exit259 + %1201 = tail call float @llvm.nvvm.ex2.approx.f32(float %1191), !dbg !108 + br label %__nv_exp2f.exit262, !dbg !108 + +__nv_exp2f.exit262: ; preds = %1198, %1200 + %.0.i261 = phi float [ %1199, %1198 ], [ %1201, %1200 ], !dbg !108 + %1202 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !108 + %.not.i263 = icmp eq i32 %1202, 0, !dbg !108 + br i1 %.not.i263, label %1205, label %1203, !dbg !108 + +1203: ; preds = %__nv_exp2f.exit262 + %1204 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1192), !dbg !108 + br label %__nv_exp2f.exit265, !dbg !108 + +1205: ; preds = %__nv_exp2f.exit262 + %1206 = tail call float @llvm.nvvm.ex2.approx.f32(float %1192), !dbg !108 + br label %__nv_exp2f.exit265, !dbg !108 + +__nv_exp2f.exit265: ; preds = %1203, %1205 + %.0.i264 = phi float [ %1204, %1203 ], [ %1206, %1205 ], !dbg !108 + %1207 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !108 + %.not.i266 = icmp eq i32 %1207, 0, !dbg !108 + br i1 %.not.i266, label %1210, label %1208, !dbg !108 + +1208: ; preds = %__nv_exp2f.exit265 + %1209 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1193), !dbg !108 + br label %__nv_exp2f.exit268, !dbg !108 + +1210: ; preds = %__nv_exp2f.exit265 + %1211 = tail call float @llvm.nvvm.ex2.approx.f32(float %1193), !dbg !108 + br label %__nv_exp2f.exit268, !dbg !108 + +__nv_exp2f.exit268: ; preds = %1208, %1210 + %.0.i267 = phi float [ %1209, %1208 ], [ %1211, %1210 ], !dbg !108 + %1212 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !108 + %.not.i269 = icmp eq i32 %1212, 0, !dbg !108 + br i1 %.not.i269, label %1215, label %1213, !dbg !108 + +1213: ; preds = %__nv_exp2f.exit268 + %1214 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1194), !dbg !108 + br label %__nv_exp2f.exit271, !dbg !108 + +1215: ; preds = %__nv_exp2f.exit268 + %1216 = tail call float @llvm.nvvm.ex2.approx.f32(float %1194), !dbg !108 + br label %__nv_exp2f.exit271, !dbg !108 + +__nv_exp2f.exit271: ; preds = %1213, %1215 + %.0.i270 = phi float [ %1214, %1213 ], [ %1216, %1215 ], !dbg !108 + %1217 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !108 + %1218 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !108 + %1219 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !108 + %.not.i278 = icmp eq i32 %1219, 0, !dbg !108 + br i1 %.not.i278, label %1222, label %1220, !dbg !108 + +1220: ; preds = %__nv_exp2f.exit271 + %1221 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1195), !dbg !108 + br label %__nv_exp2f.exit280, !dbg !108 + +1222: ; preds = %__nv_exp2f.exit271 + %1223 = tail call float @llvm.nvvm.ex2.approx.f32(float %1195), !dbg !108 + br label %__nv_exp2f.exit280, !dbg !108 + +__nv_exp2f.exit280: ; preds = %1220, %1222 + %.0.i279 = phi float [ %1221, %1220 ], [ %1223, %1222 ], !dbg !108 + %1224 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !108 + %.not.i281 = icmp eq i32 %1224, 0, !dbg !108 + br i1 %.not.i281, label %1227, label %1225, !dbg !108 + +1225: ; preds = %__nv_exp2f.exit280 + %1226 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1196), !dbg !108 + br label %__nv_exp2f.exit283, !dbg !108 + +1227: ; preds = %__nv_exp2f.exit280 + %1228 = tail call float @llvm.nvvm.ex2.approx.f32(float %1196), !dbg !108 + br label %__nv_exp2f.exit283, !dbg !108 + +__nv_exp2f.exit283: ; preds = %1225, %1227 + %.0.i282 = phi float [ %1226, %1225 ], [ %1228, %1227 ], !dbg !108 + %1229 = insertelement <4 x float> poison, float %.0.i261, i64 0, !dbg !110 + %1230 = insertelement <4 x float> %1229, float %.0.i264, i64 1, !dbg !110 + %1231 = insertelement <4 x float> %1230, float %.0.i279, i64 2, !dbg !110 + %1232 = insertelement <4 x float> %1231, float %.0.i282, i64 3, !dbg !110 + %1233 = select <4 x i1> %1190, <4 x float> %1232, <4 x float> zeroinitializer, !dbg !110 + %1234 = select i1 %1188, float %.0.i267, float 0.000000e+00, !dbg !110 + %1235 = select i1 %1189, float %.0.i270, float 0.000000e+00, !dbg !110 + %1236 = fmul float %1008, %1234, !dbg !111 + %1237 = fmul float %1009, %1235, !dbg !111 + %1238 = fmul float %1010, 0.000000e+00, !dbg !111 + %1239 = fmul float %1011, 0.000000e+00, !dbg !111 + %1240 = fmul <4 x float> %1070, %1233, !dbg !111 + %1241 = insertelement <4 x i1> poison, i1 %111, i64 0, !dbg !112 + %1242 = insertelement <4 x i1> %1241, i1 %110, i64 1, !dbg !112 + %1243 = shufflevector <4 x i1> %1242, <4 x i1> poison, <4 x i32> , !dbg !112 + %1244 = and <4 x i1> %1183, %1243, !dbg !112 + %1245 = and i1 %110, %216, !dbg !113 + %1246 = and i1 %110, %218, !dbg !113 + %1247 = and <4 x i1> %113, %1244, !dbg !113 + %1248 = fsub float %1119, %1121, !dbg !114 + %1249 = fsub float %1119, %1122, !dbg !114 + %1250 = fsub float %1120, %1121, !dbg !114 + %1251 = fsub float %1120, %1122, !dbg !114 + %1252 = fsub float %1120, %1123, !dbg !114 + %1253 = fsub float %1120, %1124, !dbg !114 + %1254 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !115 + %.not.i284 = icmp eq i32 %1254, 0, !dbg !115 + br i1 %.not.i284, label %1257, label %1255, !dbg !115 + +1255: ; preds = %__nv_exp2f.exit283 + %1256 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1248), !dbg !115 + br label %__nv_exp2f.exit286, !dbg !115 + +1257: ; preds = %__nv_exp2f.exit283 + %1258 = tail call float @llvm.nvvm.ex2.approx.f32(float %1248), !dbg !115 + br label %__nv_exp2f.exit286, !dbg !115 + +__nv_exp2f.exit286: ; preds = %1255, %1257 + %.0.i285 = phi float [ %1256, %1255 ], [ %1258, %1257 ], !dbg !115 + %1259 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !115 + %.not.i287 = icmp eq i32 %1259, 0, !dbg !115 + br i1 %.not.i287, label %1262, label %1260, !dbg !115 + +1260: ; preds = %__nv_exp2f.exit286 + %1261 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1249), !dbg !115 + br label %__nv_exp2f.exit289, !dbg !115 + +1262: ; preds = %__nv_exp2f.exit286 + %1263 = tail call float @llvm.nvvm.ex2.approx.f32(float %1249), !dbg !115 + br label %__nv_exp2f.exit289, !dbg !115 + +__nv_exp2f.exit289: ; preds = %1260, %1262 + %.0.i288 = phi float [ %1261, %1260 ], [ %1263, %1262 ], !dbg !115 + %1264 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !115 + %.not.i290 = icmp eq i32 %1264, 0, !dbg !115 + br i1 %.not.i290, label %1267, label %1265, !dbg !115 + +1265: ; preds = %__nv_exp2f.exit289 + %1266 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1250), !dbg !115 + br label %__nv_exp2f.exit292, !dbg !115 + +1267: ; preds = %__nv_exp2f.exit289 + %1268 = tail call float @llvm.nvvm.ex2.approx.f32(float %1250), !dbg !115 + br label %__nv_exp2f.exit292, !dbg !115 + +__nv_exp2f.exit292: ; preds = %1265, %1267 + %.0.i291 = phi float [ %1266, %1265 ], [ %1268, %1267 ], !dbg !115 + %1269 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !115 + %.not.i293 = icmp eq i32 %1269, 0, !dbg !115 + br i1 %.not.i293, label %1272, label %1270, !dbg !115 + +1270: ; preds = %__nv_exp2f.exit292 + %1271 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1251), !dbg !115 + br label %__nv_exp2f.exit295, !dbg !115 + +1272: ; preds = %__nv_exp2f.exit292 + %1273 = tail call float @llvm.nvvm.ex2.approx.f32(float %1251), !dbg !115 + br label %__nv_exp2f.exit295, !dbg !115 + +__nv_exp2f.exit295: ; preds = %1270, %1272 + %.0.i294 = phi float [ %1271, %1270 ], [ %1273, %1272 ], !dbg !115 + %1274 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !115 + %1275 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !115 + %1276 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !115 + %.not.i302 = icmp eq i32 %1276, 0, !dbg !115 + br i1 %.not.i302, label %1279, label %1277, !dbg !115 + +1277: ; preds = %__nv_exp2f.exit295 + %1278 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1252), !dbg !115 + br label %__nv_exp2f.exit304, !dbg !115 + +1279: ; preds = %__nv_exp2f.exit295 + %1280 = tail call float @llvm.nvvm.ex2.approx.f32(float %1252), !dbg !115 + br label %__nv_exp2f.exit304, !dbg !115 + +__nv_exp2f.exit304: ; preds = %1277, %1279 + %.0.i303 = phi float [ %1278, %1277 ], [ %1280, %1279 ], !dbg !115 + %1281 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !115 + %.not.i305 = icmp eq i32 %1281, 0, !dbg !115 + br i1 %.not.i305, label %1284, label %1282, !dbg !115 + +1282: ; preds = %__nv_exp2f.exit304 + %1283 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1253), !dbg !115 + br label %__nv_exp2f.exit307, !dbg !115 + +1284: ; preds = %__nv_exp2f.exit304 + %1285 = tail call float @llvm.nvvm.ex2.approx.f32(float %1253), !dbg !115 + br label %__nv_exp2f.exit307, !dbg !115 + +__nv_exp2f.exit307: ; preds = %1282, %1284 + %.0.i306 = phi float [ %1283, %1282 ], [ %1285, %1284 ], !dbg !115 + %1286 = insertelement <4 x float> poison, float %.0.i285, i64 0, !dbg !117 + %1287 = insertelement <4 x float> %1286, float %.0.i288, i64 1, !dbg !117 + %1288 = insertelement <4 x float> %1287, float %.0.i303, i64 2, !dbg !117 + %1289 = insertelement <4 x float> %1288, float %.0.i306, i64 3, !dbg !117 + %1290 = select <4 x i1> %1247, <4 x float> %1289, <4 x float> zeroinitializer, !dbg !117 + %1291 = select i1 %1245, float %.0.i291, float 0.000000e+00, !dbg !117 + %1292 = select i1 %1246, float %.0.i294, float 0.000000e+00, !dbg !117 + %1293 = fmul float %1012, %1291, !dbg !118 + %1294 = fmul float %1013, %1292, !dbg !118 + %1295 = fmul float %1014, 0.000000e+00, !dbg !118 + %1296 = fmul float %1015, 0.000000e+00, !dbg !118 + %1297 = fmul <4 x float> %1069, %1290, !dbg !118 + %1298 = insertelement <4 x i1> poison, i1 %121, i64 0, !dbg !119 + %1299 = insertelement <4 x i1> %1298, i1 %120, i64 1, !dbg !119 + %1300 = shufflevector <4 x i1> %1299, <4 x i1> poison, <4 x i32> , !dbg !119 + %1301 = and <4 x i1> %1139, %1300, !dbg !119 + %1302 = and i1 %120, %226, !dbg !120 + %1303 = and i1 %120, %228, !dbg !120 + %1304 = and <4 x i1> %123, %1301, !dbg !120 + %1305 = fsub float %1125, %1127, !dbg !121 + %1306 = fsub float %1125, %1128, !dbg !121 + %1307 = fsub float %1126, %1127, !dbg !121 + %1308 = fsub float %1126, %1128, !dbg !121 + %1309 = fsub float %1126, %1129, !dbg !121 + %1310 = fsub float %1126, %1130, !dbg !121 + %1311 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !122 + %.not.i308 = icmp eq i32 %1311, 0, !dbg !122 + br i1 %.not.i308, label %1314, label %1312, !dbg !122 + +1312: ; preds = %__nv_exp2f.exit307 + %1313 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1305), !dbg !122 + br label %__nv_exp2f.exit310, !dbg !122 + +1314: ; preds = %__nv_exp2f.exit307 + %1315 = tail call float @llvm.nvvm.ex2.approx.f32(float %1305), !dbg !122 + br label %__nv_exp2f.exit310, !dbg !122 + +__nv_exp2f.exit310: ; preds = %1312, %1314 + %.0.i309 = phi float [ %1313, %1312 ], [ %1315, %1314 ], !dbg !122 + %1316 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !122 + %.not.i311 = icmp eq i32 %1316, 0, !dbg !122 + br i1 %.not.i311, label %1319, label %1317, !dbg !122 + +1317: ; preds = %__nv_exp2f.exit310 + %1318 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1306), !dbg !122 + br label %__nv_exp2f.exit313, !dbg !122 + +1319: ; preds = %__nv_exp2f.exit310 + %1320 = tail call float @llvm.nvvm.ex2.approx.f32(float %1306), !dbg !122 + br label %__nv_exp2f.exit313, !dbg !122 + +__nv_exp2f.exit313: ; preds = %1317, %1319 + %.0.i312 = phi float [ %1318, %1317 ], [ %1320, %1319 ], !dbg !122 + %1321 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !122 + %.not.i314 = icmp eq i32 %1321, 0, !dbg !122 + br i1 %.not.i314, label %1324, label %1322, !dbg !122 + +1322: ; preds = %__nv_exp2f.exit313 + %1323 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1307), !dbg !122 + br label %__nv_exp2f.exit316, !dbg !122 + +1324: ; preds = %__nv_exp2f.exit313 + %1325 = tail call float @llvm.nvvm.ex2.approx.f32(float %1307), !dbg !122 + br label %__nv_exp2f.exit316, !dbg !122 + +__nv_exp2f.exit316: ; preds = %1322, %1324 + %.0.i315 = phi float [ %1323, %1322 ], [ %1325, %1324 ], !dbg !122 + %1326 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !122 + %.not.i317 = icmp eq i32 %1326, 0, !dbg !122 + br i1 %.not.i317, label %1329, label %1327, !dbg !122 + +1327: ; preds = %__nv_exp2f.exit316 + %1328 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1308), !dbg !122 + br label %__nv_exp2f.exit319, !dbg !122 + +1329: ; preds = %__nv_exp2f.exit316 + %1330 = tail call float @llvm.nvvm.ex2.approx.f32(float %1308), !dbg !122 + br label %__nv_exp2f.exit319, !dbg !122 + +__nv_exp2f.exit319: ; preds = %1327, %1329 + %.0.i318 = phi float [ %1328, %1327 ], [ %1330, %1329 ], !dbg !122 + %1331 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !122 + %1332 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !122 + %1333 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !122 + %.not.i326 = icmp eq i32 %1333, 0, !dbg !122 + br i1 %.not.i326, label %1336, label %1334, !dbg !122 + +1334: ; preds = %__nv_exp2f.exit319 + %1335 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1309), !dbg !122 + br label %__nv_exp2f.exit328, !dbg !122 + +1336: ; preds = %__nv_exp2f.exit319 + %1337 = tail call float @llvm.nvvm.ex2.approx.f32(float %1309), !dbg !122 + br label %__nv_exp2f.exit328, !dbg !122 + +__nv_exp2f.exit328: ; preds = %1334, %1336 + %.0.i327 = phi float [ %1335, %1334 ], [ %1337, %1336 ], !dbg !122 + %1338 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !122 + %.not.i329 = icmp eq i32 %1338, 0, !dbg !122 + br i1 %.not.i329, label %1341, label %1339, !dbg !122 + +1339: ; preds = %__nv_exp2f.exit328 + %1340 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1310), !dbg !122 + br label %__nv_exp2f.exit331, !dbg !122 + +1341: ; preds = %__nv_exp2f.exit328 + %1342 = tail call float @llvm.nvvm.ex2.approx.f32(float %1310), !dbg !122 + br label %__nv_exp2f.exit331, !dbg !122 + +__nv_exp2f.exit331: ; preds = %1339, %1341 + %.0.i330 = phi float [ %1340, %1339 ], [ %1342, %1341 ], !dbg !122 + %1343 = insertelement <4 x float> poison, float %.0.i309, i64 0, !dbg !124 + %1344 = insertelement <4 x float> %1343, float %.0.i312, i64 1, !dbg !124 + %1345 = insertelement <4 x float> %1344, float %.0.i327, i64 2, !dbg !124 + %1346 = insertelement <4 x float> %1345, float %.0.i330, i64 3, !dbg !124 + %1347 = select <4 x i1> %1304, <4 x float> %1346, <4 x float> zeroinitializer, !dbg !124 + %1348 = select i1 %1302, float %.0.i315, float 0.000000e+00, !dbg !124 + %1349 = select i1 %1303, float %.0.i318, float 0.000000e+00, !dbg !124 + %1350 = fmul float %1016, %1348, !dbg !125 + %1351 = fmul float %1017, %1349, !dbg !125 + %1352 = fmul float %1018, 0.000000e+00, !dbg !125 + %1353 = fmul float %1019, 0.000000e+00, !dbg !125 + %1354 = fmul <4 x float> %1068, %1347, !dbg !125 + %1355 = and i1 %100, %68, !dbg !126 + %1356 = and i1 %100, %69, !dbg !126 + %1357 = and i1 %99, %68, !dbg !126 + %1358 = and i1 %99, %69, !dbg !126 + %1359 = and i1 %100, %70, !dbg !126 + %1360 = and i1 %100, %71, !dbg !126 + %1361 = and i1 %99, %70, !dbg !126 + %1362 = and i1 %99, %71, !dbg !126 + %1363 = fsub float %1113, %1109, !dbg !127 + %1364 = fsub float %1113, %1110, !dbg !127 + %1365 = fsub float %1114, %1109, !dbg !127 + %1366 = fsub float %1114, %1110, !dbg !127 + %1367 = fsub float %1113, %1111, !dbg !127 + %1368 = fsub float %1113, %1112, !dbg !127 + %1369 = fsub float %1114, %1111, !dbg !127 + %1370 = fsub float %1114, %1112, !dbg !127 + %1371 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !128 + %.not.i332 = icmp eq i32 %1371, 0, !dbg !128 + br i1 %.not.i332, label %1374, label %1372, !dbg !128 + +1372: ; preds = %__nv_exp2f.exit331 + %1373 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1363), !dbg !128 + br label %__nv_exp2f.exit334, !dbg !128 + +1374: ; preds = %__nv_exp2f.exit331 + %1375 = tail call float @llvm.nvvm.ex2.approx.f32(float %1363), !dbg !128 + br label %__nv_exp2f.exit334, !dbg !128 + +__nv_exp2f.exit334: ; preds = %1372, %1374 + %.0.i333 = phi float [ %1373, %1372 ], [ %1375, %1374 ], !dbg !128 + %1376 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !128 + %.not.i335 = icmp eq i32 %1376, 0, !dbg !128 + br i1 %.not.i335, label %1379, label %1377, !dbg !128 + +1377: ; preds = %__nv_exp2f.exit334 + %1378 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1364), !dbg !128 + br label %__nv_exp2f.exit337, !dbg !128 + +1379: ; preds = %__nv_exp2f.exit334 + %1380 = tail call float @llvm.nvvm.ex2.approx.f32(float %1364), !dbg !128 + br label %__nv_exp2f.exit337, !dbg !128 + +__nv_exp2f.exit337: ; preds = %1377, %1379 + %.0.i336 = phi float [ %1378, %1377 ], [ %1380, %1379 ], !dbg !128 + %1381 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !128 + %.not.i338 = icmp eq i32 %1381, 0, !dbg !128 + br i1 %.not.i338, label %1384, label %1382, !dbg !128 + +1382: ; preds = %__nv_exp2f.exit337 + %1383 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1365), !dbg !128 + br label %__nv_exp2f.exit340, !dbg !128 + +1384: ; preds = %__nv_exp2f.exit337 + %1385 = tail call float @llvm.nvvm.ex2.approx.f32(float %1365), !dbg !128 + br label %__nv_exp2f.exit340, !dbg !128 + +__nv_exp2f.exit340: ; preds = %1382, %1384 + %.0.i339 = phi float [ %1383, %1382 ], [ %1385, %1384 ], !dbg !128 + %1386 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !128 + %.not.i341 = icmp eq i32 %1386, 0, !dbg !128 + br i1 %.not.i341, label %1389, label %1387, !dbg !128 + +1387: ; preds = %__nv_exp2f.exit340 + %1388 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1366), !dbg !128 + br label %__nv_exp2f.exit343, !dbg !128 + +1389: ; preds = %__nv_exp2f.exit340 + %1390 = tail call float @llvm.nvvm.ex2.approx.f32(float %1366), !dbg !128 + br label %__nv_exp2f.exit343, !dbg !128 + +__nv_exp2f.exit343: ; preds = %1387, %1389 + %.0.i342 = phi float [ %1388, %1387 ], [ %1390, %1389 ], !dbg !128 + %1391 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !128 + %.not.i344 = icmp eq i32 %1391, 0, !dbg !128 + br i1 %.not.i344, label %1394, label %1392, !dbg !128 + +1392: ; preds = %__nv_exp2f.exit343 + %1393 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1367), !dbg !128 + br label %__nv_exp2f.exit346, !dbg !128 + +1394: ; preds = %__nv_exp2f.exit343 + %1395 = tail call float @llvm.nvvm.ex2.approx.f32(float %1367), !dbg !128 + br label %__nv_exp2f.exit346, !dbg !128 + +__nv_exp2f.exit346: ; preds = %1392, %1394 + %.0.i345 = phi float [ %1393, %1392 ], [ %1395, %1394 ], !dbg !128 + %1396 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !128 + %.not.i347 = icmp eq i32 %1396, 0, !dbg !128 + br i1 %.not.i347, label %1399, label %1397, !dbg !128 + +1397: ; preds = %__nv_exp2f.exit346 + %1398 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1368), !dbg !128 + br label %__nv_exp2f.exit349, !dbg !128 + +1399: ; preds = %__nv_exp2f.exit346 + %1400 = tail call float @llvm.nvvm.ex2.approx.f32(float %1368), !dbg !128 + br label %__nv_exp2f.exit349, !dbg !128 + +__nv_exp2f.exit349: ; preds = %1397, %1399 + %.0.i348 = phi float [ %1398, %1397 ], [ %1400, %1399 ], !dbg !128 + %1401 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !128 + %.not.i350 = icmp eq i32 %1401, 0, !dbg !128 + br i1 %.not.i350, label %1404, label %1402, !dbg !128 + +1402: ; preds = %__nv_exp2f.exit349 + %1403 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1369), !dbg !128 + br label %__nv_exp2f.exit352, !dbg !128 + +1404: ; preds = %__nv_exp2f.exit349 + %1405 = tail call float @llvm.nvvm.ex2.approx.f32(float %1369), !dbg !128 + br label %__nv_exp2f.exit352, !dbg !128 + +__nv_exp2f.exit352: ; preds = %1402, %1404 + %.0.i351 = phi float [ %1403, %1402 ], [ %1405, %1404 ], !dbg !128 + %1406 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !128 + %.not.i353 = icmp eq i32 %1406, 0, !dbg !128 + br i1 %.not.i353, label %1409, label %1407, !dbg !128 + +1407: ; preds = %__nv_exp2f.exit352 + %1408 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1370), !dbg !128 + br label %__nv_exp2f.exit355, !dbg !128 + +1409: ; preds = %__nv_exp2f.exit352 + %1410 = tail call float @llvm.nvvm.ex2.approx.f32(float %1370), !dbg !128 + br label %__nv_exp2f.exit355, !dbg !128 + +__nv_exp2f.exit355: ; preds = %1407, %1409 + %.0.i354 = phi float [ %1408, %1407 ], [ %1410, %1409 ], !dbg !128 + %1411 = select i1 %1355, float %.0.i333, float 0.000000e+00, !dbg !130 + %1412 = select i1 %1356, float %.0.i336, float 0.000000e+00, !dbg !130 + %1413 = select i1 %1357, float %.0.i339, float 0.000000e+00, !dbg !130 + %1414 = select i1 %1358, float %.0.i342, float 0.000000e+00, !dbg !130 + %1415 = select i1 %1359, float %.0.i345, float 0.000000e+00, !dbg !130 + %1416 = select i1 %1360, float %.0.i348, float 0.000000e+00, !dbg !130 + %1417 = select i1 %1361, float %.0.i351, float 0.000000e+00, !dbg !130 + %1418 = select i1 %1362, float %.0.i354, float 0.000000e+00, !dbg !130 + %1419 = fmul float %1020, %1411, !dbg !131 + %1420 = fmul float %1021, %1412, !dbg !131 + %1421 = fmul float %1022, %1413, !dbg !131 + %1422 = fmul float %1023, %1414, !dbg !131 + %1423 = fmul float %1024, %1415, !dbg !131 + %1424 = fmul float %1025, %1416, !dbg !131 + %1425 = fmul float %1026, %1417, !dbg !131 + %1426 = fmul float %1027, %1418, !dbg !131 + %1427 = and i1 %111, %68, !dbg !132 + %1428 = and i1 %111, %69, !dbg !132 + %1429 = and i1 %110, %68, !dbg !132 + %1430 = and i1 %110, %69, !dbg !132 + %1431 = and i1 %111, %70, !dbg !132 + %1432 = and i1 %111, %71, !dbg !132 + %1433 = and i1 %110, %70, !dbg !132 + %1434 = and i1 %110, %71, !dbg !132 + %1435 = fsub float %1119, %1109, !dbg !133 + %1436 = fsub float %1119, %1110, !dbg !133 + %1437 = fsub float %1120, %1109, !dbg !133 + %1438 = fsub float %1120, %1110, !dbg !133 + %1439 = fsub float %1119, %1111, !dbg !133 + %1440 = fsub float %1119, %1112, !dbg !133 + %1441 = fsub float %1120, %1111, !dbg !133 + %1442 = fsub float %1120, %1112, !dbg !133 + %1443 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !134 + %.not.i356 = icmp eq i32 %1443, 0, !dbg !134 + br i1 %.not.i356, label %1446, label %1444, !dbg !134 + +1444: ; preds = %__nv_exp2f.exit355 + %1445 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1435), !dbg !134 + br label %__nv_exp2f.exit358, !dbg !134 + +1446: ; preds = %__nv_exp2f.exit355 + %1447 = tail call float @llvm.nvvm.ex2.approx.f32(float %1435), !dbg !134 + br label %__nv_exp2f.exit358, !dbg !134 + +__nv_exp2f.exit358: ; preds = %1444, %1446 + %.0.i357 = phi float [ %1445, %1444 ], [ %1447, %1446 ], !dbg !134 + %1448 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !134 + %.not.i359 = icmp eq i32 %1448, 0, !dbg !134 + br i1 %.not.i359, label %1451, label %1449, !dbg !134 + +1449: ; preds = %__nv_exp2f.exit358 + %1450 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1436), !dbg !134 + br label %__nv_exp2f.exit361, !dbg !134 + +1451: ; preds = %__nv_exp2f.exit358 + %1452 = tail call float @llvm.nvvm.ex2.approx.f32(float %1436), !dbg !134 + br label %__nv_exp2f.exit361, !dbg !134 + +__nv_exp2f.exit361: ; preds = %1449, %1451 + %.0.i360 = phi float [ %1450, %1449 ], [ %1452, %1451 ], !dbg !134 + %1453 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !134 + %.not.i362 = icmp eq i32 %1453, 0, !dbg !134 + br i1 %.not.i362, label %1456, label %1454, !dbg !134 + +1454: ; preds = %__nv_exp2f.exit361 + %1455 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1437), !dbg !134 + br label %__nv_exp2f.exit364, !dbg !134 + +1456: ; preds = %__nv_exp2f.exit361 + %1457 = tail call float @llvm.nvvm.ex2.approx.f32(float %1437), !dbg !134 + br label %__nv_exp2f.exit364, !dbg !134 + +__nv_exp2f.exit364: ; preds = %1454, %1456 + %.0.i363 = phi float [ %1455, %1454 ], [ %1457, %1456 ], !dbg !134 + %1458 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !134 + %.not.i365 = icmp eq i32 %1458, 0, !dbg !134 + br i1 %.not.i365, label %1461, label %1459, !dbg !134 + +1459: ; preds = %__nv_exp2f.exit364 + %1460 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1438), !dbg !134 + br label %__nv_exp2f.exit367, !dbg !134 + +1461: ; preds = %__nv_exp2f.exit364 + %1462 = tail call float @llvm.nvvm.ex2.approx.f32(float %1438), !dbg !134 + br label %__nv_exp2f.exit367, !dbg !134 + +__nv_exp2f.exit367: ; preds = %1459, %1461 + %.0.i366 = phi float [ %1460, %1459 ], [ %1462, %1461 ], !dbg !134 + %1463 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !134 + %.not.i368 = icmp eq i32 %1463, 0, !dbg !134 + br i1 %.not.i368, label %1466, label %1464, !dbg !134 + +1464: ; preds = %__nv_exp2f.exit367 + %1465 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1439), !dbg !134 + br label %__nv_exp2f.exit370, !dbg !134 + +1466: ; preds = %__nv_exp2f.exit367 + %1467 = tail call float @llvm.nvvm.ex2.approx.f32(float %1439), !dbg !134 + br label %__nv_exp2f.exit370, !dbg !134 + +__nv_exp2f.exit370: ; preds = %1464, %1466 + %.0.i369 = phi float [ %1465, %1464 ], [ %1467, %1466 ], !dbg !134 + %1468 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !134 + %.not.i371 = icmp eq i32 %1468, 0, !dbg !134 + br i1 %.not.i371, label %1471, label %1469, !dbg !134 + +1469: ; preds = %__nv_exp2f.exit370 + %1470 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1440), !dbg !134 + br label %__nv_exp2f.exit373, !dbg !134 + +1471: ; preds = %__nv_exp2f.exit370 + %1472 = tail call float @llvm.nvvm.ex2.approx.f32(float %1440), !dbg !134 + br label %__nv_exp2f.exit373, !dbg !134 + +__nv_exp2f.exit373: ; preds = %1469, %1471 + %.0.i372 = phi float [ %1470, %1469 ], [ %1472, %1471 ], !dbg !134 + %1473 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !134 + %.not.i374 = icmp eq i32 %1473, 0, !dbg !134 + br i1 %.not.i374, label %1476, label %1474, !dbg !134 + +1474: ; preds = %__nv_exp2f.exit373 + %1475 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1441), !dbg !134 + br label %__nv_exp2f.exit376, !dbg !134 + +1476: ; preds = %__nv_exp2f.exit373 + %1477 = tail call float @llvm.nvvm.ex2.approx.f32(float %1441), !dbg !134 + br label %__nv_exp2f.exit376, !dbg !134 + +__nv_exp2f.exit376: ; preds = %1474, %1476 + %.0.i375 = phi float [ %1475, %1474 ], [ %1477, %1476 ], !dbg !134 + %1478 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !134 + %.not.i377 = icmp eq i32 %1478, 0, !dbg !134 + br i1 %.not.i377, label %1481, label %1479, !dbg !134 + +1479: ; preds = %__nv_exp2f.exit376 + %1480 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1442), !dbg !134 + br label %__nv_exp2f.exit379, !dbg !134 + +1481: ; preds = %__nv_exp2f.exit376 + %1482 = tail call float @llvm.nvvm.ex2.approx.f32(float %1442), !dbg !134 + br label %__nv_exp2f.exit379, !dbg !134 + +__nv_exp2f.exit379: ; preds = %1479, %1481 + %.0.i378 = phi float [ %1480, %1479 ], [ %1482, %1481 ], !dbg !134 + %1483 = select i1 %1427, float %.0.i357, float 0.000000e+00, !dbg !136 + %1484 = select i1 %1428, float %.0.i360, float 0.000000e+00, !dbg !136 + %1485 = select i1 %1429, float %.0.i363, float 0.000000e+00, !dbg !136 + %1486 = select i1 %1430, float %.0.i366, float 0.000000e+00, !dbg !136 + %1487 = select i1 %1431, float %.0.i369, float 0.000000e+00, !dbg !136 + %1488 = select i1 %1432, float %.0.i372, float 0.000000e+00, !dbg !136 + %1489 = select i1 %1433, float %.0.i375, float 0.000000e+00, !dbg !136 + %1490 = select i1 %1434, float %.0.i378, float 0.000000e+00, !dbg !136 + %1491 = fmul float %1028, %1483, !dbg !137 + %1492 = fmul float %1029, %1484, !dbg !137 + %1493 = fmul float %1030, %1485, !dbg !137 + %1494 = fmul float %1031, %1486, !dbg !137 + %1495 = fmul float %1032, %1487, !dbg !137 + %1496 = fmul float %1033, %1488, !dbg !137 + %1497 = fmul float %1034, %1489, !dbg !137 + %1498 = fmul float %1035, %1490, !dbg !137 + %1499 = and i1 %111, %206, !dbg !138 + %1500 = and i1 %111, %208, !dbg !138 + %1501 = and i1 %110, %206, !dbg !138 + %1502 = and i1 %110, %208, !dbg !138 + %1503 = and i1 %111, %210, !dbg !138 + %1504 = and i1 %111, %212, !dbg !138 + %1505 = and i1 %110, %210, !dbg !138 + %1506 = and i1 %110, %212, !dbg !138 + %1507 = fsub float %1119, %1115, !dbg !139 + %1508 = fsub float %1119, %1116, !dbg !139 + %1509 = fsub float %1120, %1115, !dbg !139 + %1510 = fsub float %1120, %1116, !dbg !139 + %1511 = fsub float %1119, %1117, !dbg !139 + %1512 = fsub float %1119, %1118, !dbg !139 + %1513 = fsub float %1120, %1117, !dbg !139 + %1514 = fsub float %1120, %1118, !dbg !139 + %1515 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !140 + %.not.i380 = icmp eq i32 %1515, 0, !dbg !140 + br i1 %.not.i380, label %1518, label %1516, !dbg !140 + +1516: ; preds = %__nv_exp2f.exit379 + %1517 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1507), !dbg !140 + br label %__nv_exp2f.exit382, !dbg !140 + +1518: ; preds = %__nv_exp2f.exit379 + %1519 = tail call float @llvm.nvvm.ex2.approx.f32(float %1507), !dbg !140 + br label %__nv_exp2f.exit382, !dbg !140 + +__nv_exp2f.exit382: ; preds = %1516, %1518 + %.0.i381 = phi float [ %1517, %1516 ], [ %1519, %1518 ], !dbg !140 + %1520 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !140 + %.not.i383 = icmp eq i32 %1520, 0, !dbg !140 + br i1 %.not.i383, label %1523, label %1521, !dbg !140 + +1521: ; preds = %__nv_exp2f.exit382 + %1522 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1508), !dbg !140 + br label %__nv_exp2f.exit385, !dbg !140 + +1523: ; preds = %__nv_exp2f.exit382 + %1524 = tail call float @llvm.nvvm.ex2.approx.f32(float %1508), !dbg !140 + br label %__nv_exp2f.exit385, !dbg !140 + +__nv_exp2f.exit385: ; preds = %1521, %1523 + %.0.i384 = phi float [ %1522, %1521 ], [ %1524, %1523 ], !dbg !140 + %1525 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !140 + %.not.i386 = icmp eq i32 %1525, 0, !dbg !140 + br i1 %.not.i386, label %1528, label %1526, !dbg !140 + +1526: ; preds = %__nv_exp2f.exit385 + %1527 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1509), !dbg !140 + br label %__nv_exp2f.exit388, !dbg !140 + +1528: ; preds = %__nv_exp2f.exit385 + %1529 = tail call float @llvm.nvvm.ex2.approx.f32(float %1509), !dbg !140 + br label %__nv_exp2f.exit388, !dbg !140 + +__nv_exp2f.exit388: ; preds = %1526, %1528 + %.0.i387 = phi float [ %1527, %1526 ], [ %1529, %1528 ], !dbg !140 + %1530 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !140 + %.not.i389 = icmp eq i32 %1530, 0, !dbg !140 + br i1 %.not.i389, label %1533, label %1531, !dbg !140 + +1531: ; preds = %__nv_exp2f.exit388 + %1532 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1510), !dbg !140 + br label %__nv_exp2f.exit391, !dbg !140 + +1533: ; preds = %__nv_exp2f.exit388 + %1534 = tail call float @llvm.nvvm.ex2.approx.f32(float %1510), !dbg !140 + br label %__nv_exp2f.exit391, !dbg !140 + +__nv_exp2f.exit391: ; preds = %1531, %1533 + %.0.i390 = phi float [ %1532, %1531 ], [ %1534, %1533 ], !dbg !140 + %1535 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !140 + %.not.i392 = icmp eq i32 %1535, 0, !dbg !140 + br i1 %.not.i392, label %1538, label %1536, !dbg !140 + +1536: ; preds = %__nv_exp2f.exit391 + %1537 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1511), !dbg !140 + br label %__nv_exp2f.exit394, !dbg !140 + +1538: ; preds = %__nv_exp2f.exit391 + %1539 = tail call float @llvm.nvvm.ex2.approx.f32(float %1511), !dbg !140 + br label %__nv_exp2f.exit394, !dbg !140 + +__nv_exp2f.exit394: ; preds = %1536, %1538 + %.0.i393 = phi float [ %1537, %1536 ], [ %1539, %1538 ], !dbg !140 + %1540 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !140 + %.not.i395 = icmp eq i32 %1540, 0, !dbg !140 + br i1 %.not.i395, label %1543, label %1541, !dbg !140 + +1541: ; preds = %__nv_exp2f.exit394 + %1542 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1512), !dbg !140 + br label %__nv_exp2f.exit397, !dbg !140 + +1543: ; preds = %__nv_exp2f.exit394 + %1544 = tail call float @llvm.nvvm.ex2.approx.f32(float %1512), !dbg !140 + br label %__nv_exp2f.exit397, !dbg !140 + +__nv_exp2f.exit397: ; preds = %1541, %1543 + %.0.i396 = phi float [ %1542, %1541 ], [ %1544, %1543 ], !dbg !140 + %1545 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !140 + %.not.i398 = icmp eq i32 %1545, 0, !dbg !140 + br i1 %.not.i398, label %1548, label %1546, !dbg !140 + +1546: ; preds = %__nv_exp2f.exit397 + %1547 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1513), !dbg !140 + br label %__nv_exp2f.exit400, !dbg !140 + +1548: ; preds = %__nv_exp2f.exit397 + %1549 = tail call float @llvm.nvvm.ex2.approx.f32(float %1513), !dbg !140 + br label %__nv_exp2f.exit400, !dbg !140 + +__nv_exp2f.exit400: ; preds = %1546, %1548 + %.0.i399 = phi float [ %1547, %1546 ], [ %1549, %1548 ], !dbg !140 + %1550 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !140 + %.not.i401 = icmp eq i32 %1550, 0, !dbg !140 + br i1 %.not.i401, label %1553, label %1551, !dbg !140 + +1551: ; preds = %__nv_exp2f.exit400 + %1552 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1514), !dbg !140 + br label %__nv_exp2f.exit403, !dbg !140 + +1553: ; preds = %__nv_exp2f.exit400 + %1554 = tail call float @llvm.nvvm.ex2.approx.f32(float %1514), !dbg !140 + br label %__nv_exp2f.exit403, !dbg !140 + +__nv_exp2f.exit403: ; preds = %1551, %1553 + %.0.i402 = phi float [ %1552, %1551 ], [ %1554, %1553 ], !dbg !140 + %1555 = select i1 %1499, float %.0.i381, float 0.000000e+00, !dbg !142 + %1556 = select i1 %1500, float %.0.i384, float 0.000000e+00, !dbg !142 + %1557 = select i1 %1501, float %.0.i387, float 0.000000e+00, !dbg !142 + %1558 = select i1 %1502, float %.0.i390, float 0.000000e+00, !dbg !142 + %1559 = select i1 %1503, float %.0.i393, float 0.000000e+00, !dbg !142 + %1560 = select i1 %1504, float %.0.i396, float 0.000000e+00, !dbg !142 + %1561 = select i1 %1505, float %.0.i399, float 0.000000e+00, !dbg !142 + %1562 = select i1 %1506, float %.0.i402, float 0.000000e+00, !dbg !142 + %1563 = fmul float %1036, %1555, !dbg !143 + %1564 = fmul float %1037, %1556, !dbg !143 + %1565 = fmul float %1038, %1557, !dbg !143 + %1566 = fmul float %1039, %1558, !dbg !143 + %1567 = fmul float %1040, %1559, !dbg !143 + %1568 = fmul float %1041, %1560, !dbg !143 + %1569 = fmul float %1042, %1561, !dbg !143 + %1570 = fmul float %1043, %1562, !dbg !143 + %1571 = and i1 %121, %68, !dbg !144 + %1572 = and i1 %121, %69, !dbg !144 + %1573 = and i1 %120, %68, !dbg !144 + %1574 = and i1 %120, %69, !dbg !144 + %1575 = and i1 %121, %70, !dbg !144 + %1576 = and i1 %121, %71, !dbg !144 + %1577 = and i1 %120, %70, !dbg !144 + %1578 = and i1 %120, %71, !dbg !144 + %1579 = fsub float %1125, %1109, !dbg !145 + %1580 = fsub float %1125, %1110, !dbg !145 + %1581 = fsub float %1126, %1109, !dbg !145 + %1582 = fsub float %1126, %1110, !dbg !145 + %1583 = fsub float %1125, %1111, !dbg !145 + %1584 = fsub float %1125, %1112, !dbg !145 + %1585 = fsub float %1126, %1111, !dbg !145 + %1586 = fsub float %1126, %1112, !dbg !145 + %1587 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !146 + %.not.i404 = icmp eq i32 %1587, 0, !dbg !146 + br i1 %.not.i404, label %1590, label %1588, !dbg !146 + +1588: ; preds = %__nv_exp2f.exit403 + %1589 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1579), !dbg !146 + br label %__nv_exp2f.exit406, !dbg !146 + +1590: ; preds = %__nv_exp2f.exit403 + %1591 = tail call float @llvm.nvvm.ex2.approx.f32(float %1579), !dbg !146 + br label %__nv_exp2f.exit406, !dbg !146 + +__nv_exp2f.exit406: ; preds = %1588, %1590 + %.0.i405 = phi float [ %1589, %1588 ], [ %1591, %1590 ], !dbg !146 + %1592 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !146 + %.not.i407 = icmp eq i32 %1592, 0, !dbg !146 + br i1 %.not.i407, label %1595, label %1593, !dbg !146 + +1593: ; preds = %__nv_exp2f.exit406 + %1594 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1580), !dbg !146 + br label %__nv_exp2f.exit409, !dbg !146 + +1595: ; preds = %__nv_exp2f.exit406 + %1596 = tail call float @llvm.nvvm.ex2.approx.f32(float %1580), !dbg !146 + br label %__nv_exp2f.exit409, !dbg !146 + +__nv_exp2f.exit409: ; preds = %1593, %1595 + %.0.i408 = phi float [ %1594, %1593 ], [ %1596, %1595 ], !dbg !146 + %1597 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !146 + %.not.i410 = icmp eq i32 %1597, 0, !dbg !146 + br i1 %.not.i410, label %1600, label %1598, !dbg !146 + +1598: ; preds = %__nv_exp2f.exit409 + %1599 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1581), !dbg !146 + br label %__nv_exp2f.exit412, !dbg !146 + +1600: ; preds = %__nv_exp2f.exit409 + %1601 = tail call float @llvm.nvvm.ex2.approx.f32(float %1581), !dbg !146 + br label %__nv_exp2f.exit412, !dbg !146 + +__nv_exp2f.exit412: ; preds = %1598, %1600 + %.0.i411 = phi float [ %1599, %1598 ], [ %1601, %1600 ], !dbg !146 + %1602 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !146 + %.not.i413 = icmp eq i32 %1602, 0, !dbg !146 + br i1 %.not.i413, label %1605, label %1603, !dbg !146 + +1603: ; preds = %__nv_exp2f.exit412 + %1604 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1582), !dbg !146 + br label %__nv_exp2f.exit415, !dbg !146 + +1605: ; preds = %__nv_exp2f.exit412 + %1606 = tail call float @llvm.nvvm.ex2.approx.f32(float %1582), !dbg !146 + br label %__nv_exp2f.exit415, !dbg !146 + +__nv_exp2f.exit415: ; preds = %1603, %1605 + %.0.i414 = phi float [ %1604, %1603 ], [ %1606, %1605 ], !dbg !146 + %1607 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !146 + %.not.i416 = icmp eq i32 %1607, 0, !dbg !146 + br i1 %.not.i416, label %1610, label %1608, !dbg !146 + +1608: ; preds = %__nv_exp2f.exit415 + %1609 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1583), !dbg !146 + br label %__nv_exp2f.exit418, !dbg !146 + +1610: ; preds = %__nv_exp2f.exit415 + %1611 = tail call float @llvm.nvvm.ex2.approx.f32(float %1583), !dbg !146 + br label %__nv_exp2f.exit418, !dbg !146 + +__nv_exp2f.exit418: ; preds = %1608, %1610 + %.0.i417 = phi float [ %1609, %1608 ], [ %1611, %1610 ], !dbg !146 + %1612 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !146 + %.not.i419 = icmp eq i32 %1612, 0, !dbg !146 + br i1 %.not.i419, label %1615, label %1613, !dbg !146 + +1613: ; preds = %__nv_exp2f.exit418 + %1614 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1584), !dbg !146 + br label %__nv_exp2f.exit421, !dbg !146 + +1615: ; preds = %__nv_exp2f.exit418 + %1616 = tail call float @llvm.nvvm.ex2.approx.f32(float %1584), !dbg !146 + br label %__nv_exp2f.exit421, !dbg !146 + +__nv_exp2f.exit421: ; preds = %1613, %1615 + %.0.i420 = phi float [ %1614, %1613 ], [ %1616, %1615 ], !dbg !146 + %1617 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !146 + %.not.i422 = icmp eq i32 %1617, 0, !dbg !146 + br i1 %.not.i422, label %1620, label %1618, !dbg !146 + +1618: ; preds = %__nv_exp2f.exit421 + %1619 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1585), !dbg !146 + br label %__nv_exp2f.exit424, !dbg !146 + +1620: ; preds = %__nv_exp2f.exit421 + %1621 = tail call float @llvm.nvvm.ex2.approx.f32(float %1585), !dbg !146 + br label %__nv_exp2f.exit424, !dbg !146 + +__nv_exp2f.exit424: ; preds = %1618, %1620 + %.0.i423 = phi float [ %1619, %1618 ], [ %1621, %1620 ], !dbg !146 + %1622 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !146 + %.not.i425 = icmp eq i32 %1622, 0, !dbg !146 + br i1 %.not.i425, label %1625, label %1623, !dbg !146 + +1623: ; preds = %__nv_exp2f.exit424 + %1624 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1586), !dbg !146 + br label %__nv_exp2f.exit427, !dbg !146 + +1625: ; preds = %__nv_exp2f.exit424 + %1626 = tail call float @llvm.nvvm.ex2.approx.f32(float %1586), !dbg !146 + br label %__nv_exp2f.exit427, !dbg !146 + +__nv_exp2f.exit427: ; preds = %1623, %1625 + %.0.i426 = phi float [ %1624, %1623 ], [ %1626, %1625 ], !dbg !146 + %1627 = select i1 %1571, float %.0.i405, float 0.000000e+00, !dbg !148 + %1628 = select i1 %1572, float %.0.i408, float 0.000000e+00, !dbg !148 + %1629 = select i1 %1573, float %.0.i411, float 0.000000e+00, !dbg !148 + %1630 = select i1 %1574, float %.0.i414, float 0.000000e+00, !dbg !148 + %1631 = select i1 %1575, float %.0.i417, float 0.000000e+00, !dbg !148 + %1632 = select i1 %1576, float %.0.i420, float 0.000000e+00, !dbg !148 + %1633 = select i1 %1577, float %.0.i423, float 0.000000e+00, !dbg !148 + %1634 = select i1 %1578, float %.0.i426, float 0.000000e+00, !dbg !148 + %1635 = fmul float %1044, %1627, !dbg !149 + %1636 = fmul float %1045, %1628, !dbg !149 + %1637 = fmul float %1046, %1629, !dbg !149 + %1638 = fmul float %1047, %1630, !dbg !149 + %1639 = fmul float %1048, %1631, !dbg !149 + %1640 = fmul float %1049, %1632, !dbg !149 + %1641 = fmul float %1050, %1633, !dbg !149 + %1642 = fmul float %1051, %1634, !dbg !149 + %1643 = and i1 %121, %206, !dbg !150 + %1644 = and i1 %121, %208, !dbg !150 + %1645 = and i1 %120, %206, !dbg !150 + %1646 = and i1 %120, %208, !dbg !150 + %1647 = and i1 %121, %210, !dbg !150 + %1648 = and i1 %121, %212, !dbg !150 + %1649 = and i1 %120, %210, !dbg !150 + %1650 = and i1 %120, %212, !dbg !150 + %1651 = fsub float %1125, %1115, !dbg !151 + %1652 = fsub float %1125, %1116, !dbg !151 + %1653 = fsub float %1126, %1115, !dbg !151 + %1654 = fsub float %1126, %1116, !dbg !151 + %1655 = fsub float %1125, %1117, !dbg !151 + %1656 = fsub float %1125, %1118, !dbg !151 + %1657 = fsub float %1126, %1117, !dbg !151 + %1658 = fsub float %1126, %1118, !dbg !151 + %1659 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !152 + %.not.i428 = icmp eq i32 %1659, 0, !dbg !152 + br i1 %.not.i428, label %1662, label %1660, !dbg !152 + +1660: ; preds = %__nv_exp2f.exit427 + %1661 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1651), !dbg !152 + br label %__nv_exp2f.exit430, !dbg !152 + +1662: ; preds = %__nv_exp2f.exit427 + %1663 = tail call float @llvm.nvvm.ex2.approx.f32(float %1651), !dbg !152 + br label %__nv_exp2f.exit430, !dbg !152 + +__nv_exp2f.exit430: ; preds = %1660, %1662 + %.0.i429 = phi float [ %1661, %1660 ], [ %1663, %1662 ], !dbg !152 + %1664 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !152 + %.not.i431 = icmp eq i32 %1664, 0, !dbg !152 + br i1 %.not.i431, label %1667, label %1665, !dbg !152 + +1665: ; preds = %__nv_exp2f.exit430 + %1666 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1652), !dbg !152 + br label %__nv_exp2f.exit433, !dbg !152 + +1667: ; preds = %__nv_exp2f.exit430 + %1668 = tail call float @llvm.nvvm.ex2.approx.f32(float %1652), !dbg !152 + br label %__nv_exp2f.exit433, !dbg !152 + +__nv_exp2f.exit433: ; preds = %1665, %1667 + %.0.i432 = phi float [ %1666, %1665 ], [ %1668, %1667 ], !dbg !152 + %1669 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !152 + %.not.i434 = icmp eq i32 %1669, 0, !dbg !152 + br i1 %.not.i434, label %1672, label %1670, !dbg !152 + +1670: ; preds = %__nv_exp2f.exit433 + %1671 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1653), !dbg !152 + br label %__nv_exp2f.exit436, !dbg !152 + +1672: ; preds = %__nv_exp2f.exit433 + %1673 = tail call float @llvm.nvvm.ex2.approx.f32(float %1653), !dbg !152 + br label %__nv_exp2f.exit436, !dbg !152 + +__nv_exp2f.exit436: ; preds = %1670, %1672 + %.0.i435 = phi float [ %1671, %1670 ], [ %1673, %1672 ], !dbg !152 + %1674 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !152 + %.not.i437 = icmp eq i32 %1674, 0, !dbg !152 + br i1 %.not.i437, label %1677, label %1675, !dbg !152 + +1675: ; preds = %__nv_exp2f.exit436 + %1676 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1654), !dbg !152 + br label %__nv_exp2f.exit439, !dbg !152 + +1677: ; preds = %__nv_exp2f.exit436 + %1678 = tail call float @llvm.nvvm.ex2.approx.f32(float %1654), !dbg !152 + br label %__nv_exp2f.exit439, !dbg !152 + +__nv_exp2f.exit439: ; preds = %1675, %1677 + %.0.i438 = phi float [ %1676, %1675 ], [ %1678, %1677 ], !dbg !152 + %1679 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !152 + %.not.i440 = icmp eq i32 %1679, 0, !dbg !152 + br i1 %.not.i440, label %1682, label %1680, !dbg !152 + +1680: ; preds = %__nv_exp2f.exit439 + %1681 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1655), !dbg !152 + br label %__nv_exp2f.exit442, !dbg !152 + +1682: ; preds = %__nv_exp2f.exit439 + %1683 = tail call float @llvm.nvvm.ex2.approx.f32(float %1655), !dbg !152 + br label %__nv_exp2f.exit442, !dbg !152 + +__nv_exp2f.exit442: ; preds = %1680, %1682 + %.0.i441 = phi float [ %1681, %1680 ], [ %1683, %1682 ], !dbg !152 + %1684 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !152 + %.not.i443 = icmp eq i32 %1684, 0, !dbg !152 + br i1 %.not.i443, label %1687, label %1685, !dbg !152 + +1685: ; preds = %__nv_exp2f.exit442 + %1686 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1656), !dbg !152 + br label %__nv_exp2f.exit445, !dbg !152 + +1687: ; preds = %__nv_exp2f.exit442 + %1688 = tail call float @llvm.nvvm.ex2.approx.f32(float %1656), !dbg !152 + br label %__nv_exp2f.exit445, !dbg !152 + +__nv_exp2f.exit445: ; preds = %1685, %1687 + %.0.i444 = phi float [ %1686, %1685 ], [ %1688, %1687 ], !dbg !152 + %1689 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !152 + %.not.i446 = icmp eq i32 %1689, 0, !dbg !152 + br i1 %.not.i446, label %1692, label %1690, !dbg !152 + +1690: ; preds = %__nv_exp2f.exit445 + %1691 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1657), !dbg !152 + br label %__nv_exp2f.exit448, !dbg !152 + +1692: ; preds = %__nv_exp2f.exit445 + %1693 = tail call float @llvm.nvvm.ex2.approx.f32(float %1657), !dbg !152 + br label %__nv_exp2f.exit448, !dbg !152 + +__nv_exp2f.exit448: ; preds = %1690, %1692 + %.0.i447 = phi float [ %1691, %1690 ], [ %1693, %1692 ], !dbg !152 + %1694 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !152 + %.not.i449 = icmp eq i32 %1694, 0, !dbg !152 + br i1 %.not.i449, label %1697, label %1695, !dbg !152 + +1695: ; preds = %__nv_exp2f.exit448 + %1696 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1658), !dbg !152 + br label %__nv_exp2f.exit451, !dbg !152 + +1697: ; preds = %__nv_exp2f.exit448 + %1698 = tail call float @llvm.nvvm.ex2.approx.f32(float %1658), !dbg !152 + br label %__nv_exp2f.exit451, !dbg !152 + +__nv_exp2f.exit451: ; preds = %1695, %1697 + %.0.i450 = phi float [ %1696, %1695 ], [ %1698, %1697 ], !dbg !152 + %1699 = select i1 %1643, float %.0.i429, float 0.000000e+00, !dbg !154 + %1700 = select i1 %1644, float %.0.i432, float 0.000000e+00, !dbg !154 + %1701 = select i1 %1645, float %.0.i435, float 0.000000e+00, !dbg !154 + %1702 = select i1 %1646, float %.0.i438, float 0.000000e+00, !dbg !154 + %1703 = select i1 %1647, float %.0.i441, float 0.000000e+00, !dbg !154 + %1704 = select i1 %1648, float %.0.i444, float 0.000000e+00, !dbg !154 + %1705 = select i1 %1649, float %.0.i447, float 0.000000e+00, !dbg !154 + %1706 = select i1 %1650, float %.0.i450, float 0.000000e+00, !dbg !154 + %1707 = fmul float %1052, %1699, !dbg !155 + %1708 = fmul float %1053, %1700, !dbg !155 + %1709 = fmul float %1054, %1701, !dbg !155 + %1710 = fmul float %1055, %1702, !dbg !155 + %1711 = fmul float %1056, %1703, !dbg !155 + %1712 = fmul float %1057, %1704, !dbg !155 + %1713 = fmul float %1058, %1705, !dbg !155 + %1714 = fmul float %1059, %1706, !dbg !155 + %1715 = and i1 %121, %216, !dbg !156 + %1716 = and i1 %121, %218, !dbg !156 + %1717 = and i1 %120, %216, !dbg !156 + %1718 = and i1 %120, %218, !dbg !156 + %1719 = and i1 %121, %220, !dbg !156 + %1720 = and i1 %121, %222, !dbg !156 + %1721 = and i1 %120, %220, !dbg !156 + %1722 = and i1 %120, %222, !dbg !156 + %1723 = fsub float %1125, %1121, !dbg !157 + %1724 = fsub float %1125, %1122, !dbg !157 + %1725 = fsub float %1126, %1121, !dbg !157 + %1726 = fsub float %1126, %1122, !dbg !157 + %1727 = fsub float %1125, %1123, !dbg !157 + %1728 = fsub float %1125, %1124, !dbg !157 + %1729 = fsub float %1126, %1123, !dbg !157 + %1730 = fsub float %1126, %1124, !dbg !157 + %1731 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !158 + %.not.i452 = icmp eq i32 %1731, 0, !dbg !158 + br i1 %.not.i452, label %1734, label %1732, !dbg !158 + +1732: ; preds = %__nv_exp2f.exit451 + %1733 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1723), !dbg !158 + br label %__nv_exp2f.exit454, !dbg !158 + +1734: ; preds = %__nv_exp2f.exit451 + %1735 = tail call float @llvm.nvvm.ex2.approx.f32(float %1723), !dbg !158 + br label %__nv_exp2f.exit454, !dbg !158 + +__nv_exp2f.exit454: ; preds = %1732, %1734 + %.0.i453 = phi float [ %1733, %1732 ], [ %1735, %1734 ], !dbg !158 + %1736 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !158 + %.not.i455 = icmp eq i32 %1736, 0, !dbg !158 + br i1 %.not.i455, label %1739, label %1737, !dbg !158 + +1737: ; preds = %__nv_exp2f.exit454 + %1738 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1724), !dbg !158 + br label %__nv_exp2f.exit457, !dbg !158 + +1739: ; preds = %__nv_exp2f.exit454 + %1740 = tail call float @llvm.nvvm.ex2.approx.f32(float %1724), !dbg !158 + br label %__nv_exp2f.exit457, !dbg !158 + +__nv_exp2f.exit457: ; preds = %1737, %1739 + %.0.i456 = phi float [ %1738, %1737 ], [ %1740, %1739 ], !dbg !158 + %1741 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !158 + %.not.i458 = icmp eq i32 %1741, 0, !dbg !158 + br i1 %.not.i458, label %1744, label %1742, !dbg !158 + +1742: ; preds = %__nv_exp2f.exit457 + %1743 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1725), !dbg !158 + br label %__nv_exp2f.exit460, !dbg !158 + +1744: ; preds = %__nv_exp2f.exit457 + %1745 = tail call float @llvm.nvvm.ex2.approx.f32(float %1725), !dbg !158 + br label %__nv_exp2f.exit460, !dbg !158 + +__nv_exp2f.exit460: ; preds = %1742, %1744 + %.0.i459 = phi float [ %1743, %1742 ], [ %1745, %1744 ], !dbg !158 + %1746 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !158 + %.not.i461 = icmp eq i32 %1746, 0, !dbg !158 + br i1 %.not.i461, label %1749, label %1747, !dbg !158 + +1747: ; preds = %__nv_exp2f.exit460 + %1748 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1726), !dbg !158 + br label %__nv_exp2f.exit463, !dbg !158 + +1749: ; preds = %__nv_exp2f.exit460 + %1750 = tail call float @llvm.nvvm.ex2.approx.f32(float %1726), !dbg !158 + br label %__nv_exp2f.exit463, !dbg !158 + +__nv_exp2f.exit463: ; preds = %1747, %1749 + %.0.i462 = phi float [ %1748, %1747 ], [ %1750, %1749 ], !dbg !158 + %1751 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !158 + %.not.i464 = icmp eq i32 %1751, 0, !dbg !158 + br i1 %.not.i464, label %1754, label %1752, !dbg !158 + +1752: ; preds = %__nv_exp2f.exit463 + %1753 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1727), !dbg !158 + br label %__nv_exp2f.exit466, !dbg !158 + +1754: ; preds = %__nv_exp2f.exit463 + %1755 = tail call float @llvm.nvvm.ex2.approx.f32(float %1727), !dbg !158 + br label %__nv_exp2f.exit466, !dbg !158 + +__nv_exp2f.exit466: ; preds = %1752, %1754 + %.0.i465 = phi float [ %1753, %1752 ], [ %1755, %1754 ], !dbg !158 + %1756 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !158 + %.not.i467 = icmp eq i32 %1756, 0, !dbg !158 + br i1 %.not.i467, label %1759, label %1757, !dbg !158 + +1757: ; preds = %__nv_exp2f.exit466 + %1758 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1728), !dbg !158 + br label %__nv_exp2f.exit469, !dbg !158 + +1759: ; preds = %__nv_exp2f.exit466 + %1760 = tail call float @llvm.nvvm.ex2.approx.f32(float %1728), !dbg !158 + br label %__nv_exp2f.exit469, !dbg !158 + +__nv_exp2f.exit469: ; preds = %1757, %1759 + %.0.i468 = phi float [ %1758, %1757 ], [ %1760, %1759 ], !dbg !158 + %1761 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !158 + %.not.i470 = icmp eq i32 %1761, 0, !dbg !158 + br i1 %.not.i470, label %1764, label %1762, !dbg !158 + +1762: ; preds = %__nv_exp2f.exit469 + %1763 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1729), !dbg !158 + br label %__nv_exp2f.exit472, !dbg !158 + +1764: ; preds = %__nv_exp2f.exit469 + %1765 = tail call float @llvm.nvvm.ex2.approx.f32(float %1729), !dbg !158 + br label %__nv_exp2f.exit472, !dbg !158 + +__nv_exp2f.exit472: ; preds = %1762, %1764 + %.0.i471 = phi float [ %1763, %1762 ], [ %1765, %1764 ], !dbg !158 + %1766 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !158 + %.not.i473 = icmp eq i32 %1766, 0, !dbg !158 + br i1 %.not.i473, label %1769, label %1767, !dbg !158 + +1767: ; preds = %__nv_exp2f.exit472 + %1768 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1730), !dbg !158 + br label %__nv_exp2f.exit475, !dbg !158 + +1769: ; preds = %__nv_exp2f.exit472 + %1770 = tail call float @llvm.nvvm.ex2.approx.f32(float %1730), !dbg !158 + br label %__nv_exp2f.exit475, !dbg !158 + +__nv_exp2f.exit475: ; preds = %1767, %1769 + %.0.i474 = phi float [ %1768, %1767 ], [ %1770, %1769 ], !dbg !158 + %1771 = select i1 %1715, float %.0.i453, float 0.000000e+00, !dbg !160 + %1772 = select i1 %1716, float %.0.i456, float 0.000000e+00, !dbg !160 + %1773 = select i1 %1717, float %.0.i459, float 0.000000e+00, !dbg !160 + %1774 = select i1 %1718, float %.0.i462, float 0.000000e+00, !dbg !160 + %1775 = select i1 %1719, float %.0.i465, float 0.000000e+00, !dbg !160 + %1776 = select i1 %1720, float %.0.i468, float 0.000000e+00, !dbg !160 + %1777 = select i1 %1721, float %.0.i471, float 0.000000e+00, !dbg !160 + %1778 = select i1 %1722, float %.0.i474, float 0.000000e+00, !dbg !160 + %1779 = fmul float %1060, %1771, !dbg !161 + %1780 = fmul float %1061, %1772, !dbg !161 + %1781 = fmul float %1062, %1773, !dbg !161 + %1782 = fmul float %1063, %1774, !dbg !161 + %1783 = fmul float %1064, %1775, !dbg !161 + %1784 = fmul float %1065, %1776, !dbg !161 + %1785 = fmul float %1066, %1777, !dbg !161 + %1786 = fmul float %1067, %1778, !dbg !161 + %1787 = fmul <4 x float> %1182, %1089, !dbg !162 + %1788 = fmul <4 x float> %1240, %1094, !dbg !163 + %1789 = extractelement <4 x float> %1094, i64 2, !dbg !164 + %1790 = fmul float %1236, %1789, !dbg !163 + %1791 = fmul float %1237, %1789, !dbg !163 + %1792 = extractelement <4 x float> %1094, i64 0, !dbg !164 + %1793 = fmul float %1238, %1792, !dbg !163 + %1794 = fmul float %1239, %1792, !dbg !163 + %1795 = fmul <4 x float> %1297, %1099, !dbg !165 + %1796 = extractelement <4 x float> %1099, i64 2, !dbg !166 + %1797 = fmul float %1293, %1796, !dbg !165 + %1798 = fmul float %1294, %1796, !dbg !165 + %1799 = extractelement <4 x float> %1099, i64 0, !dbg !166 + %1800 = fmul float %1295, %1799, !dbg !165 + %1801 = fmul float %1296, %1799, !dbg !165 + %1802 = fmul <4 x float> %1354, %1104, !dbg !167 + %1803 = extractelement <4 x float> %1104, i64 2, !dbg !168 + %1804 = fmul float %1350, %1803, !dbg !167 + %1805 = fmul float %1351, %1803, !dbg !167 + %1806 = extractelement <4 x float> %1104, i64 0, !dbg !168 + %1807 = fmul float %1352, %1806, !dbg !167 + %1808 = fmul float %1353, %1806, !dbg !167 + %1809 = fmul float %1419, %1792, !dbg !164 + %1810 = fmul float %1420, %1792, !dbg !164 + %1811 = fmul float %1421, %1789, !dbg !164 + %1812 = fmul float %1422, %1789, !dbg !164 + %1813 = fmul float %1423, %1792, !dbg !164 + %1814 = fmul float %1424, %1792, !dbg !164 + %1815 = fmul float %1425, %1789, !dbg !164 + %1816 = fmul float %1426, %1789, !dbg !164 + %1817 = fmul float %1491, %1799, !dbg !169 + %1818 = fmul float %1492, %1799, !dbg !169 + %1819 = fmul float %1493, %1796, !dbg !169 + %1820 = fmul float %1494, %1796, !dbg !169 + %1821 = fmul float %1495, %1799, !dbg !169 + %1822 = fmul float %1496, %1799, !dbg !169 + %1823 = fmul float %1497, %1796, !dbg !169 + %1824 = fmul float %1498, %1796, !dbg !169 + %1825 = fmul float %1563, %1799, !dbg !166 + %1826 = fmul float %1564, %1799, !dbg !166 + %1827 = fmul float %1565, %1796, !dbg !166 + %1828 = fmul float %1566, %1796, !dbg !166 + %1829 = fmul float %1567, %1799, !dbg !166 + %1830 = fmul float %1568, %1799, !dbg !166 + %1831 = fmul float %1569, %1796, !dbg !166 + %1832 = fmul float %1570, %1796, !dbg !166 + %1833 = fmul float %1635, %1806, !dbg !170 + %1834 = fmul float %1636, %1806, !dbg !170 + %1835 = fmul float %1637, %1803, !dbg !170 + %1836 = fmul float %1638, %1803, !dbg !170 + %1837 = fmul float %1639, %1806, !dbg !170 + %1838 = fmul float %1640, %1806, !dbg !170 + %1839 = fmul float %1641, %1803, !dbg !170 + %1840 = fmul float %1642, %1803, !dbg !170 + %1841 = fmul float %1707, %1806, !dbg !171 + %1842 = fmul float %1708, %1806, !dbg !171 + %1843 = fmul float %1709, %1803, !dbg !171 + %1844 = fmul float %1710, %1803, !dbg !171 + %1845 = fmul float %1711, %1806, !dbg !171 + %1846 = fmul float %1712, %1806, !dbg !171 + %1847 = fmul float %1713, %1803, !dbg !171 + %1848 = fmul float %1714, %1803, !dbg !171 + %1849 = fmul float %1779, %1806, !dbg !168 + %1850 = fmul float %1780, %1806, !dbg !168 + %1851 = fmul float %1781, %1803, !dbg !168 + %1852 = fmul float %1782, %1803, !dbg !168 + %1853 = fmul float %1783, %1806, !dbg !168 + %1854 = fmul float %1784, %1806, !dbg !168 + %1855 = fmul float %1785, %1803, !dbg !168 + %1856 = fmul float %1786, %1803, !dbg !168 + %1857 = fsub <4 x float> zeroinitializer, %1787, !dbg !172 + %1858 = fsub float 0.000000e+00, %1790, !dbg !173 + %1859 = fsub float 0.000000e+00, %1791, !dbg !173 + %1860 = fsub float 0.000000e+00, %1793, !dbg !173 + %1861 = fsub float 0.000000e+00, %1794, !dbg !173 + %1862 = fsub <4 x float> zeroinitializer, %1788, !dbg !173 + %1863 = fsub float 0.000000e+00, %1797, !dbg !174 + %1864 = fsub float 0.000000e+00, %1798, !dbg !174 + %1865 = fsub float 0.000000e+00, %1800, !dbg !174 + %1866 = fsub float 0.000000e+00, %1801, !dbg !174 + %1867 = fsub <4 x float> zeroinitializer, %1795, !dbg !174 + %1868 = fsub float 0.000000e+00, %1804, !dbg !175 + %1869 = fsub float 0.000000e+00, %1805, !dbg !175 + %1870 = fsub float 0.000000e+00, %1807, !dbg !175 + %1871 = fsub float 0.000000e+00, %1808, !dbg !175 + %1872 = fsub <4 x float> zeroinitializer, %1802, !dbg !175 + %1873 = sub nsw i64 %10, %11, !dbg !176 + %1874 = icmp sgt i64 %1873, 2, !dbg !177 + br i1 %1874, label %.lr.ph, label %._crit_edge, !dbg !177 + +.lr.ph: ; preds = %__nv_exp2f.exit475 + %1875 = shl nuw nsw i32 %28, 4 + %1876 = lshr exact i32 %29, 3 + %1877 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %1875 + %1878 = getelementptr inbounds nuw i8, ptr addrspace(3) %1877, i32 %1876 + %1879 = getelementptr inbounds nuw i8, ptr addrspace(3) %1878, i32 8 + %1880 = shl nuw nsw i32 %27, 1 + %1881 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %1880 + %1882 = tail call i64 @llvm.umin.i64(i64 %1873, i64 16), !dbg !177 + %1883 = getelementptr inbounds nuw i8, ptr addrspace(3) %1881, i32 4 + %1884 = extractelement <4 x float> %1857, i64 0, !dbg !178 + %1885 = extractelement <4 x float> %1857, i64 1, !dbg !178 + %1886 = extractelement <4 x float> %1857, i64 2, !dbg !178 + %1887 = extractelement <4 x float> %1857, i64 3, !dbg !178 + %1888 = zext nneg <2 x i32> %73 to <2 x i64> + br label %1889, !dbg !177 + +1889: ; preds = %.lr.ph, %1889 + %1890 = phi i64 [ 2, %.lr.ph ], [ %1978, %1889 ] + %1891 = phi <4 x float> [ %1857, %.lr.ph ], [ %1977, %1889 ] + %1892 = icmp eq i64 %1890, %94, !dbg !179 + %1893 = icmp eq i64 %1890, %92, !dbg !179 + %1894 = select i1 %1892, float %1884, float 0.000000e+00, !dbg !178 + %1895 = select i1 %1892, float %1885, float 0.000000e+00, !dbg !178 + %1896 = select i1 %1893, float %1886, float 0.000000e+00, !dbg !178 + %1897 = select i1 %1893, float %1887, float 0.000000e+00, !dbg !178 + %1898 = fadd float %1894, %1896, !dbg !180 + %1899 = fadd float %1895, %1897, !dbg !180 + %1900 = bitcast float %1898 to i32, !dbg !183 + %1901 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1900, i32 16, i32 31), !dbg !183 + %1902 = bitcast i32 %1901 to float, !dbg !183 + %1903 = fadd float %1898, %1902, !dbg !180 + %1904 = bitcast float %1903 to i32, !dbg !183 + %1905 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1904, i32 8, i32 31), !dbg !183 + %1906 = bitcast i32 %1905 to float, !dbg !183 + %1907 = bitcast float %1899 to i32, !dbg !183 + %1908 = insertelement <2 x i64> poison, i64 %1890, i64 0, !dbg !185 + %1909 = shufflevector <2 x i64> %1908, <2 x i64> poison, <2 x i32> zeroinitializer, !dbg !185 + %1910 = icmp samesign ugt <2 x i64> %1909, %1888, !dbg !185 + %1911 = fadd float %1903, %1906, !dbg !180 + %1912 = bitcast float %1911 to i32, !dbg !183 + %1913 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1912, i32 4, i32 31), !dbg !183 + %1914 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1907, i32 16, i32 31), !dbg !183 + %1915 = bitcast i32 %1914 to float, !dbg !183 + %1916 = fadd float %1899, %1915, !dbg !180 + %1917 = bitcast float %1916 to i32, !dbg !183 + %1918 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1917, i32 8, i32 31), !dbg !183 + %1919 = bitcast i32 %1918 to float, !dbg !183 + %1920 = fadd float %1916, %1919, !dbg !180 + %1921 = bitcast float %1920 to i32, !dbg !183 + %1922 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1921, i32 4, i32 31), !dbg !183 + %1923 = insertelement <2 x i32> poison, i32 %1913, i64 0, !dbg !183 + %1924 = insertelement <2 x i32> %1923, i32 %1922, i64 1, !dbg !183 + %1925 = bitcast <2 x i32> %1924 to <2 x float>, !dbg !183 + %1926 = insertelement <2 x float> poison, float %1911, i64 0, !dbg !180 + %1927 = insertelement <2 x float> %1926, float %1920, i64 1, !dbg !180 + %1928 = fadd <2 x float> %1927, %1925, !dbg !180 + %1929 = select <2 x i1> %1910, <2 x float> %1928, <2 x float> zeroinitializer, !dbg !186 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !187 + %1930 = extractelement <2 x float> %1929, i64 0, !dbg !187 + %1931 = bitcast float %1930 to <1 x i32>, !dbg !187 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) %1878, <1 x i32> %1931) #2, !dbg !187 + %1932 = extractelement <2 x float> %1929, i64 1, !dbg !187 + %1933 = bitcast float %1932 to <1 x i32>, !dbg !187 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %1879, <1 x i32> %1933) #2, !dbg !187 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !187 + %1934 = load float, ptr addrspace(3) %1881, align 8, !dbg !187 + %1935 = load float, ptr addrspace(3) %1883, align 4, !dbg !187 + %1936 = extractelement <4 x float> %1891, i64 0, !dbg !188 + %1937 = fmul float %1936, %1934, !dbg !188 + %1938 = extractelement <4 x float> %1891, i64 1, !dbg !188 + %1939 = fmul float %1938, %1934, !dbg !188 + %1940 = extractelement <4 x float> %1891, i64 2, !dbg !188 + %1941 = fmul float %1940, %1935, !dbg !188 + %1942 = extractelement <4 x float> %1891, i64 3, !dbg !188 + %1943 = fmul float %1942, %1935, !dbg !188 + %1944 = fadd float %1937, %1941, !dbg !189 + %1945 = fadd float %1939, %1943, !dbg !189 + %1946 = bitcast float %1944 to i32, !dbg !190 + %1947 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1946, i32 16, i32 31), !dbg !190 + %1948 = bitcast i32 %1947 to float, !dbg !190 + %1949 = fadd float %1944, %1948, !dbg !189 + %1950 = bitcast float %1949 to i32, !dbg !190 + %1951 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1950, i32 8, i32 31), !dbg !190 + %1952 = bitcast i32 %1951 to float, !dbg !190 + %1953 = fadd float %1949, %1952, !dbg !189 + %1954 = bitcast float %1953 to i32, !dbg !190 + %1955 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1954, i32 4, i32 31), !dbg !190 + %1956 = bitcast float %1945 to i32, !dbg !190 + %1957 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1956, i32 16, i32 31), !dbg !190 + %1958 = bitcast i32 %1957 to float, !dbg !190 + %1959 = fadd float %1945, %1958, !dbg !189 + %1960 = bitcast float %1959 to i32, !dbg !190 + %1961 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1960, i32 8, i32 31), !dbg !190 + %1962 = bitcast i32 %1961 to float, !dbg !190 + %1963 = fadd float %1959, %1962, !dbg !189 + %1964 = bitcast float %1963 to i32, !dbg !190 + %1965 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1964, i32 4, i32 31), !dbg !190 + %1966 = insertelement <2 x i32> poison, i32 %1955, i64 0, !dbg !190 + %1967 = insertelement <2 x i32> %1966, i32 %1965, i64 1, !dbg !190 + %1968 = bitcast <2 x i32> %1967 to <2 x float>, !dbg !190 + %1969 = insertelement <2 x float> poison, float %1953, i64 0, !dbg !189 + %1970 = insertelement <2 x float> %1969, float %1963, i64 1, !dbg !189 + %1971 = fadd <2 x float> %1970, %1968, !dbg !189 + %1972 = fadd <2 x float> %1929, %1971, !dbg !192 + %1973 = shufflevector <2 x float> %1972, <2 x float> poison, <4 x i32> , !dbg !192 + %1974 = insertelement <4 x i1> poison, i1 %1892, i64 0, !dbg !193 + %1975 = insertelement <4 x i1> %1974, i1 %1893, i64 1, !dbg !193 + %1976 = shufflevector <4 x i1> %1975, <4 x i1> poison, <4 x i32> , !dbg !193 + %1977 = select <4 x i1> %1976, <4 x float> %1973, <4 x float> %1891, !dbg !193 + %1978 = add nuw nsw i64 %1890, 1, !dbg !177 + %exitcond.not = icmp eq i64 %1978, %1882, !dbg !177 + br i1 %exitcond.not, label %._crit_edge, label %1889, !dbg !177 + +._crit_edge: ; preds = %1889, %__nv_exp2f.exit475 + %1979 = phi <4 x float> [ %1857, %__nv_exp2f.exit475 ], [ %1977, %1889 ] + %1980 = sub nsw i64 %10, %17, !dbg !194 + %1981 = icmp sgt i64 %1980, 2, !dbg !195 + br i1 %1981, label %.lr.ph503, label %._crit_edge504, !dbg !195 + +.lr.ph503: ; preds = %._crit_edge + %1982 = and i32 %23, 4 + %1983 = icmp eq i32 %1982, 0 + %1984 = shl nuw nsw i32 %23, 3 + %1985 = and i32 %1984, 24 + %1986 = lshr i32 %23, 3 + %1987 = and i32 %1986, 3 + %1988 = or disjoint i32 %1987, %1985 + %1989 = or disjoint i32 %1988, %1982 + %1990 = xor i32 %1989, 4 + %1991 = tail call i64 @llvm.umin.i64(i64 %1980, i64 16), !dbg !195 + %1992 = extractelement <4 x float> %1862, i64 0, !dbg !196 + %1993 = extractelement <4 x float> %1862, i64 1, !dbg !196 + %1994 = extractelement <4 x float> %1862, i64 2, !dbg !196 + %1995 = extractelement <4 x float> %1862, i64 3, !dbg !196 + %.mask757 = and <4 x i32> %58, splat (i32 255) + %1996 = zext nneg <4 x i32> %.mask757 to <4 x i64> + br label %1997, !dbg !195 + +1997: ; preds = %.lr.ph503, %1997 + %.pn79499 = phi float [ %1861, %.lr.ph503 ], [ %2160, %1997 ] + %.pn81498 = phi float [ %1860, %.lr.ph503 ], [ %2158, %1997 ] + %.pn83497 = phi float [ %1859, %.lr.ph503 ], [ %2156, %1997 ] + %.pn85496 = phi float [ %1858, %.lr.ph503 ], [ %2154, %1997 ] + %1998 = phi i64 [ 2, %.lr.ph503 ], [ %2165, %1997 ] + %1999 = phi <4 x float> [ %1862, %.lr.ph503 ], [ %2164, %1997 ] + %2000 = icmp eq i64 %1998, %94, !dbg !197 + %2001 = icmp eq i64 %1998, %92, !dbg !197 + %2002 = select i1 %2000, float %1992, float 0.000000e+00, !dbg !196 + %2003 = select i1 %2000, float %1993, float 0.000000e+00, !dbg !196 + %2004 = select i1 %2001, float %1858, float 0.000000e+00, !dbg !196 + %2005 = select i1 %2001, float %1859, float 0.000000e+00, !dbg !196 + %2006 = select i1 %2000, float %1860, float 0.000000e+00, !dbg !196 + %2007 = select i1 %2000, float %1861, float 0.000000e+00, !dbg !196 + %2008 = select i1 %2001, float %1994, float 0.000000e+00, !dbg !196 + %2009 = select i1 %2001, float %1995, float 0.000000e+00, !dbg !196 + %2010 = fadd float %2002, %2004, !dbg !198 + %2011 = fadd float %2003, %2005, !dbg !198 + %2012 = fadd float %2006, %2008, !dbg !198 + %2013 = fadd float %2007, %2009, !dbg !198 + %2014 = bitcast float %2010 to i32, !dbg !199 + %2015 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2014, i32 16, i32 31), !dbg !199 + %2016 = bitcast i32 %2015 to float, !dbg !199 + %2017 = fadd float %2010, %2016, !dbg !198 + %2018 = bitcast float %2017 to i32, !dbg !199 + %2019 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2018, i32 8, i32 31), !dbg !199 + %2020 = bitcast i32 %2019 to float, !dbg !199 + %2021 = bitcast float %2011 to i32, !dbg !199 + %2022 = bitcast float %2012 to i32, !dbg !199 + %2023 = bitcast float %2013 to i32, !dbg !199 + %2024 = fadd float %2017, %2020, !dbg !198 + %2025 = bitcast float %2024 to i32, !dbg !199 + %2026 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2025, i32 4, i32 31), !dbg !199 + %2027 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2021, i32 16, i32 31), !dbg !199 + %2028 = bitcast i32 %2027 to float, !dbg !199 + %2029 = fadd float %2011, %2028, !dbg !198 + %2030 = bitcast float %2029 to i32, !dbg !199 + %2031 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2030, i32 8, i32 31), !dbg !199 + %2032 = bitcast i32 %2031 to float, !dbg !199 + %2033 = fadd float %2029, %2032, !dbg !198 + %2034 = bitcast float %2033 to i32, !dbg !199 + %2035 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2034, i32 4, i32 31), !dbg !199 + %2036 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2022, i32 16, i32 31), !dbg !199 + %2037 = bitcast i32 %2036 to float, !dbg !199 + %2038 = fadd float %2012, %2037, !dbg !198 + %2039 = bitcast float %2038 to i32, !dbg !199 + %2040 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2039, i32 8, i32 31), !dbg !199 + %2041 = bitcast i32 %2040 to float, !dbg !199 + %2042 = fadd float %2038, %2041, !dbg !198 + %2043 = bitcast float %2042 to i32, !dbg !199 + %2044 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2043, i32 4, i32 31), !dbg !199 + %2045 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2023, i32 16, i32 31), !dbg !199 + %2046 = bitcast i32 %2045 to float, !dbg !199 + %2047 = fadd float %2013, %2046, !dbg !198 + %2048 = bitcast float %2047 to i32, !dbg !199 + %2049 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2048, i32 8, i32 31), !dbg !199 + %2050 = bitcast i32 %2049 to float, !dbg !199 + %2051 = fadd float %2047, %2050, !dbg !198 + %2052 = bitcast float %2051 to i32, !dbg !199 + %2053 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2052, i32 4, i32 31), !dbg !199 + %2054 = insertelement <4 x i32> poison, i32 %2026, i64 0, !dbg !199 + %2055 = insertelement <4 x i32> %2054, i32 %2035, i64 1, !dbg !199 + %2056 = insertelement <4 x i32> %2055, i32 %2044, i64 2, !dbg !199 + %2057 = insertelement <4 x i32> %2056, i32 %2053, i64 3, !dbg !199 + %2058 = bitcast <4 x i32> %2057 to <4 x float>, !dbg !199 + %2059 = insertelement <4 x float> poison, float %2024, i64 0, !dbg !198 + %2060 = insertelement <4 x float> %2059, float %2033, i64 1, !dbg !198 + %2061 = insertelement <4 x float> %2060, float %2042, i64 2, !dbg !198 + %2062 = insertelement <4 x float> %2061, float %2051, i64 3, !dbg !198 + %2063 = fadd <4 x float> %2062, %2058, !dbg !198 + %2064 = insertelement <4 x i64> poison, i64 %1998, i64 0, !dbg !201 + %2065 = shufflevector <4 x i64> %2064, <4 x i64> poison, <4 x i32> zeroinitializer, !dbg !201 + %2066 = icmp samesign ugt <4 x i64> %2065, %1996, !dbg !201 + %2067 = select <4 x i1> %2066, <4 x float> %2063, <4 x float> zeroinitializer, !dbg !202 + %2068 = extractelement <4 x float> %2067, i64 0, !dbg !203 + %2069 = extractelement <4 x float> %2067, i64 1, !dbg !203 + %2070 = select i1 %1983, float %2068, float %2069, !dbg !203 + %2071 = select i1 %1983, float %2069, float %2068, !dbg !203 + %2072 = extractelement <4 x float> %2067, i64 2, !dbg !203 + %2073 = extractelement <4 x float> %2067, i64 3, !dbg !203 + %2074 = select i1 %1983, float %2072, float %2073, !dbg !203 + %2075 = select i1 %1983, float %2073, float %2072, !dbg !203 + %2076 = bitcast float %2070 to i32, !dbg !203 + %2077 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2076, i32 %1989, i32 31), !dbg !203 + %2078 = bitcast float %2074 to i32, !dbg !203 + %2079 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2078, i32 %1989, i32 31), !dbg !203 + %2080 = bitcast float %2071 to i32, !dbg !203 + %2081 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2080, i32 %1990, i32 31), !dbg !203 + %2082 = bitcast float %2075 to i32, !dbg !203 + %2083 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2082, i32 %1990, i32 31), !dbg !203 + %.v198 = select i1 %1983, i32 %2077, i32 %2081, !dbg !203 + %2084 = bitcast i32 %.v198 to float, !dbg !203 + %.v199 = select i1 %1983, i32 %2079, i32 %2083, !dbg !203 + %2085 = bitcast i32 %.v199 to float, !dbg !203 + %2086 = extractelement <4 x float> %1999, i64 0, !dbg !204 + %2087 = fmul float %2086, %2084, !dbg !204 + %2088 = extractelement <4 x float> %1999, i64 1, !dbg !204 + %2089 = fmul float %2088, %2084, !dbg !204 + %2090 = fmul float %.pn85496, %2085, !dbg !204 + %2091 = fmul float %.pn83497, %2085, !dbg !204 + %2092 = fmul float %.pn81498, %2084, !dbg !204 + %2093 = fmul float %.pn79499, %2084, !dbg !204 + %2094 = extractelement <4 x float> %1999, i64 2, !dbg !204 + %2095 = fmul float %2094, %2085, !dbg !204 + %2096 = extractelement <4 x float> %1999, i64 3, !dbg !204 + %2097 = fmul float %2096, %2085, !dbg !204 + %2098 = fadd float %2087, %2090, !dbg !205 + %2099 = fadd float %2089, %2091, !dbg !205 + %2100 = fadd float %2092, %2095, !dbg !205 + %2101 = fadd float %2093, %2097, !dbg !205 + %2102 = bitcast float %2098 to i32, !dbg !206 + %2103 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2102, i32 16, i32 31), !dbg !206 + %2104 = bitcast i32 %2103 to float, !dbg !206 + %2105 = fadd float %2098, %2104, !dbg !205 + %2106 = bitcast float %2105 to i32, !dbg !206 + %2107 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2106, i32 8, i32 31), !dbg !206 + %2108 = bitcast i32 %2107 to float, !dbg !206 + %2109 = fadd float %2105, %2108, !dbg !205 + %2110 = bitcast float %2109 to i32, !dbg !206 + %2111 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2110, i32 4, i32 31), !dbg !206 + %2112 = bitcast float %2099 to i32, !dbg !206 + %2113 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2112, i32 16, i32 31), !dbg !206 + %2114 = bitcast i32 %2113 to float, !dbg !206 + %2115 = fadd float %2099, %2114, !dbg !205 + %2116 = bitcast float %2115 to i32, !dbg !206 + %2117 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2116, i32 8, i32 31), !dbg !206 + %2118 = bitcast i32 %2117 to float, !dbg !206 + %2119 = fadd float %2115, %2118, !dbg !205 + %2120 = bitcast float %2119 to i32, !dbg !206 + %2121 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2120, i32 4, i32 31), !dbg !206 + %2122 = bitcast float %2100 to i32, !dbg !206 + %2123 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2122, i32 16, i32 31), !dbg !206 + %2124 = bitcast i32 %2123 to float, !dbg !206 + %2125 = fadd float %2100, %2124, !dbg !205 + %2126 = bitcast float %2125 to i32, !dbg !206 + %2127 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2126, i32 8, i32 31), !dbg !206 + %2128 = bitcast i32 %2127 to float, !dbg !206 + %2129 = fadd float %2125, %2128, !dbg !205 + %2130 = bitcast float %2129 to i32, !dbg !206 + %2131 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2130, i32 4, i32 31), !dbg !206 + %2132 = bitcast float %2101 to i32, !dbg !206 + %2133 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2132, i32 16, i32 31), !dbg !206 + %2134 = bitcast i32 %2133 to float, !dbg !206 + %2135 = fadd float %2101, %2134, !dbg !205 + %2136 = bitcast float %2135 to i32, !dbg !206 + %2137 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2136, i32 8, i32 31), !dbg !206 + %2138 = bitcast i32 %2137 to float, !dbg !206 + %2139 = fadd float %2135, %2138, !dbg !205 + %2140 = bitcast float %2139 to i32, !dbg !206 + %2141 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2140, i32 4, i32 31), !dbg !206 + %2142 = insertelement <4 x i32> poison, i32 %2111, i64 0, !dbg !206 + %2143 = insertelement <4 x i32> %2142, i32 %2121, i64 1, !dbg !206 + %2144 = insertelement <4 x i32> %2143, i32 %2131, i64 2, !dbg !206 + %2145 = insertelement <4 x i32> %2144, i32 %2141, i64 3, !dbg !206 + %2146 = bitcast <4 x i32> %2145 to <4 x float>, !dbg !206 + %2147 = insertelement <4 x float> poison, float %2109, i64 0, !dbg !205 + %2148 = insertelement <4 x float> %2147, float %2119, i64 1, !dbg !205 + %2149 = insertelement <4 x float> %2148, float %2129, i64 2, !dbg !205 + %2150 = insertelement <4 x float> %2149, float %2139, i64 3, !dbg !205 + %2151 = fadd <4 x float> %2150, %2146, !dbg !205 + %2152 = fadd <4 x float> %2067, %2151, !dbg !208 + %2153 = extractelement <4 x float> %2152, i64 0, !dbg !209 + %2154 = select i1 %2001, float %2153, float %.pn85496, !dbg !209 + %2155 = extractelement <4 x float> %2152, i64 1, !dbg !209 + %2156 = select i1 %2001, float %2155, float %.pn83497, !dbg !209 + %2157 = extractelement <4 x float> %2152, i64 2, !dbg !209 + %2158 = select i1 %2000, float %2157, float %.pn81498, !dbg !209 + %2159 = extractelement <4 x float> %2152, i64 3, !dbg !209 + %2160 = select i1 %2000, float %2159, float %.pn79499, !dbg !209 + %2161 = insertelement <4 x i1> poison, i1 %2000, i64 0, !dbg !209 + %2162 = insertelement <4 x i1> %2161, i1 %2001, i64 1, !dbg !209 + %2163 = shufflevector <4 x i1> %2162, <4 x i1> poison, <4 x i32> , !dbg !209 + %2164 = select <4 x i1> %2163, <4 x float> %2152, <4 x float> %1999, !dbg !209 + %2165 = add nuw nsw i64 %1998, 1, !dbg !195 + %exitcond597.not = icmp eq i64 %2165, %1991, !dbg !195 + br i1 %exitcond597.not, label %._crit_edge504.loopexit, label %1997, !dbg !195 + +._crit_edge504.loopexit: ; preds = %1997 + %2166 = fadd float %2154, 0.000000e+00, !dbg !210 + %2167 = fadd float %2156, 0.000000e+00, !dbg !210 + %2168 = fadd float %2158, 0.000000e+00, !dbg !210 + %2169 = fadd float %2160, 0.000000e+00, !dbg !210 + br label %._crit_edge504, !dbg !211 + +._crit_edge504: ; preds = %._crit_edge504.loopexit, %._crit_edge + %.pn85.lcssa = phi float [ %1858, %._crit_edge ], [ %2166, %._crit_edge504.loopexit ] + %.pn83.lcssa = phi float [ %1859, %._crit_edge ], [ %2167, %._crit_edge504.loopexit ] + %.pn81.lcssa = phi float [ %1860, %._crit_edge ], [ %2168, %._crit_edge504.loopexit ] + %.pn79.lcssa = phi float [ %1861, %._crit_edge ], [ %2169, %._crit_edge504.loopexit ] + %2170 = phi <4 x float> [ %1862, %._crit_edge ], [ %2164, %._crit_edge504.loopexit ] + %2171 = sub nsw i64 %10, %18, !dbg !211 + %2172 = icmp sgt i64 %2171, 2, !dbg !212 + br i1 %2172, label %.lr.ph522, label %._crit_edge523, !dbg !212 + +.lr.ph522: ; preds = %._crit_edge504 + %2173 = and i32 %23, 4 + %2174 = icmp eq i32 %2173, 0 + %2175 = shl nuw nsw i32 %23, 3 + %2176 = and i32 %2175, 24 + %2177 = lshr i32 %23, 3 + %2178 = and i32 %2177, 3 + %2179 = or disjoint i32 %2178, %2176 + %2180 = or disjoint i32 %2179, %2173 + %2181 = xor i32 %2180, 4 + %2182 = tail call i64 @llvm.umin.i64(i64 %2171, i64 16), !dbg !212 + %2183 = extractelement <4 x float> %1867, i64 0, !dbg !213 + %2184 = extractelement <4 x float> %1867, i64 1, !dbg !213 + %2185 = extractelement <4 x float> %1867, i64 2, !dbg !213 + %2186 = extractelement <4 x float> %1867, i64 3, !dbg !213 + %.mask758 = and <4 x i32> %58, splat (i32 255) + %2187 = zext nneg <4 x i32> %.mask758 to <4 x i64> + br label %2188, !dbg !212 + +2188: ; preds = %.lr.ph522, %2188 + %.pn95518 = phi float [ %1866, %.lr.ph522 ], [ %2351, %2188 ] + %.pn97517 = phi float [ %1865, %.lr.ph522 ], [ %2349, %2188 ] + %.pn99516 = phi float [ %1864, %.lr.ph522 ], [ %2347, %2188 ] + %.pn101515 = phi float [ %1863, %.lr.ph522 ], [ %2345, %2188 ] + %2189 = phi i64 [ 2, %.lr.ph522 ], [ %2356, %2188 ] + %2190 = phi <4 x float> [ %1867, %.lr.ph522 ], [ %2355, %2188 ] + %2191 = icmp eq i64 %2189, %94, !dbg !214 + %2192 = icmp eq i64 %2189, %92, !dbg !214 + %2193 = select i1 %2191, float %2183, float 0.000000e+00, !dbg !213 + %2194 = select i1 %2191, float %2184, float 0.000000e+00, !dbg !213 + %2195 = select i1 %2192, float %1863, float 0.000000e+00, !dbg !213 + %2196 = select i1 %2192, float %1864, float 0.000000e+00, !dbg !213 + %2197 = select i1 %2191, float %1865, float 0.000000e+00, !dbg !213 + %2198 = select i1 %2191, float %1866, float 0.000000e+00, !dbg !213 + %2199 = select i1 %2192, float %2185, float 0.000000e+00, !dbg !213 + %2200 = select i1 %2192, float %2186, float 0.000000e+00, !dbg !213 + %2201 = fadd float %2193, %2195, !dbg !215 + %2202 = fadd float %2194, %2196, !dbg !215 + %2203 = fadd float %2197, %2199, !dbg !215 + %2204 = fadd float %2198, %2200, !dbg !215 + %2205 = bitcast float %2201 to i32, !dbg !216 + %2206 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2205, i32 16, i32 31), !dbg !216 + %2207 = bitcast i32 %2206 to float, !dbg !216 + %2208 = fadd float %2201, %2207, !dbg !215 + %2209 = bitcast float %2208 to i32, !dbg !216 + %2210 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2209, i32 8, i32 31), !dbg !216 + %2211 = bitcast i32 %2210 to float, !dbg !216 + %2212 = bitcast float %2202 to i32, !dbg !216 + %2213 = bitcast float %2203 to i32, !dbg !216 + %2214 = bitcast float %2204 to i32, !dbg !216 + %2215 = fadd float %2208, %2211, !dbg !215 + %2216 = bitcast float %2215 to i32, !dbg !216 + %2217 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2216, i32 4, i32 31), !dbg !216 + %2218 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2212, i32 16, i32 31), !dbg !216 + %2219 = bitcast i32 %2218 to float, !dbg !216 + %2220 = fadd float %2202, %2219, !dbg !215 + %2221 = bitcast float %2220 to i32, !dbg !216 + %2222 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2221, i32 8, i32 31), !dbg !216 + %2223 = bitcast i32 %2222 to float, !dbg !216 + %2224 = fadd float %2220, %2223, !dbg !215 + %2225 = bitcast float %2224 to i32, !dbg !216 + %2226 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2225, i32 4, i32 31), !dbg !216 + %2227 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2213, i32 16, i32 31), !dbg !216 + %2228 = bitcast i32 %2227 to float, !dbg !216 + %2229 = fadd float %2203, %2228, !dbg !215 + %2230 = bitcast float %2229 to i32, !dbg !216 + %2231 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2230, i32 8, i32 31), !dbg !216 + %2232 = bitcast i32 %2231 to float, !dbg !216 + %2233 = fadd float %2229, %2232, !dbg !215 + %2234 = bitcast float %2233 to i32, !dbg !216 + %2235 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2234, i32 4, i32 31), !dbg !216 + %2236 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2214, i32 16, i32 31), !dbg !216 + %2237 = bitcast i32 %2236 to float, !dbg !216 + %2238 = fadd float %2204, %2237, !dbg !215 + %2239 = bitcast float %2238 to i32, !dbg !216 + %2240 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2239, i32 8, i32 31), !dbg !216 + %2241 = bitcast i32 %2240 to float, !dbg !216 + %2242 = fadd float %2238, %2241, !dbg !215 + %2243 = bitcast float %2242 to i32, !dbg !216 + %2244 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2243, i32 4, i32 31), !dbg !216 + %2245 = insertelement <4 x i32> poison, i32 %2217, i64 0, !dbg !216 + %2246 = insertelement <4 x i32> %2245, i32 %2226, i64 1, !dbg !216 + %2247 = insertelement <4 x i32> %2246, i32 %2235, i64 2, !dbg !216 + %2248 = insertelement <4 x i32> %2247, i32 %2244, i64 3, !dbg !216 + %2249 = bitcast <4 x i32> %2248 to <4 x float>, !dbg !216 + %2250 = insertelement <4 x float> poison, float %2215, i64 0, !dbg !215 + %2251 = insertelement <4 x float> %2250, float %2224, i64 1, !dbg !215 + %2252 = insertelement <4 x float> %2251, float %2233, i64 2, !dbg !215 + %2253 = insertelement <4 x float> %2252, float %2242, i64 3, !dbg !215 + %2254 = fadd <4 x float> %2253, %2249, !dbg !215 + %2255 = insertelement <4 x i64> poison, i64 %2189, i64 0, !dbg !218 + %2256 = shufflevector <4 x i64> %2255, <4 x i64> poison, <4 x i32> zeroinitializer, !dbg !218 + %2257 = icmp samesign ugt <4 x i64> %2256, %2187, !dbg !218 + %2258 = select <4 x i1> %2257, <4 x float> %2254, <4 x float> zeroinitializer, !dbg !219 + %2259 = extractelement <4 x float> %2258, i64 0, !dbg !220 + %2260 = extractelement <4 x float> %2258, i64 1, !dbg !220 + %2261 = select i1 %2174, float %2259, float %2260, !dbg !220 + %2262 = select i1 %2174, float %2260, float %2259, !dbg !220 + %2263 = extractelement <4 x float> %2258, i64 2, !dbg !220 + %2264 = extractelement <4 x float> %2258, i64 3, !dbg !220 + %2265 = select i1 %2174, float %2263, float %2264, !dbg !220 + %2266 = select i1 %2174, float %2264, float %2263, !dbg !220 + %2267 = bitcast float %2261 to i32, !dbg !220 + %2268 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2267, i32 %2180, i32 31), !dbg !220 + %2269 = bitcast float %2265 to i32, !dbg !220 + %2270 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2269, i32 %2180, i32 31), !dbg !220 + %2271 = bitcast float %2262 to i32, !dbg !220 + %2272 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2271, i32 %2181, i32 31), !dbg !220 + %2273 = bitcast float %2266 to i32, !dbg !220 + %2274 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2273, i32 %2181, i32 31), !dbg !220 + %.v196 = select i1 %2174, i32 %2268, i32 %2272, !dbg !220 + %2275 = bitcast i32 %.v196 to float, !dbg !220 + %.v197 = select i1 %2174, i32 %2270, i32 %2274, !dbg !220 + %2276 = bitcast i32 %.v197 to float, !dbg !220 + %2277 = extractelement <4 x float> %2190, i64 0, !dbg !221 + %2278 = fmul float %2277, %2275, !dbg !221 + %2279 = extractelement <4 x float> %2190, i64 1, !dbg !221 + %2280 = fmul float %2279, %2275, !dbg !221 + %2281 = fmul float %.pn101515, %2276, !dbg !221 + %2282 = fmul float %.pn99516, %2276, !dbg !221 + %2283 = fmul float %.pn97517, %2275, !dbg !221 + %2284 = fmul float %.pn95518, %2275, !dbg !221 + %2285 = extractelement <4 x float> %2190, i64 2, !dbg !221 + %2286 = fmul float %2285, %2276, !dbg !221 + %2287 = extractelement <4 x float> %2190, i64 3, !dbg !221 + %2288 = fmul float %2287, %2276, !dbg !221 + %2289 = fadd float %2278, %2281, !dbg !222 + %2290 = fadd float %2280, %2282, !dbg !222 + %2291 = fadd float %2283, %2286, !dbg !222 + %2292 = fadd float %2284, %2288, !dbg !222 + %2293 = bitcast float %2289 to i32, !dbg !223 + %2294 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2293, i32 16, i32 31), !dbg !223 + %2295 = bitcast i32 %2294 to float, !dbg !223 + %2296 = fadd float %2289, %2295, !dbg !222 + %2297 = bitcast float %2296 to i32, !dbg !223 + %2298 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2297, i32 8, i32 31), !dbg !223 + %2299 = bitcast i32 %2298 to float, !dbg !223 + %2300 = fadd float %2296, %2299, !dbg !222 + %2301 = bitcast float %2300 to i32, !dbg !223 + %2302 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2301, i32 4, i32 31), !dbg !223 + %2303 = bitcast float %2290 to i32, !dbg !223 + %2304 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2303, i32 16, i32 31), !dbg !223 + %2305 = bitcast i32 %2304 to float, !dbg !223 + %2306 = fadd float %2290, %2305, !dbg !222 + %2307 = bitcast float %2306 to i32, !dbg !223 + %2308 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2307, i32 8, i32 31), !dbg !223 + %2309 = bitcast i32 %2308 to float, !dbg !223 + %2310 = fadd float %2306, %2309, !dbg !222 + %2311 = bitcast float %2310 to i32, !dbg !223 + %2312 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2311, i32 4, i32 31), !dbg !223 + %2313 = bitcast float %2291 to i32, !dbg !223 + %2314 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2313, i32 16, i32 31), !dbg !223 + %2315 = bitcast i32 %2314 to float, !dbg !223 + %2316 = fadd float %2291, %2315, !dbg !222 + %2317 = bitcast float %2316 to i32, !dbg !223 + %2318 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2317, i32 8, i32 31), !dbg !223 + %2319 = bitcast i32 %2318 to float, !dbg !223 + %2320 = fadd float %2316, %2319, !dbg !222 + %2321 = bitcast float %2320 to i32, !dbg !223 + %2322 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2321, i32 4, i32 31), !dbg !223 + %2323 = bitcast float %2292 to i32, !dbg !223 + %2324 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2323, i32 16, i32 31), !dbg !223 + %2325 = bitcast i32 %2324 to float, !dbg !223 + %2326 = fadd float %2292, %2325, !dbg !222 + %2327 = bitcast float %2326 to i32, !dbg !223 + %2328 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2327, i32 8, i32 31), !dbg !223 + %2329 = bitcast i32 %2328 to float, !dbg !223 + %2330 = fadd float %2326, %2329, !dbg !222 + %2331 = bitcast float %2330 to i32, !dbg !223 + %2332 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2331, i32 4, i32 31), !dbg !223 + %2333 = insertelement <4 x i32> poison, i32 %2302, i64 0, !dbg !223 + %2334 = insertelement <4 x i32> %2333, i32 %2312, i64 1, !dbg !223 + %2335 = insertelement <4 x i32> %2334, i32 %2322, i64 2, !dbg !223 + %2336 = insertelement <4 x i32> %2335, i32 %2332, i64 3, !dbg !223 + %2337 = bitcast <4 x i32> %2336 to <4 x float>, !dbg !223 + %2338 = insertelement <4 x float> poison, float %2300, i64 0, !dbg !222 + %2339 = insertelement <4 x float> %2338, float %2310, i64 1, !dbg !222 + %2340 = insertelement <4 x float> %2339, float %2320, i64 2, !dbg !222 + %2341 = insertelement <4 x float> %2340, float %2330, i64 3, !dbg !222 + %2342 = fadd <4 x float> %2341, %2337, !dbg !222 + %2343 = fadd <4 x float> %2258, %2342, !dbg !225 + %2344 = extractelement <4 x float> %2343, i64 0, !dbg !226 + %2345 = select i1 %2192, float %2344, float %.pn101515, !dbg !226 + %2346 = extractelement <4 x float> %2343, i64 1, !dbg !226 + %2347 = select i1 %2192, float %2346, float %.pn99516, !dbg !226 + %2348 = extractelement <4 x float> %2343, i64 2, !dbg !226 + %2349 = select i1 %2191, float %2348, float %.pn97517, !dbg !226 + %2350 = extractelement <4 x float> %2343, i64 3, !dbg !226 + %2351 = select i1 %2191, float %2350, float %.pn95518, !dbg !226 + %2352 = insertelement <4 x i1> poison, i1 %2191, i64 0, !dbg !226 + %2353 = insertelement <4 x i1> %2352, i1 %2192, i64 1, !dbg !226 + %2354 = shufflevector <4 x i1> %2353, <4 x i1> poison, <4 x i32> , !dbg !226 + %2355 = select <4 x i1> %2354, <4 x float> %2343, <4 x float> %2190, !dbg !226 + %2356 = add nuw nsw i64 %2189, 1, !dbg !212 + %exitcond599.not = icmp eq i64 %2356, %2182, !dbg !212 + br i1 %exitcond599.not, label %._crit_edge523.loopexit, label %2188, !dbg !212 + +._crit_edge523.loopexit: ; preds = %2188 + %2357 = fadd float %2345, 0.000000e+00, !dbg !227 + %2358 = fadd float %2347, 0.000000e+00, !dbg !227 + %2359 = fadd float %2349, 0.000000e+00, !dbg !227 + %2360 = fadd float %2351, 0.000000e+00, !dbg !227 + br label %._crit_edge523, !dbg !228 + +._crit_edge523: ; preds = %._crit_edge523.loopexit, %._crit_edge504 + %.pn101.lcssa = phi float [ %1863, %._crit_edge504 ], [ %2357, %._crit_edge523.loopexit ] + %.pn99.lcssa = phi float [ %1864, %._crit_edge504 ], [ %2358, %._crit_edge523.loopexit ] + %.pn97.lcssa = phi float [ %1865, %._crit_edge504 ], [ %2359, %._crit_edge523.loopexit ] + %.pn95.lcssa = phi float [ %1866, %._crit_edge504 ], [ %2360, %._crit_edge523.loopexit ] + %2361 = phi <4 x float> [ %1867, %._crit_edge504 ], [ %2355, %._crit_edge523.loopexit ] + %2362 = sub nsw i64 %10, %19, !dbg !228 + %2363 = icmp sgt i64 %2362, 2, !dbg !229 + %2364 = and i32 %23, 4 + br i1 %2363, label %.lr.ph541, label %._crit_edge523.._crit_edge542_crit_edge, !dbg !229 + +._crit_edge523.._crit_edge542_crit_edge: ; preds = %._crit_edge523 + %.pre602 = shl nuw nsw i32 %23, 3, !dbg !164 + %.pre604 = and i32 %.pre602, 24, !dbg !164 + %.pre606 = lshr i32 %23, 3, !dbg !164 + %.pre608 = and i32 %.pre606, 3, !dbg !164 + br label %._crit_edge542, !dbg !229 + +.lr.ph541: ; preds = %._crit_edge523 + %2365 = icmp eq i32 %2364, 0 + %2366 = shl nuw nsw i32 %23, 3 + %2367 = and i32 %2366, 24 + %2368 = lshr i32 %23, 3 + %2369 = and i32 %2368, 3 + %2370 = or disjoint i32 %2369, %2367 + %2371 = or disjoint i32 %2370, %2364 + %2372 = xor i32 %2371, 4 + %2373 = tail call i64 @llvm.umin.i64(i64 %2362, i64 16), !dbg !229 + %2374 = extractelement <4 x float> %1872, i64 0, !dbg !230 + %2375 = extractelement <4 x float> %1872, i64 1, !dbg !230 + %2376 = extractelement <4 x float> %1872, i64 2, !dbg !230 + %2377 = extractelement <4 x float> %1872, i64 3, !dbg !230 + %.mask759 = and <4 x i32> %58, splat (i32 255) + %2378 = zext nneg <4 x i32> %.mask759 to <4 x i64> + br label %2379, !dbg !229 + +2379: ; preds = %.lr.ph541, %2379 + %.pn111537 = phi float [ %1871, %.lr.ph541 ], [ %2542, %2379 ] + %.pn113536 = phi float [ %1870, %.lr.ph541 ], [ %2540, %2379 ] + %.pn115535 = phi float [ %1869, %.lr.ph541 ], [ %2538, %2379 ] + %.pn117534 = phi float [ %1868, %.lr.ph541 ], [ %2536, %2379 ] + %2380 = phi i64 [ 2, %.lr.ph541 ], [ %2547, %2379 ] + %2381 = phi <4 x float> [ %1872, %.lr.ph541 ], [ %2546, %2379 ] + %2382 = icmp eq i64 %2380, %94, !dbg !231 + %2383 = icmp eq i64 %2380, %92, !dbg !231 + %2384 = select i1 %2382, float %2374, float 0.000000e+00, !dbg !230 + %2385 = select i1 %2382, float %2375, float 0.000000e+00, !dbg !230 + %2386 = select i1 %2383, float %1868, float 0.000000e+00, !dbg !230 + %2387 = select i1 %2383, float %1869, float 0.000000e+00, !dbg !230 + %2388 = select i1 %2382, float %1870, float 0.000000e+00, !dbg !230 + %2389 = select i1 %2382, float %1871, float 0.000000e+00, !dbg !230 + %2390 = select i1 %2383, float %2376, float 0.000000e+00, !dbg !230 + %2391 = select i1 %2383, float %2377, float 0.000000e+00, !dbg !230 + %2392 = fadd float %2384, %2386, !dbg !232 + %2393 = fadd float %2385, %2387, !dbg !232 + %2394 = fadd float %2388, %2390, !dbg !232 + %2395 = fadd float %2389, %2391, !dbg !232 + %2396 = bitcast float %2392 to i32, !dbg !233 + %2397 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2396, i32 16, i32 31), !dbg !233 + %2398 = bitcast i32 %2397 to float, !dbg !233 + %2399 = fadd float %2392, %2398, !dbg !232 + %2400 = bitcast float %2399 to i32, !dbg !233 + %2401 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2400, i32 8, i32 31), !dbg !233 + %2402 = bitcast i32 %2401 to float, !dbg !233 + %2403 = bitcast float %2393 to i32, !dbg !233 + %2404 = bitcast float %2394 to i32, !dbg !233 + %2405 = bitcast float %2395 to i32, !dbg !233 + %2406 = insertelement <4 x i64> poison, i64 %2380, i64 0, !dbg !235 + %2407 = shufflevector <4 x i64> %2406, <4 x i64> poison, <4 x i32> zeroinitializer, !dbg !235 + %2408 = icmp samesign ugt <4 x i64> %2407, %2378, !dbg !235 + %2409 = fadd float %2399, %2402, !dbg !232 + %2410 = bitcast float %2409 to i32, !dbg !233 + %2411 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2410, i32 4, i32 31), !dbg !233 + %2412 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2403, i32 16, i32 31), !dbg !233 + %2413 = bitcast i32 %2412 to float, !dbg !233 + %2414 = fadd float %2393, %2413, !dbg !232 + %2415 = bitcast float %2414 to i32, !dbg !233 + %2416 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2415, i32 8, i32 31), !dbg !233 + %2417 = bitcast i32 %2416 to float, !dbg !233 + %2418 = fadd float %2414, %2417, !dbg !232 + %2419 = bitcast float %2418 to i32, !dbg !233 + %2420 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2419, i32 4, i32 31), !dbg !233 + %2421 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2404, i32 16, i32 31), !dbg !233 + %2422 = bitcast i32 %2421 to float, !dbg !233 + %2423 = fadd float %2394, %2422, !dbg !232 + %2424 = bitcast float %2423 to i32, !dbg !233 + %2425 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2424, i32 8, i32 31), !dbg !233 + %2426 = bitcast i32 %2425 to float, !dbg !233 + %2427 = fadd float %2423, %2426, !dbg !232 + %2428 = bitcast float %2427 to i32, !dbg !233 + %2429 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2428, i32 4, i32 31), !dbg !233 + %2430 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2405, i32 16, i32 31), !dbg !233 + %2431 = bitcast i32 %2430 to float, !dbg !233 + %2432 = fadd float %2395, %2431, !dbg !232 + %2433 = bitcast float %2432 to i32, !dbg !233 + %2434 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2433, i32 8, i32 31), !dbg !233 + %2435 = bitcast i32 %2434 to float, !dbg !233 + %2436 = fadd float %2432, %2435, !dbg !232 + %2437 = bitcast float %2436 to i32, !dbg !233 + %2438 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2437, i32 4, i32 31), !dbg !233 + %2439 = insertelement <4 x i32> poison, i32 %2411, i64 0, !dbg !233 + %2440 = insertelement <4 x i32> %2439, i32 %2420, i64 1, !dbg !233 + %2441 = insertelement <4 x i32> %2440, i32 %2429, i64 2, !dbg !233 + %2442 = insertelement <4 x i32> %2441, i32 %2438, i64 3, !dbg !233 + %2443 = bitcast <4 x i32> %2442 to <4 x float>, !dbg !233 + %2444 = insertelement <4 x float> poison, float %2409, i64 0, !dbg !232 + %2445 = insertelement <4 x float> %2444, float %2418, i64 1, !dbg !232 + %2446 = insertelement <4 x float> %2445, float %2427, i64 2, !dbg !232 + %2447 = insertelement <4 x float> %2446, float %2436, i64 3, !dbg !232 + %2448 = fadd <4 x float> %2447, %2443, !dbg !232 + %2449 = select <4 x i1> %2408, <4 x float> %2448, <4 x float> zeroinitializer, !dbg !236 + %2450 = extractelement <4 x float> %2449, i64 0, !dbg !237 + %2451 = extractelement <4 x float> %2449, i64 1, !dbg !237 + %2452 = select i1 %2365, float %2450, float %2451, !dbg !237 + %2453 = select i1 %2365, float %2451, float %2450, !dbg !237 + %2454 = extractelement <4 x float> %2449, i64 2, !dbg !237 + %2455 = extractelement <4 x float> %2449, i64 3, !dbg !237 + %2456 = select i1 %2365, float %2454, float %2455, !dbg !237 + %2457 = select i1 %2365, float %2455, float %2454, !dbg !237 + %2458 = bitcast float %2452 to i32, !dbg !237 + %2459 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2458, i32 %2371, i32 31), !dbg !237 + %2460 = bitcast float %2456 to i32, !dbg !237 + %2461 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2460, i32 %2371, i32 31), !dbg !237 + %2462 = bitcast float %2453 to i32, !dbg !237 + %2463 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2462, i32 %2372, i32 31), !dbg !237 + %2464 = bitcast float %2457 to i32, !dbg !237 + %2465 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2464, i32 %2372, i32 31), !dbg !237 + %.v194 = select i1 %2365, i32 %2459, i32 %2463, !dbg !237 + %2466 = bitcast i32 %.v194 to float, !dbg !237 + %.v195 = select i1 %2365, i32 %2461, i32 %2465, !dbg !237 + %2467 = bitcast i32 %.v195 to float, !dbg !237 + %2468 = extractelement <4 x float> %2381, i64 0, !dbg !238 + %2469 = fmul float %2468, %2466, !dbg !238 + %2470 = extractelement <4 x float> %2381, i64 1, !dbg !238 + %2471 = fmul float %2470, %2466, !dbg !238 + %2472 = fmul float %.pn117534, %2467, !dbg !238 + %2473 = fmul float %.pn115535, %2467, !dbg !238 + %2474 = fmul float %.pn113536, %2466, !dbg !238 + %2475 = fmul float %.pn111537, %2466, !dbg !238 + %2476 = extractelement <4 x float> %2381, i64 2, !dbg !238 + %2477 = fmul float %2476, %2467, !dbg !238 + %2478 = extractelement <4 x float> %2381, i64 3, !dbg !238 + %2479 = fmul float %2478, %2467, !dbg !238 + %2480 = fadd float %2469, %2472, !dbg !239 + %2481 = fadd float %2471, %2473, !dbg !239 + %2482 = fadd float %2474, %2477, !dbg !239 + %2483 = fadd float %2475, %2479, !dbg !239 + %2484 = bitcast float %2480 to i32, !dbg !240 + %2485 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2484, i32 16, i32 31), !dbg !240 + %2486 = bitcast i32 %2485 to float, !dbg !240 + %2487 = fadd float %2480, %2486, !dbg !239 + %2488 = bitcast float %2487 to i32, !dbg !240 + %2489 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2488, i32 8, i32 31), !dbg !240 + %2490 = bitcast i32 %2489 to float, !dbg !240 + %2491 = fadd float %2487, %2490, !dbg !239 + %2492 = bitcast float %2491 to i32, !dbg !240 + %2493 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2492, i32 4, i32 31), !dbg !240 + %2494 = bitcast float %2481 to i32, !dbg !240 + %2495 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2494, i32 16, i32 31), !dbg !240 + %2496 = bitcast i32 %2495 to float, !dbg !240 + %2497 = fadd float %2481, %2496, !dbg !239 + %2498 = bitcast float %2497 to i32, !dbg !240 + %2499 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2498, i32 8, i32 31), !dbg !240 + %2500 = bitcast i32 %2499 to float, !dbg !240 + %2501 = fadd float %2497, %2500, !dbg !239 + %2502 = bitcast float %2501 to i32, !dbg !240 + %2503 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2502, i32 4, i32 31), !dbg !240 + %2504 = bitcast float %2482 to i32, !dbg !240 + %2505 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2504, i32 16, i32 31), !dbg !240 + %2506 = bitcast i32 %2505 to float, !dbg !240 + %2507 = fadd float %2482, %2506, !dbg !239 + %2508 = bitcast float %2507 to i32, !dbg !240 + %2509 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2508, i32 8, i32 31), !dbg !240 + %2510 = bitcast i32 %2509 to float, !dbg !240 + %2511 = fadd float %2507, %2510, !dbg !239 + %2512 = bitcast float %2511 to i32, !dbg !240 + %2513 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2512, i32 4, i32 31), !dbg !240 + %2514 = bitcast float %2483 to i32, !dbg !240 + %2515 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2514, i32 16, i32 31), !dbg !240 + %2516 = bitcast i32 %2515 to float, !dbg !240 + %2517 = fadd float %2483, %2516, !dbg !239 + %2518 = bitcast float %2517 to i32, !dbg !240 + %2519 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2518, i32 8, i32 31), !dbg !240 + %2520 = bitcast i32 %2519 to float, !dbg !240 + %2521 = fadd float %2517, %2520, !dbg !239 + %2522 = bitcast float %2521 to i32, !dbg !240 + %2523 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2522, i32 4, i32 31), !dbg !240 + %2524 = insertelement <4 x i32> poison, i32 %2493, i64 0, !dbg !240 + %2525 = insertelement <4 x i32> %2524, i32 %2503, i64 1, !dbg !240 + %2526 = insertelement <4 x i32> %2525, i32 %2513, i64 2, !dbg !240 + %2527 = insertelement <4 x i32> %2526, i32 %2523, i64 3, !dbg !240 + %2528 = bitcast <4 x i32> %2527 to <4 x float>, !dbg !240 + %2529 = insertelement <4 x float> poison, float %2491, i64 0, !dbg !239 + %2530 = insertelement <4 x float> %2529, float %2501, i64 1, !dbg !239 + %2531 = insertelement <4 x float> %2530, float %2511, i64 2, !dbg !239 + %2532 = insertelement <4 x float> %2531, float %2521, i64 3, !dbg !239 + %2533 = fadd <4 x float> %2532, %2528, !dbg !239 + %2534 = fadd <4 x float> %2449, %2533, !dbg !242 + %2535 = extractelement <4 x float> %2534, i64 0, !dbg !243 + %2536 = select i1 %2383, float %2535, float %.pn117534, !dbg !243 + %2537 = extractelement <4 x float> %2534, i64 1, !dbg !243 + %2538 = select i1 %2383, float %2537, float %.pn115535, !dbg !243 + %2539 = extractelement <4 x float> %2534, i64 2, !dbg !243 + %2540 = select i1 %2382, float %2539, float %.pn113536, !dbg !243 + %2541 = extractelement <4 x float> %2534, i64 3, !dbg !243 + %2542 = select i1 %2382, float %2541, float %.pn111537, !dbg !243 + %2543 = insertelement <4 x i1> poison, i1 %2382, i64 0, !dbg !243 + %2544 = insertelement <4 x i1> %2543, i1 %2383, i64 1, !dbg !243 + %2545 = shufflevector <4 x i1> %2544, <4 x i1> poison, <4 x i32> , !dbg !243 + %2546 = select <4 x i1> %2545, <4 x float> %2534, <4 x float> %2381, !dbg !243 + %2547 = add nuw nsw i64 %2380, 1, !dbg !229 + %exitcond601.not = icmp eq i64 %2547, %2373, !dbg !229 + br i1 %exitcond601.not, label %._crit_edge542.loopexit, label %2379, !dbg !229 + +._crit_edge542.loopexit: ; preds = %2379 + %2548 = fadd float %2536, 0.000000e+00, !dbg !244 + %2549 = fadd float %2538, 0.000000e+00, !dbg !244 + %2550 = fadd float %2540, 0.000000e+00, !dbg !244 + %2551 = fadd float %2542, 0.000000e+00, !dbg !244 + br label %._crit_edge542, !dbg !245 + +._crit_edge542: ; preds = %._crit_edge523.._crit_edge542_crit_edge, %._crit_edge542.loopexit + %.pre-phi609 = phi i32 [ %.pre608, %._crit_edge523.._crit_edge542_crit_edge ], [ %2369, %._crit_edge542.loopexit ], !dbg !164 + %.pre-phi605 = phi i32 [ %.pre604, %._crit_edge523.._crit_edge542_crit_edge ], [ %2367, %._crit_edge542.loopexit ], !dbg !164 + %.pre-phi603 = phi i32 [ %.pre602, %._crit_edge523.._crit_edge542_crit_edge ], [ %2366, %._crit_edge542.loopexit ], !dbg !164 + %.pn117.lcssa = phi float [ %1868, %._crit_edge523.._crit_edge542_crit_edge ], [ %2548, %._crit_edge542.loopexit ] + %.pn115.lcssa = phi float [ %1869, %._crit_edge523.._crit_edge542_crit_edge ], [ %2549, %._crit_edge542.loopexit ] + %.pn113.lcssa = phi float [ %1870, %._crit_edge523.._crit_edge542_crit_edge ], [ %2550, %._crit_edge542.loopexit ] + %.pn111.lcssa = phi float [ %1871, %._crit_edge523.._crit_edge542_crit_edge ], [ %2551, %._crit_edge542.loopexit ] + %2552 = phi <4 x float> [ %1872, %._crit_edge523.._crit_edge542_crit_edge ], [ %2546, %._crit_edge542.loopexit ] + %2553 = uitofp i1 %1140 to float, !dbg !245 + %2554 = uitofp i1 %1141 to float, !dbg !245 + %2555 = uitofp i1 %1142 to float, !dbg !245 + %2556 = uitofp i1 %1143 to float, !dbg !245 + %2557 = uitofp i1 %1144 to float, !dbg !245 + %2558 = uitofp i1 %1145 to float, !dbg !245 + %2559 = uitofp i1 %1146 to float, !dbg !245 + %2560 = extractelement <4 x float> %1979, i64 0, !dbg !245 + %2561 = fadd float %2560, %2556, !dbg !245 + %2562 = extractelement <4 x float> %1979, i64 1, !dbg !245 + %2563 = fadd float %2562, %2557, !dbg !245 + %2564 = extractelement <4 x float> %1979, i64 2, !dbg !245 + %2565 = fadd float %2564, %2558, !dbg !245 + %2566 = extractelement <4 x float> %1979, i64 3, !dbg !245 + %2567 = fadd float %2566, %2559, !dbg !245 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !245 + %2568 = and i32 %242, 320, !dbg !245 + %2569 = shl nuw nsw i32 %28, 3, !dbg !245 + %2570 = and i32 %23, 8, !dbg !245 + %2571 = icmp eq i32 %2570, 0, !dbg !245 + %2572 = select i1 %2571, i32 0, i32 160, !dbg !245 + %2573 = xor i32 %2572, %29, !dbg !245 + %2574 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %2568, !dbg !245 + %2575 = getelementptr inbounds nuw i8, ptr addrspace(3) %2574, i32 %2569, !dbg !245 + %2576 = getelementptr inbounds nuw i8, ptr addrspace(3) %2575, i32 %2573, !dbg !245 + %2577 = bitcast float %2561 to i32, !dbg !245 + %2578 = bitcast float %2563 to i32, !dbg !245 + tail call void asm sideeffect "st.shared.v2.b32 [ $0 + 0 ], { $1, $2 };", "r,r,r"(ptr addrspace(3) %2576, i32 %2577, i32 %2578) #2, !dbg !245 + %2579 = getelementptr inbounds nuw i8, ptr addrspace(3) %2576, i32 512, !dbg !245 + %2580 = bitcast float %2565 to i32, !dbg !245 + %2581 = bitcast float %2567 to i32, !dbg !245 + tail call void asm sideeffect "st.shared.v2.b32 [ $0 + 0 ], { $1, $2 };", "r,r,r"(ptr addrspace(3) nonnull %2579, i32 %2580, i32 %2581) #2, !dbg !245 + %2582 = extractelement <4 x float> %2170, i64 0, !dbg !210 + %2583 = fadd float %2582, %2553, !dbg !210 + %2584 = extractelement <4 x float> %2170, i64 1, !dbg !210 + %2585 = fadd float %2584, %2554, !dbg !210 + %2586 = extractelement <4 x float> %2170, i64 2, !dbg !210 + %2587 = fadd float %2586, %2553, !dbg !210 + %2588 = extractelement <4 x float> %2170, i64 3, !dbg !210 + %2589 = fadd float %2588, %2555, !dbg !210 + %2590 = extractelement <4 x float> %2361, i64 0, !dbg !227 + %2591 = fadd float %2590, %2553, !dbg !227 + %2592 = extractelement <4 x float> %2361, i64 1, !dbg !227 + %2593 = fadd float %2592, %2554, !dbg !227 + %2594 = extractelement <4 x float> %2361, i64 2, !dbg !227 + %2595 = fadd float %2594, %2553, !dbg !227 + %2596 = extractelement <4 x float> %2361, i64 3, !dbg !227 + %2597 = fadd float %2596, %2555, !dbg !227 + %2598 = extractelement <4 x float> %2552, i64 0, !dbg !244 + %2599 = fadd float %2598, %2553, !dbg !244 + %2600 = extractelement <4 x float> %2552, i64 1, !dbg !244 + %2601 = fadd float %2600, %2554, !dbg !244 + %2602 = extractelement <4 x float> %2552, i64 2, !dbg !244 + %2603 = fadd float %2602, %2553, !dbg !244 + %2604 = extractelement <4 x float> %2552, i64 3, !dbg !244 + %2605 = fadd float %2604, %2555, !dbg !244 + %2606 = icmp eq i32 %2364, 0, !dbg !164 + %2607 = select i1 %2606, float %1809, float %1810, !dbg !164 + %2608 = select i1 %2606, float %1810, float %1809, !dbg !164 + %2609 = select i1 %2606, float %1811, float %1812, !dbg !164 + %2610 = select i1 %2606, float %1812, float %1811, !dbg !164 + %2611 = select i1 %2606, float %1813, float %1814, !dbg !164 + %2612 = select i1 %2606, float %1814, float %1813, !dbg !164 + %2613 = select i1 %2606, float %1815, float %1816, !dbg !164 + %2614 = select i1 %2606, float %1816, float %1815, !dbg !164 + %2615 = or disjoint i32 %2364, %.pre-phi605, !dbg !164 + %2616 = or disjoint i32 %2615, %.pre-phi609, !dbg !164 + %2617 = bitcast float %2607 to i32, !dbg !164 + %2618 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2617, i32 %2616, i32 31), !dbg !164 + %2619 = xor i32 %2616, 4, !dbg !164 + %2620 = bitcast float %2608 to i32, !dbg !164 + %2621 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2620, i32 %2619, i32 31), !dbg !164 + %2622 = bitcast float %2609 to i32, !dbg !164 + %2623 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2622, i32 %2616, i32 31), !dbg !164 + %2624 = bitcast float %2610 to i32, !dbg !164 + %2625 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2624, i32 %2619, i32 31), !dbg !164 + %2626 = bitcast float %2611 to i32, !dbg !164 + %2627 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2626, i32 %2616, i32 31), !dbg !164 + %2628 = bitcast float %2612 to i32, !dbg !164 + %2629 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2628, i32 %2619, i32 31), !dbg !164 + %2630 = bitcast float %2613 to i32, !dbg !164 + %2631 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2630, i32 %2616, i32 31), !dbg !164 + %2632 = bitcast float %2614 to i32, !dbg !164 + %2633 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2632, i32 %2619, i32 31), !dbg !164 + %.v = select i1 %2606, i32 %2621, i32 %2618, !dbg !164 + %.v122 = select i1 %2606, i32 %2625, i32 %2623, !dbg !164 + %.v123 = select i1 %2606, i32 %2629, i32 %2627, !dbg !164 + %.v124 = select i1 %2606, i32 %2633, i32 %2631, !dbg !164 + %2634 = bitcast float %2583 to i32, !dbg !246 + %2635 = bitcast float %.pn85.lcssa to i32, !dbg !246 + %2636 = bitcast float %.pn81.lcssa to i32, !dbg !246 + %2637 = bitcast float %2587 to i32, !dbg !246 + %2638 = bitcast float %2585 to i32, !dbg !246 + %2639 = bitcast float %.pn83.lcssa to i32, !dbg !246 + %2640 = bitcast float %.pn79.lcssa to i32, !dbg !246 + %2641 = bitcast float %2589 to i32, !dbg !246 + %2642 = select i1 %2606, i32 %2618, i32 %2621, !dbg !164 + %2643 = select i1 %2606, i32 %2623, i32 %2625, !dbg !164 + %2644 = select i1 %2606, i32 %2627, i32 %2629, !dbg !164 + %2645 = select i1 %2606, i32 %2631, i32 %2633, !dbg !164 + %2646 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2634, i32 %2635, i32 %2636, i32 %2637, i32 %2642, i32 %2643) #2, !dbg !246 + %2647 = extractvalue { float, float, float, float } %2646, 0, !dbg !246 + %2648 = extractvalue { float, float, float, float } %2646, 1, !dbg !246 + %2649 = extractvalue { float, float, float, float } %2646, 2, !dbg !246 + %2650 = extractvalue { float, float, float, float } %2646, 3, !dbg !246 + %2651 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2634, i32 %2635, i32 %2636, i32 %2637, i32 %2644, i32 %2645) #2, !dbg !246 + %2652 = extractvalue { float, float, float, float } %2651, 0, !dbg !246 + %2653 = extractvalue { float, float, float, float } %2651, 1, !dbg !246 + %2654 = extractvalue { float, float, float, float } %2651, 2, !dbg !246 + %2655 = extractvalue { float, float, float, float } %2651, 3, !dbg !246 + %2656 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2647, float %2648, float %2649, float %2650, i32 %2638, i32 %2639, i32 %2640, i32 %2641, i32 %.v, i32 %.v122) #2, !dbg !246 + %2657 = extractvalue { float, float, float, float } %2656, 0, !dbg !246 + %2658 = extractvalue { float, float, float, float } %2656, 1, !dbg !246 + %2659 = extractvalue { float, float, float, float } %2656, 2, !dbg !246 + %2660 = extractvalue { float, float, float, float } %2656, 3, !dbg !246 + %2661 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2652, float %2653, float %2654, float %2655, i32 %2638, i32 %2639, i32 %2640, i32 %2641, i32 %.v123, i32 %.v124) #2, !dbg !246 + %2662 = extractvalue { float, float, float, float } %2661, 0, !dbg !246 + %2663 = extractvalue { float, float, float, float } %2661, 1, !dbg !246 + %2664 = extractvalue { float, float, float, float } %2661, 2, !dbg !246 + %2665 = extractvalue { float, float, float, float } %2661, 3, !dbg !246 + %2666 = and i32 %23, 2, !dbg !247 + %.not125 = icmp eq i32 %2666, 0, !dbg !247 + %2667 = select i1 %.not125, float %2657, float %2658, !dbg !247 + %2668 = select i1 %.not125, float %2658, float %2657, !dbg !247 + %2669 = select i1 %.not125, float %2659, float %2660, !dbg !247 + %2670 = select i1 %.not125, float %2660, float %2659, !dbg !247 + %2671 = select i1 %.not125, float %2662, float %2663, !dbg !247 + %2672 = select i1 %.not125, float %2663, float %2662, !dbg !247 + %2673 = select i1 %.not125, float %2664, float %2665, !dbg !247 + %2674 = select i1 %.not125, float %2665, float %2664, !dbg !247 + %2675 = and i32 %23, 1, !dbg !247 + %2676 = icmp eq i32 %2675, 0, !dbg !247 + %trunc = trunc i32 %23 to i2, !dbg !247 + %rev = tail call i2 @llvm.bitreverse.i2(i2 %trunc), !dbg !247 + %2677 = zext i2 %rev to i32, !dbg !247 + %2678 = or disjoint i32 %27, %2677, !dbg !247 + %2679 = bitcast float %2667 to i32, !dbg !247 + %2680 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2679, i32 %2678, i32 31), !dbg !247 + %2681 = bitcast float %2669 to i32, !dbg !247 + %2682 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2681, i32 %2678, i32 31), !dbg !247 + %2683 = xor i32 %2678, 2, !dbg !247 + %2684 = bitcast float %2668 to i32, !dbg !247 + %2685 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2684, i32 %2683, i32 31), !dbg !247 + %2686 = bitcast float %2670 to i32, !dbg !247 + %2687 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2686, i32 %2683, i32 31), !dbg !247 + %2688 = bitcast float %2671 to i32, !dbg !247 + %2689 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2688, i32 %2678, i32 31), !dbg !247 + %2690 = bitcast float %2673 to i32, !dbg !247 + %2691 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2690, i32 %2678, i32 31), !dbg !247 + %2692 = bitcast float %2672 to i32, !dbg !247 + %2693 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2692, i32 %2683, i32 31), !dbg !247 + %2694 = bitcast float %2674 to i32, !dbg !247 + %2695 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2694, i32 %2683, i32 31), !dbg !247 + %.v126 = select i1 %2676, i32 %2685, i32 %2680, !dbg !247 + %.v127 = select i1 %2676, i32 %2687, i32 %2682, !dbg !247 + %.v128 = select i1 %2676, i32 %2693, i32 %2689, !dbg !247 + %.v129 = select i1 %2676, i32 %2695, i32 %2691, !dbg !247 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !245 + %2696 = shl nuw nsw i32 %2675, 6, !dbg !245 + %2697 = select i1 %.not125, i32 0, i32 160, !dbg !245 + %2698 = or disjoint i32 %2697, %2696, !dbg !245 + %2699 = or disjoint i32 %2698, %27, !dbg !245 + %2700 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %2699, !dbg !245 + %2701 = load i32, ptr addrspace(3) %2700, align 4, !dbg !245 + %2702 = getelementptr inbounds nuw i8, ptr addrspace(3) %2700, i32 256, !dbg !245 + %2703 = load i32, ptr addrspace(3) %2702, align 4, !dbg !245 + %2704 = getelementptr inbounds nuw i8, ptr addrspace(3) %2700, i32 512, !dbg !245 + %2705 = load i32, ptr addrspace(3) %2704, align 4, !dbg !245 + %2706 = getelementptr inbounds nuw i8, ptr addrspace(3) %2700, i32 768, !dbg !245 + %2707 = load i32, ptr addrspace(3) %2706, align 4, !dbg !245 + %2708 = xor i32 %2699, 32, !dbg !245 + %2709 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %2708, !dbg !245 + %2710 = load i32, ptr addrspace(3) %2709, align 4, !dbg !245 + %2711 = getelementptr inbounds nuw i8, ptr addrspace(3) %2709, i32 256, !dbg !245 + %2712 = load i32, ptr addrspace(3) %2711, align 4, !dbg !245 + %2713 = getelementptr inbounds nuw i8, ptr addrspace(3) %2709, i32 512, !dbg !245 + %2714 = load i32, ptr addrspace(3) %2713, align 4, !dbg !245 + %2715 = getelementptr inbounds nuw i8, ptr addrspace(3) %2709, i32 768, !dbg !245 + %2716 = load i32, ptr addrspace(3) %2715, align 4, !dbg !245 + %2717 = select i1 %2676, i32 %2680, i32 %2685, !dbg !247 + %2718 = select i1 %2676, i32 %2682, i32 %2687, !dbg !247 + %2719 = select i1 %2676, i32 %2689, i32 %2693, !dbg !247 + %2720 = select i1 %2676, i32 %2691, i32 %2695, !dbg !247 + %2721 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2717, i32 %2718, i32 %.v126, i32 %.v127, i32 %2701, i32 %2703) #2, !dbg !247 + %2722 = extractvalue { float, float, float, float } %2721, 0, !dbg !247 + %2723 = extractvalue { float, float, float, float } %2721, 1, !dbg !247 + %2724 = extractvalue { float, float, float, float } %2721, 2, !dbg !247 + %2725 = extractvalue { float, float, float, float } %2721, 3, !dbg !247 + %2726 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2717, i32 %2718, i32 %.v126, i32 %.v127, i32 %2710, i32 %2712) #2, !dbg !247 + %2727 = extractvalue { float, float, float, float } %2726, 0, !dbg !247 + %2728 = extractvalue { float, float, float, float } %2726, 1, !dbg !247 + %2729 = extractvalue { float, float, float, float } %2726, 2, !dbg !247 + %2730 = extractvalue { float, float, float, float } %2726, 3, !dbg !247 + %2731 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2722, float %2723, float %2724, float %2725, i32 %2719, i32 %2720, i32 %.v128, i32 %.v129, i32 %2705, i32 %2707) #2, !dbg !247 + %2732 = extractvalue { float, float, float, float } %2731, 0, !dbg !247 + %2733 = extractvalue { float, float, float, float } %2731, 1, !dbg !247 + %2734 = extractvalue { float, float, float, float } %2731, 2, !dbg !247 + %2735 = extractvalue { float, float, float, float } %2731, 3, !dbg !247 + %2736 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2727, float %2728, float %2729, float %2730, i32 %2719, i32 %2720, i32 %.v128, i32 %.v129, i32 %2714, i32 %2716) #2, !dbg !247 + %2737 = extractvalue { float, float, float, float } %2736, 0, !dbg !247 + %2738 = extractvalue { float, float, float, float } %2736, 1, !dbg !247 + %2739 = extractvalue { float, float, float, float } %2736, 2, !dbg !247 + %2740 = extractvalue { float, float, float, float } %2736, 3, !dbg !247 + %2741 = fsub float 0.000000e+00, %2732, !dbg !248 + %2742 = fsub float 0.000000e+00, %2733, !dbg !248 + %2743 = fsub float 0.000000e+00, %2734, !dbg !248 + %2744 = fsub float 0.000000e+00, %2735, !dbg !248 + %2745 = fsub float 0.000000e+00, %2737, !dbg !248 + %2746 = fsub float 0.000000e+00, %2738, !dbg !248 + %2747 = fsub float 0.000000e+00, %2739, !dbg !248 + %2748 = fsub float 0.000000e+00, %2740, !dbg !248 + %2749 = select i1 %2606, float %1825, float %1826, !dbg !166 + %2750 = select i1 %2606, float %1826, float %1825, !dbg !166 + %2751 = select i1 %2606, float %1827, float %1828, !dbg !166 + %2752 = select i1 %2606, float %1828, float %1827, !dbg !166 + %2753 = select i1 %2606, float %1829, float %1830, !dbg !166 + %2754 = select i1 %2606, float %1830, float %1829, !dbg !166 + %2755 = select i1 %2606, float %1831, float %1832, !dbg !166 + %2756 = select i1 %2606, float %1832, float %1831, !dbg !166 + %2757 = bitcast float %2749 to i32, !dbg !166 + %2758 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2757, i32 %2616, i32 31), !dbg !166 + %2759 = bitcast float %2750 to i32, !dbg !166 + %2760 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2759, i32 %2619, i32 31), !dbg !166 + %2761 = bitcast float %2751 to i32, !dbg !166 + %2762 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2761, i32 %2616, i32 31), !dbg !166 + %2763 = bitcast float %2752 to i32, !dbg !166 + %2764 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2763, i32 %2619, i32 31), !dbg !166 + %2765 = bitcast float %2753 to i32, !dbg !166 + %2766 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2765, i32 %2616, i32 31), !dbg !166 + %2767 = bitcast float %2754 to i32, !dbg !166 + %2768 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2767, i32 %2619, i32 31), !dbg !166 + %2769 = bitcast float %2755 to i32, !dbg !166 + %2770 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2769, i32 %2616, i32 31), !dbg !166 + %2771 = bitcast float %2756 to i32, !dbg !166 + %2772 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2771, i32 %2619, i32 31), !dbg !166 + %.v130 = select i1 %2606, i32 %2760, i32 %2758, !dbg !166 + %.v131 = select i1 %2606, i32 %2764, i32 %2762, !dbg !166 + %.v132 = select i1 %2606, i32 %2768, i32 %2766, !dbg !166 + %.v133 = select i1 %2606, i32 %2772, i32 %2770, !dbg !166 + %2773 = bitcast float %2591 to i32, !dbg !249 + %2774 = bitcast float %.pn101.lcssa to i32, !dbg !249 + %2775 = bitcast float %.pn97.lcssa to i32, !dbg !249 + %2776 = bitcast float %2595 to i32, !dbg !249 + %2777 = bitcast float %2593 to i32, !dbg !249 + %2778 = bitcast float %.pn99.lcssa to i32, !dbg !249 + %2779 = bitcast float %.pn95.lcssa to i32, !dbg !249 + %2780 = bitcast float %2597 to i32, !dbg !249 + %2781 = select i1 %2606, i32 %2758, i32 %2760, !dbg !166 + %2782 = select i1 %2606, i32 %2762, i32 %2764, !dbg !166 + %2783 = select i1 %2606, i32 %2766, i32 %2768, !dbg !166 + %2784 = select i1 %2606, i32 %2770, i32 %2772, !dbg !166 + %2785 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2773, i32 %2774, i32 %2775, i32 %2776, i32 %2781, i32 %2782) #2, !dbg !249 + %2786 = extractvalue { float, float, float, float } %2785, 0, !dbg !249 + %2787 = extractvalue { float, float, float, float } %2785, 1, !dbg !249 + %2788 = extractvalue { float, float, float, float } %2785, 2, !dbg !249 + %2789 = extractvalue { float, float, float, float } %2785, 3, !dbg !249 + %2790 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2773, i32 %2774, i32 %2775, i32 %2776, i32 %2783, i32 %2784) #2, !dbg !249 + %2791 = extractvalue { float, float, float, float } %2790, 0, !dbg !249 + %2792 = extractvalue { float, float, float, float } %2790, 1, !dbg !249 + %2793 = extractvalue { float, float, float, float } %2790, 2, !dbg !249 + %2794 = extractvalue { float, float, float, float } %2790, 3, !dbg !249 + %2795 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2786, float %2787, float %2788, float %2789, i32 %2777, i32 %2778, i32 %2779, i32 %2780, i32 %.v130, i32 %.v131) #2, !dbg !249 + %2796 = extractvalue { float, float, float, float } %2795, 0, !dbg !249 + %2797 = extractvalue { float, float, float, float } %2795, 1, !dbg !249 + %2798 = extractvalue { float, float, float, float } %2795, 2, !dbg !249 + %2799 = extractvalue { float, float, float, float } %2795, 3, !dbg !249 + %2800 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2791, float %2792, float %2793, float %2794, i32 %2777, i32 %2778, i32 %2779, i32 %2780, i32 %.v132, i32 %.v133) #2, !dbg !249 + %2801 = extractvalue { float, float, float, float } %2800, 0, !dbg !249 + %2802 = extractvalue { float, float, float, float } %2800, 1, !dbg !249 + %2803 = extractvalue { float, float, float, float } %2800, 2, !dbg !249 + %2804 = extractvalue { float, float, float, float } %2800, 3, !dbg !249 + %2805 = select i1 %.not125, float %2796, float %2797, !dbg !250 + %2806 = select i1 %.not125, float %2797, float %2796, !dbg !250 + %2807 = select i1 %.not125, float %2798, float %2799, !dbg !250 + %2808 = select i1 %.not125, float %2799, float %2798, !dbg !250 + %2809 = select i1 %.not125, float %2801, float %2802, !dbg !250 + %2810 = select i1 %.not125, float %2802, float %2801, !dbg !250 + %2811 = select i1 %.not125, float %2803, float %2804, !dbg !250 + %2812 = select i1 %.not125, float %2804, float %2803, !dbg !250 + %2813 = bitcast float %2805 to i32, !dbg !250 + %2814 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2813, i32 %2678, i32 31), !dbg !250 + %2815 = bitcast float %2807 to i32, !dbg !250 + %2816 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2815, i32 %2678, i32 31), !dbg !250 + %2817 = bitcast float %2806 to i32, !dbg !250 + %2818 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2817, i32 %2683, i32 31), !dbg !250 + %2819 = bitcast float %2808 to i32, !dbg !250 + %2820 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2819, i32 %2683, i32 31), !dbg !250 + %2821 = bitcast float %2809 to i32, !dbg !250 + %2822 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2821, i32 %2678, i32 31), !dbg !250 + %2823 = bitcast float %2811 to i32, !dbg !250 + %2824 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2823, i32 %2678, i32 31), !dbg !250 + %2825 = bitcast float %2810 to i32, !dbg !250 + %2826 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2825, i32 %2683, i32 31), !dbg !250 + %2827 = bitcast float %2812 to i32, !dbg !250 + %2828 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2827, i32 %2683, i32 31), !dbg !250 + %.v134 = select i1 %2676, i32 %2818, i32 %2814, !dbg !250 + %.v135 = select i1 %2676, i32 %2820, i32 %2816, !dbg !250 + %.v136 = select i1 %2676, i32 %2826, i32 %2822, !dbg !250 + %.v137 = select i1 %2676, i32 %2828, i32 %2824, !dbg !250 + %2829 = icmp eq i32 %271, 0, !dbg !210 + %2830 = select i1 %2829, float %2583, float %2585, !dbg !210 + %2831 = select i1 %2829, float %2585, float %2583, !dbg !210 + %2832 = select i1 %2829, float %.pn85.lcssa, float %.pn83.lcssa, !dbg !210 + %2833 = select i1 %2829, float %.pn83.lcssa, float %.pn85.lcssa, !dbg !210 + %2834 = select i1 %2829, float %.pn81.lcssa, float %.pn79.lcssa, !dbg !210 + %2835 = select i1 %2829, float %.pn79.lcssa, float %.pn81.lcssa, !dbg !210 + %2836 = select i1 %2829, float %2587, float %2589, !dbg !210 + %2837 = select i1 %2829, float %2589, float %2587, !dbg !210 + %2838 = shl nuw nsw i32 %23, 2, !dbg !210 + %2839 = and i32 %2838, 28, !dbg !210 + %2840 = or disjoint i32 %2839, %.pre-phi609, !dbg !210 + %2841 = bitcast float %2830 to i32, !dbg !210 + %2842 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2841, i32 %2840, i32 31), !dbg !210 + %2843 = xor i32 %2840, 16, !dbg !210 + %2844 = bitcast float %2831 to i32, !dbg !210 + %2845 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2844, i32 %2843, i32 31), !dbg !210 + %2846 = bitcast float %2832 to i32, !dbg !210 + %2847 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2846, i32 %2840, i32 31), !dbg !210 + %2848 = bitcast float %2833 to i32, !dbg !210 + %2849 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2848, i32 %2843, i32 31), !dbg !210 + %2850 = bitcast float %2834 to i32, !dbg !210 + %2851 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2850, i32 %2840, i32 31), !dbg !210 + %2852 = bitcast float %2835 to i32, !dbg !210 + %2853 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2852, i32 %2843, i32 31), !dbg !210 + %2854 = bitcast float %2836 to i32, !dbg !210 + %2855 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2854, i32 %2840, i32 31), !dbg !210 + %2856 = bitcast float %2837 to i32, !dbg !210 + %2857 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2856, i32 %2843, i32 31), !dbg !210 + %.v138 = select i1 %2606, i32 %2845, i32 %2842, !dbg !210 + %.v139 = select i1 %2606, i32 %2849, i32 %2847, !dbg !210 + %.v140 = select i1 %2606, i32 %2853, i32 %2851, !dbg !210 + %.v141 = select i1 %2606, i32 %2857, i32 %2855, !dbg !210 + %2858 = select i1 %2676, i32 %2814, i32 %2818, !dbg !250 + %2859 = select i1 %2676, i32 %2816, i32 %2820, !dbg !250 + %2860 = select i1 %2676, i32 %2822, i32 %2826, !dbg !250 + %2861 = select i1 %2676, i32 %2824, i32 %2828, !dbg !250 + %2862 = select i1 %2606, i32 %2842, i32 %2845, !dbg !210 + %2863 = select i1 %2606, i32 %2847, i32 %2849, !dbg !210 + %2864 = select i1 %2606, i32 %2851, i32 %2853, !dbg !210 + %2865 = select i1 %2606, i32 %2855, i32 %2857, !dbg !210 + %2866 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2858, i32 %2859, i32 %.v134, i32 %.v135, i32 %2862, i32 %.v138) #2, !dbg !250 + %2867 = extractvalue { float, float, float, float } %2866, 0, !dbg !250 + %2868 = extractvalue { float, float, float, float } %2866, 1, !dbg !250 + %2869 = extractvalue { float, float, float, float } %2866, 2, !dbg !250 + %2870 = extractvalue { float, float, float, float } %2866, 3, !dbg !250 + %2871 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2858, i32 %2859, i32 %.v134, i32 %.v135, i32 %2864, i32 %.v140) #2, !dbg !250 + %2872 = extractvalue { float, float, float, float } %2871, 0, !dbg !250 + %2873 = extractvalue { float, float, float, float } %2871, 1, !dbg !250 + %2874 = extractvalue { float, float, float, float } %2871, 2, !dbg !250 + %2875 = extractvalue { float, float, float, float } %2871, 3, !dbg !250 + %2876 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2867, float %2868, float %2869, float %2870, i32 %2860, i32 %2861, i32 %.v136, i32 %.v137, i32 %2863, i32 %.v139) #2, !dbg !250 + %2877 = extractvalue { float, float, float, float } %2876, 0, !dbg !250 + %2878 = extractvalue { float, float, float, float } %2876, 1, !dbg !250 + %2879 = extractvalue { float, float, float, float } %2876, 2, !dbg !250 + %2880 = extractvalue { float, float, float, float } %2876, 3, !dbg !250 + %2881 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2872, float %2873, float %2874, float %2875, i32 %2860, i32 %2861, i32 %.v136, i32 %.v137, i32 %2865, i32 %.v141) #2, !dbg !250 + %2882 = extractvalue { float, float, float, float } %2881, 0, !dbg !250 + %2883 = extractvalue { float, float, float, float } %2881, 1, !dbg !250 + %2884 = extractvalue { float, float, float, float } %2881, 2, !dbg !250 + %2885 = extractvalue { float, float, float, float } %2881, 3, !dbg !250 + %2886 = fsub float 0.000000e+00, %2877, !dbg !251 + %2887 = fsub float 0.000000e+00, %2878, !dbg !251 + %2888 = fsub float 0.000000e+00, %2879, !dbg !251 + %2889 = fsub float 0.000000e+00, %2880, !dbg !251 + %2890 = fsub float 0.000000e+00, %2882, !dbg !251 + %2891 = fsub float 0.000000e+00, %2883, !dbg !251 + %2892 = fsub float 0.000000e+00, %2884, !dbg !251 + %2893 = fsub float 0.000000e+00, %2885, !dbg !251 + %2894 = select i1 %2606, float %1849, float %1850, !dbg !168 + %2895 = select i1 %2606, float %1850, float %1849, !dbg !168 + %2896 = select i1 %2606, float %1851, float %1852, !dbg !168 + %2897 = select i1 %2606, float %1852, float %1851, !dbg !168 + %2898 = select i1 %2606, float %1853, float %1854, !dbg !168 + %2899 = select i1 %2606, float %1854, float %1853, !dbg !168 + %2900 = select i1 %2606, float %1855, float %1856, !dbg !168 + %2901 = select i1 %2606, float %1856, float %1855, !dbg !168 + %2902 = bitcast float %2894 to i32, !dbg !168 + %2903 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2902, i32 %2616, i32 31), !dbg !168 + %2904 = bitcast float %2895 to i32, !dbg !168 + %2905 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2904, i32 %2619, i32 31), !dbg !168 + %2906 = bitcast float %2896 to i32, !dbg !168 + %2907 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2906, i32 %2616, i32 31), !dbg !168 + %2908 = bitcast float %2897 to i32, !dbg !168 + %2909 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2908, i32 %2619, i32 31), !dbg !168 + %2910 = bitcast float %2898 to i32, !dbg !168 + %2911 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2910, i32 %2616, i32 31), !dbg !168 + %2912 = bitcast float %2899 to i32, !dbg !168 + %2913 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2912, i32 %2619, i32 31), !dbg !168 + %2914 = bitcast float %2900 to i32, !dbg !168 + %2915 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2914, i32 %2616, i32 31), !dbg !168 + %2916 = bitcast float %2901 to i32, !dbg !168 + %2917 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2916, i32 %2619, i32 31), !dbg !168 + %.v142 = select i1 %2606, i32 %2905, i32 %2903, !dbg !168 + %.v143 = select i1 %2606, i32 %2909, i32 %2907, !dbg !168 + %.v144 = select i1 %2606, i32 %2913, i32 %2911, !dbg !168 + %.v145 = select i1 %2606, i32 %2917, i32 %2915, !dbg !168 + %2918 = bitcast float %2599 to i32, !dbg !252 + %2919 = bitcast float %.pn117.lcssa to i32, !dbg !252 + %2920 = bitcast float %.pn113.lcssa to i32, !dbg !252 + %2921 = bitcast float %2603 to i32, !dbg !252 + %2922 = bitcast float %2601 to i32, !dbg !252 + %2923 = bitcast float %.pn115.lcssa to i32, !dbg !252 + %2924 = bitcast float %.pn111.lcssa to i32, !dbg !252 + %2925 = bitcast float %2605 to i32, !dbg !252 + %2926 = select i1 %2606, i32 %2903, i32 %2905, !dbg !168 + %2927 = select i1 %2606, i32 %2907, i32 %2909, !dbg !168 + %2928 = select i1 %2606, i32 %2911, i32 %2913, !dbg !168 + %2929 = select i1 %2606, i32 %2915, i32 %2917, !dbg !168 + %2930 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2918, i32 %2919, i32 %2920, i32 %2921, i32 %2926, i32 %2927) #2, !dbg !252 + %2931 = extractvalue { float, float, float, float } %2930, 0, !dbg !252 + %2932 = extractvalue { float, float, float, float } %2930, 1, !dbg !252 + %2933 = extractvalue { float, float, float, float } %2930, 2, !dbg !252 + %2934 = extractvalue { float, float, float, float } %2930, 3, !dbg !252 + %2935 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2918, i32 %2919, i32 %2920, i32 %2921, i32 %2928, i32 %2929) #2, !dbg !252 + %2936 = extractvalue { float, float, float, float } %2935, 0, !dbg !252 + %2937 = extractvalue { float, float, float, float } %2935, 1, !dbg !252 + %2938 = extractvalue { float, float, float, float } %2935, 2, !dbg !252 + %2939 = extractvalue { float, float, float, float } %2935, 3, !dbg !252 + %2940 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2931, float %2932, float %2933, float %2934, i32 %2922, i32 %2923, i32 %2924, i32 %2925, i32 %.v142, i32 %.v143) #2, !dbg !252 + %2941 = extractvalue { float, float, float, float } %2940, 0, !dbg !252 + %2942 = extractvalue { float, float, float, float } %2940, 1, !dbg !252 + %2943 = extractvalue { float, float, float, float } %2940, 2, !dbg !252 + %2944 = extractvalue { float, float, float, float } %2940, 3, !dbg !252 + %2945 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2936, float %2937, float %2938, float %2939, i32 %2922, i32 %2923, i32 %2924, i32 %2925, i32 %.v144, i32 %.v145) #2, !dbg !252 + %2946 = extractvalue { float, float, float, float } %2945, 0, !dbg !252 + %2947 = extractvalue { float, float, float, float } %2945, 1, !dbg !252 + %2948 = extractvalue { float, float, float, float } %2945, 2, !dbg !252 + %2949 = extractvalue { float, float, float, float } %2945, 3, !dbg !252 + %2950 = select i1 %.not125, float %2941, float %2942, !dbg !253 + %2951 = select i1 %.not125, float %2942, float %2941, !dbg !253 + %2952 = select i1 %.not125, float %2943, float %2944, !dbg !253 + %2953 = select i1 %.not125, float %2944, float %2943, !dbg !253 + %2954 = select i1 %.not125, float %2946, float %2947, !dbg !253 + %2955 = select i1 %.not125, float %2947, float %2946, !dbg !253 + %2956 = select i1 %.not125, float %2948, float %2949, !dbg !253 + %2957 = select i1 %.not125, float %2949, float %2948, !dbg !253 + %2958 = bitcast float %2950 to i32, !dbg !253 + %2959 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2958, i32 %2678, i32 31), !dbg !253 + %2960 = bitcast float %2952 to i32, !dbg !253 + %2961 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2960, i32 %2678, i32 31), !dbg !253 + %2962 = bitcast float %2951 to i32, !dbg !253 + %2963 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2962, i32 %2683, i32 31), !dbg !253 + %2964 = bitcast float %2953 to i32, !dbg !253 + %2965 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2964, i32 %2683, i32 31), !dbg !253 + %2966 = bitcast float %2954 to i32, !dbg !253 + %2967 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2966, i32 %2678, i32 31), !dbg !253 + %2968 = bitcast float %2956 to i32, !dbg !253 + %2969 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2968, i32 %2678, i32 31), !dbg !253 + %2970 = bitcast float %2955 to i32, !dbg !253 + %2971 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2970, i32 %2683, i32 31), !dbg !253 + %2972 = bitcast float %2957 to i32, !dbg !253 + %2973 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2972, i32 %2683, i32 31), !dbg !253 + %.v146 = select i1 %2676, i32 %2963, i32 %2959, !dbg !253 + %.v147 = select i1 %2676, i32 %2965, i32 %2961, !dbg !253 + %.v148 = select i1 %2676, i32 %2971, i32 %2967, !dbg !253 + %.v149 = select i1 %2676, i32 %2973, i32 %2969, !dbg !253 + %2974 = select i1 %2829, float %2591, float %2593, !dbg !227 + %2975 = select i1 %2829, float %2593, float %2591, !dbg !227 + %2976 = select i1 %2829, float %.pn101.lcssa, float %.pn99.lcssa, !dbg !227 + %2977 = select i1 %2829, float %.pn99.lcssa, float %.pn101.lcssa, !dbg !227 + %2978 = select i1 %2829, float %.pn97.lcssa, float %.pn95.lcssa, !dbg !227 + %2979 = select i1 %2829, float %.pn95.lcssa, float %.pn97.lcssa, !dbg !227 + %2980 = select i1 %2829, float %2595, float %2597, !dbg !227 + %2981 = select i1 %2829, float %2597, float %2595, !dbg !227 + %2982 = bitcast float %2974 to i32, !dbg !227 + %2983 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2982, i32 %2840, i32 31), !dbg !227 + %2984 = bitcast float %2975 to i32, !dbg !227 + %2985 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2984, i32 %2843, i32 31), !dbg !227 + %2986 = bitcast float %2976 to i32, !dbg !227 + %2987 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2986, i32 %2840, i32 31), !dbg !227 + %2988 = bitcast float %2977 to i32, !dbg !227 + %2989 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2988, i32 %2843, i32 31), !dbg !227 + %2990 = bitcast float %2978 to i32, !dbg !227 + %2991 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2990, i32 %2840, i32 31), !dbg !227 + %2992 = bitcast float %2979 to i32, !dbg !227 + %2993 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2992, i32 %2843, i32 31), !dbg !227 + %2994 = bitcast float %2980 to i32, !dbg !227 + %2995 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2994, i32 %2840, i32 31), !dbg !227 + %2996 = bitcast float %2981 to i32, !dbg !227 + %2997 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2996, i32 %2843, i32 31), !dbg !227 + %.v150 = select i1 %2606, i32 %2985, i32 %2983, !dbg !227 + %.v151 = select i1 %2606, i32 %2989, i32 %2987, !dbg !227 + %.v152 = select i1 %2606, i32 %2993, i32 %2991, !dbg !227 + %.v153 = select i1 %2606, i32 %2997, i32 %2995, !dbg !227 + %2998 = select i1 %2676, i32 %2959, i32 %2963, !dbg !253 + %2999 = select i1 %2676, i32 %2961, i32 %2965, !dbg !253 + %3000 = select i1 %2676, i32 %2967, i32 %2971, !dbg !253 + %3001 = select i1 %2676, i32 %2969, i32 %2973, !dbg !253 + %3002 = select i1 %2606, i32 %2983, i32 %2985, !dbg !227 + %3003 = select i1 %2606, i32 %2987, i32 %2989, !dbg !227 + %3004 = select i1 %2606, i32 %2991, i32 %2993, !dbg !227 + %3005 = select i1 %2606, i32 %2995, i32 %2997, !dbg !227 + %3006 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2998, i32 %2999, i32 %.v146, i32 %.v147, i32 %3002, i32 %.v150) #2, !dbg !253 + %3007 = extractvalue { float, float, float, float } %3006, 0, !dbg !253 + %3008 = extractvalue { float, float, float, float } %3006, 1, !dbg !253 + %3009 = extractvalue { float, float, float, float } %3006, 2, !dbg !253 + %3010 = extractvalue { float, float, float, float } %3006, 3, !dbg !253 + %3011 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2998, i32 %2999, i32 %.v146, i32 %.v147, i32 %3004, i32 %.v152) #2, !dbg !253 + %3012 = extractvalue { float, float, float, float } %3011, 0, !dbg !253 + %3013 = extractvalue { float, float, float, float } %3011, 1, !dbg !253 + %3014 = extractvalue { float, float, float, float } %3011, 2, !dbg !253 + %3015 = extractvalue { float, float, float, float } %3011, 3, !dbg !253 + %3016 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3007, float %3008, float %3009, float %3010, i32 %3000, i32 %3001, i32 %.v148, i32 %.v149, i32 %3003, i32 %.v151) #2, !dbg !253 + %3017 = extractvalue { float, float, float, float } %3016, 0, !dbg !253 + %3018 = extractvalue { float, float, float, float } %3016, 1, !dbg !253 + %3019 = extractvalue { float, float, float, float } %3016, 2, !dbg !253 + %3020 = extractvalue { float, float, float, float } %3016, 3, !dbg !253 + %3021 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3012, float %3013, float %3014, float %3015, i32 %3000, i32 %3001, i32 %.v148, i32 %.v149, i32 %3005, i32 %.v153) #2, !dbg !253 + %3022 = extractvalue { float, float, float, float } %3021, 0, !dbg !253 + %3023 = extractvalue { float, float, float, float } %3021, 1, !dbg !253 + %3024 = extractvalue { float, float, float, float } %3021, 2, !dbg !253 + %3025 = extractvalue { float, float, float, float } %3021, 3, !dbg !253 + %3026 = select i1 %2676, i32 0, i32 160, !dbg !245 + %3027 = shl nuw nsw i32 %2666, 7, !dbg !245 + %3028 = or disjoint i32 %3026, %3027, !dbg !245 + %3029 = or disjoint i32 %3028, %27, !dbg !245 + %3030 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %3029, !dbg !245 + %3031 = load i32, ptr addrspace(3) %3030, align 4, !dbg !245 + %3032 = getelementptr inbounds nuw i8, ptr addrspace(3) %3030, i32 64, !dbg !245 + %3033 = load i32, ptr addrspace(3) %3032, align 4, !dbg !245 + %3034 = getelementptr inbounds nuw i8, ptr addrspace(3) %3030, i32 512, !dbg !245 + %3035 = load i32, ptr addrspace(3) %3034, align 4, !dbg !245 + %3036 = getelementptr inbounds nuw i8, ptr addrspace(3) %3030, i32 576, !dbg !245 + %3037 = load i32, ptr addrspace(3) %3036, align 4, !dbg !245 + %3038 = xor i32 %3029, 32, !dbg !245 + %3039 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %3038, !dbg !245 + %3040 = load i32, ptr addrspace(3) %3039, align 4, !dbg !245 + %3041 = getelementptr inbounds nuw i8, ptr addrspace(3) %3039, i32 64, !dbg !245 + %3042 = load i32, ptr addrspace(3) %3041, align 4, !dbg !245 + %3043 = getelementptr inbounds nuw i8, ptr addrspace(3) %3039, i32 512, !dbg !245 + %3044 = load i32, ptr addrspace(3) %3043, align 4, !dbg !245 + %3045 = getelementptr inbounds nuw i8, ptr addrspace(3) %3039, i32 576, !dbg !245 + %3046 = load i32, ptr addrspace(3) %3045, align 4, !dbg !245 + %3047 = bitcast float %1817 to i32, !dbg !254 + %3048 = bitcast float %1819 to i32, !dbg !254 + %3049 = bitcast float %1821 to i32, !dbg !254 + %3050 = bitcast float %1823 to i32, !dbg !254 + %3051 = bitcast float %1818 to i32, !dbg !254 + %3052 = bitcast float %1820 to i32, !dbg !254 + %3053 = bitcast float %1822 to i32, !dbg !254 + %3054 = bitcast float %1824 to i32, !dbg !254 + %3055 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %3047, i32 %3048, i32 %3049, i32 %3050, i32 %3031, i32 %3035) #2, !dbg !254 + %3056 = extractvalue { float, float, float, float } %3055, 0, !dbg !254 + %3057 = extractvalue { float, float, float, float } %3055, 1, !dbg !254 + %3058 = extractvalue { float, float, float, float } %3055, 2, !dbg !254 + %3059 = extractvalue { float, float, float, float } %3055, 3, !dbg !254 + %3060 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %3047, i32 %3048, i32 %3049, i32 %3050, i32 %3040, i32 %3044) #2, !dbg !254 + %3061 = extractvalue { float, float, float, float } %3060, 0, !dbg !254 + %3062 = extractvalue { float, float, float, float } %3060, 1, !dbg !254 + %3063 = extractvalue { float, float, float, float } %3060, 2, !dbg !254 + %3064 = extractvalue { float, float, float, float } %3060, 3, !dbg !254 + %3065 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3056, float %3057, float %3058, float %3059, i32 %3051, i32 %3052, i32 %3053, i32 %3054, i32 %3033, i32 %3037) #2, !dbg !254 + %3066 = extractvalue { float, float, float, float } %3065, 0, !dbg !254 + %3067 = extractvalue { float, float, float, float } %3065, 1, !dbg !254 + %3068 = extractvalue { float, float, float, float } %3065, 2, !dbg !254 + %3069 = extractvalue { float, float, float, float } %3065, 3, !dbg !254 + %3070 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3061, float %3062, float %3063, float %3064, i32 %3051, i32 %3052, i32 %3053, i32 %3054, i32 %3042, i32 %3046) #2, !dbg !254 + %3071 = extractvalue { float, float, float, float } %3070, 0, !dbg !254 + %3072 = extractvalue { float, float, float, float } %3070, 1, !dbg !254 + %3073 = extractvalue { float, float, float, float } %3070, 2, !dbg !254 + %3074 = extractvalue { float, float, float, float } %3070, 3, !dbg !254 + %3075 = select i1 %2606, float %2741, float %2742, !dbg !248 + %3076 = select i1 %2606, float %2742, float %2741, !dbg !248 + %3077 = select i1 %2606, float %2743, float %2744, !dbg !248 + %3078 = select i1 %2606, float %2744, float %2743, !dbg !248 + %3079 = select i1 %2606, float %2745, float %2746, !dbg !248 + %3080 = select i1 %2606, float %2746, float %2745, !dbg !248 + %3081 = select i1 %2606, float %2747, float %2748, !dbg !248 + %3082 = select i1 %2606, float %2748, float %2747, !dbg !248 + %3083 = bitcast float %3075 to i32, !dbg !248 + %3084 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3083, i32 %2616, i32 31), !dbg !248 + %3085 = bitcast float %3076 to i32, !dbg !248 + %3086 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3085, i32 %2619, i32 31), !dbg !248 + %3087 = bitcast float %3077 to i32, !dbg !248 + %3088 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3087, i32 %2616, i32 31), !dbg !248 + %3089 = bitcast float %3078 to i32, !dbg !248 + %3090 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3089, i32 %2619, i32 31), !dbg !248 + %3091 = bitcast float %3079 to i32, !dbg !248 + %3092 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3091, i32 %2616, i32 31), !dbg !248 + %3093 = bitcast float %3080 to i32, !dbg !248 + %3094 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3093, i32 %2619, i32 31), !dbg !248 + %3095 = bitcast float %3081 to i32, !dbg !248 + %3096 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3095, i32 %2616, i32 31), !dbg !248 + %3097 = bitcast float %3082 to i32, !dbg !248 + %3098 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3097, i32 %2619, i32 31), !dbg !248 + %.v154 = select i1 %2606, i32 %3086, i32 %3084, !dbg !248 + %.v155 = select i1 %2606, i32 %3090, i32 %3088, !dbg !248 + %.v156 = select i1 %2606, i32 %3094, i32 %3092, !dbg !248 + %.v157 = select i1 %2606, i32 %3098, i32 %3096, !dbg !248 + %3099 = bitcast float %1825 to i32, !dbg !255 + %3100 = bitcast float %1827 to i32, !dbg !255 + %3101 = bitcast float %1829 to i32, !dbg !255 + %3102 = bitcast float %1831 to i32, !dbg !255 + %3103 = bitcast float %1826 to i32, !dbg !255 + %3104 = bitcast float %1828 to i32, !dbg !255 + %3105 = bitcast float %1830 to i32, !dbg !255 + %3106 = bitcast float %1832 to i32, !dbg !255 + %3107 = select i1 %2606, i32 %3084, i32 %3086, !dbg !248 + %3108 = select i1 %2606, i32 %3088, i32 %3090, !dbg !248 + %3109 = select i1 %2606, i32 %3092, i32 %3094, !dbg !248 + %3110 = select i1 %2606, i32 %3096, i32 %3098, !dbg !248 + %3111 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3066, float %3067, float %3068, float %3069, i32 %3099, i32 %3100, i32 %3101, i32 %3102, i32 %3107, i32 %3108) #2, !dbg !255 + %3112 = extractvalue { float, float, float, float } %3111, 0, !dbg !255 + %3113 = extractvalue { float, float, float, float } %3111, 1, !dbg !255 + %3114 = extractvalue { float, float, float, float } %3111, 2, !dbg !255 + %3115 = extractvalue { float, float, float, float } %3111, 3, !dbg !255 + %3116 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3071, float %3072, float %3073, float %3074, i32 %3099, i32 %3100, i32 %3101, i32 %3102, i32 %3109, i32 %3110) #2, !dbg !255 + %3117 = extractvalue { float, float, float, float } %3116, 0, !dbg !255 + %3118 = extractvalue { float, float, float, float } %3116, 1, !dbg !255 + %3119 = extractvalue { float, float, float, float } %3116, 2, !dbg !255 + %3120 = extractvalue { float, float, float, float } %3116, 3, !dbg !255 + %3121 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3112, float %3113, float %3114, float %3115, i32 %3103, i32 %3104, i32 %3105, i32 %3106, i32 %.v154, i32 %.v155) #2, !dbg !255 + %3122 = extractvalue { float, float, float, float } %3121, 0, !dbg !255 + %3123 = extractvalue { float, float, float, float } %3121, 1, !dbg !255 + %3124 = extractvalue { float, float, float, float } %3121, 2, !dbg !255 + %3125 = extractvalue { float, float, float, float } %3121, 3, !dbg !255 + %3126 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3117, float %3118, float %3119, float %3120, i32 %3103, i32 %3104, i32 %3105, i32 %3106, i32 %.v156, i32 %.v157) #2, !dbg !255 + %3127 = extractvalue { float, float, float, float } %3126, 0, !dbg !255 + %3128 = extractvalue { float, float, float, float } %3126, 1, !dbg !255 + %3129 = extractvalue { float, float, float, float } %3126, 2, !dbg !255 + %3130 = extractvalue { float, float, float, float } %3126, 3, !dbg !255 + %3131 = select i1 %.not125, float %2591, float %2593, !dbg !227 + %3132 = select i1 %.not125, float %2593, float %2591, !dbg !227 + %3133 = select i1 %.not125, float %.pn101.lcssa, float %.pn99.lcssa, !dbg !227 + %3134 = select i1 %.not125, float %.pn99.lcssa, float %.pn101.lcssa, !dbg !227 + %3135 = select i1 %.not125, float %.pn97.lcssa, float %.pn95.lcssa, !dbg !227 + %3136 = select i1 %.not125, float %.pn95.lcssa, float %.pn97.lcssa, !dbg !227 + %3137 = select i1 %.not125, float %2595, float %2597, !dbg !227 + %3138 = select i1 %.not125, float %2597, float %2595, !dbg !227 + %3139 = bitcast float %3131 to i32, !dbg !227 + %3140 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3139, i32 %2678, i32 31), !dbg !227 + %3141 = bitcast float %3133 to i32, !dbg !227 + %3142 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3141, i32 %2678, i32 31), !dbg !227 + %3143 = bitcast float %3132 to i32, !dbg !227 + %3144 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3143, i32 %2683, i32 31), !dbg !227 + %3145 = bitcast float %3134 to i32, !dbg !227 + %3146 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3145, i32 %2683, i32 31), !dbg !227 + %3147 = bitcast float %3135 to i32, !dbg !227 + %3148 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3147, i32 %2678, i32 31), !dbg !227 + %3149 = bitcast float %3137 to i32, !dbg !227 + %3150 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3149, i32 %2678, i32 31), !dbg !227 + %3151 = bitcast float %3136 to i32, !dbg !227 + %3152 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3151, i32 %2683, i32 31), !dbg !227 + %3153 = bitcast float %3138 to i32, !dbg !227 + %3154 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3153, i32 %2683, i32 31), !dbg !227 + %.v158 = select i1 %2676, i32 %3144, i32 %3140, !dbg !227 + %.v159 = select i1 %2676, i32 %3146, i32 %3142, !dbg !227 + %.v160 = select i1 %2676, i32 %3152, i32 %3148, !dbg !227 + %.v161 = select i1 %2676, i32 %3154, i32 %3150, !dbg !227 + %3155 = select i1 %2829, float %3122, float %3123, !dbg !256 + %3156 = select i1 %2829, float %3123, float %3122, !dbg !256 + %3157 = select i1 %2829, float %3124, float %3125, !dbg !256 + %3158 = select i1 %2829, float %3125, float %3124, !dbg !256 + %3159 = select i1 %2829, float %3127, float %3128, !dbg !256 + %3160 = select i1 %2829, float %3128, float %3127, !dbg !256 + %3161 = select i1 %2829, float %3129, float %3130, !dbg !256 + %3162 = select i1 %2829, float %3130, float %3129, !dbg !256 + %3163 = bitcast float %3155 to i32, !dbg !256 + %3164 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3163, i32 %2840, i32 31), !dbg !256 + %3165 = bitcast float %3156 to i32, !dbg !256 + %3166 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3165, i32 %2843, i32 31), !dbg !256 + %3167 = bitcast float %3157 to i32, !dbg !256 + %3168 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3167, i32 %2840, i32 31), !dbg !256 + %3169 = bitcast float %3158 to i32, !dbg !256 + %3170 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3169, i32 %2843, i32 31), !dbg !256 + %3171 = bitcast float %3159 to i32, !dbg !256 + %3172 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3171, i32 %2840, i32 31), !dbg !256 + %3173 = bitcast float %3160 to i32, !dbg !256 + %3174 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3173, i32 %2843, i32 31), !dbg !256 + %3175 = bitcast float %3161 to i32, !dbg !256 + %3176 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3175, i32 %2840, i32 31), !dbg !256 + %3177 = bitcast float %3162 to i32, !dbg !256 + %3178 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3177, i32 %2843, i32 31), !dbg !256 + %.v162 = select i1 %2606, i32 %3166, i32 %3164, !dbg !256 + %.v163 = select i1 %2606, i32 %3170, i32 %3168, !dbg !256 + %.v164 = select i1 %2606, i32 %3174, i32 %3172, !dbg !256 + %.v165 = select i1 %2606, i32 %3178, i32 %3176, !dbg !256 + %3179 = select i1 %2676, i32 %3140, i32 %3144, !dbg !227 + %3180 = select i1 %2676, i32 %3142, i32 %3146, !dbg !227 + %3181 = select i1 %2676, i32 %3148, i32 %3152, !dbg !227 + %3182 = select i1 %2676, i32 %3150, i32 %3154, !dbg !227 + %3183 = select i1 %2606, i32 %3164, i32 %3166, !dbg !256 + %3184 = select i1 %2606, i32 %3168, i32 %3170, !dbg !256 + %3185 = select i1 %2606, i32 %3172, i32 %3174, !dbg !256 + %3186 = select i1 %2606, i32 %3176, i32 %3178, !dbg !256 + %3187 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %3179, i32 %3180, i32 %.v158, i32 %.v159, i32 %3183, i32 %.v162) #2, !dbg !256 + %3188 = extractvalue { float, float, float, float } %3187, 0, !dbg !256 + %3189 = extractvalue { float, float, float, float } %3187, 1, !dbg !256 + %3190 = extractvalue { float, float, float, float } %3187, 2, !dbg !256 + %3191 = extractvalue { float, float, float, float } %3187, 3, !dbg !256 + %3192 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %3179, i32 %3180, i32 %.v158, i32 %.v159, i32 %3185, i32 %.v164) #2, !dbg !256 + %3193 = extractvalue { float, float, float, float } %3192, 0, !dbg !256 + %3194 = extractvalue { float, float, float, float } %3192, 1, !dbg !256 + %3195 = extractvalue { float, float, float, float } %3192, 2, !dbg !256 + %3196 = extractvalue { float, float, float, float } %3192, 3, !dbg !256 + %3197 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3188, float %3189, float %3190, float %3191, i32 %3181, i32 %3182, i32 %.v160, i32 %.v161, i32 %3184, i32 %.v163) #2, !dbg !256 + %3198 = extractvalue { float, float, float, float } %3197, 0, !dbg !256 + %3199 = extractvalue { float, float, float, float } %3197, 1, !dbg !256 + %3200 = extractvalue { float, float, float, float } %3197, 2, !dbg !256 + %3201 = extractvalue { float, float, float, float } %3197, 3, !dbg !256 + %3202 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3193, float %3194, float %3195, float %3196, i32 %3181, i32 %3182, i32 %.v160, i32 %.v161, i32 %3186, i32 %.v165) #2, !dbg !256 + %3203 = extractvalue { float, float, float, float } %3202, 0, !dbg !256 + %3204 = extractvalue { float, float, float, float } %3202, 1, !dbg !256 + %3205 = extractvalue { float, float, float, float } %3202, 2, !dbg !256 + %3206 = extractvalue { float, float, float, float } %3202, 3, !dbg !256 + %3207 = fsub float 0.000000e+00, %3198, !dbg !257 + %3208 = fsub float 0.000000e+00, %3199, !dbg !257 + %3209 = fsub float 0.000000e+00, %3200, !dbg !257 + %3210 = fsub float 0.000000e+00, %3201, !dbg !257 + %3211 = fsub float 0.000000e+00, %3203, !dbg !257 + %3212 = fsub float 0.000000e+00, %3204, !dbg !257 + %3213 = fsub float 0.000000e+00, %3205, !dbg !257 + %3214 = fsub float 0.000000e+00, %3206, !dbg !257 + %3215 = select i1 %2606, float %2583, float %2585, !dbg !210 + %3216 = select i1 %2606, float %2585, float %2583, !dbg !210 + %3217 = select i1 %2606, float %.pn85.lcssa, float %.pn83.lcssa, !dbg !210 + %3218 = select i1 %2606, float %.pn83.lcssa, float %.pn85.lcssa, !dbg !210 + %3219 = select i1 %2606, float %.pn81.lcssa, float %.pn79.lcssa, !dbg !210 + %3220 = select i1 %2606, float %.pn79.lcssa, float %.pn81.lcssa, !dbg !210 + %3221 = select i1 %2606, float %2587, float %2589, !dbg !210 + %3222 = select i1 %2606, float %2589, float %2587, !dbg !210 + %3223 = bitcast float %3215 to i32, !dbg !210 + %3224 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3223, i32 %2616, i32 31), !dbg !210 + %3225 = bitcast float %3216 to i32, !dbg !210 + %3226 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3225, i32 %2619, i32 31), !dbg !210 + %3227 = bitcast float %3217 to i32, !dbg !210 + %3228 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3227, i32 %2616, i32 31), !dbg !210 + %3229 = bitcast float %3218 to i32, !dbg !210 + %3230 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3229, i32 %2619, i32 31), !dbg !210 + %3231 = bitcast float %3219 to i32, !dbg !210 + %3232 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3231, i32 %2616, i32 31), !dbg !210 + %3233 = bitcast float %3220 to i32, !dbg !210 + %3234 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3233, i32 %2619, i32 31), !dbg !210 + %3235 = bitcast float %3221 to i32, !dbg !210 + %3236 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3235, i32 %2616, i32 31), !dbg !210 + %3237 = bitcast float %3222 to i32, !dbg !210 + %3238 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3237, i32 %2619, i32 31), !dbg !210 + %.v166 = select i1 %2606, i32 %3226, i32 %3224, !dbg !210 + %.v167 = select i1 %2606, i32 %3230, i32 %3228, !dbg !210 + %.v168 = select i1 %2606, i32 %3234, i32 %3232, !dbg !210 + %.v169 = select i1 %2606, i32 %3238, i32 %3236, !dbg !210 + %3239 = bitcast float %1841 to i32, !dbg !258 + %3240 = bitcast float %1843 to i32, !dbg !258 + %3241 = bitcast float %1845 to i32, !dbg !258 + %3242 = bitcast float %1847 to i32, !dbg !258 + %3243 = bitcast float %1842 to i32, !dbg !258 + %3244 = bitcast float %1844 to i32, !dbg !258 + %3245 = bitcast float %1846 to i32, !dbg !258 + %3246 = bitcast float %1848 to i32, !dbg !258 + %3247 = select i1 %2606, i32 %3224, i32 %3226, !dbg !210 + %3248 = select i1 %2606, i32 %3228, i32 %3230, !dbg !210 + %3249 = select i1 %2606, i32 %3232, i32 %3234, !dbg !210 + %3250 = select i1 %2606, i32 %3236, i32 %3238, !dbg !210 + %3251 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %3239, i32 %3240, i32 %3241, i32 %3242, i32 %3247, i32 %3248) #2, !dbg !258 + %3252 = extractvalue { float, float, float, float } %3251, 0, !dbg !258 + %3253 = extractvalue { float, float, float, float } %3251, 1, !dbg !258 + %3254 = extractvalue { float, float, float, float } %3251, 2, !dbg !258 + %3255 = extractvalue { float, float, float, float } %3251, 3, !dbg !258 + %3256 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %3239, i32 %3240, i32 %3241, i32 %3242, i32 %3249, i32 %3250) #2, !dbg !258 + %3257 = extractvalue { float, float, float, float } %3256, 0, !dbg !258 + %3258 = extractvalue { float, float, float, float } %3256, 1, !dbg !258 + %3259 = extractvalue { float, float, float, float } %3256, 2, !dbg !258 + %3260 = extractvalue { float, float, float, float } %3256, 3, !dbg !258 + %3261 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3252, float %3253, float %3254, float %3255, i32 %3243, i32 %3244, i32 %3245, i32 %3246, i32 %.v166, i32 %.v167) #2, !dbg !258 + %3262 = extractvalue { float, float, float, float } %3261, 0, !dbg !258 + %3263 = extractvalue { float, float, float, float } %3261, 1, !dbg !258 + %3264 = extractvalue { float, float, float, float } %3261, 2, !dbg !258 + %3265 = extractvalue { float, float, float, float } %3261, 3, !dbg !258 + %3266 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3257, float %3258, float %3259, float %3260, i32 %3243, i32 %3244, i32 %3245, i32 %3246, i32 %.v168, i32 %.v169) #2, !dbg !258 + %3267 = extractvalue { float, float, float, float } %3266, 0, !dbg !258 + %3268 = extractvalue { float, float, float, float } %3266, 1, !dbg !258 + %3269 = extractvalue { float, float, float, float } %3266, 2, !dbg !258 + %3270 = extractvalue { float, float, float, float } %3266, 3, !dbg !258 + %3271 = select i1 %2606, float %2886, float %2887, !dbg !251 + %3272 = select i1 %2606, float %2887, float %2886, !dbg !251 + %3273 = select i1 %2606, float %2888, float %2889, !dbg !251 + %3274 = select i1 %2606, float %2889, float %2888, !dbg !251 + %3275 = select i1 %2606, float %2890, float %2891, !dbg !251 + %3276 = select i1 %2606, float %2891, float %2890, !dbg !251 + %3277 = select i1 %2606, float %2892, float %2893, !dbg !251 + %3278 = select i1 %2606, float %2893, float %2892, !dbg !251 + %3279 = bitcast float %3271 to i32, !dbg !251 + %3280 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3279, i32 %2616, i32 31), !dbg !251 + %3281 = bitcast float %3272 to i32, !dbg !251 + %3282 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3281, i32 %2619, i32 31), !dbg !251 + %3283 = bitcast float %3273 to i32, !dbg !251 + %3284 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3283, i32 %2616, i32 31), !dbg !251 + %3285 = bitcast float %3274 to i32, !dbg !251 + %3286 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3285, i32 %2619, i32 31), !dbg !251 + %3287 = bitcast float %3275 to i32, !dbg !251 + %3288 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3287, i32 %2616, i32 31), !dbg !251 + %3289 = bitcast float %3276 to i32, !dbg !251 + %3290 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3289, i32 %2619, i32 31), !dbg !251 + %3291 = bitcast float %3277 to i32, !dbg !251 + %3292 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3291, i32 %2616, i32 31), !dbg !251 + %3293 = bitcast float %3278 to i32, !dbg !251 + %3294 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3293, i32 %2619, i32 31), !dbg !251 + %.v170 = select i1 %2606, i32 %3282, i32 %3280, !dbg !251 + %.v171 = select i1 %2606, i32 %3286, i32 %3284, !dbg !251 + %.v172 = select i1 %2606, i32 %3290, i32 %3288, !dbg !251 + %.v173 = select i1 %2606, i32 %3294, i32 %3292, !dbg !251 + %3295 = bitcast float %1849 to i32, !dbg !259 + %3296 = bitcast float %1851 to i32, !dbg !259 + %3297 = bitcast float %1853 to i32, !dbg !259 + %3298 = bitcast float %1855 to i32, !dbg !259 + %3299 = bitcast float %1850 to i32, !dbg !259 + %3300 = bitcast float %1852 to i32, !dbg !259 + %3301 = bitcast float %1854 to i32, !dbg !259 + %3302 = bitcast float %1856 to i32, !dbg !259 + %3303 = select i1 %2606, i32 %3280, i32 %3282, !dbg !251 + %3304 = select i1 %2606, i32 %3284, i32 %3286, !dbg !251 + %3305 = select i1 %2606, i32 %3288, i32 %3290, !dbg !251 + %3306 = select i1 %2606, i32 %3292, i32 %3294, !dbg !251 + %3307 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3262, float %3263, float %3264, float %3265, i32 %3295, i32 %3296, i32 %3297, i32 %3298, i32 %3303, i32 %3304) #2, !dbg !259 + %3308 = extractvalue { float, float, float, float } %3307, 0, !dbg !259 + %3309 = extractvalue { float, float, float, float } %3307, 1, !dbg !259 + %3310 = extractvalue { float, float, float, float } %3307, 2, !dbg !259 + %3311 = extractvalue { float, float, float, float } %3307, 3, !dbg !259 + %3312 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3267, float %3268, float %3269, float %3270, i32 %3295, i32 %3296, i32 %3297, i32 %3298, i32 %3305, i32 %3306) #2, !dbg !259 + %3313 = extractvalue { float, float, float, float } %3312, 0, !dbg !259 + %3314 = extractvalue { float, float, float, float } %3312, 1, !dbg !259 + %3315 = extractvalue { float, float, float, float } %3312, 2, !dbg !259 + %3316 = extractvalue { float, float, float, float } %3312, 3, !dbg !259 + %3317 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3308, float %3309, float %3310, float %3311, i32 %3299, i32 %3300, i32 %3301, i32 %3302, i32 %.v170, i32 %.v171) #2, !dbg !259 + %3318 = extractvalue { float, float, float, float } %3317, 0, !dbg !259 + %3319 = extractvalue { float, float, float, float } %3317, 1, !dbg !259 + %3320 = extractvalue { float, float, float, float } %3317, 2, !dbg !259 + %3321 = extractvalue { float, float, float, float } %3317, 3, !dbg !259 + %3322 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3313, float %3314, float %3315, float %3316, i32 %3299, i32 %3300, i32 %3301, i32 %3302, i32 %.v172, i32 %.v173) #2, !dbg !259 + %3323 = extractvalue { float, float, float, float } %3322, 0, !dbg !259 + %3324 = extractvalue { float, float, float, float } %3322, 1, !dbg !259 + %3325 = extractvalue { float, float, float, float } %3322, 2, !dbg !259 + %3326 = extractvalue { float, float, float, float } %3322, 3, !dbg !259 + %3327 = select i1 %.not125, float %2599, float %2601, !dbg !244 + %3328 = select i1 %.not125, float %2601, float %2599, !dbg !244 + %3329 = select i1 %.not125, float %.pn117.lcssa, float %.pn115.lcssa, !dbg !244 + %3330 = select i1 %.not125, float %.pn115.lcssa, float %.pn117.lcssa, !dbg !244 + %3331 = select i1 %.not125, float %.pn113.lcssa, float %.pn111.lcssa, !dbg !244 + %3332 = select i1 %.not125, float %.pn111.lcssa, float %.pn113.lcssa, !dbg !244 + %3333 = select i1 %.not125, float %2603, float %2605, !dbg !244 + %3334 = select i1 %.not125, float %2605, float %2603, !dbg !244 + %3335 = bitcast float %3327 to i32, !dbg !244 + %3336 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3335, i32 %2678, i32 31), !dbg !244 + %3337 = bitcast float %3329 to i32, !dbg !244 + %3338 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3337, i32 %2678, i32 31), !dbg !244 + %3339 = bitcast float %3328 to i32, !dbg !244 + %3340 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3339, i32 %2683, i32 31), !dbg !244 + %3341 = bitcast float %3330 to i32, !dbg !244 + %3342 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3341, i32 %2683, i32 31), !dbg !244 + %3343 = bitcast float %3331 to i32, !dbg !244 + %3344 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3343, i32 %2678, i32 31), !dbg !244 + %3345 = bitcast float %3333 to i32, !dbg !244 + %3346 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3345, i32 %2678, i32 31), !dbg !244 + %3347 = bitcast float %3332 to i32, !dbg !244 + %3348 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3347, i32 %2683, i32 31), !dbg !244 + %3349 = bitcast float %3334 to i32, !dbg !244 + %3350 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3349, i32 %2683, i32 31), !dbg !244 + %.v174 = select i1 %2676, i32 %3340, i32 %3336, !dbg !244 + %.v175 = select i1 %2676, i32 %3342, i32 %3338, !dbg !244 + %.v176 = select i1 %2676, i32 %3348, i32 %3344, !dbg !244 + %.v177 = select i1 %2676, i32 %3350, i32 %3346, !dbg !244 + %3351 = select i1 %2829, float %3318, float %3319, !dbg !260 + %3352 = select i1 %2829, float %3319, float %3318, !dbg !260 + %3353 = select i1 %2829, float %3320, float %3321, !dbg !260 + %3354 = select i1 %2829, float %3321, float %3320, !dbg !260 + %3355 = select i1 %2829, float %3323, float %3324, !dbg !260 + %3356 = select i1 %2829, float %3324, float %3323, !dbg !260 + %3357 = select i1 %2829, float %3325, float %3326, !dbg !260 + %3358 = select i1 %2829, float %3326, float %3325, !dbg !260 + %3359 = bitcast float %3351 to i32, !dbg !260 + %3360 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3359, i32 %2840, i32 31), !dbg !260 + %3361 = bitcast float %3352 to i32, !dbg !260 + %3362 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3361, i32 %2843, i32 31), !dbg !260 + %3363 = bitcast float %3353 to i32, !dbg !260 + %3364 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3363, i32 %2840, i32 31), !dbg !260 + %3365 = bitcast float %3354 to i32, !dbg !260 + %3366 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3365, i32 %2843, i32 31), !dbg !260 + %3367 = bitcast float %3355 to i32, !dbg !260 + %3368 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3367, i32 %2840, i32 31), !dbg !260 + %3369 = bitcast float %3356 to i32, !dbg !260 + %3370 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3369, i32 %2843, i32 31), !dbg !260 + %3371 = bitcast float %3357 to i32, !dbg !260 + %3372 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3371, i32 %2840, i32 31), !dbg !260 + %3373 = bitcast float %3358 to i32, !dbg !260 + %3374 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3373, i32 %2843, i32 31), !dbg !260 + %.v178 = select i1 %2606, i32 %3362, i32 %3360, !dbg !260 + %.v179 = select i1 %2606, i32 %3366, i32 %3364, !dbg !260 + %.v180 = select i1 %2606, i32 %3370, i32 %3368, !dbg !260 + %.v181 = select i1 %2606, i32 %3374, i32 %3372, !dbg !260 + %3375 = select i1 %2676, i32 %3336, i32 %3340, !dbg !244 + %3376 = select i1 %2676, i32 %3338, i32 %3342, !dbg !244 + %3377 = select i1 %2676, i32 %3344, i32 %3348, !dbg !244 + %3378 = select i1 %2676, i32 %3346, i32 %3350, !dbg !244 + %3379 = select i1 %2606, i32 %3360, i32 %3362, !dbg !260 + %3380 = select i1 %2606, i32 %3364, i32 %3366, !dbg !260 + %3381 = select i1 %2606, i32 %3368, i32 %3370, !dbg !260 + %3382 = select i1 %2606, i32 %3372, i32 %3374, !dbg !260 + %3383 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %3375, i32 %3376, i32 %.v174, i32 %.v175, i32 %3379, i32 %.v178) #2, !dbg !260 + %3384 = extractvalue { float, float, float, float } %3383, 0, !dbg !260 + %3385 = extractvalue { float, float, float, float } %3383, 1, !dbg !260 + %3386 = extractvalue { float, float, float, float } %3383, 2, !dbg !260 + %3387 = extractvalue { float, float, float, float } %3383, 3, !dbg !260 + %3388 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %3375, i32 %3376, i32 %.v174, i32 %.v175, i32 %3381, i32 %.v180) #2, !dbg !260 + %3389 = extractvalue { float, float, float, float } %3388, 0, !dbg !260 + %3390 = extractvalue { float, float, float, float } %3388, 1, !dbg !260 + %3391 = extractvalue { float, float, float, float } %3388, 2, !dbg !260 + %3392 = extractvalue { float, float, float, float } %3388, 3, !dbg !260 + %3393 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3384, float %3385, float %3386, float %3387, i32 %3377, i32 %3378, i32 %.v176, i32 %.v177, i32 %3380, i32 %.v179) #2, !dbg !260 + %3394 = extractvalue { float, float, float, float } %3393, 0, !dbg !260 + %3395 = extractvalue { float, float, float, float } %3393, 1, !dbg !260 + %3396 = extractvalue { float, float, float, float } %3393, 2, !dbg !260 + %3397 = extractvalue { float, float, float, float } %3393, 3, !dbg !260 + %3398 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3389, float %3390, float %3391, float %3392, i32 %3377, i32 %3378, i32 %.v176, i32 %.v177, i32 %3382, i32 %.v181) #2, !dbg !260 + %3399 = extractvalue { float, float, float, float } %3398, 0, !dbg !260 + %3400 = extractvalue { float, float, float, float } %3398, 1, !dbg !260 + %3401 = extractvalue { float, float, float, float } %3398, 2, !dbg !260 + %3402 = extractvalue { float, float, float, float } %3398, 3, !dbg !260 + %3403 = bitcast float %1833 to i32, !dbg !261 + %3404 = bitcast float %1835 to i32, !dbg !261 + %3405 = bitcast float %1837 to i32, !dbg !261 + %3406 = bitcast float %1839 to i32, !dbg !261 + %3407 = bitcast float %1834 to i32, !dbg !261 + %3408 = bitcast float %1836 to i32, !dbg !261 + %3409 = bitcast float %1838 to i32, !dbg !261 + %3410 = bitcast float %1840 to i32, !dbg !261 + %3411 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %3403, i32 %3404, i32 %3405, i32 %3406, i32 %3031, i32 %3035) #2, !dbg !261 + %3412 = extractvalue { float, float, float, float } %3411, 0, !dbg !261 + %3413 = extractvalue { float, float, float, float } %3411, 1, !dbg !261 + %3414 = extractvalue { float, float, float, float } %3411, 2, !dbg !261 + %3415 = extractvalue { float, float, float, float } %3411, 3, !dbg !261 + %3416 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %3403, i32 %3404, i32 %3405, i32 %3406, i32 %3040, i32 %3044) #2, !dbg !261 + %3417 = extractvalue { float, float, float, float } %3416, 0, !dbg !261 + %3418 = extractvalue { float, float, float, float } %3416, 1, !dbg !261 + %3419 = extractvalue { float, float, float, float } %3416, 2, !dbg !261 + %3420 = extractvalue { float, float, float, float } %3416, 3, !dbg !261 + %3421 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3412, float %3413, float %3414, float %3415, i32 %3407, i32 %3408, i32 %3409, i32 %3410, i32 %3033, i32 %3037) #2, !dbg !261 + %3422 = extractvalue { float, float, float, float } %3421, 0, !dbg !261 + %3423 = extractvalue { float, float, float, float } %3421, 1, !dbg !261 + %3424 = extractvalue { float, float, float, float } %3421, 2, !dbg !261 + %3425 = extractvalue { float, float, float, float } %3421, 3, !dbg !261 + %3426 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3417, float %3418, float %3419, float %3420, i32 %3407, i32 %3408, i32 %3409, i32 %3410, i32 %3042, i32 %3046) #2, !dbg !261 + %3427 = extractvalue { float, float, float, float } %3426, 0, !dbg !261 + %3428 = extractvalue { float, float, float, float } %3426, 1, !dbg !261 + %3429 = extractvalue { float, float, float, float } %3426, 2, !dbg !261 + %3430 = extractvalue { float, float, float, float } %3426, 3, !dbg !261 + %3431 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3422, float %3423, float %3424, float %3425, i32 %3239, i32 %3240, i32 %3241, i32 %3242, i32 %3107, i32 %3108) #2, !dbg !262 + %3432 = extractvalue { float, float, float, float } %3431, 0, !dbg !262 + %3433 = extractvalue { float, float, float, float } %3431, 1, !dbg !262 + %3434 = extractvalue { float, float, float, float } %3431, 2, !dbg !262 + %3435 = extractvalue { float, float, float, float } %3431, 3, !dbg !262 + %3436 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3427, float %3428, float %3429, float %3430, i32 %3239, i32 %3240, i32 %3241, i32 %3242, i32 %3109, i32 %3110) #2, !dbg !262 + %3437 = extractvalue { float, float, float, float } %3436, 0, !dbg !262 + %3438 = extractvalue { float, float, float, float } %3436, 1, !dbg !262 + %3439 = extractvalue { float, float, float, float } %3436, 2, !dbg !262 + %3440 = extractvalue { float, float, float, float } %3436, 3, !dbg !262 + %3441 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3432, float %3433, float %3434, float %3435, i32 %3243, i32 %3244, i32 %3245, i32 %3246, i32 %.v154, i32 %.v155) #2, !dbg !262 + %3442 = extractvalue { float, float, float, float } %3441, 0, !dbg !262 + %3443 = extractvalue { float, float, float, float } %3441, 1, !dbg !262 + %3444 = extractvalue { float, float, float, float } %3441, 2, !dbg !262 + %3445 = extractvalue { float, float, float, float } %3441, 3, !dbg !262 + %3446 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3437, float %3438, float %3439, float %3440, i32 %3243, i32 %3244, i32 %3245, i32 %3246, i32 %.v156, i32 %.v157) #2, !dbg !262 + %3447 = extractvalue { float, float, float, float } %3446, 0, !dbg !262 + %3448 = extractvalue { float, float, float, float } %3446, 1, !dbg !262 + %3449 = extractvalue { float, float, float, float } %3446, 2, !dbg !262 + %3450 = extractvalue { float, float, float, float } %3446, 3, !dbg !262 + %3451 = select i1 %2606, float %3207, float %3208, !dbg !257 + %3452 = select i1 %2606, float %3208, float %3207, !dbg !257 + %3453 = select i1 %2606, float %3209, float %3210, !dbg !257 + %3454 = select i1 %2606, float %3210, float %3209, !dbg !257 + %3455 = select i1 %2606, float %3211, float %3212, !dbg !257 + %3456 = select i1 %2606, float %3212, float %3211, !dbg !257 + %3457 = select i1 %2606, float %3213, float %3214, !dbg !257 + %3458 = select i1 %2606, float %3214, float %3213, !dbg !257 + %3459 = bitcast float %3451 to i32, !dbg !257 + %3460 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3459, i32 %2616, i32 31), !dbg !257 + %3461 = bitcast float %3452 to i32, !dbg !257 + %3462 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3461, i32 %2619, i32 31), !dbg !257 + %3463 = bitcast float %3453 to i32, !dbg !257 + %3464 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3463, i32 %2616, i32 31), !dbg !257 + %3465 = bitcast float %3454 to i32, !dbg !257 + %3466 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3465, i32 %2619, i32 31), !dbg !257 + %3467 = bitcast float %3455 to i32, !dbg !257 + %3468 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3467, i32 %2616, i32 31), !dbg !257 + %3469 = bitcast float %3456 to i32, !dbg !257 + %3470 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3469, i32 %2619, i32 31), !dbg !257 + %3471 = bitcast float %3457 to i32, !dbg !257 + %3472 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3471, i32 %2616, i32 31), !dbg !257 + %3473 = bitcast float %3458 to i32, !dbg !257 + %3474 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3473, i32 %2619, i32 31), !dbg !257 + %.v182 = select i1 %2606, i32 %3462, i32 %3460, !dbg !257 + %.v183 = select i1 %2606, i32 %3466, i32 %3464, !dbg !257 + %.v184 = select i1 %2606, i32 %3470, i32 %3468, !dbg !257 + %.v185 = select i1 %2606, i32 %3474, i32 %3472, !dbg !257 + %3475 = select i1 %2606, i32 %3460, i32 %3462, !dbg !257 + %3476 = select i1 %2606, i32 %3464, i32 %3466, !dbg !257 + %3477 = select i1 %2606, i32 %3468, i32 %3470, !dbg !257 + %3478 = select i1 %2606, i32 %3472, i32 %3474, !dbg !257 + %3479 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3442, float %3443, float %3444, float %3445, i32 %3295, i32 %3296, i32 %3297, i32 %3298, i32 %3475, i32 %3476) #2, !dbg !263 + %3480 = extractvalue { float, float, float, float } %3479, 0, !dbg !263 + %3481 = extractvalue { float, float, float, float } %3479, 1, !dbg !263 + %3482 = extractvalue { float, float, float, float } %3479, 2, !dbg !263 + %3483 = extractvalue { float, float, float, float } %3479, 3, !dbg !263 + %3484 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3447, float %3448, float %3449, float %3450, i32 %3295, i32 %3296, i32 %3297, i32 %3298, i32 %3477, i32 %3478) #2, !dbg !263 + %3485 = extractvalue { float, float, float, float } %3484, 0, !dbg !263 + %3486 = extractvalue { float, float, float, float } %3484, 1, !dbg !263 + %3487 = extractvalue { float, float, float, float } %3484, 2, !dbg !263 + %3488 = extractvalue { float, float, float, float } %3484, 3, !dbg !263 + %3489 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3480, float %3481, float %3482, float %3483, i32 %3299, i32 %3300, i32 %3301, i32 %3302, i32 %.v182, i32 %.v183) #2, !dbg !263 + %3490 = extractvalue { float, float, float, float } %3489, 0, !dbg !263 + %3491 = extractvalue { float, float, float, float } %3489, 1, !dbg !263 + %3492 = extractvalue { float, float, float, float } %3489, 2, !dbg !263 + %3493 = extractvalue { float, float, float, float } %3489, 3, !dbg !263 + %3494 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3485, float %3486, float %3487, float %3488, i32 %3299, i32 %3300, i32 %3301, i32 %3302, i32 %.v184, i32 %.v185) #2, !dbg !263 + %3495 = extractvalue { float, float, float, float } %3494, 0, !dbg !263 + %3496 = extractvalue { float, float, float, float } %3494, 1, !dbg !263 + %3497 = extractvalue { float, float, float, float } %3494, 2, !dbg !263 + %3498 = extractvalue { float, float, float, float } %3494, 3, !dbg !263 + %3499 = select i1 %2829, float %3490, float %3491, !dbg !264 + %3500 = select i1 %2829, float %3491, float %3490, !dbg !264 + %3501 = select i1 %2829, float %3492, float %3493, !dbg !264 + %3502 = select i1 %2829, float %3493, float %3492, !dbg !264 + %3503 = select i1 %2829, float %3495, float %3496, !dbg !264 + %3504 = select i1 %2829, float %3496, float %3495, !dbg !264 + %3505 = select i1 %2829, float %3497, float %3498, !dbg !264 + %3506 = select i1 %2829, float %3498, float %3497, !dbg !264 + %3507 = bitcast float %3499 to i32, !dbg !264 + %3508 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3507, i32 %2840, i32 31), !dbg !264 + %3509 = bitcast float %3500 to i32, !dbg !264 + %3510 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3509, i32 %2843, i32 31), !dbg !264 + %3511 = bitcast float %3501 to i32, !dbg !264 + %3512 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3511, i32 %2840, i32 31), !dbg !264 + %3513 = bitcast float %3502 to i32, !dbg !264 + %3514 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3513, i32 %2843, i32 31), !dbg !264 + %3515 = bitcast float %3503 to i32, !dbg !264 + %3516 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3515, i32 %2840, i32 31), !dbg !264 + %3517 = bitcast float %3504 to i32, !dbg !264 + %3518 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3517, i32 %2843, i32 31), !dbg !264 + %3519 = bitcast float %3505 to i32, !dbg !264 + %3520 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3519, i32 %2840, i32 31), !dbg !264 + %3521 = bitcast float %3506 to i32, !dbg !264 + %3522 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3521, i32 %2843, i32 31), !dbg !264 + %.v186 = select i1 %2606, i32 %3510, i32 %3508, !dbg !264 + %.v187 = select i1 %2606, i32 %3514, i32 %3512, !dbg !264 + %.v188 = select i1 %2606, i32 %3518, i32 %3516, !dbg !264 + %.v189 = select i1 %2606, i32 %3522, i32 %3520, !dbg !264 + %3523 = select i1 %2606, i32 %3508, i32 %3510, !dbg !264 + %3524 = select i1 %2606, i32 %3512, i32 %3514, !dbg !264 + %3525 = select i1 %2606, i32 %3516, i32 %3518, !dbg !264 + %3526 = select i1 %2606, i32 %3520, i32 %3522, !dbg !264 + %3527 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %3375, i32 %3376, i32 %.v174, i32 %.v175, i32 %3523, i32 %.v186) #2, !dbg !264 + %3528 = extractvalue { float, float, float, float } %3527, 0, !dbg !264 + %3529 = extractvalue { float, float, float, float } %3527, 1, !dbg !264 + %3530 = extractvalue { float, float, float, float } %3527, 2, !dbg !264 + %3531 = extractvalue { float, float, float, float } %3527, 3, !dbg !264 + %3532 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %3375, i32 %3376, i32 %.v174, i32 %.v175, i32 %3525, i32 %.v188) #2, !dbg !264 + %3533 = extractvalue { float, float, float, float } %3532, 0, !dbg !264 + %3534 = extractvalue { float, float, float, float } %3532, 1, !dbg !264 + %3535 = extractvalue { float, float, float, float } %3532, 2, !dbg !264 + %3536 = extractvalue { float, float, float, float } %3532, 3, !dbg !264 + %3537 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3528, float %3529, float %3530, float %3531, i32 %3377, i32 %3378, i32 %.v176, i32 %.v177, i32 %3524, i32 %.v187) #2, !dbg !264 + %3538 = extractvalue { float, float, float, float } %3537, 0, !dbg !264 + %3539 = extractvalue { float, float, float, float } %3537, 1, !dbg !264 + %3540 = extractvalue { float, float, float, float } %3537, 2, !dbg !264 + %3541 = extractvalue { float, float, float, float } %3537, 3, !dbg !264 + %3542 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3533, float %3534, float %3535, float %3536, i32 %3377, i32 %3378, i32 %.v176, i32 %.v177, i32 %3526, i32 %.v189) #2, !dbg !264 + %3543 = extractvalue { float, float, float, float } %3542, 0, !dbg !264 + %3544 = extractvalue { float, float, float, float } %3542, 1, !dbg !264 + %3545 = extractvalue { float, float, float, float } %3542, 2, !dbg !264 + %3546 = extractvalue { float, float, float, float } %3542, 3, !dbg !264 + %.idx190 = shl nuw nsw i64 %1077, 12, !dbg !265 + %3547 = getelementptr i8, ptr addrspace(1) %1072, i64 %.idx190, !dbg !265 + %3548 = zext nneg i32 %1075 to i64, !dbg !266 + %3549 = getelementptr bfloat, ptr addrspace(1) %3547, i64 %3548, !dbg !266 + %.idx191 = shl nuw nsw i64 %1079, 12, !dbg !267 + %3550 = getelementptr i8, ptr addrspace(1) %1072, i64 %.idx191, !dbg !267 + %3551 = getelementptr bfloat, ptr addrspace(1) %3550, i64 %3548, !dbg !268 + %3552 = or disjoint i32 %1075, 16, !dbg !269 + %3553 = zext nneg i32 %3552 to i64, !dbg !270 + %3554 = getelementptr bfloat, ptr addrspace(1) %3550, i64 %3553, !dbg !270 + %.idx192 = shl nuw nsw i64 %1081, 12, !dbg !271 + %3555 = getelementptr i8, ptr addrspace(1) %1072, i64 %.idx192, !dbg !271 + %3556 = getelementptr bfloat, ptr addrspace(1) %3555, i64 %3548, !dbg !272 + %3557 = getelementptr bfloat, ptr addrspace(1) %3555, i64 %3553, !dbg !273 + %3558 = or disjoint i32 %1075, 32, !dbg !274 + %3559 = zext nneg i32 %3558 to i64, !dbg !275 + %3560 = getelementptr bfloat, ptr addrspace(1) %3555, i64 %3559, !dbg !275 + %.idx193 = shl nuw nsw i64 %1083, 12, !dbg !276 + %3561 = getelementptr i8, ptr addrspace(1) %1072, i64 %.idx193, !dbg !276 + %3562 = getelementptr bfloat, ptr addrspace(1) %3561, i64 %3548, !dbg !277 + %3563 = getelementptr bfloat, ptr addrspace(1) %3561, i64 %3553, !dbg !278 + %3564 = getelementptr bfloat, ptr addrspace(1) %3561, i64 %3559, !dbg !279 + %3565 = getelementptr i8, ptr addrspace(1) %3562, i64 96, !dbg !280 + %3566 = fptrunc float %2561 to bfloat, !dbg !281 + %3567 = fptrunc float %2563 to bfloat, !dbg !281 + %3568 = fptrunc float %2565 to bfloat, !dbg !281 + %3569 = fptrunc float %2567 to bfloat, !dbg !281 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !282 + %3570 = lshr i32 %23, 1, !dbg !282 + %3571 = or i32 %3570, %23, !dbg !282 + %3572 = and i32 %242, 144, !dbg !282 + %3573 = select i1 %.not125, i32 0, i32 320, !dbg !282 + %3574 = and i32 %.pre-phi603, 32, !dbg !282 + %3575 = and i32 %3571, 16, !dbg !282 + %3576 = shl nuw nsw i32 %3575, 2, !dbg !282 + %3577 = or disjoint i32 %3573, %3572, !dbg !282 + %3578 = xor i32 %3577, %3576, !dbg !282 + %3579 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %3574, !dbg !282 + %3580 = getelementptr inbounds nuw i8, ptr addrspace(3) %3579, i32 %3578, !dbg !282 + %3581 = insertelement <2 x bfloat> poison, bfloat %3566, i64 0, !dbg !282 + %3582 = insertelement <2 x bfloat> %3581, bfloat %3567, i64 1, !dbg !282 + %3583 = bitcast <2 x bfloat> %3582 to i32, !dbg !282 + %3584 = insertelement <2 x bfloat> poison, bfloat %3568, i64 0, !dbg !282 + %3585 = insertelement <2 x bfloat> %3584, bfloat %3569, i64 1, !dbg !282 + %3586 = bitcast <2 x bfloat> %3585 to i32, !dbg !282 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x2.trans.b16.p3(ptr addrspace(3) %3580, i32 %3583, i32 %3586), !dbg !282 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !282 + %3587 = and i32 %242, 112, !dbg !282 + %3588 = select i1 %2571, i32 0, i32 320, !dbg !282 + %3589 = shl nuw nsw i32 %3575, 3, !dbg !282 + %3590 = xor i32 %3588, %3587, !dbg !282 + %3591 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %3589, !dbg !282 + %3592 = getelementptr inbounds nuw i8, ptr addrspace(3) %3591, i32 %3590, !dbg !282 + %3593 = tail call { i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.trans.b16.p3(ptr addrspace(3) %3592), !dbg !282 + %3594 = extractvalue { i32, i32 } %3593, 0, !dbg !282 + %3595 = extractvalue { i32, i32 } %3593, 1, !dbg !282 + tail call void asm sideeffect "@$3 st.global.v2.b32 [ $2 + 0 ], { $0, $1 };", "r,r,l,b"(i32 %3594, i32 %3595, ptr addrspace(1) %3549, i1 %1078) #2, !dbg !282 + %3596 = fptrunc float %2741 to bfloat, !dbg !283 + %3597 = fptrunc float %2742 to bfloat, !dbg !283 + %3598 = fptrunc float %2743 to bfloat, !dbg !283 + %3599 = fptrunc float %2744 to bfloat, !dbg !283 + %3600 = fptrunc float %2745 to bfloat, !dbg !283 + %3601 = fptrunc float %2746 to bfloat, !dbg !283 + %3602 = fptrunc float %2747 to bfloat, !dbg !283 + %3603 = fptrunc float %2748 to bfloat, !dbg !283 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !284 + %3604 = and i32 %2838, 64, !dbg !284 + %3605 = xor i32 %3577, %3604, !dbg !284 + %3606 = getelementptr inbounds nuw i8, ptr addrspace(3) %3579, i32 %3605, !dbg !284 + %3607 = insertelement <2 x bfloat> poison, bfloat %3596, i64 0, !dbg !284 + %3608 = insertelement <2 x bfloat> %3607, bfloat %3597, i64 1, !dbg !284 + %3609 = bitcast <2 x bfloat> %3608 to i32, !dbg !284 + %3610 = insertelement <2 x bfloat> poison, bfloat %3598, i64 0, !dbg !284 + %3611 = insertelement <2 x bfloat> %3610, bfloat %3599, i64 1, !dbg !284 + %3612 = bitcast <2 x bfloat> %3611 to i32, !dbg !284 + %3613 = insertelement <2 x bfloat> poison, bfloat %3600, i64 0, !dbg !284 + %3614 = insertelement <2 x bfloat> %3613, bfloat %3601, i64 1, !dbg !284 + %3615 = bitcast <2 x bfloat> %3614 to i32, !dbg !284 + %3616 = insertelement <2 x bfloat> poison, bfloat %3602, i64 0, !dbg !284 + %3617 = insertelement <2 x bfloat> %3616, bfloat %3603, i64 1, !dbg !284 + %3618 = bitcast <2 x bfloat> %3617 to i32, !dbg !284 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %3606, i32 %3609, i32 %3612, i32 %3615, i32 %3618), !dbg !284 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !284 + %3619 = tail call { i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.trans.b16.p3(ptr addrspace(3) %3592), !dbg !284 + %3620 = extractvalue { i32, i32 } %3619, 0, !dbg !284 + %3621 = extractvalue { i32, i32 } %3619, 1, !dbg !284 + tail call void asm sideeffect "@$3 st.global.v2.b32 [ $2 + 0 ], { $0, $1 };", "r,r,l,b"(i32 %3620, i32 %3621, ptr addrspace(1) %3551, i1 %1080) #2, !dbg !284 + %3622 = fptrunc float %2583 to bfloat, !dbg !285 + %3623 = fptrunc float %2585 to bfloat, !dbg !285 + %3624 = fptrunc float %.pn85.lcssa to bfloat, !dbg !285 + %3625 = fptrunc float %.pn83.lcssa to bfloat, !dbg !285 + %3626 = fptrunc float %.pn81.lcssa to bfloat, !dbg !285 + %3627 = fptrunc float %.pn79.lcssa to bfloat, !dbg !285 + %3628 = fptrunc float %2587 to bfloat, !dbg !285 + %3629 = fptrunc float %2589 to bfloat, !dbg !285 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !286 + %3630 = insertelement <2 x bfloat> poison, bfloat %3622, i64 0, !dbg !286 + %3631 = insertelement <2 x bfloat> %3630, bfloat %3623, i64 1, !dbg !286 + %3632 = bitcast <2 x bfloat> %3631 to i32, !dbg !286 + %3633 = insertelement <2 x bfloat> poison, bfloat %3624, i64 0, !dbg !286 + %3634 = insertelement <2 x bfloat> %3633, bfloat %3625, i64 1, !dbg !286 + %3635 = bitcast <2 x bfloat> %3634 to i32, !dbg !286 + %3636 = insertelement <2 x bfloat> poison, bfloat %3626, i64 0, !dbg !286 + %3637 = insertelement <2 x bfloat> %3636, bfloat %3627, i64 1, !dbg !286 + %3638 = bitcast <2 x bfloat> %3637 to i32, !dbg !286 + %3639 = insertelement <2 x bfloat> poison, bfloat %3628, i64 0, !dbg !286 + %3640 = insertelement <2 x bfloat> %3639, bfloat %3629, i64 1, !dbg !286 + %3641 = bitcast <2 x bfloat> %3640 to i32, !dbg !286 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %3606, i32 %3632, i32 %3635, i32 %3638, i32 %3641), !dbg !286 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !286 + %3642 = tail call { i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.trans.b16.p3(ptr addrspace(3) %3592), !dbg !286 + %3643 = extractvalue { i32, i32 } %3642, 0, !dbg !286 + %3644 = extractvalue { i32, i32 } %3642, 1, !dbg !286 + tail call void asm sideeffect "@$3 st.global.v2.b32 [ $2 + 0 ], { $0, $1 };", "r,r,l,b"(i32 %3643, i32 %3644, ptr addrspace(1) %3554, i1 %1080) #2, !dbg !286 + %3645 = fptrunc float %3207 to bfloat, !dbg !287 + %3646 = fptrunc float %3208 to bfloat, !dbg !287 + %3647 = fptrunc float %3209 to bfloat, !dbg !287 + %3648 = fptrunc float %3210 to bfloat, !dbg !287 + %3649 = fptrunc float %3211 to bfloat, !dbg !287 + %3650 = fptrunc float %3212 to bfloat, !dbg !287 + %3651 = fptrunc float %3213 to bfloat, !dbg !287 + %3652 = fptrunc float %3214 to bfloat, !dbg !287 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !288 + %3653 = insertelement <2 x bfloat> poison, bfloat %3645, i64 0, !dbg !288 + %3654 = insertelement <2 x bfloat> %3653, bfloat %3646, i64 1, !dbg !288 + %3655 = bitcast <2 x bfloat> %3654 to i32, !dbg !288 + %3656 = insertelement <2 x bfloat> poison, bfloat %3647, i64 0, !dbg !288 + %3657 = insertelement <2 x bfloat> %3656, bfloat %3648, i64 1, !dbg !288 + %3658 = bitcast <2 x bfloat> %3657 to i32, !dbg !288 + %3659 = insertelement <2 x bfloat> poison, bfloat %3649, i64 0, !dbg !288 + %3660 = insertelement <2 x bfloat> %3659, bfloat %3650, i64 1, !dbg !288 + %3661 = bitcast <2 x bfloat> %3660 to i32, !dbg !288 + %3662 = insertelement <2 x bfloat> poison, bfloat %3651, i64 0, !dbg !288 + %3663 = insertelement <2 x bfloat> %3662, bfloat %3652, i64 1, !dbg !288 + %3664 = bitcast <2 x bfloat> %3663 to i32, !dbg !288 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %3606, i32 %3655, i32 %3658, i32 %3661, i32 %3664), !dbg !288 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !288 + %3665 = tail call { i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.trans.b16.p3(ptr addrspace(3) %3592), !dbg !288 + %3666 = extractvalue { i32, i32 } %3665, 0, !dbg !288 + %3667 = extractvalue { i32, i32 } %3665, 1, !dbg !288 + tail call void asm sideeffect "@$3 st.global.v2.b32 [ $2 + 0 ], { $0, $1 };", "r,r,l,b"(i32 %3666, i32 %3667, ptr addrspace(1) %3556, i1 %1082) #2, !dbg !288 + %3668 = fptrunc float %2886 to bfloat, !dbg !289 + %3669 = fptrunc float %2887 to bfloat, !dbg !289 + %3670 = fptrunc float %2888 to bfloat, !dbg !289 + %3671 = fptrunc float %2889 to bfloat, !dbg !289 + %3672 = fptrunc float %2890 to bfloat, !dbg !289 + %3673 = fptrunc float %2891 to bfloat, !dbg !289 + %3674 = fptrunc float %2892 to bfloat, !dbg !289 + %3675 = fptrunc float %2893 to bfloat, !dbg !289 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !290 + %3676 = insertelement <2 x bfloat> poison, bfloat %3668, i64 0, !dbg !290 + %3677 = insertelement <2 x bfloat> %3676, bfloat %3669, i64 1, !dbg !290 + %3678 = bitcast <2 x bfloat> %3677 to i32, !dbg !290 + %3679 = insertelement <2 x bfloat> poison, bfloat %3670, i64 0, !dbg !290 + %3680 = insertelement <2 x bfloat> %3679, bfloat %3671, i64 1, !dbg !290 + %3681 = bitcast <2 x bfloat> %3680 to i32, !dbg !290 + %3682 = insertelement <2 x bfloat> poison, bfloat %3672, i64 0, !dbg !290 + %3683 = insertelement <2 x bfloat> %3682, bfloat %3673, i64 1, !dbg !290 + %3684 = bitcast <2 x bfloat> %3683 to i32, !dbg !290 + %3685 = insertelement <2 x bfloat> poison, bfloat %3674, i64 0, !dbg !290 + %3686 = insertelement <2 x bfloat> %3685, bfloat %3675, i64 1, !dbg !290 + %3687 = bitcast <2 x bfloat> %3686 to i32, !dbg !290 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %3606, i32 %3678, i32 %3681, i32 %3684, i32 %3687), !dbg !290 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !290 + %3688 = tail call { i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.trans.b16.p3(ptr addrspace(3) %3592), !dbg !290 + %3689 = extractvalue { i32, i32 } %3688, 0, !dbg !290 + %3690 = extractvalue { i32, i32 } %3688, 1, !dbg !290 + tail call void asm sideeffect "@$3 st.global.v2.b32 [ $2 + 0 ], { $0, $1 };", "r,r,l,b"(i32 %3689, i32 %3690, ptr addrspace(1) %3557, i1 %1082) #2, !dbg !290 + %3691 = fptrunc float %2591 to bfloat, !dbg !291 + %3692 = fptrunc float %2593 to bfloat, !dbg !291 + %3693 = fptrunc float %.pn101.lcssa to bfloat, !dbg !291 + %3694 = fptrunc float %.pn99.lcssa to bfloat, !dbg !291 + %3695 = fptrunc float %.pn97.lcssa to bfloat, !dbg !291 + %3696 = fptrunc float %.pn95.lcssa to bfloat, !dbg !291 + %3697 = fptrunc float %2595 to bfloat, !dbg !291 + %3698 = fptrunc float %2597 to bfloat, !dbg !291 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !292 + %3699 = insertelement <2 x bfloat> poison, bfloat %3691, i64 0, !dbg !292 + %3700 = insertelement <2 x bfloat> %3699, bfloat %3692, i64 1, !dbg !292 + %3701 = bitcast <2 x bfloat> %3700 to i32, !dbg !292 + %3702 = insertelement <2 x bfloat> poison, bfloat %3693, i64 0, !dbg !292 + %3703 = insertelement <2 x bfloat> %3702, bfloat %3694, i64 1, !dbg !292 + %3704 = bitcast <2 x bfloat> %3703 to i32, !dbg !292 + %3705 = insertelement <2 x bfloat> poison, bfloat %3695, i64 0, !dbg !292 + %3706 = insertelement <2 x bfloat> %3705, bfloat %3696, i64 1, !dbg !292 + %3707 = bitcast <2 x bfloat> %3706 to i32, !dbg !292 + %3708 = insertelement <2 x bfloat> poison, bfloat %3697, i64 0, !dbg !292 + %3709 = insertelement <2 x bfloat> %3708, bfloat %3698, i64 1, !dbg !292 + %3710 = bitcast <2 x bfloat> %3709 to i32, !dbg !292 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %3606, i32 %3701, i32 %3704, i32 %3707, i32 %3710), !dbg !292 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !292 + %3711 = tail call { i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.trans.b16.p3(ptr addrspace(3) %3592), !dbg !292 + %3712 = extractvalue { i32, i32 } %3711, 0, !dbg !292 + %3713 = extractvalue { i32, i32 } %3711, 1, !dbg !292 + tail call void asm sideeffect "@$3 st.global.v2.b32 [ $2 + 0 ], { $0, $1 };", "r,r,l,b"(i32 %3712, i32 %3713, ptr addrspace(1) %3560, i1 %1082) #2, !dbg !292 + %3714 = insertelement <2 x float> poison, float %3538, i64 0, !dbg !293 + %3715 = insertelement <2 x float> %3714, float %3539, i64 1, !dbg !293 + %3716 = fsub <2 x float> zeroinitializer, %3715, !dbg !293 + %3717 = fptrunc <2 x float> %3716 to <2 x bfloat>, !dbg !294 + %3718 = insertelement <2 x float> poison, float %3540, i64 0, !dbg !293 + %3719 = insertelement <2 x float> %3718, float %3541, i64 1, !dbg !293 + %3720 = fsub <2 x float> zeroinitializer, %3719, !dbg !293 + %3721 = fptrunc <2 x float> %3720 to <2 x bfloat>, !dbg !294 + %3722 = insertelement <2 x float> poison, float %3543, i64 0, !dbg !293 + %3723 = insertelement <2 x float> %3722, float %3544, i64 1, !dbg !293 + %3724 = fsub <2 x float> zeroinitializer, %3723, !dbg !293 + %3725 = fptrunc <2 x float> %3724 to <2 x bfloat>, !dbg !294 + %3726 = insertelement <2 x float> poison, float %3545, i64 0, !dbg !293 + %3727 = insertelement <2 x float> %3726, float %3546, i64 1, !dbg !293 + %3728 = fsub <2 x float> zeroinitializer, %3727, !dbg !293 + %3729 = fptrunc <2 x float> %3728 to <2 x bfloat>, !dbg !294 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !295 + %3730 = bitcast <2 x bfloat> %3717 to i32, !dbg !295 + %3731 = bitcast <2 x bfloat> %3721 to i32, !dbg !295 + %3732 = bitcast <2 x bfloat> %3725 to i32, !dbg !295 + %3733 = bitcast <2 x bfloat> %3729 to i32, !dbg !295 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %3606, i32 %3730, i32 %3731, i32 %3732, i32 %3733), !dbg !295 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !295 + %3734 = tail call { i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.trans.b16.p3(ptr addrspace(3) %3592), !dbg !295 + %3735 = extractvalue { i32, i32 } %3734, 0, !dbg !295 + %3736 = extractvalue { i32, i32 } %3734, 1, !dbg !295 + tail call void asm sideeffect "@$3 st.global.v2.b32 [ $2 + 0 ], { $0, $1 };", "r,r,l,b"(i32 %3735, i32 %3736, ptr addrspace(1) %3562, i1 %1084) #2, !dbg !295 + %3737 = insertelement <2 x float> poison, float %3394, i64 0, !dbg !296 + %3738 = insertelement <2 x float> %3737, float %3395, i64 1, !dbg !296 + %3739 = fsub <2 x float> zeroinitializer, %3738, !dbg !296 + %3740 = fptrunc <2 x float> %3739 to <2 x bfloat>, !dbg !297 + %3741 = insertelement <2 x float> poison, float %3396, i64 0, !dbg !296 + %3742 = insertelement <2 x float> %3741, float %3397, i64 1, !dbg !296 + %3743 = fsub <2 x float> zeroinitializer, %3742, !dbg !296 + %3744 = fptrunc <2 x float> %3743 to <2 x bfloat>, !dbg !297 + %3745 = insertelement <2 x float> poison, float %3399, i64 0, !dbg !296 + %3746 = insertelement <2 x float> %3745, float %3400, i64 1, !dbg !296 + %3747 = fsub <2 x float> zeroinitializer, %3746, !dbg !296 + %3748 = fptrunc <2 x float> %3747 to <2 x bfloat>, !dbg !297 + %3749 = insertelement <2 x float> poison, float %3401, i64 0, !dbg !296 + %3750 = insertelement <2 x float> %3749, float %3402, i64 1, !dbg !296 + %3751 = fsub <2 x float> zeroinitializer, %3750, !dbg !296 + %3752 = fptrunc <2 x float> %3751 to <2 x bfloat>, !dbg !297 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !298 + %3753 = bitcast <2 x bfloat> %3740 to i32, !dbg !298 + %3754 = bitcast <2 x bfloat> %3744 to i32, !dbg !298 + %3755 = bitcast <2 x bfloat> %3748 to i32, !dbg !298 + %3756 = bitcast <2 x bfloat> %3752 to i32, !dbg !298 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %3606, i32 %3753, i32 %3754, i32 %3755, i32 %3756), !dbg !298 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !298 + %3757 = tail call { i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.trans.b16.p3(ptr addrspace(3) %3592), !dbg !298 + %3758 = extractvalue { i32, i32 } %3757, 0, !dbg !298 + %3759 = extractvalue { i32, i32 } %3757, 1, !dbg !298 + tail call void asm sideeffect "@$3 st.global.v2.b32 [ $2 + 0 ], { $0, $1 };", "r,r,l,b"(i32 %3758, i32 %3759, ptr addrspace(1) %3563, i1 %1084) #2, !dbg !298 + %3760 = insertelement <2 x float> poison, float %3017, i64 0, !dbg !299 + %3761 = insertelement <2 x float> %3760, float %3018, i64 1, !dbg !299 + %3762 = fsub <2 x float> zeroinitializer, %3761, !dbg !299 + %3763 = fptrunc <2 x float> %3762 to <2 x bfloat>, !dbg !300 + %3764 = insertelement <2 x float> poison, float %3019, i64 0, !dbg !299 + %3765 = insertelement <2 x float> %3764, float %3020, i64 1, !dbg !299 + %3766 = fsub <2 x float> zeroinitializer, %3765, !dbg !299 + %3767 = fptrunc <2 x float> %3766 to <2 x bfloat>, !dbg !300 + %3768 = insertelement <2 x float> poison, float %3022, i64 0, !dbg !299 + %3769 = insertelement <2 x float> %3768, float %3023, i64 1, !dbg !299 + %3770 = fsub <2 x float> zeroinitializer, %3769, !dbg !299 + %3771 = fptrunc <2 x float> %3770 to <2 x bfloat>, !dbg !300 + %3772 = insertelement <2 x float> poison, float %3024, i64 0, !dbg !299 + %3773 = insertelement <2 x float> %3772, float %3025, i64 1, !dbg !299 + %3774 = fsub <2 x float> zeroinitializer, %3773, !dbg !299 + %3775 = fptrunc <2 x float> %3774 to <2 x bfloat>, !dbg !300 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !301 + %3776 = bitcast <2 x bfloat> %3763 to i32, !dbg !301 + %3777 = bitcast <2 x bfloat> %3767 to i32, !dbg !301 + %3778 = bitcast <2 x bfloat> %3771 to i32, !dbg !301 + %3779 = bitcast <2 x bfloat> %3775 to i32, !dbg !301 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %3606, i32 %3776, i32 %3777, i32 %3778, i32 %3779), !dbg !301 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !301 + %3780 = tail call { i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.trans.b16.p3(ptr addrspace(3) %3592), !dbg !301 + %3781 = extractvalue { i32, i32 } %3780, 0, !dbg !301 + %3782 = extractvalue { i32, i32 } %3780, 1, !dbg !301 + tail call void asm sideeffect "@$3 st.global.v2.b32 [ $2 + 0 ], { $0, $1 };", "r,r,l,b"(i32 %3781, i32 %3782, ptr addrspace(1) %3564, i1 %1084) #2, !dbg !301 + %3783 = fptrunc float %2599 to bfloat, !dbg !302 + %3784 = fptrunc float %2601 to bfloat, !dbg !302 + %3785 = fptrunc float %.pn117.lcssa to bfloat, !dbg !302 + %3786 = fptrunc float %.pn115.lcssa to bfloat, !dbg !302 + %3787 = fptrunc float %.pn113.lcssa to bfloat, !dbg !302 + %3788 = fptrunc float %.pn111.lcssa to bfloat, !dbg !302 + %3789 = fptrunc float %2603 to bfloat, !dbg !302 + %3790 = fptrunc float %2605 to bfloat, !dbg !302 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !303 + %3791 = insertelement <2 x bfloat> poison, bfloat %3783, i64 0, !dbg !303 + %3792 = insertelement <2 x bfloat> %3791, bfloat %3784, i64 1, !dbg !303 + %3793 = bitcast <2 x bfloat> %3792 to i32, !dbg !303 + %3794 = insertelement <2 x bfloat> poison, bfloat %3785, i64 0, !dbg !303 + %3795 = insertelement <2 x bfloat> %3794, bfloat %3786, i64 1, !dbg !303 + %3796 = bitcast <2 x bfloat> %3795 to i32, !dbg !303 + %3797 = insertelement <2 x bfloat> poison, bfloat %3787, i64 0, !dbg !303 + %3798 = insertelement <2 x bfloat> %3797, bfloat %3788, i64 1, !dbg !303 + %3799 = bitcast <2 x bfloat> %3798 to i32, !dbg !303 + %3800 = insertelement <2 x bfloat> poison, bfloat %3789, i64 0, !dbg !303 + %3801 = insertelement <2 x bfloat> %3800, bfloat %3790, i64 1, !dbg !303 + %3802 = bitcast <2 x bfloat> %3801 to i32, !dbg !303 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %3606, i32 %3793, i32 %3796, i32 %3799, i32 %3802), !dbg !303 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !303 + %3803 = tail call { i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.trans.b16.p3(ptr addrspace(3) %3592), !dbg !303 + %3804 = extractvalue { i32, i32 } %3803, 0, !dbg !303 + %3805 = extractvalue { i32, i32 } %3803, 1, !dbg !303 + tail call void asm sideeffect "@$3 st.global.v2.b32 [ $2 + 0 ], { $0, $1 };", "r,r,l,b"(i32 %3804, i32 %3805, ptr addrspace(1) %3565, i1 %1084) #2, !dbg !303 + br label %common.ret, !dbg !304 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 65535) i32 @llvm.nvvm.read.ptx.sreg.ctaid.y() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: nounwind +declare void @llvm.nvvm.cp.async.commit.group() #2 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #3 + +; Function Attrs: nounwind +declare void @llvm.nvvm.cp.async.wait.group(i32 immarg) #2 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.idx.i32(i32, i32, i32, i32) #4 + +; Function Attrs: convergent nocallback nounwind memory(argmem: write) +declare void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x2.trans.b16.p3(ptr addrspace(3) writeonly captures(none), i32, i32) #5 + +; Function Attrs: convergent nocallback nofree nounwind memory(argmem: read) +declare { i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.trans.b16.p3(ptr addrspace(3) readonly captures(none)) #6 + +; Function Attrs: convergent nocallback nounwind memory(argmem: write) +declare void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) writeonly captures(none), i32, i32, i32, i32) #5 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #4 + +; Function Attrs: convergent nocallback nofree nounwind memory(argmem: read) +declare { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) readonly captures(none)) #6 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #7 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.ex2.approx.ftz.f32(float) #8 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.ex2.approx.f32(float) #8 + +; Function Attrs: nocallback nocreateundeforpoison nofree nosync nounwind speculatable willreturn memory(none) +declare i2 @llvm.bitreverse.i2(i2) #9 + +; Function Attrs: nocallback nocreateundeforpoison nofree nosync nounwind speculatable willreturn memory(none) +declare i64 @llvm.umin.i64(i64, i64) #9 + +attributes #0 = { nounwind "nvvm.reqntid"="64" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { nounwind } +attributes #3 = { convergent nocallback nounwind } +attributes #4 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #5 = { convergent nocallback nounwind memory(argmem: write) } +attributes #6 = { convergent nocallback nofree nounwind memory(argmem: read) } +attributes #7 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #8 = { mustprogress nocallback nofree nosync nounwind willreturn memory(none) } +attributes #9 = { nocallback nocreateundeforpoison nofree nosync nounwind speculatable willreturn memory(none) } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "chunk_fwd.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "chunk_gated_delta_rule_fwd_kkt_solve_kernel", linkageName: "chunk_gated_delta_rule_fwd_kkt_solve_kernel", scope: !1, file: !1, line: 40, type: !6, scopeLine: 40, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !10, !8, !8, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 70, column: 30, scope: !5) +!14 = !DILocation(line: 70, column: 36, scope: !5) +!15 = !DILocation(line: 78, column: 25, scope: !5) +!16 = !DILocation(line: 80, column: 13, scope: !5) +!17 = !DILocation(line: 80, column: 19, scope: !5) +!18 = !DILocation(line: 0, scope: !5) +!19 = !DILocation(line: 70, column: 61, scope: !5) +!20 = !DILocation(line: 70, column: 67, scope: !5) +!21 = !DILocation(line: 71, column: 34, scope: !5) +!22 = !DILocation(line: 84, column: 23, scope: !5) +!23 = !DILocation(line: 85, column: 23, scope: !5) +!24 = !DILocation(line: 86, column: 23, scope: !5) +!25 = !DILocation(line: 88, column: 16, scope: !5) +!26 = !DILocation(line: 88, column: 20, scope: !5) +!27 = !DILocation(line: 88, column: 9, scope: !5) +!28 = !DILocation(line: 91, column: 23, scope: !5) +!29 = !DILocation(line: 92, column: 21, scope: !5) +!30 = !DILocation(line: 92, column: 28, scope: !5) +!31 = !DILocation(line: 93, column: 21, scope: !5) +!32 = !DILocation(line: 93, column: 28, scope: !5) +!33 = !DILocation(line: 94, column: 21, scope: !5) +!34 = !DILocation(line: 94, column: 28, scope: !5) +!35 = !DILocation(line: 95, column: 21, scope: !5) +!36 = !DILocation(line: 95, column: 28, scope: !5) +!37 = !DILocation(line: 98, scope: !5) +!38 = !DILocation(line: 98, column: 51, scope: !5) +!39 = !DILocation(line: 98, column: 35, scope: !5) +!40 = !DILocation(line: 99, column: 51, scope: !5) +!41 = !DILocation(line: 99, column: 35, scope: !5) +!42 = !DILocation(line: 100, column: 51, scope: !5) +!43 = !DILocation(line: 100, column: 35, scope: !5) +!44 = !DILocation(line: 101, column: 51, scope: !5) +!45 = !DILocation(line: 101, column: 35, scope: !5) +!46 = !DILocation(line: 114, column: 23, scope: !5) +!47 = !DILocation(line: 102, column: 19, scope: !5) +!48 = !DILocation(line: 103, column: 19, scope: !5) +!49 = !DILocation(line: 104, column: 19, scope: !5) +!50 = !DILocation(line: 105, column: 19, scope: !5) +!51 = !DILocation(line: 109, scope: !5) +!52 = !DILocation(line: 109, column: 36, scope: !5) +!53 = !DILocation(line: 110, column: 36, scope: !5) +!54 = !DILocation(line: 111, column: 36, scope: !5) +!55 = !DILocation(line: 112, column: 36, scope: !5) +!56 = !DILocation(line: 115, column: 23, scope: !5) +!57 = !DILocation(line: 116, column: 23, scope: !5) +!58 = !DILocation(line: 117, column: 23, scope: !5) +!59 = !DILocation(line: 138, column: 38, scope: !5) +!60 = !DILocation(line: 139, column: 19, scope: !5) +!61 = !DILocation(line: 144, column: 19, scope: !5) +!62 = !DILocation(line: 139, column: 52, scope: !5) +!63 = !DILocation(line: 140, column: 23, scope: !5) +!64 = !DILocation(line: 138, column: 25, scope: !5) +!65 = !DILocation(line: 137, column: 21, scope: !5) +!66 = !DILocation(line: 142, column: 45, scope: !5) +!67 = !DILocation(line: 144, column: 11, scope: !5) +!68 = !DILocation(line: 145, column: 56, scope: !5) +!69 = !DILocation(line: 146, column: 27, scope: !5) +!70 = !DILocation(line: 148, column: 42, scope: !5) +!71 = !DILocation(line: 148, column: 49, scope: !5) +!72 = !DILocation(line: 150, column: 49, scope: !5) +!73 = !DILocation(line: 152, column: 15, scope: !5) +!74 = !DILocation(line: 153, column: 60, scope: !5) +!75 = !DILocation(line: 154, column: 31, scope: !5) +!76 = !DILocation(line: 156, column: 46, scope: !5) +!77 = !DILocation(line: 156, column: 53, scope: !5) +!78 = !DILocation(line: 158, column: 53, scope: !5) +!79 = !DILocation(line: 159, column: 53, scope: !5) +!80 = !DILocation(line: 161, column: 19, scope: !5) +!81 = !DILocation(line: 162, column: 64, scope: !5) +!82 = !DILocation(line: 163, column: 35, scope: !5) +!83 = !DILocation(line: 165, column: 50, scope: !5) +!84 = !DILocation(line: 165, column: 57, scope: !5) +!85 = !DILocation(line: 167, column: 57, scope: !5) +!86 = !DILocation(line: 168, column: 57, scope: !5) +!87 = !DILocation(line: 169, column: 57, scope: !5) +!88 = !DILocation(line: 89, column: 9, scope: !5) +!89 = !DILocation(line: 102, column: 51, scope: !5) +!90 = !DILocation(line: 103, column: 51, scope: !5) +!91 = !DILocation(line: 104, column: 51, scope: !5) +!92 = !DILocation(line: 105, column: 51, scope: !5) +!93 = !DILocation(line: 179, column: 25, scope: !5) +!94 = !DILocation(line: 180, column: 26, scope: !5) +!95 = !DILocation(line: 183, column: 32, scope: !5) +!96 = !DILocation(line: 183, column: 49, scope: !5) +!97 = !DILocation(line: 183, column: 86, scope: !5) +!98 = !DILocation(line: 47, column: 33, scope: !99, inlinedAt: !101) +!99 = distinct !DILexicalBlockFile(scope: !5, file: !100, discriminator: 0) +!100 = !DIFile(filename: "op.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils") +!101 = !DILocation(line: 183, column: 70, scope: !102) +!102 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!103 = !DILocation(line: 183, column: 102, scope: !5) +!104 = !DILocation(line: 183, column: 17, scope: !5) +!105 = !DILocation(line: 184, column: 32, scope: !5) +!106 = !DILocation(line: 184, column: 49, scope: !5) +!107 = !DILocation(line: 184, column: 86, scope: !5) +!108 = !DILocation(line: 47, column: 33, scope: !99, inlinedAt: !109) +!109 = !DILocation(line: 184, column: 70, scope: !102) +!110 = !DILocation(line: 184, column: 102, scope: !5) +!111 = !DILocation(line: 184, column: 17, scope: !5) +!112 = !DILocation(line: 185, column: 32, scope: !5) +!113 = !DILocation(line: 185, column: 49, scope: !5) +!114 = !DILocation(line: 185, column: 86, scope: !5) +!115 = !DILocation(line: 47, column: 33, scope: !99, inlinedAt: !116) +!116 = !DILocation(line: 185, column: 70, scope: !102) +!117 = !DILocation(line: 185, column: 102, scope: !5) +!118 = !DILocation(line: 185, column: 17, scope: !5) +!119 = !DILocation(line: 186, column: 32, scope: !5) +!120 = !DILocation(line: 186, column: 49, scope: !5) +!121 = !DILocation(line: 186, column: 86, scope: !5) +!122 = !DILocation(line: 47, column: 33, scope: !99, inlinedAt: !123) +!123 = !DILocation(line: 186, column: 70, scope: !102) +!124 = !DILocation(line: 186, column: 102, scope: !5) +!125 = !DILocation(line: 186, column: 17, scope: !5) +!126 = !DILocation(line: 188, column: 43, scope: !5) +!127 = !DILocation(line: 188, column: 80, scope: !5) +!128 = !DILocation(line: 47, column: 33, scope: !99, inlinedAt: !129) +!129 = !DILocation(line: 188, column: 64, scope: !102) +!130 = !DILocation(line: 188, column: 96, scope: !5) +!131 = !DILocation(line: 188, column: 17, scope: !5) +!132 = !DILocation(line: 189, column: 43, scope: !5) +!133 = !DILocation(line: 189, column: 80, scope: !5) +!134 = !DILocation(line: 47, column: 33, scope: !99, inlinedAt: !135) +!135 = !DILocation(line: 189, column: 64, scope: !102) +!136 = !DILocation(line: 189, column: 96, scope: !5) +!137 = !DILocation(line: 189, column: 17, scope: !5) +!138 = !DILocation(line: 190, column: 43, scope: !5) +!139 = !DILocation(line: 190, column: 80, scope: !5) +!140 = !DILocation(line: 47, column: 33, scope: !99, inlinedAt: !141) +!141 = !DILocation(line: 190, column: 64, scope: !102) +!142 = !DILocation(line: 190, column: 96, scope: !5) +!143 = !DILocation(line: 190, column: 17, scope: !5) +!144 = !DILocation(line: 191, column: 43, scope: !5) +!145 = !DILocation(line: 191, column: 80, scope: !5) +!146 = !DILocation(line: 47, column: 33, scope: !99, inlinedAt: !147) +!147 = !DILocation(line: 191, column: 64, scope: !102) +!148 = !DILocation(line: 191, column: 96, scope: !5) +!149 = !DILocation(line: 191, column: 17, scope: !5) +!150 = !DILocation(line: 192, column: 43, scope: !5) +!151 = !DILocation(line: 192, column: 80, scope: !5) +!152 = !DILocation(line: 47, column: 33, scope: !99, inlinedAt: !153) +!153 = !DILocation(line: 192, column: 64, scope: !102) +!154 = !DILocation(line: 192, column: 96, scope: !5) +!155 = !DILocation(line: 192, column: 17, scope: !5) +!156 = !DILocation(line: 193, column: 43, scope: !5) +!157 = !DILocation(line: 193, column: 80, scope: !5) +!158 = !DILocation(line: 47, column: 33, scope: !99, inlinedAt: !159) +!159 = !DILocation(line: 193, column: 64, scope: !102) +!160 = !DILocation(line: 193, column: 96, scope: !5) +!161 = !DILocation(line: 193, column: 17, scope: !5) +!162 = !DILocation(line: 201, column: 20, scope: !5) +!163 = !DILocation(line: 202, column: 20, scope: !5) +!164 = !DILocation(line: 207, column: 20, scope: !5) +!165 = !DILocation(line: 203, column: 20, scope: !5) +!166 = !DILocation(line: 209, column: 20, scope: !5) +!167 = !DILocation(line: 204, column: 20, scope: !5) +!168 = !DILocation(line: 212, column: 20, scope: !5) +!169 = !DILocation(line: 208, column: 20, scope: !5) +!170 = !DILocation(line: 210, column: 20, scope: !5) +!171 = !DILocation(line: 211, column: 20, scope: !5) +!172 = !DILocation(line: 222, column: 14, scope: !5) +!173 = !DILocation(line: 223, column: 14, scope: !5) +!174 = !DILocation(line: 224, column: 14, scope: !5) +!175 = !DILocation(line: 225, column: 14, scope: !5) +!176 = !DILocation(line: 227, column: 34, scope: !5) +!177 = !DILocation(line: 227, column: 22, scope: !5) +!178 = !DILocation(line: 228, column: 61, scope: !5) +!179 = !DILocation(line: 228, column: 40, scope: !5) +!180 = !DILocation(line: 263, column: 15, scope: !181, inlinedAt: !183) +!181 = distinct !DILexicalBlockFile(scope: !5, file: !182, discriminator: 0) +!182 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!183 = !DILocation(line: 293, column: 36, scope: !181, inlinedAt: !184) +!184 = !DILocation(line: 228, column: 66, scope: !102) +!185 = !DILocation(line: 229, column: 31, scope: !5) +!186 = !DILocation(line: 229, column: 41, scope: !5) +!187 = !DILocation(line: 230, column: 37, scope: !5) +!188 = !DILocation(line: 230, column: 48, scope: !5) +!189 = !DILocation(line: 263, column: 15, scope: !181, inlinedAt: !190) +!190 = !DILocation(line: 293, column: 36, scope: !181, inlinedAt: !191) +!191 = !DILocation(line: 230, column: 56, scope: !102) +!192 = !DILocation(line: 230, column: 24, scope: !5) +!193 = !DILocation(line: 231, column: 54, scope: !5) +!194 = !DILocation(line: 232, column: 34, scope: !5) +!195 = !DILocation(line: 232, column: 22, scope: !5) +!196 = !DILocation(line: 233, column: 61, scope: !5) +!197 = !DILocation(line: 233, column: 40, scope: !5) +!198 = !DILocation(line: 263, column: 15, scope: !181, inlinedAt: !199) +!199 = !DILocation(line: 293, column: 36, scope: !181, inlinedAt: !200) +!200 = !DILocation(line: 233, column: 66, scope: !102) +!201 = !DILocation(line: 234, column: 31, scope: !5) +!202 = !DILocation(line: 234, column: 41, scope: !5) +!203 = !DILocation(line: 235, column: 37, scope: !5) +!204 = !DILocation(line: 235, column: 48, scope: !5) +!205 = !DILocation(line: 263, column: 15, scope: !181, inlinedAt: !206) +!206 = !DILocation(line: 293, column: 36, scope: !181, inlinedAt: !207) +!207 = !DILocation(line: 235, column: 56, scope: !102) +!208 = !DILocation(line: 235, column: 24, scope: !5) +!209 = !DILocation(line: 236, column: 54, scope: !5) +!210 = !DILocation(line: 249, column: 14, scope: !5) +!211 = !DILocation(line: 237, column: 34, scope: !5) +!212 = !DILocation(line: 237, column: 22, scope: !5) +!213 = !DILocation(line: 238, column: 61, scope: !5) +!214 = !DILocation(line: 238, column: 40, scope: !5) +!215 = !DILocation(line: 263, column: 15, scope: !181, inlinedAt: !216) +!216 = !DILocation(line: 293, column: 36, scope: !181, inlinedAt: !217) +!217 = !DILocation(line: 238, column: 66, scope: !102) +!218 = !DILocation(line: 239, column: 31, scope: !5) +!219 = !DILocation(line: 239, column: 41, scope: !5) +!220 = !DILocation(line: 240, column: 37, scope: !5) +!221 = !DILocation(line: 240, column: 48, scope: !5) +!222 = !DILocation(line: 263, column: 15, scope: !181, inlinedAt: !223) +!223 = !DILocation(line: 293, column: 36, scope: !181, inlinedAt: !224) +!224 = !DILocation(line: 240, column: 56, scope: !102) +!225 = !DILocation(line: 240, column: 24, scope: !5) +!226 = !DILocation(line: 241, column: 54, scope: !5) +!227 = !DILocation(line: 250, column: 14, scope: !5) +!228 = !DILocation(line: 242, column: 34, scope: !5) +!229 = !DILocation(line: 242, column: 22, scope: !5) +!230 = !DILocation(line: 243, column: 61, scope: !5) +!231 = !DILocation(line: 243, column: 40, scope: !5) +!232 = !DILocation(line: 263, column: 15, scope: !181, inlinedAt: !233) +!233 = !DILocation(line: 293, column: 36, scope: !181, inlinedAt: !234) +!234 = !DILocation(line: 243, column: 66, scope: !102) +!235 = !DILocation(line: 244, column: 31, scope: !5) +!236 = !DILocation(line: 244, column: 41, scope: !5) +!237 = !DILocation(line: 245, column: 37, scope: !5) +!238 = !DILocation(line: 245, column: 48, scope: !5) +!239 = !DILocation(line: 263, column: 15, scope: !181, inlinedAt: !240) +!240 = !DILocation(line: 293, column: 36, scope: !181, inlinedAt: !241) +!241 = !DILocation(line: 245, column: 56, scope: !102) +!242 = !DILocation(line: 245, column: 24, scope: !5) +!243 = !DILocation(line: 246, column: 54, scope: !5) +!244 = !DILocation(line: 251, column: 14, scope: !5) +!245 = !DILocation(line: 248, column: 14, scope: !5) +!246 = !DILocation(line: 258, column: 23, scope: !5) +!247 = !DILocation(line: 259, column: 8, scope: !5) +!248 = !DILocation(line: 257, column: 14, scope: !5) +!249 = !DILocation(line: 263, column: 23, scope: !5) +!250 = !DILocation(line: 264, column: 8, scope: !5) +!251 = !DILocation(line: 262, column: 14, scope: !5) +!252 = !DILocation(line: 268, column: 23, scope: !5) +!253 = !DILocation(line: 269, column: 8, scope: !5) +!254 = !DILocation(line: 275, column: 22, scope: !5) +!255 = !DILocation(line: 276, column: 22, scope: !5) +!256 = !DILocation(line: 275, column: 8, scope: !5) +!257 = !DILocation(line: 273, column: 14, scope: !5) +!258 = !DILocation(line: 281, column: 22, scope: !5) +!259 = !DILocation(line: 282, column: 22, scope: !5) +!260 = !DILocation(line: 281, column: 8, scope: !5) +!261 = !DILocation(line: 287, column: 22, scope: !5) +!262 = !DILocation(line: 288, column: 22, scope: !5) +!263 = !DILocation(line: 289, column: 22, scope: !5) +!264 = !DILocation(line: 287, column: 8, scope: !5) +!265 = !DILocation(line: 297, column: 16, scope: !5) +!266 = !DILocation(line: 297, column: 51, scope: !5) +!267 = !DILocation(line: 298, column: 16, scope: !5) +!268 = !DILocation(line: 298, column: 51, scope: !5) +!269 = !DILocation(line: 299, column: 57, scope: !5) +!270 = !DILocation(line: 299, column: 51, scope: !5) +!271 = !DILocation(line: 300, column: 16, scope: !5) +!272 = !DILocation(line: 300, column: 51, scope: !5) +!273 = !DILocation(line: 301, column: 51, scope: !5) +!274 = !DILocation(line: 302, column: 59, scope: !5) +!275 = !DILocation(line: 302, column: 51, scope: !5) +!276 = !DILocation(line: 303, column: 16, scope: !5) +!277 = !DILocation(line: 303, column: 51, scope: !5) +!278 = !DILocation(line: 304, column: 51, scope: !5) +!279 = !DILocation(line: 305, column: 51, scope: !5) +!280 = !DILocation(line: 306, column: 51, scope: !5) +!281 = !DILocation(line: 319, column: 30, scope: !5) +!282 = !DILocation(line: 319, column: 20, scope: !5) +!283 = !DILocation(line: 320, column: 30, scope: !5) +!284 = !DILocation(line: 320, column: 20, scope: !5) +!285 = !DILocation(line: 321, column: 30, scope: !5) +!286 = !DILocation(line: 321, column: 20, scope: !5) +!287 = !DILocation(line: 322, column: 30, scope: !5) +!288 = !DILocation(line: 322, column: 20, scope: !5) +!289 = !DILocation(line: 323, column: 30, scope: !5) +!290 = !DILocation(line: 323, column: 20, scope: !5) +!291 = !DILocation(line: 324, column: 30, scope: !5) +!292 = !DILocation(line: 324, column: 20, scope: !5) +!293 = !DILocation(line: 285, column: 14, scope: !5) +!294 = !DILocation(line: 325, column: 30, scope: !5) +!295 = !DILocation(line: 325, column: 20, scope: !5) +!296 = !DILocation(line: 279, column: 14, scope: !5) +!297 = !DILocation(line: 326, column: 30, scope: !5) +!298 = !DILocation(line: 326, column: 20, scope: !5) +!299 = !DILocation(line: 267, column: 14, scope: !5) +!300 = !DILocation(line: 327, column: 30, scope: !5) +!301 = !DILocation(line: 327, column: 20, scope: !5) +!302 = !DILocation(line: 328, column: 30, scope: !5) +!303 = !DILocation(line: 328, column: 20, scope: !5) +!304 = !DILocation(line: 328, column: 4, scope: !5) diff --git a/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.ptx b/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..a1e0d974c22f8eb8ed0eb72146595fec1c10c9fc --- /dev/null +++ b/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.ptx @@ -0,0 +1,4100 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl chunk_gated_delta_rule_fwd_kkt_solve_kernel // -- Begin function chunk_gated_delta_rule_fwd_kkt_solve_kernel +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; + // @chunk_gated_delta_rule_fwd_kkt_solve_kernel +.visible .entry chunk_gated_delta_rule_fwd_kkt_solve_kernel( + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_fwd_kkt_solve_kernel_param_0, + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_fwd_kkt_solve_kernel_param_1, + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_fwd_kkt_solve_kernel_param_2, + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_fwd_kkt_solve_kernel_param_3, + .param .u32 chunk_gated_delta_rule_fwd_kkt_solve_kernel_param_4, + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_fwd_kkt_solve_kernel_param_5, + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_fwd_kkt_solve_kernel_param_6 +) +.reqntid 64 +{ + .reg .pred %p<97>; + .reg .b16 %rs<32>; + .reg .b32 %r<1937>; + .reg .b64 %rd<338>; + .loc 1 40 0 // chunk_fwd.py:40:0 +$L__func_begin0: + .loc 1 40 0 // chunk_fwd.py:40:0 + +// %bb.0: + ld.param.b32 %r62, [chunk_gated_delta_rule_fwd_kkt_solve_kernel_param_4]; +$L__tmp0: + .loc 1 70 30 // chunk_fwd.py:70:30 + mov.u32 %r63, %ctaid.x; + .loc 1 78 25 // chunk_fwd.py:78:25 + cvt.s64.s32 %rd2, %r62; + .loc 1 80 13 // chunk_fwd.py:80:13 + mul.wide.u32 %rd3, %r63, 64; + .loc 1 80 19 // chunk_fwd.py:80:19 + setp.gt.s64 %p2, %rd2, %rd3; + @%p2 bra $L__BB0_2; + bra.uni $L__BB0_1; +$L__BB0_2: + .loc 1 0 19 // chunk_fwd.py:0:19 + ld.param.b64 %rd66, [chunk_gated_delta_rule_fwd_kkt_solve_kernel_param_3]; + ld.param.b64 %rd65, [chunk_gated_delta_rule_fwd_kkt_solve_kernel_param_2]; + ld.param.b64 %rd64, [chunk_gated_delta_rule_fwd_kkt_solve_kernel_param_1]; + ld.param.b64 %rd63, [chunk_gated_delta_rule_fwd_kkt_solve_kernel_param_0]; + .loc 1 70 0 // chunk_fwd.py:70 + cvt.u64.u32 %rd1, %r63; + .loc 1 70 61 // chunk_fwd.py:70:61 + mov.u32 %r97, %ctaid.y; + .loc 1 70 67 // chunk_fwd.py:70:67 + cvt.u64.u32 %rd105, %r97; + .loc 1 78 25 // chunk_fwd.py:78:25 + and.b64 %rd106, %rd105, 65504; + .loc 1 71 34 // chunk_fwd.py:71:34 + and.b64 %rd107, %rd105, 31; + .loc 1 84 23 // chunk_fwd.py:84:23 + or.b64 %rd4, %rd3, 16; + .loc 1 85 23 // chunk_fwd.py:85:23 + or.b64 %rd5, %rd3, 32; + .loc 1 86 23 // chunk_fwd.py:86:23 + or.b64 %rd6, %rd3, 48; + .loc 1 88 16 // chunk_fwd.py:88:16 + mul.lo.s64 %rd108, %rd106, %rd2; + .loc 1 88 20 // chunk_fwd.py:88:20 + or.b64 %rd7, %rd108, %rd107; + .loc 1 88 9 // chunk_fwd.py:88:9 + shl.b64 %rd109, %rd7, 8; + add.s64 %rd110, %rd63, %rd109; + .loc 1 91 23 // chunk_fwd.py:91:23 + mov.u32 %r1, %tid.x; + and.b32 %r98, %r1, 56; + bfe.u32 %r99, %r1, 3, 3; + or.b32 %r100, %r99, 8; + and.b32 %r2, %r1, 28; + and.b32 %r3, %r1, 3; + and.b32 %r4, %r1, 32; + shr.u32 %r101, %r4, 2; + .loc 1 92 21 // chunk_fwd.py:92:21 + cvt.u64.u32 %rd111, %r99; + cvt.u64.u32 %rd112, %r100; + or.b64 %rd113, %rd3, %rd111; + or.b64 %rd114, %rd3, %rd112; + .loc 1 92 28 // chunk_fwd.py:92:28 + setp.lt.u64 %p29, %rd113, %rd2; + setp.lt.u64 %p30, %rd114, %rd2; + .loc 1 93 21 // chunk_fwd.py:93:21 + or.b64 %rd8, %rd4, %rd111; + or.b64 %rd9, %rd4, %rd112; + .loc 1 94 21 // chunk_fwd.py:94:21 + or.b64 %rd10, %rd5, %rd111; + or.b64 %rd11, %rd5, %rd112; + .loc 1 95 21 // chunk_fwd.py:95:21 + or.b64 %rd12, %rd6, %rd111; + or.b64 %rd13, %rd6, %rd112; + .loc 1 91 23 // chunk_fwd.py:91:23 + shl.b32 %r5, %r3, 1; + or.b32 %r6, %r5, %r101; + or.b32 %r7, %r6, 1; + or.b32 %r8, %r5, 1; + or.b32 %r102, %r5, 8; + or.b32 %r9, %r5, 9; + .loc 1 92 21 // chunk_fwd.py:92:21 + mov.b64 %rd14, {%r5, %r8}; + .loc 1 94 21 // chunk_fwd.py:94:21 + cvt.u64.u32 %rd18, %r9; + cvt.u64.u32 %rd17, %r102; + cvt.u64.u32 %rd16, %r8; + cvt.u64.u32 %rd15, %r5; + .loc 1 92 21 // chunk_fwd.py:92:21 + or.b64 %rd19, %rd3, %rd15; + or.b64 %rd20, %rd3, %rd16; + or.b64 %rd21, %rd3, %rd17; + or.b64 %rd22, %rd3, %rd18; + .loc 1 92 28 // chunk_fwd.py:92:28 + setp.lt.u64 %p13, %rd19, %rd2; + setp.lt.u64 %p14, %rd20, %rd2; + setp.lt.u64 %p15, %rd21, %rd2; + setp.lt.u64 %p16, %rd22, %rd2; + .loc 1 92 21 // chunk_fwd.py:92:21 + mov.b64 %rd23, {%r6, %r7}; + cvt.u64.u32 %rd25, %r7; + cvt.u64.u32 %rd24, %r6; + or.b64 %rd115, %rd3, %rd24; + or.b64 %rd116, %rd3, %rd25; + .loc 1 92 28 // chunk_fwd.py:92:28 + setp.lt.u64 %p12, %rd116, %rd2; + setp.lt.u64 %p11, %rd115, %rd2; + .loc 1 91 23 // chunk_fwd.py:91:23 + bfe.u32 %r10, %r1, 2, 3; + or.b32 %r11, %r10, 8; + .loc 1 92 21 // chunk_fwd.py:92:21 + mov.b64 %rd26, {%r10, %r11}; + cvt.u64.u32 %rd27, %r11; + cvt.u64.u32 %rd29, %r10; + or.b64 %rd117, %rd3, %rd29; + or.b64 %rd118, %rd3, %rd27; + .loc 1 92 28 // chunk_fwd.py:92:28 + setp.lt.u64 %p10, %rd118, %rd2; + setp.lt.u64 %p9, %rd117, %rd2; + .loc 1 93 21 // chunk_fwd.py:93:21 + or.b64 %rd28, %rd4, %rd27; + or.b64 %rd30, %rd4, %rd29; + or.b64 %rd31, %rd4, %rd15; + or.b64 %rd32, %rd4, %rd16; + or.b64 %rd33, %rd4, %rd17; + or.b64 %rd34, %rd4, %rd18; + .loc 1 93 28 // chunk_fwd.py:93:28 + setp.lt.u64 %p4, %rd28, %rd2; + setp.lt.u64 %p3, %rd30, %rd2; + setp.lt.u64 %p20, %rd34, %rd2; + setp.lt.u64 %p19, %rd33, %rd2; + setp.lt.u64 %p18, %rd32, %rd2; + setp.lt.u64 %p17, %rd31, %rd2; + .loc 1 94 21 // chunk_fwd.py:94:21 + or.b64 %rd35, %rd5, %rd27; + or.b64 %rd36, %rd5, %rd29; + or.b64 %rd37, %rd5, %rd15; + or.b64 %rd38, %rd5, %rd16; + or.b64 %rd39, %rd5, %rd17; + or.b64 %rd40, %rd5, %rd18; + .loc 1 94 28 // chunk_fwd.py:94:28 + setp.lt.u64 %p6, %rd35, %rd2; + setp.lt.u64 %p5, %rd36, %rd2; + setp.lt.u64 %p24, %rd40, %rd2; + setp.lt.u64 %p23, %rd39, %rd2; + setp.lt.u64 %p22, %rd38, %rd2; + setp.lt.u64 %p21, %rd37, %rd2; + .loc 1 95 21 // chunk_fwd.py:95:21 + or.b64 %rd41, %rd6, %rd27; + or.b64 %rd42, %rd6, %rd29; + or.b64 %rd43, %rd6, %rd15; + or.b64 %rd44, %rd6, %rd16; + or.b64 %rd45, %rd6, %rd17; + or.b64 %rd46, %rd6, %rd18; + .loc 1 95 28 // chunk_fwd.py:95:28 + setp.lt.u64 %p8, %rd41, %rd2; + setp.lt.u64 %p7, %rd42, %rd2; + setp.lt.u64 %p28, %rd46, %rd2; + setp.lt.u64 %p27, %rd45, %rd2; + setp.lt.u64 %p26, %rd44, %rd2; + setp.lt.u64 %p25, %rd43, %rd2; + .loc 1 98 0 // chunk_fwd.py:98 + shl.b64 %rd119, %rd7, 1; + add.s64 %rd120, %rd65, %rd119; + .loc 1 98 35 // chunk_fwd.py:98:35 + shl.b64 %rd121, %rd117, 6; + add.s64 %rd67, %rd120, %rd121; + shl.b64 %rd122, %rd118, 6; + add.s64 %rd68, %rd120, %rd122; + .loc 1 99 35 // chunk_fwd.py:99:35 + shl.b64 %rd123, %rd30, 6; + add.s64 %rd69, %rd120, %rd123; + shl.b64 %rd124, %rd28, 6; + add.s64 %rd70, %rd120, %rd124; + .loc 1 100 35 // chunk_fwd.py:100:35 + shl.b64 %rd125, %rd36, 6; + add.s64 %rd71, %rd120, %rd125; + shl.b64 %rd126, %rd35, 6; + add.s64 %rd72, %rd120, %rd126; + .loc 1 101 35 // chunk_fwd.py:101:35 + shl.b64 %rd127, %rd42, 6; + add.s64 %rd73, %rd120, %rd127; + shl.b64 %rd128, %rd41, 6; + add.s64 %rd74, %rd120, %rd128; + mov.b16 %rs2, 0; + .loc 1 102 19 // chunk_fwd.py:102:19 + // begin inline asm + mov.u16 %rs1, %rs2; + @%p9 ld.global.b16 { %rs1 }, [ %rd67 + 0 ]; + // end inline asm + // begin inline asm + mov.u16 %rs3, %rs2; + @%p10 ld.global.b16 { %rs3 }, [ %rd68 + 0 ]; + // end inline asm + .loc 1 103 19 // chunk_fwd.py:103:19 + // begin inline asm + mov.u16 %rs4, %rs2; + @%p3 ld.global.b16 { %rs4 }, [ %rd69 + 0 ]; + // end inline asm + // begin inline asm + mov.u16 %rs5, %rs2; + @%p4 ld.global.b16 { %rs5 }, [ %rd70 + 0 ]; + // end inline asm + .loc 1 104 19 // chunk_fwd.py:104:19 + // begin inline asm + mov.u16 %rs6, %rs2; + @%p5 ld.global.b16 { %rs6 }, [ %rd71 + 0 ]; + // end inline asm + // begin inline asm + mov.u16 %rs7, %rs2; + @%p6 ld.global.b16 { %rs7 }, [ %rd72 + 0 ]; + // end inline asm + .loc 1 105 19 // chunk_fwd.py:105:19 + // begin inline asm + mov.u16 %rs8, %rs2; + @%p7 ld.global.b16 { %rs8 }, [ %rd73 + 0 ]; + // end inline asm + // begin inline asm + mov.u16 %rs9, %rs2; + @%p8 ld.global.b16 { %rs9 }, [ %rd74 + 0 ]; + // end inline asm + .loc 1 109 0 // chunk_fwd.py:109 + shl.b64 %rd129, %rd7, 2; + add.s64 %rd130, %rd64, %rd129; + .loc 1 109 36 // chunk_fwd.py:109:36 + shl.b64 %rd131, %rd117, 7; + add.s64 %rd75, %rd130, %rd131; + shl.b64 %rd132, %rd118, 7; + add.s64 %rd76, %rd130, %rd132; + shl.b64 %rd133, %rd115, 7; + add.s64 %rd77, %rd130, %rd133; + shl.b64 %rd134, %rd116, 7; + add.s64 %rd78, %rd130, %rd134; + shl.b64 %rd135, %rd19, 7; + add.s64 %rd79, %rd130, %rd135; + shl.b64 %rd136, %rd20, 7; + add.s64 %rd80, %rd130, %rd136; + shl.b64 %rd137, %rd21, 7; + add.s64 %rd81, %rd130, %rd137; + shl.b64 %rd138, %rd22, 7; + add.s64 %rd82, %rd130, %rd138; + .loc 1 110 36 // chunk_fwd.py:110:36 + shl.b64 %rd139, %rd30, 7; + add.s64 %rd83, %rd130, %rd139; + shl.b64 %rd140, %rd28, 7; + add.s64 %rd84, %rd130, %rd140; + shl.b64 %rd141, %rd31, 7; + add.s64 %rd85, %rd130, %rd141; + shl.b64 %rd142, %rd32, 7; + add.s64 %rd86, %rd130, %rd142; + shl.b64 %rd143, %rd33, 7; + add.s64 %rd87, %rd130, %rd143; + shl.b64 %rd144, %rd34, 7; + add.s64 %rd88, %rd130, %rd144; + .loc 1 111 36 // chunk_fwd.py:111:36 + shl.b64 %rd145, %rd36, 7; + add.s64 %rd89, %rd130, %rd145; + shl.b64 %rd146, %rd35, 7; + add.s64 %rd90, %rd130, %rd146; + shl.b64 %rd147, %rd37, 7; + add.s64 %rd91, %rd130, %rd147; + shl.b64 %rd148, %rd38, 7; + add.s64 %rd92, %rd130, %rd148; + shl.b64 %rd149, %rd39, 7; + add.s64 %rd93, %rd130, %rd149; + shl.b64 %rd150, %rd40, 7; + add.s64 %rd94, %rd130, %rd150; + .loc 1 112 36 // chunk_fwd.py:112:36 + shl.b64 %rd151, %rd42, 7; + add.s64 %rd95, %rd130, %rd151; + shl.b64 %rd152, %rd41, 7; + add.s64 %rd96, %rd130, %rd152; + shl.b64 %rd153, %rd43, 7; + add.s64 %rd97, %rd130, %rd153; + shl.b64 %rd154, %rd44, 7; + add.s64 %rd98, %rd130, %rd154; + shl.b64 %rd155, %rd45, 7; + add.s64 %rd99, %rd130, %rd155; + shl.b64 %rd156, %rd46, 7; + add.s64 %rd100, %rd130, %rd156; + mov.b32 %r65, 0; + .loc 1 114 23 // chunk_fwd.py:114:23 + // begin inline asm + mov.u32 %r64, %r65; + @%p9 ld.global.b32 { %r64 }, [ %rd75 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r66, %r65; + @%p10 ld.global.b32 { %r66 }, [ %rd76 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r67, %r65; + @%p11 ld.global.b32 { %r67 }, [ %rd77 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r68, %r65; + @%p12 ld.global.b32 { %r68 }, [ %rd78 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r69, %r65; + @%p13 ld.global.b32 { %r69 }, [ %rd79 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r70, %r65; + @%p14 ld.global.b32 { %r70 }, [ %rd80 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r71, %r65; + @%p15 ld.global.b32 { %r71 }, [ %rd81 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r72, %r65; + @%p16 ld.global.b32 { %r72 }, [ %rd82 + 0 ]; + // end inline asm + .loc 1 115 23 // chunk_fwd.py:115:23 + // begin inline asm + mov.u32 %r73, %r65; + @%p3 ld.global.b32 { %r73 }, [ %rd83 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r74, %r65; + @%p4 ld.global.b32 { %r74 }, [ %rd84 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r75, %r65; + @%p17 ld.global.b32 { %r75 }, [ %rd85 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r76, %r65; + @%p18 ld.global.b32 { %r76 }, [ %rd86 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r77, %r65; + @%p19 ld.global.b32 { %r77 }, [ %rd87 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r78, %r65; + @%p20 ld.global.b32 { %r78 }, [ %rd88 + 0 ]; + // end inline asm + .loc 1 116 23 // chunk_fwd.py:116:23 + // begin inline asm + mov.u32 %r79, %r65; + @%p5 ld.global.b32 { %r79 }, [ %rd89 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r80, %r65; + @%p6 ld.global.b32 { %r80 }, [ %rd90 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r81, %r65; + @%p21 ld.global.b32 { %r81 }, [ %rd91 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r82, %r65; + @%p22 ld.global.b32 { %r82 }, [ %rd92 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r83, %r65; + @%p23 ld.global.b32 { %r83 }, [ %rd93 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r84, %r65; + @%p24 ld.global.b32 { %r84 }, [ %rd94 + 0 ]; + // end inline asm + .loc 1 117 23 // chunk_fwd.py:117:23 + // begin inline asm + mov.u32 %r85, %r65; + @%p7 ld.global.b32 { %r85 }, [ %rd95 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r86, %r65; + @%p8 ld.global.b32 { %r86 }, [ %rd96 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r87, %r65; + @%p25 ld.global.b32 { %r87 }, [ %rd97 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r88, %r65; + @%p26 ld.global.b32 { %r88 }, [ %rd98 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r89, %r65; + @%p27 ld.global.b32 { %r89 }, [ %rd99 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r90, %r65; + @%p28 ld.global.b32 { %r90 }, [ %rd100 + 0 ]; + // end inline asm + .loc 1 138 38 // chunk_fwd.py:138:38 + and.b32 %r103, %r1, 7; + shl.b32 %r104, %r103, 3; + .loc 1 139 19 // chunk_fwd.py:139:19 + shl.b64 %rd157, %rd113, 13; + add.s64 %rd47, %rd110, %rd157; + shl.b64 %rd158, %rd114, 13; + add.s64 %rd48, %rd110, %rd158; + .loc 1 139 52 // chunk_fwd.py:139:52 + cvt.u64.u32 %rd49, %r104; + mul.wide.u32 %rd159, %r104, 2; + add.s64 %rd101, %rd47, %rd159; + add.s64 %rd102, %rd48, %rd159; + .loc 1 140 23 // chunk_fwd.py:140:23 + shl.b32 %r12, %r1, 4; + and.b32 %r105, %r12, 1008; + shl.b32 %r106, %r98, 1; + xor.b32 %r107, %r105, %r106; + mov.b32 %r108, global_smem; + add.s32 %r91, %r108, %r107; + selp.b32 %r92, 16, 0, %p29; + // begin inline asm + cp.async.cg.shared.global [ %r91 + 0 ], [ %rd101 + 0 ], 0x10, %r92; + // end inline asm + add.s32 %r93, %r91, 1024; + selp.b32 %r94, 16, 0, %p30; + // begin inline asm + cp.async.cg.shared.global [ %r93 + 0 ], [ %rd102 + 0 ], 0x10, %r94; + // end inline asm + cp.async.commit_group; + .loc 1 139 52 // chunk_fwd.py:139:52 + add.s64 %rd103, %rd101, 128; + add.s64 %rd104, %rd102, 128; + .loc 1 140 23 // chunk_fwd.py:140:23 + bar.sync 0; + add.s32 %r95, %r91, 2048; + // begin inline asm + cp.async.cg.shared.global [ %r95 + 0 ], [ %rd103 + 0 ], 0x10, %r92; + // end inline asm + add.s32 %r96, %r91, 3072; + // begin inline asm + cp.async.cg.shared.global [ %r96 + 0 ], [ %rd104 + 0 ], 0x10, %r94; + // end inline asm + cp.async.commit_group; + shl.b32 %r109, %r103, 7; + shl.b32 %r110, %r103, 4; + shl.b32 %r111, %r1, 1; + and.b32 %r112, %r111, 48; + or.b32 %r113, %r109, %r110; + xor.b32 %r13, %r113, %r112; + xor.b32 %r14, %r13, 64; + shl.b32 %r114, %r4, 5; + or.b32 %r115, %r109, %r114; + xor.b32 %r116, %r110, %r112; + or.b32 %r15, %r115, %r116; + xor.b32 %r16, %r15, 64; + shl.b32 %r117, %r1, 7; + and.b32 %r118, %r117, 1920; + and.b32 %r17, %r1, 16; + xor.b32 %r119, %r110, %r17; + or.b32 %r18, %r119, %r118; + xor.b32 %r120, %r18, 32; + xor.b32 %r121, %r18, 64; + xor.b32 %r122, %r18, 96; + add.s32 %r225, %r91, 4096; + add.s32 %r230, %r91, 5120; + add.s32 %r235, %r91, 6144; + or.b32 %r123, %r116, %r109; + add.s32 %r124, %r108, %r123; + add.s32 %r19, %r124, 6144; + xor.b32 %r125, %r123, 64; + add.s32 %r126, %r108, %r125; + add.s32 %r20, %r126, 6144; + add.s32 %r127, %r108, %r18; + add.s32 %r21, %r127, 4096; + add.s32 %r22, %r108, %r120; + add.s32 %r23, %r22, 4096; + add.s32 %r24, %r108, %r121; + add.s32 %r25, %r24, 4096; + add.s32 %r26, %r108, %r122; + add.s32 %r27, %r26, 4096; + .loc 1 137 21 // chunk_fwd.py:137:21 + shl.b32 %r28, %r1, 3; + shl.b64 %rd160, %rd1, 19; + shr.u64 %rd161, %rd105, 5; + mad.lo.s64 %rd162, %rd161, %rd2, %rd111; + shl.b64 %rd163, %rd162, 13; + add.s64 %rd164, %rd160, %rd163; + shl.b64 %rd165, %rd107, 8; + or.b64 %rd166, %rd164, %rd165; + mul.wide.u32 %rd167, %r103, 16; + or.b64 %rd168, %rd166, %rd167; + add.s64 %rd169, %rd168, %rd63; + add.s64 %rd318, %rd169, 458752; + mov.b32 %r1902, 0f00000000; + mov.b64 %rd325, {%r1902, %r1902}; + mov.b32 %r1845, 1; + mov.b32 %r1844, -1; + mov.pred %p96, -1; + mov.b64 %rd319, 0; + mov.b32 %r1903, %r1902; + mov.b32 %r1904, %r1902; + mov.b32 %r1905, %r1902; + mov.b32 %r1906, %r1902; + mov.b32 %r1907, %r1902; + mov.b32 %r1908, %r1902; + mov.b32 %r1909, %r1902; + mov.b32 %r1894, %r1902; + mov.b32 %r1895, %r1902; + mov.b32 %r1896, %r1902; + mov.b32 %r1897, %r1902; + mov.b32 %r1898, %r1902; + mov.b32 %r1899, %r1902; + mov.b32 %r1900, %r1902; + mov.b32 %r1901, %r1902; + mov.b32 %r1886, %r1902; + mov.b32 %r1887, %r1902; + mov.b32 %r1888, %r1902; + mov.b32 %r1889, %r1902; + mov.b32 %r1890, %r1902; + mov.b32 %r1891, %r1902; + mov.b32 %r1892, %r1902; + mov.b32 %r1893, %r1902; + mov.b32 %r1878, %r1902; + mov.b32 %r1879, %r1902; + mov.b32 %r1880, %r1902; + mov.b32 %r1881, %r1902; + mov.b32 %r1882, %r1902; + mov.b32 %r1883, %r1902; + mov.b32 %r1884, %r1902; + mov.b32 %r1885, %r1902; + mov.b32 %r1870, %r1902; + mov.b32 %r1871, %r1902; + mov.b32 %r1872, %r1902; + mov.b32 %r1873, %r1902; + mov.b32 %r1874, %r1902; + mov.b32 %r1875, %r1902; + mov.b32 %r1876, %r1902; + mov.b32 %r1877, %r1902; + mov.b32 %r1862, %r1902; + mov.b32 %r1863, %r1902; + mov.b32 %r1864, %r1902; + mov.b32 %r1865, %r1902; + mov.b32 %r1866, %r1902; + mov.b32 %r1867, %r1902; + mov.b32 %r1868, %r1902; + mov.b32 %r1869, %r1902; + mov.b32 %r1858, %r1902; + mov.b32 %r1859, %r1902; + mov.b32 %r1860, %r1902; + mov.b32 %r1861, %r1902; + mov.b32 %r1854, %r1902; + mov.b32 %r1855, %r1902; + mov.b32 %r1856, %r1902; + mov.b32 %r1857, %r1902; + mov.b32 %r1850, %r1902; + mov.b32 %r1851, %r1902; + mov.b32 %r1852, %r1902; + mov.b32 %r1853, %r1902; + mov.b32 %r1846, %r1902; + mov.b32 %r1847, %r1902; + mov.b32 %r1848, %r1902; + mov.b32 %r1849, %r1902; + mov.b64 %rd320, %rd325; + mov.b64 %rd321, %rd325; + mov.b64 %rd322, %rd325; + mov.b64 %rd323, %rd325; + mov.b64 %rd324, %rd325; + bra.uni $L__BB0_3; +$L__BB0_7: // in Loop: Header=BB0_3 Depth=1 + .loc 1 137 21 // chunk_fwd.py:137:21 + add.s64 %rd318, %rd318, 128; + mov.pred %p96, 0; + mov.b64 %rd319, 64; + @%p1 bra $L__BB0_3; + bra.uni $L__BB0_8; +$L__BB0_3: // =>This Inner Loop Header: Depth=1 + .loc 1 0 21 // chunk_fwd.py:0:21 + mov.pred %p1, %p96; + .loc 1 144 19 // chunk_fwd.py:144:19 + setp.le.u64 %p31, %rd2, %rd4; + .loc 1 137 21 // chunk_fwd.py:137:21 + add.s32 %r154, %r1844, 1; + setp.gt.s32 %p32, %r154, 1; + selp.b32 %r1844, 0, %r154, %p32; + .loc 1 140 23 // chunk_fwd.py:140:23 + cp.async.wait_group 1; + bar.sync 0; + shl.b32 %r155, %r1844, 11; + add.s32 %r156, %r108, %r155; + .loc 1 142 45 // chunk_fwd.py:142:45 + add.s32 %r157, %r156, %r13; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r272, %r273, %r276, %r277}, [%r157]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r274, %r275, %r278, %r279}, [%r157+1024]; + add.s32 %r158, %r156, %r14; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r280, %r281, %r284, %r285}, [%r158]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r282, %r283, %r286, %r287}, [%r158+1024]; + add.s32 %r159, %r156, %r15; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r132, %r133, %r138, %r139}, [%r159]; + add.s32 %r160, %r156, %r16; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r144, %r145, %r150, %r151}, [%r160]; + .loc 1 140 23 // chunk_fwd.py:140:23 + add.s32 %r161, %r156, %r18; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r128, %r129, %r130, %r131}, [%r161]; + add.s32 %r162, %r22, %r155; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r134, %r135, %r136, %r137}, [%r162]; + add.s32 %r163, %r24, %r155; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r140, %r141, %r142, %r143}, [%r163]; + add.s32 %r164, %r26, %r155; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r146, %r147, %r148, %r149}, [%r164]; + .loc 1 142 45 // chunk_fwd.py:142:45 + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1846, %r1847, %r1848, %r1849 }, { %r128, %r129, %r130, %r131 }, { %r132, %r133 }, { %r1846, %r1847, %r1848, %r1849 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1846, %r1847, %r1848, %r1849 }, { %r134, %r135, %r136, %r137 }, { %r138, %r139 }, { %r1846, %r1847, %r1848, %r1849 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1846, %r1847, %r1848, %r1849 }, { %r140, %r141, %r142, %r143 }, { %r144, %r145 }, { %r1846, %r1847, %r1848, %r1849 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1846, %r1847, %r1848, %r1849 }, { %r146, %r147, %r148, %r149 }, { %r150, %r151 }, { %r1846, %r1847, %r1848, %r1849 }; + // end inline asm + .loc 1 137 21 // chunk_fwd.py:137:21 + add.s32 %r165, %r1845, 1; + setp.gt.s32 %p33, %r165, 1; + selp.b32 %r1845, 0, %r165, %p33; + add.s64 %rd172, %rd319, %rd49; + .loc 1 139 52 // chunk_fwd.py:139:52 + shl.b64 %rd173, %rd172, 1; + add.s64 %rd174, %rd47, %rd173; + add.s64 %rd170, %rd174, 256; + add.s64 %rd175, %rd48, %rd173; + add.s64 %rd171, %rd175, 256; + .loc 1 140 23 // chunk_fwd.py:140:23 + shl.b32 %r166, %r1845, 11; + bar.sync 0; + add.s32 %r152, %r91, %r166; + // begin inline asm + cp.async.cg.shared.global [ %r152 + 0 ], [ %rd170 + 0 ], 0x10, %r65; + // end inline asm + add.s32 %r153, %r152, 1024; + // begin inline asm + cp.async.cg.shared.global [ %r153 + 0 ], [ %rd171 + 0 ], 0x10, %r65; + // end inline asm + cp.async.commit_group; + .loc 1 144 11 // chunk_fwd.py:144:11 + @%p31 bra $L__BB0_7; +// %bb.4: // in Loop: Header=BB0_3 Depth=1 + .loc 1 0 11 // chunk_fwd.py:0:11 + setp.le.u64 %p36, %rd2, %rd5; + .loc 1 93 28 // chunk_fwd.py:93:28 + setp.lt.u64 %p35, %rd9, %rd2; + setp.lt.u64 %p34, %rd8, %rd2; + .loc 1 145 56 // chunk_fwd.py:145:56 + add.s64 %rd176, %rd318, -327680; + .loc 1 146 27 // chunk_fwd.py:146:27 + add.s64 %rd177, %rd318, -262144; + mov.b32 %r195, 0; + // begin inline asm + mov.u32 %r167, %r195; + mov.u32 %r168, %r195; + mov.u32 %r169, %r195; + mov.u32 %r170, %r195; + @%p34 ld.global.v4.b32 { %r167, %r168, %r169, %r170 }, [ %rd176 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r171, %r195; + mov.u32 %r172, %r195; + mov.u32 %r173, %r195; + mov.u32 %r174, %r195; + @%p35 ld.global.v4.b32 { %r171, %r172, %r173, %r174 }, [ %rd177 + 0 ]; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r225 + 0 ], { %r167, %r168, %r169, %r170 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r230 + 0 ], { %r171, %r172, %r173, %r174 }; + // end inline asm + .loc 1 148 42 // chunk_fwd.py:148:42 + // begin inline asm + st.shared.v4.b32 [ %r235 + 0 ], { %r167, %r168, %r169, %r170 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r288, %r289, %r292, %r293}, [%r19]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r296, %r297, %r300, %r301}, [%r20]; + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r235 + 0 ], { %r171, %r172, %r173, %r174 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r290, %r291, %r294, %r295}, [%r19]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r298, %r299, %r302, %r303}, [%r20]; + .loc 1 146 27 // chunk_fwd.py:146:27 + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r175, %r176, %r177, %r178}, [%r21]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r181, %r182, %r183, %r184}, [%r23]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r187, %r188, %r189, %r190}, [%r25]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r191, %r192, %r193, %r194}, [%r27]; + .loc 1 148 49 // chunk_fwd.py:148:49 + mov.b64 {%r179, %r180}, %rd324; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r179, %r180, %r1850, %r1851 }, { %r175, %r176, %r177, %r178 }, { %r288, %r289 }, { %r179, %r180, %r1850, %r1851 }; + // end inline asm + mov.b64 {%r185, %r186}, %rd325; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1852, %r1853, %r185, %r186 }, { %r175, %r176, %r177, %r178 }, { %r290, %r291 }, { %r1852, %r1853, %r185, %r186 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r179, %r180, %r1850, %r1851 }, { %r181, %r182, %r183, %r184 }, { %r292, %r293 }, { %r179, %r180, %r1850, %r1851 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1852, %r1853, %r185, %r186 }, { %r181, %r182, %r183, %r184 }, { %r294, %r295 }, { %r1852, %r1853, %r185, %r186 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r179, %r180, %r1850, %r1851 }, { %r187, %r188, %r189, %r190 }, { %r296, %r297 }, { %r179, %r180, %r1850, %r1851 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1852, %r1853, %r185, %r186 }, { %r187, %r188, %r189, %r190 }, { %r298, %r299 }, { %r1852, %r1853, %r185, %r186 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r179, %r180, %r1850, %r1851 }, { %r191, %r192, %r193, %r194 }, { %r300, %r301 }, { %r179, %r180, %r1850, %r1851 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1852, %r1853, %r185, %r186 }, { %r191, %r192, %r193, %r194 }, { %r302, %r303 }, { %r1852, %r1853, %r185, %r186 }; + // end inline asm + .loc 1 150 49 // chunk_fwd.py:150:49 + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1862, %r1863, %r1864, %r1865 }, { %r175, %r176, %r177, %r178 }, { %r272, %r273 }, { %r1862, %r1863, %r1864, %r1865 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1866, %r1867, %r1868, %r1869 }, { %r175, %r176, %r177, %r178 }, { %r274, %r275 }, { %r1866, %r1867, %r1868, %r1869 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1862, %r1863, %r1864, %r1865 }, { %r181, %r182, %r183, %r184 }, { %r276, %r277 }, { %r1862, %r1863, %r1864, %r1865 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1866, %r1867, %r1868, %r1869 }, { %r181, %r182, %r183, %r184 }, { %r278, %r279 }, { %r1866, %r1867, %r1868, %r1869 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1862, %r1863, %r1864, %r1865 }, { %r187, %r188, %r189, %r190 }, { %r280, %r281 }, { %r1862, %r1863, %r1864, %r1865 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1866, %r1867, %r1868, %r1869 }, { %r187, %r188, %r189, %r190 }, { %r282, %r283 }, { %r1866, %r1867, %r1868, %r1869 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1862, %r1863, %r1864, %r1865 }, { %r191, %r192, %r193, %r194 }, { %r284, %r285 }, { %r1862, %r1863, %r1864, %r1865 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1866, %r1867, %r1868, %r1869 }, { %r191, %r192, %r193, %r194 }, { %r286, %r287 }, { %r1866, %r1867, %r1868, %r1869 }; + // end inline asm + .loc 1 148 49 // chunk_fwd.py:148:49 + mov.b64 %rd324, {%r179, %r180}; + mov.b64 %rd325, {%r185, %r186}; + .loc 1 152 15 // chunk_fwd.py:152:15 + @%p36 bra $L__BB0_7; +// %bb.5: // in Loop: Header=BB0_3 Depth=1 + .loc 1 0 15 // chunk_fwd.py:0:15 + setp.le.u64 %p39, %rd2, %rd6; + .loc 1 94 28 // chunk_fwd.py:94:28 + setp.lt.u64 %p38, %rd11, %rd2; + setp.lt.u64 %p37, %rd10, %rd2; + .loc 1 153 60 // chunk_fwd.py:153:60 + add.s64 %rd178, %rd318, -196608; + .loc 1 154 31 // chunk_fwd.py:154:31 + add.s64 %rd179, %rd318, -131072; + // begin inline asm + mov.u32 %r196, %r195; + mov.u32 %r197, %r195; + mov.u32 %r198, %r195; + mov.u32 %r199, %r195; + @%p37 ld.global.v4.b32 { %r196, %r197, %r198, %r199 }, [ %rd178 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r200, %r195; + mov.u32 %r201, %r195; + mov.u32 %r202, %r195; + mov.u32 %r203, %r195; + @%p38 ld.global.v4.b32 { %r200, %r201, %r202, %r203 }, [ %rd179 + 0 ]; + // end inline asm + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r225 + 0 ], { %r196, %r197, %r198, %r199 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r230 + 0 ], { %r200, %r201, %r202, %r203 }; + // end inline asm + .loc 1 156 46 // chunk_fwd.py:156:46 + // begin inline asm + st.shared.v4.b32 [ %r235 + 0 ], { %r196, %r197, %r198, %r199 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r304, %r305, %r308, %r309}, [%r19]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r312, %r313, %r316, %r317}, [%r20]; + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r235 + 0 ], { %r200, %r201, %r202, %r203 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r306, %r307, %r310, %r311}, [%r19]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r314, %r315, %r318, %r319}, [%r20]; + .loc 1 154 31 // chunk_fwd.py:154:31 + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r204, %r205, %r206, %r207}, [%r21]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r210, %r211, %r212, %r213}, [%r23]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r216, %r217, %r218, %r219}, [%r25]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r220, %r221, %r222, %r223}, [%r27]; + .loc 1 156 53 // chunk_fwd.py:156:53 + mov.b64 {%r208, %r209}, %rd322; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r208, %r209, %r1854, %r1855 }, { %r204, %r205, %r206, %r207 }, { %r304, %r305 }, { %r208, %r209, %r1854, %r1855 }; + // end inline asm + mov.b64 {%r214, %r215}, %rd323; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1856, %r1857, %r214, %r215 }, { %r204, %r205, %r206, %r207 }, { %r306, %r307 }, { %r1856, %r1857, %r214, %r215 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r208, %r209, %r1854, %r1855 }, { %r210, %r211, %r212, %r213 }, { %r308, %r309 }, { %r208, %r209, %r1854, %r1855 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1856, %r1857, %r214, %r215 }, { %r210, %r211, %r212, %r213 }, { %r310, %r311 }, { %r1856, %r1857, %r214, %r215 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r208, %r209, %r1854, %r1855 }, { %r216, %r217, %r218, %r219 }, { %r312, %r313 }, { %r208, %r209, %r1854, %r1855 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1856, %r1857, %r214, %r215 }, { %r216, %r217, %r218, %r219 }, { %r314, %r315 }, { %r1856, %r1857, %r214, %r215 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r208, %r209, %r1854, %r1855 }, { %r220, %r221, %r222, %r223 }, { %r316, %r317 }, { %r208, %r209, %r1854, %r1855 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1856, %r1857, %r214, %r215 }, { %r220, %r221, %r222, %r223 }, { %r318, %r319 }, { %r1856, %r1857, %r214, %r215 }; + // end inline asm + .loc 1 158 53 // chunk_fwd.py:158:53 + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1870, %r1871, %r1872, %r1873 }, { %r204, %r205, %r206, %r207 }, { %r272, %r273 }, { %r1870, %r1871, %r1872, %r1873 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1874, %r1875, %r1876, %r1877 }, { %r204, %r205, %r206, %r207 }, { %r274, %r275 }, { %r1874, %r1875, %r1876, %r1877 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1870, %r1871, %r1872, %r1873 }, { %r210, %r211, %r212, %r213 }, { %r276, %r277 }, { %r1870, %r1871, %r1872, %r1873 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1874, %r1875, %r1876, %r1877 }, { %r210, %r211, %r212, %r213 }, { %r278, %r279 }, { %r1874, %r1875, %r1876, %r1877 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1870, %r1871, %r1872, %r1873 }, { %r216, %r217, %r218, %r219 }, { %r280, %r281 }, { %r1870, %r1871, %r1872, %r1873 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1874, %r1875, %r1876, %r1877 }, { %r216, %r217, %r218, %r219 }, { %r282, %r283 }, { %r1874, %r1875, %r1876, %r1877 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1870, %r1871, %r1872, %r1873 }, { %r220, %r221, %r222, %r223 }, { %r284, %r285 }, { %r1870, %r1871, %r1872, %r1873 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1874, %r1875, %r1876, %r1877 }, { %r220, %r221, %r222, %r223 }, { %r286, %r287 }, { %r1874, %r1875, %r1876, %r1877 }; + // end inline asm + .loc 1 159 53 // chunk_fwd.py:159:53 + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1878, %r1879, %r1880, %r1881 }, { %r204, %r205, %r206, %r207 }, { %r288, %r289 }, { %r1878, %r1879, %r1880, %r1881 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1882, %r1883, %r1884, %r1885 }, { %r204, %r205, %r206, %r207 }, { %r290, %r291 }, { %r1882, %r1883, %r1884, %r1885 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1878, %r1879, %r1880, %r1881 }, { %r210, %r211, %r212, %r213 }, { %r292, %r293 }, { %r1878, %r1879, %r1880, %r1881 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1882, %r1883, %r1884, %r1885 }, { %r210, %r211, %r212, %r213 }, { %r294, %r295 }, { %r1882, %r1883, %r1884, %r1885 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1878, %r1879, %r1880, %r1881 }, { %r216, %r217, %r218, %r219 }, { %r296, %r297 }, { %r1878, %r1879, %r1880, %r1881 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1882, %r1883, %r1884, %r1885 }, { %r216, %r217, %r218, %r219 }, { %r298, %r299 }, { %r1882, %r1883, %r1884, %r1885 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1878, %r1879, %r1880, %r1881 }, { %r220, %r221, %r222, %r223 }, { %r300, %r301 }, { %r1878, %r1879, %r1880, %r1881 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1882, %r1883, %r1884, %r1885 }, { %r220, %r221, %r222, %r223 }, { %r302, %r303 }, { %r1882, %r1883, %r1884, %r1885 }; + // end inline asm + .loc 1 152 15 // chunk_fwd.py:152:15 + mov.b64 %rd322, {%r208, %r209}; + mov.b64 %rd323, {%r214, %r215}; + .loc 1 161 19 // chunk_fwd.py:161:19 + @%p39 bra $L__BB0_7; +// %bb.6: // in Loop: Header=BB0_3 Depth=1 + .loc 1 95 28 // chunk_fwd.py:95:28 + setp.lt.u64 %p41, %rd13, %rd2; + setp.lt.u64 %p40, %rd12, %rd2; + .loc 1 162 64 // chunk_fwd.py:162:64 + add.s64 %rd180, %rd318, -65536; + mov.b32 %r224, 0; + .loc 1 163 35 // chunk_fwd.py:163:35 + // begin inline asm + mov.u32 %r226, %r224; + mov.u32 %r227, %r224; + mov.u32 %r228, %r224; + mov.u32 %r229, %r224; + @%p40 ld.global.v4.b32 { %r226, %r227, %r228, %r229 }, [ %rd180 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r231, %r224; + mov.u32 %r232, %r224; + mov.u32 %r233, %r224; + mov.u32 %r234, %r224; + @%p41 ld.global.v4.b32 { %r231, %r232, %r233, %r234 }, [ %rd318 + 0 ]; + // end inline asm + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r225 + 0 ], { %r226, %r227, %r228, %r229 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r230 + 0 ], { %r231, %r232, %r233, %r234 }; + // end inline asm + .loc 1 165 50 // chunk_fwd.py:165:50 + // begin inline asm + st.shared.v4.b32 [ %r235 + 0 ], { %r226, %r227, %r228, %r229 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r240, %r241, %r250, %r251}, [%r19]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r260, %r261, %r268, %r269}, [%r20]; + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r235 + 0 ], { %r231, %r232, %r233, %r234 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r242, %r243, %r254, %r255}, [%r19]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r262, %r263, %r270, %r271}, [%r20]; + .loc 1 163 35 // chunk_fwd.py:163:35 + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r236, %r237, %r238, %r239}, [%r21]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r246, %r247, %r248, %r249}, [%r23]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r256, %r257, %r258, %r259}, [%r25]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r264, %r265, %r266, %r267}, [%r27]; + .loc 1 165 57 // chunk_fwd.py:165:57 + mov.b64 {%r244, %r245}, %rd320; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r244, %r245, %r1858, %r1859 }, { %r236, %r237, %r238, %r239 }, { %r240, %r241 }, { %r244, %r245, %r1858, %r1859 }; + // end inline asm + mov.b64 {%r252, %r253}, %rd321; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1860, %r1861, %r252, %r253 }, { %r236, %r237, %r238, %r239 }, { %r242, %r243 }, { %r1860, %r1861, %r252, %r253 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r244, %r245, %r1858, %r1859 }, { %r246, %r247, %r248, %r249 }, { %r250, %r251 }, { %r244, %r245, %r1858, %r1859 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1860, %r1861, %r252, %r253 }, { %r246, %r247, %r248, %r249 }, { %r254, %r255 }, { %r1860, %r1861, %r252, %r253 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r244, %r245, %r1858, %r1859 }, { %r256, %r257, %r258, %r259 }, { %r260, %r261 }, { %r244, %r245, %r1858, %r1859 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1860, %r1861, %r252, %r253 }, { %r256, %r257, %r258, %r259 }, { %r262, %r263 }, { %r1860, %r1861, %r252, %r253 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r244, %r245, %r1858, %r1859 }, { %r264, %r265, %r266, %r267 }, { %r268, %r269 }, { %r244, %r245, %r1858, %r1859 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1860, %r1861, %r252, %r253 }, { %r264, %r265, %r266, %r267 }, { %r270, %r271 }, { %r1860, %r1861, %r252, %r253 }; + // end inline asm + .loc 1 167 57 // chunk_fwd.py:167:57 + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1886, %r1887, %r1888, %r1889 }, { %r236, %r237, %r238, %r239 }, { %r272, %r273 }, { %r1886, %r1887, %r1888, %r1889 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1890, %r1891, %r1892, %r1893 }, { %r236, %r237, %r238, %r239 }, { %r274, %r275 }, { %r1890, %r1891, %r1892, %r1893 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1886, %r1887, %r1888, %r1889 }, { %r246, %r247, %r248, %r249 }, { %r276, %r277 }, { %r1886, %r1887, %r1888, %r1889 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1890, %r1891, %r1892, %r1893 }, { %r246, %r247, %r248, %r249 }, { %r278, %r279 }, { %r1890, %r1891, %r1892, %r1893 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1886, %r1887, %r1888, %r1889 }, { %r256, %r257, %r258, %r259 }, { %r280, %r281 }, { %r1886, %r1887, %r1888, %r1889 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1890, %r1891, %r1892, %r1893 }, { %r256, %r257, %r258, %r259 }, { %r282, %r283 }, { %r1890, %r1891, %r1892, %r1893 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1886, %r1887, %r1888, %r1889 }, { %r264, %r265, %r266, %r267 }, { %r284, %r285 }, { %r1886, %r1887, %r1888, %r1889 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1890, %r1891, %r1892, %r1893 }, { %r264, %r265, %r266, %r267 }, { %r286, %r287 }, { %r1890, %r1891, %r1892, %r1893 }; + // end inline asm + .loc 1 168 57 // chunk_fwd.py:168:57 + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1894, %r1895, %r1896, %r1897 }, { %r236, %r237, %r238, %r239 }, { %r288, %r289 }, { %r1894, %r1895, %r1896, %r1897 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1898, %r1899, %r1900, %r1901 }, { %r236, %r237, %r238, %r239 }, { %r290, %r291 }, { %r1898, %r1899, %r1900, %r1901 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1894, %r1895, %r1896, %r1897 }, { %r246, %r247, %r248, %r249 }, { %r292, %r293 }, { %r1894, %r1895, %r1896, %r1897 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1898, %r1899, %r1900, %r1901 }, { %r246, %r247, %r248, %r249 }, { %r294, %r295 }, { %r1898, %r1899, %r1900, %r1901 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1894, %r1895, %r1896, %r1897 }, { %r256, %r257, %r258, %r259 }, { %r296, %r297 }, { %r1894, %r1895, %r1896, %r1897 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1898, %r1899, %r1900, %r1901 }, { %r256, %r257, %r258, %r259 }, { %r298, %r299 }, { %r1898, %r1899, %r1900, %r1901 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1894, %r1895, %r1896, %r1897 }, { %r264, %r265, %r266, %r267 }, { %r300, %r301 }, { %r1894, %r1895, %r1896, %r1897 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1898, %r1899, %r1900, %r1901 }, { %r264, %r265, %r266, %r267 }, { %r302, %r303 }, { %r1898, %r1899, %r1900, %r1901 }; + // end inline asm + .loc 1 169 57 // chunk_fwd.py:169:57 + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1902, %r1903, %r1904, %r1905 }, { %r236, %r237, %r238, %r239 }, { %r304, %r305 }, { %r1902, %r1903, %r1904, %r1905 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1906, %r1907, %r1908, %r1909 }, { %r236, %r237, %r238, %r239 }, { %r306, %r307 }, { %r1906, %r1907, %r1908, %r1909 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1902, %r1903, %r1904, %r1905 }, { %r246, %r247, %r248, %r249 }, { %r308, %r309 }, { %r1902, %r1903, %r1904, %r1905 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1906, %r1907, %r1908, %r1909 }, { %r246, %r247, %r248, %r249 }, { %r310, %r311 }, { %r1906, %r1907, %r1908, %r1909 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1902, %r1903, %r1904, %r1905 }, { %r256, %r257, %r258, %r259 }, { %r312, %r313 }, { %r1902, %r1903, %r1904, %r1905 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1906, %r1907, %r1908, %r1909 }, { %r256, %r257, %r258, %r259 }, { %r314, %r315 }, { %r1906, %r1907, %r1908, %r1909 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1902, %r1903, %r1904, %r1905 }, { %r264, %r265, %r266, %r267 }, { %r316, %r317 }, { %r1902, %r1903, %r1904, %r1905 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1906, %r1907, %r1908, %r1909 }, { %r264, %r265, %r266, %r267 }, { %r318, %r319 }, { %r1906, %r1907, %r1908, %r1909 }; + // end inline asm + .loc 1 152 15 // chunk_fwd.py:152:15 + mov.b64 %rd320, {%r244, %r245}; + mov.b64 %rd321, {%r252, %r253}; + bra.uni $L__BB0_7; +$L__BB0_8: // %__nv_exp2f.exit + .loc 1 102 51 // chunk_fwd.py:102:51 + mov.b32 %r321, {%rs1, %rs1}; + mov.b32 %r322, {%rs3, %rs3}; + mov.b32 {%rs10, %rs11}, %r322; + cvt.f32.bf16 %r323, %rs11; + cvt.f32.bf16 %r324, %rs10; + mov.b64 %rd183, {%r324, %r323}; + mov.b32 {%rs12, %rs13}, %r321; + cvt.f32.bf16 %r325, %rs13; + cvt.f32.bf16 %r326, %rs12; + mov.b64 %rd184, {%r326, %r325}; + .loc 1 103 51 // chunk_fwd.py:103:51 + mov.b32 %r327, {%rs4, %rs4}; + mov.b32 %r328, {%rs5, %rs5}; + mov.b32 {%rs14, %rs15}, %r328; + cvt.f32.bf16 %r329, %rs15; + cvt.f32.bf16 %r330, %rs14; + mov.b64 %rd185, {%r330, %r329}; + mov.b32 {%rs16, %rs17}, %r327; + cvt.f32.bf16 %r331, %rs17; + cvt.f32.bf16 %r332, %rs16; + mov.b64 %rd186, {%r332, %r331}; + .loc 1 104 51 // chunk_fwd.py:104:51 + mov.b32 %r333, {%rs6, %rs6}; + mov.b32 %r334, {%rs7, %rs7}; + mov.b32 {%rs18, %rs19}, %r334; + cvt.f32.bf16 %r335, %rs19; + cvt.f32.bf16 %r336, %rs18; + mov.b32 {%rs20, %rs21}, %r333; + cvt.f32.bf16 %r337, %rs21; + cvt.f32.bf16 %r338, %rs20; + .loc 1 105 51 // chunk_fwd.py:105:51 + mov.b32 %r339, {%rs8, %rs8}; + mov.b32 %r340, {%rs9, %rs9}; + mov.b32 {%rs22, %rs23}, %r340; + mov.b32 {%rs24, %rs25}, %r339; + .loc 1 137 21 // chunk_fwd.py:137:21 + cp.async.wait_group 0; + bar.sync 0; + .loc 1 179 25 // chunk_fwd.py:179:25 + mov.b64 {%r345, %r346}, %rd23; + mov.b64 {%r347, %r348}, %rd26; + setp.gt.u32 %p42, %r348, %r346; + setp.gt.u32 %p43, %r348, %r345; + setp.gt.u32 %p44, %r347, %r345; + setp.gt.u32 %p45, %r347, %r346; + setp.gt.u32 %p46, %r11, %r9; + mov.b64 {%r349, %r350}, %rd14; + setp.gt.u32 %p47, %r10, %r349; + setp.gt.u32 %p48, %r10, %r350; + .loc 1 183 86 // chunk_fwd.py:183:86 + sub.f32 %r351, %r64, %r67; + sub.f32 %r352, %r64, %r68; + sub.f32 %r353, %r66, %r67; + sub.f32 %r354, %r66, %r68; +$L__tmp1: + .loc 2 47 33 // op.py:47:33 @[ chunk_fwd.py:183:70 ] + ex2.approx.ftz.f32 %r355, %r351; + ex2.approx.ftz.f32 %r356, %r352; + ex2.approx.ftz.f32 %r357, %r353; + ex2.approx.ftz.f32 %r358, %r354; +$L__tmp2: + .loc 1 183 102 // chunk_fwd.py:183:102 + selp.f32 %r359, %r356, 0f00000000, %p9; + selp.f32 %r360, %r359, 0f00000000, %p45; + selp.f32 %r361, %r360, 0f00000000, %p12; + selp.f32 %r362, %r355, 0f00000000, %p9; + selp.f32 %r363, %r362, 0f00000000, %p44; + selp.f32 %r364, %r363, 0f00000000, %p11; + mov.b64 %rd191, {%r364, %r361}; + selp.f32 %r365, %r357, 0f00000000, %p10; + selp.f32 %r366, %r365, 0f00000000, %p43; + selp.f32 %r367, %r366, 0f00000000, %p11; + selp.f32 %r368, %r358, 0f00000000, %p10; + selp.f32 %r369, %r368, 0f00000000, %p42; + selp.f32 %r370, %r369, 0f00000000, %p12; + mov.b64 %rd192, {%r367, %r370}; + mov.b32 %r371, 0f00000000; + mov.b64 %rd193, {%r371, %r371}; + .loc 1 183 17 // chunk_fwd.py:183:17 + mov.b64 %rd194, {%r1846, %r1847}; + mov.b64 %rd195, {%r1848, %r1849}; + mul.f32x2 %rd196, %rd195, %rd192; + mul.f32x2 %rd197, %rd194, %rd191; + .loc 1 184 86 // chunk_fwd.py:184:86 + sub.f32 %r372, %r73, %r75; + sub.f32 %r373, %r73, %r76; + sub.f32 %r374, %r74, %r75; + sub.f32 %r375, %r74, %r76; + sub.f32 %r376, %r74, %r77; + sub.f32 %r377, %r74, %r78; +$L__tmp3: + .loc 2 47 33 // op.py:47:33 @[ chunk_fwd.py:184:70 ] + ex2.approx.ftz.f32 %r378, %r372; + ex2.approx.ftz.f32 %r379, %r373; + ex2.approx.ftz.f32 %r380, %r374; + ex2.approx.ftz.f32 %r381, %r375; + ex2.approx.ftz.f32 %r382, %r376; + ex2.approx.ftz.f32 %r383, %r377; +$L__tmp4: + .loc 1 184 102 // chunk_fwd.py:184:102 + selp.f32 %r384, %r379, 0f00000000, %p48; + selp.f32 %r385, %r384, 0f00000000, %p18; + selp.f32 %r386, %r385, 0f00000000, %p3; + selp.f32 %r387, %r378, 0f00000000, %p47; + selp.f32 %r388, %r387, 0f00000000, %p17; + selp.f32 %r389, %r388, 0f00000000, %p3; + mov.b64 %rd198, {%r389, %r386}; + selp.f32 %r390, %r382, 0f00000000, %p47; + selp.f32 %r391, %r390, 0f00000000, %p19; + selp.f32 %r392, %r391, 0f00000000, %p4; + selp.f32 %r393, %r383, 0f00000000, %p46; + selp.f32 %r394, %r393, 0f00000000, %p20; + selp.f32 %r395, %r394, 0f00000000, %p4; + mov.b64 %rd199, {%r392, %r395}; + selp.f32 %r396, %r380, 0f00000000, %p17; + selp.f32 %r397, %r396, 0f00000000, %p4; + selp.f32 %r398, %r381, 0f00000000, %p18; + selp.f32 %r399, %r398, 0f00000000, %p4; + .loc 1 184 17 // chunk_fwd.py:184:17 + mul.f32 %r400, %r1850, %r397; + mul.f32 %r401, %r1851, %r399; + mul.f32 %r402, %r1852, 0f00000000; + mul.f32 %r403, %r1853, 0f00000000; + mul.f32x2 %rd200, %rd325, %rd199; + mul.f32x2 %rd201, %rd324, %rd198; + .loc 1 185 86 // chunk_fwd.py:185:86 + sub.f32 %r404, %r79, %r81; + sub.f32 %r405, %r79, %r82; + sub.f32 %r406, %r80, %r81; + sub.f32 %r407, %r80, %r82; + sub.f32 %r408, %r80, %r83; + sub.f32 %r409, %r80, %r84; +$L__tmp5: + .loc 2 47 33 // op.py:47:33 @[ chunk_fwd.py:185:70 ] + ex2.approx.ftz.f32 %r410, %r404; + ex2.approx.ftz.f32 %r411, %r405; + ex2.approx.ftz.f32 %r412, %r406; + ex2.approx.ftz.f32 %r413, %r407; + ex2.approx.ftz.f32 %r414, %r408; + ex2.approx.ftz.f32 %r415, %r409; +$L__tmp6: + .loc 1 185 102 // chunk_fwd.py:185:102 + selp.f32 %r416, %r411, 0f00000000, %p48; + selp.f32 %r417, %r416, 0f00000000, %p22; + selp.f32 %r418, %r417, 0f00000000, %p5; + selp.f32 %r419, %r410, 0f00000000, %p47; + selp.f32 %r420, %r419, 0f00000000, %p21; + selp.f32 %r421, %r420, 0f00000000, %p5; + mov.b64 %rd202, {%r421, %r418}; + selp.f32 %r422, %r414, 0f00000000, %p47; + selp.f32 %r423, %r422, 0f00000000, %p23; + selp.f32 %r424, %r423, 0f00000000, %p6; + selp.f32 %r425, %r415, 0f00000000, %p46; + selp.f32 %r426, %r425, 0f00000000, %p24; + selp.f32 %r427, %r426, 0f00000000, %p6; + mov.b64 %rd203, {%r424, %r427}; + selp.f32 %r428, %r412, 0f00000000, %p21; + selp.f32 %r429, %r428, 0f00000000, %p6; + selp.f32 %r430, %r413, 0f00000000, %p22; + selp.f32 %r431, %r430, 0f00000000, %p6; + .loc 1 185 17 // chunk_fwd.py:185:17 + mul.f32 %r432, %r1854, %r429; + mul.f32 %r433, %r1855, %r431; + mul.f32 %r434, %r1856, 0f00000000; + mul.f32 %r435, %r1857, 0f00000000; + mul.f32x2 %rd204, %rd323, %rd203; + mul.f32x2 %rd205, %rd322, %rd202; + .loc 1 186 86 // chunk_fwd.py:186:86 + sub.f32 %r436, %r85, %r87; + sub.f32 %r437, %r85, %r88; + sub.f32 %r438, %r86, %r87; + sub.f32 %r439, %r86, %r88; + sub.f32 %r440, %r86, %r89; + sub.f32 %r441, %r86, %r90; +$L__tmp7: + .loc 2 47 33 // op.py:47:33 @[ chunk_fwd.py:186:70 ] + ex2.approx.ftz.f32 %r442, %r436; + ex2.approx.ftz.f32 %r443, %r437; + ex2.approx.ftz.f32 %r444, %r438; + ex2.approx.ftz.f32 %r445, %r439; + ex2.approx.ftz.f32 %r446, %r440; + ex2.approx.ftz.f32 %r447, %r441; +$L__tmp8: + .loc 1 186 102 // chunk_fwd.py:186:102 + selp.f32 %r448, %r443, 0f00000000, %p48; + selp.f32 %r449, %r448, 0f00000000, %p26; + selp.f32 %r450, %r449, 0f00000000, %p7; + selp.f32 %r451, %r442, 0f00000000, %p47; + selp.f32 %r452, %r451, 0f00000000, %p25; + selp.f32 %r453, %r452, 0f00000000, %p7; + mov.b64 %rd206, {%r453, %r450}; + selp.f32 %r454, %r446, 0f00000000, %p47; + selp.f32 %r455, %r454, 0f00000000, %p27; + selp.f32 %r456, %r455, 0f00000000, %p8; + selp.f32 %r457, %r447, 0f00000000, %p46; + selp.f32 %r458, %r457, 0f00000000, %p28; + selp.f32 %r459, %r458, 0f00000000, %p8; + mov.b64 %rd207, {%r456, %r459}; + selp.f32 %r460, %r444, 0f00000000, %p25; + selp.f32 %r461, %r460, 0f00000000, %p8; + selp.f32 %r462, %r445, 0f00000000, %p26; + selp.f32 %r463, %r462, 0f00000000, %p8; + .loc 1 186 17 // chunk_fwd.py:186:17 + mul.f32x2 %rd208, %rd321, %rd207; + mul.f32x2 %rd209, %rd320, %rd206; + .loc 1 188 80 // chunk_fwd.py:188:80 + sub.f32 %r468, %r73, %r69; + sub.f32 %r469, %r73, %r70; + sub.f32 %r470, %r74, %r69; + sub.f32 %r471, %r74, %r70; + sub.f32 %r472, %r73, %r71; + sub.f32 %r473, %r73, %r72; + sub.f32 %r474, %r74, %r71; + sub.f32 %r475, %r74, %r72; +$L__tmp9: + .loc 2 47 33 // op.py:47:33 @[ chunk_fwd.py:188:64 ] + ex2.approx.ftz.f32 %r476, %r468; + ex2.approx.ftz.f32 %r477, %r469; + ex2.approx.ftz.f32 %r478, %r470; + ex2.approx.ftz.f32 %r479, %r471; + ex2.approx.ftz.f32 %r480, %r472; + ex2.approx.ftz.f32 %r481, %r473; + ex2.approx.ftz.f32 %r482, %r474; + ex2.approx.ftz.f32 %r483, %r475; +$L__tmp10: + .loc 1 188 96 // chunk_fwd.py:188:96 + selp.f32 %r484, %r476, 0f00000000, %p13; + selp.f32 %r486, %r477, 0f00000000, %p14; + selp.f32 %r488, %r478, 0f00000000, %p13; + selp.f32 %r490, %r479, 0f00000000, %p14; + selp.f32 %r492, %r480, 0f00000000, %p15; + selp.f32 %r494, %r481, 0f00000000, %p16; + selp.f32 %r496, %r482, 0f00000000, %p15; + selp.f32 %r498, %r483, 0f00000000, %p16; + .loc 1 189 80 // chunk_fwd.py:189:80 + sub.f32 %r508, %r79, %r69; + sub.f32 %r509, %r79, %r70; + sub.f32 %r510, %r80, %r69; + sub.f32 %r511, %r80, %r70; + sub.f32 %r512, %r79, %r71; + sub.f32 %r513, %r79, %r72; + sub.f32 %r514, %r80, %r71; + sub.f32 %r515, %r80, %r72; +$L__tmp11: + .loc 2 47 33 // op.py:47:33 @[ chunk_fwd.py:189:64 ] + ex2.approx.ftz.f32 %r516, %r508; + ex2.approx.ftz.f32 %r517, %r509; + ex2.approx.ftz.f32 %r518, %r510; + ex2.approx.ftz.f32 %r519, %r511; + ex2.approx.ftz.f32 %r520, %r512; + ex2.approx.ftz.f32 %r521, %r513; + ex2.approx.ftz.f32 %r522, %r514; + ex2.approx.ftz.f32 %r523, %r515; +$L__tmp12: + .loc 1 189 96 // chunk_fwd.py:189:96 + selp.f32 %r524, %r516, 0f00000000, %p13; + selp.f32 %r526, %r517, 0f00000000, %p14; + selp.f32 %r528, %r518, 0f00000000, %p13; + selp.f32 %r530, %r519, 0f00000000, %p14; + selp.f32 %r532, %r520, 0f00000000, %p15; + selp.f32 %r534, %r521, 0f00000000, %p16; + selp.f32 %r536, %r522, 0f00000000, %p15; + selp.f32 %r538, %r523, 0f00000000, %p16; + .loc 1 190 80 // chunk_fwd.py:190:80 + sub.f32 %r548, %r79, %r75; + sub.f32 %r549, %r79, %r76; + sub.f32 %r550, %r80, %r75; + sub.f32 %r551, %r80, %r76; + sub.f32 %r552, %r79, %r77; + sub.f32 %r553, %r79, %r78; + sub.f32 %r554, %r80, %r77; + sub.f32 %r555, %r80, %r78; +$L__tmp13: + .loc 2 47 33 // op.py:47:33 @[ chunk_fwd.py:190:64 ] + ex2.approx.ftz.f32 %r556, %r548; + ex2.approx.ftz.f32 %r557, %r549; + ex2.approx.ftz.f32 %r558, %r550; + ex2.approx.ftz.f32 %r559, %r551; + ex2.approx.ftz.f32 %r560, %r552; + ex2.approx.ftz.f32 %r561, %r553; + ex2.approx.ftz.f32 %r562, %r554; + ex2.approx.ftz.f32 %r563, %r555; +$L__tmp14: + .loc 1 190 96 // chunk_fwd.py:190:96 + selp.f32 %r564, %r556, 0f00000000, %p17; + selp.f32 %r566, %r557, 0f00000000, %p18; + selp.f32 %r568, %r558, 0f00000000, %p17; + selp.f32 %r570, %r559, 0f00000000, %p18; + selp.f32 %r572, %r560, 0f00000000, %p19; + selp.f32 %r574, %r561, 0f00000000, %p20; + selp.f32 %r576, %r562, 0f00000000, %p19; + selp.f32 %r578, %r563, 0f00000000, %p20; + .loc 1 191 80 // chunk_fwd.py:191:80 + sub.f32 %r588, %r85, %r69; + sub.f32 %r589, %r85, %r70; + sub.f32 %r590, %r86, %r69; + sub.f32 %r591, %r86, %r70; + sub.f32 %r592, %r85, %r71; + sub.f32 %r593, %r85, %r72; + sub.f32 %r594, %r86, %r71; + sub.f32 %r595, %r86, %r72; +$L__tmp15: + .loc 2 47 33 // op.py:47:33 @[ chunk_fwd.py:191:64 ] + ex2.approx.ftz.f32 %r596, %r588; + ex2.approx.ftz.f32 %r597, %r589; + ex2.approx.ftz.f32 %r598, %r590; + ex2.approx.ftz.f32 %r599, %r591; + ex2.approx.ftz.f32 %r600, %r592; + ex2.approx.ftz.f32 %r601, %r593; + ex2.approx.ftz.f32 %r602, %r594; + ex2.approx.ftz.f32 %r603, %r595; +$L__tmp16: + .loc 1 191 96 // chunk_fwd.py:191:96 + selp.f32 %r604, %r596, 0f00000000, %p13; + selp.f32 %r606, %r597, 0f00000000, %p14; + selp.f32 %r608, %r598, 0f00000000, %p13; + selp.f32 %r610, %r599, 0f00000000, %p14; + selp.f32 %r612, %r600, 0f00000000, %p15; + selp.f32 %r614, %r601, 0f00000000, %p16; + selp.f32 %r616, %r602, 0f00000000, %p15; + selp.f32 %r618, %r603, 0f00000000, %p16; + .loc 1 192 80 // chunk_fwd.py:192:80 + sub.f32 %r628, %r85, %r75; + sub.f32 %r629, %r85, %r76; + sub.f32 %r630, %r86, %r75; + sub.f32 %r631, %r86, %r76; + sub.f32 %r632, %r85, %r77; + sub.f32 %r633, %r85, %r78; + sub.f32 %r634, %r86, %r77; + sub.f32 %r635, %r86, %r78; +$L__tmp17: + .loc 2 47 33 // op.py:47:33 @[ chunk_fwd.py:192:64 ] + ex2.approx.ftz.f32 %r636, %r628; + ex2.approx.ftz.f32 %r637, %r629; + ex2.approx.ftz.f32 %r638, %r630; + ex2.approx.ftz.f32 %r639, %r631; + ex2.approx.ftz.f32 %r640, %r632; + ex2.approx.ftz.f32 %r641, %r633; + ex2.approx.ftz.f32 %r642, %r634; + ex2.approx.ftz.f32 %r643, %r635; +$L__tmp18: + .loc 1 192 96 // chunk_fwd.py:192:96 + selp.f32 %r644, %r636, 0f00000000, %p17; + selp.f32 %r646, %r637, 0f00000000, %p18; + selp.f32 %r648, %r638, 0f00000000, %p17; + selp.f32 %r650, %r639, 0f00000000, %p18; + selp.f32 %r652, %r640, 0f00000000, %p19; + selp.f32 %r654, %r641, 0f00000000, %p20; + selp.f32 %r656, %r642, 0f00000000, %p19; + selp.f32 %r658, %r643, 0f00000000, %p20; + .loc 1 193 80 // chunk_fwd.py:193:80 + sub.f32 %r668, %r85, %r81; + sub.f32 %r669, %r85, %r82; + sub.f32 %r670, %r86, %r81; + sub.f32 %r671, %r86, %r82; + sub.f32 %r672, %r85, %r83; + sub.f32 %r673, %r85, %r84; + sub.f32 %r674, %r86, %r83; + sub.f32 %r675, %r86, %r84; +$L__tmp19: + .loc 2 47 33 // op.py:47:33 @[ chunk_fwd.py:193:64 ] + ex2.approx.ftz.f32 %r676, %r668; + ex2.approx.ftz.f32 %r677, %r669; + ex2.approx.ftz.f32 %r678, %r670; + ex2.approx.ftz.f32 %r679, %r671; + ex2.approx.ftz.f32 %r680, %r672; + ex2.approx.ftz.f32 %r681, %r673; + ex2.approx.ftz.f32 %r682, %r674; + ex2.approx.ftz.f32 %r683, %r675; +$L__tmp20: + .loc 1 193 96 // chunk_fwd.py:193:96 + selp.f32 %r684, %r676, 0f00000000, %p21; + selp.f32 %r686, %r677, 0f00000000, %p22; + selp.f32 %r688, %r678, 0f00000000, %p21; + selp.f32 %r690, %r679, 0f00000000, %p22; + selp.f32 %r692, %r680, 0f00000000, %p23; + selp.f32 %r694, %r681, 0f00000000, %p24; + selp.f32 %r696, %r682, 0f00000000, %p23; + selp.f32 %r698, %r683, 0f00000000, %p24; + .loc 1 222 14 // chunk_fwd.py:222:14 + mov.b64 {%r708, %r709}, %rd197; + neg.f32 %r710, %r709; + neg.f32 %r711, %r708; + mov.b64 %rd210, {%r711, %r710}; + fma.rn.f32x2 %rd327, %rd210, %rd184, %rd193; + mov.b64 {%r712, %r713}, %rd196; + neg.f32 %r714, %r713; + neg.f32 %r715, %r712; + mov.b64 %rd211, {%r715, %r714}; + fma.rn.f32x2 %rd328, %rd211, %rd183, %rd193; + .loc 1 223 14 // chunk_fwd.py:223:14 + neg.f32 %r716, %r400; + neg.f32 %r717, %r401; + neg.f32 %r718, %r402; + neg.f32 %r719, %r403; + mov.b64 {%r720, %r721}, %rd201; + neg.f32 %r722, %r721; + neg.f32 %r723, %r720; + mov.b64 %rd212, {%r723, %r722}; + mov.b64 {%r724, %r725}, %rd200; + neg.f32 %r726, %r725; + neg.f32 %r727, %r724; + mov.b64 %rd213, {%r727, %r726}; + .loc 1 224 14 // chunk_fwd.py:224:14 + mov.b64 {%r732, %r733}, %rd205; + neg.f32 %r734, %r733; + neg.f32 %r735, %r732; + mov.b64 {%r736, %r737}, %rd204; + neg.f32 %r738, %r737; + neg.f32 %r739, %r736; + .loc 1 225 14 // chunk_fwd.py:225:14 + mov.b64 {%r744, %r745}, %rd209; + mov.b64 {%r748, %r749}, %rd208; + .loc 1 227 34 // chunk_fwd.py:227:34 + sub.s64 %rd55, %rd2, %rd3; + .loc 1 227 22 // chunk_fwd.py:227:22 + setp.lt.s64 %p49, %rd55, 3; + @%p49 bra $L__BB0_11; +// %bb.9: // %.lr.ph + .loc 1 0 22 // chunk_fwd.py:0:22 + shl.b32 %r752, %r3, 4; + shr.u32 %r753, %r4, 3; + add.s32 %r754, %r108, %r752; + add.s32 %r756, %r754, %r753; + add.s32 %r758, %r756, 8; + shl.b32 %r755, %r2, 1; + add.s32 %r38, %r108, %r755; + .loc 1 227 22 // chunk_fwd.py:227:22 + min.u64 %rd56, %rd55, 16; + .loc 1 228 61 // chunk_fwd.py:228:61 + mov.b64 {%r39, %r40}, %rd327; + mov.b64 {%r41, %r42}, %rd328; + mov.b64 %rd326, 2; +$L__BB0_10: // =>This Inner Loop Header: Depth=1 + .loc 1 228 40 // chunk_fwd.py:228:40 + setp.eq.b64 %p50, %rd29, %rd326; + setp.eq.b64 %p51, %rd27, %rd326; + .loc 1 228 61 // chunk_fwd.py:228:61 + selp.f32 %r760, %r39, 0f00000000, %p50; + selp.f32 %r761, %r40, 0f00000000, %p50; + selp.f32 %r762, %r41, 0f00000000, %p51; + selp.f32 %r763, %r42, 0f00000000, %p51; +$L__tmp21: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:228:66 ] ] + add.f32 %r764, %r760, %r762; + add.f32 %r765, %r761, %r763; +$L__tmp22: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:228:66 ] + shfl.sync.bfly.b32 %r766, %r764, 16, 31, -1; +$L__tmp23: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:228:66 ] ] + add.f32 %r767, %r764, %r766; +$L__tmp24: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:228:66 ] + shfl.sync.bfly.b32 %r768, %r767, 8, 31, -1; +$L__tmp25: + .loc 1 229 31 // chunk_fwd.py:229:31 + setp.gt.u64 %p52, %rd326, %rd24; + setp.gt.u64 %p53, %rd326, %rd25; +$L__tmp26: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:228:66 ] ] + add.f32 %r769, %r767, %r768; +$L__tmp27: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:228:66 ] + shfl.sync.bfly.b32 %r770, %r769, 4, 31, -1; + shfl.sync.bfly.b32 %r771, %r765, 16, 31, -1; +$L__tmp28: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:228:66 ] ] + add.f32 %r772, %r765, %r771; +$L__tmp29: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:228:66 ] + shfl.sync.bfly.b32 %r773, %r772, 8, 31, -1; +$L__tmp30: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:228:66 ] ] + add.f32 %r774, %r772, %r773; +$L__tmp31: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:228:66 ] + shfl.sync.bfly.b32 %r775, %r774, 4, 31, -1; + mov.b64 %rd218, {%r770, %r775}; +$L__tmp32: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:228:66 ] ] + mov.b64 %rd219, {%r769, %r774}; + add.f32x2 %rd220, %rd219, %rd218; +$L__tmp33: + .loc 1 229 41 // chunk_fwd.py:229:41 + mov.b64 {%r776, %r777}, %rd220; + selp.f32 %r759, %r777, 0f00000000, %p53; + selp.f32 %r757, %r776, 0f00000000, %p52; + mov.b64 %rd221, {%r757, %r759}; + .loc 1 230 37 // chunk_fwd.py:230:37 + bar.sync 0; + // begin inline asm + st.shared.b32 [ %r756 + 0 ], %r757; + // end inline asm + // begin inline asm + st.shared.b32 [ %r758 + 0 ], %r759; + // end inline asm + bar.sync 0; + ld.shared.v2.b32 {%r778, %r779}, [%r38]; + .loc 1 230 48 // chunk_fwd.py:230:48 + mov.b64 {%r780, %r781}, %rd327; + mov.b64 {%r782, %r783}, %rd328; + mul.f32 %r784, %r782, %r779; + mul.f32 %r785, %r783, %r779; +$L__tmp34: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:230:56 ] ] + fma.rn.f32 %r786, %r780, %r778, %r784; + fma.rn.f32 %r787, %r781, %r778, %r785; +$L__tmp35: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:230:56 ] + shfl.sync.bfly.b32 %r788, %r786, 16, 31, -1; +$L__tmp36: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:230:56 ] ] + add.f32 %r789, %r786, %r788; +$L__tmp37: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:230:56 ] + shfl.sync.bfly.b32 %r790, %r789, 8, 31, -1; +$L__tmp38: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:230:56 ] ] + add.f32 %r791, %r789, %r790; +$L__tmp39: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:230:56 ] + shfl.sync.bfly.b32 %r792, %r791, 4, 31, -1; + shfl.sync.bfly.b32 %r793, %r787, 16, 31, -1; +$L__tmp40: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:230:56 ] ] + add.f32 %r794, %r787, %r793; +$L__tmp41: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:230:56 ] + shfl.sync.bfly.b32 %r795, %r794, 8, 31, -1; +$L__tmp42: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:230:56 ] ] + add.f32 %r796, %r794, %r795; +$L__tmp43: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:230:56 ] + shfl.sync.bfly.b32 %r797, %r796, 4, 31, -1; + mov.b64 %rd222, {%r792, %r797}; +$L__tmp44: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:230:56 ] ] + mov.b64 %rd223, {%r791, %r796}; + add.f32x2 %rd224, %rd223, %rd222; +$L__tmp45: + .loc 1 230 24 // chunk_fwd.py:230:24 + add.f32x2 %rd225, %rd221, %rd224; + .loc 1 231 54 // chunk_fwd.py:231:54 + mov.b64 {%r798, %r799}, %rd225; + selp.f32 %r800, %r799, %r783, %p51; + selp.f32 %r801, %r798, %r782, %p51; + mov.b64 %rd328, {%r801, %r800}; + selp.f32 %r802, %r799, %r781, %p50; + selp.f32 %r803, %r798, %r780, %p50; + mov.b64 %rd327, {%r803, %r802}; + .loc 1 227 22 // chunk_fwd.py:227:22 + add.s64 %rd326, %rd326, 1; + setp.ne.b64 %p54, %rd56, %rd326; + @%p54 bra $L__BB0_10; +$L__BB0_11: // %._crit_edge + .loc 1 0 0 // chunk_fwd.py:0 + bfe.u32 %r320, %r1, 2, 4; + cvt.f32.bf16 %r341, %rs23; + cvt.f32.bf16 %r342, %rs22; + cvt.f32.bf16 %r343, %rs25; + cvt.f32.bf16 %r344, %rs24; + mul.f32 %r464, %r1858, %r461; + mul.f32 %r465, %r1859, %r463; + mul.f32 %r466, %r1860, 0f00000000; + mul.f32 %r467, %r1861, 0f00000000; + selp.f32 %r485, %r484, 0f00000000, %p3; + selp.f32 %r487, %r486, 0f00000000, %p3; + selp.f32 %r489, %r488, 0f00000000, %p4; + selp.f32 %r491, %r490, 0f00000000, %p4; + selp.f32 %r493, %r492, 0f00000000, %p3; + selp.f32 %r495, %r494, 0f00000000, %p3; + selp.f32 %r497, %r496, 0f00000000, %p4; + selp.f32 %r499, %r498, 0f00000000, %p4; + selp.f32 %r525, %r524, 0f00000000, %p5; + selp.f32 %r527, %r526, 0f00000000, %p5; + selp.f32 %r529, %r528, 0f00000000, %p6; + selp.f32 %r531, %r530, 0f00000000, %p6; + selp.f32 %r533, %r532, 0f00000000, %p5; + selp.f32 %r535, %r534, 0f00000000, %p5; + selp.f32 %r537, %r536, 0f00000000, %p6; + selp.f32 %r539, %r538, 0f00000000, %p6; + selp.f32 %r565, %r564, 0f00000000, %p5; + selp.f32 %r567, %r566, 0f00000000, %p5; + selp.f32 %r569, %r568, 0f00000000, %p6; + selp.f32 %r571, %r570, 0f00000000, %p6; + selp.f32 %r573, %r572, 0f00000000, %p5; + selp.f32 %r575, %r574, 0f00000000, %p5; + selp.f32 %r577, %r576, 0f00000000, %p6; + selp.f32 %r579, %r578, 0f00000000, %p6; + selp.f32 %r605, %r604, 0f00000000, %p7; + selp.f32 %r607, %r606, 0f00000000, %p7; + selp.f32 %r609, %r608, 0f00000000, %p8; + selp.f32 %r611, %r610, 0f00000000, %p8; + selp.f32 %r613, %r612, 0f00000000, %p7; + selp.f32 %r615, %r614, 0f00000000, %p7; + selp.f32 %r617, %r616, 0f00000000, %p8; + selp.f32 %r619, %r618, 0f00000000, %p8; + selp.f32 %r645, %r644, 0f00000000, %p7; + selp.f32 %r647, %r646, 0f00000000, %p7; + selp.f32 %r649, %r648, 0f00000000, %p8; + selp.f32 %r651, %r650, 0f00000000, %p8; + selp.f32 %r653, %r652, 0f00000000, %p7; + selp.f32 %r655, %r654, 0f00000000, %p7; + selp.f32 %r657, %r656, 0f00000000, %p8; + selp.f32 %r659, %r658, 0f00000000, %p8; + selp.f32 %r685, %r684, 0f00000000, %p7; + selp.f32 %r687, %r686, 0f00000000, %p7; + selp.f32 %r689, %r688, 0f00000000, %p8; + selp.f32 %r691, %r690, 0f00000000, %p8; + selp.f32 %r693, %r692, 0f00000000, %p7; + selp.f32 %r695, %r694, 0f00000000, %p7; + selp.f32 %r697, %r696, 0f00000000, %p8; + selp.f32 %r699, %r698, 0f00000000, %p8; + neg.f32 %r746, %r745; + neg.f32 %r747, %r744; + neg.f32 %r750, %r749; + neg.f32 %r751, %r748; + mov.b64 %rd187, {%r336, %r335}; + mov.b64 %rd188, {%r338, %r337}; + neg.f32 %r728, %r432; + neg.f32 %r729, %r433; + neg.f32 %r730, %r434; + neg.f32 %r731, %r435; + mov.b64 %rd214, {%r735, %r734}; + mov.b64 %rd215, {%r739, %r738}; + fma.rn.f32 %r1914, %r716, %r330, 0f00000000; + fma.rn.f32 %r1915, %r717, %r330, 0f00000000; + fma.rn.f32 %r1916, %r718, %r332, 0f00000000; + fma.rn.f32 %r1917, %r719, %r332, 0f00000000; + fma.rn.f32x2 %rd330, %rd212, %rd186, %rd193; + fma.rn.f32x2 %rd331, %rd213, %rd185, %rd193; + .loc 1 232 34 // chunk_fwd.py:232:34 + sub.s64 %rd57, %rd2, %rd4; + .loc 1 232 22 // chunk_fwd.py:232:22 + setp.lt.s64 %p55, %rd57, 3; + @%p55 bra $L__BB0_15; +// %bb.12: // %.lr.ph503 + .loc 1 0 22 // chunk_fwd.py:0:22 + and.b32 %r43, %r1, 4; + and.b32 %r804, %r28, 24; + bfe.u32 %r805, %r1, 3, 2; + or.b32 %r806, %r805, %r804; + or.b32 %r44, %r806, %r43; + xor.b32 %r45, %r44, 4; + .loc 1 232 22 // chunk_fwd.py:232:22 + min.u64 %rd58, %rd57, 16; + .loc 1 233 61 // chunk_fwd.py:233:61 + mov.b64 {%r46, %r47}, %rd330; + mov.b64 {%r48, %r49}, %rd331; + mov.b64 %rd329, 2; + setp.eq.b32 %p56, %r43, 0; + mov.b32 %r1910, %r1917; + mov.b32 %r1911, %r1916; + mov.b32 %r1912, %r1915; + mov.b32 %r1913, %r1914; +$L__BB0_13: // =>This Inner Loop Header: Depth=1 + .loc 1 233 40 // chunk_fwd.py:233:40 + setp.eq.b64 %p57, %rd29, %rd329; + setp.eq.b64 %p58, %rd27, %rd329; + .loc 1 233 61 // chunk_fwd.py:233:61 + selp.f32 %r807, %r46, 0f00000000, %p57; + selp.f32 %r808, %r47, 0f00000000, %p57; + selp.f32 %r809, %r1914, 0f00000000, %p58; + selp.f32 %r810, %r1915, 0f00000000, %p58; + selp.f32 %r811, %r1916, 0f00000000, %p57; + selp.f32 %r812, %r1917, 0f00000000, %p57; + selp.f32 %r813, %r48, 0f00000000, %p58; + selp.f32 %r814, %r49, 0f00000000, %p58; +$L__tmp46: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:233:66 ] ] + add.f32 %r815, %r807, %r809; + add.f32 %r816, %r808, %r810; + add.f32 %r817, %r811, %r813; + add.f32 %r818, %r812, %r814; +$L__tmp47: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:233:66 ] + shfl.sync.bfly.b32 %r819, %r815, 16, 31, -1; +$L__tmp48: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:233:66 ] ] + add.f32 %r820, %r815, %r819; +$L__tmp49: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:233:66 ] + shfl.sync.bfly.b32 %r821, %r820, 8, 31, -1; +$L__tmp50: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:233:66 ] ] + add.f32 %r822, %r820, %r821; +$L__tmp51: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:233:66 ] + shfl.sync.bfly.b32 %r823, %r822, 4, 31, -1; + shfl.sync.bfly.b32 %r824, %r816, 16, 31, -1; +$L__tmp52: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:233:66 ] ] + add.f32 %r825, %r816, %r824; +$L__tmp53: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:233:66 ] + shfl.sync.bfly.b32 %r826, %r825, 8, 31, -1; +$L__tmp54: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:233:66 ] ] + add.f32 %r827, %r825, %r826; +$L__tmp55: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:233:66 ] + shfl.sync.bfly.b32 %r828, %r827, 4, 31, -1; + shfl.sync.bfly.b32 %r829, %r817, 16, 31, -1; +$L__tmp56: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:233:66 ] ] + add.f32 %r830, %r817, %r829; +$L__tmp57: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:233:66 ] + shfl.sync.bfly.b32 %r831, %r830, 8, 31, -1; +$L__tmp58: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:233:66 ] ] + add.f32 %r832, %r830, %r831; +$L__tmp59: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:233:66 ] + shfl.sync.bfly.b32 %r833, %r832, 4, 31, -1; + shfl.sync.bfly.b32 %r834, %r818, 16, 31, -1; +$L__tmp60: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:233:66 ] ] + add.f32 %r835, %r818, %r834; +$L__tmp61: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:233:66 ] + shfl.sync.bfly.b32 %r836, %r835, 8, 31, -1; +$L__tmp62: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:233:66 ] ] + add.f32 %r837, %r835, %r836; +$L__tmp63: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:233:66 ] + shfl.sync.bfly.b32 %r838, %r837, 4, 31, -1; + mov.b64 %rd226, {%r823, %r828}; + mov.b64 %rd227, {%r833, %r838}; +$L__tmp64: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:233:66 ] ] + mov.b64 %rd228, {%r822, %r827}; + mov.b64 %rd229, {%r832, %r837}; + add.f32x2 %rd230, %rd229, %rd227; + add.f32x2 %rd231, %rd228, %rd226; +$L__tmp65: + .loc 1 234 31 // chunk_fwd.py:234:31 + setp.gt.u64 %p59, %rd329, %rd17; + setp.gt.u64 %p60, %rd329, %rd18; + setp.gt.u64 %p61, %rd329, %rd15; + setp.gt.u64 %p62, %rd329, %rd16; + .loc 1 234 41 // chunk_fwd.py:234:41 + mov.b64 {%r839, %r840}, %rd231; + selp.f32 %r841, %r840, 0f00000000, %p62; + selp.f32 %r842, %r839, 0f00000000, %p61; + mov.b64 %rd232, {%r842, %r841}; + mov.b64 {%r843, %r844}, %rd230; + selp.f32 %r845, %r844, 0f00000000, %p60; + selp.f32 %r846, %r843, 0f00000000, %p59; + mov.b64 %rd233, {%r846, %r845}; + .loc 1 235 37 // chunk_fwd.py:235:37 + selp.f32 %r847, %r842, %r841, %p56; + selp.f32 %r848, %r841, %r842, %p56; + selp.f32 %r849, %r846, %r845, %p56; + selp.f32 %r850, %r845, %r846, %p56; + shfl.sync.idx.b32 %r851, %r847, %r44, 31, -1; + shfl.sync.idx.b32 %r852, %r849, %r44, 31, -1; + shfl.sync.idx.b32 %r853, %r848, %r45, 31, -1; + shfl.sync.idx.b32 %r854, %r850, %r45, 31, -1; + selp.b32 %r855, %r851, %r853, %p56; + selp.b32 %r856, %r852, %r854, %p56; + .loc 1 235 48 // chunk_fwd.py:235:48 + mov.b64 {%r857, %r858}, %rd330; + mul.f32 %r859, %r1913, %r856; + mul.f32 %r860, %r1912, %r856; + mov.b64 {%r861, %r862}, %rd331; + mul.f32 %r863, %r861, %r856; + mul.f32 %r864, %r862, %r856; +$L__tmp66: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:235:56 ] ] + fma.rn.f32 %r865, %r857, %r855, %r859; + fma.rn.f32 %r866, %r858, %r855, %r860; + fma.rn.f32 %r867, %r1911, %r855, %r863; + fma.rn.f32 %r868, %r1910, %r855, %r864; +$L__tmp67: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:235:56 ] + shfl.sync.bfly.b32 %r869, %r865, 16, 31, -1; +$L__tmp68: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:235:56 ] ] + add.f32 %r870, %r865, %r869; +$L__tmp69: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:235:56 ] + shfl.sync.bfly.b32 %r871, %r870, 8, 31, -1; +$L__tmp70: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:235:56 ] ] + add.f32 %r872, %r870, %r871; +$L__tmp71: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:235:56 ] + shfl.sync.bfly.b32 %r873, %r872, 4, 31, -1; + shfl.sync.bfly.b32 %r874, %r866, 16, 31, -1; +$L__tmp72: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:235:56 ] ] + add.f32 %r875, %r866, %r874; +$L__tmp73: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:235:56 ] + shfl.sync.bfly.b32 %r876, %r875, 8, 31, -1; +$L__tmp74: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:235:56 ] ] + add.f32 %r877, %r875, %r876; +$L__tmp75: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:235:56 ] + shfl.sync.bfly.b32 %r878, %r877, 4, 31, -1; + shfl.sync.bfly.b32 %r879, %r867, 16, 31, -1; +$L__tmp76: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:235:56 ] ] + add.f32 %r880, %r867, %r879; +$L__tmp77: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:235:56 ] + shfl.sync.bfly.b32 %r881, %r880, 8, 31, -1; +$L__tmp78: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:235:56 ] ] + add.f32 %r882, %r880, %r881; +$L__tmp79: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:235:56 ] + shfl.sync.bfly.b32 %r883, %r882, 4, 31, -1; + shfl.sync.bfly.b32 %r884, %r868, 16, 31, -1; +$L__tmp80: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:235:56 ] ] + add.f32 %r885, %r868, %r884; +$L__tmp81: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:235:56 ] + shfl.sync.bfly.b32 %r886, %r885, 8, 31, -1; +$L__tmp82: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:235:56 ] ] + add.f32 %r887, %r885, %r886; +$L__tmp83: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:235:56 ] + shfl.sync.bfly.b32 %r888, %r887, 4, 31, -1; + mov.b64 %rd234, {%r883, %r888}; + mov.b64 %rd235, {%r873, %r878}; +$L__tmp84: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:235:56 ] ] + mov.b64 %rd236, {%r882, %r887}; + mov.b64 %rd237, {%r872, %r877}; + add.f32x2 %rd238, %rd237, %rd235; + add.f32x2 %rd239, %rd236, %rd234; +$L__tmp85: + .loc 1 235 24 // chunk_fwd.py:235:24 + add.f32x2 %rd240, %rd233, %rd239; + add.f32x2 %rd241, %rd232, %rd238; + .loc 1 236 54 // chunk_fwd.py:236:54 + mov.b64 {%r889, %r890}, %rd241; + selp.f32 %r1913, %r889, %r1913, %p58; + selp.f32 %r1912, %r890, %r1912, %p58; + mov.b64 {%r891, %r892}, %rd240; + selp.f32 %r1911, %r891, %r1911, %p57; + selp.f32 %r1910, %r892, %r1910, %p57; + selp.f32 %r893, %r890, %r858, %p57; + selp.f32 %r894, %r889, %r857, %p57; + mov.b64 %rd330, {%r894, %r893}; + selp.f32 %r895, %r892, %r862, %p58; + selp.f32 %r896, %r891, %r861, %p58; + mov.b64 %rd331, {%r896, %r895}; + .loc 1 232 22 // chunk_fwd.py:232:22 + add.s64 %rd329, %rd329, 1; + setp.ne.b64 %p63, %rd58, %rd329; + @%p63 bra $L__BB0_13; +// %bb.14: // %._crit_edge504.loopexit + .loc 1 249 14 // chunk_fwd.py:249:14 + add.f32 %r1914, %r1913, 0f00000000; + add.f32 %r1915, %r1912, 0f00000000; + add.f32 %r1916, %r1911, 0f00000000; + add.f32 %r1917, %r1910, 0f00000000; +$L__BB0_15: // %._crit_edge504 + .loc 1 0 14 // chunk_fwd.py:0:14 + shl.b64 %rd181, %rd7, 7; + cvt.u64.u32 %rd182, %r320; + mov.b64 %rd189, {%r342, %r341}; + mov.b64 %rd190, {%r344, %r343}; + mul.f32 %r500, %r1862, %r485; + mul.f32 %r501, %r1863, %r487; + mul.f32 %r502, %r1864, %r489; + mul.f32 %r503, %r1865, %r491; + mul.f32 %r504, %r1866, %r493; + mul.f32 %r505, %r1867, %r495; + mul.f32 %r506, %r1868, %r497; + mul.f32 %r507, %r1869, %r499; + mul.f32 %r540, %r1870, %r525; + mul.f32 %r541, %r1871, %r527; + mul.f32 %r542, %r1872, %r529; + mul.f32 %r543, %r1873, %r531; + mul.f32 %r544, %r1874, %r533; + mul.f32 %r545, %r1875, %r535; + mul.f32 %r546, %r1876, %r537; + mul.f32 %r547, %r1877, %r539; + mul.f32 %r580, %r1878, %r565; + mul.f32 %r581, %r1879, %r567; + mul.f32 %r582, %r1880, %r569; + mul.f32 %r583, %r1881, %r571; + mul.f32 %r584, %r1882, %r573; + mul.f32 %r585, %r1883, %r575; + mul.f32 %r586, %r1884, %r577; + mul.f32 %r587, %r1885, %r579; + mul.f32 %r620, %r1886, %r605; + mul.f32 %r621, %r1887, %r607; + mul.f32 %r622, %r1888, %r609; + mul.f32 %r623, %r1889, %r611; + mul.f32 %r624, %r1890, %r613; + mul.f32 %r625, %r1891, %r615; + mul.f32 %r626, %r1892, %r617; + mul.f32 %r627, %r1893, %r619; + mul.f32 %r660, %r1894, %r645; + mul.f32 %r661, %r1895, %r647; + mul.f32 %r662, %r1896, %r649; + mul.f32 %r663, %r1897, %r651; + mul.f32 %r664, %r1898, %r653; + mul.f32 %r665, %r1899, %r655; + mul.f32 %r666, %r1900, %r657; + mul.f32 %r667, %r1901, %r659; + mul.f32 %r700, %r1902, %r685; + mul.f32 %r701, %r1903, %r687; + mul.f32 %r702, %r1904, %r689; + mul.f32 %r703, %r1905, %r691; + mul.f32 %r704, %r1906, %r693; + mul.f32 %r705, %r1907, %r695; + mul.f32 %r706, %r1908, %r697; + mul.f32 %r707, %r1909, %r699; + neg.f32 %r740, %r464; + neg.f32 %r741, %r465; + neg.f32 %r742, %r466; + neg.f32 %r743, %r467; + mov.b64 %rd216, {%r747, %r746}; + mov.b64 %rd217, {%r751, %r750}; + fma.rn.f32 %r1923, %r728, %r336, 0f00000000; + fma.rn.f32 %r1924, %r729, %r336, 0f00000000; + fma.rn.f32 %r1925, %r730, %r338, 0f00000000; + fma.rn.f32 %r1926, %r731, %r338, 0f00000000; + fma.rn.f32x2 %rd333, %rd214, %rd188, %rd193; + fma.rn.f32x2 %rd334, %rd215, %rd187, %rd193; + .loc 1 237 34 // chunk_fwd.py:237:34 + sub.s64 %rd59, %rd2, %rd5; + and.b32 %r1922, %r1, 4; + .loc 1 237 22 // chunk_fwd.py:237:22 + setp.gt.s64 %p64, %rd59, 2; + @%p64 bra $L__BB0_16; + bra.uni $L__BB0_19; +$L__BB0_16: // %.lr.ph522 + .loc 1 0 22 // chunk_fwd.py:0:22 + and.b32 %r897, %r28, 24; + bfe.u32 %r898, %r1, 3, 2; + or.b32 %r899, %r898, %r897; + or.b32 %r50, %r899, %r1922; + xor.b32 %r51, %r50, 4; + .loc 1 237 22 // chunk_fwd.py:237:22 + min.u64 %rd60, %rd59, 16; + .loc 1 238 61 // chunk_fwd.py:238:61 + mov.b64 {%r52, %r53}, %rd333; + mov.b64 {%r54, %r55}, %rd334; + mov.b64 %rd332, 2; + setp.eq.b32 %p65, %r1922, 0; + mov.b32 %r1918, %r1926; + mov.b32 %r1919, %r1925; + mov.b32 %r1920, %r1924; + mov.b32 %r1921, %r1923; +$L__BB0_17: // =>This Inner Loop Header: Depth=1 + .loc 1 238 40 // chunk_fwd.py:238:40 + setp.eq.b64 %p66, %rd29, %rd332; + setp.eq.b64 %p67, %rd27, %rd332; + .loc 1 238 61 // chunk_fwd.py:238:61 + selp.f32 %r900, %r52, 0f00000000, %p66; + selp.f32 %r901, %r53, 0f00000000, %p66; + selp.f32 %r902, %r1923, 0f00000000, %p67; + selp.f32 %r903, %r1924, 0f00000000, %p67; + selp.f32 %r904, %r1925, 0f00000000, %p66; + selp.f32 %r905, %r1926, 0f00000000, %p66; + selp.f32 %r906, %r54, 0f00000000, %p67; + selp.f32 %r907, %r55, 0f00000000, %p67; +$L__tmp86: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:238:66 ] ] + add.f32 %r908, %r900, %r902; + add.f32 %r909, %r901, %r903; + add.f32 %r910, %r904, %r906; + add.f32 %r911, %r905, %r907; +$L__tmp87: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:238:66 ] + shfl.sync.bfly.b32 %r912, %r908, 16, 31, -1; +$L__tmp88: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:238:66 ] ] + add.f32 %r913, %r908, %r912; +$L__tmp89: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:238:66 ] + shfl.sync.bfly.b32 %r914, %r913, 8, 31, -1; +$L__tmp90: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:238:66 ] ] + add.f32 %r915, %r913, %r914; +$L__tmp91: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:238:66 ] + shfl.sync.bfly.b32 %r916, %r915, 4, 31, -1; + shfl.sync.bfly.b32 %r917, %r909, 16, 31, -1; +$L__tmp92: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:238:66 ] ] + add.f32 %r918, %r909, %r917; +$L__tmp93: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:238:66 ] + shfl.sync.bfly.b32 %r919, %r918, 8, 31, -1; +$L__tmp94: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:238:66 ] ] + add.f32 %r920, %r918, %r919; +$L__tmp95: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:238:66 ] + shfl.sync.bfly.b32 %r921, %r920, 4, 31, -1; + shfl.sync.bfly.b32 %r922, %r910, 16, 31, -1; +$L__tmp96: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:238:66 ] ] + add.f32 %r923, %r910, %r922; +$L__tmp97: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:238:66 ] + shfl.sync.bfly.b32 %r924, %r923, 8, 31, -1; +$L__tmp98: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:238:66 ] ] + add.f32 %r925, %r923, %r924; +$L__tmp99: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:238:66 ] + shfl.sync.bfly.b32 %r926, %r925, 4, 31, -1; + shfl.sync.bfly.b32 %r927, %r911, 16, 31, -1; +$L__tmp100: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:238:66 ] ] + add.f32 %r928, %r911, %r927; +$L__tmp101: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:238:66 ] + shfl.sync.bfly.b32 %r929, %r928, 8, 31, -1; +$L__tmp102: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:238:66 ] ] + add.f32 %r930, %r928, %r929; +$L__tmp103: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:238:66 ] + shfl.sync.bfly.b32 %r931, %r930, 4, 31, -1; + mov.b64 %rd242, {%r916, %r921}; + mov.b64 %rd243, {%r926, %r931}; +$L__tmp104: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:238:66 ] ] + mov.b64 %rd244, {%r915, %r920}; + mov.b64 %rd245, {%r925, %r930}; + add.f32x2 %rd246, %rd245, %rd243; + add.f32x2 %rd247, %rd244, %rd242; +$L__tmp105: + .loc 1 239 31 // chunk_fwd.py:239:31 + setp.gt.u64 %p68, %rd332, %rd17; + setp.gt.u64 %p69, %rd332, %rd18; + setp.gt.u64 %p70, %rd332, %rd15; + setp.gt.u64 %p71, %rd332, %rd16; + .loc 1 239 41 // chunk_fwd.py:239:41 + mov.b64 {%r932, %r933}, %rd247; + selp.f32 %r934, %r933, 0f00000000, %p71; + selp.f32 %r935, %r932, 0f00000000, %p70; + mov.b64 %rd248, {%r935, %r934}; + mov.b64 {%r936, %r937}, %rd246; + selp.f32 %r938, %r937, 0f00000000, %p69; + selp.f32 %r939, %r936, 0f00000000, %p68; + mov.b64 %rd249, {%r939, %r938}; + .loc 1 240 37 // chunk_fwd.py:240:37 + selp.f32 %r940, %r935, %r934, %p65; + selp.f32 %r941, %r934, %r935, %p65; + selp.f32 %r942, %r939, %r938, %p65; + selp.f32 %r943, %r938, %r939, %p65; + shfl.sync.idx.b32 %r944, %r940, %r50, 31, -1; + shfl.sync.idx.b32 %r945, %r942, %r50, 31, -1; + shfl.sync.idx.b32 %r946, %r941, %r51, 31, -1; + shfl.sync.idx.b32 %r947, %r943, %r51, 31, -1; + selp.b32 %r948, %r944, %r946, %p65; + selp.b32 %r949, %r945, %r947, %p65; + .loc 1 240 48 // chunk_fwd.py:240:48 + mov.b64 {%r950, %r951}, %rd333; + mul.f32 %r952, %r1921, %r949; + mul.f32 %r953, %r1920, %r949; + mov.b64 {%r954, %r955}, %rd334; + mul.f32 %r956, %r954, %r949; + mul.f32 %r957, %r955, %r949; +$L__tmp106: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:240:56 ] ] + fma.rn.f32 %r958, %r950, %r948, %r952; + fma.rn.f32 %r959, %r951, %r948, %r953; + fma.rn.f32 %r960, %r1919, %r948, %r956; + fma.rn.f32 %r961, %r1918, %r948, %r957; +$L__tmp107: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:240:56 ] + shfl.sync.bfly.b32 %r962, %r958, 16, 31, -1; +$L__tmp108: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:240:56 ] ] + add.f32 %r963, %r958, %r962; +$L__tmp109: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:240:56 ] + shfl.sync.bfly.b32 %r964, %r963, 8, 31, -1; +$L__tmp110: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:240:56 ] ] + add.f32 %r965, %r963, %r964; +$L__tmp111: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:240:56 ] + shfl.sync.bfly.b32 %r966, %r965, 4, 31, -1; + shfl.sync.bfly.b32 %r967, %r959, 16, 31, -1; +$L__tmp112: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:240:56 ] ] + add.f32 %r968, %r959, %r967; +$L__tmp113: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:240:56 ] + shfl.sync.bfly.b32 %r969, %r968, 8, 31, -1; +$L__tmp114: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:240:56 ] ] + add.f32 %r970, %r968, %r969; +$L__tmp115: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:240:56 ] + shfl.sync.bfly.b32 %r971, %r970, 4, 31, -1; + shfl.sync.bfly.b32 %r972, %r960, 16, 31, -1; +$L__tmp116: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:240:56 ] ] + add.f32 %r973, %r960, %r972; +$L__tmp117: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:240:56 ] + shfl.sync.bfly.b32 %r974, %r973, 8, 31, -1; +$L__tmp118: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:240:56 ] ] + add.f32 %r975, %r973, %r974; +$L__tmp119: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:240:56 ] + shfl.sync.bfly.b32 %r976, %r975, 4, 31, -1; + shfl.sync.bfly.b32 %r977, %r961, 16, 31, -1; +$L__tmp120: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:240:56 ] ] + add.f32 %r978, %r961, %r977; +$L__tmp121: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:240:56 ] + shfl.sync.bfly.b32 %r979, %r978, 8, 31, -1; +$L__tmp122: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:240:56 ] ] + add.f32 %r980, %r978, %r979; +$L__tmp123: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:240:56 ] + shfl.sync.bfly.b32 %r981, %r980, 4, 31, -1; + mov.b64 %rd250, {%r976, %r981}; + mov.b64 %rd251, {%r966, %r971}; +$L__tmp124: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:240:56 ] ] + mov.b64 %rd252, {%r975, %r980}; + mov.b64 %rd253, {%r965, %r970}; + add.f32x2 %rd254, %rd253, %rd251; + add.f32x2 %rd255, %rd252, %rd250; +$L__tmp125: + .loc 1 240 24 // chunk_fwd.py:240:24 + add.f32x2 %rd256, %rd249, %rd255; + add.f32x2 %rd257, %rd248, %rd254; + .loc 1 241 54 // chunk_fwd.py:241:54 + mov.b64 {%r982, %r983}, %rd257; + selp.f32 %r1921, %r982, %r1921, %p67; + selp.f32 %r1920, %r983, %r1920, %p67; + mov.b64 {%r984, %r985}, %rd256; + selp.f32 %r1919, %r984, %r1919, %p66; + selp.f32 %r1918, %r985, %r1918, %p66; + selp.f32 %r986, %r983, %r951, %p66; + selp.f32 %r987, %r982, %r950, %p66; + mov.b64 %rd333, {%r987, %r986}; + selp.f32 %r988, %r985, %r955, %p67; + selp.f32 %r989, %r984, %r954, %p67; + mov.b64 %rd334, {%r989, %r988}; + .loc 1 237 22 // chunk_fwd.py:237:22 + add.s64 %rd332, %rd332, 1; + setp.ne.b64 %p72, %rd60, %rd332; + @%p72 bra $L__BB0_17; +// %bb.18: // %._crit_edge523.loopexit + .loc 1 250 14 // chunk_fwd.py:250:14 + add.f32 %r1923, %r1921, 0f00000000; + add.f32 %r1924, %r1920, 0f00000000; + add.f32 %r1925, %r1919, 0f00000000; + add.f32 %r1926, %r1918, 0f00000000; +$L__BB0_19: // %._crit_edge523 + .loc 1 0 14 // chunk_fwd.py:0:14 + add.s64 %rd50, %rd66, %rd181; + shl.b32 %r29, %r3, 2; + or.b64 %rd51, %rd3, %rd182; + or.b64 %rd52, %rd4, %rd182; + or.b64 %rd53, %rd5, %rd182; + or.b64 %rd54, %rd6, %rd182; + mul.f32 %r30, %r500, %r332; + mul.f32 %r31, %r501, %r332; + mul.f32 %r32, %r502, %r330; + mul.f32 %r33, %r503, %r330; + mul.f32 %r34, %r504, %r332; + mul.f32 %r35, %r505, %r332; + mul.f32 %r36, %r506, %r330; + mul.f32 %r37, %r507, %r330; + mul.f32 %r1219, %r540, %r338; + mul.f32 %r1231, %r541, %r338; + mul.f32 %r1220, %r542, %r336; + mul.f32 %r1232, %r543, %r336; + mul.f32 %r1221, %r544, %r338; + mul.f32 %r1233, %r545, %r338; + mul.f32 %r1222, %r546, %r336; + mul.f32 %r1234, %r547, %r336; + mul.f32 %r1243, %r580, %r338; + mul.f32 %r1251, %r581, %r338; + mul.f32 %r1244, %r582, %r336; + mul.f32 %r1252, %r583, %r336; + mul.f32 %r1245, %r584, %r338; + mul.f32 %r1253, %r585, %r338; + mul.f32 %r1246, %r586, %r336; + mul.f32 %r1254, %r587, %r336; + mul.f32 %r1347, %r620, %r344; + mul.f32 %r1355, %r621, %r344; + mul.f32 %r1348, %r622, %r342; + mul.f32 %r1356, %r623, %r342; + mul.f32 %r1349, %r624, %r344; + mul.f32 %r1357, %r625, %r344; + mul.f32 %r1350, %r626, %r342; + mul.f32 %r1358, %r627, %r342; + mul.f32 %r1283, %r660, %r344; + mul.f32 %r1295, %r661, %r344; + mul.f32 %r1284, %r662, %r342; + mul.f32 %r1296, %r663, %r342; + mul.f32 %r1285, %r664, %r344; + mul.f32 %r1297, %r665, %r344; + mul.f32 %r1286, %r666, %r342; + mul.f32 %r1298, %r667, %r342; + mul.f32 %r1307, %r700, %r344; + mul.f32 %r1315, %r701, %r344; + mul.f32 %r1308, %r702, %r342; + mul.f32 %r1316, %r703, %r342; + mul.f32 %r1309, %r704, %r344; + mul.f32 %r1317, %r705, %r344; + mul.f32 %r1310, %r706, %r342; + mul.f32 %r1318, %r707, %r342; + fma.rn.f32 %r1933, %r740, %r342, 0f00000000; + fma.rn.f32 %r1934, %r741, %r342, 0f00000000; + fma.rn.f32 %r1935, %r742, %r344, 0f00000000; + fma.rn.f32 %r1936, %r743, %r344, 0f00000000; + fma.rn.f32x2 %rd336, %rd216, %rd190, %rd193; + fma.rn.f32x2 %rd337, %rd217, %rd189, %rd193; + .loc 1 242 34 // chunk_fwd.py:242:34 + sub.s64 %rd61, %rd2, %rd6; + .loc 1 242 22 // chunk_fwd.py:242:22 + setp.gt.s64 %p73, %rd61, 2; + setp.eq.b32 %p95, %r1922, 0; + and.b32 %r1932, %r28, 24; + bfe.u32 %r1931, %r1, 3, 2; + @%p73 bra $L__BB0_20; + bra.uni $L__BB0_23; +$L__BB0_20: // %.lr.ph541 + .loc 1 0 22 // chunk_fwd.py:0:22 + or.b32 %r990, %r1931, %r1932; + or.b32 %r56, %r990, %r1922; + xor.b32 %r57, %r56, 4; + .loc 1 242 22 // chunk_fwd.py:242:22 + min.u64 %rd62, %rd61, 16; + .loc 1 243 61 // chunk_fwd.py:243:61 + mov.b64 {%r58, %r59}, %rd336; + mov.b64 {%r60, %r61}, %rd337; + mov.b64 %rd335, 2; + mov.b32 %r1927, %r1936; + mov.b32 %r1928, %r1935; + mov.b32 %r1929, %r1934; + mov.b32 %r1930, %r1933; +$L__BB0_21: // =>This Inner Loop Header: Depth=1 + .loc 1 243 40 // chunk_fwd.py:243:40 + setp.eq.b64 %p74, %rd29, %rd335; + setp.eq.b64 %p75, %rd27, %rd335; + .loc 1 243 61 // chunk_fwd.py:243:61 + selp.f32 %r991, %r58, 0f00000000, %p74; + selp.f32 %r992, %r59, 0f00000000, %p74; + selp.f32 %r993, %r1933, 0f00000000, %p75; + selp.f32 %r994, %r1934, 0f00000000, %p75; + selp.f32 %r995, %r1935, 0f00000000, %p74; + selp.f32 %r996, %r1936, 0f00000000, %p74; + selp.f32 %r997, %r60, 0f00000000, %p75; + selp.f32 %r998, %r61, 0f00000000, %p75; +$L__tmp126: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:243:66 ] ] + add.f32 %r999, %r991, %r993; + add.f32 %r1000, %r992, %r994; + add.f32 %r1001, %r995, %r997; + add.f32 %r1002, %r996, %r998; +$L__tmp127: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:243:66 ] + shfl.sync.bfly.b32 %r1003, %r999, 16, 31, -1; +$L__tmp128: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:243:66 ] ] + add.f32 %r1004, %r999, %r1003; +$L__tmp129: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:243:66 ] + shfl.sync.bfly.b32 %r1005, %r1004, 8, 31, -1; +$L__tmp130: + .loc 1 244 31 // chunk_fwd.py:244:31 + setp.gt.u64 %p76, %rd335, %rd17; + setp.gt.u64 %p77, %rd335, %rd18; + setp.gt.u64 %p78, %rd335, %rd15; + setp.gt.u64 %p79, %rd335, %rd16; +$L__tmp131: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:243:66 ] ] + add.f32 %r1006, %r1004, %r1005; +$L__tmp132: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:243:66 ] + shfl.sync.bfly.b32 %r1007, %r1006, 4, 31, -1; + shfl.sync.bfly.b32 %r1008, %r1000, 16, 31, -1; +$L__tmp133: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:243:66 ] ] + add.f32 %r1009, %r1000, %r1008; +$L__tmp134: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:243:66 ] + shfl.sync.bfly.b32 %r1010, %r1009, 8, 31, -1; +$L__tmp135: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:243:66 ] ] + add.f32 %r1011, %r1009, %r1010; +$L__tmp136: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:243:66 ] + shfl.sync.bfly.b32 %r1012, %r1011, 4, 31, -1; + shfl.sync.bfly.b32 %r1013, %r1001, 16, 31, -1; +$L__tmp137: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:243:66 ] ] + add.f32 %r1014, %r1001, %r1013; +$L__tmp138: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:243:66 ] + shfl.sync.bfly.b32 %r1015, %r1014, 8, 31, -1; +$L__tmp139: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:243:66 ] ] + add.f32 %r1016, %r1014, %r1015; +$L__tmp140: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:243:66 ] + shfl.sync.bfly.b32 %r1017, %r1016, 4, 31, -1; + shfl.sync.bfly.b32 %r1018, %r1002, 16, 31, -1; +$L__tmp141: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:243:66 ] ] + add.f32 %r1019, %r1002, %r1018; +$L__tmp142: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:243:66 ] + shfl.sync.bfly.b32 %r1020, %r1019, 8, 31, -1; +$L__tmp143: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:243:66 ] ] + add.f32 %r1021, %r1019, %r1020; +$L__tmp144: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:243:66 ] + shfl.sync.bfly.b32 %r1022, %r1021, 4, 31, -1; + mov.b64 %rd258, {%r1007, %r1012}; + mov.b64 %rd259, {%r1017, %r1022}; +$L__tmp145: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:243:66 ] ] + mov.b64 %rd260, {%r1006, %r1011}; + mov.b64 %rd261, {%r1016, %r1021}; + add.f32x2 %rd262, %rd261, %rd259; + add.f32x2 %rd263, %rd260, %rd258; +$L__tmp146: + .loc 1 244 41 // chunk_fwd.py:244:41 + mov.b64 {%r1023, %r1024}, %rd263; + selp.f32 %r1025, %r1024, 0f00000000, %p79; + selp.f32 %r1026, %r1023, 0f00000000, %p78; + mov.b64 %rd264, {%r1026, %r1025}; + mov.b64 {%r1027, %r1028}, %rd262; + selp.f32 %r1029, %r1028, 0f00000000, %p77; + selp.f32 %r1030, %r1027, 0f00000000, %p76; + mov.b64 %rd265, {%r1030, %r1029}; + .loc 1 245 37 // chunk_fwd.py:245:37 + selp.f32 %r1031, %r1026, %r1025, %p95; + selp.f32 %r1032, %r1025, %r1026, %p95; + selp.f32 %r1033, %r1030, %r1029, %p95; + selp.f32 %r1034, %r1029, %r1030, %p95; + shfl.sync.idx.b32 %r1035, %r1031, %r56, 31, -1; + shfl.sync.idx.b32 %r1036, %r1033, %r56, 31, -1; + shfl.sync.idx.b32 %r1037, %r1032, %r57, 31, -1; + shfl.sync.idx.b32 %r1038, %r1034, %r57, 31, -1; + selp.b32 %r1039, %r1035, %r1037, %p95; + selp.b32 %r1040, %r1036, %r1038, %p95; + .loc 1 245 48 // chunk_fwd.py:245:48 + mov.b64 {%r1041, %r1042}, %rd336; + mul.f32 %r1043, %r1930, %r1040; + mul.f32 %r1044, %r1929, %r1040; + mov.b64 {%r1045, %r1046}, %rd337; + mul.f32 %r1047, %r1045, %r1040; + mul.f32 %r1048, %r1046, %r1040; +$L__tmp147: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:245:56 ] ] + fma.rn.f32 %r1049, %r1041, %r1039, %r1043; + fma.rn.f32 %r1050, %r1042, %r1039, %r1044; + fma.rn.f32 %r1051, %r1928, %r1039, %r1047; + fma.rn.f32 %r1052, %r1927, %r1039, %r1048; +$L__tmp148: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:245:56 ] + shfl.sync.bfly.b32 %r1053, %r1049, 16, 31, -1; +$L__tmp149: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:245:56 ] ] + add.f32 %r1054, %r1049, %r1053; +$L__tmp150: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:245:56 ] + shfl.sync.bfly.b32 %r1055, %r1054, 8, 31, -1; +$L__tmp151: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:245:56 ] ] + add.f32 %r1056, %r1054, %r1055; +$L__tmp152: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:245:56 ] + shfl.sync.bfly.b32 %r1057, %r1056, 4, 31, -1; + shfl.sync.bfly.b32 %r1058, %r1050, 16, 31, -1; +$L__tmp153: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:245:56 ] ] + add.f32 %r1059, %r1050, %r1058; +$L__tmp154: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:245:56 ] + shfl.sync.bfly.b32 %r1060, %r1059, 8, 31, -1; +$L__tmp155: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:245:56 ] ] + add.f32 %r1061, %r1059, %r1060; +$L__tmp156: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:245:56 ] + shfl.sync.bfly.b32 %r1062, %r1061, 4, 31, -1; + shfl.sync.bfly.b32 %r1063, %r1051, 16, 31, -1; +$L__tmp157: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:245:56 ] ] + add.f32 %r1064, %r1051, %r1063; +$L__tmp158: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:245:56 ] + shfl.sync.bfly.b32 %r1065, %r1064, 8, 31, -1; +$L__tmp159: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:245:56 ] ] + add.f32 %r1066, %r1064, %r1065; +$L__tmp160: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:245:56 ] + shfl.sync.bfly.b32 %r1067, %r1066, 4, 31, -1; + shfl.sync.bfly.b32 %r1068, %r1052, 16, 31, -1; +$L__tmp161: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:245:56 ] ] + add.f32 %r1069, %r1052, %r1068; +$L__tmp162: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:245:56 ] + shfl.sync.bfly.b32 %r1070, %r1069, 8, 31, -1; +$L__tmp163: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:245:56 ] ] + add.f32 %r1071, %r1069, %r1070; +$L__tmp164: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:245:56 ] + shfl.sync.bfly.b32 %r1072, %r1071, 4, 31, -1; + mov.b64 %rd266, {%r1067, %r1072}; + mov.b64 %rd267, {%r1057, %r1062}; +$L__tmp165: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:245:56 ] ] + mov.b64 %rd268, {%r1066, %r1071}; + mov.b64 %rd269, {%r1056, %r1061}; + add.f32x2 %rd270, %rd269, %rd267; + add.f32x2 %rd271, %rd268, %rd266; +$L__tmp166: + .loc 1 245 24 // chunk_fwd.py:245:24 + add.f32x2 %rd272, %rd265, %rd271; + add.f32x2 %rd273, %rd264, %rd270; + .loc 1 246 54 // chunk_fwd.py:246:54 + mov.b64 {%r1073, %r1074}, %rd273; + selp.f32 %r1930, %r1073, %r1930, %p75; + selp.f32 %r1929, %r1074, %r1929, %p75; + mov.b64 {%r1075, %r1076}, %rd272; + selp.f32 %r1928, %r1075, %r1928, %p74; + selp.f32 %r1927, %r1076, %r1927, %p74; + selp.f32 %r1077, %r1074, %r1042, %p74; + selp.f32 %r1078, %r1073, %r1041, %p74; + mov.b64 %rd336, {%r1078, %r1077}; + selp.f32 %r1079, %r1076, %r1046, %p75; + selp.f32 %r1080, %r1075, %r1045, %p75; + mov.b64 %rd337, {%r1080, %r1079}; + .loc 1 242 22 // chunk_fwd.py:242:22 + add.s64 %rd335, %rd335, 1; + setp.ne.b64 %p80, %rd62, %rd335; + @%p80 bra $L__BB0_21; +// %bb.22: // %._crit_edge542.loopexit + .loc 1 251 14 // chunk_fwd.py:251:14 + add.f32 %r1933, %r1930, 0f00000000; + add.f32 %r1934, %r1929, 0f00000000; + add.f32 %r1935, %r1928, 0f00000000; + add.f32 %r1936, %r1927, 0f00000000; +$L__BB0_23: // %._crit_edge542 + .loc 1 180 26 // chunk_fwd.py:180:26 + setp.eq.b32 %p85, %r11, %r7; + setp.eq.b32 %p86, %r11, %r6; + setp.eq.b32 %p87, %r10, %r7; + setp.eq.b32 %p88, %r10, %r6; + setp.eq.b32 %p89, %r11, %r9; + setp.eq.b32 %p90, %r10, %r8; + setp.eq.b32 %p91, %r10, %r5; + .loc 1 95 28 // chunk_fwd.py:95:28 + setp.lt.u64 %p84, %rd54, %rd2; + .loc 1 94 28 // chunk_fwd.py:94:28 + setp.lt.u64 %p83, %rd53, %rd2; + .loc 1 93 28 // chunk_fwd.py:93:28 + setp.lt.u64 %p82, %rd52, %rd2; + .loc 1 92 28 // chunk_fwd.py:92:28 + setp.lt.u64 %p81, %rd51, %rd2; + .loc 1 248 14 // chunk_fwd.py:248:14 + selp.f32 %r1407, 0f3F800000, 0f00000000, %p91; + selp.f32 %r1408, 0f3F800000, 0f00000000, %p90; + selp.f32 %r1409, 0f3F800000, 0f00000000, %p89; + selp.f32 %r1410, 0f3F800000, 0f00000000, %p88; + selp.f32 %r1411, 0f3F800000, 0f00000000, %p87; + selp.f32 %r1412, 0f3F800000, 0f00000000, %p86; + selp.f32 %r1413, 0f3F800000, 0f00000000, %p85; + mov.b64 {%r1414, %r1415}, %rd327; + add.f32 %r1082, %r1414, %r1410; + add.f32 %r1083, %r1415, %r1411; + mov.b64 {%r1416, %r1417}, %rd328; + add.f32 %r1085, %r1416, %r1412; + add.f32 %r1086, %r1417, %r1413; + bar.sync 0; + and.b32 %r1418, %r12, 320; + shl.b32 %r1419, %r3, 3; + bfe.s32 %r1420, %r1, 3, 1; + and.b32 %r1421, %r1420, 160; + xor.b32 %r1422, %r1421, %r4; + add.s32 %r1423, %r108, %r1418; + add.s32 %r1424, %r1423, %r1419; + add.s32 %r1081, %r1424, %r1422; + // begin inline asm + st.shared.v2.b32 [ %r1081 + 0 ], { %r1082, %r1083 }; + // end inline asm + add.s32 %r1084, %r1081, 512; + // begin inline asm + st.shared.v2.b32 [ %r1084 + 0 ], { %r1085, %r1086 }; + // end inline asm + .loc 1 249 14 // chunk_fwd.py:249:14 + mov.b64 {%r1425, %r1426}, %rd330; + add.f32 %r1089, %r1425, %r1407; + add.f32 %r1103, %r1426, %r1408; + mov.b64 {%r1427, %r1428}, %rd331; + add.f32 %r1090, %r1427, %r1407; + add.f32 %r1104, %r1428, %r1409; + .loc 1 250 14 // chunk_fwd.py:250:14 + mov.b64 {%r1429, %r1430}, %rd333; + add.f32 %r1133, %r1429, %r1407; + add.f32 %r1147, %r1430, %r1408; + mov.b64 {%r1431, %r1432}, %rd334; + add.f32 %r1134, %r1431, %r1407; + add.f32 %r1148, %r1432, %r1409; + .loc 1 251 14 // chunk_fwd.py:251:14 + mov.b64 {%r1433, %r1434}, %rd336; + add.f32 %r1177, %r1433, %r1407; + add.f32 %r1191, %r1434, %r1408; + mov.b64 {%r1435, %r1436}, %rd337; + add.f32 %r1178, %r1435, %r1407; + add.f32 %r1192, %r1436, %r1409; + .loc 1 207 20 // chunk_fwd.py:207:20 + selp.f32 %r1437, %r30, %r31, %p95; + selp.f32 %r1438, %r31, %r30, %p95; + selp.f32 %r1439, %r32, %r33, %p95; + selp.f32 %r1440, %r33, %r32, %p95; + selp.f32 %r1441, %r34, %r35, %p95; + selp.f32 %r1442, %r35, %r34, %p95; + selp.f32 %r1443, %r36, %r37, %p95; + selp.f32 %r1444, %r37, %r36, %p95; + or.b32 %r1445, %r1922, %r1932; + or.b32 %r1446, %r1445, %r1931; + shfl.sync.idx.b32 %r1447, %r1437, %r1446, 31, -1; + xor.b32 %r1448, %r1446, 4; + shfl.sync.idx.b32 %r1449, %r1438, %r1448, 31, -1; + shfl.sync.idx.b32 %r1450, %r1439, %r1446, 31, -1; + shfl.sync.idx.b32 %r1451, %r1440, %r1448, 31, -1; + shfl.sync.idx.b32 %r1452, %r1441, %r1446, 31, -1; + shfl.sync.idx.b32 %r1453, %r1442, %r1448, 31, -1; + shfl.sync.idx.b32 %r1454, %r1443, %r1446, 31, -1; + shfl.sync.idx.b32 %r1455, %r1444, %r1448, 31, -1; + selp.b32 %r1097, %r1449, %r1447, %p95; + selp.b32 %r1098, %r1451, %r1450, %p95; + selp.b32 %r1105, %r1453, %r1452, %p95; + selp.b32 %r1106, %r1455, %r1454, %p95; + selp.b32 %r1087, %r1447, %r1449, %p95; + selp.b32 %r1088, %r1450, %r1451, %p95; + selp.b32 %r1091, %r1452, %r1453, %p95; + selp.b32 %r1092, %r1454, %r1455, %p95; + mov.b32 %r1381, 0; + .loc 1 258 23 // chunk_fwd.py:258:23 + mov.b32 %r1093, %r1381; + mov.b32 %r1094, %r1381; + mov.b32 %r1095, %r1381; + mov.b32 %r1096, %r1381; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1093, %r1094, %r1095, %r1096 }, { %r1089, %r1914, %r1916, %r1090 }, { %r1087, %r1088 }, { %r1093, %r1094, %r1095, %r1096 }; + // end inline asm + mov.b32 %r1099, %r1381; + mov.b32 %r1100, %r1381; + mov.b32 %r1101, %r1381; + mov.b32 %r1102, %r1381; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1099, %r1100, %r1101, %r1102 }, { %r1089, %r1914, %r1916, %r1090 }, { %r1091, %r1092 }, { %r1099, %r1100, %r1101, %r1102 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1093, %r1094, %r1095, %r1096 }, { %r1103, %r1915, %r1917, %r1104 }, { %r1097, %r1098 }, { %r1093, %r1094, %r1095, %r1096 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1099, %r1100, %r1101, %r1102 }, { %r1103, %r1915, %r1917, %r1104 }, { %r1105, %r1106 }, { %r1099, %r1100, %r1101, %r1102 }; + // end inline asm + .loc 1 259 8 // chunk_fwd.py:259:8 + bfe.s32 %r1456, %r1, 1, 1; + and.b32 %r1457, %r1, 2; + setp.eq.b32 %p92, %r1457, 0; + selp.f32 %r1458, %r1093, %r1094, %p92; + selp.f32 %r1459, %r1094, %r1093, %p92; + selp.f32 %r1460, %r1095, %r1096, %p92; + selp.f32 %r1461, %r1096, %r1095, %p92; + selp.f32 %r1462, %r1099, %r1100, %p92; + selp.f32 %r1463, %r1100, %r1099, %p92; + selp.f32 %r1464, %r1101, %r1102, %p92; + selp.f32 %r1465, %r1102, %r1101, %p92; + and.b32 %r1466, %r1, 1; + neg.s32 %r1467, %r1466; + setp.eq.b32 %p93, %r1466, 0; + cvt.u16.u32 %rs26, %r1; + and.b16 %rs27, %rs26, 2; + shl.b16 %rs28, %rs26, 1; + shr.u16 %rs29, %rs27, 1; + or.b16 %rs30, %rs28, %rs29; + and.b16 %rs31, %rs30, 3; + cvt.u32.u16 %r1468, %rs31; + or.b32 %r1469, %r2, %r1468; + shfl.sync.idx.b32 %r1470, %r1458, %r1469, 31, -1; + shfl.sync.idx.b32 %r1471, %r1460, %r1469, 31, -1; + xor.b32 %r1472, %r1469, 2; + shfl.sync.idx.b32 %r1473, %r1459, %r1472, 31, -1; + shfl.sync.idx.b32 %r1474, %r1461, %r1472, 31, -1; + shfl.sync.idx.b32 %r1475, %r1462, %r1469, 31, -1; + shfl.sync.idx.b32 %r1476, %r1464, %r1469, 31, -1; + shfl.sync.idx.b32 %r1477, %r1463, %r1472, 31, -1; + shfl.sync.idx.b32 %r1478, %r1465, %r1472, 31, -1; + selp.b32 %r1109, %r1473, %r1470, %p93; + selp.b32 %r1110, %r1474, %r1471, %p93; + selp.b32 %r1121, %r1477, %r1475, %p93; + selp.b32 %r1122, %r1478, %r1476, %p93; + .loc 1 248 14 // chunk_fwd.py:248:14 + bar.sync 0; + shl.b32 %r1479, %r1466, 6; + and.b32 %r1480, %r1456, 160; + or.b32 %r1481, %r1480, %r1479; + or.b32 %r1482, %r1481, %r2; + add.s32 %r1483, %r108, %r1482; + ld.shared.b32 %r1111, [%r1483]; + ld.shared.b32 %r1112, [%r1483+256]; + ld.shared.b32 %r1123, [%r1483+512]; + ld.shared.b32 %r1124, [%r1483+768]; + xor.b32 %r1484, %r1482, 32; + add.s32 %r1485, %r108, %r1484; + ld.shared.b32 %r1113, [%r1485]; + ld.shared.b32 %r1114, [%r1485+256]; + ld.shared.b32 %r1129, [%r1485+512]; + ld.shared.b32 %r1130, [%r1485+768]; + .loc 1 259 8 // chunk_fwd.py:259:8 + selp.b32 %r1107, %r1470, %r1473, %p93; + selp.b32 %r1108, %r1471, %r1474, %p93; + selp.b32 %r1119, %r1475, %r1477, %p93; + selp.b32 %r1120, %r1476, %r1478, %p93; + mov.b32 %r1115, %r1381; + mov.b32 %r1116, %r1381; + mov.b32 %r1117, %r1381; + mov.b32 %r1118, %r1381; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1115, %r1116, %r1117, %r1118 }, { %r1107, %r1108, %r1109, %r1110 }, { %r1111, %r1112 }, { %r1115, %r1116, %r1117, %r1118 }; + // end inline asm + mov.b32 %r1125, %r1381; + mov.b32 %r1126, %r1381; + mov.b32 %r1127, %r1381; + mov.b32 %r1128, %r1381; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1125, %r1126, %r1127, %r1128 }, { %r1107, %r1108, %r1109, %r1110 }, { %r1113, %r1114 }, { %r1125, %r1126, %r1127, %r1128 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1115, %r1116, %r1117, %r1118 }, { %r1119, %r1120, %r1121, %r1122 }, { %r1123, %r1124 }, { %r1115, %r1116, %r1117, %r1118 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1125, %r1126, %r1127, %r1128 }, { %r1119, %r1120, %r1121, %r1122 }, { %r1129, %r1130 }, { %r1125, %r1126, %r1127, %r1128 }; + // end inline asm + mov.b32 %r1486, 0f00000000; + .loc 1 257 14 // chunk_fwd.py:257:14 + sub.f32 %r1487, %r1486, %r1115; + sub.f32 %r1488, %r1486, %r1116; + sub.f32 %r1489, %r1486, %r1117; + sub.f32 %r1490, %r1486, %r1118; + sub.f32 %r1491, %r1486, %r1125; + sub.f32 %r1492, %r1486, %r1126; + sub.f32 %r1493, %r1486, %r1127; + sub.f32 %r1494, %r1486, %r1128; + .loc 1 209 20 // chunk_fwd.py:209:20 + selp.f32 %r1495, %r1243, %r1251, %p95; + selp.f32 %r1496, %r1251, %r1243, %p95; + selp.f32 %r1497, %r1244, %r1252, %p95; + selp.f32 %r1498, %r1252, %r1244, %p95; + selp.f32 %r1499, %r1245, %r1253, %p95; + selp.f32 %r1500, %r1253, %r1245, %p95; + selp.f32 %r1501, %r1246, %r1254, %p95; + selp.f32 %r1502, %r1254, %r1246, %p95; + shfl.sync.idx.b32 %r1503, %r1495, %r1446, 31, -1; + shfl.sync.idx.b32 %r1504, %r1496, %r1448, 31, -1; + shfl.sync.idx.b32 %r1505, %r1497, %r1446, 31, -1; + shfl.sync.idx.b32 %r1506, %r1498, %r1448, 31, -1; + shfl.sync.idx.b32 %r1507, %r1499, %r1446, 31, -1; + shfl.sync.idx.b32 %r1508, %r1500, %r1448, 31, -1; + shfl.sync.idx.b32 %r1509, %r1501, %r1446, 31, -1; + shfl.sync.idx.b32 %r1510, %r1502, %r1448, 31, -1; + selp.b32 %r1141, %r1504, %r1503, %p95; + selp.b32 %r1142, %r1506, %r1505, %p95; + selp.b32 %r1149, %r1508, %r1507, %p95; + selp.b32 %r1150, %r1510, %r1509, %p95; + selp.b32 %r1131, %r1503, %r1504, %p95; + selp.b32 %r1132, %r1505, %r1506, %p95; + selp.b32 %r1135, %r1507, %r1508, %p95; + selp.b32 %r1136, %r1509, %r1510, %p95; + .loc 1 263 23 // chunk_fwd.py:263:23 + mov.b32 %r1137, %r1381; + mov.b32 %r1138, %r1381; + mov.b32 %r1139, %r1381; + mov.b32 %r1140, %r1381; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1137, %r1138, %r1139, %r1140 }, { %r1133, %r1923, %r1925, %r1134 }, { %r1131, %r1132 }, { %r1137, %r1138, %r1139, %r1140 }; + // end inline asm + mov.b32 %r1143, %r1381; + mov.b32 %r1144, %r1381; + mov.b32 %r1145, %r1381; + mov.b32 %r1146, %r1381; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1143, %r1144, %r1145, %r1146 }, { %r1133, %r1923, %r1925, %r1134 }, { %r1135, %r1136 }, { %r1143, %r1144, %r1145, %r1146 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1137, %r1138, %r1139, %r1140 }, { %r1147, %r1924, %r1926, %r1148 }, { %r1141, %r1142 }, { %r1137, %r1138, %r1139, %r1140 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1143, %r1144, %r1145, %r1146 }, { %r1147, %r1924, %r1926, %r1148 }, { %r1149, %r1150 }, { %r1143, %r1144, %r1145, %r1146 }; + // end inline asm + .loc 1 264 8 // chunk_fwd.py:264:8 + selp.f32 %r1511, %r1137, %r1138, %p92; + selp.f32 %r1512, %r1138, %r1137, %p92; + selp.f32 %r1513, %r1139, %r1140, %p92; + selp.f32 %r1514, %r1140, %r1139, %p92; + selp.f32 %r1515, %r1143, %r1144, %p92; + selp.f32 %r1516, %r1144, %r1143, %p92; + selp.f32 %r1517, %r1145, %r1146, %p92; + selp.f32 %r1518, %r1146, %r1145, %p92; + shfl.sync.idx.b32 %r1519, %r1511, %r1469, 31, -1; + shfl.sync.idx.b32 %r1520, %r1513, %r1469, 31, -1; + shfl.sync.idx.b32 %r1521, %r1512, %r1472, 31, -1; + shfl.sync.idx.b32 %r1522, %r1514, %r1472, 31, -1; + shfl.sync.idx.b32 %r1523, %r1515, %r1469, 31, -1; + shfl.sync.idx.b32 %r1524, %r1517, %r1469, 31, -1; + shfl.sync.idx.b32 %r1525, %r1516, %r1472, 31, -1; + shfl.sync.idx.b32 %r1526, %r1518, %r1472, 31, -1; + selp.b32 %r1153, %r1521, %r1519, %p93; + selp.b32 %r1154, %r1522, %r1520, %p93; + selp.b32 %r1165, %r1525, %r1523, %p93; + selp.b32 %r1166, %r1526, %r1524, %p93; + .loc 1 249 14 // chunk_fwd.py:249:14 + setp.eq.b32 %p94, %r17, 0; + selp.f32 %r1527, %r1089, %r1103, %p94; + selp.f32 %r1528, %r1103, %r1089, %p94; + selp.f32 %r1529, %r1914, %r1915, %p94; + selp.f32 %r1530, %r1915, %r1914, %p94; + selp.f32 %r1531, %r1916, %r1917, %p94; + selp.f32 %r1532, %r1917, %r1916, %p94; + selp.f32 %r1533, %r1090, %r1104, %p94; + selp.f32 %r1534, %r1104, %r1090, %p94; + shl.b32 %r1535, %r1, 2; + and.b32 %r1536, %r1535, 28; + or.b32 %r1537, %r1536, %r1931; + shfl.sync.idx.b32 %r1538, %r1527, %r1537, 31, -1; + xor.b32 %r1539, %r1537, 16; + shfl.sync.idx.b32 %r1540, %r1528, %r1539, 31, -1; + shfl.sync.idx.b32 %r1541, %r1529, %r1537, 31, -1; + shfl.sync.idx.b32 %r1542, %r1530, %r1539, 31, -1; + shfl.sync.idx.b32 %r1543, %r1531, %r1537, 31, -1; + shfl.sync.idx.b32 %r1544, %r1532, %r1539, 31, -1; + shfl.sync.idx.b32 %r1545, %r1533, %r1537, 31, -1; + shfl.sync.idx.b32 %r1546, %r1534, %r1539, 31, -1; + selp.b32 %r1156, %r1540, %r1538, %p95; + selp.b32 %r1168, %r1542, %r1541, %p95; + selp.b32 %r1158, %r1544, %r1543, %p95; + selp.b32 %r1174, %r1546, %r1545, %p95; + .loc 1 264 8 // chunk_fwd.py:264:8 + selp.b32 %r1151, %r1519, %r1521, %p93; + selp.b32 %r1152, %r1520, %r1522, %p93; + selp.b32 %r1163, %r1523, %r1525, %p93; + selp.b32 %r1164, %r1524, %r1526, %p93; + .loc 1 249 14 // chunk_fwd.py:249:14 + selp.b32 %r1155, %r1538, %r1540, %p95; + selp.b32 %r1167, %r1541, %r1542, %p95; + selp.b32 %r1157, %r1543, %r1544, %p95; + selp.b32 %r1173, %r1545, %r1546, %p95; + .loc 1 264 8 // chunk_fwd.py:264:8 + mov.b32 %r1159, %r1381; + mov.b32 %r1160, %r1381; + mov.b32 %r1161, %r1381; + mov.b32 %r1162, %r1381; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1159, %r1160, %r1161, %r1162 }, { %r1151, %r1152, %r1153, %r1154 }, { %r1155, %r1156 }, { %r1159, %r1160, %r1161, %r1162 }; + // end inline asm + mov.b32 %r1169, %r1381; + mov.b32 %r1170, %r1381; + mov.b32 %r1171, %r1381; + mov.b32 %r1172, %r1381; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1169, %r1170, %r1171, %r1172 }, { %r1151, %r1152, %r1153, %r1154 }, { %r1157, %r1158 }, { %r1169, %r1170, %r1171, %r1172 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1159, %r1160, %r1161, %r1162 }, { %r1163, %r1164, %r1165, %r1166 }, { %r1167, %r1168 }, { %r1159, %r1160, %r1161, %r1162 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1169, %r1170, %r1171, %r1172 }, { %r1163, %r1164, %r1165, %r1166 }, { %r1173, %r1174 }, { %r1169, %r1170, %r1171, %r1172 }; + // end inline asm + .loc 1 262 14 // chunk_fwd.py:262:14 + sub.f32 %r1547, %r1486, %r1159; + sub.f32 %r1548, %r1486, %r1160; + sub.f32 %r1549, %r1486, %r1161; + sub.f32 %r1550, %r1486, %r1162; + sub.f32 %r1551, %r1486, %r1169; + sub.f32 %r1552, %r1486, %r1170; + sub.f32 %r1553, %r1486, %r1171; + sub.f32 %r1554, %r1486, %r1172; + .loc 1 212 20 // chunk_fwd.py:212:20 + selp.f32 %r1555, %r1307, %r1315, %p95; + selp.f32 %r1556, %r1315, %r1307, %p95; + selp.f32 %r1557, %r1308, %r1316, %p95; + selp.f32 %r1558, %r1316, %r1308, %p95; + selp.f32 %r1559, %r1309, %r1317, %p95; + selp.f32 %r1560, %r1317, %r1309, %p95; + selp.f32 %r1561, %r1310, %r1318, %p95; + selp.f32 %r1562, %r1318, %r1310, %p95; + shfl.sync.idx.b32 %r1563, %r1555, %r1446, 31, -1; + shfl.sync.idx.b32 %r1564, %r1556, %r1448, 31, -1; + shfl.sync.idx.b32 %r1565, %r1557, %r1446, 31, -1; + shfl.sync.idx.b32 %r1566, %r1558, %r1448, 31, -1; + shfl.sync.idx.b32 %r1567, %r1559, %r1446, 31, -1; + shfl.sync.idx.b32 %r1568, %r1560, %r1448, 31, -1; + shfl.sync.idx.b32 %r1569, %r1561, %r1446, 31, -1; + shfl.sync.idx.b32 %r1570, %r1562, %r1448, 31, -1; + selp.b32 %r1185, %r1564, %r1563, %p95; + selp.b32 %r1186, %r1566, %r1565, %p95; + selp.b32 %r1193, %r1568, %r1567, %p95; + selp.b32 %r1194, %r1570, %r1569, %p95; + selp.b32 %r1175, %r1563, %r1564, %p95; + selp.b32 %r1176, %r1565, %r1566, %p95; + selp.b32 %r1179, %r1567, %r1568, %p95; + selp.b32 %r1180, %r1569, %r1570, %p95; + .loc 1 268 23 // chunk_fwd.py:268:23 + mov.b32 %r1181, %r1381; + mov.b32 %r1182, %r1381; + mov.b32 %r1183, %r1381; + mov.b32 %r1184, %r1381; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1181, %r1182, %r1183, %r1184 }, { %r1177, %r1933, %r1935, %r1178 }, { %r1175, %r1176 }, { %r1181, %r1182, %r1183, %r1184 }; + // end inline asm + mov.b32 %r1187, %r1381; + mov.b32 %r1188, %r1381; + mov.b32 %r1189, %r1381; + mov.b32 %r1190, %r1381; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1187, %r1188, %r1189, %r1190 }, { %r1177, %r1933, %r1935, %r1178 }, { %r1179, %r1180 }, { %r1187, %r1188, %r1189, %r1190 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1181, %r1182, %r1183, %r1184 }, { %r1191, %r1934, %r1936, %r1192 }, { %r1185, %r1186 }, { %r1181, %r1182, %r1183, %r1184 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1187, %r1188, %r1189, %r1190 }, { %r1191, %r1934, %r1936, %r1192 }, { %r1193, %r1194 }, { %r1187, %r1188, %r1189, %r1190 }; + // end inline asm + .loc 1 269 8 // chunk_fwd.py:269:8 + selp.f32 %r1571, %r1181, %r1182, %p92; + selp.f32 %r1572, %r1182, %r1181, %p92; + selp.f32 %r1573, %r1183, %r1184, %p92; + selp.f32 %r1574, %r1184, %r1183, %p92; + selp.f32 %r1575, %r1187, %r1188, %p92; + selp.f32 %r1576, %r1188, %r1187, %p92; + selp.f32 %r1577, %r1189, %r1190, %p92; + selp.f32 %r1578, %r1190, %r1189, %p92; + shfl.sync.idx.b32 %r1579, %r1571, %r1469, 31, -1; + shfl.sync.idx.b32 %r1580, %r1573, %r1469, 31, -1; + shfl.sync.idx.b32 %r1581, %r1572, %r1472, 31, -1; + shfl.sync.idx.b32 %r1582, %r1574, %r1472, 31, -1; + shfl.sync.idx.b32 %r1583, %r1575, %r1469, 31, -1; + shfl.sync.idx.b32 %r1584, %r1577, %r1469, 31, -1; + shfl.sync.idx.b32 %r1585, %r1576, %r1472, 31, -1; + shfl.sync.idx.b32 %r1586, %r1578, %r1472, 31, -1; + selp.b32 %r1197, %r1581, %r1579, %p93; + selp.b32 %r1198, %r1582, %r1580, %p93; + selp.b32 %r1209, %r1585, %r1583, %p93; + selp.b32 %r1210, %r1586, %r1584, %p93; + .loc 1 250 14 // chunk_fwd.py:250:14 + selp.f32 %r1587, %r1133, %r1147, %p94; + selp.f32 %r1588, %r1147, %r1133, %p94; + selp.f32 %r1589, %r1923, %r1924, %p94; + selp.f32 %r1590, %r1924, %r1923, %p94; + selp.f32 %r1591, %r1925, %r1926, %p94; + selp.f32 %r1592, %r1926, %r1925, %p94; + selp.f32 %r1593, %r1134, %r1148, %p94; + selp.f32 %r1594, %r1148, %r1134, %p94; + shfl.sync.idx.b32 %r1595, %r1587, %r1537, 31, -1; + shfl.sync.idx.b32 %r1596, %r1588, %r1539, 31, -1; + shfl.sync.idx.b32 %r1597, %r1589, %r1537, 31, -1; + shfl.sync.idx.b32 %r1598, %r1590, %r1539, 31, -1; + shfl.sync.idx.b32 %r1599, %r1591, %r1537, 31, -1; + shfl.sync.idx.b32 %r1600, %r1592, %r1539, 31, -1; + shfl.sync.idx.b32 %r1601, %r1593, %r1537, 31, -1; + shfl.sync.idx.b32 %r1602, %r1594, %r1539, 31, -1; + selp.b32 %r1200, %r1596, %r1595, %p95; + selp.b32 %r1212, %r1598, %r1597, %p95; + selp.b32 %r1202, %r1600, %r1599, %p95; + selp.b32 %r1218, %r1602, %r1601, %p95; + .loc 1 269 8 // chunk_fwd.py:269:8 + selp.b32 %r1195, %r1579, %r1581, %p93; + selp.b32 %r1196, %r1580, %r1582, %p93; + selp.b32 %r1207, %r1583, %r1585, %p93; + selp.b32 %r1208, %r1584, %r1586, %p93; + .loc 1 250 14 // chunk_fwd.py:250:14 + selp.b32 %r1199, %r1595, %r1596, %p95; + selp.b32 %r1211, %r1597, %r1598, %p95; + selp.b32 %r1201, %r1599, %r1600, %p95; + selp.b32 %r1217, %r1601, %r1602, %p95; + .loc 1 269 8 // chunk_fwd.py:269:8 + mov.b32 %r1203, %r1381; + mov.b32 %r1204, %r1381; + mov.b32 %r1205, %r1381; + mov.b32 %r1206, %r1381; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1203, %r1204, %r1205, %r1206 }, { %r1195, %r1196, %r1197, %r1198 }, { %r1199, %r1200 }, { %r1203, %r1204, %r1205, %r1206 }; + // end inline asm + mov.b32 %r1213, %r1381; + mov.b32 %r1214, %r1381; + mov.b32 %r1215, %r1381; + mov.b32 %r1216, %r1381; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1213, %r1214, %r1215, %r1216 }, { %r1195, %r1196, %r1197, %r1198 }, { %r1201, %r1202 }, { %r1213, %r1214, %r1215, %r1216 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1203, %r1204, %r1205, %r1206 }, { %r1207, %r1208, %r1209, %r1210 }, { %r1211, %r1212 }, { %r1203, %r1204, %r1205, %r1206 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1213, %r1214, %r1215, %r1216 }, { %r1207, %r1208, %r1209, %r1210 }, { %r1217, %r1218 }, { %r1213, %r1214, %r1215, %r1216 }; + // end inline asm + .loc 1 248 14 // chunk_fwd.py:248:14 + and.b32 %r1603, %r1467, 160; + shl.b32 %r1604, %r1457, 7; + or.b32 %r1605, %r1603, %r1604; + or.b32 %r1606, %r1605, %r2; + add.s32 %r1607, %r108, %r1606; + ld.shared.b32 %r1223, [%r1607]; + ld.shared.b32 %r1235, [%r1607+64]; + ld.shared.b32 %r1224, [%r1607+512]; + ld.shared.b32 %r1236, [%r1607+576]; + xor.b32 %r1608, %r1606, 32; + add.s32 %r1609, %r108, %r1608; + ld.shared.b32 %r1225, [%r1609]; + ld.shared.b32 %r1241, [%r1609+64]; + ld.shared.b32 %r1226, [%r1609+512]; + ld.shared.b32 %r1242, [%r1609+576]; + .loc 1 275 22 // chunk_fwd.py:275:22 + mov.b32 %r1227, %r1381; + mov.b32 %r1228, %r1381; + mov.b32 %r1229, %r1381; + mov.b32 %r1230, %r1381; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1227, %r1228, %r1229, %r1230 }, { %r1219, %r1220, %r1221, %r1222 }, { %r1223, %r1224 }, { %r1227, %r1228, %r1229, %r1230 }; + // end inline asm + mov.b32 %r1237, %r1381; + mov.b32 %r1238, %r1381; + mov.b32 %r1239, %r1381; + mov.b32 %r1240, %r1381; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1237, %r1238, %r1239, %r1240 }, { %r1219, %r1220, %r1221, %r1222 }, { %r1225, %r1226 }, { %r1237, %r1238, %r1239, %r1240 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1227, %r1228, %r1229, %r1230 }, { %r1231, %r1232, %r1233, %r1234 }, { %r1235, %r1236 }, { %r1227, %r1228, %r1229, %r1230 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1237, %r1238, %r1239, %r1240 }, { %r1231, %r1232, %r1233, %r1234 }, { %r1241, %r1242 }, { %r1237, %r1238, %r1239, %r1240 }; + // end inline asm + .loc 1 257 14 // chunk_fwd.py:257:14 + selp.f32 %r1610, %r1487, %r1488, %p95; + selp.f32 %r1611, %r1488, %r1487, %p95; + selp.f32 %r1612, %r1489, %r1490, %p95; + selp.f32 %r1613, %r1490, %r1489, %p95; + selp.f32 %r1614, %r1491, %r1492, %p95; + selp.f32 %r1615, %r1492, %r1491, %p95; + selp.f32 %r1616, %r1493, %r1494, %p95; + selp.f32 %r1617, %r1494, %r1493, %p95; + shfl.sync.idx.b32 %r1618, %r1610, %r1446, 31, -1; + shfl.sync.idx.b32 %r1619, %r1611, %r1448, 31, -1; + shfl.sync.idx.b32 %r1620, %r1612, %r1446, 31, -1; + shfl.sync.idx.b32 %r1621, %r1613, %r1448, 31, -1; + shfl.sync.idx.b32 %r1622, %r1614, %r1446, 31, -1; + shfl.sync.idx.b32 %r1623, %r1615, %r1448, 31, -1; + shfl.sync.idx.b32 %r1624, %r1616, %r1446, 31, -1; + shfl.sync.idx.b32 %r1625, %r1617, %r1448, 31, -1; + selp.b32 %r1255, %r1619, %r1618, %p95; + selp.b32 %r1256, %r1621, %r1620, %p95; + selp.b32 %r1257, %r1623, %r1622, %p95; + selp.b32 %r1258, %r1625, %r1624, %p95; + selp.b32 %r1247, %r1618, %r1619, %p95; + selp.b32 %r1248, %r1620, %r1621, %p95; + selp.b32 %r1249, %r1622, %r1623, %p95; + selp.b32 %r1250, %r1624, %r1625, %p95; + .loc 1 276 22 // chunk_fwd.py:276:22 + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1227, %r1228, %r1229, %r1230 }, { %r1243, %r1244, %r1245, %r1246 }, { %r1247, %r1248 }, { %r1227, %r1228, %r1229, %r1230 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1237, %r1238, %r1239, %r1240 }, { %r1243, %r1244, %r1245, %r1246 }, { %r1249, %r1250 }, { %r1237, %r1238, %r1239, %r1240 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1227, %r1228, %r1229, %r1230 }, { %r1251, %r1252, %r1253, %r1254 }, { %r1255, %r1256 }, { %r1227, %r1228, %r1229, %r1230 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1237, %r1238, %r1239, %r1240 }, { %r1251, %r1252, %r1253, %r1254 }, { %r1257, %r1258 }, { %r1237, %r1238, %r1239, %r1240 }; + // end inline asm + .loc 1 250 14 // chunk_fwd.py:250:14 + selp.f32 %r1626, %r1133, %r1147, %p92; + selp.f32 %r1627, %r1147, %r1133, %p92; + selp.f32 %r1628, %r1923, %r1924, %p92; + selp.f32 %r1629, %r1924, %r1923, %p92; + selp.f32 %r1630, %r1925, %r1926, %p92; + selp.f32 %r1631, %r1926, %r1925, %p92; + selp.f32 %r1632, %r1134, %r1148, %p92; + selp.f32 %r1633, %r1148, %r1134, %p92; + shfl.sync.idx.b32 %r1634, %r1626, %r1469, 31, -1; + shfl.sync.idx.b32 %r1635, %r1628, %r1469, 31, -1; + shfl.sync.idx.b32 %r1636, %r1627, %r1472, 31, -1; + shfl.sync.idx.b32 %r1637, %r1629, %r1472, 31, -1; + shfl.sync.idx.b32 %r1638, %r1630, %r1469, 31, -1; + shfl.sync.idx.b32 %r1639, %r1632, %r1469, 31, -1; + shfl.sync.idx.b32 %r1640, %r1631, %r1472, 31, -1; + shfl.sync.idx.b32 %r1641, %r1633, %r1472, 31, -1; + selp.b32 %r1261, %r1636, %r1634, %p93; + selp.b32 %r1262, %r1637, %r1635, %p93; + selp.b32 %r1273, %r1640, %r1638, %p93; + selp.b32 %r1274, %r1641, %r1639, %p93; + .loc 1 275 8 // chunk_fwd.py:275:8 + selp.f32 %r1642, %r1227, %r1228, %p94; + selp.f32 %r1643, %r1228, %r1227, %p94; + selp.f32 %r1644, %r1229, %r1230, %p94; + selp.f32 %r1645, %r1230, %r1229, %p94; + selp.f32 %r1646, %r1237, %r1238, %p94; + selp.f32 %r1647, %r1238, %r1237, %p94; + selp.f32 %r1648, %r1239, %r1240, %p94; + selp.f32 %r1649, %r1240, %r1239, %p94; + shfl.sync.idx.b32 %r1650, %r1642, %r1537, 31, -1; + shfl.sync.idx.b32 %r1651, %r1643, %r1539, 31, -1; + shfl.sync.idx.b32 %r1652, %r1644, %r1537, 31, -1; + shfl.sync.idx.b32 %r1653, %r1645, %r1539, 31, -1; + shfl.sync.idx.b32 %r1654, %r1646, %r1537, 31, -1; + shfl.sync.idx.b32 %r1655, %r1647, %r1539, 31, -1; + shfl.sync.idx.b32 %r1656, %r1648, %r1537, 31, -1; + shfl.sync.idx.b32 %r1657, %r1649, %r1539, 31, -1; + selp.b32 %r1264, %r1651, %r1650, %p95; + selp.b32 %r1276, %r1653, %r1652, %p95; + selp.b32 %r1266, %r1655, %r1654, %p95; + selp.b32 %r1282, %r1657, %r1656, %p95; + .loc 1 250 14 // chunk_fwd.py:250:14 + selp.b32 %r1259, %r1634, %r1636, %p93; + selp.b32 %r1260, %r1635, %r1637, %p93; + selp.b32 %r1271, %r1638, %r1640, %p93; + selp.b32 %r1272, %r1639, %r1641, %p93; + .loc 1 275 8 // chunk_fwd.py:275:8 + selp.b32 %r1263, %r1650, %r1651, %p95; + selp.b32 %r1275, %r1652, %r1653, %p95; + selp.b32 %r1265, %r1654, %r1655, %p95; + selp.b32 %r1281, %r1656, %r1657, %p95; + mov.b32 %r1267, %r1381; + mov.b32 %r1268, %r1381; + mov.b32 %r1269, %r1381; + mov.b32 %r1270, %r1381; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1267, %r1268, %r1269, %r1270 }, { %r1259, %r1260, %r1261, %r1262 }, { %r1263, %r1264 }, { %r1267, %r1268, %r1269, %r1270 }; + // end inline asm + mov.b32 %r1277, %r1381; + mov.b32 %r1278, %r1381; + mov.b32 %r1279, %r1381; + mov.b32 %r1280, %r1381; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1277, %r1278, %r1279, %r1280 }, { %r1259, %r1260, %r1261, %r1262 }, { %r1265, %r1266 }, { %r1277, %r1278, %r1279, %r1280 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1267, %r1268, %r1269, %r1270 }, { %r1271, %r1272, %r1273, %r1274 }, { %r1275, %r1276 }, { %r1267, %r1268, %r1269, %r1270 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1277, %r1278, %r1279, %r1280 }, { %r1271, %r1272, %r1273, %r1274 }, { %r1281, %r1282 }, { %r1277, %r1278, %r1279, %r1280 }; + // end inline asm + .loc 1 273 14 // chunk_fwd.py:273:14 + sub.f32 %r1658, %r1486, %r1267; + sub.f32 %r1659, %r1486, %r1268; + sub.f32 %r1660, %r1486, %r1269; + sub.f32 %r1661, %r1486, %r1270; + sub.f32 %r1662, %r1486, %r1277; + sub.f32 %r1663, %r1486, %r1278; + sub.f32 %r1664, %r1486, %r1279; + sub.f32 %r1665, %r1486, %r1280; + .loc 1 249 14 // chunk_fwd.py:249:14 + selp.f32 %r1666, %r1089, %r1103, %p95; + selp.f32 %r1667, %r1103, %r1089, %p95; + selp.f32 %r1668, %r1914, %r1915, %p95; + selp.f32 %r1669, %r1915, %r1914, %p95; + selp.f32 %r1670, %r1916, %r1917, %p95; + selp.f32 %r1671, %r1917, %r1916, %p95; + selp.f32 %r1672, %r1090, %r1104, %p95; + selp.f32 %r1673, %r1104, %r1090, %p95; + shfl.sync.idx.b32 %r1674, %r1666, %r1446, 31, -1; + shfl.sync.idx.b32 %r1675, %r1667, %r1448, 31, -1; + shfl.sync.idx.b32 %r1676, %r1668, %r1446, 31, -1; + shfl.sync.idx.b32 %r1677, %r1669, %r1448, 31, -1; + shfl.sync.idx.b32 %r1678, %r1670, %r1446, 31, -1; + shfl.sync.idx.b32 %r1679, %r1671, %r1448, 31, -1; + shfl.sync.idx.b32 %r1680, %r1672, %r1446, 31, -1; + shfl.sync.idx.b32 %r1681, %r1673, %r1448, 31, -1; + selp.b32 %r1299, %r1675, %r1674, %p95; + selp.b32 %r1300, %r1677, %r1676, %p95; + selp.b32 %r1305, %r1679, %r1678, %p95; + selp.b32 %r1306, %r1681, %r1680, %p95; + selp.b32 %r1287, %r1674, %r1675, %p95; + selp.b32 %r1288, %r1676, %r1677, %p95; + selp.b32 %r1289, %r1678, %r1679, %p95; + selp.b32 %r1290, %r1680, %r1681, %p95; + .loc 1 281 22 // chunk_fwd.py:281:22 + mov.b32 %r1291, %r1381; + mov.b32 %r1292, %r1381; + mov.b32 %r1293, %r1381; + mov.b32 %r1294, %r1381; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1291, %r1292, %r1293, %r1294 }, { %r1283, %r1284, %r1285, %r1286 }, { %r1287, %r1288 }, { %r1291, %r1292, %r1293, %r1294 }; + // end inline asm + mov.b32 %r1301, %r1381; + mov.b32 %r1302, %r1381; + mov.b32 %r1303, %r1381; + mov.b32 %r1304, %r1381; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1301, %r1302, %r1303, %r1304 }, { %r1283, %r1284, %r1285, %r1286 }, { %r1289, %r1290 }, { %r1301, %r1302, %r1303, %r1304 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1291, %r1292, %r1293, %r1294 }, { %r1295, %r1296, %r1297, %r1298 }, { %r1299, %r1300 }, { %r1291, %r1292, %r1293, %r1294 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1301, %r1302, %r1303, %r1304 }, { %r1295, %r1296, %r1297, %r1298 }, { %r1305, %r1306 }, { %r1301, %r1302, %r1303, %r1304 }; + // end inline asm + .loc 1 262 14 // chunk_fwd.py:262:14 + selp.f32 %r1682, %r1547, %r1548, %p95; + selp.f32 %r1683, %r1548, %r1547, %p95; + selp.f32 %r1684, %r1549, %r1550, %p95; + selp.f32 %r1685, %r1550, %r1549, %p95; + selp.f32 %r1686, %r1551, %r1552, %p95; + selp.f32 %r1687, %r1552, %r1551, %p95; + selp.f32 %r1688, %r1553, %r1554, %p95; + selp.f32 %r1689, %r1554, %r1553, %p95; + shfl.sync.idx.b32 %r1690, %r1682, %r1446, 31, -1; + shfl.sync.idx.b32 %r1691, %r1683, %r1448, 31, -1; + shfl.sync.idx.b32 %r1692, %r1684, %r1446, 31, -1; + shfl.sync.idx.b32 %r1693, %r1685, %r1448, 31, -1; + shfl.sync.idx.b32 %r1694, %r1686, %r1446, 31, -1; + shfl.sync.idx.b32 %r1695, %r1687, %r1448, 31, -1; + shfl.sync.idx.b32 %r1696, %r1688, %r1446, 31, -1; + shfl.sync.idx.b32 %r1697, %r1689, %r1448, 31, -1; + selp.b32 %r1319, %r1691, %r1690, %p95; + selp.b32 %r1320, %r1693, %r1692, %p95; + selp.b32 %r1321, %r1695, %r1694, %p95; + selp.b32 %r1322, %r1697, %r1696, %p95; + selp.b32 %r1311, %r1690, %r1691, %p95; + selp.b32 %r1312, %r1692, %r1693, %p95; + selp.b32 %r1313, %r1694, %r1695, %p95; + selp.b32 %r1314, %r1696, %r1697, %p95; + .loc 1 282 22 // chunk_fwd.py:282:22 + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1291, %r1292, %r1293, %r1294 }, { %r1307, %r1308, %r1309, %r1310 }, { %r1311, %r1312 }, { %r1291, %r1292, %r1293, %r1294 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1301, %r1302, %r1303, %r1304 }, { %r1307, %r1308, %r1309, %r1310 }, { %r1313, %r1314 }, { %r1301, %r1302, %r1303, %r1304 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1291, %r1292, %r1293, %r1294 }, { %r1315, %r1316, %r1317, %r1318 }, { %r1319, %r1320 }, { %r1291, %r1292, %r1293, %r1294 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1301, %r1302, %r1303, %r1304 }, { %r1315, %r1316, %r1317, %r1318 }, { %r1321, %r1322 }, { %r1301, %r1302, %r1303, %r1304 }; + // end inline asm + .loc 1 251 14 // chunk_fwd.py:251:14 + selp.f32 %r1698, %r1177, %r1191, %p92; + selp.f32 %r1699, %r1191, %r1177, %p92; + selp.f32 %r1700, %r1933, %r1934, %p92; + selp.f32 %r1701, %r1934, %r1933, %p92; + selp.f32 %r1702, %r1935, %r1936, %p92; + selp.f32 %r1703, %r1936, %r1935, %p92; + selp.f32 %r1704, %r1178, %r1192, %p92; + selp.f32 %r1705, %r1192, %r1178, %p92; + shfl.sync.idx.b32 %r1706, %r1698, %r1469, 31, -1; + shfl.sync.idx.b32 %r1707, %r1700, %r1469, 31, -1; + shfl.sync.idx.b32 %r1708, %r1699, %r1472, 31, -1; + shfl.sync.idx.b32 %r1709, %r1701, %r1472, 31, -1; + shfl.sync.idx.b32 %r1710, %r1702, %r1469, 31, -1; + shfl.sync.idx.b32 %r1711, %r1704, %r1469, 31, -1; + shfl.sync.idx.b32 %r1712, %r1703, %r1472, 31, -1; + shfl.sync.idx.b32 %r1713, %r1705, %r1472, 31, -1; + selp.b32 %r1325, %r1708, %r1706, %p93; + selp.b32 %r1326, %r1709, %r1707, %p93; + selp.b32 %r1337, %r1712, %r1710, %p93; + selp.b32 %r1338, %r1713, %r1711, %p93; + .loc 1 281 8 // chunk_fwd.py:281:8 + selp.f32 %r1714, %r1291, %r1292, %p94; + selp.f32 %r1715, %r1292, %r1291, %p94; + selp.f32 %r1716, %r1293, %r1294, %p94; + selp.f32 %r1717, %r1294, %r1293, %p94; + selp.f32 %r1718, %r1301, %r1302, %p94; + selp.f32 %r1719, %r1302, %r1301, %p94; + selp.f32 %r1720, %r1303, %r1304, %p94; + selp.f32 %r1721, %r1304, %r1303, %p94; + shfl.sync.idx.b32 %r1722, %r1714, %r1537, 31, -1; + shfl.sync.idx.b32 %r1723, %r1715, %r1539, 31, -1; + shfl.sync.idx.b32 %r1724, %r1716, %r1537, 31, -1; + shfl.sync.idx.b32 %r1725, %r1717, %r1539, 31, -1; + shfl.sync.idx.b32 %r1726, %r1718, %r1537, 31, -1; + shfl.sync.idx.b32 %r1727, %r1719, %r1539, 31, -1; + shfl.sync.idx.b32 %r1728, %r1720, %r1537, 31, -1; + shfl.sync.idx.b32 %r1729, %r1721, %r1539, 31, -1; + selp.b32 %r1328, %r1723, %r1722, %p95; + selp.b32 %r1340, %r1725, %r1724, %p95; + selp.b32 %r1330, %r1727, %r1726, %p95; + selp.b32 %r1346, %r1729, %r1728, %p95; + .loc 1 251 14 // chunk_fwd.py:251:14 + selp.b32 %r1323, %r1706, %r1708, %p93; + selp.b32 %r1324, %r1707, %r1709, %p93; + selp.b32 %r1335, %r1710, %r1712, %p93; + selp.b32 %r1336, %r1711, %r1713, %p93; + .loc 1 281 8 // chunk_fwd.py:281:8 + selp.b32 %r1327, %r1722, %r1723, %p95; + selp.b32 %r1339, %r1724, %r1725, %p95; + selp.b32 %r1329, %r1726, %r1727, %p95; + selp.b32 %r1345, %r1728, %r1729, %p95; + mov.b32 %r1331, %r1381; + mov.b32 %r1332, %r1381; + mov.b32 %r1333, %r1381; + mov.b32 %r1334, %r1381; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1331, %r1332, %r1333, %r1334 }, { %r1323, %r1324, %r1325, %r1326 }, { %r1327, %r1328 }, { %r1331, %r1332, %r1333, %r1334 }; + // end inline asm + mov.b32 %r1341, %r1381; + mov.b32 %r1342, %r1381; + mov.b32 %r1343, %r1381; + mov.b32 %r1344, %r1381; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1341, %r1342, %r1343, %r1344 }, { %r1323, %r1324, %r1325, %r1326 }, { %r1329, %r1330 }, { %r1341, %r1342, %r1343, %r1344 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1331, %r1332, %r1333, %r1334 }, { %r1335, %r1336, %r1337, %r1338 }, { %r1339, %r1340 }, { %r1331, %r1332, %r1333, %r1334 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1341, %r1342, %r1343, %r1344 }, { %r1335, %r1336, %r1337, %r1338 }, { %r1345, %r1346 }, { %r1341, %r1342, %r1343, %r1344 }; + // end inline asm + .loc 1 287 22 // chunk_fwd.py:287:22 + mov.b32 %r1351, %r1381; + mov.b32 %r1352, %r1381; + mov.b32 %r1353, %r1381; + mov.b32 %r1354, %r1381; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1351, %r1352, %r1353, %r1354 }, { %r1347, %r1348, %r1349, %r1350 }, { %r1223, %r1224 }, { %r1351, %r1352, %r1353, %r1354 }; + // end inline asm + mov.b32 %r1359, %r1381; + mov.b32 %r1360, %r1381; + mov.b32 %r1361, %r1381; + mov.b32 %r1362, %r1381; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1359, %r1360, %r1361, %r1362 }, { %r1347, %r1348, %r1349, %r1350 }, { %r1225, %r1226 }, { %r1359, %r1360, %r1361, %r1362 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1351, %r1352, %r1353, %r1354 }, { %r1355, %r1356, %r1357, %r1358 }, { %r1235, %r1236 }, { %r1351, %r1352, %r1353, %r1354 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1359, %r1360, %r1361, %r1362 }, { %r1355, %r1356, %r1357, %r1358 }, { %r1241, %r1242 }, { %r1359, %r1360, %r1361, %r1362 }; + // end inline asm + .loc 1 288 22 // chunk_fwd.py:288:22 + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1351, %r1352, %r1353, %r1354 }, { %r1283, %r1284, %r1285, %r1286 }, { %r1247, %r1248 }, { %r1351, %r1352, %r1353, %r1354 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1359, %r1360, %r1361, %r1362 }, { %r1283, %r1284, %r1285, %r1286 }, { %r1249, %r1250 }, { %r1359, %r1360, %r1361, %r1362 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1351, %r1352, %r1353, %r1354 }, { %r1295, %r1296, %r1297, %r1298 }, { %r1255, %r1256 }, { %r1351, %r1352, %r1353, %r1354 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1359, %r1360, %r1361, %r1362 }, { %r1295, %r1296, %r1297, %r1298 }, { %r1257, %r1258 }, { %r1359, %r1360, %r1361, %r1362 }; + // end inline asm + .loc 1 273 14 // chunk_fwd.py:273:14 + selp.f32 %r1730, %r1658, %r1659, %p95; + selp.f32 %r1731, %r1659, %r1658, %p95; + selp.f32 %r1732, %r1660, %r1661, %p95; + selp.f32 %r1733, %r1661, %r1660, %p95; + selp.f32 %r1734, %r1662, %r1663, %p95; + selp.f32 %r1735, %r1663, %r1662, %p95; + selp.f32 %r1736, %r1664, %r1665, %p95; + selp.f32 %r1737, %r1665, %r1664, %p95; + shfl.sync.idx.b32 %r1738, %r1730, %r1446, 31, -1; + shfl.sync.idx.b32 %r1739, %r1731, %r1448, 31, -1; + shfl.sync.idx.b32 %r1740, %r1732, %r1446, 31, -1; + shfl.sync.idx.b32 %r1741, %r1733, %r1448, 31, -1; + shfl.sync.idx.b32 %r1742, %r1734, %r1446, 31, -1; + shfl.sync.idx.b32 %r1743, %r1735, %r1448, 31, -1; + shfl.sync.idx.b32 %r1744, %r1736, %r1446, 31, -1; + shfl.sync.idx.b32 %r1745, %r1737, %r1448, 31, -1; + selp.b32 %r1367, %r1739, %r1738, %p95; + selp.b32 %r1368, %r1741, %r1740, %p95; + selp.b32 %r1369, %r1743, %r1742, %p95; + selp.b32 %r1370, %r1745, %r1744, %p95; + selp.b32 %r1363, %r1738, %r1739, %p95; + selp.b32 %r1364, %r1740, %r1741, %p95; + selp.b32 %r1365, %r1742, %r1743, %p95; + selp.b32 %r1366, %r1744, %r1745, %p95; + .loc 1 289 22 // chunk_fwd.py:289:22 + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1351, %r1352, %r1353, %r1354 }, { %r1307, %r1308, %r1309, %r1310 }, { %r1363, %r1364 }, { %r1351, %r1352, %r1353, %r1354 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1359, %r1360, %r1361, %r1362 }, { %r1307, %r1308, %r1309, %r1310 }, { %r1365, %r1366 }, { %r1359, %r1360, %r1361, %r1362 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1351, %r1352, %r1353, %r1354 }, { %r1315, %r1316, %r1317, %r1318 }, { %r1367, %r1368 }, { %r1351, %r1352, %r1353, %r1354 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1359, %r1360, %r1361, %r1362 }, { %r1315, %r1316, %r1317, %r1318 }, { %r1369, %r1370 }, { %r1359, %r1360, %r1361, %r1362 }; + // end inline asm + .loc 1 287 8 // chunk_fwd.py:287:8 + selp.f32 %r1746, %r1351, %r1352, %p94; + selp.f32 %r1747, %r1352, %r1351, %p94; + selp.f32 %r1748, %r1353, %r1354, %p94; + selp.f32 %r1749, %r1354, %r1353, %p94; + selp.f32 %r1750, %r1359, %r1360, %p94; + selp.f32 %r1751, %r1360, %r1359, %p94; + selp.f32 %r1752, %r1361, %r1362, %p94; + selp.f32 %r1753, %r1362, %r1361, %p94; + shfl.sync.idx.b32 %r1754, %r1746, %r1537, 31, -1; + shfl.sync.idx.b32 %r1755, %r1747, %r1539, 31, -1; + shfl.sync.idx.b32 %r1756, %r1748, %r1537, 31, -1; + shfl.sync.idx.b32 %r1757, %r1749, %r1539, 31, -1; + shfl.sync.idx.b32 %r1758, %r1750, %r1537, 31, -1; + shfl.sync.idx.b32 %r1759, %r1751, %r1539, 31, -1; + shfl.sync.idx.b32 %r1760, %r1752, %r1537, 31, -1; + shfl.sync.idx.b32 %r1761, %r1753, %r1539, 31, -1; + selp.b32 %r1372, %r1755, %r1754, %p95; + selp.b32 %r1380, %r1757, %r1756, %p95; + selp.b32 %r1374, %r1759, %r1758, %p95; + selp.b32 %r1386, %r1761, %r1760, %p95; + selp.b32 %r1371, %r1754, %r1755, %p95; + selp.b32 %r1379, %r1756, %r1757, %p95; + selp.b32 %r1373, %r1758, %r1759, %p95; + selp.b32 %r1385, %r1760, %r1761, %p95; + mov.b32 %r1375, %r1381; + mov.b32 %r1376, %r1381; + mov.b32 %r1377, %r1381; + mov.b32 %r1378, %r1381; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1375, %r1376, %r1377, %r1378 }, { %r1323, %r1324, %r1325, %r1326 }, { %r1371, %r1372 }, { %r1375, %r1376, %r1377, %r1378 }; + // end inline asm + mov.b32 %r1382, %r1381; + mov.b32 %r1383, %r1381; + mov.b32 %r1384, %r1381; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1381, %r1382, %r1383, %r1384 }, { %r1323, %r1324, %r1325, %r1326 }, { %r1373, %r1374 }, { %r1381, %r1382, %r1383, %r1384 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1375, %r1376, %r1377, %r1378 }, { %r1335, %r1336, %r1337, %r1338 }, { %r1379, %r1380 }, { %r1375, %r1376, %r1377, %r1378 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1381, %r1382, %r1383, %r1384 }, { %r1335, %r1336, %r1337, %r1338 }, { %r1385, %r1386 }, { %r1381, %r1382, %r1383, %r1384 }; + // end inline asm + .loc 1 297 16 // chunk_fwd.py:297:16 + shl.b64 %rd284, %rd51, 12; + add.s64 %rd285, %rd50, %rd284; + .loc 1 297 51 // chunk_fwd.py:297:51 + mul.wide.u32 %rd286, %r29, 2; + add.s64 %rd274, %rd285, %rd286; + .loc 1 298 16 // chunk_fwd.py:298:16 + shl.b64 %rd287, %rd52, 12; + add.s64 %rd288, %rd50, %rd287; + .loc 1 298 51 // chunk_fwd.py:298:51 + add.s64 %rd275, %rd288, %rd286; + .loc 1 299 51 // chunk_fwd.py:299:51 + add.s64 %rd276, %rd275, 32; + .loc 1 300 16 // chunk_fwd.py:300:16 + shl.b64 %rd289, %rd53, 12; + add.s64 %rd290, %rd50, %rd289; + .loc 1 300 51 // chunk_fwd.py:300:51 + add.s64 %rd277, %rd290, %rd286; + .loc 1 301 51 // chunk_fwd.py:301:51 + add.s64 %rd278, %rd277, 32; + .loc 1 302 51 // chunk_fwd.py:302:51 + add.s64 %rd279, %rd277, 64; + .loc 1 303 16 // chunk_fwd.py:303:16 + shl.b64 %rd291, %rd54, 12; + add.s64 %rd292, %rd50, %rd291; + .loc 1 303 51 // chunk_fwd.py:303:51 + add.s64 %rd280, %rd292, %rd286; + .loc 1 304 51 // chunk_fwd.py:304:51 + add.s64 %rd281, %rd280, 32; + .loc 1 305 51 // chunk_fwd.py:305:51 + add.s64 %rd282, %rd280, 64; + .loc 1 306 51 // chunk_fwd.py:306:51 + add.s64 %rd283, %rd280, 96; + .loc 1 319 20 // chunk_fwd.py:319:20 + bar.sync 0; + shr.u32 %r1762, %r1, 1; + or.b32 %r1763, %r1762, %r1; + and.b32 %r1764, %r12, 144; + and.b32 %r1765, %r1456, 320; + and.b32 %r1766, %r28, 32; + and.b32 %r1767, %r1763, 16; + shl.b32 %r1768, %r1767, 2; + or.b32 %r1769, %r1765, %r1764; + xor.b32 %r1770, %r1769, %r1768; + add.s32 %r1771, %r108, %r1766; + add.s32 %r1772, %r1771, %r1770; + cvt.rn.bf16x2.f32 %r1773, %r1083, %r1082; + cvt.rn.bf16x2.f32 %r1774, %r1086, %r1085; + stmatrix.sync.aligned.m8n8.x2.trans.shared.b16 [%r1772], {%r1773, %r1774}; + bar.sync 0; + and.b32 %r1775, %r12, 112; + and.b32 %r1776, %r1420, 320; + shl.b32 %r1777, %r1767, 3; + xor.b32 %r1778, %r1776, %r1775; + add.s32 %r1779, %r108, %r1777; + add.s32 %r1780, %r1779, %r1778; + ldmatrix.sync.aligned.m8n8.x2.trans.shared.b16 {%r1387, %r1388}, [%r1780]; + // begin inline asm + @%p81 st.global.v2.b32 [ %rd274 + 0 ], { %r1387, %r1388 }; + // end inline asm + .loc 1 320 20 // chunk_fwd.py:320:20 + bar.sync 0; + and.b32 %r1781, %r1535, 64; + xor.b32 %r1782, %r1769, %r1781; + add.s32 %r1783, %r1771, %r1782; + cvt.rn.bf16x2.f32 %r1784, %r1488, %r1487; + cvt.rn.bf16x2.f32 %r1785, %r1490, %r1489; + cvt.rn.bf16x2.f32 %r1786, %r1492, %r1491; + cvt.rn.bf16x2.f32 %r1787, %r1494, %r1493; + stmatrix.sync.aligned.m8n8.x4.trans.shared.b16 [%r1783], {%r1784, %r1785, %r1786, %r1787}; + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x2.trans.shared.b16 {%r1389, %r1390}, [%r1780]; + // begin inline asm + @%p82 st.global.v2.b32 [ %rd275 + 0 ], { %r1389, %r1390 }; + // end inline asm + .loc 1 321 20 // chunk_fwd.py:321:20 + bar.sync 0; + cvt.rn.bf16x2.f32 %r1788, %r1103, %r1089; + cvt.rn.bf16x2.f32 %r1789, %r1915, %r1914; + cvt.rn.bf16x2.f32 %r1790, %r1917, %r1916; + cvt.rn.bf16x2.f32 %r1791, %r1104, %r1090; + stmatrix.sync.aligned.m8n8.x4.trans.shared.b16 [%r1783], {%r1788, %r1789, %r1790, %r1791}; + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x2.trans.shared.b16 {%r1391, %r1392}, [%r1780]; + // begin inline asm + @%p82 st.global.v2.b32 [ %rd276 + 0 ], { %r1391, %r1392 }; + // end inline asm + .loc 1 322 20 // chunk_fwd.py:322:20 + bar.sync 0; + cvt.rn.bf16x2.f32 %r1792, %r1659, %r1658; + cvt.rn.bf16x2.f32 %r1793, %r1661, %r1660; + cvt.rn.bf16x2.f32 %r1794, %r1663, %r1662; + cvt.rn.bf16x2.f32 %r1795, %r1665, %r1664; + stmatrix.sync.aligned.m8n8.x4.trans.shared.b16 [%r1783], {%r1792, %r1793, %r1794, %r1795}; + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x2.trans.shared.b16 {%r1393, %r1394}, [%r1780]; + // begin inline asm + @%p83 st.global.v2.b32 [ %rd277 + 0 ], { %r1393, %r1394 }; + // end inline asm + .loc 1 323 20 // chunk_fwd.py:323:20 + bar.sync 0; + cvt.rn.bf16x2.f32 %r1796, %r1548, %r1547; + cvt.rn.bf16x2.f32 %r1797, %r1550, %r1549; + cvt.rn.bf16x2.f32 %r1798, %r1552, %r1551; + cvt.rn.bf16x2.f32 %r1799, %r1554, %r1553; + stmatrix.sync.aligned.m8n8.x4.trans.shared.b16 [%r1783], {%r1796, %r1797, %r1798, %r1799}; + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x2.trans.shared.b16 {%r1395, %r1396}, [%r1780]; + // begin inline asm + @%p83 st.global.v2.b32 [ %rd278 + 0 ], { %r1395, %r1396 }; + // end inline asm + .loc 1 324 20 // chunk_fwd.py:324:20 + bar.sync 0; + cvt.rn.bf16x2.f32 %r1800, %r1147, %r1133; + cvt.rn.bf16x2.f32 %r1801, %r1924, %r1923; + cvt.rn.bf16x2.f32 %r1802, %r1926, %r1925; + cvt.rn.bf16x2.f32 %r1803, %r1148, %r1134; + stmatrix.sync.aligned.m8n8.x4.trans.shared.b16 [%r1783], {%r1800, %r1801, %r1802, %r1803}; + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x2.trans.shared.b16 {%r1397, %r1398}, [%r1780]; + // begin inline asm + @%p83 st.global.v2.b32 [ %rd279 + 0 ], { %r1397, %r1398 }; + // end inline asm + .loc 1 285 14 // chunk_fwd.py:285:14 + mov.b64 %rd293, {%r1375, %r1376}; + mov.b64 %rd294, {%r1486, %r1486}; + sub.f32x2 %rd295, %rd294, %rd293; + .loc 1 325 30 // chunk_fwd.py:325:30 + mov.b64 {%r1804, %r1805}, %rd295; + cvt.rn.bf16x2.f32 %r1806, %r1805, %r1804; + .loc 1 285 14 // chunk_fwd.py:285:14 + mov.b64 %rd296, {%r1377, %r1378}; + sub.f32x2 %rd297, %rd294, %rd296; + .loc 1 325 30 // chunk_fwd.py:325:30 + mov.b64 {%r1807, %r1808}, %rd297; + cvt.rn.bf16x2.f32 %r1809, %r1808, %r1807; + .loc 1 285 14 // chunk_fwd.py:285:14 + mov.b64 %rd298, {%r1381, %r1382}; + sub.f32x2 %rd299, %rd294, %rd298; + .loc 1 325 30 // chunk_fwd.py:325:30 + mov.b64 {%r1810, %r1811}, %rd299; + cvt.rn.bf16x2.f32 %r1812, %r1811, %r1810; + .loc 1 285 14 // chunk_fwd.py:285:14 + mov.b64 %rd300, {%r1383, %r1384}; + sub.f32x2 %rd301, %rd294, %rd300; + .loc 1 325 30 // chunk_fwd.py:325:30 + mov.b64 {%r1813, %r1814}, %rd301; + cvt.rn.bf16x2.f32 %r1815, %r1814, %r1813; + .loc 1 325 20 // chunk_fwd.py:325:20 + bar.sync 0; + stmatrix.sync.aligned.m8n8.x4.trans.shared.b16 [%r1783], {%r1806, %r1809, %r1812, %r1815}; + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x2.trans.shared.b16 {%r1399, %r1400}, [%r1780]; + // begin inline asm + @%p84 st.global.v2.b32 [ %rd280 + 0 ], { %r1399, %r1400 }; + // end inline asm + .loc 1 279 14 // chunk_fwd.py:279:14 + mov.b64 %rd302, {%r1331, %r1332}; + sub.f32x2 %rd303, %rd294, %rd302; + .loc 1 326 30 // chunk_fwd.py:326:30 + mov.b64 {%r1816, %r1817}, %rd303; + cvt.rn.bf16x2.f32 %r1818, %r1817, %r1816; + .loc 1 279 14 // chunk_fwd.py:279:14 + mov.b64 %rd304, {%r1333, %r1334}; + sub.f32x2 %rd305, %rd294, %rd304; + .loc 1 326 30 // chunk_fwd.py:326:30 + mov.b64 {%r1819, %r1820}, %rd305; + cvt.rn.bf16x2.f32 %r1821, %r1820, %r1819; + .loc 1 279 14 // chunk_fwd.py:279:14 + mov.b64 %rd306, {%r1341, %r1342}; + sub.f32x2 %rd307, %rd294, %rd306; + .loc 1 326 30 // chunk_fwd.py:326:30 + mov.b64 {%r1822, %r1823}, %rd307; + cvt.rn.bf16x2.f32 %r1824, %r1823, %r1822; + .loc 1 279 14 // chunk_fwd.py:279:14 + mov.b64 %rd308, {%r1343, %r1344}; + sub.f32x2 %rd309, %rd294, %rd308; + .loc 1 326 30 // chunk_fwd.py:326:30 + mov.b64 {%r1825, %r1826}, %rd309; + cvt.rn.bf16x2.f32 %r1827, %r1826, %r1825; + .loc 1 326 20 // chunk_fwd.py:326:20 + bar.sync 0; + stmatrix.sync.aligned.m8n8.x4.trans.shared.b16 [%r1783], {%r1818, %r1821, %r1824, %r1827}; + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x2.trans.shared.b16 {%r1401, %r1402}, [%r1780]; + // begin inline asm + @%p84 st.global.v2.b32 [ %rd281 + 0 ], { %r1401, %r1402 }; + // end inline asm + .loc 1 267 14 // chunk_fwd.py:267:14 + mov.b64 %rd310, {%r1203, %r1204}; + sub.f32x2 %rd311, %rd294, %rd310; + .loc 1 327 30 // chunk_fwd.py:327:30 + mov.b64 {%r1828, %r1829}, %rd311; + cvt.rn.bf16x2.f32 %r1830, %r1829, %r1828; + .loc 1 267 14 // chunk_fwd.py:267:14 + mov.b64 %rd312, {%r1205, %r1206}; + sub.f32x2 %rd313, %rd294, %rd312; + .loc 1 327 30 // chunk_fwd.py:327:30 + mov.b64 {%r1831, %r1832}, %rd313; + cvt.rn.bf16x2.f32 %r1833, %r1832, %r1831; + .loc 1 267 14 // chunk_fwd.py:267:14 + mov.b64 %rd314, {%r1213, %r1214}; + sub.f32x2 %rd315, %rd294, %rd314; + .loc 1 327 30 // chunk_fwd.py:327:30 + mov.b64 {%r1834, %r1835}, %rd315; + cvt.rn.bf16x2.f32 %r1836, %r1835, %r1834; + .loc 1 267 14 // chunk_fwd.py:267:14 + mov.b64 %rd316, {%r1215, %r1216}; + sub.f32x2 %rd317, %rd294, %rd316; + .loc 1 327 30 // chunk_fwd.py:327:30 + mov.b64 {%r1837, %r1838}, %rd317; + cvt.rn.bf16x2.f32 %r1839, %r1838, %r1837; + .loc 1 327 20 // chunk_fwd.py:327:20 + bar.sync 0; + stmatrix.sync.aligned.m8n8.x4.trans.shared.b16 [%r1783], {%r1830, %r1833, %r1836, %r1839}; + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x2.trans.shared.b16 {%r1403, %r1404}, [%r1780]; + // begin inline asm + @%p84 st.global.v2.b32 [ %rd282 + 0 ], { %r1403, %r1404 }; + // end inline asm + .loc 1 328 20 // chunk_fwd.py:328:20 + bar.sync 0; + cvt.rn.bf16x2.f32 %r1840, %r1191, %r1177; + cvt.rn.bf16x2.f32 %r1841, %r1934, %r1933; + cvt.rn.bf16x2.f32 %r1842, %r1936, %r1935; + cvt.rn.bf16x2.f32 %r1843, %r1192, %r1178; + stmatrix.sync.aligned.m8n8.x4.trans.shared.b16 [%r1783], {%r1840, %r1841, %r1842, %r1843}; + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x2.trans.shared.b16 {%r1405, %r1406}, [%r1780]; + // begin inline asm + @%p84 st.global.v2.b32 [ %rd283 + 0 ], { %r1405, %r1406 }; + // end inline asm +$L__BB0_1: // %common.ret + .loc 1 0 0 // chunk_fwd.py:0 + ret; +$L__tmp167: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py" + .file 3 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 6 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 823 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0x330 DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 99 // DW_AT_name +.b8 104 +.b8 117 +.b8 110 +.b8 107 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 111 +.b8 112 +.b8 115 +.b8 47 +.b8 103 +.b8 97 +.b8 116 +.b8 101 +.b8 100 +.b8 95 +.b8 100 +.b8 101 +.b8 108 +.b8 116 +.b8 97 +.b8 95 +.b8 114 +.b8 117 +.b8 108 +.b8 101 +.b8 0 +.b8 2 // Abbrev [2] 0x76:0x2e DW_TAG_subprogram +.b8 99 // DW_AT_name +.b8 104 +.b8 117 +.b8 110 +.b8 107 +.b8 95 +.b8 103 +.b8 97 +.b8 116 +.b8 101 +.b8 100 +.b8 95 +.b8 100 +.b8 101 +.b8 108 +.b8 116 +.b8 97 +.b8 95 +.b8 114 +.b8 117 +.b8 108 +.b8 101 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 107 +.b8 116 +.b8 95 +.b8 115 +.b8 111 +.b8 108 +.b8 118 +.b8 101 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0xa4:0x296 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 118 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0xb9:0x18 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp2 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 183 // DW_AT_call_line +.b8 70 // DW_AT_call_column +.b8 4 // Abbrev [4] 0xd1:0x18 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp3 // DW_AT_low_pc +.b64 $L__tmp4 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 184 // DW_AT_call_line +.b8 70 // DW_AT_call_column +.b8 4 // Abbrev [4] 0xe9:0x18 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp5 // DW_AT_low_pc +.b64 $L__tmp6 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 185 // DW_AT_call_line +.b8 70 // DW_AT_call_column +.b8 4 // Abbrev [4] 0x101:0x18 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp7 // DW_AT_low_pc +.b64 $L__tmp8 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 186 // DW_AT_call_line +.b8 70 // DW_AT_call_column +.b8 4 // Abbrev [4] 0x119:0x18 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp9 // DW_AT_low_pc +.b64 $L__tmp10 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 188 // DW_AT_call_line +.b8 64 // DW_AT_call_column +.b8 4 // Abbrev [4] 0x131:0x18 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp11 // DW_AT_low_pc +.b64 $L__tmp12 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 189 // DW_AT_call_line +.b8 64 // DW_AT_call_column +.b8 4 // Abbrev [4] 0x149:0x18 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp13 // DW_AT_low_pc +.b64 $L__tmp14 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 190 // DW_AT_call_line +.b8 64 // DW_AT_call_column +.b8 4 // Abbrev [4] 0x161:0x18 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp15 // DW_AT_low_pc +.b64 $L__tmp16 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 191 // DW_AT_call_line +.b8 64 // DW_AT_call_column +.b8 4 // Abbrev [4] 0x179:0x18 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp17 // DW_AT_low_pc +.b64 $L__tmp18 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 192 // DW_AT_call_line +.b8 64 // DW_AT_call_column +.b8 4 // Abbrev [4] 0x191:0x18 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp19 // DW_AT_low_pc +.b64 $L__tmp20 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 193 // DW_AT_call_line +.b8 64 // DW_AT_call_column +.b8 5 // Abbrev [5] 0x1a9:0x32 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp21 // DW_AT_low_pc +.b64 $L__tmp33 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 228 // DW_AT_call_line +.b8 66 // DW_AT_call_column +.b8 6 // Abbrev [6] 0x1c1:0x19 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp21 // DW_AT_low_pc +.b64 $L__tmp33 // DW_AT_high_pc +.b8 3 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 5 // Abbrev [5] 0x1db:0x32 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp34 // DW_AT_low_pc +.b64 $L__tmp45 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 230 // DW_AT_call_line +.b8 56 // DW_AT_call_column +.b8 6 // Abbrev [6] 0x1f3:0x19 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp34 // DW_AT_low_pc +.b64 $L__tmp45 // DW_AT_high_pc +.b8 3 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 5 // Abbrev [5] 0x20d:0x32 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp46 // DW_AT_low_pc +.b64 $L__tmp65 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 233 // DW_AT_call_line +.b8 66 // DW_AT_call_column +.b8 6 // Abbrev [6] 0x225:0x19 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp46 // DW_AT_low_pc +.b64 $L__tmp65 // DW_AT_high_pc +.b8 3 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 5 // Abbrev [5] 0x23f:0x32 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp66 // DW_AT_low_pc +.b64 $L__tmp85 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 235 // DW_AT_call_line +.b8 56 // DW_AT_call_column +.b8 6 // Abbrev [6] 0x257:0x19 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp66 // DW_AT_low_pc +.b64 $L__tmp85 // DW_AT_high_pc +.b8 3 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 5 // Abbrev [5] 0x271:0x32 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp86 // DW_AT_low_pc +.b64 $L__tmp105 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 238 // DW_AT_call_line +.b8 66 // DW_AT_call_column +.b8 6 // Abbrev [6] 0x289:0x19 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp86 // DW_AT_low_pc +.b64 $L__tmp105 // DW_AT_high_pc +.b8 3 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 5 // Abbrev [5] 0x2a3:0x32 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp106 // DW_AT_low_pc +.b64 $L__tmp125 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 240 // DW_AT_call_line +.b8 56 // DW_AT_call_column +.b8 6 // Abbrev [6] 0x2bb:0x19 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp106 // DW_AT_low_pc +.b64 $L__tmp125 // DW_AT_high_pc +.b8 3 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 5 // Abbrev [5] 0x2d5:0x32 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp126 // DW_AT_low_pc +.b64 $L__tmp146 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 243 // DW_AT_call_line +.b8 66 // DW_AT_call_column +.b8 6 // Abbrev [6] 0x2ed:0x19 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp126 // DW_AT_low_pc +.b64 $L__tmp146 // DW_AT_high_pc +.b8 3 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 5 // Abbrev [5] 0x307:0x32 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp147 // DW_AT_low_pc +.b64 $L__tmp166 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 245 // DW_AT_call_line +.b8 56 // DW_AT_call_column +.b8 6 // Abbrev [6] 0x31f:0x19 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp147 // DW_AT_low_pc +.b64 $L__tmp166 // DW_AT_high_pc +.b8 3 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.source b/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..b6993b6e33c471a1c8466dd1e35ffc9d5b77c03f --- /dev/null +++ b/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.source @@ -0,0 +1,2482 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":40:0) +#loc550 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:0) +#loc554 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc556 = loc(unknown) +#loc559 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc563 = loc("k"(#loc)) +#loc564 = loc("g"(#loc)) +#loc565 = loc("beta"(#loc)) +#loc566 = loc("A"(#loc)) +#loc567 = loc("T"(#loc)) +#loc1059 = loc("x"(#loc550)) +#loc1060 = loc("input"(#loc554)) +#loc1061 = loc("a"(#loc559)) +#loc1062 = loc("b"(#loc559)) +module { + tt.func public @chunk_gated_delta_rule_fwd_kkt_solve_kernel(%k: !tt.ptr {tt.divisibility = 16 : i32} loc("k"(#loc)), %g: !tt.ptr {tt.divisibility = 16 : i32} loc("g"(#loc)), %beta: !tt.ptr {tt.divisibility = 16 : i32} loc("beta"(#loc)), %A: !tt.ptr {tt.divisibility = 16 : i32} loc("A"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %0 = tt.get_program_id x : i32 loc(#loc1) + %1 = arith.extsi %0 : i32 to i64 loc(#loc2) + %2 = tt.get_program_id y : i32 loc(#loc3) + %3 = arith.extsi %2 : i32 to i64 loc(#loc4) + %c32_i32 = arith.constant 32 : i32 loc(#loc5) + %c32_i64 = arith.constant 32 : i64 loc(#loc5) + %4 = arith.divsi %3, %c32_i64 : i64 loc(#loc5) + %c32_i32_0 = arith.constant 32 : i32 loc(#loc6) + %c32_i64_1 = arith.constant 32 : i64 loc(#loc6) + %5 = arith.remsi %3, %c32_i64_1 : i64 loc(#loc6) + %6 = arith.extsi %T : i32 to i64 loc(#loc7) + %7 = arith.muli %4, %6 : i64 loc(#loc7) + %8 = arith.extsi %T : i32 to i64 loc(#loc8) + %9 = arith.muli %4, %8 : i64 loc(#loc8) + %10 = arith.extsi %T : i32 to i64 loc(#loc9) + %11 = arith.addi %9, %10 : i64 loc(#loc9) + %c64_i32 = arith.constant 64 : i32 loc(#loc10) + %c64_i64 = arith.constant 64 : i64 loc(#loc10) + %12 = arith.muli %1, %c64_i64 : i64 loc(#loc10) + %13 = arith.extsi %T : i32 to i64 loc(#loc11) + %14 = arith.cmpi sge, %12, %13 : i64 loc(#loc11) + cf.cond_br %14, ^bb1, ^bb2 loc(#loc11) + ^bb1: // pred: ^bb0 + tt.return loc(#loc12) + ^bb2: // pred: ^bb0 + cf.br ^bb4 loc(#loc) + ^bb3: // no predecessors + cf.br ^bb4 loc(#loc) + ^bb4: // 2 preds: ^bb2, ^bb3 + %i_tc0 = arith.constant 64 : i32 loc(#loc568) + %i_tc0_2 = arith.constant 64 : i64 loc(#loc568) + %i_tc0_3 = arith.muli %1, %i_tc0_2 : i64 loc(#loc568) + %i_tc1 = arith.constant 64 : i32 loc(#loc569) + %i_tc1_4 = arith.constant 64 : i64 loc(#loc569) + %i_tc1_5 = arith.muli %1, %i_tc1_4 : i64 loc(#loc569) + %i_tc1_6 = arith.constant 16 : i32 loc(#loc570) + %i_tc1_7 = arith.constant 16 : i64 loc(#loc570) + %i_tc1_8 = arith.addi %i_tc1_5, %i_tc1_7 : i64 loc(#loc570) + %i_tc2 = arith.constant 64 : i32 loc(#loc571) + %i_tc2_9 = arith.constant 64 : i64 loc(#loc571) + %i_tc2_10 = arith.muli %1, %i_tc2_9 : i64 loc(#loc571) + %i_tc2_11 = arith.constant 32 : i32 loc(#loc572) + %i_tc2_12 = arith.constant 32 : i64 loc(#loc572) + %i_tc2_13 = arith.addi %i_tc2_10, %i_tc2_12 : i64 loc(#loc572) + %i_tc3 = arith.constant 64 : i32 loc(#loc573) + %i_tc3_14 = arith.constant 64 : i64 loc(#loc573) + %i_tc3_15 = arith.muli %1, %i_tc3_14 : i64 loc(#loc573) + %i_tc3_16 = arith.constant 48 : i32 loc(#loc574) + %i_tc3_17 = arith.constant 48 : i64 loc(#loc574) + %i_tc3_18 = arith.addi %i_tc3_15, %i_tc3_17 : i64 loc(#loc574) + %k_19 = arith.constant 32 : i32 loc(#loc575) + %k_20 = arith.constant 32 : i64 loc(#loc575) + %k_21 = arith.muli %7, %k_20 : i64 loc(#loc575) + %k_22 = arith.constant 1 : i32 loc(#loc576) + %k_23 = arith.constant 1 : i64 loc(#loc576) + %k_24 = arith.divsi %5, %k_23 : i64 loc(#loc576) + %k_25 = arith.addi %k_21, %k_24 : i64 loc(#loc577) + %k_26 = arith.constant 128 : i32 loc(#loc578) + %k_27 = arith.constant 128 : i64 loc(#loc578) + %k_28 = arith.muli %k_25, %k_27 : i64 loc(#loc578) + %k_29 = tt.addptr %k, %k_28 : !tt.ptr, i64 loc(#loc579) + %A_30 = arith.constant 32 : i32 loc(#loc580) + %A_31 = arith.constant 32 : i64 loc(#loc580) + %A_32 = arith.muli %7, %A_31 : i64 loc(#loc580) + %A_33 = arith.addi %A_32, %5 : i64 loc(#loc581) + %A_34 = arith.constant 64 : i32 loc(#loc582) + %A_35 = arith.constant 64 : i64 loc(#loc582) + %A_36 = arith.muli %A_33, %A_35 : i64 loc(#loc582) + %A_37 = tt.addptr %A, %A_36 : !tt.ptr, i64 loc(#loc583) + %o_i = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32> loc(#loc584) + %m_tc0 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc585) + %m_tc0_38 = tt.splat %i_tc0_3 : i64 -> tensor<16xi64> loc(#loc585) + %m_tc0_39 = arith.addi %m_tc0_38, %m_tc0 : tensor<16xi64> loc(#loc585) + %m_tc0_40 = arith.extsi %T : i32 to i64 loc(#loc586) + %m_tc0_41 = tt.splat %m_tc0_40 : i64 -> tensor<16xi64> loc(#loc586) + %m_tc0_42 = arith.cmpi slt, %m_tc0_39, %m_tc0_41 : tensor<16xi64> loc(#loc586) + %m_tc1 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc587) + %m_tc1_43 = tt.splat %i_tc1_8 : i64 -> tensor<16xi64> loc(#loc587) + %m_tc1_44 = arith.addi %m_tc1_43, %m_tc1 : tensor<16xi64> loc(#loc587) + %m_tc1_45 = arith.extsi %T : i32 to i64 loc(#loc588) + %m_tc1_46 = tt.splat %m_tc1_45 : i64 -> tensor<16xi64> loc(#loc588) + %m_tc1_47 = arith.cmpi slt, %m_tc1_44, %m_tc1_46 : tensor<16xi64> loc(#loc588) + %m_tc2 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc589) + %m_tc2_48 = tt.splat %i_tc2_13 : i64 -> tensor<16xi64> loc(#loc589) + %m_tc2_49 = arith.addi %m_tc2_48, %m_tc2 : tensor<16xi64> loc(#loc589) + %m_tc2_50 = arith.extsi %T : i32 to i64 loc(#loc590) + %m_tc2_51 = tt.splat %m_tc2_50 : i64 -> tensor<16xi64> loc(#loc590) + %m_tc2_52 = arith.cmpi slt, %m_tc2_49, %m_tc2_51 : tensor<16xi64> loc(#loc590) + %m_tc3 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc591) + %m_tc3_53 = tt.splat %i_tc3_18 : i64 -> tensor<16xi64> loc(#loc591) + %m_tc3_54 = arith.addi %m_tc3_53, %m_tc3 : tensor<16xi64> loc(#loc591) + %m_tc3_55 = arith.extsi %T : i32 to i64 loc(#loc592) + %m_tc3_56 = tt.splat %m_tc3_55 : i64 -> tensor<16xi64> loc(#loc592) + %m_tc3_57 = arith.cmpi slt, %m_tc3_54, %m_tc3_56 : tensor<16xi64> loc(#loc592) + %p_b0 = arith.constant 32 : i32 loc(#loc593) + %p_b0_58 = arith.constant 32 : i64 loc(#loc593) + %p_b0_59 = arith.muli %7, %p_b0_58 : i64 loc(#loc593) + %p_b0_60 = tt.addptr %beta, %p_b0_59 : !tt.ptr, i64 loc(#loc594) + %p_b0_61 = tt.addptr %p_b0_60, %5 : !tt.ptr, i64 loc(#loc595) + %p_b0_62 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc596) + %p_b0_63 = tt.splat %i_tc0_3 : i64 -> tensor<16xi64> loc(#loc596) + %p_b0_64 = arith.addi %p_b0_63, %p_b0_62 : tensor<16xi64> loc(#loc596) + %p_b0_65 = arith.constant 32 : i32 loc(#loc597) + %p_b0_66 = arith.constant 32 : i64 loc(#loc597) + %p_b0_67 = arith.constant dense<32> : tensor<16xi64> loc(#loc597) + %p_b0_68 = arith.muli %p_b0_64, %p_b0_67 : tensor<16xi64> loc(#loc597) + %p_b0_69 = tt.splat %p_b0_61 : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc598) + %p_b0_70 = tt.addptr %p_b0_69, %p_b0_68 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc598) + %p_b1 = arith.constant 32 : i32 loc(#loc599) + %p_b1_71 = arith.constant 32 : i64 loc(#loc599) + %p_b1_72 = arith.muli %7, %p_b1_71 : i64 loc(#loc599) + %p_b1_73 = tt.addptr %beta, %p_b1_72 : !tt.ptr, i64 loc(#loc600) + %p_b1_74 = tt.addptr %p_b1_73, %5 : !tt.ptr, i64 loc(#loc601) + %p_b1_75 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc602) + %p_b1_76 = tt.splat %i_tc1_8 : i64 -> tensor<16xi64> loc(#loc602) + %p_b1_77 = arith.addi %p_b1_76, %p_b1_75 : tensor<16xi64> loc(#loc602) + %p_b1_78 = arith.constant 32 : i32 loc(#loc603) + %p_b1_79 = arith.constant 32 : i64 loc(#loc603) + %p_b1_80 = arith.constant dense<32> : tensor<16xi64> loc(#loc603) + %p_b1_81 = arith.muli %p_b1_77, %p_b1_80 : tensor<16xi64> loc(#loc603) + %p_b1_82 = tt.splat %p_b1_74 : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc604) + %p_b1_83 = tt.addptr %p_b1_82, %p_b1_81 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc604) + %p_b2 = arith.constant 32 : i32 loc(#loc605) + %p_b2_84 = arith.constant 32 : i64 loc(#loc605) + %p_b2_85 = arith.muli %7, %p_b2_84 : i64 loc(#loc605) + %p_b2_86 = tt.addptr %beta, %p_b2_85 : !tt.ptr, i64 loc(#loc606) + %p_b2_87 = tt.addptr %p_b2_86, %5 : !tt.ptr, i64 loc(#loc607) + %p_b2_88 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc608) + %p_b2_89 = tt.splat %i_tc2_13 : i64 -> tensor<16xi64> loc(#loc608) + %p_b2_90 = arith.addi %p_b2_89, %p_b2_88 : tensor<16xi64> loc(#loc608) + %p_b2_91 = arith.constant 32 : i32 loc(#loc609) + %p_b2_92 = arith.constant 32 : i64 loc(#loc609) + %p_b2_93 = arith.constant dense<32> : tensor<16xi64> loc(#loc609) + %p_b2_94 = arith.muli %p_b2_90, %p_b2_93 : tensor<16xi64> loc(#loc609) + %p_b2_95 = tt.splat %p_b2_87 : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc610) + %p_b2_96 = tt.addptr %p_b2_95, %p_b2_94 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc610) + %p_b3 = arith.constant 32 : i32 loc(#loc611) + %p_b3_97 = arith.constant 32 : i64 loc(#loc611) + %p_b3_98 = arith.muli %7, %p_b3_97 : i64 loc(#loc611) + %p_b3_99 = tt.addptr %beta, %p_b3_98 : !tt.ptr, i64 loc(#loc612) + %p_b3_100 = tt.addptr %p_b3_99, %5 : !tt.ptr, i64 loc(#loc613) + %p_b3_101 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc614) + %p_b3_102 = tt.splat %i_tc3_18 : i64 -> tensor<16xi64> loc(#loc614) + %p_b3_103 = arith.addi %p_b3_102, %p_b3_101 : tensor<16xi64> loc(#loc614) + %p_b3_104 = arith.constant 32 : i32 loc(#loc615) + %p_b3_105 = arith.constant 32 : i64 loc(#loc615) + %p_b3_106 = arith.constant dense<32> : tensor<16xi64> loc(#loc615) + %p_b3_107 = arith.muli %p_b3_103, %p_b3_106 : tensor<16xi64> loc(#loc615) + %p_b3_108 = tt.splat %p_b3_100 : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc616) + %p_b3_109 = tt.addptr %p_b3_108, %p_b3_107 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc616) + %b_b0 = arith.constant 0.000000e+00 : f32 loc(#loc617) + %b_b0_110 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc617) + %b_b0_111 = arith.truncf %b_b0_110 : tensor<16xf32> to tensor<16xbf16> loc(#loc617) + %b_b0_112 = tt.load %p_b0_70, %m_tc0_42, %b_b0_111 : tensor<16x!tt.ptr> loc(#loc617) + %b_b0_113 = arith.extf %b_b0_112 : tensor<16xbf16> to tensor<16xf32> loc(#loc618) + %b_b1 = arith.constant 0.000000e+00 : f32 loc(#loc619) + %b_b1_114 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc619) + %b_b1_115 = arith.truncf %b_b1_114 : tensor<16xf32> to tensor<16xbf16> loc(#loc619) + %b_b1_116 = tt.load %p_b1_83, %m_tc1_47, %b_b1_115 : tensor<16x!tt.ptr> loc(#loc619) + %b_b1_117 = arith.extf %b_b1_116 : tensor<16xbf16> to tensor<16xf32> loc(#loc620) + %b_b2 = arith.constant 0.000000e+00 : f32 loc(#loc621) + %b_b2_118 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc621) + %b_b2_119 = arith.truncf %b_b2_118 : tensor<16xf32> to tensor<16xbf16> loc(#loc621) + %b_b2_120 = tt.load %p_b2_96, %m_tc2_52, %b_b2_119 : tensor<16x!tt.ptr> loc(#loc621) + %b_b2_121 = arith.extf %b_b2_120 : tensor<16xbf16> to tensor<16xf32> loc(#loc622) + %b_b3 = arith.constant 0.000000e+00 : f32 loc(#loc623) + %b_b3_122 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc623) + %b_b3_123 = arith.truncf %b_b3_122 : tensor<16xf32> to tensor<16xbf16> loc(#loc623) + %b_b3_124 = tt.load %p_b3_109, %m_tc3_57, %b_b3_123 : tensor<16x!tt.ptr> loc(#loc623) + %b_b3_125 = arith.extf %b_b3_124 : tensor<16xbf16> to tensor<16xf32> loc(#loc624) + %p_g0 = arith.constant 32 : i32 loc(#loc625) + %p_g0_126 = arith.constant 32 : i64 loc(#loc625) + %p_g0_127 = arith.muli %7, %p_g0_126 : i64 loc(#loc625) + %p_g0_128 = tt.addptr %g, %p_g0_127 : !tt.ptr, i64 loc(#loc626) + %p_g0_129 = tt.addptr %p_g0_128, %5 : !tt.ptr, i64 loc(#loc627) + %p_g0_130 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc628) + %p_g0_131 = tt.splat %i_tc0_3 : i64 -> tensor<16xi64> loc(#loc628) + %p_g0_132 = arith.addi %p_g0_131, %p_g0_130 : tensor<16xi64> loc(#loc628) + %p_g0_133 = arith.constant 32 : i32 loc(#loc629) + %p_g0_134 = arith.constant 32 : i64 loc(#loc629) + %p_g0_135 = arith.constant dense<32> : tensor<16xi64> loc(#loc629) + %p_g0_136 = arith.muli %p_g0_132, %p_g0_135 : tensor<16xi64> loc(#loc629) + %p_g0_137 = tt.splat %p_g0_129 : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc630) + %p_g0_138 = tt.addptr %p_g0_137, %p_g0_136 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc630) + %p_g1 = arith.constant 32 : i32 loc(#loc631) + %p_g1_139 = arith.constant 32 : i64 loc(#loc631) + %p_g1_140 = arith.muli %7, %p_g1_139 : i64 loc(#loc631) + %p_g1_141 = tt.addptr %g, %p_g1_140 : !tt.ptr, i64 loc(#loc632) + %p_g1_142 = tt.addptr %p_g1_141, %5 : !tt.ptr, i64 loc(#loc633) + %p_g1_143 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc634) + %p_g1_144 = tt.splat %i_tc1_8 : i64 -> tensor<16xi64> loc(#loc634) + %p_g1_145 = arith.addi %p_g1_144, %p_g1_143 : tensor<16xi64> loc(#loc634) + %p_g1_146 = arith.constant 32 : i32 loc(#loc635) + %p_g1_147 = arith.constant 32 : i64 loc(#loc635) + %p_g1_148 = arith.constant dense<32> : tensor<16xi64> loc(#loc635) + %p_g1_149 = arith.muli %p_g1_145, %p_g1_148 : tensor<16xi64> loc(#loc635) + %p_g1_150 = tt.splat %p_g1_142 : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc636) + %p_g1_151 = tt.addptr %p_g1_150, %p_g1_149 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc636) + %p_g2 = arith.constant 32 : i32 loc(#loc637) + %p_g2_152 = arith.constant 32 : i64 loc(#loc637) + %p_g2_153 = arith.muli %7, %p_g2_152 : i64 loc(#loc637) + %p_g2_154 = tt.addptr %g, %p_g2_153 : !tt.ptr, i64 loc(#loc638) + %p_g2_155 = tt.addptr %p_g2_154, %5 : !tt.ptr, i64 loc(#loc639) + %p_g2_156 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc640) + %p_g2_157 = tt.splat %i_tc2_13 : i64 -> tensor<16xi64> loc(#loc640) + %p_g2_158 = arith.addi %p_g2_157, %p_g2_156 : tensor<16xi64> loc(#loc640) + %p_g2_159 = arith.constant 32 : i32 loc(#loc641) + %p_g2_160 = arith.constant 32 : i64 loc(#loc641) + %p_g2_161 = arith.constant dense<32> : tensor<16xi64> loc(#loc641) + %p_g2_162 = arith.muli %p_g2_158, %p_g2_161 : tensor<16xi64> loc(#loc641) + %p_g2_163 = tt.splat %p_g2_155 : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc642) + %p_g2_164 = tt.addptr %p_g2_163, %p_g2_162 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc642) + %p_g3 = arith.constant 32 : i32 loc(#loc643) + %p_g3_165 = arith.constant 32 : i64 loc(#loc643) + %p_g3_166 = arith.muli %7, %p_g3_165 : i64 loc(#loc643) + %p_g3_167 = tt.addptr %g, %p_g3_166 : !tt.ptr, i64 loc(#loc644) + %p_g3_168 = tt.addptr %p_g3_167, %5 : !tt.ptr, i64 loc(#loc645) + %p_g3_169 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc646) + %p_g3_170 = tt.splat %i_tc3_18 : i64 -> tensor<16xi64> loc(#loc646) + %p_g3_171 = arith.addi %p_g3_170, %p_g3_169 : tensor<16xi64> loc(#loc646) + %p_g3_172 = arith.constant 32 : i32 loc(#loc647) + %p_g3_173 = arith.constant 32 : i64 loc(#loc647) + %p_g3_174 = arith.constant dense<32> : tensor<16xi64> loc(#loc647) + %p_g3_175 = arith.muli %p_g3_171, %p_g3_174 : tensor<16xi64> loc(#loc647) + %p_g3_176 = tt.splat %p_g3_168 : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc648) + %p_g3_177 = tt.addptr %p_g3_176, %p_g3_175 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc648) + %b_g0 = arith.constant 0.000000e+00 : f32 loc(#loc649) + %b_g0_178 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc649) + %b_g0_179 = tt.load %p_g0_138, %m_tc0_42, %b_g0_178 : tensor<16x!tt.ptr> loc(#loc649) + %b_g1 = arith.constant 0.000000e+00 : f32 loc(#loc650) + %b_g1_180 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc650) + %b_g1_181 = tt.load %p_g1_151, %m_tc1_47, %b_g1_180 : tensor<16x!tt.ptr> loc(#loc650) + %b_g2 = arith.constant 0.000000e+00 : f32 loc(#loc651) + %b_g2_182 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc651) + %b_g2_183 = tt.load %p_g2_164, %m_tc2_52, %b_g2_182 : tensor<16x!tt.ptr> loc(#loc651) + %b_g3 = arith.constant 0.000000e+00 : f32 loc(#loc652) + %b_g3_184 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc652) + %b_g3_185 = tt.load %p_g3_177, %m_tc3_57, %b_g3_184 : tensor<16x!tt.ptr> loc(#loc652) + %b_A00 = tt.call @triton.language.standard.zeros__Tc16_c16T_cfp32() : () -> tensor<16x16xf32> loc(#loc653) + %b_A11 = tt.call @triton.language.standard.zeros__Tc16_c16T_cfp32() : () -> tensor<16x16xf32> loc(#loc1063) + %b_A22 = tt.call @triton.language.standard.zeros__Tc16_c16T_cfp32() : () -> tensor<16x16xf32> loc(#loc1084) + %b_A33 = tt.call @triton.language.standard.zeros__Tc16_c16T_cfp32() : () -> tensor<16x16xf32> loc(#loc1094) + %b_A10 = tt.call @triton.language.standard.zeros__Tc16_c16T_cfp32() : () -> tensor<16x16xf32> loc(#loc1066) + %b_A20 = tt.call @triton.language.standard.zeros__Tc16_c16T_cfp32() : () -> tensor<16x16xf32> loc(#loc1086) + %b_A21 = tt.call @triton.language.standard.zeros__Tc16_c16T_cfp32() : () -> tensor<16x16xf32> loc(#loc1087) + %b_A30 = tt.call @triton.language.standard.zeros__Tc16_c16T_cfp32() : () -> tensor<16x16xf32> loc(#loc1095) + %b_A31 = tt.call @triton.language.standard.zeros__Tc16_c16T_cfp32() : () -> tensor<16x16xf32> loc(#loc1096) + %b_A32 = tt.call @triton.language.standard.zeros__Tc16_c16T_cfp32() : () -> tensor<16x16xf32> loc(#loc1097) + tt.call @triton.language.standard.cdiv__c128_c64() : () -> () loc(#loc108) + tt.call @triton.language.standard.cdiv__c128_c64() : () -> () loc(#loc108) + %c0_i32 = arith.constant 0 : i32 loc(#loc109) + %c2_i32 = arith.constant 2 : i32 loc(#loc109) + %c1_i32 = arith.constant 1 : i32 loc(#loc109) + %15 = arith.bitcast %c0_i32 : i32 to i32 loc(#loc109) + %16 = arith.bitcast %c2_i32 : i32 to i32 loc(#loc109) + %17 = arith.bitcast %c1_i32 : i32 to i32 loc(#loc109) + %18 = ub.poison : i32 loc(#loc109) + %b_A32_186:10 = scf.for %i_k = %15 to %16 step %17 iter_args(%b_A00_852 = %b_A00, %b_A11_853 = %b_A11, %b_A22_854 = %b_A22, %b_A33_855 = %b_A33, %b_A10_856 = %b_A10, %b_A20_857 = %b_A20, %b_A21_858 = %b_A21, %b_A30_859 = %b_A30, %b_A31_860 = %b_A31, %b_A32_861 = %b_A32) -> (tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>) : i32 { + %o_k = arith.constant 64 : i32 loc(#loc664) + %o_k_862 = arith.constant 64 : i32 loc(#loc664) + %o_k_863 = arith.extsi %i_k : i32 to i64 loc(#loc664) + %o_k_864 = arith.extsi %o_k_862 : i32 to i64 loc(#loc664) + %o_k_865 = arith.muli %o_k_863, %o_k_864 : i64 loc(#loc664) + %o_k_866 = arith.constant 2147483647 : i64 loc(#loc664) + %o_k_867 = arith.constant -2147483648 : i64 loc(#loc664) + %o_k_868 = arith.cmpi sle, %o_k_865, %o_k_866 : i64 loc(#loc664) + %o_k_869 = arith.cmpi sge, %o_k_865, %o_k_867 : i64 loc(#loc664) + %o_k_870 = arith.andi %o_k_868, %o_k_869 : i1 loc(#loc664) + %o_k_871 = arith.muli %i_k, %o_k_862 : i32 loc(#loc664) + %o_k_872 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc665) + %o_k_873 = tt.splat %o_k_871 : i32 -> tensor<64xi32> loc(#loc666) + %o_k_874 = arith.extsi %o_k_873 : tensor<64xi32> to tensor<64xi64> loc(#loc666) + %o_k_875 = arith.extsi %o_k_872 : tensor<64xi32> to tensor<64xi64> loc(#loc666) + %o_k_876 = arith.addi %o_k_874, %o_k_875 : tensor<64xi64> loc(#loc666) + %o_k_877 = arith.constant 2147483647 : i64 loc(#loc666) + %o_k_878 = arith.constant -2147483648 : i64 loc(#loc666) + %o_k_879 = arith.constant dense<2147483647> : tensor<64xi64> loc(#loc666) + %o_k_880 = arith.cmpi sle, %o_k_876, %o_k_879 : tensor<64xi64> loc(#loc666) + %o_k_881 = arith.constant dense<-2147483648> : tensor<64xi64> loc(#loc666) + %o_k_882 = arith.cmpi sge, %o_k_876, %o_k_881 : tensor<64xi64> loc(#loc666) + %o_k_883 = arith.andi %o_k_880, %o_k_882 : tensor<64xi1> loc(#loc666) + %o_k_884 = arith.addi %o_k_873, %o_k_872 : tensor<64xi32> loc(#loc666) + %p_k0 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc667) + %p_k0_885 = tt.splat %i_tc0_3 : i64 -> tensor<16xi64> loc(#loc667) + %p_k0_886 = arith.addi %p_k0_885, %p_k0 : tensor<16xi64> loc(#loc667) + %p_k0_887 = tt.expand_dims %p_k0_886 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc668) + %p_k0_888 = arith.constant 4096 : i32 loc(#loc669) + %p_k0_889 = arith.constant 4096 : i64 loc(#loc669) + %p_k0_890 = arith.constant dense<4096> : tensor<16x1xi64> loc(#loc669) + %p_k0_891 = arith.muli %p_k0_887, %p_k0_890 : tensor<16x1xi64> loc(#loc669) + %p_k0_892 = tt.splat %k_29 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc670) + %p_k0_893 = tt.addptr %p_k0_892, %p_k0_891 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc670) + %p_k0_894 = tt.expand_dims %o_k_884 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc671) + %p_k0_895 = tt.broadcast %p_k0_893 : tensor<16x1x!tt.ptr> -> tensor<16x64x!tt.ptr> loc(#loc672) + %p_k0_896 = tt.broadcast %p_k0_894 : tensor<1x64xi32> -> tensor<16x64xi32> loc(#loc672) + %p_k0_897 = tt.addptr %p_k0_895, %p_k0_896 : tensor<16x64x!tt.ptr>, tensor<16x64xi32> loc(#loc672) + %b_k0 = tt.expand_dims %m_tc0_42 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc673) + %b_k0_898 = tt.expand_dims %o_k_884 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc674) + %b_k0_899 = arith.constant 128 : i32 loc(#loc675) + %b_k0_900 = arith.constant dense<128> : tensor<1x64xi32> loc(#loc675) + %b_k0_901 = arith.cmpi slt, %b_k0_898, %b_k0_900 : tensor<1x64xi32> loc(#loc675) + %b_k0_902 = tt.broadcast %b_k0 : tensor<16x1xi1> -> tensor<16x64xi1> loc(#loc676) + %b_k0_903 = tt.broadcast %b_k0_901 : tensor<1x64xi1> -> tensor<16x64xi1> loc(#loc676) + %b_k0_904 = arith.andi %b_k0_902, %b_k0_903 : tensor<16x64xi1> loc(#loc676) + %b_k0_905 = arith.constant 0.000000e+00 : f32 loc(#loc677) + %b_k0_906 = arith.constant dense<0.000000e+00> : tensor<16x64xf32> loc(#loc677) + %b_k0_907 = arith.truncf %b_k0_906 : tensor<16x64xf32> to tensor<16x64xbf16> loc(#loc677) + %b_k0_908 = tt.load %p_k0_897, %b_k0_904, %b_k0_907 : tensor<16x64x!tt.ptr> loc(#loc677) + %b_A00_909 = tt.trans %b_k0_908 {order = array} : tensor<16x64xbf16> -> tensor<64x16xbf16> loc(#loc678) + %b_A00_910 = arith.constant 0.000000e+00 : f32 loc(#loc679) + %b_A00_911 = tt.dot %b_k0_908, %b_A00_909, %b_A00_852, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc679) + %61 = arith.extsi %T : i32 to i64 loc(#loc126) + %62 = arith.cmpi slt, %i_tc1_8, %61 : i64 loc(#loc126) + %63:9 = scf.if %62 -> (tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>) { + %p_k1 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc680) + %p_k1_912 = tt.splat %i_tc1_8 : i64 -> tensor<16xi64> loc(#loc680) + %p_k1_913 = arith.addi %p_k1_912, %p_k1 : tensor<16xi64> loc(#loc680) + %p_k1_914 = tt.expand_dims %p_k1_913 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc681) + %p_k1_915 = arith.constant 4096 : i32 loc(#loc682) + %p_k1_916 = arith.constant 4096 : i64 loc(#loc682) + %p_k1_917 = arith.constant dense<4096> : tensor<16x1xi64> loc(#loc682) + %p_k1_918 = arith.muli %p_k1_914, %p_k1_917 : tensor<16x1xi64> loc(#loc682) + %p_k1_919 = tt.splat %k_29 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc683) + %p_k1_920 = tt.addptr %p_k1_919, %p_k1_918 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc683) + %p_k1_921 = tt.expand_dims %o_k_884 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc684) + %p_k1_922 = tt.broadcast %p_k1_920 : tensor<16x1x!tt.ptr> -> tensor<16x64x!tt.ptr> loc(#loc685) + %p_k1_923 = tt.broadcast %p_k1_921 : tensor<1x64xi32> -> tensor<16x64xi32> loc(#loc685) + %p_k1_924 = tt.addptr %p_k1_922, %p_k1_923 : tensor<16x64x!tt.ptr>, tensor<16x64xi32> loc(#loc685) + %b_k1 = tt.expand_dims %m_tc1_47 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc686) + %b_k1_925 = tt.expand_dims %o_k_884 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc687) + %b_k1_926 = arith.constant 128 : i32 loc(#loc688) + %b_k1_927 = arith.constant dense<128> : tensor<1x64xi32> loc(#loc688) + %b_k1_928 = arith.cmpi slt, %b_k1_925, %b_k1_927 : tensor<1x64xi32> loc(#loc688) + %b_k1_929 = tt.broadcast %b_k1 : tensor<16x1xi1> -> tensor<16x64xi1> loc(#loc689) + %b_k1_930 = tt.broadcast %b_k1_928 : tensor<1x64xi1> -> tensor<16x64xi1> loc(#loc689) + %b_k1_931 = arith.andi %b_k1_929, %b_k1_930 : tensor<16x64xi1> loc(#loc689) + %b_k1_932 = arith.constant 0.000000e+00 : f32 loc(#loc690) + %b_k1_933 = arith.constant dense<0.000000e+00> : tensor<16x64xf32> loc(#loc690) + %b_k1_934 = arith.truncf %b_k1_933 : tensor<16x64xf32> to tensor<16x64xbf16> loc(#loc690) + %b_k1_935 = tt.load %p_k1_924, %b_k1_931, %b_k1_934 : tensor<16x64x!tt.ptr> loc(#loc690) + %b_A11_936 = tt.trans %b_k1_935 {order = array} : tensor<16x64xbf16> -> tensor<64x16xbf16> loc(#loc691) + %b_A11_937 = arith.constant 0.000000e+00 : f32 loc(#loc692) + %b_A11_938 = tt.dot %b_k1_935, %b_A11_936, %b_A11_853, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc1073) + %b_A10_939 = tt.trans %b_k0_908 {order = array} : tensor<16x64xbf16> -> tensor<64x16xbf16> loc(#loc693) + %b_A10_940 = arith.constant 0.000000e+00 : f32 loc(#loc694) + %b_A10_941 = tt.dot %b_k1_935, %b_A10_939, %b_A10_856, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc1074) + %64 = arith.extsi %T : i32 to i64 loc(#loc143) + %65 = arith.cmpi slt, %i_tc2_13, %64 : i64 loc(#loc143) + %b_A32_942:7 = scf.if %65 -> (tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>) { + %p_k2 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc696) + %p_k2_943 = tt.splat %i_tc2_13 : i64 -> tensor<16xi64> loc(#loc696) + %p_k2_944 = arith.addi %p_k2_943, %p_k2 : tensor<16xi64> loc(#loc696) + %p_k2_945 = tt.expand_dims %p_k2_944 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc697) + %p_k2_946 = arith.constant 4096 : i32 loc(#loc698) + %p_k2_947 = arith.constant 4096 : i64 loc(#loc698) + %p_k2_948 = arith.constant dense<4096> : tensor<16x1xi64> loc(#loc698) + %p_k2_949 = arith.muli %p_k2_945, %p_k2_948 : tensor<16x1xi64> loc(#loc698) + %p_k2_950 = tt.splat %k_29 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc699) + %p_k2_951 = tt.addptr %p_k2_950, %p_k2_949 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc699) + %p_k2_952 = tt.expand_dims %o_k_884 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc700) + %p_k2_953 = tt.broadcast %p_k2_951 : tensor<16x1x!tt.ptr> -> tensor<16x64x!tt.ptr> loc(#loc701) + %p_k2_954 = tt.broadcast %p_k2_952 : tensor<1x64xi32> -> tensor<16x64xi32> loc(#loc701) + %p_k2_955 = tt.addptr %p_k2_953, %p_k2_954 : tensor<16x64x!tt.ptr>, tensor<16x64xi32> loc(#loc701) + %b_k2 = tt.expand_dims %m_tc2_52 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc702) + %b_k2_956 = tt.expand_dims %o_k_884 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc703) + %b_k2_957 = arith.constant 128 : i32 loc(#loc704) + %b_k2_958 = arith.constant dense<128> : tensor<1x64xi32> loc(#loc704) + %b_k2_959 = arith.cmpi slt, %b_k2_956, %b_k2_958 : tensor<1x64xi32> loc(#loc704) + %b_k2_960 = tt.broadcast %b_k2 : tensor<16x1xi1> -> tensor<16x64xi1> loc(#loc705) + %b_k2_961 = tt.broadcast %b_k2_959 : tensor<1x64xi1> -> tensor<16x64xi1> loc(#loc705) + %b_k2_962 = arith.andi %b_k2_960, %b_k2_961 : tensor<16x64xi1> loc(#loc705) + %b_k2_963 = arith.constant 0.000000e+00 : f32 loc(#loc706) + %b_k2_964 = arith.constant dense<0.000000e+00> : tensor<16x64xf32> loc(#loc706) + %b_k2_965 = arith.truncf %b_k2_964 : tensor<16x64xf32> to tensor<16x64xbf16> loc(#loc706) + %b_k2_966 = tt.load %p_k2_955, %b_k2_962, %b_k2_965 : tensor<16x64x!tt.ptr> loc(#loc706) + %b_A22_967 = tt.trans %b_k2_966 {order = array} : tensor<16x64xbf16> -> tensor<64x16xbf16> loc(#loc707) + %b_A22_968 = arith.constant 0.000000e+00 : f32 loc(#loc708) + %b_A22_969 = tt.dot %b_k2_966, %b_A22_967, %b_A22_854, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc1076) + %b_A20_970 = tt.trans %b_k0_908 {order = array} : tensor<16x64xbf16> -> tensor<64x16xbf16> loc(#loc709) + %b_A20_971 = arith.constant 0.000000e+00 : f32 loc(#loc710) + %b_A20_972 = tt.dot %b_k2_966, %b_A20_970, %b_A20_857, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc1077) + %b_A21_973 = tt.trans %b_k1_935 {order = array} : tensor<16x64xbf16> -> tensor<64x16xbf16> loc(#loc711) + %b_A21_974 = arith.constant 0.000000e+00 : f32 loc(#loc712) + %b_A21_975 = tt.dot %b_k2_966, %b_A21_973, %b_A21_858, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc1078) + %66 = arith.extsi %T : i32 to i64 loc(#loc162) + %67 = arith.cmpi slt, %i_tc3_18, %66 : i64 loc(#loc162) + %b_A32_976:4 = scf.if %67 -> (tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>) { + %p_k3 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc714) + %p_k3_977 = tt.splat %i_tc3_18 : i64 -> tensor<16xi64> loc(#loc714) + %p_k3_978 = arith.addi %p_k3_977, %p_k3 : tensor<16xi64> loc(#loc714) + %p_k3_979 = tt.expand_dims %p_k3_978 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc715) + %p_k3_980 = arith.constant 4096 : i32 loc(#loc716) + %p_k3_981 = arith.constant 4096 : i64 loc(#loc716) + %p_k3_982 = arith.constant dense<4096> : tensor<16x1xi64> loc(#loc716) + %p_k3_983 = arith.muli %p_k3_979, %p_k3_982 : tensor<16x1xi64> loc(#loc716) + %p_k3_984 = tt.splat %k_29 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc717) + %p_k3_985 = tt.addptr %p_k3_984, %p_k3_983 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc717) + %p_k3_986 = tt.expand_dims %o_k_884 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc718) + %p_k3_987 = tt.broadcast %p_k3_985 : tensor<16x1x!tt.ptr> -> tensor<16x64x!tt.ptr> loc(#loc719) + %p_k3_988 = tt.broadcast %p_k3_986 : tensor<1x64xi32> -> tensor<16x64xi32> loc(#loc719) + %p_k3_989 = tt.addptr %p_k3_987, %p_k3_988 : tensor<16x64x!tt.ptr>, tensor<16x64xi32> loc(#loc719) + %b_k3 = tt.expand_dims %m_tc3_57 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc720) + %b_k3_990 = tt.expand_dims %o_k_884 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc721) + %b_k3_991 = arith.constant 128 : i32 loc(#loc722) + %b_k3_992 = arith.constant dense<128> : tensor<1x64xi32> loc(#loc722) + %b_k3_993 = arith.cmpi slt, %b_k3_990, %b_k3_992 : tensor<1x64xi32> loc(#loc722) + %b_k3_994 = tt.broadcast %b_k3 : tensor<16x1xi1> -> tensor<16x64xi1> loc(#loc723) + %b_k3_995 = tt.broadcast %b_k3_993 : tensor<1x64xi1> -> tensor<16x64xi1> loc(#loc723) + %b_k3_996 = arith.andi %b_k3_994, %b_k3_995 : tensor<16x64xi1> loc(#loc723) + %b_k3_997 = arith.constant 0.000000e+00 : f32 loc(#loc724) + %b_k3_998 = arith.constant dense<0.000000e+00> : tensor<16x64xf32> loc(#loc724) + %b_k3_999 = arith.truncf %b_k3_998 : tensor<16x64xf32> to tensor<16x64xbf16> loc(#loc724) + %b_k3_1000 = tt.load %p_k3_989, %b_k3_996, %b_k3_999 : tensor<16x64x!tt.ptr> loc(#loc724) + %b_A33_1001 = tt.trans %b_k3_1000 {order = array} : tensor<16x64xbf16> -> tensor<64x16xbf16> loc(#loc725) + %b_A33_1002 = arith.constant 0.000000e+00 : f32 loc(#loc726) + %b_A33_1003 = tt.dot %b_k3_1000, %b_A33_1001, %b_A33_855, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc1080) + %b_A30_1004 = tt.trans %b_k0_908 {order = array} : tensor<16x64xbf16> -> tensor<64x16xbf16> loc(#loc727) + %b_A30_1005 = arith.constant 0.000000e+00 : f32 loc(#loc728) + %b_A30_1006 = tt.dot %b_k3_1000, %b_A30_1004, %b_A30_859, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc1081) + %b_A31_1007 = tt.trans %b_k1_935 {order = array} : tensor<16x64xbf16> -> tensor<64x16xbf16> loc(#loc729) + %b_A31_1008 = arith.constant 0.000000e+00 : f32 loc(#loc730) + %b_A31_1009 = tt.dot %b_k3_1000, %b_A31_1007, %b_A31_860, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc1082) + %b_A32_1010 = tt.trans %b_k2_966 {order = array} : tensor<16x64xbf16> -> tensor<64x16xbf16> loc(#loc731) + %b_A32_1011 = arith.constant 0.000000e+00 : f32 loc(#loc732) + %b_A32_1012 = tt.dot %b_k3_1000, %b_A32_1010, %b_A32_861, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc1083) + scf.yield %b_A33_1003, %b_A30_1006, %b_A31_1009, %b_A32_1012 : tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32> loc(#loc1083) + } else { + scf.yield %b_A33_855, %b_A30_859, %b_A31_860, %b_A32_861 : tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32> loc(#loc163) + } loc(#loc1100) + scf.yield %b_A22_969, %b_A32_976#0, %b_A20_972, %b_A21_975, %b_A32_976#1, %b_A32_976#2, %b_A32_976#3 : tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32> loc(#loc1100) + } else { + scf.yield %b_A22_854, %b_A33_855, %b_A20_857, %b_A21_858, %b_A30_859, %b_A31_860, %b_A32_861 : tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32> loc(#loc144) + } loc(#loc1106) + scf.yield %b_A11_938, %b_A32_942#0, %b_A32_942#1, %b_A10_941, %b_A32_942#2, %b_A32_942#3, %b_A32_942#4, %b_A32_942#5, %b_A32_942#6 : tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32> loc(#loc1106) + } else { + scf.yield %b_A11_853, %b_A22_854, %b_A33_855, %b_A10_856, %b_A20_857, %b_A21_858, %b_A30_859, %b_A31_860, %b_A32_861 : tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32> loc(#loc127) + } loc(#loc127) + scf.yield %b_A00_911, %63#0, %63#1, %63#2, %63#3, %63#4, %63#5, %63#6, %63#7, %63#8 : tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32> loc(#loc183) + } loc(#loc1109) + %m_d = tt.expand_dims %o_i {axis = 1 : i32} : tensor<16xi32> -> tensor<16x1xi32> loc(#loc733) + %m_d_187 = tt.expand_dims %o_i {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc734) + %m_d_188 = tt.broadcast %m_d : tensor<16x1xi32> -> tensor<16x16xi32> loc(#loc735) + %m_d_189 = tt.broadcast %m_d_187 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc735) + %m_d_190 = arith.cmpi sgt, %m_d_188, %m_d_189 : tensor<16x16xi32> loc(#loc735) + %m_I = tt.expand_dims %o_i {axis = 1 : i32} : tensor<16xi32> -> tensor<16x1xi32> loc(#loc736) + %m_I_191 = tt.expand_dims %o_i {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc737) + %m_I_192 = tt.broadcast %m_I : tensor<16x1xi32> -> tensor<16x16xi32> loc(#loc738) + %m_I_193 = tt.broadcast %m_I_191 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc738) + %m_I_194 = arith.cmpi eq, %m_I_192, %m_I_193 : tensor<16x16xi32> loc(#loc738) + %b_A00_195 = tt.expand_dims %m_tc0_42 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc739) + %b_A00_196 = tt.broadcast %b_A00_195 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc740) + %b_A00_197 = arith.andi %m_d_190, %b_A00_196 : tensor<16x16xi1> loc(#loc740) + %b_A00_198 = tt.expand_dims %m_tc0_42 {axis = 0 : i32} : tensor<16xi1> -> tensor<1x16xi1> loc(#loc741) + %b_A00_199 = tt.broadcast %b_A00_198 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc742) + %b_A00_200 = arith.andi %b_A00_197, %b_A00_199 : tensor<16x16xi1> loc(#loc742) + %b_A00_201 = tt.expand_dims %b_g0_179 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc743) + %b_A00_202 = tt.expand_dims %b_g0_179 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc744) + %b_A00_203 = tt.broadcast %b_A00_201 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc745) + %b_A00_204 = tt.broadcast %b_A00_202 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc745) + %b_A00_205 = arith.subf %b_A00_203, %b_A00_204 : tensor<16x16xf32> loc(#loc745) + %b_A00_206 = tt.call @fla.ops.utils.op.exp2__fp32S16_16S(%b_A00_205) : (tensor<16x16xf32>) -> tensor<16x16xf32> loc(#loc746) + %b_A00_207 = arith.constant 0.000000e+00 : f32 loc(#loc747) + %b_A00_208 = arith.constant 0.000000e+00 : f32 loc(#loc747) + %b_A00_209 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc747) + %b_A00_210 = arith.select %b_A00_200, %b_A00_206, %b_A00_209 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc747) + %b_A00_211 = arith.mulf %b_A32_186#0, %b_A00_210 : tensor<16x16xf32> loc(#loc748) + %b_A11_212 = tt.expand_dims %m_tc1_47 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc749) + %b_A11_213 = tt.broadcast %b_A11_212 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc750) + %b_A11_214 = arith.andi %m_d_190, %b_A11_213 : tensor<16x16xi1> loc(#loc750) + %b_A11_215 = tt.expand_dims %m_tc1_47 {axis = 0 : i32} : tensor<16xi1> -> tensor<1x16xi1> loc(#loc751) + %b_A11_216 = tt.broadcast %b_A11_215 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc752) + %b_A11_217 = arith.andi %b_A11_214, %b_A11_216 : tensor<16x16xi1> loc(#loc752) + %b_A11_218 = tt.expand_dims %b_g1_181 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc753) + %b_A11_219 = tt.expand_dims %b_g1_181 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc754) + %b_A11_220 = tt.broadcast %b_A11_218 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc755) + %b_A11_221 = tt.broadcast %b_A11_219 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc755) + %b_A11_222 = arith.subf %b_A11_220, %b_A11_221 : tensor<16x16xf32> loc(#loc755) + %b_A11_223 = tt.call @fla.ops.utils.op.exp2__fp32S16_16S(%b_A11_222) : (tensor<16x16xf32>) -> tensor<16x16xf32> loc(#loc756) + %b_A11_224 = arith.constant 0.000000e+00 : f32 loc(#loc757) + %b_A11_225 = arith.constant 0.000000e+00 : f32 loc(#loc757) + %b_A11_226 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc757) + %b_A11_227 = arith.select %b_A11_217, %b_A11_223, %b_A11_226 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc757) + %b_A11_228 = arith.mulf %b_A32_186#1, %b_A11_227 : tensor<16x16xf32> loc(#loc758) + %b_A22_229 = tt.expand_dims %m_tc2_52 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc759) + %b_A22_230 = tt.broadcast %b_A22_229 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc760) + %b_A22_231 = arith.andi %m_d_190, %b_A22_230 : tensor<16x16xi1> loc(#loc760) + %b_A22_232 = tt.expand_dims %m_tc2_52 {axis = 0 : i32} : tensor<16xi1> -> tensor<1x16xi1> loc(#loc761) + %b_A22_233 = tt.broadcast %b_A22_232 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc762) + %b_A22_234 = arith.andi %b_A22_231, %b_A22_233 : tensor<16x16xi1> loc(#loc762) + %b_A22_235 = tt.expand_dims %b_g2_183 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc763) + %b_A22_236 = tt.expand_dims %b_g2_183 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc764) + %b_A22_237 = tt.broadcast %b_A22_235 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc765) + %b_A22_238 = tt.broadcast %b_A22_236 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc765) + %b_A22_239 = arith.subf %b_A22_237, %b_A22_238 : tensor<16x16xf32> loc(#loc765) + %b_A22_240 = tt.call @fla.ops.utils.op.exp2__fp32S16_16S(%b_A22_239) : (tensor<16x16xf32>) -> tensor<16x16xf32> loc(#loc766) + %b_A22_241 = arith.constant 0.000000e+00 : f32 loc(#loc767) + %b_A22_242 = arith.constant 0.000000e+00 : f32 loc(#loc767) + %b_A22_243 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc767) + %b_A22_244 = arith.select %b_A22_234, %b_A22_240, %b_A22_243 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc767) + %b_A22_245 = arith.mulf %b_A32_186#2, %b_A22_244 : tensor<16x16xf32> loc(#loc768) + %b_A33_246 = tt.expand_dims %m_tc3_57 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc769) + %b_A33_247 = tt.broadcast %b_A33_246 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc770) + %b_A33_248 = arith.andi %m_d_190, %b_A33_247 : tensor<16x16xi1> loc(#loc770) + %b_A33_249 = tt.expand_dims %m_tc3_57 {axis = 0 : i32} : tensor<16xi1> -> tensor<1x16xi1> loc(#loc771) + %b_A33_250 = tt.broadcast %b_A33_249 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc772) + %b_A33_251 = arith.andi %b_A33_248, %b_A33_250 : tensor<16x16xi1> loc(#loc772) + %b_A33_252 = tt.expand_dims %b_g3_185 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc773) + %b_A33_253 = tt.expand_dims %b_g3_185 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc774) + %b_A33_254 = tt.broadcast %b_A33_252 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc775) + %b_A33_255 = tt.broadcast %b_A33_253 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc775) + %b_A33_256 = arith.subf %b_A33_254, %b_A33_255 : tensor<16x16xf32> loc(#loc775) + %b_A33_257 = tt.call @fla.ops.utils.op.exp2__fp32S16_16S(%b_A33_256) : (tensor<16x16xf32>) -> tensor<16x16xf32> loc(#loc776) + %b_A33_258 = arith.constant 0.000000e+00 : f32 loc(#loc777) + %b_A33_259 = arith.constant 0.000000e+00 : f32 loc(#loc777) + %b_A33_260 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc777) + %b_A33_261 = arith.select %b_A33_251, %b_A33_257, %b_A33_260 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc777) + %b_A33_262 = arith.mulf %b_A32_186#3, %b_A33_261 : tensor<16x16xf32> loc(#loc778) + %b_A10_263 = tt.expand_dims %m_tc1_47 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc779) + %b_A10_264 = tt.expand_dims %m_tc0_42 {axis = 0 : i32} : tensor<16xi1> -> tensor<1x16xi1> loc(#loc780) + %b_A10_265 = tt.broadcast %b_A10_263 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc781) + %b_A10_266 = tt.broadcast %b_A10_264 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc781) + %b_A10_267 = arith.andi %b_A10_265, %b_A10_266 : tensor<16x16xi1> loc(#loc781) + %b_A10_268 = tt.expand_dims %b_g1_181 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc782) + %b_A10_269 = tt.expand_dims %b_g0_179 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc783) + %b_A10_270 = tt.broadcast %b_A10_268 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc784) + %b_A10_271 = tt.broadcast %b_A10_269 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc784) + %b_A10_272 = arith.subf %b_A10_270, %b_A10_271 : tensor<16x16xf32> loc(#loc784) + %b_A10_273 = tt.call @fla.ops.utils.op.exp2__fp32S16_16S(%b_A10_272) : (tensor<16x16xf32>) -> tensor<16x16xf32> loc(#loc785) + %b_A10_274 = arith.constant 0.000000e+00 : f32 loc(#loc786) + %b_A10_275 = arith.constant 0.000000e+00 : f32 loc(#loc786) + %b_A10_276 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc786) + %b_A10_277 = arith.select %b_A10_267, %b_A10_273, %b_A10_276 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc786) + %b_A10_278 = arith.mulf %b_A32_186#4, %b_A10_277 : tensor<16x16xf32> loc(#loc787) + %b_A20_279 = tt.expand_dims %m_tc2_52 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc788) + %b_A20_280 = tt.expand_dims %m_tc0_42 {axis = 0 : i32} : tensor<16xi1> -> tensor<1x16xi1> loc(#loc789) + %b_A20_281 = tt.broadcast %b_A20_279 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc790) + %b_A20_282 = tt.broadcast %b_A20_280 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc790) + %b_A20_283 = arith.andi %b_A20_281, %b_A20_282 : tensor<16x16xi1> loc(#loc790) + %b_A20_284 = tt.expand_dims %b_g2_183 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc791) + %b_A20_285 = tt.expand_dims %b_g0_179 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc792) + %b_A20_286 = tt.broadcast %b_A20_284 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc793) + %b_A20_287 = tt.broadcast %b_A20_285 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc793) + %b_A20_288 = arith.subf %b_A20_286, %b_A20_287 : tensor<16x16xf32> loc(#loc793) + %b_A20_289 = tt.call @fla.ops.utils.op.exp2__fp32S16_16S(%b_A20_288) : (tensor<16x16xf32>) -> tensor<16x16xf32> loc(#loc794) + %b_A20_290 = arith.constant 0.000000e+00 : f32 loc(#loc795) + %b_A20_291 = arith.constant 0.000000e+00 : f32 loc(#loc795) + %b_A20_292 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc795) + %b_A20_293 = arith.select %b_A20_283, %b_A20_289, %b_A20_292 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc795) + %b_A20_294 = arith.mulf %b_A32_186#5, %b_A20_293 : tensor<16x16xf32> loc(#loc796) + %b_A21_295 = tt.expand_dims %m_tc2_52 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc797) + %b_A21_296 = tt.expand_dims %m_tc1_47 {axis = 0 : i32} : tensor<16xi1> -> tensor<1x16xi1> loc(#loc798) + %b_A21_297 = tt.broadcast %b_A21_295 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc799) + %b_A21_298 = tt.broadcast %b_A21_296 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc799) + %b_A21_299 = arith.andi %b_A21_297, %b_A21_298 : tensor<16x16xi1> loc(#loc799) + %b_A21_300 = tt.expand_dims %b_g2_183 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc800) + %b_A21_301 = tt.expand_dims %b_g1_181 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc801) + %b_A21_302 = tt.broadcast %b_A21_300 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc802) + %b_A21_303 = tt.broadcast %b_A21_301 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc802) + %b_A21_304 = arith.subf %b_A21_302, %b_A21_303 : tensor<16x16xf32> loc(#loc802) + %b_A21_305 = tt.call @fla.ops.utils.op.exp2__fp32S16_16S(%b_A21_304) : (tensor<16x16xf32>) -> tensor<16x16xf32> loc(#loc803) + %b_A21_306 = arith.constant 0.000000e+00 : f32 loc(#loc804) + %b_A21_307 = arith.constant 0.000000e+00 : f32 loc(#loc804) + %b_A21_308 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc804) + %b_A21_309 = arith.select %b_A21_299, %b_A21_305, %b_A21_308 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc804) + %b_A21_310 = arith.mulf %b_A32_186#6, %b_A21_309 : tensor<16x16xf32> loc(#loc805) + %b_A30_311 = tt.expand_dims %m_tc3_57 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc806) + %b_A30_312 = tt.expand_dims %m_tc0_42 {axis = 0 : i32} : tensor<16xi1> -> tensor<1x16xi1> loc(#loc807) + %b_A30_313 = tt.broadcast %b_A30_311 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc808) + %b_A30_314 = tt.broadcast %b_A30_312 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc808) + %b_A30_315 = arith.andi %b_A30_313, %b_A30_314 : tensor<16x16xi1> loc(#loc808) + %b_A30_316 = tt.expand_dims %b_g3_185 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc809) + %b_A30_317 = tt.expand_dims %b_g0_179 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc810) + %b_A30_318 = tt.broadcast %b_A30_316 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc811) + %b_A30_319 = tt.broadcast %b_A30_317 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc811) + %b_A30_320 = arith.subf %b_A30_318, %b_A30_319 : tensor<16x16xf32> loc(#loc811) + %b_A30_321 = tt.call @fla.ops.utils.op.exp2__fp32S16_16S(%b_A30_320) : (tensor<16x16xf32>) -> tensor<16x16xf32> loc(#loc812) + %b_A30_322 = arith.constant 0.000000e+00 : f32 loc(#loc813) + %b_A30_323 = arith.constant 0.000000e+00 : f32 loc(#loc813) + %b_A30_324 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc813) + %b_A30_325 = arith.select %b_A30_315, %b_A30_321, %b_A30_324 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc813) + %b_A30_326 = arith.mulf %b_A32_186#7, %b_A30_325 : tensor<16x16xf32> loc(#loc814) + %b_A31_327 = tt.expand_dims %m_tc3_57 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc815) + %b_A31_328 = tt.expand_dims %m_tc1_47 {axis = 0 : i32} : tensor<16xi1> -> tensor<1x16xi1> loc(#loc816) + %b_A31_329 = tt.broadcast %b_A31_327 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc817) + %b_A31_330 = tt.broadcast %b_A31_328 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc817) + %b_A31_331 = arith.andi %b_A31_329, %b_A31_330 : tensor<16x16xi1> loc(#loc817) + %b_A31_332 = tt.expand_dims %b_g3_185 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc818) + %b_A31_333 = tt.expand_dims %b_g1_181 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc819) + %b_A31_334 = tt.broadcast %b_A31_332 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc820) + %b_A31_335 = tt.broadcast %b_A31_333 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc820) + %b_A31_336 = arith.subf %b_A31_334, %b_A31_335 : tensor<16x16xf32> loc(#loc820) + %b_A31_337 = tt.call @fla.ops.utils.op.exp2__fp32S16_16S(%b_A31_336) : (tensor<16x16xf32>) -> tensor<16x16xf32> loc(#loc821) + %b_A31_338 = arith.constant 0.000000e+00 : f32 loc(#loc822) + %b_A31_339 = arith.constant 0.000000e+00 : f32 loc(#loc822) + %b_A31_340 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc822) + %b_A31_341 = arith.select %b_A31_331, %b_A31_337, %b_A31_340 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc822) + %b_A31_342 = arith.mulf %b_A32_186#8, %b_A31_341 : tensor<16x16xf32> loc(#loc823) + %b_A32_343 = tt.expand_dims %m_tc3_57 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc824) + %b_A32_344 = tt.expand_dims %m_tc2_52 {axis = 0 : i32} : tensor<16xi1> -> tensor<1x16xi1> loc(#loc825) + %b_A32_345 = tt.broadcast %b_A32_343 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc826) + %b_A32_346 = tt.broadcast %b_A32_344 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc826) + %b_A32_347 = arith.andi %b_A32_345, %b_A32_346 : tensor<16x16xi1> loc(#loc826) + %b_A32_348 = tt.expand_dims %b_g3_185 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc827) + %b_A32_349 = tt.expand_dims %b_g2_183 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc828) + %b_A32_350 = tt.broadcast %b_A32_348 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc829) + %b_A32_351 = tt.broadcast %b_A32_349 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc829) + %b_A32_352 = arith.subf %b_A32_350, %b_A32_351 : tensor<16x16xf32> loc(#loc829) + %b_A32_353 = tt.call @fla.ops.utils.op.exp2__fp32S16_16S(%b_A32_352) : (tensor<16x16xf32>) -> tensor<16x16xf32> loc(#loc830) + %b_A32_354 = arith.constant 0.000000e+00 : f32 loc(#loc831) + %b_A32_355 = arith.constant 0.000000e+00 : f32 loc(#loc831) + %b_A32_356 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc831) + %b_A32_357 = arith.select %b_A32_347, %b_A32_353, %b_A32_356 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc831) + %b_A32_358 = arith.mulf %b_A32_186#9, %b_A32_357 : tensor<16x16xf32> loc(#loc832) + %b_A00_359 = tt.expand_dims %b_b0_113 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc833) + %b_A00_360 = tt.broadcast %b_A00_359 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc834) + %b_A00_361 = arith.mulf %b_A00_211, %b_A00_360 : tensor<16x16xf32> loc(#loc834) + %b_A11_362 = tt.expand_dims %b_b1_117 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc835) + %b_A11_363 = tt.broadcast %b_A11_362 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc836) + %b_A11_364 = arith.mulf %b_A11_228, %b_A11_363 : tensor<16x16xf32> loc(#loc836) + %b_A22_365 = tt.expand_dims %b_b2_121 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc837) + %b_A22_366 = tt.broadcast %b_A22_365 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc838) + %b_A22_367 = arith.mulf %b_A22_245, %b_A22_366 : tensor<16x16xf32> loc(#loc838) + %b_A33_368 = tt.expand_dims %b_b3_125 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc839) + %b_A33_369 = tt.broadcast %b_A33_368 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc840) + %b_A33_370 = arith.mulf %b_A33_262, %b_A33_369 : tensor<16x16xf32> loc(#loc840) + %b_A10_371 = tt.expand_dims %b_b1_117 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc841) + %b_A10_372 = tt.broadcast %b_A10_371 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc842) + %b_A10_373 = arith.mulf %b_A10_278, %b_A10_372 : tensor<16x16xf32> loc(#loc842) + %b_A20_374 = tt.expand_dims %b_b2_121 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc843) + %b_A20_375 = tt.broadcast %b_A20_374 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc844) + %b_A20_376 = arith.mulf %b_A20_294, %b_A20_375 : tensor<16x16xf32> loc(#loc844) + %b_A21_377 = tt.expand_dims %b_b2_121 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc845) + %b_A21_378 = tt.broadcast %b_A21_377 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc846) + %b_A21_379 = arith.mulf %b_A21_310, %b_A21_378 : tensor<16x16xf32> loc(#loc846) + %b_A30_380 = tt.expand_dims %b_b3_125 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc847) + %b_A30_381 = tt.broadcast %b_A30_380 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc848) + %b_A30_382 = arith.mulf %b_A30_326, %b_A30_381 : tensor<16x16xf32> loc(#loc848) + %b_A31_383 = tt.expand_dims %b_b3_125 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc849) + %b_A31_384 = tt.broadcast %b_A31_383 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc850) + %b_A31_385 = arith.mulf %b_A31_342, %b_A31_384 : tensor<16x16xf32> loc(#loc850) + %b_A32_386 = tt.expand_dims %b_b3_125 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc851) + %b_A32_387 = tt.broadcast %b_A32_386 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc852) + %b_A32_388 = arith.mulf %b_A32_358, %b_A32_387 : tensor<16x16xf32> loc(#loc852) + %b_Ai00 = arith.constant 0.000000e+00 : f32 loc(#loc853) + %b_Ai00_389 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc853) + %b_Ai00_390 = arith.subf %b_Ai00_389, %b_A00_361 : tensor<16x16xf32> loc(#loc853) + %b_Ai11 = arith.constant 0.000000e+00 : f32 loc(#loc854) + %b_Ai11_391 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc854) + %b_Ai11_392 = arith.subf %b_Ai11_391, %b_A11_364 : tensor<16x16xf32> loc(#loc854) + %b_Ai22 = arith.constant 0.000000e+00 : f32 loc(#loc855) + %b_Ai22_393 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc855) + %b_Ai22_394 = arith.subf %b_Ai22_393, %b_A22_367 : tensor<16x16xf32> loc(#loc855) + %b_Ai33 = arith.constant 0.000000e+00 : f32 loc(#loc856) + %b_Ai33_395 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc856) + %b_Ai33_396 = arith.subf %b_Ai33_395, %b_A33_370 : tensor<16x16xf32> loc(#loc856) + %19 = arith.extsi %T : i32 to i64 loc(#loc308) + %20 = arith.subi %19, %i_tc0_3 : i64 loc(#loc308) + %c16_i32 = arith.constant 16 : i32 loc(#loc309) + %21 = arith.extsi %c16_i32 : i32 to i64 loc(#loc309) + %22 = arith.minsi %21, %20 : i64 loc(#loc309) + %c2_i32_397 = arith.constant 2 : i32 loc(#loc310) + %c1_i32_398 = arith.constant 1 : i32 loc(#loc310) + %23 = arith.extsi %c2_i32_397 : i32 to i64 loc(#loc310) + %24 = arith.bitcast %22 : i64 to i64 loc(#loc310) + %25 = arith.extsi %c1_i32_398 : i32 to i64 loc(#loc310) + %26 = ub.poison : i64 loc(#loc310) + %b_Ai00_399 = scf.for %i = %23 to %24 step %25 iter_args(%b_Ai00_852 = %b_Ai00_390) -> (tensor<16x16xf32>) : i64 { + %b_a00 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc858) + %b_a00_853 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc858) + %b_a00_854 = arith.cmpi eq, %b_a00, %b_a00_853 : tensor<16xi64> loc(#loc858) + %b_a00_855 = tt.expand_dims %b_a00_854 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc859) + %b_a00_856 = arith.constant 0.000000e+00 : f32 loc(#loc860) + %b_a00_857 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc860) + %b_a00_858 = arith.subf %b_a00_857, %b_A00_361 : tensor<16x16xf32> loc(#loc860) + %b_a00_859 = arith.constant 0.000000e+00 : f32 loc(#loc861) + %b_a00_860 = arith.constant 0.000000e+00 : f32 loc(#loc861) + %b_a00_861 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc861) + %b_a00_862 = tt.broadcast %b_a00_855 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc861) + %b_a00_863 = arith.select %b_a00_862, %b_a00_858, %b_a00_861 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc861) + %b_a00_864 = tt.call @triton.language.standard.sum__fp32S16_16S_c0_cFalse_cNone(%b_a00_863) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc862) + %b_a00_865 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc863) + %b_a00_866 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc863) + %b_a00_867 = arith.cmpi slt, %b_a00_865, %b_a00_866 : tensor<16xi64> loc(#loc863) + %b_a00_868 = arith.constant 0.000000e+00 : f32 loc(#loc864) + %b_a00_869 = arith.constant 0.000000e+00 : f32 loc(#loc864) + %b_a00_870 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc864) + %b_a00_871 = arith.select %b_a00_867, %b_a00_864, %b_a00_870 : tensor<16xi1>, tensor<16xf32> loc(#loc864) + %b_a00_872 = tt.expand_dims %b_a00_871 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc865) + %b_a00_873 = tt.broadcast %b_a00_872 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc866) + %b_a00_874 = arith.mulf %b_a00_873, %b_Ai00_852 : tensor<16x16xf32> loc(#loc866) + %b_a00_875 = tt.call @triton.language.standard.sum__fp32S16_16S_c0_cFalse_cNone(%b_a00_874) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc867) + %b_a00_876 = arith.addf %b_a00_871, %b_a00_875 : tensor<16xf32> loc(#loc868) + %b_Ai00_877 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc869) + %b_Ai00_878 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc869) + %b_Ai00_879 = arith.cmpi eq, %b_Ai00_877, %b_Ai00_878 : tensor<16xi64> loc(#loc869) + %b_Ai00_880 = tt.expand_dims %b_Ai00_879 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc870) + %b_Ai00_881 = tt.expand_dims %b_a00_876 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc871) + %b_Ai00_882 = tt.broadcast %b_Ai00_881 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc871) + %b_Ai00_883 = tt.broadcast %b_Ai00_880 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc871) + %b_Ai00_884 = arith.select %b_Ai00_883, %b_Ai00_882, %b_Ai00_852 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc871) + scf.yield %b_Ai00_884 : tensor<16x16xf32> loc(#loc325) + } loc(#loc857) + %27 = arith.extsi %T : i32 to i64 loc(#loc326) + %28 = arith.subi %27, %i_tc1_8 : i64 loc(#loc326) + %c16_i32_400 = arith.constant 16 : i32 loc(#loc327) + %29 = arith.extsi %c16_i32_400 : i32 to i64 loc(#loc327) + %30 = arith.minsi %29, %28 : i64 loc(#loc327) + %c2_i32_401 = arith.constant 2 : i32 loc(#loc328) + %c1_i32_402 = arith.constant 1 : i32 loc(#loc328) + %31 = arith.extsi %c2_i32_401 : i32 to i64 loc(#loc328) + %32 = arith.bitcast %30 : i64 to i64 loc(#loc328) + %33 = arith.extsi %c1_i32_402 : i32 to i64 loc(#loc328) + %34 = ub.poison : i64 loc(#loc328) + %b_Ai11_403 = scf.for %i = %31 to %32 step %33 iter_args(%b_Ai11_852 = %b_Ai11_392) -> (tensor<16x16xf32>) : i64 { + %b_a11 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc873) + %b_a11_853 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc873) + %b_a11_854 = arith.cmpi eq, %b_a11, %b_a11_853 : tensor<16xi64> loc(#loc873) + %b_a11_855 = tt.expand_dims %b_a11_854 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc874) + %b_a11_856 = arith.constant 0.000000e+00 : f32 loc(#loc875) + %b_a11_857 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc875) + %b_a11_858 = arith.subf %b_a11_857, %b_A11_364 : tensor<16x16xf32> loc(#loc875) + %b_a11_859 = arith.constant 0.000000e+00 : f32 loc(#loc876) + %b_a11_860 = arith.constant 0.000000e+00 : f32 loc(#loc876) + %b_a11_861 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc876) + %b_a11_862 = tt.broadcast %b_a11_855 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc876) + %b_a11_863 = arith.select %b_a11_862, %b_a11_858, %b_a11_861 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc876) + %b_a11_864 = tt.call @triton.language.standard.sum__fp32S16_16S_c0_cFalse_cNone(%b_a11_863) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc877) + %b_a11_865 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc878) + %b_a11_866 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc878) + %b_a11_867 = arith.cmpi slt, %b_a11_865, %b_a11_866 : tensor<16xi64> loc(#loc878) + %b_a11_868 = arith.constant 0.000000e+00 : f32 loc(#loc879) + %b_a11_869 = arith.constant 0.000000e+00 : f32 loc(#loc879) + %b_a11_870 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc879) + %b_a11_871 = arith.select %b_a11_867, %b_a11_864, %b_a11_870 : tensor<16xi1>, tensor<16xf32> loc(#loc879) + %b_a11_872 = tt.expand_dims %b_a11_871 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc880) + %b_a11_873 = tt.broadcast %b_a11_872 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc881) + %b_a11_874 = arith.mulf %b_a11_873, %b_Ai11_852 : tensor<16x16xf32> loc(#loc881) + %b_a11_875 = tt.call @triton.language.standard.sum__fp32S16_16S_c0_cFalse_cNone(%b_a11_874) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc882) + %b_a11_876 = arith.addf %b_a11_871, %b_a11_875 : tensor<16xf32> loc(#loc883) + %b_Ai11_877 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc884) + %b_Ai11_878 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc884) + %b_Ai11_879 = arith.cmpi eq, %b_Ai11_877, %b_Ai11_878 : tensor<16xi64> loc(#loc884) + %b_Ai11_880 = tt.expand_dims %b_Ai11_879 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc885) + %b_Ai11_881 = tt.expand_dims %b_a11_876 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc886) + %b_Ai11_882 = tt.broadcast %b_Ai11_881 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc886) + %b_Ai11_883 = tt.broadcast %b_Ai11_880 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc886) + %b_Ai11_884 = arith.select %b_Ai11_883, %b_Ai11_882, %b_Ai11_852 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc886) + scf.yield %b_Ai11_884 : tensor<16x16xf32> loc(#loc343) + } loc(#loc872) + %35 = arith.extsi %T : i32 to i64 loc(#loc344) + %36 = arith.subi %35, %i_tc2_13 : i64 loc(#loc344) + %c16_i32_404 = arith.constant 16 : i32 loc(#loc345) + %37 = arith.extsi %c16_i32_404 : i32 to i64 loc(#loc345) + %38 = arith.minsi %37, %36 : i64 loc(#loc345) + %c2_i32_405 = arith.constant 2 : i32 loc(#loc346) + %c1_i32_406 = arith.constant 1 : i32 loc(#loc346) + %39 = arith.extsi %c2_i32_405 : i32 to i64 loc(#loc346) + %40 = arith.bitcast %38 : i64 to i64 loc(#loc346) + %41 = arith.extsi %c1_i32_406 : i32 to i64 loc(#loc346) + %42 = ub.poison : i64 loc(#loc346) + %b_Ai22_407 = scf.for %i = %39 to %40 step %41 iter_args(%b_Ai22_852 = %b_Ai22_394) -> (tensor<16x16xf32>) : i64 { + %b_a22 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc888) + %b_a22_853 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc888) + %b_a22_854 = arith.cmpi eq, %b_a22, %b_a22_853 : tensor<16xi64> loc(#loc888) + %b_a22_855 = tt.expand_dims %b_a22_854 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc889) + %b_a22_856 = arith.constant 0.000000e+00 : f32 loc(#loc890) + %b_a22_857 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc890) + %b_a22_858 = arith.subf %b_a22_857, %b_A22_367 : tensor<16x16xf32> loc(#loc890) + %b_a22_859 = arith.constant 0.000000e+00 : f32 loc(#loc891) + %b_a22_860 = arith.constant 0.000000e+00 : f32 loc(#loc891) + %b_a22_861 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc891) + %b_a22_862 = tt.broadcast %b_a22_855 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc891) + %b_a22_863 = arith.select %b_a22_862, %b_a22_858, %b_a22_861 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc891) + %b_a22_864 = tt.call @triton.language.standard.sum__fp32S16_16S_c0_cFalse_cNone(%b_a22_863) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc892) + %b_a22_865 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc893) + %b_a22_866 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc893) + %b_a22_867 = arith.cmpi slt, %b_a22_865, %b_a22_866 : tensor<16xi64> loc(#loc893) + %b_a22_868 = arith.constant 0.000000e+00 : f32 loc(#loc894) + %b_a22_869 = arith.constant 0.000000e+00 : f32 loc(#loc894) + %b_a22_870 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc894) + %b_a22_871 = arith.select %b_a22_867, %b_a22_864, %b_a22_870 : tensor<16xi1>, tensor<16xf32> loc(#loc894) + %b_a22_872 = tt.expand_dims %b_a22_871 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc895) + %b_a22_873 = tt.broadcast %b_a22_872 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc896) + %b_a22_874 = arith.mulf %b_a22_873, %b_Ai22_852 : tensor<16x16xf32> loc(#loc896) + %b_a22_875 = tt.call @triton.language.standard.sum__fp32S16_16S_c0_cFalse_cNone(%b_a22_874) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc897) + %b_a22_876 = arith.addf %b_a22_871, %b_a22_875 : tensor<16xf32> loc(#loc898) + %b_Ai22_877 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc899) + %b_Ai22_878 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc899) + %b_Ai22_879 = arith.cmpi eq, %b_Ai22_877, %b_Ai22_878 : tensor<16xi64> loc(#loc899) + %b_Ai22_880 = tt.expand_dims %b_Ai22_879 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc900) + %b_Ai22_881 = tt.expand_dims %b_a22_876 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc901) + %b_Ai22_882 = tt.broadcast %b_Ai22_881 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc901) + %b_Ai22_883 = tt.broadcast %b_Ai22_880 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc901) + %b_Ai22_884 = arith.select %b_Ai22_883, %b_Ai22_882, %b_Ai22_852 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc901) + scf.yield %b_Ai22_884 : tensor<16x16xf32> loc(#loc361) + } loc(#loc887) + %43 = arith.extsi %T : i32 to i64 loc(#loc362) + %44 = arith.subi %43, %i_tc3_18 : i64 loc(#loc362) + %c16_i32_408 = arith.constant 16 : i32 loc(#loc363) + %45 = arith.extsi %c16_i32_408 : i32 to i64 loc(#loc363) + %46 = arith.minsi %45, %44 : i64 loc(#loc363) + %c2_i32_409 = arith.constant 2 : i32 loc(#loc364) + %c1_i32_410 = arith.constant 1 : i32 loc(#loc364) + %47 = arith.extsi %c2_i32_409 : i32 to i64 loc(#loc364) + %48 = arith.bitcast %46 : i64 to i64 loc(#loc364) + %49 = arith.extsi %c1_i32_410 : i32 to i64 loc(#loc364) + %50 = ub.poison : i64 loc(#loc364) + %b_Ai33_411 = scf.for %i = %47 to %48 step %49 iter_args(%b_Ai33_852 = %b_Ai33_396) -> (tensor<16x16xf32>) : i64 { + %b_a33 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc903) + %b_a33_853 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc903) + %b_a33_854 = arith.cmpi eq, %b_a33, %b_a33_853 : tensor<16xi64> loc(#loc903) + %b_a33_855 = tt.expand_dims %b_a33_854 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc904) + %b_a33_856 = arith.constant 0.000000e+00 : f32 loc(#loc905) + %b_a33_857 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc905) + %b_a33_858 = arith.subf %b_a33_857, %b_A33_370 : tensor<16x16xf32> loc(#loc905) + %b_a33_859 = arith.constant 0.000000e+00 : f32 loc(#loc906) + %b_a33_860 = arith.constant 0.000000e+00 : f32 loc(#loc906) + %b_a33_861 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc906) + %b_a33_862 = tt.broadcast %b_a33_855 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc906) + %b_a33_863 = arith.select %b_a33_862, %b_a33_858, %b_a33_861 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc906) + %b_a33_864 = tt.call @triton.language.standard.sum__fp32S16_16S_c0_cFalse_cNone(%b_a33_863) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc907) + %b_a33_865 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc908) + %b_a33_866 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc908) + %b_a33_867 = arith.cmpi slt, %b_a33_865, %b_a33_866 : tensor<16xi64> loc(#loc908) + %b_a33_868 = arith.constant 0.000000e+00 : f32 loc(#loc909) + %b_a33_869 = arith.constant 0.000000e+00 : f32 loc(#loc909) + %b_a33_870 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc909) + %b_a33_871 = arith.select %b_a33_867, %b_a33_864, %b_a33_870 : tensor<16xi1>, tensor<16xf32> loc(#loc909) + %b_a33_872 = tt.expand_dims %b_a33_871 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc910) + %b_a33_873 = tt.broadcast %b_a33_872 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc911) + %b_a33_874 = arith.mulf %b_a33_873, %b_Ai33_852 : tensor<16x16xf32> loc(#loc911) + %b_a33_875 = tt.call @triton.language.standard.sum__fp32S16_16S_c0_cFalse_cNone(%b_a33_874) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc912) + %b_a33_876 = arith.addf %b_a33_871, %b_a33_875 : tensor<16xf32> loc(#loc913) + %b_Ai33_877 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc914) + %b_Ai33_878 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc914) + %b_Ai33_879 = arith.cmpi eq, %b_Ai33_877, %b_Ai33_878 : tensor<16xi64> loc(#loc914) + %b_Ai33_880 = tt.expand_dims %b_Ai33_879 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc915) + %b_Ai33_881 = tt.expand_dims %b_a33_876 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc916) + %b_Ai33_882 = tt.broadcast %b_Ai33_881 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc916) + %b_Ai33_883 = tt.broadcast %b_Ai33_880 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc916) + %b_Ai33_884 = arith.select %b_Ai33_883, %b_Ai33_882, %b_Ai33_852 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc916) + scf.yield %b_Ai33_884 : tensor<16x16xf32> loc(#loc379) + } loc(#loc902) + %b_Ai00_412 = arith.uitofp %m_I_194 : tensor<16x16xi1> to tensor<16x16xf32> loc(#loc917) + %b_Ai00_413 = arith.addf %b_Ai00_399, %b_Ai00_412 : tensor<16x16xf32> loc(#loc917) + %b_Ai11_414 = arith.uitofp %m_I_194 : tensor<16x16xi1> to tensor<16x16xf32> loc(#loc918) + %b_Ai11_415 = arith.addf %b_Ai11_403, %b_Ai11_414 : tensor<16x16xf32> loc(#loc918) + %b_Ai22_416 = arith.uitofp %m_I_194 : tensor<16x16xi1> to tensor<16x16xf32> loc(#loc919) + %b_Ai22_417 = arith.addf %b_Ai22_407, %b_Ai22_416 : tensor<16x16xf32> loc(#loc919) + %b_Ai33_418 = arith.uitofp %m_I_194 : tensor<16x16xi1> to tensor<16x16xf32> loc(#loc920) + %b_Ai33_419 = arith.addf %b_Ai33_411, %b_Ai33_418 : tensor<16x16xf32> loc(#loc920) + %b_Ai10 = arith.constant 0.000000e+00 : f32 loc(#loc921) + %b_Ai10_420 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc921) + %b_Ai10_421 = tt.dot %b_Ai11_415, %b_A10_373, %b_Ai10_420, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc921) + %b_Ai10_422 = arith.constant 0.000000e+00 : f32 loc(#loc922) + %b_Ai10_423 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc922) + %b_Ai10_424 = tt.dot %b_Ai10_421, %b_Ai00_413, %b_Ai10_423, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc922) + %b_Ai10_425 = arith.constant 0.000000e+00 : f32 loc(#loc923) + %b_Ai10_426 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc923) + %b_Ai10_427 = arith.subf %b_Ai10_426, %b_Ai10_424 : tensor<16x16xf32> loc(#loc923) + %b_Ai21 = arith.constant 0.000000e+00 : f32 loc(#loc924) + %b_Ai21_428 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc924) + %b_Ai21_429 = tt.dot %b_Ai22_417, %b_A21_379, %b_Ai21_428, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc924) + %b_Ai21_430 = arith.constant 0.000000e+00 : f32 loc(#loc925) + %b_Ai21_431 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc925) + %b_Ai21_432 = tt.dot %b_Ai21_429, %b_Ai11_415, %b_Ai21_431, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc925) + %b_Ai21_433 = arith.constant 0.000000e+00 : f32 loc(#loc926) + %b_Ai21_434 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc926) + %b_Ai21_435 = arith.subf %b_Ai21_434, %b_Ai21_432 : tensor<16x16xf32> loc(#loc926) + %b_Ai32 = arith.constant 0.000000e+00 : f32 loc(#loc927) + %b_Ai32_436 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc927) + %b_Ai32_437 = tt.dot %b_Ai33_419, %b_A32_388, %b_Ai32_436, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc927) + %b_Ai32_438 = arith.constant 0.000000e+00 : f32 loc(#loc928) + %b_Ai32_439 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc928) + %b_Ai32_440 = tt.dot %b_Ai32_437, %b_Ai22_417, %b_Ai32_439, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc928) + %b_Ai32_441 = arith.constant 0.000000e+00 : f32 loc(#loc929) + %b_Ai32_442 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc929) + %b_Ai32_443 = arith.subf %b_Ai32_442, %b_Ai32_440 : tensor<16x16xf32> loc(#loc929) + %b_Ai20 = arith.constant 0.000000e+00 : f32 loc(#loc930) + %b_Ai20_444 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc930) + %b_Ai20_445 = tt.dot %b_A20_376, %b_Ai00_413, %b_Ai20_444, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc930) + %b_Ai20_446 = arith.constant 0.000000e+00 : f32 loc(#loc931) + %b_Ai20_447 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc931) + %b_Ai20_448 = tt.dot %b_A21_379, %b_Ai10_427, %b_Ai20_447, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc931) + %b_Ai20_449 = arith.addf %b_Ai20_445, %b_Ai20_448 : tensor<16x16xf32> loc(#loc932) + %b_Ai20_450 = arith.constant 0.000000e+00 : f32 loc(#loc933) + %b_Ai20_451 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc933) + %b_Ai20_452 = tt.dot %b_Ai22_417, %b_Ai20_449, %b_Ai20_451, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc933) + %b_Ai20_453 = arith.constant 0.000000e+00 : f32 loc(#loc934) + %b_Ai20_454 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc934) + %b_Ai20_455 = arith.subf %b_Ai20_454, %b_Ai20_452 : tensor<16x16xf32> loc(#loc934) + %b_Ai31 = arith.constant 0.000000e+00 : f32 loc(#loc935) + %b_Ai31_456 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc935) + %b_Ai31_457 = tt.dot %b_A31_385, %b_Ai11_415, %b_Ai31_456, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc935) + %b_Ai31_458 = arith.constant 0.000000e+00 : f32 loc(#loc936) + %b_Ai31_459 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc936) + %b_Ai31_460 = tt.dot %b_A32_388, %b_Ai21_435, %b_Ai31_459, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc936) + %b_Ai31_461 = arith.addf %b_Ai31_457, %b_Ai31_460 : tensor<16x16xf32> loc(#loc937) + %b_Ai31_462 = arith.constant 0.000000e+00 : f32 loc(#loc938) + %b_Ai31_463 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc938) + %b_Ai31_464 = tt.dot %b_Ai33_419, %b_Ai31_461, %b_Ai31_463, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc938) + %b_Ai31_465 = arith.constant 0.000000e+00 : f32 loc(#loc939) + %b_Ai31_466 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc939) + %b_Ai31_467 = arith.subf %b_Ai31_466, %b_Ai31_464 : tensor<16x16xf32> loc(#loc939) + %b_Ai30 = arith.constant 0.000000e+00 : f32 loc(#loc940) + %b_Ai30_468 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc940) + %b_Ai30_469 = tt.dot %b_A30_382, %b_Ai00_413, %b_Ai30_468, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc940) + %b_Ai30_470 = arith.constant 0.000000e+00 : f32 loc(#loc941) + %b_Ai30_471 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc941) + %b_Ai30_472 = tt.dot %b_A31_385, %b_Ai10_427, %b_Ai30_471, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc941) + %b_Ai30_473 = arith.addf %b_Ai30_469, %b_Ai30_472 : tensor<16x16xf32> loc(#loc942) + %b_Ai30_474 = arith.constant 0.000000e+00 : f32 loc(#loc943) + %b_Ai30_475 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc943) + %b_Ai30_476 = tt.dot %b_A32_388, %b_Ai20_455, %b_Ai30_475, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc943) + %b_Ai30_477 = arith.addf %b_Ai30_473, %b_Ai30_476 : tensor<16x16xf32> loc(#loc944) + %b_Ai30_478 = arith.constant 0.000000e+00 : f32 loc(#loc945) + %b_Ai30_479 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc945) + %b_Ai30_480 = tt.dot %b_Ai33_419, %b_Ai30_477, %b_Ai30_479, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc945) + %b_Ai30_481 = arith.constant 0.000000e+00 : f32 loc(#loc946) + %b_Ai30_482 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc946) + %b_Ai30_483 = arith.subf %b_Ai30_482, %b_Ai30_480 : tensor<16x16xf32> loc(#loc946) + %p_A00 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc947) + %p_A00_484 = tt.splat %i_tc0_3 : i64 -> tensor<16xi64> loc(#loc947) + %p_A00_485 = arith.addi %p_A00_484, %p_A00 : tensor<16xi64> loc(#loc947) + %p_A00_486 = tt.expand_dims %p_A00_485 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc948) + %p_A00_487 = arith.constant 2048 : i32 loc(#loc949) + %p_A00_488 = arith.constant 2048 : i64 loc(#loc949) + %p_A00_489 = arith.constant dense<2048> : tensor<16x1xi64> loc(#loc949) + %p_A00_490 = arith.muli %p_A00_486, %p_A00_489 : tensor<16x1xi64> loc(#loc949) + %p_A00_491 = tt.splat %A_37 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc950) + %p_A00_492 = tt.addptr %p_A00_491, %p_A00_490 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc950) + %p_A00_493 = tt.expand_dims %o_i {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc951) + %p_A00_494 = tt.broadcast %p_A00_492 : tensor<16x1x!tt.ptr> -> tensor<16x16x!tt.ptr> loc(#loc952) + %p_A00_495 = tt.broadcast %p_A00_493 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc952) + %p_A00_496 = tt.addptr %p_A00_494, %p_A00_495 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc952) + %p_A10 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc953) + %p_A10_497 = tt.splat %i_tc1_8 : i64 -> tensor<16xi64> loc(#loc953) + %p_A10_498 = arith.addi %p_A10_497, %p_A10 : tensor<16xi64> loc(#loc953) + %p_A10_499 = tt.expand_dims %p_A10_498 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc954) + %p_A10_500 = arith.constant 2048 : i32 loc(#loc955) + %p_A10_501 = arith.constant 2048 : i64 loc(#loc955) + %p_A10_502 = arith.constant dense<2048> : tensor<16x1xi64> loc(#loc955) + %p_A10_503 = arith.muli %p_A10_499, %p_A10_502 : tensor<16x1xi64> loc(#loc955) + %p_A10_504 = tt.splat %A_37 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc956) + %p_A10_505 = tt.addptr %p_A10_504, %p_A10_503 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc956) + %p_A10_506 = tt.expand_dims %o_i {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc957) + %p_A10_507 = tt.broadcast %p_A10_505 : tensor<16x1x!tt.ptr> -> tensor<16x16x!tt.ptr> loc(#loc958) + %p_A10_508 = tt.broadcast %p_A10_506 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc958) + %p_A10_509 = tt.addptr %p_A10_507, %p_A10_508 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc958) + %p_A11 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc959) + %p_A11_510 = tt.splat %i_tc1_8 : i64 -> tensor<16xi64> loc(#loc959) + %p_A11_511 = arith.addi %p_A11_510, %p_A11 : tensor<16xi64> loc(#loc959) + %p_A11_512 = tt.expand_dims %p_A11_511 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc960) + %p_A11_513 = arith.constant 2048 : i32 loc(#loc961) + %p_A11_514 = arith.constant 2048 : i64 loc(#loc961) + %p_A11_515 = arith.constant dense<2048> : tensor<16x1xi64> loc(#loc961) + %p_A11_516 = arith.muli %p_A11_512, %p_A11_515 : tensor<16x1xi64> loc(#loc961) + %p_A11_517 = tt.splat %A_37 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc962) + %p_A11_518 = tt.addptr %p_A11_517, %p_A11_516 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc962) + %p_A11_519 = arith.constant 16 : i32 loc(#loc963) + %p_A11_520 = arith.constant 16 : i32 loc(#loc963) + %p_A11_521 = arith.constant dense<16> : tensor<16xi32> loc(#loc963) + %p_A11_522 = arith.extsi %p_A11_521 : tensor<16xi32> to tensor<16xi64> loc(#loc963) + %p_A11_523 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc963) + %p_A11_524 = arith.addi %p_A11_522, %p_A11_523 : tensor<16xi64> loc(#loc963) + %p_A11_525 = arith.constant 2147483647 : i64 loc(#loc963) + %p_A11_526 = arith.constant -2147483648 : i64 loc(#loc963) + %p_A11_527 = arith.constant dense<2147483647> : tensor<16xi64> loc(#loc963) + %p_A11_528 = arith.cmpi sle, %p_A11_524, %p_A11_527 : tensor<16xi64> loc(#loc963) + %p_A11_529 = arith.constant dense<-2147483648> : tensor<16xi64> loc(#loc963) + %p_A11_530 = arith.cmpi sge, %p_A11_524, %p_A11_529 : tensor<16xi64> loc(#loc963) + %p_A11_531 = arith.andi %p_A11_528, %p_A11_530 : tensor<16xi1> loc(#loc963) + %p_A11_532 = arith.addi %p_A11_521, %o_i : tensor<16xi32> loc(#loc963) + %p_A11_533 = tt.expand_dims %p_A11_532 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc964) + %p_A11_534 = tt.broadcast %p_A11_518 : tensor<16x1x!tt.ptr> -> tensor<16x16x!tt.ptr> loc(#loc965) + %p_A11_535 = tt.broadcast %p_A11_533 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc965) + %p_A11_536 = tt.addptr %p_A11_534, %p_A11_535 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc965) + %p_A20 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc966) + %p_A20_537 = tt.splat %i_tc2_13 : i64 -> tensor<16xi64> loc(#loc966) + %p_A20_538 = arith.addi %p_A20_537, %p_A20 : tensor<16xi64> loc(#loc966) + %p_A20_539 = tt.expand_dims %p_A20_538 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc967) + %p_A20_540 = arith.constant 2048 : i32 loc(#loc968) + %p_A20_541 = arith.constant 2048 : i64 loc(#loc968) + %p_A20_542 = arith.constant dense<2048> : tensor<16x1xi64> loc(#loc968) + %p_A20_543 = arith.muli %p_A20_539, %p_A20_542 : tensor<16x1xi64> loc(#loc968) + %p_A20_544 = tt.splat %A_37 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc969) + %p_A20_545 = tt.addptr %p_A20_544, %p_A20_543 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc969) + %p_A20_546 = tt.expand_dims %o_i {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc970) + %p_A20_547 = tt.broadcast %p_A20_545 : tensor<16x1x!tt.ptr> -> tensor<16x16x!tt.ptr> loc(#loc971) + %p_A20_548 = tt.broadcast %p_A20_546 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc971) + %p_A20_549 = tt.addptr %p_A20_547, %p_A20_548 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc971) + %p_A21 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc972) + %p_A21_550 = tt.splat %i_tc2_13 : i64 -> tensor<16xi64> loc(#loc972) + %p_A21_551 = arith.addi %p_A21_550, %p_A21 : tensor<16xi64> loc(#loc972) + %p_A21_552 = tt.expand_dims %p_A21_551 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc973) + %p_A21_553 = arith.constant 2048 : i32 loc(#loc974) + %p_A21_554 = arith.constant 2048 : i64 loc(#loc974) + %p_A21_555 = arith.constant dense<2048> : tensor<16x1xi64> loc(#loc974) + %p_A21_556 = arith.muli %p_A21_552, %p_A21_555 : tensor<16x1xi64> loc(#loc974) + %p_A21_557 = tt.splat %A_37 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc975) + %p_A21_558 = tt.addptr %p_A21_557, %p_A21_556 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc975) + %p_A21_559 = arith.constant 16 : i32 loc(#loc976) + %p_A21_560 = arith.constant 16 : i32 loc(#loc976) + %p_A21_561 = arith.constant dense<16> : tensor<16xi32> loc(#loc976) + %p_A21_562 = arith.extsi %p_A21_561 : tensor<16xi32> to tensor<16xi64> loc(#loc976) + %p_A21_563 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc976) + %p_A21_564 = arith.addi %p_A21_562, %p_A21_563 : tensor<16xi64> loc(#loc976) + %p_A21_565 = arith.constant 2147483647 : i64 loc(#loc976) + %p_A21_566 = arith.constant -2147483648 : i64 loc(#loc976) + %p_A21_567 = arith.constant dense<2147483647> : tensor<16xi64> loc(#loc976) + %p_A21_568 = arith.cmpi sle, %p_A21_564, %p_A21_567 : tensor<16xi64> loc(#loc976) + %p_A21_569 = arith.constant dense<-2147483648> : tensor<16xi64> loc(#loc976) + %p_A21_570 = arith.cmpi sge, %p_A21_564, %p_A21_569 : tensor<16xi64> loc(#loc976) + %p_A21_571 = arith.andi %p_A21_568, %p_A21_570 : tensor<16xi1> loc(#loc976) + %p_A21_572 = arith.addi %p_A21_561, %o_i : tensor<16xi32> loc(#loc976) + %p_A21_573 = tt.expand_dims %p_A21_572 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc977) + %p_A21_574 = tt.broadcast %p_A21_558 : tensor<16x1x!tt.ptr> -> tensor<16x16x!tt.ptr> loc(#loc978) + %p_A21_575 = tt.broadcast %p_A21_573 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc978) + %p_A21_576 = tt.addptr %p_A21_574, %p_A21_575 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc978) + %p_A22 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc979) + %p_A22_577 = tt.splat %i_tc2_13 : i64 -> tensor<16xi64> loc(#loc979) + %p_A22_578 = arith.addi %p_A22_577, %p_A22 : tensor<16xi64> loc(#loc979) + %p_A22_579 = tt.expand_dims %p_A22_578 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc980) + %p_A22_580 = arith.constant 2048 : i32 loc(#loc981) + %p_A22_581 = arith.constant 2048 : i64 loc(#loc981) + %p_A22_582 = arith.constant dense<2048> : tensor<16x1xi64> loc(#loc981) + %p_A22_583 = arith.muli %p_A22_579, %p_A22_582 : tensor<16x1xi64> loc(#loc981) + %p_A22_584 = tt.splat %A_37 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc982) + %p_A22_585 = tt.addptr %p_A22_584, %p_A22_583 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc982) + %p_A22_586 = arith.constant 32 : i32 loc(#loc983) + %p_A22_587 = arith.constant 32 : i32 loc(#loc983) + %p_A22_588 = arith.constant dense<32> : tensor<16xi32> loc(#loc983) + %p_A22_589 = arith.extsi %p_A22_588 : tensor<16xi32> to tensor<16xi64> loc(#loc983) + %p_A22_590 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc983) + %p_A22_591 = arith.addi %p_A22_589, %p_A22_590 : tensor<16xi64> loc(#loc983) + %p_A22_592 = arith.constant 2147483647 : i64 loc(#loc983) + %p_A22_593 = arith.constant -2147483648 : i64 loc(#loc983) + %p_A22_594 = arith.constant dense<2147483647> : tensor<16xi64> loc(#loc983) + %p_A22_595 = arith.cmpi sle, %p_A22_591, %p_A22_594 : tensor<16xi64> loc(#loc983) + %p_A22_596 = arith.constant dense<-2147483648> : tensor<16xi64> loc(#loc983) + %p_A22_597 = arith.cmpi sge, %p_A22_591, %p_A22_596 : tensor<16xi64> loc(#loc983) + %p_A22_598 = arith.andi %p_A22_595, %p_A22_597 : tensor<16xi1> loc(#loc983) + %p_A22_599 = arith.addi %p_A22_588, %o_i : tensor<16xi32> loc(#loc983) + %p_A22_600 = tt.expand_dims %p_A22_599 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc984) + %p_A22_601 = tt.broadcast %p_A22_585 : tensor<16x1x!tt.ptr> -> tensor<16x16x!tt.ptr> loc(#loc985) + %p_A22_602 = tt.broadcast %p_A22_600 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc985) + %p_A22_603 = tt.addptr %p_A22_601, %p_A22_602 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc985) + %p_A30 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc986) + %p_A30_604 = tt.splat %i_tc3_18 : i64 -> tensor<16xi64> loc(#loc986) + %p_A30_605 = arith.addi %p_A30_604, %p_A30 : tensor<16xi64> loc(#loc986) + %p_A30_606 = tt.expand_dims %p_A30_605 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc987) + %p_A30_607 = arith.constant 2048 : i32 loc(#loc988) + %p_A30_608 = arith.constant 2048 : i64 loc(#loc988) + %p_A30_609 = arith.constant dense<2048> : tensor<16x1xi64> loc(#loc988) + %p_A30_610 = arith.muli %p_A30_606, %p_A30_609 : tensor<16x1xi64> loc(#loc988) + %p_A30_611 = tt.splat %A_37 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc989) + %p_A30_612 = tt.addptr %p_A30_611, %p_A30_610 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc989) + %p_A30_613 = tt.expand_dims %o_i {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc990) + %p_A30_614 = tt.broadcast %p_A30_612 : tensor<16x1x!tt.ptr> -> tensor<16x16x!tt.ptr> loc(#loc991) + %p_A30_615 = tt.broadcast %p_A30_613 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc991) + %p_A30_616 = tt.addptr %p_A30_614, %p_A30_615 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc991) + %p_A31 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc992) + %p_A31_617 = tt.splat %i_tc3_18 : i64 -> tensor<16xi64> loc(#loc992) + %p_A31_618 = arith.addi %p_A31_617, %p_A31 : tensor<16xi64> loc(#loc992) + %p_A31_619 = tt.expand_dims %p_A31_618 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc993) + %p_A31_620 = arith.constant 2048 : i32 loc(#loc994) + %p_A31_621 = arith.constant 2048 : i64 loc(#loc994) + %p_A31_622 = arith.constant dense<2048> : tensor<16x1xi64> loc(#loc994) + %p_A31_623 = arith.muli %p_A31_619, %p_A31_622 : tensor<16x1xi64> loc(#loc994) + %p_A31_624 = tt.splat %A_37 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc995) + %p_A31_625 = tt.addptr %p_A31_624, %p_A31_623 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc995) + %p_A31_626 = arith.constant 16 : i32 loc(#loc996) + %p_A31_627 = arith.constant 16 : i32 loc(#loc996) + %p_A31_628 = arith.constant dense<16> : tensor<16xi32> loc(#loc996) + %p_A31_629 = arith.extsi %p_A31_628 : tensor<16xi32> to tensor<16xi64> loc(#loc996) + %p_A31_630 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc996) + %p_A31_631 = arith.addi %p_A31_629, %p_A31_630 : tensor<16xi64> loc(#loc996) + %p_A31_632 = arith.constant 2147483647 : i64 loc(#loc996) + %p_A31_633 = arith.constant -2147483648 : i64 loc(#loc996) + %p_A31_634 = arith.constant dense<2147483647> : tensor<16xi64> loc(#loc996) + %p_A31_635 = arith.cmpi sle, %p_A31_631, %p_A31_634 : tensor<16xi64> loc(#loc996) + %p_A31_636 = arith.constant dense<-2147483648> : tensor<16xi64> loc(#loc996) + %p_A31_637 = arith.cmpi sge, %p_A31_631, %p_A31_636 : tensor<16xi64> loc(#loc996) + %p_A31_638 = arith.andi %p_A31_635, %p_A31_637 : tensor<16xi1> loc(#loc996) + %p_A31_639 = arith.addi %p_A31_628, %o_i : tensor<16xi32> loc(#loc996) + %p_A31_640 = tt.expand_dims %p_A31_639 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc997) + %p_A31_641 = tt.broadcast %p_A31_625 : tensor<16x1x!tt.ptr> -> tensor<16x16x!tt.ptr> loc(#loc998) + %p_A31_642 = tt.broadcast %p_A31_640 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc998) + %p_A31_643 = tt.addptr %p_A31_641, %p_A31_642 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc998) + %p_A32 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc999) + %p_A32_644 = tt.splat %i_tc3_18 : i64 -> tensor<16xi64> loc(#loc999) + %p_A32_645 = arith.addi %p_A32_644, %p_A32 : tensor<16xi64> loc(#loc999) + %p_A32_646 = tt.expand_dims %p_A32_645 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc1000) + %p_A32_647 = arith.constant 2048 : i32 loc(#loc1001) + %p_A32_648 = arith.constant 2048 : i64 loc(#loc1001) + %p_A32_649 = arith.constant dense<2048> : tensor<16x1xi64> loc(#loc1001) + %p_A32_650 = arith.muli %p_A32_646, %p_A32_649 : tensor<16x1xi64> loc(#loc1001) + %p_A32_651 = tt.splat %A_37 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc1002) + %p_A32_652 = tt.addptr %p_A32_651, %p_A32_650 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc1002) + %p_A32_653 = arith.constant 32 : i32 loc(#loc1003) + %p_A32_654 = arith.constant 32 : i32 loc(#loc1003) + %p_A32_655 = arith.constant dense<32> : tensor<16xi32> loc(#loc1003) + %p_A32_656 = arith.extsi %p_A32_655 : tensor<16xi32> to tensor<16xi64> loc(#loc1003) + %p_A32_657 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc1003) + %p_A32_658 = arith.addi %p_A32_656, %p_A32_657 : tensor<16xi64> loc(#loc1003) + %p_A32_659 = arith.constant 2147483647 : i64 loc(#loc1003) + %p_A32_660 = arith.constant -2147483648 : i64 loc(#loc1003) + %p_A32_661 = arith.constant dense<2147483647> : tensor<16xi64> loc(#loc1003) + %p_A32_662 = arith.cmpi sle, %p_A32_658, %p_A32_661 : tensor<16xi64> loc(#loc1003) + %p_A32_663 = arith.constant dense<-2147483648> : tensor<16xi64> loc(#loc1003) + %p_A32_664 = arith.cmpi sge, %p_A32_658, %p_A32_663 : tensor<16xi64> loc(#loc1003) + %p_A32_665 = arith.andi %p_A32_662, %p_A32_664 : tensor<16xi1> loc(#loc1003) + %p_A32_666 = arith.addi %p_A32_655, %o_i : tensor<16xi32> loc(#loc1003) + %p_A32_667 = tt.expand_dims %p_A32_666 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc1004) + %p_A32_668 = tt.broadcast %p_A32_652 : tensor<16x1x!tt.ptr> -> tensor<16x16x!tt.ptr> loc(#loc1005) + %p_A32_669 = tt.broadcast %p_A32_667 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc1005) + %p_A32_670 = tt.addptr %p_A32_668, %p_A32_669 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc1005) + %p_A33 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc1006) + %p_A33_671 = tt.splat %i_tc3_18 : i64 -> tensor<16xi64> loc(#loc1006) + %p_A33_672 = arith.addi %p_A33_671, %p_A33 : tensor<16xi64> loc(#loc1006) + %p_A33_673 = tt.expand_dims %p_A33_672 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc1007) + %p_A33_674 = arith.constant 2048 : i32 loc(#loc1008) + %p_A33_675 = arith.constant 2048 : i64 loc(#loc1008) + %p_A33_676 = arith.constant dense<2048> : tensor<16x1xi64> loc(#loc1008) + %p_A33_677 = arith.muli %p_A33_673, %p_A33_676 : tensor<16x1xi64> loc(#loc1008) + %p_A33_678 = tt.splat %A_37 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc1009) + %p_A33_679 = tt.addptr %p_A33_678, %p_A33_677 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc1009) + %p_A33_680 = arith.constant 48 : i32 loc(#loc1010) + %p_A33_681 = arith.constant 48 : i32 loc(#loc1010) + %p_A33_682 = arith.constant dense<48> : tensor<16xi32> loc(#loc1010) + %p_A33_683 = arith.extsi %p_A33_682 : tensor<16xi32> to tensor<16xi64> loc(#loc1010) + %p_A33_684 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc1010) + %p_A33_685 = arith.addi %p_A33_683, %p_A33_684 : tensor<16xi64> loc(#loc1010) + %p_A33_686 = arith.constant 2147483647 : i64 loc(#loc1010) + %p_A33_687 = arith.constant -2147483648 : i64 loc(#loc1010) + %p_A33_688 = arith.constant dense<2147483647> : tensor<16xi64> loc(#loc1010) + %p_A33_689 = arith.cmpi sle, %p_A33_685, %p_A33_688 : tensor<16xi64> loc(#loc1010) + %p_A33_690 = arith.constant dense<-2147483648> : tensor<16xi64> loc(#loc1010) + %p_A33_691 = arith.cmpi sge, %p_A33_685, %p_A33_690 : tensor<16xi64> loc(#loc1010) + %p_A33_692 = arith.andi %p_A33_689, %p_A33_691 : tensor<16xi1> loc(#loc1010) + %p_A33_693 = arith.addi %p_A33_682, %o_i : tensor<16xi32> loc(#loc1010) + %p_A33_694 = tt.expand_dims %p_A33_693 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc1011) + %p_A33_695 = tt.broadcast %p_A33_679 : tensor<16x1x!tt.ptr> -> tensor<16x16x!tt.ptr> loc(#loc1012) + %p_A33_696 = tt.broadcast %p_A33_694 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc1012) + %p_A33_697 = tt.addptr %p_A33_695, %p_A33_696 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc1012) + %m_A0 = tt.expand_dims %m_tc0_42 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc1013) + %m_A0_698 = tt.expand_dims %o_i {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc1014) + %m_A0_699 = arith.constant 64 : i32 loc(#loc1015) + %m_A0_700 = arith.constant dense<64> : tensor<1x16xi32> loc(#loc1015) + %m_A0_701 = arith.cmpi slt, %m_A0_698, %m_A0_700 : tensor<1x16xi32> loc(#loc1015) + %m_A0_702 = tt.broadcast %m_A0 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc1016) + %m_A0_703 = tt.broadcast %m_A0_701 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc1016) + %m_A0_704 = arith.andi %m_A0_702, %m_A0_703 : tensor<16x16xi1> loc(#loc1016) + %m_A1 = tt.expand_dims %m_tc1_47 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc1017) + %m_A1_705 = tt.expand_dims %o_i {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc1018) + %m_A1_706 = arith.constant 64 : i32 loc(#loc1019) + %m_A1_707 = arith.constant dense<64> : tensor<1x16xi32> loc(#loc1019) + %m_A1_708 = arith.cmpi slt, %m_A1_705, %m_A1_707 : tensor<1x16xi32> loc(#loc1019) + %m_A1_709 = tt.broadcast %m_A1 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc1020) + %m_A1_710 = tt.broadcast %m_A1_708 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc1020) + %m_A1_711 = arith.andi %m_A1_709, %m_A1_710 : tensor<16x16xi1> loc(#loc1020) + %m_A2 = tt.expand_dims %m_tc2_52 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc1021) + %m_A2_712 = tt.expand_dims %o_i {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc1022) + %m_A2_713 = arith.constant 64 : i32 loc(#loc1023) + %m_A2_714 = arith.constant dense<64> : tensor<1x16xi32> loc(#loc1023) + %m_A2_715 = arith.cmpi slt, %m_A2_712, %m_A2_714 : tensor<1x16xi32> loc(#loc1023) + %m_A2_716 = tt.broadcast %m_A2 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc1024) + %m_A2_717 = tt.broadcast %m_A2_715 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc1024) + %m_A2_718 = arith.andi %m_A2_716, %m_A2_717 : tensor<16x16xi1> loc(#loc1024) + %m_A3 = tt.expand_dims %m_tc3_57 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc1025) + %m_A3_719 = tt.expand_dims %o_i {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc1026) + %m_A3_720 = arith.constant 64 : i32 loc(#loc1027) + %m_A3_721 = arith.constant dense<64> : tensor<1x16xi32> loc(#loc1027) + %m_A3_722 = arith.cmpi slt, %m_A3_719, %m_A3_721 : tensor<1x16xi32> loc(#loc1027) + %m_A3_723 = tt.broadcast %m_A3 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc1028) + %m_A3_724 = tt.broadcast %m_A3_722 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc1028) + %m_A3_725 = arith.andi %m_A3_723, %m_A3_724 : tensor<16x16xi1> loc(#loc1028) + %m_A11 = tt.expand_dims %m_tc1_47 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc1029) + %m_A11_726 = arith.constant 16 : i32 loc(#loc1030) + %m_A11_727 = arith.constant 16 : i32 loc(#loc1030) + %m_A11_728 = arith.constant dense<16> : tensor<16xi32> loc(#loc1030) + %m_A11_729 = arith.extsi %m_A11_728 : tensor<16xi32> to tensor<16xi64> loc(#loc1030) + %m_A11_730 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc1030) + %m_A11_731 = arith.addi %m_A11_729, %m_A11_730 : tensor<16xi64> loc(#loc1030) + %m_A11_732 = arith.constant 2147483647 : i64 loc(#loc1030) + %m_A11_733 = arith.constant -2147483648 : i64 loc(#loc1030) + %m_A11_734 = arith.constant dense<2147483647> : tensor<16xi64> loc(#loc1030) + %m_A11_735 = arith.cmpi sle, %m_A11_731, %m_A11_734 : tensor<16xi64> loc(#loc1030) + %m_A11_736 = arith.constant dense<-2147483648> : tensor<16xi64> loc(#loc1030) + %m_A11_737 = arith.cmpi sge, %m_A11_731, %m_A11_736 : tensor<16xi64> loc(#loc1030) + %m_A11_738 = arith.andi %m_A11_735, %m_A11_737 : tensor<16xi1> loc(#loc1030) + %m_A11_739 = arith.addi %m_A11_728, %o_i : tensor<16xi32> loc(#loc1030) + %m_A11_740 = tt.expand_dims %m_A11_739 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc1031) + %m_A11_741 = arith.constant 64 : i32 loc(#loc1032) + %m_A11_742 = arith.constant dense<64> : tensor<1x16xi32> loc(#loc1032) + %m_A11_743 = arith.cmpi slt, %m_A11_740, %m_A11_742 : tensor<1x16xi32> loc(#loc1032) + %m_A11_744 = tt.broadcast %m_A11 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc1033) + %m_A11_745 = tt.broadcast %m_A11_743 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc1033) + %m_A11_746 = arith.andi %m_A11_744, %m_A11_745 : tensor<16x16xi1> loc(#loc1033) + %m_A21 = tt.expand_dims %m_tc2_52 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc1034) + %m_A21_747 = arith.constant 16 : i32 loc(#loc1035) + %m_A21_748 = arith.constant 16 : i32 loc(#loc1035) + %m_A21_749 = arith.constant dense<16> : tensor<16xi32> loc(#loc1035) + %m_A21_750 = arith.extsi %m_A21_749 : tensor<16xi32> to tensor<16xi64> loc(#loc1035) + %m_A21_751 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc1035) + %m_A21_752 = arith.addi %m_A21_750, %m_A21_751 : tensor<16xi64> loc(#loc1035) + %m_A21_753 = arith.constant 2147483647 : i64 loc(#loc1035) + %m_A21_754 = arith.constant -2147483648 : i64 loc(#loc1035) + %m_A21_755 = arith.constant dense<2147483647> : tensor<16xi64> loc(#loc1035) + %m_A21_756 = arith.cmpi sle, %m_A21_752, %m_A21_755 : tensor<16xi64> loc(#loc1035) + %m_A21_757 = arith.constant dense<-2147483648> : tensor<16xi64> loc(#loc1035) + %m_A21_758 = arith.cmpi sge, %m_A21_752, %m_A21_757 : tensor<16xi64> loc(#loc1035) + %m_A21_759 = arith.andi %m_A21_756, %m_A21_758 : tensor<16xi1> loc(#loc1035) + %m_A21_760 = arith.addi %m_A21_749, %o_i : tensor<16xi32> loc(#loc1035) + %m_A21_761 = tt.expand_dims %m_A21_760 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc1036) + %m_A21_762 = arith.constant 64 : i32 loc(#loc1037) + %m_A21_763 = arith.constant dense<64> : tensor<1x16xi32> loc(#loc1037) + %m_A21_764 = arith.cmpi slt, %m_A21_761, %m_A21_763 : tensor<1x16xi32> loc(#loc1037) + %m_A21_765 = tt.broadcast %m_A21 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc1038) + %m_A21_766 = tt.broadcast %m_A21_764 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc1038) + %m_A21_767 = arith.andi %m_A21_765, %m_A21_766 : tensor<16x16xi1> loc(#loc1038) + %m_A22 = tt.expand_dims %m_tc2_52 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc1039) + %m_A22_768 = arith.constant 32 : i32 loc(#loc1040) + %m_A22_769 = arith.constant 32 : i32 loc(#loc1040) + %m_A22_770 = arith.constant dense<32> : tensor<16xi32> loc(#loc1040) + %m_A22_771 = arith.extsi %m_A22_770 : tensor<16xi32> to tensor<16xi64> loc(#loc1040) + %m_A22_772 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc1040) + %m_A22_773 = arith.addi %m_A22_771, %m_A22_772 : tensor<16xi64> loc(#loc1040) + %m_A22_774 = arith.constant 2147483647 : i64 loc(#loc1040) + %m_A22_775 = arith.constant -2147483648 : i64 loc(#loc1040) + %m_A22_776 = arith.constant dense<2147483647> : tensor<16xi64> loc(#loc1040) + %m_A22_777 = arith.cmpi sle, %m_A22_773, %m_A22_776 : tensor<16xi64> loc(#loc1040) + %m_A22_778 = arith.constant dense<-2147483648> : tensor<16xi64> loc(#loc1040) + %m_A22_779 = arith.cmpi sge, %m_A22_773, %m_A22_778 : tensor<16xi64> loc(#loc1040) + %m_A22_780 = arith.andi %m_A22_777, %m_A22_779 : tensor<16xi1> loc(#loc1040) + %m_A22_781 = arith.addi %m_A22_770, %o_i : tensor<16xi32> loc(#loc1040) + %m_A22_782 = tt.expand_dims %m_A22_781 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc1041) + %m_A22_783 = arith.constant 64 : i32 loc(#loc1042) + %m_A22_784 = arith.constant dense<64> : tensor<1x16xi32> loc(#loc1042) + %m_A22_785 = arith.cmpi slt, %m_A22_782, %m_A22_784 : tensor<1x16xi32> loc(#loc1042) + %m_A22_786 = tt.broadcast %m_A22 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc1043) + %m_A22_787 = tt.broadcast %m_A22_785 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc1043) + %m_A22_788 = arith.andi %m_A22_786, %m_A22_787 : tensor<16x16xi1> loc(#loc1043) + %m_A31 = tt.expand_dims %m_tc3_57 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc1044) + %m_A31_789 = arith.constant 16 : i32 loc(#loc1045) + %m_A31_790 = arith.constant 16 : i32 loc(#loc1045) + %m_A31_791 = arith.constant dense<16> : tensor<16xi32> loc(#loc1045) + %m_A31_792 = arith.extsi %m_A31_791 : tensor<16xi32> to tensor<16xi64> loc(#loc1045) + %m_A31_793 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc1045) + %m_A31_794 = arith.addi %m_A31_792, %m_A31_793 : tensor<16xi64> loc(#loc1045) + %m_A31_795 = arith.constant 2147483647 : i64 loc(#loc1045) + %m_A31_796 = arith.constant -2147483648 : i64 loc(#loc1045) + %m_A31_797 = arith.constant dense<2147483647> : tensor<16xi64> loc(#loc1045) + %m_A31_798 = arith.cmpi sle, %m_A31_794, %m_A31_797 : tensor<16xi64> loc(#loc1045) + %m_A31_799 = arith.constant dense<-2147483648> : tensor<16xi64> loc(#loc1045) + %m_A31_800 = arith.cmpi sge, %m_A31_794, %m_A31_799 : tensor<16xi64> loc(#loc1045) + %m_A31_801 = arith.andi %m_A31_798, %m_A31_800 : tensor<16xi1> loc(#loc1045) + %m_A31_802 = arith.addi %m_A31_791, %o_i : tensor<16xi32> loc(#loc1045) + %m_A31_803 = tt.expand_dims %m_A31_802 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc1046) + %m_A31_804 = arith.constant 64 : i32 loc(#loc1047) + %m_A31_805 = arith.constant dense<64> : tensor<1x16xi32> loc(#loc1047) + %m_A31_806 = arith.cmpi slt, %m_A31_803, %m_A31_805 : tensor<1x16xi32> loc(#loc1047) + %m_A31_807 = tt.broadcast %m_A31 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc1048) + %m_A31_808 = tt.broadcast %m_A31_806 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc1048) + %m_A31_809 = arith.andi %m_A31_807, %m_A31_808 : tensor<16x16xi1> loc(#loc1048) + %m_A32 = tt.expand_dims %m_tc3_57 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc1049) + %m_A32_810 = arith.constant 32 : i32 loc(#loc1050) + %m_A32_811 = arith.constant 32 : i32 loc(#loc1050) + %m_A32_812 = arith.constant dense<32> : tensor<16xi32> loc(#loc1050) + %m_A32_813 = arith.extsi %m_A32_812 : tensor<16xi32> to tensor<16xi64> loc(#loc1050) + %m_A32_814 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc1050) + %m_A32_815 = arith.addi %m_A32_813, %m_A32_814 : tensor<16xi64> loc(#loc1050) + %m_A32_816 = arith.constant 2147483647 : i64 loc(#loc1050) + %m_A32_817 = arith.constant -2147483648 : i64 loc(#loc1050) + %m_A32_818 = arith.constant dense<2147483647> : tensor<16xi64> loc(#loc1050) + %m_A32_819 = arith.cmpi sle, %m_A32_815, %m_A32_818 : tensor<16xi64> loc(#loc1050) + %m_A32_820 = arith.constant dense<-2147483648> : tensor<16xi64> loc(#loc1050) + %m_A32_821 = arith.cmpi sge, %m_A32_815, %m_A32_820 : tensor<16xi64> loc(#loc1050) + %m_A32_822 = arith.andi %m_A32_819, %m_A32_821 : tensor<16xi1> loc(#loc1050) + %m_A32_823 = arith.addi %m_A32_812, %o_i : tensor<16xi32> loc(#loc1050) + %m_A32_824 = tt.expand_dims %m_A32_823 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc1051) + %m_A32_825 = arith.constant 64 : i32 loc(#loc1052) + %m_A32_826 = arith.constant dense<64> : tensor<1x16xi32> loc(#loc1052) + %m_A32_827 = arith.cmpi slt, %m_A32_824, %m_A32_826 : tensor<1x16xi32> loc(#loc1052) + %m_A32_828 = tt.broadcast %m_A32 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc1053) + %m_A32_829 = tt.broadcast %m_A32_827 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc1053) + %m_A32_830 = arith.andi %m_A32_828, %m_A32_829 : tensor<16x16xi1> loc(#loc1053) + %m_A33 = tt.expand_dims %m_tc3_57 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc1054) + %m_A33_831 = arith.constant 48 : i32 loc(#loc1055) + %m_A33_832 = arith.constant 48 : i32 loc(#loc1055) + %m_A33_833 = arith.constant dense<48> : tensor<16xi32> loc(#loc1055) + %m_A33_834 = arith.extsi %m_A33_833 : tensor<16xi32> to tensor<16xi64> loc(#loc1055) + %m_A33_835 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc1055) + %m_A33_836 = arith.addi %m_A33_834, %m_A33_835 : tensor<16xi64> loc(#loc1055) + %m_A33_837 = arith.constant 2147483647 : i64 loc(#loc1055) + %m_A33_838 = arith.constant -2147483648 : i64 loc(#loc1055) + %m_A33_839 = arith.constant dense<2147483647> : tensor<16xi64> loc(#loc1055) + %m_A33_840 = arith.cmpi sle, %m_A33_836, %m_A33_839 : tensor<16xi64> loc(#loc1055) + %m_A33_841 = arith.constant dense<-2147483648> : tensor<16xi64> loc(#loc1055) + %m_A33_842 = arith.cmpi sge, %m_A33_836, %m_A33_841 : tensor<16xi64> loc(#loc1055) + %m_A33_843 = arith.andi %m_A33_840, %m_A33_842 : tensor<16xi1> loc(#loc1055) + %m_A33_844 = arith.addi %m_A33_833, %o_i : tensor<16xi32> loc(#loc1055) + %m_A33_845 = tt.expand_dims %m_A33_844 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc1056) + %m_A33_846 = arith.constant 64 : i32 loc(#loc1057) + %m_A33_847 = arith.constant dense<64> : tensor<1x16xi32> loc(#loc1057) + %m_A33_848 = arith.cmpi slt, %m_A33_845, %m_A33_847 : tensor<1x16xi32> loc(#loc1057) + %m_A33_849 = tt.broadcast %m_A33 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc1058) + %m_A33_850 = tt.broadcast %m_A33_848 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc1058) + %m_A33_851 = arith.andi %m_A33_849, %m_A33_850 : tensor<16x16xi1> loc(#loc1058) + %51 = arith.truncf %b_Ai00_413 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc522) + tt.store %p_A00_496, %51, %m_A0_704 : tensor<16x16x!tt.ptr> loc(#loc523) + %52 = arith.truncf %b_Ai10_427 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc524) + tt.store %p_A10_509, %52, %m_A1_711 : tensor<16x16x!tt.ptr> loc(#loc525) + %53 = arith.truncf %b_Ai11_415 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc526) + tt.store %p_A11_536, %53, %m_A11_746 : tensor<16x16x!tt.ptr> loc(#loc527) + %54 = arith.truncf %b_Ai20_455 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc528) + tt.store %p_A20_549, %54, %m_A2_718 : tensor<16x16x!tt.ptr> loc(#loc529) + %55 = arith.truncf %b_Ai21_435 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc530) + tt.store %p_A21_576, %55, %m_A21_767 : tensor<16x16x!tt.ptr> loc(#loc531) + %56 = arith.truncf %b_Ai22_417 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc532) + tt.store %p_A22_603, %56, %m_A22_788 : tensor<16x16x!tt.ptr> loc(#loc533) + %57 = arith.truncf %b_Ai30_483 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc534) + tt.store %p_A30_616, %57, %m_A3_725 : tensor<16x16x!tt.ptr> loc(#loc535) + %58 = arith.truncf %b_Ai31_467 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc536) + tt.store %p_A31_643, %58, %m_A31_809 : tensor<16x16x!tt.ptr> loc(#loc537) + %59 = arith.truncf %b_Ai32_443 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc538) + tt.store %p_A32_670, %59, %m_A32_830 : tensor<16x16x!tt.ptr> loc(#loc539) + %60 = arith.truncf %b_Ai33_419 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc540) + tt.store %p_A33_697, %60, %m_A33_851 : tensor<16x16x!tt.ptr> loc(#loc541) + tt.return loc(#loc542) + } loc(#loc) + tt.func private @triton.language.standard.zeros__Tc16_c16T_cfp32() -> tensor<16x16xf32> attributes {noinline = false} { + %cst = arith.constant 0.000000e+00 : f32 loc(#loc544) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc544) + tt.return %cst_0 : tensor<16x16xf32> loc(#loc545) + ^bb1: // no predecessors + %0 = ub.poison : tensor<16x16xf32> loc(#loc546) + tt.return %0 : tensor<16x16xf32> loc(#loc546) + } loc(#loc543) + tt.func private @triton.language.standard.cdiv__c128_c64() attributes {noinline = false} { + tt.return loc(#loc548) + ^bb1: // no predecessors + tt.return loc(#loc549) + } loc(#loc547) + tt.func private @fla.ops.utils.op.exp2__fp32S16_16S(%x: tensor<16x16xf32> loc("x"(#loc550))) -> tensor<16x16xf32> attributes {noinline = false} { + %0 = math.exp2 %x : tensor<16x16xf32> loc(#loc551) + tt.return %0 : tensor<16x16xf32> loc(#loc552) + ^bb1: // no predecessors + %1 = ub.poison : tensor<16x16xf32> loc(#loc553) + tt.return %1 : tensor<16x16xf32> loc(#loc553) + } loc(#loc550) + tt.func private @triton.language.standard.sum__fp32S16_16S_c0_cFalse_cNone(%input: tensor<16x16xf32> loc("input"(#loc554))) -> tensor<16xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 0 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc555) + tt.reduce.return %2 : f32 loc(#loc555) + }) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc555) + tt.return %0 : tensor<16xf32> loc(#loc557) + ^bb1: // no predecessors + %1 = ub.poison : tensor<16xf32> loc(#loc558) + tt.return %1 : tensor<16xf32> loc(#loc558) + } loc(#loc554) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc559)), %b: f32 loc("b"(#loc559))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc560) + tt.return %0 : f32 loc(#loc561) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc562) + tt.return %1 : f32 loc(#loc562) + } loc(#loc559) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":70:30) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":70:36) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":70:61) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":70:67) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":71:23) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":71:34) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":78:25) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":78:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":78:38) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":80:13) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":80:19) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":81:8) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":83:18) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":84:18) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":84:23) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":85:18) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":85:23) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":86:18) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":86:23) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":88:16) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":88:28) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":88:20) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":88:40) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":88:9) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":89:16) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":89:21) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":89:28) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":89:9) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":91:23) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":92:21) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":92:28) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":93:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":93:28) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":94:21) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":94:28) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":95:21) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":95:28) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":98:24) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":98:18) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":98:29) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":98:44) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":98:51) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":98:35) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":99:24) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":99:18) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":99:29) +#loc47 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":99:44) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":99:51) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":99:35) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":100:24) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":100:18) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":100:29) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":100:44) +#loc54 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":100:51) +#loc55 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":100:35) +#loc56 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":101:24) +#loc57 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":101:18) +#loc58 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":101:29) +#loc59 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":101:44) +#loc60 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":101:51) +#loc61 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":101:35) +#loc62 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":102:19) +#loc63 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":102:51) +#loc64 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":103:19) +#loc65 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":103:51) +#loc66 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":104:19) +#loc67 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":104:51) +#loc68 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":105:19) +#loc69 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":105:51) +#loc70 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":109:25) +#loc71 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":109:19) +#loc72 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":109:30) +#loc73 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":109:45) +#loc74 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":109:52) +#loc75 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":109:36) +#loc76 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":110:25) +#loc77 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":110:19) +#loc78 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":110:30) +#loc79 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":110:45) +#loc80 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":110:52) +#loc81 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":110:36) +#loc82 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":111:25) +#loc83 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":111:19) +#loc84 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":111:30) +#loc85 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":111:45) +#loc86 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":111:52) +#loc87 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":111:36) +#loc88 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":112:25) +#loc89 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":112:19) +#loc90 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":112:30) +#loc91 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":112:45) +#loc92 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":112:52) +#loc93 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":112:36) +#loc94 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":114:23) +#loc95 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":115:23) +#loc96 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":116:23) +#loc97 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":117:23) +#loc98 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":124:21) +#loc99 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":125:21) +#loc100 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":126:21) +#loc101 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":127:21) +#loc102 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":130:21) +#loc103 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":131:21) +#loc104 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":132:21) +#loc105 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":133:21) +#loc106 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":134:21) +#loc107 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":135:21) +#loc108 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":137:32) +#loc109 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":137:21) +#loc110 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":138:20) +#loc111 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":138:38) +#loc112 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":138:25) +#loc113 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:28) +#loc114 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:33) +#loc115 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:45) +#loc116 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:19) +#loc117 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:56) +#loc118 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:52) +#loc119 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":140:40) +#loc120 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":140:56) +#loc121 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":140:67) +#loc122 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":140:52) +#loc123 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":140:23) +#loc124 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":142:38) +#loc125 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":142:45) +#loc126 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":144:19) +#loc127 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":144:11) +#loc128 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":145:32) +#loc129 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":145:37) +#loc130 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":145:49) +#loc131 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":145:23) +#loc132 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":145:60) +#loc133 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":145:56) +#loc134 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":146:44) +#loc135 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":146:60) +#loc136 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":146:71) +#loc137 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":146:56) +#loc138 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":146:27) +#loc139 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":148:42) +#loc140 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":148:49) +#loc141 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":150:42) +#loc142 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":150:49) +#loc143 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":152:23) +#loc144 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":152:15) +#loc145 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":153:36) +#loc146 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":153:41) +#loc147 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":153:53) +#loc148 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":153:27) +#loc149 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":153:64) +#loc150 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":153:60) +#loc151 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":154:48) +#loc152 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":154:64) +#loc153 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":154:75) +#loc154 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":154:60) +#loc155 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":154:31) +#loc156 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":156:46) +#loc157 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":156:53) +#loc158 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":158:46) +#loc159 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":158:53) +#loc160 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":159:46) +#loc161 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":159:53) +#loc162 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":161:27) +#loc163 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":161:19) +#loc164 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":162:40) +#loc165 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":162:45) +#loc166 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":162:57) +#loc167 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":162:31) +#loc168 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":162:68) +#loc169 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":162:64) +#loc170 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":163:52) +#loc171 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":163:68) +#loc172 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":163:79) +#loc173 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":163:64) +#loc174 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":163:35) +#loc175 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":165:50) +#loc176 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":165:57) +#loc177 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":167:50) +#loc178 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":167:57) +#loc179 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":168:50) +#loc180 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":168:57) +#loc181 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":169:50) +#loc182 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":169:57) +#loc183 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":144:8) +#loc184 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":179:14) +#loc185 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":179:29) +#loc186 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":179:25) +#loc187 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":180:14) +#loc188 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":180:30) +#loc189 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":180:26) +#loc190 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:38) +#loc191 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:32) +#loc192 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:55) +#loc193 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:49) +#loc194 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:75) +#loc195 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:91) +#loc196 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:86) +#loc197 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:70) +#loc198 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:102) +#loc199 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:17) +#loc200 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:38) +#loc201 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:32) +#loc202 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:55) +#loc203 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:49) +#loc204 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:75) +#loc205 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:91) +#loc206 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:86) +#loc207 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:70) +#loc208 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:102) +#loc209 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:17) +#loc210 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:38) +#loc211 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:32) +#loc212 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:55) +#loc213 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:49) +#loc214 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:75) +#loc215 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:91) +#loc216 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:86) +#loc217 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:70) +#loc218 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:102) +#loc219 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:17) +#loc220 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:38) +#loc221 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:32) +#loc222 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:55) +#loc223 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:49) +#loc224 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:75) +#loc225 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:91) +#loc226 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:86) +#loc227 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:70) +#loc228 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:102) +#loc229 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:17) +#loc230 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:32) +#loc231 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:49) +#loc232 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:43) +#loc233 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:69) +#loc234 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:85) +#loc235 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:80) +#loc236 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:64) +#loc237 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:96) +#loc238 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:17) +#loc239 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:32) +#loc240 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:49) +#loc241 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:43) +#loc242 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:69) +#loc243 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:85) +#loc244 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:80) +#loc245 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:64) +#loc246 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:96) +#loc247 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:17) +#loc248 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:32) +#loc249 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:49) +#loc250 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:43) +#loc251 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:69) +#loc252 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:85) +#loc253 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:80) +#loc254 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:64) +#loc255 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:96) +#loc256 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:17) +#loc257 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:32) +#loc258 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:49) +#loc259 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:43) +#loc260 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:69) +#loc261 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:85) +#loc262 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:80) +#loc263 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:64) +#loc264 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:96) +#loc265 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:17) +#loc266 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:32) +#loc267 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:49) +#loc268 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:43) +#loc269 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:69) +#loc270 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:85) +#loc271 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:80) +#loc272 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:64) +#loc273 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:96) +#loc274 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:17) +#loc275 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:32) +#loc276 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:49) +#loc277 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:43) +#loc278 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:69) +#loc279 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:85) +#loc280 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:80) +#loc281 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:64) +#loc282 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:96) +#loc283 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:17) +#loc284 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":201:25) +#loc285 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":201:20) +#loc286 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":202:25) +#loc287 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":202:20) +#loc288 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":203:25) +#loc289 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":203:20) +#loc290 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":204:25) +#loc291 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":204:20) +#loc292 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":207:25) +#loc293 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":207:20) +#loc294 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":208:25) +#loc295 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":208:20) +#loc296 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":209:25) +#loc297 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":209:20) +#loc298 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":210:25) +#loc299 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":210:20) +#loc300 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":211:25) +#loc301 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":211:20) +#loc302 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":212:25) +#loc303 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":212:20) +#loc304 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":222:14) +#loc305 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":223:14) +#loc306 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":224:14) +#loc307 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":225:14) +#loc308 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":227:34) +#loc309 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":227:30) +#loc310 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":227:22) +#loc311 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":228:40) +#loc312 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":228:43) +#loc313 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":228:54) +#loc314 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":228:61) +#loc315 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":228:66) +#loc316 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":229:31) +#loc317 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":229:41) +#loc318 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":230:37) +#loc319 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":230:48) +#loc320 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":230:56) +#loc321 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":230:24) +#loc322 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":231:34) +#loc323 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":231:37) +#loc324 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":231:54) +#loc325 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":231:8) +#loc326 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":232:34) +#loc327 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":232:30) +#loc328 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":232:22) +#loc329 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":233:40) +#loc330 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":233:43) +#loc331 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":233:54) +#loc332 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":233:61) +#loc333 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":233:66) +#loc334 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":234:31) +#loc335 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":234:41) +#loc336 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":235:37) +#loc337 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":235:48) +#loc338 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":235:56) +#loc339 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":235:24) +#loc340 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":236:34) +#loc341 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":236:37) +#loc342 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":236:54) +#loc343 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":236:8) +#loc344 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":237:34) +#loc345 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":237:30) +#loc346 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":237:22) +#loc347 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":238:40) +#loc348 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":238:43) +#loc349 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":238:54) +#loc350 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":238:61) +#loc351 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":238:66) +#loc352 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":239:31) +#loc353 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":239:41) +#loc354 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":240:37) +#loc355 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":240:48) +#loc356 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":240:56) +#loc357 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":240:24) +#loc358 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":241:34) +#loc359 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":241:37) +#loc360 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":241:54) +#loc361 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":241:8) +#loc362 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":242:34) +#loc363 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":242:30) +#loc364 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":242:22) +#loc365 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":243:40) +#loc366 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":243:43) +#loc367 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":243:54) +#loc368 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":243:61) +#loc369 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":243:66) +#loc370 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":244:31) +#loc371 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":244:41) +#loc372 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":245:37) +#loc373 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":245:48) +#loc374 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":245:56) +#loc375 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":245:24) +#loc376 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":246:34) +#loc377 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":246:37) +#loc378 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":246:54) +#loc379 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":246:8) +#loc380 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":248:14) +#loc381 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":249:14) +#loc382 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":250:14) +#loc383 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":251:14) +#loc384 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":258:23) +#loc385 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":259:8) +#loc386 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":257:14) +#loc387 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":263:23) +#loc388 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":264:8) +#loc389 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":262:14) +#loc390 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":268:23) +#loc391 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":269:8) +#loc392 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":267:14) +#loc393 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":275:22) +#loc394 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":276:22) +#loc395 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":276:8) +#loc396 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":275:8) +#loc397 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":273:14) +#loc398 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":281:22) +#loc399 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":282:22) +#loc400 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":282:8) +#loc401 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":281:8) +#loc402 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":279:14) +#loc403 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":287:22) +#loc404 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":288:22) +#loc405 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":288:8) +#loc406 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":289:22) +#loc407 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":289:8) +#loc408 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":287:8) +#loc409 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":285:14) +#loc410 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":297:25) +#loc411 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":297:30) +#loc412 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":297:42) +#loc413 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":297:16) +#loc414 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":297:55) +#loc415 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":297:51) +#loc416 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":298:25) +#loc417 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":298:30) +#loc418 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":298:42) +#loc419 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":298:16) +#loc420 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":298:55) +#loc421 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":298:51) +#loc422 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":299:25) +#loc423 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":299:30) +#loc424 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":299:42) +#loc425 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":299:16) +#loc426 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":299:57) +#loc427 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":299:62) +#loc428 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":299:51) +#loc429 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":300:25) +#loc430 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":300:30) +#loc431 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":300:42) +#loc432 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":300:16) +#loc433 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":300:55) +#loc434 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":300:51) +#loc435 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":301:25) +#loc436 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":301:30) +#loc437 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":301:42) +#loc438 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":301:16) +#loc439 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":301:57) +#loc440 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":301:62) +#loc441 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":301:51) +#loc442 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":302:25) +#loc443 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":302:30) +#loc444 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":302:42) +#loc445 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":302:16) +#loc446 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":302:59) +#loc447 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":302:64) +#loc448 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":302:51) +#loc449 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":303:25) +#loc450 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":303:30) +#loc451 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":303:42) +#loc452 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":303:16) +#loc453 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":303:55) +#loc454 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":303:51) +#loc455 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":304:25) +#loc456 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":304:30) +#loc457 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":304:42) +#loc458 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":304:16) +#loc459 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":304:57) +#loc460 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":304:62) +#loc461 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":304:51) +#loc462 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":305:25) +#loc463 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":305:30) +#loc464 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":305:42) +#loc465 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":305:16) +#loc466 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":305:59) +#loc467 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":305:64) +#loc468 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":305:51) +#loc469 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":306:25) +#loc470 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":306:30) +#loc471 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":306:42) +#loc472 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":306:16) +#loc473 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":306:59) +#loc474 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":306:64) +#loc475 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":306:51) +#loc476 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":308:17) +#loc477 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":308:33) +#loc478 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":308:44) +#loc479 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":308:29) +#loc480 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":309:17) +#loc481 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":309:33) +#loc482 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":309:44) +#loc483 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":309:29) +#loc484 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":310:17) +#loc485 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":310:33) +#loc486 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":310:44) +#loc487 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":310:29) +#loc488 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":311:17) +#loc489 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":311:33) +#loc490 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":311:44) +#loc491 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":311:29) +#loc492 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":312:18) +#loc493 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":312:36) +#loc494 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":312:41) +#loc495 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":312:52) +#loc496 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":312:30) +#loc497 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":313:18) +#loc498 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":313:36) +#loc499 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":313:41) +#loc500 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":313:52) +#loc501 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":313:30) +#loc502 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":314:18) +#loc503 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":314:38) +#loc504 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":314:43) +#loc505 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":314:54) +#loc506 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":314:30) +#loc507 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":315:18) +#loc508 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":315:36) +#loc509 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":315:41) +#loc510 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":315:52) +#loc511 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":315:30) +#loc512 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":316:18) +#loc513 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":316:38) +#loc514 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":316:43) +#loc515 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":316:54) +#loc516 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":316:30) +#loc517 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":317:18) +#loc518 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":317:38) +#loc519 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":317:43) +#loc520 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":317:54) +#loc521 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":317:30) +#loc522 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":319:30) +#loc523 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":319:20) +#loc524 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":320:30) +#loc525 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":320:20) +#loc526 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":321:30) +#loc527 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":321:20) +#loc528 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":322:30) +#loc529 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":322:20) +#loc530 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":323:30) +#loc531 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":323:20) +#loc532 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":324:30) +#loc533 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":324:20) +#loc534 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":325:30) +#loc535 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":325:20) +#loc536 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":326:30) +#loc537 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":326:20) +#loc538 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":327:30) +#loc539 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":327:20) +#loc540 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":328:30) +#loc541 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":328:20) +#loc542 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":328:4) +#loc543 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":120:0) +#loc544 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":129:31) +#loc545 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":129:11) +#loc546 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":129:4) +#loc547 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":34:0) +#loc548 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":43:11) +#loc549 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":43:4) +#loc551 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:33) +#loc552 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:20) +#loc553 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:13) +#loc555 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc557 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc558 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc560 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc561 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc562 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc568 = loc("i_tc0"(#loc13)) +#loc569 = loc("i_tc1"(#loc14)) +#loc570 = loc("i_tc1"(#loc15)) +#loc571 = loc("i_tc2"(#loc16)) +#loc572 = loc("i_tc2"(#loc17)) +#loc573 = loc("i_tc3"(#loc18)) +#loc574 = loc("i_tc3"(#loc19)) +#loc575 = loc("k"(#loc20)) +#loc576 = loc("k"(#loc21)) +#loc577 = loc("k"(#loc22)) +#loc578 = loc("k"(#loc23)) +#loc579 = loc("k"(#loc24)) +#loc580 = loc("A"(#loc25)) +#loc581 = loc("A"(#loc26)) +#loc582 = loc("A"(#loc27)) +#loc583 = loc("A"(#loc28)) +#loc584 = loc("o_i"(#loc29)) +#loc585 = loc("m_tc0"(#loc30)) +#loc586 = loc("m_tc0"(#loc31)) +#loc587 = loc("m_tc1"(#loc32)) +#loc588 = loc("m_tc1"(#loc33)) +#loc589 = loc("m_tc2"(#loc34)) +#loc590 = loc("m_tc2"(#loc35)) +#loc591 = loc("m_tc3"(#loc36)) +#loc592 = loc("m_tc3"(#loc37)) +#loc593 = loc("p_b0"(#loc38)) +#loc594 = loc("p_b0"(#loc39)) +#loc595 = loc("p_b0"(#loc40)) +#loc596 = loc("p_b0"(#loc41)) +#loc597 = loc("p_b0"(#loc42)) +#loc598 = loc("p_b0"(#loc43)) +#loc599 = loc("p_b1"(#loc44)) +#loc600 = loc("p_b1"(#loc45)) +#loc601 = loc("p_b1"(#loc46)) +#loc602 = loc("p_b1"(#loc47)) +#loc603 = loc("p_b1"(#loc48)) +#loc604 = loc("p_b1"(#loc49)) +#loc605 = loc("p_b2"(#loc50)) +#loc606 = loc("p_b2"(#loc51)) +#loc607 = loc("p_b2"(#loc52)) +#loc608 = loc("p_b2"(#loc53)) +#loc609 = loc("p_b2"(#loc54)) +#loc610 = loc("p_b2"(#loc55)) +#loc611 = loc("p_b3"(#loc56)) +#loc612 = loc("p_b3"(#loc57)) +#loc613 = loc("p_b3"(#loc58)) +#loc614 = loc("p_b3"(#loc59)) +#loc615 = loc("p_b3"(#loc60)) +#loc616 = loc("p_b3"(#loc61)) +#loc617 = loc("b_b0"(#loc62)) +#loc618 = loc("b_b0"(#loc63)) +#loc619 = loc("b_b1"(#loc64)) +#loc620 = loc("b_b1"(#loc65)) +#loc621 = loc("b_b2"(#loc66)) +#loc622 = loc("b_b2"(#loc67)) +#loc623 = loc("b_b3"(#loc68)) +#loc624 = loc("b_b3"(#loc69)) +#loc625 = loc("p_g0"(#loc70)) +#loc626 = loc("p_g0"(#loc71)) +#loc627 = loc("p_g0"(#loc72)) +#loc628 = loc("p_g0"(#loc73)) +#loc629 = loc("p_g0"(#loc74)) +#loc630 = loc("p_g0"(#loc75)) +#loc631 = loc("p_g1"(#loc76)) +#loc632 = loc("p_g1"(#loc77)) +#loc633 = loc("p_g1"(#loc78)) +#loc634 = loc("p_g1"(#loc79)) +#loc635 = loc("p_g1"(#loc80)) +#loc636 = loc("p_g1"(#loc81)) +#loc637 = loc("p_g2"(#loc82)) +#loc638 = loc("p_g2"(#loc83)) +#loc639 = loc("p_g2"(#loc84)) +#loc640 = loc("p_g2"(#loc85)) +#loc641 = loc("p_g2"(#loc86)) +#loc642 = loc("p_g2"(#loc87)) +#loc643 = loc("p_g3"(#loc88)) +#loc644 = loc("p_g3"(#loc89)) +#loc645 = loc("p_g3"(#loc90)) +#loc646 = loc("p_g3"(#loc91)) +#loc647 = loc("p_g3"(#loc92)) +#loc648 = loc("p_g3"(#loc93)) +#loc649 = loc("b_g0"(#loc94)) +#loc650 = loc("b_g1"(#loc95)) +#loc651 = loc("b_g2"(#loc96)) +#loc652 = loc("b_g3"(#loc97)) +#loc653 = loc("b_A00"(#loc98)) +#loc654 = loc("b_A11"(#loc99)) +#loc655 = loc("b_A22"(#loc100)) +#loc656 = loc("b_A33"(#loc101)) +#loc657 = loc("b_A10"(#loc102)) +#loc658 = loc("b_A20"(#loc103)) +#loc659 = loc("b_A21"(#loc104)) +#loc660 = loc("b_A30"(#loc105)) +#loc661 = loc("b_A31"(#loc106)) +#loc662 = loc("b_A32"(#loc107)) +#loc663 = loc("b_A00"(#loc109)) +#loc664 = loc("o_k"(#loc110)) +#loc665 = loc("o_k"(#loc111)) +#loc666 = loc("o_k"(#loc112)) +#loc667 = loc("p_k0"(#loc113)) +#loc668 = loc("p_k0"(#loc114)) +#loc669 = loc("p_k0"(#loc115)) +#loc670 = loc("p_k0"(#loc116)) +#loc671 = loc("p_k0"(#loc117)) +#loc672 = loc("p_k0"(#loc118)) +#loc673 = loc("b_k0"(#loc119)) +#loc674 = loc("b_k0"(#loc120)) +#loc675 = loc("b_k0"(#loc121)) +#loc676 = loc("b_k0"(#loc122)) +#loc677 = loc("b_k0"(#loc123)) +#loc678 = loc("b_A00"(#loc124)) +#loc679 = loc("b_A00"(#loc125)) +#loc680 = loc("p_k1"(#loc128)) +#loc681 = loc("p_k1"(#loc129)) +#loc682 = loc("p_k1"(#loc130)) +#loc683 = loc("p_k1"(#loc131)) +#loc684 = loc("p_k1"(#loc132)) +#loc685 = loc("p_k1"(#loc133)) +#loc686 = loc("b_k1"(#loc134)) +#loc687 = loc("b_k1"(#loc135)) +#loc688 = loc("b_k1"(#loc136)) +#loc689 = loc("b_k1"(#loc137)) +#loc690 = loc("b_k1"(#loc138)) +#loc691 = loc("b_A11"(#loc139)) +#loc692 = loc("b_A11"(#loc140)) +#loc693 = loc("b_A10"(#loc141)) +#loc694 = loc("b_A10"(#loc142)) +#loc695 = loc("b_A22"(#loc144)) +#loc696 = loc("p_k2"(#loc145)) +#loc697 = loc("p_k2"(#loc146)) +#loc698 = loc("p_k2"(#loc147)) +#loc699 = loc("p_k2"(#loc148)) +#loc700 = loc("p_k2"(#loc149)) +#loc701 = loc("p_k2"(#loc150)) +#loc702 = loc("b_k2"(#loc151)) +#loc703 = loc("b_k2"(#loc152)) +#loc704 = loc("b_k2"(#loc153)) +#loc705 = loc("b_k2"(#loc154)) +#loc706 = loc("b_k2"(#loc155)) +#loc707 = loc("b_A22"(#loc156)) +#loc708 = loc("b_A22"(#loc157)) +#loc709 = loc("b_A20"(#loc158)) +#loc710 = loc("b_A20"(#loc159)) +#loc711 = loc("b_A21"(#loc160)) +#loc712 = loc("b_A21"(#loc161)) +#loc713 = loc("b_A33"(#loc163)) +#loc714 = loc("p_k3"(#loc164)) +#loc715 = loc("p_k3"(#loc165)) +#loc716 = loc("p_k3"(#loc166)) +#loc717 = loc("p_k3"(#loc167)) +#loc718 = loc("p_k3"(#loc168)) +#loc719 = loc("p_k3"(#loc169)) +#loc720 = loc("b_k3"(#loc170)) +#loc721 = loc("b_k3"(#loc171)) +#loc722 = loc("b_k3"(#loc172)) +#loc723 = loc("b_k3"(#loc173)) +#loc724 = loc("b_k3"(#loc174)) +#loc725 = loc("b_A33"(#loc175)) +#loc726 = loc("b_A33"(#loc176)) +#loc727 = loc("b_A30"(#loc177)) +#loc728 = loc("b_A30"(#loc178)) +#loc729 = loc("b_A31"(#loc179)) +#loc730 = loc("b_A31"(#loc180)) +#loc731 = loc("b_A32"(#loc181)) +#loc732 = loc("b_A32"(#loc182)) +#loc733 = loc("m_d"(#loc184)) +#loc734 = loc("m_d"(#loc185)) +#loc735 = loc("m_d"(#loc186)) +#loc736 = loc("m_I"(#loc187)) +#loc737 = loc("m_I"(#loc188)) +#loc738 = loc("m_I"(#loc189)) +#loc739 = loc("b_A00"(#loc190)) +#loc740 = loc("b_A00"(#loc191)) +#loc741 = loc("b_A00"(#loc192)) +#loc742 = loc("b_A00"(#loc193)) +#loc743 = loc("b_A00"(#loc194)) +#loc744 = loc("b_A00"(#loc195)) +#loc745 = loc("b_A00"(#loc196)) +#loc746 = loc("b_A00"(#loc197)) +#loc747 = loc("b_A00"(#loc198)) +#loc748 = loc("b_A00"(#loc199)) +#loc749 = loc("b_A11"(#loc200)) +#loc750 = loc("b_A11"(#loc201)) +#loc751 = loc("b_A11"(#loc202)) +#loc752 = loc("b_A11"(#loc203)) +#loc753 = loc("b_A11"(#loc204)) +#loc754 = loc("b_A11"(#loc205)) +#loc755 = loc("b_A11"(#loc206)) +#loc756 = loc("b_A11"(#loc207)) +#loc757 = loc("b_A11"(#loc208)) +#loc758 = loc("b_A11"(#loc209)) +#loc759 = loc("b_A22"(#loc210)) +#loc760 = loc("b_A22"(#loc211)) +#loc761 = loc("b_A22"(#loc212)) +#loc762 = loc("b_A22"(#loc213)) +#loc763 = loc("b_A22"(#loc214)) +#loc764 = loc("b_A22"(#loc215)) +#loc765 = loc("b_A22"(#loc216)) +#loc766 = loc("b_A22"(#loc217)) +#loc767 = loc("b_A22"(#loc218)) +#loc768 = loc("b_A22"(#loc219)) +#loc769 = loc("b_A33"(#loc220)) +#loc770 = loc("b_A33"(#loc221)) +#loc771 = loc("b_A33"(#loc222)) +#loc772 = loc("b_A33"(#loc223)) +#loc773 = loc("b_A33"(#loc224)) +#loc774 = loc("b_A33"(#loc225)) +#loc775 = loc("b_A33"(#loc226)) +#loc776 = loc("b_A33"(#loc227)) +#loc777 = loc("b_A33"(#loc228)) +#loc778 = loc("b_A33"(#loc229)) +#loc779 = loc("b_A10"(#loc230)) +#loc780 = loc("b_A10"(#loc231)) +#loc781 = loc("b_A10"(#loc232)) +#loc782 = loc("b_A10"(#loc233)) +#loc783 = loc("b_A10"(#loc234)) +#loc784 = loc("b_A10"(#loc235)) +#loc785 = loc("b_A10"(#loc236)) +#loc786 = loc("b_A10"(#loc237)) +#loc787 = loc("b_A10"(#loc238)) +#loc788 = loc("b_A20"(#loc239)) +#loc789 = loc("b_A20"(#loc240)) +#loc790 = loc("b_A20"(#loc241)) +#loc791 = loc("b_A20"(#loc242)) +#loc792 = loc("b_A20"(#loc243)) +#loc793 = loc("b_A20"(#loc244)) +#loc794 = loc("b_A20"(#loc245)) +#loc795 = loc("b_A20"(#loc246)) +#loc796 = loc("b_A20"(#loc247)) +#loc797 = loc("b_A21"(#loc248)) +#loc798 = loc("b_A21"(#loc249)) +#loc799 = loc("b_A21"(#loc250)) +#loc800 = loc("b_A21"(#loc251)) +#loc801 = loc("b_A21"(#loc252)) +#loc802 = loc("b_A21"(#loc253)) +#loc803 = loc("b_A21"(#loc254)) +#loc804 = loc("b_A21"(#loc255)) +#loc805 = loc("b_A21"(#loc256)) +#loc806 = loc("b_A30"(#loc257)) +#loc807 = loc("b_A30"(#loc258)) +#loc808 = loc("b_A30"(#loc259)) +#loc809 = loc("b_A30"(#loc260)) +#loc810 = loc("b_A30"(#loc261)) +#loc811 = loc("b_A30"(#loc262)) +#loc812 = loc("b_A30"(#loc263)) +#loc813 = loc("b_A30"(#loc264)) +#loc814 = loc("b_A30"(#loc265)) +#loc815 = loc("b_A31"(#loc266)) +#loc816 = loc("b_A31"(#loc267)) +#loc817 = loc("b_A31"(#loc268)) +#loc818 = loc("b_A31"(#loc269)) +#loc819 = loc("b_A31"(#loc270)) +#loc820 = loc("b_A31"(#loc271)) +#loc821 = loc("b_A31"(#loc272)) +#loc822 = loc("b_A31"(#loc273)) +#loc823 = loc("b_A31"(#loc274)) +#loc824 = loc("b_A32"(#loc275)) +#loc825 = loc("b_A32"(#loc276)) +#loc826 = loc("b_A32"(#loc277)) +#loc827 = loc("b_A32"(#loc278)) +#loc828 = loc("b_A32"(#loc279)) +#loc829 = loc("b_A32"(#loc280)) +#loc830 = loc("b_A32"(#loc281)) +#loc831 = loc("b_A32"(#loc282)) +#loc832 = loc("b_A32"(#loc283)) +#loc833 = loc("b_A00"(#loc284)) +#loc834 = loc("b_A00"(#loc285)) +#loc835 = loc("b_A11"(#loc286)) +#loc836 = loc("b_A11"(#loc287)) +#loc837 = loc("b_A22"(#loc288)) +#loc838 = loc("b_A22"(#loc289)) +#loc839 = loc("b_A33"(#loc290)) +#loc840 = loc("b_A33"(#loc291)) +#loc841 = loc("b_A10"(#loc292)) +#loc842 = loc("b_A10"(#loc293)) +#loc843 = loc("b_A20"(#loc294)) +#loc844 = loc("b_A20"(#loc295)) +#loc845 = loc("b_A21"(#loc296)) +#loc846 = loc("b_A21"(#loc297)) +#loc847 = loc("b_A30"(#loc298)) +#loc848 = loc("b_A30"(#loc299)) +#loc849 = loc("b_A31"(#loc300)) +#loc850 = loc("b_A31"(#loc301)) +#loc851 = loc("b_A32"(#loc302)) +#loc852 = loc("b_A32"(#loc303)) +#loc853 = loc("b_Ai00"(#loc304)) +#loc854 = loc("b_Ai11"(#loc305)) +#loc855 = loc("b_Ai22"(#loc306)) +#loc856 = loc("b_Ai33"(#loc307)) +#loc857 = loc("b_Ai00"(#loc310)) +#loc858 = loc("b_a00"(#loc311)) +#loc859 = loc("b_a00"(#loc312)) +#loc860 = loc("b_a00"(#loc313)) +#loc861 = loc("b_a00"(#loc314)) +#loc862 = loc("b_a00"(#loc315)) +#loc863 = loc("b_a00"(#loc316)) +#loc864 = loc("b_a00"(#loc317)) +#loc865 = loc("b_a00"(#loc318)) +#loc866 = loc("b_a00"(#loc319)) +#loc867 = loc("b_a00"(#loc320)) +#loc868 = loc("b_a00"(#loc321)) +#loc869 = loc("b_Ai00"(#loc322)) +#loc870 = loc("b_Ai00"(#loc323)) +#loc871 = loc("b_Ai00"(#loc324)) +#loc872 = loc("b_Ai11"(#loc328)) +#loc873 = loc("b_a11"(#loc329)) +#loc874 = loc("b_a11"(#loc330)) +#loc875 = loc("b_a11"(#loc331)) +#loc876 = loc("b_a11"(#loc332)) +#loc877 = loc("b_a11"(#loc333)) +#loc878 = loc("b_a11"(#loc334)) +#loc879 = loc("b_a11"(#loc335)) +#loc880 = loc("b_a11"(#loc336)) +#loc881 = loc("b_a11"(#loc337)) +#loc882 = loc("b_a11"(#loc338)) +#loc883 = loc("b_a11"(#loc339)) +#loc884 = loc("b_Ai11"(#loc340)) +#loc885 = loc("b_Ai11"(#loc341)) +#loc886 = loc("b_Ai11"(#loc342)) +#loc887 = loc("b_Ai22"(#loc346)) +#loc888 = loc("b_a22"(#loc347)) +#loc889 = loc("b_a22"(#loc348)) +#loc890 = loc("b_a22"(#loc349)) +#loc891 = loc("b_a22"(#loc350)) +#loc892 = loc("b_a22"(#loc351)) +#loc893 = loc("b_a22"(#loc352)) +#loc894 = loc("b_a22"(#loc353)) +#loc895 = loc("b_a22"(#loc354)) +#loc896 = loc("b_a22"(#loc355)) +#loc897 = loc("b_a22"(#loc356)) +#loc898 = loc("b_a22"(#loc357)) +#loc899 = loc("b_Ai22"(#loc358)) +#loc900 = loc("b_Ai22"(#loc359)) +#loc901 = loc("b_Ai22"(#loc360)) +#loc902 = loc("b_Ai33"(#loc364)) +#loc903 = loc("b_a33"(#loc365)) +#loc904 = loc("b_a33"(#loc366)) +#loc905 = loc("b_a33"(#loc367)) +#loc906 = loc("b_a33"(#loc368)) +#loc907 = loc("b_a33"(#loc369)) +#loc908 = loc("b_a33"(#loc370)) +#loc909 = loc("b_a33"(#loc371)) +#loc910 = loc("b_a33"(#loc372)) +#loc911 = loc("b_a33"(#loc373)) +#loc912 = loc("b_a33"(#loc374)) +#loc913 = loc("b_a33"(#loc375)) +#loc914 = loc("b_Ai33"(#loc376)) +#loc915 = loc("b_Ai33"(#loc377)) +#loc916 = loc("b_Ai33"(#loc378)) +#loc917 = loc("b_Ai00"(#loc380)) +#loc918 = loc("b_Ai11"(#loc381)) +#loc919 = loc("b_Ai22"(#loc382)) +#loc920 = loc("b_Ai33"(#loc383)) +#loc921 = loc("b_Ai10"(#loc384)) +#loc922 = loc("b_Ai10"(#loc385)) +#loc923 = loc("b_Ai10"(#loc386)) +#loc924 = loc("b_Ai21"(#loc387)) +#loc925 = loc("b_Ai21"(#loc388)) +#loc926 = loc("b_Ai21"(#loc389)) +#loc927 = loc("b_Ai32"(#loc390)) +#loc928 = loc("b_Ai32"(#loc391)) +#loc929 = loc("b_Ai32"(#loc392)) +#loc930 = loc("b_Ai20"(#loc393)) +#loc931 = loc("b_Ai20"(#loc394)) +#loc932 = loc("b_Ai20"(#loc395)) +#loc933 = loc("b_Ai20"(#loc396)) +#loc934 = loc("b_Ai20"(#loc397)) +#loc935 = loc("b_Ai31"(#loc398)) +#loc936 = loc("b_Ai31"(#loc399)) +#loc937 = loc("b_Ai31"(#loc400)) +#loc938 = loc("b_Ai31"(#loc401)) +#loc939 = loc("b_Ai31"(#loc402)) +#loc940 = loc("b_Ai30"(#loc403)) +#loc941 = loc("b_Ai30"(#loc404)) +#loc942 = loc("b_Ai30"(#loc405)) +#loc943 = loc("b_Ai30"(#loc406)) +#loc944 = loc("b_Ai30"(#loc407)) +#loc945 = loc("b_Ai30"(#loc408)) +#loc946 = loc("b_Ai30"(#loc409)) +#loc947 = loc("p_A00"(#loc410)) +#loc948 = loc("p_A00"(#loc411)) +#loc949 = loc("p_A00"(#loc412)) +#loc950 = loc("p_A00"(#loc413)) +#loc951 = loc("p_A00"(#loc414)) +#loc952 = loc("p_A00"(#loc415)) +#loc953 = loc("p_A10"(#loc416)) +#loc954 = loc("p_A10"(#loc417)) +#loc955 = loc("p_A10"(#loc418)) +#loc956 = loc("p_A10"(#loc419)) +#loc957 = loc("p_A10"(#loc420)) +#loc958 = loc("p_A10"(#loc421)) +#loc959 = loc("p_A11"(#loc422)) +#loc960 = loc("p_A11"(#loc423)) +#loc961 = loc("p_A11"(#loc424)) +#loc962 = loc("p_A11"(#loc425)) +#loc963 = loc("p_A11"(#loc426)) +#loc964 = loc("p_A11"(#loc427)) +#loc965 = loc("p_A11"(#loc428)) +#loc966 = loc("p_A20"(#loc429)) +#loc967 = loc("p_A20"(#loc430)) +#loc968 = loc("p_A20"(#loc431)) +#loc969 = loc("p_A20"(#loc432)) +#loc970 = loc("p_A20"(#loc433)) +#loc971 = loc("p_A20"(#loc434)) +#loc972 = loc("p_A21"(#loc435)) +#loc973 = loc("p_A21"(#loc436)) +#loc974 = loc("p_A21"(#loc437)) +#loc975 = loc("p_A21"(#loc438)) +#loc976 = loc("p_A21"(#loc439)) +#loc977 = loc("p_A21"(#loc440)) +#loc978 = loc("p_A21"(#loc441)) +#loc979 = loc("p_A22"(#loc442)) +#loc980 = loc("p_A22"(#loc443)) +#loc981 = loc("p_A22"(#loc444)) +#loc982 = loc("p_A22"(#loc445)) +#loc983 = loc("p_A22"(#loc446)) +#loc984 = loc("p_A22"(#loc447)) +#loc985 = loc("p_A22"(#loc448)) +#loc986 = loc("p_A30"(#loc449)) +#loc987 = loc("p_A30"(#loc450)) +#loc988 = loc("p_A30"(#loc451)) +#loc989 = loc("p_A30"(#loc452)) +#loc990 = loc("p_A30"(#loc453)) +#loc991 = loc("p_A30"(#loc454)) +#loc992 = loc("p_A31"(#loc455)) +#loc993 = loc("p_A31"(#loc456)) +#loc994 = loc("p_A31"(#loc457)) +#loc995 = loc("p_A31"(#loc458)) +#loc996 = loc("p_A31"(#loc459)) +#loc997 = loc("p_A31"(#loc460)) +#loc998 = loc("p_A31"(#loc461)) +#loc999 = loc("p_A32"(#loc462)) +#loc1000 = loc("p_A32"(#loc463)) +#loc1001 = loc("p_A32"(#loc464)) +#loc1002 = loc("p_A32"(#loc465)) +#loc1003 = loc("p_A32"(#loc466)) +#loc1004 = loc("p_A32"(#loc467)) +#loc1005 = loc("p_A32"(#loc468)) +#loc1006 = loc("p_A33"(#loc469)) +#loc1007 = loc("p_A33"(#loc470)) +#loc1008 = loc("p_A33"(#loc471)) +#loc1009 = loc("p_A33"(#loc472)) +#loc1010 = loc("p_A33"(#loc473)) +#loc1011 = loc("p_A33"(#loc474)) +#loc1012 = loc("p_A33"(#loc475)) +#loc1013 = loc("m_A0"(#loc476)) +#loc1014 = loc("m_A0"(#loc477)) +#loc1015 = loc("m_A0"(#loc478)) +#loc1016 = loc("m_A0"(#loc479)) +#loc1017 = loc("m_A1"(#loc480)) +#loc1018 = loc("m_A1"(#loc481)) +#loc1019 = loc("m_A1"(#loc482)) +#loc1020 = loc("m_A1"(#loc483)) +#loc1021 = loc("m_A2"(#loc484)) +#loc1022 = loc("m_A2"(#loc485)) +#loc1023 = loc("m_A2"(#loc486)) +#loc1024 = loc("m_A2"(#loc487)) +#loc1025 = loc("m_A3"(#loc488)) +#loc1026 = loc("m_A3"(#loc489)) +#loc1027 = loc("m_A3"(#loc490)) +#loc1028 = loc("m_A3"(#loc491)) +#loc1029 = loc("m_A11"(#loc492)) +#loc1030 = loc("m_A11"(#loc493)) +#loc1031 = loc("m_A11"(#loc494)) +#loc1032 = loc("m_A11"(#loc495)) +#loc1033 = loc("m_A11"(#loc496)) +#loc1034 = loc("m_A21"(#loc497)) +#loc1035 = loc("m_A21"(#loc498)) +#loc1036 = loc("m_A21"(#loc499)) +#loc1037 = loc("m_A21"(#loc500)) +#loc1038 = loc("m_A21"(#loc501)) +#loc1039 = loc("m_A22"(#loc502)) +#loc1040 = loc("m_A22"(#loc503)) +#loc1041 = loc("m_A22"(#loc504)) +#loc1042 = loc("m_A22"(#loc505)) +#loc1043 = loc("m_A22"(#loc506)) +#loc1044 = loc("m_A31"(#loc507)) +#loc1045 = loc("m_A31"(#loc508)) +#loc1046 = loc("m_A31"(#loc509)) +#loc1047 = loc("m_A31"(#loc510)) +#loc1048 = loc("m_A31"(#loc511)) +#loc1049 = loc("m_A32"(#loc512)) +#loc1050 = loc("m_A32"(#loc513)) +#loc1051 = loc("m_A32"(#loc514)) +#loc1052 = loc("m_A32"(#loc515)) +#loc1053 = loc("m_A32"(#loc516)) +#loc1054 = loc("m_A33"(#loc517)) +#loc1055 = loc("m_A33"(#loc518)) +#loc1056 = loc("m_A33"(#loc519)) +#loc1057 = loc("m_A33"(#loc520)) +#loc1058 = loc("m_A33"(#loc521)) +#loc1063 = loc("b_A11"(#loc654)) +#loc1064 = loc("b_A22"(#loc655)) +#loc1065 = loc("b_A33"(#loc656)) +#loc1066 = loc("b_A10"(#loc657)) +#loc1067 = loc("b_A20"(#loc658)) +#loc1068 = loc("b_A21"(#loc659)) +#loc1069 = loc("b_A30"(#loc660)) +#loc1070 = loc("b_A31"(#loc661)) +#loc1071 = loc("b_A32"(#loc662)) +#loc1072 = loc("b_A11"(#loc663)) +#loc1073 = loc("b_A11"(#loc692)) +#loc1074 = loc("b_A10"(#loc694)) +#loc1075 = loc("b_A33"(#loc695)) +#loc1076 = loc("b_A22"(#loc708)) +#loc1077 = loc("b_A20"(#loc710)) +#loc1078 = loc("b_A21"(#loc712)) +#loc1079 = loc("b_A30"(#loc713)) +#loc1080 = loc("b_A33"(#loc726)) +#loc1081 = loc("b_A30"(#loc728)) +#loc1082 = loc("b_A31"(#loc730)) +#loc1083 = loc("b_A32"(#loc732)) +#loc1084 = loc("b_A22"(#loc1064)) +#loc1085 = loc("b_A33"(#loc1065)) +#loc1086 = loc("b_A20"(#loc1067)) +#loc1087 = loc("b_A21"(#loc1068)) +#loc1088 = loc("b_A30"(#loc1069)) +#loc1089 = loc("b_A31"(#loc1070)) +#loc1090 = loc("b_A32"(#loc1071)) +#loc1091 = loc("b_A22"(#loc1072)) +#loc1092 = loc("b_A20"(#loc1075)) +#loc1093 = loc("b_A31"(#loc1079)) +#loc1094 = loc("b_A33"(#loc1085)) +#loc1095 = loc("b_A30"(#loc1088)) +#loc1096 = loc("b_A31"(#loc1089)) +#loc1097 = loc("b_A32"(#loc1090)) +#loc1098 = loc("b_A33"(#loc1091)) +#loc1099 = loc("b_A21"(#loc1092)) +#loc1100 = loc("b_A32"(#loc1093)) +#loc1101 = loc("b_A10"(#loc1098)) +#loc1102 = loc("b_A30"(#loc1099)) +#loc1103 = loc("b_A20"(#loc1101)) +#loc1104 = loc("b_A31"(#loc1102)) +#loc1105 = loc("b_A21"(#loc1103)) +#loc1106 = loc("b_A32"(#loc1104)) +#loc1107 = loc("b_A30"(#loc1105)) +#loc1108 = loc("b_A31"(#loc1107)) +#loc1109 = loc("b_A32"(#loc1108)) diff --git a/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.ttgir b/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..4e93f2202ea8970480e79ff53c5d634ed6b285e8 --- /dev/null +++ b/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.ttgir @@ -0,0 +1,1442 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [4, 8], warpsPerCTA = [2, 1], order = [1, 0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1, 4], threadsPerWarp = [8, 4], warpsPerCTA = [2, 1], order = [1, 0]}> +#linear = #ttg.linear<{register = [[0, 1], [0, 8], [0, 16], [0, 32], [8, 0]], lane = [[0, 2], [0, 4], [1, 0], [2, 0], [4, 0]], warp = [[0, 0]], block = []}> +#linear1 = #ttg.linear<{register = [[0, 1], [0, 8], [0, 16], [0, 32]], lane = [[0, 2], [0, 4], [1, 0], [2, 0], [4, 0]], warp = [[8, 0]], block = []}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":40:0) +#loc1 = loc(unknown) +#loc212 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":228:66) +#loc218 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":230:56) +#loc228 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":233:66) +#loc233 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":235:56) +#loc243 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":238:66) +#loc248 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":240:56) +#loc258 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":243:66) +#loc263 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":245:56) +#mma = #ttg.nvidia_mma<{versionMajor = 2, versionMinor = 0, warpsPerCTA = [1, 2], instrShape = [16, 8]}> +#mma1 = #ttg.nvidia_mma<{versionMajor = 2, versionMinor = 0, warpsPerCTA = [2, 1], instrShape = [16, 8]}> +#shared = #ttg.swizzled_shared<{vec = 8, perPhase = 1, maxPhase = 8, order = [1, 0]}> +#shared1 = #ttg.swizzled_shared<{vec = 8, perPhase = 2, maxPhase = 2, order = [1, 0]}> +#smem = #ttg.shared_memory +#loc356 = loc("k"(#loc)) +#loc357 = loc("g"(#loc)) +#loc358 = loc("beta"(#loc)) +#loc359 = loc("A"(#loc)) +#loc360 = loc("T"(#loc)) +#loc553 = loc("b_a00"(#loc212)) +#loc558 = loc("b_a00"(#loc218)) +#loc565 = loc("b_a11"(#loc228)) +#loc570 = loc("b_a11"(#loc233)) +#loc577 = loc("b_a22"(#loc243)) +#loc582 = loc("b_a22"(#loc248)) +#loc589 = loc("b_a33"(#loc258)) +#loc594 = loc("b_a33"(#loc263)) +#loc690 = loc(callsite(#loc1 at #loc553)) +#loc692 = loc(callsite(#loc1 at #loc558)) +#loc694 = loc(callsite(#loc1 at #loc565)) +#loc696 = loc(callsite(#loc1 at #loc570)) +#loc698 = loc(callsite(#loc1 at #loc577)) +#loc700 = loc(callsite(#loc1 at #loc582)) +#loc702 = loc(callsite(#loc1 at #loc589)) +#loc704 = loc(callsite(#loc1 at #loc594)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 2 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @chunk_gated_delta_rule_fwd_kkt_solve_kernel(%k: !tt.ptr {tt.divisibility = 16 : i32} loc("k"(#loc)), %g: !tt.ptr {tt.divisibility = 16 : i32} loc("g"(#loc)), %beta: !tt.ptr {tt.divisibility = 16 : i32} loc("beta"(#loc)), %A: !tt.ptr {tt.divisibility = 16 : i32} loc("A"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<0.000000e+00> : tensor<16x16xf32, #mma> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<16x16xf32, #mma1> loc(#loc1) + %cst_1 = arith.constant dense<0.000000e+00> : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc1) + %cst_2 = arith.constant dense<0.000000e+00> : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc1) + %c0_i32 = arith.constant 0 : i32 loc(#loc1) + %c2_i32 = arith.constant 2 : i32 loc(#loc1) + %c1_i32 = arith.constant 1 : i32 loc(#loc1) + %c2_i64 = arith.constant 2 : i64 loc(#loc1) + %c1_i64 = arith.constant 1 : i64 loc(#loc1) + %cst_3 = arith.constant dense<0.000000e+00> : tensor<16x64xbf16, #blocked> loc(#loc1) + %c32_i64 = arith.constant 32 : i64 loc(#loc1) + %c64_i32 = arith.constant 64 : i32 loc(#loc1) + %c64_i64 = arith.constant 64 : i64 loc(#loc1) + %c16_i64 = arith.constant 16 : i64 loc(#loc1) + %c48_i64 = arith.constant 48 : i64 loc(#loc1) + %c128_i64 = arith.constant 128 : i64 loc(#loc1) + %cst_4 = arith.constant dense<4096> : tensor<16x1xi64, #blocked> loc(#loc1) + %cst_5 = arith.constant dense<128> : tensor<1x64xi32, #blocked> loc(#loc1) + %cst_6 = arith.constant dense<2048> : tensor<16x1xi64, #blocked1> loc(#loc1) + %cst_7 = arith.constant dense<16> : tensor<16xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc1) + %cst_8 = arith.constant dense<32> : tensor<16xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc1) + %cst_9 = arith.constant dense<48> : tensor<16xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc1) + %cst_10 = arith.constant dense<64> : tensor<1x16xi32, #blocked1> loc(#loc1) + %cst_11 = arith.constant dense<0.000000e+00> : tensor<16xbf16, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc1) + %cst_12 = arith.constant dense<0.000000e+00> : tensor<16xbf16, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc1) + %cst_13 = arith.constant dense<0.000000e+00> : tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc1) + %cst_14 = arith.constant dense<0.000000e+00> : tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc1) + %cst_15 = arith.constant dense<32> : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc1) + %cst_16 = arith.constant dense<32> : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc1) + %cst_17 = arith.constant dense<32> : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc1) + %cst_18 = arith.constant dense<32> : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc1) + %c-1_i32 = arith.constant -1 : i32 loc(#loc1) + %cst_19 = arith.constant dense<64> : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc1) + %0 = tt.get_program_id x : i32 loc(#loc2) + %1 = arith.extsi %0 : i32 to i64 loc(#loc3) + %2 = tt.get_program_id y : i32 loc(#loc4) + %3 = arith.extsi %2 : i32 to i64 loc(#loc5) + %4 = arith.divsi %3, %c32_i64 : i64 loc(#loc6) + %5 = arith.remsi %3, %c32_i64 : i64 loc(#loc7) + %6 = arith.extsi %T : i32 to i64 loc(#loc8) + %7 = arith.muli %4, %6 : i64 loc(#loc8) + %8 = arith.muli %1, %c64_i64 : i64 loc(#loc9) + %9 = arith.cmpi sge, %8, %6 : i64 loc(#loc10) + cf.cond_br %9, ^bb1, ^bb2 loc(#loc10) + ^bb1: // pred: ^bb0 + tt.return loc(#loc11) + ^bb2: // pred: ^bb0 + %i_tc1 = arith.addi %8, %c16_i64 : i64 loc(#loc361) + %i_tc2 = arith.addi %8, %c32_i64 : i64 loc(#loc362) + %i_tc3 = arith.addi %8, %c48_i64 : i64 loc(#loc363) + %k_20 = arith.muli %7, %c32_i64 : i64 loc(#loc364) + %k_21 = arith.addi %k_20, %5 : i64 loc(#loc365) + %k_22 = arith.muli %k_21, %c128_i64 : i64 loc(#loc366) + %k_23 = tt.addptr %k, %k_22 : !tt.ptr, i64 loc(#loc367) + %A_24 = arith.muli %k_21, %c64_i64 : i64 loc(#loc368) + %A_25 = tt.addptr %A, %A_24 : !tt.ptr, i64 loc(#loc369) + %o_i = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc370) + %o_i_26 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc370) + %o_i_27 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc370) + %o_i_28 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc370) + %o_i_29 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc370) + %o_i_30 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc370) + %o_i_31 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc370) + %m_tc0 = arith.extsi %o_i : tensor<16xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc371) + %m_tc0_32 = arith.extsi %o_i_26 : tensor<16xi32, #ttg.slice<{dim = 1, parent = #mma}>> to tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc371) + %m_tc0_33 = arith.extsi %o_i_27 : tensor<16xi32, #ttg.slice<{dim = 1, parent = #mma1}>> to tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc371) + %m_tc0_34 = arith.extsi %o_i_28 : tensor<16xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> to tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc371) + %m_tc0_35 = arith.extsi %o_i_29 : tensor<16xi32, #ttg.slice<{dim = 0, parent = #mma}>> to tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc371) + %m_tc0_36 = arith.extsi %o_i_30 : tensor<16xi32, #ttg.slice<{dim = 0, parent = #mma1}>> to tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc371) + %m_tc0_37 = tt.splat %8 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc371) + %m_tc0_38 = tt.splat %8 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc371) + %m_tc0_39 = tt.splat %8 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc371) + %m_tc0_40 = tt.splat %8 : i64 -> tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc371) + %m_tc0_41 = tt.splat %8 : i64 -> tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc371) + %m_tc0_42 = arith.addi %m_tc0_37, %m_tc0 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc371) + %m_tc0_43 = arith.addi %m_tc0_38, %m_tc0_32 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc371) + %m_tc0_44 = arith.addi %m_tc0_39, %m_tc0_34 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc371) + %m_tc0_45 = arith.addi %m_tc0_40, %m_tc0_35 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc371) + %m_tc0_46 = arith.addi %m_tc0_41, %m_tc0_36 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc371) + %m_tc0_47 = tt.splat %6 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc372) + %m_tc0_48 = tt.splat %6 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc372) + %m_tc0_49 = tt.splat %6 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc372) + %m_tc0_50 = tt.splat %6 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc372) + %m_tc0_51 = tt.splat %6 : i64 -> tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc372) + %m_tc0_52 = tt.splat %6 : i64 -> tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc372) + %m_tc0_53 = arith.cmpi slt, %m_tc0_42, %m_tc0_47 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc372) + %m_tc0_54 = arith.cmpi slt, %m_tc0_43, %m_tc0_48 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc372) + %m_tc0_55 = arith.cmpi slt, %m_tc0_44, %m_tc0_50 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc372) + %m_tc0_56 = arith.cmpi slt, %m_tc0_45, %m_tc0_51 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc372) + %m_tc0_57 = arith.cmpi slt, %m_tc0_46, %m_tc0_52 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc372) + %m_tc1 = tt.splat %i_tc1 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc373) + %m_tc1_58 = tt.splat %i_tc1 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc373) + %m_tc1_59 = tt.splat %i_tc1 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc373) + %m_tc1_60 = tt.splat %i_tc1 : i64 -> tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc373) + %m_tc1_61 = arith.addi %m_tc1, %m_tc0 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc373) + %m_tc1_62 = arith.addi %m_tc1_58, %m_tc0_33 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc373) + %m_tc1_63 = arith.addi %m_tc1_59, %m_tc0_34 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc373) + %m_tc1_64 = arith.addi %m_tc1_60, %m_tc0_36 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc373) + %m_tc1_65 = arith.cmpi slt, %m_tc1_61, %m_tc0_47 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc374) + %m_tc1_66 = arith.cmpi slt, %m_tc1_62, %m_tc0_49 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc374) + %m_tc1_67 = arith.cmpi slt, %m_tc1_63, %m_tc0_50 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc374) + %m_tc1_68 = arith.cmpi slt, %m_tc1_64, %m_tc0_52 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc374) + %m_tc2 = tt.splat %i_tc2 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc375) + %m_tc2_69 = tt.splat %i_tc2 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc375) + %m_tc2_70 = tt.splat %i_tc2 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc375) + %m_tc2_71 = tt.splat %i_tc2 : i64 -> tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc375) + %m_tc2_72 = arith.addi %m_tc2, %m_tc0 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc375) + %m_tc2_73 = arith.addi %m_tc2_69, %m_tc0_33 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc375) + %m_tc2_74 = arith.addi %m_tc2_70, %m_tc0_34 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc375) + %m_tc2_75 = arith.addi %m_tc2_71, %m_tc0_36 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc375) + %m_tc2_76 = arith.cmpi slt, %m_tc2_72, %m_tc0_47 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc376) + %m_tc2_77 = arith.cmpi slt, %m_tc2_73, %m_tc0_49 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc376) + %m_tc2_78 = arith.cmpi slt, %m_tc2_74, %m_tc0_50 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc376) + %m_tc2_79 = arith.cmpi slt, %m_tc2_75, %m_tc0_52 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc376) + %m_tc3 = tt.splat %i_tc3 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc377) + %m_tc3_80 = tt.splat %i_tc3 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc377) + %m_tc3_81 = tt.splat %i_tc3 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc377) + %m_tc3_82 = tt.splat %i_tc3 : i64 -> tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc377) + %m_tc3_83 = arith.addi %m_tc3, %m_tc0 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc377) + %m_tc3_84 = arith.addi %m_tc3_80, %m_tc0_33 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc377) + %m_tc3_85 = arith.addi %m_tc3_81, %m_tc0_34 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc377) + %m_tc3_86 = arith.addi %m_tc3_82, %m_tc0_36 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc377) + %m_tc3_87 = arith.cmpi slt, %m_tc3_83, %m_tc0_47 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc378) + %m_tc3_88 = arith.cmpi slt, %m_tc3_84, %m_tc0_49 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc378) + %m_tc3_89 = arith.cmpi slt, %m_tc3_85, %m_tc0_50 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc378) + %m_tc3_90 = arith.cmpi slt, %m_tc3_86, %m_tc0_52 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc378) + %p_b0 = tt.addptr %beta, %k_21 : !tt.ptr, i64 loc(#loc665) + %p_b0_91 = arith.muli %m_tc0_43, %cst_17 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc381) + %p_b0_92 = arith.muli %m_tc0_45, %cst_15 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc381) + %p_b0_93 = arith.muli %m_tc0_46, %cst_16 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc381) + %p_b0_94 = tt.splat %p_b0 : !tt.ptr -> tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc382) + %p_b0_95 = tt.splat %p_b0 : !tt.ptr -> tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc382) + %p_b0_96 = tt.addptr %p_b0_94, %p_b0_91 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma}>>, tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc382) + %p_b1 = arith.muli %m_tc1_62, %cst_18 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc383) + %p_b1_97 = arith.muli %m_tc1_64, %cst_16 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc383) + %p_b1_98 = tt.addptr %p_b0_95, %p_b1 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma1}>>, tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc384) + %p_b2 = arith.muli %m_tc2_73, %cst_18 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc385) + %p_b2_99 = arith.muli %m_tc2_75, %cst_16 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc385) + %p_b2_100 = tt.addptr %p_b0_95, %p_b2 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma1}>>, tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc386) + %p_b3 = arith.muli %m_tc3_84, %cst_18 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc387) + %p_b3_101 = arith.muli %m_tc3_86, %cst_16 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc387) + %p_b3_102 = tt.addptr %p_b0_95, %p_b3 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma1}>>, tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc388) + %b_b0 = tt.load %p_b0_96, %m_tc0_54, %cst_11 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc389) + %b_b0_103 = arith.extf %b_b0 : tensor<16xbf16, #ttg.slice<{dim = 1, parent = #mma}>> to tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc390) + %b_b1 = tt.load %p_b1_98, %m_tc1_66, %cst_12 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc391) + %b_b1_104 = arith.extf %b_b1 : tensor<16xbf16, #ttg.slice<{dim = 1, parent = #mma1}>> to tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc392) + %b_b2 = tt.load %p_b2_100, %m_tc2_77, %cst_12 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc393) + %b_b2_105 = arith.extf %b_b2 : tensor<16xbf16, #ttg.slice<{dim = 1, parent = #mma1}>> to tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc394) + %b_b3 = tt.load %p_b3_102, %m_tc3_88, %cst_12 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc395) + %b_b3_106 = arith.extf %b_b3 : tensor<16xbf16, #ttg.slice<{dim = 1, parent = #mma1}>> to tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc396) + %p_g0 = tt.addptr %g, %k_21 : !tt.ptr, i64 loc(#loc666) + %p_g0_107 = tt.splat %p_g0 : !tt.ptr -> tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc399) + %p_g0_108 = tt.splat %p_g0 : !tt.ptr -> tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc399) + %p_g0_109 = tt.splat %p_g0 : !tt.ptr -> tensor<16x!tt.ptr, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc399) + %p_g0_110 = tt.splat %p_g0 : !tt.ptr -> tensor<16x!tt.ptr, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc399) + %p_g0_111 = tt.addptr %p_g0_107, %p_b0_91 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma}>>, tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc399) + %p_g0_112 = tt.addptr %p_g0_109, %p_b0_92 : tensor<16x!tt.ptr, #ttg.slice<{dim = 0, parent = #mma}>>, tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc399) + %p_g0_113 = tt.addptr %p_g0_110, %p_b0_93 : tensor<16x!tt.ptr, #ttg.slice<{dim = 0, parent = #mma1}>>, tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc399) + %p_g1 = tt.addptr %p_g0_108, %p_b1 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma1}>>, tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc400) + %p_g1_114 = tt.addptr %p_g0_110, %p_b1_97 : tensor<16x!tt.ptr, #ttg.slice<{dim = 0, parent = #mma1}>>, tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc400) + %p_g2 = tt.addptr %p_g0_108, %p_b2 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma1}>>, tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc401) + %p_g2_115 = tt.addptr %p_g0_110, %p_b2_99 : tensor<16x!tt.ptr, #ttg.slice<{dim = 0, parent = #mma1}>>, tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc401) + %p_g3 = tt.addptr %p_g0_108, %p_b3 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma1}>>, tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc402) + %p_g3_116 = tt.addptr %p_g0_110, %p_b3_101 : tensor<16x!tt.ptr, #ttg.slice<{dim = 0, parent = #mma1}>>, tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc402) + %b_g0 = tt.load %p_g0_111, %m_tc0_54, %cst_13 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc403) + %b_g0_117 = tt.load %p_g0_112, %m_tc0_56, %cst_1 : tensor<16x!tt.ptr, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc403) + %b_g0_118 = tt.load %p_g0_113, %m_tc0_57, %cst_2 : tensor<16x!tt.ptr, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc403) + %b_g1 = tt.load %p_g1, %m_tc1_66, %cst_14 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc404) + %b_g1_119 = tt.load %p_g1_114, %m_tc1_68, %cst_2 : tensor<16x!tt.ptr, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc404) + %b_g2 = tt.load %p_g2, %m_tc2_77, %cst_14 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc405) + %b_g2_120 = tt.load %p_g2_115, %m_tc2_79, %cst_2 : tensor<16x!tt.ptr, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc405) + %b_g3 = tt.load %p_g3, %m_tc3_88, %cst_14 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc406) + %b_g3_121 = tt.load %p_g3_116, %m_tc3_90, %cst_2 : tensor<16x!tt.ptr, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc406) + %o_k = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc407) + %p_k0 = tt.expand_dims %m_tc0_42 {axis = 1 : i32} : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi64, #blocked> loc(#loc408) + %p_k0_122 = arith.muli %p_k0, %cst_4 : tensor<16x1xi64, #blocked> loc(#loc409) + %p_k0_123 = tt.splat %k_23 : !tt.ptr -> tensor<16x1x!tt.ptr, #blocked> loc(#loc410) + %p_k0_124 = tt.addptr %p_k0_123, %p_k0_122 : tensor<16x1x!tt.ptr, #blocked>, tensor<16x1xi64, #blocked> loc(#loc410) + %p_k0_125 = tt.broadcast %p_k0_124 : tensor<16x1x!tt.ptr, #blocked> -> tensor<16x64x!tt.ptr, #blocked> loc(#loc411) + %b_k0 = tt.expand_dims %m_tc0_53 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi1, #blocked> loc(#loc412) + %b_k0_126 = tt.broadcast %b_k0 : tensor<16x1xi1, #blocked> -> tensor<16x64xi1, #blocked> loc(#loc413) + %10 = arith.cmpi slt, %i_tc1, %6 : i64 loc(#loc65) + %b_k0_127 = ttg.local_alloc : () -> !ttg.memdesc<2x16x64xbf16, #shared, #smem, mutable> loc(#loc414) + %p_k0_128 = tt.expand_dims %o_k {axis = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x64xi32, #blocked> loc(#loc415) + %p_k0_129 = tt.broadcast %p_k0_128 : tensor<1x64xi32, #blocked> -> tensor<16x64xi32, #blocked> loc(#loc411) + %p_k0_130 = tt.addptr %p_k0_125, %p_k0_129 : tensor<16x64x!tt.ptr, #blocked>, tensor<16x64xi32, #blocked> loc(#loc411) + %b_k0_131 = arith.cmpi slt, %p_k0_128, %cst_5 : tensor<1x64xi32, #blocked> loc(#loc416) + %b_k0_132 = tt.broadcast %b_k0_131 : tensor<1x64xi1, #blocked> -> tensor<16x64xi1, #blocked> loc(#loc413) + %b_k0_133 = arith.andi %b_k0_126, %b_k0_132 : tensor<16x64xi1, #blocked> loc(#loc413) + %b_k0_134 = ttg.memdesc_index %b_k0_127[%c0_i32] : !ttg.memdesc<2x16x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<16x64xbf16, #shared, #smem, mutable> loc(#loc414) + %b_k0_135 = ttg.async_copy_global_to_local %p_k0_130, %b_k0_134 mask %b_k0_133 other %cst_3 {contiguity = 8 : i32} : tensor<16x64x!tt.ptr, #blocked> -> <16x64xbf16, #shared, #smem, mutable> loc(#loc414) + %b_k0_136 = ttg.async_commit_group tokens %b_k0_135 loc(#loc414) + %o_k_137 = arith.addi %o_k, %cst_19 : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc417) + %p_k0_138 = tt.expand_dims %o_k_137 {axis = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x64xi32, #blocked> loc(#loc415) + %p_k0_139 = tt.broadcast %p_k0_138 : tensor<1x64xi32, #blocked> -> tensor<16x64xi32, #blocked> loc(#loc411) + %p_k0_140 = tt.addptr %p_k0_125, %p_k0_139 : tensor<16x64x!tt.ptr, #blocked>, tensor<16x64xi32, #blocked> loc(#loc411) + %b_k0_141 = arith.cmpi slt, %p_k0_138, %cst_5 : tensor<1x64xi32, #blocked> loc(#loc416) + %b_k0_142 = tt.broadcast %b_k0_141 : tensor<1x64xi1, #blocked> -> tensor<16x64xi1, #blocked> loc(#loc413) + %b_k0_143 = arith.andi %b_k0_126, %b_k0_142 : tensor<16x64xi1, #blocked> loc(#loc413) + %b_k0_144 = ttg.memdesc_index %b_k0_127[%c1_i32] : !ttg.memdesc<2x16x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<16x64xbf16, #shared, #smem, mutable> loc(#loc414) + %b_k0_145 = ttg.async_copy_global_to_local %p_k0_140, %b_k0_144 mask %b_k0_143 other %cst_3 {contiguity = 8 : i32} : tensor<16x64x!tt.ptr, #blocked> -> <16x64xbf16, #shared, #smem, mutable> loc(#loc414) + %b_k0_146 = ttg.async_commit_group tokens %b_k0_145 loc(#loc414) + %b_A32:18 = scf.for %b_A32_355 = %c0_i32 to %c2_i32 step %c1_i32 iter_args(%arg6 = %cst, %arg7 = %cst_0, %arg8 = %cst_0, %arg9 = %cst_0, %arg10 = %cst_0, %arg11 = %cst_0, %arg12 = %cst_0, %arg13 = %cst_0, %arg14 = %cst_0, %arg15 = %cst_0, %arg16 = %c1_i32, %arg17 = %c-1_i32, %b_k0_356 = %b_k0_136, %b_k0_357 = %b_k0_146, %p_k0_358 = %p_k0_129, %p_k0_359 = %p_k0_139, %b_k0_360 = %b_k0_132, %b_k0_361 = %b_k0_142) -> (tensor<16x16xf32, #mma>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, i32, i32, !ttg.async.token, !ttg.async.token, tensor<16x64xi32, #blocked>, tensor<16x64xi32, #blocked>, tensor<16x64xi1, #blocked>, tensor<16x64xi1, #blocked>) : i32 { + %b_A32_362 = arith.cmpi slt, %b_A32_355, %c0_i32 : i32 loc(#loc727) + %b_A32_363 = arith.addi %arg17, %c1_i32 : i32 loc(#loc727) + %b_A32_364 = arith.cmpi sge, %b_A32_363, %c2_i32 : i32 loc(#loc727) + %b_A32_365 = arith.select %b_A32_364, %c0_i32, %b_A32_363 : i32 loc(#loc727) + %b_k0_366 = ttg.async_wait %b_k0_356 {num = 1 : i32} loc(#loc414) + %b_k0_367 = ttg.memdesc_index %b_k0_127[%b_A32_365] : !ttg.memdesc<2x16x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<16x64xbf16, #shared, #smem, mutable> loc(#loc414) + %b_A00_368 = ttg.local_load %b_k0_367 token %b_k0_366 : !ttg.memdesc<16x64xbf16, #shared, #smem, mutable> -> tensor<16x64xbf16, #linear> loc(#loc419) + %b_A00_369 = tt.trans %b_A00_368 {order = array} : tensor<16x64xbf16, #linear> -> tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> loc(#loc420) + %b_A00_370 = ttg.local_load %b_k0_367 token %b_k0_366 : !ttg.memdesc<16x64xbf16, #shared, #smem, mutable> -> tensor<16x64xbf16, #linear1> loc(#loc419) + %b_A00_371 = tt.trans %b_A00_370 {order = array} : tensor<16x64xbf16, #linear1> -> tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> loc(#loc420) + %b_k0_372 = ttg.local_load %b_k0_367 token %b_k0_366 : !ttg.memdesc<16x64xbf16, #shared, #smem, mutable> -> tensor<16x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma, kWidth = 2}>> loc(#loc414) + %b_A00_373 = tt.dot %b_k0_372, %b_A00_371, %arg6, inputPrecision = tf32 : tensor<16x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma, kWidth = 2}>> * tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> -> tensor<16x16xf32, #mma> loc(#loc419) + %b_A32_374 = arith.addi %arg16, %c1_i32 : i32 loc(#loc727) + %b_A32_375 = arith.cmpi sge, %b_A32_374, %c2_i32 : i32 loc(#loc727) + %b_A32_376 = arith.select %b_A32_375, %c0_i32, %b_A32_374 : i32 loc(#loc727) + %b_A32_377 = arith.addi %b_A32_355, %c2_i32 : i32 loc(#loc727) + %o_k_378 = arith.muli %b_A32_377, %c64_i32 : i32 loc(#loc421) + %o_k_379 = tt.splat %o_k_378 : i32 -> tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc417) + %o_k_380 = arith.addi %o_k_379, %o_k : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc417) + %p_k0_381 = tt.expand_dims %o_k_380 {axis = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x64xi32, #blocked> loc(#loc415) + %p_k0_382 = tt.broadcast %p_k0_381 : tensor<1x64xi32, #blocked> -> tensor<16x64xi32, #blocked> loc(#loc411) + %p_k0_383 = tt.addptr %p_k0_125, %p_k0_382 : tensor<16x64x!tt.ptr, #blocked>, tensor<16x64xi32, #blocked> loc(#loc411) + %b_k0_384 = arith.cmpi slt, %p_k0_381, %cst_5 : tensor<1x64xi32, #blocked> loc(#loc416) + %b_k0_385 = tt.broadcast %b_k0_384 : tensor<1x64xi1, #blocked> -> tensor<16x64xi1, #blocked> loc(#loc413) + %b_k0_386 = arith.andi %b_k0_126, %b_k0_385 : tensor<16x64xi1, #blocked> loc(#loc413) + %b_k0_387 = ttg.memdesc_index %b_k0_127[%b_A32_376] : !ttg.memdesc<2x16x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<16x64xbf16, #shared, #smem, mutable> loc(#loc414) + %b_A32_388 = tt.splat %b_A32_362 : i1 -> tensor<16x64xi1, #blocked> loc(#loc727) + %b_A32_389 = arith.andi %b_A32_388, %b_k0_386 : tensor<16x64xi1, #blocked> loc(#loc727) + %b_k0_390 = ttg.async_copy_global_to_local %p_k0_383, %b_k0_387 mask %b_A32_389 other %cst_3 {contiguity = 8 : i32} : tensor<16x64x!tt.ptr, #blocked> -> <16x64xbf16, #shared, #smem, mutable> loc(#loc414) + %b_k0_391 = ttg.async_commit_group tokens %b_k0_390 loc(#loc414) + %39:9 = scf.if %10 -> (tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>) { + %p_k1 = tt.expand_dims %m_tc1_61 {axis = 1 : i32} : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi64, #blocked> loc(#loc422) + %p_k1_392 = arith.muli %p_k1, %cst_4 : tensor<16x1xi64, #blocked> loc(#loc423) + %p_k1_393 = tt.addptr %p_k0_123, %p_k1_392 : tensor<16x1x!tt.ptr, #blocked>, tensor<16x1xi64, #blocked> loc(#loc424) + %p_k1_394 = tt.broadcast %p_k1_393 : tensor<16x1x!tt.ptr, #blocked> -> tensor<16x64x!tt.ptr, #blocked> loc(#loc425) + %p_k1_395 = tt.addptr %p_k1_394, %p_k0_358 : tensor<16x64x!tt.ptr, #blocked>, tensor<16x64xi32, #blocked> loc(#loc425) + %b_k1 = tt.expand_dims %m_tc1_65 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi1, #blocked> loc(#loc426) + %b_k1_396 = tt.broadcast %b_k1 : tensor<16x1xi1, #blocked> -> tensor<16x64xi1, #blocked> loc(#loc427) + %b_k1_397 = arith.andi %b_k1_396, %b_k0_360 : tensor<16x64xi1, #blocked> loc(#loc427) + %b_k1_398 = tt.load %p_k1_395, %b_k1_397, %cst_3 : tensor<16x64x!tt.ptr, #blocked> loc(#loc428) + %b_k1_399 = ttg.local_alloc %b_k1_398 : (tensor<16x64xbf16, #blocked>) -> !ttg.memdesc<16x64xbf16, #shared, #smem> loc(#loc428) + %b_A11_400 = ttg.convert_layout %b_k1_398 : tensor<16x64xbf16, #blocked> -> tensor<16x64xbf16, #linear> loc(#loc429) + %b_A11_401 = tt.trans %b_A11_400 {order = array} : tensor<16x64xbf16, #linear> -> tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> loc(#loc429) + %b_k1_402 = ttg.local_load %b_k1_399 : !ttg.memdesc<16x64xbf16, #shared, #smem> -> tensor<16x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> loc(#loc428) + %b_A11_403 = tt.dot %b_k1_402, %b_A11_401, %arg7, inputPrecision = tf32 : tensor<16x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc668) + %b_A10_404 = tt.dot %b_k1_402, %b_A00_369, %arg10, inputPrecision = tf32 : tensor<16x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc669) + %40 = arith.cmpi slt, %i_tc2, %6 : i64 loc(#loc85) + %b_A32_405:7 = scf.if %40 -> (tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>) { + %p_k2 = tt.expand_dims %m_tc2_72 {axis = 1 : i32} : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi64, #blocked> loc(#loc433) + %p_k2_406 = arith.muli %p_k2, %cst_4 : tensor<16x1xi64, #blocked> loc(#loc434) + %p_k2_407 = tt.addptr %p_k0_123, %p_k2_406 : tensor<16x1x!tt.ptr, #blocked>, tensor<16x1xi64, #blocked> loc(#loc435) + %p_k2_408 = tt.broadcast %p_k2_407 : tensor<16x1x!tt.ptr, #blocked> -> tensor<16x64x!tt.ptr, #blocked> loc(#loc436) + %p_k2_409 = tt.addptr %p_k2_408, %p_k0_358 : tensor<16x64x!tt.ptr, #blocked>, tensor<16x64xi32, #blocked> loc(#loc436) + %b_k2 = tt.expand_dims %m_tc2_76 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi1, #blocked> loc(#loc437) + %b_k2_410 = tt.broadcast %b_k2 : tensor<16x1xi1, #blocked> -> tensor<16x64xi1, #blocked> loc(#loc438) + %b_k2_411 = arith.andi %b_k2_410, %b_k0_360 : tensor<16x64xi1, #blocked> loc(#loc438) + %b_k2_412 = tt.load %p_k2_409, %b_k2_411, %cst_3 : tensor<16x64x!tt.ptr, #blocked> loc(#loc439) + %b_k2_413 = ttg.local_alloc %b_k2_412 : (tensor<16x64xbf16, #blocked>) -> !ttg.memdesc<16x64xbf16, #shared, #smem> loc(#loc439) + %b_A22_414 = ttg.convert_layout %b_k2_412 : tensor<16x64xbf16, #blocked> -> tensor<16x64xbf16, #linear> loc(#loc440) + %b_A22_415 = tt.trans %b_A22_414 {order = array} : tensor<16x64xbf16, #linear> -> tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> loc(#loc440) + %b_k2_416 = ttg.local_load %b_k2_413 : !ttg.memdesc<16x64xbf16, #shared, #smem> -> tensor<16x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> loc(#loc439) + %b_A22_417 = tt.dot %b_k2_416, %b_A22_415, %arg8, inputPrecision = tf32 : tensor<16x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc671) + %b_A20_418 = tt.dot %b_k2_416, %b_A00_369, %arg11, inputPrecision = tf32 : tensor<16x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc672) + %b_A21_419 = tt.dot %b_k2_416, %b_A11_401, %arg12, inputPrecision = tf32 : tensor<16x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc673) + %41 = arith.cmpi slt, %i_tc3, %6 : i64 loc(#loc98) + %b_A32_420:4 = scf.if %41 -> (tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>) { + %p_k3 = tt.expand_dims %m_tc3_83 {axis = 1 : i32} : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi64, #blocked> loc(#loc445) + %p_k3_421 = arith.muli %p_k3, %cst_4 : tensor<16x1xi64, #blocked> loc(#loc446) + %p_k3_422 = tt.addptr %p_k0_123, %p_k3_421 : tensor<16x1x!tt.ptr, #blocked>, tensor<16x1xi64, #blocked> loc(#loc447) + %p_k3_423 = tt.broadcast %p_k3_422 : tensor<16x1x!tt.ptr, #blocked> -> tensor<16x64x!tt.ptr, #blocked> loc(#loc448) + %p_k3_424 = tt.addptr %p_k3_423, %p_k0_358 : tensor<16x64x!tt.ptr, #blocked>, tensor<16x64xi32, #blocked> loc(#loc448) + %b_k3 = tt.expand_dims %m_tc3_87 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi1, #blocked> loc(#loc449) + %b_k3_425 = tt.broadcast %b_k3 : tensor<16x1xi1, #blocked> -> tensor<16x64xi1, #blocked> loc(#loc450) + %b_k3_426 = arith.andi %b_k3_425, %b_k0_360 : tensor<16x64xi1, #blocked> loc(#loc450) + %b_k3_427 = tt.load %p_k3_424, %b_k3_426, %cst_3 : tensor<16x64x!tt.ptr, #blocked> loc(#loc451) + %b_k3_428 = ttg.local_alloc %b_k3_427 : (tensor<16x64xbf16, #blocked>) -> !ttg.memdesc<16x64xbf16, #shared, #smem> loc(#loc451) + %b_A33_429 = ttg.convert_layout %b_k3_427 : tensor<16x64xbf16, #blocked> -> tensor<16x64xbf16, #linear> loc(#loc452) + %b_A33_430 = tt.trans %b_A33_429 {order = array} : tensor<16x64xbf16, #linear> -> tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> loc(#loc452) + %b_k3_431 = ttg.local_load %b_k3_428 : !ttg.memdesc<16x64xbf16, #shared, #smem> -> tensor<16x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> loc(#loc451) + %b_A33_432 = tt.dot %b_k3_431, %b_A33_430, %arg9, inputPrecision = tf32 : tensor<16x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc675) + %b_A30_433 = tt.dot %b_k3_431, %b_A00_369, %arg13, inputPrecision = tf32 : tensor<16x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc676) + %b_A31_434 = tt.dot %b_k3_431, %b_A11_401, %arg14, inputPrecision = tf32 : tensor<16x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc677) + %b_A32_435 = tt.dot %b_k3_431, %b_A22_415, %arg15, inputPrecision = tf32 : tensor<16x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc678) + scf.yield %b_A33_432, %b_A30_433, %b_A31_434, %b_A32_435 : tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1> loc(#loc678) + } else { + scf.yield %arg9, %arg13, %arg14, %arg15 : tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1> loc(#loc99) + } loc(#loc718) + scf.yield %b_A22_417, %b_A32_420#0, %b_A20_418, %b_A21_419, %b_A32_420#1, %b_A32_420#2, %b_A32_420#3 : tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1> loc(#loc718) + } else { + scf.yield %arg8, %arg9, %arg11, %arg12, %arg13, %arg14, %arg15 : tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1> loc(#loc86) + } loc(#loc724) + scf.yield %b_A11_403, %b_A32_405#0, %b_A32_405#1, %b_A10_404, %b_A32_405#2, %b_A32_405#3, %b_A32_405#4, %b_A32_405#5, %b_A32_405#6 : tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1> loc(#loc724) + } else { + scf.yield %arg7, %arg8, %arg9, %arg10, %arg11, %arg12, %arg13, %arg14, %arg15 : tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1> loc(#loc74) + } loc(#loc74) + scf.yield %b_A00_373, %39#0, %39#1, %39#2, %39#3, %39#4, %39#5, %39#6, %39#7, %39#8, %b_A32_376, %b_A32_365, %b_k0_357, %b_k0_391, %p_k0_359, %p_k0_382, %b_k0_361, %b_k0_385 : tensor<16x16xf32, #mma>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, i32, i32, !ttg.async.token, !ttg.async.token, tensor<16x64xi32, #blocked>, tensor<16x64xi32, #blocked>, tensor<16x64xi1, #blocked>, tensor<16x64xi1, #blocked> loc(#loc727) + } loc(#loc727) + %b_A32_147 = ttg.async_wait {num = 0 : i32} loc(#loc727) + ttg.local_dealloc %b_k0_127 : !ttg.memdesc<2x16x64xbf16, #shared, #smem, mutable> loc(#loc727) + %m_d = tt.expand_dims %o_i_26 {axis = 1 : i32} : tensor<16xi32, #ttg.slice<{dim = 1, parent = #mma}>> -> tensor<16x1xi32, #mma> loc(#loc457) + %m_d_148 = tt.expand_dims %o_i_27 {axis = 1 : i32} : tensor<16xi32, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xi32, #mma1> loc(#loc457) + %m_d_149 = tt.expand_dims %o_i_29 {axis = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 0, parent = #mma}>> -> tensor<1x16xi32, #mma> loc(#loc458) + %m_d_150 = tt.expand_dims %o_i_30 {axis = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<1x16xi32, #mma1> loc(#loc458) + %m_d_151 = tt.expand_dims %o_i_31 {axis = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x16xi32, #blocked1> loc(#loc458) + %m_d_152 = tt.broadcast %m_d : tensor<16x1xi32, #mma> -> tensor<16x16xi32, #mma> loc(#loc459) + %m_d_153 = tt.broadcast %m_d_148 : tensor<16x1xi32, #mma1> -> tensor<16x16xi32, #mma1> loc(#loc459) + %m_d_154 = tt.broadcast %m_d_149 : tensor<1x16xi32, #mma> -> tensor<16x16xi32, #mma> loc(#loc459) + %m_d_155 = tt.broadcast %m_d_150 : tensor<1x16xi32, #mma1> -> tensor<16x16xi32, #mma1> loc(#loc459) + %m_d_156 = tt.broadcast %m_d_151 : tensor<1x16xi32, #blocked1> -> tensor<16x16xi32, #blocked1> loc(#loc459) + %m_d_157 = arith.cmpi sgt, %m_d_152, %m_d_154 : tensor<16x16xi32, #mma> loc(#loc459) + %m_d_158 = arith.cmpi sgt, %m_d_153, %m_d_155 : tensor<16x16xi32, #mma1> loc(#loc459) + %m_I = arith.cmpi eq, %m_d_153, %m_d_155 : tensor<16x16xi32, #mma1> loc(#loc460) + %m_I_159 = arith.cmpi eq, %m_d_152, %m_d_154 : tensor<16x16xi32, #mma> loc(#loc460) + %b_A00 = tt.expand_dims %m_tc0_54 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #mma}>> -> tensor<16x1xi1, #mma> loc(#loc461) + %b_A00_160 = tt.expand_dims %m_tc0_55 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<16x1xi1, #blocked1> loc(#loc461) + %b_A00_161 = tt.broadcast %b_A00 : tensor<16x1xi1, #mma> -> tensor<16x16xi1, #mma> loc(#loc462) + %b_A00_162 = tt.broadcast %b_A00_160 : tensor<16x1xi1, #blocked1> -> tensor<16x16xi1, #blocked1> loc(#loc462) + %b_A00_163 = arith.andi %m_d_157, %b_A00_161 : tensor<16x16xi1, #mma> loc(#loc462) + %b_A00_164 = tt.expand_dims %m_tc0_56 {axis = 0 : i32} : tensor<16xi1, #ttg.slice<{dim = 0, parent = #mma}>> -> tensor<1x16xi1, #mma> loc(#loc463) + %b_A00_165 = tt.expand_dims %m_tc0_57 {axis = 0 : i32} : tensor<16xi1, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<1x16xi1, #mma1> loc(#loc463) + %b_A00_166 = tt.broadcast %b_A00_164 : tensor<1x16xi1, #mma> -> tensor<16x16xi1, #mma> loc(#loc464) + %b_A00_167 = tt.broadcast %b_A00_165 : tensor<1x16xi1, #mma1> -> tensor<16x16xi1, #mma1> loc(#loc464) + %b_A00_168 = arith.andi %b_A00_163, %b_A00_166 : tensor<16x16xi1, #mma> loc(#loc464) + %b_A00_169 = tt.expand_dims %b_g0 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma}>> -> tensor<16x1xf32, #mma> loc(#loc465) + %b_A00_170 = tt.expand_dims %b_g0_117 {axis = 0 : i32} : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma}>> -> tensor<1x16xf32, #mma> loc(#loc466) + %b_A00_171 = tt.expand_dims %b_g0_118 {axis = 0 : i32} : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<1x16xf32, #mma1> loc(#loc466) + %b_A00_172 = tt.broadcast %b_A00_169 : tensor<16x1xf32, #mma> -> tensor<16x16xf32, #mma> loc(#loc467) + %b_A00_173 = tt.broadcast %b_A00_170 : tensor<1x16xf32, #mma> -> tensor<16x16xf32, #mma> loc(#loc467) + %b_A00_174 = tt.broadcast %b_A00_171 : tensor<1x16xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc467) + %b_A00_175 = arith.subf %b_A00_172, %b_A00_173 : tensor<16x16xf32, #mma> loc(#loc467) + %b_A00_176 = math.exp2 %b_A00_175 : tensor<16x16xf32, #mma> loc(#loc679) + %b_A00_177 = arith.select %b_A00_168, %b_A00_176, %cst : tensor<16x16xi1, #mma>, tensor<16x16xf32, #mma> loc(#loc469) + %b_A00_178 = arith.mulf %b_A32#0, %b_A00_177 : tensor<16x16xf32, #mma> loc(#loc470) + %b_A11 = tt.expand_dims %m_tc1_66 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xi1, #mma1> loc(#loc471) + %b_A11_179 = tt.expand_dims %m_tc1_67 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<16x1xi1, #blocked1> loc(#loc471) + %b_A11_180 = tt.broadcast %b_A11 : tensor<16x1xi1, #mma1> -> tensor<16x16xi1, #mma1> loc(#loc472) + %b_A11_181 = tt.broadcast %b_A11_179 : tensor<16x1xi1, #blocked1> -> tensor<16x16xi1, #blocked1> loc(#loc472) + %b_A11_182 = arith.andi %m_d_158, %b_A11_180 : tensor<16x16xi1, #mma1> loc(#loc472) + %b_A11_183 = tt.expand_dims %m_tc1_68 {axis = 0 : i32} : tensor<16xi1, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<1x16xi1, #mma1> loc(#loc473) + %b_A11_184 = tt.broadcast %b_A11_183 : tensor<1x16xi1, #mma1> -> tensor<16x16xi1, #mma1> loc(#loc474) + %b_A11_185 = arith.andi %b_A11_182, %b_A11_184 : tensor<16x16xi1, #mma1> loc(#loc474) + %b_A11_186 = tt.expand_dims %b_g1 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xf32, #mma1> loc(#loc475) + %b_A11_187 = tt.expand_dims %b_g1_119 {axis = 0 : i32} : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<1x16xf32, #mma1> loc(#loc476) + %b_A11_188 = tt.broadcast %b_A11_186 : tensor<16x1xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc477) + %b_A11_189 = tt.broadcast %b_A11_187 : tensor<1x16xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc477) + %b_A11_190 = arith.subf %b_A11_188, %b_A11_189 : tensor<16x16xf32, #mma1> loc(#loc477) + %b_A11_191 = math.exp2 %b_A11_190 : tensor<16x16xf32, #mma1> loc(#loc680) + %b_A11_192 = arith.select %b_A11_185, %b_A11_191, %cst_0 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc479) + %b_A11_193 = arith.mulf %b_A32#1, %b_A11_192 : tensor<16x16xf32, #mma1> loc(#loc480) + %b_A22 = tt.expand_dims %m_tc2_77 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xi1, #mma1> loc(#loc481) + %b_A22_194 = tt.expand_dims %m_tc2_78 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<16x1xi1, #blocked1> loc(#loc481) + %b_A22_195 = tt.broadcast %b_A22 : tensor<16x1xi1, #mma1> -> tensor<16x16xi1, #mma1> loc(#loc482) + %b_A22_196 = tt.broadcast %b_A22_194 : tensor<16x1xi1, #blocked1> -> tensor<16x16xi1, #blocked1> loc(#loc482) + %b_A22_197 = arith.andi %m_d_158, %b_A22_195 : tensor<16x16xi1, #mma1> loc(#loc482) + %b_A22_198 = tt.expand_dims %m_tc2_79 {axis = 0 : i32} : tensor<16xi1, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<1x16xi1, #mma1> loc(#loc483) + %b_A22_199 = tt.broadcast %b_A22_198 : tensor<1x16xi1, #mma1> -> tensor<16x16xi1, #mma1> loc(#loc484) + %b_A22_200 = arith.andi %b_A22_197, %b_A22_199 : tensor<16x16xi1, #mma1> loc(#loc484) + %b_A22_201 = tt.expand_dims %b_g2 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xf32, #mma1> loc(#loc485) + %b_A22_202 = tt.expand_dims %b_g2_120 {axis = 0 : i32} : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<1x16xf32, #mma1> loc(#loc486) + %b_A22_203 = tt.broadcast %b_A22_201 : tensor<16x1xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc487) + %b_A22_204 = tt.broadcast %b_A22_202 : tensor<1x16xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc487) + %b_A22_205 = arith.subf %b_A22_203, %b_A22_204 : tensor<16x16xf32, #mma1> loc(#loc487) + %b_A22_206 = math.exp2 %b_A22_205 : tensor<16x16xf32, #mma1> loc(#loc681) + %b_A22_207 = arith.select %b_A22_200, %b_A22_206, %cst_0 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc489) + %b_A22_208 = arith.mulf %b_A32#2, %b_A22_207 : tensor<16x16xf32, #mma1> loc(#loc490) + %b_A33 = tt.expand_dims %m_tc3_88 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xi1, #mma1> loc(#loc491) + %b_A33_209 = tt.expand_dims %m_tc3_89 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<16x1xi1, #blocked1> loc(#loc491) + %b_A33_210 = tt.broadcast %b_A33 : tensor<16x1xi1, #mma1> -> tensor<16x16xi1, #mma1> loc(#loc492) + %b_A33_211 = tt.broadcast %b_A33_209 : tensor<16x1xi1, #blocked1> -> tensor<16x16xi1, #blocked1> loc(#loc492) + %b_A33_212 = arith.andi %m_d_158, %b_A33_210 : tensor<16x16xi1, #mma1> loc(#loc492) + %b_A33_213 = tt.expand_dims %m_tc3_90 {axis = 0 : i32} : tensor<16xi1, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<1x16xi1, #mma1> loc(#loc493) + %b_A33_214 = tt.broadcast %b_A33_213 : tensor<1x16xi1, #mma1> -> tensor<16x16xi1, #mma1> loc(#loc494) + %b_A33_215 = arith.andi %b_A33_212, %b_A33_214 : tensor<16x16xi1, #mma1> loc(#loc494) + %b_A33_216 = tt.expand_dims %b_g3 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xf32, #mma1> loc(#loc495) + %b_A33_217 = tt.expand_dims %b_g3_121 {axis = 0 : i32} : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<1x16xf32, #mma1> loc(#loc496) + %b_A33_218 = tt.broadcast %b_A33_216 : tensor<16x1xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc497) + %b_A33_219 = tt.broadcast %b_A33_217 : tensor<1x16xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc497) + %b_A33_220 = arith.subf %b_A33_218, %b_A33_219 : tensor<16x16xf32, #mma1> loc(#loc497) + %b_A33_221 = math.exp2 %b_A33_220 : tensor<16x16xf32, #mma1> loc(#loc682) + %b_A33_222 = arith.select %b_A33_215, %b_A33_221, %cst_0 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc499) + %b_A33_223 = arith.mulf %b_A32#3, %b_A33_222 : tensor<16x16xf32, #mma1> loc(#loc500) + %b_A10 = arith.andi %b_A11_180, %b_A00_167 : tensor<16x16xi1, #mma1> loc(#loc501) + %b_A10_224 = arith.subf %b_A11_188, %b_A00_174 : tensor<16x16xf32, #mma1> loc(#loc502) + %b_A10_225 = math.exp2 %b_A10_224 : tensor<16x16xf32, #mma1> loc(#loc683) + %b_A10_226 = arith.select %b_A10, %b_A10_225, %cst_0 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc504) + %b_A10_227 = arith.mulf %b_A32#4, %b_A10_226 : tensor<16x16xf32, #mma1> loc(#loc505) + %b_A20 = arith.andi %b_A22_195, %b_A00_167 : tensor<16x16xi1, #mma1> loc(#loc506) + %b_A20_228 = arith.subf %b_A22_203, %b_A00_174 : tensor<16x16xf32, #mma1> loc(#loc507) + %b_A20_229 = math.exp2 %b_A20_228 : tensor<16x16xf32, #mma1> loc(#loc684) + %b_A20_230 = arith.select %b_A20, %b_A20_229, %cst_0 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc509) + %b_A20_231 = arith.mulf %b_A32#5, %b_A20_230 : tensor<16x16xf32, #mma1> loc(#loc510) + %b_A21 = arith.andi %b_A22_195, %b_A11_184 : tensor<16x16xi1, #mma1> loc(#loc511) + %b_A21_232 = arith.subf %b_A22_203, %b_A11_189 : tensor<16x16xf32, #mma1> loc(#loc512) + %b_A21_233 = math.exp2 %b_A21_232 : tensor<16x16xf32, #mma1> loc(#loc685) + %b_A21_234 = arith.select %b_A21, %b_A21_233, %cst_0 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc514) + %b_A21_235 = arith.mulf %b_A32#6, %b_A21_234 : tensor<16x16xf32, #mma1> loc(#loc515) + %b_A30 = arith.andi %b_A33_210, %b_A00_167 : tensor<16x16xi1, #mma1> loc(#loc516) + %b_A30_236 = arith.subf %b_A33_218, %b_A00_174 : tensor<16x16xf32, #mma1> loc(#loc517) + %b_A30_237 = math.exp2 %b_A30_236 : tensor<16x16xf32, #mma1> loc(#loc686) + %b_A30_238 = arith.select %b_A30, %b_A30_237, %cst_0 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc519) + %b_A30_239 = arith.mulf %b_A32#7, %b_A30_238 : tensor<16x16xf32, #mma1> loc(#loc520) + %b_A31 = arith.andi %b_A33_210, %b_A11_184 : tensor<16x16xi1, #mma1> loc(#loc521) + %b_A31_240 = arith.subf %b_A33_218, %b_A11_189 : tensor<16x16xf32, #mma1> loc(#loc522) + %b_A31_241 = math.exp2 %b_A31_240 : tensor<16x16xf32, #mma1> loc(#loc687) + %b_A31_242 = arith.select %b_A31, %b_A31_241, %cst_0 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc524) + %b_A31_243 = arith.mulf %b_A32#8, %b_A31_242 : tensor<16x16xf32, #mma1> loc(#loc525) + %b_A32_244 = arith.andi %b_A33_210, %b_A22_199 : tensor<16x16xi1, #mma1> loc(#loc526) + %b_A32_245 = arith.subf %b_A33_218, %b_A22_204 : tensor<16x16xf32, #mma1> loc(#loc527) + %b_A32_246 = math.exp2 %b_A32_245 : tensor<16x16xf32, #mma1> loc(#loc688) + %b_A32_247 = arith.select %b_A32_244, %b_A32_246, %cst_0 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc529) + %b_A32_248 = arith.mulf %b_A32#9, %b_A32_247 : tensor<16x16xf32, #mma1> loc(#loc530) + %b_A00_249 = tt.expand_dims %b_b0_103 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma}>> -> tensor<16x1xf32, #mma> loc(#loc531) + %b_A00_250 = tt.broadcast %b_A00_249 : tensor<16x1xf32, #mma> -> tensor<16x16xf32, #mma> loc(#loc532) + %b_A00_251 = arith.mulf %b_A00_178, %b_A00_250 : tensor<16x16xf32, #mma> loc(#loc532) + %b_A11_252 = tt.expand_dims %b_b1_104 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xf32, #mma1> loc(#loc533) + %b_A11_253 = tt.broadcast %b_A11_252 : tensor<16x1xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc534) + %b_A11_254 = arith.mulf %b_A11_193, %b_A11_253 : tensor<16x16xf32, #mma1> loc(#loc534) + %b_A22_255 = tt.expand_dims %b_b2_105 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xf32, #mma1> loc(#loc535) + %b_A22_256 = tt.broadcast %b_A22_255 : tensor<16x1xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc536) + %b_A22_257 = arith.mulf %b_A22_208, %b_A22_256 : tensor<16x16xf32, #mma1> loc(#loc536) + %b_A33_258 = tt.expand_dims %b_b3_106 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xf32, #mma1> loc(#loc537) + %b_A33_259 = tt.broadcast %b_A33_258 : tensor<16x1xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc538) + %b_A33_260 = arith.mulf %b_A33_223, %b_A33_259 : tensor<16x16xf32, #mma1> loc(#loc538) + %b_A10_261 = arith.mulf %b_A10_227, %b_A11_253 : tensor<16x16xf32, #mma1> loc(#loc539) + %b_A20_262 = arith.mulf %b_A20_231, %b_A22_256 : tensor<16x16xf32, #mma1> loc(#loc540) + %b_A21_263 = arith.mulf %b_A21_235, %b_A22_256 : tensor<16x16xf32, #mma1> loc(#loc541) + %b_A30_264 = arith.mulf %b_A30_239, %b_A33_259 : tensor<16x16xf32, #mma1> loc(#loc542) + %b_A31_265 = arith.mulf %b_A31_243, %b_A33_259 : tensor<16x16xf32, #mma1> loc(#loc543) + %b_A32_266 = arith.mulf %b_A32_248, %b_A33_259 : tensor<16x16xf32, #mma1> loc(#loc544) + %b_Ai00 = arith.subf %cst, %b_A00_251 : tensor<16x16xf32, #mma> loc(#loc545) + %b_Ai11 = arith.subf %cst_0, %b_A11_254 : tensor<16x16xf32, #mma1> loc(#loc546) + %b_Ai22 = arith.subf %cst_0, %b_A22_257 : tensor<16x16xf32, #mma1> loc(#loc547) + %b_Ai33 = arith.subf %cst_0, %b_A33_260 : tensor<16x16xf32, #mma1> loc(#loc548) + %11 = arith.subi %6, %8 : i64 loc(#loc205) + %12 = arith.minsi %11, %c16_i64 : i64 loc(#loc206) + %b_Ai00_267 = scf.for %b_Ai00_355 = %c2_i64 to %12 step %c1_i64 iter_args(%b_Ai00_356 = %b_Ai00) -> (tensor<16x16xf32, #mma>) : i64 { + %b_a00 = tt.splat %b_Ai00_355 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc550) + %b_a00_357 = tt.splat %b_Ai00_355 : i64 -> tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc550) + %b_a00_358 = arith.cmpi eq, %m_tc0_32, %b_a00 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc550) + %b_a00_359 = tt.expand_dims %b_a00_358 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #mma}>> -> tensor<16x1xi1, #mma> loc(#loc551) + %b_a00_360 = tt.broadcast %b_a00_359 : tensor<16x1xi1, #mma> -> tensor<16x16xi1, #mma> loc(#loc552) + %b_a00_361 = arith.select %b_a00_360, %b_Ai00, %cst : tensor<16x16xi1, #mma>, tensor<16x16xf32, #mma> loc(#loc552) + %b_a00_362 = "tt.reduce"(%b_a00_361) <{axis = 0 : i32}> ({ + ^bb0(%b_a00_374: f32 loc(callsite(#loc1 at #loc553)), %b_a00_375: f32 loc(callsite(#loc1 at #loc553))): + %b_a00_376 = arith.addf %b_a00_374, %b_a00_375 : f32 loc(#loc708) + tt.reduce.return %b_a00_376 : f32 loc(#loc689) + }) : (tensor<16x16xf32, #mma>) -> tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc689) + %b_a00_363 = arith.cmpi slt, %m_tc0_35, %b_a00_357 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc554) + %b_a00_364 = arith.select %b_a00_363, %b_a00_362, %cst_1 : tensor<16xi1, #ttg.slice<{dim = 0, parent = #mma}>>, tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc555) + %b_a00_365 = ttg.convert_layout %b_a00_364 : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma}>> -> tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc556) + %b_a00_366 = tt.expand_dims %b_a00_365 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma}>> -> tensor<16x1xf32, #mma> loc(#loc556) + %b_a00_367 = tt.broadcast %b_a00_366 : tensor<16x1xf32, #mma> -> tensor<16x16xf32, #mma> loc(#loc557) + %b_a00_368 = arith.mulf %b_a00_367, %b_Ai00_356 : tensor<16x16xf32, #mma> loc(#loc557) + %b_a00_369 = "tt.reduce"(%b_a00_368) <{axis = 0 : i32}> ({ + ^bb0(%b_a00_374: f32 loc(callsite(#loc1 at #loc558)), %b_a00_375: f32 loc(callsite(#loc1 at #loc558))): + %b_a00_376 = arith.addf %b_a00_374, %b_a00_375 : f32 loc(#loc709) + tt.reduce.return %b_a00_376 : f32 loc(#loc691) + }) : (tensor<16x16xf32, #mma>) -> tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc691) + %b_a00_370 = arith.addf %b_a00_364, %b_a00_369 : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc559) + %b_Ai00_371 = tt.expand_dims %b_a00_370 {axis = 0 : i32} : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma}>> -> tensor<1x16xf32, #mma> loc(#loc560) + %b_Ai00_372 = tt.broadcast %b_Ai00_371 : tensor<1x16xf32, #mma> -> tensor<16x16xf32, #mma> loc(#loc560) + %b_Ai00_373 = arith.select %b_a00_360, %b_Ai00_372, %b_Ai00_356 : tensor<16x16xi1, #mma>, tensor<16x16xf32, #mma> loc(#loc560) + scf.yield %b_Ai00_373 : tensor<16x16xf32, #mma> loc(#loc221) + } loc(#loc549) + %13 = arith.subi %6, %i_tc1 : i64 loc(#loc222) + %14 = arith.minsi %13, %c16_i64 : i64 loc(#loc223) + %b_Ai11_268 = scf.for %b_Ai11_355 = %c2_i64 to %14 step %c1_i64 iter_args(%b_Ai11_356 = %b_Ai11) -> (tensor<16x16xf32, #mma1>) : i64 { + %b_a11 = tt.splat %b_Ai11_355 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc562) + %b_a11_357 = tt.splat %b_Ai11_355 : i64 -> tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc562) + %b_a11_358 = arith.cmpi eq, %m_tc0_33, %b_a11 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc562) + %b_a11_359 = tt.expand_dims %b_a11_358 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xi1, #mma1> loc(#loc563) + %b_a11_360 = tt.broadcast %b_a11_359 : tensor<16x1xi1, #mma1> -> tensor<16x16xi1, #mma1> loc(#loc564) + %b_a11_361 = arith.select %b_a11_360, %b_Ai11, %cst_0 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc564) + %b_a11_362 = "tt.reduce"(%b_a11_361) <{axis = 0 : i32}> ({ + ^bb0(%b_a11_374: f32 loc(callsite(#loc1 at #loc565)), %b_a11_375: f32 loc(callsite(#loc1 at #loc565))): + %b_a11_376 = arith.addf %b_a11_374, %b_a11_375 : f32 loc(#loc710) + tt.reduce.return %b_a11_376 : f32 loc(#loc693) + }) : (tensor<16x16xf32, #mma1>) -> tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc693) + %b_a11_363 = arith.cmpi slt, %m_tc0_36, %b_a11_357 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc566) + %b_a11_364 = arith.select %b_a11_363, %b_a11_362, %cst_2 : tensor<16xi1, #ttg.slice<{dim = 0, parent = #mma1}>>, tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc567) + %b_a11_365 = ttg.convert_layout %b_a11_364 : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc568) + %b_a11_366 = tt.expand_dims %b_a11_365 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xf32, #mma1> loc(#loc568) + %b_a11_367 = tt.broadcast %b_a11_366 : tensor<16x1xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc569) + %b_a11_368 = arith.mulf %b_a11_367, %b_Ai11_356 : tensor<16x16xf32, #mma1> loc(#loc569) + %b_a11_369 = "tt.reduce"(%b_a11_368) <{axis = 0 : i32}> ({ + ^bb0(%b_a11_374: f32 loc(callsite(#loc1 at #loc570)), %b_a11_375: f32 loc(callsite(#loc1 at #loc570))): + %b_a11_376 = arith.addf %b_a11_374, %b_a11_375 : f32 loc(#loc711) + tt.reduce.return %b_a11_376 : f32 loc(#loc695) + }) : (tensor<16x16xf32, #mma1>) -> tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc695) + %b_a11_370 = arith.addf %b_a11_364, %b_a11_369 : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc571) + %b_Ai11_371 = tt.expand_dims %b_a11_370 {axis = 0 : i32} : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<1x16xf32, #mma1> loc(#loc572) + %b_Ai11_372 = tt.broadcast %b_Ai11_371 : tensor<1x16xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc572) + %b_Ai11_373 = arith.select %b_a11_360, %b_Ai11_372, %b_Ai11_356 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc572) + scf.yield %b_Ai11_373 : tensor<16x16xf32, #mma1> loc(#loc236) + } loc(#loc561) + %15 = arith.subi %6, %i_tc2 : i64 loc(#loc237) + %16 = arith.minsi %15, %c16_i64 : i64 loc(#loc238) + %b_Ai22_269 = scf.for %b_Ai22_355 = %c2_i64 to %16 step %c1_i64 iter_args(%b_Ai22_356 = %b_Ai22) -> (tensor<16x16xf32, #mma1>) : i64 { + %b_a22 = tt.splat %b_Ai22_355 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc574) + %b_a22_357 = tt.splat %b_Ai22_355 : i64 -> tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc574) + %b_a22_358 = arith.cmpi eq, %m_tc0_33, %b_a22 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc574) + %b_a22_359 = tt.expand_dims %b_a22_358 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xi1, #mma1> loc(#loc575) + %b_a22_360 = tt.broadcast %b_a22_359 : tensor<16x1xi1, #mma1> -> tensor<16x16xi1, #mma1> loc(#loc576) + %b_a22_361 = arith.select %b_a22_360, %b_Ai22, %cst_0 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc576) + %b_a22_362 = "tt.reduce"(%b_a22_361) <{axis = 0 : i32}> ({ + ^bb0(%b_a22_374: f32 loc(callsite(#loc1 at #loc577)), %b_a22_375: f32 loc(callsite(#loc1 at #loc577))): + %b_a22_376 = arith.addf %b_a22_374, %b_a22_375 : f32 loc(#loc712) + tt.reduce.return %b_a22_376 : f32 loc(#loc697) + }) : (tensor<16x16xf32, #mma1>) -> tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc697) + %b_a22_363 = arith.cmpi slt, %m_tc0_36, %b_a22_357 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc578) + %b_a22_364 = arith.select %b_a22_363, %b_a22_362, %cst_2 : tensor<16xi1, #ttg.slice<{dim = 0, parent = #mma1}>>, tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc579) + %b_a22_365 = ttg.convert_layout %b_a22_364 : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc580) + %b_a22_366 = tt.expand_dims %b_a22_365 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xf32, #mma1> loc(#loc580) + %b_a22_367 = tt.broadcast %b_a22_366 : tensor<16x1xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc581) + %b_a22_368 = arith.mulf %b_a22_367, %b_Ai22_356 : tensor<16x16xf32, #mma1> loc(#loc581) + %b_a22_369 = "tt.reduce"(%b_a22_368) <{axis = 0 : i32}> ({ + ^bb0(%b_a22_374: f32 loc(callsite(#loc1 at #loc582)), %b_a22_375: f32 loc(callsite(#loc1 at #loc582))): + %b_a22_376 = arith.addf %b_a22_374, %b_a22_375 : f32 loc(#loc713) + tt.reduce.return %b_a22_376 : f32 loc(#loc699) + }) : (tensor<16x16xf32, #mma1>) -> tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc699) + %b_a22_370 = arith.addf %b_a22_364, %b_a22_369 : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc583) + %b_Ai22_371 = tt.expand_dims %b_a22_370 {axis = 0 : i32} : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<1x16xf32, #mma1> loc(#loc584) + %b_Ai22_372 = tt.broadcast %b_Ai22_371 : tensor<1x16xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc584) + %b_Ai22_373 = arith.select %b_a22_360, %b_Ai22_372, %b_Ai22_356 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc584) + scf.yield %b_Ai22_373 : tensor<16x16xf32, #mma1> loc(#loc251) + } loc(#loc573) + %17 = arith.subi %6, %i_tc3 : i64 loc(#loc252) + %18 = arith.minsi %17, %c16_i64 : i64 loc(#loc253) + %b_Ai33_270 = scf.for %b_Ai33_355 = %c2_i64 to %18 step %c1_i64 iter_args(%b_Ai33_356 = %b_Ai33) -> (tensor<16x16xf32, #mma1>) : i64 { + %b_a33 = tt.splat %b_Ai33_355 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc586) + %b_a33_357 = tt.splat %b_Ai33_355 : i64 -> tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc586) + %b_a33_358 = arith.cmpi eq, %m_tc0_33, %b_a33 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc586) + %b_a33_359 = tt.expand_dims %b_a33_358 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xi1, #mma1> loc(#loc587) + %b_a33_360 = tt.broadcast %b_a33_359 : tensor<16x1xi1, #mma1> -> tensor<16x16xi1, #mma1> loc(#loc588) + %b_a33_361 = arith.select %b_a33_360, %b_Ai33, %cst_0 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc588) + %b_a33_362 = "tt.reduce"(%b_a33_361) <{axis = 0 : i32}> ({ + ^bb0(%b_a33_374: f32 loc(callsite(#loc1 at #loc589)), %b_a33_375: f32 loc(callsite(#loc1 at #loc589))): + %b_a33_376 = arith.addf %b_a33_374, %b_a33_375 : f32 loc(#loc714) + tt.reduce.return %b_a33_376 : f32 loc(#loc701) + }) : (tensor<16x16xf32, #mma1>) -> tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc701) + %b_a33_363 = arith.cmpi slt, %m_tc0_36, %b_a33_357 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc590) + %b_a33_364 = arith.select %b_a33_363, %b_a33_362, %cst_2 : tensor<16xi1, #ttg.slice<{dim = 0, parent = #mma1}>>, tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc591) + %b_a33_365 = ttg.convert_layout %b_a33_364 : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc592) + %b_a33_366 = tt.expand_dims %b_a33_365 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xf32, #mma1> loc(#loc592) + %b_a33_367 = tt.broadcast %b_a33_366 : tensor<16x1xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc593) + %b_a33_368 = arith.mulf %b_a33_367, %b_Ai33_356 : tensor<16x16xf32, #mma1> loc(#loc593) + %b_a33_369 = "tt.reduce"(%b_a33_368) <{axis = 0 : i32}> ({ + ^bb0(%b_a33_374: f32 loc(callsite(#loc1 at #loc594)), %b_a33_375: f32 loc(callsite(#loc1 at #loc594))): + %b_a33_376 = arith.addf %b_a33_374, %b_a33_375 : f32 loc(#loc715) + tt.reduce.return %b_a33_376 : f32 loc(#loc703) + }) : (tensor<16x16xf32, #mma1>) -> tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc703) + %b_a33_370 = arith.addf %b_a33_364, %b_a33_369 : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc595) + %b_Ai33_371 = tt.expand_dims %b_a33_370 {axis = 0 : i32} : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<1x16xf32, #mma1> loc(#loc596) + %b_Ai33_372 = tt.broadcast %b_Ai33_371 : tensor<1x16xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc596) + %b_Ai33_373 = arith.select %b_a33_360, %b_Ai33_372, %b_Ai33_356 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc596) + scf.yield %b_Ai33_373 : tensor<16x16xf32, #mma1> loc(#loc266) + } loc(#loc585) + %b_Ai00_271 = arith.uitofp %m_I : tensor<16x16xi1, #mma1> to tensor<16x16xf32, #mma1> loc(#loc597) + %b_Ai00_272 = arith.uitofp %m_I_159 : tensor<16x16xi1, #mma> to tensor<16x16xf32, #mma> loc(#loc597) + %b_Ai00_273 = arith.addf %b_Ai00_267, %b_Ai00_272 : tensor<16x16xf32, #mma> loc(#loc597) + %b_Ai00_274 = ttg.local_alloc %b_Ai00_273 : (tensor<16x16xf32, #mma>) -> !ttg.memdesc<16x16xf32, #shared1, #smem> loc(#loc597) + %b_Ai11_275 = arith.addf %b_Ai11_268, %b_Ai00_271 : tensor<16x16xf32, #mma1> loc(#loc598) + %b_Ai22_276 = arith.addf %b_Ai22_269, %b_Ai00_271 : tensor<16x16xf32, #mma1> loc(#loc599) + %b_Ai33_277 = arith.addf %b_Ai33_270, %b_Ai00_271 : tensor<16x16xf32, #mma1> loc(#loc600) + %b_Ai11_278 = ttg.convert_layout %b_Ai11_275 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> loc(#loc598) + %b_A10_279 = ttg.convert_layout %b_A10_261 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> loc(#loc539) + %b_Ai10 = tt.dot %b_Ai11_278, %b_A10_279, %cst_0, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc601) + %b_Ai10_280 = ttg.convert_layout %b_Ai10 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 1}>> loc(#loc602) + %b_Ai00_281 = ttg.local_load %b_Ai00_274 : !ttg.memdesc<16x16xf32, #shared1, #smem> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 1}>> loc(#loc597) + %b_Ai10_282 = tt.dot %b_Ai10_280, %b_Ai00_281, %cst_0, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 1}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 1}>> -> tensor<16x16xf32, #mma1> loc(#loc602) + %b_Ai10_283 = arith.subf %cst_0, %b_Ai10_282 : tensor<16x16xf32, #mma1> loc(#loc603) + %b_Ai22_284 = ttg.convert_layout %b_Ai22_276 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> loc(#loc599) + %b_A21_285 = ttg.convert_layout %b_A21_263 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> loc(#loc541) + %b_Ai21 = tt.dot %b_Ai22_284, %b_A21_285, %cst_0, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc604) + %b_Ai21_286 = ttg.convert_layout %b_Ai21 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 1}>> loc(#loc605) + %b_Ai11_287 = ttg.convert_layout %b_Ai11_275 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 1}>> loc(#loc598) + %b_Ai21_288 = tt.dot %b_Ai21_286, %b_Ai11_287, %cst_0, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 1}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 1}>> -> tensor<16x16xf32, #mma1> loc(#loc605) + %b_Ai21_289 = arith.subf %cst_0, %b_Ai21_288 : tensor<16x16xf32, #mma1> loc(#loc606) + %b_Ai33_290 = ttg.convert_layout %b_Ai33_277 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> loc(#loc600) + %b_A32_291 = ttg.convert_layout %b_A32_266 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> loc(#loc544) + %b_Ai32 = tt.dot %b_Ai33_290, %b_A32_291, %cst_0, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc607) + %b_Ai32_292 = ttg.convert_layout %b_Ai32 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 1}>> loc(#loc608) + %b_Ai22_293 = ttg.convert_layout %b_Ai22_276 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 1}>> loc(#loc599) + %b_Ai32_294 = tt.dot %b_Ai32_292, %b_Ai22_293, %cst_0, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 1}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 1}>> -> tensor<16x16xf32, #mma1> loc(#loc608) + %b_Ai32_295 = arith.subf %cst_0, %b_Ai32_294 : tensor<16x16xf32, #mma1> loc(#loc609) + %b_A20_296 = ttg.convert_layout %b_A20_262 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> loc(#loc540) + %b_Ai00_297 = ttg.local_load %b_Ai00_274 : !ttg.memdesc<16x16xf32, #shared1, #smem> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> loc(#loc597) + %b_Ai20 = tt.dot %b_A20_296, %b_Ai00_297, %cst_0, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc610) + %b_A21_298 = ttg.convert_layout %b_A21_263 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> loc(#loc541) + %b_Ai10_299 = ttg.convert_layout %b_Ai10_283 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> loc(#loc603) + %b_Ai20_300 = tt.dot %b_A21_298, %b_Ai10_299, %b_Ai20, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc611) + %b_Ai22_301 = ttg.convert_layout %b_Ai22_276 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 1}>> loc(#loc599) + %b_Ai20_302 = ttg.convert_layout %b_Ai20_300 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 1}>> loc(#loc612) + %b_Ai20_303 = tt.dot %b_Ai22_301, %b_Ai20_302, %cst_0, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 1}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 1}>> -> tensor<16x16xf32, #mma1> loc(#loc612) + %b_Ai20_304 = arith.subf %cst_0, %b_Ai20_303 : tensor<16x16xf32, #mma1> loc(#loc613) + %b_A31_305 = ttg.convert_layout %b_A31_265 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> loc(#loc543) + %b_Ai11_306 = ttg.convert_layout %b_Ai11_275 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> loc(#loc598) + %b_Ai31 = tt.dot %b_A31_305, %b_Ai11_306, %cst_0, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc614) + %b_A32_307 = ttg.convert_layout %b_A32_266 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> loc(#loc544) + %b_Ai21_308 = ttg.convert_layout %b_Ai21_289 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> loc(#loc606) + %b_Ai31_309 = tt.dot %b_A32_307, %b_Ai21_308, %b_Ai31, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc615) + %b_Ai33_310 = ttg.convert_layout %b_Ai33_277 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 1}>> loc(#loc600) + %b_Ai31_311 = ttg.convert_layout %b_Ai31_309 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 1}>> loc(#loc616) + %b_Ai31_312 = tt.dot %b_Ai33_310, %b_Ai31_311, %cst_0, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 1}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 1}>> -> tensor<16x16xf32, #mma1> loc(#loc616) + %b_Ai31_313 = arith.subf %cst_0, %b_Ai31_312 : tensor<16x16xf32, #mma1> loc(#loc617) + %b_A30_314 = ttg.convert_layout %b_A30_264 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> loc(#loc542) + %b_Ai30 = tt.dot %b_A30_314, %b_Ai00_297, %cst_0, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc618) + %b_Ai30_315 = tt.dot %b_A31_305, %b_Ai10_299, %b_Ai30, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc619) + %b_Ai20_316 = ttg.convert_layout %b_Ai20_304 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> loc(#loc613) + %b_Ai30_317 = tt.dot %b_A32_307, %b_Ai20_316, %b_Ai30_315, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc620) + %b_Ai30_318 = ttg.convert_layout %b_Ai30_317 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 1}>> loc(#loc621) + %b_Ai30_319 = tt.dot %b_Ai33_310, %b_Ai30_318, %cst_0, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 1}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 1}>> -> tensor<16x16xf32, #mma1> loc(#loc621) + %b_Ai30_320 = arith.subf %cst_0, %b_Ai30_319 : tensor<16x16xf32, #mma1> loc(#loc622) + %p_A00 = tt.expand_dims %m_tc0_44 {axis = 1 : i32} : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<16x1xi64, #blocked1> loc(#loc623) + %p_A00_321 = arith.muli %p_A00, %cst_6 : tensor<16x1xi64, #blocked1> loc(#loc624) + %p_A00_322 = tt.splat %A_25 : !tt.ptr -> tensor<16x1x!tt.ptr, #blocked1> loc(#loc625) + %p_A00_323 = tt.addptr %p_A00_322, %p_A00_321 : tensor<16x1x!tt.ptr, #blocked1>, tensor<16x1xi64, #blocked1> loc(#loc625) + %p_A00_324 = tt.broadcast %p_A00_323 : tensor<16x1x!tt.ptr, #blocked1> -> tensor<16x16x!tt.ptr, #blocked1> loc(#loc626) + %p_A00_325 = tt.addptr %p_A00_324, %m_d_156 : tensor<16x16x!tt.ptr, #blocked1>, tensor<16x16xi32, #blocked1> loc(#loc626) + %p_A10 = tt.expand_dims %m_tc1_63 {axis = 1 : i32} : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<16x1xi64, #blocked1> loc(#loc627) + %p_A10_326 = arith.muli %p_A10, %cst_6 : tensor<16x1xi64, #blocked1> loc(#loc628) + %p_A10_327 = tt.addptr %p_A00_322, %p_A10_326 : tensor<16x1x!tt.ptr, #blocked1>, tensor<16x1xi64, #blocked1> loc(#loc629) + %p_A10_328 = tt.broadcast %p_A10_327 : tensor<16x1x!tt.ptr, #blocked1> -> tensor<16x16x!tt.ptr, #blocked1> loc(#loc630) + %p_A10_329 = tt.addptr %p_A10_328, %m_d_156 : tensor<16x16x!tt.ptr, #blocked1>, tensor<16x16xi32, #blocked1> loc(#loc630) + %p_A11 = arith.addi %o_i_31, %cst_7 : tensor<16xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc631) + %p_A11_330 = tt.expand_dims %p_A11 {axis = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x16xi32, #blocked1> loc(#loc632) + %p_A11_331 = tt.broadcast %p_A11_330 : tensor<1x16xi32, #blocked1> -> tensor<16x16xi32, #blocked1> loc(#loc633) + %p_A11_332 = tt.addptr %p_A10_328, %p_A11_331 : tensor<16x16x!tt.ptr, #blocked1>, tensor<16x16xi32, #blocked1> loc(#loc633) + %p_A20 = tt.expand_dims %m_tc2_74 {axis = 1 : i32} : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<16x1xi64, #blocked1> loc(#loc634) + %p_A20_333 = arith.muli %p_A20, %cst_6 : tensor<16x1xi64, #blocked1> loc(#loc635) + %p_A20_334 = tt.addptr %p_A00_322, %p_A20_333 : tensor<16x1x!tt.ptr, #blocked1>, tensor<16x1xi64, #blocked1> loc(#loc636) + %p_A20_335 = tt.broadcast %p_A20_334 : tensor<16x1x!tt.ptr, #blocked1> -> tensor<16x16x!tt.ptr, #blocked1> loc(#loc637) + %p_A20_336 = tt.addptr %p_A20_335, %m_d_156 : tensor<16x16x!tt.ptr, #blocked1>, tensor<16x16xi32, #blocked1> loc(#loc637) + %p_A21 = tt.addptr %p_A20_335, %p_A11_331 : tensor<16x16x!tt.ptr, #blocked1>, tensor<16x16xi32, #blocked1> loc(#loc638) + %p_A22 = arith.addi %o_i_31, %cst_8 : tensor<16xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc639) + %p_A22_337 = tt.expand_dims %p_A22 {axis = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x16xi32, #blocked1> loc(#loc640) + %p_A22_338 = tt.broadcast %p_A22_337 : tensor<1x16xi32, #blocked1> -> tensor<16x16xi32, #blocked1> loc(#loc641) + %p_A22_339 = tt.addptr %p_A20_335, %p_A22_338 : tensor<16x16x!tt.ptr, #blocked1>, tensor<16x16xi32, #blocked1> loc(#loc641) + %p_A30 = tt.expand_dims %m_tc3_85 {axis = 1 : i32} : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<16x1xi64, #blocked1> loc(#loc642) + %p_A30_340 = arith.muli %p_A30, %cst_6 : tensor<16x1xi64, #blocked1> loc(#loc643) + %p_A30_341 = tt.addptr %p_A00_322, %p_A30_340 : tensor<16x1x!tt.ptr, #blocked1>, tensor<16x1xi64, #blocked1> loc(#loc644) + %p_A30_342 = tt.broadcast %p_A30_341 : tensor<16x1x!tt.ptr, #blocked1> -> tensor<16x16x!tt.ptr, #blocked1> loc(#loc645) + %p_A30_343 = tt.addptr %p_A30_342, %m_d_156 : tensor<16x16x!tt.ptr, #blocked1>, tensor<16x16xi32, #blocked1> loc(#loc645) + %p_A31 = tt.addptr %p_A30_342, %p_A11_331 : tensor<16x16x!tt.ptr, #blocked1>, tensor<16x16xi32, #blocked1> loc(#loc646) + %p_A32 = tt.addptr %p_A30_342, %p_A22_338 : tensor<16x16x!tt.ptr, #blocked1>, tensor<16x16xi32, #blocked1> loc(#loc647) + %p_A33 = arith.addi %o_i_31, %cst_9 : tensor<16xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc648) + %p_A33_344 = tt.expand_dims %p_A33 {axis = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x16xi32, #blocked1> loc(#loc649) + %p_A33_345 = tt.broadcast %p_A33_344 : tensor<1x16xi32, #blocked1> -> tensor<16x16xi32, #blocked1> loc(#loc650) + %p_A33_346 = tt.addptr %p_A30_342, %p_A33_345 : tensor<16x16x!tt.ptr, #blocked1>, tensor<16x16xi32, #blocked1> loc(#loc650) + %m_A0 = arith.cmpi slt, %m_d_151, %cst_10 : tensor<1x16xi32, #blocked1> loc(#loc651) + %m_A0_347 = tt.broadcast %m_A0 : tensor<1x16xi1, #blocked1> -> tensor<16x16xi1, #blocked1> loc(#loc652) + %m_A0_348 = arith.andi %b_A00_162, %m_A0_347 : tensor<16x16xi1, #blocked1> loc(#loc652) + %m_A1 = arith.andi %b_A11_181, %m_A0_347 : tensor<16x16xi1, #blocked1> loc(#loc653) + %m_A2 = arith.andi %b_A22_196, %m_A0_347 : tensor<16x16xi1, #blocked1> loc(#loc654) + %m_A3 = arith.andi %b_A33_211, %m_A0_347 : tensor<16x16xi1, #blocked1> loc(#loc655) + %m_A11 = arith.cmpi slt, %p_A11_330, %cst_10 : tensor<1x16xi32, #blocked1> loc(#loc656) + %m_A11_349 = tt.broadcast %m_A11 : tensor<1x16xi1, #blocked1> -> tensor<16x16xi1, #blocked1> loc(#loc657) + %m_A11_350 = arith.andi %b_A11_181, %m_A11_349 : tensor<16x16xi1, #blocked1> loc(#loc657) + %m_A21 = arith.andi %b_A22_196, %m_A11_349 : tensor<16x16xi1, #blocked1> loc(#loc658) + %m_A22 = arith.cmpi slt, %p_A22_337, %cst_10 : tensor<1x16xi32, #blocked1> loc(#loc659) + %m_A22_351 = tt.broadcast %m_A22 : tensor<1x16xi1, #blocked1> -> tensor<16x16xi1, #blocked1> loc(#loc660) + %m_A22_352 = arith.andi %b_A22_196, %m_A22_351 : tensor<16x16xi1, #blocked1> loc(#loc660) + %m_A31 = arith.andi %b_A33_211, %m_A11_349 : tensor<16x16xi1, #blocked1> loc(#loc661) + %m_A32 = arith.andi %b_A33_211, %m_A22_351 : tensor<16x16xi1, #blocked1> loc(#loc662) + %m_A33 = arith.cmpi slt, %p_A33_344, %cst_10 : tensor<1x16xi32, #blocked1> loc(#loc663) + %m_A33_353 = tt.broadcast %m_A33 : tensor<1x16xi1, #blocked1> -> tensor<16x16xi1, #blocked1> loc(#loc664) + %m_A33_354 = arith.andi %b_A33_211, %m_A33_353 : tensor<16x16xi1, #blocked1> loc(#loc664) + %19 = arith.truncf %b_Ai00_273 : tensor<16x16xf32, #mma> to tensor<16x16xbf16, #mma> loc(#loc335) + %20 = ttg.convert_layout %19 : tensor<16x16xbf16, #mma> -> tensor<16x16xbf16, #blocked1> loc(#loc336) + tt.store %p_A00_325, %20, %m_A0_348 : tensor<16x16x!tt.ptr, #blocked1> loc(#loc336) + %21 = arith.truncf %b_Ai10_283 : tensor<16x16xf32, #mma1> to tensor<16x16xbf16, #mma1> loc(#loc337) + %22 = ttg.convert_layout %21 : tensor<16x16xbf16, #mma1> -> tensor<16x16xbf16, #blocked1> loc(#loc338) + tt.store %p_A10_329, %22, %m_A1 : tensor<16x16x!tt.ptr, #blocked1> loc(#loc338) + %23 = arith.truncf %b_Ai11_275 : tensor<16x16xf32, #mma1> to tensor<16x16xbf16, #mma1> loc(#loc339) + %24 = ttg.convert_layout %23 : tensor<16x16xbf16, #mma1> -> tensor<16x16xbf16, #blocked1> loc(#loc340) + tt.store %p_A11_332, %24, %m_A11_350 : tensor<16x16x!tt.ptr, #blocked1> loc(#loc340) + %25 = arith.truncf %b_Ai20_304 : tensor<16x16xf32, #mma1> to tensor<16x16xbf16, #mma1> loc(#loc341) + %26 = ttg.convert_layout %25 : tensor<16x16xbf16, #mma1> -> tensor<16x16xbf16, #blocked1> loc(#loc342) + tt.store %p_A20_336, %26, %m_A2 : tensor<16x16x!tt.ptr, #blocked1> loc(#loc342) + %27 = arith.truncf %b_Ai21_289 : tensor<16x16xf32, #mma1> to tensor<16x16xbf16, #mma1> loc(#loc343) + %28 = ttg.convert_layout %27 : tensor<16x16xbf16, #mma1> -> tensor<16x16xbf16, #blocked1> loc(#loc344) + tt.store %p_A21, %28, %m_A21 : tensor<16x16x!tt.ptr, #blocked1> loc(#loc344) + %29 = arith.truncf %b_Ai22_276 : tensor<16x16xf32, #mma1> to tensor<16x16xbf16, #mma1> loc(#loc345) + %30 = ttg.convert_layout %29 : tensor<16x16xbf16, #mma1> -> tensor<16x16xbf16, #blocked1> loc(#loc346) + tt.store %p_A22_339, %30, %m_A22_352 : tensor<16x16x!tt.ptr, #blocked1> loc(#loc346) + %31 = arith.truncf %b_Ai30_320 : tensor<16x16xf32, #mma1> to tensor<16x16xbf16, #mma1> loc(#loc347) + %32 = ttg.convert_layout %31 : tensor<16x16xbf16, #mma1> -> tensor<16x16xbf16, #blocked1> loc(#loc348) + tt.store %p_A30_343, %32, %m_A3 : tensor<16x16x!tt.ptr, #blocked1> loc(#loc348) + %33 = arith.truncf %b_Ai31_313 : tensor<16x16xf32, #mma1> to tensor<16x16xbf16, #mma1> loc(#loc349) + %34 = ttg.convert_layout %33 : tensor<16x16xbf16, #mma1> -> tensor<16x16xbf16, #blocked1> loc(#loc350) + tt.store %p_A31, %34, %m_A31 : tensor<16x16x!tt.ptr, #blocked1> loc(#loc350) + %35 = arith.truncf %b_Ai32_295 : tensor<16x16xf32, #mma1> to tensor<16x16xbf16, #mma1> loc(#loc351) + %36 = ttg.convert_layout %35 : tensor<16x16xbf16, #mma1> -> tensor<16x16xbf16, #blocked1> loc(#loc352) + tt.store %p_A32, %36, %m_A32 : tensor<16x16x!tt.ptr, #blocked1> loc(#loc352) + %37 = arith.truncf %b_Ai33_277 : tensor<16x16xf32, #mma1> to tensor<16x16xbf16, #mma1> loc(#loc353) + %38 = ttg.convert_layout %37 : tensor<16x16xbf16, #mma1> -> tensor<16x16xbf16, #blocked1> loc(#loc354) + tt.store %p_A33_346, %38, %m_A33_354 : tensor<16x16x!tt.ptr, #blocked1> loc(#loc354) + tt.return loc(#loc355) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":70:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":70:36) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":70:61) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":70:67) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":71:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":71:34) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":78:25) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":80:13) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":80:19) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":81:8) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":84:23) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":85:23) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":86:23) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":88:16) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":88:20) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":88:40) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":88:9) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":89:28) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":89:9) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":91:23) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":92:21) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":92:28) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":93:21) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":93:28) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":94:21) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":94:28) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":95:21) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":95:28) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":98:29) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":98:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":98:51) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":98:35) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":99:51) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":99:35) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":100:51) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":100:35) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":101:51) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":101:35) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":102:19) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":102:51) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":103:19) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":103:51) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":104:19) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":104:51) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":105:19) +#loc47 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":105:51) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":109:30) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":109:19) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":109:36) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":110:36) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":111:36) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":112:36) +#loc54 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":114:23) +#loc55 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":115:23) +#loc56 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":116:23) +#loc57 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":117:23) +#loc58 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":138:38) +#loc59 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:33) +#loc60 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:45) +#loc61 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:19) +#loc62 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:52) +#loc63 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":140:40) +#loc64 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":140:52) +#loc65 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":144:19) +#loc66 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":140:23) +#loc67 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:56) +#loc68 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":140:67) +#loc69 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":138:25) +#loc70 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":137:21) +#loc71 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":142:45) +#loc72 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":142:38) +#loc73 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":138:20) +#loc74 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":144:11) +#loc75 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":145:37) +#loc76 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":145:49) +#loc77 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":145:23) +#loc78 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":145:56) +#loc79 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":146:44) +#loc80 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":146:56) +#loc81 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":146:27) +#loc82 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":148:42) +#loc83 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":148:49) +#loc84 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":150:49) +#loc85 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":152:23) +#loc86 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":152:15) +#loc87 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":153:41) +#loc88 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":153:53) +#loc89 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":153:27) +#loc90 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":153:60) +#loc91 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":154:48) +#loc92 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":154:60) +#loc93 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":154:31) +#loc94 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":156:46) +#loc95 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":156:53) +#loc96 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":158:53) +#loc97 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":159:53) +#loc98 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":161:27) +#loc99 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":161:19) +#loc100 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":162:45) +#loc101 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":162:57) +#loc102 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":162:31) +#loc103 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":162:64) +#loc104 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":163:52) +#loc105 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":163:64) +#loc106 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":163:35) +#loc107 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":165:50) +#loc108 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":165:57) +#loc109 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":167:57) +#loc110 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":168:57) +#loc111 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":169:57) +#loc112 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":179:14) +#loc113 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":179:29) +#loc114 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":179:25) +#loc115 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":180:26) +#loc116 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:38) +#loc117 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:32) +#loc118 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:55) +#loc119 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:49) +#loc120 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:75) +#loc121 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:91) +#loc122 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:86) +#loc123 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:33) +#loc124 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:70) +#loc125 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:102) +#loc126 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:17) +#loc127 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:38) +#loc128 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:32) +#loc129 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:55) +#loc130 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:49) +#loc131 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:75) +#loc132 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:91) +#loc133 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:86) +#loc134 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:70) +#loc135 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:102) +#loc136 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:17) +#loc137 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:38) +#loc138 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:32) +#loc139 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:55) +#loc140 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:49) +#loc141 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:75) +#loc142 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:91) +#loc143 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:86) +#loc144 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:70) +#loc145 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:102) +#loc146 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:17) +#loc147 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:38) +#loc148 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:32) +#loc149 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:55) +#loc150 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:49) +#loc151 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:75) +#loc152 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:91) +#loc153 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:86) +#loc154 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:70) +#loc155 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:102) +#loc156 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:17) +#loc157 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:43) +#loc158 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:80) +#loc159 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:64) +#loc160 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:96) +#loc161 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:17) +#loc162 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:43) +#loc163 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:80) +#loc164 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:64) +#loc165 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:96) +#loc166 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:17) +#loc167 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:43) +#loc168 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:80) +#loc169 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:64) +#loc170 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:96) +#loc171 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:17) +#loc172 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:43) +#loc173 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:80) +#loc174 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:64) +#loc175 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:96) +#loc176 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:17) +#loc177 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:43) +#loc178 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:80) +#loc179 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:64) +#loc180 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:96) +#loc181 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:17) +#loc182 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:43) +#loc183 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:80) +#loc184 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:64) +#loc185 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:96) +#loc186 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:17) +#loc187 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":201:25) +#loc188 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":201:20) +#loc189 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":202:25) +#loc190 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":202:20) +#loc191 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":203:25) +#loc192 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":203:20) +#loc193 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":204:25) +#loc194 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":204:20) +#loc195 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":207:20) +#loc196 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":208:20) +#loc197 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":209:20) +#loc198 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":210:20) +#loc199 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":211:20) +#loc200 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":212:20) +#loc201 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":222:14) +#loc202 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":223:14) +#loc203 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":224:14) +#loc204 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":225:14) +#loc205 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":227:34) +#loc206 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":227:30) +#loc207 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":227:22) +#loc208 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":228:40) +#loc209 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":228:43) +#loc210 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":228:61) +#loc211 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc213 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc214 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":229:31) +#loc215 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":229:41) +#loc216 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":230:37) +#loc217 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":230:48) +#loc219 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":230:24) +#loc220 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":231:54) +#loc221 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":231:8) +#loc222 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":232:34) +#loc223 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":232:30) +#loc224 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":232:22) +#loc225 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":233:40) +#loc226 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":233:43) +#loc227 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":233:61) +#loc229 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":234:31) +#loc230 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":234:41) +#loc231 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":235:37) +#loc232 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":235:48) +#loc234 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":235:24) +#loc235 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":236:54) +#loc236 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":236:8) +#loc237 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":237:34) +#loc238 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":237:30) +#loc239 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":237:22) +#loc240 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":238:40) +#loc241 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":238:43) +#loc242 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":238:61) +#loc244 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":239:31) +#loc245 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":239:41) +#loc246 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":240:37) +#loc247 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":240:48) +#loc249 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":240:24) +#loc250 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":241:54) +#loc251 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":241:8) +#loc252 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":242:34) +#loc253 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":242:30) +#loc254 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":242:22) +#loc255 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":243:40) +#loc256 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":243:43) +#loc257 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":243:61) +#loc259 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":244:31) +#loc260 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":244:41) +#loc261 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":245:37) +#loc262 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":245:48) +#loc264 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":245:24) +#loc265 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":246:54) +#loc266 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":246:8) +#loc267 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":248:14) +#loc268 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":249:14) +#loc269 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":250:14) +#loc270 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":251:14) +#loc271 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":258:23) +#loc272 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":259:8) +#loc273 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":257:14) +#loc274 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":263:23) +#loc275 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":264:8) +#loc276 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":262:14) +#loc277 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":268:23) +#loc278 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":269:8) +#loc279 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":267:14) +#loc280 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":275:22) +#loc281 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":276:22) +#loc282 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":275:8) +#loc283 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":273:14) +#loc284 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":281:22) +#loc285 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":282:22) +#loc286 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":281:8) +#loc287 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":279:14) +#loc288 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":287:22) +#loc289 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":288:22) +#loc290 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":289:22) +#loc291 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":287:8) +#loc292 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":285:14) +#loc293 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":297:30) +#loc294 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":297:42) +#loc295 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":297:16) +#loc296 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":297:51) +#loc297 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":298:30) +#loc298 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":298:42) +#loc299 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":298:16) +#loc300 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":298:51) +#loc301 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":299:57) +#loc302 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":299:62) +#loc303 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":299:51) +#loc304 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":300:30) +#loc305 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":300:42) +#loc306 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":300:16) +#loc307 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":300:51) +#loc308 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":301:51) +#loc309 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":302:59) +#loc310 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":302:64) +#loc311 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":302:51) +#loc312 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":303:30) +#loc313 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":303:42) +#loc314 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":303:16) +#loc315 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":303:51) +#loc316 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":304:51) +#loc317 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":305:51) +#loc318 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":306:59) +#loc319 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":306:64) +#loc320 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":306:51) +#loc321 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":308:44) +#loc322 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":308:29) +#loc323 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":309:29) +#loc324 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":310:29) +#loc325 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":311:29) +#loc326 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":312:52) +#loc327 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":312:30) +#loc328 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":313:30) +#loc329 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":314:54) +#loc330 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":314:30) +#loc331 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":315:30) +#loc332 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":316:30) +#loc333 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":317:54) +#loc334 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":317:30) +#loc335 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":319:30) +#loc336 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":319:20) +#loc337 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":320:30) +#loc338 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":320:20) +#loc339 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":321:30) +#loc340 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":321:20) +#loc341 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":322:30) +#loc342 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":322:20) +#loc343 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":323:30) +#loc344 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":323:20) +#loc345 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":324:30) +#loc346 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":324:20) +#loc347 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":325:30) +#loc348 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":325:20) +#loc349 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":326:30) +#loc350 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":326:20) +#loc351 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":327:30) +#loc352 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":327:20) +#loc353 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":328:30) +#loc354 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":328:20) +#loc355 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":328:4) +#loc361 = loc("i_tc1"(#loc12)) +#loc362 = loc("i_tc2"(#loc13)) +#loc363 = loc("i_tc3"(#loc14)) +#loc364 = loc("k"(#loc15)) +#loc365 = loc("k"(#loc16)) +#loc366 = loc("k"(#loc17)) +#loc367 = loc("k"(#loc18)) +#loc368 = loc("A"(#loc19)) +#loc369 = loc("A"(#loc20)) +#loc370 = loc("o_i"(#loc21)) +#loc371 = loc("m_tc0"(#loc22)) +#loc372 = loc("m_tc0"(#loc23)) +#loc373 = loc("m_tc1"(#loc24)) +#loc374 = loc("m_tc1"(#loc25)) +#loc375 = loc("m_tc2"(#loc26)) +#loc376 = loc("m_tc2"(#loc27)) +#loc377 = loc("m_tc3"(#loc28)) +#loc378 = loc("m_tc3"(#loc29)) +#loc379 = loc("p_b0"(#loc30)) +#loc380 = loc("p_b0"(#loc31)) +#loc381 = loc("p_b0"(#loc32)) +#loc382 = loc("p_b0"(#loc33)) +#loc383 = loc("p_b1"(#loc34)) +#loc384 = loc("p_b1"(#loc35)) +#loc385 = loc("p_b2"(#loc36)) +#loc386 = loc("p_b2"(#loc37)) +#loc387 = loc("p_b3"(#loc38)) +#loc388 = loc("p_b3"(#loc39)) +#loc389 = loc("b_b0"(#loc40)) +#loc390 = loc("b_b0"(#loc41)) +#loc391 = loc("b_b1"(#loc42)) +#loc392 = loc("b_b1"(#loc43)) +#loc393 = loc("b_b2"(#loc44)) +#loc394 = loc("b_b2"(#loc45)) +#loc395 = loc("b_b3"(#loc46)) +#loc396 = loc("b_b3"(#loc47)) +#loc397 = loc("p_g0"(#loc48)) +#loc398 = loc("p_g0"(#loc49)) +#loc399 = loc("p_g0"(#loc50)) +#loc400 = loc("p_g1"(#loc51)) +#loc401 = loc("p_g2"(#loc52)) +#loc402 = loc("p_g3"(#loc53)) +#loc403 = loc("b_g0"(#loc54)) +#loc404 = loc("b_g1"(#loc55)) +#loc405 = loc("b_g2"(#loc56)) +#loc406 = loc("b_g3"(#loc57)) +#loc407 = loc("o_k"(#loc58)) +#loc408 = loc("p_k0"(#loc59)) +#loc409 = loc("p_k0"(#loc60)) +#loc410 = loc("p_k0"(#loc61)) +#loc411 = loc("p_k0"(#loc62)) +#loc412 = loc("b_k0"(#loc63)) +#loc413 = loc("b_k0"(#loc64)) +#loc414 = loc("b_k0"(#loc66)) +#loc415 = loc("p_k0"(#loc67)) +#loc416 = loc("b_k0"(#loc68)) +#loc417 = loc("o_k"(#loc69)) +#loc418 = loc("b_A00"(#loc70)) +#loc419 = loc("b_A00"(#loc71)) +#loc420 = loc("b_A00"(#loc72)) +#loc421 = loc("o_k"(#loc73)) +#loc422 = loc("p_k1"(#loc75)) +#loc423 = loc("p_k1"(#loc76)) +#loc424 = loc("p_k1"(#loc77)) +#loc425 = loc("p_k1"(#loc78)) +#loc426 = loc("b_k1"(#loc79)) +#loc427 = loc("b_k1"(#loc80)) +#loc428 = loc("b_k1"(#loc81)) +#loc429 = loc("b_A11"(#loc82)) +#loc430 = loc("b_A11"(#loc83)) +#loc431 = loc("b_A10"(#loc84)) +#loc432 = loc("b_A22"(#loc86)) +#loc433 = loc("p_k2"(#loc87)) +#loc434 = loc("p_k2"(#loc88)) +#loc435 = loc("p_k2"(#loc89)) +#loc436 = loc("p_k2"(#loc90)) +#loc437 = loc("b_k2"(#loc91)) +#loc438 = loc("b_k2"(#loc92)) +#loc439 = loc("b_k2"(#loc93)) +#loc440 = loc("b_A22"(#loc94)) +#loc441 = loc("b_A22"(#loc95)) +#loc442 = loc("b_A20"(#loc96)) +#loc443 = loc("b_A21"(#loc97)) +#loc444 = loc("b_A33"(#loc99)) +#loc445 = loc("p_k3"(#loc100)) +#loc446 = loc("p_k3"(#loc101)) +#loc447 = loc("p_k3"(#loc102)) +#loc448 = loc("p_k3"(#loc103)) +#loc449 = loc("b_k3"(#loc104)) +#loc450 = loc("b_k3"(#loc105)) +#loc451 = loc("b_k3"(#loc106)) +#loc452 = loc("b_A33"(#loc107)) +#loc453 = loc("b_A33"(#loc108)) +#loc454 = loc("b_A30"(#loc109)) +#loc455 = loc("b_A31"(#loc110)) +#loc456 = loc("b_A32"(#loc111)) +#loc457 = loc("m_d"(#loc112)) +#loc458 = loc("m_d"(#loc113)) +#loc459 = loc("m_d"(#loc114)) +#loc460 = loc("m_I"(#loc115)) +#loc461 = loc("b_A00"(#loc116)) +#loc462 = loc("b_A00"(#loc117)) +#loc463 = loc("b_A00"(#loc118)) +#loc464 = loc("b_A00"(#loc119)) +#loc465 = loc("b_A00"(#loc120)) +#loc466 = loc("b_A00"(#loc121)) +#loc467 = loc("b_A00"(#loc122)) +#loc468 = loc("b_A00"(#loc124)) +#loc469 = loc("b_A00"(#loc125)) +#loc470 = loc("b_A00"(#loc126)) +#loc471 = loc("b_A11"(#loc127)) +#loc472 = loc("b_A11"(#loc128)) +#loc473 = loc("b_A11"(#loc129)) +#loc474 = loc("b_A11"(#loc130)) +#loc475 = loc("b_A11"(#loc131)) +#loc476 = loc("b_A11"(#loc132)) +#loc477 = loc("b_A11"(#loc133)) +#loc478 = loc("b_A11"(#loc134)) +#loc479 = loc("b_A11"(#loc135)) +#loc480 = loc("b_A11"(#loc136)) +#loc481 = loc("b_A22"(#loc137)) +#loc482 = loc("b_A22"(#loc138)) +#loc483 = loc("b_A22"(#loc139)) +#loc484 = loc("b_A22"(#loc140)) +#loc485 = loc("b_A22"(#loc141)) +#loc486 = loc("b_A22"(#loc142)) +#loc487 = loc("b_A22"(#loc143)) +#loc488 = loc("b_A22"(#loc144)) +#loc489 = loc("b_A22"(#loc145)) +#loc490 = loc("b_A22"(#loc146)) +#loc491 = loc("b_A33"(#loc147)) +#loc492 = loc("b_A33"(#loc148)) +#loc493 = loc("b_A33"(#loc149)) +#loc494 = loc("b_A33"(#loc150)) +#loc495 = loc("b_A33"(#loc151)) +#loc496 = loc("b_A33"(#loc152)) +#loc497 = loc("b_A33"(#loc153)) +#loc498 = loc("b_A33"(#loc154)) +#loc499 = loc("b_A33"(#loc155)) +#loc500 = loc("b_A33"(#loc156)) +#loc501 = loc("b_A10"(#loc157)) +#loc502 = loc("b_A10"(#loc158)) +#loc503 = loc("b_A10"(#loc159)) +#loc504 = loc("b_A10"(#loc160)) +#loc505 = loc("b_A10"(#loc161)) +#loc506 = loc("b_A20"(#loc162)) +#loc507 = loc("b_A20"(#loc163)) +#loc508 = loc("b_A20"(#loc164)) +#loc509 = loc("b_A20"(#loc165)) +#loc510 = loc("b_A20"(#loc166)) +#loc511 = loc("b_A21"(#loc167)) +#loc512 = loc("b_A21"(#loc168)) +#loc513 = loc("b_A21"(#loc169)) +#loc514 = loc("b_A21"(#loc170)) +#loc515 = loc("b_A21"(#loc171)) +#loc516 = loc("b_A30"(#loc172)) +#loc517 = loc("b_A30"(#loc173)) +#loc518 = loc("b_A30"(#loc174)) +#loc519 = loc("b_A30"(#loc175)) +#loc520 = loc("b_A30"(#loc176)) +#loc521 = loc("b_A31"(#loc177)) +#loc522 = loc("b_A31"(#loc178)) +#loc523 = loc("b_A31"(#loc179)) +#loc524 = loc("b_A31"(#loc180)) +#loc525 = loc("b_A31"(#loc181)) +#loc526 = loc("b_A32"(#loc182)) +#loc527 = loc("b_A32"(#loc183)) +#loc528 = loc("b_A32"(#loc184)) +#loc529 = loc("b_A32"(#loc185)) +#loc530 = loc("b_A32"(#loc186)) +#loc531 = loc("b_A00"(#loc187)) +#loc532 = loc("b_A00"(#loc188)) +#loc533 = loc("b_A11"(#loc189)) +#loc534 = loc("b_A11"(#loc190)) +#loc535 = loc("b_A22"(#loc191)) +#loc536 = loc("b_A22"(#loc192)) +#loc537 = loc("b_A33"(#loc193)) +#loc538 = loc("b_A33"(#loc194)) +#loc539 = loc("b_A10"(#loc195)) +#loc540 = loc("b_A20"(#loc196)) +#loc541 = loc("b_A21"(#loc197)) +#loc542 = loc("b_A30"(#loc198)) +#loc543 = loc("b_A31"(#loc199)) +#loc544 = loc("b_A32"(#loc200)) +#loc545 = loc("b_Ai00"(#loc201)) +#loc546 = loc("b_Ai11"(#loc202)) +#loc547 = loc("b_Ai22"(#loc203)) +#loc548 = loc("b_Ai33"(#loc204)) +#loc549 = loc("b_Ai00"(#loc207)) +#loc550 = loc("b_a00"(#loc208)) +#loc551 = loc("b_a00"(#loc209)) +#loc552 = loc("b_a00"(#loc210)) +#loc554 = loc("b_a00"(#loc214)) +#loc555 = loc("b_a00"(#loc215)) +#loc556 = loc("b_a00"(#loc216)) +#loc557 = loc("b_a00"(#loc217)) +#loc559 = loc("b_a00"(#loc219)) +#loc560 = loc("b_Ai00"(#loc220)) +#loc561 = loc("b_Ai11"(#loc224)) +#loc562 = loc("b_a11"(#loc225)) +#loc563 = loc("b_a11"(#loc226)) +#loc564 = loc("b_a11"(#loc227)) +#loc566 = loc("b_a11"(#loc229)) +#loc567 = loc("b_a11"(#loc230)) +#loc568 = loc("b_a11"(#loc231)) +#loc569 = loc("b_a11"(#loc232)) +#loc571 = loc("b_a11"(#loc234)) +#loc572 = loc("b_Ai11"(#loc235)) +#loc573 = loc("b_Ai22"(#loc239)) +#loc574 = loc("b_a22"(#loc240)) +#loc575 = loc("b_a22"(#loc241)) +#loc576 = loc("b_a22"(#loc242)) +#loc578 = loc("b_a22"(#loc244)) +#loc579 = loc("b_a22"(#loc245)) +#loc580 = loc("b_a22"(#loc246)) +#loc581 = loc("b_a22"(#loc247)) +#loc583 = loc("b_a22"(#loc249)) +#loc584 = loc("b_Ai22"(#loc250)) +#loc585 = loc("b_Ai33"(#loc254)) +#loc586 = loc("b_a33"(#loc255)) +#loc587 = loc("b_a33"(#loc256)) +#loc588 = loc("b_a33"(#loc257)) +#loc590 = loc("b_a33"(#loc259)) +#loc591 = loc("b_a33"(#loc260)) +#loc592 = loc("b_a33"(#loc261)) +#loc593 = loc("b_a33"(#loc262)) +#loc595 = loc("b_a33"(#loc264)) +#loc596 = loc("b_Ai33"(#loc265)) +#loc597 = loc("b_Ai00"(#loc267)) +#loc598 = loc("b_Ai11"(#loc268)) +#loc599 = loc("b_Ai22"(#loc269)) +#loc600 = loc("b_Ai33"(#loc270)) +#loc601 = loc("b_Ai10"(#loc271)) +#loc602 = loc("b_Ai10"(#loc272)) +#loc603 = loc("b_Ai10"(#loc273)) +#loc604 = loc("b_Ai21"(#loc274)) +#loc605 = loc("b_Ai21"(#loc275)) +#loc606 = loc("b_Ai21"(#loc276)) +#loc607 = loc("b_Ai32"(#loc277)) +#loc608 = loc("b_Ai32"(#loc278)) +#loc609 = loc("b_Ai32"(#loc279)) +#loc610 = loc("b_Ai20"(#loc280)) +#loc611 = loc("b_Ai20"(#loc281)) +#loc612 = loc("b_Ai20"(#loc282)) +#loc613 = loc("b_Ai20"(#loc283)) +#loc614 = loc("b_Ai31"(#loc284)) +#loc615 = loc("b_Ai31"(#loc285)) +#loc616 = loc("b_Ai31"(#loc286)) +#loc617 = loc("b_Ai31"(#loc287)) +#loc618 = loc("b_Ai30"(#loc288)) +#loc619 = loc("b_Ai30"(#loc289)) +#loc620 = loc("b_Ai30"(#loc290)) +#loc621 = loc("b_Ai30"(#loc291)) +#loc622 = loc("b_Ai30"(#loc292)) +#loc623 = loc("p_A00"(#loc293)) +#loc624 = loc("p_A00"(#loc294)) +#loc625 = loc("p_A00"(#loc295)) +#loc626 = loc("p_A00"(#loc296)) +#loc627 = loc("p_A10"(#loc297)) +#loc628 = loc("p_A10"(#loc298)) +#loc629 = loc("p_A10"(#loc299)) +#loc630 = loc("p_A10"(#loc300)) +#loc631 = loc("p_A11"(#loc301)) +#loc632 = loc("p_A11"(#loc302)) +#loc633 = loc("p_A11"(#loc303)) +#loc634 = loc("p_A20"(#loc304)) +#loc635 = loc("p_A20"(#loc305)) +#loc636 = loc("p_A20"(#loc306)) +#loc637 = loc("p_A20"(#loc307)) +#loc638 = loc("p_A21"(#loc308)) +#loc639 = loc("p_A22"(#loc309)) +#loc640 = loc("p_A22"(#loc310)) +#loc641 = loc("p_A22"(#loc311)) +#loc642 = loc("p_A30"(#loc312)) +#loc643 = loc("p_A30"(#loc313)) +#loc644 = loc("p_A30"(#loc314)) +#loc645 = loc("p_A30"(#loc315)) +#loc646 = loc("p_A31"(#loc316)) +#loc647 = loc("p_A32"(#loc317)) +#loc648 = loc("p_A33"(#loc318)) +#loc649 = loc("p_A33"(#loc319)) +#loc650 = loc("p_A33"(#loc320)) +#loc651 = loc("m_A0"(#loc321)) +#loc652 = loc("m_A0"(#loc322)) +#loc653 = loc("m_A1"(#loc323)) +#loc654 = loc("m_A2"(#loc324)) +#loc655 = loc("m_A3"(#loc325)) +#loc656 = loc("m_A11"(#loc326)) +#loc657 = loc("m_A11"(#loc327)) +#loc658 = loc("m_A21"(#loc328)) +#loc659 = loc("m_A22"(#loc329)) +#loc660 = loc("m_A22"(#loc330)) +#loc661 = loc("m_A31"(#loc331)) +#loc662 = loc("m_A32"(#loc332)) +#loc663 = loc("m_A33"(#loc333)) +#loc664 = loc("m_A33"(#loc334)) +#loc665 = loc(fused[#loc379, #loc380]) +#loc666 = loc(fused[#loc397, #loc398]) +#loc667 = loc("b_A11"(#loc418)) +#loc668 = loc("b_A11"(#loc430)) +#loc669 = loc("b_A10"(#loc431)) +#loc670 = loc("b_A33"(#loc432)) +#loc671 = loc("b_A22"(#loc441)) +#loc672 = loc("b_A20"(#loc442)) +#loc673 = loc("b_A21"(#loc443)) +#loc674 = loc("b_A30"(#loc444)) +#loc675 = loc("b_A33"(#loc453)) +#loc676 = loc("b_A30"(#loc454)) +#loc677 = loc("b_A31"(#loc455)) +#loc678 = loc("b_A32"(#loc456)) +#loc679 = loc(callsite(#loc123 at #loc468)) +#loc680 = loc(callsite(#loc123 at #loc478)) +#loc681 = loc(callsite(#loc123 at #loc488)) +#loc682 = loc(callsite(#loc123 at #loc498)) +#loc683 = loc(callsite(#loc123 at #loc503)) +#loc684 = loc(callsite(#loc123 at #loc508)) +#loc685 = loc(callsite(#loc123 at #loc513)) +#loc686 = loc(callsite(#loc123 at #loc518)) +#loc687 = loc(callsite(#loc123 at #loc523)) +#loc688 = loc(callsite(#loc123 at #loc528)) +#loc689 = loc(callsite(#loc211 at #loc553)) +#loc691 = loc(callsite(#loc211 at #loc558)) +#loc693 = loc(callsite(#loc211 at #loc565)) +#loc695 = loc(callsite(#loc211 at #loc570)) +#loc697 = loc(callsite(#loc211 at #loc577)) +#loc699 = loc(callsite(#loc211 at #loc582)) +#loc701 = loc(callsite(#loc211 at #loc589)) +#loc703 = loc(callsite(#loc211 at #loc594)) +#loc705 = loc("b_A22"(#loc667)) +#loc706 = loc("b_A20"(#loc670)) +#loc707 = loc("b_A31"(#loc674)) +#loc708 = loc(callsite(#loc213 at #loc689)) +#loc709 = loc(callsite(#loc213 at #loc691)) +#loc710 = loc(callsite(#loc213 at #loc693)) +#loc711 = loc(callsite(#loc213 at #loc695)) +#loc712 = loc(callsite(#loc213 at #loc697)) +#loc713 = loc(callsite(#loc213 at #loc699)) +#loc714 = loc(callsite(#loc213 at #loc701)) +#loc715 = loc(callsite(#loc213 at #loc703)) +#loc716 = loc("b_A33"(#loc705)) +#loc717 = loc("b_A21"(#loc706)) +#loc718 = loc("b_A32"(#loc707)) +#loc719 = loc("b_A10"(#loc716)) +#loc720 = loc("b_A30"(#loc717)) +#loc721 = loc("b_A20"(#loc719)) +#loc722 = loc("b_A31"(#loc720)) +#loc723 = loc("b_A21"(#loc721)) +#loc724 = loc("b_A32"(#loc722)) +#loc725 = loc("b_A30"(#loc723)) +#loc726 = loc("b_A31"(#loc725)) +#loc727 = loc("b_A32"(#loc726)) diff --git a/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.ttir b/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..2e1d46d0c4277b493b19c13f2d653c6df4e97f5a --- /dev/null +++ b/triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.ttir @@ -0,0 +1,1234 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":40:0) +#loc1 = loc(unknown) +#loc213 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":228:66) +#loc219 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":230:56) +#loc229 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":233:66) +#loc234 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":235:56) +#loc244 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":238:66) +#loc249 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":240:56) +#loc259 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":243:66) +#loc264 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":245:56) +#loc357 = loc("k"(#loc)) +#loc358 = loc("g"(#loc)) +#loc359 = loc("beta"(#loc)) +#loc360 = loc("A"(#loc)) +#loc361 = loc("T"(#loc)) +#loc554 = loc("b_a00"(#loc213)) +#loc559 = loc("b_a00"(#loc219)) +#loc566 = loc("b_a11"(#loc229)) +#loc571 = loc("b_a11"(#loc234)) +#loc578 = loc("b_a22"(#loc244)) +#loc583 = loc("b_a22"(#loc249)) +#loc590 = loc("b_a33"(#loc259)) +#loc595 = loc("b_a33"(#loc264)) +#loc691 = loc(callsite(#loc1 at #loc554)) +#loc693 = loc(callsite(#loc1 at #loc559)) +#loc695 = loc(callsite(#loc1 at #loc566)) +#loc697 = loc(callsite(#loc1 at #loc571)) +#loc699 = loc(callsite(#loc1 at #loc578)) +#loc701 = loc(callsite(#loc1 at #loc583)) +#loc703 = loc(callsite(#loc1 at #loc590)) +#loc705 = loc(callsite(#loc1 at #loc595)) +module { + tt.func public @chunk_gated_delta_rule_fwd_kkt_solve_kernel(%k: !tt.ptr {tt.divisibility = 16 : i32} loc("k"(#loc)), %g: !tt.ptr {tt.divisibility = 16 : i32} loc("g"(#loc)), %beta: !tt.ptr {tt.divisibility = 16 : i32} loc("beta"(#loc)), %A: !tt.ptr {tt.divisibility = 16 : i32} loc("A"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %c1_i64 = arith.constant 1 : i64 loc(#loc1) + %c2_i64 = arith.constant 2 : i64 loc(#loc1) + %cst = arith.constant dense<0.000000e+00> : tensor<16x64xbf16> loc(#loc1) + %c1_i32 = arith.constant 1 : i32 loc(#loc1) + %c2_i32 = arith.constant 2 : i32 loc(#loc1) + %c0_i32 = arith.constant 0 : i32 loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<16xbf16> loc(#loc1) + %cst_1 = arith.constant dense<64> : tensor<1x16xi32> loc(#loc1) + %cst_2 = arith.constant dense<48> : tensor<16xi32> loc(#loc1) + %cst_3 = arith.constant dense<32> : tensor<16xi32> loc(#loc1) + %cst_4 = arith.constant dense<16> : tensor<16xi32> loc(#loc1) + %cst_5 = arith.constant dense<2048> : tensor<16x1xi64> loc(#loc1) + %cst_6 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc1) + %cst_7 = arith.constant dense<128> : tensor<1x64xi32> loc(#loc1) + %cst_8 = arith.constant dense<4096> : tensor<16x1xi64> loc(#loc1) + %cst_9 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc1) + %cst_10 = arith.constant dense<32> : tensor<16xi64> loc(#loc1) + %c128_i64 = arith.constant 128 : i64 loc(#loc1) + %c48_i64 = arith.constant 48 : i64 loc(#loc1) + %c16_i64 = arith.constant 16 : i64 loc(#loc1) + %c64_i64 = arith.constant 64 : i64 loc(#loc1) + %c64_i32 = arith.constant 64 : i32 loc(#loc1) + %c32_i64 = arith.constant 32 : i64 loc(#loc1) + %0 = tt.get_program_id x : i32 loc(#loc2) + %1 = arith.extsi %0 : i32 to i64 loc(#loc3) + %2 = tt.get_program_id y : i32 loc(#loc4) + %3 = arith.extsi %2 : i32 to i64 loc(#loc5) + %4 = arith.divsi %3, %c32_i64 : i64 loc(#loc6) + %5 = arith.remsi %3, %c32_i64 : i64 loc(#loc7) + %6 = arith.extsi %T : i32 to i64 loc(#loc8) + %7 = arith.muli %4, %6 : i64 loc(#loc8) + %8 = arith.muli %1, %c64_i64 : i64 loc(#loc9) + %9 = arith.cmpi sge, %8, %6 : i64 loc(#loc10) + cf.cond_br %9, ^bb1, ^bb2 loc(#loc10) + ^bb1: // pred: ^bb0 + tt.return loc(#loc11) + ^bb2: // pred: ^bb0 + %i_tc1 = arith.addi %8, %c16_i64 : i64 loc(#loc362) + %i_tc2 = arith.addi %8, %c32_i64 : i64 loc(#loc363) + %i_tc3 = arith.addi %8, %c48_i64 : i64 loc(#loc364) + %k_11 = arith.muli %7, %c32_i64 : i64 loc(#loc365) + %k_12 = arith.addi %k_11, %5 : i64 loc(#loc366) + %k_13 = arith.muli %k_12, %c128_i64 : i64 loc(#loc367) + %k_14 = tt.addptr %k, %k_13 : !tt.ptr, i64 loc(#loc368) + %A_15 = arith.muli %k_12, %c64_i64 : i64 loc(#loc369) + %A_16 = tt.addptr %A, %A_15 : !tt.ptr, i64 loc(#loc370) + %o_i = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32> loc(#loc371) + %m_tc0 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc372) + %m_tc0_17 = tt.splat %8 : i64 -> tensor<16xi64> loc(#loc372) + %m_tc0_18 = arith.addi %m_tc0_17, %m_tc0 : tensor<16xi64> loc(#loc372) + %m_tc0_19 = tt.splat %6 : i64 -> tensor<16xi64> loc(#loc373) + %m_tc0_20 = arith.cmpi slt, %m_tc0_18, %m_tc0_19 : tensor<16xi64> loc(#loc373) + %m_tc1 = tt.splat %i_tc1 : i64 -> tensor<16xi64> loc(#loc374) + %m_tc1_21 = arith.addi %m_tc1, %m_tc0 : tensor<16xi64> loc(#loc374) + %m_tc1_22 = arith.cmpi slt, %m_tc1_21, %m_tc0_19 : tensor<16xi64> loc(#loc375) + %m_tc2 = tt.splat %i_tc2 : i64 -> tensor<16xi64> loc(#loc376) + %m_tc2_23 = arith.addi %m_tc2, %m_tc0 : tensor<16xi64> loc(#loc376) + %m_tc2_24 = arith.cmpi slt, %m_tc2_23, %m_tc0_19 : tensor<16xi64> loc(#loc377) + %m_tc3 = tt.splat %i_tc3 : i64 -> tensor<16xi64> loc(#loc378) + %m_tc3_25 = arith.addi %m_tc3, %m_tc0 : tensor<16xi64> loc(#loc378) + %m_tc3_26 = arith.cmpi slt, %m_tc3_25, %m_tc0_19 : tensor<16xi64> loc(#loc379) + %p_b0 = tt.addptr %beta, %k_12 : !tt.ptr, i64 loc(#loc666) + %p_b0_27 = arith.muli %m_tc0_18, %cst_10 : tensor<16xi64> loc(#loc382) + %p_b0_28 = tt.splat %p_b0 : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc383) + %p_b0_29 = tt.addptr %p_b0_28, %p_b0_27 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc383) + %p_b1 = arith.muli %m_tc1_21, %cst_10 : tensor<16xi64> loc(#loc384) + %p_b1_30 = tt.addptr %p_b0_28, %p_b1 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc385) + %p_b2 = arith.muli %m_tc2_23, %cst_10 : tensor<16xi64> loc(#loc386) + %p_b2_31 = tt.addptr %p_b0_28, %p_b2 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc387) + %p_b3 = arith.muli %m_tc3_25, %cst_10 : tensor<16xi64> loc(#loc388) + %p_b3_32 = tt.addptr %p_b0_28, %p_b3 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc389) + %b_b0 = tt.load %p_b0_29, %m_tc0_20, %cst_0 : tensor<16x!tt.ptr> loc(#loc390) + %b_b0_33 = arith.extf %b_b0 : tensor<16xbf16> to tensor<16xf32> loc(#loc391) + %b_b1 = tt.load %p_b1_30, %m_tc1_22, %cst_0 : tensor<16x!tt.ptr> loc(#loc392) + %b_b1_34 = arith.extf %b_b1 : tensor<16xbf16> to tensor<16xf32> loc(#loc393) + %b_b2 = tt.load %p_b2_31, %m_tc2_24, %cst_0 : tensor<16x!tt.ptr> loc(#loc394) + %b_b2_35 = arith.extf %b_b2 : tensor<16xbf16> to tensor<16xf32> loc(#loc395) + %b_b3 = tt.load %p_b3_32, %m_tc3_26, %cst_0 : tensor<16x!tt.ptr> loc(#loc396) + %b_b3_36 = arith.extf %b_b3 : tensor<16xbf16> to tensor<16xf32> loc(#loc397) + %p_g0 = tt.addptr %g, %k_12 : !tt.ptr, i64 loc(#loc667) + %p_g0_37 = tt.splat %p_g0 : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc400) + %p_g0_38 = tt.addptr %p_g0_37, %p_b0_27 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc400) + %p_g1 = tt.addptr %p_g0_37, %p_b1 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc401) + %p_g2 = tt.addptr %p_g0_37, %p_b2 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc402) + %p_g3 = tt.addptr %p_g0_37, %p_b3 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc403) + %b_g0 = tt.load %p_g0_38, %m_tc0_20, %cst_9 : tensor<16x!tt.ptr> loc(#loc404) + %b_g1 = tt.load %p_g1, %m_tc1_22, %cst_9 : tensor<16x!tt.ptr> loc(#loc405) + %b_g2 = tt.load %p_g2, %m_tc2_24, %cst_9 : tensor<16x!tt.ptr> loc(#loc406) + %b_g3 = tt.load %p_g3, %m_tc3_26, %cst_9 : tensor<16x!tt.ptr> loc(#loc407) + %b_A32:10 = scf.for %i_k = %c0_i32 to %c2_i32 step %c1_i32 iter_args(%b_A00_197 = %cst_6, %b_A11_198 = %cst_6, %b_A22_199 = %cst_6, %b_A33_200 = %cst_6, %b_A10_201 = %cst_6, %b_A20_202 = %cst_6, %b_A21_203 = %cst_6, %b_A30_204 = %cst_6, %b_A31_205 = %cst_6, %b_A32_206 = %cst_6) -> (tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>) : i32 { + %o_k = arith.muli %i_k, %c64_i32 : i32 loc(#loc409) + %o_k_207 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc410) + %o_k_208 = tt.splat %o_k : i32 -> tensor<64xi32> loc(#loc411) + %o_k_209 = arith.addi %o_k_208, %o_k_207 : tensor<64xi32> loc(#loc411) + %p_k0 = tt.expand_dims %m_tc0_18 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc412) + %p_k0_210 = arith.muli %p_k0, %cst_8 : tensor<16x1xi64> loc(#loc413) + %p_k0_211 = tt.splat %k_14 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc414) + %p_k0_212 = tt.addptr %p_k0_211, %p_k0_210 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc414) + %p_k0_213 = tt.expand_dims %o_k_209 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc415) + %p_k0_214 = tt.broadcast %p_k0_212 : tensor<16x1x!tt.ptr> -> tensor<16x64x!tt.ptr> loc(#loc416) + %p_k0_215 = tt.broadcast %p_k0_213 : tensor<1x64xi32> -> tensor<16x64xi32> loc(#loc416) + %p_k0_216 = tt.addptr %p_k0_214, %p_k0_215 : tensor<16x64x!tt.ptr>, tensor<16x64xi32> loc(#loc416) + %b_k0 = tt.expand_dims %m_tc0_20 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc417) + %b_k0_217 = arith.cmpi slt, %p_k0_213, %cst_7 : tensor<1x64xi32> loc(#loc418) + %b_k0_218 = tt.broadcast %b_k0 : tensor<16x1xi1> -> tensor<16x64xi1> loc(#loc419) + %b_k0_219 = tt.broadcast %b_k0_217 : tensor<1x64xi1> -> tensor<16x64xi1> loc(#loc419) + %b_k0_220 = arith.andi %b_k0_218, %b_k0_219 : tensor<16x64xi1> loc(#loc419) + %b_k0_221 = tt.load %p_k0_216, %b_k0_220, %cst : tensor<16x64x!tt.ptr> loc(#loc420) + %b_A00_222 = tt.trans %b_k0_221 {order = array} : tensor<16x64xbf16> -> tensor<64x16xbf16> loc(#loc421) + %b_A00_223 = tt.dot %b_k0_221, %b_A00_222, %b_A00_197, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc422) + %28 = arith.cmpi slt, %i_tc1, %6 : i64 loc(#loc73) + %29:9 = scf.if %28 -> (tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>) { + %p_k1 = tt.expand_dims %m_tc1_21 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc423) + %p_k1_224 = arith.muli %p_k1, %cst_8 : tensor<16x1xi64> loc(#loc424) + %p_k1_225 = tt.addptr %p_k0_211, %p_k1_224 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc425) + %p_k1_226 = tt.broadcast %p_k1_225 : tensor<16x1x!tt.ptr> -> tensor<16x64x!tt.ptr> loc(#loc426) + %p_k1_227 = tt.addptr %p_k1_226, %p_k0_215 : tensor<16x64x!tt.ptr>, tensor<16x64xi32> loc(#loc426) + %b_k1 = tt.expand_dims %m_tc1_22 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc427) + %b_k1_228 = tt.broadcast %b_k1 : tensor<16x1xi1> -> tensor<16x64xi1> loc(#loc428) + %b_k1_229 = arith.andi %b_k1_228, %b_k0_219 : tensor<16x64xi1> loc(#loc428) + %b_k1_230 = tt.load %p_k1_227, %b_k1_229, %cst : tensor<16x64x!tt.ptr> loc(#loc429) + %b_A11_231 = tt.trans %b_k1_230 {order = array} : tensor<16x64xbf16> -> tensor<64x16xbf16> loc(#loc430) + %b_A11_232 = tt.dot %b_k1_230, %b_A11_231, %b_A11_198, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc669) + %b_A10_233 = tt.dot %b_k1_230, %b_A00_222, %b_A10_201, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc670) + %30 = arith.cmpi slt, %i_tc2, %6 : i64 loc(#loc85) + %b_A32_234:7 = scf.if %30 -> (tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>) { + %p_k2 = tt.expand_dims %m_tc2_23 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc434) + %p_k2_235 = arith.muli %p_k2, %cst_8 : tensor<16x1xi64> loc(#loc435) + %p_k2_236 = tt.addptr %p_k0_211, %p_k2_235 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc436) + %p_k2_237 = tt.broadcast %p_k2_236 : tensor<16x1x!tt.ptr> -> tensor<16x64x!tt.ptr> loc(#loc437) + %p_k2_238 = tt.addptr %p_k2_237, %p_k0_215 : tensor<16x64x!tt.ptr>, tensor<16x64xi32> loc(#loc437) + %b_k2 = tt.expand_dims %m_tc2_24 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc438) + %b_k2_239 = tt.broadcast %b_k2 : tensor<16x1xi1> -> tensor<16x64xi1> loc(#loc439) + %b_k2_240 = arith.andi %b_k2_239, %b_k0_219 : tensor<16x64xi1> loc(#loc439) + %b_k2_241 = tt.load %p_k2_238, %b_k2_240, %cst : tensor<16x64x!tt.ptr> loc(#loc440) + %b_A22_242 = tt.trans %b_k2_241 {order = array} : tensor<16x64xbf16> -> tensor<64x16xbf16> loc(#loc441) + %b_A22_243 = tt.dot %b_k2_241, %b_A22_242, %b_A22_199, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc672) + %b_A20_244 = tt.dot %b_k2_241, %b_A00_222, %b_A20_202, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc673) + %b_A21_245 = tt.dot %b_k2_241, %b_A11_231, %b_A21_203, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc674) + %31 = arith.cmpi slt, %i_tc3, %6 : i64 loc(#loc98) + %b_A32_246:4 = scf.if %31 -> (tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>) { + %p_k3 = tt.expand_dims %m_tc3_25 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc446) + %p_k3_247 = arith.muli %p_k3, %cst_8 : tensor<16x1xi64> loc(#loc447) + %p_k3_248 = tt.addptr %p_k0_211, %p_k3_247 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc448) + %p_k3_249 = tt.broadcast %p_k3_248 : tensor<16x1x!tt.ptr> -> tensor<16x64x!tt.ptr> loc(#loc449) + %p_k3_250 = tt.addptr %p_k3_249, %p_k0_215 : tensor<16x64x!tt.ptr>, tensor<16x64xi32> loc(#loc449) + %b_k3 = tt.expand_dims %m_tc3_26 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc450) + %b_k3_251 = tt.broadcast %b_k3 : tensor<16x1xi1> -> tensor<16x64xi1> loc(#loc451) + %b_k3_252 = arith.andi %b_k3_251, %b_k0_219 : tensor<16x64xi1> loc(#loc451) + %b_k3_253 = tt.load %p_k3_250, %b_k3_252, %cst : tensor<16x64x!tt.ptr> loc(#loc452) + %b_A33_254 = tt.trans %b_k3_253 {order = array} : tensor<16x64xbf16> -> tensor<64x16xbf16> loc(#loc453) + %b_A33_255 = tt.dot %b_k3_253, %b_A33_254, %b_A33_200, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc676) + %b_A30_256 = tt.dot %b_k3_253, %b_A00_222, %b_A30_204, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc677) + %b_A31_257 = tt.dot %b_k3_253, %b_A11_231, %b_A31_205, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc678) + %b_A32_258 = tt.dot %b_k3_253, %b_A22_242, %b_A32_206, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc679) + scf.yield %b_A33_255, %b_A30_256, %b_A31_257, %b_A32_258 : tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32> loc(#loc679) + } else { + scf.yield %b_A33_200, %b_A30_204, %b_A31_205, %b_A32_206 : tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32> loc(#loc99) + } loc(#loc719) + scf.yield %b_A22_243, %b_A32_246#0, %b_A20_244, %b_A21_245, %b_A32_246#1, %b_A32_246#2, %b_A32_246#3 : tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32> loc(#loc719) + } else { + scf.yield %b_A22_199, %b_A33_200, %b_A20_202, %b_A21_203, %b_A30_204, %b_A31_205, %b_A32_206 : tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32> loc(#loc86) + } loc(#loc725) + scf.yield %b_A11_232, %b_A32_234#0, %b_A32_234#1, %b_A10_233, %b_A32_234#2, %b_A32_234#3, %b_A32_234#4, %b_A32_234#5, %b_A32_234#6 : tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32> loc(#loc725) + } else { + scf.yield %b_A11_198, %b_A22_199, %b_A33_200, %b_A10_201, %b_A20_202, %b_A21_203, %b_A30_204, %b_A31_205, %b_A32_206 : tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32> loc(#loc74) + } loc(#loc74) + scf.yield %b_A00_223, %29#0, %29#1, %29#2, %29#3, %29#4, %29#5, %29#6, %29#7, %29#8 : tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32> loc(#loc112) + } loc(#loc728) + %m_d = tt.expand_dims %o_i {axis = 1 : i32} : tensor<16xi32> -> tensor<16x1xi32> loc(#loc458) + %m_d_39 = tt.expand_dims %o_i {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc459) + %m_d_40 = tt.broadcast %m_d : tensor<16x1xi32> -> tensor<16x16xi32> loc(#loc460) + %m_d_41 = tt.broadcast %m_d_39 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc460) + %m_d_42 = arith.cmpi sgt, %m_d_40, %m_d_41 : tensor<16x16xi32> loc(#loc460) + %m_I = arith.cmpi eq, %m_d_40, %m_d_41 : tensor<16x16xi32> loc(#loc461) + %b_A00 = tt.expand_dims %m_tc0_20 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc462) + %b_A00_43 = tt.broadcast %b_A00 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc463) + %b_A00_44 = arith.andi %m_d_42, %b_A00_43 : tensor<16x16xi1> loc(#loc463) + %b_A00_45 = tt.expand_dims %m_tc0_20 {axis = 0 : i32} : tensor<16xi1> -> tensor<1x16xi1> loc(#loc464) + %b_A00_46 = tt.broadcast %b_A00_45 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc465) + %b_A00_47 = arith.andi %b_A00_44, %b_A00_46 : tensor<16x16xi1> loc(#loc465) + %b_A00_48 = tt.expand_dims %b_g0 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc466) + %b_A00_49 = tt.expand_dims %b_g0 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc467) + %b_A00_50 = tt.broadcast %b_A00_48 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc468) + %b_A00_51 = tt.broadcast %b_A00_49 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc468) + %b_A00_52 = arith.subf %b_A00_50, %b_A00_51 : tensor<16x16xf32> loc(#loc468) + %b_A00_53 = math.exp2 %b_A00_52 : tensor<16x16xf32> loc(#loc680) + %b_A00_54 = arith.select %b_A00_47, %b_A00_53, %cst_6 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc470) + %b_A00_55 = arith.mulf %b_A32#0, %b_A00_54 : tensor<16x16xf32> loc(#loc471) + %b_A11 = tt.expand_dims %m_tc1_22 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc472) + %b_A11_56 = tt.broadcast %b_A11 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc473) + %b_A11_57 = arith.andi %m_d_42, %b_A11_56 : tensor<16x16xi1> loc(#loc473) + %b_A11_58 = tt.expand_dims %m_tc1_22 {axis = 0 : i32} : tensor<16xi1> -> tensor<1x16xi1> loc(#loc474) + %b_A11_59 = tt.broadcast %b_A11_58 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc475) + %b_A11_60 = arith.andi %b_A11_57, %b_A11_59 : tensor<16x16xi1> loc(#loc475) + %b_A11_61 = tt.expand_dims %b_g1 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc476) + %b_A11_62 = tt.expand_dims %b_g1 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc477) + %b_A11_63 = tt.broadcast %b_A11_61 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc478) + %b_A11_64 = tt.broadcast %b_A11_62 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc478) + %b_A11_65 = arith.subf %b_A11_63, %b_A11_64 : tensor<16x16xf32> loc(#loc478) + %b_A11_66 = math.exp2 %b_A11_65 : tensor<16x16xf32> loc(#loc681) + %b_A11_67 = arith.select %b_A11_60, %b_A11_66, %cst_6 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc480) + %b_A11_68 = arith.mulf %b_A32#1, %b_A11_67 : tensor<16x16xf32> loc(#loc481) + %b_A22 = tt.expand_dims %m_tc2_24 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc482) + %b_A22_69 = tt.broadcast %b_A22 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc483) + %b_A22_70 = arith.andi %m_d_42, %b_A22_69 : tensor<16x16xi1> loc(#loc483) + %b_A22_71 = tt.expand_dims %m_tc2_24 {axis = 0 : i32} : tensor<16xi1> -> tensor<1x16xi1> loc(#loc484) + %b_A22_72 = tt.broadcast %b_A22_71 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc485) + %b_A22_73 = arith.andi %b_A22_70, %b_A22_72 : tensor<16x16xi1> loc(#loc485) + %b_A22_74 = tt.expand_dims %b_g2 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc486) + %b_A22_75 = tt.expand_dims %b_g2 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc487) + %b_A22_76 = tt.broadcast %b_A22_74 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc488) + %b_A22_77 = tt.broadcast %b_A22_75 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc488) + %b_A22_78 = arith.subf %b_A22_76, %b_A22_77 : tensor<16x16xf32> loc(#loc488) + %b_A22_79 = math.exp2 %b_A22_78 : tensor<16x16xf32> loc(#loc682) + %b_A22_80 = arith.select %b_A22_73, %b_A22_79, %cst_6 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc490) + %b_A22_81 = arith.mulf %b_A32#2, %b_A22_80 : tensor<16x16xf32> loc(#loc491) + %b_A33 = tt.expand_dims %m_tc3_26 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc492) + %b_A33_82 = tt.broadcast %b_A33 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc493) + %b_A33_83 = arith.andi %m_d_42, %b_A33_82 : tensor<16x16xi1> loc(#loc493) + %b_A33_84 = tt.expand_dims %m_tc3_26 {axis = 0 : i32} : tensor<16xi1> -> tensor<1x16xi1> loc(#loc494) + %b_A33_85 = tt.broadcast %b_A33_84 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc495) + %b_A33_86 = arith.andi %b_A33_83, %b_A33_85 : tensor<16x16xi1> loc(#loc495) + %b_A33_87 = tt.expand_dims %b_g3 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc496) + %b_A33_88 = tt.expand_dims %b_g3 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc497) + %b_A33_89 = tt.broadcast %b_A33_87 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc498) + %b_A33_90 = tt.broadcast %b_A33_88 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc498) + %b_A33_91 = arith.subf %b_A33_89, %b_A33_90 : tensor<16x16xf32> loc(#loc498) + %b_A33_92 = math.exp2 %b_A33_91 : tensor<16x16xf32> loc(#loc683) + %b_A33_93 = arith.select %b_A33_86, %b_A33_92, %cst_6 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc500) + %b_A33_94 = arith.mulf %b_A32#3, %b_A33_93 : tensor<16x16xf32> loc(#loc501) + %b_A10 = arith.andi %b_A11_56, %b_A00_46 : tensor<16x16xi1> loc(#loc502) + %b_A10_95 = arith.subf %b_A11_63, %b_A00_51 : tensor<16x16xf32> loc(#loc503) + %b_A10_96 = math.exp2 %b_A10_95 : tensor<16x16xf32> loc(#loc684) + %b_A10_97 = arith.select %b_A10, %b_A10_96, %cst_6 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc505) + %b_A10_98 = arith.mulf %b_A32#4, %b_A10_97 : tensor<16x16xf32> loc(#loc506) + %b_A20 = arith.andi %b_A22_69, %b_A00_46 : tensor<16x16xi1> loc(#loc507) + %b_A20_99 = arith.subf %b_A22_76, %b_A00_51 : tensor<16x16xf32> loc(#loc508) + %b_A20_100 = math.exp2 %b_A20_99 : tensor<16x16xf32> loc(#loc685) + %b_A20_101 = arith.select %b_A20, %b_A20_100, %cst_6 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc510) + %b_A20_102 = arith.mulf %b_A32#5, %b_A20_101 : tensor<16x16xf32> loc(#loc511) + %b_A21 = arith.andi %b_A22_69, %b_A11_59 : tensor<16x16xi1> loc(#loc512) + %b_A21_103 = arith.subf %b_A22_76, %b_A11_64 : tensor<16x16xf32> loc(#loc513) + %b_A21_104 = math.exp2 %b_A21_103 : tensor<16x16xf32> loc(#loc686) + %b_A21_105 = arith.select %b_A21, %b_A21_104, %cst_6 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc515) + %b_A21_106 = arith.mulf %b_A32#6, %b_A21_105 : tensor<16x16xf32> loc(#loc516) + %b_A30 = arith.andi %b_A33_82, %b_A00_46 : tensor<16x16xi1> loc(#loc517) + %b_A30_107 = arith.subf %b_A33_89, %b_A00_51 : tensor<16x16xf32> loc(#loc518) + %b_A30_108 = math.exp2 %b_A30_107 : tensor<16x16xf32> loc(#loc687) + %b_A30_109 = arith.select %b_A30, %b_A30_108, %cst_6 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc520) + %b_A30_110 = arith.mulf %b_A32#7, %b_A30_109 : tensor<16x16xf32> loc(#loc521) + %b_A31 = arith.andi %b_A33_82, %b_A11_59 : tensor<16x16xi1> loc(#loc522) + %b_A31_111 = arith.subf %b_A33_89, %b_A11_64 : tensor<16x16xf32> loc(#loc523) + %b_A31_112 = math.exp2 %b_A31_111 : tensor<16x16xf32> loc(#loc688) + %b_A31_113 = arith.select %b_A31, %b_A31_112, %cst_6 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc525) + %b_A31_114 = arith.mulf %b_A32#8, %b_A31_113 : tensor<16x16xf32> loc(#loc526) + %b_A32_115 = arith.andi %b_A33_82, %b_A22_72 : tensor<16x16xi1> loc(#loc527) + %b_A32_116 = arith.subf %b_A33_89, %b_A22_77 : tensor<16x16xf32> loc(#loc528) + %b_A32_117 = math.exp2 %b_A32_116 : tensor<16x16xf32> loc(#loc689) + %b_A32_118 = arith.select %b_A32_115, %b_A32_117, %cst_6 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc530) + %b_A32_119 = arith.mulf %b_A32#9, %b_A32_118 : tensor<16x16xf32> loc(#loc531) + %b_A00_120 = tt.expand_dims %b_b0_33 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc532) + %b_A00_121 = tt.broadcast %b_A00_120 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc533) + %b_A00_122 = arith.mulf %b_A00_55, %b_A00_121 : tensor<16x16xf32> loc(#loc533) + %b_A11_123 = tt.expand_dims %b_b1_34 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc534) + %b_A11_124 = tt.broadcast %b_A11_123 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc535) + %b_A11_125 = arith.mulf %b_A11_68, %b_A11_124 : tensor<16x16xf32> loc(#loc535) + %b_A22_126 = tt.expand_dims %b_b2_35 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc536) + %b_A22_127 = tt.broadcast %b_A22_126 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc537) + %b_A22_128 = arith.mulf %b_A22_81, %b_A22_127 : tensor<16x16xf32> loc(#loc537) + %b_A33_129 = tt.expand_dims %b_b3_36 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc538) + %b_A33_130 = tt.broadcast %b_A33_129 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc539) + %b_A33_131 = arith.mulf %b_A33_94, %b_A33_130 : tensor<16x16xf32> loc(#loc539) + %b_A10_132 = arith.mulf %b_A10_98, %b_A11_124 : tensor<16x16xf32> loc(#loc540) + %b_A20_133 = arith.mulf %b_A20_102, %b_A22_127 : tensor<16x16xf32> loc(#loc541) + %b_A21_134 = arith.mulf %b_A21_106, %b_A22_127 : tensor<16x16xf32> loc(#loc542) + %b_A30_135 = arith.mulf %b_A30_110, %b_A33_130 : tensor<16x16xf32> loc(#loc543) + %b_A31_136 = arith.mulf %b_A31_114, %b_A33_130 : tensor<16x16xf32> loc(#loc544) + %b_A32_137 = arith.mulf %b_A32_119, %b_A33_130 : tensor<16x16xf32> loc(#loc545) + %b_Ai00 = arith.subf %cst_6, %b_A00_122 : tensor<16x16xf32> loc(#loc546) + %b_Ai11 = arith.subf %cst_6, %b_A11_125 : tensor<16x16xf32> loc(#loc547) + %b_Ai22 = arith.subf %cst_6, %b_A22_128 : tensor<16x16xf32> loc(#loc548) + %b_Ai33 = arith.subf %cst_6, %b_A33_131 : tensor<16x16xf32> loc(#loc549) + %10 = arith.subi %6, %8 : i64 loc(#loc206) + %11 = arith.minsi %10, %c16_i64 : i64 loc(#loc207) + %b_Ai00_138 = scf.for %i = %c2_i64 to %11 step %c1_i64 iter_args(%b_Ai00_197 = %b_Ai00) -> (tensor<16x16xf32>) : i64 { + %b_a00 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc551) + %b_a00_198 = arith.cmpi eq, %m_tc0, %b_a00 : tensor<16xi64> loc(#loc551) + %b_a00_199 = tt.expand_dims %b_a00_198 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc552) + %b_a00_200 = tt.broadcast %b_a00_199 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc553) + %b_a00_201 = arith.select %b_a00_200, %b_Ai00, %cst_6 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc553) + %b_a00_202 = "tt.reduce"(%b_a00_201) <{axis = 0 : i32}> ({ + ^bb0(%b_a00_213: f32 loc(callsite(#loc1 at #loc554)), %b_a00_214: f32 loc(callsite(#loc1 at #loc554))): + %b_a00_215 = arith.addf %b_a00_213, %b_a00_214 : f32 loc(#loc709) + tt.reduce.return %b_a00_215 : f32 loc(#loc690) + }) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc690) + %b_a00_203 = arith.cmpi slt, %m_tc0, %b_a00 : tensor<16xi64> loc(#loc555) + %b_a00_204 = arith.select %b_a00_203, %b_a00_202, %cst_9 : tensor<16xi1>, tensor<16xf32> loc(#loc556) + %b_a00_205 = tt.expand_dims %b_a00_204 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc557) + %b_a00_206 = tt.broadcast %b_a00_205 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc558) + %b_a00_207 = arith.mulf %b_a00_206, %b_Ai00_197 : tensor<16x16xf32> loc(#loc558) + %b_a00_208 = "tt.reduce"(%b_a00_207) <{axis = 0 : i32}> ({ + ^bb0(%b_a00_213: f32 loc(callsite(#loc1 at #loc559)), %b_a00_214: f32 loc(callsite(#loc1 at #loc559))): + %b_a00_215 = arith.addf %b_a00_213, %b_a00_214 : f32 loc(#loc710) + tt.reduce.return %b_a00_215 : f32 loc(#loc692) + }) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc692) + %b_a00_209 = arith.addf %b_a00_204, %b_a00_208 : tensor<16xf32> loc(#loc560) + %b_Ai00_210 = tt.expand_dims %b_a00_209 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc561) + %b_Ai00_211 = tt.broadcast %b_Ai00_210 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc561) + %b_Ai00_212 = arith.select %b_a00_200, %b_Ai00_211, %b_Ai00_197 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc561) + scf.yield %b_Ai00_212 : tensor<16x16xf32> loc(#loc222) + } loc(#loc550) + %12 = arith.subi %6, %i_tc1 : i64 loc(#loc223) + %13 = arith.minsi %12, %c16_i64 : i64 loc(#loc224) + %b_Ai11_139 = scf.for %i = %c2_i64 to %13 step %c1_i64 iter_args(%b_Ai11_197 = %b_Ai11) -> (tensor<16x16xf32>) : i64 { + %b_a11 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc563) + %b_a11_198 = arith.cmpi eq, %m_tc0, %b_a11 : tensor<16xi64> loc(#loc563) + %b_a11_199 = tt.expand_dims %b_a11_198 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc564) + %b_a11_200 = tt.broadcast %b_a11_199 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc565) + %b_a11_201 = arith.select %b_a11_200, %b_Ai11, %cst_6 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc565) + %b_a11_202 = "tt.reduce"(%b_a11_201) <{axis = 0 : i32}> ({ + ^bb0(%b_a11_213: f32 loc(callsite(#loc1 at #loc566)), %b_a11_214: f32 loc(callsite(#loc1 at #loc566))): + %b_a11_215 = arith.addf %b_a11_213, %b_a11_214 : f32 loc(#loc711) + tt.reduce.return %b_a11_215 : f32 loc(#loc694) + }) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc694) + %b_a11_203 = arith.cmpi slt, %m_tc0, %b_a11 : tensor<16xi64> loc(#loc567) + %b_a11_204 = arith.select %b_a11_203, %b_a11_202, %cst_9 : tensor<16xi1>, tensor<16xf32> loc(#loc568) + %b_a11_205 = tt.expand_dims %b_a11_204 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc569) + %b_a11_206 = tt.broadcast %b_a11_205 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc570) + %b_a11_207 = arith.mulf %b_a11_206, %b_Ai11_197 : tensor<16x16xf32> loc(#loc570) + %b_a11_208 = "tt.reduce"(%b_a11_207) <{axis = 0 : i32}> ({ + ^bb0(%b_a11_213: f32 loc(callsite(#loc1 at #loc571)), %b_a11_214: f32 loc(callsite(#loc1 at #loc571))): + %b_a11_215 = arith.addf %b_a11_213, %b_a11_214 : f32 loc(#loc712) + tt.reduce.return %b_a11_215 : f32 loc(#loc696) + }) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc696) + %b_a11_209 = arith.addf %b_a11_204, %b_a11_208 : tensor<16xf32> loc(#loc572) + %b_Ai11_210 = tt.expand_dims %b_a11_209 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc573) + %b_Ai11_211 = tt.broadcast %b_Ai11_210 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc573) + %b_Ai11_212 = arith.select %b_a11_200, %b_Ai11_211, %b_Ai11_197 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc573) + scf.yield %b_Ai11_212 : tensor<16x16xf32> loc(#loc237) + } loc(#loc562) + %14 = arith.subi %6, %i_tc2 : i64 loc(#loc238) + %15 = arith.minsi %14, %c16_i64 : i64 loc(#loc239) + %b_Ai22_140 = scf.for %i = %c2_i64 to %15 step %c1_i64 iter_args(%b_Ai22_197 = %b_Ai22) -> (tensor<16x16xf32>) : i64 { + %b_a22 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc575) + %b_a22_198 = arith.cmpi eq, %m_tc0, %b_a22 : tensor<16xi64> loc(#loc575) + %b_a22_199 = tt.expand_dims %b_a22_198 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc576) + %b_a22_200 = tt.broadcast %b_a22_199 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc577) + %b_a22_201 = arith.select %b_a22_200, %b_Ai22, %cst_6 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc577) + %b_a22_202 = "tt.reduce"(%b_a22_201) <{axis = 0 : i32}> ({ + ^bb0(%b_a22_213: f32 loc(callsite(#loc1 at #loc578)), %b_a22_214: f32 loc(callsite(#loc1 at #loc578))): + %b_a22_215 = arith.addf %b_a22_213, %b_a22_214 : f32 loc(#loc713) + tt.reduce.return %b_a22_215 : f32 loc(#loc698) + }) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc698) + %b_a22_203 = arith.cmpi slt, %m_tc0, %b_a22 : tensor<16xi64> loc(#loc579) + %b_a22_204 = arith.select %b_a22_203, %b_a22_202, %cst_9 : tensor<16xi1>, tensor<16xf32> loc(#loc580) + %b_a22_205 = tt.expand_dims %b_a22_204 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc581) + %b_a22_206 = tt.broadcast %b_a22_205 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc582) + %b_a22_207 = arith.mulf %b_a22_206, %b_Ai22_197 : tensor<16x16xf32> loc(#loc582) + %b_a22_208 = "tt.reduce"(%b_a22_207) <{axis = 0 : i32}> ({ + ^bb0(%b_a22_213: f32 loc(callsite(#loc1 at #loc583)), %b_a22_214: f32 loc(callsite(#loc1 at #loc583))): + %b_a22_215 = arith.addf %b_a22_213, %b_a22_214 : f32 loc(#loc714) + tt.reduce.return %b_a22_215 : f32 loc(#loc700) + }) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc700) + %b_a22_209 = arith.addf %b_a22_204, %b_a22_208 : tensor<16xf32> loc(#loc584) + %b_Ai22_210 = tt.expand_dims %b_a22_209 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc585) + %b_Ai22_211 = tt.broadcast %b_Ai22_210 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc585) + %b_Ai22_212 = arith.select %b_a22_200, %b_Ai22_211, %b_Ai22_197 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc585) + scf.yield %b_Ai22_212 : tensor<16x16xf32> loc(#loc252) + } loc(#loc574) + %16 = arith.subi %6, %i_tc3 : i64 loc(#loc253) + %17 = arith.minsi %16, %c16_i64 : i64 loc(#loc254) + %b_Ai33_141 = scf.for %i = %c2_i64 to %17 step %c1_i64 iter_args(%b_Ai33_197 = %b_Ai33) -> (tensor<16x16xf32>) : i64 { + %b_a33 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc587) + %b_a33_198 = arith.cmpi eq, %m_tc0, %b_a33 : tensor<16xi64> loc(#loc587) + %b_a33_199 = tt.expand_dims %b_a33_198 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc588) + %b_a33_200 = tt.broadcast %b_a33_199 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc589) + %b_a33_201 = arith.select %b_a33_200, %b_Ai33, %cst_6 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc589) + %b_a33_202 = "tt.reduce"(%b_a33_201) <{axis = 0 : i32}> ({ + ^bb0(%b_a33_213: f32 loc(callsite(#loc1 at #loc590)), %b_a33_214: f32 loc(callsite(#loc1 at #loc590))): + %b_a33_215 = arith.addf %b_a33_213, %b_a33_214 : f32 loc(#loc715) + tt.reduce.return %b_a33_215 : f32 loc(#loc702) + }) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc702) + %b_a33_203 = arith.cmpi slt, %m_tc0, %b_a33 : tensor<16xi64> loc(#loc591) + %b_a33_204 = arith.select %b_a33_203, %b_a33_202, %cst_9 : tensor<16xi1>, tensor<16xf32> loc(#loc592) + %b_a33_205 = tt.expand_dims %b_a33_204 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc593) + %b_a33_206 = tt.broadcast %b_a33_205 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc594) + %b_a33_207 = arith.mulf %b_a33_206, %b_Ai33_197 : tensor<16x16xf32> loc(#loc594) + %b_a33_208 = "tt.reduce"(%b_a33_207) <{axis = 0 : i32}> ({ + ^bb0(%b_a33_213: f32 loc(callsite(#loc1 at #loc595)), %b_a33_214: f32 loc(callsite(#loc1 at #loc595))): + %b_a33_215 = arith.addf %b_a33_213, %b_a33_214 : f32 loc(#loc716) + tt.reduce.return %b_a33_215 : f32 loc(#loc704) + }) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc704) + %b_a33_209 = arith.addf %b_a33_204, %b_a33_208 : tensor<16xf32> loc(#loc596) + %b_Ai33_210 = tt.expand_dims %b_a33_209 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc597) + %b_Ai33_211 = tt.broadcast %b_Ai33_210 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc597) + %b_Ai33_212 = arith.select %b_a33_200, %b_Ai33_211, %b_Ai33_197 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc597) + scf.yield %b_Ai33_212 : tensor<16x16xf32> loc(#loc267) + } loc(#loc586) + %b_Ai00_142 = arith.uitofp %m_I : tensor<16x16xi1> to tensor<16x16xf32> loc(#loc598) + %b_Ai00_143 = arith.addf %b_Ai00_138, %b_Ai00_142 : tensor<16x16xf32> loc(#loc598) + %b_Ai11_144 = arith.addf %b_Ai11_139, %b_Ai00_142 : tensor<16x16xf32> loc(#loc599) + %b_Ai22_145 = arith.addf %b_Ai22_140, %b_Ai00_142 : tensor<16x16xf32> loc(#loc600) + %b_Ai33_146 = arith.addf %b_Ai33_141, %b_Ai00_142 : tensor<16x16xf32> loc(#loc601) + %b_Ai10 = tt.dot %b_Ai11_144, %b_A10_132, %cst_6, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc602) + %b_Ai10_147 = tt.dot %b_Ai10, %b_Ai00_143, %cst_6, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc603) + %b_Ai10_148 = arith.subf %cst_6, %b_Ai10_147 : tensor<16x16xf32> loc(#loc604) + %b_Ai21 = tt.dot %b_Ai22_145, %b_A21_134, %cst_6, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc605) + %b_Ai21_149 = tt.dot %b_Ai21, %b_Ai11_144, %cst_6, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc606) + %b_Ai21_150 = arith.subf %cst_6, %b_Ai21_149 : tensor<16x16xf32> loc(#loc607) + %b_Ai32 = tt.dot %b_Ai33_146, %b_A32_137, %cst_6, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc608) + %b_Ai32_151 = tt.dot %b_Ai32, %b_Ai22_145, %cst_6, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc609) + %b_Ai32_152 = arith.subf %cst_6, %b_Ai32_151 : tensor<16x16xf32> loc(#loc610) + %b_Ai20 = tt.dot %b_A20_133, %b_Ai00_143, %cst_6, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc611) + %b_Ai20_153 = tt.dot %b_A21_134, %b_Ai10_148, %b_Ai20, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc612) + %b_Ai20_154 = tt.dot %b_Ai22_145, %b_Ai20_153, %cst_6, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc613) + %b_Ai20_155 = arith.subf %cst_6, %b_Ai20_154 : tensor<16x16xf32> loc(#loc614) + %b_Ai31 = tt.dot %b_A31_136, %b_Ai11_144, %cst_6, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc615) + %b_Ai31_156 = tt.dot %b_A32_137, %b_Ai21_150, %b_Ai31, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc616) + %b_Ai31_157 = tt.dot %b_Ai33_146, %b_Ai31_156, %cst_6, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc617) + %b_Ai31_158 = arith.subf %cst_6, %b_Ai31_157 : tensor<16x16xf32> loc(#loc618) + %b_Ai30 = tt.dot %b_A30_135, %b_Ai00_143, %cst_6, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc619) + %b_Ai30_159 = tt.dot %b_A31_136, %b_Ai10_148, %b_Ai30, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc620) + %b_Ai30_160 = tt.dot %b_A32_137, %b_Ai20_155, %b_Ai30_159, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc621) + %b_Ai30_161 = tt.dot %b_Ai33_146, %b_Ai30_160, %cst_6, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc622) + %b_Ai30_162 = arith.subf %cst_6, %b_Ai30_161 : tensor<16x16xf32> loc(#loc623) + %p_A00 = tt.expand_dims %m_tc0_18 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc624) + %p_A00_163 = arith.muli %p_A00, %cst_5 : tensor<16x1xi64> loc(#loc625) + %p_A00_164 = tt.splat %A_16 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc626) + %p_A00_165 = tt.addptr %p_A00_164, %p_A00_163 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc626) + %p_A00_166 = tt.broadcast %p_A00_165 : tensor<16x1x!tt.ptr> -> tensor<16x16x!tt.ptr> loc(#loc627) + %p_A00_167 = tt.addptr %p_A00_166, %m_d_41 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc627) + %p_A10 = tt.expand_dims %m_tc1_21 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc628) + %p_A10_168 = arith.muli %p_A10, %cst_5 : tensor<16x1xi64> loc(#loc629) + %p_A10_169 = tt.addptr %p_A00_164, %p_A10_168 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc630) + %p_A10_170 = tt.broadcast %p_A10_169 : tensor<16x1x!tt.ptr> -> tensor<16x16x!tt.ptr> loc(#loc631) + %p_A10_171 = tt.addptr %p_A10_170, %m_d_41 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc631) + %p_A11 = arith.addi %o_i, %cst_4 : tensor<16xi32> loc(#loc632) + %p_A11_172 = tt.expand_dims %p_A11 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc633) + %p_A11_173 = tt.broadcast %p_A11_172 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc634) + %p_A11_174 = tt.addptr %p_A10_170, %p_A11_173 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc634) + %p_A20 = tt.expand_dims %m_tc2_23 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc635) + %p_A20_175 = arith.muli %p_A20, %cst_5 : tensor<16x1xi64> loc(#loc636) + %p_A20_176 = tt.addptr %p_A00_164, %p_A20_175 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc637) + %p_A20_177 = tt.broadcast %p_A20_176 : tensor<16x1x!tt.ptr> -> tensor<16x16x!tt.ptr> loc(#loc638) + %p_A20_178 = tt.addptr %p_A20_177, %m_d_41 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc638) + %p_A21 = tt.addptr %p_A20_177, %p_A11_173 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc639) + %p_A22 = arith.addi %o_i, %cst_3 : tensor<16xi32> loc(#loc640) + %p_A22_179 = tt.expand_dims %p_A22 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc641) + %p_A22_180 = tt.broadcast %p_A22_179 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc642) + %p_A22_181 = tt.addptr %p_A20_177, %p_A22_180 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc642) + %p_A30 = tt.expand_dims %m_tc3_25 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc643) + %p_A30_182 = arith.muli %p_A30, %cst_5 : tensor<16x1xi64> loc(#loc644) + %p_A30_183 = tt.addptr %p_A00_164, %p_A30_182 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc645) + %p_A30_184 = tt.broadcast %p_A30_183 : tensor<16x1x!tt.ptr> -> tensor<16x16x!tt.ptr> loc(#loc646) + %p_A30_185 = tt.addptr %p_A30_184, %m_d_41 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc646) + %p_A31 = tt.addptr %p_A30_184, %p_A11_173 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc647) + %p_A32 = tt.addptr %p_A30_184, %p_A22_180 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc648) + %p_A33 = arith.addi %o_i, %cst_2 : tensor<16xi32> loc(#loc649) + %p_A33_186 = tt.expand_dims %p_A33 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc650) + %p_A33_187 = tt.broadcast %p_A33_186 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc651) + %p_A33_188 = tt.addptr %p_A30_184, %p_A33_187 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc651) + %m_A0 = arith.cmpi slt, %m_d_39, %cst_1 : tensor<1x16xi32> loc(#loc652) + %m_A0_189 = tt.broadcast %m_A0 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc653) + %m_A0_190 = arith.andi %b_A00_43, %m_A0_189 : tensor<16x16xi1> loc(#loc653) + %m_A1 = arith.andi %b_A11_56, %m_A0_189 : tensor<16x16xi1> loc(#loc654) + %m_A2 = arith.andi %b_A22_69, %m_A0_189 : tensor<16x16xi1> loc(#loc655) + %m_A3 = arith.andi %b_A33_82, %m_A0_189 : tensor<16x16xi1> loc(#loc656) + %m_A11 = arith.cmpi slt, %p_A11_172, %cst_1 : tensor<1x16xi32> loc(#loc657) + %m_A11_191 = tt.broadcast %m_A11 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc658) + %m_A11_192 = arith.andi %b_A11_56, %m_A11_191 : tensor<16x16xi1> loc(#loc658) + %m_A21 = arith.andi %b_A22_69, %m_A11_191 : tensor<16x16xi1> loc(#loc659) + %m_A22 = arith.cmpi slt, %p_A22_179, %cst_1 : tensor<1x16xi32> loc(#loc660) + %m_A22_193 = tt.broadcast %m_A22 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc661) + %m_A22_194 = arith.andi %b_A22_69, %m_A22_193 : tensor<16x16xi1> loc(#loc661) + %m_A31 = arith.andi %b_A33_82, %m_A11_191 : tensor<16x16xi1> loc(#loc662) + %m_A32 = arith.andi %b_A33_82, %m_A22_193 : tensor<16x16xi1> loc(#loc663) + %m_A33 = arith.cmpi slt, %p_A33_186, %cst_1 : tensor<1x16xi32> loc(#loc664) + %m_A33_195 = tt.broadcast %m_A33 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc665) + %m_A33_196 = arith.andi %b_A33_82, %m_A33_195 : tensor<16x16xi1> loc(#loc665) + %18 = arith.truncf %b_Ai00_143 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc336) + tt.store %p_A00_167, %18, %m_A0_190 : tensor<16x16x!tt.ptr> loc(#loc337) + %19 = arith.truncf %b_Ai10_148 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc338) + tt.store %p_A10_171, %19, %m_A1 : tensor<16x16x!tt.ptr> loc(#loc339) + %20 = arith.truncf %b_Ai11_144 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc340) + tt.store %p_A11_174, %20, %m_A11_192 : tensor<16x16x!tt.ptr> loc(#loc341) + %21 = arith.truncf %b_Ai20_155 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc342) + tt.store %p_A20_178, %21, %m_A2 : tensor<16x16x!tt.ptr> loc(#loc343) + %22 = arith.truncf %b_Ai21_150 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc344) + tt.store %p_A21, %22, %m_A21 : tensor<16x16x!tt.ptr> loc(#loc345) + %23 = arith.truncf %b_Ai22_145 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc346) + tt.store %p_A22_181, %23, %m_A22_194 : tensor<16x16x!tt.ptr> loc(#loc347) + %24 = arith.truncf %b_Ai30_162 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc348) + tt.store %p_A30_185, %24, %m_A3 : tensor<16x16x!tt.ptr> loc(#loc349) + %25 = arith.truncf %b_Ai31_158 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc350) + tt.store %p_A31, %25, %m_A31 : tensor<16x16x!tt.ptr> loc(#loc351) + %26 = arith.truncf %b_Ai32_152 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc352) + tt.store %p_A32, %26, %m_A32 : tensor<16x16x!tt.ptr> loc(#loc353) + %27 = arith.truncf %b_Ai33_146 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc354) + tt.store %p_A33_188, %27, %m_A33_196 : tensor<16x16x!tt.ptr> loc(#loc355) + tt.return loc(#loc356) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":70:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":70:36) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":70:61) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":70:67) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":71:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":71:34) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":78:25) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":80:13) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":80:19) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":81:8) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":84:23) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":85:23) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":86:23) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":88:16) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":88:20) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":88:40) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":88:9) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":89:28) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":89:9) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":91:23) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":92:21) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":92:28) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":93:21) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":93:28) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":94:21) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":94:28) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":95:21) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":95:28) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":98:29) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":98:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":98:51) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":98:35) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":99:51) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":99:35) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":100:51) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":100:35) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":101:51) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":101:35) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":102:19) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":102:51) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":103:19) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":103:51) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":104:19) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":104:51) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":105:19) +#loc47 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":105:51) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":109:30) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":109:19) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":109:36) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":110:36) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":111:36) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":112:36) +#loc54 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":114:23) +#loc55 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":115:23) +#loc56 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":116:23) +#loc57 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":117:23) +#loc58 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":137:21) +#loc59 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":138:20) +#loc60 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":138:38) +#loc61 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":138:25) +#loc62 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:33) +#loc63 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:45) +#loc64 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:19) +#loc65 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:56) +#loc66 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:52) +#loc67 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":140:40) +#loc68 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":140:67) +#loc69 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":140:52) +#loc70 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":140:23) +#loc71 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":142:38) +#loc72 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":142:45) +#loc73 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":144:19) +#loc74 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":144:11) +#loc75 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":145:37) +#loc76 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":145:49) +#loc77 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":145:23) +#loc78 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":145:56) +#loc79 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":146:44) +#loc80 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":146:56) +#loc81 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":146:27) +#loc82 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":148:42) +#loc83 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":148:49) +#loc84 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":150:49) +#loc85 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":152:23) +#loc86 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":152:15) +#loc87 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":153:41) +#loc88 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":153:53) +#loc89 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":153:27) +#loc90 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":153:60) +#loc91 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":154:48) +#loc92 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":154:60) +#loc93 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":154:31) +#loc94 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":156:46) +#loc95 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":156:53) +#loc96 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":158:53) +#loc97 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":159:53) +#loc98 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":161:27) +#loc99 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":161:19) +#loc100 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":162:45) +#loc101 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":162:57) +#loc102 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":162:31) +#loc103 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":162:64) +#loc104 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":163:52) +#loc105 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":163:64) +#loc106 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":163:35) +#loc107 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":165:50) +#loc108 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":165:57) +#loc109 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":167:57) +#loc110 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":168:57) +#loc111 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":169:57) +#loc112 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":144:8) +#loc113 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":179:14) +#loc114 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":179:29) +#loc115 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":179:25) +#loc116 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":180:26) +#loc117 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:38) +#loc118 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:32) +#loc119 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:55) +#loc120 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:49) +#loc121 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:75) +#loc122 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:91) +#loc123 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:86) +#loc124 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:33) +#loc125 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:70) +#loc126 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:102) +#loc127 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:17) +#loc128 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:38) +#loc129 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:32) +#loc130 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:55) +#loc131 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:49) +#loc132 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:75) +#loc133 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:91) +#loc134 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:86) +#loc135 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:70) +#loc136 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:102) +#loc137 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:17) +#loc138 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:38) +#loc139 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:32) +#loc140 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:55) +#loc141 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:49) +#loc142 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:75) +#loc143 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:91) +#loc144 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:86) +#loc145 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:70) +#loc146 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:102) +#loc147 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:17) +#loc148 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:38) +#loc149 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:32) +#loc150 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:55) +#loc151 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:49) +#loc152 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:75) +#loc153 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:91) +#loc154 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:86) +#loc155 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:70) +#loc156 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:102) +#loc157 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:17) +#loc158 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:43) +#loc159 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:80) +#loc160 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:64) +#loc161 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:96) +#loc162 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:17) +#loc163 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:43) +#loc164 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:80) +#loc165 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:64) +#loc166 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:96) +#loc167 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:17) +#loc168 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:43) +#loc169 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:80) +#loc170 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:64) +#loc171 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:96) +#loc172 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:17) +#loc173 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:43) +#loc174 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:80) +#loc175 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:64) +#loc176 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:96) +#loc177 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:17) +#loc178 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:43) +#loc179 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:80) +#loc180 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:64) +#loc181 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:96) +#loc182 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:17) +#loc183 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:43) +#loc184 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:80) +#loc185 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:64) +#loc186 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:96) +#loc187 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:17) +#loc188 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":201:25) +#loc189 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":201:20) +#loc190 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":202:25) +#loc191 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":202:20) +#loc192 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":203:25) +#loc193 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":203:20) +#loc194 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":204:25) +#loc195 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":204:20) +#loc196 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":207:20) +#loc197 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":208:20) +#loc198 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":209:20) +#loc199 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":210:20) +#loc200 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":211:20) +#loc201 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":212:20) +#loc202 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":222:14) +#loc203 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":223:14) +#loc204 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":224:14) +#loc205 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":225:14) +#loc206 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":227:34) +#loc207 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":227:30) +#loc208 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":227:22) +#loc209 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":228:40) +#loc210 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":228:43) +#loc211 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":228:61) +#loc212 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc214 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc215 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":229:31) +#loc216 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":229:41) +#loc217 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":230:37) +#loc218 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":230:48) +#loc220 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":230:24) +#loc221 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":231:54) +#loc222 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":231:8) +#loc223 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":232:34) +#loc224 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":232:30) +#loc225 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":232:22) +#loc226 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":233:40) +#loc227 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":233:43) +#loc228 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":233:61) +#loc230 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":234:31) +#loc231 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":234:41) +#loc232 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":235:37) +#loc233 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":235:48) +#loc235 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":235:24) +#loc236 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":236:54) +#loc237 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":236:8) +#loc238 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":237:34) +#loc239 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":237:30) +#loc240 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":237:22) +#loc241 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":238:40) +#loc242 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":238:43) +#loc243 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":238:61) +#loc245 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":239:31) +#loc246 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":239:41) +#loc247 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":240:37) +#loc248 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":240:48) +#loc250 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":240:24) +#loc251 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":241:54) +#loc252 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":241:8) +#loc253 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":242:34) +#loc254 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":242:30) +#loc255 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":242:22) +#loc256 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":243:40) +#loc257 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":243:43) +#loc258 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":243:61) +#loc260 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":244:31) +#loc261 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":244:41) +#loc262 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":245:37) +#loc263 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":245:48) +#loc265 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":245:24) +#loc266 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":246:54) +#loc267 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":246:8) +#loc268 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":248:14) +#loc269 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":249:14) +#loc270 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":250:14) +#loc271 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":251:14) +#loc272 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":258:23) +#loc273 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":259:8) +#loc274 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":257:14) +#loc275 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":263:23) +#loc276 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":264:8) +#loc277 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":262:14) +#loc278 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":268:23) +#loc279 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":269:8) +#loc280 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":267:14) +#loc281 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":275:22) +#loc282 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":276:22) +#loc283 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":275:8) +#loc284 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":273:14) +#loc285 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":281:22) +#loc286 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":282:22) +#loc287 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":281:8) +#loc288 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":279:14) +#loc289 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":287:22) +#loc290 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":288:22) +#loc291 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":289:22) +#loc292 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":287:8) +#loc293 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":285:14) +#loc294 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":297:30) +#loc295 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":297:42) +#loc296 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":297:16) +#loc297 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":297:51) +#loc298 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":298:30) +#loc299 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":298:42) +#loc300 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":298:16) +#loc301 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":298:51) +#loc302 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":299:57) +#loc303 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":299:62) +#loc304 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":299:51) +#loc305 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":300:30) +#loc306 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":300:42) +#loc307 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":300:16) +#loc308 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":300:51) +#loc309 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":301:51) +#loc310 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":302:59) +#loc311 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":302:64) +#loc312 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":302:51) +#loc313 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":303:30) +#loc314 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":303:42) +#loc315 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":303:16) +#loc316 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":303:51) +#loc317 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":304:51) +#loc318 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":305:51) +#loc319 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":306:59) +#loc320 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":306:64) +#loc321 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":306:51) +#loc322 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":308:44) +#loc323 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":308:29) +#loc324 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":309:29) +#loc325 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":310:29) +#loc326 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":311:29) +#loc327 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":312:52) +#loc328 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":312:30) +#loc329 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":313:30) +#loc330 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":314:54) +#loc331 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":314:30) +#loc332 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":315:30) +#loc333 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":316:30) +#loc334 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":317:54) +#loc335 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":317:30) +#loc336 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":319:30) +#loc337 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":319:20) +#loc338 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":320:30) +#loc339 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":320:20) +#loc340 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":321:30) +#loc341 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":321:20) +#loc342 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":322:30) +#loc343 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":322:20) +#loc344 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":323:30) +#loc345 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":323:20) +#loc346 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":324:30) +#loc347 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":324:20) +#loc348 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":325:30) +#loc349 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":325:20) +#loc350 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":326:30) +#loc351 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":326:20) +#loc352 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":327:30) +#loc353 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":327:20) +#loc354 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":328:30) +#loc355 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":328:20) +#loc356 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":328:4) +#loc362 = loc("i_tc1"(#loc12)) +#loc363 = loc("i_tc2"(#loc13)) +#loc364 = loc("i_tc3"(#loc14)) +#loc365 = loc("k"(#loc15)) +#loc366 = loc("k"(#loc16)) +#loc367 = loc("k"(#loc17)) +#loc368 = loc("k"(#loc18)) +#loc369 = loc("A"(#loc19)) +#loc370 = loc("A"(#loc20)) +#loc371 = loc("o_i"(#loc21)) +#loc372 = loc("m_tc0"(#loc22)) +#loc373 = loc("m_tc0"(#loc23)) +#loc374 = loc("m_tc1"(#loc24)) +#loc375 = loc("m_tc1"(#loc25)) +#loc376 = loc("m_tc2"(#loc26)) +#loc377 = loc("m_tc2"(#loc27)) +#loc378 = loc("m_tc3"(#loc28)) +#loc379 = loc("m_tc3"(#loc29)) +#loc380 = loc("p_b0"(#loc30)) +#loc381 = loc("p_b0"(#loc31)) +#loc382 = loc("p_b0"(#loc32)) +#loc383 = loc("p_b0"(#loc33)) +#loc384 = loc("p_b1"(#loc34)) +#loc385 = loc("p_b1"(#loc35)) +#loc386 = loc("p_b2"(#loc36)) +#loc387 = loc("p_b2"(#loc37)) +#loc388 = loc("p_b3"(#loc38)) +#loc389 = loc("p_b3"(#loc39)) +#loc390 = loc("b_b0"(#loc40)) +#loc391 = loc("b_b0"(#loc41)) +#loc392 = loc("b_b1"(#loc42)) +#loc393 = loc("b_b1"(#loc43)) +#loc394 = loc("b_b2"(#loc44)) +#loc395 = loc("b_b2"(#loc45)) +#loc396 = loc("b_b3"(#loc46)) +#loc397 = loc("b_b3"(#loc47)) +#loc398 = loc("p_g0"(#loc48)) +#loc399 = loc("p_g0"(#loc49)) +#loc400 = loc("p_g0"(#loc50)) +#loc401 = loc("p_g1"(#loc51)) +#loc402 = loc("p_g2"(#loc52)) +#loc403 = loc("p_g3"(#loc53)) +#loc404 = loc("b_g0"(#loc54)) +#loc405 = loc("b_g1"(#loc55)) +#loc406 = loc("b_g2"(#loc56)) +#loc407 = loc("b_g3"(#loc57)) +#loc408 = loc("b_A00"(#loc58)) +#loc409 = loc("o_k"(#loc59)) +#loc410 = loc("o_k"(#loc60)) +#loc411 = loc("o_k"(#loc61)) +#loc412 = loc("p_k0"(#loc62)) +#loc413 = loc("p_k0"(#loc63)) +#loc414 = loc("p_k0"(#loc64)) +#loc415 = loc("p_k0"(#loc65)) +#loc416 = loc("p_k0"(#loc66)) +#loc417 = loc("b_k0"(#loc67)) +#loc418 = loc("b_k0"(#loc68)) +#loc419 = loc("b_k0"(#loc69)) +#loc420 = loc("b_k0"(#loc70)) +#loc421 = loc("b_A00"(#loc71)) +#loc422 = loc("b_A00"(#loc72)) +#loc423 = loc("p_k1"(#loc75)) +#loc424 = loc("p_k1"(#loc76)) +#loc425 = loc("p_k1"(#loc77)) +#loc426 = loc("p_k1"(#loc78)) +#loc427 = loc("b_k1"(#loc79)) +#loc428 = loc("b_k1"(#loc80)) +#loc429 = loc("b_k1"(#loc81)) +#loc430 = loc("b_A11"(#loc82)) +#loc431 = loc("b_A11"(#loc83)) +#loc432 = loc("b_A10"(#loc84)) +#loc433 = loc("b_A22"(#loc86)) +#loc434 = loc("p_k2"(#loc87)) +#loc435 = loc("p_k2"(#loc88)) +#loc436 = loc("p_k2"(#loc89)) +#loc437 = loc("p_k2"(#loc90)) +#loc438 = loc("b_k2"(#loc91)) +#loc439 = loc("b_k2"(#loc92)) +#loc440 = loc("b_k2"(#loc93)) +#loc441 = loc("b_A22"(#loc94)) +#loc442 = loc("b_A22"(#loc95)) +#loc443 = loc("b_A20"(#loc96)) +#loc444 = loc("b_A21"(#loc97)) +#loc445 = loc("b_A33"(#loc99)) +#loc446 = loc("p_k3"(#loc100)) +#loc447 = loc("p_k3"(#loc101)) +#loc448 = loc("p_k3"(#loc102)) +#loc449 = loc("p_k3"(#loc103)) +#loc450 = loc("b_k3"(#loc104)) +#loc451 = loc("b_k3"(#loc105)) +#loc452 = loc("b_k3"(#loc106)) +#loc453 = loc("b_A33"(#loc107)) +#loc454 = loc("b_A33"(#loc108)) +#loc455 = loc("b_A30"(#loc109)) +#loc456 = loc("b_A31"(#loc110)) +#loc457 = loc("b_A32"(#loc111)) +#loc458 = loc("m_d"(#loc113)) +#loc459 = loc("m_d"(#loc114)) +#loc460 = loc("m_d"(#loc115)) +#loc461 = loc("m_I"(#loc116)) +#loc462 = loc("b_A00"(#loc117)) +#loc463 = loc("b_A00"(#loc118)) +#loc464 = loc("b_A00"(#loc119)) +#loc465 = loc("b_A00"(#loc120)) +#loc466 = loc("b_A00"(#loc121)) +#loc467 = loc("b_A00"(#loc122)) +#loc468 = loc("b_A00"(#loc123)) +#loc469 = loc("b_A00"(#loc125)) +#loc470 = loc("b_A00"(#loc126)) +#loc471 = loc("b_A00"(#loc127)) +#loc472 = loc("b_A11"(#loc128)) +#loc473 = loc("b_A11"(#loc129)) +#loc474 = loc("b_A11"(#loc130)) +#loc475 = loc("b_A11"(#loc131)) +#loc476 = loc("b_A11"(#loc132)) +#loc477 = loc("b_A11"(#loc133)) +#loc478 = loc("b_A11"(#loc134)) +#loc479 = loc("b_A11"(#loc135)) +#loc480 = loc("b_A11"(#loc136)) +#loc481 = loc("b_A11"(#loc137)) +#loc482 = loc("b_A22"(#loc138)) +#loc483 = loc("b_A22"(#loc139)) +#loc484 = loc("b_A22"(#loc140)) +#loc485 = loc("b_A22"(#loc141)) +#loc486 = loc("b_A22"(#loc142)) +#loc487 = loc("b_A22"(#loc143)) +#loc488 = loc("b_A22"(#loc144)) +#loc489 = loc("b_A22"(#loc145)) +#loc490 = loc("b_A22"(#loc146)) +#loc491 = loc("b_A22"(#loc147)) +#loc492 = loc("b_A33"(#loc148)) +#loc493 = loc("b_A33"(#loc149)) +#loc494 = loc("b_A33"(#loc150)) +#loc495 = loc("b_A33"(#loc151)) +#loc496 = loc("b_A33"(#loc152)) +#loc497 = loc("b_A33"(#loc153)) +#loc498 = loc("b_A33"(#loc154)) +#loc499 = loc("b_A33"(#loc155)) +#loc500 = loc("b_A33"(#loc156)) +#loc501 = loc("b_A33"(#loc157)) +#loc502 = loc("b_A10"(#loc158)) +#loc503 = loc("b_A10"(#loc159)) +#loc504 = loc("b_A10"(#loc160)) +#loc505 = loc("b_A10"(#loc161)) +#loc506 = loc("b_A10"(#loc162)) +#loc507 = loc("b_A20"(#loc163)) +#loc508 = loc("b_A20"(#loc164)) +#loc509 = loc("b_A20"(#loc165)) +#loc510 = loc("b_A20"(#loc166)) +#loc511 = loc("b_A20"(#loc167)) +#loc512 = loc("b_A21"(#loc168)) +#loc513 = loc("b_A21"(#loc169)) +#loc514 = loc("b_A21"(#loc170)) +#loc515 = loc("b_A21"(#loc171)) +#loc516 = loc("b_A21"(#loc172)) +#loc517 = loc("b_A30"(#loc173)) +#loc518 = loc("b_A30"(#loc174)) +#loc519 = loc("b_A30"(#loc175)) +#loc520 = loc("b_A30"(#loc176)) +#loc521 = loc("b_A30"(#loc177)) +#loc522 = loc("b_A31"(#loc178)) +#loc523 = loc("b_A31"(#loc179)) +#loc524 = loc("b_A31"(#loc180)) +#loc525 = loc("b_A31"(#loc181)) +#loc526 = loc("b_A31"(#loc182)) +#loc527 = loc("b_A32"(#loc183)) +#loc528 = loc("b_A32"(#loc184)) +#loc529 = loc("b_A32"(#loc185)) +#loc530 = loc("b_A32"(#loc186)) +#loc531 = loc("b_A32"(#loc187)) +#loc532 = loc("b_A00"(#loc188)) +#loc533 = loc("b_A00"(#loc189)) +#loc534 = loc("b_A11"(#loc190)) +#loc535 = loc("b_A11"(#loc191)) +#loc536 = loc("b_A22"(#loc192)) +#loc537 = loc("b_A22"(#loc193)) +#loc538 = loc("b_A33"(#loc194)) +#loc539 = loc("b_A33"(#loc195)) +#loc540 = loc("b_A10"(#loc196)) +#loc541 = loc("b_A20"(#loc197)) +#loc542 = loc("b_A21"(#loc198)) +#loc543 = loc("b_A30"(#loc199)) +#loc544 = loc("b_A31"(#loc200)) +#loc545 = loc("b_A32"(#loc201)) +#loc546 = loc("b_Ai00"(#loc202)) +#loc547 = loc("b_Ai11"(#loc203)) +#loc548 = loc("b_Ai22"(#loc204)) +#loc549 = loc("b_Ai33"(#loc205)) +#loc550 = loc("b_Ai00"(#loc208)) +#loc551 = loc("b_a00"(#loc209)) +#loc552 = loc("b_a00"(#loc210)) +#loc553 = loc("b_a00"(#loc211)) +#loc555 = loc("b_a00"(#loc215)) +#loc556 = loc("b_a00"(#loc216)) +#loc557 = loc("b_a00"(#loc217)) +#loc558 = loc("b_a00"(#loc218)) +#loc560 = loc("b_a00"(#loc220)) +#loc561 = loc("b_Ai00"(#loc221)) +#loc562 = loc("b_Ai11"(#loc225)) +#loc563 = loc("b_a11"(#loc226)) +#loc564 = loc("b_a11"(#loc227)) +#loc565 = loc("b_a11"(#loc228)) +#loc567 = loc("b_a11"(#loc230)) +#loc568 = loc("b_a11"(#loc231)) +#loc569 = loc("b_a11"(#loc232)) +#loc570 = loc("b_a11"(#loc233)) +#loc572 = loc("b_a11"(#loc235)) +#loc573 = loc("b_Ai11"(#loc236)) +#loc574 = loc("b_Ai22"(#loc240)) +#loc575 = loc("b_a22"(#loc241)) +#loc576 = loc("b_a22"(#loc242)) +#loc577 = loc("b_a22"(#loc243)) +#loc579 = loc("b_a22"(#loc245)) +#loc580 = loc("b_a22"(#loc246)) +#loc581 = loc("b_a22"(#loc247)) +#loc582 = loc("b_a22"(#loc248)) +#loc584 = loc("b_a22"(#loc250)) +#loc585 = loc("b_Ai22"(#loc251)) +#loc586 = loc("b_Ai33"(#loc255)) +#loc587 = loc("b_a33"(#loc256)) +#loc588 = loc("b_a33"(#loc257)) +#loc589 = loc("b_a33"(#loc258)) +#loc591 = loc("b_a33"(#loc260)) +#loc592 = loc("b_a33"(#loc261)) +#loc593 = loc("b_a33"(#loc262)) +#loc594 = loc("b_a33"(#loc263)) +#loc596 = loc("b_a33"(#loc265)) +#loc597 = loc("b_Ai33"(#loc266)) +#loc598 = loc("b_Ai00"(#loc268)) +#loc599 = loc("b_Ai11"(#loc269)) +#loc600 = loc("b_Ai22"(#loc270)) +#loc601 = loc("b_Ai33"(#loc271)) +#loc602 = loc("b_Ai10"(#loc272)) +#loc603 = loc("b_Ai10"(#loc273)) +#loc604 = loc("b_Ai10"(#loc274)) +#loc605 = loc("b_Ai21"(#loc275)) +#loc606 = loc("b_Ai21"(#loc276)) +#loc607 = loc("b_Ai21"(#loc277)) +#loc608 = loc("b_Ai32"(#loc278)) +#loc609 = loc("b_Ai32"(#loc279)) +#loc610 = loc("b_Ai32"(#loc280)) +#loc611 = loc("b_Ai20"(#loc281)) +#loc612 = loc("b_Ai20"(#loc282)) +#loc613 = loc("b_Ai20"(#loc283)) +#loc614 = loc("b_Ai20"(#loc284)) +#loc615 = loc("b_Ai31"(#loc285)) +#loc616 = loc("b_Ai31"(#loc286)) +#loc617 = loc("b_Ai31"(#loc287)) +#loc618 = loc("b_Ai31"(#loc288)) +#loc619 = loc("b_Ai30"(#loc289)) +#loc620 = loc("b_Ai30"(#loc290)) +#loc621 = loc("b_Ai30"(#loc291)) +#loc622 = loc("b_Ai30"(#loc292)) +#loc623 = loc("b_Ai30"(#loc293)) +#loc624 = loc("p_A00"(#loc294)) +#loc625 = loc("p_A00"(#loc295)) +#loc626 = loc("p_A00"(#loc296)) +#loc627 = loc("p_A00"(#loc297)) +#loc628 = loc("p_A10"(#loc298)) +#loc629 = loc("p_A10"(#loc299)) +#loc630 = loc("p_A10"(#loc300)) +#loc631 = loc("p_A10"(#loc301)) +#loc632 = loc("p_A11"(#loc302)) +#loc633 = loc("p_A11"(#loc303)) +#loc634 = loc("p_A11"(#loc304)) +#loc635 = loc("p_A20"(#loc305)) +#loc636 = loc("p_A20"(#loc306)) +#loc637 = loc("p_A20"(#loc307)) +#loc638 = loc("p_A20"(#loc308)) +#loc639 = loc("p_A21"(#loc309)) +#loc640 = loc("p_A22"(#loc310)) +#loc641 = loc("p_A22"(#loc311)) +#loc642 = loc("p_A22"(#loc312)) +#loc643 = loc("p_A30"(#loc313)) +#loc644 = loc("p_A30"(#loc314)) +#loc645 = loc("p_A30"(#loc315)) +#loc646 = loc("p_A30"(#loc316)) +#loc647 = loc("p_A31"(#loc317)) +#loc648 = loc("p_A32"(#loc318)) +#loc649 = loc("p_A33"(#loc319)) +#loc650 = loc("p_A33"(#loc320)) +#loc651 = loc("p_A33"(#loc321)) +#loc652 = loc("m_A0"(#loc322)) +#loc653 = loc("m_A0"(#loc323)) +#loc654 = loc("m_A1"(#loc324)) +#loc655 = loc("m_A2"(#loc325)) +#loc656 = loc("m_A3"(#loc326)) +#loc657 = loc("m_A11"(#loc327)) +#loc658 = loc("m_A11"(#loc328)) +#loc659 = loc("m_A21"(#loc329)) +#loc660 = loc("m_A22"(#loc330)) +#loc661 = loc("m_A22"(#loc331)) +#loc662 = loc("m_A31"(#loc332)) +#loc663 = loc("m_A32"(#loc333)) +#loc664 = loc("m_A33"(#loc334)) +#loc665 = loc("m_A33"(#loc335)) +#loc666 = loc(fused[#loc380, #loc381]) +#loc667 = loc(fused[#loc398, #loc399]) +#loc668 = loc("b_A11"(#loc408)) +#loc669 = loc("b_A11"(#loc431)) +#loc670 = loc("b_A10"(#loc432)) +#loc671 = loc("b_A33"(#loc433)) +#loc672 = loc("b_A22"(#loc442)) +#loc673 = loc("b_A20"(#loc443)) +#loc674 = loc("b_A21"(#loc444)) +#loc675 = loc("b_A30"(#loc445)) +#loc676 = loc("b_A33"(#loc454)) +#loc677 = loc("b_A30"(#loc455)) +#loc678 = loc("b_A31"(#loc456)) +#loc679 = loc("b_A32"(#loc457)) +#loc680 = loc(callsite(#loc124 at #loc469)) +#loc681 = loc(callsite(#loc124 at #loc479)) +#loc682 = loc(callsite(#loc124 at #loc489)) +#loc683 = loc(callsite(#loc124 at #loc499)) +#loc684 = loc(callsite(#loc124 at #loc504)) +#loc685 = loc(callsite(#loc124 at #loc509)) +#loc686 = loc(callsite(#loc124 at #loc514)) +#loc687 = loc(callsite(#loc124 at #loc519)) +#loc688 = loc(callsite(#loc124 at #loc524)) +#loc689 = loc(callsite(#loc124 at #loc529)) +#loc690 = loc(callsite(#loc212 at #loc554)) +#loc692 = loc(callsite(#loc212 at #loc559)) +#loc694 = loc(callsite(#loc212 at #loc566)) +#loc696 = loc(callsite(#loc212 at #loc571)) +#loc698 = loc(callsite(#loc212 at #loc578)) +#loc700 = loc(callsite(#loc212 at #loc583)) +#loc702 = loc(callsite(#loc212 at #loc590)) +#loc704 = loc(callsite(#loc212 at #loc595)) +#loc706 = loc("b_A22"(#loc668)) +#loc707 = loc("b_A20"(#loc671)) +#loc708 = loc("b_A31"(#loc675)) +#loc709 = loc(callsite(#loc214 at #loc690)) +#loc710 = loc(callsite(#loc214 at #loc692)) +#loc711 = loc(callsite(#loc214 at #loc694)) +#loc712 = loc(callsite(#loc214 at #loc696)) +#loc713 = loc(callsite(#loc214 at #loc698)) +#loc714 = loc(callsite(#loc214 at #loc700)) +#loc715 = loc(callsite(#loc214 at #loc702)) +#loc716 = loc(callsite(#loc214 at #loc704)) +#loc717 = loc("b_A33"(#loc706)) +#loc718 = loc("b_A21"(#loc707)) +#loc719 = loc("b_A32"(#loc708)) +#loc720 = loc("b_A10"(#loc717)) +#loc721 = loc("b_A30"(#loc718)) +#loc722 = loc("b_A20"(#loc720)) +#loc723 = loc("b_A31"(#loc721)) +#loc724 = loc("b_A21"(#loc722)) +#loc725 = loc("b_A32"(#loc723)) +#loc726 = loc("b_A30"(#loc724)) +#loc727 = loc("b_A31"(#loc726)) +#loc728 = loc("b_A32"(#loc727)) diff --git a/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/__grp__chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.json b/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/__grp__chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.json new file mode 100644 index 0000000000000000000000000000000000000000..dc0c79ec3c21204e52ad7d0475d312ff8c3d43f8 --- /dev/null +++ b/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/__grp__chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.json @@ -0,0 +1 @@ +{"child_paths": {"chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.source", "chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.ttir", "chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.ttgir", "chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.llir", "chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.ptx", "chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.cubin", "chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.json"}} \ No newline at end of file diff --git a/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.cubin b/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.cubin new file mode 100644 index 0000000000000000000000000000000000000000..1103355bdce2b754c45b32c195c071dd12cf2c8c --- /dev/null +++ b/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.cubin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a1a1e4327e1971883ef2e338762c0a22291f7c108bdc71077712d9f8cacd1f6d +size 152112 diff --git a/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.json b/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.json new file mode 100644 index 0000000000000000000000000000000000000000..41ad3fe48d3c3df66f7ded44e1ee40ca0ab5152c --- /dev/null +++ b/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.json @@ -0,0 +1 @@ +{"hash": "0674f764464b29029d5a9ecb1a139de5c97c76ab4b58723595ff047925066ad4", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 4, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 156696, "tmem_size": 256, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64"} \ No newline at end of file diff --git a/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.llir b/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.llir new file mode 100644 index 0000000000000000000000000000000000000000..f735a79c449f328d3f39afaad825e5fff1d6eeff --- /dev/null +++ b/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.llir @@ -0,0 +1,2956 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, ptr addrspace(1) %3, ptr addrspace(1) %4, ptr addrspace(1) %5, ptr addrspace(1) %6, ptr addrspace(1) %7, float %8, i32 %9, ptr addrspace(1) readnone captures(none) %10, ptr addrspace(1) readnone captures(none) %11) local_unnamed_addr #0 !dbg !5 { + %13 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !13 + %14 = icmp samesign ult i32 %13, 32, !dbg !13 + tail call void asm sideeffect "@$0 tcgen05.alloc.cta_group::1.sync.aligned.shared::cta.b32 [$1], 256;", "b,r"(i1 %14, ptr addrspace(3) @global_smem) #4, !dbg !13 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !13 + %15 = load i32, ptr addrspace(3) @global_smem, align 16, !dbg !13 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !13 + tail call void asm sideeffect "@$0 tcgen05.relinquish_alloc_permit.cta_group::1.sync.aligned;", "b"(i1 %14) #4, !dbg !13 + %16 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !14 + %17 = zext nneg i32 %16 to i64, !dbg !15 + %18 = lshr i64 %17, 1, !dbg !16 + %19 = lshr i64 %17, 6, !dbg !17 + %20 = and i64 %18, 31, !dbg !18 + %21 = sext i32 %9 to i64, !dbg !19 + %22 = mul nsw i64 %19, %21, !dbg !19 + %23 = add i32 %9, 63, !dbg !20 + %24 = sdiv i32 %23, 64, !dbg !25 + %25 = shl nsw i64 %22, 5, !dbg !26 + %26 = or disjoint i64 %25, %20, !dbg !27 + %27 = shl i64 %26, 7, !dbg !28 + %28 = getelementptr bfloat, ptr addrspace(1) %0, i64 %27, !dbg !29 + %29 = getelementptr bfloat, ptr addrspace(1) %1, i64 %27, !dbg !30 + %30 = getelementptr bfloat, ptr addrspace(1) %2, i64 %27, !dbg !31 + %31 = getelementptr bfloat, ptr addrspace(1) %4, i64 %27, !dbg !32 + %32 = getelementptr bfloat, ptr addrspace(1) %6, i64 %27, !dbg !33 + %33 = getelementptr bfloat, ptr addrspace(1) %7, i64 %27, !dbg !34 + %34 = shl nuw nsw i64 %17, 6, !dbg !35 + %35 = and i64 %34, 64, !dbg !35 + %36 = lshr i32 %13, 5, !dbg !36 + %37 = and i32 %13, 7, !dbg !36 + %38 = shl nuw nsw i32 %37, 3, !dbg !36 + %39 = and i32 %13, 63, !dbg !36 + %40 = and i32 %13, 120, !dbg !36 + %41 = lshr exact i32 %40, 3, !dbg !36 + %42 = or disjoint i32 %41, 16, !dbg !36 + %43 = or disjoint i32 %41, 32, !dbg !36 + %44 = or disjoint i32 %41, 48, !dbg !36 + %45 = and i32 %13, 15, !dbg !36 + %46 = and i32 %13, 96, !dbg !36 + %47 = lshr exact i32 %46, 1, !dbg !36 + %48 = or disjoint i32 %47, %45, !dbg !36 + %49 = zext nneg i32 %38 to i64, !dbg !37 + %50 = zext nneg i32 %39 to i64, !dbg !37 + %51 = zext nneg i32 %41 to i64, !dbg !37 + %52 = zext nneg i32 %42 to i64, !dbg !37 + %53 = zext nneg i32 %43 to i64, !dbg !37 + %54 = zext nneg i32 %44 to i64, !dbg !37 + %55 = zext nneg i32 %48 to i64, !dbg !37 + %56 = or disjoint i64 %35, %49, !dbg !37 + %57 = or disjoint i32 %38, 64, !dbg !38 + %58 = add nsw i32 %24, -1, !dbg !39 + %59 = getelementptr float, ptr addrspace(1) %3, i64 %26, !dbg !40 + %60 = getelementptr bfloat, ptr addrspace(1) %30, i64 %49, !dbg !41 + %61 = getelementptr bfloat, ptr addrspace(1) %28, i64 %49, !dbg !42 + %62 = zext nneg i32 %57 to i64, !dbg !43 + %63 = getelementptr bfloat, ptr addrspace(1) %28, i64 %62, !dbg !43 + %64 = getelementptr bfloat, ptr addrspace(1) %30, i64 %62, !dbg !44 + %65 = icmp sgt i32 %23, 63, !dbg !45 + %66 = sext i32 %58 to i64, !dbg !46 + %67 = shl nsw i64 %66, 6, !dbg !47 + %68 = or disjoint i64 %67, %51, !dbg !48 + %69 = or disjoint i64 %67, %52, !dbg !48 + %70 = or disjoint i64 %67, %53, !dbg !48 + %71 = or disjoint i64 %67, %54, !dbg !48 + %72 = or disjoint i64 %67, %50, !dbg !48 + %73 = icmp slt i64 %68, %21, !dbg !49 + %74 = icmp slt i64 %69, %21, !dbg !49 + %75 = icmp slt i64 %70, %21, !dbg !49 + %76 = icmp slt i64 %71, %21, !dbg !49 + %77 = icmp slt i64 %72, %21, !dbg !49 + %78 = add nsw i64 %67, 64, !dbg !50 + %79 = tail call i64 @llvm.smin.i64(i64 %78, i64 %21), !dbg !51 + %80 = add nsw i64 %22, 144115188075855871, !dbg !52 + %81 = add nsw i64 %80, %79, !dbg !53 + %.idx = shl i64 %81, 7, !dbg !54 + %82 = getelementptr i8, ptr addrspace(1) %3, i64 %.idx, !dbg !54 + %83 = getelementptr float, ptr addrspace(1) %82, i64 %20, !dbg !54 + %84 = and i32 %13, 127, !dbg !55 + %85 = icmp eq i32 %84, 0, !dbg !55 + %86 = select i1 %65, i32 4, i32 0, !dbg !55 + tail call void asm sideeffect "@$3 cp.async.ca.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x4, $2;", "r,l,r,b"(ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 156672), ptr addrspace(1) %83, i32 %86, i1 %85) #4, !dbg !55 + %87 = add i32 %15, 1048576, !dbg !56 + %88 = add i32 %15, 64, !dbg !57 + %89 = add i32 %15, 1048640, !dbg !58 + %90 = add i32 %15, 128, !dbg !59 + %91 = add i32 %15, 1048704, !dbg !60 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !55 + %.idx2 = shl nsw i64 %72, 7, !dbg !61 + %92 = getelementptr i8, ptr addrspace(1) %59, i64 %.idx2, !dbg !61 + %93 = and i1 %65, %77, !dbg !45 + %94 = and i32 %13, 64, !dbg !62 + %95 = icmp eq i32 %94, 0, !dbg !62 + %96 = shl nuw nsw i32 %39, 2, !dbg !62 + %97 = getelementptr i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 155648), i32 %96, !dbg !62 + %98 = select i1 %93, i32 4, i32 0, !dbg !62 + tail call void asm sideeffect "@$3 cp.async.ca.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x4, $2;", "r,l,r,b"(ptr addrspace(3) %97, ptr addrspace(1) %92, i32 %98, i1 %95) #4, !dbg !62 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !62 + %99 = getelementptr i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 156160), i32 %96, !dbg !62 + tail call void asm sideeffect "@$3 cp.async.ca.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x4, $2;", "r,l,r,b"(ptr addrspace(3) %99, ptr addrspace(1) %92, i32 %98, i1 %95) #4, !dbg !62 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !62 + %100 = shl nsw i64 %68, 12, !dbg !63 + %101 = shl nsw i64 %69, 12, !dbg !63 + %102 = shl nsw i64 %70, 12, !dbg !63 + %103 = shl nsw i64 %71, 12, !dbg !63 + %104 = getelementptr bfloat, ptr addrspace(1) %32, i64 %100, !dbg !64 + %105 = getelementptr bfloat, ptr addrspace(1) %32, i64 %101, !dbg !64 + %106 = getelementptr bfloat, ptr addrspace(1) %32, i64 %102, !dbg !64 + %107 = getelementptr bfloat, ptr addrspace(1) %32, i64 %103, !dbg !64 + %108 = getelementptr bfloat, ptr addrspace(1) %104, i64 %56, !dbg !65 + %109 = getelementptr bfloat, ptr addrspace(1) %105, i64 %56, !dbg !65 + %110 = getelementptr bfloat, ptr addrspace(1) %106, i64 %56, !dbg !65 + %111 = getelementptr bfloat, ptr addrspace(1) %107, i64 %56, !dbg !65 + %112 = getelementptr bfloat, ptr addrspace(1) %31, i64 %100, !dbg !66 + %113 = getelementptr bfloat, ptr addrspace(1) %31, i64 %101, !dbg !66 + %114 = getelementptr bfloat, ptr addrspace(1) %31, i64 %102, !dbg !66 + %115 = getelementptr bfloat, ptr addrspace(1) %31, i64 %103, !dbg !66 + %116 = getelementptr bfloat, ptr addrspace(1) %112, i64 %56, !dbg !67 + %117 = getelementptr bfloat, ptr addrspace(1) %113, i64 %56, !dbg !67 + %118 = getelementptr bfloat, ptr addrspace(1) %114, i64 %56, !dbg !67 + %119 = getelementptr bfloat, ptr addrspace(1) %115, i64 %56, !dbg !67 + %120 = and i1 %65, %73, !dbg !45 + %121 = and i1 %65, %74, !dbg !45 + %122 = and i1 %65, %75, !dbg !45 + %123 = and i1 %65, %76, !dbg !45 + %124 = shl nuw nsw i32 %13, 4, !dbg !68 + %125 = and i32 %124, 2032, !dbg !68 + %126 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %125, !dbg !68 + %127 = select i1 %120, i32 16, i32 0, !dbg !68 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %126, ptr addrspace(1) %116, i32 %127) #4, !dbg !68 + %128 = or disjoint i32 %125, 2048, !dbg !68 + %129 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %128, !dbg !68 + %130 = select i1 %121, i32 16, i32 0, !dbg !68 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %129, ptr addrspace(1) %117, i32 %130) #4, !dbg !68 + %131 = or disjoint i32 %125, 4096, !dbg !68 + %132 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %131, !dbg !68 + %133 = select i1 %122, i32 16, i32 0, !dbg !68 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %132, ptr addrspace(1) %118, i32 %133) #4, !dbg !68 + %134 = or disjoint i32 %125, 6144, !dbg !68 + %135 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %134, !dbg !68 + %136 = select i1 %123, i32 16, i32 0, !dbg !68 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %135, ptr addrspace(1) %119, i32 %136) #4, !dbg !68 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !68 + %137 = getelementptr bfloat, ptr addrspace(1) %29, i64 %100, !dbg !69 + %138 = getelementptr bfloat, ptr addrspace(1) %29, i64 %101, !dbg !69 + %139 = getelementptr bfloat, ptr addrspace(1) %29, i64 %102, !dbg !69 + %140 = getelementptr bfloat, ptr addrspace(1) %29, i64 %103, !dbg !69 + %141 = getelementptr bfloat, ptr addrspace(1) %137, i64 %49, !dbg !70 + %142 = getelementptr bfloat, ptr addrspace(1) %138, i64 %49, !dbg !70 + %143 = getelementptr bfloat, ptr addrspace(1) %139, i64 %49, !dbg !70 + %144 = getelementptr bfloat, ptr addrspace(1) %140, i64 %49, !dbg !70 + %145 = and i32 %13, 56, !dbg !71 + %146 = shl nuw nsw i32 %145, 1, !dbg !71 + %147 = xor i32 %125, %146, !dbg !71 + %148 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %147, !dbg !71 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %148, ptr addrspace(1) %141, i32 %127) #4, !dbg !71 + %149 = or disjoint i32 %147, 2048, !dbg !71 + %150 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %149, !dbg !71 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %150, ptr addrspace(1) %142, i32 %130) #4, !dbg !71 + %151 = or disjoint i32 %147, 4096, !dbg !71 + %152 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %151, !dbg !71 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %152, ptr addrspace(1) %143, i32 %133) #4, !dbg !71 + %153 = or disjoint i32 %147, 6144, !dbg !71 + %154 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %153, !dbg !71 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %154, ptr addrspace(1) %144, i32 %136) #4, !dbg !71 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !71 + %155 = getelementptr bfloat, ptr addrspace(1) %137, i64 %62, !dbg !72 + %156 = getelementptr bfloat, ptr addrspace(1) %138, i64 %62, !dbg !72 + %157 = getelementptr bfloat, ptr addrspace(1) %139, i64 %62, !dbg !72 + %158 = getelementptr bfloat, ptr addrspace(1) %140, i64 %62, !dbg !72 + %159 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 32768), i32 %147, !dbg !73 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %159, ptr addrspace(1) %155, i32 %127) #4, !dbg !73 + %160 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 32768), i32 %149, !dbg !73 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %160, ptr addrspace(1) %156, i32 %130) #4, !dbg !73 + %161 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 32768), i32 %151, !dbg !73 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %161, ptr addrspace(1) %157, i32 %133) #4, !dbg !73 + %162 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 32768), i32 %153, !dbg !73 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %162, ptr addrspace(1) %158, i32 %136) #4, !dbg !73 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !73 + %163 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 49152), i32 %125, !dbg !74 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %163, ptr addrspace(1) %108, i32 %127) #4, !dbg !74 + %164 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 49152), i32 %128, !dbg !74 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %164, ptr addrspace(1) %109, i32 %130) #4, !dbg !74 + %165 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 49152), i32 %131, !dbg !74 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %165, ptr addrspace(1) %110, i32 %133) #4, !dbg !74 + %166 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 49152), i32 %134, !dbg !74 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %166, ptr addrspace(1) %111, i32 %136) #4, !dbg !74 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !74 + %167 = getelementptr bfloat, ptr addrspace(1) %60, i64 %100, !dbg !75 + %168 = getelementptr bfloat, ptr addrspace(1) %60, i64 %101, !dbg !75 + %169 = getelementptr bfloat, ptr addrspace(1) %60, i64 %102, !dbg !75 + %170 = getelementptr bfloat, ptr addrspace(1) %60, i64 %103, !dbg !75 + %171 = getelementptr bfloat, ptr addrspace(1) %61, i64 %100, !dbg !76 + %172 = getelementptr bfloat, ptr addrspace(1) %61, i64 %101, !dbg !76 + %173 = getelementptr bfloat, ptr addrspace(1) %61, i64 %102, !dbg !76 + %174 = getelementptr bfloat, ptr addrspace(1) %61, i64 %103, !dbg !76 + %175 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 65536), i32 %147, !dbg !77 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %175, ptr addrspace(1) %167, i32 %127) #4, !dbg !77 + %176 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 65536), i32 %149, !dbg !77 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %176, ptr addrspace(1) %168, i32 %130) #4, !dbg !77 + %177 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 65536), i32 %151, !dbg !77 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %177, ptr addrspace(1) %169, i32 %133) #4, !dbg !77 + %178 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 65536), i32 %153, !dbg !77 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %178, ptr addrspace(1) %170, i32 %136) #4, !dbg !77 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !77 + %179 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 81920), i32 %125, !dbg !78 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %179, ptr addrspace(1) %171, i32 %127) #4, !dbg !78 + %180 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 81920), i32 %128, !dbg !78 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %180, ptr addrspace(1) %172, i32 %130) #4, !dbg !78 + %181 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 81920), i32 %131, !dbg !78 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %181, ptr addrspace(1) %173, i32 %133) #4, !dbg !78 + %182 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 81920), i32 %134, !dbg !78 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %182, ptr addrspace(1) %174, i32 %136) #4, !dbg !78 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !78 + %183 = getelementptr bfloat, ptr addrspace(1) %63, i64 %100, !dbg !79 + %184 = getelementptr bfloat, ptr addrspace(1) %63, i64 %101, !dbg !79 + %185 = getelementptr bfloat, ptr addrspace(1) %63, i64 %102, !dbg !79 + %186 = getelementptr bfloat, ptr addrspace(1) %63, i64 %103, !dbg !79 + %187 = getelementptr bfloat, ptr addrspace(1) %64, i64 %100, !dbg !80 + %188 = getelementptr bfloat, ptr addrspace(1) %64, i64 %101, !dbg !80 + %189 = getelementptr bfloat, ptr addrspace(1) %64, i64 %102, !dbg !80 + %190 = getelementptr bfloat, ptr addrspace(1) %64, i64 %103, !dbg !80 + %191 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 98304), i32 %125, !dbg !81 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %191, ptr addrspace(1) %183, i32 %127) #4, !dbg !81 + %192 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 98304), i32 %128, !dbg !81 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %192, ptr addrspace(1) %184, i32 %130) #4, !dbg !81 + %193 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 98304), i32 %131, !dbg !81 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %193, ptr addrspace(1) %185, i32 %133) #4, !dbg !81 + %194 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 98304), i32 %134, !dbg !81 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %194, ptr addrspace(1) %186, i32 %136) #4, !dbg !81 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !81 + %195 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 114688), i32 %147, !dbg !82 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %195, ptr addrspace(1) %187, i32 %127) #4, !dbg !82 + %196 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 114688), i32 %149, !dbg !82 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %196, ptr addrspace(1) %188, i32 %130) #4, !dbg !82 + %197 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 114688), i32 %151, !dbg !82 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %197, ptr addrspace(1) %189, i32 %133) #4, !dbg !82 + %198 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 114688), i32 %153, !dbg !82 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %198, ptr addrspace(1) %190, i32 %136) #4, !dbg !82 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !82 + %199 = icmp sgt i32 %23, 127, !dbg !45 + %200 = add nsw i32 %24, -2, !dbg !46 + %201 = sext i32 %200 to i64, !dbg !46 + %202 = shl nsw i64 %201, 6, !dbg !47 + %203 = or disjoint i64 %202, %51, !dbg !48 + %204 = or disjoint i64 %202, %52, !dbg !48 + %205 = or disjoint i64 %202, %53, !dbg !48 + %206 = or disjoint i64 %202, %54, !dbg !48 + %207 = or disjoint i64 %202, %50, !dbg !48 + %208 = icmp slt i64 %203, %21, !dbg !49 + %209 = icmp slt i64 %204, %21, !dbg !49 + %210 = icmp slt i64 %205, %21, !dbg !49 + %211 = icmp slt i64 %206, %21, !dbg !49 + %212 = icmp slt i64 %207, %21, !dbg !49 + %213 = add nsw i64 %202, 64, !dbg !50 + %214 = tail call i64 @llvm.smin.i64(i64 %213, i64 %21), !dbg !51 + %215 = add nsw i64 %80, %214, !dbg !53 + %.idx3 = shl i64 %215, 7, !dbg !54 + %216 = getelementptr i8, ptr addrspace(1) %3, i64 %.idx3, !dbg !54 + %217 = getelementptr float, ptr addrspace(1) %216, i64 %20, !dbg !54 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !55 + %218 = select i1 %199, i32 4, i32 0, !dbg !55 + tail call void asm sideeffect "@$3 cp.async.ca.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x4, $2;", "r,l,r,b"(ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 156676), ptr addrspace(1) %217, i32 %218, i1 %85) #4, !dbg !55 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !55 + %.idx4 = shl nsw i64 %207, 7, !dbg !61 + %219 = getelementptr i8, ptr addrspace(1) %59, i64 %.idx4, !dbg !61 + %220 = and i1 %199, %212, !dbg !45 + %221 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 155904), i32 %96, !dbg !62 + %222 = select i1 %220, i32 4, i32 0, !dbg !62 + tail call void asm sideeffect "@$3 cp.async.ca.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x4, $2;", "r,l,r,b"(ptr addrspace(3) %221, ptr addrspace(1) %219, i32 %222, i1 %95) #4, !dbg !62 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !62 + %223 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 156416), i32 %96, !dbg !62 + tail call void asm sideeffect "@$3 cp.async.ca.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x4, $2;", "r,l,r,b"(ptr addrspace(3) %223, ptr addrspace(1) %219, i32 %222, i1 %95) #4, !dbg !62 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !62 + %224 = shl nsw i64 %203, 12, !dbg !63 + %225 = shl nsw i64 %204, 12, !dbg !63 + %226 = shl nsw i64 %205, 12, !dbg !63 + %227 = shl nsw i64 %206, 12, !dbg !63 + %228 = getelementptr bfloat, ptr addrspace(1) %32, i64 %224, !dbg !64 + %229 = getelementptr bfloat, ptr addrspace(1) %32, i64 %225, !dbg !64 + %230 = getelementptr bfloat, ptr addrspace(1) %32, i64 %226, !dbg !64 + %231 = getelementptr bfloat, ptr addrspace(1) %32, i64 %227, !dbg !64 + %232 = getelementptr bfloat, ptr addrspace(1) %228, i64 %56, !dbg !65 + %233 = getelementptr bfloat, ptr addrspace(1) %229, i64 %56, !dbg !65 + %234 = getelementptr bfloat, ptr addrspace(1) %230, i64 %56, !dbg !65 + %235 = getelementptr bfloat, ptr addrspace(1) %231, i64 %56, !dbg !65 + %236 = getelementptr bfloat, ptr addrspace(1) %31, i64 %224, !dbg !66 + %237 = getelementptr bfloat, ptr addrspace(1) %31, i64 %225, !dbg !66 + %238 = getelementptr bfloat, ptr addrspace(1) %31, i64 %226, !dbg !66 + %239 = getelementptr bfloat, ptr addrspace(1) %31, i64 %227, !dbg !66 + %240 = getelementptr bfloat, ptr addrspace(1) %236, i64 %56, !dbg !67 + %241 = getelementptr bfloat, ptr addrspace(1) %237, i64 %56, !dbg !67 + %242 = getelementptr bfloat, ptr addrspace(1) %238, i64 %56, !dbg !67 + %243 = getelementptr bfloat, ptr addrspace(1) %239, i64 %56, !dbg !67 + %244 = and i1 %199, %208, !dbg !45 + %245 = and i1 %199, %209, !dbg !45 + %246 = and i1 %199, %210, !dbg !45 + %247 = and i1 %199, %211, !dbg !45 + %248 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 8192), i32 %125, !dbg !68 + %249 = select i1 %244, i32 16, i32 0, !dbg !68 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %248, ptr addrspace(1) %240, i32 %249) #4, !dbg !68 + %250 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 8192), i32 %128, !dbg !68 + %251 = select i1 %245, i32 16, i32 0, !dbg !68 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %250, ptr addrspace(1) %241, i32 %251) #4, !dbg !68 + %252 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 8192), i32 %131, !dbg !68 + %253 = select i1 %246, i32 16, i32 0, !dbg !68 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %252, ptr addrspace(1) %242, i32 %253) #4, !dbg !68 + %254 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 8192), i32 %134, !dbg !68 + %255 = select i1 %247, i32 16, i32 0, !dbg !68 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %254, ptr addrspace(1) %243, i32 %255) #4, !dbg !68 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !68 + %256 = getelementptr bfloat, ptr addrspace(1) %29, i64 %224, !dbg !69 + %257 = getelementptr bfloat, ptr addrspace(1) %29, i64 %225, !dbg !69 + %258 = getelementptr bfloat, ptr addrspace(1) %29, i64 %226, !dbg !69 + %259 = getelementptr bfloat, ptr addrspace(1) %29, i64 %227, !dbg !69 + %260 = getelementptr bfloat, ptr addrspace(1) %256, i64 %49, !dbg !70 + %261 = getelementptr bfloat, ptr addrspace(1) %257, i64 %49, !dbg !70 + %262 = getelementptr bfloat, ptr addrspace(1) %258, i64 %49, !dbg !70 + %263 = getelementptr bfloat, ptr addrspace(1) %259, i64 %49, !dbg !70 + %264 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 24576), i32 %147, !dbg !71 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %264, ptr addrspace(1) %260, i32 %249) #4, !dbg !71 + %265 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 24576), i32 %149, !dbg !71 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %265, ptr addrspace(1) %261, i32 %251) #4, !dbg !71 + %266 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 24576), i32 %151, !dbg !71 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %266, ptr addrspace(1) %262, i32 %253) #4, !dbg !71 + %267 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 24576), i32 %153, !dbg !71 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %267, ptr addrspace(1) %263, i32 %255) #4, !dbg !71 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !71 + %268 = getelementptr bfloat, ptr addrspace(1) %256, i64 %62, !dbg !72 + %269 = getelementptr bfloat, ptr addrspace(1) %257, i64 %62, !dbg !72 + %270 = getelementptr bfloat, ptr addrspace(1) %258, i64 %62, !dbg !72 + %271 = getelementptr bfloat, ptr addrspace(1) %259, i64 %62, !dbg !72 + %272 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 40960), i32 %147, !dbg !73 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %272, ptr addrspace(1) %268, i32 %249) #4, !dbg !73 + %273 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 40960), i32 %149, !dbg !73 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %273, ptr addrspace(1) %269, i32 %251) #4, !dbg !73 + %274 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 40960), i32 %151, !dbg !73 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %274, ptr addrspace(1) %270, i32 %253) #4, !dbg !73 + %275 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 40960), i32 %153, !dbg !73 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %275, ptr addrspace(1) %271, i32 %255) #4, !dbg !73 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !73 + %276 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 57344), i32 %125, !dbg !74 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %276, ptr addrspace(1) %232, i32 %249) #4, !dbg !74 + %277 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 57344), i32 %128, !dbg !74 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %277, ptr addrspace(1) %233, i32 %251) #4, !dbg !74 + %278 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 57344), i32 %131, !dbg !74 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %278, ptr addrspace(1) %234, i32 %253) #4, !dbg !74 + %279 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 57344), i32 %134, !dbg !74 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %279, ptr addrspace(1) %235, i32 %255) #4, !dbg !74 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !74 + %280 = getelementptr bfloat, ptr addrspace(1) %60, i64 %224, !dbg !75 + %281 = getelementptr bfloat, ptr addrspace(1) %60, i64 %225, !dbg !75 + %282 = getelementptr bfloat, ptr addrspace(1) %60, i64 %226, !dbg !75 + %283 = getelementptr bfloat, ptr addrspace(1) %60, i64 %227, !dbg !75 + %284 = getelementptr bfloat, ptr addrspace(1) %61, i64 %224, !dbg !76 + %285 = getelementptr bfloat, ptr addrspace(1) %61, i64 %225, !dbg !76 + %286 = getelementptr bfloat, ptr addrspace(1) %61, i64 %226, !dbg !76 + %287 = getelementptr bfloat, ptr addrspace(1) %61, i64 %227, !dbg !76 + %288 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73728), i32 %147, !dbg !77 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %288, ptr addrspace(1) %280, i32 %249) #4, !dbg !77 + %289 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73728), i32 %149, !dbg !77 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %289, ptr addrspace(1) %281, i32 %251) #4, !dbg !77 + %290 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73728), i32 %151, !dbg !77 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %290, ptr addrspace(1) %282, i32 %253) #4, !dbg !77 + %291 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73728), i32 %153, !dbg !77 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %291, ptr addrspace(1) %283, i32 %255) #4, !dbg !77 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !77 + %292 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 90112), i32 %125, !dbg !78 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %292, ptr addrspace(1) %284, i32 %249) #4, !dbg !78 + %293 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 90112), i32 %128, !dbg !78 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %293, ptr addrspace(1) %285, i32 %251) #4, !dbg !78 + %294 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 90112), i32 %131, !dbg !78 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %294, ptr addrspace(1) %286, i32 %253) #4, !dbg !78 + %295 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 90112), i32 %134, !dbg !78 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %295, ptr addrspace(1) %287, i32 %255) #4, !dbg !78 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !78 + %296 = getelementptr bfloat, ptr addrspace(1) %63, i64 %224, !dbg !79 + %297 = getelementptr bfloat, ptr addrspace(1) %63, i64 %225, !dbg !79 + %298 = getelementptr bfloat, ptr addrspace(1) %63, i64 %226, !dbg !79 + %299 = getelementptr bfloat, ptr addrspace(1) %63, i64 %227, !dbg !79 + %300 = getelementptr bfloat, ptr addrspace(1) %64, i64 %224, !dbg !80 + %301 = getelementptr bfloat, ptr addrspace(1) %64, i64 %225, !dbg !80 + %302 = getelementptr bfloat, ptr addrspace(1) %64, i64 %226, !dbg !80 + %303 = getelementptr bfloat, ptr addrspace(1) %64, i64 %227, !dbg !80 + %304 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 106496), i32 %125, !dbg !81 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %304, ptr addrspace(1) %296, i32 %249) #4, !dbg !81 + %305 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 106496), i32 %128, !dbg !81 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %305, ptr addrspace(1) %297, i32 %251) #4, !dbg !81 + %306 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 106496), i32 %131, !dbg !81 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %306, ptr addrspace(1) %298, i32 %253) #4, !dbg !81 + %307 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 106496), i32 %134, !dbg !81 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %307, ptr addrspace(1) %299, i32 %255) #4, !dbg !81 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !81 + %308 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %147, !dbg !82 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %308, ptr addrspace(1) %300, i32 %249) #4, !dbg !82 + %309 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %149, !dbg !82 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %309, ptr addrspace(1) %301, i32 %251) #4, !dbg !82 + %310 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %151, !dbg !82 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %310, ptr addrspace(1) %302, i32 %253) #4, !dbg !82 + %311 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %153, !dbg !82 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %311, ptr addrspace(1) %303, i32 %255) #4, !dbg !82 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !82 + %invariant.gep = getelementptr bfloat, ptr addrspace(1) %33, i64 %56, !dbg !45 + %invariant.gep430 = getelementptr float, ptr addrspace(1) %3, i64 %20, !dbg !45 + %invariant.gep436 = getelementptr bfloat, ptr addrspace(1) %32, i64 %56, !dbg !45 + %invariant.gep444 = getelementptr bfloat, ptr addrspace(1) %31, i64 %56, !dbg !45 + br i1 %65, label %.lr.ph, label %._crit_edge, !dbg !45 + +.lr.ph: ; preds = %12 + %312 = zext nneg i32 %24 to i64, !dbg !83 + %factor.op.mul = mul nuw nsw i64 %19, %312, !dbg !45 + %313 = shl nuw nsw i32 %44, 7, !dbg !84 + %314 = shl nuw nsw i32 %43, 7, !dbg !84 + %315 = shl nuw nsw i32 %42, 7, !dbg !84 + %316 = shl nuw nsw i32 %40, 4, !dbg !84 + %317 = add nsw i32 %24, -3 + %.reass = shl i64 %factor.op.mul, 20 + %318 = getelementptr i8, ptr addrspace(1) %5, i64 %.reass + %.idx193 = shl nuw nsw i64 %20, 15 + %319 = getelementptr i8, ptr addrspace(1) %318, i64 %.idx193 + %320 = zext nneg i32 %316 to i64 + %321 = zext nneg i32 %315 to i64 + %322 = zext nneg i32 %314 to i64 + %323 = zext nneg i32 %313 to i64 + %324 = shl nuw nsw i32 %45, 7 + %325 = shl nuw nsw i32 %46, 6 + %326 = shl nuw nsw i32 %37, 4 + %327 = and i32 %13, 16 + %328 = shl nuw nsw i32 %327, 2 + %329 = or disjoint i32 %325, %326 + %330 = xor i32 %329, %328 + %331 = or disjoint i32 %330, %324 + %332 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %331 + %333 = xor i32 %331, 16 + %334 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %333 + %335 = xor i32 %331, 32 + %336 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %335 + %337 = xor i32 %331, 48 + %338 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %337 + %339 = and i32 %124, 176 + %340 = shl nuw nsw i32 %46, 3 + %341 = and i32 %13, 4 + %342 = icmp eq i32 %341, 0 + %343 = select i1 %342, i32 0, i32 1088 + %344 = or disjoint i32 %339, %340 + %345 = xor i32 %343, %328 + %346 = or disjoint i32 %345, %344 + %347 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 139264), i32 %346 + %348 = shl nuw nsw i32 %145, 5 + %349 = shl nuw nsw i32 %13, 3 + %350 = and i32 %349, 48 + %351 = shl nuw nsw i32 %46, 1 + %352 = shl nuw nsw i32 %13, 6 + %353 = and i32 %352, 64 + %354 = or disjoint i32 %348, %350 + %355 = xor i32 %351, %353 + %356 = or disjoint i32 %354, %355 + %357 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 139264), i32 %356 + %358 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 139264), i32 %331 + %359 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 139264), i32 %333 + %360 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 139264), i32 %335 + %361 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 139264), i32 %337 + %362 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 147456), i32 %346 + %363 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 147456), i32 %356 + %364 = and i32 %124, 48 + %365 = shl nuw nsw i32 %36, 6 + %366 = or i32 %365, %349 + %367 = and i32 %366, 192 + %invariant.gep472 = getelementptr i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 155648), i32 %364, !dbg !45 + %invariant.gep474 = getelementptr i8, ptr addrspace(3) %invariant.gep472, i32 %367, !dbg !45 + %368 = shl nuw nsw i32 %327, 3 + %invariant.gep476 = getelementptr i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 156160), i32 %368, !dbg !45 + %369 = icmp eq i32 %36, 0 + %370 = lshr exact i32 ptrtoint (ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072) to i32), 4 + %371 = and i32 %370, 16383 + %372 = zext nneg i32 %371 to i64 + %373 = or disjoint i64 %372, 4611756662049472512 + %374 = add nuw nsw i64 %372, 4611756662049472640 + %375 = add nuw nsw i64 %372, 4611756662049472768 + %376 = add nuw nsw i64 %372, 4611756662049472896 + %377 = lshr exact i32 ptrtoint (ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 139264) to i32), 4 + %378 = and i32 %377, 16383 + %379 = zext nneg i32 %378 to i64 + %380 = or disjoint i64 %379, 4611756662049472512 + %381 = add nuw nsw i64 %379, 4611756662049472640 + %382 = add nuw nsw i64 %379, 4611756662049472768 + %383 = add nuw nsw i64 %379, 4611756662049472896 + %invariant.gep478 = getelementptr i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 49152), i32 %324, !dbg !45 + %invariant.gep480 = getelementptr i8, ptr addrspace(3) %invariant.gep478, i32 %325, !dbg !45 + %invariant.gep482 = getelementptr i8, ptr addrspace(3) %invariant.gep480, i32 %328, !dbg !45 + %384 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 147456), i32 %331 + %385 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 147456), i32 %333 + %386 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 147456), i32 %335 + %387 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 147456), i32 %337 + %388 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %346 + %389 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %356 + %390 = shl nuw nsw i32 %45, 1 + %391 = shl nuw nsw i32 %327, 8 + %392 = or disjoint i32 %390, %391 + %393 = or disjoint i32 %392, %46 + %394 = or disjoint i32 %393, 128 + %395 = or disjoint i32 %393, 256 + %396 = or disjoint i32 %393, 384 + %397 = or disjoint i32 %393, 512 + %398 = or disjoint i32 %393, 640 + %399 = or disjoint i32 %393, 768 + %400 = or disjoint i32 %393, 896 + %401 = or disjoint i32 %393, 1024 + %402 = or disjoint i32 %393, 1152 + %403 = or disjoint i32 %393, 1280 + %404 = or disjoint i32 %393, 1408 + %405 = or disjoint i32 %393, 1536 + %406 = or disjoint i32 %393, 1664 + %407 = or disjoint i32 %393, 1792 + %408 = or disjoint i32 %393, 1920 + %409 = or disjoint i32 %393, 2048 + %410 = or disjoint i32 %393, 2176 + %411 = or disjoint i32 %393, 2304 + %412 = or disjoint i32 %393, 2432 + %413 = or disjoint i32 %393, 2560 + %414 = or disjoint i32 %393, 2688 + %415 = or disjoint i32 %393, 2816 + %416 = or disjoint i32 %393, 2944 + %417 = or disjoint i32 %393, 3072 + %418 = or disjoint i32 %393, 3200 + %419 = or disjoint i32 %393, 3328 + %420 = or disjoint i32 %393, 3456 + %421 = or disjoint i32 %393, 3584 + %422 = or disjoint i32 %393, 3712 + %423 = or disjoint i32 %393, 3840 + %424 = or disjoint i32 %393, 3968 + %425 = shl nuw nsw i32 %37, 10 + %426 = lshr exact i32 %40, 1 + %427 = or disjoint i32 %425, %426 + %428 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %427 + %429 = getelementptr inbounds nuw i8, ptr addrspace(3) %428, i32 576 + %430 = getelementptr inbounds nuw i8, ptr addrspace(3) %428, i32 64 + %431 = getelementptr inbounds nuw i8, ptr addrspace(3) %428, i32 512 + %432 = getelementptr inbounds nuw i8, ptr addrspace(3) %428, i32 8192 + %433 = getelementptr inbounds nuw i8, ptr addrspace(3) %428, i32 8768 + %434 = getelementptr inbounds nuw i8, ptr addrspace(3) %428, i32 8256 + %435 = getelementptr inbounds nuw i8, ptr addrspace(3) %428, i32 8704 + %436 = xor i32 %427, 144 + %437 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %436 + %438 = getelementptr inbounds nuw i8, ptr addrspace(3) %437, i32 576 + %439 = getelementptr inbounds nuw i8, ptr addrspace(3) %437, i32 64 + %440 = getelementptr inbounds nuw i8, ptr addrspace(3) %437, i32 512 + %441 = getelementptr inbounds nuw i8, ptr addrspace(3) %437, i32 8192 + %442 = getelementptr inbounds nuw i8, ptr addrspace(3) %437, i32 8768 + %443 = getelementptr inbounds nuw i8, ptr addrspace(3) %437, i32 8256 + %444 = getelementptr inbounds nuw i8, ptr addrspace(3) %437, i32 8704 + %445 = xor i32 %427, 288 + %446 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %445 + %447 = getelementptr inbounds nuw i8, ptr addrspace(3) %446, i32 576 + %448 = getelementptr inbounds nuw i8, ptr addrspace(3) %446, i32 64 + %449 = getelementptr inbounds nuw i8, ptr addrspace(3) %446, i32 512 + %450 = getelementptr inbounds nuw i8, ptr addrspace(3) %446, i32 8192 + %451 = getelementptr inbounds nuw i8, ptr addrspace(3) %446, i32 8768 + %452 = getelementptr inbounds nuw i8, ptr addrspace(3) %446, i32 8256 + %453 = getelementptr inbounds nuw i8, ptr addrspace(3) %446, i32 8704 + %454 = xor i32 %427, 432 + %455 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %454 + %456 = getelementptr inbounds nuw i8, ptr addrspace(3) %455, i32 576 + %457 = getelementptr inbounds nuw i8, ptr addrspace(3) %455, i32 64 + %458 = getelementptr inbounds nuw i8, ptr addrspace(3) %455, i32 512 + %459 = getelementptr inbounds nuw i8, ptr addrspace(3) %455, i32 8192 + %460 = getelementptr inbounds nuw i8, ptr addrspace(3) %455, i32 8768 + %461 = getelementptr inbounds nuw i8, ptr addrspace(3) %455, i32 8256 + %462 = getelementptr inbounds nuw i8, ptr addrspace(3) %455, i32 8704 + %463 = add i32 %15, 192 + %464 = add nuw nsw i64 %372, 4611756662049472514 + %465 = add nuw nsw i64 %372, 4611756662049472516 + %466 = add nuw nsw i64 %372, 4611756662049472518 + %467 = add nuw nsw i64 %372, 4611756662049473024 + %468 = add nuw nsw i64 %372, 4611756662049473026 + %469 = add nuw nsw i64 %372, 4611756662049473028 + %470 = add nuw nsw i64 %372, 4611756662049473030 + %471 = lshr exact i32 ptrtoint (ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 147456) to i32), 4 + %472 = and i32 %471, 16383 + %473 = zext nneg i32 %472 to i64 + %474 = or disjoint i64 %473, 4611756662049472512 + %475 = add nuw nsw i64 %473, 4611756662049472640 + %476 = add nuw nsw i64 %473, 4611756662049472768 + %477 = add nuw nsw i64 %473, 4611756662049472896 + %478 = add nsw i32 %24, -4 + %479 = sext i32 %317 to i64, !dbg !45 + %480 = sext i32 %478 to i64, !dbg !45 + %smax = tail call i32 @llvm.smax.i32(i32 %58, i32 0), !dbg !45 + %wide.trip.count = zext nneg i32 %smax to i64, !dbg !45 + %481 = insertelement <64 x float> poison, float %8, i64 0, !dbg !85 + %482 = shufflevector <64 x float> %481, <64 x float> poison, <64 x i32> zeroinitializer, !dbg !85 + br label %483, !dbg !45 + +483: ; preds = %.lr.ph, %1658 + %indvars.iv = phi i64 [ -1, %.lr.ph ], [ %indvars.iv.next, %1658 ] + %.pn = phi i64 [ %67, %.lr.ph ], [ %.pn322462, %1658 ] + %.pn100470 = phi i1 [ %211, %.lr.ph ], [ %1675, %1658 ] + %.pn116469 = phi i1 [ %210, %.lr.ph ], [ %1674, %1658 ] + %.pn132468 = phi i1 [ %209, %.lr.ph ], [ %1673, %1658 ] + %.pn148467 = phi i1 [ %208, %.lr.ph ], [ %1672, %1658 ] + %.pn100.pn466 = phi i1 [ %76, %.lr.ph ], [ %.pn100470, %1658 ] + %.pn116.pn465 = phi i1 [ %75, %.lr.ph ], [ %.pn116469, %1658 ] + %.pn132.pn464 = phi i1 [ %74, %.lr.ph ], [ %.pn132468, %1658 ] + %.pn148.pn463 = phi i1 [ %73, %.lr.ph ], [ %.pn148467, %1658 ] + %.pn322462 = phi i64 [ %202, %.lr.ph ], [ %1666, %1658 ] + %.pn5461 = phi i64 [ %227, %.lr.ph ], [ %1959, %1658 ] + %.pn21460 = phi i64 [ %226, %.lr.ph ], [ %1958, %1658 ] + %.pn37459 = phi i64 [ %225, %.lr.ph ], [ %1957, %1658 ] + %.pn53458 = phi i64 [ %224, %.lr.ph ], [ %1956, %1658 ] + %.pn5.pn457 = phi i64 [ %103, %.lr.ph ], [ %.pn5461, %1658 ] + %.pn21.pn456 = phi i64 [ %102, %.lr.ph ], [ %.pn21460, %1658 ] + %.pn37.pn455 = phi i64 [ %101, %.lr.ph ], [ %.pn37459, %1658 ] + %.pn53.pn454 = phi i64 [ %100, %.lr.ph ], [ %.pn53458, %1658 ] + %484 = phi i64 [ %201, %.lr.ph ], [ %1665, %1658 ] + %485 = phi i64 [ %66, %.lr.ph ], [ %484, %1658 ] + %486 = phi i32 [ -1, %.lr.ph ], [ %492, %1658 ] + %487 = phi i32 [ 1, %.lr.ph ], [ %1664, %1658 ] + %488 = phi <64 x float> [ zeroinitializer, %.lr.ph ], [ %1951, %1658 ] + %.pn.pn98452.pn = or disjoint i64 %.pn, %55, !dbg !48 + %.pn96.pn471 = icmp slt i64 %.pn.pn98452.pn, %21, !dbg !49 + %489 = icmp slt i64 %indvars.iv, %479, !dbg !45 + %490 = add i32 %486, 1, !dbg !45 + %491 = icmp sgt i32 %490, 1, !dbg !45 + %492 = select i1 %491, i32 0, i32 %490, !dbg !45 + %.idx194 = shl nsw i64 %485, 20, !dbg !86 + %493 = getelementptr i8, ptr addrspace(1) %319, i64 %.idx194, !dbg !86 + %494 = getelementptr bfloat, ptr addrspace(1) %493, i64 %320, !dbg !87 + %495 = getelementptr bfloat, ptr addrspace(1) %493, i64 %321, !dbg !87 + %496 = getelementptr bfloat, ptr addrspace(1) %493, i64 %322, !dbg !87 + %497 = getelementptr bfloat, ptr addrspace(1) %493, i64 %323, !dbg !87 + %498 = getelementptr bfloat, ptr addrspace(1) %494, i64 %56, !dbg !88 + %499 = getelementptr bfloat, ptr addrspace(1) %495, i64 %56, !dbg !88 + %500 = getelementptr bfloat, ptr addrspace(1) %496, i64 %56, !dbg !88 + %501 = getelementptr bfloat, ptr addrspace(1) %497, i64 %56, !dbg !88 + %502 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !89 + %503 = fptrunc <2 x float> %502 to <2 x bfloat>, !dbg !89 + %504 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !89 + %505 = fptrunc <2 x float> %504 to <2 x bfloat>, !dbg !89 + %506 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !89 + %507 = fptrunc <2 x float> %506 to <2 x bfloat>, !dbg !89 + %508 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !89 + %509 = fptrunc <2 x float> %508 to <2 x bfloat>, !dbg !89 + %510 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !89 + %511 = fptrunc <2 x float> %510 to <2 x bfloat>, !dbg !89 + %512 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !89 + %513 = fptrunc <2 x float> %512 to <2 x bfloat>, !dbg !89 + %514 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !89 + %515 = fptrunc <2 x float> %514 to <2 x bfloat>, !dbg !89 + %516 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !89 + %517 = fptrunc <2 x float> %516 to <2 x bfloat>, !dbg !89 + %518 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !89 + %519 = fptrunc <2 x float> %518 to <2 x bfloat>, !dbg !89 + %520 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !89 + %521 = fptrunc <2 x float> %520 to <2 x bfloat>, !dbg !89 + %522 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !89 + %523 = fptrunc <2 x float> %522 to <2 x bfloat>, !dbg !89 + %524 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !89 + %525 = fptrunc <2 x float> %524 to <2 x bfloat>, !dbg !89 + %526 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !89 + %527 = fptrunc <2 x float> %526 to <2 x bfloat>, !dbg !89 + %528 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !89 + %529 = fptrunc <2 x float> %528 to <2 x bfloat>, !dbg !89 + %530 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !89 + %531 = fptrunc <2 x float> %530 to <2 x bfloat>, !dbg !89 + %532 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !89 + %533 = fptrunc <2 x float> %532 to <2 x bfloat>, !dbg !89 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !89 + %534 = bitcast <2 x bfloat> %503 to i32, !dbg !89 + %535 = bitcast <2 x bfloat> %505 to i32, !dbg !89 + %536 = bitcast <2 x bfloat> %507 to i32, !dbg !89 + %537 = bitcast <2 x bfloat> %509 to i32, !dbg !89 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %332, i32 %534, i32 %535, i32 %536, i32 %537) #4, !dbg !89 + %538 = bitcast <2 x bfloat> %511 to i32, !dbg !89 + %539 = bitcast <2 x bfloat> %513 to i32, !dbg !89 + %540 = bitcast <2 x bfloat> %515 to i32, !dbg !89 + %541 = bitcast <2 x bfloat> %517 to i32, !dbg !89 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %334, i32 %538, i32 %539, i32 %540, i32 %541) #4, !dbg !89 + %542 = bitcast <2 x bfloat> %519 to i32, !dbg !89 + %543 = bitcast <2 x bfloat> %521 to i32, !dbg !89 + %544 = bitcast <2 x bfloat> %523 to i32, !dbg !89 + %545 = bitcast <2 x bfloat> %525 to i32, !dbg !89 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %336, i32 %542, i32 %543, i32 %544, i32 %545) #4, !dbg !89 + %546 = bitcast <2 x bfloat> %527 to i32, !dbg !89 + %547 = bitcast <2 x bfloat> %529 to i32, !dbg !89 + %548 = bitcast <2 x bfloat> %531 to i32, !dbg !89 + %549 = bitcast <2 x bfloat> %533 to i32, !dbg !89 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %338, i32 %546, i32 %547, i32 %548, i32 %549) #4, !dbg !89 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %347, i32 %534, i32 %538, i32 %542, i32 %546) #4, !dbg !90 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !90 + %550 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %357), !dbg !90 + %551 = extractvalue { i32, i32, i32, i32 } %550, 0, !dbg !90 + %552 = extractvalue { i32, i32, i32, i32 } %550, 1, !dbg !90 + %553 = extractvalue { i32, i32, i32, i32 } %550, 2, !dbg !90 + %554 = extractvalue { i32, i32, i32, i32 } %550, 3, !dbg !90 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !90 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %347, i32 %535, i32 %539, i32 %543, i32 %547) #4, !dbg !90 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !90 + %555 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %357), !dbg !90 + %556 = extractvalue { i32, i32, i32, i32 } %555, 0, !dbg !90 + %557 = extractvalue { i32, i32, i32, i32 } %555, 1, !dbg !90 + %558 = extractvalue { i32, i32, i32, i32 } %555, 2, !dbg !90 + %559 = extractvalue { i32, i32, i32, i32 } %555, 3, !dbg !90 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !90 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %347, i32 %536, i32 %540, i32 %544, i32 %548) #4, !dbg !90 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !90 + %560 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %357), !dbg !90 + %561 = extractvalue { i32, i32, i32, i32 } %560, 0, !dbg !90 + %562 = extractvalue { i32, i32, i32, i32 } %560, 1, !dbg !90 + %563 = extractvalue { i32, i32, i32, i32 } %560, 2, !dbg !90 + %564 = extractvalue { i32, i32, i32, i32 } %560, 3, !dbg !90 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !90 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %347, i32 %537, i32 %541, i32 %545, i32 %549) #4, !dbg !90 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !90 + %565 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %357), !dbg !90 + %566 = extractvalue { i32, i32, i32, i32 } %565, 0, !dbg !90 + %567 = extractvalue { i32, i32, i32, i32 } %565, 1, !dbg !90 + %568 = extractvalue { i32, i32, i32, i32 } %565, 2, !dbg !90 + %569 = extractvalue { i32, i32, i32, i32 } %565, 3, !dbg !90 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %551, i32 %556, i32 %561, i32 %566, ptr addrspace(1) %498, i1 true) #4, !dbg !90 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %552, i32 %557, i32 %562, i32 %567, ptr addrspace(1) %499, i1 true) #4, !dbg !90 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %553, i32 %558, i32 %563, i32 %568, ptr addrspace(1) %500, i1 true) #4, !dbg !90 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %554, i32 %559, i32 %564, i32 %569, ptr addrspace(1) %501, i1 true) #4, !dbg !90 + %570 = getelementptr i8, ptr addrspace(1) %494, i64 16384, !dbg !91 + %571 = getelementptr i8, ptr addrspace(1) %494, i64 20480, !dbg !91 + %572 = getelementptr i8, ptr addrspace(1) %494, i64 24576, !dbg !91 + %573 = getelementptr i8, ptr addrspace(1) %494, i64 28672, !dbg !91 + %574 = getelementptr bfloat, ptr addrspace(1) %570, i64 %56, !dbg !92 + %575 = getelementptr bfloat, ptr addrspace(1) %571, i64 %56, !dbg !92 + %576 = getelementptr bfloat, ptr addrspace(1) %572, i64 %56, !dbg !92 + %577 = getelementptr bfloat, ptr addrspace(1) %573, i64 %56, !dbg !92 + %578 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !93 + %579 = fptrunc <2 x float> %578 to <2 x bfloat>, !dbg !93 + %580 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !93 + %581 = fptrunc <2 x float> %580 to <2 x bfloat>, !dbg !93 + %582 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !93 + %583 = fptrunc <2 x float> %582 to <2 x bfloat>, !dbg !93 + %584 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !93 + %585 = fptrunc <2 x float> %584 to <2 x bfloat>, !dbg !93 + %586 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !93 + %587 = fptrunc <2 x float> %586 to <2 x bfloat>, !dbg !93 + %588 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !93 + %589 = fptrunc <2 x float> %588 to <2 x bfloat>, !dbg !93 + %590 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !93 + %591 = fptrunc <2 x float> %590 to <2 x bfloat>, !dbg !93 + %592 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !93 + %593 = fptrunc <2 x float> %592 to <2 x bfloat>, !dbg !93 + %594 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !93 + %595 = fptrunc <2 x float> %594 to <2 x bfloat>, !dbg !93 + %596 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !93 + %597 = fptrunc <2 x float> %596 to <2 x bfloat>, !dbg !93 + %598 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !93 + %599 = fptrunc <2 x float> %598 to <2 x bfloat>, !dbg !93 + %600 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !93 + %601 = fptrunc <2 x float> %600 to <2 x bfloat>, !dbg !93 + %602 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !93 + %603 = fptrunc <2 x float> %602 to <2 x bfloat>, !dbg !93 + %604 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !93 + %605 = fptrunc <2 x float> %604 to <2 x bfloat>, !dbg !93 + %606 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !93 + %607 = fptrunc <2 x float> %606 to <2 x bfloat>, !dbg !93 + %608 = shufflevector <64 x float> %488, <64 x float> poison, <2 x i32> , !dbg !93 + %609 = fptrunc <2 x float> %608 to <2 x bfloat>, !dbg !93 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !93 + %610 = bitcast <2 x bfloat> %579 to i32, !dbg !93 + %611 = bitcast <2 x bfloat> %581 to i32, !dbg !93 + %612 = bitcast <2 x bfloat> %583 to i32, !dbg !93 + %613 = bitcast <2 x bfloat> %585 to i32, !dbg !93 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %358, i32 %610, i32 %611, i32 %612, i32 %613) #4, !dbg !93 + %614 = bitcast <2 x bfloat> %587 to i32, !dbg !93 + %615 = bitcast <2 x bfloat> %589 to i32, !dbg !93 + %616 = bitcast <2 x bfloat> %591 to i32, !dbg !93 + %617 = bitcast <2 x bfloat> %593 to i32, !dbg !93 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %359, i32 %614, i32 %615, i32 %616, i32 %617) #4, !dbg !93 + %618 = bitcast <2 x bfloat> %595 to i32, !dbg !93 + %619 = bitcast <2 x bfloat> %597 to i32, !dbg !93 + %620 = bitcast <2 x bfloat> %599 to i32, !dbg !93 + %621 = bitcast <2 x bfloat> %601 to i32, !dbg !93 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %360, i32 %618, i32 %619, i32 %620, i32 %621) #4, !dbg !93 + %622 = bitcast <2 x bfloat> %603 to i32, !dbg !93 + %623 = bitcast <2 x bfloat> %605 to i32, !dbg !93 + %624 = bitcast <2 x bfloat> %607 to i32, !dbg !93 + %625 = bitcast <2 x bfloat> %609 to i32, !dbg !93 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %361, i32 %622, i32 %623, i32 %624, i32 %625) #4, !dbg !93 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %362, i32 %610, i32 %614, i32 %618, i32 %622) #4, !dbg !94 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !94 + %626 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %363), !dbg !94 + %627 = extractvalue { i32, i32, i32, i32 } %626, 0, !dbg !94 + %628 = extractvalue { i32, i32, i32, i32 } %626, 1, !dbg !94 + %629 = extractvalue { i32, i32, i32, i32 } %626, 2, !dbg !94 + %630 = extractvalue { i32, i32, i32, i32 } %626, 3, !dbg !94 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !94 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %362, i32 %611, i32 %615, i32 %619, i32 %623) #4, !dbg !94 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !94 + %631 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %363), !dbg !94 + %632 = extractvalue { i32, i32, i32, i32 } %631, 0, !dbg !94 + %633 = extractvalue { i32, i32, i32, i32 } %631, 1, !dbg !94 + %634 = extractvalue { i32, i32, i32, i32 } %631, 2, !dbg !94 + %635 = extractvalue { i32, i32, i32, i32 } %631, 3, !dbg !94 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !94 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %362, i32 %612, i32 %616, i32 %620, i32 %624) #4, !dbg !94 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !94 + %636 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %363), !dbg !94 + %637 = extractvalue { i32, i32, i32, i32 } %636, 0, !dbg !94 + %638 = extractvalue { i32, i32, i32, i32 } %636, 1, !dbg !94 + %639 = extractvalue { i32, i32, i32, i32 } %636, 2, !dbg !94 + %640 = extractvalue { i32, i32, i32, i32 } %636, 3, !dbg !94 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !94 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %362, i32 %613, i32 %617, i32 %621, i32 %625) #4, !dbg !94 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !94 + %641 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %363), !dbg !94 + %642 = extractvalue { i32, i32, i32, i32 } %641, 0, !dbg !94 + %643 = extractvalue { i32, i32, i32, i32 } %641, 1, !dbg !94 + %644 = extractvalue { i32, i32, i32, i32 } %641, 2, !dbg !94 + %645 = extractvalue { i32, i32, i32, i32 } %641, 3, !dbg !94 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %627, i32 %632, i32 %637, i32 %642, ptr addrspace(1) %574, i1 true) #4, !dbg !94 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %628, i32 %633, i32 %638, i32 %643, ptr addrspace(1) %575, i1 true) #4, !dbg !94 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %629, i32 %634, i32 %639, i32 %644, ptr addrspace(1) %576, i1 true) #4, !dbg !94 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %630, i32 %635, i32 %640, i32 %645, ptr addrspace(1) %577, i1 true) #4, !dbg !94 + tail call void @llvm.nvvm.cp.async.wait.group(i32 11), !dbg !55 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !55 + %646 = getelementptr float, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 156672), i32 %492, !dbg !55 + %647 = load float, ptr addrspace(3) %646, align 4, !dbg !55 + %648 = shl i32 %492, 6, !dbg !62 + %gep475 = getelementptr float, ptr addrspace(3) %invariant.gep474, i32 %648, !dbg !62 + %649 = tail call i32 @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x1.b16.p3(ptr addrspace(3) %gep475), !dbg !62 + %650 = bitcast i32 %649 to float, !dbg !62 + %gep477 = getelementptr float, ptr addrspace(3) %invariant.gep476, i32 %648, !dbg !62 + %651 = load float, ptr addrspace(3) %gep477, align 16, !dbg !62 + %652 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 4, !dbg !62 + %653 = load float, ptr addrspace(3) %652, align 4, !dbg !62 + %654 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 8, !dbg !62 + %655 = load float, ptr addrspace(3) %654, align 8, !dbg !62 + %656 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 12, !dbg !62 + %657 = load float, ptr addrspace(3) %656, align 4, !dbg !62 + %658 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 16, !dbg !62 + %659 = load float, ptr addrspace(3) %658, align 16, !dbg !62 + %660 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 20, !dbg !62 + %661 = load float, ptr addrspace(3) %660, align 4, !dbg !62 + %662 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 24, !dbg !62 + %663 = load float, ptr addrspace(3) %662, align 8, !dbg !62 + %664 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 28, !dbg !62 + %665 = load float, ptr addrspace(3) %664, align 4, !dbg !62 + %666 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 32, !dbg !62 + %667 = load float, ptr addrspace(3) %666, align 16, !dbg !62 + %668 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 36, !dbg !62 + %669 = load float, ptr addrspace(3) %668, align 4, !dbg !62 + %670 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 40, !dbg !62 + %671 = load float, ptr addrspace(3) %670, align 8, !dbg !62 + %672 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 44, !dbg !62 + %673 = load float, ptr addrspace(3) %672, align 4, !dbg !62 + %674 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 48, !dbg !62 + %675 = load float, ptr addrspace(3) %674, align 16, !dbg !62 + %676 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 52, !dbg !62 + %677 = load float, ptr addrspace(3) %676, align 4, !dbg !62 + %678 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 56, !dbg !62 + %679 = load float, ptr addrspace(3) %678, align 8, !dbg !62 + %680 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 60, !dbg !62 + %681 = load float, ptr addrspace(3) %680, align 4, !dbg !62 + %682 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 64, !dbg !62 + %683 = load float, ptr addrspace(3) %682, align 16, !dbg !62 + %684 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 68, !dbg !62 + %685 = load float, ptr addrspace(3) %684, align 4, !dbg !62 + %686 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 72, !dbg !62 + %687 = load float, ptr addrspace(3) %686, align 8, !dbg !62 + %688 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 76, !dbg !62 + %689 = load float, ptr addrspace(3) %688, align 4, !dbg !62 + %690 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 80, !dbg !62 + %691 = load float, ptr addrspace(3) %690, align 16, !dbg !62 + %692 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 84, !dbg !62 + %693 = load float, ptr addrspace(3) %692, align 4, !dbg !62 + %694 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 88, !dbg !62 + %695 = load float, ptr addrspace(3) %694, align 8, !dbg !62 + %696 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 92, !dbg !62 + %697 = load float, ptr addrspace(3) %696, align 4, !dbg !62 + %698 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 96, !dbg !62 + %699 = load float, ptr addrspace(3) %698, align 16, !dbg !62 + %700 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 100, !dbg !62 + %701 = load float, ptr addrspace(3) %700, align 4, !dbg !62 + %702 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 104, !dbg !62 + %703 = load float, ptr addrspace(3) %702, align 8, !dbg !62 + %704 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 108, !dbg !62 + %705 = load float, ptr addrspace(3) %704, align 4, !dbg !62 + %706 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 112, !dbg !62 + %707 = load float, ptr addrspace(3) %706, align 16, !dbg !62 + %708 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 116, !dbg !62 + %709 = load float, ptr addrspace(3) %708, align 4, !dbg !62 + %710 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 120, !dbg !62 + %711 = load float, ptr addrspace(3) %710, align 8, !dbg !62 + %712 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep477, i32 124, !dbg !62 + %713 = load float, ptr addrspace(3) %712, align 4, !dbg !62 + %714 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !95 + %.not.i = icmp eq i32 %714, 0, !dbg !95 + br i1 %.not.i, label %717, label %715, !dbg !95 + +715: ; preds = %483 + %716 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %647), !dbg !95 + br label %__nv_exp2f.exit, !dbg !95 + +717: ; preds = %483 + %718 = tail call float @llvm.nvvm.ex2.approx.f32(float %647), !dbg !95 + br label %__nv_exp2f.exit, !dbg !95 + +__nv_exp2f.exit: ; preds = %715, %717 + %.0.i = phi float [ %716, %715 ], [ %718, %717 ], !dbg !95 + %719 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i325 = icmp eq i32 %719, 0, !dbg !99 + br i1 %.not.i325, label %722, label %720, !dbg !99 + +720: ; preds = %__nv_exp2f.exit + %721 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %651), !dbg !99 + br label %__nv_exp2f.exit327, !dbg !99 + +722: ; preds = %__nv_exp2f.exit + %723 = tail call float @llvm.nvvm.ex2.approx.f32(float %651), !dbg !99 + br label %__nv_exp2f.exit327, !dbg !99 + +__nv_exp2f.exit327: ; preds = %720, %722 + %.0.i326 = phi float [ %721, %720 ], [ %723, %722 ], !dbg !99 + %724 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i328 = icmp eq i32 %724, 0, !dbg !99 + br i1 %.not.i328, label %727, label %725, !dbg !99 + +725: ; preds = %__nv_exp2f.exit327 + %726 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %653), !dbg !99 + br label %__nv_exp2f.exit330, !dbg !99 + +727: ; preds = %__nv_exp2f.exit327 + %728 = tail call float @llvm.nvvm.ex2.approx.f32(float %653), !dbg !99 + br label %__nv_exp2f.exit330, !dbg !99 + +__nv_exp2f.exit330: ; preds = %725, %727 + %.0.i329 = phi float [ %726, %725 ], [ %728, %727 ], !dbg !99 + %729 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i331 = icmp eq i32 %729, 0, !dbg !99 + br i1 %.not.i331, label %732, label %730, !dbg !99 + +730: ; preds = %__nv_exp2f.exit330 + %731 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %655), !dbg !99 + br label %__nv_exp2f.exit333, !dbg !99 + +732: ; preds = %__nv_exp2f.exit330 + %733 = tail call float @llvm.nvvm.ex2.approx.f32(float %655), !dbg !99 + br label %__nv_exp2f.exit333, !dbg !99 + +__nv_exp2f.exit333: ; preds = %730, %732 + %.0.i332 = phi float [ %731, %730 ], [ %733, %732 ], !dbg !99 + %734 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i334 = icmp eq i32 %734, 0, !dbg !99 + br i1 %.not.i334, label %737, label %735, !dbg !99 + +735: ; preds = %__nv_exp2f.exit333 + %736 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %657), !dbg !99 + br label %__nv_exp2f.exit336, !dbg !99 + +737: ; preds = %__nv_exp2f.exit333 + %738 = tail call float @llvm.nvvm.ex2.approx.f32(float %657), !dbg !99 + br label %__nv_exp2f.exit336, !dbg !99 + +__nv_exp2f.exit336: ; preds = %735, %737 + %.0.i335 = phi float [ %736, %735 ], [ %738, %737 ], !dbg !99 + %739 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i337 = icmp eq i32 %739, 0, !dbg !99 + br i1 %.not.i337, label %742, label %740, !dbg !99 + +740: ; preds = %__nv_exp2f.exit336 + %741 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %659), !dbg !99 + br label %__nv_exp2f.exit339, !dbg !99 + +742: ; preds = %__nv_exp2f.exit336 + %743 = tail call float @llvm.nvvm.ex2.approx.f32(float %659), !dbg !99 + br label %__nv_exp2f.exit339, !dbg !99 + +__nv_exp2f.exit339: ; preds = %740, %742 + %.0.i338 = phi float [ %741, %740 ], [ %743, %742 ], !dbg !99 + %744 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i340 = icmp eq i32 %744, 0, !dbg !99 + br i1 %.not.i340, label %747, label %745, !dbg !99 + +745: ; preds = %__nv_exp2f.exit339 + %746 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %661), !dbg !99 + br label %__nv_exp2f.exit342, !dbg !99 + +747: ; preds = %__nv_exp2f.exit339 + %748 = tail call float @llvm.nvvm.ex2.approx.f32(float %661), !dbg !99 + br label %__nv_exp2f.exit342, !dbg !99 + +__nv_exp2f.exit342: ; preds = %745, %747 + %.0.i341 = phi float [ %746, %745 ], [ %748, %747 ], !dbg !99 + %749 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i343 = icmp eq i32 %749, 0, !dbg !99 + br i1 %.not.i343, label %752, label %750, !dbg !99 + +750: ; preds = %__nv_exp2f.exit342 + %751 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %663), !dbg !99 + br label %__nv_exp2f.exit345, !dbg !99 + +752: ; preds = %__nv_exp2f.exit342 + %753 = tail call float @llvm.nvvm.ex2.approx.f32(float %663), !dbg !99 + br label %__nv_exp2f.exit345, !dbg !99 + +__nv_exp2f.exit345: ; preds = %750, %752 + %.0.i344 = phi float [ %751, %750 ], [ %753, %752 ], !dbg !99 + %754 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i346 = icmp eq i32 %754, 0, !dbg !99 + br i1 %.not.i346, label %757, label %755, !dbg !99 + +755: ; preds = %__nv_exp2f.exit345 + %756 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %665), !dbg !99 + br label %__nv_exp2f.exit348, !dbg !99 + +757: ; preds = %__nv_exp2f.exit345 + %758 = tail call float @llvm.nvvm.ex2.approx.f32(float %665), !dbg !99 + br label %__nv_exp2f.exit348, !dbg !99 + +__nv_exp2f.exit348: ; preds = %755, %757 + %.0.i347 = phi float [ %756, %755 ], [ %758, %757 ], !dbg !99 + %759 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i349 = icmp eq i32 %759, 0, !dbg !99 + br i1 %.not.i349, label %762, label %760, !dbg !99 + +760: ; preds = %__nv_exp2f.exit348 + %761 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %667), !dbg !99 + br label %__nv_exp2f.exit351, !dbg !99 + +762: ; preds = %__nv_exp2f.exit348 + %763 = tail call float @llvm.nvvm.ex2.approx.f32(float %667), !dbg !99 + br label %__nv_exp2f.exit351, !dbg !99 + +__nv_exp2f.exit351: ; preds = %760, %762 + %.0.i350 = phi float [ %761, %760 ], [ %763, %762 ], !dbg !99 + %764 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i352 = icmp eq i32 %764, 0, !dbg !99 + br i1 %.not.i352, label %767, label %765, !dbg !99 + +765: ; preds = %__nv_exp2f.exit351 + %766 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %669), !dbg !99 + br label %__nv_exp2f.exit354, !dbg !99 + +767: ; preds = %__nv_exp2f.exit351 + %768 = tail call float @llvm.nvvm.ex2.approx.f32(float %669), !dbg !99 + br label %__nv_exp2f.exit354, !dbg !99 + +__nv_exp2f.exit354: ; preds = %765, %767 + %.0.i353 = phi float [ %766, %765 ], [ %768, %767 ], !dbg !99 + %769 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i355 = icmp eq i32 %769, 0, !dbg !99 + br i1 %.not.i355, label %772, label %770, !dbg !99 + +770: ; preds = %__nv_exp2f.exit354 + %771 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %671), !dbg !99 + br label %__nv_exp2f.exit357, !dbg !99 + +772: ; preds = %__nv_exp2f.exit354 + %773 = tail call float @llvm.nvvm.ex2.approx.f32(float %671), !dbg !99 + br label %__nv_exp2f.exit357, !dbg !99 + +__nv_exp2f.exit357: ; preds = %770, %772 + %.0.i356 = phi float [ %771, %770 ], [ %773, %772 ], !dbg !99 + %774 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i358 = icmp eq i32 %774, 0, !dbg !99 + br i1 %.not.i358, label %777, label %775, !dbg !99 + +775: ; preds = %__nv_exp2f.exit357 + %776 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %673), !dbg !99 + br label %__nv_exp2f.exit360, !dbg !99 + +777: ; preds = %__nv_exp2f.exit357 + %778 = tail call float @llvm.nvvm.ex2.approx.f32(float %673), !dbg !99 + br label %__nv_exp2f.exit360, !dbg !99 + +__nv_exp2f.exit360: ; preds = %775, %777 + %.0.i359 = phi float [ %776, %775 ], [ %778, %777 ], !dbg !99 + %779 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i361 = icmp eq i32 %779, 0, !dbg !99 + br i1 %.not.i361, label %782, label %780, !dbg !99 + +780: ; preds = %__nv_exp2f.exit360 + %781 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %675), !dbg !99 + br label %__nv_exp2f.exit363, !dbg !99 + +782: ; preds = %__nv_exp2f.exit360 + %783 = tail call float @llvm.nvvm.ex2.approx.f32(float %675), !dbg !99 + br label %__nv_exp2f.exit363, !dbg !99 + +__nv_exp2f.exit363: ; preds = %780, %782 + %.0.i362 = phi float [ %781, %780 ], [ %783, %782 ], !dbg !99 + %784 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i364 = icmp eq i32 %784, 0, !dbg !99 + br i1 %.not.i364, label %787, label %785, !dbg !99 + +785: ; preds = %__nv_exp2f.exit363 + %786 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %677), !dbg !99 + br label %__nv_exp2f.exit366, !dbg !99 + +787: ; preds = %__nv_exp2f.exit363 + %788 = tail call float @llvm.nvvm.ex2.approx.f32(float %677), !dbg !99 + br label %__nv_exp2f.exit366, !dbg !99 + +__nv_exp2f.exit366: ; preds = %785, %787 + %.0.i365 = phi float [ %786, %785 ], [ %788, %787 ], !dbg !99 + %789 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i367 = icmp eq i32 %789, 0, !dbg !99 + br i1 %.not.i367, label %792, label %790, !dbg !99 + +790: ; preds = %__nv_exp2f.exit366 + %791 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %679), !dbg !99 + br label %__nv_exp2f.exit369, !dbg !99 + +792: ; preds = %__nv_exp2f.exit366 + %793 = tail call float @llvm.nvvm.ex2.approx.f32(float %679), !dbg !99 + br label %__nv_exp2f.exit369, !dbg !99 + +__nv_exp2f.exit369: ; preds = %790, %792 + %.0.i368 = phi float [ %791, %790 ], [ %793, %792 ], !dbg !99 + %794 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i370 = icmp eq i32 %794, 0, !dbg !99 + br i1 %.not.i370, label %797, label %795, !dbg !99 + +795: ; preds = %__nv_exp2f.exit369 + %796 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %681), !dbg !99 + br label %__nv_exp2f.exit372, !dbg !99 + +797: ; preds = %__nv_exp2f.exit369 + %798 = tail call float @llvm.nvvm.ex2.approx.f32(float %681), !dbg !99 + br label %__nv_exp2f.exit372, !dbg !99 + +__nv_exp2f.exit372: ; preds = %795, %797 + %.0.i371 = phi float [ %796, %795 ], [ %798, %797 ], !dbg !99 + %799 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i373 = icmp eq i32 %799, 0, !dbg !99 + br i1 %.not.i373, label %802, label %800, !dbg !99 + +800: ; preds = %__nv_exp2f.exit372 + %801 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %683), !dbg !99 + br label %__nv_exp2f.exit375, !dbg !99 + +802: ; preds = %__nv_exp2f.exit372 + %803 = tail call float @llvm.nvvm.ex2.approx.f32(float %683), !dbg !99 + br label %__nv_exp2f.exit375, !dbg !99 + +__nv_exp2f.exit375: ; preds = %800, %802 + %.0.i374 = phi float [ %801, %800 ], [ %803, %802 ], !dbg !99 + %804 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i376 = icmp eq i32 %804, 0, !dbg !99 + br i1 %.not.i376, label %807, label %805, !dbg !99 + +805: ; preds = %__nv_exp2f.exit375 + %806 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %685), !dbg !99 + br label %__nv_exp2f.exit378, !dbg !99 + +807: ; preds = %__nv_exp2f.exit375 + %808 = tail call float @llvm.nvvm.ex2.approx.f32(float %685), !dbg !99 + br label %__nv_exp2f.exit378, !dbg !99 + +__nv_exp2f.exit378: ; preds = %805, %807 + %.0.i377 = phi float [ %806, %805 ], [ %808, %807 ], !dbg !99 + %809 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i379 = icmp eq i32 %809, 0, !dbg !99 + br i1 %.not.i379, label %812, label %810, !dbg !99 + +810: ; preds = %__nv_exp2f.exit378 + %811 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %687), !dbg !99 + br label %__nv_exp2f.exit381, !dbg !99 + +812: ; preds = %__nv_exp2f.exit378 + %813 = tail call float @llvm.nvvm.ex2.approx.f32(float %687), !dbg !99 + br label %__nv_exp2f.exit381, !dbg !99 + +__nv_exp2f.exit381: ; preds = %810, %812 + %.0.i380 = phi float [ %811, %810 ], [ %813, %812 ], !dbg !99 + %814 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i382 = icmp eq i32 %814, 0, !dbg !99 + br i1 %.not.i382, label %817, label %815, !dbg !99 + +815: ; preds = %__nv_exp2f.exit381 + %816 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %689), !dbg !99 + br label %__nv_exp2f.exit384, !dbg !99 + +817: ; preds = %__nv_exp2f.exit381 + %818 = tail call float @llvm.nvvm.ex2.approx.f32(float %689), !dbg !99 + br label %__nv_exp2f.exit384, !dbg !99 + +__nv_exp2f.exit384: ; preds = %815, %817 + %.0.i383 = phi float [ %816, %815 ], [ %818, %817 ], !dbg !99 + %819 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i385 = icmp eq i32 %819, 0, !dbg !99 + br i1 %.not.i385, label %822, label %820, !dbg !99 + +820: ; preds = %__nv_exp2f.exit384 + %821 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %691), !dbg !99 + br label %__nv_exp2f.exit387, !dbg !99 + +822: ; preds = %__nv_exp2f.exit384 + %823 = tail call float @llvm.nvvm.ex2.approx.f32(float %691), !dbg !99 + br label %__nv_exp2f.exit387, !dbg !99 + +__nv_exp2f.exit387: ; preds = %820, %822 + %.0.i386 = phi float [ %821, %820 ], [ %823, %822 ], !dbg !99 + %824 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i388 = icmp eq i32 %824, 0, !dbg !99 + br i1 %.not.i388, label %827, label %825, !dbg !99 + +825: ; preds = %__nv_exp2f.exit387 + %826 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %693), !dbg !99 + br label %__nv_exp2f.exit390, !dbg !99 + +827: ; preds = %__nv_exp2f.exit387 + %828 = tail call float @llvm.nvvm.ex2.approx.f32(float %693), !dbg !99 + br label %__nv_exp2f.exit390, !dbg !99 + +__nv_exp2f.exit390: ; preds = %825, %827 + %.0.i389 = phi float [ %826, %825 ], [ %828, %827 ], !dbg !99 + %829 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i391 = icmp eq i32 %829, 0, !dbg !99 + br i1 %.not.i391, label %832, label %830, !dbg !99 + +830: ; preds = %__nv_exp2f.exit390 + %831 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %695), !dbg !99 + br label %__nv_exp2f.exit393, !dbg !99 + +832: ; preds = %__nv_exp2f.exit390 + %833 = tail call float @llvm.nvvm.ex2.approx.f32(float %695), !dbg !99 + br label %__nv_exp2f.exit393, !dbg !99 + +__nv_exp2f.exit393: ; preds = %830, %832 + %.0.i392 = phi float [ %831, %830 ], [ %833, %832 ], !dbg !99 + %834 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i394 = icmp eq i32 %834, 0, !dbg !99 + br i1 %.not.i394, label %837, label %835, !dbg !99 + +835: ; preds = %__nv_exp2f.exit393 + %836 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %697), !dbg !99 + br label %__nv_exp2f.exit396, !dbg !99 + +837: ; preds = %__nv_exp2f.exit393 + %838 = tail call float @llvm.nvvm.ex2.approx.f32(float %697), !dbg !99 + br label %__nv_exp2f.exit396, !dbg !99 + +__nv_exp2f.exit396: ; preds = %835, %837 + %.0.i395 = phi float [ %836, %835 ], [ %838, %837 ], !dbg !99 + %839 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i397 = icmp eq i32 %839, 0, !dbg !99 + br i1 %.not.i397, label %842, label %840, !dbg !99 + +840: ; preds = %__nv_exp2f.exit396 + %841 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %699), !dbg !99 + br label %__nv_exp2f.exit399, !dbg !99 + +842: ; preds = %__nv_exp2f.exit396 + %843 = tail call float @llvm.nvvm.ex2.approx.f32(float %699), !dbg !99 + br label %__nv_exp2f.exit399, !dbg !99 + +__nv_exp2f.exit399: ; preds = %840, %842 + %.0.i398 = phi float [ %841, %840 ], [ %843, %842 ], !dbg !99 + %844 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i400 = icmp eq i32 %844, 0, !dbg !99 + br i1 %.not.i400, label %847, label %845, !dbg !99 + +845: ; preds = %__nv_exp2f.exit399 + %846 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %701), !dbg !99 + br label %__nv_exp2f.exit402, !dbg !99 + +847: ; preds = %__nv_exp2f.exit399 + %848 = tail call float @llvm.nvvm.ex2.approx.f32(float %701), !dbg !99 + br label %__nv_exp2f.exit402, !dbg !99 + +__nv_exp2f.exit402: ; preds = %845, %847 + %.0.i401 = phi float [ %846, %845 ], [ %848, %847 ], !dbg !99 + %849 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i403 = icmp eq i32 %849, 0, !dbg !99 + br i1 %.not.i403, label %852, label %850, !dbg !99 + +850: ; preds = %__nv_exp2f.exit402 + %851 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %703), !dbg !99 + br label %__nv_exp2f.exit405, !dbg !99 + +852: ; preds = %__nv_exp2f.exit402 + %853 = tail call float @llvm.nvvm.ex2.approx.f32(float %703), !dbg !99 + br label %__nv_exp2f.exit405, !dbg !99 + +__nv_exp2f.exit405: ; preds = %850, %852 + %.0.i404 = phi float [ %851, %850 ], [ %853, %852 ], !dbg !99 + %854 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i406 = icmp eq i32 %854, 0, !dbg !99 + br i1 %.not.i406, label %857, label %855, !dbg !99 + +855: ; preds = %__nv_exp2f.exit405 + %856 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %705), !dbg !99 + br label %__nv_exp2f.exit408, !dbg !99 + +857: ; preds = %__nv_exp2f.exit405 + %858 = tail call float @llvm.nvvm.ex2.approx.f32(float %705), !dbg !99 + br label %__nv_exp2f.exit408, !dbg !99 + +__nv_exp2f.exit408: ; preds = %855, %857 + %.0.i407 = phi float [ %856, %855 ], [ %858, %857 ], !dbg !99 + %859 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i409 = icmp eq i32 %859, 0, !dbg !99 + br i1 %.not.i409, label %862, label %860, !dbg !99 + +860: ; preds = %__nv_exp2f.exit408 + %861 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %707), !dbg !99 + br label %__nv_exp2f.exit411, !dbg !99 + +862: ; preds = %__nv_exp2f.exit408 + %863 = tail call float @llvm.nvvm.ex2.approx.f32(float %707), !dbg !99 + br label %__nv_exp2f.exit411, !dbg !99 + +__nv_exp2f.exit411: ; preds = %860, %862 + %.0.i410 = phi float [ %861, %860 ], [ %863, %862 ], !dbg !99 + %864 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i412 = icmp eq i32 %864, 0, !dbg !99 + br i1 %.not.i412, label %867, label %865, !dbg !99 + +865: ; preds = %__nv_exp2f.exit411 + %866 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %709), !dbg !99 + br label %__nv_exp2f.exit414, !dbg !99 + +867: ; preds = %__nv_exp2f.exit411 + %868 = tail call float @llvm.nvvm.ex2.approx.f32(float %709), !dbg !99 + br label %__nv_exp2f.exit414, !dbg !99 + +__nv_exp2f.exit414: ; preds = %865, %867 + %.0.i413 = phi float [ %866, %865 ], [ %868, %867 ], !dbg !99 + %869 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i415 = icmp eq i32 %869, 0, !dbg !99 + br i1 %.not.i415, label %872, label %870, !dbg !99 + +870: ; preds = %__nv_exp2f.exit414 + %871 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %711), !dbg !99 + br label %__nv_exp2f.exit417, !dbg !99 + +872: ; preds = %__nv_exp2f.exit414 + %873 = tail call float @llvm.nvvm.ex2.approx.f32(float %711), !dbg !99 + br label %__nv_exp2f.exit417, !dbg !99 + +__nv_exp2f.exit417: ; preds = %870, %872 + %.0.i416 = phi float [ %871, %870 ], [ %873, %872 ], !dbg !99 + %874 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !99 + %.not.i418 = icmp eq i32 %874, 0, !dbg !99 + br i1 %.not.i418, label %877, label %875, !dbg !99 + +875: ; preds = %__nv_exp2f.exit417 + %876 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %713), !dbg !99 + br label %__nv_exp2f.exit420, !dbg !99 + +877: ; preds = %__nv_exp2f.exit417 + %878 = tail call float @llvm.nvvm.ex2.approx.f32(float %713), !dbg !99 + br label %__nv_exp2f.exit420, !dbg !99 + +__nv_exp2f.exit420: ; preds = %875, %877 + %.0.i419 = phi float [ %876, %875 ], [ %878, %877 ], !dbg !99 + %gep = getelementptr bfloat, ptr addrspace(1) %invariant.gep, i64 %.pn53.pn454, !dbg !101 + %gep425 = getelementptr bfloat, ptr addrspace(1) %invariant.gep, i64 %.pn37.pn455, !dbg !101 + %gep427 = getelementptr bfloat, ptr addrspace(1) %invariant.gep, i64 %.pn21.pn456, !dbg !101 + %gep429 = getelementptr bfloat, ptr addrspace(1) %invariant.gep, i64 %.pn5.pn457, !dbg !101 + %879 = shl i32 %492, 12, !dbg !68 + %880 = getelementptr bfloat, ptr addrspace(3) @global_smem, i32 %879, !dbg !68 + %881 = getelementptr inbounds nuw i8, ptr addrspace(3) %880, i32 %125, !dbg !68 + %882 = load bfloat, ptr addrspace(3) %881, align 16, !dbg !68 + %883 = getelementptr inbounds nuw i8, ptr addrspace(3) %881, i32 2, !dbg !68 + %884 = load bfloat, ptr addrspace(3) %883, align 2, !dbg !68 + %885 = getelementptr inbounds nuw i8, ptr addrspace(3) %881, i32 4, !dbg !68 + %886 = load bfloat, ptr addrspace(3) %885, align 4, !dbg !68 + %887 = getelementptr inbounds nuw i8, ptr addrspace(3) %881, i32 6, !dbg !68 + %888 = load bfloat, ptr addrspace(3) %887, align 2, !dbg !68 + %889 = getelementptr inbounds nuw i8, ptr addrspace(3) %881, i32 8, !dbg !68 + %890 = load bfloat, ptr addrspace(3) %889, align 8, !dbg !68 + %891 = getelementptr inbounds nuw i8, ptr addrspace(3) %881, i32 10, !dbg !68 + %892 = load bfloat, ptr addrspace(3) %891, align 2, !dbg !68 + %893 = getelementptr inbounds nuw i8, ptr addrspace(3) %881, i32 12, !dbg !68 + %894 = load bfloat, ptr addrspace(3) %893, align 4, !dbg !68 + %895 = getelementptr inbounds nuw i8, ptr addrspace(3) %881, i32 14, !dbg !68 + %896 = load bfloat, ptr addrspace(3) %895, align 2, !dbg !68 + %897 = getelementptr inbounds nuw i8, ptr addrspace(3) %880, i32 %128, !dbg !68 + %898 = load bfloat, ptr addrspace(3) %897, align 16, !dbg !68 + %899 = getelementptr inbounds nuw i8, ptr addrspace(3) %897, i32 2, !dbg !68 + %900 = load bfloat, ptr addrspace(3) %899, align 2, !dbg !68 + %901 = getelementptr inbounds nuw i8, ptr addrspace(3) %897, i32 4, !dbg !68 + %902 = load bfloat, ptr addrspace(3) %901, align 4, !dbg !68 + %903 = getelementptr inbounds nuw i8, ptr addrspace(3) %897, i32 6, !dbg !68 + %904 = load bfloat, ptr addrspace(3) %903, align 2, !dbg !68 + %905 = getelementptr inbounds nuw i8, ptr addrspace(3) %897, i32 8, !dbg !68 + %906 = load bfloat, ptr addrspace(3) %905, align 8, !dbg !68 + %907 = getelementptr inbounds nuw i8, ptr addrspace(3) %897, i32 10, !dbg !68 + %908 = load bfloat, ptr addrspace(3) %907, align 2, !dbg !68 + %909 = getelementptr inbounds nuw i8, ptr addrspace(3) %897, i32 12, !dbg !68 + %910 = load bfloat, ptr addrspace(3) %909, align 4, !dbg !68 + %911 = getelementptr inbounds nuw i8, ptr addrspace(3) %897, i32 14, !dbg !68 + %912 = load bfloat, ptr addrspace(3) %911, align 2, !dbg !68 + %913 = getelementptr inbounds nuw i8, ptr addrspace(3) %880, i32 %131, !dbg !68 + %914 = load bfloat, ptr addrspace(3) %913, align 16, !dbg !68 + %915 = getelementptr inbounds nuw i8, ptr addrspace(3) %913, i32 2, !dbg !68 + %916 = load bfloat, ptr addrspace(3) %915, align 2, !dbg !68 + %917 = getelementptr inbounds nuw i8, ptr addrspace(3) %913, i32 4, !dbg !68 + %918 = load bfloat, ptr addrspace(3) %917, align 4, !dbg !68 + %919 = getelementptr inbounds nuw i8, ptr addrspace(3) %913, i32 6, !dbg !68 + %920 = load bfloat, ptr addrspace(3) %919, align 2, !dbg !68 + %921 = getelementptr inbounds nuw i8, ptr addrspace(3) %913, i32 8, !dbg !68 + %922 = load bfloat, ptr addrspace(3) %921, align 8, !dbg !68 + %923 = getelementptr inbounds nuw i8, ptr addrspace(3) %913, i32 10, !dbg !68 + %924 = load bfloat, ptr addrspace(3) %923, align 2, !dbg !68 + %925 = getelementptr inbounds nuw i8, ptr addrspace(3) %913, i32 12, !dbg !68 + %926 = load bfloat, ptr addrspace(3) %925, align 4, !dbg !68 + %927 = getelementptr inbounds nuw i8, ptr addrspace(3) %913, i32 14, !dbg !68 + %928 = load bfloat, ptr addrspace(3) %927, align 2, !dbg !68 + %929 = getelementptr inbounds nuw i8, ptr addrspace(3) %880, i32 %134, !dbg !68 + %930 = load bfloat, ptr addrspace(3) %929, align 16, !dbg !68 + %931 = getelementptr inbounds nuw i8, ptr addrspace(3) %929, i32 2, !dbg !68 + %932 = load bfloat, ptr addrspace(3) %931, align 2, !dbg !68 + %933 = getelementptr inbounds nuw i8, ptr addrspace(3) %929, i32 4, !dbg !68 + %934 = load bfloat, ptr addrspace(3) %933, align 4, !dbg !68 + %935 = getelementptr inbounds nuw i8, ptr addrspace(3) %929, i32 6, !dbg !68 + %936 = load bfloat, ptr addrspace(3) %935, align 2, !dbg !68 + %937 = getelementptr inbounds nuw i8, ptr addrspace(3) %929, i32 8, !dbg !68 + %938 = load bfloat, ptr addrspace(3) %937, align 8, !dbg !68 + %939 = getelementptr inbounds nuw i8, ptr addrspace(3) %929, i32 10, !dbg !68 + %940 = load bfloat, ptr addrspace(3) %939, align 2, !dbg !68 + %941 = getelementptr inbounds nuw i8, ptr addrspace(3) %929, i32 12, !dbg !68 + %942 = load bfloat, ptr addrspace(3) %941, align 4, !dbg !68 + %943 = getelementptr inbounds nuw i8, ptr addrspace(3) %929, i32 14, !dbg !68 + %944 = load bfloat, ptr addrspace(3) %943, align 2, !dbg !68 + tail call void @llvm.nvvm.fence.proxy.async.shared_cta(), !dbg !102 + %945 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !102 + %946 = extractvalue { i32, i1 } %945, 1, !dbg !102 + %947 = and i1 %369, %946, !dbg !102 + tail call void asm sideeffect "@$0 mbarrier.init.shared::cta.b64 [$1], 1;", "b,r"(i1 %947, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 147456)) #4, !dbg !102 + %948 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %36, i32 0, i32 31), !dbg !102 + %949 = icmp eq i32 %948, 0, !dbg !102 + br i1 %949, label %950, label %962, !dbg !102 + +950: ; preds = %__nv_exp2f.exit420 + %951 = getelementptr bfloat, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %879, !dbg !71 + %952 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !102 + %953 = extractvalue { i32, i1 } %952, 1, !dbg !102 + %954 = ptrtoint ptr addrspace(3) %951 to i32, !dbg !102 + %955 = lshr exact i32 %954, 4, !dbg !102 + %956 = and i32 %955, 16383, !dbg !102 + %957 = zext nneg i32 %956 to i64, !dbg !102 + %958 = or disjoint i64 %957, 4611756662049472512, !dbg !102 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %15, i64 %958, i64 %373, i32 68224144, i1 false, i1 %953) #4, !dbg !102 + %959 = add nuw nsw i64 %957, 4611756662049472514, !dbg !102 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %15, i64 %959, i64 %374, i32 68224144, i1 true, i1 %953) #4, !dbg !102 + %960 = add nuw nsw i64 %957, 4611756662049472516, !dbg !102 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %15, i64 %960, i64 %375, i32 68224144, i1 true, i1 %953) #4, !dbg !102 + %961 = add nuw nsw i64 %957, 4611756662049472518, !dbg !102 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %15, i64 %961, i64 %376, i32 68224144, i1 true, i1 %953) #4, !dbg !102 + tail call void asm sideeffect "@$0 tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [$1];", "b,r"(i1 %953, i32 ptrtoint (ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 147456) to i32)) #4, !dbg !102 + br label %962, !dbg !102 + +962: ; preds = %950, %__nv_exp2f.exit420 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !102 + tail call void asm sideeffect "\0A{\0A\09.reg .pred complete;\0A\09waitLoop:\0A\09mbarrier.try_wait.parity.shared::cta.b64 complete, [$0], $1;\0A\09@!complete bra.uni waitLoop;\0A}\0A", "r,r"(ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 147456), i32 0) #4, !dbg !102 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !102 + %963 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !102 + %964 = extractvalue { i32, i1 } %963, 1, !dbg !102 + %965 = and i1 %369, %964, !dbg !102 + tail call void asm sideeffect "@$0 mbarrier.inval.shared::cta.b64 [$1];", "b,r"(i1 %965, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 147456)) #4, !dbg !102 + %966 = shl i32 %948, 21, !dbg !102 + %967 = and i32 %966, 6291456, !dbg !102 + %968 = add i32 %967, %15, !dbg !102 + %969 = tail call { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } asm sideeffect "tcgen05.ld.sync.aligned.16x32bx2.x32.b32 {$0, $1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13, $14, $15, $16, $17, $18, $19, $20, $21, $22, $23, $24, $25, $26, $27, $28, $29, $30, $31}, [$32 + 0], 32;", "=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,r"(i32 %968) #4, !dbg !102 + %970 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 0, !dbg !102 + %971 = bitcast i32 %970 to float, !dbg !102 + %972 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 1, !dbg !102 + %973 = bitcast i32 %972 to float, !dbg !102 + %974 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 2, !dbg !102 + %975 = bitcast i32 %974 to float, !dbg !102 + %976 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 3, !dbg !102 + %977 = bitcast i32 %976 to float, !dbg !102 + %978 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 4, !dbg !102 + %979 = bitcast i32 %978 to float, !dbg !102 + %980 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 5, !dbg !102 + %981 = bitcast i32 %980 to float, !dbg !102 + %982 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 6, !dbg !102 + %983 = bitcast i32 %982 to float, !dbg !102 + %984 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 7, !dbg !102 + %985 = bitcast i32 %984 to float, !dbg !102 + %986 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 8, !dbg !102 + %987 = bitcast i32 %986 to float, !dbg !102 + %988 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 9, !dbg !102 + %989 = bitcast i32 %988 to float, !dbg !102 + %990 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 10, !dbg !102 + %991 = bitcast i32 %990 to float, !dbg !102 + %992 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 11, !dbg !102 + %993 = bitcast i32 %992 to float, !dbg !102 + %994 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 12, !dbg !102 + %995 = bitcast i32 %994 to float, !dbg !102 + %996 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 13, !dbg !102 + %997 = bitcast i32 %996 to float, !dbg !102 + %998 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 14, !dbg !102 + %999 = bitcast i32 %998 to float, !dbg !102 + %1000 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 15, !dbg !102 + %1001 = bitcast i32 %1000 to float, !dbg !102 + %1002 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 16, !dbg !102 + %1003 = bitcast i32 %1002 to float, !dbg !102 + %1004 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 17, !dbg !102 + %1005 = bitcast i32 %1004 to float, !dbg !102 + %1006 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 18, !dbg !102 + %1007 = bitcast i32 %1006 to float, !dbg !102 + %1008 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 19, !dbg !102 + %1009 = bitcast i32 %1008 to float, !dbg !102 + %1010 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 20, !dbg !102 + %1011 = bitcast i32 %1010 to float, !dbg !102 + %1012 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 21, !dbg !102 + %1013 = bitcast i32 %1012 to float, !dbg !102 + %1014 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 22, !dbg !102 + %1015 = bitcast i32 %1014 to float, !dbg !102 + %1016 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 23, !dbg !102 + %1017 = bitcast i32 %1016 to float, !dbg !102 + %1018 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 24, !dbg !102 + %1019 = bitcast i32 %1018 to float, !dbg !102 + %1020 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 25, !dbg !102 + %1021 = bitcast i32 %1020 to float, !dbg !102 + %1022 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 26, !dbg !102 + %1023 = bitcast i32 %1022 to float, !dbg !102 + %1024 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 27, !dbg !102 + %1025 = bitcast i32 %1024 to float, !dbg !102 + %1026 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 28, !dbg !102 + %1027 = bitcast i32 %1026 to float, !dbg !102 + %1028 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 29, !dbg !102 + %1029 = bitcast i32 %1028 to float, !dbg !102 + %1030 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 30, !dbg !102 + %1031 = bitcast i32 %1030 to float, !dbg !102 + %1032 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %969, 31, !dbg !102 + %1033 = bitcast i32 %1032 to float, !dbg !102 + tail call void @llvm.nvvm.tcgen05.wait.ld(), !dbg !102 + %1034 = add i32 %967, %87, !dbg !56 + tail call void asm sideeffect "@$0 tcgen05.st.sync.aligned.16x32bx2.x32.b32 [$1 + 0], 32, {$2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13, $14, $15, $16, $17, $18, $19, $20, $21, $22, $23, $24, $25, $26, $27, $28, $29, $30, $31, $32, $33};", "b,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r"(i1 true, i32 %1034, float %971, float %973, float %975, float %977, float %979, float %981, float %983, float %985, float %987, float %989, float %991, float %993, float %995, float %997, float %999, float %1001, float %1003, float %1005, float %1007, float %1009, float %1011, float %1013, float %1015, float %1017, float %1019, float %1021, float %1023, float %1025, float %1027, float %1029, float %1031, float %1033) #4, !dbg !56 + tail call void @llvm.nvvm.tcgen05.wait.st(), !dbg !56 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !56 + tail call void @llvm.nvvm.fence.proxy.async.shared_cta(), !dbg !56 + %1035 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !56 + %1036 = extractvalue { i32, i1 } %1035, 1, !dbg !56 + %1037 = and i1 %369, %1036, !dbg !56 + tail call void asm sideeffect "@$0 mbarrier.init.shared::cta.b64 [$1], 1;", "b,r"(i1 %1037, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072)) #4, !dbg !56 + br i1 %949, label %1038, label %1050, !dbg !56 + +1038: ; preds = %962 + %1039 = getelementptr bfloat, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 32768), i32 %879, !dbg !73 + %1040 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !56 + %1041 = extractvalue { i32, i1 } %1040, 1, !dbg !56 + %1042 = ptrtoint ptr addrspace(3) %1039 to i32, !dbg !56 + %1043 = lshr exact i32 %1042, 4, !dbg !56 + %1044 = and i32 %1043, 16383, !dbg !56 + %1045 = zext nneg i32 %1044 to i64, !dbg !56 + %1046 = or disjoint i64 %1045, 4611756662049472512, !dbg !56 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %87, i64 %1046, i64 %380, i32 68224144, i1 true, i1 %1041) #4, !dbg !56 + %1047 = add nuw nsw i64 %1045, 4611756662049472514, !dbg !56 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %87, i64 %1047, i64 %381, i32 68224144, i1 true, i1 %1041) #4, !dbg !56 + %1048 = add nuw nsw i64 %1045, 4611756662049472516, !dbg !56 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %87, i64 %1048, i64 %382, i32 68224144, i1 true, i1 %1041) #4, !dbg !56 + %1049 = add nuw nsw i64 %1045, 4611756662049472518, !dbg !56 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %87, i64 %1049, i64 %383, i32 68224144, i1 true, i1 %1041) #4, !dbg !56 + tail call void asm sideeffect "@$0 tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [$1];", "b,r"(i1 %1041, i32 ptrtoint (ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072) to i32)) #4, !dbg !56 + br label %1050, !dbg !56 + +1050: ; preds = %1038, %962 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !56 + tail call void asm sideeffect "\0A{\0A\09.reg .pred complete;\0A\09waitLoop:\0A\09mbarrier.try_wait.parity.shared::cta.b64 complete, [$0], $1;\0A\09@!complete bra.uni waitLoop;\0A}\0A", "r,r"(ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 0) #4, !dbg !56 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !56 + %1051 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !56 + %1052 = extractvalue { i32, i1 } %1051, 1, !dbg !56 + %1053 = and i1 %369, %1052, !dbg !56 + tail call void asm sideeffect "@$0 mbarrier.inval.shared::cta.b64 [$1];", "b,r"(i1 %1053, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072)) #4, !dbg !56 + %1054 = fsub float %647, %650, !dbg !103 + %1055 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #4, !dbg !104 + %.not.i421 = icmp eq i32 %1055, 0, !dbg !104 + br i1 %.not.i421, label %1058, label %1056, !dbg !104 + +1056: ; preds = %1050 + %1057 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1054), !dbg !104 + br label %__nv_exp2f.exit423, !dbg !104 + +1058: ; preds = %1050 + %1059 = tail call float @llvm.nvvm.ex2.approx.f32(float %1054), !dbg !104 + br label %__nv_exp2f.exit423, !dbg !104 + +__nv_exp2f.exit423: ; preds = %1056, %1058 + %.0.i422 = phi float [ %1057, %1056 ], [ %1059, %1058 ], !dbg !104 + %1060 = select i1 %.pn96.pn471, float %.0.i422, float 0.000000e+00, !dbg !106 + %gep483 = getelementptr bfloat, ptr addrspace(3) %invariant.gep482, i32 %879, !dbg !74 + %1061 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep483, i32 4, !dbg !74 + %1062 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep483, i32 8, !dbg !74 + %1063 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep483, i32 12, !dbg !74 + %1064 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep483, i32 16, !dbg !74 + %1065 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep483, i32 20, !dbg !74 + %1066 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep483, i32 24, !dbg !74 + %1067 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep483, i32 28, !dbg !74 + %1068 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep483, i32 32, !dbg !74 + %1069 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep483, i32 36, !dbg !74 + %1070 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep483, i32 40, !dbg !74 + %1071 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep483, i32 44, !dbg !74 + %1072 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep483, i32 48, !dbg !74 + %1073 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep483, i32 52, !dbg !74 + %1074 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep483, i32 56, !dbg !74 + %1075 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep483, i32 60, !dbg !74 + %1076 = load <2 x bfloat>, ptr addrspace(3) %1063, align 4, !dbg !74 + %1077 = fpext <2 x bfloat> %1076 to <2 x float>, !dbg !107 + %1078 = insertelement <2 x float> poison, float %1060, i64 0, !dbg !108 + %1079 = shufflevector <2 x float> %1078, <2 x float> poison, <2 x i32> zeroinitializer, !dbg !108 + %1080 = load <2 x bfloat>, ptr addrspace(3) %1062, align 8, !dbg !74 + %1081 = fpext <2 x bfloat> %1080 to <2 x float>, !dbg !107 + %1082 = load <2 x bfloat>, ptr addrspace(3) %1061, align 4, !dbg !74 + %1083 = fpext <2 x bfloat> %1082 to <2 x float>, !dbg !107 + %1084 = load <2 x bfloat>, ptr addrspace(3) %gep483, align 16, !dbg !74 + %1085 = fpext <2 x bfloat> %1084 to <2 x float>, !dbg !107 + %1086 = load <2 x bfloat>, ptr addrspace(3) %1067, align 4, !dbg !74 + %1087 = fpext <2 x bfloat> %1086 to <2 x float>, !dbg !107 + %1088 = load <2 x bfloat>, ptr addrspace(3) %1066, align 8, !dbg !74 + %1089 = fpext <2 x bfloat> %1088 to <2 x float>, !dbg !107 + %1090 = load <2 x bfloat>, ptr addrspace(3) %1065, align 4, !dbg !74 + %1091 = fpext <2 x bfloat> %1090 to <2 x float>, !dbg !107 + %1092 = load <2 x bfloat>, ptr addrspace(3) %1064, align 16, !dbg !74 + %1093 = fpext <2 x bfloat> %1092 to <2 x float>, !dbg !107 + %1094 = load <2 x bfloat>, ptr addrspace(3) %1071, align 4, !dbg !74 + %1095 = fpext <2 x bfloat> %1094 to <2 x float>, !dbg !107 + %1096 = load <2 x bfloat>, ptr addrspace(3) %1070, align 8, !dbg !74 + %1097 = fpext <2 x bfloat> %1096 to <2 x float>, !dbg !107 + %1098 = load <2 x bfloat>, ptr addrspace(3) %1069, align 4, !dbg !74 + %1099 = fpext <2 x bfloat> %1098 to <2 x float>, !dbg !107 + %1100 = load <2 x bfloat>, ptr addrspace(3) %1068, align 16, !dbg !74 + %1101 = fpext <2 x bfloat> %1100 to <2 x float>, !dbg !107 + %1102 = load <2 x bfloat>, ptr addrspace(3) %1075, align 4, !dbg !74 + %1103 = fpext <2 x bfloat> %1102 to <2 x float>, !dbg !107 + %1104 = load <2 x bfloat>, ptr addrspace(3) %1074, align 8, !dbg !74 + %1105 = fpext <2 x bfloat> %1104 to <2 x float>, !dbg !107 + %1106 = load <2 x bfloat>, ptr addrspace(3) %1073, align 4, !dbg !74 + %1107 = fpext <2 x bfloat> %1106 to <2 x float>, !dbg !107 + %1108 = load <2 x bfloat>, ptr addrspace(3) %1072, align 16, !dbg !74 + %1109 = fpext <2 x bfloat> %1108 to <2 x float>, !dbg !107 + %1110 = tail call { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } asm sideeffect "tcgen05.ld.sync.aligned.16x32bx2.x32.b32 {$0, $1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13, $14, $15, $16, $17, $18, $19, $20, $21, $22, $23, $24, $25, $26, $27, $28, $29, $30, $31}, [$32 + 0], 32;", "=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,r"(i32 %1034) #4, !dbg !56 + %1111 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 0, !dbg !56 + %1112 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 1, !dbg !56 + %1113 = insertelement <2 x i32> poison, i32 %1111, i64 0, !dbg !56 + %1114 = insertelement <2 x i32> %1113, i32 %1112, i64 1, !dbg !56 + %1115 = bitcast <2 x i32> %1114 to <2 x float>, !dbg !56 + %1116 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 2, !dbg !56 + %1117 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 3, !dbg !56 + %1118 = insertelement <2 x i32> poison, i32 %1116, i64 0, !dbg !56 + %1119 = insertelement <2 x i32> %1118, i32 %1117, i64 1, !dbg !56 + %1120 = bitcast <2 x i32> %1119 to <2 x float>, !dbg !56 + %1121 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 4, !dbg !56 + %1122 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 5, !dbg !56 + %1123 = insertelement <2 x i32> poison, i32 %1121, i64 0, !dbg !56 + %1124 = insertelement <2 x i32> %1123, i32 %1122, i64 1, !dbg !56 + %1125 = bitcast <2 x i32> %1124 to <2 x float>, !dbg !56 + %1126 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 6, !dbg !56 + %1127 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 7, !dbg !56 + %1128 = insertelement <2 x i32> poison, i32 %1126, i64 0, !dbg !56 + %1129 = insertelement <2 x i32> %1128, i32 %1127, i64 1, !dbg !56 + %1130 = bitcast <2 x i32> %1129 to <2 x float>, !dbg !56 + %1131 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 8, !dbg !56 + %1132 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 9, !dbg !56 + %1133 = insertelement <2 x i32> poison, i32 %1131, i64 0, !dbg !56 + %1134 = insertelement <2 x i32> %1133, i32 %1132, i64 1, !dbg !56 + %1135 = bitcast <2 x i32> %1134 to <2 x float>, !dbg !56 + %1136 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 10, !dbg !56 + %1137 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 11, !dbg !56 + %1138 = insertelement <2 x i32> poison, i32 %1136, i64 0, !dbg !56 + %1139 = insertelement <2 x i32> %1138, i32 %1137, i64 1, !dbg !56 + %1140 = bitcast <2 x i32> %1139 to <2 x float>, !dbg !56 + %1141 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 12, !dbg !56 + %1142 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 13, !dbg !56 + %1143 = insertelement <2 x i32> poison, i32 %1141, i64 0, !dbg !56 + %1144 = insertelement <2 x i32> %1143, i32 %1142, i64 1, !dbg !56 + %1145 = bitcast <2 x i32> %1144 to <2 x float>, !dbg !56 + %1146 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 14, !dbg !56 + %1147 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 15, !dbg !56 + %1148 = insertelement <2 x i32> poison, i32 %1146, i64 0, !dbg !56 + %1149 = insertelement <2 x i32> %1148, i32 %1147, i64 1, !dbg !56 + %1150 = bitcast <2 x i32> %1149 to <2 x float>, !dbg !56 + %1151 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 16, !dbg !56 + %1152 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 17, !dbg !56 + %1153 = insertelement <2 x i32> poison, i32 %1151, i64 0, !dbg !56 + %1154 = insertelement <2 x i32> %1153, i32 %1152, i64 1, !dbg !56 + %1155 = bitcast <2 x i32> %1154 to <2 x float>, !dbg !56 + %1156 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 18, !dbg !56 + %1157 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 19, !dbg !56 + %1158 = insertelement <2 x i32> poison, i32 %1156, i64 0, !dbg !56 + %1159 = insertelement <2 x i32> %1158, i32 %1157, i64 1, !dbg !56 + %1160 = bitcast <2 x i32> %1159 to <2 x float>, !dbg !56 + %1161 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 20, !dbg !56 + %1162 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 21, !dbg !56 + %1163 = insertelement <2 x i32> poison, i32 %1161, i64 0, !dbg !56 + %1164 = insertelement <2 x i32> %1163, i32 %1162, i64 1, !dbg !56 + %1165 = bitcast <2 x i32> %1164 to <2 x float>, !dbg !56 + %1166 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 22, !dbg !56 + %1167 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 23, !dbg !56 + %1168 = insertelement <2 x i32> poison, i32 %1166, i64 0, !dbg !56 + %1169 = insertelement <2 x i32> %1168, i32 %1167, i64 1, !dbg !56 + %1170 = bitcast <2 x i32> %1169 to <2 x float>, !dbg !56 + %1171 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 24, !dbg !56 + %1172 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 25, !dbg !56 + %1173 = insertelement <2 x i32> poison, i32 %1171, i64 0, !dbg !56 + %1174 = insertelement <2 x i32> %1173, i32 %1172, i64 1, !dbg !56 + %1175 = bitcast <2 x i32> %1174 to <2 x float>, !dbg !56 + %1176 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 26, !dbg !56 + %1177 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 27, !dbg !56 + %1178 = insertelement <2 x i32> poison, i32 %1176, i64 0, !dbg !56 + %1179 = insertelement <2 x i32> %1178, i32 %1177, i64 1, !dbg !56 + %1180 = bitcast <2 x i32> %1179 to <2 x float>, !dbg !56 + %1181 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 28, !dbg !56 + %1182 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 29, !dbg !56 + %1183 = insertelement <2 x i32> poison, i32 %1181, i64 0, !dbg !56 + %1184 = insertelement <2 x i32> %1183, i32 %1182, i64 1, !dbg !56 + %1185 = bitcast <2 x i32> %1184 to <2 x float>, !dbg !56 + %1186 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 30, !dbg !56 + %1187 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1110, 31, !dbg !56 + %1188 = insertelement <2 x i32> poison, i32 %1186, i64 0, !dbg !56 + %1189 = insertelement <2 x i32> %1188, i32 %1187, i64 1, !dbg !56 + %1190 = bitcast <2 x i32> %1189 to <2 x float>, !dbg !56 + tail call void @llvm.nvvm.tcgen05.wait.ld(), !dbg !56 + %1191 = fmul <2 x float> %1079, %1115, !dbg !108 + %1192 = fmul <2 x float> %1079, %1120, !dbg !108 + %1193 = fmul <2 x float> %1079, %1125, !dbg !108 + %1194 = fmul <2 x float> %1079, %1130, !dbg !108 + %1195 = fmul <2 x float> %1079, %1135, !dbg !108 + %1196 = fmul <2 x float> %1079, %1140, !dbg !108 + %1197 = fmul <2 x float> %1079, %1145, !dbg !108 + %1198 = fmul <2 x float> %1079, %1150, !dbg !108 + %1199 = fmul <2 x float> %1079, %1155, !dbg !108 + %1200 = fmul <2 x float> %1079, %1160, !dbg !108 + %1201 = fmul <2 x float> %1079, %1165, !dbg !108 + %1202 = fmul <2 x float> %1079, %1170, !dbg !108 + %1203 = fmul <2 x float> %1079, %1175, !dbg !108 + %1204 = fmul <2 x float> %1079, %1180, !dbg !108 + %1205 = fmul <2 x float> %1079, %1185, !dbg !108 + %1206 = fmul <2 x float> %1079, %1190, !dbg !108 + %1207 = fadd <2 x float> %1191, %1085, !dbg !107 + %1208 = fadd <2 x float> %1192, %1083, !dbg !107 + %1209 = fadd <2 x float> %1193, %1081, !dbg !107 + %1210 = fadd <2 x float> %1194, %1077, !dbg !107 + %1211 = fadd <2 x float> %1195, %1093, !dbg !107 + %1212 = fadd <2 x float> %1196, %1091, !dbg !107 + %1213 = fadd <2 x float> %1197, %1089, !dbg !107 + %1214 = fadd <2 x float> %1198, %1087, !dbg !107 + %1215 = fadd <2 x float> %1199, %1101, !dbg !107 + %1216 = fadd <2 x float> %1200, %1099, !dbg !107 + %1217 = fadd <2 x float> %1201, %1097, !dbg !107 + %1218 = fadd <2 x float> %1202, %1095, !dbg !107 + %1219 = fadd <2 x float> %1203, %1109, !dbg !107 + %1220 = fadd <2 x float> %1204, %1107, !dbg !107 + %1221 = fadd <2 x float> %1205, %1105, !dbg !107 + %1222 = fadd <2 x float> %1206, %1103, !dbg !107 + %1223 = fptrunc <2 x float> %1207 to <2 x bfloat>, !dbg !109 + %1224 = fptrunc <2 x float> %1208 to <2 x bfloat>, !dbg !109 + %1225 = fptrunc <2 x float> %1209 to <2 x bfloat>, !dbg !109 + %1226 = fptrunc <2 x float> %1210 to <2 x bfloat>, !dbg !109 + %1227 = fptrunc <2 x float> %1211 to <2 x bfloat>, !dbg !109 + %1228 = fptrunc <2 x float> %1212 to <2 x bfloat>, !dbg !109 + %1229 = fptrunc <2 x float> %1213 to <2 x bfloat>, !dbg !109 + %1230 = fptrunc <2 x float> %1214 to <2 x bfloat>, !dbg !109 + %1231 = fptrunc <2 x float> %1215 to <2 x bfloat>, !dbg !109 + %1232 = fptrunc <2 x float> %1216 to <2 x bfloat>, !dbg !109 + %1233 = fptrunc <2 x float> %1217 to <2 x bfloat>, !dbg !109 + %1234 = fptrunc <2 x float> %1218 to <2 x bfloat>, !dbg !109 + %1235 = fptrunc <2 x float> %1219 to <2 x bfloat>, !dbg !109 + %1236 = fptrunc <2 x float> %1220 to <2 x bfloat>, !dbg !109 + %1237 = fptrunc <2 x float> %1221 to <2 x bfloat>, !dbg !109 + %1238 = fptrunc <2 x float> %1222 to <2 x bfloat>, !dbg !109 + %1239 = bitcast <2 x bfloat> %1223 to i32, !dbg !109 + %1240 = bitcast <2 x bfloat> %1224 to i32, !dbg !109 + %1241 = bitcast <2 x bfloat> %1225 to i32, !dbg !109 + %1242 = bitcast <2 x bfloat> %1226 to i32, !dbg !109 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %384, i32 %1239, i32 %1240, i32 %1241, i32 %1242) #4, !dbg !109 + %1243 = bitcast <2 x bfloat> %1227 to i32, !dbg !109 + %1244 = bitcast <2 x bfloat> %1228 to i32, !dbg !109 + %1245 = bitcast <2 x bfloat> %1229 to i32, !dbg !109 + %1246 = bitcast <2 x bfloat> %1230 to i32, !dbg !109 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %385, i32 %1243, i32 %1244, i32 %1245, i32 %1246) #4, !dbg !109 + %1247 = bitcast <2 x bfloat> %1231 to i32, !dbg !109 + %1248 = bitcast <2 x bfloat> %1232 to i32, !dbg !109 + %1249 = bitcast <2 x bfloat> %1233 to i32, !dbg !109 + %1250 = bitcast <2 x bfloat> %1234 to i32, !dbg !109 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %386, i32 %1247, i32 %1248, i32 %1249, i32 %1250) #4, !dbg !109 + %1251 = bitcast <2 x bfloat> %1235 to i32, !dbg !109 + %1252 = bitcast <2 x bfloat> %1236 to i32, !dbg !109 + %1253 = bitcast <2 x bfloat> %1237 to i32, !dbg !109 + %1254 = bitcast <2 x bfloat> %1238 to i32, !dbg !109 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %387, i32 %1251, i32 %1252, i32 %1253, i32 %1254) #4, !dbg !109 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !109 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %388, i32 %1239, i32 %1243, i32 %1247, i32 %1251) #4, !dbg !109 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !109 + %1255 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %389), !dbg !109 + %1256 = extractvalue { i32, i32, i32, i32 } %1255, 0, !dbg !109 + %1257 = extractvalue { i32, i32, i32, i32 } %1255, 1, !dbg !109 + %1258 = extractvalue { i32, i32, i32, i32 } %1255, 2, !dbg !109 + %1259 = extractvalue { i32, i32, i32, i32 } %1255, 3, !dbg !109 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !109 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %388, i32 %1240, i32 %1244, i32 %1248, i32 %1252) #4, !dbg !109 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !109 + %1260 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %389), !dbg !109 + %1261 = extractvalue { i32, i32, i32, i32 } %1260, 0, !dbg !109 + %1262 = extractvalue { i32, i32, i32, i32 } %1260, 1, !dbg !109 + %1263 = extractvalue { i32, i32, i32, i32 } %1260, 2, !dbg !109 + %1264 = extractvalue { i32, i32, i32, i32 } %1260, 3, !dbg !109 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !109 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %388, i32 %1241, i32 %1245, i32 %1249, i32 %1253) #4, !dbg !109 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !109 + %1265 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %389), !dbg !109 + %1266 = extractvalue { i32, i32, i32, i32 } %1265, 0, !dbg !109 + %1267 = extractvalue { i32, i32, i32, i32 } %1265, 1, !dbg !109 + %1268 = extractvalue { i32, i32, i32, i32 } %1265, 2, !dbg !109 + %1269 = extractvalue { i32, i32, i32, i32 } %1265, 3, !dbg !109 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !109 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %388, i32 %1242, i32 %1246, i32 %1250, i32 %1254) #4, !dbg !109 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !109 + %1270 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %389), !dbg !109 + %1271 = extractvalue { i32, i32, i32, i32 } %1270, 0, !dbg !109 + %1272 = extractvalue { i32, i32, i32, i32 } %1270, 1, !dbg !109 + %1273 = extractvalue { i32, i32, i32, i32 } %1270, 2, !dbg !109 + %1274 = extractvalue { i32, i32, i32, i32 } %1270, 3, !dbg !109 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1256, i32 %1261, i32 %1266, i32 %1271, ptr addrspace(1) %gep, i1 %.pn148.pn463) #4, !dbg !110 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1257, i32 %1262, i32 %1267, i32 %1272, ptr addrspace(1) %gep425, i1 %.pn132.pn464) #4, !dbg !110 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1258, i32 %1263, i32 %1268, i32 %1273, ptr addrspace(1) %gep427, i1 %.pn116.pn465) #4, !dbg !110 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1259, i32 %1264, i32 %1269, i32 %1274, ptr addrspace(1) %gep429, i1 %.pn100.pn466) #4, !dbg !110 + %1275 = getelementptr bfloat, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 65536), i32 %879, !dbg !77 + %1276 = getelementptr bfloat, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 81920), i32 %879, !dbg !78 + %1277 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %393, !dbg !78 + %1278 = load bfloat, ptr addrspace(3) %1277, align 2, !dbg !78 + %1279 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %394, !dbg !78 + %1280 = load bfloat, ptr addrspace(3) %1279, align 2, !dbg !78 + %1281 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %395, !dbg !78 + %1282 = load bfloat, ptr addrspace(3) %1281, align 2, !dbg !78 + %1283 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %396, !dbg !78 + %1284 = load bfloat, ptr addrspace(3) %1283, align 2, !dbg !78 + %1285 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %397, !dbg !78 + %1286 = load bfloat, ptr addrspace(3) %1285, align 2, !dbg !78 + %1287 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %398, !dbg !78 + %1288 = load bfloat, ptr addrspace(3) %1287, align 2, !dbg !78 + %1289 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %399, !dbg !78 + %1290 = load bfloat, ptr addrspace(3) %1289, align 2, !dbg !78 + %1291 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %400, !dbg !78 + %1292 = load bfloat, ptr addrspace(3) %1291, align 2, !dbg !78 + %1293 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %401, !dbg !78 + %1294 = load bfloat, ptr addrspace(3) %1293, align 2, !dbg !78 + %1295 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %402, !dbg !78 + %1296 = load bfloat, ptr addrspace(3) %1295, align 2, !dbg !78 + %1297 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %403, !dbg !78 + %1298 = load bfloat, ptr addrspace(3) %1297, align 2, !dbg !78 + %1299 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %404, !dbg !78 + %1300 = load bfloat, ptr addrspace(3) %1299, align 2, !dbg !78 + %1301 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %405, !dbg !78 + %1302 = load bfloat, ptr addrspace(3) %1301, align 2, !dbg !78 + %1303 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %406, !dbg !78 + %1304 = load bfloat, ptr addrspace(3) %1303, align 2, !dbg !78 + %1305 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %407, !dbg !78 + %1306 = load bfloat, ptr addrspace(3) %1305, align 2, !dbg !78 + %1307 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %408, !dbg !78 + %1308 = load bfloat, ptr addrspace(3) %1307, align 2, !dbg !78 + %1309 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %409, !dbg !78 + %1310 = load bfloat, ptr addrspace(3) %1309, align 2, !dbg !78 + %1311 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %410, !dbg !78 + %1312 = load bfloat, ptr addrspace(3) %1311, align 2, !dbg !78 + %1313 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %411, !dbg !78 + %1314 = load bfloat, ptr addrspace(3) %1313, align 2, !dbg !78 + %1315 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %412, !dbg !78 + %1316 = load bfloat, ptr addrspace(3) %1315, align 2, !dbg !78 + %1317 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %413, !dbg !78 + %1318 = load bfloat, ptr addrspace(3) %1317, align 2, !dbg !78 + %1319 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %414, !dbg !78 + %1320 = load bfloat, ptr addrspace(3) %1319, align 2, !dbg !78 + %1321 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %415, !dbg !78 + %1322 = load bfloat, ptr addrspace(3) %1321, align 2, !dbg !78 + %1323 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %416, !dbg !78 + %1324 = load bfloat, ptr addrspace(3) %1323, align 2, !dbg !78 + %1325 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %417, !dbg !78 + %1326 = load bfloat, ptr addrspace(3) %1325, align 2, !dbg !78 + %1327 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %418, !dbg !78 + %1328 = load bfloat, ptr addrspace(3) %1327, align 2, !dbg !78 + %1329 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %419, !dbg !78 + %1330 = load bfloat, ptr addrspace(3) %1329, align 2, !dbg !78 + %1331 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %420, !dbg !78 + %1332 = load bfloat, ptr addrspace(3) %1331, align 2, !dbg !78 + %1333 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %421, !dbg !78 + %1334 = load bfloat, ptr addrspace(3) %1333, align 2, !dbg !78 + %1335 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %422, !dbg !78 + %1336 = load bfloat, ptr addrspace(3) %1335, align 2, !dbg !78 + %1337 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %423, !dbg !78 + %1338 = load bfloat, ptr addrspace(3) %1337, align 2, !dbg !78 + %1339 = getelementptr inbounds nuw i8, ptr addrspace(3) %1276, i32 %424, !dbg !78 + %1340 = load bfloat, ptr addrspace(3) %1339, align 2, !dbg !78 + %1341 = shufflevector <64 x float> %488, <64 x float> poison, <32 x i32> , !dbg !111 + %1342 = insertelement <32 x float> poison, float %.0.i, i64 0, !dbg !111 + %1343 = shufflevector <32 x float> %1342, <32 x float> poison, <32 x i32> zeroinitializer, !dbg !111 + %1344 = fmul <32 x float> %1341, %1343, !dbg !111 + %1345 = fpext bfloat %1278 to float, !dbg !112 + %1346 = fpext bfloat %1280 to float, !dbg !112 + %1347 = fpext bfloat %1282 to float, !dbg !112 + %1348 = fpext bfloat %1284 to float, !dbg !112 + %1349 = fpext bfloat %1286 to float, !dbg !112 + %1350 = fpext bfloat %1288 to float, !dbg !112 + %1351 = fpext bfloat %1290 to float, !dbg !112 + %1352 = fpext bfloat %1292 to float, !dbg !112 + %1353 = fpext bfloat %1294 to float, !dbg !112 + %1354 = fpext bfloat %1296 to float, !dbg !112 + %1355 = fpext bfloat %1298 to float, !dbg !112 + %1356 = fpext bfloat %1300 to float, !dbg !112 + %1357 = fpext bfloat %1302 to float, !dbg !112 + %1358 = fpext bfloat %1304 to float, !dbg !112 + %1359 = fpext bfloat %1306 to float, !dbg !112 + %1360 = fpext bfloat %1308 to float, !dbg !112 + %1361 = fpext bfloat %1310 to float, !dbg !112 + %1362 = fpext bfloat %1312 to float, !dbg !112 + %1363 = fpext bfloat %1314 to float, !dbg !112 + %1364 = fpext bfloat %1316 to float, !dbg !112 + %1365 = fpext bfloat %1318 to float, !dbg !112 + %1366 = fpext bfloat %1320 to float, !dbg !112 + %1367 = fpext bfloat %1322 to float, !dbg !112 + %1368 = fpext bfloat %1324 to float, !dbg !112 + %1369 = fpext bfloat %1326 to float, !dbg !112 + %1370 = fpext bfloat %1328 to float, !dbg !112 + %1371 = fpext bfloat %1330 to float, !dbg !112 + %1372 = fpext bfloat %1332 to float, !dbg !112 + %1373 = fpext bfloat %1334 to float, !dbg !112 + %1374 = fpext bfloat %1336 to float, !dbg !112 + %1375 = fpext bfloat %1338 to float, !dbg !112 + %1376 = fpext bfloat %1340 to float, !dbg !112 + %1377 = fmul float %.0.i326, %1345, !dbg !112 + %1378 = fmul float %.0.i329, %1346, !dbg !112 + %1379 = fmul float %.0.i332, %1347, !dbg !112 + %1380 = fmul float %.0.i335, %1348, !dbg !112 + %1381 = fmul float %.0.i338, %1349, !dbg !112 + %1382 = fmul float %.0.i341, %1350, !dbg !112 + %1383 = fmul float %.0.i344, %1351, !dbg !112 + %1384 = fmul float %.0.i347, %1352, !dbg !112 + %1385 = fmul float %.0.i350, %1353, !dbg !112 + %1386 = fmul float %.0.i353, %1354, !dbg !112 + %1387 = fmul float %.0.i356, %1355, !dbg !112 + %1388 = fmul float %.0.i359, %1356, !dbg !112 + %1389 = fmul float %.0.i362, %1357, !dbg !112 + %1390 = fmul float %.0.i365, %1358, !dbg !112 + %1391 = fmul float %.0.i368, %1359, !dbg !112 + %1392 = fmul float %.0.i371, %1360, !dbg !112 + %1393 = fmul float %.0.i374, %1361, !dbg !112 + %1394 = fmul float %.0.i377, %1362, !dbg !112 + %1395 = fmul float %.0.i380, %1363, !dbg !112 + %1396 = fmul float %.0.i383, %1364, !dbg !112 + %1397 = fmul float %.0.i386, %1365, !dbg !112 + %1398 = fmul float %.0.i389, %1366, !dbg !112 + %1399 = fmul float %.0.i392, %1367, !dbg !112 + %1400 = fmul float %.0.i395, %1368, !dbg !112 + %1401 = fmul float %.0.i398, %1369, !dbg !112 + %1402 = fmul float %.0.i401, %1370, !dbg !112 + %1403 = fmul float %.0.i404, %1371, !dbg !112 + %1404 = fmul float %.0.i407, %1372, !dbg !112 + %1405 = fmul float %.0.i410, %1373, !dbg !112 + %1406 = fmul float %.0.i413, %1374, !dbg !112 + %1407 = fmul float %.0.i416, %1375, !dbg !112 + %1408 = fmul float %.0.i419, %1376, !dbg !112 + %1409 = fpext bfloat %882 to float, !dbg !113 + %1410 = fpext bfloat %884 to float, !dbg !113 + %1411 = fpext bfloat %886 to float, !dbg !113 + %1412 = fpext bfloat %888 to float, !dbg !113 + %1413 = fpext bfloat %890 to float, !dbg !113 + %1414 = fpext bfloat %892 to float, !dbg !113 + %1415 = fpext bfloat %894 to float, !dbg !113 + %1416 = fpext bfloat %896 to float, !dbg !113 + %1417 = fpext bfloat %898 to float, !dbg !113 + %1418 = fpext bfloat %900 to float, !dbg !113 + %1419 = fpext bfloat %902 to float, !dbg !113 + %1420 = fpext bfloat %904 to float, !dbg !113 + %1421 = fpext bfloat %906 to float, !dbg !113 + %1422 = fpext bfloat %908 to float, !dbg !113 + %1423 = fpext bfloat %910 to float, !dbg !113 + %1424 = fpext bfloat %912 to float, !dbg !113 + %1425 = fpext bfloat %914 to float, !dbg !113 + %1426 = fpext bfloat %916 to float, !dbg !113 + %1427 = fpext bfloat %918 to float, !dbg !113 + %1428 = fpext bfloat %920 to float, !dbg !113 + %1429 = fpext bfloat %922 to float, !dbg !113 + %1430 = fpext bfloat %924 to float, !dbg !113 + %1431 = fpext bfloat %926 to float, !dbg !113 + %1432 = fpext bfloat %928 to float, !dbg !113 + %1433 = fpext bfloat %930 to float, !dbg !113 + %1434 = fpext bfloat %932 to float, !dbg !113 + %1435 = fpext bfloat %934 to float, !dbg !113 + %1436 = fpext bfloat %936 to float, !dbg !113 + %1437 = fpext bfloat %938 to float, !dbg !113 + %1438 = fpext bfloat %940 to float, !dbg !113 + %1439 = fpext bfloat %942 to float, !dbg !113 + %1440 = fpext bfloat %944 to float, !dbg !113 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !113 + %1441 = bitcast float %1409 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) %428, <1 x i32> %1441) #4, !dbg !113 + %1442 = bitcast float %1413 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %429, <1 x i32> %1442) #4, !dbg !113 + %1443 = bitcast float %1417 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %430, <1 x i32> %1443) #4, !dbg !113 + %1444 = bitcast float %1421 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %431, <1 x i32> %1444) #4, !dbg !113 + %1445 = bitcast float %1425 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %432, <1 x i32> %1445) #4, !dbg !113 + %1446 = bitcast float %1429 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %433, <1 x i32> %1446) #4, !dbg !113 + %1447 = bitcast float %1433 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %434, <1 x i32> %1447) #4, !dbg !113 + %1448 = bitcast float %1437 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %435, <1 x i32> %1448) #4, !dbg !113 + %1449 = bitcast float %1410 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %437, <1 x i32> %1449) #4, !dbg !113 + %1450 = bitcast float %1414 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %438, <1 x i32> %1450) #4, !dbg !113 + %1451 = bitcast float %1418 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %439, <1 x i32> %1451) #4, !dbg !113 + %1452 = bitcast float %1422 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %440, <1 x i32> %1452) #4, !dbg !113 + %1453 = bitcast float %1426 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %441, <1 x i32> %1453) #4, !dbg !113 + %1454 = bitcast float %1430 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %442, <1 x i32> %1454) #4, !dbg !113 + %1455 = bitcast float %1434 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %443, <1 x i32> %1455) #4, !dbg !113 + %1456 = bitcast float %1438 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %444, <1 x i32> %1456) #4, !dbg !113 + %1457 = bitcast float %1411 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %446, <1 x i32> %1457) #4, !dbg !113 + %1458 = bitcast float %1415 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %447, <1 x i32> %1458) #4, !dbg !113 + %1459 = bitcast float %1419 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %448, <1 x i32> %1459) #4, !dbg !113 + %1460 = bitcast float %1423 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %449, <1 x i32> %1460) #4, !dbg !113 + %1461 = bitcast float %1427 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %450, <1 x i32> %1461) #4, !dbg !113 + %1462 = bitcast float %1431 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %451, <1 x i32> %1462) #4, !dbg !113 + %1463 = bitcast float %1435 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %452, <1 x i32> %1463) #4, !dbg !113 + %1464 = bitcast float %1439 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %453, <1 x i32> %1464) #4, !dbg !113 + %1465 = bitcast float %1412 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %455, <1 x i32> %1465) #4, !dbg !113 + %1466 = bitcast float %1416 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %456, <1 x i32> %1466) #4, !dbg !113 + %1467 = bitcast float %1420 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %457, <1 x i32> %1467) #4, !dbg !113 + %1468 = bitcast float %1424 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %458, <1 x i32> %1468) #4, !dbg !113 + %1469 = bitcast float %1428 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %459, <1 x i32> %1469) #4, !dbg !113 + %1470 = bitcast float %1432 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %460, <1 x i32> %1470) #4, !dbg !113 + %1471 = bitcast float %1436 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %461, <1 x i32> %1471) #4, !dbg !113 + %1472 = bitcast float %1440 to <1 x i32>, !dbg !113 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %462, <1 x i32> %1472) #4, !dbg !113 + %1473 = add i32 %967, %463, !dbg !57 + tail call void asm sideeffect "@$0 tcgen05.st.sync.aligned.16x32bx2.x32.b32 [$1 + 0], 32, {$2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13, $14, $15, $16, $17, $18, $19, $20, $21, $22, $23, $24, $25, $26, $27, $28, $29, $30, $31, $32, $33};", "b,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r"(i1 true, i32 %1473, float %1377, float %1378, float %1379, float %1380, float %1381, float %1382, float %1383, float %1384, float %1385, float %1386, float %1387, float %1388, float %1389, float %1390, float %1391, float %1392, float %1393, float %1394, float %1395, float %1396, float %1397, float %1398, float %1399, float %1400, float %1401, float %1402, float %1403, float %1404, float %1405, float %1406, float %1407, float %1408) #4, !dbg !57 + tail call void @llvm.nvvm.tcgen05.wait.st(), !dbg !57 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !57 + tail call void @llvm.nvvm.fence.proxy.async.shared_cta(), !dbg !57 + %1474 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !57 + %1475 = extractvalue { i32, i1 } %1474, 1, !dbg !57 + %1476 = and i1 %369, %1475, !dbg !57 + tail call void asm sideeffect "@$0 mbarrier.init.shared::cta.b64 [$1], 1;", "b,r"(i1 %1476, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 156688)) #4, !dbg !57 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !57 + br i1 %949, label %1477, label %1480, !dbg !57 + +1477: ; preds = %__nv_exp2f.exit423 + %1478 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !57 + %1479 = extractvalue { i32, i1 } %1478, 1, !dbg !57 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::tf32 [ $0 + 0 ], [ $1 + 0 ], $2, $3, $4;", "r,r,l,r,b,b"(i32 %88, i32 %463, i64 %373, i32 68159760, i1 false, i1 %1479) #4, !dbg !57 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::tf32 [ $0 + 0 ], [ $1 + 8 ], $2, $3, $4;", "r,r,l,r,b,b"(i32 %88, i32 %463, i64 %464, i32 68159760, i1 true, i1 %1479) #4, !dbg !57 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::tf32 [ $0 + 0 ], [ $1 + 16 ], $2, $3, $4;", "r,r,l,r,b,b"(i32 %88, i32 %463, i64 %465, i32 68159760, i1 true, i1 %1479) #4, !dbg !57 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::tf32 [ $0 + 0 ], [ $1 + 24 ], $2, $3, $4;", "r,r,l,r,b,b"(i32 %88, i32 %463, i64 %466, i32 68159760, i1 true, i1 %1479) #4, !dbg !57 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::tf32 [ $0 + 0 ], [ $1 + 32 ], $2, $3, $4;", "r,r,l,r,b,b"(i32 %88, i32 %463, i64 %467, i32 68159760, i1 true, i1 %1479) #4, !dbg !57 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::tf32 [ $0 + 0 ], [ $1 + 40 ], $2, $3, $4;", "r,r,l,r,b,b"(i32 %88, i32 %463, i64 %468, i32 68159760, i1 true, i1 %1479) #4, !dbg !57 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::tf32 [ $0 + 0 ], [ $1 + 48 ], $2, $3, $4;", "r,r,l,r,b,b"(i32 %88, i32 %463, i64 %469, i32 68159760, i1 true, i1 %1479) #4, !dbg !57 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::tf32 [ $0 + 0 ], [ $1 + 56 ], $2, $3, $4;", "r,r,l,r,b,b"(i32 %88, i32 %463, i64 %470, i32 68159760, i1 true, i1 %1479) #4, !dbg !57 + tail call void asm sideeffect "@$0 tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [$1];", "b,r"(i1 %1479, i32 ptrtoint (ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 156688) to i32)) #4, !dbg !57 + br label %1480, !dbg !57 + +1480: ; preds = %1477, %__nv_exp2f.exit423 + tail call void asm sideeffect "\0A{\0A\09.reg .pred complete;\0A\09waitLoop:\0A\09mbarrier.try_wait.parity.shared::cta.b64 complete, [$0], $1;\0A\09@!complete bra.uni waitLoop;\0A}\0A", "r,r"(ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 156688), i32 0) #4, !dbg !57 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !57 + %1481 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !57 + %1482 = extractvalue { i32, i1 } %1481, 1, !dbg !57 + %1483 = and i1 %369, %1482, !dbg !57 + tail call void asm sideeffect "@$0 mbarrier.inval.shared::cta.b64 [$1];", "b,r"(i1 %1483, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 156688)) #4, !dbg !57 + tail call void @llvm.nvvm.fence.proxy.async.shared_cta(), !dbg !58 + %1484 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !58 + %1485 = extractvalue { i32, i1 } %1484, 1, !dbg !58 + %1486 = and i1 %369, %1485, !dbg !58 + tail call void asm sideeffect "@$0 mbarrier.init.shared::cta.b64 [$1], 1;", "b,r"(i1 %1486, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 156688)) #4, !dbg !58 + br i1 %949, label %1487, label %1498, !dbg !58 + +1487: ; preds = %1480 + %1488 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !58 + %1489 = extractvalue { i32, i1 } %1488, 1, !dbg !58 + %1490 = ptrtoint ptr addrspace(3) %1275 to i32, !dbg !58 + %1491 = lshr exact i32 %1490, 4, !dbg !58 + %1492 = and i32 %1491, 16383, !dbg !58 + %1493 = zext nneg i32 %1492 to i64, !dbg !58 + %1494 = or disjoint i64 %1493, 4611756662049472512, !dbg !58 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %89, i64 %1494, i64 %474, i32 68256912, i1 false, i1 %1489) #4, !dbg !58 + %1495 = add nuw nsw i64 %1493, 4611756662049472640, !dbg !58 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %89, i64 %1495, i64 %475, i32 68256912, i1 true, i1 %1489) #4, !dbg !58 + %1496 = add nuw nsw i64 %1493, 4611756662049472768, !dbg !58 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %89, i64 %1496, i64 %476, i32 68256912, i1 true, i1 %1489) #4, !dbg !58 + %1497 = add nuw nsw i64 %1493, 4611756662049472896, !dbg !58 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %89, i64 %1497, i64 %477, i32 68256912, i1 true, i1 %1489) #4, !dbg !58 + tail call void asm sideeffect "@$0 tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [$1];", "b,r"(i1 %1489, i32 ptrtoint (ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 156688) to i32)) #4, !dbg !58 + br label %1498, !dbg !58 + +1498: ; preds = %1487, %1480 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !58 + tail call void asm sideeffect "\0A{\0A\09.reg .pred complete;\0A\09waitLoop:\0A\09mbarrier.try_wait.parity.shared::cta.b64 complete, [$0], $1;\0A\09@!complete bra.uni waitLoop;\0A}\0A", "r,r"(ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 156688), i32 0) #4, !dbg !58 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !58 + %1499 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !58 + %1500 = extractvalue { i32, i1 } %1499, 1, !dbg !58 + %1501 = and i1 %369, %1500, !dbg !58 + tail call void asm sideeffect "@$0 mbarrier.inval.shared::cta.b64 [$1];", "b,r"(i1 %1501, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 156688)) #4, !dbg !58 + %1502 = getelementptr bfloat, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 98304), i32 %879, !dbg !81 + %1503 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %393, !dbg !81 + %1504 = load bfloat, ptr addrspace(3) %1503, align 2, !dbg !81 + %1505 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %394, !dbg !81 + %1506 = load bfloat, ptr addrspace(3) %1505, align 2, !dbg !81 + %1507 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %395, !dbg !81 + %1508 = load bfloat, ptr addrspace(3) %1507, align 2, !dbg !81 + %1509 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %396, !dbg !81 + %1510 = load bfloat, ptr addrspace(3) %1509, align 2, !dbg !81 + %1511 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %397, !dbg !81 + %1512 = load bfloat, ptr addrspace(3) %1511, align 2, !dbg !81 + %1513 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %398, !dbg !81 + %1514 = load bfloat, ptr addrspace(3) %1513, align 2, !dbg !81 + %1515 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %399, !dbg !81 + %1516 = load bfloat, ptr addrspace(3) %1515, align 2, !dbg !81 + %1517 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %400, !dbg !81 + %1518 = load bfloat, ptr addrspace(3) %1517, align 2, !dbg !81 + %1519 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %401, !dbg !81 + %1520 = load bfloat, ptr addrspace(3) %1519, align 2, !dbg !81 + %1521 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %402, !dbg !81 + %1522 = load bfloat, ptr addrspace(3) %1521, align 2, !dbg !81 + %1523 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %403, !dbg !81 + %1524 = load bfloat, ptr addrspace(3) %1523, align 2, !dbg !81 + %1525 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %404, !dbg !81 + %1526 = load bfloat, ptr addrspace(3) %1525, align 2, !dbg !81 + %1527 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %405, !dbg !81 + %1528 = load bfloat, ptr addrspace(3) %1527, align 2, !dbg !81 + %1529 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %406, !dbg !81 + %1530 = load bfloat, ptr addrspace(3) %1529, align 2, !dbg !81 + %1531 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %407, !dbg !81 + %1532 = load bfloat, ptr addrspace(3) %1531, align 2, !dbg !81 + %1533 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %408, !dbg !81 + %1534 = load bfloat, ptr addrspace(3) %1533, align 2, !dbg !81 + %1535 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %409, !dbg !81 + %1536 = load bfloat, ptr addrspace(3) %1535, align 2, !dbg !81 + %1537 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %410, !dbg !81 + %1538 = load bfloat, ptr addrspace(3) %1537, align 2, !dbg !81 + %1539 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %411, !dbg !81 + %1540 = load bfloat, ptr addrspace(3) %1539, align 2, !dbg !81 + %1541 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %412, !dbg !81 + %1542 = load bfloat, ptr addrspace(3) %1541, align 2, !dbg !81 + %1543 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %413, !dbg !81 + %1544 = load bfloat, ptr addrspace(3) %1543, align 2, !dbg !81 + %1545 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %414, !dbg !81 + %1546 = load bfloat, ptr addrspace(3) %1545, align 2, !dbg !81 + %1547 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %415, !dbg !81 + %1548 = load bfloat, ptr addrspace(3) %1547, align 2, !dbg !81 + %1549 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %416, !dbg !81 + %1550 = load bfloat, ptr addrspace(3) %1549, align 2, !dbg !81 + %1551 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %417, !dbg !81 + %1552 = load bfloat, ptr addrspace(3) %1551, align 2, !dbg !81 + %1553 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %418, !dbg !81 + %1554 = load bfloat, ptr addrspace(3) %1553, align 2, !dbg !81 + %1555 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %419, !dbg !81 + %1556 = load bfloat, ptr addrspace(3) %1555, align 2, !dbg !81 + %1557 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %420, !dbg !81 + %1558 = load bfloat, ptr addrspace(3) %1557, align 2, !dbg !81 + %1559 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %421, !dbg !81 + %1560 = load bfloat, ptr addrspace(3) %1559, align 2, !dbg !81 + %1561 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %422, !dbg !81 + %1562 = load bfloat, ptr addrspace(3) %1561, align 2, !dbg !81 + %1563 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %423, !dbg !81 + %1564 = load bfloat, ptr addrspace(3) %1563, align 2, !dbg !81 + %1565 = getelementptr inbounds nuw i8, ptr addrspace(3) %1502, i32 %424, !dbg !81 + %1566 = load bfloat, ptr addrspace(3) %1565, align 2, !dbg !81 + %1567 = getelementptr bfloat, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 114688), i32 %879, !dbg !82 + %1568 = shufflevector <64 x float> %488, <64 x float> poison, <32 x i32> , !dbg !114 + %1569 = fmul <32 x float> %1568, %1343, !dbg !114 + %1570 = fpext bfloat %1504 to float, !dbg !115 + %1571 = fpext bfloat %1506 to float, !dbg !115 + %1572 = fpext bfloat %1508 to float, !dbg !115 + %1573 = fpext bfloat %1510 to float, !dbg !115 + %1574 = fpext bfloat %1512 to float, !dbg !115 + %1575 = fpext bfloat %1514 to float, !dbg !115 + %1576 = fpext bfloat %1516 to float, !dbg !115 + %1577 = fpext bfloat %1518 to float, !dbg !115 + %1578 = fpext bfloat %1520 to float, !dbg !115 + %1579 = fpext bfloat %1522 to float, !dbg !115 + %1580 = fpext bfloat %1524 to float, !dbg !115 + %1581 = fpext bfloat %1526 to float, !dbg !115 + %1582 = fpext bfloat %1528 to float, !dbg !115 + %1583 = fpext bfloat %1530 to float, !dbg !115 + %1584 = fpext bfloat %1532 to float, !dbg !115 + %1585 = fpext bfloat %1534 to float, !dbg !115 + %1586 = fpext bfloat %1536 to float, !dbg !115 + %1587 = fpext bfloat %1538 to float, !dbg !115 + %1588 = fpext bfloat %1540 to float, !dbg !115 + %1589 = fpext bfloat %1542 to float, !dbg !115 + %1590 = fpext bfloat %1544 to float, !dbg !115 + %1591 = fpext bfloat %1546 to float, !dbg !115 + %1592 = fpext bfloat %1548 to float, !dbg !115 + %1593 = fpext bfloat %1550 to float, !dbg !115 + %1594 = fpext bfloat %1552 to float, !dbg !115 + %1595 = fpext bfloat %1554 to float, !dbg !115 + %1596 = fpext bfloat %1556 to float, !dbg !115 + %1597 = fpext bfloat %1558 to float, !dbg !115 + %1598 = fpext bfloat %1560 to float, !dbg !115 + %1599 = fpext bfloat %1562 to float, !dbg !115 + %1600 = fpext bfloat %1564 to float, !dbg !115 + %1601 = fpext bfloat %1566 to float, !dbg !115 + %1602 = fmul float %.0.i326, %1570, !dbg !115 + %1603 = fmul float %.0.i329, %1571, !dbg !115 + %1604 = fmul float %.0.i332, %1572, !dbg !115 + %1605 = fmul float %.0.i335, %1573, !dbg !115 + %1606 = fmul float %.0.i338, %1574, !dbg !115 + %1607 = fmul float %.0.i341, %1575, !dbg !115 + %1608 = fmul float %.0.i344, %1576, !dbg !115 + %1609 = fmul float %.0.i347, %1577, !dbg !115 + %1610 = fmul float %.0.i350, %1578, !dbg !115 + %1611 = fmul float %.0.i353, %1579, !dbg !115 + %1612 = fmul float %.0.i356, %1580, !dbg !115 + %1613 = fmul float %.0.i359, %1581, !dbg !115 + %1614 = fmul float %.0.i362, %1582, !dbg !115 + %1615 = fmul float %.0.i365, %1583, !dbg !115 + %1616 = fmul float %.0.i368, %1584, !dbg !115 + %1617 = fmul float %.0.i371, %1585, !dbg !115 + %1618 = fmul float %.0.i374, %1586, !dbg !115 + %1619 = fmul float %.0.i377, %1587, !dbg !115 + %1620 = fmul float %.0.i380, %1588, !dbg !115 + %1621 = fmul float %.0.i383, %1589, !dbg !115 + %1622 = fmul float %.0.i386, %1590, !dbg !115 + %1623 = fmul float %.0.i389, %1591, !dbg !115 + %1624 = fmul float %.0.i392, %1592, !dbg !115 + %1625 = fmul float %.0.i395, %1593, !dbg !115 + %1626 = fmul float %.0.i398, %1594, !dbg !115 + %1627 = fmul float %.0.i401, %1595, !dbg !115 + %1628 = fmul float %.0.i404, %1596, !dbg !115 + %1629 = fmul float %.0.i407, %1597, !dbg !115 + %1630 = fmul float %.0.i410, %1598, !dbg !115 + %1631 = fmul float %.0.i413, %1599, !dbg !115 + %1632 = fmul float %.0.i416, %1600, !dbg !115 + %1633 = fmul float %.0.i419, %1601, !dbg !115 + tail call void asm sideeffect "@$0 tcgen05.st.sync.aligned.16x32bx2.x32.b32 [$1 + 0], 32, {$2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13, $14, $15, $16, $17, $18, $19, $20, $21, $22, $23, $24, $25, $26, $27, $28, $29, $30, $31, $32, $33};", "b,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r"(i1 true, i32 %1473, float %1602, float %1603, float %1604, float %1605, float %1606, float %1607, float %1608, float %1609, float %1610, float %1611, float %1612, float %1613, float %1614, float %1615, float %1616, float %1617, float %1618, float %1619, float %1620, float %1621, float %1622, float %1623, float %1624, float %1625, float %1626, float %1627, float %1628, float %1629, float %1630, float %1631, float %1632, float %1633) #4, !dbg !59 + tail call void @llvm.nvvm.tcgen05.wait.st(), !dbg !59 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !59 + tail call void @llvm.nvvm.fence.proxy.async.shared_cta(), !dbg !59 + %1634 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !59 + %1635 = extractvalue { i32, i1 } %1634, 1, !dbg !59 + %1636 = and i1 %369, %1635, !dbg !59 + tail call void asm sideeffect "@$0 mbarrier.init.shared::cta.b64 [$1], 1;", "b,r"(i1 %1636, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 156688)) #4, !dbg !59 + br i1 %949, label %1637, label %1640, !dbg !59 + +1637: ; preds = %1498 + %1638 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !59 + %1639 = extractvalue { i32, i1 } %1638, 1, !dbg !59 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::tf32 [ $0 + 0 ], [ $1 + 0 ], $2, $3, $4;", "r,r,l,r,b,b"(i32 %90, i32 %463, i64 %373, i32 68159760, i1 false, i1 %1639) #4, !dbg !59 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::tf32 [ $0 + 0 ], [ $1 + 8 ], $2, $3, $4;", "r,r,l,r,b,b"(i32 %90, i32 %463, i64 %464, i32 68159760, i1 true, i1 %1639) #4, !dbg !59 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::tf32 [ $0 + 0 ], [ $1 + 16 ], $2, $3, $4;", "r,r,l,r,b,b"(i32 %90, i32 %463, i64 %465, i32 68159760, i1 true, i1 %1639) #4, !dbg !59 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::tf32 [ $0 + 0 ], [ $1 + 24 ], $2, $3, $4;", "r,r,l,r,b,b"(i32 %90, i32 %463, i64 %466, i32 68159760, i1 true, i1 %1639) #4, !dbg !59 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::tf32 [ $0 + 0 ], [ $1 + 32 ], $2, $3, $4;", "r,r,l,r,b,b"(i32 %90, i32 %463, i64 %467, i32 68159760, i1 true, i1 %1639) #4, !dbg !59 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::tf32 [ $0 + 0 ], [ $1 + 40 ], $2, $3, $4;", "r,r,l,r,b,b"(i32 %90, i32 %463, i64 %468, i32 68159760, i1 true, i1 %1639) #4, !dbg !59 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::tf32 [ $0 + 0 ], [ $1 + 48 ], $2, $3, $4;", "r,r,l,r,b,b"(i32 %90, i32 %463, i64 %469, i32 68159760, i1 true, i1 %1639) #4, !dbg !59 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::tf32 [ $0 + 0 ], [ $1 + 56 ], $2, $3, $4;", "r,r,l,r,b,b"(i32 %90, i32 %463, i64 %470, i32 68159760, i1 true, i1 %1639) #4, !dbg !59 + tail call void asm sideeffect "@$0 tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [$1];", "b,r"(i1 %1639, i32 ptrtoint (ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 156688) to i32)) #4, !dbg !59 + br label %1640, !dbg !59 + +1640: ; preds = %1637, %1498 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !59 + tail call void asm sideeffect "\0A{\0A\09.reg .pred complete;\0A\09waitLoop:\0A\09mbarrier.try_wait.parity.shared::cta.b64 complete, [$0], $1;\0A\09@!complete bra.uni waitLoop;\0A}\0A", "r,r"(ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 156688), i32 0) #4, !dbg !59 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !59 + %1641 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !59 + %1642 = extractvalue { i32, i1 } %1641, 1, !dbg !59 + %1643 = and i1 %369, %1642, !dbg !59 + tail call void asm sideeffect "@$0 mbarrier.inval.shared::cta.b64 [$1];", "b,r"(i1 %1643, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 156688)) #4, !dbg !59 + tail call void @llvm.nvvm.fence.proxy.async.shared_cta(), !dbg !60 + %1644 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !60 + %1645 = extractvalue { i32, i1 } %1644, 1, !dbg !60 + %1646 = and i1 %369, %1645, !dbg !60 + tail call void asm sideeffect "@$0 mbarrier.init.shared::cta.b64 [$1], 1;", "b,r"(i1 %1646, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072)) #4, !dbg !60 + br i1 %949, label %1647, label %1658, !dbg !60 + +1647: ; preds = %1640 + %1648 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !60 + %1649 = extractvalue { i32, i1 } %1648, 1, !dbg !60 + %1650 = ptrtoint ptr addrspace(3) %1567 to i32, !dbg !60 + %1651 = lshr exact i32 %1650, 4, !dbg !60 + %1652 = and i32 %1651, 16383, !dbg !60 + %1653 = zext nneg i32 %1652 to i64, !dbg !60 + %1654 = or disjoint i64 %1653, 4611756662049472512, !dbg !60 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %91, i64 %1654, i64 %474, i32 68256912, i1 false, i1 %1649) #4, !dbg !60 + %1655 = add nuw nsw i64 %1653, 4611756662049472640, !dbg !60 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %91, i64 %1655, i64 %475, i32 68256912, i1 true, i1 %1649) #4, !dbg !60 + %1656 = add nuw nsw i64 %1653, 4611756662049472768, !dbg !60 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %91, i64 %1656, i64 %476, i32 68256912, i1 true, i1 %1649) #4, !dbg !60 + %1657 = add nuw nsw i64 %1653, 4611756662049472896, !dbg !60 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %91, i64 %1657, i64 %477, i32 68256912, i1 true, i1 %1649) #4, !dbg !60 + tail call void asm sideeffect "@$0 tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [$1];", "b,r"(i1 %1649, i32 ptrtoint (ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072) to i32)) #4, !dbg !60 + br label %1658, !dbg !60 + +1658: ; preds = %1647, %1640 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !60 + tail call void asm sideeffect "\0A{\0A\09.reg .pred complete;\0A\09waitLoop:\0A\09mbarrier.try_wait.parity.shared::cta.b64 complete, [$0], $1;\0A\09@!complete bra.uni waitLoop;\0A}\0A", "r,r"(ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 0) #4, !dbg !60 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !60 + %1659 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !60 + %1660 = extractvalue { i32, i1 } %1659, 1, !dbg !60 + %1661 = and i1 %369, %1660, !dbg !60 + tail call void asm sideeffect "@$0 mbarrier.inval.shared::cta.b64 [$1];", "b,r"(i1 %1661, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072)) #4, !dbg !60 + %1662 = add i32 %487, 1, !dbg !45 + %1663 = icmp sgt i32 %1662, 1, !dbg !45 + %1664 = select i1 %1663, i32 0, i32 %1662, !dbg !45 + %1665 = sub nsw i64 %480, %indvars.iv, !dbg !46 + %1666 = shl nsw i64 %1665, 6, !dbg !47 + %1667 = or disjoint i64 %1666, %51, !dbg !48 + %1668 = or disjoint i64 %1666, %52, !dbg !48 + %1669 = or disjoint i64 %1666, %53, !dbg !48 + %1670 = or disjoint i64 %1666, %54, !dbg !48 + %1671 = or disjoint i64 %1666, %50, !dbg !48 + %1672 = icmp slt i64 %1667, %21, !dbg !49 + %1673 = icmp slt i64 %1668, %21, !dbg !49 + %1674 = icmp slt i64 %1669, %21, !dbg !49 + %1675 = icmp slt i64 %1670, %21, !dbg !49 + %1676 = icmp slt i64 %1671, %21, !dbg !49 + %1677 = add nsw i64 %1666, 64, !dbg !50 + %1678 = tail call i64 @llvm.smin.i64(i64 %1677, i64 %21), !dbg !51 + %1679 = add nsw i64 %80, %1678, !dbg !53 + %.idx323 = shl i64 %1679, 7, !dbg !54 + %gep431 = getelementptr i8, ptr addrspace(1) %invariant.gep430, i64 %.idx323, !dbg !54 + %1680 = getelementptr float, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 156672), i32 %1664, !dbg !55 + %1681 = select i1 %489, i32 4, i32 0, !dbg !55 + tail call void asm sideeffect "@$3 cp.async.ca.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x4, $2;", "r,l,r,b"(ptr addrspace(3) %1680, ptr addrspace(1) %gep431, i32 %1681, i1 %85) #4, !dbg !55 + %1682 = add i32 %967, %88, !dbg !57 + %1683 = tail call { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } asm sideeffect "tcgen05.ld.sync.aligned.16x32bx2.x32.b32 {$0, $1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13, $14, $15, $16, $17, $18, $19, $20, $21, $22, $23, $24, $25, $26, $27, $28, $29, $30, $31}, [$32 + 0], 32;", "=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,r"(i32 %1682) #4, !dbg !57 + %1684 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 0, !dbg !57 + %1685 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 1, !dbg !57 + %1686 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 2, !dbg !57 + %1687 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 3, !dbg !57 + %1688 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 4, !dbg !57 + %1689 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 5, !dbg !57 + %1690 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 6, !dbg !57 + %1691 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 7, !dbg !57 + %1692 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 8, !dbg !57 + %1693 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 9, !dbg !57 + %1694 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 10, !dbg !57 + %1695 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 11, !dbg !57 + %1696 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 12, !dbg !57 + %1697 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 13, !dbg !57 + %1698 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 14, !dbg !57 + %1699 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 15, !dbg !57 + %1700 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 16, !dbg !57 + %1701 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 17, !dbg !57 + %1702 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 18, !dbg !57 + %1703 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 19, !dbg !57 + %1704 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 20, !dbg !57 + %1705 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 21, !dbg !57 + %1706 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 22, !dbg !57 + %1707 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 23, !dbg !57 + %1708 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 24, !dbg !57 + %1709 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 25, !dbg !57 + %1710 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 26, !dbg !57 + %1711 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 27, !dbg !57 + %1712 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 28, !dbg !57 + %1713 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 29, !dbg !57 + %1714 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 30, !dbg !57 + %1715 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1683, 31, !dbg !57 + tail call void @llvm.nvvm.tcgen05.wait.ld(), !dbg !57 + %1716 = add i32 %967, %89, !dbg !58 + %1717 = tail call { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } asm sideeffect "tcgen05.ld.sync.aligned.16x32bx2.x32.b32 {$0, $1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13, $14, $15, $16, $17, $18, $19, $20, $21, $22, $23, $24, $25, $26, $27, $28, $29, $30, $31}, [$32 + 0], 32;", "=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,r"(i32 %1716) #4, !dbg !58 + %1718 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 0, !dbg !58 + %1719 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 1, !dbg !58 + %1720 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 2, !dbg !58 + %1721 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 3, !dbg !58 + %1722 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 4, !dbg !58 + %1723 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 5, !dbg !58 + %1724 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 6, !dbg !58 + %1725 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 7, !dbg !58 + %1726 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 8, !dbg !58 + %1727 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 9, !dbg !58 + %1728 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 10, !dbg !58 + %1729 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 11, !dbg !58 + %1730 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 12, !dbg !58 + %1731 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 13, !dbg !58 + %1732 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 14, !dbg !58 + %1733 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 15, !dbg !58 + %1734 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 16, !dbg !58 + %1735 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 17, !dbg !58 + %1736 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 18, !dbg !58 + %1737 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 19, !dbg !58 + %1738 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 20, !dbg !58 + %1739 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 21, !dbg !58 + %1740 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 22, !dbg !58 + %1741 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 23, !dbg !58 + %1742 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 24, !dbg !58 + %1743 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 25, !dbg !58 + %1744 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 26, !dbg !58 + %1745 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 27, !dbg !58 + %1746 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 28, !dbg !58 + %1747 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 29, !dbg !58 + %1748 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 30, !dbg !58 + %1749 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1717, 31, !dbg !58 + tail call void @llvm.nvvm.tcgen05.wait.ld(), !dbg !58 + %1750 = add i32 %967, %90, !dbg !59 + %1751 = tail call { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } asm sideeffect "tcgen05.ld.sync.aligned.16x32bx2.x32.b32 {$0, $1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13, $14, $15, $16, $17, $18, $19, $20, $21, $22, $23, $24, $25, $26, $27, $28, $29, $30, $31}, [$32 + 0], 32;", "=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,r"(i32 %1750) #4, !dbg !59 + %1752 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 0, !dbg !59 + %1753 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 1, !dbg !59 + %1754 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 2, !dbg !59 + %1755 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 3, !dbg !59 + %1756 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 4, !dbg !59 + %1757 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 5, !dbg !59 + %1758 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 6, !dbg !59 + %1759 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 7, !dbg !59 + %1760 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 8, !dbg !59 + %1761 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 9, !dbg !59 + %1762 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 10, !dbg !59 + %1763 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 11, !dbg !59 + %1764 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 12, !dbg !59 + %1765 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 13, !dbg !59 + %1766 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 14, !dbg !59 + %1767 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 15, !dbg !59 + %1768 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 16, !dbg !59 + %1769 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 17, !dbg !59 + %1770 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 18, !dbg !59 + %1771 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 19, !dbg !59 + %1772 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 20, !dbg !59 + %1773 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 21, !dbg !59 + %1774 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 22, !dbg !59 + %1775 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 23, !dbg !59 + %1776 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 24, !dbg !59 + %1777 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 25, !dbg !59 + %1778 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 26, !dbg !59 + %1779 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 27, !dbg !59 + %1780 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 28, !dbg !59 + %1781 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 29, !dbg !59 + %1782 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 30, !dbg !59 + %1783 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1751, 31, !dbg !59 + tail call void @llvm.nvvm.tcgen05.wait.ld(), !dbg !59 + %1784 = add i32 %967, %91, !dbg !60 + %1785 = tail call { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } asm sideeffect "tcgen05.ld.sync.aligned.16x32bx2.x32.b32 {$0, $1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13, $14, $15, $16, $17, $18, $19, $20, $21, $22, $23, $24, $25, $26, $27, $28, $29, $30, $31}, [$32 + 0], 32;", "=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,r"(i32 %1784) #4, !dbg !60 + %1786 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 0, !dbg !60 + %1787 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 1, !dbg !60 + %1788 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 2, !dbg !60 + %1789 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 3, !dbg !60 + %1790 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 4, !dbg !60 + %1791 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 5, !dbg !60 + %1792 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 6, !dbg !60 + %1793 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 7, !dbg !60 + %1794 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 8, !dbg !60 + %1795 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 9, !dbg !60 + %1796 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 10, !dbg !60 + %1797 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 11, !dbg !60 + %1798 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 12, !dbg !60 + %1799 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 13, !dbg !60 + %1800 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 14, !dbg !60 + %1801 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 15, !dbg !60 + %1802 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 16, !dbg !60 + %1803 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 17, !dbg !60 + %1804 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 18, !dbg !60 + %1805 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 19, !dbg !60 + %1806 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 20, !dbg !60 + %1807 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 21, !dbg !60 + %1808 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 22, !dbg !60 + %1809 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 23, !dbg !60 + %1810 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 24, !dbg !60 + %1811 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 25, !dbg !60 + %1812 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 26, !dbg !60 + %1813 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 27, !dbg !60 + %1814 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 28, !dbg !60 + %1815 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 29, !dbg !60 + %1816 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 30, !dbg !60 + %1817 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1785, 31, !dbg !60 + tail call void @llvm.nvvm.tcgen05.wait.ld(), !dbg !60 + %1818 = insertelement <64 x i32> poison, i32 %1752, i64 0, !dbg !59 + %1819 = insertelement <64 x i32> %1818, i32 %1753, i64 1, !dbg !59 + %1820 = insertelement <64 x i32> %1819, i32 %1754, i64 2, !dbg !59 + %1821 = insertelement <64 x i32> %1820, i32 %1755, i64 3, !dbg !59 + %1822 = insertelement <64 x i32> %1821, i32 %1756, i64 4, !dbg !59 + %1823 = insertelement <64 x i32> %1822, i32 %1757, i64 5, !dbg !59 + %1824 = insertelement <64 x i32> %1823, i32 %1758, i64 6, !dbg !59 + %1825 = insertelement <64 x i32> %1824, i32 %1759, i64 7, !dbg !59 + %1826 = insertelement <64 x i32> %1825, i32 %1760, i64 8, !dbg !59 + %1827 = insertelement <64 x i32> %1826, i32 %1761, i64 9, !dbg !59 + %1828 = insertelement <64 x i32> %1827, i32 %1762, i64 10, !dbg !59 + %1829 = insertelement <64 x i32> %1828, i32 %1763, i64 11, !dbg !59 + %1830 = insertelement <64 x i32> %1829, i32 %1764, i64 12, !dbg !59 + %1831 = insertelement <64 x i32> %1830, i32 %1765, i64 13, !dbg !59 + %1832 = insertelement <64 x i32> %1831, i32 %1766, i64 14, !dbg !59 + %1833 = insertelement <64 x i32> %1832, i32 %1767, i64 15, !dbg !59 + %1834 = insertelement <64 x i32> %1833, i32 %1768, i64 16, !dbg !59 + %1835 = insertelement <64 x i32> %1834, i32 %1769, i64 17, !dbg !59 + %1836 = insertelement <64 x i32> %1835, i32 %1770, i64 18, !dbg !59 + %1837 = insertelement <64 x i32> %1836, i32 %1771, i64 19, !dbg !59 + %1838 = insertelement <64 x i32> %1837, i32 %1772, i64 20, !dbg !59 + %1839 = insertelement <64 x i32> %1838, i32 %1773, i64 21, !dbg !59 + %1840 = insertelement <64 x i32> %1839, i32 %1774, i64 22, !dbg !59 + %1841 = insertelement <64 x i32> %1840, i32 %1775, i64 23, !dbg !59 + %1842 = insertelement <64 x i32> %1841, i32 %1776, i64 24, !dbg !59 + %1843 = insertelement <64 x i32> %1842, i32 %1777, i64 25, !dbg !59 + %1844 = insertelement <64 x i32> %1843, i32 %1778, i64 26, !dbg !59 + %1845 = insertelement <64 x i32> %1844, i32 %1779, i64 27, !dbg !59 + %1846 = insertelement <64 x i32> %1845, i32 %1780, i64 28, !dbg !59 + %1847 = insertelement <64 x i32> %1846, i32 %1781, i64 29, !dbg !59 + %1848 = insertelement <64 x i32> %1847, i32 %1782, i64 30, !dbg !59 + %1849 = insertelement <64 x i32> %1848, i32 %1783, i64 31, !dbg !59 + %1850 = insertelement <64 x i32> %1849, i32 %1684, i64 32, !dbg !59 + %1851 = insertelement <64 x i32> %1850, i32 %1685, i64 33, !dbg !59 + %1852 = insertelement <64 x i32> %1851, i32 %1686, i64 34, !dbg !59 + %1853 = insertelement <64 x i32> %1852, i32 %1687, i64 35, !dbg !59 + %1854 = insertelement <64 x i32> %1853, i32 %1688, i64 36, !dbg !59 + %1855 = insertelement <64 x i32> %1854, i32 %1689, i64 37, !dbg !59 + %1856 = insertelement <64 x i32> %1855, i32 %1690, i64 38, !dbg !59 + %1857 = insertelement <64 x i32> %1856, i32 %1691, i64 39, !dbg !59 + %1858 = insertelement <64 x i32> %1857, i32 %1692, i64 40, !dbg !59 + %1859 = insertelement <64 x i32> %1858, i32 %1693, i64 41, !dbg !59 + %1860 = insertelement <64 x i32> %1859, i32 %1694, i64 42, !dbg !59 + %1861 = insertelement <64 x i32> %1860, i32 %1695, i64 43, !dbg !59 + %1862 = insertelement <64 x i32> %1861, i32 %1696, i64 44, !dbg !59 + %1863 = insertelement <64 x i32> %1862, i32 %1697, i64 45, !dbg !59 + %1864 = insertelement <64 x i32> %1863, i32 %1698, i64 46, !dbg !59 + %1865 = insertelement <64 x i32> %1864, i32 %1699, i64 47, !dbg !59 + %1866 = insertelement <64 x i32> %1865, i32 %1700, i64 48, !dbg !59 + %1867 = insertelement <64 x i32> %1866, i32 %1701, i64 49, !dbg !59 + %1868 = insertelement <64 x i32> %1867, i32 %1702, i64 50, !dbg !59 + %1869 = insertelement <64 x i32> %1868, i32 %1703, i64 51, !dbg !59 + %1870 = insertelement <64 x i32> %1869, i32 %1704, i64 52, !dbg !59 + %1871 = insertelement <64 x i32> %1870, i32 %1705, i64 53, !dbg !59 + %1872 = insertelement <64 x i32> %1871, i32 %1706, i64 54, !dbg !59 + %1873 = insertelement <64 x i32> %1872, i32 %1707, i64 55, !dbg !59 + %1874 = insertelement <64 x i32> %1873, i32 %1708, i64 56, !dbg !59 + %1875 = insertelement <64 x i32> %1874, i32 %1709, i64 57, !dbg !59 + %1876 = insertelement <64 x i32> %1875, i32 %1710, i64 58, !dbg !59 + %1877 = insertelement <64 x i32> %1876, i32 %1711, i64 59, !dbg !59 + %1878 = insertelement <64 x i32> %1877, i32 %1712, i64 60, !dbg !59 + %1879 = insertelement <64 x i32> %1878, i32 %1713, i64 61, !dbg !59 + %1880 = insertelement <64 x i32> %1879, i32 %1714, i64 62, !dbg !59 + %1881 = insertelement <64 x i32> %1880, i32 %1715, i64 63, !dbg !59 + %1882 = bitcast <64 x i32> %1881 to <64 x float>, !dbg !59 + %1883 = fmul <64 x float> %482, %1882, !dbg !85 + %1884 = insertelement <64 x i32> poison, i32 %1786, i64 0, !dbg !60 + %1885 = insertelement <64 x i32> %1884, i32 %1787, i64 1, !dbg !60 + %1886 = insertelement <64 x i32> %1885, i32 %1788, i64 2, !dbg !60 + %1887 = insertelement <64 x i32> %1886, i32 %1789, i64 3, !dbg !60 + %1888 = insertelement <64 x i32> %1887, i32 %1790, i64 4, !dbg !60 + %1889 = insertelement <64 x i32> %1888, i32 %1791, i64 5, !dbg !60 + %1890 = insertelement <64 x i32> %1889, i32 %1792, i64 6, !dbg !60 + %1891 = insertelement <64 x i32> %1890, i32 %1793, i64 7, !dbg !60 + %1892 = insertelement <64 x i32> %1891, i32 %1794, i64 8, !dbg !60 + %1893 = insertelement <64 x i32> %1892, i32 %1795, i64 9, !dbg !60 + %1894 = insertelement <64 x i32> %1893, i32 %1796, i64 10, !dbg !60 + %1895 = insertelement <64 x i32> %1894, i32 %1797, i64 11, !dbg !60 + %1896 = insertelement <64 x i32> %1895, i32 %1798, i64 12, !dbg !60 + %1897 = insertelement <64 x i32> %1896, i32 %1799, i64 13, !dbg !60 + %1898 = insertelement <64 x i32> %1897, i32 %1800, i64 14, !dbg !60 + %1899 = insertelement <64 x i32> %1898, i32 %1801, i64 15, !dbg !60 + %1900 = insertelement <64 x i32> %1899, i32 %1802, i64 16, !dbg !60 + %1901 = insertelement <64 x i32> %1900, i32 %1803, i64 17, !dbg !60 + %1902 = insertelement <64 x i32> %1901, i32 %1804, i64 18, !dbg !60 + %1903 = insertelement <64 x i32> %1902, i32 %1805, i64 19, !dbg !60 + %1904 = insertelement <64 x i32> %1903, i32 %1806, i64 20, !dbg !60 + %1905 = insertelement <64 x i32> %1904, i32 %1807, i64 21, !dbg !60 + %1906 = insertelement <64 x i32> %1905, i32 %1808, i64 22, !dbg !60 + %1907 = insertelement <64 x i32> %1906, i32 %1809, i64 23, !dbg !60 + %1908 = insertelement <64 x i32> %1907, i32 %1810, i64 24, !dbg !60 + %1909 = insertelement <64 x i32> %1908, i32 %1811, i64 25, !dbg !60 + %1910 = insertelement <64 x i32> %1909, i32 %1812, i64 26, !dbg !60 + %1911 = insertelement <64 x i32> %1910, i32 %1813, i64 27, !dbg !60 + %1912 = insertelement <64 x i32> %1911, i32 %1814, i64 28, !dbg !60 + %1913 = insertelement <64 x i32> %1912, i32 %1815, i64 29, !dbg !60 + %1914 = insertelement <64 x i32> %1913, i32 %1816, i64 30, !dbg !60 + %1915 = insertelement <64 x i32> %1914, i32 %1817, i64 31, !dbg !60 + %1916 = insertelement <64 x i32> %1915, i32 %1718, i64 32, !dbg !60 + %1917 = insertelement <64 x i32> %1916, i32 %1719, i64 33, !dbg !60 + %1918 = insertelement <64 x i32> %1917, i32 %1720, i64 34, !dbg !60 + %1919 = insertelement <64 x i32> %1918, i32 %1721, i64 35, !dbg !60 + %1920 = insertelement <64 x i32> %1919, i32 %1722, i64 36, !dbg !60 + %1921 = insertelement <64 x i32> %1920, i32 %1723, i64 37, !dbg !60 + %1922 = insertelement <64 x i32> %1921, i32 %1724, i64 38, !dbg !60 + %1923 = insertelement <64 x i32> %1922, i32 %1725, i64 39, !dbg !60 + %1924 = insertelement <64 x i32> %1923, i32 %1726, i64 40, !dbg !60 + %1925 = insertelement <64 x i32> %1924, i32 %1727, i64 41, !dbg !60 + %1926 = insertelement <64 x i32> %1925, i32 %1728, i64 42, !dbg !60 + %1927 = insertelement <64 x i32> %1926, i32 %1729, i64 43, !dbg !60 + %1928 = insertelement <64 x i32> %1927, i32 %1730, i64 44, !dbg !60 + %1929 = insertelement <64 x i32> %1928, i32 %1731, i64 45, !dbg !60 + %1930 = insertelement <64 x i32> %1929, i32 %1732, i64 46, !dbg !60 + %1931 = insertelement <64 x i32> %1930, i32 %1733, i64 47, !dbg !60 + %1932 = insertelement <64 x i32> %1931, i32 %1734, i64 48, !dbg !60 + %1933 = insertelement <64 x i32> %1932, i32 %1735, i64 49, !dbg !60 + %1934 = insertelement <64 x i32> %1933, i32 %1736, i64 50, !dbg !60 + %1935 = insertelement <64 x i32> %1934, i32 %1737, i64 51, !dbg !60 + %1936 = insertelement <64 x i32> %1935, i32 %1738, i64 52, !dbg !60 + %1937 = insertelement <64 x i32> %1936, i32 %1739, i64 53, !dbg !60 + %1938 = insertelement <64 x i32> %1937, i32 %1740, i64 54, !dbg !60 + %1939 = insertelement <64 x i32> %1938, i32 %1741, i64 55, !dbg !60 + %1940 = insertelement <64 x i32> %1939, i32 %1742, i64 56, !dbg !60 + %1941 = insertelement <64 x i32> %1940, i32 %1743, i64 57, !dbg !60 + %1942 = insertelement <64 x i32> %1941, i32 %1744, i64 58, !dbg !60 + %1943 = insertelement <64 x i32> %1942, i32 %1745, i64 59, !dbg !60 + %1944 = insertelement <64 x i32> %1943, i32 %1746, i64 60, !dbg !60 + %1945 = insertelement <64 x i32> %1944, i32 %1747, i64 61, !dbg !60 + %1946 = insertelement <64 x i32> %1945, i32 %1748, i64 62, !dbg !60 + %1947 = insertelement <64 x i32> %1946, i32 %1749, i64 63, !dbg !60 + %1948 = bitcast <64 x i32> %1947 to <64 x float>, !dbg !60 + %1949 = fsub <64 x float> %1883, %1948, !dbg !116 + %1950 = shufflevector <32 x float> %1569, <32 x float> %1344, <64 x i32> , !dbg !117 + %1951 = fadd <64 x float> %1950, %1949, !dbg !117 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !55 + %.idx324 = shl nsw i64 %1671, 7, !dbg !61 + %1952 = getelementptr i8, ptr addrspace(1) %59, i64 %.idx324, !dbg !61 + %1953 = shl i32 %1664, 6, !dbg !62 + %1954 = and i1 %489, %1676, !dbg !45 + %gep433 = getelementptr float, ptr addrspace(3) %97, i32 %1953, !dbg !62 + %1955 = select i1 %1954, i32 4, i32 0, !dbg !62 + tail call void asm sideeffect "@$3 cp.async.ca.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x4, $2;", "r,l,r,b"(ptr addrspace(3) %gep433, ptr addrspace(1) %1952, i32 %1955, i1 %95) #4, !dbg !62 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !62 + %gep435 = getelementptr float, ptr addrspace(3) %99, i32 %1953, !dbg !62 + tail call void asm sideeffect "@$3 cp.async.ca.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x4, $2;", "r,l,r,b"(ptr addrspace(3) %gep435, ptr addrspace(1) %1952, i32 %1955, i1 %95) #4, !dbg !62 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !62 + %1956 = shl nsw i64 %1667, 12, !dbg !63 + %1957 = shl nsw i64 %1668, 12, !dbg !63 + %1958 = shl nsw i64 %1669, 12, !dbg !63 + %1959 = shl nsw i64 %1670, 12, !dbg !63 + %gep437 = getelementptr bfloat, ptr addrspace(1) %invariant.gep436, i64 %1956, !dbg !65 + %gep439 = getelementptr bfloat, ptr addrspace(1) %invariant.gep436, i64 %1957, !dbg !65 + %gep441 = getelementptr bfloat, ptr addrspace(1) %invariant.gep436, i64 %1958, !dbg !65 + %gep443 = getelementptr bfloat, ptr addrspace(1) %invariant.gep436, i64 %1959, !dbg !65 + %gep445 = getelementptr bfloat, ptr addrspace(1) %invariant.gep444, i64 %1956, !dbg !67 + %gep447 = getelementptr bfloat, ptr addrspace(1) %invariant.gep444, i64 %1957, !dbg !67 + %gep449 = getelementptr bfloat, ptr addrspace(1) %invariant.gep444, i64 %1958, !dbg !67 + %gep451 = getelementptr bfloat, ptr addrspace(1) %invariant.gep444, i64 %1959, !dbg !67 + %1960 = shl i32 %1664, 12, !dbg !68 + %1961 = getelementptr bfloat, ptr addrspace(3) @global_smem, i32 %1960, !dbg !68 + %1962 = and i1 %489, %1672, !dbg !45 + %1963 = and i1 %489, %1673, !dbg !45 + %1964 = and i1 %489, %1674, !dbg !45 + %1965 = and i1 %489, %1675, !dbg !45 + %1966 = getelementptr inbounds nuw i8, ptr addrspace(3) %1961, i32 %125, !dbg !68 + %1967 = select i1 %1962, i32 16, i32 0, !dbg !68 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %1966, ptr addrspace(1) %gep445, i32 %1967) #4, !dbg !68 + %1968 = getelementptr inbounds nuw i8, ptr addrspace(3) %1961, i32 %128, !dbg !68 + %1969 = select i1 %1963, i32 16, i32 0, !dbg !68 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %1968, ptr addrspace(1) %gep447, i32 %1969) #4, !dbg !68 + %1970 = getelementptr inbounds nuw i8, ptr addrspace(3) %1961, i32 %131, !dbg !68 + %1971 = select i1 %1964, i32 16, i32 0, !dbg !68 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %1970, ptr addrspace(1) %gep449, i32 %1971) #4, !dbg !68 + %1972 = getelementptr inbounds nuw i8, ptr addrspace(3) %1961, i32 %134, !dbg !68 + %1973 = select i1 %1965, i32 16, i32 0, !dbg !68 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %1972, ptr addrspace(1) %gep451, i32 %1973) #4, !dbg !68 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !68 + %1974 = getelementptr bfloat, ptr addrspace(1) %29, i64 %1956, !dbg !69 + %1975 = getelementptr bfloat, ptr addrspace(1) %29, i64 %1957, !dbg !69 + %1976 = getelementptr bfloat, ptr addrspace(1) %29, i64 %1958, !dbg !69 + %1977 = getelementptr bfloat, ptr addrspace(1) %29, i64 %1959, !dbg !69 + %1978 = getelementptr bfloat, ptr addrspace(1) %1974, i64 %49, !dbg !70 + %1979 = getelementptr bfloat, ptr addrspace(1) %1975, i64 %49, !dbg !70 + %1980 = getelementptr bfloat, ptr addrspace(1) %1976, i64 %49, !dbg !70 + %1981 = getelementptr bfloat, ptr addrspace(1) %1977, i64 %49, !dbg !70 + %1982 = getelementptr bfloat, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %1960, !dbg !71 + %1983 = getelementptr inbounds nuw i8, ptr addrspace(3) %1982, i32 %147, !dbg !71 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %1983, ptr addrspace(1) %1978, i32 %1967) #4, !dbg !71 + %1984 = getelementptr inbounds nuw i8, ptr addrspace(3) %1982, i32 %149, !dbg !71 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %1984, ptr addrspace(1) %1979, i32 %1969) #4, !dbg !71 + %1985 = getelementptr inbounds nuw i8, ptr addrspace(3) %1982, i32 %151, !dbg !71 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %1985, ptr addrspace(1) %1980, i32 %1971) #4, !dbg !71 + %1986 = getelementptr inbounds nuw i8, ptr addrspace(3) %1982, i32 %153, !dbg !71 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %1986, ptr addrspace(1) %1981, i32 %1973) #4, !dbg !71 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !71 + %1987 = getelementptr bfloat, ptr addrspace(1) %1974, i64 %62, !dbg !72 + %1988 = getelementptr bfloat, ptr addrspace(1) %1975, i64 %62, !dbg !72 + %1989 = getelementptr bfloat, ptr addrspace(1) %1976, i64 %62, !dbg !72 + %1990 = getelementptr bfloat, ptr addrspace(1) %1977, i64 %62, !dbg !72 + %1991 = getelementptr bfloat, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 32768), i32 %1960, !dbg !73 + %1992 = getelementptr inbounds nuw i8, ptr addrspace(3) %1991, i32 %147, !dbg !73 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %1992, ptr addrspace(1) %1987, i32 %1967) #4, !dbg !73 + %1993 = getelementptr inbounds nuw i8, ptr addrspace(3) %1991, i32 %149, !dbg !73 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %1993, ptr addrspace(1) %1988, i32 %1969) #4, !dbg !73 + %1994 = getelementptr inbounds nuw i8, ptr addrspace(3) %1991, i32 %151, !dbg !73 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %1994, ptr addrspace(1) %1989, i32 %1971) #4, !dbg !73 + %1995 = getelementptr inbounds nuw i8, ptr addrspace(3) %1991, i32 %153, !dbg !73 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %1995, ptr addrspace(1) %1990, i32 %1973) #4, !dbg !73 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !73 + %1996 = getelementptr bfloat, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 49152), i32 %1960, !dbg !74 + %1997 = getelementptr inbounds nuw i8, ptr addrspace(3) %1996, i32 %125, !dbg !74 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %1997, ptr addrspace(1) %gep437, i32 %1967) #4, !dbg !74 + %1998 = getelementptr inbounds nuw i8, ptr addrspace(3) %1996, i32 %128, !dbg !74 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %1998, ptr addrspace(1) %gep439, i32 %1969) #4, !dbg !74 + %1999 = getelementptr inbounds nuw i8, ptr addrspace(3) %1996, i32 %131, !dbg !74 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %1999, ptr addrspace(1) %gep441, i32 %1971) #4, !dbg !74 + %2000 = getelementptr inbounds nuw i8, ptr addrspace(3) %1996, i32 %134, !dbg !74 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %2000, ptr addrspace(1) %gep443, i32 %1973) #4, !dbg !74 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !74 + %2001 = getelementptr bfloat, ptr addrspace(1) %60, i64 %1956, !dbg !75 + %2002 = getelementptr bfloat, ptr addrspace(1) %60, i64 %1957, !dbg !75 + %2003 = getelementptr bfloat, ptr addrspace(1) %60, i64 %1958, !dbg !75 + %2004 = getelementptr bfloat, ptr addrspace(1) %60, i64 %1959, !dbg !75 + %2005 = getelementptr bfloat, ptr addrspace(1) %61, i64 %1956, !dbg !76 + %2006 = getelementptr bfloat, ptr addrspace(1) %61, i64 %1957, !dbg !76 + %2007 = getelementptr bfloat, ptr addrspace(1) %61, i64 %1958, !dbg !76 + %2008 = getelementptr bfloat, ptr addrspace(1) %61, i64 %1959, !dbg !76 + %2009 = getelementptr bfloat, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 65536), i32 %1960, !dbg !77 + %2010 = getelementptr inbounds nuw i8, ptr addrspace(3) %2009, i32 %147, !dbg !77 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %2010, ptr addrspace(1) %2001, i32 %1967) #4, !dbg !77 + %2011 = getelementptr inbounds nuw i8, ptr addrspace(3) %2009, i32 %149, !dbg !77 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %2011, ptr addrspace(1) %2002, i32 %1969) #4, !dbg !77 + %2012 = getelementptr inbounds nuw i8, ptr addrspace(3) %2009, i32 %151, !dbg !77 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %2012, ptr addrspace(1) %2003, i32 %1971) #4, !dbg !77 + %2013 = getelementptr inbounds nuw i8, ptr addrspace(3) %2009, i32 %153, !dbg !77 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %2013, ptr addrspace(1) %2004, i32 %1973) #4, !dbg !77 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !77 + %2014 = getelementptr bfloat, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 81920), i32 %1960, !dbg !78 + %2015 = getelementptr inbounds nuw i8, ptr addrspace(3) %2014, i32 %125, !dbg !78 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %2015, ptr addrspace(1) %2005, i32 %1967) #4, !dbg !78 + %2016 = getelementptr inbounds nuw i8, ptr addrspace(3) %2014, i32 %128, !dbg !78 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %2016, ptr addrspace(1) %2006, i32 %1969) #4, !dbg !78 + %2017 = getelementptr inbounds nuw i8, ptr addrspace(3) %2014, i32 %131, !dbg !78 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %2017, ptr addrspace(1) %2007, i32 %1971) #4, !dbg !78 + %2018 = getelementptr inbounds nuw i8, ptr addrspace(3) %2014, i32 %134, !dbg !78 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %2018, ptr addrspace(1) %2008, i32 %1973) #4, !dbg !78 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !78 + %2019 = getelementptr bfloat, ptr addrspace(1) %63, i64 %1956, !dbg !79 + %2020 = getelementptr bfloat, ptr addrspace(1) %63, i64 %1957, !dbg !79 + %2021 = getelementptr bfloat, ptr addrspace(1) %63, i64 %1958, !dbg !79 + %2022 = getelementptr bfloat, ptr addrspace(1) %63, i64 %1959, !dbg !79 + %2023 = getelementptr bfloat, ptr addrspace(1) %64, i64 %1956, !dbg !80 + %2024 = getelementptr bfloat, ptr addrspace(1) %64, i64 %1957, !dbg !80 + %2025 = getelementptr bfloat, ptr addrspace(1) %64, i64 %1958, !dbg !80 + %2026 = getelementptr bfloat, ptr addrspace(1) %64, i64 %1959, !dbg !80 + %2027 = getelementptr bfloat, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 98304), i32 %1960, !dbg !81 + %2028 = getelementptr inbounds nuw i8, ptr addrspace(3) %2027, i32 %125, !dbg !81 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %2028, ptr addrspace(1) %2019, i32 %1967) #4, !dbg !81 + %2029 = getelementptr inbounds nuw i8, ptr addrspace(3) %2027, i32 %128, !dbg !81 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %2029, ptr addrspace(1) %2020, i32 %1969) #4, !dbg !81 + %2030 = getelementptr inbounds nuw i8, ptr addrspace(3) %2027, i32 %131, !dbg !81 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %2030, ptr addrspace(1) %2021, i32 %1971) #4, !dbg !81 + %2031 = getelementptr inbounds nuw i8, ptr addrspace(3) %2027, i32 %134, !dbg !81 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %2031, ptr addrspace(1) %2022, i32 %1973) #4, !dbg !81 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !81 + %2032 = getelementptr bfloat, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 114688), i32 %1960, !dbg !82 + %2033 = getelementptr inbounds nuw i8, ptr addrspace(3) %2032, i32 %147, !dbg !82 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %2033, ptr addrspace(1) %2023, i32 %1967) #4, !dbg !82 + %2034 = getelementptr inbounds nuw i8, ptr addrspace(3) %2032, i32 %149, !dbg !82 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %2034, ptr addrspace(1) %2024, i32 %1969) #4, !dbg !82 + %2035 = getelementptr inbounds nuw i8, ptr addrspace(3) %2032, i32 %151, !dbg !82 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %2035, ptr addrspace(1) %2025, i32 %1971) #4, !dbg !82 + %2036 = getelementptr inbounds nuw i8, ptr addrspace(3) %2032, i32 %153, !dbg !82 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %2036, ptr addrspace(1) %2026, i32 %1973) #4, !dbg !82 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !82 + %indvars.iv.next = add nsw i64 %indvars.iv, 1, !dbg !45 + %exitcond.not = icmp eq i64 %indvars.iv.next, %wide.trip.count, !dbg !45 + br i1 %exitcond.not, label %._crit_edge, label %483, !dbg !45 + +._crit_edge: ; preds = %1658, %12 + %2037 = inttoptr i32 %15 to ptr addrspace(6), !dbg !13 + tail call void @llvm.nvvm.cp.async.wait.group(i32 0), !dbg !45 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !45 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !118 + tail call void asm sideeffect "@$0 tcgen05.dealloc.cta_group::1.sync.aligned.b32 $1, 256;", "b,r"(i1 %14, ptr addrspace(6) %2037) #4, !dbg !118 + ret void, !dbg !118 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #2 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind speculatable willreturn memory(none) +declare i64 @llvm.smin.i64(i64, i64) #3 + +; Function Attrs: nounwind +declare void @llvm.nvvm.cp.async.commit.group() #4 + +; Function Attrs: nounwind +declare void @llvm.nvvm.cp.async.wait.group(i32 immarg) #4 + +; Function Attrs: convergent nocallback nofree nounwind memory(argmem: read) +declare { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) readonly captures(none)) #5 + +; Function Attrs: convergent nocallback nofree nounwind memory(argmem: read) +declare i32 @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x1.b16.p3(ptr addrspace(3) readonly captures(none)) #5 + +; Function Attrs: nocallback nounwind +declare void @llvm.nvvm.fence.proxy.async.shared_cta() #6 + +; Function Attrs: convergent mustprogress nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: readwrite) +declare { i32, i1 } @llvm.nvvm.elect.sync(i32) #7 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.idx.i32(i32, i32, i32, i32) #8 + +; Function Attrs: convergent nounwind memory(inaccessiblemem: readwrite) +declare void @llvm.nvvm.tcgen05.wait.ld() #9 + +; Function Attrs: convergent nounwind memory(inaccessiblemem: readwrite) +declare void @llvm.nvvm.tcgen05.wait.st() #9 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #10 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.ex2.approx.ftz.f32(float) #11 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.ex2.approx.f32(float) #11 + +; Function Attrs: nocallback nocreateundeforpoison nofree nosync nounwind speculatable willreturn memory(none) +declare i32 @llvm.smax.i32(i32, i32) #12 + +attributes #0 = { nounwind "nvvm.reqntid"="128" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind } +attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind speculatable willreturn memory(none) } +attributes #4 = { nounwind } +attributes #5 = { convergent nocallback nofree nounwind memory(argmem: read) } +attributes #6 = { nocallback nounwind } +attributes #7 = { convergent mustprogress nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: readwrite) } +attributes #8 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #9 = { convergent nounwind memory(inaccessiblemem: readwrite) } +attributes #10 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #11 = { mustprogress nocallback nofree nosync nounwind willreturn memory(none) } +attributes #12 = { nocallback nocreateundeforpoison nofree nosync nounwind speculatable willreturn memory(none) } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "chunk_delta_h.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64", linkageName: "chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64", scope: !1, file: !1, line: 370, type: !6, scopeLine: 370, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !8, !10, !8, !8, !8, !8, !11, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 370, scope: !5) +!14 = !DILocation(line: 399, column: 24, scope: !5) +!15 = !DILocation(line: 399, column: 30, scope: !5) +!16 = !DILocation(line: 401, column: 34, scope: !5) +!17 = !DILocation(line: 402, column: 23, scope: !5) +!18 = !DILocation(line: 402, column: 34, scope: !5) +!19 = !DILocation(line: 409, column: 25, scope: !5) +!20 = !DILocation(line: 43, column: 17, scope: !21, inlinedAt: !23) +!21 = distinct !DILexicalBlockFile(scope: !5, file: !22, discriminator: 0) +!22 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!23 = !DILocation(line: 410, column: 24, scope: !24) +!24 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!25 = !DILocation(line: 43, column: 30, scope: !21, inlinedAt: !23) +!26 = !DILocation(line: 431, column: 16, scope: !5) +!27 = !DILocation(line: 431, column: 20, scope: !5) +!28 = !DILocation(line: 431, column: 53, scope: !5) +!29 = !DILocation(line: 431, column: 9, scope: !5) +!30 = !DILocation(line: 432, column: 9, scope: !5) +!31 = !DILocation(line: 433, column: 9, scope: !5) +!32 = !DILocation(line: 434, column: 10, scope: !5) +!33 = !DILocation(line: 435, column: 10, scope: !5) +!34 = !DILocation(line: 436, column: 11, scope: !5) +!35 = !DILocation(line: 446, column: 16, scope: !5) +!36 = !DILocation(line: 446, column: 34, scope: !5) +!37 = !DILocation(line: 446, column: 21, scope: !5) +!38 = !DILocation(line: 450, column: 16, scope: !5) +!39 = !DILocation(line: 489, column: 26, scope: !5) +!40 = !DILocation(line: 528, scope: !5) +!41 = !DILocation(line: 585, column: 18, scope: !5) +!42 = !DILocation(line: 586, column: 18, scope: !5) +!43 = !DILocation(line: 602, column: 22, scope: !5) +!44 = !DILocation(line: 603, column: 22, scope: !5) +!45 = !DILocation(line: 489, column: 33, scope: !5) +!46 = !DILocation(line: 490, column: 27, scope: !5) +!47 = !DILocation(line: 491, column: 26, scope: !5) +!48 = !DILocation(line: 491, column: 31, scope: !5) +!49 = !DILocation(line: 492, column: 20, scope: !5) +!50 = !DILocation(line: 525, column: 41, scope: !5) +!51 = !DILocation(line: 525, column: 45, scope: !5) +!52 = !DILocation(line: 525, column: 50, scope: !5) +!53 = !DILocation(line: 527, column: 41, scope: !5) +!54 = !DILocation(line: 527, scope: !5) +!55 = !DILocation(line: 527, column: 30, scope: !5) +!56 = !DILocation(line: 557, column: 56, scope: !5) +!57 = !DILocation(line: 600, column: 47, scope: !5) +!58 = !DILocation(line: 600, column: 89, scope: !5) +!59 = !DILocation(line: 617, column: 51, scope: !5) +!60 = !DILocation(line: 617, column: 93, scope: !5) +!61 = !DILocation(line: 528, column: 39, scope: !5) +!62 = !DILocation(line: 529, column: 26, scope: !5) +!63 = !DILocation(line: 532, column: 36, scope: !5) +!64 = !DILocation(line: 532, column: 20, scope: !5) +!65 = !DILocation(line: 532, column: 44, scope: !5) +!66 = !DILocation(line: 534, column: 20, scope: !5) +!67 = !DILocation(line: 534, column: 44, scope: !5) +!68 = !DILocation(line: 536, column: 23, scope: !5) +!69 = !DILocation(line: 539, column: 18, scope: !5) +!70 = !DILocation(line: 539, column: 41, scope: !5) +!71 = !DILocation(line: 540, column: 22, scope: !5) +!72 = !DILocation(line: 550, column: 45, scope: !5) +!73 = !DILocation(line: 551, column: 26, scope: !5) +!74 = !DILocation(line: 581, column: 24, scope: !5) +!75 = !DILocation(line: 585, column: 34, scope: !5) +!76 = !DILocation(line: 586, column: 34, scope: !5) +!77 = !DILocation(line: 587, column: 22, scope: !5) +!78 = !DILocation(line: 588, column: 22, scope: !5) +!79 = !DILocation(line: 602, column: 38, scope: !5) +!80 = !DILocation(line: 603, column: 38, scope: !5) +!81 = !DILocation(line: 604, column: 26, scope: !5) +!82 = !DILocation(line: 605, column: 26, scope: !5) +!83 = !DILocation(line: 411, column: 20, scope: !5) +!84 = !DILocation(line: 497, column: 60, scope: !5) +!85 = !DILocation(line: 617, column: 73, scope: !5) +!86 = !DILocation(line: 0, scope: !5) +!87 = !DILocation(line: 497, column: 44, scope: !5) +!88 = !DILocation(line: 497, column: 64, scope: !5) +!89 = !DILocation(line: 499, column: 33, scope: !5) +!90 = !DILocation(line: 499, column: 24, scope: !5) +!91 = !DILocation(line: 505, column: 48, scope: !5) +!92 = !DILocation(line: 505, column: 68, scope: !5) +!93 = !DILocation(line: 507, column: 37, scope: !5) +!94 = !DILocation(line: 507, column: 28, scope: !5) +!95 = !DILocation(line: 47, column: 33, scope: !96, inlinedAt: !98) +!96 = distinct !DILexicalBlockFile(scope: !5, file: !97, discriminator: 0) +!97 = !DIFile(filename: "op.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils") +!98 = !DILocation(line: 530, column: 31, scope: !24) +!99 = !DILocation(line: 47, column: 33, scope: !96, inlinedAt: !100) +!100 = !DILocation(line: 531, column: 27, scope: !24) +!101 = !DILocation(line: 533, column: 46, scope: !5) +!102 = !DILocation(line: 547, column: 31, scope: !5) +!103 = !DILocation(line: 580, column: 49, scope: !5) +!104 = !DILocation(line: 47, column: 33, scope: !96, inlinedAt: !105) +!105 = !DILocation(line: 580, column: 39, scope: !24) +!106 = !DILocation(line: 580, column: 55, scope: !5) +!107 = !DILocation(line: 581, column: 16, scope: !5) +!108 = !DILocation(line: 580, column: 20, scope: !5) +!109 = !DILocation(line: 583, column: 32, scope: !5) +!110 = !DILocation(line: 583, column: 24, scope: !5) +!111 = !DILocation(line: 590, column: 21, scope: !5) +!112 = !DILocation(line: 591, column: 24, scope: !5) +!113 = !DILocation(line: 600, column: 55, scope: !5) +!114 = !DILocation(line: 607, column: 25, scope: !5) +!115 = !DILocation(line: 608, column: 28, scope: !5) +!116 = !DILocation(line: 617, column: 81, scope: !5) +!117 = !DILocation(line: 617, column: 25, scope: !5) +!118 = !DILocation(line: 653, column: 4, scope: !5) diff --git a/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.ptx b/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.ptx new file mode 100644 index 0000000000000000000000000000000000000000..442d6dbcb879027491b3a0390252e50fbdc539d1 --- /dev/null +++ b/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.ptx @@ -0,0 +1,2956 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64 // -- Begin function chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64 +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; + // @chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64 +.visible .entry chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64( + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64_param_0, + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64_param_1, + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64_param_2, + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64_param_3, + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64_param_4, + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64_param_5, + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64_param_6, + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64_param_7, + .param .f32 chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64_param_8, + .param .u32 chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64_param_9, + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64_param_10, + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64_param_11 +) +.reqntid 128 +{ + .reg .pred %p<70>; + .reg .b16 %rs<129>; + .reg .b32 %r<1174>; + .reg .b64 %rd<571>; + .loc 1 370 0 // chunk_delta_h.py:370:0 +$L__func_begin0: + .loc 1 370 0 // chunk_delta_h.py:370:0 + +// %bb.0: + ld.param.b64 %rd66, [chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64_param_6]; + ld.param.b64 %rd64, [chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64_param_4]; + ld.param.b64 %rd63, [chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64_param_3]; + ld.param.b64 %rd62, [chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64_param_2]; + ld.param.b64 %rd61, [chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64_param_1]; + ld.param.b64 %rd60, [chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64_param_0]; +$L__tmp0: + .loc 1 370 0 // chunk_delta_h.py:370 + mov.u32 %r1, %tid.x; + setp.lt.u32 %p5, %r1, 32; + mov.b32 %r67, global_smem; + // begin inline asm + @%p5 tcgen05.alloc.cta_group::1.sync.aligned.shared::cta.b32 [%r67], 256; + // end inline asm + bar.sync 0; + ld.shared.b32 %r1171, [global_smem]; + bar.sync 0; + // begin inline asm + @%p5 tcgen05.relinquish_alloc_permit.cta_group::1.sync.aligned; + // end inline asm + .loc 1 399 24 // chunk_delta_h.py:399:24 + mov.u32 %r2, %ctaid.x; + .loc 1 399 30 // chunk_delta_h.py:399:30 + cvt.u64.u32 %rd136, %r2; + ld.param.b32 %r150, [chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64_param_9]; + .loc 1 402 23 // chunk_delta_h.py:402:23 + shr.u64 %rd1, %rd136, 6; + .loc 1 402 34 // chunk_delta_h.py:402:34 + bfe.u64 %rd2, %rd136, 1, 5; + .loc 1 409 25 // chunk_delta_h.py:409:25 + cvt.s64.s32 %rd3, %r150; + mul.lo.s64 %rd4, %rd1, %rd3; +$L__tmp1: + .loc 2 43 17 // standard.py:43:17 @[ chunk_delta_h.py:410:24 ] + add.s32 %r151, %r150, 63; + .loc 2 43 30 // standard.py:43:30 @[ chunk_delta_h.py:410:24 ] + shr.s32 %r152, %r151, 31; + shr.u32 %r153, %r152, 26; + add.s32 %r154, %r151, %r153; + shr.s32 %r3, %r154, 6; +$L__tmp2: + .loc 1 431 16 // chunk_delta_h.py:431:16 + shl.b64 %rd137, %rd4, 5; + .loc 1 431 20 // chunk_delta_h.py:431:20 + or.b64 %rd138, %rd137, %rd2; + .loc 1 431 9 // chunk_delta_h.py:431:9 + shl.b64 %rd139, %rd138, 8; + add.s64 %rd140, %rd60, %rd139; + .loc 1 432 9 // chunk_delta_h.py:432:9 + add.s64 %rd141, %rd61, %rd139; + .loc 1 433 9 // chunk_delta_h.py:433:9 + add.s64 %rd142, %rd62, %rd139; + .loc 1 434 10 // chunk_delta_h.py:434:10 + add.s64 %rd143, %rd64, %rd139; + .loc 1 435 10 // chunk_delta_h.py:435:10 + add.s64 %rd144, %rd66, %rd139; + .loc 1 446 16 // chunk_delta_h.py:446:16 + mul.wide.u32 %rd146, %r2, 64; + and.b64 %rd147, %rd146, 64; + .loc 1 446 34 // chunk_delta_h.py:446:34 + and.b32 %r5, %r1, 7; + shl.b32 %r155, %r5, 3; + and.b32 %r156, %r1, 63; + bfe.u32 %r157, %r1, 3, 4; + or.b32 %r158, %r157, 16; + or.b32 %r159, %r157, 32; + or.b32 %r160, %r157, 48; + .loc 1 446 21 // chunk_delta_h.py:446:21 + cvt.u64.u32 %rd148, %r155; + cvt.u64.u32 %rd5, %r156; + cvt.u64.u32 %rd6, %r157; + cvt.u64.u32 %rd149, %r158; + cvt.u64.u32 %rd150, %r159; + cvt.u64.u32 %rd151, %r160; + or.b64 %rd152, %rd147, %rd148; + .loc 1 489 26 // chunk_delta_h.py:489:26 + add.s32 %r10, %r3, -1; + .loc 1 528 0 // chunk_delta_h.py:528 + shl.b64 %rd153, %rd138, 2; + add.s64 %rd154, %rd63, %rd153; + .loc 1 585 18 // chunk_delta_h.py:585:18 + mul.wide.u32 %rd155, %r155, 2; + add.s64 %rd156, %rd142, %rd155; + .loc 1 586 18 // chunk_delta_h.py:586:18 + add.s64 %rd157, %rd140, %rd155; + .loc 1 489 33 // chunk_delta_h.py:489:33 + setp.lt.s32 %p8, %r151, 64; + setp.gt.s32 %p9, %r151, 63; + .loc 1 491 26 // chunk_delta_h.py:491:26 + mul.wide.s32 %rd158, %r10, 64; + .loc 1 491 31 // chunk_delta_h.py:491:31 + or.b64 %rd159, %rd158, %rd6; + or.b64 %rd160, %rd158, %rd149; + or.b64 %rd161, %rd158, %rd150; + or.b64 %rd162, %rd158, %rd151; + or.b64 %rd163, %rd158, %rd5; + .loc 1 492 20 // chunk_delta_h.py:492:20 + setp.lt.s64 %p69, %rd159, %rd3; + setp.lt.s64 %p68, %rd160, %rd3; + setp.lt.s64 %p67, %rd161, %rd3; + setp.lt.s64 %p66, %rd162, %rd3; + setp.lt.s64 %p10, %rd163, %rd3; + .loc 1 525 41 // chunk_delta_h.py:525:41 + add.s64 %rd164, %rd158, 64; + .loc 1 525 45 // chunk_delta_h.py:525:45 + min.s64 %rd165, %rd164, %rd3; + .loc 1 525 50 // chunk_delta_h.py:525:50 + add.s64 %rd7, %rd4, 144115188075855871; + .loc 1 527 41 // chunk_delta_h.py:527:41 + add.s64 %rd166, %rd7, %rd165; + .loc 1 527 0 // chunk_delta_h.py:527 + shl.b64 %rd167, %rd166, 7; + shl.b64 %rd168, %rd2, 2; + .loc 1 489 33 // chunk_delta_h.py:489:33 + add.s64 %rd9, %rd63, %rd168; + .loc 1 527 0 // chunk_delta_h.py:527 + add.s64 %rd67, %rd9, %rd167; + .loc 1 527 30 // chunk_delta_h.py:527:30 + and.b32 %r11, %r1, 127; + setp.eq.b32 %p6, %r11, 0; + selp.b32 %r69, 4, 0, %p9; + add.s32 %r68, %r67, 156672; + // begin inline asm + @%p6 cp.async.ca.shared.global [ %r68 + 0 ], [ %rd67 + 0 ], 0x4, %r69; + // end inline asm + cp.async.commit_group; + .loc 1 528 39 // chunk_delta_h.py:528:39 + shl.b64 %rd169, %rd163, 7; + add.s64 %rd68, %rd154, %rd169; + .loc 1 529 26 // chunk_delta_h.py:529:26 + and.b32 %r12, %r1, 64; + setp.eq.b32 %p7, %r12, 0; + shl.b32 %r161, %r156, 2; + add.s32 %r13, %r67, %r161; + add.s32 %r70, %r13, 155648; + selp.b32 %r162, 4, 0, %p10; + selp.b32 %r72, %r162, 0, %p9; + // begin inline asm + @%p7 cp.async.ca.shared.global [ %r70 + 0 ], [ %rd68 + 0 ], 0x4, %r72; + // end inline asm + cp.async.commit_group; + add.s32 %r71, %r13, 156160; + // begin inline asm + @%p7 cp.async.ca.shared.global [ %r71 + 0 ], [ %rd68 + 0 ], 0x4, %r72; + // end inline asm + cp.async.commit_group; + .loc 1 532 20 // chunk_delta_h.py:532:20 + shl.b64 %rd170, %rd159, 13; + add.s64 %rd171, %rd144, %rd170; + shl.b64 %rd172, %rd160, 13; + add.s64 %rd173, %rd144, %rd172; + shl.b64 %rd174, %rd161, 13; + add.s64 %rd175, %rd144, %rd174; + shl.b64 %rd176, %rd162, 13; + add.s64 %rd177, %rd144, %rd176; + .loc 1 532 44 // chunk_delta_h.py:532:44 + shl.b64 %rd178, %rd152, 1; + add.s64 %rd81, %rd171, %rd178; + add.s64 %rd82, %rd173, %rd178; + add.s64 %rd83, %rd175, %rd178; + add.s64 %rd84, %rd177, %rd178; + .loc 1 534 20 // chunk_delta_h.py:534:20 + add.s64 %rd179, %rd143, %rd170; + add.s64 %rd180, %rd143, %rd172; + add.s64 %rd181, %rd143, %rd174; + add.s64 %rd182, %rd143, %rd176; + .loc 1 534 44 // chunk_delta_h.py:534:44 + add.s64 %rd69, %rd179, %rd178; + add.s64 %rd70, %rd180, %rd178; + add.s64 %rd71, %rd181, %rd178; + add.s64 %rd72, %rd182, %rd178; + .loc 1 536 23 // chunk_delta_h.py:536:23 + shl.b32 %r14, %r1, 4; + and.b32 %r15, %r14, 2032; + add.s32 %r73, %r67, %r15; + selp.b32 %r163, 16, 0, %p69; + selp.b32 %r78, %r163, 0, %p9; + // begin inline asm + cp.async.cg.shared.global [ %r73 + 0 ], [ %rd69 + 0 ], 0x10, %r78; + // end inline asm + add.s32 %r74, %r73, 2048; + selp.b32 %r164, 16, 0, %p68; + selp.b32 %r80, %r164, 0, %p9; + // begin inline asm + cp.async.cg.shared.global [ %r74 + 0 ], [ %rd70 + 0 ], 0x10, %r80; + // end inline asm + add.s32 %r75, %r73, 4096; + selp.b32 %r165, 16, 0, %p67; + selp.b32 %r82, %r165, 0, %p9; + // begin inline asm + cp.async.cg.shared.global [ %r75 + 0 ], [ %rd71 + 0 ], 0x10, %r82; + // end inline asm + add.s32 %r76, %r73, 6144; + selp.b32 %r166, 16, 0, %p66; + selp.b32 %r84, %r166, 0, %p9; + // begin inline asm + cp.async.cg.shared.global [ %r76 + 0 ], [ %rd72 + 0 ], 0x10, %r84; + // end inline asm + cp.async.commit_group; + .loc 1 539 18 // chunk_delta_h.py:539:18 + add.s64 %rd183, %rd141, %rd170; + add.s64 %rd184, %rd141, %rd172; + add.s64 %rd185, %rd141, %rd174; + add.s64 %rd186, %rd141, %rd176; + .loc 1 539 41 // chunk_delta_h.py:539:41 + add.s64 %rd73, %rd183, %rd155; + add.s64 %rd74, %rd184, %rd155; + add.s64 %rd75, %rd185, %rd155; + add.s64 %rd76, %rd186, %rd155; + .loc 1 540 22 // chunk_delta_h.py:540:22 + and.b32 %r16, %r1, 56; + shl.b32 %r167, %r16, 1; + xor.b32 %r17, %r15, %r167; + add.s32 %r168, %r67, %r17; + add.s32 %r77, %r168, 16384; + // begin inline asm + cp.async.cg.shared.global [ %r77 + 0 ], [ %rd73 + 0 ], 0x10, %r78; + // end inline asm + add.s32 %r79, %r168, 18432; + // begin inline asm + cp.async.cg.shared.global [ %r79 + 0 ], [ %rd74 + 0 ], 0x10, %r80; + // end inline asm + add.s32 %r81, %r168, 20480; + // begin inline asm + cp.async.cg.shared.global [ %r81 + 0 ], [ %rd75 + 0 ], 0x10, %r82; + // end inline asm + add.s32 %r83, %r168, 22528; + // begin inline asm + cp.async.cg.shared.global [ %r83 + 0 ], [ %rd76 + 0 ], 0x10, %r84; + // end inline asm + cp.async.commit_group; + .loc 1 550 45 // chunk_delta_h.py:550:45 + add.s64 %rd77, %rd73, 128; + add.s64 %rd78, %rd74, 128; + add.s64 %rd79, %rd75, 128; + add.s64 %rd80, %rd76, 128; + .loc 1 551 26 // chunk_delta_h.py:551:26 + add.s32 %r85, %r168, 32768; + // begin inline asm + cp.async.cg.shared.global [ %r85 + 0 ], [ %rd77 + 0 ], 0x10, %r78; + // end inline asm + add.s32 %r86, %r168, 34816; + // begin inline asm + cp.async.cg.shared.global [ %r86 + 0 ], [ %rd78 + 0 ], 0x10, %r80; + // end inline asm + add.s32 %r87, %r168, 36864; + // begin inline asm + cp.async.cg.shared.global [ %r87 + 0 ], [ %rd79 + 0 ], 0x10, %r82; + // end inline asm + add.s32 %r88, %r168, 38912; + // begin inline asm + cp.async.cg.shared.global [ %r88 + 0 ], [ %rd80 + 0 ], 0x10, %r84; + // end inline asm + cp.async.commit_group; + .loc 1 581 24 // chunk_delta_h.py:581:24 + add.s32 %r89, %r73, 49152; + // begin inline asm + cp.async.cg.shared.global [ %r89 + 0 ], [ %rd81 + 0 ], 0x10, %r78; + // end inline asm + add.s32 %r90, %r73, 51200; + // begin inline asm + cp.async.cg.shared.global [ %r90 + 0 ], [ %rd82 + 0 ], 0x10, %r80; + // end inline asm + add.s32 %r91, %r73, 53248; + // begin inline asm + cp.async.cg.shared.global [ %r91 + 0 ], [ %rd83 + 0 ], 0x10, %r82; + // end inline asm + add.s32 %r92, %r73, 55296; + // begin inline asm + cp.async.cg.shared.global [ %r92 + 0 ], [ %rd84 + 0 ], 0x10, %r84; + // end inline asm + cp.async.commit_group; + .loc 1 585 34 // chunk_delta_h.py:585:34 + add.s64 %rd85, %rd156, %rd170; + add.s64 %rd86, %rd156, %rd172; + add.s64 %rd87, %rd156, %rd174; + add.s64 %rd88, %rd156, %rd176; + .loc 1 586 34 // chunk_delta_h.py:586:34 + add.s64 %rd89, %rd157, %rd170; + add.s64 %rd90, %rd157, %rd172; + add.s64 %rd91, %rd157, %rd174; + add.s64 %rd92, %rd157, %rd176; + .loc 1 587 22 // chunk_delta_h.py:587:22 + add.s32 %r93, %r168, 65536; + // begin inline asm + cp.async.cg.shared.global [ %r93 + 0 ], [ %rd85 + 0 ], 0x10, %r78; + // end inline asm + add.s32 %r94, %r168, 67584; + // begin inline asm + cp.async.cg.shared.global [ %r94 + 0 ], [ %rd86 + 0 ], 0x10, %r80; + // end inline asm + add.s32 %r95, %r168, 69632; + // begin inline asm + cp.async.cg.shared.global [ %r95 + 0 ], [ %rd87 + 0 ], 0x10, %r82; + // end inline asm + add.s32 %r96, %r168, 71680; + // begin inline asm + cp.async.cg.shared.global [ %r96 + 0 ], [ %rd88 + 0 ], 0x10, %r84; + // end inline asm + cp.async.commit_group; + .loc 1 588 22 // chunk_delta_h.py:588:22 + add.s32 %r97, %r73, 81920; + // begin inline asm + cp.async.cg.shared.global [ %r97 + 0 ], [ %rd89 + 0 ], 0x10, %r78; + // end inline asm + add.s32 %r98, %r73, 83968; + // begin inline asm + cp.async.cg.shared.global [ %r98 + 0 ], [ %rd90 + 0 ], 0x10, %r80; + // end inline asm + add.s32 %r99, %r73, 86016; + // begin inline asm + cp.async.cg.shared.global [ %r99 + 0 ], [ %rd91 + 0 ], 0x10, %r82; + // end inline asm + add.s32 %r100, %r73, 88064; + // begin inline asm + cp.async.cg.shared.global [ %r100 + 0 ], [ %rd92 + 0 ], 0x10, %r84; + // end inline asm + cp.async.commit_group; + .loc 1 602 38 // chunk_delta_h.py:602:38 + add.s64 %rd93, %rd89, 128; + add.s64 %rd94, %rd90, 128; + add.s64 %rd95, %rd91, 128; + add.s64 %rd96, %rd92, 128; + .loc 1 603 38 // chunk_delta_h.py:603:38 + add.s64 %rd97, %rd85, 128; + add.s64 %rd98, %rd86, 128; + add.s64 %rd99, %rd87, 128; + add.s64 %rd100, %rd88, 128; + .loc 1 604 26 // chunk_delta_h.py:604:26 + add.s32 %r101, %r73, 98304; + // begin inline asm + cp.async.cg.shared.global [ %r101 + 0 ], [ %rd93 + 0 ], 0x10, %r78; + // end inline asm + add.s32 %r102, %r73, 100352; + // begin inline asm + cp.async.cg.shared.global [ %r102 + 0 ], [ %rd94 + 0 ], 0x10, %r80; + // end inline asm + add.s32 %r103, %r73, 102400; + // begin inline asm + cp.async.cg.shared.global [ %r103 + 0 ], [ %rd95 + 0 ], 0x10, %r82; + // end inline asm + add.s32 %r104, %r73, 104448; + // begin inline asm + cp.async.cg.shared.global [ %r104 + 0 ], [ %rd96 + 0 ], 0x10, %r84; + // end inline asm + cp.async.commit_group; + .loc 1 605 26 // chunk_delta_h.py:605:26 + add.s32 %r105, %r168, 114688; + // begin inline asm + cp.async.cg.shared.global [ %r105 + 0 ], [ %rd97 + 0 ], 0x10, %r78; + // end inline asm + add.s32 %r106, %r168, 116736; + // begin inline asm + cp.async.cg.shared.global [ %r106 + 0 ], [ %rd98 + 0 ], 0x10, %r80; + // end inline asm + add.s32 %r107, %r168, 118784; + // begin inline asm + cp.async.cg.shared.global [ %r107 + 0 ], [ %rd99 + 0 ], 0x10, %r82; + // end inline asm + add.s32 %r108, %r168, 120832; + // begin inline asm + cp.async.cg.shared.global [ %r108 + 0 ], [ %rd100 + 0 ], 0x10, %r84; + // end inline asm + cp.async.commit_group; + .loc 1 489 33 // chunk_delta_h.py:489:33 + setp.gt.s32 %p11, %r151, 127; + .loc 1 490 27 // chunk_delta_h.py:490:27 + mad.wide.s32 %rd187, %r3, 64, -128; + .loc 1 491 31 // chunk_delta_h.py:491:31 + or.b64 %rd188, %rd187, %rd6; + or.b64 %rd189, %rd187, %rd149; + or.b64 %rd190, %rd187, %rd150; + or.b64 %rd191, %rd187, %rd151; + or.b64 %rd192, %rd187, %rd5; + .loc 1 492 20 // chunk_delta_h.py:492:20 + setp.lt.s64 %p65, %rd188, %rd3; + setp.lt.s64 %p64, %rd189, %rd3; + setp.lt.s64 %p63, %rd190, %rd3; + setp.lt.s64 %p62, %rd191, %rd3; + setp.lt.s64 %p12, %rd192, %rd3; + .loc 1 525 45 // chunk_delta_h.py:525:45 + min.s64 %rd193, %rd158, %rd3; + .loc 1 527 41 // chunk_delta_h.py:527:41 + add.s64 %rd194, %rd7, %rd193; + .loc 1 527 0 // chunk_delta_h.py:527 + shl.b64 %rd195, %rd194, 7; + add.s64 %rd101, %rd9, %rd195; + .loc 1 527 30 // chunk_delta_h.py:527:30 + bar.sync 0; + selp.b32 %r110, 4, 0, %p11; + add.s32 %r109, %r67, 156676; + // begin inline asm + @%p6 cp.async.ca.shared.global [ %r109 + 0 ], [ %rd101 + 0 ], 0x4, %r110; + // end inline asm + cp.async.commit_group; + .loc 1 528 39 // chunk_delta_h.py:528:39 + shl.b64 %rd196, %rd192, 7; + add.s64 %rd102, %rd154, %rd196; + .loc 1 529 26 // chunk_delta_h.py:529:26 + add.s32 %r111, %r13, 155904; + selp.b32 %r169, 4, 0, %p12; + selp.b32 %r113, %r169, 0, %p11; + // begin inline asm + @%p7 cp.async.ca.shared.global [ %r111 + 0 ], [ %rd102 + 0 ], 0x4, %r113; + // end inline asm + cp.async.commit_group; + add.s32 %r112, %r13, 156416; + // begin inline asm + @%p7 cp.async.ca.shared.global [ %r112 + 0 ], [ %rd102 + 0 ], 0x4, %r113; + // end inline asm + cp.async.commit_group; + .loc 1 532 20 // chunk_delta_h.py:532:20 + shl.b64 %rd197, %rd188, 13; + add.s64 %rd198, %rd144, %rd197; + shl.b64 %rd199, %rd189, 13; + add.s64 %rd200, %rd144, %rd199; + shl.b64 %rd201, %rd190, 13; + add.s64 %rd202, %rd144, %rd201; + shl.b64 %rd203, %rd191, 13; + add.s64 %rd204, %rd144, %rd203; + .loc 1 532 44 // chunk_delta_h.py:532:44 + add.s64 %rd115, %rd198, %rd178; + add.s64 %rd116, %rd200, %rd178; + add.s64 %rd117, %rd202, %rd178; + add.s64 %rd118, %rd204, %rd178; + .loc 1 534 20 // chunk_delta_h.py:534:20 + add.s64 %rd205, %rd143, %rd197; + add.s64 %rd206, %rd143, %rd199; + add.s64 %rd207, %rd143, %rd201; + add.s64 %rd208, %rd143, %rd203; + .loc 1 534 44 // chunk_delta_h.py:534:44 + add.s64 %rd103, %rd205, %rd178; + add.s64 %rd104, %rd206, %rd178; + add.s64 %rd105, %rd207, %rd178; + add.s64 %rd106, %rd208, %rd178; + .loc 1 536 23 // chunk_delta_h.py:536:23 + add.s32 %r114, %r73, 8192; + selp.b32 %r170, 16, 0, %p65; + selp.b32 %r119, %r170, 0, %p11; + // begin inline asm + cp.async.cg.shared.global [ %r114 + 0 ], [ %rd103 + 0 ], 0x10, %r119; + // end inline asm + add.s32 %r115, %r73, 10240; + selp.b32 %r171, 16, 0, %p64; + selp.b32 %r121, %r171, 0, %p11; + // begin inline asm + cp.async.cg.shared.global [ %r115 + 0 ], [ %rd104 + 0 ], 0x10, %r121; + // end inline asm + add.s32 %r116, %r73, 12288; + selp.b32 %r172, 16, 0, %p63; + selp.b32 %r123, %r172, 0, %p11; + // begin inline asm + cp.async.cg.shared.global [ %r116 + 0 ], [ %rd105 + 0 ], 0x10, %r123; + // end inline asm + add.s32 %r117, %r73, 14336; + selp.b32 %r173, 16, 0, %p62; + selp.b32 %r125, %r173, 0, %p11; + // begin inline asm + cp.async.cg.shared.global [ %r117 + 0 ], [ %rd106 + 0 ], 0x10, %r125; + // end inline asm + cp.async.commit_group; + .loc 1 539 18 // chunk_delta_h.py:539:18 + add.s64 %rd209, %rd141, %rd197; + add.s64 %rd210, %rd141, %rd199; + add.s64 %rd211, %rd141, %rd201; + add.s64 %rd212, %rd141, %rd203; + .loc 1 539 41 // chunk_delta_h.py:539:41 + add.s64 %rd107, %rd209, %rd155; + add.s64 %rd108, %rd210, %rd155; + add.s64 %rd109, %rd211, %rd155; + add.s64 %rd110, %rd212, %rd155; + .loc 1 540 22 // chunk_delta_h.py:540:22 + add.s32 %r118, %r168, 24576; + // begin inline asm + cp.async.cg.shared.global [ %r118 + 0 ], [ %rd107 + 0 ], 0x10, %r119; + // end inline asm + add.s32 %r120, %r168, 26624; + // begin inline asm + cp.async.cg.shared.global [ %r120 + 0 ], [ %rd108 + 0 ], 0x10, %r121; + // end inline asm + add.s32 %r122, %r168, 28672; + // begin inline asm + cp.async.cg.shared.global [ %r122 + 0 ], [ %rd109 + 0 ], 0x10, %r123; + // end inline asm + add.s32 %r124, %r168, 30720; + // begin inline asm + cp.async.cg.shared.global [ %r124 + 0 ], [ %rd110 + 0 ], 0x10, %r125; + // end inline asm + cp.async.commit_group; + .loc 1 550 45 // chunk_delta_h.py:550:45 + add.s64 %rd111, %rd107, 128; + add.s64 %rd112, %rd108, 128; + add.s64 %rd113, %rd109, 128; + add.s64 %rd114, %rd110, 128; + .loc 1 551 26 // chunk_delta_h.py:551:26 + add.s32 %r126, %r168, 40960; + // begin inline asm + cp.async.cg.shared.global [ %r126 + 0 ], [ %rd111 + 0 ], 0x10, %r119; + // end inline asm + add.s32 %r127, %r168, 43008; + // begin inline asm + cp.async.cg.shared.global [ %r127 + 0 ], [ %rd112 + 0 ], 0x10, %r121; + // end inline asm + add.s32 %r128, %r168, 45056; + // begin inline asm + cp.async.cg.shared.global [ %r128 + 0 ], [ %rd113 + 0 ], 0x10, %r123; + // end inline asm + add.s32 %r129, %r168, 47104; + // begin inline asm + cp.async.cg.shared.global [ %r129 + 0 ], [ %rd114 + 0 ], 0x10, %r125; + // end inline asm + cp.async.commit_group; + .loc 1 581 24 // chunk_delta_h.py:581:24 + add.s32 %r130, %r73, 57344; + // begin inline asm + cp.async.cg.shared.global [ %r130 + 0 ], [ %rd115 + 0 ], 0x10, %r119; + // end inline asm + add.s32 %r131, %r73, 59392; + // begin inline asm + cp.async.cg.shared.global [ %r131 + 0 ], [ %rd116 + 0 ], 0x10, %r121; + // end inline asm + add.s32 %r132, %r73, 61440; + // begin inline asm + cp.async.cg.shared.global [ %r132 + 0 ], [ %rd117 + 0 ], 0x10, %r123; + // end inline asm + add.s32 %r133, %r73, 63488; + // begin inline asm + cp.async.cg.shared.global [ %r133 + 0 ], [ %rd118 + 0 ], 0x10, %r125; + // end inline asm + cp.async.commit_group; + .loc 1 585 34 // chunk_delta_h.py:585:34 + add.s64 %rd119, %rd156, %rd197; + add.s64 %rd120, %rd156, %rd199; + add.s64 %rd121, %rd156, %rd201; + add.s64 %rd122, %rd156, %rd203; + .loc 1 586 34 // chunk_delta_h.py:586:34 + add.s64 %rd123, %rd157, %rd197; + add.s64 %rd124, %rd157, %rd199; + add.s64 %rd125, %rd157, %rd201; + add.s64 %rd126, %rd157, %rd203; + .loc 1 587 22 // chunk_delta_h.py:587:22 + add.s32 %r134, %r168, 73728; + // begin inline asm + cp.async.cg.shared.global [ %r134 + 0 ], [ %rd119 + 0 ], 0x10, %r119; + // end inline asm + add.s32 %r135, %r168, 75776; + // begin inline asm + cp.async.cg.shared.global [ %r135 + 0 ], [ %rd120 + 0 ], 0x10, %r121; + // end inline asm + add.s32 %r136, %r168, 77824; + // begin inline asm + cp.async.cg.shared.global [ %r136 + 0 ], [ %rd121 + 0 ], 0x10, %r123; + // end inline asm + add.s32 %r137, %r168, 79872; + // begin inline asm + cp.async.cg.shared.global [ %r137 + 0 ], [ %rd122 + 0 ], 0x10, %r125; + // end inline asm + cp.async.commit_group; + .loc 1 588 22 // chunk_delta_h.py:588:22 + add.s32 %r138, %r73, 90112; + // begin inline asm + cp.async.cg.shared.global [ %r138 + 0 ], [ %rd123 + 0 ], 0x10, %r119; + // end inline asm + add.s32 %r139, %r73, 92160; + // begin inline asm + cp.async.cg.shared.global [ %r139 + 0 ], [ %rd124 + 0 ], 0x10, %r121; + // end inline asm + add.s32 %r140, %r73, 94208; + // begin inline asm + cp.async.cg.shared.global [ %r140 + 0 ], [ %rd125 + 0 ], 0x10, %r123; + // end inline asm + add.s32 %r141, %r73, 96256; + // begin inline asm + cp.async.cg.shared.global [ %r141 + 0 ], [ %rd126 + 0 ], 0x10, %r125; + // end inline asm + cp.async.commit_group; + .loc 1 602 38 // chunk_delta_h.py:602:38 + add.s64 %rd127, %rd123, 128; + add.s64 %rd128, %rd124, 128; + add.s64 %rd129, %rd125, 128; + add.s64 %rd130, %rd126, 128; + .loc 1 603 38 // chunk_delta_h.py:603:38 + add.s64 %rd131, %rd119, 128; + add.s64 %rd132, %rd120, 128; + add.s64 %rd133, %rd121, 128; + add.s64 %rd134, %rd122, 128; + .loc 1 604 26 // chunk_delta_h.py:604:26 + add.s32 %r142, %r73, 106496; + // begin inline asm + cp.async.cg.shared.global [ %r142 + 0 ], [ %rd127 + 0 ], 0x10, %r119; + // end inline asm + add.s32 %r143, %r73, 108544; + // begin inline asm + cp.async.cg.shared.global [ %r143 + 0 ], [ %rd128 + 0 ], 0x10, %r121; + // end inline asm + add.s32 %r144, %r73, 110592; + // begin inline asm + cp.async.cg.shared.global [ %r144 + 0 ], [ %rd129 + 0 ], 0x10, %r123; + // end inline asm + add.s32 %r145, %r73, 112640; + // begin inline asm + cp.async.cg.shared.global [ %r145 + 0 ], [ %rd130 + 0 ], 0x10, %r125; + // end inline asm + cp.async.commit_group; + .loc 1 605 26 // chunk_delta_h.py:605:26 + add.s32 %r146, %r168, 122880; + // begin inline asm + cp.async.cg.shared.global [ %r146 + 0 ], [ %rd131 + 0 ], 0x10, %r119; + // end inline asm + add.s32 %r147, %r168, 124928; + // begin inline asm + cp.async.cg.shared.global [ %r147 + 0 ], [ %rd132 + 0 ], 0x10, %r121; + // end inline asm + add.s32 %r148, %r168, 126976; + // begin inline asm + cp.async.cg.shared.global [ %r148 + 0 ], [ %rd133 + 0 ], 0x10, %r123; + // end inline asm + add.s32 %r149, %r168, 129024; + // begin inline asm + cp.async.cg.shared.global [ %r149 + 0 ], [ %rd134 + 0 ], 0x10, %r125; + // end inline asm + cp.async.commit_group; + .loc 1 489 33 // chunk_delta_h.py:489:33 + @%p8 bra $L__BB0_15; +// %bb.1: // %.lr.ph + .loc 1 0 33 // chunk_delta_h.py:0:33 + ld.param.b32 %r66, [chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64_param_8]; + ld.param.b64 %rd65, [chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64_param_5]; + ld.param.b64 %rd135, [chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64_param_7]; + add.s64 %rd145, %rd135, %rd139; + shr.u32 %r4, %r1, 5; + and.b32 %r6, %r1, 120; + and.b32 %r7, %r1, 15; + and.b32 %r8, %r1, 96; + shr.u32 %r9, %r8, 1; + add.s32 %r476, %r1171, 1048576; + add.s32 %r766, %r1171, 64; + add.s32 %r772, %r1171, 1048640; + add.s32 %r843, %r1171, 128; + add.s32 %r851, %r1171, 1048704; + shl.b64 %rd538, %rd159, 12; + shl.b64 %rd537, %rd160, 12; + shl.b64 %rd536, %rd161, 12; + shl.b64 %rd535, %rd162, 12; + shl.b64 %rd534, %rd188, 12; + shl.b64 %rd533, %rd189, 12; + shl.b64 %rd532, %rd190, 12; + shl.b64 %rd531, %rd191, 12; + add.s64 %rd8, %rd145, %rd178; + .loc 1 411 20 // chunk_delta_h.py:411:20 + cvt.u64.u32 %rd213, %r3; + .loc 1 489 33 // chunk_delta_h.py:489:33 + mul.lo.s64 %rd214, %rd1, %rd213; + add.s32 %r174, %r3, -3; + shl.b64 %rd215, %rd214, 20; + shl.b64 %rd216, %rd2, 15; + shl.b32 %r175, %r7, 7; + shl.b32 %r176, %r8, 6; + shl.b32 %r177, %r5, 4; + and.b32 %r178, %r1, 16; + shl.b32 %r179, %r178, 2; + or.b32 %r180, %r176, %r177; + xor.b32 %r181, %r180, %r179; + or.b32 %r182, %r181, %r175; + add.s32 %r183, %r67, %r182; + add.s32 %r238, %r183, 131072; + xor.b32 %r184, %r182, 16; + add.s32 %r185, %r67, %r184; + add.s32 %r239, %r185, 131072; + xor.b32 %r186, %r182, 32; + add.s32 %r187, %r67, %r186; + add.s32 %r240, %r187, 131072; + xor.b32 %r188, %r182, 48; + add.s32 %r189, %r67, %r188; + add.s32 %r241, %r189, 131072; + and.b32 %r190, %r14, 176; + shl.b32 %r191, %r8, 3; + bfe.s32 %r192, %r1, 2, 1; + and.b32 %r193, %r192, 1088; + or.b32 %r194, %r190, %r191; + xor.b32 %r195, %r193, %r179; + or.b32 %r196, %r195, %r194; + add.s32 %r197, %r67, %r196; + add.s32 %r242, %r197, 139264; + shl.b32 %r198, %r16, 5; + shl.b32 %r199, %r1, 3; + and.b32 %r200, %r199, 48; + shl.b32 %r201, %r8, 1; + shl.b32 %r202, %r1, 6; + and.b32 %r203, %r202, 64; + or.b32 %r204, %r198, %r200; + xor.b32 %r205, %r201, %r203; + or.b32 %r206, %r204, %r205; + add.s32 %r207, %r67, %r206; + add.s32 %r18, %r207, 139264; + add.s32 %r275, %r183, 139264; + add.s32 %r276, %r185, 139264; + add.s32 %r277, %r187, 139264; + add.s32 %r278, %r189, 139264; + add.s32 %r279, %r197, 147456; + add.s32 %r19, %r207, 147456; + and.b32 %r208, %r14, 48; + shl.b32 %r209, %r4, 6; + or.b32 %r210, %r209, %r199; + and.b32 %r211, %r210, 192; + add.s32 %r212, %r67, %r208; + add.s32 %r20, %r212, %r211; + shl.b32 %r213, %r178, 3; + add.s32 %r21, %r67, %r213; + add.s32 %r855, %r67, 131072; + bfe.u32 %r214, %r855, 4, 14; + cvt.u64.u32 %rd217, %r214; + or.b64 %rd326, %rd217, 4611756662049472512; + add.s64 %rd257, %rd217, 4611756662049472640; + add.s64 %rd259, %rd217, 4611756662049472768; + add.s64 %rd261, %rd217, 4611756662049472896; + add.s32 %r215, %r67, 139264; + bfe.u32 %r216, %r215, 4, 14; + cvt.u64.u32 %rd218, %r216; + or.b64 %rd264, %rd218, 4611756662049472512; + add.s64 %rd266, %rd218, 4611756662049472640; + add.s64 %rd268, %rd218, 4611756662049472768; + add.s64 %rd270, %rd218, 4611756662049472896; + add.s32 %r217, %r67, %r175; + add.s32 %r218, %r217, %r176; + add.s32 %r22, %r218, %r179; + add.s32 %r219, %r217, %r181; + add.s32 %r514, %r219, 147456; + add.s32 %r515, %r185, 147456; + add.s32 %r516, %r187, 147456; + add.s32 %r517, %r189, 147456; + add.s32 %r518, %r197, 131072; + add.s32 %r23, %r207, 131072; + shl.b32 %r220, %r7, 1; + shl.b32 %r221, %r178, 8; + or.b32 %r222, %r220, %r221; + or.b32 %r24, %r222, %r8; + shl.b32 %r223, %r5, 10; + shr.u32 %r224, %r6, 1; + or.b32 %r225, %r223, %r224; + add.s32 %r226, %r67, %r225; + add.s32 %r551, %r226, 131072; + add.s32 %r553, %r226, 131648; + add.s32 %r555, %r226, 131136; + add.s32 %r557, %r226, 131584; + add.s32 %r559, %r226, 139264; + add.s32 %r561, %r226, 139840; + add.s32 %r563, %r226, 139328; + add.s32 %r565, %r226, 139776; + xor.b32 %r227, %r225, 16; + add.s32 %r228, %r67, %r227; + add.s32 %r567, %r228, 131200; + add.s32 %r569, %r228, 131776; + add.s32 %r571, %r228, 131264; + add.s32 %r573, %r228, 131712; + add.s32 %r575, %r228, 139392; + add.s32 %r577, %r228, 139968; + add.s32 %r579, %r228, 139456; + add.s32 %r581, %r228, 139904; + xor.b32 %r229, %r225, 32; + add.s32 %r230, %r67, %r229; + add.s32 %r583, %r230, 131328; + add.s32 %r585, %r230, 131904; + add.s32 %r587, %r230, 131392; + add.s32 %r589, %r230, 131840; + add.s32 %r591, %r230, 139520; + add.s32 %r593, %r230, 140096; + add.s32 %r595, %r230, 139584; + add.s32 %r597, %r230, 140032; + xor.b32 %r231, %r225, 48; + add.s32 %r232, %r67, %r231; + add.s32 %r599, %r232, 131456; + add.s32 %r601, %r232, 132032; + add.s32 %r603, %r232, 131520; + add.s32 %r605, %r232, 131968; + add.s32 %r607, %r232, 139648; + add.s32 %r609, %r232, 140224; + add.s32 %r611, %r232, 139712; + add.s32 %r613, %r232, 140160; + add.s32 %r844, %r1171, 192; + add.s64 %rd327, %rd217, 4611756662049472514; + add.s64 %rd328, %rd217, 4611756662049472516; + add.s64 %rd329, %rd217, 4611756662049472518; + add.s64 %rd330, %rd217, 4611756662049473024; + add.s64 %rd331, %rd217, 4611756662049473026; + add.s64 %rd332, %rd217, 4611756662049473028; + add.s64 %rd333, %rd217, 4611756662049473030; + add.s32 %r437, %r67, 147456; + bfe.u32 %r233, %r437, 4, 14; + cvt.u64.u32 %rd219, %r233; + or.b64 %rd335, %rd219, 4611756662049472512; + add.s64 %rd337, %rd219, 4611756662049472640; + add.s64 %rd339, %rd219, 4611756662049472768; + add.s64 %rd341, %rd219, 4611756662049472896; + cvt.s64.s32 %rd10, %r174; + max.s32 %r234, %r10, 0; + .loc 1 617 73 // chunk_delta_h.py:617:73 + mov.b64 %rd11, {%r66, %r66}; + .loc 1 489 33 // chunk_delta_h.py:489:33 + mad.wide.u32 %rd12, %r234, 64, 64; + mul.wide.s32 %rd13, %r3, 64; + add.s64 %rd220, %rd4, %rd5; + shl.b64 %rd221, %rd220, 7; + mad.wide.s32 %rd222, %r3, 8192, %rd221; + or.b64 %rd223, %rd222, %rd168; + add.s64 %rd224, %rd223, %rd63; + add.s64 %rd528, %rd224, -24576; + or.b64 %rd225, %rd13, %rd5; + add.s64 %rd14, %rd225, -192; + mul.wide.u32 %rd15, %r5, 16; + add.s64 %rd226, %rd4, %rd6; + shl.b64 %rd227, %rd226, 13; + mad.wide.s32 %rd228, %r3, 524288, %rd227; + shl.b64 %rd229, %rd2, 8; + or.b64 %rd230, %rd228, %rd229; + add.s64 %rd527, %rd60, %rd230; + add.s64 %rd526, %rd62, %rd230; + add.s64 %rd525, %rd61, %rd230; + and.b32 %r235, %r2, 1; + mul.wide.u32 %rd231, %r235, 128; + or.b64 %rd232, %rd230, %rd231; + add.s64 %rd524, %rd64, %rd232; + add.s64 %rd523, %rd66, %rd232; + mul.wide.s32 %rd233, %r3, 262144; + shl.b64 %rd234, %rd6, 12; + or.b64 %rd522, %rd233, %rd234; + or.b64 %rd16, %rd13, %rd6; + add.s32 %r236, %r9, %r7; + cvt.u64.u32 %rd235, %r236; + add.s64 %rd236, %rd13, %rd235; + add.s64 %rd17, %rd236, -64; + mad.wide.s32 %rd237, %r3, 1048576, %rd215; + or.b64 %rd238, %rd237, %rd216; + shl.b64 %rd239, %rd6, 8; + or.b64 %rd240, %rd238, %rd239; + or.b64 %rd241, %rd240, %rd231; + add.s64 %rd521, %rd65, %rd241; + mov.b32 %r237, 0f00000000; + mov.b64 %rd540, {%r237, %r237}; + mov.b32 %r1173, 1; + mov.b32 %r1172, -1; + mov.b64 %rd530, -1; + mov.b64 %rd529, 0; + mov.b64 %rd539, %rd540; + mov.b64 %rd541, %rd540; + mov.b64 %rd542, %rd540; + mov.b64 %rd543, %rd540; + mov.b64 %rd544, %rd540; + mov.b64 %rd545, %rd540; + mov.b64 %rd546, %rd540; + mov.b64 %rd547, %rd540; + mov.b64 %rd548, %rd540; + mov.b64 %rd549, %rd540; + mov.b64 %rd550, %rd540; + mov.b64 %rd551, %rd540; + mov.b64 %rd552, %rd540; + mov.b64 %rd553, %rd540; + mov.b64 %rd554, %rd540; + mov.b64 %rd555, %rd540; + mov.b64 %rd556, %rd540; + mov.b64 %rd557, %rd540; + mov.b64 %rd558, %rd540; + mov.b64 %rd559, %rd540; + mov.b64 %rd560, %rd540; + mov.b64 %rd561, %rd540; + mov.b64 %rd562, %rd540; + mov.b64 %rd563, %rd540; + mov.b64 %rd564, %rd540; + mov.b64 %rd565, %rd540; + mov.b64 %rd566, %rd540; + mov.b64 %rd567, %rd540; + mov.b64 %rd568, %rd540; + mov.b64 %rd569, %rd540; + mov.b64 %rd570, %rd540; + bra.uni $L__BB0_2; +$L__BB0_14: // in Loop: Header=BB0_2 Depth=1 + .loc 1 489 33 // chunk_delta_h.py:489:33 + setp.lt.s64 %p57, %rd530, %rd10; + .loc 1 617 93 // chunk_delta_h.py:617:93 + bar.sync 0; + // begin inline asm + +{ + .reg .pred complete; + waitLoop: + mbarrier.try_wait.parity.shared::cta.b64 complete, [%r855], %r848; + @!complete bra.uni waitLoop; +} + + // end inline asm + bar.sync 0; + elect.sync %r1029|%p58, -1; + and.pred %p56, %p14, %p58; + // begin inline asm + @%p56 mbarrier.inval.shared::cta.b64 [%r855]; + // end inline asm + .loc 1 489 33 // chunk_delta_h.py:489:33 + add.s32 %r1030, %r1173, 1; + setp.gt.s32 %p59, %r1030, 1; + selp.b32 %r1173, 0, %r1030, %p59; + .loc 1 491 31 // chunk_delta_h.py:491:31 + add.s64 %rd376, %rd16, %rd529; + add.s64 %rd377, %rd376, -192; + add.s64 %rd378, %rd376, -176; + add.s64 %rd379, %rd376, -160; + add.s64 %rd380, %rd376, -144; + .loc 1 492 20 // chunk_delta_h.py:492:20 + add.s64 %rd381, %rd14, %rd529; + setp.lt.s64 %p65, %rd377, %rd3; + setp.lt.s64 %p64, %rd378, %rd3; + setp.lt.s64 %p63, %rd379, %rd3; + setp.lt.s64 %p62, %rd380, %rd3; + setp.lt.s64 %p60, %rd381, %rd3; + .loc 1 525 45 // chunk_delta_h.py:525:45 + add.s64 %rd382, %rd13, %rd529; + add.s64 %rd383, %rd382, -128; + min.s64 %rd384, %rd383, %rd3; + .loc 1 527 41 // chunk_delta_h.py:527:41 + add.s64 %rd385, %rd7, %rd384; + .loc 1 527 0 // chunk_delta_h.py:527 + shl.b64 %rd386, %rd385, 7; + add.s64 %rd343, %rd9, %rd386; + .loc 1 527 30 // chunk_delta_h.py:527:30 + shl.b32 %r1031, %r1173, 2; + add.s32 %r1032, %r67, %r1031; + add.s32 %r856, %r1032, 156672; + selp.b32 %r857, 4, 0, %p57; + // begin inline asm + @%p6 cp.async.ca.shared.global [ %r856 + 0 ], [ %rd343 + 0 ], 0x4, %r857; + // end inline asm + .loc 1 600 47 // chunk_delta_h.py:600:47 + add.s32 %r890, %r471, 64; + // begin inline asm + tcgen05.ld.sync.aligned.16x32bx2.x32.b32 {%r858, %r859, %r860, %r861, %r862, %r863, %r864, %r865, %r866, %r867, %r868, %r869, %r870, %r871, %r872, %r873, %r874, %r875, %r876, %r877, %r878, %r879, %r880, %r881, %r882, %r883, %r884, %r885, %r886, %r887, %r888, %r889}, [%r890 + 0], 32; + // end inline asm + tcgen05.wait::ld.sync.aligned; + .loc 1 600 89 // chunk_delta_h.py:600:89 + add.s32 %r923, %r471, 1048640; + // begin inline asm + tcgen05.ld.sync.aligned.16x32bx2.x32.b32 {%r891, %r892, %r893, %r894, %r895, %r896, %r897, %r898, %r899, %r900, %r901, %r902, %r903, %r904, %r905, %r906, %r907, %r908, %r909, %r910, %r911, %r912, %r913, %r914, %r915, %r916, %r917, %r918, %r919, %r920, %r921, %r922}, [%r923 + 0], 32; + // end inline asm + tcgen05.wait::ld.sync.aligned; + .loc 1 617 51 // chunk_delta_h.py:617:51 + add.s32 %r956, %r471, 128; + // begin inline asm + tcgen05.ld.sync.aligned.16x32bx2.x32.b32 {%r924, %r925, %r926, %r927, %r928, %r929, %r930, %r931, %r932, %r933, %r934, %r935, %r936, %r937, %r938, %r939, %r940, %r941, %r942, %r943, %r944, %r945, %r946, %r947, %r948, %r949, %r950, %r951, %r952, %r953, %r954, %r955}, [%r956 + 0], 32; + // end inline asm + tcgen05.wait::ld.sync.aligned; + .loc 1 617 93 // chunk_delta_h.py:617:93 + add.s32 %r989, %r471, 1048704; + // begin inline asm + tcgen05.ld.sync.aligned.16x32bx2.x32.b32 {%r957, %r958, %r959, %r960, %r961, %r962, %r963, %r964, %r965, %r966, %r967, %r968, %r969, %r970, %r971, %r972, %r973, %r974, %r975, %r976, %r977, %r978, %r979, %r980, %r981, %r982, %r983, %r984, %r985, %r986, %r987, %r988}, [%r989 + 0], 32; + // end inline asm + tcgen05.wait::ld.sync.aligned; + .loc 1 617 51 // chunk_delta_h.py:617:51 + mov.b64 %rd387, {%r858, %r859}; + mov.b64 %rd388, {%r860, %r861}; + mov.b64 %rd389, {%r862, %r863}; + mov.b64 %rd390, {%r954, %r955}; + mov.b64 %rd391, {%r952, %r953}; + mov.b64 %rd392, {%r950, %r951}; + mov.b64 %rd393, {%r948, %r949}; + mov.b64 %rd394, {%r946, %r947}; + mov.b64 %rd395, {%r944, %r945}; + mov.b64 %rd396, {%r942, %r943}; + mov.b64 %rd397, {%r940, %r941}; + mov.b64 %rd398, {%r938, %r939}; + mov.b64 %rd399, {%r936, %r937}; + mov.b64 %rd400, {%r934, %r935}; + mov.b64 %rd401, {%r932, %r933}; + mov.b64 %rd402, {%r930, %r931}; + mov.b64 %rd403, {%r928, %r929}; + mov.b64 %rd404, {%r926, %r927}; + mov.b64 %rd405, {%r924, %r925}; + mov.b64 %rd406, {%r864, %r865}; + mov.b64 %rd407, {%r866, %r867}; + mov.b64 %rd408, {%r868, %r869}; + mov.b64 %rd409, {%r870, %r871}; + mov.b64 %rd410, {%r872, %r873}; + mov.b64 %rd411, {%r874, %r875}; + mov.b64 %rd412, {%r876, %r877}; + mov.b64 %rd413, {%r878, %r879}; + mov.b64 %rd414, {%r880, %r881}; + mov.b64 %rd415, {%r882, %r883}; + mov.b64 %rd416, {%r884, %r885}; + mov.b64 %rd417, {%r886, %r887}; + mov.b64 %rd418, {%r888, %r889}; + .loc 1 617 93 // chunk_delta_h.py:617:93 + mov.b64 %rd419, {%r891, %r892}; + mov.b64 %rd420, {%r893, %r894}; + mov.b64 %rd421, {%r895, %r896}; + mov.b64 %rd422, {%r987, %r988}; + mov.b64 %rd423, {%r985, %r986}; + mov.b64 %rd424, {%r983, %r984}; + mov.b64 %rd425, {%r981, %r982}; + mov.b64 %rd426, {%r979, %r980}; + mov.b64 %rd427, {%r977, %r978}; + mov.b64 %rd428, {%r975, %r976}; + mov.b64 %rd429, {%r973, %r974}; + mov.b64 %rd430, {%r971, %r972}; + mov.b64 %rd431, {%r969, %r970}; + mov.b64 %rd432, {%r967, %r968}; + mov.b64 %rd433, {%r965, %r966}; + mov.b64 %rd434, {%r963, %r964}; + mov.b64 %rd435, {%r961, %r962}; + mov.b64 %rd436, {%r959, %r960}; + mov.b64 %rd437, {%r957, %r958}; + mov.b64 %rd438, {%r897, %r898}; + mov.b64 %rd439, {%r899, %r900}; + mov.b64 %rd440, {%r901, %r902}; + mov.b64 %rd441, {%r903, %r904}; + mov.b64 %rd442, {%r905, %r906}; + mov.b64 %rd443, {%r907, %r908}; + mov.b64 %rd444, {%r909, %r910}; + mov.b64 %rd445, {%r911, %r912}; + mov.b64 %rd446, {%r913, %r914}; + mov.b64 %rd447, {%r915, %r916}; + mov.b64 %rd448, {%r917, %r918}; + mov.b64 %rd449, {%r919, %r920}; + mov.b64 %rd450, {%r921, %r922}; + .loc 1 617 81 // chunk_delta_h.py:617:81 + mov.b64 {%r1033, %r1034}, %rd450; + neg.f32 %r1035, %r1034; + neg.f32 %r1036, %r1033; + mov.b64 %rd451, {%r1036, %r1035}; + fma.rn.f32x2 %rd452, %rd11, %rd418, %rd451; + mov.b64 {%r1037, %r1038}, %rd449; + neg.f32 %r1039, %r1038; + neg.f32 %r1040, %r1037; + mov.b64 %rd453, {%r1040, %r1039}; + fma.rn.f32x2 %rd454, %rd11, %rd417, %rd453; + mov.b64 {%r1041, %r1042}, %rd448; + neg.f32 %r1043, %r1042; + neg.f32 %r1044, %r1041; + mov.b64 %rd455, {%r1044, %r1043}; + fma.rn.f32x2 %rd456, %rd11, %rd416, %rd455; + mov.b64 {%r1045, %r1046}, %rd447; + neg.f32 %r1047, %r1046; + neg.f32 %r1048, %r1045; + mov.b64 %rd457, {%r1048, %r1047}; + fma.rn.f32x2 %rd458, %rd11, %rd415, %rd457; + mov.b64 {%r1049, %r1050}, %rd446; + neg.f32 %r1051, %r1050; + neg.f32 %r1052, %r1049; + mov.b64 %rd459, {%r1052, %r1051}; + fma.rn.f32x2 %rd460, %rd11, %rd414, %rd459; + mov.b64 {%r1053, %r1054}, %rd445; + neg.f32 %r1055, %r1054; + neg.f32 %r1056, %r1053; + mov.b64 %rd461, {%r1056, %r1055}; + fma.rn.f32x2 %rd462, %rd11, %rd413, %rd461; + mov.b64 {%r1057, %r1058}, %rd444; + neg.f32 %r1059, %r1058; + neg.f32 %r1060, %r1057; + mov.b64 %rd463, {%r1060, %r1059}; + fma.rn.f32x2 %rd464, %rd11, %rd412, %rd463; + mov.b64 {%r1061, %r1062}, %rd443; + neg.f32 %r1063, %r1062; + neg.f32 %r1064, %r1061; + mov.b64 %rd465, {%r1064, %r1063}; + fma.rn.f32x2 %rd466, %rd11, %rd411, %rd465; + mov.b64 {%r1065, %r1066}, %rd442; + neg.f32 %r1067, %r1066; + neg.f32 %r1068, %r1065; + mov.b64 %rd467, {%r1068, %r1067}; + fma.rn.f32x2 %rd468, %rd11, %rd410, %rd467; + mov.b64 {%r1069, %r1070}, %rd441; + neg.f32 %r1071, %r1070; + neg.f32 %r1072, %r1069; + mov.b64 %rd469, {%r1072, %r1071}; + fma.rn.f32x2 %rd470, %rd11, %rd409, %rd469; + mov.b64 {%r1073, %r1074}, %rd440; + neg.f32 %r1075, %r1074; + neg.f32 %r1076, %r1073; + mov.b64 %rd471, {%r1076, %r1075}; + fma.rn.f32x2 %rd472, %rd11, %rd408, %rd471; + mov.b64 {%r1077, %r1078}, %rd439; + neg.f32 %r1079, %r1078; + neg.f32 %r1080, %r1077; + mov.b64 %rd473, {%r1080, %r1079}; + fma.rn.f32x2 %rd474, %rd11, %rd407, %rd473; + mov.b64 {%r1081, %r1082}, %rd438; + neg.f32 %r1083, %r1082; + neg.f32 %r1084, %r1081; + mov.b64 %rd475, {%r1084, %r1083}; + fma.rn.f32x2 %rd476, %rd11, %rd406, %rd475; + mov.b64 {%r1085, %r1086}, %rd437; + neg.f32 %r1087, %r1086; + neg.f32 %r1088, %r1085; + mov.b64 %rd477, {%r1088, %r1087}; + fma.rn.f32x2 %rd478, %rd11, %rd405, %rd477; + mov.b64 {%r1089, %r1090}, %rd436; + neg.f32 %r1091, %r1090; + neg.f32 %r1092, %r1089; + mov.b64 %rd479, {%r1092, %r1091}; + fma.rn.f32x2 %rd480, %rd11, %rd404, %rd479; + mov.b64 {%r1093, %r1094}, %rd435; + neg.f32 %r1095, %r1094; + neg.f32 %r1096, %r1093; + mov.b64 %rd481, {%r1096, %r1095}; + fma.rn.f32x2 %rd482, %rd11, %rd403, %rd481; + mov.b64 {%r1097, %r1098}, %rd434; + neg.f32 %r1099, %r1098; + neg.f32 %r1100, %r1097; + mov.b64 %rd483, {%r1100, %r1099}; + fma.rn.f32x2 %rd484, %rd11, %rd402, %rd483; + mov.b64 {%r1101, %r1102}, %rd433; + neg.f32 %r1103, %r1102; + neg.f32 %r1104, %r1101; + mov.b64 %rd485, {%r1104, %r1103}; + fma.rn.f32x2 %rd486, %rd11, %rd401, %rd485; + mov.b64 {%r1105, %r1106}, %rd432; + neg.f32 %r1107, %r1106; + neg.f32 %r1108, %r1105; + mov.b64 %rd487, {%r1108, %r1107}; + fma.rn.f32x2 %rd488, %rd11, %rd400, %rd487; + mov.b64 {%r1109, %r1110}, %rd431; + neg.f32 %r1111, %r1110; + neg.f32 %r1112, %r1109; + mov.b64 %rd489, {%r1112, %r1111}; + fma.rn.f32x2 %rd490, %rd11, %rd399, %rd489; + mov.b64 {%r1113, %r1114}, %rd430; + neg.f32 %r1115, %r1114; + neg.f32 %r1116, %r1113; + mov.b64 %rd491, {%r1116, %r1115}; + fma.rn.f32x2 %rd492, %rd11, %rd398, %rd491; + mov.b64 {%r1117, %r1118}, %rd429; + neg.f32 %r1119, %r1118; + neg.f32 %r1120, %r1117; + mov.b64 %rd493, {%r1120, %r1119}; + fma.rn.f32x2 %rd494, %rd11, %rd397, %rd493; + mov.b64 {%r1121, %r1122}, %rd428; + neg.f32 %r1123, %r1122; + neg.f32 %r1124, %r1121; + mov.b64 %rd495, {%r1124, %r1123}; + fma.rn.f32x2 %rd496, %rd11, %rd396, %rd495; + mov.b64 {%r1125, %r1126}, %rd427; + neg.f32 %r1127, %r1126; + neg.f32 %r1128, %r1125; + mov.b64 %rd497, {%r1128, %r1127}; + fma.rn.f32x2 %rd498, %rd11, %rd395, %rd497; + mov.b64 {%r1129, %r1130}, %rd426; + neg.f32 %r1131, %r1130; + neg.f32 %r1132, %r1129; + mov.b64 %rd499, {%r1132, %r1131}; + fma.rn.f32x2 %rd500, %rd11, %rd394, %rd499; + mov.b64 {%r1133, %r1134}, %rd425; + neg.f32 %r1135, %r1134; + neg.f32 %r1136, %r1133; + mov.b64 %rd501, {%r1136, %r1135}; + fma.rn.f32x2 %rd502, %rd11, %rd393, %rd501; + mov.b64 {%r1137, %r1138}, %rd424; + neg.f32 %r1139, %r1138; + neg.f32 %r1140, %r1137; + mov.b64 %rd503, {%r1140, %r1139}; + fma.rn.f32x2 %rd504, %rd11, %rd392, %rd503; + mov.b64 {%r1141, %r1142}, %rd423; + neg.f32 %r1143, %r1142; + neg.f32 %r1144, %r1141; + mov.b64 %rd505, {%r1144, %r1143}; + fma.rn.f32x2 %rd506, %rd11, %rd391, %rd505; + mov.b64 {%r1145, %r1146}, %rd422; + neg.f32 %r1147, %r1146; + neg.f32 %r1148, %r1145; + mov.b64 %rd507, {%r1148, %r1147}; + fma.rn.f32x2 %rd508, %rd11, %rd390, %rd507; + mov.b64 {%r1149, %r1150}, %rd421; + neg.f32 %r1151, %r1150; + neg.f32 %r1152, %r1149; + mov.b64 %rd509, {%r1152, %r1151}; + fma.rn.f32x2 %rd510, %rd11, %rd389, %rd509; + mov.b64 {%r1153, %r1154}, %rd420; + neg.f32 %r1155, %r1154; + neg.f32 %r1156, %r1153; + mov.b64 %rd511, {%r1156, %r1155}; + fma.rn.f32x2 %rd512, %rd11, %rd388, %rd511; + mov.b64 {%r1157, %r1158}, %rd419; + neg.f32 %r1159, %r1158; + neg.f32 %r1160, %r1157; + mov.b64 %rd513, {%r1160, %r1159}; + fma.rn.f32x2 %rd514, %rd11, %rd387, %rd513; + .loc 1 617 25 // chunk_delta_h.py:617:25 + add.f32x2 %rd555, %rd28, %rd514; + add.f32x2 %rd556, %rd29, %rd512; + add.f32x2 %rd557, %rd30, %rd510; + add.f32x2 %rd554, %rd59, %rd508; + add.f32x2 %rd553, %rd58, %rd506; + add.f32x2 %rd552, %rd57, %rd504; + add.f32x2 %rd551, %rd56, %rd502; + add.f32x2 %rd550, %rd55, %rd500; + add.f32x2 %rd549, %rd54, %rd498; + add.f32x2 %rd548, %rd53, %rd496; + add.f32x2 %rd547, %rd52, %rd494; + add.f32x2 %rd546, %rd51, %rd492; + add.f32x2 %rd545, %rd50, %rd490; + add.f32x2 %rd544, %rd49, %rd488; + add.f32x2 %rd543, %rd48, %rd486; + add.f32x2 %rd542, %rd47, %rd484; + add.f32x2 %rd541, %rd46, %rd482; + add.f32x2 %rd540, %rd45, %rd480; + add.f32x2 %rd539, %rd44, %rd478; + add.f32x2 %rd558, %rd31, %rd476; + add.f32x2 %rd559, %rd32, %rd474; + add.f32x2 %rd560, %rd33, %rd472; + add.f32x2 %rd561, %rd34, %rd470; + add.f32x2 %rd562, %rd35, %rd468; + add.f32x2 %rd563, %rd36, %rd466; + add.f32x2 %rd564, %rd37, %rd464; + add.f32x2 %rd565, %rd38, %rd462; + add.f32x2 %rd566, %rd39, %rd460; + add.f32x2 %rd567, %rd40, %rd458; + add.f32x2 %rd568, %rd41, %rd456; + add.f32x2 %rd569, %rd42, %rd454; + add.f32x2 %rd570, %rd43, %rd452; + .loc 1 527 30 // chunk_delta_h.py:527:30 + cp.async.commit_group; + .loc 1 529 26 // chunk_delta_h.py:529:26 + shl.b32 %r1161, %r1173, 8; + add.s32 %r1162, %r13, %r1161; + add.s32 %r990, %r1162, 155648; + selp.b32 %r1163, 4, 0, %p60; + selp.b32 %r992, %r1163, 0, %p57; + // begin inline asm + @%p7 cp.async.ca.shared.global [ %r990 + 0 ], [ %rd528 + 0 ], 0x4, %r992; + // end inline asm + cp.async.commit_group; + add.s32 %r991, %r1162, 156160; + // begin inline asm + @%p7 cp.async.ca.shared.global [ %r991 + 0 ], [ %rd528 + 0 ], 0x4, %r992; + // end inline asm + cp.async.commit_group; + .loc 1 532 36 // chunk_delta_h.py:532:36 + add.s64 %rd534, %rd522, -786432; + add.s64 %rd533, %rd522, -720896; + add.s64 %rd532, %rd522, -655360; + .loc 1 532 44 // chunk_delta_h.py:532:44 + add.s64 %rd531, %rd522, -589824; + add.s64 %rd515, %rd523, %rd15; + add.s64 %rd356, %rd515, -1572864; + add.s64 %rd357, %rd515, -1441792; + add.s64 %rd358, %rd515, -1310720; + .loc 1 534 44 // chunk_delta_h.py:534:44 + add.s64 %rd359, %rd515, -1179648; + add.s64 %rd516, %rd524, %rd15; + add.s64 %rd344, %rd516, -1572864; + add.s64 %rd345, %rd516, -1441792; + add.s64 %rd346, %rd516, -1310720; + .loc 1 536 23 // chunk_delta_h.py:536:23 + add.s64 %rd347, %rd516, -1179648; + shl.b32 %r1164, %r1173, 13; + add.s32 %r1165, %r67, %r1164; + add.s32 %r993, %r1165, %r15; + selp.b32 %r1166, 16, 0, %p65; + selp.b32 %r998, %r1166, 0, %p57; + // begin inline asm + cp.async.cg.shared.global [ %r993 + 0 ], [ %rd344 + 0 ], 0x10, %r998; + // end inline asm + add.s32 %r994, %r993, 2048; + selp.b32 %r1167, 16, 0, %p64; + selp.b32 %r1000, %r1167, 0, %p57; + // begin inline asm + cp.async.cg.shared.global [ %r994 + 0 ], [ %rd345 + 0 ], 0x10, %r1000; + // end inline asm + add.s32 %r995, %r993, 4096; + selp.b32 %r1168, 16, 0, %p63; + selp.b32 %r1002, %r1168, 0, %p57; + // begin inline asm + cp.async.cg.shared.global [ %r995 + 0 ], [ %rd346 + 0 ], 0x10, %r1002; + // end inline asm + add.s32 %r996, %r993, 6144; + selp.b32 %r1169, 16, 0, %p62; + selp.b32 %r1004, %r1169, 0, %p57; + // begin inline asm + cp.async.cg.shared.global [ %r996 + 0 ], [ %rd347 + 0 ], 0x10, %r1004; + // end inline asm + cp.async.commit_group; + .loc 1 539 41 // chunk_delta_h.py:539:41 + add.s64 %rd517, %rd525, %rd15; + add.s64 %rd348, %rd517, -1572864; + add.s64 %rd349, %rd517, -1441792; + add.s64 %rd350, %rd517, -1310720; + .loc 1 540 22 // chunk_delta_h.py:540:22 + add.s64 %rd351, %rd517, -1179648; + add.s32 %r1170, %r1165, %r17; + add.s32 %r997, %r1170, 16384; + // begin inline asm + cp.async.cg.shared.global [ %r997 + 0 ], [ %rd348 + 0 ], 0x10, %r998; + // end inline asm + add.s32 %r999, %r1170, 18432; + // begin inline asm + cp.async.cg.shared.global [ %r999 + 0 ], [ %rd349 + 0 ], 0x10, %r1000; + // end inline asm + add.s32 %r1001, %r1170, 20480; + // begin inline asm + cp.async.cg.shared.global [ %r1001 + 0 ], [ %rd350 + 0 ], 0x10, %r1002; + // end inline asm + add.s32 %r1003, %r1170, 22528; + // begin inline asm + cp.async.cg.shared.global [ %r1003 + 0 ], [ %rd351 + 0 ], 0x10, %r1004; + // end inline asm + cp.async.commit_group; + .loc 1 550 45 // chunk_delta_h.py:550:45 + add.s64 %rd352, %rd517, -1572736; + add.s64 %rd353, %rd517, -1441664; + add.s64 %rd354, %rd517, -1310592; + .loc 1 551 26 // chunk_delta_h.py:551:26 + add.s64 %rd355, %rd517, -1179520; + add.s32 %r1005, %r1170, 32768; + // begin inline asm + cp.async.cg.shared.global [ %r1005 + 0 ], [ %rd352 + 0 ], 0x10, %r998; + // end inline asm + add.s32 %r1006, %r1170, 34816; + // begin inline asm + cp.async.cg.shared.global [ %r1006 + 0 ], [ %rd353 + 0 ], 0x10, %r1000; + // end inline asm + add.s32 %r1007, %r1170, 36864; + // begin inline asm + cp.async.cg.shared.global [ %r1007 + 0 ], [ %rd354 + 0 ], 0x10, %r1002; + // end inline asm + add.s32 %r1008, %r1170, 38912; + // begin inline asm + cp.async.cg.shared.global [ %r1008 + 0 ], [ %rd355 + 0 ], 0x10, %r1004; + // end inline asm + cp.async.commit_group; + .loc 1 581 24 // chunk_delta_h.py:581:24 + add.s32 %r1009, %r993, 49152; + // begin inline asm + cp.async.cg.shared.global [ %r1009 + 0 ], [ %rd356 + 0 ], 0x10, %r998; + // end inline asm + add.s32 %r1010, %r993, 51200; + // begin inline asm + cp.async.cg.shared.global [ %r1010 + 0 ], [ %rd357 + 0 ], 0x10, %r1000; + // end inline asm + add.s32 %r1011, %r993, 53248; + // begin inline asm + cp.async.cg.shared.global [ %r1011 + 0 ], [ %rd358 + 0 ], 0x10, %r1002; + // end inline asm + add.s32 %r1012, %r993, 55296; + // begin inline asm + cp.async.cg.shared.global [ %r1012 + 0 ], [ %rd359 + 0 ], 0x10, %r1004; + // end inline asm + cp.async.commit_group; + .loc 1 585 34 // chunk_delta_h.py:585:34 + add.s64 %rd518, %rd526, %rd15; + add.s64 %rd360, %rd518, -1572864; + add.s64 %rd361, %rd518, -1441792; + add.s64 %rd362, %rd518, -1310720; + .loc 1 586 34 // chunk_delta_h.py:586:34 + add.s64 %rd363, %rd518, -1179648; + add.s64 %rd519, %rd527, %rd15; + add.s64 %rd364, %rd519, -1572864; + add.s64 %rd365, %rd519, -1441792; + add.s64 %rd366, %rd519, -1310720; + .loc 1 587 22 // chunk_delta_h.py:587:22 + add.s64 %rd367, %rd519, -1179648; + add.s32 %r1013, %r1170, 65536; + // begin inline asm + cp.async.cg.shared.global [ %r1013 + 0 ], [ %rd360 + 0 ], 0x10, %r998; + // end inline asm + add.s32 %r1014, %r1170, 67584; + // begin inline asm + cp.async.cg.shared.global [ %r1014 + 0 ], [ %rd361 + 0 ], 0x10, %r1000; + // end inline asm + add.s32 %r1015, %r1170, 69632; + // begin inline asm + cp.async.cg.shared.global [ %r1015 + 0 ], [ %rd362 + 0 ], 0x10, %r1002; + // end inline asm + add.s32 %r1016, %r1170, 71680; + // begin inline asm + cp.async.cg.shared.global [ %r1016 + 0 ], [ %rd363 + 0 ], 0x10, %r1004; + // end inline asm + cp.async.commit_group; + .loc 1 588 22 // chunk_delta_h.py:588:22 + add.s32 %r1017, %r993, 81920; + // begin inline asm + cp.async.cg.shared.global [ %r1017 + 0 ], [ %rd364 + 0 ], 0x10, %r998; + // end inline asm + add.s32 %r1018, %r993, 83968; + // begin inline asm + cp.async.cg.shared.global [ %r1018 + 0 ], [ %rd365 + 0 ], 0x10, %r1000; + // end inline asm + add.s32 %r1019, %r993, 86016; + // begin inline asm + cp.async.cg.shared.global [ %r1019 + 0 ], [ %rd366 + 0 ], 0x10, %r1002; + // end inline asm + add.s32 %r1020, %r993, 88064; + // begin inline asm + cp.async.cg.shared.global [ %r1020 + 0 ], [ %rd367 + 0 ], 0x10, %r1004; + // end inline asm + cp.async.commit_group; + .loc 1 602 38 // chunk_delta_h.py:602:38 + add.s64 %rd368, %rd519, -1572736; + add.s64 %rd369, %rd519, -1441664; + add.s64 %rd370, %rd519, -1310592; + .loc 1 603 38 // chunk_delta_h.py:603:38 + add.s64 %rd371, %rd519, -1179520; + add.s64 %rd372, %rd518, -1572736; + add.s64 %rd373, %rd518, -1441664; + add.s64 %rd374, %rd518, -1310592; + .loc 1 604 26 // chunk_delta_h.py:604:26 + add.s64 %rd375, %rd518, -1179520; + add.s32 %r1021, %r993, 98304; + // begin inline asm + cp.async.cg.shared.global [ %r1021 + 0 ], [ %rd368 + 0 ], 0x10, %r998; + // end inline asm + add.s32 %r1022, %r993, 100352; + // begin inline asm + cp.async.cg.shared.global [ %r1022 + 0 ], [ %rd369 + 0 ], 0x10, %r1000; + // end inline asm + add.s32 %r1023, %r993, 102400; + // begin inline asm + cp.async.cg.shared.global [ %r1023 + 0 ], [ %rd370 + 0 ], 0x10, %r1002; + // end inline asm + add.s32 %r1024, %r993, 104448; + // begin inline asm + cp.async.cg.shared.global [ %r1024 + 0 ], [ %rd371 + 0 ], 0x10, %r1004; + // end inline asm + cp.async.commit_group; + .loc 1 605 26 // chunk_delta_h.py:605:26 + add.s32 %r1025, %r1170, 114688; + // begin inline asm + cp.async.cg.shared.global [ %r1025 + 0 ], [ %rd372 + 0 ], 0x10, %r998; + // end inline asm + add.s32 %r1026, %r1170, 116736; + // begin inline asm + cp.async.cg.shared.global [ %r1026 + 0 ], [ %rd373 + 0 ], 0x10, %r1000; + // end inline asm + add.s32 %r1027, %r1170, 118784; + // begin inline asm + cp.async.cg.shared.global [ %r1027 + 0 ], [ %rd374 + 0 ], 0x10, %r1002; + // end inline asm + add.s32 %r1028, %r1170, 120832; + // begin inline asm + cp.async.cg.shared.global [ %r1028 + 0 ], [ %rd375 + 0 ], 0x10, %r1004; + // end inline asm + cp.async.commit_group; + .loc 1 489 33 // chunk_delta_h.py:489:33 + add.s64 %rd530, %rd530, 1; + add.s64 %rd529, %rd529, -64; + add.s64 %rd520, %rd12, %rd529; + add.s64 %rd528, %rd528, -8192; + add.s64 %rd527, %rd527, -524288; + add.s64 %rd526, %rd526, -524288; + add.s64 %rd525, %rd525, -524288; + add.s64 %rd524, %rd524, -524288; + add.s64 %rd523, %rd523, -524288; + add.s64 %rd522, %rd522, -262144; + add.s64 %rd521, %rd521, -1048576; + setp.ne.b64 %p61, %rd520, 0; + mov.pred %p66, %p1; + mov.pred %p67, %p2; + mov.pred %p68, %p3; + mov.pred %p69, %p4; + mov.b64 %rd535, %rd18; + mov.b64 %rd536, %rd19; + mov.b64 %rd537, %rd20; + mov.b64 %rd538, %rd21; + @%p61 bra $L__BB0_2; + bra.uni $L__BB0_15; +$L__BB0_2: // %__nv_exp2f.exit + // =>This Inner Loop Header: Depth=1 + .loc 1 0 33 // chunk_delta_h.py:0:33 + mov.b64 %rd21, %rd534; + mov.b64 %rd20, %rd533; + mov.b64 %rd19, %rd532; + mov.b64 %rd18, %rd531; + mov.pred %p4, %p65; + mov.pred %p3, %p64; + mov.pred %p2, %p63; + mov.pred %p1, %p62; + setp.eq.b32 %p14, %r4, 0; + .loc 1 489 33 // chunk_delta_h.py:489:33 + add.s32 %r312, %r1172, 1; + setp.gt.s32 %p15, %r312, 1; + selp.b32 %r1172, 0, %r312, %p15; + .loc 1 497 64 // chunk_delta_h.py:497:64 + add.s64 %rd250, %rd521, %rd15; + add.s64 %rd242, %rd250, -1048576; + add.s64 %rd243, %rd250, -1044480; + add.s64 %rd244, %rd250, -1040384; + .loc 1 499 33 // chunk_delta_h.py:499:33 + add.s64 %rd245, %rd250, -1036288; + mov.b64 {%r313, %r314}, %rd555; + cvt.rn.bf16x2.f32 %r243, %r314, %r313; + mov.b64 {%r315, %r316}, %rd556; + cvt.rn.bf16x2.f32 %r247, %r316, %r315; + mov.b64 {%r317, %r318}, %rd557; + cvt.rn.bf16x2.f32 %r251, %r318, %r317; + mov.b64 {%r319, %r320}, %rd558; + cvt.rn.bf16x2.f32 %r255, %r320, %r319; + mov.b64 {%r321, %r322}, %rd559; + cvt.rn.bf16x2.f32 %r244, %r322, %r321; + mov.b64 {%r323, %r324}, %rd560; + cvt.rn.bf16x2.f32 %r248, %r324, %r323; + mov.b64 {%r325, %r326}, %rd561; + cvt.rn.bf16x2.f32 %r252, %r326, %r325; + mov.b64 {%r327, %r328}, %rd562; + cvt.rn.bf16x2.f32 %r256, %r328, %r327; + mov.b64 {%r329, %r330}, %rd563; + cvt.rn.bf16x2.f32 %r245, %r330, %r329; + mov.b64 {%r331, %r332}, %rd564; + cvt.rn.bf16x2.f32 %r249, %r332, %r331; + mov.b64 {%r333, %r334}, %rd565; + cvt.rn.bf16x2.f32 %r253, %r334, %r333; + mov.b64 {%r335, %r336}, %rd566; + cvt.rn.bf16x2.f32 %r257, %r336, %r335; + mov.b64 {%r337, %r338}, %rd567; + cvt.rn.bf16x2.f32 %r246, %r338, %r337; + mov.b64 {%r339, %r340}, %rd568; + cvt.rn.bf16x2.f32 %r250, %r340, %r339; + mov.b64 {%r341, %r342}, %rd569; + cvt.rn.bf16x2.f32 %r254, %r342, %r341; + mov.b64 {%r343, %r344}, %rd570; + cvt.rn.bf16x2.f32 %r258, %r344, %r343; + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r238 + 0 ], { %r243, %r247, %r251, %r255 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r239 + 0 ], { %r244, %r248, %r252, %r256 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r240 + 0 ], { %r245, %r249, %r253, %r257 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r241 + 0 ], { %r246, %r250, %r254, %r258 }; + // end inline asm + .loc 1 499 24 // chunk_delta_h.py:499:24 + // begin inline asm + st.shared.v4.b32 [ %r242 + 0 ], { %r243, %r244, %r245, %r246 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r259, %r263, %r267, %r271}, [%r18]; + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r242 + 0 ], { %r247, %r248, %r249, %r250 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r260, %r264, %r268, %r272}, [%r18]; + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r242 + 0 ], { %r251, %r252, %r253, %r254 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r261, %r265, %r269, %r273}, [%r18]; + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r242 + 0 ], { %r255, %r256, %r257, %r258 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r262, %r266, %r270, %r274}, [%r18]; + mov.pred %p20, -1; + // begin inline asm + @%p20 st.global.v4.b32 [ %rd242 + 0 ], { %r259, %r260, %r261, %r262 }; + // end inline asm + // begin inline asm + @%p20 st.global.v4.b32 [ %rd243 + 0 ], { %r263, %r264, %r265, %r266 }; + // end inline asm + // begin inline asm + @%p20 st.global.v4.b32 [ %rd244 + 0 ], { %r267, %r268, %r269, %r270 }; + // end inline asm + // begin inline asm + @%p20 st.global.v4.b32 [ %rd245 + 0 ], { %r271, %r272, %r273, %r274 }; + // end inline asm + .loc 1 505 68 // chunk_delta_h.py:505:68 + add.s64 %rd246, %rd250, -1032192; + add.s64 %rd247, %rd250, -1028096; + add.s64 %rd248, %rd250, -1024000; + .loc 1 507 37 // chunk_delta_h.py:507:37 + add.s64 %rd249, %rd250, -1019904; + mov.b64 {%r345, %r346}, %rd539; + cvt.rn.bf16x2.f32 %r280, %r346, %r345; + mov.b64 {%r347, %r348}, %rd540; + cvt.rn.bf16x2.f32 %r284, %r348, %r347; + mov.b64 {%r349, %r350}, %rd541; + cvt.rn.bf16x2.f32 %r288, %r350, %r349; + mov.b64 {%r351, %r352}, %rd542; + cvt.rn.bf16x2.f32 %r292, %r352, %r351; + mov.b64 {%r353, %r354}, %rd543; + cvt.rn.bf16x2.f32 %r281, %r354, %r353; + mov.b64 {%r355, %r356}, %rd544; + cvt.rn.bf16x2.f32 %r285, %r356, %r355; + mov.b64 {%r357, %r358}, %rd545; + cvt.rn.bf16x2.f32 %r289, %r358, %r357; + mov.b64 {%r359, %r360}, %rd546; + cvt.rn.bf16x2.f32 %r293, %r360, %r359; + mov.b64 {%r361, %r362}, %rd547; + cvt.rn.bf16x2.f32 %r282, %r362, %r361; + mov.b64 {%r363, %r364}, %rd548; + cvt.rn.bf16x2.f32 %r286, %r364, %r363; + mov.b64 {%r365, %r366}, %rd549; + cvt.rn.bf16x2.f32 %r290, %r366, %r365; + mov.b64 {%r367, %r368}, %rd550; + cvt.rn.bf16x2.f32 %r294, %r368, %r367; + mov.b64 {%r369, %r370}, %rd551; + cvt.rn.bf16x2.f32 %r283, %r370, %r369; + mov.b64 {%r371, %r372}, %rd552; + cvt.rn.bf16x2.f32 %r287, %r372, %r371; + mov.b64 {%r373, %r374}, %rd553; + cvt.rn.bf16x2.f32 %r291, %r374, %r373; + mov.b64 {%r375, %r376}, %rd554; + cvt.rn.bf16x2.f32 %r295, %r376, %r375; + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r275 + 0 ], { %r280, %r284, %r288, %r292 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r276 + 0 ], { %r281, %r285, %r289, %r293 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r277 + 0 ], { %r282, %r286, %r290, %r294 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r278 + 0 ], { %r283, %r287, %r291, %r295 }; + // end inline asm + .loc 1 507 28 // chunk_delta_h.py:507:28 + // begin inline asm + st.shared.v4.b32 [ %r279 + 0 ], { %r280, %r281, %r282, %r283 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r296, %r300, %r304, %r308}, [%r19]; + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r279 + 0 ], { %r284, %r285, %r286, %r287 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r297, %r301, %r305, %r309}, [%r19]; + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r279 + 0 ], { %r288, %r289, %r290, %r291 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r298, %r302, %r306, %r310}, [%r19]; + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r279 + 0 ], { %r292, %r293, %r294, %r295 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r299, %r303, %r307, %r311}, [%r19]; + // begin inline asm + @%p20 st.global.v4.b32 [ %rd246 + 0 ], { %r296, %r297, %r298, %r299 }; + // end inline asm + // begin inline asm + @%p20 st.global.v4.b32 [ %rd247 + 0 ], { %r300, %r301, %r302, %r303 }; + // end inline asm + // begin inline asm + @%p20 st.global.v4.b32 [ %rd248 + 0 ], { %r304, %r305, %r306, %r307 }; + // end inline asm + // begin inline asm + @%p20 st.global.v4.b32 [ %rd249 + 0 ], { %r308, %r309, %r310, %r311 }; + // end inline asm + .loc 1 527 30 // chunk_delta_h.py:527:30 + cp.async.wait_group 11; + bar.sync 0; + shl.b32 %r377, %r1172, 2; + add.s32 %r378, %r67, %r377; + ld.shared.b32 %r25, [%r378+156672]; + .loc 1 529 26 // chunk_delta_h.py:529:26 + shl.b32 %r379, %r1172, 8; + add.s32 %r380, %r20, %r379; + ldmatrix.sync.aligned.m8n8.x1.shared.b16 {%r26}, [%r380+155648]; + add.s32 %r381, %r21, %r379; + ld.shared.v4.b32 {%r382, %r383, %r384, %r385}, [%r381+156160]; + ld.shared.v4.b32 {%r386, %r387, %r388, %r389}, [%r381+156176]; + ld.shared.v4.b32 {%r390, %r391, %r392, %r393}, [%r381+156192]; + ld.shared.v4.b32 {%r394, %r395, %r396, %r397}, [%r381+156208]; + ld.shared.v4.b32 {%r398, %r399, %r400, %r401}, [%r381+156224]; + ld.shared.v4.b32 {%r402, %r403, %r404, %r405}, [%r381+156240]; + ld.shared.v4.b32 {%r406, %r407, %r408, %r409}, [%r381+156256]; + ld.shared.v4.b32 {%r410, %r411, %r412, %r413}, [%r381+156272]; + .loc 1 536 23 // chunk_delta_h.py:536:23 + shl.b32 %r414, %r1172, 13; + add.s32 %r61, %r67, %r414; + add.s32 %r415, %r61, %r15; + ld.shared.v4.b32 {%r416, %r417, %r418, %r419}, [%r415]; + ld.shared.v4.b32 {%r420, %r421, %r422, %r423}, [%r415+2048]; + ld.shared.v4.b32 {%r424, %r425, %r426, %r427}, [%r415+4096]; + ld.shared.v4.b32 {%r428, %r429, %r430, %r431}, [%r415+6144]; + .loc 1 547 31 // chunk_delta_h.py:547:31 + fence.proxy.async.shared::cta; + elect.sync %r432|%p16, -1; + and.pred %p13, %p14, %p16; + // begin inline asm + @%p13 mbarrier.init.shared::cta.b64 [%r437], 1; + // end inline asm + shfl.sync.idx.b32 %r62, %r4, 0, 31, -1; + setp.ne.b32 %p17, %r62, 0; + @%p17 bra $L__BB0_4; +// %bb.3: // in Loop: Header=BB0_2 Depth=1 + .loc 1 540 22 // chunk_delta_h.py:540:22 + add.s32 %r434, %r61, 16384; + .loc 1 547 31 // chunk_delta_h.py:547:31 + elect.sync %r435|%p19, -1; + bfe.u32 %r436, %r434, 4, 14; + cvt.u64.u32 %rd262, %r436; + or.b64 %rd255, %rd262, 4611756662049472512; + mov.b32 %r433, 68224144; + mov.pred %p18, 0; + // begin inline asm + @%p19 tcgen05.mma.cta_group::1.kind::f16 [ %r1171 + 0 ], %rd255, %rd326, %r433, %p18; + // end inline asm + add.s64 %rd256, %rd262, 4611756662049472514; + // begin inline asm + @%p19 tcgen05.mma.cta_group::1.kind::f16 [ %r1171 + 0 ], %rd256, %rd257, %r433, %p20; + // end inline asm + add.s64 %rd258, %rd262, 4611756662049472516; + // begin inline asm + @%p19 tcgen05.mma.cta_group::1.kind::f16 [ %r1171 + 0 ], %rd258, %rd259, %r433, %p20; + // end inline asm + add.s64 %rd260, %rd262, 4611756662049472518; + // begin inline asm + @%p19 tcgen05.mma.cta_group::1.kind::f16 [ %r1171 + 0 ], %rd260, %rd261, %r433, %p20; + // end inline asm + // begin inline asm + @%p19 tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [%r437]; + // end inline asm +$L__BB0_4: // in Loop: Header=BB0_2 Depth=1 + .loc 1 0 0 // chunk_delta_h.py:0 + ex2.approx.ftz.f32 %r28, %r382; + ex2.approx.ftz.f32 %r29, %r383; + ex2.approx.ftz.f32 %r30, %r384; + ex2.approx.ftz.f32 %r31, %r385; + ex2.approx.ftz.f32 %r32, %r386; + ex2.approx.ftz.f32 %r33, %r387; + ex2.approx.ftz.f32 %r34, %r388; + ex2.approx.ftz.f32 %r35, %r389; + ex2.approx.ftz.f32 %r36, %r390; + ex2.approx.ftz.f32 %r37, %r391; + ex2.approx.ftz.f32 %r38, %r392; + ex2.approx.ftz.f32 %r39, %r393; + ex2.approx.ftz.f32 %r40, %r394; + ex2.approx.ftz.f32 %r41, %r395; + ex2.approx.ftz.f32 %r42, %r396; + ex2.approx.ftz.f32 %r43, %r397; + ex2.approx.ftz.f32 %r44, %r398; + ex2.approx.ftz.f32 %r45, %r399; + ex2.approx.ftz.f32 %r46, %r400; + ex2.approx.ftz.f32 %r47, %r401; + ex2.approx.ftz.f32 %r48, %r402; + ex2.approx.ftz.f32 %r49, %r403; + ex2.approx.ftz.f32 %r50, %r404; + ex2.approx.ftz.f32 %r51, %r405; + ex2.approx.ftz.f32 %r52, %r406; + ex2.approx.ftz.f32 %r53, %r407; + ex2.approx.ftz.f32 %r54, %r408; + ex2.approx.ftz.f32 %r55, %r409; + ex2.approx.ftz.f32 %r56, %r410; + ex2.approx.ftz.f32 %r57, %r411; + ex2.approx.ftz.f32 %r58, %r412; + ex2.approx.ftz.f32 %r59, %r413; + shl.b64 %rd251, %rd538, 1; + add.s64 %rd23, %rd8, %rd251; + shl.b64 %rd252, %rd537, 1; + add.s64 %rd24, %rd8, %rd252; + shl.b64 %rd253, %rd536, 1; + add.s64 %rd25, %rd8, %rd253; + shl.b64 %rd254, %rd535, 1; + add.s64 %rd26, %rd8, %rd254; + shl.b32 %r60, %r1172, 12; + mov.b32 {%rs1, %rs2}, %r416; + mov.b32 {%rs3, %rs4}, %r417; + mov.b32 {%rs5, %rs6}, %r418; + mov.b32 {%rs7, %rs8}, %r419; + mov.b32 {%rs9, %rs10}, %r420; + mov.b32 {%rs11, %rs12}, %r421; + mov.b32 {%rs13, %rs14}, %r422; + mov.b32 {%rs15, %rs16}, %r423; + mov.b32 {%rs17, %rs18}, %r424; + mov.b32 {%rs19, %rs20}, %r425; + mov.b32 {%rs21, %rs22}, %r426; + mov.b32 {%rs23, %rs24}, %r427; + mov.b32 {%rs25, %rs26}, %r428; + mov.b32 {%rs27, %rs28}, %r429; + mov.b32 {%rs29, %rs30}, %r430; + mov.b32 {%rs31, %rs32}, %r431; + .loc 1 547 31 // chunk_delta_h.py:547:31 + bar.sync 0; + mov.b32 %r438, 0; + // begin inline asm + +{ + .reg .pred complete; + waitLoop: + mbarrier.try_wait.parity.shared::cta.b64 complete, [%r437], %r438; + @!complete bra.uni waitLoop; +} + + // end inline asm + bar.sync 0; + elect.sync %r472|%p23, -1; + and.pred %p21, %p14, %p23; + // begin inline asm + @%p21 mbarrier.inval.shared::cta.b64 [%r437]; + // end inline asm + shl.b32 %r473, %r62, 21; + and.b32 %r474, %r473, 6291456; + add.s32 %r471, %r474, %r1171; + // begin inline asm + tcgen05.ld.sync.aligned.16x32bx2.x32.b32 {%r439, %r440, %r441, %r442, %r443, %r444, %r445, %r446, %r447, %r448, %r449, %r450, %r451, %r452, %r453, %r454, %r455, %r456, %r457, %r458, %r459, %r460, %r461, %r462, %r463, %r464, %r465, %r466, %r467, %r468, %r469, %r470}, [%r471 + 0], 32; + // end inline asm + tcgen05.wait::ld.sync.aligned; + .loc 1 557 56 // chunk_delta_h.py:557:56 + add.s32 %r513, %r471, 1048576; + // begin inline asm + @%p20 tcgen05.st.sync.aligned.16x32bx2.x32.b32 [%r513 + 0], 32, {%r439, %r440, %r441, %r442, %r443, %r444, %r445, %r446, %r447, %r448, %r449, %r450, %r451, %r452, %r453, %r454, %r455, %r456, %r457, %r458, %r459, %r460, %r461, %r462, %r463, %r464, %r465, %r466, %r467, %r468, %r469, %r470}; + // end inline asm + tcgen05.wait::st.sync.aligned; + bar.sync 0; + fence.proxy.async.shared::cta; + elect.sync %r475|%p24, -1; + and.pred %p22, %p14, %p24; + // begin inline asm + @%p22 mbarrier.init.shared::cta.b64 [%r855], 1; + // end inline asm + @%p17 bra $L__BB0_6; +// %bb.5: // in Loop: Header=BB0_2 Depth=1 + .loc 1 551 26 // chunk_delta_h.py:551:26 + add.s32 %r478, %r61, 32768; + .loc 1 557 56 // chunk_delta_h.py:557:56 + elect.sync %r479|%p26, -1; + bfe.u32 %r480, %r478, 4, 14; + cvt.u64.u32 %rd271, %r480; + or.b64 %rd263, %rd271, 4611756662049472512; + mov.b32 %r477, 68224144; + mov.pred %p25, -1; + // begin inline asm + @%p26 tcgen05.mma.cta_group::1.kind::f16 [ %r476 + 0 ], %rd263, %rd264, %r477, %p25; + // end inline asm + add.s64 %rd265, %rd271, 4611756662049472514; + // begin inline asm + @%p26 tcgen05.mma.cta_group::1.kind::f16 [ %r476 + 0 ], %rd265, %rd266, %r477, %p25; + // end inline asm + add.s64 %rd267, %rd271, 4611756662049472516; + // begin inline asm + @%p26 tcgen05.mma.cta_group::1.kind::f16 [ %r476 + 0 ], %rd267, %rd268, %r477, %p25; + // end inline asm + add.s64 %rd269, %rd271, 4611756662049472518; + // begin inline asm + @%p26 tcgen05.mma.cta_group::1.kind::f16 [ %r476 + 0 ], %rd269, %rd270, %r477, %p25; + // end inline asm + // begin inline asm + @%p26 tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [%r855]; + // end inline asm +$L__BB0_6: // %__nv_exp2f.exit423 + // in Loop: Header=BB0_2 Depth=1 + .loc 1 0 0 // chunk_delta_h.py:0 + add.s64 %rd22, %rd17, %rd529; + .loc 1 492 20 // chunk_delta_h.py:492:20 + setp.lt.s64 %p29, %rd22, %rd3; + .loc 1 557 56 // chunk_delta_h.py:557:56 + bar.sync 0; + // begin inline asm + +{ + .reg .pred complete; + waitLoop: + mbarrier.try_wait.parity.shared::cta.b64 complete, [%r855], %r438; + @!complete bra.uni waitLoop; +} + + // end inline asm + bar.sync 0; + elect.sync %r647|%p30, -1; + and.pred %p27, %p14, %p30; + // begin inline asm + @%p27 mbarrier.inval.shared::cta.b64 [%r855]; + // end inline asm + .loc 1 580 49 // chunk_delta_h.py:580:49 + sub.f32 %r648, %r25, %r26; +$L__tmp3: + .loc 3 47 33 // op.py:47:33 @[ chunk_delta_h.py:580:39 ] + ex2.approx.ftz.f32 %r649, %r648; +$L__tmp4: + .loc 1 580 55 // chunk_delta_h.py:580:55 + selp.f32 %r650, %r649, 0f00000000, %p29; + .loc 1 581 24 // chunk_delta_h.py:581:24 + shl.b32 %r651, %r60, 1; + add.s32 %r652, %r22, %r651; + ld.shared.v4.b32 {%r653, %r654, %r655, %r656}, [%r652+49152]; + .loc 1 581 16 // chunk_delta_h.py:581:16 + mov.b32 {%rs33, %rs34}, %r656; + cvt.f32.bf16 %r657, %rs34; + cvt.f32.bf16 %r658, %rs33; + mov.b64 %rd272, {%r658, %r657}; + .loc 1 580 20 // chunk_delta_h.py:580:20 + mov.b64 %rd273, {%r650, %r650}; + .loc 1 581 16 // chunk_delta_h.py:581:16 + mov.b32 {%rs35, %rs36}, %r655; + cvt.f32.bf16 %r659, %rs36; + cvt.f32.bf16 %r660, %rs35; + mov.b64 %rd274, {%r660, %r659}; + mov.b32 {%rs37, %rs38}, %r654; + cvt.f32.bf16 %r661, %rs38; + cvt.f32.bf16 %r662, %rs37; + mov.b64 %rd275, {%r662, %r661}; + mov.b32 {%rs39, %rs40}, %r653; + cvt.f32.bf16 %r663, %rs40; + cvt.f32.bf16 %r664, %rs39; + mov.b64 %rd276, {%r664, %r663}; + .loc 1 581 24 // chunk_delta_h.py:581:24 + ld.shared.v4.b32 {%r665, %r666, %r667, %r668}, [%r652+49168]; + .loc 1 581 16 // chunk_delta_h.py:581:16 + mov.b32 {%rs41, %rs42}, %r668; + cvt.f32.bf16 %r669, %rs42; + cvt.f32.bf16 %r670, %rs41; + mov.b64 %rd277, {%r670, %r669}; + mov.b32 {%rs43, %rs44}, %r667; + cvt.f32.bf16 %r671, %rs44; + cvt.f32.bf16 %r672, %rs43; + mov.b64 %rd278, {%r672, %r671}; + mov.b32 {%rs45, %rs46}, %r666; + cvt.f32.bf16 %r673, %rs46; + cvt.f32.bf16 %r674, %rs45; + mov.b64 %rd279, {%r674, %r673}; + mov.b32 {%rs47, %rs48}, %r665; + cvt.f32.bf16 %r675, %rs48; + cvt.f32.bf16 %r676, %rs47; + mov.b64 %rd280, {%r676, %r675}; + .loc 1 581 24 // chunk_delta_h.py:581:24 + ld.shared.v4.b32 {%r677, %r678, %r679, %r680}, [%r652+49184]; + .loc 1 581 16 // chunk_delta_h.py:581:16 + mov.b32 {%rs49, %rs50}, %r680; + cvt.f32.bf16 %r681, %rs50; + cvt.f32.bf16 %r682, %rs49; + mov.b64 %rd281, {%r682, %r681}; + mov.b32 {%rs51, %rs52}, %r679; + cvt.f32.bf16 %r683, %rs52; + cvt.f32.bf16 %r684, %rs51; + mov.b64 %rd282, {%r684, %r683}; + mov.b32 {%rs53, %rs54}, %r678; + cvt.f32.bf16 %r685, %rs54; + cvt.f32.bf16 %r686, %rs53; + mov.b64 %rd283, {%r686, %r685}; + mov.b32 {%rs55, %rs56}, %r677; + cvt.f32.bf16 %r687, %rs56; + cvt.f32.bf16 %r688, %rs55; + mov.b64 %rd284, {%r688, %r687}; + .loc 1 581 24 // chunk_delta_h.py:581:24 + ld.shared.v4.b32 {%r689, %r690, %r691, %r692}, [%r652+49200]; + .loc 1 581 16 // chunk_delta_h.py:581:16 + mov.b32 {%rs57, %rs58}, %r692; + cvt.f32.bf16 %r693, %rs58; + cvt.f32.bf16 %r694, %rs57; + mov.b64 %rd285, {%r694, %r693}; + mov.b32 {%rs59, %rs60}, %r691; + cvt.f32.bf16 %r695, %rs60; + cvt.f32.bf16 %r696, %rs59; + mov.b64 %rd286, {%r696, %r695}; + mov.b32 {%rs61, %rs62}, %r690; + cvt.f32.bf16 %r697, %rs62; + cvt.f32.bf16 %r698, %rs61; + mov.b64 %rd287, {%r698, %r697}; + mov.b32 {%rs63, %rs64}, %r689; + cvt.f32.bf16 %r699, %rs64; + cvt.f32.bf16 %r700, %rs63; + mov.b64 %rd288, {%r700, %r699}; + .loc 1 557 56 // chunk_delta_h.py:557:56 + // begin inline asm + tcgen05.ld.sync.aligned.16x32bx2.x32.b32 {%r481, %r482, %r483, %r484, %r485, %r486, %r487, %r488, %r489, %r490, %r491, %r492, %r493, %r494, %r495, %r496, %r497, %r498, %r499, %r500, %r501, %r502, %r503, %r504, %r505, %r506, %r507, %r508, %r509, %r510, %r511, %r512}, [%r513 + 0], 32; + // end inline asm + mov.b64 %rd289, {%r481, %r482}; + mov.b64 %rd290, {%r483, %r484}; + mov.b64 %rd291, {%r485, %r486}; + mov.b64 %rd292, {%r487, %r488}; + mov.b64 %rd293, {%r489, %r490}; + mov.b64 %rd294, {%r491, %r492}; + mov.b64 %rd295, {%r493, %r494}; + mov.b64 %rd296, {%r495, %r496}; + mov.b64 %rd297, {%r497, %r498}; + mov.b64 %rd298, {%r499, %r500}; + mov.b64 %rd299, {%r501, %r502}; + mov.b64 %rd300, {%r503, %r504}; + mov.b64 %rd301, {%r505, %r506}; + mov.b64 %rd302, {%r507, %r508}; + mov.b64 %rd303, {%r509, %r510}; + mov.b64 %rd304, {%r511, %r512}; + tcgen05.wait::ld.sync.aligned; + .loc 1 581 16 // chunk_delta_h.py:581:16 + fma.rn.f32x2 %rd305, %rd273, %rd289, %rd276; + fma.rn.f32x2 %rd306, %rd273, %rd290, %rd275; + fma.rn.f32x2 %rd307, %rd273, %rd291, %rd274; + fma.rn.f32x2 %rd308, %rd273, %rd292, %rd272; + fma.rn.f32x2 %rd309, %rd273, %rd293, %rd280; + fma.rn.f32x2 %rd310, %rd273, %rd294, %rd279; + fma.rn.f32x2 %rd311, %rd273, %rd295, %rd278; + fma.rn.f32x2 %rd312, %rd273, %rd296, %rd277; + fma.rn.f32x2 %rd313, %rd273, %rd297, %rd284; + fma.rn.f32x2 %rd314, %rd273, %rd298, %rd283; + fma.rn.f32x2 %rd315, %rd273, %rd299, %rd282; + fma.rn.f32x2 %rd316, %rd273, %rd300, %rd281; + fma.rn.f32x2 %rd317, %rd273, %rd301, %rd288; + fma.rn.f32x2 %rd318, %rd273, %rd302, %rd287; + fma.rn.f32x2 %rd319, %rd273, %rd303, %rd286; + fma.rn.f32x2 %rd320, %rd273, %rd304, %rd285; + .loc 1 583 32 // chunk_delta_h.py:583:32 + mov.b64 {%r701, %r702}, %rd305; + cvt.rn.bf16x2.f32 %r519, %r702, %r701; + mov.b64 {%r703, %r704}, %rd306; + cvt.rn.bf16x2.f32 %r523, %r704, %r703; + mov.b64 {%r705, %r706}, %rd307; + cvt.rn.bf16x2.f32 %r527, %r706, %r705; + mov.b64 {%r707, %r708}, %rd308; + cvt.rn.bf16x2.f32 %r531, %r708, %r707; + mov.b64 {%r709, %r710}, %rd309; + cvt.rn.bf16x2.f32 %r520, %r710, %r709; + mov.b64 {%r711, %r712}, %rd310; + cvt.rn.bf16x2.f32 %r524, %r712, %r711; + mov.b64 {%r713, %r714}, %rd311; + cvt.rn.bf16x2.f32 %r528, %r714, %r713; + mov.b64 {%r715, %r716}, %rd312; + cvt.rn.bf16x2.f32 %r532, %r716, %r715; + mov.b64 {%r717, %r718}, %rd313; + cvt.rn.bf16x2.f32 %r521, %r718, %r717; + mov.b64 {%r719, %r720}, %rd314; + cvt.rn.bf16x2.f32 %r525, %r720, %r719; + mov.b64 {%r721, %r722}, %rd315; + cvt.rn.bf16x2.f32 %r529, %r722, %r721; + mov.b64 {%r723, %r724}, %rd316; + cvt.rn.bf16x2.f32 %r533, %r724, %r723; + mov.b64 {%r725, %r726}, %rd317; + cvt.rn.bf16x2.f32 %r522, %r726, %r725; + mov.b64 {%r727, %r728}, %rd318; + cvt.rn.bf16x2.f32 %r526, %r728, %r727; + mov.b64 {%r729, %r730}, %rd319; + cvt.rn.bf16x2.f32 %r530, %r730, %r729; + mov.b64 {%r731, %r732}, %rd320; + cvt.rn.bf16x2.f32 %r534, %r732, %r731; + // begin inline asm + st.shared.v4.b32 [ %r514 + 0 ], { %r519, %r523, %r527, %r531 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r515 + 0 ], { %r520, %r524, %r528, %r532 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r516 + 0 ], { %r521, %r525, %r529, %r533 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r517 + 0 ], { %r522, %r526, %r530, %r534 }; + // end inline asm + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r518 + 0 ], { %r519, %r520, %r521, %r522 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r535, %r539, %r543, %r547}, [%r23]; + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r518 + 0 ], { %r523, %r524, %r525, %r526 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r536, %r540, %r544, %r548}, [%r23]; + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r518 + 0 ], { %r527, %r528, %r529, %r530 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r537, %r541, %r545, %r549}, [%r23]; + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r518 + 0 ], { %r531, %r532, %r533, %r534 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r538, %r542, %r546, %r550}, [%r23]; + .loc 1 583 24 // chunk_delta_h.py:583:24 + // begin inline asm + @%p69 st.global.v4.b32 [ %rd23 + 0 ], { %r535, %r536, %r537, %r538 }; + // end inline asm + // begin inline asm + @%p68 st.global.v4.b32 [ %rd24 + 0 ], { %r539, %r540, %r541, %r542 }; + // end inline asm + // begin inline asm + @%p67 st.global.v4.b32 [ %rd25 + 0 ], { %r543, %r544, %r545, %r546 }; + // end inline asm + // begin inline asm + @%p66 st.global.v4.b32 [ %rd26 + 0 ], { %r547, %r548, %r549, %r550 }; + // end inline asm + .loc 1 588 22 // chunk_delta_h.py:588:22 + add.s32 %r64, %r61, %r24; + ld.shared.b16 %rs65, [%r64+81920]; + ld.shared.b16 %rs66, [%r64+82048]; + ld.shared.b16 %rs67, [%r64+82176]; + ld.shared.b16 %rs68, [%r64+82304]; + ld.shared.b16 %rs69, [%r64+82432]; + ld.shared.b16 %rs70, [%r64+82560]; + ld.shared.b16 %rs71, [%r64+82688]; + ld.shared.b16 %rs72, [%r64+82816]; + ld.shared.b16 %rs73, [%r64+82944]; + ld.shared.b16 %rs74, [%r64+83072]; + ld.shared.b16 %rs75, [%r64+83200]; + ld.shared.b16 %rs76, [%r64+83328]; + ld.shared.b16 %rs77, [%r64+83456]; + ld.shared.b16 %rs78, [%r64+83584]; + ld.shared.b16 %rs79, [%r64+83712]; + ld.shared.b16 %rs80, [%r64+83840]; + ld.shared.b16 %rs81, [%r64+83968]; + ld.shared.b16 %rs82, [%r64+84096]; + ld.shared.b16 %rs83, [%r64+84224]; + ld.shared.b16 %rs84, [%r64+84352]; + ld.shared.b16 %rs85, [%r64+84480]; + ld.shared.b16 %rs86, [%r64+84608]; + ld.shared.b16 %rs87, [%r64+84736]; + ld.shared.b16 %rs88, [%r64+84864]; + ld.shared.b16 %rs89, [%r64+84992]; + ld.shared.b16 %rs90, [%r64+85120]; + ld.shared.b16 %rs91, [%r64+85248]; + ld.shared.b16 %rs92, [%r64+85376]; + ld.shared.b16 %rs93, [%r64+85504]; + ld.shared.b16 %rs94, [%r64+85632]; + ld.shared.b16 %rs95, [%r64+85760]; + ld.shared.b16 %rs96, [%r64+85888]; + .loc 1 591 24 // chunk_delta_h.py:591:24 + cvt.f32.bf16 %r733, %rs65; + cvt.f32.bf16 %r734, %rs66; + cvt.f32.bf16 %r735, %rs67; + cvt.f32.bf16 %r736, %rs68; + cvt.f32.bf16 %r737, %rs69; + cvt.f32.bf16 %r738, %rs70; + cvt.f32.bf16 %r739, %rs71; + cvt.f32.bf16 %r740, %rs72; + cvt.f32.bf16 %r741, %rs73; + cvt.f32.bf16 %r742, %rs74; + cvt.f32.bf16 %r743, %rs75; + cvt.f32.bf16 %r744, %rs76; + cvt.f32.bf16 %r745, %rs77; + cvt.f32.bf16 %r746, %rs78; + cvt.f32.bf16 %r747, %rs79; + cvt.f32.bf16 %r748, %rs80; + cvt.f32.bf16 %r749, %rs81; + cvt.f32.bf16 %r750, %rs82; + cvt.f32.bf16 %r751, %rs83; + cvt.f32.bf16 %r752, %rs84; + cvt.f32.bf16 %r753, %rs85; + cvt.f32.bf16 %r754, %rs86; + cvt.f32.bf16 %r755, %rs87; + cvt.f32.bf16 %r756, %rs88; + cvt.f32.bf16 %r757, %rs89; + cvt.f32.bf16 %r758, %rs90; + cvt.f32.bf16 %r759, %rs91; + cvt.f32.bf16 %r760, %rs92; + cvt.f32.bf16 %r761, %rs93; + cvt.f32.bf16 %r762, %rs94; + cvt.f32.bf16 %r763, %rs95; + cvt.f32.bf16 %r764, %rs96; + mul.f32 %r615, %r28, %r733; + mul.f32 %r616, %r29, %r734; + mul.f32 %r617, %r30, %r735; + mul.f32 %r618, %r31, %r736; + mul.f32 %r619, %r32, %r737; + mul.f32 %r620, %r33, %r738; + mul.f32 %r621, %r34, %r739; + mul.f32 %r622, %r35, %r740; + mul.f32 %r623, %r36, %r741; + mul.f32 %r624, %r37, %r742; + mul.f32 %r625, %r38, %r743; + mul.f32 %r626, %r39, %r744; + mul.f32 %r627, %r40, %r745; + mul.f32 %r628, %r41, %r746; + mul.f32 %r629, %r42, %r747; + mul.f32 %r630, %r43, %r748; + mul.f32 %r631, %r44, %r749; + mul.f32 %r632, %r45, %r750; + mul.f32 %r633, %r46, %r751; + mul.f32 %r634, %r47, %r752; + mul.f32 %r635, %r48, %r753; + mul.f32 %r636, %r49, %r754; + mul.f32 %r637, %r50, %r755; + mul.f32 %r638, %r51, %r756; + mul.f32 %r639, %r52, %r757; + mul.f32 %r640, %r53, %r758; + mul.f32 %r641, %r54, %r759; + mul.f32 %r642, %r55, %r760; + mul.f32 %r643, %r56, %r761; + mul.f32 %r644, %r57, %r762; + mul.f32 %r645, %r58, %r763; + mul.f32 %r646, %r59, %r764; + .loc 1 600 55 // chunk_delta_h.py:600:55 + cvt.f32.bf16 %r552, %rs1; + cvt.f32.bf16 %r568, %rs2; + cvt.f32.bf16 %r584, %rs3; + cvt.f32.bf16 %r600, %rs4; + cvt.f32.bf16 %r554, %rs5; + cvt.f32.bf16 %r570, %rs6; + cvt.f32.bf16 %r586, %rs7; + cvt.f32.bf16 %r602, %rs8; + cvt.f32.bf16 %r556, %rs9; + cvt.f32.bf16 %r572, %rs10; + cvt.f32.bf16 %r588, %rs11; + cvt.f32.bf16 %r604, %rs12; + cvt.f32.bf16 %r558, %rs13; + cvt.f32.bf16 %r574, %rs14; + cvt.f32.bf16 %r590, %rs15; + cvt.f32.bf16 %r606, %rs16; + cvt.f32.bf16 %r560, %rs17; + cvt.f32.bf16 %r576, %rs18; + cvt.f32.bf16 %r592, %rs19; + cvt.f32.bf16 %r608, %rs20; + cvt.f32.bf16 %r562, %rs21; + cvt.f32.bf16 %r578, %rs22; + cvt.f32.bf16 %r594, %rs23; + cvt.f32.bf16 %r610, %rs24; + cvt.f32.bf16 %r564, %rs25; + cvt.f32.bf16 %r580, %rs26; + cvt.f32.bf16 %r596, %rs27; + cvt.f32.bf16 %r612, %rs28; + cvt.f32.bf16 %r566, %rs29; + cvt.f32.bf16 %r582, %rs30; + cvt.f32.bf16 %r598, %rs31; + cvt.f32.bf16 %r614, %rs32; + bar.sync 0; + // begin inline asm + st.shared.b32 [ %r551 + 0 ], %r552; + // end inline asm + // begin inline asm + st.shared.b32 [ %r553 + 0 ], %r554; + // end inline asm + // begin inline asm + st.shared.b32 [ %r555 + 0 ], %r556; + // end inline asm + // begin inline asm + st.shared.b32 [ %r557 + 0 ], %r558; + // end inline asm + // begin inline asm + st.shared.b32 [ %r559 + 0 ], %r560; + // end inline asm + // begin inline asm + st.shared.b32 [ %r561 + 0 ], %r562; + // end inline asm + // begin inline asm + st.shared.b32 [ %r563 + 0 ], %r564; + // end inline asm + // begin inline asm + st.shared.b32 [ %r565 + 0 ], %r566; + // end inline asm + // begin inline asm + st.shared.b32 [ %r567 + 0 ], %r568; + // end inline asm + // begin inline asm + st.shared.b32 [ %r569 + 0 ], %r570; + // end inline asm + // begin inline asm + st.shared.b32 [ %r571 + 0 ], %r572; + // end inline asm + // begin inline asm + st.shared.b32 [ %r573 + 0 ], %r574; + // end inline asm + // begin inline asm + st.shared.b32 [ %r575 + 0 ], %r576; + // end inline asm + // begin inline asm + st.shared.b32 [ %r577 + 0 ], %r578; + // end inline asm + // begin inline asm + st.shared.b32 [ %r579 + 0 ], %r580; + // end inline asm + // begin inline asm + st.shared.b32 [ %r581 + 0 ], %r582; + // end inline asm + // begin inline asm + st.shared.b32 [ %r583 + 0 ], %r584; + // end inline asm + // begin inline asm + st.shared.b32 [ %r585 + 0 ], %r586; + // end inline asm + // begin inline asm + st.shared.b32 [ %r587 + 0 ], %r588; + // end inline asm + // begin inline asm + st.shared.b32 [ %r589 + 0 ], %r590; + // end inline asm + // begin inline asm + st.shared.b32 [ %r591 + 0 ], %r592; + // end inline asm + // begin inline asm + st.shared.b32 [ %r593 + 0 ], %r594; + // end inline asm + // begin inline asm + st.shared.b32 [ %r595 + 0 ], %r596; + // end inline asm + // begin inline asm + st.shared.b32 [ %r597 + 0 ], %r598; + // end inline asm + // begin inline asm + st.shared.b32 [ %r599 + 0 ], %r600; + // end inline asm + // begin inline asm + st.shared.b32 [ %r601 + 0 ], %r602; + // end inline asm + // begin inline asm + st.shared.b32 [ %r603 + 0 ], %r604; + // end inline asm + // begin inline asm + st.shared.b32 [ %r605 + 0 ], %r606; + // end inline asm + // begin inline asm + st.shared.b32 [ %r607 + 0 ], %r608; + // end inline asm + // begin inline asm + st.shared.b32 [ %r609 + 0 ], %r610; + // end inline asm + // begin inline asm + st.shared.b32 [ %r611 + 0 ], %r612; + // end inline asm + // begin inline asm + st.shared.b32 [ %r613 + 0 ], %r614; + // end inline asm + .loc 1 600 47 // chunk_delta_h.py:600:47 + add.s32 %r776, %r471, 192; + mov.pred %p34, -1; + // begin inline asm + @%p34 tcgen05.st.sync.aligned.16x32bx2.x32.b32 [%r776 + 0], 32, {%r615, %r616, %r617, %r618, %r619, %r620, %r621, %r622, %r623, %r624, %r625, %r626, %r627, %r628, %r629, %r630, %r631, %r632, %r633, %r634, %r635, %r636, %r637, %r638, %r639, %r640, %r641, %r642, %r643, %r644, %r645, %r646}; + // end inline asm + tcgen05.wait::st.sync.aligned; + bar.sync 0; + fence.proxy.async.shared::cta; + elect.sync %r765|%p31, -1; + and.pred %p28, %p14, %p31; + add.s32 %r847, %r67, 156688; + // begin inline asm + @%p28 mbarrier.init.shared::cta.b64 [%r847], 1; + // end inline asm + bar.sync 0; + @%p17 bra $L__BB0_8; +// %bb.7: // in Loop: Header=BB0_2 Depth=1 + elect.sync %r768|%p33, -1; + mov.b32 %r767, 68159760; + mov.pred %p32, 0; + // begin inline asm + @%p33 tcgen05.mma.cta_group::1.kind::tf32 [ %r766 + 0 ], [ %r844 + 0 ], %rd326, %r767, %p32; + // end inline asm + // begin inline asm + @%p33 tcgen05.mma.cta_group::1.kind::tf32 [ %r766 + 0 ], [ %r844 + 8 ], %rd327, %r767, %p34; + // end inline asm + // begin inline asm + @%p33 tcgen05.mma.cta_group::1.kind::tf32 [ %r766 + 0 ], [ %r844 + 16 ], %rd328, %r767, %p34; + // end inline asm + // begin inline asm + @%p33 tcgen05.mma.cta_group::1.kind::tf32 [ %r766 + 0 ], [ %r844 + 24 ], %rd329, %r767, %p34; + // end inline asm + // begin inline asm + @%p33 tcgen05.mma.cta_group::1.kind::tf32 [ %r766 + 0 ], [ %r844 + 32 ], %rd330, %r767, %p34; + // end inline asm + // begin inline asm + @%p33 tcgen05.mma.cta_group::1.kind::tf32 [ %r766 + 0 ], [ %r844 + 40 ], %rd331, %r767, %p34; + // end inline asm + // begin inline asm + @%p33 tcgen05.mma.cta_group::1.kind::tf32 [ %r766 + 0 ], [ %r844 + 48 ], %rd332, %r767, %p34; + // end inline asm + // begin inline asm + @%p33 tcgen05.mma.cta_group::1.kind::tf32 [ %r766 + 0 ], [ %r844 + 56 ], %rd333, %r767, %p34; + // end inline asm + // begin inline asm + @%p33 tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [%r847]; + // end inline asm +$L__BB0_8: // in Loop: Header=BB0_2 Depth=1 + .loc 1 0 47 // chunk_delta_h.py:0:47 + ex2.approx.ftz.f32 %r27, %r25; + mov.b32 %r769, 0; + .loc 1 600 47 // chunk_delta_h.py:600:47 + // begin inline asm + +{ + .reg .pred complete; + waitLoop: + mbarrier.try_wait.parity.shared::cta.b64 complete, [%r847], %r769; + @!complete bra.uni waitLoop; +} + + // end inline asm + bar.sync 0; + elect.sync %r770|%p37, -1; + and.pred %p35, %p14, %p37; + // begin inline asm + @%p35 mbarrier.inval.shared::cta.b64 [%r847]; + // end inline asm + .loc 1 600 89 // chunk_delta_h.py:600:89 + fence.proxy.async.shared::cta; + elect.sync %r771|%p38, -1; + and.pred %p36, %p14, %p38; + // begin inline asm + @%p36 mbarrier.init.shared::cta.b64 [%r847], 1; + // end inline asm + @%p17 bra $L__BB0_10; +// %bb.9: // in Loop: Header=BB0_2 Depth=1 + .loc 1 0 89 // chunk_delta_h.py:0:89 + add.s32 %r63, %r61, 65536; + .loc 1 600 89 // chunk_delta_h.py:600:89 + elect.sync %r774|%p40, -1; + bfe.u32 %r775, %r63, 4, 14; + cvt.u64.u32 %rd325, %r775; + or.b64 %rd321, %rd325, 4611756662049472512; + mov.b32 %r773, 68256912; + mov.pred %p39, 0; + // begin inline asm + @%p40 tcgen05.mma.cta_group::1.kind::f16 [ %r772 + 0 ], %rd321, %rd335, %r773, %p39; + // end inline asm + add.s64 %rd322, %rd325, 4611756662049472640; + mov.pred %p41, -1; + // begin inline asm + @%p40 tcgen05.mma.cta_group::1.kind::f16 [ %r772 + 0 ], %rd322, %rd337, %r773, %p41; + // end inline asm + add.s64 %rd323, %rd325, 4611756662049472768; + // begin inline asm + @%p40 tcgen05.mma.cta_group::1.kind::f16 [ %r772 + 0 ], %rd323, %rd339, %r773, %p41; + // end inline asm + add.s64 %rd324, %rd325, 4611756662049472896; + // begin inline asm + @%p40 tcgen05.mma.cta_group::1.kind::f16 [ %r772 + 0 ], %rd324, %rd341, %r773, %p41; + // end inline asm + // begin inline asm + @%p40 tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [%r847]; + // end inline asm +$L__BB0_10: // in Loop: Header=BB0_2 Depth=1 + .loc 1 0 89 // chunk_delta_h.py:0:89 + mov.b64 %rd27, {%r27, %r27}; + .loc 1 600 89 // chunk_delta_h.py:600:89 + bar.sync 0; + // begin inline asm + +{ + .reg .pred complete; + waitLoop: + mbarrier.try_wait.parity.shared::cta.b64 complete, [%r847], %r769; + @!complete bra.uni waitLoop; +} + + // end inline asm + bar.sync 0; + elect.sync %r809|%p44, -1; + and.pred %p42, %p14, %p44; + // begin inline asm + @%p42 mbarrier.inval.shared::cta.b64 [%r847]; + // end inline asm + .loc 1 604 26 // chunk_delta_h.py:604:26 + ld.shared.b16 %rs97, [%r64+98304]; + ld.shared.b16 %rs98, [%r64+98432]; + ld.shared.b16 %rs99, [%r64+98560]; + ld.shared.b16 %rs100, [%r64+98688]; + ld.shared.b16 %rs101, [%r64+98816]; + ld.shared.b16 %rs102, [%r64+98944]; + ld.shared.b16 %rs103, [%r64+99072]; + ld.shared.b16 %rs104, [%r64+99200]; + ld.shared.b16 %rs105, [%r64+99328]; + ld.shared.b16 %rs106, [%r64+99456]; + ld.shared.b16 %rs107, [%r64+99584]; + ld.shared.b16 %rs108, [%r64+99712]; + ld.shared.b16 %rs109, [%r64+99840]; + ld.shared.b16 %rs110, [%r64+99968]; + ld.shared.b16 %rs111, [%r64+100096]; + ld.shared.b16 %rs112, [%r64+100224]; + ld.shared.b16 %rs113, [%r64+100352]; + ld.shared.b16 %rs114, [%r64+100480]; + ld.shared.b16 %rs115, [%r64+100608]; + ld.shared.b16 %rs116, [%r64+100736]; + ld.shared.b16 %rs117, [%r64+100864]; + ld.shared.b16 %rs118, [%r64+100992]; + ld.shared.b16 %rs119, [%r64+101120]; + ld.shared.b16 %rs120, [%r64+101248]; + ld.shared.b16 %rs121, [%r64+101376]; + ld.shared.b16 %rs122, [%r64+101504]; + ld.shared.b16 %rs123, [%r64+101632]; + ld.shared.b16 %rs124, [%r64+101760]; + ld.shared.b16 %rs125, [%r64+101888]; + ld.shared.b16 %rs126, [%r64+102016]; + ld.shared.b16 %rs127, [%r64+102144]; + ld.shared.b16 %rs128, [%r64+102272]; + .loc 1 608 28 // chunk_delta_h.py:608:28 + cvt.f32.bf16 %r810, %rs97; + cvt.f32.bf16 %r811, %rs98; + cvt.f32.bf16 %r812, %rs99; + cvt.f32.bf16 %r813, %rs100; + cvt.f32.bf16 %r814, %rs101; + cvt.f32.bf16 %r815, %rs102; + cvt.f32.bf16 %r816, %rs103; + cvt.f32.bf16 %r817, %rs104; + cvt.f32.bf16 %r818, %rs105; + cvt.f32.bf16 %r819, %rs106; + cvt.f32.bf16 %r820, %rs107; + cvt.f32.bf16 %r821, %rs108; + cvt.f32.bf16 %r822, %rs109; + cvt.f32.bf16 %r823, %rs110; + cvt.f32.bf16 %r824, %rs111; + cvt.f32.bf16 %r825, %rs112; + cvt.f32.bf16 %r826, %rs113; + cvt.f32.bf16 %r827, %rs114; + cvt.f32.bf16 %r828, %rs115; + cvt.f32.bf16 %r829, %rs116; + cvt.f32.bf16 %r830, %rs117; + cvt.f32.bf16 %r831, %rs118; + cvt.f32.bf16 %r832, %rs119; + cvt.f32.bf16 %r833, %rs120; + cvt.f32.bf16 %r834, %rs121; + cvt.f32.bf16 %r835, %rs122; + cvt.f32.bf16 %r836, %rs123; + cvt.f32.bf16 %r837, %rs124; + cvt.f32.bf16 %r838, %rs125; + cvt.f32.bf16 %r839, %rs126; + cvt.f32.bf16 %r840, %rs127; + cvt.f32.bf16 %r841, %rs128; + mul.f32 %r777, %r28, %r810; + mul.f32 %r778, %r29, %r811; + mul.f32 %r779, %r30, %r812; + mul.f32 %r780, %r31, %r813; + mul.f32 %r781, %r32, %r814; + mul.f32 %r782, %r33, %r815; + mul.f32 %r783, %r34, %r816; + mul.f32 %r784, %r35, %r817; + mul.f32 %r785, %r36, %r818; + mul.f32 %r786, %r37, %r819; + mul.f32 %r787, %r38, %r820; + mul.f32 %r788, %r39, %r821; + mul.f32 %r789, %r40, %r822; + mul.f32 %r790, %r41, %r823; + mul.f32 %r791, %r42, %r824; + mul.f32 %r792, %r43, %r825; + mul.f32 %r793, %r44, %r826; + mul.f32 %r794, %r45, %r827; + mul.f32 %r795, %r46, %r828; + mul.f32 %r796, %r47, %r829; + mul.f32 %r797, %r48, %r830; + mul.f32 %r798, %r49, %r831; + mul.f32 %r799, %r50, %r832; + mul.f32 %r800, %r51, %r833; + mul.f32 %r801, %r52, %r834; + mul.f32 %r802, %r53, %r835; + mul.f32 %r803, %r54, %r836; + mul.f32 %r804, %r55, %r837; + mul.f32 %r805, %r56, %r838; + mul.f32 %r806, %r57, %r839; + mul.f32 %r807, %r58, %r840; + mul.f32 %r808, %r59, %r841; + mov.pred %p48, -1; + .loc 1 617 51 // chunk_delta_h.py:617:51 + // begin inline asm + @%p48 tcgen05.st.sync.aligned.16x32bx2.x32.b32 [%r776 + 0], 32, {%r777, %r778, %r779, %r780, %r781, %r782, %r783, %r784, %r785, %r786, %r787, %r788, %r789, %r790, %r791, %r792, %r793, %r794, %r795, %r796, %r797, %r798, %r799, %r800, %r801, %r802, %r803, %r804, %r805, %r806, %r807, %r808}; + // end inline asm + tcgen05.wait::st.sync.aligned; + bar.sync 0; + fence.proxy.async.shared::cta; + elect.sync %r842|%p45, -1; + and.pred %p43, %p14, %p45; + // begin inline asm + @%p43 mbarrier.init.shared::cta.b64 [%r847], 1; + // end inline asm + @%p17 bra $L__BB0_12; +// %bb.11: // in Loop: Header=BB0_2 Depth=1 + elect.sync %r846|%p47, -1; + mov.b32 %r845, 68159760; + mov.pred %p46, 0; + // begin inline asm + @%p47 tcgen05.mma.cta_group::1.kind::tf32 [ %r843 + 0 ], [ %r844 + 0 ], %rd326, %r845, %p46; + // end inline asm + // begin inline asm + @%p47 tcgen05.mma.cta_group::1.kind::tf32 [ %r843 + 0 ], [ %r844 + 8 ], %rd327, %r845, %p48; + // end inline asm + // begin inline asm + @%p47 tcgen05.mma.cta_group::1.kind::tf32 [ %r843 + 0 ], [ %r844 + 16 ], %rd328, %r845, %p48; + // end inline asm + // begin inline asm + @%p47 tcgen05.mma.cta_group::1.kind::tf32 [ %r843 + 0 ], [ %r844 + 24 ], %rd329, %r845, %p48; + // end inline asm + // begin inline asm + @%p47 tcgen05.mma.cta_group::1.kind::tf32 [ %r843 + 0 ], [ %r844 + 32 ], %rd330, %r845, %p48; + // end inline asm + // begin inline asm + @%p47 tcgen05.mma.cta_group::1.kind::tf32 [ %r843 + 0 ], [ %r844 + 40 ], %rd331, %r845, %p48; + // end inline asm + // begin inline asm + @%p47 tcgen05.mma.cta_group::1.kind::tf32 [ %r843 + 0 ], [ %r844 + 48 ], %rd332, %r845, %p48; + // end inline asm + // begin inline asm + @%p47 tcgen05.mma.cta_group::1.kind::tf32 [ %r843 + 0 ], [ %r844 + 56 ], %rd333, %r845, %p48; + // end inline asm + // begin inline asm + @%p47 tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [%r847]; + // end inline asm +$L__BB0_12: // in Loop: Header=BB0_2 Depth=1 + .loc 1 0 51 // chunk_delta_h.py:0:51 + mul.f32x2 %rd28, %rd555, %rd27; + mul.f32x2 %rd29, %rd556, %rd27; + mul.f32x2 %rd30, %rd557, %rd27; + mul.f32x2 %rd31, %rd558, %rd27; + mul.f32x2 %rd32, %rd559, %rd27; + mul.f32x2 %rd33, %rd560, %rd27; + mul.f32x2 %rd34, %rd561, %rd27; + mul.f32x2 %rd35, %rd562, %rd27; + mul.f32x2 %rd36, %rd563, %rd27; + mul.f32x2 %rd37, %rd564, %rd27; + mul.f32x2 %rd38, %rd565, %rd27; + mul.f32x2 %rd39, %rd566, %rd27; + mul.f32x2 %rd40, %rd567, %rd27; + mul.f32x2 %rd41, %rd568, %rd27; + mul.f32x2 %rd42, %rd569, %rd27; + mul.f32x2 %rd43, %rd570, %rd27; + mul.f32x2 %rd44, %rd539, %rd27; + mul.f32x2 %rd45, %rd540, %rd27; + mul.f32x2 %rd46, %rd541, %rd27; + mul.f32x2 %rd47, %rd542, %rd27; + mul.f32x2 %rd48, %rd543, %rd27; + mul.f32x2 %rd49, %rd544, %rd27; + mul.f32x2 %rd50, %rd545, %rd27; + mul.f32x2 %rd51, %rd546, %rd27; + mul.f32x2 %rd52, %rd547, %rd27; + mul.f32x2 %rd53, %rd548, %rd27; + mul.f32x2 %rd54, %rd549, %rd27; + mul.f32x2 %rd55, %rd550, %rd27; + mul.f32x2 %rd56, %rd551, %rd27; + mul.f32x2 %rd57, %rd552, %rd27; + mul.f32x2 %rd58, %rd553, %rd27; + mul.f32x2 %rd59, %rd554, %rd27; + .loc 1 617 51 // chunk_delta_h.py:617:51 + bar.sync 0; + mov.b32 %r848, 0; + // begin inline asm + +{ + .reg .pred complete; + waitLoop: + mbarrier.try_wait.parity.shared::cta.b64 complete, [%r847], %r848; + @!complete bra.uni waitLoop; +} + + // end inline asm + bar.sync 0; + elect.sync %r849|%p51, -1; + and.pred %p49, %p14, %p51; + // begin inline asm + @%p49 mbarrier.inval.shared::cta.b64 [%r847]; + // end inline asm + .loc 1 617 93 // chunk_delta_h.py:617:93 + fence.proxy.async.shared::cta; + elect.sync %r850|%p52, -1; + and.pred %p50, %p14, %p52; + // begin inline asm + @%p50 mbarrier.init.shared::cta.b64 [%r855], 1; + // end inline asm + @%p17 bra $L__BB0_14; +// %bb.13: // in Loop: Header=BB0_2 Depth=1 + .loc 1 0 0 // chunk_delta_h.py:0 + add.s32 %r65, %r61, 114688; + .loc 1 617 93 // chunk_delta_h.py:617:93 + elect.sync %r853|%p54, -1; + bfe.u32 %r854, %r65, 4, 14; + cvt.u64.u32 %rd342, %r854; + or.b64 %rd334, %rd342, 4611756662049472512; + mov.b32 %r852, 68256912; + mov.pred %p53, 0; + // begin inline asm + @%p54 tcgen05.mma.cta_group::1.kind::f16 [ %r851 + 0 ], %rd334, %rd335, %r852, %p53; + // end inline asm + add.s64 %rd336, %rd342, 4611756662049472640; + mov.pred %p55, -1; + // begin inline asm + @%p54 tcgen05.mma.cta_group::1.kind::f16 [ %r851 + 0 ], %rd336, %rd337, %r852, %p55; + // end inline asm + add.s64 %rd338, %rd342, 4611756662049472768; + // begin inline asm + @%p54 tcgen05.mma.cta_group::1.kind::f16 [ %r851 + 0 ], %rd338, %rd339, %r852, %p55; + // end inline asm + add.s64 %rd340, %rd342, 4611756662049472896; + // begin inline asm + @%p54 tcgen05.mma.cta_group::1.kind::f16 [ %r851 + 0 ], %rd340, %rd341, %r852, %p55; + // end inline asm + // begin inline asm + @%p54 tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [%r855]; + // end inline asm + bra.uni $L__BB0_14; +$L__BB0_15: // %._crit_edge + .loc 1 489 33 // chunk_delta_h.py:489:33 + cp.async.wait_group 0; + bar.sync 0; + .loc 1 653 4 // chunk_delta_h.py:653:4 + bar.sync 0; + // begin inline asm + @%p5 tcgen05.dealloc.cta_group::1.sync.aligned.b32 %r1171, 256; + // end inline asm + ret; +$L__tmp5: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .file 3 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 232 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xe1 DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 99 // DW_AT_name +.b8 104 +.b8 117 +.b8 110 +.b8 107 +.b8 95 +.b8 100 +.b8 101 +.b8 108 +.b8 116 +.b8 97 +.b8 95 +.b8 104 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 111 +.b8 112 +.b8 115 +.b8 47 +.b8 99 +.b8 111 +.b8 109 +.b8 109 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // Abbrev [2] 0x70:0x33 DW_TAG_subprogram +.b8 99 // DW_AT_name +.b8 104 +.b8 117 +.b8 110 +.b8 107 +.b8 95 +.b8 103 +.b8 97 +.b8 116 +.b8 101 +.b8 100 +.b8 95 +.b8 100 +.b8 101 +.b8 108 +.b8 116 +.b8 97 +.b8 95 +.b8 114 +.b8 117 +.b8 108 +.b8 101 +.b8 95 +.b8 98 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 95 +.b8 100 +.b8 104 +.b8 117 +.b8 95 +.b8 98 +.b8 108 +.b8 111 +.b8 99 +.b8 107 +.b8 100 +.b8 105 +.b8 109 +.b8 54 +.b8 52 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0xa3:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 112 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0xb8:0x19 DW_TAG_inlined_subroutine +.b32 112 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp2 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 154 // DW_AT_call_line +.b8 1 +.b8 24 // DW_AT_call_column +.b8 4 // Abbrev [4] 0xd1:0x19 DW_TAG_inlined_subroutine +.b32 112 // DW_AT_abstract_origin +.b64 $L__tmp3 // DW_AT_low_pc +.b64 $L__tmp4 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 68 // DW_AT_call_line +.b8 2 +.b8 39 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.source b/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.source new file mode 100644 index 0000000000000000000000000000000000000000..12b6bbe4054e93d9b73c1ca940d68e27edd1be8a --- /dev/null +++ b/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.source @@ -0,0 +1,1041 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":370:0) +#loc225 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":34:0) +#loc234 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:0) +#loc238 = loc("q"(#loc)) +#loc239 = loc("k"(#loc)) +#loc240 = loc("w"(#loc)) +#loc241 = loc("g"(#loc)) +#loc242 = loc("do"(#loc)) +#loc243 = loc("dh"(#loc)) +#loc244 = loc("dv"(#loc)) +#loc245 = loc("dv2"(#loc)) +#loc246 = loc("scale"(#loc)) +#loc247 = loc("T"(#loc)) +#loc453 = loc("x"(#loc225)) +#loc454 = loc("x"(#loc234)) +module { + tt.func public @chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64(%q: !tt.ptr {tt.divisibility = 16 : i32} loc("q"(#loc)), %k: !tt.ptr {tt.divisibility = 16 : i32} loc("k"(#loc)), %w: !tt.ptr {tt.divisibility = 16 : i32} loc("w"(#loc)), %g: !tt.ptr {tt.divisibility = 16 : i32} loc("g"(#loc)), %do: !tt.ptr {tt.divisibility = 16 : i32} loc("do"(#loc)), %dh: !tt.ptr {tt.divisibility = 16 : i32} loc("dh"(#loc)), %dv: !tt.ptr {tt.divisibility = 16 : i32} loc("dv"(#loc)), %dv2: !tt.ptr {tt.divisibility = 16 : i32} loc("dv2"(#loc)), %scale: f32 loc("scale"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %pid = tt.get_program_id x : i32 loc(#loc248) + %pid_0 = arith.extsi %pid : i32 to i64 loc(#loc249) + tt.call @triton.language.standard.cdiv__c128_c64() : () -> () loc(#loc250) + %NV = arith.constant 2 : i32 loc(#loc251) + %0 = arith.extsi %NV : i32 to i64 loc(#loc5) + %1 = arith.remsi %pid_0, %0 : i64 loc(#loc5) + %2 = arith.extsi %NV : i32 to i64 loc(#loc6) + %3 = arith.divsi %pid_0, %2 : i64 loc(#loc6) + %c32_i32 = arith.constant 32 : i32 loc(#loc7) + %c32_i64 = arith.constant 32 : i64 loc(#loc7) + %4 = arith.divsi %3, %c32_i64 : i64 loc(#loc7) + %c32_i32_1 = arith.constant 32 : i32 loc(#loc8) + %c32_i64_2 = arith.constant 32 : i64 loc(#loc8) + %5 = arith.remsi %3, %c32_i64_2 : i64 loc(#loc8) + %6 = arith.extsi %T : i32 to i64 loc(#loc9) + %7 = arith.muli %4, %6 : i64 loc(#loc9) + %8 = arith.extsi %T : i32 to i64 loc(#loc10) + %9 = arith.muli %4, %8 : i64 loc(#loc10) + %10 = arith.extsi %T : i32 to i64 loc(#loc11) + %11 = arith.addi %9, %10 : i64 loc(#loc11) + %NT = tt.call @triton.language.standard.cdiv__i32_c64(%T) : (i32) -> i32 loc(#loc252) + %boh = arith.extsi %NT : i32 to i64 loc(#loc253) + %boh_3 = arith.muli %4, %boh : i64 loc(#loc253) + %b_dh1 = tt.call @triton.language.standard.zeros__Tc64_c64T_cfp32() : () -> tensor<64x64xf32> loc(#loc254) + %b_dh2 = tt.call @triton.language.standard.zeros__Tc64_c64T_cfp32() : () -> tensor<64x64xf32> loc(#loc255) + %q_4 = arith.constant 32 : i32 loc(#loc256) + %q_5 = arith.constant 32 : i64 loc(#loc256) + %q_6 = arith.muli %7, %q_5 : i64 loc(#loc256) + %q_7 = arith.constant 1 : i32 loc(#loc257) + %q_8 = arith.constant 1 : i64 loc(#loc257) + %q_9 = arith.divsi %5, %q_8 : i64 loc(#loc257) + %q_10 = arith.addi %q_6, %q_9 : i64 loc(#loc258) + %q_11 = arith.constant 128 : i32 loc(#loc259) + %q_12 = arith.constant 128 : i64 loc(#loc259) + %q_13 = arith.muli %q_10, %q_12 : i64 loc(#loc259) + %q_14 = tt.addptr %q, %q_13 : !tt.ptr, i64 loc(#loc260) + %k_15 = arith.constant 32 : i32 loc(#loc261) + %k_16 = arith.constant 32 : i64 loc(#loc261) + %k_17 = arith.muli %7, %k_16 : i64 loc(#loc261) + %k_18 = arith.constant 1 : i32 loc(#loc262) + %k_19 = arith.constant 1 : i64 loc(#loc262) + %k_20 = arith.divsi %5, %k_19 : i64 loc(#loc262) + %k_21 = arith.addi %k_17, %k_20 : i64 loc(#loc263) + %k_22 = arith.constant 128 : i32 loc(#loc264) + %k_23 = arith.constant 128 : i64 loc(#loc264) + %k_24 = arith.muli %k_21, %k_23 : i64 loc(#loc264) + %k_25 = tt.addptr %k, %k_24 : !tt.ptr, i64 loc(#loc265) + %w_26 = arith.constant 32 : i32 loc(#loc266) + %w_27 = arith.constant 32 : i64 loc(#loc266) + %w_28 = arith.muli %7, %w_27 : i64 loc(#loc266) + %w_29 = arith.addi %w_28, %5 : i64 loc(#loc267) + %w_30 = arith.constant 128 : i32 loc(#loc268) + %w_31 = arith.constant 128 : i64 loc(#loc268) + %w_32 = arith.muli %w_29, %w_31 : i64 loc(#loc268) + %w_33 = tt.addptr %w, %w_32 : !tt.ptr, i64 loc(#loc269) + %do_34 = arith.constant 32 : i32 loc(#loc270) + %do_35 = arith.constant 32 : i64 loc(#loc270) + %do_36 = arith.muli %7, %do_35 : i64 loc(#loc270) + %do_37 = arith.addi %do_36, %5 : i64 loc(#loc271) + %do_38 = arith.constant 128 : i32 loc(#loc272) + %do_39 = arith.constant 128 : i64 loc(#loc272) + %do_40 = arith.muli %do_37, %do_39 : i64 loc(#loc272) + %do_41 = tt.addptr %do, %do_40 : !tt.ptr, i64 loc(#loc273) + %dv_42 = arith.constant 32 : i32 loc(#loc274) + %dv_43 = arith.constant 32 : i64 loc(#loc274) + %dv_44 = arith.muli %7, %dv_43 : i64 loc(#loc274) + %dv_45 = arith.addi %dv_44, %5 : i64 loc(#loc275) + %dv_46 = arith.constant 128 : i32 loc(#loc276) + %dv_47 = arith.constant 128 : i64 loc(#loc276) + %dv_48 = arith.muli %dv_45, %dv_47 : i64 loc(#loc276) + %dv_49 = tt.addptr %dv, %dv_48 : !tt.ptr, i64 loc(#loc277) + %dv2_50 = arith.constant 32 : i32 loc(#loc278) + %dv2_51 = arith.constant 32 : i64 loc(#loc278) + %dv2_52 = arith.muli %7, %dv2_51 : i64 loc(#loc278) + %dv2_53 = arith.addi %dv2_52, %5 : i64 loc(#loc279) + %dv2_54 = arith.constant 128 : i32 loc(#loc280) + %dv2_55 = arith.constant 128 : i64 loc(#loc280) + %dv2_56 = arith.muli %dv2_53, %dv2_55 : i64 loc(#loc280) + %dv2_57 = tt.addptr %dv2, %dv2_56 : !tt.ptr, i64 loc(#loc281) + %dh_58 = arith.constant 32 : i32 loc(#loc282) + %dh_59 = arith.constant 32 : i64 loc(#loc282) + %dh_60 = arith.muli %boh_3, %dh_59 : i64 loc(#loc282) + %dh_61 = arith.addi %dh_60, %5 : i64 loc(#loc283) + %dh_62 = arith.constant 128 : i32 loc(#loc284) + %dh_63 = arith.constant 128 : i64 loc(#loc284) + %dh_64 = arith.muli %dh_61, %dh_63 : i64 loc(#loc284) + %dh_65 = arith.constant 128 : i32 loc(#loc285) + %dh_66 = arith.constant 128 : i64 loc(#loc285) + %dh_67 = arith.muli %dh_64, %dh_66 : i64 loc(#loc285) + %dh_68 = tt.addptr %dh, %dh_67 : !tt.ptr, i64 loc(#loc286) + %o_v = arith.constant 64 : i32 loc(#loc287) + %o_v_69 = arith.constant 64 : i64 loc(#loc287) + %o_v_70 = arith.muli %1, %o_v_69 : i64 loc(#loc287) + %o_v_71 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc288) + %o_v_72 = arith.extsi %o_v_71 : tensor<64xi32> to tensor<64xi64> loc(#loc289) + %o_v_73 = tt.splat %o_v_70 : i64 -> tensor<64xi64> loc(#loc289) + %o_v_74 = arith.addi %o_v_73, %o_v_72 : tensor<64xi64> loc(#loc289) + %m_v = arith.constant 128 : i32 loc(#loc290) + %m_v_75 = arith.extsi %m_v : i32 to i64 loc(#loc290) + %m_v_76 = tt.splat %m_v_75 : i64 -> tensor<64xi64> loc(#loc290) + %m_v_77 = arith.cmpi slt, %o_v_74, %m_v_76 : tensor<64xi64> loc(#loc290) + %o_k1 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc291) + %m_k1 = arith.constant 128 : i32 loc(#loc292) + %m_k1_78 = arith.constant dense<128> : tensor<64xi32> loc(#loc292) + %m_k1_79 = arith.cmpi slt, %o_k1, %m_k1_78 : tensor<64xi32> loc(#loc292) + %o_k2 = arith.constant 64 : i32 loc(#loc293) + %o_k2_80 = arith.constant 64 : i32 loc(#loc293) + %o_k2_81 = arith.constant dense<64> : tensor<64xi32> loc(#loc293) + %o_k2_82 = arith.extsi %o_k2_81 : tensor<64xi32> to tensor<64xi64> loc(#loc293) + %o_k2_83 = arith.extsi %o_k1 : tensor<64xi32> to tensor<64xi64> loc(#loc293) + %o_k2_84 = arith.addi %o_k2_82, %o_k2_83 : tensor<64xi64> loc(#loc293) + %o_k2_85 = arith.constant 2147483647 : i64 loc(#loc293) + %o_k2_86 = arith.constant -2147483648 : i64 loc(#loc293) + %o_k2_87 = arith.constant dense<2147483647> : tensor<64xi64> loc(#loc293) + %o_k2_88 = arith.cmpi sle, %o_k2_84, %o_k2_87 : tensor<64xi64> loc(#loc293) + %o_k2_89 = arith.constant dense<-2147483648> : tensor<64xi64> loc(#loc293) + %o_k2_90 = arith.cmpi sge, %o_k2_84, %o_k2_89 : tensor<64xi64> loc(#loc293) + %o_k2_91 = arith.andi %o_k2_88, %o_k2_90 : tensor<64xi1> loc(#loc293) + %o_k2_92 = arith.addi %o_k2_81, %o_k1 : tensor<64xi32> loc(#loc293) + %m_k2 = arith.constant 128 : i32 loc(#loc294) + %m_k2_93 = arith.constant dense<128> : tensor<64xi32> loc(#loc294) + %m_k2_94 = arith.cmpi slt, %o_k2_92, %m_k2_93 : tensor<64xi32> loc(#loc294) + %o_k3 = arith.constant 128 : i32 loc(#loc295) + %o_k3_95 = arith.constant 128 : i32 loc(#loc295) + %o_k3_96 = arith.constant dense<128> : tensor<64xi32> loc(#loc295) + %o_k3_97 = arith.extsi %o_k3_96 : tensor<64xi32> to tensor<64xi64> loc(#loc295) + %o_k3_98 = arith.extsi %o_k1 : tensor<64xi32> to tensor<64xi64> loc(#loc295) + %o_k3_99 = arith.addi %o_k3_97, %o_k3_98 : tensor<64xi64> loc(#loc295) + %o_k3_100 = arith.constant 2147483647 : i64 loc(#loc295) + %o_k3_101 = arith.constant -2147483648 : i64 loc(#loc295) + %o_k3_102 = arith.constant dense<2147483647> : tensor<64xi64> loc(#loc295) + %o_k3_103 = arith.cmpi sle, %o_k3_99, %o_k3_102 : tensor<64xi64> loc(#loc295) + %o_k3_104 = arith.constant dense<-2147483648> : tensor<64xi64> loc(#loc295) + %o_k3_105 = arith.cmpi sge, %o_k3_99, %o_k3_104 : tensor<64xi64> loc(#loc295) + %o_k3_106 = arith.andi %o_k3_103, %o_k3_105 : tensor<64xi1> loc(#loc295) + %o_k3_107 = arith.addi %o_k3_96, %o_k1 : tensor<64xi32> loc(#loc295) + %m_k3 = arith.constant 128 : i32 loc(#loc296) + %m_k3_108 = arith.constant dense<128> : tensor<64xi32> loc(#loc296) + %m_k3_109 = arith.cmpi slt, %o_k3_107, %m_k3_108 : tensor<64xi32> loc(#loc296) + %o_k4 = arith.constant 192 : i32 loc(#loc297) + %o_k4_110 = arith.constant 192 : i32 loc(#loc297) + %o_k4_111 = arith.constant dense<192> : tensor<64xi32> loc(#loc297) + %o_k4_112 = arith.extsi %o_k4_111 : tensor<64xi32> to tensor<64xi64> loc(#loc297) + %o_k4_113 = arith.extsi %o_k1 : tensor<64xi32> to tensor<64xi64> loc(#loc297) + %o_k4_114 = arith.addi %o_k4_112, %o_k4_113 : tensor<64xi64> loc(#loc297) + %o_k4_115 = arith.constant 2147483647 : i64 loc(#loc297) + %o_k4_116 = arith.constant -2147483648 : i64 loc(#loc297) + %o_k4_117 = arith.constant dense<2147483647> : tensor<64xi64> loc(#loc297) + %o_k4_118 = arith.cmpi sle, %o_k4_114, %o_k4_117 : tensor<64xi64> loc(#loc297) + %o_k4_119 = arith.constant dense<-2147483648> : tensor<64xi64> loc(#loc297) + %o_k4_120 = arith.cmpi sge, %o_k4_114, %o_k4_119 : tensor<64xi64> loc(#loc297) + %o_k4_121 = arith.andi %o_k4_118, %o_k4_120 : tensor<64xi1> loc(#loc297) + %o_k4_122 = arith.addi %o_k4_111, %o_k1 : tensor<64xi32> loc(#loc297) + %m_k4 = arith.constant 128 : i32 loc(#loc298) + %m_k4_123 = arith.constant dense<128> : tensor<64xi32> loc(#loc298) + %m_k4_124 = arith.cmpi slt, %o_k4_122, %m_k4_123 : tensor<64xi32> loc(#loc298) + %c1_i32 = arith.constant 1 : i32 loc(#loc59) + %c1_i32_125 = arith.constant 1 : i32 loc(#loc59) + %12 = arith.extsi %NT : i32 to i64 loc(#loc59) + %13 = arith.extsi %c1_i32_125 : i32 to i64 loc(#loc59) + %14 = arith.subi %12, %13 : i64 loc(#loc59) + %c2147483647_i64 = arith.constant 2147483647 : i64 loc(#loc59) + %c-2147483648_i64 = arith.constant -2147483648 : i64 loc(#loc59) + %15 = arith.cmpi sle, %14, %c2147483647_i64 : i64 loc(#loc59) + %16 = arith.cmpi sge, %14, %c-2147483648_i64 : i64 loc(#loc59) + %17 = arith.andi %15, %16 : i1 loc(#loc59) + %18 = arith.subi %NT, %c1_i32_125 : i32 loc(#loc59) + %c-1_i32 = arith.constant -1 : i32 loc(#loc60) + %c1_i32_126 = arith.constant 1 : i32 loc(#loc60) + %19 = arith.bitcast %c-1_i32 : i32 to i32 loc(#loc60) + %20 = arith.bitcast %18 : i32 to i32 loc(#loc60) + %21 = arith.bitcast %c1_i32_126 : i32 to i32 loc(#loc60) + %22 = ub.poison : i32 loc(#loc60) + %b_dh2_127:2 = scf.for %arg10 = %19 to %20 step %21 iter_args(%b_dh1_128 = %b_dh1, %b_dh2_129 = %b_dh2) -> (tensor<64x64xf32>, tensor<64x64xf32>) : i32 { + %i_t_int64 = arith.subi %20, %arg10 : i32 loc(#loc300) + %i_t = arith.addi %i_t_int64, %19 : i32 loc(#loc456) + %i_t_int64_130 = arith.extsi %i_t : i32 to i64 loc(#loc300) + %o_t = arith.constant 64 : i32 loc(#loc301) + %o_t_131 = arith.constant 64 : i64 loc(#loc301) + %o_t_132 = arith.muli %i_t_int64_130, %o_t_131 : i64 loc(#loc301) + %o_t_133 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc302) + %o_t_134 = arith.extsi %o_t_133 : tensor<64xi32> to tensor<64xi64> loc(#loc303) + %o_t_135 = tt.splat %o_t_132 : i64 -> tensor<64xi64> loc(#loc303) + %o_t_136 = arith.addi %o_t_135, %o_t_134 : tensor<64xi64> loc(#loc303) + %m_t = arith.extsi %T : i32 to i64 loc(#loc304) + %m_t_137 = tt.splat %m_t : i64 -> tensor<64xi64> loc(#loc304) + %m_t_138 = arith.cmpi slt, %o_t_136, %m_t_137 : tensor<64xi64> loc(#loc304) + %p_dh1 = arith.constant 32 : i32 loc(#loc305) + %p_dh1_139 = arith.constant 32 : i64 loc(#loc305) + %p_dh1_140 = arith.muli %i_t_int64_130, %p_dh1_139 : i64 loc(#loc305) + %p_dh1_141 = arith.constant 128 : i32 loc(#loc306) + %p_dh1_142 = arith.constant 128 : i64 loc(#loc306) + %p_dh1_143 = arith.muli %p_dh1_140, %p_dh1_142 : i64 loc(#loc306) + %p_dh1_144 = arith.constant 128 : i32 loc(#loc307) + %p_dh1_145 = arith.constant 128 : i64 loc(#loc307) + %p_dh1_146 = arith.muli %p_dh1_143, %p_dh1_145 : i64 loc(#loc307) + %p_dh1_147 = tt.addptr %dh_68, %p_dh1_146 : !tt.ptr, i64 loc(#loc308) + %p_dh1_148 = tt.expand_dims %o_k1 {axis = 1 : i32} : tensor<64xi32> -> tensor<64x1xi32> loc(#loc309) + %p_dh1_149 = arith.constant 128 : i32 loc(#loc310) + %p_dh1_150 = arith.constant 128 : i32 loc(#loc310) + %p_dh1_151 = arith.constant dense<128> : tensor<64x1xi32> loc(#loc310) + %p_dh1_152 = arith.extsi %p_dh1_148 : tensor<64x1xi32> to tensor<64x1xi64> loc(#loc310) + %p_dh1_153 = arith.extsi %p_dh1_151 : tensor<64x1xi32> to tensor<64x1xi64> loc(#loc310) + %p_dh1_154 = arith.muli %p_dh1_152, %p_dh1_153 : tensor<64x1xi64> loc(#loc310) + %p_dh1_155 = arith.constant 2147483647 : i64 loc(#loc310) + %p_dh1_156 = arith.constant -2147483648 : i64 loc(#loc310) + %p_dh1_157 = arith.constant dense<2147483647> : tensor<64x1xi64> loc(#loc310) + %p_dh1_158 = arith.cmpi sle, %p_dh1_154, %p_dh1_157 : tensor<64x1xi64> loc(#loc310) + %p_dh1_159 = arith.constant dense<-2147483648> : tensor<64x1xi64> loc(#loc310) + %p_dh1_160 = arith.cmpi sge, %p_dh1_154, %p_dh1_159 : tensor<64x1xi64> loc(#loc310) + %p_dh1_161 = arith.andi %p_dh1_158, %p_dh1_160 : tensor<64x1xi1> loc(#loc310) + %p_dh1_162 = arith.muli %p_dh1_148, %p_dh1_151 : tensor<64x1xi32> loc(#loc310) + %p_dh1_163 = tt.splat %p_dh1_147 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc311) + %p_dh1_164 = tt.addptr %p_dh1_163, %p_dh1_162 : tensor<64x1x!tt.ptr>, tensor<64x1xi32> loc(#loc311) + %p_dh1_165 = tt.expand_dims %o_v_74 {axis = 0 : i32} : tensor<64xi64> -> tensor<1x64xi64> loc(#loc312) + %p_dh1_166 = tt.broadcast %p_dh1_164 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc313) + %p_dh1_167 = tt.broadcast %p_dh1_165 : tensor<1x64xi64> -> tensor<64x64xi64> loc(#loc313) + %p_dh1_168 = tt.addptr %p_dh1_166, %p_dh1_167 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc313) + %m_dh1 = tt.expand_dims %m_k1_79 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc314) + %m_dh1_169 = tt.expand_dims %m_v_77 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc315) + %m_dh1_170 = tt.broadcast %m_dh1 : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc316) + %m_dh1_171 = tt.broadcast %m_dh1_169 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc316) + %m_dh1_172 = arith.andi %m_dh1_170, %m_dh1_171 : tensor<64x64xi1> loc(#loc316) + %23 = arith.truncf %b_dh1_128 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc78) + tt.store %p_dh1_168, %23, %m_dh1_172 : tensor<64x64x!tt.ptr> loc(#loc79) + %p_dh2 = arith.constant 32 : i32 loc(#loc317) + %p_dh2_173 = arith.constant 32 : i64 loc(#loc317) + %p_dh2_174 = arith.muli %i_t_int64_130, %p_dh2_173 : i64 loc(#loc317) + %p_dh2_175 = arith.constant 128 : i32 loc(#loc318) + %p_dh2_176 = arith.constant 128 : i64 loc(#loc318) + %p_dh2_177 = arith.muli %p_dh2_174, %p_dh2_176 : i64 loc(#loc318) + %p_dh2_178 = arith.constant 128 : i32 loc(#loc319) + %p_dh2_179 = arith.constant 128 : i64 loc(#loc319) + %p_dh2_180 = arith.muli %p_dh2_177, %p_dh2_179 : i64 loc(#loc319) + %p_dh2_181 = tt.addptr %dh_68, %p_dh2_180 : !tt.ptr, i64 loc(#loc320) + %p_dh2_182 = tt.expand_dims %o_k2_92 {axis = 1 : i32} : tensor<64xi32> -> tensor<64x1xi32> loc(#loc321) + %p_dh2_183 = arith.constant 128 : i32 loc(#loc322) + %p_dh2_184 = arith.constant 128 : i32 loc(#loc322) + %p_dh2_185 = arith.constant dense<128> : tensor<64x1xi32> loc(#loc322) + %p_dh2_186 = arith.extsi %p_dh2_182 : tensor<64x1xi32> to tensor<64x1xi64> loc(#loc322) + %p_dh2_187 = arith.extsi %p_dh2_185 : tensor<64x1xi32> to tensor<64x1xi64> loc(#loc322) + %p_dh2_188 = arith.muli %p_dh2_186, %p_dh2_187 : tensor<64x1xi64> loc(#loc322) + %p_dh2_189 = arith.constant 2147483647 : i64 loc(#loc322) + %p_dh2_190 = arith.constant -2147483648 : i64 loc(#loc322) + %p_dh2_191 = arith.constant dense<2147483647> : tensor<64x1xi64> loc(#loc322) + %p_dh2_192 = arith.cmpi sle, %p_dh2_188, %p_dh2_191 : tensor<64x1xi64> loc(#loc322) + %p_dh2_193 = arith.constant dense<-2147483648> : tensor<64x1xi64> loc(#loc322) + %p_dh2_194 = arith.cmpi sge, %p_dh2_188, %p_dh2_193 : tensor<64x1xi64> loc(#loc322) + %p_dh2_195 = arith.andi %p_dh2_192, %p_dh2_194 : tensor<64x1xi1> loc(#loc322) + %p_dh2_196 = arith.muli %p_dh2_182, %p_dh2_185 : tensor<64x1xi32> loc(#loc322) + %p_dh2_197 = tt.splat %p_dh2_181 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc323) + %p_dh2_198 = tt.addptr %p_dh2_197, %p_dh2_196 : tensor<64x1x!tt.ptr>, tensor<64x1xi32> loc(#loc323) + %p_dh2_199 = tt.expand_dims %o_v_74 {axis = 0 : i32} : tensor<64xi64> -> tensor<1x64xi64> loc(#loc324) + %p_dh2_200 = tt.broadcast %p_dh2_198 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc325) + %p_dh2_201 = tt.broadcast %p_dh2_199 : tensor<1x64xi64> -> tensor<64x64xi64> loc(#loc325) + %p_dh2_202 = tt.addptr %p_dh2_200, %p_dh2_201 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc325) + %m_dh2 = tt.expand_dims %m_k2_94 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc326) + %m_dh2_203 = tt.expand_dims %m_v_77 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc327) + %m_dh2_204 = tt.broadcast %m_dh2 : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc328) + %m_dh2_205 = tt.broadcast %m_dh2_203 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc328) + %m_dh2_206 = arith.andi %m_dh2_204, %m_dh2_205 : tensor<64x64xi1> loc(#loc328) + %24 = arith.truncf %b_dh2_129 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc92) + tt.store %p_dh2_202, %24, %m_dh2_206 : tensor<64x64x!tt.ptr> loc(#loc93) + %last_idx = arith.constant 1 : i32 loc(#loc329) + %last_idx_207 = arith.constant 1 : i64 loc(#loc329) + %last_idx_208 = arith.addi %i_t_int64_130, %last_idx_207 : i64 loc(#loc329) + %last_idx_209 = arith.constant 64 : i32 loc(#loc330) + %last_idx_210 = arith.constant 64 : i64 loc(#loc330) + %last_idx_211 = arith.muli %last_idx_208, %last_idx_210 : i64 loc(#loc330) + %last_idx_212 = arith.extsi %T : i32 to i64 loc(#loc331) + %last_idx_213 = arith.minsi %last_idx_211, %last_idx_212 : i64 loc(#loc331) + %last_idx_214 = arith.constant 1 : i32 loc(#loc332) + %last_idx_215 = arith.constant 1 : i64 loc(#loc332) + %last_idx_216 = arith.subi %last_idx_213, %last_idx_215 : i64 loc(#loc332) + %bg_last = arith.addi %7, %last_idx_216 : i64 loc(#loc333) + %bg_last_217 = arith.constant 32 : i32 loc(#loc334) + %bg_last_218 = arith.constant 32 : i64 loc(#loc334) + %bg_last_219 = arith.muli %bg_last, %bg_last_218 : i64 loc(#loc334) + %bg_last_220 = tt.addptr %g, %bg_last_219 : !tt.ptr, i64 loc(#loc335) + %bg_last_221 = tt.addptr %bg_last_220, %5 : !tt.ptr, i64 loc(#loc336) + %bg_last_222 = tt.load %bg_last_221 : !tt.ptr loc(#loc337) + %p_g = arith.constant 32 : i32 loc(#loc338) + %p_g_223 = arith.constant 32 : i64 loc(#loc338) + %p_g_224 = arith.muli %7, %p_g_223 : i64 loc(#loc338) + %p_g_225 = tt.addptr %g, %p_g_224 : !tt.ptr, i64 loc(#loc339) + %p_g_226 = tt.addptr %p_g_225, %5 : !tt.ptr, i64 loc(#loc340) + %p_g_227 = arith.constant 32 : i32 loc(#loc341) + %p_g_228 = arith.constant 32 : i64 loc(#loc341) + %p_g_229 = arith.constant dense<32> : tensor<64xi64> loc(#loc341) + %p_g_230 = arith.muli %o_t_136, %p_g_229 : tensor<64xi64> loc(#loc341) + %p_g_231 = tt.splat %p_g_226 : !tt.ptr -> tensor<64x!tt.ptr> loc(#loc342) + %p_g_232 = tt.addptr %p_g_231, %p_g_230 : tensor<64x!tt.ptr>, tensor<64xi64> loc(#loc342) + %b_g = arith.constant 0.000000e+00 : f32 loc(#loc343) + %b_g_233 = arith.constant dense<0.000000e+00> : tensor<64xf32> loc(#loc343) + %b_g_234 = tt.load %p_g_232, %m_t_138, %b_g_233 : tensor<64x!tt.ptr> loc(#loc343) + %bg_last_exp = tt.call @fla.ops.utils.op.exp2__fp32(%bg_last_222) : (f32) -> f32 loc(#loc344) + %b_g_exp = tt.call @fla.ops.utils.op.exp2__fp32S64S(%b_g_234) : (tensor<64xf32>) -> tensor<64xf32> loc(#loc345) + %p_dv = tt.expand_dims %o_t_136 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc346) + %p_dv_235 = arith.constant 4096 : i32 loc(#loc347) + %p_dv_236 = arith.constant 4096 : i64 loc(#loc347) + %p_dv_237 = arith.constant dense<4096> : tensor<64x1xi64> loc(#loc347) + %p_dv_238 = arith.muli %p_dv, %p_dv_237 : tensor<64x1xi64> loc(#loc347) + %p_dv_239 = tt.splat %dv_49 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc348) + %p_dv_240 = tt.addptr %p_dv_239, %p_dv_238 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc348) + %p_dv_241 = tt.expand_dims %o_v_74 {axis = 0 : i32} : tensor<64xi64> -> tensor<1x64xi64> loc(#loc349) + %p_dv_242 = tt.broadcast %p_dv_240 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc350) + %p_dv_243 = tt.broadcast %p_dv_241 : tensor<1x64xi64> -> tensor<64x64xi64> loc(#loc350) + %p_dv_244 = tt.addptr %p_dv_242, %p_dv_243 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc350) + %p_dv2 = tt.expand_dims %o_t_136 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc351) + %p_dv2_245 = arith.constant 4096 : i32 loc(#loc352) + %p_dv2_246 = arith.constant 4096 : i64 loc(#loc352) + %p_dv2_247 = arith.constant dense<4096> : tensor<64x1xi64> loc(#loc352) + %p_dv2_248 = arith.muli %p_dv2, %p_dv2_247 : tensor<64x1xi64> loc(#loc352) + %p_dv2_249 = tt.splat %dv2_57 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc353) + %p_dv2_250 = tt.addptr %p_dv2_249, %p_dv2_248 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc353) + %p_dv2_251 = tt.expand_dims %o_v_74 {axis = 0 : i32} : tensor<64xi64> -> tensor<1x64xi64> loc(#loc354) + %p_dv2_252 = tt.broadcast %p_dv2_250 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc355) + %p_dv2_253 = tt.broadcast %p_dv2_251 : tensor<1x64xi64> -> tensor<64x64xi64> loc(#loc355) + %p_dv2_254 = tt.addptr %p_dv2_252, %p_dv2_253 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc355) + %p_do = tt.expand_dims %o_t_136 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc356) + %p_do_255 = arith.constant 4096 : i32 loc(#loc357) + %p_do_256 = arith.constant 4096 : i64 loc(#loc357) + %p_do_257 = arith.constant dense<4096> : tensor<64x1xi64> loc(#loc357) + %p_do_258 = arith.muli %p_do, %p_do_257 : tensor<64x1xi64> loc(#loc357) + %p_do_259 = tt.splat %do_41 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc358) + %p_do_260 = tt.addptr %p_do_259, %p_do_258 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc358) + %p_do_261 = tt.expand_dims %o_v_74 {axis = 0 : i32} : tensor<64xi64> -> tensor<1x64xi64> loc(#loc359) + %p_do_262 = tt.broadcast %p_do_260 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc360) + %p_do_263 = tt.broadcast %p_do_261 : tensor<1x64xi64> -> tensor<64x64xi64> loc(#loc360) + %p_do_264 = tt.addptr %p_do_262, %p_do_263 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc360) + %b_do = tt.expand_dims %m_t_138 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc361) + %b_do_265 = tt.expand_dims %m_v_77 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc362) + %b_do_266 = tt.broadcast %b_do : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc363) + %b_do_267 = tt.broadcast %b_do_265 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc363) + %b_do_268 = arith.andi %b_do_266, %b_do_267 : tensor<64x64xi1> loc(#loc363) + %b_do_269 = arith.constant 0.000000e+00 : f32 loc(#loc364) + %b_do_270 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc364) + %b_do_271 = arith.truncf %b_do_270 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc364) + %b_do_272 = tt.load %p_do_264, %b_do_268, %b_do_271 : tensor<64x64x!tt.ptr> loc(#loc364) + %p_k = tt.expand_dims %o_t_136 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc365) + %p_k_273 = arith.constant 4096 : i32 loc(#loc366) + %p_k_274 = arith.constant 4096 : i64 loc(#loc366) + %p_k_275 = arith.constant dense<4096> : tensor<64x1xi64> loc(#loc366) + %p_k_276 = arith.muli %p_k, %p_k_275 : tensor<64x1xi64> loc(#loc366) + %p_k_277 = tt.splat %k_25 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc367) + %p_k_278 = tt.addptr %p_k_277, %p_k_276 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc367) + %p_k_279 = tt.expand_dims %o_k1 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc368) + %p_k_280 = tt.broadcast %p_k_278 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc369) + %p_k_281 = tt.broadcast %p_k_279 : tensor<1x64xi32> -> tensor<64x64xi32> loc(#loc369) + %p_k_282 = tt.addptr %p_k_280, %p_k_281 : tensor<64x64x!tt.ptr>, tensor<64x64xi32> loc(#loc369) + %b_k = tt.expand_dims %m_t_138 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc370) + %b_k_283 = tt.expand_dims %m_k1_79 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc371) + %b_k_284 = tt.broadcast %b_k : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc372) + %b_k_285 = tt.broadcast %b_k_283 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc372) + %b_k_286 = arith.andi %b_k_284, %b_k_285 : tensor<64x64xi1> loc(#loc372) + %b_k_287 = arith.constant 0.000000e+00 : f32 loc(#loc373) + %b_k_288 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc373) + %b_k_289 = arith.truncf %b_k_288 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc373) + %b_k_290 = tt.load %p_k_282, %b_k_286, %b_k_289 : tensor<64x64x!tt.ptr> loc(#loc373) + %b_dv = arith.truncf %b_dh1_128 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc374) + %b_dv_291 = arith.constant 0.000000e+00 : f32 loc(#loc375) + %b_dv_292 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc375) + %b_dv_293 = tt.dot %b_k_290, %b_dv, %b_dv_292, inputPrecision = tf32 : tensor<64x64xbf16> * tensor<64x64xbf16> -> tensor<64x64xf32> loc(#loc375) + %p_k_294 = tt.expand_dims %o_t_136 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc376) + %p_k_295 = arith.constant 4096 : i32 loc(#loc377) + %p_k_296 = arith.constant 4096 : i64 loc(#loc377) + %p_k_297 = arith.constant dense<4096> : tensor<64x1xi64> loc(#loc377) + %p_k_298 = arith.muli %p_k_294, %p_k_297 : tensor<64x1xi64> loc(#loc377) + %p_k_299 = tt.splat %k_25 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc378) + %p_k_300 = tt.addptr %p_k_299, %p_k_298 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc378) + %p_k_301 = tt.expand_dims %o_k2_92 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc379) + %p_k_302 = tt.broadcast %p_k_300 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc380) + %p_k_303 = tt.broadcast %p_k_301 : tensor<1x64xi32> -> tensor<64x64xi32> loc(#loc380) + %p_k_304 = tt.addptr %p_k_302, %p_k_303 : tensor<64x64x!tt.ptr>, tensor<64x64xi32> loc(#loc380) + %b_k_305 = tt.expand_dims %m_t_138 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc381) + %b_k_306 = tt.expand_dims %m_k2_94 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc382) + %b_k_307 = tt.broadcast %b_k_305 : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc383) + %b_k_308 = tt.broadcast %b_k_306 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc383) + %b_k_309 = arith.andi %b_k_307, %b_k_308 : tensor<64x64xi1> loc(#loc383) + %b_k_310 = arith.constant 0.000000e+00 : f32 loc(#loc384) + %b_k_311 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc384) + %b_k_312 = arith.truncf %b_k_311 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc384) + %b_k_313 = tt.load %p_k_304, %b_k_309, %b_k_312 : tensor<64x64x!tt.ptr> loc(#loc384) + %b_dv_314 = arith.truncf %b_dh2_129 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc385) + %b_dv_315 = arith.constant 0.000000e+00 : f32 loc(#loc386) + %b_dv_316 = tt.dot %b_k_313, %b_dv_314, %b_dv_293, inputPrecision = tf32 : tensor<64x64xbf16> * tensor<64x64xbf16> -> tensor<64x64xf32> loc(#loc386) + %b_dv_317 = tt.splat %bg_last_222 : f32 -> tensor<64xf32> loc(#loc387) + %b_dv_318 = arith.subf %b_dv_317, %b_g_234 : tensor<64xf32> loc(#loc387) + %b_dv_319 = tt.call @fla.ops.utils.op.exp2__fp32S64S(%b_dv_318) : (tensor<64xf32>) -> tensor<64xf32> loc(#loc388) + %b_dv_320 = arith.constant 0 : i32 loc(#loc389) + %b_dv_321 = arith.constant 0.000000e+00 : f32 loc(#loc389) + %b_dv_322 = arith.constant dense<0.000000e+00> : tensor<64xf32> loc(#loc389) + %b_dv_323 = arith.select %m_t_138, %b_dv_319, %b_dv_322 : tensor<64xi1>, tensor<64xf32> loc(#loc389) + %b_dv_324 = tt.expand_dims %b_dv_323 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc390) + %b_dv_325 = tt.broadcast %b_dv_324 : tensor<64x1xf32> -> tensor<64x64xf32> loc(#loc391) + %b_dv_326 = arith.mulf %b_dv_316, %b_dv_325 : tensor<64x64xf32> loc(#loc391) + %b_dv_327 = tt.expand_dims %m_t_138 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc392) + %b_dv_328 = tt.expand_dims %m_v_77 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc393) + %b_dv_329 = tt.broadcast %b_dv_327 : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc394) + %b_dv_330 = tt.broadcast %b_dv_328 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc394) + %b_dv_331 = arith.andi %b_dv_329, %b_dv_330 : tensor<64x64xi1> loc(#loc394) + %b_dv_332 = arith.constant 0.000000e+00 : f32 loc(#loc395) + %b_dv_333 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc395) + %b_dv_334 = arith.truncf %b_dv_333 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc395) + %b_dv_335 = tt.load %p_dv_244, %b_dv_331, %b_dv_334 : tensor<64x64x!tt.ptr> loc(#loc395) + %b_dv_336 = arith.extf %b_dv_335 : tensor<64x64xbf16> to tensor<64x64xf32> loc(#loc396) + %b_dv_337 = arith.addf %b_dv_326, %b_dv_336 : tensor<64x64xf32> loc(#loc396) + %25 = tt.expand_dims %m_t_138 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc162) + %26 = tt.expand_dims %m_v_77 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc163) + %27 = tt.broadcast %25 : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc164) + %28 = tt.broadcast %26 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc164) + %29 = arith.andi %27, %28 : tensor<64x64xi1> loc(#loc164) + %30 = arith.truncf %b_dv_337 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc165) + tt.store %p_dv2_254, %30, %29 : tensor<64x64x!tt.ptr> loc(#loc166) + %p_w = tt.expand_dims %o_k1 {axis = 1 : i32} : tensor<64xi32> -> tensor<64x1xi32> loc(#loc397) + %p_w_338 = tt.splat %w_33 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc398) + %p_w_339 = tt.addptr %p_w_338, %p_w : tensor<64x1x!tt.ptr>, tensor<64x1xi32> loc(#loc398) + %p_w_340 = tt.expand_dims %o_t_136 {axis = 0 : i32} : tensor<64xi64> -> tensor<1x64xi64> loc(#loc399) + %p_w_341 = arith.constant 4096 : i32 loc(#loc400) + %p_w_342 = arith.constant 4096 : i64 loc(#loc400) + %p_w_343 = arith.constant dense<4096> : tensor<1x64xi64> loc(#loc400) + %p_w_344 = arith.muli %p_w_340, %p_w_343 : tensor<1x64xi64> loc(#loc400) + %p_w_345 = tt.broadcast %p_w_339 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc401) + %p_w_346 = tt.broadcast %p_w_344 : tensor<1x64xi64> -> tensor<64x64xi64> loc(#loc401) + %p_w_347 = tt.addptr %p_w_345, %p_w_346 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc401) + %p_q = tt.expand_dims %o_k1 {axis = 1 : i32} : tensor<64xi32> -> tensor<64x1xi32> loc(#loc402) + %p_q_348 = tt.splat %q_14 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc403) + %p_q_349 = tt.addptr %p_q_348, %p_q : tensor<64x1x!tt.ptr>, tensor<64x1xi32> loc(#loc403) + %p_q_350 = tt.expand_dims %o_t_136 {axis = 0 : i32} : tensor<64xi64> -> tensor<1x64xi64> loc(#loc404) + %p_q_351 = arith.constant 4096 : i32 loc(#loc405) + %p_q_352 = arith.constant 4096 : i64 loc(#loc405) + %p_q_353 = arith.constant dense<4096> : tensor<1x64xi64> loc(#loc405) + %p_q_354 = arith.muli %p_q_350, %p_q_353 : tensor<1x64xi64> loc(#loc405) + %p_q_355 = tt.broadcast %p_q_349 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc406) + %p_q_356 = tt.broadcast %p_q_354 : tensor<1x64xi64> -> tensor<64x64xi64> loc(#loc406) + %p_q_357 = tt.addptr %p_q_355, %p_q_356 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc406) + %b_w = tt.expand_dims %m_k1_79 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc407) + %b_w_358 = tt.expand_dims %m_t_138 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc408) + %b_w_359 = tt.broadcast %b_w : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc409) + %b_w_360 = tt.broadcast %b_w_358 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc409) + %b_w_361 = arith.andi %b_w_359, %b_w_360 : tensor<64x64xi1> loc(#loc409) + %b_w_362 = arith.constant 0.000000e+00 : f32 loc(#loc410) + %b_w_363 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc410) + %b_w_364 = arith.truncf %b_w_363 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc410) + %b_w_365 = tt.load %p_w_347, %b_w_361, %b_w_364 : tensor<64x64x!tt.ptr> loc(#loc410) + %b_q = tt.expand_dims %m_k1_79 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc411) + %b_q_366 = tt.expand_dims %m_t_138 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc412) + %b_q_367 = tt.broadcast %b_q : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc413) + %b_q_368 = tt.broadcast %b_q_366 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc413) + %b_q_369 = arith.andi %b_q_367, %b_q_368 : tensor<64x64xi1> loc(#loc413) + %b_q_370 = arith.constant 0.000000e+00 : f32 loc(#loc414) + %b_q_371 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc414) + %b_q_372 = arith.truncf %b_q_371 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc414) + %b_q_373 = tt.load %p_q_357, %b_q_369, %b_q_372 : tensor<64x64x!tt.ptr> loc(#loc414) + %b_dh1_374 = tt.splat %bg_last_exp : f32 -> tensor<64x64xf32> loc(#loc415) + %b_dh1_375 = arith.mulf %b_dh1_128, %b_dh1_374 : tensor<64x64xf32> loc(#loc415) + %b_q_376 = tt.expand_dims %b_g_exp {axis = 0 : i32} : tensor<64xf32> -> tensor<1x64xf32> loc(#loc416) + %b_q_377 = arith.extf %b_q_373 : tensor<64x64xbf16> to tensor<64x64xf32> loc(#loc417) + %b_q_378 = tt.broadcast %b_q_376 : tensor<1x64xf32> -> tensor<64x64xf32> loc(#loc417) + %b_q_379 = arith.mulf %b_q_377, %b_q_378 : tensor<64x64xf32> loc(#loc417) + %b_dh1_380 = arith.extf %b_do_272 : tensor<64x64xbf16> to tensor<64x64xf32> loc(#loc418) + %b_dh1_381 = arith.constant 0.000000e+00 : f32 loc(#loc419) + %b_dh1_382 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc419) + %b_dh1_383 = tt.dot %b_q_379, %b_dh1_380, %b_dh1_382, inputPrecision = tf32 : tensor<64x64xf32> * tensor<64x64xf32> -> tensor<64x64xf32> loc(#loc419) + %b_dh1_384 = tt.splat %scale : f32 -> tensor<64x64xf32> loc(#loc420) + %b_dh1_385 = arith.mulf %b_dh1_383, %b_dh1_384 : tensor<64x64xf32> loc(#loc420) + %b_dh1_386 = arith.truncf %b_dv_337 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc421) + %b_dh1_387 = arith.constant 0.000000e+00 : f32 loc(#loc422) + %b_dh1_388 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc422) + %b_dh1_389 = tt.dot %b_w_365, %b_dh1_386, %b_dh1_388, inputPrecision = tf32 : tensor<64x64xbf16> * tensor<64x64xbf16> -> tensor<64x64xf32> loc(#loc422) + %b_dh1_390 = arith.subf %b_dh1_385, %b_dh1_389 : tensor<64x64xf32> loc(#loc423) + %b_dh1_391 = arith.addf %b_dh1_375, %b_dh1_390 : tensor<64x64xf32> loc(#loc424) + %p_q_392 = tt.expand_dims %o_k2_92 {axis = 1 : i32} : tensor<64xi32> -> tensor<64x1xi32> loc(#loc425) + %p_q_393 = tt.splat %q_14 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc426) + %p_q_394 = tt.addptr %p_q_393, %p_q_392 : tensor<64x1x!tt.ptr>, tensor<64x1xi32> loc(#loc426) + %p_q_395 = tt.expand_dims %o_t_136 {axis = 0 : i32} : tensor<64xi64> -> tensor<1x64xi64> loc(#loc427) + %p_q_396 = arith.constant 4096 : i32 loc(#loc428) + %p_q_397 = arith.constant 4096 : i64 loc(#loc428) + %p_q_398 = arith.constant dense<4096> : tensor<1x64xi64> loc(#loc428) + %p_q_399 = arith.muli %p_q_395, %p_q_398 : tensor<1x64xi64> loc(#loc428) + %p_q_400 = tt.broadcast %p_q_394 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc429) + %p_q_401 = tt.broadcast %p_q_399 : tensor<1x64xi64> -> tensor<64x64xi64> loc(#loc429) + %p_q_402 = tt.addptr %p_q_400, %p_q_401 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc429) + %p_w_403 = tt.expand_dims %o_k2_92 {axis = 1 : i32} : tensor<64xi32> -> tensor<64x1xi32> loc(#loc430) + %p_w_404 = tt.splat %w_33 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc431) + %p_w_405 = tt.addptr %p_w_404, %p_w_403 : tensor<64x1x!tt.ptr>, tensor<64x1xi32> loc(#loc431) + %p_w_406 = tt.expand_dims %o_t_136 {axis = 0 : i32} : tensor<64xi64> -> tensor<1x64xi64> loc(#loc432) + %p_w_407 = arith.constant 4096 : i32 loc(#loc433) + %p_w_408 = arith.constant 4096 : i64 loc(#loc433) + %p_w_409 = arith.constant dense<4096> : tensor<1x64xi64> loc(#loc433) + %p_w_410 = arith.muli %p_w_406, %p_w_409 : tensor<1x64xi64> loc(#loc433) + %p_w_411 = tt.broadcast %p_w_405 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc434) + %p_w_412 = tt.broadcast %p_w_410 : tensor<1x64xi64> -> tensor<64x64xi64> loc(#loc434) + %p_w_413 = tt.addptr %p_w_411, %p_w_412 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc434) + %b_q_414 = tt.expand_dims %m_k2_94 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc435) + %b_q_415 = tt.expand_dims %m_t_138 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc436) + %b_q_416 = tt.broadcast %b_q_414 : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc437) + %b_q_417 = tt.broadcast %b_q_415 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc437) + %b_q_418 = arith.andi %b_q_416, %b_q_417 : tensor<64x64xi1> loc(#loc437) + %b_q_419 = arith.constant 0.000000e+00 : f32 loc(#loc438) + %b_q_420 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc438) + %b_q_421 = arith.truncf %b_q_420 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc438) + %b_q_422 = tt.load %p_q_402, %b_q_418, %b_q_421 : tensor<64x64x!tt.ptr> loc(#loc438) + %b_w_423 = tt.expand_dims %m_k2_94 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc439) + %b_w_424 = tt.expand_dims %m_t_138 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc440) + %b_w_425 = tt.broadcast %b_w_423 : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc441) + %b_w_426 = tt.broadcast %b_w_424 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc441) + %b_w_427 = arith.andi %b_w_425, %b_w_426 : tensor<64x64xi1> loc(#loc441) + %b_w_428 = arith.constant 0.000000e+00 : f32 loc(#loc442) + %b_w_429 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc442) + %b_w_430 = arith.truncf %b_w_429 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc442) + %b_w_431 = tt.load %p_w_413, %b_w_427, %b_w_430 : tensor<64x64x!tt.ptr> loc(#loc442) + %b_dh2_432 = tt.splat %bg_last_exp : f32 -> tensor<64x64xf32> loc(#loc443) + %b_dh2_433 = arith.mulf %b_dh2_129, %b_dh2_432 : tensor<64x64xf32> loc(#loc443) + %b_q_434 = tt.expand_dims %b_g_exp {axis = 0 : i32} : tensor<64xf32> -> tensor<1x64xf32> loc(#loc444) + %b_q_435 = arith.extf %b_q_422 : tensor<64x64xbf16> to tensor<64x64xf32> loc(#loc445) + %b_q_436 = tt.broadcast %b_q_434 : tensor<1x64xf32> -> tensor<64x64xf32> loc(#loc445) + %b_q_437 = arith.mulf %b_q_435, %b_q_436 : tensor<64x64xf32> loc(#loc445) + %b_dh2_438 = arith.extf %b_do_272 : tensor<64x64xbf16> to tensor<64x64xf32> loc(#loc446) + %b_dh2_439 = arith.constant 0.000000e+00 : f32 loc(#loc447) + %b_dh2_440 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc447) + %b_dh2_441 = tt.dot %b_q_437, %b_dh2_438, %b_dh2_440, inputPrecision = tf32 : tensor<64x64xf32> * tensor<64x64xf32> -> tensor<64x64xf32> loc(#loc447) + %b_dh2_442 = tt.splat %scale : f32 -> tensor<64x64xf32> loc(#loc448) + %b_dh2_443 = arith.mulf %b_dh2_441, %b_dh2_442 : tensor<64x64xf32> loc(#loc448) + %b_dh2_444 = arith.truncf %b_dv_337 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc449) + %b_dh2_445 = arith.constant 0.000000e+00 : f32 loc(#loc450) + %b_dh2_446 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc450) + %b_dh2_447 = tt.dot %b_w_431, %b_dh2_444, %b_dh2_446, inputPrecision = tf32 : tensor<64x64xbf16> * tensor<64x64xbf16> -> tensor<64x64xf32> loc(#loc450) + %b_dh2_448 = arith.subf %b_dh2_443, %b_dh2_447 : tensor<64x64xf32> loc(#loc451) + %b_dh2_449 = arith.addf %b_dh2_433, %b_dh2_448 : tensor<64x64xf32> loc(#loc452) + scf.yield %b_dh1_391, %b_dh2_449 : tensor<64x64xf32>, tensor<64x64xf32> loc(#loc223) + } loc(#loc455) + tt.return loc(#loc224) + } loc(#loc) + tt.func private @triton.language.standard.cdiv__c128_c64() attributes {noinline = false} { + tt.return loc(#loc226) + ^bb1: // no predecessors + tt.return loc(#loc227) + } loc(#loc225) + tt.func private @triton.language.standard.cdiv__i32_c64(%x: i32 loc("x"(#loc225))) -> i32 attributes {noinline = false} { + %c63_i32 = arith.constant 63 : i32 loc(#loc228) + %c63_i32_0 = arith.constant 63 : i32 loc(#loc228) + %0 = arith.extsi %x : i32 to i64 loc(#loc228) + %1 = arith.extsi %c63_i32_0 : i32 to i64 loc(#loc228) + %2 = arith.addi %0, %1 : i64 loc(#loc228) + %c2147483647_i64 = arith.constant 2147483647 : i64 loc(#loc228) + %c-2147483648_i64 = arith.constant -2147483648 : i64 loc(#loc228) + %3 = arith.cmpi sle, %2, %c2147483647_i64 : i64 loc(#loc228) + %4 = arith.cmpi sge, %2, %c-2147483648_i64 : i64 loc(#loc228) + %5 = arith.andi %3, %4 : i1 loc(#loc228) + %6 = arith.addi %x, %c63_i32_0 : i32 loc(#loc228) + %c64_i32 = arith.constant 64 : i32 loc(#loc229) + %c64_i32_1 = arith.constant 64 : i32 loc(#loc229) + %7 = arith.divsi %6, %c64_i32_1 : i32 loc(#loc229) + tt.return %7 : i32 loc(#loc226) + ^bb1: // no predecessors + %8 = ub.poison : i32 loc(#loc227) + tt.return %8 : i32 loc(#loc227) + } loc(#loc225) + tt.func private @triton.language.standard.zeros__Tc64_c64T_cfp32() -> tensor<64x64xf32> attributes {noinline = false} { + %cst = arith.constant 0.000000e+00 : f32 loc(#loc231) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc231) + tt.return %cst_0 : tensor<64x64xf32> loc(#loc232) + ^bb1: // no predecessors + %0 = ub.poison : tensor<64x64xf32> loc(#loc233) + tt.return %0 : tensor<64x64xf32> loc(#loc233) + } loc(#loc230) + tt.func private @fla.ops.utils.op.exp2__fp32(%x: f32 loc("x"(#loc234))) -> f32 attributes {noinline = false} { + %0 = math.exp2 %x : f32 loc(#loc235) + tt.return %0 : f32 loc(#loc236) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc237) + tt.return %1 : f32 loc(#loc237) + } loc(#loc234) + tt.func private @fla.ops.utils.op.exp2__fp32S64S(%x: tensor<64xf32> loc("x"(#loc234))) -> tensor<64xf32> attributes {noinline = false} { + %0 = math.exp2 %x : tensor<64xf32> loc(#loc235) + tt.return %0 : tensor<64xf32> loc(#loc236) + ^bb1: // no predecessors + %1 = ub.poison : tensor<64xf32> loc(#loc237) + tt.return %1 : tensor<64xf32> loc(#loc237) + } loc(#loc234) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":399:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":399:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":400:20) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":400:9) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":401:22) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":401:34) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":402:23) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":402:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":409:25) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":409:34) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":409:38) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":410:24) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":411:20) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":422:25) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":424:29) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":431:16) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":431:28) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":431:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":431:53) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":431:9) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":432:16) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":432:28) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":432:20) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":432:53) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":432:9) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":433:16) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":433:21) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":433:41) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":433:9) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":434:17) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":434:22) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":434:42) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":434:10) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":435:17) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":435:22) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":435:42) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":435:10) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":436:18) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":436:23) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":436:43) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":436:11) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":437:17) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":437:22) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":437:42) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":437:44) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":437:10) +#loc47 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":446:16) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":446:34) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":446:21) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":447:16) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":448:24) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":449:18) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":450:16) +#loc54 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":451:18) +#loc55 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":452:17) +#loc56 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":453:18) +#loc57 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":454:17) +#loc58 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":455:18) +#loc59 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":489:26) +#loc60 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":489:33) +#loc61 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":490:27) +#loc62 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":491:26) +#loc63 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":491:44) +#loc64 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":491:31) +#loc65 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":492:20) +#loc66 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":497:35) +#loc67 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":497:38) +#loc68 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":497:40) +#loc69 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":497:25) +#loc70 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":497:49) +#loc71 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":497:60) +#loc72 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":497:44) +#loc73 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":497:68) +#loc74 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":497:64) +#loc75 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":498:25) +#loc76 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":498:40) +#loc77 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":498:36) +#loc78 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":499:33) +#loc79 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":499:24) +#loc80 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":505:39) +#loc81 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":505:42) +#loc82 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":505:44) +#loc83 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":505:29) +#loc84 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":505:53) +#loc85 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":505:64) +#loc86 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":505:48) +#loc87 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":505:72) +#loc88 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":505:68) +#loc89 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":506:29) +#loc90 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":506:44) +#loc91 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":506:40) +#loc92 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":507:37) +#loc93 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":507:28) +#loc94 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":525:36) +#loc95 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":525:41) +#loc96 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":525:45) +#loc97 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":525:50) +#loc98 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":527:41) +#loc99 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":527:53) +#loc100 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":527:34) +#loc101 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":527:58) +#loc102 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":527:30) +#loc103 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":528:28) +#loc104 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":528:22) +#loc105 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":528:33) +#loc106 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":528:45) +#loc107 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":528:39) +#loc108 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":529:26) +#loc109 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":530:31) +#loc110 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":531:27) +#loc111 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":532:24) +#loc112 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":532:36) +#loc113 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":532:20) +#loc114 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":532:48) +#loc115 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":532:44) +#loc116 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":533:26) +#loc117 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":533:38) +#loc118 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":533:22) +#loc119 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":533:50) +#loc120 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":533:46) +#loc121 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":534:24) +#loc122 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":534:36) +#loc123 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":534:20) +#loc124 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":534:48) +#loc125 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":534:44) +#loc126 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":536:38) +#loc127 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":536:53) +#loc128 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":536:49) +#loc129 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":536:23) +#loc130 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":539:22) +#loc131 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":539:34) +#loc132 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":539:18) +#loc133 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":539:46) +#loc134 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":539:41) +#loc135 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":540:36) +#loc136 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":540:52) +#loc137 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":540:47) +#loc138 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":540:22) +#loc139 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":547:40) +#loc140 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":547:31) +#loc141 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":550:26) +#loc142 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":550:38) +#loc143 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":550:22) +#loc144 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":550:50) +#loc145 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":550:45) +#loc146 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":551:40) +#loc147 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":551:56) +#loc148 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":551:51) +#loc149 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":551:26) +#loc150 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":557:44) +#loc151 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":557:56) +#loc152 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":580:49) +#loc153 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":580:39) +#loc154 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":580:55) +#loc155 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":580:58) +#loc156 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":580:20) +#loc157 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":581:39) +#loc158 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":581:54) +#loc159 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":581:50) +#loc160 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":581:24) +#loc161 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":581:16) +#loc162 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":583:65) +#loc163 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":583:80) +#loc164 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":583:76) +#loc165 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":583:32) +#loc166 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":583:24) +#loc167 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":585:23) +#loc168 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":585:18) +#loc169 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":585:38) +#loc170 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":585:50) +#loc171 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":585:34) +#loc172 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":586:23) +#loc173 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":586:18) +#loc174 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":586:38) +#loc175 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":586:50) +#loc176 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":586:34) +#loc177 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":587:37) +#loc178 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":587:52) +#loc179 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":587:48) +#loc180 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":587:22) +#loc181 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":588:37) +#loc182 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":588:52) +#loc183 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":588:48) +#loc184 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":588:22) +#loc185 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":590:21) +#loc186 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":591:32) +#loc187 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":591:24) +#loc188 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":600:55) +#loc189 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":600:47) +#loc190 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":600:69) +#loc191 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":600:97) +#loc192 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":600:89) +#loc193 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":600:77) +#loc194 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":600:21) +#loc195 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":602:27) +#loc196 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":602:22) +#loc197 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":602:42) +#loc198 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":602:54) +#loc199 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":602:38) +#loc200 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":603:27) +#loc201 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":603:22) +#loc202 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":603:42) +#loc203 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":603:54) +#loc204 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":603:38) +#loc205 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":604:41) +#loc206 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":604:56) +#loc207 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":604:52) +#loc208 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":604:26) +#loc209 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":605:41) +#loc210 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":605:56) +#loc211 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":605:52) +#loc212 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":605:26) +#loc213 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":607:25) +#loc214 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":608:36) +#loc215 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":608:28) +#loc216 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":617:59) +#loc217 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":617:51) +#loc218 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":617:73) +#loc219 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":617:101) +#loc220 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":617:93) +#loc221 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":617:81) +#loc222 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":617:25) +#loc223 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":635:8) +#loc224 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":653:4) +#loc226 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":43:11) +#loc227 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":43:4) +#loc228 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":43:17) +#loc229 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":43:30) +#loc230 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":120:0) +#loc231 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":129:31) +#loc232 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":129:11) +#loc233 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":129:4) +#loc235 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:33) +#loc236 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:20) +#loc237 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:13) +#loc248 = loc("pid"(#loc1)) +#loc249 = loc("pid"(#loc2)) +#loc250 = loc("NV"(#loc3)) +#loc251 = loc("NV"(#loc4)) +#loc252 = loc("NT"(#loc12)) +#loc253 = loc("boh"(#loc13)) +#loc254 = loc("b_dh1"(#loc14)) +#loc255 = loc("b_dh2"(#loc15)) +#loc256 = loc("q"(#loc16)) +#loc257 = loc("q"(#loc17)) +#loc258 = loc("q"(#loc18)) +#loc259 = loc("q"(#loc19)) +#loc260 = loc("q"(#loc20)) +#loc261 = loc("k"(#loc21)) +#loc262 = loc("k"(#loc22)) +#loc263 = loc("k"(#loc23)) +#loc264 = loc("k"(#loc24)) +#loc265 = loc("k"(#loc25)) +#loc266 = loc("w"(#loc26)) +#loc267 = loc("w"(#loc27)) +#loc268 = loc("w"(#loc28)) +#loc269 = loc("w"(#loc29)) +#loc270 = loc("do"(#loc30)) +#loc271 = loc("do"(#loc31)) +#loc272 = loc("do"(#loc32)) +#loc273 = loc("do"(#loc33)) +#loc274 = loc("dv"(#loc34)) +#loc275 = loc("dv"(#loc35)) +#loc276 = loc("dv"(#loc36)) +#loc277 = loc("dv"(#loc37)) +#loc278 = loc("dv2"(#loc38)) +#loc279 = loc("dv2"(#loc39)) +#loc280 = loc("dv2"(#loc40)) +#loc281 = loc("dv2"(#loc41)) +#loc282 = loc("dh"(#loc42)) +#loc283 = loc("dh"(#loc43)) +#loc284 = loc("dh"(#loc44)) +#loc285 = loc("dh"(#loc45)) +#loc286 = loc("dh"(#loc46)) +#loc287 = loc("o_v"(#loc47)) +#loc288 = loc("o_v"(#loc48)) +#loc289 = loc("o_v"(#loc49)) +#loc290 = loc("m_v"(#loc50)) +#loc291 = loc("o_k1"(#loc51)) +#loc292 = loc("m_k1"(#loc52)) +#loc293 = loc("o_k2"(#loc53)) +#loc294 = loc("m_k2"(#loc54)) +#loc295 = loc("o_k3"(#loc55)) +#loc296 = loc("m_k3"(#loc56)) +#loc297 = loc("o_k4"(#loc57)) +#loc298 = loc("m_k4"(#loc58)) +#loc299 = loc("b_dh1"(#loc60)) +#loc300 = loc("i_t_int64"(#loc61)) +#loc301 = loc("o_t"(#loc62)) +#loc302 = loc("o_t"(#loc63)) +#loc303 = loc("o_t"(#loc64)) +#loc304 = loc("m_t"(#loc65)) +#loc305 = loc("p_dh1"(#loc66)) +#loc306 = loc("p_dh1"(#loc67)) +#loc307 = loc("p_dh1"(#loc68)) +#loc308 = loc("p_dh1"(#loc69)) +#loc309 = loc("p_dh1"(#loc70)) +#loc310 = loc("p_dh1"(#loc71)) +#loc311 = loc("p_dh1"(#loc72)) +#loc312 = loc("p_dh1"(#loc73)) +#loc313 = loc("p_dh1"(#loc74)) +#loc314 = loc("m_dh1"(#loc75)) +#loc315 = loc("m_dh1"(#loc76)) +#loc316 = loc("m_dh1"(#loc77)) +#loc317 = loc("p_dh2"(#loc80)) +#loc318 = loc("p_dh2"(#loc81)) +#loc319 = loc("p_dh2"(#loc82)) +#loc320 = loc("p_dh2"(#loc83)) +#loc321 = loc("p_dh2"(#loc84)) +#loc322 = loc("p_dh2"(#loc85)) +#loc323 = loc("p_dh2"(#loc86)) +#loc324 = loc("p_dh2"(#loc87)) +#loc325 = loc("p_dh2"(#loc88)) +#loc326 = loc("m_dh2"(#loc89)) +#loc327 = loc("m_dh2"(#loc90)) +#loc328 = loc("m_dh2"(#loc91)) +#loc329 = loc("last_idx"(#loc94)) +#loc330 = loc("last_idx"(#loc95)) +#loc331 = loc("last_idx"(#loc96)) +#loc332 = loc("last_idx"(#loc97)) +#loc333 = loc("bg_last"(#loc98)) +#loc334 = loc("bg_last"(#loc99)) +#loc335 = loc("bg_last"(#loc100)) +#loc336 = loc("bg_last"(#loc101)) +#loc337 = loc("bg_last"(#loc102)) +#loc338 = loc("p_g"(#loc103)) +#loc339 = loc("p_g"(#loc104)) +#loc340 = loc("p_g"(#loc105)) +#loc341 = loc("p_g"(#loc106)) +#loc342 = loc("p_g"(#loc107)) +#loc343 = loc("b_g"(#loc108)) +#loc344 = loc("bg_last_exp"(#loc109)) +#loc345 = loc("b_g_exp"(#loc110)) +#loc346 = loc("p_dv"(#loc111)) +#loc347 = loc("p_dv"(#loc112)) +#loc348 = loc("p_dv"(#loc113)) +#loc349 = loc("p_dv"(#loc114)) +#loc350 = loc("p_dv"(#loc115)) +#loc351 = loc("p_dv2"(#loc116)) +#loc352 = loc("p_dv2"(#loc117)) +#loc353 = loc("p_dv2"(#loc118)) +#loc354 = loc("p_dv2"(#loc119)) +#loc355 = loc("p_dv2"(#loc120)) +#loc356 = loc("p_do"(#loc121)) +#loc357 = loc("p_do"(#loc122)) +#loc358 = loc("p_do"(#loc123)) +#loc359 = loc("p_do"(#loc124)) +#loc360 = loc("p_do"(#loc125)) +#loc361 = loc("b_do"(#loc126)) +#loc362 = loc("b_do"(#loc127)) +#loc363 = loc("b_do"(#loc128)) +#loc364 = loc("b_do"(#loc129)) +#loc365 = loc("p_k"(#loc130)) +#loc366 = loc("p_k"(#loc131)) +#loc367 = loc("p_k"(#loc132)) +#loc368 = loc("p_k"(#loc133)) +#loc369 = loc("p_k"(#loc134)) +#loc370 = loc("b_k"(#loc135)) +#loc371 = loc("b_k"(#loc136)) +#loc372 = loc("b_k"(#loc137)) +#loc373 = loc("b_k"(#loc138)) +#loc374 = loc("b_dv"(#loc139)) +#loc375 = loc("b_dv"(#loc140)) +#loc376 = loc("p_k"(#loc141)) +#loc377 = loc("p_k"(#loc142)) +#loc378 = loc("p_k"(#loc143)) +#loc379 = loc("p_k"(#loc144)) +#loc380 = loc("p_k"(#loc145)) +#loc381 = loc("b_k"(#loc146)) +#loc382 = loc("b_k"(#loc147)) +#loc383 = loc("b_k"(#loc148)) +#loc384 = loc("b_k"(#loc149)) +#loc385 = loc("b_dv"(#loc150)) +#loc386 = loc("b_dv"(#loc151)) +#loc387 = loc("b_dv"(#loc152)) +#loc388 = loc("b_dv"(#loc153)) +#loc389 = loc("b_dv"(#loc154)) +#loc390 = loc("b_dv"(#loc155)) +#loc391 = loc("b_dv"(#loc156)) +#loc392 = loc("b_dv"(#loc157)) +#loc393 = loc("b_dv"(#loc158)) +#loc394 = loc("b_dv"(#loc159)) +#loc395 = loc("b_dv"(#loc160)) +#loc396 = loc("b_dv"(#loc161)) +#loc397 = loc("p_w"(#loc167)) +#loc398 = loc("p_w"(#loc168)) +#loc399 = loc("p_w"(#loc169)) +#loc400 = loc("p_w"(#loc170)) +#loc401 = loc("p_w"(#loc171)) +#loc402 = loc("p_q"(#loc172)) +#loc403 = loc("p_q"(#loc173)) +#loc404 = loc("p_q"(#loc174)) +#loc405 = loc("p_q"(#loc175)) +#loc406 = loc("p_q"(#loc176)) +#loc407 = loc("b_w"(#loc177)) +#loc408 = loc("b_w"(#loc178)) +#loc409 = loc("b_w"(#loc179)) +#loc410 = loc("b_w"(#loc180)) +#loc411 = loc("b_q"(#loc181)) +#loc412 = loc("b_q"(#loc182)) +#loc413 = loc("b_q"(#loc183)) +#loc414 = loc("b_q"(#loc184)) +#loc415 = loc("b_dh1"(#loc185)) +#loc416 = loc("b_q"(#loc186)) +#loc417 = loc("b_q"(#loc187)) +#loc418 = loc("b_dh1"(#loc188)) +#loc419 = loc("b_dh1"(#loc189)) +#loc420 = loc("b_dh1"(#loc190)) +#loc421 = loc("b_dh1"(#loc191)) +#loc422 = loc("b_dh1"(#loc192)) +#loc423 = loc("b_dh1"(#loc193)) +#loc424 = loc("b_dh1"(#loc194)) +#loc425 = loc("p_q"(#loc195)) +#loc426 = loc("p_q"(#loc196)) +#loc427 = loc("p_q"(#loc197)) +#loc428 = loc("p_q"(#loc198)) +#loc429 = loc("p_q"(#loc199)) +#loc430 = loc("p_w"(#loc200)) +#loc431 = loc("p_w"(#loc201)) +#loc432 = loc("p_w"(#loc202)) +#loc433 = loc("p_w"(#loc203)) +#loc434 = loc("p_w"(#loc204)) +#loc435 = loc("b_q"(#loc205)) +#loc436 = loc("b_q"(#loc206)) +#loc437 = loc("b_q"(#loc207)) +#loc438 = loc("b_q"(#loc208)) +#loc439 = loc("b_w"(#loc209)) +#loc440 = loc("b_w"(#loc210)) +#loc441 = loc("b_w"(#loc211)) +#loc442 = loc("b_w"(#loc212)) +#loc443 = loc("b_dh2"(#loc213)) +#loc444 = loc("b_q"(#loc214)) +#loc445 = loc("b_q"(#loc215)) +#loc446 = loc("b_dh2"(#loc216)) +#loc447 = loc("b_dh2"(#loc217)) +#loc448 = loc("b_dh2"(#loc218)) +#loc449 = loc("b_dh2"(#loc219)) +#loc450 = loc("b_dh2"(#loc220)) +#loc451 = loc("b_dh2"(#loc221)) +#loc452 = loc("b_dh2"(#loc222)) +#loc455 = loc("b_dh2"(#loc299)) +#loc456 = loc("i_t"(#loc300)) diff --git a/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.ttgir b/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..6bf0c439fd422701b7e7cf8391a01e165613ed7f --- /dev/null +++ b/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.ttgir @@ -0,0 +1,891 @@ +#blocked = #ttg.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [4], order = [0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [4, 8], warpsPerCTA = [4, 1], order = [1, 0]}> +#blocked2 = #ttg.blocked<{sizePerThread = [8, 1], threadsPerWarp = [8, 4], warpsPerCTA = [1, 4], order = [0, 1]}> +#linear = #ttg.linear<{register = [[0, 1], [0, 2], [0, 4], [0, 8], [0, 16]], lane = [[1, 0], [2, 0], [4, 0], [8, 0], [0, 32]], warp = [[16, 0], [32, 0]], block = []}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":370:0) +#shared = #ttg.swizzled_shared<{vec = 1, perPhase = 1, maxPhase = 1, order = [0]}> +#shared1 = #ttg.swizzled_shared<{vec = 1, perPhase = 1, maxPhase = 1, order = [1, 0]}> +#shared2 = #ttg.nvmma_shared<{swizzlingByteWidth = 128, transposed = false, elementBitWidth = 16}> +#shared3 = #ttg.nvmma_shared<{swizzlingByteWidth = 128, transposed = true, elementBitWidth = 16}> +#shared4 = #ttg.swizzled_shared<{vec = 1, perPhase = 1, maxPhase = 1, order = [0, 1]}> +#shared5 = #ttg.nvmma_shared<{swizzlingByteWidth = 128, transposed = true, elementBitWidth = 32}> +#smem = #ttg.shared_memory +#tmem = #ttng.tensor_memory_encoding +#loc144 = loc("q"(#loc)) +#loc145 = loc("k"(#loc)) +#loc146 = loc("w"(#loc)) +#loc147 = loc("g"(#loc)) +#loc148 = loc("do"(#loc)) +#loc149 = loc("dh"(#loc)) +#loc150 = loc("dv"(#loc)) +#loc151 = loc("dv2"(#loc)) +#loc152 = loc("scale"(#loc)) +#loc153 = loc("T"(#loc)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 4 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64(%q: !tt.ptr {tt.divisibility = 16 : i32} loc("q"(#loc)), %k: !tt.ptr {tt.divisibility = 16 : i32} loc("k"(#loc)), %w: !tt.ptr {tt.divisibility = 16 : i32} loc("w"(#loc)), %g: !tt.ptr {tt.divisibility = 16 : i32} loc("g"(#loc)), %do: !tt.ptr {tt.divisibility = 16 : i32} loc("do"(#loc)), %dh: !tt.ptr {tt.divisibility = 16 : i32} loc("dh"(#loc)), %dv: !tt.ptr {tt.divisibility = 16 : i32} loc("dv"(#loc)), %dv2: !tt.ptr {tt.divisibility = 16 : i32} loc("dv2"(#loc)), %scale: f32 loc("scale"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<32> : tensor<64xi64, #blocked> loc(#loc1) + %c0_i32 = arith.constant 0 : i32 loc(#loc1) + %c2_i32 = arith.constant 2 : i32 loc(#loc1) + %c-2_i32 = arith.constant -2 : i32 loc(#loc1) + %c3_i32 = arith.constant 3 : i32 loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<64xf32, #blocked> loc(#loc1) + %cst_1 = arith.constant dense<0.000000e+00> : tensor<64xf32, #ttg.slice<{dim = 1, parent = #linear}>> loc(#loc1) + %cst_2 = arith.constant dense<4096> : tensor<64x1xi64, #blocked1> loc(#loc1) + %cst_3 = arith.constant dense<128> : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc1) + %cst_4 = arith.constant dense<4096> : tensor<1x64xi64, #blocked2> loc(#loc1) + %cst_5 = arith.constant dense<128> : tensor<64x1xi32, #blocked1> loc(#loc1) + %cst_6 = arith.constant dense<64> : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc1) + %cst_7 = arith.constant dense<64> : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked2}>> loc(#loc1) + %cst_8 = arith.constant dense<64> : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc1) + %cst_9 = arith.constant dense<128> : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc1) + %cst_10 = arith.constant dense<128> : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked2}>> loc(#loc1) + %cst_11 = arith.constant dense<128> : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc1) + %c64_i64 = arith.constant 64 : i64 loc(#loc1) + %c128_i64 = arith.constant 128 : i64 loc(#loc1) + %c1_i64 = arith.constant 1 : i64 loc(#loc1) + %c1_i32 = arith.constant 1 : i32 loc(#loc1) + %c32_i64 = arith.constant 32 : i64 loc(#loc1) + %cst_12 = arith.constant dense<0.000000e+00> : tensor<64x64xbf16, #blocked1> loc(#loc1) + %cst_13 = arith.constant dense<0.000000e+00> : tensor<64x64xbf16, #blocked2> loc(#loc1) + %c63_i32 = arith.constant 63 : i32 loc(#loc1) + %c64_i32 = arith.constant 64 : i32 loc(#loc1) + %c524288_i64 = arith.constant 524288 : i64 loc(#loc1) + %c-1_i32 = arith.constant -1 : i32 loc(#loc1) + %c16384_i64 = arith.constant 16384 : i64 loc(#loc1) + %c2_i64 = arith.constant 2 : i64 loc(#loc1) + %true = arith.constant true loc(#loc1) + %cst_14 = arith.constant dense<0.000000e+00> : tensor<64x64xf32, #linear> loc(#loc1) + %false = arith.constant false loc(#loc1) + %pid = tt.get_program_id x : i32 loc(#loc154) + %pid_15 = arith.extsi %pid : i32 to i64 loc(#loc155) + %0 = arith.remsi %pid_15, %c2_i64 : i64 loc(#loc4) + %1 = arith.divsi %pid_15, %c2_i64 : i64 loc(#loc5) + %2 = arith.divsi %1, %c32_i64 : i64 loc(#loc6) + %3 = arith.remsi %1, %c32_i64 : i64 loc(#loc7) + %4 = arith.extsi %T : i32 to i64 loc(#loc8) + %5 = arith.muli %2, %4 : i64 loc(#loc8) + %NT = arith.addi %T, %c63_i32 : i32 loc(#loc280) + %NT_16 = arith.divsi %NT, %c64_i32 : i32 loc(#loc281) + %boh = arith.extsi %NT_16 : i32 to i64 loc(#loc157) + %boh_17 = arith.muli %2, %boh : i64 loc(#loc157) + %q_18 = arith.muli %5, %c32_i64 : i64 loc(#loc158) + %q_19 = arith.addi %q_18, %3 : i64 loc(#loc159) + %q_20 = arith.muli %q_19, %c128_i64 : i64 loc(#loc160) + %q_21 = tt.addptr %q, %q_20 : !tt.ptr, i64 loc(#loc161) + %k_22 = tt.addptr %k, %q_20 : !tt.ptr, i64 loc(#loc162) + %w_23 = tt.addptr %w, %q_20 : !tt.ptr, i64 loc(#loc163) + %do_24 = tt.addptr %do, %q_20 : !tt.ptr, i64 loc(#loc164) + %dv_25 = tt.addptr %dv, %q_20 : !tt.ptr, i64 loc(#loc165) + %dv2_26 = tt.addptr %dv2, %q_20 : !tt.ptr, i64 loc(#loc166) + %dh_27 = arith.muli %boh_17, %c32_i64 : i64 loc(#loc167) + %dh_28 = arith.addi %dh_27, %3 : i64 loc(#loc168) + %dh_29 = arith.muli %dh_28, %c16384_i64 : i64 loc(#loc169) + %o_v = arith.muli %0, %c64_i64 : i64 loc(#loc170) + %o_v_30 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc171) + %o_v_31 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #blocked> loc(#loc171) + %o_v_32 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked2}>> loc(#loc171) + %o_v_33 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc171) + %o_v_34 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked2}>> loc(#loc171) + %o_v_35 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 1, parent = #linear}>> loc(#loc171) + %o_v_36 = arith.extsi %o_v_30 : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> to tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc172) + %o_v_37 = arith.extsi %o_v_31 : tensor<64xi32, #blocked> to tensor<64xi64, #blocked> loc(#loc172) + %o_v_38 = arith.extsi %o_v_32 : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked2}>> to tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked2}>> loc(#loc172) + %o_v_39 = arith.extsi %o_v_33 : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> to tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc172) + %o_v_40 = arith.extsi %o_v_35 : tensor<64xi32, #ttg.slice<{dim = 1, parent = #linear}>> to tensor<64xi64, #ttg.slice<{dim = 1, parent = #linear}>> loc(#loc172) + %o_v_41 = tt.splat %o_v : i64 -> tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc172) + %o_v_42 = arith.addi %o_v_41, %o_v_36 : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc172) + %m_v = arith.cmpi slt, %o_v_42, %cst_3 : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc173) + %m_k1 = arith.cmpi slt, %o_v_33, %cst_11 : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc174) + %m_k1_43 = arith.cmpi slt, %o_v_34, %cst_10 : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked2}>> loc(#loc174) + %m_k1_44 = arith.cmpi slt, %o_v_30, %cst_9 : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc174) + %o_k2 = arith.addi %o_v_33, %cst_8 : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc175) + %o_k2_45 = arith.addi %o_v_34, %cst_7 : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked2}>> loc(#loc175) + %o_k2_46 = arith.addi %o_v_30, %cst_6 : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc175) + %m_k2 = arith.cmpi slt, %o_k2, %cst_11 : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc176) + %m_k2_47 = arith.cmpi slt, %o_k2_45, %cst_10 : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked2}>> loc(#loc176) + %m_k2_48 = arith.cmpi slt, %o_k2_46, %cst_9 : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc176) + %6 = arith.subi %NT_16, %c1_i32 : i32 loc(#loc32) + %m_t = tt.splat %4 : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc177) + %m_t_49 = tt.splat %4 : i64 -> tensor<64xi64, #blocked> loc(#loc177) + %m_t_50 = tt.splat %4 : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #linear}>> loc(#loc177) + %m_t_51 = tt.splat %4 : i64 -> tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked2}>> loc(#loc177) + %p_dh1 = tt.expand_dims %o_v_33 {axis = 1 : i32} : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<64x1xi32, #blocked1> loc(#loc178) + %p_dh1_52 = tt.expand_dims %o_v_34 {axis = 1 : i32} : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked2}>> -> tensor<64x1xi32, #blocked2> loc(#loc178) + %p_dh1_53 = arith.muli %p_dh1, %cst_5 : tensor<64x1xi32, #blocked1> loc(#loc179) + %p_dh1_54 = tt.expand_dims %o_v_42 {axis = 0 : i32} : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x64xi64, #blocked1> loc(#loc180) + %p_dh1_55 = tt.broadcast %p_dh1_54 : tensor<1x64xi64, #blocked1> -> tensor<64x64xi64, #blocked1> loc(#loc181) + %m_dh1 = tt.expand_dims %m_k1 {axis = 1 : i32} : tensor<64xi1, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<64x1xi1, #blocked1> loc(#loc182) + %m_dh1_56 = tt.expand_dims %m_k1_43 {axis = 1 : i32} : tensor<64xi1, #ttg.slice<{dim = 1, parent = #blocked2}>> -> tensor<64x1xi1, #blocked2> loc(#loc182) + %m_dh1_57 = tt.expand_dims %m_v {axis = 0 : i32} : tensor<64xi1, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x64xi1, #blocked1> loc(#loc183) + %m_dh1_58 = tt.broadcast %m_dh1 : tensor<64x1xi1, #blocked1> -> tensor<64x64xi1, #blocked1> loc(#loc184) + %m_dh1_59 = tt.broadcast %m_dh1_56 : tensor<64x1xi1, #blocked2> -> tensor<64x64xi1, #blocked2> loc(#loc184) + %m_dh1_60 = tt.broadcast %m_dh1_57 : tensor<1x64xi1, #blocked1> -> tensor<64x64xi1, #blocked1> loc(#loc184) + %m_dh1_61 = arith.andi %m_dh1_58, %m_dh1_60 : tensor<64x64xi1, #blocked1> loc(#loc184) + %p_dh2 = tt.expand_dims %o_k2 {axis = 1 : i32} : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<64x1xi32, #blocked1> loc(#loc185) + %p_dh2_62 = tt.expand_dims %o_k2_45 {axis = 1 : i32} : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked2}>> -> tensor<64x1xi32, #blocked2> loc(#loc185) + %p_dh2_63 = arith.muli %p_dh2, %cst_5 : tensor<64x1xi32, #blocked1> loc(#loc186) + %m_dh2 = tt.expand_dims %m_k2 {axis = 1 : i32} : tensor<64xi1, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<64x1xi1, #blocked1> loc(#loc187) + %m_dh2_64 = tt.expand_dims %m_k2_47 {axis = 1 : i32} : tensor<64xi1, #ttg.slice<{dim = 1, parent = #blocked2}>> -> tensor<64x1xi1, #blocked2> loc(#loc187) + %m_dh2_65 = tt.broadcast %m_dh2 : tensor<64x1xi1, #blocked1> -> tensor<64x64xi1, #blocked1> loc(#loc188) + %m_dh2_66 = tt.broadcast %m_dh2_64 : tensor<64x1xi1, #blocked2> -> tensor<64x64xi1, #blocked2> loc(#loc188) + %m_dh2_67 = arith.andi %m_dh2_65, %m_dh1_60 : tensor<64x64xi1, #blocked1> loc(#loc188) + %p_g = tt.addptr %g, %q_19 : !tt.ptr, i64 loc(#loc282) + %p_g_68 = tt.splat %p_g : !tt.ptr -> tensor<64x!tt.ptr, #blocked> loc(#loc191) + %p_dv = tt.splat %dv_25 : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked1> loc(#loc192) + %p_dv2 = tt.splat %dv2_26 : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked1> loc(#loc193) + %p_do = tt.splat %do_24 : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked1> loc(#loc194) + %p_k = tt.splat %k_22 : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked1> loc(#loc195) + %p_k_69 = tt.expand_dims %o_v_30 {axis = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x64xi32, #blocked1> loc(#loc196) + %p_k_70 = tt.broadcast %p_k_69 : tensor<1x64xi32, #blocked1> -> tensor<64x64xi32, #blocked1> loc(#loc197) + %b_k = tt.expand_dims %m_k1_44 {axis = 0 : i32} : tensor<64xi1, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x64xi1, #blocked1> loc(#loc198) + %b_k_71 = tt.broadcast %b_k : tensor<1x64xi1, #blocked1> -> tensor<64x64xi1, #blocked1> loc(#loc199) + %p_k_72 = tt.expand_dims %o_k2_46 {axis = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x64xi32, #blocked1> loc(#loc200) + %p_k_73 = tt.broadcast %p_k_72 : tensor<1x64xi32, #blocked1> -> tensor<64x64xi32, #blocked1> loc(#loc201) + %b_k_74 = tt.expand_dims %m_k2_48 {axis = 0 : i32} : tensor<64xi1, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x64xi1, #blocked1> loc(#loc202) + %b_k_75 = tt.broadcast %b_k_74 : tensor<1x64xi1, #blocked1> -> tensor<64x64xi1, #blocked1> loc(#loc203) + %p_w = tt.splat %w_23 : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked2> loc(#loc204) + %p_w_76 = tt.addptr %p_w, %p_dh1_52 : tensor<64x1x!tt.ptr, #blocked2>, tensor<64x1xi32, #blocked2> loc(#loc204) + %p_w_77 = tt.broadcast %p_w_76 : tensor<64x1x!tt.ptr, #blocked2> -> tensor<64x64x!tt.ptr, #blocked2> loc(#loc205) + %p_q = tt.splat %q_21 : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked2> loc(#loc206) + %p_q_78 = tt.addptr %p_q, %p_dh1_52 : tensor<64x1x!tt.ptr, #blocked2>, tensor<64x1xi32, #blocked2> loc(#loc206) + %p_q_79 = tt.broadcast %p_q_78 : tensor<64x1x!tt.ptr, #blocked2> -> tensor<64x64x!tt.ptr, #blocked2> loc(#loc207) + %b_dh1 = tt.splat %scale : f32 -> tensor<64x64xf32, #linear> loc(#loc208) + %p_q_80 = tt.addptr %p_q, %p_dh2_62 : tensor<64x1x!tt.ptr, #blocked2>, tensor<64x1xi32, #blocked2> loc(#loc209) + %p_q_81 = tt.broadcast %p_q_80 : tensor<64x1x!tt.ptr, #blocked2> -> tensor<64x64x!tt.ptr, #blocked2> loc(#loc210) + %p_w_82 = tt.addptr %p_w, %p_dh2_62 : tensor<64x1x!tt.ptr, #blocked2>, tensor<64x1xi32, #blocked2> loc(#loc211) + %p_w_83 = tt.broadcast %p_w_82 : tensor<64x1x!tt.ptr, #blocked2> -> tensor<64x64x!tt.ptr, #blocked2> loc(#loc212) + %bg_last = ttg.local_alloc : () -> !ttg.memdesc<2x1xf32, #shared, #smem, mutable> loc(#loc213) + %b_g = ttg.local_alloc : () -> !ttg.memdesc<2x64xf32, #shared, #smem, mutable> loc(#loc214) + %b_g_84 = ttg.local_alloc : () -> !ttg.memdesc<2x64xf32, #shared, #smem, mutable> loc(#loc214) + %b_do = ttg.local_alloc : () -> !ttg.memdesc<2x64x64xbf16, #shared1, #smem, mutable> loc(#loc215) + %b_k_85 = ttg.local_alloc : () -> !ttg.memdesc<2x64x64xbf16, #shared2, #smem, mutable> loc(#loc216) + %b_k_86 = ttg.local_alloc : () -> !ttg.memdesc<2x64x64xbf16, #shared2, #smem, mutable> loc(#loc217) + %b_dv = ttg.local_alloc : () -> !ttg.memdesc<2x64x64xbf16, #shared1, #smem, mutable> loc(#loc218) + %b_w = ttg.local_alloc : () -> !ttg.memdesc<2x64x64xbf16, #shared3, #smem, mutable> loc(#loc219) + %b_q = ttg.local_alloc : () -> !ttg.memdesc<2x64x64xbf16, #shared4, #smem, mutable> loc(#loc220) + %b_q_87 = ttg.local_alloc : () -> !ttg.memdesc<2x64x64xbf16, #shared4, #smem, mutable> loc(#loc221) + %b_w_88 = ttg.local_alloc : () -> !ttg.memdesc<2x64x64xbf16, #shared3, #smem, mutable> loc(#loc222) + %b_dh2 = arith.cmpi sgt, %6, %c-1_i32 : i32 loc(#loc283) + %i_t_int64 = arith.extsi %6 : i32 to i64 loc(#loc224) + %o_t = arith.muli %i_t_int64, %c64_i64 : i64 loc(#loc225) + %o_t_89 = tt.splat %o_t : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc226) + %o_t_90 = tt.splat %o_t : i64 -> tensor<64xi64, #blocked> loc(#loc226) + %o_t_91 = tt.splat %o_t : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #linear}>> loc(#loc226) + %o_t_92 = tt.splat %o_t : i64 -> tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked2}>> loc(#loc226) + %o_t_93 = arith.addi %o_t_89, %o_v_39 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc226) + %o_t_94 = arith.addi %o_t_90, %o_v_37 : tensor<64xi64, #blocked> loc(#loc226) + %o_t_95 = arith.addi %o_t_91, %o_v_40 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #linear}>> loc(#loc226) + %o_t_96 = arith.addi %o_t_92, %o_v_38 : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked2}>> loc(#loc226) + %m_t_97 = arith.cmpi slt, %o_t_93, %m_t : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc177) + %m_t_98 = arith.cmpi slt, %o_t_94, %m_t_49 : tensor<64xi64, #blocked> loc(#loc177) + %m_t_99 = arith.cmpi slt, %o_t_95, %m_t_50 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #linear}>> loc(#loc177) + %m_t_100 = arith.cmpi slt, %o_t_96, %m_t_51 : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked2}>> loc(#loc177) + %last_idx = arith.addi %i_t_int64, %c1_i64 : i64 loc(#loc227) + %last_idx_101 = arith.muli %last_idx, %c64_i64 : i64 loc(#loc228) + %last_idx_102 = arith.minsi %last_idx_101, %4 : i64 loc(#loc229) + %last_idx_103 = arith.subi %last_idx_102, %c1_i64 : i64 loc(#loc230) + %bg_last_104 = arith.addi %5, %last_idx_103 : i64 loc(#loc231) + %bg_last_105 = arith.muli %bg_last_104, %c32_i64 : i64 loc(#loc232) + %bg_last_106 = arith.addi %bg_last_105, %3 : i64 loc(#loc284) + %bg_last_107 = tt.addptr %g, %bg_last_106 : !tt.ptr, i64 loc(#loc284) + %bg_last_108 = tt.splat %bg_last_107 : !tt.ptr -> tensor<1x!tt.ptr, #blocked> loc(#loc213) + %bg_last_109 = ttg.memdesc_index %bg_last[%c0_i32] : !ttg.memdesc<2x1xf32, #shared, #smem, mutable> -> !ttg.memdesc<1xf32, #shared, #smem, mutable> loc(#loc213) + %b_dh2_110 = tt.splat %b_dh2 : i1 -> tensor<1xi1, #blocked> loc(#loc283) + %bg_last_111 = ttg.async_copy_global_to_local %bg_last_108, %bg_last_109 mask %b_dh2_110 : tensor<1x!tt.ptr, #blocked> -> <1xf32, #shared, #smem, mutable> loc(#loc213) + %b_dv_112 = ttng.tmem_alloc : () -> !ttg.memdesc<64x64xf32, #tmem, #ttng.tensor_memory, mutable> loc(#loc235) + %b_dv_113 = ttng.tmem_alloc : () -> !ttg.memdesc<64x64xf32, #tmem, #ttng.tensor_memory, mutable> loc(#loc236) + %b_dh1_114 = ttng.tmem_alloc : () -> !ttg.memdesc<64x64xf32, #tmem, #ttng.tensor_memory, mutable> loc(#loc237) + %b_dh1_115 = ttng.tmem_alloc : () -> !ttg.memdesc<64x64xf32, #tmem, #ttng.tensor_memory, mutable> loc(#loc238) + %b_dh2_116 = ttng.tmem_alloc : () -> !ttg.memdesc<64x64xf32, #tmem, #ttng.tensor_memory, mutable> loc(#loc239) + %b_dh2_117 = ttng.tmem_alloc : () -> !ttg.memdesc<64x64xf32, #tmem, #ttng.tensor_memory, mutable> loc(#loc240) + %bg_last_118 = ttg.async_commit_group tokens %bg_last_111 loc(#loc213) + %p_g_119 = arith.muli %o_t_94, %cst : tensor<64xi64, #blocked> loc(#loc241) + %p_g_120 = tt.addptr %p_g_68, %p_g_119 : tensor<64x!tt.ptr, #blocked>, tensor<64xi64, #blocked> loc(#loc191) + %b_g_121 = ttg.memdesc_index %b_g[%c0_i32] : !ttg.memdesc<2x64xf32, #shared, #smem, mutable> -> !ttg.memdesc<64xf32, #shared, #smem, mutable> loc(#loc214) + %b_dh2_122 = tt.splat %b_dh2 : i1 -> tensor<64xi1, #blocked> loc(#loc283) + %b_dh2_123 = arith.andi %b_dh2_122, %m_t_98 : tensor<64xi1, #blocked> loc(#loc283) + %b_g_124 = ttg.async_copy_global_to_local %p_g_120, %b_g_121 mask %b_dh2_123 other %cst_0 : tensor<64x!tt.ptr, #blocked> -> <64xf32, #shared, #smem, mutable> loc(#loc214) + %b_g_125 = ttg.async_commit_group tokens %b_g_124 loc(#loc214) + %b_g_126 = ttg.memdesc_index %b_g_84[%c0_i32] : !ttg.memdesc<2x64xf32, #shared, #smem, mutable> -> !ttg.memdesc<64xf32, #shared, #smem, mutable> loc(#loc214) + %b_g_127 = ttg.async_copy_global_to_local %p_g_120, %b_g_126 mask %b_dh2_123 other %cst_0 : tensor<64x!tt.ptr, #blocked> -> <64xf32, #shared, #smem, mutable> loc(#loc214) + %b_g_128 = ttg.async_commit_group tokens %b_g_127 loc(#loc214) + %p_dv_129 = tt.expand_dims %o_t_93 {axis = 1 : i32} : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<64x1xi64, #blocked1> loc(#loc242) + %p_dv_130 = arith.muli %p_dv_129, %cst_2 : tensor<64x1xi64, #blocked1> loc(#loc243) + %p_dv_131 = tt.addptr %p_dv, %p_dv_130 : tensor<64x1x!tt.ptr, #blocked1>, tensor<64x1xi64, #blocked1> loc(#loc192) + %p_dv_132 = tt.broadcast %p_dv_131 : tensor<64x1x!tt.ptr, #blocked1> -> tensor<64x64x!tt.ptr, #blocked1> loc(#loc244) + %p_dv_133 = tt.addptr %p_dv_132, %p_dh1_55 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi64, #blocked1> loc(#loc244) + %p_do_134 = tt.addptr %p_do, %p_dv_130 : tensor<64x1x!tt.ptr, #blocked1>, tensor<64x1xi64, #blocked1> loc(#loc194) + %p_do_135 = tt.broadcast %p_do_134 : tensor<64x1x!tt.ptr, #blocked1> -> tensor<64x64x!tt.ptr, #blocked1> loc(#loc245) + %p_do_136 = tt.addptr %p_do_135, %p_dh1_55 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi64, #blocked1> loc(#loc245) + %b_do_137 = tt.expand_dims %m_t_97 {axis = 1 : i32} : tensor<64xi1, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<64x1xi1, #blocked1> loc(#loc246) + %b_do_138 = tt.broadcast %b_do_137 : tensor<64x1xi1, #blocked1> -> tensor<64x64xi1, #blocked1> loc(#loc247) + %b_do_139 = arith.andi %b_do_138, %m_dh1_60 : tensor<64x64xi1, #blocked1> loc(#loc247) + %b_do_140 = ttg.memdesc_index %b_do[%c0_i32] : !ttg.memdesc<2x64x64xbf16, #shared1, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared1, #smem, mutable> loc(#loc215) + %b_dh2_141 = tt.splat %b_dh2 : i1 -> tensor<64x64xi1, #blocked1> loc(#loc283) + %b_dh2_142 = arith.andi %b_dh2_141, %b_do_139 : tensor<64x64xi1, #blocked1> loc(#loc283) + %b_do_143 = ttg.async_copy_global_to_local %p_do_136, %b_do_140 mask %b_dh2_142 other %cst_12 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked1> -> <64x64xbf16, #shared1, #smem, mutable> loc(#loc215) + %b_do_144 = ttg.async_commit_group tokens %b_do_143 loc(#loc215) + %p_k_145 = tt.addptr %p_k, %p_dv_130 : tensor<64x1x!tt.ptr, #blocked1>, tensor<64x1xi64, #blocked1> loc(#loc195) + %p_k_146 = tt.broadcast %p_k_145 : tensor<64x1x!tt.ptr, #blocked1> -> tensor<64x64x!tt.ptr, #blocked1> loc(#loc197) + %p_k_147 = tt.addptr %p_k_146, %p_k_70 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi32, #blocked1> loc(#loc197) + %b_k_148 = arith.andi %b_do_138, %b_k_71 : tensor<64x64xi1, #blocked1> loc(#loc199) + %b_k_149 = ttg.memdesc_index %b_k_85[%c0_i32] : !ttg.memdesc<2x64x64xbf16, #shared2, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared2, #smem, mutable> loc(#loc216) + %b_dh2_150 = arith.andi %b_dh2_141, %b_k_148 : tensor<64x64xi1, #blocked1> loc(#loc283) + %b_k_151 = ttg.async_copy_global_to_local %p_k_147, %b_k_149 mask %b_dh2_150 other %cst_12 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked1> -> <64x64xbf16, #shared2, #smem, mutable> loc(#loc216) + %b_k_152 = ttg.async_commit_group tokens %b_k_151 loc(#loc216) + %p_k_153 = tt.addptr %p_k_146, %p_k_73 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi32, #blocked1> loc(#loc201) + %b_k_154 = arith.andi %b_do_138, %b_k_75 : tensor<64x64xi1, #blocked1> loc(#loc203) + %b_k_155 = ttg.memdesc_index %b_k_86[%c0_i32] : !ttg.memdesc<2x64x64xbf16, #shared2, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared2, #smem, mutable> loc(#loc217) + %b_dh2_156 = arith.andi %b_dh2_141, %b_k_154 : tensor<64x64xi1, #blocked1> loc(#loc283) + %b_k_157 = ttg.async_copy_global_to_local %p_k_153, %b_k_155 mask %b_dh2_156 other %cst_12 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked1> -> <64x64xbf16, #shared2, #smem, mutable> loc(#loc217) + %b_k_158 = ttg.async_commit_group tokens %b_k_157 loc(#loc217) + %b_dv_159 = ttg.memdesc_index %b_dv[%c0_i32] : !ttg.memdesc<2x64x64xbf16, #shared1, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared1, #smem, mutable> loc(#loc218) + %b_dv_160 = ttg.async_copy_global_to_local %p_dv_133, %b_dv_159 mask %b_dh2_142 other %cst_12 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked1> -> <64x64xbf16, #shared1, #smem, mutable> loc(#loc218) + %b_dv_161 = ttg.async_commit_group tokens %b_dv_160 loc(#loc218) + %p_w_162 = tt.expand_dims %o_t_96 {axis = 0 : i32} : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked2}>> -> tensor<1x64xi64, #blocked2> loc(#loc248) + %p_w_163 = arith.muli %p_w_162, %cst_4 : tensor<1x64xi64, #blocked2> loc(#loc249) + %p_w_164 = tt.broadcast %p_w_163 : tensor<1x64xi64, #blocked2> -> tensor<64x64xi64, #blocked2> loc(#loc205) + %p_w_165 = tt.addptr %p_w_77, %p_w_164 : tensor<64x64x!tt.ptr, #blocked2>, tensor<64x64xi64, #blocked2> loc(#loc205) + %p_q_166 = tt.addptr %p_q_79, %p_w_164 : tensor<64x64x!tt.ptr, #blocked2>, tensor<64x64xi64, #blocked2> loc(#loc207) + %b_w_167 = tt.expand_dims %m_t_100 {axis = 0 : i32} : tensor<64xi1, #ttg.slice<{dim = 0, parent = #blocked2}>> -> tensor<1x64xi1, #blocked2> loc(#loc250) + %b_w_168 = tt.broadcast %b_w_167 : tensor<1x64xi1, #blocked2> -> tensor<64x64xi1, #blocked2> loc(#loc251) + %b_w_169 = arith.andi %m_dh1_59, %b_w_168 : tensor<64x64xi1, #blocked2> loc(#loc251) + %b_w_170 = ttg.memdesc_index %b_w[%c0_i32] : !ttg.memdesc<2x64x64xbf16, #shared3, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared3, #smem, mutable> loc(#loc219) + %b_dh2_171 = tt.splat %b_dh2 : i1 -> tensor<64x64xi1, #blocked2> loc(#loc283) + %b_dh2_172 = arith.andi %b_dh2_171, %b_w_169 : tensor<64x64xi1, #blocked2> loc(#loc283) + %b_w_173 = ttg.async_copy_global_to_local %p_w_165, %b_w_170 mask %b_dh2_172 other %cst_13 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked2> -> <64x64xbf16, #shared3, #smem, mutable> loc(#loc219) + %b_w_174 = ttg.async_commit_group tokens %b_w_173 loc(#loc219) + %b_q_175 = ttg.memdesc_index %b_q[%c0_i32] : !ttg.memdesc<2x64x64xbf16, #shared4, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared4, #smem, mutable> loc(#loc220) + %b_q_176 = ttg.async_copy_global_to_local %p_q_166, %b_q_175 mask %b_dh2_172 other %cst_13 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked2> -> <64x64xbf16, #shared4, #smem, mutable> loc(#loc220) + %b_q_177 = ttg.async_commit_group tokens %b_q_176 loc(#loc220) + %p_q_178 = tt.addptr %p_q_81, %p_w_164 : tensor<64x64x!tt.ptr, #blocked2>, tensor<64x64xi64, #blocked2> loc(#loc210) + %p_w_179 = tt.addptr %p_w_83, %p_w_164 : tensor<64x64x!tt.ptr, #blocked2>, tensor<64x64xi64, #blocked2> loc(#loc212) + %b_q_180 = arith.andi %m_dh2_66, %b_w_168 : tensor<64x64xi1, #blocked2> loc(#loc252) + %b_q_181 = ttg.memdesc_index %b_q_87[%c0_i32] : !ttg.memdesc<2x64x64xbf16, #shared4, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared4, #smem, mutable> loc(#loc221) + %b_dh2_182 = arith.andi %b_dh2_171, %b_q_180 : tensor<64x64xi1, #blocked2> loc(#loc283) + %b_q_183 = ttg.async_copy_global_to_local %p_q_178, %b_q_181 mask %b_dh2_182 other %cst_13 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked2> -> <64x64xbf16, #shared4, #smem, mutable> loc(#loc221) + %b_q_184 = ttg.async_commit_group tokens %b_q_183 loc(#loc221) + %b_w_185 = ttg.memdesc_index %b_w_88[%c0_i32] : !ttg.memdesc<2x64x64xbf16, #shared3, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared3, #smem, mutable> loc(#loc222) + %b_w_186 = ttg.async_copy_global_to_local %p_w_179, %b_w_185 mask %b_dh2_182 other %cst_13 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked2> -> <64x64xbf16, #shared3, #smem, mutable> loc(#loc222) + %b_w_187 = ttg.async_commit_group tokens %b_w_186 loc(#loc222) + %b_dh2_188 = arith.cmpi sgt, %6, %c0_i32 : i32 loc(#loc283) + %i_t = arith.addi %NT_16, %c-2_i32 : i32 loc(#loc285) + %i_t_int64_189 = arith.extsi %i_t : i32 to i64 loc(#loc224) + %o_t_190 = arith.muli %i_t_int64_189, %c64_i64 : i64 loc(#loc225) + %o_t_191 = tt.splat %o_t_190 : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc226) + %o_t_192 = tt.splat %o_t_190 : i64 -> tensor<64xi64, #blocked> loc(#loc226) + %o_t_193 = tt.splat %o_t_190 : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #linear}>> loc(#loc226) + %o_t_194 = tt.splat %o_t_190 : i64 -> tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked2}>> loc(#loc226) + %o_t_195 = arith.addi %o_t_191, %o_v_39 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc226) + %o_t_196 = arith.addi %o_t_192, %o_v_37 : tensor<64xi64, #blocked> loc(#loc226) + %o_t_197 = arith.addi %o_t_193, %o_v_40 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #linear}>> loc(#loc226) + %o_t_198 = arith.addi %o_t_194, %o_v_38 : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked2}>> loc(#loc226) + %m_t_199 = arith.cmpi slt, %o_t_195, %m_t : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc177) + %m_t_200 = arith.cmpi slt, %o_t_196, %m_t_49 : tensor<64xi64, #blocked> loc(#loc177) + %m_t_201 = arith.cmpi slt, %o_t_197, %m_t_50 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #linear}>> loc(#loc177) + %m_t_202 = arith.cmpi slt, %o_t_198, %m_t_51 : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked2}>> loc(#loc177) + %last_idx_203 = arith.addi %i_t_int64_189, %c1_i64 : i64 loc(#loc227) + %last_idx_204 = arith.muli %last_idx_203, %c64_i64 : i64 loc(#loc228) + %last_idx_205 = arith.minsi %last_idx_204, %4 : i64 loc(#loc229) + %last_idx_206 = arith.subi %last_idx_205, %c1_i64 : i64 loc(#loc230) + %bg_last_207 = arith.addi %5, %last_idx_206 : i64 loc(#loc231) + %bg_last_208 = arith.muli %bg_last_207, %c32_i64 : i64 loc(#loc232) + %bg_last_209 = arith.addi %bg_last_208, %3 : i64 loc(#loc284) + %bg_last_210 = tt.addptr %g, %bg_last_209 : !tt.ptr, i64 loc(#loc284) + %bg_last_211 = tt.splat %bg_last_210 : !tt.ptr -> tensor<1x!tt.ptr, #blocked> loc(#loc213) + %bg_last_212 = ttg.memdesc_index %bg_last[%c1_i32] : !ttg.memdesc<2x1xf32, #shared, #smem, mutable> -> !ttg.memdesc<1xf32, #shared, #smem, mutable> loc(#loc213) + %b_dh2_213 = tt.splat %b_dh2_188 : i1 -> tensor<1xi1, #blocked> loc(#loc283) + %bg_last_214 = ttg.async_copy_global_to_local %bg_last_211, %bg_last_212 mask %b_dh2_213 : tensor<1x!tt.ptr, #blocked> -> <1xf32, #shared, #smem, mutable> loc(#loc213) + %bg_last_215 = ttg.async_commit_group tokens %bg_last_214 loc(#loc213) + %p_g_216 = arith.muli %o_t_196, %cst : tensor<64xi64, #blocked> loc(#loc241) + %p_g_217 = tt.addptr %p_g_68, %p_g_216 : tensor<64x!tt.ptr, #blocked>, tensor<64xi64, #blocked> loc(#loc191) + %b_g_218 = ttg.memdesc_index %b_g[%c1_i32] : !ttg.memdesc<2x64xf32, #shared, #smem, mutable> -> !ttg.memdesc<64xf32, #shared, #smem, mutable> loc(#loc214) + %b_dh2_219 = tt.splat %b_dh2_188 : i1 -> tensor<64xi1, #blocked> loc(#loc283) + %b_dh2_220 = arith.andi %b_dh2_219, %m_t_200 : tensor<64xi1, #blocked> loc(#loc283) + %b_g_221 = ttg.async_copy_global_to_local %p_g_217, %b_g_218 mask %b_dh2_220 other %cst_0 : tensor<64x!tt.ptr, #blocked> -> <64xf32, #shared, #smem, mutable> loc(#loc214) + %b_g_222 = ttg.async_commit_group tokens %b_g_221 loc(#loc214) + %b_g_223 = ttg.memdesc_index %b_g_84[%c1_i32] : !ttg.memdesc<2x64xf32, #shared, #smem, mutable> -> !ttg.memdesc<64xf32, #shared, #smem, mutable> loc(#loc214) + %b_g_224 = ttg.async_copy_global_to_local %p_g_217, %b_g_223 mask %b_dh2_220 other %cst_0 : tensor<64x!tt.ptr, #blocked> -> <64xf32, #shared, #smem, mutable> loc(#loc214) + %b_g_225 = ttg.async_commit_group tokens %b_g_224 loc(#loc214) + %p_dv_226 = tt.expand_dims %o_t_195 {axis = 1 : i32} : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<64x1xi64, #blocked1> loc(#loc242) + %p_dv_227 = arith.muli %p_dv_226, %cst_2 : tensor<64x1xi64, #blocked1> loc(#loc243) + %p_dv_228 = tt.addptr %p_dv, %p_dv_227 : tensor<64x1x!tt.ptr, #blocked1>, tensor<64x1xi64, #blocked1> loc(#loc192) + %p_dv_229 = tt.broadcast %p_dv_228 : tensor<64x1x!tt.ptr, #blocked1> -> tensor<64x64x!tt.ptr, #blocked1> loc(#loc244) + %p_dv_230 = tt.addptr %p_dv_229, %p_dh1_55 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi64, #blocked1> loc(#loc244) + %p_do_231 = tt.addptr %p_do, %p_dv_227 : tensor<64x1x!tt.ptr, #blocked1>, tensor<64x1xi64, #blocked1> loc(#loc194) + %p_do_232 = tt.broadcast %p_do_231 : tensor<64x1x!tt.ptr, #blocked1> -> tensor<64x64x!tt.ptr, #blocked1> loc(#loc245) + %p_do_233 = tt.addptr %p_do_232, %p_dh1_55 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi64, #blocked1> loc(#loc245) + %b_do_234 = tt.expand_dims %m_t_199 {axis = 1 : i32} : tensor<64xi1, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<64x1xi1, #blocked1> loc(#loc246) + %b_do_235 = tt.broadcast %b_do_234 : tensor<64x1xi1, #blocked1> -> tensor<64x64xi1, #blocked1> loc(#loc247) + %b_do_236 = arith.andi %b_do_235, %m_dh1_60 : tensor<64x64xi1, #blocked1> loc(#loc247) + %b_do_237 = ttg.memdesc_index %b_do[%c1_i32] : !ttg.memdesc<2x64x64xbf16, #shared1, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared1, #smem, mutable> loc(#loc215) + %b_dh2_238 = tt.splat %b_dh2_188 : i1 -> tensor<64x64xi1, #blocked1> loc(#loc283) + %b_dh2_239 = arith.andi %b_dh2_238, %b_do_236 : tensor<64x64xi1, #blocked1> loc(#loc283) + %b_do_240 = ttg.async_copy_global_to_local %p_do_233, %b_do_237 mask %b_dh2_239 other %cst_12 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked1> -> <64x64xbf16, #shared1, #smem, mutable> loc(#loc215) + %b_do_241 = ttg.async_commit_group tokens %b_do_240 loc(#loc215) + %p_k_242 = tt.addptr %p_k, %p_dv_227 : tensor<64x1x!tt.ptr, #blocked1>, tensor<64x1xi64, #blocked1> loc(#loc195) + %p_k_243 = tt.broadcast %p_k_242 : tensor<64x1x!tt.ptr, #blocked1> -> tensor<64x64x!tt.ptr, #blocked1> loc(#loc197) + %p_k_244 = tt.addptr %p_k_243, %p_k_70 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi32, #blocked1> loc(#loc197) + %b_k_245 = arith.andi %b_do_235, %b_k_71 : tensor<64x64xi1, #blocked1> loc(#loc199) + %b_k_246 = ttg.memdesc_index %b_k_85[%c1_i32] : !ttg.memdesc<2x64x64xbf16, #shared2, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared2, #smem, mutable> loc(#loc216) + %b_dh2_247 = arith.andi %b_dh2_238, %b_k_245 : tensor<64x64xi1, #blocked1> loc(#loc283) + %b_k_248 = ttg.async_copy_global_to_local %p_k_244, %b_k_246 mask %b_dh2_247 other %cst_12 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked1> -> <64x64xbf16, #shared2, #smem, mutable> loc(#loc216) + %b_k_249 = ttg.async_commit_group tokens %b_k_248 loc(#loc216) + %p_k_250 = tt.addptr %p_k_243, %p_k_73 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi32, #blocked1> loc(#loc201) + %b_k_251 = arith.andi %b_do_235, %b_k_75 : tensor<64x64xi1, #blocked1> loc(#loc203) + %b_k_252 = ttg.memdesc_index %b_k_86[%c1_i32] : !ttg.memdesc<2x64x64xbf16, #shared2, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared2, #smem, mutable> loc(#loc217) + %b_dh2_253 = arith.andi %b_dh2_238, %b_k_251 : tensor<64x64xi1, #blocked1> loc(#loc283) + %b_k_254 = ttg.async_copy_global_to_local %p_k_250, %b_k_252 mask %b_dh2_253 other %cst_12 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked1> -> <64x64xbf16, #shared2, #smem, mutable> loc(#loc217) + %b_k_255 = ttg.async_commit_group tokens %b_k_254 loc(#loc217) + %b_dv_256 = ttg.memdesc_index %b_dv[%c1_i32] : !ttg.memdesc<2x64x64xbf16, #shared1, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared1, #smem, mutable> loc(#loc218) + %b_dv_257 = ttg.async_copy_global_to_local %p_dv_230, %b_dv_256 mask %b_dh2_239 other %cst_12 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked1> -> <64x64xbf16, #shared1, #smem, mutable> loc(#loc218) + %b_dv_258 = ttg.async_commit_group tokens %b_dv_257 loc(#loc218) + %p_w_259 = tt.expand_dims %o_t_198 {axis = 0 : i32} : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked2}>> -> tensor<1x64xi64, #blocked2> loc(#loc248) + %p_w_260 = arith.muli %p_w_259, %cst_4 : tensor<1x64xi64, #blocked2> loc(#loc249) + %p_w_261 = tt.broadcast %p_w_260 : tensor<1x64xi64, #blocked2> -> tensor<64x64xi64, #blocked2> loc(#loc205) + %p_w_262 = tt.addptr %p_w_77, %p_w_261 : tensor<64x64x!tt.ptr, #blocked2>, tensor<64x64xi64, #blocked2> loc(#loc205) + %p_q_263 = tt.addptr %p_q_79, %p_w_261 : tensor<64x64x!tt.ptr, #blocked2>, tensor<64x64xi64, #blocked2> loc(#loc207) + %b_w_264 = tt.expand_dims %m_t_202 {axis = 0 : i32} : tensor<64xi1, #ttg.slice<{dim = 0, parent = #blocked2}>> -> tensor<1x64xi1, #blocked2> loc(#loc250) + %b_w_265 = tt.broadcast %b_w_264 : tensor<1x64xi1, #blocked2> -> tensor<64x64xi1, #blocked2> loc(#loc251) + %b_w_266 = arith.andi %m_dh1_59, %b_w_265 : tensor<64x64xi1, #blocked2> loc(#loc251) + %b_w_267 = ttg.memdesc_index %b_w[%c1_i32] : !ttg.memdesc<2x64x64xbf16, #shared3, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared3, #smem, mutable> loc(#loc219) + %b_dh2_268 = tt.splat %b_dh2_188 : i1 -> tensor<64x64xi1, #blocked2> loc(#loc283) + %b_dh2_269 = arith.andi %b_dh2_268, %b_w_266 : tensor<64x64xi1, #blocked2> loc(#loc283) + %b_w_270 = ttg.async_copy_global_to_local %p_w_262, %b_w_267 mask %b_dh2_269 other %cst_13 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked2> -> <64x64xbf16, #shared3, #smem, mutable> loc(#loc219) + %b_w_271 = ttg.async_commit_group tokens %b_w_270 loc(#loc219) + %b_q_272 = ttg.memdesc_index %b_q[%c1_i32] : !ttg.memdesc<2x64x64xbf16, #shared4, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared4, #smem, mutable> loc(#loc220) + %b_q_273 = ttg.async_copy_global_to_local %p_q_263, %b_q_272 mask %b_dh2_269 other %cst_13 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked2> -> <64x64xbf16, #shared4, #smem, mutable> loc(#loc220) + %b_q_274 = ttg.async_commit_group tokens %b_q_273 loc(#loc220) + %p_q_275 = tt.addptr %p_q_81, %p_w_261 : tensor<64x64x!tt.ptr, #blocked2>, tensor<64x64xi64, #blocked2> loc(#loc210) + %p_w_276 = tt.addptr %p_w_83, %p_w_261 : tensor<64x64x!tt.ptr, #blocked2>, tensor<64x64xi64, #blocked2> loc(#loc212) + %b_q_277 = arith.andi %m_dh2_66, %b_w_265 : tensor<64x64xi1, #blocked2> loc(#loc252) + %b_q_278 = ttg.memdesc_index %b_q_87[%c1_i32] : !ttg.memdesc<2x64x64xbf16, #shared4, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared4, #smem, mutable> loc(#loc221) + %b_dh2_279 = arith.andi %b_dh2_268, %b_q_277 : tensor<64x64xi1, #blocked2> loc(#loc283) + %b_q_280 = ttg.async_copy_global_to_local %p_q_275, %b_q_278 mask %b_dh2_279 other %cst_13 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked2> -> <64x64xbf16, #shared4, #smem, mutable> loc(#loc221) + %b_q_281 = ttg.async_commit_group tokens %b_q_280 loc(#loc221) + %b_w_282 = ttg.memdesc_index %b_w_88[%c1_i32] : !ttg.memdesc<2x64x64xbf16, #shared3, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared3, #smem, mutable> loc(#loc222) + %b_w_283 = ttg.async_copy_global_to_local %p_w_276, %b_w_282 mask %b_dh2_279 other %cst_13 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked2> -> <64x64xbf16, #shared3, #smem, mutable> loc(#loc222) + %b_w_284 = ttg.async_commit_group tokens %b_w_283 loc(#loc222) + %b_dh2_285:34 = scf.for %b_dh2_287 = %c-1_i32 to %6 step %c1_i32 iter_args(%arg11 = %cst_14, %arg12 = %cst_14, %b_dh2_288 = %c1_i32, %b_dh2_289 = %c-1_i32, %i_t_int64_290 = %i_t_int64, %i_t_int64_291 = %i_t_int64_189, %bg_last_292 = %bg_last_118, %bg_last_293 = %bg_last_215, %b_g_294 = %b_g_125, %b_g_295 = %b_g_222, %b_g_296 = %b_g_128, %b_g_297 = %b_g_225, %p_dv_298 = %p_dv_130, %p_dv_299 = %p_dv_227, %b_do_300 = %b_do_144, %b_do_301 = %b_do_241, %b_k_302 = %b_k_152, %b_k_303 = %b_k_249, %b_k_304 = %b_k_158, %b_k_305 = %b_k_255, %m_t_306 = %m_t_99, %m_t_307 = %m_t_201, %b_dv_308 = %b_dv_161, %b_dv_309 = %b_dv_258, %b_do_310 = %b_do_139, %b_do_311 = %b_do_236, %b_w_312 = %b_w_174, %b_w_313 = %b_w_271, %b_q_314 = %b_q_177, %b_q_315 = %b_q_274, %b_q_316 = %b_q_184, %b_q_317 = %b_q_281, %b_w_318 = %b_w_187, %b_w_319 = %b_w_284) -> (tensor<64x64xf32, #linear>, tensor<64x64xf32, #linear>, i32, i32, i64, i64, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, tensor<64x1xi64, #blocked1>, tensor<64x1xi64, #blocked1>, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, tensor<64xi1, #ttg.slice<{dim = 1, parent = #linear}>>, tensor<64xi1, #ttg.slice<{dim = 1, parent = #linear}>>, !ttg.async.token, !ttg.async.token, tensor<64x64xi1, #blocked1>, tensor<64x64xi1, #blocked1>, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token) : i32 { + %b_dh2_320 = arith.subi %NT_16, %c3_i32 : i32 loc(#loc283) + %b_dh2_321 = arith.cmpi slt, %b_dh2_287, %b_dh2_320 : i32 loc(#loc283) + %b_dh2_322 = arith.addi %b_dh2_289, %c1_i32 : i32 loc(#loc283) + %b_dh2_323 = arith.cmpi sge, %b_dh2_322, %c2_i32 : i32 loc(#loc283) + %b_dh2_324 = arith.select %b_dh2_323, %c0_i32, %b_dh2_322 : i32 loc(#loc283) + %p_dh1_325 = arith.muli %i_t_int64_290, %c524288_i64 : i64 loc(#loc286) + %p_dh1_326 = arith.addi %dh_29, %p_dh1_325 : i64 loc(#loc287) + %p_dh1_327 = tt.addptr %dh, %p_dh1_326 : !tt.ptr, i64 loc(#loc287) + %p_dh1_328 = tt.splat %p_dh1_327 : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked1> loc(#loc257) + %p_dh1_329 = tt.addptr %p_dh1_328, %p_dh1_53 : tensor<64x1x!tt.ptr, #blocked1>, tensor<64x1xi32, #blocked1> loc(#loc257) + %p_dh1_330 = tt.broadcast %p_dh1_329 : tensor<64x1x!tt.ptr, #blocked1> -> tensor<64x64x!tt.ptr, #blocked1> loc(#loc181) + %p_dh1_331 = tt.addptr %p_dh1_330, %p_dh1_55 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi64, #blocked1> loc(#loc181) + %7 = arith.truncf %arg11 : tensor<64x64xf32, #linear> to tensor<64x64xbf16, #linear> loc(#loc114) + %8 = ttg.local_alloc %7 : (tensor<64x64xbf16, #linear>) -> !ttg.memdesc<64x64xbf16, #shared2, #smem> loc(#loc114) + %9 = ttg.convert_layout %7 : tensor<64x64xbf16, #linear> -> tensor<64x64xbf16, #blocked1> loc(#loc115) + tt.store %p_dh1_331, %9, %m_dh1_61 : tensor<64x64x!tt.ptr, #blocked1> loc(#loc115) + %p_dh2_332 = tt.addptr %p_dh1_328, %p_dh2_63 : tensor<64x1x!tt.ptr, #blocked1>, tensor<64x1xi32, #blocked1> loc(#loc258) + %p_dh2_333 = tt.broadcast %p_dh2_332 : tensor<64x1x!tt.ptr, #blocked1> -> tensor<64x64x!tt.ptr, #blocked1> loc(#loc259) + %p_dh2_334 = tt.addptr %p_dh2_333, %p_dh1_55 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi64, #blocked1> loc(#loc259) + %10 = arith.truncf %arg12 : tensor<64x64xf32, #linear> to tensor<64x64xbf16, #linear> loc(#loc118) + %11 = ttg.local_alloc %10 : (tensor<64x64xbf16, #linear>) -> !ttg.memdesc<64x64xbf16, #shared2, #smem> loc(#loc118) + %12 = ttg.convert_layout %10 : tensor<64x64xbf16, #linear> -> tensor<64x64xbf16, #blocked1> loc(#loc119) + tt.store %p_dh2_334, %12, %m_dh2_67 : tensor<64x64x!tt.ptr, #blocked1> loc(#loc119) + %bg_last_335 = ttg.async_wait %bg_last_292, %b_g_294, %b_g_296, %b_do_300, %b_k_302, %b_k_304, %b_dv_308, %b_w_312, %b_q_314, %b_q_316, %b_w_318 {num = 11 : i32} loc(#loc213) + %bg_last_336 = ttg.memdesc_index %bg_last[%b_dh2_324] : !ttg.memdesc<2x1xf32, #shared, #smem, mutable> -> !ttg.memdesc<1xf32, #shared, #smem, mutable> loc(#loc213) + %bg_last_337 = ttg.local_load %bg_last_336 token %bg_last_335 : !ttg.memdesc<1xf32, #shared, #smem, mutable> -> tensor<1xf32, #blocked> loc(#loc213) + %bg_last_338 = tt.unsplat %bg_last_337 : tensor<1xf32, #blocked> loc(#loc213) + %b_g_339 = ttg.memdesc_index %b_g[%b_dh2_324] : !ttg.memdesc<2x64xf32, #shared, #smem, mutable> -> !ttg.memdesc<64xf32, #shared, #smem, mutable> loc(#loc214) + %b_g_340 = ttg.local_load %b_g_339 token %bg_last_335 : !ttg.memdesc<64xf32, #shared, #smem, mutable> -> tensor<64xf32, #ttg.slice<{dim = 1, parent = #linear}>> loc(#loc214) + %b_g_341 = ttg.memdesc_index %b_g_84[%b_dh2_324] : !ttg.memdesc<2x64xf32, #shared, #smem, mutable> -> !ttg.memdesc<64xf32, #shared, #smem, mutable> loc(#loc214) + %b_g_342 = ttg.local_load %b_g_341 token %bg_last_335 : !ttg.memdesc<64xf32, #shared, #smem, mutable> -> tensor<64xf32, #ttg.slice<{dim = 0, parent = #linear}>> loc(#loc214) + %bg_last_exp = math.exp2 %bg_last_338 : f32 loc(#loc288) + %b_g_exp = math.exp2 %b_g_342 : tensor<64xf32, #ttg.slice<{dim = 0, parent = #linear}>> loc(#loc289) + %p_dv2_343 = tt.addptr %p_dv2, %p_dv_298 : tensor<64x1x!tt.ptr, #blocked1>, tensor<64x1xi64, #blocked1> loc(#loc193) + %p_dv2_344 = tt.broadcast %p_dv2_343 : tensor<64x1x!tt.ptr, #blocked1> -> tensor<64x64x!tt.ptr, #blocked1> loc(#loc262) + %p_dv2_345 = tt.addptr %p_dv2_344, %p_dh1_55 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi64, #blocked1> loc(#loc262) + %b_do_346 = ttg.memdesc_index %b_do[%b_dh2_324] : !ttg.memdesc<2x64x64xbf16, #shared1, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared1, #smem, mutable> loc(#loc215) + %b_do_347 = ttg.local_load %b_do_346 token %bg_last_335 : !ttg.memdesc<64x64xbf16, #shared1, #smem, mutable> -> tensor<64x64xbf16, #blocked1> loc(#loc215) + %b_k_348 = ttg.memdesc_index %b_k_85[%b_dh2_324] : !ttg.memdesc<2x64x64xbf16, #shared2, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared2, #smem, mutable> loc(#loc216) + ttng.fence_async_shared {bCluster = false} loc(#loc235) + %b_dv_349 = ttg.local_alloc : () -> !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc235) + ttng.init_barrier %b_dv_349, 1 : !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc235) + ttng.tc_gen5_mma %b_k_348, %8, %b_dv_112, %false, %true, %b_dv_349[%true] {is_async} : !ttg.memdesc<64x64xbf16, #shared2, #smem, mutable>, !ttg.memdesc<64x64xbf16, #shared2, #smem>, !ttg.memdesc<64x64xf32, #tmem, #ttng.tensor_memory, mutable>, !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc235) + ttng.wait_barrier %b_dv_349, %c0_i32, %true : !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc235) + ttng.inval_barrier %b_dv_349 : !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc235) + %b_k_350 = ttg.memdesc_index %b_k_86[%b_dh2_324] : !ttg.memdesc<2x64x64xbf16, #shared2, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared2, #smem, mutable> loc(#loc217) + %b_dv_351 = ttng.tmem_load %b_dv_112 : !ttg.memdesc<64x64xf32, #tmem, #ttng.tensor_memory, mutable> -> tensor<64x64xf32, #linear> loc(#loc235) + ttng.tmem_store %b_dv_351, %b_dv_113, %true : tensor<64x64xf32, #linear> -> !ttg.memdesc<64x64xf32, #tmem, #ttng.tensor_memory, mutable> loc(#loc236) + ttng.fence_async_shared {bCluster = false} loc(#loc236) + %b_dv_352 = ttg.local_alloc : () -> !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc236) + ttng.init_barrier %b_dv_352, 1 : !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc236) + ttng.tc_gen5_mma %b_k_350, %11, %b_dv_113, %true, %true, %b_dv_352[%true] {is_async} : !ttg.memdesc<64x64xbf16, #shared2, #smem, mutable>, !ttg.memdesc<64x64xbf16, #shared2, #smem>, !ttg.memdesc<64x64xf32, #tmem, #ttng.tensor_memory, mutable>, !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc236) + ttng.wait_barrier %b_dv_352, %c0_i32, %true : !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc236) + ttng.inval_barrier %b_dv_352 : !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc236) + %b_dv_353 = tt.splat %bg_last_338 : f32 -> tensor<64xf32, #ttg.slice<{dim = 1, parent = #linear}>> loc(#loc263) + %b_dv_354 = arith.subf %b_dv_353, %b_g_340 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #linear}>> loc(#loc263) + %b_dv_355 = math.exp2 %b_dv_354 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #linear}>> loc(#loc290) + %b_dv_356 = arith.select %m_t_306, %b_dv_355, %cst_1 : tensor<64xi1, #ttg.slice<{dim = 1, parent = #linear}>>, tensor<64xf32, #ttg.slice<{dim = 1, parent = #linear}>> loc(#loc265) + %b_dv_357 = tt.expand_dims %b_dv_356 {axis = 1 : i32} : tensor<64xf32, #ttg.slice<{dim = 1, parent = #linear}>> -> tensor<64x1xf32, #linear> loc(#loc266) + %b_dv_358 = tt.broadcast %b_dv_357 : tensor<64x1xf32, #linear> -> tensor<64x64xf32, #linear> loc(#loc267) + %b_dv_359 = ttg.memdesc_index %b_dv[%b_dh2_324] : !ttg.memdesc<2x64x64xbf16, #shared1, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared1, #smem, mutable> loc(#loc218) + %b_dv_360 = ttg.local_load %b_dv_359 token %bg_last_335 : !ttg.memdesc<64x64xbf16, #shared1, #smem, mutable> -> tensor<64x64xbf16, #linear> loc(#loc218) + %b_dv_361 = arith.extf %b_dv_360 : tensor<64x64xbf16, #linear> to tensor<64x64xf32, #linear> loc(#loc268) + %b_dv_362 = ttng.tmem_load %b_dv_113 : !ttg.memdesc<64x64xf32, #tmem, #ttng.tensor_memory, mutable> -> tensor<64x64xf32, #linear> loc(#loc236) + %b_dv_363 = arith.mulf %b_dv_362, %b_dv_358 : tensor<64x64xf32, #linear> loc(#loc267) + %b_dv_364 = arith.addf %b_dv_363, %b_dv_361 : tensor<64x64xf32, #linear> loc(#loc268) + %13 = arith.truncf %b_dv_364 : tensor<64x64xf32, #linear> to tensor<64x64xbf16, #linear> loc(#loc130) + %14 = ttg.local_alloc %13 : (tensor<64x64xbf16, #linear>) -> !ttg.memdesc<64x64xbf16, #shared2, #smem> loc(#loc130) + %15 = ttg.convert_layout %13 : tensor<64x64xbf16, #linear> -> tensor<64x64xbf16, #blocked1> loc(#loc130) + tt.store %p_dv2_345, %15, %b_do_310 : tensor<64x64x!tt.ptr, #blocked1> loc(#loc131) + %b_w_365 = ttg.memdesc_index %b_w[%b_dh2_324] : !ttg.memdesc<2x64x64xbf16, #shared3, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared3, #smem, mutable> loc(#loc219) + %b_q_366 = ttg.memdesc_index %b_q[%b_dh2_324] : !ttg.memdesc<2x64x64xbf16, #shared4, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared4, #smem, mutable> loc(#loc220) + %b_q_367 = ttg.local_load %b_q_366 token %bg_last_335 : !ttg.memdesc<64x64xbf16, #shared4, #smem, mutable> -> tensor<64x64xbf16, #linear> loc(#loc220) + %b_dh1_368 = tt.splat %bg_last_exp : f32 -> tensor<64x64xf32, #linear> loc(#loc269) + %b_dh1_369 = arith.mulf %arg11, %b_dh1_368 : tensor<64x64xf32, #linear> loc(#loc269) + %b_q_370 = tt.expand_dims %b_g_exp {axis = 0 : i32} : tensor<64xf32, #ttg.slice<{dim = 0, parent = #linear}>> -> tensor<1x64xf32, #linear> loc(#loc270) + %b_q_371 = arith.extf %b_q_367 : tensor<64x64xbf16, #linear> to tensor<64x64xf32, #linear> loc(#loc271) + %b_q_372 = tt.broadcast %b_q_370 : tensor<1x64xf32, #linear> -> tensor<64x64xf32, #linear> loc(#loc271) + %b_q_373 = arith.mulf %b_q_371, %b_q_372 : tensor<64x64xf32, #linear> loc(#loc271) + %b_dh1_374 = arith.extf %b_do_347 : tensor<64x64xbf16, #blocked1> to tensor<64x64xf32, #blocked1> loc(#loc272) + %b_dh1_375 = ttg.local_alloc %b_dh1_374 : (tensor<64x64xf32, #blocked1>) -> !ttg.memdesc<64x64xf32, #shared5, #smem> loc(#loc272) + %b_dh1_376 = ttng.tmem_alloc %b_q_373 : (tensor<64x64xf32, #linear>) -> !ttg.memdesc<64x64xf32, #tmem, #ttng.tensor_memory> loc(#loc237) + ttng.fence_async_shared {bCluster = false} loc(#loc237) + %b_dh1_377 = ttg.local_alloc : () -> !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc237) + ttng.init_barrier %b_dh1_377, 1 : !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc237) + ttng.tc_gen5_mma %b_dh1_376, %b_dh1_375, %b_dh1_114, %false, %true, %b_dh1_377[%true] {is_async} : !ttg.memdesc<64x64xf32, #tmem, #ttng.tensor_memory>, !ttg.memdesc<64x64xf32, #shared5, #smem>, !ttg.memdesc<64x64xf32, #tmem, #ttng.tensor_memory, mutable>, !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc237) + ttng.wait_barrier %b_dh1_377, %c0_i32, %true : !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc237) + ttng.inval_barrier %b_dh1_377 : !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc237) + ttng.fence_async_shared {bCluster = false} loc(#loc238) + %b_dh1_378 = ttg.local_alloc : () -> !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc238) + ttng.init_barrier %b_dh1_378, 1 : !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc238) + ttng.tc_gen5_mma %b_w_365, %14, %b_dh1_115, %false, %true, %b_dh1_378[%true] {is_async} : !ttg.memdesc<64x64xbf16, #shared3, #smem, mutable>, !ttg.memdesc<64x64xbf16, #shared2, #smem>, !ttg.memdesc<64x64xf32, #tmem, #ttng.tensor_memory, mutable>, !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc238) + ttng.wait_barrier %b_dh1_378, %c0_i32, %true : !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc238) + ttng.inval_barrier %b_dh1_378 : !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc238) + %b_q_379 = ttg.memdesc_index %b_q_87[%b_dh2_324] : !ttg.memdesc<2x64x64xbf16, #shared4, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared4, #smem, mutable> loc(#loc221) + %b_q_380 = ttg.local_load %b_q_379 token %bg_last_335 : !ttg.memdesc<64x64xbf16, #shared4, #smem, mutable> -> tensor<64x64xbf16, #linear> loc(#loc221) + %b_w_381 = ttg.memdesc_index %b_w_88[%b_dh2_324] : !ttg.memdesc<2x64x64xbf16, #shared3, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared3, #smem, mutable> loc(#loc222) + %b_dh2_382 = arith.mulf %arg12, %b_dh1_368 : tensor<64x64xf32, #linear> loc(#loc273) + %b_q_383 = arith.extf %b_q_380 : tensor<64x64xbf16, #linear> to tensor<64x64xf32, #linear> loc(#loc274) + %b_q_384 = arith.mulf %b_q_383, %b_q_372 : tensor<64x64xf32, #linear> loc(#loc274) + %b_dh2_385 = ttng.tmem_alloc %b_q_384 : (tensor<64x64xf32, #linear>) -> !ttg.memdesc<64x64xf32, #tmem, #ttng.tensor_memory> loc(#loc239) + ttng.fence_async_shared {bCluster = false} loc(#loc239) + %b_dh2_386 = ttg.local_alloc : () -> !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc239) + ttng.init_barrier %b_dh2_386, 1 : !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc239) + ttng.tc_gen5_mma %b_dh2_385, %b_dh1_375, %b_dh2_116, %false, %true, %b_dh2_386[%true] {is_async} : !ttg.memdesc<64x64xf32, #tmem, #ttng.tensor_memory>, !ttg.memdesc<64x64xf32, #shared5, #smem>, !ttg.memdesc<64x64xf32, #tmem, #ttng.tensor_memory, mutable>, !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc239) + ttng.wait_barrier %b_dh2_386, %c0_i32, %true : !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc239) + ttng.inval_barrier %b_dh2_386 : !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc239) + ttng.fence_async_shared {bCluster = false} loc(#loc240) + %b_dh2_387 = ttg.local_alloc : () -> !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc240) + ttng.init_barrier %b_dh2_387, 1 : !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc240) + ttng.tc_gen5_mma %b_w_381, %14, %b_dh2_117, %false, %true, %b_dh2_387[%true] {is_async} : !ttg.memdesc<64x64xbf16, #shared3, #smem, mutable>, !ttg.memdesc<64x64xbf16, #shared2, #smem>, !ttg.memdesc<64x64xf32, #tmem, #ttng.tensor_memory, mutable>, !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc240) + ttng.wait_barrier %b_dh2_387, %c0_i32, %true : !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc240) + ttng.inval_barrier %b_dh2_387 : !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc240) + %b_dh2_388 = arith.addi %b_dh2_288, %c1_i32 : i32 loc(#loc283) + %b_dh2_389 = arith.cmpi sge, %b_dh2_388, %c2_i32 : i32 loc(#loc283) + %b_dh2_390 = arith.select %b_dh2_389, %c0_i32, %b_dh2_388 : i32 loc(#loc283) + %b_dh2_391 = arith.addi %b_dh2_287, %c2_i32 : i32 loc(#loc283) + %i_t_int64_392 = arith.subi %6, %b_dh2_391 : i32 loc(#loc224) + %i_t_393 = arith.addi %i_t_int64_392, %c-1_i32 : i32 loc(#loc285) + %i_t_int64_394 = arith.extsi %i_t_393 : i32 to i64 loc(#loc224) + %o_t_395 = arith.muli %i_t_int64_394, %c64_i64 : i64 loc(#loc225) + %o_t_396 = tt.splat %o_t_395 : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc226) + %o_t_397 = tt.splat %o_t_395 : i64 -> tensor<64xi64, #blocked> loc(#loc226) + %o_t_398 = tt.splat %o_t_395 : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #linear}>> loc(#loc226) + %o_t_399 = tt.splat %o_t_395 : i64 -> tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked2}>> loc(#loc226) + %o_t_400 = arith.addi %o_t_396, %o_v_39 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc226) + %o_t_401 = arith.addi %o_t_397, %o_v_37 : tensor<64xi64, #blocked> loc(#loc226) + %o_t_402 = arith.addi %o_t_398, %o_v_40 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #linear}>> loc(#loc226) + %o_t_403 = arith.addi %o_t_399, %o_v_38 : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked2}>> loc(#loc226) + %m_t_404 = arith.cmpi slt, %o_t_400, %m_t : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc177) + %m_t_405 = arith.cmpi slt, %o_t_401, %m_t_49 : tensor<64xi64, #blocked> loc(#loc177) + %m_t_406 = arith.cmpi slt, %o_t_402, %m_t_50 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #linear}>> loc(#loc177) + %m_t_407 = arith.cmpi slt, %o_t_403, %m_t_51 : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked2}>> loc(#loc177) + %last_idx_408 = arith.addi %i_t_int64_394, %c1_i64 : i64 loc(#loc227) + %last_idx_409 = arith.muli %last_idx_408, %c64_i64 : i64 loc(#loc228) + %last_idx_410 = arith.minsi %last_idx_409, %4 : i64 loc(#loc229) + %last_idx_411 = arith.subi %last_idx_410, %c1_i64 : i64 loc(#loc230) + %bg_last_412 = arith.addi %5, %last_idx_411 : i64 loc(#loc231) + %bg_last_413 = arith.muli %bg_last_412, %c32_i64 : i64 loc(#loc232) + %bg_last_414 = arith.addi %bg_last_413, %3 : i64 loc(#loc284) + %bg_last_415 = tt.addptr %g, %bg_last_414 : !tt.ptr, i64 loc(#loc284) + %bg_last_416 = tt.splat %bg_last_415 : !tt.ptr -> tensor<1x!tt.ptr, #blocked> loc(#loc213) + %bg_last_417 = ttg.memdesc_index %bg_last[%b_dh2_390] : !ttg.memdesc<2x1xf32, #shared, #smem, mutable> -> !ttg.memdesc<1xf32, #shared, #smem, mutable> loc(#loc213) + %b_dh2_418 = tt.splat %b_dh2_321 : i1 -> tensor<1xi1, #blocked> loc(#loc283) + %bg_last_419 = ttg.async_copy_global_to_local %bg_last_416, %bg_last_417 mask %b_dh2_418 : tensor<1x!tt.ptr, #blocked> -> <1xf32, #shared, #smem, mutable> loc(#loc213) + %b_dh1_420 = ttng.tmem_load %b_dh1_114 : !ttg.memdesc<64x64xf32, #tmem, #ttng.tensor_memory, mutable> -> tensor<64x64xf32, #linear> loc(#loc237) + %b_dh1_421 = arith.mulf %b_dh1_420, %b_dh1 : tensor<64x64xf32, #linear> loc(#loc208) + %b_dh1_422 = ttng.tmem_load %b_dh1_115 : !ttg.memdesc<64x64xf32, #tmem, #ttng.tensor_memory, mutable> -> tensor<64x64xf32, #linear> loc(#loc238) + %b_dh1_423 = arith.subf %b_dh1_421, %b_dh1_422 : tensor<64x64xf32, #linear> loc(#loc275) + %b_dh1_424 = arith.addf %b_dh1_369, %b_dh1_423 : tensor<64x64xf32, #linear> loc(#loc276) + %b_dh2_425 = ttng.tmem_load %b_dh2_116 : !ttg.memdesc<64x64xf32, #tmem, #ttng.tensor_memory, mutable> -> tensor<64x64xf32, #linear> loc(#loc239) + %b_dh2_426 = arith.mulf %b_dh2_425, %b_dh1 : tensor<64x64xf32, #linear> loc(#loc277) + %b_dh2_427 = ttng.tmem_load %b_dh2_117 : !ttg.memdesc<64x64xf32, #tmem, #ttng.tensor_memory, mutable> -> tensor<64x64xf32, #linear> loc(#loc240) + %b_dh2_428 = arith.subf %b_dh2_426, %b_dh2_427 : tensor<64x64xf32, #linear> loc(#loc278) + %b_dh2_429 = arith.addf %b_dh2_382, %b_dh2_428 : tensor<64x64xf32, #linear> loc(#loc279) + %bg_last_430 = ttg.async_commit_group tokens %bg_last_419 loc(#loc213) + %p_g_431 = arith.muli %o_t_401, %cst : tensor<64xi64, #blocked> loc(#loc241) + %p_g_432 = tt.addptr %p_g_68, %p_g_431 : tensor<64x!tt.ptr, #blocked>, tensor<64xi64, #blocked> loc(#loc191) + %b_g_433 = ttg.memdesc_index %b_g[%b_dh2_390] : !ttg.memdesc<2x64xf32, #shared, #smem, mutable> -> !ttg.memdesc<64xf32, #shared, #smem, mutable> loc(#loc214) + %b_dh2_434 = tt.splat %b_dh2_321 : i1 -> tensor<64xi1, #blocked> loc(#loc283) + %b_dh2_435 = arith.andi %b_dh2_434, %m_t_405 : tensor<64xi1, #blocked> loc(#loc283) + %b_g_436 = ttg.async_copy_global_to_local %p_g_432, %b_g_433 mask %b_dh2_435 other %cst_0 : tensor<64x!tt.ptr, #blocked> -> <64xf32, #shared, #smem, mutable> loc(#loc214) + %b_g_437 = ttg.async_commit_group tokens %b_g_436 loc(#loc214) + %b_g_438 = ttg.memdesc_index %b_g_84[%b_dh2_390] : !ttg.memdesc<2x64xf32, #shared, #smem, mutable> -> !ttg.memdesc<64xf32, #shared, #smem, mutable> loc(#loc214) + %b_g_439 = ttg.async_copy_global_to_local %p_g_432, %b_g_438 mask %b_dh2_435 other %cst_0 : tensor<64x!tt.ptr, #blocked> -> <64xf32, #shared, #smem, mutable> loc(#loc214) + %b_g_440 = ttg.async_commit_group tokens %b_g_439 loc(#loc214) + %p_dv_441 = tt.expand_dims %o_t_400 {axis = 1 : i32} : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<64x1xi64, #blocked1> loc(#loc242) + %p_dv_442 = arith.muli %p_dv_441, %cst_2 : tensor<64x1xi64, #blocked1> loc(#loc243) + %p_dv_443 = tt.addptr %p_dv, %p_dv_442 : tensor<64x1x!tt.ptr, #blocked1>, tensor<64x1xi64, #blocked1> loc(#loc192) + %p_dv_444 = tt.broadcast %p_dv_443 : tensor<64x1x!tt.ptr, #blocked1> -> tensor<64x64x!tt.ptr, #blocked1> loc(#loc244) + %p_dv_445 = tt.addptr %p_dv_444, %p_dh1_55 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi64, #blocked1> loc(#loc244) + %p_do_446 = tt.addptr %p_do, %p_dv_442 : tensor<64x1x!tt.ptr, #blocked1>, tensor<64x1xi64, #blocked1> loc(#loc194) + %p_do_447 = tt.broadcast %p_do_446 : tensor<64x1x!tt.ptr, #blocked1> -> tensor<64x64x!tt.ptr, #blocked1> loc(#loc245) + %p_do_448 = tt.addptr %p_do_447, %p_dh1_55 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi64, #blocked1> loc(#loc245) + %b_do_449 = tt.expand_dims %m_t_404 {axis = 1 : i32} : tensor<64xi1, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<64x1xi1, #blocked1> loc(#loc246) + %b_do_450 = tt.broadcast %b_do_449 : tensor<64x1xi1, #blocked1> -> tensor<64x64xi1, #blocked1> loc(#loc247) + %b_do_451 = arith.andi %b_do_450, %m_dh1_60 : tensor<64x64xi1, #blocked1> loc(#loc247) + %b_do_452 = ttg.memdesc_index %b_do[%b_dh2_390] : !ttg.memdesc<2x64x64xbf16, #shared1, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared1, #smem, mutable> loc(#loc215) + %b_dh2_453 = tt.splat %b_dh2_321 : i1 -> tensor<64x64xi1, #blocked1> loc(#loc283) + %b_dh2_454 = arith.andi %b_dh2_453, %b_do_451 : tensor<64x64xi1, #blocked1> loc(#loc283) + %b_do_455 = ttg.async_copy_global_to_local %p_do_448, %b_do_452 mask %b_dh2_454 other %cst_12 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked1> -> <64x64xbf16, #shared1, #smem, mutable> loc(#loc215) + %b_do_456 = ttg.async_commit_group tokens %b_do_455 loc(#loc215) + %p_k_457 = tt.addptr %p_k, %p_dv_442 : tensor<64x1x!tt.ptr, #blocked1>, tensor<64x1xi64, #blocked1> loc(#loc195) + %p_k_458 = tt.broadcast %p_k_457 : tensor<64x1x!tt.ptr, #blocked1> -> tensor<64x64x!tt.ptr, #blocked1> loc(#loc197) + %p_k_459 = tt.addptr %p_k_458, %p_k_70 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi32, #blocked1> loc(#loc197) + %b_k_460 = arith.andi %b_do_450, %b_k_71 : tensor<64x64xi1, #blocked1> loc(#loc199) + %b_k_461 = ttg.memdesc_index %b_k_85[%b_dh2_390] : !ttg.memdesc<2x64x64xbf16, #shared2, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared2, #smem, mutable> loc(#loc216) + %b_dh2_462 = arith.andi %b_dh2_453, %b_k_460 : tensor<64x64xi1, #blocked1> loc(#loc283) + %b_k_463 = ttg.async_copy_global_to_local %p_k_459, %b_k_461 mask %b_dh2_462 other %cst_12 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked1> -> <64x64xbf16, #shared2, #smem, mutable> loc(#loc216) + %b_k_464 = ttg.async_commit_group tokens %b_k_463 loc(#loc216) + %p_k_465 = tt.addptr %p_k_458, %p_k_73 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi32, #blocked1> loc(#loc201) + %b_k_466 = arith.andi %b_do_450, %b_k_75 : tensor<64x64xi1, #blocked1> loc(#loc203) + %b_k_467 = ttg.memdesc_index %b_k_86[%b_dh2_390] : !ttg.memdesc<2x64x64xbf16, #shared2, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared2, #smem, mutable> loc(#loc217) + %b_dh2_468 = arith.andi %b_dh2_453, %b_k_466 : tensor<64x64xi1, #blocked1> loc(#loc283) + %b_k_469 = ttg.async_copy_global_to_local %p_k_465, %b_k_467 mask %b_dh2_468 other %cst_12 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked1> -> <64x64xbf16, #shared2, #smem, mutable> loc(#loc217) + %b_k_470 = ttg.async_commit_group tokens %b_k_469 loc(#loc217) + %b_dv_471 = ttg.memdesc_index %b_dv[%b_dh2_390] : !ttg.memdesc<2x64x64xbf16, #shared1, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared1, #smem, mutable> loc(#loc218) + %b_dv_472 = ttg.async_copy_global_to_local %p_dv_445, %b_dv_471 mask %b_dh2_454 other %cst_12 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked1> -> <64x64xbf16, #shared1, #smem, mutable> loc(#loc218) + %b_dv_473 = ttg.async_commit_group tokens %b_dv_472 loc(#loc218) + %p_w_474 = tt.expand_dims %o_t_403 {axis = 0 : i32} : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked2}>> -> tensor<1x64xi64, #blocked2> loc(#loc248) + %p_w_475 = arith.muli %p_w_474, %cst_4 : tensor<1x64xi64, #blocked2> loc(#loc249) + %p_w_476 = tt.broadcast %p_w_475 : tensor<1x64xi64, #blocked2> -> tensor<64x64xi64, #blocked2> loc(#loc205) + %p_w_477 = tt.addptr %p_w_77, %p_w_476 : tensor<64x64x!tt.ptr, #blocked2>, tensor<64x64xi64, #blocked2> loc(#loc205) + %p_q_478 = tt.addptr %p_q_79, %p_w_476 : tensor<64x64x!tt.ptr, #blocked2>, tensor<64x64xi64, #blocked2> loc(#loc207) + %b_w_479 = tt.expand_dims %m_t_407 {axis = 0 : i32} : tensor<64xi1, #ttg.slice<{dim = 0, parent = #blocked2}>> -> tensor<1x64xi1, #blocked2> loc(#loc250) + %b_w_480 = tt.broadcast %b_w_479 : tensor<1x64xi1, #blocked2> -> tensor<64x64xi1, #blocked2> loc(#loc251) + %b_w_481 = arith.andi %m_dh1_59, %b_w_480 : tensor<64x64xi1, #blocked2> loc(#loc251) + %b_w_482 = ttg.memdesc_index %b_w[%b_dh2_390] : !ttg.memdesc<2x64x64xbf16, #shared3, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared3, #smem, mutable> loc(#loc219) + %b_dh2_483 = tt.splat %b_dh2_321 : i1 -> tensor<64x64xi1, #blocked2> loc(#loc283) + %b_dh2_484 = arith.andi %b_dh2_483, %b_w_481 : tensor<64x64xi1, #blocked2> loc(#loc283) + %b_w_485 = ttg.async_copy_global_to_local %p_w_477, %b_w_482 mask %b_dh2_484 other %cst_13 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked2> -> <64x64xbf16, #shared3, #smem, mutable> loc(#loc219) + %b_w_486 = ttg.async_commit_group tokens %b_w_485 loc(#loc219) + %b_q_487 = ttg.memdesc_index %b_q[%b_dh2_390] : !ttg.memdesc<2x64x64xbf16, #shared4, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared4, #smem, mutable> loc(#loc220) + %b_q_488 = ttg.async_copy_global_to_local %p_q_478, %b_q_487 mask %b_dh2_484 other %cst_13 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked2> -> <64x64xbf16, #shared4, #smem, mutable> loc(#loc220) + %b_q_489 = ttg.async_commit_group tokens %b_q_488 loc(#loc220) + %p_q_490 = tt.addptr %p_q_81, %p_w_476 : tensor<64x64x!tt.ptr, #blocked2>, tensor<64x64xi64, #blocked2> loc(#loc210) + %p_w_491 = tt.addptr %p_w_83, %p_w_476 : tensor<64x64x!tt.ptr, #blocked2>, tensor<64x64xi64, #blocked2> loc(#loc212) + %b_q_492 = arith.andi %m_dh2_66, %b_w_480 : tensor<64x64xi1, #blocked2> loc(#loc252) + %b_q_493 = ttg.memdesc_index %b_q_87[%b_dh2_390] : !ttg.memdesc<2x64x64xbf16, #shared4, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared4, #smem, mutable> loc(#loc221) + %b_dh2_494 = arith.andi %b_dh2_483, %b_q_492 : tensor<64x64xi1, #blocked2> loc(#loc283) + %b_q_495 = ttg.async_copy_global_to_local %p_q_490, %b_q_493 mask %b_dh2_494 other %cst_13 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked2> -> <64x64xbf16, #shared4, #smem, mutable> loc(#loc221) + %b_q_496 = ttg.async_commit_group tokens %b_q_495 loc(#loc221) + %b_w_497 = ttg.memdesc_index %b_w_88[%b_dh2_390] : !ttg.memdesc<2x64x64xbf16, #shared3, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared3, #smem, mutable> loc(#loc222) + %b_w_498 = ttg.async_copy_global_to_local %p_w_491, %b_w_497 mask %b_dh2_494 other %cst_13 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked2> -> <64x64xbf16, #shared3, #smem, mutable> loc(#loc222) + %b_w_499 = ttg.async_commit_group tokens %b_w_498 loc(#loc222) + scf.yield %b_dh1_424, %b_dh2_429, %b_dh2_390, %b_dh2_324, %i_t_int64_291, %i_t_int64_394, %bg_last_293, %bg_last_430, %b_g_295, %b_g_437, %b_g_297, %b_g_440, %p_dv_299, %p_dv_442, %b_do_301, %b_do_456, %b_k_303, %b_k_464, %b_k_305, %b_k_470, %m_t_307, %m_t_406, %b_dv_309, %b_dv_473, %b_do_311, %b_do_451, %b_w_313, %b_w_486, %b_q_315, %b_q_489, %b_q_317, %b_q_496, %b_w_319, %b_w_499 : tensor<64x64xf32, #linear>, tensor<64x64xf32, #linear>, i32, i32, i64, i64, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, tensor<64x1xi64, #blocked1>, tensor<64x1xi64, #blocked1>, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, tensor<64xi1, #ttg.slice<{dim = 1, parent = #linear}>>, tensor<64xi1, #ttg.slice<{dim = 1, parent = #linear}>>, !ttg.async.token, !ttg.async.token, tensor<64x64xi1, #blocked1>, tensor<64x64xi1, #blocked1>, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token loc(#loc283) + } loc(#loc283) + %b_dh2_286 = ttg.async_wait {num = 0 : i32} loc(#loc283) + ttg.local_dealloc %b_w_88 : !ttg.memdesc<2x64x64xbf16, #shared3, #smem, mutable> loc(#loc283) + ttg.local_dealloc %b_q_87 : !ttg.memdesc<2x64x64xbf16, #shared4, #smem, mutable> loc(#loc283) + ttg.local_dealloc %b_q : !ttg.memdesc<2x64x64xbf16, #shared4, #smem, mutable> loc(#loc283) + ttg.local_dealloc %b_w : !ttg.memdesc<2x64x64xbf16, #shared3, #smem, mutable> loc(#loc283) + ttg.local_dealloc %b_dv : !ttg.memdesc<2x64x64xbf16, #shared1, #smem, mutable> loc(#loc283) + ttg.local_dealloc %b_k_86 : !ttg.memdesc<2x64x64xbf16, #shared2, #smem, mutable> loc(#loc283) + ttg.local_dealloc %b_k_85 : !ttg.memdesc<2x64x64xbf16, #shared2, #smem, mutable> loc(#loc283) + ttg.local_dealloc %b_do : !ttg.memdesc<2x64x64xbf16, #shared1, #smem, mutable> loc(#loc283) + ttg.local_dealloc %b_g_84 : !ttg.memdesc<2x64xf32, #shared, #smem, mutable> loc(#loc283) + ttg.local_dealloc %b_g : !ttg.memdesc<2x64xf32, #shared, #smem, mutable> loc(#loc283) + ttg.local_dealloc %bg_last : !ttg.memdesc<2x1xf32, #shared, #smem, mutable> loc(#loc283) + tt.return loc(#loc143) + } loc(#loc) +} loc(#loc) +#loc1 = loc(unknown) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":399:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":399:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":401:22) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":401:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":402:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":402:34) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":409:25) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":43:17) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":410:24) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":43:30) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":411:20) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":431:16) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":431:20) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":431:53) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":431:9) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":432:9) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":433:9) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":434:10) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":435:10) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":436:11) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":437:17) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":437:22) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":437:44) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":446:16) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":446:34) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":446:21) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":447:16) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":449:18) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":450:16) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":451:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":489:26) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":492:20) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":497:49) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":497:60) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":497:68) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":497:64) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":498:25) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":498:40) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":498:36) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":505:53) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":505:64) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":506:29) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":506:40) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":528:33) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":528:22) +#loc47 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":528:39) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":532:20) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":533:22) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":534:20) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":539:18) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":539:46) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":539:41) +#loc54 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":540:52) +#loc55 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":540:47) +#loc56 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":550:50) +#loc57 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":550:45) +#loc58 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":551:56) +#loc59 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":551:51) +#loc60 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":585:18) +#loc61 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":585:34) +#loc62 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":586:18) +#loc63 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":586:34) +#loc64 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":600:69) +#loc65 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":602:22) +#loc66 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":602:38) +#loc67 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":603:22) +#loc68 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":603:38) +#loc69 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":527:30) +#loc70 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":529:26) +#loc71 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":536:23) +#loc72 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":540:22) +#loc73 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":551:26) +#loc74 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":581:24) +#loc75 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":587:22) +#loc76 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":588:22) +#loc77 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":604:26) +#loc78 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":605:26) +#loc79 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":489:33) +#loc80 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":490:27) +#loc81 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":491:26) +#loc82 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":491:31) +#loc83 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":525:36) +#loc84 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":525:41) +#loc85 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":525:45) +#loc86 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":525:50) +#loc87 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":527:41) +#loc88 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":527:53) +#loc89 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":527:58) +#loc90 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":527:34) +#loc91 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":547:31) +#loc92 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":557:56) +#loc93 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":600:47) +#loc94 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":600:89) +#loc95 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":617:51) +#loc96 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":617:93) +#loc97 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":528:45) +#loc98 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":532:24) +#loc99 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":532:36) +#loc100 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":532:44) +#loc101 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":534:44) +#loc102 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":536:38) +#loc103 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":536:49) +#loc104 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":585:38) +#loc105 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":585:50) +#loc106 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":587:52) +#loc107 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":587:48) +#loc108 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":604:52) +#loc109 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":497:40) +#loc110 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":497:38) +#loc111 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":497:25) +#loc112 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":437:10) +#loc113 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":497:44) +#loc114 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":499:33) +#loc115 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":499:24) +#loc116 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":505:48) +#loc117 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":505:68) +#loc118 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":507:37) +#loc119 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":507:28) +#loc120 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:33) +#loc121 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":530:31) +#loc122 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":531:27) +#loc123 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":533:46) +#loc124 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":580:49) +#loc125 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":580:39) +#loc126 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":580:55) +#loc127 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":580:58) +#loc128 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":580:20) +#loc129 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":581:16) +#loc130 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":583:32) +#loc131 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":583:24) +#loc132 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":590:21) +#loc133 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":591:32) +#loc134 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":591:24) +#loc135 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":600:55) +#loc136 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":607:25) +#loc137 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":608:28) +#loc138 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":600:77) +#loc139 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":600:21) +#loc140 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":617:73) +#loc141 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":617:81) +#loc142 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":617:25) +#loc143 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":653:4) +#loc154 = loc("pid"(#loc2)) +#loc155 = loc("pid"(#loc3)) +#loc156 = loc("NT"(#loc10)) +#loc157 = loc("boh"(#loc12)) +#loc158 = loc("q"(#loc13)) +#loc159 = loc("q"(#loc14)) +#loc160 = loc("q"(#loc15)) +#loc161 = loc("q"(#loc16)) +#loc162 = loc("k"(#loc17)) +#loc163 = loc("w"(#loc18)) +#loc164 = loc("do"(#loc19)) +#loc165 = loc("dv"(#loc20)) +#loc166 = loc("dv2"(#loc21)) +#loc167 = loc("dh"(#loc22)) +#loc168 = loc("dh"(#loc23)) +#loc169 = loc("dh"(#loc24)) +#loc170 = loc("o_v"(#loc25)) +#loc171 = loc("o_v"(#loc26)) +#loc172 = loc("o_v"(#loc27)) +#loc173 = loc("m_v"(#loc28)) +#loc174 = loc("m_k1"(#loc29)) +#loc175 = loc("o_k2"(#loc30)) +#loc176 = loc("m_k2"(#loc31)) +#loc177 = loc("m_t"(#loc33)) +#loc178 = loc("p_dh1"(#loc34)) +#loc179 = loc("p_dh1"(#loc35)) +#loc180 = loc("p_dh1"(#loc36)) +#loc181 = loc("p_dh1"(#loc37)) +#loc182 = loc("m_dh1"(#loc38)) +#loc183 = loc("m_dh1"(#loc39)) +#loc184 = loc("m_dh1"(#loc40)) +#loc185 = loc("p_dh2"(#loc41)) +#loc186 = loc("p_dh2"(#loc42)) +#loc187 = loc("m_dh2"(#loc43)) +#loc188 = loc("m_dh2"(#loc44)) +#loc189 = loc("p_g"(#loc45)) +#loc190 = loc("p_g"(#loc46)) +#loc191 = loc("p_g"(#loc47)) +#loc192 = loc("p_dv"(#loc48)) +#loc193 = loc("p_dv2"(#loc49)) +#loc194 = loc("p_do"(#loc50)) +#loc195 = loc("p_k"(#loc51)) +#loc196 = loc("p_k"(#loc52)) +#loc197 = loc("p_k"(#loc53)) +#loc198 = loc("b_k"(#loc54)) +#loc199 = loc("b_k"(#loc55)) +#loc200 = loc("p_k"(#loc56)) +#loc201 = loc("p_k"(#loc57)) +#loc202 = loc("b_k"(#loc58)) +#loc203 = loc("b_k"(#loc59)) +#loc204 = loc("p_w"(#loc60)) +#loc205 = loc("p_w"(#loc61)) +#loc206 = loc("p_q"(#loc62)) +#loc207 = loc("p_q"(#loc63)) +#loc208 = loc("b_dh1"(#loc64)) +#loc209 = loc("p_q"(#loc65)) +#loc210 = loc("p_q"(#loc66)) +#loc211 = loc("p_w"(#loc67)) +#loc212 = loc("p_w"(#loc68)) +#loc213 = loc("bg_last"(#loc69)) +#loc214 = loc("b_g"(#loc70)) +#loc215 = loc("b_do"(#loc71)) +#loc216 = loc("b_k"(#loc72)) +#loc217 = loc("b_k"(#loc73)) +#loc218 = loc("b_dv"(#loc74)) +#loc219 = loc("b_w"(#loc75)) +#loc220 = loc("b_q"(#loc76)) +#loc221 = loc("b_q"(#loc77)) +#loc222 = loc("b_w"(#loc78)) +#loc223 = loc("b_dh1"(#loc79)) +#loc224 = loc("i_t_int64"(#loc80)) +#loc225 = loc("o_t"(#loc81)) +#loc226 = loc("o_t"(#loc82)) +#loc227 = loc("last_idx"(#loc83)) +#loc228 = loc("last_idx"(#loc84)) +#loc229 = loc("last_idx"(#loc85)) +#loc230 = loc("last_idx"(#loc86)) +#loc231 = loc("bg_last"(#loc87)) +#loc232 = loc("bg_last"(#loc88)) +#loc233 = loc("bg_last"(#loc89)) +#loc234 = loc("bg_last"(#loc90)) +#loc235 = loc("b_dv"(#loc91)) +#loc236 = loc("b_dv"(#loc92)) +#loc237 = loc("b_dh1"(#loc93)) +#loc238 = loc("b_dh1"(#loc94)) +#loc239 = loc("b_dh2"(#loc95)) +#loc240 = loc("b_dh2"(#loc96)) +#loc241 = loc("p_g"(#loc97)) +#loc242 = loc("p_dv"(#loc98)) +#loc243 = loc("p_dv"(#loc99)) +#loc244 = loc("p_dv"(#loc100)) +#loc245 = loc("p_do"(#loc101)) +#loc246 = loc("b_do"(#loc102)) +#loc247 = loc("b_do"(#loc103)) +#loc248 = loc("p_w"(#loc104)) +#loc249 = loc("p_w"(#loc105)) +#loc250 = loc("b_w"(#loc106)) +#loc251 = loc("b_w"(#loc107)) +#loc252 = loc("b_q"(#loc108)) +#loc253 = loc("p_dh1"(#loc109)) +#loc254 = loc("p_dh1"(#loc110)) +#loc255 = loc("p_dh1"(#loc111)) +#loc256 = loc("dh"(#loc112)) +#loc257 = loc("p_dh1"(#loc113)) +#loc258 = loc("p_dh2"(#loc116)) +#loc259 = loc("p_dh2"(#loc117)) +#loc260 = loc("bg_last_exp"(#loc121)) +#loc261 = loc("b_g_exp"(#loc122)) +#loc262 = loc("p_dv2"(#loc123)) +#loc263 = loc("b_dv"(#loc124)) +#loc264 = loc("b_dv"(#loc125)) +#loc265 = loc("b_dv"(#loc126)) +#loc266 = loc("b_dv"(#loc127)) +#loc267 = loc("b_dv"(#loc128)) +#loc268 = loc("b_dv"(#loc129)) +#loc269 = loc("b_dh1"(#loc132)) +#loc270 = loc("b_q"(#loc133)) +#loc271 = loc("b_q"(#loc134)) +#loc272 = loc("b_dh1"(#loc135)) +#loc273 = loc("b_dh2"(#loc136)) +#loc274 = loc("b_q"(#loc137)) +#loc275 = loc("b_dh1"(#loc138)) +#loc276 = loc("b_dh1"(#loc139)) +#loc277 = loc("b_dh2"(#loc140)) +#loc278 = loc("b_dh2"(#loc141)) +#loc279 = loc("b_dh2"(#loc142)) +#loc280 = loc(callsite(#loc9 at #loc156)) +#loc281 = loc(callsite(#loc11 at #loc156)) +#loc282 = loc(fused[#loc189, #loc190]) +#loc283 = loc("b_dh2"(#loc223)) +#loc284 = loc(fused[#loc233, #loc234]) +#loc285 = loc("i_t"(#loc224)) +#loc286 = loc(fused[#loc253, #loc254]) +#loc287 = loc(fused[#loc255, #loc256]) +#loc288 = loc(callsite(#loc120 at #loc260)) +#loc289 = loc(callsite(#loc120 at #loc261)) +#loc290 = loc(callsite(#loc120 at #loc264)) diff --git a/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.ttir b/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.ttir new file mode 100644 index 0000000000000000000000000000000000000000..a021e73e1058f48848c7110b1f40681f0efd40ea --- /dev/null +++ b/triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.ttir @@ -0,0 +1,503 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":370:0) +#loc145 = loc("q"(#loc)) +#loc146 = loc("k"(#loc)) +#loc147 = loc("w"(#loc)) +#loc148 = loc("g"(#loc)) +#loc149 = loc("do"(#loc)) +#loc150 = loc("dh"(#loc)) +#loc151 = loc("dv"(#loc)) +#loc152 = loc("dv2"(#loc)) +#loc153 = loc("scale"(#loc)) +#loc154 = loc("T"(#loc)) +module { + tt.func public @chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64(%q: !tt.ptr {tt.divisibility = 16 : i32} loc("q"(#loc)), %k: !tt.ptr {tt.divisibility = 16 : i32} loc("k"(#loc)), %w: !tt.ptr {tt.divisibility = 16 : i32} loc("w"(#loc)), %g: !tt.ptr {tt.divisibility = 16 : i32} loc("g"(#loc)), %do: !tt.ptr {tt.divisibility = 16 : i32} loc("do"(#loc)), %dh: !tt.ptr {tt.divisibility = 16 : i32} loc("dh"(#loc)), %dv: !tt.ptr {tt.divisibility = 16 : i32} loc("dv"(#loc)), %dv2: !tt.ptr {tt.divisibility = 16 : i32} loc("dv2"(#loc)), %scale: f32 loc("scale"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %NT = arith.constant 63 : i32 loc(#loc281) + %NT_0 = arith.constant 64 : i32 loc(#loc281) + %cst = arith.constant dense<0.000000e+00> : tensor<64x64xbf16> loc(#loc1) + %c524288_i64 = arith.constant 524288 : i64 loc(#loc1) + %c-1_i32 = arith.constant -1 : i32 loc(#loc3) + %m_v = arith.constant dense<128> : tensor<64xi64> loc(#loc156) + %dh_1 = arith.constant 16384 : i64 loc(#loc157) + %c2_i64 = arith.constant 2 : i64 loc(#loc1) + %cst_2 = arith.constant dense<4096> : tensor<1x64xi64> loc(#loc1) + %cst_3 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc1) + %cst_4 = arith.constant dense<4096> : tensor<64x1xi64> loc(#loc1) + %cst_5 = arith.constant dense<0.000000e+00> : tensor<64xf32> loc(#loc1) + %cst_6 = arith.constant dense<32> : tensor<64xi64> loc(#loc1) + %cst_7 = arith.constant dense<128> : tensor<64x1xi32> loc(#loc1) + %o_k2 = arith.constant dense<64> : tensor<64xi32> loc(#loc158) + %cst_8 = arith.constant dense<128> : tensor<64xi32> loc(#loc1) + %c64_i64 = arith.constant 64 : i64 loc(#loc1) + %c128_i64 = arith.constant 128 : i64 loc(#loc1) + %c1_i64 = arith.constant 1 : i64 loc(#loc1) + %c1_i32 = arith.constant 1 : i32 loc(#loc1) + %c32_i64 = arith.constant 32 : i64 loc(#loc1) + %pid = tt.get_program_id x : i32 loc(#loc159) + %pid_9 = arith.extsi %pid : i32 to i64 loc(#loc160) + %0 = arith.remsi %pid_9, %c2_i64 : i64 loc(#loc9) + %1 = arith.divsi %pid_9, %c2_i64 : i64 loc(#loc10) + %2 = arith.divsi %1, %c32_i64 : i64 loc(#loc11) + %3 = arith.remsi %1, %c32_i64 : i64 loc(#loc12) + %4 = arith.extsi %T : i32 to i64 loc(#loc13) + %5 = arith.muli %2, %4 : i64 loc(#loc13) + %NT_10 = arith.addi %T, %NT : i32 loc(#loc282) + %NT_11 = arith.divsi %NT_10, %NT_0 : i32 loc(#loc283) + %boh = arith.extsi %NT_11 : i32 to i64 loc(#loc161) + %boh_12 = arith.muli %2, %boh : i64 loc(#loc161) + %q_13 = arith.muli %5, %c32_i64 : i64 loc(#loc162) + %q_14 = arith.addi %q_13, %3 : i64 loc(#loc163) + %q_15 = arith.muli %q_14, %c128_i64 : i64 loc(#loc164) + %q_16 = tt.addptr %q, %q_15 : !tt.ptr, i64 loc(#loc165) + %k_17 = tt.addptr %k, %q_15 : !tt.ptr, i64 loc(#loc166) + %w_18 = tt.addptr %w, %q_15 : !tt.ptr, i64 loc(#loc167) + %do_19 = tt.addptr %do, %q_15 : !tt.ptr, i64 loc(#loc168) + %dv_20 = tt.addptr %dv, %q_15 : !tt.ptr, i64 loc(#loc169) + %dv2_21 = tt.addptr %dv2, %q_15 : !tt.ptr, i64 loc(#loc170) + %dh_22 = arith.muli %boh_12, %c32_i64 : i64 loc(#loc171) + %dh_23 = arith.addi %dh_22, %3 : i64 loc(#loc172) + %dh_24 = arith.muli %dh_23, %dh_1 : i64 loc(#loc157) + %o_v = arith.muli %0, %c64_i64 : i64 loc(#loc173) + %o_v_25 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc174) + %o_v_26 = arith.extsi %o_v_25 : tensor<64xi32> to tensor<64xi64> loc(#loc175) + %o_v_27 = tt.splat %o_v : i64 -> tensor<64xi64> loc(#loc175) + %o_v_28 = arith.addi %o_v_27, %o_v_26 : tensor<64xi64> loc(#loc175) + %m_v_29 = arith.cmpi slt, %o_v_28, %m_v : tensor<64xi64> loc(#loc156) + %m_k1 = arith.cmpi slt, %o_v_25, %cst_8 : tensor<64xi32> loc(#loc176) + %o_k2_30 = arith.addi %o_v_25, %o_k2 : tensor<64xi32> loc(#loc158) + %m_k2 = arith.cmpi slt, %o_k2_30, %cst_8 : tensor<64xi32> loc(#loc177) + %6 = arith.subi %NT_11, %c1_i32 : i32 loc(#loc33) + %b_dh2:2 = scf.for %arg10 = %c-1_i32 to %6 step %c1_i32 iter_args(%b_dh1 = %cst_3, %b_dh2_31 = %cst_3) -> (tensor<64x64xf32>, tensor<64x64xf32>) : i32 { + %i_t_int64 = arith.subi %6, %arg10 : i32 loc(#loc179) + %i_t = arith.addi %i_t_int64, %c-1_i32 : i32 loc(#loc285) + %i_t_int64_32 = arith.extsi %i_t : i32 to i64 loc(#loc179) + %o_t = arith.muli %i_t_int64_32, %c64_i64 : i64 loc(#loc180) + %o_t_33 = tt.splat %o_t : i64 -> tensor<64xi64> loc(#loc181) + %o_t_34 = arith.addi %o_t_33, %o_v_26 : tensor<64xi64> loc(#loc181) + %m_t = tt.splat %4 : i64 -> tensor<64xi64> loc(#loc182) + %m_t_35 = arith.cmpi slt, %o_t_34, %m_t : tensor<64xi64> loc(#loc182) + %p_dh1 = arith.muli %i_t_int64_32, %c524288_i64 : i64 loc(#loc286) + %p_dh1_36 = arith.addi %dh_24, %p_dh1 : i64 loc(#loc287) + %p_dh1_37 = tt.addptr %dh, %p_dh1_36 : !tt.ptr, i64 loc(#loc287) + %p_dh1_38 = tt.expand_dims %o_v_25 {axis = 1 : i32} : tensor<64xi32> -> tensor<64x1xi32> loc(#loc187) + %p_dh1_39 = arith.muli %p_dh1_38, %cst_7 : tensor<64x1xi32> loc(#loc188) + %p_dh1_40 = tt.splat %p_dh1_37 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc189) + %p_dh1_41 = tt.addptr %p_dh1_40, %p_dh1_39 : tensor<64x1x!tt.ptr>, tensor<64x1xi32> loc(#loc189) + %p_dh1_42 = tt.expand_dims %o_v_28 {axis = 0 : i32} : tensor<64xi64> -> tensor<1x64xi64> loc(#loc190) + %p_dh1_43 = tt.broadcast %p_dh1_41 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc191) + %p_dh1_44 = tt.broadcast %p_dh1_42 : tensor<1x64xi64> -> tensor<64x64xi64> loc(#loc191) + %p_dh1_45 = tt.addptr %p_dh1_43, %p_dh1_44 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc191) + %m_dh1 = tt.expand_dims %m_k1 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc192) + %m_dh1_46 = tt.expand_dims %m_v_29 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc193) + %m_dh1_47 = tt.broadcast %m_dh1 : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc194) + %m_dh1_48 = tt.broadcast %m_dh1_46 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc194) + %m_dh1_49 = arith.andi %m_dh1_47, %m_dh1_48 : tensor<64x64xi1> loc(#loc194) + %7 = arith.truncf %b_dh1 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc50) + tt.store %p_dh1_45, %7, %m_dh1_49 : tensor<64x64x!tt.ptr> loc(#loc51) + %p_dh2 = tt.expand_dims %o_k2_30 {axis = 1 : i32} : tensor<64xi32> -> tensor<64x1xi32> loc(#loc195) + %p_dh2_50 = arith.muli %p_dh2, %cst_7 : tensor<64x1xi32> loc(#loc196) + %p_dh2_51 = tt.addptr %p_dh1_40, %p_dh2_50 : tensor<64x1x!tt.ptr>, tensor<64x1xi32> loc(#loc197) + %p_dh2_52 = tt.broadcast %p_dh2_51 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc198) + %p_dh2_53 = tt.addptr %p_dh2_52, %p_dh1_44 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc198) + %m_dh2 = tt.expand_dims %m_k2 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc199) + %m_dh2_54 = tt.broadcast %m_dh2 : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc200) + %m_dh2_55 = arith.andi %m_dh2_54, %m_dh1_48 : tensor<64x64xi1> loc(#loc200) + %8 = arith.truncf %b_dh2_31 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc58) + tt.store %p_dh2_53, %8, %m_dh2_55 : tensor<64x64x!tt.ptr> loc(#loc59) + %last_idx = arith.addi %i_t_int64_32, %c1_i64 : i64 loc(#loc201) + %last_idx_56 = arith.muli %last_idx, %c64_i64 : i64 loc(#loc202) + %last_idx_57 = arith.minsi %last_idx_56, %4 : i64 loc(#loc203) + %last_idx_58 = arith.subi %last_idx_57, %c1_i64 : i64 loc(#loc204) + %bg_last = arith.addi %5, %last_idx_58 : i64 loc(#loc205) + %bg_last_59 = arith.muli %bg_last, %c32_i64 : i64 loc(#loc206) + %bg_last_60 = arith.addi %bg_last_59, %3 : i64 loc(#loc288) + %bg_last_61 = tt.addptr %g, %bg_last_60 : !tt.ptr, i64 loc(#loc288) + %bg_last_62 = tt.load %bg_last_61 : !tt.ptr loc(#loc209) + %p_g = tt.addptr %g, %q_14 : !tt.ptr, i64 loc(#loc289) + %p_g_63 = arith.muli %o_t_34, %cst_6 : tensor<64xi64> loc(#loc212) + %p_g_64 = tt.splat %p_g : !tt.ptr -> tensor<64x!tt.ptr> loc(#loc213) + %p_g_65 = tt.addptr %p_g_64, %p_g_63 : tensor<64x!tt.ptr>, tensor<64xi64> loc(#loc213) + %b_g = tt.load %p_g_65, %m_t_35, %cst_5 : tensor<64x!tt.ptr> loc(#loc214) + %bg_last_exp = math.exp2 %bg_last_62 : f32 loc(#loc290) + %b_g_exp = math.exp2 %b_g : tensor<64xf32> loc(#loc291) + %p_dv = tt.expand_dims %o_t_34 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc217) + %p_dv_66 = arith.muli %p_dv, %cst_4 : tensor<64x1xi64> loc(#loc218) + %p_dv_67 = tt.splat %dv_20 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc219) + %p_dv_68 = tt.addptr %p_dv_67, %p_dv_66 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc219) + %p_dv_69 = tt.broadcast %p_dv_68 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc220) + %p_dv_70 = tt.addptr %p_dv_69, %p_dh1_44 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc220) + %p_dv2 = tt.splat %dv2_21 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc221) + %p_dv2_71 = tt.addptr %p_dv2, %p_dv_66 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc221) + %p_dv2_72 = tt.broadcast %p_dv2_71 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc222) + %p_dv2_73 = tt.addptr %p_dv2_72, %p_dh1_44 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc222) + %p_do = tt.splat %do_19 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc223) + %p_do_74 = tt.addptr %p_do, %p_dv_66 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc223) + %p_do_75 = tt.broadcast %p_do_74 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc224) + %p_do_76 = tt.addptr %p_do_75, %p_dh1_44 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc224) + %b_do = tt.expand_dims %m_t_35 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc225) + %b_do_77 = tt.broadcast %b_do : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc226) + %b_do_78 = arith.andi %b_do_77, %m_dh1_48 : tensor<64x64xi1> loc(#loc226) + %b_do_79 = tt.load %p_do_76, %b_do_78, %cst : tensor<64x64x!tt.ptr> loc(#loc227) + %p_k = tt.splat %k_17 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc228) + %p_k_80 = tt.addptr %p_k, %p_dv_66 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc228) + %p_k_81 = tt.expand_dims %o_v_25 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc229) + %p_k_82 = tt.broadcast %p_k_80 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc230) + %p_k_83 = tt.broadcast %p_k_81 : tensor<1x64xi32> -> tensor<64x64xi32> loc(#loc230) + %p_k_84 = tt.addptr %p_k_82, %p_k_83 : tensor<64x64x!tt.ptr>, tensor<64x64xi32> loc(#loc230) + %b_k = tt.expand_dims %m_k1 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc231) + %b_k_85 = tt.broadcast %b_k : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc232) + %b_k_86 = arith.andi %b_do_77, %b_k_85 : tensor<64x64xi1> loc(#loc232) + %b_k_87 = tt.load %p_k_84, %b_k_86, %cst : tensor<64x64x!tt.ptr> loc(#loc233) + %b_dv = tt.dot %b_k_87, %7, %cst_3, inputPrecision = tf32 : tensor<64x64xbf16> * tensor<64x64xbf16> -> tensor<64x64xf32> loc(#loc234) + %p_k_88 = tt.expand_dims %o_k2_30 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc235) + %p_k_89 = tt.broadcast %p_k_88 : tensor<1x64xi32> -> tensor<64x64xi32> loc(#loc236) + %p_k_90 = tt.addptr %p_k_82, %p_k_89 : tensor<64x64x!tt.ptr>, tensor<64x64xi32> loc(#loc236) + %b_k_91 = tt.expand_dims %m_k2 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc237) + %b_k_92 = tt.broadcast %b_k_91 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc238) + %b_k_93 = arith.andi %b_do_77, %b_k_92 : tensor<64x64xi1> loc(#loc238) + %b_k_94 = tt.load %p_k_90, %b_k_93, %cst : tensor<64x64x!tt.ptr> loc(#loc239) + %b_dv_95 = tt.dot %b_k_94, %8, %b_dv, inputPrecision = tf32 : tensor<64x64xbf16> * tensor<64x64xbf16> -> tensor<64x64xf32> loc(#loc240) + %b_dv_96 = tt.splat %bg_last_62 : f32 -> tensor<64xf32> loc(#loc241) + %b_dv_97 = arith.subf %b_dv_96, %b_g : tensor<64xf32> loc(#loc241) + %b_dv_98 = math.exp2 %b_dv_97 : tensor<64xf32> loc(#loc292) + %b_dv_99 = arith.select %m_t_35, %b_dv_98, %cst_5 : tensor<64xi1>, tensor<64xf32> loc(#loc243) + %b_dv_100 = tt.expand_dims %b_dv_99 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc244) + %b_dv_101 = tt.broadcast %b_dv_100 : tensor<64x1xf32> -> tensor<64x64xf32> loc(#loc245) + %b_dv_102 = arith.mulf %b_dv_95, %b_dv_101 : tensor<64x64xf32> loc(#loc245) + %b_dv_103 = tt.load %p_dv_70, %b_do_78, %cst : tensor<64x64x!tt.ptr> loc(#loc246) + %b_dv_104 = arith.extf %b_dv_103 : tensor<64x64xbf16> to tensor<64x64xf32> loc(#loc247) + %b_dv_105 = arith.addf %b_dv_102, %b_dv_104 : tensor<64x64xf32> loc(#loc247) + %9 = arith.truncf %b_dv_105 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc108) + tt.store %p_dv2_73, %9, %b_do_78 : tensor<64x64x!tt.ptr> loc(#loc109) + %p_w = tt.splat %w_18 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc248) + %p_w_106 = tt.addptr %p_w, %p_dh1_38 : tensor<64x1x!tt.ptr>, tensor<64x1xi32> loc(#loc248) + %p_w_107 = tt.expand_dims %o_t_34 {axis = 0 : i32} : tensor<64xi64> -> tensor<1x64xi64> loc(#loc249) + %p_w_108 = arith.muli %p_w_107, %cst_2 : tensor<1x64xi64> loc(#loc250) + %p_w_109 = tt.broadcast %p_w_106 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc251) + %p_w_110 = tt.broadcast %p_w_108 : tensor<1x64xi64> -> tensor<64x64xi64> loc(#loc251) + %p_w_111 = tt.addptr %p_w_109, %p_w_110 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc251) + %p_q = tt.splat %q_16 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc252) + %p_q_112 = tt.addptr %p_q, %p_dh1_38 : tensor<64x1x!tt.ptr>, tensor<64x1xi32> loc(#loc252) + %p_q_113 = tt.broadcast %p_q_112 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc253) + %p_q_114 = tt.addptr %p_q_113, %p_w_110 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc253) + %b_w = tt.expand_dims %m_t_35 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc254) + %b_w_115 = tt.broadcast %b_w : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc255) + %b_w_116 = arith.andi %m_dh1_47, %b_w_115 : tensor<64x64xi1> loc(#loc255) + %b_w_117 = tt.load %p_w_111, %b_w_116, %cst : tensor<64x64x!tt.ptr> loc(#loc256) + %b_q = tt.load %p_q_114, %b_w_116, %cst : tensor<64x64x!tt.ptr> loc(#loc257) + %b_dh1_118 = tt.splat %bg_last_exp : f32 -> tensor<64x64xf32> loc(#loc258) + %b_dh1_119 = arith.mulf %b_dh1, %b_dh1_118 : tensor<64x64xf32> loc(#loc258) + %b_q_120 = tt.expand_dims %b_g_exp {axis = 0 : i32} : tensor<64xf32> -> tensor<1x64xf32> loc(#loc259) + %b_q_121 = arith.extf %b_q : tensor<64x64xbf16> to tensor<64x64xf32> loc(#loc260) + %b_q_122 = tt.broadcast %b_q_120 : tensor<1x64xf32> -> tensor<64x64xf32> loc(#loc260) + %b_q_123 = arith.mulf %b_q_121, %b_q_122 : tensor<64x64xf32> loc(#loc260) + %b_dh1_124 = arith.extf %b_do_79 : tensor<64x64xbf16> to tensor<64x64xf32> loc(#loc261) + %b_dh1_125 = tt.dot %b_q_123, %b_dh1_124, %cst_3, inputPrecision = tf32 : tensor<64x64xf32> * tensor<64x64xf32> -> tensor<64x64xf32> loc(#loc262) + %b_dh1_126 = tt.splat %scale : f32 -> tensor<64x64xf32> loc(#loc263) + %b_dh1_127 = arith.mulf %b_dh1_125, %b_dh1_126 : tensor<64x64xf32> loc(#loc263) + %b_dh1_128 = tt.dot %b_w_117, %9, %cst_3, inputPrecision = tf32 : tensor<64x64xbf16> * tensor<64x64xbf16> -> tensor<64x64xf32> loc(#loc264) + %b_dh1_129 = arith.subf %b_dh1_127, %b_dh1_128 : tensor<64x64xf32> loc(#loc265) + %b_dh1_130 = arith.addf %b_dh1_119, %b_dh1_129 : tensor<64x64xf32> loc(#loc266) + %p_q_131 = tt.addptr %p_q, %p_dh2 : tensor<64x1x!tt.ptr>, tensor<64x1xi32> loc(#loc267) + %p_q_132 = tt.broadcast %p_q_131 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc268) + %p_q_133 = tt.addptr %p_q_132, %p_w_110 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc268) + %p_w_134 = tt.addptr %p_w, %p_dh2 : tensor<64x1x!tt.ptr>, tensor<64x1xi32> loc(#loc269) + %p_w_135 = tt.broadcast %p_w_134 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc270) + %p_w_136 = tt.addptr %p_w_135, %p_w_110 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc270) + %b_q_137 = arith.andi %m_dh2_54, %b_w_115 : tensor<64x64xi1> loc(#loc271) + %b_q_138 = tt.load %p_q_133, %b_q_137, %cst : tensor<64x64x!tt.ptr> loc(#loc272) + %b_w_139 = tt.load %p_w_136, %b_q_137, %cst : tensor<64x64x!tt.ptr> loc(#loc273) + %b_dh2_140 = arith.mulf %b_dh2_31, %b_dh1_118 : tensor<64x64xf32> loc(#loc274) + %b_q_141 = arith.extf %b_q_138 : tensor<64x64xbf16> to tensor<64x64xf32> loc(#loc275) + %b_q_142 = arith.mulf %b_q_141, %b_q_122 : tensor<64x64xf32> loc(#loc275) + %b_dh2_143 = tt.dot %b_q_142, %b_dh1_124, %cst_3, inputPrecision = tf32 : tensor<64x64xf32> * tensor<64x64xf32> -> tensor<64x64xf32> loc(#loc276) + %b_dh2_144 = arith.mulf %b_dh2_143, %b_dh1_126 : tensor<64x64xf32> loc(#loc277) + %b_dh2_145 = tt.dot %b_w_139, %9, %cst_3, inputPrecision = tf32 : tensor<64x64xbf16> * tensor<64x64xbf16> -> tensor<64x64xf32> loc(#loc278) + %b_dh2_146 = arith.subf %b_dh2_144, %b_dh2_145 : tensor<64x64xf32> loc(#loc279) + %b_dh2_147 = arith.addf %b_dh2_140, %b_dh2_146 : tensor<64x64xf32> loc(#loc280) + scf.yield %b_dh1_130, %b_dh2_147 : tensor<64x64xf32>, tensor<64x64xf32> loc(#loc143) + } loc(#loc284) + tt.return loc(#loc144) + } loc(#loc) +} loc(#loc) +#loc1 = loc(unknown) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":410:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":489:33) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":447:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":437:44) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":450:16) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":399:24) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":399:30) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":401:22) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":401:34) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":402:23) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":402:34) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":409:25) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":43:17) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":43:30) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":411:20) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":431:16) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":431:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":431:53) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":431:9) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":432:9) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":433:9) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":434:10) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":435:10) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":436:11) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":437:17) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":437:22) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":446:16) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":446:34) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":446:21) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":449:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":451:18) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":489:26) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":490:27) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":491:26) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":491:31) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":492:20) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":497:40) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":497:38) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":497:25) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":437:10) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":497:49) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":497:60) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":497:44) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":497:68) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":497:64) +#loc47 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":498:25) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":498:40) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":498:36) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":499:33) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":499:24) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":505:53) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":505:64) +#loc54 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":505:48) +#loc55 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":505:68) +#loc56 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":506:29) +#loc57 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":506:40) +#loc58 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":507:37) +#loc59 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":507:28) +#loc60 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":525:36) +#loc61 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":525:41) +#loc62 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":525:45) +#loc63 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":525:50) +#loc64 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":527:41) +#loc65 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":527:53) +#loc66 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":527:58) +#loc67 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":527:34) +#loc68 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":527:30) +#loc69 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":528:33) +#loc70 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":528:22) +#loc71 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":528:45) +#loc72 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":528:39) +#loc73 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":529:26) +#loc74 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:33) +#loc75 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":530:31) +#loc76 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":531:27) +#loc77 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":532:24) +#loc78 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":532:36) +#loc79 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":532:20) +#loc80 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":532:44) +#loc81 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":533:22) +#loc82 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":533:46) +#loc83 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":534:20) +#loc84 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":534:44) +#loc85 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":536:38) +#loc86 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":536:49) +#loc87 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":536:23) +#loc88 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":539:18) +#loc89 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":539:46) +#loc90 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":539:41) +#loc91 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":540:52) +#loc92 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":540:47) +#loc93 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":540:22) +#loc94 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":547:31) +#loc95 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":550:50) +#loc96 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":550:45) +#loc97 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":551:56) +#loc98 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":551:51) +#loc99 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":551:26) +#loc100 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":557:56) +#loc101 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":580:49) +#loc102 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":580:39) +#loc103 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":580:55) +#loc104 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":580:58) +#loc105 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":580:20) +#loc106 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":581:24) +#loc107 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":581:16) +#loc108 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":583:32) +#loc109 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":583:24) +#loc110 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":585:18) +#loc111 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":585:38) +#loc112 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":585:50) +#loc113 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":585:34) +#loc114 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":586:18) +#loc115 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":586:34) +#loc116 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":587:52) +#loc117 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":587:48) +#loc118 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":587:22) +#loc119 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":588:22) +#loc120 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":590:21) +#loc121 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":591:32) +#loc122 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":591:24) +#loc123 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":600:55) +#loc124 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":600:47) +#loc125 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":600:69) +#loc126 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":600:89) +#loc127 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":600:77) +#loc128 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":600:21) +#loc129 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":602:22) +#loc130 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":602:38) +#loc131 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":603:22) +#loc132 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":603:38) +#loc133 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":604:52) +#loc134 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":604:26) +#loc135 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":605:26) +#loc136 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":607:25) +#loc137 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":608:28) +#loc138 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":617:51) +#loc139 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":617:73) +#loc140 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":617:93) +#loc141 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":617:81) +#loc142 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":617:25) +#loc143 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":635:8) +#loc144 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":653:4) +#loc155 = loc("NT"(#loc2)) +#loc156 = loc("m_v"(#loc4)) +#loc157 = loc("dh"(#loc5)) +#loc158 = loc("o_k2"(#loc6)) +#loc159 = loc("pid"(#loc7)) +#loc160 = loc("pid"(#loc8)) +#loc161 = loc("boh"(#loc16)) +#loc162 = loc("q"(#loc17)) +#loc163 = loc("q"(#loc18)) +#loc164 = loc("q"(#loc19)) +#loc165 = loc("q"(#loc20)) +#loc166 = loc("k"(#loc21)) +#loc167 = loc("w"(#loc22)) +#loc168 = loc("do"(#loc23)) +#loc169 = loc("dv"(#loc24)) +#loc170 = loc("dv2"(#loc25)) +#loc171 = loc("dh"(#loc26)) +#loc172 = loc("dh"(#loc27)) +#loc173 = loc("o_v"(#loc28)) +#loc174 = loc("o_v"(#loc29)) +#loc175 = loc("o_v"(#loc30)) +#loc176 = loc("m_k1"(#loc31)) +#loc177 = loc("m_k2"(#loc32)) +#loc178 = loc("b_dh1"(#loc3)) +#loc179 = loc("i_t_int64"(#loc34)) +#loc180 = loc("o_t"(#loc35)) +#loc181 = loc("o_t"(#loc36)) +#loc182 = loc("m_t"(#loc37)) +#loc183 = loc("p_dh1"(#loc38)) +#loc184 = loc("p_dh1"(#loc39)) +#loc185 = loc("p_dh1"(#loc40)) +#loc186 = loc("dh"(#loc41)) +#loc187 = loc("p_dh1"(#loc42)) +#loc188 = loc("p_dh1"(#loc43)) +#loc189 = loc("p_dh1"(#loc44)) +#loc190 = loc("p_dh1"(#loc45)) +#loc191 = loc("p_dh1"(#loc46)) +#loc192 = loc("m_dh1"(#loc47)) +#loc193 = loc("m_dh1"(#loc48)) +#loc194 = loc("m_dh1"(#loc49)) +#loc195 = loc("p_dh2"(#loc52)) +#loc196 = loc("p_dh2"(#loc53)) +#loc197 = loc("p_dh2"(#loc54)) +#loc198 = loc("p_dh2"(#loc55)) +#loc199 = loc("m_dh2"(#loc56)) +#loc200 = loc("m_dh2"(#loc57)) +#loc201 = loc("last_idx"(#loc60)) +#loc202 = loc("last_idx"(#loc61)) +#loc203 = loc("last_idx"(#loc62)) +#loc204 = loc("last_idx"(#loc63)) +#loc205 = loc("bg_last"(#loc64)) +#loc206 = loc("bg_last"(#loc65)) +#loc207 = loc("bg_last"(#loc66)) +#loc208 = loc("bg_last"(#loc67)) +#loc209 = loc("bg_last"(#loc68)) +#loc210 = loc("p_g"(#loc69)) +#loc211 = loc("p_g"(#loc70)) +#loc212 = loc("p_g"(#loc71)) +#loc213 = loc("p_g"(#loc72)) +#loc214 = loc("b_g"(#loc73)) +#loc215 = loc("bg_last_exp"(#loc75)) +#loc216 = loc("b_g_exp"(#loc76)) +#loc217 = loc("p_dv"(#loc77)) +#loc218 = loc("p_dv"(#loc78)) +#loc219 = loc("p_dv"(#loc79)) +#loc220 = loc("p_dv"(#loc80)) +#loc221 = loc("p_dv2"(#loc81)) +#loc222 = loc("p_dv2"(#loc82)) +#loc223 = loc("p_do"(#loc83)) +#loc224 = loc("p_do"(#loc84)) +#loc225 = loc("b_do"(#loc85)) +#loc226 = loc("b_do"(#loc86)) +#loc227 = loc("b_do"(#loc87)) +#loc228 = loc("p_k"(#loc88)) +#loc229 = loc("p_k"(#loc89)) +#loc230 = loc("p_k"(#loc90)) +#loc231 = loc("b_k"(#loc91)) +#loc232 = loc("b_k"(#loc92)) +#loc233 = loc("b_k"(#loc93)) +#loc234 = loc("b_dv"(#loc94)) +#loc235 = loc("p_k"(#loc95)) +#loc236 = loc("p_k"(#loc96)) +#loc237 = loc("b_k"(#loc97)) +#loc238 = loc("b_k"(#loc98)) +#loc239 = loc("b_k"(#loc99)) +#loc240 = loc("b_dv"(#loc100)) +#loc241 = loc("b_dv"(#loc101)) +#loc242 = loc("b_dv"(#loc102)) +#loc243 = loc("b_dv"(#loc103)) +#loc244 = loc("b_dv"(#loc104)) +#loc245 = loc("b_dv"(#loc105)) +#loc246 = loc("b_dv"(#loc106)) +#loc247 = loc("b_dv"(#loc107)) +#loc248 = loc("p_w"(#loc110)) +#loc249 = loc("p_w"(#loc111)) +#loc250 = loc("p_w"(#loc112)) +#loc251 = loc("p_w"(#loc113)) +#loc252 = loc("p_q"(#loc114)) +#loc253 = loc("p_q"(#loc115)) +#loc254 = loc("b_w"(#loc116)) +#loc255 = loc("b_w"(#loc117)) +#loc256 = loc("b_w"(#loc118)) +#loc257 = loc("b_q"(#loc119)) +#loc258 = loc("b_dh1"(#loc120)) +#loc259 = loc("b_q"(#loc121)) +#loc260 = loc("b_q"(#loc122)) +#loc261 = loc("b_dh1"(#loc123)) +#loc262 = loc("b_dh1"(#loc124)) +#loc263 = loc("b_dh1"(#loc125)) +#loc264 = loc("b_dh1"(#loc126)) +#loc265 = loc("b_dh1"(#loc127)) +#loc266 = loc("b_dh1"(#loc128)) +#loc267 = loc("p_q"(#loc129)) +#loc268 = loc("p_q"(#loc130)) +#loc269 = loc("p_w"(#loc131)) +#loc270 = loc("p_w"(#loc132)) +#loc271 = loc("b_q"(#loc133)) +#loc272 = loc("b_q"(#loc134)) +#loc273 = loc("b_w"(#loc135)) +#loc274 = loc("b_dh2"(#loc136)) +#loc275 = loc("b_q"(#loc137)) +#loc276 = loc("b_dh2"(#loc138)) +#loc277 = loc("b_dh2"(#loc139)) +#loc278 = loc("b_dh2"(#loc140)) +#loc279 = loc("b_dh2"(#loc141)) +#loc280 = loc("b_dh2"(#loc142)) +#loc281 = loc(callsite(#loc1 at #loc155)) +#loc282 = loc(callsite(#loc14 at #loc155)) +#loc283 = loc(callsite(#loc15 at #loc155)) +#loc284 = loc("b_dh2"(#loc178)) +#loc285 = loc("i_t"(#loc179)) +#loc286 = loc(fused[#loc183, #loc184]) +#loc287 = loc(fused[#loc185, #loc186]) +#loc288 = loc(fused[#loc207, #loc208]) +#loc289 = loc(fused[#loc210, #loc211]) +#loc290 = loc(callsite(#loc74 at #loc215)) +#loc291 = loc(callsite(#loc74 at #loc216)) +#loc292 = loc(callsite(#loc74 at #loc242)) diff --git a/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/__grp__l2norm_bwd_kernel.json b/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/__grp__l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..48774f9f83d1f1a6a60520faf5320990f0cee74e --- /dev/null +++ b/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/__grp__l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_bwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/l2norm_bwd_kernel.source", "l2norm_bwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/l2norm_bwd_kernel.ttir", "l2norm_bwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/l2norm_bwd_kernel.ttgir", "l2norm_bwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/l2norm_bwd_kernel.llir", "l2norm_bwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/l2norm_bwd_kernel.ptx", "l2norm_bwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/l2norm_bwd_kernel.cubin", "l2norm_bwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/l2norm_bwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/l2norm_bwd_kernel.cubin b/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/l2norm_bwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..5d39311f038b619b2cf7499efdbe0049a3ad390c Binary files /dev/null and b/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/l2norm_bwd_kernel.cubin differ diff --git a/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/l2norm_bwd_kernel.json b/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..368ce520cc4158a080e49b683ff8d56d935aeb10 --- /dev/null +++ b/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"hash": "064b031b2e4754e4e026cd465ce4dd1d174ce892f342095cdcec3a6da346580f", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 4, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 0, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_bwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/l2norm_bwd_kernel.llir b/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/l2norm_bwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..280756d7ca54c5cb28ca920334eb0532fa133513 --- /dev/null +++ b/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/l2norm_bwd_kernel.llir @@ -0,0 +1,673 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_bwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, ptr addrspace(1) %3, float %4, i32 %5, ptr addrspace(1) readnone captures(none) %6, ptr addrspace(1) readnone captures(none) %7) local_unnamed_addr #0 !dbg !4 { + %9 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !12 + %10 = zext nneg i32 %9 to i64, !dbg !13 + %11 = shl nuw nsw i64 %10, 5, !dbg !14 + %12 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !15 + %13 = lshr i32 %12, 4, !dbg !15 + %14 = and i32 %13, 7, !dbg !15 + %15 = or disjoint i32 %14, 8, !dbg !15 + %16 = or disjoint i32 %14, 16, !dbg !15 + %17 = or disjoint i32 %14, 24, !dbg !15 + %18 = zext nneg i32 %14 to i64, !dbg !16 + %19 = zext nneg i32 %15 to i64, !dbg !16 + %20 = zext nneg i32 %16 to i64, !dbg !16 + %21 = zext nneg i32 %17 to i64, !dbg !16 + %22 = or disjoint i64 %11, %18, !dbg !16 + %23 = or disjoint i64 %11, %19, !dbg !16 + %24 = or disjoint i64 %11, %20, !dbg !16 + %25 = or disjoint i64 %11, %21, !dbg !16 + %26 = sext i32 %5 to i64, !dbg !17 + %27 = icmp slt i64 %22, %26, !dbg !17 + %28 = icmp slt i64 %23, %26, !dbg !17 + %29 = icmp slt i64 %24, %26, !dbg !17 + %30 = icmp slt i64 %25, %26, !dbg !17 + %31 = shl nuw nsw i32 %12, 3, !dbg !18 + %32 = and i32 %31, 120, !dbg !18 + %33 = shl nuw nsw i64 %22, 7, !dbg !19 + %34 = shl nuw nsw i64 %23, 7, !dbg !19 + %35 = shl nuw nsw i64 %24, 7, !dbg !19 + %36 = shl nuw nsw i64 %25, 7, !dbg !19 + %37 = getelementptr bfloat, ptr addrspace(1) %0, i64 %33, !dbg !20 + %38 = getelementptr bfloat, ptr addrspace(1) %0, i64 %34, !dbg !20 + %39 = getelementptr bfloat, ptr addrspace(1) %0, i64 %35, !dbg !20 + %40 = getelementptr bfloat, ptr addrspace(1) %0, i64 %36, !dbg !20 + %41 = zext nneg i32 %32 to i64, !dbg !21 + %42 = getelementptr bfloat, ptr addrspace(1) %37, i64 %41, !dbg !21 + %43 = getelementptr bfloat, ptr addrspace(1) %38, i64 %41, !dbg !21 + %44 = getelementptr bfloat, ptr addrspace(1) %39, i64 %41, !dbg !21 + %45 = getelementptr bfloat, ptr addrspace(1) %40, i64 %41, !dbg !21 + %46 = getelementptr float, ptr addrspace(1) %1, i64 %22, !dbg !22 + %47 = getelementptr float, ptr addrspace(1) %1, i64 %23, !dbg !22 + %48 = getelementptr float, ptr addrspace(1) %1, i64 %24, !dbg !22 + %49 = getelementptr float, ptr addrspace(1) %1, i64 %25, !dbg !22 + %50 = getelementptr bfloat, ptr addrspace(1) %2, i64 %33, !dbg !23 + %51 = getelementptr bfloat, ptr addrspace(1) %2, i64 %34, !dbg !23 + %52 = getelementptr bfloat, ptr addrspace(1) %2, i64 %35, !dbg !23 + %53 = getelementptr bfloat, ptr addrspace(1) %2, i64 %36, !dbg !23 + %54 = getelementptr bfloat, ptr addrspace(1) %50, i64 %41, !dbg !24 + %55 = getelementptr bfloat, ptr addrspace(1) %51, i64 %41, !dbg !24 + %56 = getelementptr bfloat, ptr addrspace(1) %52, i64 %41, !dbg !24 + %57 = getelementptr bfloat, ptr addrspace(1) %53, i64 %41, !dbg !24 + %58 = getelementptr bfloat, ptr addrspace(1) %3, i64 %33, !dbg !25 + %59 = getelementptr bfloat, ptr addrspace(1) %3, i64 %34, !dbg !25 + %60 = getelementptr bfloat, ptr addrspace(1) %3, i64 %35, !dbg !25 + %61 = getelementptr bfloat, ptr addrspace(1) %3, i64 %36, !dbg !25 + %62 = getelementptr bfloat, ptr addrspace(1) %58, i64 %41, !dbg !26 + %63 = getelementptr bfloat, ptr addrspace(1) %59, i64 %41, !dbg !26 + %64 = getelementptr bfloat, ptr addrspace(1) %60, i64 %41, !dbg !26 + %65 = getelementptr bfloat, ptr addrspace(1) %61, i64 %41, !dbg !26 + %66 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %42, i1 %27) #3, !dbg !27 + %67 = extractvalue { i32, i32, i32, i32 } %66, 0, !dbg !27 + %68 = bitcast i32 %67 to <2 x bfloat>, !dbg !27 + %69 = extractvalue { i32, i32, i32, i32 } %66, 1, !dbg !27 + %70 = bitcast i32 %69 to <2 x bfloat>, !dbg !27 + %71 = extractvalue { i32, i32, i32, i32 } %66, 2, !dbg !27 + %72 = bitcast i32 %71 to <2 x bfloat>, !dbg !27 + %73 = extractvalue { i32, i32, i32, i32 } %66, 3, !dbg !27 + %74 = bitcast i32 %73 to <2 x bfloat>, !dbg !27 + %75 = extractelement <2 x bfloat> %68, i64 0, !dbg !27 + %76 = extractelement <2 x bfloat> %68, i64 1, !dbg !27 + %77 = extractelement <2 x bfloat> %70, i64 0, !dbg !27 + %78 = extractelement <2 x bfloat> %70, i64 1, !dbg !27 + %79 = extractelement <2 x bfloat> %72, i64 0, !dbg !27 + %80 = extractelement <2 x bfloat> %72, i64 1, !dbg !27 + %81 = extractelement <2 x bfloat> %74, i64 0, !dbg !27 + %82 = extractelement <2 x bfloat> %74, i64 1, !dbg !27 + %83 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %43, i1 %28) #3, !dbg !27 + %84 = extractvalue { i32, i32, i32, i32 } %83, 0, !dbg !27 + %85 = bitcast i32 %84 to <2 x bfloat>, !dbg !27 + %86 = extractvalue { i32, i32, i32, i32 } %83, 1, !dbg !27 + %87 = bitcast i32 %86 to <2 x bfloat>, !dbg !27 + %88 = extractvalue { i32, i32, i32, i32 } %83, 2, !dbg !27 + %89 = bitcast i32 %88 to <2 x bfloat>, !dbg !27 + %90 = extractvalue { i32, i32, i32, i32 } %83, 3, !dbg !27 + %91 = bitcast i32 %90 to <2 x bfloat>, !dbg !27 + %92 = extractelement <2 x bfloat> %85, i64 0, !dbg !27 + %93 = extractelement <2 x bfloat> %85, i64 1, !dbg !27 + %94 = extractelement <2 x bfloat> %87, i64 0, !dbg !27 + %95 = extractelement <2 x bfloat> %87, i64 1, !dbg !27 + %96 = extractelement <2 x bfloat> %89, i64 0, !dbg !27 + %97 = extractelement <2 x bfloat> %89, i64 1, !dbg !27 + %98 = extractelement <2 x bfloat> %91, i64 0, !dbg !27 + %99 = extractelement <2 x bfloat> %91, i64 1, !dbg !27 + %100 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %44, i1 %29) #3, !dbg !27 + %101 = extractvalue { i32, i32, i32, i32 } %100, 0, !dbg !27 + %102 = bitcast i32 %101 to <2 x bfloat>, !dbg !27 + %103 = extractvalue { i32, i32, i32, i32 } %100, 1, !dbg !27 + %104 = bitcast i32 %103 to <2 x bfloat>, !dbg !27 + %105 = extractvalue { i32, i32, i32, i32 } %100, 2, !dbg !27 + %106 = bitcast i32 %105 to <2 x bfloat>, !dbg !27 + %107 = extractvalue { i32, i32, i32, i32 } %100, 3, !dbg !27 + %108 = bitcast i32 %107 to <2 x bfloat>, !dbg !27 + %109 = extractelement <2 x bfloat> %102, i64 0, !dbg !27 + %110 = extractelement <2 x bfloat> %102, i64 1, !dbg !27 + %111 = extractelement <2 x bfloat> %104, i64 0, !dbg !27 + %112 = extractelement <2 x bfloat> %104, i64 1, !dbg !27 + %113 = extractelement <2 x bfloat> %106, i64 0, !dbg !27 + %114 = extractelement <2 x bfloat> %106, i64 1, !dbg !27 + %115 = extractelement <2 x bfloat> %108, i64 0, !dbg !27 + %116 = extractelement <2 x bfloat> %108, i64 1, !dbg !27 + %117 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %45, i1 %30) #3, !dbg !27 + %118 = extractvalue { i32, i32, i32, i32 } %117, 0, !dbg !27 + %119 = bitcast i32 %118 to <2 x bfloat>, !dbg !27 + %120 = extractvalue { i32, i32, i32, i32 } %117, 1, !dbg !27 + %121 = bitcast i32 %120 to <2 x bfloat>, !dbg !27 + %122 = extractvalue { i32, i32, i32, i32 } %117, 2, !dbg !27 + %123 = bitcast i32 %122 to <2 x bfloat>, !dbg !27 + %124 = extractvalue { i32, i32, i32, i32 } %117, 3, !dbg !27 + %125 = bitcast i32 %124 to <2 x bfloat>, !dbg !27 + %126 = extractelement <2 x bfloat> %119, i64 0, !dbg !27 + %127 = extractelement <2 x bfloat> %119, i64 1, !dbg !27 + %128 = extractelement <2 x bfloat> %121, i64 0, !dbg !27 + %129 = extractelement <2 x bfloat> %121, i64 1, !dbg !27 + %130 = extractelement <2 x bfloat> %123, i64 0, !dbg !27 + %131 = extractelement <2 x bfloat> %123, i64 1, !dbg !27 + %132 = extractelement <2 x bfloat> %125, i64 0, !dbg !27 + %133 = extractelement <2 x bfloat> %125, i64 1, !dbg !27 + %134 = fpext bfloat %75 to float, !dbg !28 + %135 = fpext bfloat %76 to float, !dbg !28 + %136 = fpext bfloat %77 to float, !dbg !28 + %137 = fpext bfloat %78 to float, !dbg !28 + %138 = fpext bfloat %79 to float, !dbg !28 + %139 = fpext bfloat %80 to float, !dbg !28 + %140 = fpext bfloat %81 to float, !dbg !28 + %141 = fpext bfloat %82 to float, !dbg !28 + %142 = fpext bfloat %92 to float, !dbg !28 + %143 = fpext bfloat %93 to float, !dbg !28 + %144 = fpext bfloat %94 to float, !dbg !28 + %145 = fpext bfloat %95 to float, !dbg !28 + %146 = fpext bfloat %96 to float, !dbg !28 + %147 = fpext bfloat %97 to float, !dbg !28 + %148 = fpext bfloat %98 to float, !dbg !28 + %149 = fpext bfloat %99 to float, !dbg !28 + %150 = fpext bfloat %109 to float, !dbg !28 + %151 = fpext bfloat %110 to float, !dbg !28 + %152 = fpext bfloat %111 to float, !dbg !28 + %153 = fpext bfloat %112 to float, !dbg !28 + %154 = fpext bfloat %113 to float, !dbg !28 + %155 = fpext bfloat %114 to float, !dbg !28 + %156 = fpext bfloat %115 to float, !dbg !28 + %157 = fpext bfloat %116 to float, !dbg !28 + %158 = fpext bfloat %126 to float, !dbg !28 + %159 = fpext bfloat %127 to float, !dbg !28 + %160 = fpext bfloat %128 to float, !dbg !28 + %161 = fpext bfloat %129 to float, !dbg !28 + %162 = fpext bfloat %130 to float, !dbg !28 + %163 = fpext bfloat %131 to float, !dbg !28 + %164 = fpext bfloat %132 to float, !dbg !28 + %165 = fpext bfloat %133 to float, !dbg !28 + %166 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %46, i1 %27) #3, !dbg !29 + %167 = bitcast i32 %166 to float, !dbg !29 + %168 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %47, i1 %28) #3, !dbg !29 + %169 = bitcast i32 %168 to float, !dbg !29 + %170 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %48, i1 %29) #3, !dbg !29 + %171 = bitcast i32 %170 to float, !dbg !29 + %172 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %49, i1 %30) #3, !dbg !29 + %173 = bitcast i32 %172 to float, !dbg !29 + %174 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %54, i1 %27) #3, !dbg !30 + %175 = extractvalue { i32, i32, i32, i32 } %174, 0, !dbg !30 + %176 = bitcast i32 %175 to <2 x bfloat>, !dbg !30 + %177 = extractvalue { i32, i32, i32, i32 } %174, 1, !dbg !30 + %178 = bitcast i32 %177 to <2 x bfloat>, !dbg !30 + %179 = extractvalue { i32, i32, i32, i32 } %174, 2, !dbg !30 + %180 = bitcast i32 %179 to <2 x bfloat>, !dbg !30 + %181 = extractvalue { i32, i32, i32, i32 } %174, 3, !dbg !30 + %182 = bitcast i32 %181 to <2 x bfloat>, !dbg !30 + %183 = extractelement <2 x bfloat> %176, i64 0, !dbg !30 + %184 = extractelement <2 x bfloat> %176, i64 1, !dbg !30 + %185 = extractelement <2 x bfloat> %178, i64 0, !dbg !30 + %186 = extractelement <2 x bfloat> %178, i64 1, !dbg !30 + %187 = extractelement <2 x bfloat> %180, i64 0, !dbg !30 + %188 = extractelement <2 x bfloat> %180, i64 1, !dbg !30 + %189 = extractelement <2 x bfloat> %182, i64 0, !dbg !30 + %190 = extractelement <2 x bfloat> %182, i64 1, !dbg !30 + %191 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %55, i1 %28) #3, !dbg !30 + %192 = extractvalue { i32, i32, i32, i32 } %191, 0, !dbg !30 + %193 = bitcast i32 %192 to <2 x bfloat>, !dbg !30 + %194 = extractvalue { i32, i32, i32, i32 } %191, 1, !dbg !30 + %195 = bitcast i32 %194 to <2 x bfloat>, !dbg !30 + %196 = extractvalue { i32, i32, i32, i32 } %191, 2, !dbg !30 + %197 = bitcast i32 %196 to <2 x bfloat>, !dbg !30 + %198 = extractvalue { i32, i32, i32, i32 } %191, 3, !dbg !30 + %199 = bitcast i32 %198 to <2 x bfloat>, !dbg !30 + %200 = extractelement <2 x bfloat> %193, i64 0, !dbg !30 + %201 = extractelement <2 x bfloat> %193, i64 1, !dbg !30 + %202 = extractelement <2 x bfloat> %195, i64 0, !dbg !30 + %203 = extractelement <2 x bfloat> %195, i64 1, !dbg !30 + %204 = extractelement <2 x bfloat> %197, i64 0, !dbg !30 + %205 = extractelement <2 x bfloat> %197, i64 1, !dbg !30 + %206 = extractelement <2 x bfloat> %199, i64 0, !dbg !30 + %207 = extractelement <2 x bfloat> %199, i64 1, !dbg !30 + %208 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %56, i1 %29) #3, !dbg !30 + %209 = extractvalue { i32, i32, i32, i32 } %208, 0, !dbg !30 + %210 = bitcast i32 %209 to <2 x bfloat>, !dbg !30 + %211 = extractvalue { i32, i32, i32, i32 } %208, 1, !dbg !30 + %212 = bitcast i32 %211 to <2 x bfloat>, !dbg !30 + %213 = extractvalue { i32, i32, i32, i32 } %208, 2, !dbg !30 + %214 = bitcast i32 %213 to <2 x bfloat>, !dbg !30 + %215 = extractvalue { i32, i32, i32, i32 } %208, 3, !dbg !30 + %216 = bitcast i32 %215 to <2 x bfloat>, !dbg !30 + %217 = extractelement <2 x bfloat> %210, i64 0, !dbg !30 + %218 = extractelement <2 x bfloat> %210, i64 1, !dbg !30 + %219 = extractelement <2 x bfloat> %212, i64 0, !dbg !30 + %220 = extractelement <2 x bfloat> %212, i64 1, !dbg !30 + %221 = extractelement <2 x bfloat> %214, i64 0, !dbg !30 + %222 = extractelement <2 x bfloat> %214, i64 1, !dbg !30 + %223 = extractelement <2 x bfloat> %216, i64 0, !dbg !30 + %224 = extractelement <2 x bfloat> %216, i64 1, !dbg !30 + %225 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %57, i1 %30) #3, !dbg !30 + %226 = extractvalue { i32, i32, i32, i32 } %225, 0, !dbg !30 + %227 = bitcast i32 %226 to <2 x bfloat>, !dbg !30 + %228 = extractvalue { i32, i32, i32, i32 } %225, 1, !dbg !30 + %229 = bitcast i32 %228 to <2 x bfloat>, !dbg !30 + %230 = extractvalue { i32, i32, i32, i32 } %225, 2, !dbg !30 + %231 = bitcast i32 %230 to <2 x bfloat>, !dbg !30 + %232 = extractvalue { i32, i32, i32, i32 } %225, 3, !dbg !30 + %233 = bitcast i32 %232 to <2 x bfloat>, !dbg !30 + %234 = extractelement <2 x bfloat> %227, i64 0, !dbg !30 + %235 = extractelement <2 x bfloat> %227, i64 1, !dbg !30 + %236 = extractelement <2 x bfloat> %229, i64 0, !dbg !30 + %237 = extractelement <2 x bfloat> %229, i64 1, !dbg !30 + %238 = extractelement <2 x bfloat> %231, i64 0, !dbg !30 + %239 = extractelement <2 x bfloat> %231, i64 1, !dbg !30 + %240 = extractelement <2 x bfloat> %233, i64 0, !dbg !30 + %241 = extractelement <2 x bfloat> %233, i64 1, !dbg !30 + %242 = fpext bfloat %183 to float, !dbg !31 + %243 = fpext bfloat %184 to float, !dbg !31 + %244 = fpext bfloat %185 to float, !dbg !31 + %245 = fpext bfloat %186 to float, !dbg !31 + %246 = fpext bfloat %187 to float, !dbg !31 + %247 = fpext bfloat %188 to float, !dbg !31 + %248 = fpext bfloat %189 to float, !dbg !31 + %249 = fpext bfloat %190 to float, !dbg !31 + %250 = fpext bfloat %200 to float, !dbg !31 + %251 = fpext bfloat %201 to float, !dbg !31 + %252 = fpext bfloat %202 to float, !dbg !31 + %253 = fpext bfloat %203 to float, !dbg !31 + %254 = fpext bfloat %204 to float, !dbg !31 + %255 = fpext bfloat %205 to float, !dbg !31 + %256 = fpext bfloat %206 to float, !dbg !31 + %257 = fpext bfloat %207 to float, !dbg !31 + %258 = fpext bfloat %217 to float, !dbg !31 + %259 = fpext bfloat %218 to float, !dbg !31 + %260 = fpext bfloat %219 to float, !dbg !31 + %261 = fpext bfloat %220 to float, !dbg !31 + %262 = fpext bfloat %221 to float, !dbg !31 + %263 = fpext bfloat %222 to float, !dbg !31 + %264 = fpext bfloat %223 to float, !dbg !31 + %265 = fpext bfloat %224 to float, !dbg !31 + %266 = fpext bfloat %234 to float, !dbg !31 + %267 = fpext bfloat %235 to float, !dbg !31 + %268 = fpext bfloat %236 to float, !dbg !31 + %269 = fpext bfloat %237 to float, !dbg !31 + %270 = fpext bfloat %238 to float, !dbg !31 + %271 = fpext bfloat %239 to float, !dbg !31 + %272 = fpext bfloat %240 to float, !dbg !31 + %273 = fpext bfloat %241 to float, !dbg !31 + %274 = fmul float %167, %242, !dbg !32 + %275 = fmul float %167, %243, !dbg !32 + %276 = fmul float %167, %244, !dbg !32 + %277 = fmul float %167, %245, !dbg !32 + %278 = fmul float %167, %246, !dbg !32 + %279 = fmul float %167, %247, !dbg !32 + %280 = fmul float %167, %248, !dbg !32 + %281 = fmul float %167, %249, !dbg !32 + %282 = fmul float %169, %250, !dbg !32 + %283 = fmul float %169, %251, !dbg !32 + %284 = fmul float %169, %252, !dbg !32 + %285 = fmul float %169, %253, !dbg !32 + %286 = fmul float %169, %254, !dbg !32 + %287 = fmul float %169, %255, !dbg !32 + %288 = fmul float %169, %256, !dbg !32 + %289 = fmul float %169, %257, !dbg !32 + %290 = fmul float %171, %258, !dbg !32 + %291 = fmul float %171, %259, !dbg !32 + %292 = fmul float %171, %260, !dbg !32 + %293 = fmul float %171, %261, !dbg !32 + %294 = fmul float %171, %262, !dbg !32 + %295 = fmul float %171, %263, !dbg !32 + %296 = fmul float %171, %264, !dbg !32 + %297 = fmul float %171, %265, !dbg !32 + %298 = fmul float %173, %266, !dbg !32 + %299 = fmul float %173, %267, !dbg !32 + %300 = fmul float %173, %268, !dbg !32 + %301 = fmul float %173, %269, !dbg !32 + %302 = fmul float %173, %270, !dbg !32 + %303 = fmul float %173, %271, !dbg !32 + %304 = fmul float %173, %272, !dbg !32 + %305 = fmul float %173, %273, !dbg !32 + %306 = fmul float %134, %242, !dbg !33 + %307 = fmul float %135, %243, !dbg !33 + %308 = fmul float %136, %244, !dbg !33 + %309 = fmul float %137, %245, !dbg !33 + %310 = fmul float %138, %246, !dbg !33 + %311 = fmul float %139, %247, !dbg !33 + %312 = fmul float %140, %248, !dbg !33 + %313 = fmul float %141, %249, !dbg !33 + %314 = fmul float %142, %250, !dbg !33 + %315 = fmul float %143, %251, !dbg !33 + %316 = fmul float %144, %252, !dbg !33 + %317 = fmul float %145, %253, !dbg !33 + %318 = fmul float %146, %254, !dbg !33 + %319 = fmul float %147, %255, !dbg !33 + %320 = fmul float %148, %256, !dbg !33 + %321 = fmul float %149, %257, !dbg !33 + %322 = fmul float %150, %258, !dbg !33 + %323 = fmul float %151, %259, !dbg !33 + %324 = fmul float %152, %260, !dbg !33 + %325 = fmul float %153, %261, !dbg !33 + %326 = fmul float %154, %262, !dbg !33 + %327 = fmul float %155, %263, !dbg !33 + %328 = fmul float %156, %264, !dbg !33 + %329 = fmul float %157, %265, !dbg !33 + %330 = fmul float %158, %266, !dbg !33 + %331 = fmul float %159, %267, !dbg !33 + %332 = fmul float %160, %268, !dbg !33 + %333 = fmul float %161, %269, !dbg !33 + %334 = fmul float %162, %270, !dbg !33 + %335 = fmul float %163, %271, !dbg !33 + %336 = fmul float %164, %272, !dbg !33 + %337 = fmul float %165, %273, !dbg !33 + %338 = fadd float %306, %307, !dbg !34 + %339 = fadd float %308, %338, !dbg !34 + %340 = fadd float %309, %339, !dbg !34 + %341 = fadd float %310, %340, !dbg !34 + %342 = fadd float %311, %341, !dbg !34 + %343 = fadd float %312, %342, !dbg !34 + %344 = fadd float %313, %343, !dbg !34 + %345 = fadd float %314, %315, !dbg !34 + %346 = fadd float %316, %345, !dbg !34 + %347 = fadd float %317, %346, !dbg !34 + %348 = fadd float %318, %347, !dbg !34 + %349 = fadd float %319, %348, !dbg !34 + %350 = fadd float %320, %349, !dbg !34 + %351 = fadd float %321, %350, !dbg !34 + %352 = fadd float %322, %323, !dbg !34 + %353 = fadd float %324, %352, !dbg !34 + %354 = fadd float %325, %353, !dbg !34 + %355 = fadd float %326, %354, !dbg !34 + %356 = fadd float %327, %355, !dbg !34 + %357 = fadd float %328, %356, !dbg !34 + %358 = fadd float %329, %357, !dbg !34 + %359 = fadd float %330, %331, !dbg !34 + %360 = fadd float %332, %359, !dbg !34 + %361 = fadd float %333, %360, !dbg !34 + %362 = fadd float %334, %361, !dbg !34 + %363 = fadd float %335, %362, !dbg !34 + %364 = fadd float %336, %363, !dbg !34 + %365 = fadd float %337, %364, !dbg !34 + %366 = bitcast float %344 to i32, !dbg !37 + %367 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %366, i32 8, i32 31), !dbg !37 + %368 = bitcast i32 %367 to float, !dbg !37 + %369 = fadd float %344, %368, !dbg !34 + %370 = bitcast float %369 to i32, !dbg !37 + %371 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %370, i32 4, i32 31), !dbg !37 + %372 = bitcast i32 %371 to float, !dbg !37 + %373 = fadd float %369, %372, !dbg !34 + %374 = bitcast float %373 to i32, !dbg !37 + %375 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %374, i32 2, i32 31), !dbg !37 + %376 = bitcast i32 %375 to float, !dbg !37 + %377 = fadd float %373, %376, !dbg !34 + %378 = bitcast float %377 to i32, !dbg !37 + %379 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %378, i32 1, i32 31), !dbg !37 + %380 = bitcast i32 %379 to float, !dbg !37 + %381 = fadd float %377, %380, !dbg !34 + %382 = bitcast float %351 to i32, !dbg !37 + %383 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %382, i32 8, i32 31), !dbg !37 + %384 = bitcast i32 %383 to float, !dbg !37 + %385 = fadd float %351, %384, !dbg !34 + %386 = bitcast float %385 to i32, !dbg !37 + %387 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %386, i32 4, i32 31), !dbg !37 + %388 = bitcast i32 %387 to float, !dbg !37 + %389 = fadd float %385, %388, !dbg !34 + %390 = bitcast float %389 to i32, !dbg !37 + %391 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %390, i32 2, i32 31), !dbg !37 + %392 = bitcast i32 %391 to float, !dbg !37 + %393 = fadd float %389, %392, !dbg !34 + %394 = bitcast float %393 to i32, !dbg !37 + %395 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %394, i32 1, i32 31), !dbg !37 + %396 = bitcast i32 %395 to float, !dbg !37 + %397 = fadd float %393, %396, !dbg !34 + %398 = bitcast float %358 to i32, !dbg !37 + %399 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %398, i32 8, i32 31), !dbg !37 + %400 = bitcast i32 %399 to float, !dbg !37 + %401 = fadd float %358, %400, !dbg !34 + %402 = bitcast float %401 to i32, !dbg !37 + %403 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %402, i32 4, i32 31), !dbg !37 + %404 = bitcast i32 %403 to float, !dbg !37 + %405 = fadd float %401, %404, !dbg !34 + %406 = bitcast float %405 to i32, !dbg !37 + %407 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %406, i32 2, i32 31), !dbg !37 + %408 = bitcast i32 %407 to float, !dbg !37 + %409 = fadd float %405, %408, !dbg !34 + %410 = bitcast float %409 to i32, !dbg !37 + %411 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %410, i32 1, i32 31), !dbg !37 + %412 = bitcast i32 %411 to float, !dbg !37 + %413 = fadd float %409, %412, !dbg !34 + %414 = bitcast float %365 to i32, !dbg !37 + %415 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %414, i32 8, i32 31), !dbg !37 + %416 = bitcast i32 %415 to float, !dbg !37 + %417 = fadd float %365, %416, !dbg !34 + %418 = bitcast float %417 to i32, !dbg !37 + %419 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %418, i32 4, i32 31), !dbg !37 + %420 = bitcast i32 %419 to float, !dbg !37 + %421 = fadd float %417, %420, !dbg !34 + %422 = bitcast float %421 to i32, !dbg !37 + %423 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %422, i32 2, i32 31), !dbg !37 + %424 = bitcast i32 %423 to float, !dbg !37 + %425 = fadd float %421, %424, !dbg !34 + %426 = bitcast float %425 to i32, !dbg !37 + %427 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %426, i32 1, i32 31), !dbg !37 + %428 = bitcast i32 %427 to float, !dbg !37 + %429 = fadd float %425, %428, !dbg !34 + %430 = fmul float %381, %134, !dbg !40 + %431 = fmul float %381, %135, !dbg !40 + %432 = fmul float %381, %136, !dbg !40 + %433 = fmul float %381, %137, !dbg !40 + %434 = fmul float %381, %138, !dbg !40 + %435 = fmul float %381, %139, !dbg !40 + %436 = fmul float %381, %140, !dbg !40 + %437 = fmul float %381, %141, !dbg !40 + %438 = fmul float %397, %142, !dbg !40 + %439 = fmul float %397, %143, !dbg !40 + %440 = fmul float %397, %144, !dbg !40 + %441 = fmul float %397, %145, !dbg !40 + %442 = fmul float %397, %146, !dbg !40 + %443 = fmul float %397, %147, !dbg !40 + %444 = fmul float %397, %148, !dbg !40 + %445 = fmul float %397, %149, !dbg !40 + %446 = fmul float %413, %150, !dbg !40 + %447 = fmul float %413, %151, !dbg !40 + %448 = fmul float %413, %152, !dbg !40 + %449 = fmul float %413, %153, !dbg !40 + %450 = fmul float %413, %154, !dbg !40 + %451 = fmul float %413, %155, !dbg !40 + %452 = fmul float %413, %156, !dbg !40 + %453 = fmul float %413, %157, !dbg !40 + %454 = fmul float %429, %158, !dbg !40 + %455 = fmul float %429, %159, !dbg !40 + %456 = fmul float %429, %160, !dbg !40 + %457 = fmul float %429, %161, !dbg !40 + %458 = fmul float %429, %162, !dbg !40 + %459 = fmul float %429, %163, !dbg !40 + %460 = fmul float %429, %164, !dbg !40 + %461 = fmul float %429, %165, !dbg !40 + %462 = fmul float %430, %167, !dbg !41 + %463 = fmul float %431, %167, !dbg !41 + %464 = fmul float %432, %167, !dbg !41 + %465 = fmul float %433, %167, !dbg !41 + %466 = fmul float %434, %167, !dbg !41 + %467 = fmul float %435, %167, !dbg !41 + %468 = fmul float %436, %167, !dbg !41 + %469 = fmul float %437, %167, !dbg !41 + %470 = fmul float %438, %169, !dbg !41 + %471 = fmul float %439, %169, !dbg !41 + %472 = fmul float %440, %169, !dbg !41 + %473 = fmul float %441, %169, !dbg !41 + %474 = fmul float %442, %169, !dbg !41 + %475 = fmul float %443, %169, !dbg !41 + %476 = fmul float %444, %169, !dbg !41 + %477 = fmul float %445, %169, !dbg !41 + %478 = fmul float %446, %171, !dbg !41 + %479 = fmul float %447, %171, !dbg !41 + %480 = fmul float %448, %171, !dbg !41 + %481 = fmul float %449, %171, !dbg !41 + %482 = fmul float %450, %171, !dbg !41 + %483 = fmul float %451, %171, !dbg !41 + %484 = fmul float %452, %171, !dbg !41 + %485 = fmul float %453, %171, !dbg !41 + %486 = fmul float %454, %173, !dbg !41 + %487 = fmul float %455, %173, !dbg !41 + %488 = fmul float %456, %173, !dbg !41 + %489 = fmul float %457, %173, !dbg !41 + %490 = fmul float %458, %173, !dbg !41 + %491 = fmul float %459, %173, !dbg !41 + %492 = fmul float %460, %173, !dbg !41 + %493 = fmul float %461, %173, !dbg !41 + %494 = fsub float %274, %462, !dbg !42 + %495 = fsub float %275, %463, !dbg !42 + %496 = fsub float %276, %464, !dbg !42 + %497 = fsub float %277, %465, !dbg !42 + %498 = fsub float %278, %466, !dbg !42 + %499 = fsub float %279, %467, !dbg !42 + %500 = fsub float %280, %468, !dbg !42 + %501 = fsub float %281, %469, !dbg !42 + %502 = fsub float %282, %470, !dbg !42 + %503 = fsub float %283, %471, !dbg !42 + %504 = fsub float %284, %472, !dbg !42 + %505 = fsub float %285, %473, !dbg !42 + %506 = fsub float %286, %474, !dbg !42 + %507 = fsub float %287, %475, !dbg !42 + %508 = fsub float %288, %476, !dbg !42 + %509 = fsub float %289, %477, !dbg !42 + %510 = fsub float %290, %478, !dbg !42 + %511 = fsub float %291, %479, !dbg !42 + %512 = fsub float %292, %480, !dbg !42 + %513 = fsub float %293, %481, !dbg !42 + %514 = fsub float %294, %482, !dbg !42 + %515 = fsub float %295, %483, !dbg !42 + %516 = fsub float %296, %484, !dbg !42 + %517 = fsub float %297, %485, !dbg !42 + %518 = fsub float %298, %486, !dbg !42 + %519 = fsub float %299, %487, !dbg !42 + %520 = fsub float %300, %488, !dbg !42 + %521 = fsub float %301, %489, !dbg !42 + %522 = fsub float %302, %490, !dbg !42 + %523 = fsub float %303, %491, !dbg !42 + %524 = fsub float %304, %492, !dbg !42 + %525 = fsub float %305, %493, !dbg !42 + %526 = fptrunc float %494 to bfloat, !dbg !43 + %527 = fptrunc float %495 to bfloat, !dbg !43 + %528 = fptrunc float %496 to bfloat, !dbg !43 + %529 = fptrunc float %497 to bfloat, !dbg !43 + %530 = fptrunc float %498 to bfloat, !dbg !43 + %531 = fptrunc float %499 to bfloat, !dbg !43 + %532 = fptrunc float %500 to bfloat, !dbg !43 + %533 = fptrunc float %501 to bfloat, !dbg !43 + %534 = fptrunc float %502 to bfloat, !dbg !43 + %535 = fptrunc float %503 to bfloat, !dbg !43 + %536 = fptrunc float %504 to bfloat, !dbg !43 + %537 = fptrunc float %505 to bfloat, !dbg !43 + %538 = fptrunc float %506 to bfloat, !dbg !43 + %539 = fptrunc float %507 to bfloat, !dbg !43 + %540 = fptrunc float %508 to bfloat, !dbg !43 + %541 = fptrunc float %509 to bfloat, !dbg !43 + %542 = fptrunc float %510 to bfloat, !dbg !43 + %543 = fptrunc float %511 to bfloat, !dbg !43 + %544 = fptrunc float %512 to bfloat, !dbg !43 + %545 = fptrunc float %513 to bfloat, !dbg !43 + %546 = fptrunc float %514 to bfloat, !dbg !43 + %547 = fptrunc float %515 to bfloat, !dbg !43 + %548 = fptrunc float %516 to bfloat, !dbg !43 + %549 = fptrunc float %517 to bfloat, !dbg !43 + %550 = fptrunc float %518 to bfloat, !dbg !43 + %551 = fptrunc float %519 to bfloat, !dbg !43 + %552 = fptrunc float %520 to bfloat, !dbg !43 + %553 = fptrunc float %521 to bfloat, !dbg !43 + %554 = fptrunc float %522 to bfloat, !dbg !43 + %555 = fptrunc float %523 to bfloat, !dbg !43 + %556 = fptrunc float %524 to bfloat, !dbg !43 + %557 = fptrunc float %525 to bfloat, !dbg !43 + %558 = insertelement <2 x bfloat> poison, bfloat %526, i64 0, !dbg !44 + %559 = insertelement <2 x bfloat> %558, bfloat %527, i64 1, !dbg !44 + %560 = bitcast <2 x bfloat> %559 to i32, !dbg !44 + %561 = insertelement <2 x bfloat> poison, bfloat %528, i64 0, !dbg !44 + %562 = insertelement <2 x bfloat> %561, bfloat %529, i64 1, !dbg !44 + %563 = bitcast <2 x bfloat> %562 to i32, !dbg !44 + %564 = insertelement <2 x bfloat> poison, bfloat %530, i64 0, !dbg !44 + %565 = insertelement <2 x bfloat> %564, bfloat %531, i64 1, !dbg !44 + %566 = bitcast <2 x bfloat> %565 to i32, !dbg !44 + %567 = insertelement <2 x bfloat> poison, bfloat %532, i64 0, !dbg !44 + %568 = insertelement <2 x bfloat> %567, bfloat %533, i64 1, !dbg !44 + %569 = bitcast <2 x bfloat> %568 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %560, i32 %563, i32 %566, i32 %569, ptr addrspace(1) %62, i1 %27) #3, !dbg !44 + %570 = insertelement <2 x bfloat> poison, bfloat %534, i64 0, !dbg !44 + %571 = insertelement <2 x bfloat> %570, bfloat %535, i64 1, !dbg !44 + %572 = bitcast <2 x bfloat> %571 to i32, !dbg !44 + %573 = insertelement <2 x bfloat> poison, bfloat %536, i64 0, !dbg !44 + %574 = insertelement <2 x bfloat> %573, bfloat %537, i64 1, !dbg !44 + %575 = bitcast <2 x bfloat> %574 to i32, !dbg !44 + %576 = insertelement <2 x bfloat> poison, bfloat %538, i64 0, !dbg !44 + %577 = insertelement <2 x bfloat> %576, bfloat %539, i64 1, !dbg !44 + %578 = bitcast <2 x bfloat> %577 to i32, !dbg !44 + %579 = insertelement <2 x bfloat> poison, bfloat %540, i64 0, !dbg !44 + %580 = insertelement <2 x bfloat> %579, bfloat %541, i64 1, !dbg !44 + %581 = bitcast <2 x bfloat> %580 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %572, i32 %575, i32 %578, i32 %581, ptr addrspace(1) %63, i1 %28) #3, !dbg !44 + %582 = insertelement <2 x bfloat> poison, bfloat %542, i64 0, !dbg !44 + %583 = insertelement <2 x bfloat> %582, bfloat %543, i64 1, !dbg !44 + %584 = bitcast <2 x bfloat> %583 to i32, !dbg !44 + %585 = insertelement <2 x bfloat> poison, bfloat %544, i64 0, !dbg !44 + %586 = insertelement <2 x bfloat> %585, bfloat %545, i64 1, !dbg !44 + %587 = bitcast <2 x bfloat> %586 to i32, !dbg !44 + %588 = insertelement <2 x bfloat> poison, bfloat %546, i64 0, !dbg !44 + %589 = insertelement <2 x bfloat> %588, bfloat %547, i64 1, !dbg !44 + %590 = bitcast <2 x bfloat> %589 to i32, !dbg !44 + %591 = insertelement <2 x bfloat> poison, bfloat %548, i64 0, !dbg !44 + %592 = insertelement <2 x bfloat> %591, bfloat %549, i64 1, !dbg !44 + %593 = bitcast <2 x bfloat> %592 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %584, i32 %587, i32 %590, i32 %593, ptr addrspace(1) %64, i1 %29) #3, !dbg !44 + %594 = insertelement <2 x bfloat> poison, bfloat %550, i64 0, !dbg !44 + %595 = insertelement <2 x bfloat> %594, bfloat %551, i64 1, !dbg !44 + %596 = bitcast <2 x bfloat> %595 to i32, !dbg !44 + %597 = insertelement <2 x bfloat> poison, bfloat %552, i64 0, !dbg !44 + %598 = insertelement <2 x bfloat> %597, bfloat %553, i64 1, !dbg !44 + %599 = bitcast <2 x bfloat> %598 to i32, !dbg !44 + %600 = insertelement <2 x bfloat> poison, bfloat %554, i64 0, !dbg !44 + %601 = insertelement <2 x bfloat> %600, bfloat %555, i64 1, !dbg !44 + %602 = bitcast <2 x bfloat> %601 to i32, !dbg !44 + %603 = insertelement <2 x bfloat> poison, bfloat %556, i64 0, !dbg !44 + %604 = insertelement <2 x bfloat> %603, bfloat %557, i64 1, !dbg !44 + %605 = bitcast <2 x bfloat> %604 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %596, i32 %599, i32 %602, i32 %605, ptr addrspace(1) %65, i1 %30) #3, !dbg !44 + ret void, !dbg !45 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +attributes #0 = { nounwind "nvvm.reqntid"="128" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = distinct !DISubprogram(name: "l2norm_bwd_kernel", linkageName: "l2norm_bwd_kernel", scope: !1, file: !1, line: 93, type: !5, scopeLine: 93, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!5 = !DISubroutineType(cc: DW_CC_normal, types: !6) +!6 = !{null, !7, !9, !7, !7, !10, !11, !7, !7} +!7 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !8, size: 64, dwarfAddressSpace: 1) +!8 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!9 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !10, size: 64, dwarfAddressSpace: 1) +!10 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!11 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!12 = !DILocation(line: 94, column: 24, scope: !4) +!13 = !DILocation(line: 94, column: 30, scope: !4) +!14 = !DILocation(line: 95, column: 16, scope: !4) +!15 = !DILocation(line: 95, column: 34, scope: !4) +!16 = !DILocation(line: 95, column: 21, scope: !4) +!17 = !DILocation(line: 97, column: 16, scope: !4) +!18 = !DILocation(line: 98, column: 30, scope: !4) +!19 = !DILocation(line: 99, column: 29, scope: !4) +!20 = !DILocation(line: 99, column: 14, scope: !4) +!21 = !DILocation(line: 99, column: 33, scope: !4) +!22 = !DILocation(line: 100, column: 20, scope: !4) +!23 = !DILocation(line: 101, column: 16, scope: !4) +!24 = !DILocation(line: 101, column: 35, scope: !4) +!25 = !DILocation(line: 102, column: 16, scope: !4) +!26 = !DILocation(line: 102, column: 35, scope: !4) +!27 = !DILocation(line: 104, column: 18, scope: !4) +!28 = !DILocation(line: 104, column: 47, scope: !4) +!29 = !DILocation(line: 105, column: 21, scope: !4) +!30 = !DILocation(line: 106, column: 19, scope: !4) +!31 = !DILocation(line: 106, column: 49, scope: !4) +!32 = !DILocation(line: 107, column: 18, scope: !4) +!33 = !DILocation(line: 107, column: 50, scope: !4) +!34 = !DILocation(line: 263, column: 15, scope: !35, inlinedAt: !37) +!35 = distinct !DILexicalBlockFile(scope: !4, file: !36, discriminator: 0) +!36 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!37 = !DILocation(line: 293, column: 36, scope: !35, inlinedAt: !38) +!38 = !DILocation(line: 107, column: 55, scope: !39) +!39 = distinct !DILexicalBlockFile(scope: !4, file: !1, discriminator: 0) +!40 = !DILocation(line: 107, column: 69, scope: !4) +!41 = !DILocation(line: 107, column: 75, scope: !4) +!42 = !DILocation(line: 107, column: 36, scope: !4) +!43 = !DILocation(line: 108, column: 27, scope: !4) +!44 = !DILocation(line: 108, column: 19, scope: !4) +!45 = !DILocation(line: 108, column: 4, scope: !4) diff --git a/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/l2norm_bwd_kernel.ptx b/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/l2norm_bwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..2d8fd8fb55b615b318b101f75d190c4ed395b1ae --- /dev/null +++ b/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/l2norm_bwd_kernel.ptx @@ -0,0 +1,796 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_bwd_kernel // -- Begin function l2norm_bwd_kernel + // @l2norm_bwd_kernel +.visible .entry l2norm_bwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_2, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_3, + .param .f32 l2norm_bwd_kernel_param_4, + .param .u32 l2norm_bwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_6, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_7 +) +.reqntid 128 +{ + .reg .pred %p<5>; + .reg .b16 %rs<65>; + .reg .b32 %r<318>; + .reg .b64 %rd<49>; + .loc 1 93 0 // l2norm.py:93:0 +$L__func_begin0: + .loc 1 93 0 // l2norm.py:93:0 + +// %bb.0: + ld.param.b64 %rd17, [l2norm_bwd_kernel_param_0]; + ld.param.b64 %rd18, [l2norm_bwd_kernel_param_1]; +$L__tmp0: + .loc 1 94 24 // l2norm.py:94:24 + mov.u32 %r54, %ctaid.x; + ld.param.b64 %rd19, [l2norm_bwd_kernel_param_2]; + .loc 1 95 16 // l2norm.py:95:16 + mul.wide.u32 %rd20, %r54, 32; + ld.param.b64 %rd21, [l2norm_bwd_kernel_param_3]; + .loc 1 95 34 // l2norm.py:95:34 + mov.u32 %r55, %tid.x; + ld.param.s32 %rd22, [l2norm_bwd_kernel_param_5]; + bfe.u32 %r56, %r55, 4, 3; + or.b32 %r57, %r56, 8; + or.b32 %r58, %r56, 16; + or.b32 %r59, %r56, 24; + .loc 1 95 21 // l2norm.py:95:21 + cvt.u64.u32 %rd23, %r56; + cvt.u64.u32 %rd24, %r57; + cvt.u64.u32 %rd25, %r58; + cvt.u64.u32 %rd26, %r59; + or.b64 %rd27, %rd20, %rd23; + or.b64 %rd28, %rd20, %rd24; + or.b64 %rd29, %rd20, %rd25; + or.b64 %rd30, %rd20, %rd26; + .loc 1 97 16 // l2norm.py:97:16 + setp.lt.s64 %p1, %rd27, %rd22; + setp.lt.s64 %p2, %rd28, %rd22; + setp.lt.s64 %p3, %rd29, %rd22; + setp.lt.s64 %p4, %rd30, %rd22; + .loc 1 98 30 // l2norm.py:98:30 + shl.b32 %r60, %r55, 3; + and.b32 %r61, %r60, 120; + .loc 1 99 14 // l2norm.py:99:14 + shl.b64 %rd31, %rd27, 8; + add.s64 %rd32, %rd17, %rd31; + shl.b64 %rd33, %rd28, 8; + add.s64 %rd34, %rd17, %rd33; + shl.b64 %rd35, %rd29, 8; + add.s64 %rd36, %rd17, %rd35; + shl.b64 %rd37, %rd30, 8; + add.s64 %rd38, %rd17, %rd37; + .loc 1 99 33 // l2norm.py:99:33 + mul.wide.u32 %rd39, %r61, 2; + add.s64 %rd1, %rd32, %rd39; + add.s64 %rd2, %rd34, %rd39; + add.s64 %rd3, %rd36, %rd39; + add.s64 %rd4, %rd38, %rd39; + .loc 1 100 20 // l2norm.py:100:20 + shl.b64 %rd40, %rd27, 2; + add.s64 %rd5, %rd18, %rd40; + add.s64 %rd6, %rd5, 32; + add.s64 %rd7, %rd5, 64; + add.s64 %rd8, %rd5, 96; + .loc 1 101 16 // l2norm.py:101:16 + add.s64 %rd41, %rd19, %rd31; + add.s64 %rd42, %rd19, %rd33; + add.s64 %rd43, %rd19, %rd35; + add.s64 %rd44, %rd19, %rd37; + .loc 1 101 35 // l2norm.py:101:35 + add.s64 %rd9, %rd41, %rd39; + add.s64 %rd10, %rd42, %rd39; + add.s64 %rd11, %rd43, %rd39; + add.s64 %rd12, %rd44, %rd39; + .loc 1 102 16 // l2norm.py:102:16 + add.s64 %rd45, %rd21, %rd31; + add.s64 %rd46, %rd21, %rd33; + add.s64 %rd47, %rd21, %rd35; + add.s64 %rd48, %rd21, %rd37; + .loc 1 102 35 // l2norm.py:102:35 + add.s64 %rd13, %rd45, %rd39; + add.s64 %rd14, %rd46, %rd39; + add.s64 %rd15, %rd47, %rd39; + add.s64 %rd16, %rd48, %rd39; + mov.b32 %r5, 0; + .loc 1 104 18 // l2norm.py:104:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + mov.b32 {%rs5, %rs6}, %r3; + mov.b32 {%rs7, %rs8}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs9, %rs10}, %r6; + mov.b32 {%rs11, %rs12}, %r7; + mov.b32 {%rs13, %rs14}, %r8; + mov.b32 {%rs15, %rs16}, %r9; + // begin inline asm + mov.u32 %r10, %r5; + mov.u32 %r11, %r5; + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + @%p3 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs17, %rs18}, %r10; + mov.b32 {%rs19, %rs20}, %r11; + mov.b32 {%rs21, %rs22}, %r12; + mov.b32 {%rs23, %rs24}, %r13; + // begin inline asm + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + @%p4 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ]; + // end inline asm + mov.b32 {%rs25, %rs26}, %r14; + mov.b32 {%rs27, %rs28}, %r15; + mov.b32 {%rs29, %rs30}, %r16; + mov.b32 {%rs31, %rs32}, %r17; + .loc 1 104 47 // l2norm.py:104:47 + cvt.f32.bf16 %r62, %rs1; + cvt.f32.bf16 %r63, %rs2; + cvt.f32.bf16 %r64, %rs3; + cvt.f32.bf16 %r65, %rs4; + cvt.f32.bf16 %r66, %rs5; + cvt.f32.bf16 %r67, %rs6; + cvt.f32.bf16 %r68, %rs7; + cvt.f32.bf16 %r69, %rs8; + cvt.f32.bf16 %r70, %rs9; + cvt.f32.bf16 %r71, %rs10; + cvt.f32.bf16 %r72, %rs11; + cvt.f32.bf16 %r73, %rs12; + cvt.f32.bf16 %r74, %rs13; + cvt.f32.bf16 %r75, %rs14; + cvt.f32.bf16 %r76, %rs15; + cvt.f32.bf16 %r77, %rs16; + cvt.f32.bf16 %r78, %rs17; + cvt.f32.bf16 %r79, %rs18; + cvt.f32.bf16 %r80, %rs19; + cvt.f32.bf16 %r81, %rs20; + cvt.f32.bf16 %r82, %rs21; + cvt.f32.bf16 %r83, %rs22; + cvt.f32.bf16 %r84, %rs23; + cvt.f32.bf16 %r85, %rs24; + cvt.f32.bf16 %r86, %rs25; + cvt.f32.bf16 %r87, %rs26; + cvt.f32.bf16 %r88, %rs27; + cvt.f32.bf16 %r89, %rs28; + cvt.f32.bf16 %r90, %rs29; + cvt.f32.bf16 %r91, %rs30; + cvt.f32.bf16 %r92, %rs31; + cvt.f32.bf16 %r93, %rs32; + .loc 1 105 21 // l2norm.py:105:21 + // begin inline asm + mov.u32 %r18, %r5; + @%p1 ld.global.b32 { %r18 }, [ %rd5 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r19, %r5; + @%p2 ld.global.b32 { %r19 }, [ %rd6 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r20, %r5; + @%p3 ld.global.b32 { %r20 }, [ %rd7 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r21, %r5; + @%p4 ld.global.b32 { %r21 }, [ %rd8 + 0 ]; + // end inline asm + .loc 1 106 19 // l2norm.py:106:19 + // begin inline asm + mov.u32 %r22, %r5; + mov.u32 %r23, %r5; + mov.u32 %r24, %r5; + mov.u32 %r25, %r5; + @%p1 ld.global.v4.b32 { %r22, %r23, %r24, %r25 }, [ %rd9 + 0 ]; + // end inline asm + mov.b32 {%rs33, %rs34}, %r22; + mov.b32 {%rs35, %rs36}, %r23; + mov.b32 {%rs37, %rs38}, %r24; + mov.b32 {%rs39, %rs40}, %r25; + // begin inline asm + mov.u32 %r26, %r5; + mov.u32 %r27, %r5; + mov.u32 %r28, %r5; + mov.u32 %r29, %r5; + @%p2 ld.global.v4.b32 { %r26, %r27, %r28, %r29 }, [ %rd10 + 0 ]; + // end inline asm + mov.b32 {%rs41, %rs42}, %r26; + mov.b32 {%rs43, %rs44}, %r27; + mov.b32 {%rs45, %rs46}, %r28; + mov.b32 {%rs47, %rs48}, %r29; + // begin inline asm + mov.u32 %r30, %r5; + mov.u32 %r31, %r5; + mov.u32 %r32, %r5; + mov.u32 %r33, %r5; + @%p3 ld.global.v4.b32 { %r30, %r31, %r32, %r33 }, [ %rd11 + 0 ]; + // end inline asm + mov.b32 {%rs49, %rs50}, %r30; + mov.b32 {%rs51, %rs52}, %r31; + mov.b32 {%rs53, %rs54}, %r32; + mov.b32 {%rs55, %rs56}, %r33; + // begin inline asm + mov.u32 %r34, %r5; + mov.u32 %r35, %r5; + mov.u32 %r36, %r5; + mov.u32 %r37, %r5; + @%p4 ld.global.v4.b32 { %r34, %r35, %r36, %r37 }, [ %rd12 + 0 ]; + // end inline asm + mov.b32 {%rs57, %rs58}, %r34; + mov.b32 {%rs59, %rs60}, %r35; + mov.b32 {%rs61, %rs62}, %r36; + mov.b32 {%rs63, %rs64}, %r37; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r94, %rs33; + cvt.f32.bf16 %r95, %rs34; + cvt.f32.bf16 %r96, %rs35; + cvt.f32.bf16 %r97, %rs36; + cvt.f32.bf16 %r98, %rs37; + cvt.f32.bf16 %r99, %rs38; + cvt.f32.bf16 %r100, %rs39; + cvt.f32.bf16 %r101, %rs40; + cvt.f32.bf16 %r102, %rs41; + cvt.f32.bf16 %r103, %rs42; + cvt.f32.bf16 %r104, %rs43; + cvt.f32.bf16 %r105, %rs44; + cvt.f32.bf16 %r106, %rs45; + cvt.f32.bf16 %r107, %rs46; + cvt.f32.bf16 %r108, %rs47; + cvt.f32.bf16 %r109, %rs48; + cvt.f32.bf16 %r110, %rs49; + cvt.f32.bf16 %r111, %rs50; + cvt.f32.bf16 %r112, %rs51; + cvt.f32.bf16 %r113, %rs52; + cvt.f32.bf16 %r114, %rs53; + cvt.f32.bf16 %r115, %rs54; + cvt.f32.bf16 %r116, %rs55; + cvt.f32.bf16 %r117, %rs56; + cvt.f32.bf16 %r118, %rs57; + cvt.f32.bf16 %r119, %rs58; + cvt.f32.bf16 %r120, %rs59; + cvt.f32.bf16 %r121, %rs60; + cvt.f32.bf16 %r122, %rs61; + cvt.f32.bf16 %r123, %rs62; + cvt.f32.bf16 %r124, %rs63; + cvt.f32.bf16 %r125, %rs64; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r126, %r63, %r95; + mul.f32 %r127, %r71, %r103; + mul.f32 %r128, %r79, %r111; + mul.f32 %r129, %r87, %r119; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r130, %rs1, %rs33, %r126; + fma.rn.f32.bf16 %r131, %rs3, %rs35, %r130; + fma.rn.f32.bf16 %r132, %rs4, %rs36, %r131; + fma.rn.f32.bf16 %r133, %rs5, %rs37, %r132; + fma.rn.f32.bf16 %r134, %rs6, %rs38, %r133; + fma.rn.f32.bf16 %r135, %rs7, %rs39, %r134; + fma.rn.f32.bf16 %r136, %rs8, %rs40, %r135; + fma.rn.f32.bf16 %r137, %rs9, %rs41, %r127; + fma.rn.f32.bf16 %r138, %rs11, %rs43, %r137; + fma.rn.f32.bf16 %r139, %rs12, %rs44, %r138; + fma.rn.f32.bf16 %r140, %rs13, %rs45, %r139; + fma.rn.f32.bf16 %r141, %rs14, %rs46, %r140; + fma.rn.f32.bf16 %r142, %rs15, %rs47, %r141; + fma.rn.f32.bf16 %r143, %rs16, %rs48, %r142; + fma.rn.f32.bf16 %r144, %rs17, %rs49, %r128; + fma.rn.f32.bf16 %r145, %rs19, %rs51, %r144; + fma.rn.f32.bf16 %r146, %rs20, %rs52, %r145; + fma.rn.f32.bf16 %r147, %rs21, %rs53, %r146; + fma.rn.f32.bf16 %r148, %rs22, %rs54, %r147; + fma.rn.f32.bf16 %r149, %rs23, %rs55, %r148; + fma.rn.f32.bf16 %r150, %rs24, %rs56, %r149; + fma.rn.f32.bf16 %r151, %rs25, %rs57, %r129; + fma.rn.f32.bf16 %r152, %rs27, %rs59, %r151; + fma.rn.f32.bf16 %r153, %rs28, %rs60, %r152; + fma.rn.f32.bf16 %r154, %rs29, %rs61, %r153; + fma.rn.f32.bf16 %r155, %rs30, %rs62, %r154; + fma.rn.f32.bf16 %r156, %rs31, %rs63, %r155; + fma.rn.f32.bf16 %r157, %rs32, %rs64, %r156; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r158, %r136, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r159, %r136, %r158; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r160, %r159, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r161, %r159, %r160; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r162, %r161, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r163, %r161, %r162; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r164, %r163, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r165, %r163, %r164; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r166, %r143, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r167, %r143, %r166; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r168, %r167, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r169, %r167, %r168; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r170, %r169, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r171, %r169, %r170; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r172, %r171, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r173, %r171, %r172; +$L__tmp18: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r174, %r150, 8, 31, -1; +$L__tmp19: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r175, %r150, %r174; +$L__tmp20: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r176, %r175, 4, 31, -1; +$L__tmp21: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r177, %r175, %r176; +$L__tmp22: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r178, %r177, 2, 31, -1; +$L__tmp23: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r179, %r177, %r178; +$L__tmp24: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r180, %r179, 1, 31, -1; +$L__tmp25: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r181, %r179, %r180; +$L__tmp26: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r182, %r157, 8, 31, -1; +$L__tmp27: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r183, %r157, %r182; +$L__tmp28: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r184, %r183, 4, 31, -1; +$L__tmp29: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r185, %r183, %r184; +$L__tmp30: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r186, %r185, 2, 31, -1; +$L__tmp31: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r187, %r185, %r186; +$L__tmp32: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r188, %r187, 1, 31, -1; +$L__tmp33: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r189, %r187, %r188; +$L__tmp34: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r190, %r165, %r62; + mul.f32 %r191, %r165, %r63; + mul.f32 %r192, %r165, %r64; + mul.f32 %r193, %r165, %r65; + mul.f32 %r194, %r165, %r66; + mul.f32 %r195, %r165, %r67; + mul.f32 %r196, %r165, %r68; + mul.f32 %r197, %r165, %r69; + mul.f32 %r198, %r173, %r70; + mul.f32 %r199, %r173, %r71; + mul.f32 %r200, %r173, %r72; + mul.f32 %r201, %r173, %r73; + mul.f32 %r202, %r173, %r74; + mul.f32 %r203, %r173, %r75; + mul.f32 %r204, %r173, %r76; + mul.f32 %r205, %r173, %r77; + mul.f32 %r206, %r181, %r78; + mul.f32 %r207, %r181, %r79; + mul.f32 %r208, %r181, %r80; + mul.f32 %r209, %r181, %r81; + mul.f32 %r210, %r181, %r82; + mul.f32 %r211, %r181, %r83; + mul.f32 %r212, %r181, %r84; + mul.f32 %r213, %r181, %r85; + mul.f32 %r214, %r189, %r86; + mul.f32 %r215, %r189, %r87; + mul.f32 %r216, %r189, %r88; + mul.f32 %r217, %r189, %r89; + mul.f32 %r218, %r189, %r90; + mul.f32 %r219, %r189, %r91; + mul.f32 %r220, %r189, %r92; + mul.f32 %r221, %r189, %r93; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r222, %r190, %r18; + mul.f32 %r223, %r191, %r18; + mul.f32 %r224, %r192, %r18; + mul.f32 %r225, %r193, %r18; + mul.f32 %r226, %r194, %r18; + mul.f32 %r227, %r195, %r18; + mul.f32 %r228, %r196, %r18; + mul.f32 %r229, %r197, %r18; + mul.f32 %r230, %r198, %r19; + mul.f32 %r231, %r199, %r19; + mul.f32 %r232, %r200, %r19; + mul.f32 %r233, %r201, %r19; + mul.f32 %r234, %r202, %r19; + mul.f32 %r235, %r203, %r19; + mul.f32 %r236, %r204, %r19; + mul.f32 %r237, %r205, %r19; + mul.f32 %r238, %r206, %r20; + mul.f32 %r239, %r207, %r20; + mul.f32 %r240, %r208, %r20; + mul.f32 %r241, %r209, %r20; + mul.f32 %r242, %r210, %r20; + mul.f32 %r243, %r211, %r20; + mul.f32 %r244, %r212, %r20; + mul.f32 %r245, %r213, %r20; + mul.f32 %r246, %r214, %r21; + mul.f32 %r247, %r215, %r21; + mul.f32 %r248, %r216, %r21; + mul.f32 %r249, %r217, %r21; + mul.f32 %r250, %r218, %r21; + mul.f32 %r251, %r219, %r21; + mul.f32 %r252, %r220, %r21; + mul.f32 %r253, %r221, %r21; + .loc 1 107 36 // l2norm.py:107:36 + neg.f32 %r254, %r222; + fma.rn.f32 %r255, %r18, %r94, %r254; + neg.f32 %r256, %r223; + fma.rn.f32 %r257, %r18, %r95, %r256; + neg.f32 %r258, %r224; + fma.rn.f32 %r259, %r18, %r96, %r258; + neg.f32 %r260, %r225; + fma.rn.f32 %r261, %r18, %r97, %r260; + neg.f32 %r262, %r226; + fma.rn.f32 %r263, %r18, %r98, %r262; + neg.f32 %r264, %r227; + fma.rn.f32 %r265, %r18, %r99, %r264; + neg.f32 %r266, %r228; + fma.rn.f32 %r267, %r18, %r100, %r266; + neg.f32 %r268, %r229; + fma.rn.f32 %r269, %r18, %r101, %r268; + neg.f32 %r270, %r230; + fma.rn.f32 %r271, %r19, %r102, %r270; + neg.f32 %r272, %r231; + fma.rn.f32 %r273, %r19, %r103, %r272; + neg.f32 %r274, %r232; + fma.rn.f32 %r275, %r19, %r104, %r274; + neg.f32 %r276, %r233; + fma.rn.f32 %r277, %r19, %r105, %r276; + neg.f32 %r278, %r234; + fma.rn.f32 %r279, %r19, %r106, %r278; + neg.f32 %r280, %r235; + fma.rn.f32 %r281, %r19, %r107, %r280; + neg.f32 %r282, %r236; + fma.rn.f32 %r283, %r19, %r108, %r282; + neg.f32 %r284, %r237; + fma.rn.f32 %r285, %r19, %r109, %r284; + neg.f32 %r286, %r238; + fma.rn.f32 %r287, %r20, %r110, %r286; + neg.f32 %r288, %r239; + fma.rn.f32 %r289, %r20, %r111, %r288; + neg.f32 %r290, %r240; + fma.rn.f32 %r291, %r20, %r112, %r290; + neg.f32 %r292, %r241; + fma.rn.f32 %r293, %r20, %r113, %r292; + neg.f32 %r294, %r242; + fma.rn.f32 %r295, %r20, %r114, %r294; + neg.f32 %r296, %r243; + fma.rn.f32 %r297, %r20, %r115, %r296; + neg.f32 %r298, %r244; + fma.rn.f32 %r299, %r20, %r116, %r298; + neg.f32 %r300, %r245; + fma.rn.f32 %r301, %r20, %r117, %r300; + neg.f32 %r302, %r246; + fma.rn.f32 %r303, %r21, %r118, %r302; + neg.f32 %r304, %r247; + fma.rn.f32 %r305, %r21, %r119, %r304; + neg.f32 %r306, %r248; + fma.rn.f32 %r307, %r21, %r120, %r306; + neg.f32 %r308, %r249; + fma.rn.f32 %r309, %r21, %r121, %r308; + neg.f32 %r310, %r250; + fma.rn.f32 %r311, %r21, %r122, %r310; + neg.f32 %r312, %r251; + fma.rn.f32 %r313, %r21, %r123, %r312; + neg.f32 %r314, %r252; + fma.rn.f32 %r315, %r21, %r124, %r314; + neg.f32 %r316, %r253; + fma.rn.f32 %r317, %r21, %r125, %r316; + .loc 1 108 19 // l2norm.py:108:19 + cvt.rn.bf16x2.f32 %r38, %r257, %r255; + cvt.rn.bf16x2.f32 %r39, %r261, %r259; + cvt.rn.bf16x2.f32 %r40, %r265, %r263; + cvt.rn.bf16x2.f32 %r41, %r269, %r267; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd13 + 0 ], { %r38, %r39, %r40, %r41 }; + // end inline asm + cvt.rn.bf16x2.f32 %r42, %r273, %r271; + cvt.rn.bf16x2.f32 %r43, %r277, %r275; + cvt.rn.bf16x2.f32 %r44, %r281, %r279; + cvt.rn.bf16x2.f32 %r45, %r285, %r283; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd14 + 0 ], { %r42, %r43, %r44, %r45 }; + // end inline asm + cvt.rn.bf16x2.f32 %r46, %r289, %r287; + cvt.rn.bf16x2.f32 %r47, %r293, %r291; + cvt.rn.bf16x2.f32 %r48, %r297, %r295; + cvt.rn.bf16x2.f32 %r49, %r301, %r299; + // begin inline asm + @%p3 st.global.v4.b32 [ %rd15 + 0 ], { %r46, %r47, %r48, %r49 }; + // end inline asm + cvt.rn.bf16x2.f32 %r50, %r305, %r303; + cvt.rn.bf16x2.f32 %r51, %r309, %r307; + cvt.rn.bf16x2.f32 %r52, %r313, %r311; + cvt.rn.bf16x2.f32 %r53, %r317, %r315; + // begin inline asm + @%p4 st.global.v4.b32 [ %rd16 + 0 ], { %r50, %r51, %r52, %r53 }; + // end inline asm + .loc 1 108 4 // l2norm.py:108:4 + ret; +$L__tmp35: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 98 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp34 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 107 // DW_AT_call_line +.b8 55 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp34 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/l2norm_bwd_kernel.source b/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/l2norm_bwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..f03b2b6a6052abdb7529f301b1a7db4396dd4fee --- /dev/null +++ b/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/l2norm_bwd_kernel.source @@ -0,0 +1,210 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc47 = loc(unknown) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc54 = loc("y"(#loc)) +#loc55 = loc("rstd"(#loc)) +#loc56 = loc("dy"(#loc)) +#loc57 = loc("dx"(#loc)) +#loc58 = loc("eps"(#loc)) +#loc59 = loc("T"(#loc)) +#loc101 = loc("input"(#loc45)) +#loc102 = loc("a"(#loc50)) +#loc103 = loc("b"(#loc50)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc60) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc61) + %o_t = arith.constant 32 : i32 loc(#loc62) + %o_t_1 = arith.constant 32 : i64 loc(#loc62) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc62) + %o_t_3 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32> loc(#loc63) + %o_t_4 = arith.extsi %o_t_3 : tensor<32xi32> to tensor<32xi64> loc(#loc64) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<32xi64> loc(#loc64) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<32xi64> loc(#loc64) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc65) + %m_t = arith.extsi %T : i32 to i64 loc(#loc66) + %m_t_7 = tt.splat %m_t : i64 -> tensor<32xi64> loc(#loc66) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<32xi64> loc(#loc66) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<32xi1> -> tensor<32x1xi1> loc(#loc67) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc68) + %m_x_10 = arith.constant 128 : i32 loc(#loc69) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc69) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc69) + %m_x_13 = tt.broadcast %m_x : tensor<32x1xi1> -> tensor<32x128xi1> loc(#loc70) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<32x128xi1> loc(#loc70) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<32x128xi1> loc(#loc70) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<32xi64> -> tensor<32x1xi64> loc(#loc71) + %p_y_16 = arith.constant 128 : i32 loc(#loc72) + %p_y_17 = arith.constant 128 : i64 loc(#loc72) + %p_y_18 = arith.constant dense<128> : tensor<32x1xi64> loc(#loc72) + %p_y_19 = arith.muli %p_y, %p_y_18 : tensor<32x1xi64> loc(#loc72) + %p_y_20 = tt.splat %y : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc73) + %p_y_21 = tt.addptr %p_y_20, %p_y_19 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc73) + %p_y_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc74) + %p_y_23 = tt.broadcast %p_y_21 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc75) + %p_y_24 = tt.broadcast %p_y_22 : tensor<1x128xi32> -> tensor<32x128xi32> loc(#loc75) + %p_y_25 = tt.addptr %p_y_23, %p_y_24 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc75) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<32x!tt.ptr> loc(#loc76) + %p_rstd_26 = tt.addptr %p_rstd, %o_t_6 : tensor<32x!tt.ptr>, tensor<32xi64> loc(#loc76) + %p_dy = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<32xi64> -> tensor<32x1xi64> loc(#loc77) + %p_dy_27 = arith.constant 128 : i32 loc(#loc78) + %p_dy_28 = arith.constant 128 : i64 loc(#loc78) + %p_dy_29 = arith.constant dense<128> : tensor<32x1xi64> loc(#loc78) + %p_dy_30 = arith.muli %p_dy, %p_dy_29 : tensor<32x1xi64> loc(#loc78) + %p_dy_31 = tt.splat %dy : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc79) + %p_dy_32 = tt.addptr %p_dy_31, %p_dy_30 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc79) + %p_dy_33 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc80) + %p_dy_34 = tt.broadcast %p_dy_32 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc81) + %p_dy_35 = tt.broadcast %p_dy_33 : tensor<1x128xi32> -> tensor<32x128xi32> loc(#loc81) + %p_dy_36 = tt.addptr %p_dy_34, %p_dy_35 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc81) + %p_dx = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<32xi64> -> tensor<32x1xi64> loc(#loc82) + %p_dx_37 = arith.constant 128 : i32 loc(#loc83) + %p_dx_38 = arith.constant 128 : i64 loc(#loc83) + %p_dx_39 = arith.constant dense<128> : tensor<32x1xi64> loc(#loc83) + %p_dx_40 = arith.muli %p_dx, %p_dx_39 : tensor<32x1xi64> loc(#loc83) + %p_dx_41 = tt.splat %dx : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc84) + %p_dx_42 = tt.addptr %p_dx_41, %p_dx_40 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc84) + %p_dx_43 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc85) + %p_dx_44 = tt.broadcast %p_dx_42 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc86) + %p_dx_45 = tt.broadcast %p_dx_43 : tensor<1x128xi32> -> tensor<32x128xi32> loc(#loc86) + %p_dx_46 = tt.addptr %p_dx_44, %p_dx_45 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc86) + %b_y = arith.constant 0.000000e+00 : f32 loc(#loc87) + %b_y_47 = arith.constant dense<0.000000e+00> : tensor<32x128xf32> loc(#loc87) + %b_y_48 = arith.truncf %b_y_47 : tensor<32x128xf32> to tensor<32x128xbf16> loc(#loc87) + %b_y_49 = tt.load %p_y_25, %m_x_15, %b_y_48 : tensor<32x128x!tt.ptr> loc(#loc87) + %b_y_50 = arith.extf %b_y_49 : tensor<32x128xbf16> to tensor<32x128xf32> loc(#loc88) + %b_rstd = arith.constant 0.000000e+00 : f32 loc(#loc89) + %b_rstd_51 = arith.constant dense<0.000000e+00> : tensor<32xf32> loc(#loc89) + %b_rstd_52 = tt.load %p_rstd_26, %m_t_8, %b_rstd_51 : tensor<32x!tt.ptr> loc(#loc89) + %b_dy = arith.constant 0.000000e+00 : f32 loc(#loc90) + %b_dy_53 = arith.constant dense<0.000000e+00> : tensor<32x128xf32> loc(#loc90) + %b_dy_54 = arith.truncf %b_dy_53 : tensor<32x128xf32> to tensor<32x128xbf16> loc(#loc90) + %b_dy_55 = tt.load %p_dy_36, %m_x_15, %b_dy_54 : tensor<32x128x!tt.ptr> loc(#loc90) + %b_dy_56 = arith.extf %b_dy_55 : tensor<32x128xbf16> to tensor<32x128xf32> loc(#loc91) + %b_dx = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<32xf32> -> tensor<32x1xf32> loc(#loc92) + %b_dx_57 = tt.broadcast %b_dx : tensor<32x1xf32> -> tensor<32x128xf32> loc(#loc93) + %b_dx_58 = arith.mulf %b_dy_56, %b_dx_57 : tensor<32x128xf32> loc(#loc93) + %b_dx_59 = arith.mulf %b_dy_56, %b_y_50 : tensor<32x128xf32> loc(#loc94) + %b_dx_60 = tt.call @triton.language.standard.sum__fp32S32_128S_c1_cFalse_cNone(%b_dx_59) : (tensor<32x128xf32>) -> tensor<32xf32> loc(#loc95) + %b_dx_61 = tt.expand_dims %b_dx_60 {axis = 1 : i32} : tensor<32xf32> -> tensor<32x1xf32> loc(#loc96) + %b_dx_62 = tt.broadcast %b_dx_61 : tensor<32x1xf32> -> tensor<32x128xf32> loc(#loc97) + %b_dx_63 = arith.mulf %b_dx_62, %b_y_50 : tensor<32x128xf32> loc(#loc97) + %b_dx_64 = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<32xf32> -> tensor<32x1xf32> loc(#loc98) + %b_dx_65 = tt.broadcast %b_dx_64 : tensor<32x1xf32> -> tensor<32x128xf32> loc(#loc99) + %b_dx_66 = arith.mulf %b_dx_63, %b_dx_65 : tensor<32x128xf32> loc(#loc99) + %b_dx_67 = arith.subf %b_dx_58, %b_dx_66 : tensor<32x128xf32> loc(#loc100) + %0 = arith.truncf %b_dx_67 : tensor<32x128xf32> to tensor<32x128xbf16> loc(#loc42) + tt.store %p_dx_46, %0, %m_x_15 : tensor<32x128x!tt.ptr> loc(#loc43) + tt.return loc(#loc44) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S32_128S_c1_cFalse_cNone(%input: tensor<32x128xf32> loc("input"(#loc45))) -> tensor<32xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc46) + tt.reduce.return %2 : f32 loc(#loc46) + }) : (tensor<32x128xf32>) -> tensor<32xf32> loc(#loc46) + tt.return %0 : tensor<32xf32> loc(#loc48) + ^bb1: // no predecessors + %1 = ub.poison : tensor<32xf32> loc(#loc49) + tt.return %1 : tensor<32xf32> loc(#loc49) + } loc(#loc45) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc50)), %b: f32 loc("b"(#loc50))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc51) + tt.return %0 : f32 loc(#loc52) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc53) + tt.return %1 : f32 loc(#loc53) + } loc(#loc50) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:31) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:39) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:20) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:31) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:39) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:82) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc60 = loc("i_t"(#loc1)) +#loc61 = loc("i_t"(#loc2)) +#loc62 = loc("o_t"(#loc3)) +#loc63 = loc("o_t"(#loc4)) +#loc64 = loc("o_t"(#loc5)) +#loc65 = loc("o_d"(#loc6)) +#loc66 = loc("m_t"(#loc7)) +#loc67 = loc("m_x"(#loc8)) +#loc68 = loc("m_x"(#loc9)) +#loc69 = loc("m_x"(#loc10)) +#loc70 = loc("m_x"(#loc11)) +#loc71 = loc("p_y"(#loc12)) +#loc72 = loc("p_y"(#loc13)) +#loc73 = loc("p_y"(#loc14)) +#loc74 = loc("p_y"(#loc15)) +#loc75 = loc("p_y"(#loc16)) +#loc76 = loc("p_rstd"(#loc17)) +#loc77 = loc("p_dy"(#loc18)) +#loc78 = loc("p_dy"(#loc19)) +#loc79 = loc("p_dy"(#loc20)) +#loc80 = loc("p_dy"(#loc21)) +#loc81 = loc("p_dy"(#loc22)) +#loc82 = loc("p_dx"(#loc23)) +#loc83 = loc("p_dx"(#loc24)) +#loc84 = loc("p_dx"(#loc25)) +#loc85 = loc("p_dx"(#loc26)) +#loc86 = loc("p_dx"(#loc27)) +#loc87 = loc("b_y"(#loc28)) +#loc88 = loc("b_y"(#loc29)) +#loc89 = loc("b_rstd"(#loc30)) +#loc90 = loc("b_dy"(#loc31)) +#loc91 = loc("b_dy"(#loc32)) +#loc92 = loc("b_dx"(#loc33)) +#loc93 = loc("b_dx"(#loc34)) +#loc94 = loc("b_dx"(#loc35)) +#loc95 = loc("b_dx"(#loc36)) +#loc96 = loc("b_dx"(#loc37)) +#loc97 = loc("b_dx"(#loc38)) +#loc98 = loc("b_dx"(#loc39)) +#loc99 = loc("b_dx"(#loc40)) +#loc100 = loc("b_dx"(#loc41)) diff --git a/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/l2norm_bwd_kernel.ttgir b/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/l2norm_bwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..bd9fdf1c908741d42eddac6a284a3ff59de3c32e --- /dev/null +++ b/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/l2norm_bwd_kernel.ttgir @@ -0,0 +1,146 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [4, 1], order = [1, 0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc39 = loc("y"(#loc)) +#loc40 = loc("rstd"(#loc)) +#loc41 = loc("dy"(#loc)) +#loc42 = loc("dx"(#loc)) +#loc43 = loc("eps"(#loc)) +#loc44 = loc("T"(#loc)) +#loc72 = loc("b_dx"(#loc30)) +#loc78 = loc(callsite(#loc1 at #loc72)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 4 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %c32_i64 = arith.constant 32 : i64 loc(#loc1) + %cst = arith.constant dense<0.000000e+00> : tensor<32x128xbf16, #blocked> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<32x1xi64, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc45) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc46) + %o_t = arith.muli %i_t_3, %c32_i64 : i64 loc(#loc47) + %o_t_4 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc48) + %o_t_5 = arith.extsi %o_t_4 : tensor<32xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %o_t_6 = tt.splat %o_t : i64 -> tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %o_t_7 = arith.addi %o_t_6, %o_t_5 : tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %m_t = arith.extsi %T : i32 to i64 loc(#loc50) + %m_t_8 = tt.splat %m_t : i64 -> tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc50) + %m_t_9 = arith.cmpi slt, %o_t_7, %m_t_8 : tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc50) + %m_x = tt.expand_dims %m_t_9 {axis = 1 : i32} : tensor<32xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32x1xi1, #blocked> loc(#loc51) + %m_x_10 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc52) + %m_x_11 = tt.expand_dims %m_x_10 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc52) + %m_x_12 = arith.cmpi slt, %m_x_11, %cst_2 : tensor<1x128xi32, #blocked> loc(#loc53) + %m_x_13 = tt.broadcast %m_x : tensor<32x1xi1, #blocked> -> tensor<32x128xi1, #blocked> loc(#loc54) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1, #blocked> -> tensor<32x128xi1, #blocked> loc(#loc54) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<32x128xi1, #blocked> loc(#loc54) + %p_y = tt.expand_dims %o_t_7 {axis = 1 : i32} : tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32x1xi64, #blocked> loc(#loc55) + %p_y_16 = arith.muli %p_y, %cst_1 : tensor<32x1xi64, #blocked> loc(#loc56) + %p_y_17 = tt.splat %y : !tt.ptr -> tensor<32x1x!tt.ptr, #blocked> loc(#loc57) + %p_y_18 = tt.addptr %p_y_17, %p_y_16 : tensor<32x1x!tt.ptr, #blocked>, tensor<32x1xi64, #blocked> loc(#loc57) + %p_y_19 = tt.broadcast %p_y_18 : tensor<32x1x!tt.ptr, #blocked> -> tensor<32x128x!tt.ptr, #blocked> loc(#loc58) + %p_y_20 = tt.broadcast %m_x_11 : tensor<1x128xi32, #blocked> -> tensor<32x128xi32, #blocked> loc(#loc58) + %p_y_21 = tt.addptr %p_y_19, %p_y_20 : tensor<32x128x!tt.ptr, #blocked>, tensor<32x128xi32, #blocked> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<32x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc59) + %p_rstd_22 = tt.addptr %p_rstd, %o_t_7 : tensor<32x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>>, tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc59) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<32x1x!tt.ptr, #blocked> loc(#loc60) + %p_dy_23 = tt.addptr %p_dy, %p_y_16 : tensor<32x1x!tt.ptr, #blocked>, tensor<32x1xi64, #blocked> loc(#loc60) + %p_dy_24 = tt.broadcast %p_dy_23 : tensor<32x1x!tt.ptr, #blocked> -> tensor<32x128x!tt.ptr, #blocked> loc(#loc61) + %p_dy_25 = tt.addptr %p_dy_24, %p_y_20 : tensor<32x128x!tt.ptr, #blocked>, tensor<32x128xi32, #blocked> loc(#loc61) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<32x1x!tt.ptr, #blocked> loc(#loc62) + %p_dx_26 = tt.addptr %p_dx, %p_y_16 : tensor<32x1x!tt.ptr, #blocked>, tensor<32x1xi64, #blocked> loc(#loc62) + %p_dx_27 = tt.broadcast %p_dx_26 : tensor<32x1x!tt.ptr, #blocked> -> tensor<32x128x!tt.ptr, #blocked> loc(#loc63) + %p_dx_28 = tt.addptr %p_dx_27, %p_y_20 : tensor<32x128x!tt.ptr, #blocked>, tensor<32x128xi32, #blocked> loc(#loc63) + %b_y = tt.load %p_y_21, %m_x_15, %cst : tensor<32x128x!tt.ptr, #blocked> loc(#loc64) + %b_y_29 = arith.extf %b_y : tensor<32x128xbf16, #blocked> to tensor<32x128xf32, #blocked> loc(#loc65) + %b_rstd = tt.load %p_rstd_22, %m_t_9, %cst_0 : tensor<32x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc66) + %b_dy = tt.load %p_dy_25, %m_x_15, %cst : tensor<32x128x!tt.ptr, #blocked> loc(#loc67) + %b_dy_30 = arith.extf %b_dy : tensor<32x128xbf16, #blocked> to tensor<32x128xf32, #blocked> loc(#loc68) + %b_dx = tt.expand_dims %b_rstd {axis = 1 : i32} : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32x1xf32, #blocked> loc(#loc69) + %b_dx_31 = tt.broadcast %b_dx : tensor<32x1xf32, #blocked> -> tensor<32x128xf32, #blocked> loc(#loc70) + %b_dx_32 = arith.mulf %b_dy_30, %b_dx_31 : tensor<32x128xf32, #blocked> loc(#loc70) + %b_dx_33 = arith.mulf %b_dy_30, %b_y_29 : tensor<32x128xf32, #blocked> loc(#loc71) + %b_dx_34 = "tt.reduce"(%b_dx_33) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_40: f32 loc(callsite(#loc1 at #loc72)), %b_dx_41: f32 loc(callsite(#loc1 at #loc72))): + %b_dx_42 = arith.addf %b_dx_40, %b_dx_41 : f32 loc(#loc79) + tt.reduce.return %b_dx_42 : f32 loc(#loc77) + }) : (tensor<32x128xf32, #blocked>) -> tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc77) + %b_dx_35 = tt.expand_dims %b_dx_34 {axis = 1 : i32} : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32x1xf32, #blocked> loc(#loc73) + %b_dx_36 = tt.broadcast %b_dx_35 : tensor<32x1xf32, #blocked> -> tensor<32x128xf32, #blocked> loc(#loc74) + %b_dx_37 = arith.mulf %b_dx_36, %b_y_29 : tensor<32x128xf32, #blocked> loc(#loc74) + %b_dx_38 = arith.mulf %b_dx_37, %b_dx_31 : tensor<32x128xf32, #blocked> loc(#loc75) + %b_dx_39 = arith.subf %b_dx_32, %b_dx_38 : tensor<32x128xf32, #blocked> loc(#loc76) + %0 = arith.truncf %b_dx_39 : tensor<32x128xf32, #blocked> to tensor<32x128xbf16, #blocked> loc(#loc36) + tt.store %p_dx_28, %0, %m_x_15 : tensor<32x128x!tt.ptr, #blocked> loc(#loc37) + tt.return loc(#loc38) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc45 = loc("i_t"(#loc2)) +#loc46 = loc("i_t"(#loc3)) +#loc47 = loc("o_t"(#loc4)) +#loc48 = loc("o_t"(#loc5)) +#loc49 = loc("o_t"(#loc6)) +#loc50 = loc("m_t"(#loc7)) +#loc51 = loc("m_x"(#loc8)) +#loc52 = loc("m_x"(#loc9)) +#loc53 = loc("m_x"(#loc10)) +#loc54 = loc("m_x"(#loc11)) +#loc55 = loc("p_y"(#loc12)) +#loc56 = loc("p_y"(#loc13)) +#loc57 = loc("p_y"(#loc14)) +#loc58 = loc("p_y"(#loc15)) +#loc59 = loc("p_rstd"(#loc16)) +#loc60 = loc("p_dy"(#loc17)) +#loc61 = loc("p_dy"(#loc18)) +#loc62 = loc("p_dx"(#loc19)) +#loc63 = loc("p_dx"(#loc20)) +#loc64 = loc("b_y"(#loc21)) +#loc65 = loc("b_y"(#loc22)) +#loc66 = loc("b_rstd"(#loc23)) +#loc67 = loc("b_dy"(#loc24)) +#loc68 = loc("b_dy"(#loc25)) +#loc69 = loc("b_dx"(#loc26)) +#loc70 = loc("b_dx"(#loc27)) +#loc71 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc32)) +#loc74 = loc("b_dx"(#loc33)) +#loc75 = loc("b_dx"(#loc34)) +#loc76 = loc("b_dx"(#loc35)) +#loc77 = loc(callsite(#loc29 at #loc72)) +#loc79 = loc(callsite(#loc31 at #loc77)) diff --git a/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/l2norm_bwd_kernel.ttir b/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/l2norm_bwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..63d8cf26fdf711c6296bb4ba74925d0b8a9334c0 --- /dev/null +++ b/triton/gpu0/AZFQGGZOI5KOJYBGZVDFZZG5DULUZ2ES6NBASXG45Q5G3I2GLAHQ/l2norm_bwd_kernel.ttir @@ -0,0 +1,147 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc40 = loc("y"(#loc)) +#loc41 = loc("rstd"(#loc)) +#loc42 = loc("dy"(#loc)) +#loc43 = loc("dx"(#loc)) +#loc44 = loc("eps"(#loc)) +#loc45 = loc("T"(#loc)) +#loc74 = loc("b_dx"(#loc31)) +#loc80 = loc(callsite(#loc1 at #loc74)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<0.000000e+00> : tensor<32x128xbf16> loc(#loc1) + %b_rstd = arith.constant dense<0.000000e+00> : tensor<32xf32> loc(#loc46) + %cst_0 = arith.constant dense<128> : tensor<32x1xi64> loc(#loc1) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc47) + %o_t = arith.constant 32 : i64 loc(#loc48) + %i_t = tt.get_program_id x : i32 loc(#loc49) + %i_t_1 = arith.extsi %i_t : i32 to i64 loc(#loc50) + %o_t_2 = arith.muli %i_t_1, %o_t : i64 loc(#loc48) + %o_t_3 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32> loc(#loc51) + %o_t_4 = arith.extsi %o_t_3 : tensor<32xi32> to tensor<32xi64> loc(#loc52) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<32xi64> loc(#loc52) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<32xi64> loc(#loc52) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc53) + %m_t = arith.extsi %T : i32 to i64 loc(#loc54) + %m_t_7 = tt.splat %m_t : i64 -> tensor<32xi64> loc(#loc54) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<32xi64> loc(#loc54) + %m_x_9 = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<32xi1> -> tensor<32x1xi1> loc(#loc55) + %m_x_10 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc56) + %m_x_11 = arith.cmpi slt, %m_x_10, %m_x : tensor<1x128xi32> loc(#loc47) + %m_x_12 = tt.broadcast %m_x_9 : tensor<32x1xi1> -> tensor<32x128xi1> loc(#loc57) + %m_x_13 = tt.broadcast %m_x_11 : tensor<1x128xi1> -> tensor<32x128xi1> loc(#loc57) + %m_x_14 = arith.andi %m_x_12, %m_x_13 : tensor<32x128xi1> loc(#loc57) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<32xi64> -> tensor<32x1xi64> loc(#loc58) + %p_y_15 = arith.muli %p_y, %cst_0 : tensor<32x1xi64> loc(#loc59) + %p_y_16 = tt.splat %y : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc60) + %p_y_17 = tt.addptr %p_y_16, %p_y_15 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc60) + %p_y_18 = tt.broadcast %p_y_17 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc61) + %p_y_19 = tt.broadcast %m_x_10 : tensor<1x128xi32> -> tensor<32x128xi32> loc(#loc61) + %p_y_20 = tt.addptr %p_y_18, %p_y_19 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc61) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<32x!tt.ptr> loc(#loc62) + %p_rstd_21 = tt.addptr %p_rstd, %o_t_6 : tensor<32x!tt.ptr>, tensor<32xi64> loc(#loc62) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc63) + %p_dy_22 = tt.addptr %p_dy, %p_y_15 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc63) + %p_dy_23 = tt.broadcast %p_dy_22 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc64) + %p_dy_24 = tt.addptr %p_dy_23, %p_y_19 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc64) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc65) + %p_dx_25 = tt.addptr %p_dx, %p_y_15 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc65) + %p_dx_26 = tt.broadcast %p_dx_25 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc66) + %p_dx_27 = tt.addptr %p_dx_26, %p_y_19 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc66) + %b_y = tt.load %p_y_20, %m_x_14, %cst : tensor<32x128x!tt.ptr> loc(#loc67) + %b_y_28 = arith.extf %b_y : tensor<32x128xbf16> to tensor<32x128xf32> loc(#loc68) + %b_rstd_29 = tt.load %p_rstd_21, %m_t_8, %b_rstd : tensor<32x!tt.ptr> loc(#loc46) + %b_dy = tt.load %p_dy_24, %m_x_14, %cst : tensor<32x128x!tt.ptr> loc(#loc69) + %b_dy_30 = arith.extf %b_dy : tensor<32x128xbf16> to tensor<32x128xf32> loc(#loc70) + %b_dx = tt.expand_dims %b_rstd_29 {axis = 1 : i32} : tensor<32xf32> -> tensor<32x1xf32> loc(#loc71) + %b_dx_31 = tt.broadcast %b_dx : tensor<32x1xf32> -> tensor<32x128xf32> loc(#loc72) + %b_dx_32 = arith.mulf %b_dy_30, %b_dx_31 : tensor<32x128xf32> loc(#loc72) + %b_dx_33 = arith.mulf %b_dy_30, %b_y_28 : tensor<32x128xf32> loc(#loc73) + %b_dx_34 = "tt.reduce"(%b_dx_33) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_40: f32 loc(callsite(#loc1 at #loc74)), %b_dx_41: f32 loc(callsite(#loc1 at #loc74))): + %b_dx_42 = arith.addf %b_dx_40, %b_dx_41 : f32 loc(#loc81) + tt.reduce.return %b_dx_42 : f32 loc(#loc79) + }) : (tensor<32x128xf32>) -> tensor<32xf32> loc(#loc79) + %b_dx_35 = tt.expand_dims %b_dx_34 {axis = 1 : i32} : tensor<32xf32> -> tensor<32x1xf32> loc(#loc75) + %b_dx_36 = tt.broadcast %b_dx_35 : tensor<32x1xf32> -> tensor<32x128xf32> loc(#loc76) + %b_dx_37 = arith.mulf %b_dx_36, %b_y_28 : tensor<32x128xf32> loc(#loc76) + %b_dx_38 = arith.mulf %b_dx_37, %b_dx_31 : tensor<32x128xf32> loc(#loc77) + %b_dx_39 = arith.subf %b_dx_32, %b_dx_38 : tensor<32x128xf32> loc(#loc78) + %0 = arith.truncf %b_dx_39 : tensor<32x128xf32> to tensor<32x128xbf16> loc(#loc37) + tt.store %p_dx_27, %0, %m_x_14 : tensor<32x128x!tt.ptr> loc(#loc38) + tt.return loc(#loc39) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("b_rstd"(#loc2)) +#loc47 = loc("m_x"(#loc3)) +#loc48 = loc("o_t"(#loc4)) +#loc49 = loc("i_t"(#loc5)) +#loc50 = loc("i_t"(#loc6)) +#loc51 = loc("o_t"(#loc7)) +#loc52 = loc("o_t"(#loc8)) +#loc53 = loc("o_d"(#loc9)) +#loc54 = loc("m_t"(#loc10)) +#loc55 = loc("m_x"(#loc11)) +#loc56 = loc("m_x"(#loc12)) +#loc57 = loc("m_x"(#loc13)) +#loc58 = loc("p_y"(#loc14)) +#loc59 = loc("p_y"(#loc15)) +#loc60 = loc("p_y"(#loc16)) +#loc61 = loc("p_y"(#loc17)) +#loc62 = loc("p_rstd"(#loc18)) +#loc63 = loc("p_dy"(#loc19)) +#loc64 = loc("p_dy"(#loc20)) +#loc65 = loc("p_dx"(#loc21)) +#loc66 = loc("p_dx"(#loc22)) +#loc67 = loc("b_y"(#loc23)) +#loc68 = loc("b_y"(#loc24)) +#loc69 = loc("b_dy"(#loc25)) +#loc70 = loc("b_dy"(#loc26)) +#loc71 = loc("b_dx"(#loc27)) +#loc72 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc29)) +#loc75 = loc("b_dx"(#loc33)) +#loc76 = loc("b_dx"(#loc34)) +#loc77 = loc("b_dx"(#loc35)) +#loc78 = loc("b_dx"(#loc36)) +#loc79 = loc(callsite(#loc30 at #loc74)) +#loc81 = loc(callsite(#loc32 at #loc79)) diff --git a/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/__grp__l2norm_bwd_kernel.json b/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/__grp__l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..242849d4323ff789a32fdb281dd6fca1668f0e8f --- /dev/null +++ b/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/__grp__l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_bwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/l2norm_bwd_kernel.source", "l2norm_bwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/l2norm_bwd_kernel.ttir", "l2norm_bwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/l2norm_bwd_kernel.ttgir", "l2norm_bwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/l2norm_bwd_kernel.llir", "l2norm_bwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/l2norm_bwd_kernel.ptx", "l2norm_bwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/l2norm_bwd_kernel.cubin", "l2norm_bwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/l2norm_bwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/l2norm_bwd_kernel.cubin b/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/l2norm_bwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..0282fa55e2db0256827cf0a6ad23ecafd8ef4573 Binary files /dev/null and b/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/l2norm_bwd_kernel.cubin differ diff --git a/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/l2norm_bwd_kernel.json b/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..04b5f4ac911c82a1a13b9a516b342978287b1c64 --- /dev/null +++ b/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"hash": "0e8894bea2f9ed905567e24913dc0bd683215e7dbbe7bd76cb8b6ba8a37c4fbc", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 16, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 64, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_bwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/l2norm_bwd_kernel.llir b/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/l2norm_bwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..11ad8ac277d542803fc872f0c93b056da9781c7f --- /dev/null +++ b/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/l2norm_bwd_kernel.llir @@ -0,0 +1,175 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_bwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, ptr addrspace(1) %3, float %4, i32 %5, ptr addrspace(1) readnone captures(none) %6, ptr addrspace(1) readnone captures(none) %7) local_unnamed_addr #0 !dbg !4 { + %9 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !12 + %10 = zext nneg i32 %9 to i64, !dbg !13 + %11 = shl nuw nsw i64 %10, 3, !dbg !14 + %12 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !15 + %13 = and i32 %12, 511, !dbg !15 + %14 = and i32 %12, 31, !dbg !15 + %15 = lshr i32 %12, 5, !dbg !15 + %16 = lshr i32 %12, 6, !dbg !15 + %17 = and i32 %16, 7, !dbg !15 + %18 = zext nneg i32 %17 to i64, !dbg !16 + %19 = or disjoint i64 %11, %18, !dbg !16 + %20 = sext i32 %5 to i64, !dbg !17 + %21 = icmp slt i64 %19, %20, !dbg !17 + %22 = shl nuw nsw i32 %12, 1, !dbg !18 + %23 = and i32 %22, 126, !dbg !18 + %24 = shl nuw nsw i64 %19, 7, !dbg !19 + %25 = getelementptr bfloat, ptr addrspace(1) %0, i64 %24, !dbg !20 + %26 = zext nneg i32 %23 to i64, !dbg !21 + %27 = getelementptr bfloat, ptr addrspace(1) %25, i64 %26, !dbg !21 + %28 = getelementptr float, ptr addrspace(1) %1, i64 %19, !dbg !22 + %29 = getelementptr bfloat, ptr addrspace(1) %2, i64 %24, !dbg !23 + %30 = getelementptr bfloat, ptr addrspace(1) %29, i64 %26, !dbg !24 + %31 = getelementptr bfloat, ptr addrspace(1) %3, i64 %24, !dbg !25 + %32 = getelementptr bfloat, ptr addrspace(1) %31, i64 %26, !dbg !26 + %33 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %27, i1 %21) #4, !dbg !27 + %34 = bitcast i32 %33 to <2 x bfloat>, !dbg !27 + %35 = extractelement <2 x bfloat> %34, i64 0, !dbg !27 + %36 = extractelement <2 x bfloat> %34, i64 1, !dbg !27 + %37 = fpext bfloat %35 to float, !dbg !28 + %38 = fpext bfloat %36 to float, !dbg !28 + %39 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %28, i1 %21) #4, !dbg !29 + %40 = bitcast i32 %39 to float, !dbg !29 + %41 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %30, i1 %21) #4, !dbg !30 + %42 = bitcast i32 %41 to <2 x bfloat>, !dbg !30 + %43 = extractelement <2 x bfloat> %42, i64 0, !dbg !30 + %44 = extractelement <2 x bfloat> %42, i64 1, !dbg !30 + %45 = fpext bfloat %43 to float, !dbg !31 + %46 = fpext bfloat %44 to float, !dbg !31 + %47 = fmul float %40, %45, !dbg !32 + %48 = fmul float %40, %46, !dbg !32 + %49 = fmul float %37, %45, !dbg !33 + %50 = fmul float %38, %46, !dbg !33 + %51 = fadd float %49, %50, !dbg !34 + %52 = bitcast float %51 to i32, !dbg !37 + %53 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %52, i32 16, i32 31), !dbg !37 + %54 = bitcast i32 %53 to float, !dbg !37 + %55 = fadd float %51, %54, !dbg !34 + %56 = bitcast float %55 to i32, !dbg !37 + %57 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %56, i32 8, i32 31), !dbg !37 + %58 = bitcast i32 %57 to float, !dbg !37 + %59 = fadd float %55, %58, !dbg !34 + %60 = bitcast float %59 to i32, !dbg !37 + %61 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %60, i32 4, i32 31), !dbg !37 + %62 = bitcast i32 %61 to float, !dbg !37 + %63 = fadd float %59, %62, !dbg !34 + %64 = bitcast float %63 to i32, !dbg !37 + %65 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %64, i32 2, i32 31), !dbg !37 + %66 = bitcast i32 %65 to float, !dbg !37 + %67 = fadd float %63, %66, !dbg !34 + %68 = bitcast float %67 to i32, !dbg !37 + %69 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %68, i32 1, i32 31), !dbg !37 + %70 = bitcast i32 %69 to float, !dbg !37 + %71 = fadd float %67, %70, !dbg !34 + %72 = and i32 %15, 1, !dbg !37 + %73 = icmp eq i32 %14, 0, !dbg !37 + %.idx = shl nuw nsw i32 %17, 3, !dbg !37 + %74 = getelementptr i8, ptr addrspace(3) @global_smem, i32 %.idx, !dbg !37 + %75 = getelementptr float, ptr addrspace(3) %74, i32 %72, !dbg !37 + %76 = bitcast float %71 to <1 x i32>, !dbg !37 + tail call void asm sideeffect "@$2 st.shared.b32 [ $0 + 0 ], $1;", "r,r,b"(ptr addrspace(3) %75, <1 x i32> %76, i1 %73) #4, !dbg !37 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !37 + %77 = icmp samesign ult i32 %13, 16, !dbg !37 + %78 = getelementptr float, ptr addrspace(3) @global_smem, i32 %13, !dbg !37 + %79 = tail call i32 asm sideeffect "@$2 ld.shared.b32 $0, [ $1 + 0 ];", "=r,r,b"(ptr addrspace(3) %78, i1 %77) #4, !dbg !37 + %80 = bitcast i32 %79 to float, !dbg !37 + %81 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %79, i32 1, i32 31), !dbg !37 + %82 = bitcast i32 %81 to float, !dbg !37 + %83 = fadd float %80, %82, !dbg !34 + %84 = and i32 %12, 1, !dbg !37 + %85 = icmp eq i32 %84, 0, !dbg !37 + %86 = and i1 %77, %85, !dbg !37 + %87 = bitcast float %83 to <1 x i32>, !dbg !37 + tail call void asm sideeffect "@$2 st.shared.b32 [ $0 + 0 ], $1;", "r,r,b"(ptr addrspace(3) %78, <1 x i32> %87, i1 %86) #4, !dbg !37 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !37 + %88 = load float, ptr addrspace(3) %74, align 8, !dbg !37 + %89 = fmul float %88, %37, !dbg !40 + %90 = fmul float %88, %38, !dbg !40 + %91 = fmul float %89, %40, !dbg !41 + %92 = fmul float %90, %40, !dbg !41 + %93 = fsub float %47, %91, !dbg !42 + %94 = fsub float %48, %92, !dbg !42 + %95 = fptrunc float %93 to bfloat, !dbg !43 + %96 = fptrunc float %94 to bfloat, !dbg !43 + %97 = insertelement <2 x bfloat> poison, bfloat %95, i64 0, !dbg !44 + %98 = insertelement <2 x bfloat> %97, bfloat %96, i64 1, !dbg !44 + %99 = bitcast <2 x bfloat> %98 to i32, !dbg !44 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %99, ptr addrspace(1) %32, i1 %21) #4, !dbg !44 + ret void, !dbg !45 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="512" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { convergent nocallback nounwind } +attributes #4 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = distinct !DISubprogram(name: "l2norm_bwd_kernel", linkageName: "l2norm_bwd_kernel", scope: !1, file: !1, line: 93, type: !5, scopeLine: 93, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!5 = !DISubroutineType(cc: DW_CC_normal, types: !6) +!6 = !{null, !7, !9, !7, !7, !10, !11, !7, !7} +!7 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !8, size: 64, dwarfAddressSpace: 1) +!8 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!9 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !10, size: 64, dwarfAddressSpace: 1) +!10 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!11 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!12 = !DILocation(line: 94, column: 24, scope: !4) +!13 = !DILocation(line: 94, column: 30, scope: !4) +!14 = !DILocation(line: 95, column: 16, scope: !4) +!15 = !DILocation(line: 95, column: 34, scope: !4) +!16 = !DILocation(line: 95, column: 21, scope: !4) +!17 = !DILocation(line: 97, column: 16, scope: !4) +!18 = !DILocation(line: 98, column: 30, scope: !4) +!19 = !DILocation(line: 99, column: 29, scope: !4) +!20 = !DILocation(line: 99, column: 14, scope: !4) +!21 = !DILocation(line: 99, column: 33, scope: !4) +!22 = !DILocation(line: 100, column: 20, scope: !4) +!23 = !DILocation(line: 101, column: 16, scope: !4) +!24 = !DILocation(line: 101, column: 35, scope: !4) +!25 = !DILocation(line: 102, column: 16, scope: !4) +!26 = !DILocation(line: 102, column: 35, scope: !4) +!27 = !DILocation(line: 104, column: 18, scope: !4) +!28 = !DILocation(line: 104, column: 47, scope: !4) +!29 = !DILocation(line: 105, column: 21, scope: !4) +!30 = !DILocation(line: 106, column: 19, scope: !4) +!31 = !DILocation(line: 106, column: 49, scope: !4) +!32 = !DILocation(line: 107, column: 18, scope: !4) +!33 = !DILocation(line: 107, column: 50, scope: !4) +!34 = !DILocation(line: 263, column: 15, scope: !35, inlinedAt: !37) +!35 = distinct !DILexicalBlockFile(scope: !4, file: !36, discriminator: 0) +!36 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!37 = !DILocation(line: 293, column: 36, scope: !35, inlinedAt: !38) +!38 = !DILocation(line: 107, column: 55, scope: !39) +!39 = distinct !DILexicalBlockFile(scope: !4, file: !1, discriminator: 0) +!40 = !DILocation(line: 107, column: 69, scope: !4) +!41 = !DILocation(line: 107, column: 75, scope: !4) +!42 = !DILocation(line: 107, column: 36, scope: !4) +!43 = !DILocation(line: 108, column: 27, scope: !4) +!44 = !DILocation(line: 108, column: 19, scope: !4) +!45 = !DILocation(line: 108, column: 4, scope: !4) diff --git a/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/l2norm_bwd_kernel.ptx b/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/l2norm_bwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..047213c6e953d43ba105d10d526ab28d584fd0e7 --- /dev/null +++ b/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/l2norm_bwd_kernel.ptx @@ -0,0 +1,407 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_bwd_kernel // -- Begin function l2norm_bwd_kernel +.extern .shared .align 16 .b8 global_smem[]; + // @l2norm_bwd_kernel +.visible .entry l2norm_bwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_2, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_3, + .param .f32 l2norm_bwd_kernel_param_4, + .param .u32 l2norm_bwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_6, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_7 +) +.reqntid 512 +{ + .reg .pred %p<7>; + .reg .b16 %rs<5>; + .reg .b32 %r<50>; + .reg .b64 %rd<19>; + .loc 1 93 0 // l2norm.py:93:0 +$L__func_begin0: + .loc 1 93 0 // l2norm.py:93:0 + +// %bb.0: + ld.param.b64 %rd5, [l2norm_bwd_kernel_param_0]; + ld.param.b64 %rd6, [l2norm_bwd_kernel_param_1]; +$L__tmp0: + .loc 1 94 24 // l2norm.py:94:24 + mov.u32 %r11, %ctaid.x; + ld.param.b64 %rd7, [l2norm_bwd_kernel_param_2]; + .loc 1 95 16 // l2norm.py:95:16 + mul.wide.u32 %rd8, %r11, 8; + ld.param.b64 %rd9, [l2norm_bwd_kernel_param_3]; + .loc 1 95 34 // l2norm.py:95:34 + mov.u32 %r12, %tid.x; + and.b32 %r13, %r12, 511; + ld.param.s32 %rd10, [l2norm_bwd_kernel_param_5]; + and.b32 %r14, %r12, 31; + bfe.u32 %r15, %r12, 6, 3; + .loc 1 95 21 // l2norm.py:95:21 + cvt.u64.u32 %rd11, %r15; + or.b64 %rd12, %rd8, %rd11; + .loc 1 97 16 // l2norm.py:97:16 + setp.lt.s64 %p1, %rd12, %rd10; + .loc 1 98 30 // l2norm.py:98:30 + shl.b32 %r16, %r12, 1; + and.b32 %r17, %r16, 126; + .loc 1 99 14 // l2norm.py:99:14 + shl.b64 %rd13, %rd12, 8; + add.s64 %rd14, %rd5, %rd13; + .loc 1 99 33 // l2norm.py:99:33 + mul.wide.u32 %rd15, %r17, 2; + add.s64 %rd1, %rd14, %rd15; + .loc 1 100 20 // l2norm.py:100:20 + shl.b64 %rd16, %rd12, 2; + add.s64 %rd2, %rd6, %rd16; + .loc 1 101 16 // l2norm.py:101:16 + add.s64 %rd17, %rd7, %rd13; + .loc 1 101 35 // l2norm.py:101:35 + add.s64 %rd3, %rd17, %rd15; + .loc 1 102 16 // l2norm.py:102:16 + add.s64 %rd18, %rd9, %rd13; + .loc 1 102 35 // l2norm.py:102:35 + add.s64 %rd4, %rd18, %rd15; + mov.b32 %r2, 0; + .loc 1 104 18 // l2norm.py:104:18 + // begin inline asm + mov.u32 %r1, %r2; + @%p1 ld.global.b32 { %r1 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + .loc 1 104 47 // l2norm.py:104:47 + cvt.f32.bf16 %r18, %rs1; + cvt.f32.bf16 %r19, %rs2; + .loc 1 105 21 // l2norm.py:105:21 + // begin inline asm + mov.u32 %r3, %r2; + @%p1 ld.global.b32 { %r3 }, [ %rd2 + 0 ]; + // end inline asm + .loc 1 106 19 // l2norm.py:106:19 + // begin inline asm + mov.u32 %r4, %r2; + @%p1 ld.global.b32 { %r4 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs3, %rs4}, %r4; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r20, %rs3; + cvt.f32.bf16 %r21, %rs4; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r22, %r19, %r21; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r23, %rs1, %rs3, %r22; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r24, %r23, 16, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r25, %r23, %r24; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r26, %r25, 8, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r27, %r25, %r26; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r28, %r27, 4, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r29, %r27, %r28; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r30, %r29, 2, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r31, %r29, %r30; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r32, %r31, 1, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r6, %r31, %r32; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + setp.eq.b32 %p2, %r14, 0; + shl.b32 %r33, %r15, 3; + mov.b32 %r34, global_smem; + add.s32 %r35, %r34, %r33; + shr.u32 %r36, %r12, 3; + and.b32 %r37, %r36, 4; + add.s32 %r5, %r35, %r37; + // begin inline asm + @%p2 st.shared.b32 [ %r5 + 0 ], %r6; + // end inline asm + bar.sync 0; + setp.lt.u32 %p3, %r13, 16; + shl.b32 %r38, %r13, 2; + add.s32 %r8, %r34, %r38; + // begin inline asm + @%p3 ld.shared.b32 %r7, [ %r8 + 0 ]; + // end inline asm + shfl.sync.bfly.b32 %r39, %r7, 1, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r9, %r7, %r39; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + and.b32 %r40, %r12, 1; + setp.ne.b32 %p5, %r40, 0; + not.pred %p6, %p5; + and.pred %p4, %p3, %p6; + // begin inline asm + @%p4 st.shared.b32 [ %r8 + 0 ], %r9; + // end inline asm + bar.sync 0; + ld.shared.b32 %r41, [%r35]; +$L__tmp15: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r42, %r41, %r18; + mul.f32 %r43, %r41, %r19; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r44, %r42, %r3; + mul.f32 %r45, %r43, %r3; + .loc 1 107 36 // l2norm.py:107:36 + neg.f32 %r46, %r44; + fma.rn.f32 %r47, %r3, %r20, %r46; + neg.f32 %r48, %r45; + fma.rn.f32 %r49, %r3, %r21, %r48; + .loc 1 108 19 // l2norm.py:108:19 + cvt.rn.bf16x2.f32 %r10, %r49, %r47; + // begin inline asm + @%p1 st.global.b32 [ %rd4 + 0 ], { %r10 }; + // end inline asm + .loc 1 108 4 // l2norm.py:108:4 + ret; +$L__tmp16: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 98 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp15 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 107 // DW_AT_call_line +.b8 55 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp14 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/l2norm_bwd_kernel.source b/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/l2norm_bwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..c23a4689ce5f60989f46cd34fecfe31d002a62d8 --- /dev/null +++ b/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/l2norm_bwd_kernel.source @@ -0,0 +1,210 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc47 = loc(unknown) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc54 = loc("y"(#loc)) +#loc55 = loc("rstd"(#loc)) +#loc56 = loc("dy"(#loc)) +#loc57 = loc("dx"(#loc)) +#loc58 = loc("eps"(#loc)) +#loc59 = loc("T"(#loc)) +#loc101 = loc("input"(#loc45)) +#loc102 = loc("a"(#loc50)) +#loc103 = loc("b"(#loc50)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc60) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc61) + %o_t = arith.constant 8 : i32 loc(#loc62) + %o_t_1 = arith.constant 8 : i64 loc(#loc62) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc62) + %o_t_3 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32> loc(#loc63) + %o_t_4 = arith.extsi %o_t_3 : tensor<8xi32> to tensor<8xi64> loc(#loc64) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<8xi64> loc(#loc64) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<8xi64> loc(#loc64) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc65) + %m_t = arith.extsi %T : i32 to i64 loc(#loc66) + %m_t_7 = tt.splat %m_t : i64 -> tensor<8xi64> loc(#loc66) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<8xi64> loc(#loc66) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<8xi1> -> tensor<8x1xi1> loc(#loc67) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc68) + %m_x_10 = arith.constant 128 : i32 loc(#loc69) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc69) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc69) + %m_x_13 = tt.broadcast %m_x : tensor<8x1xi1> -> tensor<8x128xi1> loc(#loc70) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<8x128xi1> loc(#loc70) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<8x128xi1> loc(#loc70) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc71) + %p_y_16 = arith.constant 128 : i32 loc(#loc72) + %p_y_17 = arith.constant 128 : i64 loc(#loc72) + %p_y_18 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc72) + %p_y_19 = arith.muli %p_y, %p_y_18 : tensor<8x1xi64> loc(#loc72) + %p_y_20 = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc73) + %p_y_21 = tt.addptr %p_y_20, %p_y_19 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc73) + %p_y_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc74) + %p_y_23 = tt.broadcast %p_y_21 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc75) + %p_y_24 = tt.broadcast %p_y_22 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc75) + %p_y_25 = tt.addptr %p_y_23, %p_y_24 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc75) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr> loc(#loc76) + %p_rstd_26 = tt.addptr %p_rstd, %o_t_6 : tensor<8x!tt.ptr>, tensor<8xi64> loc(#loc76) + %p_dy = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc77) + %p_dy_27 = arith.constant 128 : i32 loc(#loc78) + %p_dy_28 = arith.constant 128 : i64 loc(#loc78) + %p_dy_29 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc78) + %p_dy_30 = arith.muli %p_dy, %p_dy_29 : tensor<8x1xi64> loc(#loc78) + %p_dy_31 = tt.splat %dy : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc79) + %p_dy_32 = tt.addptr %p_dy_31, %p_dy_30 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc79) + %p_dy_33 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc80) + %p_dy_34 = tt.broadcast %p_dy_32 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc81) + %p_dy_35 = tt.broadcast %p_dy_33 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc81) + %p_dy_36 = tt.addptr %p_dy_34, %p_dy_35 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc81) + %p_dx = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc82) + %p_dx_37 = arith.constant 128 : i32 loc(#loc83) + %p_dx_38 = arith.constant 128 : i64 loc(#loc83) + %p_dx_39 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc83) + %p_dx_40 = arith.muli %p_dx, %p_dx_39 : tensor<8x1xi64> loc(#loc83) + %p_dx_41 = tt.splat %dx : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc84) + %p_dx_42 = tt.addptr %p_dx_41, %p_dx_40 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc84) + %p_dx_43 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc85) + %p_dx_44 = tt.broadcast %p_dx_42 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc86) + %p_dx_45 = tt.broadcast %p_dx_43 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc86) + %p_dx_46 = tt.addptr %p_dx_44, %p_dx_45 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc86) + %b_y = arith.constant 0.000000e+00 : f32 loc(#loc87) + %b_y_47 = arith.constant dense<0.000000e+00> : tensor<8x128xf32> loc(#loc87) + %b_y_48 = arith.truncf %b_y_47 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc87) + %b_y_49 = tt.load %p_y_25, %m_x_15, %b_y_48 : tensor<8x128x!tt.ptr> loc(#loc87) + %b_y_50 = arith.extf %b_y_49 : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc88) + %b_rstd = arith.constant 0.000000e+00 : f32 loc(#loc89) + %b_rstd_51 = arith.constant dense<0.000000e+00> : tensor<8xf32> loc(#loc89) + %b_rstd_52 = tt.load %p_rstd_26, %m_t_8, %b_rstd_51 : tensor<8x!tt.ptr> loc(#loc89) + %b_dy = arith.constant 0.000000e+00 : f32 loc(#loc90) + %b_dy_53 = arith.constant dense<0.000000e+00> : tensor<8x128xf32> loc(#loc90) + %b_dy_54 = arith.truncf %b_dy_53 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc90) + %b_dy_55 = tt.load %p_dy_36, %m_x_15, %b_dy_54 : tensor<8x128x!tt.ptr> loc(#loc90) + %b_dy_56 = arith.extf %b_dy_55 : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc91) + %b_dx = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc92) + %b_dx_57 = tt.broadcast %b_dx : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc93) + %b_dx_58 = arith.mulf %b_dy_56, %b_dx_57 : tensor<8x128xf32> loc(#loc93) + %b_dx_59 = arith.mulf %b_dy_56, %b_y_50 : tensor<8x128xf32> loc(#loc94) + %b_dx_60 = tt.call @triton.language.standard.sum__fp32S8_128S_c1_cFalse_cNone(%b_dx_59) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc95) + %b_dx_61 = tt.expand_dims %b_dx_60 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc96) + %b_dx_62 = tt.broadcast %b_dx_61 : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc97) + %b_dx_63 = arith.mulf %b_dx_62, %b_y_50 : tensor<8x128xf32> loc(#loc97) + %b_dx_64 = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc98) + %b_dx_65 = tt.broadcast %b_dx_64 : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc99) + %b_dx_66 = arith.mulf %b_dx_63, %b_dx_65 : tensor<8x128xf32> loc(#loc99) + %b_dx_67 = arith.subf %b_dx_58, %b_dx_66 : tensor<8x128xf32> loc(#loc100) + %0 = arith.truncf %b_dx_67 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc42) + tt.store %p_dx_46, %0, %m_x_15 : tensor<8x128x!tt.ptr> loc(#loc43) + tt.return loc(#loc44) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S8_128S_c1_cFalse_cNone(%input: tensor<8x128xf32> loc("input"(#loc45))) -> tensor<8xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc46) + tt.reduce.return %2 : f32 loc(#loc46) + }) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc46) + tt.return %0 : tensor<8xf32> loc(#loc48) + ^bb1: // no predecessors + %1 = ub.poison : tensor<8xf32> loc(#loc49) + tt.return %1 : tensor<8xf32> loc(#loc49) + } loc(#loc45) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc50)), %b: f32 loc("b"(#loc50))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc51) + tt.return %0 : f32 loc(#loc52) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc53) + tt.return %1 : f32 loc(#loc53) + } loc(#loc50) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:31) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:39) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:20) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:31) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:39) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:82) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc60 = loc("i_t"(#loc1)) +#loc61 = loc("i_t"(#loc2)) +#loc62 = loc("o_t"(#loc3)) +#loc63 = loc("o_t"(#loc4)) +#loc64 = loc("o_t"(#loc5)) +#loc65 = loc("o_d"(#loc6)) +#loc66 = loc("m_t"(#loc7)) +#loc67 = loc("m_x"(#loc8)) +#loc68 = loc("m_x"(#loc9)) +#loc69 = loc("m_x"(#loc10)) +#loc70 = loc("m_x"(#loc11)) +#loc71 = loc("p_y"(#loc12)) +#loc72 = loc("p_y"(#loc13)) +#loc73 = loc("p_y"(#loc14)) +#loc74 = loc("p_y"(#loc15)) +#loc75 = loc("p_y"(#loc16)) +#loc76 = loc("p_rstd"(#loc17)) +#loc77 = loc("p_dy"(#loc18)) +#loc78 = loc("p_dy"(#loc19)) +#loc79 = loc("p_dy"(#loc20)) +#loc80 = loc("p_dy"(#loc21)) +#loc81 = loc("p_dy"(#loc22)) +#loc82 = loc("p_dx"(#loc23)) +#loc83 = loc("p_dx"(#loc24)) +#loc84 = loc("p_dx"(#loc25)) +#loc85 = loc("p_dx"(#loc26)) +#loc86 = loc("p_dx"(#loc27)) +#loc87 = loc("b_y"(#loc28)) +#loc88 = loc("b_y"(#loc29)) +#loc89 = loc("b_rstd"(#loc30)) +#loc90 = loc("b_dy"(#loc31)) +#loc91 = loc("b_dy"(#loc32)) +#loc92 = loc("b_dx"(#loc33)) +#loc93 = loc("b_dx"(#loc34)) +#loc94 = loc("b_dx"(#loc35)) +#loc95 = loc("b_dx"(#loc36)) +#loc96 = loc("b_dx"(#loc37)) +#loc97 = loc("b_dx"(#loc38)) +#loc98 = loc("b_dx"(#loc39)) +#loc99 = loc("b_dx"(#loc40)) +#loc100 = loc("b_dx"(#loc41)) diff --git a/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/l2norm_bwd_kernel.ttgir b/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/l2norm_bwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..fbd57ca45a7a3b581a068d951ebad77404c70446 --- /dev/null +++ b/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/l2norm_bwd_kernel.ttgir @@ -0,0 +1,146 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 2], threadsPerWarp = [1, 32], warpsPerCTA = [8, 2], order = [1, 0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc39 = loc("y"(#loc)) +#loc40 = loc("rstd"(#loc)) +#loc41 = loc("dy"(#loc)) +#loc42 = loc("dx"(#loc)) +#loc43 = loc("eps"(#loc)) +#loc44 = loc("T"(#loc)) +#loc72 = loc("b_dx"(#loc30)) +#loc78 = loc(callsite(#loc1 at #loc72)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 16 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %c8_i64 = arith.constant 8 : i64 loc(#loc1) + %cst = arith.constant dense<0.000000e+00> : tensor<8x128xbf16, #blocked> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<8x1xi64, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc45) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc46) + %o_t = arith.muli %i_t_3, %c8_i64 : i64 loc(#loc47) + %o_t_4 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc48) + %o_t_5 = arith.extsi %o_t_4 : tensor<8xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %o_t_6 = tt.splat %o_t : i64 -> tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %o_t_7 = arith.addi %o_t_6, %o_t_5 : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %m_t = arith.extsi %T : i32 to i64 loc(#loc50) + %m_t_8 = tt.splat %m_t : i64 -> tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc50) + %m_t_9 = arith.cmpi slt, %o_t_7, %m_t_8 : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc50) + %m_x = tt.expand_dims %m_t_9 {axis = 1 : i32} : tensor<8xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xi1, #blocked> loc(#loc51) + %m_x_10 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc52) + %m_x_11 = tt.expand_dims %m_x_10 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc52) + %m_x_12 = arith.cmpi slt, %m_x_11, %cst_2 : tensor<1x128xi32, #blocked> loc(#loc53) + %m_x_13 = tt.broadcast %m_x : tensor<8x1xi1, #blocked> -> tensor<8x128xi1, #blocked> loc(#loc54) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1, #blocked> -> tensor<8x128xi1, #blocked> loc(#loc54) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<8x128xi1, #blocked> loc(#loc54) + %p_y = tt.expand_dims %o_t_7 {axis = 1 : i32} : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xi64, #blocked> loc(#loc55) + %p_y_16 = arith.muli %p_y, %cst_1 : tensor<8x1xi64, #blocked> loc(#loc56) + %p_y_17 = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr, #blocked> loc(#loc57) + %p_y_18 = tt.addptr %p_y_17, %p_y_16 : tensor<8x1x!tt.ptr, #blocked>, tensor<8x1xi64, #blocked> loc(#loc57) + %p_y_19 = tt.broadcast %p_y_18 : tensor<8x1x!tt.ptr, #blocked> -> tensor<8x128x!tt.ptr, #blocked> loc(#loc58) + %p_y_20 = tt.broadcast %m_x_11 : tensor<1x128xi32, #blocked> -> tensor<8x128xi32, #blocked> loc(#loc58) + %p_y_21 = tt.addptr %p_y_19, %p_y_20 : tensor<8x128x!tt.ptr, #blocked>, tensor<8x128xi32, #blocked> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc59) + %p_rstd_22 = tt.addptr %p_rstd, %o_t_7 : tensor<8x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>>, tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc59) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<8x1x!tt.ptr, #blocked> loc(#loc60) + %p_dy_23 = tt.addptr %p_dy, %p_y_16 : tensor<8x1x!tt.ptr, #blocked>, tensor<8x1xi64, #blocked> loc(#loc60) + %p_dy_24 = tt.broadcast %p_dy_23 : tensor<8x1x!tt.ptr, #blocked> -> tensor<8x128x!tt.ptr, #blocked> loc(#loc61) + %p_dy_25 = tt.addptr %p_dy_24, %p_y_20 : tensor<8x128x!tt.ptr, #blocked>, tensor<8x128xi32, #blocked> loc(#loc61) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<8x1x!tt.ptr, #blocked> loc(#loc62) + %p_dx_26 = tt.addptr %p_dx, %p_y_16 : tensor<8x1x!tt.ptr, #blocked>, tensor<8x1xi64, #blocked> loc(#loc62) + %p_dx_27 = tt.broadcast %p_dx_26 : tensor<8x1x!tt.ptr, #blocked> -> tensor<8x128x!tt.ptr, #blocked> loc(#loc63) + %p_dx_28 = tt.addptr %p_dx_27, %p_y_20 : tensor<8x128x!tt.ptr, #blocked>, tensor<8x128xi32, #blocked> loc(#loc63) + %b_y = tt.load %p_y_21, %m_x_15, %cst : tensor<8x128x!tt.ptr, #blocked> loc(#loc64) + %b_y_29 = arith.extf %b_y : tensor<8x128xbf16, #blocked> to tensor<8x128xf32, #blocked> loc(#loc65) + %b_rstd = tt.load %p_rstd_22, %m_t_9, %cst_0 : tensor<8x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc66) + %b_dy = tt.load %p_dy_25, %m_x_15, %cst : tensor<8x128x!tt.ptr, #blocked> loc(#loc67) + %b_dy_30 = arith.extf %b_dy : tensor<8x128xbf16, #blocked> to tensor<8x128xf32, #blocked> loc(#loc68) + %b_dx = tt.expand_dims %b_rstd {axis = 1 : i32} : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xf32, #blocked> loc(#loc69) + %b_dx_31 = tt.broadcast %b_dx : tensor<8x1xf32, #blocked> -> tensor<8x128xf32, #blocked> loc(#loc70) + %b_dx_32 = arith.mulf %b_dy_30, %b_dx_31 : tensor<8x128xf32, #blocked> loc(#loc70) + %b_dx_33 = arith.mulf %b_dy_30, %b_y_29 : tensor<8x128xf32, #blocked> loc(#loc71) + %b_dx_34 = "tt.reduce"(%b_dx_33) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_40: f32 loc(callsite(#loc1 at #loc72)), %b_dx_41: f32 loc(callsite(#loc1 at #loc72))): + %b_dx_42 = arith.addf %b_dx_40, %b_dx_41 : f32 loc(#loc79) + tt.reduce.return %b_dx_42 : f32 loc(#loc77) + }) : (tensor<8x128xf32, #blocked>) -> tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc77) + %b_dx_35 = tt.expand_dims %b_dx_34 {axis = 1 : i32} : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xf32, #blocked> loc(#loc73) + %b_dx_36 = tt.broadcast %b_dx_35 : tensor<8x1xf32, #blocked> -> tensor<8x128xf32, #blocked> loc(#loc74) + %b_dx_37 = arith.mulf %b_dx_36, %b_y_29 : tensor<8x128xf32, #blocked> loc(#loc74) + %b_dx_38 = arith.mulf %b_dx_37, %b_dx_31 : tensor<8x128xf32, #blocked> loc(#loc75) + %b_dx_39 = arith.subf %b_dx_32, %b_dx_38 : tensor<8x128xf32, #blocked> loc(#loc76) + %0 = arith.truncf %b_dx_39 : tensor<8x128xf32, #blocked> to tensor<8x128xbf16, #blocked> loc(#loc36) + tt.store %p_dx_28, %0, %m_x_15 : tensor<8x128x!tt.ptr, #blocked> loc(#loc37) + tt.return loc(#loc38) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc45 = loc("i_t"(#loc2)) +#loc46 = loc("i_t"(#loc3)) +#loc47 = loc("o_t"(#loc4)) +#loc48 = loc("o_t"(#loc5)) +#loc49 = loc("o_t"(#loc6)) +#loc50 = loc("m_t"(#loc7)) +#loc51 = loc("m_x"(#loc8)) +#loc52 = loc("m_x"(#loc9)) +#loc53 = loc("m_x"(#loc10)) +#loc54 = loc("m_x"(#loc11)) +#loc55 = loc("p_y"(#loc12)) +#loc56 = loc("p_y"(#loc13)) +#loc57 = loc("p_y"(#loc14)) +#loc58 = loc("p_y"(#loc15)) +#loc59 = loc("p_rstd"(#loc16)) +#loc60 = loc("p_dy"(#loc17)) +#loc61 = loc("p_dy"(#loc18)) +#loc62 = loc("p_dx"(#loc19)) +#loc63 = loc("p_dx"(#loc20)) +#loc64 = loc("b_y"(#loc21)) +#loc65 = loc("b_y"(#loc22)) +#loc66 = loc("b_rstd"(#loc23)) +#loc67 = loc("b_dy"(#loc24)) +#loc68 = loc("b_dy"(#loc25)) +#loc69 = loc("b_dx"(#loc26)) +#loc70 = loc("b_dx"(#loc27)) +#loc71 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc32)) +#loc74 = loc("b_dx"(#loc33)) +#loc75 = loc("b_dx"(#loc34)) +#loc76 = loc("b_dx"(#loc35)) +#loc77 = loc(callsite(#loc29 at #loc72)) +#loc79 = loc(callsite(#loc31 at #loc77)) diff --git a/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/l2norm_bwd_kernel.ttir b/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/l2norm_bwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..c4761315a53ecb33821860b30f93bdae9fbfe731 --- /dev/null +++ b/triton/gpu0/B2EJJPVC7HWZAVLH4JERHXAL22BSCXT5XPT325WLRNV2RI34J66A/l2norm_bwd_kernel.ttir @@ -0,0 +1,147 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc40 = loc("y"(#loc)) +#loc41 = loc("rstd"(#loc)) +#loc42 = loc("dy"(#loc)) +#loc43 = loc("dx"(#loc)) +#loc44 = loc("eps"(#loc)) +#loc45 = loc("T"(#loc)) +#loc74 = loc("b_dx"(#loc31)) +#loc80 = loc(callsite(#loc1 at #loc74)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<0.000000e+00> : tensor<8x128xbf16> loc(#loc1) + %b_rstd = arith.constant dense<0.000000e+00> : tensor<8xf32> loc(#loc46) + %cst_0 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc1) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc47) + %o_t = arith.constant 8 : i64 loc(#loc48) + %i_t = tt.get_program_id x : i32 loc(#loc49) + %i_t_1 = arith.extsi %i_t : i32 to i64 loc(#loc50) + %o_t_2 = arith.muli %i_t_1, %o_t : i64 loc(#loc48) + %o_t_3 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32> loc(#loc51) + %o_t_4 = arith.extsi %o_t_3 : tensor<8xi32> to tensor<8xi64> loc(#loc52) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<8xi64> loc(#loc52) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<8xi64> loc(#loc52) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc53) + %m_t = arith.extsi %T : i32 to i64 loc(#loc54) + %m_t_7 = tt.splat %m_t : i64 -> tensor<8xi64> loc(#loc54) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<8xi64> loc(#loc54) + %m_x_9 = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<8xi1> -> tensor<8x1xi1> loc(#loc55) + %m_x_10 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc56) + %m_x_11 = arith.cmpi slt, %m_x_10, %m_x : tensor<1x128xi32> loc(#loc47) + %m_x_12 = tt.broadcast %m_x_9 : tensor<8x1xi1> -> tensor<8x128xi1> loc(#loc57) + %m_x_13 = tt.broadcast %m_x_11 : tensor<1x128xi1> -> tensor<8x128xi1> loc(#loc57) + %m_x_14 = arith.andi %m_x_12, %m_x_13 : tensor<8x128xi1> loc(#loc57) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc58) + %p_y_15 = arith.muli %p_y, %cst_0 : tensor<8x1xi64> loc(#loc59) + %p_y_16 = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc60) + %p_y_17 = tt.addptr %p_y_16, %p_y_15 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc60) + %p_y_18 = tt.broadcast %p_y_17 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc61) + %p_y_19 = tt.broadcast %m_x_10 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc61) + %p_y_20 = tt.addptr %p_y_18, %p_y_19 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc61) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr> loc(#loc62) + %p_rstd_21 = tt.addptr %p_rstd, %o_t_6 : tensor<8x!tt.ptr>, tensor<8xi64> loc(#loc62) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc63) + %p_dy_22 = tt.addptr %p_dy, %p_y_15 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc63) + %p_dy_23 = tt.broadcast %p_dy_22 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc64) + %p_dy_24 = tt.addptr %p_dy_23, %p_y_19 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc64) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc65) + %p_dx_25 = tt.addptr %p_dx, %p_y_15 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc65) + %p_dx_26 = tt.broadcast %p_dx_25 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc66) + %p_dx_27 = tt.addptr %p_dx_26, %p_y_19 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc66) + %b_y = tt.load %p_y_20, %m_x_14, %cst : tensor<8x128x!tt.ptr> loc(#loc67) + %b_y_28 = arith.extf %b_y : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc68) + %b_rstd_29 = tt.load %p_rstd_21, %m_t_8, %b_rstd : tensor<8x!tt.ptr> loc(#loc46) + %b_dy = tt.load %p_dy_24, %m_x_14, %cst : tensor<8x128x!tt.ptr> loc(#loc69) + %b_dy_30 = arith.extf %b_dy : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc70) + %b_dx = tt.expand_dims %b_rstd_29 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc71) + %b_dx_31 = tt.broadcast %b_dx : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc72) + %b_dx_32 = arith.mulf %b_dy_30, %b_dx_31 : tensor<8x128xf32> loc(#loc72) + %b_dx_33 = arith.mulf %b_dy_30, %b_y_28 : tensor<8x128xf32> loc(#loc73) + %b_dx_34 = "tt.reduce"(%b_dx_33) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_40: f32 loc(callsite(#loc1 at #loc74)), %b_dx_41: f32 loc(callsite(#loc1 at #loc74))): + %b_dx_42 = arith.addf %b_dx_40, %b_dx_41 : f32 loc(#loc81) + tt.reduce.return %b_dx_42 : f32 loc(#loc79) + }) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc79) + %b_dx_35 = tt.expand_dims %b_dx_34 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc75) + %b_dx_36 = tt.broadcast %b_dx_35 : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc76) + %b_dx_37 = arith.mulf %b_dx_36, %b_y_28 : tensor<8x128xf32> loc(#loc76) + %b_dx_38 = arith.mulf %b_dx_37, %b_dx_31 : tensor<8x128xf32> loc(#loc77) + %b_dx_39 = arith.subf %b_dx_32, %b_dx_38 : tensor<8x128xf32> loc(#loc78) + %0 = arith.truncf %b_dx_39 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc37) + tt.store %p_dx_27, %0, %m_x_14 : tensor<8x128x!tt.ptr> loc(#loc38) + tt.return loc(#loc39) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("b_rstd"(#loc2)) +#loc47 = loc("m_x"(#loc3)) +#loc48 = loc("o_t"(#loc4)) +#loc49 = loc("i_t"(#loc5)) +#loc50 = loc("i_t"(#loc6)) +#loc51 = loc("o_t"(#loc7)) +#loc52 = loc("o_t"(#loc8)) +#loc53 = loc("o_d"(#loc9)) +#loc54 = loc("m_t"(#loc10)) +#loc55 = loc("m_x"(#loc11)) +#loc56 = loc("m_x"(#loc12)) +#loc57 = loc("m_x"(#loc13)) +#loc58 = loc("p_y"(#loc14)) +#loc59 = loc("p_y"(#loc15)) +#loc60 = loc("p_y"(#loc16)) +#loc61 = loc("p_y"(#loc17)) +#loc62 = loc("p_rstd"(#loc18)) +#loc63 = loc("p_dy"(#loc19)) +#loc64 = loc("p_dy"(#loc20)) +#loc65 = loc("p_dx"(#loc21)) +#loc66 = loc("p_dx"(#loc22)) +#loc67 = loc("b_y"(#loc23)) +#loc68 = loc("b_y"(#loc24)) +#loc69 = loc("b_dy"(#loc25)) +#loc70 = loc("b_dy"(#loc26)) +#loc71 = loc("b_dx"(#loc27)) +#loc72 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc29)) +#loc75 = loc("b_dx"(#loc33)) +#loc76 = loc("b_dx"(#loc34)) +#loc77 = loc("b_dx"(#loc35)) +#loc78 = loc("b_dx"(#loc36)) +#loc79 = loc(callsite(#loc30 at #loc74)) +#loc81 = loc(callsite(#loc32 at #loc79)) diff --git a/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/__grp__l2norm_fwd_kernel.json b/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/__grp__l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..5bfe6ad1f4dc221a7180f561ab173dfc46f46543 --- /dev/null +++ b/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/__grp__l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_fwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/l2norm_fwd_kernel.source", "l2norm_fwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/l2norm_fwd_kernel.ttir", "l2norm_fwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/l2norm_fwd_kernel.ttgir", "l2norm_fwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/l2norm_fwd_kernel.llir", "l2norm_fwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/l2norm_fwd_kernel.ptx", "l2norm_fwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/l2norm_fwd_kernel.cubin", "l2norm_fwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/l2norm_fwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/l2norm_fwd_kernel.cubin b/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/l2norm_fwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..37d4b20faa5ad6573c5a66dc8f920eea7bb6f004 Binary files /dev/null and b/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/l2norm_fwd_kernel.cubin differ diff --git a/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/l2norm_fwd_kernel.json b/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..27fca67c736c299685b10a7e2569453e1ad58c0c --- /dev/null +++ b/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"hash": "0e982685f4a2bc6de51d163d275aaf1b3f5067c87e84e6857b723721d69c5b80", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 1, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 64, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_fwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/l2norm_fwd_kernel.llir b/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/l2norm_fwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..6e81c10fd2dcd838a23b456dedb66e31d7e57920 --- /dev/null +++ b/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/l2norm_fwd_kernel.llir @@ -0,0 +1,1143 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 +@.str.2 = private unnamed_addr constant [17 x i8] c"__CUDA_PREC_SQRT\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_fwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, float %3, i32 %4, ptr addrspace(1) readnone captures(none) %5, ptr addrspace(1) readnone captures(none) %6) local_unnamed_addr #0 !dbg !5 { + %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %9 = zext nneg i32 %8 to i64, !dbg !14 + %10 = shl nuw nsw i64 %9, 4, !dbg !15 + %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !16 + %12 = and i32 %11, 16, !dbg !16 + %13 = icmp eq i32 %12, 0, !dbg !16 + %not. = xor i1 %13, true, !dbg !16 + %14 = zext i1 %not. to i32, !dbg !16 + %15 = or disjoint i32 %14, 2, !dbg !16 + %16 = or disjoint i32 %14, 4, !dbg !16 + %17 = or disjoint i32 %14, 6, !dbg !16 + %18 = or disjoint i32 %14, 8, !dbg !16 + %19 = or disjoint i32 %14, 10, !dbg !16 + %20 = or disjoint i32 %14, 12, !dbg !16 + %21 = or disjoint i32 %14, 14, !dbg !16 + %22 = and i32 %11, 15, !dbg !16 + %23 = zext i1 %not. to i64, !dbg !17 + %24 = zext nneg i32 %15 to i64, !dbg !17 + %25 = zext nneg i32 %16 to i64, !dbg !17 + %26 = zext nneg i32 %17 to i64, !dbg !17 + %27 = zext nneg i32 %18 to i64, !dbg !17 + %28 = zext nneg i32 %19 to i64, !dbg !17 + %29 = zext nneg i32 %20 to i64, !dbg !17 + %30 = zext nneg i32 %21 to i64, !dbg !17 + %31 = or disjoint i64 %10, %23, !dbg !17 + %32 = or disjoint i64 %10, %24, !dbg !17 + %33 = or disjoint i64 %10, %25, !dbg !17 + %34 = or disjoint i64 %10, %26, !dbg !17 + %35 = or disjoint i64 %10, %27, !dbg !17 + %36 = or disjoint i64 %10, %28, !dbg !17 + %37 = or disjoint i64 %10, %29, !dbg !17 + %38 = or disjoint i64 %10, %30, !dbg !17 + %39 = sext i32 %4 to i64, !dbg !18 + %40 = icmp slt i64 %31, %39, !dbg !18 + %41 = icmp slt i64 %32, %39, !dbg !18 + %42 = icmp slt i64 %33, %39, !dbg !18 + %43 = icmp slt i64 %34, %39, !dbg !18 + %44 = icmp slt i64 %35, %39, !dbg !18 + %45 = icmp slt i64 %36, %39, !dbg !18 + %46 = icmp slt i64 %37, %39, !dbg !18 + %47 = icmp slt i64 %38, %39, !dbg !18 + %48 = shl nuw nsw i32 %22, 3, !dbg !19 + %49 = shl nuw nsw i64 %31, 7, !dbg !20 + %50 = shl nuw nsw i64 %32, 7, !dbg !20 + %51 = shl nuw nsw i64 %33, 7, !dbg !20 + %52 = shl nuw nsw i64 %34, 7, !dbg !20 + %53 = shl nuw nsw i64 %35, 7, !dbg !20 + %54 = shl nuw nsw i64 %36, 7, !dbg !20 + %55 = shl nuw nsw i64 %37, 7, !dbg !20 + %56 = shl nuw nsw i64 %38, 7, !dbg !20 + %57 = getelementptr bfloat, ptr addrspace(1) %0, i64 %49, !dbg !21 + %58 = getelementptr bfloat, ptr addrspace(1) %0, i64 %50, !dbg !21 + %59 = getelementptr bfloat, ptr addrspace(1) %0, i64 %51, !dbg !21 + %60 = getelementptr bfloat, ptr addrspace(1) %0, i64 %52, !dbg !21 + %61 = getelementptr bfloat, ptr addrspace(1) %0, i64 %53, !dbg !21 + %62 = getelementptr bfloat, ptr addrspace(1) %0, i64 %54, !dbg !21 + %63 = getelementptr bfloat, ptr addrspace(1) %0, i64 %55, !dbg !21 + %64 = getelementptr bfloat, ptr addrspace(1) %0, i64 %56, !dbg !21 + %65 = zext nneg i32 %48 to i64, !dbg !22 + %66 = getelementptr bfloat, ptr addrspace(1) %57, i64 %65, !dbg !22 + %67 = getelementptr bfloat, ptr addrspace(1) %58, i64 %65, !dbg !22 + %68 = getelementptr bfloat, ptr addrspace(1) %59, i64 %65, !dbg !22 + %69 = getelementptr bfloat, ptr addrspace(1) %60, i64 %65, !dbg !22 + %70 = getelementptr bfloat, ptr addrspace(1) %61, i64 %65, !dbg !22 + %71 = getelementptr bfloat, ptr addrspace(1) %62, i64 %65, !dbg !22 + %72 = getelementptr bfloat, ptr addrspace(1) %63, i64 %65, !dbg !22 + %73 = getelementptr bfloat, ptr addrspace(1) %64, i64 %65, !dbg !22 + %74 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %66, i1 %40) #6, !dbg !23 + %75 = extractvalue { i32, i32, i32, i32 } %74, 0, !dbg !23 + %76 = bitcast i32 %75 to <2 x bfloat>, !dbg !23 + %77 = extractvalue { i32, i32, i32, i32 } %74, 1, !dbg !23 + %78 = bitcast i32 %77 to <2 x bfloat>, !dbg !23 + %79 = extractvalue { i32, i32, i32, i32 } %74, 2, !dbg !23 + %80 = bitcast i32 %79 to <2 x bfloat>, !dbg !23 + %81 = extractvalue { i32, i32, i32, i32 } %74, 3, !dbg !23 + %82 = bitcast i32 %81 to <2 x bfloat>, !dbg !23 + %83 = extractelement <2 x bfloat> %76, i64 0, !dbg !23 + %84 = extractelement <2 x bfloat> %76, i64 1, !dbg !23 + %85 = extractelement <2 x bfloat> %78, i64 0, !dbg !23 + %86 = extractelement <2 x bfloat> %78, i64 1, !dbg !23 + %87 = extractelement <2 x bfloat> %80, i64 0, !dbg !23 + %88 = extractelement <2 x bfloat> %80, i64 1, !dbg !23 + %89 = extractelement <2 x bfloat> %82, i64 0, !dbg !23 + %90 = extractelement <2 x bfloat> %82, i64 1, !dbg !23 + %91 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %67, i1 %41) #6, !dbg !23 + %92 = extractvalue { i32, i32, i32, i32 } %91, 0, !dbg !23 + %93 = bitcast i32 %92 to <2 x bfloat>, !dbg !23 + %94 = extractvalue { i32, i32, i32, i32 } %91, 1, !dbg !23 + %95 = bitcast i32 %94 to <2 x bfloat>, !dbg !23 + %96 = extractvalue { i32, i32, i32, i32 } %91, 2, !dbg !23 + %97 = bitcast i32 %96 to <2 x bfloat>, !dbg !23 + %98 = extractvalue { i32, i32, i32, i32 } %91, 3, !dbg !23 + %99 = bitcast i32 %98 to <2 x bfloat>, !dbg !23 + %100 = extractelement <2 x bfloat> %93, i64 0, !dbg !23 + %101 = extractelement <2 x bfloat> %93, i64 1, !dbg !23 + %102 = extractelement <2 x bfloat> %95, i64 0, !dbg !23 + %103 = extractelement <2 x bfloat> %95, i64 1, !dbg !23 + %104 = extractelement <2 x bfloat> %97, i64 0, !dbg !23 + %105 = extractelement <2 x bfloat> %97, i64 1, !dbg !23 + %106 = extractelement <2 x bfloat> %99, i64 0, !dbg !23 + %107 = extractelement <2 x bfloat> %99, i64 1, !dbg !23 + %108 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %68, i1 %42) #6, !dbg !23 + %109 = extractvalue { i32, i32, i32, i32 } %108, 0, !dbg !23 + %110 = bitcast i32 %109 to <2 x bfloat>, !dbg !23 + %111 = extractvalue { i32, i32, i32, i32 } %108, 1, !dbg !23 + %112 = bitcast i32 %111 to <2 x bfloat>, !dbg !23 + %113 = extractvalue { i32, i32, i32, i32 } %108, 2, !dbg !23 + %114 = bitcast i32 %113 to <2 x bfloat>, !dbg !23 + %115 = extractvalue { i32, i32, i32, i32 } %108, 3, !dbg !23 + %116 = bitcast i32 %115 to <2 x bfloat>, !dbg !23 + %117 = extractelement <2 x bfloat> %110, i64 0, !dbg !23 + %118 = extractelement <2 x bfloat> %110, i64 1, !dbg !23 + %119 = extractelement <2 x bfloat> %112, i64 0, !dbg !23 + %120 = extractelement <2 x bfloat> %112, i64 1, !dbg !23 + %121 = extractelement <2 x bfloat> %114, i64 0, !dbg !23 + %122 = extractelement <2 x bfloat> %114, i64 1, !dbg !23 + %123 = extractelement <2 x bfloat> %116, i64 0, !dbg !23 + %124 = extractelement <2 x bfloat> %116, i64 1, !dbg !23 + %125 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %69, i1 %43) #6, !dbg !23 + %126 = extractvalue { i32, i32, i32, i32 } %125, 0, !dbg !23 + %127 = bitcast i32 %126 to <2 x bfloat>, !dbg !23 + %128 = extractvalue { i32, i32, i32, i32 } %125, 1, !dbg !23 + %129 = bitcast i32 %128 to <2 x bfloat>, !dbg !23 + %130 = extractvalue { i32, i32, i32, i32 } %125, 2, !dbg !23 + %131 = bitcast i32 %130 to <2 x bfloat>, !dbg !23 + %132 = extractvalue { i32, i32, i32, i32 } %125, 3, !dbg !23 + %133 = bitcast i32 %132 to <2 x bfloat>, !dbg !23 + %134 = extractelement <2 x bfloat> %127, i64 0, !dbg !23 + %135 = extractelement <2 x bfloat> %127, i64 1, !dbg !23 + %136 = extractelement <2 x bfloat> %129, i64 0, !dbg !23 + %137 = extractelement <2 x bfloat> %129, i64 1, !dbg !23 + %138 = extractelement <2 x bfloat> %131, i64 0, !dbg !23 + %139 = extractelement <2 x bfloat> %131, i64 1, !dbg !23 + %140 = extractelement <2 x bfloat> %133, i64 0, !dbg !23 + %141 = extractelement <2 x bfloat> %133, i64 1, !dbg !23 + %142 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %70, i1 %44) #6, !dbg !23 + %143 = extractvalue { i32, i32, i32, i32 } %142, 0, !dbg !23 + %144 = bitcast i32 %143 to <2 x bfloat>, !dbg !23 + %145 = extractvalue { i32, i32, i32, i32 } %142, 1, !dbg !23 + %146 = bitcast i32 %145 to <2 x bfloat>, !dbg !23 + %147 = extractvalue { i32, i32, i32, i32 } %142, 2, !dbg !23 + %148 = bitcast i32 %147 to <2 x bfloat>, !dbg !23 + %149 = extractvalue { i32, i32, i32, i32 } %142, 3, !dbg !23 + %150 = bitcast i32 %149 to <2 x bfloat>, !dbg !23 + %151 = extractelement <2 x bfloat> %144, i64 0, !dbg !23 + %152 = extractelement <2 x bfloat> %144, i64 1, !dbg !23 + %153 = extractelement <2 x bfloat> %146, i64 0, !dbg !23 + %154 = extractelement <2 x bfloat> %146, i64 1, !dbg !23 + %155 = extractelement <2 x bfloat> %148, i64 0, !dbg !23 + %156 = extractelement <2 x bfloat> %148, i64 1, !dbg !23 + %157 = extractelement <2 x bfloat> %150, i64 0, !dbg !23 + %158 = extractelement <2 x bfloat> %150, i64 1, !dbg !23 + %159 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %71, i1 %45) #6, !dbg !23 + %160 = extractvalue { i32, i32, i32, i32 } %159, 0, !dbg !23 + %161 = bitcast i32 %160 to <2 x bfloat>, !dbg !23 + %162 = extractvalue { i32, i32, i32, i32 } %159, 1, !dbg !23 + %163 = bitcast i32 %162 to <2 x bfloat>, !dbg !23 + %164 = extractvalue { i32, i32, i32, i32 } %159, 2, !dbg !23 + %165 = bitcast i32 %164 to <2 x bfloat>, !dbg !23 + %166 = extractvalue { i32, i32, i32, i32 } %159, 3, !dbg !23 + %167 = bitcast i32 %166 to <2 x bfloat>, !dbg !23 + %168 = extractelement <2 x bfloat> %161, i64 0, !dbg !23 + %169 = extractelement <2 x bfloat> %161, i64 1, !dbg !23 + %170 = extractelement <2 x bfloat> %163, i64 0, !dbg !23 + %171 = extractelement <2 x bfloat> %163, i64 1, !dbg !23 + %172 = extractelement <2 x bfloat> %165, i64 0, !dbg !23 + %173 = extractelement <2 x bfloat> %165, i64 1, !dbg !23 + %174 = extractelement <2 x bfloat> %167, i64 0, !dbg !23 + %175 = extractelement <2 x bfloat> %167, i64 1, !dbg !23 + %176 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %72, i1 %46) #6, !dbg !23 + %177 = extractvalue { i32, i32, i32, i32 } %176, 0, !dbg !23 + %178 = bitcast i32 %177 to <2 x bfloat>, !dbg !23 + %179 = extractvalue { i32, i32, i32, i32 } %176, 1, !dbg !23 + %180 = bitcast i32 %179 to <2 x bfloat>, !dbg !23 + %181 = extractvalue { i32, i32, i32, i32 } %176, 2, !dbg !23 + %182 = bitcast i32 %181 to <2 x bfloat>, !dbg !23 + %183 = extractvalue { i32, i32, i32, i32 } %176, 3, !dbg !23 + %184 = bitcast i32 %183 to <2 x bfloat>, !dbg !23 + %185 = extractelement <2 x bfloat> %178, i64 0, !dbg !23 + %186 = extractelement <2 x bfloat> %178, i64 1, !dbg !23 + %187 = extractelement <2 x bfloat> %180, i64 0, !dbg !23 + %188 = extractelement <2 x bfloat> %180, i64 1, !dbg !23 + %189 = extractelement <2 x bfloat> %182, i64 0, !dbg !23 + %190 = extractelement <2 x bfloat> %182, i64 1, !dbg !23 + %191 = extractelement <2 x bfloat> %184, i64 0, !dbg !23 + %192 = extractelement <2 x bfloat> %184, i64 1, !dbg !23 + %193 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %73, i1 %47) #6, !dbg !23 + %194 = extractvalue { i32, i32, i32, i32 } %193, 0, !dbg !23 + %195 = bitcast i32 %194 to <2 x bfloat>, !dbg !23 + %196 = extractvalue { i32, i32, i32, i32 } %193, 1, !dbg !23 + %197 = bitcast i32 %196 to <2 x bfloat>, !dbg !23 + %198 = extractvalue { i32, i32, i32, i32 } %193, 2, !dbg !23 + %199 = bitcast i32 %198 to <2 x bfloat>, !dbg !23 + %200 = extractvalue { i32, i32, i32, i32 } %193, 3, !dbg !23 + %201 = bitcast i32 %200 to <2 x bfloat>, !dbg !23 + %202 = extractelement <2 x bfloat> %195, i64 0, !dbg !23 + %203 = extractelement <2 x bfloat> %195, i64 1, !dbg !23 + %204 = extractelement <2 x bfloat> %197, i64 0, !dbg !23 + %205 = extractelement <2 x bfloat> %197, i64 1, !dbg !23 + %206 = extractelement <2 x bfloat> %199, i64 0, !dbg !23 + %207 = extractelement <2 x bfloat> %199, i64 1, !dbg !23 + %208 = extractelement <2 x bfloat> %201, i64 0, !dbg !23 + %209 = extractelement <2 x bfloat> %201, i64 1, !dbg !23 + %210 = fpext bfloat %83 to float, !dbg !24 + %211 = fpext bfloat %84 to float, !dbg !24 + %212 = fpext bfloat %85 to float, !dbg !24 + %213 = fpext bfloat %86 to float, !dbg !24 + %214 = fpext bfloat %87 to float, !dbg !24 + %215 = fpext bfloat %88 to float, !dbg !24 + %216 = fpext bfloat %89 to float, !dbg !24 + %217 = fpext bfloat %90 to float, !dbg !24 + %218 = fpext bfloat %100 to float, !dbg !24 + %219 = fpext bfloat %101 to float, !dbg !24 + %220 = fpext bfloat %102 to float, !dbg !24 + %221 = fpext bfloat %103 to float, !dbg !24 + %222 = fpext bfloat %104 to float, !dbg !24 + %223 = fpext bfloat %105 to float, !dbg !24 + %224 = fpext bfloat %106 to float, !dbg !24 + %225 = fpext bfloat %107 to float, !dbg !24 + %226 = fpext bfloat %117 to float, !dbg !24 + %227 = fpext bfloat %118 to float, !dbg !24 + %228 = fpext bfloat %119 to float, !dbg !24 + %229 = fpext bfloat %120 to float, !dbg !24 + %230 = fpext bfloat %121 to float, !dbg !24 + %231 = fpext bfloat %122 to float, !dbg !24 + %232 = fpext bfloat %123 to float, !dbg !24 + %233 = fpext bfloat %124 to float, !dbg !24 + %234 = fpext bfloat %134 to float, !dbg !24 + %235 = fpext bfloat %135 to float, !dbg !24 + %236 = fpext bfloat %136 to float, !dbg !24 + %237 = fpext bfloat %137 to float, !dbg !24 + %238 = fpext bfloat %138 to float, !dbg !24 + %239 = fpext bfloat %139 to float, !dbg !24 + %240 = fpext bfloat %140 to float, !dbg !24 + %241 = fpext bfloat %141 to float, !dbg !24 + %242 = fpext bfloat %151 to float, !dbg !24 + %243 = fpext bfloat %152 to float, !dbg !24 + %244 = fpext bfloat %153 to float, !dbg !24 + %245 = fpext bfloat %154 to float, !dbg !24 + %246 = fpext bfloat %155 to float, !dbg !24 + %247 = fpext bfloat %156 to float, !dbg !24 + %248 = fpext bfloat %157 to float, !dbg !24 + %249 = fpext bfloat %158 to float, !dbg !24 + %250 = fpext bfloat %168 to float, !dbg !24 + %251 = fpext bfloat %169 to float, !dbg !24 + %252 = fpext bfloat %170 to float, !dbg !24 + %253 = fpext bfloat %171 to float, !dbg !24 + %254 = fpext bfloat %172 to float, !dbg !24 + %255 = fpext bfloat %173 to float, !dbg !24 + %256 = fpext bfloat %174 to float, !dbg !24 + %257 = fpext bfloat %175 to float, !dbg !24 + %258 = fpext bfloat %185 to float, !dbg !24 + %259 = fpext bfloat %186 to float, !dbg !24 + %260 = fpext bfloat %187 to float, !dbg !24 + %261 = fpext bfloat %188 to float, !dbg !24 + %262 = fpext bfloat %189 to float, !dbg !24 + %263 = fpext bfloat %190 to float, !dbg !24 + %264 = fpext bfloat %191 to float, !dbg !24 + %265 = fpext bfloat %192 to float, !dbg !24 + %266 = fpext bfloat %202 to float, !dbg !24 + %267 = fpext bfloat %203 to float, !dbg !24 + %268 = fpext bfloat %204 to float, !dbg !24 + %269 = fpext bfloat %205 to float, !dbg !24 + %270 = fpext bfloat %206 to float, !dbg !24 + %271 = fpext bfloat %207 to float, !dbg !24 + %272 = fpext bfloat %208 to float, !dbg !24 + %273 = fpext bfloat %209 to float, !dbg !24 + %274 = fmul float %210, %210, !dbg !25 + %275 = fmul float %211, %211, !dbg !25 + %276 = fmul float %212, %212, !dbg !25 + %277 = fmul float %213, %213, !dbg !25 + %278 = fmul float %214, %214, !dbg !25 + %279 = fmul float %215, %215, !dbg !25 + %280 = fmul float %216, %216, !dbg !25 + %281 = fmul float %217, %217, !dbg !25 + %282 = fmul float %218, %218, !dbg !25 + %283 = fmul float %219, %219, !dbg !25 + %284 = fmul float %220, %220, !dbg !25 + %285 = fmul float %221, %221, !dbg !25 + %286 = fmul float %222, %222, !dbg !25 + %287 = fmul float %223, %223, !dbg !25 + %288 = fmul float %224, %224, !dbg !25 + %289 = fmul float %225, %225, !dbg !25 + %290 = fmul float %226, %226, !dbg !25 + %291 = fmul float %227, %227, !dbg !25 + %292 = fmul float %228, %228, !dbg !25 + %293 = fmul float %229, %229, !dbg !25 + %294 = fmul float %230, %230, !dbg !25 + %295 = fmul float %231, %231, !dbg !25 + %296 = fmul float %232, %232, !dbg !25 + %297 = fmul float %233, %233, !dbg !25 + %298 = fmul float %234, %234, !dbg !25 + %299 = fmul float %235, %235, !dbg !25 + %300 = fmul float %236, %236, !dbg !25 + %301 = fmul float %237, %237, !dbg !25 + %302 = fmul float %238, %238, !dbg !25 + %303 = fmul float %239, %239, !dbg !25 + %304 = fmul float %240, %240, !dbg !25 + %305 = fmul float %241, %241, !dbg !25 + %306 = fmul float %242, %242, !dbg !25 + %307 = fmul float %243, %243, !dbg !25 + %308 = fmul float %244, %244, !dbg !25 + %309 = fmul float %245, %245, !dbg !25 + %310 = fmul float %246, %246, !dbg !25 + %311 = fmul float %247, %247, !dbg !25 + %312 = fmul float %248, %248, !dbg !25 + %313 = fmul float %249, %249, !dbg !25 + %314 = fmul float %250, %250, !dbg !25 + %315 = fmul float %251, %251, !dbg !25 + %316 = fmul float %252, %252, !dbg !25 + %317 = fmul float %253, %253, !dbg !25 + %318 = fmul float %254, %254, !dbg !25 + %319 = fmul float %255, %255, !dbg !25 + %320 = fmul float %256, %256, !dbg !25 + %321 = fmul float %257, %257, !dbg !25 + %322 = fmul float %258, %258, !dbg !25 + %323 = fmul float %259, %259, !dbg !25 + %324 = fmul float %260, %260, !dbg !25 + %325 = fmul float %261, %261, !dbg !25 + %326 = fmul float %262, %262, !dbg !25 + %327 = fmul float %263, %263, !dbg !25 + %328 = fmul float %264, %264, !dbg !25 + %329 = fmul float %265, %265, !dbg !25 + %330 = fmul float %266, %266, !dbg !25 + %331 = fmul float %267, %267, !dbg !25 + %332 = fmul float %268, %268, !dbg !25 + %333 = fmul float %269, %269, !dbg !25 + %334 = fmul float %270, %270, !dbg !25 + %335 = fmul float %271, %271, !dbg !25 + %336 = fmul float %272, %272, !dbg !25 + %337 = fmul float %273, %273, !dbg !25 + %338 = fadd float %274, %275, !dbg !26 + %339 = fadd float %276, %338, !dbg !26 + %340 = fadd float %277, %339, !dbg !26 + %341 = fadd float %278, %340, !dbg !26 + %342 = fadd float %279, %341, !dbg !26 + %343 = fadd float %280, %342, !dbg !26 + %344 = fadd float %281, %343, !dbg !26 + %345 = fadd float %282, %283, !dbg !26 + %346 = fadd float %284, %345, !dbg !26 + %347 = fadd float %285, %346, !dbg !26 + %348 = fadd float %286, %347, !dbg !26 + %349 = fadd float %287, %348, !dbg !26 + %350 = fadd float %288, %349, !dbg !26 + %351 = fadd float %289, %350, !dbg !26 + %352 = fadd float %290, %291, !dbg !26 + %353 = fadd float %292, %352, !dbg !26 + %354 = fadd float %293, %353, !dbg !26 + %355 = fadd float %294, %354, !dbg !26 + %356 = fadd float %295, %355, !dbg !26 + %357 = fadd float %296, %356, !dbg !26 + %358 = fadd float %297, %357, !dbg !26 + %359 = fadd float %298, %299, !dbg !26 + %360 = fadd float %300, %359, !dbg !26 + %361 = fadd float %301, %360, !dbg !26 + %362 = fadd float %302, %361, !dbg !26 + %363 = fadd float %303, %362, !dbg !26 + %364 = fadd float %304, %363, !dbg !26 + %365 = fadd float %305, %364, !dbg !26 + %366 = fadd float %306, %307, !dbg !26 + %367 = fadd float %308, %366, !dbg !26 + %368 = fadd float %309, %367, !dbg !26 + %369 = fadd float %310, %368, !dbg !26 + %370 = fadd float %311, %369, !dbg !26 + %371 = fadd float %312, %370, !dbg !26 + %372 = fadd float %313, %371, !dbg !26 + %373 = fadd float %314, %315, !dbg !26 + %374 = fadd float %316, %373, !dbg !26 + %375 = fadd float %317, %374, !dbg !26 + %376 = fadd float %318, %375, !dbg !26 + %377 = fadd float %319, %376, !dbg !26 + %378 = fadd float %320, %377, !dbg !26 + %379 = fadd float %321, %378, !dbg !26 + %380 = fadd float %322, %323, !dbg !26 + %381 = fadd float %324, %380, !dbg !26 + %382 = fadd float %325, %381, !dbg !26 + %383 = fadd float %326, %382, !dbg !26 + %384 = fadd float %327, %383, !dbg !26 + %385 = fadd float %328, %384, !dbg !26 + %386 = fadd float %329, %385, !dbg !26 + %387 = fadd float %330, %331, !dbg !26 + %388 = fadd float %332, %387, !dbg !26 + %389 = fadd float %333, %388, !dbg !26 + %390 = fadd float %334, %389, !dbg !26 + %391 = fadd float %335, %390, !dbg !26 + %392 = fadd float %336, %391, !dbg !26 + %393 = fadd float %337, %392, !dbg !26 + %394 = bitcast float %344 to i32, !dbg !29 + %395 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %394, i32 8, i32 31), !dbg !29 + %396 = bitcast i32 %395 to float, !dbg !29 + %397 = fadd float %344, %396, !dbg !26 + %398 = bitcast float %397 to i32, !dbg !29 + %399 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %398, i32 4, i32 31), !dbg !29 + %400 = bitcast i32 %399 to float, !dbg !29 + %401 = fadd float %397, %400, !dbg !26 + %402 = bitcast float %401 to i32, !dbg !29 + %403 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %402, i32 2, i32 31), !dbg !29 + %404 = bitcast i32 %403 to float, !dbg !29 + %405 = fadd float %401, %404, !dbg !26 + %406 = bitcast float %405 to i32, !dbg !29 + %407 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %406, i32 1, i32 31), !dbg !29 + %408 = bitcast i32 %407 to float, !dbg !29 + %409 = fadd float %405, %408, !dbg !26 + %410 = bitcast float %351 to i32, !dbg !29 + %411 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %410, i32 8, i32 31), !dbg !29 + %412 = bitcast i32 %411 to float, !dbg !29 + %413 = fadd float %351, %412, !dbg !26 + %414 = bitcast float %413 to i32, !dbg !29 + %415 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %414, i32 4, i32 31), !dbg !29 + %416 = bitcast i32 %415 to float, !dbg !29 + %417 = fadd float %413, %416, !dbg !26 + %418 = bitcast float %417 to i32, !dbg !29 + %419 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %418, i32 2, i32 31), !dbg !29 + %420 = bitcast i32 %419 to float, !dbg !29 + %421 = fadd float %417, %420, !dbg !26 + %422 = bitcast float %421 to i32, !dbg !29 + %423 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %422, i32 1, i32 31), !dbg !29 + %424 = bitcast i32 %423 to float, !dbg !29 + %425 = fadd float %421, %424, !dbg !26 + %426 = bitcast float %358 to i32, !dbg !29 + %427 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %426, i32 8, i32 31), !dbg !29 + %428 = bitcast i32 %427 to float, !dbg !29 + %429 = fadd float %358, %428, !dbg !26 + %430 = bitcast float %429 to i32, !dbg !29 + %431 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %430, i32 4, i32 31), !dbg !29 + %432 = bitcast i32 %431 to float, !dbg !29 + %433 = fadd float %429, %432, !dbg !26 + %434 = bitcast float %433 to i32, !dbg !29 + %435 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %434, i32 2, i32 31), !dbg !29 + %436 = bitcast i32 %435 to float, !dbg !29 + %437 = fadd float %433, %436, !dbg !26 + %438 = bitcast float %437 to i32, !dbg !29 + %439 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %438, i32 1, i32 31), !dbg !29 + %440 = bitcast i32 %439 to float, !dbg !29 + %441 = fadd float %437, %440, !dbg !26 + %442 = bitcast float %365 to i32, !dbg !29 + %443 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %442, i32 8, i32 31), !dbg !29 + %444 = bitcast i32 %443 to float, !dbg !29 + %445 = fadd float %365, %444, !dbg !26 + %446 = bitcast float %445 to i32, !dbg !29 + %447 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %446, i32 4, i32 31), !dbg !29 + %448 = bitcast i32 %447 to float, !dbg !29 + %449 = fadd float %445, %448, !dbg !26 + %450 = bitcast float %449 to i32, !dbg !29 + %451 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %450, i32 2, i32 31), !dbg !29 + %452 = bitcast i32 %451 to float, !dbg !29 + %453 = fadd float %449, %452, !dbg !26 + %454 = bitcast float %453 to i32, !dbg !29 + %455 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %454, i32 1, i32 31), !dbg !29 + %456 = bitcast i32 %455 to float, !dbg !29 + %457 = fadd float %453, %456, !dbg !26 + %458 = bitcast float %372 to i32, !dbg !29 + %459 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %458, i32 8, i32 31), !dbg !29 + %460 = bitcast i32 %459 to float, !dbg !29 + %461 = fadd float %372, %460, !dbg !26 + %462 = bitcast float %461 to i32, !dbg !29 + %463 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %462, i32 4, i32 31), !dbg !29 + %464 = bitcast i32 %463 to float, !dbg !29 + %465 = fadd float %461, %464, !dbg !26 + %466 = bitcast float %465 to i32, !dbg !29 + %467 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %466, i32 2, i32 31), !dbg !29 + %468 = bitcast i32 %467 to float, !dbg !29 + %469 = fadd float %465, %468, !dbg !26 + %470 = bitcast float %469 to i32, !dbg !29 + %471 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %470, i32 1, i32 31), !dbg !29 + %472 = bitcast i32 %471 to float, !dbg !29 + %473 = fadd float %469, %472, !dbg !26 + %474 = bitcast float %379 to i32, !dbg !29 + %475 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %474, i32 8, i32 31), !dbg !29 + %476 = bitcast i32 %475 to float, !dbg !29 + %477 = fadd float %379, %476, !dbg !26 + %478 = bitcast float %477 to i32, !dbg !29 + %479 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %478, i32 4, i32 31), !dbg !29 + %480 = bitcast i32 %479 to float, !dbg !29 + %481 = fadd float %477, %480, !dbg !26 + %482 = bitcast float %481 to i32, !dbg !29 + %483 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %482, i32 2, i32 31), !dbg !29 + %484 = bitcast i32 %483 to float, !dbg !29 + %485 = fadd float %481, %484, !dbg !26 + %486 = bitcast float %485 to i32, !dbg !29 + %487 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %486, i32 1, i32 31), !dbg !29 + %488 = bitcast i32 %487 to float, !dbg !29 + %489 = fadd float %485, %488, !dbg !26 + %490 = bitcast float %386 to i32, !dbg !29 + %491 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %490, i32 8, i32 31), !dbg !29 + %492 = bitcast i32 %491 to float, !dbg !29 + %493 = fadd float %386, %492, !dbg !26 + %494 = bitcast float %493 to i32, !dbg !29 + %495 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %494, i32 4, i32 31), !dbg !29 + %496 = bitcast i32 %495 to float, !dbg !29 + %497 = fadd float %493, %496, !dbg !26 + %498 = bitcast float %497 to i32, !dbg !29 + %499 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %498, i32 2, i32 31), !dbg !29 + %500 = bitcast i32 %499 to float, !dbg !29 + %501 = fadd float %497, %500, !dbg !26 + %502 = bitcast float %501 to i32, !dbg !29 + %503 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %502, i32 1, i32 31), !dbg !29 + %504 = bitcast i32 %503 to float, !dbg !29 + %505 = fadd float %501, %504, !dbg !26 + %506 = bitcast float %393 to i32, !dbg !29 + %507 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %506, i32 8, i32 31), !dbg !29 + %508 = bitcast i32 %507 to float, !dbg !29 + %509 = fadd float %393, %508, !dbg !26 + %510 = bitcast float %509 to i32, !dbg !29 + %511 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %510, i32 4, i32 31), !dbg !29 + %512 = bitcast i32 %511 to float, !dbg !29 + %513 = fadd float %509, %512, !dbg !26 + %514 = bitcast float %513 to i32, !dbg !29 + %515 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %514, i32 2, i32 31), !dbg !29 + %516 = bitcast i32 %515 to float, !dbg !29 + %517 = fadd float %513, %516, !dbg !26 + %518 = bitcast float %517 to i32, !dbg !29 + %519 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %518, i32 1, i32 31), !dbg !29 + %520 = bitcast i32 %519 to float, !dbg !29 + %521 = fadd float %517, %520, !dbg !26 + %522 = fadd float %3, %409, !dbg !32 + %523 = fadd float %3, %425, !dbg !32 + %524 = fadd float %3, %441, !dbg !32 + %525 = fadd float %3, %457, !dbg !32 + %526 = fadd float %3, %473, !dbg !32 + %527 = fadd float %3, %489, !dbg !32 + %528 = fadd float %3, %505, !dbg !32 + %529 = fadd float %3, %521, !dbg !32 + %530 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i = icmp eq i32 %530, 0, !dbg !33 + %531 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i = icmp eq i32 %531, 0, !dbg !33 + br i1 %.not.i, label %537, label %532, !dbg !33 + +532: ; preds = %7 + br i1 %.not1.i, label %535, label %533, !dbg !33 + +533: ; preds = %532 + %534 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %522) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +535: ; preds = %532 + %536 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %522) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +537: ; preds = %7 + br i1 %.not1.i, label %540, label %538, !dbg !33 + +538: ; preds = %537 + %539 = tail call float @llvm.nvvm.sqrt.rn.f(float %522) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +540: ; preds = %537 + %541 = tail call float @llvm.nvvm.sqrt.approx.f(float %522) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +__nv_sqrtf.exit: ; preds = %533, %535, %538, %540 + %.0.i = phi float [ %534, %533 ], [ %536, %535 ], [ %539, %538 ], [ %541, %540 ], !dbg !33 + %542 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i1 = icmp eq i32 %542, 0, !dbg !33 + %543 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i4 = icmp eq i32 %543, 0, !dbg !33 + br i1 %.not.i1, label %549, label %544, !dbg !33 + +544: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %547, label %545, !dbg !33 + +545: ; preds = %544 + %546 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %523) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +547: ; preds = %544 + %548 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %523) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +549: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %552, label %550, !dbg !33 + +550: ; preds = %549 + %551 = tail call float @llvm.nvvm.sqrt.rn.f(float %523) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +552: ; preds = %549 + %553 = tail call float @llvm.nvvm.sqrt.approx.f(float %523) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +__nv_sqrtf.exit5: ; preds = %545, %547, %550, %552 + %.0.i3 = phi float [ %546, %545 ], [ %548, %547 ], [ %551, %550 ], [ %553, %552 ], !dbg !33 + %554 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i6 = icmp eq i32 %554, 0, !dbg !33 + %555 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i9 = icmp eq i32 %555, 0, !dbg !33 + br i1 %.not.i6, label %561, label %556, !dbg !33 + +556: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %559, label %557, !dbg !33 + +557: ; preds = %556 + %558 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %524) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +559: ; preds = %556 + %560 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %524) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +561: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %564, label %562, !dbg !33 + +562: ; preds = %561 + %563 = tail call float @llvm.nvvm.sqrt.rn.f(float %524) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +564: ; preds = %561 + %565 = tail call float @llvm.nvvm.sqrt.approx.f(float %524) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +__nv_sqrtf.exit10: ; preds = %557, %559, %562, %564 + %.0.i8 = phi float [ %558, %557 ], [ %560, %559 ], [ %563, %562 ], [ %565, %564 ], !dbg !33 + %566 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i11 = icmp eq i32 %566, 0, !dbg !33 + %567 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i14 = icmp eq i32 %567, 0, !dbg !33 + br i1 %.not.i11, label %573, label %568, !dbg !33 + +568: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %571, label %569, !dbg !33 + +569: ; preds = %568 + %570 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %525) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +571: ; preds = %568 + %572 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %525) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +573: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %576, label %574, !dbg !33 + +574: ; preds = %573 + %575 = tail call float @llvm.nvvm.sqrt.rn.f(float %525) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +576: ; preds = %573 + %577 = tail call float @llvm.nvvm.sqrt.approx.f(float %525) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +__nv_sqrtf.exit15: ; preds = %569, %571, %574, %576 + %.0.i13 = phi float [ %570, %569 ], [ %572, %571 ], [ %575, %574 ], [ %577, %576 ], !dbg !33 + %578 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i16 = icmp eq i32 %578, 0, !dbg !33 + %579 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i19 = icmp eq i32 %579, 0, !dbg !33 + br i1 %.not.i16, label %585, label %580, !dbg !33 + +580: ; preds = %__nv_sqrtf.exit15 + br i1 %.not1.i19, label %583, label %581, !dbg !33 + +581: ; preds = %580 + %582 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %526) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +583: ; preds = %580 + %584 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %526) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +585: ; preds = %__nv_sqrtf.exit15 + br i1 %.not1.i19, label %588, label %586, !dbg !33 + +586: ; preds = %585 + %587 = tail call float @llvm.nvvm.sqrt.rn.f(float %526) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +588: ; preds = %585 + %589 = tail call float @llvm.nvvm.sqrt.approx.f(float %526) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +__nv_sqrtf.exit20: ; preds = %581, %583, %586, %588 + %.0.i18 = phi float [ %582, %581 ], [ %584, %583 ], [ %587, %586 ], [ %589, %588 ], !dbg !33 + %590 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i21 = icmp eq i32 %590, 0, !dbg !33 + %591 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i24 = icmp eq i32 %591, 0, !dbg !33 + br i1 %.not.i21, label %597, label %592, !dbg !33 + +592: ; preds = %__nv_sqrtf.exit20 + br i1 %.not1.i24, label %595, label %593, !dbg !33 + +593: ; preds = %592 + %594 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %527) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +595: ; preds = %592 + %596 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %527) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +597: ; preds = %__nv_sqrtf.exit20 + br i1 %.not1.i24, label %600, label %598, !dbg !33 + +598: ; preds = %597 + %599 = tail call float @llvm.nvvm.sqrt.rn.f(float %527) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +600: ; preds = %597 + %601 = tail call float @llvm.nvvm.sqrt.approx.f(float %527) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +__nv_sqrtf.exit25: ; preds = %593, %595, %598, %600 + %.0.i23 = phi float [ %594, %593 ], [ %596, %595 ], [ %599, %598 ], [ %601, %600 ], !dbg !33 + %602 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i26 = icmp eq i32 %602, 0, !dbg !33 + %603 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i29 = icmp eq i32 %603, 0, !dbg !33 + br i1 %.not.i26, label %609, label %604, !dbg !33 + +604: ; preds = %__nv_sqrtf.exit25 + br i1 %.not1.i29, label %607, label %605, !dbg !33 + +605: ; preds = %604 + %606 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %528) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +607: ; preds = %604 + %608 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %528) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +609: ; preds = %__nv_sqrtf.exit25 + br i1 %.not1.i29, label %612, label %610, !dbg !33 + +610: ; preds = %609 + %611 = tail call float @llvm.nvvm.sqrt.rn.f(float %528) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +612: ; preds = %609 + %613 = tail call float @llvm.nvvm.sqrt.approx.f(float %528) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +__nv_sqrtf.exit30: ; preds = %605, %607, %610, %612 + %.0.i28 = phi float [ %606, %605 ], [ %608, %607 ], [ %611, %610 ], [ %613, %612 ], !dbg !33 + %614 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i31 = icmp eq i32 %614, 0, !dbg !33 + %615 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i34 = icmp eq i32 %615, 0, !dbg !33 + br i1 %.not.i31, label %621, label %616, !dbg !33 + +616: ; preds = %__nv_sqrtf.exit30 + br i1 %.not1.i34, label %619, label %617, !dbg !33 + +617: ; preds = %616 + %618 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %529) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +619: ; preds = %616 + %620 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %529) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +621: ; preds = %__nv_sqrtf.exit30 + br i1 %.not1.i34, label %624, label %622, !dbg !33 + +622: ; preds = %621 + %623 = tail call float @llvm.nvvm.sqrt.rn.f(float %529) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +624: ; preds = %621 + %625 = tail call float @llvm.nvvm.sqrt.approx.f(float %529) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +__nv_sqrtf.exit35: ; preds = %617, %619, %622, %624 + %.0.i33 = phi float [ %618, %617 ], [ %620, %619 ], [ %623, %622 ], [ %625, %624 ], !dbg !33 + %626 = zext nneg i32 %22 to i64, !dbg !17 + %627 = or disjoint i64 %10, %626, !dbg !17 + %628 = getelementptr float, ptr addrspace(1) %2, i64 %627, !dbg !34 + %629 = getelementptr bfloat, ptr addrspace(1) %1, i64 %56, !dbg !35 + %630 = getelementptr bfloat, ptr addrspace(1) %629, i64 %65, !dbg !36 + %631 = getelementptr bfloat, ptr addrspace(1) %1, i64 %55, !dbg !35 + %632 = getelementptr bfloat, ptr addrspace(1) %631, i64 %65, !dbg !36 + %633 = getelementptr bfloat, ptr addrspace(1) %1, i64 %54, !dbg !35 + %634 = getelementptr bfloat, ptr addrspace(1) %633, i64 %65, !dbg !36 + %635 = getelementptr bfloat, ptr addrspace(1) %1, i64 %53, !dbg !35 + %636 = getelementptr bfloat, ptr addrspace(1) %635, i64 %65, !dbg !36 + %637 = getelementptr bfloat, ptr addrspace(1) %1, i64 %52, !dbg !35 + %638 = getelementptr bfloat, ptr addrspace(1) %637, i64 %65, !dbg !36 + %639 = getelementptr bfloat, ptr addrspace(1) %1, i64 %51, !dbg !35 + %640 = getelementptr bfloat, ptr addrspace(1) %639, i64 %65, !dbg !36 + %641 = getelementptr bfloat, ptr addrspace(1) %1, i64 %50, !dbg !35 + %642 = getelementptr bfloat, ptr addrspace(1) %641, i64 %65, !dbg !36 + %643 = getelementptr bfloat, ptr addrspace(1) %1, i64 %49, !dbg !35 + %644 = getelementptr bfloat, ptr addrspace(1) %643, i64 %65, !dbg !36 + %645 = icmp slt i64 %627, %39, !dbg !18 + %646 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i), !dbg !37 + %647 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i3), !dbg !37 + %648 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i8), !dbg !37 + %649 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i13), !dbg !37 + %650 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i18), !dbg !37 + %651 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i23), !dbg !37 + %652 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i28), !dbg !37 + %653 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i33), !dbg !37 + %654 = fmul float %646, %210, !dbg !38 + %655 = fmul float %646, %211, !dbg !38 + %656 = fmul float %646, %212, !dbg !38 + %657 = fmul float %646, %213, !dbg !38 + %658 = fmul float %646, %214, !dbg !38 + %659 = fmul float %646, %215, !dbg !38 + %660 = fmul float %646, %216, !dbg !38 + %661 = fmul float %646, %217, !dbg !38 + %662 = fmul float %647, %218, !dbg !38 + %663 = fmul float %647, %219, !dbg !38 + %664 = fmul float %647, %220, !dbg !38 + %665 = fmul float %647, %221, !dbg !38 + %666 = fmul float %647, %222, !dbg !38 + %667 = fmul float %647, %223, !dbg !38 + %668 = fmul float %647, %224, !dbg !38 + %669 = fmul float %647, %225, !dbg !38 + %670 = fmul float %648, %226, !dbg !38 + %671 = fmul float %648, %227, !dbg !38 + %672 = fmul float %648, %228, !dbg !38 + %673 = fmul float %648, %229, !dbg !38 + %674 = fmul float %648, %230, !dbg !38 + %675 = fmul float %648, %231, !dbg !38 + %676 = fmul float %648, %232, !dbg !38 + %677 = fmul float %648, %233, !dbg !38 + %678 = fmul float %649, %234, !dbg !38 + %679 = fmul float %649, %235, !dbg !38 + %680 = fmul float %649, %236, !dbg !38 + %681 = fmul float %649, %237, !dbg !38 + %682 = fmul float %649, %238, !dbg !38 + %683 = fmul float %649, %239, !dbg !38 + %684 = fmul float %649, %240, !dbg !38 + %685 = fmul float %649, %241, !dbg !38 + %686 = fmul float %650, %242, !dbg !38 + %687 = fmul float %650, %243, !dbg !38 + %688 = fmul float %650, %244, !dbg !38 + %689 = fmul float %650, %245, !dbg !38 + %690 = fmul float %650, %246, !dbg !38 + %691 = fmul float %650, %247, !dbg !38 + %692 = fmul float %650, %248, !dbg !38 + %693 = fmul float %650, %249, !dbg !38 + %694 = fmul float %651, %250, !dbg !38 + %695 = fmul float %651, %251, !dbg !38 + %696 = fmul float %651, %252, !dbg !38 + %697 = fmul float %651, %253, !dbg !38 + %698 = fmul float %651, %254, !dbg !38 + %699 = fmul float %651, %255, !dbg !38 + %700 = fmul float %651, %256, !dbg !38 + %701 = fmul float %651, %257, !dbg !38 + %702 = fmul float %652, %258, !dbg !38 + %703 = fmul float %652, %259, !dbg !38 + %704 = fmul float %652, %260, !dbg !38 + %705 = fmul float %652, %261, !dbg !38 + %706 = fmul float %652, %262, !dbg !38 + %707 = fmul float %652, %263, !dbg !38 + %708 = fmul float %652, %264, !dbg !38 + %709 = fmul float %652, %265, !dbg !38 + %710 = fmul float %653, %266, !dbg !38 + %711 = fmul float %653, %267, !dbg !38 + %712 = fmul float %653, %268, !dbg !38 + %713 = fmul float %653, %269, !dbg !38 + %714 = fmul float %653, %270, !dbg !38 + %715 = fmul float %653, %271, !dbg !38 + %716 = fmul float %653, %272, !dbg !38 + %717 = fmul float %653, %273, !dbg !38 + %718 = fptrunc float %654 to bfloat, !dbg !39 + %719 = fptrunc float %655 to bfloat, !dbg !39 + %720 = fptrunc float %656 to bfloat, !dbg !39 + %721 = fptrunc float %657 to bfloat, !dbg !39 + %722 = fptrunc float %658 to bfloat, !dbg !39 + %723 = fptrunc float %659 to bfloat, !dbg !39 + %724 = fptrunc float %660 to bfloat, !dbg !39 + %725 = fptrunc float %661 to bfloat, !dbg !39 + %726 = fptrunc float %662 to bfloat, !dbg !39 + %727 = fptrunc float %663 to bfloat, !dbg !39 + %728 = fptrunc float %664 to bfloat, !dbg !39 + %729 = fptrunc float %665 to bfloat, !dbg !39 + %730 = fptrunc float %666 to bfloat, !dbg !39 + %731 = fptrunc float %667 to bfloat, !dbg !39 + %732 = fptrunc float %668 to bfloat, !dbg !39 + %733 = fptrunc float %669 to bfloat, !dbg !39 + %734 = fptrunc float %670 to bfloat, !dbg !39 + %735 = fptrunc float %671 to bfloat, !dbg !39 + %736 = fptrunc float %672 to bfloat, !dbg !39 + %737 = fptrunc float %673 to bfloat, !dbg !39 + %738 = fptrunc float %674 to bfloat, !dbg !39 + %739 = fptrunc float %675 to bfloat, !dbg !39 + %740 = fptrunc float %676 to bfloat, !dbg !39 + %741 = fptrunc float %677 to bfloat, !dbg !39 + %742 = fptrunc float %678 to bfloat, !dbg !39 + %743 = fptrunc float %679 to bfloat, !dbg !39 + %744 = fptrunc float %680 to bfloat, !dbg !39 + %745 = fptrunc float %681 to bfloat, !dbg !39 + %746 = fptrunc float %682 to bfloat, !dbg !39 + %747 = fptrunc float %683 to bfloat, !dbg !39 + %748 = fptrunc float %684 to bfloat, !dbg !39 + %749 = fptrunc float %685 to bfloat, !dbg !39 + %750 = fptrunc float %686 to bfloat, !dbg !39 + %751 = fptrunc float %687 to bfloat, !dbg !39 + %752 = fptrunc float %688 to bfloat, !dbg !39 + %753 = fptrunc float %689 to bfloat, !dbg !39 + %754 = fptrunc float %690 to bfloat, !dbg !39 + %755 = fptrunc float %691 to bfloat, !dbg !39 + %756 = fptrunc float %692 to bfloat, !dbg !39 + %757 = fptrunc float %693 to bfloat, !dbg !39 + %758 = fptrunc float %694 to bfloat, !dbg !39 + %759 = fptrunc float %695 to bfloat, !dbg !39 + %760 = fptrunc float %696 to bfloat, !dbg !39 + %761 = fptrunc float %697 to bfloat, !dbg !39 + %762 = fptrunc float %698 to bfloat, !dbg !39 + %763 = fptrunc float %699 to bfloat, !dbg !39 + %764 = fptrunc float %700 to bfloat, !dbg !39 + %765 = fptrunc float %701 to bfloat, !dbg !39 + %766 = fptrunc float %702 to bfloat, !dbg !39 + %767 = fptrunc float %703 to bfloat, !dbg !39 + %768 = fptrunc float %704 to bfloat, !dbg !39 + %769 = fptrunc float %705 to bfloat, !dbg !39 + %770 = fptrunc float %706 to bfloat, !dbg !39 + %771 = fptrunc float %707 to bfloat, !dbg !39 + %772 = fptrunc float %708 to bfloat, !dbg !39 + %773 = fptrunc float %709 to bfloat, !dbg !39 + %774 = fptrunc float %710 to bfloat, !dbg !39 + %775 = fptrunc float %711 to bfloat, !dbg !39 + %776 = fptrunc float %712 to bfloat, !dbg !39 + %777 = fptrunc float %713 to bfloat, !dbg !39 + %778 = fptrunc float %714 to bfloat, !dbg !39 + %779 = fptrunc float %715 to bfloat, !dbg !39 + %780 = fptrunc float %716 to bfloat, !dbg !39 + %781 = fptrunc float %717 to bfloat, !dbg !39 + %782 = insertelement <2 x bfloat> poison, bfloat %718, i64 0, !dbg !40 + %783 = insertelement <2 x bfloat> %782, bfloat %719, i64 1, !dbg !40 + %784 = bitcast <2 x bfloat> %783 to i32, !dbg !40 + %785 = insertelement <2 x bfloat> poison, bfloat %720, i64 0, !dbg !40 + %786 = insertelement <2 x bfloat> %785, bfloat %721, i64 1, !dbg !40 + %787 = bitcast <2 x bfloat> %786 to i32, !dbg !40 + %788 = insertelement <2 x bfloat> poison, bfloat %722, i64 0, !dbg !40 + %789 = insertelement <2 x bfloat> %788, bfloat %723, i64 1, !dbg !40 + %790 = bitcast <2 x bfloat> %789 to i32, !dbg !40 + %791 = insertelement <2 x bfloat> poison, bfloat %724, i64 0, !dbg !40 + %792 = insertelement <2 x bfloat> %791, bfloat %725, i64 1, !dbg !40 + %793 = bitcast <2 x bfloat> %792 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %784, i32 %787, i32 %790, i32 %793, ptr addrspace(1) %644, i1 %40) #6, !dbg !40 + %794 = insertelement <2 x bfloat> poison, bfloat %726, i64 0, !dbg !40 + %795 = insertelement <2 x bfloat> %794, bfloat %727, i64 1, !dbg !40 + %796 = bitcast <2 x bfloat> %795 to i32, !dbg !40 + %797 = insertelement <2 x bfloat> poison, bfloat %728, i64 0, !dbg !40 + %798 = insertelement <2 x bfloat> %797, bfloat %729, i64 1, !dbg !40 + %799 = bitcast <2 x bfloat> %798 to i32, !dbg !40 + %800 = insertelement <2 x bfloat> poison, bfloat %730, i64 0, !dbg !40 + %801 = insertelement <2 x bfloat> %800, bfloat %731, i64 1, !dbg !40 + %802 = bitcast <2 x bfloat> %801 to i32, !dbg !40 + %803 = insertelement <2 x bfloat> poison, bfloat %732, i64 0, !dbg !40 + %804 = insertelement <2 x bfloat> %803, bfloat %733, i64 1, !dbg !40 + %805 = bitcast <2 x bfloat> %804 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %796, i32 %799, i32 %802, i32 %805, ptr addrspace(1) %642, i1 %41) #6, !dbg !40 + %806 = insertelement <2 x bfloat> poison, bfloat %734, i64 0, !dbg !40 + %807 = insertelement <2 x bfloat> %806, bfloat %735, i64 1, !dbg !40 + %808 = bitcast <2 x bfloat> %807 to i32, !dbg !40 + %809 = insertelement <2 x bfloat> poison, bfloat %736, i64 0, !dbg !40 + %810 = insertelement <2 x bfloat> %809, bfloat %737, i64 1, !dbg !40 + %811 = bitcast <2 x bfloat> %810 to i32, !dbg !40 + %812 = insertelement <2 x bfloat> poison, bfloat %738, i64 0, !dbg !40 + %813 = insertelement <2 x bfloat> %812, bfloat %739, i64 1, !dbg !40 + %814 = bitcast <2 x bfloat> %813 to i32, !dbg !40 + %815 = insertelement <2 x bfloat> poison, bfloat %740, i64 0, !dbg !40 + %816 = insertelement <2 x bfloat> %815, bfloat %741, i64 1, !dbg !40 + %817 = bitcast <2 x bfloat> %816 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %808, i32 %811, i32 %814, i32 %817, ptr addrspace(1) %640, i1 %42) #6, !dbg !40 + %818 = insertelement <2 x bfloat> poison, bfloat %742, i64 0, !dbg !40 + %819 = insertelement <2 x bfloat> %818, bfloat %743, i64 1, !dbg !40 + %820 = bitcast <2 x bfloat> %819 to i32, !dbg !40 + %821 = insertelement <2 x bfloat> poison, bfloat %744, i64 0, !dbg !40 + %822 = insertelement <2 x bfloat> %821, bfloat %745, i64 1, !dbg !40 + %823 = bitcast <2 x bfloat> %822 to i32, !dbg !40 + %824 = insertelement <2 x bfloat> poison, bfloat %746, i64 0, !dbg !40 + %825 = insertelement <2 x bfloat> %824, bfloat %747, i64 1, !dbg !40 + %826 = bitcast <2 x bfloat> %825 to i32, !dbg !40 + %827 = insertelement <2 x bfloat> poison, bfloat %748, i64 0, !dbg !40 + %828 = insertelement <2 x bfloat> %827, bfloat %749, i64 1, !dbg !40 + %829 = bitcast <2 x bfloat> %828 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %820, i32 %823, i32 %826, i32 %829, ptr addrspace(1) %638, i1 %43) #6, !dbg !40 + %830 = insertelement <2 x bfloat> poison, bfloat %750, i64 0, !dbg !40 + %831 = insertelement <2 x bfloat> %830, bfloat %751, i64 1, !dbg !40 + %832 = bitcast <2 x bfloat> %831 to i32, !dbg !40 + %833 = insertelement <2 x bfloat> poison, bfloat %752, i64 0, !dbg !40 + %834 = insertelement <2 x bfloat> %833, bfloat %753, i64 1, !dbg !40 + %835 = bitcast <2 x bfloat> %834 to i32, !dbg !40 + %836 = insertelement <2 x bfloat> poison, bfloat %754, i64 0, !dbg !40 + %837 = insertelement <2 x bfloat> %836, bfloat %755, i64 1, !dbg !40 + %838 = bitcast <2 x bfloat> %837 to i32, !dbg !40 + %839 = insertelement <2 x bfloat> poison, bfloat %756, i64 0, !dbg !40 + %840 = insertelement <2 x bfloat> %839, bfloat %757, i64 1, !dbg !40 + %841 = bitcast <2 x bfloat> %840 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %832, i32 %835, i32 %838, i32 %841, ptr addrspace(1) %636, i1 %44) #6, !dbg !40 + %842 = insertelement <2 x bfloat> poison, bfloat %758, i64 0, !dbg !40 + %843 = insertelement <2 x bfloat> %842, bfloat %759, i64 1, !dbg !40 + %844 = bitcast <2 x bfloat> %843 to i32, !dbg !40 + %845 = insertelement <2 x bfloat> poison, bfloat %760, i64 0, !dbg !40 + %846 = insertelement <2 x bfloat> %845, bfloat %761, i64 1, !dbg !40 + %847 = bitcast <2 x bfloat> %846 to i32, !dbg !40 + %848 = insertelement <2 x bfloat> poison, bfloat %762, i64 0, !dbg !40 + %849 = insertelement <2 x bfloat> %848, bfloat %763, i64 1, !dbg !40 + %850 = bitcast <2 x bfloat> %849 to i32, !dbg !40 + %851 = insertelement <2 x bfloat> poison, bfloat %764, i64 0, !dbg !40 + %852 = insertelement <2 x bfloat> %851, bfloat %765, i64 1, !dbg !40 + %853 = bitcast <2 x bfloat> %852 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %844, i32 %847, i32 %850, i32 %853, ptr addrspace(1) %634, i1 %45) #6, !dbg !40 + %854 = insertelement <2 x bfloat> poison, bfloat %766, i64 0, !dbg !40 + %855 = insertelement <2 x bfloat> %854, bfloat %767, i64 1, !dbg !40 + %856 = bitcast <2 x bfloat> %855 to i32, !dbg !40 + %857 = insertelement <2 x bfloat> poison, bfloat %768, i64 0, !dbg !40 + %858 = insertelement <2 x bfloat> %857, bfloat %769, i64 1, !dbg !40 + %859 = bitcast <2 x bfloat> %858 to i32, !dbg !40 + %860 = insertelement <2 x bfloat> poison, bfloat %770, i64 0, !dbg !40 + %861 = insertelement <2 x bfloat> %860, bfloat %771, i64 1, !dbg !40 + %862 = bitcast <2 x bfloat> %861 to i32, !dbg !40 + %863 = insertelement <2 x bfloat> poison, bfloat %772, i64 0, !dbg !40 + %864 = insertelement <2 x bfloat> %863, bfloat %773, i64 1, !dbg !40 + %865 = bitcast <2 x bfloat> %864 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %856, i32 %859, i32 %862, i32 %865, ptr addrspace(1) %632, i1 %46) #6, !dbg !40 + %866 = insertelement <2 x bfloat> poison, bfloat %774, i64 0, !dbg !40 + %867 = insertelement <2 x bfloat> %866, bfloat %775, i64 1, !dbg !40 + %868 = bitcast <2 x bfloat> %867 to i32, !dbg !40 + %869 = insertelement <2 x bfloat> poison, bfloat %776, i64 0, !dbg !40 + %870 = insertelement <2 x bfloat> %869, bfloat %777, i64 1, !dbg !40 + %871 = bitcast <2 x bfloat> %870 to i32, !dbg !40 + %872 = insertelement <2 x bfloat> poison, bfloat %778, i64 0, !dbg !40 + %873 = insertelement <2 x bfloat> %872, bfloat %779, i64 1, !dbg !40 + %874 = bitcast <2 x bfloat> %873 to i32, !dbg !40 + %875 = insertelement <2 x bfloat> poison, bfloat %780, i64 0, !dbg !40 + %876 = insertelement <2 x bfloat> %875, bfloat %781, i64 1, !dbg !40 + %877 = bitcast <2 x bfloat> %876 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %868, i32 %871, i32 %874, i32 %877, ptr addrspace(1) %630, i1 %47) #6, !dbg !40 + %878 = shl nuw nsw i32 %12, 1, !dbg !41 + %879 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %878, !dbg !41 + %880 = bitcast float %646 to i32, !dbg !41 + %881 = bitcast float %647 to i32, !dbg !41 + %882 = bitcast float %648 to i32, !dbg !41 + %883 = bitcast float %649 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %879, i32 %880, i32 %881, i32 %882, i32 %883) #6, !dbg !41 + %884 = getelementptr inbounds nuw i8, ptr addrspace(3) %879, i32 16, !dbg !41 + %885 = bitcast float %650 to i32, !dbg !41 + %886 = bitcast float %651 to i32, !dbg !41 + %887 = bitcast float %652 to i32, !dbg !41 + %888 = bitcast float %653 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %884, i32 %885, i32 %886, i32 %887, i32 %888) #6, !dbg !41 + tail call void @llvm.nvvm.bar.warp.sync(i32 -1), !dbg !41 + %889 = shl nuw nsw i32 %11, 1, !dbg !41 + %890 = and i32 %889, 28, !dbg !41 + %891 = shl nuw nsw i32 %11, 5, !dbg !41 + %892 = and i32 %891, 32, !dbg !41 + %893 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %892, !dbg !41 + %894 = getelementptr inbounds nuw i8, ptr addrspace(3) %893, i32 %890, !dbg !41 + %895 = load i32, ptr addrspace(3) %894, align 4, !dbg !41 + %896 = and i1 %13, %645, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %895, ptr addrspace(1) %628, i1 %896) #6, !dbg !41 + ret void, !dbg !42 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.div.full(float, float) #3 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.bar.warp.sync(i32) #4 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #5 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.f(float) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="32" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) } +attributes #4 = { convergent nocallback nounwind } +attributes #5 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #6 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "l2norm_fwd_kernel", linkageName: "l2norm_fwd_kernel", scope: !1, file: !1, line: 69, type: !6, scopeLine: 69, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !10, !11, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 70, column: 24, scope: !5) +!14 = !DILocation(line: 70, column: 30, scope: !5) +!15 = !DILocation(line: 71, column: 16, scope: !5) +!16 = !DILocation(line: 71, column: 34, scope: !5) +!17 = !DILocation(line: 71, column: 21, scope: !5) +!18 = !DILocation(line: 73, column: 16, scope: !5) +!19 = !DILocation(line: 74, column: 30, scope: !5) +!20 = !DILocation(line: 75, column: 29, scope: !5) +!21 = !DILocation(line: 75, column: 14, scope: !5) +!22 = !DILocation(line: 75, column: 33, scope: !5) +!23 = !DILocation(line: 79, column: 18, scope: !5) +!24 = !DILocation(line: 79, column: 47, scope: !5) +!25 = !DILocation(line: 80, column: 38, scope: !5) +!26 = !DILocation(line: 263, column: 15, scope: !27, inlinedAt: !29) +!27 = distinct !DILexicalBlockFile(scope: !5, file: !28, discriminator: 0) +!28 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!29 = !DILocation(line: 293, column: 36, scope: !27, inlinedAt: !30) +!30 = !DILocation(line: 80, column: 43, scope: !31) +!31 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!32 = !DILocation(line: 80, column: 48, scope: !5) +!33 = !DILocation(line: 80, column: 25, scope: !5) +!34 = !DILocation(line: 77, column: 20, scope: !5) +!35 = !DILocation(line: 76, column: 14, scope: !5) +!36 = !DILocation(line: 76, column: 33, scope: !5) +!37 = !DILocation(line: 80, column: 17, scope: !5) +!38 = !DILocation(line: 81, column: 16, scope: !5) +!39 = !DILocation(line: 83, column: 25, scope: !5) +!40 = !DILocation(line: 83, column: 18, scope: !5) +!41 = !DILocation(line: 84, column: 21, scope: !5) +!42 = !DILocation(line: 84, column: 4, scope: !5) diff --git a/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/l2norm_fwd_kernel.ptx b/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/l2norm_fwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..3dc6440506fc443481c3d02d5f0f14fdf7edf7e8 --- /dev/null +++ b/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/l2norm_fwd_kernel.ptx @@ -0,0 +1,966 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_fwd_kernel // -- Begin function l2norm_fwd_kernel +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; +.global .align 1 .b8 _$_str_$_2[17] = {95, 95, 67, 85, 68, 65, 95, 80, 82, 69, 67, 95, 83, 81, 82, 84}; + // @l2norm_fwd_kernel +.visible .entry l2norm_fwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_2, + .param .f32 l2norm_fwd_kernel_param_3, + .param .u32 l2norm_fwd_kernel_param_4, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_6 +) +.reqntid 32 +{ + .reg .pred %p<12>; + .reg .b16 %rs<65>; + .reg .b32 %r<373>; + .reg .b64 %rd<68>; + .loc 1 69 0 // l2norm.py:69:0 +$L__func_begin0: + .loc 1 69 0 // l2norm.py:69:0 + +// %bb.0: // %__nv_sqrtf.exit + ld.param.b64 %rd18, [l2norm_fwd_kernel_param_0]; + ld.param.b64 %rd19, [l2norm_fwd_kernel_param_1]; +$L__tmp0: + .loc 1 70 24 // l2norm.py:70:24 + mov.u32 %r77, %ctaid.x; + ld.param.b64 %rd20, [l2norm_fwd_kernel_param_2]; + .loc 1 71 16 // l2norm.py:71:16 + mul.wide.u32 %rd21, %r77, 16; + ld.param.b32 %r78, [l2norm_fwd_kernel_param_3]; + ld.param.s32 %rd22, [l2norm_fwd_kernel_param_4]; + .loc 1 71 34 // l2norm.py:71:34 + mov.u32 %r79, %tid.x; + and.b32 %r80, %r79, 16; + setp.eq.b32 %p10, %r80, 0; + shr.u32 %r81, %r79, 4; + bfe.u32 %r82, %r79, 4, 1; + or.b32 %r83, %r82, 2; + or.b32 %r84, %r82, 4; + or.b32 %r85, %r82, 6; + or.b32 %r86, %r82, 8; + or.b32 %r87, %r82, 10; + or.b32 %r88, %r82, 12; + or.b32 %r89, %r82, 14; + and.b32 %r90, %r79, 15; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd23, %r81; + and.b64 %rd24, %rd23, 1; + cvt.u64.u32 %rd25, %r83; + cvt.u64.u32 %rd26, %r84; + cvt.u64.u32 %rd27, %r85; + cvt.u64.u32 %rd28, %r86; + cvt.u64.u32 %rd29, %r87; + cvt.u64.u32 %rd30, %r88; + cvt.u64.u32 %rd31, %r89; + or.b64 %rd32, %rd21, %rd24; + or.b64 %rd33, %rd21, %rd25; + or.b64 %rd34, %rd21, %rd26; + or.b64 %rd35, %rd21, %rd27; + or.b64 %rd36, %rd21, %rd28; + or.b64 %rd37, %rd21, %rd29; + or.b64 %rd38, %rd21, %rd30; + or.b64 %rd39, %rd21, %rd31; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p1, %rd32, %rd22; + setp.lt.s64 %p2, %rd33, %rd22; + setp.lt.s64 %p3, %rd34, %rd22; + setp.lt.s64 %p4, %rd35, %rd22; + setp.lt.s64 %p5, %rd36, %rd22; + setp.lt.s64 %p6, %rd37, %rd22; + setp.lt.s64 %p7, %rd38, %rd22; + setp.lt.s64 %p8, %rd39, %rd22; + .loc 1 74 30 // l2norm.py:74:30 + shl.b32 %r91, %r90, 3; + .loc 1 75 14 // l2norm.py:75:14 + shl.b64 %rd40, %rd32, 8; + add.s64 %rd41, %rd18, %rd40; + shl.b64 %rd42, %rd33, 8; + add.s64 %rd43, %rd18, %rd42; + shl.b64 %rd44, %rd34, 8; + add.s64 %rd45, %rd18, %rd44; + shl.b64 %rd46, %rd35, 8; + add.s64 %rd47, %rd18, %rd46; + shl.b64 %rd48, %rd36, 8; + add.s64 %rd49, %rd18, %rd48; + shl.b64 %rd50, %rd37, 8; + add.s64 %rd51, %rd18, %rd50; + shl.b64 %rd52, %rd38, 8; + add.s64 %rd53, %rd18, %rd52; + shl.b64 %rd54, %rd39, 8; + add.s64 %rd55, %rd18, %rd54; + .loc 1 75 33 // l2norm.py:75:33 + mul.wide.u32 %rd56, %r91, 2; + add.s64 %rd1, %rd41, %rd56; + add.s64 %rd2, %rd43, %rd56; + add.s64 %rd3, %rd45, %rd56; + add.s64 %rd4, %rd47, %rd56; + add.s64 %rd5, %rd49, %rd56; + add.s64 %rd6, %rd51, %rd56; + add.s64 %rd7, %rd53, %rd56; + add.s64 %rd8, %rd55, %rd56; + mov.b32 %r5, 0; + .loc 1 79 18 // l2norm.py:79:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + mov.b32 {%rs5, %rs6}, %r3; + mov.b32 {%rs7, %rs8}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs9, %rs10}, %r6; + mov.b32 {%rs11, %rs12}, %r7; + mov.b32 {%rs13, %rs14}, %r8; + mov.b32 {%rs15, %rs16}, %r9; + // begin inline asm + mov.u32 %r10, %r5; + mov.u32 %r11, %r5; + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + @%p3 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs17, %rs18}, %r10; + mov.b32 {%rs19, %rs20}, %r11; + mov.b32 {%rs21, %rs22}, %r12; + mov.b32 {%rs23, %rs24}, %r13; + // begin inline asm + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + @%p4 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ]; + // end inline asm + mov.b32 {%rs25, %rs26}, %r14; + mov.b32 {%rs27, %rs28}, %r15; + mov.b32 {%rs29, %rs30}, %r16; + mov.b32 {%rs31, %rs32}, %r17; + // begin inline asm + mov.u32 %r18, %r5; + mov.u32 %r19, %r5; + mov.u32 %r20, %r5; + mov.u32 %r21, %r5; + @%p5 ld.global.v4.b32 { %r18, %r19, %r20, %r21 }, [ %rd5 + 0 ]; + // end inline asm + mov.b32 {%rs33, %rs34}, %r18; + mov.b32 {%rs35, %rs36}, %r19; + mov.b32 {%rs37, %rs38}, %r20; + mov.b32 {%rs39, %rs40}, %r21; + // begin inline asm + mov.u32 %r22, %r5; + mov.u32 %r23, %r5; + mov.u32 %r24, %r5; + mov.u32 %r25, %r5; + @%p6 ld.global.v4.b32 { %r22, %r23, %r24, %r25 }, [ %rd6 + 0 ]; + // end inline asm + mov.b32 {%rs41, %rs42}, %r22; + mov.b32 {%rs43, %rs44}, %r23; + mov.b32 {%rs45, %rs46}, %r24; + mov.b32 {%rs47, %rs48}, %r25; + // begin inline asm + mov.u32 %r26, %r5; + mov.u32 %r27, %r5; + mov.u32 %r28, %r5; + mov.u32 %r29, %r5; + @%p7 ld.global.v4.b32 { %r26, %r27, %r28, %r29 }, [ %rd7 + 0 ]; + // end inline asm + mov.b32 {%rs49, %rs50}, %r26; + mov.b32 {%rs51, %rs52}, %r27; + mov.b32 {%rs53, %rs54}, %r28; + mov.b32 {%rs55, %rs56}, %r29; + // begin inline asm + mov.u32 %r30, %r5; + mov.u32 %r31, %r5; + mov.u32 %r32, %r5; + mov.u32 %r33, %r5; + @%p8 ld.global.v4.b32 { %r30, %r31, %r32, %r33 }, [ %rd8 + 0 ]; + // end inline asm + mov.b32 {%rs57, %rs58}, %r30; + mov.b32 {%rs59, %rs60}, %r31; + mov.b32 {%rs61, %rs62}, %r32; + mov.b32 {%rs63, %rs64}, %r33; + .loc 1 79 47 // l2norm.py:79:47 + cvt.f32.bf16 %r92, %rs1; + cvt.f32.bf16 %r93, %rs2; + cvt.f32.bf16 %r94, %rs3; + cvt.f32.bf16 %r95, %rs4; + cvt.f32.bf16 %r96, %rs5; + cvt.f32.bf16 %r97, %rs6; + cvt.f32.bf16 %r98, %rs7; + cvt.f32.bf16 %r99, %rs8; + cvt.f32.bf16 %r100, %rs9; + cvt.f32.bf16 %r101, %rs10; + cvt.f32.bf16 %r102, %rs11; + cvt.f32.bf16 %r103, %rs12; + cvt.f32.bf16 %r104, %rs13; + cvt.f32.bf16 %r105, %rs14; + cvt.f32.bf16 %r106, %rs15; + cvt.f32.bf16 %r107, %rs16; + cvt.f32.bf16 %r108, %rs17; + cvt.f32.bf16 %r109, %rs18; + cvt.f32.bf16 %r110, %rs19; + cvt.f32.bf16 %r111, %rs20; + cvt.f32.bf16 %r112, %rs21; + cvt.f32.bf16 %r113, %rs22; + cvt.f32.bf16 %r114, %rs23; + cvt.f32.bf16 %r115, %rs24; + cvt.f32.bf16 %r116, %rs25; + cvt.f32.bf16 %r117, %rs26; + cvt.f32.bf16 %r118, %rs27; + cvt.f32.bf16 %r119, %rs28; + cvt.f32.bf16 %r120, %rs29; + cvt.f32.bf16 %r121, %rs30; + cvt.f32.bf16 %r122, %rs31; + cvt.f32.bf16 %r123, %rs32; + cvt.f32.bf16 %r124, %rs33; + cvt.f32.bf16 %r125, %rs34; + cvt.f32.bf16 %r126, %rs35; + cvt.f32.bf16 %r127, %rs36; + cvt.f32.bf16 %r128, %rs37; + cvt.f32.bf16 %r129, %rs38; + cvt.f32.bf16 %r130, %rs39; + cvt.f32.bf16 %r131, %rs40; + cvt.f32.bf16 %r132, %rs41; + cvt.f32.bf16 %r133, %rs42; + cvt.f32.bf16 %r134, %rs43; + cvt.f32.bf16 %r135, %rs44; + cvt.f32.bf16 %r136, %rs45; + cvt.f32.bf16 %r137, %rs46; + cvt.f32.bf16 %r138, %rs47; + cvt.f32.bf16 %r139, %rs48; + cvt.f32.bf16 %r140, %rs49; + cvt.f32.bf16 %r141, %rs50; + cvt.f32.bf16 %r142, %rs51; + cvt.f32.bf16 %r143, %rs52; + cvt.f32.bf16 %r144, %rs53; + cvt.f32.bf16 %r145, %rs54; + cvt.f32.bf16 %r146, %rs55; + cvt.f32.bf16 %r147, %rs56; + cvt.f32.bf16 %r148, %rs57; + cvt.f32.bf16 %r149, %rs58; + cvt.f32.bf16 %r150, %rs59; + cvt.f32.bf16 %r151, %rs60; + cvt.f32.bf16 %r152, %rs61; + cvt.f32.bf16 %r153, %rs62; + cvt.f32.bf16 %r154, %rs63; + cvt.f32.bf16 %r155, %rs64; + .loc 1 80 38 // l2norm.py:80:38 + mul.f32 %r156, %r93, %r93; + mul.f32 %r157, %r101, %r101; + mul.f32 %r158, %r109, %r109; + mul.f32 %r159, %r117, %r117; + mul.f32 %r160, %r125, %r125; + mul.f32 %r161, %r133, %r133; + mul.f32 %r162, %r141, %r141; + mul.f32 %r163, %r149, %r149; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + fma.rn.f32.bf16 %r164, %rs1, %rs1, %r156; + fma.rn.f32.bf16 %r165, %rs3, %rs3, %r164; + fma.rn.f32.bf16 %r166, %rs4, %rs4, %r165; + fma.rn.f32.bf16 %r167, %rs5, %rs5, %r166; + fma.rn.f32.bf16 %r168, %rs6, %rs6, %r167; + fma.rn.f32.bf16 %r169, %rs7, %rs7, %r168; + fma.rn.f32.bf16 %r170, %rs8, %rs8, %r169; + fma.rn.f32.bf16 %r171, %rs9, %rs9, %r157; + fma.rn.f32.bf16 %r172, %rs11, %rs11, %r171; + fma.rn.f32.bf16 %r173, %rs12, %rs12, %r172; + fma.rn.f32.bf16 %r174, %rs13, %rs13, %r173; + fma.rn.f32.bf16 %r175, %rs14, %rs14, %r174; + fma.rn.f32.bf16 %r176, %rs15, %rs15, %r175; + fma.rn.f32.bf16 %r177, %rs16, %rs16, %r176; + fma.rn.f32.bf16 %r178, %rs17, %rs17, %r158; + fma.rn.f32.bf16 %r179, %rs19, %rs19, %r178; + fma.rn.f32.bf16 %r180, %rs20, %rs20, %r179; + fma.rn.f32.bf16 %r181, %rs21, %rs21, %r180; + fma.rn.f32.bf16 %r182, %rs22, %rs22, %r181; + fma.rn.f32.bf16 %r183, %rs23, %rs23, %r182; + fma.rn.f32.bf16 %r184, %rs24, %rs24, %r183; + fma.rn.f32.bf16 %r185, %rs25, %rs25, %r159; + fma.rn.f32.bf16 %r186, %rs27, %rs27, %r185; + fma.rn.f32.bf16 %r187, %rs28, %rs28, %r186; + fma.rn.f32.bf16 %r188, %rs29, %rs29, %r187; + fma.rn.f32.bf16 %r189, %rs30, %rs30, %r188; + fma.rn.f32.bf16 %r190, %rs31, %rs31, %r189; + fma.rn.f32.bf16 %r191, %rs32, %rs32, %r190; + fma.rn.f32.bf16 %r192, %rs33, %rs33, %r160; + fma.rn.f32.bf16 %r193, %rs35, %rs35, %r192; + fma.rn.f32.bf16 %r194, %rs36, %rs36, %r193; + fma.rn.f32.bf16 %r195, %rs37, %rs37, %r194; + fma.rn.f32.bf16 %r196, %rs38, %rs38, %r195; + fma.rn.f32.bf16 %r197, %rs39, %rs39, %r196; + fma.rn.f32.bf16 %r198, %rs40, %rs40, %r197; + fma.rn.f32.bf16 %r199, %rs41, %rs41, %r161; + fma.rn.f32.bf16 %r200, %rs43, %rs43, %r199; + fma.rn.f32.bf16 %r201, %rs44, %rs44, %r200; + fma.rn.f32.bf16 %r202, %rs45, %rs45, %r201; + fma.rn.f32.bf16 %r203, %rs46, %rs46, %r202; + fma.rn.f32.bf16 %r204, %rs47, %rs47, %r203; + fma.rn.f32.bf16 %r205, %rs48, %rs48, %r204; + fma.rn.f32.bf16 %r206, %rs49, %rs49, %r162; + fma.rn.f32.bf16 %r207, %rs51, %rs51, %r206; + fma.rn.f32.bf16 %r208, %rs52, %rs52, %r207; + fma.rn.f32.bf16 %r209, %rs53, %rs53, %r208; + fma.rn.f32.bf16 %r210, %rs54, %rs54, %r209; + fma.rn.f32.bf16 %r211, %rs55, %rs55, %r210; + fma.rn.f32.bf16 %r212, %rs56, %rs56, %r211; + fma.rn.f32.bf16 %r213, %rs57, %rs57, %r163; + fma.rn.f32.bf16 %r214, %rs59, %rs59, %r213; + fma.rn.f32.bf16 %r215, %rs60, %rs60, %r214; + fma.rn.f32.bf16 %r216, %rs61, %rs61, %r215; + fma.rn.f32.bf16 %r217, %rs62, %rs62, %r216; + fma.rn.f32.bf16 %r218, %rs63, %rs63, %r217; + fma.rn.f32.bf16 %r219, %rs64, %rs64, %r218; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r220, %r170, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r221, %r170, %r220; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r222, %r221, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r223, %r221, %r222; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r224, %r223, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r225, %r223, %r224; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r226, %r225, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r227, %r225, %r226; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r228, %r177, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r229, %r177, %r228; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r230, %r229, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r231, %r229, %r230; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r232, %r231, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r233, %r231, %r232; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r234, %r233, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r235, %r233, %r234; +$L__tmp18: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r236, %r184, 8, 31, -1; +$L__tmp19: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r237, %r184, %r236; +$L__tmp20: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r238, %r237, 4, 31, -1; +$L__tmp21: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r239, %r237, %r238; +$L__tmp22: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r240, %r239, 2, 31, -1; +$L__tmp23: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r241, %r239, %r240; +$L__tmp24: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r242, %r241, 1, 31, -1; +$L__tmp25: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r243, %r241, %r242; +$L__tmp26: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r244, %r191, 8, 31, -1; +$L__tmp27: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r245, %r191, %r244; +$L__tmp28: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r246, %r245, 4, 31, -1; +$L__tmp29: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r247, %r245, %r246; +$L__tmp30: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r248, %r247, 2, 31, -1; +$L__tmp31: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r249, %r247, %r248; +$L__tmp32: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r250, %r249, 1, 31, -1; +$L__tmp33: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r251, %r249, %r250; +$L__tmp34: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r252, %r198, 8, 31, -1; +$L__tmp35: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r253, %r198, %r252; +$L__tmp36: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r254, %r253, 4, 31, -1; +$L__tmp37: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r255, %r253, %r254; +$L__tmp38: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r256, %r255, 2, 31, -1; +$L__tmp39: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r257, %r255, %r256; +$L__tmp40: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r258, %r257, 1, 31, -1; +$L__tmp41: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r259, %r257, %r258; +$L__tmp42: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r260, %r205, 8, 31, -1; +$L__tmp43: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r261, %r205, %r260; +$L__tmp44: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r262, %r261, 4, 31, -1; +$L__tmp45: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r263, %r261, %r262; +$L__tmp46: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r264, %r263, 2, 31, -1; +$L__tmp47: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r265, %r263, %r264; +$L__tmp48: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r266, %r265, 1, 31, -1; +$L__tmp49: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r267, %r265, %r266; +$L__tmp50: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r268, %r212, 8, 31, -1; +$L__tmp51: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r269, %r212, %r268; +$L__tmp52: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r270, %r269, 4, 31, -1; +$L__tmp53: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r271, %r269, %r270; +$L__tmp54: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r272, %r271, 2, 31, -1; +$L__tmp55: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r273, %r271, %r272; +$L__tmp56: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r274, %r273, 1, 31, -1; +$L__tmp57: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r275, %r273, %r274; +$L__tmp58: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r276, %r219, 8, 31, -1; +$L__tmp59: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r277, %r219, %r276; +$L__tmp60: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r278, %r277, 4, 31, -1; +$L__tmp61: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r279, %r277, %r278; +$L__tmp62: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r280, %r279, 2, 31, -1; +$L__tmp63: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r281, %r279, %r280; +$L__tmp64: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r282, %r281, 1, 31, -1; +$L__tmp65: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r283, %r281, %r282; +$L__tmp66: + .loc 1 80 48 // l2norm.py:80:48 + add.f32 %r284, %r78, %r227; + add.f32 %r285, %r78, %r235; + add.f32 %r286, %r78, %r243; + add.f32 %r287, %r78, %r251; + add.f32 %r288, %r78, %r259; + add.f32 %r289, %r78, %r267; + add.f32 %r290, %r78, %r275; + add.f32 %r291, %r78, %r283; + .loc 1 80 25 // l2norm.py:80:25 + sqrt.approx.ftz.f32 %r292, %r284; + sqrt.approx.ftz.f32 %r293, %r285; + sqrt.approx.ftz.f32 %r294, %r286; + sqrt.approx.ftz.f32 %r295, %r287; + sqrt.approx.ftz.f32 %r296, %r288; + sqrt.approx.ftz.f32 %r297, %r289; + sqrt.approx.ftz.f32 %r298, %r290; + sqrt.approx.ftz.f32 %r299, %r291; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd57, %r90; + or.b64 %rd58, %rd21, %rd57; + .loc 1 77 20 // l2norm.py:77:20 + shl.b64 %rd59, %rd58, 2; + add.s64 %rd17, %rd20, %rd59; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd60, %rd19, %rd54; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd16, %rd60, %rd56; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd61, %rd19, %rd52; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd15, %rd61, %rd56; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd62, %rd19, %rd50; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd14, %rd62, %rd56; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd63, %rd19, %rd48; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd13, %rd63, %rd56; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd64, %rd19, %rd46; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd12, %rd64, %rd56; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd65, %rd19, %rd44; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd11, %rd65, %rd56; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd66, %rd19, %rd42; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd10, %rd66, %rd56; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd67, %rd19, %rd40; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd9, %rd67, %rd56; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p11, %rd58, %rd22; + mov.b32 %r300, 0f3F800000; + .loc 1 80 17 // l2norm.py:80:17 + div.full.f32 %r67, %r300, %r292; + div.full.f32 %r68, %r300, %r293; + div.full.f32 %r69, %r300, %r294; + div.full.f32 %r70, %r300, %r295; + div.full.f32 %r72, %r300, %r296; + div.full.f32 %r73, %r300, %r297; + div.full.f32 %r74, %r300, %r298; + div.full.f32 %r75, %r300, %r299; + .loc 1 81 16 // l2norm.py:81:16 + mul.f32 %r301, %r67, %r92; + mul.f32 %r302, %r67, %r93; + mul.f32 %r303, %r67, %r94; + mul.f32 %r304, %r67, %r95; + mul.f32 %r305, %r67, %r96; + mul.f32 %r306, %r67, %r97; + mul.f32 %r307, %r67, %r98; + mul.f32 %r308, %r67, %r99; + mul.f32 %r309, %r68, %r100; + mul.f32 %r310, %r68, %r101; + mul.f32 %r311, %r68, %r102; + mul.f32 %r312, %r68, %r103; + mul.f32 %r313, %r68, %r104; + mul.f32 %r314, %r68, %r105; + mul.f32 %r315, %r68, %r106; + mul.f32 %r316, %r68, %r107; + mul.f32 %r317, %r69, %r108; + mul.f32 %r318, %r69, %r109; + mul.f32 %r319, %r69, %r110; + mul.f32 %r320, %r69, %r111; + mul.f32 %r321, %r69, %r112; + mul.f32 %r322, %r69, %r113; + mul.f32 %r323, %r69, %r114; + mul.f32 %r324, %r69, %r115; + mul.f32 %r325, %r70, %r116; + mul.f32 %r326, %r70, %r117; + mul.f32 %r327, %r70, %r118; + mul.f32 %r328, %r70, %r119; + mul.f32 %r329, %r70, %r120; + mul.f32 %r330, %r70, %r121; + mul.f32 %r331, %r70, %r122; + mul.f32 %r332, %r70, %r123; + mul.f32 %r333, %r72, %r124; + mul.f32 %r334, %r72, %r125; + mul.f32 %r335, %r72, %r126; + mul.f32 %r336, %r72, %r127; + mul.f32 %r337, %r72, %r128; + mul.f32 %r338, %r72, %r129; + mul.f32 %r339, %r72, %r130; + mul.f32 %r340, %r72, %r131; + mul.f32 %r341, %r73, %r132; + mul.f32 %r342, %r73, %r133; + mul.f32 %r343, %r73, %r134; + mul.f32 %r344, %r73, %r135; + mul.f32 %r345, %r73, %r136; + mul.f32 %r346, %r73, %r137; + mul.f32 %r347, %r73, %r138; + mul.f32 %r348, %r73, %r139; + mul.f32 %r349, %r74, %r140; + mul.f32 %r350, %r74, %r141; + mul.f32 %r351, %r74, %r142; + mul.f32 %r352, %r74, %r143; + mul.f32 %r353, %r74, %r144; + mul.f32 %r354, %r74, %r145; + mul.f32 %r355, %r74, %r146; + mul.f32 %r356, %r74, %r147; + mul.f32 %r357, %r75, %r148; + mul.f32 %r358, %r75, %r149; + mul.f32 %r359, %r75, %r150; + mul.f32 %r360, %r75, %r151; + mul.f32 %r361, %r75, %r152; + mul.f32 %r362, %r75, %r153; + mul.f32 %r363, %r75, %r154; + mul.f32 %r364, %r75, %r155; + .loc 1 83 18 // l2norm.py:83:18 + cvt.rn.bf16x2.f32 %r34, %r302, %r301; + cvt.rn.bf16x2.f32 %r35, %r304, %r303; + cvt.rn.bf16x2.f32 %r36, %r306, %r305; + cvt.rn.bf16x2.f32 %r37, %r308, %r307; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd9 + 0 ], { %r34, %r35, %r36, %r37 }; + // end inline asm + cvt.rn.bf16x2.f32 %r38, %r310, %r309; + cvt.rn.bf16x2.f32 %r39, %r312, %r311; + cvt.rn.bf16x2.f32 %r40, %r314, %r313; + cvt.rn.bf16x2.f32 %r41, %r316, %r315; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd10 + 0 ], { %r38, %r39, %r40, %r41 }; + // end inline asm + cvt.rn.bf16x2.f32 %r42, %r318, %r317; + cvt.rn.bf16x2.f32 %r43, %r320, %r319; + cvt.rn.bf16x2.f32 %r44, %r322, %r321; + cvt.rn.bf16x2.f32 %r45, %r324, %r323; + // begin inline asm + @%p3 st.global.v4.b32 [ %rd11 + 0 ], { %r42, %r43, %r44, %r45 }; + // end inline asm + cvt.rn.bf16x2.f32 %r46, %r326, %r325; + cvt.rn.bf16x2.f32 %r47, %r328, %r327; + cvt.rn.bf16x2.f32 %r48, %r330, %r329; + cvt.rn.bf16x2.f32 %r49, %r332, %r331; + // begin inline asm + @%p4 st.global.v4.b32 [ %rd12 + 0 ], { %r46, %r47, %r48, %r49 }; + // end inline asm + cvt.rn.bf16x2.f32 %r50, %r334, %r333; + cvt.rn.bf16x2.f32 %r51, %r336, %r335; + cvt.rn.bf16x2.f32 %r52, %r338, %r337; + cvt.rn.bf16x2.f32 %r53, %r340, %r339; + // begin inline asm + @%p5 st.global.v4.b32 [ %rd13 + 0 ], { %r50, %r51, %r52, %r53 }; + // end inline asm + cvt.rn.bf16x2.f32 %r54, %r342, %r341; + cvt.rn.bf16x2.f32 %r55, %r344, %r343; + cvt.rn.bf16x2.f32 %r56, %r346, %r345; + cvt.rn.bf16x2.f32 %r57, %r348, %r347; + // begin inline asm + @%p6 st.global.v4.b32 [ %rd14 + 0 ], { %r54, %r55, %r56, %r57 }; + // end inline asm + cvt.rn.bf16x2.f32 %r58, %r350, %r349; + cvt.rn.bf16x2.f32 %r59, %r352, %r351; + cvt.rn.bf16x2.f32 %r60, %r354, %r353; + cvt.rn.bf16x2.f32 %r61, %r356, %r355; + // begin inline asm + @%p7 st.global.v4.b32 [ %rd15 + 0 ], { %r58, %r59, %r60, %r61 }; + // end inline asm + cvt.rn.bf16x2.f32 %r62, %r358, %r357; + cvt.rn.bf16x2.f32 %r63, %r360, %r359; + cvt.rn.bf16x2.f32 %r64, %r362, %r361; + cvt.rn.bf16x2.f32 %r65, %r364, %r363; + // begin inline asm + @%p8 st.global.v4.b32 [ %rd16 + 0 ], { %r62, %r63, %r64, %r65 }; + // end inline asm + .loc 1 84 21 // l2norm.py:84:21 + shl.b32 %r365, %r80, 1; + mov.b32 %r366, global_smem; + add.s32 %r66, %r366, %r365; + // begin inline asm + st.shared.v4.b32 [ %r66 + 0 ], { %r67, %r68, %r69, %r70 }; + // end inline asm + add.s32 %r71, %r66, 16; + // begin inline asm + st.shared.v4.b32 [ %r71 + 0 ], { %r72, %r73, %r74, %r75 }; + // end inline asm + bar.warp.sync -1; + shl.b32 %r367, %r79, 1; + and.b32 %r368, %r367, 28; + shl.b32 %r369, %r79, 5; + and.b32 %r370, %r369, 32; + add.s32 %r371, %r366, %r370; + add.s32 %r372, %r371, %r368; + ld.shared.b32 %r76, [%r372]; + and.pred %p9, %p10, %p11; + // begin inline asm + @%p9 st.global.b32 [ %rd17 + 0 ], { %r76 }; + // end inline asm + .loc 1 84 4 // l2norm.py:84:4 + ret; +$L__tmp67: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp66 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 80 // DW_AT_call_line +.b8 43 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp66 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/l2norm_fwd_kernel.source b/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/l2norm_fwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..897f77c236e43b6c12f4512723699931dc7dd6b5 --- /dev/null +++ b/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/l2norm_fwd_kernel.source @@ -0,0 +1,172 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc38 = loc(unknown) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc45 = loc("x"(#loc)) +#loc46 = loc("y"(#loc)) +#loc47 = loc("rstd"(#loc)) +#loc48 = loc("eps"(#loc)) +#loc49 = loc("T"(#loc)) +#loc81 = loc("input"(#loc36)) +#loc82 = loc("a"(#loc41)) +#loc83 = loc("b"(#loc41)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc50) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc51) + %o_t = arith.constant 16 : i32 loc(#loc52) + %o_t_1 = arith.constant 16 : i64 loc(#loc52) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc52) + %o_t_3 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32> loc(#loc53) + %o_t_4 = arith.extsi %o_t_3 : tensor<16xi32> to tensor<16xi64> loc(#loc54) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<16xi64> loc(#loc54) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<16xi64> loc(#loc54) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc55) + %m_t = arith.extsi %T : i32 to i64 loc(#loc56) + %m_t_7 = tt.splat %m_t : i64 -> tensor<16xi64> loc(#loc56) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<16xi64> loc(#loc56) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc57) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc58) + %m_x_10 = arith.constant 128 : i32 loc(#loc59) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc59) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc59) + %m_x_13 = tt.broadcast %m_x : tensor<16x1xi1> -> tensor<16x128xi1> loc(#loc60) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<16x128xi1> loc(#loc60) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<16x128xi1> loc(#loc60) + %p_x = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc61) + %p_x_16 = arith.constant 128 : i32 loc(#loc62) + %p_x_17 = arith.constant 128 : i64 loc(#loc62) + %p_x_18 = arith.constant dense<128> : tensor<16x1xi64> loc(#loc62) + %p_x_19 = arith.muli %p_x, %p_x_18 : tensor<16x1xi64> loc(#loc62) + %p_x_20 = tt.splat %x : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc63) + %p_x_21 = tt.addptr %p_x_20, %p_x_19 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc63) + %p_x_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc64) + %p_x_23 = tt.broadcast %p_x_21 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc65) + %p_x_24 = tt.broadcast %p_x_22 : tensor<1x128xi32> -> tensor<16x128xi32> loc(#loc65) + %p_x_25 = tt.addptr %p_x_23, %p_x_24 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc65) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc66) + %p_y_26 = arith.constant 128 : i32 loc(#loc67) + %p_y_27 = arith.constant 128 : i64 loc(#loc67) + %p_y_28 = arith.constant dense<128> : tensor<16x1xi64> loc(#loc67) + %p_y_29 = arith.muli %p_y, %p_y_28 : tensor<16x1xi64> loc(#loc67) + %p_y_30 = tt.splat %y : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc68) + %p_y_31 = tt.addptr %p_y_30, %p_y_29 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc68) + %p_y_32 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc69) + %p_y_33 = tt.broadcast %p_y_31 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc70) + %p_y_34 = tt.broadcast %p_y_32 : tensor<1x128xi32> -> tensor<16x128xi32> loc(#loc70) + %p_y_35 = tt.addptr %p_y_33, %p_y_34 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc70) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc71) + %p_rstd_36 = tt.addptr %p_rstd, %o_t_6 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc71) + %b_x = arith.constant 0.000000e+00 : f32 loc(#loc72) + %b_x_37 = arith.constant dense<0.000000e+00> : tensor<16x128xf32> loc(#loc72) + %b_x_38 = arith.truncf %b_x_37 : tensor<16x128xf32> to tensor<16x128xbf16> loc(#loc72) + %b_x_39 = tt.load %p_x_25, %m_x_15, %b_x_38 : tensor<16x128x!tt.ptr> loc(#loc72) + %b_x_40 = arith.extf %b_x_39 : tensor<16x128xbf16> to tensor<16x128xf32> loc(#loc73) + %b_rstd = arith.mulf %b_x_40, %b_x_40 : tensor<16x128xf32> loc(#loc74) + %b_rstd_41 = tt.call @triton.language.standard.sum__fp32S16_128S_c1_cFalse_cNone(%b_rstd) : (tensor<16x128xf32>) -> tensor<16xf32> loc(#loc75) + %b_rstd_42 = tt.splat %eps : f32 -> tensor<16xf32> loc(#loc76) + %b_rstd_43 = arith.addf %b_rstd_41, %b_rstd_42 : tensor<16xf32> loc(#loc76) + %b_rstd_44 = math.sqrt %b_rstd_43 : tensor<16xf32> loc(#loc77) + %b_rstd_45 = arith.constant 1 : i32 loc(#loc78) + %b_rstd_46 = arith.constant 1.000000e+00 : f32 loc(#loc78) + %b_rstd_47 = arith.constant dense<1.000000e+00> : tensor<16xf32> loc(#loc78) + %b_rstd_48 = arith.divf %b_rstd_47, %b_rstd_44 : tensor<16xf32> loc(#loc78) + %b_y = tt.expand_dims %b_rstd_48 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc79) + %b_y_49 = tt.broadcast %b_y : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc80) + %b_y_50 = arith.mulf %b_x_40, %b_y_49 : tensor<16x128xf32> loc(#loc80) + %0 = arith.truncf %b_y_50 : tensor<16x128xf32> to tensor<16x128xbf16> loc(#loc32) + tt.store %p_y_35, %0, %m_x_15 : tensor<16x128x!tt.ptr> loc(#loc33) + tt.store %p_rstd_36, %b_rstd_48, %m_t_8 : tensor<16x!tt.ptr> loc(#loc34) + tt.return loc(#loc35) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S16_128S_c1_cFalse_cNone(%input: tensor<16x128xf32> loc("input"(#loc36))) -> tensor<16xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc37) + tt.reduce.return %2 : f32 loc(#loc37) + }) : (tensor<16x128xf32>) -> tensor<16xf32> loc(#loc37) + tt.return %0 : tensor<16xf32> loc(#loc39) + ^bb1: // no predecessors + %1 = ub.poison : tensor<16xf32> loc(#loc40) + tt.return %1 : tensor<16xf32> loc(#loc40) + } loc(#loc36) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc41)), %b: f32 loc("b"(#loc41))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc42) + tt.return %0 : f32 loc(#loc43) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc44) + tt.return %1 : f32 loc(#loc44) + } loc(#loc41) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:18) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:29) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:37) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc50 = loc("i_t"(#loc1)) +#loc51 = loc("i_t"(#loc2)) +#loc52 = loc("o_t"(#loc3)) +#loc53 = loc("o_t"(#loc4)) +#loc54 = loc("o_t"(#loc5)) +#loc55 = loc("o_d"(#loc6)) +#loc56 = loc("m_t"(#loc7)) +#loc57 = loc("m_x"(#loc8)) +#loc58 = loc("m_x"(#loc9)) +#loc59 = loc("m_x"(#loc10)) +#loc60 = loc("m_x"(#loc11)) +#loc61 = loc("p_x"(#loc12)) +#loc62 = loc("p_x"(#loc13)) +#loc63 = loc("p_x"(#loc14)) +#loc64 = loc("p_x"(#loc15)) +#loc65 = loc("p_x"(#loc16)) +#loc66 = loc("p_y"(#loc17)) +#loc67 = loc("p_y"(#loc18)) +#loc68 = loc("p_y"(#loc19)) +#loc69 = loc("p_y"(#loc20)) +#loc70 = loc("p_y"(#loc21)) +#loc71 = loc("p_rstd"(#loc22)) +#loc72 = loc("b_x"(#loc23)) +#loc73 = loc("b_x"(#loc24)) +#loc74 = loc("b_rstd"(#loc25)) +#loc75 = loc("b_rstd"(#loc26)) +#loc76 = loc("b_rstd"(#loc27)) +#loc77 = loc("b_rstd"(#loc28)) +#loc78 = loc("b_rstd"(#loc29)) +#loc79 = loc("b_y"(#loc30)) +#loc80 = loc("b_y"(#loc31)) diff --git a/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/l2norm_fwd_kernel.ttgir b/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/l2norm_fwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..01a521a21e54ffccd1a350e334138bd5b8f697b9 --- /dev/null +++ b/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/l2norm_fwd_kernel.ttgir @@ -0,0 +1,135 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [1, 1], order = [1, 0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [1], order = [0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc1 = loc(unknown) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc34 = loc("x"(#loc)) +#loc35 = loc("y"(#loc)) +#loc36 = loc("rstd"(#loc)) +#loc37 = loc("eps"(#loc)) +#loc38 = loc("T"(#loc)) +#loc59 = loc("b_rstd"(#loc23)) +#loc66 = loc(callsite(#loc1 at #loc59)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 1 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<1.000000e+00> : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<16x128xbf16, #blocked> loc(#loc1) + %c16_i64 = arith.constant 16 : i64 loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<16x1xi64, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc39) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc40) + %o_t = arith.muli %i_t_3, %c16_i64 : i64 loc(#loc41) + %o_t_4 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc42) + %o_t_5 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32, #blocked1> loc(#loc42) + %o_t_6 = arith.extsi %o_t_4 : tensor<16xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_7 = arith.extsi %o_t_5 : tensor<16xi32, #blocked1> to tensor<16xi64, #blocked1> loc(#loc43) + %o_t_8 = tt.splat %o_t : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_9 = tt.splat %o_t : i64 -> tensor<16xi64, #blocked1> loc(#loc43) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<16xi64, #blocked1> loc(#loc43) + %m_t = arith.extsi %T : i32 to i64 loc(#loc44) + %m_t_12 = tt.splat %m_t : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_13 = tt.splat %m_t : i64 -> tensor<16xi64, #blocked1> loc(#loc44) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<16xi64, #blocked1> loc(#loc44) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi1, #blocked> loc(#loc45) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc46) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc46) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_1 : tensor<1x128xi32, #blocked> loc(#loc47) + %m_x_19 = tt.broadcast %m_x : tensor<16x1xi1, #blocked> -> tensor<16x128xi1, #blocked> loc(#loc48) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<16x128xi1, #blocked> loc(#loc48) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<16x128xi1, #blocked> loc(#loc48) + %p_x = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi64, #blocked> loc(#loc49) + %p_x_22 = arith.muli %p_x, %cst_2 : tensor<16x1xi64, #blocked> loc(#loc50) + %p_x_23 = tt.splat %x : !tt.ptr -> tensor<16x1x!tt.ptr, #blocked> loc(#loc51) + %p_x_24 = tt.addptr %p_x_23, %p_x_22 : tensor<16x1x!tt.ptr, #blocked>, tensor<16x1xi64, #blocked> loc(#loc51) + %p_x_25 = tt.broadcast %p_x_24 : tensor<16x1x!tt.ptr, #blocked> -> tensor<16x128x!tt.ptr, #blocked> loc(#loc52) + %p_x_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<16x128xi32, #blocked> loc(#loc52) + %p_x_27 = tt.addptr %p_x_25, %p_x_26 : tensor<16x128x!tt.ptr, #blocked>, tensor<16x128xi32, #blocked> loc(#loc52) + %p_y = tt.splat %y : !tt.ptr -> tensor<16x1x!tt.ptr, #blocked> loc(#loc53) + %p_y_28 = tt.addptr %p_y, %p_x_22 : tensor<16x1x!tt.ptr, #blocked>, tensor<16x1xi64, #blocked> loc(#loc53) + %p_y_29 = tt.broadcast %p_y_28 : tensor<16x1x!tt.ptr, #blocked> -> tensor<16x128x!tt.ptr, #blocked> loc(#loc54) + %p_y_30 = tt.addptr %p_y_29, %p_x_26 : tensor<16x128x!tt.ptr, #blocked>, tensor<16x128xi32, #blocked> loc(#loc54) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<16x!tt.ptr, #blocked1> loc(#loc55) + %p_rstd_31 = tt.addptr %p_rstd, %o_t_11 : tensor<16x!tt.ptr, #blocked1>, tensor<16xi64, #blocked1> loc(#loc55) + %b_x = tt.load %p_x_27, %m_x_21, %cst_0 : tensor<16x128x!tt.ptr, #blocked> loc(#loc56) + %b_x_32 = arith.extf %b_x : tensor<16x128xbf16, #blocked> to tensor<16x128xf32, #blocked> loc(#loc57) + %b_rstd = arith.mulf %b_x_32, %b_x_32 : tensor<16x128xf32, #blocked> loc(#loc58) + %b_rstd_33 = "tt.reduce"(%b_rstd) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_40: f32 loc(callsite(#loc1 at #loc59)), %b_rstd_41: f32 loc(callsite(#loc1 at #loc59))): + %b_rstd_42 = arith.addf %b_rstd_40, %b_rstd_41 : f32 loc(#loc67) + tt.reduce.return %b_rstd_42 : f32 loc(#loc65) + }) : (tensor<16x128xf32, #blocked>) -> tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc65) + %b_rstd_34 = tt.splat %eps : f32 -> tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_36 = math.sqrt %b_rstd_35 : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc61) + %b_rstd_37 = arith.divf %cst, %b_rstd_36 : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc62) + %b_y = tt.expand_dims %b_rstd_37 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xf32, #blocked> loc(#loc63) + %b_y_38 = tt.broadcast %b_y : tensor<16x1xf32, #blocked> -> tensor<16x128xf32, #blocked> loc(#loc64) + %b_y_39 = arith.mulf %b_x_32, %b_y_38 : tensor<16x128xf32, #blocked> loc(#loc64) + %0 = arith.truncf %b_y_39 : tensor<16x128xf32, #blocked> to tensor<16x128xbf16, #blocked> loc(#loc30) + tt.store %p_y_30, %0, %m_x_21 : tensor<16x128x!tt.ptr, #blocked> loc(#loc31) + %1 = ttg.convert_layout %b_rstd_37 : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16xf32, #blocked1> loc(#loc32) + tt.store %p_rstd_31, %1, %m_t_15 : tensor<16x!tt.ptr, #blocked1> loc(#loc32) + tt.return loc(#loc33) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc39 = loc("i_t"(#loc2)) +#loc40 = loc("i_t"(#loc3)) +#loc41 = loc("o_t"(#loc4)) +#loc42 = loc("o_t"(#loc5)) +#loc43 = loc("o_t"(#loc6)) +#loc44 = loc("m_t"(#loc7)) +#loc45 = loc("m_x"(#loc8)) +#loc46 = loc("m_x"(#loc9)) +#loc47 = loc("m_x"(#loc10)) +#loc48 = loc("m_x"(#loc11)) +#loc49 = loc("p_x"(#loc12)) +#loc50 = loc("p_x"(#loc13)) +#loc51 = loc("p_x"(#loc14)) +#loc52 = loc("p_x"(#loc15)) +#loc53 = loc("p_y"(#loc16)) +#loc54 = loc("p_y"(#loc17)) +#loc55 = loc("p_rstd"(#loc18)) +#loc56 = loc("b_x"(#loc19)) +#loc57 = loc("b_x"(#loc20)) +#loc58 = loc("b_rstd"(#loc21)) +#loc60 = loc("b_rstd"(#loc25)) +#loc61 = loc("b_rstd"(#loc26)) +#loc62 = loc("b_rstd"(#loc27)) +#loc63 = loc("b_y"(#loc28)) +#loc64 = loc("b_y"(#loc29)) +#loc65 = loc(callsite(#loc22 at #loc59)) +#loc67 = loc(callsite(#loc24 at #loc65)) diff --git a/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/l2norm_fwd_kernel.ttir b/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/l2norm_fwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..d7d00d4871c5899dd9e1884d5ca3c3a665bff374 --- /dev/null +++ b/triton/gpu0/B2MCNBPUUK6G3ZI5CY6SOWVPDM7VAZ6IP2CONBL3OI3SDVU4LOAA/l2norm_fwd_kernel.ttir @@ -0,0 +1,128 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc3 = loc(unknown) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc35 = loc("x"(#loc)) +#loc36 = loc("y"(#loc)) +#loc37 = loc("rstd"(#loc)) +#loc38 = loc("eps"(#loc)) +#loc39 = loc("T"(#loc)) +#loc62 = loc("b_rstd"(#loc25)) +#loc68 = loc(callsite(#loc3 at #loc62)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %b_x = arith.constant dense<0.000000e+00> : tensor<16x128xbf16> loc(#loc40) + %b_rstd = arith.constant dense<1.000000e+00> : tensor<16xf32> loc(#loc41) + %cst = arith.constant dense<128> : tensor<16x1xi64> loc(#loc3) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc42) + %o_t = arith.constant 16 : i64 loc(#loc43) + %i_t = tt.get_program_id x : i32 loc(#loc44) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc45) + %o_t_1 = arith.muli %i_t_0, %o_t : i64 loc(#loc43) + %o_t_2 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32> loc(#loc46) + %o_t_3 = arith.extsi %o_t_2 : tensor<16xi32> to tensor<16xi64> loc(#loc47) + %o_t_4 = tt.splat %o_t_1 : i64 -> tensor<16xi64> loc(#loc47) + %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<16xi64> loc(#loc47) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc48) + %m_t = arith.extsi %T : i32 to i64 loc(#loc49) + %m_t_6 = tt.splat %m_t : i64 -> tensor<16xi64> loc(#loc49) + %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<16xi64> loc(#loc49) + %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc50) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc51) + %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc42) + %m_x_11 = tt.broadcast %m_x_8 : tensor<16x1xi1> -> tensor<16x128xi1> loc(#loc52) + %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<16x128xi1> loc(#loc52) + %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<16x128xi1> loc(#loc52) + %p_x = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc53) + %p_x_14 = arith.muli %p_x, %cst : tensor<16x1xi64> loc(#loc54) + %p_x_15 = tt.splat %x : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc55) + %p_x_16 = tt.addptr %p_x_15, %p_x_14 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc55) + %p_x_17 = tt.broadcast %p_x_16 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc56) + %p_x_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<16x128xi32> loc(#loc56) + %p_x_19 = tt.addptr %p_x_17, %p_x_18 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc56) + %p_y = tt.splat %y : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc57) + %p_y_20 = tt.addptr %p_y, %p_x_14 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc57) + %p_y_21 = tt.broadcast %p_y_20 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc58) + %p_y_22 = tt.addptr %p_y_21, %p_x_18 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc59) + %p_rstd_23 = tt.addptr %p_rstd, %o_t_5 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc59) + %b_x_24 = tt.load %p_x_19, %m_x_13, %b_x : tensor<16x128x!tt.ptr> loc(#loc40) + %b_x_25 = arith.extf %b_x_24 : tensor<16x128xbf16> to tensor<16x128xf32> loc(#loc60) + %b_rstd_26 = arith.mulf %b_x_25, %b_x_25 : tensor<16x128xf32> loc(#loc61) + %b_rstd_27 = "tt.reduce"(%b_rstd_26) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_34: f32 loc(callsite(#loc3 at #loc62)), %b_rstd_35: f32 loc(callsite(#loc3 at #loc62))): + %b_rstd_36 = arith.addf %b_rstd_34, %b_rstd_35 : f32 loc(#loc69) + tt.reduce.return %b_rstd_36 : f32 loc(#loc67) + }) : (tensor<16x128xf32>) -> tensor<16xf32> loc(#loc67) + %b_rstd_28 = tt.splat %eps : f32 -> tensor<16xf32> loc(#loc63) + %b_rstd_29 = arith.addf %b_rstd_27, %b_rstd_28 : tensor<16xf32> loc(#loc63) + %b_rstd_30 = math.sqrt %b_rstd_29 : tensor<16xf32> loc(#loc64) + %b_rstd_31 = arith.divf %b_rstd, %b_rstd_30 : tensor<16xf32> loc(#loc41) + %b_y = tt.expand_dims %b_rstd_31 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc65) + %b_y_32 = tt.broadcast %b_y : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc66) + %b_y_33 = arith.mulf %b_x_25, %b_y_32 : tensor<16x128xf32> loc(#loc66) + %0 = arith.truncf %b_y_33 : tensor<16x128xf32> to tensor<16x128xbf16> loc(#loc31) + tt.store %p_y_22, %0, %m_x_13 : tensor<16x128x!tt.ptr> loc(#loc32) + tt.store %p_rstd_23, %b_rstd_31, %m_t_7 : tensor<16x!tt.ptr> loc(#loc33) + tt.return loc(#loc34) + } loc(#loc) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc40 = loc("b_x"(#loc1)) +#loc41 = loc("b_rstd"(#loc2)) +#loc42 = loc("m_x"(#loc4)) +#loc43 = loc("o_t"(#loc5)) +#loc44 = loc("i_t"(#loc6)) +#loc45 = loc("i_t"(#loc7)) +#loc46 = loc("o_t"(#loc8)) +#loc47 = loc("o_t"(#loc9)) +#loc48 = loc("o_d"(#loc10)) +#loc49 = loc("m_t"(#loc11)) +#loc50 = loc("m_x"(#loc12)) +#loc51 = loc("m_x"(#loc13)) +#loc52 = loc("m_x"(#loc14)) +#loc53 = loc("p_x"(#loc15)) +#loc54 = loc("p_x"(#loc16)) +#loc55 = loc("p_x"(#loc17)) +#loc56 = loc("p_x"(#loc18)) +#loc57 = loc("p_y"(#loc19)) +#loc58 = loc("p_y"(#loc20)) +#loc59 = loc("p_rstd"(#loc21)) +#loc60 = loc("b_x"(#loc22)) +#loc61 = loc("b_rstd"(#loc23)) +#loc63 = loc("b_rstd"(#loc27)) +#loc64 = loc("b_rstd"(#loc28)) +#loc65 = loc("b_y"(#loc29)) +#loc66 = loc("b_y"(#loc30)) +#loc67 = loc(callsite(#loc24 at #loc62)) +#loc69 = loc(callsite(#loc26 at #loc67)) diff --git a/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/__grp__l2norm_bwd_kernel.json b/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/__grp__l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..f2b557ac3f83bc75bba5a756baf5289f41cc7aaf --- /dev/null +++ b/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/__grp__l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_bwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.source", "l2norm_bwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.ttir", "l2norm_bwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.ttgir", "l2norm_bwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.llir", "l2norm_bwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.ptx", "l2norm_bwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.cubin", "l2norm_bwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.cubin b/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..b25e756fbd7938408d7720ddf04258823cb65f79 --- /dev/null +++ b/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.cubin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3f3753afe1ef1cff9a93462255695b97534827eb81b713fe37b90efa24952b90 +size 240192 diff --git a/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.json b/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..6a18c8d7db3b10be8b94d7dba4d6e0b31154110e --- /dev/null +++ b/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"hash": "0ed46e911aefad77e8bceed2907499d6d667c2fe4b0428382c1b2f9faf6eff11", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 1, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 256, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_bwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.llir b/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..4dddaed1bcbd4d3a2266a9d657969047a347d5df --- /dev/null +++ b/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.llir @@ -0,0 +1,4778 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_bwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, ptr addrspace(1) %3, float %4, i32 %5, ptr addrspace(1) readnone captures(none) %6, ptr addrspace(1) readnone captures(none) %7) local_unnamed_addr #0 !dbg !4 { + %9 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !12 + %10 = zext nneg i32 %9 to i64, !dbg !13 + %11 = shl nuw nsw i64 %10, 6, !dbg !14 + %12 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !15 + %13 = and i32 %12, 16, !dbg !15 + %.lobit = lshr exact i32 %13, 4, !dbg !15 + %14 = or disjoint i32 %.lobit, 2, !dbg !15 + %15 = shl nuw nsw i32 %12, 1, !dbg !15 + %16 = zext nneg i32 %.lobit to i64, !dbg !16 + %17 = zext nneg i32 %14 to i64, !dbg !16 + %18 = or disjoint i64 %11, %16, !dbg !16 + %19 = or disjoint i64 %11, %17, !dbg !16 + %20 = sext i32 %5 to i64, !dbg !17 + %21 = icmp slt i64 %18, %20, !dbg !17 + %22 = icmp slt i64 %19, %20, !dbg !17 + %23 = insertelement <16 x i32> poison, i32 %.lobit, i64 0, !dbg !15 + %24 = insertelement <8 x i32> poison, i32 %.lobit, i64 0, !dbg !15 + %25 = insertelement <4 x i32> poison, i32 %.lobit, i64 0, !dbg !15 + %26 = insertelement <2 x i32> poison, i32 %.lobit, i64 0, !dbg !15 + %27 = and i32 %15, 62, !dbg !15 + %28 = or disjoint i32 %27, 1, !dbg !15 + %29 = insertelement <32 x i32> poison, i32 %28, i64 0, !dbg !16 + %30 = insertelement <32 x i32> %29, i32 %27, i64 1, !dbg !16 + %31 = shufflevector <16 x i32> %23, <16 x i32> poison, <32 x i32> , !dbg !16 + %32 = or disjoint <32 x i32> %31, , !dbg !16 + %33 = shufflevector <32 x i32> %30, <32 x i32> %32, <32 x i32> , !dbg !16 + %34 = shufflevector <8 x i32> %24, <8 x i32> poison, <32 x i32> , !dbg !16 + %35 = or disjoint <32 x i32> %34, , !dbg !16 + %36 = shufflevector <32 x i32> %33, <32 x i32> %35, <32 x i32> , !dbg !16 + %37 = shufflevector <4 x i32> %25, <4 x i32> poison, <32 x i32> , !dbg !16 + %38 = or disjoint <32 x i32> %37, , !dbg !16 + %39 = shufflevector <32 x i32> %36, <32 x i32> %38, <32 x i32> , !dbg !16 + %40 = shufflevector <2 x i32> %26, <2 x i32> poison, <32 x i32> , !dbg !16 + %41 = or disjoint <32 x i32> %40, , !dbg !16 + %42 = shufflevector <32 x i32> %39, <32 x i32> %41, <32 x i32> , !dbg !16 + %43 = zext <32 x i32> %42 to <32 x i64>, !dbg !16 + %44 = insertelement <32 x i64> poison, i64 %11, i64 0, !dbg !16 + %45 = shufflevector <32 x i64> %44, <32 x i64> poison, <32 x i32> zeroinitializer, !dbg !16 + %46 = or disjoint <32 x i64> %45, %43, !dbg !16 + %47 = insertelement <32 x i32> poison, i32 %5, i64 0, !dbg !17 + %48 = shufflevector <32 x i32> %47, <32 x i32> poison, <32 x i32> zeroinitializer, !dbg !17 + %49 = sext <32 x i32> %48 to <32 x i64>, !dbg !17 + %50 = icmp slt <32 x i64> %46, %49, !dbg !17 + %51 = and i32 %12, 15, !dbg !18 + %52 = shl nuw nsw i32 %51, 3, !dbg !18 + %53 = shl nuw nsw i64 %18, 7, !dbg !19 + %54 = shl nuw nsw i64 %19, 7, !dbg !19 + %55 = extractelement <32 x i64> %46, i64 31, !dbg !19 + %56 = shl nuw nsw i64 %55, 7, !dbg !19 + %57 = extractelement <32 x i64> %46, i64 30, !dbg !19 + %58 = shl nuw nsw i64 %57, 7, !dbg !19 + %59 = extractelement <32 x i64> %46, i64 29, !dbg !19 + %60 = shl nuw nsw i64 %59, 7, !dbg !19 + %61 = extractelement <32 x i64> %46, i64 28, !dbg !19 + %62 = shl nuw nsw i64 %61, 7, !dbg !19 + %63 = extractelement <32 x i64> %46, i64 27, !dbg !19 + %64 = shl nuw nsw i64 %63, 7, !dbg !19 + %65 = extractelement <32 x i64> %46, i64 26, !dbg !19 + %66 = shl nuw nsw i64 %65, 7, !dbg !19 + %67 = extractelement <32 x i64> %46, i64 25, !dbg !19 + %68 = shl nuw nsw i64 %67, 7, !dbg !19 + %69 = extractelement <32 x i64> %46, i64 24, !dbg !19 + %70 = shl nuw nsw i64 %69, 7, !dbg !19 + %71 = extractelement <32 x i64> %46, i64 23, !dbg !19 + %72 = shl nuw nsw i64 %71, 7, !dbg !19 + %73 = extractelement <32 x i64> %46, i64 22, !dbg !19 + %74 = shl nuw nsw i64 %73, 7, !dbg !19 + %75 = extractelement <32 x i64> %46, i64 21, !dbg !19 + %76 = shl nuw nsw i64 %75, 7, !dbg !19 + %77 = extractelement <32 x i64> %46, i64 20, !dbg !19 + %78 = shl nuw nsw i64 %77, 7, !dbg !19 + %79 = extractelement <32 x i64> %46, i64 19, !dbg !19 + %80 = shl nuw nsw i64 %79, 7, !dbg !19 + %81 = extractelement <32 x i64> %46, i64 18, !dbg !19 + %82 = shl nuw nsw i64 %81, 7, !dbg !19 + %83 = extractelement <32 x i64> %46, i64 17, !dbg !19 + %84 = shl nuw nsw i64 %83, 7, !dbg !19 + %85 = extractelement <32 x i64> %46, i64 16, !dbg !19 + %86 = shl nuw nsw i64 %85, 7, !dbg !19 + %87 = extractelement <32 x i64> %46, i64 15, !dbg !19 + %88 = shl nuw nsw i64 %87, 7, !dbg !19 + %89 = extractelement <32 x i64> %46, i64 14, !dbg !19 + %90 = shl nuw nsw i64 %89, 7, !dbg !19 + %91 = extractelement <32 x i64> %46, i64 13, !dbg !19 + %92 = shl nuw nsw i64 %91, 7, !dbg !19 + %93 = extractelement <32 x i64> %46, i64 12, !dbg !19 + %94 = shl nuw nsw i64 %93, 7, !dbg !19 + %95 = extractelement <32 x i64> %46, i64 11, !dbg !19 + %96 = shl nuw nsw i64 %95, 7, !dbg !19 + %97 = extractelement <32 x i64> %46, i64 10, !dbg !19 + %98 = shl nuw nsw i64 %97, 7, !dbg !19 + %99 = extractelement <32 x i64> %46, i64 9, !dbg !19 + %100 = shl nuw nsw i64 %99, 7, !dbg !19 + %101 = extractelement <32 x i64> %46, i64 8, !dbg !19 + %102 = shl nuw nsw i64 %101, 7, !dbg !19 + %103 = extractelement <32 x i64> %46, i64 7, !dbg !19 + %104 = shl nuw nsw i64 %103, 7, !dbg !19 + %105 = extractelement <32 x i64> %46, i64 6, !dbg !19 + %106 = shl nuw nsw i64 %105, 7, !dbg !19 + %107 = extractelement <32 x i64> %46, i64 5, !dbg !19 + %108 = shl nuw nsw i64 %107, 7, !dbg !19 + %109 = extractelement <32 x i64> %46, i64 4, !dbg !19 + %110 = shl nuw nsw i64 %109, 7, !dbg !19 + %111 = extractelement <32 x i64> %46, i64 3, !dbg !19 + %112 = shl nuw nsw i64 %111, 7, !dbg !19 + %113 = extractelement <32 x i64> %46, i64 2, !dbg !19 + %114 = shl nuw nsw i64 %113, 7, !dbg !19 + %115 = getelementptr bfloat, ptr addrspace(1) %0, i64 %53, !dbg !20 + %116 = getelementptr bfloat, ptr addrspace(1) %0, i64 %54, !dbg !20 + %117 = getelementptr bfloat, ptr addrspace(1) %0, i64 %56, !dbg !20 + %118 = getelementptr bfloat, ptr addrspace(1) %0, i64 %58, !dbg !20 + %119 = getelementptr bfloat, ptr addrspace(1) %0, i64 %60, !dbg !20 + %120 = getelementptr bfloat, ptr addrspace(1) %0, i64 %62, !dbg !20 + %121 = getelementptr bfloat, ptr addrspace(1) %0, i64 %64, !dbg !20 + %122 = getelementptr bfloat, ptr addrspace(1) %0, i64 %66, !dbg !20 + %123 = getelementptr bfloat, ptr addrspace(1) %0, i64 %68, !dbg !20 + %124 = getelementptr bfloat, ptr addrspace(1) %0, i64 %70, !dbg !20 + %125 = getelementptr bfloat, ptr addrspace(1) %0, i64 %72, !dbg !20 + %126 = getelementptr bfloat, ptr addrspace(1) %0, i64 %74, !dbg !20 + %127 = getelementptr bfloat, ptr addrspace(1) %0, i64 %76, !dbg !20 + %128 = getelementptr bfloat, ptr addrspace(1) %0, i64 %78, !dbg !20 + %129 = getelementptr bfloat, ptr addrspace(1) %0, i64 %80, !dbg !20 + %130 = getelementptr bfloat, ptr addrspace(1) %0, i64 %82, !dbg !20 + %131 = getelementptr bfloat, ptr addrspace(1) %0, i64 %84, !dbg !20 + %132 = getelementptr bfloat, ptr addrspace(1) %0, i64 %86, !dbg !20 + %133 = getelementptr bfloat, ptr addrspace(1) %0, i64 %88, !dbg !20 + %134 = getelementptr bfloat, ptr addrspace(1) %0, i64 %90, !dbg !20 + %135 = getelementptr bfloat, ptr addrspace(1) %0, i64 %92, !dbg !20 + %136 = getelementptr bfloat, ptr addrspace(1) %0, i64 %94, !dbg !20 + %137 = getelementptr bfloat, ptr addrspace(1) %0, i64 %96, !dbg !20 + %138 = getelementptr bfloat, ptr addrspace(1) %0, i64 %98, !dbg !20 + %139 = getelementptr bfloat, ptr addrspace(1) %0, i64 %100, !dbg !20 + %140 = getelementptr bfloat, ptr addrspace(1) %0, i64 %102, !dbg !20 + %141 = getelementptr bfloat, ptr addrspace(1) %0, i64 %104, !dbg !20 + %142 = getelementptr bfloat, ptr addrspace(1) %0, i64 %106, !dbg !20 + %143 = getelementptr bfloat, ptr addrspace(1) %0, i64 %108, !dbg !20 + %144 = getelementptr bfloat, ptr addrspace(1) %0, i64 %110, !dbg !20 + %145 = getelementptr bfloat, ptr addrspace(1) %0, i64 %112, !dbg !20 + %146 = getelementptr bfloat, ptr addrspace(1) %0, i64 %114, !dbg !20 + %147 = zext nneg i32 %52 to i64, !dbg !21 + %148 = getelementptr bfloat, ptr addrspace(1) %115, i64 %147, !dbg !21 + %149 = getelementptr bfloat, ptr addrspace(1) %116, i64 %147, !dbg !21 + %150 = getelementptr bfloat, ptr addrspace(1) %117, i64 %147, !dbg !21 + %151 = getelementptr bfloat, ptr addrspace(1) %118, i64 %147, !dbg !21 + %152 = getelementptr bfloat, ptr addrspace(1) %119, i64 %147, !dbg !21 + %153 = getelementptr bfloat, ptr addrspace(1) %120, i64 %147, !dbg !21 + %154 = getelementptr bfloat, ptr addrspace(1) %121, i64 %147, !dbg !21 + %155 = getelementptr bfloat, ptr addrspace(1) %122, i64 %147, !dbg !21 + %156 = getelementptr bfloat, ptr addrspace(1) %123, i64 %147, !dbg !21 + %157 = getelementptr bfloat, ptr addrspace(1) %124, i64 %147, !dbg !21 + %158 = getelementptr bfloat, ptr addrspace(1) %125, i64 %147, !dbg !21 + %159 = getelementptr bfloat, ptr addrspace(1) %126, i64 %147, !dbg !21 + %160 = getelementptr bfloat, ptr addrspace(1) %127, i64 %147, !dbg !21 + %161 = getelementptr bfloat, ptr addrspace(1) %128, i64 %147, !dbg !21 + %162 = getelementptr bfloat, ptr addrspace(1) %129, i64 %147, !dbg !21 + %163 = getelementptr bfloat, ptr addrspace(1) %130, i64 %147, !dbg !21 + %164 = getelementptr bfloat, ptr addrspace(1) %131, i64 %147, !dbg !21 + %165 = getelementptr bfloat, ptr addrspace(1) %132, i64 %147, !dbg !21 + %166 = getelementptr bfloat, ptr addrspace(1) %133, i64 %147, !dbg !21 + %167 = getelementptr bfloat, ptr addrspace(1) %134, i64 %147, !dbg !21 + %168 = getelementptr bfloat, ptr addrspace(1) %135, i64 %147, !dbg !21 + %169 = getelementptr bfloat, ptr addrspace(1) %136, i64 %147, !dbg !21 + %170 = getelementptr bfloat, ptr addrspace(1) %137, i64 %147, !dbg !21 + %171 = getelementptr bfloat, ptr addrspace(1) %138, i64 %147, !dbg !21 + %172 = getelementptr bfloat, ptr addrspace(1) %139, i64 %147, !dbg !21 + %173 = getelementptr bfloat, ptr addrspace(1) %140, i64 %147, !dbg !21 + %174 = getelementptr bfloat, ptr addrspace(1) %141, i64 %147, !dbg !21 + %175 = getelementptr bfloat, ptr addrspace(1) %142, i64 %147, !dbg !21 + %176 = getelementptr bfloat, ptr addrspace(1) %143, i64 %147, !dbg !21 + %177 = getelementptr bfloat, ptr addrspace(1) %144, i64 %147, !dbg !21 + %178 = getelementptr bfloat, ptr addrspace(1) %145, i64 %147, !dbg !21 + %179 = getelementptr bfloat, ptr addrspace(1) %146, i64 %147, !dbg !21 + %180 = extractelement <32 x i64> %46, i64 1, !dbg !22 + %181 = getelementptr float, ptr addrspace(1) %1, i64 %180, !dbg !22 + %182 = extractelement <32 x i64> %46, i64 0, !dbg !22 + %183 = getelementptr float, ptr addrspace(1) %1, i64 %182, !dbg !22 + %184 = getelementptr bfloat, ptr addrspace(1) %2, i64 %53, !dbg !23 + %185 = getelementptr bfloat, ptr addrspace(1) %2, i64 %54, !dbg !23 + %186 = getelementptr bfloat, ptr addrspace(1) %2, i64 %56, !dbg !23 + %187 = getelementptr bfloat, ptr addrspace(1) %2, i64 %58, !dbg !23 + %188 = getelementptr bfloat, ptr addrspace(1) %2, i64 %60, !dbg !23 + %189 = getelementptr bfloat, ptr addrspace(1) %2, i64 %62, !dbg !23 + %190 = getelementptr bfloat, ptr addrspace(1) %2, i64 %64, !dbg !23 + %191 = getelementptr bfloat, ptr addrspace(1) %2, i64 %66, !dbg !23 + %192 = getelementptr bfloat, ptr addrspace(1) %2, i64 %68, !dbg !23 + %193 = getelementptr bfloat, ptr addrspace(1) %2, i64 %70, !dbg !23 + %194 = getelementptr bfloat, ptr addrspace(1) %2, i64 %72, !dbg !23 + %195 = getelementptr bfloat, ptr addrspace(1) %2, i64 %74, !dbg !23 + %196 = getelementptr bfloat, ptr addrspace(1) %2, i64 %76, !dbg !23 + %197 = getelementptr bfloat, ptr addrspace(1) %2, i64 %78, !dbg !23 + %198 = getelementptr bfloat, ptr addrspace(1) %2, i64 %80, !dbg !23 + %199 = getelementptr bfloat, ptr addrspace(1) %2, i64 %82, !dbg !23 + %200 = getelementptr bfloat, ptr addrspace(1) %2, i64 %84, !dbg !23 + %201 = getelementptr bfloat, ptr addrspace(1) %2, i64 %86, !dbg !23 + %202 = getelementptr bfloat, ptr addrspace(1) %2, i64 %88, !dbg !23 + %203 = getelementptr bfloat, ptr addrspace(1) %2, i64 %90, !dbg !23 + %204 = getelementptr bfloat, ptr addrspace(1) %2, i64 %92, !dbg !23 + %205 = getelementptr bfloat, ptr addrspace(1) %2, i64 %94, !dbg !23 + %206 = getelementptr bfloat, ptr addrspace(1) %2, i64 %96, !dbg !23 + %207 = getelementptr bfloat, ptr addrspace(1) %2, i64 %98, !dbg !23 + %208 = getelementptr bfloat, ptr addrspace(1) %2, i64 %100, !dbg !23 + %209 = getelementptr bfloat, ptr addrspace(1) %2, i64 %102, !dbg !23 + %210 = getelementptr bfloat, ptr addrspace(1) %2, i64 %104, !dbg !23 + %211 = getelementptr bfloat, ptr addrspace(1) %2, i64 %106, !dbg !23 + %212 = getelementptr bfloat, ptr addrspace(1) %2, i64 %108, !dbg !23 + %213 = getelementptr bfloat, ptr addrspace(1) %2, i64 %110, !dbg !23 + %214 = getelementptr bfloat, ptr addrspace(1) %2, i64 %112, !dbg !23 + %215 = getelementptr bfloat, ptr addrspace(1) %2, i64 %114, !dbg !23 + %216 = getelementptr bfloat, ptr addrspace(1) %184, i64 %147, !dbg !24 + %217 = getelementptr bfloat, ptr addrspace(1) %185, i64 %147, !dbg !24 + %218 = getelementptr bfloat, ptr addrspace(1) %186, i64 %147, !dbg !24 + %219 = getelementptr bfloat, ptr addrspace(1) %187, i64 %147, !dbg !24 + %220 = getelementptr bfloat, ptr addrspace(1) %188, i64 %147, !dbg !24 + %221 = getelementptr bfloat, ptr addrspace(1) %189, i64 %147, !dbg !24 + %222 = getelementptr bfloat, ptr addrspace(1) %190, i64 %147, !dbg !24 + %223 = getelementptr bfloat, ptr addrspace(1) %191, i64 %147, !dbg !24 + %224 = getelementptr bfloat, ptr addrspace(1) %192, i64 %147, !dbg !24 + %225 = getelementptr bfloat, ptr addrspace(1) %193, i64 %147, !dbg !24 + %226 = getelementptr bfloat, ptr addrspace(1) %194, i64 %147, !dbg !24 + %227 = getelementptr bfloat, ptr addrspace(1) %195, i64 %147, !dbg !24 + %228 = getelementptr bfloat, ptr addrspace(1) %196, i64 %147, !dbg !24 + %229 = getelementptr bfloat, ptr addrspace(1) %197, i64 %147, !dbg !24 + %230 = getelementptr bfloat, ptr addrspace(1) %198, i64 %147, !dbg !24 + %231 = getelementptr bfloat, ptr addrspace(1) %199, i64 %147, !dbg !24 + %232 = getelementptr bfloat, ptr addrspace(1) %200, i64 %147, !dbg !24 + %233 = getelementptr bfloat, ptr addrspace(1) %201, i64 %147, !dbg !24 + %234 = getelementptr bfloat, ptr addrspace(1) %202, i64 %147, !dbg !24 + %235 = getelementptr bfloat, ptr addrspace(1) %203, i64 %147, !dbg !24 + %236 = getelementptr bfloat, ptr addrspace(1) %204, i64 %147, !dbg !24 + %237 = getelementptr bfloat, ptr addrspace(1) %205, i64 %147, !dbg !24 + %238 = getelementptr bfloat, ptr addrspace(1) %206, i64 %147, !dbg !24 + %239 = getelementptr bfloat, ptr addrspace(1) %207, i64 %147, !dbg !24 + %240 = getelementptr bfloat, ptr addrspace(1) %208, i64 %147, !dbg !24 + %241 = getelementptr bfloat, ptr addrspace(1) %209, i64 %147, !dbg !24 + %242 = getelementptr bfloat, ptr addrspace(1) %210, i64 %147, !dbg !24 + %243 = getelementptr bfloat, ptr addrspace(1) %211, i64 %147, !dbg !24 + %244 = getelementptr bfloat, ptr addrspace(1) %212, i64 %147, !dbg !24 + %245 = getelementptr bfloat, ptr addrspace(1) %213, i64 %147, !dbg !24 + %246 = getelementptr bfloat, ptr addrspace(1) %214, i64 %147, !dbg !24 + %247 = getelementptr bfloat, ptr addrspace(1) %215, i64 %147, !dbg !24 + %248 = getelementptr bfloat, ptr addrspace(1) %3, i64 %53, !dbg !25 + %249 = getelementptr bfloat, ptr addrspace(1) %3, i64 %54, !dbg !25 + %250 = getelementptr bfloat, ptr addrspace(1) %3, i64 %56, !dbg !25 + %251 = getelementptr bfloat, ptr addrspace(1) %3, i64 %58, !dbg !25 + %252 = getelementptr bfloat, ptr addrspace(1) %3, i64 %60, !dbg !25 + %253 = getelementptr bfloat, ptr addrspace(1) %3, i64 %62, !dbg !25 + %254 = getelementptr bfloat, ptr addrspace(1) %3, i64 %64, !dbg !25 + %255 = getelementptr bfloat, ptr addrspace(1) %3, i64 %66, !dbg !25 + %256 = getelementptr bfloat, ptr addrspace(1) %3, i64 %68, !dbg !25 + %257 = getelementptr bfloat, ptr addrspace(1) %3, i64 %70, !dbg !25 + %258 = getelementptr bfloat, ptr addrspace(1) %3, i64 %72, !dbg !25 + %259 = getelementptr bfloat, ptr addrspace(1) %3, i64 %74, !dbg !25 + %260 = getelementptr bfloat, ptr addrspace(1) %3, i64 %76, !dbg !25 + %261 = getelementptr bfloat, ptr addrspace(1) %3, i64 %78, !dbg !25 + %262 = getelementptr bfloat, ptr addrspace(1) %3, i64 %80, !dbg !25 + %263 = getelementptr bfloat, ptr addrspace(1) %3, i64 %82, !dbg !25 + %264 = getelementptr bfloat, ptr addrspace(1) %3, i64 %84, !dbg !25 + %265 = getelementptr bfloat, ptr addrspace(1) %3, i64 %86, !dbg !25 + %266 = getelementptr bfloat, ptr addrspace(1) %3, i64 %88, !dbg !25 + %267 = getelementptr bfloat, ptr addrspace(1) %3, i64 %90, !dbg !25 + %268 = getelementptr bfloat, ptr addrspace(1) %3, i64 %92, !dbg !25 + %269 = getelementptr bfloat, ptr addrspace(1) %3, i64 %94, !dbg !25 + %270 = getelementptr bfloat, ptr addrspace(1) %3, i64 %96, !dbg !25 + %271 = getelementptr bfloat, ptr addrspace(1) %3, i64 %98, !dbg !25 + %272 = getelementptr bfloat, ptr addrspace(1) %3, i64 %100, !dbg !25 + %273 = getelementptr bfloat, ptr addrspace(1) %3, i64 %102, !dbg !25 + %274 = getelementptr bfloat, ptr addrspace(1) %3, i64 %104, !dbg !25 + %275 = getelementptr bfloat, ptr addrspace(1) %3, i64 %106, !dbg !25 + %276 = getelementptr bfloat, ptr addrspace(1) %3, i64 %108, !dbg !25 + %277 = getelementptr bfloat, ptr addrspace(1) %3, i64 %110, !dbg !25 + %278 = getelementptr bfloat, ptr addrspace(1) %3, i64 %112, !dbg !25 + %279 = getelementptr bfloat, ptr addrspace(1) %3, i64 %114, !dbg !25 + %280 = getelementptr bfloat, ptr addrspace(1) %248, i64 %147, !dbg !26 + %281 = getelementptr bfloat, ptr addrspace(1) %249, i64 %147, !dbg !26 + %282 = getelementptr bfloat, ptr addrspace(1) %250, i64 %147, !dbg !26 + %283 = getelementptr bfloat, ptr addrspace(1) %251, i64 %147, !dbg !26 + %284 = getelementptr bfloat, ptr addrspace(1) %252, i64 %147, !dbg !26 + %285 = getelementptr bfloat, ptr addrspace(1) %253, i64 %147, !dbg !26 + %286 = getelementptr bfloat, ptr addrspace(1) %254, i64 %147, !dbg !26 + %287 = getelementptr bfloat, ptr addrspace(1) %255, i64 %147, !dbg !26 + %288 = getelementptr bfloat, ptr addrspace(1) %256, i64 %147, !dbg !26 + %289 = getelementptr bfloat, ptr addrspace(1) %257, i64 %147, !dbg !26 + %290 = getelementptr bfloat, ptr addrspace(1) %258, i64 %147, !dbg !26 + %291 = getelementptr bfloat, ptr addrspace(1) %259, i64 %147, !dbg !26 + %292 = getelementptr bfloat, ptr addrspace(1) %260, i64 %147, !dbg !26 + %293 = getelementptr bfloat, ptr addrspace(1) %261, i64 %147, !dbg !26 + %294 = getelementptr bfloat, ptr addrspace(1) %262, i64 %147, !dbg !26 + %295 = getelementptr bfloat, ptr addrspace(1) %263, i64 %147, !dbg !26 + %296 = getelementptr bfloat, ptr addrspace(1) %264, i64 %147, !dbg !26 + %297 = getelementptr bfloat, ptr addrspace(1) %265, i64 %147, !dbg !26 + %298 = getelementptr bfloat, ptr addrspace(1) %266, i64 %147, !dbg !26 + %299 = getelementptr bfloat, ptr addrspace(1) %267, i64 %147, !dbg !26 + %300 = getelementptr bfloat, ptr addrspace(1) %268, i64 %147, !dbg !26 + %301 = getelementptr bfloat, ptr addrspace(1) %269, i64 %147, !dbg !26 + %302 = getelementptr bfloat, ptr addrspace(1) %270, i64 %147, !dbg !26 + %303 = getelementptr bfloat, ptr addrspace(1) %271, i64 %147, !dbg !26 + %304 = getelementptr bfloat, ptr addrspace(1) %272, i64 %147, !dbg !26 + %305 = getelementptr bfloat, ptr addrspace(1) %273, i64 %147, !dbg !26 + %306 = getelementptr bfloat, ptr addrspace(1) %274, i64 %147, !dbg !26 + %307 = getelementptr bfloat, ptr addrspace(1) %275, i64 %147, !dbg !26 + %308 = getelementptr bfloat, ptr addrspace(1) %276, i64 %147, !dbg !26 + %309 = getelementptr bfloat, ptr addrspace(1) %277, i64 %147, !dbg !26 + %310 = getelementptr bfloat, ptr addrspace(1) %278, i64 %147, !dbg !26 + %311 = getelementptr bfloat, ptr addrspace(1) %279, i64 %147, !dbg !26 + %312 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %148, i1 %21) #5, !dbg !27 + %313 = extractvalue { i32, i32, i32, i32 } %312, 0, !dbg !27 + %314 = bitcast i32 %313 to <2 x bfloat>, !dbg !27 + %315 = extractvalue { i32, i32, i32, i32 } %312, 1, !dbg !27 + %316 = bitcast i32 %315 to <2 x bfloat>, !dbg !27 + %317 = extractvalue { i32, i32, i32, i32 } %312, 2, !dbg !27 + %318 = bitcast i32 %317 to <2 x bfloat>, !dbg !27 + %319 = extractvalue { i32, i32, i32, i32 } %312, 3, !dbg !27 + %320 = bitcast i32 %319 to <2 x bfloat>, !dbg !27 + %321 = extractelement <2 x bfloat> %314, i64 0, !dbg !27 + %322 = extractelement <2 x bfloat> %314, i64 1, !dbg !27 + %323 = extractelement <2 x bfloat> %316, i64 0, !dbg !27 + %324 = extractelement <2 x bfloat> %316, i64 1, !dbg !27 + %325 = extractelement <2 x bfloat> %318, i64 0, !dbg !27 + %326 = extractelement <2 x bfloat> %318, i64 1, !dbg !27 + %327 = extractelement <2 x bfloat> %320, i64 0, !dbg !27 + %328 = extractelement <2 x bfloat> %320, i64 1, !dbg !27 + %329 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %149, i1 %22) #5, !dbg !27 + %330 = extractvalue { i32, i32, i32, i32 } %329, 0, !dbg !27 + %331 = bitcast i32 %330 to <2 x bfloat>, !dbg !27 + %332 = extractvalue { i32, i32, i32, i32 } %329, 1, !dbg !27 + %333 = bitcast i32 %332 to <2 x bfloat>, !dbg !27 + %334 = extractvalue { i32, i32, i32, i32 } %329, 2, !dbg !27 + %335 = bitcast i32 %334 to <2 x bfloat>, !dbg !27 + %336 = extractvalue { i32, i32, i32, i32 } %329, 3, !dbg !27 + %337 = bitcast i32 %336 to <2 x bfloat>, !dbg !27 + %338 = extractelement <2 x bfloat> %331, i64 0, !dbg !27 + %339 = extractelement <2 x bfloat> %331, i64 1, !dbg !27 + %340 = extractelement <2 x bfloat> %333, i64 0, !dbg !27 + %341 = extractelement <2 x bfloat> %333, i64 1, !dbg !27 + %342 = extractelement <2 x bfloat> %335, i64 0, !dbg !27 + %343 = extractelement <2 x bfloat> %335, i64 1, !dbg !27 + %344 = extractelement <2 x bfloat> %337, i64 0, !dbg !27 + %345 = extractelement <2 x bfloat> %337, i64 1, !dbg !27 + %346 = extractelement <32 x i1> %50, i64 31, !dbg !27 + %347 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %150, i1 %346) #5, !dbg !27 + %348 = extractvalue { i32, i32, i32, i32 } %347, 0, !dbg !27 + %349 = bitcast i32 %348 to <2 x bfloat>, !dbg !27 + %350 = extractvalue { i32, i32, i32, i32 } %347, 1, !dbg !27 + %351 = bitcast i32 %350 to <2 x bfloat>, !dbg !27 + %352 = extractvalue { i32, i32, i32, i32 } %347, 2, !dbg !27 + %353 = bitcast i32 %352 to <2 x bfloat>, !dbg !27 + %354 = extractvalue { i32, i32, i32, i32 } %347, 3, !dbg !27 + %355 = bitcast i32 %354 to <2 x bfloat>, !dbg !27 + %356 = extractelement <2 x bfloat> %349, i64 0, !dbg !27 + %357 = extractelement <2 x bfloat> %349, i64 1, !dbg !27 + %358 = extractelement <2 x bfloat> %351, i64 0, !dbg !27 + %359 = extractelement <2 x bfloat> %351, i64 1, !dbg !27 + %360 = extractelement <2 x bfloat> %353, i64 0, !dbg !27 + %361 = extractelement <2 x bfloat> %353, i64 1, !dbg !27 + %362 = extractelement <2 x bfloat> %355, i64 0, !dbg !27 + %363 = extractelement <2 x bfloat> %355, i64 1, !dbg !27 + %364 = extractelement <32 x i1> %50, i64 30, !dbg !27 + %365 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %151, i1 %364) #5, !dbg !27 + %366 = extractvalue { i32, i32, i32, i32 } %365, 0, !dbg !27 + %367 = bitcast i32 %366 to <2 x bfloat>, !dbg !27 + %368 = extractvalue { i32, i32, i32, i32 } %365, 1, !dbg !27 + %369 = bitcast i32 %368 to <2 x bfloat>, !dbg !27 + %370 = extractvalue { i32, i32, i32, i32 } %365, 2, !dbg !27 + %371 = bitcast i32 %370 to <2 x bfloat>, !dbg !27 + %372 = extractvalue { i32, i32, i32, i32 } %365, 3, !dbg !27 + %373 = bitcast i32 %372 to <2 x bfloat>, !dbg !27 + %374 = extractelement <2 x bfloat> %367, i64 0, !dbg !27 + %375 = extractelement <2 x bfloat> %367, i64 1, !dbg !27 + %376 = extractelement <2 x bfloat> %369, i64 0, !dbg !27 + %377 = extractelement <2 x bfloat> %369, i64 1, !dbg !27 + %378 = extractelement <2 x bfloat> %371, i64 0, !dbg !27 + %379 = extractelement <2 x bfloat> %371, i64 1, !dbg !27 + %380 = extractelement <2 x bfloat> %373, i64 0, !dbg !27 + %381 = extractelement <2 x bfloat> %373, i64 1, !dbg !27 + %382 = extractelement <32 x i1> %50, i64 29, !dbg !27 + %383 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %152, i1 %382) #5, !dbg !27 + %384 = extractvalue { i32, i32, i32, i32 } %383, 0, !dbg !27 + %385 = bitcast i32 %384 to <2 x bfloat>, !dbg !27 + %386 = extractvalue { i32, i32, i32, i32 } %383, 1, !dbg !27 + %387 = bitcast i32 %386 to <2 x bfloat>, !dbg !27 + %388 = extractvalue { i32, i32, i32, i32 } %383, 2, !dbg !27 + %389 = bitcast i32 %388 to <2 x bfloat>, !dbg !27 + %390 = extractvalue { i32, i32, i32, i32 } %383, 3, !dbg !27 + %391 = bitcast i32 %390 to <2 x bfloat>, !dbg !27 + %392 = extractelement <2 x bfloat> %385, i64 0, !dbg !27 + %393 = extractelement <2 x bfloat> %385, i64 1, !dbg !27 + %394 = extractelement <2 x bfloat> %387, i64 0, !dbg !27 + %395 = extractelement <2 x bfloat> %387, i64 1, !dbg !27 + %396 = extractelement <2 x bfloat> %389, i64 0, !dbg !27 + %397 = extractelement <2 x bfloat> %389, i64 1, !dbg !27 + %398 = extractelement <2 x bfloat> %391, i64 0, !dbg !27 + %399 = extractelement <2 x bfloat> %391, i64 1, !dbg !27 + %400 = extractelement <32 x i1> %50, i64 28, !dbg !27 + %401 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %153, i1 %400) #5, !dbg !27 + %402 = extractvalue { i32, i32, i32, i32 } %401, 0, !dbg !27 + %403 = bitcast i32 %402 to <2 x bfloat>, !dbg !27 + %404 = extractvalue { i32, i32, i32, i32 } %401, 1, !dbg !27 + %405 = bitcast i32 %404 to <2 x bfloat>, !dbg !27 + %406 = extractvalue { i32, i32, i32, i32 } %401, 2, !dbg !27 + %407 = bitcast i32 %406 to <2 x bfloat>, !dbg !27 + %408 = extractvalue { i32, i32, i32, i32 } %401, 3, !dbg !27 + %409 = bitcast i32 %408 to <2 x bfloat>, !dbg !27 + %410 = extractelement <2 x bfloat> %403, i64 0, !dbg !27 + %411 = extractelement <2 x bfloat> %403, i64 1, !dbg !27 + %412 = extractelement <2 x bfloat> %405, i64 0, !dbg !27 + %413 = extractelement <2 x bfloat> %405, i64 1, !dbg !27 + %414 = extractelement <2 x bfloat> %407, i64 0, !dbg !27 + %415 = extractelement <2 x bfloat> %407, i64 1, !dbg !27 + %416 = extractelement <2 x bfloat> %409, i64 0, !dbg !27 + %417 = extractelement <2 x bfloat> %409, i64 1, !dbg !27 + %418 = extractelement <32 x i1> %50, i64 27, !dbg !27 + %419 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %154, i1 %418) #5, !dbg !27 + %420 = extractvalue { i32, i32, i32, i32 } %419, 0, !dbg !27 + %421 = bitcast i32 %420 to <2 x bfloat>, !dbg !27 + %422 = extractvalue { i32, i32, i32, i32 } %419, 1, !dbg !27 + %423 = bitcast i32 %422 to <2 x bfloat>, !dbg !27 + %424 = extractvalue { i32, i32, i32, i32 } %419, 2, !dbg !27 + %425 = bitcast i32 %424 to <2 x bfloat>, !dbg !27 + %426 = extractvalue { i32, i32, i32, i32 } %419, 3, !dbg !27 + %427 = bitcast i32 %426 to <2 x bfloat>, !dbg !27 + %428 = extractelement <2 x bfloat> %421, i64 0, !dbg !27 + %429 = extractelement <2 x bfloat> %421, i64 1, !dbg !27 + %430 = extractelement <2 x bfloat> %423, i64 0, !dbg !27 + %431 = extractelement <2 x bfloat> %423, i64 1, !dbg !27 + %432 = extractelement <2 x bfloat> %425, i64 0, !dbg !27 + %433 = extractelement <2 x bfloat> %425, i64 1, !dbg !27 + %434 = extractelement <2 x bfloat> %427, i64 0, !dbg !27 + %435 = extractelement <2 x bfloat> %427, i64 1, !dbg !27 + %436 = extractelement <32 x i1> %50, i64 26, !dbg !27 + %437 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %155, i1 %436) #5, !dbg !27 + %438 = extractvalue { i32, i32, i32, i32 } %437, 0, !dbg !27 + %439 = bitcast i32 %438 to <2 x bfloat>, !dbg !27 + %440 = extractvalue { i32, i32, i32, i32 } %437, 1, !dbg !27 + %441 = bitcast i32 %440 to <2 x bfloat>, !dbg !27 + %442 = extractvalue { i32, i32, i32, i32 } %437, 2, !dbg !27 + %443 = bitcast i32 %442 to <2 x bfloat>, !dbg !27 + %444 = extractvalue { i32, i32, i32, i32 } %437, 3, !dbg !27 + %445 = bitcast i32 %444 to <2 x bfloat>, !dbg !27 + %446 = extractelement <2 x bfloat> %439, i64 0, !dbg !27 + %447 = extractelement <2 x bfloat> %439, i64 1, !dbg !27 + %448 = extractelement <2 x bfloat> %441, i64 0, !dbg !27 + %449 = extractelement <2 x bfloat> %441, i64 1, !dbg !27 + %450 = extractelement <2 x bfloat> %443, i64 0, !dbg !27 + %451 = extractelement <2 x bfloat> %443, i64 1, !dbg !27 + %452 = extractelement <2 x bfloat> %445, i64 0, !dbg !27 + %453 = extractelement <2 x bfloat> %445, i64 1, !dbg !27 + %454 = extractelement <32 x i1> %50, i64 25, !dbg !27 + %455 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %156, i1 %454) #5, !dbg !27 + %456 = extractvalue { i32, i32, i32, i32 } %455, 0, !dbg !27 + %457 = bitcast i32 %456 to <2 x bfloat>, !dbg !27 + %458 = extractvalue { i32, i32, i32, i32 } %455, 1, !dbg !27 + %459 = bitcast i32 %458 to <2 x bfloat>, !dbg !27 + %460 = extractvalue { i32, i32, i32, i32 } %455, 2, !dbg !27 + %461 = bitcast i32 %460 to <2 x bfloat>, !dbg !27 + %462 = extractvalue { i32, i32, i32, i32 } %455, 3, !dbg !27 + %463 = bitcast i32 %462 to <2 x bfloat>, !dbg !27 + %464 = extractelement <2 x bfloat> %457, i64 0, !dbg !27 + %465 = extractelement <2 x bfloat> %457, i64 1, !dbg !27 + %466 = extractelement <2 x bfloat> %459, i64 0, !dbg !27 + %467 = extractelement <2 x bfloat> %459, i64 1, !dbg !27 + %468 = extractelement <2 x bfloat> %461, i64 0, !dbg !27 + %469 = extractelement <2 x bfloat> %461, i64 1, !dbg !27 + %470 = extractelement <2 x bfloat> %463, i64 0, !dbg !27 + %471 = extractelement <2 x bfloat> %463, i64 1, !dbg !27 + %472 = extractelement <32 x i1> %50, i64 24, !dbg !27 + %473 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %157, i1 %472) #5, !dbg !27 + %474 = extractvalue { i32, i32, i32, i32 } %473, 0, !dbg !27 + %475 = bitcast i32 %474 to <2 x bfloat>, !dbg !27 + %476 = extractvalue { i32, i32, i32, i32 } %473, 1, !dbg !27 + %477 = bitcast i32 %476 to <2 x bfloat>, !dbg !27 + %478 = extractvalue { i32, i32, i32, i32 } %473, 2, !dbg !27 + %479 = bitcast i32 %478 to <2 x bfloat>, !dbg !27 + %480 = extractvalue { i32, i32, i32, i32 } %473, 3, !dbg !27 + %481 = bitcast i32 %480 to <2 x bfloat>, !dbg !27 + %482 = extractelement <2 x bfloat> %475, i64 0, !dbg !27 + %483 = extractelement <2 x bfloat> %475, i64 1, !dbg !27 + %484 = extractelement <2 x bfloat> %477, i64 0, !dbg !27 + %485 = extractelement <2 x bfloat> %477, i64 1, !dbg !27 + %486 = extractelement <2 x bfloat> %479, i64 0, !dbg !27 + %487 = extractelement <2 x bfloat> %479, i64 1, !dbg !27 + %488 = extractelement <2 x bfloat> %481, i64 0, !dbg !27 + %489 = extractelement <2 x bfloat> %481, i64 1, !dbg !27 + %490 = extractelement <32 x i1> %50, i64 23, !dbg !27 + %491 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %158, i1 %490) #5, !dbg !27 + %492 = extractvalue { i32, i32, i32, i32 } %491, 0, !dbg !27 + %493 = bitcast i32 %492 to <2 x bfloat>, !dbg !27 + %494 = extractvalue { i32, i32, i32, i32 } %491, 1, !dbg !27 + %495 = bitcast i32 %494 to <2 x bfloat>, !dbg !27 + %496 = extractvalue { i32, i32, i32, i32 } %491, 2, !dbg !27 + %497 = bitcast i32 %496 to <2 x bfloat>, !dbg !27 + %498 = extractvalue { i32, i32, i32, i32 } %491, 3, !dbg !27 + %499 = bitcast i32 %498 to <2 x bfloat>, !dbg !27 + %500 = extractelement <2 x bfloat> %493, i64 0, !dbg !27 + %501 = extractelement <2 x bfloat> %493, i64 1, !dbg !27 + %502 = extractelement <2 x bfloat> %495, i64 0, !dbg !27 + %503 = extractelement <2 x bfloat> %495, i64 1, !dbg !27 + %504 = extractelement <2 x bfloat> %497, i64 0, !dbg !27 + %505 = extractelement <2 x bfloat> %497, i64 1, !dbg !27 + %506 = extractelement <2 x bfloat> %499, i64 0, !dbg !27 + %507 = extractelement <2 x bfloat> %499, i64 1, !dbg !27 + %508 = extractelement <32 x i1> %50, i64 22, !dbg !27 + %509 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %159, i1 %508) #5, !dbg !27 + %510 = extractvalue { i32, i32, i32, i32 } %509, 0, !dbg !27 + %511 = bitcast i32 %510 to <2 x bfloat>, !dbg !27 + %512 = extractvalue { i32, i32, i32, i32 } %509, 1, !dbg !27 + %513 = bitcast i32 %512 to <2 x bfloat>, !dbg !27 + %514 = extractvalue { i32, i32, i32, i32 } %509, 2, !dbg !27 + %515 = bitcast i32 %514 to <2 x bfloat>, !dbg !27 + %516 = extractvalue { i32, i32, i32, i32 } %509, 3, !dbg !27 + %517 = bitcast i32 %516 to <2 x bfloat>, !dbg !27 + %518 = extractelement <2 x bfloat> %511, i64 0, !dbg !27 + %519 = extractelement <2 x bfloat> %511, i64 1, !dbg !27 + %520 = extractelement <2 x bfloat> %513, i64 0, !dbg !27 + %521 = extractelement <2 x bfloat> %513, i64 1, !dbg !27 + %522 = extractelement <2 x bfloat> %515, i64 0, !dbg !27 + %523 = extractelement <2 x bfloat> %515, i64 1, !dbg !27 + %524 = extractelement <2 x bfloat> %517, i64 0, !dbg !27 + %525 = extractelement <2 x bfloat> %517, i64 1, !dbg !27 + %526 = extractelement <32 x i1> %50, i64 21, !dbg !27 + %527 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %160, i1 %526) #5, !dbg !27 + %528 = extractvalue { i32, i32, i32, i32 } %527, 0, !dbg !27 + %529 = bitcast i32 %528 to <2 x bfloat>, !dbg !27 + %530 = extractvalue { i32, i32, i32, i32 } %527, 1, !dbg !27 + %531 = bitcast i32 %530 to <2 x bfloat>, !dbg !27 + %532 = extractvalue { i32, i32, i32, i32 } %527, 2, !dbg !27 + %533 = bitcast i32 %532 to <2 x bfloat>, !dbg !27 + %534 = extractvalue { i32, i32, i32, i32 } %527, 3, !dbg !27 + %535 = bitcast i32 %534 to <2 x bfloat>, !dbg !27 + %536 = extractelement <2 x bfloat> %529, i64 0, !dbg !27 + %537 = extractelement <2 x bfloat> %529, i64 1, !dbg !27 + %538 = extractelement <2 x bfloat> %531, i64 0, !dbg !27 + %539 = extractelement <2 x bfloat> %531, i64 1, !dbg !27 + %540 = extractelement <2 x bfloat> %533, i64 0, !dbg !27 + %541 = extractelement <2 x bfloat> %533, i64 1, !dbg !27 + %542 = extractelement <2 x bfloat> %535, i64 0, !dbg !27 + %543 = extractelement <2 x bfloat> %535, i64 1, !dbg !27 + %544 = extractelement <32 x i1> %50, i64 20, !dbg !27 + %545 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %161, i1 %544) #5, !dbg !27 + %546 = extractvalue { i32, i32, i32, i32 } %545, 0, !dbg !27 + %547 = bitcast i32 %546 to <2 x bfloat>, !dbg !27 + %548 = extractvalue { i32, i32, i32, i32 } %545, 1, !dbg !27 + %549 = bitcast i32 %548 to <2 x bfloat>, !dbg !27 + %550 = extractvalue { i32, i32, i32, i32 } %545, 2, !dbg !27 + %551 = bitcast i32 %550 to <2 x bfloat>, !dbg !27 + %552 = extractvalue { i32, i32, i32, i32 } %545, 3, !dbg !27 + %553 = bitcast i32 %552 to <2 x bfloat>, !dbg !27 + %554 = extractelement <2 x bfloat> %547, i64 0, !dbg !27 + %555 = extractelement <2 x bfloat> %547, i64 1, !dbg !27 + %556 = extractelement <2 x bfloat> %549, i64 0, !dbg !27 + %557 = extractelement <2 x bfloat> %549, i64 1, !dbg !27 + %558 = extractelement <2 x bfloat> %551, i64 0, !dbg !27 + %559 = extractelement <2 x bfloat> %551, i64 1, !dbg !27 + %560 = extractelement <2 x bfloat> %553, i64 0, !dbg !27 + %561 = extractelement <2 x bfloat> %553, i64 1, !dbg !27 + %562 = extractelement <32 x i1> %50, i64 19, !dbg !27 + %563 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %162, i1 %562) #5, !dbg !27 + %564 = extractvalue { i32, i32, i32, i32 } %563, 0, !dbg !27 + %565 = bitcast i32 %564 to <2 x bfloat>, !dbg !27 + %566 = extractvalue { i32, i32, i32, i32 } %563, 1, !dbg !27 + %567 = bitcast i32 %566 to <2 x bfloat>, !dbg !27 + %568 = extractvalue { i32, i32, i32, i32 } %563, 2, !dbg !27 + %569 = bitcast i32 %568 to <2 x bfloat>, !dbg !27 + %570 = extractvalue { i32, i32, i32, i32 } %563, 3, !dbg !27 + %571 = bitcast i32 %570 to <2 x bfloat>, !dbg !27 + %572 = extractelement <2 x bfloat> %565, i64 0, !dbg !27 + %573 = extractelement <2 x bfloat> %565, i64 1, !dbg !27 + %574 = extractelement <2 x bfloat> %567, i64 0, !dbg !27 + %575 = extractelement <2 x bfloat> %567, i64 1, !dbg !27 + %576 = extractelement <2 x bfloat> %569, i64 0, !dbg !27 + %577 = extractelement <2 x bfloat> %569, i64 1, !dbg !27 + %578 = extractelement <2 x bfloat> %571, i64 0, !dbg !27 + %579 = extractelement <2 x bfloat> %571, i64 1, !dbg !27 + %580 = extractelement <32 x i1> %50, i64 18, !dbg !27 + %581 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %163, i1 %580) #5, !dbg !27 + %582 = extractvalue { i32, i32, i32, i32 } %581, 0, !dbg !27 + %583 = bitcast i32 %582 to <2 x bfloat>, !dbg !27 + %584 = extractvalue { i32, i32, i32, i32 } %581, 1, !dbg !27 + %585 = bitcast i32 %584 to <2 x bfloat>, !dbg !27 + %586 = extractvalue { i32, i32, i32, i32 } %581, 2, !dbg !27 + %587 = bitcast i32 %586 to <2 x bfloat>, !dbg !27 + %588 = extractvalue { i32, i32, i32, i32 } %581, 3, !dbg !27 + %589 = bitcast i32 %588 to <2 x bfloat>, !dbg !27 + %590 = extractelement <2 x bfloat> %583, i64 0, !dbg !27 + %591 = extractelement <2 x bfloat> %583, i64 1, !dbg !27 + %592 = extractelement <2 x bfloat> %585, i64 0, !dbg !27 + %593 = extractelement <2 x bfloat> %585, i64 1, !dbg !27 + %594 = extractelement <2 x bfloat> %587, i64 0, !dbg !27 + %595 = extractelement <2 x bfloat> %587, i64 1, !dbg !27 + %596 = extractelement <2 x bfloat> %589, i64 0, !dbg !27 + %597 = extractelement <2 x bfloat> %589, i64 1, !dbg !27 + %598 = extractelement <32 x i1> %50, i64 17, !dbg !27 + %599 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %164, i1 %598) #5, !dbg !27 + %600 = extractvalue { i32, i32, i32, i32 } %599, 0, !dbg !27 + %601 = bitcast i32 %600 to <2 x bfloat>, !dbg !27 + %602 = extractvalue { i32, i32, i32, i32 } %599, 1, !dbg !27 + %603 = bitcast i32 %602 to <2 x bfloat>, !dbg !27 + %604 = extractvalue { i32, i32, i32, i32 } %599, 2, !dbg !27 + %605 = bitcast i32 %604 to <2 x bfloat>, !dbg !27 + %606 = extractvalue { i32, i32, i32, i32 } %599, 3, !dbg !27 + %607 = bitcast i32 %606 to <2 x bfloat>, !dbg !27 + %608 = extractelement <2 x bfloat> %601, i64 0, !dbg !27 + %609 = extractelement <2 x bfloat> %601, i64 1, !dbg !27 + %610 = extractelement <2 x bfloat> %603, i64 0, !dbg !27 + %611 = extractelement <2 x bfloat> %603, i64 1, !dbg !27 + %612 = extractelement <2 x bfloat> %605, i64 0, !dbg !27 + %613 = extractelement <2 x bfloat> %605, i64 1, !dbg !27 + %614 = extractelement <2 x bfloat> %607, i64 0, !dbg !27 + %615 = extractelement <2 x bfloat> %607, i64 1, !dbg !27 + %616 = extractelement <32 x i1> %50, i64 16, !dbg !27 + %617 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %165, i1 %616) #5, !dbg !27 + %618 = extractvalue { i32, i32, i32, i32 } %617, 0, !dbg !27 + %619 = bitcast i32 %618 to <2 x bfloat>, !dbg !27 + %620 = extractvalue { i32, i32, i32, i32 } %617, 1, !dbg !27 + %621 = bitcast i32 %620 to <2 x bfloat>, !dbg !27 + %622 = extractvalue { i32, i32, i32, i32 } %617, 2, !dbg !27 + %623 = bitcast i32 %622 to <2 x bfloat>, !dbg !27 + %624 = extractvalue { i32, i32, i32, i32 } %617, 3, !dbg !27 + %625 = bitcast i32 %624 to <2 x bfloat>, !dbg !27 + %626 = extractelement <2 x bfloat> %619, i64 0, !dbg !27 + %627 = extractelement <2 x bfloat> %619, i64 1, !dbg !27 + %628 = extractelement <2 x bfloat> %621, i64 0, !dbg !27 + %629 = extractelement <2 x bfloat> %621, i64 1, !dbg !27 + %630 = extractelement <2 x bfloat> %623, i64 0, !dbg !27 + %631 = extractelement <2 x bfloat> %623, i64 1, !dbg !27 + %632 = extractelement <2 x bfloat> %625, i64 0, !dbg !27 + %633 = extractelement <2 x bfloat> %625, i64 1, !dbg !27 + %634 = extractelement <32 x i1> %50, i64 15, !dbg !27 + %635 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %166, i1 %634) #5, !dbg !27 + %636 = extractvalue { i32, i32, i32, i32 } %635, 0, !dbg !27 + %637 = bitcast i32 %636 to <2 x bfloat>, !dbg !27 + %638 = extractvalue { i32, i32, i32, i32 } %635, 1, !dbg !27 + %639 = bitcast i32 %638 to <2 x bfloat>, !dbg !27 + %640 = extractvalue { i32, i32, i32, i32 } %635, 2, !dbg !27 + %641 = bitcast i32 %640 to <2 x bfloat>, !dbg !27 + %642 = extractvalue { i32, i32, i32, i32 } %635, 3, !dbg !27 + %643 = bitcast i32 %642 to <2 x bfloat>, !dbg !27 + %644 = extractelement <2 x bfloat> %637, i64 0, !dbg !27 + %645 = extractelement <2 x bfloat> %637, i64 1, !dbg !27 + %646 = extractelement <2 x bfloat> %639, i64 0, !dbg !27 + %647 = extractelement <2 x bfloat> %639, i64 1, !dbg !27 + %648 = extractelement <2 x bfloat> %641, i64 0, !dbg !27 + %649 = extractelement <2 x bfloat> %641, i64 1, !dbg !27 + %650 = extractelement <2 x bfloat> %643, i64 0, !dbg !27 + %651 = extractelement <2 x bfloat> %643, i64 1, !dbg !27 + %652 = extractelement <32 x i1> %50, i64 14, !dbg !27 + %653 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %167, i1 %652) #5, !dbg !27 + %654 = extractvalue { i32, i32, i32, i32 } %653, 0, !dbg !27 + %655 = bitcast i32 %654 to <2 x bfloat>, !dbg !27 + %656 = extractvalue { i32, i32, i32, i32 } %653, 1, !dbg !27 + %657 = bitcast i32 %656 to <2 x bfloat>, !dbg !27 + %658 = extractvalue { i32, i32, i32, i32 } %653, 2, !dbg !27 + %659 = bitcast i32 %658 to <2 x bfloat>, !dbg !27 + %660 = extractvalue { i32, i32, i32, i32 } %653, 3, !dbg !27 + %661 = bitcast i32 %660 to <2 x bfloat>, !dbg !27 + %662 = extractelement <2 x bfloat> %655, i64 0, !dbg !27 + %663 = extractelement <2 x bfloat> %655, i64 1, !dbg !27 + %664 = extractelement <2 x bfloat> %657, i64 0, !dbg !27 + %665 = extractelement <2 x bfloat> %657, i64 1, !dbg !27 + %666 = extractelement <2 x bfloat> %659, i64 0, !dbg !27 + %667 = extractelement <2 x bfloat> %659, i64 1, !dbg !27 + %668 = extractelement <2 x bfloat> %661, i64 0, !dbg !27 + %669 = extractelement <2 x bfloat> %661, i64 1, !dbg !27 + %670 = extractelement <32 x i1> %50, i64 13, !dbg !27 + %671 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %168, i1 %670) #5, !dbg !27 + %672 = extractvalue { i32, i32, i32, i32 } %671, 0, !dbg !27 + %673 = bitcast i32 %672 to <2 x bfloat>, !dbg !27 + %674 = extractvalue { i32, i32, i32, i32 } %671, 1, !dbg !27 + %675 = bitcast i32 %674 to <2 x bfloat>, !dbg !27 + %676 = extractvalue { i32, i32, i32, i32 } %671, 2, !dbg !27 + %677 = bitcast i32 %676 to <2 x bfloat>, !dbg !27 + %678 = extractvalue { i32, i32, i32, i32 } %671, 3, !dbg !27 + %679 = bitcast i32 %678 to <2 x bfloat>, !dbg !27 + %680 = extractelement <2 x bfloat> %673, i64 0, !dbg !27 + %681 = extractelement <2 x bfloat> %673, i64 1, !dbg !27 + %682 = extractelement <2 x bfloat> %675, i64 0, !dbg !27 + %683 = extractelement <2 x bfloat> %675, i64 1, !dbg !27 + %684 = extractelement <2 x bfloat> %677, i64 0, !dbg !27 + %685 = extractelement <2 x bfloat> %677, i64 1, !dbg !27 + %686 = extractelement <2 x bfloat> %679, i64 0, !dbg !27 + %687 = extractelement <2 x bfloat> %679, i64 1, !dbg !27 + %688 = extractelement <32 x i1> %50, i64 12, !dbg !27 + %689 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %169, i1 %688) #5, !dbg !27 + %690 = extractvalue { i32, i32, i32, i32 } %689, 0, !dbg !27 + %691 = bitcast i32 %690 to <2 x bfloat>, !dbg !27 + %692 = extractvalue { i32, i32, i32, i32 } %689, 1, !dbg !27 + %693 = bitcast i32 %692 to <2 x bfloat>, !dbg !27 + %694 = extractvalue { i32, i32, i32, i32 } %689, 2, !dbg !27 + %695 = bitcast i32 %694 to <2 x bfloat>, !dbg !27 + %696 = extractvalue { i32, i32, i32, i32 } %689, 3, !dbg !27 + %697 = bitcast i32 %696 to <2 x bfloat>, !dbg !27 + %698 = extractelement <2 x bfloat> %691, i64 0, !dbg !27 + %699 = extractelement <2 x bfloat> %691, i64 1, !dbg !27 + %700 = extractelement <2 x bfloat> %693, i64 0, !dbg !27 + %701 = extractelement <2 x bfloat> %693, i64 1, !dbg !27 + %702 = extractelement <2 x bfloat> %695, i64 0, !dbg !27 + %703 = extractelement <2 x bfloat> %695, i64 1, !dbg !27 + %704 = extractelement <2 x bfloat> %697, i64 0, !dbg !27 + %705 = extractelement <2 x bfloat> %697, i64 1, !dbg !27 + %706 = extractelement <32 x i1> %50, i64 11, !dbg !27 + %707 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %170, i1 %706) #5, !dbg !27 + %708 = extractvalue { i32, i32, i32, i32 } %707, 0, !dbg !27 + %709 = bitcast i32 %708 to <2 x bfloat>, !dbg !27 + %710 = extractvalue { i32, i32, i32, i32 } %707, 1, !dbg !27 + %711 = bitcast i32 %710 to <2 x bfloat>, !dbg !27 + %712 = extractvalue { i32, i32, i32, i32 } %707, 2, !dbg !27 + %713 = bitcast i32 %712 to <2 x bfloat>, !dbg !27 + %714 = extractvalue { i32, i32, i32, i32 } %707, 3, !dbg !27 + %715 = bitcast i32 %714 to <2 x bfloat>, !dbg !27 + %716 = extractelement <2 x bfloat> %709, i64 0, !dbg !27 + %717 = extractelement <2 x bfloat> %709, i64 1, !dbg !27 + %718 = extractelement <2 x bfloat> %711, i64 0, !dbg !27 + %719 = extractelement <2 x bfloat> %711, i64 1, !dbg !27 + %720 = extractelement <2 x bfloat> %713, i64 0, !dbg !27 + %721 = extractelement <2 x bfloat> %713, i64 1, !dbg !27 + %722 = extractelement <2 x bfloat> %715, i64 0, !dbg !27 + %723 = extractelement <2 x bfloat> %715, i64 1, !dbg !27 + %724 = extractelement <32 x i1> %50, i64 10, !dbg !27 + %725 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %171, i1 %724) #5, !dbg !27 + %726 = extractvalue { i32, i32, i32, i32 } %725, 0, !dbg !27 + %727 = bitcast i32 %726 to <2 x bfloat>, !dbg !27 + %728 = extractvalue { i32, i32, i32, i32 } %725, 1, !dbg !27 + %729 = bitcast i32 %728 to <2 x bfloat>, !dbg !27 + %730 = extractvalue { i32, i32, i32, i32 } %725, 2, !dbg !27 + %731 = bitcast i32 %730 to <2 x bfloat>, !dbg !27 + %732 = extractvalue { i32, i32, i32, i32 } %725, 3, !dbg !27 + %733 = bitcast i32 %732 to <2 x bfloat>, !dbg !27 + %734 = extractelement <2 x bfloat> %727, i64 0, !dbg !27 + %735 = extractelement <2 x bfloat> %727, i64 1, !dbg !27 + %736 = extractelement <2 x bfloat> %729, i64 0, !dbg !27 + %737 = extractelement <2 x bfloat> %729, i64 1, !dbg !27 + %738 = extractelement <2 x bfloat> %731, i64 0, !dbg !27 + %739 = extractelement <2 x bfloat> %731, i64 1, !dbg !27 + %740 = extractelement <2 x bfloat> %733, i64 0, !dbg !27 + %741 = extractelement <2 x bfloat> %733, i64 1, !dbg !27 + %742 = extractelement <32 x i1> %50, i64 9, !dbg !27 + %743 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %172, i1 %742) #5, !dbg !27 + %744 = extractvalue { i32, i32, i32, i32 } %743, 0, !dbg !27 + %745 = bitcast i32 %744 to <2 x bfloat>, !dbg !27 + %746 = extractvalue { i32, i32, i32, i32 } %743, 1, !dbg !27 + %747 = bitcast i32 %746 to <2 x bfloat>, !dbg !27 + %748 = extractvalue { i32, i32, i32, i32 } %743, 2, !dbg !27 + %749 = bitcast i32 %748 to <2 x bfloat>, !dbg !27 + %750 = extractvalue { i32, i32, i32, i32 } %743, 3, !dbg !27 + %751 = bitcast i32 %750 to <2 x bfloat>, !dbg !27 + %752 = extractelement <2 x bfloat> %745, i64 0, !dbg !27 + %753 = extractelement <2 x bfloat> %745, i64 1, !dbg !27 + %754 = extractelement <2 x bfloat> %747, i64 0, !dbg !27 + %755 = extractelement <2 x bfloat> %747, i64 1, !dbg !27 + %756 = extractelement <2 x bfloat> %749, i64 0, !dbg !27 + %757 = extractelement <2 x bfloat> %749, i64 1, !dbg !27 + %758 = extractelement <2 x bfloat> %751, i64 0, !dbg !27 + %759 = extractelement <2 x bfloat> %751, i64 1, !dbg !27 + %760 = extractelement <32 x i1> %50, i64 8, !dbg !27 + %761 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %173, i1 %760) #5, !dbg !27 + %762 = extractvalue { i32, i32, i32, i32 } %761, 0, !dbg !27 + %763 = bitcast i32 %762 to <2 x bfloat>, !dbg !27 + %764 = extractvalue { i32, i32, i32, i32 } %761, 1, !dbg !27 + %765 = bitcast i32 %764 to <2 x bfloat>, !dbg !27 + %766 = extractvalue { i32, i32, i32, i32 } %761, 2, !dbg !27 + %767 = bitcast i32 %766 to <2 x bfloat>, !dbg !27 + %768 = extractvalue { i32, i32, i32, i32 } %761, 3, !dbg !27 + %769 = bitcast i32 %768 to <2 x bfloat>, !dbg !27 + %770 = extractelement <2 x bfloat> %763, i64 0, !dbg !27 + %771 = extractelement <2 x bfloat> %763, i64 1, !dbg !27 + %772 = extractelement <2 x bfloat> %765, i64 0, !dbg !27 + %773 = extractelement <2 x bfloat> %765, i64 1, !dbg !27 + %774 = extractelement <2 x bfloat> %767, i64 0, !dbg !27 + %775 = extractelement <2 x bfloat> %767, i64 1, !dbg !27 + %776 = extractelement <2 x bfloat> %769, i64 0, !dbg !27 + %777 = extractelement <2 x bfloat> %769, i64 1, !dbg !27 + %778 = extractelement <32 x i1> %50, i64 7, !dbg !27 + %779 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %174, i1 %778) #5, !dbg !27 + %780 = extractvalue { i32, i32, i32, i32 } %779, 0, !dbg !27 + %781 = bitcast i32 %780 to <2 x bfloat>, !dbg !27 + %782 = extractvalue { i32, i32, i32, i32 } %779, 1, !dbg !27 + %783 = bitcast i32 %782 to <2 x bfloat>, !dbg !27 + %784 = extractvalue { i32, i32, i32, i32 } %779, 2, !dbg !27 + %785 = bitcast i32 %784 to <2 x bfloat>, !dbg !27 + %786 = extractvalue { i32, i32, i32, i32 } %779, 3, !dbg !27 + %787 = bitcast i32 %786 to <2 x bfloat>, !dbg !27 + %788 = extractelement <2 x bfloat> %781, i64 0, !dbg !27 + %789 = extractelement <2 x bfloat> %781, i64 1, !dbg !27 + %790 = extractelement <2 x bfloat> %783, i64 0, !dbg !27 + %791 = extractelement <2 x bfloat> %783, i64 1, !dbg !27 + %792 = extractelement <2 x bfloat> %785, i64 0, !dbg !27 + %793 = extractelement <2 x bfloat> %785, i64 1, !dbg !27 + %794 = extractelement <2 x bfloat> %787, i64 0, !dbg !27 + %795 = extractelement <2 x bfloat> %787, i64 1, !dbg !27 + %796 = extractelement <32 x i1> %50, i64 6, !dbg !27 + %797 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %175, i1 %796) #5, !dbg !27 + %798 = extractvalue { i32, i32, i32, i32 } %797, 0, !dbg !27 + %799 = bitcast i32 %798 to <2 x bfloat>, !dbg !27 + %800 = extractvalue { i32, i32, i32, i32 } %797, 1, !dbg !27 + %801 = bitcast i32 %800 to <2 x bfloat>, !dbg !27 + %802 = extractvalue { i32, i32, i32, i32 } %797, 2, !dbg !27 + %803 = bitcast i32 %802 to <2 x bfloat>, !dbg !27 + %804 = extractvalue { i32, i32, i32, i32 } %797, 3, !dbg !27 + %805 = bitcast i32 %804 to <2 x bfloat>, !dbg !27 + %806 = extractelement <2 x bfloat> %799, i64 0, !dbg !27 + %807 = extractelement <2 x bfloat> %799, i64 1, !dbg !27 + %808 = extractelement <2 x bfloat> %801, i64 0, !dbg !27 + %809 = extractelement <2 x bfloat> %801, i64 1, !dbg !27 + %810 = extractelement <2 x bfloat> %803, i64 0, !dbg !27 + %811 = extractelement <2 x bfloat> %803, i64 1, !dbg !27 + %812 = extractelement <2 x bfloat> %805, i64 0, !dbg !27 + %813 = extractelement <2 x bfloat> %805, i64 1, !dbg !27 + %814 = extractelement <32 x i1> %50, i64 5, !dbg !27 + %815 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %176, i1 %814) #5, !dbg !27 + %816 = extractvalue { i32, i32, i32, i32 } %815, 0, !dbg !27 + %817 = bitcast i32 %816 to <2 x bfloat>, !dbg !27 + %818 = extractvalue { i32, i32, i32, i32 } %815, 1, !dbg !27 + %819 = bitcast i32 %818 to <2 x bfloat>, !dbg !27 + %820 = extractvalue { i32, i32, i32, i32 } %815, 2, !dbg !27 + %821 = bitcast i32 %820 to <2 x bfloat>, !dbg !27 + %822 = extractvalue { i32, i32, i32, i32 } %815, 3, !dbg !27 + %823 = bitcast i32 %822 to <2 x bfloat>, !dbg !27 + %824 = extractelement <2 x bfloat> %817, i64 0, !dbg !27 + %825 = extractelement <2 x bfloat> %817, i64 1, !dbg !27 + %826 = extractelement <2 x bfloat> %819, i64 0, !dbg !27 + %827 = extractelement <2 x bfloat> %819, i64 1, !dbg !27 + %828 = extractelement <2 x bfloat> %821, i64 0, !dbg !27 + %829 = extractelement <2 x bfloat> %821, i64 1, !dbg !27 + %830 = extractelement <2 x bfloat> %823, i64 0, !dbg !27 + %831 = extractelement <2 x bfloat> %823, i64 1, !dbg !27 + %832 = extractelement <32 x i1> %50, i64 4, !dbg !27 + %833 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %177, i1 %832) #5, !dbg !27 + %834 = extractvalue { i32, i32, i32, i32 } %833, 0, !dbg !27 + %835 = bitcast i32 %834 to <2 x bfloat>, !dbg !27 + %836 = extractvalue { i32, i32, i32, i32 } %833, 1, !dbg !27 + %837 = bitcast i32 %836 to <2 x bfloat>, !dbg !27 + %838 = extractvalue { i32, i32, i32, i32 } %833, 2, !dbg !27 + %839 = bitcast i32 %838 to <2 x bfloat>, !dbg !27 + %840 = extractvalue { i32, i32, i32, i32 } %833, 3, !dbg !27 + %841 = bitcast i32 %840 to <2 x bfloat>, !dbg !27 + %842 = extractelement <2 x bfloat> %835, i64 0, !dbg !27 + %843 = extractelement <2 x bfloat> %835, i64 1, !dbg !27 + %844 = extractelement <2 x bfloat> %837, i64 0, !dbg !27 + %845 = extractelement <2 x bfloat> %837, i64 1, !dbg !27 + %846 = extractelement <2 x bfloat> %839, i64 0, !dbg !27 + %847 = extractelement <2 x bfloat> %839, i64 1, !dbg !27 + %848 = extractelement <2 x bfloat> %841, i64 0, !dbg !27 + %849 = extractelement <2 x bfloat> %841, i64 1, !dbg !27 + %850 = extractelement <32 x i1> %50, i64 3, !dbg !27 + %851 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %178, i1 %850) #5, !dbg !27 + %852 = extractvalue { i32, i32, i32, i32 } %851, 0, !dbg !27 + %853 = bitcast i32 %852 to <2 x bfloat>, !dbg !27 + %854 = extractvalue { i32, i32, i32, i32 } %851, 1, !dbg !27 + %855 = bitcast i32 %854 to <2 x bfloat>, !dbg !27 + %856 = extractvalue { i32, i32, i32, i32 } %851, 2, !dbg !27 + %857 = bitcast i32 %856 to <2 x bfloat>, !dbg !27 + %858 = extractvalue { i32, i32, i32, i32 } %851, 3, !dbg !27 + %859 = bitcast i32 %858 to <2 x bfloat>, !dbg !27 + %860 = extractelement <2 x bfloat> %853, i64 0, !dbg !27 + %861 = extractelement <2 x bfloat> %853, i64 1, !dbg !27 + %862 = extractelement <2 x bfloat> %855, i64 0, !dbg !27 + %863 = extractelement <2 x bfloat> %855, i64 1, !dbg !27 + %864 = extractelement <2 x bfloat> %857, i64 0, !dbg !27 + %865 = extractelement <2 x bfloat> %857, i64 1, !dbg !27 + %866 = extractelement <2 x bfloat> %859, i64 0, !dbg !27 + %867 = extractelement <2 x bfloat> %859, i64 1, !dbg !27 + %868 = extractelement <32 x i1> %50, i64 2, !dbg !27 + %869 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %179, i1 %868) #5, !dbg !27 + %870 = extractvalue { i32, i32, i32, i32 } %869, 0, !dbg !27 + %871 = bitcast i32 %870 to <2 x bfloat>, !dbg !27 + %872 = extractvalue { i32, i32, i32, i32 } %869, 1, !dbg !27 + %873 = bitcast i32 %872 to <2 x bfloat>, !dbg !27 + %874 = extractvalue { i32, i32, i32, i32 } %869, 2, !dbg !27 + %875 = bitcast i32 %874 to <2 x bfloat>, !dbg !27 + %876 = extractvalue { i32, i32, i32, i32 } %869, 3, !dbg !27 + %877 = bitcast i32 %876 to <2 x bfloat>, !dbg !27 + %878 = extractelement <2 x bfloat> %871, i64 0, !dbg !27 + %879 = extractelement <2 x bfloat> %871, i64 1, !dbg !27 + %880 = extractelement <2 x bfloat> %873, i64 0, !dbg !27 + %881 = extractelement <2 x bfloat> %873, i64 1, !dbg !27 + %882 = extractelement <2 x bfloat> %875, i64 0, !dbg !27 + %883 = extractelement <2 x bfloat> %875, i64 1, !dbg !27 + %884 = extractelement <2 x bfloat> %877, i64 0, !dbg !27 + %885 = extractelement <2 x bfloat> %877, i64 1, !dbg !27 + %886 = fpext bfloat %321 to float, !dbg !28 + %887 = fpext bfloat %322 to float, !dbg !28 + %888 = fpext bfloat %323 to float, !dbg !28 + %889 = fpext bfloat %324 to float, !dbg !28 + %890 = fpext bfloat %325 to float, !dbg !28 + %891 = fpext bfloat %326 to float, !dbg !28 + %892 = fpext bfloat %327 to float, !dbg !28 + %893 = fpext bfloat %328 to float, !dbg !28 + %894 = fpext bfloat %338 to float, !dbg !28 + %895 = fpext bfloat %339 to float, !dbg !28 + %896 = fpext bfloat %340 to float, !dbg !28 + %897 = fpext bfloat %341 to float, !dbg !28 + %898 = fpext bfloat %342 to float, !dbg !28 + %899 = fpext bfloat %343 to float, !dbg !28 + %900 = fpext bfloat %344 to float, !dbg !28 + %901 = fpext bfloat %345 to float, !dbg !28 + %902 = fpext bfloat %356 to float, !dbg !28 + %903 = fpext bfloat %357 to float, !dbg !28 + %904 = fpext bfloat %358 to float, !dbg !28 + %905 = fpext bfloat %359 to float, !dbg !28 + %906 = fpext bfloat %360 to float, !dbg !28 + %907 = fpext bfloat %361 to float, !dbg !28 + %908 = fpext bfloat %362 to float, !dbg !28 + %909 = fpext bfloat %363 to float, !dbg !28 + %910 = fpext bfloat %374 to float, !dbg !28 + %911 = fpext bfloat %375 to float, !dbg !28 + %912 = fpext bfloat %376 to float, !dbg !28 + %913 = fpext bfloat %377 to float, !dbg !28 + %914 = fpext bfloat %378 to float, !dbg !28 + %915 = fpext bfloat %379 to float, !dbg !28 + %916 = fpext bfloat %380 to float, !dbg !28 + %917 = fpext bfloat %381 to float, !dbg !28 + %918 = fpext bfloat %392 to float, !dbg !28 + %919 = fpext bfloat %393 to float, !dbg !28 + %920 = fpext bfloat %394 to float, !dbg !28 + %921 = fpext bfloat %395 to float, !dbg !28 + %922 = fpext bfloat %396 to float, !dbg !28 + %923 = fpext bfloat %397 to float, !dbg !28 + %924 = fpext bfloat %398 to float, !dbg !28 + %925 = fpext bfloat %399 to float, !dbg !28 + %926 = fpext bfloat %410 to float, !dbg !28 + %927 = fpext bfloat %411 to float, !dbg !28 + %928 = fpext bfloat %412 to float, !dbg !28 + %929 = fpext bfloat %413 to float, !dbg !28 + %930 = fpext bfloat %414 to float, !dbg !28 + %931 = fpext bfloat %415 to float, !dbg !28 + %932 = fpext bfloat %416 to float, !dbg !28 + %933 = fpext bfloat %417 to float, !dbg !28 + %934 = fpext bfloat %428 to float, !dbg !28 + %935 = fpext bfloat %429 to float, !dbg !28 + %936 = fpext bfloat %430 to float, !dbg !28 + %937 = fpext bfloat %431 to float, !dbg !28 + %938 = fpext bfloat %432 to float, !dbg !28 + %939 = fpext bfloat %433 to float, !dbg !28 + %940 = fpext bfloat %434 to float, !dbg !28 + %941 = fpext bfloat %435 to float, !dbg !28 + %942 = fpext bfloat %446 to float, !dbg !28 + %943 = fpext bfloat %447 to float, !dbg !28 + %944 = fpext bfloat %448 to float, !dbg !28 + %945 = fpext bfloat %449 to float, !dbg !28 + %946 = fpext bfloat %450 to float, !dbg !28 + %947 = fpext bfloat %451 to float, !dbg !28 + %948 = fpext bfloat %452 to float, !dbg !28 + %949 = fpext bfloat %453 to float, !dbg !28 + %950 = fpext bfloat %464 to float, !dbg !28 + %951 = fpext bfloat %465 to float, !dbg !28 + %952 = fpext bfloat %466 to float, !dbg !28 + %953 = fpext bfloat %467 to float, !dbg !28 + %954 = fpext bfloat %468 to float, !dbg !28 + %955 = fpext bfloat %469 to float, !dbg !28 + %956 = fpext bfloat %470 to float, !dbg !28 + %957 = fpext bfloat %471 to float, !dbg !28 + %958 = fpext bfloat %482 to float, !dbg !28 + %959 = fpext bfloat %483 to float, !dbg !28 + %960 = fpext bfloat %484 to float, !dbg !28 + %961 = fpext bfloat %485 to float, !dbg !28 + %962 = fpext bfloat %486 to float, !dbg !28 + %963 = fpext bfloat %487 to float, !dbg !28 + %964 = fpext bfloat %488 to float, !dbg !28 + %965 = fpext bfloat %489 to float, !dbg !28 + %966 = fpext bfloat %500 to float, !dbg !28 + %967 = fpext bfloat %501 to float, !dbg !28 + %968 = fpext bfloat %502 to float, !dbg !28 + %969 = fpext bfloat %503 to float, !dbg !28 + %970 = fpext bfloat %504 to float, !dbg !28 + %971 = fpext bfloat %505 to float, !dbg !28 + %972 = fpext bfloat %506 to float, !dbg !28 + %973 = fpext bfloat %507 to float, !dbg !28 + %974 = fpext bfloat %518 to float, !dbg !28 + %975 = fpext bfloat %519 to float, !dbg !28 + %976 = fpext bfloat %520 to float, !dbg !28 + %977 = fpext bfloat %521 to float, !dbg !28 + %978 = fpext bfloat %522 to float, !dbg !28 + %979 = fpext bfloat %523 to float, !dbg !28 + %980 = fpext bfloat %524 to float, !dbg !28 + %981 = fpext bfloat %525 to float, !dbg !28 + %982 = fpext bfloat %536 to float, !dbg !28 + %983 = fpext bfloat %537 to float, !dbg !28 + %984 = fpext bfloat %538 to float, !dbg !28 + %985 = fpext bfloat %539 to float, !dbg !28 + %986 = fpext bfloat %540 to float, !dbg !28 + %987 = fpext bfloat %541 to float, !dbg !28 + %988 = fpext bfloat %542 to float, !dbg !28 + %989 = fpext bfloat %543 to float, !dbg !28 + %990 = fpext bfloat %554 to float, !dbg !28 + %991 = fpext bfloat %555 to float, !dbg !28 + %992 = fpext bfloat %556 to float, !dbg !28 + %993 = fpext bfloat %557 to float, !dbg !28 + %994 = fpext bfloat %558 to float, !dbg !28 + %995 = fpext bfloat %559 to float, !dbg !28 + %996 = fpext bfloat %560 to float, !dbg !28 + %997 = fpext bfloat %561 to float, !dbg !28 + %998 = fpext bfloat %572 to float, !dbg !28 + %999 = fpext bfloat %573 to float, !dbg !28 + %1000 = fpext bfloat %574 to float, !dbg !28 + %1001 = fpext bfloat %575 to float, !dbg !28 + %1002 = fpext bfloat %576 to float, !dbg !28 + %1003 = fpext bfloat %577 to float, !dbg !28 + %1004 = fpext bfloat %578 to float, !dbg !28 + %1005 = fpext bfloat %579 to float, !dbg !28 + %1006 = fpext bfloat %590 to float, !dbg !28 + %1007 = fpext bfloat %591 to float, !dbg !28 + %1008 = fpext bfloat %592 to float, !dbg !28 + %1009 = fpext bfloat %593 to float, !dbg !28 + %1010 = fpext bfloat %594 to float, !dbg !28 + %1011 = fpext bfloat %595 to float, !dbg !28 + %1012 = fpext bfloat %596 to float, !dbg !28 + %1013 = fpext bfloat %597 to float, !dbg !28 + %1014 = fpext bfloat %608 to float, !dbg !28 + %1015 = fpext bfloat %609 to float, !dbg !28 + %1016 = fpext bfloat %610 to float, !dbg !28 + %1017 = fpext bfloat %611 to float, !dbg !28 + %1018 = fpext bfloat %612 to float, !dbg !28 + %1019 = fpext bfloat %613 to float, !dbg !28 + %1020 = fpext bfloat %614 to float, !dbg !28 + %1021 = fpext bfloat %615 to float, !dbg !28 + %1022 = fpext bfloat %626 to float, !dbg !28 + %1023 = fpext bfloat %627 to float, !dbg !28 + %1024 = fpext bfloat %628 to float, !dbg !28 + %1025 = fpext bfloat %629 to float, !dbg !28 + %1026 = fpext bfloat %630 to float, !dbg !28 + %1027 = fpext bfloat %631 to float, !dbg !28 + %1028 = fpext bfloat %632 to float, !dbg !28 + %1029 = fpext bfloat %633 to float, !dbg !28 + %1030 = fpext bfloat %644 to float, !dbg !28 + %1031 = fpext bfloat %645 to float, !dbg !28 + %1032 = fpext bfloat %646 to float, !dbg !28 + %1033 = fpext bfloat %647 to float, !dbg !28 + %1034 = fpext bfloat %648 to float, !dbg !28 + %1035 = fpext bfloat %649 to float, !dbg !28 + %1036 = fpext bfloat %650 to float, !dbg !28 + %1037 = fpext bfloat %651 to float, !dbg !28 + %1038 = fpext bfloat %662 to float, !dbg !28 + %1039 = fpext bfloat %663 to float, !dbg !28 + %1040 = fpext bfloat %664 to float, !dbg !28 + %1041 = fpext bfloat %665 to float, !dbg !28 + %1042 = fpext bfloat %666 to float, !dbg !28 + %1043 = fpext bfloat %667 to float, !dbg !28 + %1044 = fpext bfloat %668 to float, !dbg !28 + %1045 = fpext bfloat %669 to float, !dbg !28 + %1046 = fpext bfloat %680 to float, !dbg !28 + %1047 = fpext bfloat %681 to float, !dbg !28 + %1048 = fpext bfloat %682 to float, !dbg !28 + %1049 = fpext bfloat %683 to float, !dbg !28 + %1050 = fpext bfloat %684 to float, !dbg !28 + %1051 = fpext bfloat %685 to float, !dbg !28 + %1052 = fpext bfloat %686 to float, !dbg !28 + %1053 = fpext bfloat %687 to float, !dbg !28 + %1054 = fpext bfloat %698 to float, !dbg !28 + %1055 = fpext bfloat %699 to float, !dbg !28 + %1056 = fpext bfloat %700 to float, !dbg !28 + %1057 = fpext bfloat %701 to float, !dbg !28 + %1058 = fpext bfloat %702 to float, !dbg !28 + %1059 = fpext bfloat %703 to float, !dbg !28 + %1060 = fpext bfloat %704 to float, !dbg !28 + %1061 = fpext bfloat %705 to float, !dbg !28 + %1062 = fpext bfloat %716 to float, !dbg !28 + %1063 = fpext bfloat %717 to float, !dbg !28 + %1064 = fpext bfloat %718 to float, !dbg !28 + %1065 = fpext bfloat %719 to float, !dbg !28 + %1066 = fpext bfloat %720 to float, !dbg !28 + %1067 = fpext bfloat %721 to float, !dbg !28 + %1068 = fpext bfloat %722 to float, !dbg !28 + %1069 = fpext bfloat %723 to float, !dbg !28 + %1070 = fpext bfloat %734 to float, !dbg !28 + %1071 = fpext bfloat %735 to float, !dbg !28 + %1072 = fpext bfloat %736 to float, !dbg !28 + %1073 = fpext bfloat %737 to float, !dbg !28 + %1074 = fpext bfloat %738 to float, !dbg !28 + %1075 = fpext bfloat %739 to float, !dbg !28 + %1076 = fpext bfloat %740 to float, !dbg !28 + %1077 = fpext bfloat %741 to float, !dbg !28 + %1078 = fpext bfloat %752 to float, !dbg !28 + %1079 = fpext bfloat %753 to float, !dbg !28 + %1080 = fpext bfloat %754 to float, !dbg !28 + %1081 = fpext bfloat %755 to float, !dbg !28 + %1082 = fpext bfloat %756 to float, !dbg !28 + %1083 = fpext bfloat %757 to float, !dbg !28 + %1084 = fpext bfloat %758 to float, !dbg !28 + %1085 = fpext bfloat %759 to float, !dbg !28 + %1086 = fpext bfloat %770 to float, !dbg !28 + %1087 = fpext bfloat %771 to float, !dbg !28 + %1088 = fpext bfloat %772 to float, !dbg !28 + %1089 = fpext bfloat %773 to float, !dbg !28 + %1090 = fpext bfloat %774 to float, !dbg !28 + %1091 = fpext bfloat %775 to float, !dbg !28 + %1092 = fpext bfloat %776 to float, !dbg !28 + %1093 = fpext bfloat %777 to float, !dbg !28 + %1094 = fpext bfloat %788 to float, !dbg !28 + %1095 = fpext bfloat %789 to float, !dbg !28 + %1096 = fpext bfloat %790 to float, !dbg !28 + %1097 = fpext bfloat %791 to float, !dbg !28 + %1098 = fpext bfloat %792 to float, !dbg !28 + %1099 = fpext bfloat %793 to float, !dbg !28 + %1100 = fpext bfloat %794 to float, !dbg !28 + %1101 = fpext bfloat %795 to float, !dbg !28 + %1102 = fpext bfloat %806 to float, !dbg !28 + %1103 = fpext bfloat %807 to float, !dbg !28 + %1104 = fpext bfloat %808 to float, !dbg !28 + %1105 = fpext bfloat %809 to float, !dbg !28 + %1106 = fpext bfloat %810 to float, !dbg !28 + %1107 = fpext bfloat %811 to float, !dbg !28 + %1108 = fpext bfloat %812 to float, !dbg !28 + %1109 = fpext bfloat %813 to float, !dbg !28 + %1110 = fpext bfloat %824 to float, !dbg !28 + %1111 = fpext bfloat %825 to float, !dbg !28 + %1112 = fpext bfloat %826 to float, !dbg !28 + %1113 = fpext bfloat %827 to float, !dbg !28 + %1114 = fpext bfloat %828 to float, !dbg !28 + %1115 = fpext bfloat %829 to float, !dbg !28 + %1116 = fpext bfloat %830 to float, !dbg !28 + %1117 = fpext bfloat %831 to float, !dbg !28 + %1118 = fpext bfloat %842 to float, !dbg !28 + %1119 = fpext bfloat %843 to float, !dbg !28 + %1120 = fpext bfloat %844 to float, !dbg !28 + %1121 = fpext bfloat %845 to float, !dbg !28 + %1122 = fpext bfloat %846 to float, !dbg !28 + %1123 = fpext bfloat %847 to float, !dbg !28 + %1124 = fpext bfloat %848 to float, !dbg !28 + %1125 = fpext bfloat %849 to float, !dbg !28 + %1126 = fpext bfloat %860 to float, !dbg !28 + %1127 = fpext bfloat %861 to float, !dbg !28 + %1128 = fpext bfloat %862 to float, !dbg !28 + %1129 = fpext bfloat %863 to float, !dbg !28 + %1130 = fpext bfloat %864 to float, !dbg !28 + %1131 = fpext bfloat %865 to float, !dbg !28 + %1132 = fpext bfloat %866 to float, !dbg !28 + %1133 = fpext bfloat %867 to float, !dbg !28 + %1134 = fpext bfloat %878 to float, !dbg !28 + %1135 = fpext bfloat %879 to float, !dbg !28 + %1136 = fpext bfloat %880 to float, !dbg !28 + %1137 = fpext bfloat %881 to float, !dbg !28 + %1138 = fpext bfloat %882 to float, !dbg !28 + %1139 = fpext bfloat %883 to float, !dbg !28 + %1140 = fpext bfloat %884 to float, !dbg !28 + %1141 = fpext bfloat %885 to float, !dbg !28 + %1142 = extractelement <32 x i1> %50, i64 1, !dbg !29 + %1143 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %181, i1 %1142) #5, !dbg !29 + %1144 = extractelement <32 x i1> %50, i64 0, !dbg !29 + %1145 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %183, i1 %1144) #5, !dbg !29 + %1146 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %216, i1 %21) #5, !dbg !30 + %1147 = extractvalue { i32, i32, i32, i32 } %1146, 0, !dbg !30 + %1148 = bitcast i32 %1147 to <2 x bfloat>, !dbg !30 + %1149 = extractvalue { i32, i32, i32, i32 } %1146, 1, !dbg !30 + %1150 = bitcast i32 %1149 to <2 x bfloat>, !dbg !30 + %1151 = extractvalue { i32, i32, i32, i32 } %1146, 2, !dbg !30 + %1152 = bitcast i32 %1151 to <2 x bfloat>, !dbg !30 + %1153 = extractvalue { i32, i32, i32, i32 } %1146, 3, !dbg !30 + %1154 = bitcast i32 %1153 to <2 x bfloat>, !dbg !30 + %1155 = extractelement <2 x bfloat> %1148, i64 0, !dbg !30 + %1156 = extractelement <2 x bfloat> %1148, i64 1, !dbg !30 + %1157 = extractelement <2 x bfloat> %1150, i64 0, !dbg !30 + %1158 = extractelement <2 x bfloat> %1150, i64 1, !dbg !30 + %1159 = extractelement <2 x bfloat> %1152, i64 0, !dbg !30 + %1160 = extractelement <2 x bfloat> %1152, i64 1, !dbg !30 + %1161 = extractelement <2 x bfloat> %1154, i64 0, !dbg !30 + %1162 = extractelement <2 x bfloat> %1154, i64 1, !dbg !30 + %1163 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %217, i1 %22) #5, !dbg !30 + %1164 = extractvalue { i32, i32, i32, i32 } %1163, 0, !dbg !30 + %1165 = bitcast i32 %1164 to <2 x bfloat>, !dbg !30 + %1166 = extractvalue { i32, i32, i32, i32 } %1163, 1, !dbg !30 + %1167 = bitcast i32 %1166 to <2 x bfloat>, !dbg !30 + %1168 = extractvalue { i32, i32, i32, i32 } %1163, 2, !dbg !30 + %1169 = bitcast i32 %1168 to <2 x bfloat>, !dbg !30 + %1170 = extractvalue { i32, i32, i32, i32 } %1163, 3, !dbg !30 + %1171 = bitcast i32 %1170 to <2 x bfloat>, !dbg !30 + %1172 = extractelement <2 x bfloat> %1165, i64 0, !dbg !30 + %1173 = extractelement <2 x bfloat> %1165, i64 1, !dbg !30 + %1174 = extractelement <2 x bfloat> %1167, i64 0, !dbg !30 + %1175 = extractelement <2 x bfloat> %1167, i64 1, !dbg !30 + %1176 = extractelement <2 x bfloat> %1169, i64 0, !dbg !30 + %1177 = extractelement <2 x bfloat> %1169, i64 1, !dbg !30 + %1178 = extractelement <2 x bfloat> %1171, i64 0, !dbg !30 + %1179 = extractelement <2 x bfloat> %1171, i64 1, !dbg !30 + %1180 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %218, i1 %346) #5, !dbg !30 + %1181 = extractvalue { i32, i32, i32, i32 } %1180, 0, !dbg !30 + %1182 = bitcast i32 %1181 to <2 x bfloat>, !dbg !30 + %1183 = extractvalue { i32, i32, i32, i32 } %1180, 1, !dbg !30 + %1184 = bitcast i32 %1183 to <2 x bfloat>, !dbg !30 + %1185 = extractvalue { i32, i32, i32, i32 } %1180, 2, !dbg !30 + %1186 = bitcast i32 %1185 to <2 x bfloat>, !dbg !30 + %1187 = extractvalue { i32, i32, i32, i32 } %1180, 3, !dbg !30 + %1188 = bitcast i32 %1187 to <2 x bfloat>, !dbg !30 + %1189 = extractelement <2 x bfloat> %1182, i64 0, !dbg !30 + %1190 = extractelement <2 x bfloat> %1182, i64 1, !dbg !30 + %1191 = extractelement <2 x bfloat> %1184, i64 0, !dbg !30 + %1192 = extractelement <2 x bfloat> %1184, i64 1, !dbg !30 + %1193 = extractelement <2 x bfloat> %1186, i64 0, !dbg !30 + %1194 = extractelement <2 x bfloat> %1186, i64 1, !dbg !30 + %1195 = extractelement <2 x bfloat> %1188, i64 0, !dbg !30 + %1196 = extractelement <2 x bfloat> %1188, i64 1, !dbg !30 + %1197 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %219, i1 %364) #5, !dbg !30 + %1198 = extractvalue { i32, i32, i32, i32 } %1197, 0, !dbg !30 + %1199 = bitcast i32 %1198 to <2 x bfloat>, !dbg !30 + %1200 = extractvalue { i32, i32, i32, i32 } %1197, 1, !dbg !30 + %1201 = bitcast i32 %1200 to <2 x bfloat>, !dbg !30 + %1202 = extractvalue { i32, i32, i32, i32 } %1197, 2, !dbg !30 + %1203 = bitcast i32 %1202 to <2 x bfloat>, !dbg !30 + %1204 = extractvalue { i32, i32, i32, i32 } %1197, 3, !dbg !30 + %1205 = bitcast i32 %1204 to <2 x bfloat>, !dbg !30 + %1206 = extractelement <2 x bfloat> %1199, i64 0, !dbg !30 + %1207 = extractelement <2 x bfloat> %1199, i64 1, !dbg !30 + %1208 = extractelement <2 x bfloat> %1201, i64 0, !dbg !30 + %1209 = extractelement <2 x bfloat> %1201, i64 1, !dbg !30 + %1210 = extractelement <2 x bfloat> %1203, i64 0, !dbg !30 + %1211 = extractelement <2 x bfloat> %1203, i64 1, !dbg !30 + %1212 = extractelement <2 x bfloat> %1205, i64 0, !dbg !30 + %1213 = extractelement <2 x bfloat> %1205, i64 1, !dbg !30 + %1214 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %220, i1 %382) #5, !dbg !30 + %1215 = extractvalue { i32, i32, i32, i32 } %1214, 0, !dbg !30 + %1216 = bitcast i32 %1215 to <2 x bfloat>, !dbg !30 + %1217 = extractvalue { i32, i32, i32, i32 } %1214, 1, !dbg !30 + %1218 = bitcast i32 %1217 to <2 x bfloat>, !dbg !30 + %1219 = extractvalue { i32, i32, i32, i32 } %1214, 2, !dbg !30 + %1220 = bitcast i32 %1219 to <2 x bfloat>, !dbg !30 + %1221 = extractvalue { i32, i32, i32, i32 } %1214, 3, !dbg !30 + %1222 = bitcast i32 %1221 to <2 x bfloat>, !dbg !30 + %1223 = extractelement <2 x bfloat> %1216, i64 0, !dbg !30 + %1224 = extractelement <2 x bfloat> %1216, i64 1, !dbg !30 + %1225 = extractelement <2 x bfloat> %1218, i64 0, !dbg !30 + %1226 = extractelement <2 x bfloat> %1218, i64 1, !dbg !30 + %1227 = extractelement <2 x bfloat> %1220, i64 0, !dbg !30 + %1228 = extractelement <2 x bfloat> %1220, i64 1, !dbg !30 + %1229 = extractelement <2 x bfloat> %1222, i64 0, !dbg !30 + %1230 = extractelement <2 x bfloat> %1222, i64 1, !dbg !30 + %1231 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %221, i1 %400) #5, !dbg !30 + %1232 = extractvalue { i32, i32, i32, i32 } %1231, 0, !dbg !30 + %1233 = bitcast i32 %1232 to <2 x bfloat>, !dbg !30 + %1234 = extractvalue { i32, i32, i32, i32 } %1231, 1, !dbg !30 + %1235 = bitcast i32 %1234 to <2 x bfloat>, !dbg !30 + %1236 = extractvalue { i32, i32, i32, i32 } %1231, 2, !dbg !30 + %1237 = bitcast i32 %1236 to <2 x bfloat>, !dbg !30 + %1238 = extractvalue { i32, i32, i32, i32 } %1231, 3, !dbg !30 + %1239 = bitcast i32 %1238 to <2 x bfloat>, !dbg !30 + %1240 = extractelement <2 x bfloat> %1233, i64 0, !dbg !30 + %1241 = extractelement <2 x bfloat> %1233, i64 1, !dbg !30 + %1242 = extractelement <2 x bfloat> %1235, i64 0, !dbg !30 + %1243 = extractelement <2 x bfloat> %1235, i64 1, !dbg !30 + %1244 = extractelement <2 x bfloat> %1237, i64 0, !dbg !30 + %1245 = extractelement <2 x bfloat> %1237, i64 1, !dbg !30 + %1246 = extractelement <2 x bfloat> %1239, i64 0, !dbg !30 + %1247 = extractelement <2 x bfloat> %1239, i64 1, !dbg !30 + %1248 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %222, i1 %418) #5, !dbg !30 + %1249 = extractvalue { i32, i32, i32, i32 } %1248, 0, !dbg !30 + %1250 = bitcast i32 %1249 to <2 x bfloat>, !dbg !30 + %1251 = extractvalue { i32, i32, i32, i32 } %1248, 1, !dbg !30 + %1252 = bitcast i32 %1251 to <2 x bfloat>, !dbg !30 + %1253 = extractvalue { i32, i32, i32, i32 } %1248, 2, !dbg !30 + %1254 = bitcast i32 %1253 to <2 x bfloat>, !dbg !30 + %1255 = extractvalue { i32, i32, i32, i32 } %1248, 3, !dbg !30 + %1256 = bitcast i32 %1255 to <2 x bfloat>, !dbg !30 + %1257 = extractelement <2 x bfloat> %1250, i64 0, !dbg !30 + %1258 = extractelement <2 x bfloat> %1250, i64 1, !dbg !30 + %1259 = extractelement <2 x bfloat> %1252, i64 0, !dbg !30 + %1260 = extractelement <2 x bfloat> %1252, i64 1, !dbg !30 + %1261 = extractelement <2 x bfloat> %1254, i64 0, !dbg !30 + %1262 = extractelement <2 x bfloat> %1254, i64 1, !dbg !30 + %1263 = extractelement <2 x bfloat> %1256, i64 0, !dbg !30 + %1264 = extractelement <2 x bfloat> %1256, i64 1, !dbg !30 + %1265 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %223, i1 %436) #5, !dbg !30 + %1266 = extractvalue { i32, i32, i32, i32 } %1265, 0, !dbg !30 + %1267 = bitcast i32 %1266 to <2 x bfloat>, !dbg !30 + %1268 = extractvalue { i32, i32, i32, i32 } %1265, 1, !dbg !30 + %1269 = bitcast i32 %1268 to <2 x bfloat>, !dbg !30 + %1270 = extractvalue { i32, i32, i32, i32 } %1265, 2, !dbg !30 + %1271 = bitcast i32 %1270 to <2 x bfloat>, !dbg !30 + %1272 = extractvalue { i32, i32, i32, i32 } %1265, 3, !dbg !30 + %1273 = bitcast i32 %1272 to <2 x bfloat>, !dbg !30 + %1274 = extractelement <2 x bfloat> %1267, i64 0, !dbg !30 + %1275 = extractelement <2 x bfloat> %1267, i64 1, !dbg !30 + %1276 = extractelement <2 x bfloat> %1269, i64 0, !dbg !30 + %1277 = extractelement <2 x bfloat> %1269, i64 1, !dbg !30 + %1278 = extractelement <2 x bfloat> %1271, i64 0, !dbg !30 + %1279 = extractelement <2 x bfloat> %1271, i64 1, !dbg !30 + %1280 = extractelement <2 x bfloat> %1273, i64 0, !dbg !30 + %1281 = extractelement <2 x bfloat> %1273, i64 1, !dbg !30 + %1282 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %224, i1 %454) #5, !dbg !30 + %1283 = extractvalue { i32, i32, i32, i32 } %1282, 0, !dbg !30 + %1284 = bitcast i32 %1283 to <2 x bfloat>, !dbg !30 + %1285 = extractvalue { i32, i32, i32, i32 } %1282, 1, !dbg !30 + %1286 = bitcast i32 %1285 to <2 x bfloat>, !dbg !30 + %1287 = extractvalue { i32, i32, i32, i32 } %1282, 2, !dbg !30 + %1288 = bitcast i32 %1287 to <2 x bfloat>, !dbg !30 + %1289 = extractvalue { i32, i32, i32, i32 } %1282, 3, !dbg !30 + %1290 = bitcast i32 %1289 to <2 x bfloat>, !dbg !30 + %1291 = extractelement <2 x bfloat> %1284, i64 0, !dbg !30 + %1292 = extractelement <2 x bfloat> %1284, i64 1, !dbg !30 + %1293 = extractelement <2 x bfloat> %1286, i64 0, !dbg !30 + %1294 = extractelement <2 x bfloat> %1286, i64 1, !dbg !30 + %1295 = extractelement <2 x bfloat> %1288, i64 0, !dbg !30 + %1296 = extractelement <2 x bfloat> %1288, i64 1, !dbg !30 + %1297 = extractelement <2 x bfloat> %1290, i64 0, !dbg !30 + %1298 = extractelement <2 x bfloat> %1290, i64 1, !dbg !30 + %1299 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %225, i1 %472) #5, !dbg !30 + %1300 = extractvalue { i32, i32, i32, i32 } %1299, 0, !dbg !30 + %1301 = bitcast i32 %1300 to <2 x bfloat>, !dbg !30 + %1302 = extractvalue { i32, i32, i32, i32 } %1299, 1, !dbg !30 + %1303 = bitcast i32 %1302 to <2 x bfloat>, !dbg !30 + %1304 = extractvalue { i32, i32, i32, i32 } %1299, 2, !dbg !30 + %1305 = bitcast i32 %1304 to <2 x bfloat>, !dbg !30 + %1306 = extractvalue { i32, i32, i32, i32 } %1299, 3, !dbg !30 + %1307 = bitcast i32 %1306 to <2 x bfloat>, !dbg !30 + %1308 = extractelement <2 x bfloat> %1301, i64 0, !dbg !30 + %1309 = extractelement <2 x bfloat> %1301, i64 1, !dbg !30 + %1310 = extractelement <2 x bfloat> %1303, i64 0, !dbg !30 + %1311 = extractelement <2 x bfloat> %1303, i64 1, !dbg !30 + %1312 = extractelement <2 x bfloat> %1305, i64 0, !dbg !30 + %1313 = extractelement <2 x bfloat> %1305, i64 1, !dbg !30 + %1314 = extractelement <2 x bfloat> %1307, i64 0, !dbg !30 + %1315 = extractelement <2 x bfloat> %1307, i64 1, !dbg !30 + %1316 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %226, i1 %490) #5, !dbg !30 + %1317 = extractvalue { i32, i32, i32, i32 } %1316, 0, !dbg !30 + %1318 = bitcast i32 %1317 to <2 x bfloat>, !dbg !30 + %1319 = extractvalue { i32, i32, i32, i32 } %1316, 1, !dbg !30 + %1320 = bitcast i32 %1319 to <2 x bfloat>, !dbg !30 + %1321 = extractvalue { i32, i32, i32, i32 } %1316, 2, !dbg !30 + %1322 = bitcast i32 %1321 to <2 x bfloat>, !dbg !30 + %1323 = extractvalue { i32, i32, i32, i32 } %1316, 3, !dbg !30 + %1324 = bitcast i32 %1323 to <2 x bfloat>, !dbg !30 + %1325 = extractelement <2 x bfloat> %1318, i64 0, !dbg !30 + %1326 = extractelement <2 x bfloat> %1318, i64 1, !dbg !30 + %1327 = extractelement <2 x bfloat> %1320, i64 0, !dbg !30 + %1328 = extractelement <2 x bfloat> %1320, i64 1, !dbg !30 + %1329 = extractelement <2 x bfloat> %1322, i64 0, !dbg !30 + %1330 = extractelement <2 x bfloat> %1322, i64 1, !dbg !30 + %1331 = extractelement <2 x bfloat> %1324, i64 0, !dbg !30 + %1332 = extractelement <2 x bfloat> %1324, i64 1, !dbg !30 + %1333 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %227, i1 %508) #5, !dbg !30 + %1334 = extractvalue { i32, i32, i32, i32 } %1333, 0, !dbg !30 + %1335 = bitcast i32 %1334 to <2 x bfloat>, !dbg !30 + %1336 = extractvalue { i32, i32, i32, i32 } %1333, 1, !dbg !30 + %1337 = bitcast i32 %1336 to <2 x bfloat>, !dbg !30 + %1338 = extractvalue { i32, i32, i32, i32 } %1333, 2, !dbg !30 + %1339 = bitcast i32 %1338 to <2 x bfloat>, !dbg !30 + %1340 = extractvalue { i32, i32, i32, i32 } %1333, 3, !dbg !30 + %1341 = bitcast i32 %1340 to <2 x bfloat>, !dbg !30 + %1342 = extractelement <2 x bfloat> %1335, i64 0, !dbg !30 + %1343 = extractelement <2 x bfloat> %1335, i64 1, !dbg !30 + %1344 = extractelement <2 x bfloat> %1337, i64 0, !dbg !30 + %1345 = extractelement <2 x bfloat> %1337, i64 1, !dbg !30 + %1346 = extractelement <2 x bfloat> %1339, i64 0, !dbg !30 + %1347 = extractelement <2 x bfloat> %1339, i64 1, !dbg !30 + %1348 = extractelement <2 x bfloat> %1341, i64 0, !dbg !30 + %1349 = extractelement <2 x bfloat> %1341, i64 1, !dbg !30 + %1350 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %228, i1 %526) #5, !dbg !30 + %1351 = extractvalue { i32, i32, i32, i32 } %1350, 0, !dbg !30 + %1352 = bitcast i32 %1351 to <2 x bfloat>, !dbg !30 + %1353 = extractvalue { i32, i32, i32, i32 } %1350, 1, !dbg !30 + %1354 = bitcast i32 %1353 to <2 x bfloat>, !dbg !30 + %1355 = extractvalue { i32, i32, i32, i32 } %1350, 2, !dbg !30 + %1356 = bitcast i32 %1355 to <2 x bfloat>, !dbg !30 + %1357 = extractvalue { i32, i32, i32, i32 } %1350, 3, !dbg !30 + %1358 = bitcast i32 %1357 to <2 x bfloat>, !dbg !30 + %1359 = extractelement <2 x bfloat> %1352, i64 0, !dbg !30 + %1360 = extractelement <2 x bfloat> %1352, i64 1, !dbg !30 + %1361 = extractelement <2 x bfloat> %1354, i64 0, !dbg !30 + %1362 = extractelement <2 x bfloat> %1354, i64 1, !dbg !30 + %1363 = extractelement <2 x bfloat> %1356, i64 0, !dbg !30 + %1364 = extractelement <2 x bfloat> %1356, i64 1, !dbg !30 + %1365 = extractelement <2 x bfloat> %1358, i64 0, !dbg !30 + %1366 = extractelement <2 x bfloat> %1358, i64 1, !dbg !30 + %1367 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %229, i1 %544) #5, !dbg !30 + %1368 = extractvalue { i32, i32, i32, i32 } %1367, 0, !dbg !30 + %1369 = bitcast i32 %1368 to <2 x bfloat>, !dbg !30 + %1370 = extractvalue { i32, i32, i32, i32 } %1367, 1, !dbg !30 + %1371 = bitcast i32 %1370 to <2 x bfloat>, !dbg !30 + %1372 = extractvalue { i32, i32, i32, i32 } %1367, 2, !dbg !30 + %1373 = bitcast i32 %1372 to <2 x bfloat>, !dbg !30 + %1374 = extractvalue { i32, i32, i32, i32 } %1367, 3, !dbg !30 + %1375 = bitcast i32 %1374 to <2 x bfloat>, !dbg !30 + %1376 = extractelement <2 x bfloat> %1369, i64 0, !dbg !30 + %1377 = extractelement <2 x bfloat> %1369, i64 1, !dbg !30 + %1378 = extractelement <2 x bfloat> %1371, i64 0, !dbg !30 + %1379 = extractelement <2 x bfloat> %1371, i64 1, !dbg !30 + %1380 = extractelement <2 x bfloat> %1373, i64 0, !dbg !30 + %1381 = extractelement <2 x bfloat> %1373, i64 1, !dbg !30 + %1382 = extractelement <2 x bfloat> %1375, i64 0, !dbg !30 + %1383 = extractelement <2 x bfloat> %1375, i64 1, !dbg !30 + %1384 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %230, i1 %562) #5, !dbg !30 + %1385 = extractvalue { i32, i32, i32, i32 } %1384, 0, !dbg !30 + %1386 = bitcast i32 %1385 to <2 x bfloat>, !dbg !30 + %1387 = extractvalue { i32, i32, i32, i32 } %1384, 1, !dbg !30 + %1388 = bitcast i32 %1387 to <2 x bfloat>, !dbg !30 + %1389 = extractvalue { i32, i32, i32, i32 } %1384, 2, !dbg !30 + %1390 = bitcast i32 %1389 to <2 x bfloat>, !dbg !30 + %1391 = extractvalue { i32, i32, i32, i32 } %1384, 3, !dbg !30 + %1392 = bitcast i32 %1391 to <2 x bfloat>, !dbg !30 + %1393 = extractelement <2 x bfloat> %1386, i64 0, !dbg !30 + %1394 = extractelement <2 x bfloat> %1386, i64 1, !dbg !30 + %1395 = extractelement <2 x bfloat> %1388, i64 0, !dbg !30 + %1396 = extractelement <2 x bfloat> %1388, i64 1, !dbg !30 + %1397 = extractelement <2 x bfloat> %1390, i64 0, !dbg !30 + %1398 = extractelement <2 x bfloat> %1390, i64 1, !dbg !30 + %1399 = extractelement <2 x bfloat> %1392, i64 0, !dbg !30 + %1400 = extractelement <2 x bfloat> %1392, i64 1, !dbg !30 + %1401 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %231, i1 %580) #5, !dbg !30 + %1402 = extractvalue { i32, i32, i32, i32 } %1401, 0, !dbg !30 + %1403 = bitcast i32 %1402 to <2 x bfloat>, !dbg !30 + %1404 = extractvalue { i32, i32, i32, i32 } %1401, 1, !dbg !30 + %1405 = bitcast i32 %1404 to <2 x bfloat>, !dbg !30 + %1406 = extractvalue { i32, i32, i32, i32 } %1401, 2, !dbg !30 + %1407 = bitcast i32 %1406 to <2 x bfloat>, !dbg !30 + %1408 = extractvalue { i32, i32, i32, i32 } %1401, 3, !dbg !30 + %1409 = bitcast i32 %1408 to <2 x bfloat>, !dbg !30 + %1410 = extractelement <2 x bfloat> %1403, i64 0, !dbg !30 + %1411 = extractelement <2 x bfloat> %1403, i64 1, !dbg !30 + %1412 = extractelement <2 x bfloat> %1405, i64 0, !dbg !30 + %1413 = extractelement <2 x bfloat> %1405, i64 1, !dbg !30 + %1414 = extractelement <2 x bfloat> %1407, i64 0, !dbg !30 + %1415 = extractelement <2 x bfloat> %1407, i64 1, !dbg !30 + %1416 = extractelement <2 x bfloat> %1409, i64 0, !dbg !30 + %1417 = extractelement <2 x bfloat> %1409, i64 1, !dbg !30 + %1418 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %232, i1 %598) #5, !dbg !30 + %1419 = extractvalue { i32, i32, i32, i32 } %1418, 0, !dbg !30 + %1420 = bitcast i32 %1419 to <2 x bfloat>, !dbg !30 + %1421 = extractvalue { i32, i32, i32, i32 } %1418, 1, !dbg !30 + %1422 = bitcast i32 %1421 to <2 x bfloat>, !dbg !30 + %1423 = extractvalue { i32, i32, i32, i32 } %1418, 2, !dbg !30 + %1424 = bitcast i32 %1423 to <2 x bfloat>, !dbg !30 + %1425 = extractvalue { i32, i32, i32, i32 } %1418, 3, !dbg !30 + %1426 = bitcast i32 %1425 to <2 x bfloat>, !dbg !30 + %1427 = extractelement <2 x bfloat> %1420, i64 0, !dbg !30 + %1428 = extractelement <2 x bfloat> %1420, i64 1, !dbg !30 + %1429 = extractelement <2 x bfloat> %1422, i64 0, !dbg !30 + %1430 = extractelement <2 x bfloat> %1422, i64 1, !dbg !30 + %1431 = extractelement <2 x bfloat> %1424, i64 0, !dbg !30 + %1432 = extractelement <2 x bfloat> %1424, i64 1, !dbg !30 + %1433 = extractelement <2 x bfloat> %1426, i64 0, !dbg !30 + %1434 = extractelement <2 x bfloat> %1426, i64 1, !dbg !30 + %1435 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %233, i1 %616) #5, !dbg !30 + %1436 = extractvalue { i32, i32, i32, i32 } %1435, 0, !dbg !30 + %1437 = bitcast i32 %1436 to <2 x bfloat>, !dbg !30 + %1438 = extractvalue { i32, i32, i32, i32 } %1435, 1, !dbg !30 + %1439 = bitcast i32 %1438 to <2 x bfloat>, !dbg !30 + %1440 = extractvalue { i32, i32, i32, i32 } %1435, 2, !dbg !30 + %1441 = bitcast i32 %1440 to <2 x bfloat>, !dbg !30 + %1442 = extractvalue { i32, i32, i32, i32 } %1435, 3, !dbg !30 + %1443 = bitcast i32 %1442 to <2 x bfloat>, !dbg !30 + %1444 = extractelement <2 x bfloat> %1437, i64 0, !dbg !30 + %1445 = extractelement <2 x bfloat> %1437, i64 1, !dbg !30 + %1446 = extractelement <2 x bfloat> %1439, i64 0, !dbg !30 + %1447 = extractelement <2 x bfloat> %1439, i64 1, !dbg !30 + %1448 = extractelement <2 x bfloat> %1441, i64 0, !dbg !30 + %1449 = extractelement <2 x bfloat> %1441, i64 1, !dbg !30 + %1450 = extractelement <2 x bfloat> %1443, i64 0, !dbg !30 + %1451 = extractelement <2 x bfloat> %1443, i64 1, !dbg !30 + %1452 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %234, i1 %634) #5, !dbg !30 + %1453 = extractvalue { i32, i32, i32, i32 } %1452, 0, !dbg !30 + %1454 = bitcast i32 %1453 to <2 x bfloat>, !dbg !30 + %1455 = extractvalue { i32, i32, i32, i32 } %1452, 1, !dbg !30 + %1456 = bitcast i32 %1455 to <2 x bfloat>, !dbg !30 + %1457 = extractvalue { i32, i32, i32, i32 } %1452, 2, !dbg !30 + %1458 = bitcast i32 %1457 to <2 x bfloat>, !dbg !30 + %1459 = extractvalue { i32, i32, i32, i32 } %1452, 3, !dbg !30 + %1460 = bitcast i32 %1459 to <2 x bfloat>, !dbg !30 + %1461 = extractelement <2 x bfloat> %1454, i64 0, !dbg !30 + %1462 = extractelement <2 x bfloat> %1454, i64 1, !dbg !30 + %1463 = extractelement <2 x bfloat> %1456, i64 0, !dbg !30 + %1464 = extractelement <2 x bfloat> %1456, i64 1, !dbg !30 + %1465 = extractelement <2 x bfloat> %1458, i64 0, !dbg !30 + %1466 = extractelement <2 x bfloat> %1458, i64 1, !dbg !30 + %1467 = extractelement <2 x bfloat> %1460, i64 0, !dbg !30 + %1468 = extractelement <2 x bfloat> %1460, i64 1, !dbg !30 + %1469 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %235, i1 %652) #5, !dbg !30 + %1470 = extractvalue { i32, i32, i32, i32 } %1469, 0, !dbg !30 + %1471 = bitcast i32 %1470 to <2 x bfloat>, !dbg !30 + %1472 = extractvalue { i32, i32, i32, i32 } %1469, 1, !dbg !30 + %1473 = bitcast i32 %1472 to <2 x bfloat>, !dbg !30 + %1474 = extractvalue { i32, i32, i32, i32 } %1469, 2, !dbg !30 + %1475 = bitcast i32 %1474 to <2 x bfloat>, !dbg !30 + %1476 = extractvalue { i32, i32, i32, i32 } %1469, 3, !dbg !30 + %1477 = bitcast i32 %1476 to <2 x bfloat>, !dbg !30 + %1478 = extractelement <2 x bfloat> %1471, i64 0, !dbg !30 + %1479 = extractelement <2 x bfloat> %1471, i64 1, !dbg !30 + %1480 = extractelement <2 x bfloat> %1473, i64 0, !dbg !30 + %1481 = extractelement <2 x bfloat> %1473, i64 1, !dbg !30 + %1482 = extractelement <2 x bfloat> %1475, i64 0, !dbg !30 + %1483 = extractelement <2 x bfloat> %1475, i64 1, !dbg !30 + %1484 = extractelement <2 x bfloat> %1477, i64 0, !dbg !30 + %1485 = extractelement <2 x bfloat> %1477, i64 1, !dbg !30 + %1486 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %236, i1 %670) #5, !dbg !30 + %1487 = extractvalue { i32, i32, i32, i32 } %1486, 0, !dbg !30 + %1488 = bitcast i32 %1487 to <2 x bfloat>, !dbg !30 + %1489 = extractvalue { i32, i32, i32, i32 } %1486, 1, !dbg !30 + %1490 = bitcast i32 %1489 to <2 x bfloat>, !dbg !30 + %1491 = extractvalue { i32, i32, i32, i32 } %1486, 2, !dbg !30 + %1492 = bitcast i32 %1491 to <2 x bfloat>, !dbg !30 + %1493 = extractvalue { i32, i32, i32, i32 } %1486, 3, !dbg !30 + %1494 = bitcast i32 %1493 to <2 x bfloat>, !dbg !30 + %1495 = extractelement <2 x bfloat> %1488, i64 0, !dbg !30 + %1496 = extractelement <2 x bfloat> %1488, i64 1, !dbg !30 + %1497 = extractelement <2 x bfloat> %1490, i64 0, !dbg !30 + %1498 = extractelement <2 x bfloat> %1490, i64 1, !dbg !30 + %1499 = extractelement <2 x bfloat> %1492, i64 0, !dbg !30 + %1500 = extractelement <2 x bfloat> %1492, i64 1, !dbg !30 + %1501 = extractelement <2 x bfloat> %1494, i64 0, !dbg !30 + %1502 = extractelement <2 x bfloat> %1494, i64 1, !dbg !30 + %1503 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %237, i1 %688) #5, !dbg !30 + %1504 = extractvalue { i32, i32, i32, i32 } %1503, 0, !dbg !30 + %1505 = bitcast i32 %1504 to <2 x bfloat>, !dbg !30 + %1506 = extractvalue { i32, i32, i32, i32 } %1503, 1, !dbg !30 + %1507 = bitcast i32 %1506 to <2 x bfloat>, !dbg !30 + %1508 = extractvalue { i32, i32, i32, i32 } %1503, 2, !dbg !30 + %1509 = bitcast i32 %1508 to <2 x bfloat>, !dbg !30 + %1510 = extractvalue { i32, i32, i32, i32 } %1503, 3, !dbg !30 + %1511 = bitcast i32 %1510 to <2 x bfloat>, !dbg !30 + %1512 = extractelement <2 x bfloat> %1505, i64 0, !dbg !30 + %1513 = extractelement <2 x bfloat> %1505, i64 1, !dbg !30 + %1514 = extractelement <2 x bfloat> %1507, i64 0, !dbg !30 + %1515 = extractelement <2 x bfloat> %1507, i64 1, !dbg !30 + %1516 = extractelement <2 x bfloat> %1509, i64 0, !dbg !30 + %1517 = extractelement <2 x bfloat> %1509, i64 1, !dbg !30 + %1518 = extractelement <2 x bfloat> %1511, i64 0, !dbg !30 + %1519 = extractelement <2 x bfloat> %1511, i64 1, !dbg !30 + %1520 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %238, i1 %706) #5, !dbg !30 + %1521 = extractvalue { i32, i32, i32, i32 } %1520, 0, !dbg !30 + %1522 = bitcast i32 %1521 to <2 x bfloat>, !dbg !30 + %1523 = extractvalue { i32, i32, i32, i32 } %1520, 1, !dbg !30 + %1524 = bitcast i32 %1523 to <2 x bfloat>, !dbg !30 + %1525 = extractvalue { i32, i32, i32, i32 } %1520, 2, !dbg !30 + %1526 = bitcast i32 %1525 to <2 x bfloat>, !dbg !30 + %1527 = extractvalue { i32, i32, i32, i32 } %1520, 3, !dbg !30 + %1528 = bitcast i32 %1527 to <2 x bfloat>, !dbg !30 + %1529 = extractelement <2 x bfloat> %1522, i64 0, !dbg !30 + %1530 = extractelement <2 x bfloat> %1522, i64 1, !dbg !30 + %1531 = extractelement <2 x bfloat> %1524, i64 0, !dbg !30 + %1532 = extractelement <2 x bfloat> %1524, i64 1, !dbg !30 + %1533 = extractelement <2 x bfloat> %1526, i64 0, !dbg !30 + %1534 = extractelement <2 x bfloat> %1526, i64 1, !dbg !30 + %1535 = extractelement <2 x bfloat> %1528, i64 0, !dbg !30 + %1536 = extractelement <2 x bfloat> %1528, i64 1, !dbg !30 + %1537 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %239, i1 %724) #5, !dbg !30 + %1538 = extractvalue { i32, i32, i32, i32 } %1537, 0, !dbg !30 + %1539 = bitcast i32 %1538 to <2 x bfloat>, !dbg !30 + %1540 = extractvalue { i32, i32, i32, i32 } %1537, 1, !dbg !30 + %1541 = bitcast i32 %1540 to <2 x bfloat>, !dbg !30 + %1542 = extractvalue { i32, i32, i32, i32 } %1537, 2, !dbg !30 + %1543 = bitcast i32 %1542 to <2 x bfloat>, !dbg !30 + %1544 = extractvalue { i32, i32, i32, i32 } %1537, 3, !dbg !30 + %1545 = bitcast i32 %1544 to <2 x bfloat>, !dbg !30 + %1546 = extractelement <2 x bfloat> %1539, i64 0, !dbg !30 + %1547 = extractelement <2 x bfloat> %1539, i64 1, !dbg !30 + %1548 = extractelement <2 x bfloat> %1541, i64 0, !dbg !30 + %1549 = extractelement <2 x bfloat> %1541, i64 1, !dbg !30 + %1550 = extractelement <2 x bfloat> %1543, i64 0, !dbg !30 + %1551 = extractelement <2 x bfloat> %1543, i64 1, !dbg !30 + %1552 = extractelement <2 x bfloat> %1545, i64 0, !dbg !30 + %1553 = extractelement <2 x bfloat> %1545, i64 1, !dbg !30 + %1554 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %240, i1 %742) #5, !dbg !30 + %1555 = extractvalue { i32, i32, i32, i32 } %1554, 0, !dbg !30 + %1556 = bitcast i32 %1555 to <2 x bfloat>, !dbg !30 + %1557 = extractvalue { i32, i32, i32, i32 } %1554, 1, !dbg !30 + %1558 = bitcast i32 %1557 to <2 x bfloat>, !dbg !30 + %1559 = extractvalue { i32, i32, i32, i32 } %1554, 2, !dbg !30 + %1560 = bitcast i32 %1559 to <2 x bfloat>, !dbg !30 + %1561 = extractvalue { i32, i32, i32, i32 } %1554, 3, !dbg !30 + %1562 = bitcast i32 %1561 to <2 x bfloat>, !dbg !30 + %1563 = extractelement <2 x bfloat> %1556, i64 0, !dbg !30 + %1564 = extractelement <2 x bfloat> %1556, i64 1, !dbg !30 + %1565 = extractelement <2 x bfloat> %1558, i64 0, !dbg !30 + %1566 = extractelement <2 x bfloat> %1558, i64 1, !dbg !30 + %1567 = extractelement <2 x bfloat> %1560, i64 0, !dbg !30 + %1568 = extractelement <2 x bfloat> %1560, i64 1, !dbg !30 + %1569 = extractelement <2 x bfloat> %1562, i64 0, !dbg !30 + %1570 = extractelement <2 x bfloat> %1562, i64 1, !dbg !30 + %1571 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %241, i1 %760) #5, !dbg !30 + %1572 = extractvalue { i32, i32, i32, i32 } %1571, 0, !dbg !30 + %1573 = bitcast i32 %1572 to <2 x bfloat>, !dbg !30 + %1574 = extractvalue { i32, i32, i32, i32 } %1571, 1, !dbg !30 + %1575 = bitcast i32 %1574 to <2 x bfloat>, !dbg !30 + %1576 = extractvalue { i32, i32, i32, i32 } %1571, 2, !dbg !30 + %1577 = bitcast i32 %1576 to <2 x bfloat>, !dbg !30 + %1578 = extractvalue { i32, i32, i32, i32 } %1571, 3, !dbg !30 + %1579 = bitcast i32 %1578 to <2 x bfloat>, !dbg !30 + %1580 = extractelement <2 x bfloat> %1573, i64 0, !dbg !30 + %1581 = extractelement <2 x bfloat> %1573, i64 1, !dbg !30 + %1582 = extractelement <2 x bfloat> %1575, i64 0, !dbg !30 + %1583 = extractelement <2 x bfloat> %1575, i64 1, !dbg !30 + %1584 = extractelement <2 x bfloat> %1577, i64 0, !dbg !30 + %1585 = extractelement <2 x bfloat> %1577, i64 1, !dbg !30 + %1586 = extractelement <2 x bfloat> %1579, i64 0, !dbg !30 + %1587 = extractelement <2 x bfloat> %1579, i64 1, !dbg !30 + %1588 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %242, i1 %778) #5, !dbg !30 + %1589 = extractvalue { i32, i32, i32, i32 } %1588, 0, !dbg !30 + %1590 = bitcast i32 %1589 to <2 x bfloat>, !dbg !30 + %1591 = extractvalue { i32, i32, i32, i32 } %1588, 1, !dbg !30 + %1592 = bitcast i32 %1591 to <2 x bfloat>, !dbg !30 + %1593 = extractvalue { i32, i32, i32, i32 } %1588, 2, !dbg !30 + %1594 = bitcast i32 %1593 to <2 x bfloat>, !dbg !30 + %1595 = extractvalue { i32, i32, i32, i32 } %1588, 3, !dbg !30 + %1596 = bitcast i32 %1595 to <2 x bfloat>, !dbg !30 + %1597 = extractelement <2 x bfloat> %1590, i64 0, !dbg !30 + %1598 = extractelement <2 x bfloat> %1590, i64 1, !dbg !30 + %1599 = extractelement <2 x bfloat> %1592, i64 0, !dbg !30 + %1600 = extractelement <2 x bfloat> %1592, i64 1, !dbg !30 + %1601 = extractelement <2 x bfloat> %1594, i64 0, !dbg !30 + %1602 = extractelement <2 x bfloat> %1594, i64 1, !dbg !30 + %1603 = extractelement <2 x bfloat> %1596, i64 0, !dbg !30 + %1604 = extractelement <2 x bfloat> %1596, i64 1, !dbg !30 + %1605 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %243, i1 %796) #5, !dbg !30 + %1606 = extractvalue { i32, i32, i32, i32 } %1605, 0, !dbg !30 + %1607 = bitcast i32 %1606 to <2 x bfloat>, !dbg !30 + %1608 = extractvalue { i32, i32, i32, i32 } %1605, 1, !dbg !30 + %1609 = bitcast i32 %1608 to <2 x bfloat>, !dbg !30 + %1610 = extractvalue { i32, i32, i32, i32 } %1605, 2, !dbg !30 + %1611 = bitcast i32 %1610 to <2 x bfloat>, !dbg !30 + %1612 = extractvalue { i32, i32, i32, i32 } %1605, 3, !dbg !30 + %1613 = bitcast i32 %1612 to <2 x bfloat>, !dbg !30 + %1614 = extractelement <2 x bfloat> %1607, i64 0, !dbg !30 + %1615 = extractelement <2 x bfloat> %1607, i64 1, !dbg !30 + %1616 = extractelement <2 x bfloat> %1609, i64 0, !dbg !30 + %1617 = extractelement <2 x bfloat> %1609, i64 1, !dbg !30 + %1618 = extractelement <2 x bfloat> %1611, i64 0, !dbg !30 + %1619 = extractelement <2 x bfloat> %1611, i64 1, !dbg !30 + %1620 = extractelement <2 x bfloat> %1613, i64 0, !dbg !30 + %1621 = extractelement <2 x bfloat> %1613, i64 1, !dbg !30 + %1622 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %244, i1 %814) #5, !dbg !30 + %1623 = extractvalue { i32, i32, i32, i32 } %1622, 0, !dbg !30 + %1624 = bitcast i32 %1623 to <2 x bfloat>, !dbg !30 + %1625 = extractvalue { i32, i32, i32, i32 } %1622, 1, !dbg !30 + %1626 = bitcast i32 %1625 to <2 x bfloat>, !dbg !30 + %1627 = extractvalue { i32, i32, i32, i32 } %1622, 2, !dbg !30 + %1628 = bitcast i32 %1627 to <2 x bfloat>, !dbg !30 + %1629 = extractvalue { i32, i32, i32, i32 } %1622, 3, !dbg !30 + %1630 = bitcast i32 %1629 to <2 x bfloat>, !dbg !30 + %1631 = extractelement <2 x bfloat> %1624, i64 0, !dbg !30 + %1632 = extractelement <2 x bfloat> %1624, i64 1, !dbg !30 + %1633 = extractelement <2 x bfloat> %1626, i64 0, !dbg !30 + %1634 = extractelement <2 x bfloat> %1626, i64 1, !dbg !30 + %1635 = extractelement <2 x bfloat> %1628, i64 0, !dbg !30 + %1636 = extractelement <2 x bfloat> %1628, i64 1, !dbg !30 + %1637 = extractelement <2 x bfloat> %1630, i64 0, !dbg !30 + %1638 = extractelement <2 x bfloat> %1630, i64 1, !dbg !30 + %1639 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %245, i1 %832) #5, !dbg !30 + %1640 = extractvalue { i32, i32, i32, i32 } %1639, 0, !dbg !30 + %1641 = bitcast i32 %1640 to <2 x bfloat>, !dbg !30 + %1642 = extractvalue { i32, i32, i32, i32 } %1639, 1, !dbg !30 + %1643 = bitcast i32 %1642 to <2 x bfloat>, !dbg !30 + %1644 = extractvalue { i32, i32, i32, i32 } %1639, 2, !dbg !30 + %1645 = bitcast i32 %1644 to <2 x bfloat>, !dbg !30 + %1646 = extractvalue { i32, i32, i32, i32 } %1639, 3, !dbg !30 + %1647 = bitcast i32 %1646 to <2 x bfloat>, !dbg !30 + %1648 = extractelement <2 x bfloat> %1641, i64 0, !dbg !30 + %1649 = extractelement <2 x bfloat> %1641, i64 1, !dbg !30 + %1650 = extractelement <2 x bfloat> %1643, i64 0, !dbg !30 + %1651 = extractelement <2 x bfloat> %1643, i64 1, !dbg !30 + %1652 = extractelement <2 x bfloat> %1645, i64 0, !dbg !30 + %1653 = extractelement <2 x bfloat> %1645, i64 1, !dbg !30 + %1654 = extractelement <2 x bfloat> %1647, i64 0, !dbg !30 + %1655 = extractelement <2 x bfloat> %1647, i64 1, !dbg !30 + %1656 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %246, i1 %850) #5, !dbg !30 + %1657 = extractvalue { i32, i32, i32, i32 } %1656, 0, !dbg !30 + %1658 = bitcast i32 %1657 to <2 x bfloat>, !dbg !30 + %1659 = extractvalue { i32, i32, i32, i32 } %1656, 1, !dbg !30 + %1660 = bitcast i32 %1659 to <2 x bfloat>, !dbg !30 + %1661 = extractvalue { i32, i32, i32, i32 } %1656, 2, !dbg !30 + %1662 = bitcast i32 %1661 to <2 x bfloat>, !dbg !30 + %1663 = extractvalue { i32, i32, i32, i32 } %1656, 3, !dbg !30 + %1664 = bitcast i32 %1663 to <2 x bfloat>, !dbg !30 + %1665 = extractelement <2 x bfloat> %1658, i64 0, !dbg !30 + %1666 = extractelement <2 x bfloat> %1658, i64 1, !dbg !30 + %1667 = extractelement <2 x bfloat> %1660, i64 0, !dbg !30 + %1668 = extractelement <2 x bfloat> %1660, i64 1, !dbg !30 + %1669 = extractelement <2 x bfloat> %1662, i64 0, !dbg !30 + %1670 = extractelement <2 x bfloat> %1662, i64 1, !dbg !30 + %1671 = extractelement <2 x bfloat> %1664, i64 0, !dbg !30 + %1672 = extractelement <2 x bfloat> %1664, i64 1, !dbg !30 + %1673 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %247, i1 %868) #5, !dbg !30 + %1674 = extractvalue { i32, i32, i32, i32 } %1673, 0, !dbg !30 + %1675 = bitcast i32 %1674 to <2 x bfloat>, !dbg !30 + %1676 = extractvalue { i32, i32, i32, i32 } %1673, 1, !dbg !30 + %1677 = bitcast i32 %1676 to <2 x bfloat>, !dbg !30 + %1678 = extractvalue { i32, i32, i32, i32 } %1673, 2, !dbg !30 + %1679 = bitcast i32 %1678 to <2 x bfloat>, !dbg !30 + %1680 = extractvalue { i32, i32, i32, i32 } %1673, 3, !dbg !30 + %1681 = bitcast i32 %1680 to <2 x bfloat>, !dbg !30 + %1682 = extractelement <2 x bfloat> %1675, i64 0, !dbg !30 + %1683 = extractelement <2 x bfloat> %1675, i64 1, !dbg !30 + %1684 = extractelement <2 x bfloat> %1677, i64 0, !dbg !30 + %1685 = extractelement <2 x bfloat> %1677, i64 1, !dbg !30 + %1686 = extractelement <2 x bfloat> %1679, i64 0, !dbg !30 + %1687 = extractelement <2 x bfloat> %1679, i64 1, !dbg !30 + %1688 = extractelement <2 x bfloat> %1681, i64 0, !dbg !30 + %1689 = extractelement <2 x bfloat> %1681, i64 1, !dbg !30 + %1690 = fpext bfloat %1155 to float, !dbg !31 + %1691 = fpext bfloat %1156 to float, !dbg !31 + %1692 = fpext bfloat %1157 to float, !dbg !31 + %1693 = fpext bfloat %1158 to float, !dbg !31 + %1694 = fpext bfloat %1159 to float, !dbg !31 + %1695 = fpext bfloat %1160 to float, !dbg !31 + %1696 = fpext bfloat %1161 to float, !dbg !31 + %1697 = fpext bfloat %1162 to float, !dbg !31 + %1698 = fpext bfloat %1172 to float, !dbg !31 + %1699 = fpext bfloat %1173 to float, !dbg !31 + %1700 = fpext bfloat %1174 to float, !dbg !31 + %1701 = fpext bfloat %1175 to float, !dbg !31 + %1702 = fpext bfloat %1176 to float, !dbg !31 + %1703 = fpext bfloat %1177 to float, !dbg !31 + %1704 = fpext bfloat %1178 to float, !dbg !31 + %1705 = fpext bfloat %1179 to float, !dbg !31 + %1706 = fpext bfloat %1189 to float, !dbg !31 + %1707 = fpext bfloat %1190 to float, !dbg !31 + %1708 = fpext bfloat %1191 to float, !dbg !31 + %1709 = fpext bfloat %1192 to float, !dbg !31 + %1710 = fpext bfloat %1193 to float, !dbg !31 + %1711 = fpext bfloat %1194 to float, !dbg !31 + %1712 = fpext bfloat %1195 to float, !dbg !31 + %1713 = fpext bfloat %1196 to float, !dbg !31 + %1714 = fpext bfloat %1206 to float, !dbg !31 + %1715 = fpext bfloat %1207 to float, !dbg !31 + %1716 = fpext bfloat %1208 to float, !dbg !31 + %1717 = fpext bfloat %1209 to float, !dbg !31 + %1718 = fpext bfloat %1210 to float, !dbg !31 + %1719 = fpext bfloat %1211 to float, !dbg !31 + %1720 = fpext bfloat %1212 to float, !dbg !31 + %1721 = fpext bfloat %1213 to float, !dbg !31 + %1722 = fpext bfloat %1223 to float, !dbg !31 + %1723 = fpext bfloat %1224 to float, !dbg !31 + %1724 = fpext bfloat %1225 to float, !dbg !31 + %1725 = fpext bfloat %1226 to float, !dbg !31 + %1726 = fpext bfloat %1227 to float, !dbg !31 + %1727 = fpext bfloat %1228 to float, !dbg !31 + %1728 = fpext bfloat %1229 to float, !dbg !31 + %1729 = fpext bfloat %1230 to float, !dbg !31 + %1730 = fpext bfloat %1240 to float, !dbg !31 + %1731 = fpext bfloat %1241 to float, !dbg !31 + %1732 = fpext bfloat %1242 to float, !dbg !31 + %1733 = fpext bfloat %1243 to float, !dbg !31 + %1734 = fpext bfloat %1244 to float, !dbg !31 + %1735 = fpext bfloat %1245 to float, !dbg !31 + %1736 = fpext bfloat %1246 to float, !dbg !31 + %1737 = fpext bfloat %1247 to float, !dbg !31 + %1738 = fpext bfloat %1257 to float, !dbg !31 + %1739 = fpext bfloat %1258 to float, !dbg !31 + %1740 = fpext bfloat %1259 to float, !dbg !31 + %1741 = fpext bfloat %1260 to float, !dbg !31 + %1742 = fpext bfloat %1261 to float, !dbg !31 + %1743 = fpext bfloat %1262 to float, !dbg !31 + %1744 = fpext bfloat %1263 to float, !dbg !31 + %1745 = fpext bfloat %1264 to float, !dbg !31 + %1746 = fpext bfloat %1274 to float, !dbg !31 + %1747 = fpext bfloat %1275 to float, !dbg !31 + %1748 = fpext bfloat %1276 to float, !dbg !31 + %1749 = fpext bfloat %1277 to float, !dbg !31 + %1750 = fpext bfloat %1278 to float, !dbg !31 + %1751 = fpext bfloat %1279 to float, !dbg !31 + %1752 = fpext bfloat %1280 to float, !dbg !31 + %1753 = fpext bfloat %1281 to float, !dbg !31 + %1754 = fpext bfloat %1291 to float, !dbg !31 + %1755 = fpext bfloat %1292 to float, !dbg !31 + %1756 = fpext bfloat %1293 to float, !dbg !31 + %1757 = fpext bfloat %1294 to float, !dbg !31 + %1758 = fpext bfloat %1295 to float, !dbg !31 + %1759 = fpext bfloat %1296 to float, !dbg !31 + %1760 = fpext bfloat %1297 to float, !dbg !31 + %1761 = fpext bfloat %1298 to float, !dbg !31 + %1762 = fpext bfloat %1308 to float, !dbg !31 + %1763 = fpext bfloat %1309 to float, !dbg !31 + %1764 = fpext bfloat %1310 to float, !dbg !31 + %1765 = fpext bfloat %1311 to float, !dbg !31 + %1766 = fpext bfloat %1312 to float, !dbg !31 + %1767 = fpext bfloat %1313 to float, !dbg !31 + %1768 = fpext bfloat %1314 to float, !dbg !31 + %1769 = fpext bfloat %1315 to float, !dbg !31 + %1770 = fpext bfloat %1325 to float, !dbg !31 + %1771 = fpext bfloat %1326 to float, !dbg !31 + %1772 = fpext bfloat %1327 to float, !dbg !31 + %1773 = fpext bfloat %1328 to float, !dbg !31 + %1774 = fpext bfloat %1329 to float, !dbg !31 + %1775 = fpext bfloat %1330 to float, !dbg !31 + %1776 = fpext bfloat %1331 to float, !dbg !31 + %1777 = fpext bfloat %1332 to float, !dbg !31 + %1778 = fpext bfloat %1342 to float, !dbg !31 + %1779 = fpext bfloat %1343 to float, !dbg !31 + %1780 = fpext bfloat %1344 to float, !dbg !31 + %1781 = fpext bfloat %1345 to float, !dbg !31 + %1782 = fpext bfloat %1346 to float, !dbg !31 + %1783 = fpext bfloat %1347 to float, !dbg !31 + %1784 = fpext bfloat %1348 to float, !dbg !31 + %1785 = fpext bfloat %1349 to float, !dbg !31 + %1786 = fpext bfloat %1359 to float, !dbg !31 + %1787 = fpext bfloat %1360 to float, !dbg !31 + %1788 = fpext bfloat %1361 to float, !dbg !31 + %1789 = fpext bfloat %1362 to float, !dbg !31 + %1790 = fpext bfloat %1363 to float, !dbg !31 + %1791 = fpext bfloat %1364 to float, !dbg !31 + %1792 = fpext bfloat %1365 to float, !dbg !31 + %1793 = fpext bfloat %1366 to float, !dbg !31 + %1794 = fpext bfloat %1376 to float, !dbg !31 + %1795 = fpext bfloat %1377 to float, !dbg !31 + %1796 = fpext bfloat %1378 to float, !dbg !31 + %1797 = fpext bfloat %1379 to float, !dbg !31 + %1798 = fpext bfloat %1380 to float, !dbg !31 + %1799 = fpext bfloat %1381 to float, !dbg !31 + %1800 = fpext bfloat %1382 to float, !dbg !31 + %1801 = fpext bfloat %1383 to float, !dbg !31 + %1802 = fpext bfloat %1393 to float, !dbg !31 + %1803 = fpext bfloat %1394 to float, !dbg !31 + %1804 = fpext bfloat %1395 to float, !dbg !31 + %1805 = fpext bfloat %1396 to float, !dbg !31 + %1806 = fpext bfloat %1397 to float, !dbg !31 + %1807 = fpext bfloat %1398 to float, !dbg !31 + %1808 = fpext bfloat %1399 to float, !dbg !31 + %1809 = fpext bfloat %1400 to float, !dbg !31 + %1810 = fpext bfloat %1410 to float, !dbg !31 + %1811 = fpext bfloat %1411 to float, !dbg !31 + %1812 = fpext bfloat %1412 to float, !dbg !31 + %1813 = fpext bfloat %1413 to float, !dbg !31 + %1814 = fpext bfloat %1414 to float, !dbg !31 + %1815 = fpext bfloat %1415 to float, !dbg !31 + %1816 = fpext bfloat %1416 to float, !dbg !31 + %1817 = fpext bfloat %1417 to float, !dbg !31 + %1818 = fpext bfloat %1427 to float, !dbg !31 + %1819 = fpext bfloat %1428 to float, !dbg !31 + %1820 = fpext bfloat %1429 to float, !dbg !31 + %1821 = fpext bfloat %1430 to float, !dbg !31 + %1822 = fpext bfloat %1431 to float, !dbg !31 + %1823 = fpext bfloat %1432 to float, !dbg !31 + %1824 = fpext bfloat %1433 to float, !dbg !31 + %1825 = fpext bfloat %1434 to float, !dbg !31 + %1826 = fpext bfloat %1444 to float, !dbg !31 + %1827 = fpext bfloat %1445 to float, !dbg !31 + %1828 = fpext bfloat %1446 to float, !dbg !31 + %1829 = fpext bfloat %1447 to float, !dbg !31 + %1830 = fpext bfloat %1448 to float, !dbg !31 + %1831 = fpext bfloat %1449 to float, !dbg !31 + %1832 = fpext bfloat %1450 to float, !dbg !31 + %1833 = fpext bfloat %1451 to float, !dbg !31 + %1834 = fpext bfloat %1461 to float, !dbg !31 + %1835 = fpext bfloat %1462 to float, !dbg !31 + %1836 = fpext bfloat %1463 to float, !dbg !31 + %1837 = fpext bfloat %1464 to float, !dbg !31 + %1838 = fpext bfloat %1465 to float, !dbg !31 + %1839 = fpext bfloat %1466 to float, !dbg !31 + %1840 = fpext bfloat %1467 to float, !dbg !31 + %1841 = fpext bfloat %1468 to float, !dbg !31 + %1842 = fpext bfloat %1478 to float, !dbg !31 + %1843 = fpext bfloat %1479 to float, !dbg !31 + %1844 = fpext bfloat %1480 to float, !dbg !31 + %1845 = fpext bfloat %1481 to float, !dbg !31 + %1846 = fpext bfloat %1482 to float, !dbg !31 + %1847 = fpext bfloat %1483 to float, !dbg !31 + %1848 = fpext bfloat %1484 to float, !dbg !31 + %1849 = fpext bfloat %1485 to float, !dbg !31 + %1850 = fpext bfloat %1495 to float, !dbg !31 + %1851 = fpext bfloat %1496 to float, !dbg !31 + %1852 = fpext bfloat %1497 to float, !dbg !31 + %1853 = fpext bfloat %1498 to float, !dbg !31 + %1854 = fpext bfloat %1499 to float, !dbg !31 + %1855 = fpext bfloat %1500 to float, !dbg !31 + %1856 = fpext bfloat %1501 to float, !dbg !31 + %1857 = fpext bfloat %1502 to float, !dbg !31 + %1858 = fpext bfloat %1512 to float, !dbg !31 + %1859 = fpext bfloat %1513 to float, !dbg !31 + %1860 = fpext bfloat %1514 to float, !dbg !31 + %1861 = fpext bfloat %1515 to float, !dbg !31 + %1862 = fpext bfloat %1516 to float, !dbg !31 + %1863 = fpext bfloat %1517 to float, !dbg !31 + %1864 = fpext bfloat %1518 to float, !dbg !31 + %1865 = fpext bfloat %1519 to float, !dbg !31 + %1866 = fpext bfloat %1529 to float, !dbg !31 + %1867 = fpext bfloat %1530 to float, !dbg !31 + %1868 = fpext bfloat %1531 to float, !dbg !31 + %1869 = fpext bfloat %1532 to float, !dbg !31 + %1870 = fpext bfloat %1533 to float, !dbg !31 + %1871 = fpext bfloat %1534 to float, !dbg !31 + %1872 = fpext bfloat %1535 to float, !dbg !31 + %1873 = fpext bfloat %1536 to float, !dbg !31 + %1874 = fpext bfloat %1546 to float, !dbg !31 + %1875 = fpext bfloat %1547 to float, !dbg !31 + %1876 = fpext bfloat %1548 to float, !dbg !31 + %1877 = fpext bfloat %1549 to float, !dbg !31 + %1878 = fpext bfloat %1550 to float, !dbg !31 + %1879 = fpext bfloat %1551 to float, !dbg !31 + %1880 = fpext bfloat %1552 to float, !dbg !31 + %1881 = fpext bfloat %1553 to float, !dbg !31 + %1882 = fpext bfloat %1563 to float, !dbg !31 + %1883 = fpext bfloat %1564 to float, !dbg !31 + %1884 = fpext bfloat %1565 to float, !dbg !31 + %1885 = fpext bfloat %1566 to float, !dbg !31 + %1886 = fpext bfloat %1567 to float, !dbg !31 + %1887 = fpext bfloat %1568 to float, !dbg !31 + %1888 = fpext bfloat %1569 to float, !dbg !31 + %1889 = fpext bfloat %1570 to float, !dbg !31 + %1890 = fpext bfloat %1580 to float, !dbg !31 + %1891 = fpext bfloat %1581 to float, !dbg !31 + %1892 = fpext bfloat %1582 to float, !dbg !31 + %1893 = fpext bfloat %1583 to float, !dbg !31 + %1894 = fpext bfloat %1584 to float, !dbg !31 + %1895 = fpext bfloat %1585 to float, !dbg !31 + %1896 = fpext bfloat %1586 to float, !dbg !31 + %1897 = fpext bfloat %1587 to float, !dbg !31 + %1898 = fpext bfloat %1597 to float, !dbg !31 + %1899 = fpext bfloat %1598 to float, !dbg !31 + %1900 = fpext bfloat %1599 to float, !dbg !31 + %1901 = fpext bfloat %1600 to float, !dbg !31 + %1902 = fpext bfloat %1601 to float, !dbg !31 + %1903 = fpext bfloat %1602 to float, !dbg !31 + %1904 = fpext bfloat %1603 to float, !dbg !31 + %1905 = fpext bfloat %1604 to float, !dbg !31 + %1906 = fpext bfloat %1614 to float, !dbg !31 + %1907 = fpext bfloat %1615 to float, !dbg !31 + %1908 = fpext bfloat %1616 to float, !dbg !31 + %1909 = fpext bfloat %1617 to float, !dbg !31 + %1910 = fpext bfloat %1618 to float, !dbg !31 + %1911 = fpext bfloat %1619 to float, !dbg !31 + %1912 = fpext bfloat %1620 to float, !dbg !31 + %1913 = fpext bfloat %1621 to float, !dbg !31 + %1914 = fpext bfloat %1631 to float, !dbg !31 + %1915 = fpext bfloat %1632 to float, !dbg !31 + %1916 = fpext bfloat %1633 to float, !dbg !31 + %1917 = fpext bfloat %1634 to float, !dbg !31 + %1918 = fpext bfloat %1635 to float, !dbg !31 + %1919 = fpext bfloat %1636 to float, !dbg !31 + %1920 = fpext bfloat %1637 to float, !dbg !31 + %1921 = fpext bfloat %1638 to float, !dbg !31 + %1922 = fpext bfloat %1648 to float, !dbg !31 + %1923 = fpext bfloat %1649 to float, !dbg !31 + %1924 = fpext bfloat %1650 to float, !dbg !31 + %1925 = fpext bfloat %1651 to float, !dbg !31 + %1926 = fpext bfloat %1652 to float, !dbg !31 + %1927 = fpext bfloat %1653 to float, !dbg !31 + %1928 = fpext bfloat %1654 to float, !dbg !31 + %1929 = fpext bfloat %1655 to float, !dbg !31 + %1930 = fpext bfloat %1665 to float, !dbg !31 + %1931 = fpext bfloat %1666 to float, !dbg !31 + %1932 = fpext bfloat %1667 to float, !dbg !31 + %1933 = fpext bfloat %1668 to float, !dbg !31 + %1934 = fpext bfloat %1669 to float, !dbg !31 + %1935 = fpext bfloat %1670 to float, !dbg !31 + %1936 = fpext bfloat %1671 to float, !dbg !31 + %1937 = fpext bfloat %1672 to float, !dbg !31 + %1938 = fpext bfloat %1682 to float, !dbg !31 + %1939 = fpext bfloat %1683 to float, !dbg !31 + %1940 = fpext bfloat %1684 to float, !dbg !31 + %1941 = fpext bfloat %1685 to float, !dbg !31 + %1942 = fpext bfloat %1686 to float, !dbg !31 + %1943 = fpext bfloat %1687 to float, !dbg !31 + %1944 = fpext bfloat %1688 to float, !dbg !31 + %1945 = fpext bfloat %1689 to float, !dbg !31 + %1946 = shl nuw nsw i32 %51, 4, !dbg !32 + %1947 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %1946, !dbg !32 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x2.b16.p3(ptr addrspace(3) %1947, i32 %1143, i32 %1145), !dbg !32 + tail call void @llvm.nvvm.bar.warp.sync(i32 -1), !dbg !32 + %1948 = shl nuw nsw i32 %13, 3, !dbg !32 + %1949 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %1948, !dbg !32 + %1950 = load float, ptr addrspace(3) %1949, align 16, !dbg !32 + %1951 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 4, !dbg !32 + %1952 = load float, ptr addrspace(3) %1951, align 4, !dbg !32 + %1953 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 8, !dbg !32 + %1954 = load float, ptr addrspace(3) %1953, align 8, !dbg !32 + %1955 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 12, !dbg !32 + %1956 = load float, ptr addrspace(3) %1955, align 4, !dbg !32 + %1957 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 16, !dbg !32 + %1958 = load float, ptr addrspace(3) %1957, align 16, !dbg !32 + %1959 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 20, !dbg !32 + %1960 = load float, ptr addrspace(3) %1959, align 4, !dbg !32 + %1961 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 24, !dbg !32 + %1962 = load float, ptr addrspace(3) %1961, align 8, !dbg !32 + %1963 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 28, !dbg !32 + %1964 = load float, ptr addrspace(3) %1963, align 4, !dbg !32 + %1965 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 32, !dbg !32 + %1966 = load float, ptr addrspace(3) %1965, align 16, !dbg !32 + %1967 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 36, !dbg !32 + %1968 = load float, ptr addrspace(3) %1967, align 4, !dbg !32 + %1969 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 40, !dbg !32 + %1970 = load float, ptr addrspace(3) %1969, align 8, !dbg !32 + %1971 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 44, !dbg !32 + %1972 = load float, ptr addrspace(3) %1971, align 4, !dbg !32 + %1973 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 48, !dbg !32 + %1974 = load float, ptr addrspace(3) %1973, align 16, !dbg !32 + %1975 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 52, !dbg !32 + %1976 = load float, ptr addrspace(3) %1975, align 4, !dbg !32 + %1977 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 56, !dbg !32 + %1978 = load float, ptr addrspace(3) %1977, align 8, !dbg !32 + %1979 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 60, !dbg !32 + %1980 = load float, ptr addrspace(3) %1979, align 4, !dbg !32 + %1981 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 64, !dbg !32 + %1982 = load float, ptr addrspace(3) %1981, align 16, !dbg !32 + %1983 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 68, !dbg !32 + %1984 = load float, ptr addrspace(3) %1983, align 4, !dbg !32 + %1985 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 72, !dbg !32 + %1986 = load float, ptr addrspace(3) %1985, align 8, !dbg !32 + %1987 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 76, !dbg !32 + %1988 = load float, ptr addrspace(3) %1987, align 4, !dbg !32 + %1989 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 80, !dbg !32 + %1990 = load float, ptr addrspace(3) %1989, align 16, !dbg !32 + %1991 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 84, !dbg !32 + %1992 = load float, ptr addrspace(3) %1991, align 4, !dbg !32 + %1993 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 88, !dbg !32 + %1994 = load float, ptr addrspace(3) %1993, align 8, !dbg !32 + %1995 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 92, !dbg !32 + %1996 = load float, ptr addrspace(3) %1995, align 4, !dbg !32 + %1997 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 96, !dbg !32 + %1998 = load float, ptr addrspace(3) %1997, align 16, !dbg !32 + %1999 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 100, !dbg !32 + %2000 = load float, ptr addrspace(3) %1999, align 4, !dbg !32 + %2001 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 104, !dbg !32 + %2002 = load float, ptr addrspace(3) %2001, align 8, !dbg !32 + %2003 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 108, !dbg !32 + %2004 = load float, ptr addrspace(3) %2003, align 4, !dbg !32 + %2005 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 112, !dbg !32 + %2006 = load float, ptr addrspace(3) %2005, align 16, !dbg !32 + %2007 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 116, !dbg !32 + %2008 = load float, ptr addrspace(3) %2007, align 4, !dbg !32 + %2009 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 120, !dbg !32 + %2010 = load float, ptr addrspace(3) %2009, align 8, !dbg !32 + %2011 = getelementptr inbounds nuw i8, ptr addrspace(3) %1949, i32 124, !dbg !32 + %2012 = load float, ptr addrspace(3) %2011, align 4, !dbg !32 + %2013 = fmul float %1950, %1690, !dbg !33 + %2014 = fmul float %1950, %1691, !dbg !33 + %2015 = fmul float %1950, %1692, !dbg !33 + %2016 = fmul float %1950, %1693, !dbg !33 + %2017 = fmul float %1950, %1694, !dbg !33 + %2018 = fmul float %1950, %1695, !dbg !33 + %2019 = fmul float %1950, %1696, !dbg !33 + %2020 = fmul float %1950, %1697, !dbg !33 + %2021 = fmul float %1952, %1698, !dbg !33 + %2022 = fmul float %1952, %1699, !dbg !33 + %2023 = fmul float %1952, %1700, !dbg !33 + %2024 = fmul float %1952, %1701, !dbg !33 + %2025 = fmul float %1952, %1702, !dbg !33 + %2026 = fmul float %1952, %1703, !dbg !33 + %2027 = fmul float %1952, %1704, !dbg !33 + %2028 = fmul float %1952, %1705, !dbg !33 + %2029 = fmul float %1954, %1706, !dbg !33 + %2030 = fmul float %1954, %1707, !dbg !33 + %2031 = fmul float %1954, %1708, !dbg !33 + %2032 = fmul float %1954, %1709, !dbg !33 + %2033 = fmul float %1954, %1710, !dbg !33 + %2034 = fmul float %1954, %1711, !dbg !33 + %2035 = fmul float %1954, %1712, !dbg !33 + %2036 = fmul float %1954, %1713, !dbg !33 + %2037 = fmul float %1956, %1714, !dbg !33 + %2038 = fmul float %1956, %1715, !dbg !33 + %2039 = fmul float %1956, %1716, !dbg !33 + %2040 = fmul float %1956, %1717, !dbg !33 + %2041 = fmul float %1956, %1718, !dbg !33 + %2042 = fmul float %1956, %1719, !dbg !33 + %2043 = fmul float %1956, %1720, !dbg !33 + %2044 = fmul float %1956, %1721, !dbg !33 + %2045 = fmul float %1958, %1722, !dbg !33 + %2046 = fmul float %1958, %1723, !dbg !33 + %2047 = fmul float %1958, %1724, !dbg !33 + %2048 = fmul float %1958, %1725, !dbg !33 + %2049 = fmul float %1958, %1726, !dbg !33 + %2050 = fmul float %1958, %1727, !dbg !33 + %2051 = fmul float %1958, %1728, !dbg !33 + %2052 = fmul float %1958, %1729, !dbg !33 + %2053 = fmul float %1960, %1730, !dbg !33 + %2054 = fmul float %1960, %1731, !dbg !33 + %2055 = fmul float %1960, %1732, !dbg !33 + %2056 = fmul float %1960, %1733, !dbg !33 + %2057 = fmul float %1960, %1734, !dbg !33 + %2058 = fmul float %1960, %1735, !dbg !33 + %2059 = fmul float %1960, %1736, !dbg !33 + %2060 = fmul float %1960, %1737, !dbg !33 + %2061 = fmul float %1962, %1738, !dbg !33 + %2062 = fmul float %1962, %1739, !dbg !33 + %2063 = fmul float %1962, %1740, !dbg !33 + %2064 = fmul float %1962, %1741, !dbg !33 + %2065 = fmul float %1962, %1742, !dbg !33 + %2066 = fmul float %1962, %1743, !dbg !33 + %2067 = fmul float %1962, %1744, !dbg !33 + %2068 = fmul float %1962, %1745, !dbg !33 + %2069 = fmul float %1964, %1746, !dbg !33 + %2070 = fmul float %1964, %1747, !dbg !33 + %2071 = fmul float %1964, %1748, !dbg !33 + %2072 = fmul float %1964, %1749, !dbg !33 + %2073 = fmul float %1964, %1750, !dbg !33 + %2074 = fmul float %1964, %1751, !dbg !33 + %2075 = fmul float %1964, %1752, !dbg !33 + %2076 = fmul float %1964, %1753, !dbg !33 + %2077 = fmul float %1966, %1754, !dbg !33 + %2078 = fmul float %1966, %1755, !dbg !33 + %2079 = fmul float %1966, %1756, !dbg !33 + %2080 = fmul float %1966, %1757, !dbg !33 + %2081 = fmul float %1966, %1758, !dbg !33 + %2082 = fmul float %1966, %1759, !dbg !33 + %2083 = fmul float %1966, %1760, !dbg !33 + %2084 = fmul float %1966, %1761, !dbg !33 + %2085 = fmul float %1968, %1762, !dbg !33 + %2086 = fmul float %1968, %1763, !dbg !33 + %2087 = fmul float %1968, %1764, !dbg !33 + %2088 = fmul float %1968, %1765, !dbg !33 + %2089 = fmul float %1968, %1766, !dbg !33 + %2090 = fmul float %1968, %1767, !dbg !33 + %2091 = fmul float %1968, %1768, !dbg !33 + %2092 = fmul float %1968, %1769, !dbg !33 + %2093 = fmul float %1970, %1770, !dbg !33 + %2094 = fmul float %1970, %1771, !dbg !33 + %2095 = fmul float %1970, %1772, !dbg !33 + %2096 = fmul float %1970, %1773, !dbg !33 + %2097 = fmul float %1970, %1774, !dbg !33 + %2098 = fmul float %1970, %1775, !dbg !33 + %2099 = fmul float %1970, %1776, !dbg !33 + %2100 = fmul float %1970, %1777, !dbg !33 + %2101 = fmul float %1972, %1778, !dbg !33 + %2102 = fmul float %1972, %1779, !dbg !33 + %2103 = fmul float %1972, %1780, !dbg !33 + %2104 = fmul float %1972, %1781, !dbg !33 + %2105 = fmul float %1972, %1782, !dbg !33 + %2106 = fmul float %1972, %1783, !dbg !33 + %2107 = fmul float %1972, %1784, !dbg !33 + %2108 = fmul float %1972, %1785, !dbg !33 + %2109 = fmul float %1974, %1786, !dbg !33 + %2110 = fmul float %1974, %1787, !dbg !33 + %2111 = fmul float %1974, %1788, !dbg !33 + %2112 = fmul float %1974, %1789, !dbg !33 + %2113 = fmul float %1974, %1790, !dbg !33 + %2114 = fmul float %1974, %1791, !dbg !33 + %2115 = fmul float %1974, %1792, !dbg !33 + %2116 = fmul float %1974, %1793, !dbg !33 + %2117 = fmul float %1976, %1794, !dbg !33 + %2118 = fmul float %1976, %1795, !dbg !33 + %2119 = fmul float %1976, %1796, !dbg !33 + %2120 = fmul float %1976, %1797, !dbg !33 + %2121 = fmul float %1976, %1798, !dbg !33 + %2122 = fmul float %1976, %1799, !dbg !33 + %2123 = fmul float %1976, %1800, !dbg !33 + %2124 = fmul float %1976, %1801, !dbg !33 + %2125 = fmul float %1978, %1802, !dbg !33 + %2126 = fmul float %1978, %1803, !dbg !33 + %2127 = fmul float %1978, %1804, !dbg !33 + %2128 = fmul float %1978, %1805, !dbg !33 + %2129 = fmul float %1978, %1806, !dbg !33 + %2130 = fmul float %1978, %1807, !dbg !33 + %2131 = fmul float %1978, %1808, !dbg !33 + %2132 = fmul float %1978, %1809, !dbg !33 + %2133 = fmul float %1980, %1810, !dbg !33 + %2134 = fmul float %1980, %1811, !dbg !33 + %2135 = fmul float %1980, %1812, !dbg !33 + %2136 = fmul float %1980, %1813, !dbg !33 + %2137 = fmul float %1980, %1814, !dbg !33 + %2138 = fmul float %1980, %1815, !dbg !33 + %2139 = fmul float %1980, %1816, !dbg !33 + %2140 = fmul float %1980, %1817, !dbg !33 + %2141 = fmul float %1982, %1818, !dbg !33 + %2142 = fmul float %1982, %1819, !dbg !33 + %2143 = fmul float %1982, %1820, !dbg !33 + %2144 = fmul float %1982, %1821, !dbg !33 + %2145 = fmul float %1982, %1822, !dbg !33 + %2146 = fmul float %1982, %1823, !dbg !33 + %2147 = fmul float %1982, %1824, !dbg !33 + %2148 = fmul float %1982, %1825, !dbg !33 + %2149 = fmul float %1984, %1826, !dbg !33 + %2150 = fmul float %1984, %1827, !dbg !33 + %2151 = fmul float %1984, %1828, !dbg !33 + %2152 = fmul float %1984, %1829, !dbg !33 + %2153 = fmul float %1984, %1830, !dbg !33 + %2154 = fmul float %1984, %1831, !dbg !33 + %2155 = fmul float %1984, %1832, !dbg !33 + %2156 = fmul float %1984, %1833, !dbg !33 + %2157 = fmul float %1986, %1834, !dbg !33 + %2158 = fmul float %1986, %1835, !dbg !33 + %2159 = fmul float %1986, %1836, !dbg !33 + %2160 = fmul float %1986, %1837, !dbg !33 + %2161 = fmul float %1986, %1838, !dbg !33 + %2162 = fmul float %1986, %1839, !dbg !33 + %2163 = fmul float %1986, %1840, !dbg !33 + %2164 = fmul float %1986, %1841, !dbg !33 + %2165 = fmul float %1988, %1842, !dbg !33 + %2166 = fmul float %1988, %1843, !dbg !33 + %2167 = fmul float %1988, %1844, !dbg !33 + %2168 = fmul float %1988, %1845, !dbg !33 + %2169 = fmul float %1988, %1846, !dbg !33 + %2170 = fmul float %1988, %1847, !dbg !33 + %2171 = fmul float %1988, %1848, !dbg !33 + %2172 = fmul float %1988, %1849, !dbg !33 + %2173 = fmul float %1990, %1850, !dbg !33 + %2174 = fmul float %1990, %1851, !dbg !33 + %2175 = fmul float %1990, %1852, !dbg !33 + %2176 = fmul float %1990, %1853, !dbg !33 + %2177 = fmul float %1990, %1854, !dbg !33 + %2178 = fmul float %1990, %1855, !dbg !33 + %2179 = fmul float %1990, %1856, !dbg !33 + %2180 = fmul float %1990, %1857, !dbg !33 + %2181 = fmul float %1992, %1858, !dbg !33 + %2182 = fmul float %1992, %1859, !dbg !33 + %2183 = fmul float %1992, %1860, !dbg !33 + %2184 = fmul float %1992, %1861, !dbg !33 + %2185 = fmul float %1992, %1862, !dbg !33 + %2186 = fmul float %1992, %1863, !dbg !33 + %2187 = fmul float %1992, %1864, !dbg !33 + %2188 = fmul float %1992, %1865, !dbg !33 + %2189 = fmul float %1994, %1866, !dbg !33 + %2190 = fmul float %1994, %1867, !dbg !33 + %2191 = fmul float %1994, %1868, !dbg !33 + %2192 = fmul float %1994, %1869, !dbg !33 + %2193 = fmul float %1994, %1870, !dbg !33 + %2194 = fmul float %1994, %1871, !dbg !33 + %2195 = fmul float %1994, %1872, !dbg !33 + %2196 = fmul float %1994, %1873, !dbg !33 + %2197 = fmul float %1996, %1874, !dbg !33 + %2198 = fmul float %1996, %1875, !dbg !33 + %2199 = fmul float %1996, %1876, !dbg !33 + %2200 = fmul float %1996, %1877, !dbg !33 + %2201 = fmul float %1996, %1878, !dbg !33 + %2202 = fmul float %1996, %1879, !dbg !33 + %2203 = fmul float %1996, %1880, !dbg !33 + %2204 = fmul float %1996, %1881, !dbg !33 + %2205 = fmul float %1998, %1882, !dbg !33 + %2206 = fmul float %1998, %1883, !dbg !33 + %2207 = fmul float %1998, %1884, !dbg !33 + %2208 = fmul float %1998, %1885, !dbg !33 + %2209 = fmul float %1998, %1886, !dbg !33 + %2210 = fmul float %1998, %1887, !dbg !33 + %2211 = fmul float %1998, %1888, !dbg !33 + %2212 = fmul float %1998, %1889, !dbg !33 + %2213 = fmul float %2000, %1890, !dbg !33 + %2214 = fmul float %2000, %1891, !dbg !33 + %2215 = fmul float %2000, %1892, !dbg !33 + %2216 = fmul float %2000, %1893, !dbg !33 + %2217 = fmul float %2000, %1894, !dbg !33 + %2218 = fmul float %2000, %1895, !dbg !33 + %2219 = fmul float %2000, %1896, !dbg !33 + %2220 = fmul float %2000, %1897, !dbg !33 + %2221 = fmul float %2002, %1898, !dbg !33 + %2222 = fmul float %2002, %1899, !dbg !33 + %2223 = fmul float %2002, %1900, !dbg !33 + %2224 = fmul float %2002, %1901, !dbg !33 + %2225 = fmul float %2002, %1902, !dbg !33 + %2226 = fmul float %2002, %1903, !dbg !33 + %2227 = fmul float %2002, %1904, !dbg !33 + %2228 = fmul float %2002, %1905, !dbg !33 + %2229 = fmul float %2004, %1906, !dbg !33 + %2230 = fmul float %2004, %1907, !dbg !33 + %2231 = fmul float %2004, %1908, !dbg !33 + %2232 = fmul float %2004, %1909, !dbg !33 + %2233 = fmul float %2004, %1910, !dbg !33 + %2234 = fmul float %2004, %1911, !dbg !33 + %2235 = fmul float %2004, %1912, !dbg !33 + %2236 = fmul float %2004, %1913, !dbg !33 + %2237 = fmul float %2006, %1914, !dbg !33 + %2238 = fmul float %2006, %1915, !dbg !33 + %2239 = fmul float %2006, %1916, !dbg !33 + %2240 = fmul float %2006, %1917, !dbg !33 + %2241 = fmul float %2006, %1918, !dbg !33 + %2242 = fmul float %2006, %1919, !dbg !33 + %2243 = fmul float %2006, %1920, !dbg !33 + %2244 = fmul float %2006, %1921, !dbg !33 + %2245 = fmul float %2008, %1922, !dbg !33 + %2246 = fmul float %2008, %1923, !dbg !33 + %2247 = fmul float %2008, %1924, !dbg !33 + %2248 = fmul float %2008, %1925, !dbg !33 + %2249 = fmul float %2008, %1926, !dbg !33 + %2250 = fmul float %2008, %1927, !dbg !33 + %2251 = fmul float %2008, %1928, !dbg !33 + %2252 = fmul float %2008, %1929, !dbg !33 + %2253 = fmul float %2010, %1930, !dbg !33 + %2254 = fmul float %2010, %1931, !dbg !33 + %2255 = fmul float %2010, %1932, !dbg !33 + %2256 = fmul float %2010, %1933, !dbg !33 + %2257 = fmul float %2010, %1934, !dbg !33 + %2258 = fmul float %2010, %1935, !dbg !33 + %2259 = fmul float %2010, %1936, !dbg !33 + %2260 = fmul float %2010, %1937, !dbg !33 + %2261 = fmul float %2012, %1938, !dbg !33 + %2262 = fmul float %2012, %1939, !dbg !33 + %2263 = fmul float %2012, %1940, !dbg !33 + %2264 = fmul float %2012, %1941, !dbg !33 + %2265 = fmul float %2012, %1942, !dbg !33 + %2266 = fmul float %2012, %1943, !dbg !33 + %2267 = fmul float %2012, %1944, !dbg !33 + %2268 = fmul float %2012, %1945, !dbg !33 + %2269 = fmul float %886, %1690, !dbg !34 + %2270 = fmul float %887, %1691, !dbg !34 + %2271 = fmul float %888, %1692, !dbg !34 + %2272 = fmul float %889, %1693, !dbg !34 + %2273 = fmul float %890, %1694, !dbg !34 + %2274 = fmul float %891, %1695, !dbg !34 + %2275 = fmul float %892, %1696, !dbg !34 + %2276 = fmul float %893, %1697, !dbg !34 + %2277 = fmul float %894, %1698, !dbg !34 + %2278 = fmul float %895, %1699, !dbg !34 + %2279 = fmul float %896, %1700, !dbg !34 + %2280 = fmul float %897, %1701, !dbg !34 + %2281 = fmul float %898, %1702, !dbg !34 + %2282 = fmul float %899, %1703, !dbg !34 + %2283 = fmul float %900, %1704, !dbg !34 + %2284 = fmul float %901, %1705, !dbg !34 + %2285 = fmul float %902, %1706, !dbg !34 + %2286 = fmul float %903, %1707, !dbg !34 + %2287 = fmul float %904, %1708, !dbg !34 + %2288 = fmul float %905, %1709, !dbg !34 + %2289 = fmul float %906, %1710, !dbg !34 + %2290 = fmul float %907, %1711, !dbg !34 + %2291 = fmul float %908, %1712, !dbg !34 + %2292 = fmul float %909, %1713, !dbg !34 + %2293 = fmul float %910, %1714, !dbg !34 + %2294 = fmul float %911, %1715, !dbg !34 + %2295 = fmul float %912, %1716, !dbg !34 + %2296 = fmul float %913, %1717, !dbg !34 + %2297 = fmul float %914, %1718, !dbg !34 + %2298 = fmul float %915, %1719, !dbg !34 + %2299 = fmul float %916, %1720, !dbg !34 + %2300 = fmul float %917, %1721, !dbg !34 + %2301 = fmul float %918, %1722, !dbg !34 + %2302 = fmul float %919, %1723, !dbg !34 + %2303 = fmul float %920, %1724, !dbg !34 + %2304 = fmul float %921, %1725, !dbg !34 + %2305 = fmul float %922, %1726, !dbg !34 + %2306 = fmul float %923, %1727, !dbg !34 + %2307 = fmul float %924, %1728, !dbg !34 + %2308 = fmul float %925, %1729, !dbg !34 + %2309 = fmul float %926, %1730, !dbg !34 + %2310 = fmul float %927, %1731, !dbg !34 + %2311 = fmul float %928, %1732, !dbg !34 + %2312 = fmul float %929, %1733, !dbg !34 + %2313 = fmul float %930, %1734, !dbg !34 + %2314 = fmul float %931, %1735, !dbg !34 + %2315 = fmul float %932, %1736, !dbg !34 + %2316 = fmul float %933, %1737, !dbg !34 + %2317 = fmul float %934, %1738, !dbg !34 + %2318 = fmul float %935, %1739, !dbg !34 + %2319 = fmul float %936, %1740, !dbg !34 + %2320 = fmul float %937, %1741, !dbg !34 + %2321 = fmul float %938, %1742, !dbg !34 + %2322 = fmul float %939, %1743, !dbg !34 + %2323 = fmul float %940, %1744, !dbg !34 + %2324 = fmul float %941, %1745, !dbg !34 + %2325 = fmul float %942, %1746, !dbg !34 + %2326 = fmul float %943, %1747, !dbg !34 + %2327 = fmul float %944, %1748, !dbg !34 + %2328 = fmul float %945, %1749, !dbg !34 + %2329 = fmul float %946, %1750, !dbg !34 + %2330 = fmul float %947, %1751, !dbg !34 + %2331 = fmul float %948, %1752, !dbg !34 + %2332 = fmul float %949, %1753, !dbg !34 + %2333 = fmul float %950, %1754, !dbg !34 + %2334 = fmul float %951, %1755, !dbg !34 + %2335 = fmul float %952, %1756, !dbg !34 + %2336 = fmul float %953, %1757, !dbg !34 + %2337 = fmul float %954, %1758, !dbg !34 + %2338 = fmul float %955, %1759, !dbg !34 + %2339 = fmul float %956, %1760, !dbg !34 + %2340 = fmul float %957, %1761, !dbg !34 + %2341 = fmul float %958, %1762, !dbg !34 + %2342 = fmul float %959, %1763, !dbg !34 + %2343 = fmul float %960, %1764, !dbg !34 + %2344 = fmul float %961, %1765, !dbg !34 + %2345 = fmul float %962, %1766, !dbg !34 + %2346 = fmul float %963, %1767, !dbg !34 + %2347 = fmul float %964, %1768, !dbg !34 + %2348 = fmul float %965, %1769, !dbg !34 + %2349 = fmul float %966, %1770, !dbg !34 + %2350 = fmul float %967, %1771, !dbg !34 + %2351 = fmul float %968, %1772, !dbg !34 + %2352 = fmul float %969, %1773, !dbg !34 + %2353 = fmul float %970, %1774, !dbg !34 + %2354 = fmul float %971, %1775, !dbg !34 + %2355 = fmul float %972, %1776, !dbg !34 + %2356 = fmul float %973, %1777, !dbg !34 + %2357 = fmul float %974, %1778, !dbg !34 + %2358 = fmul float %975, %1779, !dbg !34 + %2359 = fmul float %976, %1780, !dbg !34 + %2360 = fmul float %977, %1781, !dbg !34 + %2361 = fmul float %978, %1782, !dbg !34 + %2362 = fmul float %979, %1783, !dbg !34 + %2363 = fmul float %980, %1784, !dbg !34 + %2364 = fmul float %981, %1785, !dbg !34 + %2365 = fmul float %982, %1786, !dbg !34 + %2366 = fmul float %983, %1787, !dbg !34 + %2367 = fmul float %984, %1788, !dbg !34 + %2368 = fmul float %985, %1789, !dbg !34 + %2369 = fmul float %986, %1790, !dbg !34 + %2370 = fmul float %987, %1791, !dbg !34 + %2371 = fmul float %988, %1792, !dbg !34 + %2372 = fmul float %989, %1793, !dbg !34 + %2373 = fmul float %990, %1794, !dbg !34 + %2374 = fmul float %991, %1795, !dbg !34 + %2375 = fmul float %992, %1796, !dbg !34 + %2376 = fmul float %993, %1797, !dbg !34 + %2377 = fmul float %994, %1798, !dbg !34 + %2378 = fmul float %995, %1799, !dbg !34 + %2379 = fmul float %996, %1800, !dbg !34 + %2380 = fmul float %997, %1801, !dbg !34 + %2381 = fmul float %998, %1802, !dbg !34 + %2382 = fmul float %999, %1803, !dbg !34 + %2383 = fmul float %1000, %1804, !dbg !34 + %2384 = fmul float %1001, %1805, !dbg !34 + %2385 = fmul float %1002, %1806, !dbg !34 + %2386 = fmul float %1003, %1807, !dbg !34 + %2387 = fmul float %1004, %1808, !dbg !34 + %2388 = fmul float %1005, %1809, !dbg !34 + %2389 = fmul float %1006, %1810, !dbg !34 + %2390 = fmul float %1007, %1811, !dbg !34 + %2391 = fmul float %1008, %1812, !dbg !34 + %2392 = fmul float %1009, %1813, !dbg !34 + %2393 = fmul float %1010, %1814, !dbg !34 + %2394 = fmul float %1011, %1815, !dbg !34 + %2395 = fmul float %1012, %1816, !dbg !34 + %2396 = fmul float %1013, %1817, !dbg !34 + %2397 = fmul float %1014, %1818, !dbg !34 + %2398 = fmul float %1015, %1819, !dbg !34 + %2399 = fmul float %1016, %1820, !dbg !34 + %2400 = fmul float %1017, %1821, !dbg !34 + %2401 = fmul float %1018, %1822, !dbg !34 + %2402 = fmul float %1019, %1823, !dbg !34 + %2403 = fmul float %1020, %1824, !dbg !34 + %2404 = fmul float %1021, %1825, !dbg !34 + %2405 = fmul float %1022, %1826, !dbg !34 + %2406 = fmul float %1023, %1827, !dbg !34 + %2407 = fmul float %1024, %1828, !dbg !34 + %2408 = fmul float %1025, %1829, !dbg !34 + %2409 = fmul float %1026, %1830, !dbg !34 + %2410 = fmul float %1027, %1831, !dbg !34 + %2411 = fmul float %1028, %1832, !dbg !34 + %2412 = fmul float %1029, %1833, !dbg !34 + %2413 = fmul float %1030, %1834, !dbg !34 + %2414 = fmul float %1031, %1835, !dbg !34 + %2415 = fmul float %1032, %1836, !dbg !34 + %2416 = fmul float %1033, %1837, !dbg !34 + %2417 = fmul float %1034, %1838, !dbg !34 + %2418 = fmul float %1035, %1839, !dbg !34 + %2419 = fmul float %1036, %1840, !dbg !34 + %2420 = fmul float %1037, %1841, !dbg !34 + %2421 = fmul float %1038, %1842, !dbg !34 + %2422 = fmul float %1039, %1843, !dbg !34 + %2423 = fmul float %1040, %1844, !dbg !34 + %2424 = fmul float %1041, %1845, !dbg !34 + %2425 = fmul float %1042, %1846, !dbg !34 + %2426 = fmul float %1043, %1847, !dbg !34 + %2427 = fmul float %1044, %1848, !dbg !34 + %2428 = fmul float %1045, %1849, !dbg !34 + %2429 = fmul float %1046, %1850, !dbg !34 + %2430 = fmul float %1047, %1851, !dbg !34 + %2431 = fmul float %1048, %1852, !dbg !34 + %2432 = fmul float %1049, %1853, !dbg !34 + %2433 = fmul float %1050, %1854, !dbg !34 + %2434 = fmul float %1051, %1855, !dbg !34 + %2435 = fmul float %1052, %1856, !dbg !34 + %2436 = fmul float %1053, %1857, !dbg !34 + %2437 = fmul float %1054, %1858, !dbg !34 + %2438 = fmul float %1055, %1859, !dbg !34 + %2439 = fmul float %1056, %1860, !dbg !34 + %2440 = fmul float %1057, %1861, !dbg !34 + %2441 = fmul float %1058, %1862, !dbg !34 + %2442 = fmul float %1059, %1863, !dbg !34 + %2443 = fmul float %1060, %1864, !dbg !34 + %2444 = fmul float %1061, %1865, !dbg !34 + %2445 = fmul float %1062, %1866, !dbg !34 + %2446 = fmul float %1063, %1867, !dbg !34 + %2447 = fmul float %1064, %1868, !dbg !34 + %2448 = fmul float %1065, %1869, !dbg !34 + %2449 = fmul float %1066, %1870, !dbg !34 + %2450 = fmul float %1067, %1871, !dbg !34 + %2451 = fmul float %1068, %1872, !dbg !34 + %2452 = fmul float %1069, %1873, !dbg !34 + %2453 = fmul float %1070, %1874, !dbg !34 + %2454 = fmul float %1071, %1875, !dbg !34 + %2455 = fmul float %1072, %1876, !dbg !34 + %2456 = fmul float %1073, %1877, !dbg !34 + %2457 = fmul float %1074, %1878, !dbg !34 + %2458 = fmul float %1075, %1879, !dbg !34 + %2459 = fmul float %1076, %1880, !dbg !34 + %2460 = fmul float %1077, %1881, !dbg !34 + %2461 = fmul float %1078, %1882, !dbg !34 + %2462 = fmul float %1079, %1883, !dbg !34 + %2463 = fmul float %1080, %1884, !dbg !34 + %2464 = fmul float %1081, %1885, !dbg !34 + %2465 = fmul float %1082, %1886, !dbg !34 + %2466 = fmul float %1083, %1887, !dbg !34 + %2467 = fmul float %1084, %1888, !dbg !34 + %2468 = fmul float %1085, %1889, !dbg !34 + %2469 = fmul float %1086, %1890, !dbg !34 + %2470 = fmul float %1087, %1891, !dbg !34 + %2471 = fmul float %1088, %1892, !dbg !34 + %2472 = fmul float %1089, %1893, !dbg !34 + %2473 = fmul float %1090, %1894, !dbg !34 + %2474 = fmul float %1091, %1895, !dbg !34 + %2475 = fmul float %1092, %1896, !dbg !34 + %2476 = fmul float %1093, %1897, !dbg !34 + %2477 = fmul float %1094, %1898, !dbg !34 + %2478 = fmul float %1095, %1899, !dbg !34 + %2479 = fmul float %1096, %1900, !dbg !34 + %2480 = fmul float %1097, %1901, !dbg !34 + %2481 = fmul float %1098, %1902, !dbg !34 + %2482 = fmul float %1099, %1903, !dbg !34 + %2483 = fmul float %1100, %1904, !dbg !34 + %2484 = fmul float %1101, %1905, !dbg !34 + %2485 = fmul float %1102, %1906, !dbg !34 + %2486 = fmul float %1103, %1907, !dbg !34 + %2487 = fmul float %1104, %1908, !dbg !34 + %2488 = fmul float %1105, %1909, !dbg !34 + %2489 = fmul float %1106, %1910, !dbg !34 + %2490 = fmul float %1107, %1911, !dbg !34 + %2491 = fmul float %1108, %1912, !dbg !34 + %2492 = fmul float %1109, %1913, !dbg !34 + %2493 = fmul float %1110, %1914, !dbg !34 + %2494 = fmul float %1111, %1915, !dbg !34 + %2495 = fmul float %1112, %1916, !dbg !34 + %2496 = fmul float %1113, %1917, !dbg !34 + %2497 = fmul float %1114, %1918, !dbg !34 + %2498 = fmul float %1115, %1919, !dbg !34 + %2499 = fmul float %1116, %1920, !dbg !34 + %2500 = fmul float %1117, %1921, !dbg !34 + %2501 = fmul float %1118, %1922, !dbg !34 + %2502 = fmul float %1119, %1923, !dbg !34 + %2503 = fmul float %1120, %1924, !dbg !34 + %2504 = fmul float %1121, %1925, !dbg !34 + %2505 = fmul float %1122, %1926, !dbg !34 + %2506 = fmul float %1123, %1927, !dbg !34 + %2507 = fmul float %1124, %1928, !dbg !34 + %2508 = fmul float %1125, %1929, !dbg !34 + %2509 = fmul float %1126, %1930, !dbg !34 + %2510 = fmul float %1127, %1931, !dbg !34 + %2511 = fmul float %1128, %1932, !dbg !34 + %2512 = fmul float %1129, %1933, !dbg !34 + %2513 = fmul float %1130, %1934, !dbg !34 + %2514 = fmul float %1131, %1935, !dbg !34 + %2515 = fmul float %1132, %1936, !dbg !34 + %2516 = fmul float %1133, %1937, !dbg !34 + %2517 = fmul float %1134, %1938, !dbg !34 + %2518 = fmul float %1135, %1939, !dbg !34 + %2519 = fmul float %1136, %1940, !dbg !34 + %2520 = fmul float %1137, %1941, !dbg !34 + %2521 = fmul float %1138, %1942, !dbg !34 + %2522 = fmul float %1139, %1943, !dbg !34 + %2523 = fmul float %1140, %1944, !dbg !34 + %2524 = fmul float %1141, %1945, !dbg !34 + %2525 = fadd float %2269, %2270, !dbg !35 + %2526 = fadd float %2271, %2525, !dbg !35 + %2527 = fadd float %2272, %2526, !dbg !35 + %2528 = fadd float %2273, %2527, !dbg !35 + %2529 = fadd float %2274, %2528, !dbg !35 + %2530 = fadd float %2275, %2529, !dbg !35 + %2531 = fadd float %2276, %2530, !dbg !35 + %2532 = fadd float %2277, %2278, !dbg !35 + %2533 = fadd float %2279, %2532, !dbg !35 + %2534 = fadd float %2280, %2533, !dbg !35 + %2535 = fadd float %2281, %2534, !dbg !35 + %2536 = fadd float %2282, %2535, !dbg !35 + %2537 = fadd float %2283, %2536, !dbg !35 + %2538 = fadd float %2284, %2537, !dbg !35 + %2539 = fadd float %2285, %2286, !dbg !35 + %2540 = fadd float %2287, %2539, !dbg !35 + %2541 = fadd float %2288, %2540, !dbg !35 + %2542 = fadd float %2289, %2541, !dbg !35 + %2543 = fadd float %2290, %2542, !dbg !35 + %2544 = fadd float %2291, %2543, !dbg !35 + %2545 = fadd float %2292, %2544, !dbg !35 + %2546 = fadd float %2293, %2294, !dbg !35 + %2547 = fadd float %2295, %2546, !dbg !35 + %2548 = fadd float %2296, %2547, !dbg !35 + %2549 = fadd float %2297, %2548, !dbg !35 + %2550 = fadd float %2298, %2549, !dbg !35 + %2551 = fadd float %2299, %2550, !dbg !35 + %2552 = fadd float %2300, %2551, !dbg !35 + %2553 = fadd float %2301, %2302, !dbg !35 + %2554 = fadd float %2303, %2553, !dbg !35 + %2555 = fadd float %2304, %2554, !dbg !35 + %2556 = fadd float %2305, %2555, !dbg !35 + %2557 = fadd float %2306, %2556, !dbg !35 + %2558 = fadd float %2307, %2557, !dbg !35 + %2559 = fadd float %2308, %2558, !dbg !35 + %2560 = fadd float %2309, %2310, !dbg !35 + %2561 = fadd float %2311, %2560, !dbg !35 + %2562 = fadd float %2312, %2561, !dbg !35 + %2563 = fadd float %2313, %2562, !dbg !35 + %2564 = fadd float %2314, %2563, !dbg !35 + %2565 = fadd float %2315, %2564, !dbg !35 + %2566 = fadd float %2316, %2565, !dbg !35 + %2567 = fadd float %2317, %2318, !dbg !35 + %2568 = fadd float %2319, %2567, !dbg !35 + %2569 = fadd float %2320, %2568, !dbg !35 + %2570 = fadd float %2321, %2569, !dbg !35 + %2571 = fadd float %2322, %2570, !dbg !35 + %2572 = fadd float %2323, %2571, !dbg !35 + %2573 = fadd float %2324, %2572, !dbg !35 + %2574 = fadd float %2325, %2326, !dbg !35 + %2575 = fadd float %2327, %2574, !dbg !35 + %2576 = fadd float %2328, %2575, !dbg !35 + %2577 = fadd float %2329, %2576, !dbg !35 + %2578 = fadd float %2330, %2577, !dbg !35 + %2579 = fadd float %2331, %2578, !dbg !35 + %2580 = fadd float %2332, %2579, !dbg !35 + %2581 = fadd float %2333, %2334, !dbg !35 + %2582 = fadd float %2335, %2581, !dbg !35 + %2583 = fadd float %2336, %2582, !dbg !35 + %2584 = fadd float %2337, %2583, !dbg !35 + %2585 = fadd float %2338, %2584, !dbg !35 + %2586 = fadd float %2339, %2585, !dbg !35 + %2587 = fadd float %2340, %2586, !dbg !35 + %2588 = fadd float %2341, %2342, !dbg !35 + %2589 = fadd float %2343, %2588, !dbg !35 + %2590 = fadd float %2344, %2589, !dbg !35 + %2591 = fadd float %2345, %2590, !dbg !35 + %2592 = fadd float %2346, %2591, !dbg !35 + %2593 = fadd float %2347, %2592, !dbg !35 + %2594 = fadd float %2348, %2593, !dbg !35 + %2595 = fadd float %2349, %2350, !dbg !35 + %2596 = fadd float %2351, %2595, !dbg !35 + %2597 = fadd float %2352, %2596, !dbg !35 + %2598 = fadd float %2353, %2597, !dbg !35 + %2599 = fadd float %2354, %2598, !dbg !35 + %2600 = fadd float %2355, %2599, !dbg !35 + %2601 = fadd float %2356, %2600, !dbg !35 + %2602 = fadd float %2357, %2358, !dbg !35 + %2603 = fadd float %2359, %2602, !dbg !35 + %2604 = fadd float %2360, %2603, !dbg !35 + %2605 = fadd float %2361, %2604, !dbg !35 + %2606 = fadd float %2362, %2605, !dbg !35 + %2607 = fadd float %2363, %2606, !dbg !35 + %2608 = fadd float %2364, %2607, !dbg !35 + %2609 = fadd float %2365, %2366, !dbg !35 + %2610 = fadd float %2367, %2609, !dbg !35 + %2611 = fadd float %2368, %2610, !dbg !35 + %2612 = fadd float %2369, %2611, !dbg !35 + %2613 = fadd float %2370, %2612, !dbg !35 + %2614 = fadd float %2371, %2613, !dbg !35 + %2615 = fadd float %2372, %2614, !dbg !35 + %2616 = fadd float %2373, %2374, !dbg !35 + %2617 = fadd float %2375, %2616, !dbg !35 + %2618 = fadd float %2376, %2617, !dbg !35 + %2619 = fadd float %2377, %2618, !dbg !35 + %2620 = fadd float %2378, %2619, !dbg !35 + %2621 = fadd float %2379, %2620, !dbg !35 + %2622 = fadd float %2380, %2621, !dbg !35 + %2623 = fadd float %2381, %2382, !dbg !35 + %2624 = fadd float %2383, %2623, !dbg !35 + %2625 = fadd float %2384, %2624, !dbg !35 + %2626 = fadd float %2385, %2625, !dbg !35 + %2627 = fadd float %2386, %2626, !dbg !35 + %2628 = fadd float %2387, %2627, !dbg !35 + %2629 = fadd float %2388, %2628, !dbg !35 + %2630 = fadd float %2389, %2390, !dbg !35 + %2631 = fadd float %2391, %2630, !dbg !35 + %2632 = fadd float %2392, %2631, !dbg !35 + %2633 = fadd float %2393, %2632, !dbg !35 + %2634 = fadd float %2394, %2633, !dbg !35 + %2635 = fadd float %2395, %2634, !dbg !35 + %2636 = fadd float %2396, %2635, !dbg !35 + %2637 = fadd float %2397, %2398, !dbg !35 + %2638 = fadd float %2399, %2637, !dbg !35 + %2639 = fadd float %2400, %2638, !dbg !35 + %2640 = fadd float %2401, %2639, !dbg !35 + %2641 = fadd float %2402, %2640, !dbg !35 + %2642 = fadd float %2403, %2641, !dbg !35 + %2643 = fadd float %2404, %2642, !dbg !35 + %2644 = fadd float %2405, %2406, !dbg !35 + %2645 = fadd float %2407, %2644, !dbg !35 + %2646 = fadd float %2408, %2645, !dbg !35 + %2647 = fadd float %2409, %2646, !dbg !35 + %2648 = fadd float %2410, %2647, !dbg !35 + %2649 = fadd float %2411, %2648, !dbg !35 + %2650 = fadd float %2412, %2649, !dbg !35 + %2651 = fadd float %2413, %2414, !dbg !35 + %2652 = fadd float %2415, %2651, !dbg !35 + %2653 = fadd float %2416, %2652, !dbg !35 + %2654 = fadd float %2417, %2653, !dbg !35 + %2655 = fadd float %2418, %2654, !dbg !35 + %2656 = fadd float %2419, %2655, !dbg !35 + %2657 = fadd float %2420, %2656, !dbg !35 + %2658 = fadd float %2421, %2422, !dbg !35 + %2659 = fadd float %2423, %2658, !dbg !35 + %2660 = fadd float %2424, %2659, !dbg !35 + %2661 = fadd float %2425, %2660, !dbg !35 + %2662 = fadd float %2426, %2661, !dbg !35 + %2663 = fadd float %2427, %2662, !dbg !35 + %2664 = fadd float %2428, %2663, !dbg !35 + %2665 = fadd float %2429, %2430, !dbg !35 + %2666 = fadd float %2431, %2665, !dbg !35 + %2667 = fadd float %2432, %2666, !dbg !35 + %2668 = fadd float %2433, %2667, !dbg !35 + %2669 = fadd float %2434, %2668, !dbg !35 + %2670 = fadd float %2435, %2669, !dbg !35 + %2671 = fadd float %2436, %2670, !dbg !35 + %2672 = fadd float %2437, %2438, !dbg !35 + %2673 = fadd float %2439, %2672, !dbg !35 + %2674 = fadd float %2440, %2673, !dbg !35 + %2675 = fadd float %2441, %2674, !dbg !35 + %2676 = fadd float %2442, %2675, !dbg !35 + %2677 = fadd float %2443, %2676, !dbg !35 + %2678 = fadd float %2444, %2677, !dbg !35 + %2679 = fadd float %2445, %2446, !dbg !35 + %2680 = fadd float %2447, %2679, !dbg !35 + %2681 = fadd float %2448, %2680, !dbg !35 + %2682 = fadd float %2449, %2681, !dbg !35 + %2683 = fadd float %2450, %2682, !dbg !35 + %2684 = fadd float %2451, %2683, !dbg !35 + %2685 = fadd float %2452, %2684, !dbg !35 + %2686 = fadd float %2453, %2454, !dbg !35 + %2687 = fadd float %2455, %2686, !dbg !35 + %2688 = fadd float %2456, %2687, !dbg !35 + %2689 = fadd float %2457, %2688, !dbg !35 + %2690 = fadd float %2458, %2689, !dbg !35 + %2691 = fadd float %2459, %2690, !dbg !35 + %2692 = fadd float %2460, %2691, !dbg !35 + %2693 = fadd float %2461, %2462, !dbg !35 + %2694 = fadd float %2463, %2693, !dbg !35 + %2695 = fadd float %2464, %2694, !dbg !35 + %2696 = fadd float %2465, %2695, !dbg !35 + %2697 = fadd float %2466, %2696, !dbg !35 + %2698 = fadd float %2467, %2697, !dbg !35 + %2699 = fadd float %2468, %2698, !dbg !35 + %2700 = fadd float %2469, %2470, !dbg !35 + %2701 = fadd float %2471, %2700, !dbg !35 + %2702 = fadd float %2472, %2701, !dbg !35 + %2703 = fadd float %2473, %2702, !dbg !35 + %2704 = fadd float %2474, %2703, !dbg !35 + %2705 = fadd float %2475, %2704, !dbg !35 + %2706 = fadd float %2476, %2705, !dbg !35 + %2707 = fadd float %2477, %2478, !dbg !35 + %2708 = fadd float %2479, %2707, !dbg !35 + %2709 = fadd float %2480, %2708, !dbg !35 + %2710 = fadd float %2481, %2709, !dbg !35 + %2711 = fadd float %2482, %2710, !dbg !35 + %2712 = fadd float %2483, %2711, !dbg !35 + %2713 = fadd float %2484, %2712, !dbg !35 + %2714 = fadd float %2485, %2486, !dbg !35 + %2715 = fadd float %2487, %2714, !dbg !35 + %2716 = fadd float %2488, %2715, !dbg !35 + %2717 = fadd float %2489, %2716, !dbg !35 + %2718 = fadd float %2490, %2717, !dbg !35 + %2719 = fadd float %2491, %2718, !dbg !35 + %2720 = fadd float %2492, %2719, !dbg !35 + %2721 = fadd float %2493, %2494, !dbg !35 + %2722 = fadd float %2495, %2721, !dbg !35 + %2723 = fadd float %2496, %2722, !dbg !35 + %2724 = fadd float %2497, %2723, !dbg !35 + %2725 = fadd float %2498, %2724, !dbg !35 + %2726 = fadd float %2499, %2725, !dbg !35 + %2727 = fadd float %2500, %2726, !dbg !35 + %2728 = fadd float %2501, %2502, !dbg !35 + %2729 = fadd float %2503, %2728, !dbg !35 + %2730 = fadd float %2504, %2729, !dbg !35 + %2731 = fadd float %2505, %2730, !dbg !35 + %2732 = fadd float %2506, %2731, !dbg !35 + %2733 = fadd float %2507, %2732, !dbg !35 + %2734 = fadd float %2508, %2733, !dbg !35 + %2735 = fadd float %2509, %2510, !dbg !35 + %2736 = fadd float %2511, %2735, !dbg !35 + %2737 = fadd float %2512, %2736, !dbg !35 + %2738 = fadd float %2513, %2737, !dbg !35 + %2739 = fadd float %2514, %2738, !dbg !35 + %2740 = fadd float %2515, %2739, !dbg !35 + %2741 = fadd float %2516, %2740, !dbg !35 + %2742 = fadd float %2517, %2518, !dbg !35 + %2743 = fadd float %2519, %2742, !dbg !35 + %2744 = fadd float %2520, %2743, !dbg !35 + %2745 = fadd float %2521, %2744, !dbg !35 + %2746 = fadd float %2522, %2745, !dbg !35 + %2747 = fadd float %2523, %2746, !dbg !35 + %2748 = fadd float %2524, %2747, !dbg !35 + %2749 = bitcast float %2531 to i32, !dbg !38 + %2750 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2749, i32 8, i32 31), !dbg !38 + %2751 = bitcast i32 %2750 to float, !dbg !38 + %2752 = fadd float %2531, %2751, !dbg !35 + %2753 = bitcast float %2752 to i32, !dbg !38 + %2754 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2753, i32 4, i32 31), !dbg !38 + %2755 = bitcast i32 %2754 to float, !dbg !38 + %2756 = fadd float %2752, %2755, !dbg !35 + %2757 = bitcast float %2756 to i32, !dbg !38 + %2758 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2757, i32 2, i32 31), !dbg !38 + %2759 = bitcast i32 %2758 to float, !dbg !38 + %2760 = fadd float %2756, %2759, !dbg !35 + %2761 = bitcast float %2760 to i32, !dbg !38 + %2762 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2761, i32 1, i32 31), !dbg !38 + %2763 = bitcast i32 %2762 to float, !dbg !38 + %2764 = fadd float %2760, %2763, !dbg !35 + %2765 = bitcast float %2538 to i32, !dbg !38 + %2766 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2765, i32 8, i32 31), !dbg !38 + %2767 = bitcast i32 %2766 to float, !dbg !38 + %2768 = fadd float %2538, %2767, !dbg !35 + %2769 = bitcast float %2768 to i32, !dbg !38 + %2770 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2769, i32 4, i32 31), !dbg !38 + %2771 = bitcast i32 %2770 to float, !dbg !38 + %2772 = fadd float %2768, %2771, !dbg !35 + %2773 = bitcast float %2772 to i32, !dbg !38 + %2774 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2773, i32 2, i32 31), !dbg !38 + %2775 = bitcast i32 %2774 to float, !dbg !38 + %2776 = fadd float %2772, %2775, !dbg !35 + %2777 = bitcast float %2776 to i32, !dbg !38 + %2778 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2777, i32 1, i32 31), !dbg !38 + %2779 = bitcast i32 %2778 to float, !dbg !38 + %2780 = fadd float %2776, %2779, !dbg !35 + %2781 = bitcast float %2545 to i32, !dbg !38 + %2782 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2781, i32 8, i32 31), !dbg !38 + %2783 = bitcast i32 %2782 to float, !dbg !38 + %2784 = fadd float %2545, %2783, !dbg !35 + %2785 = bitcast float %2784 to i32, !dbg !38 + %2786 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2785, i32 4, i32 31), !dbg !38 + %2787 = bitcast i32 %2786 to float, !dbg !38 + %2788 = fadd float %2784, %2787, !dbg !35 + %2789 = bitcast float %2788 to i32, !dbg !38 + %2790 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2789, i32 2, i32 31), !dbg !38 + %2791 = bitcast i32 %2790 to float, !dbg !38 + %2792 = fadd float %2788, %2791, !dbg !35 + %2793 = bitcast float %2792 to i32, !dbg !38 + %2794 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2793, i32 1, i32 31), !dbg !38 + %2795 = bitcast i32 %2794 to float, !dbg !38 + %2796 = fadd float %2792, %2795, !dbg !35 + %2797 = bitcast float %2552 to i32, !dbg !38 + %2798 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2797, i32 8, i32 31), !dbg !38 + %2799 = bitcast i32 %2798 to float, !dbg !38 + %2800 = fadd float %2552, %2799, !dbg !35 + %2801 = bitcast float %2800 to i32, !dbg !38 + %2802 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2801, i32 4, i32 31), !dbg !38 + %2803 = bitcast i32 %2802 to float, !dbg !38 + %2804 = fadd float %2800, %2803, !dbg !35 + %2805 = bitcast float %2804 to i32, !dbg !38 + %2806 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2805, i32 2, i32 31), !dbg !38 + %2807 = bitcast i32 %2806 to float, !dbg !38 + %2808 = fadd float %2804, %2807, !dbg !35 + %2809 = bitcast float %2808 to i32, !dbg !38 + %2810 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2809, i32 1, i32 31), !dbg !38 + %2811 = bitcast i32 %2810 to float, !dbg !38 + %2812 = fadd float %2808, %2811, !dbg !35 + %2813 = bitcast float %2559 to i32, !dbg !38 + %2814 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2813, i32 8, i32 31), !dbg !38 + %2815 = bitcast i32 %2814 to float, !dbg !38 + %2816 = fadd float %2559, %2815, !dbg !35 + %2817 = bitcast float %2816 to i32, !dbg !38 + %2818 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2817, i32 4, i32 31), !dbg !38 + %2819 = bitcast i32 %2818 to float, !dbg !38 + %2820 = fadd float %2816, %2819, !dbg !35 + %2821 = bitcast float %2820 to i32, !dbg !38 + %2822 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2821, i32 2, i32 31), !dbg !38 + %2823 = bitcast i32 %2822 to float, !dbg !38 + %2824 = fadd float %2820, %2823, !dbg !35 + %2825 = bitcast float %2824 to i32, !dbg !38 + %2826 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2825, i32 1, i32 31), !dbg !38 + %2827 = bitcast i32 %2826 to float, !dbg !38 + %2828 = fadd float %2824, %2827, !dbg !35 + %2829 = bitcast float %2566 to i32, !dbg !38 + %2830 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2829, i32 8, i32 31), !dbg !38 + %2831 = bitcast i32 %2830 to float, !dbg !38 + %2832 = fadd float %2566, %2831, !dbg !35 + %2833 = bitcast float %2832 to i32, !dbg !38 + %2834 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2833, i32 4, i32 31), !dbg !38 + %2835 = bitcast i32 %2834 to float, !dbg !38 + %2836 = fadd float %2832, %2835, !dbg !35 + %2837 = bitcast float %2836 to i32, !dbg !38 + %2838 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2837, i32 2, i32 31), !dbg !38 + %2839 = bitcast i32 %2838 to float, !dbg !38 + %2840 = fadd float %2836, %2839, !dbg !35 + %2841 = bitcast float %2840 to i32, !dbg !38 + %2842 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2841, i32 1, i32 31), !dbg !38 + %2843 = bitcast i32 %2842 to float, !dbg !38 + %2844 = fadd float %2840, %2843, !dbg !35 + %2845 = bitcast float %2573 to i32, !dbg !38 + %2846 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2845, i32 8, i32 31), !dbg !38 + %2847 = bitcast i32 %2846 to float, !dbg !38 + %2848 = fadd float %2573, %2847, !dbg !35 + %2849 = bitcast float %2848 to i32, !dbg !38 + %2850 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2849, i32 4, i32 31), !dbg !38 + %2851 = bitcast i32 %2850 to float, !dbg !38 + %2852 = fadd float %2848, %2851, !dbg !35 + %2853 = bitcast float %2852 to i32, !dbg !38 + %2854 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2853, i32 2, i32 31), !dbg !38 + %2855 = bitcast i32 %2854 to float, !dbg !38 + %2856 = fadd float %2852, %2855, !dbg !35 + %2857 = bitcast float %2856 to i32, !dbg !38 + %2858 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2857, i32 1, i32 31), !dbg !38 + %2859 = bitcast i32 %2858 to float, !dbg !38 + %2860 = fadd float %2856, %2859, !dbg !35 + %2861 = bitcast float %2580 to i32, !dbg !38 + %2862 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2861, i32 8, i32 31), !dbg !38 + %2863 = bitcast i32 %2862 to float, !dbg !38 + %2864 = fadd float %2580, %2863, !dbg !35 + %2865 = bitcast float %2864 to i32, !dbg !38 + %2866 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2865, i32 4, i32 31), !dbg !38 + %2867 = bitcast i32 %2866 to float, !dbg !38 + %2868 = fadd float %2864, %2867, !dbg !35 + %2869 = bitcast float %2868 to i32, !dbg !38 + %2870 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2869, i32 2, i32 31), !dbg !38 + %2871 = bitcast i32 %2870 to float, !dbg !38 + %2872 = fadd float %2868, %2871, !dbg !35 + %2873 = bitcast float %2872 to i32, !dbg !38 + %2874 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2873, i32 1, i32 31), !dbg !38 + %2875 = bitcast i32 %2874 to float, !dbg !38 + %2876 = fadd float %2872, %2875, !dbg !35 + %2877 = bitcast float %2587 to i32, !dbg !38 + %2878 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2877, i32 8, i32 31), !dbg !38 + %2879 = bitcast i32 %2878 to float, !dbg !38 + %2880 = fadd float %2587, %2879, !dbg !35 + %2881 = bitcast float %2880 to i32, !dbg !38 + %2882 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2881, i32 4, i32 31), !dbg !38 + %2883 = bitcast i32 %2882 to float, !dbg !38 + %2884 = fadd float %2880, %2883, !dbg !35 + %2885 = bitcast float %2884 to i32, !dbg !38 + %2886 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2885, i32 2, i32 31), !dbg !38 + %2887 = bitcast i32 %2886 to float, !dbg !38 + %2888 = fadd float %2884, %2887, !dbg !35 + %2889 = bitcast float %2888 to i32, !dbg !38 + %2890 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2889, i32 1, i32 31), !dbg !38 + %2891 = bitcast i32 %2890 to float, !dbg !38 + %2892 = fadd float %2888, %2891, !dbg !35 + %2893 = bitcast float %2594 to i32, !dbg !38 + %2894 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2893, i32 8, i32 31), !dbg !38 + %2895 = bitcast i32 %2894 to float, !dbg !38 + %2896 = fadd float %2594, %2895, !dbg !35 + %2897 = bitcast float %2896 to i32, !dbg !38 + %2898 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2897, i32 4, i32 31), !dbg !38 + %2899 = bitcast i32 %2898 to float, !dbg !38 + %2900 = fadd float %2896, %2899, !dbg !35 + %2901 = bitcast float %2900 to i32, !dbg !38 + %2902 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2901, i32 2, i32 31), !dbg !38 + %2903 = bitcast i32 %2902 to float, !dbg !38 + %2904 = fadd float %2900, %2903, !dbg !35 + %2905 = bitcast float %2904 to i32, !dbg !38 + %2906 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2905, i32 1, i32 31), !dbg !38 + %2907 = bitcast i32 %2906 to float, !dbg !38 + %2908 = fadd float %2904, %2907, !dbg !35 + %2909 = bitcast float %2601 to i32, !dbg !38 + %2910 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2909, i32 8, i32 31), !dbg !38 + %2911 = bitcast i32 %2910 to float, !dbg !38 + %2912 = fadd float %2601, %2911, !dbg !35 + %2913 = bitcast float %2912 to i32, !dbg !38 + %2914 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2913, i32 4, i32 31), !dbg !38 + %2915 = bitcast i32 %2914 to float, !dbg !38 + %2916 = fadd float %2912, %2915, !dbg !35 + %2917 = bitcast float %2916 to i32, !dbg !38 + %2918 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2917, i32 2, i32 31), !dbg !38 + %2919 = bitcast i32 %2918 to float, !dbg !38 + %2920 = fadd float %2916, %2919, !dbg !35 + %2921 = bitcast float %2920 to i32, !dbg !38 + %2922 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2921, i32 1, i32 31), !dbg !38 + %2923 = bitcast i32 %2922 to float, !dbg !38 + %2924 = fadd float %2920, %2923, !dbg !35 + %2925 = bitcast float %2608 to i32, !dbg !38 + %2926 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2925, i32 8, i32 31), !dbg !38 + %2927 = bitcast i32 %2926 to float, !dbg !38 + %2928 = fadd float %2608, %2927, !dbg !35 + %2929 = bitcast float %2928 to i32, !dbg !38 + %2930 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2929, i32 4, i32 31), !dbg !38 + %2931 = bitcast i32 %2930 to float, !dbg !38 + %2932 = fadd float %2928, %2931, !dbg !35 + %2933 = bitcast float %2932 to i32, !dbg !38 + %2934 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2933, i32 2, i32 31), !dbg !38 + %2935 = bitcast i32 %2934 to float, !dbg !38 + %2936 = fadd float %2932, %2935, !dbg !35 + %2937 = bitcast float %2936 to i32, !dbg !38 + %2938 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2937, i32 1, i32 31), !dbg !38 + %2939 = bitcast i32 %2938 to float, !dbg !38 + %2940 = fadd float %2936, %2939, !dbg !35 + %2941 = bitcast float %2615 to i32, !dbg !38 + %2942 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2941, i32 8, i32 31), !dbg !38 + %2943 = bitcast i32 %2942 to float, !dbg !38 + %2944 = fadd float %2615, %2943, !dbg !35 + %2945 = bitcast float %2944 to i32, !dbg !38 + %2946 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2945, i32 4, i32 31), !dbg !38 + %2947 = bitcast i32 %2946 to float, !dbg !38 + %2948 = fadd float %2944, %2947, !dbg !35 + %2949 = bitcast float %2948 to i32, !dbg !38 + %2950 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2949, i32 2, i32 31), !dbg !38 + %2951 = bitcast i32 %2950 to float, !dbg !38 + %2952 = fadd float %2948, %2951, !dbg !35 + %2953 = bitcast float %2952 to i32, !dbg !38 + %2954 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2953, i32 1, i32 31), !dbg !38 + %2955 = bitcast i32 %2954 to float, !dbg !38 + %2956 = fadd float %2952, %2955, !dbg !35 + %2957 = bitcast float %2622 to i32, !dbg !38 + %2958 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2957, i32 8, i32 31), !dbg !38 + %2959 = bitcast i32 %2958 to float, !dbg !38 + %2960 = fadd float %2622, %2959, !dbg !35 + %2961 = bitcast float %2960 to i32, !dbg !38 + %2962 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2961, i32 4, i32 31), !dbg !38 + %2963 = bitcast i32 %2962 to float, !dbg !38 + %2964 = fadd float %2960, %2963, !dbg !35 + %2965 = bitcast float %2964 to i32, !dbg !38 + %2966 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2965, i32 2, i32 31), !dbg !38 + %2967 = bitcast i32 %2966 to float, !dbg !38 + %2968 = fadd float %2964, %2967, !dbg !35 + %2969 = bitcast float %2968 to i32, !dbg !38 + %2970 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2969, i32 1, i32 31), !dbg !38 + %2971 = bitcast i32 %2970 to float, !dbg !38 + %2972 = fadd float %2968, %2971, !dbg !35 + %2973 = bitcast float %2629 to i32, !dbg !38 + %2974 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2973, i32 8, i32 31), !dbg !38 + %2975 = bitcast i32 %2974 to float, !dbg !38 + %2976 = fadd float %2629, %2975, !dbg !35 + %2977 = bitcast float %2976 to i32, !dbg !38 + %2978 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2977, i32 4, i32 31), !dbg !38 + %2979 = bitcast i32 %2978 to float, !dbg !38 + %2980 = fadd float %2976, %2979, !dbg !35 + %2981 = bitcast float %2980 to i32, !dbg !38 + %2982 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2981, i32 2, i32 31), !dbg !38 + %2983 = bitcast i32 %2982 to float, !dbg !38 + %2984 = fadd float %2980, %2983, !dbg !35 + %2985 = bitcast float %2984 to i32, !dbg !38 + %2986 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2985, i32 1, i32 31), !dbg !38 + %2987 = bitcast i32 %2986 to float, !dbg !38 + %2988 = fadd float %2984, %2987, !dbg !35 + %2989 = bitcast float %2636 to i32, !dbg !38 + %2990 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2989, i32 8, i32 31), !dbg !38 + %2991 = bitcast i32 %2990 to float, !dbg !38 + %2992 = fadd float %2636, %2991, !dbg !35 + %2993 = bitcast float %2992 to i32, !dbg !38 + %2994 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2993, i32 4, i32 31), !dbg !38 + %2995 = bitcast i32 %2994 to float, !dbg !38 + %2996 = fadd float %2992, %2995, !dbg !35 + %2997 = bitcast float %2996 to i32, !dbg !38 + %2998 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2997, i32 2, i32 31), !dbg !38 + %2999 = bitcast i32 %2998 to float, !dbg !38 + %3000 = fadd float %2996, %2999, !dbg !35 + %3001 = bitcast float %3000 to i32, !dbg !38 + %3002 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3001, i32 1, i32 31), !dbg !38 + %3003 = bitcast i32 %3002 to float, !dbg !38 + %3004 = fadd float %3000, %3003, !dbg !35 + %3005 = bitcast float %2643 to i32, !dbg !38 + %3006 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3005, i32 8, i32 31), !dbg !38 + %3007 = bitcast i32 %3006 to float, !dbg !38 + %3008 = fadd float %2643, %3007, !dbg !35 + %3009 = bitcast float %3008 to i32, !dbg !38 + %3010 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3009, i32 4, i32 31), !dbg !38 + %3011 = bitcast i32 %3010 to float, !dbg !38 + %3012 = fadd float %3008, %3011, !dbg !35 + %3013 = bitcast float %3012 to i32, !dbg !38 + %3014 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3013, i32 2, i32 31), !dbg !38 + %3015 = bitcast i32 %3014 to float, !dbg !38 + %3016 = fadd float %3012, %3015, !dbg !35 + %3017 = bitcast float %3016 to i32, !dbg !38 + %3018 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3017, i32 1, i32 31), !dbg !38 + %3019 = bitcast i32 %3018 to float, !dbg !38 + %3020 = fadd float %3016, %3019, !dbg !35 + %3021 = bitcast float %2650 to i32, !dbg !38 + %3022 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3021, i32 8, i32 31), !dbg !38 + %3023 = bitcast i32 %3022 to float, !dbg !38 + %3024 = fadd float %2650, %3023, !dbg !35 + %3025 = bitcast float %3024 to i32, !dbg !38 + %3026 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3025, i32 4, i32 31), !dbg !38 + %3027 = bitcast i32 %3026 to float, !dbg !38 + %3028 = fadd float %3024, %3027, !dbg !35 + %3029 = bitcast float %3028 to i32, !dbg !38 + %3030 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3029, i32 2, i32 31), !dbg !38 + %3031 = bitcast i32 %3030 to float, !dbg !38 + %3032 = fadd float %3028, %3031, !dbg !35 + %3033 = bitcast float %3032 to i32, !dbg !38 + %3034 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3033, i32 1, i32 31), !dbg !38 + %3035 = bitcast i32 %3034 to float, !dbg !38 + %3036 = fadd float %3032, %3035, !dbg !35 + %3037 = bitcast float %2657 to i32, !dbg !38 + %3038 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3037, i32 8, i32 31), !dbg !38 + %3039 = bitcast i32 %3038 to float, !dbg !38 + %3040 = fadd float %2657, %3039, !dbg !35 + %3041 = bitcast float %3040 to i32, !dbg !38 + %3042 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3041, i32 4, i32 31), !dbg !38 + %3043 = bitcast i32 %3042 to float, !dbg !38 + %3044 = fadd float %3040, %3043, !dbg !35 + %3045 = bitcast float %3044 to i32, !dbg !38 + %3046 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3045, i32 2, i32 31), !dbg !38 + %3047 = bitcast i32 %3046 to float, !dbg !38 + %3048 = fadd float %3044, %3047, !dbg !35 + %3049 = bitcast float %3048 to i32, !dbg !38 + %3050 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3049, i32 1, i32 31), !dbg !38 + %3051 = bitcast i32 %3050 to float, !dbg !38 + %3052 = fadd float %3048, %3051, !dbg !35 + %3053 = bitcast float %2664 to i32, !dbg !38 + %3054 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3053, i32 8, i32 31), !dbg !38 + %3055 = bitcast i32 %3054 to float, !dbg !38 + %3056 = fadd float %2664, %3055, !dbg !35 + %3057 = bitcast float %3056 to i32, !dbg !38 + %3058 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3057, i32 4, i32 31), !dbg !38 + %3059 = bitcast i32 %3058 to float, !dbg !38 + %3060 = fadd float %3056, %3059, !dbg !35 + %3061 = bitcast float %3060 to i32, !dbg !38 + %3062 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3061, i32 2, i32 31), !dbg !38 + %3063 = bitcast i32 %3062 to float, !dbg !38 + %3064 = fadd float %3060, %3063, !dbg !35 + %3065 = bitcast float %3064 to i32, !dbg !38 + %3066 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3065, i32 1, i32 31), !dbg !38 + %3067 = bitcast i32 %3066 to float, !dbg !38 + %3068 = fadd float %3064, %3067, !dbg !35 + %3069 = bitcast float %2671 to i32, !dbg !38 + %3070 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3069, i32 8, i32 31), !dbg !38 + %3071 = bitcast i32 %3070 to float, !dbg !38 + %3072 = fadd float %2671, %3071, !dbg !35 + %3073 = bitcast float %3072 to i32, !dbg !38 + %3074 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3073, i32 4, i32 31), !dbg !38 + %3075 = bitcast i32 %3074 to float, !dbg !38 + %3076 = fadd float %3072, %3075, !dbg !35 + %3077 = bitcast float %3076 to i32, !dbg !38 + %3078 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3077, i32 2, i32 31), !dbg !38 + %3079 = bitcast i32 %3078 to float, !dbg !38 + %3080 = fadd float %3076, %3079, !dbg !35 + %3081 = bitcast float %3080 to i32, !dbg !38 + %3082 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3081, i32 1, i32 31), !dbg !38 + %3083 = bitcast i32 %3082 to float, !dbg !38 + %3084 = fadd float %3080, %3083, !dbg !35 + %3085 = bitcast float %2678 to i32, !dbg !38 + %3086 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3085, i32 8, i32 31), !dbg !38 + %3087 = bitcast i32 %3086 to float, !dbg !38 + %3088 = fadd float %2678, %3087, !dbg !35 + %3089 = bitcast float %3088 to i32, !dbg !38 + %3090 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3089, i32 4, i32 31), !dbg !38 + %3091 = bitcast i32 %3090 to float, !dbg !38 + %3092 = fadd float %3088, %3091, !dbg !35 + %3093 = bitcast float %3092 to i32, !dbg !38 + %3094 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3093, i32 2, i32 31), !dbg !38 + %3095 = bitcast i32 %3094 to float, !dbg !38 + %3096 = fadd float %3092, %3095, !dbg !35 + %3097 = bitcast float %3096 to i32, !dbg !38 + %3098 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3097, i32 1, i32 31), !dbg !38 + %3099 = bitcast i32 %3098 to float, !dbg !38 + %3100 = fadd float %3096, %3099, !dbg !35 + %3101 = bitcast float %2685 to i32, !dbg !38 + %3102 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3101, i32 8, i32 31), !dbg !38 + %3103 = bitcast i32 %3102 to float, !dbg !38 + %3104 = fadd float %2685, %3103, !dbg !35 + %3105 = bitcast float %3104 to i32, !dbg !38 + %3106 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3105, i32 4, i32 31), !dbg !38 + %3107 = bitcast i32 %3106 to float, !dbg !38 + %3108 = fadd float %3104, %3107, !dbg !35 + %3109 = bitcast float %3108 to i32, !dbg !38 + %3110 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3109, i32 2, i32 31), !dbg !38 + %3111 = bitcast i32 %3110 to float, !dbg !38 + %3112 = fadd float %3108, %3111, !dbg !35 + %3113 = bitcast float %3112 to i32, !dbg !38 + %3114 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3113, i32 1, i32 31), !dbg !38 + %3115 = bitcast i32 %3114 to float, !dbg !38 + %3116 = fadd float %3112, %3115, !dbg !35 + %3117 = bitcast float %2692 to i32, !dbg !38 + %3118 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3117, i32 8, i32 31), !dbg !38 + %3119 = bitcast i32 %3118 to float, !dbg !38 + %3120 = fadd float %2692, %3119, !dbg !35 + %3121 = bitcast float %3120 to i32, !dbg !38 + %3122 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3121, i32 4, i32 31), !dbg !38 + %3123 = bitcast i32 %3122 to float, !dbg !38 + %3124 = fadd float %3120, %3123, !dbg !35 + %3125 = bitcast float %3124 to i32, !dbg !38 + %3126 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3125, i32 2, i32 31), !dbg !38 + %3127 = bitcast i32 %3126 to float, !dbg !38 + %3128 = fadd float %3124, %3127, !dbg !35 + %3129 = bitcast float %3128 to i32, !dbg !38 + %3130 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3129, i32 1, i32 31), !dbg !38 + %3131 = bitcast i32 %3130 to float, !dbg !38 + %3132 = fadd float %3128, %3131, !dbg !35 + %3133 = bitcast float %2699 to i32, !dbg !38 + %3134 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3133, i32 8, i32 31), !dbg !38 + %3135 = bitcast i32 %3134 to float, !dbg !38 + %3136 = fadd float %2699, %3135, !dbg !35 + %3137 = bitcast float %3136 to i32, !dbg !38 + %3138 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3137, i32 4, i32 31), !dbg !38 + %3139 = bitcast i32 %3138 to float, !dbg !38 + %3140 = fadd float %3136, %3139, !dbg !35 + %3141 = bitcast float %3140 to i32, !dbg !38 + %3142 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3141, i32 2, i32 31), !dbg !38 + %3143 = bitcast i32 %3142 to float, !dbg !38 + %3144 = fadd float %3140, %3143, !dbg !35 + %3145 = bitcast float %3144 to i32, !dbg !38 + %3146 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3145, i32 1, i32 31), !dbg !38 + %3147 = bitcast i32 %3146 to float, !dbg !38 + %3148 = fadd float %3144, %3147, !dbg !35 + %3149 = bitcast float %2706 to i32, !dbg !38 + %3150 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3149, i32 8, i32 31), !dbg !38 + %3151 = bitcast i32 %3150 to float, !dbg !38 + %3152 = fadd float %2706, %3151, !dbg !35 + %3153 = bitcast float %3152 to i32, !dbg !38 + %3154 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3153, i32 4, i32 31), !dbg !38 + %3155 = bitcast i32 %3154 to float, !dbg !38 + %3156 = fadd float %3152, %3155, !dbg !35 + %3157 = bitcast float %3156 to i32, !dbg !38 + %3158 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3157, i32 2, i32 31), !dbg !38 + %3159 = bitcast i32 %3158 to float, !dbg !38 + %3160 = fadd float %3156, %3159, !dbg !35 + %3161 = bitcast float %3160 to i32, !dbg !38 + %3162 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3161, i32 1, i32 31), !dbg !38 + %3163 = bitcast i32 %3162 to float, !dbg !38 + %3164 = fadd float %3160, %3163, !dbg !35 + %3165 = bitcast float %2713 to i32, !dbg !38 + %3166 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3165, i32 8, i32 31), !dbg !38 + %3167 = bitcast i32 %3166 to float, !dbg !38 + %3168 = fadd float %2713, %3167, !dbg !35 + %3169 = bitcast float %3168 to i32, !dbg !38 + %3170 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3169, i32 4, i32 31), !dbg !38 + %3171 = bitcast i32 %3170 to float, !dbg !38 + %3172 = fadd float %3168, %3171, !dbg !35 + %3173 = bitcast float %3172 to i32, !dbg !38 + %3174 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3173, i32 2, i32 31), !dbg !38 + %3175 = bitcast i32 %3174 to float, !dbg !38 + %3176 = fadd float %3172, %3175, !dbg !35 + %3177 = bitcast float %3176 to i32, !dbg !38 + %3178 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3177, i32 1, i32 31), !dbg !38 + %3179 = bitcast i32 %3178 to float, !dbg !38 + %3180 = fadd float %3176, %3179, !dbg !35 + %3181 = bitcast float %2720 to i32, !dbg !38 + %3182 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3181, i32 8, i32 31), !dbg !38 + %3183 = bitcast i32 %3182 to float, !dbg !38 + %3184 = fadd float %2720, %3183, !dbg !35 + %3185 = bitcast float %3184 to i32, !dbg !38 + %3186 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3185, i32 4, i32 31), !dbg !38 + %3187 = bitcast i32 %3186 to float, !dbg !38 + %3188 = fadd float %3184, %3187, !dbg !35 + %3189 = bitcast float %3188 to i32, !dbg !38 + %3190 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3189, i32 2, i32 31), !dbg !38 + %3191 = bitcast i32 %3190 to float, !dbg !38 + %3192 = fadd float %3188, %3191, !dbg !35 + %3193 = bitcast float %3192 to i32, !dbg !38 + %3194 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3193, i32 1, i32 31), !dbg !38 + %3195 = bitcast i32 %3194 to float, !dbg !38 + %3196 = fadd float %3192, %3195, !dbg !35 + %3197 = bitcast float %2727 to i32, !dbg !38 + %3198 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3197, i32 8, i32 31), !dbg !38 + %3199 = bitcast i32 %3198 to float, !dbg !38 + %3200 = fadd float %2727, %3199, !dbg !35 + %3201 = bitcast float %3200 to i32, !dbg !38 + %3202 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3201, i32 4, i32 31), !dbg !38 + %3203 = bitcast i32 %3202 to float, !dbg !38 + %3204 = fadd float %3200, %3203, !dbg !35 + %3205 = bitcast float %3204 to i32, !dbg !38 + %3206 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3205, i32 2, i32 31), !dbg !38 + %3207 = bitcast i32 %3206 to float, !dbg !38 + %3208 = fadd float %3204, %3207, !dbg !35 + %3209 = bitcast float %3208 to i32, !dbg !38 + %3210 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3209, i32 1, i32 31), !dbg !38 + %3211 = bitcast i32 %3210 to float, !dbg !38 + %3212 = fadd float %3208, %3211, !dbg !35 + %3213 = bitcast float %2734 to i32, !dbg !38 + %3214 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3213, i32 8, i32 31), !dbg !38 + %3215 = bitcast i32 %3214 to float, !dbg !38 + %3216 = fadd float %2734, %3215, !dbg !35 + %3217 = bitcast float %3216 to i32, !dbg !38 + %3218 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3217, i32 4, i32 31), !dbg !38 + %3219 = bitcast i32 %3218 to float, !dbg !38 + %3220 = fadd float %3216, %3219, !dbg !35 + %3221 = bitcast float %3220 to i32, !dbg !38 + %3222 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3221, i32 2, i32 31), !dbg !38 + %3223 = bitcast i32 %3222 to float, !dbg !38 + %3224 = fadd float %3220, %3223, !dbg !35 + %3225 = bitcast float %3224 to i32, !dbg !38 + %3226 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3225, i32 1, i32 31), !dbg !38 + %3227 = bitcast i32 %3226 to float, !dbg !38 + %3228 = fadd float %3224, %3227, !dbg !35 + %3229 = bitcast float %2741 to i32, !dbg !38 + %3230 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3229, i32 8, i32 31), !dbg !38 + %3231 = bitcast i32 %3230 to float, !dbg !38 + %3232 = fadd float %2741, %3231, !dbg !35 + %3233 = bitcast float %3232 to i32, !dbg !38 + %3234 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3233, i32 4, i32 31), !dbg !38 + %3235 = bitcast i32 %3234 to float, !dbg !38 + %3236 = fadd float %3232, %3235, !dbg !35 + %3237 = bitcast float %3236 to i32, !dbg !38 + %3238 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3237, i32 2, i32 31), !dbg !38 + %3239 = bitcast i32 %3238 to float, !dbg !38 + %3240 = fadd float %3236, %3239, !dbg !35 + %3241 = bitcast float %3240 to i32, !dbg !38 + %3242 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3241, i32 1, i32 31), !dbg !38 + %3243 = bitcast i32 %3242 to float, !dbg !38 + %3244 = fadd float %3240, %3243, !dbg !35 + %3245 = bitcast float %2748 to i32, !dbg !38 + %3246 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3245, i32 8, i32 31), !dbg !38 + %3247 = bitcast i32 %3246 to float, !dbg !38 + %3248 = fadd float %2748, %3247, !dbg !35 + %3249 = bitcast float %3248 to i32, !dbg !38 + %3250 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3249, i32 4, i32 31), !dbg !38 + %3251 = bitcast i32 %3250 to float, !dbg !38 + %3252 = fadd float %3248, %3251, !dbg !35 + %3253 = bitcast float %3252 to i32, !dbg !38 + %3254 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3253, i32 2, i32 31), !dbg !38 + %3255 = bitcast i32 %3254 to float, !dbg !38 + %3256 = fadd float %3252, %3255, !dbg !35 + %3257 = bitcast float %3256 to i32, !dbg !38 + %3258 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3257, i32 1, i32 31), !dbg !38 + %3259 = bitcast i32 %3258 to float, !dbg !38 + %3260 = fadd float %3256, %3259, !dbg !35 + %3261 = fmul float %2764, %886, !dbg !41 + %3262 = fmul float %2764, %887, !dbg !41 + %3263 = fmul float %2764, %888, !dbg !41 + %3264 = fmul float %2764, %889, !dbg !41 + %3265 = fmul float %2764, %890, !dbg !41 + %3266 = fmul float %2764, %891, !dbg !41 + %3267 = fmul float %2764, %892, !dbg !41 + %3268 = fmul float %2764, %893, !dbg !41 + %3269 = fmul float %2780, %894, !dbg !41 + %3270 = fmul float %2780, %895, !dbg !41 + %3271 = fmul float %2780, %896, !dbg !41 + %3272 = fmul float %2780, %897, !dbg !41 + %3273 = fmul float %2780, %898, !dbg !41 + %3274 = fmul float %2780, %899, !dbg !41 + %3275 = fmul float %2780, %900, !dbg !41 + %3276 = fmul float %2780, %901, !dbg !41 + %3277 = fmul float %2796, %902, !dbg !41 + %3278 = fmul float %2796, %903, !dbg !41 + %3279 = fmul float %2796, %904, !dbg !41 + %3280 = fmul float %2796, %905, !dbg !41 + %3281 = fmul float %2796, %906, !dbg !41 + %3282 = fmul float %2796, %907, !dbg !41 + %3283 = fmul float %2796, %908, !dbg !41 + %3284 = fmul float %2796, %909, !dbg !41 + %3285 = fmul float %2812, %910, !dbg !41 + %3286 = fmul float %2812, %911, !dbg !41 + %3287 = fmul float %2812, %912, !dbg !41 + %3288 = fmul float %2812, %913, !dbg !41 + %3289 = fmul float %2812, %914, !dbg !41 + %3290 = fmul float %2812, %915, !dbg !41 + %3291 = fmul float %2812, %916, !dbg !41 + %3292 = fmul float %2812, %917, !dbg !41 + %3293 = fmul float %2828, %918, !dbg !41 + %3294 = fmul float %2828, %919, !dbg !41 + %3295 = fmul float %2828, %920, !dbg !41 + %3296 = fmul float %2828, %921, !dbg !41 + %3297 = fmul float %2828, %922, !dbg !41 + %3298 = fmul float %2828, %923, !dbg !41 + %3299 = fmul float %2828, %924, !dbg !41 + %3300 = fmul float %2828, %925, !dbg !41 + %3301 = fmul float %2844, %926, !dbg !41 + %3302 = fmul float %2844, %927, !dbg !41 + %3303 = fmul float %2844, %928, !dbg !41 + %3304 = fmul float %2844, %929, !dbg !41 + %3305 = fmul float %2844, %930, !dbg !41 + %3306 = fmul float %2844, %931, !dbg !41 + %3307 = fmul float %2844, %932, !dbg !41 + %3308 = fmul float %2844, %933, !dbg !41 + %3309 = fmul float %2860, %934, !dbg !41 + %3310 = fmul float %2860, %935, !dbg !41 + %3311 = fmul float %2860, %936, !dbg !41 + %3312 = fmul float %2860, %937, !dbg !41 + %3313 = fmul float %2860, %938, !dbg !41 + %3314 = fmul float %2860, %939, !dbg !41 + %3315 = fmul float %2860, %940, !dbg !41 + %3316 = fmul float %2860, %941, !dbg !41 + %3317 = fmul float %2876, %942, !dbg !41 + %3318 = fmul float %2876, %943, !dbg !41 + %3319 = fmul float %2876, %944, !dbg !41 + %3320 = fmul float %2876, %945, !dbg !41 + %3321 = fmul float %2876, %946, !dbg !41 + %3322 = fmul float %2876, %947, !dbg !41 + %3323 = fmul float %2876, %948, !dbg !41 + %3324 = fmul float %2876, %949, !dbg !41 + %3325 = fmul float %2892, %950, !dbg !41 + %3326 = fmul float %2892, %951, !dbg !41 + %3327 = fmul float %2892, %952, !dbg !41 + %3328 = fmul float %2892, %953, !dbg !41 + %3329 = fmul float %2892, %954, !dbg !41 + %3330 = fmul float %2892, %955, !dbg !41 + %3331 = fmul float %2892, %956, !dbg !41 + %3332 = fmul float %2892, %957, !dbg !41 + %3333 = fmul float %2908, %958, !dbg !41 + %3334 = fmul float %2908, %959, !dbg !41 + %3335 = fmul float %2908, %960, !dbg !41 + %3336 = fmul float %2908, %961, !dbg !41 + %3337 = fmul float %2908, %962, !dbg !41 + %3338 = fmul float %2908, %963, !dbg !41 + %3339 = fmul float %2908, %964, !dbg !41 + %3340 = fmul float %2908, %965, !dbg !41 + %3341 = fmul float %2924, %966, !dbg !41 + %3342 = fmul float %2924, %967, !dbg !41 + %3343 = fmul float %2924, %968, !dbg !41 + %3344 = fmul float %2924, %969, !dbg !41 + %3345 = fmul float %2924, %970, !dbg !41 + %3346 = fmul float %2924, %971, !dbg !41 + %3347 = fmul float %2924, %972, !dbg !41 + %3348 = fmul float %2924, %973, !dbg !41 + %3349 = fmul float %2940, %974, !dbg !41 + %3350 = fmul float %2940, %975, !dbg !41 + %3351 = fmul float %2940, %976, !dbg !41 + %3352 = fmul float %2940, %977, !dbg !41 + %3353 = fmul float %2940, %978, !dbg !41 + %3354 = fmul float %2940, %979, !dbg !41 + %3355 = fmul float %2940, %980, !dbg !41 + %3356 = fmul float %2940, %981, !dbg !41 + %3357 = fmul float %2956, %982, !dbg !41 + %3358 = fmul float %2956, %983, !dbg !41 + %3359 = fmul float %2956, %984, !dbg !41 + %3360 = fmul float %2956, %985, !dbg !41 + %3361 = fmul float %2956, %986, !dbg !41 + %3362 = fmul float %2956, %987, !dbg !41 + %3363 = fmul float %2956, %988, !dbg !41 + %3364 = fmul float %2956, %989, !dbg !41 + %3365 = fmul float %2972, %990, !dbg !41 + %3366 = fmul float %2972, %991, !dbg !41 + %3367 = fmul float %2972, %992, !dbg !41 + %3368 = fmul float %2972, %993, !dbg !41 + %3369 = fmul float %2972, %994, !dbg !41 + %3370 = fmul float %2972, %995, !dbg !41 + %3371 = fmul float %2972, %996, !dbg !41 + %3372 = fmul float %2972, %997, !dbg !41 + %3373 = fmul float %2988, %998, !dbg !41 + %3374 = fmul float %2988, %999, !dbg !41 + %3375 = fmul float %2988, %1000, !dbg !41 + %3376 = fmul float %2988, %1001, !dbg !41 + %3377 = fmul float %2988, %1002, !dbg !41 + %3378 = fmul float %2988, %1003, !dbg !41 + %3379 = fmul float %2988, %1004, !dbg !41 + %3380 = fmul float %2988, %1005, !dbg !41 + %3381 = fmul float %3004, %1006, !dbg !41 + %3382 = fmul float %3004, %1007, !dbg !41 + %3383 = fmul float %3004, %1008, !dbg !41 + %3384 = fmul float %3004, %1009, !dbg !41 + %3385 = fmul float %3004, %1010, !dbg !41 + %3386 = fmul float %3004, %1011, !dbg !41 + %3387 = fmul float %3004, %1012, !dbg !41 + %3388 = fmul float %3004, %1013, !dbg !41 + %3389 = fmul float %3020, %1014, !dbg !41 + %3390 = fmul float %3020, %1015, !dbg !41 + %3391 = fmul float %3020, %1016, !dbg !41 + %3392 = fmul float %3020, %1017, !dbg !41 + %3393 = fmul float %3020, %1018, !dbg !41 + %3394 = fmul float %3020, %1019, !dbg !41 + %3395 = fmul float %3020, %1020, !dbg !41 + %3396 = fmul float %3020, %1021, !dbg !41 + %3397 = fmul float %3036, %1022, !dbg !41 + %3398 = fmul float %3036, %1023, !dbg !41 + %3399 = fmul float %3036, %1024, !dbg !41 + %3400 = fmul float %3036, %1025, !dbg !41 + %3401 = fmul float %3036, %1026, !dbg !41 + %3402 = fmul float %3036, %1027, !dbg !41 + %3403 = fmul float %3036, %1028, !dbg !41 + %3404 = fmul float %3036, %1029, !dbg !41 + %3405 = fmul float %3052, %1030, !dbg !41 + %3406 = fmul float %3052, %1031, !dbg !41 + %3407 = fmul float %3052, %1032, !dbg !41 + %3408 = fmul float %3052, %1033, !dbg !41 + %3409 = fmul float %3052, %1034, !dbg !41 + %3410 = fmul float %3052, %1035, !dbg !41 + %3411 = fmul float %3052, %1036, !dbg !41 + %3412 = fmul float %3052, %1037, !dbg !41 + %3413 = fmul float %3068, %1038, !dbg !41 + %3414 = fmul float %3068, %1039, !dbg !41 + %3415 = fmul float %3068, %1040, !dbg !41 + %3416 = fmul float %3068, %1041, !dbg !41 + %3417 = fmul float %3068, %1042, !dbg !41 + %3418 = fmul float %3068, %1043, !dbg !41 + %3419 = fmul float %3068, %1044, !dbg !41 + %3420 = fmul float %3068, %1045, !dbg !41 + %3421 = fmul float %3084, %1046, !dbg !41 + %3422 = fmul float %3084, %1047, !dbg !41 + %3423 = fmul float %3084, %1048, !dbg !41 + %3424 = fmul float %3084, %1049, !dbg !41 + %3425 = fmul float %3084, %1050, !dbg !41 + %3426 = fmul float %3084, %1051, !dbg !41 + %3427 = fmul float %3084, %1052, !dbg !41 + %3428 = fmul float %3084, %1053, !dbg !41 + %3429 = fmul float %3100, %1054, !dbg !41 + %3430 = fmul float %3100, %1055, !dbg !41 + %3431 = fmul float %3100, %1056, !dbg !41 + %3432 = fmul float %3100, %1057, !dbg !41 + %3433 = fmul float %3100, %1058, !dbg !41 + %3434 = fmul float %3100, %1059, !dbg !41 + %3435 = fmul float %3100, %1060, !dbg !41 + %3436 = fmul float %3100, %1061, !dbg !41 + %3437 = fmul float %3116, %1062, !dbg !41 + %3438 = fmul float %3116, %1063, !dbg !41 + %3439 = fmul float %3116, %1064, !dbg !41 + %3440 = fmul float %3116, %1065, !dbg !41 + %3441 = fmul float %3116, %1066, !dbg !41 + %3442 = fmul float %3116, %1067, !dbg !41 + %3443 = fmul float %3116, %1068, !dbg !41 + %3444 = fmul float %3116, %1069, !dbg !41 + %3445 = fmul float %3132, %1070, !dbg !41 + %3446 = fmul float %3132, %1071, !dbg !41 + %3447 = fmul float %3132, %1072, !dbg !41 + %3448 = fmul float %3132, %1073, !dbg !41 + %3449 = fmul float %3132, %1074, !dbg !41 + %3450 = fmul float %3132, %1075, !dbg !41 + %3451 = fmul float %3132, %1076, !dbg !41 + %3452 = fmul float %3132, %1077, !dbg !41 + %3453 = fmul float %3148, %1078, !dbg !41 + %3454 = fmul float %3148, %1079, !dbg !41 + %3455 = fmul float %3148, %1080, !dbg !41 + %3456 = fmul float %3148, %1081, !dbg !41 + %3457 = fmul float %3148, %1082, !dbg !41 + %3458 = fmul float %3148, %1083, !dbg !41 + %3459 = fmul float %3148, %1084, !dbg !41 + %3460 = fmul float %3148, %1085, !dbg !41 + %3461 = fmul float %3164, %1086, !dbg !41 + %3462 = fmul float %3164, %1087, !dbg !41 + %3463 = fmul float %3164, %1088, !dbg !41 + %3464 = fmul float %3164, %1089, !dbg !41 + %3465 = fmul float %3164, %1090, !dbg !41 + %3466 = fmul float %3164, %1091, !dbg !41 + %3467 = fmul float %3164, %1092, !dbg !41 + %3468 = fmul float %3164, %1093, !dbg !41 + %3469 = fmul float %3180, %1094, !dbg !41 + %3470 = fmul float %3180, %1095, !dbg !41 + %3471 = fmul float %3180, %1096, !dbg !41 + %3472 = fmul float %3180, %1097, !dbg !41 + %3473 = fmul float %3180, %1098, !dbg !41 + %3474 = fmul float %3180, %1099, !dbg !41 + %3475 = fmul float %3180, %1100, !dbg !41 + %3476 = fmul float %3180, %1101, !dbg !41 + %3477 = fmul float %3196, %1102, !dbg !41 + %3478 = fmul float %3196, %1103, !dbg !41 + %3479 = fmul float %3196, %1104, !dbg !41 + %3480 = fmul float %3196, %1105, !dbg !41 + %3481 = fmul float %3196, %1106, !dbg !41 + %3482 = fmul float %3196, %1107, !dbg !41 + %3483 = fmul float %3196, %1108, !dbg !41 + %3484 = fmul float %3196, %1109, !dbg !41 + %3485 = fmul float %3212, %1110, !dbg !41 + %3486 = fmul float %3212, %1111, !dbg !41 + %3487 = fmul float %3212, %1112, !dbg !41 + %3488 = fmul float %3212, %1113, !dbg !41 + %3489 = fmul float %3212, %1114, !dbg !41 + %3490 = fmul float %3212, %1115, !dbg !41 + %3491 = fmul float %3212, %1116, !dbg !41 + %3492 = fmul float %3212, %1117, !dbg !41 + %3493 = fmul float %3228, %1118, !dbg !41 + %3494 = fmul float %3228, %1119, !dbg !41 + %3495 = fmul float %3228, %1120, !dbg !41 + %3496 = fmul float %3228, %1121, !dbg !41 + %3497 = fmul float %3228, %1122, !dbg !41 + %3498 = fmul float %3228, %1123, !dbg !41 + %3499 = fmul float %3228, %1124, !dbg !41 + %3500 = fmul float %3228, %1125, !dbg !41 + %3501 = fmul float %3244, %1126, !dbg !41 + %3502 = fmul float %3244, %1127, !dbg !41 + %3503 = fmul float %3244, %1128, !dbg !41 + %3504 = fmul float %3244, %1129, !dbg !41 + %3505 = fmul float %3244, %1130, !dbg !41 + %3506 = fmul float %3244, %1131, !dbg !41 + %3507 = fmul float %3244, %1132, !dbg !41 + %3508 = fmul float %3244, %1133, !dbg !41 + %3509 = fmul float %3260, %1134, !dbg !41 + %3510 = fmul float %3260, %1135, !dbg !41 + %3511 = fmul float %3260, %1136, !dbg !41 + %3512 = fmul float %3260, %1137, !dbg !41 + %3513 = fmul float %3260, %1138, !dbg !41 + %3514 = fmul float %3260, %1139, !dbg !41 + %3515 = fmul float %3260, %1140, !dbg !41 + %3516 = fmul float %3260, %1141, !dbg !41 + %3517 = fmul float %1950, %3261, !dbg !42 + %3518 = fmul float %1950, %3262, !dbg !42 + %3519 = fmul float %1950, %3263, !dbg !42 + %3520 = fmul float %1950, %3264, !dbg !42 + %3521 = fmul float %1950, %3265, !dbg !42 + %3522 = fmul float %1950, %3266, !dbg !42 + %3523 = fmul float %1950, %3267, !dbg !42 + %3524 = fmul float %1950, %3268, !dbg !42 + %3525 = fmul float %1952, %3269, !dbg !42 + %3526 = fmul float %1952, %3270, !dbg !42 + %3527 = fmul float %1952, %3271, !dbg !42 + %3528 = fmul float %1952, %3272, !dbg !42 + %3529 = fmul float %1952, %3273, !dbg !42 + %3530 = fmul float %1952, %3274, !dbg !42 + %3531 = fmul float %1952, %3275, !dbg !42 + %3532 = fmul float %1952, %3276, !dbg !42 + %3533 = fmul float %1954, %3277, !dbg !42 + %3534 = fmul float %1954, %3278, !dbg !42 + %3535 = fmul float %1954, %3279, !dbg !42 + %3536 = fmul float %1954, %3280, !dbg !42 + %3537 = fmul float %1954, %3281, !dbg !42 + %3538 = fmul float %1954, %3282, !dbg !42 + %3539 = fmul float %1954, %3283, !dbg !42 + %3540 = fmul float %1954, %3284, !dbg !42 + %3541 = fmul float %1956, %3285, !dbg !42 + %3542 = fmul float %1956, %3286, !dbg !42 + %3543 = fmul float %1956, %3287, !dbg !42 + %3544 = fmul float %1956, %3288, !dbg !42 + %3545 = fmul float %1956, %3289, !dbg !42 + %3546 = fmul float %1956, %3290, !dbg !42 + %3547 = fmul float %1956, %3291, !dbg !42 + %3548 = fmul float %1956, %3292, !dbg !42 + %3549 = fmul float %1958, %3293, !dbg !42 + %3550 = fmul float %1958, %3294, !dbg !42 + %3551 = fmul float %1958, %3295, !dbg !42 + %3552 = fmul float %1958, %3296, !dbg !42 + %3553 = fmul float %1958, %3297, !dbg !42 + %3554 = fmul float %1958, %3298, !dbg !42 + %3555 = fmul float %1958, %3299, !dbg !42 + %3556 = fmul float %1958, %3300, !dbg !42 + %3557 = fmul float %1960, %3301, !dbg !42 + %3558 = fmul float %1960, %3302, !dbg !42 + %3559 = fmul float %1960, %3303, !dbg !42 + %3560 = fmul float %1960, %3304, !dbg !42 + %3561 = fmul float %1960, %3305, !dbg !42 + %3562 = fmul float %1960, %3306, !dbg !42 + %3563 = fmul float %1960, %3307, !dbg !42 + %3564 = fmul float %1960, %3308, !dbg !42 + %3565 = fmul float %1962, %3309, !dbg !42 + %3566 = fmul float %1962, %3310, !dbg !42 + %3567 = fmul float %1962, %3311, !dbg !42 + %3568 = fmul float %1962, %3312, !dbg !42 + %3569 = fmul float %1962, %3313, !dbg !42 + %3570 = fmul float %1962, %3314, !dbg !42 + %3571 = fmul float %1962, %3315, !dbg !42 + %3572 = fmul float %1962, %3316, !dbg !42 + %3573 = fmul float %1964, %3317, !dbg !42 + %3574 = fmul float %1964, %3318, !dbg !42 + %3575 = fmul float %1964, %3319, !dbg !42 + %3576 = fmul float %1964, %3320, !dbg !42 + %3577 = fmul float %1964, %3321, !dbg !42 + %3578 = fmul float %1964, %3322, !dbg !42 + %3579 = fmul float %1964, %3323, !dbg !42 + %3580 = fmul float %1964, %3324, !dbg !42 + %3581 = fmul float %1966, %3325, !dbg !42 + %3582 = fmul float %1966, %3326, !dbg !42 + %3583 = fmul float %1966, %3327, !dbg !42 + %3584 = fmul float %1966, %3328, !dbg !42 + %3585 = fmul float %1966, %3329, !dbg !42 + %3586 = fmul float %1966, %3330, !dbg !42 + %3587 = fmul float %1966, %3331, !dbg !42 + %3588 = fmul float %1966, %3332, !dbg !42 + %3589 = fmul float %1968, %3333, !dbg !42 + %3590 = fmul float %1968, %3334, !dbg !42 + %3591 = fmul float %1968, %3335, !dbg !42 + %3592 = fmul float %1968, %3336, !dbg !42 + %3593 = fmul float %1968, %3337, !dbg !42 + %3594 = fmul float %1968, %3338, !dbg !42 + %3595 = fmul float %1968, %3339, !dbg !42 + %3596 = fmul float %1968, %3340, !dbg !42 + %3597 = fmul float %1970, %3341, !dbg !42 + %3598 = fmul float %1970, %3342, !dbg !42 + %3599 = fmul float %1970, %3343, !dbg !42 + %3600 = fmul float %1970, %3344, !dbg !42 + %3601 = fmul float %1970, %3345, !dbg !42 + %3602 = fmul float %1970, %3346, !dbg !42 + %3603 = fmul float %1970, %3347, !dbg !42 + %3604 = fmul float %1970, %3348, !dbg !42 + %3605 = fmul float %1972, %3349, !dbg !42 + %3606 = fmul float %1972, %3350, !dbg !42 + %3607 = fmul float %1972, %3351, !dbg !42 + %3608 = fmul float %1972, %3352, !dbg !42 + %3609 = fmul float %1972, %3353, !dbg !42 + %3610 = fmul float %1972, %3354, !dbg !42 + %3611 = fmul float %1972, %3355, !dbg !42 + %3612 = fmul float %1972, %3356, !dbg !42 + %3613 = fmul float %1974, %3357, !dbg !42 + %3614 = fmul float %1974, %3358, !dbg !42 + %3615 = fmul float %1974, %3359, !dbg !42 + %3616 = fmul float %1974, %3360, !dbg !42 + %3617 = fmul float %1974, %3361, !dbg !42 + %3618 = fmul float %1974, %3362, !dbg !42 + %3619 = fmul float %1974, %3363, !dbg !42 + %3620 = fmul float %1974, %3364, !dbg !42 + %3621 = fmul float %1976, %3365, !dbg !42 + %3622 = fmul float %1976, %3366, !dbg !42 + %3623 = fmul float %1976, %3367, !dbg !42 + %3624 = fmul float %1976, %3368, !dbg !42 + %3625 = fmul float %1976, %3369, !dbg !42 + %3626 = fmul float %1976, %3370, !dbg !42 + %3627 = fmul float %1976, %3371, !dbg !42 + %3628 = fmul float %1976, %3372, !dbg !42 + %3629 = fmul float %1978, %3373, !dbg !42 + %3630 = fmul float %1978, %3374, !dbg !42 + %3631 = fmul float %1978, %3375, !dbg !42 + %3632 = fmul float %1978, %3376, !dbg !42 + %3633 = fmul float %1978, %3377, !dbg !42 + %3634 = fmul float %1978, %3378, !dbg !42 + %3635 = fmul float %1978, %3379, !dbg !42 + %3636 = fmul float %1978, %3380, !dbg !42 + %3637 = fmul float %1980, %3381, !dbg !42 + %3638 = fmul float %1980, %3382, !dbg !42 + %3639 = fmul float %1980, %3383, !dbg !42 + %3640 = fmul float %1980, %3384, !dbg !42 + %3641 = fmul float %1980, %3385, !dbg !42 + %3642 = fmul float %1980, %3386, !dbg !42 + %3643 = fmul float %1980, %3387, !dbg !42 + %3644 = fmul float %1980, %3388, !dbg !42 + %3645 = fmul float %1982, %3389, !dbg !42 + %3646 = fmul float %1982, %3390, !dbg !42 + %3647 = fmul float %1982, %3391, !dbg !42 + %3648 = fmul float %1982, %3392, !dbg !42 + %3649 = fmul float %1982, %3393, !dbg !42 + %3650 = fmul float %1982, %3394, !dbg !42 + %3651 = fmul float %1982, %3395, !dbg !42 + %3652 = fmul float %1982, %3396, !dbg !42 + %3653 = fmul float %1984, %3397, !dbg !42 + %3654 = fmul float %1984, %3398, !dbg !42 + %3655 = fmul float %1984, %3399, !dbg !42 + %3656 = fmul float %1984, %3400, !dbg !42 + %3657 = fmul float %1984, %3401, !dbg !42 + %3658 = fmul float %1984, %3402, !dbg !42 + %3659 = fmul float %1984, %3403, !dbg !42 + %3660 = fmul float %1984, %3404, !dbg !42 + %3661 = fmul float %1986, %3405, !dbg !42 + %3662 = fmul float %1986, %3406, !dbg !42 + %3663 = fmul float %1986, %3407, !dbg !42 + %3664 = fmul float %1986, %3408, !dbg !42 + %3665 = fmul float %1986, %3409, !dbg !42 + %3666 = fmul float %1986, %3410, !dbg !42 + %3667 = fmul float %1986, %3411, !dbg !42 + %3668 = fmul float %1986, %3412, !dbg !42 + %3669 = fmul float %1988, %3413, !dbg !42 + %3670 = fmul float %1988, %3414, !dbg !42 + %3671 = fmul float %1988, %3415, !dbg !42 + %3672 = fmul float %1988, %3416, !dbg !42 + %3673 = fmul float %1988, %3417, !dbg !42 + %3674 = fmul float %1988, %3418, !dbg !42 + %3675 = fmul float %1988, %3419, !dbg !42 + %3676 = fmul float %1988, %3420, !dbg !42 + %3677 = fmul float %1990, %3421, !dbg !42 + %3678 = fmul float %1990, %3422, !dbg !42 + %3679 = fmul float %1990, %3423, !dbg !42 + %3680 = fmul float %1990, %3424, !dbg !42 + %3681 = fmul float %1990, %3425, !dbg !42 + %3682 = fmul float %1990, %3426, !dbg !42 + %3683 = fmul float %1990, %3427, !dbg !42 + %3684 = fmul float %1990, %3428, !dbg !42 + %3685 = fmul float %1992, %3429, !dbg !42 + %3686 = fmul float %1992, %3430, !dbg !42 + %3687 = fmul float %1992, %3431, !dbg !42 + %3688 = fmul float %1992, %3432, !dbg !42 + %3689 = fmul float %1992, %3433, !dbg !42 + %3690 = fmul float %1992, %3434, !dbg !42 + %3691 = fmul float %1992, %3435, !dbg !42 + %3692 = fmul float %1992, %3436, !dbg !42 + %3693 = fmul float %1994, %3437, !dbg !42 + %3694 = fmul float %1994, %3438, !dbg !42 + %3695 = fmul float %1994, %3439, !dbg !42 + %3696 = fmul float %1994, %3440, !dbg !42 + %3697 = fmul float %1994, %3441, !dbg !42 + %3698 = fmul float %1994, %3442, !dbg !42 + %3699 = fmul float %1994, %3443, !dbg !42 + %3700 = fmul float %1994, %3444, !dbg !42 + %3701 = fmul float %1996, %3445, !dbg !42 + %3702 = fmul float %1996, %3446, !dbg !42 + %3703 = fmul float %1996, %3447, !dbg !42 + %3704 = fmul float %1996, %3448, !dbg !42 + %3705 = fmul float %1996, %3449, !dbg !42 + %3706 = fmul float %1996, %3450, !dbg !42 + %3707 = fmul float %1996, %3451, !dbg !42 + %3708 = fmul float %1996, %3452, !dbg !42 + %3709 = fmul float %1998, %3453, !dbg !42 + %3710 = fmul float %1998, %3454, !dbg !42 + %3711 = fmul float %1998, %3455, !dbg !42 + %3712 = fmul float %1998, %3456, !dbg !42 + %3713 = fmul float %1998, %3457, !dbg !42 + %3714 = fmul float %1998, %3458, !dbg !42 + %3715 = fmul float %1998, %3459, !dbg !42 + %3716 = fmul float %1998, %3460, !dbg !42 + %3717 = fmul float %2000, %3461, !dbg !42 + %3718 = fmul float %2000, %3462, !dbg !42 + %3719 = fmul float %2000, %3463, !dbg !42 + %3720 = fmul float %2000, %3464, !dbg !42 + %3721 = fmul float %2000, %3465, !dbg !42 + %3722 = fmul float %2000, %3466, !dbg !42 + %3723 = fmul float %2000, %3467, !dbg !42 + %3724 = fmul float %2000, %3468, !dbg !42 + %3725 = fmul float %2002, %3469, !dbg !42 + %3726 = fmul float %2002, %3470, !dbg !42 + %3727 = fmul float %2002, %3471, !dbg !42 + %3728 = fmul float %2002, %3472, !dbg !42 + %3729 = fmul float %2002, %3473, !dbg !42 + %3730 = fmul float %2002, %3474, !dbg !42 + %3731 = fmul float %2002, %3475, !dbg !42 + %3732 = fmul float %2002, %3476, !dbg !42 + %3733 = fmul float %2004, %3477, !dbg !42 + %3734 = fmul float %2004, %3478, !dbg !42 + %3735 = fmul float %2004, %3479, !dbg !42 + %3736 = fmul float %2004, %3480, !dbg !42 + %3737 = fmul float %2004, %3481, !dbg !42 + %3738 = fmul float %2004, %3482, !dbg !42 + %3739 = fmul float %2004, %3483, !dbg !42 + %3740 = fmul float %2004, %3484, !dbg !42 + %3741 = fmul float %2006, %3485, !dbg !42 + %3742 = fmul float %2006, %3486, !dbg !42 + %3743 = fmul float %2006, %3487, !dbg !42 + %3744 = fmul float %2006, %3488, !dbg !42 + %3745 = fmul float %2006, %3489, !dbg !42 + %3746 = fmul float %2006, %3490, !dbg !42 + %3747 = fmul float %2006, %3491, !dbg !42 + %3748 = fmul float %2006, %3492, !dbg !42 + %3749 = fmul float %2008, %3493, !dbg !42 + %3750 = fmul float %2008, %3494, !dbg !42 + %3751 = fmul float %2008, %3495, !dbg !42 + %3752 = fmul float %2008, %3496, !dbg !42 + %3753 = fmul float %2008, %3497, !dbg !42 + %3754 = fmul float %2008, %3498, !dbg !42 + %3755 = fmul float %2008, %3499, !dbg !42 + %3756 = fmul float %2008, %3500, !dbg !42 + %3757 = fmul float %2010, %3501, !dbg !42 + %3758 = fmul float %2010, %3502, !dbg !42 + %3759 = fmul float %2010, %3503, !dbg !42 + %3760 = fmul float %2010, %3504, !dbg !42 + %3761 = fmul float %2010, %3505, !dbg !42 + %3762 = fmul float %2010, %3506, !dbg !42 + %3763 = fmul float %2010, %3507, !dbg !42 + %3764 = fmul float %2010, %3508, !dbg !42 + %3765 = fmul float %2012, %3509, !dbg !42 + %3766 = fmul float %2012, %3510, !dbg !42 + %3767 = fmul float %2012, %3511, !dbg !42 + %3768 = fmul float %2012, %3512, !dbg !42 + %3769 = fmul float %2012, %3513, !dbg !42 + %3770 = fmul float %2012, %3514, !dbg !42 + %3771 = fmul float %2012, %3515, !dbg !42 + %3772 = fmul float %2012, %3516, !dbg !42 + %3773 = fsub float %2013, %3517, !dbg !43 + %3774 = fsub float %2014, %3518, !dbg !43 + %3775 = fsub float %2015, %3519, !dbg !43 + %3776 = fsub float %2016, %3520, !dbg !43 + %3777 = fsub float %2017, %3521, !dbg !43 + %3778 = fsub float %2018, %3522, !dbg !43 + %3779 = fsub float %2019, %3523, !dbg !43 + %3780 = fsub float %2020, %3524, !dbg !43 + %3781 = fsub float %2021, %3525, !dbg !43 + %3782 = fsub float %2022, %3526, !dbg !43 + %3783 = fsub float %2023, %3527, !dbg !43 + %3784 = fsub float %2024, %3528, !dbg !43 + %3785 = fsub float %2025, %3529, !dbg !43 + %3786 = fsub float %2026, %3530, !dbg !43 + %3787 = fsub float %2027, %3531, !dbg !43 + %3788 = fsub float %2028, %3532, !dbg !43 + %3789 = insertelement <2 x float> poison, float %2029, i64 0, !dbg !43 + %3790 = insertelement <2 x float> %3789, float %2030, i64 1, !dbg !43 + %3791 = insertelement <2 x float> poison, float %3533, i64 0, !dbg !43 + %3792 = insertelement <2 x float> %3791, float %3534, i64 1, !dbg !43 + %3793 = fsub <2 x float> %3790, %3792, !dbg !43 + %3794 = insertelement <2 x float> poison, float %2031, i64 0, !dbg !43 + %3795 = insertelement <2 x float> %3794, float %2032, i64 1, !dbg !43 + %3796 = insertelement <2 x float> poison, float %3535, i64 0, !dbg !43 + %3797 = insertelement <2 x float> %3796, float %3536, i64 1, !dbg !43 + %3798 = fsub <2 x float> %3795, %3797, !dbg !43 + %3799 = insertelement <2 x float> poison, float %2033, i64 0, !dbg !43 + %3800 = insertelement <2 x float> %3799, float %2034, i64 1, !dbg !43 + %3801 = insertelement <2 x float> poison, float %3537, i64 0, !dbg !43 + %3802 = insertelement <2 x float> %3801, float %3538, i64 1, !dbg !43 + %3803 = fsub <2 x float> %3800, %3802, !dbg !43 + %3804 = insertelement <2 x float> poison, float %2035, i64 0, !dbg !43 + %3805 = insertelement <2 x float> %3804, float %2036, i64 1, !dbg !43 + %3806 = insertelement <2 x float> poison, float %3539, i64 0, !dbg !43 + %3807 = insertelement <2 x float> %3806, float %3540, i64 1, !dbg !43 + %3808 = fsub <2 x float> %3805, %3807, !dbg !43 + %3809 = insertelement <2 x float> poison, float %2037, i64 0, !dbg !43 + %3810 = insertelement <2 x float> %3809, float %2038, i64 1, !dbg !43 + %3811 = insertelement <2 x float> poison, float %3541, i64 0, !dbg !43 + %3812 = insertelement <2 x float> %3811, float %3542, i64 1, !dbg !43 + %3813 = fsub <2 x float> %3810, %3812, !dbg !43 + %3814 = insertelement <2 x float> poison, float %2039, i64 0, !dbg !43 + %3815 = insertelement <2 x float> %3814, float %2040, i64 1, !dbg !43 + %3816 = insertelement <2 x float> poison, float %3543, i64 0, !dbg !43 + %3817 = insertelement <2 x float> %3816, float %3544, i64 1, !dbg !43 + %3818 = fsub <2 x float> %3815, %3817, !dbg !43 + %3819 = insertelement <2 x float> poison, float %2041, i64 0, !dbg !43 + %3820 = insertelement <2 x float> %3819, float %2042, i64 1, !dbg !43 + %3821 = insertelement <2 x float> poison, float %3545, i64 0, !dbg !43 + %3822 = insertelement <2 x float> %3821, float %3546, i64 1, !dbg !43 + %3823 = fsub <2 x float> %3820, %3822, !dbg !43 + %3824 = insertelement <2 x float> poison, float %2043, i64 0, !dbg !43 + %3825 = insertelement <2 x float> %3824, float %2044, i64 1, !dbg !43 + %3826 = insertelement <2 x float> poison, float %3547, i64 0, !dbg !43 + %3827 = insertelement <2 x float> %3826, float %3548, i64 1, !dbg !43 + %3828 = fsub <2 x float> %3825, %3827, !dbg !43 + %3829 = insertelement <2 x float> poison, float %2045, i64 0, !dbg !43 + %3830 = insertelement <2 x float> %3829, float %2046, i64 1, !dbg !43 + %3831 = insertelement <2 x float> poison, float %3549, i64 0, !dbg !43 + %3832 = insertelement <2 x float> %3831, float %3550, i64 1, !dbg !43 + %3833 = fsub <2 x float> %3830, %3832, !dbg !43 + %3834 = insertelement <2 x float> poison, float %2047, i64 0, !dbg !43 + %3835 = insertelement <2 x float> %3834, float %2048, i64 1, !dbg !43 + %3836 = insertelement <2 x float> poison, float %3551, i64 0, !dbg !43 + %3837 = insertelement <2 x float> %3836, float %3552, i64 1, !dbg !43 + %3838 = fsub <2 x float> %3835, %3837, !dbg !43 + %3839 = insertelement <2 x float> poison, float %2049, i64 0, !dbg !43 + %3840 = insertelement <2 x float> %3839, float %2050, i64 1, !dbg !43 + %3841 = insertelement <2 x float> poison, float %3553, i64 0, !dbg !43 + %3842 = insertelement <2 x float> %3841, float %3554, i64 1, !dbg !43 + %3843 = fsub <2 x float> %3840, %3842, !dbg !43 + %3844 = insertelement <2 x float> poison, float %2051, i64 0, !dbg !43 + %3845 = insertelement <2 x float> %3844, float %2052, i64 1, !dbg !43 + %3846 = insertelement <2 x float> poison, float %3555, i64 0, !dbg !43 + %3847 = insertelement <2 x float> %3846, float %3556, i64 1, !dbg !43 + %3848 = fsub <2 x float> %3845, %3847, !dbg !43 + %3849 = insertelement <2 x float> poison, float %2053, i64 0, !dbg !43 + %3850 = insertelement <2 x float> %3849, float %2054, i64 1, !dbg !43 + %3851 = insertelement <2 x float> poison, float %3557, i64 0, !dbg !43 + %3852 = insertelement <2 x float> %3851, float %3558, i64 1, !dbg !43 + %3853 = fsub <2 x float> %3850, %3852, !dbg !43 + %3854 = insertelement <2 x float> poison, float %2055, i64 0, !dbg !43 + %3855 = insertelement <2 x float> %3854, float %2056, i64 1, !dbg !43 + %3856 = insertelement <2 x float> poison, float %3559, i64 0, !dbg !43 + %3857 = insertelement <2 x float> %3856, float %3560, i64 1, !dbg !43 + %3858 = fsub <2 x float> %3855, %3857, !dbg !43 + %3859 = insertelement <2 x float> poison, float %2057, i64 0, !dbg !43 + %3860 = insertelement <2 x float> %3859, float %2058, i64 1, !dbg !43 + %3861 = insertelement <2 x float> poison, float %3561, i64 0, !dbg !43 + %3862 = insertelement <2 x float> %3861, float %3562, i64 1, !dbg !43 + %3863 = fsub <2 x float> %3860, %3862, !dbg !43 + %3864 = insertelement <2 x float> poison, float %2059, i64 0, !dbg !43 + %3865 = insertelement <2 x float> %3864, float %2060, i64 1, !dbg !43 + %3866 = insertelement <2 x float> poison, float %3563, i64 0, !dbg !43 + %3867 = insertelement <2 x float> %3866, float %3564, i64 1, !dbg !43 + %3868 = fsub <2 x float> %3865, %3867, !dbg !43 + %3869 = insertelement <2 x float> poison, float %2061, i64 0, !dbg !43 + %3870 = insertelement <2 x float> %3869, float %2062, i64 1, !dbg !43 + %3871 = insertelement <2 x float> poison, float %3565, i64 0, !dbg !43 + %3872 = insertelement <2 x float> %3871, float %3566, i64 1, !dbg !43 + %3873 = fsub <2 x float> %3870, %3872, !dbg !43 + %3874 = insertelement <2 x float> poison, float %2063, i64 0, !dbg !43 + %3875 = insertelement <2 x float> %3874, float %2064, i64 1, !dbg !43 + %3876 = insertelement <2 x float> poison, float %3567, i64 0, !dbg !43 + %3877 = insertelement <2 x float> %3876, float %3568, i64 1, !dbg !43 + %3878 = fsub <2 x float> %3875, %3877, !dbg !43 + %3879 = insertelement <2 x float> poison, float %2065, i64 0, !dbg !43 + %3880 = insertelement <2 x float> %3879, float %2066, i64 1, !dbg !43 + %3881 = insertelement <2 x float> poison, float %3569, i64 0, !dbg !43 + %3882 = insertelement <2 x float> %3881, float %3570, i64 1, !dbg !43 + %3883 = fsub <2 x float> %3880, %3882, !dbg !43 + %3884 = insertelement <2 x float> poison, float %2067, i64 0, !dbg !43 + %3885 = insertelement <2 x float> %3884, float %2068, i64 1, !dbg !43 + %3886 = insertelement <2 x float> poison, float %3571, i64 0, !dbg !43 + %3887 = insertelement <2 x float> %3886, float %3572, i64 1, !dbg !43 + %3888 = fsub <2 x float> %3885, %3887, !dbg !43 + %3889 = insertelement <2 x float> poison, float %2069, i64 0, !dbg !43 + %3890 = insertelement <2 x float> %3889, float %2070, i64 1, !dbg !43 + %3891 = insertelement <2 x float> poison, float %3573, i64 0, !dbg !43 + %3892 = insertelement <2 x float> %3891, float %3574, i64 1, !dbg !43 + %3893 = fsub <2 x float> %3890, %3892, !dbg !43 + %3894 = insertelement <2 x float> poison, float %2071, i64 0, !dbg !43 + %3895 = insertelement <2 x float> %3894, float %2072, i64 1, !dbg !43 + %3896 = insertelement <2 x float> poison, float %3575, i64 0, !dbg !43 + %3897 = insertelement <2 x float> %3896, float %3576, i64 1, !dbg !43 + %3898 = fsub <2 x float> %3895, %3897, !dbg !43 + %3899 = insertelement <2 x float> poison, float %2073, i64 0, !dbg !43 + %3900 = insertelement <2 x float> %3899, float %2074, i64 1, !dbg !43 + %3901 = insertelement <2 x float> poison, float %3577, i64 0, !dbg !43 + %3902 = insertelement <2 x float> %3901, float %3578, i64 1, !dbg !43 + %3903 = fsub <2 x float> %3900, %3902, !dbg !43 + %3904 = insertelement <2 x float> poison, float %2075, i64 0, !dbg !43 + %3905 = insertelement <2 x float> %3904, float %2076, i64 1, !dbg !43 + %3906 = insertelement <2 x float> poison, float %3579, i64 0, !dbg !43 + %3907 = insertelement <2 x float> %3906, float %3580, i64 1, !dbg !43 + %3908 = fsub <2 x float> %3905, %3907, !dbg !43 + %3909 = insertelement <2 x float> poison, float %2077, i64 0, !dbg !43 + %3910 = insertelement <2 x float> %3909, float %2078, i64 1, !dbg !43 + %3911 = insertelement <2 x float> poison, float %3581, i64 0, !dbg !43 + %3912 = insertelement <2 x float> %3911, float %3582, i64 1, !dbg !43 + %3913 = fsub <2 x float> %3910, %3912, !dbg !43 + %3914 = insertelement <2 x float> poison, float %2079, i64 0, !dbg !43 + %3915 = insertelement <2 x float> %3914, float %2080, i64 1, !dbg !43 + %3916 = insertelement <2 x float> poison, float %3583, i64 0, !dbg !43 + %3917 = insertelement <2 x float> %3916, float %3584, i64 1, !dbg !43 + %3918 = fsub <2 x float> %3915, %3917, !dbg !43 + %3919 = insertelement <2 x float> poison, float %2081, i64 0, !dbg !43 + %3920 = insertelement <2 x float> %3919, float %2082, i64 1, !dbg !43 + %3921 = insertelement <2 x float> poison, float %3585, i64 0, !dbg !43 + %3922 = insertelement <2 x float> %3921, float %3586, i64 1, !dbg !43 + %3923 = fsub <2 x float> %3920, %3922, !dbg !43 + %3924 = insertelement <2 x float> poison, float %2083, i64 0, !dbg !43 + %3925 = insertelement <2 x float> %3924, float %2084, i64 1, !dbg !43 + %3926 = insertelement <2 x float> poison, float %3587, i64 0, !dbg !43 + %3927 = insertelement <2 x float> %3926, float %3588, i64 1, !dbg !43 + %3928 = fsub <2 x float> %3925, %3927, !dbg !43 + %3929 = insertelement <2 x float> poison, float %2085, i64 0, !dbg !43 + %3930 = insertelement <2 x float> %3929, float %2086, i64 1, !dbg !43 + %3931 = insertelement <2 x float> poison, float %3589, i64 0, !dbg !43 + %3932 = insertelement <2 x float> %3931, float %3590, i64 1, !dbg !43 + %3933 = fsub <2 x float> %3930, %3932, !dbg !43 + %3934 = insertelement <2 x float> poison, float %2087, i64 0, !dbg !43 + %3935 = insertelement <2 x float> %3934, float %2088, i64 1, !dbg !43 + %3936 = insertelement <2 x float> poison, float %3591, i64 0, !dbg !43 + %3937 = insertelement <2 x float> %3936, float %3592, i64 1, !dbg !43 + %3938 = fsub <2 x float> %3935, %3937, !dbg !43 + %3939 = insertelement <2 x float> poison, float %2089, i64 0, !dbg !43 + %3940 = insertelement <2 x float> %3939, float %2090, i64 1, !dbg !43 + %3941 = insertelement <2 x float> poison, float %3593, i64 0, !dbg !43 + %3942 = insertelement <2 x float> %3941, float %3594, i64 1, !dbg !43 + %3943 = fsub <2 x float> %3940, %3942, !dbg !43 + %3944 = insertelement <2 x float> poison, float %2091, i64 0, !dbg !43 + %3945 = insertelement <2 x float> %3944, float %2092, i64 1, !dbg !43 + %3946 = insertelement <2 x float> poison, float %3595, i64 0, !dbg !43 + %3947 = insertelement <2 x float> %3946, float %3596, i64 1, !dbg !43 + %3948 = fsub <2 x float> %3945, %3947, !dbg !43 + %3949 = insertelement <2 x float> poison, float %2093, i64 0, !dbg !43 + %3950 = insertelement <2 x float> %3949, float %2094, i64 1, !dbg !43 + %3951 = insertelement <2 x float> poison, float %3597, i64 0, !dbg !43 + %3952 = insertelement <2 x float> %3951, float %3598, i64 1, !dbg !43 + %3953 = fsub <2 x float> %3950, %3952, !dbg !43 + %3954 = insertelement <2 x float> poison, float %2095, i64 0, !dbg !43 + %3955 = insertelement <2 x float> %3954, float %2096, i64 1, !dbg !43 + %3956 = insertelement <2 x float> poison, float %3599, i64 0, !dbg !43 + %3957 = insertelement <2 x float> %3956, float %3600, i64 1, !dbg !43 + %3958 = fsub <2 x float> %3955, %3957, !dbg !43 + %3959 = insertelement <2 x float> poison, float %2097, i64 0, !dbg !43 + %3960 = insertelement <2 x float> %3959, float %2098, i64 1, !dbg !43 + %3961 = insertelement <2 x float> poison, float %3601, i64 0, !dbg !43 + %3962 = insertelement <2 x float> %3961, float %3602, i64 1, !dbg !43 + %3963 = fsub <2 x float> %3960, %3962, !dbg !43 + %3964 = insertelement <2 x float> poison, float %2099, i64 0, !dbg !43 + %3965 = insertelement <2 x float> %3964, float %2100, i64 1, !dbg !43 + %3966 = insertelement <2 x float> poison, float %3603, i64 0, !dbg !43 + %3967 = insertelement <2 x float> %3966, float %3604, i64 1, !dbg !43 + %3968 = fsub <2 x float> %3965, %3967, !dbg !43 + %3969 = insertelement <2 x float> poison, float %2101, i64 0, !dbg !43 + %3970 = insertelement <2 x float> %3969, float %2102, i64 1, !dbg !43 + %3971 = insertelement <2 x float> poison, float %3605, i64 0, !dbg !43 + %3972 = insertelement <2 x float> %3971, float %3606, i64 1, !dbg !43 + %3973 = fsub <2 x float> %3970, %3972, !dbg !43 + %3974 = insertelement <2 x float> poison, float %2103, i64 0, !dbg !43 + %3975 = insertelement <2 x float> %3974, float %2104, i64 1, !dbg !43 + %3976 = insertelement <2 x float> poison, float %3607, i64 0, !dbg !43 + %3977 = insertelement <2 x float> %3976, float %3608, i64 1, !dbg !43 + %3978 = fsub <2 x float> %3975, %3977, !dbg !43 + %3979 = insertelement <2 x float> poison, float %2105, i64 0, !dbg !43 + %3980 = insertelement <2 x float> %3979, float %2106, i64 1, !dbg !43 + %3981 = insertelement <2 x float> poison, float %3609, i64 0, !dbg !43 + %3982 = insertelement <2 x float> %3981, float %3610, i64 1, !dbg !43 + %3983 = fsub <2 x float> %3980, %3982, !dbg !43 + %3984 = insertelement <2 x float> poison, float %2107, i64 0, !dbg !43 + %3985 = insertelement <2 x float> %3984, float %2108, i64 1, !dbg !43 + %3986 = insertelement <2 x float> poison, float %3611, i64 0, !dbg !43 + %3987 = insertelement <2 x float> %3986, float %3612, i64 1, !dbg !43 + %3988 = fsub <2 x float> %3985, %3987, !dbg !43 + %3989 = insertelement <2 x float> poison, float %2109, i64 0, !dbg !43 + %3990 = insertelement <2 x float> %3989, float %2110, i64 1, !dbg !43 + %3991 = insertelement <2 x float> poison, float %3613, i64 0, !dbg !43 + %3992 = insertelement <2 x float> %3991, float %3614, i64 1, !dbg !43 + %3993 = fsub <2 x float> %3990, %3992, !dbg !43 + %3994 = insertelement <2 x float> poison, float %2111, i64 0, !dbg !43 + %3995 = insertelement <2 x float> %3994, float %2112, i64 1, !dbg !43 + %3996 = insertelement <2 x float> poison, float %3615, i64 0, !dbg !43 + %3997 = insertelement <2 x float> %3996, float %3616, i64 1, !dbg !43 + %3998 = fsub <2 x float> %3995, %3997, !dbg !43 + %3999 = insertelement <2 x float> poison, float %2113, i64 0, !dbg !43 + %4000 = insertelement <2 x float> %3999, float %2114, i64 1, !dbg !43 + %4001 = insertelement <2 x float> poison, float %3617, i64 0, !dbg !43 + %4002 = insertelement <2 x float> %4001, float %3618, i64 1, !dbg !43 + %4003 = fsub <2 x float> %4000, %4002, !dbg !43 + %4004 = insertelement <2 x float> poison, float %2115, i64 0, !dbg !43 + %4005 = insertelement <2 x float> %4004, float %2116, i64 1, !dbg !43 + %4006 = insertelement <2 x float> poison, float %3619, i64 0, !dbg !43 + %4007 = insertelement <2 x float> %4006, float %3620, i64 1, !dbg !43 + %4008 = fsub <2 x float> %4005, %4007, !dbg !43 + %4009 = insertelement <2 x float> poison, float %2117, i64 0, !dbg !43 + %4010 = insertelement <2 x float> %4009, float %2118, i64 1, !dbg !43 + %4011 = insertelement <2 x float> poison, float %3621, i64 0, !dbg !43 + %4012 = insertelement <2 x float> %4011, float %3622, i64 1, !dbg !43 + %4013 = fsub <2 x float> %4010, %4012, !dbg !43 + %4014 = insertelement <2 x float> poison, float %2119, i64 0, !dbg !43 + %4015 = insertelement <2 x float> %4014, float %2120, i64 1, !dbg !43 + %4016 = insertelement <2 x float> poison, float %3623, i64 0, !dbg !43 + %4017 = insertelement <2 x float> %4016, float %3624, i64 1, !dbg !43 + %4018 = fsub <2 x float> %4015, %4017, !dbg !43 + %4019 = insertelement <2 x float> poison, float %2121, i64 0, !dbg !43 + %4020 = insertelement <2 x float> %4019, float %2122, i64 1, !dbg !43 + %4021 = insertelement <2 x float> poison, float %3625, i64 0, !dbg !43 + %4022 = insertelement <2 x float> %4021, float %3626, i64 1, !dbg !43 + %4023 = fsub <2 x float> %4020, %4022, !dbg !43 + %4024 = insertelement <2 x float> poison, float %2123, i64 0, !dbg !43 + %4025 = insertelement <2 x float> %4024, float %2124, i64 1, !dbg !43 + %4026 = insertelement <2 x float> poison, float %3627, i64 0, !dbg !43 + %4027 = insertelement <2 x float> %4026, float %3628, i64 1, !dbg !43 + %4028 = fsub <2 x float> %4025, %4027, !dbg !43 + %4029 = insertelement <2 x float> poison, float %2125, i64 0, !dbg !43 + %4030 = insertelement <2 x float> %4029, float %2126, i64 1, !dbg !43 + %4031 = insertelement <2 x float> poison, float %3629, i64 0, !dbg !43 + %4032 = insertelement <2 x float> %4031, float %3630, i64 1, !dbg !43 + %4033 = fsub <2 x float> %4030, %4032, !dbg !43 + %4034 = insertelement <2 x float> poison, float %2127, i64 0, !dbg !43 + %4035 = insertelement <2 x float> %4034, float %2128, i64 1, !dbg !43 + %4036 = insertelement <2 x float> poison, float %3631, i64 0, !dbg !43 + %4037 = insertelement <2 x float> %4036, float %3632, i64 1, !dbg !43 + %4038 = fsub <2 x float> %4035, %4037, !dbg !43 + %4039 = insertelement <2 x float> poison, float %2129, i64 0, !dbg !43 + %4040 = insertelement <2 x float> %4039, float %2130, i64 1, !dbg !43 + %4041 = insertelement <2 x float> poison, float %3633, i64 0, !dbg !43 + %4042 = insertelement <2 x float> %4041, float %3634, i64 1, !dbg !43 + %4043 = fsub <2 x float> %4040, %4042, !dbg !43 + %4044 = insertelement <2 x float> poison, float %2131, i64 0, !dbg !43 + %4045 = insertelement <2 x float> %4044, float %2132, i64 1, !dbg !43 + %4046 = insertelement <2 x float> poison, float %3635, i64 0, !dbg !43 + %4047 = insertelement <2 x float> %4046, float %3636, i64 1, !dbg !43 + %4048 = fsub <2 x float> %4045, %4047, !dbg !43 + %4049 = insertelement <2 x float> poison, float %2133, i64 0, !dbg !43 + %4050 = insertelement <2 x float> %4049, float %2134, i64 1, !dbg !43 + %4051 = insertelement <2 x float> poison, float %3637, i64 0, !dbg !43 + %4052 = insertelement <2 x float> %4051, float %3638, i64 1, !dbg !43 + %4053 = fsub <2 x float> %4050, %4052, !dbg !43 + %4054 = insertelement <2 x float> poison, float %2135, i64 0, !dbg !43 + %4055 = insertelement <2 x float> %4054, float %2136, i64 1, !dbg !43 + %4056 = insertelement <2 x float> poison, float %3639, i64 0, !dbg !43 + %4057 = insertelement <2 x float> %4056, float %3640, i64 1, !dbg !43 + %4058 = fsub <2 x float> %4055, %4057, !dbg !43 + %4059 = insertelement <2 x float> poison, float %2137, i64 0, !dbg !43 + %4060 = insertelement <2 x float> %4059, float %2138, i64 1, !dbg !43 + %4061 = insertelement <2 x float> poison, float %3641, i64 0, !dbg !43 + %4062 = insertelement <2 x float> %4061, float %3642, i64 1, !dbg !43 + %4063 = fsub <2 x float> %4060, %4062, !dbg !43 + %4064 = insertelement <2 x float> poison, float %2139, i64 0, !dbg !43 + %4065 = insertelement <2 x float> %4064, float %2140, i64 1, !dbg !43 + %4066 = insertelement <2 x float> poison, float %3643, i64 0, !dbg !43 + %4067 = insertelement <2 x float> %4066, float %3644, i64 1, !dbg !43 + %4068 = fsub <2 x float> %4065, %4067, !dbg !43 + %4069 = insertelement <2 x float> poison, float %2141, i64 0, !dbg !43 + %4070 = insertelement <2 x float> %4069, float %2142, i64 1, !dbg !43 + %4071 = insertelement <2 x float> poison, float %3645, i64 0, !dbg !43 + %4072 = insertelement <2 x float> %4071, float %3646, i64 1, !dbg !43 + %4073 = fsub <2 x float> %4070, %4072, !dbg !43 + %4074 = insertelement <2 x float> poison, float %2143, i64 0, !dbg !43 + %4075 = insertelement <2 x float> %4074, float %2144, i64 1, !dbg !43 + %4076 = insertelement <2 x float> poison, float %3647, i64 0, !dbg !43 + %4077 = insertelement <2 x float> %4076, float %3648, i64 1, !dbg !43 + %4078 = fsub <2 x float> %4075, %4077, !dbg !43 + %4079 = insertelement <2 x float> poison, float %2145, i64 0, !dbg !43 + %4080 = insertelement <2 x float> %4079, float %2146, i64 1, !dbg !43 + %4081 = insertelement <2 x float> poison, float %3649, i64 0, !dbg !43 + %4082 = insertelement <2 x float> %4081, float %3650, i64 1, !dbg !43 + %4083 = fsub <2 x float> %4080, %4082, !dbg !43 + %4084 = insertelement <2 x float> poison, float %2147, i64 0, !dbg !43 + %4085 = insertelement <2 x float> %4084, float %2148, i64 1, !dbg !43 + %4086 = insertelement <2 x float> poison, float %3651, i64 0, !dbg !43 + %4087 = insertelement <2 x float> %4086, float %3652, i64 1, !dbg !43 + %4088 = fsub <2 x float> %4085, %4087, !dbg !43 + %4089 = insertelement <2 x float> poison, float %2149, i64 0, !dbg !43 + %4090 = insertelement <2 x float> %4089, float %2150, i64 1, !dbg !43 + %4091 = insertelement <2 x float> poison, float %3653, i64 0, !dbg !43 + %4092 = insertelement <2 x float> %4091, float %3654, i64 1, !dbg !43 + %4093 = fsub <2 x float> %4090, %4092, !dbg !43 + %4094 = insertelement <2 x float> poison, float %2151, i64 0, !dbg !43 + %4095 = insertelement <2 x float> %4094, float %2152, i64 1, !dbg !43 + %4096 = insertelement <2 x float> poison, float %3655, i64 0, !dbg !43 + %4097 = insertelement <2 x float> %4096, float %3656, i64 1, !dbg !43 + %4098 = fsub <2 x float> %4095, %4097, !dbg !43 + %4099 = insertelement <2 x float> poison, float %2153, i64 0, !dbg !43 + %4100 = insertelement <2 x float> %4099, float %2154, i64 1, !dbg !43 + %4101 = insertelement <2 x float> poison, float %3657, i64 0, !dbg !43 + %4102 = insertelement <2 x float> %4101, float %3658, i64 1, !dbg !43 + %4103 = fsub <2 x float> %4100, %4102, !dbg !43 + %4104 = insertelement <2 x float> poison, float %2155, i64 0, !dbg !43 + %4105 = insertelement <2 x float> %4104, float %2156, i64 1, !dbg !43 + %4106 = insertelement <2 x float> poison, float %3659, i64 0, !dbg !43 + %4107 = insertelement <2 x float> %4106, float %3660, i64 1, !dbg !43 + %4108 = fsub <2 x float> %4105, %4107, !dbg !43 + %4109 = insertelement <2 x float> poison, float %2157, i64 0, !dbg !43 + %4110 = insertelement <2 x float> %4109, float %2158, i64 1, !dbg !43 + %4111 = insertelement <2 x float> poison, float %3661, i64 0, !dbg !43 + %4112 = insertelement <2 x float> %4111, float %3662, i64 1, !dbg !43 + %4113 = fsub <2 x float> %4110, %4112, !dbg !43 + %4114 = insertelement <2 x float> poison, float %2159, i64 0, !dbg !43 + %4115 = insertelement <2 x float> %4114, float %2160, i64 1, !dbg !43 + %4116 = insertelement <2 x float> poison, float %3663, i64 0, !dbg !43 + %4117 = insertelement <2 x float> %4116, float %3664, i64 1, !dbg !43 + %4118 = fsub <2 x float> %4115, %4117, !dbg !43 + %4119 = insertelement <2 x float> poison, float %2161, i64 0, !dbg !43 + %4120 = insertelement <2 x float> %4119, float %2162, i64 1, !dbg !43 + %4121 = insertelement <2 x float> poison, float %3665, i64 0, !dbg !43 + %4122 = insertelement <2 x float> %4121, float %3666, i64 1, !dbg !43 + %4123 = fsub <2 x float> %4120, %4122, !dbg !43 + %4124 = insertelement <2 x float> poison, float %2163, i64 0, !dbg !43 + %4125 = insertelement <2 x float> %4124, float %2164, i64 1, !dbg !43 + %4126 = insertelement <2 x float> poison, float %3667, i64 0, !dbg !43 + %4127 = insertelement <2 x float> %4126, float %3668, i64 1, !dbg !43 + %4128 = fsub <2 x float> %4125, %4127, !dbg !43 + %4129 = insertelement <2 x float> poison, float %2165, i64 0, !dbg !43 + %4130 = insertelement <2 x float> %4129, float %2166, i64 1, !dbg !43 + %4131 = insertelement <2 x float> poison, float %3669, i64 0, !dbg !43 + %4132 = insertelement <2 x float> %4131, float %3670, i64 1, !dbg !43 + %4133 = fsub <2 x float> %4130, %4132, !dbg !43 + %4134 = insertelement <2 x float> poison, float %2167, i64 0, !dbg !43 + %4135 = insertelement <2 x float> %4134, float %2168, i64 1, !dbg !43 + %4136 = insertelement <2 x float> poison, float %3671, i64 0, !dbg !43 + %4137 = insertelement <2 x float> %4136, float %3672, i64 1, !dbg !43 + %4138 = fsub <2 x float> %4135, %4137, !dbg !43 + %4139 = insertelement <2 x float> poison, float %2169, i64 0, !dbg !43 + %4140 = insertelement <2 x float> %4139, float %2170, i64 1, !dbg !43 + %4141 = insertelement <2 x float> poison, float %3673, i64 0, !dbg !43 + %4142 = insertelement <2 x float> %4141, float %3674, i64 1, !dbg !43 + %4143 = fsub <2 x float> %4140, %4142, !dbg !43 + %4144 = insertelement <2 x float> poison, float %2171, i64 0, !dbg !43 + %4145 = insertelement <2 x float> %4144, float %2172, i64 1, !dbg !43 + %4146 = insertelement <2 x float> poison, float %3675, i64 0, !dbg !43 + %4147 = insertelement <2 x float> %4146, float %3676, i64 1, !dbg !43 + %4148 = fsub <2 x float> %4145, %4147, !dbg !43 + %4149 = insertelement <2 x float> poison, float %2173, i64 0, !dbg !43 + %4150 = insertelement <2 x float> %4149, float %2174, i64 1, !dbg !43 + %4151 = insertelement <2 x float> poison, float %3677, i64 0, !dbg !43 + %4152 = insertelement <2 x float> %4151, float %3678, i64 1, !dbg !43 + %4153 = fsub <2 x float> %4150, %4152, !dbg !43 + %4154 = insertelement <2 x float> poison, float %2175, i64 0, !dbg !43 + %4155 = insertelement <2 x float> %4154, float %2176, i64 1, !dbg !43 + %4156 = insertelement <2 x float> poison, float %3679, i64 0, !dbg !43 + %4157 = insertelement <2 x float> %4156, float %3680, i64 1, !dbg !43 + %4158 = fsub <2 x float> %4155, %4157, !dbg !43 + %4159 = insertelement <2 x float> poison, float %2177, i64 0, !dbg !43 + %4160 = insertelement <2 x float> %4159, float %2178, i64 1, !dbg !43 + %4161 = insertelement <2 x float> poison, float %3681, i64 0, !dbg !43 + %4162 = insertelement <2 x float> %4161, float %3682, i64 1, !dbg !43 + %4163 = fsub <2 x float> %4160, %4162, !dbg !43 + %4164 = insertelement <2 x float> poison, float %2179, i64 0, !dbg !43 + %4165 = insertelement <2 x float> %4164, float %2180, i64 1, !dbg !43 + %4166 = insertelement <2 x float> poison, float %3683, i64 0, !dbg !43 + %4167 = insertelement <2 x float> %4166, float %3684, i64 1, !dbg !43 + %4168 = fsub <2 x float> %4165, %4167, !dbg !43 + %4169 = insertelement <2 x float> poison, float %2181, i64 0, !dbg !43 + %4170 = insertelement <2 x float> %4169, float %2182, i64 1, !dbg !43 + %4171 = insertelement <2 x float> poison, float %3685, i64 0, !dbg !43 + %4172 = insertelement <2 x float> %4171, float %3686, i64 1, !dbg !43 + %4173 = fsub <2 x float> %4170, %4172, !dbg !43 + %4174 = insertelement <2 x float> poison, float %2183, i64 0, !dbg !43 + %4175 = insertelement <2 x float> %4174, float %2184, i64 1, !dbg !43 + %4176 = insertelement <2 x float> poison, float %3687, i64 0, !dbg !43 + %4177 = insertelement <2 x float> %4176, float %3688, i64 1, !dbg !43 + %4178 = fsub <2 x float> %4175, %4177, !dbg !43 + %4179 = insertelement <2 x float> poison, float %2185, i64 0, !dbg !43 + %4180 = insertelement <2 x float> %4179, float %2186, i64 1, !dbg !43 + %4181 = insertelement <2 x float> poison, float %3689, i64 0, !dbg !43 + %4182 = insertelement <2 x float> %4181, float %3690, i64 1, !dbg !43 + %4183 = fsub <2 x float> %4180, %4182, !dbg !43 + %4184 = insertelement <2 x float> poison, float %2187, i64 0, !dbg !43 + %4185 = insertelement <2 x float> %4184, float %2188, i64 1, !dbg !43 + %4186 = insertelement <2 x float> poison, float %3691, i64 0, !dbg !43 + %4187 = insertelement <2 x float> %4186, float %3692, i64 1, !dbg !43 + %4188 = fsub <2 x float> %4185, %4187, !dbg !43 + %4189 = insertelement <2 x float> poison, float %2189, i64 0, !dbg !43 + %4190 = insertelement <2 x float> %4189, float %2190, i64 1, !dbg !43 + %4191 = insertelement <2 x float> poison, float %3693, i64 0, !dbg !43 + %4192 = insertelement <2 x float> %4191, float %3694, i64 1, !dbg !43 + %4193 = fsub <2 x float> %4190, %4192, !dbg !43 + %4194 = insertelement <2 x float> poison, float %2191, i64 0, !dbg !43 + %4195 = insertelement <2 x float> %4194, float %2192, i64 1, !dbg !43 + %4196 = insertelement <2 x float> poison, float %3695, i64 0, !dbg !43 + %4197 = insertelement <2 x float> %4196, float %3696, i64 1, !dbg !43 + %4198 = fsub <2 x float> %4195, %4197, !dbg !43 + %4199 = insertelement <2 x float> poison, float %2193, i64 0, !dbg !43 + %4200 = insertelement <2 x float> %4199, float %2194, i64 1, !dbg !43 + %4201 = insertelement <2 x float> poison, float %3697, i64 0, !dbg !43 + %4202 = insertelement <2 x float> %4201, float %3698, i64 1, !dbg !43 + %4203 = fsub <2 x float> %4200, %4202, !dbg !43 + %4204 = insertelement <2 x float> poison, float %2195, i64 0, !dbg !43 + %4205 = insertelement <2 x float> %4204, float %2196, i64 1, !dbg !43 + %4206 = insertelement <2 x float> poison, float %3699, i64 0, !dbg !43 + %4207 = insertelement <2 x float> %4206, float %3700, i64 1, !dbg !43 + %4208 = fsub <2 x float> %4205, %4207, !dbg !43 + %4209 = insertelement <2 x float> poison, float %2197, i64 0, !dbg !43 + %4210 = insertelement <2 x float> %4209, float %2198, i64 1, !dbg !43 + %4211 = insertelement <2 x float> poison, float %3701, i64 0, !dbg !43 + %4212 = insertelement <2 x float> %4211, float %3702, i64 1, !dbg !43 + %4213 = fsub <2 x float> %4210, %4212, !dbg !43 + %4214 = insertelement <2 x float> poison, float %2199, i64 0, !dbg !43 + %4215 = insertelement <2 x float> %4214, float %2200, i64 1, !dbg !43 + %4216 = insertelement <2 x float> poison, float %3703, i64 0, !dbg !43 + %4217 = insertelement <2 x float> %4216, float %3704, i64 1, !dbg !43 + %4218 = fsub <2 x float> %4215, %4217, !dbg !43 + %4219 = insertelement <2 x float> poison, float %2201, i64 0, !dbg !43 + %4220 = insertelement <2 x float> %4219, float %2202, i64 1, !dbg !43 + %4221 = insertelement <2 x float> poison, float %3705, i64 0, !dbg !43 + %4222 = insertelement <2 x float> %4221, float %3706, i64 1, !dbg !43 + %4223 = fsub <2 x float> %4220, %4222, !dbg !43 + %4224 = insertelement <2 x float> poison, float %2203, i64 0, !dbg !43 + %4225 = insertelement <2 x float> %4224, float %2204, i64 1, !dbg !43 + %4226 = insertelement <2 x float> poison, float %3707, i64 0, !dbg !43 + %4227 = insertelement <2 x float> %4226, float %3708, i64 1, !dbg !43 + %4228 = fsub <2 x float> %4225, %4227, !dbg !43 + %4229 = insertelement <2 x float> poison, float %2205, i64 0, !dbg !43 + %4230 = insertelement <2 x float> %4229, float %2206, i64 1, !dbg !43 + %4231 = insertelement <2 x float> poison, float %3709, i64 0, !dbg !43 + %4232 = insertelement <2 x float> %4231, float %3710, i64 1, !dbg !43 + %4233 = fsub <2 x float> %4230, %4232, !dbg !43 + %4234 = insertelement <2 x float> poison, float %2207, i64 0, !dbg !43 + %4235 = insertelement <2 x float> %4234, float %2208, i64 1, !dbg !43 + %4236 = insertelement <2 x float> poison, float %3711, i64 0, !dbg !43 + %4237 = insertelement <2 x float> %4236, float %3712, i64 1, !dbg !43 + %4238 = fsub <2 x float> %4235, %4237, !dbg !43 + %4239 = insertelement <2 x float> poison, float %2209, i64 0, !dbg !43 + %4240 = insertelement <2 x float> %4239, float %2210, i64 1, !dbg !43 + %4241 = insertelement <2 x float> poison, float %3713, i64 0, !dbg !43 + %4242 = insertelement <2 x float> %4241, float %3714, i64 1, !dbg !43 + %4243 = fsub <2 x float> %4240, %4242, !dbg !43 + %4244 = insertelement <2 x float> poison, float %2211, i64 0, !dbg !43 + %4245 = insertelement <2 x float> %4244, float %2212, i64 1, !dbg !43 + %4246 = insertelement <2 x float> poison, float %3715, i64 0, !dbg !43 + %4247 = insertelement <2 x float> %4246, float %3716, i64 1, !dbg !43 + %4248 = fsub <2 x float> %4245, %4247, !dbg !43 + %4249 = insertelement <2 x float> poison, float %2213, i64 0, !dbg !43 + %4250 = insertelement <2 x float> %4249, float %2214, i64 1, !dbg !43 + %4251 = insertelement <2 x float> poison, float %3717, i64 0, !dbg !43 + %4252 = insertelement <2 x float> %4251, float %3718, i64 1, !dbg !43 + %4253 = fsub <2 x float> %4250, %4252, !dbg !43 + %4254 = insertelement <2 x float> poison, float %2215, i64 0, !dbg !43 + %4255 = insertelement <2 x float> %4254, float %2216, i64 1, !dbg !43 + %4256 = insertelement <2 x float> poison, float %3719, i64 0, !dbg !43 + %4257 = insertelement <2 x float> %4256, float %3720, i64 1, !dbg !43 + %4258 = fsub <2 x float> %4255, %4257, !dbg !43 + %4259 = insertelement <2 x float> poison, float %2217, i64 0, !dbg !43 + %4260 = insertelement <2 x float> %4259, float %2218, i64 1, !dbg !43 + %4261 = insertelement <2 x float> poison, float %3721, i64 0, !dbg !43 + %4262 = insertelement <2 x float> %4261, float %3722, i64 1, !dbg !43 + %4263 = fsub <2 x float> %4260, %4262, !dbg !43 + %4264 = insertelement <2 x float> poison, float %2219, i64 0, !dbg !43 + %4265 = insertelement <2 x float> %4264, float %2220, i64 1, !dbg !43 + %4266 = insertelement <2 x float> poison, float %3723, i64 0, !dbg !43 + %4267 = insertelement <2 x float> %4266, float %3724, i64 1, !dbg !43 + %4268 = fsub <2 x float> %4265, %4267, !dbg !43 + %4269 = insertelement <2 x float> poison, float %2221, i64 0, !dbg !43 + %4270 = insertelement <2 x float> %4269, float %2222, i64 1, !dbg !43 + %4271 = insertelement <2 x float> poison, float %3725, i64 0, !dbg !43 + %4272 = insertelement <2 x float> %4271, float %3726, i64 1, !dbg !43 + %4273 = fsub <2 x float> %4270, %4272, !dbg !43 + %4274 = insertelement <2 x float> poison, float %2223, i64 0, !dbg !43 + %4275 = insertelement <2 x float> %4274, float %2224, i64 1, !dbg !43 + %4276 = insertelement <2 x float> poison, float %3727, i64 0, !dbg !43 + %4277 = insertelement <2 x float> %4276, float %3728, i64 1, !dbg !43 + %4278 = fsub <2 x float> %4275, %4277, !dbg !43 + %4279 = insertelement <2 x float> poison, float %2225, i64 0, !dbg !43 + %4280 = insertelement <2 x float> %4279, float %2226, i64 1, !dbg !43 + %4281 = insertelement <2 x float> poison, float %3729, i64 0, !dbg !43 + %4282 = insertelement <2 x float> %4281, float %3730, i64 1, !dbg !43 + %4283 = fsub <2 x float> %4280, %4282, !dbg !43 + %4284 = insertelement <2 x float> poison, float %2227, i64 0, !dbg !43 + %4285 = insertelement <2 x float> %4284, float %2228, i64 1, !dbg !43 + %4286 = insertelement <2 x float> poison, float %3731, i64 0, !dbg !43 + %4287 = insertelement <2 x float> %4286, float %3732, i64 1, !dbg !43 + %4288 = fsub <2 x float> %4285, %4287, !dbg !43 + %4289 = insertelement <2 x float> poison, float %2229, i64 0, !dbg !43 + %4290 = insertelement <2 x float> %4289, float %2230, i64 1, !dbg !43 + %4291 = insertelement <2 x float> poison, float %3733, i64 0, !dbg !43 + %4292 = insertelement <2 x float> %4291, float %3734, i64 1, !dbg !43 + %4293 = fsub <2 x float> %4290, %4292, !dbg !43 + %4294 = insertelement <2 x float> poison, float %2231, i64 0, !dbg !43 + %4295 = insertelement <2 x float> %4294, float %2232, i64 1, !dbg !43 + %4296 = insertelement <2 x float> poison, float %3735, i64 0, !dbg !43 + %4297 = insertelement <2 x float> %4296, float %3736, i64 1, !dbg !43 + %4298 = fsub <2 x float> %4295, %4297, !dbg !43 + %4299 = insertelement <2 x float> poison, float %2233, i64 0, !dbg !43 + %4300 = insertelement <2 x float> %4299, float %2234, i64 1, !dbg !43 + %4301 = insertelement <2 x float> poison, float %3737, i64 0, !dbg !43 + %4302 = insertelement <2 x float> %4301, float %3738, i64 1, !dbg !43 + %4303 = fsub <2 x float> %4300, %4302, !dbg !43 + %4304 = insertelement <2 x float> poison, float %2235, i64 0, !dbg !43 + %4305 = insertelement <2 x float> %4304, float %2236, i64 1, !dbg !43 + %4306 = insertelement <2 x float> poison, float %3739, i64 0, !dbg !43 + %4307 = insertelement <2 x float> %4306, float %3740, i64 1, !dbg !43 + %4308 = fsub <2 x float> %4305, %4307, !dbg !43 + %4309 = insertelement <2 x float> poison, float %2237, i64 0, !dbg !43 + %4310 = insertelement <2 x float> %4309, float %2238, i64 1, !dbg !43 + %4311 = insertelement <2 x float> poison, float %3741, i64 0, !dbg !43 + %4312 = insertelement <2 x float> %4311, float %3742, i64 1, !dbg !43 + %4313 = fsub <2 x float> %4310, %4312, !dbg !43 + %4314 = insertelement <2 x float> poison, float %2239, i64 0, !dbg !43 + %4315 = insertelement <2 x float> %4314, float %2240, i64 1, !dbg !43 + %4316 = insertelement <2 x float> poison, float %3743, i64 0, !dbg !43 + %4317 = insertelement <2 x float> %4316, float %3744, i64 1, !dbg !43 + %4318 = fsub <2 x float> %4315, %4317, !dbg !43 + %4319 = insertelement <2 x float> poison, float %2241, i64 0, !dbg !43 + %4320 = insertelement <2 x float> %4319, float %2242, i64 1, !dbg !43 + %4321 = insertelement <2 x float> poison, float %3745, i64 0, !dbg !43 + %4322 = insertelement <2 x float> %4321, float %3746, i64 1, !dbg !43 + %4323 = fsub <2 x float> %4320, %4322, !dbg !43 + %4324 = insertelement <2 x float> poison, float %2243, i64 0, !dbg !43 + %4325 = insertelement <2 x float> %4324, float %2244, i64 1, !dbg !43 + %4326 = insertelement <2 x float> poison, float %3747, i64 0, !dbg !43 + %4327 = insertelement <2 x float> %4326, float %3748, i64 1, !dbg !43 + %4328 = fsub <2 x float> %4325, %4327, !dbg !43 + %4329 = insertelement <2 x float> poison, float %2245, i64 0, !dbg !43 + %4330 = insertelement <2 x float> %4329, float %2246, i64 1, !dbg !43 + %4331 = insertelement <2 x float> poison, float %3749, i64 0, !dbg !43 + %4332 = insertelement <2 x float> %4331, float %3750, i64 1, !dbg !43 + %4333 = fsub <2 x float> %4330, %4332, !dbg !43 + %4334 = insertelement <2 x float> poison, float %2247, i64 0, !dbg !43 + %4335 = insertelement <2 x float> %4334, float %2248, i64 1, !dbg !43 + %4336 = insertelement <2 x float> poison, float %3751, i64 0, !dbg !43 + %4337 = insertelement <2 x float> %4336, float %3752, i64 1, !dbg !43 + %4338 = fsub <2 x float> %4335, %4337, !dbg !43 + %4339 = insertelement <2 x float> poison, float %2249, i64 0, !dbg !43 + %4340 = insertelement <2 x float> %4339, float %2250, i64 1, !dbg !43 + %4341 = insertelement <2 x float> poison, float %3753, i64 0, !dbg !43 + %4342 = insertelement <2 x float> %4341, float %3754, i64 1, !dbg !43 + %4343 = fsub <2 x float> %4340, %4342, !dbg !43 + %4344 = insertelement <2 x float> poison, float %2251, i64 0, !dbg !43 + %4345 = insertelement <2 x float> %4344, float %2252, i64 1, !dbg !43 + %4346 = insertelement <2 x float> poison, float %3755, i64 0, !dbg !43 + %4347 = insertelement <2 x float> %4346, float %3756, i64 1, !dbg !43 + %4348 = fsub <2 x float> %4345, %4347, !dbg !43 + %4349 = insertelement <2 x float> poison, float %2253, i64 0, !dbg !43 + %4350 = insertelement <2 x float> %4349, float %2254, i64 1, !dbg !43 + %4351 = insertelement <2 x float> poison, float %3757, i64 0, !dbg !43 + %4352 = insertelement <2 x float> %4351, float %3758, i64 1, !dbg !43 + %4353 = fsub <2 x float> %4350, %4352, !dbg !43 + %4354 = insertelement <2 x float> poison, float %2255, i64 0, !dbg !43 + %4355 = insertelement <2 x float> %4354, float %2256, i64 1, !dbg !43 + %4356 = insertelement <2 x float> poison, float %3759, i64 0, !dbg !43 + %4357 = insertelement <2 x float> %4356, float %3760, i64 1, !dbg !43 + %4358 = fsub <2 x float> %4355, %4357, !dbg !43 + %4359 = insertelement <2 x float> poison, float %2257, i64 0, !dbg !43 + %4360 = insertelement <2 x float> %4359, float %2258, i64 1, !dbg !43 + %4361 = insertelement <2 x float> poison, float %3761, i64 0, !dbg !43 + %4362 = insertelement <2 x float> %4361, float %3762, i64 1, !dbg !43 + %4363 = fsub <2 x float> %4360, %4362, !dbg !43 + %4364 = insertelement <2 x float> poison, float %2259, i64 0, !dbg !43 + %4365 = insertelement <2 x float> %4364, float %2260, i64 1, !dbg !43 + %4366 = insertelement <2 x float> poison, float %3763, i64 0, !dbg !43 + %4367 = insertelement <2 x float> %4366, float %3764, i64 1, !dbg !43 + %4368 = fsub <2 x float> %4365, %4367, !dbg !43 + %4369 = insertelement <2 x float> poison, float %2261, i64 0, !dbg !43 + %4370 = insertelement <2 x float> %4369, float %2262, i64 1, !dbg !43 + %4371 = insertelement <2 x float> poison, float %3765, i64 0, !dbg !43 + %4372 = insertelement <2 x float> %4371, float %3766, i64 1, !dbg !43 + %4373 = fsub <2 x float> %4370, %4372, !dbg !43 + %4374 = insertelement <2 x float> poison, float %2263, i64 0, !dbg !43 + %4375 = insertelement <2 x float> %4374, float %2264, i64 1, !dbg !43 + %4376 = insertelement <2 x float> poison, float %3767, i64 0, !dbg !43 + %4377 = insertelement <2 x float> %4376, float %3768, i64 1, !dbg !43 + %4378 = fsub <2 x float> %4375, %4377, !dbg !43 + %4379 = insertelement <2 x float> poison, float %2265, i64 0, !dbg !43 + %4380 = insertelement <2 x float> %4379, float %2266, i64 1, !dbg !43 + %4381 = insertelement <2 x float> poison, float %3769, i64 0, !dbg !43 + %4382 = insertelement <2 x float> %4381, float %3770, i64 1, !dbg !43 + %4383 = fsub <2 x float> %4380, %4382, !dbg !43 + %4384 = insertelement <2 x float> poison, float %2267, i64 0, !dbg !43 + %4385 = insertelement <2 x float> %4384, float %2268, i64 1, !dbg !43 + %4386 = insertelement <2 x float> poison, float %3771, i64 0, !dbg !43 + %4387 = insertelement <2 x float> %4386, float %3772, i64 1, !dbg !43 + %4388 = fsub <2 x float> %4385, %4387, !dbg !43 + %4389 = fptrunc float %3773 to bfloat, !dbg !44 + %4390 = fptrunc float %3774 to bfloat, !dbg !44 + %4391 = fptrunc float %3775 to bfloat, !dbg !44 + %4392 = fptrunc float %3776 to bfloat, !dbg !44 + %4393 = fptrunc float %3777 to bfloat, !dbg !44 + %4394 = fptrunc float %3778 to bfloat, !dbg !44 + %4395 = fptrunc float %3779 to bfloat, !dbg !44 + %4396 = fptrunc float %3780 to bfloat, !dbg !44 + %4397 = fptrunc float %3781 to bfloat, !dbg !44 + %4398 = fptrunc float %3782 to bfloat, !dbg !44 + %4399 = fptrunc float %3783 to bfloat, !dbg !44 + %4400 = fptrunc float %3784 to bfloat, !dbg !44 + %4401 = fptrunc float %3785 to bfloat, !dbg !44 + %4402 = fptrunc float %3786 to bfloat, !dbg !44 + %4403 = fptrunc float %3787 to bfloat, !dbg !44 + %4404 = fptrunc float %3788 to bfloat, !dbg !44 + %4405 = fptrunc <2 x float> %3793 to <2 x bfloat>, !dbg !44 + %4406 = fptrunc <2 x float> %3798 to <2 x bfloat>, !dbg !44 + %4407 = fptrunc <2 x float> %3803 to <2 x bfloat>, !dbg !44 + %4408 = fptrunc <2 x float> %3808 to <2 x bfloat>, !dbg !44 + %4409 = fptrunc <2 x float> %3813 to <2 x bfloat>, !dbg !44 + %4410 = fptrunc <2 x float> %3818 to <2 x bfloat>, !dbg !44 + %4411 = fptrunc <2 x float> %3823 to <2 x bfloat>, !dbg !44 + %4412 = fptrunc <2 x float> %3828 to <2 x bfloat>, !dbg !44 + %4413 = fptrunc <2 x float> %3833 to <2 x bfloat>, !dbg !44 + %4414 = fptrunc <2 x float> %3838 to <2 x bfloat>, !dbg !44 + %4415 = fptrunc <2 x float> %3843 to <2 x bfloat>, !dbg !44 + %4416 = fptrunc <2 x float> %3848 to <2 x bfloat>, !dbg !44 + %4417 = fptrunc <2 x float> %3853 to <2 x bfloat>, !dbg !44 + %4418 = fptrunc <2 x float> %3858 to <2 x bfloat>, !dbg !44 + %4419 = fptrunc <2 x float> %3863 to <2 x bfloat>, !dbg !44 + %4420 = fptrunc <2 x float> %3868 to <2 x bfloat>, !dbg !44 + %4421 = fptrunc <2 x float> %3873 to <2 x bfloat>, !dbg !44 + %4422 = fptrunc <2 x float> %3878 to <2 x bfloat>, !dbg !44 + %4423 = fptrunc <2 x float> %3883 to <2 x bfloat>, !dbg !44 + %4424 = fptrunc <2 x float> %3888 to <2 x bfloat>, !dbg !44 + %4425 = fptrunc <2 x float> %3893 to <2 x bfloat>, !dbg !44 + %4426 = fptrunc <2 x float> %3898 to <2 x bfloat>, !dbg !44 + %4427 = fptrunc <2 x float> %3903 to <2 x bfloat>, !dbg !44 + %4428 = fptrunc <2 x float> %3908 to <2 x bfloat>, !dbg !44 + %4429 = fptrunc <2 x float> %3913 to <2 x bfloat>, !dbg !44 + %4430 = fptrunc <2 x float> %3918 to <2 x bfloat>, !dbg !44 + %4431 = fptrunc <2 x float> %3923 to <2 x bfloat>, !dbg !44 + %4432 = fptrunc <2 x float> %3928 to <2 x bfloat>, !dbg !44 + %4433 = fptrunc <2 x float> %3933 to <2 x bfloat>, !dbg !44 + %4434 = fptrunc <2 x float> %3938 to <2 x bfloat>, !dbg !44 + %4435 = fptrunc <2 x float> %3943 to <2 x bfloat>, !dbg !44 + %4436 = fptrunc <2 x float> %3948 to <2 x bfloat>, !dbg !44 + %4437 = fptrunc <2 x float> %3953 to <2 x bfloat>, !dbg !44 + %4438 = fptrunc <2 x float> %3958 to <2 x bfloat>, !dbg !44 + %4439 = fptrunc <2 x float> %3963 to <2 x bfloat>, !dbg !44 + %4440 = fptrunc <2 x float> %3968 to <2 x bfloat>, !dbg !44 + %4441 = fptrunc <2 x float> %3973 to <2 x bfloat>, !dbg !44 + %4442 = fptrunc <2 x float> %3978 to <2 x bfloat>, !dbg !44 + %4443 = fptrunc <2 x float> %3983 to <2 x bfloat>, !dbg !44 + %4444 = fptrunc <2 x float> %3988 to <2 x bfloat>, !dbg !44 + %4445 = fptrunc <2 x float> %3993 to <2 x bfloat>, !dbg !44 + %4446 = fptrunc <2 x float> %3998 to <2 x bfloat>, !dbg !44 + %4447 = fptrunc <2 x float> %4003 to <2 x bfloat>, !dbg !44 + %4448 = fptrunc <2 x float> %4008 to <2 x bfloat>, !dbg !44 + %4449 = fptrunc <2 x float> %4013 to <2 x bfloat>, !dbg !44 + %4450 = fptrunc <2 x float> %4018 to <2 x bfloat>, !dbg !44 + %4451 = fptrunc <2 x float> %4023 to <2 x bfloat>, !dbg !44 + %4452 = fptrunc <2 x float> %4028 to <2 x bfloat>, !dbg !44 + %4453 = fptrunc <2 x float> %4033 to <2 x bfloat>, !dbg !44 + %4454 = fptrunc <2 x float> %4038 to <2 x bfloat>, !dbg !44 + %4455 = fptrunc <2 x float> %4043 to <2 x bfloat>, !dbg !44 + %4456 = fptrunc <2 x float> %4048 to <2 x bfloat>, !dbg !44 + %4457 = fptrunc <2 x float> %4053 to <2 x bfloat>, !dbg !44 + %4458 = fptrunc <2 x float> %4058 to <2 x bfloat>, !dbg !44 + %4459 = fptrunc <2 x float> %4063 to <2 x bfloat>, !dbg !44 + %4460 = fptrunc <2 x float> %4068 to <2 x bfloat>, !dbg !44 + %4461 = fptrunc <2 x float> %4073 to <2 x bfloat>, !dbg !44 + %4462 = fptrunc <2 x float> %4078 to <2 x bfloat>, !dbg !44 + %4463 = fptrunc <2 x float> %4083 to <2 x bfloat>, !dbg !44 + %4464 = fptrunc <2 x float> %4088 to <2 x bfloat>, !dbg !44 + %4465 = fptrunc <2 x float> %4093 to <2 x bfloat>, !dbg !44 + %4466 = fptrunc <2 x float> %4098 to <2 x bfloat>, !dbg !44 + %4467 = fptrunc <2 x float> %4103 to <2 x bfloat>, !dbg !44 + %4468 = fptrunc <2 x float> %4108 to <2 x bfloat>, !dbg !44 + %4469 = fptrunc <2 x float> %4113 to <2 x bfloat>, !dbg !44 + %4470 = fptrunc <2 x float> %4118 to <2 x bfloat>, !dbg !44 + %4471 = fptrunc <2 x float> %4123 to <2 x bfloat>, !dbg !44 + %4472 = fptrunc <2 x float> %4128 to <2 x bfloat>, !dbg !44 + %4473 = fptrunc <2 x float> %4133 to <2 x bfloat>, !dbg !44 + %4474 = fptrunc <2 x float> %4138 to <2 x bfloat>, !dbg !44 + %4475 = fptrunc <2 x float> %4143 to <2 x bfloat>, !dbg !44 + %4476 = fptrunc <2 x float> %4148 to <2 x bfloat>, !dbg !44 + %4477 = fptrunc <2 x float> %4153 to <2 x bfloat>, !dbg !44 + %4478 = fptrunc <2 x float> %4158 to <2 x bfloat>, !dbg !44 + %4479 = fptrunc <2 x float> %4163 to <2 x bfloat>, !dbg !44 + %4480 = fptrunc <2 x float> %4168 to <2 x bfloat>, !dbg !44 + %4481 = fptrunc <2 x float> %4173 to <2 x bfloat>, !dbg !44 + %4482 = fptrunc <2 x float> %4178 to <2 x bfloat>, !dbg !44 + %4483 = fptrunc <2 x float> %4183 to <2 x bfloat>, !dbg !44 + %4484 = fptrunc <2 x float> %4188 to <2 x bfloat>, !dbg !44 + %4485 = fptrunc <2 x float> %4193 to <2 x bfloat>, !dbg !44 + %4486 = fptrunc <2 x float> %4198 to <2 x bfloat>, !dbg !44 + %4487 = fptrunc <2 x float> %4203 to <2 x bfloat>, !dbg !44 + %4488 = fptrunc <2 x float> %4208 to <2 x bfloat>, !dbg !44 + %4489 = fptrunc <2 x float> %4213 to <2 x bfloat>, !dbg !44 + %4490 = fptrunc <2 x float> %4218 to <2 x bfloat>, !dbg !44 + %4491 = fptrunc <2 x float> %4223 to <2 x bfloat>, !dbg !44 + %4492 = fptrunc <2 x float> %4228 to <2 x bfloat>, !dbg !44 + %4493 = fptrunc <2 x float> %4233 to <2 x bfloat>, !dbg !44 + %4494 = fptrunc <2 x float> %4238 to <2 x bfloat>, !dbg !44 + %4495 = fptrunc <2 x float> %4243 to <2 x bfloat>, !dbg !44 + %4496 = fptrunc <2 x float> %4248 to <2 x bfloat>, !dbg !44 + %4497 = fptrunc <2 x float> %4253 to <2 x bfloat>, !dbg !44 + %4498 = fptrunc <2 x float> %4258 to <2 x bfloat>, !dbg !44 + %4499 = fptrunc <2 x float> %4263 to <2 x bfloat>, !dbg !44 + %4500 = fptrunc <2 x float> %4268 to <2 x bfloat>, !dbg !44 + %4501 = fptrunc <2 x float> %4273 to <2 x bfloat>, !dbg !44 + %4502 = fptrunc <2 x float> %4278 to <2 x bfloat>, !dbg !44 + %4503 = fptrunc <2 x float> %4283 to <2 x bfloat>, !dbg !44 + %4504 = fptrunc <2 x float> %4288 to <2 x bfloat>, !dbg !44 + %4505 = fptrunc <2 x float> %4293 to <2 x bfloat>, !dbg !44 + %4506 = fptrunc <2 x float> %4298 to <2 x bfloat>, !dbg !44 + %4507 = fptrunc <2 x float> %4303 to <2 x bfloat>, !dbg !44 + %4508 = fptrunc <2 x float> %4308 to <2 x bfloat>, !dbg !44 + %4509 = fptrunc <2 x float> %4313 to <2 x bfloat>, !dbg !44 + %4510 = fptrunc <2 x float> %4318 to <2 x bfloat>, !dbg !44 + %4511 = fptrunc <2 x float> %4323 to <2 x bfloat>, !dbg !44 + %4512 = fptrunc <2 x float> %4328 to <2 x bfloat>, !dbg !44 + %4513 = fptrunc <2 x float> %4333 to <2 x bfloat>, !dbg !44 + %4514 = fptrunc <2 x float> %4338 to <2 x bfloat>, !dbg !44 + %4515 = fptrunc <2 x float> %4343 to <2 x bfloat>, !dbg !44 + %4516 = fptrunc <2 x float> %4348 to <2 x bfloat>, !dbg !44 + %4517 = fptrunc <2 x float> %4353 to <2 x bfloat>, !dbg !44 + %4518 = fptrunc <2 x float> %4358 to <2 x bfloat>, !dbg !44 + %4519 = fptrunc <2 x float> %4363 to <2 x bfloat>, !dbg !44 + %4520 = fptrunc <2 x float> %4368 to <2 x bfloat>, !dbg !44 + %4521 = fptrunc <2 x float> %4373 to <2 x bfloat>, !dbg !44 + %4522 = fptrunc <2 x float> %4378 to <2 x bfloat>, !dbg !44 + %4523 = fptrunc <2 x float> %4383 to <2 x bfloat>, !dbg !44 + %4524 = fptrunc <2 x float> %4388 to <2 x bfloat>, !dbg !44 + %4525 = insertelement <2 x bfloat> poison, bfloat %4389, i64 0, !dbg !45 + %4526 = insertelement <2 x bfloat> %4525, bfloat %4390, i64 1, !dbg !45 + %4527 = bitcast <2 x bfloat> %4526 to i32, !dbg !45 + %4528 = insertelement <2 x bfloat> poison, bfloat %4391, i64 0, !dbg !45 + %4529 = insertelement <2 x bfloat> %4528, bfloat %4392, i64 1, !dbg !45 + %4530 = bitcast <2 x bfloat> %4529 to i32, !dbg !45 + %4531 = insertelement <2 x bfloat> poison, bfloat %4393, i64 0, !dbg !45 + %4532 = insertelement <2 x bfloat> %4531, bfloat %4394, i64 1, !dbg !45 + %4533 = bitcast <2 x bfloat> %4532 to i32, !dbg !45 + %4534 = insertelement <2 x bfloat> poison, bfloat %4395, i64 0, !dbg !45 + %4535 = insertelement <2 x bfloat> %4534, bfloat %4396, i64 1, !dbg !45 + %4536 = bitcast <2 x bfloat> %4535 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4527, i32 %4530, i32 %4533, i32 %4536, ptr addrspace(1) %280, i1 %21) #5, !dbg !45 + %4537 = insertelement <2 x bfloat> poison, bfloat %4397, i64 0, !dbg !45 + %4538 = insertelement <2 x bfloat> %4537, bfloat %4398, i64 1, !dbg !45 + %4539 = bitcast <2 x bfloat> %4538 to i32, !dbg !45 + %4540 = insertelement <2 x bfloat> poison, bfloat %4399, i64 0, !dbg !45 + %4541 = insertelement <2 x bfloat> %4540, bfloat %4400, i64 1, !dbg !45 + %4542 = bitcast <2 x bfloat> %4541 to i32, !dbg !45 + %4543 = insertelement <2 x bfloat> poison, bfloat %4401, i64 0, !dbg !45 + %4544 = insertelement <2 x bfloat> %4543, bfloat %4402, i64 1, !dbg !45 + %4545 = bitcast <2 x bfloat> %4544 to i32, !dbg !45 + %4546 = insertelement <2 x bfloat> poison, bfloat %4403, i64 0, !dbg !45 + %4547 = insertelement <2 x bfloat> %4546, bfloat %4404, i64 1, !dbg !45 + %4548 = bitcast <2 x bfloat> %4547 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4539, i32 %4542, i32 %4545, i32 %4548, ptr addrspace(1) %281, i1 %22) #5, !dbg !45 + %4549 = bitcast <2 x bfloat> %4405 to i32, !dbg !45 + %4550 = bitcast <2 x bfloat> %4406 to i32, !dbg !45 + %4551 = bitcast <2 x bfloat> %4407 to i32, !dbg !45 + %4552 = bitcast <2 x bfloat> %4408 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4549, i32 %4550, i32 %4551, i32 %4552, ptr addrspace(1) %282, i1 %346) #5, !dbg !45 + %4553 = bitcast <2 x bfloat> %4409 to i32, !dbg !45 + %4554 = bitcast <2 x bfloat> %4410 to i32, !dbg !45 + %4555 = bitcast <2 x bfloat> %4411 to i32, !dbg !45 + %4556 = bitcast <2 x bfloat> %4412 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4553, i32 %4554, i32 %4555, i32 %4556, ptr addrspace(1) %283, i1 %364) #5, !dbg !45 + %4557 = bitcast <2 x bfloat> %4413 to i32, !dbg !45 + %4558 = bitcast <2 x bfloat> %4414 to i32, !dbg !45 + %4559 = bitcast <2 x bfloat> %4415 to i32, !dbg !45 + %4560 = bitcast <2 x bfloat> %4416 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4557, i32 %4558, i32 %4559, i32 %4560, ptr addrspace(1) %284, i1 %382) #5, !dbg !45 + %4561 = bitcast <2 x bfloat> %4417 to i32, !dbg !45 + %4562 = bitcast <2 x bfloat> %4418 to i32, !dbg !45 + %4563 = bitcast <2 x bfloat> %4419 to i32, !dbg !45 + %4564 = bitcast <2 x bfloat> %4420 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4561, i32 %4562, i32 %4563, i32 %4564, ptr addrspace(1) %285, i1 %400) #5, !dbg !45 + %4565 = bitcast <2 x bfloat> %4421 to i32, !dbg !45 + %4566 = bitcast <2 x bfloat> %4422 to i32, !dbg !45 + %4567 = bitcast <2 x bfloat> %4423 to i32, !dbg !45 + %4568 = bitcast <2 x bfloat> %4424 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4565, i32 %4566, i32 %4567, i32 %4568, ptr addrspace(1) %286, i1 %418) #5, !dbg !45 + %4569 = bitcast <2 x bfloat> %4425 to i32, !dbg !45 + %4570 = bitcast <2 x bfloat> %4426 to i32, !dbg !45 + %4571 = bitcast <2 x bfloat> %4427 to i32, !dbg !45 + %4572 = bitcast <2 x bfloat> %4428 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4569, i32 %4570, i32 %4571, i32 %4572, ptr addrspace(1) %287, i1 %436) #5, !dbg !45 + %4573 = bitcast <2 x bfloat> %4429 to i32, !dbg !45 + %4574 = bitcast <2 x bfloat> %4430 to i32, !dbg !45 + %4575 = bitcast <2 x bfloat> %4431 to i32, !dbg !45 + %4576 = bitcast <2 x bfloat> %4432 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4573, i32 %4574, i32 %4575, i32 %4576, ptr addrspace(1) %288, i1 %454) #5, !dbg !45 + %4577 = bitcast <2 x bfloat> %4433 to i32, !dbg !45 + %4578 = bitcast <2 x bfloat> %4434 to i32, !dbg !45 + %4579 = bitcast <2 x bfloat> %4435 to i32, !dbg !45 + %4580 = bitcast <2 x bfloat> %4436 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4577, i32 %4578, i32 %4579, i32 %4580, ptr addrspace(1) %289, i1 %472) #5, !dbg !45 + %4581 = bitcast <2 x bfloat> %4437 to i32, !dbg !45 + %4582 = bitcast <2 x bfloat> %4438 to i32, !dbg !45 + %4583 = bitcast <2 x bfloat> %4439 to i32, !dbg !45 + %4584 = bitcast <2 x bfloat> %4440 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4581, i32 %4582, i32 %4583, i32 %4584, ptr addrspace(1) %290, i1 %490) #5, !dbg !45 + %4585 = bitcast <2 x bfloat> %4441 to i32, !dbg !45 + %4586 = bitcast <2 x bfloat> %4442 to i32, !dbg !45 + %4587 = bitcast <2 x bfloat> %4443 to i32, !dbg !45 + %4588 = bitcast <2 x bfloat> %4444 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4585, i32 %4586, i32 %4587, i32 %4588, ptr addrspace(1) %291, i1 %508) #5, !dbg !45 + %4589 = bitcast <2 x bfloat> %4445 to i32, !dbg !45 + %4590 = bitcast <2 x bfloat> %4446 to i32, !dbg !45 + %4591 = bitcast <2 x bfloat> %4447 to i32, !dbg !45 + %4592 = bitcast <2 x bfloat> %4448 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4589, i32 %4590, i32 %4591, i32 %4592, ptr addrspace(1) %292, i1 %526) #5, !dbg !45 + %4593 = bitcast <2 x bfloat> %4449 to i32, !dbg !45 + %4594 = bitcast <2 x bfloat> %4450 to i32, !dbg !45 + %4595 = bitcast <2 x bfloat> %4451 to i32, !dbg !45 + %4596 = bitcast <2 x bfloat> %4452 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4593, i32 %4594, i32 %4595, i32 %4596, ptr addrspace(1) %293, i1 %544) #5, !dbg !45 + %4597 = bitcast <2 x bfloat> %4453 to i32, !dbg !45 + %4598 = bitcast <2 x bfloat> %4454 to i32, !dbg !45 + %4599 = bitcast <2 x bfloat> %4455 to i32, !dbg !45 + %4600 = bitcast <2 x bfloat> %4456 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4597, i32 %4598, i32 %4599, i32 %4600, ptr addrspace(1) %294, i1 %562) #5, !dbg !45 + %4601 = bitcast <2 x bfloat> %4457 to i32, !dbg !45 + %4602 = bitcast <2 x bfloat> %4458 to i32, !dbg !45 + %4603 = bitcast <2 x bfloat> %4459 to i32, !dbg !45 + %4604 = bitcast <2 x bfloat> %4460 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4601, i32 %4602, i32 %4603, i32 %4604, ptr addrspace(1) %295, i1 %580) #5, !dbg !45 + %4605 = bitcast <2 x bfloat> %4461 to i32, !dbg !45 + %4606 = bitcast <2 x bfloat> %4462 to i32, !dbg !45 + %4607 = bitcast <2 x bfloat> %4463 to i32, !dbg !45 + %4608 = bitcast <2 x bfloat> %4464 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4605, i32 %4606, i32 %4607, i32 %4608, ptr addrspace(1) %296, i1 %598) #5, !dbg !45 + %4609 = bitcast <2 x bfloat> %4465 to i32, !dbg !45 + %4610 = bitcast <2 x bfloat> %4466 to i32, !dbg !45 + %4611 = bitcast <2 x bfloat> %4467 to i32, !dbg !45 + %4612 = bitcast <2 x bfloat> %4468 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4609, i32 %4610, i32 %4611, i32 %4612, ptr addrspace(1) %297, i1 %616) #5, !dbg !45 + %4613 = bitcast <2 x bfloat> %4469 to i32, !dbg !45 + %4614 = bitcast <2 x bfloat> %4470 to i32, !dbg !45 + %4615 = bitcast <2 x bfloat> %4471 to i32, !dbg !45 + %4616 = bitcast <2 x bfloat> %4472 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4613, i32 %4614, i32 %4615, i32 %4616, ptr addrspace(1) %298, i1 %634) #5, !dbg !45 + %4617 = bitcast <2 x bfloat> %4473 to i32, !dbg !45 + %4618 = bitcast <2 x bfloat> %4474 to i32, !dbg !45 + %4619 = bitcast <2 x bfloat> %4475 to i32, !dbg !45 + %4620 = bitcast <2 x bfloat> %4476 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4617, i32 %4618, i32 %4619, i32 %4620, ptr addrspace(1) %299, i1 %652) #5, !dbg !45 + %4621 = bitcast <2 x bfloat> %4477 to i32, !dbg !45 + %4622 = bitcast <2 x bfloat> %4478 to i32, !dbg !45 + %4623 = bitcast <2 x bfloat> %4479 to i32, !dbg !45 + %4624 = bitcast <2 x bfloat> %4480 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4621, i32 %4622, i32 %4623, i32 %4624, ptr addrspace(1) %300, i1 %670) #5, !dbg !45 + %4625 = bitcast <2 x bfloat> %4481 to i32, !dbg !45 + %4626 = bitcast <2 x bfloat> %4482 to i32, !dbg !45 + %4627 = bitcast <2 x bfloat> %4483 to i32, !dbg !45 + %4628 = bitcast <2 x bfloat> %4484 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4625, i32 %4626, i32 %4627, i32 %4628, ptr addrspace(1) %301, i1 %688) #5, !dbg !45 + %4629 = bitcast <2 x bfloat> %4485 to i32, !dbg !45 + %4630 = bitcast <2 x bfloat> %4486 to i32, !dbg !45 + %4631 = bitcast <2 x bfloat> %4487 to i32, !dbg !45 + %4632 = bitcast <2 x bfloat> %4488 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4629, i32 %4630, i32 %4631, i32 %4632, ptr addrspace(1) %302, i1 %706) #5, !dbg !45 + %4633 = bitcast <2 x bfloat> %4489 to i32, !dbg !45 + %4634 = bitcast <2 x bfloat> %4490 to i32, !dbg !45 + %4635 = bitcast <2 x bfloat> %4491 to i32, !dbg !45 + %4636 = bitcast <2 x bfloat> %4492 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4633, i32 %4634, i32 %4635, i32 %4636, ptr addrspace(1) %303, i1 %724) #5, !dbg !45 + %4637 = bitcast <2 x bfloat> %4493 to i32, !dbg !45 + %4638 = bitcast <2 x bfloat> %4494 to i32, !dbg !45 + %4639 = bitcast <2 x bfloat> %4495 to i32, !dbg !45 + %4640 = bitcast <2 x bfloat> %4496 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4637, i32 %4638, i32 %4639, i32 %4640, ptr addrspace(1) %304, i1 %742) #5, !dbg !45 + %4641 = bitcast <2 x bfloat> %4497 to i32, !dbg !45 + %4642 = bitcast <2 x bfloat> %4498 to i32, !dbg !45 + %4643 = bitcast <2 x bfloat> %4499 to i32, !dbg !45 + %4644 = bitcast <2 x bfloat> %4500 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4641, i32 %4642, i32 %4643, i32 %4644, ptr addrspace(1) %305, i1 %760) #5, !dbg !45 + %4645 = bitcast <2 x bfloat> %4501 to i32, !dbg !45 + %4646 = bitcast <2 x bfloat> %4502 to i32, !dbg !45 + %4647 = bitcast <2 x bfloat> %4503 to i32, !dbg !45 + %4648 = bitcast <2 x bfloat> %4504 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4645, i32 %4646, i32 %4647, i32 %4648, ptr addrspace(1) %306, i1 %778) #5, !dbg !45 + %4649 = bitcast <2 x bfloat> %4505 to i32, !dbg !45 + %4650 = bitcast <2 x bfloat> %4506 to i32, !dbg !45 + %4651 = bitcast <2 x bfloat> %4507 to i32, !dbg !45 + %4652 = bitcast <2 x bfloat> %4508 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4649, i32 %4650, i32 %4651, i32 %4652, ptr addrspace(1) %307, i1 %796) #5, !dbg !45 + %4653 = bitcast <2 x bfloat> %4509 to i32, !dbg !45 + %4654 = bitcast <2 x bfloat> %4510 to i32, !dbg !45 + %4655 = bitcast <2 x bfloat> %4511 to i32, !dbg !45 + %4656 = bitcast <2 x bfloat> %4512 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4653, i32 %4654, i32 %4655, i32 %4656, ptr addrspace(1) %308, i1 %814) #5, !dbg !45 + %4657 = bitcast <2 x bfloat> %4513 to i32, !dbg !45 + %4658 = bitcast <2 x bfloat> %4514 to i32, !dbg !45 + %4659 = bitcast <2 x bfloat> %4515 to i32, !dbg !45 + %4660 = bitcast <2 x bfloat> %4516 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4657, i32 %4658, i32 %4659, i32 %4660, ptr addrspace(1) %309, i1 %832) #5, !dbg !45 + %4661 = bitcast <2 x bfloat> %4517 to i32, !dbg !45 + %4662 = bitcast <2 x bfloat> %4518 to i32, !dbg !45 + %4663 = bitcast <2 x bfloat> %4519 to i32, !dbg !45 + %4664 = bitcast <2 x bfloat> %4520 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4661, i32 %4662, i32 %4663, i32 %4664, ptr addrspace(1) %310, i1 %850) #5, !dbg !45 + %4665 = bitcast <2 x bfloat> %4521 to i32, !dbg !45 + %4666 = bitcast <2 x bfloat> %4522 to i32, !dbg !45 + %4667 = bitcast <2 x bfloat> %4523 to i32, !dbg !45 + %4668 = bitcast <2 x bfloat> %4524 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4665, i32 %4666, i32 %4667, i32 %4668, ptr addrspace(1) %311, i1 %868) #5, !dbg !45 + ret void, !dbg !46 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(argmem: write) +declare void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x2.b16.p3(ptr addrspace(3) writeonly captures(none), i32, i32) #2 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.bar.warp.sync(i32) #3 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #4 + +attributes #0 = { nounwind "nvvm.reqntid"="32" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(argmem: write) } +attributes #3 = { convergent nocallback nounwind } +attributes #4 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #5 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = distinct !DISubprogram(name: "l2norm_bwd_kernel", linkageName: "l2norm_bwd_kernel", scope: !1, file: !1, line: 93, type: !5, scopeLine: 93, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!5 = !DISubroutineType(cc: DW_CC_normal, types: !6) +!6 = !{null, !7, !9, !7, !7, !10, !11, !7, !7} +!7 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !8, size: 64, dwarfAddressSpace: 1) +!8 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!9 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !10, size: 64, dwarfAddressSpace: 1) +!10 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!11 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!12 = !DILocation(line: 94, column: 24, scope: !4) +!13 = !DILocation(line: 94, column: 30, scope: !4) +!14 = !DILocation(line: 95, column: 16, scope: !4) +!15 = !DILocation(line: 95, column: 34, scope: !4) +!16 = !DILocation(line: 95, column: 21, scope: !4) +!17 = !DILocation(line: 97, column: 16, scope: !4) +!18 = !DILocation(line: 98, column: 30, scope: !4) +!19 = !DILocation(line: 99, column: 29, scope: !4) +!20 = !DILocation(line: 99, column: 14, scope: !4) +!21 = !DILocation(line: 99, column: 33, scope: !4) +!22 = !DILocation(line: 100, column: 20, scope: !4) +!23 = !DILocation(line: 101, column: 16, scope: !4) +!24 = !DILocation(line: 101, column: 35, scope: !4) +!25 = !DILocation(line: 102, column: 16, scope: !4) +!26 = !DILocation(line: 102, column: 35, scope: !4) +!27 = !DILocation(line: 104, column: 18, scope: !4) +!28 = !DILocation(line: 104, column: 47, scope: !4) +!29 = !DILocation(line: 105, column: 21, scope: !4) +!30 = !DILocation(line: 106, column: 19, scope: !4) +!31 = !DILocation(line: 106, column: 49, scope: !4) +!32 = !DILocation(line: 107, column: 25, scope: !4) +!33 = !DILocation(line: 107, column: 18, scope: !4) +!34 = !DILocation(line: 107, column: 50, scope: !4) +!35 = !DILocation(line: 263, column: 15, scope: !36, inlinedAt: !38) +!36 = distinct !DILexicalBlockFile(scope: !4, file: !37, discriminator: 0) +!37 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!38 = !DILocation(line: 293, column: 36, scope: !36, inlinedAt: !39) +!39 = !DILocation(line: 107, column: 55, scope: !40) +!40 = distinct !DILexicalBlockFile(scope: !4, file: !1, discriminator: 0) +!41 = !DILocation(line: 107, column: 69, scope: !4) +!42 = !DILocation(line: 107, column: 75, scope: !4) +!43 = !DILocation(line: 107, column: 36, scope: !4) +!44 = !DILocation(line: 108, column: 27, scope: !4) +!45 = !DILocation(line: 108, column: 19, scope: !4) +!46 = !DILocation(line: 108, column: 4, scope: !4) diff --git a/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.ptx b/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..7fb24187cbfe19d4d64b56601333c36c000a1977 --- /dev/null +++ b/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.ptx @@ -0,0 +1,4518 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_bwd_kernel // -- Begin function l2norm_bwd_kernel +.extern .shared .align 16 .b8 global_smem[]; + // @l2norm_bwd_kernel +.visible .entry l2norm_bwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_2, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_3, + .param .f32 l2norm_bwd_kernel_param_4, + .param .u32 l2norm_bwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_6, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_7 +) +.reqntid 32 +{ + .reg .pred %p<50>; + .reg .b16 %rs<566>; + .reg .b32 %r<2544>; + .reg .b64 %rd<664>; + .loc 1 93 0 // l2norm.py:93:0 +$L__func_begin0: + .loc 1 93 0 // l2norm.py:93:0 + +// %bb.0: + ld.param.b64 %rd99, [l2norm_bwd_kernel_param_0]; + ld.param.b64 %rd100, [l2norm_bwd_kernel_param_1]; +$L__tmp0: + .loc 1 94 24 // l2norm.py:94:24 + mov.u32 %r388, %ctaid.x; + ld.param.b64 %rd101, [l2norm_bwd_kernel_param_2]; + .loc 1 95 16 // l2norm.py:95:16 + mul.wide.u32 %rd102, %r388, 64; + ld.param.b64 %rd103, [l2norm_bwd_kernel_param_3]; + .loc 1 95 34 // l2norm.py:95:34 + mov.u32 %r389, %tid.x; + and.b32 %r390, %r389, 16; + ld.param.s32 %rd104, [l2norm_bwd_kernel_param_5]; + bfe.u32 %r391, %r389, 4, 1; + or.b32 %r392, %r391, 2; + shl.b32 %r393, %r389, 1; + .loc 1 95 21 // l2norm.py:95:21 + cvt.u64.u32 %rd105, %r391; + cvt.u64.u32 %rd106, %r392; + or.b64 %rd107, %rd102, %rd105; + or.b64 %rd108, %rd102, %rd106; + .loc 1 97 16 // l2norm.py:97:16 + setp.lt.s64 %p1, %rd107, %rd104; + setp.lt.s64 %p2, %rd108, %rd104; + .loc 1 95 34 // l2norm.py:95:34 + and.b32 %r394, %r393, 62; + or.b32 %r395, %r394, 1; + .loc 1 95 21 // l2norm.py:95:21 + or.b32 %r396, %r391, 4; + or.b32 %r397, %r391, 6; + or.b32 %r398, %r391, 8; + or.b32 %r399, %r391, 10; + or.b32 %r400, %r391, 12; + or.b32 %r401, %r391, 14; + or.b32 %r402, %r391, 16; + or.b32 %r403, %r391, 18; + or.b32 %r404, %r391, 20; + or.b32 %r405, %r391, 22; + or.b32 %r406, %r391, 24; + or.b32 %r407, %r391, 26; + or.b32 %r408, %r391, 28; + or.b32 %r409, %r391, 30; + or.b32 %r410, %r391, 32; + or.b32 %r411, %r391, 34; + or.b32 %r412, %r391, 48; + or.b32 %r413, %r391, 50; + or.b32 %r414, %r391, 44; + or.b32 %r415, %r391, 46; + or.b32 %r416, %r391, 40; + or.b32 %r417, %r391, 42; + or.b32 %r418, %r391, 36; + or.b32 %r419, %r391, 38; + or.b32 %r420, %r391, 56; + or.b32 %r421, %r391, 58; + or.b32 %r422, %r391, 52; + or.b32 %r423, %r391, 54; + or.b32 %r424, %r391, 60; + or.b32 %r425, %r391, 62; + cvt.u64.u32 %rd109, %r419; + cvt.u64.u32 %rd110, %r418; + cvt.u64.u32 %rd111, %r417; + cvt.u64.u32 %rd112, %r416; + cvt.u64.u32 %rd113, %r415; + cvt.u64.u32 %rd114, %r414; + cvt.u64.u32 %rd115, %r413; + cvt.u64.u32 %rd116, %r412; + cvt.u64.u32 %rd117, %r423; + cvt.u64.u32 %rd118, %r422; + cvt.u64.u32 %rd119, %r421; + cvt.u64.u32 %rd120, %r420; + cvt.u64.u32 %rd121, %r425; + cvt.u64.u32 %rd122, %r424; + cvt.u64.u32 %rd123, %r394; + cvt.u64.u32 %rd124, %r395; + cvt.u64.u32 %rd125, %r411; + cvt.u64.u32 %rd126, %r410; + cvt.u64.u32 %rd127, %r409; + cvt.u64.u32 %rd128, %r408; + cvt.u64.u32 %rd129, %r407; + cvt.u64.u32 %rd130, %r406; + cvt.u64.u32 %rd131, %r405; + cvt.u64.u32 %rd132, %r404; + cvt.u64.u32 %rd133, %r403; + cvt.u64.u32 %rd134, %r402; + cvt.u64.u32 %rd135, %r401; + cvt.u64.u32 %rd136, %r400; + cvt.u64.u32 %rd137, %r399; + cvt.u64.u32 %rd138, %r398; + cvt.u64.u32 %rd139, %r397; + cvt.u64.u32 %rd140, %r396; + or.b64 %rd141, %rd102, %rd140; + or.b64 %rd142, %rd102, %rd139; + or.b64 %rd143, %rd102, %rd138; + or.b64 %rd144, %rd102, %rd137; + or.b64 %rd145, %rd102, %rd136; + or.b64 %rd146, %rd102, %rd135; + or.b64 %rd147, %rd102, %rd134; + or.b64 %rd148, %rd102, %rd133; + or.b64 %rd149, %rd102, %rd132; + or.b64 %rd150, %rd102, %rd131; + or.b64 %rd151, %rd102, %rd130; + or.b64 %rd152, %rd102, %rd129; + or.b64 %rd153, %rd102, %rd128; + or.b64 %rd154, %rd102, %rd127; + or.b64 %rd155, %rd102, %rd126; + or.b64 %rd156, %rd102, %rd125; + or.b64 %rd157, %rd102, %rd124; + or.b64 %rd158, %rd102, %rd123; + or.b64 %rd159, %rd102, %rd122; + or.b64 %rd160, %rd102, %rd121; + or.b64 %rd161, %rd102, %rd120; + or.b64 %rd162, %rd102, %rd119; + or.b64 %rd163, %rd102, %rd118; + or.b64 %rd164, %rd102, %rd117; + or.b64 %rd165, %rd102, %rd116; + or.b64 %rd166, %rd102, %rd115; + or.b64 %rd167, %rd102, %rd114; + or.b64 %rd168, %rd102, %rd113; + or.b64 %rd169, %rd102, %rd112; + or.b64 %rd170, %rd102, %rd111; + or.b64 %rd171, %rd102, %rd110; + or.b64 %rd172, %rd102, %rd109; + .loc 1 97 16 // l2norm.py:97:16 + setp.lt.s64 %p35, %rd172, %rd104; + selp.b16 %rs1, 1, 0, %p35; + shl.b16 %rs2, %rs1, 2; + setp.lt.s64 %p36, %rd171, %rd104; + selp.b16 %rs3, -1, 0, %p36; + shl.b16 %rs4, %rs3, 3; + or.b16 %rs5, %rs4, %rs2; + setp.lt.s64 %p37, %rd170, %rd104; + selp.b16 %rs6, 1, 0, %p37; + setp.lt.s64 %p38, %rd169, %rd104; + selp.b16 %rs7, -1, 0, %p38; + shl.b16 %rs8, %rs7, 1; + or.b16 %rs9, %rs6, %rs8; + and.b16 %rs10, %rs9, 3; + or.b16 %rs11, %rs10, %rs5; + shl.b16 %rs12, %rs11, 12; + setp.lt.s64 %p39, %rd168, %rd104; + selp.b16 %rs13, 1, 0, %p39; + shl.b16 %rs14, %rs13, 2; + setp.lt.s64 %p40, %rd167, %rd104; + selp.b16 %rs15, -1, 0, %p40; + shl.b16 %rs16, %rs15, 3; + or.b16 %rs17, %rs16, %rs14; + setp.lt.s64 %p41, %rd166, %rd104; + selp.b16 %rs18, 1, 0, %p41; + setp.lt.s64 %p42, %rd165, %rd104; + selp.b16 %rs19, -1, 0, %p42; + shl.b16 %rs20, %rs19, 1; + or.b16 %rs21, %rs18, %rs20; + and.b16 %rs22, %rs21, 3; + or.b16 %rs23, %rs22, %rs17; + and.b16 %rs24, %rs23, 15; + shl.b16 %rs25, %rs24, 8; + or.b16 %rs26, %rs12, %rs25; + setp.lt.s64 %p43, %rd164, %rd104; + selp.b16 %rs27, 1, 0, %p43; + shl.b16 %rs28, %rs27, 2; + setp.lt.s64 %p44, %rd163, %rd104; + selp.b16 %rs29, -1, 0, %p44; + shl.b16 %rs30, %rs29, 3; + or.b16 %rs31, %rs30, %rs28; + setp.lt.s64 %p45, %rd162, %rd104; + selp.b16 %rs32, 1, 0, %p45; + setp.lt.s64 %p46, %rd161, %rd104; + selp.b16 %rs33, -1, 0, %p46; + shl.b16 %rs34, %rs33, 1; + or.b16 %rs35, %rs32, %rs34; + and.b16 %rs36, %rs35, 3; + or.b16 %rs37, %rs36, %rs31; + shl.b16 %rs38, %rs37, 4; + setp.lt.s64 %p47, %rd160, %rd104; + selp.b16 %rs39, 1, 0, %p47; + shl.b16 %rs40, %rs39, 2; + setp.lt.s64 %p48, %rd159, %rd104; + selp.b16 %rs41, -1, 0, %p48; + shl.b16 %rs42, %rs41, 3; + or.b16 %rs43, %rs42, %rs40; + setp.lt.s64 %p49, %rd158, %rd104; + selp.b16 %rs44, -1, 0, %p49; + shl.b16 %rs45, %rs44, 1; + setp.lt.s64 %p5, %rd157, %rd104; + selp.b16 %rs46, 1, 0, %p5; + or.b16 %rs47, %rs46, %rs45; + and.b16 %rs48, %rs47, 3; + or.b16 %rs49, %rs48, %rs43; + and.b16 %rs50, %rs49, 15; + or.b16 %rs51, %rs50, %rs38; + and.b16 %rs52, %rs51, 255; + or.b16 %rs53, %rs52, %rs26; + cvt.u32.u16 %r426, %rs53; + setp.lt.s64 %p20, %rd156, %rd104; + setp.lt.s64 %p19, %rd155, %rd104; + setp.lt.s64 %p18, %rd154, %rd104; + setp.lt.s64 %p17, %rd153, %rd104; + setp.lt.s64 %p16, %rd152, %rd104; + setp.lt.s64 %p15, %rd151, %rd104; + setp.lt.s64 %p14, %rd150, %rd104; + setp.lt.s64 %p13, %rd149, %rd104; + setp.lt.s64 %p3, %rd148, %rd104; + setp.lt.s64 %p12, %rd147, %rd104; + setp.lt.s64 %p11, %rd146, %rd104; + setp.lt.s64 %p10, %rd145, %rd104; + setp.lt.s64 %p9, %rd144, %rd104; + setp.lt.s64 %p8, %rd143, %rd104; + setp.lt.s64 %p7, %rd142, %rd104; + setp.lt.s64 %p6, %rd141, %rd104; + .loc 1 98 30 // l2norm.py:98:30 + and.b32 %r427, %r389, 15; + shl.b32 %r428, %r427, 3; + .loc 1 99 14 // l2norm.py:99:14 + shl.b64 %rd173, %rd107, 8; + add.s64 %rd174, %rd99, %rd173; + shl.b64 %rd175, %rd108, 8; + add.s64 %rd176, %rd99, %rd175; + shl.b64 %rd177, %rd141, 8; + add.s64 %rd178, %rd99, %rd177; + shl.b64 %rd179, %rd142, 8; + add.s64 %rd180, %rd99, %rd179; + shl.b64 %rd181, %rd143, 8; + add.s64 %rd182, %rd99, %rd181; + shl.b64 %rd183, %rd144, 8; + add.s64 %rd184, %rd99, %rd183; + shl.b64 %rd185, %rd145, 8; + add.s64 %rd186, %rd99, %rd185; + shl.b64 %rd187, %rd146, 8; + add.s64 %rd188, %rd99, %rd187; + shl.b64 %rd189, %rd147, 8; + add.s64 %rd190, %rd99, %rd189; + shl.b64 %rd191, %rd148, 8; + add.s64 %rd192, %rd99, %rd191; + shl.b64 %rd193, %rd149, 8; + add.s64 %rd194, %rd99, %rd193; + shl.b64 %rd195, %rd150, 8; + add.s64 %rd196, %rd99, %rd195; + shl.b64 %rd197, %rd151, 8; + add.s64 %rd198, %rd99, %rd197; + shl.b64 %rd199, %rd152, 8; + add.s64 %rd200, %rd99, %rd199; + shl.b64 %rd201, %rd153, 8; + add.s64 %rd202, %rd99, %rd201; + shl.b64 %rd203, %rd154, 8; + add.s64 %rd204, %rd99, %rd203; + shl.b64 %rd205, %rd155, 8; + add.s64 %rd206, %rd99, %rd205; + shl.b64 %rd207, %rd156, 8; + add.s64 %rd208, %rd99, %rd207; + shl.b64 %rd209, %rd171, 8; + add.s64 %rd210, %rd99, %rd209; + shl.b64 %rd211, %rd172, 8; + add.s64 %rd212, %rd99, %rd211; + shl.b64 %rd213, %rd169, 8; + add.s64 %rd214, %rd99, %rd213; + shl.b64 %rd215, %rd170, 8; + add.s64 %rd216, %rd99, %rd215; + shl.b64 %rd217, %rd167, 8; + add.s64 %rd218, %rd99, %rd217; + shl.b64 %rd219, %rd168, 8; + add.s64 %rd220, %rd99, %rd219; + shl.b64 %rd221, %rd165, 8; + add.s64 %rd222, %rd99, %rd221; + shl.b64 %rd223, %rd166, 8; + add.s64 %rd224, %rd99, %rd223; + shl.b64 %rd225, %rd163, 8; + add.s64 %rd226, %rd99, %rd225; + shl.b64 %rd227, %rd164, 8; + add.s64 %rd228, %rd99, %rd227; + shl.b64 %rd229, %rd161, 8; + add.s64 %rd230, %rd99, %rd229; + shl.b64 %rd231, %rd162, 8; + add.s64 %rd232, %rd99, %rd231; + shl.b64 %rd233, %rd159, 8; + add.s64 %rd234, %rd99, %rd233; + shl.b64 %rd235, %rd160, 8; + add.s64 %rd236, %rd99, %rd235; + .loc 1 99 33 // l2norm.py:99:33 + mul.wide.u32 %rd237, %r428, 2; + add.s64 %rd1, %rd174, %rd237; + add.s64 %rd2, %rd176, %rd237; + add.s64 %rd3, %rd178, %rd237; + add.s64 %rd4, %rd180, %rd237; + add.s64 %rd5, %rd182, %rd237; + add.s64 %rd6, %rd184, %rd237; + add.s64 %rd7, %rd186, %rd237; + add.s64 %rd8, %rd188, %rd237; + add.s64 %rd9, %rd190, %rd237; + add.s64 %rd10, %rd192, %rd237; + add.s64 %rd11, %rd194, %rd237; + add.s64 %rd12, %rd196, %rd237; + add.s64 %rd13, %rd198, %rd237; + add.s64 %rd14, %rd200, %rd237; + add.s64 %rd15, %rd202, %rd237; + add.s64 %rd16, %rd204, %rd237; + add.s64 %rd17, %rd206, %rd237; + add.s64 %rd18, %rd208, %rd237; + add.s64 %rd19, %rd210, %rd237; + add.s64 %rd20, %rd212, %rd237; + add.s64 %rd21, %rd214, %rd237; + add.s64 %rd22, %rd216, %rd237; + add.s64 %rd23, %rd218, %rd237; + add.s64 %rd24, %rd220, %rd237; + add.s64 %rd25, %rd222, %rd237; + add.s64 %rd26, %rd224, %rd237; + add.s64 %rd27, %rd226, %rd237; + add.s64 %rd28, %rd228, %rd237; + add.s64 %rd29, %rd230, %rd237; + add.s64 %rd30, %rd232, %rd237; + add.s64 %rd31, %rd234, %rd237; + add.s64 %rd32, %rd236, %rd237; + .loc 1 100 20 // l2norm.py:100:20 + shl.b64 %rd238, %rd158, 2; + add.s64 %rd33, %rd100, %rd238; + shl.b64 %rd239, %rd157, 2; + add.s64 %rd34, %rd100, %rd239; + .loc 1 101 16 // l2norm.py:101:16 + add.s64 %rd240, %rd101, %rd173; + add.s64 %rd241, %rd101, %rd175; + add.s64 %rd242, %rd101, %rd177; + add.s64 %rd243, %rd101, %rd179; + add.s64 %rd244, %rd101, %rd181; + add.s64 %rd245, %rd101, %rd183; + add.s64 %rd246, %rd101, %rd185; + add.s64 %rd247, %rd101, %rd187; + add.s64 %rd248, %rd101, %rd189; + add.s64 %rd249, %rd101, %rd191; + add.s64 %rd250, %rd101, %rd193; + add.s64 %rd251, %rd101, %rd195; + add.s64 %rd252, %rd101, %rd197; + add.s64 %rd253, %rd101, %rd199; + add.s64 %rd254, %rd101, %rd201; + add.s64 %rd255, %rd101, %rd203; + add.s64 %rd256, %rd101, %rd205; + add.s64 %rd257, %rd101, %rd207; + add.s64 %rd258, %rd101, %rd209; + add.s64 %rd259, %rd101, %rd211; + add.s64 %rd260, %rd101, %rd213; + add.s64 %rd261, %rd101, %rd215; + add.s64 %rd262, %rd101, %rd217; + add.s64 %rd263, %rd101, %rd219; + add.s64 %rd264, %rd101, %rd221; + add.s64 %rd265, %rd101, %rd223; + add.s64 %rd266, %rd101, %rd225; + add.s64 %rd267, %rd101, %rd227; + add.s64 %rd268, %rd101, %rd229; + add.s64 %rd269, %rd101, %rd231; + add.s64 %rd270, %rd101, %rd233; + add.s64 %rd271, %rd101, %rd235; + .loc 1 101 35 // l2norm.py:101:35 + add.s64 %rd35, %rd240, %rd237; + add.s64 %rd36, %rd241, %rd237; + add.s64 %rd37, %rd242, %rd237; + add.s64 %rd38, %rd243, %rd237; + add.s64 %rd39, %rd244, %rd237; + add.s64 %rd40, %rd245, %rd237; + add.s64 %rd41, %rd246, %rd237; + add.s64 %rd42, %rd247, %rd237; + add.s64 %rd43, %rd248, %rd237; + add.s64 %rd44, %rd249, %rd237; + add.s64 %rd45, %rd250, %rd237; + add.s64 %rd46, %rd251, %rd237; + add.s64 %rd47, %rd252, %rd237; + add.s64 %rd48, %rd253, %rd237; + add.s64 %rd49, %rd254, %rd237; + add.s64 %rd50, %rd255, %rd237; + add.s64 %rd51, %rd256, %rd237; + add.s64 %rd52, %rd257, %rd237; + add.s64 %rd53, %rd258, %rd237; + add.s64 %rd54, %rd259, %rd237; + add.s64 %rd55, %rd260, %rd237; + add.s64 %rd56, %rd261, %rd237; + add.s64 %rd57, %rd262, %rd237; + add.s64 %rd58, %rd263, %rd237; + add.s64 %rd59, %rd264, %rd237; + add.s64 %rd60, %rd265, %rd237; + add.s64 %rd61, %rd266, %rd237; + add.s64 %rd62, %rd267, %rd237; + add.s64 %rd63, %rd268, %rd237; + add.s64 %rd64, %rd269, %rd237; + add.s64 %rd65, %rd270, %rd237; + add.s64 %rd66, %rd271, %rd237; + .loc 1 102 16 // l2norm.py:102:16 + add.s64 %rd272, %rd103, %rd173; + add.s64 %rd273, %rd103, %rd175; + add.s64 %rd274, %rd103, %rd177; + add.s64 %rd275, %rd103, %rd179; + add.s64 %rd276, %rd103, %rd181; + add.s64 %rd277, %rd103, %rd183; + add.s64 %rd278, %rd103, %rd185; + add.s64 %rd279, %rd103, %rd187; + add.s64 %rd280, %rd103, %rd189; + add.s64 %rd281, %rd103, %rd191; + add.s64 %rd282, %rd103, %rd193; + add.s64 %rd283, %rd103, %rd195; + add.s64 %rd284, %rd103, %rd197; + add.s64 %rd285, %rd103, %rd199; + add.s64 %rd286, %rd103, %rd201; + add.s64 %rd287, %rd103, %rd203; + add.s64 %rd288, %rd103, %rd205; + add.s64 %rd289, %rd103, %rd207; + add.s64 %rd290, %rd103, %rd209; + add.s64 %rd291, %rd103, %rd211; + add.s64 %rd292, %rd103, %rd213; + add.s64 %rd293, %rd103, %rd215; + add.s64 %rd294, %rd103, %rd217; + add.s64 %rd295, %rd103, %rd219; + add.s64 %rd296, %rd103, %rd221; + add.s64 %rd297, %rd103, %rd223; + add.s64 %rd298, %rd103, %rd225; + add.s64 %rd299, %rd103, %rd227; + add.s64 %rd300, %rd103, %rd229; + add.s64 %rd301, %rd103, %rd231; + add.s64 %rd302, %rd103, %rd233; + add.s64 %rd303, %rd103, %rd235; + .loc 1 102 35 // l2norm.py:102:35 + add.s64 %rd67, %rd272, %rd237; + add.s64 %rd68, %rd273, %rd237; + add.s64 %rd69, %rd274, %rd237; + add.s64 %rd70, %rd275, %rd237; + add.s64 %rd71, %rd276, %rd237; + add.s64 %rd72, %rd277, %rd237; + add.s64 %rd73, %rd278, %rd237; + add.s64 %rd74, %rd279, %rd237; + add.s64 %rd75, %rd280, %rd237; + add.s64 %rd76, %rd281, %rd237; + add.s64 %rd77, %rd282, %rd237; + add.s64 %rd78, %rd283, %rd237; + add.s64 %rd79, %rd284, %rd237; + add.s64 %rd80, %rd285, %rd237; + add.s64 %rd81, %rd286, %rd237; + add.s64 %rd82, %rd287, %rd237; + add.s64 %rd83, %rd288, %rd237; + add.s64 %rd84, %rd289, %rd237; + add.s64 %rd85, %rd290, %rd237; + add.s64 %rd86, %rd291, %rd237; + add.s64 %rd87, %rd292, %rd237; + add.s64 %rd88, %rd293, %rd237; + add.s64 %rd89, %rd294, %rd237; + add.s64 %rd90, %rd295, %rd237; + add.s64 %rd91, %rd296, %rd237; + add.s64 %rd92, %rd297, %rd237; + add.s64 %rd93, %rd298, %rd237; + add.s64 %rd94, %rd299, %rd237; + add.s64 %rd95, %rd300, %rd237; + add.s64 %rd96, %rd301, %rd237; + add.s64 %rd97, %rd302, %rd237; + add.s64 %rd98, %rd303, %rd237; + mov.b32 %r5, 0; + .loc 1 104 18 // l2norm.py:104:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs54, %rs55}, %r1; + mov.b32 {%rs56, %rs57}, %r2; + mov.b32 {%rs58, %rs59}, %r3; + mov.b32 {%rs60, %rs61}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs62, %rs63}, %r6; + mov.b32 {%rs64, %rs65}, %r7; + mov.b32 {%rs66, %rs67}, %r8; + mov.b32 {%rs68, %rs69}, %r9; + // begin inline asm + mov.u32 %r10, %r5; + mov.u32 %r11, %r5; + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + @%p6 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs70, %rs71}, %r10; + mov.b32 {%rs72, %rs73}, %r11; + mov.b32 {%rs74, %rs75}, %r12; + mov.b32 {%rs76, %rs77}, %r13; + // begin inline asm + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + @%p7 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ]; + // end inline asm + mov.b32 {%rs78, %rs79}, %r14; + mov.b32 {%rs80, %rs81}, %r15; + mov.b32 {%rs82, %rs83}, %r16; + mov.b32 {%rs84, %rs85}, %r17; + // begin inline asm + mov.u32 %r18, %r5; + mov.u32 %r19, %r5; + mov.u32 %r20, %r5; + mov.u32 %r21, %r5; + @%p8 ld.global.v4.b32 { %r18, %r19, %r20, %r21 }, [ %rd5 + 0 ]; + // end inline asm + mov.b32 {%rs86, %rs87}, %r18; + mov.b32 {%rs88, %rs89}, %r19; + mov.b32 {%rs90, %rs91}, %r20; + mov.b32 {%rs92, %rs93}, %r21; + // begin inline asm + mov.u32 %r22, %r5; + mov.u32 %r23, %r5; + mov.u32 %r24, %r5; + mov.u32 %r25, %r5; + @%p9 ld.global.v4.b32 { %r22, %r23, %r24, %r25 }, [ %rd6 + 0 ]; + // end inline asm + mov.b32 {%rs94, %rs95}, %r22; + mov.b32 {%rs96, %rs97}, %r23; + mov.b32 {%rs98, %rs99}, %r24; + mov.b32 {%rs100, %rs101}, %r25; + // begin inline asm + mov.u32 %r26, %r5; + mov.u32 %r27, %r5; + mov.u32 %r28, %r5; + mov.u32 %r29, %r5; + @%p10 ld.global.v4.b32 { %r26, %r27, %r28, %r29 }, [ %rd7 + 0 ]; + // end inline asm + mov.b32 {%rs102, %rs103}, %r26; + mov.b32 {%rs104, %rs105}, %r27; + mov.b32 {%rs106, %rs107}, %r28; + mov.b32 {%rs108, %rs109}, %r29; + // begin inline asm + mov.u32 %r30, %r5; + mov.u32 %r31, %r5; + mov.u32 %r32, %r5; + mov.u32 %r33, %r5; + @%p11 ld.global.v4.b32 { %r30, %r31, %r32, %r33 }, [ %rd8 + 0 ]; + // end inline asm + mov.b32 {%rs110, %rs111}, %r30; + mov.b32 {%rs112, %rs113}, %r31; + mov.b32 {%rs114, %rs115}, %r32; + mov.b32 {%rs116, %rs117}, %r33; + // begin inline asm + mov.u32 %r34, %r5; + mov.u32 %r35, %r5; + mov.u32 %r36, %r5; + mov.u32 %r37, %r5; + @%p12 ld.global.v4.b32 { %r34, %r35, %r36, %r37 }, [ %rd9 + 0 ]; + // end inline asm + mov.b32 {%rs118, %rs119}, %r34; + mov.b32 {%rs120, %rs121}, %r35; + mov.b32 {%rs122, %rs123}, %r36; + mov.b32 {%rs124, %rs125}, %r37; + // begin inline asm + mov.u32 %r38, %r5; + mov.u32 %r39, %r5; + mov.u32 %r40, %r5; + mov.u32 %r41, %r5; + @%p3 ld.global.v4.b32 { %r38, %r39, %r40, %r41 }, [ %rd10 + 0 ]; + // end inline asm + mov.b32 {%rs126, %rs127}, %r38; + mov.b32 {%rs128, %rs129}, %r39; + mov.b32 {%rs130, %rs131}, %r40; + mov.b32 {%rs132, %rs133}, %r41; + // begin inline asm + mov.u32 %r42, %r5; + mov.u32 %r43, %r5; + mov.u32 %r44, %r5; + mov.u32 %r45, %r5; + @%p13 ld.global.v4.b32 { %r42, %r43, %r44, %r45 }, [ %rd11 + 0 ]; + // end inline asm + mov.b32 {%rs134, %rs135}, %r42; + mov.b32 {%rs136, %rs137}, %r43; + mov.b32 {%rs138, %rs139}, %r44; + mov.b32 {%rs140, %rs141}, %r45; + // begin inline asm + mov.u32 %r46, %r5; + mov.u32 %r47, %r5; + mov.u32 %r48, %r5; + mov.u32 %r49, %r5; + @%p14 ld.global.v4.b32 { %r46, %r47, %r48, %r49 }, [ %rd12 + 0 ]; + // end inline asm + mov.b32 {%rs142, %rs143}, %r46; + mov.b32 {%rs144, %rs145}, %r47; + mov.b32 {%rs146, %rs147}, %r48; + mov.b32 {%rs148, %rs149}, %r49; + // begin inline asm + mov.u32 %r50, %r5; + mov.u32 %r51, %r5; + mov.u32 %r52, %r5; + mov.u32 %r53, %r5; + @%p15 ld.global.v4.b32 { %r50, %r51, %r52, %r53 }, [ %rd13 + 0 ]; + // end inline asm + mov.b32 {%rs150, %rs151}, %r50; + mov.b32 {%rs152, %rs153}, %r51; + mov.b32 {%rs154, %rs155}, %r52; + mov.b32 {%rs156, %rs157}, %r53; + // begin inline asm + mov.u32 %r54, %r5; + mov.u32 %r55, %r5; + mov.u32 %r56, %r5; + mov.u32 %r57, %r5; + @%p16 ld.global.v4.b32 { %r54, %r55, %r56, %r57 }, [ %rd14 + 0 ]; + // end inline asm + mov.b32 {%rs158, %rs159}, %r54; + mov.b32 {%rs160, %rs161}, %r55; + mov.b32 {%rs162, %rs163}, %r56; + mov.b32 {%rs164, %rs165}, %r57; + // begin inline asm + mov.u32 %r58, %r5; + mov.u32 %r59, %r5; + mov.u32 %r60, %r5; + mov.u32 %r61, %r5; + @%p17 ld.global.v4.b32 { %r58, %r59, %r60, %r61 }, [ %rd15 + 0 ]; + // end inline asm + mov.b32 {%rs166, %rs167}, %r58; + mov.b32 {%rs168, %rs169}, %r59; + mov.b32 {%rs170, %rs171}, %r60; + mov.b32 {%rs172, %rs173}, %r61; + // begin inline asm + mov.u32 %r62, %r5; + mov.u32 %r63, %r5; + mov.u32 %r64, %r5; + mov.u32 %r65, %r5; + @%p18 ld.global.v4.b32 { %r62, %r63, %r64, %r65 }, [ %rd16 + 0 ]; + // end inline asm + mov.b32 {%rs174, %rs175}, %r62; + mov.b32 {%rs176, %rs177}, %r63; + mov.b32 {%rs178, %rs179}, %r64; + mov.b32 {%rs180, %rs181}, %r65; + // begin inline asm + mov.u32 %r66, %r5; + mov.u32 %r67, %r5; + mov.u32 %r68, %r5; + mov.u32 %r69, %r5; + @%p19 ld.global.v4.b32 { %r66, %r67, %r68, %r69 }, [ %rd17 + 0 ]; + // end inline asm + mov.b32 {%rs182, %rs183}, %r66; + mov.b32 {%rs184, %rs185}, %r67; + mov.b32 {%rs186, %rs187}, %r68; + mov.b32 {%rs188, %rs189}, %r69; + // begin inline asm + mov.u32 %r70, %r5; + mov.u32 %r71, %r5; + mov.u32 %r72, %r5; + mov.u32 %r73, %r5; + @%p20 ld.global.v4.b32 { %r70, %r71, %r72, %r73 }, [ %rd18 + 0 ]; + // end inline asm + mov.b32 {%rs190, %rs191}, %r70; + mov.b32 {%rs192, %rs193}, %r71; + mov.b32 {%rs194, %rs195}, %r72; + mov.b32 {%rs196, %rs197}, %r73; + shr.u32 %r429, %r426, 15; + and.b32 %r430, %r429, 1; + setp.ne.b32 %p21, %r430, 0; + // begin inline asm + mov.u32 %r74, %r5; + mov.u32 %r75, %r5; + mov.u32 %r76, %r5; + mov.u32 %r77, %r5; + @%p21 ld.global.v4.b32 { %r74, %r75, %r76, %r77 }, [ %rd19 + 0 ]; + // end inline asm + mov.b32 {%rs198, %rs199}, %r74; + mov.b32 {%rs200, %rs201}, %r75; + mov.b32 {%rs202, %rs203}, %r76; + mov.b32 {%rs204, %rs205}, %r77; + shr.u32 %r431, %r426, 14; + and.b32 %r432, %r431, 1; + setp.ne.b32 %p22, %r432, 0; + // begin inline asm + mov.u32 %r78, %r5; + mov.u32 %r79, %r5; + mov.u32 %r80, %r5; + mov.u32 %r81, %r5; + @%p22 ld.global.v4.b32 { %r78, %r79, %r80, %r81 }, [ %rd20 + 0 ]; + // end inline asm + mov.b32 {%rs206, %rs207}, %r78; + mov.b32 {%rs208, %rs209}, %r79; + mov.b32 {%rs210, %rs211}, %r80; + mov.b32 {%rs212, %rs213}, %r81; + shr.u32 %r433, %r426, 13; + and.b32 %r434, %r433, 1; + setp.ne.b32 %p23, %r434, 0; + // begin inline asm + mov.u32 %r82, %r5; + mov.u32 %r83, %r5; + mov.u32 %r84, %r5; + mov.u32 %r85, %r5; + @%p23 ld.global.v4.b32 { %r82, %r83, %r84, %r85 }, [ %rd21 + 0 ]; + // end inline asm + mov.b32 {%rs214, %rs215}, %r82; + mov.b32 {%rs216, %rs217}, %r83; + mov.b32 {%rs218, %rs219}, %r84; + mov.b32 {%rs220, %rs221}, %r85; + shr.u32 %r435, %r426, 12; + and.b32 %r436, %r435, 1; + setp.ne.b32 %p24, %r436, 0; + // begin inline asm + mov.u32 %r86, %r5; + mov.u32 %r87, %r5; + mov.u32 %r88, %r5; + mov.u32 %r89, %r5; + @%p24 ld.global.v4.b32 { %r86, %r87, %r88, %r89 }, [ %rd22 + 0 ]; + // end inline asm + mov.b32 {%rs222, %rs223}, %r86; + mov.b32 {%rs224, %rs225}, %r87; + mov.b32 {%rs226, %rs227}, %r88; + mov.b32 {%rs228, %rs229}, %r89; + shr.u32 %r437, %r426, 11; + and.b32 %r438, %r437, 1; + setp.ne.b32 %p25, %r438, 0; + // begin inline asm + mov.u32 %r90, %r5; + mov.u32 %r91, %r5; + mov.u32 %r92, %r5; + mov.u32 %r93, %r5; + @%p25 ld.global.v4.b32 { %r90, %r91, %r92, %r93 }, [ %rd23 + 0 ]; + // end inline asm + mov.b32 {%rs230, %rs231}, %r90; + mov.b32 {%rs232, %rs233}, %r91; + mov.b32 {%rs234, %rs235}, %r92; + mov.b32 {%rs236, %rs237}, %r93; + shr.u32 %r439, %r426, 10; + and.b32 %r440, %r439, 1; + setp.ne.b32 %p26, %r440, 0; + // begin inline asm + mov.u32 %r94, %r5; + mov.u32 %r95, %r5; + mov.u32 %r96, %r5; + mov.u32 %r97, %r5; + @%p26 ld.global.v4.b32 { %r94, %r95, %r96, %r97 }, [ %rd24 + 0 ]; + // end inline asm + mov.b32 {%rs238, %rs239}, %r94; + mov.b32 {%rs240, %rs241}, %r95; + mov.b32 {%rs242, %rs243}, %r96; + mov.b32 {%rs244, %rs245}, %r97; + shr.u32 %r441, %r426, 9; + and.b32 %r442, %r441, 1; + setp.ne.b32 %p27, %r442, 0; + // begin inline asm + mov.u32 %r98, %r5; + mov.u32 %r99, %r5; + mov.u32 %r100, %r5; + mov.u32 %r101, %r5; + @%p27 ld.global.v4.b32 { %r98, %r99, %r100, %r101 }, [ %rd25 + 0 ]; + // end inline asm + mov.b32 {%rs246, %rs247}, %r98; + mov.b32 {%rs248, %rs249}, %r99; + mov.b32 {%rs250, %rs251}, %r100; + mov.b32 {%rs252, %rs253}, %r101; + shr.u32 %r443, %r426, 8; + and.b32 %r444, %r443, 1; + setp.ne.b32 %p28, %r444, 0; + // begin inline asm + mov.u32 %r102, %r5; + mov.u32 %r103, %r5; + mov.u32 %r104, %r5; + mov.u32 %r105, %r5; + @%p28 ld.global.v4.b32 { %r102, %r103, %r104, %r105 }, [ %rd26 + 0 ]; + // end inline asm + mov.b32 {%rs254, %rs255}, %r102; + mov.b32 {%rs256, %rs257}, %r103; + mov.b32 {%rs258, %rs259}, %r104; + mov.b32 {%rs260, %rs261}, %r105; + shr.u32 %r445, %r426, 7; + and.b32 %r446, %r445, 1; + setp.ne.b32 %p29, %r446, 0; + // begin inline asm + mov.u32 %r106, %r5; + mov.u32 %r107, %r5; + mov.u32 %r108, %r5; + mov.u32 %r109, %r5; + @%p29 ld.global.v4.b32 { %r106, %r107, %r108, %r109 }, [ %rd27 + 0 ]; + // end inline asm + mov.b32 {%rs262, %rs263}, %r106; + mov.b32 {%rs264, %rs265}, %r107; + mov.b32 {%rs266, %rs267}, %r108; + mov.b32 {%rs268, %rs269}, %r109; + shr.u32 %r447, %r426, 6; + and.b32 %r448, %r447, 1; + setp.ne.b32 %p30, %r448, 0; + // begin inline asm + mov.u32 %r110, %r5; + mov.u32 %r111, %r5; + mov.u32 %r112, %r5; + mov.u32 %r113, %r5; + @%p30 ld.global.v4.b32 { %r110, %r111, %r112, %r113 }, [ %rd28 + 0 ]; + // end inline asm + mov.b32 {%rs270, %rs271}, %r110; + mov.b32 {%rs272, %rs273}, %r111; + mov.b32 {%rs274, %rs275}, %r112; + mov.b32 {%rs276, %rs277}, %r113; + shr.u32 %r449, %r426, 5; + and.b32 %r450, %r449, 1; + setp.ne.b32 %p31, %r450, 0; + // begin inline asm + mov.u32 %r114, %r5; + mov.u32 %r115, %r5; + mov.u32 %r116, %r5; + mov.u32 %r117, %r5; + @%p31 ld.global.v4.b32 { %r114, %r115, %r116, %r117 }, [ %rd29 + 0 ]; + // end inline asm + mov.b32 {%rs278, %rs279}, %r114; + mov.b32 {%rs280, %rs281}, %r115; + mov.b32 {%rs282, %rs283}, %r116; + mov.b32 {%rs284, %rs285}, %r117; + shr.u32 %r451, %r426, 4; + and.b32 %r452, %r451, 1; + setp.ne.b32 %p32, %r452, 0; + // begin inline asm + mov.u32 %r118, %r5; + mov.u32 %r119, %r5; + mov.u32 %r120, %r5; + mov.u32 %r121, %r5; + @%p32 ld.global.v4.b32 { %r118, %r119, %r120, %r121 }, [ %rd30 + 0 ]; + // end inline asm + mov.b32 {%rs286, %rs287}, %r118; + mov.b32 {%rs288, %rs289}, %r119; + mov.b32 {%rs290, %rs291}, %r120; + mov.b32 {%rs292, %rs293}, %r121; + shr.u32 %r453, %r426, 3; + and.b32 %r454, %r453, 1; + setp.ne.b32 %p33, %r454, 0; + // begin inline asm + mov.u32 %r122, %r5; + mov.u32 %r123, %r5; + mov.u32 %r124, %r5; + mov.u32 %r125, %r5; + @%p33 ld.global.v4.b32 { %r122, %r123, %r124, %r125 }, [ %rd31 + 0 ]; + // end inline asm + mov.b32 {%rs294, %rs295}, %r122; + mov.b32 {%rs296, %rs297}, %r123; + mov.b32 {%rs298, %rs299}, %r124; + mov.b32 {%rs300, %rs301}, %r125; + shr.u32 %r455, %r426, 2; + and.b32 %r456, %r455, 1; + setp.ne.b32 %p34, %r456, 0; + // begin inline asm + mov.u32 %r126, %r5; + mov.u32 %r127, %r5; + mov.u32 %r128, %r5; + mov.u32 %r129, %r5; + @%p34 ld.global.v4.b32 { %r126, %r127, %r128, %r129 }, [ %rd32 + 0 ]; + // end inline asm + mov.b32 {%rs302, %rs303}, %r126; + mov.b32 {%rs304, %rs305}, %r127; + mov.b32 {%rs306, %rs307}, %r128; + mov.b32 {%rs308, %rs309}, %r129; + .loc 1 104 47 // l2norm.py:104:47 + cvt.f32.bf16 %r457, %rs54; + cvt.f32.bf16 %r458, %rs55; + cvt.f32.bf16 %r459, %rs56; + cvt.f32.bf16 %r460, %rs57; + cvt.f32.bf16 %r461, %rs58; + cvt.f32.bf16 %r462, %rs59; + cvt.f32.bf16 %r463, %rs60; + cvt.f32.bf16 %r464, %rs61; + cvt.f32.bf16 %r465, %rs62; + cvt.f32.bf16 %r466, %rs63; + cvt.f32.bf16 %r467, %rs64; + cvt.f32.bf16 %r468, %rs65; + cvt.f32.bf16 %r469, %rs66; + cvt.f32.bf16 %r470, %rs67; + cvt.f32.bf16 %r471, %rs68; + cvt.f32.bf16 %r472, %rs69; + cvt.f32.bf16 %r473, %rs70; + cvt.f32.bf16 %r474, %rs71; + cvt.f32.bf16 %r475, %rs72; + cvt.f32.bf16 %r476, %rs73; + cvt.f32.bf16 %r477, %rs74; + cvt.f32.bf16 %r478, %rs75; + cvt.f32.bf16 %r479, %rs76; + cvt.f32.bf16 %r480, %rs77; + cvt.f32.bf16 %r481, %rs78; + cvt.f32.bf16 %r482, %rs79; + cvt.f32.bf16 %r483, %rs80; + cvt.f32.bf16 %r484, %rs81; + cvt.f32.bf16 %r485, %rs82; + cvt.f32.bf16 %r486, %rs83; + cvt.f32.bf16 %r487, %rs84; + cvt.f32.bf16 %r488, %rs85; + cvt.f32.bf16 %r489, %rs86; + cvt.f32.bf16 %r490, %rs87; + cvt.f32.bf16 %r491, %rs88; + cvt.f32.bf16 %r492, %rs89; + cvt.f32.bf16 %r493, %rs90; + cvt.f32.bf16 %r494, %rs91; + cvt.f32.bf16 %r495, %rs92; + cvt.f32.bf16 %r496, %rs93; + cvt.f32.bf16 %r497, %rs94; + cvt.f32.bf16 %r498, %rs95; + cvt.f32.bf16 %r499, %rs96; + cvt.f32.bf16 %r500, %rs97; + cvt.f32.bf16 %r501, %rs98; + cvt.f32.bf16 %r502, %rs99; + cvt.f32.bf16 %r503, %rs100; + cvt.f32.bf16 %r504, %rs101; + cvt.f32.bf16 %r505, %rs102; + cvt.f32.bf16 %r506, %rs103; + cvt.f32.bf16 %r507, %rs104; + cvt.f32.bf16 %r508, %rs105; + cvt.f32.bf16 %r509, %rs106; + cvt.f32.bf16 %r510, %rs107; + cvt.f32.bf16 %r511, %rs108; + cvt.f32.bf16 %r512, %rs109; + cvt.f32.bf16 %r513, %rs110; + cvt.f32.bf16 %r514, %rs111; + cvt.f32.bf16 %r515, %rs112; + cvt.f32.bf16 %r516, %rs113; + cvt.f32.bf16 %r517, %rs114; + cvt.f32.bf16 %r518, %rs115; + cvt.f32.bf16 %r519, %rs116; + cvt.f32.bf16 %r520, %rs117; + cvt.f32.bf16 %r521, %rs118; + cvt.f32.bf16 %r522, %rs119; + cvt.f32.bf16 %r523, %rs120; + cvt.f32.bf16 %r524, %rs121; + cvt.f32.bf16 %r525, %rs122; + cvt.f32.bf16 %r526, %rs123; + cvt.f32.bf16 %r527, %rs124; + cvt.f32.bf16 %r528, %rs125; + cvt.f32.bf16 %r529, %rs126; + cvt.f32.bf16 %r530, %rs127; + cvt.f32.bf16 %r531, %rs128; + cvt.f32.bf16 %r532, %rs129; + cvt.f32.bf16 %r533, %rs130; + cvt.f32.bf16 %r534, %rs131; + cvt.f32.bf16 %r535, %rs132; + cvt.f32.bf16 %r536, %rs133; + cvt.f32.bf16 %r537, %rs134; + cvt.f32.bf16 %r538, %rs135; + cvt.f32.bf16 %r539, %rs136; + cvt.f32.bf16 %r540, %rs137; + cvt.f32.bf16 %r541, %rs138; + cvt.f32.bf16 %r542, %rs139; + cvt.f32.bf16 %r543, %rs140; + cvt.f32.bf16 %r544, %rs141; + cvt.f32.bf16 %r545, %rs142; + cvt.f32.bf16 %r546, %rs143; + cvt.f32.bf16 %r547, %rs144; + cvt.f32.bf16 %r548, %rs145; + cvt.f32.bf16 %r549, %rs146; + cvt.f32.bf16 %r550, %rs147; + cvt.f32.bf16 %r551, %rs148; + cvt.f32.bf16 %r552, %rs149; + cvt.f32.bf16 %r553, %rs150; + cvt.f32.bf16 %r554, %rs151; + cvt.f32.bf16 %r555, %rs152; + cvt.f32.bf16 %r556, %rs153; + cvt.f32.bf16 %r557, %rs154; + cvt.f32.bf16 %r558, %rs155; + cvt.f32.bf16 %r559, %rs156; + cvt.f32.bf16 %r560, %rs157; + cvt.f32.bf16 %r561, %rs158; + cvt.f32.bf16 %r562, %rs159; + cvt.f32.bf16 %r563, %rs160; + cvt.f32.bf16 %r564, %rs161; + cvt.f32.bf16 %r565, %rs162; + cvt.f32.bf16 %r566, %rs163; + cvt.f32.bf16 %r567, %rs164; + cvt.f32.bf16 %r568, %rs165; + cvt.f32.bf16 %r569, %rs166; + cvt.f32.bf16 %r570, %rs167; + cvt.f32.bf16 %r571, %rs168; + cvt.f32.bf16 %r572, %rs169; + cvt.f32.bf16 %r573, %rs170; + cvt.f32.bf16 %r574, %rs171; + cvt.f32.bf16 %r575, %rs172; + cvt.f32.bf16 %r576, %rs173; + cvt.f32.bf16 %r577, %rs174; + cvt.f32.bf16 %r578, %rs175; + cvt.f32.bf16 %r579, %rs176; + cvt.f32.bf16 %r580, %rs177; + cvt.f32.bf16 %r581, %rs178; + cvt.f32.bf16 %r582, %rs179; + cvt.f32.bf16 %r583, %rs180; + cvt.f32.bf16 %r584, %rs181; + cvt.f32.bf16 %r585, %rs182; + cvt.f32.bf16 %r586, %rs183; + cvt.f32.bf16 %r587, %rs184; + cvt.f32.bf16 %r588, %rs185; + cvt.f32.bf16 %r589, %rs186; + cvt.f32.bf16 %r590, %rs187; + cvt.f32.bf16 %r591, %rs188; + cvt.f32.bf16 %r592, %rs189; + cvt.f32.bf16 %r593, %rs190; + cvt.f32.bf16 %r594, %rs191; + cvt.f32.bf16 %r595, %rs192; + cvt.f32.bf16 %r596, %rs193; + cvt.f32.bf16 %r597, %rs194; + cvt.f32.bf16 %r598, %rs195; + cvt.f32.bf16 %r599, %rs196; + cvt.f32.bf16 %r600, %rs197; + cvt.f32.bf16 %r601, %rs198; + cvt.f32.bf16 %r602, %rs199; + cvt.f32.bf16 %r603, %rs200; + cvt.f32.bf16 %r604, %rs201; + cvt.f32.bf16 %r605, %rs202; + cvt.f32.bf16 %r606, %rs203; + cvt.f32.bf16 %r607, %rs204; + cvt.f32.bf16 %r608, %rs205; + cvt.f32.bf16 %r609, %rs206; + cvt.f32.bf16 %r610, %rs207; + cvt.f32.bf16 %r611, %rs208; + cvt.f32.bf16 %r612, %rs209; + cvt.f32.bf16 %r613, %rs210; + cvt.f32.bf16 %r614, %rs211; + cvt.f32.bf16 %r615, %rs212; + cvt.f32.bf16 %r616, %rs213; + cvt.f32.bf16 %r617, %rs214; + cvt.f32.bf16 %r618, %rs215; + cvt.f32.bf16 %r619, %rs216; + cvt.f32.bf16 %r620, %rs217; + cvt.f32.bf16 %r621, %rs218; + cvt.f32.bf16 %r622, %rs219; + cvt.f32.bf16 %r623, %rs220; + cvt.f32.bf16 %r624, %rs221; + cvt.f32.bf16 %r625, %rs222; + cvt.f32.bf16 %r626, %rs223; + cvt.f32.bf16 %r627, %rs224; + cvt.f32.bf16 %r628, %rs225; + cvt.f32.bf16 %r629, %rs226; + cvt.f32.bf16 %r630, %rs227; + cvt.f32.bf16 %r631, %rs228; + cvt.f32.bf16 %r632, %rs229; + cvt.f32.bf16 %r633, %rs230; + cvt.f32.bf16 %r634, %rs231; + cvt.f32.bf16 %r635, %rs232; + cvt.f32.bf16 %r636, %rs233; + cvt.f32.bf16 %r637, %rs234; + cvt.f32.bf16 %r638, %rs235; + cvt.f32.bf16 %r639, %rs236; + cvt.f32.bf16 %r640, %rs237; + cvt.f32.bf16 %r641, %rs238; + cvt.f32.bf16 %r642, %rs239; + cvt.f32.bf16 %r643, %rs240; + cvt.f32.bf16 %r644, %rs241; + cvt.f32.bf16 %r645, %rs242; + cvt.f32.bf16 %r646, %rs243; + cvt.f32.bf16 %r647, %rs244; + cvt.f32.bf16 %r648, %rs245; + cvt.f32.bf16 %r649, %rs246; + cvt.f32.bf16 %r650, %rs247; + cvt.f32.bf16 %r651, %rs248; + cvt.f32.bf16 %r652, %rs249; + cvt.f32.bf16 %r653, %rs250; + cvt.f32.bf16 %r654, %rs251; + cvt.f32.bf16 %r655, %rs252; + cvt.f32.bf16 %r656, %rs253; + cvt.f32.bf16 %r657, %rs254; + cvt.f32.bf16 %r658, %rs255; + cvt.f32.bf16 %r659, %rs256; + cvt.f32.bf16 %r660, %rs257; + cvt.f32.bf16 %r661, %rs258; + cvt.f32.bf16 %r662, %rs259; + cvt.f32.bf16 %r663, %rs260; + cvt.f32.bf16 %r664, %rs261; + cvt.f32.bf16 %r665, %rs262; + cvt.f32.bf16 %r666, %rs263; + cvt.f32.bf16 %r667, %rs264; + cvt.f32.bf16 %r668, %rs265; + cvt.f32.bf16 %r669, %rs266; + cvt.f32.bf16 %r670, %rs267; + cvt.f32.bf16 %r671, %rs268; + cvt.f32.bf16 %r672, %rs269; + cvt.f32.bf16 %r673, %rs270; + cvt.f32.bf16 %r674, %rs271; + cvt.f32.bf16 %r675, %rs272; + cvt.f32.bf16 %r676, %rs273; + cvt.f32.bf16 %r677, %rs274; + cvt.f32.bf16 %r678, %rs275; + cvt.f32.bf16 %r679, %rs276; + cvt.f32.bf16 %r680, %rs277; + cvt.f32.bf16 %r681, %rs278; + cvt.f32.bf16 %r682, %rs279; + cvt.f32.bf16 %r683, %rs280; + cvt.f32.bf16 %r684, %rs281; + cvt.f32.bf16 %r685, %rs282; + cvt.f32.bf16 %r686, %rs283; + cvt.f32.bf16 %r687, %rs284; + cvt.f32.bf16 %r688, %rs285; + cvt.f32.bf16 %r689, %rs286; + cvt.f32.bf16 %r690, %rs287; + cvt.f32.bf16 %r691, %rs288; + cvt.f32.bf16 %r692, %rs289; + cvt.f32.bf16 %r693, %rs290; + cvt.f32.bf16 %r694, %rs291; + cvt.f32.bf16 %r695, %rs292; + cvt.f32.bf16 %r696, %rs293; + cvt.f32.bf16 %r697, %rs294; + cvt.f32.bf16 %r698, %rs295; + cvt.f32.bf16 %r699, %rs296; + cvt.f32.bf16 %r700, %rs297; + cvt.f32.bf16 %r701, %rs298; + cvt.f32.bf16 %r702, %rs299; + cvt.f32.bf16 %r703, %rs300; + cvt.f32.bf16 %r704, %rs301; + cvt.f32.bf16 %r705, %rs302; + cvt.f32.bf16 %r706, %rs303; + cvt.f32.bf16 %r707, %rs304; + cvt.f32.bf16 %r708, %rs305; + cvt.f32.bf16 %r709, %rs306; + cvt.f32.bf16 %r710, %rs307; + cvt.f32.bf16 %r711, %rs308; + cvt.f32.bf16 %r712, %rs309; + .loc 1 105 21 // l2norm.py:105:21 + shr.u32 %r713, %r426, 1; + and.b32 %r714, %r713, 1; + setp.ne.b32 %p4, %r714, 0; + // begin inline asm + mov.u32 %r130, %r5; + @%p4 ld.global.b32 { %r130 }, [ %rd33 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r131, %r5; + @%p5 ld.global.b32 { %r131 }, [ %rd34 + 0 ]; + // end inline asm + .loc 1 106 19 // l2norm.py:106:19 + // begin inline asm + mov.u32 %r132, %r5; + mov.u32 %r133, %r5; + mov.u32 %r134, %r5; + mov.u32 %r135, %r5; + @%p1 ld.global.v4.b32 { %r132, %r133, %r134, %r135 }, [ %rd35 + 0 ]; + // end inline asm + mov.b32 {%rs310, %rs311}, %r132; + mov.b32 {%rs312, %rs313}, %r133; + mov.b32 {%rs314, %rs315}, %r134; + mov.b32 {%rs316, %rs317}, %r135; + // begin inline asm + mov.u32 %r136, %r5; + mov.u32 %r137, %r5; + mov.u32 %r138, %r5; + mov.u32 %r139, %r5; + @%p2 ld.global.v4.b32 { %r136, %r137, %r138, %r139 }, [ %rd36 + 0 ]; + // end inline asm + mov.b32 {%rs318, %rs319}, %r136; + mov.b32 {%rs320, %rs321}, %r137; + mov.b32 {%rs322, %rs323}, %r138; + mov.b32 {%rs324, %rs325}, %r139; + // begin inline asm + mov.u32 %r140, %r5; + mov.u32 %r141, %r5; + mov.u32 %r142, %r5; + mov.u32 %r143, %r5; + @%p6 ld.global.v4.b32 { %r140, %r141, %r142, %r143 }, [ %rd37 + 0 ]; + // end inline asm + mov.b32 {%rs326, %rs327}, %r140; + mov.b32 {%rs328, %rs329}, %r141; + mov.b32 {%rs330, %rs331}, %r142; + mov.b32 {%rs332, %rs333}, %r143; + // begin inline asm + mov.u32 %r144, %r5; + mov.u32 %r145, %r5; + mov.u32 %r146, %r5; + mov.u32 %r147, %r5; + @%p7 ld.global.v4.b32 { %r144, %r145, %r146, %r147 }, [ %rd38 + 0 ]; + // end inline asm + mov.b32 {%rs334, %rs335}, %r144; + mov.b32 {%rs336, %rs337}, %r145; + mov.b32 {%rs338, %rs339}, %r146; + mov.b32 {%rs340, %rs341}, %r147; + // begin inline asm + mov.u32 %r148, %r5; + mov.u32 %r149, %r5; + mov.u32 %r150, %r5; + mov.u32 %r151, %r5; + @%p8 ld.global.v4.b32 { %r148, %r149, %r150, %r151 }, [ %rd39 + 0 ]; + // end inline asm + mov.b32 {%rs342, %rs343}, %r148; + mov.b32 {%rs344, %rs345}, %r149; + mov.b32 {%rs346, %rs347}, %r150; + mov.b32 {%rs348, %rs349}, %r151; + // begin inline asm + mov.u32 %r152, %r5; + mov.u32 %r153, %r5; + mov.u32 %r154, %r5; + mov.u32 %r155, %r5; + @%p9 ld.global.v4.b32 { %r152, %r153, %r154, %r155 }, [ %rd40 + 0 ]; + // end inline asm + mov.b32 {%rs350, %rs351}, %r152; + mov.b32 {%rs352, %rs353}, %r153; + mov.b32 {%rs354, %rs355}, %r154; + mov.b32 {%rs356, %rs357}, %r155; + // begin inline asm + mov.u32 %r156, %r5; + mov.u32 %r157, %r5; + mov.u32 %r158, %r5; + mov.u32 %r159, %r5; + @%p10 ld.global.v4.b32 { %r156, %r157, %r158, %r159 }, [ %rd41 + 0 ]; + // end inline asm + mov.b32 {%rs358, %rs359}, %r156; + mov.b32 {%rs360, %rs361}, %r157; + mov.b32 {%rs362, %rs363}, %r158; + mov.b32 {%rs364, %rs365}, %r159; + // begin inline asm + mov.u32 %r160, %r5; + mov.u32 %r161, %r5; + mov.u32 %r162, %r5; + mov.u32 %r163, %r5; + @%p11 ld.global.v4.b32 { %r160, %r161, %r162, %r163 }, [ %rd42 + 0 ]; + // end inline asm + mov.b32 {%rs366, %rs367}, %r160; + mov.b32 {%rs368, %rs369}, %r161; + mov.b32 {%rs370, %rs371}, %r162; + mov.b32 {%rs372, %rs373}, %r163; + // begin inline asm + mov.u32 %r164, %r5; + mov.u32 %r165, %r5; + mov.u32 %r166, %r5; + mov.u32 %r167, %r5; + @%p12 ld.global.v4.b32 { %r164, %r165, %r166, %r167 }, [ %rd43 + 0 ]; + // end inline asm + mov.b32 {%rs374, %rs375}, %r164; + mov.b32 {%rs376, %rs377}, %r165; + mov.b32 {%rs378, %rs379}, %r166; + mov.b32 {%rs380, %rs381}, %r167; + // begin inline asm + mov.u32 %r168, %r5; + mov.u32 %r169, %r5; + mov.u32 %r170, %r5; + mov.u32 %r171, %r5; + @%p3 ld.global.v4.b32 { %r168, %r169, %r170, %r171 }, [ %rd44 + 0 ]; + // end inline asm + mov.b32 {%rs382, %rs383}, %r168; + mov.b32 {%rs384, %rs385}, %r169; + mov.b32 {%rs386, %rs387}, %r170; + mov.b32 {%rs388, %rs389}, %r171; + // begin inline asm + mov.u32 %r172, %r5; + mov.u32 %r173, %r5; + mov.u32 %r174, %r5; + mov.u32 %r175, %r5; + @%p13 ld.global.v4.b32 { %r172, %r173, %r174, %r175 }, [ %rd45 + 0 ]; + // end inline asm + mov.b32 {%rs390, %rs391}, %r172; + mov.b32 {%rs392, %rs393}, %r173; + mov.b32 {%rs394, %rs395}, %r174; + mov.b32 {%rs396, %rs397}, %r175; + // begin inline asm + mov.u32 %r176, %r5; + mov.u32 %r177, %r5; + mov.u32 %r178, %r5; + mov.u32 %r179, %r5; + @%p14 ld.global.v4.b32 { %r176, %r177, %r178, %r179 }, [ %rd46 + 0 ]; + // end inline asm + mov.b32 {%rs398, %rs399}, %r176; + mov.b32 {%rs400, %rs401}, %r177; + mov.b32 {%rs402, %rs403}, %r178; + mov.b32 {%rs404, %rs405}, %r179; + // begin inline asm + mov.u32 %r180, %r5; + mov.u32 %r181, %r5; + mov.u32 %r182, %r5; + mov.u32 %r183, %r5; + @%p15 ld.global.v4.b32 { %r180, %r181, %r182, %r183 }, [ %rd47 + 0 ]; + // end inline asm + mov.b32 {%rs406, %rs407}, %r180; + mov.b32 {%rs408, %rs409}, %r181; + mov.b32 {%rs410, %rs411}, %r182; + mov.b32 {%rs412, %rs413}, %r183; + // begin inline asm + mov.u32 %r184, %r5; + mov.u32 %r185, %r5; + mov.u32 %r186, %r5; + mov.u32 %r187, %r5; + @%p16 ld.global.v4.b32 { %r184, %r185, %r186, %r187 }, [ %rd48 + 0 ]; + // end inline asm + mov.b32 {%rs414, %rs415}, %r184; + mov.b32 {%rs416, %rs417}, %r185; + mov.b32 {%rs418, %rs419}, %r186; + mov.b32 {%rs420, %rs421}, %r187; + // begin inline asm + mov.u32 %r188, %r5; + mov.u32 %r189, %r5; + mov.u32 %r190, %r5; + mov.u32 %r191, %r5; + @%p17 ld.global.v4.b32 { %r188, %r189, %r190, %r191 }, [ %rd49 + 0 ]; + // end inline asm + mov.b32 {%rs422, %rs423}, %r188; + mov.b32 {%rs424, %rs425}, %r189; + mov.b32 {%rs426, %rs427}, %r190; + mov.b32 {%rs428, %rs429}, %r191; + // begin inline asm + mov.u32 %r192, %r5; + mov.u32 %r193, %r5; + mov.u32 %r194, %r5; + mov.u32 %r195, %r5; + @%p18 ld.global.v4.b32 { %r192, %r193, %r194, %r195 }, [ %rd50 + 0 ]; + // end inline asm + mov.b32 {%rs430, %rs431}, %r192; + mov.b32 {%rs432, %rs433}, %r193; + mov.b32 {%rs434, %rs435}, %r194; + mov.b32 {%rs436, %rs437}, %r195; + // begin inline asm + mov.u32 %r196, %r5; + mov.u32 %r197, %r5; + mov.u32 %r198, %r5; + mov.u32 %r199, %r5; + @%p19 ld.global.v4.b32 { %r196, %r197, %r198, %r199 }, [ %rd51 + 0 ]; + // end inline asm + mov.b32 {%rs438, %rs439}, %r196; + mov.b32 {%rs440, %rs441}, %r197; + mov.b32 {%rs442, %rs443}, %r198; + mov.b32 {%rs444, %rs445}, %r199; + // begin inline asm + mov.u32 %r200, %r5; + mov.u32 %r201, %r5; + mov.u32 %r202, %r5; + mov.u32 %r203, %r5; + @%p20 ld.global.v4.b32 { %r200, %r201, %r202, %r203 }, [ %rd52 + 0 ]; + // end inline asm + mov.b32 {%rs446, %rs447}, %r200; + mov.b32 {%rs448, %rs449}, %r201; + mov.b32 {%rs450, %rs451}, %r202; + mov.b32 {%rs452, %rs453}, %r203; + // begin inline asm + mov.u32 %r204, %r5; + mov.u32 %r205, %r5; + mov.u32 %r206, %r5; + mov.u32 %r207, %r5; + @%p21 ld.global.v4.b32 { %r204, %r205, %r206, %r207 }, [ %rd53 + 0 ]; + // end inline asm + mov.b32 {%rs454, %rs455}, %r204; + mov.b32 {%rs456, %rs457}, %r205; + mov.b32 {%rs458, %rs459}, %r206; + mov.b32 {%rs460, %rs461}, %r207; + // begin inline asm + mov.u32 %r208, %r5; + mov.u32 %r209, %r5; + mov.u32 %r210, %r5; + mov.u32 %r211, %r5; + @%p22 ld.global.v4.b32 { %r208, %r209, %r210, %r211 }, [ %rd54 + 0 ]; + // end inline asm + mov.b32 {%rs462, %rs463}, %r208; + mov.b32 {%rs464, %rs465}, %r209; + mov.b32 {%rs466, %rs467}, %r210; + mov.b32 {%rs468, %rs469}, %r211; + // begin inline asm + mov.u32 %r212, %r5; + mov.u32 %r213, %r5; + mov.u32 %r214, %r5; + mov.u32 %r215, %r5; + @%p23 ld.global.v4.b32 { %r212, %r213, %r214, %r215 }, [ %rd55 + 0 ]; + // end inline asm + mov.b32 {%rs470, %rs471}, %r212; + mov.b32 {%rs472, %rs473}, %r213; + mov.b32 {%rs474, %rs475}, %r214; + mov.b32 {%rs476, %rs477}, %r215; + // begin inline asm + mov.u32 %r216, %r5; + mov.u32 %r217, %r5; + mov.u32 %r218, %r5; + mov.u32 %r219, %r5; + @%p24 ld.global.v4.b32 { %r216, %r217, %r218, %r219 }, [ %rd56 + 0 ]; + // end inline asm + mov.b32 {%rs478, %rs479}, %r216; + mov.b32 {%rs480, %rs481}, %r217; + mov.b32 {%rs482, %rs483}, %r218; + mov.b32 {%rs484, %rs485}, %r219; + // begin inline asm + mov.u32 %r220, %r5; + mov.u32 %r221, %r5; + mov.u32 %r222, %r5; + mov.u32 %r223, %r5; + @%p25 ld.global.v4.b32 { %r220, %r221, %r222, %r223 }, [ %rd57 + 0 ]; + // end inline asm + mov.b32 {%rs486, %rs487}, %r220; + mov.b32 {%rs488, %rs489}, %r221; + mov.b32 {%rs490, %rs491}, %r222; + mov.b32 {%rs492, %rs493}, %r223; + // begin inline asm + mov.u32 %r224, %r5; + mov.u32 %r225, %r5; + mov.u32 %r226, %r5; + mov.u32 %r227, %r5; + @%p26 ld.global.v4.b32 { %r224, %r225, %r226, %r227 }, [ %rd58 + 0 ]; + // end inline asm + mov.b32 {%rs494, %rs495}, %r224; + mov.b32 {%rs496, %rs497}, %r225; + mov.b32 {%rs498, %rs499}, %r226; + mov.b32 {%rs500, %rs501}, %r227; + // begin inline asm + mov.u32 %r228, %r5; + mov.u32 %r229, %r5; + mov.u32 %r230, %r5; + mov.u32 %r231, %r5; + @%p27 ld.global.v4.b32 { %r228, %r229, %r230, %r231 }, [ %rd59 + 0 ]; + // end inline asm + mov.b32 {%rs502, %rs503}, %r228; + mov.b32 {%rs504, %rs505}, %r229; + mov.b32 {%rs506, %rs507}, %r230; + mov.b32 {%rs508, %rs509}, %r231; + // begin inline asm + mov.u32 %r232, %r5; + mov.u32 %r233, %r5; + mov.u32 %r234, %r5; + mov.u32 %r235, %r5; + @%p28 ld.global.v4.b32 { %r232, %r233, %r234, %r235 }, [ %rd60 + 0 ]; + // end inline asm + mov.b32 {%rs510, %rs511}, %r232; + mov.b32 {%rs512, %rs513}, %r233; + mov.b32 {%rs514, %rs515}, %r234; + mov.b32 {%rs516, %rs517}, %r235; + // begin inline asm + mov.u32 %r236, %r5; + mov.u32 %r237, %r5; + mov.u32 %r238, %r5; + mov.u32 %r239, %r5; + @%p29 ld.global.v4.b32 { %r236, %r237, %r238, %r239 }, [ %rd61 + 0 ]; + // end inline asm + mov.b32 {%rs518, %rs519}, %r236; + mov.b32 {%rs520, %rs521}, %r237; + mov.b32 {%rs522, %rs523}, %r238; + mov.b32 {%rs524, %rs525}, %r239; + // begin inline asm + mov.u32 %r240, %r5; + mov.u32 %r241, %r5; + mov.u32 %r242, %r5; + mov.u32 %r243, %r5; + @%p30 ld.global.v4.b32 { %r240, %r241, %r242, %r243 }, [ %rd62 + 0 ]; + // end inline asm + mov.b32 {%rs526, %rs527}, %r240; + mov.b32 {%rs528, %rs529}, %r241; + mov.b32 {%rs530, %rs531}, %r242; + mov.b32 {%rs532, %rs533}, %r243; + // begin inline asm + mov.u32 %r244, %r5; + mov.u32 %r245, %r5; + mov.u32 %r246, %r5; + mov.u32 %r247, %r5; + @%p31 ld.global.v4.b32 { %r244, %r245, %r246, %r247 }, [ %rd63 + 0 ]; + // end inline asm + mov.b32 {%rs534, %rs535}, %r244; + mov.b32 {%rs536, %rs537}, %r245; + mov.b32 {%rs538, %rs539}, %r246; + mov.b32 {%rs540, %rs541}, %r247; + // begin inline asm + mov.u32 %r248, %r5; + mov.u32 %r249, %r5; + mov.u32 %r250, %r5; + mov.u32 %r251, %r5; + @%p32 ld.global.v4.b32 { %r248, %r249, %r250, %r251 }, [ %rd64 + 0 ]; + // end inline asm + mov.b32 {%rs542, %rs543}, %r248; + mov.b32 {%rs544, %rs545}, %r249; + mov.b32 {%rs546, %rs547}, %r250; + mov.b32 {%rs548, %rs549}, %r251; + // begin inline asm + mov.u32 %r252, %r5; + mov.u32 %r253, %r5; + mov.u32 %r254, %r5; + mov.u32 %r255, %r5; + @%p33 ld.global.v4.b32 { %r252, %r253, %r254, %r255 }, [ %rd65 + 0 ]; + // end inline asm + mov.b32 {%rs550, %rs551}, %r252; + mov.b32 {%rs552, %rs553}, %r253; + mov.b32 {%rs554, %rs555}, %r254; + mov.b32 {%rs556, %rs557}, %r255; + // begin inline asm + mov.u32 %r256, %r5; + mov.u32 %r257, %r5; + mov.u32 %r258, %r5; + mov.u32 %r259, %r5; + @%p34 ld.global.v4.b32 { %r256, %r257, %r258, %r259 }, [ %rd66 + 0 ]; + // end inline asm + mov.b32 {%rs558, %rs559}, %r256; + mov.b32 {%rs560, %rs561}, %r257; + mov.b32 {%rs562, %rs563}, %r258; + mov.b32 {%rs564, %rs565}, %r259; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r715, %rs310; + cvt.f32.bf16 %r716, %rs311; + cvt.f32.bf16 %r717, %rs312; + cvt.f32.bf16 %r718, %rs313; + cvt.f32.bf16 %r719, %rs314; + cvt.f32.bf16 %r720, %rs315; + cvt.f32.bf16 %r721, %rs316; + cvt.f32.bf16 %r722, %rs317; + cvt.f32.bf16 %r723, %rs318; + cvt.f32.bf16 %r724, %rs319; + cvt.f32.bf16 %r725, %rs320; + cvt.f32.bf16 %r726, %rs321; + cvt.f32.bf16 %r727, %rs322; + cvt.f32.bf16 %r728, %rs323; + cvt.f32.bf16 %r729, %rs324; + cvt.f32.bf16 %r730, %rs325; + cvt.f32.bf16 %r731, %rs326; + cvt.f32.bf16 %r732, %rs327; + cvt.f32.bf16 %r733, %rs328; + cvt.f32.bf16 %r734, %rs329; + cvt.f32.bf16 %r735, %rs330; + cvt.f32.bf16 %r736, %rs331; + cvt.f32.bf16 %r737, %rs332; + cvt.f32.bf16 %r738, %rs333; + cvt.f32.bf16 %r739, %rs334; + cvt.f32.bf16 %r740, %rs335; + cvt.f32.bf16 %r741, %rs336; + cvt.f32.bf16 %r742, %rs337; + cvt.f32.bf16 %r743, %rs338; + cvt.f32.bf16 %r744, %rs339; + cvt.f32.bf16 %r745, %rs340; + cvt.f32.bf16 %r746, %rs341; + cvt.f32.bf16 %r747, %rs342; + cvt.f32.bf16 %r748, %rs343; + cvt.f32.bf16 %r749, %rs344; + cvt.f32.bf16 %r750, %rs345; + cvt.f32.bf16 %r751, %rs346; + cvt.f32.bf16 %r752, %rs347; + cvt.f32.bf16 %r753, %rs348; + cvt.f32.bf16 %r754, %rs349; + cvt.f32.bf16 %r755, %rs350; + cvt.f32.bf16 %r756, %rs351; + cvt.f32.bf16 %r757, %rs352; + cvt.f32.bf16 %r758, %rs353; + cvt.f32.bf16 %r759, %rs354; + cvt.f32.bf16 %r760, %rs355; + cvt.f32.bf16 %r761, %rs356; + cvt.f32.bf16 %r762, %rs357; + cvt.f32.bf16 %r763, %rs358; + cvt.f32.bf16 %r764, %rs359; + cvt.f32.bf16 %r765, %rs360; + cvt.f32.bf16 %r766, %rs361; + cvt.f32.bf16 %r767, %rs362; + cvt.f32.bf16 %r768, %rs363; + cvt.f32.bf16 %r769, %rs364; + cvt.f32.bf16 %r770, %rs365; + cvt.f32.bf16 %r771, %rs366; + cvt.f32.bf16 %r772, %rs367; + cvt.f32.bf16 %r773, %rs368; + cvt.f32.bf16 %r774, %rs369; + cvt.f32.bf16 %r775, %rs370; + cvt.f32.bf16 %r776, %rs371; + cvt.f32.bf16 %r777, %rs372; + cvt.f32.bf16 %r778, %rs373; + cvt.f32.bf16 %r779, %rs374; + cvt.f32.bf16 %r780, %rs375; + cvt.f32.bf16 %r781, %rs376; + cvt.f32.bf16 %r782, %rs377; + cvt.f32.bf16 %r783, %rs378; + cvt.f32.bf16 %r784, %rs379; + cvt.f32.bf16 %r785, %rs380; + cvt.f32.bf16 %r786, %rs381; + cvt.f32.bf16 %r787, %rs382; + cvt.f32.bf16 %r788, %rs383; + cvt.f32.bf16 %r789, %rs384; + cvt.f32.bf16 %r790, %rs385; + cvt.f32.bf16 %r791, %rs386; + cvt.f32.bf16 %r792, %rs387; + cvt.f32.bf16 %r793, %rs388; + cvt.f32.bf16 %r794, %rs389; + cvt.f32.bf16 %r795, %rs390; + cvt.f32.bf16 %r796, %rs391; + cvt.f32.bf16 %r797, %rs392; + cvt.f32.bf16 %r798, %rs393; + cvt.f32.bf16 %r799, %rs394; + cvt.f32.bf16 %r800, %rs395; + cvt.f32.bf16 %r801, %rs396; + cvt.f32.bf16 %r802, %rs397; + cvt.f32.bf16 %r803, %rs398; + cvt.f32.bf16 %r804, %rs399; + cvt.f32.bf16 %r805, %rs400; + cvt.f32.bf16 %r806, %rs401; + cvt.f32.bf16 %r807, %rs402; + cvt.f32.bf16 %r808, %rs403; + cvt.f32.bf16 %r809, %rs404; + cvt.f32.bf16 %r810, %rs405; + cvt.f32.bf16 %r811, %rs406; + cvt.f32.bf16 %r812, %rs407; + cvt.f32.bf16 %r813, %rs408; + cvt.f32.bf16 %r814, %rs409; + cvt.f32.bf16 %r815, %rs410; + cvt.f32.bf16 %r816, %rs411; + cvt.f32.bf16 %r817, %rs412; + cvt.f32.bf16 %r818, %rs413; + cvt.f32.bf16 %r819, %rs414; + cvt.f32.bf16 %r820, %rs415; + cvt.f32.bf16 %r821, %rs416; + cvt.f32.bf16 %r822, %rs417; + cvt.f32.bf16 %r823, %rs418; + cvt.f32.bf16 %r824, %rs419; + cvt.f32.bf16 %r825, %rs420; + cvt.f32.bf16 %r826, %rs421; + cvt.f32.bf16 %r827, %rs422; + cvt.f32.bf16 %r828, %rs423; + cvt.f32.bf16 %r829, %rs424; + cvt.f32.bf16 %r830, %rs425; + cvt.f32.bf16 %r831, %rs426; + cvt.f32.bf16 %r832, %rs427; + cvt.f32.bf16 %r833, %rs428; + cvt.f32.bf16 %r834, %rs429; + cvt.f32.bf16 %r835, %rs430; + cvt.f32.bf16 %r836, %rs431; + cvt.f32.bf16 %r837, %rs432; + cvt.f32.bf16 %r838, %rs433; + cvt.f32.bf16 %r839, %rs434; + cvt.f32.bf16 %r840, %rs435; + cvt.f32.bf16 %r841, %rs436; + cvt.f32.bf16 %r842, %rs437; + cvt.f32.bf16 %r843, %rs438; + cvt.f32.bf16 %r844, %rs439; + cvt.f32.bf16 %r845, %rs440; + cvt.f32.bf16 %r846, %rs441; + cvt.f32.bf16 %r847, %rs442; + cvt.f32.bf16 %r848, %rs443; + cvt.f32.bf16 %r849, %rs444; + cvt.f32.bf16 %r850, %rs445; + cvt.f32.bf16 %r851, %rs446; + cvt.f32.bf16 %r852, %rs447; + cvt.f32.bf16 %r853, %rs448; + cvt.f32.bf16 %r854, %rs449; + cvt.f32.bf16 %r855, %rs450; + cvt.f32.bf16 %r856, %rs451; + cvt.f32.bf16 %r857, %rs452; + cvt.f32.bf16 %r858, %rs453; + cvt.f32.bf16 %r859, %rs454; + cvt.f32.bf16 %r860, %rs455; + cvt.f32.bf16 %r861, %rs456; + cvt.f32.bf16 %r862, %rs457; + cvt.f32.bf16 %r863, %rs458; + cvt.f32.bf16 %r864, %rs459; + cvt.f32.bf16 %r865, %rs460; + cvt.f32.bf16 %r866, %rs461; + cvt.f32.bf16 %r867, %rs462; + cvt.f32.bf16 %r868, %rs463; + cvt.f32.bf16 %r869, %rs464; + cvt.f32.bf16 %r870, %rs465; + cvt.f32.bf16 %r871, %rs466; + cvt.f32.bf16 %r872, %rs467; + cvt.f32.bf16 %r873, %rs468; + cvt.f32.bf16 %r874, %rs469; + cvt.f32.bf16 %r875, %rs470; + cvt.f32.bf16 %r876, %rs471; + cvt.f32.bf16 %r877, %rs472; + cvt.f32.bf16 %r878, %rs473; + cvt.f32.bf16 %r879, %rs474; + cvt.f32.bf16 %r880, %rs475; + cvt.f32.bf16 %r881, %rs476; + cvt.f32.bf16 %r882, %rs477; + cvt.f32.bf16 %r883, %rs478; + cvt.f32.bf16 %r884, %rs479; + cvt.f32.bf16 %r885, %rs480; + cvt.f32.bf16 %r886, %rs481; + cvt.f32.bf16 %r887, %rs482; + cvt.f32.bf16 %r888, %rs483; + cvt.f32.bf16 %r889, %rs484; + cvt.f32.bf16 %r890, %rs485; + cvt.f32.bf16 %r891, %rs486; + cvt.f32.bf16 %r892, %rs487; + cvt.f32.bf16 %r893, %rs488; + cvt.f32.bf16 %r894, %rs489; + cvt.f32.bf16 %r895, %rs490; + cvt.f32.bf16 %r896, %rs491; + cvt.f32.bf16 %r897, %rs492; + cvt.f32.bf16 %r898, %rs493; + cvt.f32.bf16 %r899, %rs494; + cvt.f32.bf16 %r900, %rs495; + cvt.f32.bf16 %r901, %rs496; + cvt.f32.bf16 %r902, %rs497; + cvt.f32.bf16 %r903, %rs498; + cvt.f32.bf16 %r904, %rs499; + cvt.f32.bf16 %r905, %rs500; + cvt.f32.bf16 %r906, %rs501; + cvt.f32.bf16 %r907, %rs502; + cvt.f32.bf16 %r908, %rs503; + cvt.f32.bf16 %r909, %rs504; + cvt.f32.bf16 %r910, %rs505; + cvt.f32.bf16 %r911, %rs506; + cvt.f32.bf16 %r912, %rs507; + cvt.f32.bf16 %r913, %rs508; + cvt.f32.bf16 %r914, %rs509; + cvt.f32.bf16 %r915, %rs510; + cvt.f32.bf16 %r916, %rs511; + cvt.f32.bf16 %r917, %rs512; + cvt.f32.bf16 %r918, %rs513; + cvt.f32.bf16 %r919, %rs514; + cvt.f32.bf16 %r920, %rs515; + cvt.f32.bf16 %r921, %rs516; + cvt.f32.bf16 %r922, %rs517; + cvt.f32.bf16 %r923, %rs518; + cvt.f32.bf16 %r924, %rs519; + cvt.f32.bf16 %r925, %rs520; + cvt.f32.bf16 %r926, %rs521; + cvt.f32.bf16 %r927, %rs522; + cvt.f32.bf16 %r928, %rs523; + cvt.f32.bf16 %r929, %rs524; + cvt.f32.bf16 %r930, %rs525; + cvt.f32.bf16 %r931, %rs526; + cvt.f32.bf16 %r932, %rs527; + cvt.f32.bf16 %r933, %rs528; + cvt.f32.bf16 %r934, %rs529; + cvt.f32.bf16 %r935, %rs530; + cvt.f32.bf16 %r936, %rs531; + cvt.f32.bf16 %r937, %rs532; + cvt.f32.bf16 %r938, %rs533; + cvt.f32.bf16 %r939, %rs534; + cvt.f32.bf16 %r940, %rs535; + cvt.f32.bf16 %r941, %rs536; + cvt.f32.bf16 %r942, %rs537; + cvt.f32.bf16 %r943, %rs538; + cvt.f32.bf16 %r944, %rs539; + cvt.f32.bf16 %r945, %rs540; + cvt.f32.bf16 %r946, %rs541; + cvt.f32.bf16 %r947, %rs542; + cvt.f32.bf16 %r948, %rs543; + cvt.f32.bf16 %r949, %rs544; + cvt.f32.bf16 %r950, %rs545; + cvt.f32.bf16 %r951, %rs546; + cvt.f32.bf16 %r952, %rs547; + cvt.f32.bf16 %r953, %rs548; + cvt.f32.bf16 %r954, %rs549; + cvt.f32.bf16 %r955, %rs550; + cvt.f32.bf16 %r956, %rs551; + cvt.f32.bf16 %r957, %rs552; + cvt.f32.bf16 %r958, %rs553; + cvt.f32.bf16 %r959, %rs554; + cvt.f32.bf16 %r960, %rs555; + cvt.f32.bf16 %r961, %rs556; + cvt.f32.bf16 %r962, %rs557; + cvt.f32.bf16 %r963, %rs558; + cvt.f32.bf16 %r964, %rs559; + cvt.f32.bf16 %r965, %rs560; + cvt.f32.bf16 %r966, %rs561; + cvt.f32.bf16 %r967, %rs562; + cvt.f32.bf16 %r968, %rs563; + cvt.f32.bf16 %r969, %rs564; + cvt.f32.bf16 %r970, %rs565; + .loc 1 107 25 // l2norm.py:107:25 + shl.b32 %r971, %r427, 4; + mov.b32 %r972, global_smem; + add.s32 %r973, %r972, %r971; + stmatrix.sync.aligned.m8n8.x2.shared.b16 [%r973], {%r130, %r131}; + bar.warp.sync -1; + shl.b32 %r974, %r390, 3; + add.s32 %r975, %r972, %r974; + ld.shared.v4.b32 {%r976, %r977, %r978, %r979}, [%r975]; + ld.shared.v4.b32 {%r980, %r981, %r982, %r983}, [%r975+16]; + ld.shared.v4.b32 {%r984, %r985, %r986, %r987}, [%r975+32]; + ld.shared.v4.b32 {%r988, %r989, %r990, %r991}, [%r975+48]; + ld.shared.v4.b32 {%r992, %r993, %r994, %r995}, [%r975+64]; + ld.shared.v4.b32 {%r996, %r997, %r998, %r999}, [%r975+80]; + ld.shared.v4.b32 {%r1000, %r1001, %r1002, %r1003}, [%r975+96]; + ld.shared.v4.b32 {%r1004, %r1005, %r1006, %r1007}, [%r975+112]; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r1008, %r978, %r731; + mul.f32 %r1009, %r978, %r732; + mul.f32 %r1010, %r978, %r733; + mul.f32 %r1011, %r978, %r734; + mul.f32 %r1012, %r978, %r735; + mul.f32 %r1013, %r978, %r736; + mul.f32 %r1014, %r978, %r737; + mul.f32 %r1015, %r978, %r738; + mul.f32 %r1016, %r979, %r739; + mul.f32 %r1017, %r979, %r740; + mul.f32 %r1018, %r979, %r741; + mul.f32 %r1019, %r979, %r742; + mul.f32 %r1020, %r979, %r743; + mul.f32 %r1021, %r979, %r744; + mul.f32 %r1022, %r979, %r745; + mul.f32 %r1023, %r979, %r746; + mul.f32 %r1024, %r980, %r747; + mul.f32 %r1025, %r980, %r748; + mul.f32 %r1026, %r980, %r749; + mul.f32 %r1027, %r980, %r750; + mul.f32 %r1028, %r980, %r751; + mul.f32 %r1029, %r980, %r752; + mul.f32 %r1030, %r980, %r753; + mul.f32 %r1031, %r980, %r754; + mul.f32 %r1032, %r981, %r755; + mul.f32 %r1033, %r981, %r756; + mul.f32 %r1034, %r981, %r757; + mul.f32 %r1035, %r981, %r758; + mul.f32 %r1036, %r981, %r759; + mul.f32 %r1037, %r981, %r760; + mul.f32 %r1038, %r981, %r761; + mul.f32 %r1039, %r981, %r762; + mul.f32 %r1040, %r982, %r763; + mul.f32 %r1041, %r982, %r764; + mul.f32 %r1042, %r982, %r765; + mul.f32 %r1043, %r982, %r766; + mul.f32 %r1044, %r982, %r767; + mul.f32 %r1045, %r982, %r768; + mul.f32 %r1046, %r982, %r769; + mul.f32 %r1047, %r982, %r770; + mul.f32 %r1048, %r983, %r771; + mul.f32 %r1049, %r983, %r772; + mul.f32 %r1050, %r983, %r773; + mul.f32 %r1051, %r983, %r774; + mul.f32 %r1052, %r983, %r775; + mul.f32 %r1053, %r983, %r776; + mul.f32 %r1054, %r983, %r777; + mul.f32 %r1055, %r983, %r778; + mul.f32 %r1056, %r984, %r779; + mul.f32 %r1057, %r984, %r780; + mul.f32 %r1058, %r984, %r781; + mul.f32 %r1059, %r984, %r782; + mul.f32 %r1060, %r984, %r783; + mul.f32 %r1061, %r984, %r784; + mul.f32 %r1062, %r984, %r785; + mul.f32 %r1063, %r984, %r786; + mul.f32 %r1064, %r985, %r787; + mul.f32 %r1065, %r985, %r788; + mul.f32 %r1066, %r985, %r789; + mul.f32 %r1067, %r985, %r790; + mul.f32 %r1068, %r985, %r791; + mul.f32 %r1069, %r985, %r792; + mul.f32 %r1070, %r985, %r793; + mul.f32 %r1071, %r985, %r794; + mul.f32 %r1072, %r986, %r795; + mul.f32 %r1073, %r986, %r796; + mul.f32 %r1074, %r986, %r797; + mul.f32 %r1075, %r986, %r798; + mul.f32 %r1076, %r986, %r799; + mul.f32 %r1077, %r986, %r800; + mul.f32 %r1078, %r986, %r801; + mul.f32 %r1079, %r986, %r802; + mul.f32 %r1080, %r987, %r803; + mul.f32 %r1081, %r987, %r804; + mul.f32 %r1082, %r987, %r805; + mul.f32 %r1083, %r987, %r806; + mul.f32 %r1084, %r987, %r807; + mul.f32 %r1085, %r987, %r808; + mul.f32 %r1086, %r987, %r809; + mul.f32 %r1087, %r987, %r810; + mul.f32 %r1088, %r988, %r811; + mul.f32 %r1089, %r988, %r812; + mul.f32 %r1090, %r988, %r813; + mul.f32 %r1091, %r988, %r814; + mul.f32 %r1092, %r988, %r815; + mul.f32 %r1093, %r988, %r816; + mul.f32 %r1094, %r988, %r817; + mul.f32 %r1095, %r988, %r818; + mul.f32 %r1096, %r989, %r819; + mul.f32 %r1097, %r989, %r820; + mul.f32 %r1098, %r989, %r821; + mul.f32 %r1099, %r989, %r822; + mul.f32 %r1100, %r989, %r823; + mul.f32 %r1101, %r989, %r824; + mul.f32 %r1102, %r989, %r825; + mul.f32 %r1103, %r989, %r826; + mul.f32 %r1104, %r990, %r827; + mul.f32 %r1105, %r990, %r828; + mul.f32 %r1106, %r990, %r829; + mul.f32 %r1107, %r990, %r830; + mul.f32 %r1108, %r990, %r831; + mul.f32 %r1109, %r990, %r832; + mul.f32 %r1110, %r990, %r833; + mul.f32 %r1111, %r990, %r834; + mul.f32 %r1112, %r991, %r835; + mul.f32 %r1113, %r991, %r836; + mul.f32 %r1114, %r991, %r837; + mul.f32 %r1115, %r991, %r838; + mul.f32 %r1116, %r991, %r839; + mul.f32 %r1117, %r991, %r840; + mul.f32 %r1118, %r991, %r841; + mul.f32 %r1119, %r991, %r842; + mul.f32 %r1120, %r992, %r843; + mul.f32 %r1121, %r992, %r844; + mul.f32 %r1122, %r992, %r845; + mul.f32 %r1123, %r992, %r846; + mul.f32 %r1124, %r992, %r847; + mul.f32 %r1125, %r992, %r848; + mul.f32 %r1126, %r992, %r849; + mul.f32 %r1127, %r992, %r850; + mul.f32 %r1128, %r993, %r851; + mul.f32 %r1129, %r993, %r852; + mul.f32 %r1130, %r993, %r853; + mul.f32 %r1131, %r993, %r854; + mul.f32 %r1132, %r993, %r855; + mul.f32 %r1133, %r993, %r856; + mul.f32 %r1134, %r993, %r857; + mul.f32 %r1135, %r993, %r858; + mul.f32 %r1136, %r994, %r859; + mul.f32 %r1137, %r994, %r860; + mul.f32 %r1138, %r994, %r861; + mul.f32 %r1139, %r994, %r862; + mul.f32 %r1140, %r994, %r863; + mul.f32 %r1141, %r994, %r864; + mul.f32 %r1142, %r994, %r865; + mul.f32 %r1143, %r994, %r866; + mul.f32 %r1144, %r995, %r867; + mul.f32 %r1145, %r995, %r868; + mul.f32 %r1146, %r995, %r869; + mul.f32 %r1147, %r995, %r870; + mul.f32 %r1148, %r995, %r871; + mul.f32 %r1149, %r995, %r872; + mul.f32 %r1150, %r995, %r873; + mul.f32 %r1151, %r995, %r874; + mul.f32 %r1152, %r996, %r875; + mul.f32 %r1153, %r996, %r876; + mul.f32 %r1154, %r996, %r877; + mul.f32 %r1155, %r996, %r878; + mul.f32 %r1156, %r996, %r879; + mul.f32 %r1157, %r996, %r880; + mul.f32 %r1158, %r996, %r881; + mul.f32 %r1159, %r996, %r882; + mul.f32 %r1160, %r997, %r883; + mul.f32 %r1161, %r997, %r884; + mul.f32 %r1162, %r997, %r885; + mul.f32 %r1163, %r997, %r886; + mul.f32 %r1164, %r997, %r887; + mul.f32 %r1165, %r997, %r888; + mul.f32 %r1166, %r997, %r889; + mul.f32 %r1167, %r997, %r890; + mul.f32 %r1168, %r998, %r891; + mul.f32 %r1169, %r998, %r892; + mul.f32 %r1170, %r998, %r893; + mul.f32 %r1171, %r998, %r894; + mul.f32 %r1172, %r998, %r895; + mul.f32 %r1173, %r998, %r896; + mul.f32 %r1174, %r998, %r897; + mul.f32 %r1175, %r998, %r898; + mul.f32 %r1176, %r999, %r899; + mul.f32 %r1177, %r999, %r900; + mul.f32 %r1178, %r999, %r901; + mul.f32 %r1179, %r999, %r902; + mul.f32 %r1180, %r999, %r903; + mul.f32 %r1181, %r999, %r904; + mul.f32 %r1182, %r999, %r905; + mul.f32 %r1183, %r999, %r906; + mul.f32 %r1184, %r1000, %r907; + mul.f32 %r1185, %r1000, %r908; + mul.f32 %r1186, %r1000, %r909; + mul.f32 %r1187, %r1000, %r910; + mul.f32 %r1188, %r1000, %r911; + mul.f32 %r1189, %r1000, %r912; + mul.f32 %r1190, %r1000, %r913; + mul.f32 %r1191, %r1000, %r914; + mul.f32 %r1192, %r1001, %r915; + mul.f32 %r1193, %r1001, %r916; + mul.f32 %r1194, %r1001, %r917; + mul.f32 %r1195, %r1001, %r918; + mul.f32 %r1196, %r1001, %r919; + mul.f32 %r1197, %r1001, %r920; + mul.f32 %r1198, %r1001, %r921; + mul.f32 %r1199, %r1001, %r922; + mul.f32 %r1200, %r1002, %r923; + mul.f32 %r1201, %r1002, %r924; + mul.f32 %r1202, %r1002, %r925; + mul.f32 %r1203, %r1002, %r926; + mul.f32 %r1204, %r1002, %r927; + mul.f32 %r1205, %r1002, %r928; + mul.f32 %r1206, %r1002, %r929; + mul.f32 %r1207, %r1002, %r930; + mul.f32 %r1208, %r1003, %r931; + mul.f32 %r1209, %r1003, %r932; + mul.f32 %r1210, %r1003, %r933; + mul.f32 %r1211, %r1003, %r934; + mul.f32 %r1212, %r1003, %r935; + mul.f32 %r1213, %r1003, %r936; + mul.f32 %r1214, %r1003, %r937; + mul.f32 %r1215, %r1003, %r938; + mul.f32 %r1216, %r1004, %r939; + mul.f32 %r1217, %r1004, %r940; + mul.f32 %r1218, %r1004, %r941; + mul.f32 %r1219, %r1004, %r942; + mul.f32 %r1220, %r1004, %r943; + mul.f32 %r1221, %r1004, %r944; + mul.f32 %r1222, %r1004, %r945; + mul.f32 %r1223, %r1004, %r946; + mul.f32 %r1224, %r1005, %r947; + mul.f32 %r1225, %r1005, %r948; + mul.f32 %r1226, %r1005, %r949; + mul.f32 %r1227, %r1005, %r950; + mul.f32 %r1228, %r1005, %r951; + mul.f32 %r1229, %r1005, %r952; + mul.f32 %r1230, %r1005, %r953; + mul.f32 %r1231, %r1005, %r954; + mul.f32 %r1232, %r1006, %r955; + mul.f32 %r1233, %r1006, %r956; + mul.f32 %r1234, %r1006, %r957; + mul.f32 %r1235, %r1006, %r958; + mul.f32 %r1236, %r1006, %r959; + mul.f32 %r1237, %r1006, %r960; + mul.f32 %r1238, %r1006, %r961; + mul.f32 %r1239, %r1006, %r962; + mul.f32 %r1240, %r1007, %r963; + mul.f32 %r1241, %r1007, %r964; + mul.f32 %r1242, %r1007, %r965; + mul.f32 %r1243, %r1007, %r966; + mul.f32 %r1244, %r1007, %r967; + mul.f32 %r1245, %r1007, %r968; + mul.f32 %r1246, %r1007, %r969; + mul.f32 %r1247, %r1007, %r970; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r1248, %r458, %r716; + mul.f32 %r1249, %r466, %r724; + mul.f32 %r1250, %r474, %r732; + mul.f32 %r1251, %r482, %r740; + mul.f32 %r1252, %r490, %r748; + mul.f32 %r1253, %r498, %r756; + mul.f32 %r1254, %r506, %r764; + mul.f32 %r1255, %r514, %r772; + mul.f32 %r1256, %r522, %r780; + mul.f32 %r1257, %r530, %r788; + mul.f32 %r1258, %r538, %r796; + mul.f32 %r1259, %r546, %r804; + mul.f32 %r1260, %r554, %r812; + mul.f32 %r1261, %r562, %r820; + mul.f32 %r1262, %r570, %r828; + mul.f32 %r1263, %r578, %r836; + mul.f32 %r1264, %r586, %r844; + mul.f32 %r1265, %r594, %r852; + mul.f32 %r1266, %r602, %r860; + mul.f32 %r1267, %r610, %r868; + mul.f32 %r1268, %r618, %r876; + mul.f32 %r1269, %r626, %r884; + mul.f32 %r1270, %r634, %r892; + mul.f32 %r1271, %r642, %r900; + mul.f32 %r1272, %r650, %r908; + mul.f32 %r1273, %r658, %r916; + mul.f32 %r1274, %r666, %r924; + mul.f32 %r1275, %r674, %r932; + mul.f32 %r1276, %r682, %r940; + mul.f32 %r1277, %r690, %r948; + mul.f32 %r1278, %r698, %r956; + mul.f32 %r1279, %r706, %r964; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1280, %rs54, %rs310, %r1248; + fma.rn.f32.bf16 %r1281, %rs56, %rs312, %r1280; + fma.rn.f32.bf16 %r1282, %rs57, %rs313, %r1281; + fma.rn.f32.bf16 %r1283, %rs58, %rs314, %r1282; + fma.rn.f32.bf16 %r1284, %rs59, %rs315, %r1283; + fma.rn.f32.bf16 %r1285, %rs60, %rs316, %r1284; + fma.rn.f32.bf16 %r1286, %rs61, %rs317, %r1285; + fma.rn.f32.bf16 %r1287, %rs62, %rs318, %r1249; + fma.rn.f32.bf16 %r1288, %rs64, %rs320, %r1287; + fma.rn.f32.bf16 %r1289, %rs65, %rs321, %r1288; + fma.rn.f32.bf16 %r1290, %rs66, %rs322, %r1289; + fma.rn.f32.bf16 %r1291, %rs67, %rs323, %r1290; + fma.rn.f32.bf16 %r1292, %rs68, %rs324, %r1291; + fma.rn.f32.bf16 %r1293, %rs69, %rs325, %r1292; + fma.rn.f32.bf16 %r1294, %rs70, %rs326, %r1250; + fma.rn.f32.bf16 %r1295, %rs72, %rs328, %r1294; + fma.rn.f32.bf16 %r1296, %rs73, %rs329, %r1295; + fma.rn.f32.bf16 %r1297, %rs74, %rs330, %r1296; + fma.rn.f32.bf16 %r1298, %rs75, %rs331, %r1297; + fma.rn.f32.bf16 %r1299, %rs76, %rs332, %r1298; + fma.rn.f32.bf16 %r1300, %rs77, %rs333, %r1299; + fma.rn.f32.bf16 %r1301, %rs78, %rs334, %r1251; + fma.rn.f32.bf16 %r1302, %rs80, %rs336, %r1301; + fma.rn.f32.bf16 %r1303, %rs81, %rs337, %r1302; + fma.rn.f32.bf16 %r1304, %rs82, %rs338, %r1303; + fma.rn.f32.bf16 %r1305, %rs83, %rs339, %r1304; + fma.rn.f32.bf16 %r1306, %rs84, %rs340, %r1305; + fma.rn.f32.bf16 %r1307, %rs85, %rs341, %r1306; + fma.rn.f32.bf16 %r1308, %rs86, %rs342, %r1252; + fma.rn.f32.bf16 %r1309, %rs88, %rs344, %r1308; + fma.rn.f32.bf16 %r1310, %rs89, %rs345, %r1309; + fma.rn.f32.bf16 %r1311, %rs90, %rs346, %r1310; + fma.rn.f32.bf16 %r1312, %rs91, %rs347, %r1311; + fma.rn.f32.bf16 %r1313, %rs92, %rs348, %r1312; + fma.rn.f32.bf16 %r1314, %rs93, %rs349, %r1313; + fma.rn.f32.bf16 %r1315, %rs94, %rs350, %r1253; + fma.rn.f32.bf16 %r1316, %rs96, %rs352, %r1315; + fma.rn.f32.bf16 %r1317, %rs97, %rs353, %r1316; + fma.rn.f32.bf16 %r1318, %rs98, %rs354, %r1317; + fma.rn.f32.bf16 %r1319, %rs99, %rs355, %r1318; + fma.rn.f32.bf16 %r1320, %rs100, %rs356, %r1319; + fma.rn.f32.bf16 %r1321, %rs101, %rs357, %r1320; + fma.rn.f32.bf16 %r1322, %rs102, %rs358, %r1254; + fma.rn.f32.bf16 %r1323, %rs104, %rs360, %r1322; + fma.rn.f32.bf16 %r1324, %rs105, %rs361, %r1323; + fma.rn.f32.bf16 %r1325, %rs106, %rs362, %r1324; + fma.rn.f32.bf16 %r1326, %rs107, %rs363, %r1325; + fma.rn.f32.bf16 %r1327, %rs108, %rs364, %r1326; + fma.rn.f32.bf16 %r1328, %rs109, %rs365, %r1327; + fma.rn.f32.bf16 %r1329, %rs110, %rs366, %r1255; + fma.rn.f32.bf16 %r1330, %rs112, %rs368, %r1329; + fma.rn.f32.bf16 %r1331, %rs113, %rs369, %r1330; + fma.rn.f32.bf16 %r1332, %rs114, %rs370, %r1331; + fma.rn.f32.bf16 %r1333, %rs115, %rs371, %r1332; + fma.rn.f32.bf16 %r1334, %rs116, %rs372, %r1333; + fma.rn.f32.bf16 %r1335, %rs117, %rs373, %r1334; + fma.rn.f32.bf16 %r1336, %rs118, %rs374, %r1256; + fma.rn.f32.bf16 %r1337, %rs120, %rs376, %r1336; + fma.rn.f32.bf16 %r1338, %rs121, %rs377, %r1337; + fma.rn.f32.bf16 %r1339, %rs122, %rs378, %r1338; + fma.rn.f32.bf16 %r1340, %rs123, %rs379, %r1339; + fma.rn.f32.bf16 %r1341, %rs124, %rs380, %r1340; + fma.rn.f32.bf16 %r1342, %rs125, %rs381, %r1341; + fma.rn.f32.bf16 %r1343, %rs126, %rs382, %r1257; + fma.rn.f32.bf16 %r1344, %rs128, %rs384, %r1343; + fma.rn.f32.bf16 %r1345, %rs129, %rs385, %r1344; + fma.rn.f32.bf16 %r1346, %rs130, %rs386, %r1345; + fma.rn.f32.bf16 %r1347, %rs131, %rs387, %r1346; + fma.rn.f32.bf16 %r1348, %rs132, %rs388, %r1347; + fma.rn.f32.bf16 %r1349, %rs133, %rs389, %r1348; + fma.rn.f32.bf16 %r1350, %rs134, %rs390, %r1258; + fma.rn.f32.bf16 %r1351, %rs136, %rs392, %r1350; + fma.rn.f32.bf16 %r1352, %rs137, %rs393, %r1351; + fma.rn.f32.bf16 %r1353, %rs138, %rs394, %r1352; + fma.rn.f32.bf16 %r1354, %rs139, %rs395, %r1353; + fma.rn.f32.bf16 %r1355, %rs140, %rs396, %r1354; + fma.rn.f32.bf16 %r1356, %rs141, %rs397, %r1355; + fma.rn.f32.bf16 %r1357, %rs142, %rs398, %r1259; + fma.rn.f32.bf16 %r1358, %rs144, %rs400, %r1357; + fma.rn.f32.bf16 %r1359, %rs145, %rs401, %r1358; + fma.rn.f32.bf16 %r1360, %rs146, %rs402, %r1359; + fma.rn.f32.bf16 %r1361, %rs147, %rs403, %r1360; + fma.rn.f32.bf16 %r1362, %rs148, %rs404, %r1361; + fma.rn.f32.bf16 %r1363, %rs149, %rs405, %r1362; + fma.rn.f32.bf16 %r1364, %rs150, %rs406, %r1260; + fma.rn.f32.bf16 %r1365, %rs152, %rs408, %r1364; + fma.rn.f32.bf16 %r1366, %rs153, %rs409, %r1365; + fma.rn.f32.bf16 %r1367, %rs154, %rs410, %r1366; + fma.rn.f32.bf16 %r1368, %rs155, %rs411, %r1367; + fma.rn.f32.bf16 %r1369, %rs156, %rs412, %r1368; + fma.rn.f32.bf16 %r1370, %rs157, %rs413, %r1369; + fma.rn.f32.bf16 %r1371, %rs158, %rs414, %r1261; + fma.rn.f32.bf16 %r1372, %rs160, %rs416, %r1371; + fma.rn.f32.bf16 %r1373, %rs161, %rs417, %r1372; + fma.rn.f32.bf16 %r1374, %rs162, %rs418, %r1373; + fma.rn.f32.bf16 %r1375, %rs163, %rs419, %r1374; + fma.rn.f32.bf16 %r1376, %rs164, %rs420, %r1375; + fma.rn.f32.bf16 %r1377, %rs165, %rs421, %r1376; + fma.rn.f32.bf16 %r1378, %rs166, %rs422, %r1262; + fma.rn.f32.bf16 %r1379, %rs168, %rs424, %r1378; + fma.rn.f32.bf16 %r1380, %rs169, %rs425, %r1379; + fma.rn.f32.bf16 %r1381, %rs170, %rs426, %r1380; + fma.rn.f32.bf16 %r1382, %rs171, %rs427, %r1381; + fma.rn.f32.bf16 %r1383, %rs172, %rs428, %r1382; + fma.rn.f32.bf16 %r1384, %rs173, %rs429, %r1383; + fma.rn.f32.bf16 %r1385, %rs174, %rs430, %r1263; + fma.rn.f32.bf16 %r1386, %rs176, %rs432, %r1385; + fma.rn.f32.bf16 %r1387, %rs177, %rs433, %r1386; + fma.rn.f32.bf16 %r1388, %rs178, %rs434, %r1387; + fma.rn.f32.bf16 %r1389, %rs179, %rs435, %r1388; + fma.rn.f32.bf16 %r1390, %rs180, %rs436, %r1389; + fma.rn.f32.bf16 %r1391, %rs181, %rs437, %r1390; + fma.rn.f32.bf16 %r1392, %rs182, %rs438, %r1264; + fma.rn.f32.bf16 %r1393, %rs184, %rs440, %r1392; + fma.rn.f32.bf16 %r1394, %rs185, %rs441, %r1393; + fma.rn.f32.bf16 %r1395, %rs186, %rs442, %r1394; + fma.rn.f32.bf16 %r1396, %rs187, %rs443, %r1395; + fma.rn.f32.bf16 %r1397, %rs188, %rs444, %r1396; + fma.rn.f32.bf16 %r1398, %rs189, %rs445, %r1397; + fma.rn.f32.bf16 %r1399, %rs190, %rs446, %r1265; + fma.rn.f32.bf16 %r1400, %rs192, %rs448, %r1399; + fma.rn.f32.bf16 %r1401, %rs193, %rs449, %r1400; + fma.rn.f32.bf16 %r1402, %rs194, %rs450, %r1401; + fma.rn.f32.bf16 %r1403, %rs195, %rs451, %r1402; + fma.rn.f32.bf16 %r1404, %rs196, %rs452, %r1403; + fma.rn.f32.bf16 %r1405, %rs197, %rs453, %r1404; + fma.rn.f32.bf16 %r1406, %rs198, %rs454, %r1266; + fma.rn.f32.bf16 %r1407, %rs200, %rs456, %r1406; + fma.rn.f32.bf16 %r1408, %rs201, %rs457, %r1407; + fma.rn.f32.bf16 %r1409, %rs202, %rs458, %r1408; + fma.rn.f32.bf16 %r1410, %rs203, %rs459, %r1409; + fma.rn.f32.bf16 %r1411, %rs204, %rs460, %r1410; + fma.rn.f32.bf16 %r1412, %rs205, %rs461, %r1411; + fma.rn.f32.bf16 %r1413, %rs206, %rs462, %r1267; + fma.rn.f32.bf16 %r1414, %rs208, %rs464, %r1413; + fma.rn.f32.bf16 %r1415, %rs209, %rs465, %r1414; + fma.rn.f32.bf16 %r1416, %rs210, %rs466, %r1415; + fma.rn.f32.bf16 %r1417, %rs211, %rs467, %r1416; + fma.rn.f32.bf16 %r1418, %rs212, %rs468, %r1417; + fma.rn.f32.bf16 %r1419, %rs213, %rs469, %r1418; + fma.rn.f32.bf16 %r1420, %rs214, %rs470, %r1268; + fma.rn.f32.bf16 %r1421, %rs216, %rs472, %r1420; + fma.rn.f32.bf16 %r1422, %rs217, %rs473, %r1421; + fma.rn.f32.bf16 %r1423, %rs218, %rs474, %r1422; + fma.rn.f32.bf16 %r1424, %rs219, %rs475, %r1423; + fma.rn.f32.bf16 %r1425, %rs220, %rs476, %r1424; + fma.rn.f32.bf16 %r1426, %rs221, %rs477, %r1425; + fma.rn.f32.bf16 %r1427, %rs222, %rs478, %r1269; + fma.rn.f32.bf16 %r1428, %rs224, %rs480, %r1427; + fma.rn.f32.bf16 %r1429, %rs225, %rs481, %r1428; + fma.rn.f32.bf16 %r1430, %rs226, %rs482, %r1429; + fma.rn.f32.bf16 %r1431, %rs227, %rs483, %r1430; + fma.rn.f32.bf16 %r1432, %rs228, %rs484, %r1431; + fma.rn.f32.bf16 %r1433, %rs229, %rs485, %r1432; + fma.rn.f32.bf16 %r1434, %rs230, %rs486, %r1270; + fma.rn.f32.bf16 %r1435, %rs232, %rs488, %r1434; + fma.rn.f32.bf16 %r1436, %rs233, %rs489, %r1435; + fma.rn.f32.bf16 %r1437, %rs234, %rs490, %r1436; + fma.rn.f32.bf16 %r1438, %rs235, %rs491, %r1437; + fma.rn.f32.bf16 %r1439, %rs236, %rs492, %r1438; + fma.rn.f32.bf16 %r1440, %rs237, %rs493, %r1439; + fma.rn.f32.bf16 %r1441, %rs238, %rs494, %r1271; + fma.rn.f32.bf16 %r1442, %rs240, %rs496, %r1441; + fma.rn.f32.bf16 %r1443, %rs241, %rs497, %r1442; + fma.rn.f32.bf16 %r1444, %rs242, %rs498, %r1443; + fma.rn.f32.bf16 %r1445, %rs243, %rs499, %r1444; + fma.rn.f32.bf16 %r1446, %rs244, %rs500, %r1445; + fma.rn.f32.bf16 %r1447, %rs245, %rs501, %r1446; + fma.rn.f32.bf16 %r1448, %rs246, %rs502, %r1272; + fma.rn.f32.bf16 %r1449, %rs248, %rs504, %r1448; + fma.rn.f32.bf16 %r1450, %rs249, %rs505, %r1449; + fma.rn.f32.bf16 %r1451, %rs250, %rs506, %r1450; + fma.rn.f32.bf16 %r1452, %rs251, %rs507, %r1451; + fma.rn.f32.bf16 %r1453, %rs252, %rs508, %r1452; + fma.rn.f32.bf16 %r1454, %rs253, %rs509, %r1453; + fma.rn.f32.bf16 %r1455, %rs254, %rs510, %r1273; + fma.rn.f32.bf16 %r1456, %rs256, %rs512, %r1455; + fma.rn.f32.bf16 %r1457, %rs257, %rs513, %r1456; + fma.rn.f32.bf16 %r1458, %rs258, %rs514, %r1457; + fma.rn.f32.bf16 %r1459, %rs259, %rs515, %r1458; + fma.rn.f32.bf16 %r1460, %rs260, %rs516, %r1459; + fma.rn.f32.bf16 %r1461, %rs261, %rs517, %r1460; + fma.rn.f32.bf16 %r1462, %rs262, %rs518, %r1274; + fma.rn.f32.bf16 %r1463, %rs264, %rs520, %r1462; + fma.rn.f32.bf16 %r1464, %rs265, %rs521, %r1463; + fma.rn.f32.bf16 %r1465, %rs266, %rs522, %r1464; + fma.rn.f32.bf16 %r1466, %rs267, %rs523, %r1465; + fma.rn.f32.bf16 %r1467, %rs268, %rs524, %r1466; + fma.rn.f32.bf16 %r1468, %rs269, %rs525, %r1467; + fma.rn.f32.bf16 %r1469, %rs270, %rs526, %r1275; + fma.rn.f32.bf16 %r1470, %rs272, %rs528, %r1469; + fma.rn.f32.bf16 %r1471, %rs273, %rs529, %r1470; + fma.rn.f32.bf16 %r1472, %rs274, %rs530, %r1471; + fma.rn.f32.bf16 %r1473, %rs275, %rs531, %r1472; + fma.rn.f32.bf16 %r1474, %rs276, %rs532, %r1473; + fma.rn.f32.bf16 %r1475, %rs277, %rs533, %r1474; + fma.rn.f32.bf16 %r1476, %rs278, %rs534, %r1276; + fma.rn.f32.bf16 %r1477, %rs280, %rs536, %r1476; + fma.rn.f32.bf16 %r1478, %rs281, %rs537, %r1477; + fma.rn.f32.bf16 %r1479, %rs282, %rs538, %r1478; + fma.rn.f32.bf16 %r1480, %rs283, %rs539, %r1479; + fma.rn.f32.bf16 %r1481, %rs284, %rs540, %r1480; + fma.rn.f32.bf16 %r1482, %rs285, %rs541, %r1481; + fma.rn.f32.bf16 %r1483, %rs286, %rs542, %r1277; + fma.rn.f32.bf16 %r1484, %rs288, %rs544, %r1483; + fma.rn.f32.bf16 %r1485, %rs289, %rs545, %r1484; + fma.rn.f32.bf16 %r1486, %rs290, %rs546, %r1485; + fma.rn.f32.bf16 %r1487, %rs291, %rs547, %r1486; + fma.rn.f32.bf16 %r1488, %rs292, %rs548, %r1487; + fma.rn.f32.bf16 %r1489, %rs293, %rs549, %r1488; + fma.rn.f32.bf16 %r1490, %rs294, %rs550, %r1278; + fma.rn.f32.bf16 %r1491, %rs296, %rs552, %r1490; + fma.rn.f32.bf16 %r1492, %rs297, %rs553, %r1491; + fma.rn.f32.bf16 %r1493, %rs298, %rs554, %r1492; + fma.rn.f32.bf16 %r1494, %rs299, %rs555, %r1493; + fma.rn.f32.bf16 %r1495, %rs300, %rs556, %r1494; + fma.rn.f32.bf16 %r1496, %rs301, %rs557, %r1495; + fma.rn.f32.bf16 %r1497, %rs302, %rs558, %r1279; + fma.rn.f32.bf16 %r1498, %rs304, %rs560, %r1497; + fma.rn.f32.bf16 %r1499, %rs305, %rs561, %r1498; + fma.rn.f32.bf16 %r1500, %rs306, %rs562, %r1499; + fma.rn.f32.bf16 %r1501, %rs307, %rs563, %r1500; + fma.rn.f32.bf16 %r1502, %rs308, %rs564, %r1501; + fma.rn.f32.bf16 %r1503, %rs309, %rs565, %r1502; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1504, %r1286, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1505, %r1286, %r1504; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1506, %r1505, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1507, %r1505, %r1506; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1508, %r1507, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1509, %r1507, %r1508; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1510, %r1509, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1511, %r1509, %r1510; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1512, %r1293, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1513, %r1293, %r1512; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1514, %r1513, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1515, %r1513, %r1514; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1516, %r1515, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1517, %r1515, %r1516; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1518, %r1517, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1519, %r1517, %r1518; +$L__tmp18: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1520, %r1300, 8, 31, -1; +$L__tmp19: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1521, %r1300, %r1520; +$L__tmp20: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1522, %r1521, 4, 31, -1; +$L__tmp21: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1523, %r1521, %r1522; +$L__tmp22: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1524, %r1523, 2, 31, -1; +$L__tmp23: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1525, %r1523, %r1524; +$L__tmp24: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1526, %r1525, 1, 31, -1; +$L__tmp25: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1527, %r1525, %r1526; +$L__tmp26: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1528, %r1307, 8, 31, -1; +$L__tmp27: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1529, %r1307, %r1528; +$L__tmp28: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1530, %r1529, 4, 31, -1; +$L__tmp29: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1531, %r1529, %r1530; +$L__tmp30: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1532, %r1531, 2, 31, -1; +$L__tmp31: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1533, %r1531, %r1532; +$L__tmp32: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1534, %r1533, 1, 31, -1; +$L__tmp33: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1535, %r1533, %r1534; +$L__tmp34: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1536, %r1314, 8, 31, -1; +$L__tmp35: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1537, %r1314, %r1536; +$L__tmp36: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1538, %r1537, 4, 31, -1; +$L__tmp37: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1539, %r1537, %r1538; +$L__tmp38: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1540, %r1539, 2, 31, -1; +$L__tmp39: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1541, %r1539, %r1540; +$L__tmp40: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1542, %r1541, 1, 31, -1; +$L__tmp41: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1543, %r1541, %r1542; +$L__tmp42: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1544, %r1321, 8, 31, -1; +$L__tmp43: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1545, %r1321, %r1544; +$L__tmp44: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1546, %r1545, 4, 31, -1; +$L__tmp45: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1547, %r1545, %r1546; +$L__tmp46: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1548, %r1547, 2, 31, -1; +$L__tmp47: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1549, %r1547, %r1548; +$L__tmp48: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1550, %r1549, 1, 31, -1; +$L__tmp49: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1551, %r1549, %r1550; +$L__tmp50: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1552, %r1328, 8, 31, -1; +$L__tmp51: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1553, %r1328, %r1552; +$L__tmp52: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1554, %r1553, 4, 31, -1; +$L__tmp53: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1555, %r1553, %r1554; +$L__tmp54: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1556, %r1555, 2, 31, -1; +$L__tmp55: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1557, %r1555, %r1556; +$L__tmp56: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1558, %r1557, 1, 31, -1; +$L__tmp57: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1559, %r1557, %r1558; +$L__tmp58: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1560, %r1335, 8, 31, -1; +$L__tmp59: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1561, %r1335, %r1560; +$L__tmp60: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1562, %r1561, 4, 31, -1; +$L__tmp61: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1563, %r1561, %r1562; +$L__tmp62: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1564, %r1563, 2, 31, -1; +$L__tmp63: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1565, %r1563, %r1564; +$L__tmp64: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1566, %r1565, 1, 31, -1; +$L__tmp65: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1567, %r1565, %r1566; +$L__tmp66: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1568, %r1342, 8, 31, -1; +$L__tmp67: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1569, %r1342, %r1568; +$L__tmp68: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1570, %r1569, 4, 31, -1; +$L__tmp69: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1571, %r1569, %r1570; +$L__tmp70: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1572, %r1571, 2, 31, -1; +$L__tmp71: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1573, %r1571, %r1572; +$L__tmp72: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1574, %r1573, 1, 31, -1; +$L__tmp73: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1575, %r1573, %r1574; +$L__tmp74: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1576, %r1349, 8, 31, -1; +$L__tmp75: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1577, %r1349, %r1576; +$L__tmp76: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1578, %r1577, 4, 31, -1; +$L__tmp77: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1579, %r1577, %r1578; +$L__tmp78: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1580, %r1579, 2, 31, -1; +$L__tmp79: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1581, %r1579, %r1580; +$L__tmp80: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1582, %r1581, 1, 31, -1; +$L__tmp81: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1583, %r1581, %r1582; +$L__tmp82: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1584, %r1356, 8, 31, -1; +$L__tmp83: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1585, %r1356, %r1584; +$L__tmp84: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1586, %r1585, 4, 31, -1; +$L__tmp85: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1587, %r1585, %r1586; +$L__tmp86: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1588, %r1587, 2, 31, -1; +$L__tmp87: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1589, %r1587, %r1588; +$L__tmp88: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1590, %r1589, 1, 31, -1; +$L__tmp89: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1591, %r1589, %r1590; +$L__tmp90: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1592, %r1363, 8, 31, -1; +$L__tmp91: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1593, %r1363, %r1592; +$L__tmp92: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1594, %r1593, 4, 31, -1; +$L__tmp93: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1595, %r1593, %r1594; +$L__tmp94: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1596, %r1595, 2, 31, -1; +$L__tmp95: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1597, %r1595, %r1596; +$L__tmp96: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1598, %r1597, 1, 31, -1; +$L__tmp97: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1599, %r1597, %r1598; +$L__tmp98: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1600, %r1370, 8, 31, -1; +$L__tmp99: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1601, %r1370, %r1600; +$L__tmp100: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1602, %r1601, 4, 31, -1; +$L__tmp101: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1603, %r1601, %r1602; +$L__tmp102: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1604, %r1603, 2, 31, -1; +$L__tmp103: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1605, %r1603, %r1604; +$L__tmp104: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1606, %r1605, 1, 31, -1; +$L__tmp105: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1607, %r1605, %r1606; +$L__tmp106: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1608, %r1377, 8, 31, -1; +$L__tmp107: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1609, %r1377, %r1608; +$L__tmp108: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1610, %r1609, 4, 31, -1; +$L__tmp109: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1611, %r1609, %r1610; +$L__tmp110: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1612, %r1611, 2, 31, -1; +$L__tmp111: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1613, %r1611, %r1612; +$L__tmp112: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1614, %r1613, 1, 31, -1; +$L__tmp113: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1615, %r1613, %r1614; +$L__tmp114: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1616, %r1384, 8, 31, -1; +$L__tmp115: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1617, %r1384, %r1616; +$L__tmp116: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1618, %r1617, 4, 31, -1; +$L__tmp117: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1619, %r1617, %r1618; +$L__tmp118: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1620, %r1619, 2, 31, -1; +$L__tmp119: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1621, %r1619, %r1620; +$L__tmp120: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1622, %r1621, 1, 31, -1; +$L__tmp121: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1623, %r1621, %r1622; +$L__tmp122: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1624, %r1391, 8, 31, -1; +$L__tmp123: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1625, %r1391, %r1624; +$L__tmp124: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1626, %r1625, 4, 31, -1; +$L__tmp125: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1627, %r1625, %r1626; +$L__tmp126: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1628, %r1627, 2, 31, -1; +$L__tmp127: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1629, %r1627, %r1628; +$L__tmp128: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1630, %r1629, 1, 31, -1; +$L__tmp129: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1631, %r1629, %r1630; +$L__tmp130: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1632, %r1398, 8, 31, -1; +$L__tmp131: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1633, %r1398, %r1632; +$L__tmp132: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1634, %r1633, 4, 31, -1; +$L__tmp133: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1635, %r1633, %r1634; +$L__tmp134: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1636, %r1635, 2, 31, -1; +$L__tmp135: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1637, %r1635, %r1636; +$L__tmp136: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1638, %r1637, 1, 31, -1; +$L__tmp137: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1639, %r1637, %r1638; +$L__tmp138: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1640, %r1405, 8, 31, -1; +$L__tmp139: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1641, %r1405, %r1640; +$L__tmp140: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1642, %r1641, 4, 31, -1; +$L__tmp141: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1643, %r1641, %r1642; +$L__tmp142: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1644, %r1643, 2, 31, -1; +$L__tmp143: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1645, %r1643, %r1644; +$L__tmp144: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1646, %r1645, 1, 31, -1; +$L__tmp145: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1647, %r1645, %r1646; +$L__tmp146: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1648, %r1412, 8, 31, -1; +$L__tmp147: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1649, %r1412, %r1648; +$L__tmp148: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1650, %r1649, 4, 31, -1; +$L__tmp149: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1651, %r1649, %r1650; +$L__tmp150: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1652, %r1651, 2, 31, -1; +$L__tmp151: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1653, %r1651, %r1652; +$L__tmp152: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1654, %r1653, 1, 31, -1; +$L__tmp153: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1655, %r1653, %r1654; +$L__tmp154: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1656, %r1419, 8, 31, -1; +$L__tmp155: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1657, %r1419, %r1656; +$L__tmp156: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1658, %r1657, 4, 31, -1; +$L__tmp157: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1659, %r1657, %r1658; +$L__tmp158: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1660, %r1659, 2, 31, -1; +$L__tmp159: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1661, %r1659, %r1660; +$L__tmp160: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1662, %r1661, 1, 31, -1; +$L__tmp161: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1663, %r1661, %r1662; +$L__tmp162: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1664, %r1426, 8, 31, -1; +$L__tmp163: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1665, %r1426, %r1664; +$L__tmp164: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1666, %r1665, 4, 31, -1; +$L__tmp165: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1667, %r1665, %r1666; +$L__tmp166: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1668, %r1667, 2, 31, -1; +$L__tmp167: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1669, %r1667, %r1668; +$L__tmp168: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1670, %r1669, 1, 31, -1; +$L__tmp169: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1671, %r1669, %r1670; +$L__tmp170: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1672, %r1433, 8, 31, -1; +$L__tmp171: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1673, %r1433, %r1672; +$L__tmp172: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1674, %r1673, 4, 31, -1; +$L__tmp173: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1675, %r1673, %r1674; +$L__tmp174: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1676, %r1675, 2, 31, -1; +$L__tmp175: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1677, %r1675, %r1676; +$L__tmp176: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1678, %r1677, 1, 31, -1; +$L__tmp177: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1679, %r1677, %r1678; +$L__tmp178: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1680, %r1440, 8, 31, -1; +$L__tmp179: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1681, %r1440, %r1680; +$L__tmp180: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1682, %r1681, 4, 31, -1; +$L__tmp181: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1683, %r1681, %r1682; +$L__tmp182: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1684, %r1683, 2, 31, -1; +$L__tmp183: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1685, %r1683, %r1684; +$L__tmp184: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1686, %r1685, 1, 31, -1; +$L__tmp185: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1687, %r1685, %r1686; +$L__tmp186: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1688, %r1447, 8, 31, -1; +$L__tmp187: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1689, %r1447, %r1688; +$L__tmp188: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1690, %r1689, 4, 31, -1; +$L__tmp189: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1691, %r1689, %r1690; +$L__tmp190: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1692, %r1691, 2, 31, -1; +$L__tmp191: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1693, %r1691, %r1692; +$L__tmp192: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1694, %r1693, 1, 31, -1; +$L__tmp193: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1695, %r1693, %r1694; +$L__tmp194: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1696, %r1454, 8, 31, -1; +$L__tmp195: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1697, %r1454, %r1696; +$L__tmp196: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1698, %r1697, 4, 31, -1; +$L__tmp197: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1699, %r1697, %r1698; +$L__tmp198: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1700, %r1699, 2, 31, -1; +$L__tmp199: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1701, %r1699, %r1700; +$L__tmp200: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1702, %r1701, 1, 31, -1; +$L__tmp201: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1703, %r1701, %r1702; +$L__tmp202: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1704, %r1461, 8, 31, -1; +$L__tmp203: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1705, %r1461, %r1704; +$L__tmp204: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1706, %r1705, 4, 31, -1; +$L__tmp205: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1707, %r1705, %r1706; +$L__tmp206: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1708, %r1707, 2, 31, -1; +$L__tmp207: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1709, %r1707, %r1708; +$L__tmp208: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1710, %r1709, 1, 31, -1; +$L__tmp209: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1711, %r1709, %r1710; +$L__tmp210: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1712, %r1468, 8, 31, -1; +$L__tmp211: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1713, %r1468, %r1712; +$L__tmp212: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1714, %r1713, 4, 31, -1; +$L__tmp213: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1715, %r1713, %r1714; +$L__tmp214: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1716, %r1715, 2, 31, -1; +$L__tmp215: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1717, %r1715, %r1716; +$L__tmp216: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1718, %r1717, 1, 31, -1; +$L__tmp217: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1719, %r1717, %r1718; +$L__tmp218: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1720, %r1475, 8, 31, -1; +$L__tmp219: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1721, %r1475, %r1720; +$L__tmp220: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1722, %r1721, 4, 31, -1; +$L__tmp221: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1723, %r1721, %r1722; +$L__tmp222: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1724, %r1723, 2, 31, -1; +$L__tmp223: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1725, %r1723, %r1724; +$L__tmp224: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1726, %r1725, 1, 31, -1; +$L__tmp225: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1727, %r1725, %r1726; +$L__tmp226: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1728, %r1482, 8, 31, -1; +$L__tmp227: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1729, %r1482, %r1728; +$L__tmp228: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1730, %r1729, 4, 31, -1; +$L__tmp229: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1731, %r1729, %r1730; +$L__tmp230: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1732, %r1731, 2, 31, -1; +$L__tmp231: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1733, %r1731, %r1732; +$L__tmp232: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1734, %r1733, 1, 31, -1; +$L__tmp233: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1735, %r1733, %r1734; +$L__tmp234: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1736, %r1489, 8, 31, -1; +$L__tmp235: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1737, %r1489, %r1736; +$L__tmp236: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1738, %r1737, 4, 31, -1; +$L__tmp237: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1739, %r1737, %r1738; +$L__tmp238: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1740, %r1739, 2, 31, -1; +$L__tmp239: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1741, %r1739, %r1740; +$L__tmp240: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1742, %r1741, 1, 31, -1; +$L__tmp241: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1743, %r1741, %r1742; +$L__tmp242: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1744, %r1496, 8, 31, -1; +$L__tmp243: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1745, %r1496, %r1744; +$L__tmp244: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1746, %r1745, 4, 31, -1; +$L__tmp245: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1747, %r1745, %r1746; +$L__tmp246: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1748, %r1747, 2, 31, -1; +$L__tmp247: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1749, %r1747, %r1748; +$L__tmp248: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1750, %r1749, 1, 31, -1; +$L__tmp249: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1751, %r1749, %r1750; +$L__tmp250: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1752, %r1503, 8, 31, -1; +$L__tmp251: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1753, %r1503, %r1752; +$L__tmp252: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1754, %r1753, 4, 31, -1; +$L__tmp253: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1755, %r1753, %r1754; +$L__tmp254: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1756, %r1755, 2, 31, -1; +$L__tmp255: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1757, %r1755, %r1756; +$L__tmp256: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1758, %r1757, 1, 31, -1; +$L__tmp257: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1759, %r1757, %r1758; +$L__tmp258: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r1760, %r1511, %r457; + mul.f32 %r1761, %r1511, %r458; + mul.f32 %r1762, %r1511, %r459; + mul.f32 %r1763, %r1511, %r460; + mul.f32 %r1764, %r1511, %r461; + mul.f32 %r1765, %r1511, %r462; + mul.f32 %r1766, %r1511, %r463; + mul.f32 %r1767, %r1511, %r464; + mul.f32 %r1768, %r1519, %r465; + mul.f32 %r1769, %r1519, %r466; + mul.f32 %r1770, %r1519, %r467; + mul.f32 %r1771, %r1519, %r468; + mul.f32 %r1772, %r1519, %r469; + mul.f32 %r1773, %r1519, %r470; + mul.f32 %r1774, %r1519, %r471; + mul.f32 %r1775, %r1519, %r472; + mul.f32 %r1776, %r1527, %r473; + mul.f32 %r1777, %r1527, %r474; + mul.f32 %r1778, %r1527, %r475; + mul.f32 %r1779, %r1527, %r476; + mul.f32 %r1780, %r1527, %r477; + mul.f32 %r1781, %r1527, %r478; + mul.f32 %r1782, %r1527, %r479; + mul.f32 %r1783, %r1527, %r480; + mul.f32 %r1784, %r1535, %r481; + mul.f32 %r1785, %r1535, %r482; + mul.f32 %r1786, %r1535, %r483; + mul.f32 %r1787, %r1535, %r484; + mul.f32 %r1788, %r1535, %r485; + mul.f32 %r1789, %r1535, %r486; + mul.f32 %r1790, %r1535, %r487; + mul.f32 %r1791, %r1535, %r488; + mul.f32 %r1792, %r1543, %r489; + mul.f32 %r1793, %r1543, %r490; + mul.f32 %r1794, %r1543, %r491; + mul.f32 %r1795, %r1543, %r492; + mul.f32 %r1796, %r1543, %r493; + mul.f32 %r1797, %r1543, %r494; + mul.f32 %r1798, %r1543, %r495; + mul.f32 %r1799, %r1543, %r496; + mul.f32 %r1800, %r1551, %r497; + mul.f32 %r1801, %r1551, %r498; + mul.f32 %r1802, %r1551, %r499; + mul.f32 %r1803, %r1551, %r500; + mul.f32 %r1804, %r1551, %r501; + mul.f32 %r1805, %r1551, %r502; + mul.f32 %r1806, %r1551, %r503; + mul.f32 %r1807, %r1551, %r504; + mul.f32 %r1808, %r1559, %r505; + mul.f32 %r1809, %r1559, %r506; + mul.f32 %r1810, %r1559, %r507; + mul.f32 %r1811, %r1559, %r508; + mul.f32 %r1812, %r1559, %r509; + mul.f32 %r1813, %r1559, %r510; + mul.f32 %r1814, %r1559, %r511; + mul.f32 %r1815, %r1559, %r512; + mul.f32 %r1816, %r1567, %r513; + mul.f32 %r1817, %r1567, %r514; + mul.f32 %r1818, %r1567, %r515; + mul.f32 %r1819, %r1567, %r516; + mul.f32 %r1820, %r1567, %r517; + mul.f32 %r1821, %r1567, %r518; + mul.f32 %r1822, %r1567, %r519; + mul.f32 %r1823, %r1567, %r520; + mul.f32 %r1824, %r1575, %r521; + mul.f32 %r1825, %r1575, %r522; + mul.f32 %r1826, %r1575, %r523; + mul.f32 %r1827, %r1575, %r524; + mul.f32 %r1828, %r1575, %r525; + mul.f32 %r1829, %r1575, %r526; + mul.f32 %r1830, %r1575, %r527; + mul.f32 %r1831, %r1575, %r528; + mul.f32 %r1832, %r1583, %r529; + mul.f32 %r1833, %r1583, %r530; + mul.f32 %r1834, %r1583, %r531; + mul.f32 %r1835, %r1583, %r532; + mul.f32 %r1836, %r1583, %r533; + mul.f32 %r1837, %r1583, %r534; + mul.f32 %r1838, %r1583, %r535; + mul.f32 %r1839, %r1583, %r536; + mul.f32 %r1840, %r1591, %r537; + mul.f32 %r1841, %r1591, %r538; + mul.f32 %r1842, %r1591, %r539; + mul.f32 %r1843, %r1591, %r540; + mul.f32 %r1844, %r1591, %r541; + mul.f32 %r1845, %r1591, %r542; + mul.f32 %r1846, %r1591, %r543; + mul.f32 %r1847, %r1591, %r544; + mul.f32 %r1848, %r1599, %r545; + mul.f32 %r1849, %r1599, %r546; + mul.f32 %r1850, %r1599, %r547; + mul.f32 %r1851, %r1599, %r548; + mul.f32 %r1852, %r1599, %r549; + mul.f32 %r1853, %r1599, %r550; + mul.f32 %r1854, %r1599, %r551; + mul.f32 %r1855, %r1599, %r552; + mul.f32 %r1856, %r1607, %r553; + mul.f32 %r1857, %r1607, %r554; + mul.f32 %r1858, %r1607, %r555; + mul.f32 %r1859, %r1607, %r556; + mul.f32 %r1860, %r1607, %r557; + mul.f32 %r1861, %r1607, %r558; + mul.f32 %r1862, %r1607, %r559; + mul.f32 %r1863, %r1607, %r560; + mul.f32 %r1864, %r1615, %r561; + mul.f32 %r1865, %r1615, %r562; + mul.f32 %r1866, %r1615, %r563; + mul.f32 %r1867, %r1615, %r564; + mul.f32 %r1868, %r1615, %r565; + mul.f32 %r1869, %r1615, %r566; + mul.f32 %r1870, %r1615, %r567; + mul.f32 %r1871, %r1615, %r568; + mul.f32 %r1872, %r1623, %r569; + mul.f32 %r1873, %r1623, %r570; + mul.f32 %r1874, %r1623, %r571; + mul.f32 %r1875, %r1623, %r572; + mul.f32 %r1876, %r1623, %r573; + mul.f32 %r1877, %r1623, %r574; + mul.f32 %r1878, %r1623, %r575; + mul.f32 %r1879, %r1623, %r576; + mul.f32 %r1880, %r1631, %r577; + mul.f32 %r1881, %r1631, %r578; + mul.f32 %r1882, %r1631, %r579; + mul.f32 %r1883, %r1631, %r580; + mul.f32 %r1884, %r1631, %r581; + mul.f32 %r1885, %r1631, %r582; + mul.f32 %r1886, %r1631, %r583; + mul.f32 %r1887, %r1631, %r584; + mul.f32 %r1888, %r1639, %r585; + mul.f32 %r1889, %r1639, %r586; + mul.f32 %r1890, %r1639, %r587; + mul.f32 %r1891, %r1639, %r588; + mul.f32 %r1892, %r1639, %r589; + mul.f32 %r1893, %r1639, %r590; + mul.f32 %r1894, %r1639, %r591; + mul.f32 %r1895, %r1639, %r592; + mul.f32 %r1896, %r1647, %r593; + mul.f32 %r1897, %r1647, %r594; + mul.f32 %r1898, %r1647, %r595; + mul.f32 %r1899, %r1647, %r596; + mul.f32 %r1900, %r1647, %r597; + mul.f32 %r1901, %r1647, %r598; + mul.f32 %r1902, %r1647, %r599; + mul.f32 %r1903, %r1647, %r600; + mul.f32 %r1904, %r1655, %r601; + mul.f32 %r1905, %r1655, %r602; + mul.f32 %r1906, %r1655, %r603; + mul.f32 %r1907, %r1655, %r604; + mul.f32 %r1908, %r1655, %r605; + mul.f32 %r1909, %r1655, %r606; + mul.f32 %r1910, %r1655, %r607; + mul.f32 %r1911, %r1655, %r608; + mul.f32 %r1912, %r1663, %r609; + mul.f32 %r1913, %r1663, %r610; + mul.f32 %r1914, %r1663, %r611; + mul.f32 %r1915, %r1663, %r612; + mul.f32 %r1916, %r1663, %r613; + mul.f32 %r1917, %r1663, %r614; + mul.f32 %r1918, %r1663, %r615; + mul.f32 %r1919, %r1663, %r616; + mul.f32 %r1920, %r1671, %r617; + mul.f32 %r1921, %r1671, %r618; + mul.f32 %r1922, %r1671, %r619; + mul.f32 %r1923, %r1671, %r620; + mul.f32 %r1924, %r1671, %r621; + mul.f32 %r1925, %r1671, %r622; + mul.f32 %r1926, %r1671, %r623; + mul.f32 %r1927, %r1671, %r624; + mul.f32 %r1928, %r1679, %r625; + mul.f32 %r1929, %r1679, %r626; + mul.f32 %r1930, %r1679, %r627; + mul.f32 %r1931, %r1679, %r628; + mul.f32 %r1932, %r1679, %r629; + mul.f32 %r1933, %r1679, %r630; + mul.f32 %r1934, %r1679, %r631; + mul.f32 %r1935, %r1679, %r632; + mul.f32 %r1936, %r1687, %r633; + mul.f32 %r1937, %r1687, %r634; + mul.f32 %r1938, %r1687, %r635; + mul.f32 %r1939, %r1687, %r636; + mul.f32 %r1940, %r1687, %r637; + mul.f32 %r1941, %r1687, %r638; + mul.f32 %r1942, %r1687, %r639; + mul.f32 %r1943, %r1687, %r640; + mul.f32 %r1944, %r1695, %r641; + mul.f32 %r1945, %r1695, %r642; + mul.f32 %r1946, %r1695, %r643; + mul.f32 %r1947, %r1695, %r644; + mul.f32 %r1948, %r1695, %r645; + mul.f32 %r1949, %r1695, %r646; + mul.f32 %r1950, %r1695, %r647; + mul.f32 %r1951, %r1695, %r648; + mul.f32 %r1952, %r1703, %r649; + mul.f32 %r1953, %r1703, %r650; + mul.f32 %r1954, %r1703, %r651; + mul.f32 %r1955, %r1703, %r652; + mul.f32 %r1956, %r1703, %r653; + mul.f32 %r1957, %r1703, %r654; + mul.f32 %r1958, %r1703, %r655; + mul.f32 %r1959, %r1703, %r656; + mul.f32 %r1960, %r1711, %r657; + mul.f32 %r1961, %r1711, %r658; + mul.f32 %r1962, %r1711, %r659; + mul.f32 %r1963, %r1711, %r660; + mul.f32 %r1964, %r1711, %r661; + mul.f32 %r1965, %r1711, %r662; + mul.f32 %r1966, %r1711, %r663; + mul.f32 %r1967, %r1711, %r664; + mul.f32 %r1968, %r1719, %r665; + mul.f32 %r1969, %r1719, %r666; + mul.f32 %r1970, %r1719, %r667; + mul.f32 %r1971, %r1719, %r668; + mul.f32 %r1972, %r1719, %r669; + mul.f32 %r1973, %r1719, %r670; + mul.f32 %r1974, %r1719, %r671; + mul.f32 %r1975, %r1719, %r672; + mul.f32 %r1976, %r1727, %r673; + mul.f32 %r1977, %r1727, %r674; + mul.f32 %r1978, %r1727, %r675; + mul.f32 %r1979, %r1727, %r676; + mul.f32 %r1980, %r1727, %r677; + mul.f32 %r1981, %r1727, %r678; + mul.f32 %r1982, %r1727, %r679; + mul.f32 %r1983, %r1727, %r680; + mul.f32 %r1984, %r1735, %r681; + mul.f32 %r1985, %r1735, %r682; + mul.f32 %r1986, %r1735, %r683; + mul.f32 %r1987, %r1735, %r684; + mul.f32 %r1988, %r1735, %r685; + mul.f32 %r1989, %r1735, %r686; + mul.f32 %r1990, %r1735, %r687; + mul.f32 %r1991, %r1735, %r688; + mul.f32 %r1992, %r1743, %r689; + mul.f32 %r1993, %r1743, %r690; + mul.f32 %r1994, %r1743, %r691; + mul.f32 %r1995, %r1743, %r692; + mul.f32 %r1996, %r1743, %r693; + mul.f32 %r1997, %r1743, %r694; + mul.f32 %r1998, %r1743, %r695; + mul.f32 %r1999, %r1743, %r696; + mul.f32 %r2000, %r1751, %r697; + mul.f32 %r2001, %r1751, %r698; + mul.f32 %r2002, %r1751, %r699; + mul.f32 %r2003, %r1751, %r700; + mul.f32 %r2004, %r1751, %r701; + mul.f32 %r2005, %r1751, %r702; + mul.f32 %r2006, %r1751, %r703; + mul.f32 %r2007, %r1751, %r704; + mul.f32 %r2008, %r1759, %r705; + mul.f32 %r2009, %r1759, %r706; + mul.f32 %r2010, %r1759, %r707; + mul.f32 %r2011, %r1759, %r708; + mul.f32 %r2012, %r1759, %r709; + mul.f32 %r2013, %r1759, %r710; + mul.f32 %r2014, %r1759, %r711; + mul.f32 %r2015, %r1759, %r712; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r2016, %r976, %r1760; + mul.f32 %r2017, %r976, %r1761; + mul.f32 %r2018, %r976, %r1762; + mul.f32 %r2019, %r976, %r1763; + mul.f32 %r2020, %r976, %r1764; + mul.f32 %r2021, %r976, %r1765; + mul.f32 %r2022, %r976, %r1766; + mul.f32 %r2023, %r976, %r1767; + mul.f32 %r2024, %r977, %r1768; + mul.f32 %r2025, %r977, %r1769; + mul.f32 %r2026, %r977, %r1770; + mul.f32 %r2027, %r977, %r1771; + mul.f32 %r2028, %r977, %r1772; + mul.f32 %r2029, %r977, %r1773; + mul.f32 %r2030, %r977, %r1774; + mul.f32 %r2031, %r977, %r1775; + mul.f32 %r2032, %r978, %r1776; + mul.f32 %r2033, %r978, %r1777; + mul.f32 %r2034, %r978, %r1778; + mul.f32 %r2035, %r978, %r1779; + mul.f32 %r2036, %r978, %r1780; + mul.f32 %r2037, %r978, %r1781; + mul.f32 %r2038, %r978, %r1782; + mul.f32 %r2039, %r978, %r1783; + mul.f32 %r2040, %r979, %r1784; + mul.f32 %r2041, %r979, %r1785; + mul.f32 %r2042, %r979, %r1786; + mul.f32 %r2043, %r979, %r1787; + mul.f32 %r2044, %r979, %r1788; + mul.f32 %r2045, %r979, %r1789; + mul.f32 %r2046, %r979, %r1790; + mul.f32 %r2047, %r979, %r1791; + mul.f32 %r2048, %r980, %r1792; + mul.f32 %r2049, %r980, %r1793; + mul.f32 %r2050, %r980, %r1794; + mul.f32 %r2051, %r980, %r1795; + mul.f32 %r2052, %r980, %r1796; + mul.f32 %r2053, %r980, %r1797; + mul.f32 %r2054, %r980, %r1798; + mul.f32 %r2055, %r980, %r1799; + mul.f32 %r2056, %r981, %r1800; + mul.f32 %r2057, %r981, %r1801; + mul.f32 %r2058, %r981, %r1802; + mul.f32 %r2059, %r981, %r1803; + mul.f32 %r2060, %r981, %r1804; + mul.f32 %r2061, %r981, %r1805; + mul.f32 %r2062, %r981, %r1806; + mul.f32 %r2063, %r981, %r1807; + mul.f32 %r2064, %r982, %r1808; + mul.f32 %r2065, %r982, %r1809; + mul.f32 %r2066, %r982, %r1810; + mul.f32 %r2067, %r982, %r1811; + mul.f32 %r2068, %r982, %r1812; + mul.f32 %r2069, %r982, %r1813; + mul.f32 %r2070, %r982, %r1814; + mul.f32 %r2071, %r982, %r1815; + mul.f32 %r2072, %r983, %r1816; + mul.f32 %r2073, %r983, %r1817; + mul.f32 %r2074, %r983, %r1818; + mul.f32 %r2075, %r983, %r1819; + mul.f32 %r2076, %r983, %r1820; + mul.f32 %r2077, %r983, %r1821; + mul.f32 %r2078, %r983, %r1822; + mul.f32 %r2079, %r983, %r1823; + mul.f32 %r2080, %r984, %r1824; + mul.f32 %r2081, %r984, %r1825; + mul.f32 %r2082, %r984, %r1826; + mul.f32 %r2083, %r984, %r1827; + mul.f32 %r2084, %r984, %r1828; + mul.f32 %r2085, %r984, %r1829; + mul.f32 %r2086, %r984, %r1830; + mul.f32 %r2087, %r984, %r1831; + mul.f32 %r2088, %r985, %r1832; + mul.f32 %r2089, %r985, %r1833; + mul.f32 %r2090, %r985, %r1834; + mul.f32 %r2091, %r985, %r1835; + mul.f32 %r2092, %r985, %r1836; + mul.f32 %r2093, %r985, %r1837; + mul.f32 %r2094, %r985, %r1838; + mul.f32 %r2095, %r985, %r1839; + mul.f32 %r2096, %r986, %r1840; + mul.f32 %r2097, %r986, %r1841; + mul.f32 %r2098, %r986, %r1842; + mul.f32 %r2099, %r986, %r1843; + mul.f32 %r2100, %r986, %r1844; + mul.f32 %r2101, %r986, %r1845; + mul.f32 %r2102, %r986, %r1846; + mul.f32 %r2103, %r986, %r1847; + mul.f32 %r2104, %r987, %r1848; + mul.f32 %r2105, %r987, %r1849; + mul.f32 %r2106, %r987, %r1850; + mul.f32 %r2107, %r987, %r1851; + mul.f32 %r2108, %r987, %r1852; + mul.f32 %r2109, %r987, %r1853; + mul.f32 %r2110, %r987, %r1854; + mul.f32 %r2111, %r987, %r1855; + mul.f32 %r2112, %r988, %r1856; + mul.f32 %r2113, %r988, %r1857; + mul.f32 %r2114, %r988, %r1858; + mul.f32 %r2115, %r988, %r1859; + mul.f32 %r2116, %r988, %r1860; + mul.f32 %r2117, %r988, %r1861; + mul.f32 %r2118, %r988, %r1862; + mul.f32 %r2119, %r988, %r1863; + mul.f32 %r2120, %r989, %r1864; + mul.f32 %r2121, %r989, %r1865; + mul.f32 %r2122, %r989, %r1866; + mul.f32 %r2123, %r989, %r1867; + mul.f32 %r2124, %r989, %r1868; + mul.f32 %r2125, %r989, %r1869; + mul.f32 %r2126, %r989, %r1870; + mul.f32 %r2127, %r989, %r1871; + mul.f32 %r2128, %r990, %r1872; + mul.f32 %r2129, %r990, %r1873; + mul.f32 %r2130, %r990, %r1874; + mul.f32 %r2131, %r990, %r1875; + mul.f32 %r2132, %r990, %r1876; + mul.f32 %r2133, %r990, %r1877; + mul.f32 %r2134, %r990, %r1878; + mul.f32 %r2135, %r990, %r1879; + mul.f32 %r2136, %r991, %r1880; + mul.f32 %r2137, %r991, %r1881; + mul.f32 %r2138, %r991, %r1882; + mul.f32 %r2139, %r991, %r1883; + mul.f32 %r2140, %r991, %r1884; + mul.f32 %r2141, %r991, %r1885; + mul.f32 %r2142, %r991, %r1886; + mul.f32 %r2143, %r991, %r1887; + mul.f32 %r2144, %r992, %r1888; + mul.f32 %r2145, %r992, %r1889; + mul.f32 %r2146, %r992, %r1890; + mul.f32 %r2147, %r992, %r1891; + mul.f32 %r2148, %r992, %r1892; + mul.f32 %r2149, %r992, %r1893; + mul.f32 %r2150, %r992, %r1894; + mul.f32 %r2151, %r992, %r1895; + mul.f32 %r2152, %r993, %r1896; + mul.f32 %r2153, %r993, %r1897; + mul.f32 %r2154, %r993, %r1898; + mul.f32 %r2155, %r993, %r1899; + mul.f32 %r2156, %r993, %r1900; + mul.f32 %r2157, %r993, %r1901; + mul.f32 %r2158, %r993, %r1902; + mul.f32 %r2159, %r993, %r1903; + mul.f32 %r2160, %r994, %r1904; + mul.f32 %r2161, %r994, %r1905; + mul.f32 %r2162, %r994, %r1906; + mul.f32 %r2163, %r994, %r1907; + mul.f32 %r2164, %r994, %r1908; + mul.f32 %r2165, %r994, %r1909; + mul.f32 %r2166, %r994, %r1910; + mul.f32 %r2167, %r994, %r1911; + mul.f32 %r2168, %r995, %r1912; + mul.f32 %r2169, %r995, %r1913; + mul.f32 %r2170, %r995, %r1914; + mul.f32 %r2171, %r995, %r1915; + mul.f32 %r2172, %r995, %r1916; + mul.f32 %r2173, %r995, %r1917; + mul.f32 %r2174, %r995, %r1918; + mul.f32 %r2175, %r995, %r1919; + mul.f32 %r2176, %r996, %r1920; + mul.f32 %r2177, %r996, %r1921; + mul.f32 %r2178, %r996, %r1922; + mul.f32 %r2179, %r996, %r1923; + mul.f32 %r2180, %r996, %r1924; + mul.f32 %r2181, %r996, %r1925; + mul.f32 %r2182, %r996, %r1926; + mul.f32 %r2183, %r996, %r1927; + mul.f32 %r2184, %r997, %r1928; + mul.f32 %r2185, %r997, %r1929; + mul.f32 %r2186, %r997, %r1930; + mul.f32 %r2187, %r997, %r1931; + mul.f32 %r2188, %r997, %r1932; + mul.f32 %r2189, %r997, %r1933; + mul.f32 %r2190, %r997, %r1934; + mul.f32 %r2191, %r997, %r1935; + mul.f32 %r2192, %r998, %r1936; + mul.f32 %r2193, %r998, %r1937; + mul.f32 %r2194, %r998, %r1938; + mul.f32 %r2195, %r998, %r1939; + mul.f32 %r2196, %r998, %r1940; + mul.f32 %r2197, %r998, %r1941; + mul.f32 %r2198, %r998, %r1942; + mul.f32 %r2199, %r998, %r1943; + mul.f32 %r2200, %r999, %r1944; + mul.f32 %r2201, %r999, %r1945; + mul.f32 %r2202, %r999, %r1946; + mul.f32 %r2203, %r999, %r1947; + mul.f32 %r2204, %r999, %r1948; + mul.f32 %r2205, %r999, %r1949; + mul.f32 %r2206, %r999, %r1950; + mul.f32 %r2207, %r999, %r1951; + mul.f32 %r2208, %r1000, %r1952; + mul.f32 %r2209, %r1000, %r1953; + mul.f32 %r2210, %r1000, %r1954; + mul.f32 %r2211, %r1000, %r1955; + mul.f32 %r2212, %r1000, %r1956; + mul.f32 %r2213, %r1000, %r1957; + mul.f32 %r2214, %r1000, %r1958; + mul.f32 %r2215, %r1000, %r1959; + mul.f32 %r2216, %r1001, %r1960; + mul.f32 %r2217, %r1001, %r1961; + mul.f32 %r2218, %r1001, %r1962; + mul.f32 %r2219, %r1001, %r1963; + mul.f32 %r2220, %r1001, %r1964; + mul.f32 %r2221, %r1001, %r1965; + mul.f32 %r2222, %r1001, %r1966; + mul.f32 %r2223, %r1001, %r1967; + mul.f32 %r2224, %r1002, %r1968; + mul.f32 %r2225, %r1002, %r1969; + mul.f32 %r2226, %r1002, %r1970; + mul.f32 %r2227, %r1002, %r1971; + mul.f32 %r2228, %r1002, %r1972; + mul.f32 %r2229, %r1002, %r1973; + mul.f32 %r2230, %r1002, %r1974; + mul.f32 %r2231, %r1002, %r1975; + mul.f32 %r2232, %r1003, %r1976; + mul.f32 %r2233, %r1003, %r1977; + mul.f32 %r2234, %r1003, %r1978; + mul.f32 %r2235, %r1003, %r1979; + mul.f32 %r2236, %r1003, %r1980; + mul.f32 %r2237, %r1003, %r1981; + mul.f32 %r2238, %r1003, %r1982; + mul.f32 %r2239, %r1003, %r1983; + mul.f32 %r2240, %r1004, %r1984; + mul.f32 %r2241, %r1004, %r1985; + mul.f32 %r2242, %r1004, %r1986; + mul.f32 %r2243, %r1004, %r1987; + mul.f32 %r2244, %r1004, %r1988; + mul.f32 %r2245, %r1004, %r1989; + mul.f32 %r2246, %r1004, %r1990; + mul.f32 %r2247, %r1004, %r1991; + mul.f32 %r2248, %r1005, %r1992; + mul.f32 %r2249, %r1005, %r1993; + mul.f32 %r2250, %r1005, %r1994; + mul.f32 %r2251, %r1005, %r1995; + mul.f32 %r2252, %r1005, %r1996; + mul.f32 %r2253, %r1005, %r1997; + mul.f32 %r2254, %r1005, %r1998; + mul.f32 %r2255, %r1005, %r1999; + mul.f32 %r2256, %r1006, %r2000; + mul.f32 %r2257, %r1006, %r2001; + mul.f32 %r2258, %r1006, %r2002; + mul.f32 %r2259, %r1006, %r2003; + mul.f32 %r2260, %r1006, %r2004; + mul.f32 %r2261, %r1006, %r2005; + mul.f32 %r2262, %r1006, %r2006; + mul.f32 %r2263, %r1006, %r2007; + mul.f32 %r2264, %r1007, %r2008; + mul.f32 %r2265, %r1007, %r2009; + mul.f32 %r2266, %r1007, %r2010; + mul.f32 %r2267, %r1007, %r2011; + mul.f32 %r2268, %r1007, %r2012; + mul.f32 %r2269, %r1007, %r2013; + mul.f32 %r2270, %r1007, %r2014; + mul.f32 %r2271, %r1007, %r2015; + .loc 1 107 36 // l2norm.py:107:36 + neg.f32 %r2272, %r2016; + fma.rn.f32 %r2273, %r976, %r715, %r2272; + neg.f32 %r2274, %r2017; + fma.rn.f32 %r2275, %r976, %r716, %r2274; + neg.f32 %r2276, %r2018; + fma.rn.f32 %r2277, %r976, %r717, %r2276; + neg.f32 %r2278, %r2019; + fma.rn.f32 %r2279, %r976, %r718, %r2278; + neg.f32 %r2280, %r2020; + fma.rn.f32 %r2281, %r976, %r719, %r2280; + neg.f32 %r2282, %r2021; + fma.rn.f32 %r2283, %r976, %r720, %r2282; + neg.f32 %r2284, %r2022; + fma.rn.f32 %r2285, %r976, %r721, %r2284; + neg.f32 %r2286, %r2023; + fma.rn.f32 %r2287, %r976, %r722, %r2286; + neg.f32 %r2288, %r2024; + fma.rn.f32 %r2289, %r977, %r723, %r2288; + neg.f32 %r2290, %r2025; + fma.rn.f32 %r2291, %r977, %r724, %r2290; + neg.f32 %r2292, %r2026; + fma.rn.f32 %r2293, %r977, %r725, %r2292; + neg.f32 %r2294, %r2027; + fma.rn.f32 %r2295, %r977, %r726, %r2294; + neg.f32 %r2296, %r2028; + fma.rn.f32 %r2297, %r977, %r727, %r2296; + neg.f32 %r2298, %r2029; + fma.rn.f32 %r2299, %r977, %r728, %r2298; + neg.f32 %r2300, %r2030; + fma.rn.f32 %r2301, %r977, %r729, %r2300; + neg.f32 %r2302, %r2031; + fma.rn.f32 %r2303, %r977, %r730, %r2302; + mov.b64 %rd304, {%r1008, %r1009}; + mov.b64 %rd305, {%r2032, %r2033}; + sub.f32x2 %rd306, %rd304, %rd305; + mov.b64 %rd307, {%r1010, %r1011}; + mov.b64 %rd308, {%r2034, %r2035}; + sub.f32x2 %rd309, %rd307, %rd308; + mov.b64 %rd310, {%r1012, %r1013}; + mov.b64 %rd311, {%r2036, %r2037}; + sub.f32x2 %rd312, %rd310, %rd311; + mov.b64 %rd313, {%r1014, %r1015}; + mov.b64 %rd314, {%r2038, %r2039}; + sub.f32x2 %rd315, %rd313, %rd314; + mov.b64 %rd316, {%r1016, %r1017}; + mov.b64 %rd317, {%r2040, %r2041}; + sub.f32x2 %rd318, %rd316, %rd317; + mov.b64 %rd319, {%r1018, %r1019}; + mov.b64 %rd320, {%r2042, %r2043}; + sub.f32x2 %rd321, %rd319, %rd320; + mov.b64 %rd322, {%r1020, %r1021}; + mov.b64 %rd323, {%r2044, %r2045}; + sub.f32x2 %rd324, %rd322, %rd323; + mov.b64 %rd325, {%r1022, %r1023}; + mov.b64 %rd326, {%r2046, %r2047}; + sub.f32x2 %rd327, %rd325, %rd326; + mov.b64 %rd328, {%r1024, %r1025}; + mov.b64 %rd329, {%r2048, %r2049}; + sub.f32x2 %rd330, %rd328, %rd329; + mov.b64 %rd331, {%r1026, %r1027}; + mov.b64 %rd332, {%r2050, %r2051}; + sub.f32x2 %rd333, %rd331, %rd332; + mov.b64 %rd334, {%r1028, %r1029}; + mov.b64 %rd335, {%r2052, %r2053}; + sub.f32x2 %rd336, %rd334, %rd335; + mov.b64 %rd337, {%r1030, %r1031}; + mov.b64 %rd338, {%r2054, %r2055}; + sub.f32x2 %rd339, %rd337, %rd338; + mov.b64 %rd340, {%r1032, %r1033}; + mov.b64 %rd341, {%r2056, %r2057}; + sub.f32x2 %rd342, %rd340, %rd341; + mov.b64 %rd343, {%r1034, %r1035}; + mov.b64 %rd344, {%r2058, %r2059}; + sub.f32x2 %rd345, %rd343, %rd344; + mov.b64 %rd346, {%r1036, %r1037}; + mov.b64 %rd347, {%r2060, %r2061}; + sub.f32x2 %rd348, %rd346, %rd347; + mov.b64 %rd349, {%r1038, %r1039}; + mov.b64 %rd350, {%r2062, %r2063}; + sub.f32x2 %rd351, %rd349, %rd350; + mov.b64 %rd352, {%r1040, %r1041}; + mov.b64 %rd353, {%r2064, %r2065}; + sub.f32x2 %rd354, %rd352, %rd353; + mov.b64 %rd355, {%r1042, %r1043}; + mov.b64 %rd356, {%r2066, %r2067}; + sub.f32x2 %rd357, %rd355, %rd356; + mov.b64 %rd358, {%r1044, %r1045}; + mov.b64 %rd359, {%r2068, %r2069}; + sub.f32x2 %rd360, %rd358, %rd359; + mov.b64 %rd361, {%r1046, %r1047}; + mov.b64 %rd362, {%r2070, %r2071}; + sub.f32x2 %rd363, %rd361, %rd362; + mov.b64 %rd364, {%r1048, %r1049}; + mov.b64 %rd365, {%r2072, %r2073}; + sub.f32x2 %rd366, %rd364, %rd365; + mov.b64 %rd367, {%r1050, %r1051}; + mov.b64 %rd368, {%r2074, %r2075}; + sub.f32x2 %rd369, %rd367, %rd368; + mov.b64 %rd370, {%r1052, %r1053}; + mov.b64 %rd371, {%r2076, %r2077}; + sub.f32x2 %rd372, %rd370, %rd371; + mov.b64 %rd373, {%r1054, %r1055}; + mov.b64 %rd374, {%r2078, %r2079}; + sub.f32x2 %rd375, %rd373, %rd374; + mov.b64 %rd376, {%r1056, %r1057}; + mov.b64 %rd377, {%r2080, %r2081}; + sub.f32x2 %rd378, %rd376, %rd377; + mov.b64 %rd379, {%r1058, %r1059}; + mov.b64 %rd380, {%r2082, %r2083}; + sub.f32x2 %rd381, %rd379, %rd380; + mov.b64 %rd382, {%r1060, %r1061}; + mov.b64 %rd383, {%r2084, %r2085}; + sub.f32x2 %rd384, %rd382, %rd383; + mov.b64 %rd385, {%r1062, %r1063}; + mov.b64 %rd386, {%r2086, %r2087}; + sub.f32x2 %rd387, %rd385, %rd386; + mov.b64 %rd388, {%r1064, %r1065}; + mov.b64 %rd389, {%r2088, %r2089}; + sub.f32x2 %rd390, %rd388, %rd389; + mov.b64 %rd391, {%r1066, %r1067}; + mov.b64 %rd392, {%r2090, %r2091}; + sub.f32x2 %rd393, %rd391, %rd392; + mov.b64 %rd394, {%r1068, %r1069}; + mov.b64 %rd395, {%r2092, %r2093}; + sub.f32x2 %rd396, %rd394, %rd395; + mov.b64 %rd397, {%r1070, %r1071}; + mov.b64 %rd398, {%r2094, %r2095}; + sub.f32x2 %rd399, %rd397, %rd398; + mov.b64 %rd400, {%r1072, %r1073}; + mov.b64 %rd401, {%r2096, %r2097}; + sub.f32x2 %rd402, %rd400, %rd401; + mov.b64 %rd403, {%r1074, %r1075}; + mov.b64 %rd404, {%r2098, %r2099}; + sub.f32x2 %rd405, %rd403, %rd404; + mov.b64 %rd406, {%r1076, %r1077}; + mov.b64 %rd407, {%r2100, %r2101}; + sub.f32x2 %rd408, %rd406, %rd407; + mov.b64 %rd409, {%r1078, %r1079}; + mov.b64 %rd410, {%r2102, %r2103}; + sub.f32x2 %rd411, %rd409, %rd410; + mov.b64 %rd412, {%r1080, %r1081}; + mov.b64 %rd413, {%r2104, %r2105}; + sub.f32x2 %rd414, %rd412, %rd413; + mov.b64 %rd415, {%r1082, %r1083}; + mov.b64 %rd416, {%r2106, %r2107}; + sub.f32x2 %rd417, %rd415, %rd416; + mov.b64 %rd418, {%r1084, %r1085}; + mov.b64 %rd419, {%r2108, %r2109}; + sub.f32x2 %rd420, %rd418, %rd419; + mov.b64 %rd421, {%r1086, %r1087}; + mov.b64 %rd422, {%r2110, %r2111}; + sub.f32x2 %rd423, %rd421, %rd422; + mov.b64 %rd424, {%r1088, %r1089}; + mov.b64 %rd425, {%r2112, %r2113}; + sub.f32x2 %rd426, %rd424, %rd425; + mov.b64 %rd427, {%r1090, %r1091}; + mov.b64 %rd428, {%r2114, %r2115}; + sub.f32x2 %rd429, %rd427, %rd428; + mov.b64 %rd430, {%r1092, %r1093}; + mov.b64 %rd431, {%r2116, %r2117}; + sub.f32x2 %rd432, %rd430, %rd431; + mov.b64 %rd433, {%r1094, %r1095}; + mov.b64 %rd434, {%r2118, %r2119}; + sub.f32x2 %rd435, %rd433, %rd434; + mov.b64 %rd436, {%r1096, %r1097}; + mov.b64 %rd437, {%r2120, %r2121}; + sub.f32x2 %rd438, %rd436, %rd437; + mov.b64 %rd439, {%r1098, %r1099}; + mov.b64 %rd440, {%r2122, %r2123}; + sub.f32x2 %rd441, %rd439, %rd440; + mov.b64 %rd442, {%r1100, %r1101}; + mov.b64 %rd443, {%r2124, %r2125}; + sub.f32x2 %rd444, %rd442, %rd443; + mov.b64 %rd445, {%r1102, %r1103}; + mov.b64 %rd446, {%r2126, %r2127}; + sub.f32x2 %rd447, %rd445, %rd446; + mov.b64 %rd448, {%r1104, %r1105}; + mov.b64 %rd449, {%r2128, %r2129}; + sub.f32x2 %rd450, %rd448, %rd449; + mov.b64 %rd451, {%r1106, %r1107}; + mov.b64 %rd452, {%r2130, %r2131}; + sub.f32x2 %rd453, %rd451, %rd452; + mov.b64 %rd454, {%r1108, %r1109}; + mov.b64 %rd455, {%r2132, %r2133}; + sub.f32x2 %rd456, %rd454, %rd455; + mov.b64 %rd457, {%r1110, %r1111}; + mov.b64 %rd458, {%r2134, %r2135}; + sub.f32x2 %rd459, %rd457, %rd458; + mov.b64 %rd460, {%r1112, %r1113}; + mov.b64 %rd461, {%r2136, %r2137}; + sub.f32x2 %rd462, %rd460, %rd461; + mov.b64 %rd463, {%r1114, %r1115}; + mov.b64 %rd464, {%r2138, %r2139}; + sub.f32x2 %rd465, %rd463, %rd464; + mov.b64 %rd466, {%r1116, %r1117}; + mov.b64 %rd467, {%r2140, %r2141}; + sub.f32x2 %rd468, %rd466, %rd467; + mov.b64 %rd469, {%r1118, %r1119}; + mov.b64 %rd470, {%r2142, %r2143}; + sub.f32x2 %rd471, %rd469, %rd470; + mov.b64 %rd472, {%r1120, %r1121}; + mov.b64 %rd473, {%r2144, %r2145}; + sub.f32x2 %rd474, %rd472, %rd473; + mov.b64 %rd475, {%r1122, %r1123}; + mov.b64 %rd476, {%r2146, %r2147}; + sub.f32x2 %rd477, %rd475, %rd476; + mov.b64 %rd478, {%r1124, %r1125}; + mov.b64 %rd479, {%r2148, %r2149}; + sub.f32x2 %rd480, %rd478, %rd479; + mov.b64 %rd481, {%r1126, %r1127}; + mov.b64 %rd482, {%r2150, %r2151}; + sub.f32x2 %rd483, %rd481, %rd482; + mov.b64 %rd484, {%r1128, %r1129}; + mov.b64 %rd485, {%r2152, %r2153}; + sub.f32x2 %rd486, %rd484, %rd485; + mov.b64 %rd487, {%r1130, %r1131}; + mov.b64 %rd488, {%r2154, %r2155}; + sub.f32x2 %rd489, %rd487, %rd488; + mov.b64 %rd490, {%r1132, %r1133}; + mov.b64 %rd491, {%r2156, %r2157}; + sub.f32x2 %rd492, %rd490, %rd491; + mov.b64 %rd493, {%r1134, %r1135}; + mov.b64 %rd494, {%r2158, %r2159}; + sub.f32x2 %rd495, %rd493, %rd494; + mov.b64 %rd496, {%r1136, %r1137}; + mov.b64 %rd497, {%r2160, %r2161}; + sub.f32x2 %rd498, %rd496, %rd497; + mov.b64 %rd499, {%r1138, %r1139}; + mov.b64 %rd500, {%r2162, %r2163}; + sub.f32x2 %rd501, %rd499, %rd500; + mov.b64 %rd502, {%r1140, %r1141}; + mov.b64 %rd503, {%r2164, %r2165}; + sub.f32x2 %rd504, %rd502, %rd503; + mov.b64 %rd505, {%r1142, %r1143}; + mov.b64 %rd506, {%r2166, %r2167}; + sub.f32x2 %rd507, %rd505, %rd506; + mov.b64 %rd508, {%r1144, %r1145}; + mov.b64 %rd509, {%r2168, %r2169}; + sub.f32x2 %rd510, %rd508, %rd509; + mov.b64 %rd511, {%r1146, %r1147}; + mov.b64 %rd512, {%r2170, %r2171}; + sub.f32x2 %rd513, %rd511, %rd512; + mov.b64 %rd514, {%r1148, %r1149}; + mov.b64 %rd515, {%r2172, %r2173}; + sub.f32x2 %rd516, %rd514, %rd515; + mov.b64 %rd517, {%r1150, %r1151}; + mov.b64 %rd518, {%r2174, %r2175}; + sub.f32x2 %rd519, %rd517, %rd518; + mov.b64 %rd520, {%r1152, %r1153}; + mov.b64 %rd521, {%r2176, %r2177}; + sub.f32x2 %rd522, %rd520, %rd521; + mov.b64 %rd523, {%r1154, %r1155}; + mov.b64 %rd524, {%r2178, %r2179}; + sub.f32x2 %rd525, %rd523, %rd524; + mov.b64 %rd526, {%r1156, %r1157}; + mov.b64 %rd527, {%r2180, %r2181}; + sub.f32x2 %rd528, %rd526, %rd527; + mov.b64 %rd529, {%r1158, %r1159}; + mov.b64 %rd530, {%r2182, %r2183}; + sub.f32x2 %rd531, %rd529, %rd530; + mov.b64 %rd532, {%r1160, %r1161}; + mov.b64 %rd533, {%r2184, %r2185}; + sub.f32x2 %rd534, %rd532, %rd533; + mov.b64 %rd535, {%r1162, %r1163}; + mov.b64 %rd536, {%r2186, %r2187}; + sub.f32x2 %rd537, %rd535, %rd536; + mov.b64 %rd538, {%r1164, %r1165}; + mov.b64 %rd539, {%r2188, %r2189}; + sub.f32x2 %rd540, %rd538, %rd539; + mov.b64 %rd541, {%r1166, %r1167}; + mov.b64 %rd542, {%r2190, %r2191}; + sub.f32x2 %rd543, %rd541, %rd542; + mov.b64 %rd544, {%r1168, %r1169}; + mov.b64 %rd545, {%r2192, %r2193}; + sub.f32x2 %rd546, %rd544, %rd545; + mov.b64 %rd547, {%r1170, %r1171}; + mov.b64 %rd548, {%r2194, %r2195}; + sub.f32x2 %rd549, %rd547, %rd548; + mov.b64 %rd550, {%r1172, %r1173}; + mov.b64 %rd551, {%r2196, %r2197}; + sub.f32x2 %rd552, %rd550, %rd551; + mov.b64 %rd553, {%r1174, %r1175}; + mov.b64 %rd554, {%r2198, %r2199}; + sub.f32x2 %rd555, %rd553, %rd554; + mov.b64 %rd556, {%r1176, %r1177}; + mov.b64 %rd557, {%r2200, %r2201}; + sub.f32x2 %rd558, %rd556, %rd557; + mov.b64 %rd559, {%r1178, %r1179}; + mov.b64 %rd560, {%r2202, %r2203}; + sub.f32x2 %rd561, %rd559, %rd560; + mov.b64 %rd562, {%r1180, %r1181}; + mov.b64 %rd563, {%r2204, %r2205}; + sub.f32x2 %rd564, %rd562, %rd563; + mov.b64 %rd565, {%r1182, %r1183}; + mov.b64 %rd566, {%r2206, %r2207}; + sub.f32x2 %rd567, %rd565, %rd566; + mov.b64 %rd568, {%r1184, %r1185}; + mov.b64 %rd569, {%r2208, %r2209}; + sub.f32x2 %rd570, %rd568, %rd569; + mov.b64 %rd571, {%r1186, %r1187}; + mov.b64 %rd572, {%r2210, %r2211}; + sub.f32x2 %rd573, %rd571, %rd572; + mov.b64 %rd574, {%r1188, %r1189}; + mov.b64 %rd575, {%r2212, %r2213}; + sub.f32x2 %rd576, %rd574, %rd575; + mov.b64 %rd577, {%r1190, %r1191}; + mov.b64 %rd578, {%r2214, %r2215}; + sub.f32x2 %rd579, %rd577, %rd578; + mov.b64 %rd580, {%r1192, %r1193}; + mov.b64 %rd581, {%r2216, %r2217}; + sub.f32x2 %rd582, %rd580, %rd581; + mov.b64 %rd583, {%r1194, %r1195}; + mov.b64 %rd584, {%r2218, %r2219}; + sub.f32x2 %rd585, %rd583, %rd584; + mov.b64 %rd586, {%r1196, %r1197}; + mov.b64 %rd587, {%r2220, %r2221}; + sub.f32x2 %rd588, %rd586, %rd587; + mov.b64 %rd589, {%r1198, %r1199}; + mov.b64 %rd590, {%r2222, %r2223}; + sub.f32x2 %rd591, %rd589, %rd590; + mov.b64 %rd592, {%r1200, %r1201}; + mov.b64 %rd593, {%r2224, %r2225}; + sub.f32x2 %rd594, %rd592, %rd593; + mov.b64 %rd595, {%r1202, %r1203}; + mov.b64 %rd596, {%r2226, %r2227}; + sub.f32x2 %rd597, %rd595, %rd596; + mov.b64 %rd598, {%r1204, %r1205}; + mov.b64 %rd599, {%r2228, %r2229}; + sub.f32x2 %rd600, %rd598, %rd599; + mov.b64 %rd601, {%r1206, %r1207}; + mov.b64 %rd602, {%r2230, %r2231}; + sub.f32x2 %rd603, %rd601, %rd602; + mov.b64 %rd604, {%r1208, %r1209}; + mov.b64 %rd605, {%r2232, %r2233}; + sub.f32x2 %rd606, %rd604, %rd605; + mov.b64 %rd607, {%r1210, %r1211}; + mov.b64 %rd608, {%r2234, %r2235}; + sub.f32x2 %rd609, %rd607, %rd608; + mov.b64 %rd610, {%r1212, %r1213}; + mov.b64 %rd611, {%r2236, %r2237}; + sub.f32x2 %rd612, %rd610, %rd611; + mov.b64 %rd613, {%r1214, %r1215}; + mov.b64 %rd614, {%r2238, %r2239}; + sub.f32x2 %rd615, %rd613, %rd614; + mov.b64 %rd616, {%r1216, %r1217}; + mov.b64 %rd617, {%r2240, %r2241}; + sub.f32x2 %rd618, %rd616, %rd617; + mov.b64 %rd619, {%r1218, %r1219}; + mov.b64 %rd620, {%r2242, %r2243}; + sub.f32x2 %rd621, %rd619, %rd620; + mov.b64 %rd622, {%r1220, %r1221}; + mov.b64 %rd623, {%r2244, %r2245}; + sub.f32x2 %rd624, %rd622, %rd623; + mov.b64 %rd625, {%r1222, %r1223}; + mov.b64 %rd626, {%r2246, %r2247}; + sub.f32x2 %rd627, %rd625, %rd626; + mov.b64 %rd628, {%r1224, %r1225}; + mov.b64 %rd629, {%r2248, %r2249}; + sub.f32x2 %rd630, %rd628, %rd629; + mov.b64 %rd631, {%r1226, %r1227}; + mov.b64 %rd632, {%r2250, %r2251}; + sub.f32x2 %rd633, %rd631, %rd632; + mov.b64 %rd634, {%r1228, %r1229}; + mov.b64 %rd635, {%r2252, %r2253}; + sub.f32x2 %rd636, %rd634, %rd635; + mov.b64 %rd637, {%r1230, %r1231}; + mov.b64 %rd638, {%r2254, %r2255}; + sub.f32x2 %rd639, %rd637, %rd638; + mov.b64 %rd640, {%r1232, %r1233}; + mov.b64 %rd641, {%r2256, %r2257}; + sub.f32x2 %rd642, %rd640, %rd641; + mov.b64 %rd643, {%r1234, %r1235}; + mov.b64 %rd644, {%r2258, %r2259}; + sub.f32x2 %rd645, %rd643, %rd644; + mov.b64 %rd646, {%r1236, %r1237}; + mov.b64 %rd647, {%r2260, %r2261}; + sub.f32x2 %rd648, %rd646, %rd647; + mov.b64 %rd649, {%r1238, %r1239}; + mov.b64 %rd650, {%r2262, %r2263}; + sub.f32x2 %rd651, %rd649, %rd650; + mov.b64 %rd652, {%r1240, %r1241}; + mov.b64 %rd653, {%r2264, %r2265}; + sub.f32x2 %rd654, %rd652, %rd653; + mov.b64 %rd655, {%r1242, %r1243}; + mov.b64 %rd656, {%r2266, %r2267}; + sub.f32x2 %rd657, %rd655, %rd656; + mov.b64 %rd658, {%r1244, %r1245}; + mov.b64 %rd659, {%r2268, %r2269}; + sub.f32x2 %rd660, %rd658, %rd659; + mov.b64 %rd661, {%r1246, %r1247}; + mov.b64 %rd662, {%r2270, %r2271}; + sub.f32x2 %rd663, %rd661, %rd662; + .loc 1 108 27 // l2norm.py:108:27 + mov.b64 {%r2304, %r2305}, %rd306; + cvt.rn.bf16x2.f32 %r268, %r2305, %r2304; + mov.b64 {%r2306, %r2307}, %rd309; + cvt.rn.bf16x2.f32 %r269, %r2307, %r2306; + mov.b64 {%r2308, %r2309}, %rd312; + cvt.rn.bf16x2.f32 %r270, %r2309, %r2308; + mov.b64 {%r2310, %r2311}, %rd315; + cvt.rn.bf16x2.f32 %r271, %r2311, %r2310; + mov.b64 {%r2312, %r2313}, %rd318; + cvt.rn.bf16x2.f32 %r272, %r2313, %r2312; + mov.b64 {%r2314, %r2315}, %rd321; + cvt.rn.bf16x2.f32 %r273, %r2315, %r2314; + mov.b64 {%r2316, %r2317}, %rd324; + cvt.rn.bf16x2.f32 %r274, %r2317, %r2316; + mov.b64 {%r2318, %r2319}, %rd327; + cvt.rn.bf16x2.f32 %r275, %r2319, %r2318; + mov.b64 {%r2320, %r2321}, %rd330; + cvt.rn.bf16x2.f32 %r276, %r2321, %r2320; + mov.b64 {%r2322, %r2323}, %rd333; + cvt.rn.bf16x2.f32 %r277, %r2323, %r2322; + mov.b64 {%r2324, %r2325}, %rd336; + cvt.rn.bf16x2.f32 %r278, %r2325, %r2324; + mov.b64 {%r2326, %r2327}, %rd339; + cvt.rn.bf16x2.f32 %r279, %r2327, %r2326; + mov.b64 {%r2328, %r2329}, %rd342; + cvt.rn.bf16x2.f32 %r280, %r2329, %r2328; + mov.b64 {%r2330, %r2331}, %rd345; + cvt.rn.bf16x2.f32 %r281, %r2331, %r2330; + mov.b64 {%r2332, %r2333}, %rd348; + cvt.rn.bf16x2.f32 %r282, %r2333, %r2332; + mov.b64 {%r2334, %r2335}, %rd351; + cvt.rn.bf16x2.f32 %r283, %r2335, %r2334; + mov.b64 {%r2336, %r2337}, %rd354; + cvt.rn.bf16x2.f32 %r284, %r2337, %r2336; + mov.b64 {%r2338, %r2339}, %rd357; + cvt.rn.bf16x2.f32 %r285, %r2339, %r2338; + mov.b64 {%r2340, %r2341}, %rd360; + cvt.rn.bf16x2.f32 %r286, %r2341, %r2340; + mov.b64 {%r2342, %r2343}, %rd363; + cvt.rn.bf16x2.f32 %r287, %r2343, %r2342; + mov.b64 {%r2344, %r2345}, %rd366; + cvt.rn.bf16x2.f32 %r288, %r2345, %r2344; + mov.b64 {%r2346, %r2347}, %rd369; + cvt.rn.bf16x2.f32 %r289, %r2347, %r2346; + mov.b64 {%r2348, %r2349}, %rd372; + cvt.rn.bf16x2.f32 %r290, %r2349, %r2348; + mov.b64 {%r2350, %r2351}, %rd375; + cvt.rn.bf16x2.f32 %r291, %r2351, %r2350; + mov.b64 {%r2352, %r2353}, %rd378; + cvt.rn.bf16x2.f32 %r292, %r2353, %r2352; + mov.b64 {%r2354, %r2355}, %rd381; + cvt.rn.bf16x2.f32 %r293, %r2355, %r2354; + mov.b64 {%r2356, %r2357}, %rd384; + cvt.rn.bf16x2.f32 %r294, %r2357, %r2356; + mov.b64 {%r2358, %r2359}, %rd387; + cvt.rn.bf16x2.f32 %r295, %r2359, %r2358; + mov.b64 {%r2360, %r2361}, %rd390; + cvt.rn.bf16x2.f32 %r296, %r2361, %r2360; + mov.b64 {%r2362, %r2363}, %rd393; + cvt.rn.bf16x2.f32 %r297, %r2363, %r2362; + mov.b64 {%r2364, %r2365}, %rd396; + cvt.rn.bf16x2.f32 %r298, %r2365, %r2364; + mov.b64 {%r2366, %r2367}, %rd399; + cvt.rn.bf16x2.f32 %r299, %r2367, %r2366; + mov.b64 {%r2368, %r2369}, %rd402; + cvt.rn.bf16x2.f32 %r300, %r2369, %r2368; + mov.b64 {%r2370, %r2371}, %rd405; + cvt.rn.bf16x2.f32 %r301, %r2371, %r2370; + mov.b64 {%r2372, %r2373}, %rd408; + cvt.rn.bf16x2.f32 %r302, %r2373, %r2372; + mov.b64 {%r2374, %r2375}, %rd411; + cvt.rn.bf16x2.f32 %r303, %r2375, %r2374; + mov.b64 {%r2376, %r2377}, %rd414; + cvt.rn.bf16x2.f32 %r304, %r2377, %r2376; + mov.b64 {%r2378, %r2379}, %rd417; + cvt.rn.bf16x2.f32 %r305, %r2379, %r2378; + mov.b64 {%r2380, %r2381}, %rd420; + cvt.rn.bf16x2.f32 %r306, %r2381, %r2380; + mov.b64 {%r2382, %r2383}, %rd423; + cvt.rn.bf16x2.f32 %r307, %r2383, %r2382; + mov.b64 {%r2384, %r2385}, %rd426; + cvt.rn.bf16x2.f32 %r308, %r2385, %r2384; + mov.b64 {%r2386, %r2387}, %rd429; + cvt.rn.bf16x2.f32 %r309, %r2387, %r2386; + mov.b64 {%r2388, %r2389}, %rd432; + cvt.rn.bf16x2.f32 %r310, %r2389, %r2388; + mov.b64 {%r2390, %r2391}, %rd435; + cvt.rn.bf16x2.f32 %r311, %r2391, %r2390; + mov.b64 {%r2392, %r2393}, %rd438; + cvt.rn.bf16x2.f32 %r312, %r2393, %r2392; + mov.b64 {%r2394, %r2395}, %rd441; + cvt.rn.bf16x2.f32 %r313, %r2395, %r2394; + mov.b64 {%r2396, %r2397}, %rd444; + cvt.rn.bf16x2.f32 %r314, %r2397, %r2396; + mov.b64 {%r2398, %r2399}, %rd447; + cvt.rn.bf16x2.f32 %r315, %r2399, %r2398; + mov.b64 {%r2400, %r2401}, %rd450; + cvt.rn.bf16x2.f32 %r316, %r2401, %r2400; + mov.b64 {%r2402, %r2403}, %rd453; + cvt.rn.bf16x2.f32 %r317, %r2403, %r2402; + mov.b64 {%r2404, %r2405}, %rd456; + cvt.rn.bf16x2.f32 %r318, %r2405, %r2404; + mov.b64 {%r2406, %r2407}, %rd459; + cvt.rn.bf16x2.f32 %r319, %r2407, %r2406; + mov.b64 {%r2408, %r2409}, %rd462; + cvt.rn.bf16x2.f32 %r320, %r2409, %r2408; + mov.b64 {%r2410, %r2411}, %rd465; + cvt.rn.bf16x2.f32 %r321, %r2411, %r2410; + mov.b64 {%r2412, %r2413}, %rd468; + cvt.rn.bf16x2.f32 %r322, %r2413, %r2412; + mov.b64 {%r2414, %r2415}, %rd471; + cvt.rn.bf16x2.f32 %r323, %r2415, %r2414; + mov.b64 {%r2416, %r2417}, %rd474; + cvt.rn.bf16x2.f32 %r324, %r2417, %r2416; + mov.b64 {%r2418, %r2419}, %rd477; + cvt.rn.bf16x2.f32 %r325, %r2419, %r2418; + mov.b64 {%r2420, %r2421}, %rd480; + cvt.rn.bf16x2.f32 %r326, %r2421, %r2420; + mov.b64 {%r2422, %r2423}, %rd483; + cvt.rn.bf16x2.f32 %r327, %r2423, %r2422; + mov.b64 {%r2424, %r2425}, %rd486; + cvt.rn.bf16x2.f32 %r328, %r2425, %r2424; + mov.b64 {%r2426, %r2427}, %rd489; + cvt.rn.bf16x2.f32 %r329, %r2427, %r2426; + mov.b64 {%r2428, %r2429}, %rd492; + cvt.rn.bf16x2.f32 %r330, %r2429, %r2428; + mov.b64 {%r2430, %r2431}, %rd495; + cvt.rn.bf16x2.f32 %r331, %r2431, %r2430; + mov.b64 {%r2432, %r2433}, %rd498; + cvt.rn.bf16x2.f32 %r332, %r2433, %r2432; + mov.b64 {%r2434, %r2435}, %rd501; + cvt.rn.bf16x2.f32 %r333, %r2435, %r2434; + mov.b64 {%r2436, %r2437}, %rd504; + cvt.rn.bf16x2.f32 %r334, %r2437, %r2436; + mov.b64 {%r2438, %r2439}, %rd507; + cvt.rn.bf16x2.f32 %r335, %r2439, %r2438; + mov.b64 {%r2440, %r2441}, %rd510; + cvt.rn.bf16x2.f32 %r336, %r2441, %r2440; + mov.b64 {%r2442, %r2443}, %rd513; + cvt.rn.bf16x2.f32 %r337, %r2443, %r2442; + mov.b64 {%r2444, %r2445}, %rd516; + cvt.rn.bf16x2.f32 %r338, %r2445, %r2444; + mov.b64 {%r2446, %r2447}, %rd519; + cvt.rn.bf16x2.f32 %r339, %r2447, %r2446; + mov.b64 {%r2448, %r2449}, %rd522; + cvt.rn.bf16x2.f32 %r340, %r2449, %r2448; + mov.b64 {%r2450, %r2451}, %rd525; + cvt.rn.bf16x2.f32 %r341, %r2451, %r2450; + mov.b64 {%r2452, %r2453}, %rd528; + cvt.rn.bf16x2.f32 %r342, %r2453, %r2452; + mov.b64 {%r2454, %r2455}, %rd531; + cvt.rn.bf16x2.f32 %r343, %r2455, %r2454; + mov.b64 {%r2456, %r2457}, %rd534; + cvt.rn.bf16x2.f32 %r344, %r2457, %r2456; + mov.b64 {%r2458, %r2459}, %rd537; + cvt.rn.bf16x2.f32 %r345, %r2459, %r2458; + mov.b64 {%r2460, %r2461}, %rd540; + cvt.rn.bf16x2.f32 %r346, %r2461, %r2460; + mov.b64 {%r2462, %r2463}, %rd543; + cvt.rn.bf16x2.f32 %r347, %r2463, %r2462; + mov.b64 {%r2464, %r2465}, %rd546; + cvt.rn.bf16x2.f32 %r348, %r2465, %r2464; + mov.b64 {%r2466, %r2467}, %rd549; + cvt.rn.bf16x2.f32 %r349, %r2467, %r2466; + mov.b64 {%r2468, %r2469}, %rd552; + cvt.rn.bf16x2.f32 %r350, %r2469, %r2468; + mov.b64 {%r2470, %r2471}, %rd555; + cvt.rn.bf16x2.f32 %r351, %r2471, %r2470; + mov.b64 {%r2472, %r2473}, %rd558; + cvt.rn.bf16x2.f32 %r352, %r2473, %r2472; + mov.b64 {%r2474, %r2475}, %rd561; + cvt.rn.bf16x2.f32 %r353, %r2475, %r2474; + mov.b64 {%r2476, %r2477}, %rd564; + cvt.rn.bf16x2.f32 %r354, %r2477, %r2476; + mov.b64 {%r2478, %r2479}, %rd567; + cvt.rn.bf16x2.f32 %r355, %r2479, %r2478; + mov.b64 {%r2480, %r2481}, %rd570; + cvt.rn.bf16x2.f32 %r356, %r2481, %r2480; + mov.b64 {%r2482, %r2483}, %rd573; + cvt.rn.bf16x2.f32 %r357, %r2483, %r2482; + mov.b64 {%r2484, %r2485}, %rd576; + cvt.rn.bf16x2.f32 %r358, %r2485, %r2484; + mov.b64 {%r2486, %r2487}, %rd579; + cvt.rn.bf16x2.f32 %r359, %r2487, %r2486; + mov.b64 {%r2488, %r2489}, %rd582; + cvt.rn.bf16x2.f32 %r360, %r2489, %r2488; + mov.b64 {%r2490, %r2491}, %rd585; + cvt.rn.bf16x2.f32 %r361, %r2491, %r2490; + mov.b64 {%r2492, %r2493}, %rd588; + cvt.rn.bf16x2.f32 %r362, %r2493, %r2492; + mov.b64 {%r2494, %r2495}, %rd591; + cvt.rn.bf16x2.f32 %r363, %r2495, %r2494; + mov.b64 {%r2496, %r2497}, %rd594; + cvt.rn.bf16x2.f32 %r364, %r2497, %r2496; + mov.b64 {%r2498, %r2499}, %rd597; + cvt.rn.bf16x2.f32 %r365, %r2499, %r2498; + mov.b64 {%r2500, %r2501}, %rd600; + cvt.rn.bf16x2.f32 %r366, %r2501, %r2500; + mov.b64 {%r2502, %r2503}, %rd603; + cvt.rn.bf16x2.f32 %r367, %r2503, %r2502; + mov.b64 {%r2504, %r2505}, %rd606; + cvt.rn.bf16x2.f32 %r368, %r2505, %r2504; + mov.b64 {%r2506, %r2507}, %rd609; + cvt.rn.bf16x2.f32 %r369, %r2507, %r2506; + mov.b64 {%r2508, %r2509}, %rd612; + cvt.rn.bf16x2.f32 %r370, %r2509, %r2508; + mov.b64 {%r2510, %r2511}, %rd615; + cvt.rn.bf16x2.f32 %r371, %r2511, %r2510; + mov.b64 {%r2512, %r2513}, %rd618; + cvt.rn.bf16x2.f32 %r372, %r2513, %r2512; + mov.b64 {%r2514, %r2515}, %rd621; + cvt.rn.bf16x2.f32 %r373, %r2515, %r2514; + mov.b64 {%r2516, %r2517}, %rd624; + cvt.rn.bf16x2.f32 %r374, %r2517, %r2516; + mov.b64 {%r2518, %r2519}, %rd627; + cvt.rn.bf16x2.f32 %r375, %r2519, %r2518; + mov.b64 {%r2520, %r2521}, %rd630; + cvt.rn.bf16x2.f32 %r376, %r2521, %r2520; + mov.b64 {%r2522, %r2523}, %rd633; + cvt.rn.bf16x2.f32 %r377, %r2523, %r2522; + mov.b64 {%r2524, %r2525}, %rd636; + cvt.rn.bf16x2.f32 %r378, %r2525, %r2524; + mov.b64 {%r2526, %r2527}, %rd639; + cvt.rn.bf16x2.f32 %r379, %r2527, %r2526; + mov.b64 {%r2528, %r2529}, %rd642; + cvt.rn.bf16x2.f32 %r380, %r2529, %r2528; + mov.b64 {%r2530, %r2531}, %rd645; + cvt.rn.bf16x2.f32 %r381, %r2531, %r2530; + mov.b64 {%r2532, %r2533}, %rd648; + cvt.rn.bf16x2.f32 %r382, %r2533, %r2532; + mov.b64 {%r2534, %r2535}, %rd651; + cvt.rn.bf16x2.f32 %r383, %r2535, %r2534; + mov.b64 {%r2536, %r2537}, %rd654; + cvt.rn.bf16x2.f32 %r384, %r2537, %r2536; + mov.b64 {%r2538, %r2539}, %rd657; + cvt.rn.bf16x2.f32 %r385, %r2539, %r2538; + mov.b64 {%r2540, %r2541}, %rd660; + cvt.rn.bf16x2.f32 %r386, %r2541, %r2540; + mov.b64 {%r2542, %r2543}, %rd663; + cvt.rn.bf16x2.f32 %r387, %r2543, %r2542; + .loc 1 108 19 // l2norm.py:108:19 + cvt.rn.bf16x2.f32 %r260, %r2275, %r2273; + cvt.rn.bf16x2.f32 %r261, %r2279, %r2277; + cvt.rn.bf16x2.f32 %r262, %r2283, %r2281; + cvt.rn.bf16x2.f32 %r263, %r2287, %r2285; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd67 + 0 ], { %r260, %r261, %r262, %r263 }; + // end inline asm + cvt.rn.bf16x2.f32 %r264, %r2291, %r2289; + cvt.rn.bf16x2.f32 %r265, %r2295, %r2293; + cvt.rn.bf16x2.f32 %r266, %r2299, %r2297; + cvt.rn.bf16x2.f32 %r267, %r2303, %r2301; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd68 + 0 ], { %r264, %r265, %r266, %r267 }; + // end inline asm + // begin inline asm + @%p6 st.global.v4.b32 [ %rd69 + 0 ], { %r268, %r269, %r270, %r271 }; + // end inline asm + // begin inline asm + @%p7 st.global.v4.b32 [ %rd70 + 0 ], { %r272, %r273, %r274, %r275 }; + // end inline asm + // begin inline asm + @%p8 st.global.v4.b32 [ %rd71 + 0 ], { %r276, %r277, %r278, %r279 }; + // end inline asm + // begin inline asm + @%p9 st.global.v4.b32 [ %rd72 + 0 ], { %r280, %r281, %r282, %r283 }; + // end inline asm + // begin inline asm + @%p10 st.global.v4.b32 [ %rd73 + 0 ], { %r284, %r285, %r286, %r287 }; + // end inline asm + // begin inline asm + @%p11 st.global.v4.b32 [ %rd74 + 0 ], { %r288, %r289, %r290, %r291 }; + // end inline asm + // begin inline asm + @%p12 st.global.v4.b32 [ %rd75 + 0 ], { %r292, %r293, %r294, %r295 }; + // end inline asm + // begin inline asm + @%p3 st.global.v4.b32 [ %rd76 + 0 ], { %r296, %r297, %r298, %r299 }; + // end inline asm + // begin inline asm + @%p13 st.global.v4.b32 [ %rd77 + 0 ], { %r300, %r301, %r302, %r303 }; + // end inline asm + // begin inline asm + @%p14 st.global.v4.b32 [ %rd78 + 0 ], { %r304, %r305, %r306, %r307 }; + // end inline asm + // begin inline asm + @%p15 st.global.v4.b32 [ %rd79 + 0 ], { %r308, %r309, %r310, %r311 }; + // end inline asm + // begin inline asm + @%p16 st.global.v4.b32 [ %rd80 + 0 ], { %r312, %r313, %r314, %r315 }; + // end inline asm + // begin inline asm + @%p17 st.global.v4.b32 [ %rd81 + 0 ], { %r316, %r317, %r318, %r319 }; + // end inline asm + // begin inline asm + @%p18 st.global.v4.b32 [ %rd82 + 0 ], { %r320, %r321, %r322, %r323 }; + // end inline asm + // begin inline asm + @%p19 st.global.v4.b32 [ %rd83 + 0 ], { %r324, %r325, %r326, %r327 }; + // end inline asm + // begin inline asm + @%p20 st.global.v4.b32 [ %rd84 + 0 ], { %r328, %r329, %r330, %r331 }; + // end inline asm + // begin inline asm + @%p21 st.global.v4.b32 [ %rd85 + 0 ], { %r332, %r333, %r334, %r335 }; + // end inline asm + // begin inline asm + @%p22 st.global.v4.b32 [ %rd86 + 0 ], { %r336, %r337, %r338, %r339 }; + // end inline asm + // begin inline asm + @%p23 st.global.v4.b32 [ %rd87 + 0 ], { %r340, %r341, %r342, %r343 }; + // end inline asm + // begin inline asm + @%p24 st.global.v4.b32 [ %rd88 + 0 ], { %r344, %r345, %r346, %r347 }; + // end inline asm + // begin inline asm + @%p25 st.global.v4.b32 [ %rd89 + 0 ], { %r348, %r349, %r350, %r351 }; + // end inline asm + // begin inline asm + @%p26 st.global.v4.b32 [ %rd90 + 0 ], { %r352, %r353, %r354, %r355 }; + // end inline asm + // begin inline asm + @%p27 st.global.v4.b32 [ %rd91 + 0 ], { %r356, %r357, %r358, %r359 }; + // end inline asm + // begin inline asm + @%p28 st.global.v4.b32 [ %rd92 + 0 ], { %r360, %r361, %r362, %r363 }; + // end inline asm + // begin inline asm + @%p29 st.global.v4.b32 [ %rd93 + 0 ], { %r364, %r365, %r366, %r367 }; + // end inline asm + // begin inline asm + @%p30 st.global.v4.b32 [ %rd94 + 0 ], { %r368, %r369, %r370, %r371 }; + // end inline asm + // begin inline asm + @%p31 st.global.v4.b32 [ %rd95 + 0 ], { %r372, %r373, %r374, %r375 }; + // end inline asm + // begin inline asm + @%p32 st.global.v4.b32 [ %rd96 + 0 ], { %r376, %r377, %r378, %r379 }; + // end inline asm + // begin inline asm + @%p33 st.global.v4.b32 [ %rd97 + 0 ], { %r380, %r381, %r382, %r383 }; + // end inline asm + // begin inline asm + @%p34 st.global.v4.b32 [ %rd98 + 0 ], { %r384, %r385, %r386, %r387 }; + // end inline asm + .loc 1 108 4 // l2norm.py:108:4 + ret; +$L__tmp259: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 98 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp258 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 107 // DW_AT_call_line +.b8 55 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp258 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.source b/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..e5004915e0ed656b5a11e815a4f48ddf52d64251 --- /dev/null +++ b/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.source @@ -0,0 +1,210 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc47 = loc(unknown) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc54 = loc("y"(#loc)) +#loc55 = loc("rstd"(#loc)) +#loc56 = loc("dy"(#loc)) +#loc57 = loc("dx"(#loc)) +#loc58 = loc("eps"(#loc)) +#loc59 = loc("T"(#loc)) +#loc101 = loc("input"(#loc45)) +#loc102 = loc("a"(#loc50)) +#loc103 = loc("b"(#loc50)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc60) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc61) + %o_t = arith.constant 64 : i32 loc(#loc62) + %o_t_1 = arith.constant 64 : i64 loc(#loc62) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc62) + %o_t_3 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc63) + %o_t_4 = arith.extsi %o_t_3 : tensor<64xi32> to tensor<64xi64> loc(#loc64) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<64xi64> loc(#loc64) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<64xi64> loc(#loc64) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc65) + %m_t = arith.extsi %T : i32 to i64 loc(#loc66) + %m_t_7 = tt.splat %m_t : i64 -> tensor<64xi64> loc(#loc66) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<64xi64> loc(#loc66) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc67) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc68) + %m_x_10 = arith.constant 128 : i32 loc(#loc69) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc69) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc69) + %m_x_13 = tt.broadcast %m_x : tensor<64x1xi1> -> tensor<64x128xi1> loc(#loc70) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<64x128xi1> loc(#loc70) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<64x128xi1> loc(#loc70) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc71) + %p_y_16 = arith.constant 128 : i32 loc(#loc72) + %p_y_17 = arith.constant 128 : i64 loc(#loc72) + %p_y_18 = arith.constant dense<128> : tensor<64x1xi64> loc(#loc72) + %p_y_19 = arith.muli %p_y, %p_y_18 : tensor<64x1xi64> loc(#loc72) + %p_y_20 = tt.splat %y : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc73) + %p_y_21 = tt.addptr %p_y_20, %p_y_19 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc73) + %p_y_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc74) + %p_y_23 = tt.broadcast %p_y_21 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc75) + %p_y_24 = tt.broadcast %p_y_22 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc75) + %p_y_25 = tt.addptr %p_y_23, %p_y_24 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc75) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<64x!tt.ptr> loc(#loc76) + %p_rstd_26 = tt.addptr %p_rstd, %o_t_6 : tensor<64x!tt.ptr>, tensor<64xi64> loc(#loc76) + %p_dy = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc77) + %p_dy_27 = arith.constant 128 : i32 loc(#loc78) + %p_dy_28 = arith.constant 128 : i64 loc(#loc78) + %p_dy_29 = arith.constant dense<128> : tensor<64x1xi64> loc(#loc78) + %p_dy_30 = arith.muli %p_dy, %p_dy_29 : tensor<64x1xi64> loc(#loc78) + %p_dy_31 = tt.splat %dy : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc79) + %p_dy_32 = tt.addptr %p_dy_31, %p_dy_30 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc79) + %p_dy_33 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc80) + %p_dy_34 = tt.broadcast %p_dy_32 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc81) + %p_dy_35 = tt.broadcast %p_dy_33 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc81) + %p_dy_36 = tt.addptr %p_dy_34, %p_dy_35 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc81) + %p_dx = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc82) + %p_dx_37 = arith.constant 128 : i32 loc(#loc83) + %p_dx_38 = arith.constant 128 : i64 loc(#loc83) + %p_dx_39 = arith.constant dense<128> : tensor<64x1xi64> loc(#loc83) + %p_dx_40 = arith.muli %p_dx, %p_dx_39 : tensor<64x1xi64> loc(#loc83) + %p_dx_41 = tt.splat %dx : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc84) + %p_dx_42 = tt.addptr %p_dx_41, %p_dx_40 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc84) + %p_dx_43 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc85) + %p_dx_44 = tt.broadcast %p_dx_42 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc86) + %p_dx_45 = tt.broadcast %p_dx_43 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc86) + %p_dx_46 = tt.addptr %p_dx_44, %p_dx_45 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc86) + %b_y = arith.constant 0.000000e+00 : f32 loc(#loc87) + %b_y_47 = arith.constant dense<0.000000e+00> : tensor<64x128xf32> loc(#loc87) + %b_y_48 = arith.truncf %b_y_47 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc87) + %b_y_49 = tt.load %p_y_25, %m_x_15, %b_y_48 : tensor<64x128x!tt.ptr> loc(#loc87) + %b_y_50 = arith.extf %b_y_49 : tensor<64x128xbf16> to tensor<64x128xf32> loc(#loc88) + %b_rstd = arith.constant 0.000000e+00 : f32 loc(#loc89) + %b_rstd_51 = arith.constant dense<0.000000e+00> : tensor<64xf32> loc(#loc89) + %b_rstd_52 = tt.load %p_rstd_26, %m_t_8, %b_rstd_51 : tensor<64x!tt.ptr> loc(#loc89) + %b_dy = arith.constant 0.000000e+00 : f32 loc(#loc90) + %b_dy_53 = arith.constant dense<0.000000e+00> : tensor<64x128xf32> loc(#loc90) + %b_dy_54 = arith.truncf %b_dy_53 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc90) + %b_dy_55 = tt.load %p_dy_36, %m_x_15, %b_dy_54 : tensor<64x128x!tt.ptr> loc(#loc90) + %b_dy_56 = arith.extf %b_dy_55 : tensor<64x128xbf16> to tensor<64x128xf32> loc(#loc91) + %b_dx = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc92) + %b_dx_57 = tt.broadcast %b_dx : tensor<64x1xf32> -> tensor<64x128xf32> loc(#loc93) + %b_dx_58 = arith.mulf %b_dy_56, %b_dx_57 : tensor<64x128xf32> loc(#loc93) + %b_dx_59 = arith.mulf %b_dy_56, %b_y_50 : tensor<64x128xf32> loc(#loc94) + %b_dx_60 = tt.call @triton.language.standard.sum__fp32S64_128S_c1_cFalse_cNone(%b_dx_59) : (tensor<64x128xf32>) -> tensor<64xf32> loc(#loc95) + %b_dx_61 = tt.expand_dims %b_dx_60 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc96) + %b_dx_62 = tt.broadcast %b_dx_61 : tensor<64x1xf32> -> tensor<64x128xf32> loc(#loc97) + %b_dx_63 = arith.mulf %b_dx_62, %b_y_50 : tensor<64x128xf32> loc(#loc97) + %b_dx_64 = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc98) + %b_dx_65 = tt.broadcast %b_dx_64 : tensor<64x1xf32> -> tensor<64x128xf32> loc(#loc99) + %b_dx_66 = arith.mulf %b_dx_63, %b_dx_65 : tensor<64x128xf32> loc(#loc99) + %b_dx_67 = arith.subf %b_dx_58, %b_dx_66 : tensor<64x128xf32> loc(#loc100) + %0 = arith.truncf %b_dx_67 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc42) + tt.store %p_dx_46, %0, %m_x_15 : tensor<64x128x!tt.ptr> loc(#loc43) + tt.return loc(#loc44) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S64_128S_c1_cFalse_cNone(%input: tensor<64x128xf32> loc("input"(#loc45))) -> tensor<64xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc46) + tt.reduce.return %2 : f32 loc(#loc46) + }) : (tensor<64x128xf32>) -> tensor<64xf32> loc(#loc46) + tt.return %0 : tensor<64xf32> loc(#loc48) + ^bb1: // no predecessors + %1 = ub.poison : tensor<64xf32> loc(#loc49) + tt.return %1 : tensor<64xf32> loc(#loc49) + } loc(#loc45) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc50)), %b: f32 loc("b"(#loc50))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc51) + tt.return %0 : f32 loc(#loc52) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc53) + tt.return %1 : f32 loc(#loc53) + } loc(#loc50) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:31) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:39) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:20) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:31) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:39) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:82) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc60 = loc("i_t"(#loc1)) +#loc61 = loc("i_t"(#loc2)) +#loc62 = loc("o_t"(#loc3)) +#loc63 = loc("o_t"(#loc4)) +#loc64 = loc("o_t"(#loc5)) +#loc65 = loc("o_d"(#loc6)) +#loc66 = loc("m_t"(#loc7)) +#loc67 = loc("m_x"(#loc8)) +#loc68 = loc("m_x"(#loc9)) +#loc69 = loc("m_x"(#loc10)) +#loc70 = loc("m_x"(#loc11)) +#loc71 = loc("p_y"(#loc12)) +#loc72 = loc("p_y"(#loc13)) +#loc73 = loc("p_y"(#loc14)) +#loc74 = loc("p_y"(#loc15)) +#loc75 = loc("p_y"(#loc16)) +#loc76 = loc("p_rstd"(#loc17)) +#loc77 = loc("p_dy"(#loc18)) +#loc78 = loc("p_dy"(#loc19)) +#loc79 = loc("p_dy"(#loc20)) +#loc80 = loc("p_dy"(#loc21)) +#loc81 = loc("p_dy"(#loc22)) +#loc82 = loc("p_dx"(#loc23)) +#loc83 = loc("p_dx"(#loc24)) +#loc84 = loc("p_dx"(#loc25)) +#loc85 = loc("p_dx"(#loc26)) +#loc86 = loc("p_dx"(#loc27)) +#loc87 = loc("b_y"(#loc28)) +#loc88 = loc("b_y"(#loc29)) +#loc89 = loc("b_rstd"(#loc30)) +#loc90 = loc("b_dy"(#loc31)) +#loc91 = loc("b_dy"(#loc32)) +#loc92 = loc("b_dx"(#loc33)) +#loc93 = loc("b_dx"(#loc34)) +#loc94 = loc("b_dx"(#loc35)) +#loc95 = loc("b_dx"(#loc36)) +#loc96 = loc("b_dx"(#loc37)) +#loc97 = loc("b_dx"(#loc38)) +#loc98 = loc("b_dx"(#loc39)) +#loc99 = loc("b_dx"(#loc40)) +#loc100 = loc("b_dx"(#loc41)) diff --git a/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.ttgir b/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..877a1c25529bba49d8cd306b8bfde5783b216688 --- /dev/null +++ b/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.ttgir @@ -0,0 +1,154 @@ +#blocked = #ttg.blocked<{sizePerThread = [2], threadsPerWarp = [32], warpsPerCTA = [1], order = [0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [1, 1], order = [1, 0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc39 = loc("y"(#loc)) +#loc40 = loc("rstd"(#loc)) +#loc41 = loc("dy"(#loc)) +#loc42 = loc("dx"(#loc)) +#loc43 = loc("eps"(#loc)) +#loc44 = loc("T"(#loc)) +#loc72 = loc("b_dx"(#loc30)) +#loc78 = loc(callsite(#loc1 at #loc72)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 1 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %c64_i64 = arith.constant 64 : i64 loc(#loc1) + %cst = arith.constant dense<0.000000e+00> : tensor<64xf32, #blocked> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<64x128xbf16, #blocked1> loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<64x1xi64, #blocked1> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<1x128xi32, #blocked1> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc45) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc46) + %o_t = arith.muli %i_t_3, %c64_i64 : i64 loc(#loc47) + %o_t_4 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc48) + %o_t_5 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #blocked> loc(#loc48) + %o_t_6 = arith.extsi %o_t_4 : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> to tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc49) + %o_t_7 = arith.extsi %o_t_5 : tensor<64xi32, #blocked> to tensor<64xi64, #blocked> loc(#loc49) + %o_t_8 = tt.splat %o_t : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc49) + %o_t_9 = tt.splat %o_t : i64 -> tensor<64xi64, #blocked> loc(#loc49) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc49) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<64xi64, #blocked> loc(#loc49) + %m_t = arith.extsi %T : i32 to i64 loc(#loc50) + %m_t_12 = tt.splat %m_t : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc50) + %m_t_13 = tt.splat %m_t : i64 -> tensor<64xi64, #blocked> loc(#loc50) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc50) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<64xi64, #blocked> loc(#loc50) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<64xi1, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<64x1xi1, #blocked1> loc(#loc51) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc52) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x128xi32, #blocked1> loc(#loc52) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_2 : tensor<1x128xi32, #blocked1> loc(#loc53) + %m_x_19 = tt.broadcast %m_x : tensor<64x1xi1, #blocked1> -> tensor<64x128xi1, #blocked1> loc(#loc54) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked1> -> tensor<64x128xi1, #blocked1> loc(#loc54) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<64x128xi1, #blocked1> loc(#loc54) + %p_y = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<64x1xi64, #blocked1> loc(#loc55) + %p_y_22 = arith.muli %p_y, %cst_1 : tensor<64x1xi64, #blocked1> loc(#loc56) + %p_y_23 = tt.splat %y : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked1> loc(#loc57) + %p_y_24 = tt.addptr %p_y_23, %p_y_22 : tensor<64x1x!tt.ptr, #blocked1>, tensor<64x1xi64, #blocked1> loc(#loc57) + %p_y_25 = tt.broadcast %p_y_24 : tensor<64x1x!tt.ptr, #blocked1> -> tensor<64x128x!tt.ptr, #blocked1> loc(#loc58) + %p_y_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked1> -> tensor<64x128xi32, #blocked1> loc(#loc58) + %p_y_27 = tt.addptr %p_y_25, %p_y_26 : tensor<64x128x!tt.ptr, #blocked1>, tensor<64x128xi32, #blocked1> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<64x!tt.ptr, #blocked> loc(#loc59) + %p_rstd_28 = tt.addptr %p_rstd, %o_t_11 : tensor<64x!tt.ptr, #blocked>, tensor<64xi64, #blocked> loc(#loc59) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked1> loc(#loc60) + %p_dy_29 = tt.addptr %p_dy, %p_y_22 : tensor<64x1x!tt.ptr, #blocked1>, tensor<64x1xi64, #blocked1> loc(#loc60) + %p_dy_30 = tt.broadcast %p_dy_29 : tensor<64x1x!tt.ptr, #blocked1> -> tensor<64x128x!tt.ptr, #blocked1> loc(#loc61) + %p_dy_31 = tt.addptr %p_dy_30, %p_y_26 : tensor<64x128x!tt.ptr, #blocked1>, tensor<64x128xi32, #blocked1> loc(#loc61) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked1> loc(#loc62) + %p_dx_32 = tt.addptr %p_dx, %p_y_22 : tensor<64x1x!tt.ptr, #blocked1>, tensor<64x1xi64, #blocked1> loc(#loc62) + %p_dx_33 = tt.broadcast %p_dx_32 : tensor<64x1x!tt.ptr, #blocked1> -> tensor<64x128x!tt.ptr, #blocked1> loc(#loc63) + %p_dx_34 = tt.addptr %p_dx_33, %p_y_26 : tensor<64x128x!tt.ptr, #blocked1>, tensor<64x128xi32, #blocked1> loc(#loc63) + %b_y = tt.load %p_y_27, %m_x_21, %cst_0 : tensor<64x128x!tt.ptr, #blocked1> loc(#loc64) + %b_y_35 = arith.extf %b_y : tensor<64x128xbf16, #blocked1> to tensor<64x128xf32, #blocked1> loc(#loc65) + %b_rstd = tt.load %p_rstd_28, %m_t_15, %cst : tensor<64x!tt.ptr, #blocked> loc(#loc66) + %b_dy = tt.load %p_dy_31, %m_x_21, %cst_0 : tensor<64x128x!tt.ptr, #blocked1> loc(#loc67) + %b_dy_36 = arith.extf %b_dy : tensor<64x128xbf16, #blocked1> to tensor<64x128xf32, #blocked1> loc(#loc68) + %b_dx = ttg.convert_layout %b_rstd : tensor<64xf32, #blocked> -> tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc69) + %b_dx_37 = tt.expand_dims %b_dx {axis = 1 : i32} : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<64x1xf32, #blocked1> loc(#loc69) + %b_dx_38 = tt.broadcast %b_dx_37 : tensor<64x1xf32, #blocked1> -> tensor<64x128xf32, #blocked1> loc(#loc70) + %b_dx_39 = arith.mulf %b_dy_36, %b_dx_38 : tensor<64x128xf32, #blocked1> loc(#loc70) + %b_dx_40 = arith.mulf %b_dy_36, %b_y_35 : tensor<64x128xf32, #blocked1> loc(#loc71) + %b_dx_41 = "tt.reduce"(%b_dx_40) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_47: f32 loc(callsite(#loc1 at #loc72)), %b_dx_48: f32 loc(callsite(#loc1 at #loc72))): + %b_dx_49 = arith.addf %b_dx_47, %b_dx_48 : f32 loc(#loc79) + tt.reduce.return %b_dx_49 : f32 loc(#loc77) + }) : (tensor<64x128xf32, #blocked1>) -> tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc77) + %b_dx_42 = tt.expand_dims %b_dx_41 {axis = 1 : i32} : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<64x1xf32, #blocked1> loc(#loc73) + %b_dx_43 = tt.broadcast %b_dx_42 : tensor<64x1xf32, #blocked1> -> tensor<64x128xf32, #blocked1> loc(#loc74) + %b_dx_44 = arith.mulf %b_dx_43, %b_y_35 : tensor<64x128xf32, #blocked1> loc(#loc74) + %b_dx_45 = arith.mulf %b_dx_44, %b_dx_38 : tensor<64x128xf32, #blocked1> loc(#loc75) + %b_dx_46 = arith.subf %b_dx_39, %b_dx_45 : tensor<64x128xf32, #blocked1> loc(#loc76) + %0 = arith.truncf %b_dx_46 : tensor<64x128xf32, #blocked1> to tensor<64x128xbf16, #blocked1> loc(#loc36) + tt.store %p_dx_34, %0, %m_x_21 : tensor<64x128x!tt.ptr, #blocked1> loc(#loc37) + tt.return loc(#loc38) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc45 = loc("i_t"(#loc2)) +#loc46 = loc("i_t"(#loc3)) +#loc47 = loc("o_t"(#loc4)) +#loc48 = loc("o_t"(#loc5)) +#loc49 = loc("o_t"(#loc6)) +#loc50 = loc("m_t"(#loc7)) +#loc51 = loc("m_x"(#loc8)) +#loc52 = loc("m_x"(#loc9)) +#loc53 = loc("m_x"(#loc10)) +#loc54 = loc("m_x"(#loc11)) +#loc55 = loc("p_y"(#loc12)) +#loc56 = loc("p_y"(#loc13)) +#loc57 = loc("p_y"(#loc14)) +#loc58 = loc("p_y"(#loc15)) +#loc59 = loc("p_rstd"(#loc16)) +#loc60 = loc("p_dy"(#loc17)) +#loc61 = loc("p_dy"(#loc18)) +#loc62 = loc("p_dx"(#loc19)) +#loc63 = loc("p_dx"(#loc20)) +#loc64 = loc("b_y"(#loc21)) +#loc65 = loc("b_y"(#loc22)) +#loc66 = loc("b_rstd"(#loc23)) +#loc67 = loc("b_dy"(#loc24)) +#loc68 = loc("b_dy"(#loc25)) +#loc69 = loc("b_dx"(#loc26)) +#loc70 = loc("b_dx"(#loc27)) +#loc71 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc32)) +#loc74 = loc("b_dx"(#loc33)) +#loc75 = loc("b_dx"(#loc34)) +#loc76 = loc("b_dx"(#loc35)) +#loc77 = loc(callsite(#loc29 at #loc72)) +#loc79 = loc(callsite(#loc31 at #loc77)) diff --git a/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.ttir b/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..3b703edcc6d40b9b8fa5aa712507941f59d4cd5b --- /dev/null +++ b/triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.ttir @@ -0,0 +1,147 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc40 = loc("y"(#loc)) +#loc41 = loc("rstd"(#loc)) +#loc42 = loc("dy"(#loc)) +#loc43 = loc("dx"(#loc)) +#loc44 = loc("eps"(#loc)) +#loc45 = loc("T"(#loc)) +#loc74 = loc("b_dx"(#loc31)) +#loc80 = loc(callsite(#loc1 at #loc74)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<0.000000e+00> : tensor<64x128xbf16> loc(#loc1) + %b_rstd = arith.constant dense<0.000000e+00> : tensor<64xf32> loc(#loc46) + %cst_0 = arith.constant dense<128> : tensor<64x1xi64> loc(#loc1) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc47) + %o_t = arith.constant 64 : i64 loc(#loc48) + %i_t = tt.get_program_id x : i32 loc(#loc49) + %i_t_1 = arith.extsi %i_t : i32 to i64 loc(#loc50) + %o_t_2 = arith.muli %i_t_1, %o_t : i64 loc(#loc48) + %o_t_3 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc51) + %o_t_4 = arith.extsi %o_t_3 : tensor<64xi32> to tensor<64xi64> loc(#loc52) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<64xi64> loc(#loc52) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<64xi64> loc(#loc52) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc53) + %m_t = arith.extsi %T : i32 to i64 loc(#loc54) + %m_t_7 = tt.splat %m_t : i64 -> tensor<64xi64> loc(#loc54) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<64xi64> loc(#loc54) + %m_x_9 = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc55) + %m_x_10 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc56) + %m_x_11 = arith.cmpi slt, %m_x_10, %m_x : tensor<1x128xi32> loc(#loc47) + %m_x_12 = tt.broadcast %m_x_9 : tensor<64x1xi1> -> tensor<64x128xi1> loc(#loc57) + %m_x_13 = tt.broadcast %m_x_11 : tensor<1x128xi1> -> tensor<64x128xi1> loc(#loc57) + %m_x_14 = arith.andi %m_x_12, %m_x_13 : tensor<64x128xi1> loc(#loc57) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc58) + %p_y_15 = arith.muli %p_y, %cst_0 : tensor<64x1xi64> loc(#loc59) + %p_y_16 = tt.splat %y : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc60) + %p_y_17 = tt.addptr %p_y_16, %p_y_15 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc60) + %p_y_18 = tt.broadcast %p_y_17 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc61) + %p_y_19 = tt.broadcast %m_x_10 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc61) + %p_y_20 = tt.addptr %p_y_18, %p_y_19 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc61) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<64x!tt.ptr> loc(#loc62) + %p_rstd_21 = tt.addptr %p_rstd, %o_t_6 : tensor<64x!tt.ptr>, tensor<64xi64> loc(#loc62) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc63) + %p_dy_22 = tt.addptr %p_dy, %p_y_15 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc63) + %p_dy_23 = tt.broadcast %p_dy_22 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc64) + %p_dy_24 = tt.addptr %p_dy_23, %p_y_19 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc64) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc65) + %p_dx_25 = tt.addptr %p_dx, %p_y_15 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc65) + %p_dx_26 = tt.broadcast %p_dx_25 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc66) + %p_dx_27 = tt.addptr %p_dx_26, %p_y_19 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc66) + %b_y = tt.load %p_y_20, %m_x_14, %cst : tensor<64x128x!tt.ptr> loc(#loc67) + %b_y_28 = arith.extf %b_y : tensor<64x128xbf16> to tensor<64x128xf32> loc(#loc68) + %b_rstd_29 = tt.load %p_rstd_21, %m_t_8, %b_rstd : tensor<64x!tt.ptr> loc(#loc46) + %b_dy = tt.load %p_dy_24, %m_x_14, %cst : tensor<64x128x!tt.ptr> loc(#loc69) + %b_dy_30 = arith.extf %b_dy : tensor<64x128xbf16> to tensor<64x128xf32> loc(#loc70) + %b_dx = tt.expand_dims %b_rstd_29 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc71) + %b_dx_31 = tt.broadcast %b_dx : tensor<64x1xf32> -> tensor<64x128xf32> loc(#loc72) + %b_dx_32 = arith.mulf %b_dy_30, %b_dx_31 : tensor<64x128xf32> loc(#loc72) + %b_dx_33 = arith.mulf %b_dy_30, %b_y_28 : tensor<64x128xf32> loc(#loc73) + %b_dx_34 = "tt.reduce"(%b_dx_33) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_40: f32 loc(callsite(#loc1 at #loc74)), %b_dx_41: f32 loc(callsite(#loc1 at #loc74))): + %b_dx_42 = arith.addf %b_dx_40, %b_dx_41 : f32 loc(#loc81) + tt.reduce.return %b_dx_42 : f32 loc(#loc79) + }) : (tensor<64x128xf32>) -> tensor<64xf32> loc(#loc79) + %b_dx_35 = tt.expand_dims %b_dx_34 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc75) + %b_dx_36 = tt.broadcast %b_dx_35 : tensor<64x1xf32> -> tensor<64x128xf32> loc(#loc76) + %b_dx_37 = arith.mulf %b_dx_36, %b_y_28 : tensor<64x128xf32> loc(#loc76) + %b_dx_38 = arith.mulf %b_dx_37, %b_dx_31 : tensor<64x128xf32> loc(#loc77) + %b_dx_39 = arith.subf %b_dx_32, %b_dx_38 : tensor<64x128xf32> loc(#loc78) + %0 = arith.truncf %b_dx_39 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc37) + tt.store %p_dx_27, %0, %m_x_14 : tensor<64x128x!tt.ptr> loc(#loc38) + tt.return loc(#loc39) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("b_rstd"(#loc2)) +#loc47 = loc("m_x"(#loc3)) +#loc48 = loc("o_t"(#loc4)) +#loc49 = loc("i_t"(#loc5)) +#loc50 = loc("i_t"(#loc6)) +#loc51 = loc("o_t"(#loc7)) +#loc52 = loc("o_t"(#loc8)) +#loc53 = loc("o_d"(#loc9)) +#loc54 = loc("m_t"(#loc10)) +#loc55 = loc("m_x"(#loc11)) +#loc56 = loc("m_x"(#loc12)) +#loc57 = loc("m_x"(#loc13)) +#loc58 = loc("p_y"(#loc14)) +#loc59 = loc("p_y"(#loc15)) +#loc60 = loc("p_y"(#loc16)) +#loc61 = loc("p_y"(#loc17)) +#loc62 = loc("p_rstd"(#loc18)) +#loc63 = loc("p_dy"(#loc19)) +#loc64 = loc("p_dy"(#loc20)) +#loc65 = loc("p_dx"(#loc21)) +#loc66 = loc("p_dx"(#loc22)) +#loc67 = loc("b_y"(#loc23)) +#loc68 = loc("b_y"(#loc24)) +#loc69 = loc("b_dy"(#loc25)) +#loc70 = loc("b_dy"(#loc26)) +#loc71 = loc("b_dx"(#loc27)) +#loc72 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc29)) +#loc75 = loc("b_dx"(#loc33)) +#loc76 = loc("b_dx"(#loc34)) +#loc77 = loc("b_dx"(#loc35)) +#loc78 = loc("b_dx"(#loc36)) +#loc79 = loc(callsite(#loc30 at #loc74)) +#loc81 = loc(callsite(#loc32 at #loc79)) diff --git a/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/__grp__l2norm_bwd_kernel.json b/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/__grp__l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..a94649443a5b6dd847392abc68071609d404427f --- /dev/null +++ b/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/__grp__l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_bwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/l2norm_bwd_kernel.source", "l2norm_bwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/l2norm_bwd_kernel.ttir", "l2norm_bwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/l2norm_bwd_kernel.ttgir", "l2norm_bwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/l2norm_bwd_kernel.llir", "l2norm_bwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/l2norm_bwd_kernel.ptx", "l2norm_bwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/l2norm_bwd_kernel.cubin", "l2norm_bwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/l2norm_bwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/l2norm_bwd_kernel.cubin b/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/l2norm_bwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..11fbfa75b4c9e12d8aa2131cc1efaf8f6f7b2111 Binary files /dev/null and b/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/l2norm_bwd_kernel.cubin differ diff --git a/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/l2norm_bwd_kernel.json b/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..486ed77d8ccb49a32a20e3849d5517fdd6bef212 --- /dev/null +++ b/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"hash": "0f0e583220a55e8c73d14c9e83e108c97c88f95da6445b5e5dd5357bf9ad52a7", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 8, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 0, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_bwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/l2norm_bwd_kernel.llir b/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/l2norm_bwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..5aebfbe667385def730c32e647584fb126c08e0c --- /dev/null +++ b/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/l2norm_bwd_kernel.llir @@ -0,0 +1,175 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_bwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, ptr addrspace(1) %3, float %4, i32 %5, ptr addrspace(1) readnone captures(none) %6, ptr addrspace(1) readnone captures(none) %7) local_unnamed_addr #0 !dbg !4 { + %9 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !12 + %10 = zext nneg i32 %9 to i64, !dbg !13 + %11 = shl nuw nsw i64 %10, 3, !dbg !14 + %12 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !15 + %13 = lshr i32 %12, 5, !dbg !15 + %14 = and i32 %13, 7, !dbg !15 + %15 = zext nneg i32 %14 to i64, !dbg !16 + %16 = or disjoint i64 %11, %15, !dbg !16 + %17 = sext i32 %5 to i64, !dbg !17 + %18 = icmp slt i64 %16, %17, !dbg !17 + %19 = shl nuw nsw i32 %12, 2, !dbg !18 + %20 = and i32 %19, 124, !dbg !18 + %21 = shl nuw nsw i64 %16, 7, !dbg !19 + %22 = getelementptr bfloat, ptr addrspace(1) %0, i64 %21, !dbg !20 + %23 = zext nneg i32 %20 to i64, !dbg !21 + %24 = getelementptr bfloat, ptr addrspace(1) %22, i64 %23, !dbg !21 + %25 = getelementptr float, ptr addrspace(1) %1, i64 %16, !dbg !22 + %26 = getelementptr bfloat, ptr addrspace(1) %2, i64 %21, !dbg !23 + %27 = getelementptr bfloat, ptr addrspace(1) %26, i64 %23, !dbg !24 + %28 = getelementptr bfloat, ptr addrspace(1) %3, i64 %21, !dbg !25 + %29 = getelementptr bfloat, ptr addrspace(1) %28, i64 %23, !dbg !26 + %30 = tail call { i32, i32 } asm sideeffect "mov.u32 $0, $2;\0A\09mov.u32 $1, $3;\0A\09@$5 ld.global.v2.b32 { $0, $1 }, [ $4 + 0 ];", "=r,=r,r,r,l,b"(i32 0, i32 0, ptr addrspace(1) %24, i1 %18) #3, !dbg !27 + %31 = extractvalue { i32, i32 } %30, 0, !dbg !27 + %32 = bitcast i32 %31 to <2 x bfloat>, !dbg !27 + %33 = extractvalue { i32, i32 } %30, 1, !dbg !27 + %34 = bitcast i32 %33 to <2 x bfloat>, !dbg !27 + %35 = extractelement <2 x bfloat> %32, i64 0, !dbg !27 + %36 = extractelement <2 x bfloat> %32, i64 1, !dbg !27 + %37 = extractelement <2 x bfloat> %34, i64 0, !dbg !27 + %38 = extractelement <2 x bfloat> %34, i64 1, !dbg !27 + %39 = fpext bfloat %35 to float, !dbg !28 + %40 = fpext bfloat %36 to float, !dbg !28 + %41 = fpext bfloat %37 to float, !dbg !28 + %42 = fpext bfloat %38 to float, !dbg !28 + %43 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %25, i1 %18) #3, !dbg !29 + %44 = bitcast i32 %43 to float, !dbg !29 + %45 = tail call { i32, i32 } asm sideeffect "mov.u32 $0, $2;\0A\09mov.u32 $1, $3;\0A\09@$5 ld.global.v2.b32 { $0, $1 }, [ $4 + 0 ];", "=r,=r,r,r,l,b"(i32 0, i32 0, ptr addrspace(1) %27, i1 %18) #3, !dbg !30 + %46 = extractvalue { i32, i32 } %45, 0, !dbg !30 + %47 = bitcast i32 %46 to <2 x bfloat>, !dbg !30 + %48 = extractvalue { i32, i32 } %45, 1, !dbg !30 + %49 = bitcast i32 %48 to <2 x bfloat>, !dbg !30 + %50 = extractelement <2 x bfloat> %47, i64 0, !dbg !30 + %51 = extractelement <2 x bfloat> %47, i64 1, !dbg !30 + %52 = extractelement <2 x bfloat> %49, i64 0, !dbg !30 + %53 = extractelement <2 x bfloat> %49, i64 1, !dbg !30 + %54 = fpext bfloat %50 to float, !dbg !31 + %55 = fpext bfloat %51 to float, !dbg !31 + %56 = fpext bfloat %52 to float, !dbg !31 + %57 = fpext bfloat %53 to float, !dbg !31 + %58 = fmul float %44, %54, !dbg !32 + %59 = fmul float %44, %55, !dbg !32 + %60 = fmul float %44, %56, !dbg !32 + %61 = fmul float %44, %57, !dbg !32 + %62 = fmul float %39, %54, !dbg !33 + %63 = fmul float %40, %55, !dbg !33 + %64 = fmul float %41, %56, !dbg !33 + %65 = fmul float %42, %57, !dbg !33 + %66 = fadd float %62, %63, !dbg !34 + %67 = fadd float %64, %66, !dbg !34 + %68 = fadd float %65, %67, !dbg !34 + %69 = bitcast float %68 to i32, !dbg !37 + %70 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %69, i32 16, i32 31), !dbg !37 + %71 = bitcast i32 %70 to float, !dbg !37 + %72 = fadd float %68, %71, !dbg !34 + %73 = bitcast float %72 to i32, !dbg !37 + %74 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %73, i32 8, i32 31), !dbg !37 + %75 = bitcast i32 %74 to float, !dbg !37 + %76 = fadd float %72, %75, !dbg !34 + %77 = bitcast float %76 to i32, !dbg !37 + %78 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %77, i32 4, i32 31), !dbg !37 + %79 = bitcast i32 %78 to float, !dbg !37 + %80 = fadd float %76, %79, !dbg !34 + %81 = bitcast float %80 to i32, !dbg !37 + %82 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %81, i32 2, i32 31), !dbg !37 + %83 = bitcast i32 %82 to float, !dbg !37 + %84 = fadd float %80, %83, !dbg !34 + %85 = bitcast float %84 to i32, !dbg !37 + %86 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %85, i32 1, i32 31), !dbg !37 + %87 = bitcast i32 %86 to float, !dbg !37 + %88 = fadd float %84, %87, !dbg !34 + %89 = fmul float %88, %39, !dbg !40 + %90 = fmul float %88, %40, !dbg !40 + %91 = fmul float %88, %41, !dbg !40 + %92 = fmul float %88, %42, !dbg !40 + %93 = fmul float %89, %44, !dbg !41 + %94 = fmul float %90, %44, !dbg !41 + %95 = fmul float %91, %44, !dbg !41 + %96 = fmul float %92, %44, !dbg !41 + %97 = fsub float %58, %93, !dbg !42 + %98 = fsub float %59, %94, !dbg !42 + %99 = fsub float %60, %95, !dbg !42 + %100 = fsub float %61, %96, !dbg !42 + %101 = fptrunc float %97 to bfloat, !dbg !43 + %102 = fptrunc float %98 to bfloat, !dbg !43 + %103 = fptrunc float %99 to bfloat, !dbg !43 + %104 = fptrunc float %100 to bfloat, !dbg !43 + %105 = insertelement <2 x bfloat> poison, bfloat %101, i64 0, !dbg !44 + %106 = insertelement <2 x bfloat> %105, bfloat %102, i64 1, !dbg !44 + %107 = bitcast <2 x bfloat> %106 to i32, !dbg !44 + %108 = insertelement <2 x bfloat> poison, bfloat %103, i64 0, !dbg !44 + %109 = insertelement <2 x bfloat> %108, bfloat %104, i64 1, !dbg !44 + %110 = bitcast <2 x bfloat> %109 to i32, !dbg !44 + tail call void asm sideeffect "@$3 st.global.v2.b32 [ $2 + 0 ], { $0, $1 };", "r,r,l,b"(i32 %107, i32 %110, ptr addrspace(1) %29, i1 %18) #3, !dbg !44 + ret void, !dbg !45 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +attributes #0 = { nounwind "nvvm.reqntid"="256" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = distinct !DISubprogram(name: "l2norm_bwd_kernel", linkageName: "l2norm_bwd_kernel", scope: !1, file: !1, line: 93, type: !5, scopeLine: 93, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!5 = !DISubroutineType(cc: DW_CC_normal, types: !6) +!6 = !{null, !7, !9, !7, !7, !10, !11, !7, !7} +!7 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !8, size: 64, dwarfAddressSpace: 1) +!8 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!9 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !10, size: 64, dwarfAddressSpace: 1) +!10 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!11 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!12 = !DILocation(line: 94, column: 24, scope: !4) +!13 = !DILocation(line: 94, column: 30, scope: !4) +!14 = !DILocation(line: 95, column: 16, scope: !4) +!15 = !DILocation(line: 95, column: 34, scope: !4) +!16 = !DILocation(line: 95, column: 21, scope: !4) +!17 = !DILocation(line: 97, column: 16, scope: !4) +!18 = !DILocation(line: 98, column: 30, scope: !4) +!19 = !DILocation(line: 99, column: 29, scope: !4) +!20 = !DILocation(line: 99, column: 14, scope: !4) +!21 = !DILocation(line: 99, column: 33, scope: !4) +!22 = !DILocation(line: 100, column: 20, scope: !4) +!23 = !DILocation(line: 101, column: 16, scope: !4) +!24 = !DILocation(line: 101, column: 35, scope: !4) +!25 = !DILocation(line: 102, column: 16, scope: !4) +!26 = !DILocation(line: 102, column: 35, scope: !4) +!27 = !DILocation(line: 104, column: 18, scope: !4) +!28 = !DILocation(line: 104, column: 47, scope: !4) +!29 = !DILocation(line: 105, column: 21, scope: !4) +!30 = !DILocation(line: 106, column: 19, scope: !4) +!31 = !DILocation(line: 106, column: 49, scope: !4) +!32 = !DILocation(line: 107, column: 18, scope: !4) +!33 = !DILocation(line: 107, column: 50, scope: !4) +!34 = !DILocation(line: 263, column: 15, scope: !35, inlinedAt: !37) +!35 = distinct !DILexicalBlockFile(scope: !4, file: !36, discriminator: 0) +!36 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!37 = !DILocation(line: 293, column: 36, scope: !35, inlinedAt: !38) +!38 = !DILocation(line: 107, column: 55, scope: !39) +!39 = distinct !DILexicalBlockFile(scope: !4, file: !1, discriminator: 0) +!40 = !DILocation(line: 107, column: 69, scope: !4) +!41 = !DILocation(line: 107, column: 75, scope: !4) +!42 = !DILocation(line: 107, column: 36, scope: !4) +!43 = !DILocation(line: 108, column: 27, scope: !4) +!44 = !DILocation(line: 108, column: 19, scope: !4) +!45 = !DILocation(line: 108, column: 4, scope: !4) diff --git a/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/l2norm_bwd_kernel.ptx b/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/l2norm_bwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..fe7da55886e2fa5b73b208bfc4e2d72a2de54c52 --- /dev/null +++ b/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/l2norm_bwd_kernel.ptx @@ -0,0 +1,389 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_bwd_kernel // -- Begin function l2norm_bwd_kernel + // @l2norm_bwd_kernel +.visible .entry l2norm_bwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_2, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_3, + .param .f32 l2norm_bwd_kernel_param_4, + .param .u32 l2norm_bwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_6, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_7 +) +.reqntid 256 +{ + .reg .pred %p<2>; + .reg .b16 %rs<9>; + .reg .b32 %r<52>; + .reg .b64 %rd<19>; + .loc 1 93 0 // l2norm.py:93:0 +$L__func_begin0: + .loc 1 93 0 // l2norm.py:93:0 + +// %bb.0: + ld.param.b64 %rd5, [l2norm_bwd_kernel_param_0]; + ld.param.b64 %rd6, [l2norm_bwd_kernel_param_1]; +$L__tmp0: + .loc 1 94 24 // l2norm.py:94:24 + mov.u32 %r9, %ctaid.x; + ld.param.b64 %rd7, [l2norm_bwd_kernel_param_2]; + .loc 1 95 16 // l2norm.py:95:16 + mul.wide.u32 %rd8, %r9, 8; + ld.param.b64 %rd9, [l2norm_bwd_kernel_param_3]; + .loc 1 95 34 // l2norm.py:95:34 + mov.u32 %r10, %tid.x; + ld.param.s32 %rd10, [l2norm_bwd_kernel_param_5]; + bfe.u32 %r11, %r10, 5, 3; + .loc 1 95 21 // l2norm.py:95:21 + cvt.u64.u32 %rd11, %r11; + or.b64 %rd12, %rd8, %rd11; + .loc 1 97 16 // l2norm.py:97:16 + setp.lt.s64 %p1, %rd12, %rd10; + .loc 1 98 30 // l2norm.py:98:30 + shl.b32 %r12, %r10, 2; + and.b32 %r13, %r12, 124; + .loc 1 99 14 // l2norm.py:99:14 + shl.b64 %rd13, %rd12, 8; + add.s64 %rd14, %rd5, %rd13; + .loc 1 99 33 // l2norm.py:99:33 + mul.wide.u32 %rd15, %r13, 2; + add.s64 %rd1, %rd14, %rd15; + .loc 1 100 20 // l2norm.py:100:20 + shl.b64 %rd16, %rd12, 2; + add.s64 %rd2, %rd6, %rd16; + .loc 1 101 16 // l2norm.py:101:16 + add.s64 %rd17, %rd7, %rd13; + .loc 1 101 35 // l2norm.py:101:35 + add.s64 %rd3, %rd17, %rd15; + .loc 1 102 16 // l2norm.py:102:16 + add.s64 %rd18, %rd9, %rd13; + .loc 1 102 35 // l2norm.py:102:35 + add.s64 %rd4, %rd18, %rd15; + mov.b32 %r3, 0; + .loc 1 104 18 // l2norm.py:104:18 + // begin inline asm + mov.u32 %r1, %r3; + mov.u32 %r2, %r3; + @%p1 ld.global.v2.b32 { %r1, %r2 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + .loc 1 104 47 // l2norm.py:104:47 + cvt.f32.bf16 %r14, %rs1; + cvt.f32.bf16 %r15, %rs2; + cvt.f32.bf16 %r16, %rs3; + cvt.f32.bf16 %r17, %rs4; + .loc 1 105 21 // l2norm.py:105:21 + // begin inline asm + mov.u32 %r4, %r3; + @%p1 ld.global.b32 { %r4 }, [ %rd2 + 0 ]; + // end inline asm + .loc 1 106 19 // l2norm.py:106:19 + // begin inline asm + mov.u32 %r5, %r3; + mov.u32 %r6, %r3; + @%p1 ld.global.v2.b32 { %r5, %r6 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs5, %rs6}, %r5; + mov.b32 {%rs7, %rs8}, %r6; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r18, %rs5; + cvt.f32.bf16 %r19, %rs6; + cvt.f32.bf16 %r20, %rs7; + cvt.f32.bf16 %r21, %rs8; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r22, %r15, %r19; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r23, %rs1, %rs5, %r22; + fma.rn.f32.bf16 %r24, %rs3, %rs7, %r23; + fma.rn.f32.bf16 %r25, %rs4, %rs8, %r24; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r26, %r25, 16, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r27, %r25, %r26; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r28, %r27, 8, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r29, %r27, %r28; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r30, %r29, 4, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r31, %r29, %r30; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r32, %r31, 2, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r33, %r31, %r32; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r34, %r33, 1, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r35, %r33, %r34; +$L__tmp12: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r36, %r35, %r14; + mul.f32 %r37, %r35, %r15; + mul.f32 %r38, %r35, %r16; + mul.f32 %r39, %r35, %r17; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r40, %r36, %r4; + mul.f32 %r41, %r37, %r4; + mul.f32 %r42, %r38, %r4; + mul.f32 %r43, %r39, %r4; + .loc 1 107 36 // l2norm.py:107:36 + neg.f32 %r44, %r40; + fma.rn.f32 %r45, %r4, %r18, %r44; + neg.f32 %r46, %r41; + fma.rn.f32 %r47, %r4, %r19, %r46; + neg.f32 %r48, %r42; + fma.rn.f32 %r49, %r4, %r20, %r48; + neg.f32 %r50, %r43; + fma.rn.f32 %r51, %r4, %r21, %r50; + .loc 1 108 19 // l2norm.py:108:19 + cvt.rn.bf16x2.f32 %r7, %r47, %r45; + cvt.rn.bf16x2.f32 %r8, %r51, %r49; + // begin inline asm + @%p1 st.global.v2.b32 [ %rd4 + 0 ], { %r7, %r8 }; + // end inline asm + .loc 1 108 4 // l2norm.py:108:4 + ret; +$L__tmp13: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 98 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp12 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 107 // DW_AT_call_line +.b8 55 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp12 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/l2norm_bwd_kernel.source b/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/l2norm_bwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..c23a4689ce5f60989f46cd34fecfe31d002a62d8 --- /dev/null +++ b/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/l2norm_bwd_kernel.source @@ -0,0 +1,210 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc47 = loc(unknown) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc54 = loc("y"(#loc)) +#loc55 = loc("rstd"(#loc)) +#loc56 = loc("dy"(#loc)) +#loc57 = loc("dx"(#loc)) +#loc58 = loc("eps"(#loc)) +#loc59 = loc("T"(#loc)) +#loc101 = loc("input"(#loc45)) +#loc102 = loc("a"(#loc50)) +#loc103 = loc("b"(#loc50)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc60) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc61) + %o_t = arith.constant 8 : i32 loc(#loc62) + %o_t_1 = arith.constant 8 : i64 loc(#loc62) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc62) + %o_t_3 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32> loc(#loc63) + %o_t_4 = arith.extsi %o_t_3 : tensor<8xi32> to tensor<8xi64> loc(#loc64) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<8xi64> loc(#loc64) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<8xi64> loc(#loc64) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc65) + %m_t = arith.extsi %T : i32 to i64 loc(#loc66) + %m_t_7 = tt.splat %m_t : i64 -> tensor<8xi64> loc(#loc66) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<8xi64> loc(#loc66) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<8xi1> -> tensor<8x1xi1> loc(#loc67) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc68) + %m_x_10 = arith.constant 128 : i32 loc(#loc69) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc69) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc69) + %m_x_13 = tt.broadcast %m_x : tensor<8x1xi1> -> tensor<8x128xi1> loc(#loc70) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<8x128xi1> loc(#loc70) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<8x128xi1> loc(#loc70) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc71) + %p_y_16 = arith.constant 128 : i32 loc(#loc72) + %p_y_17 = arith.constant 128 : i64 loc(#loc72) + %p_y_18 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc72) + %p_y_19 = arith.muli %p_y, %p_y_18 : tensor<8x1xi64> loc(#loc72) + %p_y_20 = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc73) + %p_y_21 = tt.addptr %p_y_20, %p_y_19 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc73) + %p_y_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc74) + %p_y_23 = tt.broadcast %p_y_21 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc75) + %p_y_24 = tt.broadcast %p_y_22 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc75) + %p_y_25 = tt.addptr %p_y_23, %p_y_24 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc75) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr> loc(#loc76) + %p_rstd_26 = tt.addptr %p_rstd, %o_t_6 : tensor<8x!tt.ptr>, tensor<8xi64> loc(#loc76) + %p_dy = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc77) + %p_dy_27 = arith.constant 128 : i32 loc(#loc78) + %p_dy_28 = arith.constant 128 : i64 loc(#loc78) + %p_dy_29 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc78) + %p_dy_30 = arith.muli %p_dy, %p_dy_29 : tensor<8x1xi64> loc(#loc78) + %p_dy_31 = tt.splat %dy : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc79) + %p_dy_32 = tt.addptr %p_dy_31, %p_dy_30 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc79) + %p_dy_33 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc80) + %p_dy_34 = tt.broadcast %p_dy_32 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc81) + %p_dy_35 = tt.broadcast %p_dy_33 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc81) + %p_dy_36 = tt.addptr %p_dy_34, %p_dy_35 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc81) + %p_dx = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc82) + %p_dx_37 = arith.constant 128 : i32 loc(#loc83) + %p_dx_38 = arith.constant 128 : i64 loc(#loc83) + %p_dx_39 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc83) + %p_dx_40 = arith.muli %p_dx, %p_dx_39 : tensor<8x1xi64> loc(#loc83) + %p_dx_41 = tt.splat %dx : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc84) + %p_dx_42 = tt.addptr %p_dx_41, %p_dx_40 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc84) + %p_dx_43 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc85) + %p_dx_44 = tt.broadcast %p_dx_42 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc86) + %p_dx_45 = tt.broadcast %p_dx_43 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc86) + %p_dx_46 = tt.addptr %p_dx_44, %p_dx_45 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc86) + %b_y = arith.constant 0.000000e+00 : f32 loc(#loc87) + %b_y_47 = arith.constant dense<0.000000e+00> : tensor<8x128xf32> loc(#loc87) + %b_y_48 = arith.truncf %b_y_47 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc87) + %b_y_49 = tt.load %p_y_25, %m_x_15, %b_y_48 : tensor<8x128x!tt.ptr> loc(#loc87) + %b_y_50 = arith.extf %b_y_49 : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc88) + %b_rstd = arith.constant 0.000000e+00 : f32 loc(#loc89) + %b_rstd_51 = arith.constant dense<0.000000e+00> : tensor<8xf32> loc(#loc89) + %b_rstd_52 = tt.load %p_rstd_26, %m_t_8, %b_rstd_51 : tensor<8x!tt.ptr> loc(#loc89) + %b_dy = arith.constant 0.000000e+00 : f32 loc(#loc90) + %b_dy_53 = arith.constant dense<0.000000e+00> : tensor<8x128xf32> loc(#loc90) + %b_dy_54 = arith.truncf %b_dy_53 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc90) + %b_dy_55 = tt.load %p_dy_36, %m_x_15, %b_dy_54 : tensor<8x128x!tt.ptr> loc(#loc90) + %b_dy_56 = arith.extf %b_dy_55 : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc91) + %b_dx = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc92) + %b_dx_57 = tt.broadcast %b_dx : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc93) + %b_dx_58 = arith.mulf %b_dy_56, %b_dx_57 : tensor<8x128xf32> loc(#loc93) + %b_dx_59 = arith.mulf %b_dy_56, %b_y_50 : tensor<8x128xf32> loc(#loc94) + %b_dx_60 = tt.call @triton.language.standard.sum__fp32S8_128S_c1_cFalse_cNone(%b_dx_59) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc95) + %b_dx_61 = tt.expand_dims %b_dx_60 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc96) + %b_dx_62 = tt.broadcast %b_dx_61 : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc97) + %b_dx_63 = arith.mulf %b_dx_62, %b_y_50 : tensor<8x128xf32> loc(#loc97) + %b_dx_64 = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc98) + %b_dx_65 = tt.broadcast %b_dx_64 : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc99) + %b_dx_66 = arith.mulf %b_dx_63, %b_dx_65 : tensor<8x128xf32> loc(#loc99) + %b_dx_67 = arith.subf %b_dx_58, %b_dx_66 : tensor<8x128xf32> loc(#loc100) + %0 = arith.truncf %b_dx_67 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc42) + tt.store %p_dx_46, %0, %m_x_15 : tensor<8x128x!tt.ptr> loc(#loc43) + tt.return loc(#loc44) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S8_128S_c1_cFalse_cNone(%input: tensor<8x128xf32> loc("input"(#loc45))) -> tensor<8xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc46) + tt.reduce.return %2 : f32 loc(#loc46) + }) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc46) + tt.return %0 : tensor<8xf32> loc(#loc48) + ^bb1: // no predecessors + %1 = ub.poison : tensor<8xf32> loc(#loc49) + tt.return %1 : tensor<8xf32> loc(#loc49) + } loc(#loc45) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc50)), %b: f32 loc("b"(#loc50))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc51) + tt.return %0 : f32 loc(#loc52) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc53) + tt.return %1 : f32 loc(#loc53) + } loc(#loc50) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:31) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:39) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:20) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:31) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:39) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:82) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc60 = loc("i_t"(#loc1)) +#loc61 = loc("i_t"(#loc2)) +#loc62 = loc("o_t"(#loc3)) +#loc63 = loc("o_t"(#loc4)) +#loc64 = loc("o_t"(#loc5)) +#loc65 = loc("o_d"(#loc6)) +#loc66 = loc("m_t"(#loc7)) +#loc67 = loc("m_x"(#loc8)) +#loc68 = loc("m_x"(#loc9)) +#loc69 = loc("m_x"(#loc10)) +#loc70 = loc("m_x"(#loc11)) +#loc71 = loc("p_y"(#loc12)) +#loc72 = loc("p_y"(#loc13)) +#loc73 = loc("p_y"(#loc14)) +#loc74 = loc("p_y"(#loc15)) +#loc75 = loc("p_y"(#loc16)) +#loc76 = loc("p_rstd"(#loc17)) +#loc77 = loc("p_dy"(#loc18)) +#loc78 = loc("p_dy"(#loc19)) +#loc79 = loc("p_dy"(#loc20)) +#loc80 = loc("p_dy"(#loc21)) +#loc81 = loc("p_dy"(#loc22)) +#loc82 = loc("p_dx"(#loc23)) +#loc83 = loc("p_dx"(#loc24)) +#loc84 = loc("p_dx"(#loc25)) +#loc85 = loc("p_dx"(#loc26)) +#loc86 = loc("p_dx"(#loc27)) +#loc87 = loc("b_y"(#loc28)) +#loc88 = loc("b_y"(#loc29)) +#loc89 = loc("b_rstd"(#loc30)) +#loc90 = loc("b_dy"(#loc31)) +#loc91 = loc("b_dy"(#loc32)) +#loc92 = loc("b_dx"(#loc33)) +#loc93 = loc("b_dx"(#loc34)) +#loc94 = loc("b_dx"(#loc35)) +#loc95 = loc("b_dx"(#loc36)) +#loc96 = loc("b_dx"(#loc37)) +#loc97 = loc("b_dx"(#loc38)) +#loc98 = loc("b_dx"(#loc39)) +#loc99 = loc("b_dx"(#loc40)) +#loc100 = loc("b_dx"(#loc41)) diff --git a/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/l2norm_bwd_kernel.ttgir b/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/l2norm_bwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..726600a88e4b511224d1607f4247a538bb7712ec --- /dev/null +++ b/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/l2norm_bwd_kernel.ttgir @@ -0,0 +1,146 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 4], threadsPerWarp = [1, 32], warpsPerCTA = [8, 1], order = [1, 0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc39 = loc("y"(#loc)) +#loc40 = loc("rstd"(#loc)) +#loc41 = loc("dy"(#loc)) +#loc42 = loc("dx"(#loc)) +#loc43 = loc("eps"(#loc)) +#loc44 = loc("T"(#loc)) +#loc72 = loc("b_dx"(#loc30)) +#loc78 = loc(callsite(#loc1 at #loc72)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 8 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %c8_i64 = arith.constant 8 : i64 loc(#loc1) + %cst = arith.constant dense<0.000000e+00> : tensor<8x128xbf16, #blocked> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<8x1xi64, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc45) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc46) + %o_t = arith.muli %i_t_3, %c8_i64 : i64 loc(#loc47) + %o_t_4 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc48) + %o_t_5 = arith.extsi %o_t_4 : tensor<8xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %o_t_6 = tt.splat %o_t : i64 -> tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %o_t_7 = arith.addi %o_t_6, %o_t_5 : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %m_t = arith.extsi %T : i32 to i64 loc(#loc50) + %m_t_8 = tt.splat %m_t : i64 -> tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc50) + %m_t_9 = arith.cmpi slt, %o_t_7, %m_t_8 : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc50) + %m_x = tt.expand_dims %m_t_9 {axis = 1 : i32} : tensor<8xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xi1, #blocked> loc(#loc51) + %m_x_10 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc52) + %m_x_11 = tt.expand_dims %m_x_10 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc52) + %m_x_12 = arith.cmpi slt, %m_x_11, %cst_2 : tensor<1x128xi32, #blocked> loc(#loc53) + %m_x_13 = tt.broadcast %m_x : tensor<8x1xi1, #blocked> -> tensor<8x128xi1, #blocked> loc(#loc54) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1, #blocked> -> tensor<8x128xi1, #blocked> loc(#loc54) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<8x128xi1, #blocked> loc(#loc54) + %p_y = tt.expand_dims %o_t_7 {axis = 1 : i32} : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xi64, #blocked> loc(#loc55) + %p_y_16 = arith.muli %p_y, %cst_1 : tensor<8x1xi64, #blocked> loc(#loc56) + %p_y_17 = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr, #blocked> loc(#loc57) + %p_y_18 = tt.addptr %p_y_17, %p_y_16 : tensor<8x1x!tt.ptr, #blocked>, tensor<8x1xi64, #blocked> loc(#loc57) + %p_y_19 = tt.broadcast %p_y_18 : tensor<8x1x!tt.ptr, #blocked> -> tensor<8x128x!tt.ptr, #blocked> loc(#loc58) + %p_y_20 = tt.broadcast %m_x_11 : tensor<1x128xi32, #blocked> -> tensor<8x128xi32, #blocked> loc(#loc58) + %p_y_21 = tt.addptr %p_y_19, %p_y_20 : tensor<8x128x!tt.ptr, #blocked>, tensor<8x128xi32, #blocked> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc59) + %p_rstd_22 = tt.addptr %p_rstd, %o_t_7 : tensor<8x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>>, tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc59) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<8x1x!tt.ptr, #blocked> loc(#loc60) + %p_dy_23 = tt.addptr %p_dy, %p_y_16 : tensor<8x1x!tt.ptr, #blocked>, tensor<8x1xi64, #blocked> loc(#loc60) + %p_dy_24 = tt.broadcast %p_dy_23 : tensor<8x1x!tt.ptr, #blocked> -> tensor<8x128x!tt.ptr, #blocked> loc(#loc61) + %p_dy_25 = tt.addptr %p_dy_24, %p_y_20 : tensor<8x128x!tt.ptr, #blocked>, tensor<8x128xi32, #blocked> loc(#loc61) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<8x1x!tt.ptr, #blocked> loc(#loc62) + %p_dx_26 = tt.addptr %p_dx, %p_y_16 : tensor<8x1x!tt.ptr, #blocked>, tensor<8x1xi64, #blocked> loc(#loc62) + %p_dx_27 = tt.broadcast %p_dx_26 : tensor<8x1x!tt.ptr, #blocked> -> tensor<8x128x!tt.ptr, #blocked> loc(#loc63) + %p_dx_28 = tt.addptr %p_dx_27, %p_y_20 : tensor<8x128x!tt.ptr, #blocked>, tensor<8x128xi32, #blocked> loc(#loc63) + %b_y = tt.load %p_y_21, %m_x_15, %cst : tensor<8x128x!tt.ptr, #blocked> loc(#loc64) + %b_y_29 = arith.extf %b_y : tensor<8x128xbf16, #blocked> to tensor<8x128xf32, #blocked> loc(#loc65) + %b_rstd = tt.load %p_rstd_22, %m_t_9, %cst_0 : tensor<8x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc66) + %b_dy = tt.load %p_dy_25, %m_x_15, %cst : tensor<8x128x!tt.ptr, #blocked> loc(#loc67) + %b_dy_30 = arith.extf %b_dy : tensor<8x128xbf16, #blocked> to tensor<8x128xf32, #blocked> loc(#loc68) + %b_dx = tt.expand_dims %b_rstd {axis = 1 : i32} : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xf32, #blocked> loc(#loc69) + %b_dx_31 = tt.broadcast %b_dx : tensor<8x1xf32, #blocked> -> tensor<8x128xf32, #blocked> loc(#loc70) + %b_dx_32 = arith.mulf %b_dy_30, %b_dx_31 : tensor<8x128xf32, #blocked> loc(#loc70) + %b_dx_33 = arith.mulf %b_dy_30, %b_y_29 : tensor<8x128xf32, #blocked> loc(#loc71) + %b_dx_34 = "tt.reduce"(%b_dx_33) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_40: f32 loc(callsite(#loc1 at #loc72)), %b_dx_41: f32 loc(callsite(#loc1 at #loc72))): + %b_dx_42 = arith.addf %b_dx_40, %b_dx_41 : f32 loc(#loc79) + tt.reduce.return %b_dx_42 : f32 loc(#loc77) + }) : (tensor<8x128xf32, #blocked>) -> tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc77) + %b_dx_35 = tt.expand_dims %b_dx_34 {axis = 1 : i32} : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xf32, #blocked> loc(#loc73) + %b_dx_36 = tt.broadcast %b_dx_35 : tensor<8x1xf32, #blocked> -> tensor<8x128xf32, #blocked> loc(#loc74) + %b_dx_37 = arith.mulf %b_dx_36, %b_y_29 : tensor<8x128xf32, #blocked> loc(#loc74) + %b_dx_38 = arith.mulf %b_dx_37, %b_dx_31 : tensor<8x128xf32, #blocked> loc(#loc75) + %b_dx_39 = arith.subf %b_dx_32, %b_dx_38 : tensor<8x128xf32, #blocked> loc(#loc76) + %0 = arith.truncf %b_dx_39 : tensor<8x128xf32, #blocked> to tensor<8x128xbf16, #blocked> loc(#loc36) + tt.store %p_dx_28, %0, %m_x_15 : tensor<8x128x!tt.ptr, #blocked> loc(#loc37) + tt.return loc(#loc38) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc45 = loc("i_t"(#loc2)) +#loc46 = loc("i_t"(#loc3)) +#loc47 = loc("o_t"(#loc4)) +#loc48 = loc("o_t"(#loc5)) +#loc49 = loc("o_t"(#loc6)) +#loc50 = loc("m_t"(#loc7)) +#loc51 = loc("m_x"(#loc8)) +#loc52 = loc("m_x"(#loc9)) +#loc53 = loc("m_x"(#loc10)) +#loc54 = loc("m_x"(#loc11)) +#loc55 = loc("p_y"(#loc12)) +#loc56 = loc("p_y"(#loc13)) +#loc57 = loc("p_y"(#loc14)) +#loc58 = loc("p_y"(#loc15)) +#loc59 = loc("p_rstd"(#loc16)) +#loc60 = loc("p_dy"(#loc17)) +#loc61 = loc("p_dy"(#loc18)) +#loc62 = loc("p_dx"(#loc19)) +#loc63 = loc("p_dx"(#loc20)) +#loc64 = loc("b_y"(#loc21)) +#loc65 = loc("b_y"(#loc22)) +#loc66 = loc("b_rstd"(#loc23)) +#loc67 = loc("b_dy"(#loc24)) +#loc68 = loc("b_dy"(#loc25)) +#loc69 = loc("b_dx"(#loc26)) +#loc70 = loc("b_dx"(#loc27)) +#loc71 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc32)) +#loc74 = loc("b_dx"(#loc33)) +#loc75 = loc("b_dx"(#loc34)) +#loc76 = loc("b_dx"(#loc35)) +#loc77 = loc(callsite(#loc29 at #loc72)) +#loc79 = loc(callsite(#loc31 at #loc77)) diff --git a/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/l2norm_bwd_kernel.ttir b/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/l2norm_bwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..c4761315a53ecb33821860b30f93bdae9fbfe731 --- /dev/null +++ b/triton/gpu0/B4HFQMRAUVPIY46RJSPIHYIIZF6IR6K5UZCFWXS52U2XX6NNKKTQ/l2norm_bwd_kernel.ttir @@ -0,0 +1,147 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc40 = loc("y"(#loc)) +#loc41 = loc("rstd"(#loc)) +#loc42 = loc("dy"(#loc)) +#loc43 = loc("dx"(#loc)) +#loc44 = loc("eps"(#loc)) +#loc45 = loc("T"(#loc)) +#loc74 = loc("b_dx"(#loc31)) +#loc80 = loc(callsite(#loc1 at #loc74)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<0.000000e+00> : tensor<8x128xbf16> loc(#loc1) + %b_rstd = arith.constant dense<0.000000e+00> : tensor<8xf32> loc(#loc46) + %cst_0 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc1) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc47) + %o_t = arith.constant 8 : i64 loc(#loc48) + %i_t = tt.get_program_id x : i32 loc(#loc49) + %i_t_1 = arith.extsi %i_t : i32 to i64 loc(#loc50) + %o_t_2 = arith.muli %i_t_1, %o_t : i64 loc(#loc48) + %o_t_3 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32> loc(#loc51) + %o_t_4 = arith.extsi %o_t_3 : tensor<8xi32> to tensor<8xi64> loc(#loc52) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<8xi64> loc(#loc52) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<8xi64> loc(#loc52) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc53) + %m_t = arith.extsi %T : i32 to i64 loc(#loc54) + %m_t_7 = tt.splat %m_t : i64 -> tensor<8xi64> loc(#loc54) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<8xi64> loc(#loc54) + %m_x_9 = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<8xi1> -> tensor<8x1xi1> loc(#loc55) + %m_x_10 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc56) + %m_x_11 = arith.cmpi slt, %m_x_10, %m_x : tensor<1x128xi32> loc(#loc47) + %m_x_12 = tt.broadcast %m_x_9 : tensor<8x1xi1> -> tensor<8x128xi1> loc(#loc57) + %m_x_13 = tt.broadcast %m_x_11 : tensor<1x128xi1> -> tensor<8x128xi1> loc(#loc57) + %m_x_14 = arith.andi %m_x_12, %m_x_13 : tensor<8x128xi1> loc(#loc57) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc58) + %p_y_15 = arith.muli %p_y, %cst_0 : tensor<8x1xi64> loc(#loc59) + %p_y_16 = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc60) + %p_y_17 = tt.addptr %p_y_16, %p_y_15 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc60) + %p_y_18 = tt.broadcast %p_y_17 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc61) + %p_y_19 = tt.broadcast %m_x_10 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc61) + %p_y_20 = tt.addptr %p_y_18, %p_y_19 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc61) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr> loc(#loc62) + %p_rstd_21 = tt.addptr %p_rstd, %o_t_6 : tensor<8x!tt.ptr>, tensor<8xi64> loc(#loc62) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc63) + %p_dy_22 = tt.addptr %p_dy, %p_y_15 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc63) + %p_dy_23 = tt.broadcast %p_dy_22 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc64) + %p_dy_24 = tt.addptr %p_dy_23, %p_y_19 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc64) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc65) + %p_dx_25 = tt.addptr %p_dx, %p_y_15 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc65) + %p_dx_26 = tt.broadcast %p_dx_25 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc66) + %p_dx_27 = tt.addptr %p_dx_26, %p_y_19 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc66) + %b_y = tt.load %p_y_20, %m_x_14, %cst : tensor<8x128x!tt.ptr> loc(#loc67) + %b_y_28 = arith.extf %b_y : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc68) + %b_rstd_29 = tt.load %p_rstd_21, %m_t_8, %b_rstd : tensor<8x!tt.ptr> loc(#loc46) + %b_dy = tt.load %p_dy_24, %m_x_14, %cst : tensor<8x128x!tt.ptr> loc(#loc69) + %b_dy_30 = arith.extf %b_dy : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc70) + %b_dx = tt.expand_dims %b_rstd_29 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc71) + %b_dx_31 = tt.broadcast %b_dx : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc72) + %b_dx_32 = arith.mulf %b_dy_30, %b_dx_31 : tensor<8x128xf32> loc(#loc72) + %b_dx_33 = arith.mulf %b_dy_30, %b_y_28 : tensor<8x128xf32> loc(#loc73) + %b_dx_34 = "tt.reduce"(%b_dx_33) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_40: f32 loc(callsite(#loc1 at #loc74)), %b_dx_41: f32 loc(callsite(#loc1 at #loc74))): + %b_dx_42 = arith.addf %b_dx_40, %b_dx_41 : f32 loc(#loc81) + tt.reduce.return %b_dx_42 : f32 loc(#loc79) + }) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc79) + %b_dx_35 = tt.expand_dims %b_dx_34 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc75) + %b_dx_36 = tt.broadcast %b_dx_35 : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc76) + %b_dx_37 = arith.mulf %b_dx_36, %b_y_28 : tensor<8x128xf32> loc(#loc76) + %b_dx_38 = arith.mulf %b_dx_37, %b_dx_31 : tensor<8x128xf32> loc(#loc77) + %b_dx_39 = arith.subf %b_dx_32, %b_dx_38 : tensor<8x128xf32> loc(#loc78) + %0 = arith.truncf %b_dx_39 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc37) + tt.store %p_dx_27, %0, %m_x_14 : tensor<8x128x!tt.ptr> loc(#loc38) + tt.return loc(#loc39) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("b_rstd"(#loc2)) +#loc47 = loc("m_x"(#loc3)) +#loc48 = loc("o_t"(#loc4)) +#loc49 = loc("i_t"(#loc5)) +#loc50 = loc("i_t"(#loc6)) +#loc51 = loc("o_t"(#loc7)) +#loc52 = loc("o_t"(#loc8)) +#loc53 = loc("o_d"(#loc9)) +#loc54 = loc("m_t"(#loc10)) +#loc55 = loc("m_x"(#loc11)) +#loc56 = loc("m_x"(#loc12)) +#loc57 = loc("m_x"(#loc13)) +#loc58 = loc("p_y"(#loc14)) +#loc59 = loc("p_y"(#loc15)) +#loc60 = loc("p_y"(#loc16)) +#loc61 = loc("p_y"(#loc17)) +#loc62 = loc("p_rstd"(#loc18)) +#loc63 = loc("p_dy"(#loc19)) +#loc64 = loc("p_dy"(#loc20)) +#loc65 = loc("p_dx"(#loc21)) +#loc66 = loc("p_dx"(#loc22)) +#loc67 = loc("b_y"(#loc23)) +#loc68 = loc("b_y"(#loc24)) +#loc69 = loc("b_dy"(#loc25)) +#loc70 = loc("b_dy"(#loc26)) +#loc71 = loc("b_dx"(#loc27)) +#loc72 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc29)) +#loc75 = loc("b_dx"(#loc33)) +#loc76 = loc("b_dx"(#loc34)) +#loc77 = loc("b_dx"(#loc35)) +#loc78 = loc("b_dx"(#loc36)) +#loc79 = loc(callsite(#loc30 at #loc74)) +#loc81 = loc(callsite(#loc32 at #loc79)) diff --git a/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/__grp__l2norm_bwd_kernel.json b/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/__grp__l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..50d8a35300fa9fbe7908e87f1fc3e675819a94cb --- /dev/null +++ b/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/__grp__l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_bwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/l2norm_bwd_kernel.source", "l2norm_bwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/l2norm_bwd_kernel.ttir", "l2norm_bwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/l2norm_bwd_kernel.ttgir", "l2norm_bwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/l2norm_bwd_kernel.llir", "l2norm_bwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/l2norm_bwd_kernel.ptx", "l2norm_bwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/l2norm_bwd_kernel.cubin", "l2norm_bwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/l2norm_bwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/l2norm_bwd_kernel.cubin b/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/l2norm_bwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..1d8045f97a615914718b85f1ec37b0ada37a2ec7 Binary files /dev/null and b/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/l2norm_bwd_kernel.cubin differ diff --git a/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/l2norm_bwd_kernel.json b/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..1f5037440912bba158c1c743494b84bb576ac47e --- /dev/null +++ b/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"hash": "0fbe6f24d681fbaa2c475dc543dd059d33c38e1be7144ef1a5fce51cb4a03ca7", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 8, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 0, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_bwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/l2norm_bwd_kernel.llir b/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/l2norm_bwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..d003b5156bb8bcd488a4bb79af5c82cf4f2070c7 --- /dev/null +++ b/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/l2norm_bwd_kernel.llir @@ -0,0 +1,229 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_bwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, ptr addrspace(1) %3, float %4, i32 %5, ptr addrspace(1) readnone captures(none) %6, ptr addrspace(1) readnone captures(none) %7) local_unnamed_addr #0 !dbg !4 { + %9 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !12 + %10 = zext nneg i32 %9 to i64, !dbg !13 + %11 = shl nuw nsw i64 %10, 4, !dbg !14 + %12 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !15 + %13 = lshr i32 %12, 4, !dbg !15 + %14 = and i32 %13, 15, !dbg !15 + %15 = zext nneg i32 %14 to i64, !dbg !16 + %16 = or disjoint i64 %11, %15, !dbg !16 + %17 = sext i32 %5 to i64, !dbg !17 + %18 = icmp slt i64 %16, %17, !dbg !17 + %19 = shl nuw nsw i32 %12, 3, !dbg !18 + %20 = and i32 %19, 120, !dbg !18 + %21 = shl nuw nsw i64 %16, 7, !dbg !19 + %22 = getelementptr bfloat, ptr addrspace(1) %0, i64 %21, !dbg !20 + %23 = zext nneg i32 %20 to i64, !dbg !21 + %24 = getelementptr bfloat, ptr addrspace(1) %22, i64 %23, !dbg !21 + %25 = getelementptr float, ptr addrspace(1) %1, i64 %16, !dbg !22 + %26 = getelementptr bfloat, ptr addrspace(1) %2, i64 %21, !dbg !23 + %27 = getelementptr bfloat, ptr addrspace(1) %26, i64 %23, !dbg !24 + %28 = getelementptr bfloat, ptr addrspace(1) %3, i64 %21, !dbg !25 + %29 = getelementptr bfloat, ptr addrspace(1) %28, i64 %23, !dbg !26 + %30 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %24, i1 %18) #3, !dbg !27 + %31 = extractvalue { i32, i32, i32, i32 } %30, 0, !dbg !27 + %32 = bitcast i32 %31 to <2 x bfloat>, !dbg !27 + %33 = extractvalue { i32, i32, i32, i32 } %30, 1, !dbg !27 + %34 = bitcast i32 %33 to <2 x bfloat>, !dbg !27 + %35 = extractvalue { i32, i32, i32, i32 } %30, 2, !dbg !27 + %36 = bitcast i32 %35 to <2 x bfloat>, !dbg !27 + %37 = extractvalue { i32, i32, i32, i32 } %30, 3, !dbg !27 + %38 = bitcast i32 %37 to <2 x bfloat>, !dbg !27 + %39 = extractelement <2 x bfloat> %32, i64 0, !dbg !27 + %40 = extractelement <2 x bfloat> %32, i64 1, !dbg !27 + %41 = extractelement <2 x bfloat> %34, i64 0, !dbg !27 + %42 = extractelement <2 x bfloat> %34, i64 1, !dbg !27 + %43 = extractelement <2 x bfloat> %36, i64 0, !dbg !27 + %44 = extractelement <2 x bfloat> %36, i64 1, !dbg !27 + %45 = extractelement <2 x bfloat> %38, i64 0, !dbg !27 + %46 = extractelement <2 x bfloat> %38, i64 1, !dbg !27 + %47 = fpext bfloat %39 to float, !dbg !28 + %48 = fpext bfloat %40 to float, !dbg !28 + %49 = fpext bfloat %41 to float, !dbg !28 + %50 = fpext bfloat %42 to float, !dbg !28 + %51 = fpext bfloat %43 to float, !dbg !28 + %52 = fpext bfloat %44 to float, !dbg !28 + %53 = fpext bfloat %45 to float, !dbg !28 + %54 = fpext bfloat %46 to float, !dbg !28 + %55 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %25, i1 %18) #3, !dbg !29 + %56 = bitcast i32 %55 to float, !dbg !29 + %57 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %27, i1 %18) #3, !dbg !30 + %58 = extractvalue { i32, i32, i32, i32 } %57, 0, !dbg !30 + %59 = bitcast i32 %58 to <2 x bfloat>, !dbg !30 + %60 = extractvalue { i32, i32, i32, i32 } %57, 1, !dbg !30 + %61 = bitcast i32 %60 to <2 x bfloat>, !dbg !30 + %62 = extractvalue { i32, i32, i32, i32 } %57, 2, !dbg !30 + %63 = bitcast i32 %62 to <2 x bfloat>, !dbg !30 + %64 = extractvalue { i32, i32, i32, i32 } %57, 3, !dbg !30 + %65 = bitcast i32 %64 to <2 x bfloat>, !dbg !30 + %66 = extractelement <2 x bfloat> %59, i64 0, !dbg !30 + %67 = extractelement <2 x bfloat> %59, i64 1, !dbg !30 + %68 = extractelement <2 x bfloat> %61, i64 0, !dbg !30 + %69 = extractelement <2 x bfloat> %61, i64 1, !dbg !30 + %70 = extractelement <2 x bfloat> %63, i64 0, !dbg !30 + %71 = extractelement <2 x bfloat> %63, i64 1, !dbg !30 + %72 = extractelement <2 x bfloat> %65, i64 0, !dbg !30 + %73 = extractelement <2 x bfloat> %65, i64 1, !dbg !30 + %74 = fpext bfloat %66 to float, !dbg !31 + %75 = fpext bfloat %67 to float, !dbg !31 + %76 = fpext bfloat %68 to float, !dbg !31 + %77 = fpext bfloat %69 to float, !dbg !31 + %78 = fpext bfloat %70 to float, !dbg !31 + %79 = fpext bfloat %71 to float, !dbg !31 + %80 = fpext bfloat %72 to float, !dbg !31 + %81 = fpext bfloat %73 to float, !dbg !31 + %82 = fmul float %56, %74, !dbg !32 + %83 = fmul float %56, %75, !dbg !32 + %84 = fmul float %56, %76, !dbg !32 + %85 = fmul float %56, %77, !dbg !32 + %86 = fmul float %56, %78, !dbg !32 + %87 = fmul float %56, %79, !dbg !32 + %88 = fmul float %56, %80, !dbg !32 + %89 = fmul float %56, %81, !dbg !32 + %90 = fmul float %47, %74, !dbg !33 + %91 = fmul float %48, %75, !dbg !33 + %92 = fmul float %49, %76, !dbg !33 + %93 = fmul float %50, %77, !dbg !33 + %94 = fmul float %51, %78, !dbg !33 + %95 = fmul float %52, %79, !dbg !33 + %96 = fmul float %53, %80, !dbg !33 + %97 = fmul float %54, %81, !dbg !33 + %98 = fadd float %90, %91, !dbg !34 + %99 = fadd float %92, %98, !dbg !34 + %100 = fadd float %93, %99, !dbg !34 + %101 = fadd float %94, %100, !dbg !34 + %102 = fadd float %95, %101, !dbg !34 + %103 = fadd float %96, %102, !dbg !34 + %104 = fadd float %97, %103, !dbg !34 + %105 = bitcast float %104 to i32, !dbg !37 + %106 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %105, i32 8, i32 31), !dbg !37 + %107 = bitcast i32 %106 to float, !dbg !37 + %108 = fadd float %104, %107, !dbg !34 + %109 = bitcast float %108 to i32, !dbg !37 + %110 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %109, i32 4, i32 31), !dbg !37 + %111 = bitcast i32 %110 to float, !dbg !37 + %112 = fadd float %108, %111, !dbg !34 + %113 = bitcast float %112 to i32, !dbg !37 + %114 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %113, i32 2, i32 31), !dbg !37 + %115 = bitcast i32 %114 to float, !dbg !37 + %116 = fadd float %112, %115, !dbg !34 + %117 = bitcast float %116 to i32, !dbg !37 + %118 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %117, i32 1, i32 31), !dbg !37 + %119 = bitcast i32 %118 to float, !dbg !37 + %120 = fadd float %116, %119, !dbg !34 + %121 = fmul float %120, %47, !dbg !40 + %122 = fmul float %120, %48, !dbg !40 + %123 = fmul float %120, %49, !dbg !40 + %124 = fmul float %120, %50, !dbg !40 + %125 = fmul float %120, %51, !dbg !40 + %126 = fmul float %120, %52, !dbg !40 + %127 = fmul float %120, %53, !dbg !40 + %128 = fmul float %120, %54, !dbg !40 + %129 = fmul float %121, %56, !dbg !41 + %130 = fmul float %122, %56, !dbg !41 + %131 = fmul float %123, %56, !dbg !41 + %132 = fmul float %124, %56, !dbg !41 + %133 = fmul float %125, %56, !dbg !41 + %134 = fmul float %126, %56, !dbg !41 + %135 = fmul float %127, %56, !dbg !41 + %136 = fmul float %128, %56, !dbg !41 + %137 = fsub float %82, %129, !dbg !42 + %138 = fsub float %83, %130, !dbg !42 + %139 = fsub float %84, %131, !dbg !42 + %140 = fsub float %85, %132, !dbg !42 + %141 = fsub float %86, %133, !dbg !42 + %142 = fsub float %87, %134, !dbg !42 + %143 = fsub float %88, %135, !dbg !42 + %144 = fsub float %89, %136, !dbg !42 + %145 = fptrunc float %137 to bfloat, !dbg !43 + %146 = fptrunc float %138 to bfloat, !dbg !43 + %147 = fptrunc float %139 to bfloat, !dbg !43 + %148 = fptrunc float %140 to bfloat, !dbg !43 + %149 = fptrunc float %141 to bfloat, !dbg !43 + %150 = fptrunc float %142 to bfloat, !dbg !43 + %151 = fptrunc float %143 to bfloat, !dbg !43 + %152 = fptrunc float %144 to bfloat, !dbg !43 + %153 = insertelement <2 x bfloat> poison, bfloat %145, i64 0, !dbg !44 + %154 = insertelement <2 x bfloat> %153, bfloat %146, i64 1, !dbg !44 + %155 = bitcast <2 x bfloat> %154 to i32, !dbg !44 + %156 = insertelement <2 x bfloat> poison, bfloat %147, i64 0, !dbg !44 + %157 = insertelement <2 x bfloat> %156, bfloat %148, i64 1, !dbg !44 + %158 = bitcast <2 x bfloat> %157 to i32, !dbg !44 + %159 = insertelement <2 x bfloat> poison, bfloat %149, i64 0, !dbg !44 + %160 = insertelement <2 x bfloat> %159, bfloat %150, i64 1, !dbg !44 + %161 = bitcast <2 x bfloat> %160 to i32, !dbg !44 + %162 = insertelement <2 x bfloat> poison, bfloat %151, i64 0, !dbg !44 + %163 = insertelement <2 x bfloat> %162, bfloat %152, i64 1, !dbg !44 + %164 = bitcast <2 x bfloat> %163 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %155, i32 %158, i32 %161, i32 %164, ptr addrspace(1) %29, i1 %18) #3, !dbg !44 + ret void, !dbg !45 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +attributes #0 = { nounwind "nvvm.reqntid"="256" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = distinct !DISubprogram(name: "l2norm_bwd_kernel", linkageName: "l2norm_bwd_kernel", scope: !1, file: !1, line: 93, type: !5, scopeLine: 93, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!5 = !DISubroutineType(cc: DW_CC_normal, types: !6) +!6 = !{null, !7, !9, !7, !7, !10, !11, !7, !7} +!7 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !8, size: 64, dwarfAddressSpace: 1) +!8 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!9 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !10, size: 64, dwarfAddressSpace: 1) +!10 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!11 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!12 = !DILocation(line: 94, column: 24, scope: !4) +!13 = !DILocation(line: 94, column: 30, scope: !4) +!14 = !DILocation(line: 95, column: 16, scope: !4) +!15 = !DILocation(line: 95, column: 34, scope: !4) +!16 = !DILocation(line: 95, column: 21, scope: !4) +!17 = !DILocation(line: 97, column: 16, scope: !4) +!18 = !DILocation(line: 98, column: 30, scope: !4) +!19 = !DILocation(line: 99, column: 29, scope: !4) +!20 = !DILocation(line: 99, column: 14, scope: !4) +!21 = !DILocation(line: 99, column: 33, scope: !4) +!22 = !DILocation(line: 100, column: 20, scope: !4) +!23 = !DILocation(line: 101, column: 16, scope: !4) +!24 = !DILocation(line: 101, column: 35, scope: !4) +!25 = !DILocation(line: 102, column: 16, scope: !4) +!26 = !DILocation(line: 102, column: 35, scope: !4) +!27 = !DILocation(line: 104, column: 18, scope: !4) +!28 = !DILocation(line: 104, column: 47, scope: !4) +!29 = !DILocation(line: 105, column: 21, scope: !4) +!30 = !DILocation(line: 106, column: 19, scope: !4) +!31 = !DILocation(line: 106, column: 49, scope: !4) +!32 = !DILocation(line: 107, column: 18, scope: !4) +!33 = !DILocation(line: 107, column: 50, scope: !4) +!34 = !DILocation(line: 263, column: 15, scope: !35, inlinedAt: !37) +!35 = distinct !DILexicalBlockFile(scope: !4, file: !36, discriminator: 0) +!36 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!37 = !DILocation(line: 293, column: 36, scope: !35, inlinedAt: !38) +!38 = !DILocation(line: 107, column: 55, scope: !39) +!39 = distinct !DILexicalBlockFile(scope: !4, file: !1, discriminator: 0) +!40 = !DILocation(line: 107, column: 69, scope: !4) +!41 = !DILocation(line: 107, column: 75, scope: !4) +!42 = !DILocation(line: 107, column: 36, scope: !4) +!43 = !DILocation(line: 108, column: 27, scope: !4) +!44 = !DILocation(line: 108, column: 19, scope: !4) +!45 = !DILocation(line: 108, column: 4, scope: !4) diff --git a/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/l2norm_bwd_kernel.ptx b/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/l2norm_bwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..cfb4de60dee3c016366fd17ff2080014a2ca426a --- /dev/null +++ b/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/l2norm_bwd_kernel.ptx @@ -0,0 +1,421 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_bwd_kernel // -- Begin function l2norm_bwd_kernel + // @l2norm_bwd_kernel +.visible .entry l2norm_bwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_2, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_3, + .param .f32 l2norm_bwd_kernel_param_4, + .param .u32 l2norm_bwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_6, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_7 +) +.reqntid 256 +{ + .reg .pred %p<2>; + .reg .b16 %rs<17>; + .reg .b32 %r<84>; + .reg .b64 %rd<19>; + .loc 1 93 0 // l2norm.py:93:0 +$L__func_begin0: + .loc 1 93 0 // l2norm.py:93:0 + +// %bb.0: + ld.param.b64 %rd5, [l2norm_bwd_kernel_param_0]; + ld.param.b64 %rd6, [l2norm_bwd_kernel_param_1]; +$L__tmp0: + .loc 1 94 24 // l2norm.py:94:24 + mov.u32 %r15, %ctaid.x; + ld.param.b64 %rd7, [l2norm_bwd_kernel_param_2]; + .loc 1 95 16 // l2norm.py:95:16 + mul.wide.u32 %rd8, %r15, 16; + ld.param.b64 %rd9, [l2norm_bwd_kernel_param_3]; + .loc 1 95 34 // l2norm.py:95:34 + mov.u32 %r16, %tid.x; + ld.param.s32 %rd10, [l2norm_bwd_kernel_param_5]; + bfe.u32 %r17, %r16, 4, 4; + .loc 1 95 21 // l2norm.py:95:21 + cvt.u64.u32 %rd11, %r17; + or.b64 %rd12, %rd8, %rd11; + .loc 1 97 16 // l2norm.py:97:16 + setp.lt.s64 %p1, %rd12, %rd10; + .loc 1 98 30 // l2norm.py:98:30 + shl.b32 %r18, %r16, 3; + and.b32 %r19, %r18, 120; + .loc 1 99 14 // l2norm.py:99:14 + shl.b64 %rd13, %rd12, 8; + add.s64 %rd14, %rd5, %rd13; + .loc 1 99 33 // l2norm.py:99:33 + mul.wide.u32 %rd15, %r19, 2; + add.s64 %rd1, %rd14, %rd15; + .loc 1 100 20 // l2norm.py:100:20 + shl.b64 %rd16, %rd12, 2; + add.s64 %rd2, %rd6, %rd16; + .loc 1 101 16 // l2norm.py:101:16 + add.s64 %rd17, %rd7, %rd13; + .loc 1 101 35 // l2norm.py:101:35 + add.s64 %rd3, %rd17, %rd15; + .loc 1 102 16 // l2norm.py:102:16 + add.s64 %rd18, %rd9, %rd13; + .loc 1 102 35 // l2norm.py:102:35 + add.s64 %rd4, %rd18, %rd15; + mov.b32 %r5, 0; + .loc 1 104 18 // l2norm.py:104:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + mov.b32 {%rs5, %rs6}, %r3; + mov.b32 {%rs7, %rs8}, %r4; + .loc 1 104 47 // l2norm.py:104:47 + cvt.f32.bf16 %r20, %rs1; + cvt.f32.bf16 %r21, %rs2; + cvt.f32.bf16 %r22, %rs3; + cvt.f32.bf16 %r23, %rs4; + cvt.f32.bf16 %r24, %rs5; + cvt.f32.bf16 %r25, %rs6; + cvt.f32.bf16 %r26, %rs7; + cvt.f32.bf16 %r27, %rs8; + .loc 1 105 21 // l2norm.py:105:21 + // begin inline asm + mov.u32 %r6, %r5; + @%p1 ld.global.b32 { %r6 }, [ %rd2 + 0 ]; + // end inline asm + .loc 1 106 19 // l2norm.py:106:19 + // begin inline asm + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + mov.u32 %r10, %r5; + @%p1 ld.global.v4.b32 { %r7, %r8, %r9, %r10 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs9, %rs10}, %r7; + mov.b32 {%rs11, %rs12}, %r8; + mov.b32 {%rs13, %rs14}, %r9; + mov.b32 {%rs15, %rs16}, %r10; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r28, %rs9; + cvt.f32.bf16 %r29, %rs10; + cvt.f32.bf16 %r30, %rs11; + cvt.f32.bf16 %r31, %rs12; + cvt.f32.bf16 %r32, %rs13; + cvt.f32.bf16 %r33, %rs14; + cvt.f32.bf16 %r34, %rs15; + cvt.f32.bf16 %r35, %rs16; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r36, %r21, %r29; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r37, %rs1, %rs9, %r36; + fma.rn.f32.bf16 %r38, %rs3, %rs11, %r37; + fma.rn.f32.bf16 %r39, %rs4, %rs12, %r38; + fma.rn.f32.bf16 %r40, %rs5, %rs13, %r39; + fma.rn.f32.bf16 %r41, %rs6, %rs14, %r40; + fma.rn.f32.bf16 %r42, %rs7, %rs15, %r41; + fma.rn.f32.bf16 %r43, %rs8, %rs16, %r42; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r44, %r43, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r45, %r43, %r44; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r46, %r45, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r47, %r45, %r46; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r48, %r47, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r49, %r47, %r48; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r50, %r49, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r51, %r49, %r50; +$L__tmp10: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r52, %r51, %r20; + mul.f32 %r53, %r51, %r21; + mul.f32 %r54, %r51, %r22; + mul.f32 %r55, %r51, %r23; + mul.f32 %r56, %r51, %r24; + mul.f32 %r57, %r51, %r25; + mul.f32 %r58, %r51, %r26; + mul.f32 %r59, %r51, %r27; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r60, %r52, %r6; + mul.f32 %r61, %r53, %r6; + mul.f32 %r62, %r54, %r6; + mul.f32 %r63, %r55, %r6; + mul.f32 %r64, %r56, %r6; + mul.f32 %r65, %r57, %r6; + mul.f32 %r66, %r58, %r6; + mul.f32 %r67, %r59, %r6; + .loc 1 107 36 // l2norm.py:107:36 + neg.f32 %r68, %r60; + fma.rn.f32 %r69, %r6, %r28, %r68; + neg.f32 %r70, %r61; + fma.rn.f32 %r71, %r6, %r29, %r70; + neg.f32 %r72, %r62; + fma.rn.f32 %r73, %r6, %r30, %r72; + neg.f32 %r74, %r63; + fma.rn.f32 %r75, %r6, %r31, %r74; + neg.f32 %r76, %r64; + fma.rn.f32 %r77, %r6, %r32, %r76; + neg.f32 %r78, %r65; + fma.rn.f32 %r79, %r6, %r33, %r78; + neg.f32 %r80, %r66; + fma.rn.f32 %r81, %r6, %r34, %r80; + neg.f32 %r82, %r67; + fma.rn.f32 %r83, %r6, %r35, %r82; + .loc 1 108 19 // l2norm.py:108:19 + cvt.rn.bf16x2.f32 %r11, %r71, %r69; + cvt.rn.bf16x2.f32 %r12, %r75, %r73; + cvt.rn.bf16x2.f32 %r13, %r79, %r77; + cvt.rn.bf16x2.f32 %r14, %r83, %r81; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd4 + 0 ], { %r11, %r12, %r13, %r14 }; + // end inline asm + .loc 1 108 4 // l2norm.py:108:4 + ret; +$L__tmp11: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 98 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp10 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 107 // DW_AT_call_line +.b8 55 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp10 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/l2norm_bwd_kernel.source b/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/l2norm_bwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..25926967fd8ce8ce5392a0037c1d84708c2d91fa --- /dev/null +++ b/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/l2norm_bwd_kernel.source @@ -0,0 +1,210 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc47 = loc(unknown) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc54 = loc("y"(#loc)) +#loc55 = loc("rstd"(#loc)) +#loc56 = loc("dy"(#loc)) +#loc57 = loc("dx"(#loc)) +#loc58 = loc("eps"(#loc)) +#loc59 = loc("T"(#loc)) +#loc101 = loc("input"(#loc45)) +#loc102 = loc("a"(#loc50)) +#loc103 = loc("b"(#loc50)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc60) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc61) + %o_t = arith.constant 16 : i32 loc(#loc62) + %o_t_1 = arith.constant 16 : i64 loc(#loc62) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc62) + %o_t_3 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32> loc(#loc63) + %o_t_4 = arith.extsi %o_t_3 : tensor<16xi32> to tensor<16xi64> loc(#loc64) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<16xi64> loc(#loc64) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<16xi64> loc(#loc64) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc65) + %m_t = arith.extsi %T : i32 to i64 loc(#loc66) + %m_t_7 = tt.splat %m_t : i64 -> tensor<16xi64> loc(#loc66) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<16xi64> loc(#loc66) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc67) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc68) + %m_x_10 = arith.constant 128 : i32 loc(#loc69) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc69) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc69) + %m_x_13 = tt.broadcast %m_x : tensor<16x1xi1> -> tensor<16x128xi1> loc(#loc70) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<16x128xi1> loc(#loc70) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<16x128xi1> loc(#loc70) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc71) + %p_y_16 = arith.constant 128 : i32 loc(#loc72) + %p_y_17 = arith.constant 128 : i64 loc(#loc72) + %p_y_18 = arith.constant dense<128> : tensor<16x1xi64> loc(#loc72) + %p_y_19 = arith.muli %p_y, %p_y_18 : tensor<16x1xi64> loc(#loc72) + %p_y_20 = tt.splat %y : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc73) + %p_y_21 = tt.addptr %p_y_20, %p_y_19 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc73) + %p_y_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc74) + %p_y_23 = tt.broadcast %p_y_21 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc75) + %p_y_24 = tt.broadcast %p_y_22 : tensor<1x128xi32> -> tensor<16x128xi32> loc(#loc75) + %p_y_25 = tt.addptr %p_y_23, %p_y_24 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc75) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc76) + %p_rstd_26 = tt.addptr %p_rstd, %o_t_6 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc76) + %p_dy = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc77) + %p_dy_27 = arith.constant 128 : i32 loc(#loc78) + %p_dy_28 = arith.constant 128 : i64 loc(#loc78) + %p_dy_29 = arith.constant dense<128> : tensor<16x1xi64> loc(#loc78) + %p_dy_30 = arith.muli %p_dy, %p_dy_29 : tensor<16x1xi64> loc(#loc78) + %p_dy_31 = tt.splat %dy : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc79) + %p_dy_32 = tt.addptr %p_dy_31, %p_dy_30 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc79) + %p_dy_33 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc80) + %p_dy_34 = tt.broadcast %p_dy_32 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc81) + %p_dy_35 = tt.broadcast %p_dy_33 : tensor<1x128xi32> -> tensor<16x128xi32> loc(#loc81) + %p_dy_36 = tt.addptr %p_dy_34, %p_dy_35 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc81) + %p_dx = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc82) + %p_dx_37 = arith.constant 128 : i32 loc(#loc83) + %p_dx_38 = arith.constant 128 : i64 loc(#loc83) + %p_dx_39 = arith.constant dense<128> : tensor<16x1xi64> loc(#loc83) + %p_dx_40 = arith.muli %p_dx, %p_dx_39 : tensor<16x1xi64> loc(#loc83) + %p_dx_41 = tt.splat %dx : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc84) + %p_dx_42 = tt.addptr %p_dx_41, %p_dx_40 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc84) + %p_dx_43 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc85) + %p_dx_44 = tt.broadcast %p_dx_42 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc86) + %p_dx_45 = tt.broadcast %p_dx_43 : tensor<1x128xi32> -> tensor<16x128xi32> loc(#loc86) + %p_dx_46 = tt.addptr %p_dx_44, %p_dx_45 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc86) + %b_y = arith.constant 0.000000e+00 : f32 loc(#loc87) + %b_y_47 = arith.constant dense<0.000000e+00> : tensor<16x128xf32> loc(#loc87) + %b_y_48 = arith.truncf %b_y_47 : tensor<16x128xf32> to tensor<16x128xbf16> loc(#loc87) + %b_y_49 = tt.load %p_y_25, %m_x_15, %b_y_48 : tensor<16x128x!tt.ptr> loc(#loc87) + %b_y_50 = arith.extf %b_y_49 : tensor<16x128xbf16> to tensor<16x128xf32> loc(#loc88) + %b_rstd = arith.constant 0.000000e+00 : f32 loc(#loc89) + %b_rstd_51 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc89) + %b_rstd_52 = tt.load %p_rstd_26, %m_t_8, %b_rstd_51 : tensor<16x!tt.ptr> loc(#loc89) + %b_dy = arith.constant 0.000000e+00 : f32 loc(#loc90) + %b_dy_53 = arith.constant dense<0.000000e+00> : tensor<16x128xf32> loc(#loc90) + %b_dy_54 = arith.truncf %b_dy_53 : tensor<16x128xf32> to tensor<16x128xbf16> loc(#loc90) + %b_dy_55 = tt.load %p_dy_36, %m_x_15, %b_dy_54 : tensor<16x128x!tt.ptr> loc(#loc90) + %b_dy_56 = arith.extf %b_dy_55 : tensor<16x128xbf16> to tensor<16x128xf32> loc(#loc91) + %b_dx = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc92) + %b_dx_57 = tt.broadcast %b_dx : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc93) + %b_dx_58 = arith.mulf %b_dy_56, %b_dx_57 : tensor<16x128xf32> loc(#loc93) + %b_dx_59 = arith.mulf %b_dy_56, %b_y_50 : tensor<16x128xf32> loc(#loc94) + %b_dx_60 = tt.call @triton.language.standard.sum__fp32S16_128S_c1_cFalse_cNone(%b_dx_59) : (tensor<16x128xf32>) -> tensor<16xf32> loc(#loc95) + %b_dx_61 = tt.expand_dims %b_dx_60 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc96) + %b_dx_62 = tt.broadcast %b_dx_61 : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc97) + %b_dx_63 = arith.mulf %b_dx_62, %b_y_50 : tensor<16x128xf32> loc(#loc97) + %b_dx_64 = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc98) + %b_dx_65 = tt.broadcast %b_dx_64 : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc99) + %b_dx_66 = arith.mulf %b_dx_63, %b_dx_65 : tensor<16x128xf32> loc(#loc99) + %b_dx_67 = arith.subf %b_dx_58, %b_dx_66 : tensor<16x128xf32> loc(#loc100) + %0 = arith.truncf %b_dx_67 : tensor<16x128xf32> to tensor<16x128xbf16> loc(#loc42) + tt.store %p_dx_46, %0, %m_x_15 : tensor<16x128x!tt.ptr> loc(#loc43) + tt.return loc(#loc44) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S16_128S_c1_cFalse_cNone(%input: tensor<16x128xf32> loc("input"(#loc45))) -> tensor<16xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc46) + tt.reduce.return %2 : f32 loc(#loc46) + }) : (tensor<16x128xf32>) -> tensor<16xf32> loc(#loc46) + tt.return %0 : tensor<16xf32> loc(#loc48) + ^bb1: // no predecessors + %1 = ub.poison : tensor<16xf32> loc(#loc49) + tt.return %1 : tensor<16xf32> loc(#loc49) + } loc(#loc45) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc50)), %b: f32 loc("b"(#loc50))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc51) + tt.return %0 : f32 loc(#loc52) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc53) + tt.return %1 : f32 loc(#loc53) + } loc(#loc50) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:31) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:39) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:20) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:31) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:39) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:82) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc60 = loc("i_t"(#loc1)) +#loc61 = loc("i_t"(#loc2)) +#loc62 = loc("o_t"(#loc3)) +#loc63 = loc("o_t"(#loc4)) +#loc64 = loc("o_t"(#loc5)) +#loc65 = loc("o_d"(#loc6)) +#loc66 = loc("m_t"(#loc7)) +#loc67 = loc("m_x"(#loc8)) +#loc68 = loc("m_x"(#loc9)) +#loc69 = loc("m_x"(#loc10)) +#loc70 = loc("m_x"(#loc11)) +#loc71 = loc("p_y"(#loc12)) +#loc72 = loc("p_y"(#loc13)) +#loc73 = loc("p_y"(#loc14)) +#loc74 = loc("p_y"(#loc15)) +#loc75 = loc("p_y"(#loc16)) +#loc76 = loc("p_rstd"(#loc17)) +#loc77 = loc("p_dy"(#loc18)) +#loc78 = loc("p_dy"(#loc19)) +#loc79 = loc("p_dy"(#loc20)) +#loc80 = loc("p_dy"(#loc21)) +#loc81 = loc("p_dy"(#loc22)) +#loc82 = loc("p_dx"(#loc23)) +#loc83 = loc("p_dx"(#loc24)) +#loc84 = loc("p_dx"(#loc25)) +#loc85 = loc("p_dx"(#loc26)) +#loc86 = loc("p_dx"(#loc27)) +#loc87 = loc("b_y"(#loc28)) +#loc88 = loc("b_y"(#loc29)) +#loc89 = loc("b_rstd"(#loc30)) +#loc90 = loc("b_dy"(#loc31)) +#loc91 = loc("b_dy"(#loc32)) +#loc92 = loc("b_dx"(#loc33)) +#loc93 = loc("b_dx"(#loc34)) +#loc94 = loc("b_dx"(#loc35)) +#loc95 = loc("b_dx"(#loc36)) +#loc96 = loc("b_dx"(#loc37)) +#loc97 = loc("b_dx"(#loc38)) +#loc98 = loc("b_dx"(#loc39)) +#loc99 = loc("b_dx"(#loc40)) +#loc100 = loc("b_dx"(#loc41)) diff --git a/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/l2norm_bwd_kernel.ttgir b/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/l2norm_bwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..5c093764649316b8697e94a6e5993b80b0a138ff --- /dev/null +++ b/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/l2norm_bwd_kernel.ttgir @@ -0,0 +1,146 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [8, 1], order = [1, 0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc39 = loc("y"(#loc)) +#loc40 = loc("rstd"(#loc)) +#loc41 = loc("dy"(#loc)) +#loc42 = loc("dx"(#loc)) +#loc43 = loc("eps"(#loc)) +#loc44 = loc("T"(#loc)) +#loc72 = loc("b_dx"(#loc30)) +#loc78 = loc(callsite(#loc1 at #loc72)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 8 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %c16_i64 = arith.constant 16 : i64 loc(#loc1) + %cst = arith.constant dense<0.000000e+00> : tensor<16x128xbf16, #blocked> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<16x1xi64, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc45) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc46) + %o_t = arith.muli %i_t_3, %c16_i64 : i64 loc(#loc47) + %o_t_4 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc48) + %o_t_5 = arith.extsi %o_t_4 : tensor<16xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %o_t_6 = tt.splat %o_t : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %o_t_7 = arith.addi %o_t_6, %o_t_5 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %m_t = arith.extsi %T : i32 to i64 loc(#loc50) + %m_t_8 = tt.splat %m_t : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc50) + %m_t_9 = arith.cmpi slt, %o_t_7, %m_t_8 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc50) + %m_x = tt.expand_dims %m_t_9 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi1, #blocked> loc(#loc51) + %m_x_10 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc52) + %m_x_11 = tt.expand_dims %m_x_10 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc52) + %m_x_12 = arith.cmpi slt, %m_x_11, %cst_2 : tensor<1x128xi32, #blocked> loc(#loc53) + %m_x_13 = tt.broadcast %m_x : tensor<16x1xi1, #blocked> -> tensor<16x128xi1, #blocked> loc(#loc54) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1, #blocked> -> tensor<16x128xi1, #blocked> loc(#loc54) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<16x128xi1, #blocked> loc(#loc54) + %p_y = tt.expand_dims %o_t_7 {axis = 1 : i32} : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi64, #blocked> loc(#loc55) + %p_y_16 = arith.muli %p_y, %cst_1 : tensor<16x1xi64, #blocked> loc(#loc56) + %p_y_17 = tt.splat %y : !tt.ptr -> tensor<16x1x!tt.ptr, #blocked> loc(#loc57) + %p_y_18 = tt.addptr %p_y_17, %p_y_16 : tensor<16x1x!tt.ptr, #blocked>, tensor<16x1xi64, #blocked> loc(#loc57) + %p_y_19 = tt.broadcast %p_y_18 : tensor<16x1x!tt.ptr, #blocked> -> tensor<16x128x!tt.ptr, #blocked> loc(#loc58) + %p_y_20 = tt.broadcast %m_x_11 : tensor<1x128xi32, #blocked> -> tensor<16x128xi32, #blocked> loc(#loc58) + %p_y_21 = tt.addptr %p_y_19, %p_y_20 : tensor<16x128x!tt.ptr, #blocked>, tensor<16x128xi32, #blocked> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc59) + %p_rstd_22 = tt.addptr %p_rstd, %o_t_7 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>>, tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc59) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<16x1x!tt.ptr, #blocked> loc(#loc60) + %p_dy_23 = tt.addptr %p_dy, %p_y_16 : tensor<16x1x!tt.ptr, #blocked>, tensor<16x1xi64, #blocked> loc(#loc60) + %p_dy_24 = tt.broadcast %p_dy_23 : tensor<16x1x!tt.ptr, #blocked> -> tensor<16x128x!tt.ptr, #blocked> loc(#loc61) + %p_dy_25 = tt.addptr %p_dy_24, %p_y_20 : tensor<16x128x!tt.ptr, #blocked>, tensor<16x128xi32, #blocked> loc(#loc61) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<16x1x!tt.ptr, #blocked> loc(#loc62) + %p_dx_26 = tt.addptr %p_dx, %p_y_16 : tensor<16x1x!tt.ptr, #blocked>, tensor<16x1xi64, #blocked> loc(#loc62) + %p_dx_27 = tt.broadcast %p_dx_26 : tensor<16x1x!tt.ptr, #blocked> -> tensor<16x128x!tt.ptr, #blocked> loc(#loc63) + %p_dx_28 = tt.addptr %p_dx_27, %p_y_20 : tensor<16x128x!tt.ptr, #blocked>, tensor<16x128xi32, #blocked> loc(#loc63) + %b_y = tt.load %p_y_21, %m_x_15, %cst : tensor<16x128x!tt.ptr, #blocked> loc(#loc64) + %b_y_29 = arith.extf %b_y : tensor<16x128xbf16, #blocked> to tensor<16x128xf32, #blocked> loc(#loc65) + %b_rstd = tt.load %p_rstd_22, %m_t_9, %cst_0 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc66) + %b_dy = tt.load %p_dy_25, %m_x_15, %cst : tensor<16x128x!tt.ptr, #blocked> loc(#loc67) + %b_dy_30 = arith.extf %b_dy : tensor<16x128xbf16, #blocked> to tensor<16x128xf32, #blocked> loc(#loc68) + %b_dx = tt.expand_dims %b_rstd {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xf32, #blocked> loc(#loc69) + %b_dx_31 = tt.broadcast %b_dx : tensor<16x1xf32, #blocked> -> tensor<16x128xf32, #blocked> loc(#loc70) + %b_dx_32 = arith.mulf %b_dy_30, %b_dx_31 : tensor<16x128xf32, #blocked> loc(#loc70) + %b_dx_33 = arith.mulf %b_dy_30, %b_y_29 : tensor<16x128xf32, #blocked> loc(#loc71) + %b_dx_34 = "tt.reduce"(%b_dx_33) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_40: f32 loc(callsite(#loc1 at #loc72)), %b_dx_41: f32 loc(callsite(#loc1 at #loc72))): + %b_dx_42 = arith.addf %b_dx_40, %b_dx_41 : f32 loc(#loc79) + tt.reduce.return %b_dx_42 : f32 loc(#loc77) + }) : (tensor<16x128xf32, #blocked>) -> tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc77) + %b_dx_35 = tt.expand_dims %b_dx_34 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xf32, #blocked> loc(#loc73) + %b_dx_36 = tt.broadcast %b_dx_35 : tensor<16x1xf32, #blocked> -> tensor<16x128xf32, #blocked> loc(#loc74) + %b_dx_37 = arith.mulf %b_dx_36, %b_y_29 : tensor<16x128xf32, #blocked> loc(#loc74) + %b_dx_38 = arith.mulf %b_dx_37, %b_dx_31 : tensor<16x128xf32, #blocked> loc(#loc75) + %b_dx_39 = arith.subf %b_dx_32, %b_dx_38 : tensor<16x128xf32, #blocked> loc(#loc76) + %0 = arith.truncf %b_dx_39 : tensor<16x128xf32, #blocked> to tensor<16x128xbf16, #blocked> loc(#loc36) + tt.store %p_dx_28, %0, %m_x_15 : tensor<16x128x!tt.ptr, #blocked> loc(#loc37) + tt.return loc(#loc38) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc45 = loc("i_t"(#loc2)) +#loc46 = loc("i_t"(#loc3)) +#loc47 = loc("o_t"(#loc4)) +#loc48 = loc("o_t"(#loc5)) +#loc49 = loc("o_t"(#loc6)) +#loc50 = loc("m_t"(#loc7)) +#loc51 = loc("m_x"(#loc8)) +#loc52 = loc("m_x"(#loc9)) +#loc53 = loc("m_x"(#loc10)) +#loc54 = loc("m_x"(#loc11)) +#loc55 = loc("p_y"(#loc12)) +#loc56 = loc("p_y"(#loc13)) +#loc57 = loc("p_y"(#loc14)) +#loc58 = loc("p_y"(#loc15)) +#loc59 = loc("p_rstd"(#loc16)) +#loc60 = loc("p_dy"(#loc17)) +#loc61 = loc("p_dy"(#loc18)) +#loc62 = loc("p_dx"(#loc19)) +#loc63 = loc("p_dx"(#loc20)) +#loc64 = loc("b_y"(#loc21)) +#loc65 = loc("b_y"(#loc22)) +#loc66 = loc("b_rstd"(#loc23)) +#loc67 = loc("b_dy"(#loc24)) +#loc68 = loc("b_dy"(#loc25)) +#loc69 = loc("b_dx"(#loc26)) +#loc70 = loc("b_dx"(#loc27)) +#loc71 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc32)) +#loc74 = loc("b_dx"(#loc33)) +#loc75 = loc("b_dx"(#loc34)) +#loc76 = loc("b_dx"(#loc35)) +#loc77 = loc(callsite(#loc29 at #loc72)) +#loc79 = loc(callsite(#loc31 at #loc77)) diff --git a/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/l2norm_bwd_kernel.ttir b/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/l2norm_bwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..6c29416e2d7c989d3b096ee1e98cbabbdbd07e07 --- /dev/null +++ b/triton/gpu0/B67G6JGWQH52ULCHLXCUHXIFTUZ4HDQ344KE54NF7TSRZNFAHSTQ/l2norm_bwd_kernel.ttir @@ -0,0 +1,147 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc40 = loc("y"(#loc)) +#loc41 = loc("rstd"(#loc)) +#loc42 = loc("dy"(#loc)) +#loc43 = loc("dx"(#loc)) +#loc44 = loc("eps"(#loc)) +#loc45 = loc("T"(#loc)) +#loc74 = loc("b_dx"(#loc31)) +#loc80 = loc(callsite(#loc1 at #loc74)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<0.000000e+00> : tensor<16x128xbf16> loc(#loc1) + %b_rstd = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc46) + %cst_0 = arith.constant dense<128> : tensor<16x1xi64> loc(#loc1) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc47) + %o_t = arith.constant 16 : i64 loc(#loc48) + %i_t = tt.get_program_id x : i32 loc(#loc49) + %i_t_1 = arith.extsi %i_t : i32 to i64 loc(#loc50) + %o_t_2 = arith.muli %i_t_1, %o_t : i64 loc(#loc48) + %o_t_3 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32> loc(#loc51) + %o_t_4 = arith.extsi %o_t_3 : tensor<16xi32> to tensor<16xi64> loc(#loc52) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<16xi64> loc(#loc52) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<16xi64> loc(#loc52) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc53) + %m_t = arith.extsi %T : i32 to i64 loc(#loc54) + %m_t_7 = tt.splat %m_t : i64 -> tensor<16xi64> loc(#loc54) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<16xi64> loc(#loc54) + %m_x_9 = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc55) + %m_x_10 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc56) + %m_x_11 = arith.cmpi slt, %m_x_10, %m_x : tensor<1x128xi32> loc(#loc47) + %m_x_12 = tt.broadcast %m_x_9 : tensor<16x1xi1> -> tensor<16x128xi1> loc(#loc57) + %m_x_13 = tt.broadcast %m_x_11 : tensor<1x128xi1> -> tensor<16x128xi1> loc(#loc57) + %m_x_14 = arith.andi %m_x_12, %m_x_13 : tensor<16x128xi1> loc(#loc57) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc58) + %p_y_15 = arith.muli %p_y, %cst_0 : tensor<16x1xi64> loc(#loc59) + %p_y_16 = tt.splat %y : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc60) + %p_y_17 = tt.addptr %p_y_16, %p_y_15 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc60) + %p_y_18 = tt.broadcast %p_y_17 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc61) + %p_y_19 = tt.broadcast %m_x_10 : tensor<1x128xi32> -> tensor<16x128xi32> loc(#loc61) + %p_y_20 = tt.addptr %p_y_18, %p_y_19 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc61) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc62) + %p_rstd_21 = tt.addptr %p_rstd, %o_t_6 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc62) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc63) + %p_dy_22 = tt.addptr %p_dy, %p_y_15 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc63) + %p_dy_23 = tt.broadcast %p_dy_22 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc64) + %p_dy_24 = tt.addptr %p_dy_23, %p_y_19 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc64) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc65) + %p_dx_25 = tt.addptr %p_dx, %p_y_15 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc65) + %p_dx_26 = tt.broadcast %p_dx_25 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc66) + %p_dx_27 = tt.addptr %p_dx_26, %p_y_19 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc66) + %b_y = tt.load %p_y_20, %m_x_14, %cst : tensor<16x128x!tt.ptr> loc(#loc67) + %b_y_28 = arith.extf %b_y : tensor<16x128xbf16> to tensor<16x128xf32> loc(#loc68) + %b_rstd_29 = tt.load %p_rstd_21, %m_t_8, %b_rstd : tensor<16x!tt.ptr> loc(#loc46) + %b_dy = tt.load %p_dy_24, %m_x_14, %cst : tensor<16x128x!tt.ptr> loc(#loc69) + %b_dy_30 = arith.extf %b_dy : tensor<16x128xbf16> to tensor<16x128xf32> loc(#loc70) + %b_dx = tt.expand_dims %b_rstd_29 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc71) + %b_dx_31 = tt.broadcast %b_dx : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc72) + %b_dx_32 = arith.mulf %b_dy_30, %b_dx_31 : tensor<16x128xf32> loc(#loc72) + %b_dx_33 = arith.mulf %b_dy_30, %b_y_28 : tensor<16x128xf32> loc(#loc73) + %b_dx_34 = "tt.reduce"(%b_dx_33) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_40: f32 loc(callsite(#loc1 at #loc74)), %b_dx_41: f32 loc(callsite(#loc1 at #loc74))): + %b_dx_42 = arith.addf %b_dx_40, %b_dx_41 : f32 loc(#loc81) + tt.reduce.return %b_dx_42 : f32 loc(#loc79) + }) : (tensor<16x128xf32>) -> tensor<16xf32> loc(#loc79) + %b_dx_35 = tt.expand_dims %b_dx_34 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc75) + %b_dx_36 = tt.broadcast %b_dx_35 : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc76) + %b_dx_37 = arith.mulf %b_dx_36, %b_y_28 : tensor<16x128xf32> loc(#loc76) + %b_dx_38 = arith.mulf %b_dx_37, %b_dx_31 : tensor<16x128xf32> loc(#loc77) + %b_dx_39 = arith.subf %b_dx_32, %b_dx_38 : tensor<16x128xf32> loc(#loc78) + %0 = arith.truncf %b_dx_39 : tensor<16x128xf32> to tensor<16x128xbf16> loc(#loc37) + tt.store %p_dx_27, %0, %m_x_14 : tensor<16x128x!tt.ptr> loc(#loc38) + tt.return loc(#loc39) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("b_rstd"(#loc2)) +#loc47 = loc("m_x"(#loc3)) +#loc48 = loc("o_t"(#loc4)) +#loc49 = loc("i_t"(#loc5)) +#loc50 = loc("i_t"(#loc6)) +#loc51 = loc("o_t"(#loc7)) +#loc52 = loc("o_t"(#loc8)) +#loc53 = loc("o_d"(#loc9)) +#loc54 = loc("m_t"(#loc10)) +#loc55 = loc("m_x"(#loc11)) +#loc56 = loc("m_x"(#loc12)) +#loc57 = loc("m_x"(#loc13)) +#loc58 = loc("p_y"(#loc14)) +#loc59 = loc("p_y"(#loc15)) +#loc60 = loc("p_y"(#loc16)) +#loc61 = loc("p_y"(#loc17)) +#loc62 = loc("p_rstd"(#loc18)) +#loc63 = loc("p_dy"(#loc19)) +#loc64 = loc("p_dy"(#loc20)) +#loc65 = loc("p_dx"(#loc21)) +#loc66 = loc("p_dx"(#loc22)) +#loc67 = loc("b_y"(#loc23)) +#loc68 = loc("b_y"(#loc24)) +#loc69 = loc("b_dy"(#loc25)) +#loc70 = loc("b_dy"(#loc26)) +#loc71 = loc("b_dx"(#loc27)) +#loc72 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc29)) +#loc75 = loc("b_dx"(#loc33)) +#loc76 = loc("b_dx"(#loc34)) +#loc77 = loc("b_dx"(#loc35)) +#loc78 = loc("b_dx"(#loc36)) +#loc79 = loc(callsite(#loc30 at #loc74)) +#loc81 = loc(callsite(#loc32 at #loc79)) diff --git a/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/__grp__l2norm_fwd_kernel.json b/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/__grp__l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..7504790b398ae03754f775d69d214d79b9abe20c --- /dev/null +++ b/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/__grp__l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_fwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/l2norm_fwd_kernel.source", "l2norm_fwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/l2norm_fwd_kernel.ttir", "l2norm_fwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/l2norm_fwd_kernel.ttgir", "l2norm_fwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/l2norm_fwd_kernel.llir", "l2norm_fwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/l2norm_fwd_kernel.ptx", "l2norm_fwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/l2norm_fwd_kernel.cubin", "l2norm_fwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/l2norm_fwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/l2norm_fwd_kernel.cubin b/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/l2norm_fwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..549ecca3308808fa2a37d56972fe26cb958c7a61 Binary files /dev/null and b/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/l2norm_fwd_kernel.cubin differ diff --git a/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/l2norm_fwd_kernel.json b/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..d7d8c63611834db6268234cedf43783769fca345 --- /dev/null +++ b/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"hash": "0809f294e4f96eb0a67717f066adc91cd2bbca5e611141fbb23cb888f245a282", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 4, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 512, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_fwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/l2norm_fwd_kernel.llir b/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/l2norm_fwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..e82d4e160ac5b39c1b2a7bdc2eb731e642ebced0 --- /dev/null +++ b/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/l2norm_fwd_kernel.llir @@ -0,0 +1,2155 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 +@.str.2 = private unnamed_addr constant [17 x i8] c"__CUDA_PREC_SQRT\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_fwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, float %3, i32 %4, ptr addrspace(1) readnone captures(none) %5, ptr addrspace(1) readnone captures(none) %6) local_unnamed_addr #0 !dbg !5 { + %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %9 = zext nneg i32 %8 to i64, !dbg !14 + %10 = shl nuw nsw i64 %9, 7, !dbg !15 + %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !16 + %12 = and i32 %11, 112, !dbg !16 + %13 = lshr exact i32 %12, 4, !dbg !16 + %14 = or disjoint i32 %13, 8, !dbg !16 + %15 = insertelement <2 x i32> poison, i32 %13, i64 0, !dbg !16 + %16 = insertelement <4 x i32> poison, i32 %13, i64 0, !dbg !16 + %17 = insertelement <8 x i32> poison, i32 %13, i64 0, !dbg !16 + %18 = insertelement <16 x i32> poison, i32 %14, i64 14, !dbg !17 + %19 = insertelement <16 x i32> %18, i32 %13, i64 15, !dbg !17 + %20 = shufflevector <8 x i32> %17, <8 x i32> poison, <16 x i32> , !dbg !17 + %21 = or disjoint <16 x i32> %20, , !dbg !17 + %22 = shufflevector <16 x i32> %21, <16 x i32> %19, <16 x i32> , !dbg !17 + %23 = shufflevector <4 x i32> %16, <4 x i32> poison, <16 x i32> , !dbg !17 + %24 = or disjoint <16 x i32> %23, , !dbg !17 + %25 = shufflevector <16 x i32> %22, <16 x i32> %24, <16 x i32> , !dbg !17 + %26 = shufflevector <2 x i32> %15, <2 x i32> poison, <16 x i32> , !dbg !17 + %27 = or disjoint <16 x i32> %26, , !dbg !17 + %28 = shufflevector <16 x i32> %25, <16 x i32> %27, <16 x i32> , !dbg !17 + %29 = zext <16 x i32> %28 to <16 x i64>, !dbg !17 + %30 = insertelement <16 x i64> poison, i64 %10, i64 0, !dbg !17 + %31 = shufflevector <16 x i64> %30, <16 x i64> poison, <16 x i32> zeroinitializer, !dbg !17 + %32 = or disjoint <16 x i64> %31, %29, !dbg !17 + %33 = insertelement <16 x i32> poison, i32 %4, i64 0, !dbg !18 + %34 = shufflevector <16 x i32> %33, <16 x i32> poison, <16 x i32> zeroinitializer, !dbg !18 + %35 = sext <16 x i32> %34 to <16 x i64>, !dbg !18 + %36 = icmp slt <16 x i64> %32, %35, !dbg !18 + %37 = shl nuw nsw i32 %11, 3, !dbg !19 + %38 = and i32 %37, 120, !dbg !19 + %39 = extractelement <16 x i64> %32, i64 15, !dbg !20 + %40 = shl nuw nsw i64 %39, 7, !dbg !20 + %41 = extractelement <16 x i64> %32, i64 14, !dbg !20 + %42 = shl nuw nsw i64 %41, 7, !dbg !20 + %43 = extractelement <16 x i64> %32, i64 13, !dbg !20 + %44 = shl nuw nsw i64 %43, 7, !dbg !20 + %45 = extractelement <16 x i64> %32, i64 12, !dbg !20 + %46 = shl nuw nsw i64 %45, 7, !dbg !20 + %47 = extractelement <16 x i64> %32, i64 11, !dbg !20 + %48 = shl nuw nsw i64 %47, 7, !dbg !20 + %49 = extractelement <16 x i64> %32, i64 10, !dbg !20 + %50 = shl nuw nsw i64 %49, 7, !dbg !20 + %51 = extractelement <16 x i64> %32, i64 9, !dbg !20 + %52 = shl nuw nsw i64 %51, 7, !dbg !20 + %53 = extractelement <16 x i64> %32, i64 8, !dbg !20 + %54 = shl nuw nsw i64 %53, 7, !dbg !20 + %55 = extractelement <16 x i64> %32, i64 7, !dbg !20 + %56 = shl nuw nsw i64 %55, 7, !dbg !20 + %57 = extractelement <16 x i64> %32, i64 6, !dbg !20 + %58 = shl nuw nsw i64 %57, 7, !dbg !20 + %59 = extractelement <16 x i64> %32, i64 5, !dbg !20 + %60 = shl nuw nsw i64 %59, 7, !dbg !20 + %61 = extractelement <16 x i64> %32, i64 4, !dbg !20 + %62 = shl nuw nsw i64 %61, 7, !dbg !20 + %63 = extractelement <16 x i64> %32, i64 3, !dbg !20 + %64 = shl nuw nsw i64 %63, 7, !dbg !20 + %65 = extractelement <16 x i64> %32, i64 2, !dbg !20 + %66 = shl nuw nsw i64 %65, 7, !dbg !20 + %67 = extractelement <16 x i64> %32, i64 1, !dbg !20 + %68 = shl nuw nsw i64 %67, 7, !dbg !20 + %69 = extractelement <16 x i64> %32, i64 0, !dbg !20 + %70 = shl nuw nsw i64 %69, 7, !dbg !20 + %71 = getelementptr bfloat, ptr addrspace(1) %0, i64 %40, !dbg !21 + %72 = getelementptr bfloat, ptr addrspace(1) %0, i64 %42, !dbg !21 + %73 = getelementptr bfloat, ptr addrspace(1) %0, i64 %44, !dbg !21 + %74 = getelementptr bfloat, ptr addrspace(1) %0, i64 %46, !dbg !21 + %75 = getelementptr bfloat, ptr addrspace(1) %0, i64 %48, !dbg !21 + %76 = getelementptr bfloat, ptr addrspace(1) %0, i64 %50, !dbg !21 + %77 = getelementptr bfloat, ptr addrspace(1) %0, i64 %52, !dbg !21 + %78 = getelementptr bfloat, ptr addrspace(1) %0, i64 %54, !dbg !21 + %79 = getelementptr bfloat, ptr addrspace(1) %0, i64 %56, !dbg !21 + %80 = getelementptr bfloat, ptr addrspace(1) %0, i64 %58, !dbg !21 + %81 = getelementptr bfloat, ptr addrspace(1) %0, i64 %60, !dbg !21 + %82 = getelementptr bfloat, ptr addrspace(1) %0, i64 %62, !dbg !21 + %83 = getelementptr bfloat, ptr addrspace(1) %0, i64 %64, !dbg !21 + %84 = getelementptr bfloat, ptr addrspace(1) %0, i64 %66, !dbg !21 + %85 = getelementptr bfloat, ptr addrspace(1) %0, i64 %68, !dbg !21 + %86 = getelementptr bfloat, ptr addrspace(1) %0, i64 %70, !dbg !21 + %87 = zext nneg i32 %38 to i64, !dbg !22 + %88 = getelementptr bfloat, ptr addrspace(1) %71, i64 %87, !dbg !22 + %89 = getelementptr bfloat, ptr addrspace(1) %72, i64 %87, !dbg !22 + %90 = getelementptr bfloat, ptr addrspace(1) %73, i64 %87, !dbg !22 + %91 = getelementptr bfloat, ptr addrspace(1) %74, i64 %87, !dbg !22 + %92 = getelementptr bfloat, ptr addrspace(1) %75, i64 %87, !dbg !22 + %93 = getelementptr bfloat, ptr addrspace(1) %76, i64 %87, !dbg !22 + %94 = getelementptr bfloat, ptr addrspace(1) %77, i64 %87, !dbg !22 + %95 = getelementptr bfloat, ptr addrspace(1) %78, i64 %87, !dbg !22 + %96 = getelementptr bfloat, ptr addrspace(1) %79, i64 %87, !dbg !22 + %97 = getelementptr bfloat, ptr addrspace(1) %80, i64 %87, !dbg !22 + %98 = getelementptr bfloat, ptr addrspace(1) %81, i64 %87, !dbg !22 + %99 = getelementptr bfloat, ptr addrspace(1) %82, i64 %87, !dbg !22 + %100 = getelementptr bfloat, ptr addrspace(1) %83, i64 %87, !dbg !22 + %101 = getelementptr bfloat, ptr addrspace(1) %84, i64 %87, !dbg !22 + %102 = getelementptr bfloat, ptr addrspace(1) %85, i64 %87, !dbg !22 + %103 = getelementptr bfloat, ptr addrspace(1) %86, i64 %87, !dbg !22 + %104 = extractelement <16 x i1> %36, i64 15, !dbg !23 + %105 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %88, i1 %104) #6, !dbg !23 + %106 = extractvalue { i32, i32, i32, i32 } %105, 0, !dbg !23 + %107 = bitcast i32 %106 to <2 x bfloat>, !dbg !23 + %108 = extractvalue { i32, i32, i32, i32 } %105, 1, !dbg !23 + %109 = bitcast i32 %108 to <2 x bfloat>, !dbg !23 + %110 = extractvalue { i32, i32, i32, i32 } %105, 2, !dbg !23 + %111 = bitcast i32 %110 to <2 x bfloat>, !dbg !23 + %112 = extractvalue { i32, i32, i32, i32 } %105, 3, !dbg !23 + %113 = bitcast i32 %112 to <2 x bfloat>, !dbg !23 + %114 = extractelement <2 x bfloat> %107, i64 0, !dbg !23 + %115 = extractelement <2 x bfloat> %107, i64 1, !dbg !23 + %116 = extractelement <2 x bfloat> %109, i64 0, !dbg !23 + %117 = extractelement <2 x bfloat> %109, i64 1, !dbg !23 + %118 = extractelement <2 x bfloat> %111, i64 0, !dbg !23 + %119 = extractelement <2 x bfloat> %111, i64 1, !dbg !23 + %120 = extractelement <2 x bfloat> %113, i64 0, !dbg !23 + %121 = extractelement <2 x bfloat> %113, i64 1, !dbg !23 + %122 = extractelement <16 x i1> %36, i64 14, !dbg !23 + %123 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %89, i1 %122) #6, !dbg !23 + %124 = extractvalue { i32, i32, i32, i32 } %123, 0, !dbg !23 + %125 = bitcast i32 %124 to <2 x bfloat>, !dbg !23 + %126 = extractvalue { i32, i32, i32, i32 } %123, 1, !dbg !23 + %127 = bitcast i32 %126 to <2 x bfloat>, !dbg !23 + %128 = extractvalue { i32, i32, i32, i32 } %123, 2, !dbg !23 + %129 = bitcast i32 %128 to <2 x bfloat>, !dbg !23 + %130 = extractvalue { i32, i32, i32, i32 } %123, 3, !dbg !23 + %131 = bitcast i32 %130 to <2 x bfloat>, !dbg !23 + %132 = extractelement <2 x bfloat> %125, i64 0, !dbg !23 + %133 = extractelement <2 x bfloat> %125, i64 1, !dbg !23 + %134 = extractelement <2 x bfloat> %127, i64 0, !dbg !23 + %135 = extractelement <2 x bfloat> %127, i64 1, !dbg !23 + %136 = extractelement <2 x bfloat> %129, i64 0, !dbg !23 + %137 = extractelement <2 x bfloat> %129, i64 1, !dbg !23 + %138 = extractelement <2 x bfloat> %131, i64 0, !dbg !23 + %139 = extractelement <2 x bfloat> %131, i64 1, !dbg !23 + %140 = extractelement <16 x i1> %36, i64 13, !dbg !23 + %141 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %90, i1 %140) #6, !dbg !23 + %142 = extractvalue { i32, i32, i32, i32 } %141, 0, !dbg !23 + %143 = bitcast i32 %142 to <2 x bfloat>, !dbg !23 + %144 = extractvalue { i32, i32, i32, i32 } %141, 1, !dbg !23 + %145 = bitcast i32 %144 to <2 x bfloat>, !dbg !23 + %146 = extractvalue { i32, i32, i32, i32 } %141, 2, !dbg !23 + %147 = bitcast i32 %146 to <2 x bfloat>, !dbg !23 + %148 = extractvalue { i32, i32, i32, i32 } %141, 3, !dbg !23 + %149 = bitcast i32 %148 to <2 x bfloat>, !dbg !23 + %150 = extractelement <2 x bfloat> %143, i64 0, !dbg !23 + %151 = extractelement <2 x bfloat> %143, i64 1, !dbg !23 + %152 = extractelement <2 x bfloat> %145, i64 0, !dbg !23 + %153 = extractelement <2 x bfloat> %145, i64 1, !dbg !23 + %154 = extractelement <2 x bfloat> %147, i64 0, !dbg !23 + %155 = extractelement <2 x bfloat> %147, i64 1, !dbg !23 + %156 = extractelement <2 x bfloat> %149, i64 0, !dbg !23 + %157 = extractelement <2 x bfloat> %149, i64 1, !dbg !23 + %158 = extractelement <16 x i1> %36, i64 12, !dbg !23 + %159 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %91, i1 %158) #6, !dbg !23 + %160 = extractvalue { i32, i32, i32, i32 } %159, 0, !dbg !23 + %161 = bitcast i32 %160 to <2 x bfloat>, !dbg !23 + %162 = extractvalue { i32, i32, i32, i32 } %159, 1, !dbg !23 + %163 = bitcast i32 %162 to <2 x bfloat>, !dbg !23 + %164 = extractvalue { i32, i32, i32, i32 } %159, 2, !dbg !23 + %165 = bitcast i32 %164 to <2 x bfloat>, !dbg !23 + %166 = extractvalue { i32, i32, i32, i32 } %159, 3, !dbg !23 + %167 = bitcast i32 %166 to <2 x bfloat>, !dbg !23 + %168 = extractelement <2 x bfloat> %161, i64 0, !dbg !23 + %169 = extractelement <2 x bfloat> %161, i64 1, !dbg !23 + %170 = extractelement <2 x bfloat> %163, i64 0, !dbg !23 + %171 = extractelement <2 x bfloat> %163, i64 1, !dbg !23 + %172 = extractelement <2 x bfloat> %165, i64 0, !dbg !23 + %173 = extractelement <2 x bfloat> %165, i64 1, !dbg !23 + %174 = extractelement <2 x bfloat> %167, i64 0, !dbg !23 + %175 = extractelement <2 x bfloat> %167, i64 1, !dbg !23 + %176 = extractelement <16 x i1> %36, i64 11, !dbg !23 + %177 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %92, i1 %176) #6, !dbg !23 + %178 = extractvalue { i32, i32, i32, i32 } %177, 0, !dbg !23 + %179 = bitcast i32 %178 to <2 x bfloat>, !dbg !23 + %180 = extractvalue { i32, i32, i32, i32 } %177, 1, !dbg !23 + %181 = bitcast i32 %180 to <2 x bfloat>, !dbg !23 + %182 = extractvalue { i32, i32, i32, i32 } %177, 2, !dbg !23 + %183 = bitcast i32 %182 to <2 x bfloat>, !dbg !23 + %184 = extractvalue { i32, i32, i32, i32 } %177, 3, !dbg !23 + %185 = bitcast i32 %184 to <2 x bfloat>, !dbg !23 + %186 = extractelement <2 x bfloat> %179, i64 0, !dbg !23 + %187 = extractelement <2 x bfloat> %179, i64 1, !dbg !23 + %188 = extractelement <2 x bfloat> %181, i64 0, !dbg !23 + %189 = extractelement <2 x bfloat> %181, i64 1, !dbg !23 + %190 = extractelement <2 x bfloat> %183, i64 0, !dbg !23 + %191 = extractelement <2 x bfloat> %183, i64 1, !dbg !23 + %192 = extractelement <2 x bfloat> %185, i64 0, !dbg !23 + %193 = extractelement <2 x bfloat> %185, i64 1, !dbg !23 + %194 = extractelement <16 x i1> %36, i64 10, !dbg !23 + %195 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %93, i1 %194) #6, !dbg !23 + %196 = extractvalue { i32, i32, i32, i32 } %195, 0, !dbg !23 + %197 = bitcast i32 %196 to <2 x bfloat>, !dbg !23 + %198 = extractvalue { i32, i32, i32, i32 } %195, 1, !dbg !23 + %199 = bitcast i32 %198 to <2 x bfloat>, !dbg !23 + %200 = extractvalue { i32, i32, i32, i32 } %195, 2, !dbg !23 + %201 = bitcast i32 %200 to <2 x bfloat>, !dbg !23 + %202 = extractvalue { i32, i32, i32, i32 } %195, 3, !dbg !23 + %203 = bitcast i32 %202 to <2 x bfloat>, !dbg !23 + %204 = extractelement <2 x bfloat> %197, i64 0, !dbg !23 + %205 = extractelement <2 x bfloat> %197, i64 1, !dbg !23 + %206 = extractelement <2 x bfloat> %199, i64 0, !dbg !23 + %207 = extractelement <2 x bfloat> %199, i64 1, !dbg !23 + %208 = extractelement <2 x bfloat> %201, i64 0, !dbg !23 + %209 = extractelement <2 x bfloat> %201, i64 1, !dbg !23 + %210 = extractelement <2 x bfloat> %203, i64 0, !dbg !23 + %211 = extractelement <2 x bfloat> %203, i64 1, !dbg !23 + %212 = extractelement <16 x i1> %36, i64 9, !dbg !23 + %213 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %94, i1 %212) #6, !dbg !23 + %214 = extractvalue { i32, i32, i32, i32 } %213, 0, !dbg !23 + %215 = bitcast i32 %214 to <2 x bfloat>, !dbg !23 + %216 = extractvalue { i32, i32, i32, i32 } %213, 1, !dbg !23 + %217 = bitcast i32 %216 to <2 x bfloat>, !dbg !23 + %218 = extractvalue { i32, i32, i32, i32 } %213, 2, !dbg !23 + %219 = bitcast i32 %218 to <2 x bfloat>, !dbg !23 + %220 = extractvalue { i32, i32, i32, i32 } %213, 3, !dbg !23 + %221 = bitcast i32 %220 to <2 x bfloat>, !dbg !23 + %222 = extractelement <2 x bfloat> %215, i64 0, !dbg !23 + %223 = extractelement <2 x bfloat> %215, i64 1, !dbg !23 + %224 = extractelement <2 x bfloat> %217, i64 0, !dbg !23 + %225 = extractelement <2 x bfloat> %217, i64 1, !dbg !23 + %226 = extractelement <2 x bfloat> %219, i64 0, !dbg !23 + %227 = extractelement <2 x bfloat> %219, i64 1, !dbg !23 + %228 = extractelement <2 x bfloat> %221, i64 0, !dbg !23 + %229 = extractelement <2 x bfloat> %221, i64 1, !dbg !23 + %230 = extractelement <16 x i1> %36, i64 8, !dbg !23 + %231 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %95, i1 %230) #6, !dbg !23 + %232 = extractvalue { i32, i32, i32, i32 } %231, 0, !dbg !23 + %233 = bitcast i32 %232 to <2 x bfloat>, !dbg !23 + %234 = extractvalue { i32, i32, i32, i32 } %231, 1, !dbg !23 + %235 = bitcast i32 %234 to <2 x bfloat>, !dbg !23 + %236 = extractvalue { i32, i32, i32, i32 } %231, 2, !dbg !23 + %237 = bitcast i32 %236 to <2 x bfloat>, !dbg !23 + %238 = extractvalue { i32, i32, i32, i32 } %231, 3, !dbg !23 + %239 = bitcast i32 %238 to <2 x bfloat>, !dbg !23 + %240 = extractelement <2 x bfloat> %233, i64 0, !dbg !23 + %241 = extractelement <2 x bfloat> %233, i64 1, !dbg !23 + %242 = extractelement <2 x bfloat> %235, i64 0, !dbg !23 + %243 = extractelement <2 x bfloat> %235, i64 1, !dbg !23 + %244 = extractelement <2 x bfloat> %237, i64 0, !dbg !23 + %245 = extractelement <2 x bfloat> %237, i64 1, !dbg !23 + %246 = extractelement <2 x bfloat> %239, i64 0, !dbg !23 + %247 = extractelement <2 x bfloat> %239, i64 1, !dbg !23 + %248 = extractelement <16 x i1> %36, i64 7, !dbg !23 + %249 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %96, i1 %248) #6, !dbg !23 + %250 = extractvalue { i32, i32, i32, i32 } %249, 0, !dbg !23 + %251 = bitcast i32 %250 to <2 x bfloat>, !dbg !23 + %252 = extractvalue { i32, i32, i32, i32 } %249, 1, !dbg !23 + %253 = bitcast i32 %252 to <2 x bfloat>, !dbg !23 + %254 = extractvalue { i32, i32, i32, i32 } %249, 2, !dbg !23 + %255 = bitcast i32 %254 to <2 x bfloat>, !dbg !23 + %256 = extractvalue { i32, i32, i32, i32 } %249, 3, !dbg !23 + %257 = bitcast i32 %256 to <2 x bfloat>, !dbg !23 + %258 = extractelement <2 x bfloat> %251, i64 0, !dbg !23 + %259 = extractelement <2 x bfloat> %251, i64 1, !dbg !23 + %260 = extractelement <2 x bfloat> %253, i64 0, !dbg !23 + %261 = extractelement <2 x bfloat> %253, i64 1, !dbg !23 + %262 = extractelement <2 x bfloat> %255, i64 0, !dbg !23 + %263 = extractelement <2 x bfloat> %255, i64 1, !dbg !23 + %264 = extractelement <2 x bfloat> %257, i64 0, !dbg !23 + %265 = extractelement <2 x bfloat> %257, i64 1, !dbg !23 + %266 = extractelement <16 x i1> %36, i64 6, !dbg !23 + %267 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %97, i1 %266) #6, !dbg !23 + %268 = extractvalue { i32, i32, i32, i32 } %267, 0, !dbg !23 + %269 = bitcast i32 %268 to <2 x bfloat>, !dbg !23 + %270 = extractvalue { i32, i32, i32, i32 } %267, 1, !dbg !23 + %271 = bitcast i32 %270 to <2 x bfloat>, !dbg !23 + %272 = extractvalue { i32, i32, i32, i32 } %267, 2, !dbg !23 + %273 = bitcast i32 %272 to <2 x bfloat>, !dbg !23 + %274 = extractvalue { i32, i32, i32, i32 } %267, 3, !dbg !23 + %275 = bitcast i32 %274 to <2 x bfloat>, !dbg !23 + %276 = extractelement <2 x bfloat> %269, i64 0, !dbg !23 + %277 = extractelement <2 x bfloat> %269, i64 1, !dbg !23 + %278 = extractelement <2 x bfloat> %271, i64 0, !dbg !23 + %279 = extractelement <2 x bfloat> %271, i64 1, !dbg !23 + %280 = extractelement <2 x bfloat> %273, i64 0, !dbg !23 + %281 = extractelement <2 x bfloat> %273, i64 1, !dbg !23 + %282 = extractelement <2 x bfloat> %275, i64 0, !dbg !23 + %283 = extractelement <2 x bfloat> %275, i64 1, !dbg !23 + %284 = extractelement <16 x i1> %36, i64 5, !dbg !23 + %285 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %98, i1 %284) #6, !dbg !23 + %286 = extractvalue { i32, i32, i32, i32 } %285, 0, !dbg !23 + %287 = bitcast i32 %286 to <2 x bfloat>, !dbg !23 + %288 = extractvalue { i32, i32, i32, i32 } %285, 1, !dbg !23 + %289 = bitcast i32 %288 to <2 x bfloat>, !dbg !23 + %290 = extractvalue { i32, i32, i32, i32 } %285, 2, !dbg !23 + %291 = bitcast i32 %290 to <2 x bfloat>, !dbg !23 + %292 = extractvalue { i32, i32, i32, i32 } %285, 3, !dbg !23 + %293 = bitcast i32 %292 to <2 x bfloat>, !dbg !23 + %294 = extractelement <2 x bfloat> %287, i64 0, !dbg !23 + %295 = extractelement <2 x bfloat> %287, i64 1, !dbg !23 + %296 = extractelement <2 x bfloat> %289, i64 0, !dbg !23 + %297 = extractelement <2 x bfloat> %289, i64 1, !dbg !23 + %298 = extractelement <2 x bfloat> %291, i64 0, !dbg !23 + %299 = extractelement <2 x bfloat> %291, i64 1, !dbg !23 + %300 = extractelement <2 x bfloat> %293, i64 0, !dbg !23 + %301 = extractelement <2 x bfloat> %293, i64 1, !dbg !23 + %302 = extractelement <16 x i1> %36, i64 4, !dbg !23 + %303 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %99, i1 %302) #6, !dbg !23 + %304 = extractvalue { i32, i32, i32, i32 } %303, 0, !dbg !23 + %305 = bitcast i32 %304 to <2 x bfloat>, !dbg !23 + %306 = extractvalue { i32, i32, i32, i32 } %303, 1, !dbg !23 + %307 = bitcast i32 %306 to <2 x bfloat>, !dbg !23 + %308 = extractvalue { i32, i32, i32, i32 } %303, 2, !dbg !23 + %309 = bitcast i32 %308 to <2 x bfloat>, !dbg !23 + %310 = extractvalue { i32, i32, i32, i32 } %303, 3, !dbg !23 + %311 = bitcast i32 %310 to <2 x bfloat>, !dbg !23 + %312 = extractelement <2 x bfloat> %305, i64 0, !dbg !23 + %313 = extractelement <2 x bfloat> %305, i64 1, !dbg !23 + %314 = extractelement <2 x bfloat> %307, i64 0, !dbg !23 + %315 = extractelement <2 x bfloat> %307, i64 1, !dbg !23 + %316 = extractelement <2 x bfloat> %309, i64 0, !dbg !23 + %317 = extractelement <2 x bfloat> %309, i64 1, !dbg !23 + %318 = extractelement <2 x bfloat> %311, i64 0, !dbg !23 + %319 = extractelement <2 x bfloat> %311, i64 1, !dbg !23 + %320 = extractelement <16 x i1> %36, i64 3, !dbg !23 + %321 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %100, i1 %320) #6, !dbg !23 + %322 = extractvalue { i32, i32, i32, i32 } %321, 0, !dbg !23 + %323 = bitcast i32 %322 to <2 x bfloat>, !dbg !23 + %324 = extractvalue { i32, i32, i32, i32 } %321, 1, !dbg !23 + %325 = bitcast i32 %324 to <2 x bfloat>, !dbg !23 + %326 = extractvalue { i32, i32, i32, i32 } %321, 2, !dbg !23 + %327 = bitcast i32 %326 to <2 x bfloat>, !dbg !23 + %328 = extractvalue { i32, i32, i32, i32 } %321, 3, !dbg !23 + %329 = bitcast i32 %328 to <2 x bfloat>, !dbg !23 + %330 = extractelement <2 x bfloat> %323, i64 0, !dbg !23 + %331 = extractelement <2 x bfloat> %323, i64 1, !dbg !23 + %332 = extractelement <2 x bfloat> %325, i64 0, !dbg !23 + %333 = extractelement <2 x bfloat> %325, i64 1, !dbg !23 + %334 = extractelement <2 x bfloat> %327, i64 0, !dbg !23 + %335 = extractelement <2 x bfloat> %327, i64 1, !dbg !23 + %336 = extractelement <2 x bfloat> %329, i64 0, !dbg !23 + %337 = extractelement <2 x bfloat> %329, i64 1, !dbg !23 + %338 = extractelement <16 x i1> %36, i64 2, !dbg !23 + %339 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %101, i1 %338) #6, !dbg !23 + %340 = extractvalue { i32, i32, i32, i32 } %339, 0, !dbg !23 + %341 = bitcast i32 %340 to <2 x bfloat>, !dbg !23 + %342 = extractvalue { i32, i32, i32, i32 } %339, 1, !dbg !23 + %343 = bitcast i32 %342 to <2 x bfloat>, !dbg !23 + %344 = extractvalue { i32, i32, i32, i32 } %339, 2, !dbg !23 + %345 = bitcast i32 %344 to <2 x bfloat>, !dbg !23 + %346 = extractvalue { i32, i32, i32, i32 } %339, 3, !dbg !23 + %347 = bitcast i32 %346 to <2 x bfloat>, !dbg !23 + %348 = extractelement <2 x bfloat> %341, i64 0, !dbg !23 + %349 = extractelement <2 x bfloat> %341, i64 1, !dbg !23 + %350 = extractelement <2 x bfloat> %343, i64 0, !dbg !23 + %351 = extractelement <2 x bfloat> %343, i64 1, !dbg !23 + %352 = extractelement <2 x bfloat> %345, i64 0, !dbg !23 + %353 = extractelement <2 x bfloat> %345, i64 1, !dbg !23 + %354 = extractelement <2 x bfloat> %347, i64 0, !dbg !23 + %355 = extractelement <2 x bfloat> %347, i64 1, !dbg !23 + %356 = extractelement <16 x i1> %36, i64 1, !dbg !23 + %357 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %102, i1 %356) #6, !dbg !23 + %358 = extractvalue { i32, i32, i32, i32 } %357, 0, !dbg !23 + %359 = bitcast i32 %358 to <2 x bfloat>, !dbg !23 + %360 = extractvalue { i32, i32, i32, i32 } %357, 1, !dbg !23 + %361 = bitcast i32 %360 to <2 x bfloat>, !dbg !23 + %362 = extractvalue { i32, i32, i32, i32 } %357, 2, !dbg !23 + %363 = bitcast i32 %362 to <2 x bfloat>, !dbg !23 + %364 = extractvalue { i32, i32, i32, i32 } %357, 3, !dbg !23 + %365 = bitcast i32 %364 to <2 x bfloat>, !dbg !23 + %366 = extractelement <2 x bfloat> %359, i64 0, !dbg !23 + %367 = extractelement <2 x bfloat> %359, i64 1, !dbg !23 + %368 = extractelement <2 x bfloat> %361, i64 0, !dbg !23 + %369 = extractelement <2 x bfloat> %361, i64 1, !dbg !23 + %370 = extractelement <2 x bfloat> %363, i64 0, !dbg !23 + %371 = extractelement <2 x bfloat> %363, i64 1, !dbg !23 + %372 = extractelement <2 x bfloat> %365, i64 0, !dbg !23 + %373 = extractelement <2 x bfloat> %365, i64 1, !dbg !23 + %374 = extractelement <16 x i1> %36, i64 0, !dbg !23 + %375 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %103, i1 %374) #6, !dbg !23 + %376 = extractvalue { i32, i32, i32, i32 } %375, 0, !dbg !23 + %377 = bitcast i32 %376 to <2 x bfloat>, !dbg !23 + %378 = extractvalue { i32, i32, i32, i32 } %375, 1, !dbg !23 + %379 = bitcast i32 %378 to <2 x bfloat>, !dbg !23 + %380 = extractvalue { i32, i32, i32, i32 } %375, 2, !dbg !23 + %381 = bitcast i32 %380 to <2 x bfloat>, !dbg !23 + %382 = extractvalue { i32, i32, i32, i32 } %375, 3, !dbg !23 + %383 = bitcast i32 %382 to <2 x bfloat>, !dbg !23 + %384 = extractelement <2 x bfloat> %377, i64 0, !dbg !23 + %385 = extractelement <2 x bfloat> %377, i64 1, !dbg !23 + %386 = extractelement <2 x bfloat> %379, i64 0, !dbg !23 + %387 = extractelement <2 x bfloat> %379, i64 1, !dbg !23 + %388 = extractelement <2 x bfloat> %381, i64 0, !dbg !23 + %389 = extractelement <2 x bfloat> %381, i64 1, !dbg !23 + %390 = extractelement <2 x bfloat> %383, i64 0, !dbg !23 + %391 = extractelement <2 x bfloat> %383, i64 1, !dbg !23 + %392 = fpext bfloat %114 to float, !dbg !24 + %393 = fpext bfloat %115 to float, !dbg !24 + %394 = fpext bfloat %116 to float, !dbg !24 + %395 = fpext bfloat %117 to float, !dbg !24 + %396 = fpext bfloat %118 to float, !dbg !24 + %397 = fpext bfloat %119 to float, !dbg !24 + %398 = fpext bfloat %120 to float, !dbg !24 + %399 = fpext bfloat %121 to float, !dbg !24 + %400 = fpext bfloat %132 to float, !dbg !24 + %401 = fpext bfloat %133 to float, !dbg !24 + %402 = fpext bfloat %134 to float, !dbg !24 + %403 = fpext bfloat %135 to float, !dbg !24 + %404 = fpext bfloat %136 to float, !dbg !24 + %405 = fpext bfloat %137 to float, !dbg !24 + %406 = fpext bfloat %138 to float, !dbg !24 + %407 = fpext bfloat %139 to float, !dbg !24 + %408 = fpext bfloat %150 to float, !dbg !24 + %409 = fpext bfloat %151 to float, !dbg !24 + %410 = fpext bfloat %152 to float, !dbg !24 + %411 = fpext bfloat %153 to float, !dbg !24 + %412 = fpext bfloat %154 to float, !dbg !24 + %413 = fpext bfloat %155 to float, !dbg !24 + %414 = fpext bfloat %156 to float, !dbg !24 + %415 = fpext bfloat %157 to float, !dbg !24 + %416 = fpext bfloat %168 to float, !dbg !24 + %417 = fpext bfloat %169 to float, !dbg !24 + %418 = fpext bfloat %170 to float, !dbg !24 + %419 = fpext bfloat %171 to float, !dbg !24 + %420 = fpext bfloat %172 to float, !dbg !24 + %421 = fpext bfloat %173 to float, !dbg !24 + %422 = fpext bfloat %174 to float, !dbg !24 + %423 = fpext bfloat %175 to float, !dbg !24 + %424 = fpext bfloat %186 to float, !dbg !24 + %425 = fpext bfloat %187 to float, !dbg !24 + %426 = fpext bfloat %188 to float, !dbg !24 + %427 = fpext bfloat %189 to float, !dbg !24 + %428 = fpext bfloat %190 to float, !dbg !24 + %429 = fpext bfloat %191 to float, !dbg !24 + %430 = fpext bfloat %192 to float, !dbg !24 + %431 = fpext bfloat %193 to float, !dbg !24 + %432 = fpext bfloat %204 to float, !dbg !24 + %433 = fpext bfloat %205 to float, !dbg !24 + %434 = fpext bfloat %206 to float, !dbg !24 + %435 = fpext bfloat %207 to float, !dbg !24 + %436 = fpext bfloat %208 to float, !dbg !24 + %437 = fpext bfloat %209 to float, !dbg !24 + %438 = fpext bfloat %210 to float, !dbg !24 + %439 = fpext bfloat %211 to float, !dbg !24 + %440 = fpext bfloat %222 to float, !dbg !24 + %441 = fpext bfloat %223 to float, !dbg !24 + %442 = fpext bfloat %224 to float, !dbg !24 + %443 = fpext bfloat %225 to float, !dbg !24 + %444 = fpext bfloat %226 to float, !dbg !24 + %445 = fpext bfloat %227 to float, !dbg !24 + %446 = fpext bfloat %228 to float, !dbg !24 + %447 = fpext bfloat %229 to float, !dbg !24 + %448 = fpext bfloat %240 to float, !dbg !24 + %449 = fpext bfloat %241 to float, !dbg !24 + %450 = fpext bfloat %242 to float, !dbg !24 + %451 = fpext bfloat %243 to float, !dbg !24 + %452 = fpext bfloat %244 to float, !dbg !24 + %453 = fpext bfloat %245 to float, !dbg !24 + %454 = fpext bfloat %246 to float, !dbg !24 + %455 = fpext bfloat %247 to float, !dbg !24 + %456 = fpext bfloat %258 to float, !dbg !24 + %457 = fpext bfloat %259 to float, !dbg !24 + %458 = fpext bfloat %260 to float, !dbg !24 + %459 = fpext bfloat %261 to float, !dbg !24 + %460 = fpext bfloat %262 to float, !dbg !24 + %461 = fpext bfloat %263 to float, !dbg !24 + %462 = fpext bfloat %264 to float, !dbg !24 + %463 = fpext bfloat %265 to float, !dbg !24 + %464 = fpext bfloat %276 to float, !dbg !24 + %465 = fpext bfloat %277 to float, !dbg !24 + %466 = fpext bfloat %278 to float, !dbg !24 + %467 = fpext bfloat %279 to float, !dbg !24 + %468 = fpext bfloat %280 to float, !dbg !24 + %469 = fpext bfloat %281 to float, !dbg !24 + %470 = fpext bfloat %282 to float, !dbg !24 + %471 = fpext bfloat %283 to float, !dbg !24 + %472 = fpext bfloat %294 to float, !dbg !24 + %473 = fpext bfloat %295 to float, !dbg !24 + %474 = fpext bfloat %296 to float, !dbg !24 + %475 = fpext bfloat %297 to float, !dbg !24 + %476 = fpext bfloat %298 to float, !dbg !24 + %477 = fpext bfloat %299 to float, !dbg !24 + %478 = fpext bfloat %300 to float, !dbg !24 + %479 = fpext bfloat %301 to float, !dbg !24 + %480 = fpext bfloat %312 to float, !dbg !24 + %481 = fpext bfloat %313 to float, !dbg !24 + %482 = fpext bfloat %314 to float, !dbg !24 + %483 = fpext bfloat %315 to float, !dbg !24 + %484 = fpext bfloat %316 to float, !dbg !24 + %485 = fpext bfloat %317 to float, !dbg !24 + %486 = fpext bfloat %318 to float, !dbg !24 + %487 = fpext bfloat %319 to float, !dbg !24 + %488 = fpext bfloat %330 to float, !dbg !24 + %489 = fpext bfloat %331 to float, !dbg !24 + %490 = fpext bfloat %332 to float, !dbg !24 + %491 = fpext bfloat %333 to float, !dbg !24 + %492 = fpext bfloat %334 to float, !dbg !24 + %493 = fpext bfloat %335 to float, !dbg !24 + %494 = fpext bfloat %336 to float, !dbg !24 + %495 = fpext bfloat %337 to float, !dbg !24 + %496 = fpext bfloat %348 to float, !dbg !24 + %497 = fpext bfloat %349 to float, !dbg !24 + %498 = fpext bfloat %350 to float, !dbg !24 + %499 = fpext bfloat %351 to float, !dbg !24 + %500 = fpext bfloat %352 to float, !dbg !24 + %501 = fpext bfloat %353 to float, !dbg !24 + %502 = fpext bfloat %354 to float, !dbg !24 + %503 = fpext bfloat %355 to float, !dbg !24 + %504 = fpext bfloat %366 to float, !dbg !24 + %505 = fpext bfloat %367 to float, !dbg !24 + %506 = fpext bfloat %368 to float, !dbg !24 + %507 = fpext bfloat %369 to float, !dbg !24 + %508 = fpext bfloat %370 to float, !dbg !24 + %509 = fpext bfloat %371 to float, !dbg !24 + %510 = fpext bfloat %372 to float, !dbg !24 + %511 = fpext bfloat %373 to float, !dbg !24 + %512 = fpext bfloat %384 to float, !dbg !24 + %513 = fpext bfloat %385 to float, !dbg !24 + %514 = fpext bfloat %386 to float, !dbg !24 + %515 = fpext bfloat %387 to float, !dbg !24 + %516 = fpext bfloat %388 to float, !dbg !24 + %517 = fpext bfloat %389 to float, !dbg !24 + %518 = fpext bfloat %390 to float, !dbg !24 + %519 = fpext bfloat %391 to float, !dbg !24 + %520 = fmul float %392, %392, !dbg !25 + %521 = fmul float %393, %393, !dbg !25 + %522 = fmul float %394, %394, !dbg !25 + %523 = fmul float %395, %395, !dbg !25 + %524 = fmul float %396, %396, !dbg !25 + %525 = fmul float %397, %397, !dbg !25 + %526 = fmul float %398, %398, !dbg !25 + %527 = fmul float %399, %399, !dbg !25 + %528 = fmul float %400, %400, !dbg !25 + %529 = fmul float %401, %401, !dbg !25 + %530 = fmul float %402, %402, !dbg !25 + %531 = fmul float %403, %403, !dbg !25 + %532 = fmul float %404, %404, !dbg !25 + %533 = fmul float %405, %405, !dbg !25 + %534 = fmul float %406, %406, !dbg !25 + %535 = fmul float %407, %407, !dbg !25 + %536 = fmul float %408, %408, !dbg !25 + %537 = fmul float %409, %409, !dbg !25 + %538 = fmul float %410, %410, !dbg !25 + %539 = fmul float %411, %411, !dbg !25 + %540 = fmul float %412, %412, !dbg !25 + %541 = fmul float %413, %413, !dbg !25 + %542 = fmul float %414, %414, !dbg !25 + %543 = fmul float %415, %415, !dbg !25 + %544 = fmul float %416, %416, !dbg !25 + %545 = fmul float %417, %417, !dbg !25 + %546 = fmul float %418, %418, !dbg !25 + %547 = fmul float %419, %419, !dbg !25 + %548 = fmul float %420, %420, !dbg !25 + %549 = fmul float %421, %421, !dbg !25 + %550 = fmul float %422, %422, !dbg !25 + %551 = fmul float %423, %423, !dbg !25 + %552 = fmul float %424, %424, !dbg !25 + %553 = fmul float %425, %425, !dbg !25 + %554 = fmul float %426, %426, !dbg !25 + %555 = fmul float %427, %427, !dbg !25 + %556 = fmul float %428, %428, !dbg !25 + %557 = fmul float %429, %429, !dbg !25 + %558 = fmul float %430, %430, !dbg !25 + %559 = fmul float %431, %431, !dbg !25 + %560 = fmul float %432, %432, !dbg !25 + %561 = fmul float %433, %433, !dbg !25 + %562 = fmul float %434, %434, !dbg !25 + %563 = fmul float %435, %435, !dbg !25 + %564 = fmul float %436, %436, !dbg !25 + %565 = fmul float %437, %437, !dbg !25 + %566 = fmul float %438, %438, !dbg !25 + %567 = fmul float %439, %439, !dbg !25 + %568 = fmul float %440, %440, !dbg !25 + %569 = fmul float %441, %441, !dbg !25 + %570 = fmul float %442, %442, !dbg !25 + %571 = fmul float %443, %443, !dbg !25 + %572 = fmul float %444, %444, !dbg !25 + %573 = fmul float %445, %445, !dbg !25 + %574 = fmul float %446, %446, !dbg !25 + %575 = fmul float %447, %447, !dbg !25 + %576 = fmul float %448, %448, !dbg !25 + %577 = fmul float %449, %449, !dbg !25 + %578 = fmul float %450, %450, !dbg !25 + %579 = fmul float %451, %451, !dbg !25 + %580 = fmul float %452, %452, !dbg !25 + %581 = fmul float %453, %453, !dbg !25 + %582 = fmul float %454, %454, !dbg !25 + %583 = fmul float %455, %455, !dbg !25 + %584 = fmul float %456, %456, !dbg !25 + %585 = fmul float %457, %457, !dbg !25 + %586 = fmul float %458, %458, !dbg !25 + %587 = fmul float %459, %459, !dbg !25 + %588 = fmul float %460, %460, !dbg !25 + %589 = fmul float %461, %461, !dbg !25 + %590 = fmul float %462, %462, !dbg !25 + %591 = fmul float %463, %463, !dbg !25 + %592 = fmul float %464, %464, !dbg !25 + %593 = fmul float %465, %465, !dbg !25 + %594 = fmul float %466, %466, !dbg !25 + %595 = fmul float %467, %467, !dbg !25 + %596 = fmul float %468, %468, !dbg !25 + %597 = fmul float %469, %469, !dbg !25 + %598 = fmul float %470, %470, !dbg !25 + %599 = fmul float %471, %471, !dbg !25 + %600 = fmul float %472, %472, !dbg !25 + %601 = fmul float %473, %473, !dbg !25 + %602 = fmul float %474, %474, !dbg !25 + %603 = fmul float %475, %475, !dbg !25 + %604 = fmul float %476, %476, !dbg !25 + %605 = fmul float %477, %477, !dbg !25 + %606 = fmul float %478, %478, !dbg !25 + %607 = fmul float %479, %479, !dbg !25 + %608 = fmul float %480, %480, !dbg !25 + %609 = fmul float %481, %481, !dbg !25 + %610 = fmul float %482, %482, !dbg !25 + %611 = fmul float %483, %483, !dbg !25 + %612 = fmul float %484, %484, !dbg !25 + %613 = fmul float %485, %485, !dbg !25 + %614 = fmul float %486, %486, !dbg !25 + %615 = fmul float %487, %487, !dbg !25 + %616 = fmul float %488, %488, !dbg !25 + %617 = fmul float %489, %489, !dbg !25 + %618 = fmul float %490, %490, !dbg !25 + %619 = fmul float %491, %491, !dbg !25 + %620 = fmul float %492, %492, !dbg !25 + %621 = fmul float %493, %493, !dbg !25 + %622 = fmul float %494, %494, !dbg !25 + %623 = fmul float %495, %495, !dbg !25 + %624 = fmul float %496, %496, !dbg !25 + %625 = fmul float %497, %497, !dbg !25 + %626 = fmul float %498, %498, !dbg !25 + %627 = fmul float %499, %499, !dbg !25 + %628 = fmul float %500, %500, !dbg !25 + %629 = fmul float %501, %501, !dbg !25 + %630 = fmul float %502, %502, !dbg !25 + %631 = fmul float %503, %503, !dbg !25 + %632 = fmul float %504, %504, !dbg !25 + %633 = fmul float %505, %505, !dbg !25 + %634 = fmul float %506, %506, !dbg !25 + %635 = fmul float %507, %507, !dbg !25 + %636 = fmul float %508, %508, !dbg !25 + %637 = fmul float %509, %509, !dbg !25 + %638 = fmul float %510, %510, !dbg !25 + %639 = fmul float %511, %511, !dbg !25 + %640 = fmul float %512, %512, !dbg !25 + %641 = fmul float %513, %513, !dbg !25 + %642 = fmul float %514, %514, !dbg !25 + %643 = fmul float %515, %515, !dbg !25 + %644 = fmul float %516, %516, !dbg !25 + %645 = fmul float %517, %517, !dbg !25 + %646 = fmul float %518, %518, !dbg !25 + %647 = fmul float %519, %519, !dbg !25 + %648 = fadd float %520, %521, !dbg !26 + %649 = fadd float %522, %648, !dbg !26 + %650 = fadd float %523, %649, !dbg !26 + %651 = fadd float %524, %650, !dbg !26 + %652 = fadd float %525, %651, !dbg !26 + %653 = fadd float %526, %652, !dbg !26 + %654 = fadd float %527, %653, !dbg !26 + %655 = fadd float %528, %529, !dbg !26 + %656 = fadd float %530, %655, !dbg !26 + %657 = fadd float %531, %656, !dbg !26 + %658 = fadd float %532, %657, !dbg !26 + %659 = fadd float %533, %658, !dbg !26 + %660 = fadd float %534, %659, !dbg !26 + %661 = fadd float %535, %660, !dbg !26 + %662 = fadd float %536, %537, !dbg !26 + %663 = fadd float %538, %662, !dbg !26 + %664 = fadd float %539, %663, !dbg !26 + %665 = fadd float %540, %664, !dbg !26 + %666 = fadd float %541, %665, !dbg !26 + %667 = fadd float %542, %666, !dbg !26 + %668 = fadd float %543, %667, !dbg !26 + %669 = fadd float %544, %545, !dbg !26 + %670 = fadd float %546, %669, !dbg !26 + %671 = fadd float %547, %670, !dbg !26 + %672 = fadd float %548, %671, !dbg !26 + %673 = fadd float %549, %672, !dbg !26 + %674 = fadd float %550, %673, !dbg !26 + %675 = fadd float %551, %674, !dbg !26 + %676 = fadd float %552, %553, !dbg !26 + %677 = fadd float %554, %676, !dbg !26 + %678 = fadd float %555, %677, !dbg !26 + %679 = fadd float %556, %678, !dbg !26 + %680 = fadd float %557, %679, !dbg !26 + %681 = fadd float %558, %680, !dbg !26 + %682 = fadd float %559, %681, !dbg !26 + %683 = fadd float %560, %561, !dbg !26 + %684 = fadd float %562, %683, !dbg !26 + %685 = fadd float %563, %684, !dbg !26 + %686 = fadd float %564, %685, !dbg !26 + %687 = fadd float %565, %686, !dbg !26 + %688 = fadd float %566, %687, !dbg !26 + %689 = fadd float %567, %688, !dbg !26 + %690 = fadd float %568, %569, !dbg !26 + %691 = fadd float %570, %690, !dbg !26 + %692 = fadd float %571, %691, !dbg !26 + %693 = fadd float %572, %692, !dbg !26 + %694 = fadd float %573, %693, !dbg !26 + %695 = fadd float %574, %694, !dbg !26 + %696 = fadd float %575, %695, !dbg !26 + %697 = fadd float %576, %577, !dbg !26 + %698 = fadd float %578, %697, !dbg !26 + %699 = fadd float %579, %698, !dbg !26 + %700 = fadd float %580, %699, !dbg !26 + %701 = fadd float %581, %700, !dbg !26 + %702 = fadd float %582, %701, !dbg !26 + %703 = fadd float %583, %702, !dbg !26 + %704 = fadd float %584, %585, !dbg !26 + %705 = fadd float %586, %704, !dbg !26 + %706 = fadd float %587, %705, !dbg !26 + %707 = fadd float %588, %706, !dbg !26 + %708 = fadd float %589, %707, !dbg !26 + %709 = fadd float %590, %708, !dbg !26 + %710 = fadd float %591, %709, !dbg !26 + %711 = fadd float %592, %593, !dbg !26 + %712 = fadd float %594, %711, !dbg !26 + %713 = fadd float %595, %712, !dbg !26 + %714 = fadd float %596, %713, !dbg !26 + %715 = fadd float %597, %714, !dbg !26 + %716 = fadd float %598, %715, !dbg !26 + %717 = fadd float %599, %716, !dbg !26 + %718 = fadd float %600, %601, !dbg !26 + %719 = fadd float %602, %718, !dbg !26 + %720 = fadd float %603, %719, !dbg !26 + %721 = fadd float %604, %720, !dbg !26 + %722 = fadd float %605, %721, !dbg !26 + %723 = fadd float %606, %722, !dbg !26 + %724 = fadd float %607, %723, !dbg !26 + %725 = fadd float %608, %609, !dbg !26 + %726 = fadd float %610, %725, !dbg !26 + %727 = fadd float %611, %726, !dbg !26 + %728 = fadd float %612, %727, !dbg !26 + %729 = fadd float %613, %728, !dbg !26 + %730 = fadd float %614, %729, !dbg !26 + %731 = fadd float %615, %730, !dbg !26 + %732 = fadd float %616, %617, !dbg !26 + %733 = fadd float %618, %732, !dbg !26 + %734 = fadd float %619, %733, !dbg !26 + %735 = fadd float %620, %734, !dbg !26 + %736 = fadd float %621, %735, !dbg !26 + %737 = fadd float %622, %736, !dbg !26 + %738 = fadd float %623, %737, !dbg !26 + %739 = fadd float %624, %625, !dbg !26 + %740 = fadd float %626, %739, !dbg !26 + %741 = fadd float %627, %740, !dbg !26 + %742 = fadd float %628, %741, !dbg !26 + %743 = fadd float %629, %742, !dbg !26 + %744 = fadd float %630, %743, !dbg !26 + %745 = fadd float %631, %744, !dbg !26 + %746 = fadd float %632, %633, !dbg !26 + %747 = fadd float %634, %746, !dbg !26 + %748 = fadd float %635, %747, !dbg !26 + %749 = fadd float %636, %748, !dbg !26 + %750 = fadd float %637, %749, !dbg !26 + %751 = fadd float %638, %750, !dbg !26 + %752 = fadd float %639, %751, !dbg !26 + %753 = fadd float %640, %641, !dbg !26 + %754 = fadd float %642, %753, !dbg !26 + %755 = fadd float %643, %754, !dbg !26 + %756 = fadd float %644, %755, !dbg !26 + %757 = fadd float %645, %756, !dbg !26 + %758 = fadd float %646, %757, !dbg !26 + %759 = fadd float %647, %758, !dbg !26 + %760 = bitcast float %654 to i32, !dbg !29 + %761 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %760, i32 8, i32 31), !dbg !29 + %762 = bitcast i32 %761 to float, !dbg !29 + %763 = fadd float %654, %762, !dbg !26 + %764 = bitcast float %763 to i32, !dbg !29 + %765 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %764, i32 4, i32 31), !dbg !29 + %766 = bitcast i32 %765 to float, !dbg !29 + %767 = fadd float %763, %766, !dbg !26 + %768 = bitcast float %767 to i32, !dbg !29 + %769 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %768, i32 2, i32 31), !dbg !29 + %770 = bitcast i32 %769 to float, !dbg !29 + %771 = fadd float %767, %770, !dbg !26 + %772 = bitcast float %771 to i32, !dbg !29 + %773 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %772, i32 1, i32 31), !dbg !29 + %774 = bitcast i32 %773 to float, !dbg !29 + %775 = fadd float %771, %774, !dbg !26 + %776 = bitcast float %661 to i32, !dbg !29 + %777 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %776, i32 8, i32 31), !dbg !29 + %778 = bitcast i32 %777 to float, !dbg !29 + %779 = fadd float %661, %778, !dbg !26 + %780 = bitcast float %779 to i32, !dbg !29 + %781 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %780, i32 4, i32 31), !dbg !29 + %782 = bitcast i32 %781 to float, !dbg !29 + %783 = fadd float %779, %782, !dbg !26 + %784 = bitcast float %783 to i32, !dbg !29 + %785 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %784, i32 2, i32 31), !dbg !29 + %786 = bitcast i32 %785 to float, !dbg !29 + %787 = fadd float %783, %786, !dbg !26 + %788 = bitcast float %787 to i32, !dbg !29 + %789 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %788, i32 1, i32 31), !dbg !29 + %790 = bitcast i32 %789 to float, !dbg !29 + %791 = fadd float %787, %790, !dbg !26 + %792 = bitcast float %668 to i32, !dbg !29 + %793 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %792, i32 8, i32 31), !dbg !29 + %794 = bitcast i32 %793 to float, !dbg !29 + %795 = fadd float %668, %794, !dbg !26 + %796 = bitcast float %795 to i32, !dbg !29 + %797 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %796, i32 4, i32 31), !dbg !29 + %798 = bitcast i32 %797 to float, !dbg !29 + %799 = fadd float %795, %798, !dbg !26 + %800 = bitcast float %799 to i32, !dbg !29 + %801 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %800, i32 2, i32 31), !dbg !29 + %802 = bitcast i32 %801 to float, !dbg !29 + %803 = fadd float %799, %802, !dbg !26 + %804 = bitcast float %803 to i32, !dbg !29 + %805 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %804, i32 1, i32 31), !dbg !29 + %806 = bitcast i32 %805 to float, !dbg !29 + %807 = fadd float %803, %806, !dbg !26 + %808 = bitcast float %675 to i32, !dbg !29 + %809 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %808, i32 8, i32 31), !dbg !29 + %810 = bitcast i32 %809 to float, !dbg !29 + %811 = fadd float %675, %810, !dbg !26 + %812 = bitcast float %811 to i32, !dbg !29 + %813 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %812, i32 4, i32 31), !dbg !29 + %814 = bitcast i32 %813 to float, !dbg !29 + %815 = fadd float %811, %814, !dbg !26 + %816 = bitcast float %815 to i32, !dbg !29 + %817 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %816, i32 2, i32 31), !dbg !29 + %818 = bitcast i32 %817 to float, !dbg !29 + %819 = fadd float %815, %818, !dbg !26 + %820 = bitcast float %819 to i32, !dbg !29 + %821 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %820, i32 1, i32 31), !dbg !29 + %822 = bitcast i32 %821 to float, !dbg !29 + %823 = fadd float %819, %822, !dbg !26 + %824 = bitcast float %682 to i32, !dbg !29 + %825 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %824, i32 8, i32 31), !dbg !29 + %826 = bitcast i32 %825 to float, !dbg !29 + %827 = fadd float %682, %826, !dbg !26 + %828 = bitcast float %827 to i32, !dbg !29 + %829 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %828, i32 4, i32 31), !dbg !29 + %830 = bitcast i32 %829 to float, !dbg !29 + %831 = fadd float %827, %830, !dbg !26 + %832 = bitcast float %831 to i32, !dbg !29 + %833 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %832, i32 2, i32 31), !dbg !29 + %834 = bitcast i32 %833 to float, !dbg !29 + %835 = fadd float %831, %834, !dbg !26 + %836 = bitcast float %835 to i32, !dbg !29 + %837 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %836, i32 1, i32 31), !dbg !29 + %838 = bitcast i32 %837 to float, !dbg !29 + %839 = fadd float %835, %838, !dbg !26 + %840 = bitcast float %689 to i32, !dbg !29 + %841 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %840, i32 8, i32 31), !dbg !29 + %842 = bitcast i32 %841 to float, !dbg !29 + %843 = fadd float %689, %842, !dbg !26 + %844 = bitcast float %843 to i32, !dbg !29 + %845 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %844, i32 4, i32 31), !dbg !29 + %846 = bitcast i32 %845 to float, !dbg !29 + %847 = fadd float %843, %846, !dbg !26 + %848 = bitcast float %847 to i32, !dbg !29 + %849 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %848, i32 2, i32 31), !dbg !29 + %850 = bitcast i32 %849 to float, !dbg !29 + %851 = fadd float %847, %850, !dbg !26 + %852 = bitcast float %851 to i32, !dbg !29 + %853 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %852, i32 1, i32 31), !dbg !29 + %854 = bitcast i32 %853 to float, !dbg !29 + %855 = fadd float %851, %854, !dbg !26 + %856 = bitcast float %696 to i32, !dbg !29 + %857 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %856, i32 8, i32 31), !dbg !29 + %858 = bitcast i32 %857 to float, !dbg !29 + %859 = fadd float %696, %858, !dbg !26 + %860 = bitcast float %859 to i32, !dbg !29 + %861 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %860, i32 4, i32 31), !dbg !29 + %862 = bitcast i32 %861 to float, !dbg !29 + %863 = fadd float %859, %862, !dbg !26 + %864 = bitcast float %863 to i32, !dbg !29 + %865 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %864, i32 2, i32 31), !dbg !29 + %866 = bitcast i32 %865 to float, !dbg !29 + %867 = fadd float %863, %866, !dbg !26 + %868 = bitcast float %867 to i32, !dbg !29 + %869 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %868, i32 1, i32 31), !dbg !29 + %870 = bitcast i32 %869 to float, !dbg !29 + %871 = fadd float %867, %870, !dbg !26 + %872 = bitcast float %703 to i32, !dbg !29 + %873 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %872, i32 8, i32 31), !dbg !29 + %874 = bitcast i32 %873 to float, !dbg !29 + %875 = fadd float %703, %874, !dbg !26 + %876 = bitcast float %875 to i32, !dbg !29 + %877 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %876, i32 4, i32 31), !dbg !29 + %878 = bitcast i32 %877 to float, !dbg !29 + %879 = fadd float %875, %878, !dbg !26 + %880 = bitcast float %879 to i32, !dbg !29 + %881 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %880, i32 2, i32 31), !dbg !29 + %882 = bitcast i32 %881 to float, !dbg !29 + %883 = fadd float %879, %882, !dbg !26 + %884 = bitcast float %883 to i32, !dbg !29 + %885 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %884, i32 1, i32 31), !dbg !29 + %886 = bitcast i32 %885 to float, !dbg !29 + %887 = fadd float %883, %886, !dbg !26 + %888 = bitcast float %710 to i32, !dbg !29 + %889 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %888, i32 8, i32 31), !dbg !29 + %890 = bitcast i32 %889 to float, !dbg !29 + %891 = fadd float %710, %890, !dbg !26 + %892 = bitcast float %891 to i32, !dbg !29 + %893 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %892, i32 4, i32 31), !dbg !29 + %894 = bitcast i32 %893 to float, !dbg !29 + %895 = fadd float %891, %894, !dbg !26 + %896 = bitcast float %895 to i32, !dbg !29 + %897 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %896, i32 2, i32 31), !dbg !29 + %898 = bitcast i32 %897 to float, !dbg !29 + %899 = fadd float %895, %898, !dbg !26 + %900 = bitcast float %899 to i32, !dbg !29 + %901 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %900, i32 1, i32 31), !dbg !29 + %902 = bitcast i32 %901 to float, !dbg !29 + %903 = fadd float %899, %902, !dbg !26 + %904 = bitcast float %717 to i32, !dbg !29 + %905 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %904, i32 8, i32 31), !dbg !29 + %906 = bitcast i32 %905 to float, !dbg !29 + %907 = fadd float %717, %906, !dbg !26 + %908 = bitcast float %907 to i32, !dbg !29 + %909 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %908, i32 4, i32 31), !dbg !29 + %910 = bitcast i32 %909 to float, !dbg !29 + %911 = fadd float %907, %910, !dbg !26 + %912 = bitcast float %911 to i32, !dbg !29 + %913 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %912, i32 2, i32 31), !dbg !29 + %914 = bitcast i32 %913 to float, !dbg !29 + %915 = fadd float %911, %914, !dbg !26 + %916 = bitcast float %915 to i32, !dbg !29 + %917 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %916, i32 1, i32 31), !dbg !29 + %918 = bitcast i32 %917 to float, !dbg !29 + %919 = fadd float %915, %918, !dbg !26 + %920 = bitcast float %724 to i32, !dbg !29 + %921 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %920, i32 8, i32 31), !dbg !29 + %922 = bitcast i32 %921 to float, !dbg !29 + %923 = fadd float %724, %922, !dbg !26 + %924 = bitcast float %923 to i32, !dbg !29 + %925 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %924, i32 4, i32 31), !dbg !29 + %926 = bitcast i32 %925 to float, !dbg !29 + %927 = fadd float %923, %926, !dbg !26 + %928 = bitcast float %927 to i32, !dbg !29 + %929 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %928, i32 2, i32 31), !dbg !29 + %930 = bitcast i32 %929 to float, !dbg !29 + %931 = fadd float %927, %930, !dbg !26 + %932 = bitcast float %931 to i32, !dbg !29 + %933 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %932, i32 1, i32 31), !dbg !29 + %934 = bitcast i32 %933 to float, !dbg !29 + %935 = fadd float %931, %934, !dbg !26 + %936 = bitcast float %731 to i32, !dbg !29 + %937 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %936, i32 8, i32 31), !dbg !29 + %938 = bitcast i32 %937 to float, !dbg !29 + %939 = fadd float %731, %938, !dbg !26 + %940 = bitcast float %939 to i32, !dbg !29 + %941 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %940, i32 4, i32 31), !dbg !29 + %942 = bitcast i32 %941 to float, !dbg !29 + %943 = fadd float %939, %942, !dbg !26 + %944 = bitcast float %943 to i32, !dbg !29 + %945 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %944, i32 2, i32 31), !dbg !29 + %946 = bitcast i32 %945 to float, !dbg !29 + %947 = fadd float %943, %946, !dbg !26 + %948 = bitcast float %947 to i32, !dbg !29 + %949 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %948, i32 1, i32 31), !dbg !29 + %950 = bitcast i32 %949 to float, !dbg !29 + %951 = fadd float %947, %950, !dbg !26 + %952 = bitcast float %738 to i32, !dbg !29 + %953 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %952, i32 8, i32 31), !dbg !29 + %954 = bitcast i32 %953 to float, !dbg !29 + %955 = fadd float %738, %954, !dbg !26 + %956 = bitcast float %955 to i32, !dbg !29 + %957 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %956, i32 4, i32 31), !dbg !29 + %958 = bitcast i32 %957 to float, !dbg !29 + %959 = fadd float %955, %958, !dbg !26 + %960 = bitcast float %959 to i32, !dbg !29 + %961 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %960, i32 2, i32 31), !dbg !29 + %962 = bitcast i32 %961 to float, !dbg !29 + %963 = fadd float %959, %962, !dbg !26 + %964 = bitcast float %963 to i32, !dbg !29 + %965 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %964, i32 1, i32 31), !dbg !29 + %966 = bitcast i32 %965 to float, !dbg !29 + %967 = fadd float %963, %966, !dbg !26 + %968 = bitcast float %745 to i32, !dbg !29 + %969 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %968, i32 8, i32 31), !dbg !29 + %970 = bitcast i32 %969 to float, !dbg !29 + %971 = fadd float %745, %970, !dbg !26 + %972 = bitcast float %971 to i32, !dbg !29 + %973 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %972, i32 4, i32 31), !dbg !29 + %974 = bitcast i32 %973 to float, !dbg !29 + %975 = fadd float %971, %974, !dbg !26 + %976 = bitcast float %975 to i32, !dbg !29 + %977 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %976, i32 2, i32 31), !dbg !29 + %978 = bitcast i32 %977 to float, !dbg !29 + %979 = fadd float %975, %978, !dbg !26 + %980 = bitcast float %979 to i32, !dbg !29 + %981 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %980, i32 1, i32 31), !dbg !29 + %982 = bitcast i32 %981 to float, !dbg !29 + %983 = fadd float %979, %982, !dbg !26 + %984 = bitcast float %752 to i32, !dbg !29 + %985 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %984, i32 8, i32 31), !dbg !29 + %986 = bitcast i32 %985 to float, !dbg !29 + %987 = fadd float %752, %986, !dbg !26 + %988 = bitcast float %987 to i32, !dbg !29 + %989 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %988, i32 4, i32 31), !dbg !29 + %990 = bitcast i32 %989 to float, !dbg !29 + %991 = fadd float %987, %990, !dbg !26 + %992 = bitcast float %991 to i32, !dbg !29 + %993 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %992, i32 2, i32 31), !dbg !29 + %994 = bitcast i32 %993 to float, !dbg !29 + %995 = fadd float %991, %994, !dbg !26 + %996 = bitcast float %995 to i32, !dbg !29 + %997 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %996, i32 1, i32 31), !dbg !29 + %998 = bitcast i32 %997 to float, !dbg !29 + %999 = fadd float %995, %998, !dbg !26 + %1000 = bitcast float %759 to i32, !dbg !29 + %1001 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1000, i32 8, i32 31), !dbg !29 + %1002 = bitcast i32 %1001 to float, !dbg !29 + %1003 = fadd float %759, %1002, !dbg !26 + %1004 = bitcast float %1003 to i32, !dbg !29 + %1005 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1004, i32 4, i32 31), !dbg !29 + %1006 = bitcast i32 %1005 to float, !dbg !29 + %1007 = fadd float %1003, %1006, !dbg !26 + %1008 = bitcast float %1007 to i32, !dbg !29 + %1009 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1008, i32 2, i32 31), !dbg !29 + %1010 = bitcast i32 %1009 to float, !dbg !29 + %1011 = fadd float %1007, %1010, !dbg !26 + %1012 = bitcast float %1011 to i32, !dbg !29 + %1013 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1012, i32 1, i32 31), !dbg !29 + %1014 = bitcast i32 %1013 to float, !dbg !29 + %1015 = fadd float %1011, %1014, !dbg !26 + %1016 = fadd float %3, %775, !dbg !32 + %1017 = fadd float %3, %791, !dbg !32 + %1018 = fadd float %3, %807, !dbg !32 + %1019 = fadd float %3, %823, !dbg !32 + %1020 = fadd float %3, %839, !dbg !32 + %1021 = fadd float %3, %855, !dbg !32 + %1022 = fadd float %3, %871, !dbg !32 + %1023 = fadd float %3, %887, !dbg !32 + %1024 = fadd float %3, %903, !dbg !32 + %1025 = fadd float %3, %919, !dbg !32 + %1026 = fadd float %3, %935, !dbg !32 + %1027 = fadd float %3, %951, !dbg !32 + %1028 = fadd float %3, %967, !dbg !32 + %1029 = fadd float %3, %983, !dbg !32 + %1030 = fadd float %3, %999, !dbg !32 + %1031 = fadd float %3, %1015, !dbg !32 + %1032 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i = icmp eq i32 %1032, 0, !dbg !33 + %1033 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i = icmp eq i32 %1033, 0, !dbg !33 + br i1 %.not.i, label %1039, label %1034, !dbg !33 + +1034: ; preds = %7 + br i1 %.not1.i, label %1037, label %1035, !dbg !33 + +1035: ; preds = %1034 + %1036 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1016) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +1037: ; preds = %1034 + %1038 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1016) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +1039: ; preds = %7 + br i1 %.not1.i, label %1042, label %1040, !dbg !33 + +1040: ; preds = %1039 + %1041 = tail call float @llvm.nvvm.sqrt.rn.f(float %1016) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +1042: ; preds = %1039 + %1043 = tail call float @llvm.nvvm.sqrt.approx.f(float %1016) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +__nv_sqrtf.exit: ; preds = %1035, %1037, %1040, %1042 + %.0.i = phi float [ %1036, %1035 ], [ %1038, %1037 ], [ %1041, %1040 ], [ %1043, %1042 ], !dbg !33 + %1044 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i1 = icmp eq i32 %1044, 0, !dbg !33 + %1045 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i4 = icmp eq i32 %1045, 0, !dbg !33 + br i1 %.not.i1, label %1051, label %1046, !dbg !33 + +1046: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %1049, label %1047, !dbg !33 + +1047: ; preds = %1046 + %1048 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1017) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +1049: ; preds = %1046 + %1050 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1017) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +1051: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %1054, label %1052, !dbg !33 + +1052: ; preds = %1051 + %1053 = tail call float @llvm.nvvm.sqrt.rn.f(float %1017) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +1054: ; preds = %1051 + %1055 = tail call float @llvm.nvvm.sqrt.approx.f(float %1017) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +__nv_sqrtf.exit5: ; preds = %1047, %1049, %1052, %1054 + %.0.i3 = phi float [ %1048, %1047 ], [ %1050, %1049 ], [ %1053, %1052 ], [ %1055, %1054 ], !dbg !33 + %1056 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i6 = icmp eq i32 %1056, 0, !dbg !33 + %1057 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i9 = icmp eq i32 %1057, 0, !dbg !33 + br i1 %.not.i6, label %1063, label %1058, !dbg !33 + +1058: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %1061, label %1059, !dbg !33 + +1059: ; preds = %1058 + %1060 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1018) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +1061: ; preds = %1058 + %1062 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1018) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +1063: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %1066, label %1064, !dbg !33 + +1064: ; preds = %1063 + %1065 = tail call float @llvm.nvvm.sqrt.rn.f(float %1018) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +1066: ; preds = %1063 + %1067 = tail call float @llvm.nvvm.sqrt.approx.f(float %1018) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +__nv_sqrtf.exit10: ; preds = %1059, %1061, %1064, %1066 + %.0.i8 = phi float [ %1060, %1059 ], [ %1062, %1061 ], [ %1065, %1064 ], [ %1067, %1066 ], !dbg !33 + %1068 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i11 = icmp eq i32 %1068, 0, !dbg !33 + %1069 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i14 = icmp eq i32 %1069, 0, !dbg !33 + br i1 %.not.i11, label %1075, label %1070, !dbg !33 + +1070: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %1073, label %1071, !dbg !33 + +1071: ; preds = %1070 + %1072 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1019) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +1073: ; preds = %1070 + %1074 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1019) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +1075: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %1078, label %1076, !dbg !33 + +1076: ; preds = %1075 + %1077 = tail call float @llvm.nvvm.sqrt.rn.f(float %1019) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +1078: ; preds = %1075 + %1079 = tail call float @llvm.nvvm.sqrt.approx.f(float %1019) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +__nv_sqrtf.exit15: ; preds = %1071, %1073, %1076, %1078 + %.0.i13 = phi float [ %1072, %1071 ], [ %1074, %1073 ], [ %1077, %1076 ], [ %1079, %1078 ], !dbg !33 + %1080 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i16 = icmp eq i32 %1080, 0, !dbg !33 + %1081 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i19 = icmp eq i32 %1081, 0, !dbg !33 + br i1 %.not.i16, label %1087, label %1082, !dbg !33 + +1082: ; preds = %__nv_sqrtf.exit15 + br i1 %.not1.i19, label %1085, label %1083, !dbg !33 + +1083: ; preds = %1082 + %1084 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1020) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +1085: ; preds = %1082 + %1086 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1020) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +1087: ; preds = %__nv_sqrtf.exit15 + br i1 %.not1.i19, label %1090, label %1088, !dbg !33 + +1088: ; preds = %1087 + %1089 = tail call float @llvm.nvvm.sqrt.rn.f(float %1020) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +1090: ; preds = %1087 + %1091 = tail call float @llvm.nvvm.sqrt.approx.f(float %1020) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +__nv_sqrtf.exit20: ; preds = %1083, %1085, %1088, %1090 + %.0.i18 = phi float [ %1084, %1083 ], [ %1086, %1085 ], [ %1089, %1088 ], [ %1091, %1090 ], !dbg !33 + %1092 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i21 = icmp eq i32 %1092, 0, !dbg !33 + %1093 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i24 = icmp eq i32 %1093, 0, !dbg !33 + br i1 %.not.i21, label %1099, label %1094, !dbg !33 + +1094: ; preds = %__nv_sqrtf.exit20 + br i1 %.not1.i24, label %1097, label %1095, !dbg !33 + +1095: ; preds = %1094 + %1096 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1021) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +1097: ; preds = %1094 + %1098 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1021) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +1099: ; preds = %__nv_sqrtf.exit20 + br i1 %.not1.i24, label %1102, label %1100, !dbg !33 + +1100: ; preds = %1099 + %1101 = tail call float @llvm.nvvm.sqrt.rn.f(float %1021) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +1102: ; preds = %1099 + %1103 = tail call float @llvm.nvvm.sqrt.approx.f(float %1021) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +__nv_sqrtf.exit25: ; preds = %1095, %1097, %1100, %1102 + %.0.i23 = phi float [ %1096, %1095 ], [ %1098, %1097 ], [ %1101, %1100 ], [ %1103, %1102 ], !dbg !33 + %1104 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i26 = icmp eq i32 %1104, 0, !dbg !33 + %1105 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i29 = icmp eq i32 %1105, 0, !dbg !33 + br i1 %.not.i26, label %1111, label %1106, !dbg !33 + +1106: ; preds = %__nv_sqrtf.exit25 + br i1 %.not1.i29, label %1109, label %1107, !dbg !33 + +1107: ; preds = %1106 + %1108 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1022) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +1109: ; preds = %1106 + %1110 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1022) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +1111: ; preds = %__nv_sqrtf.exit25 + br i1 %.not1.i29, label %1114, label %1112, !dbg !33 + +1112: ; preds = %1111 + %1113 = tail call float @llvm.nvvm.sqrt.rn.f(float %1022) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +1114: ; preds = %1111 + %1115 = tail call float @llvm.nvvm.sqrt.approx.f(float %1022) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +__nv_sqrtf.exit30: ; preds = %1107, %1109, %1112, %1114 + %.0.i28 = phi float [ %1108, %1107 ], [ %1110, %1109 ], [ %1113, %1112 ], [ %1115, %1114 ], !dbg !33 + %1116 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i31 = icmp eq i32 %1116, 0, !dbg !33 + %1117 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i34 = icmp eq i32 %1117, 0, !dbg !33 + br i1 %.not.i31, label %1123, label %1118, !dbg !33 + +1118: ; preds = %__nv_sqrtf.exit30 + br i1 %.not1.i34, label %1121, label %1119, !dbg !33 + +1119: ; preds = %1118 + %1120 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1023) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +1121: ; preds = %1118 + %1122 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1023) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +1123: ; preds = %__nv_sqrtf.exit30 + br i1 %.not1.i34, label %1126, label %1124, !dbg !33 + +1124: ; preds = %1123 + %1125 = tail call float @llvm.nvvm.sqrt.rn.f(float %1023) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +1126: ; preds = %1123 + %1127 = tail call float @llvm.nvvm.sqrt.approx.f(float %1023) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +__nv_sqrtf.exit35: ; preds = %1119, %1121, %1124, %1126 + %.0.i33 = phi float [ %1120, %1119 ], [ %1122, %1121 ], [ %1125, %1124 ], [ %1127, %1126 ], !dbg !33 + %1128 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i36 = icmp eq i32 %1128, 0, !dbg !33 + %1129 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i39 = icmp eq i32 %1129, 0, !dbg !33 + br i1 %.not.i36, label %1135, label %1130, !dbg !33 + +1130: ; preds = %__nv_sqrtf.exit35 + br i1 %.not1.i39, label %1133, label %1131, !dbg !33 + +1131: ; preds = %1130 + %1132 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1024) #6, !dbg !33 + br label %__nv_sqrtf.exit40, !dbg !33 + +1133: ; preds = %1130 + %1134 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1024) #6, !dbg !33 + br label %__nv_sqrtf.exit40, !dbg !33 + +1135: ; preds = %__nv_sqrtf.exit35 + br i1 %.not1.i39, label %1138, label %1136, !dbg !33 + +1136: ; preds = %1135 + %1137 = tail call float @llvm.nvvm.sqrt.rn.f(float %1024) #6, !dbg !33 + br label %__nv_sqrtf.exit40, !dbg !33 + +1138: ; preds = %1135 + %1139 = tail call float @llvm.nvvm.sqrt.approx.f(float %1024) #6, !dbg !33 + br label %__nv_sqrtf.exit40, !dbg !33 + +__nv_sqrtf.exit40: ; preds = %1131, %1133, %1136, %1138 + %.0.i38 = phi float [ %1132, %1131 ], [ %1134, %1133 ], [ %1137, %1136 ], [ %1139, %1138 ], !dbg !33 + %1140 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i41 = icmp eq i32 %1140, 0, !dbg !33 + %1141 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i44 = icmp eq i32 %1141, 0, !dbg !33 + br i1 %.not.i41, label %1147, label %1142, !dbg !33 + +1142: ; preds = %__nv_sqrtf.exit40 + br i1 %.not1.i44, label %1145, label %1143, !dbg !33 + +1143: ; preds = %1142 + %1144 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1025) #6, !dbg !33 + br label %__nv_sqrtf.exit45, !dbg !33 + +1145: ; preds = %1142 + %1146 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1025) #6, !dbg !33 + br label %__nv_sqrtf.exit45, !dbg !33 + +1147: ; preds = %__nv_sqrtf.exit40 + br i1 %.not1.i44, label %1150, label %1148, !dbg !33 + +1148: ; preds = %1147 + %1149 = tail call float @llvm.nvvm.sqrt.rn.f(float %1025) #6, !dbg !33 + br label %__nv_sqrtf.exit45, !dbg !33 + +1150: ; preds = %1147 + %1151 = tail call float @llvm.nvvm.sqrt.approx.f(float %1025) #6, !dbg !33 + br label %__nv_sqrtf.exit45, !dbg !33 + +__nv_sqrtf.exit45: ; preds = %1143, %1145, %1148, %1150 + %.0.i43 = phi float [ %1144, %1143 ], [ %1146, %1145 ], [ %1149, %1148 ], [ %1151, %1150 ], !dbg !33 + %1152 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i46 = icmp eq i32 %1152, 0, !dbg !33 + %1153 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i49 = icmp eq i32 %1153, 0, !dbg !33 + br i1 %.not.i46, label %1159, label %1154, !dbg !33 + +1154: ; preds = %__nv_sqrtf.exit45 + br i1 %.not1.i49, label %1157, label %1155, !dbg !33 + +1155: ; preds = %1154 + %1156 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1026) #6, !dbg !33 + br label %__nv_sqrtf.exit50, !dbg !33 + +1157: ; preds = %1154 + %1158 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1026) #6, !dbg !33 + br label %__nv_sqrtf.exit50, !dbg !33 + +1159: ; preds = %__nv_sqrtf.exit45 + br i1 %.not1.i49, label %1162, label %1160, !dbg !33 + +1160: ; preds = %1159 + %1161 = tail call float @llvm.nvvm.sqrt.rn.f(float %1026) #6, !dbg !33 + br label %__nv_sqrtf.exit50, !dbg !33 + +1162: ; preds = %1159 + %1163 = tail call float @llvm.nvvm.sqrt.approx.f(float %1026) #6, !dbg !33 + br label %__nv_sqrtf.exit50, !dbg !33 + +__nv_sqrtf.exit50: ; preds = %1155, %1157, %1160, %1162 + %.0.i48 = phi float [ %1156, %1155 ], [ %1158, %1157 ], [ %1161, %1160 ], [ %1163, %1162 ], !dbg !33 + %1164 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i51 = icmp eq i32 %1164, 0, !dbg !33 + %1165 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i54 = icmp eq i32 %1165, 0, !dbg !33 + br i1 %.not.i51, label %1171, label %1166, !dbg !33 + +1166: ; preds = %__nv_sqrtf.exit50 + br i1 %.not1.i54, label %1169, label %1167, !dbg !33 + +1167: ; preds = %1166 + %1168 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1027) #6, !dbg !33 + br label %__nv_sqrtf.exit55, !dbg !33 + +1169: ; preds = %1166 + %1170 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1027) #6, !dbg !33 + br label %__nv_sqrtf.exit55, !dbg !33 + +1171: ; preds = %__nv_sqrtf.exit50 + br i1 %.not1.i54, label %1174, label %1172, !dbg !33 + +1172: ; preds = %1171 + %1173 = tail call float @llvm.nvvm.sqrt.rn.f(float %1027) #6, !dbg !33 + br label %__nv_sqrtf.exit55, !dbg !33 + +1174: ; preds = %1171 + %1175 = tail call float @llvm.nvvm.sqrt.approx.f(float %1027) #6, !dbg !33 + br label %__nv_sqrtf.exit55, !dbg !33 + +__nv_sqrtf.exit55: ; preds = %1167, %1169, %1172, %1174 + %.0.i53 = phi float [ %1168, %1167 ], [ %1170, %1169 ], [ %1173, %1172 ], [ %1175, %1174 ], !dbg !33 + %1176 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i56 = icmp eq i32 %1176, 0, !dbg !33 + %1177 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i59 = icmp eq i32 %1177, 0, !dbg !33 + br i1 %.not.i56, label %1183, label %1178, !dbg !33 + +1178: ; preds = %__nv_sqrtf.exit55 + br i1 %.not1.i59, label %1181, label %1179, !dbg !33 + +1179: ; preds = %1178 + %1180 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1028) #6, !dbg !33 + br label %__nv_sqrtf.exit60, !dbg !33 + +1181: ; preds = %1178 + %1182 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1028) #6, !dbg !33 + br label %__nv_sqrtf.exit60, !dbg !33 + +1183: ; preds = %__nv_sqrtf.exit55 + br i1 %.not1.i59, label %1186, label %1184, !dbg !33 + +1184: ; preds = %1183 + %1185 = tail call float @llvm.nvvm.sqrt.rn.f(float %1028) #6, !dbg !33 + br label %__nv_sqrtf.exit60, !dbg !33 + +1186: ; preds = %1183 + %1187 = tail call float @llvm.nvvm.sqrt.approx.f(float %1028) #6, !dbg !33 + br label %__nv_sqrtf.exit60, !dbg !33 + +__nv_sqrtf.exit60: ; preds = %1179, %1181, %1184, %1186 + %.0.i58 = phi float [ %1180, %1179 ], [ %1182, %1181 ], [ %1185, %1184 ], [ %1187, %1186 ], !dbg !33 + %1188 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i61 = icmp eq i32 %1188, 0, !dbg !33 + %1189 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i64 = icmp eq i32 %1189, 0, !dbg !33 + br i1 %.not.i61, label %1195, label %1190, !dbg !33 + +1190: ; preds = %__nv_sqrtf.exit60 + br i1 %.not1.i64, label %1193, label %1191, !dbg !33 + +1191: ; preds = %1190 + %1192 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1029) #6, !dbg !33 + br label %__nv_sqrtf.exit65, !dbg !33 + +1193: ; preds = %1190 + %1194 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1029) #6, !dbg !33 + br label %__nv_sqrtf.exit65, !dbg !33 + +1195: ; preds = %__nv_sqrtf.exit60 + br i1 %.not1.i64, label %1198, label %1196, !dbg !33 + +1196: ; preds = %1195 + %1197 = tail call float @llvm.nvvm.sqrt.rn.f(float %1029) #6, !dbg !33 + br label %__nv_sqrtf.exit65, !dbg !33 + +1198: ; preds = %1195 + %1199 = tail call float @llvm.nvvm.sqrt.approx.f(float %1029) #6, !dbg !33 + br label %__nv_sqrtf.exit65, !dbg !33 + +__nv_sqrtf.exit65: ; preds = %1191, %1193, %1196, %1198 + %.0.i63 = phi float [ %1192, %1191 ], [ %1194, %1193 ], [ %1197, %1196 ], [ %1199, %1198 ], !dbg !33 + %1200 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i66 = icmp eq i32 %1200, 0, !dbg !33 + %1201 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i69 = icmp eq i32 %1201, 0, !dbg !33 + br i1 %.not.i66, label %1207, label %1202, !dbg !33 + +1202: ; preds = %__nv_sqrtf.exit65 + br i1 %.not1.i69, label %1205, label %1203, !dbg !33 + +1203: ; preds = %1202 + %1204 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1030) #6, !dbg !33 + br label %__nv_sqrtf.exit70, !dbg !33 + +1205: ; preds = %1202 + %1206 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1030) #6, !dbg !33 + br label %__nv_sqrtf.exit70, !dbg !33 + +1207: ; preds = %__nv_sqrtf.exit65 + br i1 %.not1.i69, label %1210, label %1208, !dbg !33 + +1208: ; preds = %1207 + %1209 = tail call float @llvm.nvvm.sqrt.rn.f(float %1030) #6, !dbg !33 + br label %__nv_sqrtf.exit70, !dbg !33 + +1210: ; preds = %1207 + %1211 = tail call float @llvm.nvvm.sqrt.approx.f(float %1030) #6, !dbg !33 + br label %__nv_sqrtf.exit70, !dbg !33 + +__nv_sqrtf.exit70: ; preds = %1203, %1205, %1208, %1210 + %.0.i68 = phi float [ %1204, %1203 ], [ %1206, %1205 ], [ %1209, %1208 ], [ %1211, %1210 ], !dbg !33 + %1212 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i71 = icmp eq i32 %1212, 0, !dbg !33 + %1213 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i74 = icmp eq i32 %1213, 0, !dbg !33 + br i1 %.not.i71, label %1219, label %1214, !dbg !33 + +1214: ; preds = %__nv_sqrtf.exit70 + br i1 %.not1.i74, label %1217, label %1215, !dbg !33 + +1215: ; preds = %1214 + %1216 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1031) #6, !dbg !33 + br label %__nv_sqrtf.exit75, !dbg !33 + +1217: ; preds = %1214 + %1218 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1031) #6, !dbg !33 + br label %__nv_sqrtf.exit75, !dbg !33 + +1219: ; preds = %__nv_sqrtf.exit70 + br i1 %.not1.i74, label %1222, label %1220, !dbg !33 + +1220: ; preds = %1219 + %1221 = tail call float @llvm.nvvm.sqrt.rn.f(float %1031) #6, !dbg !33 + br label %__nv_sqrtf.exit75, !dbg !33 + +1222: ; preds = %1219 + %1223 = tail call float @llvm.nvvm.sqrt.approx.f(float %1031) #6, !dbg !33 + br label %__nv_sqrtf.exit75, !dbg !33 + +__nv_sqrtf.exit75: ; preds = %1215, %1217, %1220, %1222 + %.0.i73 = phi float [ %1216, %1215 ], [ %1218, %1217 ], [ %1221, %1220 ], [ %1223, %1222 ], !dbg !33 + %1224 = sext i32 %4 to i64, !dbg !18 + %1225 = and i32 %11, 127, !dbg !16 + %1226 = zext nneg i32 %1225 to i64, !dbg !17 + %1227 = or disjoint i64 %10, %1226, !dbg !17 + %1228 = getelementptr float, ptr addrspace(1) %2, i64 %1227, !dbg !34 + %1229 = getelementptr bfloat, ptr addrspace(1) %1, i64 %70, !dbg !35 + %1230 = getelementptr bfloat, ptr addrspace(1) %1229, i64 %87, !dbg !36 + %1231 = getelementptr bfloat, ptr addrspace(1) %1, i64 %68, !dbg !35 + %1232 = getelementptr bfloat, ptr addrspace(1) %1231, i64 %87, !dbg !36 + %1233 = getelementptr bfloat, ptr addrspace(1) %1, i64 %66, !dbg !35 + %1234 = getelementptr bfloat, ptr addrspace(1) %1233, i64 %87, !dbg !36 + %1235 = getelementptr bfloat, ptr addrspace(1) %1, i64 %64, !dbg !35 + %1236 = getelementptr bfloat, ptr addrspace(1) %1235, i64 %87, !dbg !36 + %1237 = getelementptr bfloat, ptr addrspace(1) %1, i64 %62, !dbg !35 + %1238 = getelementptr bfloat, ptr addrspace(1) %1237, i64 %87, !dbg !36 + %1239 = getelementptr bfloat, ptr addrspace(1) %1, i64 %60, !dbg !35 + %1240 = getelementptr bfloat, ptr addrspace(1) %1239, i64 %87, !dbg !36 + %1241 = getelementptr bfloat, ptr addrspace(1) %1, i64 %58, !dbg !35 + %1242 = getelementptr bfloat, ptr addrspace(1) %1241, i64 %87, !dbg !36 + %1243 = getelementptr bfloat, ptr addrspace(1) %1, i64 %56, !dbg !35 + %1244 = getelementptr bfloat, ptr addrspace(1) %1243, i64 %87, !dbg !36 + %1245 = getelementptr bfloat, ptr addrspace(1) %1, i64 %54, !dbg !35 + %1246 = getelementptr bfloat, ptr addrspace(1) %1245, i64 %87, !dbg !36 + %1247 = getelementptr bfloat, ptr addrspace(1) %1, i64 %52, !dbg !35 + %1248 = getelementptr bfloat, ptr addrspace(1) %1247, i64 %87, !dbg !36 + %1249 = getelementptr bfloat, ptr addrspace(1) %1, i64 %50, !dbg !35 + %1250 = getelementptr bfloat, ptr addrspace(1) %1249, i64 %87, !dbg !36 + %1251 = getelementptr bfloat, ptr addrspace(1) %1, i64 %48, !dbg !35 + %1252 = getelementptr bfloat, ptr addrspace(1) %1251, i64 %87, !dbg !36 + %1253 = getelementptr bfloat, ptr addrspace(1) %1, i64 %46, !dbg !35 + %1254 = getelementptr bfloat, ptr addrspace(1) %1253, i64 %87, !dbg !36 + %1255 = getelementptr bfloat, ptr addrspace(1) %1, i64 %44, !dbg !35 + %1256 = getelementptr bfloat, ptr addrspace(1) %1255, i64 %87, !dbg !36 + %1257 = getelementptr bfloat, ptr addrspace(1) %1, i64 %42, !dbg !35 + %1258 = getelementptr bfloat, ptr addrspace(1) %1257, i64 %87, !dbg !36 + %1259 = getelementptr bfloat, ptr addrspace(1) %1, i64 %40, !dbg !35 + %1260 = getelementptr bfloat, ptr addrspace(1) %1259, i64 %87, !dbg !36 + %1261 = icmp slt i64 %1227, %1224, !dbg !18 + %1262 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i), !dbg !37 + %1263 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i3), !dbg !37 + %1264 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i8), !dbg !37 + %1265 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i13), !dbg !37 + %1266 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i18), !dbg !37 + %1267 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i23), !dbg !37 + %1268 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i28), !dbg !37 + %1269 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i33), !dbg !37 + %1270 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i38), !dbg !37 + %1271 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i43), !dbg !37 + %1272 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i48), !dbg !37 + %1273 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i53), !dbg !37 + %1274 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i58), !dbg !37 + %1275 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i63), !dbg !37 + %1276 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i68), !dbg !37 + %1277 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i73), !dbg !37 + %1278 = fmul float %1262, %392, !dbg !38 + %1279 = fmul float %1262, %393, !dbg !38 + %1280 = fmul float %1262, %394, !dbg !38 + %1281 = fmul float %1262, %395, !dbg !38 + %1282 = fmul float %1262, %396, !dbg !38 + %1283 = fmul float %1262, %397, !dbg !38 + %1284 = fmul float %1262, %398, !dbg !38 + %1285 = fmul float %1262, %399, !dbg !38 + %1286 = fmul float %1263, %400, !dbg !38 + %1287 = fmul float %1263, %401, !dbg !38 + %1288 = fmul float %1263, %402, !dbg !38 + %1289 = fmul float %1263, %403, !dbg !38 + %1290 = fmul float %1263, %404, !dbg !38 + %1291 = fmul float %1263, %405, !dbg !38 + %1292 = fmul float %1263, %406, !dbg !38 + %1293 = fmul float %1263, %407, !dbg !38 + %1294 = fmul float %1264, %408, !dbg !38 + %1295 = fmul float %1264, %409, !dbg !38 + %1296 = fmul float %1264, %410, !dbg !38 + %1297 = fmul float %1264, %411, !dbg !38 + %1298 = fmul float %1264, %412, !dbg !38 + %1299 = fmul float %1264, %413, !dbg !38 + %1300 = fmul float %1264, %414, !dbg !38 + %1301 = fmul float %1264, %415, !dbg !38 + %1302 = fmul float %1265, %416, !dbg !38 + %1303 = fmul float %1265, %417, !dbg !38 + %1304 = fmul float %1265, %418, !dbg !38 + %1305 = fmul float %1265, %419, !dbg !38 + %1306 = fmul float %1265, %420, !dbg !38 + %1307 = fmul float %1265, %421, !dbg !38 + %1308 = fmul float %1265, %422, !dbg !38 + %1309 = fmul float %1265, %423, !dbg !38 + %1310 = fmul float %1266, %424, !dbg !38 + %1311 = fmul float %1266, %425, !dbg !38 + %1312 = fmul float %1266, %426, !dbg !38 + %1313 = fmul float %1266, %427, !dbg !38 + %1314 = fmul float %1266, %428, !dbg !38 + %1315 = fmul float %1266, %429, !dbg !38 + %1316 = fmul float %1266, %430, !dbg !38 + %1317 = fmul float %1266, %431, !dbg !38 + %1318 = fmul float %1267, %432, !dbg !38 + %1319 = fmul float %1267, %433, !dbg !38 + %1320 = fmul float %1267, %434, !dbg !38 + %1321 = fmul float %1267, %435, !dbg !38 + %1322 = fmul float %1267, %436, !dbg !38 + %1323 = fmul float %1267, %437, !dbg !38 + %1324 = fmul float %1267, %438, !dbg !38 + %1325 = fmul float %1267, %439, !dbg !38 + %1326 = fmul float %1268, %440, !dbg !38 + %1327 = fmul float %1268, %441, !dbg !38 + %1328 = fmul float %1268, %442, !dbg !38 + %1329 = fmul float %1268, %443, !dbg !38 + %1330 = fmul float %1268, %444, !dbg !38 + %1331 = fmul float %1268, %445, !dbg !38 + %1332 = fmul float %1268, %446, !dbg !38 + %1333 = fmul float %1268, %447, !dbg !38 + %1334 = fmul float %1269, %448, !dbg !38 + %1335 = fmul float %1269, %449, !dbg !38 + %1336 = fmul float %1269, %450, !dbg !38 + %1337 = fmul float %1269, %451, !dbg !38 + %1338 = fmul float %1269, %452, !dbg !38 + %1339 = fmul float %1269, %453, !dbg !38 + %1340 = fmul float %1269, %454, !dbg !38 + %1341 = fmul float %1269, %455, !dbg !38 + %1342 = fmul float %1270, %456, !dbg !38 + %1343 = fmul float %1270, %457, !dbg !38 + %1344 = fmul float %1270, %458, !dbg !38 + %1345 = fmul float %1270, %459, !dbg !38 + %1346 = fmul float %1270, %460, !dbg !38 + %1347 = fmul float %1270, %461, !dbg !38 + %1348 = fmul float %1270, %462, !dbg !38 + %1349 = fmul float %1270, %463, !dbg !38 + %1350 = fmul float %1271, %464, !dbg !38 + %1351 = fmul float %1271, %465, !dbg !38 + %1352 = fmul float %1271, %466, !dbg !38 + %1353 = fmul float %1271, %467, !dbg !38 + %1354 = fmul float %1271, %468, !dbg !38 + %1355 = fmul float %1271, %469, !dbg !38 + %1356 = fmul float %1271, %470, !dbg !38 + %1357 = fmul float %1271, %471, !dbg !38 + %1358 = fmul float %1272, %472, !dbg !38 + %1359 = fmul float %1272, %473, !dbg !38 + %1360 = fmul float %1272, %474, !dbg !38 + %1361 = fmul float %1272, %475, !dbg !38 + %1362 = fmul float %1272, %476, !dbg !38 + %1363 = fmul float %1272, %477, !dbg !38 + %1364 = fmul float %1272, %478, !dbg !38 + %1365 = fmul float %1272, %479, !dbg !38 + %1366 = fmul float %1273, %480, !dbg !38 + %1367 = fmul float %1273, %481, !dbg !38 + %1368 = fmul float %1273, %482, !dbg !38 + %1369 = fmul float %1273, %483, !dbg !38 + %1370 = fmul float %1273, %484, !dbg !38 + %1371 = fmul float %1273, %485, !dbg !38 + %1372 = fmul float %1273, %486, !dbg !38 + %1373 = fmul float %1273, %487, !dbg !38 + %1374 = fmul float %1274, %488, !dbg !38 + %1375 = fmul float %1274, %489, !dbg !38 + %1376 = fmul float %1274, %490, !dbg !38 + %1377 = fmul float %1274, %491, !dbg !38 + %1378 = fmul float %1274, %492, !dbg !38 + %1379 = fmul float %1274, %493, !dbg !38 + %1380 = fmul float %1274, %494, !dbg !38 + %1381 = fmul float %1274, %495, !dbg !38 + %1382 = fmul float %1275, %496, !dbg !38 + %1383 = fmul float %1275, %497, !dbg !38 + %1384 = fmul float %1275, %498, !dbg !38 + %1385 = fmul float %1275, %499, !dbg !38 + %1386 = fmul float %1275, %500, !dbg !38 + %1387 = fmul float %1275, %501, !dbg !38 + %1388 = fmul float %1275, %502, !dbg !38 + %1389 = fmul float %1275, %503, !dbg !38 + %1390 = fmul float %1276, %504, !dbg !38 + %1391 = fmul float %1276, %505, !dbg !38 + %1392 = fmul float %1276, %506, !dbg !38 + %1393 = fmul float %1276, %507, !dbg !38 + %1394 = fmul float %1276, %508, !dbg !38 + %1395 = fmul float %1276, %509, !dbg !38 + %1396 = fmul float %1276, %510, !dbg !38 + %1397 = fmul float %1276, %511, !dbg !38 + %1398 = fmul float %1277, %512, !dbg !38 + %1399 = fmul float %1277, %513, !dbg !38 + %1400 = fmul float %1277, %514, !dbg !38 + %1401 = fmul float %1277, %515, !dbg !38 + %1402 = fmul float %1277, %516, !dbg !38 + %1403 = fmul float %1277, %517, !dbg !38 + %1404 = fmul float %1277, %518, !dbg !38 + %1405 = fmul float %1277, %519, !dbg !38 + %1406 = fptrunc float %1278 to bfloat, !dbg !39 + %1407 = fptrunc float %1279 to bfloat, !dbg !39 + %1408 = fptrunc float %1280 to bfloat, !dbg !39 + %1409 = fptrunc float %1281 to bfloat, !dbg !39 + %1410 = fptrunc float %1282 to bfloat, !dbg !39 + %1411 = fptrunc float %1283 to bfloat, !dbg !39 + %1412 = fptrunc float %1284 to bfloat, !dbg !39 + %1413 = fptrunc float %1285 to bfloat, !dbg !39 + %1414 = fptrunc float %1286 to bfloat, !dbg !39 + %1415 = fptrunc float %1287 to bfloat, !dbg !39 + %1416 = fptrunc float %1288 to bfloat, !dbg !39 + %1417 = fptrunc float %1289 to bfloat, !dbg !39 + %1418 = fptrunc float %1290 to bfloat, !dbg !39 + %1419 = fptrunc float %1291 to bfloat, !dbg !39 + %1420 = fptrunc float %1292 to bfloat, !dbg !39 + %1421 = fptrunc float %1293 to bfloat, !dbg !39 + %1422 = fptrunc float %1294 to bfloat, !dbg !39 + %1423 = fptrunc float %1295 to bfloat, !dbg !39 + %1424 = fptrunc float %1296 to bfloat, !dbg !39 + %1425 = fptrunc float %1297 to bfloat, !dbg !39 + %1426 = fptrunc float %1298 to bfloat, !dbg !39 + %1427 = fptrunc float %1299 to bfloat, !dbg !39 + %1428 = fptrunc float %1300 to bfloat, !dbg !39 + %1429 = fptrunc float %1301 to bfloat, !dbg !39 + %1430 = fptrunc float %1302 to bfloat, !dbg !39 + %1431 = fptrunc float %1303 to bfloat, !dbg !39 + %1432 = fptrunc float %1304 to bfloat, !dbg !39 + %1433 = fptrunc float %1305 to bfloat, !dbg !39 + %1434 = fptrunc float %1306 to bfloat, !dbg !39 + %1435 = fptrunc float %1307 to bfloat, !dbg !39 + %1436 = fptrunc float %1308 to bfloat, !dbg !39 + %1437 = fptrunc float %1309 to bfloat, !dbg !39 + %1438 = fptrunc float %1310 to bfloat, !dbg !39 + %1439 = fptrunc float %1311 to bfloat, !dbg !39 + %1440 = fptrunc float %1312 to bfloat, !dbg !39 + %1441 = fptrunc float %1313 to bfloat, !dbg !39 + %1442 = fptrunc float %1314 to bfloat, !dbg !39 + %1443 = fptrunc float %1315 to bfloat, !dbg !39 + %1444 = fptrunc float %1316 to bfloat, !dbg !39 + %1445 = fptrunc float %1317 to bfloat, !dbg !39 + %1446 = fptrunc float %1318 to bfloat, !dbg !39 + %1447 = fptrunc float %1319 to bfloat, !dbg !39 + %1448 = fptrunc float %1320 to bfloat, !dbg !39 + %1449 = fptrunc float %1321 to bfloat, !dbg !39 + %1450 = fptrunc float %1322 to bfloat, !dbg !39 + %1451 = fptrunc float %1323 to bfloat, !dbg !39 + %1452 = fptrunc float %1324 to bfloat, !dbg !39 + %1453 = fptrunc float %1325 to bfloat, !dbg !39 + %1454 = fptrunc float %1326 to bfloat, !dbg !39 + %1455 = fptrunc float %1327 to bfloat, !dbg !39 + %1456 = fptrunc float %1328 to bfloat, !dbg !39 + %1457 = fptrunc float %1329 to bfloat, !dbg !39 + %1458 = fptrunc float %1330 to bfloat, !dbg !39 + %1459 = fptrunc float %1331 to bfloat, !dbg !39 + %1460 = fptrunc float %1332 to bfloat, !dbg !39 + %1461 = fptrunc float %1333 to bfloat, !dbg !39 + %1462 = fptrunc float %1334 to bfloat, !dbg !39 + %1463 = fptrunc float %1335 to bfloat, !dbg !39 + %1464 = fptrunc float %1336 to bfloat, !dbg !39 + %1465 = fptrunc float %1337 to bfloat, !dbg !39 + %1466 = fptrunc float %1338 to bfloat, !dbg !39 + %1467 = fptrunc float %1339 to bfloat, !dbg !39 + %1468 = fptrunc float %1340 to bfloat, !dbg !39 + %1469 = fptrunc float %1341 to bfloat, !dbg !39 + %1470 = fptrunc float %1342 to bfloat, !dbg !39 + %1471 = fptrunc float %1343 to bfloat, !dbg !39 + %1472 = fptrunc float %1344 to bfloat, !dbg !39 + %1473 = fptrunc float %1345 to bfloat, !dbg !39 + %1474 = fptrunc float %1346 to bfloat, !dbg !39 + %1475 = fptrunc float %1347 to bfloat, !dbg !39 + %1476 = fptrunc float %1348 to bfloat, !dbg !39 + %1477 = fptrunc float %1349 to bfloat, !dbg !39 + %1478 = fptrunc float %1350 to bfloat, !dbg !39 + %1479 = fptrunc float %1351 to bfloat, !dbg !39 + %1480 = fptrunc float %1352 to bfloat, !dbg !39 + %1481 = fptrunc float %1353 to bfloat, !dbg !39 + %1482 = fptrunc float %1354 to bfloat, !dbg !39 + %1483 = fptrunc float %1355 to bfloat, !dbg !39 + %1484 = fptrunc float %1356 to bfloat, !dbg !39 + %1485 = fptrunc float %1357 to bfloat, !dbg !39 + %1486 = fptrunc float %1358 to bfloat, !dbg !39 + %1487 = fptrunc float %1359 to bfloat, !dbg !39 + %1488 = fptrunc float %1360 to bfloat, !dbg !39 + %1489 = fptrunc float %1361 to bfloat, !dbg !39 + %1490 = fptrunc float %1362 to bfloat, !dbg !39 + %1491 = fptrunc float %1363 to bfloat, !dbg !39 + %1492 = fptrunc float %1364 to bfloat, !dbg !39 + %1493 = fptrunc float %1365 to bfloat, !dbg !39 + %1494 = fptrunc float %1366 to bfloat, !dbg !39 + %1495 = fptrunc float %1367 to bfloat, !dbg !39 + %1496 = fptrunc float %1368 to bfloat, !dbg !39 + %1497 = fptrunc float %1369 to bfloat, !dbg !39 + %1498 = fptrunc float %1370 to bfloat, !dbg !39 + %1499 = fptrunc float %1371 to bfloat, !dbg !39 + %1500 = fptrunc float %1372 to bfloat, !dbg !39 + %1501 = fptrunc float %1373 to bfloat, !dbg !39 + %1502 = fptrunc float %1374 to bfloat, !dbg !39 + %1503 = fptrunc float %1375 to bfloat, !dbg !39 + %1504 = fptrunc float %1376 to bfloat, !dbg !39 + %1505 = fptrunc float %1377 to bfloat, !dbg !39 + %1506 = fptrunc float %1378 to bfloat, !dbg !39 + %1507 = fptrunc float %1379 to bfloat, !dbg !39 + %1508 = fptrunc float %1380 to bfloat, !dbg !39 + %1509 = fptrunc float %1381 to bfloat, !dbg !39 + %1510 = fptrunc float %1382 to bfloat, !dbg !39 + %1511 = fptrunc float %1383 to bfloat, !dbg !39 + %1512 = fptrunc float %1384 to bfloat, !dbg !39 + %1513 = fptrunc float %1385 to bfloat, !dbg !39 + %1514 = fptrunc float %1386 to bfloat, !dbg !39 + %1515 = fptrunc float %1387 to bfloat, !dbg !39 + %1516 = fptrunc float %1388 to bfloat, !dbg !39 + %1517 = fptrunc float %1389 to bfloat, !dbg !39 + %1518 = fptrunc float %1390 to bfloat, !dbg !39 + %1519 = fptrunc float %1391 to bfloat, !dbg !39 + %1520 = fptrunc float %1392 to bfloat, !dbg !39 + %1521 = fptrunc float %1393 to bfloat, !dbg !39 + %1522 = fptrunc float %1394 to bfloat, !dbg !39 + %1523 = fptrunc float %1395 to bfloat, !dbg !39 + %1524 = fptrunc float %1396 to bfloat, !dbg !39 + %1525 = fptrunc float %1397 to bfloat, !dbg !39 + %1526 = fptrunc float %1398 to bfloat, !dbg !39 + %1527 = fptrunc float %1399 to bfloat, !dbg !39 + %1528 = fptrunc float %1400 to bfloat, !dbg !39 + %1529 = fptrunc float %1401 to bfloat, !dbg !39 + %1530 = fptrunc float %1402 to bfloat, !dbg !39 + %1531 = fptrunc float %1403 to bfloat, !dbg !39 + %1532 = fptrunc float %1404 to bfloat, !dbg !39 + %1533 = fptrunc float %1405 to bfloat, !dbg !39 + %1534 = insertelement <2 x bfloat> poison, bfloat %1406, i64 0, !dbg !40 + %1535 = insertelement <2 x bfloat> %1534, bfloat %1407, i64 1, !dbg !40 + %1536 = bitcast <2 x bfloat> %1535 to i32, !dbg !40 + %1537 = insertelement <2 x bfloat> poison, bfloat %1408, i64 0, !dbg !40 + %1538 = insertelement <2 x bfloat> %1537, bfloat %1409, i64 1, !dbg !40 + %1539 = bitcast <2 x bfloat> %1538 to i32, !dbg !40 + %1540 = insertelement <2 x bfloat> poison, bfloat %1410, i64 0, !dbg !40 + %1541 = insertelement <2 x bfloat> %1540, bfloat %1411, i64 1, !dbg !40 + %1542 = bitcast <2 x bfloat> %1541 to i32, !dbg !40 + %1543 = insertelement <2 x bfloat> poison, bfloat %1412, i64 0, !dbg !40 + %1544 = insertelement <2 x bfloat> %1543, bfloat %1413, i64 1, !dbg !40 + %1545 = bitcast <2 x bfloat> %1544 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1536, i32 %1539, i32 %1542, i32 %1545, ptr addrspace(1) %1260, i1 %104) #6, !dbg !40 + %1546 = insertelement <2 x bfloat> poison, bfloat %1414, i64 0, !dbg !40 + %1547 = insertelement <2 x bfloat> %1546, bfloat %1415, i64 1, !dbg !40 + %1548 = bitcast <2 x bfloat> %1547 to i32, !dbg !40 + %1549 = insertelement <2 x bfloat> poison, bfloat %1416, i64 0, !dbg !40 + %1550 = insertelement <2 x bfloat> %1549, bfloat %1417, i64 1, !dbg !40 + %1551 = bitcast <2 x bfloat> %1550 to i32, !dbg !40 + %1552 = insertelement <2 x bfloat> poison, bfloat %1418, i64 0, !dbg !40 + %1553 = insertelement <2 x bfloat> %1552, bfloat %1419, i64 1, !dbg !40 + %1554 = bitcast <2 x bfloat> %1553 to i32, !dbg !40 + %1555 = insertelement <2 x bfloat> poison, bfloat %1420, i64 0, !dbg !40 + %1556 = insertelement <2 x bfloat> %1555, bfloat %1421, i64 1, !dbg !40 + %1557 = bitcast <2 x bfloat> %1556 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1548, i32 %1551, i32 %1554, i32 %1557, ptr addrspace(1) %1258, i1 %122) #6, !dbg !40 + %1558 = insertelement <2 x bfloat> poison, bfloat %1422, i64 0, !dbg !40 + %1559 = insertelement <2 x bfloat> %1558, bfloat %1423, i64 1, !dbg !40 + %1560 = bitcast <2 x bfloat> %1559 to i32, !dbg !40 + %1561 = insertelement <2 x bfloat> poison, bfloat %1424, i64 0, !dbg !40 + %1562 = insertelement <2 x bfloat> %1561, bfloat %1425, i64 1, !dbg !40 + %1563 = bitcast <2 x bfloat> %1562 to i32, !dbg !40 + %1564 = insertelement <2 x bfloat> poison, bfloat %1426, i64 0, !dbg !40 + %1565 = insertelement <2 x bfloat> %1564, bfloat %1427, i64 1, !dbg !40 + %1566 = bitcast <2 x bfloat> %1565 to i32, !dbg !40 + %1567 = insertelement <2 x bfloat> poison, bfloat %1428, i64 0, !dbg !40 + %1568 = insertelement <2 x bfloat> %1567, bfloat %1429, i64 1, !dbg !40 + %1569 = bitcast <2 x bfloat> %1568 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1560, i32 %1563, i32 %1566, i32 %1569, ptr addrspace(1) %1256, i1 %140) #6, !dbg !40 + %1570 = insertelement <2 x bfloat> poison, bfloat %1430, i64 0, !dbg !40 + %1571 = insertelement <2 x bfloat> %1570, bfloat %1431, i64 1, !dbg !40 + %1572 = bitcast <2 x bfloat> %1571 to i32, !dbg !40 + %1573 = insertelement <2 x bfloat> poison, bfloat %1432, i64 0, !dbg !40 + %1574 = insertelement <2 x bfloat> %1573, bfloat %1433, i64 1, !dbg !40 + %1575 = bitcast <2 x bfloat> %1574 to i32, !dbg !40 + %1576 = insertelement <2 x bfloat> poison, bfloat %1434, i64 0, !dbg !40 + %1577 = insertelement <2 x bfloat> %1576, bfloat %1435, i64 1, !dbg !40 + %1578 = bitcast <2 x bfloat> %1577 to i32, !dbg !40 + %1579 = insertelement <2 x bfloat> poison, bfloat %1436, i64 0, !dbg !40 + %1580 = insertelement <2 x bfloat> %1579, bfloat %1437, i64 1, !dbg !40 + %1581 = bitcast <2 x bfloat> %1580 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1572, i32 %1575, i32 %1578, i32 %1581, ptr addrspace(1) %1254, i1 %158) #6, !dbg !40 + %1582 = insertelement <2 x bfloat> poison, bfloat %1438, i64 0, !dbg !40 + %1583 = insertelement <2 x bfloat> %1582, bfloat %1439, i64 1, !dbg !40 + %1584 = bitcast <2 x bfloat> %1583 to i32, !dbg !40 + %1585 = insertelement <2 x bfloat> poison, bfloat %1440, i64 0, !dbg !40 + %1586 = insertelement <2 x bfloat> %1585, bfloat %1441, i64 1, !dbg !40 + %1587 = bitcast <2 x bfloat> %1586 to i32, !dbg !40 + %1588 = insertelement <2 x bfloat> poison, bfloat %1442, i64 0, !dbg !40 + %1589 = insertelement <2 x bfloat> %1588, bfloat %1443, i64 1, !dbg !40 + %1590 = bitcast <2 x bfloat> %1589 to i32, !dbg !40 + %1591 = insertelement <2 x bfloat> poison, bfloat %1444, i64 0, !dbg !40 + %1592 = insertelement <2 x bfloat> %1591, bfloat %1445, i64 1, !dbg !40 + %1593 = bitcast <2 x bfloat> %1592 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1584, i32 %1587, i32 %1590, i32 %1593, ptr addrspace(1) %1252, i1 %176) #6, !dbg !40 + %1594 = insertelement <2 x bfloat> poison, bfloat %1446, i64 0, !dbg !40 + %1595 = insertelement <2 x bfloat> %1594, bfloat %1447, i64 1, !dbg !40 + %1596 = bitcast <2 x bfloat> %1595 to i32, !dbg !40 + %1597 = insertelement <2 x bfloat> poison, bfloat %1448, i64 0, !dbg !40 + %1598 = insertelement <2 x bfloat> %1597, bfloat %1449, i64 1, !dbg !40 + %1599 = bitcast <2 x bfloat> %1598 to i32, !dbg !40 + %1600 = insertelement <2 x bfloat> poison, bfloat %1450, i64 0, !dbg !40 + %1601 = insertelement <2 x bfloat> %1600, bfloat %1451, i64 1, !dbg !40 + %1602 = bitcast <2 x bfloat> %1601 to i32, !dbg !40 + %1603 = insertelement <2 x bfloat> poison, bfloat %1452, i64 0, !dbg !40 + %1604 = insertelement <2 x bfloat> %1603, bfloat %1453, i64 1, !dbg !40 + %1605 = bitcast <2 x bfloat> %1604 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1596, i32 %1599, i32 %1602, i32 %1605, ptr addrspace(1) %1250, i1 %194) #6, !dbg !40 + %1606 = insertelement <2 x bfloat> poison, bfloat %1454, i64 0, !dbg !40 + %1607 = insertelement <2 x bfloat> %1606, bfloat %1455, i64 1, !dbg !40 + %1608 = bitcast <2 x bfloat> %1607 to i32, !dbg !40 + %1609 = insertelement <2 x bfloat> poison, bfloat %1456, i64 0, !dbg !40 + %1610 = insertelement <2 x bfloat> %1609, bfloat %1457, i64 1, !dbg !40 + %1611 = bitcast <2 x bfloat> %1610 to i32, !dbg !40 + %1612 = insertelement <2 x bfloat> poison, bfloat %1458, i64 0, !dbg !40 + %1613 = insertelement <2 x bfloat> %1612, bfloat %1459, i64 1, !dbg !40 + %1614 = bitcast <2 x bfloat> %1613 to i32, !dbg !40 + %1615 = insertelement <2 x bfloat> poison, bfloat %1460, i64 0, !dbg !40 + %1616 = insertelement <2 x bfloat> %1615, bfloat %1461, i64 1, !dbg !40 + %1617 = bitcast <2 x bfloat> %1616 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1608, i32 %1611, i32 %1614, i32 %1617, ptr addrspace(1) %1248, i1 %212) #6, !dbg !40 + %1618 = insertelement <2 x bfloat> poison, bfloat %1462, i64 0, !dbg !40 + %1619 = insertelement <2 x bfloat> %1618, bfloat %1463, i64 1, !dbg !40 + %1620 = bitcast <2 x bfloat> %1619 to i32, !dbg !40 + %1621 = insertelement <2 x bfloat> poison, bfloat %1464, i64 0, !dbg !40 + %1622 = insertelement <2 x bfloat> %1621, bfloat %1465, i64 1, !dbg !40 + %1623 = bitcast <2 x bfloat> %1622 to i32, !dbg !40 + %1624 = insertelement <2 x bfloat> poison, bfloat %1466, i64 0, !dbg !40 + %1625 = insertelement <2 x bfloat> %1624, bfloat %1467, i64 1, !dbg !40 + %1626 = bitcast <2 x bfloat> %1625 to i32, !dbg !40 + %1627 = insertelement <2 x bfloat> poison, bfloat %1468, i64 0, !dbg !40 + %1628 = insertelement <2 x bfloat> %1627, bfloat %1469, i64 1, !dbg !40 + %1629 = bitcast <2 x bfloat> %1628 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1620, i32 %1623, i32 %1626, i32 %1629, ptr addrspace(1) %1246, i1 %230) #6, !dbg !40 + %1630 = insertelement <2 x bfloat> poison, bfloat %1470, i64 0, !dbg !40 + %1631 = insertelement <2 x bfloat> %1630, bfloat %1471, i64 1, !dbg !40 + %1632 = bitcast <2 x bfloat> %1631 to i32, !dbg !40 + %1633 = insertelement <2 x bfloat> poison, bfloat %1472, i64 0, !dbg !40 + %1634 = insertelement <2 x bfloat> %1633, bfloat %1473, i64 1, !dbg !40 + %1635 = bitcast <2 x bfloat> %1634 to i32, !dbg !40 + %1636 = insertelement <2 x bfloat> poison, bfloat %1474, i64 0, !dbg !40 + %1637 = insertelement <2 x bfloat> %1636, bfloat %1475, i64 1, !dbg !40 + %1638 = bitcast <2 x bfloat> %1637 to i32, !dbg !40 + %1639 = insertelement <2 x bfloat> poison, bfloat %1476, i64 0, !dbg !40 + %1640 = insertelement <2 x bfloat> %1639, bfloat %1477, i64 1, !dbg !40 + %1641 = bitcast <2 x bfloat> %1640 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1632, i32 %1635, i32 %1638, i32 %1641, ptr addrspace(1) %1244, i1 %248) #6, !dbg !40 + %1642 = insertelement <2 x bfloat> poison, bfloat %1478, i64 0, !dbg !40 + %1643 = insertelement <2 x bfloat> %1642, bfloat %1479, i64 1, !dbg !40 + %1644 = bitcast <2 x bfloat> %1643 to i32, !dbg !40 + %1645 = insertelement <2 x bfloat> poison, bfloat %1480, i64 0, !dbg !40 + %1646 = insertelement <2 x bfloat> %1645, bfloat %1481, i64 1, !dbg !40 + %1647 = bitcast <2 x bfloat> %1646 to i32, !dbg !40 + %1648 = insertelement <2 x bfloat> poison, bfloat %1482, i64 0, !dbg !40 + %1649 = insertelement <2 x bfloat> %1648, bfloat %1483, i64 1, !dbg !40 + %1650 = bitcast <2 x bfloat> %1649 to i32, !dbg !40 + %1651 = insertelement <2 x bfloat> poison, bfloat %1484, i64 0, !dbg !40 + %1652 = insertelement <2 x bfloat> %1651, bfloat %1485, i64 1, !dbg !40 + %1653 = bitcast <2 x bfloat> %1652 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1644, i32 %1647, i32 %1650, i32 %1653, ptr addrspace(1) %1242, i1 %266) #6, !dbg !40 + %1654 = insertelement <2 x bfloat> poison, bfloat %1486, i64 0, !dbg !40 + %1655 = insertelement <2 x bfloat> %1654, bfloat %1487, i64 1, !dbg !40 + %1656 = bitcast <2 x bfloat> %1655 to i32, !dbg !40 + %1657 = insertelement <2 x bfloat> poison, bfloat %1488, i64 0, !dbg !40 + %1658 = insertelement <2 x bfloat> %1657, bfloat %1489, i64 1, !dbg !40 + %1659 = bitcast <2 x bfloat> %1658 to i32, !dbg !40 + %1660 = insertelement <2 x bfloat> poison, bfloat %1490, i64 0, !dbg !40 + %1661 = insertelement <2 x bfloat> %1660, bfloat %1491, i64 1, !dbg !40 + %1662 = bitcast <2 x bfloat> %1661 to i32, !dbg !40 + %1663 = insertelement <2 x bfloat> poison, bfloat %1492, i64 0, !dbg !40 + %1664 = insertelement <2 x bfloat> %1663, bfloat %1493, i64 1, !dbg !40 + %1665 = bitcast <2 x bfloat> %1664 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1656, i32 %1659, i32 %1662, i32 %1665, ptr addrspace(1) %1240, i1 %284) #6, !dbg !40 + %1666 = insertelement <2 x bfloat> poison, bfloat %1494, i64 0, !dbg !40 + %1667 = insertelement <2 x bfloat> %1666, bfloat %1495, i64 1, !dbg !40 + %1668 = bitcast <2 x bfloat> %1667 to i32, !dbg !40 + %1669 = insertelement <2 x bfloat> poison, bfloat %1496, i64 0, !dbg !40 + %1670 = insertelement <2 x bfloat> %1669, bfloat %1497, i64 1, !dbg !40 + %1671 = bitcast <2 x bfloat> %1670 to i32, !dbg !40 + %1672 = insertelement <2 x bfloat> poison, bfloat %1498, i64 0, !dbg !40 + %1673 = insertelement <2 x bfloat> %1672, bfloat %1499, i64 1, !dbg !40 + %1674 = bitcast <2 x bfloat> %1673 to i32, !dbg !40 + %1675 = insertelement <2 x bfloat> poison, bfloat %1500, i64 0, !dbg !40 + %1676 = insertelement <2 x bfloat> %1675, bfloat %1501, i64 1, !dbg !40 + %1677 = bitcast <2 x bfloat> %1676 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1668, i32 %1671, i32 %1674, i32 %1677, ptr addrspace(1) %1238, i1 %302) #6, !dbg !40 + %1678 = insertelement <2 x bfloat> poison, bfloat %1502, i64 0, !dbg !40 + %1679 = insertelement <2 x bfloat> %1678, bfloat %1503, i64 1, !dbg !40 + %1680 = bitcast <2 x bfloat> %1679 to i32, !dbg !40 + %1681 = insertelement <2 x bfloat> poison, bfloat %1504, i64 0, !dbg !40 + %1682 = insertelement <2 x bfloat> %1681, bfloat %1505, i64 1, !dbg !40 + %1683 = bitcast <2 x bfloat> %1682 to i32, !dbg !40 + %1684 = insertelement <2 x bfloat> poison, bfloat %1506, i64 0, !dbg !40 + %1685 = insertelement <2 x bfloat> %1684, bfloat %1507, i64 1, !dbg !40 + %1686 = bitcast <2 x bfloat> %1685 to i32, !dbg !40 + %1687 = insertelement <2 x bfloat> poison, bfloat %1508, i64 0, !dbg !40 + %1688 = insertelement <2 x bfloat> %1687, bfloat %1509, i64 1, !dbg !40 + %1689 = bitcast <2 x bfloat> %1688 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1680, i32 %1683, i32 %1686, i32 %1689, ptr addrspace(1) %1236, i1 %320) #6, !dbg !40 + %1690 = insertelement <2 x bfloat> poison, bfloat %1510, i64 0, !dbg !40 + %1691 = insertelement <2 x bfloat> %1690, bfloat %1511, i64 1, !dbg !40 + %1692 = bitcast <2 x bfloat> %1691 to i32, !dbg !40 + %1693 = insertelement <2 x bfloat> poison, bfloat %1512, i64 0, !dbg !40 + %1694 = insertelement <2 x bfloat> %1693, bfloat %1513, i64 1, !dbg !40 + %1695 = bitcast <2 x bfloat> %1694 to i32, !dbg !40 + %1696 = insertelement <2 x bfloat> poison, bfloat %1514, i64 0, !dbg !40 + %1697 = insertelement <2 x bfloat> %1696, bfloat %1515, i64 1, !dbg !40 + %1698 = bitcast <2 x bfloat> %1697 to i32, !dbg !40 + %1699 = insertelement <2 x bfloat> poison, bfloat %1516, i64 0, !dbg !40 + %1700 = insertelement <2 x bfloat> %1699, bfloat %1517, i64 1, !dbg !40 + %1701 = bitcast <2 x bfloat> %1700 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1692, i32 %1695, i32 %1698, i32 %1701, ptr addrspace(1) %1234, i1 %338) #6, !dbg !40 + %1702 = insertelement <2 x bfloat> poison, bfloat %1518, i64 0, !dbg !40 + %1703 = insertelement <2 x bfloat> %1702, bfloat %1519, i64 1, !dbg !40 + %1704 = bitcast <2 x bfloat> %1703 to i32, !dbg !40 + %1705 = insertelement <2 x bfloat> poison, bfloat %1520, i64 0, !dbg !40 + %1706 = insertelement <2 x bfloat> %1705, bfloat %1521, i64 1, !dbg !40 + %1707 = bitcast <2 x bfloat> %1706 to i32, !dbg !40 + %1708 = insertelement <2 x bfloat> poison, bfloat %1522, i64 0, !dbg !40 + %1709 = insertelement <2 x bfloat> %1708, bfloat %1523, i64 1, !dbg !40 + %1710 = bitcast <2 x bfloat> %1709 to i32, !dbg !40 + %1711 = insertelement <2 x bfloat> poison, bfloat %1524, i64 0, !dbg !40 + %1712 = insertelement <2 x bfloat> %1711, bfloat %1525, i64 1, !dbg !40 + %1713 = bitcast <2 x bfloat> %1712 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1704, i32 %1707, i32 %1710, i32 %1713, ptr addrspace(1) %1232, i1 %356) #6, !dbg !40 + %1714 = insertelement <2 x bfloat> poison, bfloat %1526, i64 0, !dbg !40 + %1715 = insertelement <2 x bfloat> %1714, bfloat %1527, i64 1, !dbg !40 + %1716 = bitcast <2 x bfloat> %1715 to i32, !dbg !40 + %1717 = insertelement <2 x bfloat> poison, bfloat %1528, i64 0, !dbg !40 + %1718 = insertelement <2 x bfloat> %1717, bfloat %1529, i64 1, !dbg !40 + %1719 = bitcast <2 x bfloat> %1718 to i32, !dbg !40 + %1720 = insertelement <2 x bfloat> poison, bfloat %1530, i64 0, !dbg !40 + %1721 = insertelement <2 x bfloat> %1720, bfloat %1531, i64 1, !dbg !40 + %1722 = bitcast <2 x bfloat> %1721 to i32, !dbg !40 + %1723 = insertelement <2 x bfloat> poison, bfloat %1532, i64 0, !dbg !40 + %1724 = insertelement <2 x bfloat> %1723, bfloat %1533, i64 1, !dbg !40 + %1725 = bitcast <2 x bfloat> %1724 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1716, i32 %1719, i32 %1722, i32 %1725, ptr addrspace(1) %1230, i1 %374) #6, !dbg !40 + %1726 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %12, !dbg !41 + %1727 = bitcast float %1262 to i32, !dbg !41 + %1728 = bitcast float %1263 to i32, !dbg !41 + %1729 = bitcast float %1264 to i32, !dbg !41 + %1730 = bitcast float %1265 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %1726, i32 %1727, i32 %1728, i32 %1729, i32 %1730) #6, !dbg !41 + %1731 = getelementptr inbounds nuw i8, ptr addrspace(3) %1726, i32 128, !dbg !41 + %1732 = bitcast float %1266 to i32, !dbg !41 + %1733 = bitcast float %1267 to i32, !dbg !41 + %1734 = bitcast float %1268 to i32, !dbg !41 + %1735 = bitcast float %1269 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %1731, i32 %1732, i32 %1733, i32 %1734, i32 %1735) #6, !dbg !41 + %1736 = getelementptr inbounds nuw i8, ptr addrspace(3) %1726, i32 256, !dbg !41 + %1737 = bitcast float %1270 to i32, !dbg !41 + %1738 = bitcast float %1271 to i32, !dbg !41 + %1739 = bitcast float %1272 to i32, !dbg !41 + %1740 = bitcast float %1273 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %1736, i32 %1737, i32 %1738, i32 %1739, i32 %1740) #6, !dbg !41 + %1741 = getelementptr inbounds nuw i8, ptr addrspace(3) %1726, i32 384, !dbg !41 + %1742 = bitcast float %1274 to i32, !dbg !41 + %1743 = bitcast float %1275 to i32, !dbg !41 + %1744 = bitcast float %1276 to i32, !dbg !41 + %1745 = bitcast float %1277 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %1741, i32 %1742, i32 %1743, i32 %1744, i32 %1745) #6, !dbg !41 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !41 + %1746 = shl nuw nsw i32 %11, 4, !dbg !41 + %1747 = and i32 %1746, 112, !dbg !41 + %1748 = shl nuw nsw i32 %11, 2, !dbg !41 + %1749 = and i32 %1748, 384, !dbg !41 + %1750 = lshr i32 %11, 1, !dbg !41 + %1751 = and i32 %1750, 12, !dbg !41 + %1752 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %1747, !dbg !41 + %1753 = getelementptr inbounds nuw i8, ptr addrspace(3) %1752, i32 %1749, !dbg !41 + %1754 = getelementptr inbounds nuw i8, ptr addrspace(3) %1753, i32 %1751, !dbg !41 + %1755 = load i32, ptr addrspace(3) %1754, align 4, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %1755, ptr addrspace(1) %1228, i1 %1261) #6, !dbg !41 + ret void, !dbg !42 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.div.full(float, float) #3 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #4 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #5 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.f(float) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="128" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) } +attributes #4 = { convergent nocallback nounwind } +attributes #5 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #6 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "l2norm_fwd_kernel", linkageName: "l2norm_fwd_kernel", scope: !1, file: !1, line: 69, type: !6, scopeLine: 69, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !10, !11, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 70, column: 24, scope: !5) +!14 = !DILocation(line: 70, column: 30, scope: !5) +!15 = !DILocation(line: 71, column: 16, scope: !5) +!16 = !DILocation(line: 71, column: 34, scope: !5) +!17 = !DILocation(line: 71, column: 21, scope: !5) +!18 = !DILocation(line: 73, column: 16, scope: !5) +!19 = !DILocation(line: 74, column: 30, scope: !5) +!20 = !DILocation(line: 75, column: 29, scope: !5) +!21 = !DILocation(line: 75, column: 14, scope: !5) +!22 = !DILocation(line: 75, column: 33, scope: !5) +!23 = !DILocation(line: 79, column: 18, scope: !5) +!24 = !DILocation(line: 79, column: 47, scope: !5) +!25 = !DILocation(line: 80, column: 38, scope: !5) +!26 = !DILocation(line: 263, column: 15, scope: !27, inlinedAt: !29) +!27 = distinct !DILexicalBlockFile(scope: !5, file: !28, discriminator: 0) +!28 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!29 = !DILocation(line: 293, column: 36, scope: !27, inlinedAt: !30) +!30 = !DILocation(line: 80, column: 43, scope: !31) +!31 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!32 = !DILocation(line: 80, column: 48, scope: !5) +!33 = !DILocation(line: 80, column: 25, scope: !5) +!34 = !DILocation(line: 77, column: 20, scope: !5) +!35 = !DILocation(line: 76, column: 14, scope: !5) +!36 = !DILocation(line: 76, column: 33, scope: !5) +!37 = !DILocation(line: 80, column: 17, scope: !5) +!38 = !DILocation(line: 81, column: 16, scope: !5) +!39 = !DILocation(line: 83, column: 25, scope: !5) +!40 = !DILocation(line: 83, column: 18, scope: !5) +!41 = !DILocation(line: 84, column: 21, scope: !5) +!42 = !DILocation(line: 84, column: 4, scope: !5) diff --git a/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/l2norm_fwd_kernel.ptx b/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/l2norm_fwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..b8d04a0c5320916fe578bf8979a1ffa8cbae7fb1 --- /dev/null +++ b/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/l2norm_fwd_kernel.ptx @@ -0,0 +1,1614 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_fwd_kernel // -- Begin function l2norm_fwd_kernel +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; +.global .align 1 .b8 _$_str_$_2[17] = {95, 95, 67, 85, 68, 65, 95, 80, 82, 69, 67, 95, 83, 81, 82, 84}; + // @l2norm_fwd_kernel +.visible .entry l2norm_fwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_2, + .param .f32 l2norm_fwd_kernel_param_3, + .param .u32 l2norm_fwd_kernel_param_4, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_6 +) +.reqntid 128 +{ + .reg .pred %p<18>; + .reg .b16 %rs<129>; + .reg .b32 %r<729>; + .reg .b64 %rd<123>; + .loc 1 69 0 // l2norm.py:69:0 +$L__func_begin0: + .loc 1 69 0 // l2norm.py:69:0 + +// %bb.0: // %__nv_sqrtf.exit + ld.param.b64 %rd34, [l2norm_fwd_kernel_param_0]; + ld.param.b64 %rd35, [l2norm_fwd_kernel_param_1]; +$L__tmp0: + .loc 1 70 24 // l2norm.py:70:24 + mov.u32 %r151, %ctaid.x; + ld.param.b64 %rd36, [l2norm_fwd_kernel_param_2]; + .loc 1 71 16 // l2norm.py:71:16 + mul.wide.u32 %rd37, %r151, 128; + ld.param.b32 %r152, [l2norm_fwd_kernel_param_3]; + ld.param.s32 %rd38, [l2norm_fwd_kernel_param_4]; + .loc 1 71 34 // l2norm.py:71:34 + mov.u32 %r153, %tid.x; + and.b32 %r154, %r153, 112; + bfe.u32 %r155, %r153, 4, 3; + or.b32 %r156, %r155, 8; + .loc 1 71 21 // l2norm.py:71:21 + or.b32 %r157, %r155, 64; + or.b32 %r158, %r155, 72; + or.b32 %r159, %r155, 80; + or.b32 %r160, %r155, 88; + or.b32 %r161, %r155, 96; + or.b32 %r162, %r155, 104; + or.b32 %r163, %r155, 112; + or.b32 %r164, %r155, 120; + or.b32 %r165, %r155, 32; + or.b32 %r166, %r155, 40; + or.b32 %r167, %r155, 48; + or.b32 %r168, %r155, 56; + or.b32 %r169, %r155, 16; + or.b32 %r170, %r155, 24; + cvt.u64.u32 %rd39, %r164; + cvt.u64.u32 %rd40, %r163; + cvt.u64.u32 %rd41, %r162; + cvt.u64.u32 %rd42, %r161; + cvt.u64.u32 %rd43, %r160; + cvt.u64.u32 %rd44, %r159; + cvt.u64.u32 %rd45, %r158; + cvt.u64.u32 %rd46, %r157; + cvt.u64.u32 %rd47, %r168; + cvt.u64.u32 %rd48, %r167; + cvt.u64.u32 %rd49, %r166; + cvt.u64.u32 %rd50, %r165; + cvt.u64.u32 %rd51, %r170; + cvt.u64.u32 %rd52, %r169; + cvt.u64.u32 %rd53, %r156; + cvt.u64.u32 %rd54, %r155; + or.b64 %rd55, %rd37, %rd54; + or.b64 %rd56, %rd37, %rd53; + or.b64 %rd57, %rd37, %rd52; + or.b64 %rd58, %rd37, %rd51; + or.b64 %rd59, %rd37, %rd50; + or.b64 %rd60, %rd37, %rd49; + or.b64 %rd61, %rd37, %rd48; + or.b64 %rd62, %rd37, %rd47; + or.b64 %rd63, %rd37, %rd46; + or.b64 %rd64, %rd37, %rd45; + or.b64 %rd65, %rd37, %rd44; + or.b64 %rd66, %rd37, %rd43; + or.b64 %rd67, %rd37, %rd42; + or.b64 %rd68, %rd37, %rd41; + or.b64 %rd69, %rd37, %rd40; + or.b64 %rd70, %rd37, %rd39; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p16, %rd70, %rd38; + setp.lt.s64 %p14, %rd69, %rd38; + setp.lt.s64 %p13, %rd68, %rd38; + setp.lt.s64 %p12, %rd67, %rd38; + setp.lt.s64 %p11, %rd66, %rd38; + setp.lt.s64 %p10, %rd65, %rd38; + setp.lt.s64 %p9, %rd64, %rd38; + setp.lt.s64 %p8, %rd63, %rd38; + setp.lt.s64 %p15, %rd62, %rd38; + setp.lt.s64 %p7, %rd61, %rd38; + setp.lt.s64 %p6, %rd60, %rd38; + setp.lt.s64 %p5, %rd59, %rd38; + setp.lt.s64 %p4, %rd58, %rd38; + setp.lt.s64 %p3, %rd57, %rd38; + setp.lt.s64 %p2, %rd56, %rd38; + setp.lt.s64 %p1, %rd55, %rd38; + .loc 1 74 30 // l2norm.py:74:30 + shl.b32 %r171, %r153, 3; + and.b32 %r172, %r171, 120; + .loc 1 75 14 // l2norm.py:75:14 + shl.b64 %rd71, %rd55, 8; + add.s64 %rd72, %rd34, %rd71; + shl.b64 %rd73, %rd56, 8; + add.s64 %rd74, %rd34, %rd73; + shl.b64 %rd75, %rd57, 8; + add.s64 %rd76, %rd34, %rd75; + shl.b64 %rd77, %rd58, 8; + add.s64 %rd78, %rd34, %rd77; + shl.b64 %rd79, %rd59, 8; + add.s64 %rd80, %rd34, %rd79; + shl.b64 %rd81, %rd60, 8; + add.s64 %rd82, %rd34, %rd81; + shl.b64 %rd83, %rd61, 8; + add.s64 %rd84, %rd34, %rd83; + shl.b64 %rd85, %rd62, 8; + add.s64 %rd86, %rd34, %rd85; + shl.b64 %rd87, %rd63, 8; + add.s64 %rd88, %rd34, %rd87; + shl.b64 %rd89, %rd64, 8; + add.s64 %rd90, %rd34, %rd89; + shl.b64 %rd91, %rd65, 8; + add.s64 %rd92, %rd34, %rd91; + shl.b64 %rd93, %rd66, 8; + add.s64 %rd94, %rd34, %rd93; + shl.b64 %rd95, %rd67, 8; + add.s64 %rd96, %rd34, %rd95; + shl.b64 %rd97, %rd68, 8; + add.s64 %rd98, %rd34, %rd97; + shl.b64 %rd99, %rd69, 8; + add.s64 %rd100, %rd34, %rd99; + shl.b64 %rd101, %rd70, 8; + add.s64 %rd102, %rd34, %rd101; + .loc 1 75 33 // l2norm.py:75:33 + mul.wide.u32 %rd103, %r172, 2; + add.s64 %rd1, %rd72, %rd103; + add.s64 %rd2, %rd74, %rd103; + add.s64 %rd3, %rd76, %rd103; + add.s64 %rd4, %rd78, %rd103; + add.s64 %rd5, %rd80, %rd103; + add.s64 %rd6, %rd82, %rd103; + add.s64 %rd7, %rd84, %rd103; + add.s64 %rd8, %rd86, %rd103; + add.s64 %rd9, %rd88, %rd103; + add.s64 %rd10, %rd90, %rd103; + add.s64 %rd11, %rd92, %rd103; + add.s64 %rd12, %rd94, %rd103; + add.s64 %rd13, %rd96, %rd103; + add.s64 %rd14, %rd98, %rd103; + add.s64 %rd15, %rd100, %rd103; + add.s64 %rd16, %rd102, %rd103; + mov.b32 %r5, 0; + .loc 1 79 18 // l2norm.py:79:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + mov.b32 {%rs5, %rs6}, %r3; + mov.b32 {%rs7, %rs8}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs9, %rs10}, %r6; + mov.b32 {%rs11, %rs12}, %r7; + mov.b32 {%rs13, %rs14}, %r8; + mov.b32 {%rs15, %rs16}, %r9; + // begin inline asm + mov.u32 %r10, %r5; + mov.u32 %r11, %r5; + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + @%p3 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs17, %rs18}, %r10; + mov.b32 {%rs19, %rs20}, %r11; + mov.b32 {%rs21, %rs22}, %r12; + mov.b32 {%rs23, %rs24}, %r13; + // begin inline asm + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + @%p4 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ]; + // end inline asm + mov.b32 {%rs25, %rs26}, %r14; + mov.b32 {%rs27, %rs28}, %r15; + mov.b32 {%rs29, %rs30}, %r16; + mov.b32 {%rs31, %rs32}, %r17; + // begin inline asm + mov.u32 %r18, %r5; + mov.u32 %r19, %r5; + mov.u32 %r20, %r5; + mov.u32 %r21, %r5; + @%p5 ld.global.v4.b32 { %r18, %r19, %r20, %r21 }, [ %rd5 + 0 ]; + // end inline asm + mov.b32 {%rs33, %rs34}, %r18; + mov.b32 {%rs35, %rs36}, %r19; + mov.b32 {%rs37, %rs38}, %r20; + mov.b32 {%rs39, %rs40}, %r21; + // begin inline asm + mov.u32 %r22, %r5; + mov.u32 %r23, %r5; + mov.u32 %r24, %r5; + mov.u32 %r25, %r5; + @%p6 ld.global.v4.b32 { %r22, %r23, %r24, %r25 }, [ %rd6 + 0 ]; + // end inline asm + mov.b32 {%rs41, %rs42}, %r22; + mov.b32 {%rs43, %rs44}, %r23; + mov.b32 {%rs45, %rs46}, %r24; + mov.b32 {%rs47, %rs48}, %r25; + // begin inline asm + mov.u32 %r26, %r5; + mov.u32 %r27, %r5; + mov.u32 %r28, %r5; + mov.u32 %r29, %r5; + @%p7 ld.global.v4.b32 { %r26, %r27, %r28, %r29 }, [ %rd7 + 0 ]; + // end inline asm + mov.b32 {%rs49, %rs50}, %r26; + mov.b32 {%rs51, %rs52}, %r27; + mov.b32 {%rs53, %rs54}, %r28; + mov.b32 {%rs55, %rs56}, %r29; + // begin inline asm + mov.u32 %r30, %r5; + mov.u32 %r31, %r5; + mov.u32 %r32, %r5; + mov.u32 %r33, %r5; + @%p15 ld.global.v4.b32 { %r30, %r31, %r32, %r33 }, [ %rd8 + 0 ]; + // end inline asm + mov.b32 {%rs57, %rs58}, %r30; + mov.b32 {%rs59, %rs60}, %r31; + mov.b32 {%rs61, %rs62}, %r32; + mov.b32 {%rs63, %rs64}, %r33; + // begin inline asm + mov.u32 %r34, %r5; + mov.u32 %r35, %r5; + mov.u32 %r36, %r5; + mov.u32 %r37, %r5; + @%p8 ld.global.v4.b32 { %r34, %r35, %r36, %r37 }, [ %rd9 + 0 ]; + // end inline asm + mov.b32 {%rs65, %rs66}, %r34; + mov.b32 {%rs67, %rs68}, %r35; + mov.b32 {%rs69, %rs70}, %r36; + mov.b32 {%rs71, %rs72}, %r37; + // begin inline asm + mov.u32 %r38, %r5; + mov.u32 %r39, %r5; + mov.u32 %r40, %r5; + mov.u32 %r41, %r5; + @%p9 ld.global.v4.b32 { %r38, %r39, %r40, %r41 }, [ %rd10 + 0 ]; + // end inline asm + mov.b32 {%rs73, %rs74}, %r38; + mov.b32 {%rs75, %rs76}, %r39; + mov.b32 {%rs77, %rs78}, %r40; + mov.b32 {%rs79, %rs80}, %r41; + // begin inline asm + mov.u32 %r42, %r5; + mov.u32 %r43, %r5; + mov.u32 %r44, %r5; + mov.u32 %r45, %r5; + @%p10 ld.global.v4.b32 { %r42, %r43, %r44, %r45 }, [ %rd11 + 0 ]; + // end inline asm + mov.b32 {%rs81, %rs82}, %r42; + mov.b32 {%rs83, %rs84}, %r43; + mov.b32 {%rs85, %rs86}, %r44; + mov.b32 {%rs87, %rs88}, %r45; + // begin inline asm + mov.u32 %r46, %r5; + mov.u32 %r47, %r5; + mov.u32 %r48, %r5; + mov.u32 %r49, %r5; + @%p11 ld.global.v4.b32 { %r46, %r47, %r48, %r49 }, [ %rd12 + 0 ]; + // end inline asm + mov.b32 {%rs89, %rs90}, %r46; + mov.b32 {%rs91, %rs92}, %r47; + mov.b32 {%rs93, %rs94}, %r48; + mov.b32 {%rs95, %rs96}, %r49; + // begin inline asm + mov.u32 %r50, %r5; + mov.u32 %r51, %r5; + mov.u32 %r52, %r5; + mov.u32 %r53, %r5; + @%p12 ld.global.v4.b32 { %r50, %r51, %r52, %r53 }, [ %rd13 + 0 ]; + // end inline asm + mov.b32 {%rs97, %rs98}, %r50; + mov.b32 {%rs99, %rs100}, %r51; + mov.b32 {%rs101, %rs102}, %r52; + mov.b32 {%rs103, %rs104}, %r53; + // begin inline asm + mov.u32 %r54, %r5; + mov.u32 %r55, %r5; + mov.u32 %r56, %r5; + mov.u32 %r57, %r5; + @%p13 ld.global.v4.b32 { %r54, %r55, %r56, %r57 }, [ %rd14 + 0 ]; + // end inline asm + mov.b32 {%rs105, %rs106}, %r54; + mov.b32 {%rs107, %rs108}, %r55; + mov.b32 {%rs109, %rs110}, %r56; + mov.b32 {%rs111, %rs112}, %r57; + // begin inline asm + mov.u32 %r58, %r5; + mov.u32 %r59, %r5; + mov.u32 %r60, %r5; + mov.u32 %r61, %r5; + @%p14 ld.global.v4.b32 { %r58, %r59, %r60, %r61 }, [ %rd15 + 0 ]; + // end inline asm + mov.b32 {%rs113, %rs114}, %r58; + mov.b32 {%rs115, %rs116}, %r59; + mov.b32 {%rs117, %rs118}, %r60; + mov.b32 {%rs119, %rs120}, %r61; + // begin inline asm + mov.u32 %r62, %r5; + mov.u32 %r63, %r5; + mov.u32 %r64, %r5; + mov.u32 %r65, %r5; + @%p16 ld.global.v4.b32 { %r62, %r63, %r64, %r65 }, [ %rd16 + 0 ]; + // end inline asm + mov.b32 {%rs121, %rs122}, %r62; + mov.b32 {%rs123, %rs124}, %r63; + mov.b32 {%rs125, %rs126}, %r64; + mov.b32 {%rs127, %rs128}, %r65; + .loc 1 79 47 // l2norm.py:79:47 + cvt.f32.bf16 %r173, %rs1; + cvt.f32.bf16 %r174, %rs2; + cvt.f32.bf16 %r175, %rs3; + cvt.f32.bf16 %r176, %rs4; + cvt.f32.bf16 %r177, %rs5; + cvt.f32.bf16 %r178, %rs6; + cvt.f32.bf16 %r179, %rs7; + cvt.f32.bf16 %r180, %rs8; + cvt.f32.bf16 %r181, %rs9; + cvt.f32.bf16 %r182, %rs10; + cvt.f32.bf16 %r183, %rs11; + cvt.f32.bf16 %r184, %rs12; + cvt.f32.bf16 %r185, %rs13; + cvt.f32.bf16 %r186, %rs14; + cvt.f32.bf16 %r187, %rs15; + cvt.f32.bf16 %r188, %rs16; + cvt.f32.bf16 %r189, %rs17; + cvt.f32.bf16 %r190, %rs18; + cvt.f32.bf16 %r191, %rs19; + cvt.f32.bf16 %r192, %rs20; + cvt.f32.bf16 %r193, %rs21; + cvt.f32.bf16 %r194, %rs22; + cvt.f32.bf16 %r195, %rs23; + cvt.f32.bf16 %r196, %rs24; + cvt.f32.bf16 %r197, %rs25; + cvt.f32.bf16 %r198, %rs26; + cvt.f32.bf16 %r199, %rs27; + cvt.f32.bf16 %r200, %rs28; + cvt.f32.bf16 %r201, %rs29; + cvt.f32.bf16 %r202, %rs30; + cvt.f32.bf16 %r203, %rs31; + cvt.f32.bf16 %r204, %rs32; + cvt.f32.bf16 %r205, %rs33; + cvt.f32.bf16 %r206, %rs34; + cvt.f32.bf16 %r207, %rs35; + cvt.f32.bf16 %r208, %rs36; + cvt.f32.bf16 %r209, %rs37; + cvt.f32.bf16 %r210, %rs38; + cvt.f32.bf16 %r211, %rs39; + cvt.f32.bf16 %r212, %rs40; + cvt.f32.bf16 %r213, %rs41; + cvt.f32.bf16 %r214, %rs42; + cvt.f32.bf16 %r215, %rs43; + cvt.f32.bf16 %r216, %rs44; + cvt.f32.bf16 %r217, %rs45; + cvt.f32.bf16 %r218, %rs46; + cvt.f32.bf16 %r219, %rs47; + cvt.f32.bf16 %r220, %rs48; + cvt.f32.bf16 %r221, %rs49; + cvt.f32.bf16 %r222, %rs50; + cvt.f32.bf16 %r223, %rs51; + cvt.f32.bf16 %r224, %rs52; + cvt.f32.bf16 %r225, %rs53; + cvt.f32.bf16 %r226, %rs54; + cvt.f32.bf16 %r227, %rs55; + cvt.f32.bf16 %r228, %rs56; + cvt.f32.bf16 %r229, %rs57; + cvt.f32.bf16 %r230, %rs58; + cvt.f32.bf16 %r231, %rs59; + cvt.f32.bf16 %r232, %rs60; + cvt.f32.bf16 %r233, %rs61; + cvt.f32.bf16 %r234, %rs62; + cvt.f32.bf16 %r235, %rs63; + cvt.f32.bf16 %r236, %rs64; + cvt.f32.bf16 %r237, %rs65; + cvt.f32.bf16 %r238, %rs66; + cvt.f32.bf16 %r239, %rs67; + cvt.f32.bf16 %r240, %rs68; + cvt.f32.bf16 %r241, %rs69; + cvt.f32.bf16 %r242, %rs70; + cvt.f32.bf16 %r243, %rs71; + cvt.f32.bf16 %r244, %rs72; + cvt.f32.bf16 %r245, %rs73; + cvt.f32.bf16 %r246, %rs74; + cvt.f32.bf16 %r247, %rs75; + cvt.f32.bf16 %r248, %rs76; + cvt.f32.bf16 %r249, %rs77; + cvt.f32.bf16 %r250, %rs78; + cvt.f32.bf16 %r251, %rs79; + cvt.f32.bf16 %r252, %rs80; + cvt.f32.bf16 %r253, %rs81; + cvt.f32.bf16 %r254, %rs82; + cvt.f32.bf16 %r255, %rs83; + cvt.f32.bf16 %r256, %rs84; + cvt.f32.bf16 %r257, %rs85; + cvt.f32.bf16 %r258, %rs86; + cvt.f32.bf16 %r259, %rs87; + cvt.f32.bf16 %r260, %rs88; + cvt.f32.bf16 %r261, %rs89; + cvt.f32.bf16 %r262, %rs90; + cvt.f32.bf16 %r263, %rs91; + cvt.f32.bf16 %r264, %rs92; + cvt.f32.bf16 %r265, %rs93; + cvt.f32.bf16 %r266, %rs94; + cvt.f32.bf16 %r267, %rs95; + cvt.f32.bf16 %r268, %rs96; + cvt.f32.bf16 %r269, %rs97; + cvt.f32.bf16 %r270, %rs98; + cvt.f32.bf16 %r271, %rs99; + cvt.f32.bf16 %r272, %rs100; + cvt.f32.bf16 %r273, %rs101; + cvt.f32.bf16 %r274, %rs102; + cvt.f32.bf16 %r275, %rs103; + cvt.f32.bf16 %r276, %rs104; + cvt.f32.bf16 %r277, %rs105; + cvt.f32.bf16 %r278, %rs106; + cvt.f32.bf16 %r279, %rs107; + cvt.f32.bf16 %r280, %rs108; + cvt.f32.bf16 %r281, %rs109; + cvt.f32.bf16 %r282, %rs110; + cvt.f32.bf16 %r283, %rs111; + cvt.f32.bf16 %r284, %rs112; + cvt.f32.bf16 %r285, %rs113; + cvt.f32.bf16 %r286, %rs114; + cvt.f32.bf16 %r287, %rs115; + cvt.f32.bf16 %r288, %rs116; + cvt.f32.bf16 %r289, %rs117; + cvt.f32.bf16 %r290, %rs118; + cvt.f32.bf16 %r291, %rs119; + cvt.f32.bf16 %r292, %rs120; + cvt.f32.bf16 %r293, %rs121; + cvt.f32.bf16 %r294, %rs122; + cvt.f32.bf16 %r295, %rs123; + cvt.f32.bf16 %r296, %rs124; + cvt.f32.bf16 %r297, %rs125; + cvt.f32.bf16 %r298, %rs126; + cvt.f32.bf16 %r299, %rs127; + cvt.f32.bf16 %r300, %rs128; + .loc 1 80 38 // l2norm.py:80:38 + mul.f32 %r301, %r174, %r174; + mul.f32 %r302, %r182, %r182; + mul.f32 %r303, %r190, %r190; + mul.f32 %r304, %r198, %r198; + mul.f32 %r305, %r206, %r206; + mul.f32 %r306, %r214, %r214; + mul.f32 %r307, %r222, %r222; + mul.f32 %r308, %r230, %r230; + mul.f32 %r309, %r238, %r238; + mul.f32 %r310, %r246, %r246; + mul.f32 %r311, %r254, %r254; + mul.f32 %r312, %r262, %r262; + mul.f32 %r313, %r270, %r270; + mul.f32 %r314, %r278, %r278; + mul.f32 %r315, %r286, %r286; + mul.f32 %r316, %r294, %r294; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + fma.rn.f32.bf16 %r317, %rs1, %rs1, %r301; + fma.rn.f32.bf16 %r318, %rs3, %rs3, %r317; + fma.rn.f32.bf16 %r319, %rs4, %rs4, %r318; + fma.rn.f32.bf16 %r320, %rs5, %rs5, %r319; + fma.rn.f32.bf16 %r321, %rs6, %rs6, %r320; + fma.rn.f32.bf16 %r322, %rs7, %rs7, %r321; + fma.rn.f32.bf16 %r323, %rs8, %rs8, %r322; + fma.rn.f32.bf16 %r324, %rs9, %rs9, %r302; + fma.rn.f32.bf16 %r325, %rs11, %rs11, %r324; + fma.rn.f32.bf16 %r326, %rs12, %rs12, %r325; + fma.rn.f32.bf16 %r327, %rs13, %rs13, %r326; + fma.rn.f32.bf16 %r328, %rs14, %rs14, %r327; + fma.rn.f32.bf16 %r329, %rs15, %rs15, %r328; + fma.rn.f32.bf16 %r330, %rs16, %rs16, %r329; + fma.rn.f32.bf16 %r331, %rs17, %rs17, %r303; + fma.rn.f32.bf16 %r332, %rs19, %rs19, %r331; + fma.rn.f32.bf16 %r333, %rs20, %rs20, %r332; + fma.rn.f32.bf16 %r334, %rs21, %rs21, %r333; + fma.rn.f32.bf16 %r335, %rs22, %rs22, %r334; + fma.rn.f32.bf16 %r336, %rs23, %rs23, %r335; + fma.rn.f32.bf16 %r337, %rs24, %rs24, %r336; + fma.rn.f32.bf16 %r338, %rs25, %rs25, %r304; + fma.rn.f32.bf16 %r339, %rs27, %rs27, %r338; + fma.rn.f32.bf16 %r340, %rs28, %rs28, %r339; + fma.rn.f32.bf16 %r341, %rs29, %rs29, %r340; + fma.rn.f32.bf16 %r342, %rs30, %rs30, %r341; + fma.rn.f32.bf16 %r343, %rs31, %rs31, %r342; + fma.rn.f32.bf16 %r344, %rs32, %rs32, %r343; + fma.rn.f32.bf16 %r345, %rs33, %rs33, %r305; + fma.rn.f32.bf16 %r346, %rs35, %rs35, %r345; + fma.rn.f32.bf16 %r347, %rs36, %rs36, %r346; + fma.rn.f32.bf16 %r348, %rs37, %rs37, %r347; + fma.rn.f32.bf16 %r349, %rs38, %rs38, %r348; + fma.rn.f32.bf16 %r350, %rs39, %rs39, %r349; + fma.rn.f32.bf16 %r351, %rs40, %rs40, %r350; + fma.rn.f32.bf16 %r352, %rs41, %rs41, %r306; + fma.rn.f32.bf16 %r353, %rs43, %rs43, %r352; + fma.rn.f32.bf16 %r354, %rs44, %rs44, %r353; + fma.rn.f32.bf16 %r355, %rs45, %rs45, %r354; + fma.rn.f32.bf16 %r356, %rs46, %rs46, %r355; + fma.rn.f32.bf16 %r357, %rs47, %rs47, %r356; + fma.rn.f32.bf16 %r358, %rs48, %rs48, %r357; + fma.rn.f32.bf16 %r359, %rs49, %rs49, %r307; + fma.rn.f32.bf16 %r360, %rs51, %rs51, %r359; + fma.rn.f32.bf16 %r361, %rs52, %rs52, %r360; + fma.rn.f32.bf16 %r362, %rs53, %rs53, %r361; + fma.rn.f32.bf16 %r363, %rs54, %rs54, %r362; + fma.rn.f32.bf16 %r364, %rs55, %rs55, %r363; + fma.rn.f32.bf16 %r365, %rs56, %rs56, %r364; + fma.rn.f32.bf16 %r366, %rs57, %rs57, %r308; + fma.rn.f32.bf16 %r367, %rs59, %rs59, %r366; + fma.rn.f32.bf16 %r368, %rs60, %rs60, %r367; + fma.rn.f32.bf16 %r369, %rs61, %rs61, %r368; + fma.rn.f32.bf16 %r370, %rs62, %rs62, %r369; + fma.rn.f32.bf16 %r371, %rs63, %rs63, %r370; + fma.rn.f32.bf16 %r372, %rs64, %rs64, %r371; + fma.rn.f32.bf16 %r373, %rs65, %rs65, %r309; + fma.rn.f32.bf16 %r374, %rs67, %rs67, %r373; + fma.rn.f32.bf16 %r375, %rs68, %rs68, %r374; + fma.rn.f32.bf16 %r376, %rs69, %rs69, %r375; + fma.rn.f32.bf16 %r377, %rs70, %rs70, %r376; + fma.rn.f32.bf16 %r378, %rs71, %rs71, %r377; + fma.rn.f32.bf16 %r379, %rs72, %rs72, %r378; + fma.rn.f32.bf16 %r380, %rs73, %rs73, %r310; + fma.rn.f32.bf16 %r381, %rs75, %rs75, %r380; + fma.rn.f32.bf16 %r382, %rs76, %rs76, %r381; + fma.rn.f32.bf16 %r383, %rs77, %rs77, %r382; + fma.rn.f32.bf16 %r384, %rs78, %rs78, %r383; + fma.rn.f32.bf16 %r385, %rs79, %rs79, %r384; + fma.rn.f32.bf16 %r386, %rs80, %rs80, %r385; + fma.rn.f32.bf16 %r387, %rs81, %rs81, %r311; + fma.rn.f32.bf16 %r388, %rs83, %rs83, %r387; + fma.rn.f32.bf16 %r389, %rs84, %rs84, %r388; + fma.rn.f32.bf16 %r390, %rs85, %rs85, %r389; + fma.rn.f32.bf16 %r391, %rs86, %rs86, %r390; + fma.rn.f32.bf16 %r392, %rs87, %rs87, %r391; + fma.rn.f32.bf16 %r393, %rs88, %rs88, %r392; + fma.rn.f32.bf16 %r394, %rs89, %rs89, %r312; + fma.rn.f32.bf16 %r395, %rs91, %rs91, %r394; + fma.rn.f32.bf16 %r396, %rs92, %rs92, %r395; + fma.rn.f32.bf16 %r397, %rs93, %rs93, %r396; + fma.rn.f32.bf16 %r398, %rs94, %rs94, %r397; + fma.rn.f32.bf16 %r399, %rs95, %rs95, %r398; + fma.rn.f32.bf16 %r400, %rs96, %rs96, %r399; + fma.rn.f32.bf16 %r401, %rs97, %rs97, %r313; + fma.rn.f32.bf16 %r402, %rs99, %rs99, %r401; + fma.rn.f32.bf16 %r403, %rs100, %rs100, %r402; + fma.rn.f32.bf16 %r404, %rs101, %rs101, %r403; + fma.rn.f32.bf16 %r405, %rs102, %rs102, %r404; + fma.rn.f32.bf16 %r406, %rs103, %rs103, %r405; + fma.rn.f32.bf16 %r407, %rs104, %rs104, %r406; + fma.rn.f32.bf16 %r408, %rs105, %rs105, %r314; + fma.rn.f32.bf16 %r409, %rs107, %rs107, %r408; + fma.rn.f32.bf16 %r410, %rs108, %rs108, %r409; + fma.rn.f32.bf16 %r411, %rs109, %rs109, %r410; + fma.rn.f32.bf16 %r412, %rs110, %rs110, %r411; + fma.rn.f32.bf16 %r413, %rs111, %rs111, %r412; + fma.rn.f32.bf16 %r414, %rs112, %rs112, %r413; + fma.rn.f32.bf16 %r415, %rs113, %rs113, %r315; + fma.rn.f32.bf16 %r416, %rs115, %rs115, %r415; + fma.rn.f32.bf16 %r417, %rs116, %rs116, %r416; + fma.rn.f32.bf16 %r418, %rs117, %rs117, %r417; + fma.rn.f32.bf16 %r419, %rs118, %rs118, %r418; + fma.rn.f32.bf16 %r420, %rs119, %rs119, %r419; + fma.rn.f32.bf16 %r421, %rs120, %rs120, %r420; + fma.rn.f32.bf16 %r422, %rs121, %rs121, %r316; + fma.rn.f32.bf16 %r423, %rs123, %rs123, %r422; + fma.rn.f32.bf16 %r424, %rs124, %rs124, %r423; + fma.rn.f32.bf16 %r425, %rs125, %rs125, %r424; + fma.rn.f32.bf16 %r426, %rs126, %rs126, %r425; + fma.rn.f32.bf16 %r427, %rs127, %rs127, %r426; + fma.rn.f32.bf16 %r428, %rs128, %rs128, %r427; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r429, %r323, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r430, %r323, %r429; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r431, %r430, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r432, %r430, %r431; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r433, %r432, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r434, %r432, %r433; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r435, %r434, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r436, %r434, %r435; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r437, %r330, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r438, %r330, %r437; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r439, %r438, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r440, %r438, %r439; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r441, %r440, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r442, %r440, %r441; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r443, %r442, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r444, %r442, %r443; +$L__tmp18: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r445, %r337, 8, 31, -1; +$L__tmp19: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r446, %r337, %r445; +$L__tmp20: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r447, %r446, 4, 31, -1; +$L__tmp21: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r448, %r446, %r447; +$L__tmp22: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r449, %r448, 2, 31, -1; +$L__tmp23: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r450, %r448, %r449; +$L__tmp24: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r451, %r450, 1, 31, -1; +$L__tmp25: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r452, %r450, %r451; +$L__tmp26: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r453, %r344, 8, 31, -1; +$L__tmp27: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r454, %r344, %r453; +$L__tmp28: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r455, %r454, 4, 31, -1; +$L__tmp29: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r456, %r454, %r455; +$L__tmp30: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r457, %r456, 2, 31, -1; +$L__tmp31: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r458, %r456, %r457; +$L__tmp32: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r459, %r458, 1, 31, -1; +$L__tmp33: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r460, %r458, %r459; +$L__tmp34: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r461, %r351, 8, 31, -1; +$L__tmp35: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r462, %r351, %r461; +$L__tmp36: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r463, %r462, 4, 31, -1; +$L__tmp37: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r464, %r462, %r463; +$L__tmp38: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r465, %r464, 2, 31, -1; +$L__tmp39: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r466, %r464, %r465; +$L__tmp40: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r467, %r466, 1, 31, -1; +$L__tmp41: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r468, %r466, %r467; +$L__tmp42: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r469, %r358, 8, 31, -1; +$L__tmp43: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r470, %r358, %r469; +$L__tmp44: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r471, %r470, 4, 31, -1; +$L__tmp45: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r472, %r470, %r471; +$L__tmp46: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r473, %r472, 2, 31, -1; +$L__tmp47: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r474, %r472, %r473; +$L__tmp48: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r475, %r474, 1, 31, -1; +$L__tmp49: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r476, %r474, %r475; +$L__tmp50: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r477, %r365, 8, 31, -1; +$L__tmp51: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r478, %r365, %r477; +$L__tmp52: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r479, %r478, 4, 31, -1; +$L__tmp53: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r480, %r478, %r479; +$L__tmp54: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r481, %r480, 2, 31, -1; +$L__tmp55: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r482, %r480, %r481; +$L__tmp56: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r483, %r482, 1, 31, -1; +$L__tmp57: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r484, %r482, %r483; +$L__tmp58: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r485, %r372, 8, 31, -1; +$L__tmp59: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r486, %r372, %r485; +$L__tmp60: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r487, %r486, 4, 31, -1; +$L__tmp61: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r488, %r486, %r487; +$L__tmp62: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r489, %r488, 2, 31, -1; +$L__tmp63: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r490, %r488, %r489; +$L__tmp64: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r491, %r490, 1, 31, -1; +$L__tmp65: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r492, %r490, %r491; +$L__tmp66: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r493, %r379, 8, 31, -1; +$L__tmp67: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r494, %r379, %r493; +$L__tmp68: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r495, %r494, 4, 31, -1; +$L__tmp69: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r496, %r494, %r495; +$L__tmp70: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r497, %r496, 2, 31, -1; +$L__tmp71: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r498, %r496, %r497; +$L__tmp72: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r499, %r498, 1, 31, -1; +$L__tmp73: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r500, %r498, %r499; +$L__tmp74: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r501, %r386, 8, 31, -1; +$L__tmp75: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r502, %r386, %r501; +$L__tmp76: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r503, %r502, 4, 31, -1; +$L__tmp77: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r504, %r502, %r503; +$L__tmp78: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r505, %r504, 2, 31, -1; +$L__tmp79: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r506, %r504, %r505; +$L__tmp80: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r507, %r506, 1, 31, -1; +$L__tmp81: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r508, %r506, %r507; +$L__tmp82: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r509, %r393, 8, 31, -1; +$L__tmp83: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r510, %r393, %r509; +$L__tmp84: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r511, %r510, 4, 31, -1; +$L__tmp85: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r512, %r510, %r511; +$L__tmp86: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r513, %r512, 2, 31, -1; +$L__tmp87: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r514, %r512, %r513; +$L__tmp88: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r515, %r514, 1, 31, -1; +$L__tmp89: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r516, %r514, %r515; +$L__tmp90: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r517, %r400, 8, 31, -1; +$L__tmp91: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r518, %r400, %r517; +$L__tmp92: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r519, %r518, 4, 31, -1; +$L__tmp93: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r520, %r518, %r519; +$L__tmp94: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r521, %r520, 2, 31, -1; +$L__tmp95: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r522, %r520, %r521; +$L__tmp96: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r523, %r522, 1, 31, -1; +$L__tmp97: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r524, %r522, %r523; +$L__tmp98: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r525, %r407, 8, 31, -1; +$L__tmp99: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r526, %r407, %r525; +$L__tmp100: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r527, %r526, 4, 31, -1; +$L__tmp101: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r528, %r526, %r527; +$L__tmp102: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r529, %r528, 2, 31, -1; +$L__tmp103: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r530, %r528, %r529; +$L__tmp104: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r531, %r530, 1, 31, -1; +$L__tmp105: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r532, %r530, %r531; +$L__tmp106: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r533, %r414, 8, 31, -1; +$L__tmp107: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r534, %r414, %r533; +$L__tmp108: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r535, %r534, 4, 31, -1; +$L__tmp109: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r536, %r534, %r535; +$L__tmp110: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r537, %r536, 2, 31, -1; +$L__tmp111: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r538, %r536, %r537; +$L__tmp112: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r539, %r538, 1, 31, -1; +$L__tmp113: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r540, %r538, %r539; +$L__tmp114: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r541, %r421, 8, 31, -1; +$L__tmp115: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r542, %r421, %r541; +$L__tmp116: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r543, %r542, 4, 31, -1; +$L__tmp117: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r544, %r542, %r543; +$L__tmp118: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r545, %r544, 2, 31, -1; +$L__tmp119: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r546, %r544, %r545; +$L__tmp120: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r547, %r546, 1, 31, -1; +$L__tmp121: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r548, %r546, %r547; +$L__tmp122: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r549, %r428, 8, 31, -1; +$L__tmp123: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r550, %r428, %r549; +$L__tmp124: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r551, %r550, 4, 31, -1; +$L__tmp125: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r552, %r550, %r551; +$L__tmp126: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r553, %r552, 2, 31, -1; +$L__tmp127: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r554, %r552, %r553; +$L__tmp128: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r555, %r554, 1, 31, -1; +$L__tmp129: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r556, %r554, %r555; +$L__tmp130: + .loc 1 80 48 // l2norm.py:80:48 + add.f32 %r557, %r152, %r436; + add.f32 %r558, %r152, %r444; + add.f32 %r559, %r152, %r452; + add.f32 %r560, %r152, %r460; + add.f32 %r561, %r152, %r468; + add.f32 %r562, %r152, %r476; + add.f32 %r563, %r152, %r484; + add.f32 %r564, %r152, %r492; + add.f32 %r565, %r152, %r500; + add.f32 %r566, %r152, %r508; + add.f32 %r567, %r152, %r516; + add.f32 %r568, %r152, %r524; + add.f32 %r569, %r152, %r532; + add.f32 %r570, %r152, %r540; + add.f32 %r571, %r152, %r548; + add.f32 %r572, %r152, %r556; + .loc 1 80 25 // l2norm.py:80:25 + sqrt.approx.ftz.f32 %r573, %r557; + sqrt.approx.ftz.f32 %r574, %r558; + sqrt.approx.ftz.f32 %r575, %r559; + sqrt.approx.ftz.f32 %r576, %r560; + sqrt.approx.ftz.f32 %r577, %r561; + sqrt.approx.ftz.f32 %r578, %r562; + sqrt.approx.ftz.f32 %r579, %r563; + sqrt.approx.ftz.f32 %r580, %r564; + sqrt.approx.ftz.f32 %r581, %r565; + sqrt.approx.ftz.f32 %r582, %r566; + sqrt.approx.ftz.f32 %r583, %r567; + sqrt.approx.ftz.f32 %r584, %r568; + sqrt.approx.ftz.f32 %r585, %r569; + sqrt.approx.ftz.f32 %r586, %r570; + sqrt.approx.ftz.f32 %r587, %r571; + sqrt.approx.ftz.f32 %r588, %r572; + .loc 1 71 34 // l2norm.py:71:34 + and.b32 %r589, %r153, 127; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd104, %r589; + or.b64 %rd105, %rd37, %rd104; + .loc 1 77 20 // l2norm.py:77:20 + shl.b64 %rd106, %rd105, 2; + add.s64 %rd33, %rd36, %rd106; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd107, %rd35, %rd101; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd32, %rd107, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd108, %rd35, %rd99; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd31, %rd108, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd109, %rd35, %rd97; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd30, %rd109, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd110, %rd35, %rd95; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd29, %rd110, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd111, %rd35, %rd93; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd28, %rd111, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd112, %rd35, %rd91; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd27, %rd112, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd113, %rd35, %rd89; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd26, %rd113, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd114, %rd35, %rd87; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd25, %rd114, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd115, %rd35, %rd85; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd24, %rd115, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd116, %rd35, %rd83; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd23, %rd116, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd117, %rd35, %rd81; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd22, %rd117, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd118, %rd35, %rd79; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd21, %rd118, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd119, %rd35, %rd77; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd20, %rd119, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd120, %rd35, %rd75; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd19, %rd120, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd121, %rd35, %rd73; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd18, %rd121, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd122, %rd35, %rd71; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd17, %rd122, %rd103; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p17, %rd105, %rd38; + mov.b32 %r590, 0f3F800000; + .loc 1 80 17 // l2norm.py:80:17 + div.full.f32 %r131, %r590, %r573; + div.full.f32 %r132, %r590, %r574; + div.full.f32 %r133, %r590, %r575; + div.full.f32 %r134, %r590, %r576; + div.full.f32 %r136, %r590, %r577; + div.full.f32 %r137, %r590, %r578; + div.full.f32 %r138, %r590, %r579; + div.full.f32 %r139, %r590, %r580; + div.full.f32 %r141, %r590, %r581; + div.full.f32 %r142, %r590, %r582; + div.full.f32 %r143, %r590, %r583; + div.full.f32 %r144, %r590, %r584; + div.full.f32 %r146, %r590, %r585; + div.full.f32 %r147, %r590, %r586; + div.full.f32 %r148, %r590, %r587; + div.full.f32 %r149, %r590, %r588; + .loc 1 81 16 // l2norm.py:81:16 + mul.f32 %r591, %r131, %r173; + mul.f32 %r592, %r131, %r174; + mul.f32 %r593, %r131, %r175; + mul.f32 %r594, %r131, %r176; + mul.f32 %r595, %r131, %r177; + mul.f32 %r596, %r131, %r178; + mul.f32 %r597, %r131, %r179; + mul.f32 %r598, %r131, %r180; + mul.f32 %r599, %r132, %r181; + mul.f32 %r600, %r132, %r182; + mul.f32 %r601, %r132, %r183; + mul.f32 %r602, %r132, %r184; + mul.f32 %r603, %r132, %r185; + mul.f32 %r604, %r132, %r186; + mul.f32 %r605, %r132, %r187; + mul.f32 %r606, %r132, %r188; + mul.f32 %r607, %r133, %r189; + mul.f32 %r608, %r133, %r190; + mul.f32 %r609, %r133, %r191; + mul.f32 %r610, %r133, %r192; + mul.f32 %r611, %r133, %r193; + mul.f32 %r612, %r133, %r194; + mul.f32 %r613, %r133, %r195; + mul.f32 %r614, %r133, %r196; + mul.f32 %r615, %r134, %r197; + mul.f32 %r616, %r134, %r198; + mul.f32 %r617, %r134, %r199; + mul.f32 %r618, %r134, %r200; + mul.f32 %r619, %r134, %r201; + mul.f32 %r620, %r134, %r202; + mul.f32 %r621, %r134, %r203; + mul.f32 %r622, %r134, %r204; + mul.f32 %r623, %r136, %r205; + mul.f32 %r624, %r136, %r206; + mul.f32 %r625, %r136, %r207; + mul.f32 %r626, %r136, %r208; + mul.f32 %r627, %r136, %r209; + mul.f32 %r628, %r136, %r210; + mul.f32 %r629, %r136, %r211; + mul.f32 %r630, %r136, %r212; + mul.f32 %r631, %r137, %r213; + mul.f32 %r632, %r137, %r214; + mul.f32 %r633, %r137, %r215; + mul.f32 %r634, %r137, %r216; + mul.f32 %r635, %r137, %r217; + mul.f32 %r636, %r137, %r218; + mul.f32 %r637, %r137, %r219; + mul.f32 %r638, %r137, %r220; + mul.f32 %r639, %r138, %r221; + mul.f32 %r640, %r138, %r222; + mul.f32 %r641, %r138, %r223; + mul.f32 %r642, %r138, %r224; + mul.f32 %r643, %r138, %r225; + mul.f32 %r644, %r138, %r226; + mul.f32 %r645, %r138, %r227; + mul.f32 %r646, %r138, %r228; + mul.f32 %r647, %r139, %r229; + mul.f32 %r648, %r139, %r230; + mul.f32 %r649, %r139, %r231; + mul.f32 %r650, %r139, %r232; + mul.f32 %r651, %r139, %r233; + mul.f32 %r652, %r139, %r234; + mul.f32 %r653, %r139, %r235; + mul.f32 %r654, %r139, %r236; + mul.f32 %r655, %r141, %r237; + mul.f32 %r656, %r141, %r238; + mul.f32 %r657, %r141, %r239; + mul.f32 %r658, %r141, %r240; + mul.f32 %r659, %r141, %r241; + mul.f32 %r660, %r141, %r242; + mul.f32 %r661, %r141, %r243; + mul.f32 %r662, %r141, %r244; + mul.f32 %r663, %r142, %r245; + mul.f32 %r664, %r142, %r246; + mul.f32 %r665, %r142, %r247; + mul.f32 %r666, %r142, %r248; + mul.f32 %r667, %r142, %r249; + mul.f32 %r668, %r142, %r250; + mul.f32 %r669, %r142, %r251; + mul.f32 %r670, %r142, %r252; + mul.f32 %r671, %r143, %r253; + mul.f32 %r672, %r143, %r254; + mul.f32 %r673, %r143, %r255; + mul.f32 %r674, %r143, %r256; + mul.f32 %r675, %r143, %r257; + mul.f32 %r676, %r143, %r258; + mul.f32 %r677, %r143, %r259; + mul.f32 %r678, %r143, %r260; + mul.f32 %r679, %r144, %r261; + mul.f32 %r680, %r144, %r262; + mul.f32 %r681, %r144, %r263; + mul.f32 %r682, %r144, %r264; + mul.f32 %r683, %r144, %r265; + mul.f32 %r684, %r144, %r266; + mul.f32 %r685, %r144, %r267; + mul.f32 %r686, %r144, %r268; + mul.f32 %r687, %r146, %r269; + mul.f32 %r688, %r146, %r270; + mul.f32 %r689, %r146, %r271; + mul.f32 %r690, %r146, %r272; + mul.f32 %r691, %r146, %r273; + mul.f32 %r692, %r146, %r274; + mul.f32 %r693, %r146, %r275; + mul.f32 %r694, %r146, %r276; + mul.f32 %r695, %r147, %r277; + mul.f32 %r696, %r147, %r278; + mul.f32 %r697, %r147, %r279; + mul.f32 %r698, %r147, %r280; + mul.f32 %r699, %r147, %r281; + mul.f32 %r700, %r147, %r282; + mul.f32 %r701, %r147, %r283; + mul.f32 %r702, %r147, %r284; + mul.f32 %r703, %r148, %r285; + mul.f32 %r704, %r148, %r286; + mul.f32 %r705, %r148, %r287; + mul.f32 %r706, %r148, %r288; + mul.f32 %r707, %r148, %r289; + mul.f32 %r708, %r148, %r290; + mul.f32 %r709, %r148, %r291; + mul.f32 %r710, %r148, %r292; + mul.f32 %r711, %r149, %r293; + mul.f32 %r712, %r149, %r294; + mul.f32 %r713, %r149, %r295; + mul.f32 %r714, %r149, %r296; + mul.f32 %r715, %r149, %r297; + mul.f32 %r716, %r149, %r298; + mul.f32 %r717, %r149, %r299; + mul.f32 %r718, %r149, %r300; + .loc 1 83 18 // l2norm.py:83:18 + cvt.rn.bf16x2.f32 %r66, %r592, %r591; + cvt.rn.bf16x2.f32 %r67, %r594, %r593; + cvt.rn.bf16x2.f32 %r68, %r596, %r595; + cvt.rn.bf16x2.f32 %r69, %r598, %r597; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd17 + 0 ], { %r66, %r67, %r68, %r69 }; + // end inline asm + cvt.rn.bf16x2.f32 %r70, %r600, %r599; + cvt.rn.bf16x2.f32 %r71, %r602, %r601; + cvt.rn.bf16x2.f32 %r72, %r604, %r603; + cvt.rn.bf16x2.f32 %r73, %r606, %r605; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd18 + 0 ], { %r70, %r71, %r72, %r73 }; + // end inline asm + cvt.rn.bf16x2.f32 %r74, %r608, %r607; + cvt.rn.bf16x2.f32 %r75, %r610, %r609; + cvt.rn.bf16x2.f32 %r76, %r612, %r611; + cvt.rn.bf16x2.f32 %r77, %r614, %r613; + // begin inline asm + @%p3 st.global.v4.b32 [ %rd19 + 0 ], { %r74, %r75, %r76, %r77 }; + // end inline asm + cvt.rn.bf16x2.f32 %r78, %r616, %r615; + cvt.rn.bf16x2.f32 %r79, %r618, %r617; + cvt.rn.bf16x2.f32 %r80, %r620, %r619; + cvt.rn.bf16x2.f32 %r81, %r622, %r621; + // begin inline asm + @%p4 st.global.v4.b32 [ %rd20 + 0 ], { %r78, %r79, %r80, %r81 }; + // end inline asm + cvt.rn.bf16x2.f32 %r82, %r624, %r623; + cvt.rn.bf16x2.f32 %r83, %r626, %r625; + cvt.rn.bf16x2.f32 %r84, %r628, %r627; + cvt.rn.bf16x2.f32 %r85, %r630, %r629; + // begin inline asm + @%p5 st.global.v4.b32 [ %rd21 + 0 ], { %r82, %r83, %r84, %r85 }; + // end inline asm + cvt.rn.bf16x2.f32 %r86, %r632, %r631; + cvt.rn.bf16x2.f32 %r87, %r634, %r633; + cvt.rn.bf16x2.f32 %r88, %r636, %r635; + cvt.rn.bf16x2.f32 %r89, %r638, %r637; + // begin inline asm + @%p6 st.global.v4.b32 [ %rd22 + 0 ], { %r86, %r87, %r88, %r89 }; + // end inline asm + cvt.rn.bf16x2.f32 %r90, %r640, %r639; + cvt.rn.bf16x2.f32 %r91, %r642, %r641; + cvt.rn.bf16x2.f32 %r92, %r644, %r643; + cvt.rn.bf16x2.f32 %r93, %r646, %r645; + // begin inline asm + @%p7 st.global.v4.b32 [ %rd23 + 0 ], { %r90, %r91, %r92, %r93 }; + // end inline asm + cvt.rn.bf16x2.f32 %r94, %r648, %r647; + cvt.rn.bf16x2.f32 %r95, %r650, %r649; + cvt.rn.bf16x2.f32 %r96, %r652, %r651; + cvt.rn.bf16x2.f32 %r97, %r654, %r653; + // begin inline asm + @%p15 st.global.v4.b32 [ %rd24 + 0 ], { %r94, %r95, %r96, %r97 }; + // end inline asm + cvt.rn.bf16x2.f32 %r98, %r656, %r655; + cvt.rn.bf16x2.f32 %r99, %r658, %r657; + cvt.rn.bf16x2.f32 %r100, %r660, %r659; + cvt.rn.bf16x2.f32 %r101, %r662, %r661; + // begin inline asm + @%p8 st.global.v4.b32 [ %rd25 + 0 ], { %r98, %r99, %r100, %r101 }; + // end inline asm + cvt.rn.bf16x2.f32 %r102, %r664, %r663; + cvt.rn.bf16x2.f32 %r103, %r666, %r665; + cvt.rn.bf16x2.f32 %r104, %r668, %r667; + cvt.rn.bf16x2.f32 %r105, %r670, %r669; + // begin inline asm + @%p9 st.global.v4.b32 [ %rd26 + 0 ], { %r102, %r103, %r104, %r105 }; + // end inline asm + cvt.rn.bf16x2.f32 %r106, %r672, %r671; + cvt.rn.bf16x2.f32 %r107, %r674, %r673; + cvt.rn.bf16x2.f32 %r108, %r676, %r675; + cvt.rn.bf16x2.f32 %r109, %r678, %r677; + // begin inline asm + @%p10 st.global.v4.b32 [ %rd27 + 0 ], { %r106, %r107, %r108, %r109 }; + // end inline asm + cvt.rn.bf16x2.f32 %r110, %r680, %r679; + cvt.rn.bf16x2.f32 %r111, %r682, %r681; + cvt.rn.bf16x2.f32 %r112, %r684, %r683; + cvt.rn.bf16x2.f32 %r113, %r686, %r685; + // begin inline asm + @%p11 st.global.v4.b32 [ %rd28 + 0 ], { %r110, %r111, %r112, %r113 }; + // end inline asm + cvt.rn.bf16x2.f32 %r114, %r688, %r687; + cvt.rn.bf16x2.f32 %r115, %r690, %r689; + cvt.rn.bf16x2.f32 %r116, %r692, %r691; + cvt.rn.bf16x2.f32 %r117, %r694, %r693; + // begin inline asm + @%p12 st.global.v4.b32 [ %rd29 + 0 ], { %r114, %r115, %r116, %r117 }; + // end inline asm + cvt.rn.bf16x2.f32 %r118, %r696, %r695; + cvt.rn.bf16x2.f32 %r119, %r698, %r697; + cvt.rn.bf16x2.f32 %r120, %r700, %r699; + cvt.rn.bf16x2.f32 %r121, %r702, %r701; + // begin inline asm + @%p13 st.global.v4.b32 [ %rd30 + 0 ], { %r118, %r119, %r120, %r121 }; + // end inline asm + cvt.rn.bf16x2.f32 %r122, %r704, %r703; + cvt.rn.bf16x2.f32 %r123, %r706, %r705; + cvt.rn.bf16x2.f32 %r124, %r708, %r707; + cvt.rn.bf16x2.f32 %r125, %r710, %r709; + // begin inline asm + @%p14 st.global.v4.b32 [ %rd31 + 0 ], { %r122, %r123, %r124, %r125 }; + // end inline asm + cvt.rn.bf16x2.f32 %r126, %r712, %r711; + cvt.rn.bf16x2.f32 %r127, %r714, %r713; + cvt.rn.bf16x2.f32 %r128, %r716, %r715; + cvt.rn.bf16x2.f32 %r129, %r718, %r717; + // begin inline asm + @%p16 st.global.v4.b32 [ %rd32 + 0 ], { %r126, %r127, %r128, %r129 }; + // end inline asm + .loc 1 84 21 // l2norm.py:84:21 + mov.b32 %r719, global_smem; + add.s32 %r130, %r719, %r154; + // begin inline asm + st.shared.v4.b32 [ %r130 + 0 ], { %r131, %r132, %r133, %r134 }; + // end inline asm + add.s32 %r135, %r130, 128; + // begin inline asm + st.shared.v4.b32 [ %r135 + 0 ], { %r136, %r137, %r138, %r139 }; + // end inline asm + add.s32 %r140, %r130, 256; + // begin inline asm + st.shared.v4.b32 [ %r140 + 0 ], { %r141, %r142, %r143, %r144 }; + // end inline asm + add.s32 %r145, %r130, 384; + // begin inline asm + st.shared.v4.b32 [ %r145 + 0 ], { %r146, %r147, %r148, %r149 }; + // end inline asm + bar.sync 0; + shl.b32 %r720, %r153, 4; + and.b32 %r721, %r720, 112; + shl.b32 %r722, %r153, 2; + and.b32 %r723, %r722, 384; + shr.u32 %r724, %r153, 1; + and.b32 %r725, %r724, 12; + add.s32 %r726, %r719, %r721; + add.s32 %r727, %r726, %r723; + add.s32 %r728, %r727, %r725; + ld.shared.b32 %r150, [%r728]; + // begin inline asm + @%p17 st.global.b32 [ %rd33 + 0 ], { %r150 }; + // end inline asm + .loc 1 84 4 // l2norm.py:84:4 + ret; +$L__tmp131: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp130 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 80 // DW_AT_call_line +.b8 43 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp130 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/l2norm_fwd_kernel.source b/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/l2norm_fwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..65999ed34486c21231e2e10dd8e7d7c057966940 --- /dev/null +++ b/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/l2norm_fwd_kernel.source @@ -0,0 +1,172 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc38 = loc(unknown) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc45 = loc("x"(#loc)) +#loc46 = loc("y"(#loc)) +#loc47 = loc("rstd"(#loc)) +#loc48 = loc("eps"(#loc)) +#loc49 = loc("T"(#loc)) +#loc81 = loc("input"(#loc36)) +#loc82 = loc("a"(#loc41)) +#loc83 = loc("b"(#loc41)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc50) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc51) + %o_t = arith.constant 128 : i32 loc(#loc52) + %o_t_1 = arith.constant 128 : i64 loc(#loc52) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc52) + %o_t_3 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc53) + %o_t_4 = arith.extsi %o_t_3 : tensor<128xi32> to tensor<128xi64> loc(#loc54) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<128xi64> loc(#loc54) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<128xi64> loc(#loc54) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc55) + %m_t = arith.extsi %T : i32 to i64 loc(#loc56) + %m_t_7 = tt.splat %m_t : i64 -> tensor<128xi64> loc(#loc56) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<128xi64> loc(#loc56) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<128xi1> -> tensor<128x1xi1> loc(#loc57) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc58) + %m_x_10 = arith.constant 128 : i32 loc(#loc59) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc59) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc59) + %m_x_13 = tt.broadcast %m_x : tensor<128x1xi1> -> tensor<128x128xi1> loc(#loc60) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<128x128xi1> loc(#loc60) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<128x128xi1> loc(#loc60) + %p_x = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc61) + %p_x_16 = arith.constant 128 : i32 loc(#loc62) + %p_x_17 = arith.constant 128 : i64 loc(#loc62) + %p_x_18 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc62) + %p_x_19 = arith.muli %p_x, %p_x_18 : tensor<128x1xi64> loc(#loc62) + %p_x_20 = tt.splat %x : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc63) + %p_x_21 = tt.addptr %p_x_20, %p_x_19 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc63) + %p_x_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc64) + %p_x_23 = tt.broadcast %p_x_21 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc65) + %p_x_24 = tt.broadcast %p_x_22 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc65) + %p_x_25 = tt.addptr %p_x_23, %p_x_24 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc65) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc66) + %p_y_26 = arith.constant 128 : i32 loc(#loc67) + %p_y_27 = arith.constant 128 : i64 loc(#loc67) + %p_y_28 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc67) + %p_y_29 = arith.muli %p_y, %p_y_28 : tensor<128x1xi64> loc(#loc67) + %p_y_30 = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc68) + %p_y_31 = tt.addptr %p_y_30, %p_y_29 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc68) + %p_y_32 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc69) + %p_y_33 = tt.broadcast %p_y_31 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc70) + %p_y_34 = tt.broadcast %p_y_32 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc70) + %p_y_35 = tt.addptr %p_y_33, %p_y_34 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc70) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr> loc(#loc71) + %p_rstd_36 = tt.addptr %p_rstd, %o_t_6 : tensor<128x!tt.ptr>, tensor<128xi64> loc(#loc71) + %b_x = arith.constant 0.000000e+00 : f32 loc(#loc72) + %b_x_37 = arith.constant dense<0.000000e+00> : tensor<128x128xf32> loc(#loc72) + %b_x_38 = arith.truncf %b_x_37 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc72) + %b_x_39 = tt.load %p_x_25, %m_x_15, %b_x_38 : tensor<128x128x!tt.ptr> loc(#loc72) + %b_x_40 = arith.extf %b_x_39 : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc73) + %b_rstd = arith.mulf %b_x_40, %b_x_40 : tensor<128x128xf32> loc(#loc74) + %b_rstd_41 = tt.call @triton.language.standard.sum__fp32S128_128S_c1_cFalse_cNone(%b_rstd) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc75) + %b_rstd_42 = tt.splat %eps : f32 -> tensor<128xf32> loc(#loc76) + %b_rstd_43 = arith.addf %b_rstd_41, %b_rstd_42 : tensor<128xf32> loc(#loc76) + %b_rstd_44 = math.sqrt %b_rstd_43 : tensor<128xf32> loc(#loc77) + %b_rstd_45 = arith.constant 1 : i32 loc(#loc78) + %b_rstd_46 = arith.constant 1.000000e+00 : f32 loc(#loc78) + %b_rstd_47 = arith.constant dense<1.000000e+00> : tensor<128xf32> loc(#loc78) + %b_rstd_48 = arith.divf %b_rstd_47, %b_rstd_44 : tensor<128xf32> loc(#loc78) + %b_y = tt.expand_dims %b_rstd_48 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc79) + %b_y_49 = tt.broadcast %b_y : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc80) + %b_y_50 = arith.mulf %b_x_40, %b_y_49 : tensor<128x128xf32> loc(#loc80) + %0 = arith.truncf %b_y_50 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc32) + tt.store %p_y_35, %0, %m_x_15 : tensor<128x128x!tt.ptr> loc(#loc33) + tt.store %p_rstd_36, %b_rstd_48, %m_t_8 : tensor<128x!tt.ptr> loc(#loc34) + tt.return loc(#loc35) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S128_128S_c1_cFalse_cNone(%input: tensor<128x128xf32> loc("input"(#loc36))) -> tensor<128xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc37) + tt.reduce.return %2 : f32 loc(#loc37) + }) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc37) + tt.return %0 : tensor<128xf32> loc(#loc39) + ^bb1: // no predecessors + %1 = ub.poison : tensor<128xf32> loc(#loc40) + tt.return %1 : tensor<128xf32> loc(#loc40) + } loc(#loc36) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc41)), %b: f32 loc("b"(#loc41))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc42) + tt.return %0 : f32 loc(#loc43) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc44) + tt.return %1 : f32 loc(#loc44) + } loc(#loc41) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:18) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:29) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:37) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc50 = loc("i_t"(#loc1)) +#loc51 = loc("i_t"(#loc2)) +#loc52 = loc("o_t"(#loc3)) +#loc53 = loc("o_t"(#loc4)) +#loc54 = loc("o_t"(#loc5)) +#loc55 = loc("o_d"(#loc6)) +#loc56 = loc("m_t"(#loc7)) +#loc57 = loc("m_x"(#loc8)) +#loc58 = loc("m_x"(#loc9)) +#loc59 = loc("m_x"(#loc10)) +#loc60 = loc("m_x"(#loc11)) +#loc61 = loc("p_x"(#loc12)) +#loc62 = loc("p_x"(#loc13)) +#loc63 = loc("p_x"(#loc14)) +#loc64 = loc("p_x"(#loc15)) +#loc65 = loc("p_x"(#loc16)) +#loc66 = loc("p_y"(#loc17)) +#loc67 = loc("p_y"(#loc18)) +#loc68 = loc("p_y"(#loc19)) +#loc69 = loc("p_y"(#loc20)) +#loc70 = loc("p_y"(#loc21)) +#loc71 = loc("p_rstd"(#loc22)) +#loc72 = loc("b_x"(#loc23)) +#loc73 = loc("b_x"(#loc24)) +#loc74 = loc("b_rstd"(#loc25)) +#loc75 = loc("b_rstd"(#loc26)) +#loc76 = loc("b_rstd"(#loc27)) +#loc77 = loc("b_rstd"(#loc28)) +#loc78 = loc("b_rstd"(#loc29)) +#loc79 = loc("b_y"(#loc30)) +#loc80 = loc("b_y"(#loc31)) diff --git a/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/l2norm_fwd_kernel.ttgir b/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/l2norm_fwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..7c634642c2b8f78fbf435649e9ae13cf79195ce7 --- /dev/null +++ b/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/l2norm_fwd_kernel.ttgir @@ -0,0 +1,135 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [4, 1], order = [1, 0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [4], order = [0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc1 = loc(unknown) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc34 = loc("x"(#loc)) +#loc35 = loc("y"(#loc)) +#loc36 = loc("rstd"(#loc)) +#loc37 = loc("eps"(#loc)) +#loc38 = loc("T"(#loc)) +#loc59 = loc("b_rstd"(#loc23)) +#loc66 = loc(callsite(#loc1 at #loc59)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 4 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<1.000000e+00> : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<128x128xbf16, #blocked> loc(#loc1) + %c128_i64 = arith.constant 128 : i64 loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<128x1xi64, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc39) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc40) + %o_t = arith.muli %i_t_3, %c128_i64 : i64 loc(#loc41) + %o_t_4 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc42) + %o_t_5 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #blocked1> loc(#loc42) + %o_t_6 = arith.extsi %o_t_4 : tensor<128xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_7 = arith.extsi %o_t_5 : tensor<128xi32, #blocked1> to tensor<128xi64, #blocked1> loc(#loc43) + %o_t_8 = tt.splat %o_t : i64 -> tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_9 = tt.splat %o_t : i64 -> tensor<128xi64, #blocked1> loc(#loc43) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<128xi64, #blocked1> loc(#loc43) + %m_t = arith.extsi %T : i32 to i64 loc(#loc44) + %m_t_12 = tt.splat %m_t : i64 -> tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_13 = tt.splat %m_t : i64 -> tensor<128xi64, #blocked1> loc(#loc44) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<128xi64, #blocked1> loc(#loc44) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<128xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128x1xi1, #blocked> loc(#loc45) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc46) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc46) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_1 : tensor<1x128xi32, #blocked> loc(#loc47) + %m_x_19 = tt.broadcast %m_x : tensor<128x1xi1, #blocked> -> tensor<128x128xi1, #blocked> loc(#loc48) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<128x128xi1, #blocked> loc(#loc48) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<128x128xi1, #blocked> loc(#loc48) + %p_x = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128x1xi64, #blocked> loc(#loc49) + %p_x_22 = arith.muli %p_x, %cst_2 : tensor<128x1xi64, #blocked> loc(#loc50) + %p_x_23 = tt.splat %x : !tt.ptr -> tensor<128x1x!tt.ptr, #blocked> loc(#loc51) + %p_x_24 = tt.addptr %p_x_23, %p_x_22 : tensor<128x1x!tt.ptr, #blocked>, tensor<128x1xi64, #blocked> loc(#loc51) + %p_x_25 = tt.broadcast %p_x_24 : tensor<128x1x!tt.ptr, #blocked> -> tensor<128x128x!tt.ptr, #blocked> loc(#loc52) + %p_x_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<128x128xi32, #blocked> loc(#loc52) + %p_x_27 = tt.addptr %p_x_25, %p_x_26 : tensor<128x128x!tt.ptr, #blocked>, tensor<128x128xi32, #blocked> loc(#loc52) + %p_y = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr, #blocked> loc(#loc53) + %p_y_28 = tt.addptr %p_y, %p_x_22 : tensor<128x1x!tt.ptr, #blocked>, tensor<128x1xi64, #blocked> loc(#loc53) + %p_y_29 = tt.broadcast %p_y_28 : tensor<128x1x!tt.ptr, #blocked> -> tensor<128x128x!tt.ptr, #blocked> loc(#loc54) + %p_y_30 = tt.addptr %p_y_29, %p_x_26 : tensor<128x128x!tt.ptr, #blocked>, tensor<128x128xi32, #blocked> loc(#loc54) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr, #blocked1> loc(#loc55) + %p_rstd_31 = tt.addptr %p_rstd, %o_t_11 : tensor<128x!tt.ptr, #blocked1>, tensor<128xi64, #blocked1> loc(#loc55) + %b_x = tt.load %p_x_27, %m_x_21, %cst_0 : tensor<128x128x!tt.ptr, #blocked> loc(#loc56) + %b_x_32 = arith.extf %b_x : tensor<128x128xbf16, #blocked> to tensor<128x128xf32, #blocked> loc(#loc57) + %b_rstd = arith.mulf %b_x_32, %b_x_32 : tensor<128x128xf32, #blocked> loc(#loc58) + %b_rstd_33 = "tt.reduce"(%b_rstd) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_40: f32 loc(callsite(#loc1 at #loc59)), %b_rstd_41: f32 loc(callsite(#loc1 at #loc59))): + %b_rstd_42 = arith.addf %b_rstd_40, %b_rstd_41 : f32 loc(#loc67) + tt.reduce.return %b_rstd_42 : f32 loc(#loc65) + }) : (tensor<128x128xf32, #blocked>) -> tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc65) + %b_rstd_34 = tt.splat %eps : f32 -> tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_36 = math.sqrt %b_rstd_35 : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc61) + %b_rstd_37 = arith.divf %cst, %b_rstd_36 : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc62) + %b_y = tt.expand_dims %b_rstd_37 {axis = 1 : i32} : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128x1xf32, #blocked> loc(#loc63) + %b_y_38 = tt.broadcast %b_y : tensor<128x1xf32, #blocked> -> tensor<128x128xf32, #blocked> loc(#loc64) + %b_y_39 = arith.mulf %b_x_32, %b_y_38 : tensor<128x128xf32, #blocked> loc(#loc64) + %0 = arith.truncf %b_y_39 : tensor<128x128xf32, #blocked> to tensor<128x128xbf16, #blocked> loc(#loc30) + tt.store %p_y_30, %0, %m_x_21 : tensor<128x128x!tt.ptr, #blocked> loc(#loc31) + %1 = ttg.convert_layout %b_rstd_37 : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128xf32, #blocked1> loc(#loc32) + tt.store %p_rstd_31, %1, %m_t_15 : tensor<128x!tt.ptr, #blocked1> loc(#loc32) + tt.return loc(#loc33) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc39 = loc("i_t"(#loc2)) +#loc40 = loc("i_t"(#loc3)) +#loc41 = loc("o_t"(#loc4)) +#loc42 = loc("o_t"(#loc5)) +#loc43 = loc("o_t"(#loc6)) +#loc44 = loc("m_t"(#loc7)) +#loc45 = loc("m_x"(#loc8)) +#loc46 = loc("m_x"(#loc9)) +#loc47 = loc("m_x"(#loc10)) +#loc48 = loc("m_x"(#loc11)) +#loc49 = loc("p_x"(#loc12)) +#loc50 = loc("p_x"(#loc13)) +#loc51 = loc("p_x"(#loc14)) +#loc52 = loc("p_x"(#loc15)) +#loc53 = loc("p_y"(#loc16)) +#loc54 = loc("p_y"(#loc17)) +#loc55 = loc("p_rstd"(#loc18)) +#loc56 = loc("b_x"(#loc19)) +#loc57 = loc("b_x"(#loc20)) +#loc58 = loc("b_rstd"(#loc21)) +#loc60 = loc("b_rstd"(#loc25)) +#loc61 = loc("b_rstd"(#loc26)) +#loc62 = loc("b_rstd"(#loc27)) +#loc63 = loc("b_y"(#loc28)) +#loc64 = loc("b_y"(#loc29)) +#loc65 = loc(callsite(#loc22 at #loc59)) +#loc67 = loc(callsite(#loc24 at #loc65)) diff --git a/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/l2norm_fwd_kernel.ttir b/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/l2norm_fwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..8f0dfe3371851abdc6b1c20ee94003c63d3c42a8 --- /dev/null +++ b/triton/gpu0/BAE7FFHE7FXLBJTXC7YGNLOJDTJLXSS6MEIUD65SHS4IR4SFUKBA/l2norm_fwd_kernel.ttir @@ -0,0 +1,125 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc3 = loc(unknown) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc34 = loc("x"(#loc)) +#loc35 = loc("y"(#loc)) +#loc36 = loc("rstd"(#loc)) +#loc37 = loc("eps"(#loc)) +#loc38 = loc("T"(#loc)) +#loc60 = loc("b_rstd"(#loc24)) +#loc66 = loc(callsite(#loc3 at #loc60)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %b_x = arith.constant dense<0.000000e+00> : tensor<128x128xbf16> loc(#loc39) + %b_rstd = arith.constant dense<1.000000e+00> : tensor<128xf32> loc(#loc40) + %cst = arith.constant dense<128> : tensor<128x1xi64> loc(#loc3) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc41) + %c128_i64 = arith.constant 128 : i64 loc(#loc3) + %i_t = tt.get_program_id x : i32 loc(#loc42) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc43) + %o_t = arith.muli %i_t_0, %c128_i64 : i64 loc(#loc44) + %o_t_1 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc45) + %o_t_2 = arith.extsi %o_t_1 : tensor<128xi32> to tensor<128xi64> loc(#loc46) + %o_t_3 = tt.splat %o_t : i64 -> tensor<128xi64> loc(#loc46) + %o_t_4 = arith.addi %o_t_3, %o_t_2 : tensor<128xi64> loc(#loc46) + %m_t = arith.extsi %T : i32 to i64 loc(#loc47) + %m_t_5 = tt.splat %m_t : i64 -> tensor<128xi64> loc(#loc47) + %m_t_6 = arith.cmpi slt, %o_t_4, %m_t_5 : tensor<128xi64> loc(#loc47) + %m_x_7 = tt.expand_dims %m_t_6 {axis = 1 : i32} : tensor<128xi1> -> tensor<128x1xi1> loc(#loc48) + %m_x_8 = tt.expand_dims %o_t_1 {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc49) + %m_x_9 = arith.cmpi slt, %m_x_8, %m_x : tensor<1x128xi32> loc(#loc41) + %m_x_10 = tt.broadcast %m_x_7 : tensor<128x1xi1> -> tensor<128x128xi1> loc(#loc50) + %m_x_11 = tt.broadcast %m_x_9 : tensor<1x128xi1> -> tensor<128x128xi1> loc(#loc50) + %m_x_12 = arith.andi %m_x_10, %m_x_11 : tensor<128x128xi1> loc(#loc50) + %p_x = tt.expand_dims %o_t_4 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc51) + %p_x_13 = arith.muli %p_x, %cst : tensor<128x1xi64> loc(#loc52) + %p_x_14 = tt.splat %x : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc53) + %p_x_15 = tt.addptr %p_x_14, %p_x_13 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc53) + %p_x_16 = tt.broadcast %p_x_15 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc54) + %p_x_17 = tt.broadcast %m_x_8 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc54) + %p_x_18 = tt.addptr %p_x_16, %p_x_17 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc54) + %p_y = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc55) + %p_y_19 = tt.addptr %p_y, %p_x_13 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc55) + %p_y_20 = tt.broadcast %p_y_19 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc56) + %p_y_21 = tt.addptr %p_y_20, %p_x_17 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc56) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr> loc(#loc57) + %p_rstd_22 = tt.addptr %p_rstd, %o_t_4 : tensor<128x!tt.ptr>, tensor<128xi64> loc(#loc57) + %b_x_23 = tt.load %p_x_18, %m_x_12, %b_x : tensor<128x128x!tt.ptr> loc(#loc39) + %b_x_24 = arith.extf %b_x_23 : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc58) + %b_rstd_25 = arith.mulf %b_x_24, %b_x_24 : tensor<128x128xf32> loc(#loc59) + %b_rstd_26 = "tt.reduce"(%b_rstd_25) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_33: f32 loc(callsite(#loc3 at #loc60)), %b_rstd_34: f32 loc(callsite(#loc3 at #loc60))): + %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : f32 loc(#loc67) + tt.reduce.return %b_rstd_35 : f32 loc(#loc65) + }) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc65) + %b_rstd_27 = tt.splat %eps : f32 -> tensor<128xf32> loc(#loc61) + %b_rstd_28 = arith.addf %b_rstd_26, %b_rstd_27 : tensor<128xf32> loc(#loc61) + %b_rstd_29 = math.sqrt %b_rstd_28 : tensor<128xf32> loc(#loc62) + %b_rstd_30 = arith.divf %b_rstd, %b_rstd_29 : tensor<128xf32> loc(#loc40) + %b_y = tt.expand_dims %b_rstd_30 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc63) + %b_y_31 = tt.broadcast %b_y : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc64) + %b_y_32 = arith.mulf %b_x_24, %b_y_31 : tensor<128x128xf32> loc(#loc64) + %0 = arith.truncf %b_y_32 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc30) + tt.store %p_y_21, %0, %m_x_12 : tensor<128x128x!tt.ptr> loc(#loc31) + tt.store %p_rstd_22, %b_rstd_30, %m_t_6 : tensor<128x!tt.ptr> loc(#loc32) + tt.return loc(#loc33) + } loc(#loc) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc39 = loc("b_x"(#loc1)) +#loc40 = loc("b_rstd"(#loc2)) +#loc41 = loc("m_x"(#loc4)) +#loc42 = loc("i_t"(#loc5)) +#loc43 = loc("i_t"(#loc6)) +#loc44 = loc("o_t"(#loc7)) +#loc45 = loc("o_t"(#loc8)) +#loc46 = loc("o_t"(#loc9)) +#loc47 = loc("m_t"(#loc10)) +#loc48 = loc("m_x"(#loc11)) +#loc49 = loc("m_x"(#loc12)) +#loc50 = loc("m_x"(#loc13)) +#loc51 = loc("p_x"(#loc14)) +#loc52 = loc("p_x"(#loc15)) +#loc53 = loc("p_x"(#loc16)) +#loc54 = loc("p_x"(#loc17)) +#loc55 = loc("p_y"(#loc18)) +#loc56 = loc("p_y"(#loc19)) +#loc57 = loc("p_rstd"(#loc20)) +#loc58 = loc("b_x"(#loc21)) +#loc59 = loc("b_rstd"(#loc22)) +#loc61 = loc("b_rstd"(#loc26)) +#loc62 = loc("b_rstd"(#loc27)) +#loc63 = loc("b_y"(#loc28)) +#loc64 = loc("b_y"(#loc29)) +#loc65 = loc(callsite(#loc23 at #loc60)) +#loc67 = loc(callsite(#loc25 at #loc65)) diff --git a/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/__grp__recompute_w_u_fwd_kernel.json b/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/__grp__recompute_w_u_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..8bc220621766e6d51d1dae5e0a2ea7b4d7545dc1 --- /dev/null +++ b/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/__grp__recompute_w_u_fwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"recompute_w_u_fwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.source", "recompute_w_u_fwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.ttir", "recompute_w_u_fwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.ttgir", "recompute_w_u_fwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.llir", "recompute_w_u_fwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.ptx", "recompute_w_u_fwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.cubin", "recompute_w_u_fwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.cubin b/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..b5c99d05830e0b2f1cfb35475cd2068af46c6f2a --- /dev/null +++ b/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.cubin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f984a9065ff91be17fd4688fd782dac46d29fedfd378b2167a16ac8aa571501e +size 121392 diff --git a/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.json b/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..56d573675922717731cc9f34e65acc7bc2305f83 --- /dev/null +++ b/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.json @@ -0,0 +1 @@ +{"hash": "0819d21407299cfb857e24b0d3d6f74505e110be093f123673818aa2b3525f0a", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 2, "num_ctas": 1, "num_stages": 4, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 33792, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "recompute_w_u_fwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.llir b/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..6bd91e3ac3bcfd97f57827a791870f2c886001f5 --- /dev/null +++ b/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.llir @@ -0,0 +1,2349 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @recompute_w_u_fwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, ptr addrspace(1) %3, ptr addrspace(1) %4, ptr addrspace(1) %5, ptr addrspace(1) %6, i32 %7, ptr addrspace(1) readnone captures(none) %8, ptr addrspace(1) readnone captures(none) %9) local_unnamed_addr #0 !dbg !5 { + %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %12 = zext nneg i32 %11 to i64, !dbg !14 + %13 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.y(), !dbg !15 + %14 = zext nneg i32 %13 to i64, !dbg !16 + %15 = and i64 %14, 65504, !dbg !17 + %16 = and i64 %14, 31, !dbg !18 + %17 = sext i32 %7 to i64, !dbg !17 + %18 = shl nuw nsw i64 %12, 6, !dbg !19 + %19 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !20 + %20 = and i32 %19, 56, !dbg !20 + %21 = lshr exact i32 %20, 3, !dbg !20 + %22 = and i32 %19, 7, !dbg !20 + %23 = shl nuw nsw i32 %22, 3, !dbg !20 + %24 = insertelement <8 x i32> poison, i32 %19, i64 0, !dbg !20 + %25 = insertelement <8 x i32> %24, i32 %21, i64 1, !dbg !20 + %26 = shufflevector <8 x i32> %25, <8 x i32> poison, <8 x i32> , !dbg !20 + %27 = and <8 x i32> %26, , !dbg !20 + %28 = or disjoint <8 x i32> %26, , !dbg !20 + %29 = shufflevector <8 x i32> %27, <8 x i32> %28, <8 x i32> , !dbg !20 + %30 = zext nneg i32 %21 to i64, !dbg !21 + %31 = zext nneg <8 x i32> %29 to <8 x i64>, !dbg !21 + %32 = or disjoint i64 %18, %30, !dbg !21 + %33 = insertelement <8 x i64> poison, i64 %18, i64 0, !dbg !21 + %34 = shufflevector <8 x i64> %33, <8 x i64> poison, <8 x i32> zeroinitializer, !dbg !21 + %35 = or disjoint <8 x i64> %34, %31, !dbg !21 + %36 = icmp slt i64 %32, %17, !dbg !22 + %37 = insertelement <8 x i32> poison, i32 %7, i64 0, !dbg !22 + %38 = shufflevector <8 x i32> %37, <8 x i32> poison, <8 x i32> zeroinitializer, !dbg !22 + %39 = sext <8 x i32> %38 to <8 x i64>, !dbg !22 + %40 = icmp slt <8 x i64> %35, %39, !dbg !22 + %41 = mul nsw i64 %15, %17, !dbg !23 + %42 = or disjoint i64 %41, %16, !dbg !24 + %43 = getelementptr bfloat, ptr addrspace(1) %2, i64 %42, !dbg !24 + %44 = extractelement <8 x i64> %35, i64 0, !dbg !25 + %45 = shl nuw nsw i64 %44, 5, !dbg !25 + %46 = getelementptr bfloat, ptr addrspace(1) %43, i64 %45, !dbg !26 + %47 = extractelement <8 x i1> %40, i64 0, !dbg !27 + %48 = tail call i16 asm sideeffect "mov.u16 $0, $1;\0A\09@$3 ld.global.b16 { $0 }, [ $2 + 0 ];", "=c,c,l,b"(i16 0, ptr addrspace(1) %46, i1 %47) #3, !dbg !27 + %49 = shl nuw nsw i32 %19, 4, !dbg !28 + %50 = and i32 %49, 96, !dbg !28 + %51 = and i32 %19, 24, !dbg !28 + %52 = shl nuw nsw i32 %19, 2, !dbg !28 + %53 = and i32 %52, 4, !dbg !28 + %54 = and i32 %19, 32, !dbg !28 + %55 = lshr exact i32 %54, 4, !dbg !28 + %56 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %53, !dbg !28 + %57 = getelementptr inbounds nuw i8, ptr addrspace(3) %56, i32 %55, !dbg !28 + %58 = getelementptr inbounds nuw i8, ptr addrspace(3) %57, i32 %50, !dbg !28 + %59 = getelementptr inbounds nuw i8, ptr addrspace(3) %58, i32 %51, !dbg !28 + %60 = insertelement <1 x i16> poison, i16 %48, i64 0, !dbg !28 + tail call void asm sideeffect "st.shared.b16 [ $0 + 0 ], $1;", "r,h"(ptr addrspace(3) %59, <1 x i16> %60) #3, !dbg !28 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !28 + %61 = shl nuw nsw i32 %19, 5, !dbg !28 + %62 = and i32 %61, 96, !dbg !28 + %63 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %62, !dbg !28 + %64 = load bfloat, ptr addrspace(3) %63, align 16, !dbg !28 + %65 = getelementptr inbounds nuw i8, ptr addrspace(3) %63, i32 2, !dbg !28 + %66 = load bfloat, ptr addrspace(3) %65, align 2, !dbg !28 + %67 = getelementptr inbounds nuw i8, ptr addrspace(3) %63, i32 4, !dbg !28 + %68 = load bfloat, ptr addrspace(3) %67, align 4, !dbg !28 + %69 = getelementptr inbounds nuw i8, ptr addrspace(3) %63, i32 6, !dbg !28 + %70 = load bfloat, ptr addrspace(3) %69, align 2, !dbg !28 + %71 = getelementptr inbounds nuw i8, ptr addrspace(3) %63, i32 8, !dbg !28 + %72 = load bfloat, ptr addrspace(3) %71, align 8, !dbg !28 + %73 = getelementptr inbounds nuw i8, ptr addrspace(3) %63, i32 10, !dbg !28 + %74 = load bfloat, ptr addrspace(3) %73, align 2, !dbg !28 + %75 = getelementptr inbounds nuw i8, ptr addrspace(3) %63, i32 12, !dbg !28 + %76 = load bfloat, ptr addrspace(3) %75, align 4, !dbg !28 + %77 = getelementptr inbounds nuw i8, ptr addrspace(3) %63, i32 14, !dbg !28 + %78 = load bfloat, ptr addrspace(3) %77, align 2, !dbg !28 + %79 = getelementptr inbounds nuw i8, ptr addrspace(3) %63, i32 16, !dbg !28 + %80 = load bfloat, ptr addrspace(3) %79, align 16, !dbg !28 + %81 = getelementptr inbounds nuw i8, ptr addrspace(3) %63, i32 18, !dbg !28 + %82 = load bfloat, ptr addrspace(3) %81, align 2, !dbg !28 + %83 = getelementptr inbounds nuw i8, ptr addrspace(3) %63, i32 20, !dbg !28 + %84 = load bfloat, ptr addrspace(3) %83, align 4, !dbg !28 + %85 = getelementptr inbounds nuw i8, ptr addrspace(3) %63, i32 22, !dbg !28 + %86 = load bfloat, ptr addrspace(3) %85, align 2, !dbg !28 + %87 = getelementptr inbounds nuw i8, ptr addrspace(3) %63, i32 24, !dbg !28 + %88 = load bfloat, ptr addrspace(3) %87, align 8, !dbg !28 + %89 = getelementptr inbounds nuw i8, ptr addrspace(3) %63, i32 26, !dbg !28 + %90 = load bfloat, ptr addrspace(3) %89, align 2, !dbg !28 + %91 = getelementptr inbounds nuw i8, ptr addrspace(3) %63, i32 28, !dbg !28 + %92 = load bfloat, ptr addrspace(3) %91, align 4, !dbg !28 + %93 = getelementptr inbounds nuw i8, ptr addrspace(3) %63, i32 30, !dbg !28 + %94 = load bfloat, ptr addrspace(3) %93, align 2, !dbg !28 + %.idx = shl nsw i64 %42, 7, !dbg !29 + %95 = getelementptr i8, ptr addrspace(1) %5, i64 %.idx, !dbg !29 + %.idx24 = shl nuw nsw i64 %32, 12, !dbg !30 + %96 = getelementptr i8, ptr addrspace(1) %95, i64 %.idx24, !dbg !30 + %97 = extractelement <8 x i64> %35, i64 7, !dbg !30 + %.idx25 = shl nuw nsw i64 %97, 12, !dbg !30 + %98 = getelementptr i8, ptr addrspace(1) %95, i64 %.idx25, !dbg !30 + %99 = extractelement <8 x i64> %35, i64 6, !dbg !30 + %.idx26 = shl nuw nsw i64 %99, 12, !dbg !30 + %100 = getelementptr i8, ptr addrspace(1) %95, i64 %.idx26, !dbg !30 + %101 = extractelement <8 x i64> %35, i64 5, !dbg !30 + %.idx27 = shl nuw nsw i64 %101, 12, !dbg !30 + %102 = getelementptr i8, ptr addrspace(1) %95, i64 %.idx27, !dbg !30 + %103 = extractelement <8 x i64> %35, i64 4, !dbg !30 + %.idx28 = shl nuw nsw i64 %103, 12, !dbg !30 + %104 = getelementptr i8, ptr addrspace(1) %95, i64 %.idx28, !dbg !30 + %105 = extractelement <8 x i64> %35, i64 3, !dbg !30 + %.idx29 = shl nuw nsw i64 %105, 12, !dbg !30 + %106 = getelementptr i8, ptr addrspace(1) %95, i64 %.idx29, !dbg !30 + %107 = extractelement <8 x i64> %35, i64 2, !dbg !30 + %.idx30 = shl nuw nsw i64 %107, 12, !dbg !30 + %108 = getelementptr i8, ptr addrspace(1) %95, i64 %.idx30, !dbg !30 + %109 = extractelement <8 x i64> %35, i64 1, !dbg !30 + %.idx31 = shl nuw nsw i64 %109, 12, !dbg !30 + %110 = getelementptr i8, ptr addrspace(1) %95, i64 %.idx31, !dbg !30 + %111 = zext nneg i32 %23 to i64, !dbg !31 + %112 = getelementptr bfloat, ptr addrspace(1) %96, i64 %111, !dbg !31 + %113 = getelementptr bfloat, ptr addrspace(1) %98, i64 %111, !dbg !31 + %114 = getelementptr bfloat, ptr addrspace(1) %100, i64 %111, !dbg !31 + %115 = getelementptr bfloat, ptr addrspace(1) %102, i64 %111, !dbg !31 + %116 = getelementptr bfloat, ptr addrspace(1) %104, i64 %111, !dbg !31 + %117 = getelementptr bfloat, ptr addrspace(1) %106, i64 %111, !dbg !31 + %118 = getelementptr bfloat, ptr addrspace(1) %108, i64 %111, !dbg !31 + %119 = getelementptr bfloat, ptr addrspace(1) %110, i64 %111, !dbg !31 + %120 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %112, i1 %36) #3, !dbg !32 + %121 = extractvalue { i32, i32, i32, i32 } %120, 0, !dbg !32 + %122 = extractvalue { i32, i32, i32, i32 } %120, 1, !dbg !32 + %123 = extractvalue { i32, i32, i32, i32 } %120, 2, !dbg !32 + %124 = extractvalue { i32, i32, i32, i32 } %120, 3, !dbg !32 + %125 = extractelement <8 x i1> %40, i64 7, !dbg !33 + %126 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %113, i1 %125) #3, !dbg !32 + %127 = extractvalue { i32, i32, i32, i32 } %126, 0, !dbg !32 + %128 = extractvalue { i32, i32, i32, i32 } %126, 1, !dbg !32 + %129 = extractvalue { i32, i32, i32, i32 } %126, 2, !dbg !32 + %130 = extractvalue { i32, i32, i32, i32 } %126, 3, !dbg !32 + %131 = extractelement <8 x i1> %40, i64 6, !dbg !33 + %132 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %114, i1 %131) #3, !dbg !32 + %133 = extractvalue { i32, i32, i32, i32 } %132, 0, !dbg !32 + %134 = extractvalue { i32, i32, i32, i32 } %132, 1, !dbg !32 + %135 = extractvalue { i32, i32, i32, i32 } %132, 2, !dbg !32 + %136 = extractvalue { i32, i32, i32, i32 } %132, 3, !dbg !32 + %137 = extractelement <8 x i1> %40, i64 5, !dbg !33 + %138 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %115, i1 %137) #3, !dbg !32 + %139 = extractvalue { i32, i32, i32, i32 } %138, 0, !dbg !32 + %140 = extractvalue { i32, i32, i32, i32 } %138, 1, !dbg !32 + %141 = extractvalue { i32, i32, i32, i32 } %138, 2, !dbg !32 + %142 = extractvalue { i32, i32, i32, i32 } %138, 3, !dbg !32 + %143 = extractelement <8 x i1> %40, i64 4, !dbg !33 + %144 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %116, i1 %143) #3, !dbg !32 + %145 = extractvalue { i32, i32, i32, i32 } %144, 0, !dbg !32 + %146 = extractvalue { i32, i32, i32, i32 } %144, 1, !dbg !32 + %147 = extractvalue { i32, i32, i32, i32 } %144, 2, !dbg !32 + %148 = extractvalue { i32, i32, i32, i32 } %144, 3, !dbg !32 + %149 = extractelement <8 x i1> %40, i64 3, !dbg !33 + %150 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %117, i1 %149) #3, !dbg !32 + %151 = extractvalue { i32, i32, i32, i32 } %150, 0, !dbg !32 + %152 = extractvalue { i32, i32, i32, i32 } %150, 1, !dbg !32 + %153 = extractvalue { i32, i32, i32, i32 } %150, 2, !dbg !32 + %154 = extractvalue { i32, i32, i32, i32 } %150, 3, !dbg !32 + %155 = extractelement <8 x i1> %40, i64 2, !dbg !33 + %156 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %118, i1 %155) #3, !dbg !32 + %157 = extractvalue { i32, i32, i32, i32 } %156, 0, !dbg !32 + %158 = extractvalue { i32, i32, i32, i32 } %156, 1, !dbg !32 + %159 = extractvalue { i32, i32, i32, i32 } %156, 2, !dbg !32 + %160 = extractvalue { i32, i32, i32, i32 } %156, 3, !dbg !32 + %161 = extractelement <8 x i1> %40, i64 1, !dbg !33 + %162 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %119, i1 %161) #3, !dbg !32 + %163 = extractvalue { i32, i32, i32, i32 } %162, 0, !dbg !32 + %164 = extractvalue { i32, i32, i32, i32 } %162, 1, !dbg !32 + %165 = extractvalue { i32, i32, i32, i32 } %162, 2, !dbg !32 + %166 = extractvalue { i32, i32, i32, i32 } %162, 3, !dbg !32 + %167 = shl nuw nsw i32 %19, 4, !dbg !32 + %168 = and i32 %167, 1008, !dbg !32 + %169 = shl nuw nsw i32 %20, 1, !dbg !32 + %170 = xor i32 %168, %169, !dbg !32 + %171 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 25600), i32 %170, !dbg !32 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %171, i32 %121, i32 %122, i32 %123, i32 %124) #3, !dbg !32 + %172 = or disjoint i32 %170, 1024, !dbg !32 + %173 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 25600), i32 %172, !dbg !32 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %173, i32 %127, i32 %128, i32 %129, i32 %130) #3, !dbg !32 + %174 = or disjoint i32 %170, 2048, !dbg !32 + %175 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 25600), i32 %174, !dbg !32 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %175, i32 %133, i32 %134, i32 %135, i32 %136) #3, !dbg !32 + %176 = or disjoint i32 %170, 3072, !dbg !32 + %177 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 25600), i32 %176, !dbg !32 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %177, i32 %139, i32 %140, i32 %141, i32 %142) #3, !dbg !32 + %178 = or disjoint i32 %170, 4096, !dbg !32 + %179 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 25600), i32 %178, !dbg !32 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %179, i32 %145, i32 %146, i32 %147, i32 %148) #3, !dbg !32 + %180 = or disjoint i32 %170, 5120, !dbg !32 + %181 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 25600), i32 %180, !dbg !32 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %181, i32 %151, i32 %152, i32 %153, i32 %154) #3, !dbg !32 + %182 = or disjoint i32 %170, 6144, !dbg !32 + %183 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 25600), i32 %182, !dbg !32 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %183, i32 %157, i32 %158, i32 %159, i32 %160) #3, !dbg !32 + %184 = or disjoint i32 %170, 7168, !dbg !32 + %185 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 25600), i32 %184, !dbg !32 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %185, i32 %163, i32 %164, i32 %165, i32 %166) #3, !dbg !32 + %186 = getelementptr bfloat, ptr addrspace(1) %1, i64 %.idx, !dbg !34 + %187 = getelementptr bfloat, ptr addrspace(1) %186, i64 %.idx24, !dbg !35 + %188 = getelementptr bfloat, ptr addrspace(1) %186, i64 %.idx25, !dbg !35 + %189 = getelementptr bfloat, ptr addrspace(1) %186, i64 %.idx26, !dbg !35 + %190 = getelementptr bfloat, ptr addrspace(1) %186, i64 %.idx27, !dbg !35 + %191 = getelementptr bfloat, ptr addrspace(1) %186, i64 %.idx28, !dbg !35 + %192 = getelementptr bfloat, ptr addrspace(1) %186, i64 %.idx29, !dbg !35 + %193 = getelementptr bfloat, ptr addrspace(1) %186, i64 %.idx30, !dbg !35 + %194 = getelementptr bfloat, ptr addrspace(1) %186, i64 %.idx31, !dbg !35 + %195 = getelementptr bfloat, ptr addrspace(1) %4, i64 %.idx, !dbg !36 + %196 = getelementptr bfloat, ptr addrspace(1) %195, i64 %.idx24, !dbg !37 + %197 = getelementptr bfloat, ptr addrspace(1) %195, i64 %.idx25, !dbg !37 + %198 = getelementptr bfloat, ptr addrspace(1) %195, i64 %.idx26, !dbg !37 + %199 = getelementptr bfloat, ptr addrspace(1) %195, i64 %.idx27, !dbg !37 + %200 = getelementptr bfloat, ptr addrspace(1) %195, i64 %.idx28, !dbg !37 + %201 = getelementptr bfloat, ptr addrspace(1) %195, i64 %.idx29, !dbg !37 + %202 = getelementptr bfloat, ptr addrspace(1) %195, i64 %.idx30, !dbg !37 + %203 = getelementptr bfloat, ptr addrspace(1) %195, i64 %.idx31, !dbg !37 + %204 = getelementptr bfloat, ptr addrspace(1) %187, i64 %111, !dbg !38 + %205 = getelementptr bfloat, ptr addrspace(1) %188, i64 %111, !dbg !38 + %206 = getelementptr bfloat, ptr addrspace(1) %189, i64 %111, !dbg !38 + %207 = getelementptr bfloat, ptr addrspace(1) %190, i64 %111, !dbg !38 + %208 = getelementptr bfloat, ptr addrspace(1) %191, i64 %111, !dbg !38 + %209 = getelementptr bfloat, ptr addrspace(1) %192, i64 %111, !dbg !38 + %210 = getelementptr bfloat, ptr addrspace(1) %193, i64 %111, !dbg !38 + %211 = getelementptr bfloat, ptr addrspace(1) %194, i64 %111, !dbg !38 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !39 + %212 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %170, !dbg !39 + %213 = select i1 %36, i32 16, i32 0, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %212, ptr addrspace(1) %204, i32 %213) #3, !dbg !39 + %214 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %172, !dbg !39 + %215 = select i1 %125, i32 16, i32 0, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %214, ptr addrspace(1) %205, i32 %215) #3, !dbg !39 + %216 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %174, !dbg !39 + %217 = select i1 %131, i32 16, i32 0, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %216, ptr addrspace(1) %206, i32 %217) #3, !dbg !39 + %218 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %176, !dbg !39 + %219 = select i1 %137, i32 16, i32 0, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %218, ptr addrspace(1) %207, i32 %219) #3, !dbg !39 + %220 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %178, !dbg !39 + %221 = select i1 %143, i32 16, i32 0, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %220, ptr addrspace(1) %208, i32 %221) #3, !dbg !39 + %222 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %180, !dbg !39 + %223 = select i1 %149, i32 16, i32 0, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %222, ptr addrspace(1) %209, i32 %223) #3, !dbg !39 + %224 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %182, !dbg !39 + %225 = select i1 %155, i32 16, i32 0, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %224, ptr addrspace(1) %210, i32 %225) #3, !dbg !39 + %226 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %184, !dbg !39 + %227 = select i1 %161, i32 16, i32 0, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %226, ptr addrspace(1) %211, i32 %227) #3, !dbg !39 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !39 + %228 = or disjoint i32 %23, 64, !dbg !40 + %229 = zext nneg i32 %228 to i64, !dbg !38 + %230 = getelementptr bfloat, ptr addrspace(1) %187, i64 %229, !dbg !38 + %231 = getelementptr bfloat, ptr addrspace(1) %188, i64 %229, !dbg !38 + %232 = getelementptr bfloat, ptr addrspace(1) %189, i64 %229, !dbg !38 + %233 = getelementptr bfloat, ptr addrspace(1) %190, i64 %229, !dbg !38 + %234 = getelementptr bfloat, ptr addrspace(1) %191, i64 %229, !dbg !38 + %235 = getelementptr bfloat, ptr addrspace(1) %192, i64 %229, !dbg !38 + %236 = getelementptr bfloat, ptr addrspace(1) %193, i64 %229, !dbg !38 + %237 = getelementptr bfloat, ptr addrspace(1) %194, i64 %229, !dbg !38 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !39 + %238 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 8192), i32 %170, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %238, ptr addrspace(1) %230, i32 %213) #3, !dbg !39 + %239 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 8192), i32 %172, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %239, ptr addrspace(1) %231, i32 %215) #3, !dbg !39 + %240 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 8192), i32 %174, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %240, ptr addrspace(1) %232, i32 %217) #3, !dbg !39 + %241 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 8192), i32 %176, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %241, ptr addrspace(1) %233, i32 %219) #3, !dbg !39 + %242 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 8192), i32 %178, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %242, ptr addrspace(1) %234, i32 %221) #3, !dbg !39 + %243 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 8192), i32 %180, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %243, ptr addrspace(1) %235, i32 %223) #3, !dbg !39 + %244 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 8192), i32 %182, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %244, ptr addrspace(1) %236, i32 %225) #3, !dbg !39 + %245 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 8192), i32 %184, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %245, ptr addrspace(1) %237, i32 %227) #3, !dbg !39 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !39 + %246 = or disjoint i32 %23, 128, !dbg !40 + %247 = zext nneg i32 %246 to i64, !dbg !38 + %248 = getelementptr bfloat, ptr addrspace(1) %187, i64 %247, !dbg !38 + %249 = getelementptr bfloat, ptr addrspace(1) %188, i64 %247, !dbg !38 + %250 = getelementptr bfloat, ptr addrspace(1) %189, i64 %247, !dbg !38 + %251 = getelementptr bfloat, ptr addrspace(1) %190, i64 %247, !dbg !38 + %252 = getelementptr bfloat, ptr addrspace(1) %191, i64 %247, !dbg !38 + %253 = getelementptr bfloat, ptr addrspace(1) %192, i64 %247, !dbg !38 + %254 = getelementptr bfloat, ptr addrspace(1) %193, i64 %247, !dbg !38 + %255 = getelementptr bfloat, ptr addrspace(1) %194, i64 %247, !dbg !38 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !39 + %256 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %170, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %256, ptr addrspace(1) %248, i32 0) #3, !dbg !39 + %257 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %172, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %257, ptr addrspace(1) %249, i32 0) #3, !dbg !39 + %258 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %174, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %258, ptr addrspace(1) %250, i32 0) #3, !dbg !39 + %259 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %176, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %259, ptr addrspace(1) %251, i32 0) #3, !dbg !39 + %260 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %178, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %260, ptr addrspace(1) %252, i32 0) #3, !dbg !39 + %261 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %180, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %261, ptr addrspace(1) %253, i32 0) #3, !dbg !39 + %262 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %182, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %262, ptr addrspace(1) %254, i32 0) #3, !dbg !39 + %263 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %184, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %263, ptr addrspace(1) %255, i32 0) #3, !dbg !39 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !39 + %264 = shl nuw nsw i32 %19, 7 + %265 = and i32 %264, 3968 + %266 = shl nuw nsw i32 %22, 4 + %267 = lshr exact i32 %54, 1 + %268 = or disjoint i32 %265, %266 + %269 = xor i32 %268, %267 + %270 = xor i32 %269, 32 + %271 = xor i32 %269, 64 + %272 = xor i32 %269, 96 + %273 = and i32 %264, 1920 + %274 = and i32 %19, 16 + %275 = or disjoint i32 %273, %266 + %276 = xor i32 %275, %274 + %277 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 25600), i32 %276 + %278 = getelementptr inbounds nuw i8, ptr addrspace(3) %277, i32 2048 + %279 = getelementptr inbounds nuw i8, ptr addrspace(3) %277, i32 4096 + %280 = getelementptr inbounds nuw i8, ptr addrspace(3) %277, i32 6144 + %281 = xor i32 %276, 32 + %282 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 25600), i32 %281 + %283 = getelementptr inbounds nuw i8, ptr addrspace(3) %282, i32 2048 + %284 = getelementptr inbounds nuw i8, ptr addrspace(3) %282, i32 4096 + %285 = getelementptr inbounds nuw i8, ptr addrspace(3) %282, i32 6144 + %286 = xor i32 %276, 64 + %287 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 25600), i32 %286 + %288 = getelementptr inbounds nuw i8, ptr addrspace(3) %287, i32 2048 + %289 = getelementptr inbounds nuw i8, ptr addrspace(3) %287, i32 4096 + %290 = getelementptr inbounds nuw i8, ptr addrspace(3) %287, i32 6144 + %291 = xor i32 %276, 96 + %292 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 25600), i32 %291 + %293 = getelementptr inbounds nuw i8, ptr addrspace(3) %292, i32 2048 + %294 = getelementptr inbounds nuw i8, ptr addrspace(3) %292, i32 4096 + %295 = getelementptr inbounds nuw i8, ptr addrspace(3) %292, i32 6144 + %296 = shl nuw nsw i32 %22, 7 + %297 = shl nuw nsw i32 %51, 2 + %298 = xor i32 %266, %297 + %299 = xor i32 %298, %267 + %300 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 24576), i32 %296 + %301 = getelementptr inbounds nuw i8, ptr addrspace(3) %300, i32 %299 + %302 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 24576), i32 %170 + %303 = insertelement <2 x bfloat> poison, bfloat %90, i64 0, !dbg !41 + %304 = insertelement <2 x bfloat> %303, bfloat %94, i64 1, !dbg !41 + %305 = insertelement <2 x bfloat> poison, bfloat %82, i64 0, !dbg !41 + %306 = insertelement <2 x bfloat> %305, bfloat %86, i64 1, !dbg !41 + %307 = insertelement <2 x bfloat> poison, bfloat %74, i64 0, !dbg !41 + %308 = insertelement <2 x bfloat> %307, bfloat %78, i64 1, !dbg !41 + %309 = insertelement <2 x bfloat> poison, bfloat %66, i64 0, !dbg !41 + %310 = insertelement <2 x bfloat> %309, bfloat %70, i64 1, !dbg !41 + %311 = insertelement <2 x bfloat> poison, bfloat %88, i64 0, !dbg !41 + %312 = insertelement <2 x bfloat> %311, bfloat %92, i64 1, !dbg !41 + %313 = insertelement <2 x bfloat> poison, bfloat %80, i64 0, !dbg !41 + %314 = insertelement <2 x bfloat> %313, bfloat %84, i64 1, !dbg !41 + %315 = insertelement <2 x bfloat> poison, bfloat %72, i64 0, !dbg !41 + %316 = insertelement <2 x bfloat> %315, bfloat %76, i64 1, !dbg !41 + %317 = insertelement <2 x bfloat> poison, bfloat %64, i64 0, !dbg !41 + %318 = insertelement <2 x bfloat> %317, bfloat %68, i64 1, !dbg !41 + br label %319, !dbg !33 + +319: ; preds = %10, %319 + %320 = phi i1 [ true, %10 ], [ false, %319 ] + %indvars.iv = phi i64 [ 192, %10 ], [ 256, %319 ] + %321 = phi i1 [ %36, %10 ], [ false, %319 ] + %322 = phi i1 [ %125, %10 ], [ false, %319 ] + %323 = phi i1 [ %131, %10 ], [ false, %319 ] + %324 = phi i1 [ %137, %10 ], [ false, %319 ] + %325 = phi i1 [ %143, %10 ], [ false, %319 ] + %326 = phi i1 [ %149, %10 ], [ false, %319 ] + %327 = phi i1 [ %155, %10 ], [ false, %319 ] + %328 = phi i1 [ %161, %10 ], [ false, %319 ] + %329 = phi i1 [ %36, %10 ], [ %321, %319 ] + %330 = phi i1 [ %125, %10 ], [ %322, %319 ] + %331 = phi i1 [ %131, %10 ], [ %323, %319 ] + %332 = phi i1 [ %137, %10 ], [ %324, %319 ] + %333 = phi i1 [ %143, %10 ], [ %325, %319 ] + %334 = phi i1 [ %149, %10 ], [ %326, %319 ] + %335 = phi i1 [ %155, %10 ], [ %327, %319 ] + %336 = phi i1 [ %161, %10 ], [ %328, %319 ] + %337 = phi i32 [ %246, %10 ], [ %1046, %319 ] + %338 = phi i32 [ %228, %10 ], [ %337, %319 ] + %339 = phi i32 [ %23, %10 ], [ %338, %319 ] + %340 = phi i32 [ -1, %10 ], [ %344, %319 ] + %341 = phi i32 [ 2, %10 ], [ %1044, %319 ] + %342 = add i32 %340, 1, !dbg !33 + %343 = icmp sgt i32 %342, 2, !dbg !33 + %344 = select i1 %343, i32 0, i32 %342, !dbg !33 + %345 = sext i32 %339 to i64, !dbg !42 + %346 = getelementptr bfloat, ptr addrspace(1) %196, i64 %345, !dbg !42 + %347 = sext i32 %339 to i64, !dbg !42 + %348 = getelementptr bfloat, ptr addrspace(1) %197, i64 %347, !dbg !42 + %349 = sext i32 %339 to i64, !dbg !42 + %350 = getelementptr bfloat, ptr addrspace(1) %198, i64 %349, !dbg !42 + %351 = sext i32 %339 to i64, !dbg !42 + %352 = getelementptr bfloat, ptr addrspace(1) %199, i64 %351, !dbg !42 + %353 = sext i32 %339 to i64, !dbg !42 + %354 = getelementptr bfloat, ptr addrspace(1) %200, i64 %353, !dbg !42 + %355 = sext i32 %339 to i64, !dbg !42 + %356 = getelementptr bfloat, ptr addrspace(1) %201, i64 %355, !dbg !42 + %357 = sext i32 %339 to i64, !dbg !42 + %358 = getelementptr bfloat, ptr addrspace(1) %202, i64 %357, !dbg !42 + %359 = sext i32 %339 to i64, !dbg !42 + %360 = getelementptr bfloat, ptr addrspace(1) %203, i64 %359, !dbg !42 + tail call void @llvm.nvvm.cp.async.wait.group(i32 2), !dbg !39 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !39 + %.idx202 = shl i32 %344, 13, !dbg !39 + %361 = getelementptr i8, ptr addrspace(3) @global_smem, i32 %.idx202, !dbg !39 + %362 = getelementptr inbounds nuw i8, ptr addrspace(3) %361, i32 %269, !dbg !41 + %363 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %362), !dbg !41 + %364 = extractvalue { i32, i32, i32, i32 } %363, 0, !dbg !41 + %365 = bitcast i32 %364 to <2 x bfloat>, !dbg !41 + %366 = extractvalue { i32, i32, i32, i32 } %363, 1, !dbg !41 + %367 = bitcast i32 %366 to <2 x bfloat>, !dbg !41 + %368 = extractvalue { i32, i32, i32, i32 } %363, 2, !dbg !41 + %369 = bitcast i32 %368 to <2 x bfloat>, !dbg !41 + %370 = extractvalue { i32, i32, i32, i32 } %363, 3, !dbg !41 + %371 = bitcast i32 %370 to <2 x bfloat>, !dbg !41 + %372 = getelementptr inbounds nuw i8, ptr addrspace(3) %362, i32 4096, !dbg !41 + %373 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %372), !dbg !41 + %374 = extractvalue { i32, i32, i32, i32 } %373, 0, !dbg !41 + %375 = bitcast i32 %374 to <2 x bfloat>, !dbg !41 + %376 = extractvalue { i32, i32, i32, i32 } %373, 1, !dbg !41 + %377 = bitcast i32 %376 to <2 x bfloat>, !dbg !41 + %378 = extractvalue { i32, i32, i32, i32 } %373, 2, !dbg !41 + %379 = bitcast i32 %378 to <2 x bfloat>, !dbg !41 + %380 = extractvalue { i32, i32, i32, i32 } %373, 3, !dbg !41 + %381 = bitcast i32 %380 to <2 x bfloat>, !dbg !41 + %382 = getelementptr inbounds nuw i8, ptr addrspace(3) %361, i32 %270, !dbg !41 + %383 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %382), !dbg !41 + %384 = extractvalue { i32, i32, i32, i32 } %383, 0, !dbg !41 + %385 = bitcast i32 %384 to <2 x bfloat>, !dbg !41 + %386 = extractvalue { i32, i32, i32, i32 } %383, 1, !dbg !41 + %387 = bitcast i32 %386 to <2 x bfloat>, !dbg !41 + %388 = extractvalue { i32, i32, i32, i32 } %383, 2, !dbg !41 + %389 = bitcast i32 %388 to <2 x bfloat>, !dbg !41 + %390 = extractvalue { i32, i32, i32, i32 } %383, 3, !dbg !41 + %391 = bitcast i32 %390 to <2 x bfloat>, !dbg !41 + %392 = getelementptr inbounds nuw i8, ptr addrspace(3) %382, i32 4096, !dbg !41 + %393 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %392), !dbg !41 + %394 = extractvalue { i32, i32, i32, i32 } %393, 0, !dbg !41 + %395 = bitcast i32 %394 to <2 x bfloat>, !dbg !41 + %396 = extractvalue { i32, i32, i32, i32 } %393, 1, !dbg !41 + %397 = bitcast i32 %396 to <2 x bfloat>, !dbg !41 + %398 = extractvalue { i32, i32, i32, i32 } %393, 2, !dbg !41 + %399 = bitcast i32 %398 to <2 x bfloat>, !dbg !41 + %400 = extractvalue { i32, i32, i32, i32 } %393, 3, !dbg !41 + %401 = bitcast i32 %400 to <2 x bfloat>, !dbg !41 + %402 = getelementptr inbounds nuw i8, ptr addrspace(3) %361, i32 %271, !dbg !41 + %403 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %402), !dbg !41 + %404 = extractvalue { i32, i32, i32, i32 } %403, 0, !dbg !41 + %405 = bitcast i32 %404 to <2 x bfloat>, !dbg !41 + %406 = extractvalue { i32, i32, i32, i32 } %403, 1, !dbg !41 + %407 = bitcast i32 %406 to <2 x bfloat>, !dbg !41 + %408 = extractvalue { i32, i32, i32, i32 } %403, 2, !dbg !41 + %409 = bitcast i32 %408 to <2 x bfloat>, !dbg !41 + %410 = extractvalue { i32, i32, i32, i32 } %403, 3, !dbg !41 + %411 = bitcast i32 %410 to <2 x bfloat>, !dbg !41 + %412 = getelementptr inbounds nuw i8, ptr addrspace(3) %402, i32 4096, !dbg !41 + %413 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %412), !dbg !41 + %414 = extractvalue { i32, i32, i32, i32 } %413, 0, !dbg !41 + %415 = bitcast i32 %414 to <2 x bfloat>, !dbg !41 + %416 = extractvalue { i32, i32, i32, i32 } %413, 1, !dbg !41 + %417 = bitcast i32 %416 to <2 x bfloat>, !dbg !41 + %418 = extractvalue { i32, i32, i32, i32 } %413, 2, !dbg !41 + %419 = bitcast i32 %418 to <2 x bfloat>, !dbg !41 + %420 = extractvalue { i32, i32, i32, i32 } %413, 3, !dbg !41 + %421 = bitcast i32 %420 to <2 x bfloat>, !dbg !41 + %422 = getelementptr inbounds nuw i8, ptr addrspace(3) %361, i32 %272, !dbg !41 + %423 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %422), !dbg !41 + %424 = extractvalue { i32, i32, i32, i32 } %423, 0, !dbg !41 + %425 = bitcast i32 %424 to <2 x bfloat>, !dbg !41 + %426 = extractvalue { i32, i32, i32, i32 } %423, 1, !dbg !41 + %427 = bitcast i32 %426 to <2 x bfloat>, !dbg !41 + %428 = extractvalue { i32, i32, i32, i32 } %423, 2, !dbg !41 + %429 = bitcast i32 %428 to <2 x bfloat>, !dbg !41 + %430 = extractvalue { i32, i32, i32, i32 } %423, 3, !dbg !41 + %431 = bitcast i32 %430 to <2 x bfloat>, !dbg !41 + %432 = getelementptr inbounds nuw i8, ptr addrspace(3) %422, i32 4096, !dbg !41 + %433 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %432), !dbg !41 + %434 = extractvalue { i32, i32, i32, i32 } %433, 0, !dbg !41 + %435 = bitcast i32 %434 to <2 x bfloat>, !dbg !41 + %436 = extractvalue { i32, i32, i32, i32 } %433, 1, !dbg !41 + %437 = bitcast i32 %436 to <2 x bfloat>, !dbg !41 + %438 = extractvalue { i32, i32, i32, i32 } %433, 2, !dbg !41 + %439 = bitcast i32 %438 to <2 x bfloat>, !dbg !41 + %440 = extractvalue { i32, i32, i32, i32 } %433, 3, !dbg !41 + %441 = bitcast i32 %440 to <2 x bfloat>, !dbg !41 + %442 = fmul <2 x bfloat> %318, %365, !dbg !41 + %443 = fmul <2 x bfloat> %316, %367, !dbg !41 + %444 = fmul <2 x bfloat> %314, %369, !dbg !41 + %445 = fmul <2 x bfloat> %312, %371, !dbg !41 + %446 = fmul <2 x bfloat> %310, %375, !dbg !41 + %447 = fmul <2 x bfloat> %308, %377, !dbg !41 + %448 = fmul <2 x bfloat> %306, %379, !dbg !41 + %449 = fmul <2 x bfloat> %304, %381, !dbg !41 + %450 = fmul <2 x bfloat> %318, %385, !dbg !41 + %451 = fmul <2 x bfloat> %316, %387, !dbg !41 + %452 = fmul <2 x bfloat> %314, %389, !dbg !41 + %453 = fmul <2 x bfloat> %312, %391, !dbg !41 + %454 = fmul <2 x bfloat> %310, %395, !dbg !41 + %455 = fmul <2 x bfloat> %308, %397, !dbg !41 + %456 = fmul <2 x bfloat> %306, %399, !dbg !41 + %457 = fmul <2 x bfloat> %304, %401, !dbg !41 + %458 = fmul <2 x bfloat> %318, %405, !dbg !41 + %459 = fmul <2 x bfloat> %316, %407, !dbg !41 + %460 = fmul <2 x bfloat> %314, %409, !dbg !41 + %461 = fmul <2 x bfloat> %312, %411, !dbg !41 + %462 = fmul <2 x bfloat> %310, %415, !dbg !41 + %463 = fmul <2 x bfloat> %308, %417, !dbg !41 + %464 = fmul <2 x bfloat> %306, %419, !dbg !41 + %465 = fmul <2 x bfloat> %304, %421, !dbg !41 + %466 = fmul <2 x bfloat> %318, %425, !dbg !41 + %467 = fmul <2 x bfloat> %316, %427, !dbg !41 + %468 = fmul <2 x bfloat> %314, %429, !dbg !41 + %469 = fmul <2 x bfloat> %312, %431, !dbg !41 + %470 = fmul <2 x bfloat> %310, %435, !dbg !41 + %471 = fmul <2 x bfloat> %308, %437, !dbg !41 + %472 = fmul <2 x bfloat> %306, %439, !dbg !41 + %473 = fmul <2 x bfloat> %304, %441, !dbg !41 + %474 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %277), !dbg !32 + %475 = extractvalue { i32, i32, i32, i32 } %474, 0, !dbg !32 + %476 = extractvalue { i32, i32, i32, i32 } %474, 1, !dbg !32 + %477 = extractvalue { i32, i32, i32, i32 } %474, 2, !dbg !32 + %478 = extractvalue { i32, i32, i32, i32 } %474, 3, !dbg !32 + %479 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %278), !dbg !32 + %480 = extractvalue { i32, i32, i32, i32 } %479, 0, !dbg !32 + %481 = extractvalue { i32, i32, i32, i32 } %479, 1, !dbg !32 + %482 = extractvalue { i32, i32, i32, i32 } %479, 2, !dbg !32 + %483 = extractvalue { i32, i32, i32, i32 } %479, 3, !dbg !32 + %484 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %279), !dbg !32 + %485 = extractvalue { i32, i32, i32, i32 } %484, 0, !dbg !32 + %486 = extractvalue { i32, i32, i32, i32 } %484, 1, !dbg !32 + %487 = extractvalue { i32, i32, i32, i32 } %484, 2, !dbg !32 + %488 = extractvalue { i32, i32, i32, i32 } %484, 3, !dbg !32 + %489 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %280), !dbg !32 + %490 = extractvalue { i32, i32, i32, i32 } %489, 0, !dbg !32 + %491 = extractvalue { i32, i32, i32, i32 } %489, 1, !dbg !32 + %492 = extractvalue { i32, i32, i32, i32 } %489, 2, !dbg !32 + %493 = extractvalue { i32, i32, i32, i32 } %489, 3, !dbg !32 + %494 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %282), !dbg !32 + %495 = extractvalue { i32, i32, i32, i32 } %494, 0, !dbg !32 + %496 = extractvalue { i32, i32, i32, i32 } %494, 1, !dbg !32 + %497 = extractvalue { i32, i32, i32, i32 } %494, 2, !dbg !32 + %498 = extractvalue { i32, i32, i32, i32 } %494, 3, !dbg !32 + %499 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %283), !dbg !32 + %500 = extractvalue { i32, i32, i32, i32 } %499, 0, !dbg !32 + %501 = extractvalue { i32, i32, i32, i32 } %499, 1, !dbg !32 + %502 = extractvalue { i32, i32, i32, i32 } %499, 2, !dbg !32 + %503 = extractvalue { i32, i32, i32, i32 } %499, 3, !dbg !32 + %504 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %284), !dbg !32 + %505 = extractvalue { i32, i32, i32, i32 } %504, 0, !dbg !32 + %506 = extractvalue { i32, i32, i32, i32 } %504, 1, !dbg !32 + %507 = extractvalue { i32, i32, i32, i32 } %504, 2, !dbg !32 + %508 = extractvalue { i32, i32, i32, i32 } %504, 3, !dbg !32 + %509 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %285), !dbg !32 + %510 = extractvalue { i32, i32, i32, i32 } %509, 0, !dbg !32 + %511 = extractvalue { i32, i32, i32, i32 } %509, 1, !dbg !32 + %512 = extractvalue { i32, i32, i32, i32 } %509, 2, !dbg !32 + %513 = extractvalue { i32, i32, i32, i32 } %509, 3, !dbg !32 + %514 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %287), !dbg !32 + %515 = extractvalue { i32, i32, i32, i32 } %514, 0, !dbg !32 + %516 = extractvalue { i32, i32, i32, i32 } %514, 1, !dbg !32 + %517 = extractvalue { i32, i32, i32, i32 } %514, 2, !dbg !32 + %518 = extractvalue { i32, i32, i32, i32 } %514, 3, !dbg !32 + %519 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %288), !dbg !32 + %520 = extractvalue { i32, i32, i32, i32 } %519, 0, !dbg !32 + %521 = extractvalue { i32, i32, i32, i32 } %519, 1, !dbg !32 + %522 = extractvalue { i32, i32, i32, i32 } %519, 2, !dbg !32 + %523 = extractvalue { i32, i32, i32, i32 } %519, 3, !dbg !32 + %524 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %289), !dbg !32 + %525 = extractvalue { i32, i32, i32, i32 } %524, 0, !dbg !32 + %526 = extractvalue { i32, i32, i32, i32 } %524, 1, !dbg !32 + %527 = extractvalue { i32, i32, i32, i32 } %524, 2, !dbg !32 + %528 = extractvalue { i32, i32, i32, i32 } %524, 3, !dbg !32 + %529 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %290), !dbg !32 + %530 = extractvalue { i32, i32, i32, i32 } %529, 0, !dbg !32 + %531 = extractvalue { i32, i32, i32, i32 } %529, 1, !dbg !32 + %532 = extractvalue { i32, i32, i32, i32 } %529, 2, !dbg !32 + %533 = extractvalue { i32, i32, i32, i32 } %529, 3, !dbg !32 + %534 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %292), !dbg !32 + %535 = extractvalue { i32, i32, i32, i32 } %534, 0, !dbg !32 + %536 = extractvalue { i32, i32, i32, i32 } %534, 1, !dbg !32 + %537 = extractvalue { i32, i32, i32, i32 } %534, 2, !dbg !32 + %538 = extractvalue { i32, i32, i32, i32 } %534, 3, !dbg !32 + %539 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %293), !dbg !32 + %540 = extractvalue { i32, i32, i32, i32 } %539, 0, !dbg !32 + %541 = extractvalue { i32, i32, i32, i32 } %539, 1, !dbg !32 + %542 = extractvalue { i32, i32, i32, i32 } %539, 2, !dbg !32 + %543 = extractvalue { i32, i32, i32, i32 } %539, 3, !dbg !32 + %544 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %294), !dbg !32 + %545 = extractvalue { i32, i32, i32, i32 } %544, 0, !dbg !32 + %546 = extractvalue { i32, i32, i32, i32 } %544, 1, !dbg !32 + %547 = extractvalue { i32, i32, i32, i32 } %544, 2, !dbg !32 + %548 = extractvalue { i32, i32, i32, i32 } %544, 3, !dbg !32 + %549 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %295), !dbg !32 + %550 = extractvalue { i32, i32, i32, i32 } %549, 0, !dbg !32 + %551 = extractvalue { i32, i32, i32, i32 } %549, 1, !dbg !32 + %552 = extractvalue { i32, i32, i32, i32 } %549, 2, !dbg !32 + %553 = extractvalue { i32, i32, i32, i32 } %549, 3, !dbg !32 + %554 = bitcast <2 x bfloat> %442 to i32, !dbg !43 + %555 = bitcast <2 x bfloat> %443 to i32, !dbg !43 + %556 = bitcast <2 x bfloat> %444 to i32, !dbg !43 + %557 = bitcast <2 x bfloat> %445 to i32, !dbg !43 + %558 = bitcast <2 x bfloat> %446 to i32, !dbg !43 + %559 = bitcast <2 x bfloat> %447 to i32, !dbg !43 + %560 = bitcast <2 x bfloat> %448 to i32, !dbg !43 + %561 = bitcast <2 x bfloat> %449 to i32, !dbg !43 + %562 = bitcast <2 x bfloat> %450 to i32, !dbg !43 + %563 = bitcast <2 x bfloat> %451 to i32, !dbg !43 + %564 = bitcast <2 x bfloat> %452 to i32, !dbg !43 + %565 = bitcast <2 x bfloat> %453 to i32, !dbg !43 + %566 = bitcast <2 x bfloat> %454 to i32, !dbg !43 + %567 = bitcast <2 x bfloat> %455 to i32, !dbg !43 + %568 = bitcast <2 x bfloat> %456 to i32, !dbg !43 + %569 = bitcast <2 x bfloat> %457 to i32, !dbg !43 + %570 = bitcast <2 x bfloat> %458 to i32, !dbg !43 + %571 = bitcast <2 x bfloat> %459 to i32, !dbg !43 + %572 = bitcast <2 x bfloat> %460 to i32, !dbg !43 + %573 = bitcast <2 x bfloat> %461 to i32, !dbg !43 + %574 = bitcast <2 x bfloat> %462 to i32, !dbg !43 + %575 = bitcast <2 x bfloat> %463 to i32, !dbg !43 + %576 = bitcast <2 x bfloat> %464 to i32, !dbg !43 + %577 = bitcast <2 x bfloat> %465 to i32, !dbg !43 + %578 = bitcast <2 x bfloat> %466 to i32, !dbg !43 + %579 = bitcast <2 x bfloat> %467 to i32, !dbg !43 + %580 = bitcast <2 x bfloat> %468 to i32, !dbg !43 + %581 = bitcast <2 x bfloat> %469 to i32, !dbg !43 + %582 = bitcast <2 x bfloat> %470 to i32, !dbg !43 + %583 = bitcast <2 x bfloat> %471 to i32, !dbg !43 + %584 = bitcast <2 x bfloat> %472 to i32, !dbg !43 + %585 = bitcast <2 x bfloat> %473 to i32, !dbg !43 + %586 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %475, i32 %476, i32 %477, i32 %478, i32 %554, i32 %555) #3, !dbg !43 + %587 = extractvalue { float, float, float, float } %586, 0, !dbg !43 + %588 = extractvalue { float, float, float, float } %586, 1, !dbg !43 + %589 = extractvalue { float, float, float, float } %586, 2, !dbg !43 + %590 = extractvalue { float, float, float, float } %586, 3, !dbg !43 + %591 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %475, i32 %476, i32 %477, i32 %478, i32 %562, i32 %563) #3, !dbg !43 + %592 = extractvalue { float, float, float, float } %591, 0, !dbg !43 + %593 = extractvalue { float, float, float, float } %591, 1, !dbg !43 + %594 = extractvalue { float, float, float, float } %591, 2, !dbg !43 + %595 = extractvalue { float, float, float, float } %591, 3, !dbg !43 + %596 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %475, i32 %476, i32 %477, i32 %478, i32 %570, i32 %571) #3, !dbg !43 + %597 = extractvalue { float, float, float, float } %596, 0, !dbg !43 + %598 = extractvalue { float, float, float, float } %596, 1, !dbg !43 + %599 = extractvalue { float, float, float, float } %596, 2, !dbg !43 + %600 = extractvalue { float, float, float, float } %596, 3, !dbg !43 + %601 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %475, i32 %476, i32 %477, i32 %478, i32 %578, i32 %579) #3, !dbg !43 + %602 = extractvalue { float, float, float, float } %601, 0, !dbg !43 + %603 = extractvalue { float, float, float, float } %601, 1, !dbg !43 + %604 = extractvalue { float, float, float, float } %601, 2, !dbg !43 + %605 = extractvalue { float, float, float, float } %601, 3, !dbg !43 + %606 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %480, i32 %481, i32 %482, i32 %483, i32 %554, i32 %555) #3, !dbg !43 + %607 = extractvalue { float, float, float, float } %606, 0, !dbg !43 + %608 = extractvalue { float, float, float, float } %606, 1, !dbg !43 + %609 = extractvalue { float, float, float, float } %606, 2, !dbg !43 + %610 = extractvalue { float, float, float, float } %606, 3, !dbg !43 + %611 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %480, i32 %481, i32 %482, i32 %483, i32 %562, i32 %563) #3, !dbg !43 + %612 = extractvalue { float, float, float, float } %611, 0, !dbg !43 + %613 = extractvalue { float, float, float, float } %611, 1, !dbg !43 + %614 = extractvalue { float, float, float, float } %611, 2, !dbg !43 + %615 = extractvalue { float, float, float, float } %611, 3, !dbg !43 + %616 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %480, i32 %481, i32 %482, i32 %483, i32 %570, i32 %571) #3, !dbg !43 + %617 = extractvalue { float, float, float, float } %616, 0, !dbg !43 + %618 = extractvalue { float, float, float, float } %616, 1, !dbg !43 + %619 = extractvalue { float, float, float, float } %616, 2, !dbg !43 + %620 = extractvalue { float, float, float, float } %616, 3, !dbg !43 + %621 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %480, i32 %481, i32 %482, i32 %483, i32 %578, i32 %579) #3, !dbg !43 + %622 = extractvalue { float, float, float, float } %621, 0, !dbg !43 + %623 = extractvalue { float, float, float, float } %621, 1, !dbg !43 + %624 = extractvalue { float, float, float, float } %621, 2, !dbg !43 + %625 = extractvalue { float, float, float, float } %621, 3, !dbg !43 + %626 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %485, i32 %486, i32 %487, i32 %488, i32 %554, i32 %555) #3, !dbg !43 + %627 = extractvalue { float, float, float, float } %626, 0, !dbg !43 + %628 = extractvalue { float, float, float, float } %626, 1, !dbg !43 + %629 = extractvalue { float, float, float, float } %626, 2, !dbg !43 + %630 = extractvalue { float, float, float, float } %626, 3, !dbg !43 + %631 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %485, i32 %486, i32 %487, i32 %488, i32 %562, i32 %563) #3, !dbg !43 + %632 = extractvalue { float, float, float, float } %631, 0, !dbg !43 + %633 = extractvalue { float, float, float, float } %631, 1, !dbg !43 + %634 = extractvalue { float, float, float, float } %631, 2, !dbg !43 + %635 = extractvalue { float, float, float, float } %631, 3, !dbg !43 + %636 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %485, i32 %486, i32 %487, i32 %488, i32 %570, i32 %571) #3, !dbg !43 + %637 = extractvalue { float, float, float, float } %636, 0, !dbg !43 + %638 = extractvalue { float, float, float, float } %636, 1, !dbg !43 + %639 = extractvalue { float, float, float, float } %636, 2, !dbg !43 + %640 = extractvalue { float, float, float, float } %636, 3, !dbg !43 + %641 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %485, i32 %486, i32 %487, i32 %488, i32 %578, i32 %579) #3, !dbg !43 + %642 = extractvalue { float, float, float, float } %641, 0, !dbg !43 + %643 = extractvalue { float, float, float, float } %641, 1, !dbg !43 + %644 = extractvalue { float, float, float, float } %641, 2, !dbg !43 + %645 = extractvalue { float, float, float, float } %641, 3, !dbg !43 + %646 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %490, i32 %491, i32 %492, i32 %493, i32 %554, i32 %555) #3, !dbg !43 + %647 = extractvalue { float, float, float, float } %646, 0, !dbg !43 + %648 = extractvalue { float, float, float, float } %646, 1, !dbg !43 + %649 = extractvalue { float, float, float, float } %646, 2, !dbg !43 + %650 = extractvalue { float, float, float, float } %646, 3, !dbg !43 + %651 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %490, i32 %491, i32 %492, i32 %493, i32 %562, i32 %563) #3, !dbg !43 + %652 = extractvalue { float, float, float, float } %651, 0, !dbg !43 + %653 = extractvalue { float, float, float, float } %651, 1, !dbg !43 + %654 = extractvalue { float, float, float, float } %651, 2, !dbg !43 + %655 = extractvalue { float, float, float, float } %651, 3, !dbg !43 + %656 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %490, i32 %491, i32 %492, i32 %493, i32 %570, i32 %571) #3, !dbg !43 + %657 = extractvalue { float, float, float, float } %656, 0, !dbg !43 + %658 = extractvalue { float, float, float, float } %656, 1, !dbg !43 + %659 = extractvalue { float, float, float, float } %656, 2, !dbg !43 + %660 = extractvalue { float, float, float, float } %656, 3, !dbg !43 + %661 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %490, i32 %491, i32 %492, i32 %493, i32 %578, i32 %579) #3, !dbg !43 + %662 = extractvalue { float, float, float, float } %661, 0, !dbg !43 + %663 = extractvalue { float, float, float, float } %661, 1, !dbg !43 + %664 = extractvalue { float, float, float, float } %661, 2, !dbg !43 + %665 = extractvalue { float, float, float, float } %661, 3, !dbg !43 + %666 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %587, float %588, float %589, float %590, i32 %495, i32 %496, i32 %497, i32 %498, i32 %556, i32 %557) #3, !dbg !43 + %667 = extractvalue { float, float, float, float } %666, 0, !dbg !43 + %668 = extractvalue { float, float, float, float } %666, 1, !dbg !43 + %669 = extractvalue { float, float, float, float } %666, 2, !dbg !43 + %670 = extractvalue { float, float, float, float } %666, 3, !dbg !43 + %671 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %592, float %593, float %594, float %595, i32 %495, i32 %496, i32 %497, i32 %498, i32 %564, i32 %565) #3, !dbg !43 + %672 = extractvalue { float, float, float, float } %671, 0, !dbg !43 + %673 = extractvalue { float, float, float, float } %671, 1, !dbg !43 + %674 = extractvalue { float, float, float, float } %671, 2, !dbg !43 + %675 = extractvalue { float, float, float, float } %671, 3, !dbg !43 + %676 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %597, float %598, float %599, float %600, i32 %495, i32 %496, i32 %497, i32 %498, i32 %572, i32 %573) #3, !dbg !43 + %677 = extractvalue { float, float, float, float } %676, 0, !dbg !43 + %678 = extractvalue { float, float, float, float } %676, 1, !dbg !43 + %679 = extractvalue { float, float, float, float } %676, 2, !dbg !43 + %680 = extractvalue { float, float, float, float } %676, 3, !dbg !43 + %681 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %602, float %603, float %604, float %605, i32 %495, i32 %496, i32 %497, i32 %498, i32 %580, i32 %581) #3, !dbg !43 + %682 = extractvalue { float, float, float, float } %681, 0, !dbg !43 + %683 = extractvalue { float, float, float, float } %681, 1, !dbg !43 + %684 = extractvalue { float, float, float, float } %681, 2, !dbg !43 + %685 = extractvalue { float, float, float, float } %681, 3, !dbg !43 + %686 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %607, float %608, float %609, float %610, i32 %500, i32 %501, i32 %502, i32 %503, i32 %556, i32 %557) #3, !dbg !43 + %687 = extractvalue { float, float, float, float } %686, 0, !dbg !43 + %688 = extractvalue { float, float, float, float } %686, 1, !dbg !43 + %689 = extractvalue { float, float, float, float } %686, 2, !dbg !43 + %690 = extractvalue { float, float, float, float } %686, 3, !dbg !43 + %691 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %612, float %613, float %614, float %615, i32 %500, i32 %501, i32 %502, i32 %503, i32 %564, i32 %565) #3, !dbg !43 + %692 = extractvalue { float, float, float, float } %691, 0, !dbg !43 + %693 = extractvalue { float, float, float, float } %691, 1, !dbg !43 + %694 = extractvalue { float, float, float, float } %691, 2, !dbg !43 + %695 = extractvalue { float, float, float, float } %691, 3, !dbg !43 + %696 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %617, float %618, float %619, float %620, i32 %500, i32 %501, i32 %502, i32 %503, i32 %572, i32 %573) #3, !dbg !43 + %697 = extractvalue { float, float, float, float } %696, 0, !dbg !43 + %698 = extractvalue { float, float, float, float } %696, 1, !dbg !43 + %699 = extractvalue { float, float, float, float } %696, 2, !dbg !43 + %700 = extractvalue { float, float, float, float } %696, 3, !dbg !43 + %701 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %622, float %623, float %624, float %625, i32 %500, i32 %501, i32 %502, i32 %503, i32 %580, i32 %581) #3, !dbg !43 + %702 = extractvalue { float, float, float, float } %701, 0, !dbg !43 + %703 = extractvalue { float, float, float, float } %701, 1, !dbg !43 + %704 = extractvalue { float, float, float, float } %701, 2, !dbg !43 + %705 = extractvalue { float, float, float, float } %701, 3, !dbg !43 + %706 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %627, float %628, float %629, float %630, i32 %505, i32 %506, i32 %507, i32 %508, i32 %556, i32 %557) #3, !dbg !43 + %707 = extractvalue { float, float, float, float } %706, 0, !dbg !43 + %708 = extractvalue { float, float, float, float } %706, 1, !dbg !43 + %709 = extractvalue { float, float, float, float } %706, 2, !dbg !43 + %710 = extractvalue { float, float, float, float } %706, 3, !dbg !43 + %711 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %632, float %633, float %634, float %635, i32 %505, i32 %506, i32 %507, i32 %508, i32 %564, i32 %565) #3, !dbg !43 + %712 = extractvalue { float, float, float, float } %711, 0, !dbg !43 + %713 = extractvalue { float, float, float, float } %711, 1, !dbg !43 + %714 = extractvalue { float, float, float, float } %711, 2, !dbg !43 + %715 = extractvalue { float, float, float, float } %711, 3, !dbg !43 + %716 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %637, float %638, float %639, float %640, i32 %505, i32 %506, i32 %507, i32 %508, i32 %572, i32 %573) #3, !dbg !43 + %717 = extractvalue { float, float, float, float } %716, 0, !dbg !43 + %718 = extractvalue { float, float, float, float } %716, 1, !dbg !43 + %719 = extractvalue { float, float, float, float } %716, 2, !dbg !43 + %720 = extractvalue { float, float, float, float } %716, 3, !dbg !43 + %721 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %642, float %643, float %644, float %645, i32 %505, i32 %506, i32 %507, i32 %508, i32 %580, i32 %581) #3, !dbg !43 + %722 = extractvalue { float, float, float, float } %721, 0, !dbg !43 + %723 = extractvalue { float, float, float, float } %721, 1, !dbg !43 + %724 = extractvalue { float, float, float, float } %721, 2, !dbg !43 + %725 = extractvalue { float, float, float, float } %721, 3, !dbg !43 + %726 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %647, float %648, float %649, float %650, i32 %510, i32 %511, i32 %512, i32 %513, i32 %556, i32 %557) #3, !dbg !43 + %727 = extractvalue { float, float, float, float } %726, 0, !dbg !43 + %728 = extractvalue { float, float, float, float } %726, 1, !dbg !43 + %729 = extractvalue { float, float, float, float } %726, 2, !dbg !43 + %730 = extractvalue { float, float, float, float } %726, 3, !dbg !43 + %731 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %652, float %653, float %654, float %655, i32 %510, i32 %511, i32 %512, i32 %513, i32 %564, i32 %565) #3, !dbg !43 + %732 = extractvalue { float, float, float, float } %731, 0, !dbg !43 + %733 = extractvalue { float, float, float, float } %731, 1, !dbg !43 + %734 = extractvalue { float, float, float, float } %731, 2, !dbg !43 + %735 = extractvalue { float, float, float, float } %731, 3, !dbg !43 + %736 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %657, float %658, float %659, float %660, i32 %510, i32 %511, i32 %512, i32 %513, i32 %572, i32 %573) #3, !dbg !43 + %737 = extractvalue { float, float, float, float } %736, 0, !dbg !43 + %738 = extractvalue { float, float, float, float } %736, 1, !dbg !43 + %739 = extractvalue { float, float, float, float } %736, 2, !dbg !43 + %740 = extractvalue { float, float, float, float } %736, 3, !dbg !43 + %741 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %662, float %663, float %664, float %665, i32 %510, i32 %511, i32 %512, i32 %513, i32 %580, i32 %581) #3, !dbg !43 + %742 = extractvalue { float, float, float, float } %741, 0, !dbg !43 + %743 = extractvalue { float, float, float, float } %741, 1, !dbg !43 + %744 = extractvalue { float, float, float, float } %741, 2, !dbg !43 + %745 = extractvalue { float, float, float, float } %741, 3, !dbg !43 + %746 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %667, float %668, float %669, float %670, i32 %515, i32 %516, i32 %517, i32 %518, i32 %558, i32 %559) #3, !dbg !43 + %747 = extractvalue { float, float, float, float } %746, 0, !dbg !43 + %748 = extractvalue { float, float, float, float } %746, 1, !dbg !43 + %749 = extractvalue { float, float, float, float } %746, 2, !dbg !43 + %750 = extractvalue { float, float, float, float } %746, 3, !dbg !43 + %751 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %672, float %673, float %674, float %675, i32 %515, i32 %516, i32 %517, i32 %518, i32 %566, i32 %567) #3, !dbg !43 + %752 = extractvalue { float, float, float, float } %751, 0, !dbg !43 + %753 = extractvalue { float, float, float, float } %751, 1, !dbg !43 + %754 = extractvalue { float, float, float, float } %751, 2, !dbg !43 + %755 = extractvalue { float, float, float, float } %751, 3, !dbg !43 + %756 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %677, float %678, float %679, float %680, i32 %515, i32 %516, i32 %517, i32 %518, i32 %574, i32 %575) #3, !dbg !43 + %757 = extractvalue { float, float, float, float } %756, 0, !dbg !43 + %758 = extractvalue { float, float, float, float } %756, 1, !dbg !43 + %759 = extractvalue { float, float, float, float } %756, 2, !dbg !43 + %760 = extractvalue { float, float, float, float } %756, 3, !dbg !43 + %761 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %682, float %683, float %684, float %685, i32 %515, i32 %516, i32 %517, i32 %518, i32 %582, i32 %583) #3, !dbg !43 + %762 = extractvalue { float, float, float, float } %761, 0, !dbg !43 + %763 = extractvalue { float, float, float, float } %761, 1, !dbg !43 + %764 = extractvalue { float, float, float, float } %761, 2, !dbg !43 + %765 = extractvalue { float, float, float, float } %761, 3, !dbg !43 + %766 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %687, float %688, float %689, float %690, i32 %520, i32 %521, i32 %522, i32 %523, i32 %558, i32 %559) #3, !dbg !43 + %767 = extractvalue { float, float, float, float } %766, 0, !dbg !43 + %768 = extractvalue { float, float, float, float } %766, 1, !dbg !43 + %769 = extractvalue { float, float, float, float } %766, 2, !dbg !43 + %770 = extractvalue { float, float, float, float } %766, 3, !dbg !43 + %771 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %692, float %693, float %694, float %695, i32 %520, i32 %521, i32 %522, i32 %523, i32 %566, i32 %567) #3, !dbg !43 + %772 = extractvalue { float, float, float, float } %771, 0, !dbg !43 + %773 = extractvalue { float, float, float, float } %771, 1, !dbg !43 + %774 = extractvalue { float, float, float, float } %771, 2, !dbg !43 + %775 = extractvalue { float, float, float, float } %771, 3, !dbg !43 + %776 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %697, float %698, float %699, float %700, i32 %520, i32 %521, i32 %522, i32 %523, i32 %574, i32 %575) #3, !dbg !43 + %777 = extractvalue { float, float, float, float } %776, 0, !dbg !43 + %778 = extractvalue { float, float, float, float } %776, 1, !dbg !43 + %779 = extractvalue { float, float, float, float } %776, 2, !dbg !43 + %780 = extractvalue { float, float, float, float } %776, 3, !dbg !43 + %781 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %702, float %703, float %704, float %705, i32 %520, i32 %521, i32 %522, i32 %523, i32 %582, i32 %583) #3, !dbg !43 + %782 = extractvalue { float, float, float, float } %781, 0, !dbg !43 + %783 = extractvalue { float, float, float, float } %781, 1, !dbg !43 + %784 = extractvalue { float, float, float, float } %781, 2, !dbg !43 + %785 = extractvalue { float, float, float, float } %781, 3, !dbg !43 + %786 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %707, float %708, float %709, float %710, i32 %525, i32 %526, i32 %527, i32 %528, i32 %558, i32 %559) #3, !dbg !43 + %787 = extractvalue { float, float, float, float } %786, 0, !dbg !43 + %788 = extractvalue { float, float, float, float } %786, 1, !dbg !43 + %789 = extractvalue { float, float, float, float } %786, 2, !dbg !43 + %790 = extractvalue { float, float, float, float } %786, 3, !dbg !43 + %791 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %712, float %713, float %714, float %715, i32 %525, i32 %526, i32 %527, i32 %528, i32 %566, i32 %567) #3, !dbg !43 + %792 = extractvalue { float, float, float, float } %791, 0, !dbg !43 + %793 = extractvalue { float, float, float, float } %791, 1, !dbg !43 + %794 = extractvalue { float, float, float, float } %791, 2, !dbg !43 + %795 = extractvalue { float, float, float, float } %791, 3, !dbg !43 + %796 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %717, float %718, float %719, float %720, i32 %525, i32 %526, i32 %527, i32 %528, i32 %574, i32 %575) #3, !dbg !43 + %797 = extractvalue { float, float, float, float } %796, 0, !dbg !43 + %798 = extractvalue { float, float, float, float } %796, 1, !dbg !43 + %799 = extractvalue { float, float, float, float } %796, 2, !dbg !43 + %800 = extractvalue { float, float, float, float } %796, 3, !dbg !43 + %801 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %722, float %723, float %724, float %725, i32 %525, i32 %526, i32 %527, i32 %528, i32 %582, i32 %583) #3, !dbg !43 + %802 = extractvalue { float, float, float, float } %801, 0, !dbg !43 + %803 = extractvalue { float, float, float, float } %801, 1, !dbg !43 + %804 = extractvalue { float, float, float, float } %801, 2, !dbg !43 + %805 = extractvalue { float, float, float, float } %801, 3, !dbg !43 + %806 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %727, float %728, float %729, float %730, i32 %530, i32 %531, i32 %532, i32 %533, i32 %558, i32 %559) #3, !dbg !43 + %807 = extractvalue { float, float, float, float } %806, 0, !dbg !43 + %808 = extractvalue { float, float, float, float } %806, 1, !dbg !43 + %809 = extractvalue { float, float, float, float } %806, 2, !dbg !43 + %810 = extractvalue { float, float, float, float } %806, 3, !dbg !43 + %811 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %732, float %733, float %734, float %735, i32 %530, i32 %531, i32 %532, i32 %533, i32 %566, i32 %567) #3, !dbg !43 + %812 = extractvalue { float, float, float, float } %811, 0, !dbg !43 + %813 = extractvalue { float, float, float, float } %811, 1, !dbg !43 + %814 = extractvalue { float, float, float, float } %811, 2, !dbg !43 + %815 = extractvalue { float, float, float, float } %811, 3, !dbg !43 + %816 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %737, float %738, float %739, float %740, i32 %530, i32 %531, i32 %532, i32 %533, i32 %574, i32 %575) #3, !dbg !43 + %817 = extractvalue { float, float, float, float } %816, 0, !dbg !43 + %818 = extractvalue { float, float, float, float } %816, 1, !dbg !43 + %819 = extractvalue { float, float, float, float } %816, 2, !dbg !43 + %820 = extractvalue { float, float, float, float } %816, 3, !dbg !43 + %821 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %742, float %743, float %744, float %745, i32 %530, i32 %531, i32 %532, i32 %533, i32 %582, i32 %583) #3, !dbg !43 + %822 = extractvalue { float, float, float, float } %821, 0, !dbg !43 + %823 = extractvalue { float, float, float, float } %821, 1, !dbg !43 + %824 = extractvalue { float, float, float, float } %821, 2, !dbg !43 + %825 = extractvalue { float, float, float, float } %821, 3, !dbg !43 + %826 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %747, float %748, float %749, float %750, i32 %535, i32 %536, i32 %537, i32 %538, i32 %560, i32 %561) #3, !dbg !43 + %827 = extractvalue { float, float, float, float } %826, 0, !dbg !43 + %828 = extractvalue { float, float, float, float } %826, 1, !dbg !43 + %829 = extractvalue { float, float, float, float } %826, 2, !dbg !43 + %830 = extractvalue { float, float, float, float } %826, 3, !dbg !43 + %831 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %752, float %753, float %754, float %755, i32 %535, i32 %536, i32 %537, i32 %538, i32 %568, i32 %569) #3, !dbg !43 + %832 = extractvalue { float, float, float, float } %831, 0, !dbg !43 + %833 = extractvalue { float, float, float, float } %831, 1, !dbg !43 + %834 = extractvalue { float, float, float, float } %831, 2, !dbg !43 + %835 = extractvalue { float, float, float, float } %831, 3, !dbg !43 + %836 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %757, float %758, float %759, float %760, i32 %535, i32 %536, i32 %537, i32 %538, i32 %576, i32 %577) #3, !dbg !43 + %837 = extractvalue { float, float, float, float } %836, 0, !dbg !43 + %838 = extractvalue { float, float, float, float } %836, 1, !dbg !43 + %839 = extractvalue { float, float, float, float } %836, 2, !dbg !43 + %840 = extractvalue { float, float, float, float } %836, 3, !dbg !43 + %841 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %762, float %763, float %764, float %765, i32 %535, i32 %536, i32 %537, i32 %538, i32 %584, i32 %585) #3, !dbg !43 + %842 = extractvalue { float, float, float, float } %841, 0, !dbg !43 + %843 = extractvalue { float, float, float, float } %841, 1, !dbg !43 + %844 = extractvalue { float, float, float, float } %841, 2, !dbg !43 + %845 = extractvalue { float, float, float, float } %841, 3, !dbg !43 + %846 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %767, float %768, float %769, float %770, i32 %540, i32 %541, i32 %542, i32 %543, i32 %560, i32 %561) #3, !dbg !43 + %847 = extractvalue { float, float, float, float } %846, 0, !dbg !43 + %848 = extractvalue { float, float, float, float } %846, 1, !dbg !43 + %849 = extractvalue { float, float, float, float } %846, 2, !dbg !43 + %850 = extractvalue { float, float, float, float } %846, 3, !dbg !43 + %851 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %772, float %773, float %774, float %775, i32 %540, i32 %541, i32 %542, i32 %543, i32 %568, i32 %569) #3, !dbg !43 + %852 = extractvalue { float, float, float, float } %851, 0, !dbg !43 + %853 = extractvalue { float, float, float, float } %851, 1, !dbg !43 + %854 = extractvalue { float, float, float, float } %851, 2, !dbg !43 + %855 = extractvalue { float, float, float, float } %851, 3, !dbg !43 + %856 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %777, float %778, float %779, float %780, i32 %540, i32 %541, i32 %542, i32 %543, i32 %576, i32 %577) #3, !dbg !43 + %857 = extractvalue { float, float, float, float } %856, 0, !dbg !43 + %858 = extractvalue { float, float, float, float } %856, 1, !dbg !43 + %859 = extractvalue { float, float, float, float } %856, 2, !dbg !43 + %860 = extractvalue { float, float, float, float } %856, 3, !dbg !43 + %861 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %782, float %783, float %784, float %785, i32 %540, i32 %541, i32 %542, i32 %543, i32 %584, i32 %585) #3, !dbg !43 + %862 = extractvalue { float, float, float, float } %861, 0, !dbg !43 + %863 = extractvalue { float, float, float, float } %861, 1, !dbg !43 + %864 = extractvalue { float, float, float, float } %861, 2, !dbg !43 + %865 = extractvalue { float, float, float, float } %861, 3, !dbg !43 + %866 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %787, float %788, float %789, float %790, i32 %545, i32 %546, i32 %547, i32 %548, i32 %560, i32 %561) #3, !dbg !43 + %867 = extractvalue { float, float, float, float } %866, 0, !dbg !43 + %868 = extractvalue { float, float, float, float } %866, 1, !dbg !43 + %869 = extractvalue { float, float, float, float } %866, 2, !dbg !43 + %870 = extractvalue { float, float, float, float } %866, 3, !dbg !43 + %871 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %792, float %793, float %794, float %795, i32 %545, i32 %546, i32 %547, i32 %548, i32 %568, i32 %569) #3, !dbg !43 + %872 = extractvalue { float, float, float, float } %871, 0, !dbg !43 + %873 = extractvalue { float, float, float, float } %871, 1, !dbg !43 + %874 = extractvalue { float, float, float, float } %871, 2, !dbg !43 + %875 = extractvalue { float, float, float, float } %871, 3, !dbg !43 + %876 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %797, float %798, float %799, float %800, i32 %545, i32 %546, i32 %547, i32 %548, i32 %576, i32 %577) #3, !dbg !43 + %877 = extractvalue { float, float, float, float } %876, 0, !dbg !43 + %878 = extractvalue { float, float, float, float } %876, 1, !dbg !43 + %879 = extractvalue { float, float, float, float } %876, 2, !dbg !43 + %880 = extractvalue { float, float, float, float } %876, 3, !dbg !43 + %881 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %802, float %803, float %804, float %805, i32 %545, i32 %546, i32 %547, i32 %548, i32 %584, i32 %585) #3, !dbg !43 + %882 = extractvalue { float, float, float, float } %881, 0, !dbg !43 + %883 = extractvalue { float, float, float, float } %881, 1, !dbg !43 + %884 = extractvalue { float, float, float, float } %881, 2, !dbg !43 + %885 = extractvalue { float, float, float, float } %881, 3, !dbg !43 + %886 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %807, float %808, float %809, float %810, i32 %550, i32 %551, i32 %552, i32 %553, i32 %560, i32 %561) #3, !dbg !43 + %887 = extractvalue { float, float, float, float } %886, 0, !dbg !43 + %888 = extractvalue { float, float, float, float } %886, 1, !dbg !43 + %889 = extractvalue { float, float, float, float } %886, 2, !dbg !43 + %890 = extractvalue { float, float, float, float } %886, 3, !dbg !43 + %891 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %812, float %813, float %814, float %815, i32 %550, i32 %551, i32 %552, i32 %553, i32 %568, i32 %569) #3, !dbg !43 + %892 = extractvalue { float, float, float, float } %891, 0, !dbg !43 + %893 = extractvalue { float, float, float, float } %891, 1, !dbg !43 + %894 = extractvalue { float, float, float, float } %891, 2, !dbg !43 + %895 = extractvalue { float, float, float, float } %891, 3, !dbg !43 + %896 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %817, float %818, float %819, float %820, i32 %550, i32 %551, i32 %552, i32 %553, i32 %576, i32 %577) #3, !dbg !43 + %897 = extractvalue { float, float, float, float } %896, 0, !dbg !43 + %898 = extractvalue { float, float, float, float } %896, 1, !dbg !43 + %899 = extractvalue { float, float, float, float } %896, 2, !dbg !43 + %900 = extractvalue { float, float, float, float } %896, 3, !dbg !43 + %901 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %822, float %823, float %824, float %825, i32 %550, i32 %551, i32 %552, i32 %553, i32 %584, i32 %585) #3, !dbg !43 + %902 = extractvalue { float, float, float, float } %901, 0, !dbg !43 + %903 = extractvalue { float, float, float, float } %901, 1, !dbg !43 + %904 = extractvalue { float, float, float, float } %901, 2, !dbg !43 + %905 = extractvalue { float, float, float, float } %901, 3, !dbg !43 + %906 = insertelement <2 x float> poison, float %827, i64 0, !dbg !44 + %907 = insertelement <2 x float> %906, float %828, i64 1, !dbg !44 + %908 = fptrunc <2 x float> %907 to <2 x bfloat>, !dbg !44 + %909 = insertelement <2 x float> poison, float %829, i64 0, !dbg !44 + %910 = insertelement <2 x float> %909, float %830, i64 1, !dbg !44 + %911 = fptrunc <2 x float> %910 to <2 x bfloat>, !dbg !44 + %912 = insertelement <2 x float> poison, float %832, i64 0, !dbg !44 + %913 = insertelement <2 x float> %912, float %833, i64 1, !dbg !44 + %914 = fptrunc <2 x float> %913 to <2 x bfloat>, !dbg !44 + %915 = insertelement <2 x float> poison, float %834, i64 0, !dbg !44 + %916 = insertelement <2 x float> %915, float %835, i64 1, !dbg !44 + %917 = fptrunc <2 x float> %916 to <2 x bfloat>, !dbg !44 + %918 = insertelement <2 x float> poison, float %837, i64 0, !dbg !44 + %919 = insertelement <2 x float> %918, float %838, i64 1, !dbg !44 + %920 = fptrunc <2 x float> %919 to <2 x bfloat>, !dbg !44 + %921 = insertelement <2 x float> poison, float %839, i64 0, !dbg !44 + %922 = insertelement <2 x float> %921, float %840, i64 1, !dbg !44 + %923 = fptrunc <2 x float> %922 to <2 x bfloat>, !dbg !44 + %924 = insertelement <2 x float> poison, float %842, i64 0, !dbg !44 + %925 = insertelement <2 x float> %924, float %843, i64 1, !dbg !44 + %926 = fptrunc <2 x float> %925 to <2 x bfloat>, !dbg !44 + %927 = insertelement <2 x float> poison, float %844, i64 0, !dbg !44 + %928 = insertelement <2 x float> %927, float %845, i64 1, !dbg !44 + %929 = fptrunc <2 x float> %928 to <2 x bfloat>, !dbg !44 + %930 = insertelement <2 x float> poison, float %847, i64 0, !dbg !44 + %931 = insertelement <2 x float> %930, float %848, i64 1, !dbg !44 + %932 = fptrunc <2 x float> %931 to <2 x bfloat>, !dbg !44 + %933 = insertelement <2 x float> poison, float %849, i64 0, !dbg !44 + %934 = insertelement <2 x float> %933, float %850, i64 1, !dbg !44 + %935 = fptrunc <2 x float> %934 to <2 x bfloat>, !dbg !44 + %936 = insertelement <2 x float> poison, float %852, i64 0, !dbg !44 + %937 = insertelement <2 x float> %936, float %853, i64 1, !dbg !44 + %938 = fptrunc <2 x float> %937 to <2 x bfloat>, !dbg !44 + %939 = insertelement <2 x float> poison, float %854, i64 0, !dbg !44 + %940 = insertelement <2 x float> %939, float %855, i64 1, !dbg !44 + %941 = fptrunc <2 x float> %940 to <2 x bfloat>, !dbg !44 + %942 = insertelement <2 x float> poison, float %857, i64 0, !dbg !44 + %943 = insertelement <2 x float> %942, float %858, i64 1, !dbg !44 + %944 = fptrunc <2 x float> %943 to <2 x bfloat>, !dbg !44 + %945 = insertelement <2 x float> poison, float %859, i64 0, !dbg !44 + %946 = insertelement <2 x float> %945, float %860, i64 1, !dbg !44 + %947 = fptrunc <2 x float> %946 to <2 x bfloat>, !dbg !44 + %948 = insertelement <2 x float> poison, float %862, i64 0, !dbg !44 + %949 = insertelement <2 x float> %948, float %863, i64 1, !dbg !44 + %950 = fptrunc <2 x float> %949 to <2 x bfloat>, !dbg !44 + %951 = insertelement <2 x float> poison, float %864, i64 0, !dbg !44 + %952 = insertelement <2 x float> %951, float %865, i64 1, !dbg !44 + %953 = fptrunc <2 x float> %952 to <2 x bfloat>, !dbg !44 + %954 = insertelement <2 x float> poison, float %867, i64 0, !dbg !44 + %955 = insertelement <2 x float> %954, float %868, i64 1, !dbg !44 + %956 = fptrunc <2 x float> %955 to <2 x bfloat>, !dbg !44 + %957 = insertelement <2 x float> poison, float %869, i64 0, !dbg !44 + %958 = insertelement <2 x float> %957, float %870, i64 1, !dbg !44 + %959 = fptrunc <2 x float> %958 to <2 x bfloat>, !dbg !44 + %960 = insertelement <2 x float> poison, float %872, i64 0, !dbg !44 + %961 = insertelement <2 x float> %960, float %873, i64 1, !dbg !44 + %962 = fptrunc <2 x float> %961 to <2 x bfloat>, !dbg !44 + %963 = insertelement <2 x float> poison, float %874, i64 0, !dbg !44 + %964 = insertelement <2 x float> %963, float %875, i64 1, !dbg !44 + %965 = fptrunc <2 x float> %964 to <2 x bfloat>, !dbg !44 + %966 = insertelement <2 x float> poison, float %877, i64 0, !dbg !44 + %967 = insertelement <2 x float> %966, float %878, i64 1, !dbg !44 + %968 = fptrunc <2 x float> %967 to <2 x bfloat>, !dbg !44 + %969 = insertelement <2 x float> poison, float %879, i64 0, !dbg !44 + %970 = insertelement <2 x float> %969, float %880, i64 1, !dbg !44 + %971 = fptrunc <2 x float> %970 to <2 x bfloat>, !dbg !44 + %972 = insertelement <2 x float> poison, float %882, i64 0, !dbg !44 + %973 = insertelement <2 x float> %972, float %883, i64 1, !dbg !44 + %974 = fptrunc <2 x float> %973 to <2 x bfloat>, !dbg !44 + %975 = insertelement <2 x float> poison, float %884, i64 0, !dbg !44 + %976 = insertelement <2 x float> %975, float %885, i64 1, !dbg !44 + %977 = fptrunc <2 x float> %976 to <2 x bfloat>, !dbg !44 + %978 = insertelement <2 x float> poison, float %887, i64 0, !dbg !44 + %979 = insertelement <2 x float> %978, float %888, i64 1, !dbg !44 + %980 = fptrunc <2 x float> %979 to <2 x bfloat>, !dbg !44 + %981 = insertelement <2 x float> poison, float %889, i64 0, !dbg !44 + %982 = insertelement <2 x float> %981, float %890, i64 1, !dbg !44 + %983 = fptrunc <2 x float> %982 to <2 x bfloat>, !dbg !44 + %984 = insertelement <2 x float> poison, float %892, i64 0, !dbg !44 + %985 = insertelement <2 x float> %984, float %893, i64 1, !dbg !44 + %986 = fptrunc <2 x float> %985 to <2 x bfloat>, !dbg !44 + %987 = insertelement <2 x float> poison, float %894, i64 0, !dbg !44 + %988 = insertelement <2 x float> %987, float %895, i64 1, !dbg !44 + %989 = fptrunc <2 x float> %988 to <2 x bfloat>, !dbg !44 + %990 = insertelement <2 x float> poison, float %897, i64 0, !dbg !44 + %991 = insertelement <2 x float> %990, float %898, i64 1, !dbg !44 + %992 = fptrunc <2 x float> %991 to <2 x bfloat>, !dbg !44 + %993 = insertelement <2 x float> poison, float %899, i64 0, !dbg !44 + %994 = insertelement <2 x float> %993, float %900, i64 1, !dbg !44 + %995 = fptrunc <2 x float> %994 to <2 x bfloat>, !dbg !44 + %996 = insertelement <2 x float> poison, float %902, i64 0, !dbg !44 + %997 = insertelement <2 x float> %996, float %903, i64 1, !dbg !44 + %998 = fptrunc <2 x float> %997 to <2 x bfloat>, !dbg !44 + %999 = insertelement <2 x float> poison, float %904, i64 0, !dbg !44 + %1000 = insertelement <2 x float> %999, float %905, i64 1, !dbg !44 + %1001 = fptrunc <2 x float> %1000 to <2 x bfloat>, !dbg !44 + %1002 = bitcast <2 x bfloat> %908 to i32, !dbg !45 + %1003 = bitcast <2 x bfloat> %914 to i32, !dbg !45 + %1004 = bitcast <2 x bfloat> %920 to i32, !dbg !45 + %1005 = bitcast <2 x bfloat> %926 to i32, !dbg !45 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %301, i32 %1002, i32 %1003, i32 %1004, i32 %1005), !dbg !45 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !45 + %1006 = load <4 x i32>, ptr addrspace(3) %302, align 16, !dbg !45 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !45 + %1007 = bitcast <2 x bfloat> %911 to i32, !dbg !45 + %1008 = bitcast <2 x bfloat> %917 to i32, !dbg !45 + %1009 = bitcast <2 x bfloat> %923 to i32, !dbg !45 + %1010 = bitcast <2 x bfloat> %929 to i32, !dbg !45 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %301, i32 %1007, i32 %1008, i32 %1009, i32 %1010), !dbg !45 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !45 + %1011 = load <4 x i32>, ptr addrspace(3) %302, align 16, !dbg !45 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !45 + %1012 = bitcast <2 x bfloat> %932 to i32, !dbg !45 + %1013 = bitcast <2 x bfloat> %938 to i32, !dbg !45 + %1014 = bitcast <2 x bfloat> %944 to i32, !dbg !45 + %1015 = bitcast <2 x bfloat> %950 to i32, !dbg !45 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %301, i32 %1012, i32 %1013, i32 %1014, i32 %1015), !dbg !45 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !45 + %1016 = load <4 x i32>, ptr addrspace(3) %302, align 16, !dbg !45 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !45 + %1017 = bitcast <2 x bfloat> %935 to i32, !dbg !45 + %1018 = bitcast <2 x bfloat> %941 to i32, !dbg !45 + %1019 = bitcast <2 x bfloat> %947 to i32, !dbg !45 + %1020 = bitcast <2 x bfloat> %953 to i32, !dbg !45 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %301, i32 %1017, i32 %1018, i32 %1019, i32 %1020), !dbg !45 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !45 + %1021 = load <4 x i32>, ptr addrspace(3) %302, align 16, !dbg !45 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !45 + %1022 = bitcast <2 x bfloat> %956 to i32, !dbg !45 + %1023 = bitcast <2 x bfloat> %962 to i32, !dbg !45 + %1024 = bitcast <2 x bfloat> %968 to i32, !dbg !45 + %1025 = bitcast <2 x bfloat> %974 to i32, !dbg !45 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %301, i32 %1022, i32 %1023, i32 %1024, i32 %1025), !dbg !45 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !45 + %1026 = load <4 x i32>, ptr addrspace(3) %302, align 16, !dbg !45 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !45 + %1027 = bitcast <2 x bfloat> %959 to i32, !dbg !45 + %1028 = bitcast <2 x bfloat> %965 to i32, !dbg !45 + %1029 = bitcast <2 x bfloat> %971 to i32, !dbg !45 + %1030 = bitcast <2 x bfloat> %977 to i32, !dbg !45 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %301, i32 %1027, i32 %1028, i32 %1029, i32 %1030), !dbg !45 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !45 + %1031 = load <4 x i32>, ptr addrspace(3) %302, align 16, !dbg !45 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !45 + %1032 = bitcast <2 x bfloat> %980 to i32, !dbg !45 + %1033 = bitcast <2 x bfloat> %986 to i32, !dbg !45 + %1034 = bitcast <2 x bfloat> %992 to i32, !dbg !45 + %1035 = bitcast <2 x bfloat> %998 to i32, !dbg !45 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %301, i32 %1032, i32 %1033, i32 %1034, i32 %1035), !dbg !45 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !45 + %1036 = load <4 x i32>, ptr addrspace(3) %302, align 16, !dbg !45 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !45 + %1037 = bitcast <2 x bfloat> %983 to i32, !dbg !45 + %1038 = bitcast <2 x bfloat> %989 to i32, !dbg !45 + %1039 = bitcast <2 x bfloat> %995 to i32, !dbg !45 + %1040 = bitcast <2 x bfloat> %1001 to i32, !dbg !45 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %301, i32 %1037, i32 %1038, i32 %1039, i32 %1040), !dbg !45 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !45 + %1041 = load <4 x i32>, ptr addrspace(3) %302, align 16, !dbg !45 + %.extract363 = extractelement <4 x i32> %1006, i64 0, !dbg !45 + %.extract365 = extractelement <4 x i32> %1006, i64 1, !dbg !45 + %.extract367 = extractelement <4 x i32> %1006, i64 2, !dbg !45 + %.extract369 = extractelement <4 x i32> %1006, i64 3, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract363, i32 %.extract365, i32 %.extract367, i32 %.extract369, ptr addrspace(1) %346, i1 %329) #3, !dbg !45 + %.extract371 = extractelement <4 x i32> %1011, i64 0, !dbg !45 + %.extract373 = extractelement <4 x i32> %1011, i64 1, !dbg !45 + %.extract375 = extractelement <4 x i32> %1011, i64 2, !dbg !45 + %.extract377 = extractelement <4 x i32> %1011, i64 3, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract371, i32 %.extract373, i32 %.extract375, i32 %.extract377, ptr addrspace(1) %348, i1 %330) #3, !dbg !45 + %.extract379 = extractelement <4 x i32> %1016, i64 0, !dbg !45 + %.extract381 = extractelement <4 x i32> %1016, i64 1, !dbg !45 + %.extract383 = extractelement <4 x i32> %1016, i64 2, !dbg !45 + %.extract385 = extractelement <4 x i32> %1016, i64 3, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract379, i32 %.extract381, i32 %.extract383, i32 %.extract385, ptr addrspace(1) %350, i1 %331) #3, !dbg !45 + %.extract387 = extractelement <4 x i32> %1021, i64 0, !dbg !45 + %.extract389 = extractelement <4 x i32> %1021, i64 1, !dbg !45 + %.extract391 = extractelement <4 x i32> %1021, i64 2, !dbg !45 + %.extract393 = extractelement <4 x i32> %1021, i64 3, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract387, i32 %.extract389, i32 %.extract391, i32 %.extract393, ptr addrspace(1) %352, i1 %332) #3, !dbg !45 + %.extract395 = extractelement <4 x i32> %1026, i64 0, !dbg !45 + %.extract397 = extractelement <4 x i32> %1026, i64 1, !dbg !45 + %.extract399 = extractelement <4 x i32> %1026, i64 2, !dbg !45 + %.extract401 = extractelement <4 x i32> %1026, i64 3, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract395, i32 %.extract397, i32 %.extract399, i32 %.extract401, ptr addrspace(1) %354, i1 %333) #3, !dbg !45 + %.extract403 = extractelement <4 x i32> %1031, i64 0, !dbg !45 + %.extract405 = extractelement <4 x i32> %1031, i64 1, !dbg !45 + %.extract407 = extractelement <4 x i32> %1031, i64 2, !dbg !45 + %.extract409 = extractelement <4 x i32> %1031, i64 3, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract403, i32 %.extract405, i32 %.extract407, i32 %.extract409, ptr addrspace(1) %356, i1 %334) #3, !dbg !45 + %.extract411 = extractelement <4 x i32> %1036, i64 0, !dbg !45 + %.extract413 = extractelement <4 x i32> %1036, i64 1, !dbg !45 + %.extract415 = extractelement <4 x i32> %1036, i64 2, !dbg !45 + %.extract417 = extractelement <4 x i32> %1036, i64 3, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract411, i32 %.extract413, i32 %.extract415, i32 %.extract417, ptr addrspace(1) %358, i1 %335) #3, !dbg !45 + %.extract419 = extractelement <4 x i32> %1041, i64 0, !dbg !45 + %.extract421 = extractelement <4 x i32> %1041, i64 1, !dbg !45 + %.extract423 = extractelement <4 x i32> %1041, i64 2, !dbg !45 + %.extract425 = extractelement <4 x i32> %1041, i64 3, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract419, i32 %.extract421, i32 %.extract423, i32 %.extract425, ptr addrspace(1) %360, i1 %336) #3, !dbg !45 + %1042 = add i32 %341, 1, !dbg !33 + %1043 = icmp sgt i32 %1042, 2, !dbg !33 + %1044 = select i1 %1043, i32 0, i32 %1042, !dbg !33 + %1045 = or disjoint i64 %indvars.iv, %111, !dbg !40 + %1046 = trunc nuw nsw i64 %1045 to i32, !dbg !38 + %1047 = getelementptr bfloat, ptr addrspace(1) %187, i64 %1045, !dbg !38 + %1048 = getelementptr bfloat, ptr addrspace(1) %188, i64 %1045, !dbg !38 + %1049 = getelementptr bfloat, ptr addrspace(1) %189, i64 %1045, !dbg !38 + %1050 = getelementptr bfloat, ptr addrspace(1) %190, i64 %1045, !dbg !38 + %1051 = getelementptr bfloat, ptr addrspace(1) %191, i64 %1045, !dbg !38 + %1052 = getelementptr bfloat, ptr addrspace(1) %192, i64 %1045, !dbg !38 + %1053 = getelementptr bfloat, ptr addrspace(1) %193, i64 %1045, !dbg !38 + %1054 = getelementptr bfloat, ptr addrspace(1) %194, i64 %1045, !dbg !38 + %.idx299 = shl i32 %1044, 13, !dbg !39 + %1055 = getelementptr i8, ptr addrspace(3) @global_smem, i32 %.idx299, !dbg !39 + %1056 = getelementptr inbounds nuw i8, ptr addrspace(3) %1055, i32 %170, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %1056, ptr addrspace(1) %1047, i32 0) #3, !dbg !39 + %1057 = getelementptr inbounds nuw i8, ptr addrspace(3) %1055, i32 %172, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %1057, ptr addrspace(1) %1048, i32 0) #3, !dbg !39 + %1058 = getelementptr inbounds nuw i8, ptr addrspace(3) %1055, i32 %174, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %1058, ptr addrspace(1) %1049, i32 0) #3, !dbg !39 + %1059 = getelementptr inbounds nuw i8, ptr addrspace(3) %1055, i32 %176, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %1059, ptr addrspace(1) %1050, i32 0) #3, !dbg !39 + %1060 = getelementptr inbounds nuw i8, ptr addrspace(3) %1055, i32 %178, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %1060, ptr addrspace(1) %1051, i32 0) #3, !dbg !39 + %1061 = getelementptr inbounds nuw i8, ptr addrspace(3) %1055, i32 %180, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %1061, ptr addrspace(1) %1052, i32 0) #3, !dbg !39 + %1062 = getelementptr inbounds nuw i8, ptr addrspace(3) %1055, i32 %182, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %1062, ptr addrspace(1) %1053, i32 0) #3, !dbg !39 + %1063 = getelementptr inbounds nuw i8, ptr addrspace(3) %1055, i32 %184, !dbg !39 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %1063, ptr addrspace(1) %1054, i32 0) #3, !dbg !39 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !39 + br i1 %320, label %319, label %1064, !dbg !33 + +1064: ; preds = %319 + tail call void @llvm.nvvm.cp.async.wait.group(i32 0), !dbg !33 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !33 + %1065 = getelementptr float, ptr addrspace(1) %6, i64 %42, !dbg !46 + %1066 = getelementptr float, ptr addrspace(1) %1065, i64 %45, !dbg !47 + %1067 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %1066, i1 %47) #3, !dbg !48 + %1068 = bitcast i32 %1067 to float, !dbg !48 + %1069 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #3, !dbg !49 + %.not.i = icmp eq i32 %1069, 0, !dbg !49 + br i1 %.not.i, label %1072, label %1070, !dbg !49 + +1070: ; preds = %1064 + %1071 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1068), !dbg !49 + br label %__nv_exp2f.exit, !dbg !49 + +1072: ; preds = %1064 + %1073 = tail call float @llvm.nvvm.ex2.approx.f32(float %1068), !dbg !49 + br label %__nv_exp2f.exit, !dbg !49 + +__nv_exp2f.exit: ; preds = %1070, %1072 + %.0.i = phi float [ %1071, %1070 ], [ %1073, %1072 ], !dbg !49 + %1074 = and i32 %52, 132, !dbg !49 + %1075 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %50, !dbg !49 + %1076 = getelementptr inbounds nuw i8, ptr addrspace(3) %1075, i32 %1074, !dbg !49 + %1077 = getelementptr inbounds nuw i8, ptr addrspace(3) %1076, i32 %51, !dbg !49 + %1078 = bitcast float %.0.i to <1 x i32>, !dbg !49 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) %1077, <1 x i32> %1078) #3, !dbg !49 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !49 + %1079 = load <2 x float>, ptr addrspace(3) %63, align 16, !dbg !49 + %1080 = load <2 x float>, ptr addrspace(3) %71, align 8, !dbg !49 + %1081 = load <2 x float>, ptr addrspace(3) %79, align 16, !dbg !49 + %1082 = load <2 x float>, ptr addrspace(3) %87, align 8, !dbg !49 + %1083 = getelementptr inbounds nuw i8, ptr addrspace(3) %63, i32 128, !dbg !49 + %1084 = load <2 x float>, ptr addrspace(3) %1083, align 16, !dbg !49 + %1085 = getelementptr inbounds nuw i8, ptr addrspace(3) %63, i32 136, !dbg !49 + %1086 = load <2 x float>, ptr addrspace(3) %1085, align 8, !dbg !49 + %1087 = getelementptr inbounds nuw i8, ptr addrspace(3) %63, i32 144, !dbg !49 + %1088 = load <2 x float>, ptr addrspace(3) %1087, align 16, !dbg !49 + %1089 = getelementptr inbounds nuw i8, ptr addrspace(3) %63, i32 152, !dbg !49 + %1090 = load <2 x float>, ptr addrspace(3) %1089, align 8, !dbg !49 + %1091 = getelementptr bfloat, ptr addrspace(1) %0, i64 %.idx, !dbg !54 + %1092 = getelementptr bfloat, ptr addrspace(1) %1091, i64 %.idx24, !dbg !55 + %1093 = getelementptr bfloat, ptr addrspace(1) %1091, i64 %.idx25, !dbg !55 + %1094 = getelementptr bfloat, ptr addrspace(1) %1091, i64 %.idx26, !dbg !55 + %1095 = getelementptr bfloat, ptr addrspace(1) %1091, i64 %.idx27, !dbg !55 + %1096 = getelementptr bfloat, ptr addrspace(1) %1091, i64 %.idx28, !dbg !55 + %1097 = getelementptr bfloat, ptr addrspace(1) %1091, i64 %.idx29, !dbg !55 + %1098 = getelementptr bfloat, ptr addrspace(1) %1091, i64 %.idx30, !dbg !55 + %1099 = getelementptr bfloat, ptr addrspace(1) %1091, i64 %.idx31, !dbg !55 + %1100 = getelementptr bfloat, ptr addrspace(1) %3, i64 %.idx, !dbg !56 + %1101 = getelementptr bfloat, ptr addrspace(1) %1100, i64 %.idx24, !dbg !57 + %1102 = getelementptr bfloat, ptr addrspace(1) %1100, i64 %.idx25, !dbg !57 + %1103 = getelementptr bfloat, ptr addrspace(1) %1100, i64 %.idx26, !dbg !57 + %1104 = getelementptr bfloat, ptr addrspace(1) %1100, i64 %.idx27, !dbg !57 + %1105 = getelementptr bfloat, ptr addrspace(1) %1100, i64 %.idx28, !dbg !57 + %1106 = getelementptr bfloat, ptr addrspace(1) %1100, i64 %.idx29, !dbg !57 + %1107 = getelementptr bfloat, ptr addrspace(1) %1100, i64 %.idx30, !dbg !57 + %1108 = getelementptr bfloat, ptr addrspace(1) %1100, i64 %.idx31, !dbg !57 + %1109 = getelementptr bfloat, ptr addrspace(1) %1092, i64 %111, !dbg !58 + %1110 = getelementptr bfloat, ptr addrspace(1) %1093, i64 %111, !dbg !58 + %1111 = getelementptr bfloat, ptr addrspace(1) %1094, i64 %111, !dbg !58 + %1112 = getelementptr bfloat, ptr addrspace(1) %1095, i64 %111, !dbg !58 + %1113 = getelementptr bfloat, ptr addrspace(1) %1096, i64 %111, !dbg !58 + %1114 = getelementptr bfloat, ptr addrspace(1) %1097, i64 %111, !dbg !58 + %1115 = getelementptr bfloat, ptr addrspace(1) %1098, i64 %111, !dbg !58 + %1116 = getelementptr bfloat, ptr addrspace(1) %1099, i64 %111, !dbg !58 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %212, ptr addrspace(1) %1109, i32 %213) #3, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %214, ptr addrspace(1) %1110, i32 %215) #3, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %216, ptr addrspace(1) %1111, i32 %217) #3, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %218, ptr addrspace(1) %1112, i32 %219) #3, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %220, ptr addrspace(1) %1113, i32 %221) #3, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %222, ptr addrspace(1) %1114, i32 %223) #3, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %224, ptr addrspace(1) %1115, i32 %225) #3, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %226, ptr addrspace(1) %1116, i32 %227) #3, !dbg !59 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !59 + %1117 = getelementptr bfloat, ptr addrspace(1) %1092, i64 %229, !dbg !58 + %1118 = getelementptr bfloat, ptr addrspace(1) %1093, i64 %229, !dbg !58 + %1119 = getelementptr bfloat, ptr addrspace(1) %1094, i64 %229, !dbg !58 + %1120 = getelementptr bfloat, ptr addrspace(1) %1095, i64 %229, !dbg !58 + %1121 = getelementptr bfloat, ptr addrspace(1) %1096, i64 %229, !dbg !58 + %1122 = getelementptr bfloat, ptr addrspace(1) %1097, i64 %229, !dbg !58 + %1123 = getelementptr bfloat, ptr addrspace(1) %1098, i64 %229, !dbg !58 + %1124 = getelementptr bfloat, ptr addrspace(1) %1099, i64 %229, !dbg !58 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %238, ptr addrspace(1) %1117, i32 %213) #3, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %239, ptr addrspace(1) %1118, i32 %215) #3, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %240, ptr addrspace(1) %1119, i32 %217) #3, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %241, ptr addrspace(1) %1120, i32 %219) #3, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %242, ptr addrspace(1) %1121, i32 %221) #3, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %243, ptr addrspace(1) %1122, i32 %223) #3, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %244, ptr addrspace(1) %1123, i32 %225) #3, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %245, ptr addrspace(1) %1124, i32 %227) #3, !dbg !59 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !59 + %1125 = getelementptr bfloat, ptr addrspace(1) %1092, i64 %247, !dbg !58 + %1126 = getelementptr bfloat, ptr addrspace(1) %1093, i64 %247, !dbg !58 + %1127 = getelementptr bfloat, ptr addrspace(1) %1094, i64 %247, !dbg !58 + %1128 = getelementptr bfloat, ptr addrspace(1) %1095, i64 %247, !dbg !58 + %1129 = getelementptr bfloat, ptr addrspace(1) %1096, i64 %247, !dbg !58 + %1130 = getelementptr bfloat, ptr addrspace(1) %1097, i64 %247, !dbg !58 + %1131 = getelementptr bfloat, ptr addrspace(1) %1098, i64 %247, !dbg !58 + %1132 = getelementptr bfloat, ptr addrspace(1) %1099, i64 %247, !dbg !58 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %256, ptr addrspace(1) %1125, i32 0) #3, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %257, ptr addrspace(1) %1126, i32 0) #3, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %258, ptr addrspace(1) %1127, i32 0) #3, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %259, ptr addrspace(1) %1128, i32 0) #3, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %260, ptr addrspace(1) %1129, i32 0) #3, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %261, ptr addrspace(1) %1130, i32 0) #3, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %262, ptr addrspace(1) %1131, i32 0) #3, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %263, ptr addrspace(1) %1132, i32 0) #3, !dbg !59 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !59 + br label %1133, !dbg !60 + +1133: ; preds = %__nv_exp2f.exit, %1133 + %1134 = phi i1 [ true, %__nv_exp2f.exit ], [ false, %1133 ] + %indvars.iv426 = phi i64 [ 192, %__nv_exp2f.exit ], [ 256, %1133 ] + %1135 = phi i1 [ %36, %__nv_exp2f.exit ], [ false, %1133 ] + %1136 = phi i1 [ %125, %__nv_exp2f.exit ], [ false, %1133 ] + %1137 = phi i1 [ %131, %__nv_exp2f.exit ], [ false, %1133 ] + %1138 = phi i1 [ %137, %__nv_exp2f.exit ], [ false, %1133 ] + %1139 = phi i1 [ %143, %__nv_exp2f.exit ], [ false, %1133 ] + %1140 = phi i1 [ %149, %__nv_exp2f.exit ], [ false, %1133 ] + %1141 = phi i1 [ %155, %__nv_exp2f.exit ], [ false, %1133 ] + %1142 = phi i1 [ %161, %__nv_exp2f.exit ], [ false, %1133 ] + %1143 = phi i1 [ %36, %__nv_exp2f.exit ], [ %1135, %1133 ] + %1144 = phi i1 [ %125, %__nv_exp2f.exit ], [ %1136, %1133 ] + %1145 = phi i1 [ %131, %__nv_exp2f.exit ], [ %1137, %1133 ] + %1146 = phi i1 [ %137, %__nv_exp2f.exit ], [ %1138, %1133 ] + %1147 = phi i1 [ %143, %__nv_exp2f.exit ], [ %1139, %1133 ] + %1148 = phi i1 [ %149, %__nv_exp2f.exit ], [ %1140, %1133 ] + %1149 = phi i1 [ %155, %__nv_exp2f.exit ], [ %1141, %1133 ] + %1150 = phi i1 [ %161, %__nv_exp2f.exit ], [ %1142, %1133 ] + %1151 = phi i32 [ %246, %__nv_exp2f.exit ], [ %1956, %1133 ] + %1152 = phi i32 [ %228, %__nv_exp2f.exit ], [ %1151, %1133 ] + %1153 = phi i32 [ %23, %__nv_exp2f.exit ], [ %1152, %1133 ] + %1154 = phi i32 [ -1, %__nv_exp2f.exit ], [ %1158, %1133 ] + %1155 = phi i32 [ 2, %__nv_exp2f.exit ], [ %1954, %1133 ] + %1156 = add i32 %1154, 1, !dbg !60 + %1157 = icmp sgt i32 %1156, 2, !dbg !60 + %1158 = select i1 %1157, i32 0, i32 %1156, !dbg !60 + %1159 = sext i32 %1153 to i64, !dbg !61 + %1160 = getelementptr bfloat, ptr addrspace(1) %1101, i64 %1159, !dbg !61 + %1161 = sext i32 %1153 to i64, !dbg !61 + %1162 = getelementptr bfloat, ptr addrspace(1) %1102, i64 %1161, !dbg !61 + %1163 = sext i32 %1153 to i64, !dbg !61 + %1164 = getelementptr bfloat, ptr addrspace(1) %1103, i64 %1163, !dbg !61 + %1165 = sext i32 %1153 to i64, !dbg !61 + %1166 = getelementptr bfloat, ptr addrspace(1) %1104, i64 %1165, !dbg !61 + %1167 = sext i32 %1153 to i64, !dbg !61 + %1168 = getelementptr bfloat, ptr addrspace(1) %1105, i64 %1167, !dbg !61 + %1169 = sext i32 %1153 to i64, !dbg !61 + %1170 = getelementptr bfloat, ptr addrspace(1) %1106, i64 %1169, !dbg !61 + %1171 = sext i32 %1153 to i64, !dbg !61 + %1172 = getelementptr bfloat, ptr addrspace(1) %1107, i64 %1171, !dbg !61 + %1173 = sext i32 %1153 to i64, !dbg !61 + %1174 = getelementptr bfloat, ptr addrspace(1) %1108, i64 %1173, !dbg !61 + tail call void @llvm.nvvm.cp.async.wait.group(i32 2), !dbg !59 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !59 + %.idx104 = shl i32 %1158, 13, !dbg !59 + %1175 = getelementptr i8, ptr addrspace(3) @global_smem, i32 %.idx104, !dbg !59 + %1176 = getelementptr inbounds nuw i8, ptr addrspace(3) %1175, i32 %269, !dbg !62 + %1177 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %1176), !dbg !62 + %1178 = extractvalue { i32, i32, i32, i32 } %1177, 0, !dbg !62 + %1179 = bitcast i32 %1178 to <2 x bfloat>, !dbg !62 + %1180 = extractvalue { i32, i32, i32, i32 } %1177, 1, !dbg !62 + %1181 = bitcast i32 %1180 to <2 x bfloat>, !dbg !62 + %1182 = extractvalue { i32, i32, i32, i32 } %1177, 2, !dbg !62 + %1183 = bitcast i32 %1182 to <2 x bfloat>, !dbg !62 + %1184 = extractvalue { i32, i32, i32, i32 } %1177, 3, !dbg !62 + %1185 = bitcast i32 %1184 to <2 x bfloat>, !dbg !62 + %1186 = getelementptr inbounds nuw i8, ptr addrspace(3) %1176, i32 4096, !dbg !62 + %1187 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %1186), !dbg !62 + %1188 = extractvalue { i32, i32, i32, i32 } %1187, 0, !dbg !62 + %1189 = bitcast i32 %1188 to <2 x bfloat>, !dbg !62 + %1190 = extractvalue { i32, i32, i32, i32 } %1187, 1, !dbg !62 + %1191 = bitcast i32 %1190 to <2 x bfloat>, !dbg !62 + %1192 = extractvalue { i32, i32, i32, i32 } %1187, 2, !dbg !62 + %1193 = bitcast i32 %1192 to <2 x bfloat>, !dbg !62 + %1194 = extractvalue { i32, i32, i32, i32 } %1187, 3, !dbg !62 + %1195 = bitcast i32 %1194 to <2 x bfloat>, !dbg !62 + %1196 = getelementptr inbounds nuw i8, ptr addrspace(3) %1175, i32 %270, !dbg !62 + %1197 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %1196), !dbg !62 + %1198 = extractvalue { i32, i32, i32, i32 } %1197, 0, !dbg !62 + %1199 = bitcast i32 %1198 to <2 x bfloat>, !dbg !62 + %1200 = extractvalue { i32, i32, i32, i32 } %1197, 1, !dbg !62 + %1201 = bitcast i32 %1200 to <2 x bfloat>, !dbg !62 + %1202 = extractvalue { i32, i32, i32, i32 } %1197, 2, !dbg !62 + %1203 = bitcast i32 %1202 to <2 x bfloat>, !dbg !62 + %1204 = extractvalue { i32, i32, i32, i32 } %1197, 3, !dbg !62 + %1205 = bitcast i32 %1204 to <2 x bfloat>, !dbg !62 + %1206 = getelementptr inbounds nuw i8, ptr addrspace(3) %1196, i32 4096, !dbg !62 + %1207 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %1206), !dbg !62 + %1208 = extractvalue { i32, i32, i32, i32 } %1207, 0, !dbg !62 + %1209 = bitcast i32 %1208 to <2 x bfloat>, !dbg !62 + %1210 = extractvalue { i32, i32, i32, i32 } %1207, 1, !dbg !62 + %1211 = bitcast i32 %1210 to <2 x bfloat>, !dbg !62 + %1212 = extractvalue { i32, i32, i32, i32 } %1207, 2, !dbg !62 + %1213 = bitcast i32 %1212 to <2 x bfloat>, !dbg !62 + %1214 = extractvalue { i32, i32, i32, i32 } %1207, 3, !dbg !62 + %1215 = bitcast i32 %1214 to <2 x bfloat>, !dbg !62 + %1216 = getelementptr inbounds nuw i8, ptr addrspace(3) %1175, i32 %271, !dbg !62 + %1217 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %1216), !dbg !62 + %1218 = extractvalue { i32, i32, i32, i32 } %1217, 0, !dbg !62 + %1219 = bitcast i32 %1218 to <2 x bfloat>, !dbg !62 + %1220 = extractvalue { i32, i32, i32, i32 } %1217, 1, !dbg !62 + %1221 = bitcast i32 %1220 to <2 x bfloat>, !dbg !62 + %1222 = extractvalue { i32, i32, i32, i32 } %1217, 2, !dbg !62 + %1223 = bitcast i32 %1222 to <2 x bfloat>, !dbg !62 + %1224 = extractvalue { i32, i32, i32, i32 } %1217, 3, !dbg !62 + %1225 = bitcast i32 %1224 to <2 x bfloat>, !dbg !62 + %1226 = getelementptr inbounds nuw i8, ptr addrspace(3) %1216, i32 4096, !dbg !62 + %1227 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %1226), !dbg !62 + %1228 = extractvalue { i32, i32, i32, i32 } %1227, 0, !dbg !62 + %1229 = bitcast i32 %1228 to <2 x bfloat>, !dbg !62 + %1230 = extractvalue { i32, i32, i32, i32 } %1227, 1, !dbg !62 + %1231 = bitcast i32 %1230 to <2 x bfloat>, !dbg !62 + %1232 = extractvalue { i32, i32, i32, i32 } %1227, 2, !dbg !62 + %1233 = bitcast i32 %1232 to <2 x bfloat>, !dbg !62 + %1234 = extractvalue { i32, i32, i32, i32 } %1227, 3, !dbg !62 + %1235 = bitcast i32 %1234 to <2 x bfloat>, !dbg !62 + %1236 = getelementptr inbounds nuw i8, ptr addrspace(3) %1175, i32 %272, !dbg !62 + %1237 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %1236), !dbg !62 + %1238 = extractvalue { i32, i32, i32, i32 } %1237, 0, !dbg !62 + %1239 = bitcast i32 %1238 to <2 x bfloat>, !dbg !62 + %1240 = extractvalue { i32, i32, i32, i32 } %1237, 1, !dbg !62 + %1241 = bitcast i32 %1240 to <2 x bfloat>, !dbg !62 + %1242 = extractvalue { i32, i32, i32, i32 } %1237, 2, !dbg !62 + %1243 = bitcast i32 %1242 to <2 x bfloat>, !dbg !62 + %1244 = extractvalue { i32, i32, i32, i32 } %1237, 3, !dbg !62 + %1245 = bitcast i32 %1244 to <2 x bfloat>, !dbg !62 + %1246 = getelementptr inbounds nuw i8, ptr addrspace(3) %1236, i32 4096, !dbg !62 + %1247 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %1246), !dbg !62 + %1248 = extractvalue { i32, i32, i32, i32 } %1247, 0, !dbg !62 + %1249 = bitcast i32 %1248 to <2 x bfloat>, !dbg !62 + %1250 = extractvalue { i32, i32, i32, i32 } %1247, 1, !dbg !62 + %1251 = bitcast i32 %1250 to <2 x bfloat>, !dbg !62 + %1252 = extractvalue { i32, i32, i32, i32 } %1247, 2, !dbg !62 + %1253 = bitcast i32 %1252 to <2 x bfloat>, !dbg !62 + %1254 = extractvalue { i32, i32, i32, i32 } %1247, 3, !dbg !62 + %1255 = bitcast i32 %1254 to <2 x bfloat>, !dbg !62 + %1256 = fmul <2 x bfloat> %318, %1179, !dbg !28 + %1257 = fpext <2 x bfloat> %1256 to <2 x float>, !dbg !63 + %1258 = fmul <2 x float> %1079, %1257, !dbg !63 + %1259 = fptrunc <2 x float> %1258 to <2 x bfloat>, !dbg !62 + %1260 = fmul <2 x bfloat> %316, %1181, !dbg !28 + %1261 = fpext <2 x bfloat> %1260 to <2 x float>, !dbg !63 + %1262 = fmul <2 x float> %1080, %1261, !dbg !63 + %1263 = fptrunc <2 x float> %1262 to <2 x bfloat>, !dbg !62 + %1264 = fmul <2 x bfloat> %314, %1183, !dbg !28 + %1265 = fpext <2 x bfloat> %1264 to <2 x float>, !dbg !63 + %1266 = fmul <2 x float> %1081, %1265, !dbg !63 + %1267 = fptrunc <2 x float> %1266 to <2 x bfloat>, !dbg !62 + %1268 = fmul <2 x bfloat> %312, %1185, !dbg !28 + %1269 = fpext <2 x bfloat> %1268 to <2 x float>, !dbg !63 + %1270 = fmul <2 x float> %1082, %1269, !dbg !63 + %1271 = fptrunc <2 x float> %1270 to <2 x bfloat>, !dbg !62 + %1272 = fmul <2 x bfloat> %310, %1189, !dbg !28 + %1273 = fpext <2 x bfloat> %1272 to <2 x float>, !dbg !63 + %1274 = fmul <2 x float> %1084, %1273, !dbg !63 + %1275 = fptrunc <2 x float> %1274 to <2 x bfloat>, !dbg !62 + %1276 = fmul <2 x bfloat> %308, %1191, !dbg !28 + %1277 = fpext <2 x bfloat> %1276 to <2 x float>, !dbg !63 + %1278 = fmul <2 x float> %1086, %1277, !dbg !63 + %1279 = fptrunc <2 x float> %1278 to <2 x bfloat>, !dbg !62 + %1280 = fmul <2 x bfloat> %306, %1193, !dbg !28 + %1281 = fpext <2 x bfloat> %1280 to <2 x float>, !dbg !63 + %1282 = fmul <2 x float> %1088, %1281, !dbg !63 + %1283 = fptrunc <2 x float> %1282 to <2 x bfloat>, !dbg !62 + %1284 = fmul <2 x bfloat> %304, %1195, !dbg !28 + %1285 = fpext <2 x bfloat> %1284 to <2 x float>, !dbg !63 + %1286 = fmul <2 x float> %1090, %1285, !dbg !63 + %1287 = fptrunc <2 x float> %1286 to <2 x bfloat>, !dbg !62 + %1288 = fmul <2 x bfloat> %318, %1199, !dbg !28 + %1289 = fpext <2 x bfloat> %1288 to <2 x float>, !dbg !63 + %1290 = fmul <2 x float> %1079, %1289, !dbg !63 + %1291 = fptrunc <2 x float> %1290 to <2 x bfloat>, !dbg !62 + %1292 = fmul <2 x bfloat> %316, %1201, !dbg !28 + %1293 = fpext <2 x bfloat> %1292 to <2 x float>, !dbg !63 + %1294 = fmul <2 x float> %1080, %1293, !dbg !63 + %1295 = fptrunc <2 x float> %1294 to <2 x bfloat>, !dbg !62 + %1296 = fmul <2 x bfloat> %314, %1203, !dbg !28 + %1297 = fpext <2 x bfloat> %1296 to <2 x float>, !dbg !63 + %1298 = fmul <2 x float> %1081, %1297, !dbg !63 + %1299 = fptrunc <2 x float> %1298 to <2 x bfloat>, !dbg !62 + %1300 = fmul <2 x bfloat> %312, %1205, !dbg !28 + %1301 = fpext <2 x bfloat> %1300 to <2 x float>, !dbg !63 + %1302 = fmul <2 x float> %1082, %1301, !dbg !63 + %1303 = fptrunc <2 x float> %1302 to <2 x bfloat>, !dbg !62 + %1304 = fmul <2 x bfloat> %310, %1209, !dbg !28 + %1305 = fpext <2 x bfloat> %1304 to <2 x float>, !dbg !63 + %1306 = fmul <2 x float> %1084, %1305, !dbg !63 + %1307 = fptrunc <2 x float> %1306 to <2 x bfloat>, !dbg !62 + %1308 = fmul <2 x bfloat> %308, %1211, !dbg !28 + %1309 = fpext <2 x bfloat> %1308 to <2 x float>, !dbg !63 + %1310 = fmul <2 x float> %1086, %1309, !dbg !63 + %1311 = fptrunc <2 x float> %1310 to <2 x bfloat>, !dbg !62 + %1312 = fmul <2 x bfloat> %306, %1213, !dbg !28 + %1313 = fpext <2 x bfloat> %1312 to <2 x float>, !dbg !63 + %1314 = fmul <2 x float> %1088, %1313, !dbg !63 + %1315 = fptrunc <2 x float> %1314 to <2 x bfloat>, !dbg !62 + %1316 = fmul <2 x bfloat> %304, %1215, !dbg !28 + %1317 = fpext <2 x bfloat> %1316 to <2 x float>, !dbg !63 + %1318 = fmul <2 x float> %1090, %1317, !dbg !63 + %1319 = fptrunc <2 x float> %1318 to <2 x bfloat>, !dbg !62 + %1320 = fmul <2 x bfloat> %318, %1219, !dbg !28 + %1321 = fpext <2 x bfloat> %1320 to <2 x float>, !dbg !63 + %1322 = fmul <2 x float> %1079, %1321, !dbg !63 + %1323 = fptrunc <2 x float> %1322 to <2 x bfloat>, !dbg !62 + %1324 = fmul <2 x bfloat> %316, %1221, !dbg !28 + %1325 = fpext <2 x bfloat> %1324 to <2 x float>, !dbg !63 + %1326 = fmul <2 x float> %1080, %1325, !dbg !63 + %1327 = fptrunc <2 x float> %1326 to <2 x bfloat>, !dbg !62 + %1328 = fmul <2 x bfloat> %314, %1223, !dbg !28 + %1329 = fpext <2 x bfloat> %1328 to <2 x float>, !dbg !63 + %1330 = fmul <2 x float> %1081, %1329, !dbg !63 + %1331 = fptrunc <2 x float> %1330 to <2 x bfloat>, !dbg !62 + %1332 = fmul <2 x bfloat> %312, %1225, !dbg !28 + %1333 = fpext <2 x bfloat> %1332 to <2 x float>, !dbg !63 + %1334 = fmul <2 x float> %1082, %1333, !dbg !63 + %1335 = fptrunc <2 x float> %1334 to <2 x bfloat>, !dbg !62 + %1336 = fmul <2 x bfloat> %310, %1229, !dbg !28 + %1337 = fpext <2 x bfloat> %1336 to <2 x float>, !dbg !63 + %1338 = fmul <2 x float> %1084, %1337, !dbg !63 + %1339 = fptrunc <2 x float> %1338 to <2 x bfloat>, !dbg !62 + %1340 = fmul <2 x bfloat> %308, %1231, !dbg !28 + %1341 = fpext <2 x bfloat> %1340 to <2 x float>, !dbg !63 + %1342 = fmul <2 x float> %1086, %1341, !dbg !63 + %1343 = fptrunc <2 x float> %1342 to <2 x bfloat>, !dbg !62 + %1344 = fmul <2 x bfloat> %306, %1233, !dbg !28 + %1345 = fpext <2 x bfloat> %1344 to <2 x float>, !dbg !63 + %1346 = fmul <2 x float> %1088, %1345, !dbg !63 + %1347 = fptrunc <2 x float> %1346 to <2 x bfloat>, !dbg !62 + %1348 = fmul <2 x bfloat> %304, %1235, !dbg !28 + %1349 = fpext <2 x bfloat> %1348 to <2 x float>, !dbg !63 + %1350 = fmul <2 x float> %1090, %1349, !dbg !63 + %1351 = fptrunc <2 x float> %1350 to <2 x bfloat>, !dbg !62 + %1352 = fmul <2 x bfloat> %318, %1239, !dbg !28 + %1353 = fpext <2 x bfloat> %1352 to <2 x float>, !dbg !63 + %1354 = fmul <2 x float> %1079, %1353, !dbg !63 + %1355 = fptrunc <2 x float> %1354 to <2 x bfloat>, !dbg !62 + %1356 = fmul <2 x bfloat> %316, %1241, !dbg !28 + %1357 = fpext <2 x bfloat> %1356 to <2 x float>, !dbg !63 + %1358 = fmul <2 x float> %1080, %1357, !dbg !63 + %1359 = fptrunc <2 x float> %1358 to <2 x bfloat>, !dbg !62 + %1360 = fmul <2 x bfloat> %314, %1243, !dbg !28 + %1361 = fpext <2 x bfloat> %1360 to <2 x float>, !dbg !63 + %1362 = fmul <2 x float> %1081, %1361, !dbg !63 + %1363 = fptrunc <2 x float> %1362 to <2 x bfloat>, !dbg !62 + %1364 = fmul <2 x bfloat> %312, %1245, !dbg !28 + %1365 = fpext <2 x bfloat> %1364 to <2 x float>, !dbg !63 + %1366 = fmul <2 x float> %1082, %1365, !dbg !63 + %1367 = fptrunc <2 x float> %1366 to <2 x bfloat>, !dbg !62 + %1368 = fmul <2 x bfloat> %310, %1249, !dbg !28 + %1369 = fpext <2 x bfloat> %1368 to <2 x float>, !dbg !63 + %1370 = fmul <2 x float> %1084, %1369, !dbg !63 + %1371 = fptrunc <2 x float> %1370 to <2 x bfloat>, !dbg !62 + %1372 = fmul <2 x bfloat> %308, %1251, !dbg !28 + %1373 = fpext <2 x bfloat> %1372 to <2 x float>, !dbg !63 + %1374 = fmul <2 x float> %1086, %1373, !dbg !63 + %1375 = fptrunc <2 x float> %1374 to <2 x bfloat>, !dbg !62 + %1376 = fmul <2 x bfloat> %306, %1253, !dbg !28 + %1377 = fpext <2 x bfloat> %1376 to <2 x float>, !dbg !63 + %1378 = fmul <2 x float> %1088, %1377, !dbg !63 + %1379 = fptrunc <2 x float> %1378 to <2 x bfloat>, !dbg !62 + %1380 = fmul <2 x bfloat> %304, %1255, !dbg !28 + %1381 = fpext <2 x bfloat> %1380 to <2 x float>, !dbg !63 + %1382 = fmul <2 x float> %1090, %1381, !dbg !63 + %1383 = fptrunc <2 x float> %1382 to <2 x bfloat>, !dbg !62 + %1384 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %277), !dbg !32 + %1385 = extractvalue { i32, i32, i32, i32 } %1384, 0, !dbg !32 + %1386 = extractvalue { i32, i32, i32, i32 } %1384, 1, !dbg !32 + %1387 = extractvalue { i32, i32, i32, i32 } %1384, 2, !dbg !32 + %1388 = extractvalue { i32, i32, i32, i32 } %1384, 3, !dbg !32 + %1389 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %278), !dbg !32 + %1390 = extractvalue { i32, i32, i32, i32 } %1389, 0, !dbg !32 + %1391 = extractvalue { i32, i32, i32, i32 } %1389, 1, !dbg !32 + %1392 = extractvalue { i32, i32, i32, i32 } %1389, 2, !dbg !32 + %1393 = extractvalue { i32, i32, i32, i32 } %1389, 3, !dbg !32 + %1394 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %279), !dbg !32 + %1395 = extractvalue { i32, i32, i32, i32 } %1394, 0, !dbg !32 + %1396 = extractvalue { i32, i32, i32, i32 } %1394, 1, !dbg !32 + %1397 = extractvalue { i32, i32, i32, i32 } %1394, 2, !dbg !32 + %1398 = extractvalue { i32, i32, i32, i32 } %1394, 3, !dbg !32 + %1399 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %280), !dbg !32 + %1400 = extractvalue { i32, i32, i32, i32 } %1399, 0, !dbg !32 + %1401 = extractvalue { i32, i32, i32, i32 } %1399, 1, !dbg !32 + %1402 = extractvalue { i32, i32, i32, i32 } %1399, 2, !dbg !32 + %1403 = extractvalue { i32, i32, i32, i32 } %1399, 3, !dbg !32 + %1404 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %282), !dbg !32 + %1405 = extractvalue { i32, i32, i32, i32 } %1404, 0, !dbg !32 + %1406 = extractvalue { i32, i32, i32, i32 } %1404, 1, !dbg !32 + %1407 = extractvalue { i32, i32, i32, i32 } %1404, 2, !dbg !32 + %1408 = extractvalue { i32, i32, i32, i32 } %1404, 3, !dbg !32 + %1409 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %283), !dbg !32 + %1410 = extractvalue { i32, i32, i32, i32 } %1409, 0, !dbg !32 + %1411 = extractvalue { i32, i32, i32, i32 } %1409, 1, !dbg !32 + %1412 = extractvalue { i32, i32, i32, i32 } %1409, 2, !dbg !32 + %1413 = extractvalue { i32, i32, i32, i32 } %1409, 3, !dbg !32 + %1414 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %284), !dbg !32 + %1415 = extractvalue { i32, i32, i32, i32 } %1414, 0, !dbg !32 + %1416 = extractvalue { i32, i32, i32, i32 } %1414, 1, !dbg !32 + %1417 = extractvalue { i32, i32, i32, i32 } %1414, 2, !dbg !32 + %1418 = extractvalue { i32, i32, i32, i32 } %1414, 3, !dbg !32 + %1419 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %285), !dbg !32 + %1420 = extractvalue { i32, i32, i32, i32 } %1419, 0, !dbg !32 + %1421 = extractvalue { i32, i32, i32, i32 } %1419, 1, !dbg !32 + %1422 = extractvalue { i32, i32, i32, i32 } %1419, 2, !dbg !32 + %1423 = extractvalue { i32, i32, i32, i32 } %1419, 3, !dbg !32 + %1424 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %287), !dbg !32 + %1425 = extractvalue { i32, i32, i32, i32 } %1424, 0, !dbg !32 + %1426 = extractvalue { i32, i32, i32, i32 } %1424, 1, !dbg !32 + %1427 = extractvalue { i32, i32, i32, i32 } %1424, 2, !dbg !32 + %1428 = extractvalue { i32, i32, i32, i32 } %1424, 3, !dbg !32 + %1429 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %288), !dbg !32 + %1430 = extractvalue { i32, i32, i32, i32 } %1429, 0, !dbg !32 + %1431 = extractvalue { i32, i32, i32, i32 } %1429, 1, !dbg !32 + %1432 = extractvalue { i32, i32, i32, i32 } %1429, 2, !dbg !32 + %1433 = extractvalue { i32, i32, i32, i32 } %1429, 3, !dbg !32 + %1434 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %289), !dbg !32 + %1435 = extractvalue { i32, i32, i32, i32 } %1434, 0, !dbg !32 + %1436 = extractvalue { i32, i32, i32, i32 } %1434, 1, !dbg !32 + %1437 = extractvalue { i32, i32, i32, i32 } %1434, 2, !dbg !32 + %1438 = extractvalue { i32, i32, i32, i32 } %1434, 3, !dbg !32 + %1439 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %290), !dbg !32 + %1440 = extractvalue { i32, i32, i32, i32 } %1439, 0, !dbg !32 + %1441 = extractvalue { i32, i32, i32, i32 } %1439, 1, !dbg !32 + %1442 = extractvalue { i32, i32, i32, i32 } %1439, 2, !dbg !32 + %1443 = extractvalue { i32, i32, i32, i32 } %1439, 3, !dbg !32 + %1444 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %292), !dbg !32 + %1445 = extractvalue { i32, i32, i32, i32 } %1444, 0, !dbg !32 + %1446 = extractvalue { i32, i32, i32, i32 } %1444, 1, !dbg !32 + %1447 = extractvalue { i32, i32, i32, i32 } %1444, 2, !dbg !32 + %1448 = extractvalue { i32, i32, i32, i32 } %1444, 3, !dbg !32 + %1449 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %293), !dbg !32 + %1450 = extractvalue { i32, i32, i32, i32 } %1449, 0, !dbg !32 + %1451 = extractvalue { i32, i32, i32, i32 } %1449, 1, !dbg !32 + %1452 = extractvalue { i32, i32, i32, i32 } %1449, 2, !dbg !32 + %1453 = extractvalue { i32, i32, i32, i32 } %1449, 3, !dbg !32 + %1454 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %294), !dbg !32 + %1455 = extractvalue { i32, i32, i32, i32 } %1454, 0, !dbg !32 + %1456 = extractvalue { i32, i32, i32, i32 } %1454, 1, !dbg !32 + %1457 = extractvalue { i32, i32, i32, i32 } %1454, 2, !dbg !32 + %1458 = extractvalue { i32, i32, i32, i32 } %1454, 3, !dbg !32 + %1459 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %295), !dbg !32 + %1460 = extractvalue { i32, i32, i32, i32 } %1459, 0, !dbg !32 + %1461 = extractvalue { i32, i32, i32, i32 } %1459, 1, !dbg !32 + %1462 = extractvalue { i32, i32, i32, i32 } %1459, 2, !dbg !32 + %1463 = extractvalue { i32, i32, i32, i32 } %1459, 3, !dbg !32 + %1464 = bitcast <2 x bfloat> %1259 to i32, !dbg !64 + %1465 = bitcast <2 x bfloat> %1263 to i32, !dbg !64 + %1466 = bitcast <2 x bfloat> %1267 to i32, !dbg !64 + %1467 = bitcast <2 x bfloat> %1271 to i32, !dbg !64 + %1468 = bitcast <2 x bfloat> %1275 to i32, !dbg !64 + %1469 = bitcast <2 x bfloat> %1279 to i32, !dbg !64 + %1470 = bitcast <2 x bfloat> %1283 to i32, !dbg !64 + %1471 = bitcast <2 x bfloat> %1287 to i32, !dbg !64 + %1472 = bitcast <2 x bfloat> %1291 to i32, !dbg !64 + %1473 = bitcast <2 x bfloat> %1295 to i32, !dbg !64 + %1474 = bitcast <2 x bfloat> %1299 to i32, !dbg !64 + %1475 = bitcast <2 x bfloat> %1303 to i32, !dbg !64 + %1476 = bitcast <2 x bfloat> %1307 to i32, !dbg !64 + %1477 = bitcast <2 x bfloat> %1311 to i32, !dbg !64 + %1478 = bitcast <2 x bfloat> %1315 to i32, !dbg !64 + %1479 = bitcast <2 x bfloat> %1319 to i32, !dbg !64 + %1480 = bitcast <2 x bfloat> %1323 to i32, !dbg !64 + %1481 = bitcast <2 x bfloat> %1327 to i32, !dbg !64 + %1482 = bitcast <2 x bfloat> %1331 to i32, !dbg !64 + %1483 = bitcast <2 x bfloat> %1335 to i32, !dbg !64 + %1484 = bitcast <2 x bfloat> %1339 to i32, !dbg !64 + %1485 = bitcast <2 x bfloat> %1343 to i32, !dbg !64 + %1486 = bitcast <2 x bfloat> %1347 to i32, !dbg !64 + %1487 = bitcast <2 x bfloat> %1351 to i32, !dbg !64 + %1488 = bitcast <2 x bfloat> %1355 to i32, !dbg !64 + %1489 = bitcast <2 x bfloat> %1359 to i32, !dbg !64 + %1490 = bitcast <2 x bfloat> %1363 to i32, !dbg !64 + %1491 = bitcast <2 x bfloat> %1367 to i32, !dbg !64 + %1492 = bitcast <2 x bfloat> %1371 to i32, !dbg !64 + %1493 = bitcast <2 x bfloat> %1375 to i32, !dbg !64 + %1494 = bitcast <2 x bfloat> %1379 to i32, !dbg !64 + %1495 = bitcast <2 x bfloat> %1383 to i32, !dbg !64 + %1496 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1385, i32 %1386, i32 %1387, i32 %1388, i32 %1464, i32 %1465) #3, !dbg !64 + %1497 = extractvalue { float, float, float, float } %1496, 0, !dbg !64 + %1498 = extractvalue { float, float, float, float } %1496, 1, !dbg !64 + %1499 = extractvalue { float, float, float, float } %1496, 2, !dbg !64 + %1500 = extractvalue { float, float, float, float } %1496, 3, !dbg !64 + %1501 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1385, i32 %1386, i32 %1387, i32 %1388, i32 %1472, i32 %1473) #3, !dbg !64 + %1502 = extractvalue { float, float, float, float } %1501, 0, !dbg !64 + %1503 = extractvalue { float, float, float, float } %1501, 1, !dbg !64 + %1504 = extractvalue { float, float, float, float } %1501, 2, !dbg !64 + %1505 = extractvalue { float, float, float, float } %1501, 3, !dbg !64 + %1506 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1385, i32 %1386, i32 %1387, i32 %1388, i32 %1480, i32 %1481) #3, !dbg !64 + %1507 = extractvalue { float, float, float, float } %1506, 0, !dbg !64 + %1508 = extractvalue { float, float, float, float } %1506, 1, !dbg !64 + %1509 = extractvalue { float, float, float, float } %1506, 2, !dbg !64 + %1510 = extractvalue { float, float, float, float } %1506, 3, !dbg !64 + %1511 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1385, i32 %1386, i32 %1387, i32 %1388, i32 %1488, i32 %1489) #3, !dbg !64 + %1512 = extractvalue { float, float, float, float } %1511, 0, !dbg !64 + %1513 = extractvalue { float, float, float, float } %1511, 1, !dbg !64 + %1514 = extractvalue { float, float, float, float } %1511, 2, !dbg !64 + %1515 = extractvalue { float, float, float, float } %1511, 3, !dbg !64 + %1516 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1390, i32 %1391, i32 %1392, i32 %1393, i32 %1464, i32 %1465) #3, !dbg !64 + %1517 = extractvalue { float, float, float, float } %1516, 0, !dbg !64 + %1518 = extractvalue { float, float, float, float } %1516, 1, !dbg !64 + %1519 = extractvalue { float, float, float, float } %1516, 2, !dbg !64 + %1520 = extractvalue { float, float, float, float } %1516, 3, !dbg !64 + %1521 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1390, i32 %1391, i32 %1392, i32 %1393, i32 %1472, i32 %1473) #3, !dbg !64 + %1522 = extractvalue { float, float, float, float } %1521, 0, !dbg !64 + %1523 = extractvalue { float, float, float, float } %1521, 1, !dbg !64 + %1524 = extractvalue { float, float, float, float } %1521, 2, !dbg !64 + %1525 = extractvalue { float, float, float, float } %1521, 3, !dbg !64 + %1526 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1390, i32 %1391, i32 %1392, i32 %1393, i32 %1480, i32 %1481) #3, !dbg !64 + %1527 = extractvalue { float, float, float, float } %1526, 0, !dbg !64 + %1528 = extractvalue { float, float, float, float } %1526, 1, !dbg !64 + %1529 = extractvalue { float, float, float, float } %1526, 2, !dbg !64 + %1530 = extractvalue { float, float, float, float } %1526, 3, !dbg !64 + %1531 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1390, i32 %1391, i32 %1392, i32 %1393, i32 %1488, i32 %1489) #3, !dbg !64 + %1532 = extractvalue { float, float, float, float } %1531, 0, !dbg !64 + %1533 = extractvalue { float, float, float, float } %1531, 1, !dbg !64 + %1534 = extractvalue { float, float, float, float } %1531, 2, !dbg !64 + %1535 = extractvalue { float, float, float, float } %1531, 3, !dbg !64 + %1536 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1395, i32 %1396, i32 %1397, i32 %1398, i32 %1464, i32 %1465) #3, !dbg !64 + %1537 = extractvalue { float, float, float, float } %1536, 0, !dbg !64 + %1538 = extractvalue { float, float, float, float } %1536, 1, !dbg !64 + %1539 = extractvalue { float, float, float, float } %1536, 2, !dbg !64 + %1540 = extractvalue { float, float, float, float } %1536, 3, !dbg !64 + %1541 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1395, i32 %1396, i32 %1397, i32 %1398, i32 %1472, i32 %1473) #3, !dbg !64 + %1542 = extractvalue { float, float, float, float } %1541, 0, !dbg !64 + %1543 = extractvalue { float, float, float, float } %1541, 1, !dbg !64 + %1544 = extractvalue { float, float, float, float } %1541, 2, !dbg !64 + %1545 = extractvalue { float, float, float, float } %1541, 3, !dbg !64 + %1546 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1395, i32 %1396, i32 %1397, i32 %1398, i32 %1480, i32 %1481) #3, !dbg !64 + %1547 = extractvalue { float, float, float, float } %1546, 0, !dbg !64 + %1548 = extractvalue { float, float, float, float } %1546, 1, !dbg !64 + %1549 = extractvalue { float, float, float, float } %1546, 2, !dbg !64 + %1550 = extractvalue { float, float, float, float } %1546, 3, !dbg !64 + %1551 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1395, i32 %1396, i32 %1397, i32 %1398, i32 %1488, i32 %1489) #3, !dbg !64 + %1552 = extractvalue { float, float, float, float } %1551, 0, !dbg !64 + %1553 = extractvalue { float, float, float, float } %1551, 1, !dbg !64 + %1554 = extractvalue { float, float, float, float } %1551, 2, !dbg !64 + %1555 = extractvalue { float, float, float, float } %1551, 3, !dbg !64 + %1556 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1400, i32 %1401, i32 %1402, i32 %1403, i32 %1464, i32 %1465) #3, !dbg !64 + %1557 = extractvalue { float, float, float, float } %1556, 0, !dbg !64 + %1558 = extractvalue { float, float, float, float } %1556, 1, !dbg !64 + %1559 = extractvalue { float, float, float, float } %1556, 2, !dbg !64 + %1560 = extractvalue { float, float, float, float } %1556, 3, !dbg !64 + %1561 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1400, i32 %1401, i32 %1402, i32 %1403, i32 %1472, i32 %1473) #3, !dbg !64 + %1562 = extractvalue { float, float, float, float } %1561, 0, !dbg !64 + %1563 = extractvalue { float, float, float, float } %1561, 1, !dbg !64 + %1564 = extractvalue { float, float, float, float } %1561, 2, !dbg !64 + %1565 = extractvalue { float, float, float, float } %1561, 3, !dbg !64 + %1566 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1400, i32 %1401, i32 %1402, i32 %1403, i32 %1480, i32 %1481) #3, !dbg !64 + %1567 = extractvalue { float, float, float, float } %1566, 0, !dbg !64 + %1568 = extractvalue { float, float, float, float } %1566, 1, !dbg !64 + %1569 = extractvalue { float, float, float, float } %1566, 2, !dbg !64 + %1570 = extractvalue { float, float, float, float } %1566, 3, !dbg !64 + %1571 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1400, i32 %1401, i32 %1402, i32 %1403, i32 %1488, i32 %1489) #3, !dbg !64 + %1572 = extractvalue { float, float, float, float } %1571, 0, !dbg !64 + %1573 = extractvalue { float, float, float, float } %1571, 1, !dbg !64 + %1574 = extractvalue { float, float, float, float } %1571, 2, !dbg !64 + %1575 = extractvalue { float, float, float, float } %1571, 3, !dbg !64 + %1576 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1497, float %1498, float %1499, float %1500, i32 %1405, i32 %1406, i32 %1407, i32 %1408, i32 %1466, i32 %1467) #3, !dbg !64 + %1577 = extractvalue { float, float, float, float } %1576, 0, !dbg !64 + %1578 = extractvalue { float, float, float, float } %1576, 1, !dbg !64 + %1579 = extractvalue { float, float, float, float } %1576, 2, !dbg !64 + %1580 = extractvalue { float, float, float, float } %1576, 3, !dbg !64 + %1581 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1502, float %1503, float %1504, float %1505, i32 %1405, i32 %1406, i32 %1407, i32 %1408, i32 %1474, i32 %1475) #3, !dbg !64 + %1582 = extractvalue { float, float, float, float } %1581, 0, !dbg !64 + %1583 = extractvalue { float, float, float, float } %1581, 1, !dbg !64 + %1584 = extractvalue { float, float, float, float } %1581, 2, !dbg !64 + %1585 = extractvalue { float, float, float, float } %1581, 3, !dbg !64 + %1586 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1507, float %1508, float %1509, float %1510, i32 %1405, i32 %1406, i32 %1407, i32 %1408, i32 %1482, i32 %1483) #3, !dbg !64 + %1587 = extractvalue { float, float, float, float } %1586, 0, !dbg !64 + %1588 = extractvalue { float, float, float, float } %1586, 1, !dbg !64 + %1589 = extractvalue { float, float, float, float } %1586, 2, !dbg !64 + %1590 = extractvalue { float, float, float, float } %1586, 3, !dbg !64 + %1591 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1512, float %1513, float %1514, float %1515, i32 %1405, i32 %1406, i32 %1407, i32 %1408, i32 %1490, i32 %1491) #3, !dbg !64 + %1592 = extractvalue { float, float, float, float } %1591, 0, !dbg !64 + %1593 = extractvalue { float, float, float, float } %1591, 1, !dbg !64 + %1594 = extractvalue { float, float, float, float } %1591, 2, !dbg !64 + %1595 = extractvalue { float, float, float, float } %1591, 3, !dbg !64 + %1596 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1517, float %1518, float %1519, float %1520, i32 %1410, i32 %1411, i32 %1412, i32 %1413, i32 %1466, i32 %1467) #3, !dbg !64 + %1597 = extractvalue { float, float, float, float } %1596, 0, !dbg !64 + %1598 = extractvalue { float, float, float, float } %1596, 1, !dbg !64 + %1599 = extractvalue { float, float, float, float } %1596, 2, !dbg !64 + %1600 = extractvalue { float, float, float, float } %1596, 3, !dbg !64 + %1601 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1522, float %1523, float %1524, float %1525, i32 %1410, i32 %1411, i32 %1412, i32 %1413, i32 %1474, i32 %1475) #3, !dbg !64 + %1602 = extractvalue { float, float, float, float } %1601, 0, !dbg !64 + %1603 = extractvalue { float, float, float, float } %1601, 1, !dbg !64 + %1604 = extractvalue { float, float, float, float } %1601, 2, !dbg !64 + %1605 = extractvalue { float, float, float, float } %1601, 3, !dbg !64 + %1606 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1527, float %1528, float %1529, float %1530, i32 %1410, i32 %1411, i32 %1412, i32 %1413, i32 %1482, i32 %1483) #3, !dbg !64 + %1607 = extractvalue { float, float, float, float } %1606, 0, !dbg !64 + %1608 = extractvalue { float, float, float, float } %1606, 1, !dbg !64 + %1609 = extractvalue { float, float, float, float } %1606, 2, !dbg !64 + %1610 = extractvalue { float, float, float, float } %1606, 3, !dbg !64 + %1611 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1532, float %1533, float %1534, float %1535, i32 %1410, i32 %1411, i32 %1412, i32 %1413, i32 %1490, i32 %1491) #3, !dbg !64 + %1612 = extractvalue { float, float, float, float } %1611, 0, !dbg !64 + %1613 = extractvalue { float, float, float, float } %1611, 1, !dbg !64 + %1614 = extractvalue { float, float, float, float } %1611, 2, !dbg !64 + %1615 = extractvalue { float, float, float, float } %1611, 3, !dbg !64 + %1616 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1537, float %1538, float %1539, float %1540, i32 %1415, i32 %1416, i32 %1417, i32 %1418, i32 %1466, i32 %1467) #3, !dbg !64 + %1617 = extractvalue { float, float, float, float } %1616, 0, !dbg !64 + %1618 = extractvalue { float, float, float, float } %1616, 1, !dbg !64 + %1619 = extractvalue { float, float, float, float } %1616, 2, !dbg !64 + %1620 = extractvalue { float, float, float, float } %1616, 3, !dbg !64 + %1621 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1542, float %1543, float %1544, float %1545, i32 %1415, i32 %1416, i32 %1417, i32 %1418, i32 %1474, i32 %1475) #3, !dbg !64 + %1622 = extractvalue { float, float, float, float } %1621, 0, !dbg !64 + %1623 = extractvalue { float, float, float, float } %1621, 1, !dbg !64 + %1624 = extractvalue { float, float, float, float } %1621, 2, !dbg !64 + %1625 = extractvalue { float, float, float, float } %1621, 3, !dbg !64 + %1626 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1547, float %1548, float %1549, float %1550, i32 %1415, i32 %1416, i32 %1417, i32 %1418, i32 %1482, i32 %1483) #3, !dbg !64 + %1627 = extractvalue { float, float, float, float } %1626, 0, !dbg !64 + %1628 = extractvalue { float, float, float, float } %1626, 1, !dbg !64 + %1629 = extractvalue { float, float, float, float } %1626, 2, !dbg !64 + %1630 = extractvalue { float, float, float, float } %1626, 3, !dbg !64 + %1631 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1552, float %1553, float %1554, float %1555, i32 %1415, i32 %1416, i32 %1417, i32 %1418, i32 %1490, i32 %1491) #3, !dbg !64 + %1632 = extractvalue { float, float, float, float } %1631, 0, !dbg !64 + %1633 = extractvalue { float, float, float, float } %1631, 1, !dbg !64 + %1634 = extractvalue { float, float, float, float } %1631, 2, !dbg !64 + %1635 = extractvalue { float, float, float, float } %1631, 3, !dbg !64 + %1636 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1557, float %1558, float %1559, float %1560, i32 %1420, i32 %1421, i32 %1422, i32 %1423, i32 %1466, i32 %1467) #3, !dbg !64 + %1637 = extractvalue { float, float, float, float } %1636, 0, !dbg !64 + %1638 = extractvalue { float, float, float, float } %1636, 1, !dbg !64 + %1639 = extractvalue { float, float, float, float } %1636, 2, !dbg !64 + %1640 = extractvalue { float, float, float, float } %1636, 3, !dbg !64 + %1641 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1562, float %1563, float %1564, float %1565, i32 %1420, i32 %1421, i32 %1422, i32 %1423, i32 %1474, i32 %1475) #3, !dbg !64 + %1642 = extractvalue { float, float, float, float } %1641, 0, !dbg !64 + %1643 = extractvalue { float, float, float, float } %1641, 1, !dbg !64 + %1644 = extractvalue { float, float, float, float } %1641, 2, !dbg !64 + %1645 = extractvalue { float, float, float, float } %1641, 3, !dbg !64 + %1646 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1567, float %1568, float %1569, float %1570, i32 %1420, i32 %1421, i32 %1422, i32 %1423, i32 %1482, i32 %1483) #3, !dbg !64 + %1647 = extractvalue { float, float, float, float } %1646, 0, !dbg !64 + %1648 = extractvalue { float, float, float, float } %1646, 1, !dbg !64 + %1649 = extractvalue { float, float, float, float } %1646, 2, !dbg !64 + %1650 = extractvalue { float, float, float, float } %1646, 3, !dbg !64 + %1651 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1572, float %1573, float %1574, float %1575, i32 %1420, i32 %1421, i32 %1422, i32 %1423, i32 %1490, i32 %1491) #3, !dbg !64 + %1652 = extractvalue { float, float, float, float } %1651, 0, !dbg !64 + %1653 = extractvalue { float, float, float, float } %1651, 1, !dbg !64 + %1654 = extractvalue { float, float, float, float } %1651, 2, !dbg !64 + %1655 = extractvalue { float, float, float, float } %1651, 3, !dbg !64 + %1656 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1577, float %1578, float %1579, float %1580, i32 %1425, i32 %1426, i32 %1427, i32 %1428, i32 %1468, i32 %1469) #3, !dbg !64 + %1657 = extractvalue { float, float, float, float } %1656, 0, !dbg !64 + %1658 = extractvalue { float, float, float, float } %1656, 1, !dbg !64 + %1659 = extractvalue { float, float, float, float } %1656, 2, !dbg !64 + %1660 = extractvalue { float, float, float, float } %1656, 3, !dbg !64 + %1661 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1582, float %1583, float %1584, float %1585, i32 %1425, i32 %1426, i32 %1427, i32 %1428, i32 %1476, i32 %1477) #3, !dbg !64 + %1662 = extractvalue { float, float, float, float } %1661, 0, !dbg !64 + %1663 = extractvalue { float, float, float, float } %1661, 1, !dbg !64 + %1664 = extractvalue { float, float, float, float } %1661, 2, !dbg !64 + %1665 = extractvalue { float, float, float, float } %1661, 3, !dbg !64 + %1666 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1587, float %1588, float %1589, float %1590, i32 %1425, i32 %1426, i32 %1427, i32 %1428, i32 %1484, i32 %1485) #3, !dbg !64 + %1667 = extractvalue { float, float, float, float } %1666, 0, !dbg !64 + %1668 = extractvalue { float, float, float, float } %1666, 1, !dbg !64 + %1669 = extractvalue { float, float, float, float } %1666, 2, !dbg !64 + %1670 = extractvalue { float, float, float, float } %1666, 3, !dbg !64 + %1671 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1592, float %1593, float %1594, float %1595, i32 %1425, i32 %1426, i32 %1427, i32 %1428, i32 %1492, i32 %1493) #3, !dbg !64 + %1672 = extractvalue { float, float, float, float } %1671, 0, !dbg !64 + %1673 = extractvalue { float, float, float, float } %1671, 1, !dbg !64 + %1674 = extractvalue { float, float, float, float } %1671, 2, !dbg !64 + %1675 = extractvalue { float, float, float, float } %1671, 3, !dbg !64 + %1676 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1597, float %1598, float %1599, float %1600, i32 %1430, i32 %1431, i32 %1432, i32 %1433, i32 %1468, i32 %1469) #3, !dbg !64 + %1677 = extractvalue { float, float, float, float } %1676, 0, !dbg !64 + %1678 = extractvalue { float, float, float, float } %1676, 1, !dbg !64 + %1679 = extractvalue { float, float, float, float } %1676, 2, !dbg !64 + %1680 = extractvalue { float, float, float, float } %1676, 3, !dbg !64 + %1681 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1602, float %1603, float %1604, float %1605, i32 %1430, i32 %1431, i32 %1432, i32 %1433, i32 %1476, i32 %1477) #3, !dbg !64 + %1682 = extractvalue { float, float, float, float } %1681, 0, !dbg !64 + %1683 = extractvalue { float, float, float, float } %1681, 1, !dbg !64 + %1684 = extractvalue { float, float, float, float } %1681, 2, !dbg !64 + %1685 = extractvalue { float, float, float, float } %1681, 3, !dbg !64 + %1686 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1607, float %1608, float %1609, float %1610, i32 %1430, i32 %1431, i32 %1432, i32 %1433, i32 %1484, i32 %1485) #3, !dbg !64 + %1687 = extractvalue { float, float, float, float } %1686, 0, !dbg !64 + %1688 = extractvalue { float, float, float, float } %1686, 1, !dbg !64 + %1689 = extractvalue { float, float, float, float } %1686, 2, !dbg !64 + %1690 = extractvalue { float, float, float, float } %1686, 3, !dbg !64 + %1691 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1612, float %1613, float %1614, float %1615, i32 %1430, i32 %1431, i32 %1432, i32 %1433, i32 %1492, i32 %1493) #3, !dbg !64 + %1692 = extractvalue { float, float, float, float } %1691, 0, !dbg !64 + %1693 = extractvalue { float, float, float, float } %1691, 1, !dbg !64 + %1694 = extractvalue { float, float, float, float } %1691, 2, !dbg !64 + %1695 = extractvalue { float, float, float, float } %1691, 3, !dbg !64 + %1696 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1617, float %1618, float %1619, float %1620, i32 %1435, i32 %1436, i32 %1437, i32 %1438, i32 %1468, i32 %1469) #3, !dbg !64 + %1697 = extractvalue { float, float, float, float } %1696, 0, !dbg !64 + %1698 = extractvalue { float, float, float, float } %1696, 1, !dbg !64 + %1699 = extractvalue { float, float, float, float } %1696, 2, !dbg !64 + %1700 = extractvalue { float, float, float, float } %1696, 3, !dbg !64 + %1701 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1622, float %1623, float %1624, float %1625, i32 %1435, i32 %1436, i32 %1437, i32 %1438, i32 %1476, i32 %1477) #3, !dbg !64 + %1702 = extractvalue { float, float, float, float } %1701, 0, !dbg !64 + %1703 = extractvalue { float, float, float, float } %1701, 1, !dbg !64 + %1704 = extractvalue { float, float, float, float } %1701, 2, !dbg !64 + %1705 = extractvalue { float, float, float, float } %1701, 3, !dbg !64 + %1706 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1627, float %1628, float %1629, float %1630, i32 %1435, i32 %1436, i32 %1437, i32 %1438, i32 %1484, i32 %1485) #3, !dbg !64 + %1707 = extractvalue { float, float, float, float } %1706, 0, !dbg !64 + %1708 = extractvalue { float, float, float, float } %1706, 1, !dbg !64 + %1709 = extractvalue { float, float, float, float } %1706, 2, !dbg !64 + %1710 = extractvalue { float, float, float, float } %1706, 3, !dbg !64 + %1711 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1632, float %1633, float %1634, float %1635, i32 %1435, i32 %1436, i32 %1437, i32 %1438, i32 %1492, i32 %1493) #3, !dbg !64 + %1712 = extractvalue { float, float, float, float } %1711, 0, !dbg !64 + %1713 = extractvalue { float, float, float, float } %1711, 1, !dbg !64 + %1714 = extractvalue { float, float, float, float } %1711, 2, !dbg !64 + %1715 = extractvalue { float, float, float, float } %1711, 3, !dbg !64 + %1716 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1637, float %1638, float %1639, float %1640, i32 %1440, i32 %1441, i32 %1442, i32 %1443, i32 %1468, i32 %1469) #3, !dbg !64 + %1717 = extractvalue { float, float, float, float } %1716, 0, !dbg !64 + %1718 = extractvalue { float, float, float, float } %1716, 1, !dbg !64 + %1719 = extractvalue { float, float, float, float } %1716, 2, !dbg !64 + %1720 = extractvalue { float, float, float, float } %1716, 3, !dbg !64 + %1721 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1642, float %1643, float %1644, float %1645, i32 %1440, i32 %1441, i32 %1442, i32 %1443, i32 %1476, i32 %1477) #3, !dbg !64 + %1722 = extractvalue { float, float, float, float } %1721, 0, !dbg !64 + %1723 = extractvalue { float, float, float, float } %1721, 1, !dbg !64 + %1724 = extractvalue { float, float, float, float } %1721, 2, !dbg !64 + %1725 = extractvalue { float, float, float, float } %1721, 3, !dbg !64 + %1726 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1647, float %1648, float %1649, float %1650, i32 %1440, i32 %1441, i32 %1442, i32 %1443, i32 %1484, i32 %1485) #3, !dbg !64 + %1727 = extractvalue { float, float, float, float } %1726, 0, !dbg !64 + %1728 = extractvalue { float, float, float, float } %1726, 1, !dbg !64 + %1729 = extractvalue { float, float, float, float } %1726, 2, !dbg !64 + %1730 = extractvalue { float, float, float, float } %1726, 3, !dbg !64 + %1731 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1652, float %1653, float %1654, float %1655, i32 %1440, i32 %1441, i32 %1442, i32 %1443, i32 %1492, i32 %1493) #3, !dbg !64 + %1732 = extractvalue { float, float, float, float } %1731, 0, !dbg !64 + %1733 = extractvalue { float, float, float, float } %1731, 1, !dbg !64 + %1734 = extractvalue { float, float, float, float } %1731, 2, !dbg !64 + %1735 = extractvalue { float, float, float, float } %1731, 3, !dbg !64 + %1736 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1657, float %1658, float %1659, float %1660, i32 %1445, i32 %1446, i32 %1447, i32 %1448, i32 %1470, i32 %1471) #3, !dbg !64 + %1737 = extractvalue { float, float, float, float } %1736, 0, !dbg !64 + %1738 = extractvalue { float, float, float, float } %1736, 1, !dbg !64 + %1739 = extractvalue { float, float, float, float } %1736, 2, !dbg !64 + %1740 = extractvalue { float, float, float, float } %1736, 3, !dbg !64 + %1741 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1662, float %1663, float %1664, float %1665, i32 %1445, i32 %1446, i32 %1447, i32 %1448, i32 %1478, i32 %1479) #3, !dbg !64 + %1742 = extractvalue { float, float, float, float } %1741, 0, !dbg !64 + %1743 = extractvalue { float, float, float, float } %1741, 1, !dbg !64 + %1744 = extractvalue { float, float, float, float } %1741, 2, !dbg !64 + %1745 = extractvalue { float, float, float, float } %1741, 3, !dbg !64 + %1746 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1667, float %1668, float %1669, float %1670, i32 %1445, i32 %1446, i32 %1447, i32 %1448, i32 %1486, i32 %1487) #3, !dbg !64 + %1747 = extractvalue { float, float, float, float } %1746, 0, !dbg !64 + %1748 = extractvalue { float, float, float, float } %1746, 1, !dbg !64 + %1749 = extractvalue { float, float, float, float } %1746, 2, !dbg !64 + %1750 = extractvalue { float, float, float, float } %1746, 3, !dbg !64 + %1751 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1672, float %1673, float %1674, float %1675, i32 %1445, i32 %1446, i32 %1447, i32 %1448, i32 %1494, i32 %1495) #3, !dbg !64 + %1752 = extractvalue { float, float, float, float } %1751, 0, !dbg !64 + %1753 = extractvalue { float, float, float, float } %1751, 1, !dbg !64 + %1754 = extractvalue { float, float, float, float } %1751, 2, !dbg !64 + %1755 = extractvalue { float, float, float, float } %1751, 3, !dbg !64 + %1756 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1677, float %1678, float %1679, float %1680, i32 %1450, i32 %1451, i32 %1452, i32 %1453, i32 %1470, i32 %1471) #3, !dbg !64 + %1757 = extractvalue { float, float, float, float } %1756, 0, !dbg !64 + %1758 = extractvalue { float, float, float, float } %1756, 1, !dbg !64 + %1759 = extractvalue { float, float, float, float } %1756, 2, !dbg !64 + %1760 = extractvalue { float, float, float, float } %1756, 3, !dbg !64 + %1761 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1682, float %1683, float %1684, float %1685, i32 %1450, i32 %1451, i32 %1452, i32 %1453, i32 %1478, i32 %1479) #3, !dbg !64 + %1762 = extractvalue { float, float, float, float } %1761, 0, !dbg !64 + %1763 = extractvalue { float, float, float, float } %1761, 1, !dbg !64 + %1764 = extractvalue { float, float, float, float } %1761, 2, !dbg !64 + %1765 = extractvalue { float, float, float, float } %1761, 3, !dbg !64 + %1766 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1687, float %1688, float %1689, float %1690, i32 %1450, i32 %1451, i32 %1452, i32 %1453, i32 %1486, i32 %1487) #3, !dbg !64 + %1767 = extractvalue { float, float, float, float } %1766, 0, !dbg !64 + %1768 = extractvalue { float, float, float, float } %1766, 1, !dbg !64 + %1769 = extractvalue { float, float, float, float } %1766, 2, !dbg !64 + %1770 = extractvalue { float, float, float, float } %1766, 3, !dbg !64 + %1771 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1692, float %1693, float %1694, float %1695, i32 %1450, i32 %1451, i32 %1452, i32 %1453, i32 %1494, i32 %1495) #3, !dbg !64 + %1772 = extractvalue { float, float, float, float } %1771, 0, !dbg !64 + %1773 = extractvalue { float, float, float, float } %1771, 1, !dbg !64 + %1774 = extractvalue { float, float, float, float } %1771, 2, !dbg !64 + %1775 = extractvalue { float, float, float, float } %1771, 3, !dbg !64 + %1776 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1697, float %1698, float %1699, float %1700, i32 %1455, i32 %1456, i32 %1457, i32 %1458, i32 %1470, i32 %1471) #3, !dbg !64 + %1777 = extractvalue { float, float, float, float } %1776, 0, !dbg !64 + %1778 = extractvalue { float, float, float, float } %1776, 1, !dbg !64 + %1779 = extractvalue { float, float, float, float } %1776, 2, !dbg !64 + %1780 = extractvalue { float, float, float, float } %1776, 3, !dbg !64 + %1781 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1702, float %1703, float %1704, float %1705, i32 %1455, i32 %1456, i32 %1457, i32 %1458, i32 %1478, i32 %1479) #3, !dbg !64 + %1782 = extractvalue { float, float, float, float } %1781, 0, !dbg !64 + %1783 = extractvalue { float, float, float, float } %1781, 1, !dbg !64 + %1784 = extractvalue { float, float, float, float } %1781, 2, !dbg !64 + %1785 = extractvalue { float, float, float, float } %1781, 3, !dbg !64 + %1786 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1707, float %1708, float %1709, float %1710, i32 %1455, i32 %1456, i32 %1457, i32 %1458, i32 %1486, i32 %1487) #3, !dbg !64 + %1787 = extractvalue { float, float, float, float } %1786, 0, !dbg !64 + %1788 = extractvalue { float, float, float, float } %1786, 1, !dbg !64 + %1789 = extractvalue { float, float, float, float } %1786, 2, !dbg !64 + %1790 = extractvalue { float, float, float, float } %1786, 3, !dbg !64 + %1791 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1712, float %1713, float %1714, float %1715, i32 %1455, i32 %1456, i32 %1457, i32 %1458, i32 %1494, i32 %1495) #3, !dbg !64 + %1792 = extractvalue { float, float, float, float } %1791, 0, !dbg !64 + %1793 = extractvalue { float, float, float, float } %1791, 1, !dbg !64 + %1794 = extractvalue { float, float, float, float } %1791, 2, !dbg !64 + %1795 = extractvalue { float, float, float, float } %1791, 3, !dbg !64 + %1796 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1717, float %1718, float %1719, float %1720, i32 %1460, i32 %1461, i32 %1462, i32 %1463, i32 %1470, i32 %1471) #3, !dbg !64 + %1797 = extractvalue { float, float, float, float } %1796, 0, !dbg !64 + %1798 = extractvalue { float, float, float, float } %1796, 1, !dbg !64 + %1799 = extractvalue { float, float, float, float } %1796, 2, !dbg !64 + %1800 = extractvalue { float, float, float, float } %1796, 3, !dbg !64 + %1801 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1722, float %1723, float %1724, float %1725, i32 %1460, i32 %1461, i32 %1462, i32 %1463, i32 %1478, i32 %1479) #3, !dbg !64 + %1802 = extractvalue { float, float, float, float } %1801, 0, !dbg !64 + %1803 = extractvalue { float, float, float, float } %1801, 1, !dbg !64 + %1804 = extractvalue { float, float, float, float } %1801, 2, !dbg !64 + %1805 = extractvalue { float, float, float, float } %1801, 3, !dbg !64 + %1806 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1727, float %1728, float %1729, float %1730, i32 %1460, i32 %1461, i32 %1462, i32 %1463, i32 %1486, i32 %1487) #3, !dbg !64 + %1807 = extractvalue { float, float, float, float } %1806, 0, !dbg !64 + %1808 = extractvalue { float, float, float, float } %1806, 1, !dbg !64 + %1809 = extractvalue { float, float, float, float } %1806, 2, !dbg !64 + %1810 = extractvalue { float, float, float, float } %1806, 3, !dbg !64 + %1811 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1732, float %1733, float %1734, float %1735, i32 %1460, i32 %1461, i32 %1462, i32 %1463, i32 %1494, i32 %1495) #3, !dbg !64 + %1812 = extractvalue { float, float, float, float } %1811, 0, !dbg !64 + %1813 = extractvalue { float, float, float, float } %1811, 1, !dbg !64 + %1814 = extractvalue { float, float, float, float } %1811, 2, !dbg !64 + %1815 = extractvalue { float, float, float, float } %1811, 3, !dbg !64 + %1816 = insertelement <2 x float> poison, float %1737, i64 0, !dbg !65 + %1817 = insertelement <2 x float> %1816, float %1738, i64 1, !dbg !65 + %1818 = fptrunc <2 x float> %1817 to <2 x bfloat>, !dbg !65 + %1819 = insertelement <2 x float> poison, float %1739, i64 0, !dbg !65 + %1820 = insertelement <2 x float> %1819, float %1740, i64 1, !dbg !65 + %1821 = fptrunc <2 x float> %1820 to <2 x bfloat>, !dbg !65 + %1822 = insertelement <2 x float> poison, float %1742, i64 0, !dbg !65 + %1823 = insertelement <2 x float> %1822, float %1743, i64 1, !dbg !65 + %1824 = fptrunc <2 x float> %1823 to <2 x bfloat>, !dbg !65 + %1825 = insertelement <2 x float> poison, float %1744, i64 0, !dbg !65 + %1826 = insertelement <2 x float> %1825, float %1745, i64 1, !dbg !65 + %1827 = fptrunc <2 x float> %1826 to <2 x bfloat>, !dbg !65 + %1828 = insertelement <2 x float> poison, float %1747, i64 0, !dbg !65 + %1829 = insertelement <2 x float> %1828, float %1748, i64 1, !dbg !65 + %1830 = fptrunc <2 x float> %1829 to <2 x bfloat>, !dbg !65 + %1831 = insertelement <2 x float> poison, float %1749, i64 0, !dbg !65 + %1832 = insertelement <2 x float> %1831, float %1750, i64 1, !dbg !65 + %1833 = fptrunc <2 x float> %1832 to <2 x bfloat>, !dbg !65 + %1834 = insertelement <2 x float> poison, float %1752, i64 0, !dbg !65 + %1835 = insertelement <2 x float> %1834, float %1753, i64 1, !dbg !65 + %1836 = fptrunc <2 x float> %1835 to <2 x bfloat>, !dbg !65 + %1837 = insertelement <2 x float> poison, float %1754, i64 0, !dbg !65 + %1838 = insertelement <2 x float> %1837, float %1755, i64 1, !dbg !65 + %1839 = fptrunc <2 x float> %1838 to <2 x bfloat>, !dbg !65 + %1840 = insertelement <2 x float> poison, float %1757, i64 0, !dbg !65 + %1841 = insertelement <2 x float> %1840, float %1758, i64 1, !dbg !65 + %1842 = fptrunc <2 x float> %1841 to <2 x bfloat>, !dbg !65 + %1843 = insertelement <2 x float> poison, float %1759, i64 0, !dbg !65 + %1844 = insertelement <2 x float> %1843, float %1760, i64 1, !dbg !65 + %1845 = fptrunc <2 x float> %1844 to <2 x bfloat>, !dbg !65 + %1846 = insertelement <2 x float> poison, float %1762, i64 0, !dbg !65 + %1847 = insertelement <2 x float> %1846, float %1763, i64 1, !dbg !65 + %1848 = fptrunc <2 x float> %1847 to <2 x bfloat>, !dbg !65 + %1849 = insertelement <2 x float> poison, float %1764, i64 0, !dbg !65 + %1850 = insertelement <2 x float> %1849, float %1765, i64 1, !dbg !65 + %1851 = fptrunc <2 x float> %1850 to <2 x bfloat>, !dbg !65 + %1852 = insertelement <2 x float> poison, float %1767, i64 0, !dbg !65 + %1853 = insertelement <2 x float> %1852, float %1768, i64 1, !dbg !65 + %1854 = fptrunc <2 x float> %1853 to <2 x bfloat>, !dbg !65 + %1855 = insertelement <2 x float> poison, float %1769, i64 0, !dbg !65 + %1856 = insertelement <2 x float> %1855, float %1770, i64 1, !dbg !65 + %1857 = fptrunc <2 x float> %1856 to <2 x bfloat>, !dbg !65 + %1858 = insertelement <2 x float> poison, float %1772, i64 0, !dbg !65 + %1859 = insertelement <2 x float> %1858, float %1773, i64 1, !dbg !65 + %1860 = fptrunc <2 x float> %1859 to <2 x bfloat>, !dbg !65 + %1861 = insertelement <2 x float> poison, float %1774, i64 0, !dbg !65 + %1862 = insertelement <2 x float> %1861, float %1775, i64 1, !dbg !65 + %1863 = fptrunc <2 x float> %1862 to <2 x bfloat>, !dbg !65 + %1864 = insertelement <2 x float> poison, float %1777, i64 0, !dbg !65 + %1865 = insertelement <2 x float> %1864, float %1778, i64 1, !dbg !65 + %1866 = fptrunc <2 x float> %1865 to <2 x bfloat>, !dbg !65 + %1867 = insertelement <2 x float> poison, float %1779, i64 0, !dbg !65 + %1868 = insertelement <2 x float> %1867, float %1780, i64 1, !dbg !65 + %1869 = fptrunc <2 x float> %1868 to <2 x bfloat>, !dbg !65 + %1870 = insertelement <2 x float> poison, float %1782, i64 0, !dbg !65 + %1871 = insertelement <2 x float> %1870, float %1783, i64 1, !dbg !65 + %1872 = fptrunc <2 x float> %1871 to <2 x bfloat>, !dbg !65 + %1873 = insertelement <2 x float> poison, float %1784, i64 0, !dbg !65 + %1874 = insertelement <2 x float> %1873, float %1785, i64 1, !dbg !65 + %1875 = fptrunc <2 x float> %1874 to <2 x bfloat>, !dbg !65 + %1876 = insertelement <2 x float> poison, float %1787, i64 0, !dbg !65 + %1877 = insertelement <2 x float> %1876, float %1788, i64 1, !dbg !65 + %1878 = fptrunc <2 x float> %1877 to <2 x bfloat>, !dbg !65 + %1879 = insertelement <2 x float> poison, float %1789, i64 0, !dbg !65 + %1880 = insertelement <2 x float> %1879, float %1790, i64 1, !dbg !65 + %1881 = fptrunc <2 x float> %1880 to <2 x bfloat>, !dbg !65 + %1882 = insertelement <2 x float> poison, float %1792, i64 0, !dbg !65 + %1883 = insertelement <2 x float> %1882, float %1793, i64 1, !dbg !65 + %1884 = fptrunc <2 x float> %1883 to <2 x bfloat>, !dbg !65 + %1885 = insertelement <2 x float> poison, float %1794, i64 0, !dbg !65 + %1886 = insertelement <2 x float> %1885, float %1795, i64 1, !dbg !65 + %1887 = fptrunc <2 x float> %1886 to <2 x bfloat>, !dbg !65 + %1888 = insertelement <2 x float> poison, float %1797, i64 0, !dbg !65 + %1889 = insertelement <2 x float> %1888, float %1798, i64 1, !dbg !65 + %1890 = fptrunc <2 x float> %1889 to <2 x bfloat>, !dbg !65 + %1891 = insertelement <2 x float> poison, float %1799, i64 0, !dbg !65 + %1892 = insertelement <2 x float> %1891, float %1800, i64 1, !dbg !65 + %1893 = fptrunc <2 x float> %1892 to <2 x bfloat>, !dbg !65 + %1894 = insertelement <2 x float> poison, float %1802, i64 0, !dbg !65 + %1895 = insertelement <2 x float> %1894, float %1803, i64 1, !dbg !65 + %1896 = fptrunc <2 x float> %1895 to <2 x bfloat>, !dbg !65 + %1897 = insertelement <2 x float> poison, float %1804, i64 0, !dbg !65 + %1898 = insertelement <2 x float> %1897, float %1805, i64 1, !dbg !65 + %1899 = fptrunc <2 x float> %1898 to <2 x bfloat>, !dbg !65 + %1900 = insertelement <2 x float> poison, float %1807, i64 0, !dbg !65 + %1901 = insertelement <2 x float> %1900, float %1808, i64 1, !dbg !65 + %1902 = fptrunc <2 x float> %1901 to <2 x bfloat>, !dbg !65 + %1903 = insertelement <2 x float> poison, float %1809, i64 0, !dbg !65 + %1904 = insertelement <2 x float> %1903, float %1810, i64 1, !dbg !65 + %1905 = fptrunc <2 x float> %1904 to <2 x bfloat>, !dbg !65 + %1906 = insertelement <2 x float> poison, float %1812, i64 0, !dbg !65 + %1907 = insertelement <2 x float> %1906, float %1813, i64 1, !dbg !65 + %1908 = fptrunc <2 x float> %1907 to <2 x bfloat>, !dbg !65 + %1909 = insertelement <2 x float> poison, float %1814, i64 0, !dbg !65 + %1910 = insertelement <2 x float> %1909, float %1815, i64 1, !dbg !65 + %1911 = fptrunc <2 x float> %1910 to <2 x bfloat>, !dbg !65 + %1912 = bitcast <2 x bfloat> %1818 to i32, !dbg !66 + %1913 = bitcast <2 x bfloat> %1824 to i32, !dbg !66 + %1914 = bitcast <2 x bfloat> %1830 to i32, !dbg !66 + %1915 = bitcast <2 x bfloat> %1836 to i32, !dbg !66 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %301, i32 %1912, i32 %1913, i32 %1914, i32 %1915), !dbg !66 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !66 + %1916 = load <4 x i32>, ptr addrspace(3) %302, align 16, !dbg !66 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !66 + %1917 = bitcast <2 x bfloat> %1821 to i32, !dbg !66 + %1918 = bitcast <2 x bfloat> %1827 to i32, !dbg !66 + %1919 = bitcast <2 x bfloat> %1833 to i32, !dbg !66 + %1920 = bitcast <2 x bfloat> %1839 to i32, !dbg !66 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %301, i32 %1917, i32 %1918, i32 %1919, i32 %1920), !dbg !66 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !66 + %1921 = load <4 x i32>, ptr addrspace(3) %302, align 16, !dbg !66 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !66 + %1922 = bitcast <2 x bfloat> %1842 to i32, !dbg !66 + %1923 = bitcast <2 x bfloat> %1848 to i32, !dbg !66 + %1924 = bitcast <2 x bfloat> %1854 to i32, !dbg !66 + %1925 = bitcast <2 x bfloat> %1860 to i32, !dbg !66 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %301, i32 %1922, i32 %1923, i32 %1924, i32 %1925), !dbg !66 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !66 + %1926 = load <4 x i32>, ptr addrspace(3) %302, align 16, !dbg !66 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !66 + %1927 = bitcast <2 x bfloat> %1845 to i32, !dbg !66 + %1928 = bitcast <2 x bfloat> %1851 to i32, !dbg !66 + %1929 = bitcast <2 x bfloat> %1857 to i32, !dbg !66 + %1930 = bitcast <2 x bfloat> %1863 to i32, !dbg !66 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %301, i32 %1927, i32 %1928, i32 %1929, i32 %1930), !dbg !66 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !66 + %1931 = load <4 x i32>, ptr addrspace(3) %302, align 16, !dbg !66 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !66 + %1932 = bitcast <2 x bfloat> %1866 to i32, !dbg !66 + %1933 = bitcast <2 x bfloat> %1872 to i32, !dbg !66 + %1934 = bitcast <2 x bfloat> %1878 to i32, !dbg !66 + %1935 = bitcast <2 x bfloat> %1884 to i32, !dbg !66 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %301, i32 %1932, i32 %1933, i32 %1934, i32 %1935), !dbg !66 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !66 + %1936 = load <4 x i32>, ptr addrspace(3) %302, align 16, !dbg !66 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !66 + %1937 = bitcast <2 x bfloat> %1869 to i32, !dbg !66 + %1938 = bitcast <2 x bfloat> %1875 to i32, !dbg !66 + %1939 = bitcast <2 x bfloat> %1881 to i32, !dbg !66 + %1940 = bitcast <2 x bfloat> %1887 to i32, !dbg !66 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %301, i32 %1937, i32 %1938, i32 %1939, i32 %1940), !dbg !66 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !66 + %1941 = load <4 x i32>, ptr addrspace(3) %302, align 16, !dbg !66 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !66 + %1942 = bitcast <2 x bfloat> %1890 to i32, !dbg !66 + %1943 = bitcast <2 x bfloat> %1896 to i32, !dbg !66 + %1944 = bitcast <2 x bfloat> %1902 to i32, !dbg !66 + %1945 = bitcast <2 x bfloat> %1908 to i32, !dbg !66 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %301, i32 %1942, i32 %1943, i32 %1944, i32 %1945), !dbg !66 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !66 + %1946 = load <4 x i32>, ptr addrspace(3) %302, align 16, !dbg !66 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !66 + %1947 = bitcast <2 x bfloat> %1893 to i32, !dbg !66 + %1948 = bitcast <2 x bfloat> %1899 to i32, !dbg !66 + %1949 = bitcast <2 x bfloat> %1905 to i32, !dbg !66 + %1950 = bitcast <2 x bfloat> %1911 to i32, !dbg !66 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %301, i32 %1947, i32 %1948, i32 %1949, i32 %1950), !dbg !66 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !66 + %1951 = load <4 x i32>, ptr addrspace(3) %302, align 16, !dbg !66 + %.extract = extractelement <4 x i32> %1916, i64 0, !dbg !66 + %.extract301 = extractelement <4 x i32> %1916, i64 1, !dbg !66 + %.extract303 = extractelement <4 x i32> %1916, i64 2, !dbg !66 + %.extract305 = extractelement <4 x i32> %1916, i64 3, !dbg !66 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract, i32 %.extract301, i32 %.extract303, i32 %.extract305, ptr addrspace(1) %1160, i1 %1143) #3, !dbg !66 + %.extract307 = extractelement <4 x i32> %1921, i64 0, !dbg !66 + %.extract309 = extractelement <4 x i32> %1921, i64 1, !dbg !66 + %.extract311 = extractelement <4 x i32> %1921, i64 2, !dbg !66 + %.extract313 = extractelement <4 x i32> %1921, i64 3, !dbg !66 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract307, i32 %.extract309, i32 %.extract311, i32 %.extract313, ptr addrspace(1) %1162, i1 %1144) #3, !dbg !66 + %.extract315 = extractelement <4 x i32> %1926, i64 0, !dbg !66 + %.extract317 = extractelement <4 x i32> %1926, i64 1, !dbg !66 + %.extract319 = extractelement <4 x i32> %1926, i64 2, !dbg !66 + %.extract321 = extractelement <4 x i32> %1926, i64 3, !dbg !66 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract315, i32 %.extract317, i32 %.extract319, i32 %.extract321, ptr addrspace(1) %1164, i1 %1145) #3, !dbg !66 + %.extract323 = extractelement <4 x i32> %1931, i64 0, !dbg !66 + %.extract325 = extractelement <4 x i32> %1931, i64 1, !dbg !66 + %.extract327 = extractelement <4 x i32> %1931, i64 2, !dbg !66 + %.extract329 = extractelement <4 x i32> %1931, i64 3, !dbg !66 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract323, i32 %.extract325, i32 %.extract327, i32 %.extract329, ptr addrspace(1) %1166, i1 %1146) #3, !dbg !66 + %.extract331 = extractelement <4 x i32> %1936, i64 0, !dbg !66 + %.extract333 = extractelement <4 x i32> %1936, i64 1, !dbg !66 + %.extract335 = extractelement <4 x i32> %1936, i64 2, !dbg !66 + %.extract337 = extractelement <4 x i32> %1936, i64 3, !dbg !66 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract331, i32 %.extract333, i32 %.extract335, i32 %.extract337, ptr addrspace(1) %1168, i1 %1147) #3, !dbg !66 + %.extract339 = extractelement <4 x i32> %1941, i64 0, !dbg !66 + %.extract341 = extractelement <4 x i32> %1941, i64 1, !dbg !66 + %.extract343 = extractelement <4 x i32> %1941, i64 2, !dbg !66 + %.extract345 = extractelement <4 x i32> %1941, i64 3, !dbg !66 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract339, i32 %.extract341, i32 %.extract343, i32 %.extract345, ptr addrspace(1) %1170, i1 %1148) #3, !dbg !66 + %.extract347 = extractelement <4 x i32> %1946, i64 0, !dbg !66 + %.extract349 = extractelement <4 x i32> %1946, i64 1, !dbg !66 + %.extract351 = extractelement <4 x i32> %1946, i64 2, !dbg !66 + %.extract353 = extractelement <4 x i32> %1946, i64 3, !dbg !66 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract347, i32 %.extract349, i32 %.extract351, i32 %.extract353, ptr addrspace(1) %1172, i1 %1149) #3, !dbg !66 + %.extract355 = extractelement <4 x i32> %1951, i64 0, !dbg !66 + %.extract357 = extractelement <4 x i32> %1951, i64 1, !dbg !66 + %.extract359 = extractelement <4 x i32> %1951, i64 2, !dbg !66 + %.extract361 = extractelement <4 x i32> %1951, i64 3, !dbg !66 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract355, i32 %.extract357, i32 %.extract359, i32 %.extract361, ptr addrspace(1) %1174, i1 %1150) #3, !dbg !66 + %1952 = add i32 %1155, 1, !dbg !60 + %1953 = icmp sgt i32 %1952, 2, !dbg !60 + %1954 = select i1 %1953, i32 0, i32 %1952, !dbg !60 + %1955 = or disjoint i64 %indvars.iv426, %111, !dbg !67 + %1956 = trunc nuw nsw i64 %1955 to i32, !dbg !58 + %1957 = getelementptr bfloat, ptr addrspace(1) %1092, i64 %1955, !dbg !58 + %1958 = getelementptr bfloat, ptr addrspace(1) %1093, i64 %1955, !dbg !58 + %1959 = getelementptr bfloat, ptr addrspace(1) %1094, i64 %1955, !dbg !58 + %1960 = getelementptr bfloat, ptr addrspace(1) %1095, i64 %1955, !dbg !58 + %1961 = getelementptr bfloat, ptr addrspace(1) %1096, i64 %1955, !dbg !58 + %1962 = getelementptr bfloat, ptr addrspace(1) %1097, i64 %1955, !dbg !58 + %1963 = getelementptr bfloat, ptr addrspace(1) %1098, i64 %1955, !dbg !58 + %1964 = getelementptr bfloat, ptr addrspace(1) %1099, i64 %1955, !dbg !58 + %.idx201 = shl i32 %1954, 13, !dbg !59 + %1965 = getelementptr i8, ptr addrspace(3) @global_smem, i32 %.idx201, !dbg !59 + %1966 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 %170, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %1966, ptr addrspace(1) %1957, i32 0) #3, !dbg !59 + %1967 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 %172, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %1967, ptr addrspace(1) %1958, i32 0) #3, !dbg !59 + %1968 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 %174, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %1968, ptr addrspace(1) %1959, i32 0) #3, !dbg !59 + %1969 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 %176, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %1969, ptr addrspace(1) %1960, i32 0) #3, !dbg !59 + %1970 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 %178, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %1970, ptr addrspace(1) %1961, i32 0) #3, !dbg !59 + %1971 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 %180, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %1971, ptr addrspace(1) %1962, i32 0) #3, !dbg !59 + %1972 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 %182, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %1972, ptr addrspace(1) %1963, i32 0) #3, !dbg !59 + %1973 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 %184, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %1973, ptr addrspace(1) %1964, i32 0) #3, !dbg !59 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !59 + br i1 %1134, label %1133, label %1974, !dbg !60 + +1974: ; preds = %1133 + tail call void @llvm.nvvm.cp.async.wait.group(i32 0), !dbg !60 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !60 + ret void, !dbg !68 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 65535) i32 @llvm.nvvm.read.ptx.sreg.ctaid.y() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #2 + +; Function Attrs: nounwind +declare void @llvm.nvvm.cp.async.commit.group() #3 + +; Function Attrs: nounwind +declare void @llvm.nvvm.cp.async.wait.group(i32 immarg) #3 + +; Function Attrs: convergent nocallback nofree nounwind memory(argmem: read) +declare { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) readonly captures(none)) #4 + +; Function Attrs: convergent nocallback nofree nounwind memory(argmem: read) +declare { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) readonly captures(none)) #4 + +; Function Attrs: convergent nocallback nounwind memory(argmem: write) +declare void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) writeonly captures(none), i32, i32, i32, i32) #5 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #6 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.ex2.approx.ftz.f32(float) #7 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.ex2.approx.f32(float) #7 + +attributes #0 = { nounwind "nvvm.reqntid"="64" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind } +attributes #3 = { nounwind } +attributes #4 = { convergent nocallback nofree nounwind memory(argmem: read) } +attributes #5 = { convergent nocallback nounwind memory(argmem: write) } +attributes #6 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #7 = { mustprogress nocallback nofree nosync nounwind willreturn memory(none) } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "wy_fast.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "recompute_w_u_fwd_kernel", linkageName: "recompute_w_u_fwd_kernel", scope: !1, file: !1, line: 43, type: !6, scopeLine: 43, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !8, !8, !8, !8, !10, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 64, column: 30, scope: !5) +!14 = !DILocation(line: 64, column: 36, scope: !5) +!15 = !DILocation(line: 64, column: 61, scope: !5) +!16 = !DILocation(line: 64, column: 67, scope: !5) +!17 = !DILocation(line: 71, column: 25, scope: !5) +!18 = !DILocation(line: 65, column: 34, scope: !5) +!19 = !DILocation(line: 72, column: 16, scope: !5) +!20 = !DILocation(line: 72, column: 34, scope: !5) +!21 = !DILocation(line: 72, column: 21, scope: !5) +!22 = !DILocation(line: 74, column: 16, scope: !5) +!23 = !DILocation(line: 76, column: 21, scope: !5) +!24 = !DILocation(line: 76, scope: !5) +!25 = !DILocation(line: 76, column: 38, scope: !5) +!26 = !DILocation(line: 76, column: 32, scope: !5) +!27 = !DILocation(line: 77, column: 18, scope: !5) +!28 = !DILocation(line: 102, column: 21, scope: !5) +!29 = !DILocation(line: 79, column: 14, scope: !5) +!30 = !DILocation(line: 79, column: 36, scope: !5) +!31 = !DILocation(line: 79, column: 61, scope: !5) +!32 = !DILocation(line: 80, column: 18, scope: !5) +!33 = !DILocation(line: 82, column: 21, scope: !5) +!34 = !DILocation(line: 85, column: 18, scope: !5) +!35 = !DILocation(line: 85, column: 39, scope: !5) +!36 = !DILocation(line: 86, column: 18, scope: !5) +!37 = !DILocation(line: 86, column: 39, scope: !5) +!38 = !DILocation(line: 85, column: 63, scope: !5) +!39 = !DILocation(line: 87, column: 22, scope: !5) +!40 = !DILocation(line: 83, column: 25, scope: !5) +!41 = !DILocation(line: 88, column: 22, scope: !5) +!42 = !DILocation(line: 86, column: 63, scope: !5) +!43 = !DILocation(line: 89, column: 26, scope: !5) +!44 = !DILocation(line: 90, column: 29, scope: !5) +!45 = !DILocation(line: 90, column: 22, scope: !5) +!46 = !DILocation(line: 93, column: 19, scope: !5) +!47 = !DILocation(line: 93, column: 35, scope: !5) +!48 = !DILocation(line: 94, column: 27, scope: !5) +!49 = !DILocation(line: 47, column: 33, scope: !50, inlinedAt: !52) +!50 = distinct !DILexicalBlockFile(scope: !5, file: !51, discriminator: 0) +!51 = !DIFile(filename: "op.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils") +!52 = !DILocation(line: 94, column: 19, scope: !53) +!53 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!54 = !DILocation(line: 99, column: 18, scope: !5) +!55 = !DILocation(line: 99, column: 51, scope: !5) +!56 = !DILocation(line: 100, column: 18, scope: !5) +!57 = !DILocation(line: 100, column: 39, scope: !5) +!58 = !DILocation(line: 99, column: 74, scope: !5) +!59 = !DILocation(line: 101, column: 22, scope: !5) +!60 = !DILocation(line: 96, column: 21, scope: !5) +!61 = !DILocation(line: 100, column: 63, scope: !5) +!62 = !DILocation(line: 105, column: 34, scope: !5) +!63 = !DILocation(line: 104, column: 20, scope: !5) +!64 = !DILocation(line: 105, column: 26, scope: !5) +!65 = !DILocation(line: 106, column: 29, scope: !5) +!66 = !DILocation(line: 106, column: 22, scope: !5) +!67 = !DILocation(line: 97, column: 25, scope: !5) +!68 = !DILocation(line: 96, column: 4, scope: !5) diff --git a/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.ptx b/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..dd394bb653bafc3e5affe3753dfb06a327af6ba9 --- /dev/null +++ b/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.ptx @@ -0,0 +1,2267 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl recompute_w_u_fwd_kernel // -- Begin function recompute_w_u_fwd_kernel +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; + // @recompute_w_u_fwd_kernel +.visible .entry recompute_w_u_fwd_kernel( + .param .u64 .ptr .global .align 1 recompute_w_u_fwd_kernel_param_0, + .param .u64 .ptr .global .align 1 recompute_w_u_fwd_kernel_param_1, + .param .u64 .ptr .global .align 1 recompute_w_u_fwd_kernel_param_2, + .param .u64 .ptr .global .align 1 recompute_w_u_fwd_kernel_param_3, + .param .u64 .ptr .global .align 1 recompute_w_u_fwd_kernel_param_4, + .param .u64 .ptr .global .align 1 recompute_w_u_fwd_kernel_param_5, + .param .u64 .ptr .global .align 1 recompute_w_u_fwd_kernel_param_6, + .param .u32 recompute_w_u_fwd_kernel_param_7, + .param .u64 .ptr .global .align 1 recompute_w_u_fwd_kernel_param_8, + .param .u64 .ptr .global .align 1 recompute_w_u_fwd_kernel_param_9 +) +.reqntid 64 +{ + .reg .pred %p<58>; + .reg .b16 %rs<83>; + .reg .b32 %r<890>; + .reg .b64 %rd<285>; + .loc 1 43 0 // wy_fast.py:43:0 +$L__func_begin0: + .loc 1 43 0 // wy_fast.py:43:0 + +// %bb.0: + ld.param.b64 %rd55, [recompute_w_u_fwd_kernel_param_6]; + ld.param.b64 %rd54, [recompute_w_u_fwd_kernel_param_3]; + ld.param.b64 %rd53, [recompute_w_u_fwd_kernel_param_0]; + ld.param.b64 %rd89, [recompute_w_u_fwd_kernel_param_1]; +$L__tmp0: + .loc 1 64 30 // wy_fast.py:64:30 + mov.u32 %r87, %ctaid.x; + ld.param.b64 %rd90, [recompute_w_u_fwd_kernel_param_2]; + .loc 1 64 61 // wy_fast.py:64:61 + mov.u32 %r88, %ctaid.y; + .loc 1 64 67 // wy_fast.py:64:67 + cvt.u64.u32 %rd91, %r88; + ld.param.b64 %rd92, [recompute_w_u_fwd_kernel_param_4]; + .loc 1 71 25 // wy_fast.py:71:25 + and.b64 %rd93, %rd91, 65504; + ld.param.b64 %rd94, [recompute_w_u_fwd_kernel_param_5]; + .loc 1 65 34 // wy_fast.py:65:34 + and.b64 %rd95, %rd91, 31; + ld.param.s32 %rd96, [recompute_w_u_fwd_kernel_param_7]; + .loc 1 72 16 // wy_fast.py:72:16 + mul.wide.u32 %rd97, %r87, 64; + .loc 1 72 34 // wy_fast.py:72:34 + mov.u32 %r89, %tid.x; + and.b32 %r90, %r89, 56; + bfe.u32 %r91, %r89, 3, 3; + and.b32 %r92, %r89, 7; + shl.b32 %r887, %r92, 3; + and.b32 %r93, %r89, 63; + or.b32 %r94, %r91, 8; + or.b32 %r95, %r91, 16; + or.b32 %r96, %r91, 24; + or.b32 %r97, %r91, 32; + or.b32 %r98, %r91, 40; + or.b32 %r99, %r91, 48; + or.b32 %r100, %r91, 56; + .loc 1 72 21 // wy_fast.py:72:21 + cvt.u64.u32 %rd98, %r91; + cvt.u64.u32 %rd99, %r100; + cvt.u64.u32 %rd100, %r99; + cvt.u64.u32 %rd101, %r98; + cvt.u64.u32 %rd102, %r97; + cvt.u64.u32 %rd103, %r96; + cvt.u64.u32 %rd104, %r95; + cvt.u64.u32 %rd105, %r94; + cvt.u64.u32 %rd106, %r93; + or.b64 %rd107, %rd97, %rd98; + or.b64 %rd108, %rd97, %rd106; + or.b64 %rd109, %rd97, %rd105; + or.b64 %rd110, %rd97, %rd104; + or.b64 %rd111, %rd97, %rd103; + or.b64 %rd112, %rd97, %rd102; + or.b64 %rd113, %rd97, %rd101; + or.b64 %rd114, %rd97, %rd100; + or.b64 %rd115, %rd97, %rd99; + .loc 1 74 16 // wy_fast.py:74:16 + setp.lt.s64 %p42, %rd107, %rd96; + setp.lt.s64 %p49, %rd115, %rd96; + setp.lt.s64 %p48, %rd114, %rd96; + setp.lt.s64 %p47, %rd113, %rd96; + setp.lt.s64 %p46, %rd112, %rd96; + setp.lt.s64 %p45, %rd111, %rd96; + setp.lt.s64 %p44, %rd110, %rd96; + setp.lt.s64 %p43, %rd109, %rd96; + setp.lt.s64 %p19, %rd108, %rd96; + .loc 1 76 21 // wy_fast.py:76:21 + mul.lo.s64 %rd116, %rd93, %rd96; + .loc 1 76 0 // wy_fast.py:76 + or.b64 %rd1, %rd116, %rd95; + shl.b64 %rd117, %rd1, 1; + add.s64 %rd118, %rd90, %rd117; + .loc 1 76 38 // wy_fast.py:76:38 + shl.b64 %rd2, %rd108, 5; + .loc 1 76 32 // wy_fast.py:76:32 + shl.b64 %rd119, %rd108, 6; + add.s64 %rd56, %rd118, %rd119; + mov.b16 %rs1, 0; + .loc 1 77 18 // wy_fast.py:77:18 + // begin inline asm + mov.u16 %rs2, %rs1; + @%p19 ld.global.b16 { %rs2 }, [ %rd56 + 0 ]; + // end inline asm + .loc 1 102 21 // wy_fast.py:102:21 + shl.b32 %r101, %r89, 4; + and.b32 %r1, %r101, 96; + and.b32 %r2, %r89, 24; + shl.b32 %r3, %r89, 2; + and.b32 %r102, %r3, 4; + and.b32 %r103, %r89, 32; + shr.u32 %r104, %r103, 4; + mov.b32 %r105, global_smem; + add.s32 %r106, %r105, %r102; + add.s32 %r107, %r106, %r104; + add.s32 %r108, %r107, %r1; + add.s32 %r45, %r108, %r2; + // begin inline asm + st.shared.b16 [ %r45 + 0 ], %rs2; + // end inline asm + bar.sync 0; + shl.b32 %r109, %r89, 5; + and.b32 %r110, %r109, 96; + add.s32 %r4, %r105, %r110; + ld.shared.v4.b32 {%r111, %r112, %r113, %r114}, [%r4]; + ld.shared.v4.b32 {%r115, %r116, %r117, %r118}, [%r4+16]; + .loc 1 79 14 // wy_fast.py:79:14 + shl.b64 %rd3, %rd1, 7; + add.s64 %rd120, %rd94, %rd3; + .loc 1 79 36 // wy_fast.py:79:36 + shl.b64 %rd4, %rd107, 12; + add.s64 %rd121, %rd120, %rd4; + shl.b64 %rd5, %rd109, 12; + add.s64 %rd122, %rd120, %rd5; + shl.b64 %rd6, %rd110, 12; + add.s64 %rd123, %rd120, %rd6; + shl.b64 %rd7, %rd111, 12; + add.s64 %rd124, %rd120, %rd7; + shl.b64 %rd8, %rd112, 12; + add.s64 %rd125, %rd120, %rd8; + shl.b64 %rd9, %rd113, 12; + add.s64 %rd126, %rd120, %rd9; + shl.b64 %rd10, %rd114, 12; + add.s64 %rd127, %rd120, %rd10; + shl.b64 %rd11, %rd115, 12; + add.s64 %rd128, %rd120, %rd11; + .loc 1 79 61 // wy_fast.py:79:61 + cvt.u64.u32 %rd12, %r887; + mul.wide.u32 %rd129, %r887, 2; + add.s64 %rd57, %rd121, %rd129; + add.s64 %rd58, %rd122, %rd129; + add.s64 %rd59, %rd123, %rd129; + add.s64 %rd60, %rd124, %rd129; + add.s64 %rd61, %rd125, %rd129; + add.s64 %rd62, %rd126, %rd129; + add.s64 %rd63, %rd127, %rd129; + add.s64 %rd64, %rd128, %rd129; + mov.b32 %r46, 0; + .loc 1 80 18 // wy_fast.py:80:18 + // begin inline asm + mov.u32 %r48, %r46; + mov.u32 %r49, %r46; + mov.u32 %r50, %r46; + mov.u32 %r51, %r46; + @%p42 ld.global.v4.b32 { %r48, %r49, %r50, %r51 }, [ %rd57 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r53, %r46; + mov.u32 %r54, %r46; + mov.u32 %r55, %r46; + mov.u32 %r56, %r46; + @%p43 ld.global.v4.b32 { %r53, %r54, %r55, %r56 }, [ %rd58 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r58, %r46; + mov.u32 %r59, %r46; + mov.u32 %r60, %r46; + mov.u32 %r61, %r46; + @%p44 ld.global.v4.b32 { %r58, %r59, %r60, %r61 }, [ %rd59 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r63, %r46; + mov.u32 %r64, %r46; + mov.u32 %r65, %r46; + mov.u32 %r66, %r46; + @%p45 ld.global.v4.b32 { %r63, %r64, %r65, %r66 }, [ %rd60 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r68, %r46; + mov.u32 %r69, %r46; + mov.u32 %r70, %r46; + mov.u32 %r71, %r46; + @%p46 ld.global.v4.b32 { %r68, %r69, %r70, %r71 }, [ %rd61 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r73, %r46; + mov.u32 %r74, %r46; + mov.u32 %r75, %r46; + mov.u32 %r76, %r46; + @%p47 ld.global.v4.b32 { %r73, %r74, %r75, %r76 }, [ %rd62 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r78, %r46; + mov.u32 %r79, %r46; + mov.u32 %r80, %r46; + mov.u32 %r81, %r46; + @%p48 ld.global.v4.b32 { %r78, %r79, %r80, %r81 }, [ %rd63 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r83, %r46; + mov.u32 %r84, %r46; + mov.u32 %r85, %r46; + mov.u32 %r86, %r46; + @%p49 ld.global.v4.b32 { %r83, %r84, %r85, %r86 }, [ %rd64 + 0 ]; + // end inline asm + and.b32 %r119, %r101, 1008; + shl.b32 %r120, %r90, 1; + xor.b32 %r121, %r119, %r120; + add.s32 %r5, %r105, %r121; + add.s32 %r47, %r5, 25600; + // begin inline asm + st.shared.v4.b32 [ %r47 + 0 ], { %r48, %r49, %r50, %r51 }; + // end inline asm + add.s32 %r52, %r5, 26624; + // begin inline asm + st.shared.v4.b32 [ %r52 + 0 ], { %r53, %r54, %r55, %r56 }; + // end inline asm + add.s32 %r57, %r5, 27648; + // begin inline asm + st.shared.v4.b32 [ %r57 + 0 ], { %r58, %r59, %r60, %r61 }; + // end inline asm + add.s32 %r62, %r5, 28672; + // begin inline asm + st.shared.v4.b32 [ %r62 + 0 ], { %r63, %r64, %r65, %r66 }; + // end inline asm + add.s32 %r67, %r5, 29696; + // begin inline asm + st.shared.v4.b32 [ %r67 + 0 ], { %r68, %r69, %r70, %r71 }; + // end inline asm + add.s32 %r72, %r5, 30720; + // begin inline asm + st.shared.v4.b32 [ %r72 + 0 ], { %r73, %r74, %r75, %r76 }; + // end inline asm + add.s32 %r77, %r5, 31744; + // begin inline asm + st.shared.v4.b32 [ %r77 + 0 ], { %r78, %r79, %r80, %r81 }; + // end inline asm + add.s32 %r82, %r5, 32768; + // begin inline asm + st.shared.v4.b32 [ %r82 + 0 ], { %r83, %r84, %r85, %r86 }; + // end inline asm + .loc 1 85 18 // wy_fast.py:85:18 + shl.b64 %rd130, %rd1, 8; + add.s64 %rd131, %rd89, %rd130; + .loc 1 85 39 // wy_fast.py:85:39 + shl.b64 %rd132, %rd107, 13; + add.s64 %rd13, %rd131, %rd132; + shl.b64 %rd133, %rd109, 13; + add.s64 %rd14, %rd131, %rd133; + shl.b64 %rd134, %rd110, 13; + add.s64 %rd15, %rd131, %rd134; + shl.b64 %rd135, %rd111, 13; + add.s64 %rd16, %rd131, %rd135; + shl.b64 %rd136, %rd112, 13; + add.s64 %rd17, %rd131, %rd136; + shl.b64 %rd137, %rd113, 13; + add.s64 %rd18, %rd131, %rd137; + shl.b64 %rd138, %rd114, 13; + add.s64 %rd19, %rd131, %rd138; + shl.b64 %rd139, %rd115, 13; + add.s64 %rd20, %rd131, %rd139; + .loc 1 86 18 // wy_fast.py:86:18 + add.s64 %rd140, %rd92, %rd130; + .loc 1 86 39 // wy_fast.py:86:39 + add.s64 %rd21, %rd140, %rd132; + add.s64 %rd22, %rd140, %rd133; + add.s64 %rd23, %rd140, %rd134; + add.s64 %rd24, %rd140, %rd135; + add.s64 %rd25, %rd140, %rd136; + add.s64 %rd26, %rd140, %rd137; + add.s64 %rd27, %rd140, %rd138; + add.s64 %rd28, %rd140, %rd139; + .loc 1 85 63 // wy_fast.py:85:63 + add.s64 %rd65, %rd13, %rd129; + add.s64 %rd66, %rd14, %rd129; + add.s64 %rd67, %rd15, %rd129; + add.s64 %rd68, %rd16, %rd129; + add.s64 %rd69, %rd17, %rd129; + add.s64 %rd70, %rd18, %rd129; + add.s64 %rd71, %rd19, %rd129; + add.s64 %rd72, %rd20, %rd129; + .loc 1 87 22 // wy_fast.py:87:22 + bar.sync 0; + selp.b32 %r6, 16, 0, %p42; + // begin inline asm + cp.async.cg.shared.global [ %r5 + 0 ], [ %rd65 + 0 ], 0x10, %r6; + // end inline asm + add.s32 %r421, %r5, 1024; + selp.b32 %r7, 16, 0, %p43; + // begin inline asm + cp.async.cg.shared.global [ %r421 + 0 ], [ %rd66 + 0 ], 0x10, %r7; + // end inline asm + add.s32 %r422, %r5, 2048; + selp.b32 %r8, 16, 0, %p44; + // begin inline asm + cp.async.cg.shared.global [ %r422 + 0 ], [ %rd67 + 0 ], 0x10, %r8; + // end inline asm + add.s32 %r423, %r5, 3072; + selp.b32 %r9, 16, 0, %p45; + // begin inline asm + cp.async.cg.shared.global [ %r423 + 0 ], [ %rd68 + 0 ], 0x10, %r9; + // end inline asm + add.s32 %r424, %r5, 4096; + selp.b32 %r10, 16, 0, %p46; + // begin inline asm + cp.async.cg.shared.global [ %r424 + 0 ], [ %rd69 + 0 ], 0x10, %r10; + // end inline asm + add.s32 %r425, %r5, 5120; + selp.b32 %r11, 16, 0, %p47; + // begin inline asm + cp.async.cg.shared.global [ %r425 + 0 ], [ %rd70 + 0 ], 0x10, %r11; + // end inline asm + add.s32 %r426, %r5, 6144; + selp.b32 %r12, 16, 0, %p48; + // begin inline asm + cp.async.cg.shared.global [ %r426 + 0 ], [ %rd71 + 0 ], 0x10, %r12; + // end inline asm + add.s32 %r427, %r5, 7168; + selp.b32 %r13, 16, 0, %p49; + // begin inline asm + cp.async.cg.shared.global [ %r427 + 0 ], [ %rd72 + 0 ], 0x10, %r13; + // end inline asm + cp.async.commit_group; + .loc 1 83 25 // wy_fast.py:83:25 + or.b32 %r886, %r887, 64; + .loc 1 85 63 // wy_fast.py:85:63 + add.s64 %rd73, %rd65, 128; + add.s64 %rd74, %rd66, 128; + add.s64 %rd75, %rd67, 128; + add.s64 %rd76, %rd68, 128; + add.s64 %rd77, %rd69, 128; + add.s64 %rd78, %rd70, 128; + add.s64 %rd79, %rd71, 128; + add.s64 %rd80, %rd72, 128; + .loc 1 87 22 // wy_fast.py:87:22 + bar.sync 0; + add.s32 %r428, %r5, 8192; + // begin inline asm + cp.async.cg.shared.global [ %r428 + 0 ], [ %rd73 + 0 ], 0x10, %r6; + // end inline asm + add.s32 %r429, %r5, 9216; + // begin inline asm + cp.async.cg.shared.global [ %r429 + 0 ], [ %rd74 + 0 ], 0x10, %r7; + // end inline asm + add.s32 %r430, %r5, 10240; + // begin inline asm + cp.async.cg.shared.global [ %r430 + 0 ], [ %rd75 + 0 ], 0x10, %r8; + // end inline asm + add.s32 %r431, %r5, 11264; + // begin inline asm + cp.async.cg.shared.global [ %r431 + 0 ], [ %rd76 + 0 ], 0x10, %r9; + // end inline asm + add.s32 %r432, %r5, 12288; + // begin inline asm + cp.async.cg.shared.global [ %r432 + 0 ], [ %rd77 + 0 ], 0x10, %r10; + // end inline asm + add.s32 %r433, %r5, 13312; + // begin inline asm + cp.async.cg.shared.global [ %r433 + 0 ], [ %rd78 + 0 ], 0x10, %r11; + // end inline asm + add.s32 %r434, %r5, 14336; + // begin inline asm + cp.async.cg.shared.global [ %r434 + 0 ], [ %rd79 + 0 ], 0x10, %r12; + // end inline asm + add.s32 %r435, %r5, 15360; + // begin inline asm + cp.async.cg.shared.global [ %r435 + 0 ], [ %rd80 + 0 ], 0x10, %r13; + // end inline asm + cp.async.commit_group; + .loc 1 83 25 // wy_fast.py:83:25 + or.b32 %r885, %r887, 128; + .loc 1 85 63 // wy_fast.py:85:63 + add.s64 %rd81, %rd65, 256; + add.s64 %rd82, %rd66, 256; + add.s64 %rd83, %rd67, 256; + add.s64 %rd84, %rd68, 256; + add.s64 %rd85, %rd69, 256; + add.s64 %rd86, %rd70, 256; + add.s64 %rd87, %rd71, 256; + add.s64 %rd88, %rd72, 256; + .loc 1 87 22 // wy_fast.py:87:22 + bar.sync 0; + add.s32 %r436, %r5, 16384; + // begin inline asm + cp.async.cg.shared.global [ %r436 + 0 ], [ %rd81 + 0 ], 0x10, %r46; + // end inline asm + add.s32 %r437, %r5, 17408; + // begin inline asm + cp.async.cg.shared.global [ %r437 + 0 ], [ %rd82 + 0 ], 0x10, %r46; + // end inline asm + add.s32 %r438, %r5, 18432; + // begin inline asm + cp.async.cg.shared.global [ %r438 + 0 ], [ %rd83 + 0 ], 0x10, %r46; + // end inline asm + add.s32 %r439, %r5, 19456; + // begin inline asm + cp.async.cg.shared.global [ %r439 + 0 ], [ %rd84 + 0 ], 0x10, %r46; + // end inline asm + add.s32 %r440, %r5, 20480; + // begin inline asm + cp.async.cg.shared.global [ %r440 + 0 ], [ %rd85 + 0 ], 0x10, %r46; + // end inline asm + add.s32 %r441, %r5, 21504; + // begin inline asm + cp.async.cg.shared.global [ %r441 + 0 ], [ %rd86 + 0 ], 0x10, %r46; + // end inline asm + add.s32 %r442, %r5, 22528; + // begin inline asm + cp.async.cg.shared.global [ %r442 + 0 ], [ %rd87 + 0 ], 0x10, %r46; + // end inline asm + add.s32 %r443, %r5, 23552; + // begin inline asm + cp.async.cg.shared.global [ %r443 + 0 ], [ %rd88 + 0 ], 0x10, %r46; + // end inline asm + cp.async.commit_group; + shl.b32 %r122, %r89, 7; + and.b32 %r123, %r122, 3968; + shl.b32 %r124, %r92, 4; + shr.u32 %r125, %r103, 1; + or.b32 %r126, %r123, %r124; + xor.b32 %r14, %r126, %r125; + xor.b32 %r15, %r14, 32; + xor.b32 %r16, %r14, 64; + xor.b32 %r17, %r14, 96; + and.b32 %r127, %r122, 1920; + and.b32 %r128, %r89, 16; + or.b32 %r129, %r127, %r124; + xor.b32 %r130, %r129, %r128; + add.s32 %r131, %r105, %r130; + add.s32 %r18, %r131, 25600; + add.s32 %r19, %r131, 27648; + add.s32 %r20, %r131, 29696; + add.s32 %r21, %r131, 31744; + xor.b32 %r132, %r130, 32; + add.s32 %r133, %r105, %r132; + add.s32 %r22, %r133, 25600; + add.s32 %r23, %r133, 27648; + add.s32 %r24, %r133, 29696; + add.s32 %r25, %r133, 31744; + xor.b32 %r134, %r130, 64; + add.s32 %r135, %r105, %r134; + add.s32 %r26, %r135, 25600; + add.s32 %r27, %r135, 27648; + add.s32 %r28, %r135, 29696; + add.s32 %r29, %r135, 31744; + xor.b32 %r136, %r130, 96; + add.s32 %r137, %r105, %r136; + add.s32 %r30, %r137, 25600; + add.s32 %r31, %r137, 27648; + add.s32 %r32, %r137, 29696; + add.s32 %r33, %r137, 31744; + shl.b32 %r138, %r92, 7; + shl.b32 %r139, %r2, 2; + xor.b32 %r140, %r124, %r139; + xor.b32 %r141, %r140, %r125; + add.s32 %r142, %r105, %r138; + add.s32 %r143, %r142, %r141; + add.s32 %r34, %r143, 24576; + .loc 1 88 22 // wy_fast.py:88:22 + mov.b32 {%rs3, %rs4}, %r118; + mov.b32 {%rs5, %rs6}, %r117; + mov.b32 %r35, {%rs6, %rs4}; + mov.b32 {%rs7, %rs8}, %r116; + mov.b32 {%rs9, %rs10}, %r115; + mov.b32 %r36, {%rs10, %rs8}; + mov.b32 {%rs11, %rs12}, %r114; + mov.b32 {%rs13, %rs14}, %r113; + mov.b32 %r37, {%rs14, %rs12}; + mov.b32 {%rs15, %rs16}, %r112; + mov.b32 {%rs17, %rs18}, %r111; + mov.b32 %r38, {%rs18, %rs16}; + mov.b32 %r39, {%rs5, %rs3}; + mov.b32 %r40, {%rs9, %rs7}; + mov.b32 %r41, {%rs13, %rs11}; + mov.b32 %r42, {%rs17, %rs15}; + mov.b32 %r884, 2; + mov.b32 %r883, -1; + mov.b64 %rd283, 192; + mov.pred %p24, -1; + mov.pred %p25, %p42; + mov.pred %p26, %p43; + mov.pred %p27, %p44; + mov.pred %p28, %p45; + mov.pred %p29, %p46; + mov.pred %p30, %p47; + mov.pred %p31, %p48; + mov.pred %p32, %p49; + mov.pred %p33, %p42; + mov.pred %p34, %p43; + mov.pred %p35, %p44; + mov.pred %p36, %p45; + mov.pred %p37, %p46; + mov.pred %p38, %p47; + mov.pred %p39, %p48; + mov.pred %p40, %p49; + mov.b32 %r880, %r885; + mov.b32 %r881, %r886; + mov.b32 %r882, %r887; +$L__BB0_1: // =>This Inner Loop Header: Depth=1 + .loc 1 0 22 // wy_fast.py:0:22 + mov.b32 %r43, %r882; + mov.b32 %r882, %r881; + mov.b32 %r881, %r880; + mov.pred %p9, %p40; + mov.pred %p8, %p39; + mov.pred %p7, %p38; + mov.pred %p6, %p37; + mov.pred %p5, %p36; + mov.pred %p4, %p35; + mov.pred %p3, %p34; + mov.pred %p2, %p33; + mov.pred %p40, %p32; + mov.pred %p39, %p31; + mov.pred %p38, %p30; + mov.pred %p37, %p29; + mov.pred %p36, %p28; + mov.pred %p35, %p27; + mov.pred %p34, %p26; + mov.pred %p33, %p25; + mov.pred %p1, %p24; + .loc 1 82 21 // wy_fast.py:82:21 + add.s32 %r344, %r883, 1; + setp.gt.s32 %p20, %r344, 2; + selp.b32 %r883, 0, %r344, %p20; + .loc 1 86 63 // wy_fast.py:86:63 + mul.wide.s32 %rd157, %r43, 2; + add.s64 %rd141, %rd21, %rd157; + add.s64 %rd142, %rd22, %rd157; + add.s64 %rd143, %rd23, %rd157; + add.s64 %rd144, %rd24, %rd157; + add.s64 %rd145, %rd25, %rd157; + add.s64 %rd146, %rd26, %rd157; + add.s64 %rd147, %rd27, %rd157; + add.s64 %rd148, %rd28, %rd157; + .loc 1 87 22 // wy_fast.py:87:22 + cp.async.wait_group 2; + bar.sync 0; + shl.b32 %r345, %r883, 13; + add.s32 %r346, %r105, %r345; + .loc 1 88 22 // wy_fast.py:88:22 + add.s32 %r347, %r346, %r14; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r348, %r349, %r350, %r351}, [%r347]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r352, %r353, %r354, %r355}, [%r347+4096]; + add.s32 %r356, %r346, %r15; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r357, %r358, %r359, %r360}, [%r356]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r361, %r362, %r363, %r364}, [%r356+4096]; + add.s32 %r365, %r346, %r16; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r366, %r367, %r368, %r369}, [%r365]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r370, %r371, %r372, %r373}, [%r365+4096]; + add.s32 %r374, %r346, %r17; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r375, %r376, %r377, %r378}, [%r374]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r379, %r380, %r381, %r382}, [%r374+4096]; + mul.bf16x2 %r152, %r42, %r348; + mul.bf16x2 %r153, %r41, %r349; + mul.bf16x2 %r196, %r40, %r350; + mul.bf16x2 %r197, %r39, %r351; + mul.bf16x2 %r264, %r38, %r352; + mul.bf16x2 %r265, %r37, %r353; + mul.bf16x2 %r288, %r36, %r354; + mul.bf16x2 %r289, %r35, %r355; + mul.bf16x2 %r154, %r42, %r357; + mul.bf16x2 %r155, %r41, %r358; + mul.bf16x2 %r202, %r40, %r359; + mul.bf16x2 %r203, %r39, %r360; + mul.bf16x2 %r266, %r38, %r361; + mul.bf16x2 %r267, %r37, %r362; + mul.bf16x2 %r290, %r36, %r363; + mul.bf16x2 %r291, %r35, %r364; + mul.bf16x2 %r156, %r42, %r366; + mul.bf16x2 %r157, %r41, %r367; + mul.bf16x2 %r208, %r40, %r368; + mul.bf16x2 %r209, %r39, %r369; + mul.bf16x2 %r268, %r38, %r370; + mul.bf16x2 %r269, %r37, %r371; + mul.bf16x2 %r292, %r36, %r372; + mul.bf16x2 %r293, %r35, %r373; + mul.bf16x2 %r158, %r42, %r375; + mul.bf16x2 %r159, %r41, %r376; + mul.bf16x2 %r214, %r40, %r377; + mul.bf16x2 %r215, %r39, %r378; + mul.bf16x2 %r270, %r38, %r379; + mul.bf16x2 %r271, %r37, %r380; + mul.bf16x2 %r294, %r36, %r381; + mul.bf16x2 %r295, %r35, %r382; + .loc 1 80 18 // wy_fast.py:80:18 + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r144, %r145, %r146, %r147}, [%r18]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r148, %r149, %r150, %r151}, [%r19]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r160, %r161, %r162, %r163}, [%r20]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r164, %r165, %r166, %r167}, [%r21]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r172, %r173, %r174, %r175}, [%r22]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r192, %r193, %r194, %r195}, [%r23]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r220, %r221, %r222, %r223}, [%r24]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r240, %r241, %r242, %r243}, [%r25]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r256, %r257, %r258, %r259}, [%r26]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r260, %r261, %r262, %r263}, [%r27]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r272, %r273, %r274, %r275}, [%r28]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r276, %r277, %r278, %r279}, [%r29]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r280, %r281, %r282, %r283}, [%r30]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r284, %r285, %r286, %r287}, [%r31]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r296, %r297, %r298, %r299}, [%r32]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r300, %r301, %r302, %r303}, [%r33]; + .loc 1 89 26 // wy_fast.py:89:26 + mov.b32 %r168, %r46; + mov.b32 %r169, %r46; + mov.b32 %r170, %r46; + mov.b32 %r171, %r46; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r168, %r169, %r170, %r171 }, { %r144, %r145, %r146, %r147 }, { %r152, %r153 }, { %r168, %r169, %r170, %r171 }; + // end inline asm + mov.b32 %r176, %r46; + mov.b32 %r177, %r46; + mov.b32 %r178, %r46; + mov.b32 %r179, %r46; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r176, %r177, %r178, %r179 }, { %r144, %r145, %r146, %r147 }, { %r154, %r155 }, { %r176, %r177, %r178, %r179 }; + // end inline asm + mov.b32 %r180, %r46; + mov.b32 %r181, %r46; + mov.b32 %r182, %r46; + mov.b32 %r183, %r46; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r180, %r181, %r182, %r183 }, { %r144, %r145, %r146, %r147 }, { %r156, %r157 }, { %r180, %r181, %r182, %r183 }; + // end inline asm + mov.b32 %r184, %r46; + mov.b32 %r185, %r46; + mov.b32 %r186, %r46; + mov.b32 %r187, %r46; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r184, %r185, %r186, %r187 }, { %r144, %r145, %r146, %r147 }, { %r158, %r159 }, { %r184, %r185, %r186, %r187 }; + // end inline asm + mov.b32 %r188, %r46; + mov.b32 %r189, %r46; + mov.b32 %r190, %r46; + mov.b32 %r191, %r46; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r188, %r189, %r190, %r191 }, { %r148, %r149, %r150, %r151 }, { %r152, %r153 }, { %r188, %r189, %r190, %r191 }; + // end inline asm + mov.b32 %r198, %r46; + mov.b32 %r199, %r46; + mov.b32 %r200, %r46; + mov.b32 %r201, %r46; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r198, %r199, %r200, %r201 }, { %r148, %r149, %r150, %r151 }, { %r154, %r155 }, { %r198, %r199, %r200, %r201 }; + // end inline asm + mov.b32 %r204, %r46; + mov.b32 %r205, %r46; + mov.b32 %r206, %r46; + mov.b32 %r207, %r46; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r204, %r205, %r206, %r207 }, { %r148, %r149, %r150, %r151 }, { %r156, %r157 }, { %r204, %r205, %r206, %r207 }; + // end inline asm + mov.b32 %r210, %r46; + mov.b32 %r211, %r46; + mov.b32 %r212, %r46; + mov.b32 %r213, %r46; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r210, %r211, %r212, %r213 }, { %r148, %r149, %r150, %r151 }, { %r158, %r159 }, { %r210, %r211, %r212, %r213 }; + // end inline asm + mov.b32 %r216, %r46; + mov.b32 %r217, %r46; + mov.b32 %r218, %r46; + mov.b32 %r219, %r46; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r216, %r217, %r218, %r219 }, { %r160, %r161, %r162, %r163 }, { %r152, %r153 }, { %r216, %r217, %r218, %r219 }; + // end inline asm + mov.b32 %r224, %r46; + mov.b32 %r225, %r46; + mov.b32 %r226, %r46; + mov.b32 %r227, %r46; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r224, %r225, %r226, %r227 }, { %r160, %r161, %r162, %r163 }, { %r154, %r155 }, { %r224, %r225, %r226, %r227 }; + // end inline asm + mov.b32 %r228, %r46; + mov.b32 %r229, %r46; + mov.b32 %r230, %r46; + mov.b32 %r231, %r46; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r228, %r229, %r230, %r231 }, { %r160, %r161, %r162, %r163 }, { %r156, %r157 }, { %r228, %r229, %r230, %r231 }; + // end inline asm + mov.b32 %r232, %r46; + mov.b32 %r233, %r46; + mov.b32 %r234, %r46; + mov.b32 %r235, %r46; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r232, %r233, %r234, %r235 }, { %r160, %r161, %r162, %r163 }, { %r158, %r159 }, { %r232, %r233, %r234, %r235 }; + // end inline asm + mov.b32 %r236, %r46; + mov.b32 %r237, %r46; + mov.b32 %r238, %r46; + mov.b32 %r239, %r46; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r236, %r237, %r238, %r239 }, { %r164, %r165, %r166, %r167 }, { %r152, %r153 }, { %r236, %r237, %r238, %r239 }; + // end inline asm + mov.b32 %r244, %r46; + mov.b32 %r245, %r46; + mov.b32 %r246, %r46; + mov.b32 %r247, %r46; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r244, %r245, %r246, %r247 }, { %r164, %r165, %r166, %r167 }, { %r154, %r155 }, { %r244, %r245, %r246, %r247 }; + // end inline asm + mov.b32 %r248, %r46; + mov.b32 %r249, %r46; + mov.b32 %r250, %r46; + mov.b32 %r251, %r46; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r248, %r249, %r250, %r251 }, { %r164, %r165, %r166, %r167 }, { %r156, %r157 }, { %r248, %r249, %r250, %r251 }; + // end inline asm + mov.b32 %r252, %r46; + mov.b32 %r253, %r46; + mov.b32 %r254, %r46; + mov.b32 %r255, %r46; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r252, %r253, %r254, %r255 }, { %r164, %r165, %r166, %r167 }, { %r158, %r159 }, { %r252, %r253, %r254, %r255 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r168, %r169, %r170, %r171 }, { %r172, %r173, %r174, %r175 }, { %r196, %r197 }, { %r168, %r169, %r170, %r171 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r176, %r177, %r178, %r179 }, { %r172, %r173, %r174, %r175 }, { %r202, %r203 }, { %r176, %r177, %r178, %r179 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r180, %r181, %r182, %r183 }, { %r172, %r173, %r174, %r175 }, { %r208, %r209 }, { %r180, %r181, %r182, %r183 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r184, %r185, %r186, %r187 }, { %r172, %r173, %r174, %r175 }, { %r214, %r215 }, { %r184, %r185, %r186, %r187 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r188, %r189, %r190, %r191 }, { %r192, %r193, %r194, %r195 }, { %r196, %r197 }, { %r188, %r189, %r190, %r191 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r198, %r199, %r200, %r201 }, { %r192, %r193, %r194, %r195 }, { %r202, %r203 }, { %r198, %r199, %r200, %r201 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r204, %r205, %r206, %r207 }, { %r192, %r193, %r194, %r195 }, { %r208, %r209 }, { %r204, %r205, %r206, %r207 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r210, %r211, %r212, %r213 }, { %r192, %r193, %r194, %r195 }, { %r214, %r215 }, { %r210, %r211, %r212, %r213 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r216, %r217, %r218, %r219 }, { %r220, %r221, %r222, %r223 }, { %r196, %r197 }, { %r216, %r217, %r218, %r219 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r224, %r225, %r226, %r227 }, { %r220, %r221, %r222, %r223 }, { %r202, %r203 }, { %r224, %r225, %r226, %r227 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r228, %r229, %r230, %r231 }, { %r220, %r221, %r222, %r223 }, { %r208, %r209 }, { %r228, %r229, %r230, %r231 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r232, %r233, %r234, %r235 }, { %r220, %r221, %r222, %r223 }, { %r214, %r215 }, { %r232, %r233, %r234, %r235 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r236, %r237, %r238, %r239 }, { %r240, %r241, %r242, %r243 }, { %r196, %r197 }, { %r236, %r237, %r238, %r239 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r244, %r245, %r246, %r247 }, { %r240, %r241, %r242, %r243 }, { %r202, %r203 }, { %r244, %r245, %r246, %r247 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r248, %r249, %r250, %r251 }, { %r240, %r241, %r242, %r243 }, { %r208, %r209 }, { %r248, %r249, %r250, %r251 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r252, %r253, %r254, %r255 }, { %r240, %r241, %r242, %r243 }, { %r214, %r215 }, { %r252, %r253, %r254, %r255 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r168, %r169, %r170, %r171 }, { %r256, %r257, %r258, %r259 }, { %r264, %r265 }, { %r168, %r169, %r170, %r171 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r176, %r177, %r178, %r179 }, { %r256, %r257, %r258, %r259 }, { %r266, %r267 }, { %r176, %r177, %r178, %r179 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r180, %r181, %r182, %r183 }, { %r256, %r257, %r258, %r259 }, { %r268, %r269 }, { %r180, %r181, %r182, %r183 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r184, %r185, %r186, %r187 }, { %r256, %r257, %r258, %r259 }, { %r270, %r271 }, { %r184, %r185, %r186, %r187 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r188, %r189, %r190, %r191 }, { %r260, %r261, %r262, %r263 }, { %r264, %r265 }, { %r188, %r189, %r190, %r191 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r198, %r199, %r200, %r201 }, { %r260, %r261, %r262, %r263 }, { %r266, %r267 }, { %r198, %r199, %r200, %r201 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r204, %r205, %r206, %r207 }, { %r260, %r261, %r262, %r263 }, { %r268, %r269 }, { %r204, %r205, %r206, %r207 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r210, %r211, %r212, %r213 }, { %r260, %r261, %r262, %r263 }, { %r270, %r271 }, { %r210, %r211, %r212, %r213 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r216, %r217, %r218, %r219 }, { %r272, %r273, %r274, %r275 }, { %r264, %r265 }, { %r216, %r217, %r218, %r219 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r224, %r225, %r226, %r227 }, { %r272, %r273, %r274, %r275 }, { %r266, %r267 }, { %r224, %r225, %r226, %r227 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r228, %r229, %r230, %r231 }, { %r272, %r273, %r274, %r275 }, { %r268, %r269 }, { %r228, %r229, %r230, %r231 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r232, %r233, %r234, %r235 }, { %r272, %r273, %r274, %r275 }, { %r270, %r271 }, { %r232, %r233, %r234, %r235 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r236, %r237, %r238, %r239 }, { %r276, %r277, %r278, %r279 }, { %r264, %r265 }, { %r236, %r237, %r238, %r239 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r244, %r245, %r246, %r247 }, { %r276, %r277, %r278, %r279 }, { %r266, %r267 }, { %r244, %r245, %r246, %r247 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r248, %r249, %r250, %r251 }, { %r276, %r277, %r278, %r279 }, { %r268, %r269 }, { %r248, %r249, %r250, %r251 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r252, %r253, %r254, %r255 }, { %r276, %r277, %r278, %r279 }, { %r270, %r271 }, { %r252, %r253, %r254, %r255 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r168, %r169, %r170, %r171 }, { %r280, %r281, %r282, %r283 }, { %r288, %r289 }, { %r168, %r169, %r170, %r171 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r176, %r177, %r178, %r179 }, { %r280, %r281, %r282, %r283 }, { %r290, %r291 }, { %r176, %r177, %r178, %r179 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r180, %r181, %r182, %r183 }, { %r280, %r281, %r282, %r283 }, { %r292, %r293 }, { %r180, %r181, %r182, %r183 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r184, %r185, %r186, %r187 }, { %r280, %r281, %r282, %r283 }, { %r294, %r295 }, { %r184, %r185, %r186, %r187 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r188, %r189, %r190, %r191 }, { %r284, %r285, %r286, %r287 }, { %r288, %r289 }, { %r188, %r189, %r190, %r191 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r198, %r199, %r200, %r201 }, { %r284, %r285, %r286, %r287 }, { %r290, %r291 }, { %r198, %r199, %r200, %r201 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r204, %r205, %r206, %r207 }, { %r284, %r285, %r286, %r287 }, { %r292, %r293 }, { %r204, %r205, %r206, %r207 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r210, %r211, %r212, %r213 }, { %r284, %r285, %r286, %r287 }, { %r294, %r295 }, { %r210, %r211, %r212, %r213 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r216, %r217, %r218, %r219 }, { %r296, %r297, %r298, %r299 }, { %r288, %r289 }, { %r216, %r217, %r218, %r219 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r224, %r225, %r226, %r227 }, { %r296, %r297, %r298, %r299 }, { %r290, %r291 }, { %r224, %r225, %r226, %r227 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r228, %r229, %r230, %r231 }, { %r296, %r297, %r298, %r299 }, { %r292, %r293 }, { %r228, %r229, %r230, %r231 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r232, %r233, %r234, %r235 }, { %r296, %r297, %r298, %r299 }, { %r294, %r295 }, { %r232, %r233, %r234, %r235 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r236, %r237, %r238, %r239 }, { %r300, %r301, %r302, %r303 }, { %r288, %r289 }, { %r236, %r237, %r238, %r239 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r244, %r245, %r246, %r247 }, { %r300, %r301, %r302, %r303 }, { %r290, %r291 }, { %r244, %r245, %r246, %r247 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r248, %r249, %r250, %r251 }, { %r300, %r301, %r302, %r303 }, { %r292, %r293 }, { %r248, %r249, %r250, %r251 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r252, %r253, %r254, %r255 }, { %r300, %r301, %r302, %r303 }, { %r294, %r295 }, { %r252, %r253, %r254, %r255 }; + // end inline asm + .loc 1 90 29 // wy_fast.py:90:29 + cvt.rn.bf16x2.f32 %r383, %r169, %r168; + cvt.rn.bf16x2.f32 %r384, %r171, %r170; + cvt.rn.bf16x2.f32 %r385, %r177, %r176; + cvt.rn.bf16x2.f32 %r386, %r179, %r178; + cvt.rn.bf16x2.f32 %r387, %r181, %r180; + cvt.rn.bf16x2.f32 %r388, %r183, %r182; + cvt.rn.bf16x2.f32 %r389, %r185, %r184; + cvt.rn.bf16x2.f32 %r390, %r187, %r186; + cvt.rn.bf16x2.f32 %r391, %r189, %r188; + cvt.rn.bf16x2.f32 %r392, %r191, %r190; + cvt.rn.bf16x2.f32 %r393, %r199, %r198; + cvt.rn.bf16x2.f32 %r394, %r201, %r200; + cvt.rn.bf16x2.f32 %r395, %r205, %r204; + cvt.rn.bf16x2.f32 %r396, %r207, %r206; + cvt.rn.bf16x2.f32 %r397, %r211, %r210; + cvt.rn.bf16x2.f32 %r398, %r213, %r212; + cvt.rn.bf16x2.f32 %r399, %r217, %r216; + cvt.rn.bf16x2.f32 %r400, %r219, %r218; + cvt.rn.bf16x2.f32 %r401, %r225, %r224; + cvt.rn.bf16x2.f32 %r402, %r227, %r226; + cvt.rn.bf16x2.f32 %r403, %r229, %r228; + cvt.rn.bf16x2.f32 %r404, %r231, %r230; + cvt.rn.bf16x2.f32 %r405, %r233, %r232; + cvt.rn.bf16x2.f32 %r406, %r235, %r234; + cvt.rn.bf16x2.f32 %r407, %r237, %r236; + cvt.rn.bf16x2.f32 %r408, %r239, %r238; + cvt.rn.bf16x2.f32 %r409, %r245, %r244; + cvt.rn.bf16x2.f32 %r410, %r247, %r246; + cvt.rn.bf16x2.f32 %r411, %r249, %r248; + cvt.rn.bf16x2.f32 %r412, %r251, %r250; + cvt.rn.bf16x2.f32 %r413, %r253, %r252; + cvt.rn.bf16x2.f32 %r414, %r255, %r254; + .loc 1 90 22 // wy_fast.py:90:22 + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r34], {%r383, %r385, %r387, %r389}; + bar.sync 0; + ld.shared.v4.b32 {%r304, %r305, %r306, %r307}, [%r5+24576]; + bar.sync 0; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r34], {%r384, %r386, %r388, %r390}; + bar.sync 0; + ld.shared.v4.b32 {%r308, %r309, %r310, %r311}, [%r5+24576]; + bar.sync 0; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r34], {%r391, %r393, %r395, %r397}; + bar.sync 0; + ld.shared.v4.b32 {%r312, %r313, %r314, %r315}, [%r5+24576]; + bar.sync 0; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r34], {%r392, %r394, %r396, %r398}; + bar.sync 0; + ld.shared.v4.b32 {%r316, %r317, %r318, %r319}, [%r5+24576]; + bar.sync 0; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r34], {%r399, %r401, %r403, %r405}; + bar.sync 0; + ld.shared.v4.b32 {%r320, %r321, %r322, %r323}, [%r5+24576]; + bar.sync 0; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r34], {%r400, %r402, %r404, %r406}; + bar.sync 0; + ld.shared.v4.b32 {%r324, %r325, %r326, %r327}, [%r5+24576]; + bar.sync 0; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r34], {%r407, %r409, %r411, %r413}; + bar.sync 0; + ld.shared.v4.b32 {%r328, %r329, %r330, %r331}, [%r5+24576]; + bar.sync 0; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r34], {%r408, %r410, %r412, %r414}; + bar.sync 0; + ld.shared.v4.b32 {%r332, %r333, %r334, %r335}, [%r5+24576]; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd141 + 0 ], { %r304, %r305, %r306, %r307 }; + // end inline asm + // begin inline asm + @%p3 st.global.v4.b32 [ %rd142 + 0 ], { %r308, %r309, %r310, %r311 }; + // end inline asm + // begin inline asm + @%p4 st.global.v4.b32 [ %rd143 + 0 ], { %r312, %r313, %r314, %r315 }; + // end inline asm + // begin inline asm + @%p5 st.global.v4.b32 [ %rd144 + 0 ], { %r316, %r317, %r318, %r319 }; + // end inline asm + // begin inline asm + @%p6 st.global.v4.b32 [ %rd145 + 0 ], { %r320, %r321, %r322, %r323 }; + // end inline asm + // begin inline asm + @%p7 st.global.v4.b32 [ %rd146 + 0 ], { %r324, %r325, %r326, %r327 }; + // end inline asm + // begin inline asm + @%p8 st.global.v4.b32 [ %rd147 + 0 ], { %r328, %r329, %r330, %r331 }; + // end inline asm + // begin inline asm + @%p9 st.global.v4.b32 [ %rd148 + 0 ], { %r332, %r333, %r334, %r335 }; + // end inline asm + .loc 1 82 21 // wy_fast.py:82:21 + add.s32 %r415, %r884, 1; + setp.gt.s32 %p21, %r415, 2; + selp.b32 %r884, 0, %r415, %p21; + .loc 1 83 25 // wy_fast.py:83:25 + or.b64 %rd158, %rd283, %rd12; + .loc 1 85 63 // wy_fast.py:85:63 + cvt.u32.u64 %r880, %rd158; + shl.b64 %rd159, %rd158, 1; + add.s64 %rd149, %rd13, %rd159; + add.s64 %rd150, %rd14, %rd159; + add.s64 %rd151, %rd15, %rd159; + add.s64 %rd152, %rd16, %rd159; + add.s64 %rd153, %rd17, %rd159; + add.s64 %rd154, %rd18, %rd159; + add.s64 %rd155, %rd19, %rd159; + add.s64 %rd156, %rd20, %rd159; + .loc 1 87 22 // wy_fast.py:87:22 + shl.b32 %r416, %r884, 13; + add.s32 %r336, %r5, %r416; + // begin inline asm + cp.async.cg.shared.global [ %r336 + 0 ], [ %rd149 + 0 ], 0x10, %r46; + // end inline asm + add.s32 %r337, %r336, 1024; + // begin inline asm + cp.async.cg.shared.global [ %r337 + 0 ], [ %rd150 + 0 ], 0x10, %r46; + // end inline asm + add.s32 %r338, %r336, 2048; + // begin inline asm + cp.async.cg.shared.global [ %r338 + 0 ], [ %rd151 + 0 ], 0x10, %r46; + // end inline asm + add.s32 %r339, %r336, 3072; + // begin inline asm + cp.async.cg.shared.global [ %r339 + 0 ], [ %rd152 + 0 ], 0x10, %r46; + // end inline asm + add.s32 %r340, %r336, 4096; + // begin inline asm + cp.async.cg.shared.global [ %r340 + 0 ], [ %rd153 + 0 ], 0x10, %r46; + // end inline asm + add.s32 %r341, %r336, 5120; + // begin inline asm + cp.async.cg.shared.global [ %r341 + 0 ], [ %rd154 + 0 ], 0x10, %r46; + // end inline asm + add.s32 %r342, %r336, 6144; + // begin inline asm + cp.async.cg.shared.global [ %r342 + 0 ], [ %rd155 + 0 ], 0x10, %r46; + // end inline asm + add.s32 %r343, %r336, 7168; + // begin inline asm + cp.async.cg.shared.global [ %r343 + 0 ], [ %rd156 + 0 ], 0x10, %r46; + // end inline asm + cp.async.commit_group; + mov.b64 %rd283, 256; + mov.pred %p24, 0; + mov.pred %p25, %p24; + mov.pred %p26, %p24; + mov.pred %p27, %p24; + mov.pred %p28, %p24; + mov.pred %p29, %p24; + mov.pred %p30, %p24; + mov.pred %p31, %p24; + mov.pred %p32, %p24; + .loc 1 82 21 // wy_fast.py:82:21 + @%p1 bra $L__BB0_1; +// %bb.2: // %__nv_exp2f.exit + cp.async.wait_group 0; + bar.sync 0; + .loc 1 93 19 // wy_fast.py:93:19 + shl.b64 %rd185, %rd1, 2; + add.s64 %rd186, %rd55, %rd185; + .loc 1 93 35 // wy_fast.py:93:35 + shl.b64 %rd187, %rd2, 2; + add.s64 %rd160, %rd186, %rd187; + mov.b32 %r418, 0; + .loc 1 94 27 // wy_fast.py:94:27 + // begin inline asm + mov.u32 %r417, %r418; + @%p19 ld.global.b32 { %r417 }, [ %rd160 + 0 ]; + // end inline asm +$L__tmp1: + .loc 2 47 33 // op.py:47:33 @[ wy_fast.py:94:19 ] + ex2.approx.ftz.f32 %r420, %r417; + and.b32 %r444, %r3, 132; + add.s32 %r445, %r105, %r1; + add.s32 %r446, %r445, %r444; + add.s32 %r419, %r446, %r2; + // begin inline asm + st.shared.b32 [ %r419 + 0 ], %r420; + // end inline asm + bar.sync 0; + ld.shared.v2.b64 {%rd29, %rd30}, [%r4]; + ld.shared.v2.b64 {%rd31, %rd32}, [%r4+16]; + ld.shared.v2.b64 {%rd33, %rd34}, [%r4+128]; + ld.shared.v2.b64 {%rd35, %rd36}, [%r4+144]; +$L__tmp2: + .loc 1 99 18 // wy_fast.py:99:18 + shl.b64 %rd188, %rd3, 1; + add.s64 %rd189, %rd53, %rd188; + .loc 1 99 51 // wy_fast.py:99:51 + shl.b64 %rd190, %rd4, 1; + add.s64 %rd37, %rd189, %rd190; + shl.b64 %rd191, %rd5, 1; + add.s64 %rd38, %rd189, %rd191; + shl.b64 %rd192, %rd6, 1; + add.s64 %rd39, %rd189, %rd192; + shl.b64 %rd193, %rd7, 1; + add.s64 %rd40, %rd189, %rd193; + shl.b64 %rd194, %rd8, 1; + add.s64 %rd41, %rd189, %rd194; + shl.b64 %rd195, %rd9, 1; + add.s64 %rd42, %rd189, %rd195; + shl.b64 %rd196, %rd10, 1; + add.s64 %rd43, %rd189, %rd196; + shl.b64 %rd197, %rd11, 1; + add.s64 %rd44, %rd189, %rd197; + .loc 1 100 18 // wy_fast.py:100:18 + add.s64 %rd198, %rd54, %rd188; + .loc 1 100 39 // wy_fast.py:100:39 + add.s64 %rd45, %rd198, %rd190; + add.s64 %rd46, %rd198, %rd191; + add.s64 %rd47, %rd198, %rd192; + add.s64 %rd48, %rd198, %rd193; + add.s64 %rd49, %rd198, %rd194; + add.s64 %rd50, %rd198, %rd195; + add.s64 %rd51, %rd198, %rd196; + add.s64 %rd52, %rd198, %rd197; + .loc 1 99 74 // wy_fast.py:99:74 + shl.b64 %rd199, %rd12, 1; + add.s64 %rd161, %rd37, %rd199; + add.s64 %rd162, %rd38, %rd199; + add.s64 %rd163, %rd39, %rd199; + add.s64 %rd164, %rd40, %rd199; + add.s64 %rd165, %rd41, %rd199; + add.s64 %rd166, %rd42, %rd199; + add.s64 %rd167, %rd43, %rd199; + add.s64 %rd168, %rd44, %rd199; + .loc 1 101 22 // wy_fast.py:101:22 + bar.sync 0; + // begin inline asm + cp.async.cg.shared.global [ %r5 + 0 ], [ %rd161 + 0 ], 0x10, %r6; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r421 + 0 ], [ %rd162 + 0 ], 0x10, %r7; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r422 + 0 ], [ %rd163 + 0 ], 0x10, %r8; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r423 + 0 ], [ %rd164 + 0 ], 0x10, %r9; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r424 + 0 ], [ %rd165 + 0 ], 0x10, %r10; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r425 + 0 ], [ %rd166 + 0 ], 0x10, %r11; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r426 + 0 ], [ %rd167 + 0 ], 0x10, %r12; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r427 + 0 ], [ %rd168 + 0 ], 0x10, %r13; + // end inline asm + cp.async.commit_group; + .loc 1 99 74 // wy_fast.py:99:74 + add.s64 %rd169, %rd161, 128; + add.s64 %rd170, %rd162, 128; + add.s64 %rd171, %rd163, 128; + add.s64 %rd172, %rd164, 128; + add.s64 %rd173, %rd165, 128; + add.s64 %rd174, %rd166, 128; + add.s64 %rd175, %rd167, 128; + add.s64 %rd176, %rd168, 128; + .loc 1 101 22 // wy_fast.py:101:22 + bar.sync 0; + // begin inline asm + cp.async.cg.shared.global [ %r428 + 0 ], [ %rd169 + 0 ], 0x10, %r6; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r429 + 0 ], [ %rd170 + 0 ], 0x10, %r7; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r430 + 0 ], [ %rd171 + 0 ], 0x10, %r8; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r431 + 0 ], [ %rd172 + 0 ], 0x10, %r9; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r432 + 0 ], [ %rd173 + 0 ], 0x10, %r10; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r433 + 0 ], [ %rd174 + 0 ], 0x10, %r11; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r434 + 0 ], [ %rd175 + 0 ], 0x10, %r12; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r435 + 0 ], [ %rd176 + 0 ], 0x10, %r13; + // end inline asm + cp.async.commit_group; + .loc 1 99 74 // wy_fast.py:99:74 + add.s64 %rd177, %rd161, 256; + add.s64 %rd178, %rd162, 256; + add.s64 %rd179, %rd163, 256; + add.s64 %rd180, %rd164, 256; + add.s64 %rd181, %rd165, 256; + add.s64 %rd182, %rd166, 256; + add.s64 %rd183, %rd167, 256; + add.s64 %rd184, %rd168, 256; + .loc 1 101 22 // wy_fast.py:101:22 + bar.sync 0; + // begin inline asm + cp.async.cg.shared.global [ %r436 + 0 ], [ %rd177 + 0 ], 0x10, %r418; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r437 + 0 ], [ %rd178 + 0 ], 0x10, %r418; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r438 + 0 ], [ %rd179 + 0 ], 0x10, %r418; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r439 + 0 ], [ %rd180 + 0 ], 0x10, %r418; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r440 + 0 ], [ %rd181 + 0 ], 0x10, %r418; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r441 + 0 ], [ %rd182 + 0 ], 0x10, %r418; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r442 + 0 ], [ %rd183 + 0 ], 0x10, %r418; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r443 + 0 ], [ %rd184 + 0 ], 0x10, %r418; + // end inline asm + cp.async.commit_group; + mov.b32 %r889, 2; + mov.b32 %r888, -1; + mov.b64 %rd284, 192; + mov.pred %p41, -1; + mov.pred %p50, %p42; + mov.pred %p51, %p43; + mov.pred %p52, %p44; + mov.pred %p53, %p45; + mov.pred %p54, %p46; + mov.pred %p55, %p47; + mov.pred %p56, %p48; + mov.pred %p57, %p49; +$L__BB0_3: // =>This Inner Loop Header: Depth=1 + .loc 1 0 22 // wy_fast.py:0:22 + mov.b32 %r44, %r887; + mov.b32 %r887, %r886; + mov.b32 %r886, %r885; + mov.pred %p18, %p57; + mov.pred %p17, %p56; + mov.pred %p16, %p55; + mov.pred %p15, %p54; + mov.pred %p14, %p53; + mov.pred %p13, %p52; + mov.pred %p12, %p51; + mov.pred %p11, %p50; + mov.pred %p57, %p49; + mov.pred %p56, %p48; + mov.pred %p55, %p47; + mov.pred %p54, %p46; + mov.pred %p53, %p45; + mov.pred %p52, %p44; + mov.pred %p51, %p43; + mov.pred %p50, %p42; + mov.pred %p10, %p41; + .loc 1 96 21 // wy_fast.py:96:21 + add.s32 %r647, %r888, 1; + setp.gt.s32 %p22, %r647, 2; + selp.b32 %r888, 0, %r647, %p22; + .loc 1 100 63 // wy_fast.py:100:63 + mul.wide.s32 %rd216, %r44, 2; + add.s64 %rd200, %rd45, %rd216; + add.s64 %rd201, %rd46, %rd216; + add.s64 %rd202, %rd47, %rd216; + add.s64 %rd203, %rd48, %rd216; + add.s64 %rd204, %rd49, %rd216; + add.s64 %rd205, %rd50, %rd216; + add.s64 %rd206, %rd51, %rd216; + add.s64 %rd207, %rd52, %rd216; + .loc 1 101 22 // wy_fast.py:101:22 + cp.async.wait_group 2; + bar.sync 0; + shl.b32 %r648, %r888, 13; + add.s32 %r649, %r105, %r648; + .loc 1 105 34 // wy_fast.py:105:34 + add.s32 %r650, %r649, %r14; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r651, %r652, %r653, %r654}, [%r650]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r655, %r656, %r657, %r658}, [%r650+4096]; + add.s32 %r659, %r649, %r15; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r660, %r661, %r662, %r663}, [%r659]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r664, %r665, %r666, %r667}, [%r659+4096]; + add.s32 %r668, %r649, %r16; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r669, %r670, %r671, %r672}, [%r668]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r673, %r674, %r675, %r676}, [%r668+4096]; + add.s32 %r677, %r649, %r17; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r678, %r679, %r680, %r681}, [%r677]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r682, %r683, %r684, %r685}, [%r677+4096]; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r686, %r42, %r651; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs19, %rs20}, %r686; + cvt.f32.bf16 %r687, %rs20; + cvt.f32.bf16 %r688, %rs19; + mov.b64 %rd217, {%r688, %r687}; + mul.f32x2 %rd218, %rd29, %rd217; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r689, %r690}, %rd218; + cvt.rn.bf16x2.f32 %r455, %r690, %r689; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r691, %r41, %r652; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs21, %rs22}, %r691; + cvt.f32.bf16 %r692, %rs22; + cvt.f32.bf16 %r693, %rs21; + mov.b64 %rd219, {%r693, %r692}; + mul.f32x2 %rd220, %rd30, %rd219; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r694, %r695}, %rd220; + cvt.rn.bf16x2.f32 %r456, %r695, %r694; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r696, %r40, %r653; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs23, %rs24}, %r696; + cvt.f32.bf16 %r697, %rs24; + cvt.f32.bf16 %r698, %rs23; + mov.b64 %rd221, {%r698, %r697}; + mul.f32x2 %rd222, %rd31, %rd221; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r699, %r700}, %rd222; + cvt.rn.bf16x2.f32 %r499, %r700, %r699; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r701, %r39, %r654; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs25, %rs26}, %r701; + cvt.f32.bf16 %r702, %rs26; + cvt.f32.bf16 %r703, %rs25; + mov.b64 %rd223, {%r703, %r702}; + mul.f32x2 %rd224, %rd32, %rd223; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r704, %r705}, %rd224; + cvt.rn.bf16x2.f32 %r500, %r705, %r704; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r706, %r38, %r655; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs27, %rs28}, %r706; + cvt.f32.bf16 %r707, %rs28; + cvt.f32.bf16 %r708, %rs27; + mov.b64 %rd225, {%r708, %r707}; + mul.f32x2 %rd226, %rd33, %rd225; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r709, %r710}, %rd226; + cvt.rn.bf16x2.f32 %r567, %r710, %r709; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r711, %r37, %r656; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs29, %rs30}, %r711; + cvt.f32.bf16 %r712, %rs30; + cvt.f32.bf16 %r713, %rs29; + mov.b64 %rd227, {%r713, %r712}; + mul.f32x2 %rd228, %rd34, %rd227; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r714, %r715}, %rd228; + cvt.rn.bf16x2.f32 %r568, %r715, %r714; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r716, %r36, %r657; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs31, %rs32}, %r716; + cvt.f32.bf16 %r717, %rs32; + cvt.f32.bf16 %r718, %rs31; + mov.b64 %rd229, {%r718, %r717}; + mul.f32x2 %rd230, %rd35, %rd229; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r719, %r720}, %rd230; + cvt.rn.bf16x2.f32 %r591, %r720, %r719; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r721, %r35, %r658; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs33, %rs34}, %r721; + cvt.f32.bf16 %r722, %rs34; + cvt.f32.bf16 %r723, %rs33; + mov.b64 %rd231, {%r723, %r722}; + mul.f32x2 %rd232, %rd36, %rd231; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r724, %r725}, %rd232; + cvt.rn.bf16x2.f32 %r592, %r725, %r724; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r726, %r42, %r660; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs35, %rs36}, %r726; + cvt.f32.bf16 %r727, %rs36; + cvt.f32.bf16 %r728, %rs35; + mov.b64 %rd233, {%r728, %r727}; + mul.f32x2 %rd234, %rd29, %rd233; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r729, %r730}, %rd234; + cvt.rn.bf16x2.f32 %r457, %r730, %r729; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r731, %r41, %r661; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs37, %rs38}, %r731; + cvt.f32.bf16 %r732, %rs38; + cvt.f32.bf16 %r733, %rs37; + mov.b64 %rd235, {%r733, %r732}; + mul.f32x2 %rd236, %rd30, %rd235; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r734, %r735}, %rd236; + cvt.rn.bf16x2.f32 %r458, %r735, %r734; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r736, %r40, %r662; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs39, %rs40}, %r736; + cvt.f32.bf16 %r737, %rs40; + cvt.f32.bf16 %r738, %rs39; + mov.b64 %rd237, {%r738, %r737}; + mul.f32x2 %rd238, %rd31, %rd237; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r739, %r740}, %rd238; + cvt.rn.bf16x2.f32 %r505, %r740, %r739; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r741, %r39, %r663; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs41, %rs42}, %r741; + cvt.f32.bf16 %r742, %rs42; + cvt.f32.bf16 %r743, %rs41; + mov.b64 %rd239, {%r743, %r742}; + mul.f32x2 %rd240, %rd32, %rd239; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r744, %r745}, %rd240; + cvt.rn.bf16x2.f32 %r506, %r745, %r744; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r746, %r38, %r664; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs43, %rs44}, %r746; + cvt.f32.bf16 %r747, %rs44; + cvt.f32.bf16 %r748, %rs43; + mov.b64 %rd241, {%r748, %r747}; + mul.f32x2 %rd242, %rd33, %rd241; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r749, %r750}, %rd242; + cvt.rn.bf16x2.f32 %r569, %r750, %r749; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r751, %r37, %r665; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs45, %rs46}, %r751; + cvt.f32.bf16 %r752, %rs46; + cvt.f32.bf16 %r753, %rs45; + mov.b64 %rd243, {%r753, %r752}; + mul.f32x2 %rd244, %rd34, %rd243; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r754, %r755}, %rd244; + cvt.rn.bf16x2.f32 %r570, %r755, %r754; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r756, %r36, %r666; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs47, %rs48}, %r756; + cvt.f32.bf16 %r757, %rs48; + cvt.f32.bf16 %r758, %rs47; + mov.b64 %rd245, {%r758, %r757}; + mul.f32x2 %rd246, %rd35, %rd245; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r759, %r760}, %rd246; + cvt.rn.bf16x2.f32 %r593, %r760, %r759; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r761, %r35, %r667; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs49, %rs50}, %r761; + cvt.f32.bf16 %r762, %rs50; + cvt.f32.bf16 %r763, %rs49; + mov.b64 %rd247, {%r763, %r762}; + mul.f32x2 %rd248, %rd36, %rd247; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r764, %r765}, %rd248; + cvt.rn.bf16x2.f32 %r594, %r765, %r764; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r766, %r42, %r669; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs51, %rs52}, %r766; + cvt.f32.bf16 %r767, %rs52; + cvt.f32.bf16 %r768, %rs51; + mov.b64 %rd249, {%r768, %r767}; + mul.f32x2 %rd250, %rd29, %rd249; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r769, %r770}, %rd250; + cvt.rn.bf16x2.f32 %r459, %r770, %r769; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r771, %r41, %r670; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs53, %rs54}, %r771; + cvt.f32.bf16 %r772, %rs54; + cvt.f32.bf16 %r773, %rs53; + mov.b64 %rd251, {%r773, %r772}; + mul.f32x2 %rd252, %rd30, %rd251; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r774, %r775}, %rd252; + cvt.rn.bf16x2.f32 %r460, %r775, %r774; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r776, %r40, %r671; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs55, %rs56}, %r776; + cvt.f32.bf16 %r777, %rs56; + cvt.f32.bf16 %r778, %rs55; + mov.b64 %rd253, {%r778, %r777}; + mul.f32x2 %rd254, %rd31, %rd253; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r779, %r780}, %rd254; + cvt.rn.bf16x2.f32 %r511, %r780, %r779; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r781, %r39, %r672; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs57, %rs58}, %r781; + cvt.f32.bf16 %r782, %rs58; + cvt.f32.bf16 %r783, %rs57; + mov.b64 %rd255, {%r783, %r782}; + mul.f32x2 %rd256, %rd32, %rd255; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r784, %r785}, %rd256; + cvt.rn.bf16x2.f32 %r512, %r785, %r784; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r786, %r38, %r673; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs59, %rs60}, %r786; + cvt.f32.bf16 %r787, %rs60; + cvt.f32.bf16 %r788, %rs59; + mov.b64 %rd257, {%r788, %r787}; + mul.f32x2 %rd258, %rd33, %rd257; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r789, %r790}, %rd258; + cvt.rn.bf16x2.f32 %r571, %r790, %r789; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r791, %r37, %r674; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs61, %rs62}, %r791; + cvt.f32.bf16 %r792, %rs62; + cvt.f32.bf16 %r793, %rs61; + mov.b64 %rd259, {%r793, %r792}; + mul.f32x2 %rd260, %rd34, %rd259; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r794, %r795}, %rd260; + cvt.rn.bf16x2.f32 %r572, %r795, %r794; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r796, %r36, %r675; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs63, %rs64}, %r796; + cvt.f32.bf16 %r797, %rs64; + cvt.f32.bf16 %r798, %rs63; + mov.b64 %rd261, {%r798, %r797}; + mul.f32x2 %rd262, %rd35, %rd261; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r799, %r800}, %rd262; + cvt.rn.bf16x2.f32 %r595, %r800, %r799; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r801, %r35, %r676; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs65, %rs66}, %r801; + cvt.f32.bf16 %r802, %rs66; + cvt.f32.bf16 %r803, %rs65; + mov.b64 %rd263, {%r803, %r802}; + mul.f32x2 %rd264, %rd36, %rd263; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r804, %r805}, %rd264; + cvt.rn.bf16x2.f32 %r596, %r805, %r804; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r806, %r42, %r678; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs67, %rs68}, %r806; + cvt.f32.bf16 %r807, %rs68; + cvt.f32.bf16 %r808, %rs67; + mov.b64 %rd265, {%r808, %r807}; + mul.f32x2 %rd266, %rd29, %rd265; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r809, %r810}, %rd266; + cvt.rn.bf16x2.f32 %r461, %r810, %r809; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r811, %r41, %r679; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs69, %rs70}, %r811; + cvt.f32.bf16 %r812, %rs70; + cvt.f32.bf16 %r813, %rs69; + mov.b64 %rd267, {%r813, %r812}; + mul.f32x2 %rd268, %rd30, %rd267; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r814, %r815}, %rd268; + cvt.rn.bf16x2.f32 %r462, %r815, %r814; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r816, %r40, %r680; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs71, %rs72}, %r816; + cvt.f32.bf16 %r817, %rs72; + cvt.f32.bf16 %r818, %rs71; + mov.b64 %rd269, {%r818, %r817}; + mul.f32x2 %rd270, %rd31, %rd269; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r819, %r820}, %rd270; + cvt.rn.bf16x2.f32 %r517, %r820, %r819; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r821, %r39, %r681; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs73, %rs74}, %r821; + cvt.f32.bf16 %r822, %rs74; + cvt.f32.bf16 %r823, %rs73; + mov.b64 %rd271, {%r823, %r822}; + mul.f32x2 %rd272, %rd32, %rd271; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r824, %r825}, %rd272; + cvt.rn.bf16x2.f32 %r518, %r825, %r824; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r826, %r38, %r682; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs75, %rs76}, %r826; + cvt.f32.bf16 %r827, %rs76; + cvt.f32.bf16 %r828, %rs75; + mov.b64 %rd273, {%r828, %r827}; + mul.f32x2 %rd274, %rd33, %rd273; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r829, %r830}, %rd274; + cvt.rn.bf16x2.f32 %r573, %r830, %r829; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r831, %r37, %r683; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs77, %rs78}, %r831; + cvt.f32.bf16 %r832, %rs78; + cvt.f32.bf16 %r833, %rs77; + mov.b64 %rd275, {%r833, %r832}; + mul.f32x2 %rd276, %rd34, %rd275; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r834, %r835}, %rd276; + cvt.rn.bf16x2.f32 %r574, %r835, %r834; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r836, %r36, %r684; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs79, %rs80}, %r836; + cvt.f32.bf16 %r837, %rs80; + cvt.f32.bf16 %r838, %rs79; + mov.b64 %rd277, {%r838, %r837}; + mul.f32x2 %rd278, %rd35, %rd277; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r839, %r840}, %rd278; + cvt.rn.bf16x2.f32 %r597, %r840, %r839; + .loc 1 102 21 // wy_fast.py:102:21 + mul.bf16x2 %r841, %r35, %r685; + .loc 1 104 20 // wy_fast.py:104:20 + mov.b32 {%rs81, %rs82}, %r841; + cvt.f32.bf16 %r842, %rs82; + cvt.f32.bf16 %r843, %rs81; + mov.b64 %rd279, {%r843, %r842}; + mul.f32x2 %rd280, %rd36, %rd279; + .loc 1 105 34 // wy_fast.py:105:34 + mov.b64 {%r844, %r845}, %rd280; + cvt.rn.bf16x2.f32 %r598, %r845, %r844; + .loc 1 80 18 // wy_fast.py:80:18 + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r447, %r448, %r449, %r450}, [%r18]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r451, %r452, %r453, %r454}, [%r19]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r463, %r464, %r465, %r466}, [%r20]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r467, %r468, %r469, %r470}, [%r21]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r475, %r476, %r477, %r478}, [%r22]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r495, %r496, %r497, %r498}, [%r23]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r523, %r524, %r525, %r526}, [%r24]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r543, %r544, %r545, %r546}, [%r25]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r559, %r560, %r561, %r562}, [%r26]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r563, %r564, %r565, %r566}, [%r27]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r575, %r576, %r577, %r578}, [%r28]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r579, %r580, %r581, %r582}, [%r29]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r583, %r584, %r585, %r586}, [%r30]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r587, %r588, %r589, %r590}, [%r31]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r599, %r600, %r601, %r602}, [%r32]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r603, %r604, %r605, %r606}, [%r33]; + .loc 1 105 26 // wy_fast.py:105:26 + mov.b32 %r471, %r418; + mov.b32 %r472, %r418; + mov.b32 %r473, %r418; + mov.b32 %r474, %r418; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r471, %r472, %r473, %r474 }, { %r447, %r448, %r449, %r450 }, { %r455, %r456 }, { %r471, %r472, %r473, %r474 }; + // end inline asm + mov.b32 %r479, %r418; + mov.b32 %r480, %r418; + mov.b32 %r481, %r418; + mov.b32 %r482, %r418; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r479, %r480, %r481, %r482 }, { %r447, %r448, %r449, %r450 }, { %r457, %r458 }, { %r479, %r480, %r481, %r482 }; + // end inline asm + mov.b32 %r483, %r418; + mov.b32 %r484, %r418; + mov.b32 %r485, %r418; + mov.b32 %r486, %r418; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r483, %r484, %r485, %r486 }, { %r447, %r448, %r449, %r450 }, { %r459, %r460 }, { %r483, %r484, %r485, %r486 }; + // end inline asm + mov.b32 %r487, %r418; + mov.b32 %r488, %r418; + mov.b32 %r489, %r418; + mov.b32 %r490, %r418; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r487, %r488, %r489, %r490 }, { %r447, %r448, %r449, %r450 }, { %r461, %r462 }, { %r487, %r488, %r489, %r490 }; + // end inline asm + mov.b32 %r491, %r418; + mov.b32 %r492, %r418; + mov.b32 %r493, %r418; + mov.b32 %r494, %r418; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r491, %r492, %r493, %r494 }, { %r451, %r452, %r453, %r454 }, { %r455, %r456 }, { %r491, %r492, %r493, %r494 }; + // end inline asm + mov.b32 %r501, %r418; + mov.b32 %r502, %r418; + mov.b32 %r503, %r418; + mov.b32 %r504, %r418; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r501, %r502, %r503, %r504 }, { %r451, %r452, %r453, %r454 }, { %r457, %r458 }, { %r501, %r502, %r503, %r504 }; + // end inline asm + mov.b32 %r507, %r418; + mov.b32 %r508, %r418; + mov.b32 %r509, %r418; + mov.b32 %r510, %r418; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r507, %r508, %r509, %r510 }, { %r451, %r452, %r453, %r454 }, { %r459, %r460 }, { %r507, %r508, %r509, %r510 }; + // end inline asm + mov.b32 %r513, %r418; + mov.b32 %r514, %r418; + mov.b32 %r515, %r418; + mov.b32 %r516, %r418; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r513, %r514, %r515, %r516 }, { %r451, %r452, %r453, %r454 }, { %r461, %r462 }, { %r513, %r514, %r515, %r516 }; + // end inline asm + mov.b32 %r519, %r418; + mov.b32 %r520, %r418; + mov.b32 %r521, %r418; + mov.b32 %r522, %r418; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r519, %r520, %r521, %r522 }, { %r463, %r464, %r465, %r466 }, { %r455, %r456 }, { %r519, %r520, %r521, %r522 }; + // end inline asm + mov.b32 %r527, %r418; + mov.b32 %r528, %r418; + mov.b32 %r529, %r418; + mov.b32 %r530, %r418; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r527, %r528, %r529, %r530 }, { %r463, %r464, %r465, %r466 }, { %r457, %r458 }, { %r527, %r528, %r529, %r530 }; + // end inline asm + mov.b32 %r531, %r418; + mov.b32 %r532, %r418; + mov.b32 %r533, %r418; + mov.b32 %r534, %r418; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r531, %r532, %r533, %r534 }, { %r463, %r464, %r465, %r466 }, { %r459, %r460 }, { %r531, %r532, %r533, %r534 }; + // end inline asm + mov.b32 %r535, %r418; + mov.b32 %r536, %r418; + mov.b32 %r537, %r418; + mov.b32 %r538, %r418; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r535, %r536, %r537, %r538 }, { %r463, %r464, %r465, %r466 }, { %r461, %r462 }, { %r535, %r536, %r537, %r538 }; + // end inline asm + mov.b32 %r539, %r418; + mov.b32 %r540, %r418; + mov.b32 %r541, %r418; + mov.b32 %r542, %r418; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r539, %r540, %r541, %r542 }, { %r467, %r468, %r469, %r470 }, { %r455, %r456 }, { %r539, %r540, %r541, %r542 }; + // end inline asm + mov.b32 %r547, %r418; + mov.b32 %r548, %r418; + mov.b32 %r549, %r418; + mov.b32 %r550, %r418; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r547, %r548, %r549, %r550 }, { %r467, %r468, %r469, %r470 }, { %r457, %r458 }, { %r547, %r548, %r549, %r550 }; + // end inline asm + mov.b32 %r551, %r418; + mov.b32 %r552, %r418; + mov.b32 %r553, %r418; + mov.b32 %r554, %r418; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r551, %r552, %r553, %r554 }, { %r467, %r468, %r469, %r470 }, { %r459, %r460 }, { %r551, %r552, %r553, %r554 }; + // end inline asm + mov.b32 %r555, %r418; + mov.b32 %r556, %r418; + mov.b32 %r557, %r418; + mov.b32 %r558, %r418; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r555, %r556, %r557, %r558 }, { %r467, %r468, %r469, %r470 }, { %r461, %r462 }, { %r555, %r556, %r557, %r558 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r471, %r472, %r473, %r474 }, { %r475, %r476, %r477, %r478 }, { %r499, %r500 }, { %r471, %r472, %r473, %r474 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r479, %r480, %r481, %r482 }, { %r475, %r476, %r477, %r478 }, { %r505, %r506 }, { %r479, %r480, %r481, %r482 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r483, %r484, %r485, %r486 }, { %r475, %r476, %r477, %r478 }, { %r511, %r512 }, { %r483, %r484, %r485, %r486 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r487, %r488, %r489, %r490 }, { %r475, %r476, %r477, %r478 }, { %r517, %r518 }, { %r487, %r488, %r489, %r490 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r491, %r492, %r493, %r494 }, { %r495, %r496, %r497, %r498 }, { %r499, %r500 }, { %r491, %r492, %r493, %r494 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r501, %r502, %r503, %r504 }, { %r495, %r496, %r497, %r498 }, { %r505, %r506 }, { %r501, %r502, %r503, %r504 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r507, %r508, %r509, %r510 }, { %r495, %r496, %r497, %r498 }, { %r511, %r512 }, { %r507, %r508, %r509, %r510 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r513, %r514, %r515, %r516 }, { %r495, %r496, %r497, %r498 }, { %r517, %r518 }, { %r513, %r514, %r515, %r516 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r519, %r520, %r521, %r522 }, { %r523, %r524, %r525, %r526 }, { %r499, %r500 }, { %r519, %r520, %r521, %r522 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r527, %r528, %r529, %r530 }, { %r523, %r524, %r525, %r526 }, { %r505, %r506 }, { %r527, %r528, %r529, %r530 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r531, %r532, %r533, %r534 }, { %r523, %r524, %r525, %r526 }, { %r511, %r512 }, { %r531, %r532, %r533, %r534 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r535, %r536, %r537, %r538 }, { %r523, %r524, %r525, %r526 }, { %r517, %r518 }, { %r535, %r536, %r537, %r538 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r539, %r540, %r541, %r542 }, { %r543, %r544, %r545, %r546 }, { %r499, %r500 }, { %r539, %r540, %r541, %r542 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r547, %r548, %r549, %r550 }, { %r543, %r544, %r545, %r546 }, { %r505, %r506 }, { %r547, %r548, %r549, %r550 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r551, %r552, %r553, %r554 }, { %r543, %r544, %r545, %r546 }, { %r511, %r512 }, { %r551, %r552, %r553, %r554 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r555, %r556, %r557, %r558 }, { %r543, %r544, %r545, %r546 }, { %r517, %r518 }, { %r555, %r556, %r557, %r558 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r471, %r472, %r473, %r474 }, { %r559, %r560, %r561, %r562 }, { %r567, %r568 }, { %r471, %r472, %r473, %r474 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r479, %r480, %r481, %r482 }, { %r559, %r560, %r561, %r562 }, { %r569, %r570 }, { %r479, %r480, %r481, %r482 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r483, %r484, %r485, %r486 }, { %r559, %r560, %r561, %r562 }, { %r571, %r572 }, { %r483, %r484, %r485, %r486 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r487, %r488, %r489, %r490 }, { %r559, %r560, %r561, %r562 }, { %r573, %r574 }, { %r487, %r488, %r489, %r490 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r491, %r492, %r493, %r494 }, { %r563, %r564, %r565, %r566 }, { %r567, %r568 }, { %r491, %r492, %r493, %r494 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r501, %r502, %r503, %r504 }, { %r563, %r564, %r565, %r566 }, { %r569, %r570 }, { %r501, %r502, %r503, %r504 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r507, %r508, %r509, %r510 }, { %r563, %r564, %r565, %r566 }, { %r571, %r572 }, { %r507, %r508, %r509, %r510 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r513, %r514, %r515, %r516 }, { %r563, %r564, %r565, %r566 }, { %r573, %r574 }, { %r513, %r514, %r515, %r516 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r519, %r520, %r521, %r522 }, { %r575, %r576, %r577, %r578 }, { %r567, %r568 }, { %r519, %r520, %r521, %r522 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r527, %r528, %r529, %r530 }, { %r575, %r576, %r577, %r578 }, { %r569, %r570 }, { %r527, %r528, %r529, %r530 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r531, %r532, %r533, %r534 }, { %r575, %r576, %r577, %r578 }, { %r571, %r572 }, { %r531, %r532, %r533, %r534 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r535, %r536, %r537, %r538 }, { %r575, %r576, %r577, %r578 }, { %r573, %r574 }, { %r535, %r536, %r537, %r538 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r539, %r540, %r541, %r542 }, { %r579, %r580, %r581, %r582 }, { %r567, %r568 }, { %r539, %r540, %r541, %r542 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r547, %r548, %r549, %r550 }, { %r579, %r580, %r581, %r582 }, { %r569, %r570 }, { %r547, %r548, %r549, %r550 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r551, %r552, %r553, %r554 }, { %r579, %r580, %r581, %r582 }, { %r571, %r572 }, { %r551, %r552, %r553, %r554 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r555, %r556, %r557, %r558 }, { %r579, %r580, %r581, %r582 }, { %r573, %r574 }, { %r555, %r556, %r557, %r558 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r471, %r472, %r473, %r474 }, { %r583, %r584, %r585, %r586 }, { %r591, %r592 }, { %r471, %r472, %r473, %r474 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r479, %r480, %r481, %r482 }, { %r583, %r584, %r585, %r586 }, { %r593, %r594 }, { %r479, %r480, %r481, %r482 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r483, %r484, %r485, %r486 }, { %r583, %r584, %r585, %r586 }, { %r595, %r596 }, { %r483, %r484, %r485, %r486 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r487, %r488, %r489, %r490 }, { %r583, %r584, %r585, %r586 }, { %r597, %r598 }, { %r487, %r488, %r489, %r490 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r491, %r492, %r493, %r494 }, { %r587, %r588, %r589, %r590 }, { %r591, %r592 }, { %r491, %r492, %r493, %r494 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r501, %r502, %r503, %r504 }, { %r587, %r588, %r589, %r590 }, { %r593, %r594 }, { %r501, %r502, %r503, %r504 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r507, %r508, %r509, %r510 }, { %r587, %r588, %r589, %r590 }, { %r595, %r596 }, { %r507, %r508, %r509, %r510 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r513, %r514, %r515, %r516 }, { %r587, %r588, %r589, %r590 }, { %r597, %r598 }, { %r513, %r514, %r515, %r516 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r519, %r520, %r521, %r522 }, { %r599, %r600, %r601, %r602 }, { %r591, %r592 }, { %r519, %r520, %r521, %r522 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r527, %r528, %r529, %r530 }, { %r599, %r600, %r601, %r602 }, { %r593, %r594 }, { %r527, %r528, %r529, %r530 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r531, %r532, %r533, %r534 }, { %r599, %r600, %r601, %r602 }, { %r595, %r596 }, { %r531, %r532, %r533, %r534 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r535, %r536, %r537, %r538 }, { %r599, %r600, %r601, %r602 }, { %r597, %r598 }, { %r535, %r536, %r537, %r538 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r539, %r540, %r541, %r542 }, { %r603, %r604, %r605, %r606 }, { %r591, %r592 }, { %r539, %r540, %r541, %r542 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r547, %r548, %r549, %r550 }, { %r603, %r604, %r605, %r606 }, { %r593, %r594 }, { %r547, %r548, %r549, %r550 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r551, %r552, %r553, %r554 }, { %r603, %r604, %r605, %r606 }, { %r595, %r596 }, { %r551, %r552, %r553, %r554 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r555, %r556, %r557, %r558 }, { %r603, %r604, %r605, %r606 }, { %r597, %r598 }, { %r555, %r556, %r557, %r558 }; + // end inline asm + .loc 1 106 29 // wy_fast.py:106:29 + cvt.rn.bf16x2.f32 %r846, %r472, %r471; + cvt.rn.bf16x2.f32 %r847, %r474, %r473; + cvt.rn.bf16x2.f32 %r848, %r480, %r479; + cvt.rn.bf16x2.f32 %r849, %r482, %r481; + cvt.rn.bf16x2.f32 %r850, %r484, %r483; + cvt.rn.bf16x2.f32 %r851, %r486, %r485; + cvt.rn.bf16x2.f32 %r852, %r488, %r487; + cvt.rn.bf16x2.f32 %r853, %r490, %r489; + cvt.rn.bf16x2.f32 %r854, %r492, %r491; + cvt.rn.bf16x2.f32 %r855, %r494, %r493; + cvt.rn.bf16x2.f32 %r856, %r502, %r501; + cvt.rn.bf16x2.f32 %r857, %r504, %r503; + cvt.rn.bf16x2.f32 %r858, %r508, %r507; + cvt.rn.bf16x2.f32 %r859, %r510, %r509; + cvt.rn.bf16x2.f32 %r860, %r514, %r513; + cvt.rn.bf16x2.f32 %r861, %r516, %r515; + cvt.rn.bf16x2.f32 %r862, %r520, %r519; + cvt.rn.bf16x2.f32 %r863, %r522, %r521; + cvt.rn.bf16x2.f32 %r864, %r528, %r527; + cvt.rn.bf16x2.f32 %r865, %r530, %r529; + cvt.rn.bf16x2.f32 %r866, %r532, %r531; + cvt.rn.bf16x2.f32 %r867, %r534, %r533; + cvt.rn.bf16x2.f32 %r868, %r536, %r535; + cvt.rn.bf16x2.f32 %r869, %r538, %r537; + cvt.rn.bf16x2.f32 %r870, %r540, %r539; + cvt.rn.bf16x2.f32 %r871, %r542, %r541; + cvt.rn.bf16x2.f32 %r872, %r548, %r547; + cvt.rn.bf16x2.f32 %r873, %r550, %r549; + cvt.rn.bf16x2.f32 %r874, %r552, %r551; + cvt.rn.bf16x2.f32 %r875, %r554, %r553; + cvt.rn.bf16x2.f32 %r876, %r556, %r555; + cvt.rn.bf16x2.f32 %r877, %r558, %r557; + .loc 1 106 22 // wy_fast.py:106:22 + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r34], {%r846, %r848, %r850, %r852}; + bar.sync 0; + ld.shared.v4.b32 {%r607, %r608, %r609, %r610}, [%r5+24576]; + bar.sync 0; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r34], {%r847, %r849, %r851, %r853}; + bar.sync 0; + ld.shared.v4.b32 {%r611, %r612, %r613, %r614}, [%r5+24576]; + bar.sync 0; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r34], {%r854, %r856, %r858, %r860}; + bar.sync 0; + ld.shared.v4.b32 {%r615, %r616, %r617, %r618}, [%r5+24576]; + bar.sync 0; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r34], {%r855, %r857, %r859, %r861}; + bar.sync 0; + ld.shared.v4.b32 {%r619, %r620, %r621, %r622}, [%r5+24576]; + bar.sync 0; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r34], {%r862, %r864, %r866, %r868}; + bar.sync 0; + ld.shared.v4.b32 {%r623, %r624, %r625, %r626}, [%r5+24576]; + bar.sync 0; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r34], {%r863, %r865, %r867, %r869}; + bar.sync 0; + ld.shared.v4.b32 {%r627, %r628, %r629, %r630}, [%r5+24576]; + bar.sync 0; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r34], {%r870, %r872, %r874, %r876}; + bar.sync 0; + ld.shared.v4.b32 {%r631, %r632, %r633, %r634}, [%r5+24576]; + bar.sync 0; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r34], {%r871, %r873, %r875, %r877}; + bar.sync 0; + ld.shared.v4.b32 {%r635, %r636, %r637, %r638}, [%r5+24576]; + // begin inline asm + @%p11 st.global.v4.b32 [ %rd200 + 0 ], { %r607, %r608, %r609, %r610 }; + // end inline asm + // begin inline asm + @%p12 st.global.v4.b32 [ %rd201 + 0 ], { %r611, %r612, %r613, %r614 }; + // end inline asm + // begin inline asm + @%p13 st.global.v4.b32 [ %rd202 + 0 ], { %r615, %r616, %r617, %r618 }; + // end inline asm + // begin inline asm + @%p14 st.global.v4.b32 [ %rd203 + 0 ], { %r619, %r620, %r621, %r622 }; + // end inline asm + // begin inline asm + @%p15 st.global.v4.b32 [ %rd204 + 0 ], { %r623, %r624, %r625, %r626 }; + // end inline asm + // begin inline asm + @%p16 st.global.v4.b32 [ %rd205 + 0 ], { %r627, %r628, %r629, %r630 }; + // end inline asm + // begin inline asm + @%p17 st.global.v4.b32 [ %rd206 + 0 ], { %r631, %r632, %r633, %r634 }; + // end inline asm + // begin inline asm + @%p18 st.global.v4.b32 [ %rd207 + 0 ], { %r635, %r636, %r637, %r638 }; + // end inline asm + .loc 1 96 21 // wy_fast.py:96:21 + add.s32 %r878, %r889, 1; + setp.gt.s32 %p23, %r878, 2; + selp.b32 %r889, 0, %r878, %p23; + .loc 1 97 25 // wy_fast.py:97:25 + or.b64 %rd281, %rd284, %rd12; + .loc 1 99 74 // wy_fast.py:99:74 + cvt.u32.u64 %r885, %rd281; + shl.b64 %rd282, %rd281, 1; + add.s64 %rd208, %rd37, %rd282; + add.s64 %rd209, %rd38, %rd282; + add.s64 %rd210, %rd39, %rd282; + add.s64 %rd211, %rd40, %rd282; + add.s64 %rd212, %rd41, %rd282; + add.s64 %rd213, %rd42, %rd282; + add.s64 %rd214, %rd43, %rd282; + add.s64 %rd215, %rd44, %rd282; + .loc 1 101 22 // wy_fast.py:101:22 + shl.b32 %r879, %r889, 13; + add.s32 %r639, %r5, %r879; + // begin inline asm + cp.async.cg.shared.global [ %r639 + 0 ], [ %rd208 + 0 ], 0x10, %r418; + // end inline asm + add.s32 %r640, %r639, 1024; + // begin inline asm + cp.async.cg.shared.global [ %r640 + 0 ], [ %rd209 + 0 ], 0x10, %r418; + // end inline asm + add.s32 %r641, %r639, 2048; + // begin inline asm + cp.async.cg.shared.global [ %r641 + 0 ], [ %rd210 + 0 ], 0x10, %r418; + // end inline asm + add.s32 %r642, %r639, 3072; + // begin inline asm + cp.async.cg.shared.global [ %r642 + 0 ], [ %rd211 + 0 ], 0x10, %r418; + // end inline asm + add.s32 %r643, %r639, 4096; + // begin inline asm + cp.async.cg.shared.global [ %r643 + 0 ], [ %rd212 + 0 ], 0x10, %r418; + // end inline asm + add.s32 %r644, %r639, 5120; + // begin inline asm + cp.async.cg.shared.global [ %r644 + 0 ], [ %rd213 + 0 ], 0x10, %r418; + // end inline asm + add.s32 %r645, %r639, 6144; + // begin inline asm + cp.async.cg.shared.global [ %r645 + 0 ], [ %rd214 + 0 ], 0x10, %r418; + // end inline asm + add.s32 %r646, %r639, 7168; + // begin inline asm + cp.async.cg.shared.global [ %r646 + 0 ], [ %rd215 + 0 ], 0x10, %r418; + // end inline asm + cp.async.commit_group; + mov.b64 %rd284, 256; + mov.pred %p41, 0; + mov.pred %p42, %p41; + mov.pred %p43, %p41; + mov.pred %p44, %p41; + mov.pred %p45, %p41; + mov.pred %p46, %p41; + mov.pred %p47, %p41; + mov.pred %p48, %p41; + mov.pred %p49, %p41; + .loc 1 96 21 // wy_fast.py:96:21 + @%p10 bra $L__BB0_3; +// %bb.4: + cp.async.wait_group 0; + bar.sync 0; + .loc 1 96 4 // wy_fast.py:96:4 + ret; +$L__tmp3: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 186 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xb3 DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 119 // DW_AT_name +.b8 121 +.b8 95 +.b8 102 +.b8 97 +.b8 115 +.b8 116 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 111 +.b8 112 +.b8 115 +.b8 47 +.b8 103 +.b8 97 +.b8 116 +.b8 101 +.b8 100 +.b8 95 +.b8 100 +.b8 101 +.b8 108 +.b8 116 +.b8 97 +.b8 95 +.b8 114 +.b8 117 +.b8 108 +.b8 101 +.b8 0 +.b8 2 // Abbrev [2] 0x74:0x1b DW_TAG_subprogram +.b8 114 // DW_AT_name +.b8 101 +.b8 99 +.b8 111 +.b8 109 +.b8 112 +.b8 117 +.b8 116 +.b8 101 +.b8 95 +.b8 119 +.b8 95 +.b8 117 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x8f:0x2e DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 116 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0xa4:0x18 DW_TAG_inlined_subroutine +.b32 116 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp2 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 94 // DW_AT_call_line +.b8 19 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.source b/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..cd868e779d6d2210f35f43955df407019fa40f9f --- /dev/null +++ b/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.source @@ -0,0 +1,522 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":43:0) +#loc116 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:0) +#loc120 = loc("k"(#loc)) +#loc121 = loc("v"(#loc)) +#loc122 = loc("beta"(#loc)) +#loc123 = loc("w"(#loc)) +#loc124 = loc("u"(#loc)) +#loc125 = loc("A"(#loc)) +#loc126 = loc("g"(#loc)) +#loc127 = loc("T"(#loc)) +#loc222 = loc("x"(#loc116)) +module { + tt.func public @recompute_w_u_fwd_kernel(%k: !tt.ptr {tt.divisibility = 16 : i32} loc("k"(#loc)), %v: !tt.ptr {tt.divisibility = 16 : i32} loc("v"(#loc)), %beta: !tt.ptr {tt.divisibility = 16 : i32} loc("beta"(#loc)), %w: !tt.ptr {tt.divisibility = 16 : i32} loc("w"(#loc)), %u: !tt.ptr {tt.divisibility = 16 : i32} loc("u"(#loc)), %A: !tt.ptr {tt.divisibility = 16 : i32} loc("A"(#loc)), %g: !tt.ptr {tt.divisibility = 16 : i32} loc("g"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %0 = tt.get_program_id x : i32 loc(#loc1) + %1 = arith.extsi %0 : i32 to i64 loc(#loc2) + %2 = tt.get_program_id y : i32 loc(#loc3) + %3 = arith.extsi %2 : i32 to i64 loc(#loc4) + %c32_i32 = arith.constant 32 : i32 loc(#loc5) + %c32_i64 = arith.constant 32 : i64 loc(#loc5) + %4 = arith.divsi %3, %c32_i64 : i64 loc(#loc5) + %c32_i32_0 = arith.constant 32 : i32 loc(#loc6) + %c32_i64_1 = arith.constant 32 : i64 loc(#loc6) + %5 = arith.remsi %3, %c32_i64_1 : i64 loc(#loc6) + %6 = arith.extsi %T : i32 to i64 loc(#loc7) + %7 = arith.muli %4, %6 : i64 loc(#loc7) + %8 = arith.extsi %T : i32 to i64 loc(#loc8) + %9 = arith.muli %4, %8 : i64 loc(#loc8) + %10 = arith.extsi %T : i32 to i64 loc(#loc9) + %11 = arith.addi %9, %10 : i64 loc(#loc9) + %o_t = arith.constant 64 : i32 loc(#loc128) + %o_t_2 = arith.constant 64 : i64 loc(#loc128) + %o_t_3 = arith.muli %1, %o_t_2 : i64 loc(#loc128) + %o_t_4 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc129) + %o_t_5 = arith.extsi %o_t_4 : tensor<64xi32> to tensor<64xi64> loc(#loc130) + %o_t_6 = tt.splat %o_t_3 : i64 -> tensor<64xi64> loc(#loc130) + %o_t_7 = arith.addi %o_t_6, %o_t_5 : tensor<64xi64> loc(#loc130) + %o_A = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc131) + %m_t = arith.extsi %T : i32 to i64 loc(#loc132) + %m_t_8 = tt.splat %m_t : i64 -> tensor<64xi64> loc(#loc132) + %m_t_9 = arith.cmpi slt, %o_t_7, %m_t_8 : tensor<64xi64> loc(#loc132) + %m_A = tt.expand_dims %m_t_9 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc133) + %m_A_10 = tt.expand_dims %o_A {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc134) + %m_A_11 = arith.constant 64 : i32 loc(#loc135) + %m_A_12 = arith.constant dense<64> : tensor<1x64xi32> loc(#loc135) + %m_A_13 = arith.cmpi slt, %m_A_10, %m_A_12 : tensor<1x64xi32> loc(#loc135) + %m_A_14 = tt.broadcast %m_A : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc136) + %m_A_15 = tt.broadcast %m_A_13 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc136) + %m_A_16 = arith.andi %m_A_14, %m_A_15 : tensor<64x64xi1> loc(#loc136) + %p_b = arith.constant 32 : i32 loc(#loc137) + %p_b_17 = arith.constant 32 : i64 loc(#loc137) + %p_b_18 = arith.muli %7, %p_b_17 : i64 loc(#loc137) + %p_b_19 = tt.addptr %beta, %p_b_18 : !tt.ptr, i64 loc(#loc138) + %p_b_20 = tt.addptr %p_b_19, %5 : !tt.ptr, i64 loc(#loc139) + %p_b_21 = arith.constant 32 : i32 loc(#loc140) + %p_b_22 = arith.constant 32 : i64 loc(#loc140) + %p_b_23 = arith.constant dense<32> : tensor<64xi64> loc(#loc140) + %p_b_24 = arith.muli %o_t_7, %p_b_23 : tensor<64xi64> loc(#loc140) + %p_b_25 = tt.splat %p_b_20 : !tt.ptr -> tensor<64x!tt.ptr> loc(#loc141) + %p_b_26 = tt.addptr %p_b_25, %p_b_24 : tensor<64x!tt.ptr>, tensor<64xi64> loc(#loc141) + %b_b = arith.constant 0.000000e+00 : f32 loc(#loc142) + %b_b_27 = arith.constant dense<0.000000e+00> : tensor<64xf32> loc(#loc142) + %b_b_28 = arith.truncf %b_b_27 : tensor<64xf32> to tensor<64xbf16> loc(#loc142) + %b_b_29 = tt.load %p_b_26, %m_t_9, %b_b_28 : tensor<64x!tt.ptr> loc(#loc142) + %p_A = arith.constant 32 : i32 loc(#loc143) + %p_A_30 = arith.constant 32 : i64 loc(#loc143) + %p_A_31 = arith.muli %7, %p_A_30 : i64 loc(#loc143) + %p_A_32 = arith.addi %p_A_31, %5 : i64 loc(#loc144) + %p_A_33 = arith.constant 64 : i32 loc(#loc145) + %p_A_34 = arith.constant 64 : i64 loc(#loc145) + %p_A_35 = arith.muli %p_A_32, %p_A_34 : i64 loc(#loc145) + %p_A_36 = tt.addptr %A, %p_A_35 : !tt.ptr, i64 loc(#loc146) + %p_A_37 = tt.expand_dims %o_t_7 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc147) + %p_A_38 = arith.constant 2048 : i32 loc(#loc148) + %p_A_39 = arith.constant 2048 : i64 loc(#loc148) + %p_A_40 = arith.constant dense<2048> : tensor<64x1xi64> loc(#loc148) + %p_A_41 = arith.muli %p_A_37, %p_A_40 : tensor<64x1xi64> loc(#loc148) + %p_A_42 = tt.splat %p_A_36 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc149) + %p_A_43 = tt.addptr %p_A_42, %p_A_41 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc149) + %p_A_44 = tt.expand_dims %o_A {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc150) + %p_A_45 = tt.broadcast %p_A_43 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc151) + %p_A_46 = tt.broadcast %p_A_44 : tensor<1x64xi32> -> tensor<64x64xi32> loc(#loc151) + %p_A_47 = tt.addptr %p_A_45, %p_A_46 : tensor<64x64x!tt.ptr>, tensor<64x64xi32> loc(#loc151) + %b_A = arith.constant 0.000000e+00 : f32 loc(#loc152) + %b_A_48 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc152) + %b_A_49 = arith.truncf %b_A_48 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc152) + %b_A_50 = tt.load %p_A_47, %m_A_16, %b_A_49 : tensor<64x64x!tt.ptr> loc(#loc152) + tt.call @triton.language.standard.cdiv__c128_c64() : () -> () loc(#loc35) + tt.call @triton.language.standard.cdiv__c128_c64() : () -> () loc(#loc35) + %c0_i32 = arith.constant 0 : i32 loc(#loc36) + %c2_i32 = arith.constant 2 : i32 loc(#loc36) + %c1_i32 = arith.constant 1 : i32 loc(#loc36) + %12 = arith.bitcast %c0_i32 : i32 to i32 loc(#loc36) + %13 = arith.bitcast %c2_i32 : i32 to i32 loc(#loc36) + %14 = arith.bitcast %c1_i32 : i32 to i32 loc(#loc36) + %15 = ub.poison : i32 loc(#loc36) + scf.for %i_v = %12 to %13 step %14 : i32 { + %o_v = arith.constant 64 : i32 loc(#loc153) + %o_v_67 = arith.constant 64 : i32 loc(#loc153) + %o_v_68 = arith.extsi %i_v : i32 to i64 loc(#loc153) + %o_v_69 = arith.extsi %o_v_67 : i32 to i64 loc(#loc153) + %o_v_70 = arith.muli %o_v_68, %o_v_69 : i64 loc(#loc153) + %o_v_71 = arith.constant 2147483647 : i64 loc(#loc153) + %o_v_72 = arith.constant -2147483648 : i64 loc(#loc153) + %o_v_73 = arith.cmpi sle, %o_v_70, %o_v_71 : i64 loc(#loc153) + %o_v_74 = arith.cmpi sge, %o_v_70, %o_v_72 : i64 loc(#loc153) + %o_v_75 = arith.andi %o_v_73, %o_v_74 : i1 loc(#loc153) + %o_v_76 = arith.muli %i_v, %o_v_67 : i32 loc(#loc153) + %o_v_77 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc154) + %o_v_78 = tt.splat %o_v_76 : i32 -> tensor<64xi32> loc(#loc155) + %o_v_79 = arith.extsi %o_v_78 : tensor<64xi32> to tensor<64xi64> loc(#loc155) + %o_v_80 = arith.extsi %o_v_77 : tensor<64xi32> to tensor<64xi64> loc(#loc155) + %o_v_81 = arith.addi %o_v_79, %o_v_80 : tensor<64xi64> loc(#loc155) + %o_v_82 = arith.constant 2147483647 : i64 loc(#loc155) + %o_v_83 = arith.constant -2147483648 : i64 loc(#loc155) + %o_v_84 = arith.constant dense<2147483647> : tensor<64xi64> loc(#loc155) + %o_v_85 = arith.cmpi sle, %o_v_81, %o_v_84 : tensor<64xi64> loc(#loc155) + %o_v_86 = arith.constant dense<-2147483648> : tensor<64xi64> loc(#loc155) + %o_v_87 = arith.cmpi sge, %o_v_81, %o_v_86 : tensor<64xi64> loc(#loc155) + %o_v_88 = arith.andi %o_v_85, %o_v_87 : tensor<64xi1> loc(#loc155) + %o_v_89 = arith.addi %o_v_78, %o_v_77 : tensor<64xi32> loc(#loc155) + %m_v = tt.expand_dims %m_t_9 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc156) + %m_v_90 = tt.expand_dims %o_v_89 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc157) + %m_v_91 = arith.constant 128 : i32 loc(#loc158) + %m_v_92 = arith.constant dense<128> : tensor<1x64xi32> loc(#loc158) + %m_v_93 = arith.cmpi slt, %m_v_90, %m_v_92 : tensor<1x64xi32> loc(#loc158) + %m_v_94 = tt.broadcast %m_v : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc159) + %m_v_95 = tt.broadcast %m_v_93 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc159) + %m_v_96 = arith.andi %m_v_94, %m_v_95 : tensor<64x64xi1> loc(#loc159) + %p_v = arith.constant 32 : i32 loc(#loc160) + %p_v_97 = arith.constant 32 : i64 loc(#loc160) + %p_v_98 = arith.muli %7, %p_v_97 : i64 loc(#loc160) + %p_v_99 = arith.addi %p_v_98, %5 : i64 loc(#loc161) + %p_v_100 = arith.constant 128 : i32 loc(#loc162) + %p_v_101 = arith.constant 128 : i64 loc(#loc162) + %p_v_102 = arith.muli %p_v_99, %p_v_101 : i64 loc(#loc162) + %p_v_103 = tt.addptr %v, %p_v_102 : !tt.ptr, i64 loc(#loc163) + %p_v_104 = tt.expand_dims %o_t_7 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc164) + %p_v_105 = arith.constant 4096 : i32 loc(#loc165) + %p_v_106 = arith.constant 4096 : i64 loc(#loc165) + %p_v_107 = arith.constant dense<4096> : tensor<64x1xi64> loc(#loc165) + %p_v_108 = arith.muli %p_v_104, %p_v_107 : tensor<64x1xi64> loc(#loc165) + %p_v_109 = tt.splat %p_v_103 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc166) + %p_v_110 = tt.addptr %p_v_109, %p_v_108 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc166) + %p_v_111 = tt.expand_dims %o_v_89 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc167) + %p_v_112 = tt.broadcast %p_v_110 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc168) + %p_v_113 = tt.broadcast %p_v_111 : tensor<1x64xi32> -> tensor<64x64xi32> loc(#loc168) + %p_v_114 = tt.addptr %p_v_112, %p_v_113 : tensor<64x64x!tt.ptr>, tensor<64x64xi32> loc(#loc168) + %p_u = arith.constant 32 : i32 loc(#loc169) + %p_u_115 = arith.constant 32 : i64 loc(#loc169) + %p_u_116 = arith.muli %7, %p_u_115 : i64 loc(#loc169) + %p_u_117 = arith.addi %p_u_116, %5 : i64 loc(#loc170) + %p_u_118 = arith.constant 128 : i32 loc(#loc171) + %p_u_119 = arith.constant 128 : i64 loc(#loc171) + %p_u_120 = arith.muli %p_u_117, %p_u_119 : i64 loc(#loc171) + %p_u_121 = tt.addptr %u, %p_u_120 : !tt.ptr, i64 loc(#loc172) + %p_u_122 = tt.expand_dims %o_t_7 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc173) + %p_u_123 = arith.constant 4096 : i32 loc(#loc174) + %p_u_124 = arith.constant 4096 : i64 loc(#loc174) + %p_u_125 = arith.constant dense<4096> : tensor<64x1xi64> loc(#loc174) + %p_u_126 = arith.muli %p_u_122, %p_u_125 : tensor<64x1xi64> loc(#loc174) + %p_u_127 = tt.splat %p_u_121 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc175) + %p_u_128 = tt.addptr %p_u_127, %p_u_126 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc175) + %p_u_129 = tt.expand_dims %o_v_89 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc176) + %p_u_130 = tt.broadcast %p_u_128 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc177) + %p_u_131 = tt.broadcast %p_u_129 : tensor<1x64xi32> -> tensor<64x64xi32> loc(#loc177) + %p_u_132 = tt.addptr %p_u_130, %p_u_131 : tensor<64x64x!tt.ptr>, tensor<64x64xi32> loc(#loc177) + %b_v = arith.constant 0.000000e+00 : f32 loc(#loc178) + %b_v_133 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc178) + %b_v_134 = arith.truncf %b_v_133 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc178) + %b_v_135 = tt.load %p_v_114, %m_v_96, %b_v_134 : tensor<64x64x!tt.ptr> loc(#loc178) + %b_vb = tt.expand_dims %b_b_29 {axis = 1 : i32} : tensor<64xbf16> -> tensor<64x1xbf16> loc(#loc179) + %b_vb_136 = tt.broadcast %b_vb : tensor<64x1xbf16> -> tensor<64x64xbf16> loc(#loc180) + %b_vb_137 = arith.mulf %b_v_135, %b_vb_136 : tensor<64x64xbf16> loc(#loc180) + %b_u = arith.constant 0.000000e+00 : f32 loc(#loc181) + %b_u_138 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc181) + %b_u_139 = tt.dot %b_A_50, %b_vb_137, %b_u_138 : tensor<64x64xbf16> * tensor<64x64xbf16> -> tensor<64x64xf32> loc(#loc181) + %20 = arith.truncf %b_u_139 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc66) + tt.store %p_u_132, %20, %m_v_96 : tensor<64x64x!tt.ptr> loc(#loc67) + } loc(#loc36) + %p_g = arith.constant 32 : i32 loc(#loc182) + %p_g_51 = arith.constant 32 : i64 loc(#loc182) + %p_g_52 = arith.muli %7, %p_g_51 : i64 loc(#loc182) + %p_g_53 = arith.addi %p_g_52, %5 : i64 loc(#loc183) + %p_g_54 = tt.addptr %g, %p_g_53 : !tt.ptr, i64 loc(#loc184) + %p_g_55 = arith.constant 32 : i32 loc(#loc185) + %p_g_56 = arith.constant 32 : i64 loc(#loc185) + %p_g_57 = arith.constant dense<32> : tensor<64xi64> loc(#loc185) + %p_g_58 = arith.muli %o_t_7, %p_g_57 : tensor<64xi64> loc(#loc185) + %p_g_59 = tt.splat %p_g_54 : !tt.ptr -> tensor<64x!tt.ptr> loc(#loc186) + %p_g_60 = tt.addptr %p_g_59, %p_g_58 : tensor<64x!tt.ptr>, tensor<64xi64> loc(#loc186) + %b_g = arith.constant 0.000000e+00 : f32 loc(#loc187) + %b_g_61 = arith.constant dense<0.000000e+00> : tensor<64xf32> loc(#loc187) + %b_g_62 = tt.load %p_g_60, %m_t_9, %b_g_61 : tensor<64x!tt.ptr> loc(#loc187) + %b_g_63 = tt.call @fla.ops.utils.op.exp2__fp32S64S(%b_g_62) : (tensor<64xf32>) -> tensor<64xf32> loc(#loc188) + tt.call @triton.language.standard.cdiv__c128_c64() : () -> () loc(#loc75) + tt.call @triton.language.standard.cdiv__c128_c64() : () -> () loc(#loc75) + %c0_i32_64 = arith.constant 0 : i32 loc(#loc76) + %c2_i32_65 = arith.constant 2 : i32 loc(#loc76) + %c1_i32_66 = arith.constant 1 : i32 loc(#loc76) + %16 = arith.bitcast %c0_i32_64 : i32 to i32 loc(#loc76) + %17 = arith.bitcast %c2_i32_65 : i32 to i32 loc(#loc76) + %18 = arith.bitcast %c1_i32_66 : i32 to i32 loc(#loc76) + %19 = ub.poison : i32 loc(#loc76) + scf.for %i_k = %16 to %17 step %18 : i32 { + %o_k = arith.constant 64 : i32 loc(#loc189) + %o_k_67 = arith.constant 64 : i32 loc(#loc189) + %o_k_68 = arith.extsi %i_k : i32 to i64 loc(#loc189) + %o_k_69 = arith.extsi %o_k_67 : i32 to i64 loc(#loc189) + %o_k_70 = arith.muli %o_k_68, %o_k_69 : i64 loc(#loc189) + %o_k_71 = arith.constant 2147483647 : i64 loc(#loc189) + %o_k_72 = arith.constant -2147483648 : i64 loc(#loc189) + %o_k_73 = arith.cmpi sle, %o_k_70, %o_k_71 : i64 loc(#loc189) + %o_k_74 = arith.cmpi sge, %o_k_70, %o_k_72 : i64 loc(#loc189) + %o_k_75 = arith.andi %o_k_73, %o_k_74 : i1 loc(#loc189) + %o_k_76 = arith.muli %i_k, %o_k_67 : i32 loc(#loc189) + %o_k_77 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc190) + %o_k_78 = tt.splat %o_k_76 : i32 -> tensor<64xi32> loc(#loc191) + %o_k_79 = arith.extsi %o_k_78 : tensor<64xi32> to tensor<64xi64> loc(#loc191) + %o_k_80 = arith.extsi %o_k_77 : tensor<64xi32> to tensor<64xi64> loc(#loc191) + %o_k_81 = arith.addi %o_k_79, %o_k_80 : tensor<64xi64> loc(#loc191) + %o_k_82 = arith.constant 2147483647 : i64 loc(#loc191) + %o_k_83 = arith.constant -2147483648 : i64 loc(#loc191) + %o_k_84 = arith.constant dense<2147483647> : tensor<64xi64> loc(#loc191) + %o_k_85 = arith.cmpi sle, %o_k_81, %o_k_84 : tensor<64xi64> loc(#loc191) + %o_k_86 = arith.constant dense<-2147483648> : tensor<64xi64> loc(#loc191) + %o_k_87 = arith.cmpi sge, %o_k_81, %o_k_86 : tensor<64xi64> loc(#loc191) + %o_k_88 = arith.andi %o_k_85, %o_k_87 : tensor<64xi1> loc(#loc191) + %o_k_89 = arith.addi %o_k_78, %o_k_77 : tensor<64xi32> loc(#loc191) + %m_k = tt.expand_dims %m_t_9 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc192) + %m_k_90 = tt.expand_dims %o_k_89 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc193) + %m_k_91 = arith.constant 128 : i32 loc(#loc194) + %m_k_92 = arith.constant dense<128> : tensor<1x64xi32> loc(#loc194) + %m_k_93 = arith.cmpi slt, %m_k_90, %m_k_92 : tensor<1x64xi32> loc(#loc194) + %m_k_94 = tt.broadcast %m_k : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc195) + %m_k_95 = tt.broadcast %m_k_93 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc195) + %m_k_96 = arith.andi %m_k_94, %m_k_95 : tensor<64x64xi1> loc(#loc195) + %p_k = arith.constant 32 : i32 loc(#loc196) + %p_k_97 = arith.constant 32 : i64 loc(#loc196) + %p_k_98 = arith.muli %7, %p_k_97 : i64 loc(#loc196) + %p_k_99 = arith.constant 1 : i32 loc(#loc197) + %p_k_100 = arith.constant 1 : i64 loc(#loc197) + %p_k_101 = arith.divsi %5, %p_k_100 : i64 loc(#loc197) + %p_k_102 = arith.addi %p_k_98, %p_k_101 : i64 loc(#loc198) + %p_k_103 = arith.constant 128 : i32 loc(#loc199) + %p_k_104 = arith.constant 128 : i64 loc(#loc199) + %p_k_105 = arith.muli %p_k_102, %p_k_104 : i64 loc(#loc199) + %p_k_106 = tt.addptr %k, %p_k_105 : !tt.ptr, i64 loc(#loc200) + %p_k_107 = tt.expand_dims %o_t_7 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc201) + %p_k_108 = arith.constant 4096 : i32 loc(#loc202) + %p_k_109 = arith.constant 4096 : i64 loc(#loc202) + %p_k_110 = arith.constant dense<4096> : tensor<64x1xi64> loc(#loc202) + %p_k_111 = arith.muli %p_k_107, %p_k_110 : tensor<64x1xi64> loc(#loc202) + %p_k_112 = tt.splat %p_k_106 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc203) + %p_k_113 = tt.addptr %p_k_112, %p_k_111 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc203) + %p_k_114 = tt.expand_dims %o_k_89 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc204) + %p_k_115 = tt.broadcast %p_k_113 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc205) + %p_k_116 = tt.broadcast %p_k_114 : tensor<1x64xi32> -> tensor<64x64xi32> loc(#loc205) + %p_k_117 = tt.addptr %p_k_115, %p_k_116 : tensor<64x64x!tt.ptr>, tensor<64x64xi32> loc(#loc205) + %p_w = arith.constant 32 : i32 loc(#loc206) + %p_w_118 = arith.constant 32 : i64 loc(#loc206) + %p_w_119 = arith.muli %7, %p_w_118 : i64 loc(#loc206) + %p_w_120 = arith.addi %p_w_119, %5 : i64 loc(#loc207) + %p_w_121 = arith.constant 128 : i32 loc(#loc208) + %p_w_122 = arith.constant 128 : i64 loc(#loc208) + %p_w_123 = arith.muli %p_w_120, %p_w_122 : i64 loc(#loc208) + %p_w_124 = tt.addptr %w, %p_w_123 : !tt.ptr, i64 loc(#loc209) + %p_w_125 = tt.expand_dims %o_t_7 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc210) + %p_w_126 = arith.constant 4096 : i32 loc(#loc211) + %p_w_127 = arith.constant 4096 : i64 loc(#loc211) + %p_w_128 = arith.constant dense<4096> : tensor<64x1xi64> loc(#loc211) + %p_w_129 = arith.muli %p_w_125, %p_w_128 : tensor<64x1xi64> loc(#loc211) + %p_w_130 = tt.splat %p_w_124 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc212) + %p_w_131 = tt.addptr %p_w_130, %p_w_129 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc212) + %p_w_132 = tt.expand_dims %o_k_89 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc213) + %p_w_133 = tt.broadcast %p_w_131 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc214) + %p_w_134 = tt.broadcast %p_w_132 : tensor<1x64xi32> -> tensor<64x64xi32> loc(#loc214) + %p_w_135 = tt.addptr %p_w_133, %p_w_134 : tensor<64x64x!tt.ptr>, tensor<64x64xi32> loc(#loc214) + %b_k = arith.constant 0.000000e+00 : f32 loc(#loc215) + %b_k_136 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc215) + %b_k_137 = arith.truncf %b_k_136 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc215) + %b_k_138 = tt.load %p_k_117, %m_k_96, %b_k_137 : tensor<64x64x!tt.ptr> loc(#loc215) + %b_kb = tt.expand_dims %b_b_29 {axis = 1 : i32} : tensor<64xbf16> -> tensor<64x1xbf16> loc(#loc216) + %b_kb_139 = tt.broadcast %b_kb : tensor<64x1xbf16> -> tensor<64x64xbf16> loc(#loc217) + %b_kb_140 = arith.mulf %b_k_138, %b_kb_139 : tensor<64x64xbf16> loc(#loc217) + %b_kb_141 = tt.expand_dims %b_g_63 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc218) + %b_kb_142 = arith.extf %b_kb_140 : tensor<64x64xbf16> to tensor<64x64xf32> loc(#loc219) + %b_kb_143 = tt.broadcast %b_kb_141 : tensor<64x1xf32> -> tensor<64x64xf32> loc(#loc219) + %b_kb_144 = arith.mulf %b_kb_142, %b_kb_143 : tensor<64x64xf32> loc(#loc219) + %b_w = arith.truncf %b_kb_144 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc220) + %b_w_145 = arith.constant 0.000000e+00 : f32 loc(#loc221) + %b_w_146 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc221) + %b_w_147 = tt.dot %b_A_50, %b_w, %b_w_146, inputPrecision = tf32 : tensor<64x64xbf16> * tensor<64x64xbf16> -> tensor<64x64xf32> loc(#loc221) + %20 = arith.truncf %b_w_147 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc110) + tt.store %p_w_135, %20, %m_k_96 : tensor<64x64x!tt.ptr> loc(#loc111) + } loc(#loc76) + tt.return loc(#loc112) + } loc(#loc) + tt.func private @triton.language.standard.cdiv__c128_c64() attributes {noinline = false} { + tt.return loc(#loc114) + ^bb1: // no predecessors + tt.return loc(#loc115) + } loc(#loc113) + tt.func private @fla.ops.utils.op.exp2__fp32S64S(%x: tensor<64xf32> loc("x"(#loc116))) -> tensor<64xf32> attributes {noinline = false} { + %0 = math.exp2 %x : tensor<64xf32> loc(#loc117) + tt.return %0 : tensor<64xf32> loc(#loc118) + ^bb1: // no predecessors + %1 = ub.poison : tensor<64xf32> loc(#loc119) + tt.return %1 : tensor<64xf32> loc(#loc119) + } loc(#loc116) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":64:30) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":64:36) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":64:61) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":64:67) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":65:23) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":65:34) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":71:25) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":71:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":71:38) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":72:16) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":72:34) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":72:21) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":73:23) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":74:16) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":75:14) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":75:30) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":75:41) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":75:26) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":76:21) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":76:17) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":76:26) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":76:38) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":76:32) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":77:18) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":79:19) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":79:24) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":79:31) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":79:14) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":79:40) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":79:52) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":79:36) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":79:65) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":79:61) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":80:18) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":82:32) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":82:21) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":83:20) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":83:38) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":83:25) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":84:18) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":84:34) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":84:45) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":84:30) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":85:23) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":85:28) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":85:35) +#loc47 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":85:18) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":85:43) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":85:55) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":85:39) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":85:67) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":85:63) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":86:23) +#loc54 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":86:28) +#loc55 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":86:35) +#loc56 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":86:18) +#loc57 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":86:43) +#loc58 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":86:55) +#loc59 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":86:39) +#loc60 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":86:67) +#loc61 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":86:63) +#loc62 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":87:22) +#loc63 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":88:26) +#loc64 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":88:22) +#loc65 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":89:26) +#loc66 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":90:29) +#loc67 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":90:22) +#loc68 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":93:23) +#loc69 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":93:28) +#loc70 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":93:19) +#loc71 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":93:41) +#loc72 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":93:35) +#loc73 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":94:27) +#loc74 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":94:19) +#loc75 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":96:32) +#loc76 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":96:21) +#loc77 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":97:20) +#loc78 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":97:38) +#loc79 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":97:25) +#loc80 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":98:18) +#loc81 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":98:34) +#loc82 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":98:45) +#loc83 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":98:30) +#loc84 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":99:23) +#loc85 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":99:35) +#loc86 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":99:27) +#loc87 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":99:47) +#loc88 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":99:18) +#loc89 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":99:55) +#loc90 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":99:67) +#loc91 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":99:51) +#loc92 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":99:78) +#loc93 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":99:74) +#loc94 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":100:23) +#loc95 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":100:28) +#loc96 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":100:35) +#loc97 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":100:18) +#loc98 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":100:43) +#loc99 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":100:55) +#loc100 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":100:39) +#loc101 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":100:67) +#loc102 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":100:63) +#loc103 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":101:22) +#loc104 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":102:25) +#loc105 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":102:21) +#loc106 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":104:24) +#loc107 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":104:20) +#loc108 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":105:34) +#loc109 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":105:26) +#loc110 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":106:29) +#loc111 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":106:22) +#loc112 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":96:4) +#loc113 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":34:0) +#loc114 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":43:11) +#loc115 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":43:4) +#loc117 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:33) +#loc118 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:20) +#loc119 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:13) +#loc128 = loc("o_t"(#loc10)) +#loc129 = loc("o_t"(#loc11)) +#loc130 = loc("o_t"(#loc12)) +#loc131 = loc("o_A"(#loc13)) +#loc132 = loc("m_t"(#loc14)) +#loc133 = loc("m_A"(#loc15)) +#loc134 = loc("m_A"(#loc16)) +#loc135 = loc("m_A"(#loc17)) +#loc136 = loc("m_A"(#loc18)) +#loc137 = loc("p_b"(#loc19)) +#loc138 = loc("p_b"(#loc20)) +#loc139 = loc("p_b"(#loc21)) +#loc140 = loc("p_b"(#loc22)) +#loc141 = loc("p_b"(#loc23)) +#loc142 = loc("b_b"(#loc24)) +#loc143 = loc("p_A"(#loc25)) +#loc144 = loc("p_A"(#loc26)) +#loc145 = loc("p_A"(#loc27)) +#loc146 = loc("p_A"(#loc28)) +#loc147 = loc("p_A"(#loc29)) +#loc148 = loc("p_A"(#loc30)) +#loc149 = loc("p_A"(#loc31)) +#loc150 = loc("p_A"(#loc32)) +#loc151 = loc("p_A"(#loc33)) +#loc152 = loc("b_A"(#loc34)) +#loc153 = loc("o_v"(#loc37)) +#loc154 = loc("o_v"(#loc38)) +#loc155 = loc("o_v"(#loc39)) +#loc156 = loc("m_v"(#loc40)) +#loc157 = loc("m_v"(#loc41)) +#loc158 = loc("m_v"(#loc42)) +#loc159 = loc("m_v"(#loc43)) +#loc160 = loc("p_v"(#loc44)) +#loc161 = loc("p_v"(#loc45)) +#loc162 = loc("p_v"(#loc46)) +#loc163 = loc("p_v"(#loc47)) +#loc164 = loc("p_v"(#loc48)) +#loc165 = loc("p_v"(#loc49)) +#loc166 = loc("p_v"(#loc50)) +#loc167 = loc("p_v"(#loc51)) +#loc168 = loc("p_v"(#loc52)) +#loc169 = loc("p_u"(#loc53)) +#loc170 = loc("p_u"(#loc54)) +#loc171 = loc("p_u"(#loc55)) +#loc172 = loc("p_u"(#loc56)) +#loc173 = loc("p_u"(#loc57)) +#loc174 = loc("p_u"(#loc58)) +#loc175 = loc("p_u"(#loc59)) +#loc176 = loc("p_u"(#loc60)) +#loc177 = loc("p_u"(#loc61)) +#loc178 = loc("b_v"(#loc62)) +#loc179 = loc("b_vb"(#loc63)) +#loc180 = loc("b_vb"(#loc64)) +#loc181 = loc("b_u"(#loc65)) +#loc182 = loc("p_g"(#loc68)) +#loc183 = loc("p_g"(#loc69)) +#loc184 = loc("p_g"(#loc70)) +#loc185 = loc("p_g"(#loc71)) +#loc186 = loc("p_g"(#loc72)) +#loc187 = loc("b_g"(#loc73)) +#loc188 = loc("b_g"(#loc74)) +#loc189 = loc("o_k"(#loc77)) +#loc190 = loc("o_k"(#loc78)) +#loc191 = loc("o_k"(#loc79)) +#loc192 = loc("m_k"(#loc80)) +#loc193 = loc("m_k"(#loc81)) +#loc194 = loc("m_k"(#loc82)) +#loc195 = loc("m_k"(#loc83)) +#loc196 = loc("p_k"(#loc84)) +#loc197 = loc("p_k"(#loc85)) +#loc198 = loc("p_k"(#loc86)) +#loc199 = loc("p_k"(#loc87)) +#loc200 = loc("p_k"(#loc88)) +#loc201 = loc("p_k"(#loc89)) +#loc202 = loc("p_k"(#loc90)) +#loc203 = loc("p_k"(#loc91)) +#loc204 = loc("p_k"(#loc92)) +#loc205 = loc("p_k"(#loc93)) +#loc206 = loc("p_w"(#loc94)) +#loc207 = loc("p_w"(#loc95)) +#loc208 = loc("p_w"(#loc96)) +#loc209 = loc("p_w"(#loc97)) +#loc210 = loc("p_w"(#loc98)) +#loc211 = loc("p_w"(#loc99)) +#loc212 = loc("p_w"(#loc100)) +#loc213 = loc("p_w"(#loc101)) +#loc214 = loc("p_w"(#loc102)) +#loc215 = loc("b_k"(#loc103)) +#loc216 = loc("b_kb"(#loc104)) +#loc217 = loc("b_kb"(#loc105)) +#loc218 = loc("b_kb"(#loc106)) +#loc219 = loc("b_kb"(#loc107)) +#loc220 = loc("b_w"(#loc108)) +#loc221 = loc("b_w"(#loc109)) diff --git a/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.ttgir b/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..2cdd0d58f6f1ac77623384e40d7f54110ea855f6 --- /dev/null +++ b/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.ttgir @@ -0,0 +1,369 @@ +#blocked = #ttg.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [2], order = [0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [4, 8], warpsPerCTA = [2, 1], order = [1, 0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":43:0) +#mma = #ttg.nvidia_mma<{versionMajor = 2, versionMinor = 0, warpsPerCTA = [1, 2], instrShape = [16, 8]}> +#shared = #ttg.swizzled_shared<{vec = 8, perPhase = 1, maxPhase = 8, order = [1, 0]}> +#smem = #ttg.shared_memory +#loc76 = loc("k"(#loc)) +#loc77 = loc("v"(#loc)) +#loc78 = loc("beta"(#loc)) +#loc79 = loc("w"(#loc)) +#loc80 = loc("u"(#loc)) +#loc81 = loc("A"(#loc)) +#loc82 = loc("g"(#loc)) +#loc83 = loc("T"(#loc)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 2 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @recompute_w_u_fwd_kernel(%k: !tt.ptr {tt.divisibility = 16 : i32} loc("k"(#loc)), %v: !tt.ptr {tt.divisibility = 16 : i32} loc("v"(#loc)), %beta: !tt.ptr {tt.divisibility = 16 : i32} loc("beta"(#loc)), %w: !tt.ptr {tt.divisibility = 16 : i32} loc("w"(#loc)), %u: !tt.ptr {tt.divisibility = 16 : i32} loc("u"(#loc)), %A: !tt.ptr {tt.divisibility = 16 : i32} loc("A"(#loc)), %g: !tt.ptr {tt.divisibility = 16 : i32} loc("g"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<0.000000e+00> : tensor<64x64xf32, #mma> loc(#loc1) + %c32_i64 = arith.constant 32 : i64 loc(#loc1) + %c64_i32 = arith.constant 64 : i32 loc(#loc1) + %c64_i64 = arith.constant 64 : i64 loc(#loc1) + %cst_0 = arith.constant dense<32> : tensor<64xi64, #blocked> loc(#loc1) + %cst_1 = arith.constant dense<0.000000e+00> : tensor<64xf32, #blocked> loc(#loc1) + %c128_i64 = arith.constant 128 : i64 loc(#loc1) + %cst_2 = arith.constant dense<0.000000e+00> : tensor<64x64xbf16, #blocked1> loc(#loc1) + %c1_i32 = arith.constant 1 : i32 loc(#loc1) + %c2_i32 = arith.constant 2 : i32 loc(#loc1) + %c0_i32 = arith.constant 0 : i32 loc(#loc1) + %cst_3 = arith.constant dense<0.000000e+00> : tensor<64xbf16, #blocked> loc(#loc1) + %cst_4 = arith.constant dense<4096> : tensor<64x1xi64, #blocked1> loc(#loc1) + %cst_5 = arith.constant dense<128> : tensor<1x64xi32, #blocked1> loc(#loc1) + %cst_6 = arith.constant dense<2048> : tensor<64x1xi64, #blocked1> loc(#loc1) + %cst_7 = arith.constant dense<64> : tensor<1x64xi32, #blocked1> loc(#loc1) + %c-1_i32 = arith.constant -1 : i32 loc(#loc1) + %c3_i32 = arith.constant 3 : i32 loc(#loc1) + %cst_8 = arith.constant dense<64> : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc1) + %cst_9 = arith.constant dense<128> : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc1) + %cst_10 = arith.constant dense : tensor<64x64xi1, #blocked1> loc(#loc1) + %0 = tt.get_program_id x : i32 loc(#loc2) + %1 = arith.extsi %0 : i32 to i64 loc(#loc3) + %2 = tt.get_program_id y : i32 loc(#loc4) + %3 = arith.extsi %2 : i32 to i64 loc(#loc5) + %4 = arith.divsi %3, %c32_i64 : i64 loc(#loc6) + %5 = arith.remsi %3, %c32_i64 : i64 loc(#loc7) + %6 = arith.extsi %T : i32 to i64 loc(#loc8) + %7 = arith.muli %4, %6 : i64 loc(#loc8) + %o_t = arith.muli %1, %c64_i64 : i64 loc(#loc84) + %o_t_11 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc85) + %o_t_12 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc85) + %o_t_13 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #blocked> loc(#loc85) + %o_t_14 = arith.extsi %o_t_11 : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> to tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc86) + %o_t_15 = arith.extsi %o_t_13 : tensor<64xi32, #blocked> to tensor<64xi64, #blocked> loc(#loc86) + %o_t_16 = tt.splat %o_t : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc86) + %o_t_17 = tt.splat %o_t : i64 -> tensor<64xi64, #blocked> loc(#loc86) + %o_t_18 = arith.addi %o_t_16, %o_t_14 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc86) + %o_t_19 = arith.addi %o_t_17, %o_t_15 : tensor<64xi64, #blocked> loc(#loc86) + %m_t = tt.splat %6 : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc87) + %m_t_20 = tt.splat %6 : i64 -> tensor<64xi64, #blocked> loc(#loc87) + %m_t_21 = arith.cmpi slt, %o_t_18, %m_t : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc87) + %m_t_22 = arith.cmpi slt, %o_t_19, %m_t_20 : tensor<64xi64, #blocked> loc(#loc87) + %m_A = tt.expand_dims %m_t_21 {axis = 1 : i32} : tensor<64xi1, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<64x1xi1, #blocked1> loc(#loc88) + %m_A_23 = tt.expand_dims %o_t_12 {axis = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x64xi32, #blocked1> loc(#loc89) + %m_A_24 = arith.cmpi slt, %m_A_23, %cst_7 : tensor<1x64xi32, #blocked1> loc(#loc90) + %m_A_25 = tt.broadcast %m_A : tensor<64x1xi1, #blocked1> -> tensor<64x64xi1, #blocked1> loc(#loc91) + %m_A_26 = tt.broadcast %m_A_24 : tensor<1x64xi1, #blocked1> -> tensor<64x64xi1, #blocked1> loc(#loc91) + %m_A_27 = arith.andi %m_A_25, %m_A_26 : tensor<64x64xi1, #blocked1> loc(#loc91) + %p_b = arith.muli %7, %c32_i64 : i64 loc(#loc92) + %p_b_28 = arith.addi %p_b, %5 : i64 loc(#loc143) + %p_b_29 = tt.addptr %beta, %p_b_28 : !tt.ptr, i64 loc(#loc143) + %p_b_30 = arith.muli %o_t_19, %cst_0 : tensor<64xi64, #blocked> loc(#loc95) + %p_b_31 = tt.splat %p_b_29 : !tt.ptr -> tensor<64x!tt.ptr, #blocked> loc(#loc96) + %p_b_32 = tt.addptr %p_b_31, %p_b_30 : tensor<64x!tt.ptr, #blocked>, tensor<64xi64, #blocked> loc(#loc96) + %b_b = tt.load %p_b_32, %m_t_22, %cst_3 : tensor<64x!tt.ptr, #blocked> loc(#loc97) + %b_kb = ttg.convert_layout %b_b : tensor<64xbf16, #blocked> -> tensor<64xbf16, #ttg.slice<{dim = 1, parent = #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>}>> loc(#loc98) + %p_A = arith.muli %p_b_28, %c64_i64 : i64 loc(#loc99) + %p_A_33 = tt.addptr %A, %p_A : !tt.ptr, i64 loc(#loc100) + %p_A_34 = tt.expand_dims %o_t_18 {axis = 1 : i32} : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<64x1xi64, #blocked1> loc(#loc101) + %p_A_35 = arith.muli %p_A_34, %cst_6 : tensor<64x1xi64, #blocked1> loc(#loc102) + %p_A_36 = tt.splat %p_A_33 : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked1> loc(#loc103) + %p_A_37 = tt.addptr %p_A_36, %p_A_35 : tensor<64x1x!tt.ptr, #blocked1>, tensor<64x1xi64, #blocked1> loc(#loc103) + %p_A_38 = tt.broadcast %p_A_37 : tensor<64x1x!tt.ptr, #blocked1> -> tensor<64x64x!tt.ptr, #blocked1> loc(#loc104) + %p_A_39 = tt.broadcast %m_A_23 : tensor<1x64xi32, #blocked1> -> tensor<64x64xi32, #blocked1> loc(#loc104) + %p_A_40 = tt.addptr %p_A_38, %p_A_39 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi32, #blocked1> loc(#loc104) + %b_A = tt.load %p_A_40, %m_A_27, %cst_2 : tensor<64x64x!tt.ptr, #blocked1> loc(#loc105) + %b_A_41 = ttg.local_alloc %b_A : (tensor<64x64xbf16, #blocked1>) -> !ttg.memdesc<64x64xbf16, #shared, #smem> loc(#loc105) + %p_v = arith.muli %p_b_28, %c128_i64 : i64 loc(#loc106) + %p_v_42 = tt.addptr %v, %p_v : !tt.ptr, i64 loc(#loc107) + %p_v_43 = arith.muli %p_A_34, %cst_4 : tensor<64x1xi64, #blocked1> loc(#loc108) + %p_v_44 = tt.splat %p_v_42 : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked1> loc(#loc109) + %p_v_45 = tt.addptr %p_v_44, %p_v_43 : tensor<64x1x!tt.ptr, #blocked1>, tensor<64x1xi64, #blocked1> loc(#loc109) + %p_v_46 = tt.broadcast %p_v_45 : tensor<64x1x!tt.ptr, #blocked1> -> tensor<64x64x!tt.ptr, #blocked1> loc(#loc110) + %p_u = tt.addptr %u, %p_v : !tt.ptr, i64 loc(#loc111) + %p_u_47 = tt.splat %p_u : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked1> loc(#loc112) + %p_u_48 = tt.addptr %p_u_47, %p_v_43 : tensor<64x1x!tt.ptr, #blocked1>, tensor<64x1xi64, #blocked1> loc(#loc112) + %p_u_49 = tt.broadcast %p_u_48 : tensor<64x1x!tt.ptr, #blocked1> -> tensor<64x64x!tt.ptr, #blocked1> loc(#loc113) + %b_vb = tt.expand_dims %b_kb {axis = 1 : i32} : tensor<64xbf16, #ttg.slice<{dim = 1, parent = #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>}>> -> tensor<64x1xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> loc(#loc114) + %b_vb_50 = tt.broadcast %b_vb : tensor<64x1xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> -> tensor<64x64xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> loc(#loc115) + %b_v = ttg.local_alloc : () -> !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> loc(#loc116) + %m_v = arith.cmpi slt, %m_A_23, %cst_5 : tensor<1x64xi32, #blocked1> loc(#loc117) + %m_v_51 = tt.broadcast %m_v : tensor<1x64xi1, #blocked1> -> tensor<64x64xi1, #blocked1> loc(#loc118) + %m_v_52 = arith.andi %m_A_25, %m_v_51 : tensor<64x64xi1, #blocked1> loc(#loc118) + %p_v_53 = tt.addptr %p_v_46, %p_A_39 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi32, #blocked1> loc(#loc110) + %b_v_54 = ttg.memdesc_index %b_v[%c0_i32] : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> loc(#loc116) + %b_v_55 = ttg.async_copy_global_to_local %p_v_53, %b_v_54 mask %m_v_52 other %cst_2 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked1> -> <64x64xbf16, #shared, #smem, mutable> loc(#loc116) + %b_v_56 = ttg.async_commit_group tokens %b_v_55 loc(#loc116) + %o_v = arith.addi %o_t_12, %cst_8 : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc119) + %m_v_57 = tt.expand_dims %o_v {axis = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x64xi32, #blocked1> loc(#loc120) + %m_v_58 = arith.cmpi slt, %m_v_57, %cst_5 : tensor<1x64xi32, #blocked1> loc(#loc117) + %m_v_59 = tt.broadcast %m_v_58 : tensor<1x64xi1, #blocked1> -> tensor<64x64xi1, #blocked1> loc(#loc118) + %m_v_60 = arith.andi %m_A_25, %m_v_59 : tensor<64x64xi1, #blocked1> loc(#loc118) + %p_v_61 = tt.broadcast %m_v_57 : tensor<1x64xi32, #blocked1> -> tensor<64x64xi32, #blocked1> loc(#loc110) + %p_v_62 = tt.addptr %p_v_46, %p_v_61 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi32, #blocked1> loc(#loc110) + %b_v_63 = ttg.memdesc_index %b_v[%c1_i32] : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> loc(#loc116) + %b_v_64 = ttg.async_copy_global_to_local %p_v_62, %b_v_63 mask %m_v_60 other %cst_2 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked1> -> <64x64xbf16, #shared, #smem, mutable> loc(#loc116) + %b_v_65 = ttg.async_commit_group tokens %b_v_64 loc(#loc116) + %o_v_66 = arith.addi %o_t_12, %cst_9 : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc119) + %m_v_67 = tt.expand_dims %o_v_66 {axis = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x64xi32, #blocked1> loc(#loc120) + %m_v_68 = arith.cmpi slt, %m_v_67, %cst_5 : tensor<1x64xi32, #blocked1> loc(#loc117) + %m_v_69 = tt.broadcast %m_v_68 : tensor<1x64xi1, #blocked1> -> tensor<64x64xi1, #blocked1> loc(#loc118) + %m_v_70 = arith.andi %m_A_25, %m_v_69 : tensor<64x64xi1, #blocked1> loc(#loc118) + %p_v_71 = tt.broadcast %m_v_67 : tensor<1x64xi32, #blocked1> -> tensor<64x64xi32, #blocked1> loc(#loc110) + %p_v_72 = tt.addptr %p_v_46, %p_v_71 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi32, #blocked1> loc(#loc110) + %b_v_73 = ttg.memdesc_index %b_v[%c2_i32] : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> loc(#loc116) + %b_v_74 = ttg.async_copy_global_to_local %p_v_72, %b_v_73 mask %cst_10 other %cst_2 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked1> -> <64x64xbf16, #shared, #smem, mutable> loc(#loc116) + %b_v_75 = ttg.async_commit_group tokens %b_v_74 loc(#loc116) + %8:11 = scf.for %arg8 = %c0_i32 to %c2_i32 step %c1_i32 iter_args(%arg9 = %c2_i32, %arg10 = %c-1_i32, %p_A_100 = %p_A_39, %p_v_101 = %p_v_61, %p_v_102 = %p_v_71, %b_v_103 = %b_v_56, %b_v_104 = %b_v_65, %b_v_105 = %b_v_75, %m_v_106 = %m_v_52, %m_v_107 = %m_v_60, %m_v_108 = %m_v_70) -> (i32, i32, tensor<64x64xi32, #blocked1>, tensor<64x64xi32, #blocked1>, tensor<64x64xi32, #blocked1>, !ttg.async.token, !ttg.async.token, !ttg.async.token, tensor<64x64xi1, #blocked1>, tensor<64x64xi1, #blocked1>, tensor<64x64xi1, #blocked1>) : i32 { + %12 = arith.cmpi slt, %arg8, %c-1_i32 : i32 loc(#loc46) + %13 = arith.addi %arg10, %c1_i32 : i32 loc(#loc46) + %14 = arith.cmpi sge, %13, %c3_i32 : i32 loc(#loc46) + %15 = arith.select %14, %c0_i32, %13 : i32 loc(#loc46) + %p_u_109 = tt.addptr %p_u_49, %p_A_100 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi32, #blocked1> loc(#loc113) + %b_v_110 = ttg.async_wait %b_v_103 {num = 2 : i32} loc(#loc116) + %b_v_111 = ttg.memdesc_index %b_v[%15] : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> loc(#loc116) + %b_vb_112 = ttg.local_load %b_v_111 token %b_v_110 : !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> -> tensor<64x64xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> loc(#loc115) + %b_vb_113 = arith.mulf %b_vb_112, %b_vb_50 : tensor<64x64xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> loc(#loc115) + %b_A_114 = ttg.local_load %b_A_41 : !ttg.memdesc<64x64xbf16, #shared, #smem> -> tensor<64x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma, kWidth = 2}>> loc(#loc105) + %b_u = tt.dot %b_A_114, %b_vb_113, %cst : tensor<64x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma, kWidth = 2}>> * tensor<64x64xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> -> tensor<64x64xf32, #mma> loc(#loc121) + %16 = arith.truncf %b_u : tensor<64x64xf32, #mma> to tensor<64x64xbf16, #mma> loc(#loc48) + %17 = ttg.convert_layout %16 : tensor<64x64xbf16, #mma> -> tensor<64x64xbf16, #blocked1> loc(#loc49) + tt.store %p_u_109, %17, %m_v_106 : tensor<64x64x!tt.ptr, #blocked1> loc(#loc49) + %18 = arith.addi %arg9, %c1_i32 : i32 loc(#loc46) + %19 = arith.cmpi sge, %18, %c3_i32 : i32 loc(#loc46) + %20 = arith.select %19, %c0_i32, %18 : i32 loc(#loc46) + %21 = arith.addi %arg8, %c3_i32 : i32 loc(#loc46) + %o_v_115 = arith.muli %21, %c64_i32 : i32 loc(#loc122) + %o_v_116 = tt.splat %o_v_115 : i32 -> tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc119) + %o_v_117 = arith.addi %o_v_116, %o_t_12 : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc119) + %m_v_118 = tt.expand_dims %o_v_117 {axis = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x64xi32, #blocked1> loc(#loc120) + %m_v_119 = arith.cmpi slt, %m_v_118, %cst_5 : tensor<1x64xi32, #blocked1> loc(#loc117) + %m_v_120 = tt.broadcast %m_v_119 : tensor<1x64xi1, #blocked1> -> tensor<64x64xi1, #blocked1> loc(#loc118) + %m_v_121 = arith.andi %m_A_25, %m_v_120 : tensor<64x64xi1, #blocked1> loc(#loc118) + %p_v_122 = tt.broadcast %m_v_118 : tensor<1x64xi32, #blocked1> -> tensor<64x64xi32, #blocked1> loc(#loc110) + %p_v_123 = tt.addptr %p_v_46, %p_v_122 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi32, #blocked1> loc(#loc110) + %b_v_124 = ttg.memdesc_index %b_v[%20] : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> loc(#loc116) + %22 = tt.splat %12 : i1 -> tensor<64x64xi1, #blocked1> loc(#loc46) + %23 = arith.andi %22, %m_v_121 : tensor<64x64xi1, #blocked1> loc(#loc46) + %b_v_125 = ttg.async_copy_global_to_local %p_v_123, %b_v_124 mask %23 other %cst_2 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked1> -> <64x64xbf16, #shared, #smem, mutable> loc(#loc116) + %b_v_126 = ttg.async_commit_group tokens %b_v_125 loc(#loc116) + scf.yield %20, %15, %p_v_101, %p_v_102, %p_v_122, %b_v_104, %b_v_105, %b_v_126, %m_v_107, %m_v_108, %m_v_121 : i32, i32, tensor<64x64xi32, #blocked1>, tensor<64x64xi32, #blocked1>, tensor<64x64xi32, #blocked1>, !ttg.async.token, !ttg.async.token, !ttg.async.token, tensor<64x64xi1, #blocked1>, tensor<64x64xi1, #blocked1>, tensor<64x64xi1, #blocked1> loc(#loc46) + } loc(#loc46) + %9 = ttg.async_wait {num = 0 : i32} loc(#loc46) + ttg.local_dealloc %b_v : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> loc(#loc46) + %p_g = tt.addptr %g, %p_b_28 : !tt.ptr, i64 loc(#loc123) + %p_g_76 = tt.splat %p_g : !tt.ptr -> tensor<64x!tt.ptr, #blocked> loc(#loc124) + %p_g_77 = tt.addptr %p_g_76, %p_b_30 : tensor<64x!tt.ptr, #blocked>, tensor<64xi64, #blocked> loc(#loc124) + %b_g = tt.load %p_g_77, %m_t_22, %cst_1 : tensor<64x!tt.ptr, #blocked> loc(#loc125) + %b_g_78 = math.exp2 %b_g : tensor<64xf32, #blocked> loc(#loc144) + %b_g_79 = ttg.convert_layout %b_g_78 : tensor<64xf32, #blocked> -> tensor<64xf32, #ttg.slice<{dim = 1, parent = #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>}>> loc(#loc144) + %p_k = tt.addptr %k, %p_v : !tt.ptr, i64 loc(#loc127) + %p_k_80 = tt.splat %p_k : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked1> loc(#loc128) + %p_k_81 = tt.addptr %p_k_80, %p_v_43 : tensor<64x1x!tt.ptr, #blocked1>, tensor<64x1xi64, #blocked1> loc(#loc128) + %p_k_82 = tt.broadcast %p_k_81 : tensor<64x1x!tt.ptr, #blocked1> -> tensor<64x64x!tt.ptr, #blocked1> loc(#loc129) + %p_w = tt.addptr %w, %p_v : !tt.ptr, i64 loc(#loc130) + %p_w_83 = tt.splat %p_w : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked1> loc(#loc131) + %p_w_84 = tt.addptr %p_w_83, %p_v_43 : tensor<64x1x!tt.ptr, #blocked1>, tensor<64x1xi64, #blocked1> loc(#loc131) + %p_w_85 = tt.broadcast %p_w_84 : tensor<64x1x!tt.ptr, #blocked1> -> tensor<64x64x!tt.ptr, #blocked1> loc(#loc132) + %b_kb_86 = tt.expand_dims %b_g_79 {axis = 1 : i32} : tensor<64xf32, #ttg.slice<{dim = 1, parent = #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>}>> -> tensor<64x1xf32, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> loc(#loc133) + %b_kb_87 = tt.broadcast %b_kb_86 : tensor<64x1xf32, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> -> tensor<64x64xf32, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> loc(#loc134) + %b_k = ttg.local_alloc : () -> !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> loc(#loc135) + %p_k_88 = tt.addptr %p_k_82, %p_A_39 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi32, #blocked1> loc(#loc129) + %b_k_89 = ttg.memdesc_index %b_k[%c0_i32] : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> loc(#loc135) + %b_k_90 = ttg.async_copy_global_to_local %p_k_88, %b_k_89 mask %m_v_52 other %cst_2 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked1> -> <64x64xbf16, #shared, #smem, mutable> loc(#loc135) + %b_k_91 = ttg.async_commit_group tokens %b_k_90 loc(#loc135) + %p_k_92 = tt.addptr %p_k_82, %p_v_61 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi32, #blocked1> loc(#loc129) + %b_k_93 = ttg.memdesc_index %b_k[%c1_i32] : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> loc(#loc135) + %b_k_94 = ttg.async_copy_global_to_local %p_k_92, %b_k_93 mask %m_v_60 other %cst_2 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked1> -> <64x64xbf16, #shared, #smem, mutable> loc(#loc135) + %b_k_95 = ttg.async_commit_group tokens %b_k_94 loc(#loc135) + %p_k_96 = tt.addptr %p_k_82, %p_v_71 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi32, #blocked1> loc(#loc129) + %b_k_97 = ttg.memdesc_index %b_k[%c2_i32] : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> loc(#loc135) + %b_k_98 = ttg.async_copy_global_to_local %p_k_96, %b_k_97 mask %cst_10 other %cst_2 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked1> -> <64x64xbf16, #shared, #smem, mutable> loc(#loc135) + %b_k_99 = ttg.async_commit_group tokens %b_k_98 loc(#loc135) + %10:11 = scf.for %arg8 = %c0_i32 to %c2_i32 step %c1_i32 iter_args(%arg9 = %c2_i32, %arg10 = %c-1_i32, %p_A_100 = %p_A_39, %p_v_101 = %p_v_61, %p_v_102 = %p_v_71, %b_k_103 = %b_k_91, %b_k_104 = %b_k_95, %b_k_105 = %b_k_99, %m_v_106 = %m_v_52, %m_v_107 = %m_v_60, %m_v_108 = %m_v_70) -> (i32, i32, tensor<64x64xi32, #blocked1>, tensor<64x64xi32, #blocked1>, tensor<64x64xi32, #blocked1>, !ttg.async.token, !ttg.async.token, !ttg.async.token, tensor<64x64xi1, #blocked1>, tensor<64x64xi1, #blocked1>, tensor<64x64xi1, #blocked1>) : i32 { + %12 = arith.cmpi slt, %arg8, %c-1_i32 : i32 loc(#loc65) + %13 = arith.addi %arg10, %c1_i32 : i32 loc(#loc65) + %14 = arith.cmpi sge, %13, %c3_i32 : i32 loc(#loc65) + %15 = arith.select %14, %c0_i32, %13 : i32 loc(#loc65) + %p_w_109 = tt.addptr %p_w_85, %p_A_100 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi32, #blocked1> loc(#loc132) + %b_k_110 = ttg.async_wait %b_k_103 {num = 2 : i32} loc(#loc135) + %b_k_111 = ttg.memdesc_index %b_k[%15] : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> loc(#loc135) + %b_w = ttg.local_load %b_k_111 token %b_k_110 : !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> -> tensor<64x64xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> loc(#loc136) + %b_kb_112 = arith.mulf %b_w, %b_vb_50 : tensor<64x64xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> loc(#loc98) + %b_kb_113 = arith.extf %b_kb_112 : tensor<64x64xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> to tensor<64x64xf32, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> loc(#loc134) + %b_kb_114 = arith.mulf %b_kb_113, %b_kb_87 : tensor<64x64xf32, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> loc(#loc134) + %b_w_115 = arith.truncf %b_kb_114 : tensor<64x64xf32, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> to tensor<64x64xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> loc(#loc136) + %b_A_116 = ttg.local_load %b_A_41 : !ttg.memdesc<64x64xbf16, #shared, #smem> -> tensor<64x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma, kWidth = 2}>> loc(#loc105) + %b_w_117 = tt.dot %b_A_116, %b_w_115, %cst, inputPrecision = tf32 : tensor<64x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma, kWidth = 2}>> * tensor<64x64xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> -> tensor<64x64xf32, #mma> loc(#loc137) + %16 = arith.truncf %b_w_117 : tensor<64x64xf32, #mma> to tensor<64x64xbf16, #mma> loc(#loc68) + %17 = ttg.convert_layout %16 : tensor<64x64xbf16, #mma> -> tensor<64x64xbf16, #blocked1> loc(#loc69) + tt.store %p_w_109, %17, %m_v_106 : tensor<64x64x!tt.ptr, #blocked1> loc(#loc69) + %18 = arith.addi %arg9, %c1_i32 : i32 loc(#loc65) + %19 = arith.cmpi sge, %18, %c3_i32 : i32 loc(#loc65) + %20 = arith.select %19, %c0_i32, %18 : i32 loc(#loc65) + %21 = arith.addi %arg8, %c3_i32 : i32 loc(#loc65) + %o_k = arith.muli %21, %c64_i32 : i32 loc(#loc138) + %o_k_118 = tt.splat %o_k : i32 -> tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc139) + %o_k_119 = arith.addi %o_k_118, %o_t_12 : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc139) + %m_k = tt.expand_dims %o_k_119 {axis = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x64xi32, #blocked1> loc(#loc140) + %m_k_120 = arith.cmpi slt, %m_k, %cst_5 : tensor<1x64xi32, #blocked1> loc(#loc141) + %m_k_121 = tt.broadcast %m_k_120 : tensor<1x64xi1, #blocked1> -> tensor<64x64xi1, #blocked1> loc(#loc142) + %m_k_122 = arith.andi %m_A_25, %m_k_121 : tensor<64x64xi1, #blocked1> loc(#loc142) + %p_k_123 = tt.broadcast %m_k : tensor<1x64xi32, #blocked1> -> tensor<64x64xi32, #blocked1> loc(#loc129) + %p_k_124 = tt.addptr %p_k_82, %p_k_123 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi32, #blocked1> loc(#loc129) + %b_k_125 = ttg.memdesc_index %b_k[%20] : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> loc(#loc135) + %22 = tt.splat %12 : i1 -> tensor<64x64xi1, #blocked1> loc(#loc65) + %23 = arith.andi %22, %m_k_122 : tensor<64x64xi1, #blocked1> loc(#loc65) + %b_k_126 = ttg.async_copy_global_to_local %p_k_124, %b_k_125 mask %23 other %cst_2 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked1> -> <64x64xbf16, #shared, #smem, mutable> loc(#loc135) + %b_k_127 = ttg.async_commit_group tokens %b_k_126 loc(#loc135) + scf.yield %20, %15, %p_v_101, %p_v_102, %p_k_123, %b_k_104, %b_k_105, %b_k_127, %m_v_107, %m_v_108, %m_k_122 : i32, i32, tensor<64x64xi32, #blocked1>, tensor<64x64xi32, #blocked1>, tensor<64x64xi32, #blocked1>, !ttg.async.token, !ttg.async.token, !ttg.async.token, tensor<64x64xi1, #blocked1>, tensor<64x64xi1, #blocked1>, tensor<64x64xi1, #blocked1> loc(#loc65) + } loc(#loc65) + %11 = ttg.async_wait {num = 0 : i32} loc(#loc65) + ttg.local_dealloc %b_k : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> loc(#loc65) + tt.return loc(#loc75) + } loc(#loc) +} loc(#loc) +#loc1 = loc(unknown) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":64:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":64:36) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":64:61) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":64:67) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":65:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":65:34) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":71:25) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":72:16) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":72:34) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":72:21) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":74:16) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":75:14) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":75:30) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":75:41) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":75:26) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":76:21) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":76:26) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":76:17) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":76:38) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":76:32) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":77:18) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":102:21) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":79:31) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":79:14) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":79:40) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":79:52) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":79:36) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":79:61) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":80:18) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":85:35) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":85:18) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":85:55) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":85:39) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":85:63) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":86:18) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":86:39) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":86:63) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":88:26) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":88:22) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":87:22) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":84:45) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":84:30) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":83:25) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":84:34) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":82:21) +#loc47 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":89:26) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":90:29) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":90:22) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":83:20) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":93:19) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":93:35) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":94:27) +#loc54 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:33) +#loc55 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":94:19) +#loc56 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":99:18) +#loc57 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":99:51) +#loc58 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":99:74) +#loc59 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":100:18) +#loc60 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":100:39) +#loc61 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":100:63) +#loc62 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":104:24) +#loc63 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":104:20) +#loc64 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":101:22) +#loc65 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":96:21) +#loc66 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":105:34) +#loc67 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":105:26) +#loc68 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":106:29) +#loc69 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":106:22) +#loc70 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":97:20) +#loc71 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":97:25) +#loc72 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":98:34) +#loc73 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":98:45) +#loc74 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":98:30) +#loc75 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":96:4) +#loc84 = loc("o_t"(#loc9)) +#loc85 = loc("o_t"(#loc10)) +#loc86 = loc("o_t"(#loc11)) +#loc87 = loc("m_t"(#loc12)) +#loc88 = loc("m_A"(#loc13)) +#loc89 = loc("m_A"(#loc14)) +#loc90 = loc("m_A"(#loc15)) +#loc91 = loc("m_A"(#loc16)) +#loc92 = loc("p_b"(#loc17)) +#loc93 = loc("p_b"(#loc18)) +#loc94 = loc("p_b"(#loc19)) +#loc95 = loc("p_b"(#loc20)) +#loc96 = loc("p_b"(#loc21)) +#loc97 = loc("b_b"(#loc22)) +#loc98 = loc("b_kb"(#loc23)) +#loc99 = loc("p_A"(#loc24)) +#loc100 = loc("p_A"(#loc25)) +#loc101 = loc("p_A"(#loc26)) +#loc102 = loc("p_A"(#loc27)) +#loc103 = loc("p_A"(#loc28)) +#loc104 = loc("p_A"(#loc29)) +#loc105 = loc("b_A"(#loc30)) +#loc106 = loc("p_v"(#loc31)) +#loc107 = loc("p_v"(#loc32)) +#loc108 = loc("p_v"(#loc33)) +#loc109 = loc("p_v"(#loc34)) +#loc110 = loc("p_v"(#loc35)) +#loc111 = loc("p_u"(#loc36)) +#loc112 = loc("p_u"(#loc37)) +#loc113 = loc("p_u"(#loc38)) +#loc114 = loc("b_vb"(#loc39)) +#loc115 = loc("b_vb"(#loc40)) +#loc116 = loc("b_v"(#loc41)) +#loc117 = loc("m_v"(#loc42)) +#loc118 = loc("m_v"(#loc43)) +#loc119 = loc("o_v"(#loc44)) +#loc120 = loc("m_v"(#loc45)) +#loc121 = loc("b_u"(#loc47)) +#loc122 = loc("o_v"(#loc50)) +#loc123 = loc("p_g"(#loc51)) +#loc124 = loc("p_g"(#loc52)) +#loc125 = loc("b_g"(#loc53)) +#loc126 = loc("b_g"(#loc55)) +#loc127 = loc("p_k"(#loc56)) +#loc128 = loc("p_k"(#loc57)) +#loc129 = loc("p_k"(#loc58)) +#loc130 = loc("p_w"(#loc59)) +#loc131 = loc("p_w"(#loc60)) +#loc132 = loc("p_w"(#loc61)) +#loc133 = loc("b_kb"(#loc62)) +#loc134 = loc("b_kb"(#loc63)) +#loc135 = loc("b_k"(#loc64)) +#loc136 = loc("b_w"(#loc66)) +#loc137 = loc("b_w"(#loc67)) +#loc138 = loc("o_k"(#loc70)) +#loc139 = loc("o_k"(#loc71)) +#loc140 = loc("m_k"(#loc72)) +#loc141 = loc("m_k"(#loc73)) +#loc142 = loc("m_k"(#loc74)) +#loc143 = loc(fused[#loc93, #loc94]) +#loc144 = loc(callsite(#loc54 at #loc126)) diff --git a/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.ttir b/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..78155515097592189bb3c0a9bccd3efdefe0ffaf --- /dev/null +++ b/triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.ttir @@ -0,0 +1,278 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":43:0) +#loc79 = loc("k"(#loc)) +#loc80 = loc("v"(#loc)) +#loc81 = loc("beta"(#loc)) +#loc82 = loc("w"(#loc)) +#loc83 = loc("u"(#loc)) +#loc84 = loc("A"(#loc)) +#loc85 = loc("g"(#loc)) +#loc86 = loc("T"(#loc)) +module { + tt.func public @recompute_w_u_fwd_kernel(%k: !tt.ptr {tt.divisibility = 16 : i32} loc("k"(#loc)), %v: !tt.ptr {tt.divisibility = 16 : i32} loc("v"(#loc)), %beta: !tt.ptr {tt.divisibility = 16 : i32} loc("beta"(#loc)), %w: !tt.ptr {tt.divisibility = 16 : i32} loc("w"(#loc)), %u: !tt.ptr {tt.divisibility = 16 : i32} loc("u"(#loc)), %A: !tt.ptr {tt.divisibility = 16 : i32} loc("A"(#loc)), %g: !tt.ptr {tt.divisibility = 16 : i32} loc("g"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %c1_i32 = arith.constant 1 : i32 loc(#loc1) + %c2_i32 = arith.constant 2 : i32 loc(#loc1) + %c0_i32 = arith.constant 0 : i32 loc(#loc1) + %cst = arith.constant dense<0.000000e+00> : tensor<64x64xbf16> loc(#loc1) + %b_b = arith.constant dense<0.000000e+00> : tensor<64xbf16> loc(#loc87) + %cst_0 = arith.constant dense<4096> : tensor<64x1xi64> loc(#loc1) + %c128_i64 = arith.constant 128 : i64 loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<1x64xi32> loc(#loc1) + %cst_2 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc1) + %p_A = arith.constant dense<2048> : tensor<64x1xi64> loc(#loc88) + %cst_3 = arith.constant dense<0.000000e+00> : tensor<64xf32> loc(#loc1) + %cst_4 = arith.constant dense<32> : tensor<64xi64> loc(#loc1) + %m_A = arith.constant dense<64> : tensor<1x64xi32> loc(#loc89) + %c64_i64 = arith.constant 64 : i64 loc(#loc1) + %c64_i32 = arith.constant 64 : i32 loc(#loc1) + %c32_i64 = arith.constant 32 : i64 loc(#loc1) + %0 = tt.get_program_id x : i32 loc(#loc5) + %1 = arith.extsi %0 : i32 to i64 loc(#loc6) + %2 = tt.get_program_id y : i32 loc(#loc7) + %3 = arith.extsi %2 : i32 to i64 loc(#loc8) + %4 = arith.divsi %3, %c32_i64 : i64 loc(#loc9) + %5 = arith.remsi %3, %c32_i64 : i64 loc(#loc10) + %6 = arith.extsi %T : i32 to i64 loc(#loc11) + %7 = arith.muli %4, %6 : i64 loc(#loc11) + %o_t = arith.muli %1, %c64_i64 : i64 loc(#loc90) + %o_t_5 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc91) + %o_t_6 = arith.extsi %o_t_5 : tensor<64xi32> to tensor<64xi64> loc(#loc92) + %o_t_7 = tt.splat %o_t : i64 -> tensor<64xi64> loc(#loc92) + %o_t_8 = arith.addi %o_t_7, %o_t_6 : tensor<64xi64> loc(#loc92) + %m_t = tt.splat %6 : i64 -> tensor<64xi64> loc(#loc93) + %m_t_9 = arith.cmpi slt, %o_t_8, %m_t : tensor<64xi64> loc(#loc93) + %m_A_10 = tt.expand_dims %m_t_9 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc94) + %m_A_11 = tt.expand_dims %o_t_5 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc95) + %m_A_12 = arith.cmpi slt, %m_A_11, %m_A : tensor<1x64xi32> loc(#loc89) + %m_A_13 = tt.broadcast %m_A_10 : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc96) + %m_A_14 = tt.broadcast %m_A_12 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc96) + %m_A_15 = arith.andi %m_A_13, %m_A_14 : tensor<64x64xi1> loc(#loc96) + %p_b = arith.muli %7, %c32_i64 : i64 loc(#loc97) + %p_b_16 = arith.addi %p_b, %5 : i64 loc(#loc149) + %p_b_17 = tt.addptr %beta, %p_b_16 : !tt.ptr, i64 loc(#loc149) + %p_b_18 = arith.muli %o_t_8, %cst_4 : tensor<64xi64> loc(#loc100) + %p_b_19 = tt.splat %p_b_17 : !tt.ptr -> tensor<64x!tt.ptr> loc(#loc101) + %p_b_20 = tt.addptr %p_b_19, %p_b_18 : tensor<64x!tt.ptr>, tensor<64xi64> loc(#loc101) + %b_b_21 = tt.load %p_b_20, %m_t_9, %b_b : tensor<64x!tt.ptr> loc(#loc87) + %p_A_22 = arith.muli %p_b_16, %c64_i64 : i64 loc(#loc102) + %p_A_23 = tt.addptr %A, %p_A_22 : !tt.ptr, i64 loc(#loc103) + %p_A_24 = tt.expand_dims %o_t_8 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc104) + %p_A_25 = arith.muli %p_A_24, %p_A : tensor<64x1xi64> loc(#loc88) + %p_A_26 = tt.splat %p_A_23 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc105) + %p_A_27 = tt.addptr %p_A_26, %p_A_25 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc105) + %p_A_28 = tt.broadcast %p_A_27 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc106) + %p_A_29 = tt.broadcast %m_A_11 : tensor<1x64xi32> -> tensor<64x64xi32> loc(#loc106) + %p_A_30 = tt.addptr %p_A_28, %p_A_29 : tensor<64x64x!tt.ptr>, tensor<64x64xi32> loc(#loc106) + %b_A = tt.load %p_A_30, %m_A_15, %cst : tensor<64x64x!tt.ptr> loc(#loc107) + scf.for %i_v = %c0_i32 to %c2_i32 step %c1_i32 : i32 { + %o_v = arith.muli %i_v, %c64_i32 : i32 loc(#loc108) + %o_v_34 = tt.splat %o_v : i32 -> tensor<64xi32> loc(#loc109) + %o_v_35 = arith.addi %o_v_34, %o_t_5 : tensor<64xi32> loc(#loc109) + %m_v = tt.expand_dims %o_v_35 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc110) + %m_v_36 = arith.cmpi slt, %m_v, %cst_1 : tensor<1x64xi32> loc(#loc111) + %m_v_37 = tt.broadcast %m_v_36 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc112) + %m_v_38 = arith.andi %m_A_13, %m_v_37 : tensor<64x64xi1> loc(#loc112) + %p_v = arith.muli %p_b_16, %c128_i64 : i64 loc(#loc113) + %p_v_39 = tt.addptr %v, %p_v : !tt.ptr, i64 loc(#loc114) + %p_v_40 = arith.muli %p_A_24, %cst_0 : tensor<64x1xi64> loc(#loc115) + %p_v_41 = tt.splat %p_v_39 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc116) + %p_v_42 = tt.addptr %p_v_41, %p_v_40 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc116) + %p_v_43 = tt.broadcast %p_v_42 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc117) + %p_v_44 = tt.broadcast %m_v : tensor<1x64xi32> -> tensor<64x64xi32> loc(#loc117) + %p_v_45 = tt.addptr %p_v_43, %p_v_44 : tensor<64x64x!tt.ptr>, tensor<64x64xi32> loc(#loc117) + %p_u = tt.addptr %u, %p_v : !tt.ptr, i64 loc(#loc118) + %p_u_46 = tt.splat %p_u : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc119) + %p_u_47 = tt.addptr %p_u_46, %p_v_40 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc119) + %p_u_48 = tt.broadcast %p_u_47 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc120) + %p_u_49 = tt.addptr %p_u_48, %p_v_44 : tensor<64x64x!tt.ptr>, tensor<64x64xi32> loc(#loc120) + %b_v = tt.load %p_v_45, %m_v_38, %cst : tensor<64x64x!tt.ptr> loc(#loc121) + %b_vb = tt.expand_dims %b_b_21 {axis = 1 : i32} : tensor<64xbf16> -> tensor<64x1xbf16> loc(#loc122) + %b_vb_50 = tt.broadcast %b_vb : tensor<64x1xbf16> -> tensor<64x64xbf16> loc(#loc123) + %b_vb_51 = arith.mulf %b_v, %b_vb_50 : tensor<64x64xbf16> loc(#loc123) + %b_u = tt.dot %b_A, %b_vb_51, %cst_2 : tensor<64x64xbf16> * tensor<64x64xbf16> -> tensor<64x64xf32> loc(#loc124) + %8 = arith.truncf %b_u : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc48) + tt.store %p_u_49, %8, %m_v_38 : tensor<64x64x!tt.ptr> loc(#loc49) + } loc(#loc30) + %p_g = tt.addptr %g, %p_b_16 : !tt.ptr, i64 loc(#loc125) + %p_g_31 = tt.splat %p_g : !tt.ptr -> tensor<64x!tt.ptr> loc(#loc126) + %p_g_32 = tt.addptr %p_g_31, %p_b_18 : tensor<64x!tt.ptr>, tensor<64xi64> loc(#loc126) + %b_g = tt.load %p_g_32, %m_t_9, %cst_3 : tensor<64x!tt.ptr> loc(#loc127) + %b_g_33 = math.exp2 %b_g : tensor<64xf32> loc(#loc150) + scf.for %i_k = %c0_i32 to %c2_i32 step %c1_i32 : i32 { + %o_k = arith.muli %i_k, %c64_i32 : i32 loc(#loc129) + %o_k_34 = tt.splat %o_k : i32 -> tensor<64xi32> loc(#loc130) + %o_k_35 = arith.addi %o_k_34, %o_t_5 : tensor<64xi32> loc(#loc130) + %m_k = tt.expand_dims %o_k_35 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc131) + %m_k_36 = arith.cmpi slt, %m_k, %cst_1 : tensor<1x64xi32> loc(#loc132) + %m_k_37 = tt.broadcast %m_k_36 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc133) + %m_k_38 = arith.andi %m_A_13, %m_k_37 : tensor<64x64xi1> loc(#loc133) + %p_k = arith.muli %p_b_16, %c128_i64 : i64 loc(#loc134) + %p_k_39 = tt.addptr %k, %p_k : !tt.ptr, i64 loc(#loc135) + %p_k_40 = arith.muli %p_A_24, %cst_0 : tensor<64x1xi64> loc(#loc136) + %p_k_41 = tt.splat %p_k_39 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc137) + %p_k_42 = tt.addptr %p_k_41, %p_k_40 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc137) + %p_k_43 = tt.broadcast %p_k_42 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc138) + %p_k_44 = tt.broadcast %m_k : tensor<1x64xi32> -> tensor<64x64xi32> loc(#loc138) + %p_k_45 = tt.addptr %p_k_43, %p_k_44 : tensor<64x64x!tt.ptr>, tensor<64x64xi32> loc(#loc138) + %p_w = tt.addptr %w, %p_k : !tt.ptr, i64 loc(#loc139) + %p_w_46 = tt.splat %p_w : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc140) + %p_w_47 = tt.addptr %p_w_46, %p_k_40 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc140) + %p_w_48 = tt.broadcast %p_w_47 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc141) + %p_w_49 = tt.addptr %p_w_48, %p_k_44 : tensor<64x64x!tt.ptr>, tensor<64x64xi32> loc(#loc141) + %b_k = tt.load %p_k_45, %m_k_38, %cst : tensor<64x64x!tt.ptr> loc(#loc142) + %b_kb = tt.expand_dims %b_b_21 {axis = 1 : i32} : tensor<64xbf16> -> tensor<64x1xbf16> loc(#loc143) + %b_kb_50 = tt.broadcast %b_kb : tensor<64x1xbf16> -> tensor<64x64xbf16> loc(#loc144) + %b_kb_51 = arith.mulf %b_k, %b_kb_50 : tensor<64x64xbf16> loc(#loc144) + %b_kb_52 = tt.expand_dims %b_g_33 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc145) + %b_kb_53 = arith.extf %b_kb_51 : tensor<64x64xbf16> to tensor<64x64xf32> loc(#loc146) + %b_kb_54 = tt.broadcast %b_kb_52 : tensor<64x1xf32> -> tensor<64x64xf32> loc(#loc146) + %b_kb_55 = arith.mulf %b_kb_53, %b_kb_54 : tensor<64x64xf32> loc(#loc146) + %b_w = arith.truncf %b_kb_55 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc147) + %b_w_56 = tt.dot %b_A, %b_w, %cst_2, inputPrecision = tf32 : tensor<64x64xbf16> * tensor<64x64xbf16> -> tensor<64x64xf32> loc(#loc148) + %8 = arith.truncf %b_w_56 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc76) + tt.store %p_w_49, %8, %m_k_38 : tensor<64x64x!tt.ptr> loc(#loc77) + } loc(#loc55) + tt.return loc(#loc78) + } loc(#loc) +} loc(#loc) +#loc1 = loc(unknown) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":77:18) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":79:52) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":75:41) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":64:30) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":64:36) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":64:61) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":64:67) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":65:23) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":65:34) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":71:25) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":72:16) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":72:34) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":72:21) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":74:16) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":75:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":75:30) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":75:26) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":76:21) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":76:26) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":76:17) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":76:38) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":76:32) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":79:31) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":79:14) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":79:40) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":79:36) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":79:61) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":80:18) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":82:21) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":83:20) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":83:25) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":84:34) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":84:45) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":84:30) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":85:35) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":85:18) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":85:55) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":85:39) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":85:63) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":86:18) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":86:39) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":86:63) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":87:22) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":88:26) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":88:22) +#loc47 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":89:26) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":90:29) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":90:22) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":93:19) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":93:35) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":94:27) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:33) +#loc54 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":94:19) +#loc55 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":96:21) +#loc56 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":97:20) +#loc57 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":97:25) +#loc58 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":98:34) +#loc59 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":98:45) +#loc60 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":98:30) +#loc61 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":99:47) +#loc62 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":99:18) +#loc63 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":99:67) +#loc64 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":99:51) +#loc65 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":99:74) +#loc66 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":100:18) +#loc67 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":100:39) +#loc68 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":100:63) +#loc69 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":101:22) +#loc70 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":102:25) +#loc71 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":102:21) +#loc72 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":104:24) +#loc73 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":104:20) +#loc74 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":105:34) +#loc75 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":105:26) +#loc76 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":106:29) +#loc77 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":106:22) +#loc78 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/wy_fast.py":96:4) +#loc87 = loc("b_b"(#loc2)) +#loc88 = loc("p_A"(#loc3)) +#loc89 = loc("m_A"(#loc4)) +#loc90 = loc("o_t"(#loc12)) +#loc91 = loc("o_t"(#loc13)) +#loc92 = loc("o_t"(#loc14)) +#loc93 = loc("m_t"(#loc15)) +#loc94 = loc("m_A"(#loc16)) +#loc95 = loc("m_A"(#loc17)) +#loc96 = loc("m_A"(#loc18)) +#loc97 = loc("p_b"(#loc19)) +#loc98 = loc("p_b"(#loc20)) +#loc99 = loc("p_b"(#loc21)) +#loc100 = loc("p_b"(#loc22)) +#loc101 = loc("p_b"(#loc23)) +#loc102 = loc("p_A"(#loc24)) +#loc103 = loc("p_A"(#loc25)) +#loc104 = loc("p_A"(#loc26)) +#loc105 = loc("p_A"(#loc27)) +#loc106 = loc("p_A"(#loc28)) +#loc107 = loc("b_A"(#loc29)) +#loc108 = loc("o_v"(#loc31)) +#loc109 = loc("o_v"(#loc32)) +#loc110 = loc("m_v"(#loc33)) +#loc111 = loc("m_v"(#loc34)) +#loc112 = loc("m_v"(#loc35)) +#loc113 = loc("p_v"(#loc36)) +#loc114 = loc("p_v"(#loc37)) +#loc115 = loc("p_v"(#loc38)) +#loc116 = loc("p_v"(#loc39)) +#loc117 = loc("p_v"(#loc40)) +#loc118 = loc("p_u"(#loc41)) +#loc119 = loc("p_u"(#loc42)) +#loc120 = loc("p_u"(#loc43)) +#loc121 = loc("b_v"(#loc44)) +#loc122 = loc("b_vb"(#loc45)) +#loc123 = loc("b_vb"(#loc46)) +#loc124 = loc("b_u"(#loc47)) +#loc125 = loc("p_g"(#loc50)) +#loc126 = loc("p_g"(#loc51)) +#loc127 = loc("b_g"(#loc52)) +#loc128 = loc("b_g"(#loc54)) +#loc129 = loc("o_k"(#loc56)) +#loc130 = loc("o_k"(#loc57)) +#loc131 = loc("m_k"(#loc58)) +#loc132 = loc("m_k"(#loc59)) +#loc133 = loc("m_k"(#loc60)) +#loc134 = loc("p_k"(#loc61)) +#loc135 = loc("p_k"(#loc62)) +#loc136 = loc("p_k"(#loc63)) +#loc137 = loc("p_k"(#loc64)) +#loc138 = loc("p_k"(#loc65)) +#loc139 = loc("p_w"(#loc66)) +#loc140 = loc("p_w"(#loc67)) +#loc141 = loc("p_w"(#loc68)) +#loc142 = loc("b_k"(#loc69)) +#loc143 = loc("b_kb"(#loc70)) +#loc144 = loc("b_kb"(#loc71)) +#loc145 = loc("b_kb"(#loc72)) +#loc146 = loc("b_kb"(#loc73)) +#loc147 = loc("b_w"(#loc74)) +#loc148 = loc("b_w"(#loc75)) +#loc149 = loc(fused[#loc98, #loc99]) +#loc150 = loc(callsite(#loc53 at #loc128)) diff --git a/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/__grp__l2norm_fwd_kernel.json b/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/__grp__l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..7f6700570944c6d570c616854136df6b21678d20 --- /dev/null +++ b/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/__grp__l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_fwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/l2norm_fwd_kernel.source", "l2norm_fwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/l2norm_fwd_kernel.ttir", "l2norm_fwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/l2norm_fwd_kernel.ttgir", "l2norm_fwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/l2norm_fwd_kernel.llir", "l2norm_fwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/l2norm_fwd_kernel.ptx", "l2norm_fwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/l2norm_fwd_kernel.cubin", "l2norm_fwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/l2norm_fwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/l2norm_fwd_kernel.cubin b/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/l2norm_fwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..12fd1fc4ca105f402363c59d6b1169b54bdf3721 Binary files /dev/null and b/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/l2norm_fwd_kernel.cubin differ diff --git a/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/l2norm_fwd_kernel.json b/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..638589f1f7c518f23d2dc8d869fc7bb94925d900 --- /dev/null +++ b/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"hash": "086075513d142f286e25e948ea9ac6a0043467fd0888f071e19f024553260629", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 2, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 32, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_fwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/l2norm_fwd_kernel.llir b/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/l2norm_fwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..613b94d07d8ced702c39bf1e6cb755c151616cba --- /dev/null +++ b/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/l2norm_fwd_kernel.llir @@ -0,0 +1,378 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 +@.str.2 = private unnamed_addr constant [17 x i8] c"__CUDA_PREC_SQRT\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_fwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, float %3, i32 %4, ptr addrspace(1) readnone captures(none) %5, ptr addrspace(1) readnone captures(none) %6) local_unnamed_addr #0 !dbg !5 { + %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %9 = zext nneg i32 %8 to i64, !dbg !14 + %10 = shl nuw nsw i64 %9, 3, !dbg !15 + %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !16 + %12 = and i32 %11, 48, !dbg !16 + %13 = lshr exact i32 %12, 4, !dbg !16 + %14 = or disjoint i32 %13, 4, !dbg !16 + %15 = zext nneg i32 %13 to i64, !dbg !17 + %16 = zext nneg i32 %14 to i64, !dbg !17 + %17 = or disjoint i64 %10, %15, !dbg !17 + %18 = or disjoint i64 %10, %16, !dbg !17 + %19 = sext i32 %4 to i64, !dbg !18 + %20 = icmp slt i64 %17, %19, !dbg !18 + %21 = icmp slt i64 %18, %19, !dbg !18 + %22 = shl nuw nsw i32 %11, 3, !dbg !19 + %23 = and i32 %22, 120, !dbg !19 + %24 = shl nuw nsw i64 %17, 7, !dbg !20 + %25 = shl nuw nsw i64 %18, 7, !dbg !20 + %26 = getelementptr bfloat, ptr addrspace(1) %0, i64 %24, !dbg !21 + %27 = getelementptr bfloat, ptr addrspace(1) %0, i64 %25, !dbg !21 + %28 = zext nneg i32 %23 to i64, !dbg !22 + %29 = getelementptr bfloat, ptr addrspace(1) %26, i64 %28, !dbg !22 + %30 = getelementptr bfloat, ptr addrspace(1) %27, i64 %28, !dbg !22 + %31 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %29, i1 %20) #6, !dbg !23 + %32 = extractvalue { i32, i32, i32, i32 } %31, 0, !dbg !23 + %33 = bitcast i32 %32 to <2 x bfloat>, !dbg !23 + %34 = extractvalue { i32, i32, i32, i32 } %31, 1, !dbg !23 + %35 = bitcast i32 %34 to <2 x bfloat>, !dbg !23 + %36 = extractvalue { i32, i32, i32, i32 } %31, 2, !dbg !23 + %37 = bitcast i32 %36 to <2 x bfloat>, !dbg !23 + %38 = extractvalue { i32, i32, i32, i32 } %31, 3, !dbg !23 + %39 = bitcast i32 %38 to <2 x bfloat>, !dbg !23 + %40 = extractelement <2 x bfloat> %33, i64 0, !dbg !23 + %41 = extractelement <2 x bfloat> %33, i64 1, !dbg !23 + %42 = extractelement <2 x bfloat> %35, i64 0, !dbg !23 + %43 = extractelement <2 x bfloat> %35, i64 1, !dbg !23 + %44 = extractelement <2 x bfloat> %37, i64 0, !dbg !23 + %45 = extractelement <2 x bfloat> %37, i64 1, !dbg !23 + %46 = extractelement <2 x bfloat> %39, i64 0, !dbg !23 + %47 = extractelement <2 x bfloat> %39, i64 1, !dbg !23 + %48 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %30, i1 %21) #6, !dbg !23 + %49 = extractvalue { i32, i32, i32, i32 } %48, 0, !dbg !23 + %50 = bitcast i32 %49 to <2 x bfloat>, !dbg !23 + %51 = extractvalue { i32, i32, i32, i32 } %48, 1, !dbg !23 + %52 = bitcast i32 %51 to <2 x bfloat>, !dbg !23 + %53 = extractvalue { i32, i32, i32, i32 } %48, 2, !dbg !23 + %54 = bitcast i32 %53 to <2 x bfloat>, !dbg !23 + %55 = extractvalue { i32, i32, i32, i32 } %48, 3, !dbg !23 + %56 = bitcast i32 %55 to <2 x bfloat>, !dbg !23 + %57 = extractelement <2 x bfloat> %50, i64 0, !dbg !23 + %58 = extractelement <2 x bfloat> %50, i64 1, !dbg !23 + %59 = extractelement <2 x bfloat> %52, i64 0, !dbg !23 + %60 = extractelement <2 x bfloat> %52, i64 1, !dbg !23 + %61 = extractelement <2 x bfloat> %54, i64 0, !dbg !23 + %62 = extractelement <2 x bfloat> %54, i64 1, !dbg !23 + %63 = extractelement <2 x bfloat> %56, i64 0, !dbg !23 + %64 = extractelement <2 x bfloat> %56, i64 1, !dbg !23 + %65 = fpext bfloat %40 to float, !dbg !24 + %66 = fpext bfloat %41 to float, !dbg !24 + %67 = fpext bfloat %42 to float, !dbg !24 + %68 = fpext bfloat %43 to float, !dbg !24 + %69 = fpext bfloat %44 to float, !dbg !24 + %70 = fpext bfloat %45 to float, !dbg !24 + %71 = fpext bfloat %46 to float, !dbg !24 + %72 = fpext bfloat %47 to float, !dbg !24 + %73 = fpext bfloat %57 to float, !dbg !24 + %74 = fpext bfloat %58 to float, !dbg !24 + %75 = fpext bfloat %59 to float, !dbg !24 + %76 = fpext bfloat %60 to float, !dbg !24 + %77 = fpext bfloat %61 to float, !dbg !24 + %78 = fpext bfloat %62 to float, !dbg !24 + %79 = fpext bfloat %63 to float, !dbg !24 + %80 = fpext bfloat %64 to float, !dbg !24 + %81 = fmul float %65, %65, !dbg !25 + %82 = fmul float %66, %66, !dbg !25 + %83 = fmul float %67, %67, !dbg !25 + %84 = fmul float %68, %68, !dbg !25 + %85 = fmul float %69, %69, !dbg !25 + %86 = fmul float %70, %70, !dbg !25 + %87 = fmul float %71, %71, !dbg !25 + %88 = fmul float %72, %72, !dbg !25 + %89 = fmul float %73, %73, !dbg !25 + %90 = fmul float %74, %74, !dbg !25 + %91 = fmul float %75, %75, !dbg !25 + %92 = fmul float %76, %76, !dbg !25 + %93 = fmul float %77, %77, !dbg !25 + %94 = fmul float %78, %78, !dbg !25 + %95 = fmul float %79, %79, !dbg !25 + %96 = fmul float %80, %80, !dbg !25 + %97 = fadd float %81, %82, !dbg !26 + %98 = fadd float %83, %97, !dbg !26 + %99 = fadd float %84, %98, !dbg !26 + %100 = fadd float %85, %99, !dbg !26 + %101 = fadd float %86, %100, !dbg !26 + %102 = fadd float %87, %101, !dbg !26 + %103 = fadd float %88, %102, !dbg !26 + %104 = fadd float %89, %90, !dbg !26 + %105 = fadd float %91, %104, !dbg !26 + %106 = fadd float %92, %105, !dbg !26 + %107 = fadd float %93, %106, !dbg !26 + %108 = fadd float %94, %107, !dbg !26 + %109 = fadd float %95, %108, !dbg !26 + %110 = fadd float %96, %109, !dbg !26 + %111 = bitcast float %103 to i32, !dbg !29 + %112 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %111, i32 8, i32 31), !dbg !29 + %113 = bitcast i32 %112 to float, !dbg !29 + %114 = fadd float %103, %113, !dbg !26 + %115 = bitcast float %114 to i32, !dbg !29 + %116 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %115, i32 4, i32 31), !dbg !29 + %117 = bitcast i32 %116 to float, !dbg !29 + %118 = fadd float %114, %117, !dbg !26 + %119 = bitcast float %118 to i32, !dbg !29 + %120 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %119, i32 2, i32 31), !dbg !29 + %121 = bitcast i32 %120 to float, !dbg !29 + %122 = fadd float %118, %121, !dbg !26 + %123 = bitcast float %122 to i32, !dbg !29 + %124 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %123, i32 1, i32 31), !dbg !29 + %125 = bitcast i32 %124 to float, !dbg !29 + %126 = fadd float %122, %125, !dbg !26 + %127 = bitcast float %110 to i32, !dbg !29 + %128 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %127, i32 8, i32 31), !dbg !29 + %129 = bitcast i32 %128 to float, !dbg !29 + %130 = fadd float %110, %129, !dbg !26 + %131 = bitcast float %130 to i32, !dbg !29 + %132 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %131, i32 4, i32 31), !dbg !29 + %133 = bitcast i32 %132 to float, !dbg !29 + %134 = fadd float %130, %133, !dbg !26 + %135 = bitcast float %134 to i32, !dbg !29 + %136 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %135, i32 2, i32 31), !dbg !29 + %137 = bitcast i32 %136 to float, !dbg !29 + %138 = fadd float %134, %137, !dbg !26 + %139 = bitcast float %138 to i32, !dbg !29 + %140 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %139, i32 1, i32 31), !dbg !29 + %141 = bitcast i32 %140 to float, !dbg !29 + %142 = fadd float %138, %141, !dbg !26 + %143 = fadd float %3, %126, !dbg !32 + %144 = fadd float %3, %142, !dbg !32 + %145 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i = icmp eq i32 %145, 0, !dbg !33 + %146 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i = icmp eq i32 %146, 0, !dbg !33 + br i1 %.not.i, label %152, label %147, !dbg !33 + +147: ; preds = %7 + br i1 %.not1.i, label %150, label %148, !dbg !33 + +148: ; preds = %147 + %149 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %143) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +150: ; preds = %147 + %151 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %143) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +152: ; preds = %7 + br i1 %.not1.i, label %155, label %153, !dbg !33 + +153: ; preds = %152 + %154 = tail call float @llvm.nvvm.sqrt.rn.f(float %143) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +155: ; preds = %152 + %156 = tail call float @llvm.nvvm.sqrt.approx.f(float %143) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +__nv_sqrtf.exit: ; preds = %148, %150, %153, %155 + %.0.i = phi float [ %149, %148 ], [ %151, %150 ], [ %154, %153 ], [ %156, %155 ], !dbg !33 + %157 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i1 = icmp eq i32 %157, 0, !dbg !33 + %158 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i4 = icmp eq i32 %158, 0, !dbg !33 + br i1 %.not.i1, label %164, label %159, !dbg !33 + +159: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %162, label %160, !dbg !33 + +160: ; preds = %159 + %161 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %144) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +162: ; preds = %159 + %163 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %144) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +164: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %167, label %165, !dbg !33 + +165: ; preds = %164 + %166 = tail call float @llvm.nvvm.sqrt.rn.f(float %144) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +167: ; preds = %164 + %168 = tail call float @llvm.nvvm.sqrt.approx.f(float %144) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +__nv_sqrtf.exit5: ; preds = %160, %162, %165, %167 + %.0.i3 = phi float [ %161, %160 ], [ %163, %162 ], [ %166, %165 ], [ %168, %167 ], !dbg !33 + %169 = and i32 %11, 7, !dbg !16 + %170 = zext nneg i32 %169 to i64, !dbg !17 + %171 = or disjoint i64 %10, %170, !dbg !17 + %172 = getelementptr float, ptr addrspace(1) %2, i64 %171, !dbg !34 + %173 = getelementptr bfloat, ptr addrspace(1) %1, i64 %25, !dbg !35 + %174 = getelementptr bfloat, ptr addrspace(1) %173, i64 %28, !dbg !36 + %175 = getelementptr bfloat, ptr addrspace(1) %1, i64 %24, !dbg !35 + %176 = getelementptr bfloat, ptr addrspace(1) %175, i64 %28, !dbg !36 + %177 = icmp slt i64 %171, %19, !dbg !18 + %178 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i), !dbg !37 + %179 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i3), !dbg !37 + %180 = fmul float %178, %65, !dbg !38 + %181 = fmul float %178, %66, !dbg !38 + %182 = fmul float %178, %67, !dbg !38 + %183 = fmul float %178, %68, !dbg !38 + %184 = fmul float %178, %69, !dbg !38 + %185 = fmul float %178, %70, !dbg !38 + %186 = fmul float %178, %71, !dbg !38 + %187 = fmul float %178, %72, !dbg !38 + %188 = fmul float %179, %73, !dbg !38 + %189 = fmul float %179, %74, !dbg !38 + %190 = fmul float %179, %75, !dbg !38 + %191 = fmul float %179, %76, !dbg !38 + %192 = fmul float %179, %77, !dbg !38 + %193 = fmul float %179, %78, !dbg !38 + %194 = fmul float %179, %79, !dbg !38 + %195 = fmul float %179, %80, !dbg !38 + %196 = fptrunc float %180 to bfloat, !dbg !39 + %197 = fptrunc float %181 to bfloat, !dbg !39 + %198 = fptrunc float %182 to bfloat, !dbg !39 + %199 = fptrunc float %183 to bfloat, !dbg !39 + %200 = fptrunc float %184 to bfloat, !dbg !39 + %201 = fptrunc float %185 to bfloat, !dbg !39 + %202 = fptrunc float %186 to bfloat, !dbg !39 + %203 = fptrunc float %187 to bfloat, !dbg !39 + %204 = fptrunc float %188 to bfloat, !dbg !39 + %205 = fptrunc float %189 to bfloat, !dbg !39 + %206 = fptrunc float %190 to bfloat, !dbg !39 + %207 = fptrunc float %191 to bfloat, !dbg !39 + %208 = fptrunc float %192 to bfloat, !dbg !39 + %209 = fptrunc float %193 to bfloat, !dbg !39 + %210 = fptrunc float %194 to bfloat, !dbg !39 + %211 = fptrunc float %195 to bfloat, !dbg !39 + %212 = insertelement <2 x bfloat> poison, bfloat %196, i64 0, !dbg !40 + %213 = insertelement <2 x bfloat> %212, bfloat %197, i64 1, !dbg !40 + %214 = bitcast <2 x bfloat> %213 to i32, !dbg !40 + %215 = insertelement <2 x bfloat> poison, bfloat %198, i64 0, !dbg !40 + %216 = insertelement <2 x bfloat> %215, bfloat %199, i64 1, !dbg !40 + %217 = bitcast <2 x bfloat> %216 to i32, !dbg !40 + %218 = insertelement <2 x bfloat> poison, bfloat %200, i64 0, !dbg !40 + %219 = insertelement <2 x bfloat> %218, bfloat %201, i64 1, !dbg !40 + %220 = bitcast <2 x bfloat> %219 to i32, !dbg !40 + %221 = insertelement <2 x bfloat> poison, bfloat %202, i64 0, !dbg !40 + %222 = insertelement <2 x bfloat> %221, bfloat %203, i64 1, !dbg !40 + %223 = bitcast <2 x bfloat> %222 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %214, i32 %217, i32 %220, i32 %223, ptr addrspace(1) %176, i1 %20) #6, !dbg !40 + %224 = insertelement <2 x bfloat> poison, bfloat %204, i64 0, !dbg !40 + %225 = insertelement <2 x bfloat> %224, bfloat %205, i64 1, !dbg !40 + %226 = bitcast <2 x bfloat> %225 to i32, !dbg !40 + %227 = insertelement <2 x bfloat> poison, bfloat %206, i64 0, !dbg !40 + %228 = insertelement <2 x bfloat> %227, bfloat %207, i64 1, !dbg !40 + %229 = bitcast <2 x bfloat> %228 to i32, !dbg !40 + %230 = insertelement <2 x bfloat> poison, bfloat %208, i64 0, !dbg !40 + %231 = insertelement <2 x bfloat> %230, bfloat %209, i64 1, !dbg !40 + %232 = bitcast <2 x bfloat> %231 to i32, !dbg !40 + %233 = insertelement <2 x bfloat> poison, bfloat %210, i64 0, !dbg !40 + %234 = insertelement <2 x bfloat> %233, bfloat %211, i64 1, !dbg !40 + %235 = bitcast <2 x bfloat> %234 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %226, i32 %229, i32 %232, i32 %235, ptr addrspace(1) %174, i1 %21) #6, !dbg !40 + %236 = lshr exact i32 %12, 1, !dbg !41 + %237 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %236, !dbg !41 + %238 = bitcast float %178 to i32, !dbg !41 + %239 = bitcast float %179 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v2.b32 [ $0 + 0 ], { $1, $2 };", "r,r,r"(ptr addrspace(3) %237, i32 %238, i32 %239) #6, !dbg !41 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !41 + %240 = and i32 %22, 24, !dbg !41 + %241 = and i32 %11, 4, !dbg !41 + %242 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %240, !dbg !41 + %243 = getelementptr inbounds nuw i8, ptr addrspace(3) %242, i32 %241, !dbg !41 + %244 = load i32, ptr addrspace(3) %243, align 4, !dbg !41 + %245 = and i32 %11, 56, !dbg !41 + %246 = icmp eq i32 %245, 0, !dbg !41 + %247 = and i1 %246, %177, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %244, ptr addrspace(1) %172, i1 %247) #6, !dbg !41 + ret void, !dbg !42 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.div.full(float, float) #3 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #4 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #5 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.f(float) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="64" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) } +attributes #4 = { convergent nocallback nounwind } +attributes #5 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #6 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "l2norm_fwd_kernel", linkageName: "l2norm_fwd_kernel", scope: !1, file: !1, line: 69, type: !6, scopeLine: 69, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !10, !11, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 70, column: 24, scope: !5) +!14 = !DILocation(line: 70, column: 30, scope: !5) +!15 = !DILocation(line: 71, column: 16, scope: !5) +!16 = !DILocation(line: 71, column: 34, scope: !5) +!17 = !DILocation(line: 71, column: 21, scope: !5) +!18 = !DILocation(line: 73, column: 16, scope: !5) +!19 = !DILocation(line: 74, column: 30, scope: !5) +!20 = !DILocation(line: 75, column: 29, scope: !5) +!21 = !DILocation(line: 75, column: 14, scope: !5) +!22 = !DILocation(line: 75, column: 33, scope: !5) +!23 = !DILocation(line: 79, column: 18, scope: !5) +!24 = !DILocation(line: 79, column: 47, scope: !5) +!25 = !DILocation(line: 80, column: 38, scope: !5) +!26 = !DILocation(line: 263, column: 15, scope: !27, inlinedAt: !29) +!27 = distinct !DILexicalBlockFile(scope: !5, file: !28, discriminator: 0) +!28 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!29 = !DILocation(line: 293, column: 36, scope: !27, inlinedAt: !30) +!30 = !DILocation(line: 80, column: 43, scope: !31) +!31 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!32 = !DILocation(line: 80, column: 48, scope: !5) +!33 = !DILocation(line: 80, column: 25, scope: !5) +!34 = !DILocation(line: 77, column: 20, scope: !5) +!35 = !DILocation(line: 76, column: 14, scope: !5) +!36 = !DILocation(line: 76, column: 33, scope: !5) +!37 = !DILocation(line: 80, column: 17, scope: !5) +!38 = !DILocation(line: 81, column: 16, scope: !5) +!39 = !DILocation(line: 83, column: 25, scope: !5) +!40 = !DILocation(line: 83, column: 18, scope: !5) +!41 = !DILocation(line: 84, column: 21, scope: !5) +!42 = !DILocation(line: 84, column: 4, scope: !5) diff --git a/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/l2norm_fwd_kernel.ptx b/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/l2norm_fwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..f9207d5b8d1c4a52f67da1a1409492d5285e7e90 --- /dev/null +++ b/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/l2norm_fwd_kernel.ptx @@ -0,0 +1,481 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_fwd_kernel // -- Begin function l2norm_fwd_kernel +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; +.global .align 1 .b8 _$_str_$_2[17] = {95, 95, 67, 85, 68, 65, 95, 80, 82, 69, 67, 95, 83, 81, 82, 84}; + // @l2norm_fwd_kernel +.visible .entry l2norm_fwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_2, + .param .f32 l2norm_fwd_kernel_param_3, + .param .u32 l2norm_fwd_kernel_param_4, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_6 +) +.reqntid 64 +{ + .reg .pred %p<6>; + .reg .b16 %rs<17>; + .reg .b32 %r<107>; + .reg .b64 %rd<25>; + .loc 1 69 0 // l2norm.py:69:0 +$L__func_begin0: + .loc 1 69 0 // l2norm.py:69:0 + +// %bb.0: // %__nv_sqrtf.exit + ld.param.b64 %rd6, [l2norm_fwd_kernel_param_0]; + ld.param.b64 %rd7, [l2norm_fwd_kernel_param_1]; +$L__tmp0: + .loc 1 70 24 // l2norm.py:70:24 + mov.u32 %r22, %ctaid.x; + ld.param.b64 %rd8, [l2norm_fwd_kernel_param_2]; + .loc 1 71 16 // l2norm.py:71:16 + mul.wide.u32 %rd9, %r22, 8; + ld.param.b32 %r23, [l2norm_fwd_kernel_param_3]; + ld.param.s32 %rd10, [l2norm_fwd_kernel_param_4]; + .loc 1 71 34 // l2norm.py:71:34 + mov.u32 %r24, %tid.x; + and.b32 %r25, %r24, 48; + bfe.u32 %r26, %r24, 4, 2; + or.b32 %r27, %r26, 4; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd11, %r26; + cvt.u64.u32 %rd12, %r27; + or.b64 %rd13, %rd9, %rd11; + or.b64 %rd14, %rd9, %rd12; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p1, %rd13, %rd10; + setp.lt.s64 %p2, %rd14, %rd10; + .loc 1 74 30 // l2norm.py:74:30 + shl.b32 %r28, %r24, 3; + and.b32 %r29, %r28, 120; + .loc 1 75 14 // l2norm.py:75:14 + shl.b64 %rd15, %rd13, 8; + add.s64 %rd16, %rd6, %rd15; + shl.b64 %rd17, %rd14, 8; + add.s64 %rd18, %rd6, %rd17; + .loc 1 75 33 // l2norm.py:75:33 + mul.wide.u32 %rd19, %r29, 2; + add.s64 %rd1, %rd16, %rd19; + add.s64 %rd2, %rd18, %rd19; + mov.b32 %r5, 0; + .loc 1 79 18 // l2norm.py:79:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + mov.b32 {%rs5, %rs6}, %r3; + mov.b32 {%rs7, %rs8}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs9, %rs10}, %r6; + mov.b32 {%rs11, %rs12}, %r7; + mov.b32 {%rs13, %rs14}, %r8; + mov.b32 {%rs15, %rs16}, %r9; + .loc 1 79 47 // l2norm.py:79:47 + cvt.f32.bf16 %r30, %rs1; + cvt.f32.bf16 %r31, %rs2; + cvt.f32.bf16 %r32, %rs3; + cvt.f32.bf16 %r33, %rs4; + cvt.f32.bf16 %r34, %rs5; + cvt.f32.bf16 %r35, %rs6; + cvt.f32.bf16 %r36, %rs7; + cvt.f32.bf16 %r37, %rs8; + cvt.f32.bf16 %r38, %rs9; + cvt.f32.bf16 %r39, %rs10; + cvt.f32.bf16 %r40, %rs11; + cvt.f32.bf16 %r41, %rs12; + cvt.f32.bf16 %r42, %rs13; + cvt.f32.bf16 %r43, %rs14; + cvt.f32.bf16 %r44, %rs15; + cvt.f32.bf16 %r45, %rs16; + .loc 1 80 38 // l2norm.py:80:38 + mul.f32 %r46, %r31, %r31; + mul.f32 %r47, %r39, %r39; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + fma.rn.f32.bf16 %r48, %rs1, %rs1, %r46; + fma.rn.f32.bf16 %r49, %rs3, %rs3, %r48; + fma.rn.f32.bf16 %r50, %rs4, %rs4, %r49; + fma.rn.f32.bf16 %r51, %rs5, %rs5, %r50; + fma.rn.f32.bf16 %r52, %rs6, %rs6, %r51; + fma.rn.f32.bf16 %r53, %rs7, %rs7, %r52; + fma.rn.f32.bf16 %r54, %rs8, %rs8, %r53; + fma.rn.f32.bf16 %r55, %rs9, %rs9, %r47; + fma.rn.f32.bf16 %r56, %rs11, %rs11, %r55; + fma.rn.f32.bf16 %r57, %rs12, %rs12, %r56; + fma.rn.f32.bf16 %r58, %rs13, %rs13, %r57; + fma.rn.f32.bf16 %r59, %rs14, %rs14, %r58; + fma.rn.f32.bf16 %r60, %rs15, %rs15, %r59; + fma.rn.f32.bf16 %r61, %rs16, %rs16, %r60; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r62, %r54, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r63, %r54, %r62; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r64, %r63, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r65, %r63, %r64; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r66, %r65, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r67, %r65, %r66; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r68, %r67, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r69, %r67, %r68; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r70, %r61, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r71, %r61, %r70; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r72, %r71, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r73, %r71, %r72; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r74, %r73, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r75, %r73, %r74; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r76, %r75, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r77, %r75, %r76; +$L__tmp18: + .loc 1 80 48 // l2norm.py:80:48 + add.f32 %r78, %r23, %r69; + add.f32 %r79, %r23, %r77; + .loc 1 80 25 // l2norm.py:80:25 + sqrt.approx.ftz.f32 %r80, %r78; + sqrt.approx.ftz.f32 %r81, %r79; + .loc 1 71 34 // l2norm.py:71:34 + and.b32 %r82, %r24, 7; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd20, %r82; + or.b64 %rd21, %rd9, %rd20; + .loc 1 77 20 // l2norm.py:77:20 + shl.b64 %rd22, %rd21, 2; + add.s64 %rd5, %rd8, %rd22; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd23, %rd7, %rd17; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd4, %rd23, %rd19; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd24, %rd7, %rd15; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd3, %rd24, %rd19; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p4, %rd21, %rd10; + mov.b32 %r83, 0f3F800000; + .loc 1 80 17 // l2norm.py:80:17 + div.full.f32 %r19, %r83, %r80; + div.full.f32 %r20, %r83, %r81; + .loc 1 81 16 // l2norm.py:81:16 + mul.f32 %r84, %r19, %r30; + mul.f32 %r85, %r19, %r31; + mul.f32 %r86, %r19, %r32; + mul.f32 %r87, %r19, %r33; + mul.f32 %r88, %r19, %r34; + mul.f32 %r89, %r19, %r35; + mul.f32 %r90, %r19, %r36; + mul.f32 %r91, %r19, %r37; + mul.f32 %r92, %r20, %r38; + mul.f32 %r93, %r20, %r39; + mul.f32 %r94, %r20, %r40; + mul.f32 %r95, %r20, %r41; + mul.f32 %r96, %r20, %r42; + mul.f32 %r97, %r20, %r43; + mul.f32 %r98, %r20, %r44; + mul.f32 %r99, %r20, %r45; + .loc 1 83 18 // l2norm.py:83:18 + cvt.rn.bf16x2.f32 %r10, %r85, %r84; + cvt.rn.bf16x2.f32 %r11, %r87, %r86; + cvt.rn.bf16x2.f32 %r12, %r89, %r88; + cvt.rn.bf16x2.f32 %r13, %r91, %r90; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd3 + 0 ], { %r10, %r11, %r12, %r13 }; + // end inline asm + cvt.rn.bf16x2.f32 %r14, %r93, %r92; + cvt.rn.bf16x2.f32 %r15, %r95, %r94; + cvt.rn.bf16x2.f32 %r16, %r97, %r96; + cvt.rn.bf16x2.f32 %r17, %r99, %r98; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd4 + 0 ], { %r14, %r15, %r16, %r17 }; + // end inline asm + .loc 1 84 21 // l2norm.py:84:21 + shr.u32 %r100, %r25, 1; + mov.b32 %r101, global_smem; + add.s32 %r18, %r101, %r100; + // begin inline asm + st.shared.v2.b32 [ %r18 + 0 ], { %r19, %r20 }; + // end inline asm + bar.sync 0; + and.b32 %r102, %r28, 24; + and.b32 %r103, %r24, 4; + add.s32 %r104, %r101, %r102; + add.s32 %r105, %r104, %r103; + ld.shared.b32 %r21, [%r105]; + and.b32 %r106, %r24, 56; + setp.eq.b32 %p5, %r106, 0; + and.pred %p3, %p5, %p4; + // begin inline asm + @%p3 st.global.b32 [ %rd5 + 0 ], { %r21 }; + // end inline asm + .loc 1 84 4 // l2norm.py:84:4 + ret; +$L__tmp19: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp18 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 80 // DW_AT_call_line +.b8 43 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp18 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/l2norm_fwd_kernel.source b/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/l2norm_fwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..c7c5a735c81381ccb0efd0d2a8f8d86240068e11 --- /dev/null +++ b/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/l2norm_fwd_kernel.source @@ -0,0 +1,172 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc38 = loc(unknown) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc45 = loc("x"(#loc)) +#loc46 = loc("y"(#loc)) +#loc47 = loc("rstd"(#loc)) +#loc48 = loc("eps"(#loc)) +#loc49 = loc("T"(#loc)) +#loc81 = loc("input"(#loc36)) +#loc82 = loc("a"(#loc41)) +#loc83 = loc("b"(#loc41)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc50) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc51) + %o_t = arith.constant 8 : i32 loc(#loc52) + %o_t_1 = arith.constant 8 : i64 loc(#loc52) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc52) + %o_t_3 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32> loc(#loc53) + %o_t_4 = arith.extsi %o_t_3 : tensor<8xi32> to tensor<8xi64> loc(#loc54) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<8xi64> loc(#loc54) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<8xi64> loc(#loc54) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc55) + %m_t = arith.extsi %T : i32 to i64 loc(#loc56) + %m_t_7 = tt.splat %m_t : i64 -> tensor<8xi64> loc(#loc56) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<8xi64> loc(#loc56) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<8xi1> -> tensor<8x1xi1> loc(#loc57) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc58) + %m_x_10 = arith.constant 128 : i32 loc(#loc59) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc59) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc59) + %m_x_13 = tt.broadcast %m_x : tensor<8x1xi1> -> tensor<8x128xi1> loc(#loc60) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<8x128xi1> loc(#loc60) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<8x128xi1> loc(#loc60) + %p_x = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc61) + %p_x_16 = arith.constant 128 : i32 loc(#loc62) + %p_x_17 = arith.constant 128 : i64 loc(#loc62) + %p_x_18 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc62) + %p_x_19 = arith.muli %p_x, %p_x_18 : tensor<8x1xi64> loc(#loc62) + %p_x_20 = tt.splat %x : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc63) + %p_x_21 = tt.addptr %p_x_20, %p_x_19 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc63) + %p_x_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc64) + %p_x_23 = tt.broadcast %p_x_21 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc65) + %p_x_24 = tt.broadcast %p_x_22 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc65) + %p_x_25 = tt.addptr %p_x_23, %p_x_24 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc65) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc66) + %p_y_26 = arith.constant 128 : i32 loc(#loc67) + %p_y_27 = arith.constant 128 : i64 loc(#loc67) + %p_y_28 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc67) + %p_y_29 = arith.muli %p_y, %p_y_28 : tensor<8x1xi64> loc(#loc67) + %p_y_30 = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc68) + %p_y_31 = tt.addptr %p_y_30, %p_y_29 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc68) + %p_y_32 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc69) + %p_y_33 = tt.broadcast %p_y_31 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc70) + %p_y_34 = tt.broadcast %p_y_32 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc70) + %p_y_35 = tt.addptr %p_y_33, %p_y_34 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc70) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr> loc(#loc71) + %p_rstd_36 = tt.addptr %p_rstd, %o_t_6 : tensor<8x!tt.ptr>, tensor<8xi64> loc(#loc71) + %b_x = arith.constant 0.000000e+00 : f32 loc(#loc72) + %b_x_37 = arith.constant dense<0.000000e+00> : tensor<8x128xf32> loc(#loc72) + %b_x_38 = arith.truncf %b_x_37 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc72) + %b_x_39 = tt.load %p_x_25, %m_x_15, %b_x_38 : tensor<8x128x!tt.ptr> loc(#loc72) + %b_x_40 = arith.extf %b_x_39 : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc73) + %b_rstd = arith.mulf %b_x_40, %b_x_40 : tensor<8x128xf32> loc(#loc74) + %b_rstd_41 = tt.call @triton.language.standard.sum__fp32S8_128S_c1_cFalse_cNone(%b_rstd) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc75) + %b_rstd_42 = tt.splat %eps : f32 -> tensor<8xf32> loc(#loc76) + %b_rstd_43 = arith.addf %b_rstd_41, %b_rstd_42 : tensor<8xf32> loc(#loc76) + %b_rstd_44 = math.sqrt %b_rstd_43 : tensor<8xf32> loc(#loc77) + %b_rstd_45 = arith.constant 1 : i32 loc(#loc78) + %b_rstd_46 = arith.constant 1.000000e+00 : f32 loc(#loc78) + %b_rstd_47 = arith.constant dense<1.000000e+00> : tensor<8xf32> loc(#loc78) + %b_rstd_48 = arith.divf %b_rstd_47, %b_rstd_44 : tensor<8xf32> loc(#loc78) + %b_y = tt.expand_dims %b_rstd_48 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc79) + %b_y_49 = tt.broadcast %b_y : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc80) + %b_y_50 = arith.mulf %b_x_40, %b_y_49 : tensor<8x128xf32> loc(#loc80) + %0 = arith.truncf %b_y_50 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc32) + tt.store %p_y_35, %0, %m_x_15 : tensor<8x128x!tt.ptr> loc(#loc33) + tt.store %p_rstd_36, %b_rstd_48, %m_t_8 : tensor<8x!tt.ptr> loc(#loc34) + tt.return loc(#loc35) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S8_128S_c1_cFalse_cNone(%input: tensor<8x128xf32> loc("input"(#loc36))) -> tensor<8xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc37) + tt.reduce.return %2 : f32 loc(#loc37) + }) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc37) + tt.return %0 : tensor<8xf32> loc(#loc39) + ^bb1: // no predecessors + %1 = ub.poison : tensor<8xf32> loc(#loc40) + tt.return %1 : tensor<8xf32> loc(#loc40) + } loc(#loc36) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc41)), %b: f32 loc("b"(#loc41))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc42) + tt.return %0 : f32 loc(#loc43) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc44) + tt.return %1 : f32 loc(#loc44) + } loc(#loc41) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:18) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:29) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:37) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc50 = loc("i_t"(#loc1)) +#loc51 = loc("i_t"(#loc2)) +#loc52 = loc("o_t"(#loc3)) +#loc53 = loc("o_t"(#loc4)) +#loc54 = loc("o_t"(#loc5)) +#loc55 = loc("o_d"(#loc6)) +#loc56 = loc("m_t"(#loc7)) +#loc57 = loc("m_x"(#loc8)) +#loc58 = loc("m_x"(#loc9)) +#loc59 = loc("m_x"(#loc10)) +#loc60 = loc("m_x"(#loc11)) +#loc61 = loc("p_x"(#loc12)) +#loc62 = loc("p_x"(#loc13)) +#loc63 = loc("p_x"(#loc14)) +#loc64 = loc("p_x"(#loc15)) +#loc65 = loc("p_x"(#loc16)) +#loc66 = loc("p_y"(#loc17)) +#loc67 = loc("p_y"(#loc18)) +#loc68 = loc("p_y"(#loc19)) +#loc69 = loc("p_y"(#loc20)) +#loc70 = loc("p_y"(#loc21)) +#loc71 = loc("p_rstd"(#loc22)) +#loc72 = loc("b_x"(#loc23)) +#loc73 = loc("b_x"(#loc24)) +#loc74 = loc("b_rstd"(#loc25)) +#loc75 = loc("b_rstd"(#loc26)) +#loc76 = loc("b_rstd"(#loc27)) +#loc77 = loc("b_rstd"(#loc28)) +#loc78 = loc("b_rstd"(#loc29)) +#loc79 = loc("b_y"(#loc30)) +#loc80 = loc("b_y"(#loc31)) diff --git a/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/l2norm_fwd_kernel.ttgir b/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/l2norm_fwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..c6d3fdcbb1bdd697f920b817b4d4e82b8e993287 --- /dev/null +++ b/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/l2norm_fwd_kernel.ttgir @@ -0,0 +1,135 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [2, 1], order = [1, 0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [2], order = [0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc1 = loc(unknown) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc34 = loc("x"(#loc)) +#loc35 = loc("y"(#loc)) +#loc36 = loc("rstd"(#loc)) +#loc37 = loc("eps"(#loc)) +#loc38 = loc("T"(#loc)) +#loc59 = loc("b_rstd"(#loc23)) +#loc66 = loc(callsite(#loc1 at #loc59)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 2 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<1.000000e+00> : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<8x128xbf16, #blocked> loc(#loc1) + %c8_i64 = arith.constant 8 : i64 loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<8x1xi64, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc39) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc40) + %o_t = arith.muli %i_t_3, %c8_i64 : i64 loc(#loc41) + %o_t_4 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc42) + %o_t_5 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32, #blocked1> loc(#loc42) + %o_t_6 = arith.extsi %o_t_4 : tensor<8xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_7 = arith.extsi %o_t_5 : tensor<8xi32, #blocked1> to tensor<8xi64, #blocked1> loc(#loc43) + %o_t_8 = tt.splat %o_t : i64 -> tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_9 = tt.splat %o_t : i64 -> tensor<8xi64, #blocked1> loc(#loc43) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<8xi64, #blocked1> loc(#loc43) + %m_t = arith.extsi %T : i32 to i64 loc(#loc44) + %m_t_12 = tt.splat %m_t : i64 -> tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_13 = tt.splat %m_t : i64 -> tensor<8xi64, #blocked1> loc(#loc44) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<8xi64, #blocked1> loc(#loc44) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<8xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xi1, #blocked> loc(#loc45) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc46) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc46) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_1 : tensor<1x128xi32, #blocked> loc(#loc47) + %m_x_19 = tt.broadcast %m_x : tensor<8x1xi1, #blocked> -> tensor<8x128xi1, #blocked> loc(#loc48) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<8x128xi1, #blocked> loc(#loc48) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<8x128xi1, #blocked> loc(#loc48) + %p_x = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xi64, #blocked> loc(#loc49) + %p_x_22 = arith.muli %p_x, %cst_2 : tensor<8x1xi64, #blocked> loc(#loc50) + %p_x_23 = tt.splat %x : !tt.ptr -> tensor<8x1x!tt.ptr, #blocked> loc(#loc51) + %p_x_24 = tt.addptr %p_x_23, %p_x_22 : tensor<8x1x!tt.ptr, #blocked>, tensor<8x1xi64, #blocked> loc(#loc51) + %p_x_25 = tt.broadcast %p_x_24 : tensor<8x1x!tt.ptr, #blocked> -> tensor<8x128x!tt.ptr, #blocked> loc(#loc52) + %p_x_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<8x128xi32, #blocked> loc(#loc52) + %p_x_27 = tt.addptr %p_x_25, %p_x_26 : tensor<8x128x!tt.ptr, #blocked>, tensor<8x128xi32, #blocked> loc(#loc52) + %p_y = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr, #blocked> loc(#loc53) + %p_y_28 = tt.addptr %p_y, %p_x_22 : tensor<8x1x!tt.ptr, #blocked>, tensor<8x1xi64, #blocked> loc(#loc53) + %p_y_29 = tt.broadcast %p_y_28 : tensor<8x1x!tt.ptr, #blocked> -> tensor<8x128x!tt.ptr, #blocked> loc(#loc54) + %p_y_30 = tt.addptr %p_y_29, %p_x_26 : tensor<8x128x!tt.ptr, #blocked>, tensor<8x128xi32, #blocked> loc(#loc54) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr, #blocked1> loc(#loc55) + %p_rstd_31 = tt.addptr %p_rstd, %o_t_11 : tensor<8x!tt.ptr, #blocked1>, tensor<8xi64, #blocked1> loc(#loc55) + %b_x = tt.load %p_x_27, %m_x_21, %cst_0 : tensor<8x128x!tt.ptr, #blocked> loc(#loc56) + %b_x_32 = arith.extf %b_x : tensor<8x128xbf16, #blocked> to tensor<8x128xf32, #blocked> loc(#loc57) + %b_rstd = arith.mulf %b_x_32, %b_x_32 : tensor<8x128xf32, #blocked> loc(#loc58) + %b_rstd_33 = "tt.reduce"(%b_rstd) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_40: f32 loc(callsite(#loc1 at #loc59)), %b_rstd_41: f32 loc(callsite(#loc1 at #loc59))): + %b_rstd_42 = arith.addf %b_rstd_40, %b_rstd_41 : f32 loc(#loc67) + tt.reduce.return %b_rstd_42 : f32 loc(#loc65) + }) : (tensor<8x128xf32, #blocked>) -> tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc65) + %b_rstd_34 = tt.splat %eps : f32 -> tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_36 = math.sqrt %b_rstd_35 : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc61) + %b_rstd_37 = arith.divf %cst, %b_rstd_36 : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc62) + %b_y = tt.expand_dims %b_rstd_37 {axis = 1 : i32} : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xf32, #blocked> loc(#loc63) + %b_y_38 = tt.broadcast %b_y : tensor<8x1xf32, #blocked> -> tensor<8x128xf32, #blocked> loc(#loc64) + %b_y_39 = arith.mulf %b_x_32, %b_y_38 : tensor<8x128xf32, #blocked> loc(#loc64) + %0 = arith.truncf %b_y_39 : tensor<8x128xf32, #blocked> to tensor<8x128xbf16, #blocked> loc(#loc30) + tt.store %p_y_30, %0, %m_x_21 : tensor<8x128x!tt.ptr, #blocked> loc(#loc31) + %1 = ttg.convert_layout %b_rstd_37 : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8xf32, #blocked1> loc(#loc32) + tt.store %p_rstd_31, %1, %m_t_15 : tensor<8x!tt.ptr, #blocked1> loc(#loc32) + tt.return loc(#loc33) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc39 = loc("i_t"(#loc2)) +#loc40 = loc("i_t"(#loc3)) +#loc41 = loc("o_t"(#loc4)) +#loc42 = loc("o_t"(#loc5)) +#loc43 = loc("o_t"(#loc6)) +#loc44 = loc("m_t"(#loc7)) +#loc45 = loc("m_x"(#loc8)) +#loc46 = loc("m_x"(#loc9)) +#loc47 = loc("m_x"(#loc10)) +#loc48 = loc("m_x"(#loc11)) +#loc49 = loc("p_x"(#loc12)) +#loc50 = loc("p_x"(#loc13)) +#loc51 = loc("p_x"(#loc14)) +#loc52 = loc("p_x"(#loc15)) +#loc53 = loc("p_y"(#loc16)) +#loc54 = loc("p_y"(#loc17)) +#loc55 = loc("p_rstd"(#loc18)) +#loc56 = loc("b_x"(#loc19)) +#loc57 = loc("b_x"(#loc20)) +#loc58 = loc("b_rstd"(#loc21)) +#loc60 = loc("b_rstd"(#loc25)) +#loc61 = loc("b_rstd"(#loc26)) +#loc62 = loc("b_rstd"(#loc27)) +#loc63 = loc("b_y"(#loc28)) +#loc64 = loc("b_y"(#loc29)) +#loc65 = loc(callsite(#loc22 at #loc59)) +#loc67 = loc(callsite(#loc24 at #loc65)) diff --git a/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/l2norm_fwd_kernel.ttir b/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/l2norm_fwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..8bbf17da02489a0d195ed0bd020180f9e4ac2496 --- /dev/null +++ b/triton/gpu0/BBQHKUJ5CQXSQ3RF5FEOVGWGUACDIZ75BCEPA4PBT4BEKUZGAYUQ/l2norm_fwd_kernel.ttir @@ -0,0 +1,128 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc3 = loc(unknown) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc35 = loc("x"(#loc)) +#loc36 = loc("y"(#loc)) +#loc37 = loc("rstd"(#loc)) +#loc38 = loc("eps"(#loc)) +#loc39 = loc("T"(#loc)) +#loc62 = loc("b_rstd"(#loc25)) +#loc68 = loc(callsite(#loc3 at #loc62)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %b_x = arith.constant dense<0.000000e+00> : tensor<8x128xbf16> loc(#loc40) + %b_rstd = arith.constant dense<1.000000e+00> : tensor<8xf32> loc(#loc41) + %cst = arith.constant dense<128> : tensor<8x1xi64> loc(#loc3) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc42) + %o_t = arith.constant 8 : i64 loc(#loc43) + %i_t = tt.get_program_id x : i32 loc(#loc44) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc45) + %o_t_1 = arith.muli %i_t_0, %o_t : i64 loc(#loc43) + %o_t_2 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32> loc(#loc46) + %o_t_3 = arith.extsi %o_t_2 : tensor<8xi32> to tensor<8xi64> loc(#loc47) + %o_t_4 = tt.splat %o_t_1 : i64 -> tensor<8xi64> loc(#loc47) + %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<8xi64> loc(#loc47) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc48) + %m_t = arith.extsi %T : i32 to i64 loc(#loc49) + %m_t_6 = tt.splat %m_t : i64 -> tensor<8xi64> loc(#loc49) + %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<8xi64> loc(#loc49) + %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<8xi1> -> tensor<8x1xi1> loc(#loc50) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc51) + %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc42) + %m_x_11 = tt.broadcast %m_x_8 : tensor<8x1xi1> -> tensor<8x128xi1> loc(#loc52) + %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<8x128xi1> loc(#loc52) + %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<8x128xi1> loc(#loc52) + %p_x = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc53) + %p_x_14 = arith.muli %p_x, %cst : tensor<8x1xi64> loc(#loc54) + %p_x_15 = tt.splat %x : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc55) + %p_x_16 = tt.addptr %p_x_15, %p_x_14 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc55) + %p_x_17 = tt.broadcast %p_x_16 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc56) + %p_x_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc56) + %p_x_19 = tt.addptr %p_x_17, %p_x_18 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc56) + %p_y = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc57) + %p_y_20 = tt.addptr %p_y, %p_x_14 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc57) + %p_y_21 = tt.broadcast %p_y_20 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc58) + %p_y_22 = tt.addptr %p_y_21, %p_x_18 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr> loc(#loc59) + %p_rstd_23 = tt.addptr %p_rstd, %o_t_5 : tensor<8x!tt.ptr>, tensor<8xi64> loc(#loc59) + %b_x_24 = tt.load %p_x_19, %m_x_13, %b_x : tensor<8x128x!tt.ptr> loc(#loc40) + %b_x_25 = arith.extf %b_x_24 : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc60) + %b_rstd_26 = arith.mulf %b_x_25, %b_x_25 : tensor<8x128xf32> loc(#loc61) + %b_rstd_27 = "tt.reduce"(%b_rstd_26) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_34: f32 loc(callsite(#loc3 at #loc62)), %b_rstd_35: f32 loc(callsite(#loc3 at #loc62))): + %b_rstd_36 = arith.addf %b_rstd_34, %b_rstd_35 : f32 loc(#loc69) + tt.reduce.return %b_rstd_36 : f32 loc(#loc67) + }) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc67) + %b_rstd_28 = tt.splat %eps : f32 -> tensor<8xf32> loc(#loc63) + %b_rstd_29 = arith.addf %b_rstd_27, %b_rstd_28 : tensor<8xf32> loc(#loc63) + %b_rstd_30 = math.sqrt %b_rstd_29 : tensor<8xf32> loc(#loc64) + %b_rstd_31 = arith.divf %b_rstd, %b_rstd_30 : tensor<8xf32> loc(#loc41) + %b_y = tt.expand_dims %b_rstd_31 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc65) + %b_y_32 = tt.broadcast %b_y : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc66) + %b_y_33 = arith.mulf %b_x_25, %b_y_32 : tensor<8x128xf32> loc(#loc66) + %0 = arith.truncf %b_y_33 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc31) + tt.store %p_y_22, %0, %m_x_13 : tensor<8x128x!tt.ptr> loc(#loc32) + tt.store %p_rstd_23, %b_rstd_31, %m_t_7 : tensor<8x!tt.ptr> loc(#loc33) + tt.return loc(#loc34) + } loc(#loc) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc40 = loc("b_x"(#loc1)) +#loc41 = loc("b_rstd"(#loc2)) +#loc42 = loc("m_x"(#loc4)) +#loc43 = loc("o_t"(#loc5)) +#loc44 = loc("i_t"(#loc6)) +#loc45 = loc("i_t"(#loc7)) +#loc46 = loc("o_t"(#loc8)) +#loc47 = loc("o_t"(#loc9)) +#loc48 = loc("o_d"(#loc10)) +#loc49 = loc("m_t"(#loc11)) +#loc50 = loc("m_x"(#loc12)) +#loc51 = loc("m_x"(#loc13)) +#loc52 = loc("m_x"(#loc14)) +#loc53 = loc("p_x"(#loc15)) +#loc54 = loc("p_x"(#loc16)) +#loc55 = loc("p_x"(#loc17)) +#loc56 = loc("p_x"(#loc18)) +#loc57 = loc("p_y"(#loc19)) +#loc58 = loc("p_y"(#loc20)) +#loc59 = loc("p_rstd"(#loc21)) +#loc60 = loc("b_x"(#loc22)) +#loc61 = loc("b_rstd"(#loc23)) +#loc63 = loc("b_rstd"(#loc27)) +#loc64 = loc("b_rstd"(#loc28)) +#loc65 = loc("b_y"(#loc29)) +#loc66 = loc("b_y"(#loc30)) +#loc67 = loc(callsite(#loc24 at #loc62)) +#loc69 = loc(callsite(#loc26 at #loc67)) diff --git a/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/__grp__l2norm_bwd_kernel.json b/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/__grp__l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..c9891a599895cd4cd0eb12f80c4108b56621c869 --- /dev/null +++ b/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/__grp__l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_bwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.source", "l2norm_bwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.ttir", "l2norm_bwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.ttgir", "l2norm_bwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.llir", "l2norm_bwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.ptx", "l2norm_bwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.cubin", "l2norm_bwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.cubin b/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..d144bd686bcb0724cb3f120aa07fb76de9d9095e --- /dev/null +++ b/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.cubin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:891a986d506009060f5cc64e4ab10f2b7bbcaa1fbdb0463144c7091e004da2f5 +size 241800 diff --git a/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.json b/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..7be5485354b8ffe7dd157afc3ec5175922807d43 --- /dev/null +++ b/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"hash": "08e442931eaf32d97c54446575a4dd4cb68096eb485b64ecdb61a2f886f99847", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 2, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 512, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_bwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.llir b/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..a77fd8a23fa93b6b096cc3aab501ebbd80e2193c --- /dev/null +++ b/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.llir @@ -0,0 +1,4790 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_bwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, ptr addrspace(1) %3, float %4, i32 %5, ptr addrspace(1) readnone captures(none) %6, ptr addrspace(1) readnone captures(none) %7) local_unnamed_addr #0 !dbg !4 { + %9 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !12 + %10 = zext nneg i32 %9 to i64, !dbg !13 + %11 = shl nuw nsw i64 %10, 7, !dbg !14 + %12 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !15 + %13 = lshr i32 %12, 4, !dbg !15 + %14 = and i32 %13, 3, !dbg !15 + %15 = or disjoint i32 %14, 4, !dbg !15 + %16 = shl nuw nsw i32 %12, 1, !dbg !15 + %17 = zext nneg i32 %14 to i64, !dbg !16 + %18 = zext nneg i32 %15 to i64, !dbg !16 + %19 = or disjoint i64 %11, %17, !dbg !16 + %20 = or disjoint i64 %11, %18, !dbg !16 + %21 = sext i32 %5 to i64, !dbg !17 + %22 = icmp slt i64 %19, %21, !dbg !17 + %23 = icmp slt i64 %20, %21, !dbg !17 + %24 = insertelement <16 x i32> poison, i32 %14, i64 0, !dbg !15 + %25 = insertelement <16 x i32> %24, i32 %13, i64 1, !dbg !15 + %26 = insertelement <8 x i32> poison, i32 %14, i64 0, !dbg !15 + %27 = insertelement <4 x i32> poison, i32 %14, i64 0, !dbg !15 + %28 = or disjoint i32 %14, 120, !dbg !15 + %29 = or i32 %13, 124, !dbg !15 + %30 = and i32 %16, 126, !dbg !15 + %31 = or disjoint i32 %30, 1, !dbg !15 + %32 = insertelement <32 x i32> poison, i32 %31, i64 0, !dbg !16 + %33 = insertelement <32 x i32> %32, i32 %30, i64 1, !dbg !16 + %34 = insertelement <32 x i32> %33, i32 %29, i64 2, !dbg !16 + %35 = insertelement <32 x i32> %34, i32 %28, i64 3, !dbg !16 + %36 = shufflevector <16 x i32> %25, <16 x i32> poison, <32 x i32> , !dbg !16 + %37 = or <32 x i32> %36, , !dbg !16 + %38 = shufflevector <32 x i32> %35, <32 x i32> %37, <32 x i32> , !dbg !16 + %39 = shufflevector <8 x i32> %26, <8 x i32> poison, <32 x i32> , !dbg !16 + %40 = or disjoint <32 x i32> %39, , !dbg !16 + %41 = shufflevector <32 x i32> %38, <32 x i32> %40, <32 x i32> , !dbg !16 + %42 = shufflevector <4 x i32> %27, <4 x i32> poison, <32 x i32> , !dbg !16 + %43 = or disjoint <32 x i32> %42, , !dbg !16 + %44 = shufflevector <32 x i32> %41, <32 x i32> %43, <32 x i32> , !dbg !16 + %45 = zext <32 x i32> %44 to <32 x i64>, !dbg !16 + %46 = insertelement <32 x i64> poison, i64 %11, i64 0, !dbg !16 + %47 = shufflevector <32 x i64> %46, <32 x i64> poison, <32 x i32> zeroinitializer, !dbg !16 + %48 = or disjoint <32 x i64> %47, %45, !dbg !16 + %49 = insertelement <32 x i32> poison, i32 %5, i64 0, !dbg !17 + %50 = shufflevector <32 x i32> %49, <32 x i32> poison, <32 x i32> zeroinitializer, !dbg !17 + %51 = sext <32 x i32> %50 to <32 x i64>, !dbg !17 + %52 = icmp slt <32 x i64> %48, %51, !dbg !17 + %53 = shl nuw nsw i32 %12, 3, !dbg !18 + %54 = and i32 %53, 120, !dbg !18 + %55 = shl nuw nsw i64 %19, 7, !dbg !19 + %56 = shl nuw nsw i64 %20, 7, !dbg !19 + %57 = extractelement <32 x i64> %48, i64 31, !dbg !19 + %58 = shl nuw nsw i64 %57, 7, !dbg !19 + %59 = extractelement <32 x i64> %48, i64 30, !dbg !19 + %60 = shl nuw nsw i64 %59, 7, !dbg !19 + %61 = extractelement <32 x i64> %48, i64 29, !dbg !19 + %62 = shl nuw nsw i64 %61, 7, !dbg !19 + %63 = extractelement <32 x i64> %48, i64 28, !dbg !19 + %64 = shl nuw nsw i64 %63, 7, !dbg !19 + %65 = extractelement <32 x i64> %48, i64 27, !dbg !19 + %66 = shl nuw nsw i64 %65, 7, !dbg !19 + %67 = extractelement <32 x i64> %48, i64 26, !dbg !19 + %68 = shl nuw nsw i64 %67, 7, !dbg !19 + %69 = extractelement <32 x i64> %48, i64 25, !dbg !19 + %70 = shl nuw nsw i64 %69, 7, !dbg !19 + %71 = extractelement <32 x i64> %48, i64 24, !dbg !19 + %72 = shl nuw nsw i64 %71, 7, !dbg !19 + %73 = extractelement <32 x i64> %48, i64 23, !dbg !19 + %74 = shl nuw nsw i64 %73, 7, !dbg !19 + %75 = extractelement <32 x i64> %48, i64 22, !dbg !19 + %76 = shl nuw nsw i64 %75, 7, !dbg !19 + %77 = extractelement <32 x i64> %48, i64 21, !dbg !19 + %78 = shl nuw nsw i64 %77, 7, !dbg !19 + %79 = extractelement <32 x i64> %48, i64 20, !dbg !19 + %80 = shl nuw nsw i64 %79, 7, !dbg !19 + %81 = extractelement <32 x i64> %48, i64 19, !dbg !19 + %82 = shl nuw nsw i64 %81, 7, !dbg !19 + %83 = extractelement <32 x i64> %48, i64 18, !dbg !19 + %84 = shl nuw nsw i64 %83, 7, !dbg !19 + %85 = extractelement <32 x i64> %48, i64 17, !dbg !19 + %86 = shl nuw nsw i64 %85, 7, !dbg !19 + %87 = extractelement <32 x i64> %48, i64 16, !dbg !19 + %88 = shl nuw nsw i64 %87, 7, !dbg !19 + %89 = extractelement <32 x i64> %48, i64 15, !dbg !19 + %90 = shl nuw nsw i64 %89, 7, !dbg !19 + %91 = extractelement <32 x i64> %48, i64 14, !dbg !19 + %92 = shl nuw nsw i64 %91, 7, !dbg !19 + %93 = extractelement <32 x i64> %48, i64 13, !dbg !19 + %94 = shl nuw nsw i64 %93, 7, !dbg !19 + %95 = extractelement <32 x i64> %48, i64 12, !dbg !19 + %96 = shl nuw nsw i64 %95, 7, !dbg !19 + %97 = extractelement <32 x i64> %48, i64 11, !dbg !19 + %98 = shl nuw nsw i64 %97, 7, !dbg !19 + %99 = extractelement <32 x i64> %48, i64 10, !dbg !19 + %100 = shl nuw nsw i64 %99, 7, !dbg !19 + %101 = extractelement <32 x i64> %48, i64 9, !dbg !19 + %102 = shl nuw nsw i64 %101, 7, !dbg !19 + %103 = extractelement <32 x i64> %48, i64 8, !dbg !19 + %104 = shl nuw nsw i64 %103, 7, !dbg !19 + %105 = extractelement <32 x i64> %48, i64 7, !dbg !19 + %106 = shl nuw nsw i64 %105, 7, !dbg !19 + %107 = extractelement <32 x i64> %48, i64 6, !dbg !19 + %108 = shl nuw nsw i64 %107, 7, !dbg !19 + %109 = extractelement <32 x i64> %48, i64 5, !dbg !19 + %110 = shl nuw nsw i64 %109, 7, !dbg !19 + %111 = extractelement <32 x i64> %48, i64 4, !dbg !19 + %112 = shl nuw nsw i64 %111, 7, !dbg !19 + %113 = extractelement <32 x i64> %48, i64 3, !dbg !19 + %114 = shl nuw nsw i64 %113, 7, !dbg !19 + %115 = extractelement <32 x i64> %48, i64 2, !dbg !19 + %116 = shl nuw nsw i64 %115, 7, !dbg !19 + %117 = getelementptr bfloat, ptr addrspace(1) %0, i64 %55, !dbg !20 + %118 = getelementptr bfloat, ptr addrspace(1) %0, i64 %56, !dbg !20 + %119 = getelementptr bfloat, ptr addrspace(1) %0, i64 %58, !dbg !20 + %120 = getelementptr bfloat, ptr addrspace(1) %0, i64 %60, !dbg !20 + %121 = getelementptr bfloat, ptr addrspace(1) %0, i64 %62, !dbg !20 + %122 = getelementptr bfloat, ptr addrspace(1) %0, i64 %64, !dbg !20 + %123 = getelementptr bfloat, ptr addrspace(1) %0, i64 %66, !dbg !20 + %124 = getelementptr bfloat, ptr addrspace(1) %0, i64 %68, !dbg !20 + %125 = getelementptr bfloat, ptr addrspace(1) %0, i64 %70, !dbg !20 + %126 = getelementptr bfloat, ptr addrspace(1) %0, i64 %72, !dbg !20 + %127 = getelementptr bfloat, ptr addrspace(1) %0, i64 %74, !dbg !20 + %128 = getelementptr bfloat, ptr addrspace(1) %0, i64 %76, !dbg !20 + %129 = getelementptr bfloat, ptr addrspace(1) %0, i64 %78, !dbg !20 + %130 = getelementptr bfloat, ptr addrspace(1) %0, i64 %80, !dbg !20 + %131 = getelementptr bfloat, ptr addrspace(1) %0, i64 %82, !dbg !20 + %132 = getelementptr bfloat, ptr addrspace(1) %0, i64 %84, !dbg !20 + %133 = getelementptr bfloat, ptr addrspace(1) %0, i64 %86, !dbg !20 + %134 = getelementptr bfloat, ptr addrspace(1) %0, i64 %88, !dbg !20 + %135 = getelementptr bfloat, ptr addrspace(1) %0, i64 %90, !dbg !20 + %136 = getelementptr bfloat, ptr addrspace(1) %0, i64 %92, !dbg !20 + %137 = getelementptr bfloat, ptr addrspace(1) %0, i64 %94, !dbg !20 + %138 = getelementptr bfloat, ptr addrspace(1) %0, i64 %96, !dbg !20 + %139 = getelementptr bfloat, ptr addrspace(1) %0, i64 %98, !dbg !20 + %140 = getelementptr bfloat, ptr addrspace(1) %0, i64 %100, !dbg !20 + %141 = getelementptr bfloat, ptr addrspace(1) %0, i64 %102, !dbg !20 + %142 = getelementptr bfloat, ptr addrspace(1) %0, i64 %104, !dbg !20 + %143 = getelementptr bfloat, ptr addrspace(1) %0, i64 %106, !dbg !20 + %144 = getelementptr bfloat, ptr addrspace(1) %0, i64 %108, !dbg !20 + %145 = getelementptr bfloat, ptr addrspace(1) %0, i64 %110, !dbg !20 + %146 = getelementptr bfloat, ptr addrspace(1) %0, i64 %112, !dbg !20 + %147 = getelementptr bfloat, ptr addrspace(1) %0, i64 %114, !dbg !20 + %148 = getelementptr bfloat, ptr addrspace(1) %0, i64 %116, !dbg !20 + %149 = zext nneg i32 %54 to i64, !dbg !21 + %150 = getelementptr bfloat, ptr addrspace(1) %117, i64 %149, !dbg !21 + %151 = getelementptr bfloat, ptr addrspace(1) %118, i64 %149, !dbg !21 + %152 = getelementptr bfloat, ptr addrspace(1) %119, i64 %149, !dbg !21 + %153 = getelementptr bfloat, ptr addrspace(1) %120, i64 %149, !dbg !21 + %154 = getelementptr bfloat, ptr addrspace(1) %121, i64 %149, !dbg !21 + %155 = getelementptr bfloat, ptr addrspace(1) %122, i64 %149, !dbg !21 + %156 = getelementptr bfloat, ptr addrspace(1) %123, i64 %149, !dbg !21 + %157 = getelementptr bfloat, ptr addrspace(1) %124, i64 %149, !dbg !21 + %158 = getelementptr bfloat, ptr addrspace(1) %125, i64 %149, !dbg !21 + %159 = getelementptr bfloat, ptr addrspace(1) %126, i64 %149, !dbg !21 + %160 = getelementptr bfloat, ptr addrspace(1) %127, i64 %149, !dbg !21 + %161 = getelementptr bfloat, ptr addrspace(1) %128, i64 %149, !dbg !21 + %162 = getelementptr bfloat, ptr addrspace(1) %129, i64 %149, !dbg !21 + %163 = getelementptr bfloat, ptr addrspace(1) %130, i64 %149, !dbg !21 + %164 = getelementptr bfloat, ptr addrspace(1) %131, i64 %149, !dbg !21 + %165 = getelementptr bfloat, ptr addrspace(1) %132, i64 %149, !dbg !21 + %166 = getelementptr bfloat, ptr addrspace(1) %133, i64 %149, !dbg !21 + %167 = getelementptr bfloat, ptr addrspace(1) %134, i64 %149, !dbg !21 + %168 = getelementptr bfloat, ptr addrspace(1) %135, i64 %149, !dbg !21 + %169 = getelementptr bfloat, ptr addrspace(1) %136, i64 %149, !dbg !21 + %170 = getelementptr bfloat, ptr addrspace(1) %137, i64 %149, !dbg !21 + %171 = getelementptr bfloat, ptr addrspace(1) %138, i64 %149, !dbg !21 + %172 = getelementptr bfloat, ptr addrspace(1) %139, i64 %149, !dbg !21 + %173 = getelementptr bfloat, ptr addrspace(1) %140, i64 %149, !dbg !21 + %174 = getelementptr bfloat, ptr addrspace(1) %141, i64 %149, !dbg !21 + %175 = getelementptr bfloat, ptr addrspace(1) %142, i64 %149, !dbg !21 + %176 = getelementptr bfloat, ptr addrspace(1) %143, i64 %149, !dbg !21 + %177 = getelementptr bfloat, ptr addrspace(1) %144, i64 %149, !dbg !21 + %178 = getelementptr bfloat, ptr addrspace(1) %145, i64 %149, !dbg !21 + %179 = getelementptr bfloat, ptr addrspace(1) %146, i64 %149, !dbg !21 + %180 = getelementptr bfloat, ptr addrspace(1) %147, i64 %149, !dbg !21 + %181 = getelementptr bfloat, ptr addrspace(1) %148, i64 %149, !dbg !21 + %182 = extractelement <32 x i64> %48, i64 1, !dbg !22 + %183 = getelementptr float, ptr addrspace(1) %1, i64 %182, !dbg !22 + %184 = extractelement <32 x i64> %48, i64 0, !dbg !22 + %185 = getelementptr float, ptr addrspace(1) %1, i64 %184, !dbg !22 + %186 = getelementptr bfloat, ptr addrspace(1) %2, i64 %55, !dbg !23 + %187 = getelementptr bfloat, ptr addrspace(1) %2, i64 %56, !dbg !23 + %188 = getelementptr bfloat, ptr addrspace(1) %2, i64 %58, !dbg !23 + %189 = getelementptr bfloat, ptr addrspace(1) %2, i64 %60, !dbg !23 + %190 = getelementptr bfloat, ptr addrspace(1) %2, i64 %62, !dbg !23 + %191 = getelementptr bfloat, ptr addrspace(1) %2, i64 %64, !dbg !23 + %192 = getelementptr bfloat, ptr addrspace(1) %2, i64 %66, !dbg !23 + %193 = getelementptr bfloat, ptr addrspace(1) %2, i64 %68, !dbg !23 + %194 = getelementptr bfloat, ptr addrspace(1) %2, i64 %70, !dbg !23 + %195 = getelementptr bfloat, ptr addrspace(1) %2, i64 %72, !dbg !23 + %196 = getelementptr bfloat, ptr addrspace(1) %2, i64 %74, !dbg !23 + %197 = getelementptr bfloat, ptr addrspace(1) %2, i64 %76, !dbg !23 + %198 = getelementptr bfloat, ptr addrspace(1) %2, i64 %78, !dbg !23 + %199 = getelementptr bfloat, ptr addrspace(1) %2, i64 %80, !dbg !23 + %200 = getelementptr bfloat, ptr addrspace(1) %2, i64 %82, !dbg !23 + %201 = getelementptr bfloat, ptr addrspace(1) %2, i64 %84, !dbg !23 + %202 = getelementptr bfloat, ptr addrspace(1) %2, i64 %86, !dbg !23 + %203 = getelementptr bfloat, ptr addrspace(1) %2, i64 %88, !dbg !23 + %204 = getelementptr bfloat, ptr addrspace(1) %2, i64 %90, !dbg !23 + %205 = getelementptr bfloat, ptr addrspace(1) %2, i64 %92, !dbg !23 + %206 = getelementptr bfloat, ptr addrspace(1) %2, i64 %94, !dbg !23 + %207 = getelementptr bfloat, ptr addrspace(1) %2, i64 %96, !dbg !23 + %208 = getelementptr bfloat, ptr addrspace(1) %2, i64 %98, !dbg !23 + %209 = getelementptr bfloat, ptr addrspace(1) %2, i64 %100, !dbg !23 + %210 = getelementptr bfloat, ptr addrspace(1) %2, i64 %102, !dbg !23 + %211 = getelementptr bfloat, ptr addrspace(1) %2, i64 %104, !dbg !23 + %212 = getelementptr bfloat, ptr addrspace(1) %2, i64 %106, !dbg !23 + %213 = getelementptr bfloat, ptr addrspace(1) %2, i64 %108, !dbg !23 + %214 = getelementptr bfloat, ptr addrspace(1) %2, i64 %110, !dbg !23 + %215 = getelementptr bfloat, ptr addrspace(1) %2, i64 %112, !dbg !23 + %216 = getelementptr bfloat, ptr addrspace(1) %2, i64 %114, !dbg !23 + %217 = getelementptr bfloat, ptr addrspace(1) %2, i64 %116, !dbg !23 + %218 = getelementptr bfloat, ptr addrspace(1) %186, i64 %149, !dbg !24 + %219 = getelementptr bfloat, ptr addrspace(1) %187, i64 %149, !dbg !24 + %220 = getelementptr bfloat, ptr addrspace(1) %188, i64 %149, !dbg !24 + %221 = getelementptr bfloat, ptr addrspace(1) %189, i64 %149, !dbg !24 + %222 = getelementptr bfloat, ptr addrspace(1) %190, i64 %149, !dbg !24 + %223 = getelementptr bfloat, ptr addrspace(1) %191, i64 %149, !dbg !24 + %224 = getelementptr bfloat, ptr addrspace(1) %192, i64 %149, !dbg !24 + %225 = getelementptr bfloat, ptr addrspace(1) %193, i64 %149, !dbg !24 + %226 = getelementptr bfloat, ptr addrspace(1) %194, i64 %149, !dbg !24 + %227 = getelementptr bfloat, ptr addrspace(1) %195, i64 %149, !dbg !24 + %228 = getelementptr bfloat, ptr addrspace(1) %196, i64 %149, !dbg !24 + %229 = getelementptr bfloat, ptr addrspace(1) %197, i64 %149, !dbg !24 + %230 = getelementptr bfloat, ptr addrspace(1) %198, i64 %149, !dbg !24 + %231 = getelementptr bfloat, ptr addrspace(1) %199, i64 %149, !dbg !24 + %232 = getelementptr bfloat, ptr addrspace(1) %200, i64 %149, !dbg !24 + %233 = getelementptr bfloat, ptr addrspace(1) %201, i64 %149, !dbg !24 + %234 = getelementptr bfloat, ptr addrspace(1) %202, i64 %149, !dbg !24 + %235 = getelementptr bfloat, ptr addrspace(1) %203, i64 %149, !dbg !24 + %236 = getelementptr bfloat, ptr addrspace(1) %204, i64 %149, !dbg !24 + %237 = getelementptr bfloat, ptr addrspace(1) %205, i64 %149, !dbg !24 + %238 = getelementptr bfloat, ptr addrspace(1) %206, i64 %149, !dbg !24 + %239 = getelementptr bfloat, ptr addrspace(1) %207, i64 %149, !dbg !24 + %240 = getelementptr bfloat, ptr addrspace(1) %208, i64 %149, !dbg !24 + %241 = getelementptr bfloat, ptr addrspace(1) %209, i64 %149, !dbg !24 + %242 = getelementptr bfloat, ptr addrspace(1) %210, i64 %149, !dbg !24 + %243 = getelementptr bfloat, ptr addrspace(1) %211, i64 %149, !dbg !24 + %244 = getelementptr bfloat, ptr addrspace(1) %212, i64 %149, !dbg !24 + %245 = getelementptr bfloat, ptr addrspace(1) %213, i64 %149, !dbg !24 + %246 = getelementptr bfloat, ptr addrspace(1) %214, i64 %149, !dbg !24 + %247 = getelementptr bfloat, ptr addrspace(1) %215, i64 %149, !dbg !24 + %248 = getelementptr bfloat, ptr addrspace(1) %216, i64 %149, !dbg !24 + %249 = getelementptr bfloat, ptr addrspace(1) %217, i64 %149, !dbg !24 + %250 = getelementptr bfloat, ptr addrspace(1) %3, i64 %55, !dbg !25 + %251 = getelementptr bfloat, ptr addrspace(1) %3, i64 %56, !dbg !25 + %252 = getelementptr bfloat, ptr addrspace(1) %3, i64 %58, !dbg !25 + %253 = getelementptr bfloat, ptr addrspace(1) %3, i64 %60, !dbg !25 + %254 = getelementptr bfloat, ptr addrspace(1) %3, i64 %62, !dbg !25 + %255 = getelementptr bfloat, ptr addrspace(1) %3, i64 %64, !dbg !25 + %256 = getelementptr bfloat, ptr addrspace(1) %3, i64 %66, !dbg !25 + %257 = getelementptr bfloat, ptr addrspace(1) %3, i64 %68, !dbg !25 + %258 = getelementptr bfloat, ptr addrspace(1) %3, i64 %70, !dbg !25 + %259 = getelementptr bfloat, ptr addrspace(1) %3, i64 %72, !dbg !25 + %260 = getelementptr bfloat, ptr addrspace(1) %3, i64 %74, !dbg !25 + %261 = getelementptr bfloat, ptr addrspace(1) %3, i64 %76, !dbg !25 + %262 = getelementptr bfloat, ptr addrspace(1) %3, i64 %78, !dbg !25 + %263 = getelementptr bfloat, ptr addrspace(1) %3, i64 %80, !dbg !25 + %264 = getelementptr bfloat, ptr addrspace(1) %3, i64 %82, !dbg !25 + %265 = getelementptr bfloat, ptr addrspace(1) %3, i64 %84, !dbg !25 + %266 = getelementptr bfloat, ptr addrspace(1) %3, i64 %86, !dbg !25 + %267 = getelementptr bfloat, ptr addrspace(1) %3, i64 %88, !dbg !25 + %268 = getelementptr bfloat, ptr addrspace(1) %3, i64 %90, !dbg !25 + %269 = getelementptr bfloat, ptr addrspace(1) %3, i64 %92, !dbg !25 + %270 = getelementptr bfloat, ptr addrspace(1) %3, i64 %94, !dbg !25 + %271 = getelementptr bfloat, ptr addrspace(1) %3, i64 %96, !dbg !25 + %272 = getelementptr bfloat, ptr addrspace(1) %3, i64 %98, !dbg !25 + %273 = getelementptr bfloat, ptr addrspace(1) %3, i64 %100, !dbg !25 + %274 = getelementptr bfloat, ptr addrspace(1) %3, i64 %102, !dbg !25 + %275 = getelementptr bfloat, ptr addrspace(1) %3, i64 %104, !dbg !25 + %276 = getelementptr bfloat, ptr addrspace(1) %3, i64 %106, !dbg !25 + %277 = getelementptr bfloat, ptr addrspace(1) %3, i64 %108, !dbg !25 + %278 = getelementptr bfloat, ptr addrspace(1) %3, i64 %110, !dbg !25 + %279 = getelementptr bfloat, ptr addrspace(1) %3, i64 %112, !dbg !25 + %280 = getelementptr bfloat, ptr addrspace(1) %3, i64 %114, !dbg !25 + %281 = getelementptr bfloat, ptr addrspace(1) %3, i64 %116, !dbg !25 + %282 = getelementptr bfloat, ptr addrspace(1) %250, i64 %149, !dbg !26 + %283 = getelementptr bfloat, ptr addrspace(1) %251, i64 %149, !dbg !26 + %284 = getelementptr bfloat, ptr addrspace(1) %252, i64 %149, !dbg !26 + %285 = getelementptr bfloat, ptr addrspace(1) %253, i64 %149, !dbg !26 + %286 = getelementptr bfloat, ptr addrspace(1) %254, i64 %149, !dbg !26 + %287 = getelementptr bfloat, ptr addrspace(1) %255, i64 %149, !dbg !26 + %288 = getelementptr bfloat, ptr addrspace(1) %256, i64 %149, !dbg !26 + %289 = getelementptr bfloat, ptr addrspace(1) %257, i64 %149, !dbg !26 + %290 = getelementptr bfloat, ptr addrspace(1) %258, i64 %149, !dbg !26 + %291 = getelementptr bfloat, ptr addrspace(1) %259, i64 %149, !dbg !26 + %292 = getelementptr bfloat, ptr addrspace(1) %260, i64 %149, !dbg !26 + %293 = getelementptr bfloat, ptr addrspace(1) %261, i64 %149, !dbg !26 + %294 = getelementptr bfloat, ptr addrspace(1) %262, i64 %149, !dbg !26 + %295 = getelementptr bfloat, ptr addrspace(1) %263, i64 %149, !dbg !26 + %296 = getelementptr bfloat, ptr addrspace(1) %264, i64 %149, !dbg !26 + %297 = getelementptr bfloat, ptr addrspace(1) %265, i64 %149, !dbg !26 + %298 = getelementptr bfloat, ptr addrspace(1) %266, i64 %149, !dbg !26 + %299 = getelementptr bfloat, ptr addrspace(1) %267, i64 %149, !dbg !26 + %300 = getelementptr bfloat, ptr addrspace(1) %268, i64 %149, !dbg !26 + %301 = getelementptr bfloat, ptr addrspace(1) %269, i64 %149, !dbg !26 + %302 = getelementptr bfloat, ptr addrspace(1) %270, i64 %149, !dbg !26 + %303 = getelementptr bfloat, ptr addrspace(1) %271, i64 %149, !dbg !26 + %304 = getelementptr bfloat, ptr addrspace(1) %272, i64 %149, !dbg !26 + %305 = getelementptr bfloat, ptr addrspace(1) %273, i64 %149, !dbg !26 + %306 = getelementptr bfloat, ptr addrspace(1) %274, i64 %149, !dbg !26 + %307 = getelementptr bfloat, ptr addrspace(1) %275, i64 %149, !dbg !26 + %308 = getelementptr bfloat, ptr addrspace(1) %276, i64 %149, !dbg !26 + %309 = getelementptr bfloat, ptr addrspace(1) %277, i64 %149, !dbg !26 + %310 = getelementptr bfloat, ptr addrspace(1) %278, i64 %149, !dbg !26 + %311 = getelementptr bfloat, ptr addrspace(1) %279, i64 %149, !dbg !26 + %312 = getelementptr bfloat, ptr addrspace(1) %280, i64 %149, !dbg !26 + %313 = getelementptr bfloat, ptr addrspace(1) %281, i64 %149, !dbg !26 + %314 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %150, i1 %22) #4, !dbg !27 + %315 = extractvalue { i32, i32, i32, i32 } %314, 0, !dbg !27 + %316 = bitcast i32 %315 to <2 x bfloat>, !dbg !27 + %317 = extractvalue { i32, i32, i32, i32 } %314, 1, !dbg !27 + %318 = bitcast i32 %317 to <2 x bfloat>, !dbg !27 + %319 = extractvalue { i32, i32, i32, i32 } %314, 2, !dbg !27 + %320 = bitcast i32 %319 to <2 x bfloat>, !dbg !27 + %321 = extractvalue { i32, i32, i32, i32 } %314, 3, !dbg !27 + %322 = bitcast i32 %321 to <2 x bfloat>, !dbg !27 + %323 = extractelement <2 x bfloat> %316, i64 0, !dbg !27 + %324 = extractelement <2 x bfloat> %316, i64 1, !dbg !27 + %325 = extractelement <2 x bfloat> %318, i64 0, !dbg !27 + %326 = extractelement <2 x bfloat> %318, i64 1, !dbg !27 + %327 = extractelement <2 x bfloat> %320, i64 0, !dbg !27 + %328 = extractelement <2 x bfloat> %320, i64 1, !dbg !27 + %329 = extractelement <2 x bfloat> %322, i64 0, !dbg !27 + %330 = extractelement <2 x bfloat> %322, i64 1, !dbg !27 + %331 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %151, i1 %23) #4, !dbg !27 + %332 = extractvalue { i32, i32, i32, i32 } %331, 0, !dbg !27 + %333 = bitcast i32 %332 to <2 x bfloat>, !dbg !27 + %334 = extractvalue { i32, i32, i32, i32 } %331, 1, !dbg !27 + %335 = bitcast i32 %334 to <2 x bfloat>, !dbg !27 + %336 = extractvalue { i32, i32, i32, i32 } %331, 2, !dbg !27 + %337 = bitcast i32 %336 to <2 x bfloat>, !dbg !27 + %338 = extractvalue { i32, i32, i32, i32 } %331, 3, !dbg !27 + %339 = bitcast i32 %338 to <2 x bfloat>, !dbg !27 + %340 = extractelement <2 x bfloat> %333, i64 0, !dbg !27 + %341 = extractelement <2 x bfloat> %333, i64 1, !dbg !27 + %342 = extractelement <2 x bfloat> %335, i64 0, !dbg !27 + %343 = extractelement <2 x bfloat> %335, i64 1, !dbg !27 + %344 = extractelement <2 x bfloat> %337, i64 0, !dbg !27 + %345 = extractelement <2 x bfloat> %337, i64 1, !dbg !27 + %346 = extractelement <2 x bfloat> %339, i64 0, !dbg !27 + %347 = extractelement <2 x bfloat> %339, i64 1, !dbg !27 + %348 = extractelement <32 x i1> %52, i64 31, !dbg !27 + %349 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %152, i1 %348) #4, !dbg !27 + %350 = extractvalue { i32, i32, i32, i32 } %349, 0, !dbg !27 + %351 = bitcast i32 %350 to <2 x bfloat>, !dbg !27 + %352 = extractvalue { i32, i32, i32, i32 } %349, 1, !dbg !27 + %353 = bitcast i32 %352 to <2 x bfloat>, !dbg !27 + %354 = extractvalue { i32, i32, i32, i32 } %349, 2, !dbg !27 + %355 = bitcast i32 %354 to <2 x bfloat>, !dbg !27 + %356 = extractvalue { i32, i32, i32, i32 } %349, 3, !dbg !27 + %357 = bitcast i32 %356 to <2 x bfloat>, !dbg !27 + %358 = extractelement <2 x bfloat> %351, i64 0, !dbg !27 + %359 = extractelement <2 x bfloat> %351, i64 1, !dbg !27 + %360 = extractelement <2 x bfloat> %353, i64 0, !dbg !27 + %361 = extractelement <2 x bfloat> %353, i64 1, !dbg !27 + %362 = extractelement <2 x bfloat> %355, i64 0, !dbg !27 + %363 = extractelement <2 x bfloat> %355, i64 1, !dbg !27 + %364 = extractelement <2 x bfloat> %357, i64 0, !dbg !27 + %365 = extractelement <2 x bfloat> %357, i64 1, !dbg !27 + %366 = extractelement <32 x i1> %52, i64 30, !dbg !27 + %367 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %153, i1 %366) #4, !dbg !27 + %368 = extractvalue { i32, i32, i32, i32 } %367, 0, !dbg !27 + %369 = bitcast i32 %368 to <2 x bfloat>, !dbg !27 + %370 = extractvalue { i32, i32, i32, i32 } %367, 1, !dbg !27 + %371 = bitcast i32 %370 to <2 x bfloat>, !dbg !27 + %372 = extractvalue { i32, i32, i32, i32 } %367, 2, !dbg !27 + %373 = bitcast i32 %372 to <2 x bfloat>, !dbg !27 + %374 = extractvalue { i32, i32, i32, i32 } %367, 3, !dbg !27 + %375 = bitcast i32 %374 to <2 x bfloat>, !dbg !27 + %376 = extractelement <2 x bfloat> %369, i64 0, !dbg !27 + %377 = extractelement <2 x bfloat> %369, i64 1, !dbg !27 + %378 = extractelement <2 x bfloat> %371, i64 0, !dbg !27 + %379 = extractelement <2 x bfloat> %371, i64 1, !dbg !27 + %380 = extractelement <2 x bfloat> %373, i64 0, !dbg !27 + %381 = extractelement <2 x bfloat> %373, i64 1, !dbg !27 + %382 = extractelement <2 x bfloat> %375, i64 0, !dbg !27 + %383 = extractelement <2 x bfloat> %375, i64 1, !dbg !27 + %384 = extractelement <32 x i1> %52, i64 29, !dbg !27 + %385 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %154, i1 %384) #4, !dbg !27 + %386 = extractvalue { i32, i32, i32, i32 } %385, 0, !dbg !27 + %387 = bitcast i32 %386 to <2 x bfloat>, !dbg !27 + %388 = extractvalue { i32, i32, i32, i32 } %385, 1, !dbg !27 + %389 = bitcast i32 %388 to <2 x bfloat>, !dbg !27 + %390 = extractvalue { i32, i32, i32, i32 } %385, 2, !dbg !27 + %391 = bitcast i32 %390 to <2 x bfloat>, !dbg !27 + %392 = extractvalue { i32, i32, i32, i32 } %385, 3, !dbg !27 + %393 = bitcast i32 %392 to <2 x bfloat>, !dbg !27 + %394 = extractelement <2 x bfloat> %387, i64 0, !dbg !27 + %395 = extractelement <2 x bfloat> %387, i64 1, !dbg !27 + %396 = extractelement <2 x bfloat> %389, i64 0, !dbg !27 + %397 = extractelement <2 x bfloat> %389, i64 1, !dbg !27 + %398 = extractelement <2 x bfloat> %391, i64 0, !dbg !27 + %399 = extractelement <2 x bfloat> %391, i64 1, !dbg !27 + %400 = extractelement <2 x bfloat> %393, i64 0, !dbg !27 + %401 = extractelement <2 x bfloat> %393, i64 1, !dbg !27 + %402 = extractelement <32 x i1> %52, i64 28, !dbg !27 + %403 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %155, i1 %402) #4, !dbg !27 + %404 = extractvalue { i32, i32, i32, i32 } %403, 0, !dbg !27 + %405 = bitcast i32 %404 to <2 x bfloat>, !dbg !27 + %406 = extractvalue { i32, i32, i32, i32 } %403, 1, !dbg !27 + %407 = bitcast i32 %406 to <2 x bfloat>, !dbg !27 + %408 = extractvalue { i32, i32, i32, i32 } %403, 2, !dbg !27 + %409 = bitcast i32 %408 to <2 x bfloat>, !dbg !27 + %410 = extractvalue { i32, i32, i32, i32 } %403, 3, !dbg !27 + %411 = bitcast i32 %410 to <2 x bfloat>, !dbg !27 + %412 = extractelement <2 x bfloat> %405, i64 0, !dbg !27 + %413 = extractelement <2 x bfloat> %405, i64 1, !dbg !27 + %414 = extractelement <2 x bfloat> %407, i64 0, !dbg !27 + %415 = extractelement <2 x bfloat> %407, i64 1, !dbg !27 + %416 = extractelement <2 x bfloat> %409, i64 0, !dbg !27 + %417 = extractelement <2 x bfloat> %409, i64 1, !dbg !27 + %418 = extractelement <2 x bfloat> %411, i64 0, !dbg !27 + %419 = extractelement <2 x bfloat> %411, i64 1, !dbg !27 + %420 = extractelement <32 x i1> %52, i64 27, !dbg !27 + %421 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %156, i1 %420) #4, !dbg !27 + %422 = extractvalue { i32, i32, i32, i32 } %421, 0, !dbg !27 + %423 = bitcast i32 %422 to <2 x bfloat>, !dbg !27 + %424 = extractvalue { i32, i32, i32, i32 } %421, 1, !dbg !27 + %425 = bitcast i32 %424 to <2 x bfloat>, !dbg !27 + %426 = extractvalue { i32, i32, i32, i32 } %421, 2, !dbg !27 + %427 = bitcast i32 %426 to <2 x bfloat>, !dbg !27 + %428 = extractvalue { i32, i32, i32, i32 } %421, 3, !dbg !27 + %429 = bitcast i32 %428 to <2 x bfloat>, !dbg !27 + %430 = extractelement <2 x bfloat> %423, i64 0, !dbg !27 + %431 = extractelement <2 x bfloat> %423, i64 1, !dbg !27 + %432 = extractelement <2 x bfloat> %425, i64 0, !dbg !27 + %433 = extractelement <2 x bfloat> %425, i64 1, !dbg !27 + %434 = extractelement <2 x bfloat> %427, i64 0, !dbg !27 + %435 = extractelement <2 x bfloat> %427, i64 1, !dbg !27 + %436 = extractelement <2 x bfloat> %429, i64 0, !dbg !27 + %437 = extractelement <2 x bfloat> %429, i64 1, !dbg !27 + %438 = extractelement <32 x i1> %52, i64 26, !dbg !27 + %439 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %157, i1 %438) #4, !dbg !27 + %440 = extractvalue { i32, i32, i32, i32 } %439, 0, !dbg !27 + %441 = bitcast i32 %440 to <2 x bfloat>, !dbg !27 + %442 = extractvalue { i32, i32, i32, i32 } %439, 1, !dbg !27 + %443 = bitcast i32 %442 to <2 x bfloat>, !dbg !27 + %444 = extractvalue { i32, i32, i32, i32 } %439, 2, !dbg !27 + %445 = bitcast i32 %444 to <2 x bfloat>, !dbg !27 + %446 = extractvalue { i32, i32, i32, i32 } %439, 3, !dbg !27 + %447 = bitcast i32 %446 to <2 x bfloat>, !dbg !27 + %448 = extractelement <2 x bfloat> %441, i64 0, !dbg !27 + %449 = extractelement <2 x bfloat> %441, i64 1, !dbg !27 + %450 = extractelement <2 x bfloat> %443, i64 0, !dbg !27 + %451 = extractelement <2 x bfloat> %443, i64 1, !dbg !27 + %452 = extractelement <2 x bfloat> %445, i64 0, !dbg !27 + %453 = extractelement <2 x bfloat> %445, i64 1, !dbg !27 + %454 = extractelement <2 x bfloat> %447, i64 0, !dbg !27 + %455 = extractelement <2 x bfloat> %447, i64 1, !dbg !27 + %456 = extractelement <32 x i1> %52, i64 25, !dbg !27 + %457 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %158, i1 %456) #4, !dbg !27 + %458 = extractvalue { i32, i32, i32, i32 } %457, 0, !dbg !27 + %459 = bitcast i32 %458 to <2 x bfloat>, !dbg !27 + %460 = extractvalue { i32, i32, i32, i32 } %457, 1, !dbg !27 + %461 = bitcast i32 %460 to <2 x bfloat>, !dbg !27 + %462 = extractvalue { i32, i32, i32, i32 } %457, 2, !dbg !27 + %463 = bitcast i32 %462 to <2 x bfloat>, !dbg !27 + %464 = extractvalue { i32, i32, i32, i32 } %457, 3, !dbg !27 + %465 = bitcast i32 %464 to <2 x bfloat>, !dbg !27 + %466 = extractelement <2 x bfloat> %459, i64 0, !dbg !27 + %467 = extractelement <2 x bfloat> %459, i64 1, !dbg !27 + %468 = extractelement <2 x bfloat> %461, i64 0, !dbg !27 + %469 = extractelement <2 x bfloat> %461, i64 1, !dbg !27 + %470 = extractelement <2 x bfloat> %463, i64 0, !dbg !27 + %471 = extractelement <2 x bfloat> %463, i64 1, !dbg !27 + %472 = extractelement <2 x bfloat> %465, i64 0, !dbg !27 + %473 = extractelement <2 x bfloat> %465, i64 1, !dbg !27 + %474 = extractelement <32 x i1> %52, i64 24, !dbg !27 + %475 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %159, i1 %474) #4, !dbg !27 + %476 = extractvalue { i32, i32, i32, i32 } %475, 0, !dbg !27 + %477 = bitcast i32 %476 to <2 x bfloat>, !dbg !27 + %478 = extractvalue { i32, i32, i32, i32 } %475, 1, !dbg !27 + %479 = bitcast i32 %478 to <2 x bfloat>, !dbg !27 + %480 = extractvalue { i32, i32, i32, i32 } %475, 2, !dbg !27 + %481 = bitcast i32 %480 to <2 x bfloat>, !dbg !27 + %482 = extractvalue { i32, i32, i32, i32 } %475, 3, !dbg !27 + %483 = bitcast i32 %482 to <2 x bfloat>, !dbg !27 + %484 = extractelement <2 x bfloat> %477, i64 0, !dbg !27 + %485 = extractelement <2 x bfloat> %477, i64 1, !dbg !27 + %486 = extractelement <2 x bfloat> %479, i64 0, !dbg !27 + %487 = extractelement <2 x bfloat> %479, i64 1, !dbg !27 + %488 = extractelement <2 x bfloat> %481, i64 0, !dbg !27 + %489 = extractelement <2 x bfloat> %481, i64 1, !dbg !27 + %490 = extractelement <2 x bfloat> %483, i64 0, !dbg !27 + %491 = extractelement <2 x bfloat> %483, i64 1, !dbg !27 + %492 = extractelement <32 x i1> %52, i64 23, !dbg !27 + %493 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %160, i1 %492) #4, !dbg !27 + %494 = extractvalue { i32, i32, i32, i32 } %493, 0, !dbg !27 + %495 = bitcast i32 %494 to <2 x bfloat>, !dbg !27 + %496 = extractvalue { i32, i32, i32, i32 } %493, 1, !dbg !27 + %497 = bitcast i32 %496 to <2 x bfloat>, !dbg !27 + %498 = extractvalue { i32, i32, i32, i32 } %493, 2, !dbg !27 + %499 = bitcast i32 %498 to <2 x bfloat>, !dbg !27 + %500 = extractvalue { i32, i32, i32, i32 } %493, 3, !dbg !27 + %501 = bitcast i32 %500 to <2 x bfloat>, !dbg !27 + %502 = extractelement <2 x bfloat> %495, i64 0, !dbg !27 + %503 = extractelement <2 x bfloat> %495, i64 1, !dbg !27 + %504 = extractelement <2 x bfloat> %497, i64 0, !dbg !27 + %505 = extractelement <2 x bfloat> %497, i64 1, !dbg !27 + %506 = extractelement <2 x bfloat> %499, i64 0, !dbg !27 + %507 = extractelement <2 x bfloat> %499, i64 1, !dbg !27 + %508 = extractelement <2 x bfloat> %501, i64 0, !dbg !27 + %509 = extractelement <2 x bfloat> %501, i64 1, !dbg !27 + %510 = extractelement <32 x i1> %52, i64 22, !dbg !27 + %511 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %161, i1 %510) #4, !dbg !27 + %512 = extractvalue { i32, i32, i32, i32 } %511, 0, !dbg !27 + %513 = bitcast i32 %512 to <2 x bfloat>, !dbg !27 + %514 = extractvalue { i32, i32, i32, i32 } %511, 1, !dbg !27 + %515 = bitcast i32 %514 to <2 x bfloat>, !dbg !27 + %516 = extractvalue { i32, i32, i32, i32 } %511, 2, !dbg !27 + %517 = bitcast i32 %516 to <2 x bfloat>, !dbg !27 + %518 = extractvalue { i32, i32, i32, i32 } %511, 3, !dbg !27 + %519 = bitcast i32 %518 to <2 x bfloat>, !dbg !27 + %520 = extractelement <2 x bfloat> %513, i64 0, !dbg !27 + %521 = extractelement <2 x bfloat> %513, i64 1, !dbg !27 + %522 = extractelement <2 x bfloat> %515, i64 0, !dbg !27 + %523 = extractelement <2 x bfloat> %515, i64 1, !dbg !27 + %524 = extractelement <2 x bfloat> %517, i64 0, !dbg !27 + %525 = extractelement <2 x bfloat> %517, i64 1, !dbg !27 + %526 = extractelement <2 x bfloat> %519, i64 0, !dbg !27 + %527 = extractelement <2 x bfloat> %519, i64 1, !dbg !27 + %528 = extractelement <32 x i1> %52, i64 21, !dbg !27 + %529 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %162, i1 %528) #4, !dbg !27 + %530 = extractvalue { i32, i32, i32, i32 } %529, 0, !dbg !27 + %531 = bitcast i32 %530 to <2 x bfloat>, !dbg !27 + %532 = extractvalue { i32, i32, i32, i32 } %529, 1, !dbg !27 + %533 = bitcast i32 %532 to <2 x bfloat>, !dbg !27 + %534 = extractvalue { i32, i32, i32, i32 } %529, 2, !dbg !27 + %535 = bitcast i32 %534 to <2 x bfloat>, !dbg !27 + %536 = extractvalue { i32, i32, i32, i32 } %529, 3, !dbg !27 + %537 = bitcast i32 %536 to <2 x bfloat>, !dbg !27 + %538 = extractelement <2 x bfloat> %531, i64 0, !dbg !27 + %539 = extractelement <2 x bfloat> %531, i64 1, !dbg !27 + %540 = extractelement <2 x bfloat> %533, i64 0, !dbg !27 + %541 = extractelement <2 x bfloat> %533, i64 1, !dbg !27 + %542 = extractelement <2 x bfloat> %535, i64 0, !dbg !27 + %543 = extractelement <2 x bfloat> %535, i64 1, !dbg !27 + %544 = extractelement <2 x bfloat> %537, i64 0, !dbg !27 + %545 = extractelement <2 x bfloat> %537, i64 1, !dbg !27 + %546 = extractelement <32 x i1> %52, i64 20, !dbg !27 + %547 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %163, i1 %546) #4, !dbg !27 + %548 = extractvalue { i32, i32, i32, i32 } %547, 0, !dbg !27 + %549 = bitcast i32 %548 to <2 x bfloat>, !dbg !27 + %550 = extractvalue { i32, i32, i32, i32 } %547, 1, !dbg !27 + %551 = bitcast i32 %550 to <2 x bfloat>, !dbg !27 + %552 = extractvalue { i32, i32, i32, i32 } %547, 2, !dbg !27 + %553 = bitcast i32 %552 to <2 x bfloat>, !dbg !27 + %554 = extractvalue { i32, i32, i32, i32 } %547, 3, !dbg !27 + %555 = bitcast i32 %554 to <2 x bfloat>, !dbg !27 + %556 = extractelement <2 x bfloat> %549, i64 0, !dbg !27 + %557 = extractelement <2 x bfloat> %549, i64 1, !dbg !27 + %558 = extractelement <2 x bfloat> %551, i64 0, !dbg !27 + %559 = extractelement <2 x bfloat> %551, i64 1, !dbg !27 + %560 = extractelement <2 x bfloat> %553, i64 0, !dbg !27 + %561 = extractelement <2 x bfloat> %553, i64 1, !dbg !27 + %562 = extractelement <2 x bfloat> %555, i64 0, !dbg !27 + %563 = extractelement <2 x bfloat> %555, i64 1, !dbg !27 + %564 = extractelement <32 x i1> %52, i64 19, !dbg !27 + %565 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %164, i1 %564) #4, !dbg !27 + %566 = extractvalue { i32, i32, i32, i32 } %565, 0, !dbg !27 + %567 = bitcast i32 %566 to <2 x bfloat>, !dbg !27 + %568 = extractvalue { i32, i32, i32, i32 } %565, 1, !dbg !27 + %569 = bitcast i32 %568 to <2 x bfloat>, !dbg !27 + %570 = extractvalue { i32, i32, i32, i32 } %565, 2, !dbg !27 + %571 = bitcast i32 %570 to <2 x bfloat>, !dbg !27 + %572 = extractvalue { i32, i32, i32, i32 } %565, 3, !dbg !27 + %573 = bitcast i32 %572 to <2 x bfloat>, !dbg !27 + %574 = extractelement <2 x bfloat> %567, i64 0, !dbg !27 + %575 = extractelement <2 x bfloat> %567, i64 1, !dbg !27 + %576 = extractelement <2 x bfloat> %569, i64 0, !dbg !27 + %577 = extractelement <2 x bfloat> %569, i64 1, !dbg !27 + %578 = extractelement <2 x bfloat> %571, i64 0, !dbg !27 + %579 = extractelement <2 x bfloat> %571, i64 1, !dbg !27 + %580 = extractelement <2 x bfloat> %573, i64 0, !dbg !27 + %581 = extractelement <2 x bfloat> %573, i64 1, !dbg !27 + %582 = extractelement <32 x i1> %52, i64 18, !dbg !27 + %583 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %165, i1 %582) #4, !dbg !27 + %584 = extractvalue { i32, i32, i32, i32 } %583, 0, !dbg !27 + %585 = bitcast i32 %584 to <2 x bfloat>, !dbg !27 + %586 = extractvalue { i32, i32, i32, i32 } %583, 1, !dbg !27 + %587 = bitcast i32 %586 to <2 x bfloat>, !dbg !27 + %588 = extractvalue { i32, i32, i32, i32 } %583, 2, !dbg !27 + %589 = bitcast i32 %588 to <2 x bfloat>, !dbg !27 + %590 = extractvalue { i32, i32, i32, i32 } %583, 3, !dbg !27 + %591 = bitcast i32 %590 to <2 x bfloat>, !dbg !27 + %592 = extractelement <2 x bfloat> %585, i64 0, !dbg !27 + %593 = extractelement <2 x bfloat> %585, i64 1, !dbg !27 + %594 = extractelement <2 x bfloat> %587, i64 0, !dbg !27 + %595 = extractelement <2 x bfloat> %587, i64 1, !dbg !27 + %596 = extractelement <2 x bfloat> %589, i64 0, !dbg !27 + %597 = extractelement <2 x bfloat> %589, i64 1, !dbg !27 + %598 = extractelement <2 x bfloat> %591, i64 0, !dbg !27 + %599 = extractelement <2 x bfloat> %591, i64 1, !dbg !27 + %600 = extractelement <32 x i1> %52, i64 17, !dbg !27 + %601 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %166, i1 %600) #4, !dbg !27 + %602 = extractvalue { i32, i32, i32, i32 } %601, 0, !dbg !27 + %603 = bitcast i32 %602 to <2 x bfloat>, !dbg !27 + %604 = extractvalue { i32, i32, i32, i32 } %601, 1, !dbg !27 + %605 = bitcast i32 %604 to <2 x bfloat>, !dbg !27 + %606 = extractvalue { i32, i32, i32, i32 } %601, 2, !dbg !27 + %607 = bitcast i32 %606 to <2 x bfloat>, !dbg !27 + %608 = extractvalue { i32, i32, i32, i32 } %601, 3, !dbg !27 + %609 = bitcast i32 %608 to <2 x bfloat>, !dbg !27 + %610 = extractelement <2 x bfloat> %603, i64 0, !dbg !27 + %611 = extractelement <2 x bfloat> %603, i64 1, !dbg !27 + %612 = extractelement <2 x bfloat> %605, i64 0, !dbg !27 + %613 = extractelement <2 x bfloat> %605, i64 1, !dbg !27 + %614 = extractelement <2 x bfloat> %607, i64 0, !dbg !27 + %615 = extractelement <2 x bfloat> %607, i64 1, !dbg !27 + %616 = extractelement <2 x bfloat> %609, i64 0, !dbg !27 + %617 = extractelement <2 x bfloat> %609, i64 1, !dbg !27 + %618 = extractelement <32 x i1> %52, i64 16, !dbg !27 + %619 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %167, i1 %618) #4, !dbg !27 + %620 = extractvalue { i32, i32, i32, i32 } %619, 0, !dbg !27 + %621 = bitcast i32 %620 to <2 x bfloat>, !dbg !27 + %622 = extractvalue { i32, i32, i32, i32 } %619, 1, !dbg !27 + %623 = bitcast i32 %622 to <2 x bfloat>, !dbg !27 + %624 = extractvalue { i32, i32, i32, i32 } %619, 2, !dbg !27 + %625 = bitcast i32 %624 to <2 x bfloat>, !dbg !27 + %626 = extractvalue { i32, i32, i32, i32 } %619, 3, !dbg !27 + %627 = bitcast i32 %626 to <2 x bfloat>, !dbg !27 + %628 = extractelement <2 x bfloat> %621, i64 0, !dbg !27 + %629 = extractelement <2 x bfloat> %621, i64 1, !dbg !27 + %630 = extractelement <2 x bfloat> %623, i64 0, !dbg !27 + %631 = extractelement <2 x bfloat> %623, i64 1, !dbg !27 + %632 = extractelement <2 x bfloat> %625, i64 0, !dbg !27 + %633 = extractelement <2 x bfloat> %625, i64 1, !dbg !27 + %634 = extractelement <2 x bfloat> %627, i64 0, !dbg !27 + %635 = extractelement <2 x bfloat> %627, i64 1, !dbg !27 + %636 = extractelement <32 x i1> %52, i64 15, !dbg !27 + %637 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %168, i1 %636) #4, !dbg !27 + %638 = extractvalue { i32, i32, i32, i32 } %637, 0, !dbg !27 + %639 = bitcast i32 %638 to <2 x bfloat>, !dbg !27 + %640 = extractvalue { i32, i32, i32, i32 } %637, 1, !dbg !27 + %641 = bitcast i32 %640 to <2 x bfloat>, !dbg !27 + %642 = extractvalue { i32, i32, i32, i32 } %637, 2, !dbg !27 + %643 = bitcast i32 %642 to <2 x bfloat>, !dbg !27 + %644 = extractvalue { i32, i32, i32, i32 } %637, 3, !dbg !27 + %645 = bitcast i32 %644 to <2 x bfloat>, !dbg !27 + %646 = extractelement <2 x bfloat> %639, i64 0, !dbg !27 + %647 = extractelement <2 x bfloat> %639, i64 1, !dbg !27 + %648 = extractelement <2 x bfloat> %641, i64 0, !dbg !27 + %649 = extractelement <2 x bfloat> %641, i64 1, !dbg !27 + %650 = extractelement <2 x bfloat> %643, i64 0, !dbg !27 + %651 = extractelement <2 x bfloat> %643, i64 1, !dbg !27 + %652 = extractelement <2 x bfloat> %645, i64 0, !dbg !27 + %653 = extractelement <2 x bfloat> %645, i64 1, !dbg !27 + %654 = extractelement <32 x i1> %52, i64 14, !dbg !27 + %655 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %169, i1 %654) #4, !dbg !27 + %656 = extractvalue { i32, i32, i32, i32 } %655, 0, !dbg !27 + %657 = bitcast i32 %656 to <2 x bfloat>, !dbg !27 + %658 = extractvalue { i32, i32, i32, i32 } %655, 1, !dbg !27 + %659 = bitcast i32 %658 to <2 x bfloat>, !dbg !27 + %660 = extractvalue { i32, i32, i32, i32 } %655, 2, !dbg !27 + %661 = bitcast i32 %660 to <2 x bfloat>, !dbg !27 + %662 = extractvalue { i32, i32, i32, i32 } %655, 3, !dbg !27 + %663 = bitcast i32 %662 to <2 x bfloat>, !dbg !27 + %664 = extractelement <2 x bfloat> %657, i64 0, !dbg !27 + %665 = extractelement <2 x bfloat> %657, i64 1, !dbg !27 + %666 = extractelement <2 x bfloat> %659, i64 0, !dbg !27 + %667 = extractelement <2 x bfloat> %659, i64 1, !dbg !27 + %668 = extractelement <2 x bfloat> %661, i64 0, !dbg !27 + %669 = extractelement <2 x bfloat> %661, i64 1, !dbg !27 + %670 = extractelement <2 x bfloat> %663, i64 0, !dbg !27 + %671 = extractelement <2 x bfloat> %663, i64 1, !dbg !27 + %672 = extractelement <32 x i1> %52, i64 13, !dbg !27 + %673 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %170, i1 %672) #4, !dbg !27 + %674 = extractvalue { i32, i32, i32, i32 } %673, 0, !dbg !27 + %675 = bitcast i32 %674 to <2 x bfloat>, !dbg !27 + %676 = extractvalue { i32, i32, i32, i32 } %673, 1, !dbg !27 + %677 = bitcast i32 %676 to <2 x bfloat>, !dbg !27 + %678 = extractvalue { i32, i32, i32, i32 } %673, 2, !dbg !27 + %679 = bitcast i32 %678 to <2 x bfloat>, !dbg !27 + %680 = extractvalue { i32, i32, i32, i32 } %673, 3, !dbg !27 + %681 = bitcast i32 %680 to <2 x bfloat>, !dbg !27 + %682 = extractelement <2 x bfloat> %675, i64 0, !dbg !27 + %683 = extractelement <2 x bfloat> %675, i64 1, !dbg !27 + %684 = extractelement <2 x bfloat> %677, i64 0, !dbg !27 + %685 = extractelement <2 x bfloat> %677, i64 1, !dbg !27 + %686 = extractelement <2 x bfloat> %679, i64 0, !dbg !27 + %687 = extractelement <2 x bfloat> %679, i64 1, !dbg !27 + %688 = extractelement <2 x bfloat> %681, i64 0, !dbg !27 + %689 = extractelement <2 x bfloat> %681, i64 1, !dbg !27 + %690 = extractelement <32 x i1> %52, i64 12, !dbg !27 + %691 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %171, i1 %690) #4, !dbg !27 + %692 = extractvalue { i32, i32, i32, i32 } %691, 0, !dbg !27 + %693 = bitcast i32 %692 to <2 x bfloat>, !dbg !27 + %694 = extractvalue { i32, i32, i32, i32 } %691, 1, !dbg !27 + %695 = bitcast i32 %694 to <2 x bfloat>, !dbg !27 + %696 = extractvalue { i32, i32, i32, i32 } %691, 2, !dbg !27 + %697 = bitcast i32 %696 to <2 x bfloat>, !dbg !27 + %698 = extractvalue { i32, i32, i32, i32 } %691, 3, !dbg !27 + %699 = bitcast i32 %698 to <2 x bfloat>, !dbg !27 + %700 = extractelement <2 x bfloat> %693, i64 0, !dbg !27 + %701 = extractelement <2 x bfloat> %693, i64 1, !dbg !27 + %702 = extractelement <2 x bfloat> %695, i64 0, !dbg !27 + %703 = extractelement <2 x bfloat> %695, i64 1, !dbg !27 + %704 = extractelement <2 x bfloat> %697, i64 0, !dbg !27 + %705 = extractelement <2 x bfloat> %697, i64 1, !dbg !27 + %706 = extractelement <2 x bfloat> %699, i64 0, !dbg !27 + %707 = extractelement <2 x bfloat> %699, i64 1, !dbg !27 + %708 = extractelement <32 x i1> %52, i64 11, !dbg !27 + %709 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %172, i1 %708) #4, !dbg !27 + %710 = extractvalue { i32, i32, i32, i32 } %709, 0, !dbg !27 + %711 = bitcast i32 %710 to <2 x bfloat>, !dbg !27 + %712 = extractvalue { i32, i32, i32, i32 } %709, 1, !dbg !27 + %713 = bitcast i32 %712 to <2 x bfloat>, !dbg !27 + %714 = extractvalue { i32, i32, i32, i32 } %709, 2, !dbg !27 + %715 = bitcast i32 %714 to <2 x bfloat>, !dbg !27 + %716 = extractvalue { i32, i32, i32, i32 } %709, 3, !dbg !27 + %717 = bitcast i32 %716 to <2 x bfloat>, !dbg !27 + %718 = extractelement <2 x bfloat> %711, i64 0, !dbg !27 + %719 = extractelement <2 x bfloat> %711, i64 1, !dbg !27 + %720 = extractelement <2 x bfloat> %713, i64 0, !dbg !27 + %721 = extractelement <2 x bfloat> %713, i64 1, !dbg !27 + %722 = extractelement <2 x bfloat> %715, i64 0, !dbg !27 + %723 = extractelement <2 x bfloat> %715, i64 1, !dbg !27 + %724 = extractelement <2 x bfloat> %717, i64 0, !dbg !27 + %725 = extractelement <2 x bfloat> %717, i64 1, !dbg !27 + %726 = extractelement <32 x i1> %52, i64 10, !dbg !27 + %727 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %173, i1 %726) #4, !dbg !27 + %728 = extractvalue { i32, i32, i32, i32 } %727, 0, !dbg !27 + %729 = bitcast i32 %728 to <2 x bfloat>, !dbg !27 + %730 = extractvalue { i32, i32, i32, i32 } %727, 1, !dbg !27 + %731 = bitcast i32 %730 to <2 x bfloat>, !dbg !27 + %732 = extractvalue { i32, i32, i32, i32 } %727, 2, !dbg !27 + %733 = bitcast i32 %732 to <2 x bfloat>, !dbg !27 + %734 = extractvalue { i32, i32, i32, i32 } %727, 3, !dbg !27 + %735 = bitcast i32 %734 to <2 x bfloat>, !dbg !27 + %736 = extractelement <2 x bfloat> %729, i64 0, !dbg !27 + %737 = extractelement <2 x bfloat> %729, i64 1, !dbg !27 + %738 = extractelement <2 x bfloat> %731, i64 0, !dbg !27 + %739 = extractelement <2 x bfloat> %731, i64 1, !dbg !27 + %740 = extractelement <2 x bfloat> %733, i64 0, !dbg !27 + %741 = extractelement <2 x bfloat> %733, i64 1, !dbg !27 + %742 = extractelement <2 x bfloat> %735, i64 0, !dbg !27 + %743 = extractelement <2 x bfloat> %735, i64 1, !dbg !27 + %744 = extractelement <32 x i1> %52, i64 9, !dbg !27 + %745 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %174, i1 %744) #4, !dbg !27 + %746 = extractvalue { i32, i32, i32, i32 } %745, 0, !dbg !27 + %747 = bitcast i32 %746 to <2 x bfloat>, !dbg !27 + %748 = extractvalue { i32, i32, i32, i32 } %745, 1, !dbg !27 + %749 = bitcast i32 %748 to <2 x bfloat>, !dbg !27 + %750 = extractvalue { i32, i32, i32, i32 } %745, 2, !dbg !27 + %751 = bitcast i32 %750 to <2 x bfloat>, !dbg !27 + %752 = extractvalue { i32, i32, i32, i32 } %745, 3, !dbg !27 + %753 = bitcast i32 %752 to <2 x bfloat>, !dbg !27 + %754 = extractelement <2 x bfloat> %747, i64 0, !dbg !27 + %755 = extractelement <2 x bfloat> %747, i64 1, !dbg !27 + %756 = extractelement <2 x bfloat> %749, i64 0, !dbg !27 + %757 = extractelement <2 x bfloat> %749, i64 1, !dbg !27 + %758 = extractelement <2 x bfloat> %751, i64 0, !dbg !27 + %759 = extractelement <2 x bfloat> %751, i64 1, !dbg !27 + %760 = extractelement <2 x bfloat> %753, i64 0, !dbg !27 + %761 = extractelement <2 x bfloat> %753, i64 1, !dbg !27 + %762 = extractelement <32 x i1> %52, i64 8, !dbg !27 + %763 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %175, i1 %762) #4, !dbg !27 + %764 = extractvalue { i32, i32, i32, i32 } %763, 0, !dbg !27 + %765 = bitcast i32 %764 to <2 x bfloat>, !dbg !27 + %766 = extractvalue { i32, i32, i32, i32 } %763, 1, !dbg !27 + %767 = bitcast i32 %766 to <2 x bfloat>, !dbg !27 + %768 = extractvalue { i32, i32, i32, i32 } %763, 2, !dbg !27 + %769 = bitcast i32 %768 to <2 x bfloat>, !dbg !27 + %770 = extractvalue { i32, i32, i32, i32 } %763, 3, !dbg !27 + %771 = bitcast i32 %770 to <2 x bfloat>, !dbg !27 + %772 = extractelement <2 x bfloat> %765, i64 0, !dbg !27 + %773 = extractelement <2 x bfloat> %765, i64 1, !dbg !27 + %774 = extractelement <2 x bfloat> %767, i64 0, !dbg !27 + %775 = extractelement <2 x bfloat> %767, i64 1, !dbg !27 + %776 = extractelement <2 x bfloat> %769, i64 0, !dbg !27 + %777 = extractelement <2 x bfloat> %769, i64 1, !dbg !27 + %778 = extractelement <2 x bfloat> %771, i64 0, !dbg !27 + %779 = extractelement <2 x bfloat> %771, i64 1, !dbg !27 + %780 = extractelement <32 x i1> %52, i64 7, !dbg !27 + %781 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %176, i1 %780) #4, !dbg !27 + %782 = extractvalue { i32, i32, i32, i32 } %781, 0, !dbg !27 + %783 = bitcast i32 %782 to <2 x bfloat>, !dbg !27 + %784 = extractvalue { i32, i32, i32, i32 } %781, 1, !dbg !27 + %785 = bitcast i32 %784 to <2 x bfloat>, !dbg !27 + %786 = extractvalue { i32, i32, i32, i32 } %781, 2, !dbg !27 + %787 = bitcast i32 %786 to <2 x bfloat>, !dbg !27 + %788 = extractvalue { i32, i32, i32, i32 } %781, 3, !dbg !27 + %789 = bitcast i32 %788 to <2 x bfloat>, !dbg !27 + %790 = extractelement <2 x bfloat> %783, i64 0, !dbg !27 + %791 = extractelement <2 x bfloat> %783, i64 1, !dbg !27 + %792 = extractelement <2 x bfloat> %785, i64 0, !dbg !27 + %793 = extractelement <2 x bfloat> %785, i64 1, !dbg !27 + %794 = extractelement <2 x bfloat> %787, i64 0, !dbg !27 + %795 = extractelement <2 x bfloat> %787, i64 1, !dbg !27 + %796 = extractelement <2 x bfloat> %789, i64 0, !dbg !27 + %797 = extractelement <2 x bfloat> %789, i64 1, !dbg !27 + %798 = extractelement <32 x i1> %52, i64 6, !dbg !27 + %799 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %177, i1 %798) #4, !dbg !27 + %800 = extractvalue { i32, i32, i32, i32 } %799, 0, !dbg !27 + %801 = bitcast i32 %800 to <2 x bfloat>, !dbg !27 + %802 = extractvalue { i32, i32, i32, i32 } %799, 1, !dbg !27 + %803 = bitcast i32 %802 to <2 x bfloat>, !dbg !27 + %804 = extractvalue { i32, i32, i32, i32 } %799, 2, !dbg !27 + %805 = bitcast i32 %804 to <2 x bfloat>, !dbg !27 + %806 = extractvalue { i32, i32, i32, i32 } %799, 3, !dbg !27 + %807 = bitcast i32 %806 to <2 x bfloat>, !dbg !27 + %808 = extractelement <2 x bfloat> %801, i64 0, !dbg !27 + %809 = extractelement <2 x bfloat> %801, i64 1, !dbg !27 + %810 = extractelement <2 x bfloat> %803, i64 0, !dbg !27 + %811 = extractelement <2 x bfloat> %803, i64 1, !dbg !27 + %812 = extractelement <2 x bfloat> %805, i64 0, !dbg !27 + %813 = extractelement <2 x bfloat> %805, i64 1, !dbg !27 + %814 = extractelement <2 x bfloat> %807, i64 0, !dbg !27 + %815 = extractelement <2 x bfloat> %807, i64 1, !dbg !27 + %816 = extractelement <32 x i1> %52, i64 5, !dbg !27 + %817 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %178, i1 %816) #4, !dbg !27 + %818 = extractvalue { i32, i32, i32, i32 } %817, 0, !dbg !27 + %819 = bitcast i32 %818 to <2 x bfloat>, !dbg !27 + %820 = extractvalue { i32, i32, i32, i32 } %817, 1, !dbg !27 + %821 = bitcast i32 %820 to <2 x bfloat>, !dbg !27 + %822 = extractvalue { i32, i32, i32, i32 } %817, 2, !dbg !27 + %823 = bitcast i32 %822 to <2 x bfloat>, !dbg !27 + %824 = extractvalue { i32, i32, i32, i32 } %817, 3, !dbg !27 + %825 = bitcast i32 %824 to <2 x bfloat>, !dbg !27 + %826 = extractelement <2 x bfloat> %819, i64 0, !dbg !27 + %827 = extractelement <2 x bfloat> %819, i64 1, !dbg !27 + %828 = extractelement <2 x bfloat> %821, i64 0, !dbg !27 + %829 = extractelement <2 x bfloat> %821, i64 1, !dbg !27 + %830 = extractelement <2 x bfloat> %823, i64 0, !dbg !27 + %831 = extractelement <2 x bfloat> %823, i64 1, !dbg !27 + %832 = extractelement <2 x bfloat> %825, i64 0, !dbg !27 + %833 = extractelement <2 x bfloat> %825, i64 1, !dbg !27 + %834 = extractelement <32 x i1> %52, i64 4, !dbg !27 + %835 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %179, i1 %834) #4, !dbg !27 + %836 = extractvalue { i32, i32, i32, i32 } %835, 0, !dbg !27 + %837 = bitcast i32 %836 to <2 x bfloat>, !dbg !27 + %838 = extractvalue { i32, i32, i32, i32 } %835, 1, !dbg !27 + %839 = bitcast i32 %838 to <2 x bfloat>, !dbg !27 + %840 = extractvalue { i32, i32, i32, i32 } %835, 2, !dbg !27 + %841 = bitcast i32 %840 to <2 x bfloat>, !dbg !27 + %842 = extractvalue { i32, i32, i32, i32 } %835, 3, !dbg !27 + %843 = bitcast i32 %842 to <2 x bfloat>, !dbg !27 + %844 = extractelement <2 x bfloat> %837, i64 0, !dbg !27 + %845 = extractelement <2 x bfloat> %837, i64 1, !dbg !27 + %846 = extractelement <2 x bfloat> %839, i64 0, !dbg !27 + %847 = extractelement <2 x bfloat> %839, i64 1, !dbg !27 + %848 = extractelement <2 x bfloat> %841, i64 0, !dbg !27 + %849 = extractelement <2 x bfloat> %841, i64 1, !dbg !27 + %850 = extractelement <2 x bfloat> %843, i64 0, !dbg !27 + %851 = extractelement <2 x bfloat> %843, i64 1, !dbg !27 + %852 = extractelement <32 x i1> %52, i64 3, !dbg !27 + %853 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %180, i1 %852) #4, !dbg !27 + %854 = extractvalue { i32, i32, i32, i32 } %853, 0, !dbg !27 + %855 = bitcast i32 %854 to <2 x bfloat>, !dbg !27 + %856 = extractvalue { i32, i32, i32, i32 } %853, 1, !dbg !27 + %857 = bitcast i32 %856 to <2 x bfloat>, !dbg !27 + %858 = extractvalue { i32, i32, i32, i32 } %853, 2, !dbg !27 + %859 = bitcast i32 %858 to <2 x bfloat>, !dbg !27 + %860 = extractvalue { i32, i32, i32, i32 } %853, 3, !dbg !27 + %861 = bitcast i32 %860 to <2 x bfloat>, !dbg !27 + %862 = extractelement <2 x bfloat> %855, i64 0, !dbg !27 + %863 = extractelement <2 x bfloat> %855, i64 1, !dbg !27 + %864 = extractelement <2 x bfloat> %857, i64 0, !dbg !27 + %865 = extractelement <2 x bfloat> %857, i64 1, !dbg !27 + %866 = extractelement <2 x bfloat> %859, i64 0, !dbg !27 + %867 = extractelement <2 x bfloat> %859, i64 1, !dbg !27 + %868 = extractelement <2 x bfloat> %861, i64 0, !dbg !27 + %869 = extractelement <2 x bfloat> %861, i64 1, !dbg !27 + %870 = extractelement <32 x i1> %52, i64 2, !dbg !27 + %871 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %181, i1 %870) #4, !dbg !27 + %872 = extractvalue { i32, i32, i32, i32 } %871, 0, !dbg !27 + %873 = bitcast i32 %872 to <2 x bfloat>, !dbg !27 + %874 = extractvalue { i32, i32, i32, i32 } %871, 1, !dbg !27 + %875 = bitcast i32 %874 to <2 x bfloat>, !dbg !27 + %876 = extractvalue { i32, i32, i32, i32 } %871, 2, !dbg !27 + %877 = bitcast i32 %876 to <2 x bfloat>, !dbg !27 + %878 = extractvalue { i32, i32, i32, i32 } %871, 3, !dbg !27 + %879 = bitcast i32 %878 to <2 x bfloat>, !dbg !27 + %880 = extractelement <2 x bfloat> %873, i64 0, !dbg !27 + %881 = extractelement <2 x bfloat> %873, i64 1, !dbg !27 + %882 = extractelement <2 x bfloat> %875, i64 0, !dbg !27 + %883 = extractelement <2 x bfloat> %875, i64 1, !dbg !27 + %884 = extractelement <2 x bfloat> %877, i64 0, !dbg !27 + %885 = extractelement <2 x bfloat> %877, i64 1, !dbg !27 + %886 = extractelement <2 x bfloat> %879, i64 0, !dbg !27 + %887 = extractelement <2 x bfloat> %879, i64 1, !dbg !27 + %888 = fpext bfloat %323 to float, !dbg !28 + %889 = fpext bfloat %324 to float, !dbg !28 + %890 = fpext bfloat %325 to float, !dbg !28 + %891 = fpext bfloat %326 to float, !dbg !28 + %892 = fpext bfloat %327 to float, !dbg !28 + %893 = fpext bfloat %328 to float, !dbg !28 + %894 = fpext bfloat %329 to float, !dbg !28 + %895 = fpext bfloat %330 to float, !dbg !28 + %896 = fpext bfloat %340 to float, !dbg !28 + %897 = fpext bfloat %341 to float, !dbg !28 + %898 = fpext bfloat %342 to float, !dbg !28 + %899 = fpext bfloat %343 to float, !dbg !28 + %900 = fpext bfloat %344 to float, !dbg !28 + %901 = fpext bfloat %345 to float, !dbg !28 + %902 = fpext bfloat %346 to float, !dbg !28 + %903 = fpext bfloat %347 to float, !dbg !28 + %904 = fpext bfloat %358 to float, !dbg !28 + %905 = fpext bfloat %359 to float, !dbg !28 + %906 = fpext bfloat %360 to float, !dbg !28 + %907 = fpext bfloat %361 to float, !dbg !28 + %908 = fpext bfloat %362 to float, !dbg !28 + %909 = fpext bfloat %363 to float, !dbg !28 + %910 = fpext bfloat %364 to float, !dbg !28 + %911 = fpext bfloat %365 to float, !dbg !28 + %912 = fpext bfloat %376 to float, !dbg !28 + %913 = fpext bfloat %377 to float, !dbg !28 + %914 = fpext bfloat %378 to float, !dbg !28 + %915 = fpext bfloat %379 to float, !dbg !28 + %916 = fpext bfloat %380 to float, !dbg !28 + %917 = fpext bfloat %381 to float, !dbg !28 + %918 = fpext bfloat %382 to float, !dbg !28 + %919 = fpext bfloat %383 to float, !dbg !28 + %920 = fpext bfloat %394 to float, !dbg !28 + %921 = fpext bfloat %395 to float, !dbg !28 + %922 = fpext bfloat %396 to float, !dbg !28 + %923 = fpext bfloat %397 to float, !dbg !28 + %924 = fpext bfloat %398 to float, !dbg !28 + %925 = fpext bfloat %399 to float, !dbg !28 + %926 = fpext bfloat %400 to float, !dbg !28 + %927 = fpext bfloat %401 to float, !dbg !28 + %928 = fpext bfloat %412 to float, !dbg !28 + %929 = fpext bfloat %413 to float, !dbg !28 + %930 = fpext bfloat %414 to float, !dbg !28 + %931 = fpext bfloat %415 to float, !dbg !28 + %932 = fpext bfloat %416 to float, !dbg !28 + %933 = fpext bfloat %417 to float, !dbg !28 + %934 = fpext bfloat %418 to float, !dbg !28 + %935 = fpext bfloat %419 to float, !dbg !28 + %936 = fpext bfloat %430 to float, !dbg !28 + %937 = fpext bfloat %431 to float, !dbg !28 + %938 = fpext bfloat %432 to float, !dbg !28 + %939 = fpext bfloat %433 to float, !dbg !28 + %940 = fpext bfloat %434 to float, !dbg !28 + %941 = fpext bfloat %435 to float, !dbg !28 + %942 = fpext bfloat %436 to float, !dbg !28 + %943 = fpext bfloat %437 to float, !dbg !28 + %944 = fpext bfloat %448 to float, !dbg !28 + %945 = fpext bfloat %449 to float, !dbg !28 + %946 = fpext bfloat %450 to float, !dbg !28 + %947 = fpext bfloat %451 to float, !dbg !28 + %948 = fpext bfloat %452 to float, !dbg !28 + %949 = fpext bfloat %453 to float, !dbg !28 + %950 = fpext bfloat %454 to float, !dbg !28 + %951 = fpext bfloat %455 to float, !dbg !28 + %952 = fpext bfloat %466 to float, !dbg !28 + %953 = fpext bfloat %467 to float, !dbg !28 + %954 = fpext bfloat %468 to float, !dbg !28 + %955 = fpext bfloat %469 to float, !dbg !28 + %956 = fpext bfloat %470 to float, !dbg !28 + %957 = fpext bfloat %471 to float, !dbg !28 + %958 = fpext bfloat %472 to float, !dbg !28 + %959 = fpext bfloat %473 to float, !dbg !28 + %960 = fpext bfloat %484 to float, !dbg !28 + %961 = fpext bfloat %485 to float, !dbg !28 + %962 = fpext bfloat %486 to float, !dbg !28 + %963 = fpext bfloat %487 to float, !dbg !28 + %964 = fpext bfloat %488 to float, !dbg !28 + %965 = fpext bfloat %489 to float, !dbg !28 + %966 = fpext bfloat %490 to float, !dbg !28 + %967 = fpext bfloat %491 to float, !dbg !28 + %968 = fpext bfloat %502 to float, !dbg !28 + %969 = fpext bfloat %503 to float, !dbg !28 + %970 = fpext bfloat %504 to float, !dbg !28 + %971 = fpext bfloat %505 to float, !dbg !28 + %972 = fpext bfloat %506 to float, !dbg !28 + %973 = fpext bfloat %507 to float, !dbg !28 + %974 = fpext bfloat %508 to float, !dbg !28 + %975 = fpext bfloat %509 to float, !dbg !28 + %976 = fpext bfloat %520 to float, !dbg !28 + %977 = fpext bfloat %521 to float, !dbg !28 + %978 = fpext bfloat %522 to float, !dbg !28 + %979 = fpext bfloat %523 to float, !dbg !28 + %980 = fpext bfloat %524 to float, !dbg !28 + %981 = fpext bfloat %525 to float, !dbg !28 + %982 = fpext bfloat %526 to float, !dbg !28 + %983 = fpext bfloat %527 to float, !dbg !28 + %984 = fpext bfloat %538 to float, !dbg !28 + %985 = fpext bfloat %539 to float, !dbg !28 + %986 = fpext bfloat %540 to float, !dbg !28 + %987 = fpext bfloat %541 to float, !dbg !28 + %988 = fpext bfloat %542 to float, !dbg !28 + %989 = fpext bfloat %543 to float, !dbg !28 + %990 = fpext bfloat %544 to float, !dbg !28 + %991 = fpext bfloat %545 to float, !dbg !28 + %992 = fpext bfloat %556 to float, !dbg !28 + %993 = fpext bfloat %557 to float, !dbg !28 + %994 = fpext bfloat %558 to float, !dbg !28 + %995 = fpext bfloat %559 to float, !dbg !28 + %996 = fpext bfloat %560 to float, !dbg !28 + %997 = fpext bfloat %561 to float, !dbg !28 + %998 = fpext bfloat %562 to float, !dbg !28 + %999 = fpext bfloat %563 to float, !dbg !28 + %1000 = fpext bfloat %574 to float, !dbg !28 + %1001 = fpext bfloat %575 to float, !dbg !28 + %1002 = fpext bfloat %576 to float, !dbg !28 + %1003 = fpext bfloat %577 to float, !dbg !28 + %1004 = fpext bfloat %578 to float, !dbg !28 + %1005 = fpext bfloat %579 to float, !dbg !28 + %1006 = fpext bfloat %580 to float, !dbg !28 + %1007 = fpext bfloat %581 to float, !dbg !28 + %1008 = fpext bfloat %592 to float, !dbg !28 + %1009 = fpext bfloat %593 to float, !dbg !28 + %1010 = fpext bfloat %594 to float, !dbg !28 + %1011 = fpext bfloat %595 to float, !dbg !28 + %1012 = fpext bfloat %596 to float, !dbg !28 + %1013 = fpext bfloat %597 to float, !dbg !28 + %1014 = fpext bfloat %598 to float, !dbg !28 + %1015 = fpext bfloat %599 to float, !dbg !28 + %1016 = fpext bfloat %610 to float, !dbg !28 + %1017 = fpext bfloat %611 to float, !dbg !28 + %1018 = fpext bfloat %612 to float, !dbg !28 + %1019 = fpext bfloat %613 to float, !dbg !28 + %1020 = fpext bfloat %614 to float, !dbg !28 + %1021 = fpext bfloat %615 to float, !dbg !28 + %1022 = fpext bfloat %616 to float, !dbg !28 + %1023 = fpext bfloat %617 to float, !dbg !28 + %1024 = fpext bfloat %628 to float, !dbg !28 + %1025 = fpext bfloat %629 to float, !dbg !28 + %1026 = fpext bfloat %630 to float, !dbg !28 + %1027 = fpext bfloat %631 to float, !dbg !28 + %1028 = fpext bfloat %632 to float, !dbg !28 + %1029 = fpext bfloat %633 to float, !dbg !28 + %1030 = fpext bfloat %634 to float, !dbg !28 + %1031 = fpext bfloat %635 to float, !dbg !28 + %1032 = fpext bfloat %646 to float, !dbg !28 + %1033 = fpext bfloat %647 to float, !dbg !28 + %1034 = fpext bfloat %648 to float, !dbg !28 + %1035 = fpext bfloat %649 to float, !dbg !28 + %1036 = fpext bfloat %650 to float, !dbg !28 + %1037 = fpext bfloat %651 to float, !dbg !28 + %1038 = fpext bfloat %652 to float, !dbg !28 + %1039 = fpext bfloat %653 to float, !dbg !28 + %1040 = fpext bfloat %664 to float, !dbg !28 + %1041 = fpext bfloat %665 to float, !dbg !28 + %1042 = fpext bfloat %666 to float, !dbg !28 + %1043 = fpext bfloat %667 to float, !dbg !28 + %1044 = fpext bfloat %668 to float, !dbg !28 + %1045 = fpext bfloat %669 to float, !dbg !28 + %1046 = fpext bfloat %670 to float, !dbg !28 + %1047 = fpext bfloat %671 to float, !dbg !28 + %1048 = fpext bfloat %682 to float, !dbg !28 + %1049 = fpext bfloat %683 to float, !dbg !28 + %1050 = fpext bfloat %684 to float, !dbg !28 + %1051 = fpext bfloat %685 to float, !dbg !28 + %1052 = fpext bfloat %686 to float, !dbg !28 + %1053 = fpext bfloat %687 to float, !dbg !28 + %1054 = fpext bfloat %688 to float, !dbg !28 + %1055 = fpext bfloat %689 to float, !dbg !28 + %1056 = fpext bfloat %700 to float, !dbg !28 + %1057 = fpext bfloat %701 to float, !dbg !28 + %1058 = fpext bfloat %702 to float, !dbg !28 + %1059 = fpext bfloat %703 to float, !dbg !28 + %1060 = fpext bfloat %704 to float, !dbg !28 + %1061 = fpext bfloat %705 to float, !dbg !28 + %1062 = fpext bfloat %706 to float, !dbg !28 + %1063 = fpext bfloat %707 to float, !dbg !28 + %1064 = fpext bfloat %718 to float, !dbg !28 + %1065 = fpext bfloat %719 to float, !dbg !28 + %1066 = fpext bfloat %720 to float, !dbg !28 + %1067 = fpext bfloat %721 to float, !dbg !28 + %1068 = fpext bfloat %722 to float, !dbg !28 + %1069 = fpext bfloat %723 to float, !dbg !28 + %1070 = fpext bfloat %724 to float, !dbg !28 + %1071 = fpext bfloat %725 to float, !dbg !28 + %1072 = fpext bfloat %736 to float, !dbg !28 + %1073 = fpext bfloat %737 to float, !dbg !28 + %1074 = fpext bfloat %738 to float, !dbg !28 + %1075 = fpext bfloat %739 to float, !dbg !28 + %1076 = fpext bfloat %740 to float, !dbg !28 + %1077 = fpext bfloat %741 to float, !dbg !28 + %1078 = fpext bfloat %742 to float, !dbg !28 + %1079 = fpext bfloat %743 to float, !dbg !28 + %1080 = fpext bfloat %754 to float, !dbg !28 + %1081 = fpext bfloat %755 to float, !dbg !28 + %1082 = fpext bfloat %756 to float, !dbg !28 + %1083 = fpext bfloat %757 to float, !dbg !28 + %1084 = fpext bfloat %758 to float, !dbg !28 + %1085 = fpext bfloat %759 to float, !dbg !28 + %1086 = fpext bfloat %760 to float, !dbg !28 + %1087 = fpext bfloat %761 to float, !dbg !28 + %1088 = fpext bfloat %772 to float, !dbg !28 + %1089 = fpext bfloat %773 to float, !dbg !28 + %1090 = fpext bfloat %774 to float, !dbg !28 + %1091 = fpext bfloat %775 to float, !dbg !28 + %1092 = fpext bfloat %776 to float, !dbg !28 + %1093 = fpext bfloat %777 to float, !dbg !28 + %1094 = fpext bfloat %778 to float, !dbg !28 + %1095 = fpext bfloat %779 to float, !dbg !28 + %1096 = fpext bfloat %790 to float, !dbg !28 + %1097 = fpext bfloat %791 to float, !dbg !28 + %1098 = fpext bfloat %792 to float, !dbg !28 + %1099 = fpext bfloat %793 to float, !dbg !28 + %1100 = fpext bfloat %794 to float, !dbg !28 + %1101 = fpext bfloat %795 to float, !dbg !28 + %1102 = fpext bfloat %796 to float, !dbg !28 + %1103 = fpext bfloat %797 to float, !dbg !28 + %1104 = fpext bfloat %808 to float, !dbg !28 + %1105 = fpext bfloat %809 to float, !dbg !28 + %1106 = fpext bfloat %810 to float, !dbg !28 + %1107 = fpext bfloat %811 to float, !dbg !28 + %1108 = fpext bfloat %812 to float, !dbg !28 + %1109 = fpext bfloat %813 to float, !dbg !28 + %1110 = fpext bfloat %814 to float, !dbg !28 + %1111 = fpext bfloat %815 to float, !dbg !28 + %1112 = fpext bfloat %826 to float, !dbg !28 + %1113 = fpext bfloat %827 to float, !dbg !28 + %1114 = fpext bfloat %828 to float, !dbg !28 + %1115 = fpext bfloat %829 to float, !dbg !28 + %1116 = fpext bfloat %830 to float, !dbg !28 + %1117 = fpext bfloat %831 to float, !dbg !28 + %1118 = fpext bfloat %832 to float, !dbg !28 + %1119 = fpext bfloat %833 to float, !dbg !28 + %1120 = fpext bfloat %844 to float, !dbg !28 + %1121 = fpext bfloat %845 to float, !dbg !28 + %1122 = fpext bfloat %846 to float, !dbg !28 + %1123 = fpext bfloat %847 to float, !dbg !28 + %1124 = fpext bfloat %848 to float, !dbg !28 + %1125 = fpext bfloat %849 to float, !dbg !28 + %1126 = fpext bfloat %850 to float, !dbg !28 + %1127 = fpext bfloat %851 to float, !dbg !28 + %1128 = fpext bfloat %862 to float, !dbg !28 + %1129 = fpext bfloat %863 to float, !dbg !28 + %1130 = fpext bfloat %864 to float, !dbg !28 + %1131 = fpext bfloat %865 to float, !dbg !28 + %1132 = fpext bfloat %866 to float, !dbg !28 + %1133 = fpext bfloat %867 to float, !dbg !28 + %1134 = fpext bfloat %868 to float, !dbg !28 + %1135 = fpext bfloat %869 to float, !dbg !28 + %1136 = fpext bfloat %880 to float, !dbg !28 + %1137 = fpext bfloat %881 to float, !dbg !28 + %1138 = fpext bfloat %882 to float, !dbg !28 + %1139 = fpext bfloat %883 to float, !dbg !28 + %1140 = fpext bfloat %884 to float, !dbg !28 + %1141 = fpext bfloat %885 to float, !dbg !28 + %1142 = fpext bfloat %886 to float, !dbg !28 + %1143 = fpext bfloat %887 to float, !dbg !28 + %1144 = extractelement <32 x i1> %52, i64 1, !dbg !29 + %1145 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %183, i1 %1144) #4, !dbg !29 + %1146 = extractelement <32 x i1> %52, i64 0, !dbg !29 + %1147 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %185, i1 %1146) #4, !dbg !29 + %1148 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %218, i1 %22) #4, !dbg !30 + %1149 = extractvalue { i32, i32, i32, i32 } %1148, 0, !dbg !30 + %1150 = bitcast i32 %1149 to <2 x bfloat>, !dbg !30 + %1151 = extractvalue { i32, i32, i32, i32 } %1148, 1, !dbg !30 + %1152 = bitcast i32 %1151 to <2 x bfloat>, !dbg !30 + %1153 = extractvalue { i32, i32, i32, i32 } %1148, 2, !dbg !30 + %1154 = bitcast i32 %1153 to <2 x bfloat>, !dbg !30 + %1155 = extractvalue { i32, i32, i32, i32 } %1148, 3, !dbg !30 + %1156 = bitcast i32 %1155 to <2 x bfloat>, !dbg !30 + %1157 = extractelement <2 x bfloat> %1150, i64 0, !dbg !30 + %1158 = extractelement <2 x bfloat> %1150, i64 1, !dbg !30 + %1159 = extractelement <2 x bfloat> %1152, i64 0, !dbg !30 + %1160 = extractelement <2 x bfloat> %1152, i64 1, !dbg !30 + %1161 = extractelement <2 x bfloat> %1154, i64 0, !dbg !30 + %1162 = extractelement <2 x bfloat> %1154, i64 1, !dbg !30 + %1163 = extractelement <2 x bfloat> %1156, i64 0, !dbg !30 + %1164 = extractelement <2 x bfloat> %1156, i64 1, !dbg !30 + %1165 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %219, i1 %23) #4, !dbg !30 + %1166 = extractvalue { i32, i32, i32, i32 } %1165, 0, !dbg !30 + %1167 = bitcast i32 %1166 to <2 x bfloat>, !dbg !30 + %1168 = extractvalue { i32, i32, i32, i32 } %1165, 1, !dbg !30 + %1169 = bitcast i32 %1168 to <2 x bfloat>, !dbg !30 + %1170 = extractvalue { i32, i32, i32, i32 } %1165, 2, !dbg !30 + %1171 = bitcast i32 %1170 to <2 x bfloat>, !dbg !30 + %1172 = extractvalue { i32, i32, i32, i32 } %1165, 3, !dbg !30 + %1173 = bitcast i32 %1172 to <2 x bfloat>, !dbg !30 + %1174 = extractelement <2 x bfloat> %1167, i64 0, !dbg !30 + %1175 = extractelement <2 x bfloat> %1167, i64 1, !dbg !30 + %1176 = extractelement <2 x bfloat> %1169, i64 0, !dbg !30 + %1177 = extractelement <2 x bfloat> %1169, i64 1, !dbg !30 + %1178 = extractelement <2 x bfloat> %1171, i64 0, !dbg !30 + %1179 = extractelement <2 x bfloat> %1171, i64 1, !dbg !30 + %1180 = extractelement <2 x bfloat> %1173, i64 0, !dbg !30 + %1181 = extractelement <2 x bfloat> %1173, i64 1, !dbg !30 + %1182 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %220, i1 %348) #4, !dbg !30 + %1183 = extractvalue { i32, i32, i32, i32 } %1182, 0, !dbg !30 + %1184 = bitcast i32 %1183 to <2 x bfloat>, !dbg !30 + %1185 = extractvalue { i32, i32, i32, i32 } %1182, 1, !dbg !30 + %1186 = bitcast i32 %1185 to <2 x bfloat>, !dbg !30 + %1187 = extractvalue { i32, i32, i32, i32 } %1182, 2, !dbg !30 + %1188 = bitcast i32 %1187 to <2 x bfloat>, !dbg !30 + %1189 = extractvalue { i32, i32, i32, i32 } %1182, 3, !dbg !30 + %1190 = bitcast i32 %1189 to <2 x bfloat>, !dbg !30 + %1191 = extractelement <2 x bfloat> %1184, i64 0, !dbg !30 + %1192 = extractelement <2 x bfloat> %1184, i64 1, !dbg !30 + %1193 = extractelement <2 x bfloat> %1186, i64 0, !dbg !30 + %1194 = extractelement <2 x bfloat> %1186, i64 1, !dbg !30 + %1195 = extractelement <2 x bfloat> %1188, i64 0, !dbg !30 + %1196 = extractelement <2 x bfloat> %1188, i64 1, !dbg !30 + %1197 = extractelement <2 x bfloat> %1190, i64 0, !dbg !30 + %1198 = extractelement <2 x bfloat> %1190, i64 1, !dbg !30 + %1199 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %221, i1 %366) #4, !dbg !30 + %1200 = extractvalue { i32, i32, i32, i32 } %1199, 0, !dbg !30 + %1201 = bitcast i32 %1200 to <2 x bfloat>, !dbg !30 + %1202 = extractvalue { i32, i32, i32, i32 } %1199, 1, !dbg !30 + %1203 = bitcast i32 %1202 to <2 x bfloat>, !dbg !30 + %1204 = extractvalue { i32, i32, i32, i32 } %1199, 2, !dbg !30 + %1205 = bitcast i32 %1204 to <2 x bfloat>, !dbg !30 + %1206 = extractvalue { i32, i32, i32, i32 } %1199, 3, !dbg !30 + %1207 = bitcast i32 %1206 to <2 x bfloat>, !dbg !30 + %1208 = extractelement <2 x bfloat> %1201, i64 0, !dbg !30 + %1209 = extractelement <2 x bfloat> %1201, i64 1, !dbg !30 + %1210 = extractelement <2 x bfloat> %1203, i64 0, !dbg !30 + %1211 = extractelement <2 x bfloat> %1203, i64 1, !dbg !30 + %1212 = extractelement <2 x bfloat> %1205, i64 0, !dbg !30 + %1213 = extractelement <2 x bfloat> %1205, i64 1, !dbg !30 + %1214 = extractelement <2 x bfloat> %1207, i64 0, !dbg !30 + %1215 = extractelement <2 x bfloat> %1207, i64 1, !dbg !30 + %1216 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %222, i1 %384) #4, !dbg !30 + %1217 = extractvalue { i32, i32, i32, i32 } %1216, 0, !dbg !30 + %1218 = bitcast i32 %1217 to <2 x bfloat>, !dbg !30 + %1219 = extractvalue { i32, i32, i32, i32 } %1216, 1, !dbg !30 + %1220 = bitcast i32 %1219 to <2 x bfloat>, !dbg !30 + %1221 = extractvalue { i32, i32, i32, i32 } %1216, 2, !dbg !30 + %1222 = bitcast i32 %1221 to <2 x bfloat>, !dbg !30 + %1223 = extractvalue { i32, i32, i32, i32 } %1216, 3, !dbg !30 + %1224 = bitcast i32 %1223 to <2 x bfloat>, !dbg !30 + %1225 = extractelement <2 x bfloat> %1218, i64 0, !dbg !30 + %1226 = extractelement <2 x bfloat> %1218, i64 1, !dbg !30 + %1227 = extractelement <2 x bfloat> %1220, i64 0, !dbg !30 + %1228 = extractelement <2 x bfloat> %1220, i64 1, !dbg !30 + %1229 = extractelement <2 x bfloat> %1222, i64 0, !dbg !30 + %1230 = extractelement <2 x bfloat> %1222, i64 1, !dbg !30 + %1231 = extractelement <2 x bfloat> %1224, i64 0, !dbg !30 + %1232 = extractelement <2 x bfloat> %1224, i64 1, !dbg !30 + %1233 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %223, i1 %402) #4, !dbg !30 + %1234 = extractvalue { i32, i32, i32, i32 } %1233, 0, !dbg !30 + %1235 = bitcast i32 %1234 to <2 x bfloat>, !dbg !30 + %1236 = extractvalue { i32, i32, i32, i32 } %1233, 1, !dbg !30 + %1237 = bitcast i32 %1236 to <2 x bfloat>, !dbg !30 + %1238 = extractvalue { i32, i32, i32, i32 } %1233, 2, !dbg !30 + %1239 = bitcast i32 %1238 to <2 x bfloat>, !dbg !30 + %1240 = extractvalue { i32, i32, i32, i32 } %1233, 3, !dbg !30 + %1241 = bitcast i32 %1240 to <2 x bfloat>, !dbg !30 + %1242 = extractelement <2 x bfloat> %1235, i64 0, !dbg !30 + %1243 = extractelement <2 x bfloat> %1235, i64 1, !dbg !30 + %1244 = extractelement <2 x bfloat> %1237, i64 0, !dbg !30 + %1245 = extractelement <2 x bfloat> %1237, i64 1, !dbg !30 + %1246 = extractelement <2 x bfloat> %1239, i64 0, !dbg !30 + %1247 = extractelement <2 x bfloat> %1239, i64 1, !dbg !30 + %1248 = extractelement <2 x bfloat> %1241, i64 0, !dbg !30 + %1249 = extractelement <2 x bfloat> %1241, i64 1, !dbg !30 + %1250 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %224, i1 %420) #4, !dbg !30 + %1251 = extractvalue { i32, i32, i32, i32 } %1250, 0, !dbg !30 + %1252 = bitcast i32 %1251 to <2 x bfloat>, !dbg !30 + %1253 = extractvalue { i32, i32, i32, i32 } %1250, 1, !dbg !30 + %1254 = bitcast i32 %1253 to <2 x bfloat>, !dbg !30 + %1255 = extractvalue { i32, i32, i32, i32 } %1250, 2, !dbg !30 + %1256 = bitcast i32 %1255 to <2 x bfloat>, !dbg !30 + %1257 = extractvalue { i32, i32, i32, i32 } %1250, 3, !dbg !30 + %1258 = bitcast i32 %1257 to <2 x bfloat>, !dbg !30 + %1259 = extractelement <2 x bfloat> %1252, i64 0, !dbg !30 + %1260 = extractelement <2 x bfloat> %1252, i64 1, !dbg !30 + %1261 = extractelement <2 x bfloat> %1254, i64 0, !dbg !30 + %1262 = extractelement <2 x bfloat> %1254, i64 1, !dbg !30 + %1263 = extractelement <2 x bfloat> %1256, i64 0, !dbg !30 + %1264 = extractelement <2 x bfloat> %1256, i64 1, !dbg !30 + %1265 = extractelement <2 x bfloat> %1258, i64 0, !dbg !30 + %1266 = extractelement <2 x bfloat> %1258, i64 1, !dbg !30 + %1267 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %225, i1 %438) #4, !dbg !30 + %1268 = extractvalue { i32, i32, i32, i32 } %1267, 0, !dbg !30 + %1269 = bitcast i32 %1268 to <2 x bfloat>, !dbg !30 + %1270 = extractvalue { i32, i32, i32, i32 } %1267, 1, !dbg !30 + %1271 = bitcast i32 %1270 to <2 x bfloat>, !dbg !30 + %1272 = extractvalue { i32, i32, i32, i32 } %1267, 2, !dbg !30 + %1273 = bitcast i32 %1272 to <2 x bfloat>, !dbg !30 + %1274 = extractvalue { i32, i32, i32, i32 } %1267, 3, !dbg !30 + %1275 = bitcast i32 %1274 to <2 x bfloat>, !dbg !30 + %1276 = extractelement <2 x bfloat> %1269, i64 0, !dbg !30 + %1277 = extractelement <2 x bfloat> %1269, i64 1, !dbg !30 + %1278 = extractelement <2 x bfloat> %1271, i64 0, !dbg !30 + %1279 = extractelement <2 x bfloat> %1271, i64 1, !dbg !30 + %1280 = extractelement <2 x bfloat> %1273, i64 0, !dbg !30 + %1281 = extractelement <2 x bfloat> %1273, i64 1, !dbg !30 + %1282 = extractelement <2 x bfloat> %1275, i64 0, !dbg !30 + %1283 = extractelement <2 x bfloat> %1275, i64 1, !dbg !30 + %1284 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %226, i1 %456) #4, !dbg !30 + %1285 = extractvalue { i32, i32, i32, i32 } %1284, 0, !dbg !30 + %1286 = bitcast i32 %1285 to <2 x bfloat>, !dbg !30 + %1287 = extractvalue { i32, i32, i32, i32 } %1284, 1, !dbg !30 + %1288 = bitcast i32 %1287 to <2 x bfloat>, !dbg !30 + %1289 = extractvalue { i32, i32, i32, i32 } %1284, 2, !dbg !30 + %1290 = bitcast i32 %1289 to <2 x bfloat>, !dbg !30 + %1291 = extractvalue { i32, i32, i32, i32 } %1284, 3, !dbg !30 + %1292 = bitcast i32 %1291 to <2 x bfloat>, !dbg !30 + %1293 = extractelement <2 x bfloat> %1286, i64 0, !dbg !30 + %1294 = extractelement <2 x bfloat> %1286, i64 1, !dbg !30 + %1295 = extractelement <2 x bfloat> %1288, i64 0, !dbg !30 + %1296 = extractelement <2 x bfloat> %1288, i64 1, !dbg !30 + %1297 = extractelement <2 x bfloat> %1290, i64 0, !dbg !30 + %1298 = extractelement <2 x bfloat> %1290, i64 1, !dbg !30 + %1299 = extractelement <2 x bfloat> %1292, i64 0, !dbg !30 + %1300 = extractelement <2 x bfloat> %1292, i64 1, !dbg !30 + %1301 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %227, i1 %474) #4, !dbg !30 + %1302 = extractvalue { i32, i32, i32, i32 } %1301, 0, !dbg !30 + %1303 = bitcast i32 %1302 to <2 x bfloat>, !dbg !30 + %1304 = extractvalue { i32, i32, i32, i32 } %1301, 1, !dbg !30 + %1305 = bitcast i32 %1304 to <2 x bfloat>, !dbg !30 + %1306 = extractvalue { i32, i32, i32, i32 } %1301, 2, !dbg !30 + %1307 = bitcast i32 %1306 to <2 x bfloat>, !dbg !30 + %1308 = extractvalue { i32, i32, i32, i32 } %1301, 3, !dbg !30 + %1309 = bitcast i32 %1308 to <2 x bfloat>, !dbg !30 + %1310 = extractelement <2 x bfloat> %1303, i64 0, !dbg !30 + %1311 = extractelement <2 x bfloat> %1303, i64 1, !dbg !30 + %1312 = extractelement <2 x bfloat> %1305, i64 0, !dbg !30 + %1313 = extractelement <2 x bfloat> %1305, i64 1, !dbg !30 + %1314 = extractelement <2 x bfloat> %1307, i64 0, !dbg !30 + %1315 = extractelement <2 x bfloat> %1307, i64 1, !dbg !30 + %1316 = extractelement <2 x bfloat> %1309, i64 0, !dbg !30 + %1317 = extractelement <2 x bfloat> %1309, i64 1, !dbg !30 + %1318 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %228, i1 %492) #4, !dbg !30 + %1319 = extractvalue { i32, i32, i32, i32 } %1318, 0, !dbg !30 + %1320 = bitcast i32 %1319 to <2 x bfloat>, !dbg !30 + %1321 = extractvalue { i32, i32, i32, i32 } %1318, 1, !dbg !30 + %1322 = bitcast i32 %1321 to <2 x bfloat>, !dbg !30 + %1323 = extractvalue { i32, i32, i32, i32 } %1318, 2, !dbg !30 + %1324 = bitcast i32 %1323 to <2 x bfloat>, !dbg !30 + %1325 = extractvalue { i32, i32, i32, i32 } %1318, 3, !dbg !30 + %1326 = bitcast i32 %1325 to <2 x bfloat>, !dbg !30 + %1327 = extractelement <2 x bfloat> %1320, i64 0, !dbg !30 + %1328 = extractelement <2 x bfloat> %1320, i64 1, !dbg !30 + %1329 = extractelement <2 x bfloat> %1322, i64 0, !dbg !30 + %1330 = extractelement <2 x bfloat> %1322, i64 1, !dbg !30 + %1331 = extractelement <2 x bfloat> %1324, i64 0, !dbg !30 + %1332 = extractelement <2 x bfloat> %1324, i64 1, !dbg !30 + %1333 = extractelement <2 x bfloat> %1326, i64 0, !dbg !30 + %1334 = extractelement <2 x bfloat> %1326, i64 1, !dbg !30 + %1335 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %229, i1 %510) #4, !dbg !30 + %1336 = extractvalue { i32, i32, i32, i32 } %1335, 0, !dbg !30 + %1337 = bitcast i32 %1336 to <2 x bfloat>, !dbg !30 + %1338 = extractvalue { i32, i32, i32, i32 } %1335, 1, !dbg !30 + %1339 = bitcast i32 %1338 to <2 x bfloat>, !dbg !30 + %1340 = extractvalue { i32, i32, i32, i32 } %1335, 2, !dbg !30 + %1341 = bitcast i32 %1340 to <2 x bfloat>, !dbg !30 + %1342 = extractvalue { i32, i32, i32, i32 } %1335, 3, !dbg !30 + %1343 = bitcast i32 %1342 to <2 x bfloat>, !dbg !30 + %1344 = extractelement <2 x bfloat> %1337, i64 0, !dbg !30 + %1345 = extractelement <2 x bfloat> %1337, i64 1, !dbg !30 + %1346 = extractelement <2 x bfloat> %1339, i64 0, !dbg !30 + %1347 = extractelement <2 x bfloat> %1339, i64 1, !dbg !30 + %1348 = extractelement <2 x bfloat> %1341, i64 0, !dbg !30 + %1349 = extractelement <2 x bfloat> %1341, i64 1, !dbg !30 + %1350 = extractelement <2 x bfloat> %1343, i64 0, !dbg !30 + %1351 = extractelement <2 x bfloat> %1343, i64 1, !dbg !30 + %1352 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %230, i1 %528) #4, !dbg !30 + %1353 = extractvalue { i32, i32, i32, i32 } %1352, 0, !dbg !30 + %1354 = bitcast i32 %1353 to <2 x bfloat>, !dbg !30 + %1355 = extractvalue { i32, i32, i32, i32 } %1352, 1, !dbg !30 + %1356 = bitcast i32 %1355 to <2 x bfloat>, !dbg !30 + %1357 = extractvalue { i32, i32, i32, i32 } %1352, 2, !dbg !30 + %1358 = bitcast i32 %1357 to <2 x bfloat>, !dbg !30 + %1359 = extractvalue { i32, i32, i32, i32 } %1352, 3, !dbg !30 + %1360 = bitcast i32 %1359 to <2 x bfloat>, !dbg !30 + %1361 = extractelement <2 x bfloat> %1354, i64 0, !dbg !30 + %1362 = extractelement <2 x bfloat> %1354, i64 1, !dbg !30 + %1363 = extractelement <2 x bfloat> %1356, i64 0, !dbg !30 + %1364 = extractelement <2 x bfloat> %1356, i64 1, !dbg !30 + %1365 = extractelement <2 x bfloat> %1358, i64 0, !dbg !30 + %1366 = extractelement <2 x bfloat> %1358, i64 1, !dbg !30 + %1367 = extractelement <2 x bfloat> %1360, i64 0, !dbg !30 + %1368 = extractelement <2 x bfloat> %1360, i64 1, !dbg !30 + %1369 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %231, i1 %546) #4, !dbg !30 + %1370 = extractvalue { i32, i32, i32, i32 } %1369, 0, !dbg !30 + %1371 = bitcast i32 %1370 to <2 x bfloat>, !dbg !30 + %1372 = extractvalue { i32, i32, i32, i32 } %1369, 1, !dbg !30 + %1373 = bitcast i32 %1372 to <2 x bfloat>, !dbg !30 + %1374 = extractvalue { i32, i32, i32, i32 } %1369, 2, !dbg !30 + %1375 = bitcast i32 %1374 to <2 x bfloat>, !dbg !30 + %1376 = extractvalue { i32, i32, i32, i32 } %1369, 3, !dbg !30 + %1377 = bitcast i32 %1376 to <2 x bfloat>, !dbg !30 + %1378 = extractelement <2 x bfloat> %1371, i64 0, !dbg !30 + %1379 = extractelement <2 x bfloat> %1371, i64 1, !dbg !30 + %1380 = extractelement <2 x bfloat> %1373, i64 0, !dbg !30 + %1381 = extractelement <2 x bfloat> %1373, i64 1, !dbg !30 + %1382 = extractelement <2 x bfloat> %1375, i64 0, !dbg !30 + %1383 = extractelement <2 x bfloat> %1375, i64 1, !dbg !30 + %1384 = extractelement <2 x bfloat> %1377, i64 0, !dbg !30 + %1385 = extractelement <2 x bfloat> %1377, i64 1, !dbg !30 + %1386 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %232, i1 %564) #4, !dbg !30 + %1387 = extractvalue { i32, i32, i32, i32 } %1386, 0, !dbg !30 + %1388 = bitcast i32 %1387 to <2 x bfloat>, !dbg !30 + %1389 = extractvalue { i32, i32, i32, i32 } %1386, 1, !dbg !30 + %1390 = bitcast i32 %1389 to <2 x bfloat>, !dbg !30 + %1391 = extractvalue { i32, i32, i32, i32 } %1386, 2, !dbg !30 + %1392 = bitcast i32 %1391 to <2 x bfloat>, !dbg !30 + %1393 = extractvalue { i32, i32, i32, i32 } %1386, 3, !dbg !30 + %1394 = bitcast i32 %1393 to <2 x bfloat>, !dbg !30 + %1395 = extractelement <2 x bfloat> %1388, i64 0, !dbg !30 + %1396 = extractelement <2 x bfloat> %1388, i64 1, !dbg !30 + %1397 = extractelement <2 x bfloat> %1390, i64 0, !dbg !30 + %1398 = extractelement <2 x bfloat> %1390, i64 1, !dbg !30 + %1399 = extractelement <2 x bfloat> %1392, i64 0, !dbg !30 + %1400 = extractelement <2 x bfloat> %1392, i64 1, !dbg !30 + %1401 = extractelement <2 x bfloat> %1394, i64 0, !dbg !30 + %1402 = extractelement <2 x bfloat> %1394, i64 1, !dbg !30 + %1403 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %233, i1 %582) #4, !dbg !30 + %1404 = extractvalue { i32, i32, i32, i32 } %1403, 0, !dbg !30 + %1405 = bitcast i32 %1404 to <2 x bfloat>, !dbg !30 + %1406 = extractvalue { i32, i32, i32, i32 } %1403, 1, !dbg !30 + %1407 = bitcast i32 %1406 to <2 x bfloat>, !dbg !30 + %1408 = extractvalue { i32, i32, i32, i32 } %1403, 2, !dbg !30 + %1409 = bitcast i32 %1408 to <2 x bfloat>, !dbg !30 + %1410 = extractvalue { i32, i32, i32, i32 } %1403, 3, !dbg !30 + %1411 = bitcast i32 %1410 to <2 x bfloat>, !dbg !30 + %1412 = extractelement <2 x bfloat> %1405, i64 0, !dbg !30 + %1413 = extractelement <2 x bfloat> %1405, i64 1, !dbg !30 + %1414 = extractelement <2 x bfloat> %1407, i64 0, !dbg !30 + %1415 = extractelement <2 x bfloat> %1407, i64 1, !dbg !30 + %1416 = extractelement <2 x bfloat> %1409, i64 0, !dbg !30 + %1417 = extractelement <2 x bfloat> %1409, i64 1, !dbg !30 + %1418 = extractelement <2 x bfloat> %1411, i64 0, !dbg !30 + %1419 = extractelement <2 x bfloat> %1411, i64 1, !dbg !30 + %1420 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %234, i1 %600) #4, !dbg !30 + %1421 = extractvalue { i32, i32, i32, i32 } %1420, 0, !dbg !30 + %1422 = bitcast i32 %1421 to <2 x bfloat>, !dbg !30 + %1423 = extractvalue { i32, i32, i32, i32 } %1420, 1, !dbg !30 + %1424 = bitcast i32 %1423 to <2 x bfloat>, !dbg !30 + %1425 = extractvalue { i32, i32, i32, i32 } %1420, 2, !dbg !30 + %1426 = bitcast i32 %1425 to <2 x bfloat>, !dbg !30 + %1427 = extractvalue { i32, i32, i32, i32 } %1420, 3, !dbg !30 + %1428 = bitcast i32 %1427 to <2 x bfloat>, !dbg !30 + %1429 = extractelement <2 x bfloat> %1422, i64 0, !dbg !30 + %1430 = extractelement <2 x bfloat> %1422, i64 1, !dbg !30 + %1431 = extractelement <2 x bfloat> %1424, i64 0, !dbg !30 + %1432 = extractelement <2 x bfloat> %1424, i64 1, !dbg !30 + %1433 = extractelement <2 x bfloat> %1426, i64 0, !dbg !30 + %1434 = extractelement <2 x bfloat> %1426, i64 1, !dbg !30 + %1435 = extractelement <2 x bfloat> %1428, i64 0, !dbg !30 + %1436 = extractelement <2 x bfloat> %1428, i64 1, !dbg !30 + %1437 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %235, i1 %618) #4, !dbg !30 + %1438 = extractvalue { i32, i32, i32, i32 } %1437, 0, !dbg !30 + %1439 = bitcast i32 %1438 to <2 x bfloat>, !dbg !30 + %1440 = extractvalue { i32, i32, i32, i32 } %1437, 1, !dbg !30 + %1441 = bitcast i32 %1440 to <2 x bfloat>, !dbg !30 + %1442 = extractvalue { i32, i32, i32, i32 } %1437, 2, !dbg !30 + %1443 = bitcast i32 %1442 to <2 x bfloat>, !dbg !30 + %1444 = extractvalue { i32, i32, i32, i32 } %1437, 3, !dbg !30 + %1445 = bitcast i32 %1444 to <2 x bfloat>, !dbg !30 + %1446 = extractelement <2 x bfloat> %1439, i64 0, !dbg !30 + %1447 = extractelement <2 x bfloat> %1439, i64 1, !dbg !30 + %1448 = extractelement <2 x bfloat> %1441, i64 0, !dbg !30 + %1449 = extractelement <2 x bfloat> %1441, i64 1, !dbg !30 + %1450 = extractelement <2 x bfloat> %1443, i64 0, !dbg !30 + %1451 = extractelement <2 x bfloat> %1443, i64 1, !dbg !30 + %1452 = extractelement <2 x bfloat> %1445, i64 0, !dbg !30 + %1453 = extractelement <2 x bfloat> %1445, i64 1, !dbg !30 + %1454 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %236, i1 %636) #4, !dbg !30 + %1455 = extractvalue { i32, i32, i32, i32 } %1454, 0, !dbg !30 + %1456 = bitcast i32 %1455 to <2 x bfloat>, !dbg !30 + %1457 = extractvalue { i32, i32, i32, i32 } %1454, 1, !dbg !30 + %1458 = bitcast i32 %1457 to <2 x bfloat>, !dbg !30 + %1459 = extractvalue { i32, i32, i32, i32 } %1454, 2, !dbg !30 + %1460 = bitcast i32 %1459 to <2 x bfloat>, !dbg !30 + %1461 = extractvalue { i32, i32, i32, i32 } %1454, 3, !dbg !30 + %1462 = bitcast i32 %1461 to <2 x bfloat>, !dbg !30 + %1463 = extractelement <2 x bfloat> %1456, i64 0, !dbg !30 + %1464 = extractelement <2 x bfloat> %1456, i64 1, !dbg !30 + %1465 = extractelement <2 x bfloat> %1458, i64 0, !dbg !30 + %1466 = extractelement <2 x bfloat> %1458, i64 1, !dbg !30 + %1467 = extractelement <2 x bfloat> %1460, i64 0, !dbg !30 + %1468 = extractelement <2 x bfloat> %1460, i64 1, !dbg !30 + %1469 = extractelement <2 x bfloat> %1462, i64 0, !dbg !30 + %1470 = extractelement <2 x bfloat> %1462, i64 1, !dbg !30 + %1471 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %237, i1 %654) #4, !dbg !30 + %1472 = extractvalue { i32, i32, i32, i32 } %1471, 0, !dbg !30 + %1473 = bitcast i32 %1472 to <2 x bfloat>, !dbg !30 + %1474 = extractvalue { i32, i32, i32, i32 } %1471, 1, !dbg !30 + %1475 = bitcast i32 %1474 to <2 x bfloat>, !dbg !30 + %1476 = extractvalue { i32, i32, i32, i32 } %1471, 2, !dbg !30 + %1477 = bitcast i32 %1476 to <2 x bfloat>, !dbg !30 + %1478 = extractvalue { i32, i32, i32, i32 } %1471, 3, !dbg !30 + %1479 = bitcast i32 %1478 to <2 x bfloat>, !dbg !30 + %1480 = extractelement <2 x bfloat> %1473, i64 0, !dbg !30 + %1481 = extractelement <2 x bfloat> %1473, i64 1, !dbg !30 + %1482 = extractelement <2 x bfloat> %1475, i64 0, !dbg !30 + %1483 = extractelement <2 x bfloat> %1475, i64 1, !dbg !30 + %1484 = extractelement <2 x bfloat> %1477, i64 0, !dbg !30 + %1485 = extractelement <2 x bfloat> %1477, i64 1, !dbg !30 + %1486 = extractelement <2 x bfloat> %1479, i64 0, !dbg !30 + %1487 = extractelement <2 x bfloat> %1479, i64 1, !dbg !30 + %1488 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %238, i1 %672) #4, !dbg !30 + %1489 = extractvalue { i32, i32, i32, i32 } %1488, 0, !dbg !30 + %1490 = bitcast i32 %1489 to <2 x bfloat>, !dbg !30 + %1491 = extractvalue { i32, i32, i32, i32 } %1488, 1, !dbg !30 + %1492 = bitcast i32 %1491 to <2 x bfloat>, !dbg !30 + %1493 = extractvalue { i32, i32, i32, i32 } %1488, 2, !dbg !30 + %1494 = bitcast i32 %1493 to <2 x bfloat>, !dbg !30 + %1495 = extractvalue { i32, i32, i32, i32 } %1488, 3, !dbg !30 + %1496 = bitcast i32 %1495 to <2 x bfloat>, !dbg !30 + %1497 = extractelement <2 x bfloat> %1490, i64 0, !dbg !30 + %1498 = extractelement <2 x bfloat> %1490, i64 1, !dbg !30 + %1499 = extractelement <2 x bfloat> %1492, i64 0, !dbg !30 + %1500 = extractelement <2 x bfloat> %1492, i64 1, !dbg !30 + %1501 = extractelement <2 x bfloat> %1494, i64 0, !dbg !30 + %1502 = extractelement <2 x bfloat> %1494, i64 1, !dbg !30 + %1503 = extractelement <2 x bfloat> %1496, i64 0, !dbg !30 + %1504 = extractelement <2 x bfloat> %1496, i64 1, !dbg !30 + %1505 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %239, i1 %690) #4, !dbg !30 + %1506 = extractvalue { i32, i32, i32, i32 } %1505, 0, !dbg !30 + %1507 = bitcast i32 %1506 to <2 x bfloat>, !dbg !30 + %1508 = extractvalue { i32, i32, i32, i32 } %1505, 1, !dbg !30 + %1509 = bitcast i32 %1508 to <2 x bfloat>, !dbg !30 + %1510 = extractvalue { i32, i32, i32, i32 } %1505, 2, !dbg !30 + %1511 = bitcast i32 %1510 to <2 x bfloat>, !dbg !30 + %1512 = extractvalue { i32, i32, i32, i32 } %1505, 3, !dbg !30 + %1513 = bitcast i32 %1512 to <2 x bfloat>, !dbg !30 + %1514 = extractelement <2 x bfloat> %1507, i64 0, !dbg !30 + %1515 = extractelement <2 x bfloat> %1507, i64 1, !dbg !30 + %1516 = extractelement <2 x bfloat> %1509, i64 0, !dbg !30 + %1517 = extractelement <2 x bfloat> %1509, i64 1, !dbg !30 + %1518 = extractelement <2 x bfloat> %1511, i64 0, !dbg !30 + %1519 = extractelement <2 x bfloat> %1511, i64 1, !dbg !30 + %1520 = extractelement <2 x bfloat> %1513, i64 0, !dbg !30 + %1521 = extractelement <2 x bfloat> %1513, i64 1, !dbg !30 + %1522 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %240, i1 %708) #4, !dbg !30 + %1523 = extractvalue { i32, i32, i32, i32 } %1522, 0, !dbg !30 + %1524 = bitcast i32 %1523 to <2 x bfloat>, !dbg !30 + %1525 = extractvalue { i32, i32, i32, i32 } %1522, 1, !dbg !30 + %1526 = bitcast i32 %1525 to <2 x bfloat>, !dbg !30 + %1527 = extractvalue { i32, i32, i32, i32 } %1522, 2, !dbg !30 + %1528 = bitcast i32 %1527 to <2 x bfloat>, !dbg !30 + %1529 = extractvalue { i32, i32, i32, i32 } %1522, 3, !dbg !30 + %1530 = bitcast i32 %1529 to <2 x bfloat>, !dbg !30 + %1531 = extractelement <2 x bfloat> %1524, i64 0, !dbg !30 + %1532 = extractelement <2 x bfloat> %1524, i64 1, !dbg !30 + %1533 = extractelement <2 x bfloat> %1526, i64 0, !dbg !30 + %1534 = extractelement <2 x bfloat> %1526, i64 1, !dbg !30 + %1535 = extractelement <2 x bfloat> %1528, i64 0, !dbg !30 + %1536 = extractelement <2 x bfloat> %1528, i64 1, !dbg !30 + %1537 = extractelement <2 x bfloat> %1530, i64 0, !dbg !30 + %1538 = extractelement <2 x bfloat> %1530, i64 1, !dbg !30 + %1539 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %241, i1 %726) #4, !dbg !30 + %1540 = extractvalue { i32, i32, i32, i32 } %1539, 0, !dbg !30 + %1541 = bitcast i32 %1540 to <2 x bfloat>, !dbg !30 + %1542 = extractvalue { i32, i32, i32, i32 } %1539, 1, !dbg !30 + %1543 = bitcast i32 %1542 to <2 x bfloat>, !dbg !30 + %1544 = extractvalue { i32, i32, i32, i32 } %1539, 2, !dbg !30 + %1545 = bitcast i32 %1544 to <2 x bfloat>, !dbg !30 + %1546 = extractvalue { i32, i32, i32, i32 } %1539, 3, !dbg !30 + %1547 = bitcast i32 %1546 to <2 x bfloat>, !dbg !30 + %1548 = extractelement <2 x bfloat> %1541, i64 0, !dbg !30 + %1549 = extractelement <2 x bfloat> %1541, i64 1, !dbg !30 + %1550 = extractelement <2 x bfloat> %1543, i64 0, !dbg !30 + %1551 = extractelement <2 x bfloat> %1543, i64 1, !dbg !30 + %1552 = extractelement <2 x bfloat> %1545, i64 0, !dbg !30 + %1553 = extractelement <2 x bfloat> %1545, i64 1, !dbg !30 + %1554 = extractelement <2 x bfloat> %1547, i64 0, !dbg !30 + %1555 = extractelement <2 x bfloat> %1547, i64 1, !dbg !30 + %1556 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %242, i1 %744) #4, !dbg !30 + %1557 = extractvalue { i32, i32, i32, i32 } %1556, 0, !dbg !30 + %1558 = bitcast i32 %1557 to <2 x bfloat>, !dbg !30 + %1559 = extractvalue { i32, i32, i32, i32 } %1556, 1, !dbg !30 + %1560 = bitcast i32 %1559 to <2 x bfloat>, !dbg !30 + %1561 = extractvalue { i32, i32, i32, i32 } %1556, 2, !dbg !30 + %1562 = bitcast i32 %1561 to <2 x bfloat>, !dbg !30 + %1563 = extractvalue { i32, i32, i32, i32 } %1556, 3, !dbg !30 + %1564 = bitcast i32 %1563 to <2 x bfloat>, !dbg !30 + %1565 = extractelement <2 x bfloat> %1558, i64 0, !dbg !30 + %1566 = extractelement <2 x bfloat> %1558, i64 1, !dbg !30 + %1567 = extractelement <2 x bfloat> %1560, i64 0, !dbg !30 + %1568 = extractelement <2 x bfloat> %1560, i64 1, !dbg !30 + %1569 = extractelement <2 x bfloat> %1562, i64 0, !dbg !30 + %1570 = extractelement <2 x bfloat> %1562, i64 1, !dbg !30 + %1571 = extractelement <2 x bfloat> %1564, i64 0, !dbg !30 + %1572 = extractelement <2 x bfloat> %1564, i64 1, !dbg !30 + %1573 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %243, i1 %762) #4, !dbg !30 + %1574 = extractvalue { i32, i32, i32, i32 } %1573, 0, !dbg !30 + %1575 = bitcast i32 %1574 to <2 x bfloat>, !dbg !30 + %1576 = extractvalue { i32, i32, i32, i32 } %1573, 1, !dbg !30 + %1577 = bitcast i32 %1576 to <2 x bfloat>, !dbg !30 + %1578 = extractvalue { i32, i32, i32, i32 } %1573, 2, !dbg !30 + %1579 = bitcast i32 %1578 to <2 x bfloat>, !dbg !30 + %1580 = extractvalue { i32, i32, i32, i32 } %1573, 3, !dbg !30 + %1581 = bitcast i32 %1580 to <2 x bfloat>, !dbg !30 + %1582 = extractelement <2 x bfloat> %1575, i64 0, !dbg !30 + %1583 = extractelement <2 x bfloat> %1575, i64 1, !dbg !30 + %1584 = extractelement <2 x bfloat> %1577, i64 0, !dbg !30 + %1585 = extractelement <2 x bfloat> %1577, i64 1, !dbg !30 + %1586 = extractelement <2 x bfloat> %1579, i64 0, !dbg !30 + %1587 = extractelement <2 x bfloat> %1579, i64 1, !dbg !30 + %1588 = extractelement <2 x bfloat> %1581, i64 0, !dbg !30 + %1589 = extractelement <2 x bfloat> %1581, i64 1, !dbg !30 + %1590 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %244, i1 %780) #4, !dbg !30 + %1591 = extractvalue { i32, i32, i32, i32 } %1590, 0, !dbg !30 + %1592 = bitcast i32 %1591 to <2 x bfloat>, !dbg !30 + %1593 = extractvalue { i32, i32, i32, i32 } %1590, 1, !dbg !30 + %1594 = bitcast i32 %1593 to <2 x bfloat>, !dbg !30 + %1595 = extractvalue { i32, i32, i32, i32 } %1590, 2, !dbg !30 + %1596 = bitcast i32 %1595 to <2 x bfloat>, !dbg !30 + %1597 = extractvalue { i32, i32, i32, i32 } %1590, 3, !dbg !30 + %1598 = bitcast i32 %1597 to <2 x bfloat>, !dbg !30 + %1599 = extractelement <2 x bfloat> %1592, i64 0, !dbg !30 + %1600 = extractelement <2 x bfloat> %1592, i64 1, !dbg !30 + %1601 = extractelement <2 x bfloat> %1594, i64 0, !dbg !30 + %1602 = extractelement <2 x bfloat> %1594, i64 1, !dbg !30 + %1603 = extractelement <2 x bfloat> %1596, i64 0, !dbg !30 + %1604 = extractelement <2 x bfloat> %1596, i64 1, !dbg !30 + %1605 = extractelement <2 x bfloat> %1598, i64 0, !dbg !30 + %1606 = extractelement <2 x bfloat> %1598, i64 1, !dbg !30 + %1607 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %245, i1 %798) #4, !dbg !30 + %1608 = extractvalue { i32, i32, i32, i32 } %1607, 0, !dbg !30 + %1609 = bitcast i32 %1608 to <2 x bfloat>, !dbg !30 + %1610 = extractvalue { i32, i32, i32, i32 } %1607, 1, !dbg !30 + %1611 = bitcast i32 %1610 to <2 x bfloat>, !dbg !30 + %1612 = extractvalue { i32, i32, i32, i32 } %1607, 2, !dbg !30 + %1613 = bitcast i32 %1612 to <2 x bfloat>, !dbg !30 + %1614 = extractvalue { i32, i32, i32, i32 } %1607, 3, !dbg !30 + %1615 = bitcast i32 %1614 to <2 x bfloat>, !dbg !30 + %1616 = extractelement <2 x bfloat> %1609, i64 0, !dbg !30 + %1617 = extractelement <2 x bfloat> %1609, i64 1, !dbg !30 + %1618 = extractelement <2 x bfloat> %1611, i64 0, !dbg !30 + %1619 = extractelement <2 x bfloat> %1611, i64 1, !dbg !30 + %1620 = extractelement <2 x bfloat> %1613, i64 0, !dbg !30 + %1621 = extractelement <2 x bfloat> %1613, i64 1, !dbg !30 + %1622 = extractelement <2 x bfloat> %1615, i64 0, !dbg !30 + %1623 = extractelement <2 x bfloat> %1615, i64 1, !dbg !30 + %1624 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %246, i1 %816) #4, !dbg !30 + %1625 = extractvalue { i32, i32, i32, i32 } %1624, 0, !dbg !30 + %1626 = bitcast i32 %1625 to <2 x bfloat>, !dbg !30 + %1627 = extractvalue { i32, i32, i32, i32 } %1624, 1, !dbg !30 + %1628 = bitcast i32 %1627 to <2 x bfloat>, !dbg !30 + %1629 = extractvalue { i32, i32, i32, i32 } %1624, 2, !dbg !30 + %1630 = bitcast i32 %1629 to <2 x bfloat>, !dbg !30 + %1631 = extractvalue { i32, i32, i32, i32 } %1624, 3, !dbg !30 + %1632 = bitcast i32 %1631 to <2 x bfloat>, !dbg !30 + %1633 = extractelement <2 x bfloat> %1626, i64 0, !dbg !30 + %1634 = extractelement <2 x bfloat> %1626, i64 1, !dbg !30 + %1635 = extractelement <2 x bfloat> %1628, i64 0, !dbg !30 + %1636 = extractelement <2 x bfloat> %1628, i64 1, !dbg !30 + %1637 = extractelement <2 x bfloat> %1630, i64 0, !dbg !30 + %1638 = extractelement <2 x bfloat> %1630, i64 1, !dbg !30 + %1639 = extractelement <2 x bfloat> %1632, i64 0, !dbg !30 + %1640 = extractelement <2 x bfloat> %1632, i64 1, !dbg !30 + %1641 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %247, i1 %834) #4, !dbg !30 + %1642 = extractvalue { i32, i32, i32, i32 } %1641, 0, !dbg !30 + %1643 = bitcast i32 %1642 to <2 x bfloat>, !dbg !30 + %1644 = extractvalue { i32, i32, i32, i32 } %1641, 1, !dbg !30 + %1645 = bitcast i32 %1644 to <2 x bfloat>, !dbg !30 + %1646 = extractvalue { i32, i32, i32, i32 } %1641, 2, !dbg !30 + %1647 = bitcast i32 %1646 to <2 x bfloat>, !dbg !30 + %1648 = extractvalue { i32, i32, i32, i32 } %1641, 3, !dbg !30 + %1649 = bitcast i32 %1648 to <2 x bfloat>, !dbg !30 + %1650 = extractelement <2 x bfloat> %1643, i64 0, !dbg !30 + %1651 = extractelement <2 x bfloat> %1643, i64 1, !dbg !30 + %1652 = extractelement <2 x bfloat> %1645, i64 0, !dbg !30 + %1653 = extractelement <2 x bfloat> %1645, i64 1, !dbg !30 + %1654 = extractelement <2 x bfloat> %1647, i64 0, !dbg !30 + %1655 = extractelement <2 x bfloat> %1647, i64 1, !dbg !30 + %1656 = extractelement <2 x bfloat> %1649, i64 0, !dbg !30 + %1657 = extractelement <2 x bfloat> %1649, i64 1, !dbg !30 + %1658 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %248, i1 %852) #4, !dbg !30 + %1659 = extractvalue { i32, i32, i32, i32 } %1658, 0, !dbg !30 + %1660 = bitcast i32 %1659 to <2 x bfloat>, !dbg !30 + %1661 = extractvalue { i32, i32, i32, i32 } %1658, 1, !dbg !30 + %1662 = bitcast i32 %1661 to <2 x bfloat>, !dbg !30 + %1663 = extractvalue { i32, i32, i32, i32 } %1658, 2, !dbg !30 + %1664 = bitcast i32 %1663 to <2 x bfloat>, !dbg !30 + %1665 = extractvalue { i32, i32, i32, i32 } %1658, 3, !dbg !30 + %1666 = bitcast i32 %1665 to <2 x bfloat>, !dbg !30 + %1667 = extractelement <2 x bfloat> %1660, i64 0, !dbg !30 + %1668 = extractelement <2 x bfloat> %1660, i64 1, !dbg !30 + %1669 = extractelement <2 x bfloat> %1662, i64 0, !dbg !30 + %1670 = extractelement <2 x bfloat> %1662, i64 1, !dbg !30 + %1671 = extractelement <2 x bfloat> %1664, i64 0, !dbg !30 + %1672 = extractelement <2 x bfloat> %1664, i64 1, !dbg !30 + %1673 = extractelement <2 x bfloat> %1666, i64 0, !dbg !30 + %1674 = extractelement <2 x bfloat> %1666, i64 1, !dbg !30 + %1675 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %249, i1 %870) #4, !dbg !30 + %1676 = extractvalue { i32, i32, i32, i32 } %1675, 0, !dbg !30 + %1677 = bitcast i32 %1676 to <2 x bfloat>, !dbg !30 + %1678 = extractvalue { i32, i32, i32, i32 } %1675, 1, !dbg !30 + %1679 = bitcast i32 %1678 to <2 x bfloat>, !dbg !30 + %1680 = extractvalue { i32, i32, i32, i32 } %1675, 2, !dbg !30 + %1681 = bitcast i32 %1680 to <2 x bfloat>, !dbg !30 + %1682 = extractvalue { i32, i32, i32, i32 } %1675, 3, !dbg !30 + %1683 = bitcast i32 %1682 to <2 x bfloat>, !dbg !30 + %1684 = extractelement <2 x bfloat> %1677, i64 0, !dbg !30 + %1685 = extractelement <2 x bfloat> %1677, i64 1, !dbg !30 + %1686 = extractelement <2 x bfloat> %1679, i64 0, !dbg !30 + %1687 = extractelement <2 x bfloat> %1679, i64 1, !dbg !30 + %1688 = extractelement <2 x bfloat> %1681, i64 0, !dbg !30 + %1689 = extractelement <2 x bfloat> %1681, i64 1, !dbg !30 + %1690 = extractelement <2 x bfloat> %1683, i64 0, !dbg !30 + %1691 = extractelement <2 x bfloat> %1683, i64 1, !dbg !30 + %1692 = fpext bfloat %1157 to float, !dbg !31 + %1693 = fpext bfloat %1158 to float, !dbg !31 + %1694 = fpext bfloat %1159 to float, !dbg !31 + %1695 = fpext bfloat %1160 to float, !dbg !31 + %1696 = fpext bfloat %1161 to float, !dbg !31 + %1697 = fpext bfloat %1162 to float, !dbg !31 + %1698 = fpext bfloat %1163 to float, !dbg !31 + %1699 = fpext bfloat %1164 to float, !dbg !31 + %1700 = fpext bfloat %1174 to float, !dbg !31 + %1701 = fpext bfloat %1175 to float, !dbg !31 + %1702 = fpext bfloat %1176 to float, !dbg !31 + %1703 = fpext bfloat %1177 to float, !dbg !31 + %1704 = fpext bfloat %1178 to float, !dbg !31 + %1705 = fpext bfloat %1179 to float, !dbg !31 + %1706 = fpext bfloat %1180 to float, !dbg !31 + %1707 = fpext bfloat %1181 to float, !dbg !31 + %1708 = fpext bfloat %1191 to float, !dbg !31 + %1709 = fpext bfloat %1192 to float, !dbg !31 + %1710 = fpext bfloat %1193 to float, !dbg !31 + %1711 = fpext bfloat %1194 to float, !dbg !31 + %1712 = fpext bfloat %1195 to float, !dbg !31 + %1713 = fpext bfloat %1196 to float, !dbg !31 + %1714 = fpext bfloat %1197 to float, !dbg !31 + %1715 = fpext bfloat %1198 to float, !dbg !31 + %1716 = fpext bfloat %1208 to float, !dbg !31 + %1717 = fpext bfloat %1209 to float, !dbg !31 + %1718 = fpext bfloat %1210 to float, !dbg !31 + %1719 = fpext bfloat %1211 to float, !dbg !31 + %1720 = fpext bfloat %1212 to float, !dbg !31 + %1721 = fpext bfloat %1213 to float, !dbg !31 + %1722 = fpext bfloat %1214 to float, !dbg !31 + %1723 = fpext bfloat %1215 to float, !dbg !31 + %1724 = fpext bfloat %1225 to float, !dbg !31 + %1725 = fpext bfloat %1226 to float, !dbg !31 + %1726 = fpext bfloat %1227 to float, !dbg !31 + %1727 = fpext bfloat %1228 to float, !dbg !31 + %1728 = fpext bfloat %1229 to float, !dbg !31 + %1729 = fpext bfloat %1230 to float, !dbg !31 + %1730 = fpext bfloat %1231 to float, !dbg !31 + %1731 = fpext bfloat %1232 to float, !dbg !31 + %1732 = fpext bfloat %1242 to float, !dbg !31 + %1733 = fpext bfloat %1243 to float, !dbg !31 + %1734 = fpext bfloat %1244 to float, !dbg !31 + %1735 = fpext bfloat %1245 to float, !dbg !31 + %1736 = fpext bfloat %1246 to float, !dbg !31 + %1737 = fpext bfloat %1247 to float, !dbg !31 + %1738 = fpext bfloat %1248 to float, !dbg !31 + %1739 = fpext bfloat %1249 to float, !dbg !31 + %1740 = fpext bfloat %1259 to float, !dbg !31 + %1741 = fpext bfloat %1260 to float, !dbg !31 + %1742 = fpext bfloat %1261 to float, !dbg !31 + %1743 = fpext bfloat %1262 to float, !dbg !31 + %1744 = fpext bfloat %1263 to float, !dbg !31 + %1745 = fpext bfloat %1264 to float, !dbg !31 + %1746 = fpext bfloat %1265 to float, !dbg !31 + %1747 = fpext bfloat %1266 to float, !dbg !31 + %1748 = fpext bfloat %1276 to float, !dbg !31 + %1749 = fpext bfloat %1277 to float, !dbg !31 + %1750 = fpext bfloat %1278 to float, !dbg !31 + %1751 = fpext bfloat %1279 to float, !dbg !31 + %1752 = fpext bfloat %1280 to float, !dbg !31 + %1753 = fpext bfloat %1281 to float, !dbg !31 + %1754 = fpext bfloat %1282 to float, !dbg !31 + %1755 = fpext bfloat %1283 to float, !dbg !31 + %1756 = fpext bfloat %1293 to float, !dbg !31 + %1757 = fpext bfloat %1294 to float, !dbg !31 + %1758 = fpext bfloat %1295 to float, !dbg !31 + %1759 = fpext bfloat %1296 to float, !dbg !31 + %1760 = fpext bfloat %1297 to float, !dbg !31 + %1761 = fpext bfloat %1298 to float, !dbg !31 + %1762 = fpext bfloat %1299 to float, !dbg !31 + %1763 = fpext bfloat %1300 to float, !dbg !31 + %1764 = fpext bfloat %1310 to float, !dbg !31 + %1765 = fpext bfloat %1311 to float, !dbg !31 + %1766 = fpext bfloat %1312 to float, !dbg !31 + %1767 = fpext bfloat %1313 to float, !dbg !31 + %1768 = fpext bfloat %1314 to float, !dbg !31 + %1769 = fpext bfloat %1315 to float, !dbg !31 + %1770 = fpext bfloat %1316 to float, !dbg !31 + %1771 = fpext bfloat %1317 to float, !dbg !31 + %1772 = fpext bfloat %1327 to float, !dbg !31 + %1773 = fpext bfloat %1328 to float, !dbg !31 + %1774 = fpext bfloat %1329 to float, !dbg !31 + %1775 = fpext bfloat %1330 to float, !dbg !31 + %1776 = fpext bfloat %1331 to float, !dbg !31 + %1777 = fpext bfloat %1332 to float, !dbg !31 + %1778 = fpext bfloat %1333 to float, !dbg !31 + %1779 = fpext bfloat %1334 to float, !dbg !31 + %1780 = fpext bfloat %1344 to float, !dbg !31 + %1781 = fpext bfloat %1345 to float, !dbg !31 + %1782 = fpext bfloat %1346 to float, !dbg !31 + %1783 = fpext bfloat %1347 to float, !dbg !31 + %1784 = fpext bfloat %1348 to float, !dbg !31 + %1785 = fpext bfloat %1349 to float, !dbg !31 + %1786 = fpext bfloat %1350 to float, !dbg !31 + %1787 = fpext bfloat %1351 to float, !dbg !31 + %1788 = fpext bfloat %1361 to float, !dbg !31 + %1789 = fpext bfloat %1362 to float, !dbg !31 + %1790 = fpext bfloat %1363 to float, !dbg !31 + %1791 = fpext bfloat %1364 to float, !dbg !31 + %1792 = fpext bfloat %1365 to float, !dbg !31 + %1793 = fpext bfloat %1366 to float, !dbg !31 + %1794 = fpext bfloat %1367 to float, !dbg !31 + %1795 = fpext bfloat %1368 to float, !dbg !31 + %1796 = fpext bfloat %1378 to float, !dbg !31 + %1797 = fpext bfloat %1379 to float, !dbg !31 + %1798 = fpext bfloat %1380 to float, !dbg !31 + %1799 = fpext bfloat %1381 to float, !dbg !31 + %1800 = fpext bfloat %1382 to float, !dbg !31 + %1801 = fpext bfloat %1383 to float, !dbg !31 + %1802 = fpext bfloat %1384 to float, !dbg !31 + %1803 = fpext bfloat %1385 to float, !dbg !31 + %1804 = fpext bfloat %1395 to float, !dbg !31 + %1805 = fpext bfloat %1396 to float, !dbg !31 + %1806 = fpext bfloat %1397 to float, !dbg !31 + %1807 = fpext bfloat %1398 to float, !dbg !31 + %1808 = fpext bfloat %1399 to float, !dbg !31 + %1809 = fpext bfloat %1400 to float, !dbg !31 + %1810 = fpext bfloat %1401 to float, !dbg !31 + %1811 = fpext bfloat %1402 to float, !dbg !31 + %1812 = fpext bfloat %1412 to float, !dbg !31 + %1813 = fpext bfloat %1413 to float, !dbg !31 + %1814 = fpext bfloat %1414 to float, !dbg !31 + %1815 = fpext bfloat %1415 to float, !dbg !31 + %1816 = fpext bfloat %1416 to float, !dbg !31 + %1817 = fpext bfloat %1417 to float, !dbg !31 + %1818 = fpext bfloat %1418 to float, !dbg !31 + %1819 = fpext bfloat %1419 to float, !dbg !31 + %1820 = fpext bfloat %1429 to float, !dbg !31 + %1821 = fpext bfloat %1430 to float, !dbg !31 + %1822 = fpext bfloat %1431 to float, !dbg !31 + %1823 = fpext bfloat %1432 to float, !dbg !31 + %1824 = fpext bfloat %1433 to float, !dbg !31 + %1825 = fpext bfloat %1434 to float, !dbg !31 + %1826 = fpext bfloat %1435 to float, !dbg !31 + %1827 = fpext bfloat %1436 to float, !dbg !31 + %1828 = fpext bfloat %1446 to float, !dbg !31 + %1829 = fpext bfloat %1447 to float, !dbg !31 + %1830 = fpext bfloat %1448 to float, !dbg !31 + %1831 = fpext bfloat %1449 to float, !dbg !31 + %1832 = fpext bfloat %1450 to float, !dbg !31 + %1833 = fpext bfloat %1451 to float, !dbg !31 + %1834 = fpext bfloat %1452 to float, !dbg !31 + %1835 = fpext bfloat %1453 to float, !dbg !31 + %1836 = fpext bfloat %1463 to float, !dbg !31 + %1837 = fpext bfloat %1464 to float, !dbg !31 + %1838 = fpext bfloat %1465 to float, !dbg !31 + %1839 = fpext bfloat %1466 to float, !dbg !31 + %1840 = fpext bfloat %1467 to float, !dbg !31 + %1841 = fpext bfloat %1468 to float, !dbg !31 + %1842 = fpext bfloat %1469 to float, !dbg !31 + %1843 = fpext bfloat %1470 to float, !dbg !31 + %1844 = fpext bfloat %1480 to float, !dbg !31 + %1845 = fpext bfloat %1481 to float, !dbg !31 + %1846 = fpext bfloat %1482 to float, !dbg !31 + %1847 = fpext bfloat %1483 to float, !dbg !31 + %1848 = fpext bfloat %1484 to float, !dbg !31 + %1849 = fpext bfloat %1485 to float, !dbg !31 + %1850 = fpext bfloat %1486 to float, !dbg !31 + %1851 = fpext bfloat %1487 to float, !dbg !31 + %1852 = fpext bfloat %1497 to float, !dbg !31 + %1853 = fpext bfloat %1498 to float, !dbg !31 + %1854 = fpext bfloat %1499 to float, !dbg !31 + %1855 = fpext bfloat %1500 to float, !dbg !31 + %1856 = fpext bfloat %1501 to float, !dbg !31 + %1857 = fpext bfloat %1502 to float, !dbg !31 + %1858 = fpext bfloat %1503 to float, !dbg !31 + %1859 = fpext bfloat %1504 to float, !dbg !31 + %1860 = fpext bfloat %1514 to float, !dbg !31 + %1861 = fpext bfloat %1515 to float, !dbg !31 + %1862 = fpext bfloat %1516 to float, !dbg !31 + %1863 = fpext bfloat %1517 to float, !dbg !31 + %1864 = fpext bfloat %1518 to float, !dbg !31 + %1865 = fpext bfloat %1519 to float, !dbg !31 + %1866 = fpext bfloat %1520 to float, !dbg !31 + %1867 = fpext bfloat %1521 to float, !dbg !31 + %1868 = fpext bfloat %1531 to float, !dbg !31 + %1869 = fpext bfloat %1532 to float, !dbg !31 + %1870 = fpext bfloat %1533 to float, !dbg !31 + %1871 = fpext bfloat %1534 to float, !dbg !31 + %1872 = fpext bfloat %1535 to float, !dbg !31 + %1873 = fpext bfloat %1536 to float, !dbg !31 + %1874 = fpext bfloat %1537 to float, !dbg !31 + %1875 = fpext bfloat %1538 to float, !dbg !31 + %1876 = fpext bfloat %1548 to float, !dbg !31 + %1877 = fpext bfloat %1549 to float, !dbg !31 + %1878 = fpext bfloat %1550 to float, !dbg !31 + %1879 = fpext bfloat %1551 to float, !dbg !31 + %1880 = fpext bfloat %1552 to float, !dbg !31 + %1881 = fpext bfloat %1553 to float, !dbg !31 + %1882 = fpext bfloat %1554 to float, !dbg !31 + %1883 = fpext bfloat %1555 to float, !dbg !31 + %1884 = fpext bfloat %1565 to float, !dbg !31 + %1885 = fpext bfloat %1566 to float, !dbg !31 + %1886 = fpext bfloat %1567 to float, !dbg !31 + %1887 = fpext bfloat %1568 to float, !dbg !31 + %1888 = fpext bfloat %1569 to float, !dbg !31 + %1889 = fpext bfloat %1570 to float, !dbg !31 + %1890 = fpext bfloat %1571 to float, !dbg !31 + %1891 = fpext bfloat %1572 to float, !dbg !31 + %1892 = fpext bfloat %1582 to float, !dbg !31 + %1893 = fpext bfloat %1583 to float, !dbg !31 + %1894 = fpext bfloat %1584 to float, !dbg !31 + %1895 = fpext bfloat %1585 to float, !dbg !31 + %1896 = fpext bfloat %1586 to float, !dbg !31 + %1897 = fpext bfloat %1587 to float, !dbg !31 + %1898 = fpext bfloat %1588 to float, !dbg !31 + %1899 = fpext bfloat %1589 to float, !dbg !31 + %1900 = fpext bfloat %1599 to float, !dbg !31 + %1901 = fpext bfloat %1600 to float, !dbg !31 + %1902 = fpext bfloat %1601 to float, !dbg !31 + %1903 = fpext bfloat %1602 to float, !dbg !31 + %1904 = fpext bfloat %1603 to float, !dbg !31 + %1905 = fpext bfloat %1604 to float, !dbg !31 + %1906 = fpext bfloat %1605 to float, !dbg !31 + %1907 = fpext bfloat %1606 to float, !dbg !31 + %1908 = fpext bfloat %1616 to float, !dbg !31 + %1909 = fpext bfloat %1617 to float, !dbg !31 + %1910 = fpext bfloat %1618 to float, !dbg !31 + %1911 = fpext bfloat %1619 to float, !dbg !31 + %1912 = fpext bfloat %1620 to float, !dbg !31 + %1913 = fpext bfloat %1621 to float, !dbg !31 + %1914 = fpext bfloat %1622 to float, !dbg !31 + %1915 = fpext bfloat %1623 to float, !dbg !31 + %1916 = fpext bfloat %1633 to float, !dbg !31 + %1917 = fpext bfloat %1634 to float, !dbg !31 + %1918 = fpext bfloat %1635 to float, !dbg !31 + %1919 = fpext bfloat %1636 to float, !dbg !31 + %1920 = fpext bfloat %1637 to float, !dbg !31 + %1921 = fpext bfloat %1638 to float, !dbg !31 + %1922 = fpext bfloat %1639 to float, !dbg !31 + %1923 = fpext bfloat %1640 to float, !dbg !31 + %1924 = fpext bfloat %1650 to float, !dbg !31 + %1925 = fpext bfloat %1651 to float, !dbg !31 + %1926 = fpext bfloat %1652 to float, !dbg !31 + %1927 = fpext bfloat %1653 to float, !dbg !31 + %1928 = fpext bfloat %1654 to float, !dbg !31 + %1929 = fpext bfloat %1655 to float, !dbg !31 + %1930 = fpext bfloat %1656 to float, !dbg !31 + %1931 = fpext bfloat %1657 to float, !dbg !31 + %1932 = fpext bfloat %1667 to float, !dbg !31 + %1933 = fpext bfloat %1668 to float, !dbg !31 + %1934 = fpext bfloat %1669 to float, !dbg !31 + %1935 = fpext bfloat %1670 to float, !dbg !31 + %1936 = fpext bfloat %1671 to float, !dbg !31 + %1937 = fpext bfloat %1672 to float, !dbg !31 + %1938 = fpext bfloat %1673 to float, !dbg !31 + %1939 = fpext bfloat %1674 to float, !dbg !31 + %1940 = fpext bfloat %1684 to float, !dbg !31 + %1941 = fpext bfloat %1685 to float, !dbg !31 + %1942 = fpext bfloat %1686 to float, !dbg !31 + %1943 = fpext bfloat %1687 to float, !dbg !31 + %1944 = fpext bfloat %1688 to float, !dbg !31 + %1945 = fpext bfloat %1689 to float, !dbg !31 + %1946 = fpext bfloat %1690 to float, !dbg !31 + %1947 = fpext bfloat %1691 to float, !dbg !31 + %1948 = and i32 %16, 60, !dbg !32 + %1949 = shl nuw nsw i32 %12, 6, !dbg !32 + %1950 = and i32 %1949, 64, !dbg !32 + %1951 = and i32 %12, 32, !dbg !32 + %1952 = shl nuw nsw i32 %1951, 2, !dbg !32 + %1953 = or disjoint i32 %1950, %1952, !dbg !32 + %1954 = or disjoint i32 %1953, %1948, !dbg !32 + %1955 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %1954, !dbg !32 + %1956 = insertelement <1 x i32> poison, i32 %1145, i64 0, !dbg !32 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) %1955, <1 x i32> %1956) #4, !dbg !32 + %1957 = xor i32 %1954, 320, !dbg !32 + %1958 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %1957, !dbg !32 + %1959 = insertelement <1 x i32> poison, i32 %1147, i64 0, !dbg !32 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) %1958, <1 x i32> %1959) #4, !dbg !32 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !32 + %1960 = and i32 %12, 16, !dbg !32 + %1961 = icmp eq i32 %1960, 0, !dbg !32 + %1962 = select i1 %1961, i32 0, i32 320, !dbg !32 + %1963 = shl nuw nsw i32 %1951, 1, !dbg !32 + %1964 = xor i32 %1962, %1963, !dbg !32 + %1965 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %1964, !dbg !32 + %1966 = load float, ptr addrspace(3) %1965, align 16, !dbg !32 + %1967 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 4, !dbg !32 + %1968 = load float, ptr addrspace(3) %1967, align 4, !dbg !32 + %1969 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 8, !dbg !32 + %1970 = load float, ptr addrspace(3) %1969, align 8, !dbg !32 + %1971 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 12, !dbg !32 + %1972 = load float, ptr addrspace(3) %1971, align 4, !dbg !32 + %1973 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 16, !dbg !32 + %1974 = load float, ptr addrspace(3) %1973, align 16, !dbg !32 + %1975 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 20, !dbg !32 + %1976 = load float, ptr addrspace(3) %1975, align 4, !dbg !32 + %1977 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 24, !dbg !32 + %1978 = load float, ptr addrspace(3) %1977, align 8, !dbg !32 + %1979 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 28, !dbg !32 + %1980 = load float, ptr addrspace(3) %1979, align 4, !dbg !32 + %1981 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 32, !dbg !32 + %1982 = load float, ptr addrspace(3) %1981, align 16, !dbg !32 + %1983 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 36, !dbg !32 + %1984 = load float, ptr addrspace(3) %1983, align 4, !dbg !32 + %1985 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 40, !dbg !32 + %1986 = load float, ptr addrspace(3) %1985, align 8, !dbg !32 + %1987 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 44, !dbg !32 + %1988 = load float, ptr addrspace(3) %1987, align 4, !dbg !32 + %1989 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 48, !dbg !32 + %1990 = load float, ptr addrspace(3) %1989, align 16, !dbg !32 + %1991 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 52, !dbg !32 + %1992 = load float, ptr addrspace(3) %1991, align 4, !dbg !32 + %1993 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 56, !dbg !32 + %1994 = load float, ptr addrspace(3) %1993, align 8, !dbg !32 + %1995 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 60, !dbg !32 + %1996 = load float, ptr addrspace(3) %1995, align 4, !dbg !32 + %1997 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 128, !dbg !32 + %1998 = load float, ptr addrspace(3) %1997, align 16, !dbg !32 + %1999 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 132, !dbg !32 + %2000 = load float, ptr addrspace(3) %1999, align 4, !dbg !32 + %2001 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 136, !dbg !32 + %2002 = load float, ptr addrspace(3) %2001, align 8, !dbg !32 + %2003 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 140, !dbg !32 + %2004 = load float, ptr addrspace(3) %2003, align 4, !dbg !32 + %2005 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 144, !dbg !32 + %2006 = load float, ptr addrspace(3) %2005, align 16, !dbg !32 + %2007 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 148, !dbg !32 + %2008 = load float, ptr addrspace(3) %2007, align 4, !dbg !32 + %2009 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 152, !dbg !32 + %2010 = load float, ptr addrspace(3) %2009, align 8, !dbg !32 + %2011 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 156, !dbg !32 + %2012 = load float, ptr addrspace(3) %2011, align 4, !dbg !32 + %2013 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 160, !dbg !32 + %2014 = load float, ptr addrspace(3) %2013, align 16, !dbg !32 + %2015 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 164, !dbg !32 + %2016 = load float, ptr addrspace(3) %2015, align 4, !dbg !32 + %2017 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 168, !dbg !32 + %2018 = load float, ptr addrspace(3) %2017, align 8, !dbg !32 + %2019 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 172, !dbg !32 + %2020 = load float, ptr addrspace(3) %2019, align 4, !dbg !32 + %2021 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 176, !dbg !32 + %2022 = load float, ptr addrspace(3) %2021, align 16, !dbg !32 + %2023 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 180, !dbg !32 + %2024 = load float, ptr addrspace(3) %2023, align 4, !dbg !32 + %2025 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 184, !dbg !32 + %2026 = load float, ptr addrspace(3) %2025, align 8, !dbg !32 + %2027 = getelementptr inbounds nuw i8, ptr addrspace(3) %1965, i32 188, !dbg !32 + %2028 = load float, ptr addrspace(3) %2027, align 4, !dbg !32 + %2029 = fmul float %1966, %1692, !dbg !33 + %2030 = fmul float %1966, %1693, !dbg !33 + %2031 = fmul float %1966, %1694, !dbg !33 + %2032 = fmul float %1966, %1695, !dbg !33 + %2033 = fmul float %1966, %1696, !dbg !33 + %2034 = fmul float %1966, %1697, !dbg !33 + %2035 = fmul float %1966, %1698, !dbg !33 + %2036 = fmul float %1966, %1699, !dbg !33 + %2037 = fmul float %1968, %1700, !dbg !33 + %2038 = fmul float %1968, %1701, !dbg !33 + %2039 = fmul float %1968, %1702, !dbg !33 + %2040 = fmul float %1968, %1703, !dbg !33 + %2041 = fmul float %1968, %1704, !dbg !33 + %2042 = fmul float %1968, %1705, !dbg !33 + %2043 = fmul float %1968, %1706, !dbg !33 + %2044 = fmul float %1968, %1707, !dbg !33 + %2045 = fmul float %1970, %1708, !dbg !33 + %2046 = fmul float %1970, %1709, !dbg !33 + %2047 = fmul float %1970, %1710, !dbg !33 + %2048 = fmul float %1970, %1711, !dbg !33 + %2049 = fmul float %1970, %1712, !dbg !33 + %2050 = fmul float %1970, %1713, !dbg !33 + %2051 = fmul float %1970, %1714, !dbg !33 + %2052 = fmul float %1970, %1715, !dbg !33 + %2053 = fmul float %1972, %1716, !dbg !33 + %2054 = fmul float %1972, %1717, !dbg !33 + %2055 = fmul float %1972, %1718, !dbg !33 + %2056 = fmul float %1972, %1719, !dbg !33 + %2057 = fmul float %1972, %1720, !dbg !33 + %2058 = fmul float %1972, %1721, !dbg !33 + %2059 = fmul float %1972, %1722, !dbg !33 + %2060 = fmul float %1972, %1723, !dbg !33 + %2061 = fmul float %1974, %1724, !dbg !33 + %2062 = fmul float %1974, %1725, !dbg !33 + %2063 = fmul float %1974, %1726, !dbg !33 + %2064 = fmul float %1974, %1727, !dbg !33 + %2065 = fmul float %1974, %1728, !dbg !33 + %2066 = fmul float %1974, %1729, !dbg !33 + %2067 = fmul float %1974, %1730, !dbg !33 + %2068 = fmul float %1974, %1731, !dbg !33 + %2069 = fmul float %1976, %1732, !dbg !33 + %2070 = fmul float %1976, %1733, !dbg !33 + %2071 = fmul float %1976, %1734, !dbg !33 + %2072 = fmul float %1976, %1735, !dbg !33 + %2073 = fmul float %1976, %1736, !dbg !33 + %2074 = fmul float %1976, %1737, !dbg !33 + %2075 = fmul float %1976, %1738, !dbg !33 + %2076 = fmul float %1976, %1739, !dbg !33 + %2077 = fmul float %1978, %1740, !dbg !33 + %2078 = fmul float %1978, %1741, !dbg !33 + %2079 = fmul float %1978, %1742, !dbg !33 + %2080 = fmul float %1978, %1743, !dbg !33 + %2081 = fmul float %1978, %1744, !dbg !33 + %2082 = fmul float %1978, %1745, !dbg !33 + %2083 = fmul float %1978, %1746, !dbg !33 + %2084 = fmul float %1978, %1747, !dbg !33 + %2085 = fmul float %1980, %1748, !dbg !33 + %2086 = fmul float %1980, %1749, !dbg !33 + %2087 = fmul float %1980, %1750, !dbg !33 + %2088 = fmul float %1980, %1751, !dbg !33 + %2089 = fmul float %1980, %1752, !dbg !33 + %2090 = fmul float %1980, %1753, !dbg !33 + %2091 = fmul float %1980, %1754, !dbg !33 + %2092 = fmul float %1980, %1755, !dbg !33 + %2093 = fmul float %1982, %1756, !dbg !33 + %2094 = fmul float %1982, %1757, !dbg !33 + %2095 = fmul float %1982, %1758, !dbg !33 + %2096 = fmul float %1982, %1759, !dbg !33 + %2097 = fmul float %1982, %1760, !dbg !33 + %2098 = fmul float %1982, %1761, !dbg !33 + %2099 = fmul float %1982, %1762, !dbg !33 + %2100 = fmul float %1982, %1763, !dbg !33 + %2101 = fmul float %1984, %1764, !dbg !33 + %2102 = fmul float %1984, %1765, !dbg !33 + %2103 = fmul float %1984, %1766, !dbg !33 + %2104 = fmul float %1984, %1767, !dbg !33 + %2105 = fmul float %1984, %1768, !dbg !33 + %2106 = fmul float %1984, %1769, !dbg !33 + %2107 = fmul float %1984, %1770, !dbg !33 + %2108 = fmul float %1984, %1771, !dbg !33 + %2109 = fmul float %1986, %1772, !dbg !33 + %2110 = fmul float %1986, %1773, !dbg !33 + %2111 = fmul float %1986, %1774, !dbg !33 + %2112 = fmul float %1986, %1775, !dbg !33 + %2113 = fmul float %1986, %1776, !dbg !33 + %2114 = fmul float %1986, %1777, !dbg !33 + %2115 = fmul float %1986, %1778, !dbg !33 + %2116 = fmul float %1986, %1779, !dbg !33 + %2117 = fmul float %1988, %1780, !dbg !33 + %2118 = fmul float %1988, %1781, !dbg !33 + %2119 = fmul float %1988, %1782, !dbg !33 + %2120 = fmul float %1988, %1783, !dbg !33 + %2121 = fmul float %1988, %1784, !dbg !33 + %2122 = fmul float %1988, %1785, !dbg !33 + %2123 = fmul float %1988, %1786, !dbg !33 + %2124 = fmul float %1988, %1787, !dbg !33 + %2125 = fmul float %1990, %1788, !dbg !33 + %2126 = fmul float %1990, %1789, !dbg !33 + %2127 = fmul float %1990, %1790, !dbg !33 + %2128 = fmul float %1990, %1791, !dbg !33 + %2129 = fmul float %1990, %1792, !dbg !33 + %2130 = fmul float %1990, %1793, !dbg !33 + %2131 = fmul float %1990, %1794, !dbg !33 + %2132 = fmul float %1990, %1795, !dbg !33 + %2133 = fmul float %1992, %1796, !dbg !33 + %2134 = fmul float %1992, %1797, !dbg !33 + %2135 = fmul float %1992, %1798, !dbg !33 + %2136 = fmul float %1992, %1799, !dbg !33 + %2137 = fmul float %1992, %1800, !dbg !33 + %2138 = fmul float %1992, %1801, !dbg !33 + %2139 = fmul float %1992, %1802, !dbg !33 + %2140 = fmul float %1992, %1803, !dbg !33 + %2141 = fmul float %1994, %1804, !dbg !33 + %2142 = fmul float %1994, %1805, !dbg !33 + %2143 = fmul float %1994, %1806, !dbg !33 + %2144 = fmul float %1994, %1807, !dbg !33 + %2145 = fmul float %1994, %1808, !dbg !33 + %2146 = fmul float %1994, %1809, !dbg !33 + %2147 = fmul float %1994, %1810, !dbg !33 + %2148 = fmul float %1994, %1811, !dbg !33 + %2149 = fmul float %1996, %1812, !dbg !33 + %2150 = fmul float %1996, %1813, !dbg !33 + %2151 = fmul float %1996, %1814, !dbg !33 + %2152 = fmul float %1996, %1815, !dbg !33 + %2153 = fmul float %1996, %1816, !dbg !33 + %2154 = fmul float %1996, %1817, !dbg !33 + %2155 = fmul float %1996, %1818, !dbg !33 + %2156 = fmul float %1996, %1819, !dbg !33 + %2157 = fmul float %1998, %1820, !dbg !33 + %2158 = fmul float %1998, %1821, !dbg !33 + %2159 = fmul float %1998, %1822, !dbg !33 + %2160 = fmul float %1998, %1823, !dbg !33 + %2161 = fmul float %1998, %1824, !dbg !33 + %2162 = fmul float %1998, %1825, !dbg !33 + %2163 = fmul float %1998, %1826, !dbg !33 + %2164 = fmul float %1998, %1827, !dbg !33 + %2165 = fmul float %2000, %1828, !dbg !33 + %2166 = fmul float %2000, %1829, !dbg !33 + %2167 = fmul float %2000, %1830, !dbg !33 + %2168 = fmul float %2000, %1831, !dbg !33 + %2169 = fmul float %2000, %1832, !dbg !33 + %2170 = fmul float %2000, %1833, !dbg !33 + %2171 = fmul float %2000, %1834, !dbg !33 + %2172 = fmul float %2000, %1835, !dbg !33 + %2173 = fmul float %2002, %1836, !dbg !33 + %2174 = fmul float %2002, %1837, !dbg !33 + %2175 = fmul float %2002, %1838, !dbg !33 + %2176 = fmul float %2002, %1839, !dbg !33 + %2177 = fmul float %2002, %1840, !dbg !33 + %2178 = fmul float %2002, %1841, !dbg !33 + %2179 = fmul float %2002, %1842, !dbg !33 + %2180 = fmul float %2002, %1843, !dbg !33 + %2181 = fmul float %2004, %1844, !dbg !33 + %2182 = fmul float %2004, %1845, !dbg !33 + %2183 = fmul float %2004, %1846, !dbg !33 + %2184 = fmul float %2004, %1847, !dbg !33 + %2185 = fmul float %2004, %1848, !dbg !33 + %2186 = fmul float %2004, %1849, !dbg !33 + %2187 = fmul float %2004, %1850, !dbg !33 + %2188 = fmul float %2004, %1851, !dbg !33 + %2189 = fmul float %2006, %1852, !dbg !33 + %2190 = fmul float %2006, %1853, !dbg !33 + %2191 = fmul float %2006, %1854, !dbg !33 + %2192 = fmul float %2006, %1855, !dbg !33 + %2193 = fmul float %2006, %1856, !dbg !33 + %2194 = fmul float %2006, %1857, !dbg !33 + %2195 = fmul float %2006, %1858, !dbg !33 + %2196 = fmul float %2006, %1859, !dbg !33 + %2197 = fmul float %2008, %1860, !dbg !33 + %2198 = fmul float %2008, %1861, !dbg !33 + %2199 = fmul float %2008, %1862, !dbg !33 + %2200 = fmul float %2008, %1863, !dbg !33 + %2201 = fmul float %2008, %1864, !dbg !33 + %2202 = fmul float %2008, %1865, !dbg !33 + %2203 = fmul float %2008, %1866, !dbg !33 + %2204 = fmul float %2008, %1867, !dbg !33 + %2205 = fmul float %2010, %1868, !dbg !33 + %2206 = fmul float %2010, %1869, !dbg !33 + %2207 = fmul float %2010, %1870, !dbg !33 + %2208 = fmul float %2010, %1871, !dbg !33 + %2209 = fmul float %2010, %1872, !dbg !33 + %2210 = fmul float %2010, %1873, !dbg !33 + %2211 = fmul float %2010, %1874, !dbg !33 + %2212 = fmul float %2010, %1875, !dbg !33 + %2213 = fmul float %2012, %1876, !dbg !33 + %2214 = fmul float %2012, %1877, !dbg !33 + %2215 = fmul float %2012, %1878, !dbg !33 + %2216 = fmul float %2012, %1879, !dbg !33 + %2217 = fmul float %2012, %1880, !dbg !33 + %2218 = fmul float %2012, %1881, !dbg !33 + %2219 = fmul float %2012, %1882, !dbg !33 + %2220 = fmul float %2012, %1883, !dbg !33 + %2221 = fmul float %2014, %1884, !dbg !33 + %2222 = fmul float %2014, %1885, !dbg !33 + %2223 = fmul float %2014, %1886, !dbg !33 + %2224 = fmul float %2014, %1887, !dbg !33 + %2225 = fmul float %2014, %1888, !dbg !33 + %2226 = fmul float %2014, %1889, !dbg !33 + %2227 = fmul float %2014, %1890, !dbg !33 + %2228 = fmul float %2014, %1891, !dbg !33 + %2229 = fmul float %2016, %1892, !dbg !33 + %2230 = fmul float %2016, %1893, !dbg !33 + %2231 = fmul float %2016, %1894, !dbg !33 + %2232 = fmul float %2016, %1895, !dbg !33 + %2233 = fmul float %2016, %1896, !dbg !33 + %2234 = fmul float %2016, %1897, !dbg !33 + %2235 = fmul float %2016, %1898, !dbg !33 + %2236 = fmul float %2016, %1899, !dbg !33 + %2237 = fmul float %2018, %1900, !dbg !33 + %2238 = fmul float %2018, %1901, !dbg !33 + %2239 = fmul float %2018, %1902, !dbg !33 + %2240 = fmul float %2018, %1903, !dbg !33 + %2241 = fmul float %2018, %1904, !dbg !33 + %2242 = fmul float %2018, %1905, !dbg !33 + %2243 = fmul float %2018, %1906, !dbg !33 + %2244 = fmul float %2018, %1907, !dbg !33 + %2245 = fmul float %2020, %1908, !dbg !33 + %2246 = fmul float %2020, %1909, !dbg !33 + %2247 = fmul float %2020, %1910, !dbg !33 + %2248 = fmul float %2020, %1911, !dbg !33 + %2249 = fmul float %2020, %1912, !dbg !33 + %2250 = fmul float %2020, %1913, !dbg !33 + %2251 = fmul float %2020, %1914, !dbg !33 + %2252 = fmul float %2020, %1915, !dbg !33 + %2253 = fmul float %2022, %1916, !dbg !33 + %2254 = fmul float %2022, %1917, !dbg !33 + %2255 = fmul float %2022, %1918, !dbg !33 + %2256 = fmul float %2022, %1919, !dbg !33 + %2257 = fmul float %2022, %1920, !dbg !33 + %2258 = fmul float %2022, %1921, !dbg !33 + %2259 = fmul float %2022, %1922, !dbg !33 + %2260 = fmul float %2022, %1923, !dbg !33 + %2261 = fmul float %2024, %1924, !dbg !33 + %2262 = fmul float %2024, %1925, !dbg !33 + %2263 = fmul float %2024, %1926, !dbg !33 + %2264 = fmul float %2024, %1927, !dbg !33 + %2265 = fmul float %2024, %1928, !dbg !33 + %2266 = fmul float %2024, %1929, !dbg !33 + %2267 = fmul float %2024, %1930, !dbg !33 + %2268 = fmul float %2024, %1931, !dbg !33 + %2269 = fmul float %2026, %1932, !dbg !33 + %2270 = fmul float %2026, %1933, !dbg !33 + %2271 = fmul float %2026, %1934, !dbg !33 + %2272 = fmul float %2026, %1935, !dbg !33 + %2273 = fmul float %2026, %1936, !dbg !33 + %2274 = fmul float %2026, %1937, !dbg !33 + %2275 = fmul float %2026, %1938, !dbg !33 + %2276 = fmul float %2026, %1939, !dbg !33 + %2277 = fmul float %2028, %1940, !dbg !33 + %2278 = fmul float %2028, %1941, !dbg !33 + %2279 = fmul float %2028, %1942, !dbg !33 + %2280 = fmul float %2028, %1943, !dbg !33 + %2281 = fmul float %2028, %1944, !dbg !33 + %2282 = fmul float %2028, %1945, !dbg !33 + %2283 = fmul float %2028, %1946, !dbg !33 + %2284 = fmul float %2028, %1947, !dbg !33 + %2285 = fmul float %888, %1692, !dbg !34 + %2286 = fmul float %889, %1693, !dbg !34 + %2287 = fmul float %890, %1694, !dbg !34 + %2288 = fmul float %891, %1695, !dbg !34 + %2289 = fmul float %892, %1696, !dbg !34 + %2290 = fmul float %893, %1697, !dbg !34 + %2291 = fmul float %894, %1698, !dbg !34 + %2292 = fmul float %895, %1699, !dbg !34 + %2293 = fmul float %896, %1700, !dbg !34 + %2294 = fmul float %897, %1701, !dbg !34 + %2295 = fmul float %898, %1702, !dbg !34 + %2296 = fmul float %899, %1703, !dbg !34 + %2297 = fmul float %900, %1704, !dbg !34 + %2298 = fmul float %901, %1705, !dbg !34 + %2299 = fmul float %902, %1706, !dbg !34 + %2300 = fmul float %903, %1707, !dbg !34 + %2301 = fmul float %904, %1708, !dbg !34 + %2302 = fmul float %905, %1709, !dbg !34 + %2303 = fmul float %906, %1710, !dbg !34 + %2304 = fmul float %907, %1711, !dbg !34 + %2305 = fmul float %908, %1712, !dbg !34 + %2306 = fmul float %909, %1713, !dbg !34 + %2307 = fmul float %910, %1714, !dbg !34 + %2308 = fmul float %911, %1715, !dbg !34 + %2309 = fmul float %912, %1716, !dbg !34 + %2310 = fmul float %913, %1717, !dbg !34 + %2311 = fmul float %914, %1718, !dbg !34 + %2312 = fmul float %915, %1719, !dbg !34 + %2313 = fmul float %916, %1720, !dbg !34 + %2314 = fmul float %917, %1721, !dbg !34 + %2315 = fmul float %918, %1722, !dbg !34 + %2316 = fmul float %919, %1723, !dbg !34 + %2317 = fmul float %920, %1724, !dbg !34 + %2318 = fmul float %921, %1725, !dbg !34 + %2319 = fmul float %922, %1726, !dbg !34 + %2320 = fmul float %923, %1727, !dbg !34 + %2321 = fmul float %924, %1728, !dbg !34 + %2322 = fmul float %925, %1729, !dbg !34 + %2323 = fmul float %926, %1730, !dbg !34 + %2324 = fmul float %927, %1731, !dbg !34 + %2325 = fmul float %928, %1732, !dbg !34 + %2326 = fmul float %929, %1733, !dbg !34 + %2327 = fmul float %930, %1734, !dbg !34 + %2328 = fmul float %931, %1735, !dbg !34 + %2329 = fmul float %932, %1736, !dbg !34 + %2330 = fmul float %933, %1737, !dbg !34 + %2331 = fmul float %934, %1738, !dbg !34 + %2332 = fmul float %935, %1739, !dbg !34 + %2333 = fmul float %936, %1740, !dbg !34 + %2334 = fmul float %937, %1741, !dbg !34 + %2335 = fmul float %938, %1742, !dbg !34 + %2336 = fmul float %939, %1743, !dbg !34 + %2337 = fmul float %940, %1744, !dbg !34 + %2338 = fmul float %941, %1745, !dbg !34 + %2339 = fmul float %942, %1746, !dbg !34 + %2340 = fmul float %943, %1747, !dbg !34 + %2341 = fmul float %944, %1748, !dbg !34 + %2342 = fmul float %945, %1749, !dbg !34 + %2343 = fmul float %946, %1750, !dbg !34 + %2344 = fmul float %947, %1751, !dbg !34 + %2345 = fmul float %948, %1752, !dbg !34 + %2346 = fmul float %949, %1753, !dbg !34 + %2347 = fmul float %950, %1754, !dbg !34 + %2348 = fmul float %951, %1755, !dbg !34 + %2349 = fmul float %952, %1756, !dbg !34 + %2350 = fmul float %953, %1757, !dbg !34 + %2351 = fmul float %954, %1758, !dbg !34 + %2352 = fmul float %955, %1759, !dbg !34 + %2353 = fmul float %956, %1760, !dbg !34 + %2354 = fmul float %957, %1761, !dbg !34 + %2355 = fmul float %958, %1762, !dbg !34 + %2356 = fmul float %959, %1763, !dbg !34 + %2357 = fmul float %960, %1764, !dbg !34 + %2358 = fmul float %961, %1765, !dbg !34 + %2359 = fmul float %962, %1766, !dbg !34 + %2360 = fmul float %963, %1767, !dbg !34 + %2361 = fmul float %964, %1768, !dbg !34 + %2362 = fmul float %965, %1769, !dbg !34 + %2363 = fmul float %966, %1770, !dbg !34 + %2364 = fmul float %967, %1771, !dbg !34 + %2365 = fmul float %968, %1772, !dbg !34 + %2366 = fmul float %969, %1773, !dbg !34 + %2367 = fmul float %970, %1774, !dbg !34 + %2368 = fmul float %971, %1775, !dbg !34 + %2369 = fmul float %972, %1776, !dbg !34 + %2370 = fmul float %973, %1777, !dbg !34 + %2371 = fmul float %974, %1778, !dbg !34 + %2372 = fmul float %975, %1779, !dbg !34 + %2373 = fmul float %976, %1780, !dbg !34 + %2374 = fmul float %977, %1781, !dbg !34 + %2375 = fmul float %978, %1782, !dbg !34 + %2376 = fmul float %979, %1783, !dbg !34 + %2377 = fmul float %980, %1784, !dbg !34 + %2378 = fmul float %981, %1785, !dbg !34 + %2379 = fmul float %982, %1786, !dbg !34 + %2380 = fmul float %983, %1787, !dbg !34 + %2381 = fmul float %984, %1788, !dbg !34 + %2382 = fmul float %985, %1789, !dbg !34 + %2383 = fmul float %986, %1790, !dbg !34 + %2384 = fmul float %987, %1791, !dbg !34 + %2385 = fmul float %988, %1792, !dbg !34 + %2386 = fmul float %989, %1793, !dbg !34 + %2387 = fmul float %990, %1794, !dbg !34 + %2388 = fmul float %991, %1795, !dbg !34 + %2389 = fmul float %992, %1796, !dbg !34 + %2390 = fmul float %993, %1797, !dbg !34 + %2391 = fmul float %994, %1798, !dbg !34 + %2392 = fmul float %995, %1799, !dbg !34 + %2393 = fmul float %996, %1800, !dbg !34 + %2394 = fmul float %997, %1801, !dbg !34 + %2395 = fmul float %998, %1802, !dbg !34 + %2396 = fmul float %999, %1803, !dbg !34 + %2397 = fmul float %1000, %1804, !dbg !34 + %2398 = fmul float %1001, %1805, !dbg !34 + %2399 = fmul float %1002, %1806, !dbg !34 + %2400 = fmul float %1003, %1807, !dbg !34 + %2401 = fmul float %1004, %1808, !dbg !34 + %2402 = fmul float %1005, %1809, !dbg !34 + %2403 = fmul float %1006, %1810, !dbg !34 + %2404 = fmul float %1007, %1811, !dbg !34 + %2405 = fmul float %1008, %1812, !dbg !34 + %2406 = fmul float %1009, %1813, !dbg !34 + %2407 = fmul float %1010, %1814, !dbg !34 + %2408 = fmul float %1011, %1815, !dbg !34 + %2409 = fmul float %1012, %1816, !dbg !34 + %2410 = fmul float %1013, %1817, !dbg !34 + %2411 = fmul float %1014, %1818, !dbg !34 + %2412 = fmul float %1015, %1819, !dbg !34 + %2413 = fmul float %1016, %1820, !dbg !34 + %2414 = fmul float %1017, %1821, !dbg !34 + %2415 = fmul float %1018, %1822, !dbg !34 + %2416 = fmul float %1019, %1823, !dbg !34 + %2417 = fmul float %1020, %1824, !dbg !34 + %2418 = fmul float %1021, %1825, !dbg !34 + %2419 = fmul float %1022, %1826, !dbg !34 + %2420 = fmul float %1023, %1827, !dbg !34 + %2421 = fmul float %1024, %1828, !dbg !34 + %2422 = fmul float %1025, %1829, !dbg !34 + %2423 = fmul float %1026, %1830, !dbg !34 + %2424 = fmul float %1027, %1831, !dbg !34 + %2425 = fmul float %1028, %1832, !dbg !34 + %2426 = fmul float %1029, %1833, !dbg !34 + %2427 = fmul float %1030, %1834, !dbg !34 + %2428 = fmul float %1031, %1835, !dbg !34 + %2429 = fmul float %1032, %1836, !dbg !34 + %2430 = fmul float %1033, %1837, !dbg !34 + %2431 = fmul float %1034, %1838, !dbg !34 + %2432 = fmul float %1035, %1839, !dbg !34 + %2433 = fmul float %1036, %1840, !dbg !34 + %2434 = fmul float %1037, %1841, !dbg !34 + %2435 = fmul float %1038, %1842, !dbg !34 + %2436 = fmul float %1039, %1843, !dbg !34 + %2437 = fmul float %1040, %1844, !dbg !34 + %2438 = fmul float %1041, %1845, !dbg !34 + %2439 = fmul float %1042, %1846, !dbg !34 + %2440 = fmul float %1043, %1847, !dbg !34 + %2441 = fmul float %1044, %1848, !dbg !34 + %2442 = fmul float %1045, %1849, !dbg !34 + %2443 = fmul float %1046, %1850, !dbg !34 + %2444 = fmul float %1047, %1851, !dbg !34 + %2445 = fmul float %1048, %1852, !dbg !34 + %2446 = fmul float %1049, %1853, !dbg !34 + %2447 = fmul float %1050, %1854, !dbg !34 + %2448 = fmul float %1051, %1855, !dbg !34 + %2449 = fmul float %1052, %1856, !dbg !34 + %2450 = fmul float %1053, %1857, !dbg !34 + %2451 = fmul float %1054, %1858, !dbg !34 + %2452 = fmul float %1055, %1859, !dbg !34 + %2453 = fmul float %1056, %1860, !dbg !34 + %2454 = fmul float %1057, %1861, !dbg !34 + %2455 = fmul float %1058, %1862, !dbg !34 + %2456 = fmul float %1059, %1863, !dbg !34 + %2457 = fmul float %1060, %1864, !dbg !34 + %2458 = fmul float %1061, %1865, !dbg !34 + %2459 = fmul float %1062, %1866, !dbg !34 + %2460 = fmul float %1063, %1867, !dbg !34 + %2461 = fmul float %1064, %1868, !dbg !34 + %2462 = fmul float %1065, %1869, !dbg !34 + %2463 = fmul float %1066, %1870, !dbg !34 + %2464 = fmul float %1067, %1871, !dbg !34 + %2465 = fmul float %1068, %1872, !dbg !34 + %2466 = fmul float %1069, %1873, !dbg !34 + %2467 = fmul float %1070, %1874, !dbg !34 + %2468 = fmul float %1071, %1875, !dbg !34 + %2469 = fmul float %1072, %1876, !dbg !34 + %2470 = fmul float %1073, %1877, !dbg !34 + %2471 = fmul float %1074, %1878, !dbg !34 + %2472 = fmul float %1075, %1879, !dbg !34 + %2473 = fmul float %1076, %1880, !dbg !34 + %2474 = fmul float %1077, %1881, !dbg !34 + %2475 = fmul float %1078, %1882, !dbg !34 + %2476 = fmul float %1079, %1883, !dbg !34 + %2477 = fmul float %1080, %1884, !dbg !34 + %2478 = fmul float %1081, %1885, !dbg !34 + %2479 = fmul float %1082, %1886, !dbg !34 + %2480 = fmul float %1083, %1887, !dbg !34 + %2481 = fmul float %1084, %1888, !dbg !34 + %2482 = fmul float %1085, %1889, !dbg !34 + %2483 = fmul float %1086, %1890, !dbg !34 + %2484 = fmul float %1087, %1891, !dbg !34 + %2485 = fmul float %1088, %1892, !dbg !34 + %2486 = fmul float %1089, %1893, !dbg !34 + %2487 = fmul float %1090, %1894, !dbg !34 + %2488 = fmul float %1091, %1895, !dbg !34 + %2489 = fmul float %1092, %1896, !dbg !34 + %2490 = fmul float %1093, %1897, !dbg !34 + %2491 = fmul float %1094, %1898, !dbg !34 + %2492 = fmul float %1095, %1899, !dbg !34 + %2493 = fmul float %1096, %1900, !dbg !34 + %2494 = fmul float %1097, %1901, !dbg !34 + %2495 = fmul float %1098, %1902, !dbg !34 + %2496 = fmul float %1099, %1903, !dbg !34 + %2497 = fmul float %1100, %1904, !dbg !34 + %2498 = fmul float %1101, %1905, !dbg !34 + %2499 = fmul float %1102, %1906, !dbg !34 + %2500 = fmul float %1103, %1907, !dbg !34 + %2501 = fmul float %1104, %1908, !dbg !34 + %2502 = fmul float %1105, %1909, !dbg !34 + %2503 = fmul float %1106, %1910, !dbg !34 + %2504 = fmul float %1107, %1911, !dbg !34 + %2505 = fmul float %1108, %1912, !dbg !34 + %2506 = fmul float %1109, %1913, !dbg !34 + %2507 = fmul float %1110, %1914, !dbg !34 + %2508 = fmul float %1111, %1915, !dbg !34 + %2509 = fmul float %1112, %1916, !dbg !34 + %2510 = fmul float %1113, %1917, !dbg !34 + %2511 = fmul float %1114, %1918, !dbg !34 + %2512 = fmul float %1115, %1919, !dbg !34 + %2513 = fmul float %1116, %1920, !dbg !34 + %2514 = fmul float %1117, %1921, !dbg !34 + %2515 = fmul float %1118, %1922, !dbg !34 + %2516 = fmul float %1119, %1923, !dbg !34 + %2517 = fmul float %1120, %1924, !dbg !34 + %2518 = fmul float %1121, %1925, !dbg !34 + %2519 = fmul float %1122, %1926, !dbg !34 + %2520 = fmul float %1123, %1927, !dbg !34 + %2521 = fmul float %1124, %1928, !dbg !34 + %2522 = fmul float %1125, %1929, !dbg !34 + %2523 = fmul float %1126, %1930, !dbg !34 + %2524 = fmul float %1127, %1931, !dbg !34 + %2525 = fmul float %1128, %1932, !dbg !34 + %2526 = fmul float %1129, %1933, !dbg !34 + %2527 = fmul float %1130, %1934, !dbg !34 + %2528 = fmul float %1131, %1935, !dbg !34 + %2529 = fmul float %1132, %1936, !dbg !34 + %2530 = fmul float %1133, %1937, !dbg !34 + %2531 = fmul float %1134, %1938, !dbg !34 + %2532 = fmul float %1135, %1939, !dbg !34 + %2533 = fmul float %1136, %1940, !dbg !34 + %2534 = fmul float %1137, %1941, !dbg !34 + %2535 = fmul float %1138, %1942, !dbg !34 + %2536 = fmul float %1139, %1943, !dbg !34 + %2537 = fmul float %1140, %1944, !dbg !34 + %2538 = fmul float %1141, %1945, !dbg !34 + %2539 = fmul float %1142, %1946, !dbg !34 + %2540 = fmul float %1143, %1947, !dbg !34 + %2541 = fadd float %2285, %2286, !dbg !35 + %2542 = fadd float %2287, %2541, !dbg !35 + %2543 = fadd float %2288, %2542, !dbg !35 + %2544 = fadd float %2289, %2543, !dbg !35 + %2545 = fadd float %2290, %2544, !dbg !35 + %2546 = fadd float %2291, %2545, !dbg !35 + %2547 = fadd float %2292, %2546, !dbg !35 + %2548 = fadd float %2293, %2294, !dbg !35 + %2549 = fadd float %2295, %2548, !dbg !35 + %2550 = fadd float %2296, %2549, !dbg !35 + %2551 = fadd float %2297, %2550, !dbg !35 + %2552 = fadd float %2298, %2551, !dbg !35 + %2553 = fadd float %2299, %2552, !dbg !35 + %2554 = fadd float %2300, %2553, !dbg !35 + %2555 = fadd float %2301, %2302, !dbg !35 + %2556 = fadd float %2303, %2555, !dbg !35 + %2557 = fadd float %2304, %2556, !dbg !35 + %2558 = fadd float %2305, %2557, !dbg !35 + %2559 = fadd float %2306, %2558, !dbg !35 + %2560 = fadd float %2307, %2559, !dbg !35 + %2561 = fadd float %2308, %2560, !dbg !35 + %2562 = fadd float %2309, %2310, !dbg !35 + %2563 = fadd float %2311, %2562, !dbg !35 + %2564 = fadd float %2312, %2563, !dbg !35 + %2565 = fadd float %2313, %2564, !dbg !35 + %2566 = fadd float %2314, %2565, !dbg !35 + %2567 = fadd float %2315, %2566, !dbg !35 + %2568 = fadd float %2316, %2567, !dbg !35 + %2569 = fadd float %2317, %2318, !dbg !35 + %2570 = fadd float %2319, %2569, !dbg !35 + %2571 = fadd float %2320, %2570, !dbg !35 + %2572 = fadd float %2321, %2571, !dbg !35 + %2573 = fadd float %2322, %2572, !dbg !35 + %2574 = fadd float %2323, %2573, !dbg !35 + %2575 = fadd float %2324, %2574, !dbg !35 + %2576 = fadd float %2325, %2326, !dbg !35 + %2577 = fadd float %2327, %2576, !dbg !35 + %2578 = fadd float %2328, %2577, !dbg !35 + %2579 = fadd float %2329, %2578, !dbg !35 + %2580 = fadd float %2330, %2579, !dbg !35 + %2581 = fadd float %2331, %2580, !dbg !35 + %2582 = fadd float %2332, %2581, !dbg !35 + %2583 = fadd float %2333, %2334, !dbg !35 + %2584 = fadd float %2335, %2583, !dbg !35 + %2585 = fadd float %2336, %2584, !dbg !35 + %2586 = fadd float %2337, %2585, !dbg !35 + %2587 = fadd float %2338, %2586, !dbg !35 + %2588 = fadd float %2339, %2587, !dbg !35 + %2589 = fadd float %2340, %2588, !dbg !35 + %2590 = fadd float %2341, %2342, !dbg !35 + %2591 = fadd float %2343, %2590, !dbg !35 + %2592 = fadd float %2344, %2591, !dbg !35 + %2593 = fadd float %2345, %2592, !dbg !35 + %2594 = fadd float %2346, %2593, !dbg !35 + %2595 = fadd float %2347, %2594, !dbg !35 + %2596 = fadd float %2348, %2595, !dbg !35 + %2597 = fadd float %2349, %2350, !dbg !35 + %2598 = fadd float %2351, %2597, !dbg !35 + %2599 = fadd float %2352, %2598, !dbg !35 + %2600 = fadd float %2353, %2599, !dbg !35 + %2601 = fadd float %2354, %2600, !dbg !35 + %2602 = fadd float %2355, %2601, !dbg !35 + %2603 = fadd float %2356, %2602, !dbg !35 + %2604 = fadd float %2357, %2358, !dbg !35 + %2605 = fadd float %2359, %2604, !dbg !35 + %2606 = fadd float %2360, %2605, !dbg !35 + %2607 = fadd float %2361, %2606, !dbg !35 + %2608 = fadd float %2362, %2607, !dbg !35 + %2609 = fadd float %2363, %2608, !dbg !35 + %2610 = fadd float %2364, %2609, !dbg !35 + %2611 = fadd float %2365, %2366, !dbg !35 + %2612 = fadd float %2367, %2611, !dbg !35 + %2613 = fadd float %2368, %2612, !dbg !35 + %2614 = fadd float %2369, %2613, !dbg !35 + %2615 = fadd float %2370, %2614, !dbg !35 + %2616 = fadd float %2371, %2615, !dbg !35 + %2617 = fadd float %2372, %2616, !dbg !35 + %2618 = fadd float %2373, %2374, !dbg !35 + %2619 = fadd float %2375, %2618, !dbg !35 + %2620 = fadd float %2376, %2619, !dbg !35 + %2621 = fadd float %2377, %2620, !dbg !35 + %2622 = fadd float %2378, %2621, !dbg !35 + %2623 = fadd float %2379, %2622, !dbg !35 + %2624 = fadd float %2380, %2623, !dbg !35 + %2625 = fadd float %2381, %2382, !dbg !35 + %2626 = fadd float %2383, %2625, !dbg !35 + %2627 = fadd float %2384, %2626, !dbg !35 + %2628 = fadd float %2385, %2627, !dbg !35 + %2629 = fadd float %2386, %2628, !dbg !35 + %2630 = fadd float %2387, %2629, !dbg !35 + %2631 = fadd float %2388, %2630, !dbg !35 + %2632 = fadd float %2389, %2390, !dbg !35 + %2633 = fadd float %2391, %2632, !dbg !35 + %2634 = fadd float %2392, %2633, !dbg !35 + %2635 = fadd float %2393, %2634, !dbg !35 + %2636 = fadd float %2394, %2635, !dbg !35 + %2637 = fadd float %2395, %2636, !dbg !35 + %2638 = fadd float %2396, %2637, !dbg !35 + %2639 = fadd float %2397, %2398, !dbg !35 + %2640 = fadd float %2399, %2639, !dbg !35 + %2641 = fadd float %2400, %2640, !dbg !35 + %2642 = fadd float %2401, %2641, !dbg !35 + %2643 = fadd float %2402, %2642, !dbg !35 + %2644 = fadd float %2403, %2643, !dbg !35 + %2645 = fadd float %2404, %2644, !dbg !35 + %2646 = fadd float %2405, %2406, !dbg !35 + %2647 = fadd float %2407, %2646, !dbg !35 + %2648 = fadd float %2408, %2647, !dbg !35 + %2649 = fadd float %2409, %2648, !dbg !35 + %2650 = fadd float %2410, %2649, !dbg !35 + %2651 = fadd float %2411, %2650, !dbg !35 + %2652 = fadd float %2412, %2651, !dbg !35 + %2653 = fadd float %2413, %2414, !dbg !35 + %2654 = fadd float %2415, %2653, !dbg !35 + %2655 = fadd float %2416, %2654, !dbg !35 + %2656 = fadd float %2417, %2655, !dbg !35 + %2657 = fadd float %2418, %2656, !dbg !35 + %2658 = fadd float %2419, %2657, !dbg !35 + %2659 = fadd float %2420, %2658, !dbg !35 + %2660 = fadd float %2421, %2422, !dbg !35 + %2661 = fadd float %2423, %2660, !dbg !35 + %2662 = fadd float %2424, %2661, !dbg !35 + %2663 = fadd float %2425, %2662, !dbg !35 + %2664 = fadd float %2426, %2663, !dbg !35 + %2665 = fadd float %2427, %2664, !dbg !35 + %2666 = fadd float %2428, %2665, !dbg !35 + %2667 = fadd float %2429, %2430, !dbg !35 + %2668 = fadd float %2431, %2667, !dbg !35 + %2669 = fadd float %2432, %2668, !dbg !35 + %2670 = fadd float %2433, %2669, !dbg !35 + %2671 = fadd float %2434, %2670, !dbg !35 + %2672 = fadd float %2435, %2671, !dbg !35 + %2673 = fadd float %2436, %2672, !dbg !35 + %2674 = fadd float %2437, %2438, !dbg !35 + %2675 = fadd float %2439, %2674, !dbg !35 + %2676 = fadd float %2440, %2675, !dbg !35 + %2677 = fadd float %2441, %2676, !dbg !35 + %2678 = fadd float %2442, %2677, !dbg !35 + %2679 = fadd float %2443, %2678, !dbg !35 + %2680 = fadd float %2444, %2679, !dbg !35 + %2681 = fadd float %2445, %2446, !dbg !35 + %2682 = fadd float %2447, %2681, !dbg !35 + %2683 = fadd float %2448, %2682, !dbg !35 + %2684 = fadd float %2449, %2683, !dbg !35 + %2685 = fadd float %2450, %2684, !dbg !35 + %2686 = fadd float %2451, %2685, !dbg !35 + %2687 = fadd float %2452, %2686, !dbg !35 + %2688 = fadd float %2453, %2454, !dbg !35 + %2689 = fadd float %2455, %2688, !dbg !35 + %2690 = fadd float %2456, %2689, !dbg !35 + %2691 = fadd float %2457, %2690, !dbg !35 + %2692 = fadd float %2458, %2691, !dbg !35 + %2693 = fadd float %2459, %2692, !dbg !35 + %2694 = fadd float %2460, %2693, !dbg !35 + %2695 = fadd float %2461, %2462, !dbg !35 + %2696 = fadd float %2463, %2695, !dbg !35 + %2697 = fadd float %2464, %2696, !dbg !35 + %2698 = fadd float %2465, %2697, !dbg !35 + %2699 = fadd float %2466, %2698, !dbg !35 + %2700 = fadd float %2467, %2699, !dbg !35 + %2701 = fadd float %2468, %2700, !dbg !35 + %2702 = fadd float %2469, %2470, !dbg !35 + %2703 = fadd float %2471, %2702, !dbg !35 + %2704 = fadd float %2472, %2703, !dbg !35 + %2705 = fadd float %2473, %2704, !dbg !35 + %2706 = fadd float %2474, %2705, !dbg !35 + %2707 = fadd float %2475, %2706, !dbg !35 + %2708 = fadd float %2476, %2707, !dbg !35 + %2709 = fadd float %2477, %2478, !dbg !35 + %2710 = fadd float %2479, %2709, !dbg !35 + %2711 = fadd float %2480, %2710, !dbg !35 + %2712 = fadd float %2481, %2711, !dbg !35 + %2713 = fadd float %2482, %2712, !dbg !35 + %2714 = fadd float %2483, %2713, !dbg !35 + %2715 = fadd float %2484, %2714, !dbg !35 + %2716 = fadd float %2485, %2486, !dbg !35 + %2717 = fadd float %2487, %2716, !dbg !35 + %2718 = fadd float %2488, %2717, !dbg !35 + %2719 = fadd float %2489, %2718, !dbg !35 + %2720 = fadd float %2490, %2719, !dbg !35 + %2721 = fadd float %2491, %2720, !dbg !35 + %2722 = fadd float %2492, %2721, !dbg !35 + %2723 = fadd float %2493, %2494, !dbg !35 + %2724 = fadd float %2495, %2723, !dbg !35 + %2725 = fadd float %2496, %2724, !dbg !35 + %2726 = fadd float %2497, %2725, !dbg !35 + %2727 = fadd float %2498, %2726, !dbg !35 + %2728 = fadd float %2499, %2727, !dbg !35 + %2729 = fadd float %2500, %2728, !dbg !35 + %2730 = fadd float %2501, %2502, !dbg !35 + %2731 = fadd float %2503, %2730, !dbg !35 + %2732 = fadd float %2504, %2731, !dbg !35 + %2733 = fadd float %2505, %2732, !dbg !35 + %2734 = fadd float %2506, %2733, !dbg !35 + %2735 = fadd float %2507, %2734, !dbg !35 + %2736 = fadd float %2508, %2735, !dbg !35 + %2737 = fadd float %2509, %2510, !dbg !35 + %2738 = fadd float %2511, %2737, !dbg !35 + %2739 = fadd float %2512, %2738, !dbg !35 + %2740 = fadd float %2513, %2739, !dbg !35 + %2741 = fadd float %2514, %2740, !dbg !35 + %2742 = fadd float %2515, %2741, !dbg !35 + %2743 = fadd float %2516, %2742, !dbg !35 + %2744 = fadd float %2517, %2518, !dbg !35 + %2745 = fadd float %2519, %2744, !dbg !35 + %2746 = fadd float %2520, %2745, !dbg !35 + %2747 = fadd float %2521, %2746, !dbg !35 + %2748 = fadd float %2522, %2747, !dbg !35 + %2749 = fadd float %2523, %2748, !dbg !35 + %2750 = fadd float %2524, %2749, !dbg !35 + %2751 = fadd float %2525, %2526, !dbg !35 + %2752 = fadd float %2527, %2751, !dbg !35 + %2753 = fadd float %2528, %2752, !dbg !35 + %2754 = fadd float %2529, %2753, !dbg !35 + %2755 = fadd float %2530, %2754, !dbg !35 + %2756 = fadd float %2531, %2755, !dbg !35 + %2757 = fadd float %2532, %2756, !dbg !35 + %2758 = fadd float %2533, %2534, !dbg !35 + %2759 = fadd float %2535, %2758, !dbg !35 + %2760 = fadd float %2536, %2759, !dbg !35 + %2761 = fadd float %2537, %2760, !dbg !35 + %2762 = fadd float %2538, %2761, !dbg !35 + %2763 = fadd float %2539, %2762, !dbg !35 + %2764 = fadd float %2540, %2763, !dbg !35 + %2765 = bitcast float %2547 to i32, !dbg !38 + %2766 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2765, i32 8, i32 31), !dbg !38 + %2767 = bitcast i32 %2766 to float, !dbg !38 + %2768 = fadd float %2547, %2767, !dbg !35 + %2769 = bitcast float %2768 to i32, !dbg !38 + %2770 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2769, i32 4, i32 31), !dbg !38 + %2771 = bitcast i32 %2770 to float, !dbg !38 + %2772 = fadd float %2768, %2771, !dbg !35 + %2773 = bitcast float %2772 to i32, !dbg !38 + %2774 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2773, i32 2, i32 31), !dbg !38 + %2775 = bitcast i32 %2774 to float, !dbg !38 + %2776 = fadd float %2772, %2775, !dbg !35 + %2777 = bitcast float %2776 to i32, !dbg !38 + %2778 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2777, i32 1, i32 31), !dbg !38 + %2779 = bitcast i32 %2778 to float, !dbg !38 + %2780 = fadd float %2776, %2779, !dbg !35 + %2781 = bitcast float %2554 to i32, !dbg !38 + %2782 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2781, i32 8, i32 31), !dbg !38 + %2783 = bitcast i32 %2782 to float, !dbg !38 + %2784 = fadd float %2554, %2783, !dbg !35 + %2785 = bitcast float %2784 to i32, !dbg !38 + %2786 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2785, i32 4, i32 31), !dbg !38 + %2787 = bitcast i32 %2786 to float, !dbg !38 + %2788 = fadd float %2784, %2787, !dbg !35 + %2789 = bitcast float %2788 to i32, !dbg !38 + %2790 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2789, i32 2, i32 31), !dbg !38 + %2791 = bitcast i32 %2790 to float, !dbg !38 + %2792 = fadd float %2788, %2791, !dbg !35 + %2793 = bitcast float %2792 to i32, !dbg !38 + %2794 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2793, i32 1, i32 31), !dbg !38 + %2795 = bitcast i32 %2794 to float, !dbg !38 + %2796 = fadd float %2792, %2795, !dbg !35 + %2797 = bitcast float %2561 to i32, !dbg !38 + %2798 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2797, i32 8, i32 31), !dbg !38 + %2799 = bitcast i32 %2798 to float, !dbg !38 + %2800 = fadd float %2561, %2799, !dbg !35 + %2801 = bitcast float %2800 to i32, !dbg !38 + %2802 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2801, i32 4, i32 31), !dbg !38 + %2803 = bitcast i32 %2802 to float, !dbg !38 + %2804 = fadd float %2800, %2803, !dbg !35 + %2805 = bitcast float %2804 to i32, !dbg !38 + %2806 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2805, i32 2, i32 31), !dbg !38 + %2807 = bitcast i32 %2806 to float, !dbg !38 + %2808 = fadd float %2804, %2807, !dbg !35 + %2809 = bitcast float %2808 to i32, !dbg !38 + %2810 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2809, i32 1, i32 31), !dbg !38 + %2811 = bitcast i32 %2810 to float, !dbg !38 + %2812 = fadd float %2808, %2811, !dbg !35 + %2813 = bitcast float %2568 to i32, !dbg !38 + %2814 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2813, i32 8, i32 31), !dbg !38 + %2815 = bitcast i32 %2814 to float, !dbg !38 + %2816 = fadd float %2568, %2815, !dbg !35 + %2817 = bitcast float %2816 to i32, !dbg !38 + %2818 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2817, i32 4, i32 31), !dbg !38 + %2819 = bitcast i32 %2818 to float, !dbg !38 + %2820 = fadd float %2816, %2819, !dbg !35 + %2821 = bitcast float %2820 to i32, !dbg !38 + %2822 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2821, i32 2, i32 31), !dbg !38 + %2823 = bitcast i32 %2822 to float, !dbg !38 + %2824 = fadd float %2820, %2823, !dbg !35 + %2825 = bitcast float %2824 to i32, !dbg !38 + %2826 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2825, i32 1, i32 31), !dbg !38 + %2827 = bitcast i32 %2826 to float, !dbg !38 + %2828 = fadd float %2824, %2827, !dbg !35 + %2829 = bitcast float %2575 to i32, !dbg !38 + %2830 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2829, i32 8, i32 31), !dbg !38 + %2831 = bitcast i32 %2830 to float, !dbg !38 + %2832 = fadd float %2575, %2831, !dbg !35 + %2833 = bitcast float %2832 to i32, !dbg !38 + %2834 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2833, i32 4, i32 31), !dbg !38 + %2835 = bitcast i32 %2834 to float, !dbg !38 + %2836 = fadd float %2832, %2835, !dbg !35 + %2837 = bitcast float %2836 to i32, !dbg !38 + %2838 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2837, i32 2, i32 31), !dbg !38 + %2839 = bitcast i32 %2838 to float, !dbg !38 + %2840 = fadd float %2836, %2839, !dbg !35 + %2841 = bitcast float %2840 to i32, !dbg !38 + %2842 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2841, i32 1, i32 31), !dbg !38 + %2843 = bitcast i32 %2842 to float, !dbg !38 + %2844 = fadd float %2840, %2843, !dbg !35 + %2845 = bitcast float %2582 to i32, !dbg !38 + %2846 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2845, i32 8, i32 31), !dbg !38 + %2847 = bitcast i32 %2846 to float, !dbg !38 + %2848 = fadd float %2582, %2847, !dbg !35 + %2849 = bitcast float %2848 to i32, !dbg !38 + %2850 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2849, i32 4, i32 31), !dbg !38 + %2851 = bitcast i32 %2850 to float, !dbg !38 + %2852 = fadd float %2848, %2851, !dbg !35 + %2853 = bitcast float %2852 to i32, !dbg !38 + %2854 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2853, i32 2, i32 31), !dbg !38 + %2855 = bitcast i32 %2854 to float, !dbg !38 + %2856 = fadd float %2852, %2855, !dbg !35 + %2857 = bitcast float %2856 to i32, !dbg !38 + %2858 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2857, i32 1, i32 31), !dbg !38 + %2859 = bitcast i32 %2858 to float, !dbg !38 + %2860 = fadd float %2856, %2859, !dbg !35 + %2861 = bitcast float %2589 to i32, !dbg !38 + %2862 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2861, i32 8, i32 31), !dbg !38 + %2863 = bitcast i32 %2862 to float, !dbg !38 + %2864 = fadd float %2589, %2863, !dbg !35 + %2865 = bitcast float %2864 to i32, !dbg !38 + %2866 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2865, i32 4, i32 31), !dbg !38 + %2867 = bitcast i32 %2866 to float, !dbg !38 + %2868 = fadd float %2864, %2867, !dbg !35 + %2869 = bitcast float %2868 to i32, !dbg !38 + %2870 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2869, i32 2, i32 31), !dbg !38 + %2871 = bitcast i32 %2870 to float, !dbg !38 + %2872 = fadd float %2868, %2871, !dbg !35 + %2873 = bitcast float %2872 to i32, !dbg !38 + %2874 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2873, i32 1, i32 31), !dbg !38 + %2875 = bitcast i32 %2874 to float, !dbg !38 + %2876 = fadd float %2872, %2875, !dbg !35 + %2877 = bitcast float %2596 to i32, !dbg !38 + %2878 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2877, i32 8, i32 31), !dbg !38 + %2879 = bitcast i32 %2878 to float, !dbg !38 + %2880 = fadd float %2596, %2879, !dbg !35 + %2881 = bitcast float %2880 to i32, !dbg !38 + %2882 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2881, i32 4, i32 31), !dbg !38 + %2883 = bitcast i32 %2882 to float, !dbg !38 + %2884 = fadd float %2880, %2883, !dbg !35 + %2885 = bitcast float %2884 to i32, !dbg !38 + %2886 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2885, i32 2, i32 31), !dbg !38 + %2887 = bitcast i32 %2886 to float, !dbg !38 + %2888 = fadd float %2884, %2887, !dbg !35 + %2889 = bitcast float %2888 to i32, !dbg !38 + %2890 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2889, i32 1, i32 31), !dbg !38 + %2891 = bitcast i32 %2890 to float, !dbg !38 + %2892 = fadd float %2888, %2891, !dbg !35 + %2893 = bitcast float %2603 to i32, !dbg !38 + %2894 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2893, i32 8, i32 31), !dbg !38 + %2895 = bitcast i32 %2894 to float, !dbg !38 + %2896 = fadd float %2603, %2895, !dbg !35 + %2897 = bitcast float %2896 to i32, !dbg !38 + %2898 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2897, i32 4, i32 31), !dbg !38 + %2899 = bitcast i32 %2898 to float, !dbg !38 + %2900 = fadd float %2896, %2899, !dbg !35 + %2901 = bitcast float %2900 to i32, !dbg !38 + %2902 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2901, i32 2, i32 31), !dbg !38 + %2903 = bitcast i32 %2902 to float, !dbg !38 + %2904 = fadd float %2900, %2903, !dbg !35 + %2905 = bitcast float %2904 to i32, !dbg !38 + %2906 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2905, i32 1, i32 31), !dbg !38 + %2907 = bitcast i32 %2906 to float, !dbg !38 + %2908 = fadd float %2904, %2907, !dbg !35 + %2909 = bitcast float %2610 to i32, !dbg !38 + %2910 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2909, i32 8, i32 31), !dbg !38 + %2911 = bitcast i32 %2910 to float, !dbg !38 + %2912 = fadd float %2610, %2911, !dbg !35 + %2913 = bitcast float %2912 to i32, !dbg !38 + %2914 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2913, i32 4, i32 31), !dbg !38 + %2915 = bitcast i32 %2914 to float, !dbg !38 + %2916 = fadd float %2912, %2915, !dbg !35 + %2917 = bitcast float %2916 to i32, !dbg !38 + %2918 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2917, i32 2, i32 31), !dbg !38 + %2919 = bitcast i32 %2918 to float, !dbg !38 + %2920 = fadd float %2916, %2919, !dbg !35 + %2921 = bitcast float %2920 to i32, !dbg !38 + %2922 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2921, i32 1, i32 31), !dbg !38 + %2923 = bitcast i32 %2922 to float, !dbg !38 + %2924 = fadd float %2920, %2923, !dbg !35 + %2925 = bitcast float %2617 to i32, !dbg !38 + %2926 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2925, i32 8, i32 31), !dbg !38 + %2927 = bitcast i32 %2926 to float, !dbg !38 + %2928 = fadd float %2617, %2927, !dbg !35 + %2929 = bitcast float %2928 to i32, !dbg !38 + %2930 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2929, i32 4, i32 31), !dbg !38 + %2931 = bitcast i32 %2930 to float, !dbg !38 + %2932 = fadd float %2928, %2931, !dbg !35 + %2933 = bitcast float %2932 to i32, !dbg !38 + %2934 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2933, i32 2, i32 31), !dbg !38 + %2935 = bitcast i32 %2934 to float, !dbg !38 + %2936 = fadd float %2932, %2935, !dbg !35 + %2937 = bitcast float %2936 to i32, !dbg !38 + %2938 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2937, i32 1, i32 31), !dbg !38 + %2939 = bitcast i32 %2938 to float, !dbg !38 + %2940 = fadd float %2936, %2939, !dbg !35 + %2941 = bitcast float %2624 to i32, !dbg !38 + %2942 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2941, i32 8, i32 31), !dbg !38 + %2943 = bitcast i32 %2942 to float, !dbg !38 + %2944 = fadd float %2624, %2943, !dbg !35 + %2945 = bitcast float %2944 to i32, !dbg !38 + %2946 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2945, i32 4, i32 31), !dbg !38 + %2947 = bitcast i32 %2946 to float, !dbg !38 + %2948 = fadd float %2944, %2947, !dbg !35 + %2949 = bitcast float %2948 to i32, !dbg !38 + %2950 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2949, i32 2, i32 31), !dbg !38 + %2951 = bitcast i32 %2950 to float, !dbg !38 + %2952 = fadd float %2948, %2951, !dbg !35 + %2953 = bitcast float %2952 to i32, !dbg !38 + %2954 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2953, i32 1, i32 31), !dbg !38 + %2955 = bitcast i32 %2954 to float, !dbg !38 + %2956 = fadd float %2952, %2955, !dbg !35 + %2957 = bitcast float %2631 to i32, !dbg !38 + %2958 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2957, i32 8, i32 31), !dbg !38 + %2959 = bitcast i32 %2958 to float, !dbg !38 + %2960 = fadd float %2631, %2959, !dbg !35 + %2961 = bitcast float %2960 to i32, !dbg !38 + %2962 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2961, i32 4, i32 31), !dbg !38 + %2963 = bitcast i32 %2962 to float, !dbg !38 + %2964 = fadd float %2960, %2963, !dbg !35 + %2965 = bitcast float %2964 to i32, !dbg !38 + %2966 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2965, i32 2, i32 31), !dbg !38 + %2967 = bitcast i32 %2966 to float, !dbg !38 + %2968 = fadd float %2964, %2967, !dbg !35 + %2969 = bitcast float %2968 to i32, !dbg !38 + %2970 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2969, i32 1, i32 31), !dbg !38 + %2971 = bitcast i32 %2970 to float, !dbg !38 + %2972 = fadd float %2968, %2971, !dbg !35 + %2973 = bitcast float %2638 to i32, !dbg !38 + %2974 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2973, i32 8, i32 31), !dbg !38 + %2975 = bitcast i32 %2974 to float, !dbg !38 + %2976 = fadd float %2638, %2975, !dbg !35 + %2977 = bitcast float %2976 to i32, !dbg !38 + %2978 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2977, i32 4, i32 31), !dbg !38 + %2979 = bitcast i32 %2978 to float, !dbg !38 + %2980 = fadd float %2976, %2979, !dbg !35 + %2981 = bitcast float %2980 to i32, !dbg !38 + %2982 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2981, i32 2, i32 31), !dbg !38 + %2983 = bitcast i32 %2982 to float, !dbg !38 + %2984 = fadd float %2980, %2983, !dbg !35 + %2985 = bitcast float %2984 to i32, !dbg !38 + %2986 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2985, i32 1, i32 31), !dbg !38 + %2987 = bitcast i32 %2986 to float, !dbg !38 + %2988 = fadd float %2984, %2987, !dbg !35 + %2989 = bitcast float %2645 to i32, !dbg !38 + %2990 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2989, i32 8, i32 31), !dbg !38 + %2991 = bitcast i32 %2990 to float, !dbg !38 + %2992 = fadd float %2645, %2991, !dbg !35 + %2993 = bitcast float %2992 to i32, !dbg !38 + %2994 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2993, i32 4, i32 31), !dbg !38 + %2995 = bitcast i32 %2994 to float, !dbg !38 + %2996 = fadd float %2992, %2995, !dbg !35 + %2997 = bitcast float %2996 to i32, !dbg !38 + %2998 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2997, i32 2, i32 31), !dbg !38 + %2999 = bitcast i32 %2998 to float, !dbg !38 + %3000 = fadd float %2996, %2999, !dbg !35 + %3001 = bitcast float %3000 to i32, !dbg !38 + %3002 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3001, i32 1, i32 31), !dbg !38 + %3003 = bitcast i32 %3002 to float, !dbg !38 + %3004 = fadd float %3000, %3003, !dbg !35 + %3005 = bitcast float %2652 to i32, !dbg !38 + %3006 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3005, i32 8, i32 31), !dbg !38 + %3007 = bitcast i32 %3006 to float, !dbg !38 + %3008 = fadd float %2652, %3007, !dbg !35 + %3009 = bitcast float %3008 to i32, !dbg !38 + %3010 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3009, i32 4, i32 31), !dbg !38 + %3011 = bitcast i32 %3010 to float, !dbg !38 + %3012 = fadd float %3008, %3011, !dbg !35 + %3013 = bitcast float %3012 to i32, !dbg !38 + %3014 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3013, i32 2, i32 31), !dbg !38 + %3015 = bitcast i32 %3014 to float, !dbg !38 + %3016 = fadd float %3012, %3015, !dbg !35 + %3017 = bitcast float %3016 to i32, !dbg !38 + %3018 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3017, i32 1, i32 31), !dbg !38 + %3019 = bitcast i32 %3018 to float, !dbg !38 + %3020 = fadd float %3016, %3019, !dbg !35 + %3021 = bitcast float %2659 to i32, !dbg !38 + %3022 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3021, i32 8, i32 31), !dbg !38 + %3023 = bitcast i32 %3022 to float, !dbg !38 + %3024 = fadd float %2659, %3023, !dbg !35 + %3025 = bitcast float %3024 to i32, !dbg !38 + %3026 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3025, i32 4, i32 31), !dbg !38 + %3027 = bitcast i32 %3026 to float, !dbg !38 + %3028 = fadd float %3024, %3027, !dbg !35 + %3029 = bitcast float %3028 to i32, !dbg !38 + %3030 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3029, i32 2, i32 31), !dbg !38 + %3031 = bitcast i32 %3030 to float, !dbg !38 + %3032 = fadd float %3028, %3031, !dbg !35 + %3033 = bitcast float %3032 to i32, !dbg !38 + %3034 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3033, i32 1, i32 31), !dbg !38 + %3035 = bitcast i32 %3034 to float, !dbg !38 + %3036 = fadd float %3032, %3035, !dbg !35 + %3037 = bitcast float %2666 to i32, !dbg !38 + %3038 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3037, i32 8, i32 31), !dbg !38 + %3039 = bitcast i32 %3038 to float, !dbg !38 + %3040 = fadd float %2666, %3039, !dbg !35 + %3041 = bitcast float %3040 to i32, !dbg !38 + %3042 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3041, i32 4, i32 31), !dbg !38 + %3043 = bitcast i32 %3042 to float, !dbg !38 + %3044 = fadd float %3040, %3043, !dbg !35 + %3045 = bitcast float %3044 to i32, !dbg !38 + %3046 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3045, i32 2, i32 31), !dbg !38 + %3047 = bitcast i32 %3046 to float, !dbg !38 + %3048 = fadd float %3044, %3047, !dbg !35 + %3049 = bitcast float %3048 to i32, !dbg !38 + %3050 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3049, i32 1, i32 31), !dbg !38 + %3051 = bitcast i32 %3050 to float, !dbg !38 + %3052 = fadd float %3048, %3051, !dbg !35 + %3053 = bitcast float %2673 to i32, !dbg !38 + %3054 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3053, i32 8, i32 31), !dbg !38 + %3055 = bitcast i32 %3054 to float, !dbg !38 + %3056 = fadd float %2673, %3055, !dbg !35 + %3057 = bitcast float %3056 to i32, !dbg !38 + %3058 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3057, i32 4, i32 31), !dbg !38 + %3059 = bitcast i32 %3058 to float, !dbg !38 + %3060 = fadd float %3056, %3059, !dbg !35 + %3061 = bitcast float %3060 to i32, !dbg !38 + %3062 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3061, i32 2, i32 31), !dbg !38 + %3063 = bitcast i32 %3062 to float, !dbg !38 + %3064 = fadd float %3060, %3063, !dbg !35 + %3065 = bitcast float %3064 to i32, !dbg !38 + %3066 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3065, i32 1, i32 31), !dbg !38 + %3067 = bitcast i32 %3066 to float, !dbg !38 + %3068 = fadd float %3064, %3067, !dbg !35 + %3069 = bitcast float %2680 to i32, !dbg !38 + %3070 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3069, i32 8, i32 31), !dbg !38 + %3071 = bitcast i32 %3070 to float, !dbg !38 + %3072 = fadd float %2680, %3071, !dbg !35 + %3073 = bitcast float %3072 to i32, !dbg !38 + %3074 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3073, i32 4, i32 31), !dbg !38 + %3075 = bitcast i32 %3074 to float, !dbg !38 + %3076 = fadd float %3072, %3075, !dbg !35 + %3077 = bitcast float %3076 to i32, !dbg !38 + %3078 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3077, i32 2, i32 31), !dbg !38 + %3079 = bitcast i32 %3078 to float, !dbg !38 + %3080 = fadd float %3076, %3079, !dbg !35 + %3081 = bitcast float %3080 to i32, !dbg !38 + %3082 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3081, i32 1, i32 31), !dbg !38 + %3083 = bitcast i32 %3082 to float, !dbg !38 + %3084 = fadd float %3080, %3083, !dbg !35 + %3085 = bitcast float %2687 to i32, !dbg !38 + %3086 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3085, i32 8, i32 31), !dbg !38 + %3087 = bitcast i32 %3086 to float, !dbg !38 + %3088 = fadd float %2687, %3087, !dbg !35 + %3089 = bitcast float %3088 to i32, !dbg !38 + %3090 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3089, i32 4, i32 31), !dbg !38 + %3091 = bitcast i32 %3090 to float, !dbg !38 + %3092 = fadd float %3088, %3091, !dbg !35 + %3093 = bitcast float %3092 to i32, !dbg !38 + %3094 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3093, i32 2, i32 31), !dbg !38 + %3095 = bitcast i32 %3094 to float, !dbg !38 + %3096 = fadd float %3092, %3095, !dbg !35 + %3097 = bitcast float %3096 to i32, !dbg !38 + %3098 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3097, i32 1, i32 31), !dbg !38 + %3099 = bitcast i32 %3098 to float, !dbg !38 + %3100 = fadd float %3096, %3099, !dbg !35 + %3101 = bitcast float %2694 to i32, !dbg !38 + %3102 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3101, i32 8, i32 31), !dbg !38 + %3103 = bitcast i32 %3102 to float, !dbg !38 + %3104 = fadd float %2694, %3103, !dbg !35 + %3105 = bitcast float %3104 to i32, !dbg !38 + %3106 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3105, i32 4, i32 31), !dbg !38 + %3107 = bitcast i32 %3106 to float, !dbg !38 + %3108 = fadd float %3104, %3107, !dbg !35 + %3109 = bitcast float %3108 to i32, !dbg !38 + %3110 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3109, i32 2, i32 31), !dbg !38 + %3111 = bitcast i32 %3110 to float, !dbg !38 + %3112 = fadd float %3108, %3111, !dbg !35 + %3113 = bitcast float %3112 to i32, !dbg !38 + %3114 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3113, i32 1, i32 31), !dbg !38 + %3115 = bitcast i32 %3114 to float, !dbg !38 + %3116 = fadd float %3112, %3115, !dbg !35 + %3117 = bitcast float %2701 to i32, !dbg !38 + %3118 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3117, i32 8, i32 31), !dbg !38 + %3119 = bitcast i32 %3118 to float, !dbg !38 + %3120 = fadd float %2701, %3119, !dbg !35 + %3121 = bitcast float %3120 to i32, !dbg !38 + %3122 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3121, i32 4, i32 31), !dbg !38 + %3123 = bitcast i32 %3122 to float, !dbg !38 + %3124 = fadd float %3120, %3123, !dbg !35 + %3125 = bitcast float %3124 to i32, !dbg !38 + %3126 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3125, i32 2, i32 31), !dbg !38 + %3127 = bitcast i32 %3126 to float, !dbg !38 + %3128 = fadd float %3124, %3127, !dbg !35 + %3129 = bitcast float %3128 to i32, !dbg !38 + %3130 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3129, i32 1, i32 31), !dbg !38 + %3131 = bitcast i32 %3130 to float, !dbg !38 + %3132 = fadd float %3128, %3131, !dbg !35 + %3133 = bitcast float %2708 to i32, !dbg !38 + %3134 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3133, i32 8, i32 31), !dbg !38 + %3135 = bitcast i32 %3134 to float, !dbg !38 + %3136 = fadd float %2708, %3135, !dbg !35 + %3137 = bitcast float %3136 to i32, !dbg !38 + %3138 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3137, i32 4, i32 31), !dbg !38 + %3139 = bitcast i32 %3138 to float, !dbg !38 + %3140 = fadd float %3136, %3139, !dbg !35 + %3141 = bitcast float %3140 to i32, !dbg !38 + %3142 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3141, i32 2, i32 31), !dbg !38 + %3143 = bitcast i32 %3142 to float, !dbg !38 + %3144 = fadd float %3140, %3143, !dbg !35 + %3145 = bitcast float %3144 to i32, !dbg !38 + %3146 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3145, i32 1, i32 31), !dbg !38 + %3147 = bitcast i32 %3146 to float, !dbg !38 + %3148 = fadd float %3144, %3147, !dbg !35 + %3149 = bitcast float %2715 to i32, !dbg !38 + %3150 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3149, i32 8, i32 31), !dbg !38 + %3151 = bitcast i32 %3150 to float, !dbg !38 + %3152 = fadd float %2715, %3151, !dbg !35 + %3153 = bitcast float %3152 to i32, !dbg !38 + %3154 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3153, i32 4, i32 31), !dbg !38 + %3155 = bitcast i32 %3154 to float, !dbg !38 + %3156 = fadd float %3152, %3155, !dbg !35 + %3157 = bitcast float %3156 to i32, !dbg !38 + %3158 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3157, i32 2, i32 31), !dbg !38 + %3159 = bitcast i32 %3158 to float, !dbg !38 + %3160 = fadd float %3156, %3159, !dbg !35 + %3161 = bitcast float %3160 to i32, !dbg !38 + %3162 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3161, i32 1, i32 31), !dbg !38 + %3163 = bitcast i32 %3162 to float, !dbg !38 + %3164 = fadd float %3160, %3163, !dbg !35 + %3165 = bitcast float %2722 to i32, !dbg !38 + %3166 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3165, i32 8, i32 31), !dbg !38 + %3167 = bitcast i32 %3166 to float, !dbg !38 + %3168 = fadd float %2722, %3167, !dbg !35 + %3169 = bitcast float %3168 to i32, !dbg !38 + %3170 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3169, i32 4, i32 31), !dbg !38 + %3171 = bitcast i32 %3170 to float, !dbg !38 + %3172 = fadd float %3168, %3171, !dbg !35 + %3173 = bitcast float %3172 to i32, !dbg !38 + %3174 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3173, i32 2, i32 31), !dbg !38 + %3175 = bitcast i32 %3174 to float, !dbg !38 + %3176 = fadd float %3172, %3175, !dbg !35 + %3177 = bitcast float %3176 to i32, !dbg !38 + %3178 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3177, i32 1, i32 31), !dbg !38 + %3179 = bitcast i32 %3178 to float, !dbg !38 + %3180 = fadd float %3176, %3179, !dbg !35 + %3181 = bitcast float %2729 to i32, !dbg !38 + %3182 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3181, i32 8, i32 31), !dbg !38 + %3183 = bitcast i32 %3182 to float, !dbg !38 + %3184 = fadd float %2729, %3183, !dbg !35 + %3185 = bitcast float %3184 to i32, !dbg !38 + %3186 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3185, i32 4, i32 31), !dbg !38 + %3187 = bitcast i32 %3186 to float, !dbg !38 + %3188 = fadd float %3184, %3187, !dbg !35 + %3189 = bitcast float %3188 to i32, !dbg !38 + %3190 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3189, i32 2, i32 31), !dbg !38 + %3191 = bitcast i32 %3190 to float, !dbg !38 + %3192 = fadd float %3188, %3191, !dbg !35 + %3193 = bitcast float %3192 to i32, !dbg !38 + %3194 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3193, i32 1, i32 31), !dbg !38 + %3195 = bitcast i32 %3194 to float, !dbg !38 + %3196 = fadd float %3192, %3195, !dbg !35 + %3197 = bitcast float %2736 to i32, !dbg !38 + %3198 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3197, i32 8, i32 31), !dbg !38 + %3199 = bitcast i32 %3198 to float, !dbg !38 + %3200 = fadd float %2736, %3199, !dbg !35 + %3201 = bitcast float %3200 to i32, !dbg !38 + %3202 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3201, i32 4, i32 31), !dbg !38 + %3203 = bitcast i32 %3202 to float, !dbg !38 + %3204 = fadd float %3200, %3203, !dbg !35 + %3205 = bitcast float %3204 to i32, !dbg !38 + %3206 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3205, i32 2, i32 31), !dbg !38 + %3207 = bitcast i32 %3206 to float, !dbg !38 + %3208 = fadd float %3204, %3207, !dbg !35 + %3209 = bitcast float %3208 to i32, !dbg !38 + %3210 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3209, i32 1, i32 31), !dbg !38 + %3211 = bitcast i32 %3210 to float, !dbg !38 + %3212 = fadd float %3208, %3211, !dbg !35 + %3213 = bitcast float %2743 to i32, !dbg !38 + %3214 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3213, i32 8, i32 31), !dbg !38 + %3215 = bitcast i32 %3214 to float, !dbg !38 + %3216 = fadd float %2743, %3215, !dbg !35 + %3217 = bitcast float %3216 to i32, !dbg !38 + %3218 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3217, i32 4, i32 31), !dbg !38 + %3219 = bitcast i32 %3218 to float, !dbg !38 + %3220 = fadd float %3216, %3219, !dbg !35 + %3221 = bitcast float %3220 to i32, !dbg !38 + %3222 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3221, i32 2, i32 31), !dbg !38 + %3223 = bitcast i32 %3222 to float, !dbg !38 + %3224 = fadd float %3220, %3223, !dbg !35 + %3225 = bitcast float %3224 to i32, !dbg !38 + %3226 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3225, i32 1, i32 31), !dbg !38 + %3227 = bitcast i32 %3226 to float, !dbg !38 + %3228 = fadd float %3224, %3227, !dbg !35 + %3229 = bitcast float %2750 to i32, !dbg !38 + %3230 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3229, i32 8, i32 31), !dbg !38 + %3231 = bitcast i32 %3230 to float, !dbg !38 + %3232 = fadd float %2750, %3231, !dbg !35 + %3233 = bitcast float %3232 to i32, !dbg !38 + %3234 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3233, i32 4, i32 31), !dbg !38 + %3235 = bitcast i32 %3234 to float, !dbg !38 + %3236 = fadd float %3232, %3235, !dbg !35 + %3237 = bitcast float %3236 to i32, !dbg !38 + %3238 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3237, i32 2, i32 31), !dbg !38 + %3239 = bitcast i32 %3238 to float, !dbg !38 + %3240 = fadd float %3236, %3239, !dbg !35 + %3241 = bitcast float %3240 to i32, !dbg !38 + %3242 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3241, i32 1, i32 31), !dbg !38 + %3243 = bitcast i32 %3242 to float, !dbg !38 + %3244 = fadd float %3240, %3243, !dbg !35 + %3245 = bitcast float %2757 to i32, !dbg !38 + %3246 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3245, i32 8, i32 31), !dbg !38 + %3247 = bitcast i32 %3246 to float, !dbg !38 + %3248 = fadd float %2757, %3247, !dbg !35 + %3249 = bitcast float %3248 to i32, !dbg !38 + %3250 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3249, i32 4, i32 31), !dbg !38 + %3251 = bitcast i32 %3250 to float, !dbg !38 + %3252 = fadd float %3248, %3251, !dbg !35 + %3253 = bitcast float %3252 to i32, !dbg !38 + %3254 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3253, i32 2, i32 31), !dbg !38 + %3255 = bitcast i32 %3254 to float, !dbg !38 + %3256 = fadd float %3252, %3255, !dbg !35 + %3257 = bitcast float %3256 to i32, !dbg !38 + %3258 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3257, i32 1, i32 31), !dbg !38 + %3259 = bitcast i32 %3258 to float, !dbg !38 + %3260 = fadd float %3256, %3259, !dbg !35 + %3261 = bitcast float %2764 to i32, !dbg !38 + %3262 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3261, i32 8, i32 31), !dbg !38 + %3263 = bitcast i32 %3262 to float, !dbg !38 + %3264 = fadd float %2764, %3263, !dbg !35 + %3265 = bitcast float %3264 to i32, !dbg !38 + %3266 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3265, i32 4, i32 31), !dbg !38 + %3267 = bitcast i32 %3266 to float, !dbg !38 + %3268 = fadd float %3264, %3267, !dbg !35 + %3269 = bitcast float %3268 to i32, !dbg !38 + %3270 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3269, i32 2, i32 31), !dbg !38 + %3271 = bitcast i32 %3270 to float, !dbg !38 + %3272 = fadd float %3268, %3271, !dbg !35 + %3273 = bitcast float %3272 to i32, !dbg !38 + %3274 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3273, i32 1, i32 31), !dbg !38 + %3275 = bitcast i32 %3274 to float, !dbg !38 + %3276 = fadd float %3272, %3275, !dbg !35 + %3277 = fmul float %2780, %888, !dbg !41 + %3278 = fmul float %2780, %889, !dbg !41 + %3279 = fmul float %2780, %890, !dbg !41 + %3280 = fmul float %2780, %891, !dbg !41 + %3281 = fmul float %2780, %892, !dbg !41 + %3282 = fmul float %2780, %893, !dbg !41 + %3283 = fmul float %2780, %894, !dbg !41 + %3284 = fmul float %2780, %895, !dbg !41 + %3285 = fmul float %2796, %896, !dbg !41 + %3286 = fmul float %2796, %897, !dbg !41 + %3287 = fmul float %2796, %898, !dbg !41 + %3288 = fmul float %2796, %899, !dbg !41 + %3289 = fmul float %2796, %900, !dbg !41 + %3290 = fmul float %2796, %901, !dbg !41 + %3291 = fmul float %2796, %902, !dbg !41 + %3292 = fmul float %2796, %903, !dbg !41 + %3293 = fmul float %2812, %904, !dbg !41 + %3294 = fmul float %2812, %905, !dbg !41 + %3295 = fmul float %2812, %906, !dbg !41 + %3296 = fmul float %2812, %907, !dbg !41 + %3297 = fmul float %2812, %908, !dbg !41 + %3298 = fmul float %2812, %909, !dbg !41 + %3299 = fmul float %2812, %910, !dbg !41 + %3300 = fmul float %2812, %911, !dbg !41 + %3301 = fmul float %2828, %912, !dbg !41 + %3302 = fmul float %2828, %913, !dbg !41 + %3303 = fmul float %2828, %914, !dbg !41 + %3304 = fmul float %2828, %915, !dbg !41 + %3305 = fmul float %2828, %916, !dbg !41 + %3306 = fmul float %2828, %917, !dbg !41 + %3307 = fmul float %2828, %918, !dbg !41 + %3308 = fmul float %2828, %919, !dbg !41 + %3309 = fmul float %2844, %920, !dbg !41 + %3310 = fmul float %2844, %921, !dbg !41 + %3311 = fmul float %2844, %922, !dbg !41 + %3312 = fmul float %2844, %923, !dbg !41 + %3313 = fmul float %2844, %924, !dbg !41 + %3314 = fmul float %2844, %925, !dbg !41 + %3315 = fmul float %2844, %926, !dbg !41 + %3316 = fmul float %2844, %927, !dbg !41 + %3317 = fmul float %2860, %928, !dbg !41 + %3318 = fmul float %2860, %929, !dbg !41 + %3319 = fmul float %2860, %930, !dbg !41 + %3320 = fmul float %2860, %931, !dbg !41 + %3321 = fmul float %2860, %932, !dbg !41 + %3322 = fmul float %2860, %933, !dbg !41 + %3323 = fmul float %2860, %934, !dbg !41 + %3324 = fmul float %2860, %935, !dbg !41 + %3325 = fmul float %2876, %936, !dbg !41 + %3326 = fmul float %2876, %937, !dbg !41 + %3327 = fmul float %2876, %938, !dbg !41 + %3328 = fmul float %2876, %939, !dbg !41 + %3329 = fmul float %2876, %940, !dbg !41 + %3330 = fmul float %2876, %941, !dbg !41 + %3331 = fmul float %2876, %942, !dbg !41 + %3332 = fmul float %2876, %943, !dbg !41 + %3333 = fmul float %2892, %944, !dbg !41 + %3334 = fmul float %2892, %945, !dbg !41 + %3335 = fmul float %2892, %946, !dbg !41 + %3336 = fmul float %2892, %947, !dbg !41 + %3337 = fmul float %2892, %948, !dbg !41 + %3338 = fmul float %2892, %949, !dbg !41 + %3339 = fmul float %2892, %950, !dbg !41 + %3340 = fmul float %2892, %951, !dbg !41 + %3341 = fmul float %2908, %952, !dbg !41 + %3342 = fmul float %2908, %953, !dbg !41 + %3343 = fmul float %2908, %954, !dbg !41 + %3344 = fmul float %2908, %955, !dbg !41 + %3345 = fmul float %2908, %956, !dbg !41 + %3346 = fmul float %2908, %957, !dbg !41 + %3347 = fmul float %2908, %958, !dbg !41 + %3348 = fmul float %2908, %959, !dbg !41 + %3349 = fmul float %2924, %960, !dbg !41 + %3350 = fmul float %2924, %961, !dbg !41 + %3351 = fmul float %2924, %962, !dbg !41 + %3352 = fmul float %2924, %963, !dbg !41 + %3353 = fmul float %2924, %964, !dbg !41 + %3354 = fmul float %2924, %965, !dbg !41 + %3355 = fmul float %2924, %966, !dbg !41 + %3356 = fmul float %2924, %967, !dbg !41 + %3357 = fmul float %2940, %968, !dbg !41 + %3358 = fmul float %2940, %969, !dbg !41 + %3359 = fmul float %2940, %970, !dbg !41 + %3360 = fmul float %2940, %971, !dbg !41 + %3361 = fmul float %2940, %972, !dbg !41 + %3362 = fmul float %2940, %973, !dbg !41 + %3363 = fmul float %2940, %974, !dbg !41 + %3364 = fmul float %2940, %975, !dbg !41 + %3365 = fmul float %2956, %976, !dbg !41 + %3366 = fmul float %2956, %977, !dbg !41 + %3367 = fmul float %2956, %978, !dbg !41 + %3368 = fmul float %2956, %979, !dbg !41 + %3369 = fmul float %2956, %980, !dbg !41 + %3370 = fmul float %2956, %981, !dbg !41 + %3371 = fmul float %2956, %982, !dbg !41 + %3372 = fmul float %2956, %983, !dbg !41 + %3373 = fmul float %2972, %984, !dbg !41 + %3374 = fmul float %2972, %985, !dbg !41 + %3375 = fmul float %2972, %986, !dbg !41 + %3376 = fmul float %2972, %987, !dbg !41 + %3377 = fmul float %2972, %988, !dbg !41 + %3378 = fmul float %2972, %989, !dbg !41 + %3379 = fmul float %2972, %990, !dbg !41 + %3380 = fmul float %2972, %991, !dbg !41 + %3381 = fmul float %2988, %992, !dbg !41 + %3382 = fmul float %2988, %993, !dbg !41 + %3383 = fmul float %2988, %994, !dbg !41 + %3384 = fmul float %2988, %995, !dbg !41 + %3385 = fmul float %2988, %996, !dbg !41 + %3386 = fmul float %2988, %997, !dbg !41 + %3387 = fmul float %2988, %998, !dbg !41 + %3388 = fmul float %2988, %999, !dbg !41 + %3389 = fmul float %3004, %1000, !dbg !41 + %3390 = fmul float %3004, %1001, !dbg !41 + %3391 = fmul float %3004, %1002, !dbg !41 + %3392 = fmul float %3004, %1003, !dbg !41 + %3393 = fmul float %3004, %1004, !dbg !41 + %3394 = fmul float %3004, %1005, !dbg !41 + %3395 = fmul float %3004, %1006, !dbg !41 + %3396 = fmul float %3004, %1007, !dbg !41 + %3397 = fmul float %3020, %1008, !dbg !41 + %3398 = fmul float %3020, %1009, !dbg !41 + %3399 = fmul float %3020, %1010, !dbg !41 + %3400 = fmul float %3020, %1011, !dbg !41 + %3401 = fmul float %3020, %1012, !dbg !41 + %3402 = fmul float %3020, %1013, !dbg !41 + %3403 = fmul float %3020, %1014, !dbg !41 + %3404 = fmul float %3020, %1015, !dbg !41 + %3405 = fmul float %3036, %1016, !dbg !41 + %3406 = fmul float %3036, %1017, !dbg !41 + %3407 = fmul float %3036, %1018, !dbg !41 + %3408 = fmul float %3036, %1019, !dbg !41 + %3409 = fmul float %3036, %1020, !dbg !41 + %3410 = fmul float %3036, %1021, !dbg !41 + %3411 = fmul float %3036, %1022, !dbg !41 + %3412 = fmul float %3036, %1023, !dbg !41 + %3413 = fmul float %3052, %1024, !dbg !41 + %3414 = fmul float %3052, %1025, !dbg !41 + %3415 = fmul float %3052, %1026, !dbg !41 + %3416 = fmul float %3052, %1027, !dbg !41 + %3417 = fmul float %3052, %1028, !dbg !41 + %3418 = fmul float %3052, %1029, !dbg !41 + %3419 = fmul float %3052, %1030, !dbg !41 + %3420 = fmul float %3052, %1031, !dbg !41 + %3421 = fmul float %3068, %1032, !dbg !41 + %3422 = fmul float %3068, %1033, !dbg !41 + %3423 = fmul float %3068, %1034, !dbg !41 + %3424 = fmul float %3068, %1035, !dbg !41 + %3425 = fmul float %3068, %1036, !dbg !41 + %3426 = fmul float %3068, %1037, !dbg !41 + %3427 = fmul float %3068, %1038, !dbg !41 + %3428 = fmul float %3068, %1039, !dbg !41 + %3429 = fmul float %3084, %1040, !dbg !41 + %3430 = fmul float %3084, %1041, !dbg !41 + %3431 = fmul float %3084, %1042, !dbg !41 + %3432 = fmul float %3084, %1043, !dbg !41 + %3433 = fmul float %3084, %1044, !dbg !41 + %3434 = fmul float %3084, %1045, !dbg !41 + %3435 = fmul float %3084, %1046, !dbg !41 + %3436 = fmul float %3084, %1047, !dbg !41 + %3437 = fmul float %3100, %1048, !dbg !41 + %3438 = fmul float %3100, %1049, !dbg !41 + %3439 = fmul float %3100, %1050, !dbg !41 + %3440 = fmul float %3100, %1051, !dbg !41 + %3441 = fmul float %3100, %1052, !dbg !41 + %3442 = fmul float %3100, %1053, !dbg !41 + %3443 = fmul float %3100, %1054, !dbg !41 + %3444 = fmul float %3100, %1055, !dbg !41 + %3445 = fmul float %3116, %1056, !dbg !41 + %3446 = fmul float %3116, %1057, !dbg !41 + %3447 = fmul float %3116, %1058, !dbg !41 + %3448 = fmul float %3116, %1059, !dbg !41 + %3449 = fmul float %3116, %1060, !dbg !41 + %3450 = fmul float %3116, %1061, !dbg !41 + %3451 = fmul float %3116, %1062, !dbg !41 + %3452 = fmul float %3116, %1063, !dbg !41 + %3453 = fmul float %3132, %1064, !dbg !41 + %3454 = fmul float %3132, %1065, !dbg !41 + %3455 = fmul float %3132, %1066, !dbg !41 + %3456 = fmul float %3132, %1067, !dbg !41 + %3457 = fmul float %3132, %1068, !dbg !41 + %3458 = fmul float %3132, %1069, !dbg !41 + %3459 = fmul float %3132, %1070, !dbg !41 + %3460 = fmul float %3132, %1071, !dbg !41 + %3461 = fmul float %3148, %1072, !dbg !41 + %3462 = fmul float %3148, %1073, !dbg !41 + %3463 = fmul float %3148, %1074, !dbg !41 + %3464 = fmul float %3148, %1075, !dbg !41 + %3465 = fmul float %3148, %1076, !dbg !41 + %3466 = fmul float %3148, %1077, !dbg !41 + %3467 = fmul float %3148, %1078, !dbg !41 + %3468 = fmul float %3148, %1079, !dbg !41 + %3469 = fmul float %3164, %1080, !dbg !41 + %3470 = fmul float %3164, %1081, !dbg !41 + %3471 = fmul float %3164, %1082, !dbg !41 + %3472 = fmul float %3164, %1083, !dbg !41 + %3473 = fmul float %3164, %1084, !dbg !41 + %3474 = fmul float %3164, %1085, !dbg !41 + %3475 = fmul float %3164, %1086, !dbg !41 + %3476 = fmul float %3164, %1087, !dbg !41 + %3477 = fmul float %3180, %1088, !dbg !41 + %3478 = fmul float %3180, %1089, !dbg !41 + %3479 = fmul float %3180, %1090, !dbg !41 + %3480 = fmul float %3180, %1091, !dbg !41 + %3481 = fmul float %3180, %1092, !dbg !41 + %3482 = fmul float %3180, %1093, !dbg !41 + %3483 = fmul float %3180, %1094, !dbg !41 + %3484 = fmul float %3180, %1095, !dbg !41 + %3485 = fmul float %3196, %1096, !dbg !41 + %3486 = fmul float %3196, %1097, !dbg !41 + %3487 = fmul float %3196, %1098, !dbg !41 + %3488 = fmul float %3196, %1099, !dbg !41 + %3489 = fmul float %3196, %1100, !dbg !41 + %3490 = fmul float %3196, %1101, !dbg !41 + %3491 = fmul float %3196, %1102, !dbg !41 + %3492 = fmul float %3196, %1103, !dbg !41 + %3493 = fmul float %3212, %1104, !dbg !41 + %3494 = fmul float %3212, %1105, !dbg !41 + %3495 = fmul float %3212, %1106, !dbg !41 + %3496 = fmul float %3212, %1107, !dbg !41 + %3497 = fmul float %3212, %1108, !dbg !41 + %3498 = fmul float %3212, %1109, !dbg !41 + %3499 = fmul float %3212, %1110, !dbg !41 + %3500 = fmul float %3212, %1111, !dbg !41 + %3501 = fmul float %3228, %1112, !dbg !41 + %3502 = fmul float %3228, %1113, !dbg !41 + %3503 = fmul float %3228, %1114, !dbg !41 + %3504 = fmul float %3228, %1115, !dbg !41 + %3505 = fmul float %3228, %1116, !dbg !41 + %3506 = fmul float %3228, %1117, !dbg !41 + %3507 = fmul float %3228, %1118, !dbg !41 + %3508 = fmul float %3228, %1119, !dbg !41 + %3509 = fmul float %3244, %1120, !dbg !41 + %3510 = fmul float %3244, %1121, !dbg !41 + %3511 = fmul float %3244, %1122, !dbg !41 + %3512 = fmul float %3244, %1123, !dbg !41 + %3513 = fmul float %3244, %1124, !dbg !41 + %3514 = fmul float %3244, %1125, !dbg !41 + %3515 = fmul float %3244, %1126, !dbg !41 + %3516 = fmul float %3244, %1127, !dbg !41 + %3517 = fmul float %3260, %1128, !dbg !41 + %3518 = fmul float %3260, %1129, !dbg !41 + %3519 = fmul float %3260, %1130, !dbg !41 + %3520 = fmul float %3260, %1131, !dbg !41 + %3521 = fmul float %3260, %1132, !dbg !41 + %3522 = fmul float %3260, %1133, !dbg !41 + %3523 = fmul float %3260, %1134, !dbg !41 + %3524 = fmul float %3260, %1135, !dbg !41 + %3525 = fmul float %3276, %1136, !dbg !41 + %3526 = fmul float %3276, %1137, !dbg !41 + %3527 = fmul float %3276, %1138, !dbg !41 + %3528 = fmul float %3276, %1139, !dbg !41 + %3529 = fmul float %3276, %1140, !dbg !41 + %3530 = fmul float %3276, %1141, !dbg !41 + %3531 = fmul float %3276, %1142, !dbg !41 + %3532 = fmul float %3276, %1143, !dbg !41 + %3533 = fmul float %1966, %3277, !dbg !42 + %3534 = fmul float %1966, %3278, !dbg !42 + %3535 = fmul float %1966, %3279, !dbg !42 + %3536 = fmul float %1966, %3280, !dbg !42 + %3537 = fmul float %1966, %3281, !dbg !42 + %3538 = fmul float %1966, %3282, !dbg !42 + %3539 = fmul float %1966, %3283, !dbg !42 + %3540 = fmul float %1966, %3284, !dbg !42 + %3541 = fmul float %1968, %3285, !dbg !42 + %3542 = fmul float %1968, %3286, !dbg !42 + %3543 = fmul float %1968, %3287, !dbg !42 + %3544 = fmul float %1968, %3288, !dbg !42 + %3545 = fmul float %1968, %3289, !dbg !42 + %3546 = fmul float %1968, %3290, !dbg !42 + %3547 = fmul float %1968, %3291, !dbg !42 + %3548 = fmul float %1968, %3292, !dbg !42 + %3549 = fmul float %1970, %3293, !dbg !42 + %3550 = fmul float %1970, %3294, !dbg !42 + %3551 = fmul float %1970, %3295, !dbg !42 + %3552 = fmul float %1970, %3296, !dbg !42 + %3553 = fmul float %1970, %3297, !dbg !42 + %3554 = fmul float %1970, %3298, !dbg !42 + %3555 = fmul float %1970, %3299, !dbg !42 + %3556 = fmul float %1970, %3300, !dbg !42 + %3557 = fmul float %1972, %3301, !dbg !42 + %3558 = fmul float %1972, %3302, !dbg !42 + %3559 = fmul float %1972, %3303, !dbg !42 + %3560 = fmul float %1972, %3304, !dbg !42 + %3561 = fmul float %1972, %3305, !dbg !42 + %3562 = fmul float %1972, %3306, !dbg !42 + %3563 = fmul float %1972, %3307, !dbg !42 + %3564 = fmul float %1972, %3308, !dbg !42 + %3565 = fmul float %1974, %3309, !dbg !42 + %3566 = fmul float %1974, %3310, !dbg !42 + %3567 = fmul float %1974, %3311, !dbg !42 + %3568 = fmul float %1974, %3312, !dbg !42 + %3569 = fmul float %1974, %3313, !dbg !42 + %3570 = fmul float %1974, %3314, !dbg !42 + %3571 = fmul float %1974, %3315, !dbg !42 + %3572 = fmul float %1974, %3316, !dbg !42 + %3573 = fmul float %1976, %3317, !dbg !42 + %3574 = fmul float %1976, %3318, !dbg !42 + %3575 = fmul float %1976, %3319, !dbg !42 + %3576 = fmul float %1976, %3320, !dbg !42 + %3577 = fmul float %1976, %3321, !dbg !42 + %3578 = fmul float %1976, %3322, !dbg !42 + %3579 = fmul float %1976, %3323, !dbg !42 + %3580 = fmul float %1976, %3324, !dbg !42 + %3581 = fmul float %1978, %3325, !dbg !42 + %3582 = fmul float %1978, %3326, !dbg !42 + %3583 = fmul float %1978, %3327, !dbg !42 + %3584 = fmul float %1978, %3328, !dbg !42 + %3585 = fmul float %1978, %3329, !dbg !42 + %3586 = fmul float %1978, %3330, !dbg !42 + %3587 = fmul float %1978, %3331, !dbg !42 + %3588 = fmul float %1978, %3332, !dbg !42 + %3589 = fmul float %1980, %3333, !dbg !42 + %3590 = fmul float %1980, %3334, !dbg !42 + %3591 = fmul float %1980, %3335, !dbg !42 + %3592 = fmul float %1980, %3336, !dbg !42 + %3593 = fmul float %1980, %3337, !dbg !42 + %3594 = fmul float %1980, %3338, !dbg !42 + %3595 = fmul float %1980, %3339, !dbg !42 + %3596 = fmul float %1980, %3340, !dbg !42 + %3597 = fmul float %1982, %3341, !dbg !42 + %3598 = fmul float %1982, %3342, !dbg !42 + %3599 = fmul float %1982, %3343, !dbg !42 + %3600 = fmul float %1982, %3344, !dbg !42 + %3601 = fmul float %1982, %3345, !dbg !42 + %3602 = fmul float %1982, %3346, !dbg !42 + %3603 = fmul float %1982, %3347, !dbg !42 + %3604 = fmul float %1982, %3348, !dbg !42 + %3605 = fmul float %1984, %3349, !dbg !42 + %3606 = fmul float %1984, %3350, !dbg !42 + %3607 = fmul float %1984, %3351, !dbg !42 + %3608 = fmul float %1984, %3352, !dbg !42 + %3609 = fmul float %1984, %3353, !dbg !42 + %3610 = fmul float %1984, %3354, !dbg !42 + %3611 = fmul float %1984, %3355, !dbg !42 + %3612 = fmul float %1984, %3356, !dbg !42 + %3613 = fmul float %1986, %3357, !dbg !42 + %3614 = fmul float %1986, %3358, !dbg !42 + %3615 = fmul float %1986, %3359, !dbg !42 + %3616 = fmul float %1986, %3360, !dbg !42 + %3617 = fmul float %1986, %3361, !dbg !42 + %3618 = fmul float %1986, %3362, !dbg !42 + %3619 = fmul float %1986, %3363, !dbg !42 + %3620 = fmul float %1986, %3364, !dbg !42 + %3621 = fmul float %1988, %3365, !dbg !42 + %3622 = fmul float %1988, %3366, !dbg !42 + %3623 = fmul float %1988, %3367, !dbg !42 + %3624 = fmul float %1988, %3368, !dbg !42 + %3625 = fmul float %1988, %3369, !dbg !42 + %3626 = fmul float %1988, %3370, !dbg !42 + %3627 = fmul float %1988, %3371, !dbg !42 + %3628 = fmul float %1988, %3372, !dbg !42 + %3629 = fmul float %1990, %3373, !dbg !42 + %3630 = fmul float %1990, %3374, !dbg !42 + %3631 = fmul float %1990, %3375, !dbg !42 + %3632 = fmul float %1990, %3376, !dbg !42 + %3633 = fmul float %1990, %3377, !dbg !42 + %3634 = fmul float %1990, %3378, !dbg !42 + %3635 = fmul float %1990, %3379, !dbg !42 + %3636 = fmul float %1990, %3380, !dbg !42 + %3637 = fmul float %1992, %3381, !dbg !42 + %3638 = fmul float %1992, %3382, !dbg !42 + %3639 = fmul float %1992, %3383, !dbg !42 + %3640 = fmul float %1992, %3384, !dbg !42 + %3641 = fmul float %1992, %3385, !dbg !42 + %3642 = fmul float %1992, %3386, !dbg !42 + %3643 = fmul float %1992, %3387, !dbg !42 + %3644 = fmul float %1992, %3388, !dbg !42 + %3645 = fmul float %1994, %3389, !dbg !42 + %3646 = fmul float %1994, %3390, !dbg !42 + %3647 = fmul float %1994, %3391, !dbg !42 + %3648 = fmul float %1994, %3392, !dbg !42 + %3649 = fmul float %1994, %3393, !dbg !42 + %3650 = fmul float %1994, %3394, !dbg !42 + %3651 = fmul float %1994, %3395, !dbg !42 + %3652 = fmul float %1994, %3396, !dbg !42 + %3653 = fmul float %1996, %3397, !dbg !42 + %3654 = fmul float %1996, %3398, !dbg !42 + %3655 = fmul float %1996, %3399, !dbg !42 + %3656 = fmul float %1996, %3400, !dbg !42 + %3657 = fmul float %1996, %3401, !dbg !42 + %3658 = fmul float %1996, %3402, !dbg !42 + %3659 = fmul float %1996, %3403, !dbg !42 + %3660 = fmul float %1996, %3404, !dbg !42 + %3661 = fmul float %1998, %3405, !dbg !42 + %3662 = fmul float %1998, %3406, !dbg !42 + %3663 = fmul float %1998, %3407, !dbg !42 + %3664 = fmul float %1998, %3408, !dbg !42 + %3665 = fmul float %1998, %3409, !dbg !42 + %3666 = fmul float %1998, %3410, !dbg !42 + %3667 = fmul float %1998, %3411, !dbg !42 + %3668 = fmul float %1998, %3412, !dbg !42 + %3669 = fmul float %2000, %3413, !dbg !42 + %3670 = fmul float %2000, %3414, !dbg !42 + %3671 = fmul float %2000, %3415, !dbg !42 + %3672 = fmul float %2000, %3416, !dbg !42 + %3673 = fmul float %2000, %3417, !dbg !42 + %3674 = fmul float %2000, %3418, !dbg !42 + %3675 = fmul float %2000, %3419, !dbg !42 + %3676 = fmul float %2000, %3420, !dbg !42 + %3677 = fmul float %2002, %3421, !dbg !42 + %3678 = fmul float %2002, %3422, !dbg !42 + %3679 = fmul float %2002, %3423, !dbg !42 + %3680 = fmul float %2002, %3424, !dbg !42 + %3681 = fmul float %2002, %3425, !dbg !42 + %3682 = fmul float %2002, %3426, !dbg !42 + %3683 = fmul float %2002, %3427, !dbg !42 + %3684 = fmul float %2002, %3428, !dbg !42 + %3685 = fmul float %2004, %3429, !dbg !42 + %3686 = fmul float %2004, %3430, !dbg !42 + %3687 = fmul float %2004, %3431, !dbg !42 + %3688 = fmul float %2004, %3432, !dbg !42 + %3689 = fmul float %2004, %3433, !dbg !42 + %3690 = fmul float %2004, %3434, !dbg !42 + %3691 = fmul float %2004, %3435, !dbg !42 + %3692 = fmul float %2004, %3436, !dbg !42 + %3693 = fmul float %2006, %3437, !dbg !42 + %3694 = fmul float %2006, %3438, !dbg !42 + %3695 = fmul float %2006, %3439, !dbg !42 + %3696 = fmul float %2006, %3440, !dbg !42 + %3697 = fmul float %2006, %3441, !dbg !42 + %3698 = fmul float %2006, %3442, !dbg !42 + %3699 = fmul float %2006, %3443, !dbg !42 + %3700 = fmul float %2006, %3444, !dbg !42 + %3701 = fmul float %2008, %3445, !dbg !42 + %3702 = fmul float %2008, %3446, !dbg !42 + %3703 = fmul float %2008, %3447, !dbg !42 + %3704 = fmul float %2008, %3448, !dbg !42 + %3705 = fmul float %2008, %3449, !dbg !42 + %3706 = fmul float %2008, %3450, !dbg !42 + %3707 = fmul float %2008, %3451, !dbg !42 + %3708 = fmul float %2008, %3452, !dbg !42 + %3709 = fmul float %2010, %3453, !dbg !42 + %3710 = fmul float %2010, %3454, !dbg !42 + %3711 = fmul float %2010, %3455, !dbg !42 + %3712 = fmul float %2010, %3456, !dbg !42 + %3713 = fmul float %2010, %3457, !dbg !42 + %3714 = fmul float %2010, %3458, !dbg !42 + %3715 = fmul float %2010, %3459, !dbg !42 + %3716 = fmul float %2010, %3460, !dbg !42 + %3717 = fmul float %2012, %3461, !dbg !42 + %3718 = fmul float %2012, %3462, !dbg !42 + %3719 = fmul float %2012, %3463, !dbg !42 + %3720 = fmul float %2012, %3464, !dbg !42 + %3721 = fmul float %2012, %3465, !dbg !42 + %3722 = fmul float %2012, %3466, !dbg !42 + %3723 = fmul float %2012, %3467, !dbg !42 + %3724 = fmul float %2012, %3468, !dbg !42 + %3725 = fmul float %2014, %3469, !dbg !42 + %3726 = fmul float %2014, %3470, !dbg !42 + %3727 = fmul float %2014, %3471, !dbg !42 + %3728 = fmul float %2014, %3472, !dbg !42 + %3729 = fmul float %2014, %3473, !dbg !42 + %3730 = fmul float %2014, %3474, !dbg !42 + %3731 = fmul float %2014, %3475, !dbg !42 + %3732 = fmul float %2014, %3476, !dbg !42 + %3733 = fmul float %2016, %3477, !dbg !42 + %3734 = fmul float %2016, %3478, !dbg !42 + %3735 = fmul float %2016, %3479, !dbg !42 + %3736 = fmul float %2016, %3480, !dbg !42 + %3737 = fmul float %2016, %3481, !dbg !42 + %3738 = fmul float %2016, %3482, !dbg !42 + %3739 = fmul float %2016, %3483, !dbg !42 + %3740 = fmul float %2016, %3484, !dbg !42 + %3741 = fmul float %2018, %3485, !dbg !42 + %3742 = fmul float %2018, %3486, !dbg !42 + %3743 = fmul float %2018, %3487, !dbg !42 + %3744 = fmul float %2018, %3488, !dbg !42 + %3745 = fmul float %2018, %3489, !dbg !42 + %3746 = fmul float %2018, %3490, !dbg !42 + %3747 = fmul float %2018, %3491, !dbg !42 + %3748 = fmul float %2018, %3492, !dbg !42 + %3749 = fmul float %2020, %3493, !dbg !42 + %3750 = fmul float %2020, %3494, !dbg !42 + %3751 = fmul float %2020, %3495, !dbg !42 + %3752 = fmul float %2020, %3496, !dbg !42 + %3753 = fmul float %2020, %3497, !dbg !42 + %3754 = fmul float %2020, %3498, !dbg !42 + %3755 = fmul float %2020, %3499, !dbg !42 + %3756 = fmul float %2020, %3500, !dbg !42 + %3757 = fmul float %2022, %3501, !dbg !42 + %3758 = fmul float %2022, %3502, !dbg !42 + %3759 = fmul float %2022, %3503, !dbg !42 + %3760 = fmul float %2022, %3504, !dbg !42 + %3761 = fmul float %2022, %3505, !dbg !42 + %3762 = fmul float %2022, %3506, !dbg !42 + %3763 = fmul float %2022, %3507, !dbg !42 + %3764 = fmul float %2022, %3508, !dbg !42 + %3765 = fmul float %2024, %3509, !dbg !42 + %3766 = fmul float %2024, %3510, !dbg !42 + %3767 = fmul float %2024, %3511, !dbg !42 + %3768 = fmul float %2024, %3512, !dbg !42 + %3769 = fmul float %2024, %3513, !dbg !42 + %3770 = fmul float %2024, %3514, !dbg !42 + %3771 = fmul float %2024, %3515, !dbg !42 + %3772 = fmul float %2024, %3516, !dbg !42 + %3773 = fmul float %2026, %3517, !dbg !42 + %3774 = fmul float %2026, %3518, !dbg !42 + %3775 = fmul float %2026, %3519, !dbg !42 + %3776 = fmul float %2026, %3520, !dbg !42 + %3777 = fmul float %2026, %3521, !dbg !42 + %3778 = fmul float %2026, %3522, !dbg !42 + %3779 = fmul float %2026, %3523, !dbg !42 + %3780 = fmul float %2026, %3524, !dbg !42 + %3781 = fmul float %2028, %3525, !dbg !42 + %3782 = fmul float %2028, %3526, !dbg !42 + %3783 = fmul float %2028, %3527, !dbg !42 + %3784 = fmul float %2028, %3528, !dbg !42 + %3785 = fmul float %2028, %3529, !dbg !42 + %3786 = fmul float %2028, %3530, !dbg !42 + %3787 = fmul float %2028, %3531, !dbg !42 + %3788 = fmul float %2028, %3532, !dbg !42 + %3789 = fsub float %2029, %3533, !dbg !43 + %3790 = fsub float %2030, %3534, !dbg !43 + %3791 = fsub float %2031, %3535, !dbg !43 + %3792 = fsub float %2032, %3536, !dbg !43 + %3793 = fsub float %2033, %3537, !dbg !43 + %3794 = fsub float %2034, %3538, !dbg !43 + %3795 = fsub float %2035, %3539, !dbg !43 + %3796 = fsub float %2036, %3540, !dbg !43 + %3797 = fsub float %2037, %3541, !dbg !43 + %3798 = fsub float %2038, %3542, !dbg !43 + %3799 = fsub float %2039, %3543, !dbg !43 + %3800 = fsub float %2040, %3544, !dbg !43 + %3801 = fsub float %2041, %3545, !dbg !43 + %3802 = fsub float %2042, %3546, !dbg !43 + %3803 = fsub float %2043, %3547, !dbg !43 + %3804 = fsub float %2044, %3548, !dbg !43 + %3805 = insertelement <2 x float> poison, float %2045, i64 0, !dbg !43 + %3806 = insertelement <2 x float> %3805, float %2046, i64 1, !dbg !43 + %3807 = insertelement <2 x float> poison, float %3549, i64 0, !dbg !43 + %3808 = insertelement <2 x float> %3807, float %3550, i64 1, !dbg !43 + %3809 = fsub <2 x float> %3806, %3808, !dbg !43 + %3810 = insertelement <2 x float> poison, float %2047, i64 0, !dbg !43 + %3811 = insertelement <2 x float> %3810, float %2048, i64 1, !dbg !43 + %3812 = insertelement <2 x float> poison, float %3551, i64 0, !dbg !43 + %3813 = insertelement <2 x float> %3812, float %3552, i64 1, !dbg !43 + %3814 = fsub <2 x float> %3811, %3813, !dbg !43 + %3815 = insertelement <2 x float> poison, float %2049, i64 0, !dbg !43 + %3816 = insertelement <2 x float> %3815, float %2050, i64 1, !dbg !43 + %3817 = insertelement <2 x float> poison, float %3553, i64 0, !dbg !43 + %3818 = insertelement <2 x float> %3817, float %3554, i64 1, !dbg !43 + %3819 = fsub <2 x float> %3816, %3818, !dbg !43 + %3820 = insertelement <2 x float> poison, float %2051, i64 0, !dbg !43 + %3821 = insertelement <2 x float> %3820, float %2052, i64 1, !dbg !43 + %3822 = insertelement <2 x float> poison, float %3555, i64 0, !dbg !43 + %3823 = insertelement <2 x float> %3822, float %3556, i64 1, !dbg !43 + %3824 = fsub <2 x float> %3821, %3823, !dbg !43 + %3825 = insertelement <2 x float> poison, float %2053, i64 0, !dbg !43 + %3826 = insertelement <2 x float> %3825, float %2054, i64 1, !dbg !43 + %3827 = insertelement <2 x float> poison, float %3557, i64 0, !dbg !43 + %3828 = insertelement <2 x float> %3827, float %3558, i64 1, !dbg !43 + %3829 = fsub <2 x float> %3826, %3828, !dbg !43 + %3830 = insertelement <2 x float> poison, float %2055, i64 0, !dbg !43 + %3831 = insertelement <2 x float> %3830, float %2056, i64 1, !dbg !43 + %3832 = insertelement <2 x float> poison, float %3559, i64 0, !dbg !43 + %3833 = insertelement <2 x float> %3832, float %3560, i64 1, !dbg !43 + %3834 = fsub <2 x float> %3831, %3833, !dbg !43 + %3835 = insertelement <2 x float> poison, float %2057, i64 0, !dbg !43 + %3836 = insertelement <2 x float> %3835, float %2058, i64 1, !dbg !43 + %3837 = insertelement <2 x float> poison, float %3561, i64 0, !dbg !43 + %3838 = insertelement <2 x float> %3837, float %3562, i64 1, !dbg !43 + %3839 = fsub <2 x float> %3836, %3838, !dbg !43 + %3840 = insertelement <2 x float> poison, float %2059, i64 0, !dbg !43 + %3841 = insertelement <2 x float> %3840, float %2060, i64 1, !dbg !43 + %3842 = insertelement <2 x float> poison, float %3563, i64 0, !dbg !43 + %3843 = insertelement <2 x float> %3842, float %3564, i64 1, !dbg !43 + %3844 = fsub <2 x float> %3841, %3843, !dbg !43 + %3845 = insertelement <2 x float> poison, float %2061, i64 0, !dbg !43 + %3846 = insertelement <2 x float> %3845, float %2062, i64 1, !dbg !43 + %3847 = insertelement <2 x float> poison, float %3565, i64 0, !dbg !43 + %3848 = insertelement <2 x float> %3847, float %3566, i64 1, !dbg !43 + %3849 = fsub <2 x float> %3846, %3848, !dbg !43 + %3850 = insertelement <2 x float> poison, float %2063, i64 0, !dbg !43 + %3851 = insertelement <2 x float> %3850, float %2064, i64 1, !dbg !43 + %3852 = insertelement <2 x float> poison, float %3567, i64 0, !dbg !43 + %3853 = insertelement <2 x float> %3852, float %3568, i64 1, !dbg !43 + %3854 = fsub <2 x float> %3851, %3853, !dbg !43 + %3855 = insertelement <2 x float> poison, float %2065, i64 0, !dbg !43 + %3856 = insertelement <2 x float> %3855, float %2066, i64 1, !dbg !43 + %3857 = insertelement <2 x float> poison, float %3569, i64 0, !dbg !43 + %3858 = insertelement <2 x float> %3857, float %3570, i64 1, !dbg !43 + %3859 = fsub <2 x float> %3856, %3858, !dbg !43 + %3860 = insertelement <2 x float> poison, float %2067, i64 0, !dbg !43 + %3861 = insertelement <2 x float> %3860, float %2068, i64 1, !dbg !43 + %3862 = insertelement <2 x float> poison, float %3571, i64 0, !dbg !43 + %3863 = insertelement <2 x float> %3862, float %3572, i64 1, !dbg !43 + %3864 = fsub <2 x float> %3861, %3863, !dbg !43 + %3865 = insertelement <2 x float> poison, float %2069, i64 0, !dbg !43 + %3866 = insertelement <2 x float> %3865, float %2070, i64 1, !dbg !43 + %3867 = insertelement <2 x float> poison, float %3573, i64 0, !dbg !43 + %3868 = insertelement <2 x float> %3867, float %3574, i64 1, !dbg !43 + %3869 = fsub <2 x float> %3866, %3868, !dbg !43 + %3870 = insertelement <2 x float> poison, float %2071, i64 0, !dbg !43 + %3871 = insertelement <2 x float> %3870, float %2072, i64 1, !dbg !43 + %3872 = insertelement <2 x float> poison, float %3575, i64 0, !dbg !43 + %3873 = insertelement <2 x float> %3872, float %3576, i64 1, !dbg !43 + %3874 = fsub <2 x float> %3871, %3873, !dbg !43 + %3875 = insertelement <2 x float> poison, float %2073, i64 0, !dbg !43 + %3876 = insertelement <2 x float> %3875, float %2074, i64 1, !dbg !43 + %3877 = insertelement <2 x float> poison, float %3577, i64 0, !dbg !43 + %3878 = insertelement <2 x float> %3877, float %3578, i64 1, !dbg !43 + %3879 = fsub <2 x float> %3876, %3878, !dbg !43 + %3880 = insertelement <2 x float> poison, float %2075, i64 0, !dbg !43 + %3881 = insertelement <2 x float> %3880, float %2076, i64 1, !dbg !43 + %3882 = insertelement <2 x float> poison, float %3579, i64 0, !dbg !43 + %3883 = insertelement <2 x float> %3882, float %3580, i64 1, !dbg !43 + %3884 = fsub <2 x float> %3881, %3883, !dbg !43 + %3885 = insertelement <2 x float> poison, float %2077, i64 0, !dbg !43 + %3886 = insertelement <2 x float> %3885, float %2078, i64 1, !dbg !43 + %3887 = insertelement <2 x float> poison, float %3581, i64 0, !dbg !43 + %3888 = insertelement <2 x float> %3887, float %3582, i64 1, !dbg !43 + %3889 = fsub <2 x float> %3886, %3888, !dbg !43 + %3890 = insertelement <2 x float> poison, float %2079, i64 0, !dbg !43 + %3891 = insertelement <2 x float> %3890, float %2080, i64 1, !dbg !43 + %3892 = insertelement <2 x float> poison, float %3583, i64 0, !dbg !43 + %3893 = insertelement <2 x float> %3892, float %3584, i64 1, !dbg !43 + %3894 = fsub <2 x float> %3891, %3893, !dbg !43 + %3895 = insertelement <2 x float> poison, float %2081, i64 0, !dbg !43 + %3896 = insertelement <2 x float> %3895, float %2082, i64 1, !dbg !43 + %3897 = insertelement <2 x float> poison, float %3585, i64 0, !dbg !43 + %3898 = insertelement <2 x float> %3897, float %3586, i64 1, !dbg !43 + %3899 = fsub <2 x float> %3896, %3898, !dbg !43 + %3900 = insertelement <2 x float> poison, float %2083, i64 0, !dbg !43 + %3901 = insertelement <2 x float> %3900, float %2084, i64 1, !dbg !43 + %3902 = insertelement <2 x float> poison, float %3587, i64 0, !dbg !43 + %3903 = insertelement <2 x float> %3902, float %3588, i64 1, !dbg !43 + %3904 = fsub <2 x float> %3901, %3903, !dbg !43 + %3905 = insertelement <2 x float> poison, float %2085, i64 0, !dbg !43 + %3906 = insertelement <2 x float> %3905, float %2086, i64 1, !dbg !43 + %3907 = insertelement <2 x float> poison, float %3589, i64 0, !dbg !43 + %3908 = insertelement <2 x float> %3907, float %3590, i64 1, !dbg !43 + %3909 = fsub <2 x float> %3906, %3908, !dbg !43 + %3910 = insertelement <2 x float> poison, float %2087, i64 0, !dbg !43 + %3911 = insertelement <2 x float> %3910, float %2088, i64 1, !dbg !43 + %3912 = insertelement <2 x float> poison, float %3591, i64 0, !dbg !43 + %3913 = insertelement <2 x float> %3912, float %3592, i64 1, !dbg !43 + %3914 = fsub <2 x float> %3911, %3913, !dbg !43 + %3915 = insertelement <2 x float> poison, float %2089, i64 0, !dbg !43 + %3916 = insertelement <2 x float> %3915, float %2090, i64 1, !dbg !43 + %3917 = insertelement <2 x float> poison, float %3593, i64 0, !dbg !43 + %3918 = insertelement <2 x float> %3917, float %3594, i64 1, !dbg !43 + %3919 = fsub <2 x float> %3916, %3918, !dbg !43 + %3920 = insertelement <2 x float> poison, float %2091, i64 0, !dbg !43 + %3921 = insertelement <2 x float> %3920, float %2092, i64 1, !dbg !43 + %3922 = insertelement <2 x float> poison, float %3595, i64 0, !dbg !43 + %3923 = insertelement <2 x float> %3922, float %3596, i64 1, !dbg !43 + %3924 = fsub <2 x float> %3921, %3923, !dbg !43 + %3925 = insertelement <2 x float> poison, float %2093, i64 0, !dbg !43 + %3926 = insertelement <2 x float> %3925, float %2094, i64 1, !dbg !43 + %3927 = insertelement <2 x float> poison, float %3597, i64 0, !dbg !43 + %3928 = insertelement <2 x float> %3927, float %3598, i64 1, !dbg !43 + %3929 = fsub <2 x float> %3926, %3928, !dbg !43 + %3930 = insertelement <2 x float> poison, float %2095, i64 0, !dbg !43 + %3931 = insertelement <2 x float> %3930, float %2096, i64 1, !dbg !43 + %3932 = insertelement <2 x float> poison, float %3599, i64 0, !dbg !43 + %3933 = insertelement <2 x float> %3932, float %3600, i64 1, !dbg !43 + %3934 = fsub <2 x float> %3931, %3933, !dbg !43 + %3935 = insertelement <2 x float> poison, float %2097, i64 0, !dbg !43 + %3936 = insertelement <2 x float> %3935, float %2098, i64 1, !dbg !43 + %3937 = insertelement <2 x float> poison, float %3601, i64 0, !dbg !43 + %3938 = insertelement <2 x float> %3937, float %3602, i64 1, !dbg !43 + %3939 = fsub <2 x float> %3936, %3938, !dbg !43 + %3940 = insertelement <2 x float> poison, float %2099, i64 0, !dbg !43 + %3941 = insertelement <2 x float> %3940, float %2100, i64 1, !dbg !43 + %3942 = insertelement <2 x float> poison, float %3603, i64 0, !dbg !43 + %3943 = insertelement <2 x float> %3942, float %3604, i64 1, !dbg !43 + %3944 = fsub <2 x float> %3941, %3943, !dbg !43 + %3945 = insertelement <2 x float> poison, float %2101, i64 0, !dbg !43 + %3946 = insertelement <2 x float> %3945, float %2102, i64 1, !dbg !43 + %3947 = insertelement <2 x float> poison, float %3605, i64 0, !dbg !43 + %3948 = insertelement <2 x float> %3947, float %3606, i64 1, !dbg !43 + %3949 = fsub <2 x float> %3946, %3948, !dbg !43 + %3950 = insertelement <2 x float> poison, float %2103, i64 0, !dbg !43 + %3951 = insertelement <2 x float> %3950, float %2104, i64 1, !dbg !43 + %3952 = insertelement <2 x float> poison, float %3607, i64 0, !dbg !43 + %3953 = insertelement <2 x float> %3952, float %3608, i64 1, !dbg !43 + %3954 = fsub <2 x float> %3951, %3953, !dbg !43 + %3955 = insertelement <2 x float> poison, float %2105, i64 0, !dbg !43 + %3956 = insertelement <2 x float> %3955, float %2106, i64 1, !dbg !43 + %3957 = insertelement <2 x float> poison, float %3609, i64 0, !dbg !43 + %3958 = insertelement <2 x float> %3957, float %3610, i64 1, !dbg !43 + %3959 = fsub <2 x float> %3956, %3958, !dbg !43 + %3960 = insertelement <2 x float> poison, float %2107, i64 0, !dbg !43 + %3961 = insertelement <2 x float> %3960, float %2108, i64 1, !dbg !43 + %3962 = insertelement <2 x float> poison, float %3611, i64 0, !dbg !43 + %3963 = insertelement <2 x float> %3962, float %3612, i64 1, !dbg !43 + %3964 = fsub <2 x float> %3961, %3963, !dbg !43 + %3965 = insertelement <2 x float> poison, float %2109, i64 0, !dbg !43 + %3966 = insertelement <2 x float> %3965, float %2110, i64 1, !dbg !43 + %3967 = insertelement <2 x float> poison, float %3613, i64 0, !dbg !43 + %3968 = insertelement <2 x float> %3967, float %3614, i64 1, !dbg !43 + %3969 = fsub <2 x float> %3966, %3968, !dbg !43 + %3970 = insertelement <2 x float> poison, float %2111, i64 0, !dbg !43 + %3971 = insertelement <2 x float> %3970, float %2112, i64 1, !dbg !43 + %3972 = insertelement <2 x float> poison, float %3615, i64 0, !dbg !43 + %3973 = insertelement <2 x float> %3972, float %3616, i64 1, !dbg !43 + %3974 = fsub <2 x float> %3971, %3973, !dbg !43 + %3975 = insertelement <2 x float> poison, float %2113, i64 0, !dbg !43 + %3976 = insertelement <2 x float> %3975, float %2114, i64 1, !dbg !43 + %3977 = insertelement <2 x float> poison, float %3617, i64 0, !dbg !43 + %3978 = insertelement <2 x float> %3977, float %3618, i64 1, !dbg !43 + %3979 = fsub <2 x float> %3976, %3978, !dbg !43 + %3980 = insertelement <2 x float> poison, float %2115, i64 0, !dbg !43 + %3981 = insertelement <2 x float> %3980, float %2116, i64 1, !dbg !43 + %3982 = insertelement <2 x float> poison, float %3619, i64 0, !dbg !43 + %3983 = insertelement <2 x float> %3982, float %3620, i64 1, !dbg !43 + %3984 = fsub <2 x float> %3981, %3983, !dbg !43 + %3985 = insertelement <2 x float> poison, float %2117, i64 0, !dbg !43 + %3986 = insertelement <2 x float> %3985, float %2118, i64 1, !dbg !43 + %3987 = insertelement <2 x float> poison, float %3621, i64 0, !dbg !43 + %3988 = insertelement <2 x float> %3987, float %3622, i64 1, !dbg !43 + %3989 = fsub <2 x float> %3986, %3988, !dbg !43 + %3990 = insertelement <2 x float> poison, float %2119, i64 0, !dbg !43 + %3991 = insertelement <2 x float> %3990, float %2120, i64 1, !dbg !43 + %3992 = insertelement <2 x float> poison, float %3623, i64 0, !dbg !43 + %3993 = insertelement <2 x float> %3992, float %3624, i64 1, !dbg !43 + %3994 = fsub <2 x float> %3991, %3993, !dbg !43 + %3995 = insertelement <2 x float> poison, float %2121, i64 0, !dbg !43 + %3996 = insertelement <2 x float> %3995, float %2122, i64 1, !dbg !43 + %3997 = insertelement <2 x float> poison, float %3625, i64 0, !dbg !43 + %3998 = insertelement <2 x float> %3997, float %3626, i64 1, !dbg !43 + %3999 = fsub <2 x float> %3996, %3998, !dbg !43 + %4000 = insertelement <2 x float> poison, float %2123, i64 0, !dbg !43 + %4001 = insertelement <2 x float> %4000, float %2124, i64 1, !dbg !43 + %4002 = insertelement <2 x float> poison, float %3627, i64 0, !dbg !43 + %4003 = insertelement <2 x float> %4002, float %3628, i64 1, !dbg !43 + %4004 = fsub <2 x float> %4001, %4003, !dbg !43 + %4005 = insertelement <2 x float> poison, float %2125, i64 0, !dbg !43 + %4006 = insertelement <2 x float> %4005, float %2126, i64 1, !dbg !43 + %4007 = insertelement <2 x float> poison, float %3629, i64 0, !dbg !43 + %4008 = insertelement <2 x float> %4007, float %3630, i64 1, !dbg !43 + %4009 = fsub <2 x float> %4006, %4008, !dbg !43 + %4010 = insertelement <2 x float> poison, float %2127, i64 0, !dbg !43 + %4011 = insertelement <2 x float> %4010, float %2128, i64 1, !dbg !43 + %4012 = insertelement <2 x float> poison, float %3631, i64 0, !dbg !43 + %4013 = insertelement <2 x float> %4012, float %3632, i64 1, !dbg !43 + %4014 = fsub <2 x float> %4011, %4013, !dbg !43 + %4015 = insertelement <2 x float> poison, float %2129, i64 0, !dbg !43 + %4016 = insertelement <2 x float> %4015, float %2130, i64 1, !dbg !43 + %4017 = insertelement <2 x float> poison, float %3633, i64 0, !dbg !43 + %4018 = insertelement <2 x float> %4017, float %3634, i64 1, !dbg !43 + %4019 = fsub <2 x float> %4016, %4018, !dbg !43 + %4020 = insertelement <2 x float> poison, float %2131, i64 0, !dbg !43 + %4021 = insertelement <2 x float> %4020, float %2132, i64 1, !dbg !43 + %4022 = insertelement <2 x float> poison, float %3635, i64 0, !dbg !43 + %4023 = insertelement <2 x float> %4022, float %3636, i64 1, !dbg !43 + %4024 = fsub <2 x float> %4021, %4023, !dbg !43 + %4025 = insertelement <2 x float> poison, float %2133, i64 0, !dbg !43 + %4026 = insertelement <2 x float> %4025, float %2134, i64 1, !dbg !43 + %4027 = insertelement <2 x float> poison, float %3637, i64 0, !dbg !43 + %4028 = insertelement <2 x float> %4027, float %3638, i64 1, !dbg !43 + %4029 = fsub <2 x float> %4026, %4028, !dbg !43 + %4030 = insertelement <2 x float> poison, float %2135, i64 0, !dbg !43 + %4031 = insertelement <2 x float> %4030, float %2136, i64 1, !dbg !43 + %4032 = insertelement <2 x float> poison, float %3639, i64 0, !dbg !43 + %4033 = insertelement <2 x float> %4032, float %3640, i64 1, !dbg !43 + %4034 = fsub <2 x float> %4031, %4033, !dbg !43 + %4035 = insertelement <2 x float> poison, float %2137, i64 0, !dbg !43 + %4036 = insertelement <2 x float> %4035, float %2138, i64 1, !dbg !43 + %4037 = insertelement <2 x float> poison, float %3641, i64 0, !dbg !43 + %4038 = insertelement <2 x float> %4037, float %3642, i64 1, !dbg !43 + %4039 = fsub <2 x float> %4036, %4038, !dbg !43 + %4040 = insertelement <2 x float> poison, float %2139, i64 0, !dbg !43 + %4041 = insertelement <2 x float> %4040, float %2140, i64 1, !dbg !43 + %4042 = insertelement <2 x float> poison, float %3643, i64 0, !dbg !43 + %4043 = insertelement <2 x float> %4042, float %3644, i64 1, !dbg !43 + %4044 = fsub <2 x float> %4041, %4043, !dbg !43 + %4045 = insertelement <2 x float> poison, float %2141, i64 0, !dbg !43 + %4046 = insertelement <2 x float> %4045, float %2142, i64 1, !dbg !43 + %4047 = insertelement <2 x float> poison, float %3645, i64 0, !dbg !43 + %4048 = insertelement <2 x float> %4047, float %3646, i64 1, !dbg !43 + %4049 = fsub <2 x float> %4046, %4048, !dbg !43 + %4050 = insertelement <2 x float> poison, float %2143, i64 0, !dbg !43 + %4051 = insertelement <2 x float> %4050, float %2144, i64 1, !dbg !43 + %4052 = insertelement <2 x float> poison, float %3647, i64 0, !dbg !43 + %4053 = insertelement <2 x float> %4052, float %3648, i64 1, !dbg !43 + %4054 = fsub <2 x float> %4051, %4053, !dbg !43 + %4055 = insertelement <2 x float> poison, float %2145, i64 0, !dbg !43 + %4056 = insertelement <2 x float> %4055, float %2146, i64 1, !dbg !43 + %4057 = insertelement <2 x float> poison, float %3649, i64 0, !dbg !43 + %4058 = insertelement <2 x float> %4057, float %3650, i64 1, !dbg !43 + %4059 = fsub <2 x float> %4056, %4058, !dbg !43 + %4060 = insertelement <2 x float> poison, float %2147, i64 0, !dbg !43 + %4061 = insertelement <2 x float> %4060, float %2148, i64 1, !dbg !43 + %4062 = insertelement <2 x float> poison, float %3651, i64 0, !dbg !43 + %4063 = insertelement <2 x float> %4062, float %3652, i64 1, !dbg !43 + %4064 = fsub <2 x float> %4061, %4063, !dbg !43 + %4065 = insertelement <2 x float> poison, float %2149, i64 0, !dbg !43 + %4066 = insertelement <2 x float> %4065, float %2150, i64 1, !dbg !43 + %4067 = insertelement <2 x float> poison, float %3653, i64 0, !dbg !43 + %4068 = insertelement <2 x float> %4067, float %3654, i64 1, !dbg !43 + %4069 = fsub <2 x float> %4066, %4068, !dbg !43 + %4070 = insertelement <2 x float> poison, float %2151, i64 0, !dbg !43 + %4071 = insertelement <2 x float> %4070, float %2152, i64 1, !dbg !43 + %4072 = insertelement <2 x float> poison, float %3655, i64 0, !dbg !43 + %4073 = insertelement <2 x float> %4072, float %3656, i64 1, !dbg !43 + %4074 = fsub <2 x float> %4071, %4073, !dbg !43 + %4075 = insertelement <2 x float> poison, float %2153, i64 0, !dbg !43 + %4076 = insertelement <2 x float> %4075, float %2154, i64 1, !dbg !43 + %4077 = insertelement <2 x float> poison, float %3657, i64 0, !dbg !43 + %4078 = insertelement <2 x float> %4077, float %3658, i64 1, !dbg !43 + %4079 = fsub <2 x float> %4076, %4078, !dbg !43 + %4080 = insertelement <2 x float> poison, float %2155, i64 0, !dbg !43 + %4081 = insertelement <2 x float> %4080, float %2156, i64 1, !dbg !43 + %4082 = insertelement <2 x float> poison, float %3659, i64 0, !dbg !43 + %4083 = insertelement <2 x float> %4082, float %3660, i64 1, !dbg !43 + %4084 = fsub <2 x float> %4081, %4083, !dbg !43 + %4085 = insertelement <2 x float> poison, float %2157, i64 0, !dbg !43 + %4086 = insertelement <2 x float> %4085, float %2158, i64 1, !dbg !43 + %4087 = insertelement <2 x float> poison, float %3661, i64 0, !dbg !43 + %4088 = insertelement <2 x float> %4087, float %3662, i64 1, !dbg !43 + %4089 = fsub <2 x float> %4086, %4088, !dbg !43 + %4090 = insertelement <2 x float> poison, float %2159, i64 0, !dbg !43 + %4091 = insertelement <2 x float> %4090, float %2160, i64 1, !dbg !43 + %4092 = insertelement <2 x float> poison, float %3663, i64 0, !dbg !43 + %4093 = insertelement <2 x float> %4092, float %3664, i64 1, !dbg !43 + %4094 = fsub <2 x float> %4091, %4093, !dbg !43 + %4095 = insertelement <2 x float> poison, float %2161, i64 0, !dbg !43 + %4096 = insertelement <2 x float> %4095, float %2162, i64 1, !dbg !43 + %4097 = insertelement <2 x float> poison, float %3665, i64 0, !dbg !43 + %4098 = insertelement <2 x float> %4097, float %3666, i64 1, !dbg !43 + %4099 = fsub <2 x float> %4096, %4098, !dbg !43 + %4100 = insertelement <2 x float> poison, float %2163, i64 0, !dbg !43 + %4101 = insertelement <2 x float> %4100, float %2164, i64 1, !dbg !43 + %4102 = insertelement <2 x float> poison, float %3667, i64 0, !dbg !43 + %4103 = insertelement <2 x float> %4102, float %3668, i64 1, !dbg !43 + %4104 = fsub <2 x float> %4101, %4103, !dbg !43 + %4105 = insertelement <2 x float> poison, float %2165, i64 0, !dbg !43 + %4106 = insertelement <2 x float> %4105, float %2166, i64 1, !dbg !43 + %4107 = insertelement <2 x float> poison, float %3669, i64 0, !dbg !43 + %4108 = insertelement <2 x float> %4107, float %3670, i64 1, !dbg !43 + %4109 = fsub <2 x float> %4106, %4108, !dbg !43 + %4110 = insertelement <2 x float> poison, float %2167, i64 0, !dbg !43 + %4111 = insertelement <2 x float> %4110, float %2168, i64 1, !dbg !43 + %4112 = insertelement <2 x float> poison, float %3671, i64 0, !dbg !43 + %4113 = insertelement <2 x float> %4112, float %3672, i64 1, !dbg !43 + %4114 = fsub <2 x float> %4111, %4113, !dbg !43 + %4115 = insertelement <2 x float> poison, float %2169, i64 0, !dbg !43 + %4116 = insertelement <2 x float> %4115, float %2170, i64 1, !dbg !43 + %4117 = insertelement <2 x float> poison, float %3673, i64 0, !dbg !43 + %4118 = insertelement <2 x float> %4117, float %3674, i64 1, !dbg !43 + %4119 = fsub <2 x float> %4116, %4118, !dbg !43 + %4120 = insertelement <2 x float> poison, float %2171, i64 0, !dbg !43 + %4121 = insertelement <2 x float> %4120, float %2172, i64 1, !dbg !43 + %4122 = insertelement <2 x float> poison, float %3675, i64 0, !dbg !43 + %4123 = insertelement <2 x float> %4122, float %3676, i64 1, !dbg !43 + %4124 = fsub <2 x float> %4121, %4123, !dbg !43 + %4125 = insertelement <2 x float> poison, float %2173, i64 0, !dbg !43 + %4126 = insertelement <2 x float> %4125, float %2174, i64 1, !dbg !43 + %4127 = insertelement <2 x float> poison, float %3677, i64 0, !dbg !43 + %4128 = insertelement <2 x float> %4127, float %3678, i64 1, !dbg !43 + %4129 = fsub <2 x float> %4126, %4128, !dbg !43 + %4130 = insertelement <2 x float> poison, float %2175, i64 0, !dbg !43 + %4131 = insertelement <2 x float> %4130, float %2176, i64 1, !dbg !43 + %4132 = insertelement <2 x float> poison, float %3679, i64 0, !dbg !43 + %4133 = insertelement <2 x float> %4132, float %3680, i64 1, !dbg !43 + %4134 = fsub <2 x float> %4131, %4133, !dbg !43 + %4135 = insertelement <2 x float> poison, float %2177, i64 0, !dbg !43 + %4136 = insertelement <2 x float> %4135, float %2178, i64 1, !dbg !43 + %4137 = insertelement <2 x float> poison, float %3681, i64 0, !dbg !43 + %4138 = insertelement <2 x float> %4137, float %3682, i64 1, !dbg !43 + %4139 = fsub <2 x float> %4136, %4138, !dbg !43 + %4140 = insertelement <2 x float> poison, float %2179, i64 0, !dbg !43 + %4141 = insertelement <2 x float> %4140, float %2180, i64 1, !dbg !43 + %4142 = insertelement <2 x float> poison, float %3683, i64 0, !dbg !43 + %4143 = insertelement <2 x float> %4142, float %3684, i64 1, !dbg !43 + %4144 = fsub <2 x float> %4141, %4143, !dbg !43 + %4145 = insertelement <2 x float> poison, float %2181, i64 0, !dbg !43 + %4146 = insertelement <2 x float> %4145, float %2182, i64 1, !dbg !43 + %4147 = insertelement <2 x float> poison, float %3685, i64 0, !dbg !43 + %4148 = insertelement <2 x float> %4147, float %3686, i64 1, !dbg !43 + %4149 = fsub <2 x float> %4146, %4148, !dbg !43 + %4150 = insertelement <2 x float> poison, float %2183, i64 0, !dbg !43 + %4151 = insertelement <2 x float> %4150, float %2184, i64 1, !dbg !43 + %4152 = insertelement <2 x float> poison, float %3687, i64 0, !dbg !43 + %4153 = insertelement <2 x float> %4152, float %3688, i64 1, !dbg !43 + %4154 = fsub <2 x float> %4151, %4153, !dbg !43 + %4155 = insertelement <2 x float> poison, float %2185, i64 0, !dbg !43 + %4156 = insertelement <2 x float> %4155, float %2186, i64 1, !dbg !43 + %4157 = insertelement <2 x float> poison, float %3689, i64 0, !dbg !43 + %4158 = insertelement <2 x float> %4157, float %3690, i64 1, !dbg !43 + %4159 = fsub <2 x float> %4156, %4158, !dbg !43 + %4160 = insertelement <2 x float> poison, float %2187, i64 0, !dbg !43 + %4161 = insertelement <2 x float> %4160, float %2188, i64 1, !dbg !43 + %4162 = insertelement <2 x float> poison, float %3691, i64 0, !dbg !43 + %4163 = insertelement <2 x float> %4162, float %3692, i64 1, !dbg !43 + %4164 = fsub <2 x float> %4161, %4163, !dbg !43 + %4165 = insertelement <2 x float> poison, float %2189, i64 0, !dbg !43 + %4166 = insertelement <2 x float> %4165, float %2190, i64 1, !dbg !43 + %4167 = insertelement <2 x float> poison, float %3693, i64 0, !dbg !43 + %4168 = insertelement <2 x float> %4167, float %3694, i64 1, !dbg !43 + %4169 = fsub <2 x float> %4166, %4168, !dbg !43 + %4170 = insertelement <2 x float> poison, float %2191, i64 0, !dbg !43 + %4171 = insertelement <2 x float> %4170, float %2192, i64 1, !dbg !43 + %4172 = insertelement <2 x float> poison, float %3695, i64 0, !dbg !43 + %4173 = insertelement <2 x float> %4172, float %3696, i64 1, !dbg !43 + %4174 = fsub <2 x float> %4171, %4173, !dbg !43 + %4175 = insertelement <2 x float> poison, float %2193, i64 0, !dbg !43 + %4176 = insertelement <2 x float> %4175, float %2194, i64 1, !dbg !43 + %4177 = insertelement <2 x float> poison, float %3697, i64 0, !dbg !43 + %4178 = insertelement <2 x float> %4177, float %3698, i64 1, !dbg !43 + %4179 = fsub <2 x float> %4176, %4178, !dbg !43 + %4180 = insertelement <2 x float> poison, float %2195, i64 0, !dbg !43 + %4181 = insertelement <2 x float> %4180, float %2196, i64 1, !dbg !43 + %4182 = insertelement <2 x float> poison, float %3699, i64 0, !dbg !43 + %4183 = insertelement <2 x float> %4182, float %3700, i64 1, !dbg !43 + %4184 = fsub <2 x float> %4181, %4183, !dbg !43 + %4185 = insertelement <2 x float> poison, float %2197, i64 0, !dbg !43 + %4186 = insertelement <2 x float> %4185, float %2198, i64 1, !dbg !43 + %4187 = insertelement <2 x float> poison, float %3701, i64 0, !dbg !43 + %4188 = insertelement <2 x float> %4187, float %3702, i64 1, !dbg !43 + %4189 = fsub <2 x float> %4186, %4188, !dbg !43 + %4190 = insertelement <2 x float> poison, float %2199, i64 0, !dbg !43 + %4191 = insertelement <2 x float> %4190, float %2200, i64 1, !dbg !43 + %4192 = insertelement <2 x float> poison, float %3703, i64 0, !dbg !43 + %4193 = insertelement <2 x float> %4192, float %3704, i64 1, !dbg !43 + %4194 = fsub <2 x float> %4191, %4193, !dbg !43 + %4195 = insertelement <2 x float> poison, float %2201, i64 0, !dbg !43 + %4196 = insertelement <2 x float> %4195, float %2202, i64 1, !dbg !43 + %4197 = insertelement <2 x float> poison, float %3705, i64 0, !dbg !43 + %4198 = insertelement <2 x float> %4197, float %3706, i64 1, !dbg !43 + %4199 = fsub <2 x float> %4196, %4198, !dbg !43 + %4200 = insertelement <2 x float> poison, float %2203, i64 0, !dbg !43 + %4201 = insertelement <2 x float> %4200, float %2204, i64 1, !dbg !43 + %4202 = insertelement <2 x float> poison, float %3707, i64 0, !dbg !43 + %4203 = insertelement <2 x float> %4202, float %3708, i64 1, !dbg !43 + %4204 = fsub <2 x float> %4201, %4203, !dbg !43 + %4205 = insertelement <2 x float> poison, float %2205, i64 0, !dbg !43 + %4206 = insertelement <2 x float> %4205, float %2206, i64 1, !dbg !43 + %4207 = insertelement <2 x float> poison, float %3709, i64 0, !dbg !43 + %4208 = insertelement <2 x float> %4207, float %3710, i64 1, !dbg !43 + %4209 = fsub <2 x float> %4206, %4208, !dbg !43 + %4210 = insertelement <2 x float> poison, float %2207, i64 0, !dbg !43 + %4211 = insertelement <2 x float> %4210, float %2208, i64 1, !dbg !43 + %4212 = insertelement <2 x float> poison, float %3711, i64 0, !dbg !43 + %4213 = insertelement <2 x float> %4212, float %3712, i64 1, !dbg !43 + %4214 = fsub <2 x float> %4211, %4213, !dbg !43 + %4215 = insertelement <2 x float> poison, float %2209, i64 0, !dbg !43 + %4216 = insertelement <2 x float> %4215, float %2210, i64 1, !dbg !43 + %4217 = insertelement <2 x float> poison, float %3713, i64 0, !dbg !43 + %4218 = insertelement <2 x float> %4217, float %3714, i64 1, !dbg !43 + %4219 = fsub <2 x float> %4216, %4218, !dbg !43 + %4220 = insertelement <2 x float> poison, float %2211, i64 0, !dbg !43 + %4221 = insertelement <2 x float> %4220, float %2212, i64 1, !dbg !43 + %4222 = insertelement <2 x float> poison, float %3715, i64 0, !dbg !43 + %4223 = insertelement <2 x float> %4222, float %3716, i64 1, !dbg !43 + %4224 = fsub <2 x float> %4221, %4223, !dbg !43 + %4225 = insertelement <2 x float> poison, float %2213, i64 0, !dbg !43 + %4226 = insertelement <2 x float> %4225, float %2214, i64 1, !dbg !43 + %4227 = insertelement <2 x float> poison, float %3717, i64 0, !dbg !43 + %4228 = insertelement <2 x float> %4227, float %3718, i64 1, !dbg !43 + %4229 = fsub <2 x float> %4226, %4228, !dbg !43 + %4230 = insertelement <2 x float> poison, float %2215, i64 0, !dbg !43 + %4231 = insertelement <2 x float> %4230, float %2216, i64 1, !dbg !43 + %4232 = insertelement <2 x float> poison, float %3719, i64 0, !dbg !43 + %4233 = insertelement <2 x float> %4232, float %3720, i64 1, !dbg !43 + %4234 = fsub <2 x float> %4231, %4233, !dbg !43 + %4235 = insertelement <2 x float> poison, float %2217, i64 0, !dbg !43 + %4236 = insertelement <2 x float> %4235, float %2218, i64 1, !dbg !43 + %4237 = insertelement <2 x float> poison, float %3721, i64 0, !dbg !43 + %4238 = insertelement <2 x float> %4237, float %3722, i64 1, !dbg !43 + %4239 = fsub <2 x float> %4236, %4238, !dbg !43 + %4240 = insertelement <2 x float> poison, float %2219, i64 0, !dbg !43 + %4241 = insertelement <2 x float> %4240, float %2220, i64 1, !dbg !43 + %4242 = insertelement <2 x float> poison, float %3723, i64 0, !dbg !43 + %4243 = insertelement <2 x float> %4242, float %3724, i64 1, !dbg !43 + %4244 = fsub <2 x float> %4241, %4243, !dbg !43 + %4245 = insertelement <2 x float> poison, float %2221, i64 0, !dbg !43 + %4246 = insertelement <2 x float> %4245, float %2222, i64 1, !dbg !43 + %4247 = insertelement <2 x float> poison, float %3725, i64 0, !dbg !43 + %4248 = insertelement <2 x float> %4247, float %3726, i64 1, !dbg !43 + %4249 = fsub <2 x float> %4246, %4248, !dbg !43 + %4250 = insertelement <2 x float> poison, float %2223, i64 0, !dbg !43 + %4251 = insertelement <2 x float> %4250, float %2224, i64 1, !dbg !43 + %4252 = insertelement <2 x float> poison, float %3727, i64 0, !dbg !43 + %4253 = insertelement <2 x float> %4252, float %3728, i64 1, !dbg !43 + %4254 = fsub <2 x float> %4251, %4253, !dbg !43 + %4255 = insertelement <2 x float> poison, float %2225, i64 0, !dbg !43 + %4256 = insertelement <2 x float> %4255, float %2226, i64 1, !dbg !43 + %4257 = insertelement <2 x float> poison, float %3729, i64 0, !dbg !43 + %4258 = insertelement <2 x float> %4257, float %3730, i64 1, !dbg !43 + %4259 = fsub <2 x float> %4256, %4258, !dbg !43 + %4260 = insertelement <2 x float> poison, float %2227, i64 0, !dbg !43 + %4261 = insertelement <2 x float> %4260, float %2228, i64 1, !dbg !43 + %4262 = insertelement <2 x float> poison, float %3731, i64 0, !dbg !43 + %4263 = insertelement <2 x float> %4262, float %3732, i64 1, !dbg !43 + %4264 = fsub <2 x float> %4261, %4263, !dbg !43 + %4265 = insertelement <2 x float> poison, float %2229, i64 0, !dbg !43 + %4266 = insertelement <2 x float> %4265, float %2230, i64 1, !dbg !43 + %4267 = insertelement <2 x float> poison, float %3733, i64 0, !dbg !43 + %4268 = insertelement <2 x float> %4267, float %3734, i64 1, !dbg !43 + %4269 = fsub <2 x float> %4266, %4268, !dbg !43 + %4270 = insertelement <2 x float> poison, float %2231, i64 0, !dbg !43 + %4271 = insertelement <2 x float> %4270, float %2232, i64 1, !dbg !43 + %4272 = insertelement <2 x float> poison, float %3735, i64 0, !dbg !43 + %4273 = insertelement <2 x float> %4272, float %3736, i64 1, !dbg !43 + %4274 = fsub <2 x float> %4271, %4273, !dbg !43 + %4275 = insertelement <2 x float> poison, float %2233, i64 0, !dbg !43 + %4276 = insertelement <2 x float> %4275, float %2234, i64 1, !dbg !43 + %4277 = insertelement <2 x float> poison, float %3737, i64 0, !dbg !43 + %4278 = insertelement <2 x float> %4277, float %3738, i64 1, !dbg !43 + %4279 = fsub <2 x float> %4276, %4278, !dbg !43 + %4280 = insertelement <2 x float> poison, float %2235, i64 0, !dbg !43 + %4281 = insertelement <2 x float> %4280, float %2236, i64 1, !dbg !43 + %4282 = insertelement <2 x float> poison, float %3739, i64 0, !dbg !43 + %4283 = insertelement <2 x float> %4282, float %3740, i64 1, !dbg !43 + %4284 = fsub <2 x float> %4281, %4283, !dbg !43 + %4285 = insertelement <2 x float> poison, float %2237, i64 0, !dbg !43 + %4286 = insertelement <2 x float> %4285, float %2238, i64 1, !dbg !43 + %4287 = insertelement <2 x float> poison, float %3741, i64 0, !dbg !43 + %4288 = insertelement <2 x float> %4287, float %3742, i64 1, !dbg !43 + %4289 = fsub <2 x float> %4286, %4288, !dbg !43 + %4290 = insertelement <2 x float> poison, float %2239, i64 0, !dbg !43 + %4291 = insertelement <2 x float> %4290, float %2240, i64 1, !dbg !43 + %4292 = insertelement <2 x float> poison, float %3743, i64 0, !dbg !43 + %4293 = insertelement <2 x float> %4292, float %3744, i64 1, !dbg !43 + %4294 = fsub <2 x float> %4291, %4293, !dbg !43 + %4295 = insertelement <2 x float> poison, float %2241, i64 0, !dbg !43 + %4296 = insertelement <2 x float> %4295, float %2242, i64 1, !dbg !43 + %4297 = insertelement <2 x float> poison, float %3745, i64 0, !dbg !43 + %4298 = insertelement <2 x float> %4297, float %3746, i64 1, !dbg !43 + %4299 = fsub <2 x float> %4296, %4298, !dbg !43 + %4300 = insertelement <2 x float> poison, float %2243, i64 0, !dbg !43 + %4301 = insertelement <2 x float> %4300, float %2244, i64 1, !dbg !43 + %4302 = insertelement <2 x float> poison, float %3747, i64 0, !dbg !43 + %4303 = insertelement <2 x float> %4302, float %3748, i64 1, !dbg !43 + %4304 = fsub <2 x float> %4301, %4303, !dbg !43 + %4305 = insertelement <2 x float> poison, float %2245, i64 0, !dbg !43 + %4306 = insertelement <2 x float> %4305, float %2246, i64 1, !dbg !43 + %4307 = insertelement <2 x float> poison, float %3749, i64 0, !dbg !43 + %4308 = insertelement <2 x float> %4307, float %3750, i64 1, !dbg !43 + %4309 = fsub <2 x float> %4306, %4308, !dbg !43 + %4310 = insertelement <2 x float> poison, float %2247, i64 0, !dbg !43 + %4311 = insertelement <2 x float> %4310, float %2248, i64 1, !dbg !43 + %4312 = insertelement <2 x float> poison, float %3751, i64 0, !dbg !43 + %4313 = insertelement <2 x float> %4312, float %3752, i64 1, !dbg !43 + %4314 = fsub <2 x float> %4311, %4313, !dbg !43 + %4315 = insertelement <2 x float> poison, float %2249, i64 0, !dbg !43 + %4316 = insertelement <2 x float> %4315, float %2250, i64 1, !dbg !43 + %4317 = insertelement <2 x float> poison, float %3753, i64 0, !dbg !43 + %4318 = insertelement <2 x float> %4317, float %3754, i64 1, !dbg !43 + %4319 = fsub <2 x float> %4316, %4318, !dbg !43 + %4320 = insertelement <2 x float> poison, float %2251, i64 0, !dbg !43 + %4321 = insertelement <2 x float> %4320, float %2252, i64 1, !dbg !43 + %4322 = insertelement <2 x float> poison, float %3755, i64 0, !dbg !43 + %4323 = insertelement <2 x float> %4322, float %3756, i64 1, !dbg !43 + %4324 = fsub <2 x float> %4321, %4323, !dbg !43 + %4325 = insertelement <2 x float> poison, float %2253, i64 0, !dbg !43 + %4326 = insertelement <2 x float> %4325, float %2254, i64 1, !dbg !43 + %4327 = insertelement <2 x float> poison, float %3757, i64 0, !dbg !43 + %4328 = insertelement <2 x float> %4327, float %3758, i64 1, !dbg !43 + %4329 = fsub <2 x float> %4326, %4328, !dbg !43 + %4330 = insertelement <2 x float> poison, float %2255, i64 0, !dbg !43 + %4331 = insertelement <2 x float> %4330, float %2256, i64 1, !dbg !43 + %4332 = insertelement <2 x float> poison, float %3759, i64 0, !dbg !43 + %4333 = insertelement <2 x float> %4332, float %3760, i64 1, !dbg !43 + %4334 = fsub <2 x float> %4331, %4333, !dbg !43 + %4335 = insertelement <2 x float> poison, float %2257, i64 0, !dbg !43 + %4336 = insertelement <2 x float> %4335, float %2258, i64 1, !dbg !43 + %4337 = insertelement <2 x float> poison, float %3761, i64 0, !dbg !43 + %4338 = insertelement <2 x float> %4337, float %3762, i64 1, !dbg !43 + %4339 = fsub <2 x float> %4336, %4338, !dbg !43 + %4340 = insertelement <2 x float> poison, float %2259, i64 0, !dbg !43 + %4341 = insertelement <2 x float> %4340, float %2260, i64 1, !dbg !43 + %4342 = insertelement <2 x float> poison, float %3763, i64 0, !dbg !43 + %4343 = insertelement <2 x float> %4342, float %3764, i64 1, !dbg !43 + %4344 = fsub <2 x float> %4341, %4343, !dbg !43 + %4345 = insertelement <2 x float> poison, float %2261, i64 0, !dbg !43 + %4346 = insertelement <2 x float> %4345, float %2262, i64 1, !dbg !43 + %4347 = insertelement <2 x float> poison, float %3765, i64 0, !dbg !43 + %4348 = insertelement <2 x float> %4347, float %3766, i64 1, !dbg !43 + %4349 = fsub <2 x float> %4346, %4348, !dbg !43 + %4350 = insertelement <2 x float> poison, float %2263, i64 0, !dbg !43 + %4351 = insertelement <2 x float> %4350, float %2264, i64 1, !dbg !43 + %4352 = insertelement <2 x float> poison, float %3767, i64 0, !dbg !43 + %4353 = insertelement <2 x float> %4352, float %3768, i64 1, !dbg !43 + %4354 = fsub <2 x float> %4351, %4353, !dbg !43 + %4355 = insertelement <2 x float> poison, float %2265, i64 0, !dbg !43 + %4356 = insertelement <2 x float> %4355, float %2266, i64 1, !dbg !43 + %4357 = insertelement <2 x float> poison, float %3769, i64 0, !dbg !43 + %4358 = insertelement <2 x float> %4357, float %3770, i64 1, !dbg !43 + %4359 = fsub <2 x float> %4356, %4358, !dbg !43 + %4360 = insertelement <2 x float> poison, float %2267, i64 0, !dbg !43 + %4361 = insertelement <2 x float> %4360, float %2268, i64 1, !dbg !43 + %4362 = insertelement <2 x float> poison, float %3771, i64 0, !dbg !43 + %4363 = insertelement <2 x float> %4362, float %3772, i64 1, !dbg !43 + %4364 = fsub <2 x float> %4361, %4363, !dbg !43 + %4365 = insertelement <2 x float> poison, float %2269, i64 0, !dbg !43 + %4366 = insertelement <2 x float> %4365, float %2270, i64 1, !dbg !43 + %4367 = insertelement <2 x float> poison, float %3773, i64 0, !dbg !43 + %4368 = insertelement <2 x float> %4367, float %3774, i64 1, !dbg !43 + %4369 = fsub <2 x float> %4366, %4368, !dbg !43 + %4370 = insertelement <2 x float> poison, float %2271, i64 0, !dbg !43 + %4371 = insertelement <2 x float> %4370, float %2272, i64 1, !dbg !43 + %4372 = insertelement <2 x float> poison, float %3775, i64 0, !dbg !43 + %4373 = insertelement <2 x float> %4372, float %3776, i64 1, !dbg !43 + %4374 = fsub <2 x float> %4371, %4373, !dbg !43 + %4375 = insertelement <2 x float> poison, float %2273, i64 0, !dbg !43 + %4376 = insertelement <2 x float> %4375, float %2274, i64 1, !dbg !43 + %4377 = insertelement <2 x float> poison, float %3777, i64 0, !dbg !43 + %4378 = insertelement <2 x float> %4377, float %3778, i64 1, !dbg !43 + %4379 = fsub <2 x float> %4376, %4378, !dbg !43 + %4380 = insertelement <2 x float> poison, float %2275, i64 0, !dbg !43 + %4381 = insertelement <2 x float> %4380, float %2276, i64 1, !dbg !43 + %4382 = insertelement <2 x float> poison, float %3779, i64 0, !dbg !43 + %4383 = insertelement <2 x float> %4382, float %3780, i64 1, !dbg !43 + %4384 = fsub <2 x float> %4381, %4383, !dbg !43 + %4385 = insertelement <2 x float> poison, float %2277, i64 0, !dbg !43 + %4386 = insertelement <2 x float> %4385, float %2278, i64 1, !dbg !43 + %4387 = insertelement <2 x float> poison, float %3781, i64 0, !dbg !43 + %4388 = insertelement <2 x float> %4387, float %3782, i64 1, !dbg !43 + %4389 = fsub <2 x float> %4386, %4388, !dbg !43 + %4390 = insertelement <2 x float> poison, float %2279, i64 0, !dbg !43 + %4391 = insertelement <2 x float> %4390, float %2280, i64 1, !dbg !43 + %4392 = insertelement <2 x float> poison, float %3783, i64 0, !dbg !43 + %4393 = insertelement <2 x float> %4392, float %3784, i64 1, !dbg !43 + %4394 = fsub <2 x float> %4391, %4393, !dbg !43 + %4395 = insertelement <2 x float> poison, float %2281, i64 0, !dbg !43 + %4396 = insertelement <2 x float> %4395, float %2282, i64 1, !dbg !43 + %4397 = insertelement <2 x float> poison, float %3785, i64 0, !dbg !43 + %4398 = insertelement <2 x float> %4397, float %3786, i64 1, !dbg !43 + %4399 = fsub <2 x float> %4396, %4398, !dbg !43 + %4400 = insertelement <2 x float> poison, float %2283, i64 0, !dbg !43 + %4401 = insertelement <2 x float> %4400, float %2284, i64 1, !dbg !43 + %4402 = insertelement <2 x float> poison, float %3787, i64 0, !dbg !43 + %4403 = insertelement <2 x float> %4402, float %3788, i64 1, !dbg !43 + %4404 = fsub <2 x float> %4401, %4403, !dbg !43 + %4405 = fptrunc float %3789 to bfloat, !dbg !44 + %4406 = fptrunc float %3790 to bfloat, !dbg !44 + %4407 = fptrunc float %3791 to bfloat, !dbg !44 + %4408 = fptrunc float %3792 to bfloat, !dbg !44 + %4409 = fptrunc float %3793 to bfloat, !dbg !44 + %4410 = fptrunc float %3794 to bfloat, !dbg !44 + %4411 = fptrunc float %3795 to bfloat, !dbg !44 + %4412 = fptrunc float %3796 to bfloat, !dbg !44 + %4413 = fptrunc float %3797 to bfloat, !dbg !44 + %4414 = fptrunc float %3798 to bfloat, !dbg !44 + %4415 = fptrunc float %3799 to bfloat, !dbg !44 + %4416 = fptrunc float %3800 to bfloat, !dbg !44 + %4417 = fptrunc float %3801 to bfloat, !dbg !44 + %4418 = fptrunc float %3802 to bfloat, !dbg !44 + %4419 = fptrunc float %3803 to bfloat, !dbg !44 + %4420 = fptrunc float %3804 to bfloat, !dbg !44 + %4421 = fptrunc <2 x float> %3809 to <2 x bfloat>, !dbg !44 + %4422 = fptrunc <2 x float> %3814 to <2 x bfloat>, !dbg !44 + %4423 = fptrunc <2 x float> %3819 to <2 x bfloat>, !dbg !44 + %4424 = fptrunc <2 x float> %3824 to <2 x bfloat>, !dbg !44 + %4425 = fptrunc <2 x float> %3829 to <2 x bfloat>, !dbg !44 + %4426 = fptrunc <2 x float> %3834 to <2 x bfloat>, !dbg !44 + %4427 = fptrunc <2 x float> %3839 to <2 x bfloat>, !dbg !44 + %4428 = fptrunc <2 x float> %3844 to <2 x bfloat>, !dbg !44 + %4429 = fptrunc <2 x float> %3849 to <2 x bfloat>, !dbg !44 + %4430 = fptrunc <2 x float> %3854 to <2 x bfloat>, !dbg !44 + %4431 = fptrunc <2 x float> %3859 to <2 x bfloat>, !dbg !44 + %4432 = fptrunc <2 x float> %3864 to <2 x bfloat>, !dbg !44 + %4433 = fptrunc <2 x float> %3869 to <2 x bfloat>, !dbg !44 + %4434 = fptrunc <2 x float> %3874 to <2 x bfloat>, !dbg !44 + %4435 = fptrunc <2 x float> %3879 to <2 x bfloat>, !dbg !44 + %4436 = fptrunc <2 x float> %3884 to <2 x bfloat>, !dbg !44 + %4437 = fptrunc <2 x float> %3889 to <2 x bfloat>, !dbg !44 + %4438 = fptrunc <2 x float> %3894 to <2 x bfloat>, !dbg !44 + %4439 = fptrunc <2 x float> %3899 to <2 x bfloat>, !dbg !44 + %4440 = fptrunc <2 x float> %3904 to <2 x bfloat>, !dbg !44 + %4441 = fptrunc <2 x float> %3909 to <2 x bfloat>, !dbg !44 + %4442 = fptrunc <2 x float> %3914 to <2 x bfloat>, !dbg !44 + %4443 = fptrunc <2 x float> %3919 to <2 x bfloat>, !dbg !44 + %4444 = fptrunc <2 x float> %3924 to <2 x bfloat>, !dbg !44 + %4445 = fptrunc <2 x float> %3929 to <2 x bfloat>, !dbg !44 + %4446 = fptrunc <2 x float> %3934 to <2 x bfloat>, !dbg !44 + %4447 = fptrunc <2 x float> %3939 to <2 x bfloat>, !dbg !44 + %4448 = fptrunc <2 x float> %3944 to <2 x bfloat>, !dbg !44 + %4449 = fptrunc <2 x float> %3949 to <2 x bfloat>, !dbg !44 + %4450 = fptrunc <2 x float> %3954 to <2 x bfloat>, !dbg !44 + %4451 = fptrunc <2 x float> %3959 to <2 x bfloat>, !dbg !44 + %4452 = fptrunc <2 x float> %3964 to <2 x bfloat>, !dbg !44 + %4453 = fptrunc <2 x float> %3969 to <2 x bfloat>, !dbg !44 + %4454 = fptrunc <2 x float> %3974 to <2 x bfloat>, !dbg !44 + %4455 = fptrunc <2 x float> %3979 to <2 x bfloat>, !dbg !44 + %4456 = fptrunc <2 x float> %3984 to <2 x bfloat>, !dbg !44 + %4457 = fptrunc <2 x float> %3989 to <2 x bfloat>, !dbg !44 + %4458 = fptrunc <2 x float> %3994 to <2 x bfloat>, !dbg !44 + %4459 = fptrunc <2 x float> %3999 to <2 x bfloat>, !dbg !44 + %4460 = fptrunc <2 x float> %4004 to <2 x bfloat>, !dbg !44 + %4461 = fptrunc <2 x float> %4009 to <2 x bfloat>, !dbg !44 + %4462 = fptrunc <2 x float> %4014 to <2 x bfloat>, !dbg !44 + %4463 = fptrunc <2 x float> %4019 to <2 x bfloat>, !dbg !44 + %4464 = fptrunc <2 x float> %4024 to <2 x bfloat>, !dbg !44 + %4465 = fptrunc <2 x float> %4029 to <2 x bfloat>, !dbg !44 + %4466 = fptrunc <2 x float> %4034 to <2 x bfloat>, !dbg !44 + %4467 = fptrunc <2 x float> %4039 to <2 x bfloat>, !dbg !44 + %4468 = fptrunc <2 x float> %4044 to <2 x bfloat>, !dbg !44 + %4469 = fptrunc <2 x float> %4049 to <2 x bfloat>, !dbg !44 + %4470 = fptrunc <2 x float> %4054 to <2 x bfloat>, !dbg !44 + %4471 = fptrunc <2 x float> %4059 to <2 x bfloat>, !dbg !44 + %4472 = fptrunc <2 x float> %4064 to <2 x bfloat>, !dbg !44 + %4473 = fptrunc <2 x float> %4069 to <2 x bfloat>, !dbg !44 + %4474 = fptrunc <2 x float> %4074 to <2 x bfloat>, !dbg !44 + %4475 = fptrunc <2 x float> %4079 to <2 x bfloat>, !dbg !44 + %4476 = fptrunc <2 x float> %4084 to <2 x bfloat>, !dbg !44 + %4477 = fptrunc <2 x float> %4089 to <2 x bfloat>, !dbg !44 + %4478 = fptrunc <2 x float> %4094 to <2 x bfloat>, !dbg !44 + %4479 = fptrunc <2 x float> %4099 to <2 x bfloat>, !dbg !44 + %4480 = fptrunc <2 x float> %4104 to <2 x bfloat>, !dbg !44 + %4481 = fptrunc <2 x float> %4109 to <2 x bfloat>, !dbg !44 + %4482 = fptrunc <2 x float> %4114 to <2 x bfloat>, !dbg !44 + %4483 = fptrunc <2 x float> %4119 to <2 x bfloat>, !dbg !44 + %4484 = fptrunc <2 x float> %4124 to <2 x bfloat>, !dbg !44 + %4485 = fptrunc <2 x float> %4129 to <2 x bfloat>, !dbg !44 + %4486 = fptrunc <2 x float> %4134 to <2 x bfloat>, !dbg !44 + %4487 = fptrunc <2 x float> %4139 to <2 x bfloat>, !dbg !44 + %4488 = fptrunc <2 x float> %4144 to <2 x bfloat>, !dbg !44 + %4489 = fptrunc <2 x float> %4149 to <2 x bfloat>, !dbg !44 + %4490 = fptrunc <2 x float> %4154 to <2 x bfloat>, !dbg !44 + %4491 = fptrunc <2 x float> %4159 to <2 x bfloat>, !dbg !44 + %4492 = fptrunc <2 x float> %4164 to <2 x bfloat>, !dbg !44 + %4493 = fptrunc <2 x float> %4169 to <2 x bfloat>, !dbg !44 + %4494 = fptrunc <2 x float> %4174 to <2 x bfloat>, !dbg !44 + %4495 = fptrunc <2 x float> %4179 to <2 x bfloat>, !dbg !44 + %4496 = fptrunc <2 x float> %4184 to <2 x bfloat>, !dbg !44 + %4497 = fptrunc <2 x float> %4189 to <2 x bfloat>, !dbg !44 + %4498 = fptrunc <2 x float> %4194 to <2 x bfloat>, !dbg !44 + %4499 = fptrunc <2 x float> %4199 to <2 x bfloat>, !dbg !44 + %4500 = fptrunc <2 x float> %4204 to <2 x bfloat>, !dbg !44 + %4501 = fptrunc <2 x float> %4209 to <2 x bfloat>, !dbg !44 + %4502 = fptrunc <2 x float> %4214 to <2 x bfloat>, !dbg !44 + %4503 = fptrunc <2 x float> %4219 to <2 x bfloat>, !dbg !44 + %4504 = fptrunc <2 x float> %4224 to <2 x bfloat>, !dbg !44 + %4505 = fptrunc <2 x float> %4229 to <2 x bfloat>, !dbg !44 + %4506 = fptrunc <2 x float> %4234 to <2 x bfloat>, !dbg !44 + %4507 = fptrunc <2 x float> %4239 to <2 x bfloat>, !dbg !44 + %4508 = fptrunc <2 x float> %4244 to <2 x bfloat>, !dbg !44 + %4509 = fptrunc <2 x float> %4249 to <2 x bfloat>, !dbg !44 + %4510 = fptrunc <2 x float> %4254 to <2 x bfloat>, !dbg !44 + %4511 = fptrunc <2 x float> %4259 to <2 x bfloat>, !dbg !44 + %4512 = fptrunc <2 x float> %4264 to <2 x bfloat>, !dbg !44 + %4513 = fptrunc <2 x float> %4269 to <2 x bfloat>, !dbg !44 + %4514 = fptrunc <2 x float> %4274 to <2 x bfloat>, !dbg !44 + %4515 = fptrunc <2 x float> %4279 to <2 x bfloat>, !dbg !44 + %4516 = fptrunc <2 x float> %4284 to <2 x bfloat>, !dbg !44 + %4517 = fptrunc <2 x float> %4289 to <2 x bfloat>, !dbg !44 + %4518 = fptrunc <2 x float> %4294 to <2 x bfloat>, !dbg !44 + %4519 = fptrunc <2 x float> %4299 to <2 x bfloat>, !dbg !44 + %4520 = fptrunc <2 x float> %4304 to <2 x bfloat>, !dbg !44 + %4521 = fptrunc <2 x float> %4309 to <2 x bfloat>, !dbg !44 + %4522 = fptrunc <2 x float> %4314 to <2 x bfloat>, !dbg !44 + %4523 = fptrunc <2 x float> %4319 to <2 x bfloat>, !dbg !44 + %4524 = fptrunc <2 x float> %4324 to <2 x bfloat>, !dbg !44 + %4525 = fptrunc <2 x float> %4329 to <2 x bfloat>, !dbg !44 + %4526 = fptrunc <2 x float> %4334 to <2 x bfloat>, !dbg !44 + %4527 = fptrunc <2 x float> %4339 to <2 x bfloat>, !dbg !44 + %4528 = fptrunc <2 x float> %4344 to <2 x bfloat>, !dbg !44 + %4529 = fptrunc <2 x float> %4349 to <2 x bfloat>, !dbg !44 + %4530 = fptrunc <2 x float> %4354 to <2 x bfloat>, !dbg !44 + %4531 = fptrunc <2 x float> %4359 to <2 x bfloat>, !dbg !44 + %4532 = fptrunc <2 x float> %4364 to <2 x bfloat>, !dbg !44 + %4533 = fptrunc <2 x float> %4369 to <2 x bfloat>, !dbg !44 + %4534 = fptrunc <2 x float> %4374 to <2 x bfloat>, !dbg !44 + %4535 = fptrunc <2 x float> %4379 to <2 x bfloat>, !dbg !44 + %4536 = fptrunc <2 x float> %4384 to <2 x bfloat>, !dbg !44 + %4537 = fptrunc <2 x float> %4389 to <2 x bfloat>, !dbg !44 + %4538 = fptrunc <2 x float> %4394 to <2 x bfloat>, !dbg !44 + %4539 = fptrunc <2 x float> %4399 to <2 x bfloat>, !dbg !44 + %4540 = fptrunc <2 x float> %4404 to <2 x bfloat>, !dbg !44 + %4541 = insertelement <2 x bfloat> poison, bfloat %4405, i64 0, !dbg !45 + %4542 = insertelement <2 x bfloat> %4541, bfloat %4406, i64 1, !dbg !45 + %4543 = bitcast <2 x bfloat> %4542 to i32, !dbg !45 + %4544 = insertelement <2 x bfloat> poison, bfloat %4407, i64 0, !dbg !45 + %4545 = insertelement <2 x bfloat> %4544, bfloat %4408, i64 1, !dbg !45 + %4546 = bitcast <2 x bfloat> %4545 to i32, !dbg !45 + %4547 = insertelement <2 x bfloat> poison, bfloat %4409, i64 0, !dbg !45 + %4548 = insertelement <2 x bfloat> %4547, bfloat %4410, i64 1, !dbg !45 + %4549 = bitcast <2 x bfloat> %4548 to i32, !dbg !45 + %4550 = insertelement <2 x bfloat> poison, bfloat %4411, i64 0, !dbg !45 + %4551 = insertelement <2 x bfloat> %4550, bfloat %4412, i64 1, !dbg !45 + %4552 = bitcast <2 x bfloat> %4551 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4543, i32 %4546, i32 %4549, i32 %4552, ptr addrspace(1) %282, i1 %22) #4, !dbg !45 + %4553 = insertelement <2 x bfloat> poison, bfloat %4413, i64 0, !dbg !45 + %4554 = insertelement <2 x bfloat> %4553, bfloat %4414, i64 1, !dbg !45 + %4555 = bitcast <2 x bfloat> %4554 to i32, !dbg !45 + %4556 = insertelement <2 x bfloat> poison, bfloat %4415, i64 0, !dbg !45 + %4557 = insertelement <2 x bfloat> %4556, bfloat %4416, i64 1, !dbg !45 + %4558 = bitcast <2 x bfloat> %4557 to i32, !dbg !45 + %4559 = insertelement <2 x bfloat> poison, bfloat %4417, i64 0, !dbg !45 + %4560 = insertelement <2 x bfloat> %4559, bfloat %4418, i64 1, !dbg !45 + %4561 = bitcast <2 x bfloat> %4560 to i32, !dbg !45 + %4562 = insertelement <2 x bfloat> poison, bfloat %4419, i64 0, !dbg !45 + %4563 = insertelement <2 x bfloat> %4562, bfloat %4420, i64 1, !dbg !45 + %4564 = bitcast <2 x bfloat> %4563 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4555, i32 %4558, i32 %4561, i32 %4564, ptr addrspace(1) %283, i1 %23) #4, !dbg !45 + %4565 = bitcast <2 x bfloat> %4421 to i32, !dbg !45 + %4566 = bitcast <2 x bfloat> %4422 to i32, !dbg !45 + %4567 = bitcast <2 x bfloat> %4423 to i32, !dbg !45 + %4568 = bitcast <2 x bfloat> %4424 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4565, i32 %4566, i32 %4567, i32 %4568, ptr addrspace(1) %284, i1 %348) #4, !dbg !45 + %4569 = bitcast <2 x bfloat> %4425 to i32, !dbg !45 + %4570 = bitcast <2 x bfloat> %4426 to i32, !dbg !45 + %4571 = bitcast <2 x bfloat> %4427 to i32, !dbg !45 + %4572 = bitcast <2 x bfloat> %4428 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4569, i32 %4570, i32 %4571, i32 %4572, ptr addrspace(1) %285, i1 %366) #4, !dbg !45 + %4573 = bitcast <2 x bfloat> %4429 to i32, !dbg !45 + %4574 = bitcast <2 x bfloat> %4430 to i32, !dbg !45 + %4575 = bitcast <2 x bfloat> %4431 to i32, !dbg !45 + %4576 = bitcast <2 x bfloat> %4432 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4573, i32 %4574, i32 %4575, i32 %4576, ptr addrspace(1) %286, i1 %384) #4, !dbg !45 + %4577 = bitcast <2 x bfloat> %4433 to i32, !dbg !45 + %4578 = bitcast <2 x bfloat> %4434 to i32, !dbg !45 + %4579 = bitcast <2 x bfloat> %4435 to i32, !dbg !45 + %4580 = bitcast <2 x bfloat> %4436 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4577, i32 %4578, i32 %4579, i32 %4580, ptr addrspace(1) %287, i1 %402) #4, !dbg !45 + %4581 = bitcast <2 x bfloat> %4437 to i32, !dbg !45 + %4582 = bitcast <2 x bfloat> %4438 to i32, !dbg !45 + %4583 = bitcast <2 x bfloat> %4439 to i32, !dbg !45 + %4584 = bitcast <2 x bfloat> %4440 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4581, i32 %4582, i32 %4583, i32 %4584, ptr addrspace(1) %288, i1 %420) #4, !dbg !45 + %4585 = bitcast <2 x bfloat> %4441 to i32, !dbg !45 + %4586 = bitcast <2 x bfloat> %4442 to i32, !dbg !45 + %4587 = bitcast <2 x bfloat> %4443 to i32, !dbg !45 + %4588 = bitcast <2 x bfloat> %4444 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4585, i32 %4586, i32 %4587, i32 %4588, ptr addrspace(1) %289, i1 %438) #4, !dbg !45 + %4589 = bitcast <2 x bfloat> %4445 to i32, !dbg !45 + %4590 = bitcast <2 x bfloat> %4446 to i32, !dbg !45 + %4591 = bitcast <2 x bfloat> %4447 to i32, !dbg !45 + %4592 = bitcast <2 x bfloat> %4448 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4589, i32 %4590, i32 %4591, i32 %4592, ptr addrspace(1) %290, i1 %456) #4, !dbg !45 + %4593 = bitcast <2 x bfloat> %4449 to i32, !dbg !45 + %4594 = bitcast <2 x bfloat> %4450 to i32, !dbg !45 + %4595 = bitcast <2 x bfloat> %4451 to i32, !dbg !45 + %4596 = bitcast <2 x bfloat> %4452 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4593, i32 %4594, i32 %4595, i32 %4596, ptr addrspace(1) %291, i1 %474) #4, !dbg !45 + %4597 = bitcast <2 x bfloat> %4453 to i32, !dbg !45 + %4598 = bitcast <2 x bfloat> %4454 to i32, !dbg !45 + %4599 = bitcast <2 x bfloat> %4455 to i32, !dbg !45 + %4600 = bitcast <2 x bfloat> %4456 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4597, i32 %4598, i32 %4599, i32 %4600, ptr addrspace(1) %292, i1 %492) #4, !dbg !45 + %4601 = bitcast <2 x bfloat> %4457 to i32, !dbg !45 + %4602 = bitcast <2 x bfloat> %4458 to i32, !dbg !45 + %4603 = bitcast <2 x bfloat> %4459 to i32, !dbg !45 + %4604 = bitcast <2 x bfloat> %4460 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4601, i32 %4602, i32 %4603, i32 %4604, ptr addrspace(1) %293, i1 %510) #4, !dbg !45 + %4605 = bitcast <2 x bfloat> %4461 to i32, !dbg !45 + %4606 = bitcast <2 x bfloat> %4462 to i32, !dbg !45 + %4607 = bitcast <2 x bfloat> %4463 to i32, !dbg !45 + %4608 = bitcast <2 x bfloat> %4464 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4605, i32 %4606, i32 %4607, i32 %4608, ptr addrspace(1) %294, i1 %528) #4, !dbg !45 + %4609 = bitcast <2 x bfloat> %4465 to i32, !dbg !45 + %4610 = bitcast <2 x bfloat> %4466 to i32, !dbg !45 + %4611 = bitcast <2 x bfloat> %4467 to i32, !dbg !45 + %4612 = bitcast <2 x bfloat> %4468 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4609, i32 %4610, i32 %4611, i32 %4612, ptr addrspace(1) %295, i1 %546) #4, !dbg !45 + %4613 = bitcast <2 x bfloat> %4469 to i32, !dbg !45 + %4614 = bitcast <2 x bfloat> %4470 to i32, !dbg !45 + %4615 = bitcast <2 x bfloat> %4471 to i32, !dbg !45 + %4616 = bitcast <2 x bfloat> %4472 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4613, i32 %4614, i32 %4615, i32 %4616, ptr addrspace(1) %296, i1 %564) #4, !dbg !45 + %4617 = bitcast <2 x bfloat> %4473 to i32, !dbg !45 + %4618 = bitcast <2 x bfloat> %4474 to i32, !dbg !45 + %4619 = bitcast <2 x bfloat> %4475 to i32, !dbg !45 + %4620 = bitcast <2 x bfloat> %4476 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4617, i32 %4618, i32 %4619, i32 %4620, ptr addrspace(1) %297, i1 %582) #4, !dbg !45 + %4621 = bitcast <2 x bfloat> %4477 to i32, !dbg !45 + %4622 = bitcast <2 x bfloat> %4478 to i32, !dbg !45 + %4623 = bitcast <2 x bfloat> %4479 to i32, !dbg !45 + %4624 = bitcast <2 x bfloat> %4480 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4621, i32 %4622, i32 %4623, i32 %4624, ptr addrspace(1) %298, i1 %600) #4, !dbg !45 + %4625 = bitcast <2 x bfloat> %4481 to i32, !dbg !45 + %4626 = bitcast <2 x bfloat> %4482 to i32, !dbg !45 + %4627 = bitcast <2 x bfloat> %4483 to i32, !dbg !45 + %4628 = bitcast <2 x bfloat> %4484 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4625, i32 %4626, i32 %4627, i32 %4628, ptr addrspace(1) %299, i1 %618) #4, !dbg !45 + %4629 = bitcast <2 x bfloat> %4485 to i32, !dbg !45 + %4630 = bitcast <2 x bfloat> %4486 to i32, !dbg !45 + %4631 = bitcast <2 x bfloat> %4487 to i32, !dbg !45 + %4632 = bitcast <2 x bfloat> %4488 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4629, i32 %4630, i32 %4631, i32 %4632, ptr addrspace(1) %300, i1 %636) #4, !dbg !45 + %4633 = bitcast <2 x bfloat> %4489 to i32, !dbg !45 + %4634 = bitcast <2 x bfloat> %4490 to i32, !dbg !45 + %4635 = bitcast <2 x bfloat> %4491 to i32, !dbg !45 + %4636 = bitcast <2 x bfloat> %4492 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4633, i32 %4634, i32 %4635, i32 %4636, ptr addrspace(1) %301, i1 %654) #4, !dbg !45 + %4637 = bitcast <2 x bfloat> %4493 to i32, !dbg !45 + %4638 = bitcast <2 x bfloat> %4494 to i32, !dbg !45 + %4639 = bitcast <2 x bfloat> %4495 to i32, !dbg !45 + %4640 = bitcast <2 x bfloat> %4496 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4637, i32 %4638, i32 %4639, i32 %4640, ptr addrspace(1) %302, i1 %672) #4, !dbg !45 + %4641 = bitcast <2 x bfloat> %4497 to i32, !dbg !45 + %4642 = bitcast <2 x bfloat> %4498 to i32, !dbg !45 + %4643 = bitcast <2 x bfloat> %4499 to i32, !dbg !45 + %4644 = bitcast <2 x bfloat> %4500 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4641, i32 %4642, i32 %4643, i32 %4644, ptr addrspace(1) %303, i1 %690) #4, !dbg !45 + %4645 = bitcast <2 x bfloat> %4501 to i32, !dbg !45 + %4646 = bitcast <2 x bfloat> %4502 to i32, !dbg !45 + %4647 = bitcast <2 x bfloat> %4503 to i32, !dbg !45 + %4648 = bitcast <2 x bfloat> %4504 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4645, i32 %4646, i32 %4647, i32 %4648, ptr addrspace(1) %304, i1 %708) #4, !dbg !45 + %4649 = bitcast <2 x bfloat> %4505 to i32, !dbg !45 + %4650 = bitcast <2 x bfloat> %4506 to i32, !dbg !45 + %4651 = bitcast <2 x bfloat> %4507 to i32, !dbg !45 + %4652 = bitcast <2 x bfloat> %4508 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4649, i32 %4650, i32 %4651, i32 %4652, ptr addrspace(1) %305, i1 %726) #4, !dbg !45 + %4653 = bitcast <2 x bfloat> %4509 to i32, !dbg !45 + %4654 = bitcast <2 x bfloat> %4510 to i32, !dbg !45 + %4655 = bitcast <2 x bfloat> %4511 to i32, !dbg !45 + %4656 = bitcast <2 x bfloat> %4512 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4653, i32 %4654, i32 %4655, i32 %4656, ptr addrspace(1) %306, i1 %744) #4, !dbg !45 + %4657 = bitcast <2 x bfloat> %4513 to i32, !dbg !45 + %4658 = bitcast <2 x bfloat> %4514 to i32, !dbg !45 + %4659 = bitcast <2 x bfloat> %4515 to i32, !dbg !45 + %4660 = bitcast <2 x bfloat> %4516 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4657, i32 %4658, i32 %4659, i32 %4660, ptr addrspace(1) %307, i1 %762) #4, !dbg !45 + %4661 = bitcast <2 x bfloat> %4517 to i32, !dbg !45 + %4662 = bitcast <2 x bfloat> %4518 to i32, !dbg !45 + %4663 = bitcast <2 x bfloat> %4519 to i32, !dbg !45 + %4664 = bitcast <2 x bfloat> %4520 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4661, i32 %4662, i32 %4663, i32 %4664, ptr addrspace(1) %308, i1 %780) #4, !dbg !45 + %4665 = bitcast <2 x bfloat> %4521 to i32, !dbg !45 + %4666 = bitcast <2 x bfloat> %4522 to i32, !dbg !45 + %4667 = bitcast <2 x bfloat> %4523 to i32, !dbg !45 + %4668 = bitcast <2 x bfloat> %4524 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4665, i32 %4666, i32 %4667, i32 %4668, ptr addrspace(1) %309, i1 %798) #4, !dbg !45 + %4669 = bitcast <2 x bfloat> %4525 to i32, !dbg !45 + %4670 = bitcast <2 x bfloat> %4526 to i32, !dbg !45 + %4671 = bitcast <2 x bfloat> %4527 to i32, !dbg !45 + %4672 = bitcast <2 x bfloat> %4528 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4669, i32 %4670, i32 %4671, i32 %4672, ptr addrspace(1) %310, i1 %816) #4, !dbg !45 + %4673 = bitcast <2 x bfloat> %4529 to i32, !dbg !45 + %4674 = bitcast <2 x bfloat> %4530 to i32, !dbg !45 + %4675 = bitcast <2 x bfloat> %4531 to i32, !dbg !45 + %4676 = bitcast <2 x bfloat> %4532 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4673, i32 %4674, i32 %4675, i32 %4676, ptr addrspace(1) %311, i1 %834) #4, !dbg !45 + %4677 = bitcast <2 x bfloat> %4533 to i32, !dbg !45 + %4678 = bitcast <2 x bfloat> %4534 to i32, !dbg !45 + %4679 = bitcast <2 x bfloat> %4535 to i32, !dbg !45 + %4680 = bitcast <2 x bfloat> %4536 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4677, i32 %4678, i32 %4679, i32 %4680, ptr addrspace(1) %312, i1 %852) #4, !dbg !45 + %4681 = bitcast <2 x bfloat> %4537 to i32, !dbg !45 + %4682 = bitcast <2 x bfloat> %4538 to i32, !dbg !45 + %4683 = bitcast <2 x bfloat> %4539 to i32, !dbg !45 + %4684 = bitcast <2 x bfloat> %4540 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %4681, i32 %4682, i32 %4683, i32 %4684, ptr addrspace(1) %313, i1 %870) #4, !dbg !45 + ret void, !dbg !46 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #2 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="64" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind } +attributes #3 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #4 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = distinct !DISubprogram(name: "l2norm_bwd_kernel", linkageName: "l2norm_bwd_kernel", scope: !1, file: !1, line: 93, type: !5, scopeLine: 93, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!5 = !DISubroutineType(cc: DW_CC_normal, types: !6) +!6 = !{null, !7, !9, !7, !7, !10, !11, !7, !7} +!7 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !8, size: 64, dwarfAddressSpace: 1) +!8 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!9 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !10, size: 64, dwarfAddressSpace: 1) +!10 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!11 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!12 = !DILocation(line: 94, column: 24, scope: !4) +!13 = !DILocation(line: 94, column: 30, scope: !4) +!14 = !DILocation(line: 95, column: 16, scope: !4) +!15 = !DILocation(line: 95, column: 34, scope: !4) +!16 = !DILocation(line: 95, column: 21, scope: !4) +!17 = !DILocation(line: 97, column: 16, scope: !4) +!18 = !DILocation(line: 98, column: 30, scope: !4) +!19 = !DILocation(line: 99, column: 29, scope: !4) +!20 = !DILocation(line: 99, column: 14, scope: !4) +!21 = !DILocation(line: 99, column: 33, scope: !4) +!22 = !DILocation(line: 100, column: 20, scope: !4) +!23 = !DILocation(line: 101, column: 16, scope: !4) +!24 = !DILocation(line: 101, column: 35, scope: !4) +!25 = !DILocation(line: 102, column: 16, scope: !4) +!26 = !DILocation(line: 102, column: 35, scope: !4) +!27 = !DILocation(line: 104, column: 18, scope: !4) +!28 = !DILocation(line: 104, column: 47, scope: !4) +!29 = !DILocation(line: 105, column: 21, scope: !4) +!30 = !DILocation(line: 106, column: 19, scope: !4) +!31 = !DILocation(line: 106, column: 49, scope: !4) +!32 = !DILocation(line: 107, column: 25, scope: !4) +!33 = !DILocation(line: 107, column: 18, scope: !4) +!34 = !DILocation(line: 107, column: 50, scope: !4) +!35 = !DILocation(line: 263, column: 15, scope: !36, inlinedAt: !38) +!36 = distinct !DILexicalBlockFile(scope: !4, file: !37, discriminator: 0) +!37 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!38 = !DILocation(line: 293, column: 36, scope: !36, inlinedAt: !39) +!39 = !DILocation(line: 107, column: 55, scope: !40) +!40 = distinct !DILexicalBlockFile(scope: !4, file: !1, discriminator: 0) +!41 = !DILocation(line: 107, column: 69, scope: !4) +!42 = !DILocation(line: 107, column: 75, scope: !4) +!43 = !DILocation(line: 107, column: 36, scope: !4) +!44 = !DILocation(line: 108, column: 27, scope: !4) +!45 = !DILocation(line: 108, column: 19, scope: !4) +!46 = !DILocation(line: 108, column: 4, scope: !4) diff --git a/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.ptx b/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..11d855bd123cb50f0f26a193fefd6b54cd056c8a --- /dev/null +++ b/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.ptx @@ -0,0 +1,4535 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_bwd_kernel // -- Begin function l2norm_bwd_kernel +.extern .shared .align 16 .b8 global_smem[]; + // @l2norm_bwd_kernel +.visible .entry l2norm_bwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_2, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_3, + .param .f32 l2norm_bwd_kernel_param_4, + .param .u32 l2norm_bwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_6, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_7 +) +.reqntid 64 +{ + .reg .pred %p<50>; + .reg .b16 %rs<566>; + .reg .b32 %r<2556>; + .reg .b64 %rd<664>; + .loc 1 93 0 // l2norm.py:93:0 +$L__func_begin0: + .loc 1 93 0 // l2norm.py:93:0 + +// %bb.0: + ld.param.b64 %rd99, [l2norm_bwd_kernel_param_0]; + ld.param.b64 %rd100, [l2norm_bwd_kernel_param_1]; +$L__tmp0: + .loc 1 94 24 // l2norm.py:94:24 + mov.u32 %r390, %ctaid.x; + ld.param.b64 %rd101, [l2norm_bwd_kernel_param_2]; + .loc 1 95 16 // l2norm.py:95:16 + mul.wide.u32 %rd102, %r390, 128; + ld.param.b64 %rd103, [l2norm_bwd_kernel_param_3]; + .loc 1 95 34 // l2norm.py:95:34 + mov.u32 %r391, %tid.x; + shr.u32 %r392, %r391, 4; + ld.param.s32 %rd104, [l2norm_bwd_kernel_param_5]; + bfe.u32 %r393, %r391, 4, 2; + or.b32 %r394, %r393, 4; + shl.b32 %r395, %r391, 1; + .loc 1 95 21 // l2norm.py:95:21 + cvt.u64.u32 %rd105, %r393; + cvt.u64.u32 %rd106, %r394; + or.b64 %rd107, %rd102, %rd105; + or.b64 %rd108, %rd102, %rd106; + .loc 1 97 16 // l2norm.py:97:16 + setp.lt.s64 %p1, %rd107, %rd104; + setp.lt.s64 %p2, %rd108, %rd104; + .loc 1 95 34 // l2norm.py:95:34 + or.b32 %r396, %r393, 120; + or.b32 %r397, %r392, 124; + and.b32 %r398, %r395, 126; + or.b32 %r399, %r398, 1; + .loc 1 95 21 // l2norm.py:95:21 + or.b32 %r400, %r393, 8; + or.b32 %r401, %r393, 12; + or.b32 %r402, %r393, 16; + or.b32 %r403, %r393, 20; + or.b32 %r404, %r393, 24; + or.b32 %r405, %r393, 28; + or.b32 %r406, %r393, 32; + or.b32 %r407, %r393, 36; + or.b32 %r408, %r393, 40; + or.b32 %r409, %r393, 44; + or.b32 %r410, %r393, 48; + or.b32 %r411, %r393, 52; + or.b32 %r412, %r393, 56; + or.b32 %r413, %r392, 60; + or.b32 %r414, %r393, 64; + or.b32 %r415, %r393, 68; + or.b32 %r416, %r393, 96; + or.b32 %r417, %r393, 100; + or.b32 %r418, %r393, 88; + or.b32 %r419, %r393, 92; + or.b32 %r420, %r393, 80; + or.b32 %r421, %r393, 84; + or.b32 %r422, %r393, 72; + or.b32 %r423, %r393, 76; + or.b32 %r424, %r393, 112; + or.b32 %r425, %r393, 116; + or.b32 %r426, %r393, 104; + or.b32 %r427, %r393, 108; + cvt.u64.u32 %rd109, %r423; + cvt.u64.u32 %rd110, %r422; + cvt.u64.u32 %rd111, %r421; + cvt.u64.u32 %rd112, %r420; + cvt.u64.u32 %rd113, %r419; + cvt.u64.u32 %rd114, %r418; + cvt.u64.u32 %rd115, %r417; + cvt.u64.u32 %rd116, %r416; + cvt.u64.u32 %rd117, %r427; + cvt.u64.u32 %rd118, %r426; + cvt.u64.u32 %rd119, %r425; + cvt.u64.u32 %rd120, %r424; + cvt.u64.u32 %rd121, %r397; + cvt.u64.u32 %rd122, %r396; + cvt.u64.u32 %rd123, %r398; + cvt.u64.u32 %rd124, %r399; + cvt.u64.u32 %rd125, %r415; + cvt.u64.u32 %rd126, %r414; + cvt.u64.u32 %rd127, %r413; + cvt.u64.u32 %rd128, %r412; + cvt.u64.u32 %rd129, %r411; + cvt.u64.u32 %rd130, %r410; + cvt.u64.u32 %rd131, %r409; + cvt.u64.u32 %rd132, %r408; + cvt.u64.u32 %rd133, %r407; + cvt.u64.u32 %rd134, %r406; + cvt.u64.u32 %rd135, %r405; + cvt.u64.u32 %rd136, %r404; + cvt.u64.u32 %rd137, %r403; + cvt.u64.u32 %rd138, %r402; + cvt.u64.u32 %rd139, %r401; + cvt.u64.u32 %rd140, %r400; + or.b64 %rd141, %rd102, %rd140; + or.b64 %rd142, %rd102, %rd139; + or.b64 %rd143, %rd102, %rd138; + or.b64 %rd144, %rd102, %rd137; + or.b64 %rd145, %rd102, %rd136; + or.b64 %rd146, %rd102, %rd135; + or.b64 %rd147, %rd102, %rd134; + or.b64 %rd148, %rd102, %rd133; + or.b64 %rd149, %rd102, %rd132; + or.b64 %rd150, %rd102, %rd131; + or.b64 %rd151, %rd102, %rd130; + or.b64 %rd152, %rd102, %rd129; + or.b64 %rd153, %rd102, %rd128; + or.b64 %rd154, %rd102, %rd127; + or.b64 %rd155, %rd102, %rd126; + or.b64 %rd156, %rd102, %rd125; + or.b64 %rd157, %rd102, %rd124; + or.b64 %rd158, %rd102, %rd123; + or.b64 %rd159, %rd102, %rd122; + or.b64 %rd160, %rd102, %rd121; + or.b64 %rd161, %rd102, %rd120; + or.b64 %rd162, %rd102, %rd119; + or.b64 %rd163, %rd102, %rd118; + or.b64 %rd164, %rd102, %rd117; + or.b64 %rd165, %rd102, %rd116; + or.b64 %rd166, %rd102, %rd115; + or.b64 %rd167, %rd102, %rd114; + or.b64 %rd168, %rd102, %rd113; + or.b64 %rd169, %rd102, %rd112; + or.b64 %rd170, %rd102, %rd111; + or.b64 %rd171, %rd102, %rd110; + or.b64 %rd172, %rd102, %rd109; + .loc 1 97 16 // l2norm.py:97:16 + setp.lt.s64 %p35, %rd172, %rd104; + selp.b16 %rs1, 1, 0, %p35; + shl.b16 %rs2, %rs1, 2; + setp.lt.s64 %p36, %rd171, %rd104; + selp.b16 %rs3, -1, 0, %p36; + shl.b16 %rs4, %rs3, 3; + or.b16 %rs5, %rs4, %rs2; + setp.lt.s64 %p37, %rd170, %rd104; + selp.b16 %rs6, 1, 0, %p37; + setp.lt.s64 %p38, %rd169, %rd104; + selp.b16 %rs7, -1, 0, %p38; + shl.b16 %rs8, %rs7, 1; + or.b16 %rs9, %rs6, %rs8; + and.b16 %rs10, %rs9, 3; + or.b16 %rs11, %rs10, %rs5; + shl.b16 %rs12, %rs11, 12; + setp.lt.s64 %p39, %rd168, %rd104; + selp.b16 %rs13, 1, 0, %p39; + shl.b16 %rs14, %rs13, 2; + setp.lt.s64 %p40, %rd167, %rd104; + selp.b16 %rs15, -1, 0, %p40; + shl.b16 %rs16, %rs15, 3; + or.b16 %rs17, %rs16, %rs14; + setp.lt.s64 %p41, %rd166, %rd104; + selp.b16 %rs18, 1, 0, %p41; + setp.lt.s64 %p42, %rd165, %rd104; + selp.b16 %rs19, -1, 0, %p42; + shl.b16 %rs20, %rs19, 1; + or.b16 %rs21, %rs18, %rs20; + and.b16 %rs22, %rs21, 3; + or.b16 %rs23, %rs22, %rs17; + and.b16 %rs24, %rs23, 15; + shl.b16 %rs25, %rs24, 8; + or.b16 %rs26, %rs12, %rs25; + setp.lt.s64 %p43, %rd164, %rd104; + selp.b16 %rs27, 1, 0, %p43; + shl.b16 %rs28, %rs27, 2; + setp.lt.s64 %p44, %rd163, %rd104; + selp.b16 %rs29, -1, 0, %p44; + shl.b16 %rs30, %rs29, 3; + or.b16 %rs31, %rs30, %rs28; + setp.lt.s64 %p45, %rd162, %rd104; + selp.b16 %rs32, 1, 0, %p45; + setp.lt.s64 %p46, %rd161, %rd104; + selp.b16 %rs33, -1, 0, %p46; + shl.b16 %rs34, %rs33, 1; + or.b16 %rs35, %rs32, %rs34; + and.b16 %rs36, %rs35, 3; + or.b16 %rs37, %rs36, %rs31; + shl.b16 %rs38, %rs37, 4; + setp.lt.s64 %p47, %rd160, %rd104; + selp.b16 %rs39, 1, 0, %p47; + shl.b16 %rs40, %rs39, 2; + setp.lt.s64 %p48, %rd159, %rd104; + selp.b16 %rs41, -1, 0, %p48; + shl.b16 %rs42, %rs41, 3; + or.b16 %rs43, %rs42, %rs40; + setp.lt.s64 %p49, %rd158, %rd104; + selp.b16 %rs44, -1, 0, %p49; + shl.b16 %rs45, %rs44, 1; + setp.lt.s64 %p5, %rd157, %rd104; + selp.b16 %rs46, 1, 0, %p5; + or.b16 %rs47, %rs46, %rs45; + and.b16 %rs48, %rs47, 3; + or.b16 %rs49, %rs48, %rs43; + and.b16 %rs50, %rs49, 15; + or.b16 %rs51, %rs50, %rs38; + and.b16 %rs52, %rs51, 255; + or.b16 %rs53, %rs52, %rs26; + cvt.u32.u16 %r428, %rs53; + setp.lt.s64 %p20, %rd156, %rd104; + setp.lt.s64 %p19, %rd155, %rd104; + setp.lt.s64 %p18, %rd154, %rd104; + setp.lt.s64 %p17, %rd153, %rd104; + setp.lt.s64 %p16, %rd152, %rd104; + setp.lt.s64 %p15, %rd151, %rd104; + setp.lt.s64 %p14, %rd150, %rd104; + setp.lt.s64 %p13, %rd149, %rd104; + setp.lt.s64 %p3, %rd148, %rd104; + setp.lt.s64 %p12, %rd147, %rd104; + setp.lt.s64 %p11, %rd146, %rd104; + setp.lt.s64 %p10, %rd145, %rd104; + setp.lt.s64 %p9, %rd144, %rd104; + setp.lt.s64 %p8, %rd143, %rd104; + setp.lt.s64 %p7, %rd142, %rd104; + setp.lt.s64 %p6, %rd141, %rd104; + .loc 1 98 30 // l2norm.py:98:30 + shl.b32 %r429, %r391, 3; + and.b32 %r430, %r429, 120; + .loc 1 99 14 // l2norm.py:99:14 + shl.b64 %rd173, %rd107, 8; + add.s64 %rd174, %rd99, %rd173; + shl.b64 %rd175, %rd108, 8; + add.s64 %rd176, %rd99, %rd175; + shl.b64 %rd177, %rd141, 8; + add.s64 %rd178, %rd99, %rd177; + shl.b64 %rd179, %rd142, 8; + add.s64 %rd180, %rd99, %rd179; + shl.b64 %rd181, %rd143, 8; + add.s64 %rd182, %rd99, %rd181; + shl.b64 %rd183, %rd144, 8; + add.s64 %rd184, %rd99, %rd183; + shl.b64 %rd185, %rd145, 8; + add.s64 %rd186, %rd99, %rd185; + shl.b64 %rd187, %rd146, 8; + add.s64 %rd188, %rd99, %rd187; + shl.b64 %rd189, %rd147, 8; + add.s64 %rd190, %rd99, %rd189; + shl.b64 %rd191, %rd148, 8; + add.s64 %rd192, %rd99, %rd191; + shl.b64 %rd193, %rd149, 8; + add.s64 %rd194, %rd99, %rd193; + shl.b64 %rd195, %rd150, 8; + add.s64 %rd196, %rd99, %rd195; + shl.b64 %rd197, %rd151, 8; + add.s64 %rd198, %rd99, %rd197; + shl.b64 %rd199, %rd152, 8; + add.s64 %rd200, %rd99, %rd199; + shl.b64 %rd201, %rd153, 8; + add.s64 %rd202, %rd99, %rd201; + shl.b64 %rd203, %rd154, 8; + add.s64 %rd204, %rd99, %rd203; + shl.b64 %rd205, %rd155, 8; + add.s64 %rd206, %rd99, %rd205; + shl.b64 %rd207, %rd156, 8; + add.s64 %rd208, %rd99, %rd207; + shl.b64 %rd209, %rd171, 8; + add.s64 %rd210, %rd99, %rd209; + shl.b64 %rd211, %rd172, 8; + add.s64 %rd212, %rd99, %rd211; + shl.b64 %rd213, %rd169, 8; + add.s64 %rd214, %rd99, %rd213; + shl.b64 %rd215, %rd170, 8; + add.s64 %rd216, %rd99, %rd215; + shl.b64 %rd217, %rd167, 8; + add.s64 %rd218, %rd99, %rd217; + shl.b64 %rd219, %rd168, 8; + add.s64 %rd220, %rd99, %rd219; + shl.b64 %rd221, %rd165, 8; + add.s64 %rd222, %rd99, %rd221; + shl.b64 %rd223, %rd166, 8; + add.s64 %rd224, %rd99, %rd223; + shl.b64 %rd225, %rd163, 8; + add.s64 %rd226, %rd99, %rd225; + shl.b64 %rd227, %rd164, 8; + add.s64 %rd228, %rd99, %rd227; + shl.b64 %rd229, %rd161, 8; + add.s64 %rd230, %rd99, %rd229; + shl.b64 %rd231, %rd162, 8; + add.s64 %rd232, %rd99, %rd231; + shl.b64 %rd233, %rd159, 8; + add.s64 %rd234, %rd99, %rd233; + shl.b64 %rd235, %rd160, 8; + add.s64 %rd236, %rd99, %rd235; + .loc 1 99 33 // l2norm.py:99:33 + mul.wide.u32 %rd237, %r430, 2; + add.s64 %rd1, %rd174, %rd237; + add.s64 %rd2, %rd176, %rd237; + add.s64 %rd3, %rd178, %rd237; + add.s64 %rd4, %rd180, %rd237; + add.s64 %rd5, %rd182, %rd237; + add.s64 %rd6, %rd184, %rd237; + add.s64 %rd7, %rd186, %rd237; + add.s64 %rd8, %rd188, %rd237; + add.s64 %rd9, %rd190, %rd237; + add.s64 %rd10, %rd192, %rd237; + add.s64 %rd11, %rd194, %rd237; + add.s64 %rd12, %rd196, %rd237; + add.s64 %rd13, %rd198, %rd237; + add.s64 %rd14, %rd200, %rd237; + add.s64 %rd15, %rd202, %rd237; + add.s64 %rd16, %rd204, %rd237; + add.s64 %rd17, %rd206, %rd237; + add.s64 %rd18, %rd208, %rd237; + add.s64 %rd19, %rd210, %rd237; + add.s64 %rd20, %rd212, %rd237; + add.s64 %rd21, %rd214, %rd237; + add.s64 %rd22, %rd216, %rd237; + add.s64 %rd23, %rd218, %rd237; + add.s64 %rd24, %rd220, %rd237; + add.s64 %rd25, %rd222, %rd237; + add.s64 %rd26, %rd224, %rd237; + add.s64 %rd27, %rd226, %rd237; + add.s64 %rd28, %rd228, %rd237; + add.s64 %rd29, %rd230, %rd237; + add.s64 %rd30, %rd232, %rd237; + add.s64 %rd31, %rd234, %rd237; + add.s64 %rd32, %rd236, %rd237; + .loc 1 100 20 // l2norm.py:100:20 + shl.b64 %rd238, %rd158, 2; + add.s64 %rd33, %rd100, %rd238; + shl.b64 %rd239, %rd157, 2; + add.s64 %rd34, %rd100, %rd239; + .loc 1 101 16 // l2norm.py:101:16 + add.s64 %rd240, %rd101, %rd173; + add.s64 %rd241, %rd101, %rd175; + add.s64 %rd242, %rd101, %rd177; + add.s64 %rd243, %rd101, %rd179; + add.s64 %rd244, %rd101, %rd181; + add.s64 %rd245, %rd101, %rd183; + add.s64 %rd246, %rd101, %rd185; + add.s64 %rd247, %rd101, %rd187; + add.s64 %rd248, %rd101, %rd189; + add.s64 %rd249, %rd101, %rd191; + add.s64 %rd250, %rd101, %rd193; + add.s64 %rd251, %rd101, %rd195; + add.s64 %rd252, %rd101, %rd197; + add.s64 %rd253, %rd101, %rd199; + add.s64 %rd254, %rd101, %rd201; + add.s64 %rd255, %rd101, %rd203; + add.s64 %rd256, %rd101, %rd205; + add.s64 %rd257, %rd101, %rd207; + add.s64 %rd258, %rd101, %rd209; + add.s64 %rd259, %rd101, %rd211; + add.s64 %rd260, %rd101, %rd213; + add.s64 %rd261, %rd101, %rd215; + add.s64 %rd262, %rd101, %rd217; + add.s64 %rd263, %rd101, %rd219; + add.s64 %rd264, %rd101, %rd221; + add.s64 %rd265, %rd101, %rd223; + add.s64 %rd266, %rd101, %rd225; + add.s64 %rd267, %rd101, %rd227; + add.s64 %rd268, %rd101, %rd229; + add.s64 %rd269, %rd101, %rd231; + add.s64 %rd270, %rd101, %rd233; + add.s64 %rd271, %rd101, %rd235; + .loc 1 101 35 // l2norm.py:101:35 + add.s64 %rd35, %rd240, %rd237; + add.s64 %rd36, %rd241, %rd237; + add.s64 %rd37, %rd242, %rd237; + add.s64 %rd38, %rd243, %rd237; + add.s64 %rd39, %rd244, %rd237; + add.s64 %rd40, %rd245, %rd237; + add.s64 %rd41, %rd246, %rd237; + add.s64 %rd42, %rd247, %rd237; + add.s64 %rd43, %rd248, %rd237; + add.s64 %rd44, %rd249, %rd237; + add.s64 %rd45, %rd250, %rd237; + add.s64 %rd46, %rd251, %rd237; + add.s64 %rd47, %rd252, %rd237; + add.s64 %rd48, %rd253, %rd237; + add.s64 %rd49, %rd254, %rd237; + add.s64 %rd50, %rd255, %rd237; + add.s64 %rd51, %rd256, %rd237; + add.s64 %rd52, %rd257, %rd237; + add.s64 %rd53, %rd258, %rd237; + add.s64 %rd54, %rd259, %rd237; + add.s64 %rd55, %rd260, %rd237; + add.s64 %rd56, %rd261, %rd237; + add.s64 %rd57, %rd262, %rd237; + add.s64 %rd58, %rd263, %rd237; + add.s64 %rd59, %rd264, %rd237; + add.s64 %rd60, %rd265, %rd237; + add.s64 %rd61, %rd266, %rd237; + add.s64 %rd62, %rd267, %rd237; + add.s64 %rd63, %rd268, %rd237; + add.s64 %rd64, %rd269, %rd237; + add.s64 %rd65, %rd270, %rd237; + add.s64 %rd66, %rd271, %rd237; + .loc 1 102 16 // l2norm.py:102:16 + add.s64 %rd272, %rd103, %rd173; + add.s64 %rd273, %rd103, %rd175; + add.s64 %rd274, %rd103, %rd177; + add.s64 %rd275, %rd103, %rd179; + add.s64 %rd276, %rd103, %rd181; + add.s64 %rd277, %rd103, %rd183; + add.s64 %rd278, %rd103, %rd185; + add.s64 %rd279, %rd103, %rd187; + add.s64 %rd280, %rd103, %rd189; + add.s64 %rd281, %rd103, %rd191; + add.s64 %rd282, %rd103, %rd193; + add.s64 %rd283, %rd103, %rd195; + add.s64 %rd284, %rd103, %rd197; + add.s64 %rd285, %rd103, %rd199; + add.s64 %rd286, %rd103, %rd201; + add.s64 %rd287, %rd103, %rd203; + add.s64 %rd288, %rd103, %rd205; + add.s64 %rd289, %rd103, %rd207; + add.s64 %rd290, %rd103, %rd209; + add.s64 %rd291, %rd103, %rd211; + add.s64 %rd292, %rd103, %rd213; + add.s64 %rd293, %rd103, %rd215; + add.s64 %rd294, %rd103, %rd217; + add.s64 %rd295, %rd103, %rd219; + add.s64 %rd296, %rd103, %rd221; + add.s64 %rd297, %rd103, %rd223; + add.s64 %rd298, %rd103, %rd225; + add.s64 %rd299, %rd103, %rd227; + add.s64 %rd300, %rd103, %rd229; + add.s64 %rd301, %rd103, %rd231; + add.s64 %rd302, %rd103, %rd233; + add.s64 %rd303, %rd103, %rd235; + .loc 1 102 35 // l2norm.py:102:35 + add.s64 %rd67, %rd272, %rd237; + add.s64 %rd68, %rd273, %rd237; + add.s64 %rd69, %rd274, %rd237; + add.s64 %rd70, %rd275, %rd237; + add.s64 %rd71, %rd276, %rd237; + add.s64 %rd72, %rd277, %rd237; + add.s64 %rd73, %rd278, %rd237; + add.s64 %rd74, %rd279, %rd237; + add.s64 %rd75, %rd280, %rd237; + add.s64 %rd76, %rd281, %rd237; + add.s64 %rd77, %rd282, %rd237; + add.s64 %rd78, %rd283, %rd237; + add.s64 %rd79, %rd284, %rd237; + add.s64 %rd80, %rd285, %rd237; + add.s64 %rd81, %rd286, %rd237; + add.s64 %rd82, %rd287, %rd237; + add.s64 %rd83, %rd288, %rd237; + add.s64 %rd84, %rd289, %rd237; + add.s64 %rd85, %rd290, %rd237; + add.s64 %rd86, %rd291, %rd237; + add.s64 %rd87, %rd292, %rd237; + add.s64 %rd88, %rd293, %rd237; + add.s64 %rd89, %rd294, %rd237; + add.s64 %rd90, %rd295, %rd237; + add.s64 %rd91, %rd296, %rd237; + add.s64 %rd92, %rd297, %rd237; + add.s64 %rd93, %rd298, %rd237; + add.s64 %rd94, %rd299, %rd237; + add.s64 %rd95, %rd300, %rd237; + add.s64 %rd96, %rd301, %rd237; + add.s64 %rd97, %rd302, %rd237; + add.s64 %rd98, %rd303, %rd237; + mov.b32 %r5, 0; + .loc 1 104 18 // l2norm.py:104:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs54, %rs55}, %r1; + mov.b32 {%rs56, %rs57}, %r2; + mov.b32 {%rs58, %rs59}, %r3; + mov.b32 {%rs60, %rs61}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs62, %rs63}, %r6; + mov.b32 {%rs64, %rs65}, %r7; + mov.b32 {%rs66, %rs67}, %r8; + mov.b32 {%rs68, %rs69}, %r9; + // begin inline asm + mov.u32 %r10, %r5; + mov.u32 %r11, %r5; + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + @%p6 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs70, %rs71}, %r10; + mov.b32 {%rs72, %rs73}, %r11; + mov.b32 {%rs74, %rs75}, %r12; + mov.b32 {%rs76, %rs77}, %r13; + // begin inline asm + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + @%p7 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ]; + // end inline asm + mov.b32 {%rs78, %rs79}, %r14; + mov.b32 {%rs80, %rs81}, %r15; + mov.b32 {%rs82, %rs83}, %r16; + mov.b32 {%rs84, %rs85}, %r17; + // begin inline asm + mov.u32 %r18, %r5; + mov.u32 %r19, %r5; + mov.u32 %r20, %r5; + mov.u32 %r21, %r5; + @%p8 ld.global.v4.b32 { %r18, %r19, %r20, %r21 }, [ %rd5 + 0 ]; + // end inline asm + mov.b32 {%rs86, %rs87}, %r18; + mov.b32 {%rs88, %rs89}, %r19; + mov.b32 {%rs90, %rs91}, %r20; + mov.b32 {%rs92, %rs93}, %r21; + // begin inline asm + mov.u32 %r22, %r5; + mov.u32 %r23, %r5; + mov.u32 %r24, %r5; + mov.u32 %r25, %r5; + @%p9 ld.global.v4.b32 { %r22, %r23, %r24, %r25 }, [ %rd6 + 0 ]; + // end inline asm + mov.b32 {%rs94, %rs95}, %r22; + mov.b32 {%rs96, %rs97}, %r23; + mov.b32 {%rs98, %rs99}, %r24; + mov.b32 {%rs100, %rs101}, %r25; + // begin inline asm + mov.u32 %r26, %r5; + mov.u32 %r27, %r5; + mov.u32 %r28, %r5; + mov.u32 %r29, %r5; + @%p10 ld.global.v4.b32 { %r26, %r27, %r28, %r29 }, [ %rd7 + 0 ]; + // end inline asm + mov.b32 {%rs102, %rs103}, %r26; + mov.b32 {%rs104, %rs105}, %r27; + mov.b32 {%rs106, %rs107}, %r28; + mov.b32 {%rs108, %rs109}, %r29; + // begin inline asm + mov.u32 %r30, %r5; + mov.u32 %r31, %r5; + mov.u32 %r32, %r5; + mov.u32 %r33, %r5; + @%p11 ld.global.v4.b32 { %r30, %r31, %r32, %r33 }, [ %rd8 + 0 ]; + // end inline asm + mov.b32 {%rs110, %rs111}, %r30; + mov.b32 {%rs112, %rs113}, %r31; + mov.b32 {%rs114, %rs115}, %r32; + mov.b32 {%rs116, %rs117}, %r33; + // begin inline asm + mov.u32 %r34, %r5; + mov.u32 %r35, %r5; + mov.u32 %r36, %r5; + mov.u32 %r37, %r5; + @%p12 ld.global.v4.b32 { %r34, %r35, %r36, %r37 }, [ %rd9 + 0 ]; + // end inline asm + mov.b32 {%rs118, %rs119}, %r34; + mov.b32 {%rs120, %rs121}, %r35; + mov.b32 {%rs122, %rs123}, %r36; + mov.b32 {%rs124, %rs125}, %r37; + // begin inline asm + mov.u32 %r38, %r5; + mov.u32 %r39, %r5; + mov.u32 %r40, %r5; + mov.u32 %r41, %r5; + @%p3 ld.global.v4.b32 { %r38, %r39, %r40, %r41 }, [ %rd10 + 0 ]; + // end inline asm + mov.b32 {%rs126, %rs127}, %r38; + mov.b32 {%rs128, %rs129}, %r39; + mov.b32 {%rs130, %rs131}, %r40; + mov.b32 {%rs132, %rs133}, %r41; + // begin inline asm + mov.u32 %r42, %r5; + mov.u32 %r43, %r5; + mov.u32 %r44, %r5; + mov.u32 %r45, %r5; + @%p13 ld.global.v4.b32 { %r42, %r43, %r44, %r45 }, [ %rd11 + 0 ]; + // end inline asm + mov.b32 {%rs134, %rs135}, %r42; + mov.b32 {%rs136, %rs137}, %r43; + mov.b32 {%rs138, %rs139}, %r44; + mov.b32 {%rs140, %rs141}, %r45; + // begin inline asm + mov.u32 %r46, %r5; + mov.u32 %r47, %r5; + mov.u32 %r48, %r5; + mov.u32 %r49, %r5; + @%p14 ld.global.v4.b32 { %r46, %r47, %r48, %r49 }, [ %rd12 + 0 ]; + // end inline asm + mov.b32 {%rs142, %rs143}, %r46; + mov.b32 {%rs144, %rs145}, %r47; + mov.b32 {%rs146, %rs147}, %r48; + mov.b32 {%rs148, %rs149}, %r49; + // begin inline asm + mov.u32 %r50, %r5; + mov.u32 %r51, %r5; + mov.u32 %r52, %r5; + mov.u32 %r53, %r5; + @%p15 ld.global.v4.b32 { %r50, %r51, %r52, %r53 }, [ %rd13 + 0 ]; + // end inline asm + mov.b32 {%rs150, %rs151}, %r50; + mov.b32 {%rs152, %rs153}, %r51; + mov.b32 {%rs154, %rs155}, %r52; + mov.b32 {%rs156, %rs157}, %r53; + // begin inline asm + mov.u32 %r54, %r5; + mov.u32 %r55, %r5; + mov.u32 %r56, %r5; + mov.u32 %r57, %r5; + @%p16 ld.global.v4.b32 { %r54, %r55, %r56, %r57 }, [ %rd14 + 0 ]; + // end inline asm + mov.b32 {%rs158, %rs159}, %r54; + mov.b32 {%rs160, %rs161}, %r55; + mov.b32 {%rs162, %rs163}, %r56; + mov.b32 {%rs164, %rs165}, %r57; + // begin inline asm + mov.u32 %r58, %r5; + mov.u32 %r59, %r5; + mov.u32 %r60, %r5; + mov.u32 %r61, %r5; + @%p17 ld.global.v4.b32 { %r58, %r59, %r60, %r61 }, [ %rd15 + 0 ]; + // end inline asm + mov.b32 {%rs166, %rs167}, %r58; + mov.b32 {%rs168, %rs169}, %r59; + mov.b32 {%rs170, %rs171}, %r60; + mov.b32 {%rs172, %rs173}, %r61; + // begin inline asm + mov.u32 %r62, %r5; + mov.u32 %r63, %r5; + mov.u32 %r64, %r5; + mov.u32 %r65, %r5; + @%p18 ld.global.v4.b32 { %r62, %r63, %r64, %r65 }, [ %rd16 + 0 ]; + // end inline asm + mov.b32 {%rs174, %rs175}, %r62; + mov.b32 {%rs176, %rs177}, %r63; + mov.b32 {%rs178, %rs179}, %r64; + mov.b32 {%rs180, %rs181}, %r65; + // begin inline asm + mov.u32 %r66, %r5; + mov.u32 %r67, %r5; + mov.u32 %r68, %r5; + mov.u32 %r69, %r5; + @%p19 ld.global.v4.b32 { %r66, %r67, %r68, %r69 }, [ %rd17 + 0 ]; + // end inline asm + mov.b32 {%rs182, %rs183}, %r66; + mov.b32 {%rs184, %rs185}, %r67; + mov.b32 {%rs186, %rs187}, %r68; + mov.b32 {%rs188, %rs189}, %r69; + // begin inline asm + mov.u32 %r70, %r5; + mov.u32 %r71, %r5; + mov.u32 %r72, %r5; + mov.u32 %r73, %r5; + @%p20 ld.global.v4.b32 { %r70, %r71, %r72, %r73 }, [ %rd18 + 0 ]; + // end inline asm + mov.b32 {%rs190, %rs191}, %r70; + mov.b32 {%rs192, %rs193}, %r71; + mov.b32 {%rs194, %rs195}, %r72; + mov.b32 {%rs196, %rs197}, %r73; + shr.u32 %r431, %r428, 15; + and.b32 %r432, %r431, 1; + setp.ne.b32 %p21, %r432, 0; + // begin inline asm + mov.u32 %r74, %r5; + mov.u32 %r75, %r5; + mov.u32 %r76, %r5; + mov.u32 %r77, %r5; + @%p21 ld.global.v4.b32 { %r74, %r75, %r76, %r77 }, [ %rd19 + 0 ]; + // end inline asm + mov.b32 {%rs198, %rs199}, %r74; + mov.b32 {%rs200, %rs201}, %r75; + mov.b32 {%rs202, %rs203}, %r76; + mov.b32 {%rs204, %rs205}, %r77; + shr.u32 %r433, %r428, 14; + and.b32 %r434, %r433, 1; + setp.ne.b32 %p22, %r434, 0; + // begin inline asm + mov.u32 %r78, %r5; + mov.u32 %r79, %r5; + mov.u32 %r80, %r5; + mov.u32 %r81, %r5; + @%p22 ld.global.v4.b32 { %r78, %r79, %r80, %r81 }, [ %rd20 + 0 ]; + // end inline asm + mov.b32 {%rs206, %rs207}, %r78; + mov.b32 {%rs208, %rs209}, %r79; + mov.b32 {%rs210, %rs211}, %r80; + mov.b32 {%rs212, %rs213}, %r81; + shr.u32 %r435, %r428, 13; + and.b32 %r436, %r435, 1; + setp.ne.b32 %p23, %r436, 0; + // begin inline asm + mov.u32 %r82, %r5; + mov.u32 %r83, %r5; + mov.u32 %r84, %r5; + mov.u32 %r85, %r5; + @%p23 ld.global.v4.b32 { %r82, %r83, %r84, %r85 }, [ %rd21 + 0 ]; + // end inline asm + mov.b32 {%rs214, %rs215}, %r82; + mov.b32 {%rs216, %rs217}, %r83; + mov.b32 {%rs218, %rs219}, %r84; + mov.b32 {%rs220, %rs221}, %r85; + shr.u32 %r437, %r428, 12; + and.b32 %r438, %r437, 1; + setp.ne.b32 %p24, %r438, 0; + // begin inline asm + mov.u32 %r86, %r5; + mov.u32 %r87, %r5; + mov.u32 %r88, %r5; + mov.u32 %r89, %r5; + @%p24 ld.global.v4.b32 { %r86, %r87, %r88, %r89 }, [ %rd22 + 0 ]; + // end inline asm + mov.b32 {%rs222, %rs223}, %r86; + mov.b32 {%rs224, %rs225}, %r87; + mov.b32 {%rs226, %rs227}, %r88; + mov.b32 {%rs228, %rs229}, %r89; + shr.u32 %r439, %r428, 11; + and.b32 %r440, %r439, 1; + setp.ne.b32 %p25, %r440, 0; + // begin inline asm + mov.u32 %r90, %r5; + mov.u32 %r91, %r5; + mov.u32 %r92, %r5; + mov.u32 %r93, %r5; + @%p25 ld.global.v4.b32 { %r90, %r91, %r92, %r93 }, [ %rd23 + 0 ]; + // end inline asm + mov.b32 {%rs230, %rs231}, %r90; + mov.b32 {%rs232, %rs233}, %r91; + mov.b32 {%rs234, %rs235}, %r92; + mov.b32 {%rs236, %rs237}, %r93; + shr.u32 %r441, %r428, 10; + and.b32 %r442, %r441, 1; + setp.ne.b32 %p26, %r442, 0; + // begin inline asm + mov.u32 %r94, %r5; + mov.u32 %r95, %r5; + mov.u32 %r96, %r5; + mov.u32 %r97, %r5; + @%p26 ld.global.v4.b32 { %r94, %r95, %r96, %r97 }, [ %rd24 + 0 ]; + // end inline asm + mov.b32 {%rs238, %rs239}, %r94; + mov.b32 {%rs240, %rs241}, %r95; + mov.b32 {%rs242, %rs243}, %r96; + mov.b32 {%rs244, %rs245}, %r97; + shr.u32 %r443, %r428, 9; + and.b32 %r444, %r443, 1; + setp.ne.b32 %p27, %r444, 0; + // begin inline asm + mov.u32 %r98, %r5; + mov.u32 %r99, %r5; + mov.u32 %r100, %r5; + mov.u32 %r101, %r5; + @%p27 ld.global.v4.b32 { %r98, %r99, %r100, %r101 }, [ %rd25 + 0 ]; + // end inline asm + mov.b32 {%rs246, %rs247}, %r98; + mov.b32 {%rs248, %rs249}, %r99; + mov.b32 {%rs250, %rs251}, %r100; + mov.b32 {%rs252, %rs253}, %r101; + shr.u32 %r445, %r428, 8; + and.b32 %r446, %r445, 1; + setp.ne.b32 %p28, %r446, 0; + // begin inline asm + mov.u32 %r102, %r5; + mov.u32 %r103, %r5; + mov.u32 %r104, %r5; + mov.u32 %r105, %r5; + @%p28 ld.global.v4.b32 { %r102, %r103, %r104, %r105 }, [ %rd26 + 0 ]; + // end inline asm + mov.b32 {%rs254, %rs255}, %r102; + mov.b32 {%rs256, %rs257}, %r103; + mov.b32 {%rs258, %rs259}, %r104; + mov.b32 {%rs260, %rs261}, %r105; + shr.u32 %r447, %r428, 7; + and.b32 %r448, %r447, 1; + setp.ne.b32 %p29, %r448, 0; + // begin inline asm + mov.u32 %r106, %r5; + mov.u32 %r107, %r5; + mov.u32 %r108, %r5; + mov.u32 %r109, %r5; + @%p29 ld.global.v4.b32 { %r106, %r107, %r108, %r109 }, [ %rd27 + 0 ]; + // end inline asm + mov.b32 {%rs262, %rs263}, %r106; + mov.b32 {%rs264, %rs265}, %r107; + mov.b32 {%rs266, %rs267}, %r108; + mov.b32 {%rs268, %rs269}, %r109; + shr.u32 %r449, %r428, 6; + and.b32 %r450, %r449, 1; + setp.ne.b32 %p30, %r450, 0; + // begin inline asm + mov.u32 %r110, %r5; + mov.u32 %r111, %r5; + mov.u32 %r112, %r5; + mov.u32 %r113, %r5; + @%p30 ld.global.v4.b32 { %r110, %r111, %r112, %r113 }, [ %rd28 + 0 ]; + // end inline asm + mov.b32 {%rs270, %rs271}, %r110; + mov.b32 {%rs272, %rs273}, %r111; + mov.b32 {%rs274, %rs275}, %r112; + mov.b32 {%rs276, %rs277}, %r113; + shr.u32 %r451, %r428, 5; + and.b32 %r452, %r451, 1; + setp.ne.b32 %p31, %r452, 0; + // begin inline asm + mov.u32 %r114, %r5; + mov.u32 %r115, %r5; + mov.u32 %r116, %r5; + mov.u32 %r117, %r5; + @%p31 ld.global.v4.b32 { %r114, %r115, %r116, %r117 }, [ %rd29 + 0 ]; + // end inline asm + mov.b32 {%rs278, %rs279}, %r114; + mov.b32 {%rs280, %rs281}, %r115; + mov.b32 {%rs282, %rs283}, %r116; + mov.b32 {%rs284, %rs285}, %r117; + shr.u32 %r453, %r428, 4; + and.b32 %r454, %r453, 1; + setp.ne.b32 %p32, %r454, 0; + // begin inline asm + mov.u32 %r118, %r5; + mov.u32 %r119, %r5; + mov.u32 %r120, %r5; + mov.u32 %r121, %r5; + @%p32 ld.global.v4.b32 { %r118, %r119, %r120, %r121 }, [ %rd30 + 0 ]; + // end inline asm + mov.b32 {%rs286, %rs287}, %r118; + mov.b32 {%rs288, %rs289}, %r119; + mov.b32 {%rs290, %rs291}, %r120; + mov.b32 {%rs292, %rs293}, %r121; + shr.u32 %r455, %r428, 3; + and.b32 %r456, %r455, 1; + setp.ne.b32 %p33, %r456, 0; + // begin inline asm + mov.u32 %r122, %r5; + mov.u32 %r123, %r5; + mov.u32 %r124, %r5; + mov.u32 %r125, %r5; + @%p33 ld.global.v4.b32 { %r122, %r123, %r124, %r125 }, [ %rd31 + 0 ]; + // end inline asm + mov.b32 {%rs294, %rs295}, %r122; + mov.b32 {%rs296, %rs297}, %r123; + mov.b32 {%rs298, %rs299}, %r124; + mov.b32 {%rs300, %rs301}, %r125; + shr.u32 %r457, %r428, 2; + and.b32 %r458, %r457, 1; + setp.ne.b32 %p34, %r458, 0; + // begin inline asm + mov.u32 %r126, %r5; + mov.u32 %r127, %r5; + mov.u32 %r128, %r5; + mov.u32 %r129, %r5; + @%p34 ld.global.v4.b32 { %r126, %r127, %r128, %r129 }, [ %rd32 + 0 ]; + // end inline asm + mov.b32 {%rs302, %rs303}, %r126; + mov.b32 {%rs304, %rs305}, %r127; + mov.b32 {%rs306, %rs307}, %r128; + mov.b32 {%rs308, %rs309}, %r129; + .loc 1 104 47 // l2norm.py:104:47 + cvt.f32.bf16 %r459, %rs54; + cvt.f32.bf16 %r460, %rs55; + cvt.f32.bf16 %r461, %rs56; + cvt.f32.bf16 %r462, %rs57; + cvt.f32.bf16 %r463, %rs58; + cvt.f32.bf16 %r464, %rs59; + cvt.f32.bf16 %r465, %rs60; + cvt.f32.bf16 %r466, %rs61; + cvt.f32.bf16 %r467, %rs62; + cvt.f32.bf16 %r468, %rs63; + cvt.f32.bf16 %r469, %rs64; + cvt.f32.bf16 %r470, %rs65; + cvt.f32.bf16 %r471, %rs66; + cvt.f32.bf16 %r472, %rs67; + cvt.f32.bf16 %r473, %rs68; + cvt.f32.bf16 %r474, %rs69; + cvt.f32.bf16 %r475, %rs70; + cvt.f32.bf16 %r476, %rs71; + cvt.f32.bf16 %r477, %rs72; + cvt.f32.bf16 %r478, %rs73; + cvt.f32.bf16 %r479, %rs74; + cvt.f32.bf16 %r480, %rs75; + cvt.f32.bf16 %r481, %rs76; + cvt.f32.bf16 %r482, %rs77; + cvt.f32.bf16 %r483, %rs78; + cvt.f32.bf16 %r484, %rs79; + cvt.f32.bf16 %r485, %rs80; + cvt.f32.bf16 %r486, %rs81; + cvt.f32.bf16 %r487, %rs82; + cvt.f32.bf16 %r488, %rs83; + cvt.f32.bf16 %r489, %rs84; + cvt.f32.bf16 %r490, %rs85; + cvt.f32.bf16 %r491, %rs86; + cvt.f32.bf16 %r492, %rs87; + cvt.f32.bf16 %r493, %rs88; + cvt.f32.bf16 %r494, %rs89; + cvt.f32.bf16 %r495, %rs90; + cvt.f32.bf16 %r496, %rs91; + cvt.f32.bf16 %r497, %rs92; + cvt.f32.bf16 %r498, %rs93; + cvt.f32.bf16 %r499, %rs94; + cvt.f32.bf16 %r500, %rs95; + cvt.f32.bf16 %r501, %rs96; + cvt.f32.bf16 %r502, %rs97; + cvt.f32.bf16 %r503, %rs98; + cvt.f32.bf16 %r504, %rs99; + cvt.f32.bf16 %r505, %rs100; + cvt.f32.bf16 %r506, %rs101; + cvt.f32.bf16 %r507, %rs102; + cvt.f32.bf16 %r508, %rs103; + cvt.f32.bf16 %r509, %rs104; + cvt.f32.bf16 %r510, %rs105; + cvt.f32.bf16 %r511, %rs106; + cvt.f32.bf16 %r512, %rs107; + cvt.f32.bf16 %r513, %rs108; + cvt.f32.bf16 %r514, %rs109; + cvt.f32.bf16 %r515, %rs110; + cvt.f32.bf16 %r516, %rs111; + cvt.f32.bf16 %r517, %rs112; + cvt.f32.bf16 %r518, %rs113; + cvt.f32.bf16 %r519, %rs114; + cvt.f32.bf16 %r520, %rs115; + cvt.f32.bf16 %r521, %rs116; + cvt.f32.bf16 %r522, %rs117; + cvt.f32.bf16 %r523, %rs118; + cvt.f32.bf16 %r524, %rs119; + cvt.f32.bf16 %r525, %rs120; + cvt.f32.bf16 %r526, %rs121; + cvt.f32.bf16 %r527, %rs122; + cvt.f32.bf16 %r528, %rs123; + cvt.f32.bf16 %r529, %rs124; + cvt.f32.bf16 %r530, %rs125; + cvt.f32.bf16 %r531, %rs126; + cvt.f32.bf16 %r532, %rs127; + cvt.f32.bf16 %r533, %rs128; + cvt.f32.bf16 %r534, %rs129; + cvt.f32.bf16 %r535, %rs130; + cvt.f32.bf16 %r536, %rs131; + cvt.f32.bf16 %r537, %rs132; + cvt.f32.bf16 %r538, %rs133; + cvt.f32.bf16 %r539, %rs134; + cvt.f32.bf16 %r540, %rs135; + cvt.f32.bf16 %r541, %rs136; + cvt.f32.bf16 %r542, %rs137; + cvt.f32.bf16 %r543, %rs138; + cvt.f32.bf16 %r544, %rs139; + cvt.f32.bf16 %r545, %rs140; + cvt.f32.bf16 %r546, %rs141; + cvt.f32.bf16 %r547, %rs142; + cvt.f32.bf16 %r548, %rs143; + cvt.f32.bf16 %r549, %rs144; + cvt.f32.bf16 %r550, %rs145; + cvt.f32.bf16 %r551, %rs146; + cvt.f32.bf16 %r552, %rs147; + cvt.f32.bf16 %r553, %rs148; + cvt.f32.bf16 %r554, %rs149; + cvt.f32.bf16 %r555, %rs150; + cvt.f32.bf16 %r556, %rs151; + cvt.f32.bf16 %r557, %rs152; + cvt.f32.bf16 %r558, %rs153; + cvt.f32.bf16 %r559, %rs154; + cvt.f32.bf16 %r560, %rs155; + cvt.f32.bf16 %r561, %rs156; + cvt.f32.bf16 %r562, %rs157; + cvt.f32.bf16 %r563, %rs158; + cvt.f32.bf16 %r564, %rs159; + cvt.f32.bf16 %r565, %rs160; + cvt.f32.bf16 %r566, %rs161; + cvt.f32.bf16 %r567, %rs162; + cvt.f32.bf16 %r568, %rs163; + cvt.f32.bf16 %r569, %rs164; + cvt.f32.bf16 %r570, %rs165; + cvt.f32.bf16 %r571, %rs166; + cvt.f32.bf16 %r572, %rs167; + cvt.f32.bf16 %r573, %rs168; + cvt.f32.bf16 %r574, %rs169; + cvt.f32.bf16 %r575, %rs170; + cvt.f32.bf16 %r576, %rs171; + cvt.f32.bf16 %r577, %rs172; + cvt.f32.bf16 %r578, %rs173; + cvt.f32.bf16 %r579, %rs174; + cvt.f32.bf16 %r580, %rs175; + cvt.f32.bf16 %r581, %rs176; + cvt.f32.bf16 %r582, %rs177; + cvt.f32.bf16 %r583, %rs178; + cvt.f32.bf16 %r584, %rs179; + cvt.f32.bf16 %r585, %rs180; + cvt.f32.bf16 %r586, %rs181; + cvt.f32.bf16 %r587, %rs182; + cvt.f32.bf16 %r588, %rs183; + cvt.f32.bf16 %r589, %rs184; + cvt.f32.bf16 %r590, %rs185; + cvt.f32.bf16 %r591, %rs186; + cvt.f32.bf16 %r592, %rs187; + cvt.f32.bf16 %r593, %rs188; + cvt.f32.bf16 %r594, %rs189; + cvt.f32.bf16 %r595, %rs190; + cvt.f32.bf16 %r596, %rs191; + cvt.f32.bf16 %r597, %rs192; + cvt.f32.bf16 %r598, %rs193; + cvt.f32.bf16 %r599, %rs194; + cvt.f32.bf16 %r600, %rs195; + cvt.f32.bf16 %r601, %rs196; + cvt.f32.bf16 %r602, %rs197; + cvt.f32.bf16 %r603, %rs198; + cvt.f32.bf16 %r604, %rs199; + cvt.f32.bf16 %r605, %rs200; + cvt.f32.bf16 %r606, %rs201; + cvt.f32.bf16 %r607, %rs202; + cvt.f32.bf16 %r608, %rs203; + cvt.f32.bf16 %r609, %rs204; + cvt.f32.bf16 %r610, %rs205; + cvt.f32.bf16 %r611, %rs206; + cvt.f32.bf16 %r612, %rs207; + cvt.f32.bf16 %r613, %rs208; + cvt.f32.bf16 %r614, %rs209; + cvt.f32.bf16 %r615, %rs210; + cvt.f32.bf16 %r616, %rs211; + cvt.f32.bf16 %r617, %rs212; + cvt.f32.bf16 %r618, %rs213; + cvt.f32.bf16 %r619, %rs214; + cvt.f32.bf16 %r620, %rs215; + cvt.f32.bf16 %r621, %rs216; + cvt.f32.bf16 %r622, %rs217; + cvt.f32.bf16 %r623, %rs218; + cvt.f32.bf16 %r624, %rs219; + cvt.f32.bf16 %r625, %rs220; + cvt.f32.bf16 %r626, %rs221; + cvt.f32.bf16 %r627, %rs222; + cvt.f32.bf16 %r628, %rs223; + cvt.f32.bf16 %r629, %rs224; + cvt.f32.bf16 %r630, %rs225; + cvt.f32.bf16 %r631, %rs226; + cvt.f32.bf16 %r632, %rs227; + cvt.f32.bf16 %r633, %rs228; + cvt.f32.bf16 %r634, %rs229; + cvt.f32.bf16 %r635, %rs230; + cvt.f32.bf16 %r636, %rs231; + cvt.f32.bf16 %r637, %rs232; + cvt.f32.bf16 %r638, %rs233; + cvt.f32.bf16 %r639, %rs234; + cvt.f32.bf16 %r640, %rs235; + cvt.f32.bf16 %r641, %rs236; + cvt.f32.bf16 %r642, %rs237; + cvt.f32.bf16 %r643, %rs238; + cvt.f32.bf16 %r644, %rs239; + cvt.f32.bf16 %r645, %rs240; + cvt.f32.bf16 %r646, %rs241; + cvt.f32.bf16 %r647, %rs242; + cvt.f32.bf16 %r648, %rs243; + cvt.f32.bf16 %r649, %rs244; + cvt.f32.bf16 %r650, %rs245; + cvt.f32.bf16 %r651, %rs246; + cvt.f32.bf16 %r652, %rs247; + cvt.f32.bf16 %r653, %rs248; + cvt.f32.bf16 %r654, %rs249; + cvt.f32.bf16 %r655, %rs250; + cvt.f32.bf16 %r656, %rs251; + cvt.f32.bf16 %r657, %rs252; + cvt.f32.bf16 %r658, %rs253; + cvt.f32.bf16 %r659, %rs254; + cvt.f32.bf16 %r660, %rs255; + cvt.f32.bf16 %r661, %rs256; + cvt.f32.bf16 %r662, %rs257; + cvt.f32.bf16 %r663, %rs258; + cvt.f32.bf16 %r664, %rs259; + cvt.f32.bf16 %r665, %rs260; + cvt.f32.bf16 %r666, %rs261; + cvt.f32.bf16 %r667, %rs262; + cvt.f32.bf16 %r668, %rs263; + cvt.f32.bf16 %r669, %rs264; + cvt.f32.bf16 %r670, %rs265; + cvt.f32.bf16 %r671, %rs266; + cvt.f32.bf16 %r672, %rs267; + cvt.f32.bf16 %r673, %rs268; + cvt.f32.bf16 %r674, %rs269; + cvt.f32.bf16 %r675, %rs270; + cvt.f32.bf16 %r676, %rs271; + cvt.f32.bf16 %r677, %rs272; + cvt.f32.bf16 %r678, %rs273; + cvt.f32.bf16 %r679, %rs274; + cvt.f32.bf16 %r680, %rs275; + cvt.f32.bf16 %r681, %rs276; + cvt.f32.bf16 %r682, %rs277; + cvt.f32.bf16 %r683, %rs278; + cvt.f32.bf16 %r684, %rs279; + cvt.f32.bf16 %r685, %rs280; + cvt.f32.bf16 %r686, %rs281; + cvt.f32.bf16 %r687, %rs282; + cvt.f32.bf16 %r688, %rs283; + cvt.f32.bf16 %r689, %rs284; + cvt.f32.bf16 %r690, %rs285; + cvt.f32.bf16 %r691, %rs286; + cvt.f32.bf16 %r692, %rs287; + cvt.f32.bf16 %r693, %rs288; + cvt.f32.bf16 %r694, %rs289; + cvt.f32.bf16 %r695, %rs290; + cvt.f32.bf16 %r696, %rs291; + cvt.f32.bf16 %r697, %rs292; + cvt.f32.bf16 %r698, %rs293; + cvt.f32.bf16 %r699, %rs294; + cvt.f32.bf16 %r700, %rs295; + cvt.f32.bf16 %r701, %rs296; + cvt.f32.bf16 %r702, %rs297; + cvt.f32.bf16 %r703, %rs298; + cvt.f32.bf16 %r704, %rs299; + cvt.f32.bf16 %r705, %rs300; + cvt.f32.bf16 %r706, %rs301; + cvt.f32.bf16 %r707, %rs302; + cvt.f32.bf16 %r708, %rs303; + cvt.f32.bf16 %r709, %rs304; + cvt.f32.bf16 %r710, %rs305; + cvt.f32.bf16 %r711, %rs306; + cvt.f32.bf16 %r712, %rs307; + cvt.f32.bf16 %r713, %rs308; + cvt.f32.bf16 %r714, %rs309; + .loc 1 105 21 // l2norm.py:105:21 + shr.u32 %r715, %r428, 1; + and.b32 %r716, %r715, 1; + setp.ne.b32 %p4, %r716, 0; + // begin inline asm + mov.u32 %r259, %r5; + @%p4 ld.global.b32 { %r259 }, [ %rd33 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r261, %r5; + @%p5 ld.global.b32 { %r261 }, [ %rd34 + 0 ]; + // end inline asm + .loc 1 106 19 // l2norm.py:106:19 + // begin inline asm + mov.u32 %r130, %r5; + mov.u32 %r131, %r5; + mov.u32 %r132, %r5; + mov.u32 %r133, %r5; + @%p1 ld.global.v4.b32 { %r130, %r131, %r132, %r133 }, [ %rd35 + 0 ]; + // end inline asm + mov.b32 {%rs310, %rs311}, %r130; + mov.b32 {%rs312, %rs313}, %r131; + mov.b32 {%rs314, %rs315}, %r132; + mov.b32 {%rs316, %rs317}, %r133; + // begin inline asm + mov.u32 %r134, %r5; + mov.u32 %r135, %r5; + mov.u32 %r136, %r5; + mov.u32 %r137, %r5; + @%p2 ld.global.v4.b32 { %r134, %r135, %r136, %r137 }, [ %rd36 + 0 ]; + // end inline asm + mov.b32 {%rs318, %rs319}, %r134; + mov.b32 {%rs320, %rs321}, %r135; + mov.b32 {%rs322, %rs323}, %r136; + mov.b32 {%rs324, %rs325}, %r137; + // begin inline asm + mov.u32 %r138, %r5; + mov.u32 %r139, %r5; + mov.u32 %r140, %r5; + mov.u32 %r141, %r5; + @%p6 ld.global.v4.b32 { %r138, %r139, %r140, %r141 }, [ %rd37 + 0 ]; + // end inline asm + mov.b32 {%rs326, %rs327}, %r138; + mov.b32 {%rs328, %rs329}, %r139; + mov.b32 {%rs330, %rs331}, %r140; + mov.b32 {%rs332, %rs333}, %r141; + // begin inline asm + mov.u32 %r142, %r5; + mov.u32 %r143, %r5; + mov.u32 %r144, %r5; + mov.u32 %r145, %r5; + @%p7 ld.global.v4.b32 { %r142, %r143, %r144, %r145 }, [ %rd38 + 0 ]; + // end inline asm + mov.b32 {%rs334, %rs335}, %r142; + mov.b32 {%rs336, %rs337}, %r143; + mov.b32 {%rs338, %rs339}, %r144; + mov.b32 {%rs340, %rs341}, %r145; + // begin inline asm + mov.u32 %r146, %r5; + mov.u32 %r147, %r5; + mov.u32 %r148, %r5; + mov.u32 %r149, %r5; + @%p8 ld.global.v4.b32 { %r146, %r147, %r148, %r149 }, [ %rd39 + 0 ]; + // end inline asm + mov.b32 {%rs342, %rs343}, %r146; + mov.b32 {%rs344, %rs345}, %r147; + mov.b32 {%rs346, %rs347}, %r148; + mov.b32 {%rs348, %rs349}, %r149; + // begin inline asm + mov.u32 %r150, %r5; + mov.u32 %r151, %r5; + mov.u32 %r152, %r5; + mov.u32 %r153, %r5; + @%p9 ld.global.v4.b32 { %r150, %r151, %r152, %r153 }, [ %rd40 + 0 ]; + // end inline asm + mov.b32 {%rs350, %rs351}, %r150; + mov.b32 {%rs352, %rs353}, %r151; + mov.b32 {%rs354, %rs355}, %r152; + mov.b32 {%rs356, %rs357}, %r153; + // begin inline asm + mov.u32 %r154, %r5; + mov.u32 %r155, %r5; + mov.u32 %r156, %r5; + mov.u32 %r157, %r5; + @%p10 ld.global.v4.b32 { %r154, %r155, %r156, %r157 }, [ %rd41 + 0 ]; + // end inline asm + mov.b32 {%rs358, %rs359}, %r154; + mov.b32 {%rs360, %rs361}, %r155; + mov.b32 {%rs362, %rs363}, %r156; + mov.b32 {%rs364, %rs365}, %r157; + // begin inline asm + mov.u32 %r158, %r5; + mov.u32 %r159, %r5; + mov.u32 %r160, %r5; + mov.u32 %r161, %r5; + @%p11 ld.global.v4.b32 { %r158, %r159, %r160, %r161 }, [ %rd42 + 0 ]; + // end inline asm + mov.b32 {%rs366, %rs367}, %r158; + mov.b32 {%rs368, %rs369}, %r159; + mov.b32 {%rs370, %rs371}, %r160; + mov.b32 {%rs372, %rs373}, %r161; + // begin inline asm + mov.u32 %r162, %r5; + mov.u32 %r163, %r5; + mov.u32 %r164, %r5; + mov.u32 %r165, %r5; + @%p12 ld.global.v4.b32 { %r162, %r163, %r164, %r165 }, [ %rd43 + 0 ]; + // end inline asm + mov.b32 {%rs374, %rs375}, %r162; + mov.b32 {%rs376, %rs377}, %r163; + mov.b32 {%rs378, %rs379}, %r164; + mov.b32 {%rs380, %rs381}, %r165; + // begin inline asm + mov.u32 %r166, %r5; + mov.u32 %r167, %r5; + mov.u32 %r168, %r5; + mov.u32 %r169, %r5; + @%p3 ld.global.v4.b32 { %r166, %r167, %r168, %r169 }, [ %rd44 + 0 ]; + // end inline asm + mov.b32 {%rs382, %rs383}, %r166; + mov.b32 {%rs384, %rs385}, %r167; + mov.b32 {%rs386, %rs387}, %r168; + mov.b32 {%rs388, %rs389}, %r169; + // begin inline asm + mov.u32 %r170, %r5; + mov.u32 %r171, %r5; + mov.u32 %r172, %r5; + mov.u32 %r173, %r5; + @%p13 ld.global.v4.b32 { %r170, %r171, %r172, %r173 }, [ %rd45 + 0 ]; + // end inline asm + mov.b32 {%rs390, %rs391}, %r170; + mov.b32 {%rs392, %rs393}, %r171; + mov.b32 {%rs394, %rs395}, %r172; + mov.b32 {%rs396, %rs397}, %r173; + // begin inline asm + mov.u32 %r174, %r5; + mov.u32 %r175, %r5; + mov.u32 %r176, %r5; + mov.u32 %r177, %r5; + @%p14 ld.global.v4.b32 { %r174, %r175, %r176, %r177 }, [ %rd46 + 0 ]; + // end inline asm + mov.b32 {%rs398, %rs399}, %r174; + mov.b32 {%rs400, %rs401}, %r175; + mov.b32 {%rs402, %rs403}, %r176; + mov.b32 {%rs404, %rs405}, %r177; + // begin inline asm + mov.u32 %r178, %r5; + mov.u32 %r179, %r5; + mov.u32 %r180, %r5; + mov.u32 %r181, %r5; + @%p15 ld.global.v4.b32 { %r178, %r179, %r180, %r181 }, [ %rd47 + 0 ]; + // end inline asm + mov.b32 {%rs406, %rs407}, %r178; + mov.b32 {%rs408, %rs409}, %r179; + mov.b32 {%rs410, %rs411}, %r180; + mov.b32 {%rs412, %rs413}, %r181; + // begin inline asm + mov.u32 %r182, %r5; + mov.u32 %r183, %r5; + mov.u32 %r184, %r5; + mov.u32 %r185, %r5; + @%p16 ld.global.v4.b32 { %r182, %r183, %r184, %r185 }, [ %rd48 + 0 ]; + // end inline asm + mov.b32 {%rs414, %rs415}, %r182; + mov.b32 {%rs416, %rs417}, %r183; + mov.b32 {%rs418, %rs419}, %r184; + mov.b32 {%rs420, %rs421}, %r185; + // begin inline asm + mov.u32 %r186, %r5; + mov.u32 %r187, %r5; + mov.u32 %r188, %r5; + mov.u32 %r189, %r5; + @%p17 ld.global.v4.b32 { %r186, %r187, %r188, %r189 }, [ %rd49 + 0 ]; + // end inline asm + mov.b32 {%rs422, %rs423}, %r186; + mov.b32 {%rs424, %rs425}, %r187; + mov.b32 {%rs426, %rs427}, %r188; + mov.b32 {%rs428, %rs429}, %r189; + // begin inline asm + mov.u32 %r190, %r5; + mov.u32 %r191, %r5; + mov.u32 %r192, %r5; + mov.u32 %r193, %r5; + @%p18 ld.global.v4.b32 { %r190, %r191, %r192, %r193 }, [ %rd50 + 0 ]; + // end inline asm + mov.b32 {%rs430, %rs431}, %r190; + mov.b32 {%rs432, %rs433}, %r191; + mov.b32 {%rs434, %rs435}, %r192; + mov.b32 {%rs436, %rs437}, %r193; + // begin inline asm + mov.u32 %r194, %r5; + mov.u32 %r195, %r5; + mov.u32 %r196, %r5; + mov.u32 %r197, %r5; + @%p19 ld.global.v4.b32 { %r194, %r195, %r196, %r197 }, [ %rd51 + 0 ]; + // end inline asm + mov.b32 {%rs438, %rs439}, %r194; + mov.b32 {%rs440, %rs441}, %r195; + mov.b32 {%rs442, %rs443}, %r196; + mov.b32 {%rs444, %rs445}, %r197; + // begin inline asm + mov.u32 %r198, %r5; + mov.u32 %r199, %r5; + mov.u32 %r200, %r5; + mov.u32 %r201, %r5; + @%p20 ld.global.v4.b32 { %r198, %r199, %r200, %r201 }, [ %rd52 + 0 ]; + // end inline asm + mov.b32 {%rs446, %rs447}, %r198; + mov.b32 {%rs448, %rs449}, %r199; + mov.b32 {%rs450, %rs451}, %r200; + mov.b32 {%rs452, %rs453}, %r201; + // begin inline asm + mov.u32 %r202, %r5; + mov.u32 %r203, %r5; + mov.u32 %r204, %r5; + mov.u32 %r205, %r5; + @%p21 ld.global.v4.b32 { %r202, %r203, %r204, %r205 }, [ %rd53 + 0 ]; + // end inline asm + mov.b32 {%rs454, %rs455}, %r202; + mov.b32 {%rs456, %rs457}, %r203; + mov.b32 {%rs458, %rs459}, %r204; + mov.b32 {%rs460, %rs461}, %r205; + // begin inline asm + mov.u32 %r206, %r5; + mov.u32 %r207, %r5; + mov.u32 %r208, %r5; + mov.u32 %r209, %r5; + @%p22 ld.global.v4.b32 { %r206, %r207, %r208, %r209 }, [ %rd54 + 0 ]; + // end inline asm + mov.b32 {%rs462, %rs463}, %r206; + mov.b32 {%rs464, %rs465}, %r207; + mov.b32 {%rs466, %rs467}, %r208; + mov.b32 {%rs468, %rs469}, %r209; + // begin inline asm + mov.u32 %r210, %r5; + mov.u32 %r211, %r5; + mov.u32 %r212, %r5; + mov.u32 %r213, %r5; + @%p23 ld.global.v4.b32 { %r210, %r211, %r212, %r213 }, [ %rd55 + 0 ]; + // end inline asm + mov.b32 {%rs470, %rs471}, %r210; + mov.b32 {%rs472, %rs473}, %r211; + mov.b32 {%rs474, %rs475}, %r212; + mov.b32 {%rs476, %rs477}, %r213; + // begin inline asm + mov.u32 %r214, %r5; + mov.u32 %r215, %r5; + mov.u32 %r216, %r5; + mov.u32 %r217, %r5; + @%p24 ld.global.v4.b32 { %r214, %r215, %r216, %r217 }, [ %rd56 + 0 ]; + // end inline asm + mov.b32 {%rs478, %rs479}, %r214; + mov.b32 {%rs480, %rs481}, %r215; + mov.b32 {%rs482, %rs483}, %r216; + mov.b32 {%rs484, %rs485}, %r217; + // begin inline asm + mov.u32 %r218, %r5; + mov.u32 %r219, %r5; + mov.u32 %r220, %r5; + mov.u32 %r221, %r5; + @%p25 ld.global.v4.b32 { %r218, %r219, %r220, %r221 }, [ %rd57 + 0 ]; + // end inline asm + mov.b32 {%rs486, %rs487}, %r218; + mov.b32 {%rs488, %rs489}, %r219; + mov.b32 {%rs490, %rs491}, %r220; + mov.b32 {%rs492, %rs493}, %r221; + // begin inline asm + mov.u32 %r222, %r5; + mov.u32 %r223, %r5; + mov.u32 %r224, %r5; + mov.u32 %r225, %r5; + @%p26 ld.global.v4.b32 { %r222, %r223, %r224, %r225 }, [ %rd58 + 0 ]; + // end inline asm + mov.b32 {%rs494, %rs495}, %r222; + mov.b32 {%rs496, %rs497}, %r223; + mov.b32 {%rs498, %rs499}, %r224; + mov.b32 {%rs500, %rs501}, %r225; + // begin inline asm + mov.u32 %r226, %r5; + mov.u32 %r227, %r5; + mov.u32 %r228, %r5; + mov.u32 %r229, %r5; + @%p27 ld.global.v4.b32 { %r226, %r227, %r228, %r229 }, [ %rd59 + 0 ]; + // end inline asm + mov.b32 {%rs502, %rs503}, %r226; + mov.b32 {%rs504, %rs505}, %r227; + mov.b32 {%rs506, %rs507}, %r228; + mov.b32 {%rs508, %rs509}, %r229; + // begin inline asm + mov.u32 %r230, %r5; + mov.u32 %r231, %r5; + mov.u32 %r232, %r5; + mov.u32 %r233, %r5; + @%p28 ld.global.v4.b32 { %r230, %r231, %r232, %r233 }, [ %rd60 + 0 ]; + // end inline asm + mov.b32 {%rs510, %rs511}, %r230; + mov.b32 {%rs512, %rs513}, %r231; + mov.b32 {%rs514, %rs515}, %r232; + mov.b32 {%rs516, %rs517}, %r233; + // begin inline asm + mov.u32 %r234, %r5; + mov.u32 %r235, %r5; + mov.u32 %r236, %r5; + mov.u32 %r237, %r5; + @%p29 ld.global.v4.b32 { %r234, %r235, %r236, %r237 }, [ %rd61 + 0 ]; + // end inline asm + mov.b32 {%rs518, %rs519}, %r234; + mov.b32 {%rs520, %rs521}, %r235; + mov.b32 {%rs522, %rs523}, %r236; + mov.b32 {%rs524, %rs525}, %r237; + // begin inline asm + mov.u32 %r238, %r5; + mov.u32 %r239, %r5; + mov.u32 %r240, %r5; + mov.u32 %r241, %r5; + @%p30 ld.global.v4.b32 { %r238, %r239, %r240, %r241 }, [ %rd62 + 0 ]; + // end inline asm + mov.b32 {%rs526, %rs527}, %r238; + mov.b32 {%rs528, %rs529}, %r239; + mov.b32 {%rs530, %rs531}, %r240; + mov.b32 {%rs532, %rs533}, %r241; + // begin inline asm + mov.u32 %r242, %r5; + mov.u32 %r243, %r5; + mov.u32 %r244, %r5; + mov.u32 %r245, %r5; + @%p31 ld.global.v4.b32 { %r242, %r243, %r244, %r245 }, [ %rd63 + 0 ]; + // end inline asm + mov.b32 {%rs534, %rs535}, %r242; + mov.b32 {%rs536, %rs537}, %r243; + mov.b32 {%rs538, %rs539}, %r244; + mov.b32 {%rs540, %rs541}, %r245; + // begin inline asm + mov.u32 %r246, %r5; + mov.u32 %r247, %r5; + mov.u32 %r248, %r5; + mov.u32 %r249, %r5; + @%p32 ld.global.v4.b32 { %r246, %r247, %r248, %r249 }, [ %rd64 + 0 ]; + // end inline asm + mov.b32 {%rs542, %rs543}, %r246; + mov.b32 {%rs544, %rs545}, %r247; + mov.b32 {%rs546, %rs547}, %r248; + mov.b32 {%rs548, %rs549}, %r249; + // begin inline asm + mov.u32 %r250, %r5; + mov.u32 %r251, %r5; + mov.u32 %r252, %r5; + mov.u32 %r253, %r5; + @%p33 ld.global.v4.b32 { %r250, %r251, %r252, %r253 }, [ %rd65 + 0 ]; + // end inline asm + mov.b32 {%rs550, %rs551}, %r250; + mov.b32 {%rs552, %rs553}, %r251; + mov.b32 {%rs554, %rs555}, %r252; + mov.b32 {%rs556, %rs557}, %r253; + // begin inline asm + mov.u32 %r254, %r5; + mov.u32 %r255, %r5; + mov.u32 %r256, %r5; + mov.u32 %r257, %r5; + @%p34 ld.global.v4.b32 { %r254, %r255, %r256, %r257 }, [ %rd66 + 0 ]; + // end inline asm + mov.b32 {%rs558, %rs559}, %r254; + mov.b32 {%rs560, %rs561}, %r255; + mov.b32 {%rs562, %rs563}, %r256; + mov.b32 {%rs564, %rs565}, %r257; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r717, %rs310; + cvt.f32.bf16 %r718, %rs311; + cvt.f32.bf16 %r719, %rs312; + cvt.f32.bf16 %r720, %rs313; + cvt.f32.bf16 %r721, %rs314; + cvt.f32.bf16 %r722, %rs315; + cvt.f32.bf16 %r723, %rs316; + cvt.f32.bf16 %r724, %rs317; + cvt.f32.bf16 %r725, %rs318; + cvt.f32.bf16 %r726, %rs319; + cvt.f32.bf16 %r727, %rs320; + cvt.f32.bf16 %r728, %rs321; + cvt.f32.bf16 %r729, %rs322; + cvt.f32.bf16 %r730, %rs323; + cvt.f32.bf16 %r731, %rs324; + cvt.f32.bf16 %r732, %rs325; + cvt.f32.bf16 %r733, %rs326; + cvt.f32.bf16 %r734, %rs327; + cvt.f32.bf16 %r735, %rs328; + cvt.f32.bf16 %r736, %rs329; + cvt.f32.bf16 %r737, %rs330; + cvt.f32.bf16 %r738, %rs331; + cvt.f32.bf16 %r739, %rs332; + cvt.f32.bf16 %r740, %rs333; + cvt.f32.bf16 %r741, %rs334; + cvt.f32.bf16 %r742, %rs335; + cvt.f32.bf16 %r743, %rs336; + cvt.f32.bf16 %r744, %rs337; + cvt.f32.bf16 %r745, %rs338; + cvt.f32.bf16 %r746, %rs339; + cvt.f32.bf16 %r747, %rs340; + cvt.f32.bf16 %r748, %rs341; + cvt.f32.bf16 %r749, %rs342; + cvt.f32.bf16 %r750, %rs343; + cvt.f32.bf16 %r751, %rs344; + cvt.f32.bf16 %r752, %rs345; + cvt.f32.bf16 %r753, %rs346; + cvt.f32.bf16 %r754, %rs347; + cvt.f32.bf16 %r755, %rs348; + cvt.f32.bf16 %r756, %rs349; + cvt.f32.bf16 %r757, %rs350; + cvt.f32.bf16 %r758, %rs351; + cvt.f32.bf16 %r759, %rs352; + cvt.f32.bf16 %r760, %rs353; + cvt.f32.bf16 %r761, %rs354; + cvt.f32.bf16 %r762, %rs355; + cvt.f32.bf16 %r763, %rs356; + cvt.f32.bf16 %r764, %rs357; + cvt.f32.bf16 %r765, %rs358; + cvt.f32.bf16 %r766, %rs359; + cvt.f32.bf16 %r767, %rs360; + cvt.f32.bf16 %r768, %rs361; + cvt.f32.bf16 %r769, %rs362; + cvt.f32.bf16 %r770, %rs363; + cvt.f32.bf16 %r771, %rs364; + cvt.f32.bf16 %r772, %rs365; + cvt.f32.bf16 %r773, %rs366; + cvt.f32.bf16 %r774, %rs367; + cvt.f32.bf16 %r775, %rs368; + cvt.f32.bf16 %r776, %rs369; + cvt.f32.bf16 %r777, %rs370; + cvt.f32.bf16 %r778, %rs371; + cvt.f32.bf16 %r779, %rs372; + cvt.f32.bf16 %r780, %rs373; + cvt.f32.bf16 %r781, %rs374; + cvt.f32.bf16 %r782, %rs375; + cvt.f32.bf16 %r783, %rs376; + cvt.f32.bf16 %r784, %rs377; + cvt.f32.bf16 %r785, %rs378; + cvt.f32.bf16 %r786, %rs379; + cvt.f32.bf16 %r787, %rs380; + cvt.f32.bf16 %r788, %rs381; + cvt.f32.bf16 %r789, %rs382; + cvt.f32.bf16 %r790, %rs383; + cvt.f32.bf16 %r791, %rs384; + cvt.f32.bf16 %r792, %rs385; + cvt.f32.bf16 %r793, %rs386; + cvt.f32.bf16 %r794, %rs387; + cvt.f32.bf16 %r795, %rs388; + cvt.f32.bf16 %r796, %rs389; + cvt.f32.bf16 %r797, %rs390; + cvt.f32.bf16 %r798, %rs391; + cvt.f32.bf16 %r799, %rs392; + cvt.f32.bf16 %r800, %rs393; + cvt.f32.bf16 %r801, %rs394; + cvt.f32.bf16 %r802, %rs395; + cvt.f32.bf16 %r803, %rs396; + cvt.f32.bf16 %r804, %rs397; + cvt.f32.bf16 %r805, %rs398; + cvt.f32.bf16 %r806, %rs399; + cvt.f32.bf16 %r807, %rs400; + cvt.f32.bf16 %r808, %rs401; + cvt.f32.bf16 %r809, %rs402; + cvt.f32.bf16 %r810, %rs403; + cvt.f32.bf16 %r811, %rs404; + cvt.f32.bf16 %r812, %rs405; + cvt.f32.bf16 %r813, %rs406; + cvt.f32.bf16 %r814, %rs407; + cvt.f32.bf16 %r815, %rs408; + cvt.f32.bf16 %r816, %rs409; + cvt.f32.bf16 %r817, %rs410; + cvt.f32.bf16 %r818, %rs411; + cvt.f32.bf16 %r819, %rs412; + cvt.f32.bf16 %r820, %rs413; + cvt.f32.bf16 %r821, %rs414; + cvt.f32.bf16 %r822, %rs415; + cvt.f32.bf16 %r823, %rs416; + cvt.f32.bf16 %r824, %rs417; + cvt.f32.bf16 %r825, %rs418; + cvt.f32.bf16 %r826, %rs419; + cvt.f32.bf16 %r827, %rs420; + cvt.f32.bf16 %r828, %rs421; + cvt.f32.bf16 %r829, %rs422; + cvt.f32.bf16 %r830, %rs423; + cvt.f32.bf16 %r831, %rs424; + cvt.f32.bf16 %r832, %rs425; + cvt.f32.bf16 %r833, %rs426; + cvt.f32.bf16 %r834, %rs427; + cvt.f32.bf16 %r835, %rs428; + cvt.f32.bf16 %r836, %rs429; + cvt.f32.bf16 %r837, %rs430; + cvt.f32.bf16 %r838, %rs431; + cvt.f32.bf16 %r839, %rs432; + cvt.f32.bf16 %r840, %rs433; + cvt.f32.bf16 %r841, %rs434; + cvt.f32.bf16 %r842, %rs435; + cvt.f32.bf16 %r843, %rs436; + cvt.f32.bf16 %r844, %rs437; + cvt.f32.bf16 %r845, %rs438; + cvt.f32.bf16 %r846, %rs439; + cvt.f32.bf16 %r847, %rs440; + cvt.f32.bf16 %r848, %rs441; + cvt.f32.bf16 %r849, %rs442; + cvt.f32.bf16 %r850, %rs443; + cvt.f32.bf16 %r851, %rs444; + cvt.f32.bf16 %r852, %rs445; + cvt.f32.bf16 %r853, %rs446; + cvt.f32.bf16 %r854, %rs447; + cvt.f32.bf16 %r855, %rs448; + cvt.f32.bf16 %r856, %rs449; + cvt.f32.bf16 %r857, %rs450; + cvt.f32.bf16 %r858, %rs451; + cvt.f32.bf16 %r859, %rs452; + cvt.f32.bf16 %r860, %rs453; + cvt.f32.bf16 %r861, %rs454; + cvt.f32.bf16 %r862, %rs455; + cvt.f32.bf16 %r863, %rs456; + cvt.f32.bf16 %r864, %rs457; + cvt.f32.bf16 %r865, %rs458; + cvt.f32.bf16 %r866, %rs459; + cvt.f32.bf16 %r867, %rs460; + cvt.f32.bf16 %r868, %rs461; + cvt.f32.bf16 %r869, %rs462; + cvt.f32.bf16 %r870, %rs463; + cvt.f32.bf16 %r871, %rs464; + cvt.f32.bf16 %r872, %rs465; + cvt.f32.bf16 %r873, %rs466; + cvt.f32.bf16 %r874, %rs467; + cvt.f32.bf16 %r875, %rs468; + cvt.f32.bf16 %r876, %rs469; + cvt.f32.bf16 %r877, %rs470; + cvt.f32.bf16 %r878, %rs471; + cvt.f32.bf16 %r879, %rs472; + cvt.f32.bf16 %r880, %rs473; + cvt.f32.bf16 %r881, %rs474; + cvt.f32.bf16 %r882, %rs475; + cvt.f32.bf16 %r883, %rs476; + cvt.f32.bf16 %r884, %rs477; + cvt.f32.bf16 %r885, %rs478; + cvt.f32.bf16 %r886, %rs479; + cvt.f32.bf16 %r887, %rs480; + cvt.f32.bf16 %r888, %rs481; + cvt.f32.bf16 %r889, %rs482; + cvt.f32.bf16 %r890, %rs483; + cvt.f32.bf16 %r891, %rs484; + cvt.f32.bf16 %r892, %rs485; + cvt.f32.bf16 %r893, %rs486; + cvt.f32.bf16 %r894, %rs487; + cvt.f32.bf16 %r895, %rs488; + cvt.f32.bf16 %r896, %rs489; + cvt.f32.bf16 %r897, %rs490; + cvt.f32.bf16 %r898, %rs491; + cvt.f32.bf16 %r899, %rs492; + cvt.f32.bf16 %r900, %rs493; + cvt.f32.bf16 %r901, %rs494; + cvt.f32.bf16 %r902, %rs495; + cvt.f32.bf16 %r903, %rs496; + cvt.f32.bf16 %r904, %rs497; + cvt.f32.bf16 %r905, %rs498; + cvt.f32.bf16 %r906, %rs499; + cvt.f32.bf16 %r907, %rs500; + cvt.f32.bf16 %r908, %rs501; + cvt.f32.bf16 %r909, %rs502; + cvt.f32.bf16 %r910, %rs503; + cvt.f32.bf16 %r911, %rs504; + cvt.f32.bf16 %r912, %rs505; + cvt.f32.bf16 %r913, %rs506; + cvt.f32.bf16 %r914, %rs507; + cvt.f32.bf16 %r915, %rs508; + cvt.f32.bf16 %r916, %rs509; + cvt.f32.bf16 %r917, %rs510; + cvt.f32.bf16 %r918, %rs511; + cvt.f32.bf16 %r919, %rs512; + cvt.f32.bf16 %r920, %rs513; + cvt.f32.bf16 %r921, %rs514; + cvt.f32.bf16 %r922, %rs515; + cvt.f32.bf16 %r923, %rs516; + cvt.f32.bf16 %r924, %rs517; + cvt.f32.bf16 %r925, %rs518; + cvt.f32.bf16 %r926, %rs519; + cvt.f32.bf16 %r927, %rs520; + cvt.f32.bf16 %r928, %rs521; + cvt.f32.bf16 %r929, %rs522; + cvt.f32.bf16 %r930, %rs523; + cvt.f32.bf16 %r931, %rs524; + cvt.f32.bf16 %r932, %rs525; + cvt.f32.bf16 %r933, %rs526; + cvt.f32.bf16 %r934, %rs527; + cvt.f32.bf16 %r935, %rs528; + cvt.f32.bf16 %r936, %rs529; + cvt.f32.bf16 %r937, %rs530; + cvt.f32.bf16 %r938, %rs531; + cvt.f32.bf16 %r939, %rs532; + cvt.f32.bf16 %r940, %rs533; + cvt.f32.bf16 %r941, %rs534; + cvt.f32.bf16 %r942, %rs535; + cvt.f32.bf16 %r943, %rs536; + cvt.f32.bf16 %r944, %rs537; + cvt.f32.bf16 %r945, %rs538; + cvt.f32.bf16 %r946, %rs539; + cvt.f32.bf16 %r947, %rs540; + cvt.f32.bf16 %r948, %rs541; + cvt.f32.bf16 %r949, %rs542; + cvt.f32.bf16 %r950, %rs543; + cvt.f32.bf16 %r951, %rs544; + cvt.f32.bf16 %r952, %rs545; + cvt.f32.bf16 %r953, %rs546; + cvt.f32.bf16 %r954, %rs547; + cvt.f32.bf16 %r955, %rs548; + cvt.f32.bf16 %r956, %rs549; + cvt.f32.bf16 %r957, %rs550; + cvt.f32.bf16 %r958, %rs551; + cvt.f32.bf16 %r959, %rs552; + cvt.f32.bf16 %r960, %rs553; + cvt.f32.bf16 %r961, %rs554; + cvt.f32.bf16 %r962, %rs555; + cvt.f32.bf16 %r963, %rs556; + cvt.f32.bf16 %r964, %rs557; + cvt.f32.bf16 %r965, %rs558; + cvt.f32.bf16 %r966, %rs559; + cvt.f32.bf16 %r967, %rs560; + cvt.f32.bf16 %r968, %rs561; + cvt.f32.bf16 %r969, %rs562; + cvt.f32.bf16 %r970, %rs563; + cvt.f32.bf16 %r971, %rs564; + cvt.f32.bf16 %r972, %rs565; + .loc 1 107 25 // l2norm.py:107:25 + and.b32 %r973, %r395, 60; + shl.b32 %r974, %r391, 6; + and.b32 %r975, %r974, 64; + and.b32 %r976, %r391, 32; + shl.b32 %r977, %r976, 2; + or.b32 %r978, %r975, %r977; + or.b32 %r979, %r978, %r973; + mov.b32 %r980, global_smem; + add.s32 %r258, %r980, %r979; + // begin inline asm + st.shared.b32 [ %r258 + 0 ], %r259; + // end inline asm + xor.b32 %r981, %r979, 64; + add.s32 %r982, %r980, %r981; + add.s32 %r260, %r982, 256; + // begin inline asm + st.shared.b32 [ %r260 + 0 ], %r261; + // end inline asm + bar.sync 0; + bfe.s32 %r983, %r391, 4, 1; + and.b32 %r984, %r983, 320; + shl.b32 %r985, %r976, 1; + xor.b32 %r986, %r984, %r985; + add.s32 %r987, %r980, %r986; + ld.shared.v4.b32 {%r988, %r989, %r990, %r991}, [%r987]; + ld.shared.v4.b32 {%r992, %r993, %r994, %r995}, [%r987+16]; + ld.shared.v4.b32 {%r996, %r997, %r998, %r999}, [%r987+32]; + ld.shared.v4.b32 {%r1000, %r1001, %r1002, %r1003}, [%r987+48]; + ld.shared.v4.b32 {%r1004, %r1005, %r1006, %r1007}, [%r987+128]; + ld.shared.v4.b32 {%r1008, %r1009, %r1010, %r1011}, [%r987+144]; + ld.shared.v4.b32 {%r1012, %r1013, %r1014, %r1015}, [%r987+160]; + ld.shared.v4.b32 {%r1016, %r1017, %r1018, %r1019}, [%r987+176]; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r1020, %r990, %r733; + mul.f32 %r1021, %r990, %r734; + mul.f32 %r1022, %r990, %r735; + mul.f32 %r1023, %r990, %r736; + mul.f32 %r1024, %r990, %r737; + mul.f32 %r1025, %r990, %r738; + mul.f32 %r1026, %r990, %r739; + mul.f32 %r1027, %r990, %r740; + mul.f32 %r1028, %r991, %r741; + mul.f32 %r1029, %r991, %r742; + mul.f32 %r1030, %r991, %r743; + mul.f32 %r1031, %r991, %r744; + mul.f32 %r1032, %r991, %r745; + mul.f32 %r1033, %r991, %r746; + mul.f32 %r1034, %r991, %r747; + mul.f32 %r1035, %r991, %r748; + mul.f32 %r1036, %r992, %r749; + mul.f32 %r1037, %r992, %r750; + mul.f32 %r1038, %r992, %r751; + mul.f32 %r1039, %r992, %r752; + mul.f32 %r1040, %r992, %r753; + mul.f32 %r1041, %r992, %r754; + mul.f32 %r1042, %r992, %r755; + mul.f32 %r1043, %r992, %r756; + mul.f32 %r1044, %r993, %r757; + mul.f32 %r1045, %r993, %r758; + mul.f32 %r1046, %r993, %r759; + mul.f32 %r1047, %r993, %r760; + mul.f32 %r1048, %r993, %r761; + mul.f32 %r1049, %r993, %r762; + mul.f32 %r1050, %r993, %r763; + mul.f32 %r1051, %r993, %r764; + mul.f32 %r1052, %r994, %r765; + mul.f32 %r1053, %r994, %r766; + mul.f32 %r1054, %r994, %r767; + mul.f32 %r1055, %r994, %r768; + mul.f32 %r1056, %r994, %r769; + mul.f32 %r1057, %r994, %r770; + mul.f32 %r1058, %r994, %r771; + mul.f32 %r1059, %r994, %r772; + mul.f32 %r1060, %r995, %r773; + mul.f32 %r1061, %r995, %r774; + mul.f32 %r1062, %r995, %r775; + mul.f32 %r1063, %r995, %r776; + mul.f32 %r1064, %r995, %r777; + mul.f32 %r1065, %r995, %r778; + mul.f32 %r1066, %r995, %r779; + mul.f32 %r1067, %r995, %r780; + mul.f32 %r1068, %r996, %r781; + mul.f32 %r1069, %r996, %r782; + mul.f32 %r1070, %r996, %r783; + mul.f32 %r1071, %r996, %r784; + mul.f32 %r1072, %r996, %r785; + mul.f32 %r1073, %r996, %r786; + mul.f32 %r1074, %r996, %r787; + mul.f32 %r1075, %r996, %r788; + mul.f32 %r1076, %r997, %r789; + mul.f32 %r1077, %r997, %r790; + mul.f32 %r1078, %r997, %r791; + mul.f32 %r1079, %r997, %r792; + mul.f32 %r1080, %r997, %r793; + mul.f32 %r1081, %r997, %r794; + mul.f32 %r1082, %r997, %r795; + mul.f32 %r1083, %r997, %r796; + mul.f32 %r1084, %r998, %r797; + mul.f32 %r1085, %r998, %r798; + mul.f32 %r1086, %r998, %r799; + mul.f32 %r1087, %r998, %r800; + mul.f32 %r1088, %r998, %r801; + mul.f32 %r1089, %r998, %r802; + mul.f32 %r1090, %r998, %r803; + mul.f32 %r1091, %r998, %r804; + mul.f32 %r1092, %r999, %r805; + mul.f32 %r1093, %r999, %r806; + mul.f32 %r1094, %r999, %r807; + mul.f32 %r1095, %r999, %r808; + mul.f32 %r1096, %r999, %r809; + mul.f32 %r1097, %r999, %r810; + mul.f32 %r1098, %r999, %r811; + mul.f32 %r1099, %r999, %r812; + mul.f32 %r1100, %r1000, %r813; + mul.f32 %r1101, %r1000, %r814; + mul.f32 %r1102, %r1000, %r815; + mul.f32 %r1103, %r1000, %r816; + mul.f32 %r1104, %r1000, %r817; + mul.f32 %r1105, %r1000, %r818; + mul.f32 %r1106, %r1000, %r819; + mul.f32 %r1107, %r1000, %r820; + mul.f32 %r1108, %r1001, %r821; + mul.f32 %r1109, %r1001, %r822; + mul.f32 %r1110, %r1001, %r823; + mul.f32 %r1111, %r1001, %r824; + mul.f32 %r1112, %r1001, %r825; + mul.f32 %r1113, %r1001, %r826; + mul.f32 %r1114, %r1001, %r827; + mul.f32 %r1115, %r1001, %r828; + mul.f32 %r1116, %r1002, %r829; + mul.f32 %r1117, %r1002, %r830; + mul.f32 %r1118, %r1002, %r831; + mul.f32 %r1119, %r1002, %r832; + mul.f32 %r1120, %r1002, %r833; + mul.f32 %r1121, %r1002, %r834; + mul.f32 %r1122, %r1002, %r835; + mul.f32 %r1123, %r1002, %r836; + mul.f32 %r1124, %r1003, %r837; + mul.f32 %r1125, %r1003, %r838; + mul.f32 %r1126, %r1003, %r839; + mul.f32 %r1127, %r1003, %r840; + mul.f32 %r1128, %r1003, %r841; + mul.f32 %r1129, %r1003, %r842; + mul.f32 %r1130, %r1003, %r843; + mul.f32 %r1131, %r1003, %r844; + mul.f32 %r1132, %r1004, %r845; + mul.f32 %r1133, %r1004, %r846; + mul.f32 %r1134, %r1004, %r847; + mul.f32 %r1135, %r1004, %r848; + mul.f32 %r1136, %r1004, %r849; + mul.f32 %r1137, %r1004, %r850; + mul.f32 %r1138, %r1004, %r851; + mul.f32 %r1139, %r1004, %r852; + mul.f32 %r1140, %r1005, %r853; + mul.f32 %r1141, %r1005, %r854; + mul.f32 %r1142, %r1005, %r855; + mul.f32 %r1143, %r1005, %r856; + mul.f32 %r1144, %r1005, %r857; + mul.f32 %r1145, %r1005, %r858; + mul.f32 %r1146, %r1005, %r859; + mul.f32 %r1147, %r1005, %r860; + mul.f32 %r1148, %r1006, %r861; + mul.f32 %r1149, %r1006, %r862; + mul.f32 %r1150, %r1006, %r863; + mul.f32 %r1151, %r1006, %r864; + mul.f32 %r1152, %r1006, %r865; + mul.f32 %r1153, %r1006, %r866; + mul.f32 %r1154, %r1006, %r867; + mul.f32 %r1155, %r1006, %r868; + mul.f32 %r1156, %r1007, %r869; + mul.f32 %r1157, %r1007, %r870; + mul.f32 %r1158, %r1007, %r871; + mul.f32 %r1159, %r1007, %r872; + mul.f32 %r1160, %r1007, %r873; + mul.f32 %r1161, %r1007, %r874; + mul.f32 %r1162, %r1007, %r875; + mul.f32 %r1163, %r1007, %r876; + mul.f32 %r1164, %r1008, %r877; + mul.f32 %r1165, %r1008, %r878; + mul.f32 %r1166, %r1008, %r879; + mul.f32 %r1167, %r1008, %r880; + mul.f32 %r1168, %r1008, %r881; + mul.f32 %r1169, %r1008, %r882; + mul.f32 %r1170, %r1008, %r883; + mul.f32 %r1171, %r1008, %r884; + mul.f32 %r1172, %r1009, %r885; + mul.f32 %r1173, %r1009, %r886; + mul.f32 %r1174, %r1009, %r887; + mul.f32 %r1175, %r1009, %r888; + mul.f32 %r1176, %r1009, %r889; + mul.f32 %r1177, %r1009, %r890; + mul.f32 %r1178, %r1009, %r891; + mul.f32 %r1179, %r1009, %r892; + mul.f32 %r1180, %r1010, %r893; + mul.f32 %r1181, %r1010, %r894; + mul.f32 %r1182, %r1010, %r895; + mul.f32 %r1183, %r1010, %r896; + mul.f32 %r1184, %r1010, %r897; + mul.f32 %r1185, %r1010, %r898; + mul.f32 %r1186, %r1010, %r899; + mul.f32 %r1187, %r1010, %r900; + mul.f32 %r1188, %r1011, %r901; + mul.f32 %r1189, %r1011, %r902; + mul.f32 %r1190, %r1011, %r903; + mul.f32 %r1191, %r1011, %r904; + mul.f32 %r1192, %r1011, %r905; + mul.f32 %r1193, %r1011, %r906; + mul.f32 %r1194, %r1011, %r907; + mul.f32 %r1195, %r1011, %r908; + mul.f32 %r1196, %r1012, %r909; + mul.f32 %r1197, %r1012, %r910; + mul.f32 %r1198, %r1012, %r911; + mul.f32 %r1199, %r1012, %r912; + mul.f32 %r1200, %r1012, %r913; + mul.f32 %r1201, %r1012, %r914; + mul.f32 %r1202, %r1012, %r915; + mul.f32 %r1203, %r1012, %r916; + mul.f32 %r1204, %r1013, %r917; + mul.f32 %r1205, %r1013, %r918; + mul.f32 %r1206, %r1013, %r919; + mul.f32 %r1207, %r1013, %r920; + mul.f32 %r1208, %r1013, %r921; + mul.f32 %r1209, %r1013, %r922; + mul.f32 %r1210, %r1013, %r923; + mul.f32 %r1211, %r1013, %r924; + mul.f32 %r1212, %r1014, %r925; + mul.f32 %r1213, %r1014, %r926; + mul.f32 %r1214, %r1014, %r927; + mul.f32 %r1215, %r1014, %r928; + mul.f32 %r1216, %r1014, %r929; + mul.f32 %r1217, %r1014, %r930; + mul.f32 %r1218, %r1014, %r931; + mul.f32 %r1219, %r1014, %r932; + mul.f32 %r1220, %r1015, %r933; + mul.f32 %r1221, %r1015, %r934; + mul.f32 %r1222, %r1015, %r935; + mul.f32 %r1223, %r1015, %r936; + mul.f32 %r1224, %r1015, %r937; + mul.f32 %r1225, %r1015, %r938; + mul.f32 %r1226, %r1015, %r939; + mul.f32 %r1227, %r1015, %r940; + mul.f32 %r1228, %r1016, %r941; + mul.f32 %r1229, %r1016, %r942; + mul.f32 %r1230, %r1016, %r943; + mul.f32 %r1231, %r1016, %r944; + mul.f32 %r1232, %r1016, %r945; + mul.f32 %r1233, %r1016, %r946; + mul.f32 %r1234, %r1016, %r947; + mul.f32 %r1235, %r1016, %r948; + mul.f32 %r1236, %r1017, %r949; + mul.f32 %r1237, %r1017, %r950; + mul.f32 %r1238, %r1017, %r951; + mul.f32 %r1239, %r1017, %r952; + mul.f32 %r1240, %r1017, %r953; + mul.f32 %r1241, %r1017, %r954; + mul.f32 %r1242, %r1017, %r955; + mul.f32 %r1243, %r1017, %r956; + mul.f32 %r1244, %r1018, %r957; + mul.f32 %r1245, %r1018, %r958; + mul.f32 %r1246, %r1018, %r959; + mul.f32 %r1247, %r1018, %r960; + mul.f32 %r1248, %r1018, %r961; + mul.f32 %r1249, %r1018, %r962; + mul.f32 %r1250, %r1018, %r963; + mul.f32 %r1251, %r1018, %r964; + mul.f32 %r1252, %r1019, %r965; + mul.f32 %r1253, %r1019, %r966; + mul.f32 %r1254, %r1019, %r967; + mul.f32 %r1255, %r1019, %r968; + mul.f32 %r1256, %r1019, %r969; + mul.f32 %r1257, %r1019, %r970; + mul.f32 %r1258, %r1019, %r971; + mul.f32 %r1259, %r1019, %r972; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r1260, %r460, %r718; + mul.f32 %r1261, %r468, %r726; + mul.f32 %r1262, %r476, %r734; + mul.f32 %r1263, %r484, %r742; + mul.f32 %r1264, %r492, %r750; + mul.f32 %r1265, %r500, %r758; + mul.f32 %r1266, %r508, %r766; + mul.f32 %r1267, %r516, %r774; + mul.f32 %r1268, %r524, %r782; + mul.f32 %r1269, %r532, %r790; + mul.f32 %r1270, %r540, %r798; + mul.f32 %r1271, %r548, %r806; + mul.f32 %r1272, %r556, %r814; + mul.f32 %r1273, %r564, %r822; + mul.f32 %r1274, %r572, %r830; + mul.f32 %r1275, %r580, %r838; + mul.f32 %r1276, %r588, %r846; + mul.f32 %r1277, %r596, %r854; + mul.f32 %r1278, %r604, %r862; + mul.f32 %r1279, %r612, %r870; + mul.f32 %r1280, %r620, %r878; + mul.f32 %r1281, %r628, %r886; + mul.f32 %r1282, %r636, %r894; + mul.f32 %r1283, %r644, %r902; + mul.f32 %r1284, %r652, %r910; + mul.f32 %r1285, %r660, %r918; + mul.f32 %r1286, %r668, %r926; + mul.f32 %r1287, %r676, %r934; + mul.f32 %r1288, %r684, %r942; + mul.f32 %r1289, %r692, %r950; + mul.f32 %r1290, %r700, %r958; + mul.f32 %r1291, %r708, %r966; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1292, %rs54, %rs310, %r1260; + fma.rn.f32.bf16 %r1293, %rs56, %rs312, %r1292; + fma.rn.f32.bf16 %r1294, %rs57, %rs313, %r1293; + fma.rn.f32.bf16 %r1295, %rs58, %rs314, %r1294; + fma.rn.f32.bf16 %r1296, %rs59, %rs315, %r1295; + fma.rn.f32.bf16 %r1297, %rs60, %rs316, %r1296; + fma.rn.f32.bf16 %r1298, %rs61, %rs317, %r1297; + fma.rn.f32.bf16 %r1299, %rs62, %rs318, %r1261; + fma.rn.f32.bf16 %r1300, %rs64, %rs320, %r1299; + fma.rn.f32.bf16 %r1301, %rs65, %rs321, %r1300; + fma.rn.f32.bf16 %r1302, %rs66, %rs322, %r1301; + fma.rn.f32.bf16 %r1303, %rs67, %rs323, %r1302; + fma.rn.f32.bf16 %r1304, %rs68, %rs324, %r1303; + fma.rn.f32.bf16 %r1305, %rs69, %rs325, %r1304; + fma.rn.f32.bf16 %r1306, %rs70, %rs326, %r1262; + fma.rn.f32.bf16 %r1307, %rs72, %rs328, %r1306; + fma.rn.f32.bf16 %r1308, %rs73, %rs329, %r1307; + fma.rn.f32.bf16 %r1309, %rs74, %rs330, %r1308; + fma.rn.f32.bf16 %r1310, %rs75, %rs331, %r1309; + fma.rn.f32.bf16 %r1311, %rs76, %rs332, %r1310; + fma.rn.f32.bf16 %r1312, %rs77, %rs333, %r1311; + fma.rn.f32.bf16 %r1313, %rs78, %rs334, %r1263; + fma.rn.f32.bf16 %r1314, %rs80, %rs336, %r1313; + fma.rn.f32.bf16 %r1315, %rs81, %rs337, %r1314; + fma.rn.f32.bf16 %r1316, %rs82, %rs338, %r1315; + fma.rn.f32.bf16 %r1317, %rs83, %rs339, %r1316; + fma.rn.f32.bf16 %r1318, %rs84, %rs340, %r1317; + fma.rn.f32.bf16 %r1319, %rs85, %rs341, %r1318; + fma.rn.f32.bf16 %r1320, %rs86, %rs342, %r1264; + fma.rn.f32.bf16 %r1321, %rs88, %rs344, %r1320; + fma.rn.f32.bf16 %r1322, %rs89, %rs345, %r1321; + fma.rn.f32.bf16 %r1323, %rs90, %rs346, %r1322; + fma.rn.f32.bf16 %r1324, %rs91, %rs347, %r1323; + fma.rn.f32.bf16 %r1325, %rs92, %rs348, %r1324; + fma.rn.f32.bf16 %r1326, %rs93, %rs349, %r1325; + fma.rn.f32.bf16 %r1327, %rs94, %rs350, %r1265; + fma.rn.f32.bf16 %r1328, %rs96, %rs352, %r1327; + fma.rn.f32.bf16 %r1329, %rs97, %rs353, %r1328; + fma.rn.f32.bf16 %r1330, %rs98, %rs354, %r1329; + fma.rn.f32.bf16 %r1331, %rs99, %rs355, %r1330; + fma.rn.f32.bf16 %r1332, %rs100, %rs356, %r1331; + fma.rn.f32.bf16 %r1333, %rs101, %rs357, %r1332; + fma.rn.f32.bf16 %r1334, %rs102, %rs358, %r1266; + fma.rn.f32.bf16 %r1335, %rs104, %rs360, %r1334; + fma.rn.f32.bf16 %r1336, %rs105, %rs361, %r1335; + fma.rn.f32.bf16 %r1337, %rs106, %rs362, %r1336; + fma.rn.f32.bf16 %r1338, %rs107, %rs363, %r1337; + fma.rn.f32.bf16 %r1339, %rs108, %rs364, %r1338; + fma.rn.f32.bf16 %r1340, %rs109, %rs365, %r1339; + fma.rn.f32.bf16 %r1341, %rs110, %rs366, %r1267; + fma.rn.f32.bf16 %r1342, %rs112, %rs368, %r1341; + fma.rn.f32.bf16 %r1343, %rs113, %rs369, %r1342; + fma.rn.f32.bf16 %r1344, %rs114, %rs370, %r1343; + fma.rn.f32.bf16 %r1345, %rs115, %rs371, %r1344; + fma.rn.f32.bf16 %r1346, %rs116, %rs372, %r1345; + fma.rn.f32.bf16 %r1347, %rs117, %rs373, %r1346; + fma.rn.f32.bf16 %r1348, %rs118, %rs374, %r1268; + fma.rn.f32.bf16 %r1349, %rs120, %rs376, %r1348; + fma.rn.f32.bf16 %r1350, %rs121, %rs377, %r1349; + fma.rn.f32.bf16 %r1351, %rs122, %rs378, %r1350; + fma.rn.f32.bf16 %r1352, %rs123, %rs379, %r1351; + fma.rn.f32.bf16 %r1353, %rs124, %rs380, %r1352; + fma.rn.f32.bf16 %r1354, %rs125, %rs381, %r1353; + fma.rn.f32.bf16 %r1355, %rs126, %rs382, %r1269; + fma.rn.f32.bf16 %r1356, %rs128, %rs384, %r1355; + fma.rn.f32.bf16 %r1357, %rs129, %rs385, %r1356; + fma.rn.f32.bf16 %r1358, %rs130, %rs386, %r1357; + fma.rn.f32.bf16 %r1359, %rs131, %rs387, %r1358; + fma.rn.f32.bf16 %r1360, %rs132, %rs388, %r1359; + fma.rn.f32.bf16 %r1361, %rs133, %rs389, %r1360; + fma.rn.f32.bf16 %r1362, %rs134, %rs390, %r1270; + fma.rn.f32.bf16 %r1363, %rs136, %rs392, %r1362; + fma.rn.f32.bf16 %r1364, %rs137, %rs393, %r1363; + fma.rn.f32.bf16 %r1365, %rs138, %rs394, %r1364; + fma.rn.f32.bf16 %r1366, %rs139, %rs395, %r1365; + fma.rn.f32.bf16 %r1367, %rs140, %rs396, %r1366; + fma.rn.f32.bf16 %r1368, %rs141, %rs397, %r1367; + fma.rn.f32.bf16 %r1369, %rs142, %rs398, %r1271; + fma.rn.f32.bf16 %r1370, %rs144, %rs400, %r1369; + fma.rn.f32.bf16 %r1371, %rs145, %rs401, %r1370; + fma.rn.f32.bf16 %r1372, %rs146, %rs402, %r1371; + fma.rn.f32.bf16 %r1373, %rs147, %rs403, %r1372; + fma.rn.f32.bf16 %r1374, %rs148, %rs404, %r1373; + fma.rn.f32.bf16 %r1375, %rs149, %rs405, %r1374; + fma.rn.f32.bf16 %r1376, %rs150, %rs406, %r1272; + fma.rn.f32.bf16 %r1377, %rs152, %rs408, %r1376; + fma.rn.f32.bf16 %r1378, %rs153, %rs409, %r1377; + fma.rn.f32.bf16 %r1379, %rs154, %rs410, %r1378; + fma.rn.f32.bf16 %r1380, %rs155, %rs411, %r1379; + fma.rn.f32.bf16 %r1381, %rs156, %rs412, %r1380; + fma.rn.f32.bf16 %r1382, %rs157, %rs413, %r1381; + fma.rn.f32.bf16 %r1383, %rs158, %rs414, %r1273; + fma.rn.f32.bf16 %r1384, %rs160, %rs416, %r1383; + fma.rn.f32.bf16 %r1385, %rs161, %rs417, %r1384; + fma.rn.f32.bf16 %r1386, %rs162, %rs418, %r1385; + fma.rn.f32.bf16 %r1387, %rs163, %rs419, %r1386; + fma.rn.f32.bf16 %r1388, %rs164, %rs420, %r1387; + fma.rn.f32.bf16 %r1389, %rs165, %rs421, %r1388; + fma.rn.f32.bf16 %r1390, %rs166, %rs422, %r1274; + fma.rn.f32.bf16 %r1391, %rs168, %rs424, %r1390; + fma.rn.f32.bf16 %r1392, %rs169, %rs425, %r1391; + fma.rn.f32.bf16 %r1393, %rs170, %rs426, %r1392; + fma.rn.f32.bf16 %r1394, %rs171, %rs427, %r1393; + fma.rn.f32.bf16 %r1395, %rs172, %rs428, %r1394; + fma.rn.f32.bf16 %r1396, %rs173, %rs429, %r1395; + fma.rn.f32.bf16 %r1397, %rs174, %rs430, %r1275; + fma.rn.f32.bf16 %r1398, %rs176, %rs432, %r1397; + fma.rn.f32.bf16 %r1399, %rs177, %rs433, %r1398; + fma.rn.f32.bf16 %r1400, %rs178, %rs434, %r1399; + fma.rn.f32.bf16 %r1401, %rs179, %rs435, %r1400; + fma.rn.f32.bf16 %r1402, %rs180, %rs436, %r1401; + fma.rn.f32.bf16 %r1403, %rs181, %rs437, %r1402; + fma.rn.f32.bf16 %r1404, %rs182, %rs438, %r1276; + fma.rn.f32.bf16 %r1405, %rs184, %rs440, %r1404; + fma.rn.f32.bf16 %r1406, %rs185, %rs441, %r1405; + fma.rn.f32.bf16 %r1407, %rs186, %rs442, %r1406; + fma.rn.f32.bf16 %r1408, %rs187, %rs443, %r1407; + fma.rn.f32.bf16 %r1409, %rs188, %rs444, %r1408; + fma.rn.f32.bf16 %r1410, %rs189, %rs445, %r1409; + fma.rn.f32.bf16 %r1411, %rs190, %rs446, %r1277; + fma.rn.f32.bf16 %r1412, %rs192, %rs448, %r1411; + fma.rn.f32.bf16 %r1413, %rs193, %rs449, %r1412; + fma.rn.f32.bf16 %r1414, %rs194, %rs450, %r1413; + fma.rn.f32.bf16 %r1415, %rs195, %rs451, %r1414; + fma.rn.f32.bf16 %r1416, %rs196, %rs452, %r1415; + fma.rn.f32.bf16 %r1417, %rs197, %rs453, %r1416; + fma.rn.f32.bf16 %r1418, %rs198, %rs454, %r1278; + fma.rn.f32.bf16 %r1419, %rs200, %rs456, %r1418; + fma.rn.f32.bf16 %r1420, %rs201, %rs457, %r1419; + fma.rn.f32.bf16 %r1421, %rs202, %rs458, %r1420; + fma.rn.f32.bf16 %r1422, %rs203, %rs459, %r1421; + fma.rn.f32.bf16 %r1423, %rs204, %rs460, %r1422; + fma.rn.f32.bf16 %r1424, %rs205, %rs461, %r1423; + fma.rn.f32.bf16 %r1425, %rs206, %rs462, %r1279; + fma.rn.f32.bf16 %r1426, %rs208, %rs464, %r1425; + fma.rn.f32.bf16 %r1427, %rs209, %rs465, %r1426; + fma.rn.f32.bf16 %r1428, %rs210, %rs466, %r1427; + fma.rn.f32.bf16 %r1429, %rs211, %rs467, %r1428; + fma.rn.f32.bf16 %r1430, %rs212, %rs468, %r1429; + fma.rn.f32.bf16 %r1431, %rs213, %rs469, %r1430; + fma.rn.f32.bf16 %r1432, %rs214, %rs470, %r1280; + fma.rn.f32.bf16 %r1433, %rs216, %rs472, %r1432; + fma.rn.f32.bf16 %r1434, %rs217, %rs473, %r1433; + fma.rn.f32.bf16 %r1435, %rs218, %rs474, %r1434; + fma.rn.f32.bf16 %r1436, %rs219, %rs475, %r1435; + fma.rn.f32.bf16 %r1437, %rs220, %rs476, %r1436; + fma.rn.f32.bf16 %r1438, %rs221, %rs477, %r1437; + fma.rn.f32.bf16 %r1439, %rs222, %rs478, %r1281; + fma.rn.f32.bf16 %r1440, %rs224, %rs480, %r1439; + fma.rn.f32.bf16 %r1441, %rs225, %rs481, %r1440; + fma.rn.f32.bf16 %r1442, %rs226, %rs482, %r1441; + fma.rn.f32.bf16 %r1443, %rs227, %rs483, %r1442; + fma.rn.f32.bf16 %r1444, %rs228, %rs484, %r1443; + fma.rn.f32.bf16 %r1445, %rs229, %rs485, %r1444; + fma.rn.f32.bf16 %r1446, %rs230, %rs486, %r1282; + fma.rn.f32.bf16 %r1447, %rs232, %rs488, %r1446; + fma.rn.f32.bf16 %r1448, %rs233, %rs489, %r1447; + fma.rn.f32.bf16 %r1449, %rs234, %rs490, %r1448; + fma.rn.f32.bf16 %r1450, %rs235, %rs491, %r1449; + fma.rn.f32.bf16 %r1451, %rs236, %rs492, %r1450; + fma.rn.f32.bf16 %r1452, %rs237, %rs493, %r1451; + fma.rn.f32.bf16 %r1453, %rs238, %rs494, %r1283; + fma.rn.f32.bf16 %r1454, %rs240, %rs496, %r1453; + fma.rn.f32.bf16 %r1455, %rs241, %rs497, %r1454; + fma.rn.f32.bf16 %r1456, %rs242, %rs498, %r1455; + fma.rn.f32.bf16 %r1457, %rs243, %rs499, %r1456; + fma.rn.f32.bf16 %r1458, %rs244, %rs500, %r1457; + fma.rn.f32.bf16 %r1459, %rs245, %rs501, %r1458; + fma.rn.f32.bf16 %r1460, %rs246, %rs502, %r1284; + fma.rn.f32.bf16 %r1461, %rs248, %rs504, %r1460; + fma.rn.f32.bf16 %r1462, %rs249, %rs505, %r1461; + fma.rn.f32.bf16 %r1463, %rs250, %rs506, %r1462; + fma.rn.f32.bf16 %r1464, %rs251, %rs507, %r1463; + fma.rn.f32.bf16 %r1465, %rs252, %rs508, %r1464; + fma.rn.f32.bf16 %r1466, %rs253, %rs509, %r1465; + fma.rn.f32.bf16 %r1467, %rs254, %rs510, %r1285; + fma.rn.f32.bf16 %r1468, %rs256, %rs512, %r1467; + fma.rn.f32.bf16 %r1469, %rs257, %rs513, %r1468; + fma.rn.f32.bf16 %r1470, %rs258, %rs514, %r1469; + fma.rn.f32.bf16 %r1471, %rs259, %rs515, %r1470; + fma.rn.f32.bf16 %r1472, %rs260, %rs516, %r1471; + fma.rn.f32.bf16 %r1473, %rs261, %rs517, %r1472; + fma.rn.f32.bf16 %r1474, %rs262, %rs518, %r1286; + fma.rn.f32.bf16 %r1475, %rs264, %rs520, %r1474; + fma.rn.f32.bf16 %r1476, %rs265, %rs521, %r1475; + fma.rn.f32.bf16 %r1477, %rs266, %rs522, %r1476; + fma.rn.f32.bf16 %r1478, %rs267, %rs523, %r1477; + fma.rn.f32.bf16 %r1479, %rs268, %rs524, %r1478; + fma.rn.f32.bf16 %r1480, %rs269, %rs525, %r1479; + fma.rn.f32.bf16 %r1481, %rs270, %rs526, %r1287; + fma.rn.f32.bf16 %r1482, %rs272, %rs528, %r1481; + fma.rn.f32.bf16 %r1483, %rs273, %rs529, %r1482; + fma.rn.f32.bf16 %r1484, %rs274, %rs530, %r1483; + fma.rn.f32.bf16 %r1485, %rs275, %rs531, %r1484; + fma.rn.f32.bf16 %r1486, %rs276, %rs532, %r1485; + fma.rn.f32.bf16 %r1487, %rs277, %rs533, %r1486; + fma.rn.f32.bf16 %r1488, %rs278, %rs534, %r1288; + fma.rn.f32.bf16 %r1489, %rs280, %rs536, %r1488; + fma.rn.f32.bf16 %r1490, %rs281, %rs537, %r1489; + fma.rn.f32.bf16 %r1491, %rs282, %rs538, %r1490; + fma.rn.f32.bf16 %r1492, %rs283, %rs539, %r1491; + fma.rn.f32.bf16 %r1493, %rs284, %rs540, %r1492; + fma.rn.f32.bf16 %r1494, %rs285, %rs541, %r1493; + fma.rn.f32.bf16 %r1495, %rs286, %rs542, %r1289; + fma.rn.f32.bf16 %r1496, %rs288, %rs544, %r1495; + fma.rn.f32.bf16 %r1497, %rs289, %rs545, %r1496; + fma.rn.f32.bf16 %r1498, %rs290, %rs546, %r1497; + fma.rn.f32.bf16 %r1499, %rs291, %rs547, %r1498; + fma.rn.f32.bf16 %r1500, %rs292, %rs548, %r1499; + fma.rn.f32.bf16 %r1501, %rs293, %rs549, %r1500; + fma.rn.f32.bf16 %r1502, %rs294, %rs550, %r1290; + fma.rn.f32.bf16 %r1503, %rs296, %rs552, %r1502; + fma.rn.f32.bf16 %r1504, %rs297, %rs553, %r1503; + fma.rn.f32.bf16 %r1505, %rs298, %rs554, %r1504; + fma.rn.f32.bf16 %r1506, %rs299, %rs555, %r1505; + fma.rn.f32.bf16 %r1507, %rs300, %rs556, %r1506; + fma.rn.f32.bf16 %r1508, %rs301, %rs557, %r1507; + fma.rn.f32.bf16 %r1509, %rs302, %rs558, %r1291; + fma.rn.f32.bf16 %r1510, %rs304, %rs560, %r1509; + fma.rn.f32.bf16 %r1511, %rs305, %rs561, %r1510; + fma.rn.f32.bf16 %r1512, %rs306, %rs562, %r1511; + fma.rn.f32.bf16 %r1513, %rs307, %rs563, %r1512; + fma.rn.f32.bf16 %r1514, %rs308, %rs564, %r1513; + fma.rn.f32.bf16 %r1515, %rs309, %rs565, %r1514; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1516, %r1298, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1517, %r1298, %r1516; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1518, %r1517, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1519, %r1517, %r1518; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1520, %r1519, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1521, %r1519, %r1520; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1522, %r1521, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1523, %r1521, %r1522; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1524, %r1305, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1525, %r1305, %r1524; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1526, %r1525, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1527, %r1525, %r1526; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1528, %r1527, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1529, %r1527, %r1528; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1530, %r1529, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1531, %r1529, %r1530; +$L__tmp18: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1532, %r1312, 8, 31, -1; +$L__tmp19: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1533, %r1312, %r1532; +$L__tmp20: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1534, %r1533, 4, 31, -1; +$L__tmp21: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1535, %r1533, %r1534; +$L__tmp22: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1536, %r1535, 2, 31, -1; +$L__tmp23: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1537, %r1535, %r1536; +$L__tmp24: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1538, %r1537, 1, 31, -1; +$L__tmp25: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1539, %r1537, %r1538; +$L__tmp26: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1540, %r1319, 8, 31, -1; +$L__tmp27: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1541, %r1319, %r1540; +$L__tmp28: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1542, %r1541, 4, 31, -1; +$L__tmp29: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1543, %r1541, %r1542; +$L__tmp30: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1544, %r1543, 2, 31, -1; +$L__tmp31: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1545, %r1543, %r1544; +$L__tmp32: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1546, %r1545, 1, 31, -1; +$L__tmp33: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1547, %r1545, %r1546; +$L__tmp34: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1548, %r1326, 8, 31, -1; +$L__tmp35: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1549, %r1326, %r1548; +$L__tmp36: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1550, %r1549, 4, 31, -1; +$L__tmp37: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1551, %r1549, %r1550; +$L__tmp38: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1552, %r1551, 2, 31, -1; +$L__tmp39: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1553, %r1551, %r1552; +$L__tmp40: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1554, %r1553, 1, 31, -1; +$L__tmp41: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1555, %r1553, %r1554; +$L__tmp42: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1556, %r1333, 8, 31, -1; +$L__tmp43: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1557, %r1333, %r1556; +$L__tmp44: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1558, %r1557, 4, 31, -1; +$L__tmp45: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1559, %r1557, %r1558; +$L__tmp46: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1560, %r1559, 2, 31, -1; +$L__tmp47: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1561, %r1559, %r1560; +$L__tmp48: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1562, %r1561, 1, 31, -1; +$L__tmp49: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1563, %r1561, %r1562; +$L__tmp50: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1564, %r1340, 8, 31, -1; +$L__tmp51: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1565, %r1340, %r1564; +$L__tmp52: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1566, %r1565, 4, 31, -1; +$L__tmp53: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1567, %r1565, %r1566; +$L__tmp54: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1568, %r1567, 2, 31, -1; +$L__tmp55: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1569, %r1567, %r1568; +$L__tmp56: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1570, %r1569, 1, 31, -1; +$L__tmp57: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1571, %r1569, %r1570; +$L__tmp58: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1572, %r1347, 8, 31, -1; +$L__tmp59: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1573, %r1347, %r1572; +$L__tmp60: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1574, %r1573, 4, 31, -1; +$L__tmp61: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1575, %r1573, %r1574; +$L__tmp62: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1576, %r1575, 2, 31, -1; +$L__tmp63: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1577, %r1575, %r1576; +$L__tmp64: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1578, %r1577, 1, 31, -1; +$L__tmp65: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1579, %r1577, %r1578; +$L__tmp66: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1580, %r1354, 8, 31, -1; +$L__tmp67: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1581, %r1354, %r1580; +$L__tmp68: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1582, %r1581, 4, 31, -1; +$L__tmp69: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1583, %r1581, %r1582; +$L__tmp70: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1584, %r1583, 2, 31, -1; +$L__tmp71: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1585, %r1583, %r1584; +$L__tmp72: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1586, %r1585, 1, 31, -1; +$L__tmp73: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1587, %r1585, %r1586; +$L__tmp74: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1588, %r1361, 8, 31, -1; +$L__tmp75: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1589, %r1361, %r1588; +$L__tmp76: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1590, %r1589, 4, 31, -1; +$L__tmp77: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1591, %r1589, %r1590; +$L__tmp78: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1592, %r1591, 2, 31, -1; +$L__tmp79: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1593, %r1591, %r1592; +$L__tmp80: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1594, %r1593, 1, 31, -1; +$L__tmp81: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1595, %r1593, %r1594; +$L__tmp82: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1596, %r1368, 8, 31, -1; +$L__tmp83: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1597, %r1368, %r1596; +$L__tmp84: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1598, %r1597, 4, 31, -1; +$L__tmp85: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1599, %r1597, %r1598; +$L__tmp86: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1600, %r1599, 2, 31, -1; +$L__tmp87: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1601, %r1599, %r1600; +$L__tmp88: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1602, %r1601, 1, 31, -1; +$L__tmp89: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1603, %r1601, %r1602; +$L__tmp90: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1604, %r1375, 8, 31, -1; +$L__tmp91: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1605, %r1375, %r1604; +$L__tmp92: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1606, %r1605, 4, 31, -1; +$L__tmp93: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1607, %r1605, %r1606; +$L__tmp94: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1608, %r1607, 2, 31, -1; +$L__tmp95: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1609, %r1607, %r1608; +$L__tmp96: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1610, %r1609, 1, 31, -1; +$L__tmp97: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1611, %r1609, %r1610; +$L__tmp98: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1612, %r1382, 8, 31, -1; +$L__tmp99: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1613, %r1382, %r1612; +$L__tmp100: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1614, %r1613, 4, 31, -1; +$L__tmp101: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1615, %r1613, %r1614; +$L__tmp102: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1616, %r1615, 2, 31, -1; +$L__tmp103: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1617, %r1615, %r1616; +$L__tmp104: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1618, %r1617, 1, 31, -1; +$L__tmp105: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1619, %r1617, %r1618; +$L__tmp106: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1620, %r1389, 8, 31, -1; +$L__tmp107: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1621, %r1389, %r1620; +$L__tmp108: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1622, %r1621, 4, 31, -1; +$L__tmp109: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1623, %r1621, %r1622; +$L__tmp110: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1624, %r1623, 2, 31, -1; +$L__tmp111: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1625, %r1623, %r1624; +$L__tmp112: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1626, %r1625, 1, 31, -1; +$L__tmp113: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1627, %r1625, %r1626; +$L__tmp114: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1628, %r1396, 8, 31, -1; +$L__tmp115: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1629, %r1396, %r1628; +$L__tmp116: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1630, %r1629, 4, 31, -1; +$L__tmp117: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1631, %r1629, %r1630; +$L__tmp118: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1632, %r1631, 2, 31, -1; +$L__tmp119: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1633, %r1631, %r1632; +$L__tmp120: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1634, %r1633, 1, 31, -1; +$L__tmp121: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1635, %r1633, %r1634; +$L__tmp122: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1636, %r1403, 8, 31, -1; +$L__tmp123: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1637, %r1403, %r1636; +$L__tmp124: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1638, %r1637, 4, 31, -1; +$L__tmp125: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1639, %r1637, %r1638; +$L__tmp126: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1640, %r1639, 2, 31, -1; +$L__tmp127: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1641, %r1639, %r1640; +$L__tmp128: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1642, %r1641, 1, 31, -1; +$L__tmp129: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1643, %r1641, %r1642; +$L__tmp130: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1644, %r1410, 8, 31, -1; +$L__tmp131: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1645, %r1410, %r1644; +$L__tmp132: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1646, %r1645, 4, 31, -1; +$L__tmp133: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1647, %r1645, %r1646; +$L__tmp134: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1648, %r1647, 2, 31, -1; +$L__tmp135: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1649, %r1647, %r1648; +$L__tmp136: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1650, %r1649, 1, 31, -1; +$L__tmp137: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1651, %r1649, %r1650; +$L__tmp138: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1652, %r1417, 8, 31, -1; +$L__tmp139: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1653, %r1417, %r1652; +$L__tmp140: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1654, %r1653, 4, 31, -1; +$L__tmp141: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1655, %r1653, %r1654; +$L__tmp142: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1656, %r1655, 2, 31, -1; +$L__tmp143: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1657, %r1655, %r1656; +$L__tmp144: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1658, %r1657, 1, 31, -1; +$L__tmp145: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1659, %r1657, %r1658; +$L__tmp146: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1660, %r1424, 8, 31, -1; +$L__tmp147: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1661, %r1424, %r1660; +$L__tmp148: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1662, %r1661, 4, 31, -1; +$L__tmp149: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1663, %r1661, %r1662; +$L__tmp150: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1664, %r1663, 2, 31, -1; +$L__tmp151: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1665, %r1663, %r1664; +$L__tmp152: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1666, %r1665, 1, 31, -1; +$L__tmp153: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1667, %r1665, %r1666; +$L__tmp154: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1668, %r1431, 8, 31, -1; +$L__tmp155: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1669, %r1431, %r1668; +$L__tmp156: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1670, %r1669, 4, 31, -1; +$L__tmp157: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1671, %r1669, %r1670; +$L__tmp158: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1672, %r1671, 2, 31, -1; +$L__tmp159: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1673, %r1671, %r1672; +$L__tmp160: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1674, %r1673, 1, 31, -1; +$L__tmp161: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1675, %r1673, %r1674; +$L__tmp162: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1676, %r1438, 8, 31, -1; +$L__tmp163: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1677, %r1438, %r1676; +$L__tmp164: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1678, %r1677, 4, 31, -1; +$L__tmp165: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1679, %r1677, %r1678; +$L__tmp166: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1680, %r1679, 2, 31, -1; +$L__tmp167: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1681, %r1679, %r1680; +$L__tmp168: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1682, %r1681, 1, 31, -1; +$L__tmp169: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1683, %r1681, %r1682; +$L__tmp170: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1684, %r1445, 8, 31, -1; +$L__tmp171: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1685, %r1445, %r1684; +$L__tmp172: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1686, %r1685, 4, 31, -1; +$L__tmp173: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1687, %r1685, %r1686; +$L__tmp174: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1688, %r1687, 2, 31, -1; +$L__tmp175: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1689, %r1687, %r1688; +$L__tmp176: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1690, %r1689, 1, 31, -1; +$L__tmp177: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1691, %r1689, %r1690; +$L__tmp178: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1692, %r1452, 8, 31, -1; +$L__tmp179: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1693, %r1452, %r1692; +$L__tmp180: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1694, %r1693, 4, 31, -1; +$L__tmp181: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1695, %r1693, %r1694; +$L__tmp182: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1696, %r1695, 2, 31, -1; +$L__tmp183: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1697, %r1695, %r1696; +$L__tmp184: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1698, %r1697, 1, 31, -1; +$L__tmp185: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1699, %r1697, %r1698; +$L__tmp186: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1700, %r1459, 8, 31, -1; +$L__tmp187: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1701, %r1459, %r1700; +$L__tmp188: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1702, %r1701, 4, 31, -1; +$L__tmp189: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1703, %r1701, %r1702; +$L__tmp190: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1704, %r1703, 2, 31, -1; +$L__tmp191: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1705, %r1703, %r1704; +$L__tmp192: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1706, %r1705, 1, 31, -1; +$L__tmp193: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1707, %r1705, %r1706; +$L__tmp194: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1708, %r1466, 8, 31, -1; +$L__tmp195: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1709, %r1466, %r1708; +$L__tmp196: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1710, %r1709, 4, 31, -1; +$L__tmp197: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1711, %r1709, %r1710; +$L__tmp198: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1712, %r1711, 2, 31, -1; +$L__tmp199: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1713, %r1711, %r1712; +$L__tmp200: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1714, %r1713, 1, 31, -1; +$L__tmp201: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1715, %r1713, %r1714; +$L__tmp202: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1716, %r1473, 8, 31, -1; +$L__tmp203: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1717, %r1473, %r1716; +$L__tmp204: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1718, %r1717, 4, 31, -1; +$L__tmp205: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1719, %r1717, %r1718; +$L__tmp206: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1720, %r1719, 2, 31, -1; +$L__tmp207: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1721, %r1719, %r1720; +$L__tmp208: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1722, %r1721, 1, 31, -1; +$L__tmp209: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1723, %r1721, %r1722; +$L__tmp210: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1724, %r1480, 8, 31, -1; +$L__tmp211: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1725, %r1480, %r1724; +$L__tmp212: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1726, %r1725, 4, 31, -1; +$L__tmp213: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1727, %r1725, %r1726; +$L__tmp214: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1728, %r1727, 2, 31, -1; +$L__tmp215: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1729, %r1727, %r1728; +$L__tmp216: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1730, %r1729, 1, 31, -1; +$L__tmp217: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1731, %r1729, %r1730; +$L__tmp218: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1732, %r1487, 8, 31, -1; +$L__tmp219: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1733, %r1487, %r1732; +$L__tmp220: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1734, %r1733, 4, 31, -1; +$L__tmp221: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1735, %r1733, %r1734; +$L__tmp222: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1736, %r1735, 2, 31, -1; +$L__tmp223: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1737, %r1735, %r1736; +$L__tmp224: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1738, %r1737, 1, 31, -1; +$L__tmp225: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1739, %r1737, %r1738; +$L__tmp226: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1740, %r1494, 8, 31, -1; +$L__tmp227: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1741, %r1494, %r1740; +$L__tmp228: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1742, %r1741, 4, 31, -1; +$L__tmp229: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1743, %r1741, %r1742; +$L__tmp230: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1744, %r1743, 2, 31, -1; +$L__tmp231: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1745, %r1743, %r1744; +$L__tmp232: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1746, %r1745, 1, 31, -1; +$L__tmp233: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1747, %r1745, %r1746; +$L__tmp234: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1748, %r1501, 8, 31, -1; +$L__tmp235: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1749, %r1501, %r1748; +$L__tmp236: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1750, %r1749, 4, 31, -1; +$L__tmp237: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1751, %r1749, %r1750; +$L__tmp238: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1752, %r1751, 2, 31, -1; +$L__tmp239: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1753, %r1751, %r1752; +$L__tmp240: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1754, %r1753, 1, 31, -1; +$L__tmp241: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1755, %r1753, %r1754; +$L__tmp242: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1756, %r1508, 8, 31, -1; +$L__tmp243: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1757, %r1508, %r1756; +$L__tmp244: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1758, %r1757, 4, 31, -1; +$L__tmp245: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1759, %r1757, %r1758; +$L__tmp246: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1760, %r1759, 2, 31, -1; +$L__tmp247: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1761, %r1759, %r1760; +$L__tmp248: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1762, %r1761, 1, 31, -1; +$L__tmp249: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1763, %r1761, %r1762; +$L__tmp250: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1764, %r1515, 8, 31, -1; +$L__tmp251: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1765, %r1515, %r1764; +$L__tmp252: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1766, %r1765, 4, 31, -1; +$L__tmp253: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1767, %r1765, %r1766; +$L__tmp254: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1768, %r1767, 2, 31, -1; +$L__tmp255: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1769, %r1767, %r1768; +$L__tmp256: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1770, %r1769, 1, 31, -1; +$L__tmp257: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1771, %r1769, %r1770; +$L__tmp258: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r1772, %r1523, %r459; + mul.f32 %r1773, %r1523, %r460; + mul.f32 %r1774, %r1523, %r461; + mul.f32 %r1775, %r1523, %r462; + mul.f32 %r1776, %r1523, %r463; + mul.f32 %r1777, %r1523, %r464; + mul.f32 %r1778, %r1523, %r465; + mul.f32 %r1779, %r1523, %r466; + mul.f32 %r1780, %r1531, %r467; + mul.f32 %r1781, %r1531, %r468; + mul.f32 %r1782, %r1531, %r469; + mul.f32 %r1783, %r1531, %r470; + mul.f32 %r1784, %r1531, %r471; + mul.f32 %r1785, %r1531, %r472; + mul.f32 %r1786, %r1531, %r473; + mul.f32 %r1787, %r1531, %r474; + mul.f32 %r1788, %r1539, %r475; + mul.f32 %r1789, %r1539, %r476; + mul.f32 %r1790, %r1539, %r477; + mul.f32 %r1791, %r1539, %r478; + mul.f32 %r1792, %r1539, %r479; + mul.f32 %r1793, %r1539, %r480; + mul.f32 %r1794, %r1539, %r481; + mul.f32 %r1795, %r1539, %r482; + mul.f32 %r1796, %r1547, %r483; + mul.f32 %r1797, %r1547, %r484; + mul.f32 %r1798, %r1547, %r485; + mul.f32 %r1799, %r1547, %r486; + mul.f32 %r1800, %r1547, %r487; + mul.f32 %r1801, %r1547, %r488; + mul.f32 %r1802, %r1547, %r489; + mul.f32 %r1803, %r1547, %r490; + mul.f32 %r1804, %r1555, %r491; + mul.f32 %r1805, %r1555, %r492; + mul.f32 %r1806, %r1555, %r493; + mul.f32 %r1807, %r1555, %r494; + mul.f32 %r1808, %r1555, %r495; + mul.f32 %r1809, %r1555, %r496; + mul.f32 %r1810, %r1555, %r497; + mul.f32 %r1811, %r1555, %r498; + mul.f32 %r1812, %r1563, %r499; + mul.f32 %r1813, %r1563, %r500; + mul.f32 %r1814, %r1563, %r501; + mul.f32 %r1815, %r1563, %r502; + mul.f32 %r1816, %r1563, %r503; + mul.f32 %r1817, %r1563, %r504; + mul.f32 %r1818, %r1563, %r505; + mul.f32 %r1819, %r1563, %r506; + mul.f32 %r1820, %r1571, %r507; + mul.f32 %r1821, %r1571, %r508; + mul.f32 %r1822, %r1571, %r509; + mul.f32 %r1823, %r1571, %r510; + mul.f32 %r1824, %r1571, %r511; + mul.f32 %r1825, %r1571, %r512; + mul.f32 %r1826, %r1571, %r513; + mul.f32 %r1827, %r1571, %r514; + mul.f32 %r1828, %r1579, %r515; + mul.f32 %r1829, %r1579, %r516; + mul.f32 %r1830, %r1579, %r517; + mul.f32 %r1831, %r1579, %r518; + mul.f32 %r1832, %r1579, %r519; + mul.f32 %r1833, %r1579, %r520; + mul.f32 %r1834, %r1579, %r521; + mul.f32 %r1835, %r1579, %r522; + mul.f32 %r1836, %r1587, %r523; + mul.f32 %r1837, %r1587, %r524; + mul.f32 %r1838, %r1587, %r525; + mul.f32 %r1839, %r1587, %r526; + mul.f32 %r1840, %r1587, %r527; + mul.f32 %r1841, %r1587, %r528; + mul.f32 %r1842, %r1587, %r529; + mul.f32 %r1843, %r1587, %r530; + mul.f32 %r1844, %r1595, %r531; + mul.f32 %r1845, %r1595, %r532; + mul.f32 %r1846, %r1595, %r533; + mul.f32 %r1847, %r1595, %r534; + mul.f32 %r1848, %r1595, %r535; + mul.f32 %r1849, %r1595, %r536; + mul.f32 %r1850, %r1595, %r537; + mul.f32 %r1851, %r1595, %r538; + mul.f32 %r1852, %r1603, %r539; + mul.f32 %r1853, %r1603, %r540; + mul.f32 %r1854, %r1603, %r541; + mul.f32 %r1855, %r1603, %r542; + mul.f32 %r1856, %r1603, %r543; + mul.f32 %r1857, %r1603, %r544; + mul.f32 %r1858, %r1603, %r545; + mul.f32 %r1859, %r1603, %r546; + mul.f32 %r1860, %r1611, %r547; + mul.f32 %r1861, %r1611, %r548; + mul.f32 %r1862, %r1611, %r549; + mul.f32 %r1863, %r1611, %r550; + mul.f32 %r1864, %r1611, %r551; + mul.f32 %r1865, %r1611, %r552; + mul.f32 %r1866, %r1611, %r553; + mul.f32 %r1867, %r1611, %r554; + mul.f32 %r1868, %r1619, %r555; + mul.f32 %r1869, %r1619, %r556; + mul.f32 %r1870, %r1619, %r557; + mul.f32 %r1871, %r1619, %r558; + mul.f32 %r1872, %r1619, %r559; + mul.f32 %r1873, %r1619, %r560; + mul.f32 %r1874, %r1619, %r561; + mul.f32 %r1875, %r1619, %r562; + mul.f32 %r1876, %r1627, %r563; + mul.f32 %r1877, %r1627, %r564; + mul.f32 %r1878, %r1627, %r565; + mul.f32 %r1879, %r1627, %r566; + mul.f32 %r1880, %r1627, %r567; + mul.f32 %r1881, %r1627, %r568; + mul.f32 %r1882, %r1627, %r569; + mul.f32 %r1883, %r1627, %r570; + mul.f32 %r1884, %r1635, %r571; + mul.f32 %r1885, %r1635, %r572; + mul.f32 %r1886, %r1635, %r573; + mul.f32 %r1887, %r1635, %r574; + mul.f32 %r1888, %r1635, %r575; + mul.f32 %r1889, %r1635, %r576; + mul.f32 %r1890, %r1635, %r577; + mul.f32 %r1891, %r1635, %r578; + mul.f32 %r1892, %r1643, %r579; + mul.f32 %r1893, %r1643, %r580; + mul.f32 %r1894, %r1643, %r581; + mul.f32 %r1895, %r1643, %r582; + mul.f32 %r1896, %r1643, %r583; + mul.f32 %r1897, %r1643, %r584; + mul.f32 %r1898, %r1643, %r585; + mul.f32 %r1899, %r1643, %r586; + mul.f32 %r1900, %r1651, %r587; + mul.f32 %r1901, %r1651, %r588; + mul.f32 %r1902, %r1651, %r589; + mul.f32 %r1903, %r1651, %r590; + mul.f32 %r1904, %r1651, %r591; + mul.f32 %r1905, %r1651, %r592; + mul.f32 %r1906, %r1651, %r593; + mul.f32 %r1907, %r1651, %r594; + mul.f32 %r1908, %r1659, %r595; + mul.f32 %r1909, %r1659, %r596; + mul.f32 %r1910, %r1659, %r597; + mul.f32 %r1911, %r1659, %r598; + mul.f32 %r1912, %r1659, %r599; + mul.f32 %r1913, %r1659, %r600; + mul.f32 %r1914, %r1659, %r601; + mul.f32 %r1915, %r1659, %r602; + mul.f32 %r1916, %r1667, %r603; + mul.f32 %r1917, %r1667, %r604; + mul.f32 %r1918, %r1667, %r605; + mul.f32 %r1919, %r1667, %r606; + mul.f32 %r1920, %r1667, %r607; + mul.f32 %r1921, %r1667, %r608; + mul.f32 %r1922, %r1667, %r609; + mul.f32 %r1923, %r1667, %r610; + mul.f32 %r1924, %r1675, %r611; + mul.f32 %r1925, %r1675, %r612; + mul.f32 %r1926, %r1675, %r613; + mul.f32 %r1927, %r1675, %r614; + mul.f32 %r1928, %r1675, %r615; + mul.f32 %r1929, %r1675, %r616; + mul.f32 %r1930, %r1675, %r617; + mul.f32 %r1931, %r1675, %r618; + mul.f32 %r1932, %r1683, %r619; + mul.f32 %r1933, %r1683, %r620; + mul.f32 %r1934, %r1683, %r621; + mul.f32 %r1935, %r1683, %r622; + mul.f32 %r1936, %r1683, %r623; + mul.f32 %r1937, %r1683, %r624; + mul.f32 %r1938, %r1683, %r625; + mul.f32 %r1939, %r1683, %r626; + mul.f32 %r1940, %r1691, %r627; + mul.f32 %r1941, %r1691, %r628; + mul.f32 %r1942, %r1691, %r629; + mul.f32 %r1943, %r1691, %r630; + mul.f32 %r1944, %r1691, %r631; + mul.f32 %r1945, %r1691, %r632; + mul.f32 %r1946, %r1691, %r633; + mul.f32 %r1947, %r1691, %r634; + mul.f32 %r1948, %r1699, %r635; + mul.f32 %r1949, %r1699, %r636; + mul.f32 %r1950, %r1699, %r637; + mul.f32 %r1951, %r1699, %r638; + mul.f32 %r1952, %r1699, %r639; + mul.f32 %r1953, %r1699, %r640; + mul.f32 %r1954, %r1699, %r641; + mul.f32 %r1955, %r1699, %r642; + mul.f32 %r1956, %r1707, %r643; + mul.f32 %r1957, %r1707, %r644; + mul.f32 %r1958, %r1707, %r645; + mul.f32 %r1959, %r1707, %r646; + mul.f32 %r1960, %r1707, %r647; + mul.f32 %r1961, %r1707, %r648; + mul.f32 %r1962, %r1707, %r649; + mul.f32 %r1963, %r1707, %r650; + mul.f32 %r1964, %r1715, %r651; + mul.f32 %r1965, %r1715, %r652; + mul.f32 %r1966, %r1715, %r653; + mul.f32 %r1967, %r1715, %r654; + mul.f32 %r1968, %r1715, %r655; + mul.f32 %r1969, %r1715, %r656; + mul.f32 %r1970, %r1715, %r657; + mul.f32 %r1971, %r1715, %r658; + mul.f32 %r1972, %r1723, %r659; + mul.f32 %r1973, %r1723, %r660; + mul.f32 %r1974, %r1723, %r661; + mul.f32 %r1975, %r1723, %r662; + mul.f32 %r1976, %r1723, %r663; + mul.f32 %r1977, %r1723, %r664; + mul.f32 %r1978, %r1723, %r665; + mul.f32 %r1979, %r1723, %r666; + mul.f32 %r1980, %r1731, %r667; + mul.f32 %r1981, %r1731, %r668; + mul.f32 %r1982, %r1731, %r669; + mul.f32 %r1983, %r1731, %r670; + mul.f32 %r1984, %r1731, %r671; + mul.f32 %r1985, %r1731, %r672; + mul.f32 %r1986, %r1731, %r673; + mul.f32 %r1987, %r1731, %r674; + mul.f32 %r1988, %r1739, %r675; + mul.f32 %r1989, %r1739, %r676; + mul.f32 %r1990, %r1739, %r677; + mul.f32 %r1991, %r1739, %r678; + mul.f32 %r1992, %r1739, %r679; + mul.f32 %r1993, %r1739, %r680; + mul.f32 %r1994, %r1739, %r681; + mul.f32 %r1995, %r1739, %r682; + mul.f32 %r1996, %r1747, %r683; + mul.f32 %r1997, %r1747, %r684; + mul.f32 %r1998, %r1747, %r685; + mul.f32 %r1999, %r1747, %r686; + mul.f32 %r2000, %r1747, %r687; + mul.f32 %r2001, %r1747, %r688; + mul.f32 %r2002, %r1747, %r689; + mul.f32 %r2003, %r1747, %r690; + mul.f32 %r2004, %r1755, %r691; + mul.f32 %r2005, %r1755, %r692; + mul.f32 %r2006, %r1755, %r693; + mul.f32 %r2007, %r1755, %r694; + mul.f32 %r2008, %r1755, %r695; + mul.f32 %r2009, %r1755, %r696; + mul.f32 %r2010, %r1755, %r697; + mul.f32 %r2011, %r1755, %r698; + mul.f32 %r2012, %r1763, %r699; + mul.f32 %r2013, %r1763, %r700; + mul.f32 %r2014, %r1763, %r701; + mul.f32 %r2015, %r1763, %r702; + mul.f32 %r2016, %r1763, %r703; + mul.f32 %r2017, %r1763, %r704; + mul.f32 %r2018, %r1763, %r705; + mul.f32 %r2019, %r1763, %r706; + mul.f32 %r2020, %r1771, %r707; + mul.f32 %r2021, %r1771, %r708; + mul.f32 %r2022, %r1771, %r709; + mul.f32 %r2023, %r1771, %r710; + mul.f32 %r2024, %r1771, %r711; + mul.f32 %r2025, %r1771, %r712; + mul.f32 %r2026, %r1771, %r713; + mul.f32 %r2027, %r1771, %r714; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r2028, %r988, %r1772; + mul.f32 %r2029, %r988, %r1773; + mul.f32 %r2030, %r988, %r1774; + mul.f32 %r2031, %r988, %r1775; + mul.f32 %r2032, %r988, %r1776; + mul.f32 %r2033, %r988, %r1777; + mul.f32 %r2034, %r988, %r1778; + mul.f32 %r2035, %r988, %r1779; + mul.f32 %r2036, %r989, %r1780; + mul.f32 %r2037, %r989, %r1781; + mul.f32 %r2038, %r989, %r1782; + mul.f32 %r2039, %r989, %r1783; + mul.f32 %r2040, %r989, %r1784; + mul.f32 %r2041, %r989, %r1785; + mul.f32 %r2042, %r989, %r1786; + mul.f32 %r2043, %r989, %r1787; + mul.f32 %r2044, %r990, %r1788; + mul.f32 %r2045, %r990, %r1789; + mul.f32 %r2046, %r990, %r1790; + mul.f32 %r2047, %r990, %r1791; + mul.f32 %r2048, %r990, %r1792; + mul.f32 %r2049, %r990, %r1793; + mul.f32 %r2050, %r990, %r1794; + mul.f32 %r2051, %r990, %r1795; + mul.f32 %r2052, %r991, %r1796; + mul.f32 %r2053, %r991, %r1797; + mul.f32 %r2054, %r991, %r1798; + mul.f32 %r2055, %r991, %r1799; + mul.f32 %r2056, %r991, %r1800; + mul.f32 %r2057, %r991, %r1801; + mul.f32 %r2058, %r991, %r1802; + mul.f32 %r2059, %r991, %r1803; + mul.f32 %r2060, %r992, %r1804; + mul.f32 %r2061, %r992, %r1805; + mul.f32 %r2062, %r992, %r1806; + mul.f32 %r2063, %r992, %r1807; + mul.f32 %r2064, %r992, %r1808; + mul.f32 %r2065, %r992, %r1809; + mul.f32 %r2066, %r992, %r1810; + mul.f32 %r2067, %r992, %r1811; + mul.f32 %r2068, %r993, %r1812; + mul.f32 %r2069, %r993, %r1813; + mul.f32 %r2070, %r993, %r1814; + mul.f32 %r2071, %r993, %r1815; + mul.f32 %r2072, %r993, %r1816; + mul.f32 %r2073, %r993, %r1817; + mul.f32 %r2074, %r993, %r1818; + mul.f32 %r2075, %r993, %r1819; + mul.f32 %r2076, %r994, %r1820; + mul.f32 %r2077, %r994, %r1821; + mul.f32 %r2078, %r994, %r1822; + mul.f32 %r2079, %r994, %r1823; + mul.f32 %r2080, %r994, %r1824; + mul.f32 %r2081, %r994, %r1825; + mul.f32 %r2082, %r994, %r1826; + mul.f32 %r2083, %r994, %r1827; + mul.f32 %r2084, %r995, %r1828; + mul.f32 %r2085, %r995, %r1829; + mul.f32 %r2086, %r995, %r1830; + mul.f32 %r2087, %r995, %r1831; + mul.f32 %r2088, %r995, %r1832; + mul.f32 %r2089, %r995, %r1833; + mul.f32 %r2090, %r995, %r1834; + mul.f32 %r2091, %r995, %r1835; + mul.f32 %r2092, %r996, %r1836; + mul.f32 %r2093, %r996, %r1837; + mul.f32 %r2094, %r996, %r1838; + mul.f32 %r2095, %r996, %r1839; + mul.f32 %r2096, %r996, %r1840; + mul.f32 %r2097, %r996, %r1841; + mul.f32 %r2098, %r996, %r1842; + mul.f32 %r2099, %r996, %r1843; + mul.f32 %r2100, %r997, %r1844; + mul.f32 %r2101, %r997, %r1845; + mul.f32 %r2102, %r997, %r1846; + mul.f32 %r2103, %r997, %r1847; + mul.f32 %r2104, %r997, %r1848; + mul.f32 %r2105, %r997, %r1849; + mul.f32 %r2106, %r997, %r1850; + mul.f32 %r2107, %r997, %r1851; + mul.f32 %r2108, %r998, %r1852; + mul.f32 %r2109, %r998, %r1853; + mul.f32 %r2110, %r998, %r1854; + mul.f32 %r2111, %r998, %r1855; + mul.f32 %r2112, %r998, %r1856; + mul.f32 %r2113, %r998, %r1857; + mul.f32 %r2114, %r998, %r1858; + mul.f32 %r2115, %r998, %r1859; + mul.f32 %r2116, %r999, %r1860; + mul.f32 %r2117, %r999, %r1861; + mul.f32 %r2118, %r999, %r1862; + mul.f32 %r2119, %r999, %r1863; + mul.f32 %r2120, %r999, %r1864; + mul.f32 %r2121, %r999, %r1865; + mul.f32 %r2122, %r999, %r1866; + mul.f32 %r2123, %r999, %r1867; + mul.f32 %r2124, %r1000, %r1868; + mul.f32 %r2125, %r1000, %r1869; + mul.f32 %r2126, %r1000, %r1870; + mul.f32 %r2127, %r1000, %r1871; + mul.f32 %r2128, %r1000, %r1872; + mul.f32 %r2129, %r1000, %r1873; + mul.f32 %r2130, %r1000, %r1874; + mul.f32 %r2131, %r1000, %r1875; + mul.f32 %r2132, %r1001, %r1876; + mul.f32 %r2133, %r1001, %r1877; + mul.f32 %r2134, %r1001, %r1878; + mul.f32 %r2135, %r1001, %r1879; + mul.f32 %r2136, %r1001, %r1880; + mul.f32 %r2137, %r1001, %r1881; + mul.f32 %r2138, %r1001, %r1882; + mul.f32 %r2139, %r1001, %r1883; + mul.f32 %r2140, %r1002, %r1884; + mul.f32 %r2141, %r1002, %r1885; + mul.f32 %r2142, %r1002, %r1886; + mul.f32 %r2143, %r1002, %r1887; + mul.f32 %r2144, %r1002, %r1888; + mul.f32 %r2145, %r1002, %r1889; + mul.f32 %r2146, %r1002, %r1890; + mul.f32 %r2147, %r1002, %r1891; + mul.f32 %r2148, %r1003, %r1892; + mul.f32 %r2149, %r1003, %r1893; + mul.f32 %r2150, %r1003, %r1894; + mul.f32 %r2151, %r1003, %r1895; + mul.f32 %r2152, %r1003, %r1896; + mul.f32 %r2153, %r1003, %r1897; + mul.f32 %r2154, %r1003, %r1898; + mul.f32 %r2155, %r1003, %r1899; + mul.f32 %r2156, %r1004, %r1900; + mul.f32 %r2157, %r1004, %r1901; + mul.f32 %r2158, %r1004, %r1902; + mul.f32 %r2159, %r1004, %r1903; + mul.f32 %r2160, %r1004, %r1904; + mul.f32 %r2161, %r1004, %r1905; + mul.f32 %r2162, %r1004, %r1906; + mul.f32 %r2163, %r1004, %r1907; + mul.f32 %r2164, %r1005, %r1908; + mul.f32 %r2165, %r1005, %r1909; + mul.f32 %r2166, %r1005, %r1910; + mul.f32 %r2167, %r1005, %r1911; + mul.f32 %r2168, %r1005, %r1912; + mul.f32 %r2169, %r1005, %r1913; + mul.f32 %r2170, %r1005, %r1914; + mul.f32 %r2171, %r1005, %r1915; + mul.f32 %r2172, %r1006, %r1916; + mul.f32 %r2173, %r1006, %r1917; + mul.f32 %r2174, %r1006, %r1918; + mul.f32 %r2175, %r1006, %r1919; + mul.f32 %r2176, %r1006, %r1920; + mul.f32 %r2177, %r1006, %r1921; + mul.f32 %r2178, %r1006, %r1922; + mul.f32 %r2179, %r1006, %r1923; + mul.f32 %r2180, %r1007, %r1924; + mul.f32 %r2181, %r1007, %r1925; + mul.f32 %r2182, %r1007, %r1926; + mul.f32 %r2183, %r1007, %r1927; + mul.f32 %r2184, %r1007, %r1928; + mul.f32 %r2185, %r1007, %r1929; + mul.f32 %r2186, %r1007, %r1930; + mul.f32 %r2187, %r1007, %r1931; + mul.f32 %r2188, %r1008, %r1932; + mul.f32 %r2189, %r1008, %r1933; + mul.f32 %r2190, %r1008, %r1934; + mul.f32 %r2191, %r1008, %r1935; + mul.f32 %r2192, %r1008, %r1936; + mul.f32 %r2193, %r1008, %r1937; + mul.f32 %r2194, %r1008, %r1938; + mul.f32 %r2195, %r1008, %r1939; + mul.f32 %r2196, %r1009, %r1940; + mul.f32 %r2197, %r1009, %r1941; + mul.f32 %r2198, %r1009, %r1942; + mul.f32 %r2199, %r1009, %r1943; + mul.f32 %r2200, %r1009, %r1944; + mul.f32 %r2201, %r1009, %r1945; + mul.f32 %r2202, %r1009, %r1946; + mul.f32 %r2203, %r1009, %r1947; + mul.f32 %r2204, %r1010, %r1948; + mul.f32 %r2205, %r1010, %r1949; + mul.f32 %r2206, %r1010, %r1950; + mul.f32 %r2207, %r1010, %r1951; + mul.f32 %r2208, %r1010, %r1952; + mul.f32 %r2209, %r1010, %r1953; + mul.f32 %r2210, %r1010, %r1954; + mul.f32 %r2211, %r1010, %r1955; + mul.f32 %r2212, %r1011, %r1956; + mul.f32 %r2213, %r1011, %r1957; + mul.f32 %r2214, %r1011, %r1958; + mul.f32 %r2215, %r1011, %r1959; + mul.f32 %r2216, %r1011, %r1960; + mul.f32 %r2217, %r1011, %r1961; + mul.f32 %r2218, %r1011, %r1962; + mul.f32 %r2219, %r1011, %r1963; + mul.f32 %r2220, %r1012, %r1964; + mul.f32 %r2221, %r1012, %r1965; + mul.f32 %r2222, %r1012, %r1966; + mul.f32 %r2223, %r1012, %r1967; + mul.f32 %r2224, %r1012, %r1968; + mul.f32 %r2225, %r1012, %r1969; + mul.f32 %r2226, %r1012, %r1970; + mul.f32 %r2227, %r1012, %r1971; + mul.f32 %r2228, %r1013, %r1972; + mul.f32 %r2229, %r1013, %r1973; + mul.f32 %r2230, %r1013, %r1974; + mul.f32 %r2231, %r1013, %r1975; + mul.f32 %r2232, %r1013, %r1976; + mul.f32 %r2233, %r1013, %r1977; + mul.f32 %r2234, %r1013, %r1978; + mul.f32 %r2235, %r1013, %r1979; + mul.f32 %r2236, %r1014, %r1980; + mul.f32 %r2237, %r1014, %r1981; + mul.f32 %r2238, %r1014, %r1982; + mul.f32 %r2239, %r1014, %r1983; + mul.f32 %r2240, %r1014, %r1984; + mul.f32 %r2241, %r1014, %r1985; + mul.f32 %r2242, %r1014, %r1986; + mul.f32 %r2243, %r1014, %r1987; + mul.f32 %r2244, %r1015, %r1988; + mul.f32 %r2245, %r1015, %r1989; + mul.f32 %r2246, %r1015, %r1990; + mul.f32 %r2247, %r1015, %r1991; + mul.f32 %r2248, %r1015, %r1992; + mul.f32 %r2249, %r1015, %r1993; + mul.f32 %r2250, %r1015, %r1994; + mul.f32 %r2251, %r1015, %r1995; + mul.f32 %r2252, %r1016, %r1996; + mul.f32 %r2253, %r1016, %r1997; + mul.f32 %r2254, %r1016, %r1998; + mul.f32 %r2255, %r1016, %r1999; + mul.f32 %r2256, %r1016, %r2000; + mul.f32 %r2257, %r1016, %r2001; + mul.f32 %r2258, %r1016, %r2002; + mul.f32 %r2259, %r1016, %r2003; + mul.f32 %r2260, %r1017, %r2004; + mul.f32 %r2261, %r1017, %r2005; + mul.f32 %r2262, %r1017, %r2006; + mul.f32 %r2263, %r1017, %r2007; + mul.f32 %r2264, %r1017, %r2008; + mul.f32 %r2265, %r1017, %r2009; + mul.f32 %r2266, %r1017, %r2010; + mul.f32 %r2267, %r1017, %r2011; + mul.f32 %r2268, %r1018, %r2012; + mul.f32 %r2269, %r1018, %r2013; + mul.f32 %r2270, %r1018, %r2014; + mul.f32 %r2271, %r1018, %r2015; + mul.f32 %r2272, %r1018, %r2016; + mul.f32 %r2273, %r1018, %r2017; + mul.f32 %r2274, %r1018, %r2018; + mul.f32 %r2275, %r1018, %r2019; + mul.f32 %r2276, %r1019, %r2020; + mul.f32 %r2277, %r1019, %r2021; + mul.f32 %r2278, %r1019, %r2022; + mul.f32 %r2279, %r1019, %r2023; + mul.f32 %r2280, %r1019, %r2024; + mul.f32 %r2281, %r1019, %r2025; + mul.f32 %r2282, %r1019, %r2026; + mul.f32 %r2283, %r1019, %r2027; + .loc 1 107 36 // l2norm.py:107:36 + neg.f32 %r2284, %r2028; + fma.rn.f32 %r2285, %r988, %r717, %r2284; + neg.f32 %r2286, %r2029; + fma.rn.f32 %r2287, %r988, %r718, %r2286; + neg.f32 %r2288, %r2030; + fma.rn.f32 %r2289, %r988, %r719, %r2288; + neg.f32 %r2290, %r2031; + fma.rn.f32 %r2291, %r988, %r720, %r2290; + neg.f32 %r2292, %r2032; + fma.rn.f32 %r2293, %r988, %r721, %r2292; + neg.f32 %r2294, %r2033; + fma.rn.f32 %r2295, %r988, %r722, %r2294; + neg.f32 %r2296, %r2034; + fma.rn.f32 %r2297, %r988, %r723, %r2296; + neg.f32 %r2298, %r2035; + fma.rn.f32 %r2299, %r988, %r724, %r2298; + neg.f32 %r2300, %r2036; + fma.rn.f32 %r2301, %r989, %r725, %r2300; + neg.f32 %r2302, %r2037; + fma.rn.f32 %r2303, %r989, %r726, %r2302; + neg.f32 %r2304, %r2038; + fma.rn.f32 %r2305, %r989, %r727, %r2304; + neg.f32 %r2306, %r2039; + fma.rn.f32 %r2307, %r989, %r728, %r2306; + neg.f32 %r2308, %r2040; + fma.rn.f32 %r2309, %r989, %r729, %r2308; + neg.f32 %r2310, %r2041; + fma.rn.f32 %r2311, %r989, %r730, %r2310; + neg.f32 %r2312, %r2042; + fma.rn.f32 %r2313, %r989, %r731, %r2312; + neg.f32 %r2314, %r2043; + fma.rn.f32 %r2315, %r989, %r732, %r2314; + mov.b64 %rd304, {%r1020, %r1021}; + mov.b64 %rd305, {%r2044, %r2045}; + sub.f32x2 %rd306, %rd304, %rd305; + mov.b64 %rd307, {%r1022, %r1023}; + mov.b64 %rd308, {%r2046, %r2047}; + sub.f32x2 %rd309, %rd307, %rd308; + mov.b64 %rd310, {%r1024, %r1025}; + mov.b64 %rd311, {%r2048, %r2049}; + sub.f32x2 %rd312, %rd310, %rd311; + mov.b64 %rd313, {%r1026, %r1027}; + mov.b64 %rd314, {%r2050, %r2051}; + sub.f32x2 %rd315, %rd313, %rd314; + mov.b64 %rd316, {%r1028, %r1029}; + mov.b64 %rd317, {%r2052, %r2053}; + sub.f32x2 %rd318, %rd316, %rd317; + mov.b64 %rd319, {%r1030, %r1031}; + mov.b64 %rd320, {%r2054, %r2055}; + sub.f32x2 %rd321, %rd319, %rd320; + mov.b64 %rd322, {%r1032, %r1033}; + mov.b64 %rd323, {%r2056, %r2057}; + sub.f32x2 %rd324, %rd322, %rd323; + mov.b64 %rd325, {%r1034, %r1035}; + mov.b64 %rd326, {%r2058, %r2059}; + sub.f32x2 %rd327, %rd325, %rd326; + mov.b64 %rd328, {%r1036, %r1037}; + mov.b64 %rd329, {%r2060, %r2061}; + sub.f32x2 %rd330, %rd328, %rd329; + mov.b64 %rd331, {%r1038, %r1039}; + mov.b64 %rd332, {%r2062, %r2063}; + sub.f32x2 %rd333, %rd331, %rd332; + mov.b64 %rd334, {%r1040, %r1041}; + mov.b64 %rd335, {%r2064, %r2065}; + sub.f32x2 %rd336, %rd334, %rd335; + mov.b64 %rd337, {%r1042, %r1043}; + mov.b64 %rd338, {%r2066, %r2067}; + sub.f32x2 %rd339, %rd337, %rd338; + mov.b64 %rd340, {%r1044, %r1045}; + mov.b64 %rd341, {%r2068, %r2069}; + sub.f32x2 %rd342, %rd340, %rd341; + mov.b64 %rd343, {%r1046, %r1047}; + mov.b64 %rd344, {%r2070, %r2071}; + sub.f32x2 %rd345, %rd343, %rd344; + mov.b64 %rd346, {%r1048, %r1049}; + mov.b64 %rd347, {%r2072, %r2073}; + sub.f32x2 %rd348, %rd346, %rd347; + mov.b64 %rd349, {%r1050, %r1051}; + mov.b64 %rd350, {%r2074, %r2075}; + sub.f32x2 %rd351, %rd349, %rd350; + mov.b64 %rd352, {%r1052, %r1053}; + mov.b64 %rd353, {%r2076, %r2077}; + sub.f32x2 %rd354, %rd352, %rd353; + mov.b64 %rd355, {%r1054, %r1055}; + mov.b64 %rd356, {%r2078, %r2079}; + sub.f32x2 %rd357, %rd355, %rd356; + mov.b64 %rd358, {%r1056, %r1057}; + mov.b64 %rd359, {%r2080, %r2081}; + sub.f32x2 %rd360, %rd358, %rd359; + mov.b64 %rd361, {%r1058, %r1059}; + mov.b64 %rd362, {%r2082, %r2083}; + sub.f32x2 %rd363, %rd361, %rd362; + mov.b64 %rd364, {%r1060, %r1061}; + mov.b64 %rd365, {%r2084, %r2085}; + sub.f32x2 %rd366, %rd364, %rd365; + mov.b64 %rd367, {%r1062, %r1063}; + mov.b64 %rd368, {%r2086, %r2087}; + sub.f32x2 %rd369, %rd367, %rd368; + mov.b64 %rd370, {%r1064, %r1065}; + mov.b64 %rd371, {%r2088, %r2089}; + sub.f32x2 %rd372, %rd370, %rd371; + mov.b64 %rd373, {%r1066, %r1067}; + mov.b64 %rd374, {%r2090, %r2091}; + sub.f32x2 %rd375, %rd373, %rd374; + mov.b64 %rd376, {%r1068, %r1069}; + mov.b64 %rd377, {%r2092, %r2093}; + sub.f32x2 %rd378, %rd376, %rd377; + mov.b64 %rd379, {%r1070, %r1071}; + mov.b64 %rd380, {%r2094, %r2095}; + sub.f32x2 %rd381, %rd379, %rd380; + mov.b64 %rd382, {%r1072, %r1073}; + mov.b64 %rd383, {%r2096, %r2097}; + sub.f32x2 %rd384, %rd382, %rd383; + mov.b64 %rd385, {%r1074, %r1075}; + mov.b64 %rd386, {%r2098, %r2099}; + sub.f32x2 %rd387, %rd385, %rd386; + mov.b64 %rd388, {%r1076, %r1077}; + mov.b64 %rd389, {%r2100, %r2101}; + sub.f32x2 %rd390, %rd388, %rd389; + mov.b64 %rd391, {%r1078, %r1079}; + mov.b64 %rd392, {%r2102, %r2103}; + sub.f32x2 %rd393, %rd391, %rd392; + mov.b64 %rd394, {%r1080, %r1081}; + mov.b64 %rd395, {%r2104, %r2105}; + sub.f32x2 %rd396, %rd394, %rd395; + mov.b64 %rd397, {%r1082, %r1083}; + mov.b64 %rd398, {%r2106, %r2107}; + sub.f32x2 %rd399, %rd397, %rd398; + mov.b64 %rd400, {%r1084, %r1085}; + mov.b64 %rd401, {%r2108, %r2109}; + sub.f32x2 %rd402, %rd400, %rd401; + mov.b64 %rd403, {%r1086, %r1087}; + mov.b64 %rd404, {%r2110, %r2111}; + sub.f32x2 %rd405, %rd403, %rd404; + mov.b64 %rd406, {%r1088, %r1089}; + mov.b64 %rd407, {%r2112, %r2113}; + sub.f32x2 %rd408, %rd406, %rd407; + mov.b64 %rd409, {%r1090, %r1091}; + mov.b64 %rd410, {%r2114, %r2115}; + sub.f32x2 %rd411, %rd409, %rd410; + mov.b64 %rd412, {%r1092, %r1093}; + mov.b64 %rd413, {%r2116, %r2117}; + sub.f32x2 %rd414, %rd412, %rd413; + mov.b64 %rd415, {%r1094, %r1095}; + mov.b64 %rd416, {%r2118, %r2119}; + sub.f32x2 %rd417, %rd415, %rd416; + mov.b64 %rd418, {%r1096, %r1097}; + mov.b64 %rd419, {%r2120, %r2121}; + sub.f32x2 %rd420, %rd418, %rd419; + mov.b64 %rd421, {%r1098, %r1099}; + mov.b64 %rd422, {%r2122, %r2123}; + sub.f32x2 %rd423, %rd421, %rd422; + mov.b64 %rd424, {%r1100, %r1101}; + mov.b64 %rd425, {%r2124, %r2125}; + sub.f32x2 %rd426, %rd424, %rd425; + mov.b64 %rd427, {%r1102, %r1103}; + mov.b64 %rd428, {%r2126, %r2127}; + sub.f32x2 %rd429, %rd427, %rd428; + mov.b64 %rd430, {%r1104, %r1105}; + mov.b64 %rd431, {%r2128, %r2129}; + sub.f32x2 %rd432, %rd430, %rd431; + mov.b64 %rd433, {%r1106, %r1107}; + mov.b64 %rd434, {%r2130, %r2131}; + sub.f32x2 %rd435, %rd433, %rd434; + mov.b64 %rd436, {%r1108, %r1109}; + mov.b64 %rd437, {%r2132, %r2133}; + sub.f32x2 %rd438, %rd436, %rd437; + mov.b64 %rd439, {%r1110, %r1111}; + mov.b64 %rd440, {%r2134, %r2135}; + sub.f32x2 %rd441, %rd439, %rd440; + mov.b64 %rd442, {%r1112, %r1113}; + mov.b64 %rd443, {%r2136, %r2137}; + sub.f32x2 %rd444, %rd442, %rd443; + mov.b64 %rd445, {%r1114, %r1115}; + mov.b64 %rd446, {%r2138, %r2139}; + sub.f32x2 %rd447, %rd445, %rd446; + mov.b64 %rd448, {%r1116, %r1117}; + mov.b64 %rd449, {%r2140, %r2141}; + sub.f32x2 %rd450, %rd448, %rd449; + mov.b64 %rd451, {%r1118, %r1119}; + mov.b64 %rd452, {%r2142, %r2143}; + sub.f32x2 %rd453, %rd451, %rd452; + mov.b64 %rd454, {%r1120, %r1121}; + mov.b64 %rd455, {%r2144, %r2145}; + sub.f32x2 %rd456, %rd454, %rd455; + mov.b64 %rd457, {%r1122, %r1123}; + mov.b64 %rd458, {%r2146, %r2147}; + sub.f32x2 %rd459, %rd457, %rd458; + mov.b64 %rd460, {%r1124, %r1125}; + mov.b64 %rd461, {%r2148, %r2149}; + sub.f32x2 %rd462, %rd460, %rd461; + mov.b64 %rd463, {%r1126, %r1127}; + mov.b64 %rd464, {%r2150, %r2151}; + sub.f32x2 %rd465, %rd463, %rd464; + mov.b64 %rd466, {%r1128, %r1129}; + mov.b64 %rd467, {%r2152, %r2153}; + sub.f32x2 %rd468, %rd466, %rd467; + mov.b64 %rd469, {%r1130, %r1131}; + mov.b64 %rd470, {%r2154, %r2155}; + sub.f32x2 %rd471, %rd469, %rd470; + mov.b64 %rd472, {%r1132, %r1133}; + mov.b64 %rd473, {%r2156, %r2157}; + sub.f32x2 %rd474, %rd472, %rd473; + mov.b64 %rd475, {%r1134, %r1135}; + mov.b64 %rd476, {%r2158, %r2159}; + sub.f32x2 %rd477, %rd475, %rd476; + mov.b64 %rd478, {%r1136, %r1137}; + mov.b64 %rd479, {%r2160, %r2161}; + sub.f32x2 %rd480, %rd478, %rd479; + mov.b64 %rd481, {%r1138, %r1139}; + mov.b64 %rd482, {%r2162, %r2163}; + sub.f32x2 %rd483, %rd481, %rd482; + mov.b64 %rd484, {%r1140, %r1141}; + mov.b64 %rd485, {%r2164, %r2165}; + sub.f32x2 %rd486, %rd484, %rd485; + mov.b64 %rd487, {%r1142, %r1143}; + mov.b64 %rd488, {%r2166, %r2167}; + sub.f32x2 %rd489, %rd487, %rd488; + mov.b64 %rd490, {%r1144, %r1145}; + mov.b64 %rd491, {%r2168, %r2169}; + sub.f32x2 %rd492, %rd490, %rd491; + mov.b64 %rd493, {%r1146, %r1147}; + mov.b64 %rd494, {%r2170, %r2171}; + sub.f32x2 %rd495, %rd493, %rd494; + mov.b64 %rd496, {%r1148, %r1149}; + mov.b64 %rd497, {%r2172, %r2173}; + sub.f32x2 %rd498, %rd496, %rd497; + mov.b64 %rd499, {%r1150, %r1151}; + mov.b64 %rd500, {%r2174, %r2175}; + sub.f32x2 %rd501, %rd499, %rd500; + mov.b64 %rd502, {%r1152, %r1153}; + mov.b64 %rd503, {%r2176, %r2177}; + sub.f32x2 %rd504, %rd502, %rd503; + mov.b64 %rd505, {%r1154, %r1155}; + mov.b64 %rd506, {%r2178, %r2179}; + sub.f32x2 %rd507, %rd505, %rd506; + mov.b64 %rd508, {%r1156, %r1157}; + mov.b64 %rd509, {%r2180, %r2181}; + sub.f32x2 %rd510, %rd508, %rd509; + mov.b64 %rd511, {%r1158, %r1159}; + mov.b64 %rd512, {%r2182, %r2183}; + sub.f32x2 %rd513, %rd511, %rd512; + mov.b64 %rd514, {%r1160, %r1161}; + mov.b64 %rd515, {%r2184, %r2185}; + sub.f32x2 %rd516, %rd514, %rd515; + mov.b64 %rd517, {%r1162, %r1163}; + mov.b64 %rd518, {%r2186, %r2187}; + sub.f32x2 %rd519, %rd517, %rd518; + mov.b64 %rd520, {%r1164, %r1165}; + mov.b64 %rd521, {%r2188, %r2189}; + sub.f32x2 %rd522, %rd520, %rd521; + mov.b64 %rd523, {%r1166, %r1167}; + mov.b64 %rd524, {%r2190, %r2191}; + sub.f32x2 %rd525, %rd523, %rd524; + mov.b64 %rd526, {%r1168, %r1169}; + mov.b64 %rd527, {%r2192, %r2193}; + sub.f32x2 %rd528, %rd526, %rd527; + mov.b64 %rd529, {%r1170, %r1171}; + mov.b64 %rd530, {%r2194, %r2195}; + sub.f32x2 %rd531, %rd529, %rd530; + mov.b64 %rd532, {%r1172, %r1173}; + mov.b64 %rd533, {%r2196, %r2197}; + sub.f32x2 %rd534, %rd532, %rd533; + mov.b64 %rd535, {%r1174, %r1175}; + mov.b64 %rd536, {%r2198, %r2199}; + sub.f32x2 %rd537, %rd535, %rd536; + mov.b64 %rd538, {%r1176, %r1177}; + mov.b64 %rd539, {%r2200, %r2201}; + sub.f32x2 %rd540, %rd538, %rd539; + mov.b64 %rd541, {%r1178, %r1179}; + mov.b64 %rd542, {%r2202, %r2203}; + sub.f32x2 %rd543, %rd541, %rd542; + mov.b64 %rd544, {%r1180, %r1181}; + mov.b64 %rd545, {%r2204, %r2205}; + sub.f32x2 %rd546, %rd544, %rd545; + mov.b64 %rd547, {%r1182, %r1183}; + mov.b64 %rd548, {%r2206, %r2207}; + sub.f32x2 %rd549, %rd547, %rd548; + mov.b64 %rd550, {%r1184, %r1185}; + mov.b64 %rd551, {%r2208, %r2209}; + sub.f32x2 %rd552, %rd550, %rd551; + mov.b64 %rd553, {%r1186, %r1187}; + mov.b64 %rd554, {%r2210, %r2211}; + sub.f32x2 %rd555, %rd553, %rd554; + mov.b64 %rd556, {%r1188, %r1189}; + mov.b64 %rd557, {%r2212, %r2213}; + sub.f32x2 %rd558, %rd556, %rd557; + mov.b64 %rd559, {%r1190, %r1191}; + mov.b64 %rd560, {%r2214, %r2215}; + sub.f32x2 %rd561, %rd559, %rd560; + mov.b64 %rd562, {%r1192, %r1193}; + mov.b64 %rd563, {%r2216, %r2217}; + sub.f32x2 %rd564, %rd562, %rd563; + mov.b64 %rd565, {%r1194, %r1195}; + mov.b64 %rd566, {%r2218, %r2219}; + sub.f32x2 %rd567, %rd565, %rd566; + mov.b64 %rd568, {%r1196, %r1197}; + mov.b64 %rd569, {%r2220, %r2221}; + sub.f32x2 %rd570, %rd568, %rd569; + mov.b64 %rd571, {%r1198, %r1199}; + mov.b64 %rd572, {%r2222, %r2223}; + sub.f32x2 %rd573, %rd571, %rd572; + mov.b64 %rd574, {%r1200, %r1201}; + mov.b64 %rd575, {%r2224, %r2225}; + sub.f32x2 %rd576, %rd574, %rd575; + mov.b64 %rd577, {%r1202, %r1203}; + mov.b64 %rd578, {%r2226, %r2227}; + sub.f32x2 %rd579, %rd577, %rd578; + mov.b64 %rd580, {%r1204, %r1205}; + mov.b64 %rd581, {%r2228, %r2229}; + sub.f32x2 %rd582, %rd580, %rd581; + mov.b64 %rd583, {%r1206, %r1207}; + mov.b64 %rd584, {%r2230, %r2231}; + sub.f32x2 %rd585, %rd583, %rd584; + mov.b64 %rd586, {%r1208, %r1209}; + mov.b64 %rd587, {%r2232, %r2233}; + sub.f32x2 %rd588, %rd586, %rd587; + mov.b64 %rd589, {%r1210, %r1211}; + mov.b64 %rd590, {%r2234, %r2235}; + sub.f32x2 %rd591, %rd589, %rd590; + mov.b64 %rd592, {%r1212, %r1213}; + mov.b64 %rd593, {%r2236, %r2237}; + sub.f32x2 %rd594, %rd592, %rd593; + mov.b64 %rd595, {%r1214, %r1215}; + mov.b64 %rd596, {%r2238, %r2239}; + sub.f32x2 %rd597, %rd595, %rd596; + mov.b64 %rd598, {%r1216, %r1217}; + mov.b64 %rd599, {%r2240, %r2241}; + sub.f32x2 %rd600, %rd598, %rd599; + mov.b64 %rd601, {%r1218, %r1219}; + mov.b64 %rd602, {%r2242, %r2243}; + sub.f32x2 %rd603, %rd601, %rd602; + mov.b64 %rd604, {%r1220, %r1221}; + mov.b64 %rd605, {%r2244, %r2245}; + sub.f32x2 %rd606, %rd604, %rd605; + mov.b64 %rd607, {%r1222, %r1223}; + mov.b64 %rd608, {%r2246, %r2247}; + sub.f32x2 %rd609, %rd607, %rd608; + mov.b64 %rd610, {%r1224, %r1225}; + mov.b64 %rd611, {%r2248, %r2249}; + sub.f32x2 %rd612, %rd610, %rd611; + mov.b64 %rd613, {%r1226, %r1227}; + mov.b64 %rd614, {%r2250, %r2251}; + sub.f32x2 %rd615, %rd613, %rd614; + mov.b64 %rd616, {%r1228, %r1229}; + mov.b64 %rd617, {%r2252, %r2253}; + sub.f32x2 %rd618, %rd616, %rd617; + mov.b64 %rd619, {%r1230, %r1231}; + mov.b64 %rd620, {%r2254, %r2255}; + sub.f32x2 %rd621, %rd619, %rd620; + mov.b64 %rd622, {%r1232, %r1233}; + mov.b64 %rd623, {%r2256, %r2257}; + sub.f32x2 %rd624, %rd622, %rd623; + mov.b64 %rd625, {%r1234, %r1235}; + mov.b64 %rd626, {%r2258, %r2259}; + sub.f32x2 %rd627, %rd625, %rd626; + mov.b64 %rd628, {%r1236, %r1237}; + mov.b64 %rd629, {%r2260, %r2261}; + sub.f32x2 %rd630, %rd628, %rd629; + mov.b64 %rd631, {%r1238, %r1239}; + mov.b64 %rd632, {%r2262, %r2263}; + sub.f32x2 %rd633, %rd631, %rd632; + mov.b64 %rd634, {%r1240, %r1241}; + mov.b64 %rd635, {%r2264, %r2265}; + sub.f32x2 %rd636, %rd634, %rd635; + mov.b64 %rd637, {%r1242, %r1243}; + mov.b64 %rd638, {%r2266, %r2267}; + sub.f32x2 %rd639, %rd637, %rd638; + mov.b64 %rd640, {%r1244, %r1245}; + mov.b64 %rd641, {%r2268, %r2269}; + sub.f32x2 %rd642, %rd640, %rd641; + mov.b64 %rd643, {%r1246, %r1247}; + mov.b64 %rd644, {%r2270, %r2271}; + sub.f32x2 %rd645, %rd643, %rd644; + mov.b64 %rd646, {%r1248, %r1249}; + mov.b64 %rd647, {%r2272, %r2273}; + sub.f32x2 %rd648, %rd646, %rd647; + mov.b64 %rd649, {%r1250, %r1251}; + mov.b64 %rd650, {%r2274, %r2275}; + sub.f32x2 %rd651, %rd649, %rd650; + mov.b64 %rd652, {%r1252, %r1253}; + mov.b64 %rd653, {%r2276, %r2277}; + sub.f32x2 %rd654, %rd652, %rd653; + mov.b64 %rd655, {%r1254, %r1255}; + mov.b64 %rd656, {%r2278, %r2279}; + sub.f32x2 %rd657, %rd655, %rd656; + mov.b64 %rd658, {%r1256, %r1257}; + mov.b64 %rd659, {%r2280, %r2281}; + sub.f32x2 %rd660, %rd658, %rd659; + mov.b64 %rd661, {%r1258, %r1259}; + mov.b64 %rd662, {%r2282, %r2283}; + sub.f32x2 %rd663, %rd661, %rd662; + .loc 1 108 27 // l2norm.py:108:27 + mov.b64 {%r2316, %r2317}, %rd306; + cvt.rn.bf16x2.f32 %r270, %r2317, %r2316; + mov.b64 {%r2318, %r2319}, %rd309; + cvt.rn.bf16x2.f32 %r271, %r2319, %r2318; + mov.b64 {%r2320, %r2321}, %rd312; + cvt.rn.bf16x2.f32 %r272, %r2321, %r2320; + mov.b64 {%r2322, %r2323}, %rd315; + cvt.rn.bf16x2.f32 %r273, %r2323, %r2322; + mov.b64 {%r2324, %r2325}, %rd318; + cvt.rn.bf16x2.f32 %r274, %r2325, %r2324; + mov.b64 {%r2326, %r2327}, %rd321; + cvt.rn.bf16x2.f32 %r275, %r2327, %r2326; + mov.b64 {%r2328, %r2329}, %rd324; + cvt.rn.bf16x2.f32 %r276, %r2329, %r2328; + mov.b64 {%r2330, %r2331}, %rd327; + cvt.rn.bf16x2.f32 %r277, %r2331, %r2330; + mov.b64 {%r2332, %r2333}, %rd330; + cvt.rn.bf16x2.f32 %r278, %r2333, %r2332; + mov.b64 {%r2334, %r2335}, %rd333; + cvt.rn.bf16x2.f32 %r279, %r2335, %r2334; + mov.b64 {%r2336, %r2337}, %rd336; + cvt.rn.bf16x2.f32 %r280, %r2337, %r2336; + mov.b64 {%r2338, %r2339}, %rd339; + cvt.rn.bf16x2.f32 %r281, %r2339, %r2338; + mov.b64 {%r2340, %r2341}, %rd342; + cvt.rn.bf16x2.f32 %r282, %r2341, %r2340; + mov.b64 {%r2342, %r2343}, %rd345; + cvt.rn.bf16x2.f32 %r283, %r2343, %r2342; + mov.b64 {%r2344, %r2345}, %rd348; + cvt.rn.bf16x2.f32 %r284, %r2345, %r2344; + mov.b64 {%r2346, %r2347}, %rd351; + cvt.rn.bf16x2.f32 %r285, %r2347, %r2346; + mov.b64 {%r2348, %r2349}, %rd354; + cvt.rn.bf16x2.f32 %r286, %r2349, %r2348; + mov.b64 {%r2350, %r2351}, %rd357; + cvt.rn.bf16x2.f32 %r287, %r2351, %r2350; + mov.b64 {%r2352, %r2353}, %rd360; + cvt.rn.bf16x2.f32 %r288, %r2353, %r2352; + mov.b64 {%r2354, %r2355}, %rd363; + cvt.rn.bf16x2.f32 %r289, %r2355, %r2354; + mov.b64 {%r2356, %r2357}, %rd366; + cvt.rn.bf16x2.f32 %r290, %r2357, %r2356; + mov.b64 {%r2358, %r2359}, %rd369; + cvt.rn.bf16x2.f32 %r291, %r2359, %r2358; + mov.b64 {%r2360, %r2361}, %rd372; + cvt.rn.bf16x2.f32 %r292, %r2361, %r2360; + mov.b64 {%r2362, %r2363}, %rd375; + cvt.rn.bf16x2.f32 %r293, %r2363, %r2362; + mov.b64 {%r2364, %r2365}, %rd378; + cvt.rn.bf16x2.f32 %r294, %r2365, %r2364; + mov.b64 {%r2366, %r2367}, %rd381; + cvt.rn.bf16x2.f32 %r295, %r2367, %r2366; + mov.b64 {%r2368, %r2369}, %rd384; + cvt.rn.bf16x2.f32 %r296, %r2369, %r2368; + mov.b64 {%r2370, %r2371}, %rd387; + cvt.rn.bf16x2.f32 %r297, %r2371, %r2370; + mov.b64 {%r2372, %r2373}, %rd390; + cvt.rn.bf16x2.f32 %r298, %r2373, %r2372; + mov.b64 {%r2374, %r2375}, %rd393; + cvt.rn.bf16x2.f32 %r299, %r2375, %r2374; + mov.b64 {%r2376, %r2377}, %rd396; + cvt.rn.bf16x2.f32 %r300, %r2377, %r2376; + mov.b64 {%r2378, %r2379}, %rd399; + cvt.rn.bf16x2.f32 %r301, %r2379, %r2378; + mov.b64 {%r2380, %r2381}, %rd402; + cvt.rn.bf16x2.f32 %r302, %r2381, %r2380; + mov.b64 {%r2382, %r2383}, %rd405; + cvt.rn.bf16x2.f32 %r303, %r2383, %r2382; + mov.b64 {%r2384, %r2385}, %rd408; + cvt.rn.bf16x2.f32 %r304, %r2385, %r2384; + mov.b64 {%r2386, %r2387}, %rd411; + cvt.rn.bf16x2.f32 %r305, %r2387, %r2386; + mov.b64 {%r2388, %r2389}, %rd414; + cvt.rn.bf16x2.f32 %r306, %r2389, %r2388; + mov.b64 {%r2390, %r2391}, %rd417; + cvt.rn.bf16x2.f32 %r307, %r2391, %r2390; + mov.b64 {%r2392, %r2393}, %rd420; + cvt.rn.bf16x2.f32 %r308, %r2393, %r2392; + mov.b64 {%r2394, %r2395}, %rd423; + cvt.rn.bf16x2.f32 %r309, %r2395, %r2394; + mov.b64 {%r2396, %r2397}, %rd426; + cvt.rn.bf16x2.f32 %r310, %r2397, %r2396; + mov.b64 {%r2398, %r2399}, %rd429; + cvt.rn.bf16x2.f32 %r311, %r2399, %r2398; + mov.b64 {%r2400, %r2401}, %rd432; + cvt.rn.bf16x2.f32 %r312, %r2401, %r2400; + mov.b64 {%r2402, %r2403}, %rd435; + cvt.rn.bf16x2.f32 %r313, %r2403, %r2402; + mov.b64 {%r2404, %r2405}, %rd438; + cvt.rn.bf16x2.f32 %r314, %r2405, %r2404; + mov.b64 {%r2406, %r2407}, %rd441; + cvt.rn.bf16x2.f32 %r315, %r2407, %r2406; + mov.b64 {%r2408, %r2409}, %rd444; + cvt.rn.bf16x2.f32 %r316, %r2409, %r2408; + mov.b64 {%r2410, %r2411}, %rd447; + cvt.rn.bf16x2.f32 %r317, %r2411, %r2410; + mov.b64 {%r2412, %r2413}, %rd450; + cvt.rn.bf16x2.f32 %r318, %r2413, %r2412; + mov.b64 {%r2414, %r2415}, %rd453; + cvt.rn.bf16x2.f32 %r319, %r2415, %r2414; + mov.b64 {%r2416, %r2417}, %rd456; + cvt.rn.bf16x2.f32 %r320, %r2417, %r2416; + mov.b64 {%r2418, %r2419}, %rd459; + cvt.rn.bf16x2.f32 %r321, %r2419, %r2418; + mov.b64 {%r2420, %r2421}, %rd462; + cvt.rn.bf16x2.f32 %r322, %r2421, %r2420; + mov.b64 {%r2422, %r2423}, %rd465; + cvt.rn.bf16x2.f32 %r323, %r2423, %r2422; + mov.b64 {%r2424, %r2425}, %rd468; + cvt.rn.bf16x2.f32 %r324, %r2425, %r2424; + mov.b64 {%r2426, %r2427}, %rd471; + cvt.rn.bf16x2.f32 %r325, %r2427, %r2426; + mov.b64 {%r2428, %r2429}, %rd474; + cvt.rn.bf16x2.f32 %r326, %r2429, %r2428; + mov.b64 {%r2430, %r2431}, %rd477; + cvt.rn.bf16x2.f32 %r327, %r2431, %r2430; + mov.b64 {%r2432, %r2433}, %rd480; + cvt.rn.bf16x2.f32 %r328, %r2433, %r2432; + mov.b64 {%r2434, %r2435}, %rd483; + cvt.rn.bf16x2.f32 %r329, %r2435, %r2434; + mov.b64 {%r2436, %r2437}, %rd486; + cvt.rn.bf16x2.f32 %r330, %r2437, %r2436; + mov.b64 {%r2438, %r2439}, %rd489; + cvt.rn.bf16x2.f32 %r331, %r2439, %r2438; + mov.b64 {%r2440, %r2441}, %rd492; + cvt.rn.bf16x2.f32 %r332, %r2441, %r2440; + mov.b64 {%r2442, %r2443}, %rd495; + cvt.rn.bf16x2.f32 %r333, %r2443, %r2442; + mov.b64 {%r2444, %r2445}, %rd498; + cvt.rn.bf16x2.f32 %r334, %r2445, %r2444; + mov.b64 {%r2446, %r2447}, %rd501; + cvt.rn.bf16x2.f32 %r335, %r2447, %r2446; + mov.b64 {%r2448, %r2449}, %rd504; + cvt.rn.bf16x2.f32 %r336, %r2449, %r2448; + mov.b64 {%r2450, %r2451}, %rd507; + cvt.rn.bf16x2.f32 %r337, %r2451, %r2450; + mov.b64 {%r2452, %r2453}, %rd510; + cvt.rn.bf16x2.f32 %r338, %r2453, %r2452; + mov.b64 {%r2454, %r2455}, %rd513; + cvt.rn.bf16x2.f32 %r339, %r2455, %r2454; + mov.b64 {%r2456, %r2457}, %rd516; + cvt.rn.bf16x2.f32 %r340, %r2457, %r2456; + mov.b64 {%r2458, %r2459}, %rd519; + cvt.rn.bf16x2.f32 %r341, %r2459, %r2458; + mov.b64 {%r2460, %r2461}, %rd522; + cvt.rn.bf16x2.f32 %r342, %r2461, %r2460; + mov.b64 {%r2462, %r2463}, %rd525; + cvt.rn.bf16x2.f32 %r343, %r2463, %r2462; + mov.b64 {%r2464, %r2465}, %rd528; + cvt.rn.bf16x2.f32 %r344, %r2465, %r2464; + mov.b64 {%r2466, %r2467}, %rd531; + cvt.rn.bf16x2.f32 %r345, %r2467, %r2466; + mov.b64 {%r2468, %r2469}, %rd534; + cvt.rn.bf16x2.f32 %r346, %r2469, %r2468; + mov.b64 {%r2470, %r2471}, %rd537; + cvt.rn.bf16x2.f32 %r347, %r2471, %r2470; + mov.b64 {%r2472, %r2473}, %rd540; + cvt.rn.bf16x2.f32 %r348, %r2473, %r2472; + mov.b64 {%r2474, %r2475}, %rd543; + cvt.rn.bf16x2.f32 %r349, %r2475, %r2474; + mov.b64 {%r2476, %r2477}, %rd546; + cvt.rn.bf16x2.f32 %r350, %r2477, %r2476; + mov.b64 {%r2478, %r2479}, %rd549; + cvt.rn.bf16x2.f32 %r351, %r2479, %r2478; + mov.b64 {%r2480, %r2481}, %rd552; + cvt.rn.bf16x2.f32 %r352, %r2481, %r2480; + mov.b64 {%r2482, %r2483}, %rd555; + cvt.rn.bf16x2.f32 %r353, %r2483, %r2482; + mov.b64 {%r2484, %r2485}, %rd558; + cvt.rn.bf16x2.f32 %r354, %r2485, %r2484; + mov.b64 {%r2486, %r2487}, %rd561; + cvt.rn.bf16x2.f32 %r355, %r2487, %r2486; + mov.b64 {%r2488, %r2489}, %rd564; + cvt.rn.bf16x2.f32 %r356, %r2489, %r2488; + mov.b64 {%r2490, %r2491}, %rd567; + cvt.rn.bf16x2.f32 %r357, %r2491, %r2490; + mov.b64 {%r2492, %r2493}, %rd570; + cvt.rn.bf16x2.f32 %r358, %r2493, %r2492; + mov.b64 {%r2494, %r2495}, %rd573; + cvt.rn.bf16x2.f32 %r359, %r2495, %r2494; + mov.b64 {%r2496, %r2497}, %rd576; + cvt.rn.bf16x2.f32 %r360, %r2497, %r2496; + mov.b64 {%r2498, %r2499}, %rd579; + cvt.rn.bf16x2.f32 %r361, %r2499, %r2498; + mov.b64 {%r2500, %r2501}, %rd582; + cvt.rn.bf16x2.f32 %r362, %r2501, %r2500; + mov.b64 {%r2502, %r2503}, %rd585; + cvt.rn.bf16x2.f32 %r363, %r2503, %r2502; + mov.b64 {%r2504, %r2505}, %rd588; + cvt.rn.bf16x2.f32 %r364, %r2505, %r2504; + mov.b64 {%r2506, %r2507}, %rd591; + cvt.rn.bf16x2.f32 %r365, %r2507, %r2506; + mov.b64 {%r2508, %r2509}, %rd594; + cvt.rn.bf16x2.f32 %r366, %r2509, %r2508; + mov.b64 {%r2510, %r2511}, %rd597; + cvt.rn.bf16x2.f32 %r367, %r2511, %r2510; + mov.b64 {%r2512, %r2513}, %rd600; + cvt.rn.bf16x2.f32 %r368, %r2513, %r2512; + mov.b64 {%r2514, %r2515}, %rd603; + cvt.rn.bf16x2.f32 %r369, %r2515, %r2514; + mov.b64 {%r2516, %r2517}, %rd606; + cvt.rn.bf16x2.f32 %r370, %r2517, %r2516; + mov.b64 {%r2518, %r2519}, %rd609; + cvt.rn.bf16x2.f32 %r371, %r2519, %r2518; + mov.b64 {%r2520, %r2521}, %rd612; + cvt.rn.bf16x2.f32 %r372, %r2521, %r2520; + mov.b64 {%r2522, %r2523}, %rd615; + cvt.rn.bf16x2.f32 %r373, %r2523, %r2522; + mov.b64 {%r2524, %r2525}, %rd618; + cvt.rn.bf16x2.f32 %r374, %r2525, %r2524; + mov.b64 {%r2526, %r2527}, %rd621; + cvt.rn.bf16x2.f32 %r375, %r2527, %r2526; + mov.b64 {%r2528, %r2529}, %rd624; + cvt.rn.bf16x2.f32 %r376, %r2529, %r2528; + mov.b64 {%r2530, %r2531}, %rd627; + cvt.rn.bf16x2.f32 %r377, %r2531, %r2530; + mov.b64 {%r2532, %r2533}, %rd630; + cvt.rn.bf16x2.f32 %r378, %r2533, %r2532; + mov.b64 {%r2534, %r2535}, %rd633; + cvt.rn.bf16x2.f32 %r379, %r2535, %r2534; + mov.b64 {%r2536, %r2537}, %rd636; + cvt.rn.bf16x2.f32 %r380, %r2537, %r2536; + mov.b64 {%r2538, %r2539}, %rd639; + cvt.rn.bf16x2.f32 %r381, %r2539, %r2538; + mov.b64 {%r2540, %r2541}, %rd642; + cvt.rn.bf16x2.f32 %r382, %r2541, %r2540; + mov.b64 {%r2542, %r2543}, %rd645; + cvt.rn.bf16x2.f32 %r383, %r2543, %r2542; + mov.b64 {%r2544, %r2545}, %rd648; + cvt.rn.bf16x2.f32 %r384, %r2545, %r2544; + mov.b64 {%r2546, %r2547}, %rd651; + cvt.rn.bf16x2.f32 %r385, %r2547, %r2546; + mov.b64 {%r2548, %r2549}, %rd654; + cvt.rn.bf16x2.f32 %r386, %r2549, %r2548; + mov.b64 {%r2550, %r2551}, %rd657; + cvt.rn.bf16x2.f32 %r387, %r2551, %r2550; + mov.b64 {%r2552, %r2553}, %rd660; + cvt.rn.bf16x2.f32 %r388, %r2553, %r2552; + mov.b64 {%r2554, %r2555}, %rd663; + cvt.rn.bf16x2.f32 %r389, %r2555, %r2554; + .loc 1 108 19 // l2norm.py:108:19 + cvt.rn.bf16x2.f32 %r262, %r2287, %r2285; + cvt.rn.bf16x2.f32 %r263, %r2291, %r2289; + cvt.rn.bf16x2.f32 %r264, %r2295, %r2293; + cvt.rn.bf16x2.f32 %r265, %r2299, %r2297; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd67 + 0 ], { %r262, %r263, %r264, %r265 }; + // end inline asm + cvt.rn.bf16x2.f32 %r266, %r2303, %r2301; + cvt.rn.bf16x2.f32 %r267, %r2307, %r2305; + cvt.rn.bf16x2.f32 %r268, %r2311, %r2309; + cvt.rn.bf16x2.f32 %r269, %r2315, %r2313; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd68 + 0 ], { %r266, %r267, %r268, %r269 }; + // end inline asm + // begin inline asm + @%p6 st.global.v4.b32 [ %rd69 + 0 ], { %r270, %r271, %r272, %r273 }; + // end inline asm + // begin inline asm + @%p7 st.global.v4.b32 [ %rd70 + 0 ], { %r274, %r275, %r276, %r277 }; + // end inline asm + // begin inline asm + @%p8 st.global.v4.b32 [ %rd71 + 0 ], { %r278, %r279, %r280, %r281 }; + // end inline asm + // begin inline asm + @%p9 st.global.v4.b32 [ %rd72 + 0 ], { %r282, %r283, %r284, %r285 }; + // end inline asm + // begin inline asm + @%p10 st.global.v4.b32 [ %rd73 + 0 ], { %r286, %r287, %r288, %r289 }; + // end inline asm + // begin inline asm + @%p11 st.global.v4.b32 [ %rd74 + 0 ], { %r290, %r291, %r292, %r293 }; + // end inline asm + // begin inline asm + @%p12 st.global.v4.b32 [ %rd75 + 0 ], { %r294, %r295, %r296, %r297 }; + // end inline asm + // begin inline asm + @%p3 st.global.v4.b32 [ %rd76 + 0 ], { %r298, %r299, %r300, %r301 }; + // end inline asm + // begin inline asm + @%p13 st.global.v4.b32 [ %rd77 + 0 ], { %r302, %r303, %r304, %r305 }; + // end inline asm + // begin inline asm + @%p14 st.global.v4.b32 [ %rd78 + 0 ], { %r306, %r307, %r308, %r309 }; + // end inline asm + // begin inline asm + @%p15 st.global.v4.b32 [ %rd79 + 0 ], { %r310, %r311, %r312, %r313 }; + // end inline asm + // begin inline asm + @%p16 st.global.v4.b32 [ %rd80 + 0 ], { %r314, %r315, %r316, %r317 }; + // end inline asm + // begin inline asm + @%p17 st.global.v4.b32 [ %rd81 + 0 ], { %r318, %r319, %r320, %r321 }; + // end inline asm + // begin inline asm + @%p18 st.global.v4.b32 [ %rd82 + 0 ], { %r322, %r323, %r324, %r325 }; + // end inline asm + // begin inline asm + @%p19 st.global.v4.b32 [ %rd83 + 0 ], { %r326, %r327, %r328, %r329 }; + // end inline asm + // begin inline asm + @%p20 st.global.v4.b32 [ %rd84 + 0 ], { %r330, %r331, %r332, %r333 }; + // end inline asm + // begin inline asm + @%p21 st.global.v4.b32 [ %rd85 + 0 ], { %r334, %r335, %r336, %r337 }; + // end inline asm + // begin inline asm + @%p22 st.global.v4.b32 [ %rd86 + 0 ], { %r338, %r339, %r340, %r341 }; + // end inline asm + // begin inline asm + @%p23 st.global.v4.b32 [ %rd87 + 0 ], { %r342, %r343, %r344, %r345 }; + // end inline asm + // begin inline asm + @%p24 st.global.v4.b32 [ %rd88 + 0 ], { %r346, %r347, %r348, %r349 }; + // end inline asm + // begin inline asm + @%p25 st.global.v4.b32 [ %rd89 + 0 ], { %r350, %r351, %r352, %r353 }; + // end inline asm + // begin inline asm + @%p26 st.global.v4.b32 [ %rd90 + 0 ], { %r354, %r355, %r356, %r357 }; + // end inline asm + // begin inline asm + @%p27 st.global.v4.b32 [ %rd91 + 0 ], { %r358, %r359, %r360, %r361 }; + // end inline asm + // begin inline asm + @%p28 st.global.v4.b32 [ %rd92 + 0 ], { %r362, %r363, %r364, %r365 }; + // end inline asm + // begin inline asm + @%p29 st.global.v4.b32 [ %rd93 + 0 ], { %r366, %r367, %r368, %r369 }; + // end inline asm + // begin inline asm + @%p30 st.global.v4.b32 [ %rd94 + 0 ], { %r370, %r371, %r372, %r373 }; + // end inline asm + // begin inline asm + @%p31 st.global.v4.b32 [ %rd95 + 0 ], { %r374, %r375, %r376, %r377 }; + // end inline asm + // begin inline asm + @%p32 st.global.v4.b32 [ %rd96 + 0 ], { %r378, %r379, %r380, %r381 }; + // end inline asm + // begin inline asm + @%p33 st.global.v4.b32 [ %rd97 + 0 ], { %r382, %r383, %r384, %r385 }; + // end inline asm + // begin inline asm + @%p34 st.global.v4.b32 [ %rd98 + 0 ], { %r386, %r387, %r388, %r389 }; + // end inline asm + .loc 1 108 4 // l2norm.py:108:4 + ret; +$L__tmp259: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 98 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp258 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 107 // DW_AT_call_line +.b8 55 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp258 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.source b/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..35265513bc03aa2afd3347f28178370dc92e754e --- /dev/null +++ b/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.source @@ -0,0 +1,210 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc47 = loc(unknown) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc54 = loc("y"(#loc)) +#loc55 = loc("rstd"(#loc)) +#loc56 = loc("dy"(#loc)) +#loc57 = loc("dx"(#loc)) +#loc58 = loc("eps"(#loc)) +#loc59 = loc("T"(#loc)) +#loc101 = loc("input"(#loc45)) +#loc102 = loc("a"(#loc50)) +#loc103 = loc("b"(#loc50)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc60) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc61) + %o_t = arith.constant 128 : i32 loc(#loc62) + %o_t_1 = arith.constant 128 : i64 loc(#loc62) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc62) + %o_t_3 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc63) + %o_t_4 = arith.extsi %o_t_3 : tensor<128xi32> to tensor<128xi64> loc(#loc64) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<128xi64> loc(#loc64) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<128xi64> loc(#loc64) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc65) + %m_t = arith.extsi %T : i32 to i64 loc(#loc66) + %m_t_7 = tt.splat %m_t : i64 -> tensor<128xi64> loc(#loc66) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<128xi64> loc(#loc66) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<128xi1> -> tensor<128x1xi1> loc(#loc67) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc68) + %m_x_10 = arith.constant 128 : i32 loc(#loc69) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc69) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc69) + %m_x_13 = tt.broadcast %m_x : tensor<128x1xi1> -> tensor<128x128xi1> loc(#loc70) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<128x128xi1> loc(#loc70) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<128x128xi1> loc(#loc70) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc71) + %p_y_16 = arith.constant 128 : i32 loc(#loc72) + %p_y_17 = arith.constant 128 : i64 loc(#loc72) + %p_y_18 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc72) + %p_y_19 = arith.muli %p_y, %p_y_18 : tensor<128x1xi64> loc(#loc72) + %p_y_20 = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc73) + %p_y_21 = tt.addptr %p_y_20, %p_y_19 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc73) + %p_y_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc74) + %p_y_23 = tt.broadcast %p_y_21 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc75) + %p_y_24 = tt.broadcast %p_y_22 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc75) + %p_y_25 = tt.addptr %p_y_23, %p_y_24 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc75) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr> loc(#loc76) + %p_rstd_26 = tt.addptr %p_rstd, %o_t_6 : tensor<128x!tt.ptr>, tensor<128xi64> loc(#loc76) + %p_dy = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc77) + %p_dy_27 = arith.constant 128 : i32 loc(#loc78) + %p_dy_28 = arith.constant 128 : i64 loc(#loc78) + %p_dy_29 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc78) + %p_dy_30 = arith.muli %p_dy, %p_dy_29 : tensor<128x1xi64> loc(#loc78) + %p_dy_31 = tt.splat %dy : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc79) + %p_dy_32 = tt.addptr %p_dy_31, %p_dy_30 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc79) + %p_dy_33 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc80) + %p_dy_34 = tt.broadcast %p_dy_32 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc81) + %p_dy_35 = tt.broadcast %p_dy_33 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc81) + %p_dy_36 = tt.addptr %p_dy_34, %p_dy_35 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc81) + %p_dx = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc82) + %p_dx_37 = arith.constant 128 : i32 loc(#loc83) + %p_dx_38 = arith.constant 128 : i64 loc(#loc83) + %p_dx_39 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc83) + %p_dx_40 = arith.muli %p_dx, %p_dx_39 : tensor<128x1xi64> loc(#loc83) + %p_dx_41 = tt.splat %dx : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc84) + %p_dx_42 = tt.addptr %p_dx_41, %p_dx_40 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc84) + %p_dx_43 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc85) + %p_dx_44 = tt.broadcast %p_dx_42 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc86) + %p_dx_45 = tt.broadcast %p_dx_43 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc86) + %p_dx_46 = tt.addptr %p_dx_44, %p_dx_45 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc86) + %b_y = arith.constant 0.000000e+00 : f32 loc(#loc87) + %b_y_47 = arith.constant dense<0.000000e+00> : tensor<128x128xf32> loc(#loc87) + %b_y_48 = arith.truncf %b_y_47 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc87) + %b_y_49 = tt.load %p_y_25, %m_x_15, %b_y_48 : tensor<128x128x!tt.ptr> loc(#loc87) + %b_y_50 = arith.extf %b_y_49 : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc88) + %b_rstd = arith.constant 0.000000e+00 : f32 loc(#loc89) + %b_rstd_51 = arith.constant dense<0.000000e+00> : tensor<128xf32> loc(#loc89) + %b_rstd_52 = tt.load %p_rstd_26, %m_t_8, %b_rstd_51 : tensor<128x!tt.ptr> loc(#loc89) + %b_dy = arith.constant 0.000000e+00 : f32 loc(#loc90) + %b_dy_53 = arith.constant dense<0.000000e+00> : tensor<128x128xf32> loc(#loc90) + %b_dy_54 = arith.truncf %b_dy_53 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc90) + %b_dy_55 = tt.load %p_dy_36, %m_x_15, %b_dy_54 : tensor<128x128x!tt.ptr> loc(#loc90) + %b_dy_56 = arith.extf %b_dy_55 : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc91) + %b_dx = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc92) + %b_dx_57 = tt.broadcast %b_dx : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc93) + %b_dx_58 = arith.mulf %b_dy_56, %b_dx_57 : tensor<128x128xf32> loc(#loc93) + %b_dx_59 = arith.mulf %b_dy_56, %b_y_50 : tensor<128x128xf32> loc(#loc94) + %b_dx_60 = tt.call @triton.language.standard.sum__fp32S128_128S_c1_cFalse_cNone(%b_dx_59) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc95) + %b_dx_61 = tt.expand_dims %b_dx_60 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc96) + %b_dx_62 = tt.broadcast %b_dx_61 : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc97) + %b_dx_63 = arith.mulf %b_dx_62, %b_y_50 : tensor<128x128xf32> loc(#loc97) + %b_dx_64 = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc98) + %b_dx_65 = tt.broadcast %b_dx_64 : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc99) + %b_dx_66 = arith.mulf %b_dx_63, %b_dx_65 : tensor<128x128xf32> loc(#loc99) + %b_dx_67 = arith.subf %b_dx_58, %b_dx_66 : tensor<128x128xf32> loc(#loc100) + %0 = arith.truncf %b_dx_67 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc42) + tt.store %p_dx_46, %0, %m_x_15 : tensor<128x128x!tt.ptr> loc(#loc43) + tt.return loc(#loc44) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S128_128S_c1_cFalse_cNone(%input: tensor<128x128xf32> loc("input"(#loc45))) -> tensor<128xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc46) + tt.reduce.return %2 : f32 loc(#loc46) + }) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc46) + tt.return %0 : tensor<128xf32> loc(#loc48) + ^bb1: // no predecessors + %1 = ub.poison : tensor<128xf32> loc(#loc49) + tt.return %1 : tensor<128xf32> loc(#loc49) + } loc(#loc45) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc50)), %b: f32 loc("b"(#loc50))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc51) + tt.return %0 : f32 loc(#loc52) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc53) + tt.return %1 : f32 loc(#loc53) + } loc(#loc50) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:31) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:39) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:20) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:31) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:39) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:82) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc60 = loc("i_t"(#loc1)) +#loc61 = loc("i_t"(#loc2)) +#loc62 = loc("o_t"(#loc3)) +#loc63 = loc("o_t"(#loc4)) +#loc64 = loc("o_t"(#loc5)) +#loc65 = loc("o_d"(#loc6)) +#loc66 = loc("m_t"(#loc7)) +#loc67 = loc("m_x"(#loc8)) +#loc68 = loc("m_x"(#loc9)) +#loc69 = loc("m_x"(#loc10)) +#loc70 = loc("m_x"(#loc11)) +#loc71 = loc("p_y"(#loc12)) +#loc72 = loc("p_y"(#loc13)) +#loc73 = loc("p_y"(#loc14)) +#loc74 = loc("p_y"(#loc15)) +#loc75 = loc("p_y"(#loc16)) +#loc76 = loc("p_rstd"(#loc17)) +#loc77 = loc("p_dy"(#loc18)) +#loc78 = loc("p_dy"(#loc19)) +#loc79 = loc("p_dy"(#loc20)) +#loc80 = loc("p_dy"(#loc21)) +#loc81 = loc("p_dy"(#loc22)) +#loc82 = loc("p_dx"(#loc23)) +#loc83 = loc("p_dx"(#loc24)) +#loc84 = loc("p_dx"(#loc25)) +#loc85 = loc("p_dx"(#loc26)) +#loc86 = loc("p_dx"(#loc27)) +#loc87 = loc("b_y"(#loc28)) +#loc88 = loc("b_y"(#loc29)) +#loc89 = loc("b_rstd"(#loc30)) +#loc90 = loc("b_dy"(#loc31)) +#loc91 = loc("b_dy"(#loc32)) +#loc92 = loc("b_dx"(#loc33)) +#loc93 = loc("b_dx"(#loc34)) +#loc94 = loc("b_dx"(#loc35)) +#loc95 = loc("b_dx"(#loc36)) +#loc96 = loc("b_dx"(#loc37)) +#loc97 = loc("b_dx"(#loc38)) +#loc98 = loc("b_dx"(#loc39)) +#loc99 = loc("b_dx"(#loc40)) +#loc100 = loc("b_dx"(#loc41)) diff --git a/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.ttgir b/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..bfa2ef918972bb7aa416f60a993d58f4b42040eb --- /dev/null +++ b/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.ttgir @@ -0,0 +1,154 @@ +#blocked = #ttg.blocked<{sizePerThread = [2], threadsPerWarp = [32], warpsPerCTA = [2], order = [0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [2, 1], order = [1, 0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc39 = loc("y"(#loc)) +#loc40 = loc("rstd"(#loc)) +#loc41 = loc("dy"(#loc)) +#loc42 = loc("dx"(#loc)) +#loc43 = loc("eps"(#loc)) +#loc44 = loc("T"(#loc)) +#loc72 = loc("b_dx"(#loc30)) +#loc78 = loc(callsite(#loc1 at #loc72)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 2 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %c128_i64 = arith.constant 128 : i64 loc(#loc1) + %cst = arith.constant dense<0.000000e+00> : tensor<128xf32, #blocked> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<128x128xbf16, #blocked1> loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<128x1xi64, #blocked1> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<1x128xi32, #blocked1> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc45) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc46) + %o_t = arith.muli %i_t_3, %c128_i64 : i64 loc(#loc47) + %o_t_4 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc48) + %o_t_5 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #blocked> loc(#loc48) + %o_t_6 = arith.extsi %o_t_4 : tensor<128xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> to tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc49) + %o_t_7 = arith.extsi %o_t_5 : tensor<128xi32, #blocked> to tensor<128xi64, #blocked> loc(#loc49) + %o_t_8 = tt.splat %o_t : i64 -> tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc49) + %o_t_9 = tt.splat %o_t : i64 -> tensor<128xi64, #blocked> loc(#loc49) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc49) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<128xi64, #blocked> loc(#loc49) + %m_t = arith.extsi %T : i32 to i64 loc(#loc50) + %m_t_12 = tt.splat %m_t : i64 -> tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc50) + %m_t_13 = tt.splat %m_t : i64 -> tensor<128xi64, #blocked> loc(#loc50) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc50) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<128xi64, #blocked> loc(#loc50) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<128xi1, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<128x1xi1, #blocked1> loc(#loc51) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc52) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x128xi32, #blocked1> loc(#loc52) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_2 : tensor<1x128xi32, #blocked1> loc(#loc53) + %m_x_19 = tt.broadcast %m_x : tensor<128x1xi1, #blocked1> -> tensor<128x128xi1, #blocked1> loc(#loc54) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked1> -> tensor<128x128xi1, #blocked1> loc(#loc54) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<128x128xi1, #blocked1> loc(#loc54) + %p_y = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<128x1xi64, #blocked1> loc(#loc55) + %p_y_22 = arith.muli %p_y, %cst_1 : tensor<128x1xi64, #blocked1> loc(#loc56) + %p_y_23 = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr, #blocked1> loc(#loc57) + %p_y_24 = tt.addptr %p_y_23, %p_y_22 : tensor<128x1x!tt.ptr, #blocked1>, tensor<128x1xi64, #blocked1> loc(#loc57) + %p_y_25 = tt.broadcast %p_y_24 : tensor<128x1x!tt.ptr, #blocked1> -> tensor<128x128x!tt.ptr, #blocked1> loc(#loc58) + %p_y_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked1> -> tensor<128x128xi32, #blocked1> loc(#loc58) + %p_y_27 = tt.addptr %p_y_25, %p_y_26 : tensor<128x128x!tt.ptr, #blocked1>, tensor<128x128xi32, #blocked1> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr, #blocked> loc(#loc59) + %p_rstd_28 = tt.addptr %p_rstd, %o_t_11 : tensor<128x!tt.ptr, #blocked>, tensor<128xi64, #blocked> loc(#loc59) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<128x1x!tt.ptr, #blocked1> loc(#loc60) + %p_dy_29 = tt.addptr %p_dy, %p_y_22 : tensor<128x1x!tt.ptr, #blocked1>, tensor<128x1xi64, #blocked1> loc(#loc60) + %p_dy_30 = tt.broadcast %p_dy_29 : tensor<128x1x!tt.ptr, #blocked1> -> tensor<128x128x!tt.ptr, #blocked1> loc(#loc61) + %p_dy_31 = tt.addptr %p_dy_30, %p_y_26 : tensor<128x128x!tt.ptr, #blocked1>, tensor<128x128xi32, #blocked1> loc(#loc61) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<128x1x!tt.ptr, #blocked1> loc(#loc62) + %p_dx_32 = tt.addptr %p_dx, %p_y_22 : tensor<128x1x!tt.ptr, #blocked1>, tensor<128x1xi64, #blocked1> loc(#loc62) + %p_dx_33 = tt.broadcast %p_dx_32 : tensor<128x1x!tt.ptr, #blocked1> -> tensor<128x128x!tt.ptr, #blocked1> loc(#loc63) + %p_dx_34 = tt.addptr %p_dx_33, %p_y_26 : tensor<128x128x!tt.ptr, #blocked1>, tensor<128x128xi32, #blocked1> loc(#loc63) + %b_y = tt.load %p_y_27, %m_x_21, %cst_0 : tensor<128x128x!tt.ptr, #blocked1> loc(#loc64) + %b_y_35 = arith.extf %b_y : tensor<128x128xbf16, #blocked1> to tensor<128x128xf32, #blocked1> loc(#loc65) + %b_rstd = tt.load %p_rstd_28, %m_t_15, %cst : tensor<128x!tt.ptr, #blocked> loc(#loc66) + %b_dy = tt.load %p_dy_31, %m_x_21, %cst_0 : tensor<128x128x!tt.ptr, #blocked1> loc(#loc67) + %b_dy_36 = arith.extf %b_dy : tensor<128x128xbf16, #blocked1> to tensor<128x128xf32, #blocked1> loc(#loc68) + %b_dx = ttg.convert_layout %b_rstd : tensor<128xf32, #blocked> -> tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc69) + %b_dx_37 = tt.expand_dims %b_dx {axis = 1 : i32} : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<128x1xf32, #blocked1> loc(#loc69) + %b_dx_38 = tt.broadcast %b_dx_37 : tensor<128x1xf32, #blocked1> -> tensor<128x128xf32, #blocked1> loc(#loc70) + %b_dx_39 = arith.mulf %b_dy_36, %b_dx_38 : tensor<128x128xf32, #blocked1> loc(#loc70) + %b_dx_40 = arith.mulf %b_dy_36, %b_y_35 : tensor<128x128xf32, #blocked1> loc(#loc71) + %b_dx_41 = "tt.reduce"(%b_dx_40) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_47: f32 loc(callsite(#loc1 at #loc72)), %b_dx_48: f32 loc(callsite(#loc1 at #loc72))): + %b_dx_49 = arith.addf %b_dx_47, %b_dx_48 : f32 loc(#loc79) + tt.reduce.return %b_dx_49 : f32 loc(#loc77) + }) : (tensor<128x128xf32, #blocked1>) -> tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc77) + %b_dx_42 = tt.expand_dims %b_dx_41 {axis = 1 : i32} : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<128x1xf32, #blocked1> loc(#loc73) + %b_dx_43 = tt.broadcast %b_dx_42 : tensor<128x1xf32, #blocked1> -> tensor<128x128xf32, #blocked1> loc(#loc74) + %b_dx_44 = arith.mulf %b_dx_43, %b_y_35 : tensor<128x128xf32, #blocked1> loc(#loc74) + %b_dx_45 = arith.mulf %b_dx_44, %b_dx_38 : tensor<128x128xf32, #blocked1> loc(#loc75) + %b_dx_46 = arith.subf %b_dx_39, %b_dx_45 : tensor<128x128xf32, #blocked1> loc(#loc76) + %0 = arith.truncf %b_dx_46 : tensor<128x128xf32, #blocked1> to tensor<128x128xbf16, #blocked1> loc(#loc36) + tt.store %p_dx_34, %0, %m_x_21 : tensor<128x128x!tt.ptr, #blocked1> loc(#loc37) + tt.return loc(#loc38) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc45 = loc("i_t"(#loc2)) +#loc46 = loc("i_t"(#loc3)) +#loc47 = loc("o_t"(#loc4)) +#loc48 = loc("o_t"(#loc5)) +#loc49 = loc("o_t"(#loc6)) +#loc50 = loc("m_t"(#loc7)) +#loc51 = loc("m_x"(#loc8)) +#loc52 = loc("m_x"(#loc9)) +#loc53 = loc("m_x"(#loc10)) +#loc54 = loc("m_x"(#loc11)) +#loc55 = loc("p_y"(#loc12)) +#loc56 = loc("p_y"(#loc13)) +#loc57 = loc("p_y"(#loc14)) +#loc58 = loc("p_y"(#loc15)) +#loc59 = loc("p_rstd"(#loc16)) +#loc60 = loc("p_dy"(#loc17)) +#loc61 = loc("p_dy"(#loc18)) +#loc62 = loc("p_dx"(#loc19)) +#loc63 = loc("p_dx"(#loc20)) +#loc64 = loc("b_y"(#loc21)) +#loc65 = loc("b_y"(#loc22)) +#loc66 = loc("b_rstd"(#loc23)) +#loc67 = loc("b_dy"(#loc24)) +#loc68 = loc("b_dy"(#loc25)) +#loc69 = loc("b_dx"(#loc26)) +#loc70 = loc("b_dx"(#loc27)) +#loc71 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc32)) +#loc74 = loc("b_dx"(#loc33)) +#loc75 = loc("b_dx"(#loc34)) +#loc76 = loc("b_dx"(#loc35)) +#loc77 = loc(callsite(#loc29 at #loc72)) +#loc79 = loc(callsite(#loc31 at #loc77)) diff --git a/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.ttir b/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..f242766466ced3ec80bd72ef6238b6accc66dbe7 --- /dev/null +++ b/triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.ttir @@ -0,0 +1,144 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc39 = loc("y"(#loc)) +#loc40 = loc("rstd"(#loc)) +#loc41 = loc("dy"(#loc)) +#loc42 = loc("dx"(#loc)) +#loc43 = loc("eps"(#loc)) +#loc44 = loc("T"(#loc)) +#loc72 = loc("b_dx"(#loc30)) +#loc78 = loc(callsite(#loc1 at #loc72)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<0.000000e+00> : tensor<128x128xbf16> loc(#loc1) + %b_rstd = arith.constant dense<0.000000e+00> : tensor<128xf32> loc(#loc45) + %cst_0 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc1) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc46) + %c128_i64 = arith.constant 128 : i64 loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc47) + %i_t_1 = arith.extsi %i_t : i32 to i64 loc(#loc48) + %o_t = arith.muli %i_t_1, %c128_i64 : i64 loc(#loc49) + %o_t_2 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc50) + %o_t_3 = arith.extsi %o_t_2 : tensor<128xi32> to tensor<128xi64> loc(#loc51) + %o_t_4 = tt.splat %o_t : i64 -> tensor<128xi64> loc(#loc51) + %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<128xi64> loc(#loc51) + %m_t = arith.extsi %T : i32 to i64 loc(#loc52) + %m_t_6 = tt.splat %m_t : i64 -> tensor<128xi64> loc(#loc52) + %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<128xi64> loc(#loc52) + %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<128xi1> -> tensor<128x1xi1> loc(#loc53) + %m_x_9 = tt.expand_dims %o_t_2 {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc54) + %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc46) + %m_x_11 = tt.broadcast %m_x_8 : tensor<128x1xi1> -> tensor<128x128xi1> loc(#loc55) + %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<128x128xi1> loc(#loc55) + %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<128x128xi1> loc(#loc55) + %p_y = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc56) + %p_y_14 = arith.muli %p_y, %cst_0 : tensor<128x1xi64> loc(#loc57) + %p_y_15 = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc58) + %p_y_16 = tt.addptr %p_y_15, %p_y_14 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc58) + %p_y_17 = tt.broadcast %p_y_16 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc59) + %p_y_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc59) + %p_y_19 = tt.addptr %p_y_17, %p_y_18 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc59) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr> loc(#loc60) + %p_rstd_20 = tt.addptr %p_rstd, %o_t_5 : tensor<128x!tt.ptr>, tensor<128xi64> loc(#loc60) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc61) + %p_dy_21 = tt.addptr %p_dy, %p_y_14 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc61) + %p_dy_22 = tt.broadcast %p_dy_21 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc62) + %p_dy_23 = tt.addptr %p_dy_22, %p_y_18 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc62) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc63) + %p_dx_24 = tt.addptr %p_dx, %p_y_14 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc63) + %p_dx_25 = tt.broadcast %p_dx_24 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc64) + %p_dx_26 = tt.addptr %p_dx_25, %p_y_18 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc64) + %b_y = tt.load %p_y_19, %m_x_13, %cst : tensor<128x128x!tt.ptr> loc(#loc65) + %b_y_27 = arith.extf %b_y : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc66) + %b_rstd_28 = tt.load %p_rstd_20, %m_t_7, %b_rstd : tensor<128x!tt.ptr> loc(#loc45) + %b_dy = tt.load %p_dy_23, %m_x_13, %cst : tensor<128x128x!tt.ptr> loc(#loc67) + %b_dy_29 = arith.extf %b_dy : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc68) + %b_dx = tt.expand_dims %b_rstd_28 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc69) + %b_dx_30 = tt.broadcast %b_dx : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc70) + %b_dx_31 = arith.mulf %b_dy_29, %b_dx_30 : tensor<128x128xf32> loc(#loc70) + %b_dx_32 = arith.mulf %b_dy_29, %b_y_27 : tensor<128x128xf32> loc(#loc71) + %b_dx_33 = "tt.reduce"(%b_dx_32) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_39: f32 loc(callsite(#loc1 at #loc72)), %b_dx_40: f32 loc(callsite(#loc1 at #loc72))): + %b_dx_41 = arith.addf %b_dx_39, %b_dx_40 : f32 loc(#loc79) + tt.reduce.return %b_dx_41 : f32 loc(#loc77) + }) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc77) + %b_dx_34 = tt.expand_dims %b_dx_33 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc73) + %b_dx_35 = tt.broadcast %b_dx_34 : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc74) + %b_dx_36 = arith.mulf %b_dx_35, %b_y_27 : tensor<128x128xf32> loc(#loc74) + %b_dx_37 = arith.mulf %b_dx_36, %b_dx_30 : tensor<128x128xf32> loc(#loc75) + %b_dx_38 = arith.subf %b_dx_31, %b_dx_37 : tensor<128x128xf32> loc(#loc76) + %0 = arith.truncf %b_dx_38 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc36) + tt.store %p_dx_26, %0, %m_x_13 : tensor<128x128x!tt.ptr> loc(#loc37) + tt.return loc(#loc38) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc45 = loc("b_rstd"(#loc2)) +#loc46 = loc("m_x"(#loc3)) +#loc47 = loc("i_t"(#loc4)) +#loc48 = loc("i_t"(#loc5)) +#loc49 = loc("o_t"(#loc6)) +#loc50 = loc("o_t"(#loc7)) +#loc51 = loc("o_t"(#loc8)) +#loc52 = loc("m_t"(#loc9)) +#loc53 = loc("m_x"(#loc10)) +#loc54 = loc("m_x"(#loc11)) +#loc55 = loc("m_x"(#loc12)) +#loc56 = loc("p_y"(#loc13)) +#loc57 = loc("p_y"(#loc14)) +#loc58 = loc("p_y"(#loc15)) +#loc59 = loc("p_y"(#loc16)) +#loc60 = loc("p_rstd"(#loc17)) +#loc61 = loc("p_dy"(#loc18)) +#loc62 = loc("p_dy"(#loc19)) +#loc63 = loc("p_dx"(#loc20)) +#loc64 = loc("p_dx"(#loc21)) +#loc65 = loc("b_y"(#loc22)) +#loc66 = loc("b_y"(#loc23)) +#loc67 = loc("b_dy"(#loc24)) +#loc68 = loc("b_dy"(#loc25)) +#loc69 = loc("b_dx"(#loc26)) +#loc70 = loc("b_dx"(#loc27)) +#loc71 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc32)) +#loc74 = loc("b_dx"(#loc33)) +#loc75 = loc("b_dx"(#loc34)) +#loc76 = loc("b_dx"(#loc35)) +#loc77 = loc(callsite(#loc29 at #loc72)) +#loc79 = loc(callsite(#loc31 at #loc77)) diff --git a/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/__grp__l2norm_bwd_kernel.json b/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/__grp__l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..b57ffb1e01483306b3bad7fa565175e627bdbd71 --- /dev/null +++ b/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/__grp__l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_bwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/l2norm_bwd_kernel.source", "l2norm_bwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/l2norm_bwd_kernel.ttir", "l2norm_bwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/l2norm_bwd_kernel.ttgir", "l2norm_bwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/l2norm_bwd_kernel.llir", "l2norm_bwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/l2norm_bwd_kernel.ptx", "l2norm_bwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/l2norm_bwd_kernel.cubin", "l2norm_bwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/l2norm_bwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/l2norm_bwd_kernel.cubin b/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/l2norm_bwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..0282fa55e2db0256827cf0a6ad23ecafd8ef4573 Binary files /dev/null and b/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/l2norm_bwd_kernel.cubin differ diff --git a/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/l2norm_bwd_kernel.json b/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..474562ee662da5a2a2ffb82938e3c7707b4affec --- /dev/null +++ b/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"hash": "0940c12b9b5f8c45379a5ee4656ae0aef018142e2fe00a3f14fda41d4c510fd6", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 16, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 64, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_bwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/l2norm_bwd_kernel.llir b/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/l2norm_bwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..11ad8ac277d542803fc872f0c93b056da9781c7f --- /dev/null +++ b/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/l2norm_bwd_kernel.llir @@ -0,0 +1,175 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_bwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, ptr addrspace(1) %3, float %4, i32 %5, ptr addrspace(1) readnone captures(none) %6, ptr addrspace(1) readnone captures(none) %7) local_unnamed_addr #0 !dbg !4 { + %9 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !12 + %10 = zext nneg i32 %9 to i64, !dbg !13 + %11 = shl nuw nsw i64 %10, 3, !dbg !14 + %12 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !15 + %13 = and i32 %12, 511, !dbg !15 + %14 = and i32 %12, 31, !dbg !15 + %15 = lshr i32 %12, 5, !dbg !15 + %16 = lshr i32 %12, 6, !dbg !15 + %17 = and i32 %16, 7, !dbg !15 + %18 = zext nneg i32 %17 to i64, !dbg !16 + %19 = or disjoint i64 %11, %18, !dbg !16 + %20 = sext i32 %5 to i64, !dbg !17 + %21 = icmp slt i64 %19, %20, !dbg !17 + %22 = shl nuw nsw i32 %12, 1, !dbg !18 + %23 = and i32 %22, 126, !dbg !18 + %24 = shl nuw nsw i64 %19, 7, !dbg !19 + %25 = getelementptr bfloat, ptr addrspace(1) %0, i64 %24, !dbg !20 + %26 = zext nneg i32 %23 to i64, !dbg !21 + %27 = getelementptr bfloat, ptr addrspace(1) %25, i64 %26, !dbg !21 + %28 = getelementptr float, ptr addrspace(1) %1, i64 %19, !dbg !22 + %29 = getelementptr bfloat, ptr addrspace(1) %2, i64 %24, !dbg !23 + %30 = getelementptr bfloat, ptr addrspace(1) %29, i64 %26, !dbg !24 + %31 = getelementptr bfloat, ptr addrspace(1) %3, i64 %24, !dbg !25 + %32 = getelementptr bfloat, ptr addrspace(1) %31, i64 %26, !dbg !26 + %33 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %27, i1 %21) #4, !dbg !27 + %34 = bitcast i32 %33 to <2 x bfloat>, !dbg !27 + %35 = extractelement <2 x bfloat> %34, i64 0, !dbg !27 + %36 = extractelement <2 x bfloat> %34, i64 1, !dbg !27 + %37 = fpext bfloat %35 to float, !dbg !28 + %38 = fpext bfloat %36 to float, !dbg !28 + %39 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %28, i1 %21) #4, !dbg !29 + %40 = bitcast i32 %39 to float, !dbg !29 + %41 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %30, i1 %21) #4, !dbg !30 + %42 = bitcast i32 %41 to <2 x bfloat>, !dbg !30 + %43 = extractelement <2 x bfloat> %42, i64 0, !dbg !30 + %44 = extractelement <2 x bfloat> %42, i64 1, !dbg !30 + %45 = fpext bfloat %43 to float, !dbg !31 + %46 = fpext bfloat %44 to float, !dbg !31 + %47 = fmul float %40, %45, !dbg !32 + %48 = fmul float %40, %46, !dbg !32 + %49 = fmul float %37, %45, !dbg !33 + %50 = fmul float %38, %46, !dbg !33 + %51 = fadd float %49, %50, !dbg !34 + %52 = bitcast float %51 to i32, !dbg !37 + %53 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %52, i32 16, i32 31), !dbg !37 + %54 = bitcast i32 %53 to float, !dbg !37 + %55 = fadd float %51, %54, !dbg !34 + %56 = bitcast float %55 to i32, !dbg !37 + %57 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %56, i32 8, i32 31), !dbg !37 + %58 = bitcast i32 %57 to float, !dbg !37 + %59 = fadd float %55, %58, !dbg !34 + %60 = bitcast float %59 to i32, !dbg !37 + %61 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %60, i32 4, i32 31), !dbg !37 + %62 = bitcast i32 %61 to float, !dbg !37 + %63 = fadd float %59, %62, !dbg !34 + %64 = bitcast float %63 to i32, !dbg !37 + %65 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %64, i32 2, i32 31), !dbg !37 + %66 = bitcast i32 %65 to float, !dbg !37 + %67 = fadd float %63, %66, !dbg !34 + %68 = bitcast float %67 to i32, !dbg !37 + %69 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %68, i32 1, i32 31), !dbg !37 + %70 = bitcast i32 %69 to float, !dbg !37 + %71 = fadd float %67, %70, !dbg !34 + %72 = and i32 %15, 1, !dbg !37 + %73 = icmp eq i32 %14, 0, !dbg !37 + %.idx = shl nuw nsw i32 %17, 3, !dbg !37 + %74 = getelementptr i8, ptr addrspace(3) @global_smem, i32 %.idx, !dbg !37 + %75 = getelementptr float, ptr addrspace(3) %74, i32 %72, !dbg !37 + %76 = bitcast float %71 to <1 x i32>, !dbg !37 + tail call void asm sideeffect "@$2 st.shared.b32 [ $0 + 0 ], $1;", "r,r,b"(ptr addrspace(3) %75, <1 x i32> %76, i1 %73) #4, !dbg !37 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !37 + %77 = icmp samesign ult i32 %13, 16, !dbg !37 + %78 = getelementptr float, ptr addrspace(3) @global_smem, i32 %13, !dbg !37 + %79 = tail call i32 asm sideeffect "@$2 ld.shared.b32 $0, [ $1 + 0 ];", "=r,r,b"(ptr addrspace(3) %78, i1 %77) #4, !dbg !37 + %80 = bitcast i32 %79 to float, !dbg !37 + %81 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %79, i32 1, i32 31), !dbg !37 + %82 = bitcast i32 %81 to float, !dbg !37 + %83 = fadd float %80, %82, !dbg !34 + %84 = and i32 %12, 1, !dbg !37 + %85 = icmp eq i32 %84, 0, !dbg !37 + %86 = and i1 %77, %85, !dbg !37 + %87 = bitcast float %83 to <1 x i32>, !dbg !37 + tail call void asm sideeffect "@$2 st.shared.b32 [ $0 + 0 ], $1;", "r,r,b"(ptr addrspace(3) %78, <1 x i32> %87, i1 %86) #4, !dbg !37 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !37 + %88 = load float, ptr addrspace(3) %74, align 8, !dbg !37 + %89 = fmul float %88, %37, !dbg !40 + %90 = fmul float %88, %38, !dbg !40 + %91 = fmul float %89, %40, !dbg !41 + %92 = fmul float %90, %40, !dbg !41 + %93 = fsub float %47, %91, !dbg !42 + %94 = fsub float %48, %92, !dbg !42 + %95 = fptrunc float %93 to bfloat, !dbg !43 + %96 = fptrunc float %94 to bfloat, !dbg !43 + %97 = insertelement <2 x bfloat> poison, bfloat %95, i64 0, !dbg !44 + %98 = insertelement <2 x bfloat> %97, bfloat %96, i64 1, !dbg !44 + %99 = bitcast <2 x bfloat> %98 to i32, !dbg !44 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %99, ptr addrspace(1) %32, i1 %21) #4, !dbg !44 + ret void, !dbg !45 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="512" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { convergent nocallback nounwind } +attributes #4 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = distinct !DISubprogram(name: "l2norm_bwd_kernel", linkageName: "l2norm_bwd_kernel", scope: !1, file: !1, line: 93, type: !5, scopeLine: 93, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!5 = !DISubroutineType(cc: DW_CC_normal, types: !6) +!6 = !{null, !7, !9, !7, !7, !10, !11, !7, !7} +!7 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !8, size: 64, dwarfAddressSpace: 1) +!8 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!9 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !10, size: 64, dwarfAddressSpace: 1) +!10 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!11 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!12 = !DILocation(line: 94, column: 24, scope: !4) +!13 = !DILocation(line: 94, column: 30, scope: !4) +!14 = !DILocation(line: 95, column: 16, scope: !4) +!15 = !DILocation(line: 95, column: 34, scope: !4) +!16 = !DILocation(line: 95, column: 21, scope: !4) +!17 = !DILocation(line: 97, column: 16, scope: !4) +!18 = !DILocation(line: 98, column: 30, scope: !4) +!19 = !DILocation(line: 99, column: 29, scope: !4) +!20 = !DILocation(line: 99, column: 14, scope: !4) +!21 = !DILocation(line: 99, column: 33, scope: !4) +!22 = !DILocation(line: 100, column: 20, scope: !4) +!23 = !DILocation(line: 101, column: 16, scope: !4) +!24 = !DILocation(line: 101, column: 35, scope: !4) +!25 = !DILocation(line: 102, column: 16, scope: !4) +!26 = !DILocation(line: 102, column: 35, scope: !4) +!27 = !DILocation(line: 104, column: 18, scope: !4) +!28 = !DILocation(line: 104, column: 47, scope: !4) +!29 = !DILocation(line: 105, column: 21, scope: !4) +!30 = !DILocation(line: 106, column: 19, scope: !4) +!31 = !DILocation(line: 106, column: 49, scope: !4) +!32 = !DILocation(line: 107, column: 18, scope: !4) +!33 = !DILocation(line: 107, column: 50, scope: !4) +!34 = !DILocation(line: 263, column: 15, scope: !35, inlinedAt: !37) +!35 = distinct !DILexicalBlockFile(scope: !4, file: !36, discriminator: 0) +!36 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!37 = !DILocation(line: 293, column: 36, scope: !35, inlinedAt: !38) +!38 = !DILocation(line: 107, column: 55, scope: !39) +!39 = distinct !DILexicalBlockFile(scope: !4, file: !1, discriminator: 0) +!40 = !DILocation(line: 107, column: 69, scope: !4) +!41 = !DILocation(line: 107, column: 75, scope: !4) +!42 = !DILocation(line: 107, column: 36, scope: !4) +!43 = !DILocation(line: 108, column: 27, scope: !4) +!44 = !DILocation(line: 108, column: 19, scope: !4) +!45 = !DILocation(line: 108, column: 4, scope: !4) diff --git a/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/l2norm_bwd_kernel.ptx b/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/l2norm_bwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..047213c6e953d43ba105d10d526ab28d584fd0e7 --- /dev/null +++ b/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/l2norm_bwd_kernel.ptx @@ -0,0 +1,407 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_bwd_kernel // -- Begin function l2norm_bwd_kernel +.extern .shared .align 16 .b8 global_smem[]; + // @l2norm_bwd_kernel +.visible .entry l2norm_bwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_2, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_3, + .param .f32 l2norm_bwd_kernel_param_4, + .param .u32 l2norm_bwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_6, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_7 +) +.reqntid 512 +{ + .reg .pred %p<7>; + .reg .b16 %rs<5>; + .reg .b32 %r<50>; + .reg .b64 %rd<19>; + .loc 1 93 0 // l2norm.py:93:0 +$L__func_begin0: + .loc 1 93 0 // l2norm.py:93:0 + +// %bb.0: + ld.param.b64 %rd5, [l2norm_bwd_kernel_param_0]; + ld.param.b64 %rd6, [l2norm_bwd_kernel_param_1]; +$L__tmp0: + .loc 1 94 24 // l2norm.py:94:24 + mov.u32 %r11, %ctaid.x; + ld.param.b64 %rd7, [l2norm_bwd_kernel_param_2]; + .loc 1 95 16 // l2norm.py:95:16 + mul.wide.u32 %rd8, %r11, 8; + ld.param.b64 %rd9, [l2norm_bwd_kernel_param_3]; + .loc 1 95 34 // l2norm.py:95:34 + mov.u32 %r12, %tid.x; + and.b32 %r13, %r12, 511; + ld.param.s32 %rd10, [l2norm_bwd_kernel_param_5]; + and.b32 %r14, %r12, 31; + bfe.u32 %r15, %r12, 6, 3; + .loc 1 95 21 // l2norm.py:95:21 + cvt.u64.u32 %rd11, %r15; + or.b64 %rd12, %rd8, %rd11; + .loc 1 97 16 // l2norm.py:97:16 + setp.lt.s64 %p1, %rd12, %rd10; + .loc 1 98 30 // l2norm.py:98:30 + shl.b32 %r16, %r12, 1; + and.b32 %r17, %r16, 126; + .loc 1 99 14 // l2norm.py:99:14 + shl.b64 %rd13, %rd12, 8; + add.s64 %rd14, %rd5, %rd13; + .loc 1 99 33 // l2norm.py:99:33 + mul.wide.u32 %rd15, %r17, 2; + add.s64 %rd1, %rd14, %rd15; + .loc 1 100 20 // l2norm.py:100:20 + shl.b64 %rd16, %rd12, 2; + add.s64 %rd2, %rd6, %rd16; + .loc 1 101 16 // l2norm.py:101:16 + add.s64 %rd17, %rd7, %rd13; + .loc 1 101 35 // l2norm.py:101:35 + add.s64 %rd3, %rd17, %rd15; + .loc 1 102 16 // l2norm.py:102:16 + add.s64 %rd18, %rd9, %rd13; + .loc 1 102 35 // l2norm.py:102:35 + add.s64 %rd4, %rd18, %rd15; + mov.b32 %r2, 0; + .loc 1 104 18 // l2norm.py:104:18 + // begin inline asm + mov.u32 %r1, %r2; + @%p1 ld.global.b32 { %r1 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + .loc 1 104 47 // l2norm.py:104:47 + cvt.f32.bf16 %r18, %rs1; + cvt.f32.bf16 %r19, %rs2; + .loc 1 105 21 // l2norm.py:105:21 + // begin inline asm + mov.u32 %r3, %r2; + @%p1 ld.global.b32 { %r3 }, [ %rd2 + 0 ]; + // end inline asm + .loc 1 106 19 // l2norm.py:106:19 + // begin inline asm + mov.u32 %r4, %r2; + @%p1 ld.global.b32 { %r4 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs3, %rs4}, %r4; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r20, %rs3; + cvt.f32.bf16 %r21, %rs4; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r22, %r19, %r21; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r23, %rs1, %rs3, %r22; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r24, %r23, 16, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r25, %r23, %r24; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r26, %r25, 8, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r27, %r25, %r26; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r28, %r27, 4, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r29, %r27, %r28; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r30, %r29, 2, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r31, %r29, %r30; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r32, %r31, 1, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r6, %r31, %r32; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + setp.eq.b32 %p2, %r14, 0; + shl.b32 %r33, %r15, 3; + mov.b32 %r34, global_smem; + add.s32 %r35, %r34, %r33; + shr.u32 %r36, %r12, 3; + and.b32 %r37, %r36, 4; + add.s32 %r5, %r35, %r37; + // begin inline asm + @%p2 st.shared.b32 [ %r5 + 0 ], %r6; + // end inline asm + bar.sync 0; + setp.lt.u32 %p3, %r13, 16; + shl.b32 %r38, %r13, 2; + add.s32 %r8, %r34, %r38; + // begin inline asm + @%p3 ld.shared.b32 %r7, [ %r8 + 0 ]; + // end inline asm + shfl.sync.bfly.b32 %r39, %r7, 1, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r9, %r7, %r39; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + and.b32 %r40, %r12, 1; + setp.ne.b32 %p5, %r40, 0; + not.pred %p6, %p5; + and.pred %p4, %p3, %p6; + // begin inline asm + @%p4 st.shared.b32 [ %r8 + 0 ], %r9; + // end inline asm + bar.sync 0; + ld.shared.b32 %r41, [%r35]; +$L__tmp15: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r42, %r41, %r18; + mul.f32 %r43, %r41, %r19; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r44, %r42, %r3; + mul.f32 %r45, %r43, %r3; + .loc 1 107 36 // l2norm.py:107:36 + neg.f32 %r46, %r44; + fma.rn.f32 %r47, %r3, %r20, %r46; + neg.f32 %r48, %r45; + fma.rn.f32 %r49, %r3, %r21, %r48; + .loc 1 108 19 // l2norm.py:108:19 + cvt.rn.bf16x2.f32 %r10, %r49, %r47; + // begin inline asm + @%p1 st.global.b32 [ %rd4 + 0 ], { %r10 }; + // end inline asm + .loc 1 108 4 // l2norm.py:108:4 + ret; +$L__tmp16: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 98 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp15 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 107 // DW_AT_call_line +.b8 55 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp14 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/l2norm_bwd_kernel.source b/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/l2norm_bwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..c23a4689ce5f60989f46cd34fecfe31d002a62d8 --- /dev/null +++ b/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/l2norm_bwd_kernel.source @@ -0,0 +1,210 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc47 = loc(unknown) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc54 = loc("y"(#loc)) +#loc55 = loc("rstd"(#loc)) +#loc56 = loc("dy"(#loc)) +#loc57 = loc("dx"(#loc)) +#loc58 = loc("eps"(#loc)) +#loc59 = loc("T"(#loc)) +#loc101 = loc("input"(#loc45)) +#loc102 = loc("a"(#loc50)) +#loc103 = loc("b"(#loc50)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc60) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc61) + %o_t = arith.constant 8 : i32 loc(#loc62) + %o_t_1 = arith.constant 8 : i64 loc(#loc62) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc62) + %o_t_3 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32> loc(#loc63) + %o_t_4 = arith.extsi %o_t_3 : tensor<8xi32> to tensor<8xi64> loc(#loc64) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<8xi64> loc(#loc64) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<8xi64> loc(#loc64) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc65) + %m_t = arith.extsi %T : i32 to i64 loc(#loc66) + %m_t_7 = tt.splat %m_t : i64 -> tensor<8xi64> loc(#loc66) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<8xi64> loc(#loc66) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<8xi1> -> tensor<8x1xi1> loc(#loc67) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc68) + %m_x_10 = arith.constant 128 : i32 loc(#loc69) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc69) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc69) + %m_x_13 = tt.broadcast %m_x : tensor<8x1xi1> -> tensor<8x128xi1> loc(#loc70) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<8x128xi1> loc(#loc70) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<8x128xi1> loc(#loc70) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc71) + %p_y_16 = arith.constant 128 : i32 loc(#loc72) + %p_y_17 = arith.constant 128 : i64 loc(#loc72) + %p_y_18 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc72) + %p_y_19 = arith.muli %p_y, %p_y_18 : tensor<8x1xi64> loc(#loc72) + %p_y_20 = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc73) + %p_y_21 = tt.addptr %p_y_20, %p_y_19 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc73) + %p_y_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc74) + %p_y_23 = tt.broadcast %p_y_21 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc75) + %p_y_24 = tt.broadcast %p_y_22 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc75) + %p_y_25 = tt.addptr %p_y_23, %p_y_24 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc75) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr> loc(#loc76) + %p_rstd_26 = tt.addptr %p_rstd, %o_t_6 : tensor<8x!tt.ptr>, tensor<8xi64> loc(#loc76) + %p_dy = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc77) + %p_dy_27 = arith.constant 128 : i32 loc(#loc78) + %p_dy_28 = arith.constant 128 : i64 loc(#loc78) + %p_dy_29 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc78) + %p_dy_30 = arith.muli %p_dy, %p_dy_29 : tensor<8x1xi64> loc(#loc78) + %p_dy_31 = tt.splat %dy : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc79) + %p_dy_32 = tt.addptr %p_dy_31, %p_dy_30 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc79) + %p_dy_33 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc80) + %p_dy_34 = tt.broadcast %p_dy_32 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc81) + %p_dy_35 = tt.broadcast %p_dy_33 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc81) + %p_dy_36 = tt.addptr %p_dy_34, %p_dy_35 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc81) + %p_dx = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc82) + %p_dx_37 = arith.constant 128 : i32 loc(#loc83) + %p_dx_38 = arith.constant 128 : i64 loc(#loc83) + %p_dx_39 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc83) + %p_dx_40 = arith.muli %p_dx, %p_dx_39 : tensor<8x1xi64> loc(#loc83) + %p_dx_41 = tt.splat %dx : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc84) + %p_dx_42 = tt.addptr %p_dx_41, %p_dx_40 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc84) + %p_dx_43 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc85) + %p_dx_44 = tt.broadcast %p_dx_42 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc86) + %p_dx_45 = tt.broadcast %p_dx_43 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc86) + %p_dx_46 = tt.addptr %p_dx_44, %p_dx_45 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc86) + %b_y = arith.constant 0.000000e+00 : f32 loc(#loc87) + %b_y_47 = arith.constant dense<0.000000e+00> : tensor<8x128xf32> loc(#loc87) + %b_y_48 = arith.truncf %b_y_47 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc87) + %b_y_49 = tt.load %p_y_25, %m_x_15, %b_y_48 : tensor<8x128x!tt.ptr> loc(#loc87) + %b_y_50 = arith.extf %b_y_49 : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc88) + %b_rstd = arith.constant 0.000000e+00 : f32 loc(#loc89) + %b_rstd_51 = arith.constant dense<0.000000e+00> : tensor<8xf32> loc(#loc89) + %b_rstd_52 = tt.load %p_rstd_26, %m_t_8, %b_rstd_51 : tensor<8x!tt.ptr> loc(#loc89) + %b_dy = arith.constant 0.000000e+00 : f32 loc(#loc90) + %b_dy_53 = arith.constant dense<0.000000e+00> : tensor<8x128xf32> loc(#loc90) + %b_dy_54 = arith.truncf %b_dy_53 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc90) + %b_dy_55 = tt.load %p_dy_36, %m_x_15, %b_dy_54 : tensor<8x128x!tt.ptr> loc(#loc90) + %b_dy_56 = arith.extf %b_dy_55 : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc91) + %b_dx = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc92) + %b_dx_57 = tt.broadcast %b_dx : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc93) + %b_dx_58 = arith.mulf %b_dy_56, %b_dx_57 : tensor<8x128xf32> loc(#loc93) + %b_dx_59 = arith.mulf %b_dy_56, %b_y_50 : tensor<8x128xf32> loc(#loc94) + %b_dx_60 = tt.call @triton.language.standard.sum__fp32S8_128S_c1_cFalse_cNone(%b_dx_59) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc95) + %b_dx_61 = tt.expand_dims %b_dx_60 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc96) + %b_dx_62 = tt.broadcast %b_dx_61 : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc97) + %b_dx_63 = arith.mulf %b_dx_62, %b_y_50 : tensor<8x128xf32> loc(#loc97) + %b_dx_64 = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc98) + %b_dx_65 = tt.broadcast %b_dx_64 : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc99) + %b_dx_66 = arith.mulf %b_dx_63, %b_dx_65 : tensor<8x128xf32> loc(#loc99) + %b_dx_67 = arith.subf %b_dx_58, %b_dx_66 : tensor<8x128xf32> loc(#loc100) + %0 = arith.truncf %b_dx_67 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc42) + tt.store %p_dx_46, %0, %m_x_15 : tensor<8x128x!tt.ptr> loc(#loc43) + tt.return loc(#loc44) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S8_128S_c1_cFalse_cNone(%input: tensor<8x128xf32> loc("input"(#loc45))) -> tensor<8xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc46) + tt.reduce.return %2 : f32 loc(#loc46) + }) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc46) + tt.return %0 : tensor<8xf32> loc(#loc48) + ^bb1: // no predecessors + %1 = ub.poison : tensor<8xf32> loc(#loc49) + tt.return %1 : tensor<8xf32> loc(#loc49) + } loc(#loc45) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc50)), %b: f32 loc("b"(#loc50))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc51) + tt.return %0 : f32 loc(#loc52) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc53) + tt.return %1 : f32 loc(#loc53) + } loc(#loc50) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:31) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:39) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:20) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:31) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:39) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:82) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc60 = loc("i_t"(#loc1)) +#loc61 = loc("i_t"(#loc2)) +#loc62 = loc("o_t"(#loc3)) +#loc63 = loc("o_t"(#loc4)) +#loc64 = loc("o_t"(#loc5)) +#loc65 = loc("o_d"(#loc6)) +#loc66 = loc("m_t"(#loc7)) +#loc67 = loc("m_x"(#loc8)) +#loc68 = loc("m_x"(#loc9)) +#loc69 = loc("m_x"(#loc10)) +#loc70 = loc("m_x"(#loc11)) +#loc71 = loc("p_y"(#loc12)) +#loc72 = loc("p_y"(#loc13)) +#loc73 = loc("p_y"(#loc14)) +#loc74 = loc("p_y"(#loc15)) +#loc75 = loc("p_y"(#loc16)) +#loc76 = loc("p_rstd"(#loc17)) +#loc77 = loc("p_dy"(#loc18)) +#loc78 = loc("p_dy"(#loc19)) +#loc79 = loc("p_dy"(#loc20)) +#loc80 = loc("p_dy"(#loc21)) +#loc81 = loc("p_dy"(#loc22)) +#loc82 = loc("p_dx"(#loc23)) +#loc83 = loc("p_dx"(#loc24)) +#loc84 = loc("p_dx"(#loc25)) +#loc85 = loc("p_dx"(#loc26)) +#loc86 = loc("p_dx"(#loc27)) +#loc87 = loc("b_y"(#loc28)) +#loc88 = loc("b_y"(#loc29)) +#loc89 = loc("b_rstd"(#loc30)) +#loc90 = loc("b_dy"(#loc31)) +#loc91 = loc("b_dy"(#loc32)) +#loc92 = loc("b_dx"(#loc33)) +#loc93 = loc("b_dx"(#loc34)) +#loc94 = loc("b_dx"(#loc35)) +#loc95 = loc("b_dx"(#loc36)) +#loc96 = loc("b_dx"(#loc37)) +#loc97 = loc("b_dx"(#loc38)) +#loc98 = loc("b_dx"(#loc39)) +#loc99 = loc("b_dx"(#loc40)) +#loc100 = loc("b_dx"(#loc41)) diff --git a/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/l2norm_bwd_kernel.ttgir b/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/l2norm_bwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..fbd57ca45a7a3b581a068d951ebad77404c70446 --- /dev/null +++ b/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/l2norm_bwd_kernel.ttgir @@ -0,0 +1,146 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 2], threadsPerWarp = [1, 32], warpsPerCTA = [8, 2], order = [1, 0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc39 = loc("y"(#loc)) +#loc40 = loc("rstd"(#loc)) +#loc41 = loc("dy"(#loc)) +#loc42 = loc("dx"(#loc)) +#loc43 = loc("eps"(#loc)) +#loc44 = loc("T"(#loc)) +#loc72 = loc("b_dx"(#loc30)) +#loc78 = loc(callsite(#loc1 at #loc72)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 16 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %c8_i64 = arith.constant 8 : i64 loc(#loc1) + %cst = arith.constant dense<0.000000e+00> : tensor<8x128xbf16, #blocked> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<8x1xi64, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc45) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc46) + %o_t = arith.muli %i_t_3, %c8_i64 : i64 loc(#loc47) + %o_t_4 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc48) + %o_t_5 = arith.extsi %o_t_4 : tensor<8xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %o_t_6 = tt.splat %o_t : i64 -> tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %o_t_7 = arith.addi %o_t_6, %o_t_5 : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %m_t = arith.extsi %T : i32 to i64 loc(#loc50) + %m_t_8 = tt.splat %m_t : i64 -> tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc50) + %m_t_9 = arith.cmpi slt, %o_t_7, %m_t_8 : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc50) + %m_x = tt.expand_dims %m_t_9 {axis = 1 : i32} : tensor<8xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xi1, #blocked> loc(#loc51) + %m_x_10 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc52) + %m_x_11 = tt.expand_dims %m_x_10 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc52) + %m_x_12 = arith.cmpi slt, %m_x_11, %cst_2 : tensor<1x128xi32, #blocked> loc(#loc53) + %m_x_13 = tt.broadcast %m_x : tensor<8x1xi1, #blocked> -> tensor<8x128xi1, #blocked> loc(#loc54) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1, #blocked> -> tensor<8x128xi1, #blocked> loc(#loc54) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<8x128xi1, #blocked> loc(#loc54) + %p_y = tt.expand_dims %o_t_7 {axis = 1 : i32} : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xi64, #blocked> loc(#loc55) + %p_y_16 = arith.muli %p_y, %cst_1 : tensor<8x1xi64, #blocked> loc(#loc56) + %p_y_17 = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr, #blocked> loc(#loc57) + %p_y_18 = tt.addptr %p_y_17, %p_y_16 : tensor<8x1x!tt.ptr, #blocked>, tensor<8x1xi64, #blocked> loc(#loc57) + %p_y_19 = tt.broadcast %p_y_18 : tensor<8x1x!tt.ptr, #blocked> -> tensor<8x128x!tt.ptr, #blocked> loc(#loc58) + %p_y_20 = tt.broadcast %m_x_11 : tensor<1x128xi32, #blocked> -> tensor<8x128xi32, #blocked> loc(#loc58) + %p_y_21 = tt.addptr %p_y_19, %p_y_20 : tensor<8x128x!tt.ptr, #blocked>, tensor<8x128xi32, #blocked> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc59) + %p_rstd_22 = tt.addptr %p_rstd, %o_t_7 : tensor<8x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>>, tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc59) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<8x1x!tt.ptr, #blocked> loc(#loc60) + %p_dy_23 = tt.addptr %p_dy, %p_y_16 : tensor<8x1x!tt.ptr, #blocked>, tensor<8x1xi64, #blocked> loc(#loc60) + %p_dy_24 = tt.broadcast %p_dy_23 : tensor<8x1x!tt.ptr, #blocked> -> tensor<8x128x!tt.ptr, #blocked> loc(#loc61) + %p_dy_25 = tt.addptr %p_dy_24, %p_y_20 : tensor<8x128x!tt.ptr, #blocked>, tensor<8x128xi32, #blocked> loc(#loc61) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<8x1x!tt.ptr, #blocked> loc(#loc62) + %p_dx_26 = tt.addptr %p_dx, %p_y_16 : tensor<8x1x!tt.ptr, #blocked>, tensor<8x1xi64, #blocked> loc(#loc62) + %p_dx_27 = tt.broadcast %p_dx_26 : tensor<8x1x!tt.ptr, #blocked> -> tensor<8x128x!tt.ptr, #blocked> loc(#loc63) + %p_dx_28 = tt.addptr %p_dx_27, %p_y_20 : tensor<8x128x!tt.ptr, #blocked>, tensor<8x128xi32, #blocked> loc(#loc63) + %b_y = tt.load %p_y_21, %m_x_15, %cst : tensor<8x128x!tt.ptr, #blocked> loc(#loc64) + %b_y_29 = arith.extf %b_y : tensor<8x128xbf16, #blocked> to tensor<8x128xf32, #blocked> loc(#loc65) + %b_rstd = tt.load %p_rstd_22, %m_t_9, %cst_0 : tensor<8x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc66) + %b_dy = tt.load %p_dy_25, %m_x_15, %cst : tensor<8x128x!tt.ptr, #blocked> loc(#loc67) + %b_dy_30 = arith.extf %b_dy : tensor<8x128xbf16, #blocked> to tensor<8x128xf32, #blocked> loc(#loc68) + %b_dx = tt.expand_dims %b_rstd {axis = 1 : i32} : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xf32, #blocked> loc(#loc69) + %b_dx_31 = tt.broadcast %b_dx : tensor<8x1xf32, #blocked> -> tensor<8x128xf32, #blocked> loc(#loc70) + %b_dx_32 = arith.mulf %b_dy_30, %b_dx_31 : tensor<8x128xf32, #blocked> loc(#loc70) + %b_dx_33 = arith.mulf %b_dy_30, %b_y_29 : tensor<8x128xf32, #blocked> loc(#loc71) + %b_dx_34 = "tt.reduce"(%b_dx_33) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_40: f32 loc(callsite(#loc1 at #loc72)), %b_dx_41: f32 loc(callsite(#loc1 at #loc72))): + %b_dx_42 = arith.addf %b_dx_40, %b_dx_41 : f32 loc(#loc79) + tt.reduce.return %b_dx_42 : f32 loc(#loc77) + }) : (tensor<8x128xf32, #blocked>) -> tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc77) + %b_dx_35 = tt.expand_dims %b_dx_34 {axis = 1 : i32} : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xf32, #blocked> loc(#loc73) + %b_dx_36 = tt.broadcast %b_dx_35 : tensor<8x1xf32, #blocked> -> tensor<8x128xf32, #blocked> loc(#loc74) + %b_dx_37 = arith.mulf %b_dx_36, %b_y_29 : tensor<8x128xf32, #blocked> loc(#loc74) + %b_dx_38 = arith.mulf %b_dx_37, %b_dx_31 : tensor<8x128xf32, #blocked> loc(#loc75) + %b_dx_39 = arith.subf %b_dx_32, %b_dx_38 : tensor<8x128xf32, #blocked> loc(#loc76) + %0 = arith.truncf %b_dx_39 : tensor<8x128xf32, #blocked> to tensor<8x128xbf16, #blocked> loc(#loc36) + tt.store %p_dx_28, %0, %m_x_15 : tensor<8x128x!tt.ptr, #blocked> loc(#loc37) + tt.return loc(#loc38) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc45 = loc("i_t"(#loc2)) +#loc46 = loc("i_t"(#loc3)) +#loc47 = loc("o_t"(#loc4)) +#loc48 = loc("o_t"(#loc5)) +#loc49 = loc("o_t"(#loc6)) +#loc50 = loc("m_t"(#loc7)) +#loc51 = loc("m_x"(#loc8)) +#loc52 = loc("m_x"(#loc9)) +#loc53 = loc("m_x"(#loc10)) +#loc54 = loc("m_x"(#loc11)) +#loc55 = loc("p_y"(#loc12)) +#loc56 = loc("p_y"(#loc13)) +#loc57 = loc("p_y"(#loc14)) +#loc58 = loc("p_y"(#loc15)) +#loc59 = loc("p_rstd"(#loc16)) +#loc60 = loc("p_dy"(#loc17)) +#loc61 = loc("p_dy"(#loc18)) +#loc62 = loc("p_dx"(#loc19)) +#loc63 = loc("p_dx"(#loc20)) +#loc64 = loc("b_y"(#loc21)) +#loc65 = loc("b_y"(#loc22)) +#loc66 = loc("b_rstd"(#loc23)) +#loc67 = loc("b_dy"(#loc24)) +#loc68 = loc("b_dy"(#loc25)) +#loc69 = loc("b_dx"(#loc26)) +#loc70 = loc("b_dx"(#loc27)) +#loc71 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc32)) +#loc74 = loc("b_dx"(#loc33)) +#loc75 = loc("b_dx"(#loc34)) +#loc76 = loc("b_dx"(#loc35)) +#loc77 = loc(callsite(#loc29 at #loc72)) +#loc79 = loc(callsite(#loc31 at #loc77)) diff --git a/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/l2norm_bwd_kernel.ttir b/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/l2norm_bwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..c4761315a53ecb33821860b30f93bdae9fbfe731 --- /dev/null +++ b/triton/gpu0/BFAMCK43L6GEKN42L3SGK2XAV3YBQFBOF7QAUPYU7WSB2TCRB7LA/l2norm_bwd_kernel.ttir @@ -0,0 +1,147 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc40 = loc("y"(#loc)) +#loc41 = loc("rstd"(#loc)) +#loc42 = loc("dy"(#loc)) +#loc43 = loc("dx"(#loc)) +#loc44 = loc("eps"(#loc)) +#loc45 = loc("T"(#loc)) +#loc74 = loc("b_dx"(#loc31)) +#loc80 = loc(callsite(#loc1 at #loc74)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<0.000000e+00> : tensor<8x128xbf16> loc(#loc1) + %b_rstd = arith.constant dense<0.000000e+00> : tensor<8xf32> loc(#loc46) + %cst_0 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc1) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc47) + %o_t = arith.constant 8 : i64 loc(#loc48) + %i_t = tt.get_program_id x : i32 loc(#loc49) + %i_t_1 = arith.extsi %i_t : i32 to i64 loc(#loc50) + %o_t_2 = arith.muli %i_t_1, %o_t : i64 loc(#loc48) + %o_t_3 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32> loc(#loc51) + %o_t_4 = arith.extsi %o_t_3 : tensor<8xi32> to tensor<8xi64> loc(#loc52) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<8xi64> loc(#loc52) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<8xi64> loc(#loc52) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc53) + %m_t = arith.extsi %T : i32 to i64 loc(#loc54) + %m_t_7 = tt.splat %m_t : i64 -> tensor<8xi64> loc(#loc54) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<8xi64> loc(#loc54) + %m_x_9 = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<8xi1> -> tensor<8x1xi1> loc(#loc55) + %m_x_10 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc56) + %m_x_11 = arith.cmpi slt, %m_x_10, %m_x : tensor<1x128xi32> loc(#loc47) + %m_x_12 = tt.broadcast %m_x_9 : tensor<8x1xi1> -> tensor<8x128xi1> loc(#loc57) + %m_x_13 = tt.broadcast %m_x_11 : tensor<1x128xi1> -> tensor<8x128xi1> loc(#loc57) + %m_x_14 = arith.andi %m_x_12, %m_x_13 : tensor<8x128xi1> loc(#loc57) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc58) + %p_y_15 = arith.muli %p_y, %cst_0 : tensor<8x1xi64> loc(#loc59) + %p_y_16 = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc60) + %p_y_17 = tt.addptr %p_y_16, %p_y_15 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc60) + %p_y_18 = tt.broadcast %p_y_17 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc61) + %p_y_19 = tt.broadcast %m_x_10 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc61) + %p_y_20 = tt.addptr %p_y_18, %p_y_19 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc61) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr> loc(#loc62) + %p_rstd_21 = tt.addptr %p_rstd, %o_t_6 : tensor<8x!tt.ptr>, tensor<8xi64> loc(#loc62) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc63) + %p_dy_22 = tt.addptr %p_dy, %p_y_15 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc63) + %p_dy_23 = tt.broadcast %p_dy_22 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc64) + %p_dy_24 = tt.addptr %p_dy_23, %p_y_19 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc64) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc65) + %p_dx_25 = tt.addptr %p_dx, %p_y_15 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc65) + %p_dx_26 = tt.broadcast %p_dx_25 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc66) + %p_dx_27 = tt.addptr %p_dx_26, %p_y_19 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc66) + %b_y = tt.load %p_y_20, %m_x_14, %cst : tensor<8x128x!tt.ptr> loc(#loc67) + %b_y_28 = arith.extf %b_y : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc68) + %b_rstd_29 = tt.load %p_rstd_21, %m_t_8, %b_rstd : tensor<8x!tt.ptr> loc(#loc46) + %b_dy = tt.load %p_dy_24, %m_x_14, %cst : tensor<8x128x!tt.ptr> loc(#loc69) + %b_dy_30 = arith.extf %b_dy : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc70) + %b_dx = tt.expand_dims %b_rstd_29 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc71) + %b_dx_31 = tt.broadcast %b_dx : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc72) + %b_dx_32 = arith.mulf %b_dy_30, %b_dx_31 : tensor<8x128xf32> loc(#loc72) + %b_dx_33 = arith.mulf %b_dy_30, %b_y_28 : tensor<8x128xf32> loc(#loc73) + %b_dx_34 = "tt.reduce"(%b_dx_33) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_40: f32 loc(callsite(#loc1 at #loc74)), %b_dx_41: f32 loc(callsite(#loc1 at #loc74))): + %b_dx_42 = arith.addf %b_dx_40, %b_dx_41 : f32 loc(#loc81) + tt.reduce.return %b_dx_42 : f32 loc(#loc79) + }) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc79) + %b_dx_35 = tt.expand_dims %b_dx_34 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc75) + %b_dx_36 = tt.broadcast %b_dx_35 : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc76) + %b_dx_37 = arith.mulf %b_dx_36, %b_y_28 : tensor<8x128xf32> loc(#loc76) + %b_dx_38 = arith.mulf %b_dx_37, %b_dx_31 : tensor<8x128xf32> loc(#loc77) + %b_dx_39 = arith.subf %b_dx_32, %b_dx_38 : tensor<8x128xf32> loc(#loc78) + %0 = arith.truncf %b_dx_39 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc37) + tt.store %p_dx_27, %0, %m_x_14 : tensor<8x128x!tt.ptr> loc(#loc38) + tt.return loc(#loc39) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("b_rstd"(#loc2)) +#loc47 = loc("m_x"(#loc3)) +#loc48 = loc("o_t"(#loc4)) +#loc49 = loc("i_t"(#loc5)) +#loc50 = loc("i_t"(#loc6)) +#loc51 = loc("o_t"(#loc7)) +#loc52 = loc("o_t"(#loc8)) +#loc53 = loc("o_d"(#loc9)) +#loc54 = loc("m_t"(#loc10)) +#loc55 = loc("m_x"(#loc11)) +#loc56 = loc("m_x"(#loc12)) +#loc57 = loc("m_x"(#loc13)) +#loc58 = loc("p_y"(#loc14)) +#loc59 = loc("p_y"(#loc15)) +#loc60 = loc("p_y"(#loc16)) +#loc61 = loc("p_y"(#loc17)) +#loc62 = loc("p_rstd"(#loc18)) +#loc63 = loc("p_dy"(#loc19)) +#loc64 = loc("p_dy"(#loc20)) +#loc65 = loc("p_dx"(#loc21)) +#loc66 = loc("p_dx"(#loc22)) +#loc67 = loc("b_y"(#loc23)) +#loc68 = loc("b_y"(#loc24)) +#loc69 = loc("b_dy"(#loc25)) +#loc70 = loc("b_dy"(#loc26)) +#loc71 = loc("b_dx"(#loc27)) +#loc72 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc29)) +#loc75 = loc("b_dx"(#loc33)) +#loc76 = loc("b_dx"(#loc34)) +#loc77 = loc("b_dx"(#loc35)) +#loc78 = loc("b_dx"(#loc36)) +#loc79 = loc(callsite(#loc30 at #loc74)) +#loc81 = loc(callsite(#loc32 at #loc79)) diff --git a/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/__grp__l2norm_bwd_kernel.json b/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/__grp__l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..658982d0159cf9b06ffa82fb6be56ba6cb687050 --- /dev/null +++ b/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/__grp__l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_bwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/l2norm_bwd_kernel.source", "l2norm_bwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/l2norm_bwd_kernel.ttir", "l2norm_bwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/l2norm_bwd_kernel.ttgir", "l2norm_bwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/l2norm_bwd_kernel.llir", "l2norm_bwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/l2norm_bwd_kernel.ptx", "l2norm_bwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/l2norm_bwd_kernel.cubin", "l2norm_bwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/l2norm_bwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/l2norm_bwd_kernel.cubin b/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/l2norm_bwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..7d983c42ab235cca77589520bc84b6224077598e Binary files /dev/null and b/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/l2norm_bwd_kernel.cubin differ diff --git a/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/l2norm_bwd_kernel.json b/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..78604234c1d0638a727402f0b95ad1ea0f91f189 --- /dev/null +++ b/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"hash": "09c0f2622d04109d24c0acc1dc46c84fc8e5f95a29f1bf0d5700d934b1a619fa", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 8, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 0, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_bwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/l2norm_bwd_kernel.llir b/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/l2norm_bwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..34ac243461b951de8e152baed95400f872c9ccdc --- /dev/null +++ b/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/l2norm_bwd_kernel.llir @@ -0,0 +1,1269 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_bwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, ptr addrspace(1) %3, float %4, i32 %5, ptr addrspace(1) readnone captures(none) %6, ptr addrspace(1) readnone captures(none) %7) local_unnamed_addr #0 !dbg !4 { + %9 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !12 + %10 = zext nneg i32 %9 to i64, !dbg !13 + %11 = shl nuw nsw i64 %10, 7, !dbg !14 + %12 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !15 + %13 = lshr i32 %12, 4, !dbg !15 + %14 = and i32 %13, 15, !dbg !15 + %15 = or disjoint i32 %14, 16, !dbg !15 + %16 = insertelement <2 x i32> poison, i32 %13, i64 0, !dbg !15 + %17 = insertelement <2 x i32> %16, i32 %14, i64 1, !dbg !15 + %18 = or <2 x i32> %17, , !dbg !15 + %19 = insertelement <4 x i32> poison, i32 %13, i64 0, !dbg !15 + %20 = insertelement <4 x i32> %19, i32 %14, i64 1, !dbg !15 + %21 = shufflevector <2 x i32> %18, <2 x i32> poison, <8 x i32> , !dbg !16 + %22 = insertelement <8 x i32> %21, i32 %15, i64 6, !dbg !16 + %23 = insertelement <8 x i32> %22, i32 %14, i64 7, !dbg !16 + %24 = shufflevector <4 x i32> %20, <4 x i32> poison, <8 x i32> , !dbg !16 + %25 = or <8 x i32> %24, , !dbg !16 + %26 = shufflevector <8 x i32> %25, <8 x i32> %23, <8 x i32> , !dbg !16 + %27 = zext <8 x i32> %26 to <8 x i64>, !dbg !16 + %28 = insertelement <8 x i64> poison, i64 %11, i64 0, !dbg !16 + %29 = shufflevector <8 x i64> %28, <8 x i64> poison, <8 x i32> zeroinitializer, !dbg !16 + %30 = or disjoint <8 x i64> %29, %27, !dbg !16 + %31 = insertelement <8 x i32> poison, i32 %5, i64 0, !dbg !17 + %32 = shufflevector <8 x i32> %31, <8 x i32> poison, <8 x i32> zeroinitializer, !dbg !17 + %33 = sext <8 x i32> %32 to <8 x i64>, !dbg !17 + %34 = icmp slt <8 x i64> %30, %33, !dbg !17 + %35 = shl nuw nsw i32 %12, 3, !dbg !18 + %36 = and i32 %35, 120, !dbg !18 + %37 = extractelement <8 x i64> %30, i64 7, !dbg !19 + %38 = shl nuw nsw i64 %37, 7, !dbg !19 + %39 = extractelement <8 x i64> %30, i64 6, !dbg !19 + %40 = shl nuw nsw i64 %39, 7, !dbg !19 + %41 = extractelement <8 x i64> %30, i64 5, !dbg !19 + %42 = shl nuw nsw i64 %41, 7, !dbg !19 + %43 = extractelement <8 x i64> %30, i64 4, !dbg !19 + %44 = shl nuw nsw i64 %43, 7, !dbg !19 + %45 = extractelement <8 x i64> %30, i64 3, !dbg !19 + %46 = shl nuw nsw i64 %45, 7, !dbg !19 + %47 = extractelement <8 x i64> %30, i64 2, !dbg !19 + %48 = shl nuw nsw i64 %47, 7, !dbg !19 + %49 = extractelement <8 x i64> %30, i64 1, !dbg !19 + %50 = shl nuw nsw i64 %49, 7, !dbg !19 + %51 = extractelement <8 x i64> %30, i64 0, !dbg !19 + %52 = shl nuw nsw i64 %51, 7, !dbg !19 + %53 = getelementptr bfloat, ptr addrspace(1) %0, i64 %38, !dbg !20 + %54 = getelementptr bfloat, ptr addrspace(1) %0, i64 %40, !dbg !20 + %55 = getelementptr bfloat, ptr addrspace(1) %0, i64 %42, !dbg !20 + %56 = getelementptr bfloat, ptr addrspace(1) %0, i64 %44, !dbg !20 + %57 = getelementptr bfloat, ptr addrspace(1) %0, i64 %46, !dbg !20 + %58 = getelementptr bfloat, ptr addrspace(1) %0, i64 %48, !dbg !20 + %59 = getelementptr bfloat, ptr addrspace(1) %0, i64 %50, !dbg !20 + %60 = getelementptr bfloat, ptr addrspace(1) %0, i64 %52, !dbg !20 + %61 = zext nneg i32 %36 to i64, !dbg !21 + %62 = getelementptr bfloat, ptr addrspace(1) %53, i64 %61, !dbg !21 + %63 = getelementptr bfloat, ptr addrspace(1) %54, i64 %61, !dbg !21 + %64 = getelementptr bfloat, ptr addrspace(1) %55, i64 %61, !dbg !21 + %65 = getelementptr bfloat, ptr addrspace(1) %56, i64 %61, !dbg !21 + %66 = getelementptr bfloat, ptr addrspace(1) %57, i64 %61, !dbg !21 + %67 = getelementptr bfloat, ptr addrspace(1) %58, i64 %61, !dbg !21 + %68 = getelementptr bfloat, ptr addrspace(1) %59, i64 %61, !dbg !21 + %69 = getelementptr bfloat, ptr addrspace(1) %60, i64 %61, !dbg !21 + %70 = getelementptr float, ptr addrspace(1) %1, i64 %37, !dbg !22 + %71 = getelementptr float, ptr addrspace(1) %1, i64 %39, !dbg !22 + %72 = getelementptr float, ptr addrspace(1) %1, i64 %41, !dbg !22 + %73 = getelementptr float, ptr addrspace(1) %1, i64 %43, !dbg !22 + %74 = getelementptr float, ptr addrspace(1) %1, i64 %45, !dbg !22 + %75 = getelementptr float, ptr addrspace(1) %1, i64 %47, !dbg !22 + %76 = getelementptr float, ptr addrspace(1) %1, i64 %49, !dbg !22 + %77 = getelementptr float, ptr addrspace(1) %1, i64 %51, !dbg !22 + %78 = getelementptr bfloat, ptr addrspace(1) %2, i64 %38, !dbg !23 + %79 = getelementptr bfloat, ptr addrspace(1) %2, i64 %40, !dbg !23 + %80 = getelementptr bfloat, ptr addrspace(1) %2, i64 %42, !dbg !23 + %81 = getelementptr bfloat, ptr addrspace(1) %2, i64 %44, !dbg !23 + %82 = getelementptr bfloat, ptr addrspace(1) %2, i64 %46, !dbg !23 + %83 = getelementptr bfloat, ptr addrspace(1) %2, i64 %48, !dbg !23 + %84 = getelementptr bfloat, ptr addrspace(1) %2, i64 %50, !dbg !23 + %85 = getelementptr bfloat, ptr addrspace(1) %2, i64 %52, !dbg !23 + %86 = getelementptr bfloat, ptr addrspace(1) %78, i64 %61, !dbg !24 + %87 = getelementptr bfloat, ptr addrspace(1) %79, i64 %61, !dbg !24 + %88 = getelementptr bfloat, ptr addrspace(1) %80, i64 %61, !dbg !24 + %89 = getelementptr bfloat, ptr addrspace(1) %81, i64 %61, !dbg !24 + %90 = getelementptr bfloat, ptr addrspace(1) %82, i64 %61, !dbg !24 + %91 = getelementptr bfloat, ptr addrspace(1) %83, i64 %61, !dbg !24 + %92 = getelementptr bfloat, ptr addrspace(1) %84, i64 %61, !dbg !24 + %93 = getelementptr bfloat, ptr addrspace(1) %85, i64 %61, !dbg !24 + %94 = getelementptr bfloat, ptr addrspace(1) %3, i64 %38, !dbg !25 + %95 = getelementptr bfloat, ptr addrspace(1) %3, i64 %40, !dbg !25 + %96 = getelementptr bfloat, ptr addrspace(1) %3, i64 %42, !dbg !25 + %97 = getelementptr bfloat, ptr addrspace(1) %3, i64 %44, !dbg !25 + %98 = getelementptr bfloat, ptr addrspace(1) %3, i64 %46, !dbg !25 + %99 = getelementptr bfloat, ptr addrspace(1) %3, i64 %48, !dbg !25 + %100 = getelementptr bfloat, ptr addrspace(1) %3, i64 %50, !dbg !25 + %101 = getelementptr bfloat, ptr addrspace(1) %3, i64 %52, !dbg !25 + %102 = getelementptr bfloat, ptr addrspace(1) %94, i64 %61, !dbg !26 + %103 = getelementptr bfloat, ptr addrspace(1) %95, i64 %61, !dbg !26 + %104 = getelementptr bfloat, ptr addrspace(1) %96, i64 %61, !dbg !26 + %105 = getelementptr bfloat, ptr addrspace(1) %97, i64 %61, !dbg !26 + %106 = getelementptr bfloat, ptr addrspace(1) %98, i64 %61, !dbg !26 + %107 = getelementptr bfloat, ptr addrspace(1) %99, i64 %61, !dbg !26 + %108 = getelementptr bfloat, ptr addrspace(1) %100, i64 %61, !dbg !26 + %109 = getelementptr bfloat, ptr addrspace(1) %101, i64 %61, !dbg !26 + %110 = extractelement <8 x i1> %34, i64 7, !dbg !27 + %111 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %62, i1 %110) #3, !dbg !27 + %112 = extractvalue { i32, i32, i32, i32 } %111, 0, !dbg !27 + %113 = bitcast i32 %112 to <2 x bfloat>, !dbg !27 + %114 = extractvalue { i32, i32, i32, i32 } %111, 1, !dbg !27 + %115 = bitcast i32 %114 to <2 x bfloat>, !dbg !27 + %116 = extractvalue { i32, i32, i32, i32 } %111, 2, !dbg !27 + %117 = bitcast i32 %116 to <2 x bfloat>, !dbg !27 + %118 = extractvalue { i32, i32, i32, i32 } %111, 3, !dbg !27 + %119 = bitcast i32 %118 to <2 x bfloat>, !dbg !27 + %120 = extractelement <2 x bfloat> %113, i64 0, !dbg !27 + %121 = extractelement <2 x bfloat> %113, i64 1, !dbg !27 + %122 = extractelement <2 x bfloat> %115, i64 0, !dbg !27 + %123 = extractelement <2 x bfloat> %115, i64 1, !dbg !27 + %124 = extractelement <2 x bfloat> %117, i64 0, !dbg !27 + %125 = extractelement <2 x bfloat> %117, i64 1, !dbg !27 + %126 = extractelement <2 x bfloat> %119, i64 0, !dbg !27 + %127 = extractelement <2 x bfloat> %119, i64 1, !dbg !27 + %128 = extractelement <8 x i1> %34, i64 6, !dbg !27 + %129 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %63, i1 %128) #3, !dbg !27 + %130 = extractvalue { i32, i32, i32, i32 } %129, 0, !dbg !27 + %131 = bitcast i32 %130 to <2 x bfloat>, !dbg !27 + %132 = extractvalue { i32, i32, i32, i32 } %129, 1, !dbg !27 + %133 = bitcast i32 %132 to <2 x bfloat>, !dbg !27 + %134 = extractvalue { i32, i32, i32, i32 } %129, 2, !dbg !27 + %135 = bitcast i32 %134 to <2 x bfloat>, !dbg !27 + %136 = extractvalue { i32, i32, i32, i32 } %129, 3, !dbg !27 + %137 = bitcast i32 %136 to <2 x bfloat>, !dbg !27 + %138 = extractelement <2 x bfloat> %131, i64 0, !dbg !27 + %139 = extractelement <2 x bfloat> %131, i64 1, !dbg !27 + %140 = extractelement <2 x bfloat> %133, i64 0, !dbg !27 + %141 = extractelement <2 x bfloat> %133, i64 1, !dbg !27 + %142 = extractelement <2 x bfloat> %135, i64 0, !dbg !27 + %143 = extractelement <2 x bfloat> %135, i64 1, !dbg !27 + %144 = extractelement <2 x bfloat> %137, i64 0, !dbg !27 + %145 = extractelement <2 x bfloat> %137, i64 1, !dbg !27 + %146 = extractelement <8 x i1> %34, i64 5, !dbg !27 + %147 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %64, i1 %146) #3, !dbg !27 + %148 = extractvalue { i32, i32, i32, i32 } %147, 0, !dbg !27 + %149 = bitcast i32 %148 to <2 x bfloat>, !dbg !27 + %150 = extractvalue { i32, i32, i32, i32 } %147, 1, !dbg !27 + %151 = bitcast i32 %150 to <2 x bfloat>, !dbg !27 + %152 = extractvalue { i32, i32, i32, i32 } %147, 2, !dbg !27 + %153 = bitcast i32 %152 to <2 x bfloat>, !dbg !27 + %154 = extractvalue { i32, i32, i32, i32 } %147, 3, !dbg !27 + %155 = bitcast i32 %154 to <2 x bfloat>, !dbg !27 + %156 = extractelement <2 x bfloat> %149, i64 0, !dbg !27 + %157 = extractelement <2 x bfloat> %149, i64 1, !dbg !27 + %158 = extractelement <2 x bfloat> %151, i64 0, !dbg !27 + %159 = extractelement <2 x bfloat> %151, i64 1, !dbg !27 + %160 = extractelement <2 x bfloat> %153, i64 0, !dbg !27 + %161 = extractelement <2 x bfloat> %153, i64 1, !dbg !27 + %162 = extractelement <2 x bfloat> %155, i64 0, !dbg !27 + %163 = extractelement <2 x bfloat> %155, i64 1, !dbg !27 + %164 = extractelement <8 x i1> %34, i64 4, !dbg !27 + %165 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %65, i1 %164) #3, !dbg !27 + %166 = extractvalue { i32, i32, i32, i32 } %165, 0, !dbg !27 + %167 = bitcast i32 %166 to <2 x bfloat>, !dbg !27 + %168 = extractvalue { i32, i32, i32, i32 } %165, 1, !dbg !27 + %169 = bitcast i32 %168 to <2 x bfloat>, !dbg !27 + %170 = extractvalue { i32, i32, i32, i32 } %165, 2, !dbg !27 + %171 = bitcast i32 %170 to <2 x bfloat>, !dbg !27 + %172 = extractvalue { i32, i32, i32, i32 } %165, 3, !dbg !27 + %173 = bitcast i32 %172 to <2 x bfloat>, !dbg !27 + %174 = extractelement <2 x bfloat> %167, i64 0, !dbg !27 + %175 = extractelement <2 x bfloat> %167, i64 1, !dbg !27 + %176 = extractelement <2 x bfloat> %169, i64 0, !dbg !27 + %177 = extractelement <2 x bfloat> %169, i64 1, !dbg !27 + %178 = extractelement <2 x bfloat> %171, i64 0, !dbg !27 + %179 = extractelement <2 x bfloat> %171, i64 1, !dbg !27 + %180 = extractelement <2 x bfloat> %173, i64 0, !dbg !27 + %181 = extractelement <2 x bfloat> %173, i64 1, !dbg !27 + %182 = extractelement <8 x i1> %34, i64 3, !dbg !27 + %183 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %66, i1 %182) #3, !dbg !27 + %184 = extractvalue { i32, i32, i32, i32 } %183, 0, !dbg !27 + %185 = bitcast i32 %184 to <2 x bfloat>, !dbg !27 + %186 = extractvalue { i32, i32, i32, i32 } %183, 1, !dbg !27 + %187 = bitcast i32 %186 to <2 x bfloat>, !dbg !27 + %188 = extractvalue { i32, i32, i32, i32 } %183, 2, !dbg !27 + %189 = bitcast i32 %188 to <2 x bfloat>, !dbg !27 + %190 = extractvalue { i32, i32, i32, i32 } %183, 3, !dbg !27 + %191 = bitcast i32 %190 to <2 x bfloat>, !dbg !27 + %192 = extractelement <2 x bfloat> %185, i64 0, !dbg !27 + %193 = extractelement <2 x bfloat> %185, i64 1, !dbg !27 + %194 = extractelement <2 x bfloat> %187, i64 0, !dbg !27 + %195 = extractelement <2 x bfloat> %187, i64 1, !dbg !27 + %196 = extractelement <2 x bfloat> %189, i64 0, !dbg !27 + %197 = extractelement <2 x bfloat> %189, i64 1, !dbg !27 + %198 = extractelement <2 x bfloat> %191, i64 0, !dbg !27 + %199 = extractelement <2 x bfloat> %191, i64 1, !dbg !27 + %200 = extractelement <8 x i1> %34, i64 2, !dbg !27 + %201 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %67, i1 %200) #3, !dbg !27 + %202 = extractvalue { i32, i32, i32, i32 } %201, 0, !dbg !27 + %203 = bitcast i32 %202 to <2 x bfloat>, !dbg !27 + %204 = extractvalue { i32, i32, i32, i32 } %201, 1, !dbg !27 + %205 = bitcast i32 %204 to <2 x bfloat>, !dbg !27 + %206 = extractvalue { i32, i32, i32, i32 } %201, 2, !dbg !27 + %207 = bitcast i32 %206 to <2 x bfloat>, !dbg !27 + %208 = extractvalue { i32, i32, i32, i32 } %201, 3, !dbg !27 + %209 = bitcast i32 %208 to <2 x bfloat>, !dbg !27 + %210 = extractelement <2 x bfloat> %203, i64 0, !dbg !27 + %211 = extractelement <2 x bfloat> %203, i64 1, !dbg !27 + %212 = extractelement <2 x bfloat> %205, i64 0, !dbg !27 + %213 = extractelement <2 x bfloat> %205, i64 1, !dbg !27 + %214 = extractelement <2 x bfloat> %207, i64 0, !dbg !27 + %215 = extractelement <2 x bfloat> %207, i64 1, !dbg !27 + %216 = extractelement <2 x bfloat> %209, i64 0, !dbg !27 + %217 = extractelement <2 x bfloat> %209, i64 1, !dbg !27 + %218 = extractelement <8 x i1> %34, i64 1, !dbg !27 + %219 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %68, i1 %218) #3, !dbg !27 + %220 = extractvalue { i32, i32, i32, i32 } %219, 0, !dbg !27 + %221 = bitcast i32 %220 to <2 x bfloat>, !dbg !27 + %222 = extractvalue { i32, i32, i32, i32 } %219, 1, !dbg !27 + %223 = bitcast i32 %222 to <2 x bfloat>, !dbg !27 + %224 = extractvalue { i32, i32, i32, i32 } %219, 2, !dbg !27 + %225 = bitcast i32 %224 to <2 x bfloat>, !dbg !27 + %226 = extractvalue { i32, i32, i32, i32 } %219, 3, !dbg !27 + %227 = bitcast i32 %226 to <2 x bfloat>, !dbg !27 + %228 = extractelement <2 x bfloat> %221, i64 0, !dbg !27 + %229 = extractelement <2 x bfloat> %221, i64 1, !dbg !27 + %230 = extractelement <2 x bfloat> %223, i64 0, !dbg !27 + %231 = extractelement <2 x bfloat> %223, i64 1, !dbg !27 + %232 = extractelement <2 x bfloat> %225, i64 0, !dbg !27 + %233 = extractelement <2 x bfloat> %225, i64 1, !dbg !27 + %234 = extractelement <2 x bfloat> %227, i64 0, !dbg !27 + %235 = extractelement <2 x bfloat> %227, i64 1, !dbg !27 + %236 = extractelement <8 x i1> %34, i64 0, !dbg !27 + %237 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %69, i1 %236) #3, !dbg !27 + %238 = extractvalue { i32, i32, i32, i32 } %237, 0, !dbg !27 + %239 = bitcast i32 %238 to <2 x bfloat>, !dbg !27 + %240 = extractvalue { i32, i32, i32, i32 } %237, 1, !dbg !27 + %241 = bitcast i32 %240 to <2 x bfloat>, !dbg !27 + %242 = extractvalue { i32, i32, i32, i32 } %237, 2, !dbg !27 + %243 = bitcast i32 %242 to <2 x bfloat>, !dbg !27 + %244 = extractvalue { i32, i32, i32, i32 } %237, 3, !dbg !27 + %245 = bitcast i32 %244 to <2 x bfloat>, !dbg !27 + %246 = extractelement <2 x bfloat> %239, i64 0, !dbg !27 + %247 = extractelement <2 x bfloat> %239, i64 1, !dbg !27 + %248 = extractelement <2 x bfloat> %241, i64 0, !dbg !27 + %249 = extractelement <2 x bfloat> %241, i64 1, !dbg !27 + %250 = extractelement <2 x bfloat> %243, i64 0, !dbg !27 + %251 = extractelement <2 x bfloat> %243, i64 1, !dbg !27 + %252 = extractelement <2 x bfloat> %245, i64 0, !dbg !27 + %253 = extractelement <2 x bfloat> %245, i64 1, !dbg !27 + %254 = fpext bfloat %120 to float, !dbg !28 + %255 = fpext bfloat %121 to float, !dbg !28 + %256 = fpext bfloat %122 to float, !dbg !28 + %257 = fpext bfloat %123 to float, !dbg !28 + %258 = fpext bfloat %124 to float, !dbg !28 + %259 = fpext bfloat %125 to float, !dbg !28 + %260 = fpext bfloat %126 to float, !dbg !28 + %261 = fpext bfloat %127 to float, !dbg !28 + %262 = fpext bfloat %138 to float, !dbg !28 + %263 = fpext bfloat %139 to float, !dbg !28 + %264 = fpext bfloat %140 to float, !dbg !28 + %265 = fpext bfloat %141 to float, !dbg !28 + %266 = fpext bfloat %142 to float, !dbg !28 + %267 = fpext bfloat %143 to float, !dbg !28 + %268 = fpext bfloat %144 to float, !dbg !28 + %269 = fpext bfloat %145 to float, !dbg !28 + %270 = fpext bfloat %156 to float, !dbg !28 + %271 = fpext bfloat %157 to float, !dbg !28 + %272 = fpext bfloat %158 to float, !dbg !28 + %273 = fpext bfloat %159 to float, !dbg !28 + %274 = fpext bfloat %160 to float, !dbg !28 + %275 = fpext bfloat %161 to float, !dbg !28 + %276 = fpext bfloat %162 to float, !dbg !28 + %277 = fpext bfloat %163 to float, !dbg !28 + %278 = fpext bfloat %174 to float, !dbg !28 + %279 = fpext bfloat %175 to float, !dbg !28 + %280 = fpext bfloat %176 to float, !dbg !28 + %281 = fpext bfloat %177 to float, !dbg !28 + %282 = fpext bfloat %178 to float, !dbg !28 + %283 = fpext bfloat %179 to float, !dbg !28 + %284 = fpext bfloat %180 to float, !dbg !28 + %285 = fpext bfloat %181 to float, !dbg !28 + %286 = fpext bfloat %192 to float, !dbg !28 + %287 = fpext bfloat %193 to float, !dbg !28 + %288 = fpext bfloat %194 to float, !dbg !28 + %289 = fpext bfloat %195 to float, !dbg !28 + %290 = fpext bfloat %196 to float, !dbg !28 + %291 = fpext bfloat %197 to float, !dbg !28 + %292 = fpext bfloat %198 to float, !dbg !28 + %293 = fpext bfloat %199 to float, !dbg !28 + %294 = fpext bfloat %210 to float, !dbg !28 + %295 = fpext bfloat %211 to float, !dbg !28 + %296 = fpext bfloat %212 to float, !dbg !28 + %297 = fpext bfloat %213 to float, !dbg !28 + %298 = fpext bfloat %214 to float, !dbg !28 + %299 = fpext bfloat %215 to float, !dbg !28 + %300 = fpext bfloat %216 to float, !dbg !28 + %301 = fpext bfloat %217 to float, !dbg !28 + %302 = fpext bfloat %228 to float, !dbg !28 + %303 = fpext bfloat %229 to float, !dbg !28 + %304 = fpext bfloat %230 to float, !dbg !28 + %305 = fpext bfloat %231 to float, !dbg !28 + %306 = fpext bfloat %232 to float, !dbg !28 + %307 = fpext bfloat %233 to float, !dbg !28 + %308 = fpext bfloat %234 to float, !dbg !28 + %309 = fpext bfloat %235 to float, !dbg !28 + %310 = fpext bfloat %246 to float, !dbg !28 + %311 = fpext bfloat %247 to float, !dbg !28 + %312 = fpext bfloat %248 to float, !dbg !28 + %313 = fpext bfloat %249 to float, !dbg !28 + %314 = fpext bfloat %250 to float, !dbg !28 + %315 = fpext bfloat %251 to float, !dbg !28 + %316 = fpext bfloat %252 to float, !dbg !28 + %317 = fpext bfloat %253 to float, !dbg !28 + %318 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %70, i1 %110) #3, !dbg !29 + %319 = bitcast i32 %318 to float, !dbg !29 + %320 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %71, i1 %128) #3, !dbg !29 + %321 = bitcast i32 %320 to float, !dbg !29 + %322 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %72, i1 %146) #3, !dbg !29 + %323 = bitcast i32 %322 to float, !dbg !29 + %324 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %73, i1 %164) #3, !dbg !29 + %325 = bitcast i32 %324 to float, !dbg !29 + %326 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %74, i1 %182) #3, !dbg !29 + %327 = bitcast i32 %326 to float, !dbg !29 + %328 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %75, i1 %200) #3, !dbg !29 + %329 = bitcast i32 %328 to float, !dbg !29 + %330 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %76, i1 %218) #3, !dbg !29 + %331 = bitcast i32 %330 to float, !dbg !29 + %332 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %77, i1 %236) #3, !dbg !29 + %333 = bitcast i32 %332 to float, !dbg !29 + %334 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %86, i1 %110) #3, !dbg !30 + %335 = extractvalue { i32, i32, i32, i32 } %334, 0, !dbg !30 + %336 = bitcast i32 %335 to <2 x bfloat>, !dbg !30 + %337 = extractvalue { i32, i32, i32, i32 } %334, 1, !dbg !30 + %338 = bitcast i32 %337 to <2 x bfloat>, !dbg !30 + %339 = extractvalue { i32, i32, i32, i32 } %334, 2, !dbg !30 + %340 = bitcast i32 %339 to <2 x bfloat>, !dbg !30 + %341 = extractvalue { i32, i32, i32, i32 } %334, 3, !dbg !30 + %342 = bitcast i32 %341 to <2 x bfloat>, !dbg !30 + %343 = extractelement <2 x bfloat> %336, i64 0, !dbg !30 + %344 = extractelement <2 x bfloat> %336, i64 1, !dbg !30 + %345 = extractelement <2 x bfloat> %338, i64 0, !dbg !30 + %346 = extractelement <2 x bfloat> %338, i64 1, !dbg !30 + %347 = extractelement <2 x bfloat> %340, i64 0, !dbg !30 + %348 = extractelement <2 x bfloat> %340, i64 1, !dbg !30 + %349 = extractelement <2 x bfloat> %342, i64 0, !dbg !30 + %350 = extractelement <2 x bfloat> %342, i64 1, !dbg !30 + %351 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %87, i1 %128) #3, !dbg !30 + %352 = extractvalue { i32, i32, i32, i32 } %351, 0, !dbg !30 + %353 = bitcast i32 %352 to <2 x bfloat>, !dbg !30 + %354 = extractvalue { i32, i32, i32, i32 } %351, 1, !dbg !30 + %355 = bitcast i32 %354 to <2 x bfloat>, !dbg !30 + %356 = extractvalue { i32, i32, i32, i32 } %351, 2, !dbg !30 + %357 = bitcast i32 %356 to <2 x bfloat>, !dbg !30 + %358 = extractvalue { i32, i32, i32, i32 } %351, 3, !dbg !30 + %359 = bitcast i32 %358 to <2 x bfloat>, !dbg !30 + %360 = extractelement <2 x bfloat> %353, i64 0, !dbg !30 + %361 = extractelement <2 x bfloat> %353, i64 1, !dbg !30 + %362 = extractelement <2 x bfloat> %355, i64 0, !dbg !30 + %363 = extractelement <2 x bfloat> %355, i64 1, !dbg !30 + %364 = extractelement <2 x bfloat> %357, i64 0, !dbg !30 + %365 = extractelement <2 x bfloat> %357, i64 1, !dbg !30 + %366 = extractelement <2 x bfloat> %359, i64 0, !dbg !30 + %367 = extractelement <2 x bfloat> %359, i64 1, !dbg !30 + %368 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %88, i1 %146) #3, !dbg !30 + %369 = extractvalue { i32, i32, i32, i32 } %368, 0, !dbg !30 + %370 = bitcast i32 %369 to <2 x bfloat>, !dbg !30 + %371 = extractvalue { i32, i32, i32, i32 } %368, 1, !dbg !30 + %372 = bitcast i32 %371 to <2 x bfloat>, !dbg !30 + %373 = extractvalue { i32, i32, i32, i32 } %368, 2, !dbg !30 + %374 = bitcast i32 %373 to <2 x bfloat>, !dbg !30 + %375 = extractvalue { i32, i32, i32, i32 } %368, 3, !dbg !30 + %376 = bitcast i32 %375 to <2 x bfloat>, !dbg !30 + %377 = extractelement <2 x bfloat> %370, i64 0, !dbg !30 + %378 = extractelement <2 x bfloat> %370, i64 1, !dbg !30 + %379 = extractelement <2 x bfloat> %372, i64 0, !dbg !30 + %380 = extractelement <2 x bfloat> %372, i64 1, !dbg !30 + %381 = extractelement <2 x bfloat> %374, i64 0, !dbg !30 + %382 = extractelement <2 x bfloat> %374, i64 1, !dbg !30 + %383 = extractelement <2 x bfloat> %376, i64 0, !dbg !30 + %384 = extractelement <2 x bfloat> %376, i64 1, !dbg !30 + %385 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %89, i1 %164) #3, !dbg !30 + %386 = extractvalue { i32, i32, i32, i32 } %385, 0, !dbg !30 + %387 = bitcast i32 %386 to <2 x bfloat>, !dbg !30 + %388 = extractvalue { i32, i32, i32, i32 } %385, 1, !dbg !30 + %389 = bitcast i32 %388 to <2 x bfloat>, !dbg !30 + %390 = extractvalue { i32, i32, i32, i32 } %385, 2, !dbg !30 + %391 = bitcast i32 %390 to <2 x bfloat>, !dbg !30 + %392 = extractvalue { i32, i32, i32, i32 } %385, 3, !dbg !30 + %393 = bitcast i32 %392 to <2 x bfloat>, !dbg !30 + %394 = extractelement <2 x bfloat> %387, i64 0, !dbg !30 + %395 = extractelement <2 x bfloat> %387, i64 1, !dbg !30 + %396 = extractelement <2 x bfloat> %389, i64 0, !dbg !30 + %397 = extractelement <2 x bfloat> %389, i64 1, !dbg !30 + %398 = extractelement <2 x bfloat> %391, i64 0, !dbg !30 + %399 = extractelement <2 x bfloat> %391, i64 1, !dbg !30 + %400 = extractelement <2 x bfloat> %393, i64 0, !dbg !30 + %401 = extractelement <2 x bfloat> %393, i64 1, !dbg !30 + %402 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %90, i1 %182) #3, !dbg !30 + %403 = extractvalue { i32, i32, i32, i32 } %402, 0, !dbg !30 + %404 = bitcast i32 %403 to <2 x bfloat>, !dbg !30 + %405 = extractvalue { i32, i32, i32, i32 } %402, 1, !dbg !30 + %406 = bitcast i32 %405 to <2 x bfloat>, !dbg !30 + %407 = extractvalue { i32, i32, i32, i32 } %402, 2, !dbg !30 + %408 = bitcast i32 %407 to <2 x bfloat>, !dbg !30 + %409 = extractvalue { i32, i32, i32, i32 } %402, 3, !dbg !30 + %410 = bitcast i32 %409 to <2 x bfloat>, !dbg !30 + %411 = extractelement <2 x bfloat> %404, i64 0, !dbg !30 + %412 = extractelement <2 x bfloat> %404, i64 1, !dbg !30 + %413 = extractelement <2 x bfloat> %406, i64 0, !dbg !30 + %414 = extractelement <2 x bfloat> %406, i64 1, !dbg !30 + %415 = extractelement <2 x bfloat> %408, i64 0, !dbg !30 + %416 = extractelement <2 x bfloat> %408, i64 1, !dbg !30 + %417 = extractelement <2 x bfloat> %410, i64 0, !dbg !30 + %418 = extractelement <2 x bfloat> %410, i64 1, !dbg !30 + %419 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %91, i1 %200) #3, !dbg !30 + %420 = extractvalue { i32, i32, i32, i32 } %419, 0, !dbg !30 + %421 = bitcast i32 %420 to <2 x bfloat>, !dbg !30 + %422 = extractvalue { i32, i32, i32, i32 } %419, 1, !dbg !30 + %423 = bitcast i32 %422 to <2 x bfloat>, !dbg !30 + %424 = extractvalue { i32, i32, i32, i32 } %419, 2, !dbg !30 + %425 = bitcast i32 %424 to <2 x bfloat>, !dbg !30 + %426 = extractvalue { i32, i32, i32, i32 } %419, 3, !dbg !30 + %427 = bitcast i32 %426 to <2 x bfloat>, !dbg !30 + %428 = extractelement <2 x bfloat> %421, i64 0, !dbg !30 + %429 = extractelement <2 x bfloat> %421, i64 1, !dbg !30 + %430 = extractelement <2 x bfloat> %423, i64 0, !dbg !30 + %431 = extractelement <2 x bfloat> %423, i64 1, !dbg !30 + %432 = extractelement <2 x bfloat> %425, i64 0, !dbg !30 + %433 = extractelement <2 x bfloat> %425, i64 1, !dbg !30 + %434 = extractelement <2 x bfloat> %427, i64 0, !dbg !30 + %435 = extractelement <2 x bfloat> %427, i64 1, !dbg !30 + %436 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %92, i1 %218) #3, !dbg !30 + %437 = extractvalue { i32, i32, i32, i32 } %436, 0, !dbg !30 + %438 = bitcast i32 %437 to <2 x bfloat>, !dbg !30 + %439 = extractvalue { i32, i32, i32, i32 } %436, 1, !dbg !30 + %440 = bitcast i32 %439 to <2 x bfloat>, !dbg !30 + %441 = extractvalue { i32, i32, i32, i32 } %436, 2, !dbg !30 + %442 = bitcast i32 %441 to <2 x bfloat>, !dbg !30 + %443 = extractvalue { i32, i32, i32, i32 } %436, 3, !dbg !30 + %444 = bitcast i32 %443 to <2 x bfloat>, !dbg !30 + %445 = extractelement <2 x bfloat> %438, i64 0, !dbg !30 + %446 = extractelement <2 x bfloat> %438, i64 1, !dbg !30 + %447 = extractelement <2 x bfloat> %440, i64 0, !dbg !30 + %448 = extractelement <2 x bfloat> %440, i64 1, !dbg !30 + %449 = extractelement <2 x bfloat> %442, i64 0, !dbg !30 + %450 = extractelement <2 x bfloat> %442, i64 1, !dbg !30 + %451 = extractelement <2 x bfloat> %444, i64 0, !dbg !30 + %452 = extractelement <2 x bfloat> %444, i64 1, !dbg !30 + %453 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %93, i1 %236) #3, !dbg !30 + %454 = extractvalue { i32, i32, i32, i32 } %453, 0, !dbg !30 + %455 = bitcast i32 %454 to <2 x bfloat>, !dbg !30 + %456 = extractvalue { i32, i32, i32, i32 } %453, 1, !dbg !30 + %457 = bitcast i32 %456 to <2 x bfloat>, !dbg !30 + %458 = extractvalue { i32, i32, i32, i32 } %453, 2, !dbg !30 + %459 = bitcast i32 %458 to <2 x bfloat>, !dbg !30 + %460 = extractvalue { i32, i32, i32, i32 } %453, 3, !dbg !30 + %461 = bitcast i32 %460 to <2 x bfloat>, !dbg !30 + %462 = extractelement <2 x bfloat> %455, i64 0, !dbg !30 + %463 = extractelement <2 x bfloat> %455, i64 1, !dbg !30 + %464 = extractelement <2 x bfloat> %457, i64 0, !dbg !30 + %465 = extractelement <2 x bfloat> %457, i64 1, !dbg !30 + %466 = extractelement <2 x bfloat> %459, i64 0, !dbg !30 + %467 = extractelement <2 x bfloat> %459, i64 1, !dbg !30 + %468 = extractelement <2 x bfloat> %461, i64 0, !dbg !30 + %469 = extractelement <2 x bfloat> %461, i64 1, !dbg !30 + %470 = fpext bfloat %343 to float, !dbg !31 + %471 = fpext bfloat %344 to float, !dbg !31 + %472 = fpext bfloat %345 to float, !dbg !31 + %473 = fpext bfloat %346 to float, !dbg !31 + %474 = fpext bfloat %347 to float, !dbg !31 + %475 = fpext bfloat %348 to float, !dbg !31 + %476 = fpext bfloat %349 to float, !dbg !31 + %477 = fpext bfloat %350 to float, !dbg !31 + %478 = fpext bfloat %360 to float, !dbg !31 + %479 = fpext bfloat %361 to float, !dbg !31 + %480 = fpext bfloat %362 to float, !dbg !31 + %481 = fpext bfloat %363 to float, !dbg !31 + %482 = fpext bfloat %364 to float, !dbg !31 + %483 = fpext bfloat %365 to float, !dbg !31 + %484 = fpext bfloat %366 to float, !dbg !31 + %485 = fpext bfloat %367 to float, !dbg !31 + %486 = fpext bfloat %377 to float, !dbg !31 + %487 = fpext bfloat %378 to float, !dbg !31 + %488 = fpext bfloat %379 to float, !dbg !31 + %489 = fpext bfloat %380 to float, !dbg !31 + %490 = fpext bfloat %381 to float, !dbg !31 + %491 = fpext bfloat %382 to float, !dbg !31 + %492 = fpext bfloat %383 to float, !dbg !31 + %493 = fpext bfloat %384 to float, !dbg !31 + %494 = fpext bfloat %394 to float, !dbg !31 + %495 = fpext bfloat %395 to float, !dbg !31 + %496 = fpext bfloat %396 to float, !dbg !31 + %497 = fpext bfloat %397 to float, !dbg !31 + %498 = fpext bfloat %398 to float, !dbg !31 + %499 = fpext bfloat %399 to float, !dbg !31 + %500 = fpext bfloat %400 to float, !dbg !31 + %501 = fpext bfloat %401 to float, !dbg !31 + %502 = fpext bfloat %411 to float, !dbg !31 + %503 = fpext bfloat %412 to float, !dbg !31 + %504 = fpext bfloat %413 to float, !dbg !31 + %505 = fpext bfloat %414 to float, !dbg !31 + %506 = fpext bfloat %415 to float, !dbg !31 + %507 = fpext bfloat %416 to float, !dbg !31 + %508 = fpext bfloat %417 to float, !dbg !31 + %509 = fpext bfloat %418 to float, !dbg !31 + %510 = fpext bfloat %428 to float, !dbg !31 + %511 = fpext bfloat %429 to float, !dbg !31 + %512 = fpext bfloat %430 to float, !dbg !31 + %513 = fpext bfloat %431 to float, !dbg !31 + %514 = fpext bfloat %432 to float, !dbg !31 + %515 = fpext bfloat %433 to float, !dbg !31 + %516 = fpext bfloat %434 to float, !dbg !31 + %517 = fpext bfloat %435 to float, !dbg !31 + %518 = fpext bfloat %445 to float, !dbg !31 + %519 = fpext bfloat %446 to float, !dbg !31 + %520 = fpext bfloat %447 to float, !dbg !31 + %521 = fpext bfloat %448 to float, !dbg !31 + %522 = fpext bfloat %449 to float, !dbg !31 + %523 = fpext bfloat %450 to float, !dbg !31 + %524 = fpext bfloat %451 to float, !dbg !31 + %525 = fpext bfloat %452 to float, !dbg !31 + %526 = fpext bfloat %462 to float, !dbg !31 + %527 = fpext bfloat %463 to float, !dbg !31 + %528 = fpext bfloat %464 to float, !dbg !31 + %529 = fpext bfloat %465 to float, !dbg !31 + %530 = fpext bfloat %466 to float, !dbg !31 + %531 = fpext bfloat %467 to float, !dbg !31 + %532 = fpext bfloat %468 to float, !dbg !31 + %533 = fpext bfloat %469 to float, !dbg !31 + %534 = fmul float %319, %470, !dbg !32 + %535 = fmul float %319, %471, !dbg !32 + %536 = fmul float %319, %472, !dbg !32 + %537 = fmul float %319, %473, !dbg !32 + %538 = fmul float %319, %474, !dbg !32 + %539 = fmul float %319, %475, !dbg !32 + %540 = fmul float %319, %476, !dbg !32 + %541 = fmul float %319, %477, !dbg !32 + %542 = fmul float %321, %478, !dbg !32 + %543 = fmul float %321, %479, !dbg !32 + %544 = fmul float %321, %480, !dbg !32 + %545 = fmul float %321, %481, !dbg !32 + %546 = fmul float %321, %482, !dbg !32 + %547 = fmul float %321, %483, !dbg !32 + %548 = fmul float %321, %484, !dbg !32 + %549 = fmul float %321, %485, !dbg !32 + %550 = fmul float %323, %486, !dbg !32 + %551 = fmul float %323, %487, !dbg !32 + %552 = fmul float %323, %488, !dbg !32 + %553 = fmul float %323, %489, !dbg !32 + %554 = fmul float %323, %490, !dbg !32 + %555 = fmul float %323, %491, !dbg !32 + %556 = fmul float %323, %492, !dbg !32 + %557 = fmul float %323, %493, !dbg !32 + %558 = fmul float %325, %494, !dbg !32 + %559 = fmul float %325, %495, !dbg !32 + %560 = fmul float %325, %496, !dbg !32 + %561 = fmul float %325, %497, !dbg !32 + %562 = fmul float %325, %498, !dbg !32 + %563 = fmul float %325, %499, !dbg !32 + %564 = fmul float %325, %500, !dbg !32 + %565 = fmul float %325, %501, !dbg !32 + %566 = fmul float %327, %502, !dbg !32 + %567 = fmul float %327, %503, !dbg !32 + %568 = fmul float %327, %504, !dbg !32 + %569 = fmul float %327, %505, !dbg !32 + %570 = fmul float %327, %506, !dbg !32 + %571 = fmul float %327, %507, !dbg !32 + %572 = fmul float %327, %508, !dbg !32 + %573 = fmul float %327, %509, !dbg !32 + %574 = fmul float %329, %510, !dbg !32 + %575 = fmul float %329, %511, !dbg !32 + %576 = fmul float %329, %512, !dbg !32 + %577 = fmul float %329, %513, !dbg !32 + %578 = fmul float %329, %514, !dbg !32 + %579 = fmul float %329, %515, !dbg !32 + %580 = fmul float %329, %516, !dbg !32 + %581 = fmul float %329, %517, !dbg !32 + %582 = fmul float %331, %518, !dbg !32 + %583 = fmul float %331, %519, !dbg !32 + %584 = fmul float %331, %520, !dbg !32 + %585 = fmul float %331, %521, !dbg !32 + %586 = fmul float %331, %522, !dbg !32 + %587 = fmul float %331, %523, !dbg !32 + %588 = fmul float %331, %524, !dbg !32 + %589 = fmul float %331, %525, !dbg !32 + %590 = fmul float %333, %526, !dbg !32 + %591 = fmul float %333, %527, !dbg !32 + %592 = fmul float %333, %528, !dbg !32 + %593 = fmul float %333, %529, !dbg !32 + %594 = fmul float %333, %530, !dbg !32 + %595 = fmul float %333, %531, !dbg !32 + %596 = fmul float %333, %532, !dbg !32 + %597 = fmul float %333, %533, !dbg !32 + %598 = fmul float %254, %470, !dbg !33 + %599 = fmul float %255, %471, !dbg !33 + %600 = fmul float %256, %472, !dbg !33 + %601 = fmul float %257, %473, !dbg !33 + %602 = fmul float %258, %474, !dbg !33 + %603 = fmul float %259, %475, !dbg !33 + %604 = fmul float %260, %476, !dbg !33 + %605 = fmul float %261, %477, !dbg !33 + %606 = fmul float %262, %478, !dbg !33 + %607 = fmul float %263, %479, !dbg !33 + %608 = fmul float %264, %480, !dbg !33 + %609 = fmul float %265, %481, !dbg !33 + %610 = fmul float %266, %482, !dbg !33 + %611 = fmul float %267, %483, !dbg !33 + %612 = fmul float %268, %484, !dbg !33 + %613 = fmul float %269, %485, !dbg !33 + %614 = fmul float %270, %486, !dbg !33 + %615 = fmul float %271, %487, !dbg !33 + %616 = fmul float %272, %488, !dbg !33 + %617 = fmul float %273, %489, !dbg !33 + %618 = fmul float %274, %490, !dbg !33 + %619 = fmul float %275, %491, !dbg !33 + %620 = fmul float %276, %492, !dbg !33 + %621 = fmul float %277, %493, !dbg !33 + %622 = fmul float %278, %494, !dbg !33 + %623 = fmul float %279, %495, !dbg !33 + %624 = fmul float %280, %496, !dbg !33 + %625 = fmul float %281, %497, !dbg !33 + %626 = fmul float %282, %498, !dbg !33 + %627 = fmul float %283, %499, !dbg !33 + %628 = fmul float %284, %500, !dbg !33 + %629 = fmul float %285, %501, !dbg !33 + %630 = fmul float %286, %502, !dbg !33 + %631 = fmul float %287, %503, !dbg !33 + %632 = fmul float %288, %504, !dbg !33 + %633 = fmul float %289, %505, !dbg !33 + %634 = fmul float %290, %506, !dbg !33 + %635 = fmul float %291, %507, !dbg !33 + %636 = fmul float %292, %508, !dbg !33 + %637 = fmul float %293, %509, !dbg !33 + %638 = fmul float %294, %510, !dbg !33 + %639 = fmul float %295, %511, !dbg !33 + %640 = fmul float %296, %512, !dbg !33 + %641 = fmul float %297, %513, !dbg !33 + %642 = fmul float %298, %514, !dbg !33 + %643 = fmul float %299, %515, !dbg !33 + %644 = fmul float %300, %516, !dbg !33 + %645 = fmul float %301, %517, !dbg !33 + %646 = fmul float %302, %518, !dbg !33 + %647 = fmul float %303, %519, !dbg !33 + %648 = fmul float %304, %520, !dbg !33 + %649 = fmul float %305, %521, !dbg !33 + %650 = fmul float %306, %522, !dbg !33 + %651 = fmul float %307, %523, !dbg !33 + %652 = fmul float %308, %524, !dbg !33 + %653 = fmul float %309, %525, !dbg !33 + %654 = fmul float %310, %526, !dbg !33 + %655 = fmul float %311, %527, !dbg !33 + %656 = fmul float %312, %528, !dbg !33 + %657 = fmul float %313, %529, !dbg !33 + %658 = fmul float %314, %530, !dbg !33 + %659 = fmul float %315, %531, !dbg !33 + %660 = fmul float %316, %532, !dbg !33 + %661 = fmul float %317, %533, !dbg !33 + %662 = fadd float %598, %599, !dbg !34 + %663 = fadd float %600, %662, !dbg !34 + %664 = fadd float %601, %663, !dbg !34 + %665 = fadd float %602, %664, !dbg !34 + %666 = fadd float %603, %665, !dbg !34 + %667 = fadd float %604, %666, !dbg !34 + %668 = fadd float %605, %667, !dbg !34 + %669 = fadd float %606, %607, !dbg !34 + %670 = fadd float %608, %669, !dbg !34 + %671 = fadd float %609, %670, !dbg !34 + %672 = fadd float %610, %671, !dbg !34 + %673 = fadd float %611, %672, !dbg !34 + %674 = fadd float %612, %673, !dbg !34 + %675 = fadd float %613, %674, !dbg !34 + %676 = fadd float %614, %615, !dbg !34 + %677 = fadd float %616, %676, !dbg !34 + %678 = fadd float %617, %677, !dbg !34 + %679 = fadd float %618, %678, !dbg !34 + %680 = fadd float %619, %679, !dbg !34 + %681 = fadd float %620, %680, !dbg !34 + %682 = fadd float %621, %681, !dbg !34 + %683 = fadd float %622, %623, !dbg !34 + %684 = fadd float %624, %683, !dbg !34 + %685 = fadd float %625, %684, !dbg !34 + %686 = fadd float %626, %685, !dbg !34 + %687 = fadd float %627, %686, !dbg !34 + %688 = fadd float %628, %687, !dbg !34 + %689 = fadd float %629, %688, !dbg !34 + %690 = fadd float %630, %631, !dbg !34 + %691 = fadd float %632, %690, !dbg !34 + %692 = fadd float %633, %691, !dbg !34 + %693 = fadd float %634, %692, !dbg !34 + %694 = fadd float %635, %693, !dbg !34 + %695 = fadd float %636, %694, !dbg !34 + %696 = fadd float %637, %695, !dbg !34 + %697 = fadd float %638, %639, !dbg !34 + %698 = fadd float %640, %697, !dbg !34 + %699 = fadd float %641, %698, !dbg !34 + %700 = fadd float %642, %699, !dbg !34 + %701 = fadd float %643, %700, !dbg !34 + %702 = fadd float %644, %701, !dbg !34 + %703 = fadd float %645, %702, !dbg !34 + %704 = fadd float %646, %647, !dbg !34 + %705 = fadd float %648, %704, !dbg !34 + %706 = fadd float %649, %705, !dbg !34 + %707 = fadd float %650, %706, !dbg !34 + %708 = fadd float %651, %707, !dbg !34 + %709 = fadd float %652, %708, !dbg !34 + %710 = fadd float %653, %709, !dbg !34 + %711 = fadd float %654, %655, !dbg !34 + %712 = fadd float %656, %711, !dbg !34 + %713 = fadd float %657, %712, !dbg !34 + %714 = fadd float %658, %713, !dbg !34 + %715 = fadd float %659, %714, !dbg !34 + %716 = fadd float %660, %715, !dbg !34 + %717 = fadd float %661, %716, !dbg !34 + %718 = bitcast float %668 to i32, !dbg !37 + %719 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %718, i32 8, i32 31), !dbg !37 + %720 = bitcast i32 %719 to float, !dbg !37 + %721 = fadd float %668, %720, !dbg !34 + %722 = bitcast float %721 to i32, !dbg !37 + %723 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %722, i32 4, i32 31), !dbg !37 + %724 = bitcast i32 %723 to float, !dbg !37 + %725 = fadd float %721, %724, !dbg !34 + %726 = bitcast float %725 to i32, !dbg !37 + %727 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %726, i32 2, i32 31), !dbg !37 + %728 = bitcast i32 %727 to float, !dbg !37 + %729 = fadd float %725, %728, !dbg !34 + %730 = bitcast float %729 to i32, !dbg !37 + %731 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %730, i32 1, i32 31), !dbg !37 + %732 = bitcast i32 %731 to float, !dbg !37 + %733 = fadd float %729, %732, !dbg !34 + %734 = bitcast float %675 to i32, !dbg !37 + %735 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %734, i32 8, i32 31), !dbg !37 + %736 = bitcast i32 %735 to float, !dbg !37 + %737 = fadd float %675, %736, !dbg !34 + %738 = bitcast float %737 to i32, !dbg !37 + %739 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %738, i32 4, i32 31), !dbg !37 + %740 = bitcast i32 %739 to float, !dbg !37 + %741 = fadd float %737, %740, !dbg !34 + %742 = bitcast float %741 to i32, !dbg !37 + %743 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %742, i32 2, i32 31), !dbg !37 + %744 = bitcast i32 %743 to float, !dbg !37 + %745 = fadd float %741, %744, !dbg !34 + %746 = bitcast float %745 to i32, !dbg !37 + %747 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %746, i32 1, i32 31), !dbg !37 + %748 = bitcast i32 %747 to float, !dbg !37 + %749 = fadd float %745, %748, !dbg !34 + %750 = bitcast float %682 to i32, !dbg !37 + %751 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %750, i32 8, i32 31), !dbg !37 + %752 = bitcast i32 %751 to float, !dbg !37 + %753 = fadd float %682, %752, !dbg !34 + %754 = bitcast float %753 to i32, !dbg !37 + %755 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %754, i32 4, i32 31), !dbg !37 + %756 = bitcast i32 %755 to float, !dbg !37 + %757 = fadd float %753, %756, !dbg !34 + %758 = bitcast float %757 to i32, !dbg !37 + %759 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %758, i32 2, i32 31), !dbg !37 + %760 = bitcast i32 %759 to float, !dbg !37 + %761 = fadd float %757, %760, !dbg !34 + %762 = bitcast float %761 to i32, !dbg !37 + %763 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %762, i32 1, i32 31), !dbg !37 + %764 = bitcast i32 %763 to float, !dbg !37 + %765 = fadd float %761, %764, !dbg !34 + %766 = bitcast float %689 to i32, !dbg !37 + %767 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %766, i32 8, i32 31), !dbg !37 + %768 = bitcast i32 %767 to float, !dbg !37 + %769 = fadd float %689, %768, !dbg !34 + %770 = bitcast float %769 to i32, !dbg !37 + %771 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %770, i32 4, i32 31), !dbg !37 + %772 = bitcast i32 %771 to float, !dbg !37 + %773 = fadd float %769, %772, !dbg !34 + %774 = bitcast float %773 to i32, !dbg !37 + %775 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %774, i32 2, i32 31), !dbg !37 + %776 = bitcast i32 %775 to float, !dbg !37 + %777 = fadd float %773, %776, !dbg !34 + %778 = bitcast float %777 to i32, !dbg !37 + %779 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %778, i32 1, i32 31), !dbg !37 + %780 = bitcast i32 %779 to float, !dbg !37 + %781 = fadd float %777, %780, !dbg !34 + %782 = bitcast float %696 to i32, !dbg !37 + %783 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %782, i32 8, i32 31), !dbg !37 + %784 = bitcast i32 %783 to float, !dbg !37 + %785 = fadd float %696, %784, !dbg !34 + %786 = bitcast float %785 to i32, !dbg !37 + %787 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %786, i32 4, i32 31), !dbg !37 + %788 = bitcast i32 %787 to float, !dbg !37 + %789 = fadd float %785, %788, !dbg !34 + %790 = bitcast float %789 to i32, !dbg !37 + %791 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %790, i32 2, i32 31), !dbg !37 + %792 = bitcast i32 %791 to float, !dbg !37 + %793 = fadd float %789, %792, !dbg !34 + %794 = bitcast float %793 to i32, !dbg !37 + %795 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %794, i32 1, i32 31), !dbg !37 + %796 = bitcast i32 %795 to float, !dbg !37 + %797 = fadd float %793, %796, !dbg !34 + %798 = bitcast float %703 to i32, !dbg !37 + %799 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %798, i32 8, i32 31), !dbg !37 + %800 = bitcast i32 %799 to float, !dbg !37 + %801 = fadd float %703, %800, !dbg !34 + %802 = bitcast float %801 to i32, !dbg !37 + %803 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %802, i32 4, i32 31), !dbg !37 + %804 = bitcast i32 %803 to float, !dbg !37 + %805 = fadd float %801, %804, !dbg !34 + %806 = bitcast float %805 to i32, !dbg !37 + %807 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %806, i32 2, i32 31), !dbg !37 + %808 = bitcast i32 %807 to float, !dbg !37 + %809 = fadd float %805, %808, !dbg !34 + %810 = bitcast float %809 to i32, !dbg !37 + %811 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %810, i32 1, i32 31), !dbg !37 + %812 = bitcast i32 %811 to float, !dbg !37 + %813 = fadd float %809, %812, !dbg !34 + %814 = bitcast float %710 to i32, !dbg !37 + %815 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %814, i32 8, i32 31), !dbg !37 + %816 = bitcast i32 %815 to float, !dbg !37 + %817 = fadd float %710, %816, !dbg !34 + %818 = bitcast float %817 to i32, !dbg !37 + %819 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %818, i32 4, i32 31), !dbg !37 + %820 = bitcast i32 %819 to float, !dbg !37 + %821 = fadd float %817, %820, !dbg !34 + %822 = bitcast float %821 to i32, !dbg !37 + %823 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %822, i32 2, i32 31), !dbg !37 + %824 = bitcast i32 %823 to float, !dbg !37 + %825 = fadd float %821, %824, !dbg !34 + %826 = bitcast float %825 to i32, !dbg !37 + %827 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %826, i32 1, i32 31), !dbg !37 + %828 = bitcast i32 %827 to float, !dbg !37 + %829 = fadd float %825, %828, !dbg !34 + %830 = bitcast float %717 to i32, !dbg !37 + %831 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %830, i32 8, i32 31), !dbg !37 + %832 = bitcast i32 %831 to float, !dbg !37 + %833 = fadd float %717, %832, !dbg !34 + %834 = bitcast float %833 to i32, !dbg !37 + %835 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %834, i32 4, i32 31), !dbg !37 + %836 = bitcast i32 %835 to float, !dbg !37 + %837 = fadd float %833, %836, !dbg !34 + %838 = bitcast float %837 to i32, !dbg !37 + %839 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %838, i32 2, i32 31), !dbg !37 + %840 = bitcast i32 %839 to float, !dbg !37 + %841 = fadd float %837, %840, !dbg !34 + %842 = bitcast float %841 to i32, !dbg !37 + %843 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %842, i32 1, i32 31), !dbg !37 + %844 = bitcast i32 %843 to float, !dbg !37 + %845 = fadd float %841, %844, !dbg !34 + %846 = fmul float %733, %254, !dbg !40 + %847 = fmul float %733, %255, !dbg !40 + %848 = fmul float %733, %256, !dbg !40 + %849 = fmul float %733, %257, !dbg !40 + %850 = fmul float %733, %258, !dbg !40 + %851 = fmul float %733, %259, !dbg !40 + %852 = fmul float %733, %260, !dbg !40 + %853 = fmul float %733, %261, !dbg !40 + %854 = fmul float %749, %262, !dbg !40 + %855 = fmul float %749, %263, !dbg !40 + %856 = fmul float %749, %264, !dbg !40 + %857 = fmul float %749, %265, !dbg !40 + %858 = fmul float %749, %266, !dbg !40 + %859 = fmul float %749, %267, !dbg !40 + %860 = fmul float %749, %268, !dbg !40 + %861 = fmul float %749, %269, !dbg !40 + %862 = fmul float %765, %270, !dbg !40 + %863 = fmul float %765, %271, !dbg !40 + %864 = fmul float %765, %272, !dbg !40 + %865 = fmul float %765, %273, !dbg !40 + %866 = fmul float %765, %274, !dbg !40 + %867 = fmul float %765, %275, !dbg !40 + %868 = fmul float %765, %276, !dbg !40 + %869 = fmul float %765, %277, !dbg !40 + %870 = fmul float %781, %278, !dbg !40 + %871 = fmul float %781, %279, !dbg !40 + %872 = fmul float %781, %280, !dbg !40 + %873 = fmul float %781, %281, !dbg !40 + %874 = fmul float %781, %282, !dbg !40 + %875 = fmul float %781, %283, !dbg !40 + %876 = fmul float %781, %284, !dbg !40 + %877 = fmul float %781, %285, !dbg !40 + %878 = fmul float %797, %286, !dbg !40 + %879 = fmul float %797, %287, !dbg !40 + %880 = fmul float %797, %288, !dbg !40 + %881 = fmul float %797, %289, !dbg !40 + %882 = fmul float %797, %290, !dbg !40 + %883 = fmul float %797, %291, !dbg !40 + %884 = fmul float %797, %292, !dbg !40 + %885 = fmul float %797, %293, !dbg !40 + %886 = fmul float %813, %294, !dbg !40 + %887 = fmul float %813, %295, !dbg !40 + %888 = fmul float %813, %296, !dbg !40 + %889 = fmul float %813, %297, !dbg !40 + %890 = fmul float %813, %298, !dbg !40 + %891 = fmul float %813, %299, !dbg !40 + %892 = fmul float %813, %300, !dbg !40 + %893 = fmul float %813, %301, !dbg !40 + %894 = fmul float %829, %302, !dbg !40 + %895 = fmul float %829, %303, !dbg !40 + %896 = fmul float %829, %304, !dbg !40 + %897 = fmul float %829, %305, !dbg !40 + %898 = fmul float %829, %306, !dbg !40 + %899 = fmul float %829, %307, !dbg !40 + %900 = fmul float %829, %308, !dbg !40 + %901 = fmul float %829, %309, !dbg !40 + %902 = fmul float %845, %310, !dbg !40 + %903 = fmul float %845, %311, !dbg !40 + %904 = fmul float %845, %312, !dbg !40 + %905 = fmul float %845, %313, !dbg !40 + %906 = fmul float %845, %314, !dbg !40 + %907 = fmul float %845, %315, !dbg !40 + %908 = fmul float %845, %316, !dbg !40 + %909 = fmul float %845, %317, !dbg !40 + %910 = fmul float %846, %319, !dbg !41 + %911 = fmul float %847, %319, !dbg !41 + %912 = fmul float %848, %319, !dbg !41 + %913 = fmul float %849, %319, !dbg !41 + %914 = fmul float %850, %319, !dbg !41 + %915 = fmul float %851, %319, !dbg !41 + %916 = fmul float %852, %319, !dbg !41 + %917 = fmul float %853, %319, !dbg !41 + %918 = fmul float %854, %321, !dbg !41 + %919 = fmul float %855, %321, !dbg !41 + %920 = fmul float %856, %321, !dbg !41 + %921 = fmul float %857, %321, !dbg !41 + %922 = fmul float %858, %321, !dbg !41 + %923 = fmul float %859, %321, !dbg !41 + %924 = fmul float %860, %321, !dbg !41 + %925 = fmul float %861, %321, !dbg !41 + %926 = fmul float %862, %323, !dbg !41 + %927 = fmul float %863, %323, !dbg !41 + %928 = fmul float %864, %323, !dbg !41 + %929 = fmul float %865, %323, !dbg !41 + %930 = fmul float %866, %323, !dbg !41 + %931 = fmul float %867, %323, !dbg !41 + %932 = fmul float %868, %323, !dbg !41 + %933 = fmul float %869, %323, !dbg !41 + %934 = fmul float %870, %325, !dbg !41 + %935 = fmul float %871, %325, !dbg !41 + %936 = fmul float %872, %325, !dbg !41 + %937 = fmul float %873, %325, !dbg !41 + %938 = fmul float %874, %325, !dbg !41 + %939 = fmul float %875, %325, !dbg !41 + %940 = fmul float %876, %325, !dbg !41 + %941 = fmul float %877, %325, !dbg !41 + %942 = fmul float %878, %327, !dbg !41 + %943 = fmul float %879, %327, !dbg !41 + %944 = fmul float %880, %327, !dbg !41 + %945 = fmul float %881, %327, !dbg !41 + %946 = fmul float %882, %327, !dbg !41 + %947 = fmul float %883, %327, !dbg !41 + %948 = fmul float %884, %327, !dbg !41 + %949 = fmul float %885, %327, !dbg !41 + %950 = fmul float %886, %329, !dbg !41 + %951 = fmul float %887, %329, !dbg !41 + %952 = fmul float %888, %329, !dbg !41 + %953 = fmul float %889, %329, !dbg !41 + %954 = fmul float %890, %329, !dbg !41 + %955 = fmul float %891, %329, !dbg !41 + %956 = fmul float %892, %329, !dbg !41 + %957 = fmul float %893, %329, !dbg !41 + %958 = fmul float %894, %331, !dbg !41 + %959 = fmul float %895, %331, !dbg !41 + %960 = fmul float %896, %331, !dbg !41 + %961 = fmul float %897, %331, !dbg !41 + %962 = fmul float %898, %331, !dbg !41 + %963 = fmul float %899, %331, !dbg !41 + %964 = fmul float %900, %331, !dbg !41 + %965 = fmul float %901, %331, !dbg !41 + %966 = fmul float %902, %333, !dbg !41 + %967 = fmul float %903, %333, !dbg !41 + %968 = fmul float %904, %333, !dbg !41 + %969 = fmul float %905, %333, !dbg !41 + %970 = fmul float %906, %333, !dbg !41 + %971 = fmul float %907, %333, !dbg !41 + %972 = fmul float %908, %333, !dbg !41 + %973 = fmul float %909, %333, !dbg !41 + %974 = fsub float %534, %910, !dbg !42 + %975 = fsub float %535, %911, !dbg !42 + %976 = fsub float %536, %912, !dbg !42 + %977 = fsub float %537, %913, !dbg !42 + %978 = fsub float %538, %914, !dbg !42 + %979 = fsub float %539, %915, !dbg !42 + %980 = fsub float %540, %916, !dbg !42 + %981 = fsub float %541, %917, !dbg !42 + %982 = fsub float %542, %918, !dbg !42 + %983 = fsub float %543, %919, !dbg !42 + %984 = fsub float %544, %920, !dbg !42 + %985 = fsub float %545, %921, !dbg !42 + %986 = fsub float %546, %922, !dbg !42 + %987 = fsub float %547, %923, !dbg !42 + %988 = fsub float %548, %924, !dbg !42 + %989 = fsub float %549, %925, !dbg !42 + %990 = fsub float %550, %926, !dbg !42 + %991 = fsub float %551, %927, !dbg !42 + %992 = fsub float %552, %928, !dbg !42 + %993 = fsub float %553, %929, !dbg !42 + %994 = fsub float %554, %930, !dbg !42 + %995 = fsub float %555, %931, !dbg !42 + %996 = fsub float %556, %932, !dbg !42 + %997 = fsub float %557, %933, !dbg !42 + %998 = fsub float %558, %934, !dbg !42 + %999 = fsub float %559, %935, !dbg !42 + %1000 = fsub float %560, %936, !dbg !42 + %1001 = fsub float %561, %937, !dbg !42 + %1002 = fsub float %562, %938, !dbg !42 + %1003 = fsub float %563, %939, !dbg !42 + %1004 = fsub float %564, %940, !dbg !42 + %1005 = fsub float %565, %941, !dbg !42 + %1006 = fsub float %566, %942, !dbg !42 + %1007 = fsub float %567, %943, !dbg !42 + %1008 = fsub float %568, %944, !dbg !42 + %1009 = fsub float %569, %945, !dbg !42 + %1010 = fsub float %570, %946, !dbg !42 + %1011 = fsub float %571, %947, !dbg !42 + %1012 = fsub float %572, %948, !dbg !42 + %1013 = fsub float %573, %949, !dbg !42 + %1014 = fsub float %574, %950, !dbg !42 + %1015 = fsub float %575, %951, !dbg !42 + %1016 = fsub float %576, %952, !dbg !42 + %1017 = fsub float %577, %953, !dbg !42 + %1018 = fsub float %578, %954, !dbg !42 + %1019 = fsub float %579, %955, !dbg !42 + %1020 = fsub float %580, %956, !dbg !42 + %1021 = fsub float %581, %957, !dbg !42 + %1022 = insertelement <2 x float> poison, float %582, i64 0, !dbg !42 + %1023 = insertelement <2 x float> %1022, float %583, i64 1, !dbg !42 + %1024 = insertelement <2 x float> poison, float %958, i64 0, !dbg !42 + %1025 = insertelement <2 x float> %1024, float %959, i64 1, !dbg !42 + %1026 = fsub <2 x float> %1023, %1025, !dbg !42 + %1027 = insertelement <2 x float> poison, float %584, i64 0, !dbg !42 + %1028 = insertelement <2 x float> %1027, float %585, i64 1, !dbg !42 + %1029 = insertelement <2 x float> poison, float %960, i64 0, !dbg !42 + %1030 = insertelement <2 x float> %1029, float %961, i64 1, !dbg !42 + %1031 = fsub <2 x float> %1028, %1030, !dbg !42 + %1032 = insertelement <2 x float> poison, float %586, i64 0, !dbg !42 + %1033 = insertelement <2 x float> %1032, float %587, i64 1, !dbg !42 + %1034 = insertelement <2 x float> poison, float %962, i64 0, !dbg !42 + %1035 = insertelement <2 x float> %1034, float %963, i64 1, !dbg !42 + %1036 = fsub <2 x float> %1033, %1035, !dbg !42 + %1037 = insertelement <2 x float> poison, float %588, i64 0, !dbg !42 + %1038 = insertelement <2 x float> %1037, float %589, i64 1, !dbg !42 + %1039 = insertelement <2 x float> poison, float %964, i64 0, !dbg !42 + %1040 = insertelement <2 x float> %1039, float %965, i64 1, !dbg !42 + %1041 = fsub <2 x float> %1038, %1040, !dbg !42 + %1042 = insertelement <2 x float> poison, float %590, i64 0, !dbg !42 + %1043 = insertelement <2 x float> %1042, float %591, i64 1, !dbg !42 + %1044 = insertelement <2 x float> poison, float %966, i64 0, !dbg !42 + %1045 = insertelement <2 x float> %1044, float %967, i64 1, !dbg !42 + %1046 = fsub <2 x float> %1043, %1045, !dbg !42 + %1047 = insertelement <2 x float> poison, float %592, i64 0, !dbg !42 + %1048 = insertelement <2 x float> %1047, float %593, i64 1, !dbg !42 + %1049 = insertelement <2 x float> poison, float %968, i64 0, !dbg !42 + %1050 = insertelement <2 x float> %1049, float %969, i64 1, !dbg !42 + %1051 = fsub <2 x float> %1048, %1050, !dbg !42 + %1052 = insertelement <2 x float> poison, float %594, i64 0, !dbg !42 + %1053 = insertelement <2 x float> %1052, float %595, i64 1, !dbg !42 + %1054 = insertelement <2 x float> poison, float %970, i64 0, !dbg !42 + %1055 = insertelement <2 x float> %1054, float %971, i64 1, !dbg !42 + %1056 = fsub <2 x float> %1053, %1055, !dbg !42 + %1057 = insertelement <2 x float> poison, float %596, i64 0, !dbg !42 + %1058 = insertelement <2 x float> %1057, float %597, i64 1, !dbg !42 + %1059 = insertelement <2 x float> poison, float %972, i64 0, !dbg !42 + %1060 = insertelement <2 x float> %1059, float %973, i64 1, !dbg !42 + %1061 = fsub <2 x float> %1058, %1060, !dbg !42 + %1062 = fptrunc float %974 to bfloat, !dbg !43 + %1063 = fptrunc float %975 to bfloat, !dbg !43 + %1064 = fptrunc float %976 to bfloat, !dbg !43 + %1065 = fptrunc float %977 to bfloat, !dbg !43 + %1066 = fptrunc float %978 to bfloat, !dbg !43 + %1067 = fptrunc float %979 to bfloat, !dbg !43 + %1068 = fptrunc float %980 to bfloat, !dbg !43 + %1069 = fptrunc float %981 to bfloat, !dbg !43 + %1070 = fptrunc float %982 to bfloat, !dbg !43 + %1071 = fptrunc float %983 to bfloat, !dbg !43 + %1072 = fptrunc float %984 to bfloat, !dbg !43 + %1073 = fptrunc float %985 to bfloat, !dbg !43 + %1074 = fptrunc float %986 to bfloat, !dbg !43 + %1075 = fptrunc float %987 to bfloat, !dbg !43 + %1076 = fptrunc float %988 to bfloat, !dbg !43 + %1077 = fptrunc float %989 to bfloat, !dbg !43 + %1078 = fptrunc float %990 to bfloat, !dbg !43 + %1079 = fptrunc float %991 to bfloat, !dbg !43 + %1080 = fptrunc float %992 to bfloat, !dbg !43 + %1081 = fptrunc float %993 to bfloat, !dbg !43 + %1082 = fptrunc float %994 to bfloat, !dbg !43 + %1083 = fptrunc float %995 to bfloat, !dbg !43 + %1084 = fptrunc float %996 to bfloat, !dbg !43 + %1085 = fptrunc float %997 to bfloat, !dbg !43 + %1086 = fptrunc float %998 to bfloat, !dbg !43 + %1087 = fptrunc float %999 to bfloat, !dbg !43 + %1088 = fptrunc float %1000 to bfloat, !dbg !43 + %1089 = fptrunc float %1001 to bfloat, !dbg !43 + %1090 = fptrunc float %1002 to bfloat, !dbg !43 + %1091 = fptrunc float %1003 to bfloat, !dbg !43 + %1092 = fptrunc float %1004 to bfloat, !dbg !43 + %1093 = fptrunc float %1005 to bfloat, !dbg !43 + %1094 = fptrunc float %1006 to bfloat, !dbg !43 + %1095 = fptrunc float %1007 to bfloat, !dbg !43 + %1096 = fptrunc float %1008 to bfloat, !dbg !43 + %1097 = fptrunc float %1009 to bfloat, !dbg !43 + %1098 = fptrunc float %1010 to bfloat, !dbg !43 + %1099 = fptrunc float %1011 to bfloat, !dbg !43 + %1100 = fptrunc float %1012 to bfloat, !dbg !43 + %1101 = fptrunc float %1013 to bfloat, !dbg !43 + %1102 = fptrunc float %1014 to bfloat, !dbg !43 + %1103 = fptrunc float %1015 to bfloat, !dbg !43 + %1104 = fptrunc float %1016 to bfloat, !dbg !43 + %1105 = fptrunc float %1017 to bfloat, !dbg !43 + %1106 = fptrunc float %1018 to bfloat, !dbg !43 + %1107 = fptrunc float %1019 to bfloat, !dbg !43 + %1108 = fptrunc float %1020 to bfloat, !dbg !43 + %1109 = fptrunc float %1021 to bfloat, !dbg !43 + %1110 = fptrunc <2 x float> %1026 to <2 x bfloat>, !dbg !43 + %1111 = fptrunc <2 x float> %1031 to <2 x bfloat>, !dbg !43 + %1112 = fptrunc <2 x float> %1036 to <2 x bfloat>, !dbg !43 + %1113 = fptrunc <2 x float> %1041 to <2 x bfloat>, !dbg !43 + %1114 = fptrunc <2 x float> %1046 to <2 x bfloat>, !dbg !43 + %1115 = fptrunc <2 x float> %1051 to <2 x bfloat>, !dbg !43 + %1116 = fptrunc <2 x float> %1056 to <2 x bfloat>, !dbg !43 + %1117 = fptrunc <2 x float> %1061 to <2 x bfloat>, !dbg !43 + %1118 = insertelement <2 x bfloat> poison, bfloat %1062, i64 0, !dbg !44 + %1119 = insertelement <2 x bfloat> %1118, bfloat %1063, i64 1, !dbg !44 + %1120 = bitcast <2 x bfloat> %1119 to i32, !dbg !44 + %1121 = insertelement <2 x bfloat> poison, bfloat %1064, i64 0, !dbg !44 + %1122 = insertelement <2 x bfloat> %1121, bfloat %1065, i64 1, !dbg !44 + %1123 = bitcast <2 x bfloat> %1122 to i32, !dbg !44 + %1124 = insertelement <2 x bfloat> poison, bfloat %1066, i64 0, !dbg !44 + %1125 = insertelement <2 x bfloat> %1124, bfloat %1067, i64 1, !dbg !44 + %1126 = bitcast <2 x bfloat> %1125 to i32, !dbg !44 + %1127 = insertelement <2 x bfloat> poison, bfloat %1068, i64 0, !dbg !44 + %1128 = insertelement <2 x bfloat> %1127, bfloat %1069, i64 1, !dbg !44 + %1129 = bitcast <2 x bfloat> %1128 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1120, i32 %1123, i32 %1126, i32 %1129, ptr addrspace(1) %102, i1 %110) #3, !dbg !44 + %1130 = insertelement <2 x bfloat> poison, bfloat %1070, i64 0, !dbg !44 + %1131 = insertelement <2 x bfloat> %1130, bfloat %1071, i64 1, !dbg !44 + %1132 = bitcast <2 x bfloat> %1131 to i32, !dbg !44 + %1133 = insertelement <2 x bfloat> poison, bfloat %1072, i64 0, !dbg !44 + %1134 = insertelement <2 x bfloat> %1133, bfloat %1073, i64 1, !dbg !44 + %1135 = bitcast <2 x bfloat> %1134 to i32, !dbg !44 + %1136 = insertelement <2 x bfloat> poison, bfloat %1074, i64 0, !dbg !44 + %1137 = insertelement <2 x bfloat> %1136, bfloat %1075, i64 1, !dbg !44 + %1138 = bitcast <2 x bfloat> %1137 to i32, !dbg !44 + %1139 = insertelement <2 x bfloat> poison, bfloat %1076, i64 0, !dbg !44 + %1140 = insertelement <2 x bfloat> %1139, bfloat %1077, i64 1, !dbg !44 + %1141 = bitcast <2 x bfloat> %1140 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1132, i32 %1135, i32 %1138, i32 %1141, ptr addrspace(1) %103, i1 %128) #3, !dbg !44 + %1142 = insertelement <2 x bfloat> poison, bfloat %1078, i64 0, !dbg !44 + %1143 = insertelement <2 x bfloat> %1142, bfloat %1079, i64 1, !dbg !44 + %1144 = bitcast <2 x bfloat> %1143 to i32, !dbg !44 + %1145 = insertelement <2 x bfloat> poison, bfloat %1080, i64 0, !dbg !44 + %1146 = insertelement <2 x bfloat> %1145, bfloat %1081, i64 1, !dbg !44 + %1147 = bitcast <2 x bfloat> %1146 to i32, !dbg !44 + %1148 = insertelement <2 x bfloat> poison, bfloat %1082, i64 0, !dbg !44 + %1149 = insertelement <2 x bfloat> %1148, bfloat %1083, i64 1, !dbg !44 + %1150 = bitcast <2 x bfloat> %1149 to i32, !dbg !44 + %1151 = insertelement <2 x bfloat> poison, bfloat %1084, i64 0, !dbg !44 + %1152 = insertelement <2 x bfloat> %1151, bfloat %1085, i64 1, !dbg !44 + %1153 = bitcast <2 x bfloat> %1152 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1144, i32 %1147, i32 %1150, i32 %1153, ptr addrspace(1) %104, i1 %146) #3, !dbg !44 + %1154 = insertelement <2 x bfloat> poison, bfloat %1086, i64 0, !dbg !44 + %1155 = insertelement <2 x bfloat> %1154, bfloat %1087, i64 1, !dbg !44 + %1156 = bitcast <2 x bfloat> %1155 to i32, !dbg !44 + %1157 = insertelement <2 x bfloat> poison, bfloat %1088, i64 0, !dbg !44 + %1158 = insertelement <2 x bfloat> %1157, bfloat %1089, i64 1, !dbg !44 + %1159 = bitcast <2 x bfloat> %1158 to i32, !dbg !44 + %1160 = insertelement <2 x bfloat> poison, bfloat %1090, i64 0, !dbg !44 + %1161 = insertelement <2 x bfloat> %1160, bfloat %1091, i64 1, !dbg !44 + %1162 = bitcast <2 x bfloat> %1161 to i32, !dbg !44 + %1163 = insertelement <2 x bfloat> poison, bfloat %1092, i64 0, !dbg !44 + %1164 = insertelement <2 x bfloat> %1163, bfloat %1093, i64 1, !dbg !44 + %1165 = bitcast <2 x bfloat> %1164 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1156, i32 %1159, i32 %1162, i32 %1165, ptr addrspace(1) %105, i1 %164) #3, !dbg !44 + %1166 = insertelement <2 x bfloat> poison, bfloat %1094, i64 0, !dbg !44 + %1167 = insertelement <2 x bfloat> %1166, bfloat %1095, i64 1, !dbg !44 + %1168 = bitcast <2 x bfloat> %1167 to i32, !dbg !44 + %1169 = insertelement <2 x bfloat> poison, bfloat %1096, i64 0, !dbg !44 + %1170 = insertelement <2 x bfloat> %1169, bfloat %1097, i64 1, !dbg !44 + %1171 = bitcast <2 x bfloat> %1170 to i32, !dbg !44 + %1172 = insertelement <2 x bfloat> poison, bfloat %1098, i64 0, !dbg !44 + %1173 = insertelement <2 x bfloat> %1172, bfloat %1099, i64 1, !dbg !44 + %1174 = bitcast <2 x bfloat> %1173 to i32, !dbg !44 + %1175 = insertelement <2 x bfloat> poison, bfloat %1100, i64 0, !dbg !44 + %1176 = insertelement <2 x bfloat> %1175, bfloat %1101, i64 1, !dbg !44 + %1177 = bitcast <2 x bfloat> %1176 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1168, i32 %1171, i32 %1174, i32 %1177, ptr addrspace(1) %106, i1 %182) #3, !dbg !44 + %1178 = insertelement <2 x bfloat> poison, bfloat %1102, i64 0, !dbg !44 + %1179 = insertelement <2 x bfloat> %1178, bfloat %1103, i64 1, !dbg !44 + %1180 = bitcast <2 x bfloat> %1179 to i32, !dbg !44 + %1181 = insertelement <2 x bfloat> poison, bfloat %1104, i64 0, !dbg !44 + %1182 = insertelement <2 x bfloat> %1181, bfloat %1105, i64 1, !dbg !44 + %1183 = bitcast <2 x bfloat> %1182 to i32, !dbg !44 + %1184 = insertelement <2 x bfloat> poison, bfloat %1106, i64 0, !dbg !44 + %1185 = insertelement <2 x bfloat> %1184, bfloat %1107, i64 1, !dbg !44 + %1186 = bitcast <2 x bfloat> %1185 to i32, !dbg !44 + %1187 = insertelement <2 x bfloat> poison, bfloat %1108, i64 0, !dbg !44 + %1188 = insertelement <2 x bfloat> %1187, bfloat %1109, i64 1, !dbg !44 + %1189 = bitcast <2 x bfloat> %1188 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1180, i32 %1183, i32 %1186, i32 %1189, ptr addrspace(1) %107, i1 %200) #3, !dbg !44 + %1190 = bitcast <2 x bfloat> %1110 to i32, !dbg !44 + %1191 = bitcast <2 x bfloat> %1111 to i32, !dbg !44 + %1192 = bitcast <2 x bfloat> %1112 to i32, !dbg !44 + %1193 = bitcast <2 x bfloat> %1113 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1190, i32 %1191, i32 %1192, i32 %1193, ptr addrspace(1) %108, i1 %218) #3, !dbg !44 + %1194 = bitcast <2 x bfloat> %1114 to i32, !dbg !44 + %1195 = bitcast <2 x bfloat> %1115 to i32, !dbg !44 + %1196 = bitcast <2 x bfloat> %1116 to i32, !dbg !44 + %1197 = bitcast <2 x bfloat> %1117 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1194, i32 %1195, i32 %1196, i32 %1197, ptr addrspace(1) %109, i1 %236) #3, !dbg !44 + ret void, !dbg !45 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +attributes #0 = { nounwind "nvvm.reqntid"="256" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = distinct !DISubprogram(name: "l2norm_bwd_kernel", linkageName: "l2norm_bwd_kernel", scope: !1, file: !1, line: 93, type: !5, scopeLine: 93, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!5 = !DISubroutineType(cc: DW_CC_normal, types: !6) +!6 = !{null, !7, !9, !7, !7, !10, !11, !7, !7} +!7 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !8, size: 64, dwarfAddressSpace: 1) +!8 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!9 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !10, size: 64, dwarfAddressSpace: 1) +!10 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!11 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!12 = !DILocation(line: 94, column: 24, scope: !4) +!13 = !DILocation(line: 94, column: 30, scope: !4) +!14 = !DILocation(line: 95, column: 16, scope: !4) +!15 = !DILocation(line: 95, column: 34, scope: !4) +!16 = !DILocation(line: 95, column: 21, scope: !4) +!17 = !DILocation(line: 97, column: 16, scope: !4) +!18 = !DILocation(line: 98, column: 30, scope: !4) +!19 = !DILocation(line: 99, column: 29, scope: !4) +!20 = !DILocation(line: 99, column: 14, scope: !4) +!21 = !DILocation(line: 99, column: 33, scope: !4) +!22 = !DILocation(line: 100, column: 20, scope: !4) +!23 = !DILocation(line: 101, column: 16, scope: !4) +!24 = !DILocation(line: 101, column: 35, scope: !4) +!25 = !DILocation(line: 102, column: 16, scope: !4) +!26 = !DILocation(line: 102, column: 35, scope: !4) +!27 = !DILocation(line: 104, column: 18, scope: !4) +!28 = !DILocation(line: 104, column: 47, scope: !4) +!29 = !DILocation(line: 105, column: 21, scope: !4) +!30 = !DILocation(line: 106, column: 19, scope: !4) +!31 = !DILocation(line: 106, column: 49, scope: !4) +!32 = !DILocation(line: 107, column: 18, scope: !4) +!33 = !DILocation(line: 107, column: 50, scope: !4) +!34 = !DILocation(line: 263, column: 15, scope: !35, inlinedAt: !37) +!35 = distinct !DILexicalBlockFile(scope: !4, file: !36, discriminator: 0) +!36 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!37 = !DILocation(line: 293, column: 36, scope: !35, inlinedAt: !38) +!38 = !DILocation(line: 107, column: 55, scope: !39) +!39 = distinct !DILexicalBlockFile(scope: !4, file: !1, discriminator: 0) +!40 = !DILocation(line: 107, column: 69, scope: !4) +!41 = !DILocation(line: 107, column: 75, scope: !4) +!42 = !DILocation(line: 107, column: 36, scope: !4) +!43 = !DILocation(line: 108, column: 27, scope: !4) +!44 = !DILocation(line: 108, column: 19, scope: !4) +!45 = !DILocation(line: 108, column: 4, scope: !4) diff --git a/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/l2norm_bwd_kernel.ptx b/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/l2norm_bwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..4c3848dc801d6b9b5bb800ef00d70ae1b18234b4 --- /dev/null +++ b/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/l2norm_bwd_kernel.ptx @@ -0,0 +1,1322 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_bwd_kernel // -- Begin function l2norm_bwd_kernel + // @l2norm_bwd_kernel +.visible .entry l2norm_bwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_2, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_3, + .param .f32 l2norm_bwd_kernel_param_4, + .param .u32 l2norm_bwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_6, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_7 +) +.reqntid 256 +{ + .reg .pred %p<9>; + .reg .b16 %rs<129>; + .reg .b32 %r<631>; + .reg .b64 %rd<120>; + .loc 1 93 0 // l2norm.py:93:0 +$L__func_begin0: + .loc 1 93 0 // l2norm.py:93:0 + +// %bb.0: + ld.param.b64 %rd33, [l2norm_bwd_kernel_param_0]; + ld.param.b64 %rd34, [l2norm_bwd_kernel_param_1]; +$L__tmp0: + .loc 1 94 24 // l2norm.py:94:24 + mov.u32 %r106, %ctaid.x; + ld.param.b64 %rd35, [l2norm_bwd_kernel_param_2]; + .loc 1 95 16 // l2norm.py:95:16 + mul.wide.u32 %rd36, %r106, 128; + ld.param.b64 %rd37, [l2norm_bwd_kernel_param_3]; + .loc 1 95 34 // l2norm.py:95:34 + mov.u32 %r107, %tid.x; + shr.u32 %r108, %r107, 4; + ld.param.s32 %rd38, [l2norm_bwd_kernel_param_5]; + bfe.u32 %r109, %r107, 4, 4; + or.b32 %r110, %r109, 16; + or.b32 %r111, %r109, 32; + or.b32 %r112, %r108, 48; + .loc 1 95 21 // l2norm.py:95:21 + or.b32 %r113, %r109, 64; + or.b32 %r114, %r109, 80; + or.b32 %r115, %r109, 96; + or.b32 %r116, %r108, 112; + cvt.u64.u32 %rd39, %r116; + cvt.u64.u32 %rd40, %r115; + cvt.u64.u32 %rd41, %r114; + cvt.u64.u32 %rd42, %r113; + cvt.u64.u32 %rd43, %r112; + cvt.u64.u32 %rd44, %r111; + cvt.u64.u32 %rd45, %r110; + cvt.u64.u32 %rd46, %r109; + or.b64 %rd47, %rd36, %rd46; + or.b64 %rd48, %rd36, %rd45; + or.b64 %rd49, %rd36, %rd44; + or.b64 %rd50, %rd36, %rd43; + or.b64 %rd51, %rd36, %rd42; + or.b64 %rd52, %rd36, %rd41; + or.b64 %rd53, %rd36, %rd40; + or.b64 %rd54, %rd36, %rd39; + .loc 1 97 16 // l2norm.py:97:16 + setp.lt.s64 %p8, %rd54, %rd38; + setp.lt.s64 %p7, %rd53, %rd38; + setp.lt.s64 %p6, %rd52, %rd38; + setp.lt.s64 %p5, %rd51, %rd38; + setp.lt.s64 %p4, %rd50, %rd38; + setp.lt.s64 %p3, %rd49, %rd38; + setp.lt.s64 %p2, %rd48, %rd38; + setp.lt.s64 %p1, %rd47, %rd38; + .loc 1 98 30 // l2norm.py:98:30 + shl.b32 %r117, %r107, 3; + and.b32 %r118, %r117, 120; + .loc 1 99 14 // l2norm.py:99:14 + shl.b64 %rd55, %rd47, 8; + add.s64 %rd56, %rd33, %rd55; + shl.b64 %rd57, %rd48, 8; + add.s64 %rd58, %rd33, %rd57; + shl.b64 %rd59, %rd49, 8; + add.s64 %rd60, %rd33, %rd59; + shl.b64 %rd61, %rd50, 8; + add.s64 %rd62, %rd33, %rd61; + shl.b64 %rd63, %rd51, 8; + add.s64 %rd64, %rd33, %rd63; + shl.b64 %rd65, %rd52, 8; + add.s64 %rd66, %rd33, %rd65; + shl.b64 %rd67, %rd53, 8; + add.s64 %rd68, %rd33, %rd67; + shl.b64 %rd69, %rd54, 8; + add.s64 %rd70, %rd33, %rd69; + .loc 1 99 33 // l2norm.py:99:33 + mul.wide.u32 %rd71, %r118, 2; + add.s64 %rd1, %rd56, %rd71; + add.s64 %rd2, %rd58, %rd71; + add.s64 %rd3, %rd60, %rd71; + add.s64 %rd4, %rd62, %rd71; + add.s64 %rd5, %rd64, %rd71; + add.s64 %rd6, %rd66, %rd71; + add.s64 %rd7, %rd68, %rd71; + add.s64 %rd8, %rd70, %rd71; + .loc 1 100 20 // l2norm.py:100:20 + shl.b64 %rd72, %rd47, 2; + add.s64 %rd9, %rd34, %rd72; + shl.b64 %rd73, %rd48, 2; + add.s64 %rd10, %rd34, %rd73; + shl.b64 %rd74, %rd49, 2; + add.s64 %rd11, %rd34, %rd74; + shl.b64 %rd75, %rd50, 2; + add.s64 %rd12, %rd34, %rd75; + shl.b64 %rd76, %rd51, 2; + add.s64 %rd13, %rd34, %rd76; + shl.b64 %rd77, %rd52, 2; + add.s64 %rd14, %rd34, %rd77; + shl.b64 %rd78, %rd53, 2; + add.s64 %rd15, %rd34, %rd78; + shl.b64 %rd79, %rd54, 2; + add.s64 %rd16, %rd34, %rd79; + .loc 1 101 16 // l2norm.py:101:16 + add.s64 %rd80, %rd35, %rd55; + add.s64 %rd81, %rd35, %rd57; + add.s64 %rd82, %rd35, %rd59; + add.s64 %rd83, %rd35, %rd61; + add.s64 %rd84, %rd35, %rd63; + add.s64 %rd85, %rd35, %rd65; + add.s64 %rd86, %rd35, %rd67; + add.s64 %rd87, %rd35, %rd69; + .loc 1 101 35 // l2norm.py:101:35 + add.s64 %rd17, %rd80, %rd71; + add.s64 %rd18, %rd81, %rd71; + add.s64 %rd19, %rd82, %rd71; + add.s64 %rd20, %rd83, %rd71; + add.s64 %rd21, %rd84, %rd71; + add.s64 %rd22, %rd85, %rd71; + add.s64 %rd23, %rd86, %rd71; + add.s64 %rd24, %rd87, %rd71; + .loc 1 102 16 // l2norm.py:102:16 + add.s64 %rd88, %rd37, %rd55; + add.s64 %rd89, %rd37, %rd57; + add.s64 %rd90, %rd37, %rd59; + add.s64 %rd91, %rd37, %rd61; + add.s64 %rd92, %rd37, %rd63; + add.s64 %rd93, %rd37, %rd65; + add.s64 %rd94, %rd37, %rd67; + add.s64 %rd95, %rd37, %rd69; + .loc 1 102 35 // l2norm.py:102:35 + add.s64 %rd25, %rd88, %rd71; + add.s64 %rd26, %rd89, %rd71; + add.s64 %rd27, %rd90, %rd71; + add.s64 %rd28, %rd91, %rd71; + add.s64 %rd29, %rd92, %rd71; + add.s64 %rd30, %rd93, %rd71; + add.s64 %rd31, %rd94, %rd71; + add.s64 %rd32, %rd95, %rd71; + mov.b32 %r5, 0; + .loc 1 104 18 // l2norm.py:104:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + mov.b32 {%rs5, %rs6}, %r3; + mov.b32 {%rs7, %rs8}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs9, %rs10}, %r6; + mov.b32 {%rs11, %rs12}, %r7; + mov.b32 {%rs13, %rs14}, %r8; + mov.b32 {%rs15, %rs16}, %r9; + // begin inline asm + mov.u32 %r10, %r5; + mov.u32 %r11, %r5; + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + @%p3 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs17, %rs18}, %r10; + mov.b32 {%rs19, %rs20}, %r11; + mov.b32 {%rs21, %rs22}, %r12; + mov.b32 {%rs23, %rs24}, %r13; + // begin inline asm + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + @%p4 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ]; + // end inline asm + mov.b32 {%rs25, %rs26}, %r14; + mov.b32 {%rs27, %rs28}, %r15; + mov.b32 {%rs29, %rs30}, %r16; + mov.b32 {%rs31, %rs32}, %r17; + // begin inline asm + mov.u32 %r18, %r5; + mov.u32 %r19, %r5; + mov.u32 %r20, %r5; + mov.u32 %r21, %r5; + @%p5 ld.global.v4.b32 { %r18, %r19, %r20, %r21 }, [ %rd5 + 0 ]; + // end inline asm + mov.b32 {%rs33, %rs34}, %r18; + mov.b32 {%rs35, %rs36}, %r19; + mov.b32 {%rs37, %rs38}, %r20; + mov.b32 {%rs39, %rs40}, %r21; + // begin inline asm + mov.u32 %r22, %r5; + mov.u32 %r23, %r5; + mov.u32 %r24, %r5; + mov.u32 %r25, %r5; + @%p6 ld.global.v4.b32 { %r22, %r23, %r24, %r25 }, [ %rd6 + 0 ]; + // end inline asm + mov.b32 {%rs41, %rs42}, %r22; + mov.b32 {%rs43, %rs44}, %r23; + mov.b32 {%rs45, %rs46}, %r24; + mov.b32 {%rs47, %rs48}, %r25; + // begin inline asm + mov.u32 %r26, %r5; + mov.u32 %r27, %r5; + mov.u32 %r28, %r5; + mov.u32 %r29, %r5; + @%p7 ld.global.v4.b32 { %r26, %r27, %r28, %r29 }, [ %rd7 + 0 ]; + // end inline asm + mov.b32 {%rs49, %rs50}, %r26; + mov.b32 {%rs51, %rs52}, %r27; + mov.b32 {%rs53, %rs54}, %r28; + mov.b32 {%rs55, %rs56}, %r29; + // begin inline asm + mov.u32 %r30, %r5; + mov.u32 %r31, %r5; + mov.u32 %r32, %r5; + mov.u32 %r33, %r5; + @%p8 ld.global.v4.b32 { %r30, %r31, %r32, %r33 }, [ %rd8 + 0 ]; + // end inline asm + mov.b32 {%rs57, %rs58}, %r30; + mov.b32 {%rs59, %rs60}, %r31; + mov.b32 {%rs61, %rs62}, %r32; + mov.b32 {%rs63, %rs64}, %r33; + .loc 1 104 47 // l2norm.py:104:47 + cvt.f32.bf16 %r119, %rs1; + cvt.f32.bf16 %r120, %rs2; + cvt.f32.bf16 %r121, %rs3; + cvt.f32.bf16 %r122, %rs4; + cvt.f32.bf16 %r123, %rs5; + cvt.f32.bf16 %r124, %rs6; + cvt.f32.bf16 %r125, %rs7; + cvt.f32.bf16 %r126, %rs8; + cvt.f32.bf16 %r127, %rs9; + cvt.f32.bf16 %r128, %rs10; + cvt.f32.bf16 %r129, %rs11; + cvt.f32.bf16 %r130, %rs12; + cvt.f32.bf16 %r131, %rs13; + cvt.f32.bf16 %r132, %rs14; + cvt.f32.bf16 %r133, %rs15; + cvt.f32.bf16 %r134, %rs16; + cvt.f32.bf16 %r135, %rs17; + cvt.f32.bf16 %r136, %rs18; + cvt.f32.bf16 %r137, %rs19; + cvt.f32.bf16 %r138, %rs20; + cvt.f32.bf16 %r139, %rs21; + cvt.f32.bf16 %r140, %rs22; + cvt.f32.bf16 %r141, %rs23; + cvt.f32.bf16 %r142, %rs24; + cvt.f32.bf16 %r143, %rs25; + cvt.f32.bf16 %r144, %rs26; + cvt.f32.bf16 %r145, %rs27; + cvt.f32.bf16 %r146, %rs28; + cvt.f32.bf16 %r147, %rs29; + cvt.f32.bf16 %r148, %rs30; + cvt.f32.bf16 %r149, %rs31; + cvt.f32.bf16 %r150, %rs32; + cvt.f32.bf16 %r151, %rs33; + cvt.f32.bf16 %r152, %rs34; + cvt.f32.bf16 %r153, %rs35; + cvt.f32.bf16 %r154, %rs36; + cvt.f32.bf16 %r155, %rs37; + cvt.f32.bf16 %r156, %rs38; + cvt.f32.bf16 %r157, %rs39; + cvt.f32.bf16 %r158, %rs40; + cvt.f32.bf16 %r159, %rs41; + cvt.f32.bf16 %r160, %rs42; + cvt.f32.bf16 %r161, %rs43; + cvt.f32.bf16 %r162, %rs44; + cvt.f32.bf16 %r163, %rs45; + cvt.f32.bf16 %r164, %rs46; + cvt.f32.bf16 %r165, %rs47; + cvt.f32.bf16 %r166, %rs48; + cvt.f32.bf16 %r167, %rs49; + cvt.f32.bf16 %r168, %rs50; + cvt.f32.bf16 %r169, %rs51; + cvt.f32.bf16 %r170, %rs52; + cvt.f32.bf16 %r171, %rs53; + cvt.f32.bf16 %r172, %rs54; + cvt.f32.bf16 %r173, %rs55; + cvt.f32.bf16 %r174, %rs56; + cvt.f32.bf16 %r175, %rs57; + cvt.f32.bf16 %r176, %rs58; + cvt.f32.bf16 %r177, %rs59; + cvt.f32.bf16 %r178, %rs60; + cvt.f32.bf16 %r179, %rs61; + cvt.f32.bf16 %r180, %rs62; + cvt.f32.bf16 %r181, %rs63; + cvt.f32.bf16 %r182, %rs64; + .loc 1 105 21 // l2norm.py:105:21 + // begin inline asm + mov.u32 %r34, %r5; + @%p1 ld.global.b32 { %r34 }, [ %rd9 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r35, %r5; + @%p2 ld.global.b32 { %r35 }, [ %rd10 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r36, %r5; + @%p3 ld.global.b32 { %r36 }, [ %rd11 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r37, %r5; + @%p4 ld.global.b32 { %r37 }, [ %rd12 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r38, %r5; + @%p5 ld.global.b32 { %r38 }, [ %rd13 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r39, %r5; + @%p6 ld.global.b32 { %r39 }, [ %rd14 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r40, %r5; + @%p7 ld.global.b32 { %r40 }, [ %rd15 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r41, %r5; + @%p8 ld.global.b32 { %r41 }, [ %rd16 + 0 ]; + // end inline asm + .loc 1 106 19 // l2norm.py:106:19 + // begin inline asm + mov.u32 %r42, %r5; + mov.u32 %r43, %r5; + mov.u32 %r44, %r5; + mov.u32 %r45, %r5; + @%p1 ld.global.v4.b32 { %r42, %r43, %r44, %r45 }, [ %rd17 + 0 ]; + // end inline asm + mov.b32 {%rs65, %rs66}, %r42; + mov.b32 {%rs67, %rs68}, %r43; + mov.b32 {%rs69, %rs70}, %r44; + mov.b32 {%rs71, %rs72}, %r45; + // begin inline asm + mov.u32 %r46, %r5; + mov.u32 %r47, %r5; + mov.u32 %r48, %r5; + mov.u32 %r49, %r5; + @%p2 ld.global.v4.b32 { %r46, %r47, %r48, %r49 }, [ %rd18 + 0 ]; + // end inline asm + mov.b32 {%rs73, %rs74}, %r46; + mov.b32 {%rs75, %rs76}, %r47; + mov.b32 {%rs77, %rs78}, %r48; + mov.b32 {%rs79, %rs80}, %r49; + // begin inline asm + mov.u32 %r50, %r5; + mov.u32 %r51, %r5; + mov.u32 %r52, %r5; + mov.u32 %r53, %r5; + @%p3 ld.global.v4.b32 { %r50, %r51, %r52, %r53 }, [ %rd19 + 0 ]; + // end inline asm + mov.b32 {%rs81, %rs82}, %r50; + mov.b32 {%rs83, %rs84}, %r51; + mov.b32 {%rs85, %rs86}, %r52; + mov.b32 {%rs87, %rs88}, %r53; + // begin inline asm + mov.u32 %r54, %r5; + mov.u32 %r55, %r5; + mov.u32 %r56, %r5; + mov.u32 %r57, %r5; + @%p4 ld.global.v4.b32 { %r54, %r55, %r56, %r57 }, [ %rd20 + 0 ]; + // end inline asm + mov.b32 {%rs89, %rs90}, %r54; + mov.b32 {%rs91, %rs92}, %r55; + mov.b32 {%rs93, %rs94}, %r56; + mov.b32 {%rs95, %rs96}, %r57; + // begin inline asm + mov.u32 %r58, %r5; + mov.u32 %r59, %r5; + mov.u32 %r60, %r5; + mov.u32 %r61, %r5; + @%p5 ld.global.v4.b32 { %r58, %r59, %r60, %r61 }, [ %rd21 + 0 ]; + // end inline asm + mov.b32 {%rs97, %rs98}, %r58; + mov.b32 {%rs99, %rs100}, %r59; + mov.b32 {%rs101, %rs102}, %r60; + mov.b32 {%rs103, %rs104}, %r61; + // begin inline asm + mov.u32 %r62, %r5; + mov.u32 %r63, %r5; + mov.u32 %r64, %r5; + mov.u32 %r65, %r5; + @%p6 ld.global.v4.b32 { %r62, %r63, %r64, %r65 }, [ %rd22 + 0 ]; + // end inline asm + mov.b32 {%rs105, %rs106}, %r62; + mov.b32 {%rs107, %rs108}, %r63; + mov.b32 {%rs109, %rs110}, %r64; + mov.b32 {%rs111, %rs112}, %r65; + // begin inline asm + mov.u32 %r66, %r5; + mov.u32 %r67, %r5; + mov.u32 %r68, %r5; + mov.u32 %r69, %r5; + @%p7 ld.global.v4.b32 { %r66, %r67, %r68, %r69 }, [ %rd23 + 0 ]; + // end inline asm + mov.b32 {%rs113, %rs114}, %r66; + mov.b32 {%rs115, %rs116}, %r67; + mov.b32 {%rs117, %rs118}, %r68; + mov.b32 {%rs119, %rs120}, %r69; + // begin inline asm + mov.u32 %r70, %r5; + mov.u32 %r71, %r5; + mov.u32 %r72, %r5; + mov.u32 %r73, %r5; + @%p8 ld.global.v4.b32 { %r70, %r71, %r72, %r73 }, [ %rd24 + 0 ]; + // end inline asm + mov.b32 {%rs121, %rs122}, %r70; + mov.b32 {%rs123, %rs124}, %r71; + mov.b32 {%rs125, %rs126}, %r72; + mov.b32 {%rs127, %rs128}, %r73; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r183, %rs65; + cvt.f32.bf16 %r184, %rs66; + cvt.f32.bf16 %r185, %rs67; + cvt.f32.bf16 %r186, %rs68; + cvt.f32.bf16 %r187, %rs69; + cvt.f32.bf16 %r188, %rs70; + cvt.f32.bf16 %r189, %rs71; + cvt.f32.bf16 %r190, %rs72; + cvt.f32.bf16 %r191, %rs73; + cvt.f32.bf16 %r192, %rs74; + cvt.f32.bf16 %r193, %rs75; + cvt.f32.bf16 %r194, %rs76; + cvt.f32.bf16 %r195, %rs77; + cvt.f32.bf16 %r196, %rs78; + cvt.f32.bf16 %r197, %rs79; + cvt.f32.bf16 %r198, %rs80; + cvt.f32.bf16 %r199, %rs81; + cvt.f32.bf16 %r200, %rs82; + cvt.f32.bf16 %r201, %rs83; + cvt.f32.bf16 %r202, %rs84; + cvt.f32.bf16 %r203, %rs85; + cvt.f32.bf16 %r204, %rs86; + cvt.f32.bf16 %r205, %rs87; + cvt.f32.bf16 %r206, %rs88; + cvt.f32.bf16 %r207, %rs89; + cvt.f32.bf16 %r208, %rs90; + cvt.f32.bf16 %r209, %rs91; + cvt.f32.bf16 %r210, %rs92; + cvt.f32.bf16 %r211, %rs93; + cvt.f32.bf16 %r212, %rs94; + cvt.f32.bf16 %r213, %rs95; + cvt.f32.bf16 %r214, %rs96; + cvt.f32.bf16 %r215, %rs97; + cvt.f32.bf16 %r216, %rs98; + cvt.f32.bf16 %r217, %rs99; + cvt.f32.bf16 %r218, %rs100; + cvt.f32.bf16 %r219, %rs101; + cvt.f32.bf16 %r220, %rs102; + cvt.f32.bf16 %r221, %rs103; + cvt.f32.bf16 %r222, %rs104; + cvt.f32.bf16 %r223, %rs105; + cvt.f32.bf16 %r224, %rs106; + cvt.f32.bf16 %r225, %rs107; + cvt.f32.bf16 %r226, %rs108; + cvt.f32.bf16 %r227, %rs109; + cvt.f32.bf16 %r228, %rs110; + cvt.f32.bf16 %r229, %rs111; + cvt.f32.bf16 %r230, %rs112; + cvt.f32.bf16 %r231, %rs113; + cvt.f32.bf16 %r232, %rs114; + cvt.f32.bf16 %r233, %rs115; + cvt.f32.bf16 %r234, %rs116; + cvt.f32.bf16 %r235, %rs117; + cvt.f32.bf16 %r236, %rs118; + cvt.f32.bf16 %r237, %rs119; + cvt.f32.bf16 %r238, %rs120; + cvt.f32.bf16 %r239, %rs121; + cvt.f32.bf16 %r240, %rs122; + cvt.f32.bf16 %r241, %rs123; + cvt.f32.bf16 %r242, %rs124; + cvt.f32.bf16 %r243, %rs125; + cvt.f32.bf16 %r244, %rs126; + cvt.f32.bf16 %r245, %rs127; + cvt.f32.bf16 %r246, %rs128; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r247, %r40, %r231; + mul.f32 %r248, %r40, %r232; + mul.f32 %r249, %r40, %r233; + mul.f32 %r250, %r40, %r234; + mul.f32 %r251, %r40, %r235; + mul.f32 %r252, %r40, %r236; + mul.f32 %r253, %r40, %r237; + mul.f32 %r254, %r40, %r238; + mul.f32 %r255, %r41, %r239; + mul.f32 %r256, %r41, %r240; + mul.f32 %r257, %r41, %r241; + mul.f32 %r258, %r41, %r242; + mul.f32 %r259, %r41, %r243; + mul.f32 %r260, %r41, %r244; + mul.f32 %r261, %r41, %r245; + mul.f32 %r262, %r41, %r246; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r263, %r120, %r184; + mul.f32 %r264, %r128, %r192; + mul.f32 %r265, %r136, %r200; + mul.f32 %r266, %r144, %r208; + mul.f32 %r267, %r152, %r216; + mul.f32 %r268, %r160, %r224; + mul.f32 %r269, %r168, %r232; + mul.f32 %r270, %r176, %r240; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r271, %rs1, %rs65, %r263; + fma.rn.f32.bf16 %r272, %rs3, %rs67, %r271; + fma.rn.f32.bf16 %r273, %rs4, %rs68, %r272; + fma.rn.f32.bf16 %r274, %rs5, %rs69, %r273; + fma.rn.f32.bf16 %r275, %rs6, %rs70, %r274; + fma.rn.f32.bf16 %r276, %rs7, %rs71, %r275; + fma.rn.f32.bf16 %r277, %rs8, %rs72, %r276; + fma.rn.f32.bf16 %r278, %rs9, %rs73, %r264; + fma.rn.f32.bf16 %r279, %rs11, %rs75, %r278; + fma.rn.f32.bf16 %r280, %rs12, %rs76, %r279; + fma.rn.f32.bf16 %r281, %rs13, %rs77, %r280; + fma.rn.f32.bf16 %r282, %rs14, %rs78, %r281; + fma.rn.f32.bf16 %r283, %rs15, %rs79, %r282; + fma.rn.f32.bf16 %r284, %rs16, %rs80, %r283; + fma.rn.f32.bf16 %r285, %rs17, %rs81, %r265; + fma.rn.f32.bf16 %r286, %rs19, %rs83, %r285; + fma.rn.f32.bf16 %r287, %rs20, %rs84, %r286; + fma.rn.f32.bf16 %r288, %rs21, %rs85, %r287; + fma.rn.f32.bf16 %r289, %rs22, %rs86, %r288; + fma.rn.f32.bf16 %r290, %rs23, %rs87, %r289; + fma.rn.f32.bf16 %r291, %rs24, %rs88, %r290; + fma.rn.f32.bf16 %r292, %rs25, %rs89, %r266; + fma.rn.f32.bf16 %r293, %rs27, %rs91, %r292; + fma.rn.f32.bf16 %r294, %rs28, %rs92, %r293; + fma.rn.f32.bf16 %r295, %rs29, %rs93, %r294; + fma.rn.f32.bf16 %r296, %rs30, %rs94, %r295; + fma.rn.f32.bf16 %r297, %rs31, %rs95, %r296; + fma.rn.f32.bf16 %r298, %rs32, %rs96, %r297; + fma.rn.f32.bf16 %r299, %rs33, %rs97, %r267; + fma.rn.f32.bf16 %r300, %rs35, %rs99, %r299; + fma.rn.f32.bf16 %r301, %rs36, %rs100, %r300; + fma.rn.f32.bf16 %r302, %rs37, %rs101, %r301; + fma.rn.f32.bf16 %r303, %rs38, %rs102, %r302; + fma.rn.f32.bf16 %r304, %rs39, %rs103, %r303; + fma.rn.f32.bf16 %r305, %rs40, %rs104, %r304; + fma.rn.f32.bf16 %r306, %rs41, %rs105, %r268; + fma.rn.f32.bf16 %r307, %rs43, %rs107, %r306; + fma.rn.f32.bf16 %r308, %rs44, %rs108, %r307; + fma.rn.f32.bf16 %r309, %rs45, %rs109, %r308; + fma.rn.f32.bf16 %r310, %rs46, %rs110, %r309; + fma.rn.f32.bf16 %r311, %rs47, %rs111, %r310; + fma.rn.f32.bf16 %r312, %rs48, %rs112, %r311; + fma.rn.f32.bf16 %r313, %rs49, %rs113, %r269; + fma.rn.f32.bf16 %r314, %rs51, %rs115, %r313; + fma.rn.f32.bf16 %r315, %rs52, %rs116, %r314; + fma.rn.f32.bf16 %r316, %rs53, %rs117, %r315; + fma.rn.f32.bf16 %r317, %rs54, %rs118, %r316; + fma.rn.f32.bf16 %r318, %rs55, %rs119, %r317; + fma.rn.f32.bf16 %r319, %rs56, %rs120, %r318; + fma.rn.f32.bf16 %r320, %rs57, %rs121, %r270; + fma.rn.f32.bf16 %r321, %rs59, %rs123, %r320; + fma.rn.f32.bf16 %r322, %rs60, %rs124, %r321; + fma.rn.f32.bf16 %r323, %rs61, %rs125, %r322; + fma.rn.f32.bf16 %r324, %rs62, %rs126, %r323; + fma.rn.f32.bf16 %r325, %rs63, %rs127, %r324; + fma.rn.f32.bf16 %r326, %rs64, %rs128, %r325; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r327, %r277, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r328, %r277, %r327; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r329, %r328, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r330, %r328, %r329; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r331, %r330, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r332, %r330, %r331; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r333, %r332, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r334, %r332, %r333; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r335, %r284, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r336, %r284, %r335; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r337, %r336, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r338, %r336, %r337; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r339, %r338, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r340, %r338, %r339; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r341, %r340, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r342, %r340, %r341; +$L__tmp18: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r343, %r291, 8, 31, -1; +$L__tmp19: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r344, %r291, %r343; +$L__tmp20: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r345, %r344, 4, 31, -1; +$L__tmp21: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r346, %r344, %r345; +$L__tmp22: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r347, %r346, 2, 31, -1; +$L__tmp23: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r348, %r346, %r347; +$L__tmp24: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r349, %r348, 1, 31, -1; +$L__tmp25: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r350, %r348, %r349; +$L__tmp26: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r351, %r298, 8, 31, -1; +$L__tmp27: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r352, %r298, %r351; +$L__tmp28: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r353, %r352, 4, 31, -1; +$L__tmp29: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r354, %r352, %r353; +$L__tmp30: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r355, %r354, 2, 31, -1; +$L__tmp31: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r356, %r354, %r355; +$L__tmp32: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r357, %r356, 1, 31, -1; +$L__tmp33: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r358, %r356, %r357; +$L__tmp34: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r359, %r305, 8, 31, -1; +$L__tmp35: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r360, %r305, %r359; +$L__tmp36: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r361, %r360, 4, 31, -1; +$L__tmp37: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r362, %r360, %r361; +$L__tmp38: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r363, %r362, 2, 31, -1; +$L__tmp39: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r364, %r362, %r363; +$L__tmp40: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r365, %r364, 1, 31, -1; +$L__tmp41: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r366, %r364, %r365; +$L__tmp42: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r367, %r312, 8, 31, -1; +$L__tmp43: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r368, %r312, %r367; +$L__tmp44: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r369, %r368, 4, 31, -1; +$L__tmp45: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r370, %r368, %r369; +$L__tmp46: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r371, %r370, 2, 31, -1; +$L__tmp47: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r372, %r370, %r371; +$L__tmp48: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r373, %r372, 1, 31, -1; +$L__tmp49: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r374, %r372, %r373; +$L__tmp50: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r375, %r319, 8, 31, -1; +$L__tmp51: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r376, %r319, %r375; +$L__tmp52: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r377, %r376, 4, 31, -1; +$L__tmp53: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r378, %r376, %r377; +$L__tmp54: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r379, %r378, 2, 31, -1; +$L__tmp55: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r380, %r378, %r379; +$L__tmp56: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r381, %r380, 1, 31, -1; +$L__tmp57: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r382, %r380, %r381; +$L__tmp58: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r383, %r326, 8, 31, -1; +$L__tmp59: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r384, %r326, %r383; +$L__tmp60: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r385, %r384, 4, 31, -1; +$L__tmp61: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r386, %r384, %r385; +$L__tmp62: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r387, %r386, 2, 31, -1; +$L__tmp63: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r388, %r386, %r387; +$L__tmp64: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r389, %r388, 1, 31, -1; +$L__tmp65: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r390, %r388, %r389; +$L__tmp66: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r391, %r334, %r119; + mul.f32 %r392, %r334, %r120; + mul.f32 %r393, %r334, %r121; + mul.f32 %r394, %r334, %r122; + mul.f32 %r395, %r334, %r123; + mul.f32 %r396, %r334, %r124; + mul.f32 %r397, %r334, %r125; + mul.f32 %r398, %r334, %r126; + mul.f32 %r399, %r342, %r127; + mul.f32 %r400, %r342, %r128; + mul.f32 %r401, %r342, %r129; + mul.f32 %r402, %r342, %r130; + mul.f32 %r403, %r342, %r131; + mul.f32 %r404, %r342, %r132; + mul.f32 %r405, %r342, %r133; + mul.f32 %r406, %r342, %r134; + mul.f32 %r407, %r350, %r135; + mul.f32 %r408, %r350, %r136; + mul.f32 %r409, %r350, %r137; + mul.f32 %r410, %r350, %r138; + mul.f32 %r411, %r350, %r139; + mul.f32 %r412, %r350, %r140; + mul.f32 %r413, %r350, %r141; + mul.f32 %r414, %r350, %r142; + mul.f32 %r415, %r358, %r143; + mul.f32 %r416, %r358, %r144; + mul.f32 %r417, %r358, %r145; + mul.f32 %r418, %r358, %r146; + mul.f32 %r419, %r358, %r147; + mul.f32 %r420, %r358, %r148; + mul.f32 %r421, %r358, %r149; + mul.f32 %r422, %r358, %r150; + mul.f32 %r423, %r366, %r151; + mul.f32 %r424, %r366, %r152; + mul.f32 %r425, %r366, %r153; + mul.f32 %r426, %r366, %r154; + mul.f32 %r427, %r366, %r155; + mul.f32 %r428, %r366, %r156; + mul.f32 %r429, %r366, %r157; + mul.f32 %r430, %r366, %r158; + mul.f32 %r431, %r374, %r159; + mul.f32 %r432, %r374, %r160; + mul.f32 %r433, %r374, %r161; + mul.f32 %r434, %r374, %r162; + mul.f32 %r435, %r374, %r163; + mul.f32 %r436, %r374, %r164; + mul.f32 %r437, %r374, %r165; + mul.f32 %r438, %r374, %r166; + mul.f32 %r439, %r382, %r167; + mul.f32 %r440, %r382, %r168; + mul.f32 %r441, %r382, %r169; + mul.f32 %r442, %r382, %r170; + mul.f32 %r443, %r382, %r171; + mul.f32 %r444, %r382, %r172; + mul.f32 %r445, %r382, %r173; + mul.f32 %r446, %r382, %r174; + mul.f32 %r447, %r390, %r175; + mul.f32 %r448, %r390, %r176; + mul.f32 %r449, %r390, %r177; + mul.f32 %r450, %r390, %r178; + mul.f32 %r451, %r390, %r179; + mul.f32 %r452, %r390, %r180; + mul.f32 %r453, %r390, %r181; + mul.f32 %r454, %r390, %r182; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r455, %r391, %r34; + mul.f32 %r456, %r392, %r34; + mul.f32 %r457, %r393, %r34; + mul.f32 %r458, %r394, %r34; + mul.f32 %r459, %r395, %r34; + mul.f32 %r460, %r396, %r34; + mul.f32 %r461, %r397, %r34; + mul.f32 %r462, %r398, %r34; + mul.f32 %r463, %r399, %r35; + mul.f32 %r464, %r400, %r35; + mul.f32 %r465, %r401, %r35; + mul.f32 %r466, %r402, %r35; + mul.f32 %r467, %r403, %r35; + mul.f32 %r468, %r404, %r35; + mul.f32 %r469, %r405, %r35; + mul.f32 %r470, %r406, %r35; + mul.f32 %r471, %r407, %r36; + mul.f32 %r472, %r408, %r36; + mul.f32 %r473, %r409, %r36; + mul.f32 %r474, %r410, %r36; + mul.f32 %r475, %r411, %r36; + mul.f32 %r476, %r412, %r36; + mul.f32 %r477, %r413, %r36; + mul.f32 %r478, %r414, %r36; + mul.f32 %r479, %r415, %r37; + mul.f32 %r480, %r416, %r37; + mul.f32 %r481, %r417, %r37; + mul.f32 %r482, %r418, %r37; + mul.f32 %r483, %r419, %r37; + mul.f32 %r484, %r420, %r37; + mul.f32 %r485, %r421, %r37; + mul.f32 %r486, %r422, %r37; + mul.f32 %r487, %r423, %r38; + mul.f32 %r488, %r424, %r38; + mul.f32 %r489, %r425, %r38; + mul.f32 %r490, %r426, %r38; + mul.f32 %r491, %r427, %r38; + mul.f32 %r492, %r428, %r38; + mul.f32 %r493, %r429, %r38; + mul.f32 %r494, %r430, %r38; + mul.f32 %r495, %r431, %r39; + mul.f32 %r496, %r432, %r39; + mul.f32 %r497, %r433, %r39; + mul.f32 %r498, %r434, %r39; + mul.f32 %r499, %r435, %r39; + mul.f32 %r500, %r436, %r39; + mul.f32 %r501, %r437, %r39; + mul.f32 %r502, %r438, %r39; + mul.f32 %r503, %r439, %r40; + mul.f32 %r504, %r440, %r40; + mul.f32 %r505, %r441, %r40; + mul.f32 %r506, %r442, %r40; + mul.f32 %r507, %r443, %r40; + mul.f32 %r508, %r444, %r40; + mul.f32 %r509, %r445, %r40; + mul.f32 %r510, %r446, %r40; + mul.f32 %r511, %r447, %r41; + mul.f32 %r512, %r448, %r41; + mul.f32 %r513, %r449, %r41; + mul.f32 %r514, %r450, %r41; + mul.f32 %r515, %r451, %r41; + mul.f32 %r516, %r452, %r41; + mul.f32 %r517, %r453, %r41; + mul.f32 %r518, %r454, %r41; + .loc 1 107 36 // l2norm.py:107:36 + neg.f32 %r519, %r455; + fma.rn.f32 %r520, %r34, %r183, %r519; + neg.f32 %r521, %r456; + fma.rn.f32 %r522, %r34, %r184, %r521; + neg.f32 %r523, %r457; + fma.rn.f32 %r524, %r34, %r185, %r523; + neg.f32 %r525, %r458; + fma.rn.f32 %r526, %r34, %r186, %r525; + neg.f32 %r527, %r459; + fma.rn.f32 %r528, %r34, %r187, %r527; + neg.f32 %r529, %r460; + fma.rn.f32 %r530, %r34, %r188, %r529; + neg.f32 %r531, %r461; + fma.rn.f32 %r532, %r34, %r189, %r531; + neg.f32 %r533, %r462; + fma.rn.f32 %r534, %r34, %r190, %r533; + neg.f32 %r535, %r463; + fma.rn.f32 %r536, %r35, %r191, %r535; + neg.f32 %r537, %r464; + fma.rn.f32 %r538, %r35, %r192, %r537; + neg.f32 %r539, %r465; + fma.rn.f32 %r540, %r35, %r193, %r539; + neg.f32 %r541, %r466; + fma.rn.f32 %r542, %r35, %r194, %r541; + neg.f32 %r543, %r467; + fma.rn.f32 %r544, %r35, %r195, %r543; + neg.f32 %r545, %r468; + fma.rn.f32 %r546, %r35, %r196, %r545; + neg.f32 %r547, %r469; + fma.rn.f32 %r548, %r35, %r197, %r547; + neg.f32 %r549, %r470; + fma.rn.f32 %r550, %r35, %r198, %r549; + neg.f32 %r551, %r471; + fma.rn.f32 %r552, %r36, %r199, %r551; + neg.f32 %r553, %r472; + fma.rn.f32 %r554, %r36, %r200, %r553; + neg.f32 %r555, %r473; + fma.rn.f32 %r556, %r36, %r201, %r555; + neg.f32 %r557, %r474; + fma.rn.f32 %r558, %r36, %r202, %r557; + neg.f32 %r559, %r475; + fma.rn.f32 %r560, %r36, %r203, %r559; + neg.f32 %r561, %r476; + fma.rn.f32 %r562, %r36, %r204, %r561; + neg.f32 %r563, %r477; + fma.rn.f32 %r564, %r36, %r205, %r563; + neg.f32 %r565, %r478; + fma.rn.f32 %r566, %r36, %r206, %r565; + neg.f32 %r567, %r479; + fma.rn.f32 %r568, %r37, %r207, %r567; + neg.f32 %r569, %r480; + fma.rn.f32 %r570, %r37, %r208, %r569; + neg.f32 %r571, %r481; + fma.rn.f32 %r572, %r37, %r209, %r571; + neg.f32 %r573, %r482; + fma.rn.f32 %r574, %r37, %r210, %r573; + neg.f32 %r575, %r483; + fma.rn.f32 %r576, %r37, %r211, %r575; + neg.f32 %r577, %r484; + fma.rn.f32 %r578, %r37, %r212, %r577; + neg.f32 %r579, %r485; + fma.rn.f32 %r580, %r37, %r213, %r579; + neg.f32 %r581, %r486; + fma.rn.f32 %r582, %r37, %r214, %r581; + neg.f32 %r583, %r487; + fma.rn.f32 %r584, %r38, %r215, %r583; + neg.f32 %r585, %r488; + fma.rn.f32 %r586, %r38, %r216, %r585; + neg.f32 %r587, %r489; + fma.rn.f32 %r588, %r38, %r217, %r587; + neg.f32 %r589, %r490; + fma.rn.f32 %r590, %r38, %r218, %r589; + neg.f32 %r591, %r491; + fma.rn.f32 %r592, %r38, %r219, %r591; + neg.f32 %r593, %r492; + fma.rn.f32 %r594, %r38, %r220, %r593; + neg.f32 %r595, %r493; + fma.rn.f32 %r596, %r38, %r221, %r595; + neg.f32 %r597, %r494; + fma.rn.f32 %r598, %r38, %r222, %r597; + neg.f32 %r599, %r495; + fma.rn.f32 %r600, %r39, %r223, %r599; + neg.f32 %r601, %r496; + fma.rn.f32 %r602, %r39, %r224, %r601; + neg.f32 %r603, %r497; + fma.rn.f32 %r604, %r39, %r225, %r603; + neg.f32 %r605, %r498; + fma.rn.f32 %r606, %r39, %r226, %r605; + neg.f32 %r607, %r499; + fma.rn.f32 %r608, %r39, %r227, %r607; + neg.f32 %r609, %r500; + fma.rn.f32 %r610, %r39, %r228, %r609; + neg.f32 %r611, %r501; + fma.rn.f32 %r612, %r39, %r229, %r611; + neg.f32 %r613, %r502; + fma.rn.f32 %r614, %r39, %r230, %r613; + mov.b64 %rd96, {%r247, %r248}; + mov.b64 %rd97, {%r503, %r504}; + sub.f32x2 %rd98, %rd96, %rd97; + mov.b64 %rd99, {%r249, %r250}; + mov.b64 %rd100, {%r505, %r506}; + sub.f32x2 %rd101, %rd99, %rd100; + mov.b64 %rd102, {%r251, %r252}; + mov.b64 %rd103, {%r507, %r508}; + sub.f32x2 %rd104, %rd102, %rd103; + mov.b64 %rd105, {%r253, %r254}; + mov.b64 %rd106, {%r509, %r510}; + sub.f32x2 %rd107, %rd105, %rd106; + mov.b64 %rd108, {%r255, %r256}; + mov.b64 %rd109, {%r511, %r512}; + sub.f32x2 %rd110, %rd108, %rd109; + mov.b64 %rd111, {%r257, %r258}; + mov.b64 %rd112, {%r513, %r514}; + sub.f32x2 %rd113, %rd111, %rd112; + mov.b64 %rd114, {%r259, %r260}; + mov.b64 %rd115, {%r515, %r516}; + sub.f32x2 %rd116, %rd114, %rd115; + mov.b64 %rd117, {%r261, %r262}; + mov.b64 %rd118, {%r517, %r518}; + sub.f32x2 %rd119, %rd117, %rd118; + .loc 1 108 27 // l2norm.py:108:27 + mov.b64 {%r615, %r616}, %rd98; + cvt.rn.bf16x2.f32 %r98, %r616, %r615; + mov.b64 {%r617, %r618}, %rd101; + cvt.rn.bf16x2.f32 %r99, %r618, %r617; + mov.b64 {%r619, %r620}, %rd104; + cvt.rn.bf16x2.f32 %r100, %r620, %r619; + mov.b64 {%r621, %r622}, %rd107; + cvt.rn.bf16x2.f32 %r101, %r622, %r621; + mov.b64 {%r623, %r624}, %rd110; + cvt.rn.bf16x2.f32 %r102, %r624, %r623; + mov.b64 {%r625, %r626}, %rd113; + cvt.rn.bf16x2.f32 %r103, %r626, %r625; + mov.b64 {%r627, %r628}, %rd116; + cvt.rn.bf16x2.f32 %r104, %r628, %r627; + mov.b64 {%r629, %r630}, %rd119; + cvt.rn.bf16x2.f32 %r105, %r630, %r629; + .loc 1 108 19 // l2norm.py:108:19 + cvt.rn.bf16x2.f32 %r74, %r522, %r520; + cvt.rn.bf16x2.f32 %r75, %r526, %r524; + cvt.rn.bf16x2.f32 %r76, %r530, %r528; + cvt.rn.bf16x2.f32 %r77, %r534, %r532; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd25 + 0 ], { %r74, %r75, %r76, %r77 }; + // end inline asm + cvt.rn.bf16x2.f32 %r78, %r538, %r536; + cvt.rn.bf16x2.f32 %r79, %r542, %r540; + cvt.rn.bf16x2.f32 %r80, %r546, %r544; + cvt.rn.bf16x2.f32 %r81, %r550, %r548; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd26 + 0 ], { %r78, %r79, %r80, %r81 }; + // end inline asm + cvt.rn.bf16x2.f32 %r82, %r554, %r552; + cvt.rn.bf16x2.f32 %r83, %r558, %r556; + cvt.rn.bf16x2.f32 %r84, %r562, %r560; + cvt.rn.bf16x2.f32 %r85, %r566, %r564; + // begin inline asm + @%p3 st.global.v4.b32 [ %rd27 + 0 ], { %r82, %r83, %r84, %r85 }; + // end inline asm + cvt.rn.bf16x2.f32 %r86, %r570, %r568; + cvt.rn.bf16x2.f32 %r87, %r574, %r572; + cvt.rn.bf16x2.f32 %r88, %r578, %r576; + cvt.rn.bf16x2.f32 %r89, %r582, %r580; + // begin inline asm + @%p4 st.global.v4.b32 [ %rd28 + 0 ], { %r86, %r87, %r88, %r89 }; + // end inline asm + cvt.rn.bf16x2.f32 %r90, %r586, %r584; + cvt.rn.bf16x2.f32 %r91, %r590, %r588; + cvt.rn.bf16x2.f32 %r92, %r594, %r592; + cvt.rn.bf16x2.f32 %r93, %r598, %r596; + // begin inline asm + @%p5 st.global.v4.b32 [ %rd29 + 0 ], { %r90, %r91, %r92, %r93 }; + // end inline asm + cvt.rn.bf16x2.f32 %r94, %r602, %r600; + cvt.rn.bf16x2.f32 %r95, %r606, %r604; + cvt.rn.bf16x2.f32 %r96, %r610, %r608; + cvt.rn.bf16x2.f32 %r97, %r614, %r612; + // begin inline asm + @%p6 st.global.v4.b32 [ %rd30 + 0 ], { %r94, %r95, %r96, %r97 }; + // end inline asm + // begin inline asm + @%p7 st.global.v4.b32 [ %rd31 + 0 ], { %r98, %r99, %r100, %r101 }; + // end inline asm + // begin inline asm + @%p8 st.global.v4.b32 [ %rd32 + 0 ], { %r102, %r103, %r104, %r105 }; + // end inline asm + .loc 1 108 4 // l2norm.py:108:4 + ret; +$L__tmp67: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 98 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp66 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 107 // DW_AT_call_line +.b8 55 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp66 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/l2norm_bwd_kernel.source b/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/l2norm_bwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..35265513bc03aa2afd3347f28178370dc92e754e --- /dev/null +++ b/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/l2norm_bwd_kernel.source @@ -0,0 +1,210 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc47 = loc(unknown) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc54 = loc("y"(#loc)) +#loc55 = loc("rstd"(#loc)) +#loc56 = loc("dy"(#loc)) +#loc57 = loc("dx"(#loc)) +#loc58 = loc("eps"(#loc)) +#loc59 = loc("T"(#loc)) +#loc101 = loc("input"(#loc45)) +#loc102 = loc("a"(#loc50)) +#loc103 = loc("b"(#loc50)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc60) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc61) + %o_t = arith.constant 128 : i32 loc(#loc62) + %o_t_1 = arith.constant 128 : i64 loc(#loc62) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc62) + %o_t_3 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc63) + %o_t_4 = arith.extsi %o_t_3 : tensor<128xi32> to tensor<128xi64> loc(#loc64) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<128xi64> loc(#loc64) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<128xi64> loc(#loc64) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc65) + %m_t = arith.extsi %T : i32 to i64 loc(#loc66) + %m_t_7 = tt.splat %m_t : i64 -> tensor<128xi64> loc(#loc66) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<128xi64> loc(#loc66) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<128xi1> -> tensor<128x1xi1> loc(#loc67) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc68) + %m_x_10 = arith.constant 128 : i32 loc(#loc69) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc69) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc69) + %m_x_13 = tt.broadcast %m_x : tensor<128x1xi1> -> tensor<128x128xi1> loc(#loc70) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<128x128xi1> loc(#loc70) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<128x128xi1> loc(#loc70) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc71) + %p_y_16 = arith.constant 128 : i32 loc(#loc72) + %p_y_17 = arith.constant 128 : i64 loc(#loc72) + %p_y_18 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc72) + %p_y_19 = arith.muli %p_y, %p_y_18 : tensor<128x1xi64> loc(#loc72) + %p_y_20 = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc73) + %p_y_21 = tt.addptr %p_y_20, %p_y_19 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc73) + %p_y_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc74) + %p_y_23 = tt.broadcast %p_y_21 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc75) + %p_y_24 = tt.broadcast %p_y_22 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc75) + %p_y_25 = tt.addptr %p_y_23, %p_y_24 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc75) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr> loc(#loc76) + %p_rstd_26 = tt.addptr %p_rstd, %o_t_6 : tensor<128x!tt.ptr>, tensor<128xi64> loc(#loc76) + %p_dy = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc77) + %p_dy_27 = arith.constant 128 : i32 loc(#loc78) + %p_dy_28 = arith.constant 128 : i64 loc(#loc78) + %p_dy_29 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc78) + %p_dy_30 = arith.muli %p_dy, %p_dy_29 : tensor<128x1xi64> loc(#loc78) + %p_dy_31 = tt.splat %dy : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc79) + %p_dy_32 = tt.addptr %p_dy_31, %p_dy_30 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc79) + %p_dy_33 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc80) + %p_dy_34 = tt.broadcast %p_dy_32 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc81) + %p_dy_35 = tt.broadcast %p_dy_33 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc81) + %p_dy_36 = tt.addptr %p_dy_34, %p_dy_35 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc81) + %p_dx = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc82) + %p_dx_37 = arith.constant 128 : i32 loc(#loc83) + %p_dx_38 = arith.constant 128 : i64 loc(#loc83) + %p_dx_39 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc83) + %p_dx_40 = arith.muli %p_dx, %p_dx_39 : tensor<128x1xi64> loc(#loc83) + %p_dx_41 = tt.splat %dx : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc84) + %p_dx_42 = tt.addptr %p_dx_41, %p_dx_40 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc84) + %p_dx_43 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc85) + %p_dx_44 = tt.broadcast %p_dx_42 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc86) + %p_dx_45 = tt.broadcast %p_dx_43 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc86) + %p_dx_46 = tt.addptr %p_dx_44, %p_dx_45 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc86) + %b_y = arith.constant 0.000000e+00 : f32 loc(#loc87) + %b_y_47 = arith.constant dense<0.000000e+00> : tensor<128x128xf32> loc(#loc87) + %b_y_48 = arith.truncf %b_y_47 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc87) + %b_y_49 = tt.load %p_y_25, %m_x_15, %b_y_48 : tensor<128x128x!tt.ptr> loc(#loc87) + %b_y_50 = arith.extf %b_y_49 : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc88) + %b_rstd = arith.constant 0.000000e+00 : f32 loc(#loc89) + %b_rstd_51 = arith.constant dense<0.000000e+00> : tensor<128xf32> loc(#loc89) + %b_rstd_52 = tt.load %p_rstd_26, %m_t_8, %b_rstd_51 : tensor<128x!tt.ptr> loc(#loc89) + %b_dy = arith.constant 0.000000e+00 : f32 loc(#loc90) + %b_dy_53 = arith.constant dense<0.000000e+00> : tensor<128x128xf32> loc(#loc90) + %b_dy_54 = arith.truncf %b_dy_53 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc90) + %b_dy_55 = tt.load %p_dy_36, %m_x_15, %b_dy_54 : tensor<128x128x!tt.ptr> loc(#loc90) + %b_dy_56 = arith.extf %b_dy_55 : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc91) + %b_dx = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc92) + %b_dx_57 = tt.broadcast %b_dx : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc93) + %b_dx_58 = arith.mulf %b_dy_56, %b_dx_57 : tensor<128x128xf32> loc(#loc93) + %b_dx_59 = arith.mulf %b_dy_56, %b_y_50 : tensor<128x128xf32> loc(#loc94) + %b_dx_60 = tt.call @triton.language.standard.sum__fp32S128_128S_c1_cFalse_cNone(%b_dx_59) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc95) + %b_dx_61 = tt.expand_dims %b_dx_60 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc96) + %b_dx_62 = tt.broadcast %b_dx_61 : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc97) + %b_dx_63 = arith.mulf %b_dx_62, %b_y_50 : tensor<128x128xf32> loc(#loc97) + %b_dx_64 = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc98) + %b_dx_65 = tt.broadcast %b_dx_64 : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc99) + %b_dx_66 = arith.mulf %b_dx_63, %b_dx_65 : tensor<128x128xf32> loc(#loc99) + %b_dx_67 = arith.subf %b_dx_58, %b_dx_66 : tensor<128x128xf32> loc(#loc100) + %0 = arith.truncf %b_dx_67 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc42) + tt.store %p_dx_46, %0, %m_x_15 : tensor<128x128x!tt.ptr> loc(#loc43) + tt.return loc(#loc44) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S128_128S_c1_cFalse_cNone(%input: tensor<128x128xf32> loc("input"(#loc45))) -> tensor<128xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc46) + tt.reduce.return %2 : f32 loc(#loc46) + }) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc46) + tt.return %0 : tensor<128xf32> loc(#loc48) + ^bb1: // no predecessors + %1 = ub.poison : tensor<128xf32> loc(#loc49) + tt.return %1 : tensor<128xf32> loc(#loc49) + } loc(#loc45) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc50)), %b: f32 loc("b"(#loc50))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc51) + tt.return %0 : f32 loc(#loc52) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc53) + tt.return %1 : f32 loc(#loc53) + } loc(#loc50) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:31) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:39) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:20) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:31) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:39) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:82) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc60 = loc("i_t"(#loc1)) +#loc61 = loc("i_t"(#loc2)) +#loc62 = loc("o_t"(#loc3)) +#loc63 = loc("o_t"(#loc4)) +#loc64 = loc("o_t"(#loc5)) +#loc65 = loc("o_d"(#loc6)) +#loc66 = loc("m_t"(#loc7)) +#loc67 = loc("m_x"(#loc8)) +#loc68 = loc("m_x"(#loc9)) +#loc69 = loc("m_x"(#loc10)) +#loc70 = loc("m_x"(#loc11)) +#loc71 = loc("p_y"(#loc12)) +#loc72 = loc("p_y"(#loc13)) +#loc73 = loc("p_y"(#loc14)) +#loc74 = loc("p_y"(#loc15)) +#loc75 = loc("p_y"(#loc16)) +#loc76 = loc("p_rstd"(#loc17)) +#loc77 = loc("p_dy"(#loc18)) +#loc78 = loc("p_dy"(#loc19)) +#loc79 = loc("p_dy"(#loc20)) +#loc80 = loc("p_dy"(#loc21)) +#loc81 = loc("p_dy"(#loc22)) +#loc82 = loc("p_dx"(#loc23)) +#loc83 = loc("p_dx"(#loc24)) +#loc84 = loc("p_dx"(#loc25)) +#loc85 = loc("p_dx"(#loc26)) +#loc86 = loc("p_dx"(#loc27)) +#loc87 = loc("b_y"(#loc28)) +#loc88 = loc("b_y"(#loc29)) +#loc89 = loc("b_rstd"(#loc30)) +#loc90 = loc("b_dy"(#loc31)) +#loc91 = loc("b_dy"(#loc32)) +#loc92 = loc("b_dx"(#loc33)) +#loc93 = loc("b_dx"(#loc34)) +#loc94 = loc("b_dx"(#loc35)) +#loc95 = loc("b_dx"(#loc36)) +#loc96 = loc("b_dx"(#loc37)) +#loc97 = loc("b_dx"(#loc38)) +#loc98 = loc("b_dx"(#loc39)) +#loc99 = loc("b_dx"(#loc40)) +#loc100 = loc("b_dx"(#loc41)) diff --git a/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/l2norm_bwd_kernel.ttgir b/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/l2norm_bwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..cf4819b5fa01984c7a6220e72e56e103945fa67d --- /dev/null +++ b/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/l2norm_bwd_kernel.ttgir @@ -0,0 +1,146 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [8, 1], order = [1, 0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc39 = loc("y"(#loc)) +#loc40 = loc("rstd"(#loc)) +#loc41 = loc("dy"(#loc)) +#loc42 = loc("dx"(#loc)) +#loc43 = loc("eps"(#loc)) +#loc44 = loc("T"(#loc)) +#loc72 = loc("b_dx"(#loc30)) +#loc78 = loc(callsite(#loc1 at #loc72)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 8 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %c128_i64 = arith.constant 128 : i64 loc(#loc1) + %cst = arith.constant dense<0.000000e+00> : tensor<128x128xbf16, #blocked> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<128x1xi64, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc45) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc46) + %o_t = arith.muli %i_t_3, %c128_i64 : i64 loc(#loc47) + %o_t_4 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc48) + %o_t_5 = arith.extsi %o_t_4 : tensor<128xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %o_t_6 = tt.splat %o_t : i64 -> tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %o_t_7 = arith.addi %o_t_6, %o_t_5 : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %m_t = arith.extsi %T : i32 to i64 loc(#loc50) + %m_t_8 = tt.splat %m_t : i64 -> tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc50) + %m_t_9 = arith.cmpi slt, %o_t_7, %m_t_8 : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc50) + %m_x = tt.expand_dims %m_t_9 {axis = 1 : i32} : tensor<128xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128x1xi1, #blocked> loc(#loc51) + %m_x_10 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc52) + %m_x_11 = tt.expand_dims %m_x_10 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc52) + %m_x_12 = arith.cmpi slt, %m_x_11, %cst_2 : tensor<1x128xi32, #blocked> loc(#loc53) + %m_x_13 = tt.broadcast %m_x : tensor<128x1xi1, #blocked> -> tensor<128x128xi1, #blocked> loc(#loc54) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1, #blocked> -> tensor<128x128xi1, #blocked> loc(#loc54) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<128x128xi1, #blocked> loc(#loc54) + %p_y = tt.expand_dims %o_t_7 {axis = 1 : i32} : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128x1xi64, #blocked> loc(#loc55) + %p_y_16 = arith.muli %p_y, %cst_1 : tensor<128x1xi64, #blocked> loc(#loc56) + %p_y_17 = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr, #blocked> loc(#loc57) + %p_y_18 = tt.addptr %p_y_17, %p_y_16 : tensor<128x1x!tt.ptr, #blocked>, tensor<128x1xi64, #blocked> loc(#loc57) + %p_y_19 = tt.broadcast %p_y_18 : tensor<128x1x!tt.ptr, #blocked> -> tensor<128x128x!tt.ptr, #blocked> loc(#loc58) + %p_y_20 = tt.broadcast %m_x_11 : tensor<1x128xi32, #blocked> -> tensor<128x128xi32, #blocked> loc(#loc58) + %p_y_21 = tt.addptr %p_y_19, %p_y_20 : tensor<128x128x!tt.ptr, #blocked>, tensor<128x128xi32, #blocked> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc59) + %p_rstd_22 = tt.addptr %p_rstd, %o_t_7 : tensor<128x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>>, tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc59) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<128x1x!tt.ptr, #blocked> loc(#loc60) + %p_dy_23 = tt.addptr %p_dy, %p_y_16 : tensor<128x1x!tt.ptr, #blocked>, tensor<128x1xi64, #blocked> loc(#loc60) + %p_dy_24 = tt.broadcast %p_dy_23 : tensor<128x1x!tt.ptr, #blocked> -> tensor<128x128x!tt.ptr, #blocked> loc(#loc61) + %p_dy_25 = tt.addptr %p_dy_24, %p_y_20 : tensor<128x128x!tt.ptr, #blocked>, tensor<128x128xi32, #blocked> loc(#loc61) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<128x1x!tt.ptr, #blocked> loc(#loc62) + %p_dx_26 = tt.addptr %p_dx, %p_y_16 : tensor<128x1x!tt.ptr, #blocked>, tensor<128x1xi64, #blocked> loc(#loc62) + %p_dx_27 = tt.broadcast %p_dx_26 : tensor<128x1x!tt.ptr, #blocked> -> tensor<128x128x!tt.ptr, #blocked> loc(#loc63) + %p_dx_28 = tt.addptr %p_dx_27, %p_y_20 : tensor<128x128x!tt.ptr, #blocked>, tensor<128x128xi32, #blocked> loc(#loc63) + %b_y = tt.load %p_y_21, %m_x_15, %cst : tensor<128x128x!tt.ptr, #blocked> loc(#loc64) + %b_y_29 = arith.extf %b_y : tensor<128x128xbf16, #blocked> to tensor<128x128xf32, #blocked> loc(#loc65) + %b_rstd = tt.load %p_rstd_22, %m_t_9, %cst_0 : tensor<128x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc66) + %b_dy = tt.load %p_dy_25, %m_x_15, %cst : tensor<128x128x!tt.ptr, #blocked> loc(#loc67) + %b_dy_30 = arith.extf %b_dy : tensor<128x128xbf16, #blocked> to tensor<128x128xf32, #blocked> loc(#loc68) + %b_dx = tt.expand_dims %b_rstd {axis = 1 : i32} : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128x1xf32, #blocked> loc(#loc69) + %b_dx_31 = tt.broadcast %b_dx : tensor<128x1xf32, #blocked> -> tensor<128x128xf32, #blocked> loc(#loc70) + %b_dx_32 = arith.mulf %b_dy_30, %b_dx_31 : tensor<128x128xf32, #blocked> loc(#loc70) + %b_dx_33 = arith.mulf %b_dy_30, %b_y_29 : tensor<128x128xf32, #blocked> loc(#loc71) + %b_dx_34 = "tt.reduce"(%b_dx_33) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_40: f32 loc(callsite(#loc1 at #loc72)), %b_dx_41: f32 loc(callsite(#loc1 at #loc72))): + %b_dx_42 = arith.addf %b_dx_40, %b_dx_41 : f32 loc(#loc79) + tt.reduce.return %b_dx_42 : f32 loc(#loc77) + }) : (tensor<128x128xf32, #blocked>) -> tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc77) + %b_dx_35 = tt.expand_dims %b_dx_34 {axis = 1 : i32} : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128x1xf32, #blocked> loc(#loc73) + %b_dx_36 = tt.broadcast %b_dx_35 : tensor<128x1xf32, #blocked> -> tensor<128x128xf32, #blocked> loc(#loc74) + %b_dx_37 = arith.mulf %b_dx_36, %b_y_29 : tensor<128x128xf32, #blocked> loc(#loc74) + %b_dx_38 = arith.mulf %b_dx_37, %b_dx_31 : tensor<128x128xf32, #blocked> loc(#loc75) + %b_dx_39 = arith.subf %b_dx_32, %b_dx_38 : tensor<128x128xf32, #blocked> loc(#loc76) + %0 = arith.truncf %b_dx_39 : tensor<128x128xf32, #blocked> to tensor<128x128xbf16, #blocked> loc(#loc36) + tt.store %p_dx_28, %0, %m_x_15 : tensor<128x128x!tt.ptr, #blocked> loc(#loc37) + tt.return loc(#loc38) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc45 = loc("i_t"(#loc2)) +#loc46 = loc("i_t"(#loc3)) +#loc47 = loc("o_t"(#loc4)) +#loc48 = loc("o_t"(#loc5)) +#loc49 = loc("o_t"(#loc6)) +#loc50 = loc("m_t"(#loc7)) +#loc51 = loc("m_x"(#loc8)) +#loc52 = loc("m_x"(#loc9)) +#loc53 = loc("m_x"(#loc10)) +#loc54 = loc("m_x"(#loc11)) +#loc55 = loc("p_y"(#loc12)) +#loc56 = loc("p_y"(#loc13)) +#loc57 = loc("p_y"(#loc14)) +#loc58 = loc("p_y"(#loc15)) +#loc59 = loc("p_rstd"(#loc16)) +#loc60 = loc("p_dy"(#loc17)) +#loc61 = loc("p_dy"(#loc18)) +#loc62 = loc("p_dx"(#loc19)) +#loc63 = loc("p_dx"(#loc20)) +#loc64 = loc("b_y"(#loc21)) +#loc65 = loc("b_y"(#loc22)) +#loc66 = loc("b_rstd"(#loc23)) +#loc67 = loc("b_dy"(#loc24)) +#loc68 = loc("b_dy"(#loc25)) +#loc69 = loc("b_dx"(#loc26)) +#loc70 = loc("b_dx"(#loc27)) +#loc71 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc32)) +#loc74 = loc("b_dx"(#loc33)) +#loc75 = loc("b_dx"(#loc34)) +#loc76 = loc("b_dx"(#loc35)) +#loc77 = loc(callsite(#loc29 at #loc72)) +#loc79 = loc(callsite(#loc31 at #loc77)) diff --git a/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/l2norm_bwd_kernel.ttir b/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/l2norm_bwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..f242766466ced3ec80bd72ef6238b6accc66dbe7 --- /dev/null +++ b/triton/gpu0/BHAPEYRNAQIJ2JGAVTA5YRWIJ7EOL6K2FHY36DKXADMTJMNGDH5A/l2norm_bwd_kernel.ttir @@ -0,0 +1,144 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc39 = loc("y"(#loc)) +#loc40 = loc("rstd"(#loc)) +#loc41 = loc("dy"(#loc)) +#loc42 = loc("dx"(#loc)) +#loc43 = loc("eps"(#loc)) +#loc44 = loc("T"(#loc)) +#loc72 = loc("b_dx"(#loc30)) +#loc78 = loc(callsite(#loc1 at #loc72)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<0.000000e+00> : tensor<128x128xbf16> loc(#loc1) + %b_rstd = arith.constant dense<0.000000e+00> : tensor<128xf32> loc(#loc45) + %cst_0 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc1) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc46) + %c128_i64 = arith.constant 128 : i64 loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc47) + %i_t_1 = arith.extsi %i_t : i32 to i64 loc(#loc48) + %o_t = arith.muli %i_t_1, %c128_i64 : i64 loc(#loc49) + %o_t_2 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc50) + %o_t_3 = arith.extsi %o_t_2 : tensor<128xi32> to tensor<128xi64> loc(#loc51) + %o_t_4 = tt.splat %o_t : i64 -> tensor<128xi64> loc(#loc51) + %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<128xi64> loc(#loc51) + %m_t = arith.extsi %T : i32 to i64 loc(#loc52) + %m_t_6 = tt.splat %m_t : i64 -> tensor<128xi64> loc(#loc52) + %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<128xi64> loc(#loc52) + %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<128xi1> -> tensor<128x1xi1> loc(#loc53) + %m_x_9 = tt.expand_dims %o_t_2 {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc54) + %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc46) + %m_x_11 = tt.broadcast %m_x_8 : tensor<128x1xi1> -> tensor<128x128xi1> loc(#loc55) + %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<128x128xi1> loc(#loc55) + %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<128x128xi1> loc(#loc55) + %p_y = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc56) + %p_y_14 = arith.muli %p_y, %cst_0 : tensor<128x1xi64> loc(#loc57) + %p_y_15 = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc58) + %p_y_16 = tt.addptr %p_y_15, %p_y_14 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc58) + %p_y_17 = tt.broadcast %p_y_16 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc59) + %p_y_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc59) + %p_y_19 = tt.addptr %p_y_17, %p_y_18 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc59) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr> loc(#loc60) + %p_rstd_20 = tt.addptr %p_rstd, %o_t_5 : tensor<128x!tt.ptr>, tensor<128xi64> loc(#loc60) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc61) + %p_dy_21 = tt.addptr %p_dy, %p_y_14 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc61) + %p_dy_22 = tt.broadcast %p_dy_21 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc62) + %p_dy_23 = tt.addptr %p_dy_22, %p_y_18 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc62) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc63) + %p_dx_24 = tt.addptr %p_dx, %p_y_14 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc63) + %p_dx_25 = tt.broadcast %p_dx_24 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc64) + %p_dx_26 = tt.addptr %p_dx_25, %p_y_18 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc64) + %b_y = tt.load %p_y_19, %m_x_13, %cst : tensor<128x128x!tt.ptr> loc(#loc65) + %b_y_27 = arith.extf %b_y : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc66) + %b_rstd_28 = tt.load %p_rstd_20, %m_t_7, %b_rstd : tensor<128x!tt.ptr> loc(#loc45) + %b_dy = tt.load %p_dy_23, %m_x_13, %cst : tensor<128x128x!tt.ptr> loc(#loc67) + %b_dy_29 = arith.extf %b_dy : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc68) + %b_dx = tt.expand_dims %b_rstd_28 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc69) + %b_dx_30 = tt.broadcast %b_dx : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc70) + %b_dx_31 = arith.mulf %b_dy_29, %b_dx_30 : tensor<128x128xf32> loc(#loc70) + %b_dx_32 = arith.mulf %b_dy_29, %b_y_27 : tensor<128x128xf32> loc(#loc71) + %b_dx_33 = "tt.reduce"(%b_dx_32) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_39: f32 loc(callsite(#loc1 at #loc72)), %b_dx_40: f32 loc(callsite(#loc1 at #loc72))): + %b_dx_41 = arith.addf %b_dx_39, %b_dx_40 : f32 loc(#loc79) + tt.reduce.return %b_dx_41 : f32 loc(#loc77) + }) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc77) + %b_dx_34 = tt.expand_dims %b_dx_33 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc73) + %b_dx_35 = tt.broadcast %b_dx_34 : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc74) + %b_dx_36 = arith.mulf %b_dx_35, %b_y_27 : tensor<128x128xf32> loc(#loc74) + %b_dx_37 = arith.mulf %b_dx_36, %b_dx_30 : tensor<128x128xf32> loc(#loc75) + %b_dx_38 = arith.subf %b_dx_31, %b_dx_37 : tensor<128x128xf32> loc(#loc76) + %0 = arith.truncf %b_dx_38 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc36) + tt.store %p_dx_26, %0, %m_x_13 : tensor<128x128x!tt.ptr> loc(#loc37) + tt.return loc(#loc38) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc45 = loc("b_rstd"(#loc2)) +#loc46 = loc("m_x"(#loc3)) +#loc47 = loc("i_t"(#loc4)) +#loc48 = loc("i_t"(#loc5)) +#loc49 = loc("o_t"(#loc6)) +#loc50 = loc("o_t"(#loc7)) +#loc51 = loc("o_t"(#loc8)) +#loc52 = loc("m_t"(#loc9)) +#loc53 = loc("m_x"(#loc10)) +#loc54 = loc("m_x"(#loc11)) +#loc55 = loc("m_x"(#loc12)) +#loc56 = loc("p_y"(#loc13)) +#loc57 = loc("p_y"(#loc14)) +#loc58 = loc("p_y"(#loc15)) +#loc59 = loc("p_y"(#loc16)) +#loc60 = loc("p_rstd"(#loc17)) +#loc61 = loc("p_dy"(#loc18)) +#loc62 = loc("p_dy"(#loc19)) +#loc63 = loc("p_dx"(#loc20)) +#loc64 = loc("p_dx"(#loc21)) +#loc65 = loc("b_y"(#loc22)) +#loc66 = loc("b_y"(#loc23)) +#loc67 = loc("b_dy"(#loc24)) +#loc68 = loc("b_dy"(#loc25)) +#loc69 = loc("b_dx"(#loc26)) +#loc70 = loc("b_dx"(#loc27)) +#loc71 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc32)) +#loc74 = loc("b_dx"(#loc33)) +#loc75 = loc("b_dx"(#loc34)) +#loc76 = loc("b_dx"(#loc35)) +#loc77 = loc(callsite(#loc29 at #loc72)) +#loc79 = loc(callsite(#loc31 at #loc77)) diff --git a/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/__grp__l2norm_fwd_kernel.json b/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/__grp__l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..a2213c16b9c7c0ab79fb72aefc38c9e68dc5d86b --- /dev/null +++ b/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/__grp__l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_fwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/l2norm_fwd_kernel.source", "l2norm_fwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/l2norm_fwd_kernel.ttir", "l2norm_fwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/l2norm_fwd_kernel.ttgir", "l2norm_fwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/l2norm_fwd_kernel.llir", "l2norm_fwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/l2norm_fwd_kernel.ptx", "l2norm_fwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/l2norm_fwd_kernel.cubin", "l2norm_fwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/l2norm_fwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/l2norm_fwd_kernel.cubin b/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/l2norm_fwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..8bfc2e9fa91e80ecdbc7c2c84508a1036b0830a0 Binary files /dev/null and b/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/l2norm_fwd_kernel.cubin differ diff --git a/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/l2norm_fwd_kernel.json b/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..8d7c1b562b689bbe0bcdf547a872c2687c0d83ee --- /dev/null +++ b/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"hash": "09c8cc4332be29e06c7f2a10b9b216ecc8c3fce52681f26686b4ff094094f478", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 2, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 256, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_fwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/l2norm_fwd_kernel.llir b/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/l2norm_fwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..9b57de57013c4abfc4fbb540f66a5c82af16a1d8 --- /dev/null +++ b/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/l2norm_fwd_kernel.llir @@ -0,0 +1,2155 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 +@.str.2 = private unnamed_addr constant [17 x i8] c"__CUDA_PREC_SQRT\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_fwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, float %3, i32 %4, ptr addrspace(1) readnone captures(none) %5, ptr addrspace(1) readnone captures(none) %6) local_unnamed_addr #0 !dbg !5 { + %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %9 = zext nneg i32 %8 to i64, !dbg !14 + %10 = shl nuw nsw i64 %9, 6, !dbg !15 + %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !16 + %12 = and i32 %11, 48, !dbg !16 + %13 = lshr exact i32 %12, 4, !dbg !16 + %14 = or disjoint i32 %13, 4, !dbg !16 + %15 = insertelement <2 x i32> poison, i32 %13, i64 0, !dbg !16 + %16 = insertelement <4 x i32> poison, i32 %13, i64 0, !dbg !16 + %17 = insertelement <8 x i32> poison, i32 %13, i64 0, !dbg !16 + %18 = insertelement <16 x i32> poison, i32 %14, i64 14, !dbg !17 + %19 = insertelement <16 x i32> %18, i32 %13, i64 15, !dbg !17 + %20 = shufflevector <8 x i32> %17, <8 x i32> poison, <16 x i32> , !dbg !17 + %21 = or disjoint <16 x i32> %20, , !dbg !17 + %22 = shufflevector <16 x i32> %21, <16 x i32> %19, <16 x i32> , !dbg !17 + %23 = shufflevector <4 x i32> %16, <4 x i32> poison, <16 x i32> , !dbg !17 + %24 = or disjoint <16 x i32> %23, , !dbg !17 + %25 = shufflevector <16 x i32> %22, <16 x i32> %24, <16 x i32> , !dbg !17 + %26 = shufflevector <2 x i32> %15, <2 x i32> poison, <16 x i32> , !dbg !17 + %27 = or disjoint <16 x i32> %26, , !dbg !17 + %28 = shufflevector <16 x i32> %25, <16 x i32> %27, <16 x i32> , !dbg !17 + %29 = zext <16 x i32> %28 to <16 x i64>, !dbg !17 + %30 = insertelement <16 x i64> poison, i64 %10, i64 0, !dbg !17 + %31 = shufflevector <16 x i64> %30, <16 x i64> poison, <16 x i32> zeroinitializer, !dbg !17 + %32 = or disjoint <16 x i64> %31, %29, !dbg !17 + %33 = insertelement <16 x i32> poison, i32 %4, i64 0, !dbg !18 + %34 = shufflevector <16 x i32> %33, <16 x i32> poison, <16 x i32> zeroinitializer, !dbg !18 + %35 = sext <16 x i32> %34 to <16 x i64>, !dbg !18 + %36 = icmp slt <16 x i64> %32, %35, !dbg !18 + %37 = shl nuw nsw i32 %11, 3, !dbg !19 + %38 = and i32 %37, 120, !dbg !19 + %39 = extractelement <16 x i64> %32, i64 15, !dbg !20 + %40 = shl nuw nsw i64 %39, 7, !dbg !20 + %41 = extractelement <16 x i64> %32, i64 14, !dbg !20 + %42 = shl nuw nsw i64 %41, 7, !dbg !20 + %43 = extractelement <16 x i64> %32, i64 13, !dbg !20 + %44 = shl nuw nsw i64 %43, 7, !dbg !20 + %45 = extractelement <16 x i64> %32, i64 12, !dbg !20 + %46 = shl nuw nsw i64 %45, 7, !dbg !20 + %47 = extractelement <16 x i64> %32, i64 11, !dbg !20 + %48 = shl nuw nsw i64 %47, 7, !dbg !20 + %49 = extractelement <16 x i64> %32, i64 10, !dbg !20 + %50 = shl nuw nsw i64 %49, 7, !dbg !20 + %51 = extractelement <16 x i64> %32, i64 9, !dbg !20 + %52 = shl nuw nsw i64 %51, 7, !dbg !20 + %53 = extractelement <16 x i64> %32, i64 8, !dbg !20 + %54 = shl nuw nsw i64 %53, 7, !dbg !20 + %55 = extractelement <16 x i64> %32, i64 7, !dbg !20 + %56 = shl nuw nsw i64 %55, 7, !dbg !20 + %57 = extractelement <16 x i64> %32, i64 6, !dbg !20 + %58 = shl nuw nsw i64 %57, 7, !dbg !20 + %59 = extractelement <16 x i64> %32, i64 5, !dbg !20 + %60 = shl nuw nsw i64 %59, 7, !dbg !20 + %61 = extractelement <16 x i64> %32, i64 4, !dbg !20 + %62 = shl nuw nsw i64 %61, 7, !dbg !20 + %63 = extractelement <16 x i64> %32, i64 3, !dbg !20 + %64 = shl nuw nsw i64 %63, 7, !dbg !20 + %65 = extractelement <16 x i64> %32, i64 2, !dbg !20 + %66 = shl nuw nsw i64 %65, 7, !dbg !20 + %67 = extractelement <16 x i64> %32, i64 1, !dbg !20 + %68 = shl nuw nsw i64 %67, 7, !dbg !20 + %69 = extractelement <16 x i64> %32, i64 0, !dbg !20 + %70 = shl nuw nsw i64 %69, 7, !dbg !20 + %71 = getelementptr bfloat, ptr addrspace(1) %0, i64 %40, !dbg !21 + %72 = getelementptr bfloat, ptr addrspace(1) %0, i64 %42, !dbg !21 + %73 = getelementptr bfloat, ptr addrspace(1) %0, i64 %44, !dbg !21 + %74 = getelementptr bfloat, ptr addrspace(1) %0, i64 %46, !dbg !21 + %75 = getelementptr bfloat, ptr addrspace(1) %0, i64 %48, !dbg !21 + %76 = getelementptr bfloat, ptr addrspace(1) %0, i64 %50, !dbg !21 + %77 = getelementptr bfloat, ptr addrspace(1) %0, i64 %52, !dbg !21 + %78 = getelementptr bfloat, ptr addrspace(1) %0, i64 %54, !dbg !21 + %79 = getelementptr bfloat, ptr addrspace(1) %0, i64 %56, !dbg !21 + %80 = getelementptr bfloat, ptr addrspace(1) %0, i64 %58, !dbg !21 + %81 = getelementptr bfloat, ptr addrspace(1) %0, i64 %60, !dbg !21 + %82 = getelementptr bfloat, ptr addrspace(1) %0, i64 %62, !dbg !21 + %83 = getelementptr bfloat, ptr addrspace(1) %0, i64 %64, !dbg !21 + %84 = getelementptr bfloat, ptr addrspace(1) %0, i64 %66, !dbg !21 + %85 = getelementptr bfloat, ptr addrspace(1) %0, i64 %68, !dbg !21 + %86 = getelementptr bfloat, ptr addrspace(1) %0, i64 %70, !dbg !21 + %87 = zext nneg i32 %38 to i64, !dbg !22 + %88 = getelementptr bfloat, ptr addrspace(1) %71, i64 %87, !dbg !22 + %89 = getelementptr bfloat, ptr addrspace(1) %72, i64 %87, !dbg !22 + %90 = getelementptr bfloat, ptr addrspace(1) %73, i64 %87, !dbg !22 + %91 = getelementptr bfloat, ptr addrspace(1) %74, i64 %87, !dbg !22 + %92 = getelementptr bfloat, ptr addrspace(1) %75, i64 %87, !dbg !22 + %93 = getelementptr bfloat, ptr addrspace(1) %76, i64 %87, !dbg !22 + %94 = getelementptr bfloat, ptr addrspace(1) %77, i64 %87, !dbg !22 + %95 = getelementptr bfloat, ptr addrspace(1) %78, i64 %87, !dbg !22 + %96 = getelementptr bfloat, ptr addrspace(1) %79, i64 %87, !dbg !22 + %97 = getelementptr bfloat, ptr addrspace(1) %80, i64 %87, !dbg !22 + %98 = getelementptr bfloat, ptr addrspace(1) %81, i64 %87, !dbg !22 + %99 = getelementptr bfloat, ptr addrspace(1) %82, i64 %87, !dbg !22 + %100 = getelementptr bfloat, ptr addrspace(1) %83, i64 %87, !dbg !22 + %101 = getelementptr bfloat, ptr addrspace(1) %84, i64 %87, !dbg !22 + %102 = getelementptr bfloat, ptr addrspace(1) %85, i64 %87, !dbg !22 + %103 = getelementptr bfloat, ptr addrspace(1) %86, i64 %87, !dbg !22 + %104 = extractelement <16 x i1> %36, i64 15, !dbg !23 + %105 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %88, i1 %104) #6, !dbg !23 + %106 = extractvalue { i32, i32, i32, i32 } %105, 0, !dbg !23 + %107 = bitcast i32 %106 to <2 x bfloat>, !dbg !23 + %108 = extractvalue { i32, i32, i32, i32 } %105, 1, !dbg !23 + %109 = bitcast i32 %108 to <2 x bfloat>, !dbg !23 + %110 = extractvalue { i32, i32, i32, i32 } %105, 2, !dbg !23 + %111 = bitcast i32 %110 to <2 x bfloat>, !dbg !23 + %112 = extractvalue { i32, i32, i32, i32 } %105, 3, !dbg !23 + %113 = bitcast i32 %112 to <2 x bfloat>, !dbg !23 + %114 = extractelement <2 x bfloat> %107, i64 0, !dbg !23 + %115 = extractelement <2 x bfloat> %107, i64 1, !dbg !23 + %116 = extractelement <2 x bfloat> %109, i64 0, !dbg !23 + %117 = extractelement <2 x bfloat> %109, i64 1, !dbg !23 + %118 = extractelement <2 x bfloat> %111, i64 0, !dbg !23 + %119 = extractelement <2 x bfloat> %111, i64 1, !dbg !23 + %120 = extractelement <2 x bfloat> %113, i64 0, !dbg !23 + %121 = extractelement <2 x bfloat> %113, i64 1, !dbg !23 + %122 = extractelement <16 x i1> %36, i64 14, !dbg !23 + %123 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %89, i1 %122) #6, !dbg !23 + %124 = extractvalue { i32, i32, i32, i32 } %123, 0, !dbg !23 + %125 = bitcast i32 %124 to <2 x bfloat>, !dbg !23 + %126 = extractvalue { i32, i32, i32, i32 } %123, 1, !dbg !23 + %127 = bitcast i32 %126 to <2 x bfloat>, !dbg !23 + %128 = extractvalue { i32, i32, i32, i32 } %123, 2, !dbg !23 + %129 = bitcast i32 %128 to <2 x bfloat>, !dbg !23 + %130 = extractvalue { i32, i32, i32, i32 } %123, 3, !dbg !23 + %131 = bitcast i32 %130 to <2 x bfloat>, !dbg !23 + %132 = extractelement <2 x bfloat> %125, i64 0, !dbg !23 + %133 = extractelement <2 x bfloat> %125, i64 1, !dbg !23 + %134 = extractelement <2 x bfloat> %127, i64 0, !dbg !23 + %135 = extractelement <2 x bfloat> %127, i64 1, !dbg !23 + %136 = extractelement <2 x bfloat> %129, i64 0, !dbg !23 + %137 = extractelement <2 x bfloat> %129, i64 1, !dbg !23 + %138 = extractelement <2 x bfloat> %131, i64 0, !dbg !23 + %139 = extractelement <2 x bfloat> %131, i64 1, !dbg !23 + %140 = extractelement <16 x i1> %36, i64 13, !dbg !23 + %141 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %90, i1 %140) #6, !dbg !23 + %142 = extractvalue { i32, i32, i32, i32 } %141, 0, !dbg !23 + %143 = bitcast i32 %142 to <2 x bfloat>, !dbg !23 + %144 = extractvalue { i32, i32, i32, i32 } %141, 1, !dbg !23 + %145 = bitcast i32 %144 to <2 x bfloat>, !dbg !23 + %146 = extractvalue { i32, i32, i32, i32 } %141, 2, !dbg !23 + %147 = bitcast i32 %146 to <2 x bfloat>, !dbg !23 + %148 = extractvalue { i32, i32, i32, i32 } %141, 3, !dbg !23 + %149 = bitcast i32 %148 to <2 x bfloat>, !dbg !23 + %150 = extractelement <2 x bfloat> %143, i64 0, !dbg !23 + %151 = extractelement <2 x bfloat> %143, i64 1, !dbg !23 + %152 = extractelement <2 x bfloat> %145, i64 0, !dbg !23 + %153 = extractelement <2 x bfloat> %145, i64 1, !dbg !23 + %154 = extractelement <2 x bfloat> %147, i64 0, !dbg !23 + %155 = extractelement <2 x bfloat> %147, i64 1, !dbg !23 + %156 = extractelement <2 x bfloat> %149, i64 0, !dbg !23 + %157 = extractelement <2 x bfloat> %149, i64 1, !dbg !23 + %158 = extractelement <16 x i1> %36, i64 12, !dbg !23 + %159 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %91, i1 %158) #6, !dbg !23 + %160 = extractvalue { i32, i32, i32, i32 } %159, 0, !dbg !23 + %161 = bitcast i32 %160 to <2 x bfloat>, !dbg !23 + %162 = extractvalue { i32, i32, i32, i32 } %159, 1, !dbg !23 + %163 = bitcast i32 %162 to <2 x bfloat>, !dbg !23 + %164 = extractvalue { i32, i32, i32, i32 } %159, 2, !dbg !23 + %165 = bitcast i32 %164 to <2 x bfloat>, !dbg !23 + %166 = extractvalue { i32, i32, i32, i32 } %159, 3, !dbg !23 + %167 = bitcast i32 %166 to <2 x bfloat>, !dbg !23 + %168 = extractelement <2 x bfloat> %161, i64 0, !dbg !23 + %169 = extractelement <2 x bfloat> %161, i64 1, !dbg !23 + %170 = extractelement <2 x bfloat> %163, i64 0, !dbg !23 + %171 = extractelement <2 x bfloat> %163, i64 1, !dbg !23 + %172 = extractelement <2 x bfloat> %165, i64 0, !dbg !23 + %173 = extractelement <2 x bfloat> %165, i64 1, !dbg !23 + %174 = extractelement <2 x bfloat> %167, i64 0, !dbg !23 + %175 = extractelement <2 x bfloat> %167, i64 1, !dbg !23 + %176 = extractelement <16 x i1> %36, i64 11, !dbg !23 + %177 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %92, i1 %176) #6, !dbg !23 + %178 = extractvalue { i32, i32, i32, i32 } %177, 0, !dbg !23 + %179 = bitcast i32 %178 to <2 x bfloat>, !dbg !23 + %180 = extractvalue { i32, i32, i32, i32 } %177, 1, !dbg !23 + %181 = bitcast i32 %180 to <2 x bfloat>, !dbg !23 + %182 = extractvalue { i32, i32, i32, i32 } %177, 2, !dbg !23 + %183 = bitcast i32 %182 to <2 x bfloat>, !dbg !23 + %184 = extractvalue { i32, i32, i32, i32 } %177, 3, !dbg !23 + %185 = bitcast i32 %184 to <2 x bfloat>, !dbg !23 + %186 = extractelement <2 x bfloat> %179, i64 0, !dbg !23 + %187 = extractelement <2 x bfloat> %179, i64 1, !dbg !23 + %188 = extractelement <2 x bfloat> %181, i64 0, !dbg !23 + %189 = extractelement <2 x bfloat> %181, i64 1, !dbg !23 + %190 = extractelement <2 x bfloat> %183, i64 0, !dbg !23 + %191 = extractelement <2 x bfloat> %183, i64 1, !dbg !23 + %192 = extractelement <2 x bfloat> %185, i64 0, !dbg !23 + %193 = extractelement <2 x bfloat> %185, i64 1, !dbg !23 + %194 = extractelement <16 x i1> %36, i64 10, !dbg !23 + %195 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %93, i1 %194) #6, !dbg !23 + %196 = extractvalue { i32, i32, i32, i32 } %195, 0, !dbg !23 + %197 = bitcast i32 %196 to <2 x bfloat>, !dbg !23 + %198 = extractvalue { i32, i32, i32, i32 } %195, 1, !dbg !23 + %199 = bitcast i32 %198 to <2 x bfloat>, !dbg !23 + %200 = extractvalue { i32, i32, i32, i32 } %195, 2, !dbg !23 + %201 = bitcast i32 %200 to <2 x bfloat>, !dbg !23 + %202 = extractvalue { i32, i32, i32, i32 } %195, 3, !dbg !23 + %203 = bitcast i32 %202 to <2 x bfloat>, !dbg !23 + %204 = extractelement <2 x bfloat> %197, i64 0, !dbg !23 + %205 = extractelement <2 x bfloat> %197, i64 1, !dbg !23 + %206 = extractelement <2 x bfloat> %199, i64 0, !dbg !23 + %207 = extractelement <2 x bfloat> %199, i64 1, !dbg !23 + %208 = extractelement <2 x bfloat> %201, i64 0, !dbg !23 + %209 = extractelement <2 x bfloat> %201, i64 1, !dbg !23 + %210 = extractelement <2 x bfloat> %203, i64 0, !dbg !23 + %211 = extractelement <2 x bfloat> %203, i64 1, !dbg !23 + %212 = extractelement <16 x i1> %36, i64 9, !dbg !23 + %213 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %94, i1 %212) #6, !dbg !23 + %214 = extractvalue { i32, i32, i32, i32 } %213, 0, !dbg !23 + %215 = bitcast i32 %214 to <2 x bfloat>, !dbg !23 + %216 = extractvalue { i32, i32, i32, i32 } %213, 1, !dbg !23 + %217 = bitcast i32 %216 to <2 x bfloat>, !dbg !23 + %218 = extractvalue { i32, i32, i32, i32 } %213, 2, !dbg !23 + %219 = bitcast i32 %218 to <2 x bfloat>, !dbg !23 + %220 = extractvalue { i32, i32, i32, i32 } %213, 3, !dbg !23 + %221 = bitcast i32 %220 to <2 x bfloat>, !dbg !23 + %222 = extractelement <2 x bfloat> %215, i64 0, !dbg !23 + %223 = extractelement <2 x bfloat> %215, i64 1, !dbg !23 + %224 = extractelement <2 x bfloat> %217, i64 0, !dbg !23 + %225 = extractelement <2 x bfloat> %217, i64 1, !dbg !23 + %226 = extractelement <2 x bfloat> %219, i64 0, !dbg !23 + %227 = extractelement <2 x bfloat> %219, i64 1, !dbg !23 + %228 = extractelement <2 x bfloat> %221, i64 0, !dbg !23 + %229 = extractelement <2 x bfloat> %221, i64 1, !dbg !23 + %230 = extractelement <16 x i1> %36, i64 8, !dbg !23 + %231 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %95, i1 %230) #6, !dbg !23 + %232 = extractvalue { i32, i32, i32, i32 } %231, 0, !dbg !23 + %233 = bitcast i32 %232 to <2 x bfloat>, !dbg !23 + %234 = extractvalue { i32, i32, i32, i32 } %231, 1, !dbg !23 + %235 = bitcast i32 %234 to <2 x bfloat>, !dbg !23 + %236 = extractvalue { i32, i32, i32, i32 } %231, 2, !dbg !23 + %237 = bitcast i32 %236 to <2 x bfloat>, !dbg !23 + %238 = extractvalue { i32, i32, i32, i32 } %231, 3, !dbg !23 + %239 = bitcast i32 %238 to <2 x bfloat>, !dbg !23 + %240 = extractelement <2 x bfloat> %233, i64 0, !dbg !23 + %241 = extractelement <2 x bfloat> %233, i64 1, !dbg !23 + %242 = extractelement <2 x bfloat> %235, i64 0, !dbg !23 + %243 = extractelement <2 x bfloat> %235, i64 1, !dbg !23 + %244 = extractelement <2 x bfloat> %237, i64 0, !dbg !23 + %245 = extractelement <2 x bfloat> %237, i64 1, !dbg !23 + %246 = extractelement <2 x bfloat> %239, i64 0, !dbg !23 + %247 = extractelement <2 x bfloat> %239, i64 1, !dbg !23 + %248 = extractelement <16 x i1> %36, i64 7, !dbg !23 + %249 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %96, i1 %248) #6, !dbg !23 + %250 = extractvalue { i32, i32, i32, i32 } %249, 0, !dbg !23 + %251 = bitcast i32 %250 to <2 x bfloat>, !dbg !23 + %252 = extractvalue { i32, i32, i32, i32 } %249, 1, !dbg !23 + %253 = bitcast i32 %252 to <2 x bfloat>, !dbg !23 + %254 = extractvalue { i32, i32, i32, i32 } %249, 2, !dbg !23 + %255 = bitcast i32 %254 to <2 x bfloat>, !dbg !23 + %256 = extractvalue { i32, i32, i32, i32 } %249, 3, !dbg !23 + %257 = bitcast i32 %256 to <2 x bfloat>, !dbg !23 + %258 = extractelement <2 x bfloat> %251, i64 0, !dbg !23 + %259 = extractelement <2 x bfloat> %251, i64 1, !dbg !23 + %260 = extractelement <2 x bfloat> %253, i64 0, !dbg !23 + %261 = extractelement <2 x bfloat> %253, i64 1, !dbg !23 + %262 = extractelement <2 x bfloat> %255, i64 0, !dbg !23 + %263 = extractelement <2 x bfloat> %255, i64 1, !dbg !23 + %264 = extractelement <2 x bfloat> %257, i64 0, !dbg !23 + %265 = extractelement <2 x bfloat> %257, i64 1, !dbg !23 + %266 = extractelement <16 x i1> %36, i64 6, !dbg !23 + %267 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %97, i1 %266) #6, !dbg !23 + %268 = extractvalue { i32, i32, i32, i32 } %267, 0, !dbg !23 + %269 = bitcast i32 %268 to <2 x bfloat>, !dbg !23 + %270 = extractvalue { i32, i32, i32, i32 } %267, 1, !dbg !23 + %271 = bitcast i32 %270 to <2 x bfloat>, !dbg !23 + %272 = extractvalue { i32, i32, i32, i32 } %267, 2, !dbg !23 + %273 = bitcast i32 %272 to <2 x bfloat>, !dbg !23 + %274 = extractvalue { i32, i32, i32, i32 } %267, 3, !dbg !23 + %275 = bitcast i32 %274 to <2 x bfloat>, !dbg !23 + %276 = extractelement <2 x bfloat> %269, i64 0, !dbg !23 + %277 = extractelement <2 x bfloat> %269, i64 1, !dbg !23 + %278 = extractelement <2 x bfloat> %271, i64 0, !dbg !23 + %279 = extractelement <2 x bfloat> %271, i64 1, !dbg !23 + %280 = extractelement <2 x bfloat> %273, i64 0, !dbg !23 + %281 = extractelement <2 x bfloat> %273, i64 1, !dbg !23 + %282 = extractelement <2 x bfloat> %275, i64 0, !dbg !23 + %283 = extractelement <2 x bfloat> %275, i64 1, !dbg !23 + %284 = extractelement <16 x i1> %36, i64 5, !dbg !23 + %285 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %98, i1 %284) #6, !dbg !23 + %286 = extractvalue { i32, i32, i32, i32 } %285, 0, !dbg !23 + %287 = bitcast i32 %286 to <2 x bfloat>, !dbg !23 + %288 = extractvalue { i32, i32, i32, i32 } %285, 1, !dbg !23 + %289 = bitcast i32 %288 to <2 x bfloat>, !dbg !23 + %290 = extractvalue { i32, i32, i32, i32 } %285, 2, !dbg !23 + %291 = bitcast i32 %290 to <2 x bfloat>, !dbg !23 + %292 = extractvalue { i32, i32, i32, i32 } %285, 3, !dbg !23 + %293 = bitcast i32 %292 to <2 x bfloat>, !dbg !23 + %294 = extractelement <2 x bfloat> %287, i64 0, !dbg !23 + %295 = extractelement <2 x bfloat> %287, i64 1, !dbg !23 + %296 = extractelement <2 x bfloat> %289, i64 0, !dbg !23 + %297 = extractelement <2 x bfloat> %289, i64 1, !dbg !23 + %298 = extractelement <2 x bfloat> %291, i64 0, !dbg !23 + %299 = extractelement <2 x bfloat> %291, i64 1, !dbg !23 + %300 = extractelement <2 x bfloat> %293, i64 0, !dbg !23 + %301 = extractelement <2 x bfloat> %293, i64 1, !dbg !23 + %302 = extractelement <16 x i1> %36, i64 4, !dbg !23 + %303 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %99, i1 %302) #6, !dbg !23 + %304 = extractvalue { i32, i32, i32, i32 } %303, 0, !dbg !23 + %305 = bitcast i32 %304 to <2 x bfloat>, !dbg !23 + %306 = extractvalue { i32, i32, i32, i32 } %303, 1, !dbg !23 + %307 = bitcast i32 %306 to <2 x bfloat>, !dbg !23 + %308 = extractvalue { i32, i32, i32, i32 } %303, 2, !dbg !23 + %309 = bitcast i32 %308 to <2 x bfloat>, !dbg !23 + %310 = extractvalue { i32, i32, i32, i32 } %303, 3, !dbg !23 + %311 = bitcast i32 %310 to <2 x bfloat>, !dbg !23 + %312 = extractelement <2 x bfloat> %305, i64 0, !dbg !23 + %313 = extractelement <2 x bfloat> %305, i64 1, !dbg !23 + %314 = extractelement <2 x bfloat> %307, i64 0, !dbg !23 + %315 = extractelement <2 x bfloat> %307, i64 1, !dbg !23 + %316 = extractelement <2 x bfloat> %309, i64 0, !dbg !23 + %317 = extractelement <2 x bfloat> %309, i64 1, !dbg !23 + %318 = extractelement <2 x bfloat> %311, i64 0, !dbg !23 + %319 = extractelement <2 x bfloat> %311, i64 1, !dbg !23 + %320 = extractelement <16 x i1> %36, i64 3, !dbg !23 + %321 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %100, i1 %320) #6, !dbg !23 + %322 = extractvalue { i32, i32, i32, i32 } %321, 0, !dbg !23 + %323 = bitcast i32 %322 to <2 x bfloat>, !dbg !23 + %324 = extractvalue { i32, i32, i32, i32 } %321, 1, !dbg !23 + %325 = bitcast i32 %324 to <2 x bfloat>, !dbg !23 + %326 = extractvalue { i32, i32, i32, i32 } %321, 2, !dbg !23 + %327 = bitcast i32 %326 to <2 x bfloat>, !dbg !23 + %328 = extractvalue { i32, i32, i32, i32 } %321, 3, !dbg !23 + %329 = bitcast i32 %328 to <2 x bfloat>, !dbg !23 + %330 = extractelement <2 x bfloat> %323, i64 0, !dbg !23 + %331 = extractelement <2 x bfloat> %323, i64 1, !dbg !23 + %332 = extractelement <2 x bfloat> %325, i64 0, !dbg !23 + %333 = extractelement <2 x bfloat> %325, i64 1, !dbg !23 + %334 = extractelement <2 x bfloat> %327, i64 0, !dbg !23 + %335 = extractelement <2 x bfloat> %327, i64 1, !dbg !23 + %336 = extractelement <2 x bfloat> %329, i64 0, !dbg !23 + %337 = extractelement <2 x bfloat> %329, i64 1, !dbg !23 + %338 = extractelement <16 x i1> %36, i64 2, !dbg !23 + %339 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %101, i1 %338) #6, !dbg !23 + %340 = extractvalue { i32, i32, i32, i32 } %339, 0, !dbg !23 + %341 = bitcast i32 %340 to <2 x bfloat>, !dbg !23 + %342 = extractvalue { i32, i32, i32, i32 } %339, 1, !dbg !23 + %343 = bitcast i32 %342 to <2 x bfloat>, !dbg !23 + %344 = extractvalue { i32, i32, i32, i32 } %339, 2, !dbg !23 + %345 = bitcast i32 %344 to <2 x bfloat>, !dbg !23 + %346 = extractvalue { i32, i32, i32, i32 } %339, 3, !dbg !23 + %347 = bitcast i32 %346 to <2 x bfloat>, !dbg !23 + %348 = extractelement <2 x bfloat> %341, i64 0, !dbg !23 + %349 = extractelement <2 x bfloat> %341, i64 1, !dbg !23 + %350 = extractelement <2 x bfloat> %343, i64 0, !dbg !23 + %351 = extractelement <2 x bfloat> %343, i64 1, !dbg !23 + %352 = extractelement <2 x bfloat> %345, i64 0, !dbg !23 + %353 = extractelement <2 x bfloat> %345, i64 1, !dbg !23 + %354 = extractelement <2 x bfloat> %347, i64 0, !dbg !23 + %355 = extractelement <2 x bfloat> %347, i64 1, !dbg !23 + %356 = extractelement <16 x i1> %36, i64 1, !dbg !23 + %357 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %102, i1 %356) #6, !dbg !23 + %358 = extractvalue { i32, i32, i32, i32 } %357, 0, !dbg !23 + %359 = bitcast i32 %358 to <2 x bfloat>, !dbg !23 + %360 = extractvalue { i32, i32, i32, i32 } %357, 1, !dbg !23 + %361 = bitcast i32 %360 to <2 x bfloat>, !dbg !23 + %362 = extractvalue { i32, i32, i32, i32 } %357, 2, !dbg !23 + %363 = bitcast i32 %362 to <2 x bfloat>, !dbg !23 + %364 = extractvalue { i32, i32, i32, i32 } %357, 3, !dbg !23 + %365 = bitcast i32 %364 to <2 x bfloat>, !dbg !23 + %366 = extractelement <2 x bfloat> %359, i64 0, !dbg !23 + %367 = extractelement <2 x bfloat> %359, i64 1, !dbg !23 + %368 = extractelement <2 x bfloat> %361, i64 0, !dbg !23 + %369 = extractelement <2 x bfloat> %361, i64 1, !dbg !23 + %370 = extractelement <2 x bfloat> %363, i64 0, !dbg !23 + %371 = extractelement <2 x bfloat> %363, i64 1, !dbg !23 + %372 = extractelement <2 x bfloat> %365, i64 0, !dbg !23 + %373 = extractelement <2 x bfloat> %365, i64 1, !dbg !23 + %374 = extractelement <16 x i1> %36, i64 0, !dbg !23 + %375 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %103, i1 %374) #6, !dbg !23 + %376 = extractvalue { i32, i32, i32, i32 } %375, 0, !dbg !23 + %377 = bitcast i32 %376 to <2 x bfloat>, !dbg !23 + %378 = extractvalue { i32, i32, i32, i32 } %375, 1, !dbg !23 + %379 = bitcast i32 %378 to <2 x bfloat>, !dbg !23 + %380 = extractvalue { i32, i32, i32, i32 } %375, 2, !dbg !23 + %381 = bitcast i32 %380 to <2 x bfloat>, !dbg !23 + %382 = extractvalue { i32, i32, i32, i32 } %375, 3, !dbg !23 + %383 = bitcast i32 %382 to <2 x bfloat>, !dbg !23 + %384 = extractelement <2 x bfloat> %377, i64 0, !dbg !23 + %385 = extractelement <2 x bfloat> %377, i64 1, !dbg !23 + %386 = extractelement <2 x bfloat> %379, i64 0, !dbg !23 + %387 = extractelement <2 x bfloat> %379, i64 1, !dbg !23 + %388 = extractelement <2 x bfloat> %381, i64 0, !dbg !23 + %389 = extractelement <2 x bfloat> %381, i64 1, !dbg !23 + %390 = extractelement <2 x bfloat> %383, i64 0, !dbg !23 + %391 = extractelement <2 x bfloat> %383, i64 1, !dbg !23 + %392 = fpext bfloat %114 to float, !dbg !24 + %393 = fpext bfloat %115 to float, !dbg !24 + %394 = fpext bfloat %116 to float, !dbg !24 + %395 = fpext bfloat %117 to float, !dbg !24 + %396 = fpext bfloat %118 to float, !dbg !24 + %397 = fpext bfloat %119 to float, !dbg !24 + %398 = fpext bfloat %120 to float, !dbg !24 + %399 = fpext bfloat %121 to float, !dbg !24 + %400 = fpext bfloat %132 to float, !dbg !24 + %401 = fpext bfloat %133 to float, !dbg !24 + %402 = fpext bfloat %134 to float, !dbg !24 + %403 = fpext bfloat %135 to float, !dbg !24 + %404 = fpext bfloat %136 to float, !dbg !24 + %405 = fpext bfloat %137 to float, !dbg !24 + %406 = fpext bfloat %138 to float, !dbg !24 + %407 = fpext bfloat %139 to float, !dbg !24 + %408 = fpext bfloat %150 to float, !dbg !24 + %409 = fpext bfloat %151 to float, !dbg !24 + %410 = fpext bfloat %152 to float, !dbg !24 + %411 = fpext bfloat %153 to float, !dbg !24 + %412 = fpext bfloat %154 to float, !dbg !24 + %413 = fpext bfloat %155 to float, !dbg !24 + %414 = fpext bfloat %156 to float, !dbg !24 + %415 = fpext bfloat %157 to float, !dbg !24 + %416 = fpext bfloat %168 to float, !dbg !24 + %417 = fpext bfloat %169 to float, !dbg !24 + %418 = fpext bfloat %170 to float, !dbg !24 + %419 = fpext bfloat %171 to float, !dbg !24 + %420 = fpext bfloat %172 to float, !dbg !24 + %421 = fpext bfloat %173 to float, !dbg !24 + %422 = fpext bfloat %174 to float, !dbg !24 + %423 = fpext bfloat %175 to float, !dbg !24 + %424 = fpext bfloat %186 to float, !dbg !24 + %425 = fpext bfloat %187 to float, !dbg !24 + %426 = fpext bfloat %188 to float, !dbg !24 + %427 = fpext bfloat %189 to float, !dbg !24 + %428 = fpext bfloat %190 to float, !dbg !24 + %429 = fpext bfloat %191 to float, !dbg !24 + %430 = fpext bfloat %192 to float, !dbg !24 + %431 = fpext bfloat %193 to float, !dbg !24 + %432 = fpext bfloat %204 to float, !dbg !24 + %433 = fpext bfloat %205 to float, !dbg !24 + %434 = fpext bfloat %206 to float, !dbg !24 + %435 = fpext bfloat %207 to float, !dbg !24 + %436 = fpext bfloat %208 to float, !dbg !24 + %437 = fpext bfloat %209 to float, !dbg !24 + %438 = fpext bfloat %210 to float, !dbg !24 + %439 = fpext bfloat %211 to float, !dbg !24 + %440 = fpext bfloat %222 to float, !dbg !24 + %441 = fpext bfloat %223 to float, !dbg !24 + %442 = fpext bfloat %224 to float, !dbg !24 + %443 = fpext bfloat %225 to float, !dbg !24 + %444 = fpext bfloat %226 to float, !dbg !24 + %445 = fpext bfloat %227 to float, !dbg !24 + %446 = fpext bfloat %228 to float, !dbg !24 + %447 = fpext bfloat %229 to float, !dbg !24 + %448 = fpext bfloat %240 to float, !dbg !24 + %449 = fpext bfloat %241 to float, !dbg !24 + %450 = fpext bfloat %242 to float, !dbg !24 + %451 = fpext bfloat %243 to float, !dbg !24 + %452 = fpext bfloat %244 to float, !dbg !24 + %453 = fpext bfloat %245 to float, !dbg !24 + %454 = fpext bfloat %246 to float, !dbg !24 + %455 = fpext bfloat %247 to float, !dbg !24 + %456 = fpext bfloat %258 to float, !dbg !24 + %457 = fpext bfloat %259 to float, !dbg !24 + %458 = fpext bfloat %260 to float, !dbg !24 + %459 = fpext bfloat %261 to float, !dbg !24 + %460 = fpext bfloat %262 to float, !dbg !24 + %461 = fpext bfloat %263 to float, !dbg !24 + %462 = fpext bfloat %264 to float, !dbg !24 + %463 = fpext bfloat %265 to float, !dbg !24 + %464 = fpext bfloat %276 to float, !dbg !24 + %465 = fpext bfloat %277 to float, !dbg !24 + %466 = fpext bfloat %278 to float, !dbg !24 + %467 = fpext bfloat %279 to float, !dbg !24 + %468 = fpext bfloat %280 to float, !dbg !24 + %469 = fpext bfloat %281 to float, !dbg !24 + %470 = fpext bfloat %282 to float, !dbg !24 + %471 = fpext bfloat %283 to float, !dbg !24 + %472 = fpext bfloat %294 to float, !dbg !24 + %473 = fpext bfloat %295 to float, !dbg !24 + %474 = fpext bfloat %296 to float, !dbg !24 + %475 = fpext bfloat %297 to float, !dbg !24 + %476 = fpext bfloat %298 to float, !dbg !24 + %477 = fpext bfloat %299 to float, !dbg !24 + %478 = fpext bfloat %300 to float, !dbg !24 + %479 = fpext bfloat %301 to float, !dbg !24 + %480 = fpext bfloat %312 to float, !dbg !24 + %481 = fpext bfloat %313 to float, !dbg !24 + %482 = fpext bfloat %314 to float, !dbg !24 + %483 = fpext bfloat %315 to float, !dbg !24 + %484 = fpext bfloat %316 to float, !dbg !24 + %485 = fpext bfloat %317 to float, !dbg !24 + %486 = fpext bfloat %318 to float, !dbg !24 + %487 = fpext bfloat %319 to float, !dbg !24 + %488 = fpext bfloat %330 to float, !dbg !24 + %489 = fpext bfloat %331 to float, !dbg !24 + %490 = fpext bfloat %332 to float, !dbg !24 + %491 = fpext bfloat %333 to float, !dbg !24 + %492 = fpext bfloat %334 to float, !dbg !24 + %493 = fpext bfloat %335 to float, !dbg !24 + %494 = fpext bfloat %336 to float, !dbg !24 + %495 = fpext bfloat %337 to float, !dbg !24 + %496 = fpext bfloat %348 to float, !dbg !24 + %497 = fpext bfloat %349 to float, !dbg !24 + %498 = fpext bfloat %350 to float, !dbg !24 + %499 = fpext bfloat %351 to float, !dbg !24 + %500 = fpext bfloat %352 to float, !dbg !24 + %501 = fpext bfloat %353 to float, !dbg !24 + %502 = fpext bfloat %354 to float, !dbg !24 + %503 = fpext bfloat %355 to float, !dbg !24 + %504 = fpext bfloat %366 to float, !dbg !24 + %505 = fpext bfloat %367 to float, !dbg !24 + %506 = fpext bfloat %368 to float, !dbg !24 + %507 = fpext bfloat %369 to float, !dbg !24 + %508 = fpext bfloat %370 to float, !dbg !24 + %509 = fpext bfloat %371 to float, !dbg !24 + %510 = fpext bfloat %372 to float, !dbg !24 + %511 = fpext bfloat %373 to float, !dbg !24 + %512 = fpext bfloat %384 to float, !dbg !24 + %513 = fpext bfloat %385 to float, !dbg !24 + %514 = fpext bfloat %386 to float, !dbg !24 + %515 = fpext bfloat %387 to float, !dbg !24 + %516 = fpext bfloat %388 to float, !dbg !24 + %517 = fpext bfloat %389 to float, !dbg !24 + %518 = fpext bfloat %390 to float, !dbg !24 + %519 = fpext bfloat %391 to float, !dbg !24 + %520 = fmul float %392, %392, !dbg !25 + %521 = fmul float %393, %393, !dbg !25 + %522 = fmul float %394, %394, !dbg !25 + %523 = fmul float %395, %395, !dbg !25 + %524 = fmul float %396, %396, !dbg !25 + %525 = fmul float %397, %397, !dbg !25 + %526 = fmul float %398, %398, !dbg !25 + %527 = fmul float %399, %399, !dbg !25 + %528 = fmul float %400, %400, !dbg !25 + %529 = fmul float %401, %401, !dbg !25 + %530 = fmul float %402, %402, !dbg !25 + %531 = fmul float %403, %403, !dbg !25 + %532 = fmul float %404, %404, !dbg !25 + %533 = fmul float %405, %405, !dbg !25 + %534 = fmul float %406, %406, !dbg !25 + %535 = fmul float %407, %407, !dbg !25 + %536 = fmul float %408, %408, !dbg !25 + %537 = fmul float %409, %409, !dbg !25 + %538 = fmul float %410, %410, !dbg !25 + %539 = fmul float %411, %411, !dbg !25 + %540 = fmul float %412, %412, !dbg !25 + %541 = fmul float %413, %413, !dbg !25 + %542 = fmul float %414, %414, !dbg !25 + %543 = fmul float %415, %415, !dbg !25 + %544 = fmul float %416, %416, !dbg !25 + %545 = fmul float %417, %417, !dbg !25 + %546 = fmul float %418, %418, !dbg !25 + %547 = fmul float %419, %419, !dbg !25 + %548 = fmul float %420, %420, !dbg !25 + %549 = fmul float %421, %421, !dbg !25 + %550 = fmul float %422, %422, !dbg !25 + %551 = fmul float %423, %423, !dbg !25 + %552 = fmul float %424, %424, !dbg !25 + %553 = fmul float %425, %425, !dbg !25 + %554 = fmul float %426, %426, !dbg !25 + %555 = fmul float %427, %427, !dbg !25 + %556 = fmul float %428, %428, !dbg !25 + %557 = fmul float %429, %429, !dbg !25 + %558 = fmul float %430, %430, !dbg !25 + %559 = fmul float %431, %431, !dbg !25 + %560 = fmul float %432, %432, !dbg !25 + %561 = fmul float %433, %433, !dbg !25 + %562 = fmul float %434, %434, !dbg !25 + %563 = fmul float %435, %435, !dbg !25 + %564 = fmul float %436, %436, !dbg !25 + %565 = fmul float %437, %437, !dbg !25 + %566 = fmul float %438, %438, !dbg !25 + %567 = fmul float %439, %439, !dbg !25 + %568 = fmul float %440, %440, !dbg !25 + %569 = fmul float %441, %441, !dbg !25 + %570 = fmul float %442, %442, !dbg !25 + %571 = fmul float %443, %443, !dbg !25 + %572 = fmul float %444, %444, !dbg !25 + %573 = fmul float %445, %445, !dbg !25 + %574 = fmul float %446, %446, !dbg !25 + %575 = fmul float %447, %447, !dbg !25 + %576 = fmul float %448, %448, !dbg !25 + %577 = fmul float %449, %449, !dbg !25 + %578 = fmul float %450, %450, !dbg !25 + %579 = fmul float %451, %451, !dbg !25 + %580 = fmul float %452, %452, !dbg !25 + %581 = fmul float %453, %453, !dbg !25 + %582 = fmul float %454, %454, !dbg !25 + %583 = fmul float %455, %455, !dbg !25 + %584 = fmul float %456, %456, !dbg !25 + %585 = fmul float %457, %457, !dbg !25 + %586 = fmul float %458, %458, !dbg !25 + %587 = fmul float %459, %459, !dbg !25 + %588 = fmul float %460, %460, !dbg !25 + %589 = fmul float %461, %461, !dbg !25 + %590 = fmul float %462, %462, !dbg !25 + %591 = fmul float %463, %463, !dbg !25 + %592 = fmul float %464, %464, !dbg !25 + %593 = fmul float %465, %465, !dbg !25 + %594 = fmul float %466, %466, !dbg !25 + %595 = fmul float %467, %467, !dbg !25 + %596 = fmul float %468, %468, !dbg !25 + %597 = fmul float %469, %469, !dbg !25 + %598 = fmul float %470, %470, !dbg !25 + %599 = fmul float %471, %471, !dbg !25 + %600 = fmul float %472, %472, !dbg !25 + %601 = fmul float %473, %473, !dbg !25 + %602 = fmul float %474, %474, !dbg !25 + %603 = fmul float %475, %475, !dbg !25 + %604 = fmul float %476, %476, !dbg !25 + %605 = fmul float %477, %477, !dbg !25 + %606 = fmul float %478, %478, !dbg !25 + %607 = fmul float %479, %479, !dbg !25 + %608 = fmul float %480, %480, !dbg !25 + %609 = fmul float %481, %481, !dbg !25 + %610 = fmul float %482, %482, !dbg !25 + %611 = fmul float %483, %483, !dbg !25 + %612 = fmul float %484, %484, !dbg !25 + %613 = fmul float %485, %485, !dbg !25 + %614 = fmul float %486, %486, !dbg !25 + %615 = fmul float %487, %487, !dbg !25 + %616 = fmul float %488, %488, !dbg !25 + %617 = fmul float %489, %489, !dbg !25 + %618 = fmul float %490, %490, !dbg !25 + %619 = fmul float %491, %491, !dbg !25 + %620 = fmul float %492, %492, !dbg !25 + %621 = fmul float %493, %493, !dbg !25 + %622 = fmul float %494, %494, !dbg !25 + %623 = fmul float %495, %495, !dbg !25 + %624 = fmul float %496, %496, !dbg !25 + %625 = fmul float %497, %497, !dbg !25 + %626 = fmul float %498, %498, !dbg !25 + %627 = fmul float %499, %499, !dbg !25 + %628 = fmul float %500, %500, !dbg !25 + %629 = fmul float %501, %501, !dbg !25 + %630 = fmul float %502, %502, !dbg !25 + %631 = fmul float %503, %503, !dbg !25 + %632 = fmul float %504, %504, !dbg !25 + %633 = fmul float %505, %505, !dbg !25 + %634 = fmul float %506, %506, !dbg !25 + %635 = fmul float %507, %507, !dbg !25 + %636 = fmul float %508, %508, !dbg !25 + %637 = fmul float %509, %509, !dbg !25 + %638 = fmul float %510, %510, !dbg !25 + %639 = fmul float %511, %511, !dbg !25 + %640 = fmul float %512, %512, !dbg !25 + %641 = fmul float %513, %513, !dbg !25 + %642 = fmul float %514, %514, !dbg !25 + %643 = fmul float %515, %515, !dbg !25 + %644 = fmul float %516, %516, !dbg !25 + %645 = fmul float %517, %517, !dbg !25 + %646 = fmul float %518, %518, !dbg !25 + %647 = fmul float %519, %519, !dbg !25 + %648 = fadd float %520, %521, !dbg !26 + %649 = fadd float %522, %648, !dbg !26 + %650 = fadd float %523, %649, !dbg !26 + %651 = fadd float %524, %650, !dbg !26 + %652 = fadd float %525, %651, !dbg !26 + %653 = fadd float %526, %652, !dbg !26 + %654 = fadd float %527, %653, !dbg !26 + %655 = fadd float %528, %529, !dbg !26 + %656 = fadd float %530, %655, !dbg !26 + %657 = fadd float %531, %656, !dbg !26 + %658 = fadd float %532, %657, !dbg !26 + %659 = fadd float %533, %658, !dbg !26 + %660 = fadd float %534, %659, !dbg !26 + %661 = fadd float %535, %660, !dbg !26 + %662 = fadd float %536, %537, !dbg !26 + %663 = fadd float %538, %662, !dbg !26 + %664 = fadd float %539, %663, !dbg !26 + %665 = fadd float %540, %664, !dbg !26 + %666 = fadd float %541, %665, !dbg !26 + %667 = fadd float %542, %666, !dbg !26 + %668 = fadd float %543, %667, !dbg !26 + %669 = fadd float %544, %545, !dbg !26 + %670 = fadd float %546, %669, !dbg !26 + %671 = fadd float %547, %670, !dbg !26 + %672 = fadd float %548, %671, !dbg !26 + %673 = fadd float %549, %672, !dbg !26 + %674 = fadd float %550, %673, !dbg !26 + %675 = fadd float %551, %674, !dbg !26 + %676 = fadd float %552, %553, !dbg !26 + %677 = fadd float %554, %676, !dbg !26 + %678 = fadd float %555, %677, !dbg !26 + %679 = fadd float %556, %678, !dbg !26 + %680 = fadd float %557, %679, !dbg !26 + %681 = fadd float %558, %680, !dbg !26 + %682 = fadd float %559, %681, !dbg !26 + %683 = fadd float %560, %561, !dbg !26 + %684 = fadd float %562, %683, !dbg !26 + %685 = fadd float %563, %684, !dbg !26 + %686 = fadd float %564, %685, !dbg !26 + %687 = fadd float %565, %686, !dbg !26 + %688 = fadd float %566, %687, !dbg !26 + %689 = fadd float %567, %688, !dbg !26 + %690 = fadd float %568, %569, !dbg !26 + %691 = fadd float %570, %690, !dbg !26 + %692 = fadd float %571, %691, !dbg !26 + %693 = fadd float %572, %692, !dbg !26 + %694 = fadd float %573, %693, !dbg !26 + %695 = fadd float %574, %694, !dbg !26 + %696 = fadd float %575, %695, !dbg !26 + %697 = fadd float %576, %577, !dbg !26 + %698 = fadd float %578, %697, !dbg !26 + %699 = fadd float %579, %698, !dbg !26 + %700 = fadd float %580, %699, !dbg !26 + %701 = fadd float %581, %700, !dbg !26 + %702 = fadd float %582, %701, !dbg !26 + %703 = fadd float %583, %702, !dbg !26 + %704 = fadd float %584, %585, !dbg !26 + %705 = fadd float %586, %704, !dbg !26 + %706 = fadd float %587, %705, !dbg !26 + %707 = fadd float %588, %706, !dbg !26 + %708 = fadd float %589, %707, !dbg !26 + %709 = fadd float %590, %708, !dbg !26 + %710 = fadd float %591, %709, !dbg !26 + %711 = fadd float %592, %593, !dbg !26 + %712 = fadd float %594, %711, !dbg !26 + %713 = fadd float %595, %712, !dbg !26 + %714 = fadd float %596, %713, !dbg !26 + %715 = fadd float %597, %714, !dbg !26 + %716 = fadd float %598, %715, !dbg !26 + %717 = fadd float %599, %716, !dbg !26 + %718 = fadd float %600, %601, !dbg !26 + %719 = fadd float %602, %718, !dbg !26 + %720 = fadd float %603, %719, !dbg !26 + %721 = fadd float %604, %720, !dbg !26 + %722 = fadd float %605, %721, !dbg !26 + %723 = fadd float %606, %722, !dbg !26 + %724 = fadd float %607, %723, !dbg !26 + %725 = fadd float %608, %609, !dbg !26 + %726 = fadd float %610, %725, !dbg !26 + %727 = fadd float %611, %726, !dbg !26 + %728 = fadd float %612, %727, !dbg !26 + %729 = fadd float %613, %728, !dbg !26 + %730 = fadd float %614, %729, !dbg !26 + %731 = fadd float %615, %730, !dbg !26 + %732 = fadd float %616, %617, !dbg !26 + %733 = fadd float %618, %732, !dbg !26 + %734 = fadd float %619, %733, !dbg !26 + %735 = fadd float %620, %734, !dbg !26 + %736 = fadd float %621, %735, !dbg !26 + %737 = fadd float %622, %736, !dbg !26 + %738 = fadd float %623, %737, !dbg !26 + %739 = fadd float %624, %625, !dbg !26 + %740 = fadd float %626, %739, !dbg !26 + %741 = fadd float %627, %740, !dbg !26 + %742 = fadd float %628, %741, !dbg !26 + %743 = fadd float %629, %742, !dbg !26 + %744 = fadd float %630, %743, !dbg !26 + %745 = fadd float %631, %744, !dbg !26 + %746 = fadd float %632, %633, !dbg !26 + %747 = fadd float %634, %746, !dbg !26 + %748 = fadd float %635, %747, !dbg !26 + %749 = fadd float %636, %748, !dbg !26 + %750 = fadd float %637, %749, !dbg !26 + %751 = fadd float %638, %750, !dbg !26 + %752 = fadd float %639, %751, !dbg !26 + %753 = fadd float %640, %641, !dbg !26 + %754 = fadd float %642, %753, !dbg !26 + %755 = fadd float %643, %754, !dbg !26 + %756 = fadd float %644, %755, !dbg !26 + %757 = fadd float %645, %756, !dbg !26 + %758 = fadd float %646, %757, !dbg !26 + %759 = fadd float %647, %758, !dbg !26 + %760 = bitcast float %654 to i32, !dbg !29 + %761 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %760, i32 8, i32 31), !dbg !29 + %762 = bitcast i32 %761 to float, !dbg !29 + %763 = fadd float %654, %762, !dbg !26 + %764 = bitcast float %763 to i32, !dbg !29 + %765 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %764, i32 4, i32 31), !dbg !29 + %766 = bitcast i32 %765 to float, !dbg !29 + %767 = fadd float %763, %766, !dbg !26 + %768 = bitcast float %767 to i32, !dbg !29 + %769 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %768, i32 2, i32 31), !dbg !29 + %770 = bitcast i32 %769 to float, !dbg !29 + %771 = fadd float %767, %770, !dbg !26 + %772 = bitcast float %771 to i32, !dbg !29 + %773 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %772, i32 1, i32 31), !dbg !29 + %774 = bitcast i32 %773 to float, !dbg !29 + %775 = fadd float %771, %774, !dbg !26 + %776 = bitcast float %661 to i32, !dbg !29 + %777 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %776, i32 8, i32 31), !dbg !29 + %778 = bitcast i32 %777 to float, !dbg !29 + %779 = fadd float %661, %778, !dbg !26 + %780 = bitcast float %779 to i32, !dbg !29 + %781 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %780, i32 4, i32 31), !dbg !29 + %782 = bitcast i32 %781 to float, !dbg !29 + %783 = fadd float %779, %782, !dbg !26 + %784 = bitcast float %783 to i32, !dbg !29 + %785 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %784, i32 2, i32 31), !dbg !29 + %786 = bitcast i32 %785 to float, !dbg !29 + %787 = fadd float %783, %786, !dbg !26 + %788 = bitcast float %787 to i32, !dbg !29 + %789 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %788, i32 1, i32 31), !dbg !29 + %790 = bitcast i32 %789 to float, !dbg !29 + %791 = fadd float %787, %790, !dbg !26 + %792 = bitcast float %668 to i32, !dbg !29 + %793 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %792, i32 8, i32 31), !dbg !29 + %794 = bitcast i32 %793 to float, !dbg !29 + %795 = fadd float %668, %794, !dbg !26 + %796 = bitcast float %795 to i32, !dbg !29 + %797 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %796, i32 4, i32 31), !dbg !29 + %798 = bitcast i32 %797 to float, !dbg !29 + %799 = fadd float %795, %798, !dbg !26 + %800 = bitcast float %799 to i32, !dbg !29 + %801 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %800, i32 2, i32 31), !dbg !29 + %802 = bitcast i32 %801 to float, !dbg !29 + %803 = fadd float %799, %802, !dbg !26 + %804 = bitcast float %803 to i32, !dbg !29 + %805 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %804, i32 1, i32 31), !dbg !29 + %806 = bitcast i32 %805 to float, !dbg !29 + %807 = fadd float %803, %806, !dbg !26 + %808 = bitcast float %675 to i32, !dbg !29 + %809 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %808, i32 8, i32 31), !dbg !29 + %810 = bitcast i32 %809 to float, !dbg !29 + %811 = fadd float %675, %810, !dbg !26 + %812 = bitcast float %811 to i32, !dbg !29 + %813 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %812, i32 4, i32 31), !dbg !29 + %814 = bitcast i32 %813 to float, !dbg !29 + %815 = fadd float %811, %814, !dbg !26 + %816 = bitcast float %815 to i32, !dbg !29 + %817 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %816, i32 2, i32 31), !dbg !29 + %818 = bitcast i32 %817 to float, !dbg !29 + %819 = fadd float %815, %818, !dbg !26 + %820 = bitcast float %819 to i32, !dbg !29 + %821 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %820, i32 1, i32 31), !dbg !29 + %822 = bitcast i32 %821 to float, !dbg !29 + %823 = fadd float %819, %822, !dbg !26 + %824 = bitcast float %682 to i32, !dbg !29 + %825 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %824, i32 8, i32 31), !dbg !29 + %826 = bitcast i32 %825 to float, !dbg !29 + %827 = fadd float %682, %826, !dbg !26 + %828 = bitcast float %827 to i32, !dbg !29 + %829 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %828, i32 4, i32 31), !dbg !29 + %830 = bitcast i32 %829 to float, !dbg !29 + %831 = fadd float %827, %830, !dbg !26 + %832 = bitcast float %831 to i32, !dbg !29 + %833 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %832, i32 2, i32 31), !dbg !29 + %834 = bitcast i32 %833 to float, !dbg !29 + %835 = fadd float %831, %834, !dbg !26 + %836 = bitcast float %835 to i32, !dbg !29 + %837 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %836, i32 1, i32 31), !dbg !29 + %838 = bitcast i32 %837 to float, !dbg !29 + %839 = fadd float %835, %838, !dbg !26 + %840 = bitcast float %689 to i32, !dbg !29 + %841 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %840, i32 8, i32 31), !dbg !29 + %842 = bitcast i32 %841 to float, !dbg !29 + %843 = fadd float %689, %842, !dbg !26 + %844 = bitcast float %843 to i32, !dbg !29 + %845 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %844, i32 4, i32 31), !dbg !29 + %846 = bitcast i32 %845 to float, !dbg !29 + %847 = fadd float %843, %846, !dbg !26 + %848 = bitcast float %847 to i32, !dbg !29 + %849 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %848, i32 2, i32 31), !dbg !29 + %850 = bitcast i32 %849 to float, !dbg !29 + %851 = fadd float %847, %850, !dbg !26 + %852 = bitcast float %851 to i32, !dbg !29 + %853 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %852, i32 1, i32 31), !dbg !29 + %854 = bitcast i32 %853 to float, !dbg !29 + %855 = fadd float %851, %854, !dbg !26 + %856 = bitcast float %696 to i32, !dbg !29 + %857 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %856, i32 8, i32 31), !dbg !29 + %858 = bitcast i32 %857 to float, !dbg !29 + %859 = fadd float %696, %858, !dbg !26 + %860 = bitcast float %859 to i32, !dbg !29 + %861 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %860, i32 4, i32 31), !dbg !29 + %862 = bitcast i32 %861 to float, !dbg !29 + %863 = fadd float %859, %862, !dbg !26 + %864 = bitcast float %863 to i32, !dbg !29 + %865 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %864, i32 2, i32 31), !dbg !29 + %866 = bitcast i32 %865 to float, !dbg !29 + %867 = fadd float %863, %866, !dbg !26 + %868 = bitcast float %867 to i32, !dbg !29 + %869 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %868, i32 1, i32 31), !dbg !29 + %870 = bitcast i32 %869 to float, !dbg !29 + %871 = fadd float %867, %870, !dbg !26 + %872 = bitcast float %703 to i32, !dbg !29 + %873 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %872, i32 8, i32 31), !dbg !29 + %874 = bitcast i32 %873 to float, !dbg !29 + %875 = fadd float %703, %874, !dbg !26 + %876 = bitcast float %875 to i32, !dbg !29 + %877 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %876, i32 4, i32 31), !dbg !29 + %878 = bitcast i32 %877 to float, !dbg !29 + %879 = fadd float %875, %878, !dbg !26 + %880 = bitcast float %879 to i32, !dbg !29 + %881 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %880, i32 2, i32 31), !dbg !29 + %882 = bitcast i32 %881 to float, !dbg !29 + %883 = fadd float %879, %882, !dbg !26 + %884 = bitcast float %883 to i32, !dbg !29 + %885 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %884, i32 1, i32 31), !dbg !29 + %886 = bitcast i32 %885 to float, !dbg !29 + %887 = fadd float %883, %886, !dbg !26 + %888 = bitcast float %710 to i32, !dbg !29 + %889 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %888, i32 8, i32 31), !dbg !29 + %890 = bitcast i32 %889 to float, !dbg !29 + %891 = fadd float %710, %890, !dbg !26 + %892 = bitcast float %891 to i32, !dbg !29 + %893 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %892, i32 4, i32 31), !dbg !29 + %894 = bitcast i32 %893 to float, !dbg !29 + %895 = fadd float %891, %894, !dbg !26 + %896 = bitcast float %895 to i32, !dbg !29 + %897 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %896, i32 2, i32 31), !dbg !29 + %898 = bitcast i32 %897 to float, !dbg !29 + %899 = fadd float %895, %898, !dbg !26 + %900 = bitcast float %899 to i32, !dbg !29 + %901 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %900, i32 1, i32 31), !dbg !29 + %902 = bitcast i32 %901 to float, !dbg !29 + %903 = fadd float %899, %902, !dbg !26 + %904 = bitcast float %717 to i32, !dbg !29 + %905 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %904, i32 8, i32 31), !dbg !29 + %906 = bitcast i32 %905 to float, !dbg !29 + %907 = fadd float %717, %906, !dbg !26 + %908 = bitcast float %907 to i32, !dbg !29 + %909 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %908, i32 4, i32 31), !dbg !29 + %910 = bitcast i32 %909 to float, !dbg !29 + %911 = fadd float %907, %910, !dbg !26 + %912 = bitcast float %911 to i32, !dbg !29 + %913 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %912, i32 2, i32 31), !dbg !29 + %914 = bitcast i32 %913 to float, !dbg !29 + %915 = fadd float %911, %914, !dbg !26 + %916 = bitcast float %915 to i32, !dbg !29 + %917 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %916, i32 1, i32 31), !dbg !29 + %918 = bitcast i32 %917 to float, !dbg !29 + %919 = fadd float %915, %918, !dbg !26 + %920 = bitcast float %724 to i32, !dbg !29 + %921 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %920, i32 8, i32 31), !dbg !29 + %922 = bitcast i32 %921 to float, !dbg !29 + %923 = fadd float %724, %922, !dbg !26 + %924 = bitcast float %923 to i32, !dbg !29 + %925 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %924, i32 4, i32 31), !dbg !29 + %926 = bitcast i32 %925 to float, !dbg !29 + %927 = fadd float %923, %926, !dbg !26 + %928 = bitcast float %927 to i32, !dbg !29 + %929 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %928, i32 2, i32 31), !dbg !29 + %930 = bitcast i32 %929 to float, !dbg !29 + %931 = fadd float %927, %930, !dbg !26 + %932 = bitcast float %931 to i32, !dbg !29 + %933 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %932, i32 1, i32 31), !dbg !29 + %934 = bitcast i32 %933 to float, !dbg !29 + %935 = fadd float %931, %934, !dbg !26 + %936 = bitcast float %731 to i32, !dbg !29 + %937 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %936, i32 8, i32 31), !dbg !29 + %938 = bitcast i32 %937 to float, !dbg !29 + %939 = fadd float %731, %938, !dbg !26 + %940 = bitcast float %939 to i32, !dbg !29 + %941 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %940, i32 4, i32 31), !dbg !29 + %942 = bitcast i32 %941 to float, !dbg !29 + %943 = fadd float %939, %942, !dbg !26 + %944 = bitcast float %943 to i32, !dbg !29 + %945 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %944, i32 2, i32 31), !dbg !29 + %946 = bitcast i32 %945 to float, !dbg !29 + %947 = fadd float %943, %946, !dbg !26 + %948 = bitcast float %947 to i32, !dbg !29 + %949 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %948, i32 1, i32 31), !dbg !29 + %950 = bitcast i32 %949 to float, !dbg !29 + %951 = fadd float %947, %950, !dbg !26 + %952 = bitcast float %738 to i32, !dbg !29 + %953 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %952, i32 8, i32 31), !dbg !29 + %954 = bitcast i32 %953 to float, !dbg !29 + %955 = fadd float %738, %954, !dbg !26 + %956 = bitcast float %955 to i32, !dbg !29 + %957 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %956, i32 4, i32 31), !dbg !29 + %958 = bitcast i32 %957 to float, !dbg !29 + %959 = fadd float %955, %958, !dbg !26 + %960 = bitcast float %959 to i32, !dbg !29 + %961 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %960, i32 2, i32 31), !dbg !29 + %962 = bitcast i32 %961 to float, !dbg !29 + %963 = fadd float %959, %962, !dbg !26 + %964 = bitcast float %963 to i32, !dbg !29 + %965 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %964, i32 1, i32 31), !dbg !29 + %966 = bitcast i32 %965 to float, !dbg !29 + %967 = fadd float %963, %966, !dbg !26 + %968 = bitcast float %745 to i32, !dbg !29 + %969 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %968, i32 8, i32 31), !dbg !29 + %970 = bitcast i32 %969 to float, !dbg !29 + %971 = fadd float %745, %970, !dbg !26 + %972 = bitcast float %971 to i32, !dbg !29 + %973 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %972, i32 4, i32 31), !dbg !29 + %974 = bitcast i32 %973 to float, !dbg !29 + %975 = fadd float %971, %974, !dbg !26 + %976 = bitcast float %975 to i32, !dbg !29 + %977 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %976, i32 2, i32 31), !dbg !29 + %978 = bitcast i32 %977 to float, !dbg !29 + %979 = fadd float %975, %978, !dbg !26 + %980 = bitcast float %979 to i32, !dbg !29 + %981 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %980, i32 1, i32 31), !dbg !29 + %982 = bitcast i32 %981 to float, !dbg !29 + %983 = fadd float %979, %982, !dbg !26 + %984 = bitcast float %752 to i32, !dbg !29 + %985 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %984, i32 8, i32 31), !dbg !29 + %986 = bitcast i32 %985 to float, !dbg !29 + %987 = fadd float %752, %986, !dbg !26 + %988 = bitcast float %987 to i32, !dbg !29 + %989 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %988, i32 4, i32 31), !dbg !29 + %990 = bitcast i32 %989 to float, !dbg !29 + %991 = fadd float %987, %990, !dbg !26 + %992 = bitcast float %991 to i32, !dbg !29 + %993 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %992, i32 2, i32 31), !dbg !29 + %994 = bitcast i32 %993 to float, !dbg !29 + %995 = fadd float %991, %994, !dbg !26 + %996 = bitcast float %995 to i32, !dbg !29 + %997 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %996, i32 1, i32 31), !dbg !29 + %998 = bitcast i32 %997 to float, !dbg !29 + %999 = fadd float %995, %998, !dbg !26 + %1000 = bitcast float %759 to i32, !dbg !29 + %1001 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1000, i32 8, i32 31), !dbg !29 + %1002 = bitcast i32 %1001 to float, !dbg !29 + %1003 = fadd float %759, %1002, !dbg !26 + %1004 = bitcast float %1003 to i32, !dbg !29 + %1005 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1004, i32 4, i32 31), !dbg !29 + %1006 = bitcast i32 %1005 to float, !dbg !29 + %1007 = fadd float %1003, %1006, !dbg !26 + %1008 = bitcast float %1007 to i32, !dbg !29 + %1009 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1008, i32 2, i32 31), !dbg !29 + %1010 = bitcast i32 %1009 to float, !dbg !29 + %1011 = fadd float %1007, %1010, !dbg !26 + %1012 = bitcast float %1011 to i32, !dbg !29 + %1013 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1012, i32 1, i32 31), !dbg !29 + %1014 = bitcast i32 %1013 to float, !dbg !29 + %1015 = fadd float %1011, %1014, !dbg !26 + %1016 = fadd float %3, %775, !dbg !32 + %1017 = fadd float %3, %791, !dbg !32 + %1018 = fadd float %3, %807, !dbg !32 + %1019 = fadd float %3, %823, !dbg !32 + %1020 = fadd float %3, %839, !dbg !32 + %1021 = fadd float %3, %855, !dbg !32 + %1022 = fadd float %3, %871, !dbg !32 + %1023 = fadd float %3, %887, !dbg !32 + %1024 = fadd float %3, %903, !dbg !32 + %1025 = fadd float %3, %919, !dbg !32 + %1026 = fadd float %3, %935, !dbg !32 + %1027 = fadd float %3, %951, !dbg !32 + %1028 = fadd float %3, %967, !dbg !32 + %1029 = fadd float %3, %983, !dbg !32 + %1030 = fadd float %3, %999, !dbg !32 + %1031 = fadd float %3, %1015, !dbg !32 + %1032 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i = icmp eq i32 %1032, 0, !dbg !33 + %1033 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i = icmp eq i32 %1033, 0, !dbg !33 + br i1 %.not.i, label %1039, label %1034, !dbg !33 + +1034: ; preds = %7 + br i1 %.not1.i, label %1037, label %1035, !dbg !33 + +1035: ; preds = %1034 + %1036 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1016) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +1037: ; preds = %1034 + %1038 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1016) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +1039: ; preds = %7 + br i1 %.not1.i, label %1042, label %1040, !dbg !33 + +1040: ; preds = %1039 + %1041 = tail call float @llvm.nvvm.sqrt.rn.f(float %1016) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +1042: ; preds = %1039 + %1043 = tail call float @llvm.nvvm.sqrt.approx.f(float %1016) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +__nv_sqrtf.exit: ; preds = %1035, %1037, %1040, %1042 + %.0.i = phi float [ %1036, %1035 ], [ %1038, %1037 ], [ %1041, %1040 ], [ %1043, %1042 ], !dbg !33 + %1044 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i1 = icmp eq i32 %1044, 0, !dbg !33 + %1045 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i4 = icmp eq i32 %1045, 0, !dbg !33 + br i1 %.not.i1, label %1051, label %1046, !dbg !33 + +1046: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %1049, label %1047, !dbg !33 + +1047: ; preds = %1046 + %1048 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1017) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +1049: ; preds = %1046 + %1050 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1017) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +1051: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %1054, label %1052, !dbg !33 + +1052: ; preds = %1051 + %1053 = tail call float @llvm.nvvm.sqrt.rn.f(float %1017) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +1054: ; preds = %1051 + %1055 = tail call float @llvm.nvvm.sqrt.approx.f(float %1017) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +__nv_sqrtf.exit5: ; preds = %1047, %1049, %1052, %1054 + %.0.i3 = phi float [ %1048, %1047 ], [ %1050, %1049 ], [ %1053, %1052 ], [ %1055, %1054 ], !dbg !33 + %1056 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i6 = icmp eq i32 %1056, 0, !dbg !33 + %1057 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i9 = icmp eq i32 %1057, 0, !dbg !33 + br i1 %.not.i6, label %1063, label %1058, !dbg !33 + +1058: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %1061, label %1059, !dbg !33 + +1059: ; preds = %1058 + %1060 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1018) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +1061: ; preds = %1058 + %1062 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1018) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +1063: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %1066, label %1064, !dbg !33 + +1064: ; preds = %1063 + %1065 = tail call float @llvm.nvvm.sqrt.rn.f(float %1018) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +1066: ; preds = %1063 + %1067 = tail call float @llvm.nvvm.sqrt.approx.f(float %1018) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +__nv_sqrtf.exit10: ; preds = %1059, %1061, %1064, %1066 + %.0.i8 = phi float [ %1060, %1059 ], [ %1062, %1061 ], [ %1065, %1064 ], [ %1067, %1066 ], !dbg !33 + %1068 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i11 = icmp eq i32 %1068, 0, !dbg !33 + %1069 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i14 = icmp eq i32 %1069, 0, !dbg !33 + br i1 %.not.i11, label %1075, label %1070, !dbg !33 + +1070: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %1073, label %1071, !dbg !33 + +1071: ; preds = %1070 + %1072 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1019) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +1073: ; preds = %1070 + %1074 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1019) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +1075: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %1078, label %1076, !dbg !33 + +1076: ; preds = %1075 + %1077 = tail call float @llvm.nvvm.sqrt.rn.f(float %1019) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +1078: ; preds = %1075 + %1079 = tail call float @llvm.nvvm.sqrt.approx.f(float %1019) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +__nv_sqrtf.exit15: ; preds = %1071, %1073, %1076, %1078 + %.0.i13 = phi float [ %1072, %1071 ], [ %1074, %1073 ], [ %1077, %1076 ], [ %1079, %1078 ], !dbg !33 + %1080 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i16 = icmp eq i32 %1080, 0, !dbg !33 + %1081 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i19 = icmp eq i32 %1081, 0, !dbg !33 + br i1 %.not.i16, label %1087, label %1082, !dbg !33 + +1082: ; preds = %__nv_sqrtf.exit15 + br i1 %.not1.i19, label %1085, label %1083, !dbg !33 + +1083: ; preds = %1082 + %1084 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1020) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +1085: ; preds = %1082 + %1086 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1020) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +1087: ; preds = %__nv_sqrtf.exit15 + br i1 %.not1.i19, label %1090, label %1088, !dbg !33 + +1088: ; preds = %1087 + %1089 = tail call float @llvm.nvvm.sqrt.rn.f(float %1020) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +1090: ; preds = %1087 + %1091 = tail call float @llvm.nvvm.sqrt.approx.f(float %1020) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +__nv_sqrtf.exit20: ; preds = %1083, %1085, %1088, %1090 + %.0.i18 = phi float [ %1084, %1083 ], [ %1086, %1085 ], [ %1089, %1088 ], [ %1091, %1090 ], !dbg !33 + %1092 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i21 = icmp eq i32 %1092, 0, !dbg !33 + %1093 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i24 = icmp eq i32 %1093, 0, !dbg !33 + br i1 %.not.i21, label %1099, label %1094, !dbg !33 + +1094: ; preds = %__nv_sqrtf.exit20 + br i1 %.not1.i24, label %1097, label %1095, !dbg !33 + +1095: ; preds = %1094 + %1096 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1021) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +1097: ; preds = %1094 + %1098 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1021) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +1099: ; preds = %__nv_sqrtf.exit20 + br i1 %.not1.i24, label %1102, label %1100, !dbg !33 + +1100: ; preds = %1099 + %1101 = tail call float @llvm.nvvm.sqrt.rn.f(float %1021) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +1102: ; preds = %1099 + %1103 = tail call float @llvm.nvvm.sqrt.approx.f(float %1021) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +__nv_sqrtf.exit25: ; preds = %1095, %1097, %1100, %1102 + %.0.i23 = phi float [ %1096, %1095 ], [ %1098, %1097 ], [ %1101, %1100 ], [ %1103, %1102 ], !dbg !33 + %1104 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i26 = icmp eq i32 %1104, 0, !dbg !33 + %1105 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i29 = icmp eq i32 %1105, 0, !dbg !33 + br i1 %.not.i26, label %1111, label %1106, !dbg !33 + +1106: ; preds = %__nv_sqrtf.exit25 + br i1 %.not1.i29, label %1109, label %1107, !dbg !33 + +1107: ; preds = %1106 + %1108 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1022) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +1109: ; preds = %1106 + %1110 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1022) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +1111: ; preds = %__nv_sqrtf.exit25 + br i1 %.not1.i29, label %1114, label %1112, !dbg !33 + +1112: ; preds = %1111 + %1113 = tail call float @llvm.nvvm.sqrt.rn.f(float %1022) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +1114: ; preds = %1111 + %1115 = tail call float @llvm.nvvm.sqrt.approx.f(float %1022) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +__nv_sqrtf.exit30: ; preds = %1107, %1109, %1112, %1114 + %.0.i28 = phi float [ %1108, %1107 ], [ %1110, %1109 ], [ %1113, %1112 ], [ %1115, %1114 ], !dbg !33 + %1116 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i31 = icmp eq i32 %1116, 0, !dbg !33 + %1117 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i34 = icmp eq i32 %1117, 0, !dbg !33 + br i1 %.not.i31, label %1123, label %1118, !dbg !33 + +1118: ; preds = %__nv_sqrtf.exit30 + br i1 %.not1.i34, label %1121, label %1119, !dbg !33 + +1119: ; preds = %1118 + %1120 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1023) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +1121: ; preds = %1118 + %1122 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1023) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +1123: ; preds = %__nv_sqrtf.exit30 + br i1 %.not1.i34, label %1126, label %1124, !dbg !33 + +1124: ; preds = %1123 + %1125 = tail call float @llvm.nvvm.sqrt.rn.f(float %1023) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +1126: ; preds = %1123 + %1127 = tail call float @llvm.nvvm.sqrt.approx.f(float %1023) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +__nv_sqrtf.exit35: ; preds = %1119, %1121, %1124, %1126 + %.0.i33 = phi float [ %1120, %1119 ], [ %1122, %1121 ], [ %1125, %1124 ], [ %1127, %1126 ], !dbg !33 + %1128 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i36 = icmp eq i32 %1128, 0, !dbg !33 + %1129 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i39 = icmp eq i32 %1129, 0, !dbg !33 + br i1 %.not.i36, label %1135, label %1130, !dbg !33 + +1130: ; preds = %__nv_sqrtf.exit35 + br i1 %.not1.i39, label %1133, label %1131, !dbg !33 + +1131: ; preds = %1130 + %1132 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1024) #6, !dbg !33 + br label %__nv_sqrtf.exit40, !dbg !33 + +1133: ; preds = %1130 + %1134 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1024) #6, !dbg !33 + br label %__nv_sqrtf.exit40, !dbg !33 + +1135: ; preds = %__nv_sqrtf.exit35 + br i1 %.not1.i39, label %1138, label %1136, !dbg !33 + +1136: ; preds = %1135 + %1137 = tail call float @llvm.nvvm.sqrt.rn.f(float %1024) #6, !dbg !33 + br label %__nv_sqrtf.exit40, !dbg !33 + +1138: ; preds = %1135 + %1139 = tail call float @llvm.nvvm.sqrt.approx.f(float %1024) #6, !dbg !33 + br label %__nv_sqrtf.exit40, !dbg !33 + +__nv_sqrtf.exit40: ; preds = %1131, %1133, %1136, %1138 + %.0.i38 = phi float [ %1132, %1131 ], [ %1134, %1133 ], [ %1137, %1136 ], [ %1139, %1138 ], !dbg !33 + %1140 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i41 = icmp eq i32 %1140, 0, !dbg !33 + %1141 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i44 = icmp eq i32 %1141, 0, !dbg !33 + br i1 %.not.i41, label %1147, label %1142, !dbg !33 + +1142: ; preds = %__nv_sqrtf.exit40 + br i1 %.not1.i44, label %1145, label %1143, !dbg !33 + +1143: ; preds = %1142 + %1144 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1025) #6, !dbg !33 + br label %__nv_sqrtf.exit45, !dbg !33 + +1145: ; preds = %1142 + %1146 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1025) #6, !dbg !33 + br label %__nv_sqrtf.exit45, !dbg !33 + +1147: ; preds = %__nv_sqrtf.exit40 + br i1 %.not1.i44, label %1150, label %1148, !dbg !33 + +1148: ; preds = %1147 + %1149 = tail call float @llvm.nvvm.sqrt.rn.f(float %1025) #6, !dbg !33 + br label %__nv_sqrtf.exit45, !dbg !33 + +1150: ; preds = %1147 + %1151 = tail call float @llvm.nvvm.sqrt.approx.f(float %1025) #6, !dbg !33 + br label %__nv_sqrtf.exit45, !dbg !33 + +__nv_sqrtf.exit45: ; preds = %1143, %1145, %1148, %1150 + %.0.i43 = phi float [ %1144, %1143 ], [ %1146, %1145 ], [ %1149, %1148 ], [ %1151, %1150 ], !dbg !33 + %1152 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i46 = icmp eq i32 %1152, 0, !dbg !33 + %1153 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i49 = icmp eq i32 %1153, 0, !dbg !33 + br i1 %.not.i46, label %1159, label %1154, !dbg !33 + +1154: ; preds = %__nv_sqrtf.exit45 + br i1 %.not1.i49, label %1157, label %1155, !dbg !33 + +1155: ; preds = %1154 + %1156 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1026) #6, !dbg !33 + br label %__nv_sqrtf.exit50, !dbg !33 + +1157: ; preds = %1154 + %1158 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1026) #6, !dbg !33 + br label %__nv_sqrtf.exit50, !dbg !33 + +1159: ; preds = %__nv_sqrtf.exit45 + br i1 %.not1.i49, label %1162, label %1160, !dbg !33 + +1160: ; preds = %1159 + %1161 = tail call float @llvm.nvvm.sqrt.rn.f(float %1026) #6, !dbg !33 + br label %__nv_sqrtf.exit50, !dbg !33 + +1162: ; preds = %1159 + %1163 = tail call float @llvm.nvvm.sqrt.approx.f(float %1026) #6, !dbg !33 + br label %__nv_sqrtf.exit50, !dbg !33 + +__nv_sqrtf.exit50: ; preds = %1155, %1157, %1160, %1162 + %.0.i48 = phi float [ %1156, %1155 ], [ %1158, %1157 ], [ %1161, %1160 ], [ %1163, %1162 ], !dbg !33 + %1164 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i51 = icmp eq i32 %1164, 0, !dbg !33 + %1165 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i54 = icmp eq i32 %1165, 0, !dbg !33 + br i1 %.not.i51, label %1171, label %1166, !dbg !33 + +1166: ; preds = %__nv_sqrtf.exit50 + br i1 %.not1.i54, label %1169, label %1167, !dbg !33 + +1167: ; preds = %1166 + %1168 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1027) #6, !dbg !33 + br label %__nv_sqrtf.exit55, !dbg !33 + +1169: ; preds = %1166 + %1170 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1027) #6, !dbg !33 + br label %__nv_sqrtf.exit55, !dbg !33 + +1171: ; preds = %__nv_sqrtf.exit50 + br i1 %.not1.i54, label %1174, label %1172, !dbg !33 + +1172: ; preds = %1171 + %1173 = tail call float @llvm.nvvm.sqrt.rn.f(float %1027) #6, !dbg !33 + br label %__nv_sqrtf.exit55, !dbg !33 + +1174: ; preds = %1171 + %1175 = tail call float @llvm.nvvm.sqrt.approx.f(float %1027) #6, !dbg !33 + br label %__nv_sqrtf.exit55, !dbg !33 + +__nv_sqrtf.exit55: ; preds = %1167, %1169, %1172, %1174 + %.0.i53 = phi float [ %1168, %1167 ], [ %1170, %1169 ], [ %1173, %1172 ], [ %1175, %1174 ], !dbg !33 + %1176 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i56 = icmp eq i32 %1176, 0, !dbg !33 + %1177 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i59 = icmp eq i32 %1177, 0, !dbg !33 + br i1 %.not.i56, label %1183, label %1178, !dbg !33 + +1178: ; preds = %__nv_sqrtf.exit55 + br i1 %.not1.i59, label %1181, label %1179, !dbg !33 + +1179: ; preds = %1178 + %1180 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1028) #6, !dbg !33 + br label %__nv_sqrtf.exit60, !dbg !33 + +1181: ; preds = %1178 + %1182 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1028) #6, !dbg !33 + br label %__nv_sqrtf.exit60, !dbg !33 + +1183: ; preds = %__nv_sqrtf.exit55 + br i1 %.not1.i59, label %1186, label %1184, !dbg !33 + +1184: ; preds = %1183 + %1185 = tail call float @llvm.nvvm.sqrt.rn.f(float %1028) #6, !dbg !33 + br label %__nv_sqrtf.exit60, !dbg !33 + +1186: ; preds = %1183 + %1187 = tail call float @llvm.nvvm.sqrt.approx.f(float %1028) #6, !dbg !33 + br label %__nv_sqrtf.exit60, !dbg !33 + +__nv_sqrtf.exit60: ; preds = %1179, %1181, %1184, %1186 + %.0.i58 = phi float [ %1180, %1179 ], [ %1182, %1181 ], [ %1185, %1184 ], [ %1187, %1186 ], !dbg !33 + %1188 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i61 = icmp eq i32 %1188, 0, !dbg !33 + %1189 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i64 = icmp eq i32 %1189, 0, !dbg !33 + br i1 %.not.i61, label %1195, label %1190, !dbg !33 + +1190: ; preds = %__nv_sqrtf.exit60 + br i1 %.not1.i64, label %1193, label %1191, !dbg !33 + +1191: ; preds = %1190 + %1192 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1029) #6, !dbg !33 + br label %__nv_sqrtf.exit65, !dbg !33 + +1193: ; preds = %1190 + %1194 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1029) #6, !dbg !33 + br label %__nv_sqrtf.exit65, !dbg !33 + +1195: ; preds = %__nv_sqrtf.exit60 + br i1 %.not1.i64, label %1198, label %1196, !dbg !33 + +1196: ; preds = %1195 + %1197 = tail call float @llvm.nvvm.sqrt.rn.f(float %1029) #6, !dbg !33 + br label %__nv_sqrtf.exit65, !dbg !33 + +1198: ; preds = %1195 + %1199 = tail call float @llvm.nvvm.sqrt.approx.f(float %1029) #6, !dbg !33 + br label %__nv_sqrtf.exit65, !dbg !33 + +__nv_sqrtf.exit65: ; preds = %1191, %1193, %1196, %1198 + %.0.i63 = phi float [ %1192, %1191 ], [ %1194, %1193 ], [ %1197, %1196 ], [ %1199, %1198 ], !dbg !33 + %1200 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i66 = icmp eq i32 %1200, 0, !dbg !33 + %1201 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i69 = icmp eq i32 %1201, 0, !dbg !33 + br i1 %.not.i66, label %1207, label %1202, !dbg !33 + +1202: ; preds = %__nv_sqrtf.exit65 + br i1 %.not1.i69, label %1205, label %1203, !dbg !33 + +1203: ; preds = %1202 + %1204 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1030) #6, !dbg !33 + br label %__nv_sqrtf.exit70, !dbg !33 + +1205: ; preds = %1202 + %1206 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1030) #6, !dbg !33 + br label %__nv_sqrtf.exit70, !dbg !33 + +1207: ; preds = %__nv_sqrtf.exit65 + br i1 %.not1.i69, label %1210, label %1208, !dbg !33 + +1208: ; preds = %1207 + %1209 = tail call float @llvm.nvvm.sqrt.rn.f(float %1030) #6, !dbg !33 + br label %__nv_sqrtf.exit70, !dbg !33 + +1210: ; preds = %1207 + %1211 = tail call float @llvm.nvvm.sqrt.approx.f(float %1030) #6, !dbg !33 + br label %__nv_sqrtf.exit70, !dbg !33 + +__nv_sqrtf.exit70: ; preds = %1203, %1205, %1208, %1210 + %.0.i68 = phi float [ %1204, %1203 ], [ %1206, %1205 ], [ %1209, %1208 ], [ %1211, %1210 ], !dbg !33 + %1212 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i71 = icmp eq i32 %1212, 0, !dbg !33 + %1213 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i74 = icmp eq i32 %1213, 0, !dbg !33 + br i1 %.not.i71, label %1219, label %1214, !dbg !33 + +1214: ; preds = %__nv_sqrtf.exit70 + br i1 %.not1.i74, label %1217, label %1215, !dbg !33 + +1215: ; preds = %1214 + %1216 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1031) #6, !dbg !33 + br label %__nv_sqrtf.exit75, !dbg !33 + +1217: ; preds = %1214 + %1218 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1031) #6, !dbg !33 + br label %__nv_sqrtf.exit75, !dbg !33 + +1219: ; preds = %__nv_sqrtf.exit70 + br i1 %.not1.i74, label %1222, label %1220, !dbg !33 + +1220: ; preds = %1219 + %1221 = tail call float @llvm.nvvm.sqrt.rn.f(float %1031) #6, !dbg !33 + br label %__nv_sqrtf.exit75, !dbg !33 + +1222: ; preds = %1219 + %1223 = tail call float @llvm.nvvm.sqrt.approx.f(float %1031) #6, !dbg !33 + br label %__nv_sqrtf.exit75, !dbg !33 + +__nv_sqrtf.exit75: ; preds = %1215, %1217, %1220, %1222 + %.0.i73 = phi float [ %1216, %1215 ], [ %1218, %1217 ], [ %1221, %1220 ], [ %1223, %1222 ], !dbg !33 + %1224 = sext i32 %4 to i64, !dbg !18 + %1225 = and i32 %11, 63, !dbg !16 + %1226 = zext nneg i32 %1225 to i64, !dbg !17 + %1227 = or disjoint i64 %10, %1226, !dbg !17 + %1228 = getelementptr float, ptr addrspace(1) %2, i64 %1227, !dbg !34 + %1229 = getelementptr bfloat, ptr addrspace(1) %1, i64 %70, !dbg !35 + %1230 = getelementptr bfloat, ptr addrspace(1) %1229, i64 %87, !dbg !36 + %1231 = getelementptr bfloat, ptr addrspace(1) %1, i64 %68, !dbg !35 + %1232 = getelementptr bfloat, ptr addrspace(1) %1231, i64 %87, !dbg !36 + %1233 = getelementptr bfloat, ptr addrspace(1) %1, i64 %66, !dbg !35 + %1234 = getelementptr bfloat, ptr addrspace(1) %1233, i64 %87, !dbg !36 + %1235 = getelementptr bfloat, ptr addrspace(1) %1, i64 %64, !dbg !35 + %1236 = getelementptr bfloat, ptr addrspace(1) %1235, i64 %87, !dbg !36 + %1237 = getelementptr bfloat, ptr addrspace(1) %1, i64 %62, !dbg !35 + %1238 = getelementptr bfloat, ptr addrspace(1) %1237, i64 %87, !dbg !36 + %1239 = getelementptr bfloat, ptr addrspace(1) %1, i64 %60, !dbg !35 + %1240 = getelementptr bfloat, ptr addrspace(1) %1239, i64 %87, !dbg !36 + %1241 = getelementptr bfloat, ptr addrspace(1) %1, i64 %58, !dbg !35 + %1242 = getelementptr bfloat, ptr addrspace(1) %1241, i64 %87, !dbg !36 + %1243 = getelementptr bfloat, ptr addrspace(1) %1, i64 %56, !dbg !35 + %1244 = getelementptr bfloat, ptr addrspace(1) %1243, i64 %87, !dbg !36 + %1245 = getelementptr bfloat, ptr addrspace(1) %1, i64 %54, !dbg !35 + %1246 = getelementptr bfloat, ptr addrspace(1) %1245, i64 %87, !dbg !36 + %1247 = getelementptr bfloat, ptr addrspace(1) %1, i64 %52, !dbg !35 + %1248 = getelementptr bfloat, ptr addrspace(1) %1247, i64 %87, !dbg !36 + %1249 = getelementptr bfloat, ptr addrspace(1) %1, i64 %50, !dbg !35 + %1250 = getelementptr bfloat, ptr addrspace(1) %1249, i64 %87, !dbg !36 + %1251 = getelementptr bfloat, ptr addrspace(1) %1, i64 %48, !dbg !35 + %1252 = getelementptr bfloat, ptr addrspace(1) %1251, i64 %87, !dbg !36 + %1253 = getelementptr bfloat, ptr addrspace(1) %1, i64 %46, !dbg !35 + %1254 = getelementptr bfloat, ptr addrspace(1) %1253, i64 %87, !dbg !36 + %1255 = getelementptr bfloat, ptr addrspace(1) %1, i64 %44, !dbg !35 + %1256 = getelementptr bfloat, ptr addrspace(1) %1255, i64 %87, !dbg !36 + %1257 = getelementptr bfloat, ptr addrspace(1) %1, i64 %42, !dbg !35 + %1258 = getelementptr bfloat, ptr addrspace(1) %1257, i64 %87, !dbg !36 + %1259 = getelementptr bfloat, ptr addrspace(1) %1, i64 %40, !dbg !35 + %1260 = getelementptr bfloat, ptr addrspace(1) %1259, i64 %87, !dbg !36 + %1261 = icmp slt i64 %1227, %1224, !dbg !18 + %1262 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i), !dbg !37 + %1263 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i3), !dbg !37 + %1264 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i8), !dbg !37 + %1265 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i13), !dbg !37 + %1266 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i18), !dbg !37 + %1267 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i23), !dbg !37 + %1268 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i28), !dbg !37 + %1269 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i33), !dbg !37 + %1270 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i38), !dbg !37 + %1271 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i43), !dbg !37 + %1272 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i48), !dbg !37 + %1273 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i53), !dbg !37 + %1274 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i58), !dbg !37 + %1275 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i63), !dbg !37 + %1276 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i68), !dbg !37 + %1277 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i73), !dbg !37 + %1278 = fmul float %1262, %392, !dbg !38 + %1279 = fmul float %1262, %393, !dbg !38 + %1280 = fmul float %1262, %394, !dbg !38 + %1281 = fmul float %1262, %395, !dbg !38 + %1282 = fmul float %1262, %396, !dbg !38 + %1283 = fmul float %1262, %397, !dbg !38 + %1284 = fmul float %1262, %398, !dbg !38 + %1285 = fmul float %1262, %399, !dbg !38 + %1286 = fmul float %1263, %400, !dbg !38 + %1287 = fmul float %1263, %401, !dbg !38 + %1288 = fmul float %1263, %402, !dbg !38 + %1289 = fmul float %1263, %403, !dbg !38 + %1290 = fmul float %1263, %404, !dbg !38 + %1291 = fmul float %1263, %405, !dbg !38 + %1292 = fmul float %1263, %406, !dbg !38 + %1293 = fmul float %1263, %407, !dbg !38 + %1294 = fmul float %1264, %408, !dbg !38 + %1295 = fmul float %1264, %409, !dbg !38 + %1296 = fmul float %1264, %410, !dbg !38 + %1297 = fmul float %1264, %411, !dbg !38 + %1298 = fmul float %1264, %412, !dbg !38 + %1299 = fmul float %1264, %413, !dbg !38 + %1300 = fmul float %1264, %414, !dbg !38 + %1301 = fmul float %1264, %415, !dbg !38 + %1302 = fmul float %1265, %416, !dbg !38 + %1303 = fmul float %1265, %417, !dbg !38 + %1304 = fmul float %1265, %418, !dbg !38 + %1305 = fmul float %1265, %419, !dbg !38 + %1306 = fmul float %1265, %420, !dbg !38 + %1307 = fmul float %1265, %421, !dbg !38 + %1308 = fmul float %1265, %422, !dbg !38 + %1309 = fmul float %1265, %423, !dbg !38 + %1310 = fmul float %1266, %424, !dbg !38 + %1311 = fmul float %1266, %425, !dbg !38 + %1312 = fmul float %1266, %426, !dbg !38 + %1313 = fmul float %1266, %427, !dbg !38 + %1314 = fmul float %1266, %428, !dbg !38 + %1315 = fmul float %1266, %429, !dbg !38 + %1316 = fmul float %1266, %430, !dbg !38 + %1317 = fmul float %1266, %431, !dbg !38 + %1318 = fmul float %1267, %432, !dbg !38 + %1319 = fmul float %1267, %433, !dbg !38 + %1320 = fmul float %1267, %434, !dbg !38 + %1321 = fmul float %1267, %435, !dbg !38 + %1322 = fmul float %1267, %436, !dbg !38 + %1323 = fmul float %1267, %437, !dbg !38 + %1324 = fmul float %1267, %438, !dbg !38 + %1325 = fmul float %1267, %439, !dbg !38 + %1326 = fmul float %1268, %440, !dbg !38 + %1327 = fmul float %1268, %441, !dbg !38 + %1328 = fmul float %1268, %442, !dbg !38 + %1329 = fmul float %1268, %443, !dbg !38 + %1330 = fmul float %1268, %444, !dbg !38 + %1331 = fmul float %1268, %445, !dbg !38 + %1332 = fmul float %1268, %446, !dbg !38 + %1333 = fmul float %1268, %447, !dbg !38 + %1334 = fmul float %1269, %448, !dbg !38 + %1335 = fmul float %1269, %449, !dbg !38 + %1336 = fmul float %1269, %450, !dbg !38 + %1337 = fmul float %1269, %451, !dbg !38 + %1338 = fmul float %1269, %452, !dbg !38 + %1339 = fmul float %1269, %453, !dbg !38 + %1340 = fmul float %1269, %454, !dbg !38 + %1341 = fmul float %1269, %455, !dbg !38 + %1342 = fmul float %1270, %456, !dbg !38 + %1343 = fmul float %1270, %457, !dbg !38 + %1344 = fmul float %1270, %458, !dbg !38 + %1345 = fmul float %1270, %459, !dbg !38 + %1346 = fmul float %1270, %460, !dbg !38 + %1347 = fmul float %1270, %461, !dbg !38 + %1348 = fmul float %1270, %462, !dbg !38 + %1349 = fmul float %1270, %463, !dbg !38 + %1350 = fmul float %1271, %464, !dbg !38 + %1351 = fmul float %1271, %465, !dbg !38 + %1352 = fmul float %1271, %466, !dbg !38 + %1353 = fmul float %1271, %467, !dbg !38 + %1354 = fmul float %1271, %468, !dbg !38 + %1355 = fmul float %1271, %469, !dbg !38 + %1356 = fmul float %1271, %470, !dbg !38 + %1357 = fmul float %1271, %471, !dbg !38 + %1358 = fmul float %1272, %472, !dbg !38 + %1359 = fmul float %1272, %473, !dbg !38 + %1360 = fmul float %1272, %474, !dbg !38 + %1361 = fmul float %1272, %475, !dbg !38 + %1362 = fmul float %1272, %476, !dbg !38 + %1363 = fmul float %1272, %477, !dbg !38 + %1364 = fmul float %1272, %478, !dbg !38 + %1365 = fmul float %1272, %479, !dbg !38 + %1366 = fmul float %1273, %480, !dbg !38 + %1367 = fmul float %1273, %481, !dbg !38 + %1368 = fmul float %1273, %482, !dbg !38 + %1369 = fmul float %1273, %483, !dbg !38 + %1370 = fmul float %1273, %484, !dbg !38 + %1371 = fmul float %1273, %485, !dbg !38 + %1372 = fmul float %1273, %486, !dbg !38 + %1373 = fmul float %1273, %487, !dbg !38 + %1374 = fmul float %1274, %488, !dbg !38 + %1375 = fmul float %1274, %489, !dbg !38 + %1376 = fmul float %1274, %490, !dbg !38 + %1377 = fmul float %1274, %491, !dbg !38 + %1378 = fmul float %1274, %492, !dbg !38 + %1379 = fmul float %1274, %493, !dbg !38 + %1380 = fmul float %1274, %494, !dbg !38 + %1381 = fmul float %1274, %495, !dbg !38 + %1382 = fmul float %1275, %496, !dbg !38 + %1383 = fmul float %1275, %497, !dbg !38 + %1384 = fmul float %1275, %498, !dbg !38 + %1385 = fmul float %1275, %499, !dbg !38 + %1386 = fmul float %1275, %500, !dbg !38 + %1387 = fmul float %1275, %501, !dbg !38 + %1388 = fmul float %1275, %502, !dbg !38 + %1389 = fmul float %1275, %503, !dbg !38 + %1390 = fmul float %1276, %504, !dbg !38 + %1391 = fmul float %1276, %505, !dbg !38 + %1392 = fmul float %1276, %506, !dbg !38 + %1393 = fmul float %1276, %507, !dbg !38 + %1394 = fmul float %1276, %508, !dbg !38 + %1395 = fmul float %1276, %509, !dbg !38 + %1396 = fmul float %1276, %510, !dbg !38 + %1397 = fmul float %1276, %511, !dbg !38 + %1398 = fmul float %1277, %512, !dbg !38 + %1399 = fmul float %1277, %513, !dbg !38 + %1400 = fmul float %1277, %514, !dbg !38 + %1401 = fmul float %1277, %515, !dbg !38 + %1402 = fmul float %1277, %516, !dbg !38 + %1403 = fmul float %1277, %517, !dbg !38 + %1404 = fmul float %1277, %518, !dbg !38 + %1405 = fmul float %1277, %519, !dbg !38 + %1406 = fptrunc float %1278 to bfloat, !dbg !39 + %1407 = fptrunc float %1279 to bfloat, !dbg !39 + %1408 = fptrunc float %1280 to bfloat, !dbg !39 + %1409 = fptrunc float %1281 to bfloat, !dbg !39 + %1410 = fptrunc float %1282 to bfloat, !dbg !39 + %1411 = fptrunc float %1283 to bfloat, !dbg !39 + %1412 = fptrunc float %1284 to bfloat, !dbg !39 + %1413 = fptrunc float %1285 to bfloat, !dbg !39 + %1414 = fptrunc float %1286 to bfloat, !dbg !39 + %1415 = fptrunc float %1287 to bfloat, !dbg !39 + %1416 = fptrunc float %1288 to bfloat, !dbg !39 + %1417 = fptrunc float %1289 to bfloat, !dbg !39 + %1418 = fptrunc float %1290 to bfloat, !dbg !39 + %1419 = fptrunc float %1291 to bfloat, !dbg !39 + %1420 = fptrunc float %1292 to bfloat, !dbg !39 + %1421 = fptrunc float %1293 to bfloat, !dbg !39 + %1422 = fptrunc float %1294 to bfloat, !dbg !39 + %1423 = fptrunc float %1295 to bfloat, !dbg !39 + %1424 = fptrunc float %1296 to bfloat, !dbg !39 + %1425 = fptrunc float %1297 to bfloat, !dbg !39 + %1426 = fptrunc float %1298 to bfloat, !dbg !39 + %1427 = fptrunc float %1299 to bfloat, !dbg !39 + %1428 = fptrunc float %1300 to bfloat, !dbg !39 + %1429 = fptrunc float %1301 to bfloat, !dbg !39 + %1430 = fptrunc float %1302 to bfloat, !dbg !39 + %1431 = fptrunc float %1303 to bfloat, !dbg !39 + %1432 = fptrunc float %1304 to bfloat, !dbg !39 + %1433 = fptrunc float %1305 to bfloat, !dbg !39 + %1434 = fptrunc float %1306 to bfloat, !dbg !39 + %1435 = fptrunc float %1307 to bfloat, !dbg !39 + %1436 = fptrunc float %1308 to bfloat, !dbg !39 + %1437 = fptrunc float %1309 to bfloat, !dbg !39 + %1438 = fptrunc float %1310 to bfloat, !dbg !39 + %1439 = fptrunc float %1311 to bfloat, !dbg !39 + %1440 = fptrunc float %1312 to bfloat, !dbg !39 + %1441 = fptrunc float %1313 to bfloat, !dbg !39 + %1442 = fptrunc float %1314 to bfloat, !dbg !39 + %1443 = fptrunc float %1315 to bfloat, !dbg !39 + %1444 = fptrunc float %1316 to bfloat, !dbg !39 + %1445 = fptrunc float %1317 to bfloat, !dbg !39 + %1446 = fptrunc float %1318 to bfloat, !dbg !39 + %1447 = fptrunc float %1319 to bfloat, !dbg !39 + %1448 = fptrunc float %1320 to bfloat, !dbg !39 + %1449 = fptrunc float %1321 to bfloat, !dbg !39 + %1450 = fptrunc float %1322 to bfloat, !dbg !39 + %1451 = fptrunc float %1323 to bfloat, !dbg !39 + %1452 = fptrunc float %1324 to bfloat, !dbg !39 + %1453 = fptrunc float %1325 to bfloat, !dbg !39 + %1454 = fptrunc float %1326 to bfloat, !dbg !39 + %1455 = fptrunc float %1327 to bfloat, !dbg !39 + %1456 = fptrunc float %1328 to bfloat, !dbg !39 + %1457 = fptrunc float %1329 to bfloat, !dbg !39 + %1458 = fptrunc float %1330 to bfloat, !dbg !39 + %1459 = fptrunc float %1331 to bfloat, !dbg !39 + %1460 = fptrunc float %1332 to bfloat, !dbg !39 + %1461 = fptrunc float %1333 to bfloat, !dbg !39 + %1462 = fptrunc float %1334 to bfloat, !dbg !39 + %1463 = fptrunc float %1335 to bfloat, !dbg !39 + %1464 = fptrunc float %1336 to bfloat, !dbg !39 + %1465 = fptrunc float %1337 to bfloat, !dbg !39 + %1466 = fptrunc float %1338 to bfloat, !dbg !39 + %1467 = fptrunc float %1339 to bfloat, !dbg !39 + %1468 = fptrunc float %1340 to bfloat, !dbg !39 + %1469 = fptrunc float %1341 to bfloat, !dbg !39 + %1470 = fptrunc float %1342 to bfloat, !dbg !39 + %1471 = fptrunc float %1343 to bfloat, !dbg !39 + %1472 = fptrunc float %1344 to bfloat, !dbg !39 + %1473 = fptrunc float %1345 to bfloat, !dbg !39 + %1474 = fptrunc float %1346 to bfloat, !dbg !39 + %1475 = fptrunc float %1347 to bfloat, !dbg !39 + %1476 = fptrunc float %1348 to bfloat, !dbg !39 + %1477 = fptrunc float %1349 to bfloat, !dbg !39 + %1478 = fptrunc float %1350 to bfloat, !dbg !39 + %1479 = fptrunc float %1351 to bfloat, !dbg !39 + %1480 = fptrunc float %1352 to bfloat, !dbg !39 + %1481 = fptrunc float %1353 to bfloat, !dbg !39 + %1482 = fptrunc float %1354 to bfloat, !dbg !39 + %1483 = fptrunc float %1355 to bfloat, !dbg !39 + %1484 = fptrunc float %1356 to bfloat, !dbg !39 + %1485 = fptrunc float %1357 to bfloat, !dbg !39 + %1486 = fptrunc float %1358 to bfloat, !dbg !39 + %1487 = fptrunc float %1359 to bfloat, !dbg !39 + %1488 = fptrunc float %1360 to bfloat, !dbg !39 + %1489 = fptrunc float %1361 to bfloat, !dbg !39 + %1490 = fptrunc float %1362 to bfloat, !dbg !39 + %1491 = fptrunc float %1363 to bfloat, !dbg !39 + %1492 = fptrunc float %1364 to bfloat, !dbg !39 + %1493 = fptrunc float %1365 to bfloat, !dbg !39 + %1494 = fptrunc float %1366 to bfloat, !dbg !39 + %1495 = fptrunc float %1367 to bfloat, !dbg !39 + %1496 = fptrunc float %1368 to bfloat, !dbg !39 + %1497 = fptrunc float %1369 to bfloat, !dbg !39 + %1498 = fptrunc float %1370 to bfloat, !dbg !39 + %1499 = fptrunc float %1371 to bfloat, !dbg !39 + %1500 = fptrunc float %1372 to bfloat, !dbg !39 + %1501 = fptrunc float %1373 to bfloat, !dbg !39 + %1502 = fptrunc float %1374 to bfloat, !dbg !39 + %1503 = fptrunc float %1375 to bfloat, !dbg !39 + %1504 = fptrunc float %1376 to bfloat, !dbg !39 + %1505 = fptrunc float %1377 to bfloat, !dbg !39 + %1506 = fptrunc float %1378 to bfloat, !dbg !39 + %1507 = fptrunc float %1379 to bfloat, !dbg !39 + %1508 = fptrunc float %1380 to bfloat, !dbg !39 + %1509 = fptrunc float %1381 to bfloat, !dbg !39 + %1510 = fptrunc float %1382 to bfloat, !dbg !39 + %1511 = fptrunc float %1383 to bfloat, !dbg !39 + %1512 = fptrunc float %1384 to bfloat, !dbg !39 + %1513 = fptrunc float %1385 to bfloat, !dbg !39 + %1514 = fptrunc float %1386 to bfloat, !dbg !39 + %1515 = fptrunc float %1387 to bfloat, !dbg !39 + %1516 = fptrunc float %1388 to bfloat, !dbg !39 + %1517 = fptrunc float %1389 to bfloat, !dbg !39 + %1518 = fptrunc float %1390 to bfloat, !dbg !39 + %1519 = fptrunc float %1391 to bfloat, !dbg !39 + %1520 = fptrunc float %1392 to bfloat, !dbg !39 + %1521 = fptrunc float %1393 to bfloat, !dbg !39 + %1522 = fptrunc float %1394 to bfloat, !dbg !39 + %1523 = fptrunc float %1395 to bfloat, !dbg !39 + %1524 = fptrunc float %1396 to bfloat, !dbg !39 + %1525 = fptrunc float %1397 to bfloat, !dbg !39 + %1526 = fptrunc float %1398 to bfloat, !dbg !39 + %1527 = fptrunc float %1399 to bfloat, !dbg !39 + %1528 = fptrunc float %1400 to bfloat, !dbg !39 + %1529 = fptrunc float %1401 to bfloat, !dbg !39 + %1530 = fptrunc float %1402 to bfloat, !dbg !39 + %1531 = fptrunc float %1403 to bfloat, !dbg !39 + %1532 = fptrunc float %1404 to bfloat, !dbg !39 + %1533 = fptrunc float %1405 to bfloat, !dbg !39 + %1534 = insertelement <2 x bfloat> poison, bfloat %1406, i64 0, !dbg !40 + %1535 = insertelement <2 x bfloat> %1534, bfloat %1407, i64 1, !dbg !40 + %1536 = bitcast <2 x bfloat> %1535 to i32, !dbg !40 + %1537 = insertelement <2 x bfloat> poison, bfloat %1408, i64 0, !dbg !40 + %1538 = insertelement <2 x bfloat> %1537, bfloat %1409, i64 1, !dbg !40 + %1539 = bitcast <2 x bfloat> %1538 to i32, !dbg !40 + %1540 = insertelement <2 x bfloat> poison, bfloat %1410, i64 0, !dbg !40 + %1541 = insertelement <2 x bfloat> %1540, bfloat %1411, i64 1, !dbg !40 + %1542 = bitcast <2 x bfloat> %1541 to i32, !dbg !40 + %1543 = insertelement <2 x bfloat> poison, bfloat %1412, i64 0, !dbg !40 + %1544 = insertelement <2 x bfloat> %1543, bfloat %1413, i64 1, !dbg !40 + %1545 = bitcast <2 x bfloat> %1544 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1536, i32 %1539, i32 %1542, i32 %1545, ptr addrspace(1) %1260, i1 %104) #6, !dbg !40 + %1546 = insertelement <2 x bfloat> poison, bfloat %1414, i64 0, !dbg !40 + %1547 = insertelement <2 x bfloat> %1546, bfloat %1415, i64 1, !dbg !40 + %1548 = bitcast <2 x bfloat> %1547 to i32, !dbg !40 + %1549 = insertelement <2 x bfloat> poison, bfloat %1416, i64 0, !dbg !40 + %1550 = insertelement <2 x bfloat> %1549, bfloat %1417, i64 1, !dbg !40 + %1551 = bitcast <2 x bfloat> %1550 to i32, !dbg !40 + %1552 = insertelement <2 x bfloat> poison, bfloat %1418, i64 0, !dbg !40 + %1553 = insertelement <2 x bfloat> %1552, bfloat %1419, i64 1, !dbg !40 + %1554 = bitcast <2 x bfloat> %1553 to i32, !dbg !40 + %1555 = insertelement <2 x bfloat> poison, bfloat %1420, i64 0, !dbg !40 + %1556 = insertelement <2 x bfloat> %1555, bfloat %1421, i64 1, !dbg !40 + %1557 = bitcast <2 x bfloat> %1556 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1548, i32 %1551, i32 %1554, i32 %1557, ptr addrspace(1) %1258, i1 %122) #6, !dbg !40 + %1558 = insertelement <2 x bfloat> poison, bfloat %1422, i64 0, !dbg !40 + %1559 = insertelement <2 x bfloat> %1558, bfloat %1423, i64 1, !dbg !40 + %1560 = bitcast <2 x bfloat> %1559 to i32, !dbg !40 + %1561 = insertelement <2 x bfloat> poison, bfloat %1424, i64 0, !dbg !40 + %1562 = insertelement <2 x bfloat> %1561, bfloat %1425, i64 1, !dbg !40 + %1563 = bitcast <2 x bfloat> %1562 to i32, !dbg !40 + %1564 = insertelement <2 x bfloat> poison, bfloat %1426, i64 0, !dbg !40 + %1565 = insertelement <2 x bfloat> %1564, bfloat %1427, i64 1, !dbg !40 + %1566 = bitcast <2 x bfloat> %1565 to i32, !dbg !40 + %1567 = insertelement <2 x bfloat> poison, bfloat %1428, i64 0, !dbg !40 + %1568 = insertelement <2 x bfloat> %1567, bfloat %1429, i64 1, !dbg !40 + %1569 = bitcast <2 x bfloat> %1568 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1560, i32 %1563, i32 %1566, i32 %1569, ptr addrspace(1) %1256, i1 %140) #6, !dbg !40 + %1570 = insertelement <2 x bfloat> poison, bfloat %1430, i64 0, !dbg !40 + %1571 = insertelement <2 x bfloat> %1570, bfloat %1431, i64 1, !dbg !40 + %1572 = bitcast <2 x bfloat> %1571 to i32, !dbg !40 + %1573 = insertelement <2 x bfloat> poison, bfloat %1432, i64 0, !dbg !40 + %1574 = insertelement <2 x bfloat> %1573, bfloat %1433, i64 1, !dbg !40 + %1575 = bitcast <2 x bfloat> %1574 to i32, !dbg !40 + %1576 = insertelement <2 x bfloat> poison, bfloat %1434, i64 0, !dbg !40 + %1577 = insertelement <2 x bfloat> %1576, bfloat %1435, i64 1, !dbg !40 + %1578 = bitcast <2 x bfloat> %1577 to i32, !dbg !40 + %1579 = insertelement <2 x bfloat> poison, bfloat %1436, i64 0, !dbg !40 + %1580 = insertelement <2 x bfloat> %1579, bfloat %1437, i64 1, !dbg !40 + %1581 = bitcast <2 x bfloat> %1580 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1572, i32 %1575, i32 %1578, i32 %1581, ptr addrspace(1) %1254, i1 %158) #6, !dbg !40 + %1582 = insertelement <2 x bfloat> poison, bfloat %1438, i64 0, !dbg !40 + %1583 = insertelement <2 x bfloat> %1582, bfloat %1439, i64 1, !dbg !40 + %1584 = bitcast <2 x bfloat> %1583 to i32, !dbg !40 + %1585 = insertelement <2 x bfloat> poison, bfloat %1440, i64 0, !dbg !40 + %1586 = insertelement <2 x bfloat> %1585, bfloat %1441, i64 1, !dbg !40 + %1587 = bitcast <2 x bfloat> %1586 to i32, !dbg !40 + %1588 = insertelement <2 x bfloat> poison, bfloat %1442, i64 0, !dbg !40 + %1589 = insertelement <2 x bfloat> %1588, bfloat %1443, i64 1, !dbg !40 + %1590 = bitcast <2 x bfloat> %1589 to i32, !dbg !40 + %1591 = insertelement <2 x bfloat> poison, bfloat %1444, i64 0, !dbg !40 + %1592 = insertelement <2 x bfloat> %1591, bfloat %1445, i64 1, !dbg !40 + %1593 = bitcast <2 x bfloat> %1592 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1584, i32 %1587, i32 %1590, i32 %1593, ptr addrspace(1) %1252, i1 %176) #6, !dbg !40 + %1594 = insertelement <2 x bfloat> poison, bfloat %1446, i64 0, !dbg !40 + %1595 = insertelement <2 x bfloat> %1594, bfloat %1447, i64 1, !dbg !40 + %1596 = bitcast <2 x bfloat> %1595 to i32, !dbg !40 + %1597 = insertelement <2 x bfloat> poison, bfloat %1448, i64 0, !dbg !40 + %1598 = insertelement <2 x bfloat> %1597, bfloat %1449, i64 1, !dbg !40 + %1599 = bitcast <2 x bfloat> %1598 to i32, !dbg !40 + %1600 = insertelement <2 x bfloat> poison, bfloat %1450, i64 0, !dbg !40 + %1601 = insertelement <2 x bfloat> %1600, bfloat %1451, i64 1, !dbg !40 + %1602 = bitcast <2 x bfloat> %1601 to i32, !dbg !40 + %1603 = insertelement <2 x bfloat> poison, bfloat %1452, i64 0, !dbg !40 + %1604 = insertelement <2 x bfloat> %1603, bfloat %1453, i64 1, !dbg !40 + %1605 = bitcast <2 x bfloat> %1604 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1596, i32 %1599, i32 %1602, i32 %1605, ptr addrspace(1) %1250, i1 %194) #6, !dbg !40 + %1606 = insertelement <2 x bfloat> poison, bfloat %1454, i64 0, !dbg !40 + %1607 = insertelement <2 x bfloat> %1606, bfloat %1455, i64 1, !dbg !40 + %1608 = bitcast <2 x bfloat> %1607 to i32, !dbg !40 + %1609 = insertelement <2 x bfloat> poison, bfloat %1456, i64 0, !dbg !40 + %1610 = insertelement <2 x bfloat> %1609, bfloat %1457, i64 1, !dbg !40 + %1611 = bitcast <2 x bfloat> %1610 to i32, !dbg !40 + %1612 = insertelement <2 x bfloat> poison, bfloat %1458, i64 0, !dbg !40 + %1613 = insertelement <2 x bfloat> %1612, bfloat %1459, i64 1, !dbg !40 + %1614 = bitcast <2 x bfloat> %1613 to i32, !dbg !40 + %1615 = insertelement <2 x bfloat> poison, bfloat %1460, i64 0, !dbg !40 + %1616 = insertelement <2 x bfloat> %1615, bfloat %1461, i64 1, !dbg !40 + %1617 = bitcast <2 x bfloat> %1616 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1608, i32 %1611, i32 %1614, i32 %1617, ptr addrspace(1) %1248, i1 %212) #6, !dbg !40 + %1618 = insertelement <2 x bfloat> poison, bfloat %1462, i64 0, !dbg !40 + %1619 = insertelement <2 x bfloat> %1618, bfloat %1463, i64 1, !dbg !40 + %1620 = bitcast <2 x bfloat> %1619 to i32, !dbg !40 + %1621 = insertelement <2 x bfloat> poison, bfloat %1464, i64 0, !dbg !40 + %1622 = insertelement <2 x bfloat> %1621, bfloat %1465, i64 1, !dbg !40 + %1623 = bitcast <2 x bfloat> %1622 to i32, !dbg !40 + %1624 = insertelement <2 x bfloat> poison, bfloat %1466, i64 0, !dbg !40 + %1625 = insertelement <2 x bfloat> %1624, bfloat %1467, i64 1, !dbg !40 + %1626 = bitcast <2 x bfloat> %1625 to i32, !dbg !40 + %1627 = insertelement <2 x bfloat> poison, bfloat %1468, i64 0, !dbg !40 + %1628 = insertelement <2 x bfloat> %1627, bfloat %1469, i64 1, !dbg !40 + %1629 = bitcast <2 x bfloat> %1628 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1620, i32 %1623, i32 %1626, i32 %1629, ptr addrspace(1) %1246, i1 %230) #6, !dbg !40 + %1630 = insertelement <2 x bfloat> poison, bfloat %1470, i64 0, !dbg !40 + %1631 = insertelement <2 x bfloat> %1630, bfloat %1471, i64 1, !dbg !40 + %1632 = bitcast <2 x bfloat> %1631 to i32, !dbg !40 + %1633 = insertelement <2 x bfloat> poison, bfloat %1472, i64 0, !dbg !40 + %1634 = insertelement <2 x bfloat> %1633, bfloat %1473, i64 1, !dbg !40 + %1635 = bitcast <2 x bfloat> %1634 to i32, !dbg !40 + %1636 = insertelement <2 x bfloat> poison, bfloat %1474, i64 0, !dbg !40 + %1637 = insertelement <2 x bfloat> %1636, bfloat %1475, i64 1, !dbg !40 + %1638 = bitcast <2 x bfloat> %1637 to i32, !dbg !40 + %1639 = insertelement <2 x bfloat> poison, bfloat %1476, i64 0, !dbg !40 + %1640 = insertelement <2 x bfloat> %1639, bfloat %1477, i64 1, !dbg !40 + %1641 = bitcast <2 x bfloat> %1640 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1632, i32 %1635, i32 %1638, i32 %1641, ptr addrspace(1) %1244, i1 %248) #6, !dbg !40 + %1642 = insertelement <2 x bfloat> poison, bfloat %1478, i64 0, !dbg !40 + %1643 = insertelement <2 x bfloat> %1642, bfloat %1479, i64 1, !dbg !40 + %1644 = bitcast <2 x bfloat> %1643 to i32, !dbg !40 + %1645 = insertelement <2 x bfloat> poison, bfloat %1480, i64 0, !dbg !40 + %1646 = insertelement <2 x bfloat> %1645, bfloat %1481, i64 1, !dbg !40 + %1647 = bitcast <2 x bfloat> %1646 to i32, !dbg !40 + %1648 = insertelement <2 x bfloat> poison, bfloat %1482, i64 0, !dbg !40 + %1649 = insertelement <2 x bfloat> %1648, bfloat %1483, i64 1, !dbg !40 + %1650 = bitcast <2 x bfloat> %1649 to i32, !dbg !40 + %1651 = insertelement <2 x bfloat> poison, bfloat %1484, i64 0, !dbg !40 + %1652 = insertelement <2 x bfloat> %1651, bfloat %1485, i64 1, !dbg !40 + %1653 = bitcast <2 x bfloat> %1652 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1644, i32 %1647, i32 %1650, i32 %1653, ptr addrspace(1) %1242, i1 %266) #6, !dbg !40 + %1654 = insertelement <2 x bfloat> poison, bfloat %1486, i64 0, !dbg !40 + %1655 = insertelement <2 x bfloat> %1654, bfloat %1487, i64 1, !dbg !40 + %1656 = bitcast <2 x bfloat> %1655 to i32, !dbg !40 + %1657 = insertelement <2 x bfloat> poison, bfloat %1488, i64 0, !dbg !40 + %1658 = insertelement <2 x bfloat> %1657, bfloat %1489, i64 1, !dbg !40 + %1659 = bitcast <2 x bfloat> %1658 to i32, !dbg !40 + %1660 = insertelement <2 x bfloat> poison, bfloat %1490, i64 0, !dbg !40 + %1661 = insertelement <2 x bfloat> %1660, bfloat %1491, i64 1, !dbg !40 + %1662 = bitcast <2 x bfloat> %1661 to i32, !dbg !40 + %1663 = insertelement <2 x bfloat> poison, bfloat %1492, i64 0, !dbg !40 + %1664 = insertelement <2 x bfloat> %1663, bfloat %1493, i64 1, !dbg !40 + %1665 = bitcast <2 x bfloat> %1664 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1656, i32 %1659, i32 %1662, i32 %1665, ptr addrspace(1) %1240, i1 %284) #6, !dbg !40 + %1666 = insertelement <2 x bfloat> poison, bfloat %1494, i64 0, !dbg !40 + %1667 = insertelement <2 x bfloat> %1666, bfloat %1495, i64 1, !dbg !40 + %1668 = bitcast <2 x bfloat> %1667 to i32, !dbg !40 + %1669 = insertelement <2 x bfloat> poison, bfloat %1496, i64 0, !dbg !40 + %1670 = insertelement <2 x bfloat> %1669, bfloat %1497, i64 1, !dbg !40 + %1671 = bitcast <2 x bfloat> %1670 to i32, !dbg !40 + %1672 = insertelement <2 x bfloat> poison, bfloat %1498, i64 0, !dbg !40 + %1673 = insertelement <2 x bfloat> %1672, bfloat %1499, i64 1, !dbg !40 + %1674 = bitcast <2 x bfloat> %1673 to i32, !dbg !40 + %1675 = insertelement <2 x bfloat> poison, bfloat %1500, i64 0, !dbg !40 + %1676 = insertelement <2 x bfloat> %1675, bfloat %1501, i64 1, !dbg !40 + %1677 = bitcast <2 x bfloat> %1676 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1668, i32 %1671, i32 %1674, i32 %1677, ptr addrspace(1) %1238, i1 %302) #6, !dbg !40 + %1678 = insertelement <2 x bfloat> poison, bfloat %1502, i64 0, !dbg !40 + %1679 = insertelement <2 x bfloat> %1678, bfloat %1503, i64 1, !dbg !40 + %1680 = bitcast <2 x bfloat> %1679 to i32, !dbg !40 + %1681 = insertelement <2 x bfloat> poison, bfloat %1504, i64 0, !dbg !40 + %1682 = insertelement <2 x bfloat> %1681, bfloat %1505, i64 1, !dbg !40 + %1683 = bitcast <2 x bfloat> %1682 to i32, !dbg !40 + %1684 = insertelement <2 x bfloat> poison, bfloat %1506, i64 0, !dbg !40 + %1685 = insertelement <2 x bfloat> %1684, bfloat %1507, i64 1, !dbg !40 + %1686 = bitcast <2 x bfloat> %1685 to i32, !dbg !40 + %1687 = insertelement <2 x bfloat> poison, bfloat %1508, i64 0, !dbg !40 + %1688 = insertelement <2 x bfloat> %1687, bfloat %1509, i64 1, !dbg !40 + %1689 = bitcast <2 x bfloat> %1688 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1680, i32 %1683, i32 %1686, i32 %1689, ptr addrspace(1) %1236, i1 %320) #6, !dbg !40 + %1690 = insertelement <2 x bfloat> poison, bfloat %1510, i64 0, !dbg !40 + %1691 = insertelement <2 x bfloat> %1690, bfloat %1511, i64 1, !dbg !40 + %1692 = bitcast <2 x bfloat> %1691 to i32, !dbg !40 + %1693 = insertelement <2 x bfloat> poison, bfloat %1512, i64 0, !dbg !40 + %1694 = insertelement <2 x bfloat> %1693, bfloat %1513, i64 1, !dbg !40 + %1695 = bitcast <2 x bfloat> %1694 to i32, !dbg !40 + %1696 = insertelement <2 x bfloat> poison, bfloat %1514, i64 0, !dbg !40 + %1697 = insertelement <2 x bfloat> %1696, bfloat %1515, i64 1, !dbg !40 + %1698 = bitcast <2 x bfloat> %1697 to i32, !dbg !40 + %1699 = insertelement <2 x bfloat> poison, bfloat %1516, i64 0, !dbg !40 + %1700 = insertelement <2 x bfloat> %1699, bfloat %1517, i64 1, !dbg !40 + %1701 = bitcast <2 x bfloat> %1700 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1692, i32 %1695, i32 %1698, i32 %1701, ptr addrspace(1) %1234, i1 %338) #6, !dbg !40 + %1702 = insertelement <2 x bfloat> poison, bfloat %1518, i64 0, !dbg !40 + %1703 = insertelement <2 x bfloat> %1702, bfloat %1519, i64 1, !dbg !40 + %1704 = bitcast <2 x bfloat> %1703 to i32, !dbg !40 + %1705 = insertelement <2 x bfloat> poison, bfloat %1520, i64 0, !dbg !40 + %1706 = insertelement <2 x bfloat> %1705, bfloat %1521, i64 1, !dbg !40 + %1707 = bitcast <2 x bfloat> %1706 to i32, !dbg !40 + %1708 = insertelement <2 x bfloat> poison, bfloat %1522, i64 0, !dbg !40 + %1709 = insertelement <2 x bfloat> %1708, bfloat %1523, i64 1, !dbg !40 + %1710 = bitcast <2 x bfloat> %1709 to i32, !dbg !40 + %1711 = insertelement <2 x bfloat> poison, bfloat %1524, i64 0, !dbg !40 + %1712 = insertelement <2 x bfloat> %1711, bfloat %1525, i64 1, !dbg !40 + %1713 = bitcast <2 x bfloat> %1712 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1704, i32 %1707, i32 %1710, i32 %1713, ptr addrspace(1) %1232, i1 %356) #6, !dbg !40 + %1714 = insertelement <2 x bfloat> poison, bfloat %1526, i64 0, !dbg !40 + %1715 = insertelement <2 x bfloat> %1714, bfloat %1527, i64 1, !dbg !40 + %1716 = bitcast <2 x bfloat> %1715 to i32, !dbg !40 + %1717 = insertelement <2 x bfloat> poison, bfloat %1528, i64 0, !dbg !40 + %1718 = insertelement <2 x bfloat> %1717, bfloat %1529, i64 1, !dbg !40 + %1719 = bitcast <2 x bfloat> %1718 to i32, !dbg !40 + %1720 = insertelement <2 x bfloat> poison, bfloat %1530, i64 0, !dbg !40 + %1721 = insertelement <2 x bfloat> %1720, bfloat %1531, i64 1, !dbg !40 + %1722 = bitcast <2 x bfloat> %1721 to i32, !dbg !40 + %1723 = insertelement <2 x bfloat> poison, bfloat %1532, i64 0, !dbg !40 + %1724 = insertelement <2 x bfloat> %1723, bfloat %1533, i64 1, !dbg !40 + %1725 = bitcast <2 x bfloat> %1724 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1716, i32 %1719, i32 %1722, i32 %1725, ptr addrspace(1) %1230, i1 %374) #6, !dbg !40 + %1726 = shl nuw nsw i32 %12, 1, !dbg !41 + %1727 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %1726, !dbg !41 + %1728 = bitcast float %1262 to i32, !dbg !41 + %1729 = bitcast float %1263 to i32, !dbg !41 + %1730 = bitcast float %1264 to i32, !dbg !41 + %1731 = bitcast float %1265 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %1727, i32 %1728, i32 %1729, i32 %1730, i32 %1731) #6, !dbg !41 + %1732 = getelementptr inbounds nuw i8, ptr addrspace(3) %1727, i32 16, !dbg !41 + %1733 = bitcast float %1266 to i32, !dbg !41 + %1734 = bitcast float %1267 to i32, !dbg !41 + %1735 = bitcast float %1268 to i32, !dbg !41 + %1736 = bitcast float %1269 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %1732, i32 %1733, i32 %1734, i32 %1735, i32 %1736) #6, !dbg !41 + %1737 = getelementptr inbounds nuw i8, ptr addrspace(3) %1727, i32 128, !dbg !41 + %1738 = bitcast float %1270 to i32, !dbg !41 + %1739 = bitcast float %1271 to i32, !dbg !41 + %1740 = bitcast float %1272 to i32, !dbg !41 + %1741 = bitcast float %1273 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %1737, i32 %1738, i32 %1739, i32 %1740, i32 %1741) #6, !dbg !41 + %1742 = getelementptr inbounds nuw i8, ptr addrspace(3) %1727, i32 144, !dbg !41 + %1743 = bitcast float %1274 to i32, !dbg !41 + %1744 = bitcast float %1275 to i32, !dbg !41 + %1745 = bitcast float %1276 to i32, !dbg !41 + %1746 = bitcast float %1277 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %1742, i32 %1743, i32 %1744, i32 %1745, i32 %1746) #6, !dbg !41 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !41 + %1747 = shl nuw nsw i32 %11, 5, !dbg !41 + %1748 = and i32 %1747, 96, !dbg !41 + %1749 = and i32 %11, 28, !dbg !41 + %1750 = shl nuw nsw i32 %11, 2, !dbg !41 + %1751 = and i32 %1750, 128, !dbg !41 + %1752 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %1748, !dbg !41 + %1753 = getelementptr inbounds nuw i8, ptr addrspace(3) %1752, i32 %1751, !dbg !41 + %1754 = getelementptr inbounds nuw i8, ptr addrspace(3) %1753, i32 %1749, !dbg !41 + %1755 = load i32, ptr addrspace(3) %1754, align 4, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %1755, ptr addrspace(1) %1228, i1 %1261) #6, !dbg !41 + ret void, !dbg !42 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.div.full(float, float) #3 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #4 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #5 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.f(float) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="64" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) } +attributes #4 = { convergent nocallback nounwind } +attributes #5 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #6 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "l2norm_fwd_kernel", linkageName: "l2norm_fwd_kernel", scope: !1, file: !1, line: 69, type: !6, scopeLine: 69, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !10, !11, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 70, column: 24, scope: !5) +!14 = !DILocation(line: 70, column: 30, scope: !5) +!15 = !DILocation(line: 71, column: 16, scope: !5) +!16 = !DILocation(line: 71, column: 34, scope: !5) +!17 = !DILocation(line: 71, column: 21, scope: !5) +!18 = !DILocation(line: 73, column: 16, scope: !5) +!19 = !DILocation(line: 74, column: 30, scope: !5) +!20 = !DILocation(line: 75, column: 29, scope: !5) +!21 = !DILocation(line: 75, column: 14, scope: !5) +!22 = !DILocation(line: 75, column: 33, scope: !5) +!23 = !DILocation(line: 79, column: 18, scope: !5) +!24 = !DILocation(line: 79, column: 47, scope: !5) +!25 = !DILocation(line: 80, column: 38, scope: !5) +!26 = !DILocation(line: 263, column: 15, scope: !27, inlinedAt: !29) +!27 = distinct !DILexicalBlockFile(scope: !5, file: !28, discriminator: 0) +!28 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!29 = !DILocation(line: 293, column: 36, scope: !27, inlinedAt: !30) +!30 = !DILocation(line: 80, column: 43, scope: !31) +!31 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!32 = !DILocation(line: 80, column: 48, scope: !5) +!33 = !DILocation(line: 80, column: 25, scope: !5) +!34 = !DILocation(line: 77, column: 20, scope: !5) +!35 = !DILocation(line: 76, column: 14, scope: !5) +!36 = !DILocation(line: 76, column: 33, scope: !5) +!37 = !DILocation(line: 80, column: 17, scope: !5) +!38 = !DILocation(line: 81, column: 16, scope: !5) +!39 = !DILocation(line: 83, column: 25, scope: !5) +!40 = !DILocation(line: 83, column: 18, scope: !5) +!41 = !DILocation(line: 84, column: 21, scope: !5) +!42 = !DILocation(line: 84, column: 4, scope: !5) diff --git a/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/l2norm_fwd_kernel.ptx b/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/l2norm_fwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..a639d4f0c3e5907e14b8e26f48642ed216e80526 --- /dev/null +++ b/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/l2norm_fwd_kernel.ptx @@ -0,0 +1,1614 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_fwd_kernel // -- Begin function l2norm_fwd_kernel +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; +.global .align 1 .b8 _$_str_$_2[17] = {95, 95, 67, 85, 68, 65, 95, 80, 82, 69, 67, 95, 83, 81, 82, 84}; + // @l2norm_fwd_kernel +.visible .entry l2norm_fwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_2, + .param .f32 l2norm_fwd_kernel_param_3, + .param .u32 l2norm_fwd_kernel_param_4, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_6 +) +.reqntid 64 +{ + .reg .pred %p<18>; + .reg .b16 %rs<129>; + .reg .b32 %r<729>; + .reg .b64 %rd<123>; + .loc 1 69 0 // l2norm.py:69:0 +$L__func_begin0: + .loc 1 69 0 // l2norm.py:69:0 + +// %bb.0: // %__nv_sqrtf.exit + ld.param.b64 %rd34, [l2norm_fwd_kernel_param_0]; + ld.param.b64 %rd35, [l2norm_fwd_kernel_param_1]; +$L__tmp0: + .loc 1 70 24 // l2norm.py:70:24 + mov.u32 %r151, %ctaid.x; + ld.param.b64 %rd36, [l2norm_fwd_kernel_param_2]; + .loc 1 71 16 // l2norm.py:71:16 + mul.wide.u32 %rd37, %r151, 64; + ld.param.b32 %r152, [l2norm_fwd_kernel_param_3]; + ld.param.s32 %rd38, [l2norm_fwd_kernel_param_4]; + .loc 1 71 34 // l2norm.py:71:34 + mov.u32 %r153, %tid.x; + and.b32 %r154, %r153, 48; + bfe.u32 %r155, %r153, 4, 2; + or.b32 %r156, %r155, 4; + .loc 1 71 21 // l2norm.py:71:21 + or.b32 %r157, %r155, 32; + or.b32 %r158, %r155, 36; + or.b32 %r159, %r155, 40; + or.b32 %r160, %r155, 44; + or.b32 %r161, %r155, 48; + or.b32 %r162, %r155, 52; + or.b32 %r163, %r155, 56; + or.b32 %r164, %r155, 60; + or.b32 %r165, %r155, 16; + or.b32 %r166, %r155, 20; + or.b32 %r167, %r155, 24; + or.b32 %r168, %r155, 28; + or.b32 %r169, %r155, 8; + or.b32 %r170, %r155, 12; + cvt.u64.u32 %rd39, %r164; + cvt.u64.u32 %rd40, %r163; + cvt.u64.u32 %rd41, %r162; + cvt.u64.u32 %rd42, %r161; + cvt.u64.u32 %rd43, %r160; + cvt.u64.u32 %rd44, %r159; + cvt.u64.u32 %rd45, %r158; + cvt.u64.u32 %rd46, %r157; + cvt.u64.u32 %rd47, %r168; + cvt.u64.u32 %rd48, %r167; + cvt.u64.u32 %rd49, %r166; + cvt.u64.u32 %rd50, %r165; + cvt.u64.u32 %rd51, %r170; + cvt.u64.u32 %rd52, %r169; + cvt.u64.u32 %rd53, %r156; + cvt.u64.u32 %rd54, %r155; + or.b64 %rd55, %rd37, %rd54; + or.b64 %rd56, %rd37, %rd53; + or.b64 %rd57, %rd37, %rd52; + or.b64 %rd58, %rd37, %rd51; + or.b64 %rd59, %rd37, %rd50; + or.b64 %rd60, %rd37, %rd49; + or.b64 %rd61, %rd37, %rd48; + or.b64 %rd62, %rd37, %rd47; + or.b64 %rd63, %rd37, %rd46; + or.b64 %rd64, %rd37, %rd45; + or.b64 %rd65, %rd37, %rd44; + or.b64 %rd66, %rd37, %rd43; + or.b64 %rd67, %rd37, %rd42; + or.b64 %rd68, %rd37, %rd41; + or.b64 %rd69, %rd37, %rd40; + or.b64 %rd70, %rd37, %rd39; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p16, %rd70, %rd38; + setp.lt.s64 %p14, %rd69, %rd38; + setp.lt.s64 %p13, %rd68, %rd38; + setp.lt.s64 %p12, %rd67, %rd38; + setp.lt.s64 %p11, %rd66, %rd38; + setp.lt.s64 %p10, %rd65, %rd38; + setp.lt.s64 %p9, %rd64, %rd38; + setp.lt.s64 %p8, %rd63, %rd38; + setp.lt.s64 %p15, %rd62, %rd38; + setp.lt.s64 %p7, %rd61, %rd38; + setp.lt.s64 %p6, %rd60, %rd38; + setp.lt.s64 %p5, %rd59, %rd38; + setp.lt.s64 %p4, %rd58, %rd38; + setp.lt.s64 %p3, %rd57, %rd38; + setp.lt.s64 %p2, %rd56, %rd38; + setp.lt.s64 %p1, %rd55, %rd38; + .loc 1 74 30 // l2norm.py:74:30 + shl.b32 %r171, %r153, 3; + and.b32 %r172, %r171, 120; + .loc 1 75 14 // l2norm.py:75:14 + shl.b64 %rd71, %rd55, 8; + add.s64 %rd72, %rd34, %rd71; + shl.b64 %rd73, %rd56, 8; + add.s64 %rd74, %rd34, %rd73; + shl.b64 %rd75, %rd57, 8; + add.s64 %rd76, %rd34, %rd75; + shl.b64 %rd77, %rd58, 8; + add.s64 %rd78, %rd34, %rd77; + shl.b64 %rd79, %rd59, 8; + add.s64 %rd80, %rd34, %rd79; + shl.b64 %rd81, %rd60, 8; + add.s64 %rd82, %rd34, %rd81; + shl.b64 %rd83, %rd61, 8; + add.s64 %rd84, %rd34, %rd83; + shl.b64 %rd85, %rd62, 8; + add.s64 %rd86, %rd34, %rd85; + shl.b64 %rd87, %rd63, 8; + add.s64 %rd88, %rd34, %rd87; + shl.b64 %rd89, %rd64, 8; + add.s64 %rd90, %rd34, %rd89; + shl.b64 %rd91, %rd65, 8; + add.s64 %rd92, %rd34, %rd91; + shl.b64 %rd93, %rd66, 8; + add.s64 %rd94, %rd34, %rd93; + shl.b64 %rd95, %rd67, 8; + add.s64 %rd96, %rd34, %rd95; + shl.b64 %rd97, %rd68, 8; + add.s64 %rd98, %rd34, %rd97; + shl.b64 %rd99, %rd69, 8; + add.s64 %rd100, %rd34, %rd99; + shl.b64 %rd101, %rd70, 8; + add.s64 %rd102, %rd34, %rd101; + .loc 1 75 33 // l2norm.py:75:33 + mul.wide.u32 %rd103, %r172, 2; + add.s64 %rd1, %rd72, %rd103; + add.s64 %rd2, %rd74, %rd103; + add.s64 %rd3, %rd76, %rd103; + add.s64 %rd4, %rd78, %rd103; + add.s64 %rd5, %rd80, %rd103; + add.s64 %rd6, %rd82, %rd103; + add.s64 %rd7, %rd84, %rd103; + add.s64 %rd8, %rd86, %rd103; + add.s64 %rd9, %rd88, %rd103; + add.s64 %rd10, %rd90, %rd103; + add.s64 %rd11, %rd92, %rd103; + add.s64 %rd12, %rd94, %rd103; + add.s64 %rd13, %rd96, %rd103; + add.s64 %rd14, %rd98, %rd103; + add.s64 %rd15, %rd100, %rd103; + add.s64 %rd16, %rd102, %rd103; + mov.b32 %r5, 0; + .loc 1 79 18 // l2norm.py:79:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + mov.b32 {%rs5, %rs6}, %r3; + mov.b32 {%rs7, %rs8}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs9, %rs10}, %r6; + mov.b32 {%rs11, %rs12}, %r7; + mov.b32 {%rs13, %rs14}, %r8; + mov.b32 {%rs15, %rs16}, %r9; + // begin inline asm + mov.u32 %r10, %r5; + mov.u32 %r11, %r5; + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + @%p3 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs17, %rs18}, %r10; + mov.b32 {%rs19, %rs20}, %r11; + mov.b32 {%rs21, %rs22}, %r12; + mov.b32 {%rs23, %rs24}, %r13; + // begin inline asm + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + @%p4 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ]; + // end inline asm + mov.b32 {%rs25, %rs26}, %r14; + mov.b32 {%rs27, %rs28}, %r15; + mov.b32 {%rs29, %rs30}, %r16; + mov.b32 {%rs31, %rs32}, %r17; + // begin inline asm + mov.u32 %r18, %r5; + mov.u32 %r19, %r5; + mov.u32 %r20, %r5; + mov.u32 %r21, %r5; + @%p5 ld.global.v4.b32 { %r18, %r19, %r20, %r21 }, [ %rd5 + 0 ]; + // end inline asm + mov.b32 {%rs33, %rs34}, %r18; + mov.b32 {%rs35, %rs36}, %r19; + mov.b32 {%rs37, %rs38}, %r20; + mov.b32 {%rs39, %rs40}, %r21; + // begin inline asm + mov.u32 %r22, %r5; + mov.u32 %r23, %r5; + mov.u32 %r24, %r5; + mov.u32 %r25, %r5; + @%p6 ld.global.v4.b32 { %r22, %r23, %r24, %r25 }, [ %rd6 + 0 ]; + // end inline asm + mov.b32 {%rs41, %rs42}, %r22; + mov.b32 {%rs43, %rs44}, %r23; + mov.b32 {%rs45, %rs46}, %r24; + mov.b32 {%rs47, %rs48}, %r25; + // begin inline asm + mov.u32 %r26, %r5; + mov.u32 %r27, %r5; + mov.u32 %r28, %r5; + mov.u32 %r29, %r5; + @%p7 ld.global.v4.b32 { %r26, %r27, %r28, %r29 }, [ %rd7 + 0 ]; + // end inline asm + mov.b32 {%rs49, %rs50}, %r26; + mov.b32 {%rs51, %rs52}, %r27; + mov.b32 {%rs53, %rs54}, %r28; + mov.b32 {%rs55, %rs56}, %r29; + // begin inline asm + mov.u32 %r30, %r5; + mov.u32 %r31, %r5; + mov.u32 %r32, %r5; + mov.u32 %r33, %r5; + @%p15 ld.global.v4.b32 { %r30, %r31, %r32, %r33 }, [ %rd8 + 0 ]; + // end inline asm + mov.b32 {%rs57, %rs58}, %r30; + mov.b32 {%rs59, %rs60}, %r31; + mov.b32 {%rs61, %rs62}, %r32; + mov.b32 {%rs63, %rs64}, %r33; + // begin inline asm + mov.u32 %r34, %r5; + mov.u32 %r35, %r5; + mov.u32 %r36, %r5; + mov.u32 %r37, %r5; + @%p8 ld.global.v4.b32 { %r34, %r35, %r36, %r37 }, [ %rd9 + 0 ]; + // end inline asm + mov.b32 {%rs65, %rs66}, %r34; + mov.b32 {%rs67, %rs68}, %r35; + mov.b32 {%rs69, %rs70}, %r36; + mov.b32 {%rs71, %rs72}, %r37; + // begin inline asm + mov.u32 %r38, %r5; + mov.u32 %r39, %r5; + mov.u32 %r40, %r5; + mov.u32 %r41, %r5; + @%p9 ld.global.v4.b32 { %r38, %r39, %r40, %r41 }, [ %rd10 + 0 ]; + // end inline asm + mov.b32 {%rs73, %rs74}, %r38; + mov.b32 {%rs75, %rs76}, %r39; + mov.b32 {%rs77, %rs78}, %r40; + mov.b32 {%rs79, %rs80}, %r41; + // begin inline asm + mov.u32 %r42, %r5; + mov.u32 %r43, %r5; + mov.u32 %r44, %r5; + mov.u32 %r45, %r5; + @%p10 ld.global.v4.b32 { %r42, %r43, %r44, %r45 }, [ %rd11 + 0 ]; + // end inline asm + mov.b32 {%rs81, %rs82}, %r42; + mov.b32 {%rs83, %rs84}, %r43; + mov.b32 {%rs85, %rs86}, %r44; + mov.b32 {%rs87, %rs88}, %r45; + // begin inline asm + mov.u32 %r46, %r5; + mov.u32 %r47, %r5; + mov.u32 %r48, %r5; + mov.u32 %r49, %r5; + @%p11 ld.global.v4.b32 { %r46, %r47, %r48, %r49 }, [ %rd12 + 0 ]; + // end inline asm + mov.b32 {%rs89, %rs90}, %r46; + mov.b32 {%rs91, %rs92}, %r47; + mov.b32 {%rs93, %rs94}, %r48; + mov.b32 {%rs95, %rs96}, %r49; + // begin inline asm + mov.u32 %r50, %r5; + mov.u32 %r51, %r5; + mov.u32 %r52, %r5; + mov.u32 %r53, %r5; + @%p12 ld.global.v4.b32 { %r50, %r51, %r52, %r53 }, [ %rd13 + 0 ]; + // end inline asm + mov.b32 {%rs97, %rs98}, %r50; + mov.b32 {%rs99, %rs100}, %r51; + mov.b32 {%rs101, %rs102}, %r52; + mov.b32 {%rs103, %rs104}, %r53; + // begin inline asm + mov.u32 %r54, %r5; + mov.u32 %r55, %r5; + mov.u32 %r56, %r5; + mov.u32 %r57, %r5; + @%p13 ld.global.v4.b32 { %r54, %r55, %r56, %r57 }, [ %rd14 + 0 ]; + // end inline asm + mov.b32 {%rs105, %rs106}, %r54; + mov.b32 {%rs107, %rs108}, %r55; + mov.b32 {%rs109, %rs110}, %r56; + mov.b32 {%rs111, %rs112}, %r57; + // begin inline asm + mov.u32 %r58, %r5; + mov.u32 %r59, %r5; + mov.u32 %r60, %r5; + mov.u32 %r61, %r5; + @%p14 ld.global.v4.b32 { %r58, %r59, %r60, %r61 }, [ %rd15 + 0 ]; + // end inline asm + mov.b32 {%rs113, %rs114}, %r58; + mov.b32 {%rs115, %rs116}, %r59; + mov.b32 {%rs117, %rs118}, %r60; + mov.b32 {%rs119, %rs120}, %r61; + // begin inline asm + mov.u32 %r62, %r5; + mov.u32 %r63, %r5; + mov.u32 %r64, %r5; + mov.u32 %r65, %r5; + @%p16 ld.global.v4.b32 { %r62, %r63, %r64, %r65 }, [ %rd16 + 0 ]; + // end inline asm + mov.b32 {%rs121, %rs122}, %r62; + mov.b32 {%rs123, %rs124}, %r63; + mov.b32 {%rs125, %rs126}, %r64; + mov.b32 {%rs127, %rs128}, %r65; + .loc 1 79 47 // l2norm.py:79:47 + cvt.f32.bf16 %r173, %rs1; + cvt.f32.bf16 %r174, %rs2; + cvt.f32.bf16 %r175, %rs3; + cvt.f32.bf16 %r176, %rs4; + cvt.f32.bf16 %r177, %rs5; + cvt.f32.bf16 %r178, %rs6; + cvt.f32.bf16 %r179, %rs7; + cvt.f32.bf16 %r180, %rs8; + cvt.f32.bf16 %r181, %rs9; + cvt.f32.bf16 %r182, %rs10; + cvt.f32.bf16 %r183, %rs11; + cvt.f32.bf16 %r184, %rs12; + cvt.f32.bf16 %r185, %rs13; + cvt.f32.bf16 %r186, %rs14; + cvt.f32.bf16 %r187, %rs15; + cvt.f32.bf16 %r188, %rs16; + cvt.f32.bf16 %r189, %rs17; + cvt.f32.bf16 %r190, %rs18; + cvt.f32.bf16 %r191, %rs19; + cvt.f32.bf16 %r192, %rs20; + cvt.f32.bf16 %r193, %rs21; + cvt.f32.bf16 %r194, %rs22; + cvt.f32.bf16 %r195, %rs23; + cvt.f32.bf16 %r196, %rs24; + cvt.f32.bf16 %r197, %rs25; + cvt.f32.bf16 %r198, %rs26; + cvt.f32.bf16 %r199, %rs27; + cvt.f32.bf16 %r200, %rs28; + cvt.f32.bf16 %r201, %rs29; + cvt.f32.bf16 %r202, %rs30; + cvt.f32.bf16 %r203, %rs31; + cvt.f32.bf16 %r204, %rs32; + cvt.f32.bf16 %r205, %rs33; + cvt.f32.bf16 %r206, %rs34; + cvt.f32.bf16 %r207, %rs35; + cvt.f32.bf16 %r208, %rs36; + cvt.f32.bf16 %r209, %rs37; + cvt.f32.bf16 %r210, %rs38; + cvt.f32.bf16 %r211, %rs39; + cvt.f32.bf16 %r212, %rs40; + cvt.f32.bf16 %r213, %rs41; + cvt.f32.bf16 %r214, %rs42; + cvt.f32.bf16 %r215, %rs43; + cvt.f32.bf16 %r216, %rs44; + cvt.f32.bf16 %r217, %rs45; + cvt.f32.bf16 %r218, %rs46; + cvt.f32.bf16 %r219, %rs47; + cvt.f32.bf16 %r220, %rs48; + cvt.f32.bf16 %r221, %rs49; + cvt.f32.bf16 %r222, %rs50; + cvt.f32.bf16 %r223, %rs51; + cvt.f32.bf16 %r224, %rs52; + cvt.f32.bf16 %r225, %rs53; + cvt.f32.bf16 %r226, %rs54; + cvt.f32.bf16 %r227, %rs55; + cvt.f32.bf16 %r228, %rs56; + cvt.f32.bf16 %r229, %rs57; + cvt.f32.bf16 %r230, %rs58; + cvt.f32.bf16 %r231, %rs59; + cvt.f32.bf16 %r232, %rs60; + cvt.f32.bf16 %r233, %rs61; + cvt.f32.bf16 %r234, %rs62; + cvt.f32.bf16 %r235, %rs63; + cvt.f32.bf16 %r236, %rs64; + cvt.f32.bf16 %r237, %rs65; + cvt.f32.bf16 %r238, %rs66; + cvt.f32.bf16 %r239, %rs67; + cvt.f32.bf16 %r240, %rs68; + cvt.f32.bf16 %r241, %rs69; + cvt.f32.bf16 %r242, %rs70; + cvt.f32.bf16 %r243, %rs71; + cvt.f32.bf16 %r244, %rs72; + cvt.f32.bf16 %r245, %rs73; + cvt.f32.bf16 %r246, %rs74; + cvt.f32.bf16 %r247, %rs75; + cvt.f32.bf16 %r248, %rs76; + cvt.f32.bf16 %r249, %rs77; + cvt.f32.bf16 %r250, %rs78; + cvt.f32.bf16 %r251, %rs79; + cvt.f32.bf16 %r252, %rs80; + cvt.f32.bf16 %r253, %rs81; + cvt.f32.bf16 %r254, %rs82; + cvt.f32.bf16 %r255, %rs83; + cvt.f32.bf16 %r256, %rs84; + cvt.f32.bf16 %r257, %rs85; + cvt.f32.bf16 %r258, %rs86; + cvt.f32.bf16 %r259, %rs87; + cvt.f32.bf16 %r260, %rs88; + cvt.f32.bf16 %r261, %rs89; + cvt.f32.bf16 %r262, %rs90; + cvt.f32.bf16 %r263, %rs91; + cvt.f32.bf16 %r264, %rs92; + cvt.f32.bf16 %r265, %rs93; + cvt.f32.bf16 %r266, %rs94; + cvt.f32.bf16 %r267, %rs95; + cvt.f32.bf16 %r268, %rs96; + cvt.f32.bf16 %r269, %rs97; + cvt.f32.bf16 %r270, %rs98; + cvt.f32.bf16 %r271, %rs99; + cvt.f32.bf16 %r272, %rs100; + cvt.f32.bf16 %r273, %rs101; + cvt.f32.bf16 %r274, %rs102; + cvt.f32.bf16 %r275, %rs103; + cvt.f32.bf16 %r276, %rs104; + cvt.f32.bf16 %r277, %rs105; + cvt.f32.bf16 %r278, %rs106; + cvt.f32.bf16 %r279, %rs107; + cvt.f32.bf16 %r280, %rs108; + cvt.f32.bf16 %r281, %rs109; + cvt.f32.bf16 %r282, %rs110; + cvt.f32.bf16 %r283, %rs111; + cvt.f32.bf16 %r284, %rs112; + cvt.f32.bf16 %r285, %rs113; + cvt.f32.bf16 %r286, %rs114; + cvt.f32.bf16 %r287, %rs115; + cvt.f32.bf16 %r288, %rs116; + cvt.f32.bf16 %r289, %rs117; + cvt.f32.bf16 %r290, %rs118; + cvt.f32.bf16 %r291, %rs119; + cvt.f32.bf16 %r292, %rs120; + cvt.f32.bf16 %r293, %rs121; + cvt.f32.bf16 %r294, %rs122; + cvt.f32.bf16 %r295, %rs123; + cvt.f32.bf16 %r296, %rs124; + cvt.f32.bf16 %r297, %rs125; + cvt.f32.bf16 %r298, %rs126; + cvt.f32.bf16 %r299, %rs127; + cvt.f32.bf16 %r300, %rs128; + .loc 1 80 38 // l2norm.py:80:38 + mul.f32 %r301, %r174, %r174; + mul.f32 %r302, %r182, %r182; + mul.f32 %r303, %r190, %r190; + mul.f32 %r304, %r198, %r198; + mul.f32 %r305, %r206, %r206; + mul.f32 %r306, %r214, %r214; + mul.f32 %r307, %r222, %r222; + mul.f32 %r308, %r230, %r230; + mul.f32 %r309, %r238, %r238; + mul.f32 %r310, %r246, %r246; + mul.f32 %r311, %r254, %r254; + mul.f32 %r312, %r262, %r262; + mul.f32 %r313, %r270, %r270; + mul.f32 %r314, %r278, %r278; + mul.f32 %r315, %r286, %r286; + mul.f32 %r316, %r294, %r294; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + fma.rn.f32.bf16 %r317, %rs1, %rs1, %r301; + fma.rn.f32.bf16 %r318, %rs3, %rs3, %r317; + fma.rn.f32.bf16 %r319, %rs4, %rs4, %r318; + fma.rn.f32.bf16 %r320, %rs5, %rs5, %r319; + fma.rn.f32.bf16 %r321, %rs6, %rs6, %r320; + fma.rn.f32.bf16 %r322, %rs7, %rs7, %r321; + fma.rn.f32.bf16 %r323, %rs8, %rs8, %r322; + fma.rn.f32.bf16 %r324, %rs9, %rs9, %r302; + fma.rn.f32.bf16 %r325, %rs11, %rs11, %r324; + fma.rn.f32.bf16 %r326, %rs12, %rs12, %r325; + fma.rn.f32.bf16 %r327, %rs13, %rs13, %r326; + fma.rn.f32.bf16 %r328, %rs14, %rs14, %r327; + fma.rn.f32.bf16 %r329, %rs15, %rs15, %r328; + fma.rn.f32.bf16 %r330, %rs16, %rs16, %r329; + fma.rn.f32.bf16 %r331, %rs17, %rs17, %r303; + fma.rn.f32.bf16 %r332, %rs19, %rs19, %r331; + fma.rn.f32.bf16 %r333, %rs20, %rs20, %r332; + fma.rn.f32.bf16 %r334, %rs21, %rs21, %r333; + fma.rn.f32.bf16 %r335, %rs22, %rs22, %r334; + fma.rn.f32.bf16 %r336, %rs23, %rs23, %r335; + fma.rn.f32.bf16 %r337, %rs24, %rs24, %r336; + fma.rn.f32.bf16 %r338, %rs25, %rs25, %r304; + fma.rn.f32.bf16 %r339, %rs27, %rs27, %r338; + fma.rn.f32.bf16 %r340, %rs28, %rs28, %r339; + fma.rn.f32.bf16 %r341, %rs29, %rs29, %r340; + fma.rn.f32.bf16 %r342, %rs30, %rs30, %r341; + fma.rn.f32.bf16 %r343, %rs31, %rs31, %r342; + fma.rn.f32.bf16 %r344, %rs32, %rs32, %r343; + fma.rn.f32.bf16 %r345, %rs33, %rs33, %r305; + fma.rn.f32.bf16 %r346, %rs35, %rs35, %r345; + fma.rn.f32.bf16 %r347, %rs36, %rs36, %r346; + fma.rn.f32.bf16 %r348, %rs37, %rs37, %r347; + fma.rn.f32.bf16 %r349, %rs38, %rs38, %r348; + fma.rn.f32.bf16 %r350, %rs39, %rs39, %r349; + fma.rn.f32.bf16 %r351, %rs40, %rs40, %r350; + fma.rn.f32.bf16 %r352, %rs41, %rs41, %r306; + fma.rn.f32.bf16 %r353, %rs43, %rs43, %r352; + fma.rn.f32.bf16 %r354, %rs44, %rs44, %r353; + fma.rn.f32.bf16 %r355, %rs45, %rs45, %r354; + fma.rn.f32.bf16 %r356, %rs46, %rs46, %r355; + fma.rn.f32.bf16 %r357, %rs47, %rs47, %r356; + fma.rn.f32.bf16 %r358, %rs48, %rs48, %r357; + fma.rn.f32.bf16 %r359, %rs49, %rs49, %r307; + fma.rn.f32.bf16 %r360, %rs51, %rs51, %r359; + fma.rn.f32.bf16 %r361, %rs52, %rs52, %r360; + fma.rn.f32.bf16 %r362, %rs53, %rs53, %r361; + fma.rn.f32.bf16 %r363, %rs54, %rs54, %r362; + fma.rn.f32.bf16 %r364, %rs55, %rs55, %r363; + fma.rn.f32.bf16 %r365, %rs56, %rs56, %r364; + fma.rn.f32.bf16 %r366, %rs57, %rs57, %r308; + fma.rn.f32.bf16 %r367, %rs59, %rs59, %r366; + fma.rn.f32.bf16 %r368, %rs60, %rs60, %r367; + fma.rn.f32.bf16 %r369, %rs61, %rs61, %r368; + fma.rn.f32.bf16 %r370, %rs62, %rs62, %r369; + fma.rn.f32.bf16 %r371, %rs63, %rs63, %r370; + fma.rn.f32.bf16 %r372, %rs64, %rs64, %r371; + fma.rn.f32.bf16 %r373, %rs65, %rs65, %r309; + fma.rn.f32.bf16 %r374, %rs67, %rs67, %r373; + fma.rn.f32.bf16 %r375, %rs68, %rs68, %r374; + fma.rn.f32.bf16 %r376, %rs69, %rs69, %r375; + fma.rn.f32.bf16 %r377, %rs70, %rs70, %r376; + fma.rn.f32.bf16 %r378, %rs71, %rs71, %r377; + fma.rn.f32.bf16 %r379, %rs72, %rs72, %r378; + fma.rn.f32.bf16 %r380, %rs73, %rs73, %r310; + fma.rn.f32.bf16 %r381, %rs75, %rs75, %r380; + fma.rn.f32.bf16 %r382, %rs76, %rs76, %r381; + fma.rn.f32.bf16 %r383, %rs77, %rs77, %r382; + fma.rn.f32.bf16 %r384, %rs78, %rs78, %r383; + fma.rn.f32.bf16 %r385, %rs79, %rs79, %r384; + fma.rn.f32.bf16 %r386, %rs80, %rs80, %r385; + fma.rn.f32.bf16 %r387, %rs81, %rs81, %r311; + fma.rn.f32.bf16 %r388, %rs83, %rs83, %r387; + fma.rn.f32.bf16 %r389, %rs84, %rs84, %r388; + fma.rn.f32.bf16 %r390, %rs85, %rs85, %r389; + fma.rn.f32.bf16 %r391, %rs86, %rs86, %r390; + fma.rn.f32.bf16 %r392, %rs87, %rs87, %r391; + fma.rn.f32.bf16 %r393, %rs88, %rs88, %r392; + fma.rn.f32.bf16 %r394, %rs89, %rs89, %r312; + fma.rn.f32.bf16 %r395, %rs91, %rs91, %r394; + fma.rn.f32.bf16 %r396, %rs92, %rs92, %r395; + fma.rn.f32.bf16 %r397, %rs93, %rs93, %r396; + fma.rn.f32.bf16 %r398, %rs94, %rs94, %r397; + fma.rn.f32.bf16 %r399, %rs95, %rs95, %r398; + fma.rn.f32.bf16 %r400, %rs96, %rs96, %r399; + fma.rn.f32.bf16 %r401, %rs97, %rs97, %r313; + fma.rn.f32.bf16 %r402, %rs99, %rs99, %r401; + fma.rn.f32.bf16 %r403, %rs100, %rs100, %r402; + fma.rn.f32.bf16 %r404, %rs101, %rs101, %r403; + fma.rn.f32.bf16 %r405, %rs102, %rs102, %r404; + fma.rn.f32.bf16 %r406, %rs103, %rs103, %r405; + fma.rn.f32.bf16 %r407, %rs104, %rs104, %r406; + fma.rn.f32.bf16 %r408, %rs105, %rs105, %r314; + fma.rn.f32.bf16 %r409, %rs107, %rs107, %r408; + fma.rn.f32.bf16 %r410, %rs108, %rs108, %r409; + fma.rn.f32.bf16 %r411, %rs109, %rs109, %r410; + fma.rn.f32.bf16 %r412, %rs110, %rs110, %r411; + fma.rn.f32.bf16 %r413, %rs111, %rs111, %r412; + fma.rn.f32.bf16 %r414, %rs112, %rs112, %r413; + fma.rn.f32.bf16 %r415, %rs113, %rs113, %r315; + fma.rn.f32.bf16 %r416, %rs115, %rs115, %r415; + fma.rn.f32.bf16 %r417, %rs116, %rs116, %r416; + fma.rn.f32.bf16 %r418, %rs117, %rs117, %r417; + fma.rn.f32.bf16 %r419, %rs118, %rs118, %r418; + fma.rn.f32.bf16 %r420, %rs119, %rs119, %r419; + fma.rn.f32.bf16 %r421, %rs120, %rs120, %r420; + fma.rn.f32.bf16 %r422, %rs121, %rs121, %r316; + fma.rn.f32.bf16 %r423, %rs123, %rs123, %r422; + fma.rn.f32.bf16 %r424, %rs124, %rs124, %r423; + fma.rn.f32.bf16 %r425, %rs125, %rs125, %r424; + fma.rn.f32.bf16 %r426, %rs126, %rs126, %r425; + fma.rn.f32.bf16 %r427, %rs127, %rs127, %r426; + fma.rn.f32.bf16 %r428, %rs128, %rs128, %r427; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r429, %r323, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r430, %r323, %r429; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r431, %r430, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r432, %r430, %r431; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r433, %r432, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r434, %r432, %r433; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r435, %r434, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r436, %r434, %r435; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r437, %r330, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r438, %r330, %r437; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r439, %r438, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r440, %r438, %r439; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r441, %r440, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r442, %r440, %r441; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r443, %r442, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r444, %r442, %r443; +$L__tmp18: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r445, %r337, 8, 31, -1; +$L__tmp19: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r446, %r337, %r445; +$L__tmp20: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r447, %r446, 4, 31, -1; +$L__tmp21: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r448, %r446, %r447; +$L__tmp22: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r449, %r448, 2, 31, -1; +$L__tmp23: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r450, %r448, %r449; +$L__tmp24: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r451, %r450, 1, 31, -1; +$L__tmp25: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r452, %r450, %r451; +$L__tmp26: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r453, %r344, 8, 31, -1; +$L__tmp27: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r454, %r344, %r453; +$L__tmp28: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r455, %r454, 4, 31, -1; +$L__tmp29: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r456, %r454, %r455; +$L__tmp30: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r457, %r456, 2, 31, -1; +$L__tmp31: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r458, %r456, %r457; +$L__tmp32: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r459, %r458, 1, 31, -1; +$L__tmp33: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r460, %r458, %r459; +$L__tmp34: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r461, %r351, 8, 31, -1; +$L__tmp35: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r462, %r351, %r461; +$L__tmp36: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r463, %r462, 4, 31, -1; +$L__tmp37: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r464, %r462, %r463; +$L__tmp38: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r465, %r464, 2, 31, -1; +$L__tmp39: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r466, %r464, %r465; +$L__tmp40: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r467, %r466, 1, 31, -1; +$L__tmp41: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r468, %r466, %r467; +$L__tmp42: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r469, %r358, 8, 31, -1; +$L__tmp43: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r470, %r358, %r469; +$L__tmp44: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r471, %r470, 4, 31, -1; +$L__tmp45: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r472, %r470, %r471; +$L__tmp46: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r473, %r472, 2, 31, -1; +$L__tmp47: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r474, %r472, %r473; +$L__tmp48: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r475, %r474, 1, 31, -1; +$L__tmp49: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r476, %r474, %r475; +$L__tmp50: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r477, %r365, 8, 31, -1; +$L__tmp51: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r478, %r365, %r477; +$L__tmp52: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r479, %r478, 4, 31, -1; +$L__tmp53: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r480, %r478, %r479; +$L__tmp54: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r481, %r480, 2, 31, -1; +$L__tmp55: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r482, %r480, %r481; +$L__tmp56: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r483, %r482, 1, 31, -1; +$L__tmp57: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r484, %r482, %r483; +$L__tmp58: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r485, %r372, 8, 31, -1; +$L__tmp59: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r486, %r372, %r485; +$L__tmp60: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r487, %r486, 4, 31, -1; +$L__tmp61: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r488, %r486, %r487; +$L__tmp62: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r489, %r488, 2, 31, -1; +$L__tmp63: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r490, %r488, %r489; +$L__tmp64: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r491, %r490, 1, 31, -1; +$L__tmp65: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r492, %r490, %r491; +$L__tmp66: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r493, %r379, 8, 31, -1; +$L__tmp67: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r494, %r379, %r493; +$L__tmp68: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r495, %r494, 4, 31, -1; +$L__tmp69: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r496, %r494, %r495; +$L__tmp70: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r497, %r496, 2, 31, -1; +$L__tmp71: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r498, %r496, %r497; +$L__tmp72: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r499, %r498, 1, 31, -1; +$L__tmp73: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r500, %r498, %r499; +$L__tmp74: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r501, %r386, 8, 31, -1; +$L__tmp75: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r502, %r386, %r501; +$L__tmp76: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r503, %r502, 4, 31, -1; +$L__tmp77: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r504, %r502, %r503; +$L__tmp78: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r505, %r504, 2, 31, -1; +$L__tmp79: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r506, %r504, %r505; +$L__tmp80: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r507, %r506, 1, 31, -1; +$L__tmp81: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r508, %r506, %r507; +$L__tmp82: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r509, %r393, 8, 31, -1; +$L__tmp83: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r510, %r393, %r509; +$L__tmp84: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r511, %r510, 4, 31, -1; +$L__tmp85: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r512, %r510, %r511; +$L__tmp86: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r513, %r512, 2, 31, -1; +$L__tmp87: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r514, %r512, %r513; +$L__tmp88: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r515, %r514, 1, 31, -1; +$L__tmp89: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r516, %r514, %r515; +$L__tmp90: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r517, %r400, 8, 31, -1; +$L__tmp91: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r518, %r400, %r517; +$L__tmp92: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r519, %r518, 4, 31, -1; +$L__tmp93: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r520, %r518, %r519; +$L__tmp94: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r521, %r520, 2, 31, -1; +$L__tmp95: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r522, %r520, %r521; +$L__tmp96: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r523, %r522, 1, 31, -1; +$L__tmp97: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r524, %r522, %r523; +$L__tmp98: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r525, %r407, 8, 31, -1; +$L__tmp99: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r526, %r407, %r525; +$L__tmp100: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r527, %r526, 4, 31, -1; +$L__tmp101: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r528, %r526, %r527; +$L__tmp102: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r529, %r528, 2, 31, -1; +$L__tmp103: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r530, %r528, %r529; +$L__tmp104: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r531, %r530, 1, 31, -1; +$L__tmp105: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r532, %r530, %r531; +$L__tmp106: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r533, %r414, 8, 31, -1; +$L__tmp107: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r534, %r414, %r533; +$L__tmp108: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r535, %r534, 4, 31, -1; +$L__tmp109: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r536, %r534, %r535; +$L__tmp110: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r537, %r536, 2, 31, -1; +$L__tmp111: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r538, %r536, %r537; +$L__tmp112: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r539, %r538, 1, 31, -1; +$L__tmp113: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r540, %r538, %r539; +$L__tmp114: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r541, %r421, 8, 31, -1; +$L__tmp115: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r542, %r421, %r541; +$L__tmp116: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r543, %r542, 4, 31, -1; +$L__tmp117: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r544, %r542, %r543; +$L__tmp118: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r545, %r544, 2, 31, -1; +$L__tmp119: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r546, %r544, %r545; +$L__tmp120: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r547, %r546, 1, 31, -1; +$L__tmp121: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r548, %r546, %r547; +$L__tmp122: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r549, %r428, 8, 31, -1; +$L__tmp123: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r550, %r428, %r549; +$L__tmp124: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r551, %r550, 4, 31, -1; +$L__tmp125: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r552, %r550, %r551; +$L__tmp126: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r553, %r552, 2, 31, -1; +$L__tmp127: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r554, %r552, %r553; +$L__tmp128: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r555, %r554, 1, 31, -1; +$L__tmp129: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r556, %r554, %r555; +$L__tmp130: + .loc 1 80 48 // l2norm.py:80:48 + add.f32 %r557, %r152, %r436; + add.f32 %r558, %r152, %r444; + add.f32 %r559, %r152, %r452; + add.f32 %r560, %r152, %r460; + add.f32 %r561, %r152, %r468; + add.f32 %r562, %r152, %r476; + add.f32 %r563, %r152, %r484; + add.f32 %r564, %r152, %r492; + add.f32 %r565, %r152, %r500; + add.f32 %r566, %r152, %r508; + add.f32 %r567, %r152, %r516; + add.f32 %r568, %r152, %r524; + add.f32 %r569, %r152, %r532; + add.f32 %r570, %r152, %r540; + add.f32 %r571, %r152, %r548; + add.f32 %r572, %r152, %r556; + .loc 1 80 25 // l2norm.py:80:25 + sqrt.approx.ftz.f32 %r573, %r557; + sqrt.approx.ftz.f32 %r574, %r558; + sqrt.approx.ftz.f32 %r575, %r559; + sqrt.approx.ftz.f32 %r576, %r560; + sqrt.approx.ftz.f32 %r577, %r561; + sqrt.approx.ftz.f32 %r578, %r562; + sqrt.approx.ftz.f32 %r579, %r563; + sqrt.approx.ftz.f32 %r580, %r564; + sqrt.approx.ftz.f32 %r581, %r565; + sqrt.approx.ftz.f32 %r582, %r566; + sqrt.approx.ftz.f32 %r583, %r567; + sqrt.approx.ftz.f32 %r584, %r568; + sqrt.approx.ftz.f32 %r585, %r569; + sqrt.approx.ftz.f32 %r586, %r570; + sqrt.approx.ftz.f32 %r587, %r571; + sqrt.approx.ftz.f32 %r588, %r572; + .loc 1 71 34 // l2norm.py:71:34 + and.b32 %r589, %r153, 63; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd104, %r589; + or.b64 %rd105, %rd37, %rd104; + .loc 1 77 20 // l2norm.py:77:20 + shl.b64 %rd106, %rd105, 2; + add.s64 %rd33, %rd36, %rd106; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd107, %rd35, %rd101; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd32, %rd107, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd108, %rd35, %rd99; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd31, %rd108, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd109, %rd35, %rd97; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd30, %rd109, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd110, %rd35, %rd95; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd29, %rd110, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd111, %rd35, %rd93; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd28, %rd111, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd112, %rd35, %rd91; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd27, %rd112, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd113, %rd35, %rd89; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd26, %rd113, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd114, %rd35, %rd87; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd25, %rd114, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd115, %rd35, %rd85; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd24, %rd115, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd116, %rd35, %rd83; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd23, %rd116, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd117, %rd35, %rd81; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd22, %rd117, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd118, %rd35, %rd79; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd21, %rd118, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd119, %rd35, %rd77; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd20, %rd119, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd120, %rd35, %rd75; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd19, %rd120, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd121, %rd35, %rd73; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd18, %rd121, %rd103; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd122, %rd35, %rd71; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd17, %rd122, %rd103; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p17, %rd105, %rd38; + mov.b32 %r590, 0f3F800000; + .loc 1 80 17 // l2norm.py:80:17 + div.full.f32 %r131, %r590, %r573; + div.full.f32 %r132, %r590, %r574; + div.full.f32 %r133, %r590, %r575; + div.full.f32 %r134, %r590, %r576; + div.full.f32 %r136, %r590, %r577; + div.full.f32 %r137, %r590, %r578; + div.full.f32 %r138, %r590, %r579; + div.full.f32 %r139, %r590, %r580; + div.full.f32 %r141, %r590, %r581; + div.full.f32 %r142, %r590, %r582; + div.full.f32 %r143, %r590, %r583; + div.full.f32 %r144, %r590, %r584; + div.full.f32 %r146, %r590, %r585; + div.full.f32 %r147, %r590, %r586; + div.full.f32 %r148, %r590, %r587; + div.full.f32 %r149, %r590, %r588; + .loc 1 81 16 // l2norm.py:81:16 + mul.f32 %r591, %r131, %r173; + mul.f32 %r592, %r131, %r174; + mul.f32 %r593, %r131, %r175; + mul.f32 %r594, %r131, %r176; + mul.f32 %r595, %r131, %r177; + mul.f32 %r596, %r131, %r178; + mul.f32 %r597, %r131, %r179; + mul.f32 %r598, %r131, %r180; + mul.f32 %r599, %r132, %r181; + mul.f32 %r600, %r132, %r182; + mul.f32 %r601, %r132, %r183; + mul.f32 %r602, %r132, %r184; + mul.f32 %r603, %r132, %r185; + mul.f32 %r604, %r132, %r186; + mul.f32 %r605, %r132, %r187; + mul.f32 %r606, %r132, %r188; + mul.f32 %r607, %r133, %r189; + mul.f32 %r608, %r133, %r190; + mul.f32 %r609, %r133, %r191; + mul.f32 %r610, %r133, %r192; + mul.f32 %r611, %r133, %r193; + mul.f32 %r612, %r133, %r194; + mul.f32 %r613, %r133, %r195; + mul.f32 %r614, %r133, %r196; + mul.f32 %r615, %r134, %r197; + mul.f32 %r616, %r134, %r198; + mul.f32 %r617, %r134, %r199; + mul.f32 %r618, %r134, %r200; + mul.f32 %r619, %r134, %r201; + mul.f32 %r620, %r134, %r202; + mul.f32 %r621, %r134, %r203; + mul.f32 %r622, %r134, %r204; + mul.f32 %r623, %r136, %r205; + mul.f32 %r624, %r136, %r206; + mul.f32 %r625, %r136, %r207; + mul.f32 %r626, %r136, %r208; + mul.f32 %r627, %r136, %r209; + mul.f32 %r628, %r136, %r210; + mul.f32 %r629, %r136, %r211; + mul.f32 %r630, %r136, %r212; + mul.f32 %r631, %r137, %r213; + mul.f32 %r632, %r137, %r214; + mul.f32 %r633, %r137, %r215; + mul.f32 %r634, %r137, %r216; + mul.f32 %r635, %r137, %r217; + mul.f32 %r636, %r137, %r218; + mul.f32 %r637, %r137, %r219; + mul.f32 %r638, %r137, %r220; + mul.f32 %r639, %r138, %r221; + mul.f32 %r640, %r138, %r222; + mul.f32 %r641, %r138, %r223; + mul.f32 %r642, %r138, %r224; + mul.f32 %r643, %r138, %r225; + mul.f32 %r644, %r138, %r226; + mul.f32 %r645, %r138, %r227; + mul.f32 %r646, %r138, %r228; + mul.f32 %r647, %r139, %r229; + mul.f32 %r648, %r139, %r230; + mul.f32 %r649, %r139, %r231; + mul.f32 %r650, %r139, %r232; + mul.f32 %r651, %r139, %r233; + mul.f32 %r652, %r139, %r234; + mul.f32 %r653, %r139, %r235; + mul.f32 %r654, %r139, %r236; + mul.f32 %r655, %r141, %r237; + mul.f32 %r656, %r141, %r238; + mul.f32 %r657, %r141, %r239; + mul.f32 %r658, %r141, %r240; + mul.f32 %r659, %r141, %r241; + mul.f32 %r660, %r141, %r242; + mul.f32 %r661, %r141, %r243; + mul.f32 %r662, %r141, %r244; + mul.f32 %r663, %r142, %r245; + mul.f32 %r664, %r142, %r246; + mul.f32 %r665, %r142, %r247; + mul.f32 %r666, %r142, %r248; + mul.f32 %r667, %r142, %r249; + mul.f32 %r668, %r142, %r250; + mul.f32 %r669, %r142, %r251; + mul.f32 %r670, %r142, %r252; + mul.f32 %r671, %r143, %r253; + mul.f32 %r672, %r143, %r254; + mul.f32 %r673, %r143, %r255; + mul.f32 %r674, %r143, %r256; + mul.f32 %r675, %r143, %r257; + mul.f32 %r676, %r143, %r258; + mul.f32 %r677, %r143, %r259; + mul.f32 %r678, %r143, %r260; + mul.f32 %r679, %r144, %r261; + mul.f32 %r680, %r144, %r262; + mul.f32 %r681, %r144, %r263; + mul.f32 %r682, %r144, %r264; + mul.f32 %r683, %r144, %r265; + mul.f32 %r684, %r144, %r266; + mul.f32 %r685, %r144, %r267; + mul.f32 %r686, %r144, %r268; + mul.f32 %r687, %r146, %r269; + mul.f32 %r688, %r146, %r270; + mul.f32 %r689, %r146, %r271; + mul.f32 %r690, %r146, %r272; + mul.f32 %r691, %r146, %r273; + mul.f32 %r692, %r146, %r274; + mul.f32 %r693, %r146, %r275; + mul.f32 %r694, %r146, %r276; + mul.f32 %r695, %r147, %r277; + mul.f32 %r696, %r147, %r278; + mul.f32 %r697, %r147, %r279; + mul.f32 %r698, %r147, %r280; + mul.f32 %r699, %r147, %r281; + mul.f32 %r700, %r147, %r282; + mul.f32 %r701, %r147, %r283; + mul.f32 %r702, %r147, %r284; + mul.f32 %r703, %r148, %r285; + mul.f32 %r704, %r148, %r286; + mul.f32 %r705, %r148, %r287; + mul.f32 %r706, %r148, %r288; + mul.f32 %r707, %r148, %r289; + mul.f32 %r708, %r148, %r290; + mul.f32 %r709, %r148, %r291; + mul.f32 %r710, %r148, %r292; + mul.f32 %r711, %r149, %r293; + mul.f32 %r712, %r149, %r294; + mul.f32 %r713, %r149, %r295; + mul.f32 %r714, %r149, %r296; + mul.f32 %r715, %r149, %r297; + mul.f32 %r716, %r149, %r298; + mul.f32 %r717, %r149, %r299; + mul.f32 %r718, %r149, %r300; + .loc 1 83 18 // l2norm.py:83:18 + cvt.rn.bf16x2.f32 %r66, %r592, %r591; + cvt.rn.bf16x2.f32 %r67, %r594, %r593; + cvt.rn.bf16x2.f32 %r68, %r596, %r595; + cvt.rn.bf16x2.f32 %r69, %r598, %r597; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd17 + 0 ], { %r66, %r67, %r68, %r69 }; + // end inline asm + cvt.rn.bf16x2.f32 %r70, %r600, %r599; + cvt.rn.bf16x2.f32 %r71, %r602, %r601; + cvt.rn.bf16x2.f32 %r72, %r604, %r603; + cvt.rn.bf16x2.f32 %r73, %r606, %r605; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd18 + 0 ], { %r70, %r71, %r72, %r73 }; + // end inline asm + cvt.rn.bf16x2.f32 %r74, %r608, %r607; + cvt.rn.bf16x2.f32 %r75, %r610, %r609; + cvt.rn.bf16x2.f32 %r76, %r612, %r611; + cvt.rn.bf16x2.f32 %r77, %r614, %r613; + // begin inline asm + @%p3 st.global.v4.b32 [ %rd19 + 0 ], { %r74, %r75, %r76, %r77 }; + // end inline asm + cvt.rn.bf16x2.f32 %r78, %r616, %r615; + cvt.rn.bf16x2.f32 %r79, %r618, %r617; + cvt.rn.bf16x2.f32 %r80, %r620, %r619; + cvt.rn.bf16x2.f32 %r81, %r622, %r621; + // begin inline asm + @%p4 st.global.v4.b32 [ %rd20 + 0 ], { %r78, %r79, %r80, %r81 }; + // end inline asm + cvt.rn.bf16x2.f32 %r82, %r624, %r623; + cvt.rn.bf16x2.f32 %r83, %r626, %r625; + cvt.rn.bf16x2.f32 %r84, %r628, %r627; + cvt.rn.bf16x2.f32 %r85, %r630, %r629; + // begin inline asm + @%p5 st.global.v4.b32 [ %rd21 + 0 ], { %r82, %r83, %r84, %r85 }; + // end inline asm + cvt.rn.bf16x2.f32 %r86, %r632, %r631; + cvt.rn.bf16x2.f32 %r87, %r634, %r633; + cvt.rn.bf16x2.f32 %r88, %r636, %r635; + cvt.rn.bf16x2.f32 %r89, %r638, %r637; + // begin inline asm + @%p6 st.global.v4.b32 [ %rd22 + 0 ], { %r86, %r87, %r88, %r89 }; + // end inline asm + cvt.rn.bf16x2.f32 %r90, %r640, %r639; + cvt.rn.bf16x2.f32 %r91, %r642, %r641; + cvt.rn.bf16x2.f32 %r92, %r644, %r643; + cvt.rn.bf16x2.f32 %r93, %r646, %r645; + // begin inline asm + @%p7 st.global.v4.b32 [ %rd23 + 0 ], { %r90, %r91, %r92, %r93 }; + // end inline asm + cvt.rn.bf16x2.f32 %r94, %r648, %r647; + cvt.rn.bf16x2.f32 %r95, %r650, %r649; + cvt.rn.bf16x2.f32 %r96, %r652, %r651; + cvt.rn.bf16x2.f32 %r97, %r654, %r653; + // begin inline asm + @%p15 st.global.v4.b32 [ %rd24 + 0 ], { %r94, %r95, %r96, %r97 }; + // end inline asm + cvt.rn.bf16x2.f32 %r98, %r656, %r655; + cvt.rn.bf16x2.f32 %r99, %r658, %r657; + cvt.rn.bf16x2.f32 %r100, %r660, %r659; + cvt.rn.bf16x2.f32 %r101, %r662, %r661; + // begin inline asm + @%p8 st.global.v4.b32 [ %rd25 + 0 ], { %r98, %r99, %r100, %r101 }; + // end inline asm + cvt.rn.bf16x2.f32 %r102, %r664, %r663; + cvt.rn.bf16x2.f32 %r103, %r666, %r665; + cvt.rn.bf16x2.f32 %r104, %r668, %r667; + cvt.rn.bf16x2.f32 %r105, %r670, %r669; + // begin inline asm + @%p9 st.global.v4.b32 [ %rd26 + 0 ], { %r102, %r103, %r104, %r105 }; + // end inline asm + cvt.rn.bf16x2.f32 %r106, %r672, %r671; + cvt.rn.bf16x2.f32 %r107, %r674, %r673; + cvt.rn.bf16x2.f32 %r108, %r676, %r675; + cvt.rn.bf16x2.f32 %r109, %r678, %r677; + // begin inline asm + @%p10 st.global.v4.b32 [ %rd27 + 0 ], { %r106, %r107, %r108, %r109 }; + // end inline asm + cvt.rn.bf16x2.f32 %r110, %r680, %r679; + cvt.rn.bf16x2.f32 %r111, %r682, %r681; + cvt.rn.bf16x2.f32 %r112, %r684, %r683; + cvt.rn.bf16x2.f32 %r113, %r686, %r685; + // begin inline asm + @%p11 st.global.v4.b32 [ %rd28 + 0 ], { %r110, %r111, %r112, %r113 }; + // end inline asm + cvt.rn.bf16x2.f32 %r114, %r688, %r687; + cvt.rn.bf16x2.f32 %r115, %r690, %r689; + cvt.rn.bf16x2.f32 %r116, %r692, %r691; + cvt.rn.bf16x2.f32 %r117, %r694, %r693; + // begin inline asm + @%p12 st.global.v4.b32 [ %rd29 + 0 ], { %r114, %r115, %r116, %r117 }; + // end inline asm + cvt.rn.bf16x2.f32 %r118, %r696, %r695; + cvt.rn.bf16x2.f32 %r119, %r698, %r697; + cvt.rn.bf16x2.f32 %r120, %r700, %r699; + cvt.rn.bf16x2.f32 %r121, %r702, %r701; + // begin inline asm + @%p13 st.global.v4.b32 [ %rd30 + 0 ], { %r118, %r119, %r120, %r121 }; + // end inline asm + cvt.rn.bf16x2.f32 %r122, %r704, %r703; + cvt.rn.bf16x2.f32 %r123, %r706, %r705; + cvt.rn.bf16x2.f32 %r124, %r708, %r707; + cvt.rn.bf16x2.f32 %r125, %r710, %r709; + // begin inline asm + @%p14 st.global.v4.b32 [ %rd31 + 0 ], { %r122, %r123, %r124, %r125 }; + // end inline asm + cvt.rn.bf16x2.f32 %r126, %r712, %r711; + cvt.rn.bf16x2.f32 %r127, %r714, %r713; + cvt.rn.bf16x2.f32 %r128, %r716, %r715; + cvt.rn.bf16x2.f32 %r129, %r718, %r717; + // begin inline asm + @%p16 st.global.v4.b32 [ %rd32 + 0 ], { %r126, %r127, %r128, %r129 }; + // end inline asm + .loc 1 84 21 // l2norm.py:84:21 + shl.b32 %r719, %r154, 1; + mov.b32 %r720, global_smem; + add.s32 %r130, %r720, %r719; + // begin inline asm + st.shared.v4.b32 [ %r130 + 0 ], { %r131, %r132, %r133, %r134 }; + // end inline asm + add.s32 %r135, %r130, 16; + // begin inline asm + st.shared.v4.b32 [ %r135 + 0 ], { %r136, %r137, %r138, %r139 }; + // end inline asm + add.s32 %r140, %r130, 128; + // begin inline asm + st.shared.v4.b32 [ %r140 + 0 ], { %r141, %r142, %r143, %r144 }; + // end inline asm + add.s32 %r145, %r130, 144; + // begin inline asm + st.shared.v4.b32 [ %r145 + 0 ], { %r146, %r147, %r148, %r149 }; + // end inline asm + bar.sync 0; + shl.b32 %r721, %r153, 5; + and.b32 %r722, %r721, 96; + and.b32 %r723, %r153, 28; + shl.b32 %r724, %r153, 2; + and.b32 %r725, %r724, 128; + add.s32 %r726, %r720, %r722; + add.s32 %r727, %r726, %r725; + add.s32 %r728, %r727, %r723; + ld.shared.b32 %r150, [%r728]; + // begin inline asm + @%p17 st.global.b32 [ %rd33 + 0 ], { %r150 }; + // end inline asm + .loc 1 84 4 // l2norm.py:84:4 + ret; +$L__tmp131: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp130 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 80 // DW_AT_call_line +.b8 43 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp130 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/l2norm_fwd_kernel.source b/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/l2norm_fwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..b5000c470f5e0683aea9e8045b2d0a8c059ea3d1 --- /dev/null +++ b/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/l2norm_fwd_kernel.source @@ -0,0 +1,172 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc38 = loc(unknown) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc45 = loc("x"(#loc)) +#loc46 = loc("y"(#loc)) +#loc47 = loc("rstd"(#loc)) +#loc48 = loc("eps"(#loc)) +#loc49 = loc("T"(#loc)) +#loc81 = loc("input"(#loc36)) +#loc82 = loc("a"(#loc41)) +#loc83 = loc("b"(#loc41)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc50) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc51) + %o_t = arith.constant 64 : i32 loc(#loc52) + %o_t_1 = arith.constant 64 : i64 loc(#loc52) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc52) + %o_t_3 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc53) + %o_t_4 = arith.extsi %o_t_3 : tensor<64xi32> to tensor<64xi64> loc(#loc54) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<64xi64> loc(#loc54) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<64xi64> loc(#loc54) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc55) + %m_t = arith.extsi %T : i32 to i64 loc(#loc56) + %m_t_7 = tt.splat %m_t : i64 -> tensor<64xi64> loc(#loc56) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<64xi64> loc(#loc56) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc57) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc58) + %m_x_10 = arith.constant 128 : i32 loc(#loc59) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc59) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc59) + %m_x_13 = tt.broadcast %m_x : tensor<64x1xi1> -> tensor<64x128xi1> loc(#loc60) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<64x128xi1> loc(#loc60) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<64x128xi1> loc(#loc60) + %p_x = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc61) + %p_x_16 = arith.constant 128 : i32 loc(#loc62) + %p_x_17 = arith.constant 128 : i64 loc(#loc62) + %p_x_18 = arith.constant dense<128> : tensor<64x1xi64> loc(#loc62) + %p_x_19 = arith.muli %p_x, %p_x_18 : tensor<64x1xi64> loc(#loc62) + %p_x_20 = tt.splat %x : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc63) + %p_x_21 = tt.addptr %p_x_20, %p_x_19 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc63) + %p_x_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc64) + %p_x_23 = tt.broadcast %p_x_21 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc65) + %p_x_24 = tt.broadcast %p_x_22 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc65) + %p_x_25 = tt.addptr %p_x_23, %p_x_24 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc65) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc66) + %p_y_26 = arith.constant 128 : i32 loc(#loc67) + %p_y_27 = arith.constant 128 : i64 loc(#loc67) + %p_y_28 = arith.constant dense<128> : tensor<64x1xi64> loc(#loc67) + %p_y_29 = arith.muli %p_y, %p_y_28 : tensor<64x1xi64> loc(#loc67) + %p_y_30 = tt.splat %y : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc68) + %p_y_31 = tt.addptr %p_y_30, %p_y_29 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc68) + %p_y_32 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc69) + %p_y_33 = tt.broadcast %p_y_31 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc70) + %p_y_34 = tt.broadcast %p_y_32 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc70) + %p_y_35 = tt.addptr %p_y_33, %p_y_34 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc70) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<64x!tt.ptr> loc(#loc71) + %p_rstd_36 = tt.addptr %p_rstd, %o_t_6 : tensor<64x!tt.ptr>, tensor<64xi64> loc(#loc71) + %b_x = arith.constant 0.000000e+00 : f32 loc(#loc72) + %b_x_37 = arith.constant dense<0.000000e+00> : tensor<64x128xf32> loc(#loc72) + %b_x_38 = arith.truncf %b_x_37 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc72) + %b_x_39 = tt.load %p_x_25, %m_x_15, %b_x_38 : tensor<64x128x!tt.ptr> loc(#loc72) + %b_x_40 = arith.extf %b_x_39 : tensor<64x128xbf16> to tensor<64x128xf32> loc(#loc73) + %b_rstd = arith.mulf %b_x_40, %b_x_40 : tensor<64x128xf32> loc(#loc74) + %b_rstd_41 = tt.call @triton.language.standard.sum__fp32S64_128S_c1_cFalse_cNone(%b_rstd) : (tensor<64x128xf32>) -> tensor<64xf32> loc(#loc75) + %b_rstd_42 = tt.splat %eps : f32 -> tensor<64xf32> loc(#loc76) + %b_rstd_43 = arith.addf %b_rstd_41, %b_rstd_42 : tensor<64xf32> loc(#loc76) + %b_rstd_44 = math.sqrt %b_rstd_43 : tensor<64xf32> loc(#loc77) + %b_rstd_45 = arith.constant 1 : i32 loc(#loc78) + %b_rstd_46 = arith.constant 1.000000e+00 : f32 loc(#loc78) + %b_rstd_47 = arith.constant dense<1.000000e+00> : tensor<64xf32> loc(#loc78) + %b_rstd_48 = arith.divf %b_rstd_47, %b_rstd_44 : tensor<64xf32> loc(#loc78) + %b_y = tt.expand_dims %b_rstd_48 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc79) + %b_y_49 = tt.broadcast %b_y : tensor<64x1xf32> -> tensor<64x128xf32> loc(#loc80) + %b_y_50 = arith.mulf %b_x_40, %b_y_49 : tensor<64x128xf32> loc(#loc80) + %0 = arith.truncf %b_y_50 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc32) + tt.store %p_y_35, %0, %m_x_15 : tensor<64x128x!tt.ptr> loc(#loc33) + tt.store %p_rstd_36, %b_rstd_48, %m_t_8 : tensor<64x!tt.ptr> loc(#loc34) + tt.return loc(#loc35) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S64_128S_c1_cFalse_cNone(%input: tensor<64x128xf32> loc("input"(#loc36))) -> tensor<64xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc37) + tt.reduce.return %2 : f32 loc(#loc37) + }) : (tensor<64x128xf32>) -> tensor<64xf32> loc(#loc37) + tt.return %0 : tensor<64xf32> loc(#loc39) + ^bb1: // no predecessors + %1 = ub.poison : tensor<64xf32> loc(#loc40) + tt.return %1 : tensor<64xf32> loc(#loc40) + } loc(#loc36) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc41)), %b: f32 loc("b"(#loc41))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc42) + tt.return %0 : f32 loc(#loc43) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc44) + tt.return %1 : f32 loc(#loc44) + } loc(#loc41) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:18) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:29) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:37) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc50 = loc("i_t"(#loc1)) +#loc51 = loc("i_t"(#loc2)) +#loc52 = loc("o_t"(#loc3)) +#loc53 = loc("o_t"(#loc4)) +#loc54 = loc("o_t"(#loc5)) +#loc55 = loc("o_d"(#loc6)) +#loc56 = loc("m_t"(#loc7)) +#loc57 = loc("m_x"(#loc8)) +#loc58 = loc("m_x"(#loc9)) +#loc59 = loc("m_x"(#loc10)) +#loc60 = loc("m_x"(#loc11)) +#loc61 = loc("p_x"(#loc12)) +#loc62 = loc("p_x"(#loc13)) +#loc63 = loc("p_x"(#loc14)) +#loc64 = loc("p_x"(#loc15)) +#loc65 = loc("p_x"(#loc16)) +#loc66 = loc("p_y"(#loc17)) +#loc67 = loc("p_y"(#loc18)) +#loc68 = loc("p_y"(#loc19)) +#loc69 = loc("p_y"(#loc20)) +#loc70 = loc("p_y"(#loc21)) +#loc71 = loc("p_rstd"(#loc22)) +#loc72 = loc("b_x"(#loc23)) +#loc73 = loc("b_x"(#loc24)) +#loc74 = loc("b_rstd"(#loc25)) +#loc75 = loc("b_rstd"(#loc26)) +#loc76 = loc("b_rstd"(#loc27)) +#loc77 = loc("b_rstd"(#loc28)) +#loc78 = loc("b_rstd"(#loc29)) +#loc79 = loc("b_y"(#loc30)) +#loc80 = loc("b_y"(#loc31)) diff --git a/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/l2norm_fwd_kernel.ttgir b/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/l2norm_fwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..50a41b1ce790ba9584c49729ad676f2abc60b7ce --- /dev/null +++ b/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/l2norm_fwd_kernel.ttgir @@ -0,0 +1,135 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [2, 1], order = [1, 0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [2], order = [0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc1 = loc(unknown) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc34 = loc("x"(#loc)) +#loc35 = loc("y"(#loc)) +#loc36 = loc("rstd"(#loc)) +#loc37 = loc("eps"(#loc)) +#loc38 = loc("T"(#loc)) +#loc59 = loc("b_rstd"(#loc23)) +#loc66 = loc(callsite(#loc1 at #loc59)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 2 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<1.000000e+00> : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<64x128xbf16, #blocked> loc(#loc1) + %c64_i64 = arith.constant 64 : i64 loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<64x1xi64, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc39) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc40) + %o_t = arith.muli %i_t_3, %c64_i64 : i64 loc(#loc41) + %o_t_4 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc42) + %o_t_5 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #blocked1> loc(#loc42) + %o_t_6 = arith.extsi %o_t_4 : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_7 = arith.extsi %o_t_5 : tensor<64xi32, #blocked1> to tensor<64xi64, #blocked1> loc(#loc43) + %o_t_8 = tt.splat %o_t : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_9 = tt.splat %o_t : i64 -> tensor<64xi64, #blocked1> loc(#loc43) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<64xi64, #blocked1> loc(#loc43) + %m_t = arith.extsi %T : i32 to i64 loc(#loc44) + %m_t_12 = tt.splat %m_t : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_13 = tt.splat %m_t : i64 -> tensor<64xi64, #blocked1> loc(#loc44) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<64xi64, #blocked1> loc(#loc44) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<64xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64x1xi1, #blocked> loc(#loc45) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc46) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc46) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_1 : tensor<1x128xi32, #blocked> loc(#loc47) + %m_x_19 = tt.broadcast %m_x : tensor<64x1xi1, #blocked> -> tensor<64x128xi1, #blocked> loc(#loc48) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<64x128xi1, #blocked> loc(#loc48) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<64x128xi1, #blocked> loc(#loc48) + %p_x = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64x1xi64, #blocked> loc(#loc49) + %p_x_22 = arith.muli %p_x, %cst_2 : tensor<64x1xi64, #blocked> loc(#loc50) + %p_x_23 = tt.splat %x : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked> loc(#loc51) + %p_x_24 = tt.addptr %p_x_23, %p_x_22 : tensor<64x1x!tt.ptr, #blocked>, tensor<64x1xi64, #blocked> loc(#loc51) + %p_x_25 = tt.broadcast %p_x_24 : tensor<64x1x!tt.ptr, #blocked> -> tensor<64x128x!tt.ptr, #blocked> loc(#loc52) + %p_x_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<64x128xi32, #blocked> loc(#loc52) + %p_x_27 = tt.addptr %p_x_25, %p_x_26 : tensor<64x128x!tt.ptr, #blocked>, tensor<64x128xi32, #blocked> loc(#loc52) + %p_y = tt.splat %y : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked> loc(#loc53) + %p_y_28 = tt.addptr %p_y, %p_x_22 : tensor<64x1x!tt.ptr, #blocked>, tensor<64x1xi64, #blocked> loc(#loc53) + %p_y_29 = tt.broadcast %p_y_28 : tensor<64x1x!tt.ptr, #blocked> -> tensor<64x128x!tt.ptr, #blocked> loc(#loc54) + %p_y_30 = tt.addptr %p_y_29, %p_x_26 : tensor<64x128x!tt.ptr, #blocked>, tensor<64x128xi32, #blocked> loc(#loc54) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<64x!tt.ptr, #blocked1> loc(#loc55) + %p_rstd_31 = tt.addptr %p_rstd, %o_t_11 : tensor<64x!tt.ptr, #blocked1>, tensor<64xi64, #blocked1> loc(#loc55) + %b_x = tt.load %p_x_27, %m_x_21, %cst_0 : tensor<64x128x!tt.ptr, #blocked> loc(#loc56) + %b_x_32 = arith.extf %b_x : tensor<64x128xbf16, #blocked> to tensor<64x128xf32, #blocked> loc(#loc57) + %b_rstd = arith.mulf %b_x_32, %b_x_32 : tensor<64x128xf32, #blocked> loc(#loc58) + %b_rstd_33 = "tt.reduce"(%b_rstd) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_40: f32 loc(callsite(#loc1 at #loc59)), %b_rstd_41: f32 loc(callsite(#loc1 at #loc59))): + %b_rstd_42 = arith.addf %b_rstd_40, %b_rstd_41 : f32 loc(#loc67) + tt.reduce.return %b_rstd_42 : f32 loc(#loc65) + }) : (tensor<64x128xf32, #blocked>) -> tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc65) + %b_rstd_34 = tt.splat %eps : f32 -> tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_36 = math.sqrt %b_rstd_35 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc61) + %b_rstd_37 = arith.divf %cst, %b_rstd_36 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc62) + %b_y = tt.expand_dims %b_rstd_37 {axis = 1 : i32} : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64x1xf32, #blocked> loc(#loc63) + %b_y_38 = tt.broadcast %b_y : tensor<64x1xf32, #blocked> -> tensor<64x128xf32, #blocked> loc(#loc64) + %b_y_39 = arith.mulf %b_x_32, %b_y_38 : tensor<64x128xf32, #blocked> loc(#loc64) + %0 = arith.truncf %b_y_39 : tensor<64x128xf32, #blocked> to tensor<64x128xbf16, #blocked> loc(#loc30) + tt.store %p_y_30, %0, %m_x_21 : tensor<64x128x!tt.ptr, #blocked> loc(#loc31) + %1 = ttg.convert_layout %b_rstd_37 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64xf32, #blocked1> loc(#loc32) + tt.store %p_rstd_31, %1, %m_t_15 : tensor<64x!tt.ptr, #blocked1> loc(#loc32) + tt.return loc(#loc33) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc39 = loc("i_t"(#loc2)) +#loc40 = loc("i_t"(#loc3)) +#loc41 = loc("o_t"(#loc4)) +#loc42 = loc("o_t"(#loc5)) +#loc43 = loc("o_t"(#loc6)) +#loc44 = loc("m_t"(#loc7)) +#loc45 = loc("m_x"(#loc8)) +#loc46 = loc("m_x"(#loc9)) +#loc47 = loc("m_x"(#loc10)) +#loc48 = loc("m_x"(#loc11)) +#loc49 = loc("p_x"(#loc12)) +#loc50 = loc("p_x"(#loc13)) +#loc51 = loc("p_x"(#loc14)) +#loc52 = loc("p_x"(#loc15)) +#loc53 = loc("p_y"(#loc16)) +#loc54 = loc("p_y"(#loc17)) +#loc55 = loc("p_rstd"(#loc18)) +#loc56 = loc("b_x"(#loc19)) +#loc57 = loc("b_x"(#loc20)) +#loc58 = loc("b_rstd"(#loc21)) +#loc60 = loc("b_rstd"(#loc25)) +#loc61 = loc("b_rstd"(#loc26)) +#loc62 = loc("b_rstd"(#loc27)) +#loc63 = loc("b_y"(#loc28)) +#loc64 = loc("b_y"(#loc29)) +#loc65 = loc(callsite(#loc22 at #loc59)) +#loc67 = loc(callsite(#loc24 at #loc65)) diff --git a/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/l2norm_fwd_kernel.ttir b/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/l2norm_fwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..85ebb8b128ceed91f07ab82f5c0fe10a876f6335 --- /dev/null +++ b/triton/gpu0/BHEMYQZSXYU6A3D7FIILTMQW5TEMH7HFE2A7EZUGWT7QSQEU6R4A/l2norm_fwd_kernel.ttir @@ -0,0 +1,128 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc3 = loc(unknown) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc35 = loc("x"(#loc)) +#loc36 = loc("y"(#loc)) +#loc37 = loc("rstd"(#loc)) +#loc38 = loc("eps"(#loc)) +#loc39 = loc("T"(#loc)) +#loc62 = loc("b_rstd"(#loc25)) +#loc68 = loc(callsite(#loc3 at #loc62)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %b_x = arith.constant dense<0.000000e+00> : tensor<64x128xbf16> loc(#loc40) + %b_rstd = arith.constant dense<1.000000e+00> : tensor<64xf32> loc(#loc41) + %cst = arith.constant dense<128> : tensor<64x1xi64> loc(#loc3) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc42) + %o_t = arith.constant 64 : i64 loc(#loc43) + %i_t = tt.get_program_id x : i32 loc(#loc44) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc45) + %o_t_1 = arith.muli %i_t_0, %o_t : i64 loc(#loc43) + %o_t_2 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc46) + %o_t_3 = arith.extsi %o_t_2 : tensor<64xi32> to tensor<64xi64> loc(#loc47) + %o_t_4 = tt.splat %o_t_1 : i64 -> tensor<64xi64> loc(#loc47) + %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<64xi64> loc(#loc47) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc48) + %m_t = arith.extsi %T : i32 to i64 loc(#loc49) + %m_t_6 = tt.splat %m_t : i64 -> tensor<64xi64> loc(#loc49) + %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<64xi64> loc(#loc49) + %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc50) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc51) + %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc42) + %m_x_11 = tt.broadcast %m_x_8 : tensor<64x1xi1> -> tensor<64x128xi1> loc(#loc52) + %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<64x128xi1> loc(#loc52) + %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<64x128xi1> loc(#loc52) + %p_x = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc53) + %p_x_14 = arith.muli %p_x, %cst : tensor<64x1xi64> loc(#loc54) + %p_x_15 = tt.splat %x : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc55) + %p_x_16 = tt.addptr %p_x_15, %p_x_14 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc55) + %p_x_17 = tt.broadcast %p_x_16 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc56) + %p_x_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc56) + %p_x_19 = tt.addptr %p_x_17, %p_x_18 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc56) + %p_y = tt.splat %y : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc57) + %p_y_20 = tt.addptr %p_y, %p_x_14 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc57) + %p_y_21 = tt.broadcast %p_y_20 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc58) + %p_y_22 = tt.addptr %p_y_21, %p_x_18 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<64x!tt.ptr> loc(#loc59) + %p_rstd_23 = tt.addptr %p_rstd, %o_t_5 : tensor<64x!tt.ptr>, tensor<64xi64> loc(#loc59) + %b_x_24 = tt.load %p_x_19, %m_x_13, %b_x : tensor<64x128x!tt.ptr> loc(#loc40) + %b_x_25 = arith.extf %b_x_24 : tensor<64x128xbf16> to tensor<64x128xf32> loc(#loc60) + %b_rstd_26 = arith.mulf %b_x_25, %b_x_25 : tensor<64x128xf32> loc(#loc61) + %b_rstd_27 = "tt.reduce"(%b_rstd_26) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_34: f32 loc(callsite(#loc3 at #loc62)), %b_rstd_35: f32 loc(callsite(#loc3 at #loc62))): + %b_rstd_36 = arith.addf %b_rstd_34, %b_rstd_35 : f32 loc(#loc69) + tt.reduce.return %b_rstd_36 : f32 loc(#loc67) + }) : (tensor<64x128xf32>) -> tensor<64xf32> loc(#loc67) + %b_rstd_28 = tt.splat %eps : f32 -> tensor<64xf32> loc(#loc63) + %b_rstd_29 = arith.addf %b_rstd_27, %b_rstd_28 : tensor<64xf32> loc(#loc63) + %b_rstd_30 = math.sqrt %b_rstd_29 : tensor<64xf32> loc(#loc64) + %b_rstd_31 = arith.divf %b_rstd, %b_rstd_30 : tensor<64xf32> loc(#loc41) + %b_y = tt.expand_dims %b_rstd_31 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc65) + %b_y_32 = tt.broadcast %b_y : tensor<64x1xf32> -> tensor<64x128xf32> loc(#loc66) + %b_y_33 = arith.mulf %b_x_25, %b_y_32 : tensor<64x128xf32> loc(#loc66) + %0 = arith.truncf %b_y_33 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc31) + tt.store %p_y_22, %0, %m_x_13 : tensor<64x128x!tt.ptr> loc(#loc32) + tt.store %p_rstd_23, %b_rstd_31, %m_t_7 : tensor<64x!tt.ptr> loc(#loc33) + tt.return loc(#loc34) + } loc(#loc) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc40 = loc("b_x"(#loc1)) +#loc41 = loc("b_rstd"(#loc2)) +#loc42 = loc("m_x"(#loc4)) +#loc43 = loc("o_t"(#loc5)) +#loc44 = loc("i_t"(#loc6)) +#loc45 = loc("i_t"(#loc7)) +#loc46 = loc("o_t"(#loc8)) +#loc47 = loc("o_t"(#loc9)) +#loc48 = loc("o_d"(#loc10)) +#loc49 = loc("m_t"(#loc11)) +#loc50 = loc("m_x"(#loc12)) +#loc51 = loc("m_x"(#loc13)) +#loc52 = loc("m_x"(#loc14)) +#loc53 = loc("p_x"(#loc15)) +#loc54 = loc("p_x"(#loc16)) +#loc55 = loc("p_x"(#loc17)) +#loc56 = loc("p_x"(#loc18)) +#loc57 = loc("p_y"(#loc19)) +#loc58 = loc("p_y"(#loc20)) +#loc59 = loc("p_rstd"(#loc21)) +#loc60 = loc("b_x"(#loc22)) +#loc61 = loc("b_rstd"(#loc23)) +#loc63 = loc("b_rstd"(#loc27)) +#loc64 = loc("b_rstd"(#loc28)) +#loc65 = loc("b_y"(#loc29)) +#loc66 = loc("b_y"(#loc30)) +#loc67 = loc(callsite(#loc24 at #loc62)) +#loc69 = loc(callsite(#loc26 at #loc67)) diff --git a/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/__grp__l2norm_fwd_kernel.json b/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/__grp__l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..1eec865266a0824d909b98d210711e3c328767b4 --- /dev/null +++ b/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/__grp__l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_fwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/l2norm_fwd_kernel.source", "l2norm_fwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/l2norm_fwd_kernel.ttir", "l2norm_fwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/l2norm_fwd_kernel.ttgir", "l2norm_fwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/l2norm_fwd_kernel.llir", "l2norm_fwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/l2norm_fwd_kernel.ptx", "l2norm_fwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/l2norm_fwd_kernel.cubin", "l2norm_fwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/l2norm_fwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/l2norm_fwd_kernel.cubin b/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/l2norm_fwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..5b9f0fc4ab53ba30814328b6e8f3b1abd3ee478b Binary files /dev/null and b/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/l2norm_fwd_kernel.cubin differ diff --git a/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/l2norm_fwd_kernel.json b/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..2b285bb3f34532a970f449570235024996a9f18b --- /dev/null +++ b/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"hash": "0a93425db861505c2fa5d637b6fcac973dd29d3c8d58b1a1cb1ca10cbf955261", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 16, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 128, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_fwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/l2norm_fwd_kernel.llir b/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/l2norm_fwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..ab0bd2952bb707429d2ddc480f587e9e07c4ecd1 --- /dev/null +++ b/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/l2norm_fwd_kernel.llir @@ -0,0 +1,249 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 +@.str.2 = private unnamed_addr constant [17 x i8] c"__CUDA_PREC_SQRT\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_fwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, float %3, i32 %4, ptr addrspace(1) readnone captures(none) %5, ptr addrspace(1) readnone captures(none) %6) local_unnamed_addr #0 !dbg !5 { + %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %9 = zext nneg i32 %8 to i64, !dbg !14 + %10 = shl nuw nsw i64 %9, 5, !dbg !15 + %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !16 + %12 = and i32 %11, 496, !dbg !16 + %13 = lshr exact i32 %12, 4, !dbg !16 + %14 = zext nneg i32 %13 to i64, !dbg !17 + %15 = or disjoint i64 %10, %14, !dbg !17 + %16 = sext i32 %4 to i64, !dbg !18 + %17 = icmp slt i64 %15, %16, !dbg !18 + %18 = shl nuw nsw i32 %11, 3, !dbg !19 + %19 = and i32 %18, 120, !dbg !19 + %20 = shl nuw nsw i64 %15, 7, !dbg !20 + %21 = getelementptr bfloat, ptr addrspace(1) %0, i64 %20, !dbg !21 + %22 = zext nneg i32 %19 to i64, !dbg !22 + %23 = getelementptr bfloat, ptr addrspace(1) %21, i64 %22, !dbg !22 + %24 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %23, i1 %17) #6, !dbg !23 + %25 = extractvalue { i32, i32, i32, i32 } %24, 0, !dbg !23 + %26 = bitcast i32 %25 to <2 x bfloat>, !dbg !23 + %27 = extractvalue { i32, i32, i32, i32 } %24, 1, !dbg !23 + %28 = bitcast i32 %27 to <2 x bfloat>, !dbg !23 + %29 = extractvalue { i32, i32, i32, i32 } %24, 2, !dbg !23 + %30 = bitcast i32 %29 to <2 x bfloat>, !dbg !23 + %31 = extractvalue { i32, i32, i32, i32 } %24, 3, !dbg !23 + %32 = bitcast i32 %31 to <2 x bfloat>, !dbg !23 + %33 = extractelement <2 x bfloat> %26, i64 0, !dbg !23 + %34 = extractelement <2 x bfloat> %26, i64 1, !dbg !23 + %35 = extractelement <2 x bfloat> %28, i64 0, !dbg !23 + %36 = extractelement <2 x bfloat> %28, i64 1, !dbg !23 + %37 = extractelement <2 x bfloat> %30, i64 0, !dbg !23 + %38 = extractelement <2 x bfloat> %30, i64 1, !dbg !23 + %39 = extractelement <2 x bfloat> %32, i64 0, !dbg !23 + %40 = extractelement <2 x bfloat> %32, i64 1, !dbg !23 + %41 = fpext bfloat %33 to float, !dbg !24 + %42 = fpext bfloat %34 to float, !dbg !24 + %43 = fpext bfloat %35 to float, !dbg !24 + %44 = fpext bfloat %36 to float, !dbg !24 + %45 = fpext bfloat %37 to float, !dbg !24 + %46 = fpext bfloat %38 to float, !dbg !24 + %47 = fpext bfloat %39 to float, !dbg !24 + %48 = fpext bfloat %40 to float, !dbg !24 + %49 = fmul float %41, %41, !dbg !25 + %50 = fmul float %42, %42, !dbg !25 + %51 = fmul float %43, %43, !dbg !25 + %52 = fmul float %44, %44, !dbg !25 + %53 = fmul float %45, %45, !dbg !25 + %54 = fmul float %46, %46, !dbg !25 + %55 = fmul float %47, %47, !dbg !25 + %56 = fmul float %48, %48, !dbg !25 + %57 = fadd float %49, %50, !dbg !26 + %58 = fadd float %51, %57, !dbg !26 + %59 = fadd float %52, %58, !dbg !26 + %60 = fadd float %53, %59, !dbg !26 + %61 = fadd float %54, %60, !dbg !26 + %62 = fadd float %55, %61, !dbg !26 + %63 = fadd float %56, %62, !dbg !26 + %64 = bitcast float %63 to i32, !dbg !29 + %65 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %64, i32 8, i32 31), !dbg !29 + %66 = bitcast i32 %65 to float, !dbg !29 + %67 = fadd float %63, %66, !dbg !26 + %68 = bitcast float %67 to i32, !dbg !29 + %69 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %68, i32 4, i32 31), !dbg !29 + %70 = bitcast i32 %69 to float, !dbg !29 + %71 = fadd float %67, %70, !dbg !26 + %72 = bitcast float %71 to i32, !dbg !29 + %73 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %72, i32 2, i32 31), !dbg !29 + %74 = bitcast i32 %73 to float, !dbg !29 + %75 = fadd float %71, %74, !dbg !26 + %76 = bitcast float %75 to i32, !dbg !29 + %77 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %76, i32 1, i32 31), !dbg !29 + %78 = bitcast i32 %77 to float, !dbg !29 + %79 = fadd float %75, %78, !dbg !26 + %80 = fadd float %3, %79, !dbg !32 + %81 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i = icmp eq i32 %81, 0, !dbg !33 + %82 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i = icmp eq i32 %82, 0, !dbg !33 + br i1 %.not.i, label %88, label %83, !dbg !33 + +83: ; preds = %7 + br i1 %.not1.i, label %86, label %84, !dbg !33 + +84: ; preds = %83 + %85 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %80) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +86: ; preds = %83 + %87 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %80) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +88: ; preds = %7 + br i1 %.not1.i, label %91, label %89, !dbg !33 + +89: ; preds = %88 + %90 = tail call float @llvm.nvvm.sqrt.rn.f(float %80) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +91: ; preds = %88 + %92 = tail call float @llvm.nvvm.sqrt.approx.f(float %80) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +__nv_sqrtf.exit: ; preds = %84, %86, %89, %91 + %.0.i = phi float [ %85, %84 ], [ %87, %86 ], [ %90, %89 ], [ %92, %91 ], !dbg !33 + %93 = and i32 %11, 31, !dbg !16 + %94 = zext nneg i32 %93 to i64, !dbg !17 + %95 = or disjoint i64 %10, %94, !dbg !17 + %96 = getelementptr float, ptr addrspace(1) %2, i64 %95, !dbg !34 + %97 = getelementptr bfloat, ptr addrspace(1) %1, i64 %20, !dbg !35 + %98 = getelementptr bfloat, ptr addrspace(1) %97, i64 %22, !dbg !36 + %99 = icmp slt i64 %95, %16, !dbg !18 + %100 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i), !dbg !37 + %101 = fmul float %100, %41, !dbg !38 + %102 = fmul float %100, %42, !dbg !38 + %103 = fmul float %100, %43, !dbg !38 + %104 = fmul float %100, %44, !dbg !38 + %105 = fmul float %100, %45, !dbg !38 + %106 = fmul float %100, %46, !dbg !38 + %107 = fmul float %100, %47, !dbg !38 + %108 = fmul float %100, %48, !dbg !38 + %109 = fptrunc float %101 to bfloat, !dbg !39 + %110 = fptrunc float %102 to bfloat, !dbg !39 + %111 = fptrunc float %103 to bfloat, !dbg !39 + %112 = fptrunc float %104 to bfloat, !dbg !39 + %113 = fptrunc float %105 to bfloat, !dbg !39 + %114 = fptrunc float %106 to bfloat, !dbg !39 + %115 = fptrunc float %107 to bfloat, !dbg !39 + %116 = fptrunc float %108 to bfloat, !dbg !39 + %117 = insertelement <2 x bfloat> poison, bfloat %109, i64 0, !dbg !40 + %118 = insertelement <2 x bfloat> %117, bfloat %110, i64 1, !dbg !40 + %119 = bitcast <2 x bfloat> %118 to i32, !dbg !40 + %120 = insertelement <2 x bfloat> poison, bfloat %111, i64 0, !dbg !40 + %121 = insertelement <2 x bfloat> %120, bfloat %112, i64 1, !dbg !40 + %122 = bitcast <2 x bfloat> %121 to i32, !dbg !40 + %123 = insertelement <2 x bfloat> poison, bfloat %113, i64 0, !dbg !40 + %124 = insertelement <2 x bfloat> %123, bfloat %114, i64 1, !dbg !40 + %125 = bitcast <2 x bfloat> %124 to i32, !dbg !40 + %126 = insertelement <2 x bfloat> poison, bfloat %115, i64 0, !dbg !40 + %127 = insertelement <2 x bfloat> %126, bfloat %116, i64 1, !dbg !40 + %128 = bitcast <2 x bfloat> %127 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %119, i32 %122, i32 %125, i32 %128, ptr addrspace(1) %98, i1 %17) #6, !dbg !40 + %129 = lshr exact i32 %12, 2, !dbg !41 + %130 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %129, !dbg !41 + %131 = bitcast float %100 to <1 x i32>, !dbg !41 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) %130, <1 x i32> %131) #6, !dbg !41 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !41 + %132 = shl nuw nsw i32 %93, 2, !dbg !41 + %133 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %132, !dbg !41 + %134 = load i32, ptr addrspace(3) %133, align 4, !dbg !41 + %135 = and i32 %11, 480, !dbg !41 + %136 = icmp eq i32 %135, 0, !dbg !41 + %137 = and i1 %136, %99, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %134, ptr addrspace(1) %96, i1 %137) #6, !dbg !41 + ret void, !dbg !42 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.div.full(float, float) #3 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #4 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #5 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.f(float) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="512" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) } +attributes #4 = { convergent nocallback nounwind } +attributes #5 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #6 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "l2norm_fwd_kernel", linkageName: "l2norm_fwd_kernel", scope: !1, file: !1, line: 69, type: !6, scopeLine: 69, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !10, !11, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 70, column: 24, scope: !5) +!14 = !DILocation(line: 70, column: 30, scope: !5) +!15 = !DILocation(line: 71, column: 16, scope: !5) +!16 = !DILocation(line: 71, column: 34, scope: !5) +!17 = !DILocation(line: 71, column: 21, scope: !5) +!18 = !DILocation(line: 73, column: 16, scope: !5) +!19 = !DILocation(line: 74, column: 30, scope: !5) +!20 = !DILocation(line: 75, column: 29, scope: !5) +!21 = !DILocation(line: 75, column: 14, scope: !5) +!22 = !DILocation(line: 75, column: 33, scope: !5) +!23 = !DILocation(line: 79, column: 18, scope: !5) +!24 = !DILocation(line: 79, column: 47, scope: !5) +!25 = !DILocation(line: 80, column: 38, scope: !5) +!26 = !DILocation(line: 263, column: 15, scope: !27, inlinedAt: !29) +!27 = distinct !DILexicalBlockFile(scope: !5, file: !28, discriminator: 0) +!28 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!29 = !DILocation(line: 293, column: 36, scope: !27, inlinedAt: !30) +!30 = !DILocation(line: 80, column: 43, scope: !31) +!31 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!32 = !DILocation(line: 80, column: 48, scope: !5) +!33 = !DILocation(line: 80, column: 25, scope: !5) +!34 = !DILocation(line: 77, column: 20, scope: !5) +!35 = !DILocation(line: 76, column: 14, scope: !5) +!36 = !DILocation(line: 76, column: 33, scope: !5) +!37 = !DILocation(line: 80, column: 17, scope: !5) +!38 = !DILocation(line: 81, column: 16, scope: !5) +!39 = !DILocation(line: 83, column: 25, scope: !5) +!40 = !DILocation(line: 83, column: 18, scope: !5) +!41 = !DILocation(line: 84, column: 21, scope: !5) +!42 = !DILocation(line: 84, column: 4, scope: !5) diff --git a/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/l2norm_fwd_kernel.ptx b/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/l2norm_fwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..b613c24c0ffa7882b56068d70d6595ae30d1cb19 --- /dev/null +++ b/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/l2norm_fwd_kernel.ptx @@ -0,0 +1,399 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_fwd_kernel // -- Begin function l2norm_fwd_kernel +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; +.global .align 1 .b8 _$_str_$_2[17] = {95, 95, 67, 85, 68, 65, 95, 80, 82, 69, 67, 95, 83, 81, 82, 84}; + // @l2norm_fwd_kernel +.visible .entry l2norm_fwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_2, + .param .f32 l2norm_fwd_kernel_param_3, + .param .u32 l2norm_fwd_kernel_param_4, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_6 +) +.reqntid 512 +{ + .reg .pred %p<5>; + .reg .b16 %rs<9>; + .reg .b32 %r<61>; + .reg .b64 %rd<18>; + .loc 1 69 0 // l2norm.py:69:0 +$L__func_begin0: + .loc 1 69 0 // l2norm.py:69:0 + +// %bb.0: // %__nv_sqrtf.exit + ld.param.b64 %rd4, [l2norm_fwd_kernel_param_0]; + ld.param.b64 %rd5, [l2norm_fwd_kernel_param_1]; +$L__tmp0: + .loc 1 70 24 // l2norm.py:70:24 + mov.u32 %r13, %ctaid.x; + ld.param.b64 %rd6, [l2norm_fwd_kernel_param_2]; + .loc 1 71 16 // l2norm.py:71:16 + mul.wide.u32 %rd7, %r13, 32; + ld.param.b32 %r14, [l2norm_fwd_kernel_param_3]; + ld.param.s32 %rd8, [l2norm_fwd_kernel_param_4]; + .loc 1 71 34 // l2norm.py:71:34 + mov.u32 %r15, %tid.x; + and.b32 %r16, %r15, 496; + bfe.u32 %r17, %r15, 4, 5; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd9, %r17; + or.b64 %rd10, %rd7, %rd9; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p1, %rd10, %rd8; + .loc 1 74 30 // l2norm.py:74:30 + shl.b32 %r18, %r15, 3; + and.b32 %r19, %r18, 120; + .loc 1 75 14 // l2norm.py:75:14 + shl.b64 %rd11, %rd10, 8; + add.s64 %rd12, %rd4, %rd11; + .loc 1 75 33 // l2norm.py:75:33 + mul.wide.u32 %rd13, %r19, 2; + add.s64 %rd1, %rd12, %rd13; + mov.b32 %r5, 0; + .loc 1 79 18 // l2norm.py:79:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + mov.b32 {%rs5, %rs6}, %r3; + mov.b32 {%rs7, %rs8}, %r4; + .loc 1 79 47 // l2norm.py:79:47 + cvt.f32.bf16 %r20, %rs1; + cvt.f32.bf16 %r21, %rs2; + cvt.f32.bf16 %r22, %rs3; + cvt.f32.bf16 %r23, %rs4; + cvt.f32.bf16 %r24, %rs5; + cvt.f32.bf16 %r25, %rs6; + cvt.f32.bf16 %r26, %rs7; + cvt.f32.bf16 %r27, %rs8; + .loc 1 80 38 // l2norm.py:80:38 + mul.f32 %r28, %r21, %r21; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + fma.rn.f32.bf16 %r29, %rs1, %rs1, %r28; + fma.rn.f32.bf16 %r30, %rs3, %rs3, %r29; + fma.rn.f32.bf16 %r31, %rs4, %rs4, %r30; + fma.rn.f32.bf16 %r32, %rs5, %rs5, %r31; + fma.rn.f32.bf16 %r33, %rs6, %rs6, %r32; + fma.rn.f32.bf16 %r34, %rs7, %rs7, %r33; + fma.rn.f32.bf16 %r35, %rs8, %rs8, %r34; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r36, %r35, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r37, %r35, %r36; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r38, %r37, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r39, %r37, %r38; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r40, %r39, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r41, %r39, %r40; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r42, %r41, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r43, %r41, %r42; +$L__tmp10: + .loc 1 80 48 // l2norm.py:80:48 + add.f32 %r44, %r14, %r43; + .loc 1 80 25 // l2norm.py:80:25 + sqrt.approx.ftz.f32 %r45, %r44; + .loc 1 71 34 // l2norm.py:71:34 + and.b32 %r46, %r15, 31; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd14, %r46; + or.b64 %rd15, %rd7, %rd14; + .loc 1 77 20 // l2norm.py:77:20 + shl.b64 %rd16, %rd15, 2; + add.s64 %rd3, %rd6, %rd16; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd17, %rd5, %rd11; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd2, %rd17, %rd13; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p3, %rd15, %rd8; + mov.b32 %r47, 0f3F800000; + .loc 1 80 17 // l2norm.py:80:17 + div.full.f32 %r11, %r47, %r45; + .loc 1 81 16 // l2norm.py:81:16 + mul.f32 %r48, %r11, %r20; + mul.f32 %r49, %r11, %r21; + mul.f32 %r50, %r11, %r22; + mul.f32 %r51, %r11, %r23; + mul.f32 %r52, %r11, %r24; + mul.f32 %r53, %r11, %r25; + mul.f32 %r54, %r11, %r26; + mul.f32 %r55, %r11, %r27; + .loc 1 83 18 // l2norm.py:83:18 + cvt.rn.bf16x2.f32 %r6, %r49, %r48; + cvt.rn.bf16x2.f32 %r7, %r51, %r50; + cvt.rn.bf16x2.f32 %r8, %r53, %r52; + cvt.rn.bf16x2.f32 %r9, %r55, %r54; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd2 + 0 ], { %r6, %r7, %r8, %r9 }; + // end inline asm + .loc 1 84 21 // l2norm.py:84:21 + shr.u32 %r56, %r16, 2; + mov.b32 %r57, global_smem; + add.s32 %r10, %r57, %r56; + // begin inline asm + st.shared.b32 [ %r10 + 0 ], %r11; + // end inline asm + bar.sync 0; + shl.b32 %r58, %r46, 2; + add.s32 %r59, %r57, %r58; + ld.shared.b32 %r12, [%r59]; + and.b32 %r60, %r15, 480; + setp.eq.b32 %p4, %r60, 0; + and.pred %p2, %p4, %p3; + // begin inline asm + @%p2 st.global.b32 [ %rd3 + 0 ], { %r12 }; + // end inline asm + .loc 1 84 4 // l2norm.py:84:4 + ret; +$L__tmp11: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp10 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 80 // DW_AT_call_line +.b8 43 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp10 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/l2norm_fwd_kernel.source b/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/l2norm_fwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..d9fa83aac0600b1e5d35d95ed8aaad5c5636e577 --- /dev/null +++ b/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/l2norm_fwd_kernel.source @@ -0,0 +1,172 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc38 = loc(unknown) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc45 = loc("x"(#loc)) +#loc46 = loc("y"(#loc)) +#loc47 = loc("rstd"(#loc)) +#loc48 = loc("eps"(#loc)) +#loc49 = loc("T"(#loc)) +#loc81 = loc("input"(#loc36)) +#loc82 = loc("a"(#loc41)) +#loc83 = loc("b"(#loc41)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc50) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc51) + %o_t = arith.constant 32 : i32 loc(#loc52) + %o_t_1 = arith.constant 32 : i64 loc(#loc52) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc52) + %o_t_3 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32> loc(#loc53) + %o_t_4 = arith.extsi %o_t_3 : tensor<32xi32> to tensor<32xi64> loc(#loc54) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<32xi64> loc(#loc54) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<32xi64> loc(#loc54) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc55) + %m_t = arith.extsi %T : i32 to i64 loc(#loc56) + %m_t_7 = tt.splat %m_t : i64 -> tensor<32xi64> loc(#loc56) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<32xi64> loc(#loc56) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<32xi1> -> tensor<32x1xi1> loc(#loc57) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc58) + %m_x_10 = arith.constant 128 : i32 loc(#loc59) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc59) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc59) + %m_x_13 = tt.broadcast %m_x : tensor<32x1xi1> -> tensor<32x128xi1> loc(#loc60) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<32x128xi1> loc(#loc60) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<32x128xi1> loc(#loc60) + %p_x = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<32xi64> -> tensor<32x1xi64> loc(#loc61) + %p_x_16 = arith.constant 128 : i32 loc(#loc62) + %p_x_17 = arith.constant 128 : i64 loc(#loc62) + %p_x_18 = arith.constant dense<128> : tensor<32x1xi64> loc(#loc62) + %p_x_19 = arith.muli %p_x, %p_x_18 : tensor<32x1xi64> loc(#loc62) + %p_x_20 = tt.splat %x : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc63) + %p_x_21 = tt.addptr %p_x_20, %p_x_19 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc63) + %p_x_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc64) + %p_x_23 = tt.broadcast %p_x_21 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc65) + %p_x_24 = tt.broadcast %p_x_22 : tensor<1x128xi32> -> tensor<32x128xi32> loc(#loc65) + %p_x_25 = tt.addptr %p_x_23, %p_x_24 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc65) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<32xi64> -> tensor<32x1xi64> loc(#loc66) + %p_y_26 = arith.constant 128 : i32 loc(#loc67) + %p_y_27 = arith.constant 128 : i64 loc(#loc67) + %p_y_28 = arith.constant dense<128> : tensor<32x1xi64> loc(#loc67) + %p_y_29 = arith.muli %p_y, %p_y_28 : tensor<32x1xi64> loc(#loc67) + %p_y_30 = tt.splat %y : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc68) + %p_y_31 = tt.addptr %p_y_30, %p_y_29 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc68) + %p_y_32 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc69) + %p_y_33 = tt.broadcast %p_y_31 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc70) + %p_y_34 = tt.broadcast %p_y_32 : tensor<1x128xi32> -> tensor<32x128xi32> loc(#loc70) + %p_y_35 = tt.addptr %p_y_33, %p_y_34 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc70) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<32x!tt.ptr> loc(#loc71) + %p_rstd_36 = tt.addptr %p_rstd, %o_t_6 : tensor<32x!tt.ptr>, tensor<32xi64> loc(#loc71) + %b_x = arith.constant 0.000000e+00 : f32 loc(#loc72) + %b_x_37 = arith.constant dense<0.000000e+00> : tensor<32x128xf32> loc(#loc72) + %b_x_38 = arith.truncf %b_x_37 : tensor<32x128xf32> to tensor<32x128xbf16> loc(#loc72) + %b_x_39 = tt.load %p_x_25, %m_x_15, %b_x_38 : tensor<32x128x!tt.ptr> loc(#loc72) + %b_x_40 = arith.extf %b_x_39 : tensor<32x128xbf16> to tensor<32x128xf32> loc(#loc73) + %b_rstd = arith.mulf %b_x_40, %b_x_40 : tensor<32x128xf32> loc(#loc74) + %b_rstd_41 = tt.call @triton.language.standard.sum__fp32S32_128S_c1_cFalse_cNone(%b_rstd) : (tensor<32x128xf32>) -> tensor<32xf32> loc(#loc75) + %b_rstd_42 = tt.splat %eps : f32 -> tensor<32xf32> loc(#loc76) + %b_rstd_43 = arith.addf %b_rstd_41, %b_rstd_42 : tensor<32xf32> loc(#loc76) + %b_rstd_44 = math.sqrt %b_rstd_43 : tensor<32xf32> loc(#loc77) + %b_rstd_45 = arith.constant 1 : i32 loc(#loc78) + %b_rstd_46 = arith.constant 1.000000e+00 : f32 loc(#loc78) + %b_rstd_47 = arith.constant dense<1.000000e+00> : tensor<32xf32> loc(#loc78) + %b_rstd_48 = arith.divf %b_rstd_47, %b_rstd_44 : tensor<32xf32> loc(#loc78) + %b_y = tt.expand_dims %b_rstd_48 {axis = 1 : i32} : tensor<32xf32> -> tensor<32x1xf32> loc(#loc79) + %b_y_49 = tt.broadcast %b_y : tensor<32x1xf32> -> tensor<32x128xf32> loc(#loc80) + %b_y_50 = arith.mulf %b_x_40, %b_y_49 : tensor<32x128xf32> loc(#loc80) + %0 = arith.truncf %b_y_50 : tensor<32x128xf32> to tensor<32x128xbf16> loc(#loc32) + tt.store %p_y_35, %0, %m_x_15 : tensor<32x128x!tt.ptr> loc(#loc33) + tt.store %p_rstd_36, %b_rstd_48, %m_t_8 : tensor<32x!tt.ptr> loc(#loc34) + tt.return loc(#loc35) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S32_128S_c1_cFalse_cNone(%input: tensor<32x128xf32> loc("input"(#loc36))) -> tensor<32xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc37) + tt.reduce.return %2 : f32 loc(#loc37) + }) : (tensor<32x128xf32>) -> tensor<32xf32> loc(#loc37) + tt.return %0 : tensor<32xf32> loc(#loc39) + ^bb1: // no predecessors + %1 = ub.poison : tensor<32xf32> loc(#loc40) + tt.return %1 : tensor<32xf32> loc(#loc40) + } loc(#loc36) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc41)), %b: f32 loc("b"(#loc41))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc42) + tt.return %0 : f32 loc(#loc43) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc44) + tt.return %1 : f32 loc(#loc44) + } loc(#loc41) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:18) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:29) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:37) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc50 = loc("i_t"(#loc1)) +#loc51 = loc("i_t"(#loc2)) +#loc52 = loc("o_t"(#loc3)) +#loc53 = loc("o_t"(#loc4)) +#loc54 = loc("o_t"(#loc5)) +#loc55 = loc("o_d"(#loc6)) +#loc56 = loc("m_t"(#loc7)) +#loc57 = loc("m_x"(#loc8)) +#loc58 = loc("m_x"(#loc9)) +#loc59 = loc("m_x"(#loc10)) +#loc60 = loc("m_x"(#loc11)) +#loc61 = loc("p_x"(#loc12)) +#loc62 = loc("p_x"(#loc13)) +#loc63 = loc("p_x"(#loc14)) +#loc64 = loc("p_x"(#loc15)) +#loc65 = loc("p_x"(#loc16)) +#loc66 = loc("p_y"(#loc17)) +#loc67 = loc("p_y"(#loc18)) +#loc68 = loc("p_y"(#loc19)) +#loc69 = loc("p_y"(#loc20)) +#loc70 = loc("p_y"(#loc21)) +#loc71 = loc("p_rstd"(#loc22)) +#loc72 = loc("b_x"(#loc23)) +#loc73 = loc("b_x"(#loc24)) +#loc74 = loc("b_rstd"(#loc25)) +#loc75 = loc("b_rstd"(#loc26)) +#loc76 = loc("b_rstd"(#loc27)) +#loc77 = loc("b_rstd"(#loc28)) +#loc78 = loc("b_rstd"(#loc29)) +#loc79 = loc("b_y"(#loc30)) +#loc80 = loc("b_y"(#loc31)) diff --git a/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/l2norm_fwd_kernel.ttgir b/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/l2norm_fwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..db58b001d1884266b691359240132106e3d85aec --- /dev/null +++ b/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/l2norm_fwd_kernel.ttgir @@ -0,0 +1,135 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [16, 1], order = [1, 0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [16], order = [0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc1 = loc(unknown) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc34 = loc("x"(#loc)) +#loc35 = loc("y"(#loc)) +#loc36 = loc("rstd"(#loc)) +#loc37 = loc("eps"(#loc)) +#loc38 = loc("T"(#loc)) +#loc59 = loc("b_rstd"(#loc23)) +#loc66 = loc(callsite(#loc1 at #loc59)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 16 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<1.000000e+00> : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<32x128xbf16, #blocked> loc(#loc1) + %c32_i64 = arith.constant 32 : i64 loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<32x1xi64, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc39) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc40) + %o_t = arith.muli %i_t_3, %c32_i64 : i64 loc(#loc41) + %o_t_4 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc42) + %o_t_5 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32, #blocked1> loc(#loc42) + %o_t_6 = arith.extsi %o_t_4 : tensor<32xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_7 = arith.extsi %o_t_5 : tensor<32xi32, #blocked1> to tensor<32xi64, #blocked1> loc(#loc43) + %o_t_8 = tt.splat %o_t : i64 -> tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_9 = tt.splat %o_t : i64 -> tensor<32xi64, #blocked1> loc(#loc43) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<32xi64, #blocked1> loc(#loc43) + %m_t = arith.extsi %T : i32 to i64 loc(#loc44) + %m_t_12 = tt.splat %m_t : i64 -> tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_13 = tt.splat %m_t : i64 -> tensor<32xi64, #blocked1> loc(#loc44) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<32xi64, #blocked1> loc(#loc44) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<32xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32x1xi1, #blocked> loc(#loc45) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc46) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc46) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_1 : tensor<1x128xi32, #blocked> loc(#loc47) + %m_x_19 = tt.broadcast %m_x : tensor<32x1xi1, #blocked> -> tensor<32x128xi1, #blocked> loc(#loc48) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<32x128xi1, #blocked> loc(#loc48) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<32x128xi1, #blocked> loc(#loc48) + %p_x = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32x1xi64, #blocked> loc(#loc49) + %p_x_22 = arith.muli %p_x, %cst_2 : tensor<32x1xi64, #blocked> loc(#loc50) + %p_x_23 = tt.splat %x : !tt.ptr -> tensor<32x1x!tt.ptr, #blocked> loc(#loc51) + %p_x_24 = tt.addptr %p_x_23, %p_x_22 : tensor<32x1x!tt.ptr, #blocked>, tensor<32x1xi64, #blocked> loc(#loc51) + %p_x_25 = tt.broadcast %p_x_24 : tensor<32x1x!tt.ptr, #blocked> -> tensor<32x128x!tt.ptr, #blocked> loc(#loc52) + %p_x_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<32x128xi32, #blocked> loc(#loc52) + %p_x_27 = tt.addptr %p_x_25, %p_x_26 : tensor<32x128x!tt.ptr, #blocked>, tensor<32x128xi32, #blocked> loc(#loc52) + %p_y = tt.splat %y : !tt.ptr -> tensor<32x1x!tt.ptr, #blocked> loc(#loc53) + %p_y_28 = tt.addptr %p_y, %p_x_22 : tensor<32x1x!tt.ptr, #blocked>, tensor<32x1xi64, #blocked> loc(#loc53) + %p_y_29 = tt.broadcast %p_y_28 : tensor<32x1x!tt.ptr, #blocked> -> tensor<32x128x!tt.ptr, #blocked> loc(#loc54) + %p_y_30 = tt.addptr %p_y_29, %p_x_26 : tensor<32x128x!tt.ptr, #blocked>, tensor<32x128xi32, #blocked> loc(#loc54) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<32x!tt.ptr, #blocked1> loc(#loc55) + %p_rstd_31 = tt.addptr %p_rstd, %o_t_11 : tensor<32x!tt.ptr, #blocked1>, tensor<32xi64, #blocked1> loc(#loc55) + %b_x = tt.load %p_x_27, %m_x_21, %cst_0 : tensor<32x128x!tt.ptr, #blocked> loc(#loc56) + %b_x_32 = arith.extf %b_x : tensor<32x128xbf16, #blocked> to tensor<32x128xf32, #blocked> loc(#loc57) + %b_rstd = arith.mulf %b_x_32, %b_x_32 : tensor<32x128xf32, #blocked> loc(#loc58) + %b_rstd_33 = "tt.reduce"(%b_rstd) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_40: f32 loc(callsite(#loc1 at #loc59)), %b_rstd_41: f32 loc(callsite(#loc1 at #loc59))): + %b_rstd_42 = arith.addf %b_rstd_40, %b_rstd_41 : f32 loc(#loc67) + tt.reduce.return %b_rstd_42 : f32 loc(#loc65) + }) : (tensor<32x128xf32, #blocked>) -> tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc65) + %b_rstd_34 = tt.splat %eps : f32 -> tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_36 = math.sqrt %b_rstd_35 : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc61) + %b_rstd_37 = arith.divf %cst, %b_rstd_36 : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc62) + %b_y = tt.expand_dims %b_rstd_37 {axis = 1 : i32} : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32x1xf32, #blocked> loc(#loc63) + %b_y_38 = tt.broadcast %b_y : tensor<32x1xf32, #blocked> -> tensor<32x128xf32, #blocked> loc(#loc64) + %b_y_39 = arith.mulf %b_x_32, %b_y_38 : tensor<32x128xf32, #blocked> loc(#loc64) + %0 = arith.truncf %b_y_39 : tensor<32x128xf32, #blocked> to tensor<32x128xbf16, #blocked> loc(#loc30) + tt.store %p_y_30, %0, %m_x_21 : tensor<32x128x!tt.ptr, #blocked> loc(#loc31) + %1 = ttg.convert_layout %b_rstd_37 : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32xf32, #blocked1> loc(#loc32) + tt.store %p_rstd_31, %1, %m_t_15 : tensor<32x!tt.ptr, #blocked1> loc(#loc32) + tt.return loc(#loc33) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc39 = loc("i_t"(#loc2)) +#loc40 = loc("i_t"(#loc3)) +#loc41 = loc("o_t"(#loc4)) +#loc42 = loc("o_t"(#loc5)) +#loc43 = loc("o_t"(#loc6)) +#loc44 = loc("m_t"(#loc7)) +#loc45 = loc("m_x"(#loc8)) +#loc46 = loc("m_x"(#loc9)) +#loc47 = loc("m_x"(#loc10)) +#loc48 = loc("m_x"(#loc11)) +#loc49 = loc("p_x"(#loc12)) +#loc50 = loc("p_x"(#loc13)) +#loc51 = loc("p_x"(#loc14)) +#loc52 = loc("p_x"(#loc15)) +#loc53 = loc("p_y"(#loc16)) +#loc54 = loc("p_y"(#loc17)) +#loc55 = loc("p_rstd"(#loc18)) +#loc56 = loc("b_x"(#loc19)) +#loc57 = loc("b_x"(#loc20)) +#loc58 = loc("b_rstd"(#loc21)) +#loc60 = loc("b_rstd"(#loc25)) +#loc61 = loc("b_rstd"(#loc26)) +#loc62 = loc("b_rstd"(#loc27)) +#loc63 = loc("b_y"(#loc28)) +#loc64 = loc("b_y"(#loc29)) +#loc65 = loc(callsite(#loc22 at #loc59)) +#loc67 = loc(callsite(#loc24 at #loc65)) diff --git a/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/l2norm_fwd_kernel.ttir b/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/l2norm_fwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..3688281bceaf7f8abdf8568104090be5e0170141 --- /dev/null +++ b/triton/gpu0/BKJUEXNYMFIFYL5F2Y33N7FMS465FHJ4RVMLDIOLDSQQZP4VKJQQ/l2norm_fwd_kernel.ttir @@ -0,0 +1,128 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc3 = loc(unknown) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc35 = loc("x"(#loc)) +#loc36 = loc("y"(#loc)) +#loc37 = loc("rstd"(#loc)) +#loc38 = loc("eps"(#loc)) +#loc39 = loc("T"(#loc)) +#loc62 = loc("b_rstd"(#loc25)) +#loc68 = loc(callsite(#loc3 at #loc62)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %b_x = arith.constant dense<0.000000e+00> : tensor<32x128xbf16> loc(#loc40) + %b_rstd = arith.constant dense<1.000000e+00> : tensor<32xf32> loc(#loc41) + %cst = arith.constant dense<128> : tensor<32x1xi64> loc(#loc3) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc42) + %o_t = arith.constant 32 : i64 loc(#loc43) + %i_t = tt.get_program_id x : i32 loc(#loc44) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc45) + %o_t_1 = arith.muli %i_t_0, %o_t : i64 loc(#loc43) + %o_t_2 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32> loc(#loc46) + %o_t_3 = arith.extsi %o_t_2 : tensor<32xi32> to tensor<32xi64> loc(#loc47) + %o_t_4 = tt.splat %o_t_1 : i64 -> tensor<32xi64> loc(#loc47) + %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<32xi64> loc(#loc47) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc48) + %m_t = arith.extsi %T : i32 to i64 loc(#loc49) + %m_t_6 = tt.splat %m_t : i64 -> tensor<32xi64> loc(#loc49) + %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<32xi64> loc(#loc49) + %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<32xi1> -> tensor<32x1xi1> loc(#loc50) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc51) + %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc42) + %m_x_11 = tt.broadcast %m_x_8 : tensor<32x1xi1> -> tensor<32x128xi1> loc(#loc52) + %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<32x128xi1> loc(#loc52) + %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<32x128xi1> loc(#loc52) + %p_x = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<32xi64> -> tensor<32x1xi64> loc(#loc53) + %p_x_14 = arith.muli %p_x, %cst : tensor<32x1xi64> loc(#loc54) + %p_x_15 = tt.splat %x : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc55) + %p_x_16 = tt.addptr %p_x_15, %p_x_14 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc55) + %p_x_17 = tt.broadcast %p_x_16 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc56) + %p_x_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<32x128xi32> loc(#loc56) + %p_x_19 = tt.addptr %p_x_17, %p_x_18 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc56) + %p_y = tt.splat %y : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc57) + %p_y_20 = tt.addptr %p_y, %p_x_14 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc57) + %p_y_21 = tt.broadcast %p_y_20 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc58) + %p_y_22 = tt.addptr %p_y_21, %p_x_18 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<32x!tt.ptr> loc(#loc59) + %p_rstd_23 = tt.addptr %p_rstd, %o_t_5 : tensor<32x!tt.ptr>, tensor<32xi64> loc(#loc59) + %b_x_24 = tt.load %p_x_19, %m_x_13, %b_x : tensor<32x128x!tt.ptr> loc(#loc40) + %b_x_25 = arith.extf %b_x_24 : tensor<32x128xbf16> to tensor<32x128xf32> loc(#loc60) + %b_rstd_26 = arith.mulf %b_x_25, %b_x_25 : tensor<32x128xf32> loc(#loc61) + %b_rstd_27 = "tt.reduce"(%b_rstd_26) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_34: f32 loc(callsite(#loc3 at #loc62)), %b_rstd_35: f32 loc(callsite(#loc3 at #loc62))): + %b_rstd_36 = arith.addf %b_rstd_34, %b_rstd_35 : f32 loc(#loc69) + tt.reduce.return %b_rstd_36 : f32 loc(#loc67) + }) : (tensor<32x128xf32>) -> tensor<32xf32> loc(#loc67) + %b_rstd_28 = tt.splat %eps : f32 -> tensor<32xf32> loc(#loc63) + %b_rstd_29 = arith.addf %b_rstd_27, %b_rstd_28 : tensor<32xf32> loc(#loc63) + %b_rstd_30 = math.sqrt %b_rstd_29 : tensor<32xf32> loc(#loc64) + %b_rstd_31 = arith.divf %b_rstd, %b_rstd_30 : tensor<32xf32> loc(#loc41) + %b_y = tt.expand_dims %b_rstd_31 {axis = 1 : i32} : tensor<32xf32> -> tensor<32x1xf32> loc(#loc65) + %b_y_32 = tt.broadcast %b_y : tensor<32x1xf32> -> tensor<32x128xf32> loc(#loc66) + %b_y_33 = arith.mulf %b_x_25, %b_y_32 : tensor<32x128xf32> loc(#loc66) + %0 = arith.truncf %b_y_33 : tensor<32x128xf32> to tensor<32x128xbf16> loc(#loc31) + tt.store %p_y_22, %0, %m_x_13 : tensor<32x128x!tt.ptr> loc(#loc32) + tt.store %p_rstd_23, %b_rstd_31, %m_t_7 : tensor<32x!tt.ptr> loc(#loc33) + tt.return loc(#loc34) + } loc(#loc) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc40 = loc("b_x"(#loc1)) +#loc41 = loc("b_rstd"(#loc2)) +#loc42 = loc("m_x"(#loc4)) +#loc43 = loc("o_t"(#loc5)) +#loc44 = loc("i_t"(#loc6)) +#loc45 = loc("i_t"(#loc7)) +#loc46 = loc("o_t"(#loc8)) +#loc47 = loc("o_t"(#loc9)) +#loc48 = loc("o_d"(#loc10)) +#loc49 = loc("m_t"(#loc11)) +#loc50 = loc("m_x"(#loc12)) +#loc51 = loc("m_x"(#loc13)) +#loc52 = loc("m_x"(#loc14)) +#loc53 = loc("p_x"(#loc15)) +#loc54 = loc("p_x"(#loc16)) +#loc55 = loc("p_x"(#loc17)) +#loc56 = loc("p_x"(#loc18)) +#loc57 = loc("p_y"(#loc19)) +#loc58 = loc("p_y"(#loc20)) +#loc59 = loc("p_rstd"(#loc21)) +#loc60 = loc("b_x"(#loc22)) +#loc61 = loc("b_rstd"(#loc23)) +#loc63 = loc("b_rstd"(#loc27)) +#loc64 = loc("b_rstd"(#loc28)) +#loc65 = loc("b_y"(#loc29)) +#loc66 = loc("b_y"(#loc30)) +#loc67 = loc(callsite(#loc24 at #loc62)) +#loc69 = loc(callsite(#loc26 at #loc67)) diff --git a/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/__grp__l2norm_bwd_kernel.json b/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/__grp__l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..02c7a6b7fbe509da81c20a4edcd18e5b232c79f6 --- /dev/null +++ b/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/__grp__l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_bwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/l2norm_bwd_kernel.source", "l2norm_bwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/l2norm_bwd_kernel.ttir", "l2norm_bwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/l2norm_bwd_kernel.ttgir", "l2norm_bwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/l2norm_bwd_kernel.llir", "l2norm_bwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/l2norm_bwd_kernel.ptx", "l2norm_bwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/l2norm_bwd_kernel.cubin", "l2norm_bwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/l2norm_bwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/l2norm_bwd_kernel.cubin b/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/l2norm_bwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..b487da32db330462db72f8ceb19a476ed5318a19 Binary files /dev/null and b/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/l2norm_bwd_kernel.cubin differ diff --git a/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/l2norm_bwd_kernel.json b/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..27f9df2bda4be25f132c772e1a1afdab46573307 --- /dev/null +++ b/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"hash": "0a9444b143c63655acf57e512db7c4a7e315536f76040534df048793e7b77490", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 2, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 0, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_bwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/l2norm_bwd_kernel.llir b/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/l2norm_bwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..849296bcccd7cb177c37382bd0efbc0b6eacd50c --- /dev/null +++ b/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/l2norm_bwd_kernel.llir @@ -0,0 +1,673 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_bwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, ptr addrspace(1) %3, float %4, i32 %5, ptr addrspace(1) readnone captures(none) %6, ptr addrspace(1) readnone captures(none) %7) local_unnamed_addr #0 !dbg !4 { + %9 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !12 + %10 = zext nneg i32 %9 to i64, !dbg !13 + %11 = shl nuw nsw i64 %10, 4, !dbg !14 + %12 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !15 + %13 = lshr i32 %12, 4, !dbg !15 + %14 = and i32 %13, 3, !dbg !15 + %15 = or disjoint i32 %14, 4, !dbg !15 + %16 = or disjoint i32 %14, 8, !dbg !15 + %17 = or disjoint i32 %14, 12, !dbg !15 + %18 = zext nneg i32 %14 to i64, !dbg !16 + %19 = zext nneg i32 %15 to i64, !dbg !16 + %20 = zext nneg i32 %16 to i64, !dbg !16 + %21 = zext nneg i32 %17 to i64, !dbg !16 + %22 = or disjoint i64 %11, %18, !dbg !16 + %23 = or disjoint i64 %11, %19, !dbg !16 + %24 = or disjoint i64 %11, %20, !dbg !16 + %25 = or disjoint i64 %11, %21, !dbg !16 + %26 = sext i32 %5 to i64, !dbg !17 + %27 = icmp slt i64 %22, %26, !dbg !17 + %28 = icmp slt i64 %23, %26, !dbg !17 + %29 = icmp slt i64 %24, %26, !dbg !17 + %30 = icmp slt i64 %25, %26, !dbg !17 + %31 = shl nuw nsw i32 %12, 3, !dbg !18 + %32 = and i32 %31, 120, !dbg !18 + %33 = shl nuw nsw i64 %22, 7, !dbg !19 + %34 = shl nuw nsw i64 %23, 7, !dbg !19 + %35 = shl nuw nsw i64 %24, 7, !dbg !19 + %36 = shl nuw nsw i64 %25, 7, !dbg !19 + %37 = getelementptr bfloat, ptr addrspace(1) %0, i64 %33, !dbg !20 + %38 = getelementptr bfloat, ptr addrspace(1) %0, i64 %34, !dbg !20 + %39 = getelementptr bfloat, ptr addrspace(1) %0, i64 %35, !dbg !20 + %40 = getelementptr bfloat, ptr addrspace(1) %0, i64 %36, !dbg !20 + %41 = zext nneg i32 %32 to i64, !dbg !21 + %42 = getelementptr bfloat, ptr addrspace(1) %37, i64 %41, !dbg !21 + %43 = getelementptr bfloat, ptr addrspace(1) %38, i64 %41, !dbg !21 + %44 = getelementptr bfloat, ptr addrspace(1) %39, i64 %41, !dbg !21 + %45 = getelementptr bfloat, ptr addrspace(1) %40, i64 %41, !dbg !21 + %46 = getelementptr float, ptr addrspace(1) %1, i64 %22, !dbg !22 + %47 = getelementptr float, ptr addrspace(1) %1, i64 %23, !dbg !22 + %48 = getelementptr float, ptr addrspace(1) %1, i64 %24, !dbg !22 + %49 = getelementptr float, ptr addrspace(1) %1, i64 %25, !dbg !22 + %50 = getelementptr bfloat, ptr addrspace(1) %2, i64 %33, !dbg !23 + %51 = getelementptr bfloat, ptr addrspace(1) %2, i64 %34, !dbg !23 + %52 = getelementptr bfloat, ptr addrspace(1) %2, i64 %35, !dbg !23 + %53 = getelementptr bfloat, ptr addrspace(1) %2, i64 %36, !dbg !23 + %54 = getelementptr bfloat, ptr addrspace(1) %50, i64 %41, !dbg !24 + %55 = getelementptr bfloat, ptr addrspace(1) %51, i64 %41, !dbg !24 + %56 = getelementptr bfloat, ptr addrspace(1) %52, i64 %41, !dbg !24 + %57 = getelementptr bfloat, ptr addrspace(1) %53, i64 %41, !dbg !24 + %58 = getelementptr bfloat, ptr addrspace(1) %3, i64 %33, !dbg !25 + %59 = getelementptr bfloat, ptr addrspace(1) %3, i64 %34, !dbg !25 + %60 = getelementptr bfloat, ptr addrspace(1) %3, i64 %35, !dbg !25 + %61 = getelementptr bfloat, ptr addrspace(1) %3, i64 %36, !dbg !25 + %62 = getelementptr bfloat, ptr addrspace(1) %58, i64 %41, !dbg !26 + %63 = getelementptr bfloat, ptr addrspace(1) %59, i64 %41, !dbg !26 + %64 = getelementptr bfloat, ptr addrspace(1) %60, i64 %41, !dbg !26 + %65 = getelementptr bfloat, ptr addrspace(1) %61, i64 %41, !dbg !26 + %66 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %42, i1 %27) #3, !dbg !27 + %67 = extractvalue { i32, i32, i32, i32 } %66, 0, !dbg !27 + %68 = bitcast i32 %67 to <2 x bfloat>, !dbg !27 + %69 = extractvalue { i32, i32, i32, i32 } %66, 1, !dbg !27 + %70 = bitcast i32 %69 to <2 x bfloat>, !dbg !27 + %71 = extractvalue { i32, i32, i32, i32 } %66, 2, !dbg !27 + %72 = bitcast i32 %71 to <2 x bfloat>, !dbg !27 + %73 = extractvalue { i32, i32, i32, i32 } %66, 3, !dbg !27 + %74 = bitcast i32 %73 to <2 x bfloat>, !dbg !27 + %75 = extractelement <2 x bfloat> %68, i64 0, !dbg !27 + %76 = extractelement <2 x bfloat> %68, i64 1, !dbg !27 + %77 = extractelement <2 x bfloat> %70, i64 0, !dbg !27 + %78 = extractelement <2 x bfloat> %70, i64 1, !dbg !27 + %79 = extractelement <2 x bfloat> %72, i64 0, !dbg !27 + %80 = extractelement <2 x bfloat> %72, i64 1, !dbg !27 + %81 = extractelement <2 x bfloat> %74, i64 0, !dbg !27 + %82 = extractelement <2 x bfloat> %74, i64 1, !dbg !27 + %83 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %43, i1 %28) #3, !dbg !27 + %84 = extractvalue { i32, i32, i32, i32 } %83, 0, !dbg !27 + %85 = bitcast i32 %84 to <2 x bfloat>, !dbg !27 + %86 = extractvalue { i32, i32, i32, i32 } %83, 1, !dbg !27 + %87 = bitcast i32 %86 to <2 x bfloat>, !dbg !27 + %88 = extractvalue { i32, i32, i32, i32 } %83, 2, !dbg !27 + %89 = bitcast i32 %88 to <2 x bfloat>, !dbg !27 + %90 = extractvalue { i32, i32, i32, i32 } %83, 3, !dbg !27 + %91 = bitcast i32 %90 to <2 x bfloat>, !dbg !27 + %92 = extractelement <2 x bfloat> %85, i64 0, !dbg !27 + %93 = extractelement <2 x bfloat> %85, i64 1, !dbg !27 + %94 = extractelement <2 x bfloat> %87, i64 0, !dbg !27 + %95 = extractelement <2 x bfloat> %87, i64 1, !dbg !27 + %96 = extractelement <2 x bfloat> %89, i64 0, !dbg !27 + %97 = extractelement <2 x bfloat> %89, i64 1, !dbg !27 + %98 = extractelement <2 x bfloat> %91, i64 0, !dbg !27 + %99 = extractelement <2 x bfloat> %91, i64 1, !dbg !27 + %100 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %44, i1 %29) #3, !dbg !27 + %101 = extractvalue { i32, i32, i32, i32 } %100, 0, !dbg !27 + %102 = bitcast i32 %101 to <2 x bfloat>, !dbg !27 + %103 = extractvalue { i32, i32, i32, i32 } %100, 1, !dbg !27 + %104 = bitcast i32 %103 to <2 x bfloat>, !dbg !27 + %105 = extractvalue { i32, i32, i32, i32 } %100, 2, !dbg !27 + %106 = bitcast i32 %105 to <2 x bfloat>, !dbg !27 + %107 = extractvalue { i32, i32, i32, i32 } %100, 3, !dbg !27 + %108 = bitcast i32 %107 to <2 x bfloat>, !dbg !27 + %109 = extractelement <2 x bfloat> %102, i64 0, !dbg !27 + %110 = extractelement <2 x bfloat> %102, i64 1, !dbg !27 + %111 = extractelement <2 x bfloat> %104, i64 0, !dbg !27 + %112 = extractelement <2 x bfloat> %104, i64 1, !dbg !27 + %113 = extractelement <2 x bfloat> %106, i64 0, !dbg !27 + %114 = extractelement <2 x bfloat> %106, i64 1, !dbg !27 + %115 = extractelement <2 x bfloat> %108, i64 0, !dbg !27 + %116 = extractelement <2 x bfloat> %108, i64 1, !dbg !27 + %117 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %45, i1 %30) #3, !dbg !27 + %118 = extractvalue { i32, i32, i32, i32 } %117, 0, !dbg !27 + %119 = bitcast i32 %118 to <2 x bfloat>, !dbg !27 + %120 = extractvalue { i32, i32, i32, i32 } %117, 1, !dbg !27 + %121 = bitcast i32 %120 to <2 x bfloat>, !dbg !27 + %122 = extractvalue { i32, i32, i32, i32 } %117, 2, !dbg !27 + %123 = bitcast i32 %122 to <2 x bfloat>, !dbg !27 + %124 = extractvalue { i32, i32, i32, i32 } %117, 3, !dbg !27 + %125 = bitcast i32 %124 to <2 x bfloat>, !dbg !27 + %126 = extractelement <2 x bfloat> %119, i64 0, !dbg !27 + %127 = extractelement <2 x bfloat> %119, i64 1, !dbg !27 + %128 = extractelement <2 x bfloat> %121, i64 0, !dbg !27 + %129 = extractelement <2 x bfloat> %121, i64 1, !dbg !27 + %130 = extractelement <2 x bfloat> %123, i64 0, !dbg !27 + %131 = extractelement <2 x bfloat> %123, i64 1, !dbg !27 + %132 = extractelement <2 x bfloat> %125, i64 0, !dbg !27 + %133 = extractelement <2 x bfloat> %125, i64 1, !dbg !27 + %134 = fpext bfloat %75 to float, !dbg !28 + %135 = fpext bfloat %76 to float, !dbg !28 + %136 = fpext bfloat %77 to float, !dbg !28 + %137 = fpext bfloat %78 to float, !dbg !28 + %138 = fpext bfloat %79 to float, !dbg !28 + %139 = fpext bfloat %80 to float, !dbg !28 + %140 = fpext bfloat %81 to float, !dbg !28 + %141 = fpext bfloat %82 to float, !dbg !28 + %142 = fpext bfloat %92 to float, !dbg !28 + %143 = fpext bfloat %93 to float, !dbg !28 + %144 = fpext bfloat %94 to float, !dbg !28 + %145 = fpext bfloat %95 to float, !dbg !28 + %146 = fpext bfloat %96 to float, !dbg !28 + %147 = fpext bfloat %97 to float, !dbg !28 + %148 = fpext bfloat %98 to float, !dbg !28 + %149 = fpext bfloat %99 to float, !dbg !28 + %150 = fpext bfloat %109 to float, !dbg !28 + %151 = fpext bfloat %110 to float, !dbg !28 + %152 = fpext bfloat %111 to float, !dbg !28 + %153 = fpext bfloat %112 to float, !dbg !28 + %154 = fpext bfloat %113 to float, !dbg !28 + %155 = fpext bfloat %114 to float, !dbg !28 + %156 = fpext bfloat %115 to float, !dbg !28 + %157 = fpext bfloat %116 to float, !dbg !28 + %158 = fpext bfloat %126 to float, !dbg !28 + %159 = fpext bfloat %127 to float, !dbg !28 + %160 = fpext bfloat %128 to float, !dbg !28 + %161 = fpext bfloat %129 to float, !dbg !28 + %162 = fpext bfloat %130 to float, !dbg !28 + %163 = fpext bfloat %131 to float, !dbg !28 + %164 = fpext bfloat %132 to float, !dbg !28 + %165 = fpext bfloat %133 to float, !dbg !28 + %166 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %46, i1 %27) #3, !dbg !29 + %167 = bitcast i32 %166 to float, !dbg !29 + %168 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %47, i1 %28) #3, !dbg !29 + %169 = bitcast i32 %168 to float, !dbg !29 + %170 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %48, i1 %29) #3, !dbg !29 + %171 = bitcast i32 %170 to float, !dbg !29 + %172 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %49, i1 %30) #3, !dbg !29 + %173 = bitcast i32 %172 to float, !dbg !29 + %174 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %54, i1 %27) #3, !dbg !30 + %175 = extractvalue { i32, i32, i32, i32 } %174, 0, !dbg !30 + %176 = bitcast i32 %175 to <2 x bfloat>, !dbg !30 + %177 = extractvalue { i32, i32, i32, i32 } %174, 1, !dbg !30 + %178 = bitcast i32 %177 to <2 x bfloat>, !dbg !30 + %179 = extractvalue { i32, i32, i32, i32 } %174, 2, !dbg !30 + %180 = bitcast i32 %179 to <2 x bfloat>, !dbg !30 + %181 = extractvalue { i32, i32, i32, i32 } %174, 3, !dbg !30 + %182 = bitcast i32 %181 to <2 x bfloat>, !dbg !30 + %183 = extractelement <2 x bfloat> %176, i64 0, !dbg !30 + %184 = extractelement <2 x bfloat> %176, i64 1, !dbg !30 + %185 = extractelement <2 x bfloat> %178, i64 0, !dbg !30 + %186 = extractelement <2 x bfloat> %178, i64 1, !dbg !30 + %187 = extractelement <2 x bfloat> %180, i64 0, !dbg !30 + %188 = extractelement <2 x bfloat> %180, i64 1, !dbg !30 + %189 = extractelement <2 x bfloat> %182, i64 0, !dbg !30 + %190 = extractelement <2 x bfloat> %182, i64 1, !dbg !30 + %191 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %55, i1 %28) #3, !dbg !30 + %192 = extractvalue { i32, i32, i32, i32 } %191, 0, !dbg !30 + %193 = bitcast i32 %192 to <2 x bfloat>, !dbg !30 + %194 = extractvalue { i32, i32, i32, i32 } %191, 1, !dbg !30 + %195 = bitcast i32 %194 to <2 x bfloat>, !dbg !30 + %196 = extractvalue { i32, i32, i32, i32 } %191, 2, !dbg !30 + %197 = bitcast i32 %196 to <2 x bfloat>, !dbg !30 + %198 = extractvalue { i32, i32, i32, i32 } %191, 3, !dbg !30 + %199 = bitcast i32 %198 to <2 x bfloat>, !dbg !30 + %200 = extractelement <2 x bfloat> %193, i64 0, !dbg !30 + %201 = extractelement <2 x bfloat> %193, i64 1, !dbg !30 + %202 = extractelement <2 x bfloat> %195, i64 0, !dbg !30 + %203 = extractelement <2 x bfloat> %195, i64 1, !dbg !30 + %204 = extractelement <2 x bfloat> %197, i64 0, !dbg !30 + %205 = extractelement <2 x bfloat> %197, i64 1, !dbg !30 + %206 = extractelement <2 x bfloat> %199, i64 0, !dbg !30 + %207 = extractelement <2 x bfloat> %199, i64 1, !dbg !30 + %208 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %56, i1 %29) #3, !dbg !30 + %209 = extractvalue { i32, i32, i32, i32 } %208, 0, !dbg !30 + %210 = bitcast i32 %209 to <2 x bfloat>, !dbg !30 + %211 = extractvalue { i32, i32, i32, i32 } %208, 1, !dbg !30 + %212 = bitcast i32 %211 to <2 x bfloat>, !dbg !30 + %213 = extractvalue { i32, i32, i32, i32 } %208, 2, !dbg !30 + %214 = bitcast i32 %213 to <2 x bfloat>, !dbg !30 + %215 = extractvalue { i32, i32, i32, i32 } %208, 3, !dbg !30 + %216 = bitcast i32 %215 to <2 x bfloat>, !dbg !30 + %217 = extractelement <2 x bfloat> %210, i64 0, !dbg !30 + %218 = extractelement <2 x bfloat> %210, i64 1, !dbg !30 + %219 = extractelement <2 x bfloat> %212, i64 0, !dbg !30 + %220 = extractelement <2 x bfloat> %212, i64 1, !dbg !30 + %221 = extractelement <2 x bfloat> %214, i64 0, !dbg !30 + %222 = extractelement <2 x bfloat> %214, i64 1, !dbg !30 + %223 = extractelement <2 x bfloat> %216, i64 0, !dbg !30 + %224 = extractelement <2 x bfloat> %216, i64 1, !dbg !30 + %225 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %57, i1 %30) #3, !dbg !30 + %226 = extractvalue { i32, i32, i32, i32 } %225, 0, !dbg !30 + %227 = bitcast i32 %226 to <2 x bfloat>, !dbg !30 + %228 = extractvalue { i32, i32, i32, i32 } %225, 1, !dbg !30 + %229 = bitcast i32 %228 to <2 x bfloat>, !dbg !30 + %230 = extractvalue { i32, i32, i32, i32 } %225, 2, !dbg !30 + %231 = bitcast i32 %230 to <2 x bfloat>, !dbg !30 + %232 = extractvalue { i32, i32, i32, i32 } %225, 3, !dbg !30 + %233 = bitcast i32 %232 to <2 x bfloat>, !dbg !30 + %234 = extractelement <2 x bfloat> %227, i64 0, !dbg !30 + %235 = extractelement <2 x bfloat> %227, i64 1, !dbg !30 + %236 = extractelement <2 x bfloat> %229, i64 0, !dbg !30 + %237 = extractelement <2 x bfloat> %229, i64 1, !dbg !30 + %238 = extractelement <2 x bfloat> %231, i64 0, !dbg !30 + %239 = extractelement <2 x bfloat> %231, i64 1, !dbg !30 + %240 = extractelement <2 x bfloat> %233, i64 0, !dbg !30 + %241 = extractelement <2 x bfloat> %233, i64 1, !dbg !30 + %242 = fpext bfloat %183 to float, !dbg !31 + %243 = fpext bfloat %184 to float, !dbg !31 + %244 = fpext bfloat %185 to float, !dbg !31 + %245 = fpext bfloat %186 to float, !dbg !31 + %246 = fpext bfloat %187 to float, !dbg !31 + %247 = fpext bfloat %188 to float, !dbg !31 + %248 = fpext bfloat %189 to float, !dbg !31 + %249 = fpext bfloat %190 to float, !dbg !31 + %250 = fpext bfloat %200 to float, !dbg !31 + %251 = fpext bfloat %201 to float, !dbg !31 + %252 = fpext bfloat %202 to float, !dbg !31 + %253 = fpext bfloat %203 to float, !dbg !31 + %254 = fpext bfloat %204 to float, !dbg !31 + %255 = fpext bfloat %205 to float, !dbg !31 + %256 = fpext bfloat %206 to float, !dbg !31 + %257 = fpext bfloat %207 to float, !dbg !31 + %258 = fpext bfloat %217 to float, !dbg !31 + %259 = fpext bfloat %218 to float, !dbg !31 + %260 = fpext bfloat %219 to float, !dbg !31 + %261 = fpext bfloat %220 to float, !dbg !31 + %262 = fpext bfloat %221 to float, !dbg !31 + %263 = fpext bfloat %222 to float, !dbg !31 + %264 = fpext bfloat %223 to float, !dbg !31 + %265 = fpext bfloat %224 to float, !dbg !31 + %266 = fpext bfloat %234 to float, !dbg !31 + %267 = fpext bfloat %235 to float, !dbg !31 + %268 = fpext bfloat %236 to float, !dbg !31 + %269 = fpext bfloat %237 to float, !dbg !31 + %270 = fpext bfloat %238 to float, !dbg !31 + %271 = fpext bfloat %239 to float, !dbg !31 + %272 = fpext bfloat %240 to float, !dbg !31 + %273 = fpext bfloat %241 to float, !dbg !31 + %274 = fmul float %167, %242, !dbg !32 + %275 = fmul float %167, %243, !dbg !32 + %276 = fmul float %167, %244, !dbg !32 + %277 = fmul float %167, %245, !dbg !32 + %278 = fmul float %167, %246, !dbg !32 + %279 = fmul float %167, %247, !dbg !32 + %280 = fmul float %167, %248, !dbg !32 + %281 = fmul float %167, %249, !dbg !32 + %282 = fmul float %169, %250, !dbg !32 + %283 = fmul float %169, %251, !dbg !32 + %284 = fmul float %169, %252, !dbg !32 + %285 = fmul float %169, %253, !dbg !32 + %286 = fmul float %169, %254, !dbg !32 + %287 = fmul float %169, %255, !dbg !32 + %288 = fmul float %169, %256, !dbg !32 + %289 = fmul float %169, %257, !dbg !32 + %290 = fmul float %171, %258, !dbg !32 + %291 = fmul float %171, %259, !dbg !32 + %292 = fmul float %171, %260, !dbg !32 + %293 = fmul float %171, %261, !dbg !32 + %294 = fmul float %171, %262, !dbg !32 + %295 = fmul float %171, %263, !dbg !32 + %296 = fmul float %171, %264, !dbg !32 + %297 = fmul float %171, %265, !dbg !32 + %298 = fmul float %173, %266, !dbg !32 + %299 = fmul float %173, %267, !dbg !32 + %300 = fmul float %173, %268, !dbg !32 + %301 = fmul float %173, %269, !dbg !32 + %302 = fmul float %173, %270, !dbg !32 + %303 = fmul float %173, %271, !dbg !32 + %304 = fmul float %173, %272, !dbg !32 + %305 = fmul float %173, %273, !dbg !32 + %306 = fmul float %134, %242, !dbg !33 + %307 = fmul float %135, %243, !dbg !33 + %308 = fmul float %136, %244, !dbg !33 + %309 = fmul float %137, %245, !dbg !33 + %310 = fmul float %138, %246, !dbg !33 + %311 = fmul float %139, %247, !dbg !33 + %312 = fmul float %140, %248, !dbg !33 + %313 = fmul float %141, %249, !dbg !33 + %314 = fmul float %142, %250, !dbg !33 + %315 = fmul float %143, %251, !dbg !33 + %316 = fmul float %144, %252, !dbg !33 + %317 = fmul float %145, %253, !dbg !33 + %318 = fmul float %146, %254, !dbg !33 + %319 = fmul float %147, %255, !dbg !33 + %320 = fmul float %148, %256, !dbg !33 + %321 = fmul float %149, %257, !dbg !33 + %322 = fmul float %150, %258, !dbg !33 + %323 = fmul float %151, %259, !dbg !33 + %324 = fmul float %152, %260, !dbg !33 + %325 = fmul float %153, %261, !dbg !33 + %326 = fmul float %154, %262, !dbg !33 + %327 = fmul float %155, %263, !dbg !33 + %328 = fmul float %156, %264, !dbg !33 + %329 = fmul float %157, %265, !dbg !33 + %330 = fmul float %158, %266, !dbg !33 + %331 = fmul float %159, %267, !dbg !33 + %332 = fmul float %160, %268, !dbg !33 + %333 = fmul float %161, %269, !dbg !33 + %334 = fmul float %162, %270, !dbg !33 + %335 = fmul float %163, %271, !dbg !33 + %336 = fmul float %164, %272, !dbg !33 + %337 = fmul float %165, %273, !dbg !33 + %338 = fadd float %306, %307, !dbg !34 + %339 = fadd float %308, %338, !dbg !34 + %340 = fadd float %309, %339, !dbg !34 + %341 = fadd float %310, %340, !dbg !34 + %342 = fadd float %311, %341, !dbg !34 + %343 = fadd float %312, %342, !dbg !34 + %344 = fadd float %313, %343, !dbg !34 + %345 = fadd float %314, %315, !dbg !34 + %346 = fadd float %316, %345, !dbg !34 + %347 = fadd float %317, %346, !dbg !34 + %348 = fadd float %318, %347, !dbg !34 + %349 = fadd float %319, %348, !dbg !34 + %350 = fadd float %320, %349, !dbg !34 + %351 = fadd float %321, %350, !dbg !34 + %352 = fadd float %322, %323, !dbg !34 + %353 = fadd float %324, %352, !dbg !34 + %354 = fadd float %325, %353, !dbg !34 + %355 = fadd float %326, %354, !dbg !34 + %356 = fadd float %327, %355, !dbg !34 + %357 = fadd float %328, %356, !dbg !34 + %358 = fadd float %329, %357, !dbg !34 + %359 = fadd float %330, %331, !dbg !34 + %360 = fadd float %332, %359, !dbg !34 + %361 = fadd float %333, %360, !dbg !34 + %362 = fadd float %334, %361, !dbg !34 + %363 = fadd float %335, %362, !dbg !34 + %364 = fadd float %336, %363, !dbg !34 + %365 = fadd float %337, %364, !dbg !34 + %366 = bitcast float %344 to i32, !dbg !37 + %367 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %366, i32 8, i32 31), !dbg !37 + %368 = bitcast i32 %367 to float, !dbg !37 + %369 = fadd float %344, %368, !dbg !34 + %370 = bitcast float %369 to i32, !dbg !37 + %371 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %370, i32 4, i32 31), !dbg !37 + %372 = bitcast i32 %371 to float, !dbg !37 + %373 = fadd float %369, %372, !dbg !34 + %374 = bitcast float %373 to i32, !dbg !37 + %375 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %374, i32 2, i32 31), !dbg !37 + %376 = bitcast i32 %375 to float, !dbg !37 + %377 = fadd float %373, %376, !dbg !34 + %378 = bitcast float %377 to i32, !dbg !37 + %379 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %378, i32 1, i32 31), !dbg !37 + %380 = bitcast i32 %379 to float, !dbg !37 + %381 = fadd float %377, %380, !dbg !34 + %382 = bitcast float %351 to i32, !dbg !37 + %383 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %382, i32 8, i32 31), !dbg !37 + %384 = bitcast i32 %383 to float, !dbg !37 + %385 = fadd float %351, %384, !dbg !34 + %386 = bitcast float %385 to i32, !dbg !37 + %387 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %386, i32 4, i32 31), !dbg !37 + %388 = bitcast i32 %387 to float, !dbg !37 + %389 = fadd float %385, %388, !dbg !34 + %390 = bitcast float %389 to i32, !dbg !37 + %391 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %390, i32 2, i32 31), !dbg !37 + %392 = bitcast i32 %391 to float, !dbg !37 + %393 = fadd float %389, %392, !dbg !34 + %394 = bitcast float %393 to i32, !dbg !37 + %395 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %394, i32 1, i32 31), !dbg !37 + %396 = bitcast i32 %395 to float, !dbg !37 + %397 = fadd float %393, %396, !dbg !34 + %398 = bitcast float %358 to i32, !dbg !37 + %399 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %398, i32 8, i32 31), !dbg !37 + %400 = bitcast i32 %399 to float, !dbg !37 + %401 = fadd float %358, %400, !dbg !34 + %402 = bitcast float %401 to i32, !dbg !37 + %403 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %402, i32 4, i32 31), !dbg !37 + %404 = bitcast i32 %403 to float, !dbg !37 + %405 = fadd float %401, %404, !dbg !34 + %406 = bitcast float %405 to i32, !dbg !37 + %407 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %406, i32 2, i32 31), !dbg !37 + %408 = bitcast i32 %407 to float, !dbg !37 + %409 = fadd float %405, %408, !dbg !34 + %410 = bitcast float %409 to i32, !dbg !37 + %411 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %410, i32 1, i32 31), !dbg !37 + %412 = bitcast i32 %411 to float, !dbg !37 + %413 = fadd float %409, %412, !dbg !34 + %414 = bitcast float %365 to i32, !dbg !37 + %415 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %414, i32 8, i32 31), !dbg !37 + %416 = bitcast i32 %415 to float, !dbg !37 + %417 = fadd float %365, %416, !dbg !34 + %418 = bitcast float %417 to i32, !dbg !37 + %419 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %418, i32 4, i32 31), !dbg !37 + %420 = bitcast i32 %419 to float, !dbg !37 + %421 = fadd float %417, %420, !dbg !34 + %422 = bitcast float %421 to i32, !dbg !37 + %423 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %422, i32 2, i32 31), !dbg !37 + %424 = bitcast i32 %423 to float, !dbg !37 + %425 = fadd float %421, %424, !dbg !34 + %426 = bitcast float %425 to i32, !dbg !37 + %427 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %426, i32 1, i32 31), !dbg !37 + %428 = bitcast i32 %427 to float, !dbg !37 + %429 = fadd float %425, %428, !dbg !34 + %430 = fmul float %381, %134, !dbg !40 + %431 = fmul float %381, %135, !dbg !40 + %432 = fmul float %381, %136, !dbg !40 + %433 = fmul float %381, %137, !dbg !40 + %434 = fmul float %381, %138, !dbg !40 + %435 = fmul float %381, %139, !dbg !40 + %436 = fmul float %381, %140, !dbg !40 + %437 = fmul float %381, %141, !dbg !40 + %438 = fmul float %397, %142, !dbg !40 + %439 = fmul float %397, %143, !dbg !40 + %440 = fmul float %397, %144, !dbg !40 + %441 = fmul float %397, %145, !dbg !40 + %442 = fmul float %397, %146, !dbg !40 + %443 = fmul float %397, %147, !dbg !40 + %444 = fmul float %397, %148, !dbg !40 + %445 = fmul float %397, %149, !dbg !40 + %446 = fmul float %413, %150, !dbg !40 + %447 = fmul float %413, %151, !dbg !40 + %448 = fmul float %413, %152, !dbg !40 + %449 = fmul float %413, %153, !dbg !40 + %450 = fmul float %413, %154, !dbg !40 + %451 = fmul float %413, %155, !dbg !40 + %452 = fmul float %413, %156, !dbg !40 + %453 = fmul float %413, %157, !dbg !40 + %454 = fmul float %429, %158, !dbg !40 + %455 = fmul float %429, %159, !dbg !40 + %456 = fmul float %429, %160, !dbg !40 + %457 = fmul float %429, %161, !dbg !40 + %458 = fmul float %429, %162, !dbg !40 + %459 = fmul float %429, %163, !dbg !40 + %460 = fmul float %429, %164, !dbg !40 + %461 = fmul float %429, %165, !dbg !40 + %462 = fmul float %430, %167, !dbg !41 + %463 = fmul float %431, %167, !dbg !41 + %464 = fmul float %432, %167, !dbg !41 + %465 = fmul float %433, %167, !dbg !41 + %466 = fmul float %434, %167, !dbg !41 + %467 = fmul float %435, %167, !dbg !41 + %468 = fmul float %436, %167, !dbg !41 + %469 = fmul float %437, %167, !dbg !41 + %470 = fmul float %438, %169, !dbg !41 + %471 = fmul float %439, %169, !dbg !41 + %472 = fmul float %440, %169, !dbg !41 + %473 = fmul float %441, %169, !dbg !41 + %474 = fmul float %442, %169, !dbg !41 + %475 = fmul float %443, %169, !dbg !41 + %476 = fmul float %444, %169, !dbg !41 + %477 = fmul float %445, %169, !dbg !41 + %478 = fmul float %446, %171, !dbg !41 + %479 = fmul float %447, %171, !dbg !41 + %480 = fmul float %448, %171, !dbg !41 + %481 = fmul float %449, %171, !dbg !41 + %482 = fmul float %450, %171, !dbg !41 + %483 = fmul float %451, %171, !dbg !41 + %484 = fmul float %452, %171, !dbg !41 + %485 = fmul float %453, %171, !dbg !41 + %486 = fmul float %454, %173, !dbg !41 + %487 = fmul float %455, %173, !dbg !41 + %488 = fmul float %456, %173, !dbg !41 + %489 = fmul float %457, %173, !dbg !41 + %490 = fmul float %458, %173, !dbg !41 + %491 = fmul float %459, %173, !dbg !41 + %492 = fmul float %460, %173, !dbg !41 + %493 = fmul float %461, %173, !dbg !41 + %494 = fsub float %274, %462, !dbg !42 + %495 = fsub float %275, %463, !dbg !42 + %496 = fsub float %276, %464, !dbg !42 + %497 = fsub float %277, %465, !dbg !42 + %498 = fsub float %278, %466, !dbg !42 + %499 = fsub float %279, %467, !dbg !42 + %500 = fsub float %280, %468, !dbg !42 + %501 = fsub float %281, %469, !dbg !42 + %502 = fsub float %282, %470, !dbg !42 + %503 = fsub float %283, %471, !dbg !42 + %504 = fsub float %284, %472, !dbg !42 + %505 = fsub float %285, %473, !dbg !42 + %506 = fsub float %286, %474, !dbg !42 + %507 = fsub float %287, %475, !dbg !42 + %508 = fsub float %288, %476, !dbg !42 + %509 = fsub float %289, %477, !dbg !42 + %510 = fsub float %290, %478, !dbg !42 + %511 = fsub float %291, %479, !dbg !42 + %512 = fsub float %292, %480, !dbg !42 + %513 = fsub float %293, %481, !dbg !42 + %514 = fsub float %294, %482, !dbg !42 + %515 = fsub float %295, %483, !dbg !42 + %516 = fsub float %296, %484, !dbg !42 + %517 = fsub float %297, %485, !dbg !42 + %518 = fsub float %298, %486, !dbg !42 + %519 = fsub float %299, %487, !dbg !42 + %520 = fsub float %300, %488, !dbg !42 + %521 = fsub float %301, %489, !dbg !42 + %522 = fsub float %302, %490, !dbg !42 + %523 = fsub float %303, %491, !dbg !42 + %524 = fsub float %304, %492, !dbg !42 + %525 = fsub float %305, %493, !dbg !42 + %526 = fptrunc float %494 to bfloat, !dbg !43 + %527 = fptrunc float %495 to bfloat, !dbg !43 + %528 = fptrunc float %496 to bfloat, !dbg !43 + %529 = fptrunc float %497 to bfloat, !dbg !43 + %530 = fptrunc float %498 to bfloat, !dbg !43 + %531 = fptrunc float %499 to bfloat, !dbg !43 + %532 = fptrunc float %500 to bfloat, !dbg !43 + %533 = fptrunc float %501 to bfloat, !dbg !43 + %534 = fptrunc float %502 to bfloat, !dbg !43 + %535 = fptrunc float %503 to bfloat, !dbg !43 + %536 = fptrunc float %504 to bfloat, !dbg !43 + %537 = fptrunc float %505 to bfloat, !dbg !43 + %538 = fptrunc float %506 to bfloat, !dbg !43 + %539 = fptrunc float %507 to bfloat, !dbg !43 + %540 = fptrunc float %508 to bfloat, !dbg !43 + %541 = fptrunc float %509 to bfloat, !dbg !43 + %542 = fptrunc float %510 to bfloat, !dbg !43 + %543 = fptrunc float %511 to bfloat, !dbg !43 + %544 = fptrunc float %512 to bfloat, !dbg !43 + %545 = fptrunc float %513 to bfloat, !dbg !43 + %546 = fptrunc float %514 to bfloat, !dbg !43 + %547 = fptrunc float %515 to bfloat, !dbg !43 + %548 = fptrunc float %516 to bfloat, !dbg !43 + %549 = fptrunc float %517 to bfloat, !dbg !43 + %550 = fptrunc float %518 to bfloat, !dbg !43 + %551 = fptrunc float %519 to bfloat, !dbg !43 + %552 = fptrunc float %520 to bfloat, !dbg !43 + %553 = fptrunc float %521 to bfloat, !dbg !43 + %554 = fptrunc float %522 to bfloat, !dbg !43 + %555 = fptrunc float %523 to bfloat, !dbg !43 + %556 = fptrunc float %524 to bfloat, !dbg !43 + %557 = fptrunc float %525 to bfloat, !dbg !43 + %558 = insertelement <2 x bfloat> poison, bfloat %526, i64 0, !dbg !44 + %559 = insertelement <2 x bfloat> %558, bfloat %527, i64 1, !dbg !44 + %560 = bitcast <2 x bfloat> %559 to i32, !dbg !44 + %561 = insertelement <2 x bfloat> poison, bfloat %528, i64 0, !dbg !44 + %562 = insertelement <2 x bfloat> %561, bfloat %529, i64 1, !dbg !44 + %563 = bitcast <2 x bfloat> %562 to i32, !dbg !44 + %564 = insertelement <2 x bfloat> poison, bfloat %530, i64 0, !dbg !44 + %565 = insertelement <2 x bfloat> %564, bfloat %531, i64 1, !dbg !44 + %566 = bitcast <2 x bfloat> %565 to i32, !dbg !44 + %567 = insertelement <2 x bfloat> poison, bfloat %532, i64 0, !dbg !44 + %568 = insertelement <2 x bfloat> %567, bfloat %533, i64 1, !dbg !44 + %569 = bitcast <2 x bfloat> %568 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %560, i32 %563, i32 %566, i32 %569, ptr addrspace(1) %62, i1 %27) #3, !dbg !44 + %570 = insertelement <2 x bfloat> poison, bfloat %534, i64 0, !dbg !44 + %571 = insertelement <2 x bfloat> %570, bfloat %535, i64 1, !dbg !44 + %572 = bitcast <2 x bfloat> %571 to i32, !dbg !44 + %573 = insertelement <2 x bfloat> poison, bfloat %536, i64 0, !dbg !44 + %574 = insertelement <2 x bfloat> %573, bfloat %537, i64 1, !dbg !44 + %575 = bitcast <2 x bfloat> %574 to i32, !dbg !44 + %576 = insertelement <2 x bfloat> poison, bfloat %538, i64 0, !dbg !44 + %577 = insertelement <2 x bfloat> %576, bfloat %539, i64 1, !dbg !44 + %578 = bitcast <2 x bfloat> %577 to i32, !dbg !44 + %579 = insertelement <2 x bfloat> poison, bfloat %540, i64 0, !dbg !44 + %580 = insertelement <2 x bfloat> %579, bfloat %541, i64 1, !dbg !44 + %581 = bitcast <2 x bfloat> %580 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %572, i32 %575, i32 %578, i32 %581, ptr addrspace(1) %63, i1 %28) #3, !dbg !44 + %582 = insertelement <2 x bfloat> poison, bfloat %542, i64 0, !dbg !44 + %583 = insertelement <2 x bfloat> %582, bfloat %543, i64 1, !dbg !44 + %584 = bitcast <2 x bfloat> %583 to i32, !dbg !44 + %585 = insertelement <2 x bfloat> poison, bfloat %544, i64 0, !dbg !44 + %586 = insertelement <2 x bfloat> %585, bfloat %545, i64 1, !dbg !44 + %587 = bitcast <2 x bfloat> %586 to i32, !dbg !44 + %588 = insertelement <2 x bfloat> poison, bfloat %546, i64 0, !dbg !44 + %589 = insertelement <2 x bfloat> %588, bfloat %547, i64 1, !dbg !44 + %590 = bitcast <2 x bfloat> %589 to i32, !dbg !44 + %591 = insertelement <2 x bfloat> poison, bfloat %548, i64 0, !dbg !44 + %592 = insertelement <2 x bfloat> %591, bfloat %549, i64 1, !dbg !44 + %593 = bitcast <2 x bfloat> %592 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %584, i32 %587, i32 %590, i32 %593, ptr addrspace(1) %64, i1 %29) #3, !dbg !44 + %594 = insertelement <2 x bfloat> poison, bfloat %550, i64 0, !dbg !44 + %595 = insertelement <2 x bfloat> %594, bfloat %551, i64 1, !dbg !44 + %596 = bitcast <2 x bfloat> %595 to i32, !dbg !44 + %597 = insertelement <2 x bfloat> poison, bfloat %552, i64 0, !dbg !44 + %598 = insertelement <2 x bfloat> %597, bfloat %553, i64 1, !dbg !44 + %599 = bitcast <2 x bfloat> %598 to i32, !dbg !44 + %600 = insertelement <2 x bfloat> poison, bfloat %554, i64 0, !dbg !44 + %601 = insertelement <2 x bfloat> %600, bfloat %555, i64 1, !dbg !44 + %602 = bitcast <2 x bfloat> %601 to i32, !dbg !44 + %603 = insertelement <2 x bfloat> poison, bfloat %556, i64 0, !dbg !44 + %604 = insertelement <2 x bfloat> %603, bfloat %557, i64 1, !dbg !44 + %605 = bitcast <2 x bfloat> %604 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %596, i32 %599, i32 %602, i32 %605, ptr addrspace(1) %65, i1 %30) #3, !dbg !44 + ret void, !dbg !45 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +attributes #0 = { nounwind "nvvm.reqntid"="64" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = distinct !DISubprogram(name: "l2norm_bwd_kernel", linkageName: "l2norm_bwd_kernel", scope: !1, file: !1, line: 93, type: !5, scopeLine: 93, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!5 = !DISubroutineType(cc: DW_CC_normal, types: !6) +!6 = !{null, !7, !9, !7, !7, !10, !11, !7, !7} +!7 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !8, size: 64, dwarfAddressSpace: 1) +!8 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!9 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !10, size: 64, dwarfAddressSpace: 1) +!10 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!11 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!12 = !DILocation(line: 94, column: 24, scope: !4) +!13 = !DILocation(line: 94, column: 30, scope: !4) +!14 = !DILocation(line: 95, column: 16, scope: !4) +!15 = !DILocation(line: 95, column: 34, scope: !4) +!16 = !DILocation(line: 95, column: 21, scope: !4) +!17 = !DILocation(line: 97, column: 16, scope: !4) +!18 = !DILocation(line: 98, column: 30, scope: !4) +!19 = !DILocation(line: 99, column: 29, scope: !4) +!20 = !DILocation(line: 99, column: 14, scope: !4) +!21 = !DILocation(line: 99, column: 33, scope: !4) +!22 = !DILocation(line: 100, column: 20, scope: !4) +!23 = !DILocation(line: 101, column: 16, scope: !4) +!24 = !DILocation(line: 101, column: 35, scope: !4) +!25 = !DILocation(line: 102, column: 16, scope: !4) +!26 = !DILocation(line: 102, column: 35, scope: !4) +!27 = !DILocation(line: 104, column: 18, scope: !4) +!28 = !DILocation(line: 104, column: 47, scope: !4) +!29 = !DILocation(line: 105, column: 21, scope: !4) +!30 = !DILocation(line: 106, column: 19, scope: !4) +!31 = !DILocation(line: 106, column: 49, scope: !4) +!32 = !DILocation(line: 107, column: 18, scope: !4) +!33 = !DILocation(line: 107, column: 50, scope: !4) +!34 = !DILocation(line: 263, column: 15, scope: !35, inlinedAt: !37) +!35 = distinct !DILexicalBlockFile(scope: !4, file: !36, discriminator: 0) +!36 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!37 = !DILocation(line: 293, column: 36, scope: !35, inlinedAt: !38) +!38 = !DILocation(line: 107, column: 55, scope: !39) +!39 = distinct !DILexicalBlockFile(scope: !4, file: !1, discriminator: 0) +!40 = !DILocation(line: 107, column: 69, scope: !4) +!41 = !DILocation(line: 107, column: 75, scope: !4) +!42 = !DILocation(line: 107, column: 36, scope: !4) +!43 = !DILocation(line: 108, column: 27, scope: !4) +!44 = !DILocation(line: 108, column: 19, scope: !4) +!45 = !DILocation(line: 108, column: 4, scope: !4) diff --git a/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/l2norm_bwd_kernel.ptx b/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/l2norm_bwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..3b4f28ad51afee3d5218430b04f95c6a373a3cf6 --- /dev/null +++ b/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/l2norm_bwd_kernel.ptx @@ -0,0 +1,796 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_bwd_kernel // -- Begin function l2norm_bwd_kernel + // @l2norm_bwd_kernel +.visible .entry l2norm_bwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_2, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_3, + .param .f32 l2norm_bwd_kernel_param_4, + .param .u32 l2norm_bwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_6, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_7 +) +.reqntid 64 +{ + .reg .pred %p<5>; + .reg .b16 %rs<65>; + .reg .b32 %r<318>; + .reg .b64 %rd<49>; + .loc 1 93 0 // l2norm.py:93:0 +$L__func_begin0: + .loc 1 93 0 // l2norm.py:93:0 + +// %bb.0: + ld.param.b64 %rd17, [l2norm_bwd_kernel_param_0]; + ld.param.b64 %rd18, [l2norm_bwd_kernel_param_1]; +$L__tmp0: + .loc 1 94 24 // l2norm.py:94:24 + mov.u32 %r54, %ctaid.x; + ld.param.b64 %rd19, [l2norm_bwd_kernel_param_2]; + .loc 1 95 16 // l2norm.py:95:16 + mul.wide.u32 %rd20, %r54, 16; + ld.param.b64 %rd21, [l2norm_bwd_kernel_param_3]; + .loc 1 95 34 // l2norm.py:95:34 + mov.u32 %r55, %tid.x; + ld.param.s32 %rd22, [l2norm_bwd_kernel_param_5]; + bfe.u32 %r56, %r55, 4, 2; + or.b32 %r57, %r56, 4; + or.b32 %r58, %r56, 8; + or.b32 %r59, %r56, 12; + .loc 1 95 21 // l2norm.py:95:21 + cvt.u64.u32 %rd23, %r56; + cvt.u64.u32 %rd24, %r57; + cvt.u64.u32 %rd25, %r58; + cvt.u64.u32 %rd26, %r59; + or.b64 %rd27, %rd20, %rd23; + or.b64 %rd28, %rd20, %rd24; + or.b64 %rd29, %rd20, %rd25; + or.b64 %rd30, %rd20, %rd26; + .loc 1 97 16 // l2norm.py:97:16 + setp.lt.s64 %p1, %rd27, %rd22; + setp.lt.s64 %p2, %rd28, %rd22; + setp.lt.s64 %p3, %rd29, %rd22; + setp.lt.s64 %p4, %rd30, %rd22; + .loc 1 98 30 // l2norm.py:98:30 + shl.b32 %r60, %r55, 3; + and.b32 %r61, %r60, 120; + .loc 1 99 14 // l2norm.py:99:14 + shl.b64 %rd31, %rd27, 8; + add.s64 %rd32, %rd17, %rd31; + shl.b64 %rd33, %rd28, 8; + add.s64 %rd34, %rd17, %rd33; + shl.b64 %rd35, %rd29, 8; + add.s64 %rd36, %rd17, %rd35; + shl.b64 %rd37, %rd30, 8; + add.s64 %rd38, %rd17, %rd37; + .loc 1 99 33 // l2norm.py:99:33 + mul.wide.u32 %rd39, %r61, 2; + add.s64 %rd1, %rd32, %rd39; + add.s64 %rd2, %rd34, %rd39; + add.s64 %rd3, %rd36, %rd39; + add.s64 %rd4, %rd38, %rd39; + .loc 1 100 20 // l2norm.py:100:20 + shl.b64 %rd40, %rd27, 2; + add.s64 %rd5, %rd18, %rd40; + add.s64 %rd6, %rd5, 16; + add.s64 %rd7, %rd5, 32; + add.s64 %rd8, %rd5, 48; + .loc 1 101 16 // l2norm.py:101:16 + add.s64 %rd41, %rd19, %rd31; + add.s64 %rd42, %rd19, %rd33; + add.s64 %rd43, %rd19, %rd35; + add.s64 %rd44, %rd19, %rd37; + .loc 1 101 35 // l2norm.py:101:35 + add.s64 %rd9, %rd41, %rd39; + add.s64 %rd10, %rd42, %rd39; + add.s64 %rd11, %rd43, %rd39; + add.s64 %rd12, %rd44, %rd39; + .loc 1 102 16 // l2norm.py:102:16 + add.s64 %rd45, %rd21, %rd31; + add.s64 %rd46, %rd21, %rd33; + add.s64 %rd47, %rd21, %rd35; + add.s64 %rd48, %rd21, %rd37; + .loc 1 102 35 // l2norm.py:102:35 + add.s64 %rd13, %rd45, %rd39; + add.s64 %rd14, %rd46, %rd39; + add.s64 %rd15, %rd47, %rd39; + add.s64 %rd16, %rd48, %rd39; + mov.b32 %r5, 0; + .loc 1 104 18 // l2norm.py:104:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + mov.b32 {%rs5, %rs6}, %r3; + mov.b32 {%rs7, %rs8}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs9, %rs10}, %r6; + mov.b32 {%rs11, %rs12}, %r7; + mov.b32 {%rs13, %rs14}, %r8; + mov.b32 {%rs15, %rs16}, %r9; + // begin inline asm + mov.u32 %r10, %r5; + mov.u32 %r11, %r5; + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + @%p3 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs17, %rs18}, %r10; + mov.b32 {%rs19, %rs20}, %r11; + mov.b32 {%rs21, %rs22}, %r12; + mov.b32 {%rs23, %rs24}, %r13; + // begin inline asm + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + @%p4 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ]; + // end inline asm + mov.b32 {%rs25, %rs26}, %r14; + mov.b32 {%rs27, %rs28}, %r15; + mov.b32 {%rs29, %rs30}, %r16; + mov.b32 {%rs31, %rs32}, %r17; + .loc 1 104 47 // l2norm.py:104:47 + cvt.f32.bf16 %r62, %rs1; + cvt.f32.bf16 %r63, %rs2; + cvt.f32.bf16 %r64, %rs3; + cvt.f32.bf16 %r65, %rs4; + cvt.f32.bf16 %r66, %rs5; + cvt.f32.bf16 %r67, %rs6; + cvt.f32.bf16 %r68, %rs7; + cvt.f32.bf16 %r69, %rs8; + cvt.f32.bf16 %r70, %rs9; + cvt.f32.bf16 %r71, %rs10; + cvt.f32.bf16 %r72, %rs11; + cvt.f32.bf16 %r73, %rs12; + cvt.f32.bf16 %r74, %rs13; + cvt.f32.bf16 %r75, %rs14; + cvt.f32.bf16 %r76, %rs15; + cvt.f32.bf16 %r77, %rs16; + cvt.f32.bf16 %r78, %rs17; + cvt.f32.bf16 %r79, %rs18; + cvt.f32.bf16 %r80, %rs19; + cvt.f32.bf16 %r81, %rs20; + cvt.f32.bf16 %r82, %rs21; + cvt.f32.bf16 %r83, %rs22; + cvt.f32.bf16 %r84, %rs23; + cvt.f32.bf16 %r85, %rs24; + cvt.f32.bf16 %r86, %rs25; + cvt.f32.bf16 %r87, %rs26; + cvt.f32.bf16 %r88, %rs27; + cvt.f32.bf16 %r89, %rs28; + cvt.f32.bf16 %r90, %rs29; + cvt.f32.bf16 %r91, %rs30; + cvt.f32.bf16 %r92, %rs31; + cvt.f32.bf16 %r93, %rs32; + .loc 1 105 21 // l2norm.py:105:21 + // begin inline asm + mov.u32 %r18, %r5; + @%p1 ld.global.b32 { %r18 }, [ %rd5 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r19, %r5; + @%p2 ld.global.b32 { %r19 }, [ %rd6 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r20, %r5; + @%p3 ld.global.b32 { %r20 }, [ %rd7 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r21, %r5; + @%p4 ld.global.b32 { %r21 }, [ %rd8 + 0 ]; + // end inline asm + .loc 1 106 19 // l2norm.py:106:19 + // begin inline asm + mov.u32 %r22, %r5; + mov.u32 %r23, %r5; + mov.u32 %r24, %r5; + mov.u32 %r25, %r5; + @%p1 ld.global.v4.b32 { %r22, %r23, %r24, %r25 }, [ %rd9 + 0 ]; + // end inline asm + mov.b32 {%rs33, %rs34}, %r22; + mov.b32 {%rs35, %rs36}, %r23; + mov.b32 {%rs37, %rs38}, %r24; + mov.b32 {%rs39, %rs40}, %r25; + // begin inline asm + mov.u32 %r26, %r5; + mov.u32 %r27, %r5; + mov.u32 %r28, %r5; + mov.u32 %r29, %r5; + @%p2 ld.global.v4.b32 { %r26, %r27, %r28, %r29 }, [ %rd10 + 0 ]; + // end inline asm + mov.b32 {%rs41, %rs42}, %r26; + mov.b32 {%rs43, %rs44}, %r27; + mov.b32 {%rs45, %rs46}, %r28; + mov.b32 {%rs47, %rs48}, %r29; + // begin inline asm + mov.u32 %r30, %r5; + mov.u32 %r31, %r5; + mov.u32 %r32, %r5; + mov.u32 %r33, %r5; + @%p3 ld.global.v4.b32 { %r30, %r31, %r32, %r33 }, [ %rd11 + 0 ]; + // end inline asm + mov.b32 {%rs49, %rs50}, %r30; + mov.b32 {%rs51, %rs52}, %r31; + mov.b32 {%rs53, %rs54}, %r32; + mov.b32 {%rs55, %rs56}, %r33; + // begin inline asm + mov.u32 %r34, %r5; + mov.u32 %r35, %r5; + mov.u32 %r36, %r5; + mov.u32 %r37, %r5; + @%p4 ld.global.v4.b32 { %r34, %r35, %r36, %r37 }, [ %rd12 + 0 ]; + // end inline asm + mov.b32 {%rs57, %rs58}, %r34; + mov.b32 {%rs59, %rs60}, %r35; + mov.b32 {%rs61, %rs62}, %r36; + mov.b32 {%rs63, %rs64}, %r37; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r94, %rs33; + cvt.f32.bf16 %r95, %rs34; + cvt.f32.bf16 %r96, %rs35; + cvt.f32.bf16 %r97, %rs36; + cvt.f32.bf16 %r98, %rs37; + cvt.f32.bf16 %r99, %rs38; + cvt.f32.bf16 %r100, %rs39; + cvt.f32.bf16 %r101, %rs40; + cvt.f32.bf16 %r102, %rs41; + cvt.f32.bf16 %r103, %rs42; + cvt.f32.bf16 %r104, %rs43; + cvt.f32.bf16 %r105, %rs44; + cvt.f32.bf16 %r106, %rs45; + cvt.f32.bf16 %r107, %rs46; + cvt.f32.bf16 %r108, %rs47; + cvt.f32.bf16 %r109, %rs48; + cvt.f32.bf16 %r110, %rs49; + cvt.f32.bf16 %r111, %rs50; + cvt.f32.bf16 %r112, %rs51; + cvt.f32.bf16 %r113, %rs52; + cvt.f32.bf16 %r114, %rs53; + cvt.f32.bf16 %r115, %rs54; + cvt.f32.bf16 %r116, %rs55; + cvt.f32.bf16 %r117, %rs56; + cvt.f32.bf16 %r118, %rs57; + cvt.f32.bf16 %r119, %rs58; + cvt.f32.bf16 %r120, %rs59; + cvt.f32.bf16 %r121, %rs60; + cvt.f32.bf16 %r122, %rs61; + cvt.f32.bf16 %r123, %rs62; + cvt.f32.bf16 %r124, %rs63; + cvt.f32.bf16 %r125, %rs64; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r126, %r63, %r95; + mul.f32 %r127, %r71, %r103; + mul.f32 %r128, %r79, %r111; + mul.f32 %r129, %r87, %r119; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r130, %rs1, %rs33, %r126; + fma.rn.f32.bf16 %r131, %rs3, %rs35, %r130; + fma.rn.f32.bf16 %r132, %rs4, %rs36, %r131; + fma.rn.f32.bf16 %r133, %rs5, %rs37, %r132; + fma.rn.f32.bf16 %r134, %rs6, %rs38, %r133; + fma.rn.f32.bf16 %r135, %rs7, %rs39, %r134; + fma.rn.f32.bf16 %r136, %rs8, %rs40, %r135; + fma.rn.f32.bf16 %r137, %rs9, %rs41, %r127; + fma.rn.f32.bf16 %r138, %rs11, %rs43, %r137; + fma.rn.f32.bf16 %r139, %rs12, %rs44, %r138; + fma.rn.f32.bf16 %r140, %rs13, %rs45, %r139; + fma.rn.f32.bf16 %r141, %rs14, %rs46, %r140; + fma.rn.f32.bf16 %r142, %rs15, %rs47, %r141; + fma.rn.f32.bf16 %r143, %rs16, %rs48, %r142; + fma.rn.f32.bf16 %r144, %rs17, %rs49, %r128; + fma.rn.f32.bf16 %r145, %rs19, %rs51, %r144; + fma.rn.f32.bf16 %r146, %rs20, %rs52, %r145; + fma.rn.f32.bf16 %r147, %rs21, %rs53, %r146; + fma.rn.f32.bf16 %r148, %rs22, %rs54, %r147; + fma.rn.f32.bf16 %r149, %rs23, %rs55, %r148; + fma.rn.f32.bf16 %r150, %rs24, %rs56, %r149; + fma.rn.f32.bf16 %r151, %rs25, %rs57, %r129; + fma.rn.f32.bf16 %r152, %rs27, %rs59, %r151; + fma.rn.f32.bf16 %r153, %rs28, %rs60, %r152; + fma.rn.f32.bf16 %r154, %rs29, %rs61, %r153; + fma.rn.f32.bf16 %r155, %rs30, %rs62, %r154; + fma.rn.f32.bf16 %r156, %rs31, %rs63, %r155; + fma.rn.f32.bf16 %r157, %rs32, %rs64, %r156; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r158, %r136, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r159, %r136, %r158; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r160, %r159, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r161, %r159, %r160; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r162, %r161, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r163, %r161, %r162; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r164, %r163, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r165, %r163, %r164; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r166, %r143, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r167, %r143, %r166; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r168, %r167, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r169, %r167, %r168; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r170, %r169, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r171, %r169, %r170; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r172, %r171, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r173, %r171, %r172; +$L__tmp18: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r174, %r150, 8, 31, -1; +$L__tmp19: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r175, %r150, %r174; +$L__tmp20: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r176, %r175, 4, 31, -1; +$L__tmp21: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r177, %r175, %r176; +$L__tmp22: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r178, %r177, 2, 31, -1; +$L__tmp23: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r179, %r177, %r178; +$L__tmp24: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r180, %r179, 1, 31, -1; +$L__tmp25: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r181, %r179, %r180; +$L__tmp26: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r182, %r157, 8, 31, -1; +$L__tmp27: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r183, %r157, %r182; +$L__tmp28: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r184, %r183, 4, 31, -1; +$L__tmp29: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r185, %r183, %r184; +$L__tmp30: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r186, %r185, 2, 31, -1; +$L__tmp31: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r187, %r185, %r186; +$L__tmp32: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r188, %r187, 1, 31, -1; +$L__tmp33: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r189, %r187, %r188; +$L__tmp34: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r190, %r165, %r62; + mul.f32 %r191, %r165, %r63; + mul.f32 %r192, %r165, %r64; + mul.f32 %r193, %r165, %r65; + mul.f32 %r194, %r165, %r66; + mul.f32 %r195, %r165, %r67; + mul.f32 %r196, %r165, %r68; + mul.f32 %r197, %r165, %r69; + mul.f32 %r198, %r173, %r70; + mul.f32 %r199, %r173, %r71; + mul.f32 %r200, %r173, %r72; + mul.f32 %r201, %r173, %r73; + mul.f32 %r202, %r173, %r74; + mul.f32 %r203, %r173, %r75; + mul.f32 %r204, %r173, %r76; + mul.f32 %r205, %r173, %r77; + mul.f32 %r206, %r181, %r78; + mul.f32 %r207, %r181, %r79; + mul.f32 %r208, %r181, %r80; + mul.f32 %r209, %r181, %r81; + mul.f32 %r210, %r181, %r82; + mul.f32 %r211, %r181, %r83; + mul.f32 %r212, %r181, %r84; + mul.f32 %r213, %r181, %r85; + mul.f32 %r214, %r189, %r86; + mul.f32 %r215, %r189, %r87; + mul.f32 %r216, %r189, %r88; + mul.f32 %r217, %r189, %r89; + mul.f32 %r218, %r189, %r90; + mul.f32 %r219, %r189, %r91; + mul.f32 %r220, %r189, %r92; + mul.f32 %r221, %r189, %r93; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r222, %r190, %r18; + mul.f32 %r223, %r191, %r18; + mul.f32 %r224, %r192, %r18; + mul.f32 %r225, %r193, %r18; + mul.f32 %r226, %r194, %r18; + mul.f32 %r227, %r195, %r18; + mul.f32 %r228, %r196, %r18; + mul.f32 %r229, %r197, %r18; + mul.f32 %r230, %r198, %r19; + mul.f32 %r231, %r199, %r19; + mul.f32 %r232, %r200, %r19; + mul.f32 %r233, %r201, %r19; + mul.f32 %r234, %r202, %r19; + mul.f32 %r235, %r203, %r19; + mul.f32 %r236, %r204, %r19; + mul.f32 %r237, %r205, %r19; + mul.f32 %r238, %r206, %r20; + mul.f32 %r239, %r207, %r20; + mul.f32 %r240, %r208, %r20; + mul.f32 %r241, %r209, %r20; + mul.f32 %r242, %r210, %r20; + mul.f32 %r243, %r211, %r20; + mul.f32 %r244, %r212, %r20; + mul.f32 %r245, %r213, %r20; + mul.f32 %r246, %r214, %r21; + mul.f32 %r247, %r215, %r21; + mul.f32 %r248, %r216, %r21; + mul.f32 %r249, %r217, %r21; + mul.f32 %r250, %r218, %r21; + mul.f32 %r251, %r219, %r21; + mul.f32 %r252, %r220, %r21; + mul.f32 %r253, %r221, %r21; + .loc 1 107 36 // l2norm.py:107:36 + neg.f32 %r254, %r222; + fma.rn.f32 %r255, %r18, %r94, %r254; + neg.f32 %r256, %r223; + fma.rn.f32 %r257, %r18, %r95, %r256; + neg.f32 %r258, %r224; + fma.rn.f32 %r259, %r18, %r96, %r258; + neg.f32 %r260, %r225; + fma.rn.f32 %r261, %r18, %r97, %r260; + neg.f32 %r262, %r226; + fma.rn.f32 %r263, %r18, %r98, %r262; + neg.f32 %r264, %r227; + fma.rn.f32 %r265, %r18, %r99, %r264; + neg.f32 %r266, %r228; + fma.rn.f32 %r267, %r18, %r100, %r266; + neg.f32 %r268, %r229; + fma.rn.f32 %r269, %r18, %r101, %r268; + neg.f32 %r270, %r230; + fma.rn.f32 %r271, %r19, %r102, %r270; + neg.f32 %r272, %r231; + fma.rn.f32 %r273, %r19, %r103, %r272; + neg.f32 %r274, %r232; + fma.rn.f32 %r275, %r19, %r104, %r274; + neg.f32 %r276, %r233; + fma.rn.f32 %r277, %r19, %r105, %r276; + neg.f32 %r278, %r234; + fma.rn.f32 %r279, %r19, %r106, %r278; + neg.f32 %r280, %r235; + fma.rn.f32 %r281, %r19, %r107, %r280; + neg.f32 %r282, %r236; + fma.rn.f32 %r283, %r19, %r108, %r282; + neg.f32 %r284, %r237; + fma.rn.f32 %r285, %r19, %r109, %r284; + neg.f32 %r286, %r238; + fma.rn.f32 %r287, %r20, %r110, %r286; + neg.f32 %r288, %r239; + fma.rn.f32 %r289, %r20, %r111, %r288; + neg.f32 %r290, %r240; + fma.rn.f32 %r291, %r20, %r112, %r290; + neg.f32 %r292, %r241; + fma.rn.f32 %r293, %r20, %r113, %r292; + neg.f32 %r294, %r242; + fma.rn.f32 %r295, %r20, %r114, %r294; + neg.f32 %r296, %r243; + fma.rn.f32 %r297, %r20, %r115, %r296; + neg.f32 %r298, %r244; + fma.rn.f32 %r299, %r20, %r116, %r298; + neg.f32 %r300, %r245; + fma.rn.f32 %r301, %r20, %r117, %r300; + neg.f32 %r302, %r246; + fma.rn.f32 %r303, %r21, %r118, %r302; + neg.f32 %r304, %r247; + fma.rn.f32 %r305, %r21, %r119, %r304; + neg.f32 %r306, %r248; + fma.rn.f32 %r307, %r21, %r120, %r306; + neg.f32 %r308, %r249; + fma.rn.f32 %r309, %r21, %r121, %r308; + neg.f32 %r310, %r250; + fma.rn.f32 %r311, %r21, %r122, %r310; + neg.f32 %r312, %r251; + fma.rn.f32 %r313, %r21, %r123, %r312; + neg.f32 %r314, %r252; + fma.rn.f32 %r315, %r21, %r124, %r314; + neg.f32 %r316, %r253; + fma.rn.f32 %r317, %r21, %r125, %r316; + .loc 1 108 19 // l2norm.py:108:19 + cvt.rn.bf16x2.f32 %r38, %r257, %r255; + cvt.rn.bf16x2.f32 %r39, %r261, %r259; + cvt.rn.bf16x2.f32 %r40, %r265, %r263; + cvt.rn.bf16x2.f32 %r41, %r269, %r267; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd13 + 0 ], { %r38, %r39, %r40, %r41 }; + // end inline asm + cvt.rn.bf16x2.f32 %r42, %r273, %r271; + cvt.rn.bf16x2.f32 %r43, %r277, %r275; + cvt.rn.bf16x2.f32 %r44, %r281, %r279; + cvt.rn.bf16x2.f32 %r45, %r285, %r283; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd14 + 0 ], { %r42, %r43, %r44, %r45 }; + // end inline asm + cvt.rn.bf16x2.f32 %r46, %r289, %r287; + cvt.rn.bf16x2.f32 %r47, %r293, %r291; + cvt.rn.bf16x2.f32 %r48, %r297, %r295; + cvt.rn.bf16x2.f32 %r49, %r301, %r299; + // begin inline asm + @%p3 st.global.v4.b32 [ %rd15 + 0 ], { %r46, %r47, %r48, %r49 }; + // end inline asm + cvt.rn.bf16x2.f32 %r50, %r305, %r303; + cvt.rn.bf16x2.f32 %r51, %r309, %r307; + cvt.rn.bf16x2.f32 %r52, %r313, %r311; + cvt.rn.bf16x2.f32 %r53, %r317, %r315; + // begin inline asm + @%p4 st.global.v4.b32 [ %rd16 + 0 ], { %r50, %r51, %r52, %r53 }; + // end inline asm + .loc 1 108 4 // l2norm.py:108:4 + ret; +$L__tmp35: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 98 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp34 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 107 // DW_AT_call_line +.b8 55 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp34 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/l2norm_bwd_kernel.source b/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/l2norm_bwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..25926967fd8ce8ce5392a0037c1d84708c2d91fa --- /dev/null +++ b/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/l2norm_bwd_kernel.source @@ -0,0 +1,210 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc47 = loc(unknown) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc54 = loc("y"(#loc)) +#loc55 = loc("rstd"(#loc)) +#loc56 = loc("dy"(#loc)) +#loc57 = loc("dx"(#loc)) +#loc58 = loc("eps"(#loc)) +#loc59 = loc("T"(#loc)) +#loc101 = loc("input"(#loc45)) +#loc102 = loc("a"(#loc50)) +#loc103 = loc("b"(#loc50)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc60) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc61) + %o_t = arith.constant 16 : i32 loc(#loc62) + %o_t_1 = arith.constant 16 : i64 loc(#loc62) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc62) + %o_t_3 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32> loc(#loc63) + %o_t_4 = arith.extsi %o_t_3 : tensor<16xi32> to tensor<16xi64> loc(#loc64) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<16xi64> loc(#loc64) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<16xi64> loc(#loc64) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc65) + %m_t = arith.extsi %T : i32 to i64 loc(#loc66) + %m_t_7 = tt.splat %m_t : i64 -> tensor<16xi64> loc(#loc66) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<16xi64> loc(#loc66) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc67) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc68) + %m_x_10 = arith.constant 128 : i32 loc(#loc69) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc69) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc69) + %m_x_13 = tt.broadcast %m_x : tensor<16x1xi1> -> tensor<16x128xi1> loc(#loc70) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<16x128xi1> loc(#loc70) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<16x128xi1> loc(#loc70) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc71) + %p_y_16 = arith.constant 128 : i32 loc(#loc72) + %p_y_17 = arith.constant 128 : i64 loc(#loc72) + %p_y_18 = arith.constant dense<128> : tensor<16x1xi64> loc(#loc72) + %p_y_19 = arith.muli %p_y, %p_y_18 : tensor<16x1xi64> loc(#loc72) + %p_y_20 = tt.splat %y : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc73) + %p_y_21 = tt.addptr %p_y_20, %p_y_19 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc73) + %p_y_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc74) + %p_y_23 = tt.broadcast %p_y_21 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc75) + %p_y_24 = tt.broadcast %p_y_22 : tensor<1x128xi32> -> tensor<16x128xi32> loc(#loc75) + %p_y_25 = tt.addptr %p_y_23, %p_y_24 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc75) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc76) + %p_rstd_26 = tt.addptr %p_rstd, %o_t_6 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc76) + %p_dy = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc77) + %p_dy_27 = arith.constant 128 : i32 loc(#loc78) + %p_dy_28 = arith.constant 128 : i64 loc(#loc78) + %p_dy_29 = arith.constant dense<128> : tensor<16x1xi64> loc(#loc78) + %p_dy_30 = arith.muli %p_dy, %p_dy_29 : tensor<16x1xi64> loc(#loc78) + %p_dy_31 = tt.splat %dy : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc79) + %p_dy_32 = tt.addptr %p_dy_31, %p_dy_30 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc79) + %p_dy_33 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc80) + %p_dy_34 = tt.broadcast %p_dy_32 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc81) + %p_dy_35 = tt.broadcast %p_dy_33 : tensor<1x128xi32> -> tensor<16x128xi32> loc(#loc81) + %p_dy_36 = tt.addptr %p_dy_34, %p_dy_35 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc81) + %p_dx = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc82) + %p_dx_37 = arith.constant 128 : i32 loc(#loc83) + %p_dx_38 = arith.constant 128 : i64 loc(#loc83) + %p_dx_39 = arith.constant dense<128> : tensor<16x1xi64> loc(#loc83) + %p_dx_40 = arith.muli %p_dx, %p_dx_39 : tensor<16x1xi64> loc(#loc83) + %p_dx_41 = tt.splat %dx : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc84) + %p_dx_42 = tt.addptr %p_dx_41, %p_dx_40 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc84) + %p_dx_43 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc85) + %p_dx_44 = tt.broadcast %p_dx_42 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc86) + %p_dx_45 = tt.broadcast %p_dx_43 : tensor<1x128xi32> -> tensor<16x128xi32> loc(#loc86) + %p_dx_46 = tt.addptr %p_dx_44, %p_dx_45 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc86) + %b_y = arith.constant 0.000000e+00 : f32 loc(#loc87) + %b_y_47 = arith.constant dense<0.000000e+00> : tensor<16x128xf32> loc(#loc87) + %b_y_48 = arith.truncf %b_y_47 : tensor<16x128xf32> to tensor<16x128xbf16> loc(#loc87) + %b_y_49 = tt.load %p_y_25, %m_x_15, %b_y_48 : tensor<16x128x!tt.ptr> loc(#loc87) + %b_y_50 = arith.extf %b_y_49 : tensor<16x128xbf16> to tensor<16x128xf32> loc(#loc88) + %b_rstd = arith.constant 0.000000e+00 : f32 loc(#loc89) + %b_rstd_51 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc89) + %b_rstd_52 = tt.load %p_rstd_26, %m_t_8, %b_rstd_51 : tensor<16x!tt.ptr> loc(#loc89) + %b_dy = arith.constant 0.000000e+00 : f32 loc(#loc90) + %b_dy_53 = arith.constant dense<0.000000e+00> : tensor<16x128xf32> loc(#loc90) + %b_dy_54 = arith.truncf %b_dy_53 : tensor<16x128xf32> to tensor<16x128xbf16> loc(#loc90) + %b_dy_55 = tt.load %p_dy_36, %m_x_15, %b_dy_54 : tensor<16x128x!tt.ptr> loc(#loc90) + %b_dy_56 = arith.extf %b_dy_55 : tensor<16x128xbf16> to tensor<16x128xf32> loc(#loc91) + %b_dx = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc92) + %b_dx_57 = tt.broadcast %b_dx : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc93) + %b_dx_58 = arith.mulf %b_dy_56, %b_dx_57 : tensor<16x128xf32> loc(#loc93) + %b_dx_59 = arith.mulf %b_dy_56, %b_y_50 : tensor<16x128xf32> loc(#loc94) + %b_dx_60 = tt.call @triton.language.standard.sum__fp32S16_128S_c1_cFalse_cNone(%b_dx_59) : (tensor<16x128xf32>) -> tensor<16xf32> loc(#loc95) + %b_dx_61 = tt.expand_dims %b_dx_60 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc96) + %b_dx_62 = tt.broadcast %b_dx_61 : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc97) + %b_dx_63 = arith.mulf %b_dx_62, %b_y_50 : tensor<16x128xf32> loc(#loc97) + %b_dx_64 = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc98) + %b_dx_65 = tt.broadcast %b_dx_64 : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc99) + %b_dx_66 = arith.mulf %b_dx_63, %b_dx_65 : tensor<16x128xf32> loc(#loc99) + %b_dx_67 = arith.subf %b_dx_58, %b_dx_66 : tensor<16x128xf32> loc(#loc100) + %0 = arith.truncf %b_dx_67 : tensor<16x128xf32> to tensor<16x128xbf16> loc(#loc42) + tt.store %p_dx_46, %0, %m_x_15 : tensor<16x128x!tt.ptr> loc(#loc43) + tt.return loc(#loc44) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S16_128S_c1_cFalse_cNone(%input: tensor<16x128xf32> loc("input"(#loc45))) -> tensor<16xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc46) + tt.reduce.return %2 : f32 loc(#loc46) + }) : (tensor<16x128xf32>) -> tensor<16xf32> loc(#loc46) + tt.return %0 : tensor<16xf32> loc(#loc48) + ^bb1: // no predecessors + %1 = ub.poison : tensor<16xf32> loc(#loc49) + tt.return %1 : tensor<16xf32> loc(#loc49) + } loc(#loc45) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc50)), %b: f32 loc("b"(#loc50))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc51) + tt.return %0 : f32 loc(#loc52) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc53) + tt.return %1 : f32 loc(#loc53) + } loc(#loc50) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:31) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:39) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:20) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:31) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:39) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:82) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc60 = loc("i_t"(#loc1)) +#loc61 = loc("i_t"(#loc2)) +#loc62 = loc("o_t"(#loc3)) +#loc63 = loc("o_t"(#loc4)) +#loc64 = loc("o_t"(#loc5)) +#loc65 = loc("o_d"(#loc6)) +#loc66 = loc("m_t"(#loc7)) +#loc67 = loc("m_x"(#loc8)) +#loc68 = loc("m_x"(#loc9)) +#loc69 = loc("m_x"(#loc10)) +#loc70 = loc("m_x"(#loc11)) +#loc71 = loc("p_y"(#loc12)) +#loc72 = loc("p_y"(#loc13)) +#loc73 = loc("p_y"(#loc14)) +#loc74 = loc("p_y"(#loc15)) +#loc75 = loc("p_y"(#loc16)) +#loc76 = loc("p_rstd"(#loc17)) +#loc77 = loc("p_dy"(#loc18)) +#loc78 = loc("p_dy"(#loc19)) +#loc79 = loc("p_dy"(#loc20)) +#loc80 = loc("p_dy"(#loc21)) +#loc81 = loc("p_dy"(#loc22)) +#loc82 = loc("p_dx"(#loc23)) +#loc83 = loc("p_dx"(#loc24)) +#loc84 = loc("p_dx"(#loc25)) +#loc85 = loc("p_dx"(#loc26)) +#loc86 = loc("p_dx"(#loc27)) +#loc87 = loc("b_y"(#loc28)) +#loc88 = loc("b_y"(#loc29)) +#loc89 = loc("b_rstd"(#loc30)) +#loc90 = loc("b_dy"(#loc31)) +#loc91 = loc("b_dy"(#loc32)) +#loc92 = loc("b_dx"(#loc33)) +#loc93 = loc("b_dx"(#loc34)) +#loc94 = loc("b_dx"(#loc35)) +#loc95 = loc("b_dx"(#loc36)) +#loc96 = loc("b_dx"(#loc37)) +#loc97 = loc("b_dx"(#loc38)) +#loc98 = loc("b_dx"(#loc39)) +#loc99 = loc("b_dx"(#loc40)) +#loc100 = loc("b_dx"(#loc41)) diff --git a/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/l2norm_bwd_kernel.ttgir b/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/l2norm_bwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..fa83cd562f03a22e30c42f4fe750c97366cebd41 --- /dev/null +++ b/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/l2norm_bwd_kernel.ttgir @@ -0,0 +1,146 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [2, 1], order = [1, 0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc39 = loc("y"(#loc)) +#loc40 = loc("rstd"(#loc)) +#loc41 = loc("dy"(#loc)) +#loc42 = loc("dx"(#loc)) +#loc43 = loc("eps"(#loc)) +#loc44 = loc("T"(#loc)) +#loc72 = loc("b_dx"(#loc30)) +#loc78 = loc(callsite(#loc1 at #loc72)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 2 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %c16_i64 = arith.constant 16 : i64 loc(#loc1) + %cst = arith.constant dense<0.000000e+00> : tensor<16x128xbf16, #blocked> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<16x1xi64, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc45) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc46) + %o_t = arith.muli %i_t_3, %c16_i64 : i64 loc(#loc47) + %o_t_4 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc48) + %o_t_5 = arith.extsi %o_t_4 : tensor<16xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %o_t_6 = tt.splat %o_t : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %o_t_7 = arith.addi %o_t_6, %o_t_5 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %m_t = arith.extsi %T : i32 to i64 loc(#loc50) + %m_t_8 = tt.splat %m_t : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc50) + %m_t_9 = arith.cmpi slt, %o_t_7, %m_t_8 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc50) + %m_x = tt.expand_dims %m_t_9 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi1, #blocked> loc(#loc51) + %m_x_10 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc52) + %m_x_11 = tt.expand_dims %m_x_10 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc52) + %m_x_12 = arith.cmpi slt, %m_x_11, %cst_2 : tensor<1x128xi32, #blocked> loc(#loc53) + %m_x_13 = tt.broadcast %m_x : tensor<16x1xi1, #blocked> -> tensor<16x128xi1, #blocked> loc(#loc54) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1, #blocked> -> tensor<16x128xi1, #blocked> loc(#loc54) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<16x128xi1, #blocked> loc(#loc54) + %p_y = tt.expand_dims %o_t_7 {axis = 1 : i32} : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi64, #blocked> loc(#loc55) + %p_y_16 = arith.muli %p_y, %cst_1 : tensor<16x1xi64, #blocked> loc(#loc56) + %p_y_17 = tt.splat %y : !tt.ptr -> tensor<16x1x!tt.ptr, #blocked> loc(#loc57) + %p_y_18 = tt.addptr %p_y_17, %p_y_16 : tensor<16x1x!tt.ptr, #blocked>, tensor<16x1xi64, #blocked> loc(#loc57) + %p_y_19 = tt.broadcast %p_y_18 : tensor<16x1x!tt.ptr, #blocked> -> tensor<16x128x!tt.ptr, #blocked> loc(#loc58) + %p_y_20 = tt.broadcast %m_x_11 : tensor<1x128xi32, #blocked> -> tensor<16x128xi32, #blocked> loc(#loc58) + %p_y_21 = tt.addptr %p_y_19, %p_y_20 : tensor<16x128x!tt.ptr, #blocked>, tensor<16x128xi32, #blocked> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc59) + %p_rstd_22 = tt.addptr %p_rstd, %o_t_7 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>>, tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc59) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<16x1x!tt.ptr, #blocked> loc(#loc60) + %p_dy_23 = tt.addptr %p_dy, %p_y_16 : tensor<16x1x!tt.ptr, #blocked>, tensor<16x1xi64, #blocked> loc(#loc60) + %p_dy_24 = tt.broadcast %p_dy_23 : tensor<16x1x!tt.ptr, #blocked> -> tensor<16x128x!tt.ptr, #blocked> loc(#loc61) + %p_dy_25 = tt.addptr %p_dy_24, %p_y_20 : tensor<16x128x!tt.ptr, #blocked>, tensor<16x128xi32, #blocked> loc(#loc61) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<16x1x!tt.ptr, #blocked> loc(#loc62) + %p_dx_26 = tt.addptr %p_dx, %p_y_16 : tensor<16x1x!tt.ptr, #blocked>, tensor<16x1xi64, #blocked> loc(#loc62) + %p_dx_27 = tt.broadcast %p_dx_26 : tensor<16x1x!tt.ptr, #blocked> -> tensor<16x128x!tt.ptr, #blocked> loc(#loc63) + %p_dx_28 = tt.addptr %p_dx_27, %p_y_20 : tensor<16x128x!tt.ptr, #blocked>, tensor<16x128xi32, #blocked> loc(#loc63) + %b_y = tt.load %p_y_21, %m_x_15, %cst : tensor<16x128x!tt.ptr, #blocked> loc(#loc64) + %b_y_29 = arith.extf %b_y : tensor<16x128xbf16, #blocked> to tensor<16x128xf32, #blocked> loc(#loc65) + %b_rstd = tt.load %p_rstd_22, %m_t_9, %cst_0 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc66) + %b_dy = tt.load %p_dy_25, %m_x_15, %cst : tensor<16x128x!tt.ptr, #blocked> loc(#loc67) + %b_dy_30 = arith.extf %b_dy : tensor<16x128xbf16, #blocked> to tensor<16x128xf32, #blocked> loc(#loc68) + %b_dx = tt.expand_dims %b_rstd {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xf32, #blocked> loc(#loc69) + %b_dx_31 = tt.broadcast %b_dx : tensor<16x1xf32, #blocked> -> tensor<16x128xf32, #blocked> loc(#loc70) + %b_dx_32 = arith.mulf %b_dy_30, %b_dx_31 : tensor<16x128xf32, #blocked> loc(#loc70) + %b_dx_33 = arith.mulf %b_dy_30, %b_y_29 : tensor<16x128xf32, #blocked> loc(#loc71) + %b_dx_34 = "tt.reduce"(%b_dx_33) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_40: f32 loc(callsite(#loc1 at #loc72)), %b_dx_41: f32 loc(callsite(#loc1 at #loc72))): + %b_dx_42 = arith.addf %b_dx_40, %b_dx_41 : f32 loc(#loc79) + tt.reduce.return %b_dx_42 : f32 loc(#loc77) + }) : (tensor<16x128xf32, #blocked>) -> tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc77) + %b_dx_35 = tt.expand_dims %b_dx_34 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xf32, #blocked> loc(#loc73) + %b_dx_36 = tt.broadcast %b_dx_35 : tensor<16x1xf32, #blocked> -> tensor<16x128xf32, #blocked> loc(#loc74) + %b_dx_37 = arith.mulf %b_dx_36, %b_y_29 : tensor<16x128xf32, #blocked> loc(#loc74) + %b_dx_38 = arith.mulf %b_dx_37, %b_dx_31 : tensor<16x128xf32, #blocked> loc(#loc75) + %b_dx_39 = arith.subf %b_dx_32, %b_dx_38 : tensor<16x128xf32, #blocked> loc(#loc76) + %0 = arith.truncf %b_dx_39 : tensor<16x128xf32, #blocked> to tensor<16x128xbf16, #blocked> loc(#loc36) + tt.store %p_dx_28, %0, %m_x_15 : tensor<16x128x!tt.ptr, #blocked> loc(#loc37) + tt.return loc(#loc38) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc45 = loc("i_t"(#loc2)) +#loc46 = loc("i_t"(#loc3)) +#loc47 = loc("o_t"(#loc4)) +#loc48 = loc("o_t"(#loc5)) +#loc49 = loc("o_t"(#loc6)) +#loc50 = loc("m_t"(#loc7)) +#loc51 = loc("m_x"(#loc8)) +#loc52 = loc("m_x"(#loc9)) +#loc53 = loc("m_x"(#loc10)) +#loc54 = loc("m_x"(#loc11)) +#loc55 = loc("p_y"(#loc12)) +#loc56 = loc("p_y"(#loc13)) +#loc57 = loc("p_y"(#loc14)) +#loc58 = loc("p_y"(#loc15)) +#loc59 = loc("p_rstd"(#loc16)) +#loc60 = loc("p_dy"(#loc17)) +#loc61 = loc("p_dy"(#loc18)) +#loc62 = loc("p_dx"(#loc19)) +#loc63 = loc("p_dx"(#loc20)) +#loc64 = loc("b_y"(#loc21)) +#loc65 = loc("b_y"(#loc22)) +#loc66 = loc("b_rstd"(#loc23)) +#loc67 = loc("b_dy"(#loc24)) +#loc68 = loc("b_dy"(#loc25)) +#loc69 = loc("b_dx"(#loc26)) +#loc70 = loc("b_dx"(#loc27)) +#loc71 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc32)) +#loc74 = loc("b_dx"(#loc33)) +#loc75 = loc("b_dx"(#loc34)) +#loc76 = loc("b_dx"(#loc35)) +#loc77 = loc(callsite(#loc29 at #loc72)) +#loc79 = loc(callsite(#loc31 at #loc77)) diff --git a/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/l2norm_bwd_kernel.ttir b/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/l2norm_bwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..6c29416e2d7c989d3b096ee1e98cbabbdbd07e07 --- /dev/null +++ b/triton/gpu0/BKKEJMKDYY3FLLHVPZIS3N6EU7RRKU3POYCAKNG7ASDZHZ5XOSIA/l2norm_bwd_kernel.ttir @@ -0,0 +1,147 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc40 = loc("y"(#loc)) +#loc41 = loc("rstd"(#loc)) +#loc42 = loc("dy"(#loc)) +#loc43 = loc("dx"(#loc)) +#loc44 = loc("eps"(#loc)) +#loc45 = loc("T"(#loc)) +#loc74 = loc("b_dx"(#loc31)) +#loc80 = loc(callsite(#loc1 at #loc74)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<0.000000e+00> : tensor<16x128xbf16> loc(#loc1) + %b_rstd = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc46) + %cst_0 = arith.constant dense<128> : tensor<16x1xi64> loc(#loc1) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc47) + %o_t = arith.constant 16 : i64 loc(#loc48) + %i_t = tt.get_program_id x : i32 loc(#loc49) + %i_t_1 = arith.extsi %i_t : i32 to i64 loc(#loc50) + %o_t_2 = arith.muli %i_t_1, %o_t : i64 loc(#loc48) + %o_t_3 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32> loc(#loc51) + %o_t_4 = arith.extsi %o_t_3 : tensor<16xi32> to tensor<16xi64> loc(#loc52) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<16xi64> loc(#loc52) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<16xi64> loc(#loc52) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc53) + %m_t = arith.extsi %T : i32 to i64 loc(#loc54) + %m_t_7 = tt.splat %m_t : i64 -> tensor<16xi64> loc(#loc54) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<16xi64> loc(#loc54) + %m_x_9 = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc55) + %m_x_10 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc56) + %m_x_11 = arith.cmpi slt, %m_x_10, %m_x : tensor<1x128xi32> loc(#loc47) + %m_x_12 = tt.broadcast %m_x_9 : tensor<16x1xi1> -> tensor<16x128xi1> loc(#loc57) + %m_x_13 = tt.broadcast %m_x_11 : tensor<1x128xi1> -> tensor<16x128xi1> loc(#loc57) + %m_x_14 = arith.andi %m_x_12, %m_x_13 : tensor<16x128xi1> loc(#loc57) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc58) + %p_y_15 = arith.muli %p_y, %cst_0 : tensor<16x1xi64> loc(#loc59) + %p_y_16 = tt.splat %y : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc60) + %p_y_17 = tt.addptr %p_y_16, %p_y_15 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc60) + %p_y_18 = tt.broadcast %p_y_17 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc61) + %p_y_19 = tt.broadcast %m_x_10 : tensor<1x128xi32> -> tensor<16x128xi32> loc(#loc61) + %p_y_20 = tt.addptr %p_y_18, %p_y_19 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc61) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc62) + %p_rstd_21 = tt.addptr %p_rstd, %o_t_6 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc62) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc63) + %p_dy_22 = tt.addptr %p_dy, %p_y_15 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc63) + %p_dy_23 = tt.broadcast %p_dy_22 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc64) + %p_dy_24 = tt.addptr %p_dy_23, %p_y_19 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc64) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc65) + %p_dx_25 = tt.addptr %p_dx, %p_y_15 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc65) + %p_dx_26 = tt.broadcast %p_dx_25 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc66) + %p_dx_27 = tt.addptr %p_dx_26, %p_y_19 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc66) + %b_y = tt.load %p_y_20, %m_x_14, %cst : tensor<16x128x!tt.ptr> loc(#loc67) + %b_y_28 = arith.extf %b_y : tensor<16x128xbf16> to tensor<16x128xf32> loc(#loc68) + %b_rstd_29 = tt.load %p_rstd_21, %m_t_8, %b_rstd : tensor<16x!tt.ptr> loc(#loc46) + %b_dy = tt.load %p_dy_24, %m_x_14, %cst : tensor<16x128x!tt.ptr> loc(#loc69) + %b_dy_30 = arith.extf %b_dy : tensor<16x128xbf16> to tensor<16x128xf32> loc(#loc70) + %b_dx = tt.expand_dims %b_rstd_29 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc71) + %b_dx_31 = tt.broadcast %b_dx : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc72) + %b_dx_32 = arith.mulf %b_dy_30, %b_dx_31 : tensor<16x128xf32> loc(#loc72) + %b_dx_33 = arith.mulf %b_dy_30, %b_y_28 : tensor<16x128xf32> loc(#loc73) + %b_dx_34 = "tt.reduce"(%b_dx_33) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_40: f32 loc(callsite(#loc1 at #loc74)), %b_dx_41: f32 loc(callsite(#loc1 at #loc74))): + %b_dx_42 = arith.addf %b_dx_40, %b_dx_41 : f32 loc(#loc81) + tt.reduce.return %b_dx_42 : f32 loc(#loc79) + }) : (tensor<16x128xf32>) -> tensor<16xf32> loc(#loc79) + %b_dx_35 = tt.expand_dims %b_dx_34 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc75) + %b_dx_36 = tt.broadcast %b_dx_35 : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc76) + %b_dx_37 = arith.mulf %b_dx_36, %b_y_28 : tensor<16x128xf32> loc(#loc76) + %b_dx_38 = arith.mulf %b_dx_37, %b_dx_31 : tensor<16x128xf32> loc(#loc77) + %b_dx_39 = arith.subf %b_dx_32, %b_dx_38 : tensor<16x128xf32> loc(#loc78) + %0 = arith.truncf %b_dx_39 : tensor<16x128xf32> to tensor<16x128xbf16> loc(#loc37) + tt.store %p_dx_27, %0, %m_x_14 : tensor<16x128x!tt.ptr> loc(#loc38) + tt.return loc(#loc39) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("b_rstd"(#loc2)) +#loc47 = loc("m_x"(#loc3)) +#loc48 = loc("o_t"(#loc4)) +#loc49 = loc("i_t"(#loc5)) +#loc50 = loc("i_t"(#loc6)) +#loc51 = loc("o_t"(#loc7)) +#loc52 = loc("o_t"(#loc8)) +#loc53 = loc("o_d"(#loc9)) +#loc54 = loc("m_t"(#loc10)) +#loc55 = loc("m_x"(#loc11)) +#loc56 = loc("m_x"(#loc12)) +#loc57 = loc("m_x"(#loc13)) +#loc58 = loc("p_y"(#loc14)) +#loc59 = loc("p_y"(#loc15)) +#loc60 = loc("p_y"(#loc16)) +#loc61 = loc("p_y"(#loc17)) +#loc62 = loc("p_rstd"(#loc18)) +#loc63 = loc("p_dy"(#loc19)) +#loc64 = loc("p_dy"(#loc20)) +#loc65 = loc("p_dx"(#loc21)) +#loc66 = loc("p_dx"(#loc22)) +#loc67 = loc("b_y"(#loc23)) +#loc68 = loc("b_y"(#loc24)) +#loc69 = loc("b_dy"(#loc25)) +#loc70 = loc("b_dy"(#loc26)) +#loc71 = loc("b_dx"(#loc27)) +#loc72 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc29)) +#loc75 = loc("b_dx"(#loc33)) +#loc76 = loc("b_dx"(#loc34)) +#loc77 = loc("b_dx"(#loc35)) +#loc78 = loc("b_dx"(#loc36)) +#loc79 = loc(callsite(#loc30 at #loc74)) +#loc81 = loc(callsite(#loc32 at #loc79)) diff --git a/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/__grp__l2norm_bwd_kernel.json b/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/__grp__l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..45f1aac7f8c55cbdbd435c40703308ed550b002f --- /dev/null +++ b/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/__grp__l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_bwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/l2norm_bwd_kernel.source", "l2norm_bwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/l2norm_bwd_kernel.ttir", "l2norm_bwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/l2norm_bwd_kernel.ttgir", "l2norm_bwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/l2norm_bwd_kernel.llir", "l2norm_bwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/l2norm_bwd_kernel.ptx", "l2norm_bwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/l2norm_bwd_kernel.cubin", "l2norm_bwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/l2norm_bwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/l2norm_bwd_kernel.cubin b/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/l2norm_bwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..7e1b04550b51a5079e8749ac25d9411e3943bd88 Binary files /dev/null and b/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/l2norm_bwd_kernel.cubin differ diff --git a/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/l2norm_bwd_kernel.json b/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..34998dbddadd4e2aeb980733598ae6420d338371 --- /dev/null +++ b/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"hash": "0b4ceea756f13bf793321935c284c2bfc415c7b68a76bd08aa7951d5fd8a8de2", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 1, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 0, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_bwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/l2norm_bwd_kernel.llir b/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/l2norm_bwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..6c6e883877d482f36857a2b69535d3994cc60282 --- /dev/null +++ b/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/l2norm_bwd_kernel.llir @@ -0,0 +1,1265 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_bwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, ptr addrspace(1) %3, float %4, i32 %5, ptr addrspace(1) readnone captures(none) %6, ptr addrspace(1) readnone captures(none) %7) local_unnamed_addr #0 !dbg !4 { + %9 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !12 + %10 = zext nneg i32 %9 to i64, !dbg !13 + %11 = shl nuw nsw i64 %10, 4, !dbg !14 + %12 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !15 + %13 = lshr i32 %12, 4, !dbg !15 + %.lobit = and i32 %13, 1, !dbg !15 + %14 = or disjoint i32 %.lobit, 2, !dbg !15 + %15 = or disjoint i32 %.lobit, 4, !dbg !15 + %16 = or disjoint i32 %.lobit, 6, !dbg !15 + %17 = or disjoint i32 %.lobit, 8, !dbg !15 + %18 = or disjoint i32 %.lobit, 10, !dbg !15 + %19 = or disjoint i32 %.lobit, 12, !dbg !15 + %20 = or disjoint i32 %.lobit, 14, !dbg !15 + %21 = zext nneg i32 %.lobit to i64, !dbg !16 + %22 = zext nneg i32 %14 to i64, !dbg !16 + %23 = zext nneg i32 %15 to i64, !dbg !16 + %24 = zext nneg i32 %16 to i64, !dbg !16 + %25 = zext nneg i32 %17 to i64, !dbg !16 + %26 = zext nneg i32 %18 to i64, !dbg !16 + %27 = zext nneg i32 %19 to i64, !dbg !16 + %28 = zext nneg i32 %20 to i64, !dbg !16 + %29 = or disjoint i64 %11, %21, !dbg !16 + %30 = or disjoint i64 %11, %22, !dbg !16 + %31 = or disjoint i64 %11, %23, !dbg !16 + %32 = or disjoint i64 %11, %24, !dbg !16 + %33 = or disjoint i64 %11, %25, !dbg !16 + %34 = or disjoint i64 %11, %26, !dbg !16 + %35 = or disjoint i64 %11, %27, !dbg !16 + %36 = or disjoint i64 %11, %28, !dbg !16 + %37 = sext i32 %5 to i64, !dbg !17 + %38 = icmp slt i64 %29, %37, !dbg !17 + %39 = icmp slt i64 %30, %37, !dbg !17 + %40 = icmp slt i64 %31, %37, !dbg !17 + %41 = icmp slt i64 %32, %37, !dbg !17 + %42 = icmp slt i64 %33, %37, !dbg !17 + %43 = icmp slt i64 %34, %37, !dbg !17 + %44 = icmp slt i64 %35, %37, !dbg !17 + %45 = icmp slt i64 %36, %37, !dbg !17 + %46 = shl nuw nsw i32 %12, 3, !dbg !18 + %47 = and i32 %46, 120, !dbg !18 + %48 = shl nuw nsw i64 %29, 7, !dbg !19 + %49 = shl nuw nsw i64 %30, 7, !dbg !19 + %50 = shl nuw nsw i64 %31, 7, !dbg !19 + %51 = shl nuw nsw i64 %32, 7, !dbg !19 + %52 = shl nuw nsw i64 %33, 7, !dbg !19 + %53 = shl nuw nsw i64 %34, 7, !dbg !19 + %54 = shl nuw nsw i64 %35, 7, !dbg !19 + %55 = shl nuw nsw i64 %36, 7, !dbg !19 + %56 = getelementptr bfloat, ptr addrspace(1) %0, i64 %48, !dbg !20 + %57 = getelementptr bfloat, ptr addrspace(1) %0, i64 %49, !dbg !20 + %58 = getelementptr bfloat, ptr addrspace(1) %0, i64 %50, !dbg !20 + %59 = getelementptr bfloat, ptr addrspace(1) %0, i64 %51, !dbg !20 + %60 = getelementptr bfloat, ptr addrspace(1) %0, i64 %52, !dbg !20 + %61 = getelementptr bfloat, ptr addrspace(1) %0, i64 %53, !dbg !20 + %62 = getelementptr bfloat, ptr addrspace(1) %0, i64 %54, !dbg !20 + %63 = getelementptr bfloat, ptr addrspace(1) %0, i64 %55, !dbg !20 + %64 = zext nneg i32 %47 to i64, !dbg !21 + %65 = getelementptr bfloat, ptr addrspace(1) %56, i64 %64, !dbg !21 + %66 = getelementptr bfloat, ptr addrspace(1) %57, i64 %64, !dbg !21 + %67 = getelementptr bfloat, ptr addrspace(1) %58, i64 %64, !dbg !21 + %68 = getelementptr bfloat, ptr addrspace(1) %59, i64 %64, !dbg !21 + %69 = getelementptr bfloat, ptr addrspace(1) %60, i64 %64, !dbg !21 + %70 = getelementptr bfloat, ptr addrspace(1) %61, i64 %64, !dbg !21 + %71 = getelementptr bfloat, ptr addrspace(1) %62, i64 %64, !dbg !21 + %72 = getelementptr bfloat, ptr addrspace(1) %63, i64 %64, !dbg !21 + %73 = getelementptr float, ptr addrspace(1) %1, i64 %29, !dbg !22 + %74 = getelementptr float, ptr addrspace(1) %1, i64 %30, !dbg !22 + %75 = getelementptr float, ptr addrspace(1) %1, i64 %31, !dbg !22 + %76 = getelementptr float, ptr addrspace(1) %1, i64 %32, !dbg !22 + %77 = getelementptr float, ptr addrspace(1) %1, i64 %33, !dbg !22 + %78 = getelementptr float, ptr addrspace(1) %1, i64 %34, !dbg !22 + %79 = getelementptr float, ptr addrspace(1) %1, i64 %35, !dbg !22 + %80 = getelementptr float, ptr addrspace(1) %1, i64 %36, !dbg !22 + %81 = getelementptr bfloat, ptr addrspace(1) %2, i64 %48, !dbg !23 + %82 = getelementptr bfloat, ptr addrspace(1) %2, i64 %49, !dbg !23 + %83 = getelementptr bfloat, ptr addrspace(1) %2, i64 %50, !dbg !23 + %84 = getelementptr bfloat, ptr addrspace(1) %2, i64 %51, !dbg !23 + %85 = getelementptr bfloat, ptr addrspace(1) %2, i64 %52, !dbg !23 + %86 = getelementptr bfloat, ptr addrspace(1) %2, i64 %53, !dbg !23 + %87 = getelementptr bfloat, ptr addrspace(1) %2, i64 %54, !dbg !23 + %88 = getelementptr bfloat, ptr addrspace(1) %2, i64 %55, !dbg !23 + %89 = getelementptr bfloat, ptr addrspace(1) %81, i64 %64, !dbg !24 + %90 = getelementptr bfloat, ptr addrspace(1) %82, i64 %64, !dbg !24 + %91 = getelementptr bfloat, ptr addrspace(1) %83, i64 %64, !dbg !24 + %92 = getelementptr bfloat, ptr addrspace(1) %84, i64 %64, !dbg !24 + %93 = getelementptr bfloat, ptr addrspace(1) %85, i64 %64, !dbg !24 + %94 = getelementptr bfloat, ptr addrspace(1) %86, i64 %64, !dbg !24 + %95 = getelementptr bfloat, ptr addrspace(1) %87, i64 %64, !dbg !24 + %96 = getelementptr bfloat, ptr addrspace(1) %88, i64 %64, !dbg !24 + %97 = getelementptr bfloat, ptr addrspace(1) %3, i64 %48, !dbg !25 + %98 = getelementptr bfloat, ptr addrspace(1) %3, i64 %49, !dbg !25 + %99 = getelementptr bfloat, ptr addrspace(1) %3, i64 %50, !dbg !25 + %100 = getelementptr bfloat, ptr addrspace(1) %3, i64 %51, !dbg !25 + %101 = getelementptr bfloat, ptr addrspace(1) %3, i64 %52, !dbg !25 + %102 = getelementptr bfloat, ptr addrspace(1) %3, i64 %53, !dbg !25 + %103 = getelementptr bfloat, ptr addrspace(1) %3, i64 %54, !dbg !25 + %104 = getelementptr bfloat, ptr addrspace(1) %3, i64 %55, !dbg !25 + %105 = getelementptr bfloat, ptr addrspace(1) %97, i64 %64, !dbg !26 + %106 = getelementptr bfloat, ptr addrspace(1) %98, i64 %64, !dbg !26 + %107 = getelementptr bfloat, ptr addrspace(1) %99, i64 %64, !dbg !26 + %108 = getelementptr bfloat, ptr addrspace(1) %100, i64 %64, !dbg !26 + %109 = getelementptr bfloat, ptr addrspace(1) %101, i64 %64, !dbg !26 + %110 = getelementptr bfloat, ptr addrspace(1) %102, i64 %64, !dbg !26 + %111 = getelementptr bfloat, ptr addrspace(1) %103, i64 %64, !dbg !26 + %112 = getelementptr bfloat, ptr addrspace(1) %104, i64 %64, !dbg !26 + %113 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %65, i1 %38) #3, !dbg !27 + %114 = extractvalue { i32, i32, i32, i32 } %113, 0, !dbg !27 + %115 = bitcast i32 %114 to <2 x bfloat>, !dbg !27 + %116 = extractvalue { i32, i32, i32, i32 } %113, 1, !dbg !27 + %117 = bitcast i32 %116 to <2 x bfloat>, !dbg !27 + %118 = extractvalue { i32, i32, i32, i32 } %113, 2, !dbg !27 + %119 = bitcast i32 %118 to <2 x bfloat>, !dbg !27 + %120 = extractvalue { i32, i32, i32, i32 } %113, 3, !dbg !27 + %121 = bitcast i32 %120 to <2 x bfloat>, !dbg !27 + %122 = extractelement <2 x bfloat> %115, i64 0, !dbg !27 + %123 = extractelement <2 x bfloat> %115, i64 1, !dbg !27 + %124 = extractelement <2 x bfloat> %117, i64 0, !dbg !27 + %125 = extractelement <2 x bfloat> %117, i64 1, !dbg !27 + %126 = extractelement <2 x bfloat> %119, i64 0, !dbg !27 + %127 = extractelement <2 x bfloat> %119, i64 1, !dbg !27 + %128 = extractelement <2 x bfloat> %121, i64 0, !dbg !27 + %129 = extractelement <2 x bfloat> %121, i64 1, !dbg !27 + %130 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %66, i1 %39) #3, !dbg !27 + %131 = extractvalue { i32, i32, i32, i32 } %130, 0, !dbg !27 + %132 = bitcast i32 %131 to <2 x bfloat>, !dbg !27 + %133 = extractvalue { i32, i32, i32, i32 } %130, 1, !dbg !27 + %134 = bitcast i32 %133 to <2 x bfloat>, !dbg !27 + %135 = extractvalue { i32, i32, i32, i32 } %130, 2, !dbg !27 + %136 = bitcast i32 %135 to <2 x bfloat>, !dbg !27 + %137 = extractvalue { i32, i32, i32, i32 } %130, 3, !dbg !27 + %138 = bitcast i32 %137 to <2 x bfloat>, !dbg !27 + %139 = extractelement <2 x bfloat> %132, i64 0, !dbg !27 + %140 = extractelement <2 x bfloat> %132, i64 1, !dbg !27 + %141 = extractelement <2 x bfloat> %134, i64 0, !dbg !27 + %142 = extractelement <2 x bfloat> %134, i64 1, !dbg !27 + %143 = extractelement <2 x bfloat> %136, i64 0, !dbg !27 + %144 = extractelement <2 x bfloat> %136, i64 1, !dbg !27 + %145 = extractelement <2 x bfloat> %138, i64 0, !dbg !27 + %146 = extractelement <2 x bfloat> %138, i64 1, !dbg !27 + %147 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %67, i1 %40) #3, !dbg !27 + %148 = extractvalue { i32, i32, i32, i32 } %147, 0, !dbg !27 + %149 = bitcast i32 %148 to <2 x bfloat>, !dbg !27 + %150 = extractvalue { i32, i32, i32, i32 } %147, 1, !dbg !27 + %151 = bitcast i32 %150 to <2 x bfloat>, !dbg !27 + %152 = extractvalue { i32, i32, i32, i32 } %147, 2, !dbg !27 + %153 = bitcast i32 %152 to <2 x bfloat>, !dbg !27 + %154 = extractvalue { i32, i32, i32, i32 } %147, 3, !dbg !27 + %155 = bitcast i32 %154 to <2 x bfloat>, !dbg !27 + %156 = extractelement <2 x bfloat> %149, i64 0, !dbg !27 + %157 = extractelement <2 x bfloat> %149, i64 1, !dbg !27 + %158 = extractelement <2 x bfloat> %151, i64 0, !dbg !27 + %159 = extractelement <2 x bfloat> %151, i64 1, !dbg !27 + %160 = extractelement <2 x bfloat> %153, i64 0, !dbg !27 + %161 = extractelement <2 x bfloat> %153, i64 1, !dbg !27 + %162 = extractelement <2 x bfloat> %155, i64 0, !dbg !27 + %163 = extractelement <2 x bfloat> %155, i64 1, !dbg !27 + %164 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %68, i1 %41) #3, !dbg !27 + %165 = extractvalue { i32, i32, i32, i32 } %164, 0, !dbg !27 + %166 = bitcast i32 %165 to <2 x bfloat>, !dbg !27 + %167 = extractvalue { i32, i32, i32, i32 } %164, 1, !dbg !27 + %168 = bitcast i32 %167 to <2 x bfloat>, !dbg !27 + %169 = extractvalue { i32, i32, i32, i32 } %164, 2, !dbg !27 + %170 = bitcast i32 %169 to <2 x bfloat>, !dbg !27 + %171 = extractvalue { i32, i32, i32, i32 } %164, 3, !dbg !27 + %172 = bitcast i32 %171 to <2 x bfloat>, !dbg !27 + %173 = extractelement <2 x bfloat> %166, i64 0, !dbg !27 + %174 = extractelement <2 x bfloat> %166, i64 1, !dbg !27 + %175 = extractelement <2 x bfloat> %168, i64 0, !dbg !27 + %176 = extractelement <2 x bfloat> %168, i64 1, !dbg !27 + %177 = extractelement <2 x bfloat> %170, i64 0, !dbg !27 + %178 = extractelement <2 x bfloat> %170, i64 1, !dbg !27 + %179 = extractelement <2 x bfloat> %172, i64 0, !dbg !27 + %180 = extractelement <2 x bfloat> %172, i64 1, !dbg !27 + %181 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %69, i1 %42) #3, !dbg !27 + %182 = extractvalue { i32, i32, i32, i32 } %181, 0, !dbg !27 + %183 = bitcast i32 %182 to <2 x bfloat>, !dbg !27 + %184 = extractvalue { i32, i32, i32, i32 } %181, 1, !dbg !27 + %185 = bitcast i32 %184 to <2 x bfloat>, !dbg !27 + %186 = extractvalue { i32, i32, i32, i32 } %181, 2, !dbg !27 + %187 = bitcast i32 %186 to <2 x bfloat>, !dbg !27 + %188 = extractvalue { i32, i32, i32, i32 } %181, 3, !dbg !27 + %189 = bitcast i32 %188 to <2 x bfloat>, !dbg !27 + %190 = extractelement <2 x bfloat> %183, i64 0, !dbg !27 + %191 = extractelement <2 x bfloat> %183, i64 1, !dbg !27 + %192 = extractelement <2 x bfloat> %185, i64 0, !dbg !27 + %193 = extractelement <2 x bfloat> %185, i64 1, !dbg !27 + %194 = extractelement <2 x bfloat> %187, i64 0, !dbg !27 + %195 = extractelement <2 x bfloat> %187, i64 1, !dbg !27 + %196 = extractelement <2 x bfloat> %189, i64 0, !dbg !27 + %197 = extractelement <2 x bfloat> %189, i64 1, !dbg !27 + %198 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %70, i1 %43) #3, !dbg !27 + %199 = extractvalue { i32, i32, i32, i32 } %198, 0, !dbg !27 + %200 = bitcast i32 %199 to <2 x bfloat>, !dbg !27 + %201 = extractvalue { i32, i32, i32, i32 } %198, 1, !dbg !27 + %202 = bitcast i32 %201 to <2 x bfloat>, !dbg !27 + %203 = extractvalue { i32, i32, i32, i32 } %198, 2, !dbg !27 + %204 = bitcast i32 %203 to <2 x bfloat>, !dbg !27 + %205 = extractvalue { i32, i32, i32, i32 } %198, 3, !dbg !27 + %206 = bitcast i32 %205 to <2 x bfloat>, !dbg !27 + %207 = extractelement <2 x bfloat> %200, i64 0, !dbg !27 + %208 = extractelement <2 x bfloat> %200, i64 1, !dbg !27 + %209 = extractelement <2 x bfloat> %202, i64 0, !dbg !27 + %210 = extractelement <2 x bfloat> %202, i64 1, !dbg !27 + %211 = extractelement <2 x bfloat> %204, i64 0, !dbg !27 + %212 = extractelement <2 x bfloat> %204, i64 1, !dbg !27 + %213 = extractelement <2 x bfloat> %206, i64 0, !dbg !27 + %214 = extractelement <2 x bfloat> %206, i64 1, !dbg !27 + %215 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %71, i1 %44) #3, !dbg !27 + %216 = extractvalue { i32, i32, i32, i32 } %215, 0, !dbg !27 + %217 = bitcast i32 %216 to <2 x bfloat>, !dbg !27 + %218 = extractvalue { i32, i32, i32, i32 } %215, 1, !dbg !27 + %219 = bitcast i32 %218 to <2 x bfloat>, !dbg !27 + %220 = extractvalue { i32, i32, i32, i32 } %215, 2, !dbg !27 + %221 = bitcast i32 %220 to <2 x bfloat>, !dbg !27 + %222 = extractvalue { i32, i32, i32, i32 } %215, 3, !dbg !27 + %223 = bitcast i32 %222 to <2 x bfloat>, !dbg !27 + %224 = extractelement <2 x bfloat> %217, i64 0, !dbg !27 + %225 = extractelement <2 x bfloat> %217, i64 1, !dbg !27 + %226 = extractelement <2 x bfloat> %219, i64 0, !dbg !27 + %227 = extractelement <2 x bfloat> %219, i64 1, !dbg !27 + %228 = extractelement <2 x bfloat> %221, i64 0, !dbg !27 + %229 = extractelement <2 x bfloat> %221, i64 1, !dbg !27 + %230 = extractelement <2 x bfloat> %223, i64 0, !dbg !27 + %231 = extractelement <2 x bfloat> %223, i64 1, !dbg !27 + %232 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %72, i1 %45) #3, !dbg !27 + %233 = extractvalue { i32, i32, i32, i32 } %232, 0, !dbg !27 + %234 = bitcast i32 %233 to <2 x bfloat>, !dbg !27 + %235 = extractvalue { i32, i32, i32, i32 } %232, 1, !dbg !27 + %236 = bitcast i32 %235 to <2 x bfloat>, !dbg !27 + %237 = extractvalue { i32, i32, i32, i32 } %232, 2, !dbg !27 + %238 = bitcast i32 %237 to <2 x bfloat>, !dbg !27 + %239 = extractvalue { i32, i32, i32, i32 } %232, 3, !dbg !27 + %240 = bitcast i32 %239 to <2 x bfloat>, !dbg !27 + %241 = extractelement <2 x bfloat> %234, i64 0, !dbg !27 + %242 = extractelement <2 x bfloat> %234, i64 1, !dbg !27 + %243 = extractelement <2 x bfloat> %236, i64 0, !dbg !27 + %244 = extractelement <2 x bfloat> %236, i64 1, !dbg !27 + %245 = extractelement <2 x bfloat> %238, i64 0, !dbg !27 + %246 = extractelement <2 x bfloat> %238, i64 1, !dbg !27 + %247 = extractelement <2 x bfloat> %240, i64 0, !dbg !27 + %248 = extractelement <2 x bfloat> %240, i64 1, !dbg !27 + %249 = fpext bfloat %122 to float, !dbg !28 + %250 = fpext bfloat %123 to float, !dbg !28 + %251 = fpext bfloat %124 to float, !dbg !28 + %252 = fpext bfloat %125 to float, !dbg !28 + %253 = fpext bfloat %126 to float, !dbg !28 + %254 = fpext bfloat %127 to float, !dbg !28 + %255 = fpext bfloat %128 to float, !dbg !28 + %256 = fpext bfloat %129 to float, !dbg !28 + %257 = fpext bfloat %139 to float, !dbg !28 + %258 = fpext bfloat %140 to float, !dbg !28 + %259 = fpext bfloat %141 to float, !dbg !28 + %260 = fpext bfloat %142 to float, !dbg !28 + %261 = fpext bfloat %143 to float, !dbg !28 + %262 = fpext bfloat %144 to float, !dbg !28 + %263 = fpext bfloat %145 to float, !dbg !28 + %264 = fpext bfloat %146 to float, !dbg !28 + %265 = fpext bfloat %156 to float, !dbg !28 + %266 = fpext bfloat %157 to float, !dbg !28 + %267 = fpext bfloat %158 to float, !dbg !28 + %268 = fpext bfloat %159 to float, !dbg !28 + %269 = fpext bfloat %160 to float, !dbg !28 + %270 = fpext bfloat %161 to float, !dbg !28 + %271 = fpext bfloat %162 to float, !dbg !28 + %272 = fpext bfloat %163 to float, !dbg !28 + %273 = fpext bfloat %173 to float, !dbg !28 + %274 = fpext bfloat %174 to float, !dbg !28 + %275 = fpext bfloat %175 to float, !dbg !28 + %276 = fpext bfloat %176 to float, !dbg !28 + %277 = fpext bfloat %177 to float, !dbg !28 + %278 = fpext bfloat %178 to float, !dbg !28 + %279 = fpext bfloat %179 to float, !dbg !28 + %280 = fpext bfloat %180 to float, !dbg !28 + %281 = fpext bfloat %190 to float, !dbg !28 + %282 = fpext bfloat %191 to float, !dbg !28 + %283 = fpext bfloat %192 to float, !dbg !28 + %284 = fpext bfloat %193 to float, !dbg !28 + %285 = fpext bfloat %194 to float, !dbg !28 + %286 = fpext bfloat %195 to float, !dbg !28 + %287 = fpext bfloat %196 to float, !dbg !28 + %288 = fpext bfloat %197 to float, !dbg !28 + %289 = fpext bfloat %207 to float, !dbg !28 + %290 = fpext bfloat %208 to float, !dbg !28 + %291 = fpext bfloat %209 to float, !dbg !28 + %292 = fpext bfloat %210 to float, !dbg !28 + %293 = fpext bfloat %211 to float, !dbg !28 + %294 = fpext bfloat %212 to float, !dbg !28 + %295 = fpext bfloat %213 to float, !dbg !28 + %296 = fpext bfloat %214 to float, !dbg !28 + %297 = fpext bfloat %224 to float, !dbg !28 + %298 = fpext bfloat %225 to float, !dbg !28 + %299 = fpext bfloat %226 to float, !dbg !28 + %300 = fpext bfloat %227 to float, !dbg !28 + %301 = fpext bfloat %228 to float, !dbg !28 + %302 = fpext bfloat %229 to float, !dbg !28 + %303 = fpext bfloat %230 to float, !dbg !28 + %304 = fpext bfloat %231 to float, !dbg !28 + %305 = fpext bfloat %241 to float, !dbg !28 + %306 = fpext bfloat %242 to float, !dbg !28 + %307 = fpext bfloat %243 to float, !dbg !28 + %308 = fpext bfloat %244 to float, !dbg !28 + %309 = fpext bfloat %245 to float, !dbg !28 + %310 = fpext bfloat %246 to float, !dbg !28 + %311 = fpext bfloat %247 to float, !dbg !28 + %312 = fpext bfloat %248 to float, !dbg !28 + %313 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %73, i1 %38) #3, !dbg !29 + %314 = bitcast i32 %313 to float, !dbg !29 + %315 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %74, i1 %39) #3, !dbg !29 + %316 = bitcast i32 %315 to float, !dbg !29 + %317 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %75, i1 %40) #3, !dbg !29 + %318 = bitcast i32 %317 to float, !dbg !29 + %319 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %76, i1 %41) #3, !dbg !29 + %320 = bitcast i32 %319 to float, !dbg !29 + %321 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %77, i1 %42) #3, !dbg !29 + %322 = bitcast i32 %321 to float, !dbg !29 + %323 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %78, i1 %43) #3, !dbg !29 + %324 = bitcast i32 %323 to float, !dbg !29 + %325 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %79, i1 %44) #3, !dbg !29 + %326 = bitcast i32 %325 to float, !dbg !29 + %327 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %80, i1 %45) #3, !dbg !29 + %328 = bitcast i32 %327 to float, !dbg !29 + %329 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %89, i1 %38) #3, !dbg !30 + %330 = extractvalue { i32, i32, i32, i32 } %329, 0, !dbg !30 + %331 = bitcast i32 %330 to <2 x bfloat>, !dbg !30 + %332 = extractvalue { i32, i32, i32, i32 } %329, 1, !dbg !30 + %333 = bitcast i32 %332 to <2 x bfloat>, !dbg !30 + %334 = extractvalue { i32, i32, i32, i32 } %329, 2, !dbg !30 + %335 = bitcast i32 %334 to <2 x bfloat>, !dbg !30 + %336 = extractvalue { i32, i32, i32, i32 } %329, 3, !dbg !30 + %337 = bitcast i32 %336 to <2 x bfloat>, !dbg !30 + %338 = extractelement <2 x bfloat> %331, i64 0, !dbg !30 + %339 = extractelement <2 x bfloat> %331, i64 1, !dbg !30 + %340 = extractelement <2 x bfloat> %333, i64 0, !dbg !30 + %341 = extractelement <2 x bfloat> %333, i64 1, !dbg !30 + %342 = extractelement <2 x bfloat> %335, i64 0, !dbg !30 + %343 = extractelement <2 x bfloat> %335, i64 1, !dbg !30 + %344 = extractelement <2 x bfloat> %337, i64 0, !dbg !30 + %345 = extractelement <2 x bfloat> %337, i64 1, !dbg !30 + %346 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %90, i1 %39) #3, !dbg !30 + %347 = extractvalue { i32, i32, i32, i32 } %346, 0, !dbg !30 + %348 = bitcast i32 %347 to <2 x bfloat>, !dbg !30 + %349 = extractvalue { i32, i32, i32, i32 } %346, 1, !dbg !30 + %350 = bitcast i32 %349 to <2 x bfloat>, !dbg !30 + %351 = extractvalue { i32, i32, i32, i32 } %346, 2, !dbg !30 + %352 = bitcast i32 %351 to <2 x bfloat>, !dbg !30 + %353 = extractvalue { i32, i32, i32, i32 } %346, 3, !dbg !30 + %354 = bitcast i32 %353 to <2 x bfloat>, !dbg !30 + %355 = extractelement <2 x bfloat> %348, i64 0, !dbg !30 + %356 = extractelement <2 x bfloat> %348, i64 1, !dbg !30 + %357 = extractelement <2 x bfloat> %350, i64 0, !dbg !30 + %358 = extractelement <2 x bfloat> %350, i64 1, !dbg !30 + %359 = extractelement <2 x bfloat> %352, i64 0, !dbg !30 + %360 = extractelement <2 x bfloat> %352, i64 1, !dbg !30 + %361 = extractelement <2 x bfloat> %354, i64 0, !dbg !30 + %362 = extractelement <2 x bfloat> %354, i64 1, !dbg !30 + %363 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %91, i1 %40) #3, !dbg !30 + %364 = extractvalue { i32, i32, i32, i32 } %363, 0, !dbg !30 + %365 = bitcast i32 %364 to <2 x bfloat>, !dbg !30 + %366 = extractvalue { i32, i32, i32, i32 } %363, 1, !dbg !30 + %367 = bitcast i32 %366 to <2 x bfloat>, !dbg !30 + %368 = extractvalue { i32, i32, i32, i32 } %363, 2, !dbg !30 + %369 = bitcast i32 %368 to <2 x bfloat>, !dbg !30 + %370 = extractvalue { i32, i32, i32, i32 } %363, 3, !dbg !30 + %371 = bitcast i32 %370 to <2 x bfloat>, !dbg !30 + %372 = extractelement <2 x bfloat> %365, i64 0, !dbg !30 + %373 = extractelement <2 x bfloat> %365, i64 1, !dbg !30 + %374 = extractelement <2 x bfloat> %367, i64 0, !dbg !30 + %375 = extractelement <2 x bfloat> %367, i64 1, !dbg !30 + %376 = extractelement <2 x bfloat> %369, i64 0, !dbg !30 + %377 = extractelement <2 x bfloat> %369, i64 1, !dbg !30 + %378 = extractelement <2 x bfloat> %371, i64 0, !dbg !30 + %379 = extractelement <2 x bfloat> %371, i64 1, !dbg !30 + %380 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %92, i1 %41) #3, !dbg !30 + %381 = extractvalue { i32, i32, i32, i32 } %380, 0, !dbg !30 + %382 = bitcast i32 %381 to <2 x bfloat>, !dbg !30 + %383 = extractvalue { i32, i32, i32, i32 } %380, 1, !dbg !30 + %384 = bitcast i32 %383 to <2 x bfloat>, !dbg !30 + %385 = extractvalue { i32, i32, i32, i32 } %380, 2, !dbg !30 + %386 = bitcast i32 %385 to <2 x bfloat>, !dbg !30 + %387 = extractvalue { i32, i32, i32, i32 } %380, 3, !dbg !30 + %388 = bitcast i32 %387 to <2 x bfloat>, !dbg !30 + %389 = extractelement <2 x bfloat> %382, i64 0, !dbg !30 + %390 = extractelement <2 x bfloat> %382, i64 1, !dbg !30 + %391 = extractelement <2 x bfloat> %384, i64 0, !dbg !30 + %392 = extractelement <2 x bfloat> %384, i64 1, !dbg !30 + %393 = extractelement <2 x bfloat> %386, i64 0, !dbg !30 + %394 = extractelement <2 x bfloat> %386, i64 1, !dbg !30 + %395 = extractelement <2 x bfloat> %388, i64 0, !dbg !30 + %396 = extractelement <2 x bfloat> %388, i64 1, !dbg !30 + %397 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %93, i1 %42) #3, !dbg !30 + %398 = extractvalue { i32, i32, i32, i32 } %397, 0, !dbg !30 + %399 = bitcast i32 %398 to <2 x bfloat>, !dbg !30 + %400 = extractvalue { i32, i32, i32, i32 } %397, 1, !dbg !30 + %401 = bitcast i32 %400 to <2 x bfloat>, !dbg !30 + %402 = extractvalue { i32, i32, i32, i32 } %397, 2, !dbg !30 + %403 = bitcast i32 %402 to <2 x bfloat>, !dbg !30 + %404 = extractvalue { i32, i32, i32, i32 } %397, 3, !dbg !30 + %405 = bitcast i32 %404 to <2 x bfloat>, !dbg !30 + %406 = extractelement <2 x bfloat> %399, i64 0, !dbg !30 + %407 = extractelement <2 x bfloat> %399, i64 1, !dbg !30 + %408 = extractelement <2 x bfloat> %401, i64 0, !dbg !30 + %409 = extractelement <2 x bfloat> %401, i64 1, !dbg !30 + %410 = extractelement <2 x bfloat> %403, i64 0, !dbg !30 + %411 = extractelement <2 x bfloat> %403, i64 1, !dbg !30 + %412 = extractelement <2 x bfloat> %405, i64 0, !dbg !30 + %413 = extractelement <2 x bfloat> %405, i64 1, !dbg !30 + %414 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %94, i1 %43) #3, !dbg !30 + %415 = extractvalue { i32, i32, i32, i32 } %414, 0, !dbg !30 + %416 = bitcast i32 %415 to <2 x bfloat>, !dbg !30 + %417 = extractvalue { i32, i32, i32, i32 } %414, 1, !dbg !30 + %418 = bitcast i32 %417 to <2 x bfloat>, !dbg !30 + %419 = extractvalue { i32, i32, i32, i32 } %414, 2, !dbg !30 + %420 = bitcast i32 %419 to <2 x bfloat>, !dbg !30 + %421 = extractvalue { i32, i32, i32, i32 } %414, 3, !dbg !30 + %422 = bitcast i32 %421 to <2 x bfloat>, !dbg !30 + %423 = extractelement <2 x bfloat> %416, i64 0, !dbg !30 + %424 = extractelement <2 x bfloat> %416, i64 1, !dbg !30 + %425 = extractelement <2 x bfloat> %418, i64 0, !dbg !30 + %426 = extractelement <2 x bfloat> %418, i64 1, !dbg !30 + %427 = extractelement <2 x bfloat> %420, i64 0, !dbg !30 + %428 = extractelement <2 x bfloat> %420, i64 1, !dbg !30 + %429 = extractelement <2 x bfloat> %422, i64 0, !dbg !30 + %430 = extractelement <2 x bfloat> %422, i64 1, !dbg !30 + %431 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %95, i1 %44) #3, !dbg !30 + %432 = extractvalue { i32, i32, i32, i32 } %431, 0, !dbg !30 + %433 = bitcast i32 %432 to <2 x bfloat>, !dbg !30 + %434 = extractvalue { i32, i32, i32, i32 } %431, 1, !dbg !30 + %435 = bitcast i32 %434 to <2 x bfloat>, !dbg !30 + %436 = extractvalue { i32, i32, i32, i32 } %431, 2, !dbg !30 + %437 = bitcast i32 %436 to <2 x bfloat>, !dbg !30 + %438 = extractvalue { i32, i32, i32, i32 } %431, 3, !dbg !30 + %439 = bitcast i32 %438 to <2 x bfloat>, !dbg !30 + %440 = extractelement <2 x bfloat> %433, i64 0, !dbg !30 + %441 = extractelement <2 x bfloat> %433, i64 1, !dbg !30 + %442 = extractelement <2 x bfloat> %435, i64 0, !dbg !30 + %443 = extractelement <2 x bfloat> %435, i64 1, !dbg !30 + %444 = extractelement <2 x bfloat> %437, i64 0, !dbg !30 + %445 = extractelement <2 x bfloat> %437, i64 1, !dbg !30 + %446 = extractelement <2 x bfloat> %439, i64 0, !dbg !30 + %447 = extractelement <2 x bfloat> %439, i64 1, !dbg !30 + %448 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %96, i1 %45) #3, !dbg !30 + %449 = extractvalue { i32, i32, i32, i32 } %448, 0, !dbg !30 + %450 = bitcast i32 %449 to <2 x bfloat>, !dbg !30 + %451 = extractvalue { i32, i32, i32, i32 } %448, 1, !dbg !30 + %452 = bitcast i32 %451 to <2 x bfloat>, !dbg !30 + %453 = extractvalue { i32, i32, i32, i32 } %448, 2, !dbg !30 + %454 = bitcast i32 %453 to <2 x bfloat>, !dbg !30 + %455 = extractvalue { i32, i32, i32, i32 } %448, 3, !dbg !30 + %456 = bitcast i32 %455 to <2 x bfloat>, !dbg !30 + %457 = extractelement <2 x bfloat> %450, i64 0, !dbg !30 + %458 = extractelement <2 x bfloat> %450, i64 1, !dbg !30 + %459 = extractelement <2 x bfloat> %452, i64 0, !dbg !30 + %460 = extractelement <2 x bfloat> %452, i64 1, !dbg !30 + %461 = extractelement <2 x bfloat> %454, i64 0, !dbg !30 + %462 = extractelement <2 x bfloat> %454, i64 1, !dbg !30 + %463 = extractelement <2 x bfloat> %456, i64 0, !dbg !30 + %464 = extractelement <2 x bfloat> %456, i64 1, !dbg !30 + %465 = fpext bfloat %338 to float, !dbg !31 + %466 = fpext bfloat %339 to float, !dbg !31 + %467 = fpext bfloat %340 to float, !dbg !31 + %468 = fpext bfloat %341 to float, !dbg !31 + %469 = fpext bfloat %342 to float, !dbg !31 + %470 = fpext bfloat %343 to float, !dbg !31 + %471 = fpext bfloat %344 to float, !dbg !31 + %472 = fpext bfloat %345 to float, !dbg !31 + %473 = fpext bfloat %355 to float, !dbg !31 + %474 = fpext bfloat %356 to float, !dbg !31 + %475 = fpext bfloat %357 to float, !dbg !31 + %476 = fpext bfloat %358 to float, !dbg !31 + %477 = fpext bfloat %359 to float, !dbg !31 + %478 = fpext bfloat %360 to float, !dbg !31 + %479 = fpext bfloat %361 to float, !dbg !31 + %480 = fpext bfloat %362 to float, !dbg !31 + %481 = fpext bfloat %372 to float, !dbg !31 + %482 = fpext bfloat %373 to float, !dbg !31 + %483 = fpext bfloat %374 to float, !dbg !31 + %484 = fpext bfloat %375 to float, !dbg !31 + %485 = fpext bfloat %376 to float, !dbg !31 + %486 = fpext bfloat %377 to float, !dbg !31 + %487 = fpext bfloat %378 to float, !dbg !31 + %488 = fpext bfloat %379 to float, !dbg !31 + %489 = fpext bfloat %389 to float, !dbg !31 + %490 = fpext bfloat %390 to float, !dbg !31 + %491 = fpext bfloat %391 to float, !dbg !31 + %492 = fpext bfloat %392 to float, !dbg !31 + %493 = fpext bfloat %393 to float, !dbg !31 + %494 = fpext bfloat %394 to float, !dbg !31 + %495 = fpext bfloat %395 to float, !dbg !31 + %496 = fpext bfloat %396 to float, !dbg !31 + %497 = fpext bfloat %406 to float, !dbg !31 + %498 = fpext bfloat %407 to float, !dbg !31 + %499 = fpext bfloat %408 to float, !dbg !31 + %500 = fpext bfloat %409 to float, !dbg !31 + %501 = fpext bfloat %410 to float, !dbg !31 + %502 = fpext bfloat %411 to float, !dbg !31 + %503 = fpext bfloat %412 to float, !dbg !31 + %504 = fpext bfloat %413 to float, !dbg !31 + %505 = fpext bfloat %423 to float, !dbg !31 + %506 = fpext bfloat %424 to float, !dbg !31 + %507 = fpext bfloat %425 to float, !dbg !31 + %508 = fpext bfloat %426 to float, !dbg !31 + %509 = fpext bfloat %427 to float, !dbg !31 + %510 = fpext bfloat %428 to float, !dbg !31 + %511 = fpext bfloat %429 to float, !dbg !31 + %512 = fpext bfloat %430 to float, !dbg !31 + %513 = fpext bfloat %440 to float, !dbg !31 + %514 = fpext bfloat %441 to float, !dbg !31 + %515 = fpext bfloat %442 to float, !dbg !31 + %516 = fpext bfloat %443 to float, !dbg !31 + %517 = fpext bfloat %444 to float, !dbg !31 + %518 = fpext bfloat %445 to float, !dbg !31 + %519 = fpext bfloat %446 to float, !dbg !31 + %520 = fpext bfloat %447 to float, !dbg !31 + %521 = fpext bfloat %457 to float, !dbg !31 + %522 = fpext bfloat %458 to float, !dbg !31 + %523 = fpext bfloat %459 to float, !dbg !31 + %524 = fpext bfloat %460 to float, !dbg !31 + %525 = fpext bfloat %461 to float, !dbg !31 + %526 = fpext bfloat %462 to float, !dbg !31 + %527 = fpext bfloat %463 to float, !dbg !31 + %528 = fpext bfloat %464 to float, !dbg !31 + %529 = fmul float %314, %465, !dbg !32 + %530 = fmul float %314, %466, !dbg !32 + %531 = fmul float %314, %467, !dbg !32 + %532 = fmul float %314, %468, !dbg !32 + %533 = fmul float %314, %469, !dbg !32 + %534 = fmul float %314, %470, !dbg !32 + %535 = fmul float %314, %471, !dbg !32 + %536 = fmul float %314, %472, !dbg !32 + %537 = fmul float %316, %473, !dbg !32 + %538 = fmul float %316, %474, !dbg !32 + %539 = fmul float %316, %475, !dbg !32 + %540 = fmul float %316, %476, !dbg !32 + %541 = fmul float %316, %477, !dbg !32 + %542 = fmul float %316, %478, !dbg !32 + %543 = fmul float %316, %479, !dbg !32 + %544 = fmul float %316, %480, !dbg !32 + %545 = fmul float %318, %481, !dbg !32 + %546 = fmul float %318, %482, !dbg !32 + %547 = fmul float %318, %483, !dbg !32 + %548 = fmul float %318, %484, !dbg !32 + %549 = fmul float %318, %485, !dbg !32 + %550 = fmul float %318, %486, !dbg !32 + %551 = fmul float %318, %487, !dbg !32 + %552 = fmul float %318, %488, !dbg !32 + %553 = fmul float %320, %489, !dbg !32 + %554 = fmul float %320, %490, !dbg !32 + %555 = fmul float %320, %491, !dbg !32 + %556 = fmul float %320, %492, !dbg !32 + %557 = fmul float %320, %493, !dbg !32 + %558 = fmul float %320, %494, !dbg !32 + %559 = fmul float %320, %495, !dbg !32 + %560 = fmul float %320, %496, !dbg !32 + %561 = fmul float %322, %497, !dbg !32 + %562 = fmul float %322, %498, !dbg !32 + %563 = fmul float %322, %499, !dbg !32 + %564 = fmul float %322, %500, !dbg !32 + %565 = fmul float %322, %501, !dbg !32 + %566 = fmul float %322, %502, !dbg !32 + %567 = fmul float %322, %503, !dbg !32 + %568 = fmul float %322, %504, !dbg !32 + %569 = fmul float %324, %505, !dbg !32 + %570 = fmul float %324, %506, !dbg !32 + %571 = fmul float %324, %507, !dbg !32 + %572 = fmul float %324, %508, !dbg !32 + %573 = fmul float %324, %509, !dbg !32 + %574 = fmul float %324, %510, !dbg !32 + %575 = fmul float %324, %511, !dbg !32 + %576 = fmul float %324, %512, !dbg !32 + %577 = fmul float %326, %513, !dbg !32 + %578 = fmul float %326, %514, !dbg !32 + %579 = fmul float %326, %515, !dbg !32 + %580 = fmul float %326, %516, !dbg !32 + %581 = fmul float %326, %517, !dbg !32 + %582 = fmul float %326, %518, !dbg !32 + %583 = fmul float %326, %519, !dbg !32 + %584 = fmul float %326, %520, !dbg !32 + %585 = fmul float %328, %521, !dbg !32 + %586 = fmul float %328, %522, !dbg !32 + %587 = fmul float %328, %523, !dbg !32 + %588 = fmul float %328, %524, !dbg !32 + %589 = fmul float %328, %525, !dbg !32 + %590 = fmul float %328, %526, !dbg !32 + %591 = fmul float %328, %527, !dbg !32 + %592 = fmul float %328, %528, !dbg !32 + %593 = fmul float %249, %465, !dbg !33 + %594 = fmul float %250, %466, !dbg !33 + %595 = fmul float %251, %467, !dbg !33 + %596 = fmul float %252, %468, !dbg !33 + %597 = fmul float %253, %469, !dbg !33 + %598 = fmul float %254, %470, !dbg !33 + %599 = fmul float %255, %471, !dbg !33 + %600 = fmul float %256, %472, !dbg !33 + %601 = fmul float %257, %473, !dbg !33 + %602 = fmul float %258, %474, !dbg !33 + %603 = fmul float %259, %475, !dbg !33 + %604 = fmul float %260, %476, !dbg !33 + %605 = fmul float %261, %477, !dbg !33 + %606 = fmul float %262, %478, !dbg !33 + %607 = fmul float %263, %479, !dbg !33 + %608 = fmul float %264, %480, !dbg !33 + %609 = fmul float %265, %481, !dbg !33 + %610 = fmul float %266, %482, !dbg !33 + %611 = fmul float %267, %483, !dbg !33 + %612 = fmul float %268, %484, !dbg !33 + %613 = fmul float %269, %485, !dbg !33 + %614 = fmul float %270, %486, !dbg !33 + %615 = fmul float %271, %487, !dbg !33 + %616 = fmul float %272, %488, !dbg !33 + %617 = fmul float %273, %489, !dbg !33 + %618 = fmul float %274, %490, !dbg !33 + %619 = fmul float %275, %491, !dbg !33 + %620 = fmul float %276, %492, !dbg !33 + %621 = fmul float %277, %493, !dbg !33 + %622 = fmul float %278, %494, !dbg !33 + %623 = fmul float %279, %495, !dbg !33 + %624 = fmul float %280, %496, !dbg !33 + %625 = fmul float %281, %497, !dbg !33 + %626 = fmul float %282, %498, !dbg !33 + %627 = fmul float %283, %499, !dbg !33 + %628 = fmul float %284, %500, !dbg !33 + %629 = fmul float %285, %501, !dbg !33 + %630 = fmul float %286, %502, !dbg !33 + %631 = fmul float %287, %503, !dbg !33 + %632 = fmul float %288, %504, !dbg !33 + %633 = fmul float %289, %505, !dbg !33 + %634 = fmul float %290, %506, !dbg !33 + %635 = fmul float %291, %507, !dbg !33 + %636 = fmul float %292, %508, !dbg !33 + %637 = fmul float %293, %509, !dbg !33 + %638 = fmul float %294, %510, !dbg !33 + %639 = fmul float %295, %511, !dbg !33 + %640 = fmul float %296, %512, !dbg !33 + %641 = fmul float %297, %513, !dbg !33 + %642 = fmul float %298, %514, !dbg !33 + %643 = fmul float %299, %515, !dbg !33 + %644 = fmul float %300, %516, !dbg !33 + %645 = fmul float %301, %517, !dbg !33 + %646 = fmul float %302, %518, !dbg !33 + %647 = fmul float %303, %519, !dbg !33 + %648 = fmul float %304, %520, !dbg !33 + %649 = fmul float %305, %521, !dbg !33 + %650 = fmul float %306, %522, !dbg !33 + %651 = fmul float %307, %523, !dbg !33 + %652 = fmul float %308, %524, !dbg !33 + %653 = fmul float %309, %525, !dbg !33 + %654 = fmul float %310, %526, !dbg !33 + %655 = fmul float %311, %527, !dbg !33 + %656 = fmul float %312, %528, !dbg !33 + %657 = fadd float %593, %594, !dbg !34 + %658 = fadd float %595, %657, !dbg !34 + %659 = fadd float %596, %658, !dbg !34 + %660 = fadd float %597, %659, !dbg !34 + %661 = fadd float %598, %660, !dbg !34 + %662 = fadd float %599, %661, !dbg !34 + %663 = fadd float %600, %662, !dbg !34 + %664 = fadd float %601, %602, !dbg !34 + %665 = fadd float %603, %664, !dbg !34 + %666 = fadd float %604, %665, !dbg !34 + %667 = fadd float %605, %666, !dbg !34 + %668 = fadd float %606, %667, !dbg !34 + %669 = fadd float %607, %668, !dbg !34 + %670 = fadd float %608, %669, !dbg !34 + %671 = fadd float %609, %610, !dbg !34 + %672 = fadd float %611, %671, !dbg !34 + %673 = fadd float %612, %672, !dbg !34 + %674 = fadd float %613, %673, !dbg !34 + %675 = fadd float %614, %674, !dbg !34 + %676 = fadd float %615, %675, !dbg !34 + %677 = fadd float %616, %676, !dbg !34 + %678 = fadd float %617, %618, !dbg !34 + %679 = fadd float %619, %678, !dbg !34 + %680 = fadd float %620, %679, !dbg !34 + %681 = fadd float %621, %680, !dbg !34 + %682 = fadd float %622, %681, !dbg !34 + %683 = fadd float %623, %682, !dbg !34 + %684 = fadd float %624, %683, !dbg !34 + %685 = fadd float %625, %626, !dbg !34 + %686 = fadd float %627, %685, !dbg !34 + %687 = fadd float %628, %686, !dbg !34 + %688 = fadd float %629, %687, !dbg !34 + %689 = fadd float %630, %688, !dbg !34 + %690 = fadd float %631, %689, !dbg !34 + %691 = fadd float %632, %690, !dbg !34 + %692 = fadd float %633, %634, !dbg !34 + %693 = fadd float %635, %692, !dbg !34 + %694 = fadd float %636, %693, !dbg !34 + %695 = fadd float %637, %694, !dbg !34 + %696 = fadd float %638, %695, !dbg !34 + %697 = fadd float %639, %696, !dbg !34 + %698 = fadd float %640, %697, !dbg !34 + %699 = fadd float %641, %642, !dbg !34 + %700 = fadd float %643, %699, !dbg !34 + %701 = fadd float %644, %700, !dbg !34 + %702 = fadd float %645, %701, !dbg !34 + %703 = fadd float %646, %702, !dbg !34 + %704 = fadd float %647, %703, !dbg !34 + %705 = fadd float %648, %704, !dbg !34 + %706 = fadd float %649, %650, !dbg !34 + %707 = fadd float %651, %706, !dbg !34 + %708 = fadd float %652, %707, !dbg !34 + %709 = fadd float %653, %708, !dbg !34 + %710 = fadd float %654, %709, !dbg !34 + %711 = fadd float %655, %710, !dbg !34 + %712 = fadd float %656, %711, !dbg !34 + %713 = bitcast float %663 to i32, !dbg !37 + %714 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %713, i32 8, i32 31), !dbg !37 + %715 = bitcast i32 %714 to float, !dbg !37 + %716 = fadd float %663, %715, !dbg !34 + %717 = bitcast float %716 to i32, !dbg !37 + %718 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %717, i32 4, i32 31), !dbg !37 + %719 = bitcast i32 %718 to float, !dbg !37 + %720 = fadd float %716, %719, !dbg !34 + %721 = bitcast float %720 to i32, !dbg !37 + %722 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %721, i32 2, i32 31), !dbg !37 + %723 = bitcast i32 %722 to float, !dbg !37 + %724 = fadd float %720, %723, !dbg !34 + %725 = bitcast float %724 to i32, !dbg !37 + %726 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %725, i32 1, i32 31), !dbg !37 + %727 = bitcast i32 %726 to float, !dbg !37 + %728 = fadd float %724, %727, !dbg !34 + %729 = bitcast float %670 to i32, !dbg !37 + %730 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %729, i32 8, i32 31), !dbg !37 + %731 = bitcast i32 %730 to float, !dbg !37 + %732 = fadd float %670, %731, !dbg !34 + %733 = bitcast float %732 to i32, !dbg !37 + %734 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %733, i32 4, i32 31), !dbg !37 + %735 = bitcast i32 %734 to float, !dbg !37 + %736 = fadd float %732, %735, !dbg !34 + %737 = bitcast float %736 to i32, !dbg !37 + %738 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %737, i32 2, i32 31), !dbg !37 + %739 = bitcast i32 %738 to float, !dbg !37 + %740 = fadd float %736, %739, !dbg !34 + %741 = bitcast float %740 to i32, !dbg !37 + %742 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %741, i32 1, i32 31), !dbg !37 + %743 = bitcast i32 %742 to float, !dbg !37 + %744 = fadd float %740, %743, !dbg !34 + %745 = bitcast float %677 to i32, !dbg !37 + %746 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %745, i32 8, i32 31), !dbg !37 + %747 = bitcast i32 %746 to float, !dbg !37 + %748 = fadd float %677, %747, !dbg !34 + %749 = bitcast float %748 to i32, !dbg !37 + %750 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %749, i32 4, i32 31), !dbg !37 + %751 = bitcast i32 %750 to float, !dbg !37 + %752 = fadd float %748, %751, !dbg !34 + %753 = bitcast float %752 to i32, !dbg !37 + %754 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %753, i32 2, i32 31), !dbg !37 + %755 = bitcast i32 %754 to float, !dbg !37 + %756 = fadd float %752, %755, !dbg !34 + %757 = bitcast float %756 to i32, !dbg !37 + %758 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %757, i32 1, i32 31), !dbg !37 + %759 = bitcast i32 %758 to float, !dbg !37 + %760 = fadd float %756, %759, !dbg !34 + %761 = bitcast float %684 to i32, !dbg !37 + %762 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %761, i32 8, i32 31), !dbg !37 + %763 = bitcast i32 %762 to float, !dbg !37 + %764 = fadd float %684, %763, !dbg !34 + %765 = bitcast float %764 to i32, !dbg !37 + %766 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %765, i32 4, i32 31), !dbg !37 + %767 = bitcast i32 %766 to float, !dbg !37 + %768 = fadd float %764, %767, !dbg !34 + %769 = bitcast float %768 to i32, !dbg !37 + %770 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %769, i32 2, i32 31), !dbg !37 + %771 = bitcast i32 %770 to float, !dbg !37 + %772 = fadd float %768, %771, !dbg !34 + %773 = bitcast float %772 to i32, !dbg !37 + %774 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %773, i32 1, i32 31), !dbg !37 + %775 = bitcast i32 %774 to float, !dbg !37 + %776 = fadd float %772, %775, !dbg !34 + %777 = bitcast float %691 to i32, !dbg !37 + %778 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %777, i32 8, i32 31), !dbg !37 + %779 = bitcast i32 %778 to float, !dbg !37 + %780 = fadd float %691, %779, !dbg !34 + %781 = bitcast float %780 to i32, !dbg !37 + %782 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %781, i32 4, i32 31), !dbg !37 + %783 = bitcast i32 %782 to float, !dbg !37 + %784 = fadd float %780, %783, !dbg !34 + %785 = bitcast float %784 to i32, !dbg !37 + %786 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %785, i32 2, i32 31), !dbg !37 + %787 = bitcast i32 %786 to float, !dbg !37 + %788 = fadd float %784, %787, !dbg !34 + %789 = bitcast float %788 to i32, !dbg !37 + %790 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %789, i32 1, i32 31), !dbg !37 + %791 = bitcast i32 %790 to float, !dbg !37 + %792 = fadd float %788, %791, !dbg !34 + %793 = bitcast float %698 to i32, !dbg !37 + %794 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %793, i32 8, i32 31), !dbg !37 + %795 = bitcast i32 %794 to float, !dbg !37 + %796 = fadd float %698, %795, !dbg !34 + %797 = bitcast float %796 to i32, !dbg !37 + %798 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %797, i32 4, i32 31), !dbg !37 + %799 = bitcast i32 %798 to float, !dbg !37 + %800 = fadd float %796, %799, !dbg !34 + %801 = bitcast float %800 to i32, !dbg !37 + %802 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %801, i32 2, i32 31), !dbg !37 + %803 = bitcast i32 %802 to float, !dbg !37 + %804 = fadd float %800, %803, !dbg !34 + %805 = bitcast float %804 to i32, !dbg !37 + %806 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %805, i32 1, i32 31), !dbg !37 + %807 = bitcast i32 %806 to float, !dbg !37 + %808 = fadd float %804, %807, !dbg !34 + %809 = bitcast float %705 to i32, !dbg !37 + %810 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %809, i32 8, i32 31), !dbg !37 + %811 = bitcast i32 %810 to float, !dbg !37 + %812 = fadd float %705, %811, !dbg !34 + %813 = bitcast float %812 to i32, !dbg !37 + %814 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %813, i32 4, i32 31), !dbg !37 + %815 = bitcast i32 %814 to float, !dbg !37 + %816 = fadd float %812, %815, !dbg !34 + %817 = bitcast float %816 to i32, !dbg !37 + %818 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %817, i32 2, i32 31), !dbg !37 + %819 = bitcast i32 %818 to float, !dbg !37 + %820 = fadd float %816, %819, !dbg !34 + %821 = bitcast float %820 to i32, !dbg !37 + %822 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %821, i32 1, i32 31), !dbg !37 + %823 = bitcast i32 %822 to float, !dbg !37 + %824 = fadd float %820, %823, !dbg !34 + %825 = bitcast float %712 to i32, !dbg !37 + %826 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %825, i32 8, i32 31), !dbg !37 + %827 = bitcast i32 %826 to float, !dbg !37 + %828 = fadd float %712, %827, !dbg !34 + %829 = bitcast float %828 to i32, !dbg !37 + %830 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %829, i32 4, i32 31), !dbg !37 + %831 = bitcast i32 %830 to float, !dbg !37 + %832 = fadd float %828, %831, !dbg !34 + %833 = bitcast float %832 to i32, !dbg !37 + %834 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %833, i32 2, i32 31), !dbg !37 + %835 = bitcast i32 %834 to float, !dbg !37 + %836 = fadd float %832, %835, !dbg !34 + %837 = bitcast float %836 to i32, !dbg !37 + %838 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %837, i32 1, i32 31), !dbg !37 + %839 = bitcast i32 %838 to float, !dbg !37 + %840 = fadd float %836, %839, !dbg !34 + %841 = fmul float %728, %249, !dbg !40 + %842 = fmul float %728, %250, !dbg !40 + %843 = fmul float %728, %251, !dbg !40 + %844 = fmul float %728, %252, !dbg !40 + %845 = fmul float %728, %253, !dbg !40 + %846 = fmul float %728, %254, !dbg !40 + %847 = fmul float %728, %255, !dbg !40 + %848 = fmul float %728, %256, !dbg !40 + %849 = fmul float %744, %257, !dbg !40 + %850 = fmul float %744, %258, !dbg !40 + %851 = fmul float %744, %259, !dbg !40 + %852 = fmul float %744, %260, !dbg !40 + %853 = fmul float %744, %261, !dbg !40 + %854 = fmul float %744, %262, !dbg !40 + %855 = fmul float %744, %263, !dbg !40 + %856 = fmul float %744, %264, !dbg !40 + %857 = fmul float %760, %265, !dbg !40 + %858 = fmul float %760, %266, !dbg !40 + %859 = fmul float %760, %267, !dbg !40 + %860 = fmul float %760, %268, !dbg !40 + %861 = fmul float %760, %269, !dbg !40 + %862 = fmul float %760, %270, !dbg !40 + %863 = fmul float %760, %271, !dbg !40 + %864 = fmul float %760, %272, !dbg !40 + %865 = fmul float %776, %273, !dbg !40 + %866 = fmul float %776, %274, !dbg !40 + %867 = fmul float %776, %275, !dbg !40 + %868 = fmul float %776, %276, !dbg !40 + %869 = fmul float %776, %277, !dbg !40 + %870 = fmul float %776, %278, !dbg !40 + %871 = fmul float %776, %279, !dbg !40 + %872 = fmul float %776, %280, !dbg !40 + %873 = fmul float %792, %281, !dbg !40 + %874 = fmul float %792, %282, !dbg !40 + %875 = fmul float %792, %283, !dbg !40 + %876 = fmul float %792, %284, !dbg !40 + %877 = fmul float %792, %285, !dbg !40 + %878 = fmul float %792, %286, !dbg !40 + %879 = fmul float %792, %287, !dbg !40 + %880 = fmul float %792, %288, !dbg !40 + %881 = fmul float %808, %289, !dbg !40 + %882 = fmul float %808, %290, !dbg !40 + %883 = fmul float %808, %291, !dbg !40 + %884 = fmul float %808, %292, !dbg !40 + %885 = fmul float %808, %293, !dbg !40 + %886 = fmul float %808, %294, !dbg !40 + %887 = fmul float %808, %295, !dbg !40 + %888 = fmul float %808, %296, !dbg !40 + %889 = fmul float %824, %297, !dbg !40 + %890 = fmul float %824, %298, !dbg !40 + %891 = fmul float %824, %299, !dbg !40 + %892 = fmul float %824, %300, !dbg !40 + %893 = fmul float %824, %301, !dbg !40 + %894 = fmul float %824, %302, !dbg !40 + %895 = fmul float %824, %303, !dbg !40 + %896 = fmul float %824, %304, !dbg !40 + %897 = fmul float %840, %305, !dbg !40 + %898 = fmul float %840, %306, !dbg !40 + %899 = fmul float %840, %307, !dbg !40 + %900 = fmul float %840, %308, !dbg !40 + %901 = fmul float %840, %309, !dbg !40 + %902 = fmul float %840, %310, !dbg !40 + %903 = fmul float %840, %311, !dbg !40 + %904 = fmul float %840, %312, !dbg !40 + %905 = fmul float %841, %314, !dbg !41 + %906 = fmul float %842, %314, !dbg !41 + %907 = fmul float %843, %314, !dbg !41 + %908 = fmul float %844, %314, !dbg !41 + %909 = fmul float %845, %314, !dbg !41 + %910 = fmul float %846, %314, !dbg !41 + %911 = fmul float %847, %314, !dbg !41 + %912 = fmul float %848, %314, !dbg !41 + %913 = fmul float %849, %316, !dbg !41 + %914 = fmul float %850, %316, !dbg !41 + %915 = fmul float %851, %316, !dbg !41 + %916 = fmul float %852, %316, !dbg !41 + %917 = fmul float %853, %316, !dbg !41 + %918 = fmul float %854, %316, !dbg !41 + %919 = fmul float %855, %316, !dbg !41 + %920 = fmul float %856, %316, !dbg !41 + %921 = fmul float %857, %318, !dbg !41 + %922 = fmul float %858, %318, !dbg !41 + %923 = fmul float %859, %318, !dbg !41 + %924 = fmul float %860, %318, !dbg !41 + %925 = fmul float %861, %318, !dbg !41 + %926 = fmul float %862, %318, !dbg !41 + %927 = fmul float %863, %318, !dbg !41 + %928 = fmul float %864, %318, !dbg !41 + %929 = fmul float %865, %320, !dbg !41 + %930 = fmul float %866, %320, !dbg !41 + %931 = fmul float %867, %320, !dbg !41 + %932 = fmul float %868, %320, !dbg !41 + %933 = fmul float %869, %320, !dbg !41 + %934 = fmul float %870, %320, !dbg !41 + %935 = fmul float %871, %320, !dbg !41 + %936 = fmul float %872, %320, !dbg !41 + %937 = fmul float %873, %322, !dbg !41 + %938 = fmul float %874, %322, !dbg !41 + %939 = fmul float %875, %322, !dbg !41 + %940 = fmul float %876, %322, !dbg !41 + %941 = fmul float %877, %322, !dbg !41 + %942 = fmul float %878, %322, !dbg !41 + %943 = fmul float %879, %322, !dbg !41 + %944 = fmul float %880, %322, !dbg !41 + %945 = fmul float %881, %324, !dbg !41 + %946 = fmul float %882, %324, !dbg !41 + %947 = fmul float %883, %324, !dbg !41 + %948 = fmul float %884, %324, !dbg !41 + %949 = fmul float %885, %324, !dbg !41 + %950 = fmul float %886, %324, !dbg !41 + %951 = fmul float %887, %324, !dbg !41 + %952 = fmul float %888, %324, !dbg !41 + %953 = fmul float %889, %326, !dbg !41 + %954 = fmul float %890, %326, !dbg !41 + %955 = fmul float %891, %326, !dbg !41 + %956 = fmul float %892, %326, !dbg !41 + %957 = fmul float %893, %326, !dbg !41 + %958 = fmul float %894, %326, !dbg !41 + %959 = fmul float %895, %326, !dbg !41 + %960 = fmul float %896, %326, !dbg !41 + %961 = fmul float %897, %328, !dbg !41 + %962 = fmul float %898, %328, !dbg !41 + %963 = fmul float %899, %328, !dbg !41 + %964 = fmul float %900, %328, !dbg !41 + %965 = fmul float %901, %328, !dbg !41 + %966 = fmul float %902, %328, !dbg !41 + %967 = fmul float %903, %328, !dbg !41 + %968 = fmul float %904, %328, !dbg !41 + %969 = fsub float %529, %905, !dbg !42 + %970 = fsub float %530, %906, !dbg !42 + %971 = fsub float %531, %907, !dbg !42 + %972 = fsub float %532, %908, !dbg !42 + %973 = fsub float %533, %909, !dbg !42 + %974 = fsub float %534, %910, !dbg !42 + %975 = fsub float %535, %911, !dbg !42 + %976 = fsub float %536, %912, !dbg !42 + %977 = fsub float %537, %913, !dbg !42 + %978 = fsub float %538, %914, !dbg !42 + %979 = fsub float %539, %915, !dbg !42 + %980 = fsub float %540, %916, !dbg !42 + %981 = fsub float %541, %917, !dbg !42 + %982 = fsub float %542, %918, !dbg !42 + %983 = fsub float %543, %919, !dbg !42 + %984 = fsub float %544, %920, !dbg !42 + %985 = fsub float %545, %921, !dbg !42 + %986 = fsub float %546, %922, !dbg !42 + %987 = fsub float %547, %923, !dbg !42 + %988 = fsub float %548, %924, !dbg !42 + %989 = fsub float %549, %925, !dbg !42 + %990 = fsub float %550, %926, !dbg !42 + %991 = fsub float %551, %927, !dbg !42 + %992 = fsub float %552, %928, !dbg !42 + %993 = fsub float %553, %929, !dbg !42 + %994 = fsub float %554, %930, !dbg !42 + %995 = fsub float %555, %931, !dbg !42 + %996 = fsub float %556, %932, !dbg !42 + %997 = fsub float %557, %933, !dbg !42 + %998 = fsub float %558, %934, !dbg !42 + %999 = fsub float %559, %935, !dbg !42 + %1000 = fsub float %560, %936, !dbg !42 + %1001 = fsub float %561, %937, !dbg !42 + %1002 = fsub float %562, %938, !dbg !42 + %1003 = fsub float %563, %939, !dbg !42 + %1004 = fsub float %564, %940, !dbg !42 + %1005 = fsub float %565, %941, !dbg !42 + %1006 = fsub float %566, %942, !dbg !42 + %1007 = fsub float %567, %943, !dbg !42 + %1008 = fsub float %568, %944, !dbg !42 + %1009 = fsub float %569, %945, !dbg !42 + %1010 = fsub float %570, %946, !dbg !42 + %1011 = fsub float %571, %947, !dbg !42 + %1012 = fsub float %572, %948, !dbg !42 + %1013 = fsub float %573, %949, !dbg !42 + %1014 = fsub float %574, %950, !dbg !42 + %1015 = fsub float %575, %951, !dbg !42 + %1016 = fsub float %576, %952, !dbg !42 + %1017 = insertelement <2 x float> poison, float %577, i64 0, !dbg !42 + %1018 = insertelement <2 x float> %1017, float %578, i64 1, !dbg !42 + %1019 = insertelement <2 x float> poison, float %953, i64 0, !dbg !42 + %1020 = insertelement <2 x float> %1019, float %954, i64 1, !dbg !42 + %1021 = fsub <2 x float> %1018, %1020, !dbg !42 + %1022 = insertelement <2 x float> poison, float %579, i64 0, !dbg !42 + %1023 = insertelement <2 x float> %1022, float %580, i64 1, !dbg !42 + %1024 = insertelement <2 x float> poison, float %955, i64 0, !dbg !42 + %1025 = insertelement <2 x float> %1024, float %956, i64 1, !dbg !42 + %1026 = fsub <2 x float> %1023, %1025, !dbg !42 + %1027 = insertelement <2 x float> poison, float %581, i64 0, !dbg !42 + %1028 = insertelement <2 x float> %1027, float %582, i64 1, !dbg !42 + %1029 = insertelement <2 x float> poison, float %957, i64 0, !dbg !42 + %1030 = insertelement <2 x float> %1029, float %958, i64 1, !dbg !42 + %1031 = fsub <2 x float> %1028, %1030, !dbg !42 + %1032 = insertelement <2 x float> poison, float %583, i64 0, !dbg !42 + %1033 = insertelement <2 x float> %1032, float %584, i64 1, !dbg !42 + %1034 = insertelement <2 x float> poison, float %959, i64 0, !dbg !42 + %1035 = insertelement <2 x float> %1034, float %960, i64 1, !dbg !42 + %1036 = fsub <2 x float> %1033, %1035, !dbg !42 + %1037 = insertelement <2 x float> poison, float %585, i64 0, !dbg !42 + %1038 = insertelement <2 x float> %1037, float %586, i64 1, !dbg !42 + %1039 = insertelement <2 x float> poison, float %961, i64 0, !dbg !42 + %1040 = insertelement <2 x float> %1039, float %962, i64 1, !dbg !42 + %1041 = fsub <2 x float> %1038, %1040, !dbg !42 + %1042 = insertelement <2 x float> poison, float %587, i64 0, !dbg !42 + %1043 = insertelement <2 x float> %1042, float %588, i64 1, !dbg !42 + %1044 = insertelement <2 x float> poison, float %963, i64 0, !dbg !42 + %1045 = insertelement <2 x float> %1044, float %964, i64 1, !dbg !42 + %1046 = fsub <2 x float> %1043, %1045, !dbg !42 + %1047 = insertelement <2 x float> poison, float %589, i64 0, !dbg !42 + %1048 = insertelement <2 x float> %1047, float %590, i64 1, !dbg !42 + %1049 = insertelement <2 x float> poison, float %965, i64 0, !dbg !42 + %1050 = insertelement <2 x float> %1049, float %966, i64 1, !dbg !42 + %1051 = fsub <2 x float> %1048, %1050, !dbg !42 + %1052 = insertelement <2 x float> poison, float %591, i64 0, !dbg !42 + %1053 = insertelement <2 x float> %1052, float %592, i64 1, !dbg !42 + %1054 = insertelement <2 x float> poison, float %967, i64 0, !dbg !42 + %1055 = insertelement <2 x float> %1054, float %968, i64 1, !dbg !42 + %1056 = fsub <2 x float> %1053, %1055, !dbg !42 + %1057 = fptrunc float %969 to bfloat, !dbg !43 + %1058 = fptrunc float %970 to bfloat, !dbg !43 + %1059 = fptrunc float %971 to bfloat, !dbg !43 + %1060 = fptrunc float %972 to bfloat, !dbg !43 + %1061 = fptrunc float %973 to bfloat, !dbg !43 + %1062 = fptrunc float %974 to bfloat, !dbg !43 + %1063 = fptrunc float %975 to bfloat, !dbg !43 + %1064 = fptrunc float %976 to bfloat, !dbg !43 + %1065 = fptrunc float %977 to bfloat, !dbg !43 + %1066 = fptrunc float %978 to bfloat, !dbg !43 + %1067 = fptrunc float %979 to bfloat, !dbg !43 + %1068 = fptrunc float %980 to bfloat, !dbg !43 + %1069 = fptrunc float %981 to bfloat, !dbg !43 + %1070 = fptrunc float %982 to bfloat, !dbg !43 + %1071 = fptrunc float %983 to bfloat, !dbg !43 + %1072 = fptrunc float %984 to bfloat, !dbg !43 + %1073 = fptrunc float %985 to bfloat, !dbg !43 + %1074 = fptrunc float %986 to bfloat, !dbg !43 + %1075 = fptrunc float %987 to bfloat, !dbg !43 + %1076 = fptrunc float %988 to bfloat, !dbg !43 + %1077 = fptrunc float %989 to bfloat, !dbg !43 + %1078 = fptrunc float %990 to bfloat, !dbg !43 + %1079 = fptrunc float %991 to bfloat, !dbg !43 + %1080 = fptrunc float %992 to bfloat, !dbg !43 + %1081 = fptrunc float %993 to bfloat, !dbg !43 + %1082 = fptrunc float %994 to bfloat, !dbg !43 + %1083 = fptrunc float %995 to bfloat, !dbg !43 + %1084 = fptrunc float %996 to bfloat, !dbg !43 + %1085 = fptrunc float %997 to bfloat, !dbg !43 + %1086 = fptrunc float %998 to bfloat, !dbg !43 + %1087 = fptrunc float %999 to bfloat, !dbg !43 + %1088 = fptrunc float %1000 to bfloat, !dbg !43 + %1089 = fptrunc float %1001 to bfloat, !dbg !43 + %1090 = fptrunc float %1002 to bfloat, !dbg !43 + %1091 = fptrunc float %1003 to bfloat, !dbg !43 + %1092 = fptrunc float %1004 to bfloat, !dbg !43 + %1093 = fptrunc float %1005 to bfloat, !dbg !43 + %1094 = fptrunc float %1006 to bfloat, !dbg !43 + %1095 = fptrunc float %1007 to bfloat, !dbg !43 + %1096 = fptrunc float %1008 to bfloat, !dbg !43 + %1097 = fptrunc float %1009 to bfloat, !dbg !43 + %1098 = fptrunc float %1010 to bfloat, !dbg !43 + %1099 = fptrunc float %1011 to bfloat, !dbg !43 + %1100 = fptrunc float %1012 to bfloat, !dbg !43 + %1101 = fptrunc float %1013 to bfloat, !dbg !43 + %1102 = fptrunc float %1014 to bfloat, !dbg !43 + %1103 = fptrunc float %1015 to bfloat, !dbg !43 + %1104 = fptrunc float %1016 to bfloat, !dbg !43 + %1105 = fptrunc <2 x float> %1021 to <2 x bfloat>, !dbg !43 + %1106 = fptrunc <2 x float> %1026 to <2 x bfloat>, !dbg !43 + %1107 = fptrunc <2 x float> %1031 to <2 x bfloat>, !dbg !43 + %1108 = fptrunc <2 x float> %1036 to <2 x bfloat>, !dbg !43 + %1109 = fptrunc <2 x float> %1041 to <2 x bfloat>, !dbg !43 + %1110 = fptrunc <2 x float> %1046 to <2 x bfloat>, !dbg !43 + %1111 = fptrunc <2 x float> %1051 to <2 x bfloat>, !dbg !43 + %1112 = fptrunc <2 x float> %1056 to <2 x bfloat>, !dbg !43 + %1113 = insertelement <2 x bfloat> poison, bfloat %1057, i64 0, !dbg !44 + %1114 = insertelement <2 x bfloat> %1113, bfloat %1058, i64 1, !dbg !44 + %1115 = bitcast <2 x bfloat> %1114 to i32, !dbg !44 + %1116 = insertelement <2 x bfloat> poison, bfloat %1059, i64 0, !dbg !44 + %1117 = insertelement <2 x bfloat> %1116, bfloat %1060, i64 1, !dbg !44 + %1118 = bitcast <2 x bfloat> %1117 to i32, !dbg !44 + %1119 = insertelement <2 x bfloat> poison, bfloat %1061, i64 0, !dbg !44 + %1120 = insertelement <2 x bfloat> %1119, bfloat %1062, i64 1, !dbg !44 + %1121 = bitcast <2 x bfloat> %1120 to i32, !dbg !44 + %1122 = insertelement <2 x bfloat> poison, bfloat %1063, i64 0, !dbg !44 + %1123 = insertelement <2 x bfloat> %1122, bfloat %1064, i64 1, !dbg !44 + %1124 = bitcast <2 x bfloat> %1123 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1115, i32 %1118, i32 %1121, i32 %1124, ptr addrspace(1) %105, i1 %38) #3, !dbg !44 + %1125 = insertelement <2 x bfloat> poison, bfloat %1065, i64 0, !dbg !44 + %1126 = insertelement <2 x bfloat> %1125, bfloat %1066, i64 1, !dbg !44 + %1127 = bitcast <2 x bfloat> %1126 to i32, !dbg !44 + %1128 = insertelement <2 x bfloat> poison, bfloat %1067, i64 0, !dbg !44 + %1129 = insertelement <2 x bfloat> %1128, bfloat %1068, i64 1, !dbg !44 + %1130 = bitcast <2 x bfloat> %1129 to i32, !dbg !44 + %1131 = insertelement <2 x bfloat> poison, bfloat %1069, i64 0, !dbg !44 + %1132 = insertelement <2 x bfloat> %1131, bfloat %1070, i64 1, !dbg !44 + %1133 = bitcast <2 x bfloat> %1132 to i32, !dbg !44 + %1134 = insertelement <2 x bfloat> poison, bfloat %1071, i64 0, !dbg !44 + %1135 = insertelement <2 x bfloat> %1134, bfloat %1072, i64 1, !dbg !44 + %1136 = bitcast <2 x bfloat> %1135 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1127, i32 %1130, i32 %1133, i32 %1136, ptr addrspace(1) %106, i1 %39) #3, !dbg !44 + %1137 = insertelement <2 x bfloat> poison, bfloat %1073, i64 0, !dbg !44 + %1138 = insertelement <2 x bfloat> %1137, bfloat %1074, i64 1, !dbg !44 + %1139 = bitcast <2 x bfloat> %1138 to i32, !dbg !44 + %1140 = insertelement <2 x bfloat> poison, bfloat %1075, i64 0, !dbg !44 + %1141 = insertelement <2 x bfloat> %1140, bfloat %1076, i64 1, !dbg !44 + %1142 = bitcast <2 x bfloat> %1141 to i32, !dbg !44 + %1143 = insertelement <2 x bfloat> poison, bfloat %1077, i64 0, !dbg !44 + %1144 = insertelement <2 x bfloat> %1143, bfloat %1078, i64 1, !dbg !44 + %1145 = bitcast <2 x bfloat> %1144 to i32, !dbg !44 + %1146 = insertelement <2 x bfloat> poison, bfloat %1079, i64 0, !dbg !44 + %1147 = insertelement <2 x bfloat> %1146, bfloat %1080, i64 1, !dbg !44 + %1148 = bitcast <2 x bfloat> %1147 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1139, i32 %1142, i32 %1145, i32 %1148, ptr addrspace(1) %107, i1 %40) #3, !dbg !44 + %1149 = insertelement <2 x bfloat> poison, bfloat %1081, i64 0, !dbg !44 + %1150 = insertelement <2 x bfloat> %1149, bfloat %1082, i64 1, !dbg !44 + %1151 = bitcast <2 x bfloat> %1150 to i32, !dbg !44 + %1152 = insertelement <2 x bfloat> poison, bfloat %1083, i64 0, !dbg !44 + %1153 = insertelement <2 x bfloat> %1152, bfloat %1084, i64 1, !dbg !44 + %1154 = bitcast <2 x bfloat> %1153 to i32, !dbg !44 + %1155 = insertelement <2 x bfloat> poison, bfloat %1085, i64 0, !dbg !44 + %1156 = insertelement <2 x bfloat> %1155, bfloat %1086, i64 1, !dbg !44 + %1157 = bitcast <2 x bfloat> %1156 to i32, !dbg !44 + %1158 = insertelement <2 x bfloat> poison, bfloat %1087, i64 0, !dbg !44 + %1159 = insertelement <2 x bfloat> %1158, bfloat %1088, i64 1, !dbg !44 + %1160 = bitcast <2 x bfloat> %1159 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1151, i32 %1154, i32 %1157, i32 %1160, ptr addrspace(1) %108, i1 %41) #3, !dbg !44 + %1161 = insertelement <2 x bfloat> poison, bfloat %1089, i64 0, !dbg !44 + %1162 = insertelement <2 x bfloat> %1161, bfloat %1090, i64 1, !dbg !44 + %1163 = bitcast <2 x bfloat> %1162 to i32, !dbg !44 + %1164 = insertelement <2 x bfloat> poison, bfloat %1091, i64 0, !dbg !44 + %1165 = insertelement <2 x bfloat> %1164, bfloat %1092, i64 1, !dbg !44 + %1166 = bitcast <2 x bfloat> %1165 to i32, !dbg !44 + %1167 = insertelement <2 x bfloat> poison, bfloat %1093, i64 0, !dbg !44 + %1168 = insertelement <2 x bfloat> %1167, bfloat %1094, i64 1, !dbg !44 + %1169 = bitcast <2 x bfloat> %1168 to i32, !dbg !44 + %1170 = insertelement <2 x bfloat> poison, bfloat %1095, i64 0, !dbg !44 + %1171 = insertelement <2 x bfloat> %1170, bfloat %1096, i64 1, !dbg !44 + %1172 = bitcast <2 x bfloat> %1171 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1163, i32 %1166, i32 %1169, i32 %1172, ptr addrspace(1) %109, i1 %42) #3, !dbg !44 + %1173 = insertelement <2 x bfloat> poison, bfloat %1097, i64 0, !dbg !44 + %1174 = insertelement <2 x bfloat> %1173, bfloat %1098, i64 1, !dbg !44 + %1175 = bitcast <2 x bfloat> %1174 to i32, !dbg !44 + %1176 = insertelement <2 x bfloat> poison, bfloat %1099, i64 0, !dbg !44 + %1177 = insertelement <2 x bfloat> %1176, bfloat %1100, i64 1, !dbg !44 + %1178 = bitcast <2 x bfloat> %1177 to i32, !dbg !44 + %1179 = insertelement <2 x bfloat> poison, bfloat %1101, i64 0, !dbg !44 + %1180 = insertelement <2 x bfloat> %1179, bfloat %1102, i64 1, !dbg !44 + %1181 = bitcast <2 x bfloat> %1180 to i32, !dbg !44 + %1182 = insertelement <2 x bfloat> poison, bfloat %1103, i64 0, !dbg !44 + %1183 = insertelement <2 x bfloat> %1182, bfloat %1104, i64 1, !dbg !44 + %1184 = bitcast <2 x bfloat> %1183 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1175, i32 %1178, i32 %1181, i32 %1184, ptr addrspace(1) %110, i1 %43) #3, !dbg !44 + %1185 = bitcast <2 x bfloat> %1105 to i32, !dbg !44 + %1186 = bitcast <2 x bfloat> %1106 to i32, !dbg !44 + %1187 = bitcast <2 x bfloat> %1107 to i32, !dbg !44 + %1188 = bitcast <2 x bfloat> %1108 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1185, i32 %1186, i32 %1187, i32 %1188, ptr addrspace(1) %111, i1 %44) #3, !dbg !44 + %1189 = bitcast <2 x bfloat> %1109 to i32, !dbg !44 + %1190 = bitcast <2 x bfloat> %1110 to i32, !dbg !44 + %1191 = bitcast <2 x bfloat> %1111 to i32, !dbg !44 + %1192 = bitcast <2 x bfloat> %1112 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1189, i32 %1190, i32 %1191, i32 %1192, ptr addrspace(1) %112, i1 %45) #3, !dbg !44 + ret void, !dbg !45 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +attributes #0 = { nounwind "nvvm.reqntid"="32" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = distinct !DISubprogram(name: "l2norm_bwd_kernel", linkageName: "l2norm_bwd_kernel", scope: !1, file: !1, line: 93, type: !5, scopeLine: 93, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!5 = !DISubroutineType(cc: DW_CC_normal, types: !6) +!6 = !{null, !7, !9, !7, !7, !10, !11, !7, !7} +!7 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !8, size: 64, dwarfAddressSpace: 1) +!8 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!9 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !10, size: 64, dwarfAddressSpace: 1) +!10 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!11 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!12 = !DILocation(line: 94, column: 24, scope: !4) +!13 = !DILocation(line: 94, column: 30, scope: !4) +!14 = !DILocation(line: 95, column: 16, scope: !4) +!15 = !DILocation(line: 95, column: 34, scope: !4) +!16 = !DILocation(line: 95, column: 21, scope: !4) +!17 = !DILocation(line: 97, column: 16, scope: !4) +!18 = !DILocation(line: 98, column: 30, scope: !4) +!19 = !DILocation(line: 99, column: 29, scope: !4) +!20 = !DILocation(line: 99, column: 14, scope: !4) +!21 = !DILocation(line: 99, column: 33, scope: !4) +!22 = !DILocation(line: 100, column: 20, scope: !4) +!23 = !DILocation(line: 101, column: 16, scope: !4) +!24 = !DILocation(line: 101, column: 35, scope: !4) +!25 = !DILocation(line: 102, column: 16, scope: !4) +!26 = !DILocation(line: 102, column: 35, scope: !4) +!27 = !DILocation(line: 104, column: 18, scope: !4) +!28 = !DILocation(line: 104, column: 47, scope: !4) +!29 = !DILocation(line: 105, column: 21, scope: !4) +!30 = !DILocation(line: 106, column: 19, scope: !4) +!31 = !DILocation(line: 106, column: 49, scope: !4) +!32 = !DILocation(line: 107, column: 18, scope: !4) +!33 = !DILocation(line: 107, column: 50, scope: !4) +!34 = !DILocation(line: 263, column: 15, scope: !35, inlinedAt: !37) +!35 = distinct !DILexicalBlockFile(scope: !4, file: !36, discriminator: 0) +!36 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!37 = !DILocation(line: 293, column: 36, scope: !35, inlinedAt: !38) +!38 = !DILocation(line: 107, column: 55, scope: !39) +!39 = distinct !DILexicalBlockFile(scope: !4, file: !1, discriminator: 0) +!40 = !DILocation(line: 107, column: 69, scope: !4) +!41 = !DILocation(line: 107, column: 75, scope: !4) +!42 = !DILocation(line: 107, column: 36, scope: !4) +!43 = !DILocation(line: 108, column: 27, scope: !4) +!44 = !DILocation(line: 108, column: 19, scope: !4) +!45 = !DILocation(line: 108, column: 4, scope: !4) diff --git a/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/l2norm_bwd_kernel.ptx b/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/l2norm_bwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..ebcd3a0a8c7c49ffa8afcd03bbba5a9892a776e4 --- /dev/null +++ b/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/l2norm_bwd_kernel.ptx @@ -0,0 +1,1314 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_bwd_kernel // -- Begin function l2norm_bwd_kernel + // @l2norm_bwd_kernel +.visible .entry l2norm_bwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_2, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_3, + .param .f32 l2norm_bwd_kernel_param_4, + .param .u32 l2norm_bwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_6, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_7 +) +.reqntid 32 +{ + .reg .pred %p<9>; + .reg .b16 %rs<129>; + .reg .b32 %r<630>; + .reg .b64 %rd<113>; + .loc 1 93 0 // l2norm.py:93:0 +$L__func_begin0: + .loc 1 93 0 // l2norm.py:93:0 + +// %bb.0: + ld.param.b64 %rd33, [l2norm_bwd_kernel_param_0]; + ld.param.b64 %rd34, [l2norm_bwd_kernel_param_1]; +$L__tmp0: + .loc 1 94 24 // l2norm.py:94:24 + mov.u32 %r106, %ctaid.x; + ld.param.b64 %rd35, [l2norm_bwd_kernel_param_2]; + .loc 1 95 16 // l2norm.py:95:16 + mul.wide.u32 %rd36, %r106, 16; + ld.param.b64 %rd37, [l2norm_bwd_kernel_param_3]; + .loc 1 95 34 // l2norm.py:95:34 + mov.u32 %r107, %tid.x; + ld.param.s32 %rd38, [l2norm_bwd_kernel_param_5]; + bfe.u32 %r108, %r107, 4, 1; + or.b32 %r109, %r108, 2; + or.b32 %r110, %r108, 4; + or.b32 %r111, %r108, 6; + or.b32 %r112, %r108, 8; + or.b32 %r113, %r108, 10; + or.b32 %r114, %r108, 12; + or.b32 %r115, %r108, 14; + .loc 1 95 21 // l2norm.py:95:21 + cvt.u64.u32 %rd39, %r108; + cvt.u64.u32 %rd40, %r109; + cvt.u64.u32 %rd41, %r110; + cvt.u64.u32 %rd42, %r111; + cvt.u64.u32 %rd43, %r112; + cvt.u64.u32 %rd44, %r113; + cvt.u64.u32 %rd45, %r114; + cvt.u64.u32 %rd46, %r115; + or.b64 %rd47, %rd36, %rd39; + or.b64 %rd48, %rd36, %rd40; + or.b64 %rd49, %rd36, %rd41; + or.b64 %rd50, %rd36, %rd42; + or.b64 %rd51, %rd36, %rd43; + or.b64 %rd52, %rd36, %rd44; + or.b64 %rd53, %rd36, %rd45; + or.b64 %rd54, %rd36, %rd46; + .loc 1 97 16 // l2norm.py:97:16 + setp.lt.s64 %p1, %rd47, %rd38; + setp.lt.s64 %p2, %rd48, %rd38; + setp.lt.s64 %p3, %rd49, %rd38; + setp.lt.s64 %p4, %rd50, %rd38; + setp.lt.s64 %p5, %rd51, %rd38; + setp.lt.s64 %p6, %rd52, %rd38; + setp.lt.s64 %p7, %rd53, %rd38; + setp.lt.s64 %p8, %rd54, %rd38; + .loc 1 98 30 // l2norm.py:98:30 + shl.b32 %r116, %r107, 3; + and.b32 %r117, %r116, 120; + .loc 1 99 14 // l2norm.py:99:14 + shl.b64 %rd55, %rd47, 8; + add.s64 %rd56, %rd33, %rd55; + shl.b64 %rd57, %rd48, 8; + add.s64 %rd58, %rd33, %rd57; + shl.b64 %rd59, %rd49, 8; + add.s64 %rd60, %rd33, %rd59; + shl.b64 %rd61, %rd50, 8; + add.s64 %rd62, %rd33, %rd61; + shl.b64 %rd63, %rd51, 8; + add.s64 %rd64, %rd33, %rd63; + shl.b64 %rd65, %rd52, 8; + add.s64 %rd66, %rd33, %rd65; + shl.b64 %rd67, %rd53, 8; + add.s64 %rd68, %rd33, %rd67; + shl.b64 %rd69, %rd54, 8; + add.s64 %rd70, %rd33, %rd69; + .loc 1 99 33 // l2norm.py:99:33 + mul.wide.u32 %rd71, %r117, 2; + add.s64 %rd1, %rd56, %rd71; + add.s64 %rd2, %rd58, %rd71; + add.s64 %rd3, %rd60, %rd71; + add.s64 %rd4, %rd62, %rd71; + add.s64 %rd5, %rd64, %rd71; + add.s64 %rd6, %rd66, %rd71; + add.s64 %rd7, %rd68, %rd71; + add.s64 %rd8, %rd70, %rd71; + .loc 1 100 20 // l2norm.py:100:20 + shl.b64 %rd72, %rd47, 2; + add.s64 %rd9, %rd34, %rd72; + add.s64 %rd10, %rd9, 8; + add.s64 %rd11, %rd9, 16; + add.s64 %rd12, %rd9, 24; + add.s64 %rd13, %rd9, 32; + add.s64 %rd14, %rd9, 40; + add.s64 %rd15, %rd9, 48; + add.s64 %rd16, %rd9, 56; + .loc 1 101 16 // l2norm.py:101:16 + add.s64 %rd73, %rd35, %rd55; + add.s64 %rd74, %rd35, %rd57; + add.s64 %rd75, %rd35, %rd59; + add.s64 %rd76, %rd35, %rd61; + add.s64 %rd77, %rd35, %rd63; + add.s64 %rd78, %rd35, %rd65; + add.s64 %rd79, %rd35, %rd67; + add.s64 %rd80, %rd35, %rd69; + .loc 1 101 35 // l2norm.py:101:35 + add.s64 %rd17, %rd73, %rd71; + add.s64 %rd18, %rd74, %rd71; + add.s64 %rd19, %rd75, %rd71; + add.s64 %rd20, %rd76, %rd71; + add.s64 %rd21, %rd77, %rd71; + add.s64 %rd22, %rd78, %rd71; + add.s64 %rd23, %rd79, %rd71; + add.s64 %rd24, %rd80, %rd71; + .loc 1 102 16 // l2norm.py:102:16 + add.s64 %rd81, %rd37, %rd55; + add.s64 %rd82, %rd37, %rd57; + add.s64 %rd83, %rd37, %rd59; + add.s64 %rd84, %rd37, %rd61; + add.s64 %rd85, %rd37, %rd63; + add.s64 %rd86, %rd37, %rd65; + add.s64 %rd87, %rd37, %rd67; + add.s64 %rd88, %rd37, %rd69; + .loc 1 102 35 // l2norm.py:102:35 + add.s64 %rd25, %rd81, %rd71; + add.s64 %rd26, %rd82, %rd71; + add.s64 %rd27, %rd83, %rd71; + add.s64 %rd28, %rd84, %rd71; + add.s64 %rd29, %rd85, %rd71; + add.s64 %rd30, %rd86, %rd71; + add.s64 %rd31, %rd87, %rd71; + add.s64 %rd32, %rd88, %rd71; + mov.b32 %r5, 0; + .loc 1 104 18 // l2norm.py:104:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + mov.b32 {%rs5, %rs6}, %r3; + mov.b32 {%rs7, %rs8}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs9, %rs10}, %r6; + mov.b32 {%rs11, %rs12}, %r7; + mov.b32 {%rs13, %rs14}, %r8; + mov.b32 {%rs15, %rs16}, %r9; + // begin inline asm + mov.u32 %r10, %r5; + mov.u32 %r11, %r5; + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + @%p3 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs17, %rs18}, %r10; + mov.b32 {%rs19, %rs20}, %r11; + mov.b32 {%rs21, %rs22}, %r12; + mov.b32 {%rs23, %rs24}, %r13; + // begin inline asm + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + @%p4 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ]; + // end inline asm + mov.b32 {%rs25, %rs26}, %r14; + mov.b32 {%rs27, %rs28}, %r15; + mov.b32 {%rs29, %rs30}, %r16; + mov.b32 {%rs31, %rs32}, %r17; + // begin inline asm + mov.u32 %r18, %r5; + mov.u32 %r19, %r5; + mov.u32 %r20, %r5; + mov.u32 %r21, %r5; + @%p5 ld.global.v4.b32 { %r18, %r19, %r20, %r21 }, [ %rd5 + 0 ]; + // end inline asm + mov.b32 {%rs33, %rs34}, %r18; + mov.b32 {%rs35, %rs36}, %r19; + mov.b32 {%rs37, %rs38}, %r20; + mov.b32 {%rs39, %rs40}, %r21; + // begin inline asm + mov.u32 %r22, %r5; + mov.u32 %r23, %r5; + mov.u32 %r24, %r5; + mov.u32 %r25, %r5; + @%p6 ld.global.v4.b32 { %r22, %r23, %r24, %r25 }, [ %rd6 + 0 ]; + // end inline asm + mov.b32 {%rs41, %rs42}, %r22; + mov.b32 {%rs43, %rs44}, %r23; + mov.b32 {%rs45, %rs46}, %r24; + mov.b32 {%rs47, %rs48}, %r25; + // begin inline asm + mov.u32 %r26, %r5; + mov.u32 %r27, %r5; + mov.u32 %r28, %r5; + mov.u32 %r29, %r5; + @%p7 ld.global.v4.b32 { %r26, %r27, %r28, %r29 }, [ %rd7 + 0 ]; + // end inline asm + mov.b32 {%rs49, %rs50}, %r26; + mov.b32 {%rs51, %rs52}, %r27; + mov.b32 {%rs53, %rs54}, %r28; + mov.b32 {%rs55, %rs56}, %r29; + // begin inline asm + mov.u32 %r30, %r5; + mov.u32 %r31, %r5; + mov.u32 %r32, %r5; + mov.u32 %r33, %r5; + @%p8 ld.global.v4.b32 { %r30, %r31, %r32, %r33 }, [ %rd8 + 0 ]; + // end inline asm + mov.b32 {%rs57, %rs58}, %r30; + mov.b32 {%rs59, %rs60}, %r31; + mov.b32 {%rs61, %rs62}, %r32; + mov.b32 {%rs63, %rs64}, %r33; + .loc 1 104 47 // l2norm.py:104:47 + cvt.f32.bf16 %r118, %rs1; + cvt.f32.bf16 %r119, %rs2; + cvt.f32.bf16 %r120, %rs3; + cvt.f32.bf16 %r121, %rs4; + cvt.f32.bf16 %r122, %rs5; + cvt.f32.bf16 %r123, %rs6; + cvt.f32.bf16 %r124, %rs7; + cvt.f32.bf16 %r125, %rs8; + cvt.f32.bf16 %r126, %rs9; + cvt.f32.bf16 %r127, %rs10; + cvt.f32.bf16 %r128, %rs11; + cvt.f32.bf16 %r129, %rs12; + cvt.f32.bf16 %r130, %rs13; + cvt.f32.bf16 %r131, %rs14; + cvt.f32.bf16 %r132, %rs15; + cvt.f32.bf16 %r133, %rs16; + cvt.f32.bf16 %r134, %rs17; + cvt.f32.bf16 %r135, %rs18; + cvt.f32.bf16 %r136, %rs19; + cvt.f32.bf16 %r137, %rs20; + cvt.f32.bf16 %r138, %rs21; + cvt.f32.bf16 %r139, %rs22; + cvt.f32.bf16 %r140, %rs23; + cvt.f32.bf16 %r141, %rs24; + cvt.f32.bf16 %r142, %rs25; + cvt.f32.bf16 %r143, %rs26; + cvt.f32.bf16 %r144, %rs27; + cvt.f32.bf16 %r145, %rs28; + cvt.f32.bf16 %r146, %rs29; + cvt.f32.bf16 %r147, %rs30; + cvt.f32.bf16 %r148, %rs31; + cvt.f32.bf16 %r149, %rs32; + cvt.f32.bf16 %r150, %rs33; + cvt.f32.bf16 %r151, %rs34; + cvt.f32.bf16 %r152, %rs35; + cvt.f32.bf16 %r153, %rs36; + cvt.f32.bf16 %r154, %rs37; + cvt.f32.bf16 %r155, %rs38; + cvt.f32.bf16 %r156, %rs39; + cvt.f32.bf16 %r157, %rs40; + cvt.f32.bf16 %r158, %rs41; + cvt.f32.bf16 %r159, %rs42; + cvt.f32.bf16 %r160, %rs43; + cvt.f32.bf16 %r161, %rs44; + cvt.f32.bf16 %r162, %rs45; + cvt.f32.bf16 %r163, %rs46; + cvt.f32.bf16 %r164, %rs47; + cvt.f32.bf16 %r165, %rs48; + cvt.f32.bf16 %r166, %rs49; + cvt.f32.bf16 %r167, %rs50; + cvt.f32.bf16 %r168, %rs51; + cvt.f32.bf16 %r169, %rs52; + cvt.f32.bf16 %r170, %rs53; + cvt.f32.bf16 %r171, %rs54; + cvt.f32.bf16 %r172, %rs55; + cvt.f32.bf16 %r173, %rs56; + cvt.f32.bf16 %r174, %rs57; + cvt.f32.bf16 %r175, %rs58; + cvt.f32.bf16 %r176, %rs59; + cvt.f32.bf16 %r177, %rs60; + cvt.f32.bf16 %r178, %rs61; + cvt.f32.bf16 %r179, %rs62; + cvt.f32.bf16 %r180, %rs63; + cvt.f32.bf16 %r181, %rs64; + .loc 1 105 21 // l2norm.py:105:21 + // begin inline asm + mov.u32 %r34, %r5; + @%p1 ld.global.b32 { %r34 }, [ %rd9 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r35, %r5; + @%p2 ld.global.b32 { %r35 }, [ %rd10 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r36, %r5; + @%p3 ld.global.b32 { %r36 }, [ %rd11 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r37, %r5; + @%p4 ld.global.b32 { %r37 }, [ %rd12 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r38, %r5; + @%p5 ld.global.b32 { %r38 }, [ %rd13 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r39, %r5; + @%p6 ld.global.b32 { %r39 }, [ %rd14 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r40, %r5; + @%p7 ld.global.b32 { %r40 }, [ %rd15 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r41, %r5; + @%p8 ld.global.b32 { %r41 }, [ %rd16 + 0 ]; + // end inline asm + .loc 1 106 19 // l2norm.py:106:19 + // begin inline asm + mov.u32 %r42, %r5; + mov.u32 %r43, %r5; + mov.u32 %r44, %r5; + mov.u32 %r45, %r5; + @%p1 ld.global.v4.b32 { %r42, %r43, %r44, %r45 }, [ %rd17 + 0 ]; + // end inline asm + mov.b32 {%rs65, %rs66}, %r42; + mov.b32 {%rs67, %rs68}, %r43; + mov.b32 {%rs69, %rs70}, %r44; + mov.b32 {%rs71, %rs72}, %r45; + // begin inline asm + mov.u32 %r46, %r5; + mov.u32 %r47, %r5; + mov.u32 %r48, %r5; + mov.u32 %r49, %r5; + @%p2 ld.global.v4.b32 { %r46, %r47, %r48, %r49 }, [ %rd18 + 0 ]; + // end inline asm + mov.b32 {%rs73, %rs74}, %r46; + mov.b32 {%rs75, %rs76}, %r47; + mov.b32 {%rs77, %rs78}, %r48; + mov.b32 {%rs79, %rs80}, %r49; + // begin inline asm + mov.u32 %r50, %r5; + mov.u32 %r51, %r5; + mov.u32 %r52, %r5; + mov.u32 %r53, %r5; + @%p3 ld.global.v4.b32 { %r50, %r51, %r52, %r53 }, [ %rd19 + 0 ]; + // end inline asm + mov.b32 {%rs81, %rs82}, %r50; + mov.b32 {%rs83, %rs84}, %r51; + mov.b32 {%rs85, %rs86}, %r52; + mov.b32 {%rs87, %rs88}, %r53; + // begin inline asm + mov.u32 %r54, %r5; + mov.u32 %r55, %r5; + mov.u32 %r56, %r5; + mov.u32 %r57, %r5; + @%p4 ld.global.v4.b32 { %r54, %r55, %r56, %r57 }, [ %rd20 + 0 ]; + // end inline asm + mov.b32 {%rs89, %rs90}, %r54; + mov.b32 {%rs91, %rs92}, %r55; + mov.b32 {%rs93, %rs94}, %r56; + mov.b32 {%rs95, %rs96}, %r57; + // begin inline asm + mov.u32 %r58, %r5; + mov.u32 %r59, %r5; + mov.u32 %r60, %r5; + mov.u32 %r61, %r5; + @%p5 ld.global.v4.b32 { %r58, %r59, %r60, %r61 }, [ %rd21 + 0 ]; + // end inline asm + mov.b32 {%rs97, %rs98}, %r58; + mov.b32 {%rs99, %rs100}, %r59; + mov.b32 {%rs101, %rs102}, %r60; + mov.b32 {%rs103, %rs104}, %r61; + // begin inline asm + mov.u32 %r62, %r5; + mov.u32 %r63, %r5; + mov.u32 %r64, %r5; + mov.u32 %r65, %r5; + @%p6 ld.global.v4.b32 { %r62, %r63, %r64, %r65 }, [ %rd22 + 0 ]; + // end inline asm + mov.b32 {%rs105, %rs106}, %r62; + mov.b32 {%rs107, %rs108}, %r63; + mov.b32 {%rs109, %rs110}, %r64; + mov.b32 {%rs111, %rs112}, %r65; + // begin inline asm + mov.u32 %r66, %r5; + mov.u32 %r67, %r5; + mov.u32 %r68, %r5; + mov.u32 %r69, %r5; + @%p7 ld.global.v4.b32 { %r66, %r67, %r68, %r69 }, [ %rd23 + 0 ]; + // end inline asm + mov.b32 {%rs113, %rs114}, %r66; + mov.b32 {%rs115, %rs116}, %r67; + mov.b32 {%rs117, %rs118}, %r68; + mov.b32 {%rs119, %rs120}, %r69; + // begin inline asm + mov.u32 %r70, %r5; + mov.u32 %r71, %r5; + mov.u32 %r72, %r5; + mov.u32 %r73, %r5; + @%p8 ld.global.v4.b32 { %r70, %r71, %r72, %r73 }, [ %rd24 + 0 ]; + // end inline asm + mov.b32 {%rs121, %rs122}, %r70; + mov.b32 {%rs123, %rs124}, %r71; + mov.b32 {%rs125, %rs126}, %r72; + mov.b32 {%rs127, %rs128}, %r73; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r182, %rs65; + cvt.f32.bf16 %r183, %rs66; + cvt.f32.bf16 %r184, %rs67; + cvt.f32.bf16 %r185, %rs68; + cvt.f32.bf16 %r186, %rs69; + cvt.f32.bf16 %r187, %rs70; + cvt.f32.bf16 %r188, %rs71; + cvt.f32.bf16 %r189, %rs72; + cvt.f32.bf16 %r190, %rs73; + cvt.f32.bf16 %r191, %rs74; + cvt.f32.bf16 %r192, %rs75; + cvt.f32.bf16 %r193, %rs76; + cvt.f32.bf16 %r194, %rs77; + cvt.f32.bf16 %r195, %rs78; + cvt.f32.bf16 %r196, %rs79; + cvt.f32.bf16 %r197, %rs80; + cvt.f32.bf16 %r198, %rs81; + cvt.f32.bf16 %r199, %rs82; + cvt.f32.bf16 %r200, %rs83; + cvt.f32.bf16 %r201, %rs84; + cvt.f32.bf16 %r202, %rs85; + cvt.f32.bf16 %r203, %rs86; + cvt.f32.bf16 %r204, %rs87; + cvt.f32.bf16 %r205, %rs88; + cvt.f32.bf16 %r206, %rs89; + cvt.f32.bf16 %r207, %rs90; + cvt.f32.bf16 %r208, %rs91; + cvt.f32.bf16 %r209, %rs92; + cvt.f32.bf16 %r210, %rs93; + cvt.f32.bf16 %r211, %rs94; + cvt.f32.bf16 %r212, %rs95; + cvt.f32.bf16 %r213, %rs96; + cvt.f32.bf16 %r214, %rs97; + cvt.f32.bf16 %r215, %rs98; + cvt.f32.bf16 %r216, %rs99; + cvt.f32.bf16 %r217, %rs100; + cvt.f32.bf16 %r218, %rs101; + cvt.f32.bf16 %r219, %rs102; + cvt.f32.bf16 %r220, %rs103; + cvt.f32.bf16 %r221, %rs104; + cvt.f32.bf16 %r222, %rs105; + cvt.f32.bf16 %r223, %rs106; + cvt.f32.bf16 %r224, %rs107; + cvt.f32.bf16 %r225, %rs108; + cvt.f32.bf16 %r226, %rs109; + cvt.f32.bf16 %r227, %rs110; + cvt.f32.bf16 %r228, %rs111; + cvt.f32.bf16 %r229, %rs112; + cvt.f32.bf16 %r230, %rs113; + cvt.f32.bf16 %r231, %rs114; + cvt.f32.bf16 %r232, %rs115; + cvt.f32.bf16 %r233, %rs116; + cvt.f32.bf16 %r234, %rs117; + cvt.f32.bf16 %r235, %rs118; + cvt.f32.bf16 %r236, %rs119; + cvt.f32.bf16 %r237, %rs120; + cvt.f32.bf16 %r238, %rs121; + cvt.f32.bf16 %r239, %rs122; + cvt.f32.bf16 %r240, %rs123; + cvt.f32.bf16 %r241, %rs124; + cvt.f32.bf16 %r242, %rs125; + cvt.f32.bf16 %r243, %rs126; + cvt.f32.bf16 %r244, %rs127; + cvt.f32.bf16 %r245, %rs128; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r246, %r40, %r230; + mul.f32 %r247, %r40, %r231; + mul.f32 %r248, %r40, %r232; + mul.f32 %r249, %r40, %r233; + mul.f32 %r250, %r40, %r234; + mul.f32 %r251, %r40, %r235; + mul.f32 %r252, %r40, %r236; + mul.f32 %r253, %r40, %r237; + mul.f32 %r254, %r41, %r238; + mul.f32 %r255, %r41, %r239; + mul.f32 %r256, %r41, %r240; + mul.f32 %r257, %r41, %r241; + mul.f32 %r258, %r41, %r242; + mul.f32 %r259, %r41, %r243; + mul.f32 %r260, %r41, %r244; + mul.f32 %r261, %r41, %r245; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r262, %r119, %r183; + mul.f32 %r263, %r127, %r191; + mul.f32 %r264, %r135, %r199; + mul.f32 %r265, %r143, %r207; + mul.f32 %r266, %r151, %r215; + mul.f32 %r267, %r159, %r223; + mul.f32 %r268, %r167, %r231; + mul.f32 %r269, %r175, %r239; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r270, %rs1, %rs65, %r262; + fma.rn.f32.bf16 %r271, %rs3, %rs67, %r270; + fma.rn.f32.bf16 %r272, %rs4, %rs68, %r271; + fma.rn.f32.bf16 %r273, %rs5, %rs69, %r272; + fma.rn.f32.bf16 %r274, %rs6, %rs70, %r273; + fma.rn.f32.bf16 %r275, %rs7, %rs71, %r274; + fma.rn.f32.bf16 %r276, %rs8, %rs72, %r275; + fma.rn.f32.bf16 %r277, %rs9, %rs73, %r263; + fma.rn.f32.bf16 %r278, %rs11, %rs75, %r277; + fma.rn.f32.bf16 %r279, %rs12, %rs76, %r278; + fma.rn.f32.bf16 %r280, %rs13, %rs77, %r279; + fma.rn.f32.bf16 %r281, %rs14, %rs78, %r280; + fma.rn.f32.bf16 %r282, %rs15, %rs79, %r281; + fma.rn.f32.bf16 %r283, %rs16, %rs80, %r282; + fma.rn.f32.bf16 %r284, %rs17, %rs81, %r264; + fma.rn.f32.bf16 %r285, %rs19, %rs83, %r284; + fma.rn.f32.bf16 %r286, %rs20, %rs84, %r285; + fma.rn.f32.bf16 %r287, %rs21, %rs85, %r286; + fma.rn.f32.bf16 %r288, %rs22, %rs86, %r287; + fma.rn.f32.bf16 %r289, %rs23, %rs87, %r288; + fma.rn.f32.bf16 %r290, %rs24, %rs88, %r289; + fma.rn.f32.bf16 %r291, %rs25, %rs89, %r265; + fma.rn.f32.bf16 %r292, %rs27, %rs91, %r291; + fma.rn.f32.bf16 %r293, %rs28, %rs92, %r292; + fma.rn.f32.bf16 %r294, %rs29, %rs93, %r293; + fma.rn.f32.bf16 %r295, %rs30, %rs94, %r294; + fma.rn.f32.bf16 %r296, %rs31, %rs95, %r295; + fma.rn.f32.bf16 %r297, %rs32, %rs96, %r296; + fma.rn.f32.bf16 %r298, %rs33, %rs97, %r266; + fma.rn.f32.bf16 %r299, %rs35, %rs99, %r298; + fma.rn.f32.bf16 %r300, %rs36, %rs100, %r299; + fma.rn.f32.bf16 %r301, %rs37, %rs101, %r300; + fma.rn.f32.bf16 %r302, %rs38, %rs102, %r301; + fma.rn.f32.bf16 %r303, %rs39, %rs103, %r302; + fma.rn.f32.bf16 %r304, %rs40, %rs104, %r303; + fma.rn.f32.bf16 %r305, %rs41, %rs105, %r267; + fma.rn.f32.bf16 %r306, %rs43, %rs107, %r305; + fma.rn.f32.bf16 %r307, %rs44, %rs108, %r306; + fma.rn.f32.bf16 %r308, %rs45, %rs109, %r307; + fma.rn.f32.bf16 %r309, %rs46, %rs110, %r308; + fma.rn.f32.bf16 %r310, %rs47, %rs111, %r309; + fma.rn.f32.bf16 %r311, %rs48, %rs112, %r310; + fma.rn.f32.bf16 %r312, %rs49, %rs113, %r268; + fma.rn.f32.bf16 %r313, %rs51, %rs115, %r312; + fma.rn.f32.bf16 %r314, %rs52, %rs116, %r313; + fma.rn.f32.bf16 %r315, %rs53, %rs117, %r314; + fma.rn.f32.bf16 %r316, %rs54, %rs118, %r315; + fma.rn.f32.bf16 %r317, %rs55, %rs119, %r316; + fma.rn.f32.bf16 %r318, %rs56, %rs120, %r317; + fma.rn.f32.bf16 %r319, %rs57, %rs121, %r269; + fma.rn.f32.bf16 %r320, %rs59, %rs123, %r319; + fma.rn.f32.bf16 %r321, %rs60, %rs124, %r320; + fma.rn.f32.bf16 %r322, %rs61, %rs125, %r321; + fma.rn.f32.bf16 %r323, %rs62, %rs126, %r322; + fma.rn.f32.bf16 %r324, %rs63, %rs127, %r323; + fma.rn.f32.bf16 %r325, %rs64, %rs128, %r324; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r326, %r276, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r327, %r276, %r326; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r328, %r327, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r329, %r327, %r328; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r330, %r329, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r331, %r329, %r330; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r332, %r331, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r333, %r331, %r332; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r334, %r283, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r335, %r283, %r334; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r336, %r335, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r337, %r335, %r336; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r338, %r337, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r339, %r337, %r338; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r340, %r339, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r341, %r339, %r340; +$L__tmp18: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r342, %r290, 8, 31, -1; +$L__tmp19: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r343, %r290, %r342; +$L__tmp20: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r344, %r343, 4, 31, -1; +$L__tmp21: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r345, %r343, %r344; +$L__tmp22: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r346, %r345, 2, 31, -1; +$L__tmp23: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r347, %r345, %r346; +$L__tmp24: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r348, %r347, 1, 31, -1; +$L__tmp25: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r349, %r347, %r348; +$L__tmp26: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r350, %r297, 8, 31, -1; +$L__tmp27: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r351, %r297, %r350; +$L__tmp28: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r352, %r351, 4, 31, -1; +$L__tmp29: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r353, %r351, %r352; +$L__tmp30: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r354, %r353, 2, 31, -1; +$L__tmp31: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r355, %r353, %r354; +$L__tmp32: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r356, %r355, 1, 31, -1; +$L__tmp33: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r357, %r355, %r356; +$L__tmp34: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r358, %r304, 8, 31, -1; +$L__tmp35: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r359, %r304, %r358; +$L__tmp36: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r360, %r359, 4, 31, -1; +$L__tmp37: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r361, %r359, %r360; +$L__tmp38: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r362, %r361, 2, 31, -1; +$L__tmp39: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r363, %r361, %r362; +$L__tmp40: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r364, %r363, 1, 31, -1; +$L__tmp41: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r365, %r363, %r364; +$L__tmp42: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r366, %r311, 8, 31, -1; +$L__tmp43: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r367, %r311, %r366; +$L__tmp44: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r368, %r367, 4, 31, -1; +$L__tmp45: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r369, %r367, %r368; +$L__tmp46: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r370, %r369, 2, 31, -1; +$L__tmp47: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r371, %r369, %r370; +$L__tmp48: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r372, %r371, 1, 31, -1; +$L__tmp49: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r373, %r371, %r372; +$L__tmp50: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r374, %r318, 8, 31, -1; +$L__tmp51: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r375, %r318, %r374; +$L__tmp52: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r376, %r375, 4, 31, -1; +$L__tmp53: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r377, %r375, %r376; +$L__tmp54: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r378, %r377, 2, 31, -1; +$L__tmp55: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r379, %r377, %r378; +$L__tmp56: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r380, %r379, 1, 31, -1; +$L__tmp57: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r381, %r379, %r380; +$L__tmp58: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r382, %r325, 8, 31, -1; +$L__tmp59: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r383, %r325, %r382; +$L__tmp60: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r384, %r383, 4, 31, -1; +$L__tmp61: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r385, %r383, %r384; +$L__tmp62: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r386, %r385, 2, 31, -1; +$L__tmp63: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r387, %r385, %r386; +$L__tmp64: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r388, %r387, 1, 31, -1; +$L__tmp65: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r389, %r387, %r388; +$L__tmp66: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r390, %r333, %r118; + mul.f32 %r391, %r333, %r119; + mul.f32 %r392, %r333, %r120; + mul.f32 %r393, %r333, %r121; + mul.f32 %r394, %r333, %r122; + mul.f32 %r395, %r333, %r123; + mul.f32 %r396, %r333, %r124; + mul.f32 %r397, %r333, %r125; + mul.f32 %r398, %r341, %r126; + mul.f32 %r399, %r341, %r127; + mul.f32 %r400, %r341, %r128; + mul.f32 %r401, %r341, %r129; + mul.f32 %r402, %r341, %r130; + mul.f32 %r403, %r341, %r131; + mul.f32 %r404, %r341, %r132; + mul.f32 %r405, %r341, %r133; + mul.f32 %r406, %r349, %r134; + mul.f32 %r407, %r349, %r135; + mul.f32 %r408, %r349, %r136; + mul.f32 %r409, %r349, %r137; + mul.f32 %r410, %r349, %r138; + mul.f32 %r411, %r349, %r139; + mul.f32 %r412, %r349, %r140; + mul.f32 %r413, %r349, %r141; + mul.f32 %r414, %r357, %r142; + mul.f32 %r415, %r357, %r143; + mul.f32 %r416, %r357, %r144; + mul.f32 %r417, %r357, %r145; + mul.f32 %r418, %r357, %r146; + mul.f32 %r419, %r357, %r147; + mul.f32 %r420, %r357, %r148; + mul.f32 %r421, %r357, %r149; + mul.f32 %r422, %r365, %r150; + mul.f32 %r423, %r365, %r151; + mul.f32 %r424, %r365, %r152; + mul.f32 %r425, %r365, %r153; + mul.f32 %r426, %r365, %r154; + mul.f32 %r427, %r365, %r155; + mul.f32 %r428, %r365, %r156; + mul.f32 %r429, %r365, %r157; + mul.f32 %r430, %r373, %r158; + mul.f32 %r431, %r373, %r159; + mul.f32 %r432, %r373, %r160; + mul.f32 %r433, %r373, %r161; + mul.f32 %r434, %r373, %r162; + mul.f32 %r435, %r373, %r163; + mul.f32 %r436, %r373, %r164; + mul.f32 %r437, %r373, %r165; + mul.f32 %r438, %r381, %r166; + mul.f32 %r439, %r381, %r167; + mul.f32 %r440, %r381, %r168; + mul.f32 %r441, %r381, %r169; + mul.f32 %r442, %r381, %r170; + mul.f32 %r443, %r381, %r171; + mul.f32 %r444, %r381, %r172; + mul.f32 %r445, %r381, %r173; + mul.f32 %r446, %r389, %r174; + mul.f32 %r447, %r389, %r175; + mul.f32 %r448, %r389, %r176; + mul.f32 %r449, %r389, %r177; + mul.f32 %r450, %r389, %r178; + mul.f32 %r451, %r389, %r179; + mul.f32 %r452, %r389, %r180; + mul.f32 %r453, %r389, %r181; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r454, %r390, %r34; + mul.f32 %r455, %r391, %r34; + mul.f32 %r456, %r392, %r34; + mul.f32 %r457, %r393, %r34; + mul.f32 %r458, %r394, %r34; + mul.f32 %r459, %r395, %r34; + mul.f32 %r460, %r396, %r34; + mul.f32 %r461, %r397, %r34; + mul.f32 %r462, %r398, %r35; + mul.f32 %r463, %r399, %r35; + mul.f32 %r464, %r400, %r35; + mul.f32 %r465, %r401, %r35; + mul.f32 %r466, %r402, %r35; + mul.f32 %r467, %r403, %r35; + mul.f32 %r468, %r404, %r35; + mul.f32 %r469, %r405, %r35; + mul.f32 %r470, %r406, %r36; + mul.f32 %r471, %r407, %r36; + mul.f32 %r472, %r408, %r36; + mul.f32 %r473, %r409, %r36; + mul.f32 %r474, %r410, %r36; + mul.f32 %r475, %r411, %r36; + mul.f32 %r476, %r412, %r36; + mul.f32 %r477, %r413, %r36; + mul.f32 %r478, %r414, %r37; + mul.f32 %r479, %r415, %r37; + mul.f32 %r480, %r416, %r37; + mul.f32 %r481, %r417, %r37; + mul.f32 %r482, %r418, %r37; + mul.f32 %r483, %r419, %r37; + mul.f32 %r484, %r420, %r37; + mul.f32 %r485, %r421, %r37; + mul.f32 %r486, %r422, %r38; + mul.f32 %r487, %r423, %r38; + mul.f32 %r488, %r424, %r38; + mul.f32 %r489, %r425, %r38; + mul.f32 %r490, %r426, %r38; + mul.f32 %r491, %r427, %r38; + mul.f32 %r492, %r428, %r38; + mul.f32 %r493, %r429, %r38; + mul.f32 %r494, %r430, %r39; + mul.f32 %r495, %r431, %r39; + mul.f32 %r496, %r432, %r39; + mul.f32 %r497, %r433, %r39; + mul.f32 %r498, %r434, %r39; + mul.f32 %r499, %r435, %r39; + mul.f32 %r500, %r436, %r39; + mul.f32 %r501, %r437, %r39; + mul.f32 %r502, %r438, %r40; + mul.f32 %r503, %r439, %r40; + mul.f32 %r504, %r440, %r40; + mul.f32 %r505, %r441, %r40; + mul.f32 %r506, %r442, %r40; + mul.f32 %r507, %r443, %r40; + mul.f32 %r508, %r444, %r40; + mul.f32 %r509, %r445, %r40; + mul.f32 %r510, %r446, %r41; + mul.f32 %r511, %r447, %r41; + mul.f32 %r512, %r448, %r41; + mul.f32 %r513, %r449, %r41; + mul.f32 %r514, %r450, %r41; + mul.f32 %r515, %r451, %r41; + mul.f32 %r516, %r452, %r41; + mul.f32 %r517, %r453, %r41; + .loc 1 107 36 // l2norm.py:107:36 + neg.f32 %r518, %r454; + fma.rn.f32 %r519, %r34, %r182, %r518; + neg.f32 %r520, %r455; + fma.rn.f32 %r521, %r34, %r183, %r520; + neg.f32 %r522, %r456; + fma.rn.f32 %r523, %r34, %r184, %r522; + neg.f32 %r524, %r457; + fma.rn.f32 %r525, %r34, %r185, %r524; + neg.f32 %r526, %r458; + fma.rn.f32 %r527, %r34, %r186, %r526; + neg.f32 %r528, %r459; + fma.rn.f32 %r529, %r34, %r187, %r528; + neg.f32 %r530, %r460; + fma.rn.f32 %r531, %r34, %r188, %r530; + neg.f32 %r532, %r461; + fma.rn.f32 %r533, %r34, %r189, %r532; + neg.f32 %r534, %r462; + fma.rn.f32 %r535, %r35, %r190, %r534; + neg.f32 %r536, %r463; + fma.rn.f32 %r537, %r35, %r191, %r536; + neg.f32 %r538, %r464; + fma.rn.f32 %r539, %r35, %r192, %r538; + neg.f32 %r540, %r465; + fma.rn.f32 %r541, %r35, %r193, %r540; + neg.f32 %r542, %r466; + fma.rn.f32 %r543, %r35, %r194, %r542; + neg.f32 %r544, %r467; + fma.rn.f32 %r545, %r35, %r195, %r544; + neg.f32 %r546, %r468; + fma.rn.f32 %r547, %r35, %r196, %r546; + neg.f32 %r548, %r469; + fma.rn.f32 %r549, %r35, %r197, %r548; + neg.f32 %r550, %r470; + fma.rn.f32 %r551, %r36, %r198, %r550; + neg.f32 %r552, %r471; + fma.rn.f32 %r553, %r36, %r199, %r552; + neg.f32 %r554, %r472; + fma.rn.f32 %r555, %r36, %r200, %r554; + neg.f32 %r556, %r473; + fma.rn.f32 %r557, %r36, %r201, %r556; + neg.f32 %r558, %r474; + fma.rn.f32 %r559, %r36, %r202, %r558; + neg.f32 %r560, %r475; + fma.rn.f32 %r561, %r36, %r203, %r560; + neg.f32 %r562, %r476; + fma.rn.f32 %r563, %r36, %r204, %r562; + neg.f32 %r564, %r477; + fma.rn.f32 %r565, %r36, %r205, %r564; + neg.f32 %r566, %r478; + fma.rn.f32 %r567, %r37, %r206, %r566; + neg.f32 %r568, %r479; + fma.rn.f32 %r569, %r37, %r207, %r568; + neg.f32 %r570, %r480; + fma.rn.f32 %r571, %r37, %r208, %r570; + neg.f32 %r572, %r481; + fma.rn.f32 %r573, %r37, %r209, %r572; + neg.f32 %r574, %r482; + fma.rn.f32 %r575, %r37, %r210, %r574; + neg.f32 %r576, %r483; + fma.rn.f32 %r577, %r37, %r211, %r576; + neg.f32 %r578, %r484; + fma.rn.f32 %r579, %r37, %r212, %r578; + neg.f32 %r580, %r485; + fma.rn.f32 %r581, %r37, %r213, %r580; + neg.f32 %r582, %r486; + fma.rn.f32 %r583, %r38, %r214, %r582; + neg.f32 %r584, %r487; + fma.rn.f32 %r585, %r38, %r215, %r584; + neg.f32 %r586, %r488; + fma.rn.f32 %r587, %r38, %r216, %r586; + neg.f32 %r588, %r489; + fma.rn.f32 %r589, %r38, %r217, %r588; + neg.f32 %r590, %r490; + fma.rn.f32 %r591, %r38, %r218, %r590; + neg.f32 %r592, %r491; + fma.rn.f32 %r593, %r38, %r219, %r592; + neg.f32 %r594, %r492; + fma.rn.f32 %r595, %r38, %r220, %r594; + neg.f32 %r596, %r493; + fma.rn.f32 %r597, %r38, %r221, %r596; + neg.f32 %r598, %r494; + fma.rn.f32 %r599, %r39, %r222, %r598; + neg.f32 %r600, %r495; + fma.rn.f32 %r601, %r39, %r223, %r600; + neg.f32 %r602, %r496; + fma.rn.f32 %r603, %r39, %r224, %r602; + neg.f32 %r604, %r497; + fma.rn.f32 %r605, %r39, %r225, %r604; + neg.f32 %r606, %r498; + fma.rn.f32 %r607, %r39, %r226, %r606; + neg.f32 %r608, %r499; + fma.rn.f32 %r609, %r39, %r227, %r608; + neg.f32 %r610, %r500; + fma.rn.f32 %r611, %r39, %r228, %r610; + neg.f32 %r612, %r501; + fma.rn.f32 %r613, %r39, %r229, %r612; + mov.b64 %rd89, {%r246, %r247}; + mov.b64 %rd90, {%r502, %r503}; + sub.f32x2 %rd91, %rd89, %rd90; + mov.b64 %rd92, {%r248, %r249}; + mov.b64 %rd93, {%r504, %r505}; + sub.f32x2 %rd94, %rd92, %rd93; + mov.b64 %rd95, {%r250, %r251}; + mov.b64 %rd96, {%r506, %r507}; + sub.f32x2 %rd97, %rd95, %rd96; + mov.b64 %rd98, {%r252, %r253}; + mov.b64 %rd99, {%r508, %r509}; + sub.f32x2 %rd100, %rd98, %rd99; + mov.b64 %rd101, {%r254, %r255}; + mov.b64 %rd102, {%r510, %r511}; + sub.f32x2 %rd103, %rd101, %rd102; + mov.b64 %rd104, {%r256, %r257}; + mov.b64 %rd105, {%r512, %r513}; + sub.f32x2 %rd106, %rd104, %rd105; + mov.b64 %rd107, {%r258, %r259}; + mov.b64 %rd108, {%r514, %r515}; + sub.f32x2 %rd109, %rd107, %rd108; + mov.b64 %rd110, {%r260, %r261}; + mov.b64 %rd111, {%r516, %r517}; + sub.f32x2 %rd112, %rd110, %rd111; + .loc 1 108 27 // l2norm.py:108:27 + mov.b64 {%r614, %r615}, %rd91; + cvt.rn.bf16x2.f32 %r98, %r615, %r614; + mov.b64 {%r616, %r617}, %rd94; + cvt.rn.bf16x2.f32 %r99, %r617, %r616; + mov.b64 {%r618, %r619}, %rd97; + cvt.rn.bf16x2.f32 %r100, %r619, %r618; + mov.b64 {%r620, %r621}, %rd100; + cvt.rn.bf16x2.f32 %r101, %r621, %r620; + mov.b64 {%r622, %r623}, %rd103; + cvt.rn.bf16x2.f32 %r102, %r623, %r622; + mov.b64 {%r624, %r625}, %rd106; + cvt.rn.bf16x2.f32 %r103, %r625, %r624; + mov.b64 {%r626, %r627}, %rd109; + cvt.rn.bf16x2.f32 %r104, %r627, %r626; + mov.b64 {%r628, %r629}, %rd112; + cvt.rn.bf16x2.f32 %r105, %r629, %r628; + .loc 1 108 19 // l2norm.py:108:19 + cvt.rn.bf16x2.f32 %r74, %r521, %r519; + cvt.rn.bf16x2.f32 %r75, %r525, %r523; + cvt.rn.bf16x2.f32 %r76, %r529, %r527; + cvt.rn.bf16x2.f32 %r77, %r533, %r531; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd25 + 0 ], { %r74, %r75, %r76, %r77 }; + // end inline asm + cvt.rn.bf16x2.f32 %r78, %r537, %r535; + cvt.rn.bf16x2.f32 %r79, %r541, %r539; + cvt.rn.bf16x2.f32 %r80, %r545, %r543; + cvt.rn.bf16x2.f32 %r81, %r549, %r547; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd26 + 0 ], { %r78, %r79, %r80, %r81 }; + // end inline asm + cvt.rn.bf16x2.f32 %r82, %r553, %r551; + cvt.rn.bf16x2.f32 %r83, %r557, %r555; + cvt.rn.bf16x2.f32 %r84, %r561, %r559; + cvt.rn.bf16x2.f32 %r85, %r565, %r563; + // begin inline asm + @%p3 st.global.v4.b32 [ %rd27 + 0 ], { %r82, %r83, %r84, %r85 }; + // end inline asm + cvt.rn.bf16x2.f32 %r86, %r569, %r567; + cvt.rn.bf16x2.f32 %r87, %r573, %r571; + cvt.rn.bf16x2.f32 %r88, %r577, %r575; + cvt.rn.bf16x2.f32 %r89, %r581, %r579; + // begin inline asm + @%p4 st.global.v4.b32 [ %rd28 + 0 ], { %r86, %r87, %r88, %r89 }; + // end inline asm + cvt.rn.bf16x2.f32 %r90, %r585, %r583; + cvt.rn.bf16x2.f32 %r91, %r589, %r587; + cvt.rn.bf16x2.f32 %r92, %r593, %r591; + cvt.rn.bf16x2.f32 %r93, %r597, %r595; + // begin inline asm + @%p5 st.global.v4.b32 [ %rd29 + 0 ], { %r90, %r91, %r92, %r93 }; + // end inline asm + cvt.rn.bf16x2.f32 %r94, %r601, %r599; + cvt.rn.bf16x2.f32 %r95, %r605, %r603; + cvt.rn.bf16x2.f32 %r96, %r609, %r607; + cvt.rn.bf16x2.f32 %r97, %r613, %r611; + // begin inline asm + @%p6 st.global.v4.b32 [ %rd30 + 0 ], { %r94, %r95, %r96, %r97 }; + // end inline asm + // begin inline asm + @%p7 st.global.v4.b32 [ %rd31 + 0 ], { %r98, %r99, %r100, %r101 }; + // end inline asm + // begin inline asm + @%p8 st.global.v4.b32 [ %rd32 + 0 ], { %r102, %r103, %r104, %r105 }; + // end inline asm + .loc 1 108 4 // l2norm.py:108:4 + ret; +$L__tmp67: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 98 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp66 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 107 // DW_AT_call_line +.b8 55 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp66 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/l2norm_bwd_kernel.source b/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/l2norm_bwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..25926967fd8ce8ce5392a0037c1d84708c2d91fa --- /dev/null +++ b/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/l2norm_bwd_kernel.source @@ -0,0 +1,210 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc47 = loc(unknown) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc54 = loc("y"(#loc)) +#loc55 = loc("rstd"(#loc)) +#loc56 = loc("dy"(#loc)) +#loc57 = loc("dx"(#loc)) +#loc58 = loc("eps"(#loc)) +#loc59 = loc("T"(#loc)) +#loc101 = loc("input"(#loc45)) +#loc102 = loc("a"(#loc50)) +#loc103 = loc("b"(#loc50)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc60) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc61) + %o_t = arith.constant 16 : i32 loc(#loc62) + %o_t_1 = arith.constant 16 : i64 loc(#loc62) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc62) + %o_t_3 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32> loc(#loc63) + %o_t_4 = arith.extsi %o_t_3 : tensor<16xi32> to tensor<16xi64> loc(#loc64) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<16xi64> loc(#loc64) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<16xi64> loc(#loc64) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc65) + %m_t = arith.extsi %T : i32 to i64 loc(#loc66) + %m_t_7 = tt.splat %m_t : i64 -> tensor<16xi64> loc(#loc66) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<16xi64> loc(#loc66) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc67) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc68) + %m_x_10 = arith.constant 128 : i32 loc(#loc69) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc69) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc69) + %m_x_13 = tt.broadcast %m_x : tensor<16x1xi1> -> tensor<16x128xi1> loc(#loc70) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<16x128xi1> loc(#loc70) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<16x128xi1> loc(#loc70) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc71) + %p_y_16 = arith.constant 128 : i32 loc(#loc72) + %p_y_17 = arith.constant 128 : i64 loc(#loc72) + %p_y_18 = arith.constant dense<128> : tensor<16x1xi64> loc(#loc72) + %p_y_19 = arith.muli %p_y, %p_y_18 : tensor<16x1xi64> loc(#loc72) + %p_y_20 = tt.splat %y : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc73) + %p_y_21 = tt.addptr %p_y_20, %p_y_19 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc73) + %p_y_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc74) + %p_y_23 = tt.broadcast %p_y_21 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc75) + %p_y_24 = tt.broadcast %p_y_22 : tensor<1x128xi32> -> tensor<16x128xi32> loc(#loc75) + %p_y_25 = tt.addptr %p_y_23, %p_y_24 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc75) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc76) + %p_rstd_26 = tt.addptr %p_rstd, %o_t_6 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc76) + %p_dy = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc77) + %p_dy_27 = arith.constant 128 : i32 loc(#loc78) + %p_dy_28 = arith.constant 128 : i64 loc(#loc78) + %p_dy_29 = arith.constant dense<128> : tensor<16x1xi64> loc(#loc78) + %p_dy_30 = arith.muli %p_dy, %p_dy_29 : tensor<16x1xi64> loc(#loc78) + %p_dy_31 = tt.splat %dy : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc79) + %p_dy_32 = tt.addptr %p_dy_31, %p_dy_30 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc79) + %p_dy_33 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc80) + %p_dy_34 = tt.broadcast %p_dy_32 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc81) + %p_dy_35 = tt.broadcast %p_dy_33 : tensor<1x128xi32> -> tensor<16x128xi32> loc(#loc81) + %p_dy_36 = tt.addptr %p_dy_34, %p_dy_35 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc81) + %p_dx = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc82) + %p_dx_37 = arith.constant 128 : i32 loc(#loc83) + %p_dx_38 = arith.constant 128 : i64 loc(#loc83) + %p_dx_39 = arith.constant dense<128> : tensor<16x1xi64> loc(#loc83) + %p_dx_40 = arith.muli %p_dx, %p_dx_39 : tensor<16x1xi64> loc(#loc83) + %p_dx_41 = tt.splat %dx : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc84) + %p_dx_42 = tt.addptr %p_dx_41, %p_dx_40 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc84) + %p_dx_43 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc85) + %p_dx_44 = tt.broadcast %p_dx_42 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc86) + %p_dx_45 = tt.broadcast %p_dx_43 : tensor<1x128xi32> -> tensor<16x128xi32> loc(#loc86) + %p_dx_46 = tt.addptr %p_dx_44, %p_dx_45 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc86) + %b_y = arith.constant 0.000000e+00 : f32 loc(#loc87) + %b_y_47 = arith.constant dense<0.000000e+00> : tensor<16x128xf32> loc(#loc87) + %b_y_48 = arith.truncf %b_y_47 : tensor<16x128xf32> to tensor<16x128xbf16> loc(#loc87) + %b_y_49 = tt.load %p_y_25, %m_x_15, %b_y_48 : tensor<16x128x!tt.ptr> loc(#loc87) + %b_y_50 = arith.extf %b_y_49 : tensor<16x128xbf16> to tensor<16x128xf32> loc(#loc88) + %b_rstd = arith.constant 0.000000e+00 : f32 loc(#loc89) + %b_rstd_51 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc89) + %b_rstd_52 = tt.load %p_rstd_26, %m_t_8, %b_rstd_51 : tensor<16x!tt.ptr> loc(#loc89) + %b_dy = arith.constant 0.000000e+00 : f32 loc(#loc90) + %b_dy_53 = arith.constant dense<0.000000e+00> : tensor<16x128xf32> loc(#loc90) + %b_dy_54 = arith.truncf %b_dy_53 : tensor<16x128xf32> to tensor<16x128xbf16> loc(#loc90) + %b_dy_55 = tt.load %p_dy_36, %m_x_15, %b_dy_54 : tensor<16x128x!tt.ptr> loc(#loc90) + %b_dy_56 = arith.extf %b_dy_55 : tensor<16x128xbf16> to tensor<16x128xf32> loc(#loc91) + %b_dx = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc92) + %b_dx_57 = tt.broadcast %b_dx : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc93) + %b_dx_58 = arith.mulf %b_dy_56, %b_dx_57 : tensor<16x128xf32> loc(#loc93) + %b_dx_59 = arith.mulf %b_dy_56, %b_y_50 : tensor<16x128xf32> loc(#loc94) + %b_dx_60 = tt.call @triton.language.standard.sum__fp32S16_128S_c1_cFalse_cNone(%b_dx_59) : (tensor<16x128xf32>) -> tensor<16xf32> loc(#loc95) + %b_dx_61 = tt.expand_dims %b_dx_60 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc96) + %b_dx_62 = tt.broadcast %b_dx_61 : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc97) + %b_dx_63 = arith.mulf %b_dx_62, %b_y_50 : tensor<16x128xf32> loc(#loc97) + %b_dx_64 = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc98) + %b_dx_65 = tt.broadcast %b_dx_64 : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc99) + %b_dx_66 = arith.mulf %b_dx_63, %b_dx_65 : tensor<16x128xf32> loc(#loc99) + %b_dx_67 = arith.subf %b_dx_58, %b_dx_66 : tensor<16x128xf32> loc(#loc100) + %0 = arith.truncf %b_dx_67 : tensor<16x128xf32> to tensor<16x128xbf16> loc(#loc42) + tt.store %p_dx_46, %0, %m_x_15 : tensor<16x128x!tt.ptr> loc(#loc43) + tt.return loc(#loc44) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S16_128S_c1_cFalse_cNone(%input: tensor<16x128xf32> loc("input"(#loc45))) -> tensor<16xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc46) + tt.reduce.return %2 : f32 loc(#loc46) + }) : (tensor<16x128xf32>) -> tensor<16xf32> loc(#loc46) + tt.return %0 : tensor<16xf32> loc(#loc48) + ^bb1: // no predecessors + %1 = ub.poison : tensor<16xf32> loc(#loc49) + tt.return %1 : tensor<16xf32> loc(#loc49) + } loc(#loc45) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc50)), %b: f32 loc("b"(#loc50))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc51) + tt.return %0 : f32 loc(#loc52) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc53) + tt.return %1 : f32 loc(#loc53) + } loc(#loc50) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:31) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:39) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:20) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:31) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:39) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:82) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc60 = loc("i_t"(#loc1)) +#loc61 = loc("i_t"(#loc2)) +#loc62 = loc("o_t"(#loc3)) +#loc63 = loc("o_t"(#loc4)) +#loc64 = loc("o_t"(#loc5)) +#loc65 = loc("o_d"(#loc6)) +#loc66 = loc("m_t"(#loc7)) +#loc67 = loc("m_x"(#loc8)) +#loc68 = loc("m_x"(#loc9)) +#loc69 = loc("m_x"(#loc10)) +#loc70 = loc("m_x"(#loc11)) +#loc71 = loc("p_y"(#loc12)) +#loc72 = loc("p_y"(#loc13)) +#loc73 = loc("p_y"(#loc14)) +#loc74 = loc("p_y"(#loc15)) +#loc75 = loc("p_y"(#loc16)) +#loc76 = loc("p_rstd"(#loc17)) +#loc77 = loc("p_dy"(#loc18)) +#loc78 = loc("p_dy"(#loc19)) +#loc79 = loc("p_dy"(#loc20)) +#loc80 = loc("p_dy"(#loc21)) +#loc81 = loc("p_dy"(#loc22)) +#loc82 = loc("p_dx"(#loc23)) +#loc83 = loc("p_dx"(#loc24)) +#loc84 = loc("p_dx"(#loc25)) +#loc85 = loc("p_dx"(#loc26)) +#loc86 = loc("p_dx"(#loc27)) +#loc87 = loc("b_y"(#loc28)) +#loc88 = loc("b_y"(#loc29)) +#loc89 = loc("b_rstd"(#loc30)) +#loc90 = loc("b_dy"(#loc31)) +#loc91 = loc("b_dy"(#loc32)) +#loc92 = loc("b_dx"(#loc33)) +#loc93 = loc("b_dx"(#loc34)) +#loc94 = loc("b_dx"(#loc35)) +#loc95 = loc("b_dx"(#loc36)) +#loc96 = loc("b_dx"(#loc37)) +#loc97 = loc("b_dx"(#loc38)) +#loc98 = loc("b_dx"(#loc39)) +#loc99 = loc("b_dx"(#loc40)) +#loc100 = loc("b_dx"(#loc41)) diff --git a/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/l2norm_bwd_kernel.ttgir b/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/l2norm_bwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..b9040654b17f6fbd2cc090c4417fff4f6813ed17 --- /dev/null +++ b/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/l2norm_bwd_kernel.ttgir @@ -0,0 +1,146 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [1, 1], order = [1, 0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc39 = loc("y"(#loc)) +#loc40 = loc("rstd"(#loc)) +#loc41 = loc("dy"(#loc)) +#loc42 = loc("dx"(#loc)) +#loc43 = loc("eps"(#loc)) +#loc44 = loc("T"(#loc)) +#loc72 = loc("b_dx"(#loc30)) +#loc78 = loc(callsite(#loc1 at #loc72)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 1 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %c16_i64 = arith.constant 16 : i64 loc(#loc1) + %cst = arith.constant dense<0.000000e+00> : tensor<16x128xbf16, #blocked> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<16x1xi64, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc45) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc46) + %o_t = arith.muli %i_t_3, %c16_i64 : i64 loc(#loc47) + %o_t_4 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc48) + %o_t_5 = arith.extsi %o_t_4 : tensor<16xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %o_t_6 = tt.splat %o_t : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %o_t_7 = arith.addi %o_t_6, %o_t_5 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %m_t = arith.extsi %T : i32 to i64 loc(#loc50) + %m_t_8 = tt.splat %m_t : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc50) + %m_t_9 = arith.cmpi slt, %o_t_7, %m_t_8 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc50) + %m_x = tt.expand_dims %m_t_9 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi1, #blocked> loc(#loc51) + %m_x_10 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc52) + %m_x_11 = tt.expand_dims %m_x_10 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc52) + %m_x_12 = arith.cmpi slt, %m_x_11, %cst_2 : tensor<1x128xi32, #blocked> loc(#loc53) + %m_x_13 = tt.broadcast %m_x : tensor<16x1xi1, #blocked> -> tensor<16x128xi1, #blocked> loc(#loc54) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1, #blocked> -> tensor<16x128xi1, #blocked> loc(#loc54) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<16x128xi1, #blocked> loc(#loc54) + %p_y = tt.expand_dims %o_t_7 {axis = 1 : i32} : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi64, #blocked> loc(#loc55) + %p_y_16 = arith.muli %p_y, %cst_1 : tensor<16x1xi64, #blocked> loc(#loc56) + %p_y_17 = tt.splat %y : !tt.ptr -> tensor<16x1x!tt.ptr, #blocked> loc(#loc57) + %p_y_18 = tt.addptr %p_y_17, %p_y_16 : tensor<16x1x!tt.ptr, #blocked>, tensor<16x1xi64, #blocked> loc(#loc57) + %p_y_19 = tt.broadcast %p_y_18 : tensor<16x1x!tt.ptr, #blocked> -> tensor<16x128x!tt.ptr, #blocked> loc(#loc58) + %p_y_20 = tt.broadcast %m_x_11 : tensor<1x128xi32, #blocked> -> tensor<16x128xi32, #blocked> loc(#loc58) + %p_y_21 = tt.addptr %p_y_19, %p_y_20 : tensor<16x128x!tt.ptr, #blocked>, tensor<16x128xi32, #blocked> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc59) + %p_rstd_22 = tt.addptr %p_rstd, %o_t_7 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>>, tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc59) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<16x1x!tt.ptr, #blocked> loc(#loc60) + %p_dy_23 = tt.addptr %p_dy, %p_y_16 : tensor<16x1x!tt.ptr, #blocked>, tensor<16x1xi64, #blocked> loc(#loc60) + %p_dy_24 = tt.broadcast %p_dy_23 : tensor<16x1x!tt.ptr, #blocked> -> tensor<16x128x!tt.ptr, #blocked> loc(#loc61) + %p_dy_25 = tt.addptr %p_dy_24, %p_y_20 : tensor<16x128x!tt.ptr, #blocked>, tensor<16x128xi32, #blocked> loc(#loc61) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<16x1x!tt.ptr, #blocked> loc(#loc62) + %p_dx_26 = tt.addptr %p_dx, %p_y_16 : tensor<16x1x!tt.ptr, #blocked>, tensor<16x1xi64, #blocked> loc(#loc62) + %p_dx_27 = tt.broadcast %p_dx_26 : tensor<16x1x!tt.ptr, #blocked> -> tensor<16x128x!tt.ptr, #blocked> loc(#loc63) + %p_dx_28 = tt.addptr %p_dx_27, %p_y_20 : tensor<16x128x!tt.ptr, #blocked>, tensor<16x128xi32, #blocked> loc(#loc63) + %b_y = tt.load %p_y_21, %m_x_15, %cst : tensor<16x128x!tt.ptr, #blocked> loc(#loc64) + %b_y_29 = arith.extf %b_y : tensor<16x128xbf16, #blocked> to tensor<16x128xf32, #blocked> loc(#loc65) + %b_rstd = tt.load %p_rstd_22, %m_t_9, %cst_0 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc66) + %b_dy = tt.load %p_dy_25, %m_x_15, %cst : tensor<16x128x!tt.ptr, #blocked> loc(#loc67) + %b_dy_30 = arith.extf %b_dy : tensor<16x128xbf16, #blocked> to tensor<16x128xf32, #blocked> loc(#loc68) + %b_dx = tt.expand_dims %b_rstd {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xf32, #blocked> loc(#loc69) + %b_dx_31 = tt.broadcast %b_dx : tensor<16x1xf32, #blocked> -> tensor<16x128xf32, #blocked> loc(#loc70) + %b_dx_32 = arith.mulf %b_dy_30, %b_dx_31 : tensor<16x128xf32, #blocked> loc(#loc70) + %b_dx_33 = arith.mulf %b_dy_30, %b_y_29 : tensor<16x128xf32, #blocked> loc(#loc71) + %b_dx_34 = "tt.reduce"(%b_dx_33) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_40: f32 loc(callsite(#loc1 at #loc72)), %b_dx_41: f32 loc(callsite(#loc1 at #loc72))): + %b_dx_42 = arith.addf %b_dx_40, %b_dx_41 : f32 loc(#loc79) + tt.reduce.return %b_dx_42 : f32 loc(#loc77) + }) : (tensor<16x128xf32, #blocked>) -> tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc77) + %b_dx_35 = tt.expand_dims %b_dx_34 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xf32, #blocked> loc(#loc73) + %b_dx_36 = tt.broadcast %b_dx_35 : tensor<16x1xf32, #blocked> -> tensor<16x128xf32, #blocked> loc(#loc74) + %b_dx_37 = arith.mulf %b_dx_36, %b_y_29 : tensor<16x128xf32, #blocked> loc(#loc74) + %b_dx_38 = arith.mulf %b_dx_37, %b_dx_31 : tensor<16x128xf32, #blocked> loc(#loc75) + %b_dx_39 = arith.subf %b_dx_32, %b_dx_38 : tensor<16x128xf32, #blocked> loc(#loc76) + %0 = arith.truncf %b_dx_39 : tensor<16x128xf32, #blocked> to tensor<16x128xbf16, #blocked> loc(#loc36) + tt.store %p_dx_28, %0, %m_x_15 : tensor<16x128x!tt.ptr, #blocked> loc(#loc37) + tt.return loc(#loc38) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc45 = loc("i_t"(#loc2)) +#loc46 = loc("i_t"(#loc3)) +#loc47 = loc("o_t"(#loc4)) +#loc48 = loc("o_t"(#loc5)) +#loc49 = loc("o_t"(#loc6)) +#loc50 = loc("m_t"(#loc7)) +#loc51 = loc("m_x"(#loc8)) +#loc52 = loc("m_x"(#loc9)) +#loc53 = loc("m_x"(#loc10)) +#loc54 = loc("m_x"(#loc11)) +#loc55 = loc("p_y"(#loc12)) +#loc56 = loc("p_y"(#loc13)) +#loc57 = loc("p_y"(#loc14)) +#loc58 = loc("p_y"(#loc15)) +#loc59 = loc("p_rstd"(#loc16)) +#loc60 = loc("p_dy"(#loc17)) +#loc61 = loc("p_dy"(#loc18)) +#loc62 = loc("p_dx"(#loc19)) +#loc63 = loc("p_dx"(#loc20)) +#loc64 = loc("b_y"(#loc21)) +#loc65 = loc("b_y"(#loc22)) +#loc66 = loc("b_rstd"(#loc23)) +#loc67 = loc("b_dy"(#loc24)) +#loc68 = loc("b_dy"(#loc25)) +#loc69 = loc("b_dx"(#loc26)) +#loc70 = loc("b_dx"(#loc27)) +#loc71 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc32)) +#loc74 = loc("b_dx"(#loc33)) +#loc75 = loc("b_dx"(#loc34)) +#loc76 = loc("b_dx"(#loc35)) +#loc77 = loc(callsite(#loc29 at #loc72)) +#loc79 = loc(callsite(#loc31 at #loc77)) diff --git a/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/l2norm_bwd_kernel.ttir b/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/l2norm_bwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..6c29416e2d7c989d3b096ee1e98cbabbdbd07e07 --- /dev/null +++ b/triton/gpu0/BNGO5J2W6E57PEZSDE24FBGCX7CBLR5WRJ3L2CFKPFI5L7MKRXRA/l2norm_bwd_kernel.ttir @@ -0,0 +1,147 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc40 = loc("y"(#loc)) +#loc41 = loc("rstd"(#loc)) +#loc42 = loc("dy"(#loc)) +#loc43 = loc("dx"(#loc)) +#loc44 = loc("eps"(#loc)) +#loc45 = loc("T"(#loc)) +#loc74 = loc("b_dx"(#loc31)) +#loc80 = loc(callsite(#loc1 at #loc74)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<0.000000e+00> : tensor<16x128xbf16> loc(#loc1) + %b_rstd = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc46) + %cst_0 = arith.constant dense<128> : tensor<16x1xi64> loc(#loc1) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc47) + %o_t = arith.constant 16 : i64 loc(#loc48) + %i_t = tt.get_program_id x : i32 loc(#loc49) + %i_t_1 = arith.extsi %i_t : i32 to i64 loc(#loc50) + %o_t_2 = arith.muli %i_t_1, %o_t : i64 loc(#loc48) + %o_t_3 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32> loc(#loc51) + %o_t_4 = arith.extsi %o_t_3 : tensor<16xi32> to tensor<16xi64> loc(#loc52) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<16xi64> loc(#loc52) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<16xi64> loc(#loc52) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc53) + %m_t = arith.extsi %T : i32 to i64 loc(#loc54) + %m_t_7 = tt.splat %m_t : i64 -> tensor<16xi64> loc(#loc54) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<16xi64> loc(#loc54) + %m_x_9 = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc55) + %m_x_10 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc56) + %m_x_11 = arith.cmpi slt, %m_x_10, %m_x : tensor<1x128xi32> loc(#loc47) + %m_x_12 = tt.broadcast %m_x_9 : tensor<16x1xi1> -> tensor<16x128xi1> loc(#loc57) + %m_x_13 = tt.broadcast %m_x_11 : tensor<1x128xi1> -> tensor<16x128xi1> loc(#loc57) + %m_x_14 = arith.andi %m_x_12, %m_x_13 : tensor<16x128xi1> loc(#loc57) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc58) + %p_y_15 = arith.muli %p_y, %cst_0 : tensor<16x1xi64> loc(#loc59) + %p_y_16 = tt.splat %y : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc60) + %p_y_17 = tt.addptr %p_y_16, %p_y_15 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc60) + %p_y_18 = tt.broadcast %p_y_17 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc61) + %p_y_19 = tt.broadcast %m_x_10 : tensor<1x128xi32> -> tensor<16x128xi32> loc(#loc61) + %p_y_20 = tt.addptr %p_y_18, %p_y_19 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc61) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc62) + %p_rstd_21 = tt.addptr %p_rstd, %o_t_6 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc62) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc63) + %p_dy_22 = tt.addptr %p_dy, %p_y_15 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc63) + %p_dy_23 = tt.broadcast %p_dy_22 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc64) + %p_dy_24 = tt.addptr %p_dy_23, %p_y_19 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc64) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc65) + %p_dx_25 = tt.addptr %p_dx, %p_y_15 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc65) + %p_dx_26 = tt.broadcast %p_dx_25 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc66) + %p_dx_27 = tt.addptr %p_dx_26, %p_y_19 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc66) + %b_y = tt.load %p_y_20, %m_x_14, %cst : tensor<16x128x!tt.ptr> loc(#loc67) + %b_y_28 = arith.extf %b_y : tensor<16x128xbf16> to tensor<16x128xf32> loc(#loc68) + %b_rstd_29 = tt.load %p_rstd_21, %m_t_8, %b_rstd : tensor<16x!tt.ptr> loc(#loc46) + %b_dy = tt.load %p_dy_24, %m_x_14, %cst : tensor<16x128x!tt.ptr> loc(#loc69) + %b_dy_30 = arith.extf %b_dy : tensor<16x128xbf16> to tensor<16x128xf32> loc(#loc70) + %b_dx = tt.expand_dims %b_rstd_29 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc71) + %b_dx_31 = tt.broadcast %b_dx : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc72) + %b_dx_32 = arith.mulf %b_dy_30, %b_dx_31 : tensor<16x128xf32> loc(#loc72) + %b_dx_33 = arith.mulf %b_dy_30, %b_y_28 : tensor<16x128xf32> loc(#loc73) + %b_dx_34 = "tt.reduce"(%b_dx_33) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_40: f32 loc(callsite(#loc1 at #loc74)), %b_dx_41: f32 loc(callsite(#loc1 at #loc74))): + %b_dx_42 = arith.addf %b_dx_40, %b_dx_41 : f32 loc(#loc81) + tt.reduce.return %b_dx_42 : f32 loc(#loc79) + }) : (tensor<16x128xf32>) -> tensor<16xf32> loc(#loc79) + %b_dx_35 = tt.expand_dims %b_dx_34 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc75) + %b_dx_36 = tt.broadcast %b_dx_35 : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc76) + %b_dx_37 = arith.mulf %b_dx_36, %b_y_28 : tensor<16x128xf32> loc(#loc76) + %b_dx_38 = arith.mulf %b_dx_37, %b_dx_31 : tensor<16x128xf32> loc(#loc77) + %b_dx_39 = arith.subf %b_dx_32, %b_dx_38 : tensor<16x128xf32> loc(#loc78) + %0 = arith.truncf %b_dx_39 : tensor<16x128xf32> to tensor<16x128xbf16> loc(#loc37) + tt.store %p_dx_27, %0, %m_x_14 : tensor<16x128x!tt.ptr> loc(#loc38) + tt.return loc(#loc39) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("b_rstd"(#loc2)) +#loc47 = loc("m_x"(#loc3)) +#loc48 = loc("o_t"(#loc4)) +#loc49 = loc("i_t"(#loc5)) +#loc50 = loc("i_t"(#loc6)) +#loc51 = loc("o_t"(#loc7)) +#loc52 = loc("o_t"(#loc8)) +#loc53 = loc("o_d"(#loc9)) +#loc54 = loc("m_t"(#loc10)) +#loc55 = loc("m_x"(#loc11)) +#loc56 = loc("m_x"(#loc12)) +#loc57 = loc("m_x"(#loc13)) +#loc58 = loc("p_y"(#loc14)) +#loc59 = loc("p_y"(#loc15)) +#loc60 = loc("p_y"(#loc16)) +#loc61 = loc("p_y"(#loc17)) +#loc62 = loc("p_rstd"(#loc18)) +#loc63 = loc("p_dy"(#loc19)) +#loc64 = loc("p_dy"(#loc20)) +#loc65 = loc("p_dx"(#loc21)) +#loc66 = loc("p_dx"(#loc22)) +#loc67 = loc("b_y"(#loc23)) +#loc68 = loc("b_y"(#loc24)) +#loc69 = loc("b_dy"(#loc25)) +#loc70 = loc("b_dy"(#loc26)) +#loc71 = loc("b_dx"(#loc27)) +#loc72 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc29)) +#loc75 = loc("b_dx"(#loc33)) +#loc76 = loc("b_dx"(#loc34)) +#loc77 = loc("b_dx"(#loc35)) +#loc78 = loc("b_dx"(#loc36)) +#loc79 = loc(callsite(#loc30 at #loc74)) +#loc81 = loc(callsite(#loc32 at #loc79)) diff --git a/triton/gpu0/BNNETK7G7W5ON3JOXOQDIYPTUKIDYYRLJ7TSGUPGL4ZAETIBRK6A/l2norm_bwd_kernel.autotune.json b/triton/gpu0/BNNETK7G7W5ON3JOXOQDIYPTUKIDYYRLJ7TSGUPGL4ZAETIBRK6A/l2norm_bwd_kernel.autotune.json new file mode 100644 index 0000000000000000000000000000000000000000..7ff472c9e8a9914a2a36f5102de761552e0621f0 --- /dev/null +++ b/triton/gpu0/BNNETK7G7W5ON3JOXOQDIYPTUKIDYYRLJ7TSGUPGL4ZAETIBRK6A/l2norm_bwd_kernel.autotune.json @@ -0,0 +1 @@ +{"key": [128, 9, "torch.bfloat16", "torch.float32", "torch.bfloat16", "torch.bfloat16"], "configs_timings": [[{"kwargs": {"BT": 8}, "num_warps": 1, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.07168000191450119, 0.07158400118350983, 0.07360000163316727]], [{"kwargs": {"BT": 16}, "num_warps": 1, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.07360000163316727, 0.07168000191450119, 0.0737600028514862]], [{"kwargs": {"BT": 32}, "num_warps": 1, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.0798719972372055, 0.07977599650621414, 0.07993599772453308]], [{"kwargs": {"BT": 64}, "num_warps": 1, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.10639999806880951, 0.10444799810647964, 0.10652799904346466]], [{"kwargs": {"BT": 128}, "num_warps": 1, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.35734400153160095, 0.3563520014286041, 0.3584960103034973]], [{"kwargs": {"BT": 8}, "num_warps": 2, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.0716480016708374, 0.07158400118350983, 0.07174400240182877]], [{"kwargs": {"BT": 16}, "num_warps": 2, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.07174400240182877, 0.07161600142717361, 0.07369600236415863]], [{"kwargs": {"BT": 32}, "num_warps": 2, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.07174400240182877, 0.0716480016708374, 0.07369600236415863]], [{"kwargs": {"BT": 64}, "num_warps": 2, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.0798719972372055, 0.07980799674987793, 0.07996799796819687]], [{"kwargs": {"BT": 128}, "num_warps": 2, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.11260800063610077, 0.1125119999051094, 0.11270400136709213]], [{"kwargs": {"BT": 8}, "num_warps": 4, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.0716480016708374, 0.07158400118350983, 0.07171200215816498]], [{"kwargs": {"BT": 16}, "num_warps": 4, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.07168000191450119, 0.07161600142717361, 0.07177600264549255]], [{"kwargs": {"BT": 32}, "num_warps": 4, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.07366400212049484, 0.07174400240182877, 0.0737600028514862]], [{"kwargs": {"BT": 64}, "num_warps": 4, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.07177600264549255, 0.0716480016708374, 0.07372800260782242]], [{"kwargs": {"BT": 128}, "num_warps": 4, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.08191999793052673, 0.08185599744319916, 0.08198399841785431]], [{"kwargs": {"BT": 8}, "num_warps": 8, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.08803199976682663, 0.08793599903583527, 0.0880960002541542]], [{"kwargs": {"BT": 16}, "num_warps": 8, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.0716480016708374, 0.07155200093984604, 0.07171200215816498]], [{"kwargs": {"BT": 32}, "num_warps": 8, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.07168000191450119, 0.07161600142717361, 0.07174400240182877]], [{"kwargs": {"BT": 64}, "num_warps": 8, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.07372800260782242, 0.07364480197429657, 0.07379200309515]], [{"kwargs": {"BT": 128}, "num_warps": 8, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.07577600330114365, 0.07571200281381607, 0.07583999633789062]], [{"kwargs": {"BT": 8}, "num_warps": 16, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.16176000237464905, 0.16169600188732147, 0.16185599565505981]], [{"kwargs": {"BT": 16}, "num_warps": 16, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.0920960009098053, 0.09203200042247772, 0.09219200164079666]], [{"kwargs": {"BT": 32}, "num_warps": 16, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.07369600236415863, 0.07360000163316727, 0.0737600028514862]], [{"kwargs": {"BT": 64}, "num_warps": 16, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.0716480016708374, 0.07158400118350983, 0.07171200215816498]], [{"kwargs": {"BT": 128}, "num_warps": 16, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.07372800260782242, 0.07366400212049484, 0.07379200309515]]]} \ No newline at end of file diff --git a/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/__grp__l2norm_fwd_kernel.json b/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/__grp__l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..064ebbfb69a598d982ed9931a3ca9815865e79df --- /dev/null +++ b/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/__grp__l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_fwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/l2norm_fwd_kernel.source", "l2norm_fwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/l2norm_fwd_kernel.ttir", "l2norm_fwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/l2norm_fwd_kernel.ttgir", "l2norm_fwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/l2norm_fwd_kernel.llir", "l2norm_fwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/l2norm_fwd_kernel.ptx", "l2norm_fwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/l2norm_fwd_kernel.cubin", "l2norm_fwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/l2norm_fwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/l2norm_fwd_kernel.cubin b/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/l2norm_fwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..9abef151dae160ade70af4a66e9b2ae99e0c9dd7 Binary files /dev/null and b/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/l2norm_fwd_kernel.cubin differ diff --git a/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/l2norm_fwd_kernel.json b/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..d7d60f620156dd1638b1b9fb7d56e6a0227e543f --- /dev/null +++ b/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"hash": "0be7ffe159ec0caf3d9d3f1ba9182f3878c89a8f48cd18073bda9cc0666b565a", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 8, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 256, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_fwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/l2norm_fwd_kernel.llir b/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/l2norm_fwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..a9eb789100099aee9135e15092e32ed65fbb1fea --- /dev/null +++ b/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/l2norm_fwd_kernel.llir @@ -0,0 +1,637 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 +@.str.2 = private unnamed_addr constant [17 x i8] c"__CUDA_PREC_SQRT\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_fwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, float %3, i32 %4, ptr addrspace(1) readnone captures(none) %5, ptr addrspace(1) readnone captures(none) %6) local_unnamed_addr #0 !dbg !5 { + %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %9 = zext nneg i32 %8 to i64, !dbg !14 + %10 = shl nuw nsw i64 %9, 6, !dbg !15 + %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !16 + %12 = and i32 %11, 240, !dbg !16 + %13 = lshr exact i32 %12, 4, !dbg !16 + %14 = or disjoint i32 %13, 16, !dbg !16 + %15 = or disjoint i32 %13, 32, !dbg !16 + %16 = or disjoint i32 %13, 48, !dbg !16 + %17 = zext nneg i32 %13 to i64, !dbg !17 + %18 = zext nneg i32 %14 to i64, !dbg !17 + %19 = zext nneg i32 %15 to i64, !dbg !17 + %20 = zext nneg i32 %16 to i64, !dbg !17 + %21 = or disjoint i64 %10, %17, !dbg !17 + %22 = or disjoint i64 %10, %18, !dbg !17 + %23 = or disjoint i64 %10, %19, !dbg !17 + %24 = or disjoint i64 %10, %20, !dbg !17 + %25 = sext i32 %4 to i64, !dbg !18 + %26 = icmp slt i64 %21, %25, !dbg !18 + %27 = icmp slt i64 %22, %25, !dbg !18 + %28 = icmp slt i64 %23, %25, !dbg !18 + %29 = icmp slt i64 %24, %25, !dbg !18 + %30 = shl nuw nsw i32 %11, 3, !dbg !19 + %31 = and i32 %30, 120, !dbg !19 + %32 = shl nuw nsw i64 %21, 7, !dbg !20 + %33 = shl nuw nsw i64 %22, 7, !dbg !20 + %34 = shl nuw nsw i64 %23, 7, !dbg !20 + %35 = shl nuw nsw i64 %24, 7, !dbg !20 + %36 = getelementptr bfloat, ptr addrspace(1) %0, i64 %32, !dbg !21 + %37 = getelementptr bfloat, ptr addrspace(1) %0, i64 %33, !dbg !21 + %38 = getelementptr bfloat, ptr addrspace(1) %0, i64 %34, !dbg !21 + %39 = getelementptr bfloat, ptr addrspace(1) %0, i64 %35, !dbg !21 + %40 = zext nneg i32 %31 to i64, !dbg !22 + %41 = getelementptr bfloat, ptr addrspace(1) %36, i64 %40, !dbg !22 + %42 = getelementptr bfloat, ptr addrspace(1) %37, i64 %40, !dbg !22 + %43 = getelementptr bfloat, ptr addrspace(1) %38, i64 %40, !dbg !22 + %44 = getelementptr bfloat, ptr addrspace(1) %39, i64 %40, !dbg !22 + %45 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %41, i1 %26) #6, !dbg !23 + %46 = extractvalue { i32, i32, i32, i32 } %45, 0, !dbg !23 + %47 = bitcast i32 %46 to <2 x bfloat>, !dbg !23 + %48 = extractvalue { i32, i32, i32, i32 } %45, 1, !dbg !23 + %49 = bitcast i32 %48 to <2 x bfloat>, !dbg !23 + %50 = extractvalue { i32, i32, i32, i32 } %45, 2, !dbg !23 + %51 = bitcast i32 %50 to <2 x bfloat>, !dbg !23 + %52 = extractvalue { i32, i32, i32, i32 } %45, 3, !dbg !23 + %53 = bitcast i32 %52 to <2 x bfloat>, !dbg !23 + %54 = extractelement <2 x bfloat> %47, i64 0, !dbg !23 + %55 = extractelement <2 x bfloat> %47, i64 1, !dbg !23 + %56 = extractelement <2 x bfloat> %49, i64 0, !dbg !23 + %57 = extractelement <2 x bfloat> %49, i64 1, !dbg !23 + %58 = extractelement <2 x bfloat> %51, i64 0, !dbg !23 + %59 = extractelement <2 x bfloat> %51, i64 1, !dbg !23 + %60 = extractelement <2 x bfloat> %53, i64 0, !dbg !23 + %61 = extractelement <2 x bfloat> %53, i64 1, !dbg !23 + %62 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %42, i1 %27) #6, !dbg !23 + %63 = extractvalue { i32, i32, i32, i32 } %62, 0, !dbg !23 + %64 = bitcast i32 %63 to <2 x bfloat>, !dbg !23 + %65 = extractvalue { i32, i32, i32, i32 } %62, 1, !dbg !23 + %66 = bitcast i32 %65 to <2 x bfloat>, !dbg !23 + %67 = extractvalue { i32, i32, i32, i32 } %62, 2, !dbg !23 + %68 = bitcast i32 %67 to <2 x bfloat>, !dbg !23 + %69 = extractvalue { i32, i32, i32, i32 } %62, 3, !dbg !23 + %70 = bitcast i32 %69 to <2 x bfloat>, !dbg !23 + %71 = extractelement <2 x bfloat> %64, i64 0, !dbg !23 + %72 = extractelement <2 x bfloat> %64, i64 1, !dbg !23 + %73 = extractelement <2 x bfloat> %66, i64 0, !dbg !23 + %74 = extractelement <2 x bfloat> %66, i64 1, !dbg !23 + %75 = extractelement <2 x bfloat> %68, i64 0, !dbg !23 + %76 = extractelement <2 x bfloat> %68, i64 1, !dbg !23 + %77 = extractelement <2 x bfloat> %70, i64 0, !dbg !23 + %78 = extractelement <2 x bfloat> %70, i64 1, !dbg !23 + %79 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %43, i1 %28) #6, !dbg !23 + %80 = extractvalue { i32, i32, i32, i32 } %79, 0, !dbg !23 + %81 = bitcast i32 %80 to <2 x bfloat>, !dbg !23 + %82 = extractvalue { i32, i32, i32, i32 } %79, 1, !dbg !23 + %83 = bitcast i32 %82 to <2 x bfloat>, !dbg !23 + %84 = extractvalue { i32, i32, i32, i32 } %79, 2, !dbg !23 + %85 = bitcast i32 %84 to <2 x bfloat>, !dbg !23 + %86 = extractvalue { i32, i32, i32, i32 } %79, 3, !dbg !23 + %87 = bitcast i32 %86 to <2 x bfloat>, !dbg !23 + %88 = extractelement <2 x bfloat> %81, i64 0, !dbg !23 + %89 = extractelement <2 x bfloat> %81, i64 1, !dbg !23 + %90 = extractelement <2 x bfloat> %83, i64 0, !dbg !23 + %91 = extractelement <2 x bfloat> %83, i64 1, !dbg !23 + %92 = extractelement <2 x bfloat> %85, i64 0, !dbg !23 + %93 = extractelement <2 x bfloat> %85, i64 1, !dbg !23 + %94 = extractelement <2 x bfloat> %87, i64 0, !dbg !23 + %95 = extractelement <2 x bfloat> %87, i64 1, !dbg !23 + %96 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %44, i1 %29) #6, !dbg !23 + %97 = extractvalue { i32, i32, i32, i32 } %96, 0, !dbg !23 + %98 = bitcast i32 %97 to <2 x bfloat>, !dbg !23 + %99 = extractvalue { i32, i32, i32, i32 } %96, 1, !dbg !23 + %100 = bitcast i32 %99 to <2 x bfloat>, !dbg !23 + %101 = extractvalue { i32, i32, i32, i32 } %96, 2, !dbg !23 + %102 = bitcast i32 %101 to <2 x bfloat>, !dbg !23 + %103 = extractvalue { i32, i32, i32, i32 } %96, 3, !dbg !23 + %104 = bitcast i32 %103 to <2 x bfloat>, !dbg !23 + %105 = extractelement <2 x bfloat> %98, i64 0, !dbg !23 + %106 = extractelement <2 x bfloat> %98, i64 1, !dbg !23 + %107 = extractelement <2 x bfloat> %100, i64 0, !dbg !23 + %108 = extractelement <2 x bfloat> %100, i64 1, !dbg !23 + %109 = extractelement <2 x bfloat> %102, i64 0, !dbg !23 + %110 = extractelement <2 x bfloat> %102, i64 1, !dbg !23 + %111 = extractelement <2 x bfloat> %104, i64 0, !dbg !23 + %112 = extractelement <2 x bfloat> %104, i64 1, !dbg !23 + %113 = fpext bfloat %54 to float, !dbg !24 + %114 = fpext bfloat %55 to float, !dbg !24 + %115 = fpext bfloat %56 to float, !dbg !24 + %116 = fpext bfloat %57 to float, !dbg !24 + %117 = fpext bfloat %58 to float, !dbg !24 + %118 = fpext bfloat %59 to float, !dbg !24 + %119 = fpext bfloat %60 to float, !dbg !24 + %120 = fpext bfloat %61 to float, !dbg !24 + %121 = fpext bfloat %71 to float, !dbg !24 + %122 = fpext bfloat %72 to float, !dbg !24 + %123 = fpext bfloat %73 to float, !dbg !24 + %124 = fpext bfloat %74 to float, !dbg !24 + %125 = fpext bfloat %75 to float, !dbg !24 + %126 = fpext bfloat %76 to float, !dbg !24 + %127 = fpext bfloat %77 to float, !dbg !24 + %128 = fpext bfloat %78 to float, !dbg !24 + %129 = fpext bfloat %88 to float, !dbg !24 + %130 = fpext bfloat %89 to float, !dbg !24 + %131 = fpext bfloat %90 to float, !dbg !24 + %132 = fpext bfloat %91 to float, !dbg !24 + %133 = fpext bfloat %92 to float, !dbg !24 + %134 = fpext bfloat %93 to float, !dbg !24 + %135 = fpext bfloat %94 to float, !dbg !24 + %136 = fpext bfloat %95 to float, !dbg !24 + %137 = fpext bfloat %105 to float, !dbg !24 + %138 = fpext bfloat %106 to float, !dbg !24 + %139 = fpext bfloat %107 to float, !dbg !24 + %140 = fpext bfloat %108 to float, !dbg !24 + %141 = fpext bfloat %109 to float, !dbg !24 + %142 = fpext bfloat %110 to float, !dbg !24 + %143 = fpext bfloat %111 to float, !dbg !24 + %144 = fpext bfloat %112 to float, !dbg !24 + %145 = fmul float %113, %113, !dbg !25 + %146 = fmul float %114, %114, !dbg !25 + %147 = fmul float %115, %115, !dbg !25 + %148 = fmul float %116, %116, !dbg !25 + %149 = fmul float %117, %117, !dbg !25 + %150 = fmul float %118, %118, !dbg !25 + %151 = fmul float %119, %119, !dbg !25 + %152 = fmul float %120, %120, !dbg !25 + %153 = fmul float %121, %121, !dbg !25 + %154 = fmul float %122, %122, !dbg !25 + %155 = fmul float %123, %123, !dbg !25 + %156 = fmul float %124, %124, !dbg !25 + %157 = fmul float %125, %125, !dbg !25 + %158 = fmul float %126, %126, !dbg !25 + %159 = fmul float %127, %127, !dbg !25 + %160 = fmul float %128, %128, !dbg !25 + %161 = fmul float %129, %129, !dbg !25 + %162 = fmul float %130, %130, !dbg !25 + %163 = fmul float %131, %131, !dbg !25 + %164 = fmul float %132, %132, !dbg !25 + %165 = fmul float %133, %133, !dbg !25 + %166 = fmul float %134, %134, !dbg !25 + %167 = fmul float %135, %135, !dbg !25 + %168 = fmul float %136, %136, !dbg !25 + %169 = fmul float %137, %137, !dbg !25 + %170 = fmul float %138, %138, !dbg !25 + %171 = fmul float %139, %139, !dbg !25 + %172 = fmul float %140, %140, !dbg !25 + %173 = fmul float %141, %141, !dbg !25 + %174 = fmul float %142, %142, !dbg !25 + %175 = fmul float %143, %143, !dbg !25 + %176 = fmul float %144, %144, !dbg !25 + %177 = fadd float %145, %146, !dbg !26 + %178 = fadd float %147, %177, !dbg !26 + %179 = fadd float %148, %178, !dbg !26 + %180 = fadd float %149, %179, !dbg !26 + %181 = fadd float %150, %180, !dbg !26 + %182 = fadd float %151, %181, !dbg !26 + %183 = fadd float %152, %182, !dbg !26 + %184 = fadd float %153, %154, !dbg !26 + %185 = fadd float %155, %184, !dbg !26 + %186 = fadd float %156, %185, !dbg !26 + %187 = fadd float %157, %186, !dbg !26 + %188 = fadd float %158, %187, !dbg !26 + %189 = fadd float %159, %188, !dbg !26 + %190 = fadd float %160, %189, !dbg !26 + %191 = fadd float %161, %162, !dbg !26 + %192 = fadd float %163, %191, !dbg !26 + %193 = fadd float %164, %192, !dbg !26 + %194 = fadd float %165, %193, !dbg !26 + %195 = fadd float %166, %194, !dbg !26 + %196 = fadd float %167, %195, !dbg !26 + %197 = fadd float %168, %196, !dbg !26 + %198 = fadd float %169, %170, !dbg !26 + %199 = fadd float %171, %198, !dbg !26 + %200 = fadd float %172, %199, !dbg !26 + %201 = fadd float %173, %200, !dbg !26 + %202 = fadd float %174, %201, !dbg !26 + %203 = fadd float %175, %202, !dbg !26 + %204 = fadd float %176, %203, !dbg !26 + %205 = bitcast float %183 to i32, !dbg !29 + %206 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %205, i32 8, i32 31), !dbg !29 + %207 = bitcast i32 %206 to float, !dbg !29 + %208 = fadd float %183, %207, !dbg !26 + %209 = bitcast float %208 to i32, !dbg !29 + %210 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %209, i32 4, i32 31), !dbg !29 + %211 = bitcast i32 %210 to float, !dbg !29 + %212 = fadd float %208, %211, !dbg !26 + %213 = bitcast float %212 to i32, !dbg !29 + %214 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %213, i32 2, i32 31), !dbg !29 + %215 = bitcast i32 %214 to float, !dbg !29 + %216 = fadd float %212, %215, !dbg !26 + %217 = bitcast float %216 to i32, !dbg !29 + %218 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %217, i32 1, i32 31), !dbg !29 + %219 = bitcast i32 %218 to float, !dbg !29 + %220 = fadd float %216, %219, !dbg !26 + %221 = bitcast float %190 to i32, !dbg !29 + %222 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %221, i32 8, i32 31), !dbg !29 + %223 = bitcast i32 %222 to float, !dbg !29 + %224 = fadd float %190, %223, !dbg !26 + %225 = bitcast float %224 to i32, !dbg !29 + %226 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %225, i32 4, i32 31), !dbg !29 + %227 = bitcast i32 %226 to float, !dbg !29 + %228 = fadd float %224, %227, !dbg !26 + %229 = bitcast float %228 to i32, !dbg !29 + %230 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %229, i32 2, i32 31), !dbg !29 + %231 = bitcast i32 %230 to float, !dbg !29 + %232 = fadd float %228, %231, !dbg !26 + %233 = bitcast float %232 to i32, !dbg !29 + %234 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %233, i32 1, i32 31), !dbg !29 + %235 = bitcast i32 %234 to float, !dbg !29 + %236 = fadd float %232, %235, !dbg !26 + %237 = bitcast float %197 to i32, !dbg !29 + %238 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %237, i32 8, i32 31), !dbg !29 + %239 = bitcast i32 %238 to float, !dbg !29 + %240 = fadd float %197, %239, !dbg !26 + %241 = bitcast float %240 to i32, !dbg !29 + %242 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %241, i32 4, i32 31), !dbg !29 + %243 = bitcast i32 %242 to float, !dbg !29 + %244 = fadd float %240, %243, !dbg !26 + %245 = bitcast float %244 to i32, !dbg !29 + %246 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %245, i32 2, i32 31), !dbg !29 + %247 = bitcast i32 %246 to float, !dbg !29 + %248 = fadd float %244, %247, !dbg !26 + %249 = bitcast float %248 to i32, !dbg !29 + %250 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %249, i32 1, i32 31), !dbg !29 + %251 = bitcast i32 %250 to float, !dbg !29 + %252 = fadd float %248, %251, !dbg !26 + %253 = bitcast float %204 to i32, !dbg !29 + %254 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %253, i32 8, i32 31), !dbg !29 + %255 = bitcast i32 %254 to float, !dbg !29 + %256 = fadd float %204, %255, !dbg !26 + %257 = bitcast float %256 to i32, !dbg !29 + %258 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %257, i32 4, i32 31), !dbg !29 + %259 = bitcast i32 %258 to float, !dbg !29 + %260 = fadd float %256, %259, !dbg !26 + %261 = bitcast float %260 to i32, !dbg !29 + %262 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %261, i32 2, i32 31), !dbg !29 + %263 = bitcast i32 %262 to float, !dbg !29 + %264 = fadd float %260, %263, !dbg !26 + %265 = bitcast float %264 to i32, !dbg !29 + %266 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %265, i32 1, i32 31), !dbg !29 + %267 = bitcast i32 %266 to float, !dbg !29 + %268 = fadd float %264, %267, !dbg !26 + %269 = fadd float %3, %220, !dbg !32 + %270 = fadd float %3, %236, !dbg !32 + %271 = fadd float %3, %252, !dbg !32 + %272 = fadd float %3, %268, !dbg !32 + %273 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i = icmp eq i32 %273, 0, !dbg !33 + %274 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i = icmp eq i32 %274, 0, !dbg !33 + br i1 %.not.i, label %280, label %275, !dbg !33 + +275: ; preds = %7 + br i1 %.not1.i, label %278, label %276, !dbg !33 + +276: ; preds = %275 + %277 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %269) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +278: ; preds = %275 + %279 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %269) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +280: ; preds = %7 + br i1 %.not1.i, label %283, label %281, !dbg !33 + +281: ; preds = %280 + %282 = tail call float @llvm.nvvm.sqrt.rn.f(float %269) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +283: ; preds = %280 + %284 = tail call float @llvm.nvvm.sqrt.approx.f(float %269) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +__nv_sqrtf.exit: ; preds = %276, %278, %281, %283 + %.0.i = phi float [ %277, %276 ], [ %279, %278 ], [ %282, %281 ], [ %284, %283 ], !dbg !33 + %285 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i1 = icmp eq i32 %285, 0, !dbg !33 + %286 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i4 = icmp eq i32 %286, 0, !dbg !33 + br i1 %.not.i1, label %292, label %287, !dbg !33 + +287: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %290, label %288, !dbg !33 + +288: ; preds = %287 + %289 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %270) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +290: ; preds = %287 + %291 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %270) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +292: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %295, label %293, !dbg !33 + +293: ; preds = %292 + %294 = tail call float @llvm.nvvm.sqrt.rn.f(float %270) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +295: ; preds = %292 + %296 = tail call float @llvm.nvvm.sqrt.approx.f(float %270) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +__nv_sqrtf.exit5: ; preds = %288, %290, %293, %295 + %.0.i3 = phi float [ %289, %288 ], [ %291, %290 ], [ %294, %293 ], [ %296, %295 ], !dbg !33 + %297 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i6 = icmp eq i32 %297, 0, !dbg !33 + %298 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i9 = icmp eq i32 %298, 0, !dbg !33 + br i1 %.not.i6, label %304, label %299, !dbg !33 + +299: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %302, label %300, !dbg !33 + +300: ; preds = %299 + %301 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %271) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +302: ; preds = %299 + %303 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %271) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +304: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %307, label %305, !dbg !33 + +305: ; preds = %304 + %306 = tail call float @llvm.nvvm.sqrt.rn.f(float %271) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +307: ; preds = %304 + %308 = tail call float @llvm.nvvm.sqrt.approx.f(float %271) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +__nv_sqrtf.exit10: ; preds = %300, %302, %305, %307 + %.0.i8 = phi float [ %301, %300 ], [ %303, %302 ], [ %306, %305 ], [ %308, %307 ], !dbg !33 + %309 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i11 = icmp eq i32 %309, 0, !dbg !33 + %310 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i14 = icmp eq i32 %310, 0, !dbg !33 + br i1 %.not.i11, label %316, label %311, !dbg !33 + +311: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %314, label %312, !dbg !33 + +312: ; preds = %311 + %313 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %272) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +314: ; preds = %311 + %315 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %272) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +316: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %319, label %317, !dbg !33 + +317: ; preds = %316 + %318 = tail call float @llvm.nvvm.sqrt.rn.f(float %272) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +319: ; preds = %316 + %320 = tail call float @llvm.nvvm.sqrt.approx.f(float %272) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +__nv_sqrtf.exit15: ; preds = %312, %314, %317, %319 + %.0.i13 = phi float [ %313, %312 ], [ %315, %314 ], [ %318, %317 ], [ %320, %319 ], !dbg !33 + %321 = and i32 %11, 63, !dbg !16 + %322 = zext nneg i32 %321 to i64, !dbg !17 + %323 = or disjoint i64 %10, %322, !dbg !17 + %324 = getelementptr float, ptr addrspace(1) %2, i64 %323, !dbg !34 + %325 = getelementptr bfloat, ptr addrspace(1) %1, i64 %35, !dbg !35 + %326 = getelementptr bfloat, ptr addrspace(1) %325, i64 %40, !dbg !36 + %327 = getelementptr bfloat, ptr addrspace(1) %1, i64 %34, !dbg !35 + %328 = getelementptr bfloat, ptr addrspace(1) %327, i64 %40, !dbg !36 + %329 = getelementptr bfloat, ptr addrspace(1) %1, i64 %33, !dbg !35 + %330 = getelementptr bfloat, ptr addrspace(1) %329, i64 %40, !dbg !36 + %331 = getelementptr bfloat, ptr addrspace(1) %1, i64 %32, !dbg !35 + %332 = getelementptr bfloat, ptr addrspace(1) %331, i64 %40, !dbg !36 + %333 = icmp slt i64 %323, %25, !dbg !18 + %334 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i), !dbg !37 + %335 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i3), !dbg !37 + %336 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i8), !dbg !37 + %337 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i13), !dbg !37 + %338 = fmul float %334, %113, !dbg !38 + %339 = fmul float %334, %114, !dbg !38 + %340 = fmul float %334, %115, !dbg !38 + %341 = fmul float %334, %116, !dbg !38 + %342 = fmul float %334, %117, !dbg !38 + %343 = fmul float %334, %118, !dbg !38 + %344 = fmul float %334, %119, !dbg !38 + %345 = fmul float %334, %120, !dbg !38 + %346 = fmul float %335, %121, !dbg !38 + %347 = fmul float %335, %122, !dbg !38 + %348 = fmul float %335, %123, !dbg !38 + %349 = fmul float %335, %124, !dbg !38 + %350 = fmul float %335, %125, !dbg !38 + %351 = fmul float %335, %126, !dbg !38 + %352 = fmul float %335, %127, !dbg !38 + %353 = fmul float %335, %128, !dbg !38 + %354 = fmul float %336, %129, !dbg !38 + %355 = fmul float %336, %130, !dbg !38 + %356 = fmul float %336, %131, !dbg !38 + %357 = fmul float %336, %132, !dbg !38 + %358 = fmul float %336, %133, !dbg !38 + %359 = fmul float %336, %134, !dbg !38 + %360 = fmul float %336, %135, !dbg !38 + %361 = fmul float %336, %136, !dbg !38 + %362 = fmul float %337, %137, !dbg !38 + %363 = fmul float %337, %138, !dbg !38 + %364 = fmul float %337, %139, !dbg !38 + %365 = fmul float %337, %140, !dbg !38 + %366 = fmul float %337, %141, !dbg !38 + %367 = fmul float %337, %142, !dbg !38 + %368 = fmul float %337, %143, !dbg !38 + %369 = fmul float %337, %144, !dbg !38 + %370 = fptrunc float %338 to bfloat, !dbg !39 + %371 = fptrunc float %339 to bfloat, !dbg !39 + %372 = fptrunc float %340 to bfloat, !dbg !39 + %373 = fptrunc float %341 to bfloat, !dbg !39 + %374 = fptrunc float %342 to bfloat, !dbg !39 + %375 = fptrunc float %343 to bfloat, !dbg !39 + %376 = fptrunc float %344 to bfloat, !dbg !39 + %377 = fptrunc float %345 to bfloat, !dbg !39 + %378 = fptrunc float %346 to bfloat, !dbg !39 + %379 = fptrunc float %347 to bfloat, !dbg !39 + %380 = fptrunc float %348 to bfloat, !dbg !39 + %381 = fptrunc float %349 to bfloat, !dbg !39 + %382 = fptrunc float %350 to bfloat, !dbg !39 + %383 = fptrunc float %351 to bfloat, !dbg !39 + %384 = fptrunc float %352 to bfloat, !dbg !39 + %385 = fptrunc float %353 to bfloat, !dbg !39 + %386 = fptrunc float %354 to bfloat, !dbg !39 + %387 = fptrunc float %355 to bfloat, !dbg !39 + %388 = fptrunc float %356 to bfloat, !dbg !39 + %389 = fptrunc float %357 to bfloat, !dbg !39 + %390 = fptrunc float %358 to bfloat, !dbg !39 + %391 = fptrunc float %359 to bfloat, !dbg !39 + %392 = fptrunc float %360 to bfloat, !dbg !39 + %393 = fptrunc float %361 to bfloat, !dbg !39 + %394 = fptrunc float %362 to bfloat, !dbg !39 + %395 = fptrunc float %363 to bfloat, !dbg !39 + %396 = fptrunc float %364 to bfloat, !dbg !39 + %397 = fptrunc float %365 to bfloat, !dbg !39 + %398 = fptrunc float %366 to bfloat, !dbg !39 + %399 = fptrunc float %367 to bfloat, !dbg !39 + %400 = fptrunc float %368 to bfloat, !dbg !39 + %401 = fptrunc float %369 to bfloat, !dbg !39 + %402 = insertelement <2 x bfloat> poison, bfloat %370, i64 0, !dbg !40 + %403 = insertelement <2 x bfloat> %402, bfloat %371, i64 1, !dbg !40 + %404 = bitcast <2 x bfloat> %403 to i32, !dbg !40 + %405 = insertelement <2 x bfloat> poison, bfloat %372, i64 0, !dbg !40 + %406 = insertelement <2 x bfloat> %405, bfloat %373, i64 1, !dbg !40 + %407 = bitcast <2 x bfloat> %406 to i32, !dbg !40 + %408 = insertelement <2 x bfloat> poison, bfloat %374, i64 0, !dbg !40 + %409 = insertelement <2 x bfloat> %408, bfloat %375, i64 1, !dbg !40 + %410 = bitcast <2 x bfloat> %409 to i32, !dbg !40 + %411 = insertelement <2 x bfloat> poison, bfloat %376, i64 0, !dbg !40 + %412 = insertelement <2 x bfloat> %411, bfloat %377, i64 1, !dbg !40 + %413 = bitcast <2 x bfloat> %412 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %404, i32 %407, i32 %410, i32 %413, ptr addrspace(1) %332, i1 %26) #6, !dbg !40 + %414 = insertelement <2 x bfloat> poison, bfloat %378, i64 0, !dbg !40 + %415 = insertelement <2 x bfloat> %414, bfloat %379, i64 1, !dbg !40 + %416 = bitcast <2 x bfloat> %415 to i32, !dbg !40 + %417 = insertelement <2 x bfloat> poison, bfloat %380, i64 0, !dbg !40 + %418 = insertelement <2 x bfloat> %417, bfloat %381, i64 1, !dbg !40 + %419 = bitcast <2 x bfloat> %418 to i32, !dbg !40 + %420 = insertelement <2 x bfloat> poison, bfloat %382, i64 0, !dbg !40 + %421 = insertelement <2 x bfloat> %420, bfloat %383, i64 1, !dbg !40 + %422 = bitcast <2 x bfloat> %421 to i32, !dbg !40 + %423 = insertelement <2 x bfloat> poison, bfloat %384, i64 0, !dbg !40 + %424 = insertelement <2 x bfloat> %423, bfloat %385, i64 1, !dbg !40 + %425 = bitcast <2 x bfloat> %424 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %416, i32 %419, i32 %422, i32 %425, ptr addrspace(1) %330, i1 %27) #6, !dbg !40 + %426 = insertelement <2 x bfloat> poison, bfloat %386, i64 0, !dbg !40 + %427 = insertelement <2 x bfloat> %426, bfloat %387, i64 1, !dbg !40 + %428 = bitcast <2 x bfloat> %427 to i32, !dbg !40 + %429 = insertelement <2 x bfloat> poison, bfloat %388, i64 0, !dbg !40 + %430 = insertelement <2 x bfloat> %429, bfloat %389, i64 1, !dbg !40 + %431 = bitcast <2 x bfloat> %430 to i32, !dbg !40 + %432 = insertelement <2 x bfloat> poison, bfloat %390, i64 0, !dbg !40 + %433 = insertelement <2 x bfloat> %432, bfloat %391, i64 1, !dbg !40 + %434 = bitcast <2 x bfloat> %433 to i32, !dbg !40 + %435 = insertelement <2 x bfloat> poison, bfloat %392, i64 0, !dbg !40 + %436 = insertelement <2 x bfloat> %435, bfloat %393, i64 1, !dbg !40 + %437 = bitcast <2 x bfloat> %436 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %428, i32 %431, i32 %434, i32 %437, ptr addrspace(1) %328, i1 %28) #6, !dbg !40 + %438 = insertelement <2 x bfloat> poison, bfloat %394, i64 0, !dbg !40 + %439 = insertelement <2 x bfloat> %438, bfloat %395, i64 1, !dbg !40 + %440 = bitcast <2 x bfloat> %439 to i32, !dbg !40 + %441 = insertelement <2 x bfloat> poison, bfloat %396, i64 0, !dbg !40 + %442 = insertelement <2 x bfloat> %441, bfloat %397, i64 1, !dbg !40 + %443 = bitcast <2 x bfloat> %442 to i32, !dbg !40 + %444 = insertelement <2 x bfloat> poison, bfloat %398, i64 0, !dbg !40 + %445 = insertelement <2 x bfloat> %444, bfloat %399, i64 1, !dbg !40 + %446 = bitcast <2 x bfloat> %445 to i32, !dbg !40 + %447 = insertelement <2 x bfloat> poison, bfloat %400, i64 0, !dbg !40 + %448 = insertelement <2 x bfloat> %447, bfloat %401, i64 1, !dbg !40 + %449 = bitcast <2 x bfloat> %448 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %440, i32 %443, i32 %446, i32 %449, ptr addrspace(1) %326, i1 %29) #6, !dbg !40 + %450 = lshr exact i32 %12, 1, !dbg !41 + %451 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %450, !dbg !41 + %452 = bitcast float %334 to i32, !dbg !41 + %453 = bitcast float %335 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v2.b32 [ $0 + 0 ], { $1, $2 };", "r,r,r"(ptr addrspace(3) %451, i32 %452, i32 %453) #6, !dbg !41 + %454 = getelementptr inbounds nuw i8, ptr addrspace(3) %451, i32 128, !dbg !41 + %455 = bitcast float %336 to i32, !dbg !41 + %456 = bitcast float %337 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v2.b32 [ $0 + 0 ], { $1, $2 };", "r,r,r"(ptr addrspace(3) nonnull %454, i32 %455, i32 %456) #6, !dbg !41 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !41 + %457 = lshr i32 %11, 2, !dbg !41 + %458 = and i32 %457, 4, !dbg !41 + %459 = shl nuw nsw i32 %11, 2, !dbg !41 + %460 = and i32 %459, 128, !dbg !41 + %461 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %458, !dbg !41 + %462 = getelementptr inbounds nuw i8, ptr addrspace(3) %461, i32 %460, !dbg !41 + %463 = getelementptr inbounds nuw i8, ptr addrspace(3) %462, i32 %31, !dbg !41 + %464 = load i32, ptr addrspace(3) %463, align 4, !dbg !41 + %465 = and i32 %11, 192, !dbg !41 + %466 = icmp eq i32 %465, 0, !dbg !41 + %467 = and i1 %466, %333, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %464, ptr addrspace(1) %324, i1 %467) #6, !dbg !41 + ret void, !dbg !42 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.div.full(float, float) #3 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #4 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #5 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.f(float) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="256" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) } +attributes #4 = { convergent nocallback nounwind } +attributes #5 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #6 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "l2norm_fwd_kernel", linkageName: "l2norm_fwd_kernel", scope: !1, file: !1, line: 69, type: !6, scopeLine: 69, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !10, !11, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 70, column: 24, scope: !5) +!14 = !DILocation(line: 70, column: 30, scope: !5) +!15 = !DILocation(line: 71, column: 16, scope: !5) +!16 = !DILocation(line: 71, column: 34, scope: !5) +!17 = !DILocation(line: 71, column: 21, scope: !5) +!18 = !DILocation(line: 73, column: 16, scope: !5) +!19 = !DILocation(line: 74, column: 30, scope: !5) +!20 = !DILocation(line: 75, column: 29, scope: !5) +!21 = !DILocation(line: 75, column: 14, scope: !5) +!22 = !DILocation(line: 75, column: 33, scope: !5) +!23 = !DILocation(line: 79, column: 18, scope: !5) +!24 = !DILocation(line: 79, column: 47, scope: !5) +!25 = !DILocation(line: 80, column: 38, scope: !5) +!26 = !DILocation(line: 263, column: 15, scope: !27, inlinedAt: !29) +!27 = distinct !DILexicalBlockFile(scope: !5, file: !28, discriminator: 0) +!28 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!29 = !DILocation(line: 293, column: 36, scope: !27, inlinedAt: !30) +!30 = !DILocation(line: 80, column: 43, scope: !31) +!31 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!32 = !DILocation(line: 80, column: 48, scope: !5) +!33 = !DILocation(line: 80, column: 25, scope: !5) +!34 = !DILocation(line: 77, column: 20, scope: !5) +!35 = !DILocation(line: 76, column: 14, scope: !5) +!36 = !DILocation(line: 76, column: 33, scope: !5) +!37 = !DILocation(line: 80, column: 17, scope: !5) +!38 = !DILocation(line: 81, column: 16, scope: !5) +!39 = !DILocation(line: 83, column: 25, scope: !5) +!40 = !DILocation(line: 83, column: 18, scope: !5) +!41 = !DILocation(line: 84, column: 21, scope: !5) +!42 = !DILocation(line: 84, column: 4, scope: !5) diff --git a/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/l2norm_fwd_kernel.ptx b/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/l2norm_fwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..fecbe610272e5b85fe448594c8a7eecfa8c24ddb --- /dev/null +++ b/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/l2norm_fwd_kernel.ptx @@ -0,0 +1,648 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_fwd_kernel // -- Begin function l2norm_fwd_kernel +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; +.global .align 1 .b8 _$_str_$_2[17] = {95, 95, 67, 85, 68, 65, 95, 80, 82, 69, 67, 95, 83, 81, 82, 84}; + // @l2norm_fwd_kernel +.visible .entry l2norm_fwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_2, + .param .f32 l2norm_fwd_kernel_param_3, + .param .u32 l2norm_fwd_kernel_param_4, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_6 +) +.reqntid 256 +{ + .reg .pred %p<8>; + .reg .b16 %rs<33>; + .reg .b32 %r<199>; + .reg .b64 %rd<39>; + .loc 1 69 0 // l2norm.py:69:0 +$L__func_begin0: + .loc 1 69 0 // l2norm.py:69:0 + +// %bb.0: // %__nv_sqrtf.exit + ld.param.b64 %rd10, [l2norm_fwd_kernel_param_0]; + ld.param.b64 %rd11, [l2norm_fwd_kernel_param_1]; +$L__tmp0: + .loc 1 70 24 // l2norm.py:70:24 + mov.u32 %r41, %ctaid.x; + ld.param.b64 %rd12, [l2norm_fwd_kernel_param_2]; + .loc 1 71 16 // l2norm.py:71:16 + mul.wide.u32 %rd13, %r41, 64; + ld.param.b32 %r42, [l2norm_fwd_kernel_param_3]; + ld.param.s32 %rd14, [l2norm_fwd_kernel_param_4]; + .loc 1 71 34 // l2norm.py:71:34 + mov.u32 %r43, %tid.x; + and.b32 %r44, %r43, 240; + bfe.u32 %r45, %r43, 4, 4; + or.b32 %r46, %r45, 16; + or.b32 %r47, %r45, 32; + or.b32 %r48, %r45, 48; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd15, %r45; + cvt.u64.u32 %rd16, %r46; + cvt.u64.u32 %rd17, %r47; + cvt.u64.u32 %rd18, %r48; + or.b64 %rd19, %rd13, %rd15; + or.b64 %rd20, %rd13, %rd16; + or.b64 %rd21, %rd13, %rd17; + or.b64 %rd22, %rd13, %rd18; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p1, %rd19, %rd14; + setp.lt.s64 %p2, %rd20, %rd14; + setp.lt.s64 %p3, %rd21, %rd14; + setp.lt.s64 %p4, %rd22, %rd14; + .loc 1 74 30 // l2norm.py:74:30 + shl.b32 %r49, %r43, 3; + and.b32 %r50, %r49, 120; + .loc 1 75 14 // l2norm.py:75:14 + shl.b64 %rd23, %rd19, 8; + add.s64 %rd24, %rd10, %rd23; + shl.b64 %rd25, %rd20, 8; + add.s64 %rd26, %rd10, %rd25; + shl.b64 %rd27, %rd21, 8; + add.s64 %rd28, %rd10, %rd27; + shl.b64 %rd29, %rd22, 8; + add.s64 %rd30, %rd10, %rd29; + .loc 1 75 33 // l2norm.py:75:33 + mul.wide.u32 %rd31, %r50, 2; + add.s64 %rd1, %rd24, %rd31; + add.s64 %rd2, %rd26, %rd31; + add.s64 %rd3, %rd28, %rd31; + add.s64 %rd4, %rd30, %rd31; + mov.b32 %r5, 0; + .loc 1 79 18 // l2norm.py:79:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + mov.b32 {%rs5, %rs6}, %r3; + mov.b32 {%rs7, %rs8}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs9, %rs10}, %r6; + mov.b32 {%rs11, %rs12}, %r7; + mov.b32 {%rs13, %rs14}, %r8; + mov.b32 {%rs15, %rs16}, %r9; + // begin inline asm + mov.u32 %r10, %r5; + mov.u32 %r11, %r5; + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + @%p3 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs17, %rs18}, %r10; + mov.b32 {%rs19, %rs20}, %r11; + mov.b32 {%rs21, %rs22}, %r12; + mov.b32 {%rs23, %rs24}, %r13; + // begin inline asm + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + @%p4 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ]; + // end inline asm + mov.b32 {%rs25, %rs26}, %r14; + mov.b32 {%rs27, %rs28}, %r15; + mov.b32 {%rs29, %rs30}, %r16; + mov.b32 {%rs31, %rs32}, %r17; + .loc 1 79 47 // l2norm.py:79:47 + cvt.f32.bf16 %r51, %rs1; + cvt.f32.bf16 %r52, %rs2; + cvt.f32.bf16 %r53, %rs3; + cvt.f32.bf16 %r54, %rs4; + cvt.f32.bf16 %r55, %rs5; + cvt.f32.bf16 %r56, %rs6; + cvt.f32.bf16 %r57, %rs7; + cvt.f32.bf16 %r58, %rs8; + cvt.f32.bf16 %r59, %rs9; + cvt.f32.bf16 %r60, %rs10; + cvt.f32.bf16 %r61, %rs11; + cvt.f32.bf16 %r62, %rs12; + cvt.f32.bf16 %r63, %rs13; + cvt.f32.bf16 %r64, %rs14; + cvt.f32.bf16 %r65, %rs15; + cvt.f32.bf16 %r66, %rs16; + cvt.f32.bf16 %r67, %rs17; + cvt.f32.bf16 %r68, %rs18; + cvt.f32.bf16 %r69, %rs19; + cvt.f32.bf16 %r70, %rs20; + cvt.f32.bf16 %r71, %rs21; + cvt.f32.bf16 %r72, %rs22; + cvt.f32.bf16 %r73, %rs23; + cvt.f32.bf16 %r74, %rs24; + cvt.f32.bf16 %r75, %rs25; + cvt.f32.bf16 %r76, %rs26; + cvt.f32.bf16 %r77, %rs27; + cvt.f32.bf16 %r78, %rs28; + cvt.f32.bf16 %r79, %rs29; + cvt.f32.bf16 %r80, %rs30; + cvt.f32.bf16 %r81, %rs31; + cvt.f32.bf16 %r82, %rs32; + .loc 1 80 38 // l2norm.py:80:38 + mul.f32 %r83, %r52, %r52; + mul.f32 %r84, %r60, %r60; + mul.f32 %r85, %r68, %r68; + mul.f32 %r86, %r76, %r76; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + fma.rn.f32.bf16 %r87, %rs1, %rs1, %r83; + fma.rn.f32.bf16 %r88, %rs3, %rs3, %r87; + fma.rn.f32.bf16 %r89, %rs4, %rs4, %r88; + fma.rn.f32.bf16 %r90, %rs5, %rs5, %r89; + fma.rn.f32.bf16 %r91, %rs6, %rs6, %r90; + fma.rn.f32.bf16 %r92, %rs7, %rs7, %r91; + fma.rn.f32.bf16 %r93, %rs8, %rs8, %r92; + fma.rn.f32.bf16 %r94, %rs9, %rs9, %r84; + fma.rn.f32.bf16 %r95, %rs11, %rs11, %r94; + fma.rn.f32.bf16 %r96, %rs12, %rs12, %r95; + fma.rn.f32.bf16 %r97, %rs13, %rs13, %r96; + fma.rn.f32.bf16 %r98, %rs14, %rs14, %r97; + fma.rn.f32.bf16 %r99, %rs15, %rs15, %r98; + fma.rn.f32.bf16 %r100, %rs16, %rs16, %r99; + fma.rn.f32.bf16 %r101, %rs17, %rs17, %r85; + fma.rn.f32.bf16 %r102, %rs19, %rs19, %r101; + fma.rn.f32.bf16 %r103, %rs20, %rs20, %r102; + fma.rn.f32.bf16 %r104, %rs21, %rs21, %r103; + fma.rn.f32.bf16 %r105, %rs22, %rs22, %r104; + fma.rn.f32.bf16 %r106, %rs23, %rs23, %r105; + fma.rn.f32.bf16 %r107, %rs24, %rs24, %r106; + fma.rn.f32.bf16 %r108, %rs25, %rs25, %r86; + fma.rn.f32.bf16 %r109, %rs27, %rs27, %r108; + fma.rn.f32.bf16 %r110, %rs28, %rs28, %r109; + fma.rn.f32.bf16 %r111, %rs29, %rs29, %r110; + fma.rn.f32.bf16 %r112, %rs30, %rs30, %r111; + fma.rn.f32.bf16 %r113, %rs31, %rs31, %r112; + fma.rn.f32.bf16 %r114, %rs32, %rs32, %r113; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r115, %r93, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r116, %r93, %r115; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r117, %r116, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r118, %r116, %r117; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r119, %r118, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r120, %r118, %r119; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r121, %r120, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r122, %r120, %r121; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r123, %r100, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r124, %r100, %r123; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r125, %r124, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r126, %r124, %r125; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r127, %r126, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r128, %r126, %r127; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r129, %r128, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r130, %r128, %r129; +$L__tmp18: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r131, %r107, 8, 31, -1; +$L__tmp19: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r132, %r107, %r131; +$L__tmp20: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r133, %r132, 4, 31, -1; +$L__tmp21: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r134, %r132, %r133; +$L__tmp22: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r135, %r134, 2, 31, -1; +$L__tmp23: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r136, %r134, %r135; +$L__tmp24: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r137, %r136, 1, 31, -1; +$L__tmp25: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r138, %r136, %r137; +$L__tmp26: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r139, %r114, 8, 31, -1; +$L__tmp27: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r140, %r114, %r139; +$L__tmp28: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r141, %r140, 4, 31, -1; +$L__tmp29: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r142, %r140, %r141; +$L__tmp30: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r143, %r142, 2, 31, -1; +$L__tmp31: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r144, %r142, %r143; +$L__tmp32: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r145, %r144, 1, 31, -1; +$L__tmp33: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r146, %r144, %r145; +$L__tmp34: + .loc 1 80 48 // l2norm.py:80:48 + add.f32 %r147, %r42, %r122; + add.f32 %r148, %r42, %r130; + add.f32 %r149, %r42, %r138; + add.f32 %r150, %r42, %r146; + .loc 1 80 25 // l2norm.py:80:25 + sqrt.approx.ftz.f32 %r151, %r147; + sqrt.approx.ftz.f32 %r152, %r148; + sqrt.approx.ftz.f32 %r153, %r149; + sqrt.approx.ftz.f32 %r154, %r150; + .loc 1 71 34 // l2norm.py:71:34 + and.b32 %r155, %r43, 63; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd32, %r155; + or.b64 %rd33, %rd13, %rd32; + .loc 1 77 20 // l2norm.py:77:20 + shl.b64 %rd34, %rd33, 2; + add.s64 %rd9, %rd12, %rd34; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd35, %rd11, %rd29; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd8, %rd35, %rd31; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd36, %rd11, %rd27; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd7, %rd36, %rd31; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd37, %rd11, %rd25; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd6, %rd37, %rd31; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd38, %rd11, %rd23; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd5, %rd38, %rd31; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p6, %rd33, %rd14; + mov.b32 %r156, 0f3F800000; + .loc 1 80 17 // l2norm.py:80:17 + div.full.f32 %r35, %r156, %r151; + div.full.f32 %r36, %r156, %r152; + div.full.f32 %r38, %r156, %r153; + div.full.f32 %r39, %r156, %r154; + .loc 1 81 16 // l2norm.py:81:16 + mul.f32 %r157, %r35, %r51; + mul.f32 %r158, %r35, %r52; + mul.f32 %r159, %r35, %r53; + mul.f32 %r160, %r35, %r54; + mul.f32 %r161, %r35, %r55; + mul.f32 %r162, %r35, %r56; + mul.f32 %r163, %r35, %r57; + mul.f32 %r164, %r35, %r58; + mul.f32 %r165, %r36, %r59; + mul.f32 %r166, %r36, %r60; + mul.f32 %r167, %r36, %r61; + mul.f32 %r168, %r36, %r62; + mul.f32 %r169, %r36, %r63; + mul.f32 %r170, %r36, %r64; + mul.f32 %r171, %r36, %r65; + mul.f32 %r172, %r36, %r66; + mul.f32 %r173, %r38, %r67; + mul.f32 %r174, %r38, %r68; + mul.f32 %r175, %r38, %r69; + mul.f32 %r176, %r38, %r70; + mul.f32 %r177, %r38, %r71; + mul.f32 %r178, %r38, %r72; + mul.f32 %r179, %r38, %r73; + mul.f32 %r180, %r38, %r74; + mul.f32 %r181, %r39, %r75; + mul.f32 %r182, %r39, %r76; + mul.f32 %r183, %r39, %r77; + mul.f32 %r184, %r39, %r78; + mul.f32 %r185, %r39, %r79; + mul.f32 %r186, %r39, %r80; + mul.f32 %r187, %r39, %r81; + mul.f32 %r188, %r39, %r82; + .loc 1 83 18 // l2norm.py:83:18 + cvt.rn.bf16x2.f32 %r18, %r158, %r157; + cvt.rn.bf16x2.f32 %r19, %r160, %r159; + cvt.rn.bf16x2.f32 %r20, %r162, %r161; + cvt.rn.bf16x2.f32 %r21, %r164, %r163; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd5 + 0 ], { %r18, %r19, %r20, %r21 }; + // end inline asm + cvt.rn.bf16x2.f32 %r22, %r166, %r165; + cvt.rn.bf16x2.f32 %r23, %r168, %r167; + cvt.rn.bf16x2.f32 %r24, %r170, %r169; + cvt.rn.bf16x2.f32 %r25, %r172, %r171; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd6 + 0 ], { %r22, %r23, %r24, %r25 }; + // end inline asm + cvt.rn.bf16x2.f32 %r26, %r174, %r173; + cvt.rn.bf16x2.f32 %r27, %r176, %r175; + cvt.rn.bf16x2.f32 %r28, %r178, %r177; + cvt.rn.bf16x2.f32 %r29, %r180, %r179; + // begin inline asm + @%p3 st.global.v4.b32 [ %rd7 + 0 ], { %r26, %r27, %r28, %r29 }; + // end inline asm + cvt.rn.bf16x2.f32 %r30, %r182, %r181; + cvt.rn.bf16x2.f32 %r31, %r184, %r183; + cvt.rn.bf16x2.f32 %r32, %r186, %r185; + cvt.rn.bf16x2.f32 %r33, %r188, %r187; + // begin inline asm + @%p4 st.global.v4.b32 [ %rd8 + 0 ], { %r30, %r31, %r32, %r33 }; + // end inline asm + .loc 1 84 21 // l2norm.py:84:21 + shr.u32 %r189, %r44, 1; + mov.b32 %r190, global_smem; + add.s32 %r34, %r190, %r189; + // begin inline asm + st.shared.v2.b32 [ %r34 + 0 ], { %r35, %r36 }; + // end inline asm + add.s32 %r37, %r34, 128; + // begin inline asm + st.shared.v2.b32 [ %r37 + 0 ], { %r38, %r39 }; + // end inline asm + bar.sync 0; + shr.u32 %r191, %r43, 2; + and.b32 %r192, %r191, 4; + shl.b32 %r193, %r43, 2; + and.b32 %r194, %r193, 128; + add.s32 %r195, %r190, %r192; + add.s32 %r196, %r195, %r194; + add.s32 %r197, %r196, %r50; + ld.shared.b32 %r40, [%r197]; + and.b32 %r198, %r43, 192; + setp.eq.b32 %p7, %r198, 0; + and.pred %p5, %p7, %p6; + // begin inline asm + @%p5 st.global.b32 [ %rd9 + 0 ], { %r40 }; + // end inline asm + .loc 1 84 4 // l2norm.py:84:4 + ret; +$L__tmp35: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp34 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 80 // DW_AT_call_line +.b8 43 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp34 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/l2norm_fwd_kernel.source b/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/l2norm_fwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..b5000c470f5e0683aea9e8045b2d0a8c059ea3d1 --- /dev/null +++ b/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/l2norm_fwd_kernel.source @@ -0,0 +1,172 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc38 = loc(unknown) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc45 = loc("x"(#loc)) +#loc46 = loc("y"(#loc)) +#loc47 = loc("rstd"(#loc)) +#loc48 = loc("eps"(#loc)) +#loc49 = loc("T"(#loc)) +#loc81 = loc("input"(#loc36)) +#loc82 = loc("a"(#loc41)) +#loc83 = loc("b"(#loc41)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc50) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc51) + %o_t = arith.constant 64 : i32 loc(#loc52) + %o_t_1 = arith.constant 64 : i64 loc(#loc52) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc52) + %o_t_3 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc53) + %o_t_4 = arith.extsi %o_t_3 : tensor<64xi32> to tensor<64xi64> loc(#loc54) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<64xi64> loc(#loc54) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<64xi64> loc(#loc54) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc55) + %m_t = arith.extsi %T : i32 to i64 loc(#loc56) + %m_t_7 = tt.splat %m_t : i64 -> tensor<64xi64> loc(#loc56) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<64xi64> loc(#loc56) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc57) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc58) + %m_x_10 = arith.constant 128 : i32 loc(#loc59) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc59) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc59) + %m_x_13 = tt.broadcast %m_x : tensor<64x1xi1> -> tensor<64x128xi1> loc(#loc60) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<64x128xi1> loc(#loc60) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<64x128xi1> loc(#loc60) + %p_x = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc61) + %p_x_16 = arith.constant 128 : i32 loc(#loc62) + %p_x_17 = arith.constant 128 : i64 loc(#loc62) + %p_x_18 = arith.constant dense<128> : tensor<64x1xi64> loc(#loc62) + %p_x_19 = arith.muli %p_x, %p_x_18 : tensor<64x1xi64> loc(#loc62) + %p_x_20 = tt.splat %x : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc63) + %p_x_21 = tt.addptr %p_x_20, %p_x_19 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc63) + %p_x_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc64) + %p_x_23 = tt.broadcast %p_x_21 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc65) + %p_x_24 = tt.broadcast %p_x_22 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc65) + %p_x_25 = tt.addptr %p_x_23, %p_x_24 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc65) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc66) + %p_y_26 = arith.constant 128 : i32 loc(#loc67) + %p_y_27 = arith.constant 128 : i64 loc(#loc67) + %p_y_28 = arith.constant dense<128> : tensor<64x1xi64> loc(#loc67) + %p_y_29 = arith.muli %p_y, %p_y_28 : tensor<64x1xi64> loc(#loc67) + %p_y_30 = tt.splat %y : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc68) + %p_y_31 = tt.addptr %p_y_30, %p_y_29 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc68) + %p_y_32 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc69) + %p_y_33 = tt.broadcast %p_y_31 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc70) + %p_y_34 = tt.broadcast %p_y_32 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc70) + %p_y_35 = tt.addptr %p_y_33, %p_y_34 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc70) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<64x!tt.ptr> loc(#loc71) + %p_rstd_36 = tt.addptr %p_rstd, %o_t_6 : tensor<64x!tt.ptr>, tensor<64xi64> loc(#loc71) + %b_x = arith.constant 0.000000e+00 : f32 loc(#loc72) + %b_x_37 = arith.constant dense<0.000000e+00> : tensor<64x128xf32> loc(#loc72) + %b_x_38 = arith.truncf %b_x_37 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc72) + %b_x_39 = tt.load %p_x_25, %m_x_15, %b_x_38 : tensor<64x128x!tt.ptr> loc(#loc72) + %b_x_40 = arith.extf %b_x_39 : tensor<64x128xbf16> to tensor<64x128xf32> loc(#loc73) + %b_rstd = arith.mulf %b_x_40, %b_x_40 : tensor<64x128xf32> loc(#loc74) + %b_rstd_41 = tt.call @triton.language.standard.sum__fp32S64_128S_c1_cFalse_cNone(%b_rstd) : (tensor<64x128xf32>) -> tensor<64xf32> loc(#loc75) + %b_rstd_42 = tt.splat %eps : f32 -> tensor<64xf32> loc(#loc76) + %b_rstd_43 = arith.addf %b_rstd_41, %b_rstd_42 : tensor<64xf32> loc(#loc76) + %b_rstd_44 = math.sqrt %b_rstd_43 : tensor<64xf32> loc(#loc77) + %b_rstd_45 = arith.constant 1 : i32 loc(#loc78) + %b_rstd_46 = arith.constant 1.000000e+00 : f32 loc(#loc78) + %b_rstd_47 = arith.constant dense<1.000000e+00> : tensor<64xf32> loc(#loc78) + %b_rstd_48 = arith.divf %b_rstd_47, %b_rstd_44 : tensor<64xf32> loc(#loc78) + %b_y = tt.expand_dims %b_rstd_48 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc79) + %b_y_49 = tt.broadcast %b_y : tensor<64x1xf32> -> tensor<64x128xf32> loc(#loc80) + %b_y_50 = arith.mulf %b_x_40, %b_y_49 : tensor<64x128xf32> loc(#loc80) + %0 = arith.truncf %b_y_50 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc32) + tt.store %p_y_35, %0, %m_x_15 : tensor<64x128x!tt.ptr> loc(#loc33) + tt.store %p_rstd_36, %b_rstd_48, %m_t_8 : tensor<64x!tt.ptr> loc(#loc34) + tt.return loc(#loc35) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S64_128S_c1_cFalse_cNone(%input: tensor<64x128xf32> loc("input"(#loc36))) -> tensor<64xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc37) + tt.reduce.return %2 : f32 loc(#loc37) + }) : (tensor<64x128xf32>) -> tensor<64xf32> loc(#loc37) + tt.return %0 : tensor<64xf32> loc(#loc39) + ^bb1: // no predecessors + %1 = ub.poison : tensor<64xf32> loc(#loc40) + tt.return %1 : tensor<64xf32> loc(#loc40) + } loc(#loc36) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc41)), %b: f32 loc("b"(#loc41))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc42) + tt.return %0 : f32 loc(#loc43) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc44) + tt.return %1 : f32 loc(#loc44) + } loc(#loc41) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:18) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:29) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:37) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc50 = loc("i_t"(#loc1)) +#loc51 = loc("i_t"(#loc2)) +#loc52 = loc("o_t"(#loc3)) +#loc53 = loc("o_t"(#loc4)) +#loc54 = loc("o_t"(#loc5)) +#loc55 = loc("o_d"(#loc6)) +#loc56 = loc("m_t"(#loc7)) +#loc57 = loc("m_x"(#loc8)) +#loc58 = loc("m_x"(#loc9)) +#loc59 = loc("m_x"(#loc10)) +#loc60 = loc("m_x"(#loc11)) +#loc61 = loc("p_x"(#loc12)) +#loc62 = loc("p_x"(#loc13)) +#loc63 = loc("p_x"(#loc14)) +#loc64 = loc("p_x"(#loc15)) +#loc65 = loc("p_x"(#loc16)) +#loc66 = loc("p_y"(#loc17)) +#loc67 = loc("p_y"(#loc18)) +#loc68 = loc("p_y"(#loc19)) +#loc69 = loc("p_y"(#loc20)) +#loc70 = loc("p_y"(#loc21)) +#loc71 = loc("p_rstd"(#loc22)) +#loc72 = loc("b_x"(#loc23)) +#loc73 = loc("b_x"(#loc24)) +#loc74 = loc("b_rstd"(#loc25)) +#loc75 = loc("b_rstd"(#loc26)) +#loc76 = loc("b_rstd"(#loc27)) +#loc77 = loc("b_rstd"(#loc28)) +#loc78 = loc("b_rstd"(#loc29)) +#loc79 = loc("b_y"(#loc30)) +#loc80 = loc("b_y"(#loc31)) diff --git a/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/l2norm_fwd_kernel.ttgir b/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/l2norm_fwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..33748d09ca551567f81a51bf937094e585232745 --- /dev/null +++ b/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/l2norm_fwd_kernel.ttgir @@ -0,0 +1,135 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [8, 1], order = [1, 0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [8], order = [0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc1 = loc(unknown) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc34 = loc("x"(#loc)) +#loc35 = loc("y"(#loc)) +#loc36 = loc("rstd"(#loc)) +#loc37 = loc("eps"(#loc)) +#loc38 = loc("T"(#loc)) +#loc59 = loc("b_rstd"(#loc23)) +#loc66 = loc(callsite(#loc1 at #loc59)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 8 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<1.000000e+00> : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<64x128xbf16, #blocked> loc(#loc1) + %c64_i64 = arith.constant 64 : i64 loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<64x1xi64, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc39) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc40) + %o_t = arith.muli %i_t_3, %c64_i64 : i64 loc(#loc41) + %o_t_4 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc42) + %o_t_5 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #blocked1> loc(#loc42) + %o_t_6 = arith.extsi %o_t_4 : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_7 = arith.extsi %o_t_5 : tensor<64xi32, #blocked1> to tensor<64xi64, #blocked1> loc(#loc43) + %o_t_8 = tt.splat %o_t : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_9 = tt.splat %o_t : i64 -> tensor<64xi64, #blocked1> loc(#loc43) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<64xi64, #blocked1> loc(#loc43) + %m_t = arith.extsi %T : i32 to i64 loc(#loc44) + %m_t_12 = tt.splat %m_t : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_13 = tt.splat %m_t : i64 -> tensor<64xi64, #blocked1> loc(#loc44) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<64xi64, #blocked1> loc(#loc44) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<64xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64x1xi1, #blocked> loc(#loc45) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc46) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc46) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_1 : tensor<1x128xi32, #blocked> loc(#loc47) + %m_x_19 = tt.broadcast %m_x : tensor<64x1xi1, #blocked> -> tensor<64x128xi1, #blocked> loc(#loc48) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<64x128xi1, #blocked> loc(#loc48) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<64x128xi1, #blocked> loc(#loc48) + %p_x = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64x1xi64, #blocked> loc(#loc49) + %p_x_22 = arith.muli %p_x, %cst_2 : tensor<64x1xi64, #blocked> loc(#loc50) + %p_x_23 = tt.splat %x : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked> loc(#loc51) + %p_x_24 = tt.addptr %p_x_23, %p_x_22 : tensor<64x1x!tt.ptr, #blocked>, tensor<64x1xi64, #blocked> loc(#loc51) + %p_x_25 = tt.broadcast %p_x_24 : tensor<64x1x!tt.ptr, #blocked> -> tensor<64x128x!tt.ptr, #blocked> loc(#loc52) + %p_x_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<64x128xi32, #blocked> loc(#loc52) + %p_x_27 = tt.addptr %p_x_25, %p_x_26 : tensor<64x128x!tt.ptr, #blocked>, tensor<64x128xi32, #blocked> loc(#loc52) + %p_y = tt.splat %y : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked> loc(#loc53) + %p_y_28 = tt.addptr %p_y, %p_x_22 : tensor<64x1x!tt.ptr, #blocked>, tensor<64x1xi64, #blocked> loc(#loc53) + %p_y_29 = tt.broadcast %p_y_28 : tensor<64x1x!tt.ptr, #blocked> -> tensor<64x128x!tt.ptr, #blocked> loc(#loc54) + %p_y_30 = tt.addptr %p_y_29, %p_x_26 : tensor<64x128x!tt.ptr, #blocked>, tensor<64x128xi32, #blocked> loc(#loc54) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<64x!tt.ptr, #blocked1> loc(#loc55) + %p_rstd_31 = tt.addptr %p_rstd, %o_t_11 : tensor<64x!tt.ptr, #blocked1>, tensor<64xi64, #blocked1> loc(#loc55) + %b_x = tt.load %p_x_27, %m_x_21, %cst_0 : tensor<64x128x!tt.ptr, #blocked> loc(#loc56) + %b_x_32 = arith.extf %b_x : tensor<64x128xbf16, #blocked> to tensor<64x128xf32, #blocked> loc(#loc57) + %b_rstd = arith.mulf %b_x_32, %b_x_32 : tensor<64x128xf32, #blocked> loc(#loc58) + %b_rstd_33 = "tt.reduce"(%b_rstd) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_40: f32 loc(callsite(#loc1 at #loc59)), %b_rstd_41: f32 loc(callsite(#loc1 at #loc59))): + %b_rstd_42 = arith.addf %b_rstd_40, %b_rstd_41 : f32 loc(#loc67) + tt.reduce.return %b_rstd_42 : f32 loc(#loc65) + }) : (tensor<64x128xf32, #blocked>) -> tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc65) + %b_rstd_34 = tt.splat %eps : f32 -> tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_36 = math.sqrt %b_rstd_35 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc61) + %b_rstd_37 = arith.divf %cst, %b_rstd_36 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc62) + %b_y = tt.expand_dims %b_rstd_37 {axis = 1 : i32} : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64x1xf32, #blocked> loc(#loc63) + %b_y_38 = tt.broadcast %b_y : tensor<64x1xf32, #blocked> -> tensor<64x128xf32, #blocked> loc(#loc64) + %b_y_39 = arith.mulf %b_x_32, %b_y_38 : tensor<64x128xf32, #blocked> loc(#loc64) + %0 = arith.truncf %b_y_39 : tensor<64x128xf32, #blocked> to tensor<64x128xbf16, #blocked> loc(#loc30) + tt.store %p_y_30, %0, %m_x_21 : tensor<64x128x!tt.ptr, #blocked> loc(#loc31) + %1 = ttg.convert_layout %b_rstd_37 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64xf32, #blocked1> loc(#loc32) + tt.store %p_rstd_31, %1, %m_t_15 : tensor<64x!tt.ptr, #blocked1> loc(#loc32) + tt.return loc(#loc33) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc39 = loc("i_t"(#loc2)) +#loc40 = loc("i_t"(#loc3)) +#loc41 = loc("o_t"(#loc4)) +#loc42 = loc("o_t"(#loc5)) +#loc43 = loc("o_t"(#loc6)) +#loc44 = loc("m_t"(#loc7)) +#loc45 = loc("m_x"(#loc8)) +#loc46 = loc("m_x"(#loc9)) +#loc47 = loc("m_x"(#loc10)) +#loc48 = loc("m_x"(#loc11)) +#loc49 = loc("p_x"(#loc12)) +#loc50 = loc("p_x"(#loc13)) +#loc51 = loc("p_x"(#loc14)) +#loc52 = loc("p_x"(#loc15)) +#loc53 = loc("p_y"(#loc16)) +#loc54 = loc("p_y"(#loc17)) +#loc55 = loc("p_rstd"(#loc18)) +#loc56 = loc("b_x"(#loc19)) +#loc57 = loc("b_x"(#loc20)) +#loc58 = loc("b_rstd"(#loc21)) +#loc60 = loc("b_rstd"(#loc25)) +#loc61 = loc("b_rstd"(#loc26)) +#loc62 = loc("b_rstd"(#loc27)) +#loc63 = loc("b_y"(#loc28)) +#loc64 = loc("b_y"(#loc29)) +#loc65 = loc(callsite(#loc22 at #loc59)) +#loc67 = loc(callsite(#loc24 at #loc65)) diff --git a/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/l2norm_fwd_kernel.ttir b/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/l2norm_fwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..85ebb8b128ceed91f07ab82f5c0fe10a876f6335 --- /dev/null +++ b/triton/gpu0/BPT77YKZ5QGK6PM5H4N2SGBPHB4MRGUPJDGRQBZ33KOMAZTLKZNA/l2norm_fwd_kernel.ttir @@ -0,0 +1,128 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc3 = loc(unknown) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc35 = loc("x"(#loc)) +#loc36 = loc("y"(#loc)) +#loc37 = loc("rstd"(#loc)) +#loc38 = loc("eps"(#loc)) +#loc39 = loc("T"(#loc)) +#loc62 = loc("b_rstd"(#loc25)) +#loc68 = loc(callsite(#loc3 at #loc62)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %b_x = arith.constant dense<0.000000e+00> : tensor<64x128xbf16> loc(#loc40) + %b_rstd = arith.constant dense<1.000000e+00> : tensor<64xf32> loc(#loc41) + %cst = arith.constant dense<128> : tensor<64x1xi64> loc(#loc3) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc42) + %o_t = arith.constant 64 : i64 loc(#loc43) + %i_t = tt.get_program_id x : i32 loc(#loc44) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc45) + %o_t_1 = arith.muli %i_t_0, %o_t : i64 loc(#loc43) + %o_t_2 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc46) + %o_t_3 = arith.extsi %o_t_2 : tensor<64xi32> to tensor<64xi64> loc(#loc47) + %o_t_4 = tt.splat %o_t_1 : i64 -> tensor<64xi64> loc(#loc47) + %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<64xi64> loc(#loc47) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc48) + %m_t = arith.extsi %T : i32 to i64 loc(#loc49) + %m_t_6 = tt.splat %m_t : i64 -> tensor<64xi64> loc(#loc49) + %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<64xi64> loc(#loc49) + %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc50) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc51) + %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc42) + %m_x_11 = tt.broadcast %m_x_8 : tensor<64x1xi1> -> tensor<64x128xi1> loc(#loc52) + %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<64x128xi1> loc(#loc52) + %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<64x128xi1> loc(#loc52) + %p_x = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc53) + %p_x_14 = arith.muli %p_x, %cst : tensor<64x1xi64> loc(#loc54) + %p_x_15 = tt.splat %x : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc55) + %p_x_16 = tt.addptr %p_x_15, %p_x_14 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc55) + %p_x_17 = tt.broadcast %p_x_16 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc56) + %p_x_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc56) + %p_x_19 = tt.addptr %p_x_17, %p_x_18 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc56) + %p_y = tt.splat %y : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc57) + %p_y_20 = tt.addptr %p_y, %p_x_14 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc57) + %p_y_21 = tt.broadcast %p_y_20 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc58) + %p_y_22 = tt.addptr %p_y_21, %p_x_18 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<64x!tt.ptr> loc(#loc59) + %p_rstd_23 = tt.addptr %p_rstd, %o_t_5 : tensor<64x!tt.ptr>, tensor<64xi64> loc(#loc59) + %b_x_24 = tt.load %p_x_19, %m_x_13, %b_x : tensor<64x128x!tt.ptr> loc(#loc40) + %b_x_25 = arith.extf %b_x_24 : tensor<64x128xbf16> to tensor<64x128xf32> loc(#loc60) + %b_rstd_26 = arith.mulf %b_x_25, %b_x_25 : tensor<64x128xf32> loc(#loc61) + %b_rstd_27 = "tt.reduce"(%b_rstd_26) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_34: f32 loc(callsite(#loc3 at #loc62)), %b_rstd_35: f32 loc(callsite(#loc3 at #loc62))): + %b_rstd_36 = arith.addf %b_rstd_34, %b_rstd_35 : f32 loc(#loc69) + tt.reduce.return %b_rstd_36 : f32 loc(#loc67) + }) : (tensor<64x128xf32>) -> tensor<64xf32> loc(#loc67) + %b_rstd_28 = tt.splat %eps : f32 -> tensor<64xf32> loc(#loc63) + %b_rstd_29 = arith.addf %b_rstd_27, %b_rstd_28 : tensor<64xf32> loc(#loc63) + %b_rstd_30 = math.sqrt %b_rstd_29 : tensor<64xf32> loc(#loc64) + %b_rstd_31 = arith.divf %b_rstd, %b_rstd_30 : tensor<64xf32> loc(#loc41) + %b_y = tt.expand_dims %b_rstd_31 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc65) + %b_y_32 = tt.broadcast %b_y : tensor<64x1xf32> -> tensor<64x128xf32> loc(#loc66) + %b_y_33 = arith.mulf %b_x_25, %b_y_32 : tensor<64x128xf32> loc(#loc66) + %0 = arith.truncf %b_y_33 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc31) + tt.store %p_y_22, %0, %m_x_13 : tensor<64x128x!tt.ptr> loc(#loc32) + tt.store %p_rstd_23, %b_rstd_31, %m_t_7 : tensor<64x!tt.ptr> loc(#loc33) + tt.return loc(#loc34) + } loc(#loc) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc40 = loc("b_x"(#loc1)) +#loc41 = loc("b_rstd"(#loc2)) +#loc42 = loc("m_x"(#loc4)) +#loc43 = loc("o_t"(#loc5)) +#loc44 = loc("i_t"(#loc6)) +#loc45 = loc("i_t"(#loc7)) +#loc46 = loc("o_t"(#loc8)) +#loc47 = loc("o_t"(#loc9)) +#loc48 = loc("o_d"(#loc10)) +#loc49 = loc("m_t"(#loc11)) +#loc50 = loc("m_x"(#loc12)) +#loc51 = loc("m_x"(#loc13)) +#loc52 = loc("m_x"(#loc14)) +#loc53 = loc("p_x"(#loc15)) +#loc54 = loc("p_x"(#loc16)) +#loc55 = loc("p_x"(#loc17)) +#loc56 = loc("p_x"(#loc18)) +#loc57 = loc("p_y"(#loc19)) +#loc58 = loc("p_y"(#loc20)) +#loc59 = loc("p_rstd"(#loc21)) +#loc60 = loc("b_x"(#loc22)) +#loc61 = loc("b_rstd"(#loc23)) +#loc63 = loc("b_rstd"(#loc27)) +#loc64 = loc("b_rstd"(#loc28)) +#loc65 = loc("b_y"(#loc29)) +#loc66 = loc("b_y"(#loc30)) +#loc67 = loc(callsite(#loc24 at #loc62)) +#loc69 = loc(callsite(#loc26 at #loc67)) diff --git a/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/__grp__l2norm_fwd_kernel.json b/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/__grp__l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..827e53f44cb2ad92d8c2ddee51487b935792e824 --- /dev/null +++ b/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/__grp__l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_fwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/l2norm_fwd_kernel.source", "l2norm_fwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/l2norm_fwd_kernel.ttir", "l2norm_fwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/l2norm_fwd_kernel.ttgir", "l2norm_fwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/l2norm_fwd_kernel.llir", "l2norm_fwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/l2norm_fwd_kernel.ptx", "l2norm_fwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/l2norm_fwd_kernel.cubin", "l2norm_fwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/l2norm_fwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/l2norm_fwd_kernel.cubin b/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/l2norm_fwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..0b1e3fa6498e0cb9477ed46f6d01f449c2b92fee Binary files /dev/null and b/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/l2norm_fwd_kernel.cubin differ diff --git a/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/l2norm_fwd_kernel.json b/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..8219e484ef8cc220170d5c42b06682ad8e50c1d4 --- /dev/null +++ b/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"hash": "0c88700877aa8a1e5f7d8fd8197e7002f72c8e0c013c3cbc7e94c44df9095b08", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 16, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 256, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_fwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/l2norm_fwd_kernel.llir b/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/l2norm_fwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..8b4267a71f3e80975afef384b840d05aa485e03e --- /dev/null +++ b/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/l2norm_fwd_kernel.llir @@ -0,0 +1,378 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 +@.str.2 = private unnamed_addr constant [17 x i8] c"__CUDA_PREC_SQRT\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_fwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, float %3, i32 %4, ptr addrspace(1) readnone captures(none) %5, ptr addrspace(1) readnone captures(none) %6) local_unnamed_addr #0 !dbg !5 { + %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %9 = zext nneg i32 %8 to i64, !dbg !14 + %10 = shl nuw nsw i64 %9, 6, !dbg !15 + %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !16 + %12 = and i32 %11, 496, !dbg !16 + %13 = lshr exact i32 %12, 4, !dbg !16 + %14 = or disjoint i32 %13, 32, !dbg !16 + %15 = zext nneg i32 %13 to i64, !dbg !17 + %16 = zext nneg i32 %14 to i64, !dbg !17 + %17 = or disjoint i64 %10, %15, !dbg !17 + %18 = or disjoint i64 %10, %16, !dbg !17 + %19 = sext i32 %4 to i64, !dbg !18 + %20 = icmp slt i64 %17, %19, !dbg !18 + %21 = icmp slt i64 %18, %19, !dbg !18 + %22 = shl nuw nsw i32 %11, 3, !dbg !19 + %23 = and i32 %22, 120, !dbg !19 + %24 = shl nuw nsw i64 %17, 7, !dbg !20 + %25 = shl nuw nsw i64 %18, 7, !dbg !20 + %26 = getelementptr bfloat, ptr addrspace(1) %0, i64 %24, !dbg !21 + %27 = getelementptr bfloat, ptr addrspace(1) %0, i64 %25, !dbg !21 + %28 = zext nneg i32 %23 to i64, !dbg !22 + %29 = getelementptr bfloat, ptr addrspace(1) %26, i64 %28, !dbg !22 + %30 = getelementptr bfloat, ptr addrspace(1) %27, i64 %28, !dbg !22 + %31 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %29, i1 %20) #6, !dbg !23 + %32 = extractvalue { i32, i32, i32, i32 } %31, 0, !dbg !23 + %33 = bitcast i32 %32 to <2 x bfloat>, !dbg !23 + %34 = extractvalue { i32, i32, i32, i32 } %31, 1, !dbg !23 + %35 = bitcast i32 %34 to <2 x bfloat>, !dbg !23 + %36 = extractvalue { i32, i32, i32, i32 } %31, 2, !dbg !23 + %37 = bitcast i32 %36 to <2 x bfloat>, !dbg !23 + %38 = extractvalue { i32, i32, i32, i32 } %31, 3, !dbg !23 + %39 = bitcast i32 %38 to <2 x bfloat>, !dbg !23 + %40 = extractelement <2 x bfloat> %33, i64 0, !dbg !23 + %41 = extractelement <2 x bfloat> %33, i64 1, !dbg !23 + %42 = extractelement <2 x bfloat> %35, i64 0, !dbg !23 + %43 = extractelement <2 x bfloat> %35, i64 1, !dbg !23 + %44 = extractelement <2 x bfloat> %37, i64 0, !dbg !23 + %45 = extractelement <2 x bfloat> %37, i64 1, !dbg !23 + %46 = extractelement <2 x bfloat> %39, i64 0, !dbg !23 + %47 = extractelement <2 x bfloat> %39, i64 1, !dbg !23 + %48 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %30, i1 %21) #6, !dbg !23 + %49 = extractvalue { i32, i32, i32, i32 } %48, 0, !dbg !23 + %50 = bitcast i32 %49 to <2 x bfloat>, !dbg !23 + %51 = extractvalue { i32, i32, i32, i32 } %48, 1, !dbg !23 + %52 = bitcast i32 %51 to <2 x bfloat>, !dbg !23 + %53 = extractvalue { i32, i32, i32, i32 } %48, 2, !dbg !23 + %54 = bitcast i32 %53 to <2 x bfloat>, !dbg !23 + %55 = extractvalue { i32, i32, i32, i32 } %48, 3, !dbg !23 + %56 = bitcast i32 %55 to <2 x bfloat>, !dbg !23 + %57 = extractelement <2 x bfloat> %50, i64 0, !dbg !23 + %58 = extractelement <2 x bfloat> %50, i64 1, !dbg !23 + %59 = extractelement <2 x bfloat> %52, i64 0, !dbg !23 + %60 = extractelement <2 x bfloat> %52, i64 1, !dbg !23 + %61 = extractelement <2 x bfloat> %54, i64 0, !dbg !23 + %62 = extractelement <2 x bfloat> %54, i64 1, !dbg !23 + %63 = extractelement <2 x bfloat> %56, i64 0, !dbg !23 + %64 = extractelement <2 x bfloat> %56, i64 1, !dbg !23 + %65 = fpext bfloat %40 to float, !dbg !24 + %66 = fpext bfloat %41 to float, !dbg !24 + %67 = fpext bfloat %42 to float, !dbg !24 + %68 = fpext bfloat %43 to float, !dbg !24 + %69 = fpext bfloat %44 to float, !dbg !24 + %70 = fpext bfloat %45 to float, !dbg !24 + %71 = fpext bfloat %46 to float, !dbg !24 + %72 = fpext bfloat %47 to float, !dbg !24 + %73 = fpext bfloat %57 to float, !dbg !24 + %74 = fpext bfloat %58 to float, !dbg !24 + %75 = fpext bfloat %59 to float, !dbg !24 + %76 = fpext bfloat %60 to float, !dbg !24 + %77 = fpext bfloat %61 to float, !dbg !24 + %78 = fpext bfloat %62 to float, !dbg !24 + %79 = fpext bfloat %63 to float, !dbg !24 + %80 = fpext bfloat %64 to float, !dbg !24 + %81 = fmul float %65, %65, !dbg !25 + %82 = fmul float %66, %66, !dbg !25 + %83 = fmul float %67, %67, !dbg !25 + %84 = fmul float %68, %68, !dbg !25 + %85 = fmul float %69, %69, !dbg !25 + %86 = fmul float %70, %70, !dbg !25 + %87 = fmul float %71, %71, !dbg !25 + %88 = fmul float %72, %72, !dbg !25 + %89 = fmul float %73, %73, !dbg !25 + %90 = fmul float %74, %74, !dbg !25 + %91 = fmul float %75, %75, !dbg !25 + %92 = fmul float %76, %76, !dbg !25 + %93 = fmul float %77, %77, !dbg !25 + %94 = fmul float %78, %78, !dbg !25 + %95 = fmul float %79, %79, !dbg !25 + %96 = fmul float %80, %80, !dbg !25 + %97 = fadd float %81, %82, !dbg !26 + %98 = fadd float %83, %97, !dbg !26 + %99 = fadd float %84, %98, !dbg !26 + %100 = fadd float %85, %99, !dbg !26 + %101 = fadd float %86, %100, !dbg !26 + %102 = fadd float %87, %101, !dbg !26 + %103 = fadd float %88, %102, !dbg !26 + %104 = fadd float %89, %90, !dbg !26 + %105 = fadd float %91, %104, !dbg !26 + %106 = fadd float %92, %105, !dbg !26 + %107 = fadd float %93, %106, !dbg !26 + %108 = fadd float %94, %107, !dbg !26 + %109 = fadd float %95, %108, !dbg !26 + %110 = fadd float %96, %109, !dbg !26 + %111 = bitcast float %103 to i32, !dbg !29 + %112 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %111, i32 8, i32 31), !dbg !29 + %113 = bitcast i32 %112 to float, !dbg !29 + %114 = fadd float %103, %113, !dbg !26 + %115 = bitcast float %114 to i32, !dbg !29 + %116 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %115, i32 4, i32 31), !dbg !29 + %117 = bitcast i32 %116 to float, !dbg !29 + %118 = fadd float %114, %117, !dbg !26 + %119 = bitcast float %118 to i32, !dbg !29 + %120 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %119, i32 2, i32 31), !dbg !29 + %121 = bitcast i32 %120 to float, !dbg !29 + %122 = fadd float %118, %121, !dbg !26 + %123 = bitcast float %122 to i32, !dbg !29 + %124 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %123, i32 1, i32 31), !dbg !29 + %125 = bitcast i32 %124 to float, !dbg !29 + %126 = fadd float %122, %125, !dbg !26 + %127 = bitcast float %110 to i32, !dbg !29 + %128 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %127, i32 8, i32 31), !dbg !29 + %129 = bitcast i32 %128 to float, !dbg !29 + %130 = fadd float %110, %129, !dbg !26 + %131 = bitcast float %130 to i32, !dbg !29 + %132 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %131, i32 4, i32 31), !dbg !29 + %133 = bitcast i32 %132 to float, !dbg !29 + %134 = fadd float %130, %133, !dbg !26 + %135 = bitcast float %134 to i32, !dbg !29 + %136 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %135, i32 2, i32 31), !dbg !29 + %137 = bitcast i32 %136 to float, !dbg !29 + %138 = fadd float %134, %137, !dbg !26 + %139 = bitcast float %138 to i32, !dbg !29 + %140 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %139, i32 1, i32 31), !dbg !29 + %141 = bitcast i32 %140 to float, !dbg !29 + %142 = fadd float %138, %141, !dbg !26 + %143 = fadd float %3, %126, !dbg !32 + %144 = fadd float %3, %142, !dbg !32 + %145 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i = icmp eq i32 %145, 0, !dbg !33 + %146 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i = icmp eq i32 %146, 0, !dbg !33 + br i1 %.not.i, label %152, label %147, !dbg !33 + +147: ; preds = %7 + br i1 %.not1.i, label %150, label %148, !dbg !33 + +148: ; preds = %147 + %149 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %143) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +150: ; preds = %147 + %151 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %143) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +152: ; preds = %7 + br i1 %.not1.i, label %155, label %153, !dbg !33 + +153: ; preds = %152 + %154 = tail call float @llvm.nvvm.sqrt.rn.f(float %143) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +155: ; preds = %152 + %156 = tail call float @llvm.nvvm.sqrt.approx.f(float %143) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +__nv_sqrtf.exit: ; preds = %148, %150, %153, %155 + %.0.i = phi float [ %149, %148 ], [ %151, %150 ], [ %154, %153 ], [ %156, %155 ], !dbg !33 + %157 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i1 = icmp eq i32 %157, 0, !dbg !33 + %158 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i4 = icmp eq i32 %158, 0, !dbg !33 + br i1 %.not.i1, label %164, label %159, !dbg !33 + +159: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %162, label %160, !dbg !33 + +160: ; preds = %159 + %161 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %144) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +162: ; preds = %159 + %163 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %144) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +164: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %167, label %165, !dbg !33 + +165: ; preds = %164 + %166 = tail call float @llvm.nvvm.sqrt.rn.f(float %144) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +167: ; preds = %164 + %168 = tail call float @llvm.nvvm.sqrt.approx.f(float %144) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +__nv_sqrtf.exit5: ; preds = %160, %162, %165, %167 + %.0.i3 = phi float [ %161, %160 ], [ %163, %162 ], [ %166, %165 ], [ %168, %167 ], !dbg !33 + %169 = and i32 %11, 63, !dbg !16 + %170 = zext nneg i32 %169 to i64, !dbg !17 + %171 = or disjoint i64 %10, %170, !dbg !17 + %172 = getelementptr float, ptr addrspace(1) %2, i64 %171, !dbg !34 + %173 = getelementptr bfloat, ptr addrspace(1) %1, i64 %25, !dbg !35 + %174 = getelementptr bfloat, ptr addrspace(1) %173, i64 %28, !dbg !36 + %175 = getelementptr bfloat, ptr addrspace(1) %1, i64 %24, !dbg !35 + %176 = getelementptr bfloat, ptr addrspace(1) %175, i64 %28, !dbg !36 + %177 = icmp slt i64 %171, %19, !dbg !18 + %178 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i), !dbg !37 + %179 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i3), !dbg !37 + %180 = fmul float %178, %65, !dbg !38 + %181 = fmul float %178, %66, !dbg !38 + %182 = fmul float %178, %67, !dbg !38 + %183 = fmul float %178, %68, !dbg !38 + %184 = fmul float %178, %69, !dbg !38 + %185 = fmul float %178, %70, !dbg !38 + %186 = fmul float %178, %71, !dbg !38 + %187 = fmul float %178, %72, !dbg !38 + %188 = fmul float %179, %73, !dbg !38 + %189 = fmul float %179, %74, !dbg !38 + %190 = fmul float %179, %75, !dbg !38 + %191 = fmul float %179, %76, !dbg !38 + %192 = fmul float %179, %77, !dbg !38 + %193 = fmul float %179, %78, !dbg !38 + %194 = fmul float %179, %79, !dbg !38 + %195 = fmul float %179, %80, !dbg !38 + %196 = fptrunc float %180 to bfloat, !dbg !39 + %197 = fptrunc float %181 to bfloat, !dbg !39 + %198 = fptrunc float %182 to bfloat, !dbg !39 + %199 = fptrunc float %183 to bfloat, !dbg !39 + %200 = fptrunc float %184 to bfloat, !dbg !39 + %201 = fptrunc float %185 to bfloat, !dbg !39 + %202 = fptrunc float %186 to bfloat, !dbg !39 + %203 = fptrunc float %187 to bfloat, !dbg !39 + %204 = fptrunc float %188 to bfloat, !dbg !39 + %205 = fptrunc float %189 to bfloat, !dbg !39 + %206 = fptrunc float %190 to bfloat, !dbg !39 + %207 = fptrunc float %191 to bfloat, !dbg !39 + %208 = fptrunc float %192 to bfloat, !dbg !39 + %209 = fptrunc float %193 to bfloat, !dbg !39 + %210 = fptrunc float %194 to bfloat, !dbg !39 + %211 = fptrunc float %195 to bfloat, !dbg !39 + %212 = insertelement <2 x bfloat> poison, bfloat %196, i64 0, !dbg !40 + %213 = insertelement <2 x bfloat> %212, bfloat %197, i64 1, !dbg !40 + %214 = bitcast <2 x bfloat> %213 to i32, !dbg !40 + %215 = insertelement <2 x bfloat> poison, bfloat %198, i64 0, !dbg !40 + %216 = insertelement <2 x bfloat> %215, bfloat %199, i64 1, !dbg !40 + %217 = bitcast <2 x bfloat> %216 to i32, !dbg !40 + %218 = insertelement <2 x bfloat> poison, bfloat %200, i64 0, !dbg !40 + %219 = insertelement <2 x bfloat> %218, bfloat %201, i64 1, !dbg !40 + %220 = bitcast <2 x bfloat> %219 to i32, !dbg !40 + %221 = insertelement <2 x bfloat> poison, bfloat %202, i64 0, !dbg !40 + %222 = insertelement <2 x bfloat> %221, bfloat %203, i64 1, !dbg !40 + %223 = bitcast <2 x bfloat> %222 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %214, i32 %217, i32 %220, i32 %223, ptr addrspace(1) %176, i1 %20) #6, !dbg !40 + %224 = insertelement <2 x bfloat> poison, bfloat %204, i64 0, !dbg !40 + %225 = insertelement <2 x bfloat> %224, bfloat %205, i64 1, !dbg !40 + %226 = bitcast <2 x bfloat> %225 to i32, !dbg !40 + %227 = insertelement <2 x bfloat> poison, bfloat %206, i64 0, !dbg !40 + %228 = insertelement <2 x bfloat> %227, bfloat %207, i64 1, !dbg !40 + %229 = bitcast <2 x bfloat> %228 to i32, !dbg !40 + %230 = insertelement <2 x bfloat> poison, bfloat %208, i64 0, !dbg !40 + %231 = insertelement <2 x bfloat> %230, bfloat %209, i64 1, !dbg !40 + %232 = bitcast <2 x bfloat> %231 to i32, !dbg !40 + %233 = insertelement <2 x bfloat> poison, bfloat %210, i64 0, !dbg !40 + %234 = insertelement <2 x bfloat> %233, bfloat %211, i64 1, !dbg !40 + %235 = bitcast <2 x bfloat> %234 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %226, i32 %229, i32 %232, i32 %235, ptr addrspace(1) %174, i1 %21) #6, !dbg !40 + %236 = lshr exact i32 %12, 2, !dbg !41 + %237 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %236, !dbg !41 + %238 = bitcast float %178 to <1 x i32>, !dbg !41 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) %237, <1 x i32> %238) #6, !dbg !41 + %239 = getelementptr inbounds nuw i8, ptr addrspace(3) %237, i32 128, !dbg !41 + %240 = bitcast float %179 to <1 x i32>, !dbg !41 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %239, <1 x i32> %240) #6, !dbg !41 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !41 + %241 = shl nuw nsw i32 %169, 2, !dbg !41 + %242 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %241, !dbg !41 + %243 = load i32, ptr addrspace(3) %242, align 4, !dbg !41 + %244 = and i32 %11, 448, !dbg !41 + %245 = icmp eq i32 %244, 0, !dbg !41 + %246 = and i1 %245, %177, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %243, ptr addrspace(1) %172, i1 %246) #6, !dbg !41 + ret void, !dbg !42 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.div.full(float, float) #3 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #4 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #5 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.f(float) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="512" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) } +attributes #4 = { convergent nocallback nounwind } +attributes #5 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #6 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "l2norm_fwd_kernel", linkageName: "l2norm_fwd_kernel", scope: !1, file: !1, line: 69, type: !6, scopeLine: 69, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !10, !11, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 70, column: 24, scope: !5) +!14 = !DILocation(line: 70, column: 30, scope: !5) +!15 = !DILocation(line: 71, column: 16, scope: !5) +!16 = !DILocation(line: 71, column: 34, scope: !5) +!17 = !DILocation(line: 71, column: 21, scope: !5) +!18 = !DILocation(line: 73, column: 16, scope: !5) +!19 = !DILocation(line: 74, column: 30, scope: !5) +!20 = !DILocation(line: 75, column: 29, scope: !5) +!21 = !DILocation(line: 75, column: 14, scope: !5) +!22 = !DILocation(line: 75, column: 33, scope: !5) +!23 = !DILocation(line: 79, column: 18, scope: !5) +!24 = !DILocation(line: 79, column: 47, scope: !5) +!25 = !DILocation(line: 80, column: 38, scope: !5) +!26 = !DILocation(line: 263, column: 15, scope: !27, inlinedAt: !29) +!27 = distinct !DILexicalBlockFile(scope: !5, file: !28, discriminator: 0) +!28 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!29 = !DILocation(line: 293, column: 36, scope: !27, inlinedAt: !30) +!30 = !DILocation(line: 80, column: 43, scope: !31) +!31 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!32 = !DILocation(line: 80, column: 48, scope: !5) +!33 = !DILocation(line: 80, column: 25, scope: !5) +!34 = !DILocation(line: 77, column: 20, scope: !5) +!35 = !DILocation(line: 76, column: 14, scope: !5) +!36 = !DILocation(line: 76, column: 33, scope: !5) +!37 = !DILocation(line: 80, column: 17, scope: !5) +!38 = !DILocation(line: 81, column: 16, scope: !5) +!39 = !DILocation(line: 83, column: 25, scope: !5) +!40 = !DILocation(line: 83, column: 18, scope: !5) +!41 = !DILocation(line: 84, column: 21, scope: !5) +!42 = !DILocation(line: 84, column: 4, scope: !5) diff --git a/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/l2norm_fwd_kernel.ptx b/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/l2norm_fwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..c66b4263d0ae3d8f81b7b3f7774a13790b5f3136 --- /dev/null +++ b/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/l2norm_fwd_kernel.ptx @@ -0,0 +1,483 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_fwd_kernel // -- Begin function l2norm_fwd_kernel +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; +.global .align 1 .b8 _$_str_$_2[17] = {95, 95, 67, 85, 68, 65, 95, 80, 82, 69, 67, 95, 83, 81, 82, 84}; + // @l2norm_fwd_kernel +.visible .entry l2norm_fwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_2, + .param .f32 l2norm_fwd_kernel_param_3, + .param .u32 l2norm_fwd_kernel_param_4, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_6 +) +.reqntid 512 +{ + .reg .pred %p<6>; + .reg .b16 %rs<17>; + .reg .b32 %r<106>; + .reg .b64 %rd<25>; + .loc 1 69 0 // l2norm.py:69:0 +$L__func_begin0: + .loc 1 69 0 // l2norm.py:69:0 + +// %bb.0: // %__nv_sqrtf.exit + ld.param.b64 %rd6, [l2norm_fwd_kernel_param_0]; + ld.param.b64 %rd7, [l2norm_fwd_kernel_param_1]; +$L__tmp0: + .loc 1 70 24 // l2norm.py:70:24 + mov.u32 %r23, %ctaid.x; + ld.param.b64 %rd8, [l2norm_fwd_kernel_param_2]; + .loc 1 71 16 // l2norm.py:71:16 + mul.wide.u32 %rd9, %r23, 64; + ld.param.b32 %r24, [l2norm_fwd_kernel_param_3]; + ld.param.s32 %rd10, [l2norm_fwd_kernel_param_4]; + .loc 1 71 34 // l2norm.py:71:34 + mov.u32 %r25, %tid.x; + and.b32 %r26, %r25, 496; + bfe.u32 %r27, %r25, 4, 5; + or.b32 %r28, %r27, 32; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd11, %r27; + cvt.u64.u32 %rd12, %r28; + or.b64 %rd13, %rd9, %rd11; + or.b64 %rd14, %rd9, %rd12; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p1, %rd13, %rd10; + setp.lt.s64 %p2, %rd14, %rd10; + .loc 1 74 30 // l2norm.py:74:30 + shl.b32 %r29, %r25, 3; + and.b32 %r30, %r29, 120; + .loc 1 75 14 // l2norm.py:75:14 + shl.b64 %rd15, %rd13, 8; + add.s64 %rd16, %rd6, %rd15; + shl.b64 %rd17, %rd14, 8; + add.s64 %rd18, %rd6, %rd17; + .loc 1 75 33 // l2norm.py:75:33 + mul.wide.u32 %rd19, %r30, 2; + add.s64 %rd1, %rd16, %rd19; + add.s64 %rd2, %rd18, %rd19; + mov.b32 %r5, 0; + .loc 1 79 18 // l2norm.py:79:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + mov.b32 {%rs5, %rs6}, %r3; + mov.b32 {%rs7, %rs8}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs9, %rs10}, %r6; + mov.b32 {%rs11, %rs12}, %r7; + mov.b32 {%rs13, %rs14}, %r8; + mov.b32 {%rs15, %rs16}, %r9; + .loc 1 79 47 // l2norm.py:79:47 + cvt.f32.bf16 %r31, %rs1; + cvt.f32.bf16 %r32, %rs2; + cvt.f32.bf16 %r33, %rs3; + cvt.f32.bf16 %r34, %rs4; + cvt.f32.bf16 %r35, %rs5; + cvt.f32.bf16 %r36, %rs6; + cvt.f32.bf16 %r37, %rs7; + cvt.f32.bf16 %r38, %rs8; + cvt.f32.bf16 %r39, %rs9; + cvt.f32.bf16 %r40, %rs10; + cvt.f32.bf16 %r41, %rs11; + cvt.f32.bf16 %r42, %rs12; + cvt.f32.bf16 %r43, %rs13; + cvt.f32.bf16 %r44, %rs14; + cvt.f32.bf16 %r45, %rs15; + cvt.f32.bf16 %r46, %rs16; + .loc 1 80 38 // l2norm.py:80:38 + mul.f32 %r47, %r32, %r32; + mul.f32 %r48, %r40, %r40; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + fma.rn.f32.bf16 %r49, %rs1, %rs1, %r47; + fma.rn.f32.bf16 %r50, %rs3, %rs3, %r49; + fma.rn.f32.bf16 %r51, %rs4, %rs4, %r50; + fma.rn.f32.bf16 %r52, %rs5, %rs5, %r51; + fma.rn.f32.bf16 %r53, %rs6, %rs6, %r52; + fma.rn.f32.bf16 %r54, %rs7, %rs7, %r53; + fma.rn.f32.bf16 %r55, %rs8, %rs8, %r54; + fma.rn.f32.bf16 %r56, %rs9, %rs9, %r48; + fma.rn.f32.bf16 %r57, %rs11, %rs11, %r56; + fma.rn.f32.bf16 %r58, %rs12, %rs12, %r57; + fma.rn.f32.bf16 %r59, %rs13, %rs13, %r58; + fma.rn.f32.bf16 %r60, %rs14, %rs14, %r59; + fma.rn.f32.bf16 %r61, %rs15, %rs15, %r60; + fma.rn.f32.bf16 %r62, %rs16, %rs16, %r61; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r63, %r55, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r64, %r55, %r63; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r65, %r64, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r66, %r64, %r65; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r67, %r66, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r68, %r66, %r67; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r69, %r68, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r70, %r68, %r69; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r71, %r62, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r72, %r62, %r71; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r73, %r72, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r74, %r72, %r73; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r75, %r74, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r76, %r74, %r75; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r77, %r76, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r78, %r76, %r77; +$L__tmp18: + .loc 1 80 48 // l2norm.py:80:48 + add.f32 %r79, %r24, %r70; + add.f32 %r80, %r24, %r78; + .loc 1 80 25 // l2norm.py:80:25 + sqrt.approx.ftz.f32 %r81, %r79; + sqrt.approx.ftz.f32 %r82, %r80; + .loc 1 71 34 // l2norm.py:71:34 + and.b32 %r83, %r25, 63; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd20, %r83; + or.b64 %rd21, %rd9, %rd20; + .loc 1 77 20 // l2norm.py:77:20 + shl.b64 %rd22, %rd21, 2; + add.s64 %rd5, %rd8, %rd22; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd23, %rd7, %rd17; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd4, %rd23, %rd19; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd24, %rd7, %rd15; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd3, %rd24, %rd19; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p4, %rd21, %rd10; + mov.b32 %r84, 0f3F800000; + .loc 1 80 17 // l2norm.py:80:17 + div.full.f32 %r19, %r84, %r81; + div.full.f32 %r21, %r84, %r82; + .loc 1 81 16 // l2norm.py:81:16 + mul.f32 %r85, %r19, %r31; + mul.f32 %r86, %r19, %r32; + mul.f32 %r87, %r19, %r33; + mul.f32 %r88, %r19, %r34; + mul.f32 %r89, %r19, %r35; + mul.f32 %r90, %r19, %r36; + mul.f32 %r91, %r19, %r37; + mul.f32 %r92, %r19, %r38; + mul.f32 %r93, %r21, %r39; + mul.f32 %r94, %r21, %r40; + mul.f32 %r95, %r21, %r41; + mul.f32 %r96, %r21, %r42; + mul.f32 %r97, %r21, %r43; + mul.f32 %r98, %r21, %r44; + mul.f32 %r99, %r21, %r45; + mul.f32 %r100, %r21, %r46; + .loc 1 83 18 // l2norm.py:83:18 + cvt.rn.bf16x2.f32 %r10, %r86, %r85; + cvt.rn.bf16x2.f32 %r11, %r88, %r87; + cvt.rn.bf16x2.f32 %r12, %r90, %r89; + cvt.rn.bf16x2.f32 %r13, %r92, %r91; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd3 + 0 ], { %r10, %r11, %r12, %r13 }; + // end inline asm + cvt.rn.bf16x2.f32 %r14, %r94, %r93; + cvt.rn.bf16x2.f32 %r15, %r96, %r95; + cvt.rn.bf16x2.f32 %r16, %r98, %r97; + cvt.rn.bf16x2.f32 %r17, %r100, %r99; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd4 + 0 ], { %r14, %r15, %r16, %r17 }; + // end inline asm + .loc 1 84 21 // l2norm.py:84:21 + shr.u32 %r101, %r26, 2; + mov.b32 %r102, global_smem; + add.s32 %r18, %r102, %r101; + // begin inline asm + st.shared.b32 [ %r18 + 0 ], %r19; + // end inline asm + add.s32 %r20, %r18, 128; + // begin inline asm + st.shared.b32 [ %r20 + 0 ], %r21; + // end inline asm + bar.sync 0; + shl.b32 %r103, %r83, 2; + add.s32 %r104, %r102, %r103; + ld.shared.b32 %r22, [%r104]; + and.b32 %r105, %r25, 448; + setp.eq.b32 %p5, %r105, 0; + and.pred %p3, %p5, %p4; + // begin inline asm + @%p3 st.global.b32 [ %rd5 + 0 ], { %r22 }; + // end inline asm + .loc 1 84 4 // l2norm.py:84:4 + ret; +$L__tmp19: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp18 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 80 // DW_AT_call_line +.b8 43 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp18 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/l2norm_fwd_kernel.source b/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/l2norm_fwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..b5000c470f5e0683aea9e8045b2d0a8c059ea3d1 --- /dev/null +++ b/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/l2norm_fwd_kernel.source @@ -0,0 +1,172 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc38 = loc(unknown) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc45 = loc("x"(#loc)) +#loc46 = loc("y"(#loc)) +#loc47 = loc("rstd"(#loc)) +#loc48 = loc("eps"(#loc)) +#loc49 = loc("T"(#loc)) +#loc81 = loc("input"(#loc36)) +#loc82 = loc("a"(#loc41)) +#loc83 = loc("b"(#loc41)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc50) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc51) + %o_t = arith.constant 64 : i32 loc(#loc52) + %o_t_1 = arith.constant 64 : i64 loc(#loc52) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc52) + %o_t_3 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc53) + %o_t_4 = arith.extsi %o_t_3 : tensor<64xi32> to tensor<64xi64> loc(#loc54) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<64xi64> loc(#loc54) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<64xi64> loc(#loc54) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc55) + %m_t = arith.extsi %T : i32 to i64 loc(#loc56) + %m_t_7 = tt.splat %m_t : i64 -> tensor<64xi64> loc(#loc56) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<64xi64> loc(#loc56) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc57) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc58) + %m_x_10 = arith.constant 128 : i32 loc(#loc59) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc59) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc59) + %m_x_13 = tt.broadcast %m_x : tensor<64x1xi1> -> tensor<64x128xi1> loc(#loc60) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<64x128xi1> loc(#loc60) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<64x128xi1> loc(#loc60) + %p_x = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc61) + %p_x_16 = arith.constant 128 : i32 loc(#loc62) + %p_x_17 = arith.constant 128 : i64 loc(#loc62) + %p_x_18 = arith.constant dense<128> : tensor<64x1xi64> loc(#loc62) + %p_x_19 = arith.muli %p_x, %p_x_18 : tensor<64x1xi64> loc(#loc62) + %p_x_20 = tt.splat %x : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc63) + %p_x_21 = tt.addptr %p_x_20, %p_x_19 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc63) + %p_x_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc64) + %p_x_23 = tt.broadcast %p_x_21 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc65) + %p_x_24 = tt.broadcast %p_x_22 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc65) + %p_x_25 = tt.addptr %p_x_23, %p_x_24 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc65) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc66) + %p_y_26 = arith.constant 128 : i32 loc(#loc67) + %p_y_27 = arith.constant 128 : i64 loc(#loc67) + %p_y_28 = arith.constant dense<128> : tensor<64x1xi64> loc(#loc67) + %p_y_29 = arith.muli %p_y, %p_y_28 : tensor<64x1xi64> loc(#loc67) + %p_y_30 = tt.splat %y : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc68) + %p_y_31 = tt.addptr %p_y_30, %p_y_29 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc68) + %p_y_32 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc69) + %p_y_33 = tt.broadcast %p_y_31 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc70) + %p_y_34 = tt.broadcast %p_y_32 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc70) + %p_y_35 = tt.addptr %p_y_33, %p_y_34 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc70) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<64x!tt.ptr> loc(#loc71) + %p_rstd_36 = tt.addptr %p_rstd, %o_t_6 : tensor<64x!tt.ptr>, tensor<64xi64> loc(#loc71) + %b_x = arith.constant 0.000000e+00 : f32 loc(#loc72) + %b_x_37 = arith.constant dense<0.000000e+00> : tensor<64x128xf32> loc(#loc72) + %b_x_38 = arith.truncf %b_x_37 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc72) + %b_x_39 = tt.load %p_x_25, %m_x_15, %b_x_38 : tensor<64x128x!tt.ptr> loc(#loc72) + %b_x_40 = arith.extf %b_x_39 : tensor<64x128xbf16> to tensor<64x128xf32> loc(#loc73) + %b_rstd = arith.mulf %b_x_40, %b_x_40 : tensor<64x128xf32> loc(#loc74) + %b_rstd_41 = tt.call @triton.language.standard.sum__fp32S64_128S_c1_cFalse_cNone(%b_rstd) : (tensor<64x128xf32>) -> tensor<64xf32> loc(#loc75) + %b_rstd_42 = tt.splat %eps : f32 -> tensor<64xf32> loc(#loc76) + %b_rstd_43 = arith.addf %b_rstd_41, %b_rstd_42 : tensor<64xf32> loc(#loc76) + %b_rstd_44 = math.sqrt %b_rstd_43 : tensor<64xf32> loc(#loc77) + %b_rstd_45 = arith.constant 1 : i32 loc(#loc78) + %b_rstd_46 = arith.constant 1.000000e+00 : f32 loc(#loc78) + %b_rstd_47 = arith.constant dense<1.000000e+00> : tensor<64xf32> loc(#loc78) + %b_rstd_48 = arith.divf %b_rstd_47, %b_rstd_44 : tensor<64xf32> loc(#loc78) + %b_y = tt.expand_dims %b_rstd_48 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc79) + %b_y_49 = tt.broadcast %b_y : tensor<64x1xf32> -> tensor<64x128xf32> loc(#loc80) + %b_y_50 = arith.mulf %b_x_40, %b_y_49 : tensor<64x128xf32> loc(#loc80) + %0 = arith.truncf %b_y_50 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc32) + tt.store %p_y_35, %0, %m_x_15 : tensor<64x128x!tt.ptr> loc(#loc33) + tt.store %p_rstd_36, %b_rstd_48, %m_t_8 : tensor<64x!tt.ptr> loc(#loc34) + tt.return loc(#loc35) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S64_128S_c1_cFalse_cNone(%input: tensor<64x128xf32> loc("input"(#loc36))) -> tensor<64xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc37) + tt.reduce.return %2 : f32 loc(#loc37) + }) : (tensor<64x128xf32>) -> tensor<64xf32> loc(#loc37) + tt.return %0 : tensor<64xf32> loc(#loc39) + ^bb1: // no predecessors + %1 = ub.poison : tensor<64xf32> loc(#loc40) + tt.return %1 : tensor<64xf32> loc(#loc40) + } loc(#loc36) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc41)), %b: f32 loc("b"(#loc41))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc42) + tt.return %0 : f32 loc(#loc43) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc44) + tt.return %1 : f32 loc(#loc44) + } loc(#loc41) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:18) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:29) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:37) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc50 = loc("i_t"(#loc1)) +#loc51 = loc("i_t"(#loc2)) +#loc52 = loc("o_t"(#loc3)) +#loc53 = loc("o_t"(#loc4)) +#loc54 = loc("o_t"(#loc5)) +#loc55 = loc("o_d"(#loc6)) +#loc56 = loc("m_t"(#loc7)) +#loc57 = loc("m_x"(#loc8)) +#loc58 = loc("m_x"(#loc9)) +#loc59 = loc("m_x"(#loc10)) +#loc60 = loc("m_x"(#loc11)) +#loc61 = loc("p_x"(#loc12)) +#loc62 = loc("p_x"(#loc13)) +#loc63 = loc("p_x"(#loc14)) +#loc64 = loc("p_x"(#loc15)) +#loc65 = loc("p_x"(#loc16)) +#loc66 = loc("p_y"(#loc17)) +#loc67 = loc("p_y"(#loc18)) +#loc68 = loc("p_y"(#loc19)) +#loc69 = loc("p_y"(#loc20)) +#loc70 = loc("p_y"(#loc21)) +#loc71 = loc("p_rstd"(#loc22)) +#loc72 = loc("b_x"(#loc23)) +#loc73 = loc("b_x"(#loc24)) +#loc74 = loc("b_rstd"(#loc25)) +#loc75 = loc("b_rstd"(#loc26)) +#loc76 = loc("b_rstd"(#loc27)) +#loc77 = loc("b_rstd"(#loc28)) +#loc78 = loc("b_rstd"(#loc29)) +#loc79 = loc("b_y"(#loc30)) +#loc80 = loc("b_y"(#loc31)) diff --git a/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/l2norm_fwd_kernel.ttgir b/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/l2norm_fwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..fae852765a0ea146175350c2c53e78f2102c41cf --- /dev/null +++ b/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/l2norm_fwd_kernel.ttgir @@ -0,0 +1,135 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [16, 1], order = [1, 0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [16], order = [0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc1 = loc(unknown) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc34 = loc("x"(#loc)) +#loc35 = loc("y"(#loc)) +#loc36 = loc("rstd"(#loc)) +#loc37 = loc("eps"(#loc)) +#loc38 = loc("T"(#loc)) +#loc59 = loc("b_rstd"(#loc23)) +#loc66 = loc(callsite(#loc1 at #loc59)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 16 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<1.000000e+00> : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<64x128xbf16, #blocked> loc(#loc1) + %c64_i64 = arith.constant 64 : i64 loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<64x1xi64, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc39) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc40) + %o_t = arith.muli %i_t_3, %c64_i64 : i64 loc(#loc41) + %o_t_4 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc42) + %o_t_5 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #blocked1> loc(#loc42) + %o_t_6 = arith.extsi %o_t_4 : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_7 = arith.extsi %o_t_5 : tensor<64xi32, #blocked1> to tensor<64xi64, #blocked1> loc(#loc43) + %o_t_8 = tt.splat %o_t : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_9 = tt.splat %o_t : i64 -> tensor<64xi64, #blocked1> loc(#loc43) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<64xi64, #blocked1> loc(#loc43) + %m_t = arith.extsi %T : i32 to i64 loc(#loc44) + %m_t_12 = tt.splat %m_t : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_13 = tt.splat %m_t : i64 -> tensor<64xi64, #blocked1> loc(#loc44) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<64xi64, #blocked1> loc(#loc44) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<64xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64x1xi1, #blocked> loc(#loc45) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc46) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc46) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_1 : tensor<1x128xi32, #blocked> loc(#loc47) + %m_x_19 = tt.broadcast %m_x : tensor<64x1xi1, #blocked> -> tensor<64x128xi1, #blocked> loc(#loc48) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<64x128xi1, #blocked> loc(#loc48) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<64x128xi1, #blocked> loc(#loc48) + %p_x = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64x1xi64, #blocked> loc(#loc49) + %p_x_22 = arith.muli %p_x, %cst_2 : tensor<64x1xi64, #blocked> loc(#loc50) + %p_x_23 = tt.splat %x : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked> loc(#loc51) + %p_x_24 = tt.addptr %p_x_23, %p_x_22 : tensor<64x1x!tt.ptr, #blocked>, tensor<64x1xi64, #blocked> loc(#loc51) + %p_x_25 = tt.broadcast %p_x_24 : tensor<64x1x!tt.ptr, #blocked> -> tensor<64x128x!tt.ptr, #blocked> loc(#loc52) + %p_x_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<64x128xi32, #blocked> loc(#loc52) + %p_x_27 = tt.addptr %p_x_25, %p_x_26 : tensor<64x128x!tt.ptr, #blocked>, tensor<64x128xi32, #blocked> loc(#loc52) + %p_y = tt.splat %y : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked> loc(#loc53) + %p_y_28 = tt.addptr %p_y, %p_x_22 : tensor<64x1x!tt.ptr, #blocked>, tensor<64x1xi64, #blocked> loc(#loc53) + %p_y_29 = tt.broadcast %p_y_28 : tensor<64x1x!tt.ptr, #blocked> -> tensor<64x128x!tt.ptr, #blocked> loc(#loc54) + %p_y_30 = tt.addptr %p_y_29, %p_x_26 : tensor<64x128x!tt.ptr, #blocked>, tensor<64x128xi32, #blocked> loc(#loc54) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<64x!tt.ptr, #blocked1> loc(#loc55) + %p_rstd_31 = tt.addptr %p_rstd, %o_t_11 : tensor<64x!tt.ptr, #blocked1>, tensor<64xi64, #blocked1> loc(#loc55) + %b_x = tt.load %p_x_27, %m_x_21, %cst_0 : tensor<64x128x!tt.ptr, #blocked> loc(#loc56) + %b_x_32 = arith.extf %b_x : tensor<64x128xbf16, #blocked> to tensor<64x128xf32, #blocked> loc(#loc57) + %b_rstd = arith.mulf %b_x_32, %b_x_32 : tensor<64x128xf32, #blocked> loc(#loc58) + %b_rstd_33 = "tt.reduce"(%b_rstd) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_40: f32 loc(callsite(#loc1 at #loc59)), %b_rstd_41: f32 loc(callsite(#loc1 at #loc59))): + %b_rstd_42 = arith.addf %b_rstd_40, %b_rstd_41 : f32 loc(#loc67) + tt.reduce.return %b_rstd_42 : f32 loc(#loc65) + }) : (tensor<64x128xf32, #blocked>) -> tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc65) + %b_rstd_34 = tt.splat %eps : f32 -> tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_36 = math.sqrt %b_rstd_35 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc61) + %b_rstd_37 = arith.divf %cst, %b_rstd_36 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc62) + %b_y = tt.expand_dims %b_rstd_37 {axis = 1 : i32} : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64x1xf32, #blocked> loc(#loc63) + %b_y_38 = tt.broadcast %b_y : tensor<64x1xf32, #blocked> -> tensor<64x128xf32, #blocked> loc(#loc64) + %b_y_39 = arith.mulf %b_x_32, %b_y_38 : tensor<64x128xf32, #blocked> loc(#loc64) + %0 = arith.truncf %b_y_39 : tensor<64x128xf32, #blocked> to tensor<64x128xbf16, #blocked> loc(#loc30) + tt.store %p_y_30, %0, %m_x_21 : tensor<64x128x!tt.ptr, #blocked> loc(#loc31) + %1 = ttg.convert_layout %b_rstd_37 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64xf32, #blocked1> loc(#loc32) + tt.store %p_rstd_31, %1, %m_t_15 : tensor<64x!tt.ptr, #blocked1> loc(#loc32) + tt.return loc(#loc33) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc39 = loc("i_t"(#loc2)) +#loc40 = loc("i_t"(#loc3)) +#loc41 = loc("o_t"(#loc4)) +#loc42 = loc("o_t"(#loc5)) +#loc43 = loc("o_t"(#loc6)) +#loc44 = loc("m_t"(#loc7)) +#loc45 = loc("m_x"(#loc8)) +#loc46 = loc("m_x"(#loc9)) +#loc47 = loc("m_x"(#loc10)) +#loc48 = loc("m_x"(#loc11)) +#loc49 = loc("p_x"(#loc12)) +#loc50 = loc("p_x"(#loc13)) +#loc51 = loc("p_x"(#loc14)) +#loc52 = loc("p_x"(#loc15)) +#loc53 = loc("p_y"(#loc16)) +#loc54 = loc("p_y"(#loc17)) +#loc55 = loc("p_rstd"(#loc18)) +#loc56 = loc("b_x"(#loc19)) +#loc57 = loc("b_x"(#loc20)) +#loc58 = loc("b_rstd"(#loc21)) +#loc60 = loc("b_rstd"(#loc25)) +#loc61 = loc("b_rstd"(#loc26)) +#loc62 = loc("b_rstd"(#loc27)) +#loc63 = loc("b_y"(#loc28)) +#loc64 = loc("b_y"(#loc29)) +#loc65 = loc(callsite(#loc22 at #loc59)) +#loc67 = loc(callsite(#loc24 at #loc65)) diff --git a/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/l2norm_fwd_kernel.ttir b/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/l2norm_fwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..85ebb8b128ceed91f07ab82f5c0fe10a876f6335 --- /dev/null +++ b/triton/gpu0/BSEHACDXVKFB4X35R7MBS7TQAL3SZDQMAE6DZPD6STCE36IJLMEA/l2norm_fwd_kernel.ttir @@ -0,0 +1,128 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc3 = loc(unknown) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc35 = loc("x"(#loc)) +#loc36 = loc("y"(#loc)) +#loc37 = loc("rstd"(#loc)) +#loc38 = loc("eps"(#loc)) +#loc39 = loc("T"(#loc)) +#loc62 = loc("b_rstd"(#loc25)) +#loc68 = loc(callsite(#loc3 at #loc62)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %b_x = arith.constant dense<0.000000e+00> : tensor<64x128xbf16> loc(#loc40) + %b_rstd = arith.constant dense<1.000000e+00> : tensor<64xf32> loc(#loc41) + %cst = arith.constant dense<128> : tensor<64x1xi64> loc(#loc3) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc42) + %o_t = arith.constant 64 : i64 loc(#loc43) + %i_t = tt.get_program_id x : i32 loc(#loc44) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc45) + %o_t_1 = arith.muli %i_t_0, %o_t : i64 loc(#loc43) + %o_t_2 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc46) + %o_t_3 = arith.extsi %o_t_2 : tensor<64xi32> to tensor<64xi64> loc(#loc47) + %o_t_4 = tt.splat %o_t_1 : i64 -> tensor<64xi64> loc(#loc47) + %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<64xi64> loc(#loc47) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc48) + %m_t = arith.extsi %T : i32 to i64 loc(#loc49) + %m_t_6 = tt.splat %m_t : i64 -> tensor<64xi64> loc(#loc49) + %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<64xi64> loc(#loc49) + %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc50) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc51) + %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc42) + %m_x_11 = tt.broadcast %m_x_8 : tensor<64x1xi1> -> tensor<64x128xi1> loc(#loc52) + %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<64x128xi1> loc(#loc52) + %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<64x128xi1> loc(#loc52) + %p_x = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc53) + %p_x_14 = arith.muli %p_x, %cst : tensor<64x1xi64> loc(#loc54) + %p_x_15 = tt.splat %x : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc55) + %p_x_16 = tt.addptr %p_x_15, %p_x_14 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc55) + %p_x_17 = tt.broadcast %p_x_16 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc56) + %p_x_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc56) + %p_x_19 = tt.addptr %p_x_17, %p_x_18 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc56) + %p_y = tt.splat %y : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc57) + %p_y_20 = tt.addptr %p_y, %p_x_14 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc57) + %p_y_21 = tt.broadcast %p_y_20 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc58) + %p_y_22 = tt.addptr %p_y_21, %p_x_18 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<64x!tt.ptr> loc(#loc59) + %p_rstd_23 = tt.addptr %p_rstd, %o_t_5 : tensor<64x!tt.ptr>, tensor<64xi64> loc(#loc59) + %b_x_24 = tt.load %p_x_19, %m_x_13, %b_x : tensor<64x128x!tt.ptr> loc(#loc40) + %b_x_25 = arith.extf %b_x_24 : tensor<64x128xbf16> to tensor<64x128xf32> loc(#loc60) + %b_rstd_26 = arith.mulf %b_x_25, %b_x_25 : tensor<64x128xf32> loc(#loc61) + %b_rstd_27 = "tt.reduce"(%b_rstd_26) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_34: f32 loc(callsite(#loc3 at #loc62)), %b_rstd_35: f32 loc(callsite(#loc3 at #loc62))): + %b_rstd_36 = arith.addf %b_rstd_34, %b_rstd_35 : f32 loc(#loc69) + tt.reduce.return %b_rstd_36 : f32 loc(#loc67) + }) : (tensor<64x128xf32>) -> tensor<64xf32> loc(#loc67) + %b_rstd_28 = tt.splat %eps : f32 -> tensor<64xf32> loc(#loc63) + %b_rstd_29 = arith.addf %b_rstd_27, %b_rstd_28 : tensor<64xf32> loc(#loc63) + %b_rstd_30 = math.sqrt %b_rstd_29 : tensor<64xf32> loc(#loc64) + %b_rstd_31 = arith.divf %b_rstd, %b_rstd_30 : tensor<64xf32> loc(#loc41) + %b_y = tt.expand_dims %b_rstd_31 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc65) + %b_y_32 = tt.broadcast %b_y : tensor<64x1xf32> -> tensor<64x128xf32> loc(#loc66) + %b_y_33 = arith.mulf %b_x_25, %b_y_32 : tensor<64x128xf32> loc(#loc66) + %0 = arith.truncf %b_y_33 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc31) + tt.store %p_y_22, %0, %m_x_13 : tensor<64x128x!tt.ptr> loc(#loc32) + tt.store %p_rstd_23, %b_rstd_31, %m_t_7 : tensor<64x!tt.ptr> loc(#loc33) + tt.return loc(#loc34) + } loc(#loc) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc40 = loc("b_x"(#loc1)) +#loc41 = loc("b_rstd"(#loc2)) +#loc42 = loc("m_x"(#loc4)) +#loc43 = loc("o_t"(#loc5)) +#loc44 = loc("i_t"(#loc6)) +#loc45 = loc("i_t"(#loc7)) +#loc46 = loc("o_t"(#loc8)) +#loc47 = loc("o_t"(#loc9)) +#loc48 = loc("o_d"(#loc10)) +#loc49 = loc("m_t"(#loc11)) +#loc50 = loc("m_x"(#loc12)) +#loc51 = loc("m_x"(#loc13)) +#loc52 = loc("m_x"(#loc14)) +#loc53 = loc("p_x"(#loc15)) +#loc54 = loc("p_x"(#loc16)) +#loc55 = loc("p_x"(#loc17)) +#loc56 = loc("p_x"(#loc18)) +#loc57 = loc("p_y"(#loc19)) +#loc58 = loc("p_y"(#loc20)) +#loc59 = loc("p_rstd"(#loc21)) +#loc60 = loc("b_x"(#loc22)) +#loc61 = loc("b_rstd"(#loc23)) +#loc63 = loc("b_rstd"(#loc27)) +#loc64 = loc("b_rstd"(#loc28)) +#loc65 = loc("b_y"(#loc29)) +#loc66 = loc("b_y"(#loc30)) +#loc67 = loc(callsite(#loc24 at #loc62)) +#loc69 = loc(callsite(#loc26 at #loc67)) diff --git a/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/__grp__l2norm_fwd_kernel.json b/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/__grp__l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..aa53c67677fbf8000eff5680c5bc67d1cb71879b --- /dev/null +++ b/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/__grp__l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_fwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/l2norm_fwd_kernel.source", "l2norm_fwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/l2norm_fwd_kernel.ttir", "l2norm_fwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/l2norm_fwd_kernel.ttgir", "l2norm_fwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/l2norm_fwd_kernel.llir", "l2norm_fwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/l2norm_fwd_kernel.ptx", "l2norm_fwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/l2norm_fwd_kernel.cubin", "l2norm_fwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/l2norm_fwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/l2norm_fwd_kernel.cubin b/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/l2norm_fwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..f0c87b49c11949a0803e6250aba0a6d0f32f274e Binary files /dev/null and b/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/l2norm_fwd_kernel.cubin differ diff --git a/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/l2norm_fwd_kernel.json b/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..d13d913bf014058b4ad1ee2e0248fb60b4230c19 --- /dev/null +++ b/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"hash": "0d381f0f87de816f21488910979817f16cac2d2c73e8c73f5052457f06066826", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 1, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 32, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_fwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/l2norm_fwd_kernel.llir b/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/l2norm_fwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..4bd998a916deea1dcc376f2130c5dd7440fb9787 --- /dev/null +++ b/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/l2norm_fwd_kernel.llir @@ -0,0 +1,633 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 +@.str.2 = private unnamed_addr constant [17 x i8] c"__CUDA_PREC_SQRT\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_fwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, float %3, i32 %4, ptr addrspace(1) readnone captures(none) %5, ptr addrspace(1) readnone captures(none) %6) local_unnamed_addr #0 !dbg !5 { + %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %9 = zext nneg i32 %8 to i64, !dbg !14 + %10 = shl nuw nsw i64 %9, 3, !dbg !15 + %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !16 + %12 = and i32 %11, 16, !dbg !16 + %.lobit = lshr exact i32 %12, 4, !dbg !16 + %13 = or disjoint i32 %.lobit, 2, !dbg !16 + %14 = or disjoint i32 %.lobit, 4, !dbg !16 + %15 = or disjoint i32 %.lobit, 6, !dbg !16 + %16 = zext nneg i32 %.lobit to i64, !dbg !17 + %17 = zext nneg i32 %13 to i64, !dbg !17 + %18 = zext nneg i32 %14 to i64, !dbg !17 + %19 = zext nneg i32 %15 to i64, !dbg !17 + %20 = or disjoint i64 %10, %16, !dbg !17 + %21 = or disjoint i64 %10, %17, !dbg !17 + %22 = or disjoint i64 %10, %18, !dbg !17 + %23 = or disjoint i64 %10, %19, !dbg !17 + %24 = sext i32 %4 to i64, !dbg !18 + %25 = icmp slt i64 %20, %24, !dbg !18 + %26 = icmp slt i64 %21, %24, !dbg !18 + %27 = icmp slt i64 %22, %24, !dbg !18 + %28 = icmp slt i64 %23, %24, !dbg !18 + %29 = shl nuw nsw i32 %11, 3, !dbg !19 + %30 = and i32 %29, 120, !dbg !19 + %31 = shl nuw nsw i64 %20, 7, !dbg !20 + %32 = shl nuw nsw i64 %21, 7, !dbg !20 + %33 = shl nuw nsw i64 %22, 7, !dbg !20 + %34 = shl nuw nsw i64 %23, 7, !dbg !20 + %35 = getelementptr bfloat, ptr addrspace(1) %0, i64 %31, !dbg !21 + %36 = getelementptr bfloat, ptr addrspace(1) %0, i64 %32, !dbg !21 + %37 = getelementptr bfloat, ptr addrspace(1) %0, i64 %33, !dbg !21 + %38 = getelementptr bfloat, ptr addrspace(1) %0, i64 %34, !dbg !21 + %39 = zext nneg i32 %30 to i64, !dbg !22 + %40 = getelementptr bfloat, ptr addrspace(1) %35, i64 %39, !dbg !22 + %41 = getelementptr bfloat, ptr addrspace(1) %36, i64 %39, !dbg !22 + %42 = getelementptr bfloat, ptr addrspace(1) %37, i64 %39, !dbg !22 + %43 = getelementptr bfloat, ptr addrspace(1) %38, i64 %39, !dbg !22 + %44 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %40, i1 %25) #6, !dbg !23 + %45 = extractvalue { i32, i32, i32, i32 } %44, 0, !dbg !23 + %46 = bitcast i32 %45 to <2 x bfloat>, !dbg !23 + %47 = extractvalue { i32, i32, i32, i32 } %44, 1, !dbg !23 + %48 = bitcast i32 %47 to <2 x bfloat>, !dbg !23 + %49 = extractvalue { i32, i32, i32, i32 } %44, 2, !dbg !23 + %50 = bitcast i32 %49 to <2 x bfloat>, !dbg !23 + %51 = extractvalue { i32, i32, i32, i32 } %44, 3, !dbg !23 + %52 = bitcast i32 %51 to <2 x bfloat>, !dbg !23 + %53 = extractelement <2 x bfloat> %46, i64 0, !dbg !23 + %54 = extractelement <2 x bfloat> %46, i64 1, !dbg !23 + %55 = extractelement <2 x bfloat> %48, i64 0, !dbg !23 + %56 = extractelement <2 x bfloat> %48, i64 1, !dbg !23 + %57 = extractelement <2 x bfloat> %50, i64 0, !dbg !23 + %58 = extractelement <2 x bfloat> %50, i64 1, !dbg !23 + %59 = extractelement <2 x bfloat> %52, i64 0, !dbg !23 + %60 = extractelement <2 x bfloat> %52, i64 1, !dbg !23 + %61 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %41, i1 %26) #6, !dbg !23 + %62 = extractvalue { i32, i32, i32, i32 } %61, 0, !dbg !23 + %63 = bitcast i32 %62 to <2 x bfloat>, !dbg !23 + %64 = extractvalue { i32, i32, i32, i32 } %61, 1, !dbg !23 + %65 = bitcast i32 %64 to <2 x bfloat>, !dbg !23 + %66 = extractvalue { i32, i32, i32, i32 } %61, 2, !dbg !23 + %67 = bitcast i32 %66 to <2 x bfloat>, !dbg !23 + %68 = extractvalue { i32, i32, i32, i32 } %61, 3, !dbg !23 + %69 = bitcast i32 %68 to <2 x bfloat>, !dbg !23 + %70 = extractelement <2 x bfloat> %63, i64 0, !dbg !23 + %71 = extractelement <2 x bfloat> %63, i64 1, !dbg !23 + %72 = extractelement <2 x bfloat> %65, i64 0, !dbg !23 + %73 = extractelement <2 x bfloat> %65, i64 1, !dbg !23 + %74 = extractelement <2 x bfloat> %67, i64 0, !dbg !23 + %75 = extractelement <2 x bfloat> %67, i64 1, !dbg !23 + %76 = extractelement <2 x bfloat> %69, i64 0, !dbg !23 + %77 = extractelement <2 x bfloat> %69, i64 1, !dbg !23 + %78 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %42, i1 %27) #6, !dbg !23 + %79 = extractvalue { i32, i32, i32, i32 } %78, 0, !dbg !23 + %80 = bitcast i32 %79 to <2 x bfloat>, !dbg !23 + %81 = extractvalue { i32, i32, i32, i32 } %78, 1, !dbg !23 + %82 = bitcast i32 %81 to <2 x bfloat>, !dbg !23 + %83 = extractvalue { i32, i32, i32, i32 } %78, 2, !dbg !23 + %84 = bitcast i32 %83 to <2 x bfloat>, !dbg !23 + %85 = extractvalue { i32, i32, i32, i32 } %78, 3, !dbg !23 + %86 = bitcast i32 %85 to <2 x bfloat>, !dbg !23 + %87 = extractelement <2 x bfloat> %80, i64 0, !dbg !23 + %88 = extractelement <2 x bfloat> %80, i64 1, !dbg !23 + %89 = extractelement <2 x bfloat> %82, i64 0, !dbg !23 + %90 = extractelement <2 x bfloat> %82, i64 1, !dbg !23 + %91 = extractelement <2 x bfloat> %84, i64 0, !dbg !23 + %92 = extractelement <2 x bfloat> %84, i64 1, !dbg !23 + %93 = extractelement <2 x bfloat> %86, i64 0, !dbg !23 + %94 = extractelement <2 x bfloat> %86, i64 1, !dbg !23 + %95 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %43, i1 %28) #6, !dbg !23 + %96 = extractvalue { i32, i32, i32, i32 } %95, 0, !dbg !23 + %97 = bitcast i32 %96 to <2 x bfloat>, !dbg !23 + %98 = extractvalue { i32, i32, i32, i32 } %95, 1, !dbg !23 + %99 = bitcast i32 %98 to <2 x bfloat>, !dbg !23 + %100 = extractvalue { i32, i32, i32, i32 } %95, 2, !dbg !23 + %101 = bitcast i32 %100 to <2 x bfloat>, !dbg !23 + %102 = extractvalue { i32, i32, i32, i32 } %95, 3, !dbg !23 + %103 = bitcast i32 %102 to <2 x bfloat>, !dbg !23 + %104 = extractelement <2 x bfloat> %97, i64 0, !dbg !23 + %105 = extractelement <2 x bfloat> %97, i64 1, !dbg !23 + %106 = extractelement <2 x bfloat> %99, i64 0, !dbg !23 + %107 = extractelement <2 x bfloat> %99, i64 1, !dbg !23 + %108 = extractelement <2 x bfloat> %101, i64 0, !dbg !23 + %109 = extractelement <2 x bfloat> %101, i64 1, !dbg !23 + %110 = extractelement <2 x bfloat> %103, i64 0, !dbg !23 + %111 = extractelement <2 x bfloat> %103, i64 1, !dbg !23 + %112 = fpext bfloat %53 to float, !dbg !24 + %113 = fpext bfloat %54 to float, !dbg !24 + %114 = fpext bfloat %55 to float, !dbg !24 + %115 = fpext bfloat %56 to float, !dbg !24 + %116 = fpext bfloat %57 to float, !dbg !24 + %117 = fpext bfloat %58 to float, !dbg !24 + %118 = fpext bfloat %59 to float, !dbg !24 + %119 = fpext bfloat %60 to float, !dbg !24 + %120 = fpext bfloat %70 to float, !dbg !24 + %121 = fpext bfloat %71 to float, !dbg !24 + %122 = fpext bfloat %72 to float, !dbg !24 + %123 = fpext bfloat %73 to float, !dbg !24 + %124 = fpext bfloat %74 to float, !dbg !24 + %125 = fpext bfloat %75 to float, !dbg !24 + %126 = fpext bfloat %76 to float, !dbg !24 + %127 = fpext bfloat %77 to float, !dbg !24 + %128 = fpext bfloat %87 to float, !dbg !24 + %129 = fpext bfloat %88 to float, !dbg !24 + %130 = fpext bfloat %89 to float, !dbg !24 + %131 = fpext bfloat %90 to float, !dbg !24 + %132 = fpext bfloat %91 to float, !dbg !24 + %133 = fpext bfloat %92 to float, !dbg !24 + %134 = fpext bfloat %93 to float, !dbg !24 + %135 = fpext bfloat %94 to float, !dbg !24 + %136 = fpext bfloat %104 to float, !dbg !24 + %137 = fpext bfloat %105 to float, !dbg !24 + %138 = fpext bfloat %106 to float, !dbg !24 + %139 = fpext bfloat %107 to float, !dbg !24 + %140 = fpext bfloat %108 to float, !dbg !24 + %141 = fpext bfloat %109 to float, !dbg !24 + %142 = fpext bfloat %110 to float, !dbg !24 + %143 = fpext bfloat %111 to float, !dbg !24 + %144 = fmul float %112, %112, !dbg !25 + %145 = fmul float %113, %113, !dbg !25 + %146 = fmul float %114, %114, !dbg !25 + %147 = fmul float %115, %115, !dbg !25 + %148 = fmul float %116, %116, !dbg !25 + %149 = fmul float %117, %117, !dbg !25 + %150 = fmul float %118, %118, !dbg !25 + %151 = fmul float %119, %119, !dbg !25 + %152 = fmul float %120, %120, !dbg !25 + %153 = fmul float %121, %121, !dbg !25 + %154 = fmul float %122, %122, !dbg !25 + %155 = fmul float %123, %123, !dbg !25 + %156 = fmul float %124, %124, !dbg !25 + %157 = fmul float %125, %125, !dbg !25 + %158 = fmul float %126, %126, !dbg !25 + %159 = fmul float %127, %127, !dbg !25 + %160 = fmul float %128, %128, !dbg !25 + %161 = fmul float %129, %129, !dbg !25 + %162 = fmul float %130, %130, !dbg !25 + %163 = fmul float %131, %131, !dbg !25 + %164 = fmul float %132, %132, !dbg !25 + %165 = fmul float %133, %133, !dbg !25 + %166 = fmul float %134, %134, !dbg !25 + %167 = fmul float %135, %135, !dbg !25 + %168 = fmul float %136, %136, !dbg !25 + %169 = fmul float %137, %137, !dbg !25 + %170 = fmul float %138, %138, !dbg !25 + %171 = fmul float %139, %139, !dbg !25 + %172 = fmul float %140, %140, !dbg !25 + %173 = fmul float %141, %141, !dbg !25 + %174 = fmul float %142, %142, !dbg !25 + %175 = fmul float %143, %143, !dbg !25 + %176 = fadd float %144, %145, !dbg !26 + %177 = fadd float %146, %176, !dbg !26 + %178 = fadd float %147, %177, !dbg !26 + %179 = fadd float %148, %178, !dbg !26 + %180 = fadd float %149, %179, !dbg !26 + %181 = fadd float %150, %180, !dbg !26 + %182 = fadd float %151, %181, !dbg !26 + %183 = fadd float %152, %153, !dbg !26 + %184 = fadd float %154, %183, !dbg !26 + %185 = fadd float %155, %184, !dbg !26 + %186 = fadd float %156, %185, !dbg !26 + %187 = fadd float %157, %186, !dbg !26 + %188 = fadd float %158, %187, !dbg !26 + %189 = fadd float %159, %188, !dbg !26 + %190 = fadd float %160, %161, !dbg !26 + %191 = fadd float %162, %190, !dbg !26 + %192 = fadd float %163, %191, !dbg !26 + %193 = fadd float %164, %192, !dbg !26 + %194 = fadd float %165, %193, !dbg !26 + %195 = fadd float %166, %194, !dbg !26 + %196 = fadd float %167, %195, !dbg !26 + %197 = fadd float %168, %169, !dbg !26 + %198 = fadd float %170, %197, !dbg !26 + %199 = fadd float %171, %198, !dbg !26 + %200 = fadd float %172, %199, !dbg !26 + %201 = fadd float %173, %200, !dbg !26 + %202 = fadd float %174, %201, !dbg !26 + %203 = fadd float %175, %202, !dbg !26 + %204 = bitcast float %182 to i32, !dbg !29 + %205 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %204, i32 8, i32 31), !dbg !29 + %206 = bitcast i32 %205 to float, !dbg !29 + %207 = fadd float %182, %206, !dbg !26 + %208 = bitcast float %207 to i32, !dbg !29 + %209 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %208, i32 4, i32 31), !dbg !29 + %210 = bitcast i32 %209 to float, !dbg !29 + %211 = fadd float %207, %210, !dbg !26 + %212 = bitcast float %211 to i32, !dbg !29 + %213 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %212, i32 2, i32 31), !dbg !29 + %214 = bitcast i32 %213 to float, !dbg !29 + %215 = fadd float %211, %214, !dbg !26 + %216 = bitcast float %215 to i32, !dbg !29 + %217 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %216, i32 1, i32 31), !dbg !29 + %218 = bitcast i32 %217 to float, !dbg !29 + %219 = fadd float %215, %218, !dbg !26 + %220 = bitcast float %189 to i32, !dbg !29 + %221 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %220, i32 8, i32 31), !dbg !29 + %222 = bitcast i32 %221 to float, !dbg !29 + %223 = fadd float %189, %222, !dbg !26 + %224 = bitcast float %223 to i32, !dbg !29 + %225 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %224, i32 4, i32 31), !dbg !29 + %226 = bitcast i32 %225 to float, !dbg !29 + %227 = fadd float %223, %226, !dbg !26 + %228 = bitcast float %227 to i32, !dbg !29 + %229 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %228, i32 2, i32 31), !dbg !29 + %230 = bitcast i32 %229 to float, !dbg !29 + %231 = fadd float %227, %230, !dbg !26 + %232 = bitcast float %231 to i32, !dbg !29 + %233 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %232, i32 1, i32 31), !dbg !29 + %234 = bitcast i32 %233 to float, !dbg !29 + %235 = fadd float %231, %234, !dbg !26 + %236 = bitcast float %196 to i32, !dbg !29 + %237 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %236, i32 8, i32 31), !dbg !29 + %238 = bitcast i32 %237 to float, !dbg !29 + %239 = fadd float %196, %238, !dbg !26 + %240 = bitcast float %239 to i32, !dbg !29 + %241 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %240, i32 4, i32 31), !dbg !29 + %242 = bitcast i32 %241 to float, !dbg !29 + %243 = fadd float %239, %242, !dbg !26 + %244 = bitcast float %243 to i32, !dbg !29 + %245 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %244, i32 2, i32 31), !dbg !29 + %246 = bitcast i32 %245 to float, !dbg !29 + %247 = fadd float %243, %246, !dbg !26 + %248 = bitcast float %247 to i32, !dbg !29 + %249 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %248, i32 1, i32 31), !dbg !29 + %250 = bitcast i32 %249 to float, !dbg !29 + %251 = fadd float %247, %250, !dbg !26 + %252 = bitcast float %203 to i32, !dbg !29 + %253 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %252, i32 8, i32 31), !dbg !29 + %254 = bitcast i32 %253 to float, !dbg !29 + %255 = fadd float %203, %254, !dbg !26 + %256 = bitcast float %255 to i32, !dbg !29 + %257 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %256, i32 4, i32 31), !dbg !29 + %258 = bitcast i32 %257 to float, !dbg !29 + %259 = fadd float %255, %258, !dbg !26 + %260 = bitcast float %259 to i32, !dbg !29 + %261 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %260, i32 2, i32 31), !dbg !29 + %262 = bitcast i32 %261 to float, !dbg !29 + %263 = fadd float %259, %262, !dbg !26 + %264 = bitcast float %263 to i32, !dbg !29 + %265 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %264, i32 1, i32 31), !dbg !29 + %266 = bitcast i32 %265 to float, !dbg !29 + %267 = fadd float %263, %266, !dbg !26 + %268 = fadd float %3, %219, !dbg !32 + %269 = fadd float %3, %235, !dbg !32 + %270 = fadd float %3, %251, !dbg !32 + %271 = fadd float %3, %267, !dbg !32 + %272 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i = icmp eq i32 %272, 0, !dbg !33 + %273 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i = icmp eq i32 %273, 0, !dbg !33 + br i1 %.not.i, label %279, label %274, !dbg !33 + +274: ; preds = %7 + br i1 %.not1.i, label %277, label %275, !dbg !33 + +275: ; preds = %274 + %276 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %268) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +277: ; preds = %274 + %278 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %268) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +279: ; preds = %7 + br i1 %.not1.i, label %282, label %280, !dbg !33 + +280: ; preds = %279 + %281 = tail call float @llvm.nvvm.sqrt.rn.f(float %268) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +282: ; preds = %279 + %283 = tail call float @llvm.nvvm.sqrt.approx.f(float %268) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +__nv_sqrtf.exit: ; preds = %275, %277, %280, %282 + %.0.i = phi float [ %276, %275 ], [ %278, %277 ], [ %281, %280 ], [ %283, %282 ], !dbg !33 + %284 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i1 = icmp eq i32 %284, 0, !dbg !33 + %285 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i4 = icmp eq i32 %285, 0, !dbg !33 + br i1 %.not.i1, label %291, label %286, !dbg !33 + +286: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %289, label %287, !dbg !33 + +287: ; preds = %286 + %288 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %269) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +289: ; preds = %286 + %290 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %269) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +291: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %294, label %292, !dbg !33 + +292: ; preds = %291 + %293 = tail call float @llvm.nvvm.sqrt.rn.f(float %269) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +294: ; preds = %291 + %295 = tail call float @llvm.nvvm.sqrt.approx.f(float %269) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +__nv_sqrtf.exit5: ; preds = %287, %289, %292, %294 + %.0.i3 = phi float [ %288, %287 ], [ %290, %289 ], [ %293, %292 ], [ %295, %294 ], !dbg !33 + %296 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i6 = icmp eq i32 %296, 0, !dbg !33 + %297 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i9 = icmp eq i32 %297, 0, !dbg !33 + br i1 %.not.i6, label %303, label %298, !dbg !33 + +298: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %301, label %299, !dbg !33 + +299: ; preds = %298 + %300 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %270) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +301: ; preds = %298 + %302 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %270) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +303: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %306, label %304, !dbg !33 + +304: ; preds = %303 + %305 = tail call float @llvm.nvvm.sqrt.rn.f(float %270) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +306: ; preds = %303 + %307 = tail call float @llvm.nvvm.sqrt.approx.f(float %270) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +__nv_sqrtf.exit10: ; preds = %299, %301, %304, %306 + %.0.i8 = phi float [ %300, %299 ], [ %302, %301 ], [ %305, %304 ], [ %307, %306 ], !dbg !33 + %308 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i11 = icmp eq i32 %308, 0, !dbg !33 + %309 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i14 = icmp eq i32 %309, 0, !dbg !33 + br i1 %.not.i11, label %315, label %310, !dbg !33 + +310: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %313, label %311, !dbg !33 + +311: ; preds = %310 + %312 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %271) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +313: ; preds = %310 + %314 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %271) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +315: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %318, label %316, !dbg !33 + +316: ; preds = %315 + %317 = tail call float @llvm.nvvm.sqrt.rn.f(float %271) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +318: ; preds = %315 + %319 = tail call float @llvm.nvvm.sqrt.approx.f(float %271) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +__nv_sqrtf.exit15: ; preds = %311, %313, %316, %318 + %.0.i13 = phi float [ %312, %311 ], [ %314, %313 ], [ %317, %316 ], [ %319, %318 ], !dbg !33 + %320 = and i32 %11, 7, !dbg !16 + %321 = zext nneg i32 %320 to i64, !dbg !17 + %322 = or disjoint i64 %10, %321, !dbg !17 + %323 = getelementptr float, ptr addrspace(1) %2, i64 %322, !dbg !34 + %324 = getelementptr bfloat, ptr addrspace(1) %1, i64 %34, !dbg !35 + %325 = getelementptr bfloat, ptr addrspace(1) %324, i64 %39, !dbg !36 + %326 = getelementptr bfloat, ptr addrspace(1) %1, i64 %33, !dbg !35 + %327 = getelementptr bfloat, ptr addrspace(1) %326, i64 %39, !dbg !36 + %328 = getelementptr bfloat, ptr addrspace(1) %1, i64 %32, !dbg !35 + %329 = getelementptr bfloat, ptr addrspace(1) %328, i64 %39, !dbg !36 + %330 = getelementptr bfloat, ptr addrspace(1) %1, i64 %31, !dbg !35 + %331 = getelementptr bfloat, ptr addrspace(1) %330, i64 %39, !dbg !36 + %332 = icmp slt i64 %322, %24, !dbg !18 + %333 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i), !dbg !37 + %334 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i3), !dbg !37 + %335 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i8), !dbg !37 + %336 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i13), !dbg !37 + %337 = fmul float %333, %112, !dbg !38 + %338 = fmul float %333, %113, !dbg !38 + %339 = fmul float %333, %114, !dbg !38 + %340 = fmul float %333, %115, !dbg !38 + %341 = fmul float %333, %116, !dbg !38 + %342 = fmul float %333, %117, !dbg !38 + %343 = fmul float %333, %118, !dbg !38 + %344 = fmul float %333, %119, !dbg !38 + %345 = fmul float %334, %120, !dbg !38 + %346 = fmul float %334, %121, !dbg !38 + %347 = fmul float %334, %122, !dbg !38 + %348 = fmul float %334, %123, !dbg !38 + %349 = fmul float %334, %124, !dbg !38 + %350 = fmul float %334, %125, !dbg !38 + %351 = fmul float %334, %126, !dbg !38 + %352 = fmul float %334, %127, !dbg !38 + %353 = fmul float %335, %128, !dbg !38 + %354 = fmul float %335, %129, !dbg !38 + %355 = fmul float %335, %130, !dbg !38 + %356 = fmul float %335, %131, !dbg !38 + %357 = fmul float %335, %132, !dbg !38 + %358 = fmul float %335, %133, !dbg !38 + %359 = fmul float %335, %134, !dbg !38 + %360 = fmul float %335, %135, !dbg !38 + %361 = fmul float %336, %136, !dbg !38 + %362 = fmul float %336, %137, !dbg !38 + %363 = fmul float %336, %138, !dbg !38 + %364 = fmul float %336, %139, !dbg !38 + %365 = fmul float %336, %140, !dbg !38 + %366 = fmul float %336, %141, !dbg !38 + %367 = fmul float %336, %142, !dbg !38 + %368 = fmul float %336, %143, !dbg !38 + %369 = fptrunc float %337 to bfloat, !dbg !39 + %370 = fptrunc float %338 to bfloat, !dbg !39 + %371 = fptrunc float %339 to bfloat, !dbg !39 + %372 = fptrunc float %340 to bfloat, !dbg !39 + %373 = fptrunc float %341 to bfloat, !dbg !39 + %374 = fptrunc float %342 to bfloat, !dbg !39 + %375 = fptrunc float %343 to bfloat, !dbg !39 + %376 = fptrunc float %344 to bfloat, !dbg !39 + %377 = fptrunc float %345 to bfloat, !dbg !39 + %378 = fptrunc float %346 to bfloat, !dbg !39 + %379 = fptrunc float %347 to bfloat, !dbg !39 + %380 = fptrunc float %348 to bfloat, !dbg !39 + %381 = fptrunc float %349 to bfloat, !dbg !39 + %382 = fptrunc float %350 to bfloat, !dbg !39 + %383 = fptrunc float %351 to bfloat, !dbg !39 + %384 = fptrunc float %352 to bfloat, !dbg !39 + %385 = fptrunc float %353 to bfloat, !dbg !39 + %386 = fptrunc float %354 to bfloat, !dbg !39 + %387 = fptrunc float %355 to bfloat, !dbg !39 + %388 = fptrunc float %356 to bfloat, !dbg !39 + %389 = fptrunc float %357 to bfloat, !dbg !39 + %390 = fptrunc float %358 to bfloat, !dbg !39 + %391 = fptrunc float %359 to bfloat, !dbg !39 + %392 = fptrunc float %360 to bfloat, !dbg !39 + %393 = fptrunc float %361 to bfloat, !dbg !39 + %394 = fptrunc float %362 to bfloat, !dbg !39 + %395 = fptrunc float %363 to bfloat, !dbg !39 + %396 = fptrunc float %364 to bfloat, !dbg !39 + %397 = fptrunc float %365 to bfloat, !dbg !39 + %398 = fptrunc float %366 to bfloat, !dbg !39 + %399 = fptrunc float %367 to bfloat, !dbg !39 + %400 = fptrunc float %368 to bfloat, !dbg !39 + %401 = insertelement <2 x bfloat> poison, bfloat %369, i64 0, !dbg !40 + %402 = insertelement <2 x bfloat> %401, bfloat %370, i64 1, !dbg !40 + %403 = bitcast <2 x bfloat> %402 to i32, !dbg !40 + %404 = insertelement <2 x bfloat> poison, bfloat %371, i64 0, !dbg !40 + %405 = insertelement <2 x bfloat> %404, bfloat %372, i64 1, !dbg !40 + %406 = bitcast <2 x bfloat> %405 to i32, !dbg !40 + %407 = insertelement <2 x bfloat> poison, bfloat %373, i64 0, !dbg !40 + %408 = insertelement <2 x bfloat> %407, bfloat %374, i64 1, !dbg !40 + %409 = bitcast <2 x bfloat> %408 to i32, !dbg !40 + %410 = insertelement <2 x bfloat> poison, bfloat %375, i64 0, !dbg !40 + %411 = insertelement <2 x bfloat> %410, bfloat %376, i64 1, !dbg !40 + %412 = bitcast <2 x bfloat> %411 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %403, i32 %406, i32 %409, i32 %412, ptr addrspace(1) %331, i1 %25) #6, !dbg !40 + %413 = insertelement <2 x bfloat> poison, bfloat %377, i64 0, !dbg !40 + %414 = insertelement <2 x bfloat> %413, bfloat %378, i64 1, !dbg !40 + %415 = bitcast <2 x bfloat> %414 to i32, !dbg !40 + %416 = insertelement <2 x bfloat> poison, bfloat %379, i64 0, !dbg !40 + %417 = insertelement <2 x bfloat> %416, bfloat %380, i64 1, !dbg !40 + %418 = bitcast <2 x bfloat> %417 to i32, !dbg !40 + %419 = insertelement <2 x bfloat> poison, bfloat %381, i64 0, !dbg !40 + %420 = insertelement <2 x bfloat> %419, bfloat %382, i64 1, !dbg !40 + %421 = bitcast <2 x bfloat> %420 to i32, !dbg !40 + %422 = insertelement <2 x bfloat> poison, bfloat %383, i64 0, !dbg !40 + %423 = insertelement <2 x bfloat> %422, bfloat %384, i64 1, !dbg !40 + %424 = bitcast <2 x bfloat> %423 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %415, i32 %418, i32 %421, i32 %424, ptr addrspace(1) %329, i1 %26) #6, !dbg !40 + %425 = insertelement <2 x bfloat> poison, bfloat %385, i64 0, !dbg !40 + %426 = insertelement <2 x bfloat> %425, bfloat %386, i64 1, !dbg !40 + %427 = bitcast <2 x bfloat> %426 to i32, !dbg !40 + %428 = insertelement <2 x bfloat> poison, bfloat %387, i64 0, !dbg !40 + %429 = insertelement <2 x bfloat> %428, bfloat %388, i64 1, !dbg !40 + %430 = bitcast <2 x bfloat> %429 to i32, !dbg !40 + %431 = insertelement <2 x bfloat> poison, bfloat %389, i64 0, !dbg !40 + %432 = insertelement <2 x bfloat> %431, bfloat %390, i64 1, !dbg !40 + %433 = bitcast <2 x bfloat> %432 to i32, !dbg !40 + %434 = insertelement <2 x bfloat> poison, bfloat %391, i64 0, !dbg !40 + %435 = insertelement <2 x bfloat> %434, bfloat %392, i64 1, !dbg !40 + %436 = bitcast <2 x bfloat> %435 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %427, i32 %430, i32 %433, i32 %436, ptr addrspace(1) %327, i1 %27) #6, !dbg !40 + %437 = insertelement <2 x bfloat> poison, bfloat %393, i64 0, !dbg !40 + %438 = insertelement <2 x bfloat> %437, bfloat %394, i64 1, !dbg !40 + %439 = bitcast <2 x bfloat> %438 to i32, !dbg !40 + %440 = insertelement <2 x bfloat> poison, bfloat %395, i64 0, !dbg !40 + %441 = insertelement <2 x bfloat> %440, bfloat %396, i64 1, !dbg !40 + %442 = bitcast <2 x bfloat> %441 to i32, !dbg !40 + %443 = insertelement <2 x bfloat> poison, bfloat %397, i64 0, !dbg !40 + %444 = insertelement <2 x bfloat> %443, bfloat %398, i64 1, !dbg !40 + %445 = bitcast <2 x bfloat> %444 to i32, !dbg !40 + %446 = insertelement <2 x bfloat> poison, bfloat %399, i64 0, !dbg !40 + %447 = insertelement <2 x bfloat> %446, bfloat %400, i64 1, !dbg !40 + %448 = bitcast <2 x bfloat> %447 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %439, i32 %442, i32 %445, i32 %448, ptr addrspace(1) %325, i1 %28) #6, !dbg !40 + %449 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %12, !dbg !41 + %450 = bitcast float %333 to i32, !dbg !41 + %451 = bitcast float %334 to i32, !dbg !41 + %452 = bitcast float %335 to i32, !dbg !41 + %453 = bitcast float %336 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %449, i32 %450, i32 %451, i32 %452, i32 %453) #6, !dbg !41 + tail call void @llvm.nvvm.bar.warp.sync(i32 -1), !dbg !41 + %454 = shl nuw nsw i32 %11, 1, !dbg !41 + %455 = and i32 %454, 12, !dbg !41 + %456 = shl nuw nsw i32 %11, 4, !dbg !41 + %457 = and i32 %456, 16, !dbg !41 + %458 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %457, !dbg !41 + %459 = getelementptr inbounds nuw i8, ptr addrspace(3) %458, i32 %455, !dbg !41 + %460 = load i32, ptr addrspace(3) %459, align 4, !dbg !41 + %461 = and i32 %11, 24, !dbg !41 + %462 = icmp eq i32 %461, 0, !dbg !41 + %463 = and i1 %462, %332, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %460, ptr addrspace(1) %323, i1 %463) #6, !dbg !41 + ret void, !dbg !42 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.div.full(float, float) #3 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.bar.warp.sync(i32) #4 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #5 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.f(float) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="32" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) } +attributes #4 = { convergent nocallback nounwind } +attributes #5 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #6 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "l2norm_fwd_kernel", linkageName: "l2norm_fwd_kernel", scope: !1, file: !1, line: 69, type: !6, scopeLine: 69, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !10, !11, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 70, column: 24, scope: !5) +!14 = !DILocation(line: 70, column: 30, scope: !5) +!15 = !DILocation(line: 71, column: 16, scope: !5) +!16 = !DILocation(line: 71, column: 34, scope: !5) +!17 = !DILocation(line: 71, column: 21, scope: !5) +!18 = !DILocation(line: 73, column: 16, scope: !5) +!19 = !DILocation(line: 74, column: 30, scope: !5) +!20 = !DILocation(line: 75, column: 29, scope: !5) +!21 = !DILocation(line: 75, column: 14, scope: !5) +!22 = !DILocation(line: 75, column: 33, scope: !5) +!23 = !DILocation(line: 79, column: 18, scope: !5) +!24 = !DILocation(line: 79, column: 47, scope: !5) +!25 = !DILocation(line: 80, column: 38, scope: !5) +!26 = !DILocation(line: 263, column: 15, scope: !27, inlinedAt: !29) +!27 = distinct !DILexicalBlockFile(scope: !5, file: !28, discriminator: 0) +!28 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!29 = !DILocation(line: 293, column: 36, scope: !27, inlinedAt: !30) +!30 = !DILocation(line: 80, column: 43, scope: !31) +!31 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!32 = !DILocation(line: 80, column: 48, scope: !5) +!33 = !DILocation(line: 80, column: 25, scope: !5) +!34 = !DILocation(line: 77, column: 20, scope: !5) +!35 = !DILocation(line: 76, column: 14, scope: !5) +!36 = !DILocation(line: 76, column: 33, scope: !5) +!37 = !DILocation(line: 80, column: 17, scope: !5) +!38 = !DILocation(line: 81, column: 16, scope: !5) +!39 = !DILocation(line: 83, column: 25, scope: !5) +!40 = !DILocation(line: 83, column: 18, scope: !5) +!41 = !DILocation(line: 84, column: 21, scope: !5) +!42 = !DILocation(line: 84, column: 4, scope: !5) diff --git a/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/l2norm_fwd_kernel.ptx b/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/l2norm_fwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..3fc2508145d10f5bed3aa6f423050f6d381ce762 --- /dev/null +++ b/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/l2norm_fwd_kernel.ptx @@ -0,0 +1,642 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_fwd_kernel // -- Begin function l2norm_fwd_kernel +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; +.global .align 1 .b8 _$_str_$_2[17] = {95, 95, 67, 85, 68, 65, 95, 80, 82, 69, 67, 95, 83, 81, 82, 84}; + // @l2norm_fwd_kernel +.visible .entry l2norm_fwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_2, + .param .f32 l2norm_fwd_kernel_param_3, + .param .u32 l2norm_fwd_kernel_param_4, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_6 +) +.reqntid 32 +{ + .reg .pred %p<8>; + .reg .b16 %rs<33>; + .reg .b32 %r<196>; + .reg .b64 %rd<39>; + .loc 1 69 0 // l2norm.py:69:0 +$L__func_begin0: + .loc 1 69 0 // l2norm.py:69:0 + +// %bb.0: // %__nv_sqrtf.exit + ld.param.b64 %rd10, [l2norm_fwd_kernel_param_0]; + ld.param.b64 %rd11, [l2norm_fwd_kernel_param_1]; +$L__tmp0: + .loc 1 70 24 // l2norm.py:70:24 + mov.u32 %r40, %ctaid.x; + ld.param.b64 %rd12, [l2norm_fwd_kernel_param_2]; + .loc 1 71 16 // l2norm.py:71:16 + mul.wide.u32 %rd13, %r40, 8; + ld.param.b32 %r41, [l2norm_fwd_kernel_param_3]; + ld.param.s32 %rd14, [l2norm_fwd_kernel_param_4]; + .loc 1 71 34 // l2norm.py:71:34 + mov.u32 %r42, %tid.x; + and.b32 %r43, %r42, 16; + bfe.u32 %r44, %r42, 4, 1; + or.b32 %r45, %r44, 2; + or.b32 %r46, %r44, 4; + or.b32 %r47, %r44, 6; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd15, %r44; + cvt.u64.u32 %rd16, %r45; + cvt.u64.u32 %rd17, %r46; + cvt.u64.u32 %rd18, %r47; + or.b64 %rd19, %rd13, %rd15; + or.b64 %rd20, %rd13, %rd16; + or.b64 %rd21, %rd13, %rd17; + or.b64 %rd22, %rd13, %rd18; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p1, %rd19, %rd14; + setp.lt.s64 %p2, %rd20, %rd14; + setp.lt.s64 %p3, %rd21, %rd14; + setp.lt.s64 %p4, %rd22, %rd14; + .loc 1 74 30 // l2norm.py:74:30 + shl.b32 %r48, %r42, 3; + and.b32 %r49, %r48, 120; + .loc 1 75 14 // l2norm.py:75:14 + shl.b64 %rd23, %rd19, 8; + add.s64 %rd24, %rd10, %rd23; + shl.b64 %rd25, %rd20, 8; + add.s64 %rd26, %rd10, %rd25; + shl.b64 %rd27, %rd21, 8; + add.s64 %rd28, %rd10, %rd27; + shl.b64 %rd29, %rd22, 8; + add.s64 %rd30, %rd10, %rd29; + .loc 1 75 33 // l2norm.py:75:33 + mul.wide.u32 %rd31, %r49, 2; + add.s64 %rd1, %rd24, %rd31; + add.s64 %rd2, %rd26, %rd31; + add.s64 %rd3, %rd28, %rd31; + add.s64 %rd4, %rd30, %rd31; + mov.b32 %r5, 0; + .loc 1 79 18 // l2norm.py:79:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + mov.b32 {%rs5, %rs6}, %r3; + mov.b32 {%rs7, %rs8}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs9, %rs10}, %r6; + mov.b32 {%rs11, %rs12}, %r7; + mov.b32 {%rs13, %rs14}, %r8; + mov.b32 {%rs15, %rs16}, %r9; + // begin inline asm + mov.u32 %r10, %r5; + mov.u32 %r11, %r5; + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + @%p3 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs17, %rs18}, %r10; + mov.b32 {%rs19, %rs20}, %r11; + mov.b32 {%rs21, %rs22}, %r12; + mov.b32 {%rs23, %rs24}, %r13; + // begin inline asm + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + @%p4 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ]; + // end inline asm + mov.b32 {%rs25, %rs26}, %r14; + mov.b32 {%rs27, %rs28}, %r15; + mov.b32 {%rs29, %rs30}, %r16; + mov.b32 {%rs31, %rs32}, %r17; + .loc 1 79 47 // l2norm.py:79:47 + cvt.f32.bf16 %r50, %rs1; + cvt.f32.bf16 %r51, %rs2; + cvt.f32.bf16 %r52, %rs3; + cvt.f32.bf16 %r53, %rs4; + cvt.f32.bf16 %r54, %rs5; + cvt.f32.bf16 %r55, %rs6; + cvt.f32.bf16 %r56, %rs7; + cvt.f32.bf16 %r57, %rs8; + cvt.f32.bf16 %r58, %rs9; + cvt.f32.bf16 %r59, %rs10; + cvt.f32.bf16 %r60, %rs11; + cvt.f32.bf16 %r61, %rs12; + cvt.f32.bf16 %r62, %rs13; + cvt.f32.bf16 %r63, %rs14; + cvt.f32.bf16 %r64, %rs15; + cvt.f32.bf16 %r65, %rs16; + cvt.f32.bf16 %r66, %rs17; + cvt.f32.bf16 %r67, %rs18; + cvt.f32.bf16 %r68, %rs19; + cvt.f32.bf16 %r69, %rs20; + cvt.f32.bf16 %r70, %rs21; + cvt.f32.bf16 %r71, %rs22; + cvt.f32.bf16 %r72, %rs23; + cvt.f32.bf16 %r73, %rs24; + cvt.f32.bf16 %r74, %rs25; + cvt.f32.bf16 %r75, %rs26; + cvt.f32.bf16 %r76, %rs27; + cvt.f32.bf16 %r77, %rs28; + cvt.f32.bf16 %r78, %rs29; + cvt.f32.bf16 %r79, %rs30; + cvt.f32.bf16 %r80, %rs31; + cvt.f32.bf16 %r81, %rs32; + .loc 1 80 38 // l2norm.py:80:38 + mul.f32 %r82, %r51, %r51; + mul.f32 %r83, %r59, %r59; + mul.f32 %r84, %r67, %r67; + mul.f32 %r85, %r75, %r75; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + fma.rn.f32.bf16 %r86, %rs1, %rs1, %r82; + fma.rn.f32.bf16 %r87, %rs3, %rs3, %r86; + fma.rn.f32.bf16 %r88, %rs4, %rs4, %r87; + fma.rn.f32.bf16 %r89, %rs5, %rs5, %r88; + fma.rn.f32.bf16 %r90, %rs6, %rs6, %r89; + fma.rn.f32.bf16 %r91, %rs7, %rs7, %r90; + fma.rn.f32.bf16 %r92, %rs8, %rs8, %r91; + fma.rn.f32.bf16 %r93, %rs9, %rs9, %r83; + fma.rn.f32.bf16 %r94, %rs11, %rs11, %r93; + fma.rn.f32.bf16 %r95, %rs12, %rs12, %r94; + fma.rn.f32.bf16 %r96, %rs13, %rs13, %r95; + fma.rn.f32.bf16 %r97, %rs14, %rs14, %r96; + fma.rn.f32.bf16 %r98, %rs15, %rs15, %r97; + fma.rn.f32.bf16 %r99, %rs16, %rs16, %r98; + fma.rn.f32.bf16 %r100, %rs17, %rs17, %r84; + fma.rn.f32.bf16 %r101, %rs19, %rs19, %r100; + fma.rn.f32.bf16 %r102, %rs20, %rs20, %r101; + fma.rn.f32.bf16 %r103, %rs21, %rs21, %r102; + fma.rn.f32.bf16 %r104, %rs22, %rs22, %r103; + fma.rn.f32.bf16 %r105, %rs23, %rs23, %r104; + fma.rn.f32.bf16 %r106, %rs24, %rs24, %r105; + fma.rn.f32.bf16 %r107, %rs25, %rs25, %r85; + fma.rn.f32.bf16 %r108, %rs27, %rs27, %r107; + fma.rn.f32.bf16 %r109, %rs28, %rs28, %r108; + fma.rn.f32.bf16 %r110, %rs29, %rs29, %r109; + fma.rn.f32.bf16 %r111, %rs30, %rs30, %r110; + fma.rn.f32.bf16 %r112, %rs31, %rs31, %r111; + fma.rn.f32.bf16 %r113, %rs32, %rs32, %r112; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r114, %r92, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r115, %r92, %r114; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r116, %r115, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r117, %r115, %r116; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r118, %r117, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r119, %r117, %r118; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r120, %r119, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r121, %r119, %r120; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r122, %r99, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r123, %r99, %r122; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r124, %r123, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r125, %r123, %r124; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r126, %r125, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r127, %r125, %r126; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r128, %r127, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r129, %r127, %r128; +$L__tmp18: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r130, %r106, 8, 31, -1; +$L__tmp19: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r131, %r106, %r130; +$L__tmp20: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r132, %r131, 4, 31, -1; +$L__tmp21: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r133, %r131, %r132; +$L__tmp22: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r134, %r133, 2, 31, -1; +$L__tmp23: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r135, %r133, %r134; +$L__tmp24: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r136, %r135, 1, 31, -1; +$L__tmp25: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r137, %r135, %r136; +$L__tmp26: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r138, %r113, 8, 31, -1; +$L__tmp27: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r139, %r113, %r138; +$L__tmp28: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r140, %r139, 4, 31, -1; +$L__tmp29: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r141, %r139, %r140; +$L__tmp30: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r142, %r141, 2, 31, -1; +$L__tmp31: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r143, %r141, %r142; +$L__tmp32: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r144, %r143, 1, 31, -1; +$L__tmp33: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r145, %r143, %r144; +$L__tmp34: + .loc 1 80 48 // l2norm.py:80:48 + add.f32 %r146, %r41, %r121; + add.f32 %r147, %r41, %r129; + add.f32 %r148, %r41, %r137; + add.f32 %r149, %r41, %r145; + .loc 1 80 25 // l2norm.py:80:25 + sqrt.approx.ftz.f32 %r150, %r146; + sqrt.approx.ftz.f32 %r151, %r147; + sqrt.approx.ftz.f32 %r152, %r148; + sqrt.approx.ftz.f32 %r153, %r149; + .loc 1 71 34 // l2norm.py:71:34 + and.b32 %r154, %r42, 7; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd32, %r154; + or.b64 %rd33, %rd13, %rd32; + .loc 1 77 20 // l2norm.py:77:20 + shl.b64 %rd34, %rd33, 2; + add.s64 %rd9, %rd12, %rd34; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd35, %rd11, %rd29; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd8, %rd35, %rd31; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd36, %rd11, %rd27; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd7, %rd36, %rd31; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd37, %rd11, %rd25; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd6, %rd37, %rd31; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd38, %rd11, %rd23; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd5, %rd38, %rd31; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p6, %rd33, %rd14; + mov.b32 %r155, 0f3F800000; + .loc 1 80 17 // l2norm.py:80:17 + div.full.f32 %r35, %r155, %r150; + div.full.f32 %r36, %r155, %r151; + div.full.f32 %r37, %r155, %r152; + div.full.f32 %r38, %r155, %r153; + .loc 1 81 16 // l2norm.py:81:16 + mul.f32 %r156, %r35, %r50; + mul.f32 %r157, %r35, %r51; + mul.f32 %r158, %r35, %r52; + mul.f32 %r159, %r35, %r53; + mul.f32 %r160, %r35, %r54; + mul.f32 %r161, %r35, %r55; + mul.f32 %r162, %r35, %r56; + mul.f32 %r163, %r35, %r57; + mul.f32 %r164, %r36, %r58; + mul.f32 %r165, %r36, %r59; + mul.f32 %r166, %r36, %r60; + mul.f32 %r167, %r36, %r61; + mul.f32 %r168, %r36, %r62; + mul.f32 %r169, %r36, %r63; + mul.f32 %r170, %r36, %r64; + mul.f32 %r171, %r36, %r65; + mul.f32 %r172, %r37, %r66; + mul.f32 %r173, %r37, %r67; + mul.f32 %r174, %r37, %r68; + mul.f32 %r175, %r37, %r69; + mul.f32 %r176, %r37, %r70; + mul.f32 %r177, %r37, %r71; + mul.f32 %r178, %r37, %r72; + mul.f32 %r179, %r37, %r73; + mul.f32 %r180, %r38, %r74; + mul.f32 %r181, %r38, %r75; + mul.f32 %r182, %r38, %r76; + mul.f32 %r183, %r38, %r77; + mul.f32 %r184, %r38, %r78; + mul.f32 %r185, %r38, %r79; + mul.f32 %r186, %r38, %r80; + mul.f32 %r187, %r38, %r81; + .loc 1 83 18 // l2norm.py:83:18 + cvt.rn.bf16x2.f32 %r18, %r157, %r156; + cvt.rn.bf16x2.f32 %r19, %r159, %r158; + cvt.rn.bf16x2.f32 %r20, %r161, %r160; + cvt.rn.bf16x2.f32 %r21, %r163, %r162; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd5 + 0 ], { %r18, %r19, %r20, %r21 }; + // end inline asm + cvt.rn.bf16x2.f32 %r22, %r165, %r164; + cvt.rn.bf16x2.f32 %r23, %r167, %r166; + cvt.rn.bf16x2.f32 %r24, %r169, %r168; + cvt.rn.bf16x2.f32 %r25, %r171, %r170; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd6 + 0 ], { %r22, %r23, %r24, %r25 }; + // end inline asm + cvt.rn.bf16x2.f32 %r26, %r173, %r172; + cvt.rn.bf16x2.f32 %r27, %r175, %r174; + cvt.rn.bf16x2.f32 %r28, %r177, %r176; + cvt.rn.bf16x2.f32 %r29, %r179, %r178; + // begin inline asm + @%p3 st.global.v4.b32 [ %rd7 + 0 ], { %r26, %r27, %r28, %r29 }; + // end inline asm + cvt.rn.bf16x2.f32 %r30, %r181, %r180; + cvt.rn.bf16x2.f32 %r31, %r183, %r182; + cvt.rn.bf16x2.f32 %r32, %r185, %r184; + cvt.rn.bf16x2.f32 %r33, %r187, %r186; + // begin inline asm + @%p4 st.global.v4.b32 [ %rd8 + 0 ], { %r30, %r31, %r32, %r33 }; + // end inline asm + .loc 1 84 21 // l2norm.py:84:21 + mov.b32 %r188, global_smem; + add.s32 %r34, %r188, %r43; + // begin inline asm + st.shared.v4.b32 [ %r34 + 0 ], { %r35, %r36, %r37, %r38 }; + // end inline asm + bar.warp.sync -1; + shl.b32 %r189, %r42, 1; + and.b32 %r190, %r189, 12; + shl.b32 %r191, %r42, 4; + and.b32 %r192, %r191, 16; + add.s32 %r193, %r188, %r192; + add.s32 %r194, %r193, %r190; + ld.shared.b32 %r39, [%r194]; + and.b32 %r195, %r42, 24; + setp.eq.b32 %p7, %r195, 0; + and.pred %p5, %p7, %p6; + // begin inline asm + @%p5 st.global.b32 [ %rd9 + 0 ], { %r39 }; + // end inline asm + .loc 1 84 4 // l2norm.py:84:4 + ret; +$L__tmp35: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp34 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 80 // DW_AT_call_line +.b8 43 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp34 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/l2norm_fwd_kernel.source b/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/l2norm_fwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..c7c5a735c81381ccb0efd0d2a8f8d86240068e11 --- /dev/null +++ b/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/l2norm_fwd_kernel.source @@ -0,0 +1,172 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc38 = loc(unknown) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc45 = loc("x"(#loc)) +#loc46 = loc("y"(#loc)) +#loc47 = loc("rstd"(#loc)) +#loc48 = loc("eps"(#loc)) +#loc49 = loc("T"(#loc)) +#loc81 = loc("input"(#loc36)) +#loc82 = loc("a"(#loc41)) +#loc83 = loc("b"(#loc41)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc50) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc51) + %o_t = arith.constant 8 : i32 loc(#loc52) + %o_t_1 = arith.constant 8 : i64 loc(#loc52) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc52) + %o_t_3 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32> loc(#loc53) + %o_t_4 = arith.extsi %o_t_3 : tensor<8xi32> to tensor<8xi64> loc(#loc54) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<8xi64> loc(#loc54) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<8xi64> loc(#loc54) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc55) + %m_t = arith.extsi %T : i32 to i64 loc(#loc56) + %m_t_7 = tt.splat %m_t : i64 -> tensor<8xi64> loc(#loc56) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<8xi64> loc(#loc56) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<8xi1> -> tensor<8x1xi1> loc(#loc57) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc58) + %m_x_10 = arith.constant 128 : i32 loc(#loc59) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc59) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc59) + %m_x_13 = tt.broadcast %m_x : tensor<8x1xi1> -> tensor<8x128xi1> loc(#loc60) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<8x128xi1> loc(#loc60) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<8x128xi1> loc(#loc60) + %p_x = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc61) + %p_x_16 = arith.constant 128 : i32 loc(#loc62) + %p_x_17 = arith.constant 128 : i64 loc(#loc62) + %p_x_18 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc62) + %p_x_19 = arith.muli %p_x, %p_x_18 : tensor<8x1xi64> loc(#loc62) + %p_x_20 = tt.splat %x : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc63) + %p_x_21 = tt.addptr %p_x_20, %p_x_19 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc63) + %p_x_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc64) + %p_x_23 = tt.broadcast %p_x_21 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc65) + %p_x_24 = tt.broadcast %p_x_22 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc65) + %p_x_25 = tt.addptr %p_x_23, %p_x_24 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc65) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc66) + %p_y_26 = arith.constant 128 : i32 loc(#loc67) + %p_y_27 = arith.constant 128 : i64 loc(#loc67) + %p_y_28 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc67) + %p_y_29 = arith.muli %p_y, %p_y_28 : tensor<8x1xi64> loc(#loc67) + %p_y_30 = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc68) + %p_y_31 = tt.addptr %p_y_30, %p_y_29 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc68) + %p_y_32 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc69) + %p_y_33 = tt.broadcast %p_y_31 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc70) + %p_y_34 = tt.broadcast %p_y_32 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc70) + %p_y_35 = tt.addptr %p_y_33, %p_y_34 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc70) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr> loc(#loc71) + %p_rstd_36 = tt.addptr %p_rstd, %o_t_6 : tensor<8x!tt.ptr>, tensor<8xi64> loc(#loc71) + %b_x = arith.constant 0.000000e+00 : f32 loc(#loc72) + %b_x_37 = arith.constant dense<0.000000e+00> : tensor<8x128xf32> loc(#loc72) + %b_x_38 = arith.truncf %b_x_37 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc72) + %b_x_39 = tt.load %p_x_25, %m_x_15, %b_x_38 : tensor<8x128x!tt.ptr> loc(#loc72) + %b_x_40 = arith.extf %b_x_39 : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc73) + %b_rstd = arith.mulf %b_x_40, %b_x_40 : tensor<8x128xf32> loc(#loc74) + %b_rstd_41 = tt.call @triton.language.standard.sum__fp32S8_128S_c1_cFalse_cNone(%b_rstd) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc75) + %b_rstd_42 = tt.splat %eps : f32 -> tensor<8xf32> loc(#loc76) + %b_rstd_43 = arith.addf %b_rstd_41, %b_rstd_42 : tensor<8xf32> loc(#loc76) + %b_rstd_44 = math.sqrt %b_rstd_43 : tensor<8xf32> loc(#loc77) + %b_rstd_45 = arith.constant 1 : i32 loc(#loc78) + %b_rstd_46 = arith.constant 1.000000e+00 : f32 loc(#loc78) + %b_rstd_47 = arith.constant dense<1.000000e+00> : tensor<8xf32> loc(#loc78) + %b_rstd_48 = arith.divf %b_rstd_47, %b_rstd_44 : tensor<8xf32> loc(#loc78) + %b_y = tt.expand_dims %b_rstd_48 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc79) + %b_y_49 = tt.broadcast %b_y : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc80) + %b_y_50 = arith.mulf %b_x_40, %b_y_49 : tensor<8x128xf32> loc(#loc80) + %0 = arith.truncf %b_y_50 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc32) + tt.store %p_y_35, %0, %m_x_15 : tensor<8x128x!tt.ptr> loc(#loc33) + tt.store %p_rstd_36, %b_rstd_48, %m_t_8 : tensor<8x!tt.ptr> loc(#loc34) + tt.return loc(#loc35) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S8_128S_c1_cFalse_cNone(%input: tensor<8x128xf32> loc("input"(#loc36))) -> tensor<8xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc37) + tt.reduce.return %2 : f32 loc(#loc37) + }) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc37) + tt.return %0 : tensor<8xf32> loc(#loc39) + ^bb1: // no predecessors + %1 = ub.poison : tensor<8xf32> loc(#loc40) + tt.return %1 : tensor<8xf32> loc(#loc40) + } loc(#loc36) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc41)), %b: f32 loc("b"(#loc41))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc42) + tt.return %0 : f32 loc(#loc43) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc44) + tt.return %1 : f32 loc(#loc44) + } loc(#loc41) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:18) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:29) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:37) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc50 = loc("i_t"(#loc1)) +#loc51 = loc("i_t"(#loc2)) +#loc52 = loc("o_t"(#loc3)) +#loc53 = loc("o_t"(#loc4)) +#loc54 = loc("o_t"(#loc5)) +#loc55 = loc("o_d"(#loc6)) +#loc56 = loc("m_t"(#loc7)) +#loc57 = loc("m_x"(#loc8)) +#loc58 = loc("m_x"(#loc9)) +#loc59 = loc("m_x"(#loc10)) +#loc60 = loc("m_x"(#loc11)) +#loc61 = loc("p_x"(#loc12)) +#loc62 = loc("p_x"(#loc13)) +#loc63 = loc("p_x"(#loc14)) +#loc64 = loc("p_x"(#loc15)) +#loc65 = loc("p_x"(#loc16)) +#loc66 = loc("p_y"(#loc17)) +#loc67 = loc("p_y"(#loc18)) +#loc68 = loc("p_y"(#loc19)) +#loc69 = loc("p_y"(#loc20)) +#loc70 = loc("p_y"(#loc21)) +#loc71 = loc("p_rstd"(#loc22)) +#loc72 = loc("b_x"(#loc23)) +#loc73 = loc("b_x"(#loc24)) +#loc74 = loc("b_rstd"(#loc25)) +#loc75 = loc("b_rstd"(#loc26)) +#loc76 = loc("b_rstd"(#loc27)) +#loc77 = loc("b_rstd"(#loc28)) +#loc78 = loc("b_rstd"(#loc29)) +#loc79 = loc("b_y"(#loc30)) +#loc80 = loc("b_y"(#loc31)) diff --git a/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/l2norm_fwd_kernel.ttgir b/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/l2norm_fwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..1c4d2f1ae1e45290e174028273575be07fb7713b --- /dev/null +++ b/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/l2norm_fwd_kernel.ttgir @@ -0,0 +1,135 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [1, 1], order = [1, 0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [1], order = [0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc1 = loc(unknown) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc34 = loc("x"(#loc)) +#loc35 = loc("y"(#loc)) +#loc36 = loc("rstd"(#loc)) +#loc37 = loc("eps"(#loc)) +#loc38 = loc("T"(#loc)) +#loc59 = loc("b_rstd"(#loc23)) +#loc66 = loc(callsite(#loc1 at #loc59)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 1 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<1.000000e+00> : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<8x128xbf16, #blocked> loc(#loc1) + %c8_i64 = arith.constant 8 : i64 loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<8x1xi64, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc39) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc40) + %o_t = arith.muli %i_t_3, %c8_i64 : i64 loc(#loc41) + %o_t_4 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc42) + %o_t_5 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32, #blocked1> loc(#loc42) + %o_t_6 = arith.extsi %o_t_4 : tensor<8xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_7 = arith.extsi %o_t_5 : tensor<8xi32, #blocked1> to tensor<8xi64, #blocked1> loc(#loc43) + %o_t_8 = tt.splat %o_t : i64 -> tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_9 = tt.splat %o_t : i64 -> tensor<8xi64, #blocked1> loc(#loc43) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<8xi64, #blocked1> loc(#loc43) + %m_t = arith.extsi %T : i32 to i64 loc(#loc44) + %m_t_12 = tt.splat %m_t : i64 -> tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_13 = tt.splat %m_t : i64 -> tensor<8xi64, #blocked1> loc(#loc44) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<8xi64, #blocked1> loc(#loc44) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<8xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xi1, #blocked> loc(#loc45) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc46) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc46) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_1 : tensor<1x128xi32, #blocked> loc(#loc47) + %m_x_19 = tt.broadcast %m_x : tensor<8x1xi1, #blocked> -> tensor<8x128xi1, #blocked> loc(#loc48) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<8x128xi1, #blocked> loc(#loc48) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<8x128xi1, #blocked> loc(#loc48) + %p_x = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xi64, #blocked> loc(#loc49) + %p_x_22 = arith.muli %p_x, %cst_2 : tensor<8x1xi64, #blocked> loc(#loc50) + %p_x_23 = tt.splat %x : !tt.ptr -> tensor<8x1x!tt.ptr, #blocked> loc(#loc51) + %p_x_24 = tt.addptr %p_x_23, %p_x_22 : tensor<8x1x!tt.ptr, #blocked>, tensor<8x1xi64, #blocked> loc(#loc51) + %p_x_25 = tt.broadcast %p_x_24 : tensor<8x1x!tt.ptr, #blocked> -> tensor<8x128x!tt.ptr, #blocked> loc(#loc52) + %p_x_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<8x128xi32, #blocked> loc(#loc52) + %p_x_27 = tt.addptr %p_x_25, %p_x_26 : tensor<8x128x!tt.ptr, #blocked>, tensor<8x128xi32, #blocked> loc(#loc52) + %p_y = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr, #blocked> loc(#loc53) + %p_y_28 = tt.addptr %p_y, %p_x_22 : tensor<8x1x!tt.ptr, #blocked>, tensor<8x1xi64, #blocked> loc(#loc53) + %p_y_29 = tt.broadcast %p_y_28 : tensor<8x1x!tt.ptr, #blocked> -> tensor<8x128x!tt.ptr, #blocked> loc(#loc54) + %p_y_30 = tt.addptr %p_y_29, %p_x_26 : tensor<8x128x!tt.ptr, #blocked>, tensor<8x128xi32, #blocked> loc(#loc54) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr, #blocked1> loc(#loc55) + %p_rstd_31 = tt.addptr %p_rstd, %o_t_11 : tensor<8x!tt.ptr, #blocked1>, tensor<8xi64, #blocked1> loc(#loc55) + %b_x = tt.load %p_x_27, %m_x_21, %cst_0 : tensor<8x128x!tt.ptr, #blocked> loc(#loc56) + %b_x_32 = arith.extf %b_x : tensor<8x128xbf16, #blocked> to tensor<8x128xf32, #blocked> loc(#loc57) + %b_rstd = arith.mulf %b_x_32, %b_x_32 : tensor<8x128xf32, #blocked> loc(#loc58) + %b_rstd_33 = "tt.reduce"(%b_rstd) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_40: f32 loc(callsite(#loc1 at #loc59)), %b_rstd_41: f32 loc(callsite(#loc1 at #loc59))): + %b_rstd_42 = arith.addf %b_rstd_40, %b_rstd_41 : f32 loc(#loc67) + tt.reduce.return %b_rstd_42 : f32 loc(#loc65) + }) : (tensor<8x128xf32, #blocked>) -> tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc65) + %b_rstd_34 = tt.splat %eps : f32 -> tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_36 = math.sqrt %b_rstd_35 : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc61) + %b_rstd_37 = arith.divf %cst, %b_rstd_36 : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc62) + %b_y = tt.expand_dims %b_rstd_37 {axis = 1 : i32} : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xf32, #blocked> loc(#loc63) + %b_y_38 = tt.broadcast %b_y : tensor<8x1xf32, #blocked> -> tensor<8x128xf32, #blocked> loc(#loc64) + %b_y_39 = arith.mulf %b_x_32, %b_y_38 : tensor<8x128xf32, #blocked> loc(#loc64) + %0 = arith.truncf %b_y_39 : tensor<8x128xf32, #blocked> to tensor<8x128xbf16, #blocked> loc(#loc30) + tt.store %p_y_30, %0, %m_x_21 : tensor<8x128x!tt.ptr, #blocked> loc(#loc31) + %1 = ttg.convert_layout %b_rstd_37 : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8xf32, #blocked1> loc(#loc32) + tt.store %p_rstd_31, %1, %m_t_15 : tensor<8x!tt.ptr, #blocked1> loc(#loc32) + tt.return loc(#loc33) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc39 = loc("i_t"(#loc2)) +#loc40 = loc("i_t"(#loc3)) +#loc41 = loc("o_t"(#loc4)) +#loc42 = loc("o_t"(#loc5)) +#loc43 = loc("o_t"(#loc6)) +#loc44 = loc("m_t"(#loc7)) +#loc45 = loc("m_x"(#loc8)) +#loc46 = loc("m_x"(#loc9)) +#loc47 = loc("m_x"(#loc10)) +#loc48 = loc("m_x"(#loc11)) +#loc49 = loc("p_x"(#loc12)) +#loc50 = loc("p_x"(#loc13)) +#loc51 = loc("p_x"(#loc14)) +#loc52 = loc("p_x"(#loc15)) +#loc53 = loc("p_y"(#loc16)) +#loc54 = loc("p_y"(#loc17)) +#loc55 = loc("p_rstd"(#loc18)) +#loc56 = loc("b_x"(#loc19)) +#loc57 = loc("b_x"(#loc20)) +#loc58 = loc("b_rstd"(#loc21)) +#loc60 = loc("b_rstd"(#loc25)) +#loc61 = loc("b_rstd"(#loc26)) +#loc62 = loc("b_rstd"(#loc27)) +#loc63 = loc("b_y"(#loc28)) +#loc64 = loc("b_y"(#loc29)) +#loc65 = loc(callsite(#loc22 at #loc59)) +#loc67 = loc(callsite(#loc24 at #loc65)) diff --git a/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/l2norm_fwd_kernel.ttir b/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/l2norm_fwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..8bbf17da02489a0d195ed0bd020180f9e4ac2496 --- /dev/null +++ b/triton/gpu0/BU4B6D4H32AW6IKIREIJPGAX6FWKYLJMOPUMOP2QKJCX6BQGNATA/l2norm_fwd_kernel.ttir @@ -0,0 +1,128 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc3 = loc(unknown) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc35 = loc("x"(#loc)) +#loc36 = loc("y"(#loc)) +#loc37 = loc("rstd"(#loc)) +#loc38 = loc("eps"(#loc)) +#loc39 = loc("T"(#loc)) +#loc62 = loc("b_rstd"(#loc25)) +#loc68 = loc(callsite(#loc3 at #loc62)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %b_x = arith.constant dense<0.000000e+00> : tensor<8x128xbf16> loc(#loc40) + %b_rstd = arith.constant dense<1.000000e+00> : tensor<8xf32> loc(#loc41) + %cst = arith.constant dense<128> : tensor<8x1xi64> loc(#loc3) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc42) + %o_t = arith.constant 8 : i64 loc(#loc43) + %i_t = tt.get_program_id x : i32 loc(#loc44) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc45) + %o_t_1 = arith.muli %i_t_0, %o_t : i64 loc(#loc43) + %o_t_2 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32> loc(#loc46) + %o_t_3 = arith.extsi %o_t_2 : tensor<8xi32> to tensor<8xi64> loc(#loc47) + %o_t_4 = tt.splat %o_t_1 : i64 -> tensor<8xi64> loc(#loc47) + %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<8xi64> loc(#loc47) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc48) + %m_t = arith.extsi %T : i32 to i64 loc(#loc49) + %m_t_6 = tt.splat %m_t : i64 -> tensor<8xi64> loc(#loc49) + %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<8xi64> loc(#loc49) + %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<8xi1> -> tensor<8x1xi1> loc(#loc50) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc51) + %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc42) + %m_x_11 = tt.broadcast %m_x_8 : tensor<8x1xi1> -> tensor<8x128xi1> loc(#loc52) + %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<8x128xi1> loc(#loc52) + %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<8x128xi1> loc(#loc52) + %p_x = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc53) + %p_x_14 = arith.muli %p_x, %cst : tensor<8x1xi64> loc(#loc54) + %p_x_15 = tt.splat %x : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc55) + %p_x_16 = tt.addptr %p_x_15, %p_x_14 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc55) + %p_x_17 = tt.broadcast %p_x_16 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc56) + %p_x_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc56) + %p_x_19 = tt.addptr %p_x_17, %p_x_18 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc56) + %p_y = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc57) + %p_y_20 = tt.addptr %p_y, %p_x_14 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc57) + %p_y_21 = tt.broadcast %p_y_20 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc58) + %p_y_22 = tt.addptr %p_y_21, %p_x_18 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr> loc(#loc59) + %p_rstd_23 = tt.addptr %p_rstd, %o_t_5 : tensor<8x!tt.ptr>, tensor<8xi64> loc(#loc59) + %b_x_24 = tt.load %p_x_19, %m_x_13, %b_x : tensor<8x128x!tt.ptr> loc(#loc40) + %b_x_25 = arith.extf %b_x_24 : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc60) + %b_rstd_26 = arith.mulf %b_x_25, %b_x_25 : tensor<8x128xf32> loc(#loc61) + %b_rstd_27 = "tt.reduce"(%b_rstd_26) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_34: f32 loc(callsite(#loc3 at #loc62)), %b_rstd_35: f32 loc(callsite(#loc3 at #loc62))): + %b_rstd_36 = arith.addf %b_rstd_34, %b_rstd_35 : f32 loc(#loc69) + tt.reduce.return %b_rstd_36 : f32 loc(#loc67) + }) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc67) + %b_rstd_28 = tt.splat %eps : f32 -> tensor<8xf32> loc(#loc63) + %b_rstd_29 = arith.addf %b_rstd_27, %b_rstd_28 : tensor<8xf32> loc(#loc63) + %b_rstd_30 = math.sqrt %b_rstd_29 : tensor<8xf32> loc(#loc64) + %b_rstd_31 = arith.divf %b_rstd, %b_rstd_30 : tensor<8xf32> loc(#loc41) + %b_y = tt.expand_dims %b_rstd_31 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc65) + %b_y_32 = tt.broadcast %b_y : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc66) + %b_y_33 = arith.mulf %b_x_25, %b_y_32 : tensor<8x128xf32> loc(#loc66) + %0 = arith.truncf %b_y_33 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc31) + tt.store %p_y_22, %0, %m_x_13 : tensor<8x128x!tt.ptr> loc(#loc32) + tt.store %p_rstd_23, %b_rstd_31, %m_t_7 : tensor<8x!tt.ptr> loc(#loc33) + tt.return loc(#loc34) + } loc(#loc) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc40 = loc("b_x"(#loc1)) +#loc41 = loc("b_rstd"(#loc2)) +#loc42 = loc("m_x"(#loc4)) +#loc43 = loc("o_t"(#loc5)) +#loc44 = loc("i_t"(#loc6)) +#loc45 = loc("i_t"(#loc7)) +#loc46 = loc("o_t"(#loc8)) +#loc47 = loc("o_t"(#loc9)) +#loc48 = loc("o_d"(#loc10)) +#loc49 = loc("m_t"(#loc11)) +#loc50 = loc("m_x"(#loc12)) +#loc51 = loc("m_x"(#loc13)) +#loc52 = loc("m_x"(#loc14)) +#loc53 = loc("p_x"(#loc15)) +#loc54 = loc("p_x"(#loc16)) +#loc55 = loc("p_x"(#loc17)) +#loc56 = loc("p_x"(#loc18)) +#loc57 = loc("p_y"(#loc19)) +#loc58 = loc("p_y"(#loc20)) +#loc59 = loc("p_rstd"(#loc21)) +#loc60 = loc("b_x"(#loc22)) +#loc61 = loc("b_rstd"(#loc23)) +#loc63 = loc("b_rstd"(#loc27)) +#loc64 = loc("b_rstd"(#loc28)) +#loc65 = loc("b_y"(#loc29)) +#loc66 = loc("b_y"(#loc30)) +#loc67 = loc(callsite(#loc24 at #loc62)) +#loc69 = loc(callsite(#loc26 at #loc67)) diff --git a/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/__grp__l2norm_fwd_kernel.json b/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/__grp__l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..6e3f6e59479f71a6681b891d274d86f0900eab9e --- /dev/null +++ b/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/__grp__l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_fwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.source", "l2norm_fwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.ttir", "l2norm_fwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.ttgir", "l2norm_fwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.llir", "l2norm_fwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.ptx", "l2norm_fwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.cubin", "l2norm_fwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.cubin b/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..c2bbf30463f772f88a1e53baca6effdc9897e60d --- /dev/null +++ b/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.cubin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ab59f85efd0fac634f2920e68426f45435a7a7c70c23e4d36ab45badbe8b6b2e +size 334184 diff --git a/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.json b/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..1cde75bff018c62247f2533dbd207c6fd4ed3e05 --- /dev/null +++ b/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"hash": "0d6978a4b75d98f9df38c5f391aa65f8d7a98e2e0dd999104c57ef6b77caa72e", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 1, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 256, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_fwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.llir b/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..008bb885fb1f9b602483fdc6754177d8d0e1c302 --- /dev/null +++ b/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.llir @@ -0,0 +1,8142 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 +@.str.2 = private unnamed_addr constant [17 x i8] c"__CUDA_PREC_SQRT\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_fwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, float %3, i32 %4, ptr addrspace(1) readnone captures(none) %5, ptr addrspace(1) readnone captures(none) %6) local_unnamed_addr #0 !dbg !5 { + %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %9 = zext nneg i32 %8 to i64, !dbg !14 + %10 = shl nuw nsw i64 %9, 7, !dbg !15 + %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !16 + %12 = and i32 %11, 16, !dbg !16 + %.lobit = lshr exact i32 %12, 4, !dbg !16 + %13 = or disjoint i32 %.lobit, 2, !dbg !16 + %14 = insertelement <2 x i32> poison, i32 %.lobit, i64 0, !dbg !16 + %15 = insertelement <4 x i32> poison, i32 %.lobit, i64 0, !dbg !16 + %16 = insertelement <8 x i32> poison, i32 %.lobit, i64 0, !dbg !16 + %17 = insertelement <16 x i32> poison, i32 %.lobit, i64 0, !dbg !16 + %18 = insertelement <32 x i32> poison, i32 %.lobit, i64 0, !dbg !16 + %19 = insertelement <64 x i32> poison, i32 %13, i64 62, !dbg !17 + %20 = insertelement <64 x i32> %19, i32 %.lobit, i64 63, !dbg !17 + %21 = shufflevector <32 x i32> %18, <32 x i32> poison, <64 x i32> , !dbg !17 + %22 = or disjoint <64 x i32> %21, , !dbg !17 + %23 = shufflevector <64 x i32> %22, <64 x i32> %20, <64 x i32> , !dbg !17 + %24 = shufflevector <16 x i32> %17, <16 x i32> poison, <64 x i32> , !dbg !17 + %25 = or disjoint <64 x i32> %24, , !dbg !17 + %26 = shufflevector <64 x i32> %23, <64 x i32> %25, <64 x i32> , !dbg !17 + %27 = shufflevector <8 x i32> %16, <8 x i32> poison, <64 x i32> , !dbg !17 + %28 = or disjoint <64 x i32> %27, , !dbg !17 + %29 = shufflevector <64 x i32> %26, <64 x i32> %28, <64 x i32> , !dbg !17 + %30 = shufflevector <4 x i32> %15, <4 x i32> poison, <64 x i32> , !dbg !17 + %31 = or disjoint <64 x i32> %30, , !dbg !17 + %32 = shufflevector <64 x i32> %29, <64 x i32> %31, <64 x i32> , !dbg !17 + %33 = shufflevector <2 x i32> %14, <2 x i32> poison, <64 x i32> , !dbg !17 + %34 = or disjoint <64 x i32> %33, , !dbg !17 + %35 = shufflevector <64 x i32> %32, <64 x i32> %34, <64 x i32> , !dbg !17 + %36 = zext <64 x i32> %35 to <64 x i64>, !dbg !17 + %37 = insertelement <64 x i64> poison, i64 %10, i64 0, !dbg !17 + %38 = shufflevector <64 x i64> %37, <64 x i64> poison, <64 x i32> zeroinitializer, !dbg !17 + %39 = or disjoint <64 x i64> %38, %36, !dbg !17 + %40 = insertelement <64 x i32> poison, i32 %4, i64 0, !dbg !18 + %41 = shufflevector <64 x i32> %40, <64 x i32> poison, <64 x i32> zeroinitializer, !dbg !18 + %42 = sext <64 x i32> %41 to <64 x i64>, !dbg !18 + %43 = icmp slt <64 x i64> %39, %42, !dbg !18 + %44 = and i32 %11, 15, !dbg !19 + %45 = shl nuw nsw i32 %44, 3, !dbg !19 + %46 = extractelement <64 x i64> %39, i64 63, !dbg !20 + %47 = shl nuw nsw i64 %46, 7, !dbg !20 + %48 = extractelement <64 x i64> %39, i64 62, !dbg !20 + %49 = shl nuw nsw i64 %48, 7, !dbg !20 + %50 = extractelement <64 x i64> %39, i64 61, !dbg !20 + %51 = shl nuw nsw i64 %50, 7, !dbg !20 + %52 = extractelement <64 x i64> %39, i64 60, !dbg !20 + %53 = shl nuw nsw i64 %52, 7, !dbg !20 + %54 = extractelement <64 x i64> %39, i64 59, !dbg !20 + %55 = shl nuw nsw i64 %54, 7, !dbg !20 + %56 = extractelement <64 x i64> %39, i64 58, !dbg !20 + %57 = shl nuw nsw i64 %56, 7, !dbg !20 + %58 = extractelement <64 x i64> %39, i64 57, !dbg !20 + %59 = shl nuw nsw i64 %58, 7, !dbg !20 + %60 = extractelement <64 x i64> %39, i64 56, !dbg !20 + %61 = shl nuw nsw i64 %60, 7, !dbg !20 + %62 = extractelement <64 x i64> %39, i64 55, !dbg !20 + %63 = shl nuw nsw i64 %62, 7, !dbg !20 + %64 = extractelement <64 x i64> %39, i64 54, !dbg !20 + %65 = shl nuw nsw i64 %64, 7, !dbg !20 + %66 = extractelement <64 x i64> %39, i64 53, !dbg !20 + %67 = shl nuw nsw i64 %66, 7, !dbg !20 + %68 = extractelement <64 x i64> %39, i64 52, !dbg !20 + %69 = shl nuw nsw i64 %68, 7, !dbg !20 + %70 = extractelement <64 x i64> %39, i64 51, !dbg !20 + %71 = shl nuw nsw i64 %70, 7, !dbg !20 + %72 = extractelement <64 x i64> %39, i64 50, !dbg !20 + %73 = shl nuw nsw i64 %72, 7, !dbg !20 + %74 = extractelement <64 x i64> %39, i64 49, !dbg !20 + %75 = shl nuw nsw i64 %74, 7, !dbg !20 + %76 = extractelement <64 x i64> %39, i64 48, !dbg !20 + %77 = shl nuw nsw i64 %76, 7, !dbg !20 + %78 = extractelement <64 x i64> %39, i64 47, !dbg !20 + %79 = shl nuw nsw i64 %78, 7, !dbg !20 + %80 = extractelement <64 x i64> %39, i64 46, !dbg !20 + %81 = shl nuw nsw i64 %80, 7, !dbg !20 + %82 = extractelement <64 x i64> %39, i64 45, !dbg !20 + %83 = shl nuw nsw i64 %82, 7, !dbg !20 + %84 = extractelement <64 x i64> %39, i64 44, !dbg !20 + %85 = shl nuw nsw i64 %84, 7, !dbg !20 + %86 = extractelement <64 x i64> %39, i64 43, !dbg !20 + %87 = shl nuw nsw i64 %86, 7, !dbg !20 + %88 = extractelement <64 x i64> %39, i64 42, !dbg !20 + %89 = shl nuw nsw i64 %88, 7, !dbg !20 + %90 = extractelement <64 x i64> %39, i64 41, !dbg !20 + %91 = shl nuw nsw i64 %90, 7, !dbg !20 + %92 = extractelement <64 x i64> %39, i64 40, !dbg !20 + %93 = shl nuw nsw i64 %92, 7, !dbg !20 + %94 = extractelement <64 x i64> %39, i64 39, !dbg !20 + %95 = shl nuw nsw i64 %94, 7, !dbg !20 + %96 = extractelement <64 x i64> %39, i64 38, !dbg !20 + %97 = shl nuw nsw i64 %96, 7, !dbg !20 + %98 = extractelement <64 x i64> %39, i64 37, !dbg !20 + %99 = shl nuw nsw i64 %98, 7, !dbg !20 + %100 = extractelement <64 x i64> %39, i64 36, !dbg !20 + %101 = shl nuw nsw i64 %100, 7, !dbg !20 + %102 = extractelement <64 x i64> %39, i64 35, !dbg !20 + %103 = shl nuw nsw i64 %102, 7, !dbg !20 + %104 = extractelement <64 x i64> %39, i64 34, !dbg !20 + %105 = shl nuw nsw i64 %104, 7, !dbg !20 + %106 = extractelement <64 x i64> %39, i64 33, !dbg !20 + %107 = shl nuw nsw i64 %106, 7, !dbg !20 + %108 = extractelement <64 x i64> %39, i64 32, !dbg !20 + %109 = shl nuw nsw i64 %108, 7, !dbg !20 + %110 = extractelement <64 x i64> %39, i64 31, !dbg !20 + %111 = shl nuw nsw i64 %110, 7, !dbg !20 + %112 = extractelement <64 x i64> %39, i64 30, !dbg !20 + %113 = shl nuw nsw i64 %112, 7, !dbg !20 + %114 = extractelement <64 x i64> %39, i64 29, !dbg !20 + %115 = shl nuw nsw i64 %114, 7, !dbg !20 + %116 = extractelement <64 x i64> %39, i64 28, !dbg !20 + %117 = shl nuw nsw i64 %116, 7, !dbg !20 + %118 = extractelement <64 x i64> %39, i64 27, !dbg !20 + %119 = shl nuw nsw i64 %118, 7, !dbg !20 + %120 = extractelement <64 x i64> %39, i64 26, !dbg !20 + %121 = shl nuw nsw i64 %120, 7, !dbg !20 + %122 = extractelement <64 x i64> %39, i64 25, !dbg !20 + %123 = shl nuw nsw i64 %122, 7, !dbg !20 + %124 = extractelement <64 x i64> %39, i64 24, !dbg !20 + %125 = shl nuw nsw i64 %124, 7, !dbg !20 + %126 = extractelement <64 x i64> %39, i64 23, !dbg !20 + %127 = shl nuw nsw i64 %126, 7, !dbg !20 + %128 = extractelement <64 x i64> %39, i64 22, !dbg !20 + %129 = shl nuw nsw i64 %128, 7, !dbg !20 + %130 = extractelement <64 x i64> %39, i64 21, !dbg !20 + %131 = shl nuw nsw i64 %130, 7, !dbg !20 + %132 = extractelement <64 x i64> %39, i64 20, !dbg !20 + %133 = shl nuw nsw i64 %132, 7, !dbg !20 + %134 = extractelement <64 x i64> %39, i64 19, !dbg !20 + %135 = shl nuw nsw i64 %134, 7, !dbg !20 + %136 = extractelement <64 x i64> %39, i64 18, !dbg !20 + %137 = shl nuw nsw i64 %136, 7, !dbg !20 + %138 = extractelement <64 x i64> %39, i64 17, !dbg !20 + %139 = shl nuw nsw i64 %138, 7, !dbg !20 + %140 = extractelement <64 x i64> %39, i64 16, !dbg !20 + %141 = shl nuw nsw i64 %140, 7, !dbg !20 + %142 = extractelement <64 x i64> %39, i64 15, !dbg !20 + %143 = shl nuw nsw i64 %142, 7, !dbg !20 + %144 = extractelement <64 x i64> %39, i64 14, !dbg !20 + %145 = shl nuw nsw i64 %144, 7, !dbg !20 + %146 = extractelement <64 x i64> %39, i64 13, !dbg !20 + %147 = shl nuw nsw i64 %146, 7, !dbg !20 + %148 = extractelement <64 x i64> %39, i64 12, !dbg !20 + %149 = shl nuw nsw i64 %148, 7, !dbg !20 + %150 = extractelement <64 x i64> %39, i64 11, !dbg !20 + %151 = shl nuw nsw i64 %150, 7, !dbg !20 + %152 = extractelement <64 x i64> %39, i64 10, !dbg !20 + %153 = shl nuw nsw i64 %152, 7, !dbg !20 + %154 = extractelement <64 x i64> %39, i64 9, !dbg !20 + %155 = shl nuw nsw i64 %154, 7, !dbg !20 + %156 = extractelement <64 x i64> %39, i64 8, !dbg !20 + %157 = shl nuw nsw i64 %156, 7, !dbg !20 + %158 = extractelement <64 x i64> %39, i64 7, !dbg !20 + %159 = shl nuw nsw i64 %158, 7, !dbg !20 + %160 = extractelement <64 x i64> %39, i64 6, !dbg !20 + %161 = shl nuw nsw i64 %160, 7, !dbg !20 + %162 = extractelement <64 x i64> %39, i64 5, !dbg !20 + %163 = shl nuw nsw i64 %162, 7, !dbg !20 + %164 = extractelement <64 x i64> %39, i64 4, !dbg !20 + %165 = shl nuw nsw i64 %164, 7, !dbg !20 + %166 = extractelement <64 x i64> %39, i64 3, !dbg !20 + %167 = shl nuw nsw i64 %166, 7, !dbg !20 + %168 = extractelement <64 x i64> %39, i64 2, !dbg !20 + %169 = shl nuw nsw i64 %168, 7, !dbg !20 + %170 = extractelement <64 x i64> %39, i64 1, !dbg !20 + %171 = shl nuw nsw i64 %170, 7, !dbg !20 + %172 = extractelement <64 x i64> %39, i64 0, !dbg !20 + %173 = shl nuw nsw i64 %172, 7, !dbg !20 + %174 = getelementptr bfloat, ptr addrspace(1) %0, i64 %47, !dbg !21 + %175 = getelementptr bfloat, ptr addrspace(1) %0, i64 %49, !dbg !21 + %176 = getelementptr bfloat, ptr addrspace(1) %0, i64 %51, !dbg !21 + %177 = getelementptr bfloat, ptr addrspace(1) %0, i64 %53, !dbg !21 + %178 = getelementptr bfloat, ptr addrspace(1) %0, i64 %55, !dbg !21 + %179 = getelementptr bfloat, ptr addrspace(1) %0, i64 %57, !dbg !21 + %180 = getelementptr bfloat, ptr addrspace(1) %0, i64 %59, !dbg !21 + %181 = getelementptr bfloat, ptr addrspace(1) %0, i64 %61, !dbg !21 + %182 = getelementptr bfloat, ptr addrspace(1) %0, i64 %63, !dbg !21 + %183 = getelementptr bfloat, ptr addrspace(1) %0, i64 %65, !dbg !21 + %184 = getelementptr bfloat, ptr addrspace(1) %0, i64 %67, !dbg !21 + %185 = getelementptr bfloat, ptr addrspace(1) %0, i64 %69, !dbg !21 + %186 = getelementptr bfloat, ptr addrspace(1) %0, i64 %71, !dbg !21 + %187 = getelementptr bfloat, ptr addrspace(1) %0, i64 %73, !dbg !21 + %188 = getelementptr bfloat, ptr addrspace(1) %0, i64 %75, !dbg !21 + %189 = getelementptr bfloat, ptr addrspace(1) %0, i64 %77, !dbg !21 + %190 = getelementptr bfloat, ptr addrspace(1) %0, i64 %79, !dbg !21 + %191 = getelementptr bfloat, ptr addrspace(1) %0, i64 %81, !dbg !21 + %192 = getelementptr bfloat, ptr addrspace(1) %0, i64 %83, !dbg !21 + %193 = getelementptr bfloat, ptr addrspace(1) %0, i64 %85, !dbg !21 + %194 = getelementptr bfloat, ptr addrspace(1) %0, i64 %87, !dbg !21 + %195 = getelementptr bfloat, ptr addrspace(1) %0, i64 %89, !dbg !21 + %196 = getelementptr bfloat, ptr addrspace(1) %0, i64 %91, !dbg !21 + %197 = getelementptr bfloat, ptr addrspace(1) %0, i64 %93, !dbg !21 + %198 = getelementptr bfloat, ptr addrspace(1) %0, i64 %95, !dbg !21 + %199 = getelementptr bfloat, ptr addrspace(1) %0, i64 %97, !dbg !21 + %200 = getelementptr bfloat, ptr addrspace(1) %0, i64 %99, !dbg !21 + %201 = getelementptr bfloat, ptr addrspace(1) %0, i64 %101, !dbg !21 + %202 = getelementptr bfloat, ptr addrspace(1) %0, i64 %103, !dbg !21 + %203 = getelementptr bfloat, ptr addrspace(1) %0, i64 %105, !dbg !21 + %204 = getelementptr bfloat, ptr addrspace(1) %0, i64 %107, !dbg !21 + %205 = getelementptr bfloat, ptr addrspace(1) %0, i64 %109, !dbg !21 + %206 = getelementptr bfloat, ptr addrspace(1) %0, i64 %111, !dbg !21 + %207 = getelementptr bfloat, ptr addrspace(1) %0, i64 %113, !dbg !21 + %208 = getelementptr bfloat, ptr addrspace(1) %0, i64 %115, !dbg !21 + %209 = getelementptr bfloat, ptr addrspace(1) %0, i64 %117, !dbg !21 + %210 = getelementptr bfloat, ptr addrspace(1) %0, i64 %119, !dbg !21 + %211 = getelementptr bfloat, ptr addrspace(1) %0, i64 %121, !dbg !21 + %212 = getelementptr bfloat, ptr addrspace(1) %0, i64 %123, !dbg !21 + %213 = getelementptr bfloat, ptr addrspace(1) %0, i64 %125, !dbg !21 + %214 = getelementptr bfloat, ptr addrspace(1) %0, i64 %127, !dbg !21 + %215 = getelementptr bfloat, ptr addrspace(1) %0, i64 %129, !dbg !21 + %216 = getelementptr bfloat, ptr addrspace(1) %0, i64 %131, !dbg !21 + %217 = getelementptr bfloat, ptr addrspace(1) %0, i64 %133, !dbg !21 + %218 = getelementptr bfloat, ptr addrspace(1) %0, i64 %135, !dbg !21 + %219 = getelementptr bfloat, ptr addrspace(1) %0, i64 %137, !dbg !21 + %220 = getelementptr bfloat, ptr addrspace(1) %0, i64 %139, !dbg !21 + %221 = getelementptr bfloat, ptr addrspace(1) %0, i64 %141, !dbg !21 + %222 = getelementptr bfloat, ptr addrspace(1) %0, i64 %143, !dbg !21 + %223 = getelementptr bfloat, ptr addrspace(1) %0, i64 %145, !dbg !21 + %224 = getelementptr bfloat, ptr addrspace(1) %0, i64 %147, !dbg !21 + %225 = getelementptr bfloat, ptr addrspace(1) %0, i64 %149, !dbg !21 + %226 = getelementptr bfloat, ptr addrspace(1) %0, i64 %151, !dbg !21 + %227 = getelementptr bfloat, ptr addrspace(1) %0, i64 %153, !dbg !21 + %228 = getelementptr bfloat, ptr addrspace(1) %0, i64 %155, !dbg !21 + %229 = getelementptr bfloat, ptr addrspace(1) %0, i64 %157, !dbg !21 + %230 = getelementptr bfloat, ptr addrspace(1) %0, i64 %159, !dbg !21 + %231 = getelementptr bfloat, ptr addrspace(1) %0, i64 %161, !dbg !21 + %232 = getelementptr bfloat, ptr addrspace(1) %0, i64 %163, !dbg !21 + %233 = getelementptr bfloat, ptr addrspace(1) %0, i64 %165, !dbg !21 + %234 = getelementptr bfloat, ptr addrspace(1) %0, i64 %167, !dbg !21 + %235 = getelementptr bfloat, ptr addrspace(1) %0, i64 %169, !dbg !21 + %236 = getelementptr bfloat, ptr addrspace(1) %0, i64 %171, !dbg !21 + %237 = getelementptr bfloat, ptr addrspace(1) %0, i64 %173, !dbg !21 + %238 = zext nneg i32 %45 to i64, !dbg !22 + %239 = getelementptr bfloat, ptr addrspace(1) %174, i64 %238, !dbg !22 + %240 = getelementptr bfloat, ptr addrspace(1) %175, i64 %238, !dbg !22 + %241 = getelementptr bfloat, ptr addrspace(1) %176, i64 %238, !dbg !22 + %242 = getelementptr bfloat, ptr addrspace(1) %177, i64 %238, !dbg !22 + %243 = getelementptr bfloat, ptr addrspace(1) %178, i64 %238, !dbg !22 + %244 = getelementptr bfloat, ptr addrspace(1) %179, i64 %238, !dbg !22 + %245 = getelementptr bfloat, ptr addrspace(1) %180, i64 %238, !dbg !22 + %246 = getelementptr bfloat, ptr addrspace(1) %181, i64 %238, !dbg !22 + %247 = getelementptr bfloat, ptr addrspace(1) %182, i64 %238, !dbg !22 + %248 = getelementptr bfloat, ptr addrspace(1) %183, i64 %238, !dbg !22 + %249 = getelementptr bfloat, ptr addrspace(1) %184, i64 %238, !dbg !22 + %250 = getelementptr bfloat, ptr addrspace(1) %185, i64 %238, !dbg !22 + %251 = getelementptr bfloat, ptr addrspace(1) %186, i64 %238, !dbg !22 + %252 = getelementptr bfloat, ptr addrspace(1) %187, i64 %238, !dbg !22 + %253 = getelementptr bfloat, ptr addrspace(1) %188, i64 %238, !dbg !22 + %254 = getelementptr bfloat, ptr addrspace(1) %189, i64 %238, !dbg !22 + %255 = getelementptr bfloat, ptr addrspace(1) %190, i64 %238, !dbg !22 + %256 = getelementptr bfloat, ptr addrspace(1) %191, i64 %238, !dbg !22 + %257 = getelementptr bfloat, ptr addrspace(1) %192, i64 %238, !dbg !22 + %258 = getelementptr bfloat, ptr addrspace(1) %193, i64 %238, !dbg !22 + %259 = getelementptr bfloat, ptr addrspace(1) %194, i64 %238, !dbg !22 + %260 = getelementptr bfloat, ptr addrspace(1) %195, i64 %238, !dbg !22 + %261 = getelementptr bfloat, ptr addrspace(1) %196, i64 %238, !dbg !22 + %262 = getelementptr bfloat, ptr addrspace(1) %197, i64 %238, !dbg !22 + %263 = getelementptr bfloat, ptr addrspace(1) %198, i64 %238, !dbg !22 + %264 = getelementptr bfloat, ptr addrspace(1) %199, i64 %238, !dbg !22 + %265 = getelementptr bfloat, ptr addrspace(1) %200, i64 %238, !dbg !22 + %266 = getelementptr bfloat, ptr addrspace(1) %201, i64 %238, !dbg !22 + %267 = getelementptr bfloat, ptr addrspace(1) %202, i64 %238, !dbg !22 + %268 = getelementptr bfloat, ptr addrspace(1) %203, i64 %238, !dbg !22 + %269 = getelementptr bfloat, ptr addrspace(1) %204, i64 %238, !dbg !22 + %270 = getelementptr bfloat, ptr addrspace(1) %205, i64 %238, !dbg !22 + %271 = getelementptr bfloat, ptr addrspace(1) %206, i64 %238, !dbg !22 + %272 = getelementptr bfloat, ptr addrspace(1) %207, i64 %238, !dbg !22 + %273 = getelementptr bfloat, ptr addrspace(1) %208, i64 %238, !dbg !22 + %274 = getelementptr bfloat, ptr addrspace(1) %209, i64 %238, !dbg !22 + %275 = getelementptr bfloat, ptr addrspace(1) %210, i64 %238, !dbg !22 + %276 = getelementptr bfloat, ptr addrspace(1) %211, i64 %238, !dbg !22 + %277 = getelementptr bfloat, ptr addrspace(1) %212, i64 %238, !dbg !22 + %278 = getelementptr bfloat, ptr addrspace(1) %213, i64 %238, !dbg !22 + %279 = getelementptr bfloat, ptr addrspace(1) %214, i64 %238, !dbg !22 + %280 = getelementptr bfloat, ptr addrspace(1) %215, i64 %238, !dbg !22 + %281 = getelementptr bfloat, ptr addrspace(1) %216, i64 %238, !dbg !22 + %282 = getelementptr bfloat, ptr addrspace(1) %217, i64 %238, !dbg !22 + %283 = getelementptr bfloat, ptr addrspace(1) %218, i64 %238, !dbg !22 + %284 = getelementptr bfloat, ptr addrspace(1) %219, i64 %238, !dbg !22 + %285 = getelementptr bfloat, ptr addrspace(1) %220, i64 %238, !dbg !22 + %286 = getelementptr bfloat, ptr addrspace(1) %221, i64 %238, !dbg !22 + %287 = getelementptr bfloat, ptr addrspace(1) %222, i64 %238, !dbg !22 + %288 = getelementptr bfloat, ptr addrspace(1) %223, i64 %238, !dbg !22 + %289 = getelementptr bfloat, ptr addrspace(1) %224, i64 %238, !dbg !22 + %290 = getelementptr bfloat, ptr addrspace(1) %225, i64 %238, !dbg !22 + %291 = getelementptr bfloat, ptr addrspace(1) %226, i64 %238, !dbg !22 + %292 = getelementptr bfloat, ptr addrspace(1) %227, i64 %238, !dbg !22 + %293 = getelementptr bfloat, ptr addrspace(1) %228, i64 %238, !dbg !22 + %294 = getelementptr bfloat, ptr addrspace(1) %229, i64 %238, !dbg !22 + %295 = getelementptr bfloat, ptr addrspace(1) %230, i64 %238, !dbg !22 + %296 = getelementptr bfloat, ptr addrspace(1) %231, i64 %238, !dbg !22 + %297 = getelementptr bfloat, ptr addrspace(1) %232, i64 %238, !dbg !22 + %298 = getelementptr bfloat, ptr addrspace(1) %233, i64 %238, !dbg !22 + %299 = getelementptr bfloat, ptr addrspace(1) %234, i64 %238, !dbg !22 + %300 = getelementptr bfloat, ptr addrspace(1) %235, i64 %238, !dbg !22 + %301 = getelementptr bfloat, ptr addrspace(1) %236, i64 %238, !dbg !22 + %302 = getelementptr bfloat, ptr addrspace(1) %237, i64 %238, !dbg !22 + %303 = extractelement <64 x i1> %43, i64 63, !dbg !23 + %304 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %239, i1 %303) #7, !dbg !23 + %305 = extractvalue { i32, i32, i32, i32 } %304, 0, !dbg !23 + %306 = bitcast i32 %305 to <2 x bfloat>, !dbg !23 + %307 = extractvalue { i32, i32, i32, i32 } %304, 1, !dbg !23 + %308 = bitcast i32 %307 to <2 x bfloat>, !dbg !23 + %309 = extractvalue { i32, i32, i32, i32 } %304, 2, !dbg !23 + %310 = bitcast i32 %309 to <2 x bfloat>, !dbg !23 + %311 = extractvalue { i32, i32, i32, i32 } %304, 3, !dbg !23 + %312 = bitcast i32 %311 to <2 x bfloat>, !dbg !23 + %313 = extractelement <2 x bfloat> %306, i64 0, !dbg !23 + %314 = extractelement <2 x bfloat> %306, i64 1, !dbg !23 + %315 = extractelement <2 x bfloat> %308, i64 0, !dbg !23 + %316 = extractelement <2 x bfloat> %308, i64 1, !dbg !23 + %317 = extractelement <2 x bfloat> %310, i64 0, !dbg !23 + %318 = extractelement <2 x bfloat> %310, i64 1, !dbg !23 + %319 = extractelement <2 x bfloat> %312, i64 0, !dbg !23 + %320 = extractelement <2 x bfloat> %312, i64 1, !dbg !23 + %321 = extractelement <64 x i1> %43, i64 62, !dbg !23 + %322 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %240, i1 %321) #7, !dbg !23 + %323 = extractvalue { i32, i32, i32, i32 } %322, 0, !dbg !23 + %324 = bitcast i32 %323 to <2 x bfloat>, !dbg !23 + %325 = extractvalue { i32, i32, i32, i32 } %322, 1, !dbg !23 + %326 = bitcast i32 %325 to <2 x bfloat>, !dbg !23 + %327 = extractvalue { i32, i32, i32, i32 } %322, 2, !dbg !23 + %328 = bitcast i32 %327 to <2 x bfloat>, !dbg !23 + %329 = extractvalue { i32, i32, i32, i32 } %322, 3, !dbg !23 + %330 = bitcast i32 %329 to <2 x bfloat>, !dbg !23 + %331 = extractelement <2 x bfloat> %324, i64 0, !dbg !23 + %332 = extractelement <2 x bfloat> %324, i64 1, !dbg !23 + %333 = extractelement <2 x bfloat> %326, i64 0, !dbg !23 + %334 = extractelement <2 x bfloat> %326, i64 1, !dbg !23 + %335 = extractelement <2 x bfloat> %328, i64 0, !dbg !23 + %336 = extractelement <2 x bfloat> %328, i64 1, !dbg !23 + %337 = extractelement <2 x bfloat> %330, i64 0, !dbg !23 + %338 = extractelement <2 x bfloat> %330, i64 1, !dbg !23 + %339 = extractelement <64 x i1> %43, i64 61, !dbg !23 + %340 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %241, i1 %339) #7, !dbg !23 + %341 = extractvalue { i32, i32, i32, i32 } %340, 0, !dbg !23 + %342 = bitcast i32 %341 to <2 x bfloat>, !dbg !23 + %343 = extractvalue { i32, i32, i32, i32 } %340, 1, !dbg !23 + %344 = bitcast i32 %343 to <2 x bfloat>, !dbg !23 + %345 = extractvalue { i32, i32, i32, i32 } %340, 2, !dbg !23 + %346 = bitcast i32 %345 to <2 x bfloat>, !dbg !23 + %347 = extractvalue { i32, i32, i32, i32 } %340, 3, !dbg !23 + %348 = bitcast i32 %347 to <2 x bfloat>, !dbg !23 + %349 = extractelement <2 x bfloat> %342, i64 0, !dbg !23 + %350 = extractelement <2 x bfloat> %342, i64 1, !dbg !23 + %351 = extractelement <2 x bfloat> %344, i64 0, !dbg !23 + %352 = extractelement <2 x bfloat> %344, i64 1, !dbg !23 + %353 = extractelement <2 x bfloat> %346, i64 0, !dbg !23 + %354 = extractelement <2 x bfloat> %346, i64 1, !dbg !23 + %355 = extractelement <2 x bfloat> %348, i64 0, !dbg !23 + %356 = extractelement <2 x bfloat> %348, i64 1, !dbg !23 + %357 = extractelement <64 x i1> %43, i64 60, !dbg !23 + %358 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %242, i1 %357) #7, !dbg !23 + %359 = extractvalue { i32, i32, i32, i32 } %358, 0, !dbg !23 + %360 = bitcast i32 %359 to <2 x bfloat>, !dbg !23 + %361 = extractvalue { i32, i32, i32, i32 } %358, 1, !dbg !23 + %362 = bitcast i32 %361 to <2 x bfloat>, !dbg !23 + %363 = extractvalue { i32, i32, i32, i32 } %358, 2, !dbg !23 + %364 = bitcast i32 %363 to <2 x bfloat>, !dbg !23 + %365 = extractvalue { i32, i32, i32, i32 } %358, 3, !dbg !23 + %366 = bitcast i32 %365 to <2 x bfloat>, !dbg !23 + %367 = extractelement <2 x bfloat> %360, i64 0, !dbg !23 + %368 = extractelement <2 x bfloat> %360, i64 1, !dbg !23 + %369 = extractelement <2 x bfloat> %362, i64 0, !dbg !23 + %370 = extractelement <2 x bfloat> %362, i64 1, !dbg !23 + %371 = extractelement <2 x bfloat> %364, i64 0, !dbg !23 + %372 = extractelement <2 x bfloat> %364, i64 1, !dbg !23 + %373 = extractelement <2 x bfloat> %366, i64 0, !dbg !23 + %374 = extractelement <2 x bfloat> %366, i64 1, !dbg !23 + %375 = extractelement <64 x i1> %43, i64 59, !dbg !23 + %376 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %243, i1 %375) #7, !dbg !23 + %377 = extractvalue { i32, i32, i32, i32 } %376, 0, !dbg !23 + %378 = bitcast i32 %377 to <2 x bfloat>, !dbg !23 + %379 = extractvalue { i32, i32, i32, i32 } %376, 1, !dbg !23 + %380 = bitcast i32 %379 to <2 x bfloat>, !dbg !23 + %381 = extractvalue { i32, i32, i32, i32 } %376, 2, !dbg !23 + %382 = bitcast i32 %381 to <2 x bfloat>, !dbg !23 + %383 = extractvalue { i32, i32, i32, i32 } %376, 3, !dbg !23 + %384 = bitcast i32 %383 to <2 x bfloat>, !dbg !23 + %385 = extractelement <2 x bfloat> %378, i64 0, !dbg !23 + %386 = extractelement <2 x bfloat> %378, i64 1, !dbg !23 + %387 = extractelement <2 x bfloat> %380, i64 0, !dbg !23 + %388 = extractelement <2 x bfloat> %380, i64 1, !dbg !23 + %389 = extractelement <2 x bfloat> %382, i64 0, !dbg !23 + %390 = extractelement <2 x bfloat> %382, i64 1, !dbg !23 + %391 = extractelement <2 x bfloat> %384, i64 0, !dbg !23 + %392 = extractelement <2 x bfloat> %384, i64 1, !dbg !23 + %393 = extractelement <64 x i1> %43, i64 58, !dbg !23 + %394 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %244, i1 %393) #7, !dbg !23 + %395 = extractvalue { i32, i32, i32, i32 } %394, 0, !dbg !23 + %396 = bitcast i32 %395 to <2 x bfloat>, !dbg !23 + %397 = extractvalue { i32, i32, i32, i32 } %394, 1, !dbg !23 + %398 = bitcast i32 %397 to <2 x bfloat>, !dbg !23 + %399 = extractvalue { i32, i32, i32, i32 } %394, 2, !dbg !23 + %400 = bitcast i32 %399 to <2 x bfloat>, !dbg !23 + %401 = extractvalue { i32, i32, i32, i32 } %394, 3, !dbg !23 + %402 = bitcast i32 %401 to <2 x bfloat>, !dbg !23 + %403 = extractelement <2 x bfloat> %396, i64 0, !dbg !23 + %404 = extractelement <2 x bfloat> %396, i64 1, !dbg !23 + %405 = extractelement <2 x bfloat> %398, i64 0, !dbg !23 + %406 = extractelement <2 x bfloat> %398, i64 1, !dbg !23 + %407 = extractelement <2 x bfloat> %400, i64 0, !dbg !23 + %408 = extractelement <2 x bfloat> %400, i64 1, !dbg !23 + %409 = extractelement <2 x bfloat> %402, i64 0, !dbg !23 + %410 = extractelement <2 x bfloat> %402, i64 1, !dbg !23 + %411 = extractelement <64 x i1> %43, i64 57, !dbg !23 + %412 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %245, i1 %411) #7, !dbg !23 + %413 = extractvalue { i32, i32, i32, i32 } %412, 0, !dbg !23 + %414 = bitcast i32 %413 to <2 x bfloat>, !dbg !23 + %415 = extractvalue { i32, i32, i32, i32 } %412, 1, !dbg !23 + %416 = bitcast i32 %415 to <2 x bfloat>, !dbg !23 + %417 = extractvalue { i32, i32, i32, i32 } %412, 2, !dbg !23 + %418 = bitcast i32 %417 to <2 x bfloat>, !dbg !23 + %419 = extractvalue { i32, i32, i32, i32 } %412, 3, !dbg !23 + %420 = bitcast i32 %419 to <2 x bfloat>, !dbg !23 + %421 = extractelement <2 x bfloat> %414, i64 0, !dbg !23 + %422 = extractelement <2 x bfloat> %414, i64 1, !dbg !23 + %423 = extractelement <2 x bfloat> %416, i64 0, !dbg !23 + %424 = extractelement <2 x bfloat> %416, i64 1, !dbg !23 + %425 = extractelement <2 x bfloat> %418, i64 0, !dbg !23 + %426 = extractelement <2 x bfloat> %418, i64 1, !dbg !23 + %427 = extractelement <2 x bfloat> %420, i64 0, !dbg !23 + %428 = extractelement <2 x bfloat> %420, i64 1, !dbg !23 + %429 = extractelement <64 x i1> %43, i64 56, !dbg !23 + %430 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %246, i1 %429) #7, !dbg !23 + %431 = extractvalue { i32, i32, i32, i32 } %430, 0, !dbg !23 + %432 = bitcast i32 %431 to <2 x bfloat>, !dbg !23 + %433 = extractvalue { i32, i32, i32, i32 } %430, 1, !dbg !23 + %434 = bitcast i32 %433 to <2 x bfloat>, !dbg !23 + %435 = extractvalue { i32, i32, i32, i32 } %430, 2, !dbg !23 + %436 = bitcast i32 %435 to <2 x bfloat>, !dbg !23 + %437 = extractvalue { i32, i32, i32, i32 } %430, 3, !dbg !23 + %438 = bitcast i32 %437 to <2 x bfloat>, !dbg !23 + %439 = extractelement <2 x bfloat> %432, i64 0, !dbg !23 + %440 = extractelement <2 x bfloat> %432, i64 1, !dbg !23 + %441 = extractelement <2 x bfloat> %434, i64 0, !dbg !23 + %442 = extractelement <2 x bfloat> %434, i64 1, !dbg !23 + %443 = extractelement <2 x bfloat> %436, i64 0, !dbg !23 + %444 = extractelement <2 x bfloat> %436, i64 1, !dbg !23 + %445 = extractelement <2 x bfloat> %438, i64 0, !dbg !23 + %446 = extractelement <2 x bfloat> %438, i64 1, !dbg !23 + %447 = extractelement <64 x i1> %43, i64 55, !dbg !23 + %448 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %247, i1 %447) #7, !dbg !23 + %449 = extractvalue { i32, i32, i32, i32 } %448, 0, !dbg !23 + %450 = bitcast i32 %449 to <2 x bfloat>, !dbg !23 + %451 = extractvalue { i32, i32, i32, i32 } %448, 1, !dbg !23 + %452 = bitcast i32 %451 to <2 x bfloat>, !dbg !23 + %453 = extractvalue { i32, i32, i32, i32 } %448, 2, !dbg !23 + %454 = bitcast i32 %453 to <2 x bfloat>, !dbg !23 + %455 = extractvalue { i32, i32, i32, i32 } %448, 3, !dbg !23 + %456 = bitcast i32 %455 to <2 x bfloat>, !dbg !23 + %457 = extractelement <2 x bfloat> %450, i64 0, !dbg !23 + %458 = extractelement <2 x bfloat> %450, i64 1, !dbg !23 + %459 = extractelement <2 x bfloat> %452, i64 0, !dbg !23 + %460 = extractelement <2 x bfloat> %452, i64 1, !dbg !23 + %461 = extractelement <2 x bfloat> %454, i64 0, !dbg !23 + %462 = extractelement <2 x bfloat> %454, i64 1, !dbg !23 + %463 = extractelement <2 x bfloat> %456, i64 0, !dbg !23 + %464 = extractelement <2 x bfloat> %456, i64 1, !dbg !23 + %465 = extractelement <64 x i1> %43, i64 54, !dbg !23 + %466 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %248, i1 %465) #7, !dbg !23 + %467 = extractvalue { i32, i32, i32, i32 } %466, 0, !dbg !23 + %468 = bitcast i32 %467 to <2 x bfloat>, !dbg !23 + %469 = extractvalue { i32, i32, i32, i32 } %466, 1, !dbg !23 + %470 = bitcast i32 %469 to <2 x bfloat>, !dbg !23 + %471 = extractvalue { i32, i32, i32, i32 } %466, 2, !dbg !23 + %472 = bitcast i32 %471 to <2 x bfloat>, !dbg !23 + %473 = extractvalue { i32, i32, i32, i32 } %466, 3, !dbg !23 + %474 = bitcast i32 %473 to <2 x bfloat>, !dbg !23 + %475 = extractelement <2 x bfloat> %468, i64 0, !dbg !23 + %476 = extractelement <2 x bfloat> %468, i64 1, !dbg !23 + %477 = extractelement <2 x bfloat> %470, i64 0, !dbg !23 + %478 = extractelement <2 x bfloat> %470, i64 1, !dbg !23 + %479 = extractelement <2 x bfloat> %472, i64 0, !dbg !23 + %480 = extractelement <2 x bfloat> %472, i64 1, !dbg !23 + %481 = extractelement <2 x bfloat> %474, i64 0, !dbg !23 + %482 = extractelement <2 x bfloat> %474, i64 1, !dbg !23 + %483 = extractelement <64 x i1> %43, i64 53, !dbg !23 + %484 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %249, i1 %483) #7, !dbg !23 + %485 = extractvalue { i32, i32, i32, i32 } %484, 0, !dbg !23 + %486 = bitcast i32 %485 to <2 x bfloat>, !dbg !23 + %487 = extractvalue { i32, i32, i32, i32 } %484, 1, !dbg !23 + %488 = bitcast i32 %487 to <2 x bfloat>, !dbg !23 + %489 = extractvalue { i32, i32, i32, i32 } %484, 2, !dbg !23 + %490 = bitcast i32 %489 to <2 x bfloat>, !dbg !23 + %491 = extractvalue { i32, i32, i32, i32 } %484, 3, !dbg !23 + %492 = bitcast i32 %491 to <2 x bfloat>, !dbg !23 + %493 = extractelement <2 x bfloat> %486, i64 0, !dbg !23 + %494 = extractelement <2 x bfloat> %486, i64 1, !dbg !23 + %495 = extractelement <2 x bfloat> %488, i64 0, !dbg !23 + %496 = extractelement <2 x bfloat> %488, i64 1, !dbg !23 + %497 = extractelement <2 x bfloat> %490, i64 0, !dbg !23 + %498 = extractelement <2 x bfloat> %490, i64 1, !dbg !23 + %499 = extractelement <2 x bfloat> %492, i64 0, !dbg !23 + %500 = extractelement <2 x bfloat> %492, i64 1, !dbg !23 + %501 = extractelement <64 x i1> %43, i64 52, !dbg !23 + %502 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %250, i1 %501) #7, !dbg !23 + %503 = extractvalue { i32, i32, i32, i32 } %502, 0, !dbg !23 + %504 = bitcast i32 %503 to <2 x bfloat>, !dbg !23 + %505 = extractvalue { i32, i32, i32, i32 } %502, 1, !dbg !23 + %506 = bitcast i32 %505 to <2 x bfloat>, !dbg !23 + %507 = extractvalue { i32, i32, i32, i32 } %502, 2, !dbg !23 + %508 = bitcast i32 %507 to <2 x bfloat>, !dbg !23 + %509 = extractvalue { i32, i32, i32, i32 } %502, 3, !dbg !23 + %510 = bitcast i32 %509 to <2 x bfloat>, !dbg !23 + %511 = extractelement <2 x bfloat> %504, i64 0, !dbg !23 + %512 = extractelement <2 x bfloat> %504, i64 1, !dbg !23 + %513 = extractelement <2 x bfloat> %506, i64 0, !dbg !23 + %514 = extractelement <2 x bfloat> %506, i64 1, !dbg !23 + %515 = extractelement <2 x bfloat> %508, i64 0, !dbg !23 + %516 = extractelement <2 x bfloat> %508, i64 1, !dbg !23 + %517 = extractelement <2 x bfloat> %510, i64 0, !dbg !23 + %518 = extractelement <2 x bfloat> %510, i64 1, !dbg !23 + %519 = extractelement <64 x i1> %43, i64 51, !dbg !23 + %520 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %251, i1 %519) #7, !dbg !23 + %521 = extractvalue { i32, i32, i32, i32 } %520, 0, !dbg !23 + %522 = bitcast i32 %521 to <2 x bfloat>, !dbg !23 + %523 = extractvalue { i32, i32, i32, i32 } %520, 1, !dbg !23 + %524 = bitcast i32 %523 to <2 x bfloat>, !dbg !23 + %525 = extractvalue { i32, i32, i32, i32 } %520, 2, !dbg !23 + %526 = bitcast i32 %525 to <2 x bfloat>, !dbg !23 + %527 = extractvalue { i32, i32, i32, i32 } %520, 3, !dbg !23 + %528 = bitcast i32 %527 to <2 x bfloat>, !dbg !23 + %529 = extractelement <2 x bfloat> %522, i64 0, !dbg !23 + %530 = extractelement <2 x bfloat> %522, i64 1, !dbg !23 + %531 = extractelement <2 x bfloat> %524, i64 0, !dbg !23 + %532 = extractelement <2 x bfloat> %524, i64 1, !dbg !23 + %533 = extractelement <2 x bfloat> %526, i64 0, !dbg !23 + %534 = extractelement <2 x bfloat> %526, i64 1, !dbg !23 + %535 = extractelement <2 x bfloat> %528, i64 0, !dbg !23 + %536 = extractelement <2 x bfloat> %528, i64 1, !dbg !23 + %537 = extractelement <64 x i1> %43, i64 50, !dbg !23 + %538 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %252, i1 %537) #7, !dbg !23 + %539 = extractvalue { i32, i32, i32, i32 } %538, 0, !dbg !23 + %540 = bitcast i32 %539 to <2 x bfloat>, !dbg !23 + %541 = extractvalue { i32, i32, i32, i32 } %538, 1, !dbg !23 + %542 = bitcast i32 %541 to <2 x bfloat>, !dbg !23 + %543 = extractvalue { i32, i32, i32, i32 } %538, 2, !dbg !23 + %544 = bitcast i32 %543 to <2 x bfloat>, !dbg !23 + %545 = extractvalue { i32, i32, i32, i32 } %538, 3, !dbg !23 + %546 = bitcast i32 %545 to <2 x bfloat>, !dbg !23 + %547 = extractelement <2 x bfloat> %540, i64 0, !dbg !23 + %548 = extractelement <2 x bfloat> %540, i64 1, !dbg !23 + %549 = extractelement <2 x bfloat> %542, i64 0, !dbg !23 + %550 = extractelement <2 x bfloat> %542, i64 1, !dbg !23 + %551 = extractelement <2 x bfloat> %544, i64 0, !dbg !23 + %552 = extractelement <2 x bfloat> %544, i64 1, !dbg !23 + %553 = extractelement <2 x bfloat> %546, i64 0, !dbg !23 + %554 = extractelement <2 x bfloat> %546, i64 1, !dbg !23 + %555 = extractelement <64 x i1> %43, i64 49, !dbg !23 + %556 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %253, i1 %555) #7, !dbg !23 + %557 = extractvalue { i32, i32, i32, i32 } %556, 0, !dbg !23 + %558 = bitcast i32 %557 to <2 x bfloat>, !dbg !23 + %559 = extractvalue { i32, i32, i32, i32 } %556, 1, !dbg !23 + %560 = bitcast i32 %559 to <2 x bfloat>, !dbg !23 + %561 = extractvalue { i32, i32, i32, i32 } %556, 2, !dbg !23 + %562 = bitcast i32 %561 to <2 x bfloat>, !dbg !23 + %563 = extractvalue { i32, i32, i32, i32 } %556, 3, !dbg !23 + %564 = bitcast i32 %563 to <2 x bfloat>, !dbg !23 + %565 = extractelement <2 x bfloat> %558, i64 0, !dbg !23 + %566 = extractelement <2 x bfloat> %558, i64 1, !dbg !23 + %567 = extractelement <2 x bfloat> %560, i64 0, !dbg !23 + %568 = extractelement <2 x bfloat> %560, i64 1, !dbg !23 + %569 = extractelement <2 x bfloat> %562, i64 0, !dbg !23 + %570 = extractelement <2 x bfloat> %562, i64 1, !dbg !23 + %571 = extractelement <2 x bfloat> %564, i64 0, !dbg !23 + %572 = extractelement <2 x bfloat> %564, i64 1, !dbg !23 + %573 = extractelement <64 x i1> %43, i64 48, !dbg !23 + %574 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %254, i1 %573) #7, !dbg !23 + %575 = extractvalue { i32, i32, i32, i32 } %574, 0, !dbg !23 + %576 = bitcast i32 %575 to <2 x bfloat>, !dbg !23 + %577 = extractvalue { i32, i32, i32, i32 } %574, 1, !dbg !23 + %578 = bitcast i32 %577 to <2 x bfloat>, !dbg !23 + %579 = extractvalue { i32, i32, i32, i32 } %574, 2, !dbg !23 + %580 = bitcast i32 %579 to <2 x bfloat>, !dbg !23 + %581 = extractvalue { i32, i32, i32, i32 } %574, 3, !dbg !23 + %582 = bitcast i32 %581 to <2 x bfloat>, !dbg !23 + %583 = extractelement <2 x bfloat> %576, i64 0, !dbg !23 + %584 = extractelement <2 x bfloat> %576, i64 1, !dbg !23 + %585 = extractelement <2 x bfloat> %578, i64 0, !dbg !23 + %586 = extractelement <2 x bfloat> %578, i64 1, !dbg !23 + %587 = extractelement <2 x bfloat> %580, i64 0, !dbg !23 + %588 = extractelement <2 x bfloat> %580, i64 1, !dbg !23 + %589 = extractelement <2 x bfloat> %582, i64 0, !dbg !23 + %590 = extractelement <2 x bfloat> %582, i64 1, !dbg !23 + %591 = extractelement <64 x i1> %43, i64 47, !dbg !23 + %592 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %255, i1 %591) #7, !dbg !23 + %593 = extractvalue { i32, i32, i32, i32 } %592, 0, !dbg !23 + %594 = bitcast i32 %593 to <2 x bfloat>, !dbg !23 + %595 = extractvalue { i32, i32, i32, i32 } %592, 1, !dbg !23 + %596 = bitcast i32 %595 to <2 x bfloat>, !dbg !23 + %597 = extractvalue { i32, i32, i32, i32 } %592, 2, !dbg !23 + %598 = bitcast i32 %597 to <2 x bfloat>, !dbg !23 + %599 = extractvalue { i32, i32, i32, i32 } %592, 3, !dbg !23 + %600 = bitcast i32 %599 to <2 x bfloat>, !dbg !23 + %601 = extractelement <2 x bfloat> %594, i64 0, !dbg !23 + %602 = extractelement <2 x bfloat> %594, i64 1, !dbg !23 + %603 = extractelement <2 x bfloat> %596, i64 0, !dbg !23 + %604 = extractelement <2 x bfloat> %596, i64 1, !dbg !23 + %605 = extractelement <2 x bfloat> %598, i64 0, !dbg !23 + %606 = extractelement <2 x bfloat> %598, i64 1, !dbg !23 + %607 = extractelement <2 x bfloat> %600, i64 0, !dbg !23 + %608 = extractelement <2 x bfloat> %600, i64 1, !dbg !23 + %609 = extractelement <64 x i1> %43, i64 46, !dbg !23 + %610 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %256, i1 %609) #7, !dbg !23 + %611 = extractvalue { i32, i32, i32, i32 } %610, 0, !dbg !23 + %612 = bitcast i32 %611 to <2 x bfloat>, !dbg !23 + %613 = extractvalue { i32, i32, i32, i32 } %610, 1, !dbg !23 + %614 = bitcast i32 %613 to <2 x bfloat>, !dbg !23 + %615 = extractvalue { i32, i32, i32, i32 } %610, 2, !dbg !23 + %616 = bitcast i32 %615 to <2 x bfloat>, !dbg !23 + %617 = extractvalue { i32, i32, i32, i32 } %610, 3, !dbg !23 + %618 = bitcast i32 %617 to <2 x bfloat>, !dbg !23 + %619 = extractelement <2 x bfloat> %612, i64 0, !dbg !23 + %620 = extractelement <2 x bfloat> %612, i64 1, !dbg !23 + %621 = extractelement <2 x bfloat> %614, i64 0, !dbg !23 + %622 = extractelement <2 x bfloat> %614, i64 1, !dbg !23 + %623 = extractelement <2 x bfloat> %616, i64 0, !dbg !23 + %624 = extractelement <2 x bfloat> %616, i64 1, !dbg !23 + %625 = extractelement <2 x bfloat> %618, i64 0, !dbg !23 + %626 = extractelement <2 x bfloat> %618, i64 1, !dbg !23 + %627 = extractelement <64 x i1> %43, i64 45, !dbg !23 + %628 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %257, i1 %627) #7, !dbg !23 + %629 = extractvalue { i32, i32, i32, i32 } %628, 0, !dbg !23 + %630 = bitcast i32 %629 to <2 x bfloat>, !dbg !23 + %631 = extractvalue { i32, i32, i32, i32 } %628, 1, !dbg !23 + %632 = bitcast i32 %631 to <2 x bfloat>, !dbg !23 + %633 = extractvalue { i32, i32, i32, i32 } %628, 2, !dbg !23 + %634 = bitcast i32 %633 to <2 x bfloat>, !dbg !23 + %635 = extractvalue { i32, i32, i32, i32 } %628, 3, !dbg !23 + %636 = bitcast i32 %635 to <2 x bfloat>, !dbg !23 + %637 = extractelement <2 x bfloat> %630, i64 0, !dbg !23 + %638 = extractelement <2 x bfloat> %630, i64 1, !dbg !23 + %639 = extractelement <2 x bfloat> %632, i64 0, !dbg !23 + %640 = extractelement <2 x bfloat> %632, i64 1, !dbg !23 + %641 = extractelement <2 x bfloat> %634, i64 0, !dbg !23 + %642 = extractelement <2 x bfloat> %634, i64 1, !dbg !23 + %643 = extractelement <2 x bfloat> %636, i64 0, !dbg !23 + %644 = extractelement <2 x bfloat> %636, i64 1, !dbg !23 + %645 = extractelement <64 x i1> %43, i64 44, !dbg !23 + %646 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %258, i1 %645) #7, !dbg !23 + %647 = extractvalue { i32, i32, i32, i32 } %646, 0, !dbg !23 + %648 = bitcast i32 %647 to <2 x bfloat>, !dbg !23 + %649 = extractvalue { i32, i32, i32, i32 } %646, 1, !dbg !23 + %650 = bitcast i32 %649 to <2 x bfloat>, !dbg !23 + %651 = extractvalue { i32, i32, i32, i32 } %646, 2, !dbg !23 + %652 = bitcast i32 %651 to <2 x bfloat>, !dbg !23 + %653 = extractvalue { i32, i32, i32, i32 } %646, 3, !dbg !23 + %654 = bitcast i32 %653 to <2 x bfloat>, !dbg !23 + %655 = extractelement <2 x bfloat> %648, i64 0, !dbg !23 + %656 = extractelement <2 x bfloat> %648, i64 1, !dbg !23 + %657 = extractelement <2 x bfloat> %650, i64 0, !dbg !23 + %658 = extractelement <2 x bfloat> %650, i64 1, !dbg !23 + %659 = extractelement <2 x bfloat> %652, i64 0, !dbg !23 + %660 = extractelement <2 x bfloat> %652, i64 1, !dbg !23 + %661 = extractelement <2 x bfloat> %654, i64 0, !dbg !23 + %662 = extractelement <2 x bfloat> %654, i64 1, !dbg !23 + %663 = extractelement <64 x i1> %43, i64 43, !dbg !23 + %664 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %259, i1 %663) #7, !dbg !23 + %665 = extractvalue { i32, i32, i32, i32 } %664, 0, !dbg !23 + %666 = bitcast i32 %665 to <2 x bfloat>, !dbg !23 + %667 = extractvalue { i32, i32, i32, i32 } %664, 1, !dbg !23 + %668 = bitcast i32 %667 to <2 x bfloat>, !dbg !23 + %669 = extractvalue { i32, i32, i32, i32 } %664, 2, !dbg !23 + %670 = bitcast i32 %669 to <2 x bfloat>, !dbg !23 + %671 = extractvalue { i32, i32, i32, i32 } %664, 3, !dbg !23 + %672 = bitcast i32 %671 to <2 x bfloat>, !dbg !23 + %673 = extractelement <2 x bfloat> %666, i64 0, !dbg !23 + %674 = extractelement <2 x bfloat> %666, i64 1, !dbg !23 + %675 = extractelement <2 x bfloat> %668, i64 0, !dbg !23 + %676 = extractelement <2 x bfloat> %668, i64 1, !dbg !23 + %677 = extractelement <2 x bfloat> %670, i64 0, !dbg !23 + %678 = extractelement <2 x bfloat> %670, i64 1, !dbg !23 + %679 = extractelement <2 x bfloat> %672, i64 0, !dbg !23 + %680 = extractelement <2 x bfloat> %672, i64 1, !dbg !23 + %681 = extractelement <64 x i1> %43, i64 42, !dbg !23 + %682 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %260, i1 %681) #7, !dbg !23 + %683 = extractvalue { i32, i32, i32, i32 } %682, 0, !dbg !23 + %684 = bitcast i32 %683 to <2 x bfloat>, !dbg !23 + %685 = extractvalue { i32, i32, i32, i32 } %682, 1, !dbg !23 + %686 = bitcast i32 %685 to <2 x bfloat>, !dbg !23 + %687 = extractvalue { i32, i32, i32, i32 } %682, 2, !dbg !23 + %688 = bitcast i32 %687 to <2 x bfloat>, !dbg !23 + %689 = extractvalue { i32, i32, i32, i32 } %682, 3, !dbg !23 + %690 = bitcast i32 %689 to <2 x bfloat>, !dbg !23 + %691 = extractelement <2 x bfloat> %684, i64 0, !dbg !23 + %692 = extractelement <2 x bfloat> %684, i64 1, !dbg !23 + %693 = extractelement <2 x bfloat> %686, i64 0, !dbg !23 + %694 = extractelement <2 x bfloat> %686, i64 1, !dbg !23 + %695 = extractelement <2 x bfloat> %688, i64 0, !dbg !23 + %696 = extractelement <2 x bfloat> %688, i64 1, !dbg !23 + %697 = extractelement <2 x bfloat> %690, i64 0, !dbg !23 + %698 = extractelement <2 x bfloat> %690, i64 1, !dbg !23 + %699 = extractelement <64 x i1> %43, i64 41, !dbg !23 + %700 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %261, i1 %699) #7, !dbg !23 + %701 = extractvalue { i32, i32, i32, i32 } %700, 0, !dbg !23 + %702 = bitcast i32 %701 to <2 x bfloat>, !dbg !23 + %703 = extractvalue { i32, i32, i32, i32 } %700, 1, !dbg !23 + %704 = bitcast i32 %703 to <2 x bfloat>, !dbg !23 + %705 = extractvalue { i32, i32, i32, i32 } %700, 2, !dbg !23 + %706 = bitcast i32 %705 to <2 x bfloat>, !dbg !23 + %707 = extractvalue { i32, i32, i32, i32 } %700, 3, !dbg !23 + %708 = bitcast i32 %707 to <2 x bfloat>, !dbg !23 + %709 = extractelement <2 x bfloat> %702, i64 0, !dbg !23 + %710 = extractelement <2 x bfloat> %702, i64 1, !dbg !23 + %711 = extractelement <2 x bfloat> %704, i64 0, !dbg !23 + %712 = extractelement <2 x bfloat> %704, i64 1, !dbg !23 + %713 = extractelement <2 x bfloat> %706, i64 0, !dbg !23 + %714 = extractelement <2 x bfloat> %706, i64 1, !dbg !23 + %715 = extractelement <2 x bfloat> %708, i64 0, !dbg !23 + %716 = extractelement <2 x bfloat> %708, i64 1, !dbg !23 + %717 = extractelement <64 x i1> %43, i64 40, !dbg !23 + %718 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %262, i1 %717) #7, !dbg !23 + %719 = extractvalue { i32, i32, i32, i32 } %718, 0, !dbg !23 + %720 = bitcast i32 %719 to <2 x bfloat>, !dbg !23 + %721 = extractvalue { i32, i32, i32, i32 } %718, 1, !dbg !23 + %722 = bitcast i32 %721 to <2 x bfloat>, !dbg !23 + %723 = extractvalue { i32, i32, i32, i32 } %718, 2, !dbg !23 + %724 = bitcast i32 %723 to <2 x bfloat>, !dbg !23 + %725 = extractvalue { i32, i32, i32, i32 } %718, 3, !dbg !23 + %726 = bitcast i32 %725 to <2 x bfloat>, !dbg !23 + %727 = extractelement <2 x bfloat> %720, i64 0, !dbg !23 + %728 = extractelement <2 x bfloat> %720, i64 1, !dbg !23 + %729 = extractelement <2 x bfloat> %722, i64 0, !dbg !23 + %730 = extractelement <2 x bfloat> %722, i64 1, !dbg !23 + %731 = extractelement <2 x bfloat> %724, i64 0, !dbg !23 + %732 = extractelement <2 x bfloat> %724, i64 1, !dbg !23 + %733 = extractelement <2 x bfloat> %726, i64 0, !dbg !23 + %734 = extractelement <2 x bfloat> %726, i64 1, !dbg !23 + %735 = extractelement <64 x i1> %43, i64 39, !dbg !23 + %736 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %263, i1 %735) #7, !dbg !23 + %737 = extractvalue { i32, i32, i32, i32 } %736, 0, !dbg !23 + %738 = bitcast i32 %737 to <2 x bfloat>, !dbg !23 + %739 = extractvalue { i32, i32, i32, i32 } %736, 1, !dbg !23 + %740 = bitcast i32 %739 to <2 x bfloat>, !dbg !23 + %741 = extractvalue { i32, i32, i32, i32 } %736, 2, !dbg !23 + %742 = bitcast i32 %741 to <2 x bfloat>, !dbg !23 + %743 = extractvalue { i32, i32, i32, i32 } %736, 3, !dbg !23 + %744 = bitcast i32 %743 to <2 x bfloat>, !dbg !23 + %745 = extractelement <2 x bfloat> %738, i64 0, !dbg !23 + %746 = extractelement <2 x bfloat> %738, i64 1, !dbg !23 + %747 = extractelement <2 x bfloat> %740, i64 0, !dbg !23 + %748 = extractelement <2 x bfloat> %740, i64 1, !dbg !23 + %749 = extractelement <2 x bfloat> %742, i64 0, !dbg !23 + %750 = extractelement <2 x bfloat> %742, i64 1, !dbg !23 + %751 = extractelement <2 x bfloat> %744, i64 0, !dbg !23 + %752 = extractelement <2 x bfloat> %744, i64 1, !dbg !23 + %753 = extractelement <64 x i1> %43, i64 38, !dbg !23 + %754 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %264, i1 %753) #7, !dbg !23 + %755 = extractvalue { i32, i32, i32, i32 } %754, 0, !dbg !23 + %756 = bitcast i32 %755 to <2 x bfloat>, !dbg !23 + %757 = extractvalue { i32, i32, i32, i32 } %754, 1, !dbg !23 + %758 = bitcast i32 %757 to <2 x bfloat>, !dbg !23 + %759 = extractvalue { i32, i32, i32, i32 } %754, 2, !dbg !23 + %760 = bitcast i32 %759 to <2 x bfloat>, !dbg !23 + %761 = extractvalue { i32, i32, i32, i32 } %754, 3, !dbg !23 + %762 = bitcast i32 %761 to <2 x bfloat>, !dbg !23 + %763 = extractelement <2 x bfloat> %756, i64 0, !dbg !23 + %764 = extractelement <2 x bfloat> %756, i64 1, !dbg !23 + %765 = extractelement <2 x bfloat> %758, i64 0, !dbg !23 + %766 = extractelement <2 x bfloat> %758, i64 1, !dbg !23 + %767 = extractelement <2 x bfloat> %760, i64 0, !dbg !23 + %768 = extractelement <2 x bfloat> %760, i64 1, !dbg !23 + %769 = extractelement <2 x bfloat> %762, i64 0, !dbg !23 + %770 = extractelement <2 x bfloat> %762, i64 1, !dbg !23 + %771 = extractelement <64 x i1> %43, i64 37, !dbg !23 + %772 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %265, i1 %771) #7, !dbg !23 + %773 = extractvalue { i32, i32, i32, i32 } %772, 0, !dbg !23 + %774 = bitcast i32 %773 to <2 x bfloat>, !dbg !23 + %775 = extractvalue { i32, i32, i32, i32 } %772, 1, !dbg !23 + %776 = bitcast i32 %775 to <2 x bfloat>, !dbg !23 + %777 = extractvalue { i32, i32, i32, i32 } %772, 2, !dbg !23 + %778 = bitcast i32 %777 to <2 x bfloat>, !dbg !23 + %779 = extractvalue { i32, i32, i32, i32 } %772, 3, !dbg !23 + %780 = bitcast i32 %779 to <2 x bfloat>, !dbg !23 + %781 = extractelement <2 x bfloat> %774, i64 0, !dbg !23 + %782 = extractelement <2 x bfloat> %774, i64 1, !dbg !23 + %783 = extractelement <2 x bfloat> %776, i64 0, !dbg !23 + %784 = extractelement <2 x bfloat> %776, i64 1, !dbg !23 + %785 = extractelement <2 x bfloat> %778, i64 0, !dbg !23 + %786 = extractelement <2 x bfloat> %778, i64 1, !dbg !23 + %787 = extractelement <2 x bfloat> %780, i64 0, !dbg !23 + %788 = extractelement <2 x bfloat> %780, i64 1, !dbg !23 + %789 = extractelement <64 x i1> %43, i64 36, !dbg !23 + %790 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %266, i1 %789) #7, !dbg !23 + %791 = extractvalue { i32, i32, i32, i32 } %790, 0, !dbg !23 + %792 = bitcast i32 %791 to <2 x bfloat>, !dbg !23 + %793 = extractvalue { i32, i32, i32, i32 } %790, 1, !dbg !23 + %794 = bitcast i32 %793 to <2 x bfloat>, !dbg !23 + %795 = extractvalue { i32, i32, i32, i32 } %790, 2, !dbg !23 + %796 = bitcast i32 %795 to <2 x bfloat>, !dbg !23 + %797 = extractvalue { i32, i32, i32, i32 } %790, 3, !dbg !23 + %798 = bitcast i32 %797 to <2 x bfloat>, !dbg !23 + %799 = extractelement <2 x bfloat> %792, i64 0, !dbg !23 + %800 = extractelement <2 x bfloat> %792, i64 1, !dbg !23 + %801 = extractelement <2 x bfloat> %794, i64 0, !dbg !23 + %802 = extractelement <2 x bfloat> %794, i64 1, !dbg !23 + %803 = extractelement <2 x bfloat> %796, i64 0, !dbg !23 + %804 = extractelement <2 x bfloat> %796, i64 1, !dbg !23 + %805 = extractelement <2 x bfloat> %798, i64 0, !dbg !23 + %806 = extractelement <2 x bfloat> %798, i64 1, !dbg !23 + %807 = extractelement <64 x i1> %43, i64 35, !dbg !23 + %808 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %267, i1 %807) #7, !dbg !23 + %809 = extractvalue { i32, i32, i32, i32 } %808, 0, !dbg !23 + %810 = bitcast i32 %809 to <2 x bfloat>, !dbg !23 + %811 = extractvalue { i32, i32, i32, i32 } %808, 1, !dbg !23 + %812 = bitcast i32 %811 to <2 x bfloat>, !dbg !23 + %813 = extractvalue { i32, i32, i32, i32 } %808, 2, !dbg !23 + %814 = bitcast i32 %813 to <2 x bfloat>, !dbg !23 + %815 = extractvalue { i32, i32, i32, i32 } %808, 3, !dbg !23 + %816 = bitcast i32 %815 to <2 x bfloat>, !dbg !23 + %817 = extractelement <2 x bfloat> %810, i64 0, !dbg !23 + %818 = extractelement <2 x bfloat> %810, i64 1, !dbg !23 + %819 = extractelement <2 x bfloat> %812, i64 0, !dbg !23 + %820 = extractelement <2 x bfloat> %812, i64 1, !dbg !23 + %821 = extractelement <2 x bfloat> %814, i64 0, !dbg !23 + %822 = extractelement <2 x bfloat> %814, i64 1, !dbg !23 + %823 = extractelement <2 x bfloat> %816, i64 0, !dbg !23 + %824 = extractelement <2 x bfloat> %816, i64 1, !dbg !23 + %825 = extractelement <64 x i1> %43, i64 34, !dbg !23 + %826 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %268, i1 %825) #7, !dbg !23 + %827 = extractvalue { i32, i32, i32, i32 } %826, 0, !dbg !23 + %828 = bitcast i32 %827 to <2 x bfloat>, !dbg !23 + %829 = extractvalue { i32, i32, i32, i32 } %826, 1, !dbg !23 + %830 = bitcast i32 %829 to <2 x bfloat>, !dbg !23 + %831 = extractvalue { i32, i32, i32, i32 } %826, 2, !dbg !23 + %832 = bitcast i32 %831 to <2 x bfloat>, !dbg !23 + %833 = extractvalue { i32, i32, i32, i32 } %826, 3, !dbg !23 + %834 = bitcast i32 %833 to <2 x bfloat>, !dbg !23 + %835 = extractelement <2 x bfloat> %828, i64 0, !dbg !23 + %836 = extractelement <2 x bfloat> %828, i64 1, !dbg !23 + %837 = extractelement <2 x bfloat> %830, i64 0, !dbg !23 + %838 = extractelement <2 x bfloat> %830, i64 1, !dbg !23 + %839 = extractelement <2 x bfloat> %832, i64 0, !dbg !23 + %840 = extractelement <2 x bfloat> %832, i64 1, !dbg !23 + %841 = extractelement <2 x bfloat> %834, i64 0, !dbg !23 + %842 = extractelement <2 x bfloat> %834, i64 1, !dbg !23 + %843 = extractelement <64 x i1> %43, i64 33, !dbg !23 + %844 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %269, i1 %843) #7, !dbg !23 + %845 = extractvalue { i32, i32, i32, i32 } %844, 0, !dbg !23 + %846 = bitcast i32 %845 to <2 x bfloat>, !dbg !23 + %847 = extractvalue { i32, i32, i32, i32 } %844, 1, !dbg !23 + %848 = bitcast i32 %847 to <2 x bfloat>, !dbg !23 + %849 = extractvalue { i32, i32, i32, i32 } %844, 2, !dbg !23 + %850 = bitcast i32 %849 to <2 x bfloat>, !dbg !23 + %851 = extractvalue { i32, i32, i32, i32 } %844, 3, !dbg !23 + %852 = bitcast i32 %851 to <2 x bfloat>, !dbg !23 + %853 = extractelement <2 x bfloat> %846, i64 0, !dbg !23 + %854 = extractelement <2 x bfloat> %846, i64 1, !dbg !23 + %855 = extractelement <2 x bfloat> %848, i64 0, !dbg !23 + %856 = extractelement <2 x bfloat> %848, i64 1, !dbg !23 + %857 = extractelement <2 x bfloat> %850, i64 0, !dbg !23 + %858 = extractelement <2 x bfloat> %850, i64 1, !dbg !23 + %859 = extractelement <2 x bfloat> %852, i64 0, !dbg !23 + %860 = extractelement <2 x bfloat> %852, i64 1, !dbg !23 + %861 = extractelement <64 x i1> %43, i64 32, !dbg !23 + %862 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %270, i1 %861) #7, !dbg !23 + %863 = extractvalue { i32, i32, i32, i32 } %862, 0, !dbg !23 + %864 = bitcast i32 %863 to <2 x bfloat>, !dbg !23 + %865 = extractvalue { i32, i32, i32, i32 } %862, 1, !dbg !23 + %866 = bitcast i32 %865 to <2 x bfloat>, !dbg !23 + %867 = extractvalue { i32, i32, i32, i32 } %862, 2, !dbg !23 + %868 = bitcast i32 %867 to <2 x bfloat>, !dbg !23 + %869 = extractvalue { i32, i32, i32, i32 } %862, 3, !dbg !23 + %870 = bitcast i32 %869 to <2 x bfloat>, !dbg !23 + %871 = extractelement <2 x bfloat> %864, i64 0, !dbg !23 + %872 = extractelement <2 x bfloat> %864, i64 1, !dbg !23 + %873 = extractelement <2 x bfloat> %866, i64 0, !dbg !23 + %874 = extractelement <2 x bfloat> %866, i64 1, !dbg !23 + %875 = extractelement <2 x bfloat> %868, i64 0, !dbg !23 + %876 = extractelement <2 x bfloat> %868, i64 1, !dbg !23 + %877 = extractelement <2 x bfloat> %870, i64 0, !dbg !23 + %878 = extractelement <2 x bfloat> %870, i64 1, !dbg !23 + %879 = extractelement <64 x i1> %43, i64 31, !dbg !23 + %880 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %271, i1 %879) #7, !dbg !23 + %881 = extractvalue { i32, i32, i32, i32 } %880, 0, !dbg !23 + %882 = bitcast i32 %881 to <2 x bfloat>, !dbg !23 + %883 = extractvalue { i32, i32, i32, i32 } %880, 1, !dbg !23 + %884 = bitcast i32 %883 to <2 x bfloat>, !dbg !23 + %885 = extractvalue { i32, i32, i32, i32 } %880, 2, !dbg !23 + %886 = bitcast i32 %885 to <2 x bfloat>, !dbg !23 + %887 = extractvalue { i32, i32, i32, i32 } %880, 3, !dbg !23 + %888 = bitcast i32 %887 to <2 x bfloat>, !dbg !23 + %889 = extractelement <2 x bfloat> %882, i64 0, !dbg !23 + %890 = extractelement <2 x bfloat> %882, i64 1, !dbg !23 + %891 = extractelement <2 x bfloat> %884, i64 0, !dbg !23 + %892 = extractelement <2 x bfloat> %884, i64 1, !dbg !23 + %893 = extractelement <2 x bfloat> %886, i64 0, !dbg !23 + %894 = extractelement <2 x bfloat> %886, i64 1, !dbg !23 + %895 = extractelement <2 x bfloat> %888, i64 0, !dbg !23 + %896 = extractelement <2 x bfloat> %888, i64 1, !dbg !23 + %897 = extractelement <64 x i1> %43, i64 30, !dbg !23 + %898 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %272, i1 %897) #7, !dbg !23 + %899 = extractvalue { i32, i32, i32, i32 } %898, 0, !dbg !23 + %900 = bitcast i32 %899 to <2 x bfloat>, !dbg !23 + %901 = extractvalue { i32, i32, i32, i32 } %898, 1, !dbg !23 + %902 = bitcast i32 %901 to <2 x bfloat>, !dbg !23 + %903 = extractvalue { i32, i32, i32, i32 } %898, 2, !dbg !23 + %904 = bitcast i32 %903 to <2 x bfloat>, !dbg !23 + %905 = extractvalue { i32, i32, i32, i32 } %898, 3, !dbg !23 + %906 = bitcast i32 %905 to <2 x bfloat>, !dbg !23 + %907 = extractelement <2 x bfloat> %900, i64 0, !dbg !23 + %908 = extractelement <2 x bfloat> %900, i64 1, !dbg !23 + %909 = extractelement <2 x bfloat> %902, i64 0, !dbg !23 + %910 = extractelement <2 x bfloat> %902, i64 1, !dbg !23 + %911 = extractelement <2 x bfloat> %904, i64 0, !dbg !23 + %912 = extractelement <2 x bfloat> %904, i64 1, !dbg !23 + %913 = extractelement <2 x bfloat> %906, i64 0, !dbg !23 + %914 = extractelement <2 x bfloat> %906, i64 1, !dbg !23 + %915 = extractelement <64 x i1> %43, i64 29, !dbg !23 + %916 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %273, i1 %915) #7, !dbg !23 + %917 = extractvalue { i32, i32, i32, i32 } %916, 0, !dbg !23 + %918 = bitcast i32 %917 to <2 x bfloat>, !dbg !23 + %919 = extractvalue { i32, i32, i32, i32 } %916, 1, !dbg !23 + %920 = bitcast i32 %919 to <2 x bfloat>, !dbg !23 + %921 = extractvalue { i32, i32, i32, i32 } %916, 2, !dbg !23 + %922 = bitcast i32 %921 to <2 x bfloat>, !dbg !23 + %923 = extractvalue { i32, i32, i32, i32 } %916, 3, !dbg !23 + %924 = bitcast i32 %923 to <2 x bfloat>, !dbg !23 + %925 = extractelement <2 x bfloat> %918, i64 0, !dbg !23 + %926 = extractelement <2 x bfloat> %918, i64 1, !dbg !23 + %927 = extractelement <2 x bfloat> %920, i64 0, !dbg !23 + %928 = extractelement <2 x bfloat> %920, i64 1, !dbg !23 + %929 = extractelement <2 x bfloat> %922, i64 0, !dbg !23 + %930 = extractelement <2 x bfloat> %922, i64 1, !dbg !23 + %931 = extractelement <2 x bfloat> %924, i64 0, !dbg !23 + %932 = extractelement <2 x bfloat> %924, i64 1, !dbg !23 + %933 = extractelement <64 x i1> %43, i64 28, !dbg !23 + %934 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %274, i1 %933) #7, !dbg !23 + %935 = extractvalue { i32, i32, i32, i32 } %934, 0, !dbg !23 + %936 = bitcast i32 %935 to <2 x bfloat>, !dbg !23 + %937 = extractvalue { i32, i32, i32, i32 } %934, 1, !dbg !23 + %938 = bitcast i32 %937 to <2 x bfloat>, !dbg !23 + %939 = extractvalue { i32, i32, i32, i32 } %934, 2, !dbg !23 + %940 = bitcast i32 %939 to <2 x bfloat>, !dbg !23 + %941 = extractvalue { i32, i32, i32, i32 } %934, 3, !dbg !23 + %942 = bitcast i32 %941 to <2 x bfloat>, !dbg !23 + %943 = extractelement <2 x bfloat> %936, i64 0, !dbg !23 + %944 = extractelement <2 x bfloat> %936, i64 1, !dbg !23 + %945 = extractelement <2 x bfloat> %938, i64 0, !dbg !23 + %946 = extractelement <2 x bfloat> %938, i64 1, !dbg !23 + %947 = extractelement <2 x bfloat> %940, i64 0, !dbg !23 + %948 = extractelement <2 x bfloat> %940, i64 1, !dbg !23 + %949 = extractelement <2 x bfloat> %942, i64 0, !dbg !23 + %950 = extractelement <2 x bfloat> %942, i64 1, !dbg !23 + %951 = extractelement <64 x i1> %43, i64 27, !dbg !23 + %952 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %275, i1 %951) #7, !dbg !23 + %953 = extractvalue { i32, i32, i32, i32 } %952, 0, !dbg !23 + %954 = bitcast i32 %953 to <2 x bfloat>, !dbg !23 + %955 = extractvalue { i32, i32, i32, i32 } %952, 1, !dbg !23 + %956 = bitcast i32 %955 to <2 x bfloat>, !dbg !23 + %957 = extractvalue { i32, i32, i32, i32 } %952, 2, !dbg !23 + %958 = bitcast i32 %957 to <2 x bfloat>, !dbg !23 + %959 = extractvalue { i32, i32, i32, i32 } %952, 3, !dbg !23 + %960 = bitcast i32 %959 to <2 x bfloat>, !dbg !23 + %961 = extractelement <2 x bfloat> %954, i64 0, !dbg !23 + %962 = extractelement <2 x bfloat> %954, i64 1, !dbg !23 + %963 = extractelement <2 x bfloat> %956, i64 0, !dbg !23 + %964 = extractelement <2 x bfloat> %956, i64 1, !dbg !23 + %965 = extractelement <2 x bfloat> %958, i64 0, !dbg !23 + %966 = extractelement <2 x bfloat> %958, i64 1, !dbg !23 + %967 = extractelement <2 x bfloat> %960, i64 0, !dbg !23 + %968 = extractelement <2 x bfloat> %960, i64 1, !dbg !23 + %969 = extractelement <64 x i1> %43, i64 26, !dbg !23 + %970 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %276, i1 %969) #7, !dbg !23 + %971 = extractvalue { i32, i32, i32, i32 } %970, 0, !dbg !23 + %972 = bitcast i32 %971 to <2 x bfloat>, !dbg !23 + %973 = extractvalue { i32, i32, i32, i32 } %970, 1, !dbg !23 + %974 = bitcast i32 %973 to <2 x bfloat>, !dbg !23 + %975 = extractvalue { i32, i32, i32, i32 } %970, 2, !dbg !23 + %976 = bitcast i32 %975 to <2 x bfloat>, !dbg !23 + %977 = extractvalue { i32, i32, i32, i32 } %970, 3, !dbg !23 + %978 = bitcast i32 %977 to <2 x bfloat>, !dbg !23 + %979 = extractelement <2 x bfloat> %972, i64 0, !dbg !23 + %980 = extractelement <2 x bfloat> %972, i64 1, !dbg !23 + %981 = extractelement <2 x bfloat> %974, i64 0, !dbg !23 + %982 = extractelement <2 x bfloat> %974, i64 1, !dbg !23 + %983 = extractelement <2 x bfloat> %976, i64 0, !dbg !23 + %984 = extractelement <2 x bfloat> %976, i64 1, !dbg !23 + %985 = extractelement <2 x bfloat> %978, i64 0, !dbg !23 + %986 = extractelement <2 x bfloat> %978, i64 1, !dbg !23 + %987 = extractelement <64 x i1> %43, i64 25, !dbg !23 + %988 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %277, i1 %987) #7, !dbg !23 + %989 = extractvalue { i32, i32, i32, i32 } %988, 0, !dbg !23 + %990 = bitcast i32 %989 to <2 x bfloat>, !dbg !23 + %991 = extractvalue { i32, i32, i32, i32 } %988, 1, !dbg !23 + %992 = bitcast i32 %991 to <2 x bfloat>, !dbg !23 + %993 = extractvalue { i32, i32, i32, i32 } %988, 2, !dbg !23 + %994 = bitcast i32 %993 to <2 x bfloat>, !dbg !23 + %995 = extractvalue { i32, i32, i32, i32 } %988, 3, !dbg !23 + %996 = bitcast i32 %995 to <2 x bfloat>, !dbg !23 + %997 = extractelement <2 x bfloat> %990, i64 0, !dbg !23 + %998 = extractelement <2 x bfloat> %990, i64 1, !dbg !23 + %999 = extractelement <2 x bfloat> %992, i64 0, !dbg !23 + %1000 = extractelement <2 x bfloat> %992, i64 1, !dbg !23 + %1001 = extractelement <2 x bfloat> %994, i64 0, !dbg !23 + %1002 = extractelement <2 x bfloat> %994, i64 1, !dbg !23 + %1003 = extractelement <2 x bfloat> %996, i64 0, !dbg !23 + %1004 = extractelement <2 x bfloat> %996, i64 1, !dbg !23 + %1005 = extractelement <64 x i1> %43, i64 24, !dbg !23 + %1006 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %278, i1 %1005) #7, !dbg !23 + %1007 = extractvalue { i32, i32, i32, i32 } %1006, 0, !dbg !23 + %1008 = bitcast i32 %1007 to <2 x bfloat>, !dbg !23 + %1009 = extractvalue { i32, i32, i32, i32 } %1006, 1, !dbg !23 + %1010 = bitcast i32 %1009 to <2 x bfloat>, !dbg !23 + %1011 = extractvalue { i32, i32, i32, i32 } %1006, 2, !dbg !23 + %1012 = bitcast i32 %1011 to <2 x bfloat>, !dbg !23 + %1013 = extractvalue { i32, i32, i32, i32 } %1006, 3, !dbg !23 + %1014 = bitcast i32 %1013 to <2 x bfloat>, !dbg !23 + %1015 = extractelement <2 x bfloat> %1008, i64 0, !dbg !23 + %1016 = extractelement <2 x bfloat> %1008, i64 1, !dbg !23 + %1017 = extractelement <2 x bfloat> %1010, i64 0, !dbg !23 + %1018 = extractelement <2 x bfloat> %1010, i64 1, !dbg !23 + %1019 = extractelement <2 x bfloat> %1012, i64 0, !dbg !23 + %1020 = extractelement <2 x bfloat> %1012, i64 1, !dbg !23 + %1021 = extractelement <2 x bfloat> %1014, i64 0, !dbg !23 + %1022 = extractelement <2 x bfloat> %1014, i64 1, !dbg !23 + %1023 = extractelement <64 x i1> %43, i64 23, !dbg !23 + %1024 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %279, i1 %1023) #7, !dbg !23 + %1025 = extractvalue { i32, i32, i32, i32 } %1024, 0, !dbg !23 + %1026 = bitcast i32 %1025 to <2 x bfloat>, !dbg !23 + %1027 = extractvalue { i32, i32, i32, i32 } %1024, 1, !dbg !23 + %1028 = bitcast i32 %1027 to <2 x bfloat>, !dbg !23 + %1029 = extractvalue { i32, i32, i32, i32 } %1024, 2, !dbg !23 + %1030 = bitcast i32 %1029 to <2 x bfloat>, !dbg !23 + %1031 = extractvalue { i32, i32, i32, i32 } %1024, 3, !dbg !23 + %1032 = bitcast i32 %1031 to <2 x bfloat>, !dbg !23 + %1033 = extractelement <2 x bfloat> %1026, i64 0, !dbg !23 + %1034 = extractelement <2 x bfloat> %1026, i64 1, !dbg !23 + %1035 = extractelement <2 x bfloat> %1028, i64 0, !dbg !23 + %1036 = extractelement <2 x bfloat> %1028, i64 1, !dbg !23 + %1037 = extractelement <2 x bfloat> %1030, i64 0, !dbg !23 + %1038 = extractelement <2 x bfloat> %1030, i64 1, !dbg !23 + %1039 = extractelement <2 x bfloat> %1032, i64 0, !dbg !23 + %1040 = extractelement <2 x bfloat> %1032, i64 1, !dbg !23 + %1041 = extractelement <64 x i1> %43, i64 22, !dbg !23 + %1042 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %280, i1 %1041) #7, !dbg !23 + %1043 = extractvalue { i32, i32, i32, i32 } %1042, 0, !dbg !23 + %1044 = bitcast i32 %1043 to <2 x bfloat>, !dbg !23 + %1045 = extractvalue { i32, i32, i32, i32 } %1042, 1, !dbg !23 + %1046 = bitcast i32 %1045 to <2 x bfloat>, !dbg !23 + %1047 = extractvalue { i32, i32, i32, i32 } %1042, 2, !dbg !23 + %1048 = bitcast i32 %1047 to <2 x bfloat>, !dbg !23 + %1049 = extractvalue { i32, i32, i32, i32 } %1042, 3, !dbg !23 + %1050 = bitcast i32 %1049 to <2 x bfloat>, !dbg !23 + %1051 = extractelement <2 x bfloat> %1044, i64 0, !dbg !23 + %1052 = extractelement <2 x bfloat> %1044, i64 1, !dbg !23 + %1053 = extractelement <2 x bfloat> %1046, i64 0, !dbg !23 + %1054 = extractelement <2 x bfloat> %1046, i64 1, !dbg !23 + %1055 = extractelement <2 x bfloat> %1048, i64 0, !dbg !23 + %1056 = extractelement <2 x bfloat> %1048, i64 1, !dbg !23 + %1057 = extractelement <2 x bfloat> %1050, i64 0, !dbg !23 + %1058 = extractelement <2 x bfloat> %1050, i64 1, !dbg !23 + %1059 = extractelement <64 x i1> %43, i64 21, !dbg !23 + %1060 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %281, i1 %1059) #7, !dbg !23 + %1061 = extractvalue { i32, i32, i32, i32 } %1060, 0, !dbg !23 + %1062 = bitcast i32 %1061 to <2 x bfloat>, !dbg !23 + %1063 = extractvalue { i32, i32, i32, i32 } %1060, 1, !dbg !23 + %1064 = bitcast i32 %1063 to <2 x bfloat>, !dbg !23 + %1065 = extractvalue { i32, i32, i32, i32 } %1060, 2, !dbg !23 + %1066 = bitcast i32 %1065 to <2 x bfloat>, !dbg !23 + %1067 = extractvalue { i32, i32, i32, i32 } %1060, 3, !dbg !23 + %1068 = bitcast i32 %1067 to <2 x bfloat>, !dbg !23 + %1069 = extractelement <2 x bfloat> %1062, i64 0, !dbg !23 + %1070 = extractelement <2 x bfloat> %1062, i64 1, !dbg !23 + %1071 = extractelement <2 x bfloat> %1064, i64 0, !dbg !23 + %1072 = extractelement <2 x bfloat> %1064, i64 1, !dbg !23 + %1073 = extractelement <2 x bfloat> %1066, i64 0, !dbg !23 + %1074 = extractelement <2 x bfloat> %1066, i64 1, !dbg !23 + %1075 = extractelement <2 x bfloat> %1068, i64 0, !dbg !23 + %1076 = extractelement <2 x bfloat> %1068, i64 1, !dbg !23 + %1077 = extractelement <64 x i1> %43, i64 20, !dbg !23 + %1078 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %282, i1 %1077) #7, !dbg !23 + %1079 = extractvalue { i32, i32, i32, i32 } %1078, 0, !dbg !23 + %1080 = bitcast i32 %1079 to <2 x bfloat>, !dbg !23 + %1081 = extractvalue { i32, i32, i32, i32 } %1078, 1, !dbg !23 + %1082 = bitcast i32 %1081 to <2 x bfloat>, !dbg !23 + %1083 = extractvalue { i32, i32, i32, i32 } %1078, 2, !dbg !23 + %1084 = bitcast i32 %1083 to <2 x bfloat>, !dbg !23 + %1085 = extractvalue { i32, i32, i32, i32 } %1078, 3, !dbg !23 + %1086 = bitcast i32 %1085 to <2 x bfloat>, !dbg !23 + %1087 = extractelement <2 x bfloat> %1080, i64 0, !dbg !23 + %1088 = extractelement <2 x bfloat> %1080, i64 1, !dbg !23 + %1089 = extractelement <2 x bfloat> %1082, i64 0, !dbg !23 + %1090 = extractelement <2 x bfloat> %1082, i64 1, !dbg !23 + %1091 = extractelement <2 x bfloat> %1084, i64 0, !dbg !23 + %1092 = extractelement <2 x bfloat> %1084, i64 1, !dbg !23 + %1093 = extractelement <2 x bfloat> %1086, i64 0, !dbg !23 + %1094 = extractelement <2 x bfloat> %1086, i64 1, !dbg !23 + %1095 = extractelement <64 x i1> %43, i64 19, !dbg !23 + %1096 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %283, i1 %1095) #7, !dbg !23 + %1097 = extractvalue { i32, i32, i32, i32 } %1096, 0, !dbg !23 + %1098 = bitcast i32 %1097 to <2 x bfloat>, !dbg !23 + %1099 = extractvalue { i32, i32, i32, i32 } %1096, 1, !dbg !23 + %1100 = bitcast i32 %1099 to <2 x bfloat>, !dbg !23 + %1101 = extractvalue { i32, i32, i32, i32 } %1096, 2, !dbg !23 + %1102 = bitcast i32 %1101 to <2 x bfloat>, !dbg !23 + %1103 = extractvalue { i32, i32, i32, i32 } %1096, 3, !dbg !23 + %1104 = bitcast i32 %1103 to <2 x bfloat>, !dbg !23 + %1105 = extractelement <2 x bfloat> %1098, i64 0, !dbg !23 + %1106 = extractelement <2 x bfloat> %1098, i64 1, !dbg !23 + %1107 = extractelement <2 x bfloat> %1100, i64 0, !dbg !23 + %1108 = extractelement <2 x bfloat> %1100, i64 1, !dbg !23 + %1109 = extractelement <2 x bfloat> %1102, i64 0, !dbg !23 + %1110 = extractelement <2 x bfloat> %1102, i64 1, !dbg !23 + %1111 = extractelement <2 x bfloat> %1104, i64 0, !dbg !23 + %1112 = extractelement <2 x bfloat> %1104, i64 1, !dbg !23 + %1113 = extractelement <64 x i1> %43, i64 18, !dbg !23 + %1114 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %284, i1 %1113) #7, !dbg !23 + %1115 = extractvalue { i32, i32, i32, i32 } %1114, 0, !dbg !23 + %1116 = bitcast i32 %1115 to <2 x bfloat>, !dbg !23 + %1117 = extractvalue { i32, i32, i32, i32 } %1114, 1, !dbg !23 + %1118 = bitcast i32 %1117 to <2 x bfloat>, !dbg !23 + %1119 = extractvalue { i32, i32, i32, i32 } %1114, 2, !dbg !23 + %1120 = bitcast i32 %1119 to <2 x bfloat>, !dbg !23 + %1121 = extractvalue { i32, i32, i32, i32 } %1114, 3, !dbg !23 + %1122 = bitcast i32 %1121 to <2 x bfloat>, !dbg !23 + %1123 = extractelement <2 x bfloat> %1116, i64 0, !dbg !23 + %1124 = extractelement <2 x bfloat> %1116, i64 1, !dbg !23 + %1125 = extractelement <2 x bfloat> %1118, i64 0, !dbg !23 + %1126 = extractelement <2 x bfloat> %1118, i64 1, !dbg !23 + %1127 = extractelement <2 x bfloat> %1120, i64 0, !dbg !23 + %1128 = extractelement <2 x bfloat> %1120, i64 1, !dbg !23 + %1129 = extractelement <2 x bfloat> %1122, i64 0, !dbg !23 + %1130 = extractelement <2 x bfloat> %1122, i64 1, !dbg !23 + %1131 = extractelement <64 x i1> %43, i64 17, !dbg !23 + %1132 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %285, i1 %1131) #7, !dbg !23 + %1133 = extractvalue { i32, i32, i32, i32 } %1132, 0, !dbg !23 + %1134 = bitcast i32 %1133 to <2 x bfloat>, !dbg !23 + %1135 = extractvalue { i32, i32, i32, i32 } %1132, 1, !dbg !23 + %1136 = bitcast i32 %1135 to <2 x bfloat>, !dbg !23 + %1137 = extractvalue { i32, i32, i32, i32 } %1132, 2, !dbg !23 + %1138 = bitcast i32 %1137 to <2 x bfloat>, !dbg !23 + %1139 = extractvalue { i32, i32, i32, i32 } %1132, 3, !dbg !23 + %1140 = bitcast i32 %1139 to <2 x bfloat>, !dbg !23 + %1141 = extractelement <2 x bfloat> %1134, i64 0, !dbg !23 + %1142 = extractelement <2 x bfloat> %1134, i64 1, !dbg !23 + %1143 = extractelement <2 x bfloat> %1136, i64 0, !dbg !23 + %1144 = extractelement <2 x bfloat> %1136, i64 1, !dbg !23 + %1145 = extractelement <2 x bfloat> %1138, i64 0, !dbg !23 + %1146 = extractelement <2 x bfloat> %1138, i64 1, !dbg !23 + %1147 = extractelement <2 x bfloat> %1140, i64 0, !dbg !23 + %1148 = extractelement <2 x bfloat> %1140, i64 1, !dbg !23 + %1149 = extractelement <64 x i1> %43, i64 16, !dbg !23 + %1150 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %286, i1 %1149) #7, !dbg !23 + %1151 = extractvalue { i32, i32, i32, i32 } %1150, 0, !dbg !23 + %1152 = bitcast i32 %1151 to <2 x bfloat>, !dbg !23 + %1153 = extractvalue { i32, i32, i32, i32 } %1150, 1, !dbg !23 + %1154 = bitcast i32 %1153 to <2 x bfloat>, !dbg !23 + %1155 = extractvalue { i32, i32, i32, i32 } %1150, 2, !dbg !23 + %1156 = bitcast i32 %1155 to <2 x bfloat>, !dbg !23 + %1157 = extractvalue { i32, i32, i32, i32 } %1150, 3, !dbg !23 + %1158 = bitcast i32 %1157 to <2 x bfloat>, !dbg !23 + %1159 = extractelement <2 x bfloat> %1152, i64 0, !dbg !23 + %1160 = extractelement <2 x bfloat> %1152, i64 1, !dbg !23 + %1161 = extractelement <2 x bfloat> %1154, i64 0, !dbg !23 + %1162 = extractelement <2 x bfloat> %1154, i64 1, !dbg !23 + %1163 = extractelement <2 x bfloat> %1156, i64 0, !dbg !23 + %1164 = extractelement <2 x bfloat> %1156, i64 1, !dbg !23 + %1165 = extractelement <2 x bfloat> %1158, i64 0, !dbg !23 + %1166 = extractelement <2 x bfloat> %1158, i64 1, !dbg !23 + %1167 = extractelement <64 x i1> %43, i64 15, !dbg !23 + %1168 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %287, i1 %1167) #7, !dbg !23 + %1169 = extractvalue { i32, i32, i32, i32 } %1168, 0, !dbg !23 + %1170 = bitcast i32 %1169 to <2 x bfloat>, !dbg !23 + %1171 = extractvalue { i32, i32, i32, i32 } %1168, 1, !dbg !23 + %1172 = bitcast i32 %1171 to <2 x bfloat>, !dbg !23 + %1173 = extractvalue { i32, i32, i32, i32 } %1168, 2, !dbg !23 + %1174 = bitcast i32 %1173 to <2 x bfloat>, !dbg !23 + %1175 = extractvalue { i32, i32, i32, i32 } %1168, 3, !dbg !23 + %1176 = bitcast i32 %1175 to <2 x bfloat>, !dbg !23 + %1177 = extractelement <2 x bfloat> %1170, i64 0, !dbg !23 + %1178 = extractelement <2 x bfloat> %1170, i64 1, !dbg !23 + %1179 = extractelement <2 x bfloat> %1172, i64 0, !dbg !23 + %1180 = extractelement <2 x bfloat> %1172, i64 1, !dbg !23 + %1181 = extractelement <2 x bfloat> %1174, i64 0, !dbg !23 + %1182 = extractelement <2 x bfloat> %1174, i64 1, !dbg !23 + %1183 = extractelement <2 x bfloat> %1176, i64 0, !dbg !23 + %1184 = extractelement <2 x bfloat> %1176, i64 1, !dbg !23 + %1185 = extractelement <64 x i1> %43, i64 14, !dbg !23 + %1186 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %288, i1 %1185) #7, !dbg !23 + %1187 = extractvalue { i32, i32, i32, i32 } %1186, 0, !dbg !23 + %1188 = bitcast i32 %1187 to <2 x bfloat>, !dbg !23 + %1189 = extractvalue { i32, i32, i32, i32 } %1186, 1, !dbg !23 + %1190 = bitcast i32 %1189 to <2 x bfloat>, !dbg !23 + %1191 = extractvalue { i32, i32, i32, i32 } %1186, 2, !dbg !23 + %1192 = bitcast i32 %1191 to <2 x bfloat>, !dbg !23 + %1193 = extractvalue { i32, i32, i32, i32 } %1186, 3, !dbg !23 + %1194 = bitcast i32 %1193 to <2 x bfloat>, !dbg !23 + %1195 = extractelement <2 x bfloat> %1188, i64 0, !dbg !23 + %1196 = extractelement <2 x bfloat> %1188, i64 1, !dbg !23 + %1197 = extractelement <2 x bfloat> %1190, i64 0, !dbg !23 + %1198 = extractelement <2 x bfloat> %1190, i64 1, !dbg !23 + %1199 = extractelement <2 x bfloat> %1192, i64 0, !dbg !23 + %1200 = extractelement <2 x bfloat> %1192, i64 1, !dbg !23 + %1201 = extractelement <2 x bfloat> %1194, i64 0, !dbg !23 + %1202 = extractelement <2 x bfloat> %1194, i64 1, !dbg !23 + %1203 = extractelement <64 x i1> %43, i64 13, !dbg !23 + %1204 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %289, i1 %1203) #7, !dbg !23 + %1205 = extractvalue { i32, i32, i32, i32 } %1204, 0, !dbg !23 + %1206 = bitcast i32 %1205 to <2 x bfloat>, !dbg !23 + %1207 = extractvalue { i32, i32, i32, i32 } %1204, 1, !dbg !23 + %1208 = bitcast i32 %1207 to <2 x bfloat>, !dbg !23 + %1209 = extractvalue { i32, i32, i32, i32 } %1204, 2, !dbg !23 + %1210 = bitcast i32 %1209 to <2 x bfloat>, !dbg !23 + %1211 = extractvalue { i32, i32, i32, i32 } %1204, 3, !dbg !23 + %1212 = bitcast i32 %1211 to <2 x bfloat>, !dbg !23 + %1213 = extractelement <2 x bfloat> %1206, i64 0, !dbg !23 + %1214 = extractelement <2 x bfloat> %1206, i64 1, !dbg !23 + %1215 = extractelement <2 x bfloat> %1208, i64 0, !dbg !23 + %1216 = extractelement <2 x bfloat> %1208, i64 1, !dbg !23 + %1217 = extractelement <2 x bfloat> %1210, i64 0, !dbg !23 + %1218 = extractelement <2 x bfloat> %1210, i64 1, !dbg !23 + %1219 = extractelement <2 x bfloat> %1212, i64 0, !dbg !23 + %1220 = extractelement <2 x bfloat> %1212, i64 1, !dbg !23 + %1221 = extractelement <64 x i1> %43, i64 12, !dbg !23 + %1222 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %290, i1 %1221) #7, !dbg !23 + %1223 = extractvalue { i32, i32, i32, i32 } %1222, 0, !dbg !23 + %1224 = bitcast i32 %1223 to <2 x bfloat>, !dbg !23 + %1225 = extractvalue { i32, i32, i32, i32 } %1222, 1, !dbg !23 + %1226 = bitcast i32 %1225 to <2 x bfloat>, !dbg !23 + %1227 = extractvalue { i32, i32, i32, i32 } %1222, 2, !dbg !23 + %1228 = bitcast i32 %1227 to <2 x bfloat>, !dbg !23 + %1229 = extractvalue { i32, i32, i32, i32 } %1222, 3, !dbg !23 + %1230 = bitcast i32 %1229 to <2 x bfloat>, !dbg !23 + %1231 = extractelement <2 x bfloat> %1224, i64 0, !dbg !23 + %1232 = extractelement <2 x bfloat> %1224, i64 1, !dbg !23 + %1233 = extractelement <2 x bfloat> %1226, i64 0, !dbg !23 + %1234 = extractelement <2 x bfloat> %1226, i64 1, !dbg !23 + %1235 = extractelement <2 x bfloat> %1228, i64 0, !dbg !23 + %1236 = extractelement <2 x bfloat> %1228, i64 1, !dbg !23 + %1237 = extractelement <2 x bfloat> %1230, i64 0, !dbg !23 + %1238 = extractelement <2 x bfloat> %1230, i64 1, !dbg !23 + %1239 = extractelement <64 x i1> %43, i64 11, !dbg !23 + %1240 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %291, i1 %1239) #7, !dbg !23 + %1241 = extractvalue { i32, i32, i32, i32 } %1240, 0, !dbg !23 + %1242 = bitcast i32 %1241 to <2 x bfloat>, !dbg !23 + %1243 = extractvalue { i32, i32, i32, i32 } %1240, 1, !dbg !23 + %1244 = bitcast i32 %1243 to <2 x bfloat>, !dbg !23 + %1245 = extractvalue { i32, i32, i32, i32 } %1240, 2, !dbg !23 + %1246 = bitcast i32 %1245 to <2 x bfloat>, !dbg !23 + %1247 = extractvalue { i32, i32, i32, i32 } %1240, 3, !dbg !23 + %1248 = bitcast i32 %1247 to <2 x bfloat>, !dbg !23 + %1249 = extractelement <2 x bfloat> %1242, i64 0, !dbg !23 + %1250 = extractelement <2 x bfloat> %1242, i64 1, !dbg !23 + %1251 = extractelement <2 x bfloat> %1244, i64 0, !dbg !23 + %1252 = extractelement <2 x bfloat> %1244, i64 1, !dbg !23 + %1253 = extractelement <2 x bfloat> %1246, i64 0, !dbg !23 + %1254 = extractelement <2 x bfloat> %1246, i64 1, !dbg !23 + %1255 = extractelement <2 x bfloat> %1248, i64 0, !dbg !23 + %1256 = extractelement <2 x bfloat> %1248, i64 1, !dbg !23 + %1257 = extractelement <64 x i1> %43, i64 10, !dbg !23 + %1258 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %292, i1 %1257) #7, !dbg !23 + %1259 = extractvalue { i32, i32, i32, i32 } %1258, 0, !dbg !23 + %1260 = bitcast i32 %1259 to <2 x bfloat>, !dbg !23 + %1261 = extractvalue { i32, i32, i32, i32 } %1258, 1, !dbg !23 + %1262 = bitcast i32 %1261 to <2 x bfloat>, !dbg !23 + %1263 = extractvalue { i32, i32, i32, i32 } %1258, 2, !dbg !23 + %1264 = bitcast i32 %1263 to <2 x bfloat>, !dbg !23 + %1265 = extractvalue { i32, i32, i32, i32 } %1258, 3, !dbg !23 + %1266 = bitcast i32 %1265 to <2 x bfloat>, !dbg !23 + %1267 = extractelement <2 x bfloat> %1260, i64 0, !dbg !23 + %1268 = extractelement <2 x bfloat> %1260, i64 1, !dbg !23 + %1269 = extractelement <2 x bfloat> %1262, i64 0, !dbg !23 + %1270 = extractelement <2 x bfloat> %1262, i64 1, !dbg !23 + %1271 = extractelement <2 x bfloat> %1264, i64 0, !dbg !23 + %1272 = extractelement <2 x bfloat> %1264, i64 1, !dbg !23 + %1273 = extractelement <2 x bfloat> %1266, i64 0, !dbg !23 + %1274 = extractelement <2 x bfloat> %1266, i64 1, !dbg !23 + %1275 = extractelement <64 x i1> %43, i64 9, !dbg !23 + %1276 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %293, i1 %1275) #7, !dbg !23 + %1277 = extractvalue { i32, i32, i32, i32 } %1276, 0, !dbg !23 + %1278 = bitcast i32 %1277 to <2 x bfloat>, !dbg !23 + %1279 = extractvalue { i32, i32, i32, i32 } %1276, 1, !dbg !23 + %1280 = bitcast i32 %1279 to <2 x bfloat>, !dbg !23 + %1281 = extractvalue { i32, i32, i32, i32 } %1276, 2, !dbg !23 + %1282 = bitcast i32 %1281 to <2 x bfloat>, !dbg !23 + %1283 = extractvalue { i32, i32, i32, i32 } %1276, 3, !dbg !23 + %1284 = bitcast i32 %1283 to <2 x bfloat>, !dbg !23 + %1285 = extractelement <2 x bfloat> %1278, i64 0, !dbg !23 + %1286 = extractelement <2 x bfloat> %1278, i64 1, !dbg !23 + %1287 = extractelement <2 x bfloat> %1280, i64 0, !dbg !23 + %1288 = extractelement <2 x bfloat> %1280, i64 1, !dbg !23 + %1289 = extractelement <2 x bfloat> %1282, i64 0, !dbg !23 + %1290 = extractelement <2 x bfloat> %1282, i64 1, !dbg !23 + %1291 = extractelement <2 x bfloat> %1284, i64 0, !dbg !23 + %1292 = extractelement <2 x bfloat> %1284, i64 1, !dbg !23 + %1293 = extractelement <64 x i1> %43, i64 8, !dbg !23 + %1294 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %294, i1 %1293) #7, !dbg !23 + %1295 = extractvalue { i32, i32, i32, i32 } %1294, 0, !dbg !23 + %1296 = bitcast i32 %1295 to <2 x bfloat>, !dbg !23 + %1297 = extractvalue { i32, i32, i32, i32 } %1294, 1, !dbg !23 + %1298 = bitcast i32 %1297 to <2 x bfloat>, !dbg !23 + %1299 = extractvalue { i32, i32, i32, i32 } %1294, 2, !dbg !23 + %1300 = bitcast i32 %1299 to <2 x bfloat>, !dbg !23 + %1301 = extractvalue { i32, i32, i32, i32 } %1294, 3, !dbg !23 + %1302 = bitcast i32 %1301 to <2 x bfloat>, !dbg !23 + %1303 = extractelement <2 x bfloat> %1296, i64 0, !dbg !23 + %1304 = extractelement <2 x bfloat> %1296, i64 1, !dbg !23 + %1305 = extractelement <2 x bfloat> %1298, i64 0, !dbg !23 + %1306 = extractelement <2 x bfloat> %1298, i64 1, !dbg !23 + %1307 = extractelement <2 x bfloat> %1300, i64 0, !dbg !23 + %1308 = extractelement <2 x bfloat> %1300, i64 1, !dbg !23 + %1309 = extractelement <2 x bfloat> %1302, i64 0, !dbg !23 + %1310 = extractelement <2 x bfloat> %1302, i64 1, !dbg !23 + %1311 = extractelement <64 x i1> %43, i64 7, !dbg !23 + %1312 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %295, i1 %1311) #7, !dbg !23 + %1313 = extractvalue { i32, i32, i32, i32 } %1312, 0, !dbg !23 + %1314 = bitcast i32 %1313 to <2 x bfloat>, !dbg !23 + %1315 = extractvalue { i32, i32, i32, i32 } %1312, 1, !dbg !23 + %1316 = bitcast i32 %1315 to <2 x bfloat>, !dbg !23 + %1317 = extractvalue { i32, i32, i32, i32 } %1312, 2, !dbg !23 + %1318 = bitcast i32 %1317 to <2 x bfloat>, !dbg !23 + %1319 = extractvalue { i32, i32, i32, i32 } %1312, 3, !dbg !23 + %1320 = bitcast i32 %1319 to <2 x bfloat>, !dbg !23 + %1321 = extractelement <2 x bfloat> %1314, i64 0, !dbg !23 + %1322 = extractelement <2 x bfloat> %1314, i64 1, !dbg !23 + %1323 = extractelement <2 x bfloat> %1316, i64 0, !dbg !23 + %1324 = extractelement <2 x bfloat> %1316, i64 1, !dbg !23 + %1325 = extractelement <2 x bfloat> %1318, i64 0, !dbg !23 + %1326 = extractelement <2 x bfloat> %1318, i64 1, !dbg !23 + %1327 = extractelement <2 x bfloat> %1320, i64 0, !dbg !23 + %1328 = extractelement <2 x bfloat> %1320, i64 1, !dbg !23 + %1329 = extractelement <64 x i1> %43, i64 6, !dbg !23 + %1330 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %296, i1 %1329) #7, !dbg !23 + %1331 = extractvalue { i32, i32, i32, i32 } %1330, 0, !dbg !23 + %1332 = bitcast i32 %1331 to <2 x bfloat>, !dbg !23 + %1333 = extractvalue { i32, i32, i32, i32 } %1330, 1, !dbg !23 + %1334 = bitcast i32 %1333 to <2 x bfloat>, !dbg !23 + %1335 = extractvalue { i32, i32, i32, i32 } %1330, 2, !dbg !23 + %1336 = bitcast i32 %1335 to <2 x bfloat>, !dbg !23 + %1337 = extractvalue { i32, i32, i32, i32 } %1330, 3, !dbg !23 + %1338 = bitcast i32 %1337 to <2 x bfloat>, !dbg !23 + %1339 = extractelement <2 x bfloat> %1332, i64 0, !dbg !23 + %1340 = extractelement <2 x bfloat> %1332, i64 1, !dbg !23 + %1341 = extractelement <2 x bfloat> %1334, i64 0, !dbg !23 + %1342 = extractelement <2 x bfloat> %1334, i64 1, !dbg !23 + %1343 = extractelement <2 x bfloat> %1336, i64 0, !dbg !23 + %1344 = extractelement <2 x bfloat> %1336, i64 1, !dbg !23 + %1345 = extractelement <2 x bfloat> %1338, i64 0, !dbg !23 + %1346 = extractelement <2 x bfloat> %1338, i64 1, !dbg !23 + %1347 = extractelement <64 x i1> %43, i64 5, !dbg !23 + %1348 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %297, i1 %1347) #7, !dbg !23 + %1349 = extractvalue { i32, i32, i32, i32 } %1348, 0, !dbg !23 + %1350 = bitcast i32 %1349 to <2 x bfloat>, !dbg !23 + %1351 = extractvalue { i32, i32, i32, i32 } %1348, 1, !dbg !23 + %1352 = bitcast i32 %1351 to <2 x bfloat>, !dbg !23 + %1353 = extractvalue { i32, i32, i32, i32 } %1348, 2, !dbg !23 + %1354 = bitcast i32 %1353 to <2 x bfloat>, !dbg !23 + %1355 = extractvalue { i32, i32, i32, i32 } %1348, 3, !dbg !23 + %1356 = bitcast i32 %1355 to <2 x bfloat>, !dbg !23 + %1357 = extractelement <2 x bfloat> %1350, i64 0, !dbg !23 + %1358 = extractelement <2 x bfloat> %1350, i64 1, !dbg !23 + %1359 = extractelement <2 x bfloat> %1352, i64 0, !dbg !23 + %1360 = extractelement <2 x bfloat> %1352, i64 1, !dbg !23 + %1361 = extractelement <2 x bfloat> %1354, i64 0, !dbg !23 + %1362 = extractelement <2 x bfloat> %1354, i64 1, !dbg !23 + %1363 = extractelement <2 x bfloat> %1356, i64 0, !dbg !23 + %1364 = extractelement <2 x bfloat> %1356, i64 1, !dbg !23 + %1365 = extractelement <64 x i1> %43, i64 4, !dbg !23 + %1366 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %298, i1 %1365) #7, !dbg !23 + %1367 = extractvalue { i32, i32, i32, i32 } %1366, 0, !dbg !23 + %1368 = bitcast i32 %1367 to <2 x bfloat>, !dbg !23 + %1369 = extractvalue { i32, i32, i32, i32 } %1366, 1, !dbg !23 + %1370 = bitcast i32 %1369 to <2 x bfloat>, !dbg !23 + %1371 = extractvalue { i32, i32, i32, i32 } %1366, 2, !dbg !23 + %1372 = bitcast i32 %1371 to <2 x bfloat>, !dbg !23 + %1373 = extractvalue { i32, i32, i32, i32 } %1366, 3, !dbg !23 + %1374 = bitcast i32 %1373 to <2 x bfloat>, !dbg !23 + %1375 = extractelement <2 x bfloat> %1368, i64 0, !dbg !23 + %1376 = extractelement <2 x bfloat> %1368, i64 1, !dbg !23 + %1377 = extractelement <2 x bfloat> %1370, i64 0, !dbg !23 + %1378 = extractelement <2 x bfloat> %1370, i64 1, !dbg !23 + %1379 = extractelement <2 x bfloat> %1372, i64 0, !dbg !23 + %1380 = extractelement <2 x bfloat> %1372, i64 1, !dbg !23 + %1381 = extractelement <2 x bfloat> %1374, i64 0, !dbg !23 + %1382 = extractelement <2 x bfloat> %1374, i64 1, !dbg !23 + %1383 = extractelement <64 x i1> %43, i64 3, !dbg !23 + %1384 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %299, i1 %1383) #7, !dbg !23 + %1385 = extractvalue { i32, i32, i32, i32 } %1384, 0, !dbg !23 + %1386 = bitcast i32 %1385 to <2 x bfloat>, !dbg !23 + %1387 = extractvalue { i32, i32, i32, i32 } %1384, 1, !dbg !23 + %1388 = bitcast i32 %1387 to <2 x bfloat>, !dbg !23 + %1389 = extractvalue { i32, i32, i32, i32 } %1384, 2, !dbg !23 + %1390 = bitcast i32 %1389 to <2 x bfloat>, !dbg !23 + %1391 = extractvalue { i32, i32, i32, i32 } %1384, 3, !dbg !23 + %1392 = bitcast i32 %1391 to <2 x bfloat>, !dbg !23 + %1393 = extractelement <2 x bfloat> %1386, i64 0, !dbg !23 + %1394 = extractelement <2 x bfloat> %1386, i64 1, !dbg !23 + %1395 = extractelement <2 x bfloat> %1388, i64 0, !dbg !23 + %1396 = extractelement <2 x bfloat> %1388, i64 1, !dbg !23 + %1397 = extractelement <2 x bfloat> %1390, i64 0, !dbg !23 + %1398 = extractelement <2 x bfloat> %1390, i64 1, !dbg !23 + %1399 = extractelement <2 x bfloat> %1392, i64 0, !dbg !23 + %1400 = extractelement <2 x bfloat> %1392, i64 1, !dbg !23 + %1401 = extractelement <64 x i1> %43, i64 2, !dbg !23 + %1402 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %300, i1 %1401) #7, !dbg !23 + %1403 = extractvalue { i32, i32, i32, i32 } %1402, 0, !dbg !23 + %1404 = bitcast i32 %1403 to <2 x bfloat>, !dbg !23 + %1405 = extractvalue { i32, i32, i32, i32 } %1402, 1, !dbg !23 + %1406 = bitcast i32 %1405 to <2 x bfloat>, !dbg !23 + %1407 = extractvalue { i32, i32, i32, i32 } %1402, 2, !dbg !23 + %1408 = bitcast i32 %1407 to <2 x bfloat>, !dbg !23 + %1409 = extractvalue { i32, i32, i32, i32 } %1402, 3, !dbg !23 + %1410 = bitcast i32 %1409 to <2 x bfloat>, !dbg !23 + %1411 = extractelement <2 x bfloat> %1404, i64 0, !dbg !23 + %1412 = extractelement <2 x bfloat> %1404, i64 1, !dbg !23 + %1413 = extractelement <2 x bfloat> %1406, i64 0, !dbg !23 + %1414 = extractelement <2 x bfloat> %1406, i64 1, !dbg !23 + %1415 = extractelement <2 x bfloat> %1408, i64 0, !dbg !23 + %1416 = extractelement <2 x bfloat> %1408, i64 1, !dbg !23 + %1417 = extractelement <2 x bfloat> %1410, i64 0, !dbg !23 + %1418 = extractelement <2 x bfloat> %1410, i64 1, !dbg !23 + %1419 = extractelement <64 x i1> %43, i64 1, !dbg !23 + %1420 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %301, i1 %1419) #7, !dbg !23 + %1421 = extractvalue { i32, i32, i32, i32 } %1420, 0, !dbg !23 + %1422 = bitcast i32 %1421 to <2 x bfloat>, !dbg !23 + %1423 = extractvalue { i32, i32, i32, i32 } %1420, 1, !dbg !23 + %1424 = bitcast i32 %1423 to <2 x bfloat>, !dbg !23 + %1425 = extractvalue { i32, i32, i32, i32 } %1420, 2, !dbg !23 + %1426 = bitcast i32 %1425 to <2 x bfloat>, !dbg !23 + %1427 = extractvalue { i32, i32, i32, i32 } %1420, 3, !dbg !23 + %1428 = bitcast i32 %1427 to <2 x bfloat>, !dbg !23 + %1429 = extractelement <2 x bfloat> %1422, i64 0, !dbg !23 + %1430 = extractelement <2 x bfloat> %1422, i64 1, !dbg !23 + %1431 = extractelement <2 x bfloat> %1424, i64 0, !dbg !23 + %1432 = extractelement <2 x bfloat> %1424, i64 1, !dbg !23 + %1433 = extractelement <2 x bfloat> %1426, i64 0, !dbg !23 + %1434 = extractelement <2 x bfloat> %1426, i64 1, !dbg !23 + %1435 = extractelement <2 x bfloat> %1428, i64 0, !dbg !23 + %1436 = extractelement <2 x bfloat> %1428, i64 1, !dbg !23 + %1437 = extractelement <64 x i1> %43, i64 0, !dbg !23 + %1438 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %302, i1 %1437) #7, !dbg !23 + %1439 = extractvalue { i32, i32, i32, i32 } %1438, 0, !dbg !23 + %1440 = bitcast i32 %1439 to <2 x bfloat>, !dbg !23 + %1441 = extractvalue { i32, i32, i32, i32 } %1438, 1, !dbg !23 + %1442 = bitcast i32 %1441 to <2 x bfloat>, !dbg !23 + %1443 = extractvalue { i32, i32, i32, i32 } %1438, 2, !dbg !23 + %1444 = bitcast i32 %1443 to <2 x bfloat>, !dbg !23 + %1445 = extractvalue { i32, i32, i32, i32 } %1438, 3, !dbg !23 + %1446 = bitcast i32 %1445 to <2 x bfloat>, !dbg !23 + %1447 = extractelement <2 x bfloat> %1440, i64 0, !dbg !23 + %1448 = extractelement <2 x bfloat> %1440, i64 1, !dbg !23 + %1449 = extractelement <2 x bfloat> %1442, i64 0, !dbg !23 + %1450 = extractelement <2 x bfloat> %1442, i64 1, !dbg !23 + %1451 = extractelement <2 x bfloat> %1444, i64 0, !dbg !23 + %1452 = extractelement <2 x bfloat> %1444, i64 1, !dbg !23 + %1453 = extractelement <2 x bfloat> %1446, i64 0, !dbg !23 + %1454 = extractelement <2 x bfloat> %1446, i64 1, !dbg !23 + %1455 = fpext bfloat %313 to float, !dbg !24 + %1456 = fpext bfloat %314 to float, !dbg !24 + %1457 = fpext bfloat %315 to float, !dbg !24 + %1458 = fpext bfloat %316 to float, !dbg !24 + %1459 = fpext bfloat %317 to float, !dbg !24 + %1460 = fpext bfloat %318 to float, !dbg !24 + %1461 = fpext bfloat %319 to float, !dbg !24 + %1462 = fpext bfloat %320 to float, !dbg !24 + %1463 = fpext bfloat %331 to float, !dbg !24 + %1464 = fpext bfloat %332 to float, !dbg !24 + %1465 = fpext bfloat %333 to float, !dbg !24 + %1466 = fpext bfloat %334 to float, !dbg !24 + %1467 = fpext bfloat %335 to float, !dbg !24 + %1468 = fpext bfloat %336 to float, !dbg !24 + %1469 = fpext bfloat %337 to float, !dbg !24 + %1470 = fpext bfloat %338 to float, !dbg !24 + %1471 = fpext bfloat %349 to float, !dbg !24 + %1472 = fpext bfloat %350 to float, !dbg !24 + %1473 = fpext bfloat %351 to float, !dbg !24 + %1474 = fpext bfloat %352 to float, !dbg !24 + %1475 = fpext bfloat %353 to float, !dbg !24 + %1476 = fpext bfloat %354 to float, !dbg !24 + %1477 = fpext bfloat %355 to float, !dbg !24 + %1478 = fpext bfloat %356 to float, !dbg !24 + %1479 = fpext bfloat %367 to float, !dbg !24 + %1480 = fpext bfloat %368 to float, !dbg !24 + %1481 = fpext bfloat %369 to float, !dbg !24 + %1482 = fpext bfloat %370 to float, !dbg !24 + %1483 = fpext bfloat %371 to float, !dbg !24 + %1484 = fpext bfloat %372 to float, !dbg !24 + %1485 = fpext bfloat %373 to float, !dbg !24 + %1486 = fpext bfloat %374 to float, !dbg !24 + %1487 = fpext bfloat %385 to float, !dbg !24 + %1488 = fpext bfloat %386 to float, !dbg !24 + %1489 = fpext bfloat %387 to float, !dbg !24 + %1490 = fpext bfloat %388 to float, !dbg !24 + %1491 = fpext bfloat %389 to float, !dbg !24 + %1492 = fpext bfloat %390 to float, !dbg !24 + %1493 = fpext bfloat %391 to float, !dbg !24 + %1494 = fpext bfloat %392 to float, !dbg !24 + %1495 = fpext bfloat %403 to float, !dbg !24 + %1496 = fpext bfloat %404 to float, !dbg !24 + %1497 = fpext bfloat %405 to float, !dbg !24 + %1498 = fpext bfloat %406 to float, !dbg !24 + %1499 = fpext bfloat %407 to float, !dbg !24 + %1500 = fpext bfloat %408 to float, !dbg !24 + %1501 = fpext bfloat %409 to float, !dbg !24 + %1502 = fpext bfloat %410 to float, !dbg !24 + %1503 = fpext bfloat %421 to float, !dbg !24 + %1504 = fpext bfloat %422 to float, !dbg !24 + %1505 = fpext bfloat %423 to float, !dbg !24 + %1506 = fpext bfloat %424 to float, !dbg !24 + %1507 = fpext bfloat %425 to float, !dbg !24 + %1508 = fpext bfloat %426 to float, !dbg !24 + %1509 = fpext bfloat %427 to float, !dbg !24 + %1510 = fpext bfloat %428 to float, !dbg !24 + %1511 = fpext bfloat %439 to float, !dbg !24 + %1512 = fpext bfloat %440 to float, !dbg !24 + %1513 = fpext bfloat %441 to float, !dbg !24 + %1514 = fpext bfloat %442 to float, !dbg !24 + %1515 = fpext bfloat %443 to float, !dbg !24 + %1516 = fpext bfloat %444 to float, !dbg !24 + %1517 = fpext bfloat %445 to float, !dbg !24 + %1518 = fpext bfloat %446 to float, !dbg !24 + %1519 = fpext bfloat %457 to float, !dbg !24 + %1520 = fpext bfloat %458 to float, !dbg !24 + %1521 = fpext bfloat %459 to float, !dbg !24 + %1522 = fpext bfloat %460 to float, !dbg !24 + %1523 = fpext bfloat %461 to float, !dbg !24 + %1524 = fpext bfloat %462 to float, !dbg !24 + %1525 = fpext bfloat %463 to float, !dbg !24 + %1526 = fpext bfloat %464 to float, !dbg !24 + %1527 = fpext bfloat %475 to float, !dbg !24 + %1528 = fpext bfloat %476 to float, !dbg !24 + %1529 = fpext bfloat %477 to float, !dbg !24 + %1530 = fpext bfloat %478 to float, !dbg !24 + %1531 = fpext bfloat %479 to float, !dbg !24 + %1532 = fpext bfloat %480 to float, !dbg !24 + %1533 = fpext bfloat %481 to float, !dbg !24 + %1534 = fpext bfloat %482 to float, !dbg !24 + %1535 = fpext bfloat %493 to float, !dbg !24 + %1536 = fpext bfloat %494 to float, !dbg !24 + %1537 = fpext bfloat %495 to float, !dbg !24 + %1538 = fpext bfloat %496 to float, !dbg !24 + %1539 = fpext bfloat %497 to float, !dbg !24 + %1540 = fpext bfloat %498 to float, !dbg !24 + %1541 = fpext bfloat %499 to float, !dbg !24 + %1542 = fpext bfloat %500 to float, !dbg !24 + %1543 = fpext bfloat %511 to float, !dbg !24 + %1544 = fpext bfloat %512 to float, !dbg !24 + %1545 = fpext bfloat %513 to float, !dbg !24 + %1546 = fpext bfloat %514 to float, !dbg !24 + %1547 = fpext bfloat %515 to float, !dbg !24 + %1548 = fpext bfloat %516 to float, !dbg !24 + %1549 = fpext bfloat %517 to float, !dbg !24 + %1550 = fpext bfloat %518 to float, !dbg !24 + %1551 = fpext bfloat %529 to float, !dbg !24 + %1552 = fpext bfloat %530 to float, !dbg !24 + %1553 = fpext bfloat %531 to float, !dbg !24 + %1554 = fpext bfloat %532 to float, !dbg !24 + %1555 = fpext bfloat %533 to float, !dbg !24 + %1556 = fpext bfloat %534 to float, !dbg !24 + %1557 = fpext bfloat %535 to float, !dbg !24 + %1558 = fpext bfloat %536 to float, !dbg !24 + %1559 = fpext bfloat %547 to float, !dbg !24 + %1560 = fpext bfloat %548 to float, !dbg !24 + %1561 = fpext bfloat %549 to float, !dbg !24 + %1562 = fpext bfloat %550 to float, !dbg !24 + %1563 = fpext bfloat %551 to float, !dbg !24 + %1564 = fpext bfloat %552 to float, !dbg !24 + %1565 = fpext bfloat %553 to float, !dbg !24 + %1566 = fpext bfloat %554 to float, !dbg !24 + %1567 = fpext bfloat %565 to float, !dbg !24 + %1568 = fpext bfloat %566 to float, !dbg !24 + %1569 = fpext bfloat %567 to float, !dbg !24 + %1570 = fpext bfloat %568 to float, !dbg !24 + %1571 = fpext bfloat %569 to float, !dbg !24 + %1572 = fpext bfloat %570 to float, !dbg !24 + %1573 = fpext bfloat %571 to float, !dbg !24 + %1574 = fpext bfloat %572 to float, !dbg !24 + %1575 = fpext bfloat %583 to float, !dbg !24 + %1576 = fpext bfloat %584 to float, !dbg !24 + %1577 = fpext bfloat %585 to float, !dbg !24 + %1578 = fpext bfloat %586 to float, !dbg !24 + %1579 = fpext bfloat %587 to float, !dbg !24 + %1580 = fpext bfloat %588 to float, !dbg !24 + %1581 = fpext bfloat %589 to float, !dbg !24 + %1582 = fpext bfloat %590 to float, !dbg !24 + %1583 = fpext bfloat %601 to float, !dbg !24 + %1584 = fpext bfloat %602 to float, !dbg !24 + %1585 = fpext bfloat %603 to float, !dbg !24 + %1586 = fpext bfloat %604 to float, !dbg !24 + %1587 = fpext bfloat %605 to float, !dbg !24 + %1588 = fpext bfloat %606 to float, !dbg !24 + %1589 = fpext bfloat %607 to float, !dbg !24 + %1590 = fpext bfloat %608 to float, !dbg !24 + %1591 = fpext bfloat %619 to float, !dbg !24 + %1592 = fpext bfloat %620 to float, !dbg !24 + %1593 = fpext bfloat %621 to float, !dbg !24 + %1594 = fpext bfloat %622 to float, !dbg !24 + %1595 = fpext bfloat %623 to float, !dbg !24 + %1596 = fpext bfloat %624 to float, !dbg !24 + %1597 = fpext bfloat %625 to float, !dbg !24 + %1598 = fpext bfloat %626 to float, !dbg !24 + %1599 = fpext bfloat %637 to float, !dbg !24 + %1600 = fpext bfloat %638 to float, !dbg !24 + %1601 = fpext bfloat %639 to float, !dbg !24 + %1602 = fpext bfloat %640 to float, !dbg !24 + %1603 = fpext bfloat %641 to float, !dbg !24 + %1604 = fpext bfloat %642 to float, !dbg !24 + %1605 = fpext bfloat %643 to float, !dbg !24 + %1606 = fpext bfloat %644 to float, !dbg !24 + %1607 = fpext bfloat %655 to float, !dbg !24 + %1608 = fpext bfloat %656 to float, !dbg !24 + %1609 = fpext bfloat %657 to float, !dbg !24 + %1610 = fpext bfloat %658 to float, !dbg !24 + %1611 = fpext bfloat %659 to float, !dbg !24 + %1612 = fpext bfloat %660 to float, !dbg !24 + %1613 = fpext bfloat %661 to float, !dbg !24 + %1614 = fpext bfloat %662 to float, !dbg !24 + %1615 = fpext bfloat %673 to float, !dbg !24 + %1616 = fpext bfloat %674 to float, !dbg !24 + %1617 = fpext bfloat %675 to float, !dbg !24 + %1618 = fpext bfloat %676 to float, !dbg !24 + %1619 = fpext bfloat %677 to float, !dbg !24 + %1620 = fpext bfloat %678 to float, !dbg !24 + %1621 = fpext bfloat %679 to float, !dbg !24 + %1622 = fpext bfloat %680 to float, !dbg !24 + %1623 = fpext bfloat %691 to float, !dbg !24 + %1624 = fpext bfloat %692 to float, !dbg !24 + %1625 = fpext bfloat %693 to float, !dbg !24 + %1626 = fpext bfloat %694 to float, !dbg !24 + %1627 = fpext bfloat %695 to float, !dbg !24 + %1628 = fpext bfloat %696 to float, !dbg !24 + %1629 = fpext bfloat %697 to float, !dbg !24 + %1630 = fpext bfloat %698 to float, !dbg !24 + %1631 = fpext bfloat %709 to float, !dbg !24 + %1632 = fpext bfloat %710 to float, !dbg !24 + %1633 = fpext bfloat %711 to float, !dbg !24 + %1634 = fpext bfloat %712 to float, !dbg !24 + %1635 = fpext bfloat %713 to float, !dbg !24 + %1636 = fpext bfloat %714 to float, !dbg !24 + %1637 = fpext bfloat %715 to float, !dbg !24 + %1638 = fpext bfloat %716 to float, !dbg !24 + %1639 = fpext bfloat %727 to float, !dbg !24 + %1640 = fpext bfloat %728 to float, !dbg !24 + %1641 = fpext bfloat %729 to float, !dbg !24 + %1642 = fpext bfloat %730 to float, !dbg !24 + %1643 = fpext bfloat %731 to float, !dbg !24 + %1644 = fpext bfloat %732 to float, !dbg !24 + %1645 = fpext bfloat %733 to float, !dbg !24 + %1646 = fpext bfloat %734 to float, !dbg !24 + %1647 = fpext bfloat %745 to float, !dbg !24 + %1648 = fpext bfloat %746 to float, !dbg !24 + %1649 = fpext bfloat %747 to float, !dbg !24 + %1650 = fpext bfloat %748 to float, !dbg !24 + %1651 = fpext bfloat %749 to float, !dbg !24 + %1652 = fpext bfloat %750 to float, !dbg !24 + %1653 = fpext bfloat %751 to float, !dbg !24 + %1654 = fpext bfloat %752 to float, !dbg !24 + %1655 = fpext bfloat %763 to float, !dbg !24 + %1656 = fpext bfloat %764 to float, !dbg !24 + %1657 = fpext bfloat %765 to float, !dbg !24 + %1658 = fpext bfloat %766 to float, !dbg !24 + %1659 = fpext bfloat %767 to float, !dbg !24 + %1660 = fpext bfloat %768 to float, !dbg !24 + %1661 = fpext bfloat %769 to float, !dbg !24 + %1662 = fpext bfloat %770 to float, !dbg !24 + %1663 = fpext bfloat %781 to float, !dbg !24 + %1664 = fpext bfloat %782 to float, !dbg !24 + %1665 = fpext bfloat %783 to float, !dbg !24 + %1666 = fpext bfloat %784 to float, !dbg !24 + %1667 = fpext bfloat %785 to float, !dbg !24 + %1668 = fpext bfloat %786 to float, !dbg !24 + %1669 = fpext bfloat %787 to float, !dbg !24 + %1670 = fpext bfloat %788 to float, !dbg !24 + %1671 = fpext bfloat %799 to float, !dbg !24 + %1672 = fpext bfloat %800 to float, !dbg !24 + %1673 = fpext bfloat %801 to float, !dbg !24 + %1674 = fpext bfloat %802 to float, !dbg !24 + %1675 = fpext bfloat %803 to float, !dbg !24 + %1676 = fpext bfloat %804 to float, !dbg !24 + %1677 = fpext bfloat %805 to float, !dbg !24 + %1678 = fpext bfloat %806 to float, !dbg !24 + %1679 = fpext bfloat %817 to float, !dbg !24 + %1680 = fpext bfloat %818 to float, !dbg !24 + %1681 = fpext bfloat %819 to float, !dbg !24 + %1682 = fpext bfloat %820 to float, !dbg !24 + %1683 = fpext bfloat %821 to float, !dbg !24 + %1684 = fpext bfloat %822 to float, !dbg !24 + %1685 = fpext bfloat %823 to float, !dbg !24 + %1686 = fpext bfloat %824 to float, !dbg !24 + %1687 = fpext bfloat %835 to float, !dbg !24 + %1688 = fpext bfloat %836 to float, !dbg !24 + %1689 = fpext bfloat %837 to float, !dbg !24 + %1690 = fpext bfloat %838 to float, !dbg !24 + %1691 = fpext bfloat %839 to float, !dbg !24 + %1692 = fpext bfloat %840 to float, !dbg !24 + %1693 = fpext bfloat %841 to float, !dbg !24 + %1694 = fpext bfloat %842 to float, !dbg !24 + %1695 = fpext bfloat %853 to float, !dbg !24 + %1696 = fpext bfloat %854 to float, !dbg !24 + %1697 = fpext bfloat %855 to float, !dbg !24 + %1698 = fpext bfloat %856 to float, !dbg !24 + %1699 = fpext bfloat %857 to float, !dbg !24 + %1700 = fpext bfloat %858 to float, !dbg !24 + %1701 = fpext bfloat %859 to float, !dbg !24 + %1702 = fpext bfloat %860 to float, !dbg !24 + %1703 = fpext bfloat %871 to float, !dbg !24 + %1704 = fpext bfloat %872 to float, !dbg !24 + %1705 = fpext bfloat %873 to float, !dbg !24 + %1706 = fpext bfloat %874 to float, !dbg !24 + %1707 = fpext bfloat %875 to float, !dbg !24 + %1708 = fpext bfloat %876 to float, !dbg !24 + %1709 = fpext bfloat %877 to float, !dbg !24 + %1710 = fpext bfloat %878 to float, !dbg !24 + %1711 = fpext bfloat %889 to float, !dbg !24 + %1712 = fpext bfloat %890 to float, !dbg !24 + %1713 = fpext bfloat %891 to float, !dbg !24 + %1714 = fpext bfloat %892 to float, !dbg !24 + %1715 = fpext bfloat %893 to float, !dbg !24 + %1716 = fpext bfloat %894 to float, !dbg !24 + %1717 = fpext bfloat %895 to float, !dbg !24 + %1718 = fpext bfloat %896 to float, !dbg !24 + %1719 = fpext bfloat %907 to float, !dbg !24 + %1720 = fpext bfloat %908 to float, !dbg !24 + %1721 = fpext bfloat %909 to float, !dbg !24 + %1722 = fpext bfloat %910 to float, !dbg !24 + %1723 = fpext bfloat %911 to float, !dbg !24 + %1724 = fpext bfloat %912 to float, !dbg !24 + %1725 = fpext bfloat %913 to float, !dbg !24 + %1726 = fpext bfloat %914 to float, !dbg !24 + %1727 = fpext bfloat %925 to float, !dbg !24 + %1728 = fpext bfloat %926 to float, !dbg !24 + %1729 = fpext bfloat %927 to float, !dbg !24 + %1730 = fpext bfloat %928 to float, !dbg !24 + %1731 = fpext bfloat %929 to float, !dbg !24 + %1732 = fpext bfloat %930 to float, !dbg !24 + %1733 = fpext bfloat %931 to float, !dbg !24 + %1734 = fpext bfloat %932 to float, !dbg !24 + %1735 = fpext bfloat %943 to float, !dbg !24 + %1736 = fpext bfloat %944 to float, !dbg !24 + %1737 = fpext bfloat %945 to float, !dbg !24 + %1738 = fpext bfloat %946 to float, !dbg !24 + %1739 = fpext bfloat %947 to float, !dbg !24 + %1740 = fpext bfloat %948 to float, !dbg !24 + %1741 = fpext bfloat %949 to float, !dbg !24 + %1742 = fpext bfloat %950 to float, !dbg !24 + %1743 = fpext bfloat %961 to float, !dbg !24 + %1744 = fpext bfloat %962 to float, !dbg !24 + %1745 = fpext bfloat %963 to float, !dbg !24 + %1746 = fpext bfloat %964 to float, !dbg !24 + %1747 = fpext bfloat %965 to float, !dbg !24 + %1748 = fpext bfloat %966 to float, !dbg !24 + %1749 = fpext bfloat %967 to float, !dbg !24 + %1750 = fpext bfloat %968 to float, !dbg !24 + %1751 = fpext bfloat %979 to float, !dbg !24 + %1752 = fpext bfloat %980 to float, !dbg !24 + %1753 = fpext bfloat %981 to float, !dbg !24 + %1754 = fpext bfloat %982 to float, !dbg !24 + %1755 = fpext bfloat %983 to float, !dbg !24 + %1756 = fpext bfloat %984 to float, !dbg !24 + %1757 = fpext bfloat %985 to float, !dbg !24 + %1758 = fpext bfloat %986 to float, !dbg !24 + %1759 = fpext bfloat %997 to float, !dbg !24 + %1760 = fpext bfloat %998 to float, !dbg !24 + %1761 = fpext bfloat %999 to float, !dbg !24 + %1762 = fpext bfloat %1000 to float, !dbg !24 + %1763 = fpext bfloat %1001 to float, !dbg !24 + %1764 = fpext bfloat %1002 to float, !dbg !24 + %1765 = fpext bfloat %1003 to float, !dbg !24 + %1766 = fpext bfloat %1004 to float, !dbg !24 + %1767 = fpext bfloat %1015 to float, !dbg !24 + %1768 = fpext bfloat %1016 to float, !dbg !24 + %1769 = fpext bfloat %1017 to float, !dbg !24 + %1770 = fpext bfloat %1018 to float, !dbg !24 + %1771 = fpext bfloat %1019 to float, !dbg !24 + %1772 = fpext bfloat %1020 to float, !dbg !24 + %1773 = fpext bfloat %1021 to float, !dbg !24 + %1774 = fpext bfloat %1022 to float, !dbg !24 + %1775 = fpext bfloat %1033 to float, !dbg !24 + %1776 = fpext bfloat %1034 to float, !dbg !24 + %1777 = fpext bfloat %1035 to float, !dbg !24 + %1778 = fpext bfloat %1036 to float, !dbg !24 + %1779 = fpext bfloat %1037 to float, !dbg !24 + %1780 = fpext bfloat %1038 to float, !dbg !24 + %1781 = fpext bfloat %1039 to float, !dbg !24 + %1782 = fpext bfloat %1040 to float, !dbg !24 + %1783 = fpext bfloat %1051 to float, !dbg !24 + %1784 = fpext bfloat %1052 to float, !dbg !24 + %1785 = fpext bfloat %1053 to float, !dbg !24 + %1786 = fpext bfloat %1054 to float, !dbg !24 + %1787 = fpext bfloat %1055 to float, !dbg !24 + %1788 = fpext bfloat %1056 to float, !dbg !24 + %1789 = fpext bfloat %1057 to float, !dbg !24 + %1790 = fpext bfloat %1058 to float, !dbg !24 + %1791 = fpext bfloat %1069 to float, !dbg !24 + %1792 = fpext bfloat %1070 to float, !dbg !24 + %1793 = fpext bfloat %1071 to float, !dbg !24 + %1794 = fpext bfloat %1072 to float, !dbg !24 + %1795 = fpext bfloat %1073 to float, !dbg !24 + %1796 = fpext bfloat %1074 to float, !dbg !24 + %1797 = fpext bfloat %1075 to float, !dbg !24 + %1798 = fpext bfloat %1076 to float, !dbg !24 + %1799 = fpext bfloat %1087 to float, !dbg !24 + %1800 = fpext bfloat %1088 to float, !dbg !24 + %1801 = fpext bfloat %1089 to float, !dbg !24 + %1802 = fpext bfloat %1090 to float, !dbg !24 + %1803 = fpext bfloat %1091 to float, !dbg !24 + %1804 = fpext bfloat %1092 to float, !dbg !24 + %1805 = fpext bfloat %1093 to float, !dbg !24 + %1806 = fpext bfloat %1094 to float, !dbg !24 + %1807 = fpext bfloat %1105 to float, !dbg !24 + %1808 = fpext bfloat %1106 to float, !dbg !24 + %1809 = fpext bfloat %1107 to float, !dbg !24 + %1810 = fpext bfloat %1108 to float, !dbg !24 + %1811 = fpext bfloat %1109 to float, !dbg !24 + %1812 = fpext bfloat %1110 to float, !dbg !24 + %1813 = fpext bfloat %1111 to float, !dbg !24 + %1814 = fpext bfloat %1112 to float, !dbg !24 + %1815 = fpext bfloat %1123 to float, !dbg !24 + %1816 = fpext bfloat %1124 to float, !dbg !24 + %1817 = fpext bfloat %1125 to float, !dbg !24 + %1818 = fpext bfloat %1126 to float, !dbg !24 + %1819 = fpext bfloat %1127 to float, !dbg !24 + %1820 = fpext bfloat %1128 to float, !dbg !24 + %1821 = fpext bfloat %1129 to float, !dbg !24 + %1822 = fpext bfloat %1130 to float, !dbg !24 + %1823 = fpext bfloat %1141 to float, !dbg !24 + %1824 = fpext bfloat %1142 to float, !dbg !24 + %1825 = fpext bfloat %1143 to float, !dbg !24 + %1826 = fpext bfloat %1144 to float, !dbg !24 + %1827 = fpext bfloat %1145 to float, !dbg !24 + %1828 = fpext bfloat %1146 to float, !dbg !24 + %1829 = fpext bfloat %1147 to float, !dbg !24 + %1830 = fpext bfloat %1148 to float, !dbg !24 + %1831 = fpext bfloat %1159 to float, !dbg !24 + %1832 = fpext bfloat %1160 to float, !dbg !24 + %1833 = fpext bfloat %1161 to float, !dbg !24 + %1834 = fpext bfloat %1162 to float, !dbg !24 + %1835 = fpext bfloat %1163 to float, !dbg !24 + %1836 = fpext bfloat %1164 to float, !dbg !24 + %1837 = fpext bfloat %1165 to float, !dbg !24 + %1838 = fpext bfloat %1166 to float, !dbg !24 + %1839 = fpext bfloat %1177 to float, !dbg !24 + %1840 = fpext bfloat %1178 to float, !dbg !24 + %1841 = fpext bfloat %1179 to float, !dbg !24 + %1842 = fpext bfloat %1180 to float, !dbg !24 + %1843 = fpext bfloat %1181 to float, !dbg !24 + %1844 = fpext bfloat %1182 to float, !dbg !24 + %1845 = fpext bfloat %1183 to float, !dbg !24 + %1846 = fpext bfloat %1184 to float, !dbg !24 + %1847 = fpext bfloat %1195 to float, !dbg !24 + %1848 = fpext bfloat %1196 to float, !dbg !24 + %1849 = fpext bfloat %1197 to float, !dbg !24 + %1850 = fpext bfloat %1198 to float, !dbg !24 + %1851 = fpext bfloat %1199 to float, !dbg !24 + %1852 = fpext bfloat %1200 to float, !dbg !24 + %1853 = fpext bfloat %1201 to float, !dbg !24 + %1854 = fpext bfloat %1202 to float, !dbg !24 + %1855 = fpext bfloat %1213 to float, !dbg !24 + %1856 = fpext bfloat %1214 to float, !dbg !24 + %1857 = fpext bfloat %1215 to float, !dbg !24 + %1858 = fpext bfloat %1216 to float, !dbg !24 + %1859 = fpext bfloat %1217 to float, !dbg !24 + %1860 = fpext bfloat %1218 to float, !dbg !24 + %1861 = fpext bfloat %1219 to float, !dbg !24 + %1862 = fpext bfloat %1220 to float, !dbg !24 + %1863 = fpext bfloat %1231 to float, !dbg !24 + %1864 = fpext bfloat %1232 to float, !dbg !24 + %1865 = fpext bfloat %1233 to float, !dbg !24 + %1866 = fpext bfloat %1234 to float, !dbg !24 + %1867 = fpext bfloat %1235 to float, !dbg !24 + %1868 = fpext bfloat %1236 to float, !dbg !24 + %1869 = fpext bfloat %1237 to float, !dbg !24 + %1870 = fpext bfloat %1238 to float, !dbg !24 + %1871 = fpext bfloat %1249 to float, !dbg !24 + %1872 = fpext bfloat %1250 to float, !dbg !24 + %1873 = fpext bfloat %1251 to float, !dbg !24 + %1874 = fpext bfloat %1252 to float, !dbg !24 + %1875 = fpext bfloat %1253 to float, !dbg !24 + %1876 = fpext bfloat %1254 to float, !dbg !24 + %1877 = fpext bfloat %1255 to float, !dbg !24 + %1878 = fpext bfloat %1256 to float, !dbg !24 + %1879 = fpext bfloat %1267 to float, !dbg !24 + %1880 = fpext bfloat %1268 to float, !dbg !24 + %1881 = fpext bfloat %1269 to float, !dbg !24 + %1882 = fpext bfloat %1270 to float, !dbg !24 + %1883 = fpext bfloat %1271 to float, !dbg !24 + %1884 = fpext bfloat %1272 to float, !dbg !24 + %1885 = fpext bfloat %1273 to float, !dbg !24 + %1886 = fpext bfloat %1274 to float, !dbg !24 + %1887 = fpext bfloat %1285 to float, !dbg !24 + %1888 = fpext bfloat %1286 to float, !dbg !24 + %1889 = fpext bfloat %1287 to float, !dbg !24 + %1890 = fpext bfloat %1288 to float, !dbg !24 + %1891 = fpext bfloat %1289 to float, !dbg !24 + %1892 = fpext bfloat %1290 to float, !dbg !24 + %1893 = fpext bfloat %1291 to float, !dbg !24 + %1894 = fpext bfloat %1292 to float, !dbg !24 + %1895 = fpext bfloat %1303 to float, !dbg !24 + %1896 = fpext bfloat %1304 to float, !dbg !24 + %1897 = fpext bfloat %1305 to float, !dbg !24 + %1898 = fpext bfloat %1306 to float, !dbg !24 + %1899 = fpext bfloat %1307 to float, !dbg !24 + %1900 = fpext bfloat %1308 to float, !dbg !24 + %1901 = fpext bfloat %1309 to float, !dbg !24 + %1902 = fpext bfloat %1310 to float, !dbg !24 + %1903 = fpext bfloat %1321 to float, !dbg !24 + %1904 = fpext bfloat %1322 to float, !dbg !24 + %1905 = fpext bfloat %1323 to float, !dbg !24 + %1906 = fpext bfloat %1324 to float, !dbg !24 + %1907 = fpext bfloat %1325 to float, !dbg !24 + %1908 = fpext bfloat %1326 to float, !dbg !24 + %1909 = fpext bfloat %1327 to float, !dbg !24 + %1910 = fpext bfloat %1328 to float, !dbg !24 + %1911 = fpext bfloat %1339 to float, !dbg !24 + %1912 = fpext bfloat %1340 to float, !dbg !24 + %1913 = fpext bfloat %1341 to float, !dbg !24 + %1914 = fpext bfloat %1342 to float, !dbg !24 + %1915 = fpext bfloat %1343 to float, !dbg !24 + %1916 = fpext bfloat %1344 to float, !dbg !24 + %1917 = fpext bfloat %1345 to float, !dbg !24 + %1918 = fpext bfloat %1346 to float, !dbg !24 + %1919 = fpext bfloat %1357 to float, !dbg !24 + %1920 = fpext bfloat %1358 to float, !dbg !24 + %1921 = fpext bfloat %1359 to float, !dbg !24 + %1922 = fpext bfloat %1360 to float, !dbg !24 + %1923 = fpext bfloat %1361 to float, !dbg !24 + %1924 = fpext bfloat %1362 to float, !dbg !24 + %1925 = fpext bfloat %1363 to float, !dbg !24 + %1926 = fpext bfloat %1364 to float, !dbg !24 + %1927 = fpext bfloat %1375 to float, !dbg !24 + %1928 = fpext bfloat %1376 to float, !dbg !24 + %1929 = fpext bfloat %1377 to float, !dbg !24 + %1930 = fpext bfloat %1378 to float, !dbg !24 + %1931 = fpext bfloat %1379 to float, !dbg !24 + %1932 = fpext bfloat %1380 to float, !dbg !24 + %1933 = fpext bfloat %1381 to float, !dbg !24 + %1934 = fpext bfloat %1382 to float, !dbg !24 + %1935 = fpext bfloat %1393 to float, !dbg !24 + %1936 = fpext bfloat %1394 to float, !dbg !24 + %1937 = fpext bfloat %1395 to float, !dbg !24 + %1938 = fpext bfloat %1396 to float, !dbg !24 + %1939 = fpext bfloat %1397 to float, !dbg !24 + %1940 = fpext bfloat %1398 to float, !dbg !24 + %1941 = fpext bfloat %1399 to float, !dbg !24 + %1942 = fpext bfloat %1400 to float, !dbg !24 + %1943 = fpext bfloat %1411 to float, !dbg !24 + %1944 = fpext bfloat %1412 to float, !dbg !24 + %1945 = fpext bfloat %1413 to float, !dbg !24 + %1946 = fpext bfloat %1414 to float, !dbg !24 + %1947 = fpext bfloat %1415 to float, !dbg !24 + %1948 = fpext bfloat %1416 to float, !dbg !24 + %1949 = fpext bfloat %1417 to float, !dbg !24 + %1950 = fpext bfloat %1418 to float, !dbg !24 + %1951 = fpext bfloat %1429 to float, !dbg !24 + %1952 = fpext bfloat %1430 to float, !dbg !24 + %1953 = fpext bfloat %1431 to float, !dbg !24 + %1954 = fpext bfloat %1432 to float, !dbg !24 + %1955 = fpext bfloat %1433 to float, !dbg !24 + %1956 = fpext bfloat %1434 to float, !dbg !24 + %1957 = fpext bfloat %1435 to float, !dbg !24 + %1958 = fpext bfloat %1436 to float, !dbg !24 + %1959 = fpext bfloat %1447 to float, !dbg !24 + %1960 = fpext bfloat %1448 to float, !dbg !24 + %1961 = fpext bfloat %1449 to float, !dbg !24 + %1962 = fpext bfloat %1450 to float, !dbg !24 + %1963 = fpext bfloat %1451 to float, !dbg !24 + %1964 = fpext bfloat %1452 to float, !dbg !24 + %1965 = fpext bfloat %1453 to float, !dbg !24 + %1966 = fpext bfloat %1454 to float, !dbg !24 + %1967 = fmul float %1455, %1455, !dbg !25 + %1968 = fmul float %1456, %1456, !dbg !25 + %1969 = fmul float %1457, %1457, !dbg !25 + %1970 = fmul float %1458, %1458, !dbg !25 + %1971 = fmul float %1459, %1459, !dbg !25 + %1972 = fmul float %1460, %1460, !dbg !25 + %1973 = fmul float %1461, %1461, !dbg !25 + %1974 = fmul float %1462, %1462, !dbg !25 + %1975 = fmul float %1463, %1463, !dbg !25 + %1976 = fmul float %1464, %1464, !dbg !25 + %1977 = fmul float %1465, %1465, !dbg !25 + %1978 = fmul float %1466, %1466, !dbg !25 + %1979 = fmul float %1467, %1467, !dbg !25 + %1980 = fmul float %1468, %1468, !dbg !25 + %1981 = fmul float %1469, %1469, !dbg !25 + %1982 = fmul float %1470, %1470, !dbg !25 + %1983 = fmul float %1471, %1471, !dbg !25 + %1984 = fmul float %1472, %1472, !dbg !25 + %1985 = fmul float %1473, %1473, !dbg !25 + %1986 = fmul float %1474, %1474, !dbg !25 + %1987 = fmul float %1475, %1475, !dbg !25 + %1988 = fmul float %1476, %1476, !dbg !25 + %1989 = fmul float %1477, %1477, !dbg !25 + %1990 = fmul float %1478, %1478, !dbg !25 + %1991 = fmul float %1479, %1479, !dbg !25 + %1992 = fmul float %1480, %1480, !dbg !25 + %1993 = fmul float %1481, %1481, !dbg !25 + %1994 = fmul float %1482, %1482, !dbg !25 + %1995 = fmul float %1483, %1483, !dbg !25 + %1996 = fmul float %1484, %1484, !dbg !25 + %1997 = fmul float %1485, %1485, !dbg !25 + %1998 = fmul float %1486, %1486, !dbg !25 + %1999 = fmul float %1487, %1487, !dbg !25 + %2000 = fmul float %1488, %1488, !dbg !25 + %2001 = fmul float %1489, %1489, !dbg !25 + %2002 = fmul float %1490, %1490, !dbg !25 + %2003 = fmul float %1491, %1491, !dbg !25 + %2004 = fmul float %1492, %1492, !dbg !25 + %2005 = fmul float %1493, %1493, !dbg !25 + %2006 = fmul float %1494, %1494, !dbg !25 + %2007 = fmul float %1495, %1495, !dbg !25 + %2008 = fmul float %1496, %1496, !dbg !25 + %2009 = fmul float %1497, %1497, !dbg !25 + %2010 = fmul float %1498, %1498, !dbg !25 + %2011 = fmul float %1499, %1499, !dbg !25 + %2012 = fmul float %1500, %1500, !dbg !25 + %2013 = fmul float %1501, %1501, !dbg !25 + %2014 = fmul float %1502, %1502, !dbg !25 + %2015 = fmul float %1503, %1503, !dbg !25 + %2016 = fmul float %1504, %1504, !dbg !25 + %2017 = fmul float %1505, %1505, !dbg !25 + %2018 = fmul float %1506, %1506, !dbg !25 + %2019 = fmul float %1507, %1507, !dbg !25 + %2020 = fmul float %1508, %1508, !dbg !25 + %2021 = fmul float %1509, %1509, !dbg !25 + %2022 = fmul float %1510, %1510, !dbg !25 + %2023 = fmul float %1511, %1511, !dbg !25 + %2024 = fmul float %1512, %1512, !dbg !25 + %2025 = fmul float %1513, %1513, !dbg !25 + %2026 = fmul float %1514, %1514, !dbg !25 + %2027 = fmul float %1515, %1515, !dbg !25 + %2028 = fmul float %1516, %1516, !dbg !25 + %2029 = fmul float %1517, %1517, !dbg !25 + %2030 = fmul float %1518, %1518, !dbg !25 + %2031 = fmul float %1519, %1519, !dbg !25 + %2032 = fmul float %1520, %1520, !dbg !25 + %2033 = fmul float %1521, %1521, !dbg !25 + %2034 = fmul float %1522, %1522, !dbg !25 + %2035 = fmul float %1523, %1523, !dbg !25 + %2036 = fmul float %1524, %1524, !dbg !25 + %2037 = fmul float %1525, %1525, !dbg !25 + %2038 = fmul float %1526, %1526, !dbg !25 + %2039 = fmul float %1527, %1527, !dbg !25 + %2040 = fmul float %1528, %1528, !dbg !25 + %2041 = fmul float %1529, %1529, !dbg !25 + %2042 = fmul float %1530, %1530, !dbg !25 + %2043 = fmul float %1531, %1531, !dbg !25 + %2044 = fmul float %1532, %1532, !dbg !25 + %2045 = fmul float %1533, %1533, !dbg !25 + %2046 = fmul float %1534, %1534, !dbg !25 + %2047 = fmul float %1535, %1535, !dbg !25 + %2048 = fmul float %1536, %1536, !dbg !25 + %2049 = fmul float %1537, %1537, !dbg !25 + %2050 = fmul float %1538, %1538, !dbg !25 + %2051 = fmul float %1539, %1539, !dbg !25 + %2052 = fmul float %1540, %1540, !dbg !25 + %2053 = fmul float %1541, %1541, !dbg !25 + %2054 = fmul float %1542, %1542, !dbg !25 + %2055 = fmul float %1543, %1543, !dbg !25 + %2056 = fmul float %1544, %1544, !dbg !25 + %2057 = fmul float %1545, %1545, !dbg !25 + %2058 = fmul float %1546, %1546, !dbg !25 + %2059 = fmul float %1547, %1547, !dbg !25 + %2060 = fmul float %1548, %1548, !dbg !25 + %2061 = fmul float %1549, %1549, !dbg !25 + %2062 = fmul float %1550, %1550, !dbg !25 + %2063 = fmul float %1551, %1551, !dbg !25 + %2064 = fmul float %1552, %1552, !dbg !25 + %2065 = fmul float %1553, %1553, !dbg !25 + %2066 = fmul float %1554, %1554, !dbg !25 + %2067 = fmul float %1555, %1555, !dbg !25 + %2068 = fmul float %1556, %1556, !dbg !25 + %2069 = fmul float %1557, %1557, !dbg !25 + %2070 = fmul float %1558, %1558, !dbg !25 + %2071 = fmul float %1559, %1559, !dbg !25 + %2072 = fmul float %1560, %1560, !dbg !25 + %2073 = fmul float %1561, %1561, !dbg !25 + %2074 = fmul float %1562, %1562, !dbg !25 + %2075 = fmul float %1563, %1563, !dbg !25 + %2076 = fmul float %1564, %1564, !dbg !25 + %2077 = fmul float %1565, %1565, !dbg !25 + %2078 = fmul float %1566, %1566, !dbg !25 + %2079 = fmul float %1567, %1567, !dbg !25 + %2080 = fmul float %1568, %1568, !dbg !25 + %2081 = fmul float %1569, %1569, !dbg !25 + %2082 = fmul float %1570, %1570, !dbg !25 + %2083 = fmul float %1571, %1571, !dbg !25 + %2084 = fmul float %1572, %1572, !dbg !25 + %2085 = fmul float %1573, %1573, !dbg !25 + %2086 = fmul float %1574, %1574, !dbg !25 + %2087 = fmul float %1575, %1575, !dbg !25 + %2088 = fmul float %1576, %1576, !dbg !25 + %2089 = fmul float %1577, %1577, !dbg !25 + %2090 = fmul float %1578, %1578, !dbg !25 + %2091 = fmul float %1579, %1579, !dbg !25 + %2092 = fmul float %1580, %1580, !dbg !25 + %2093 = fmul float %1581, %1581, !dbg !25 + %2094 = fmul float %1582, %1582, !dbg !25 + %2095 = fmul float %1583, %1583, !dbg !25 + %2096 = fmul float %1584, %1584, !dbg !25 + %2097 = fmul float %1585, %1585, !dbg !25 + %2098 = fmul float %1586, %1586, !dbg !25 + %2099 = fmul float %1587, %1587, !dbg !25 + %2100 = fmul float %1588, %1588, !dbg !25 + %2101 = fmul float %1589, %1589, !dbg !25 + %2102 = fmul float %1590, %1590, !dbg !25 + %2103 = fmul float %1591, %1591, !dbg !25 + %2104 = fmul float %1592, %1592, !dbg !25 + %2105 = fmul float %1593, %1593, !dbg !25 + %2106 = fmul float %1594, %1594, !dbg !25 + %2107 = fmul float %1595, %1595, !dbg !25 + %2108 = fmul float %1596, %1596, !dbg !25 + %2109 = fmul float %1597, %1597, !dbg !25 + %2110 = fmul float %1598, %1598, !dbg !25 + %2111 = fmul float %1599, %1599, !dbg !25 + %2112 = fmul float %1600, %1600, !dbg !25 + %2113 = fmul float %1601, %1601, !dbg !25 + %2114 = fmul float %1602, %1602, !dbg !25 + %2115 = fmul float %1603, %1603, !dbg !25 + %2116 = fmul float %1604, %1604, !dbg !25 + %2117 = fmul float %1605, %1605, !dbg !25 + %2118 = fmul float %1606, %1606, !dbg !25 + %2119 = fmul float %1607, %1607, !dbg !25 + %2120 = fmul float %1608, %1608, !dbg !25 + %2121 = fmul float %1609, %1609, !dbg !25 + %2122 = fmul float %1610, %1610, !dbg !25 + %2123 = fmul float %1611, %1611, !dbg !25 + %2124 = fmul float %1612, %1612, !dbg !25 + %2125 = fmul float %1613, %1613, !dbg !25 + %2126 = fmul float %1614, %1614, !dbg !25 + %2127 = fmul float %1615, %1615, !dbg !25 + %2128 = fmul float %1616, %1616, !dbg !25 + %2129 = fmul float %1617, %1617, !dbg !25 + %2130 = fmul float %1618, %1618, !dbg !25 + %2131 = fmul float %1619, %1619, !dbg !25 + %2132 = fmul float %1620, %1620, !dbg !25 + %2133 = fmul float %1621, %1621, !dbg !25 + %2134 = fmul float %1622, %1622, !dbg !25 + %2135 = fmul float %1623, %1623, !dbg !25 + %2136 = fmul float %1624, %1624, !dbg !25 + %2137 = fmul float %1625, %1625, !dbg !25 + %2138 = fmul float %1626, %1626, !dbg !25 + %2139 = fmul float %1627, %1627, !dbg !25 + %2140 = fmul float %1628, %1628, !dbg !25 + %2141 = fmul float %1629, %1629, !dbg !25 + %2142 = fmul float %1630, %1630, !dbg !25 + %2143 = fmul float %1631, %1631, !dbg !25 + %2144 = fmul float %1632, %1632, !dbg !25 + %2145 = fmul float %1633, %1633, !dbg !25 + %2146 = fmul float %1634, %1634, !dbg !25 + %2147 = fmul float %1635, %1635, !dbg !25 + %2148 = fmul float %1636, %1636, !dbg !25 + %2149 = fmul float %1637, %1637, !dbg !25 + %2150 = fmul float %1638, %1638, !dbg !25 + %2151 = fmul float %1639, %1639, !dbg !25 + %2152 = fmul float %1640, %1640, !dbg !25 + %2153 = fmul float %1641, %1641, !dbg !25 + %2154 = fmul float %1642, %1642, !dbg !25 + %2155 = fmul float %1643, %1643, !dbg !25 + %2156 = fmul float %1644, %1644, !dbg !25 + %2157 = fmul float %1645, %1645, !dbg !25 + %2158 = fmul float %1646, %1646, !dbg !25 + %2159 = fmul float %1647, %1647, !dbg !25 + %2160 = fmul float %1648, %1648, !dbg !25 + %2161 = fmul float %1649, %1649, !dbg !25 + %2162 = fmul float %1650, %1650, !dbg !25 + %2163 = fmul float %1651, %1651, !dbg !25 + %2164 = fmul float %1652, %1652, !dbg !25 + %2165 = fmul float %1653, %1653, !dbg !25 + %2166 = fmul float %1654, %1654, !dbg !25 + %2167 = fmul float %1655, %1655, !dbg !25 + %2168 = fmul float %1656, %1656, !dbg !25 + %2169 = fmul float %1657, %1657, !dbg !25 + %2170 = fmul float %1658, %1658, !dbg !25 + %2171 = fmul float %1659, %1659, !dbg !25 + %2172 = fmul float %1660, %1660, !dbg !25 + %2173 = fmul float %1661, %1661, !dbg !25 + %2174 = fmul float %1662, %1662, !dbg !25 + %2175 = fmul float %1663, %1663, !dbg !25 + %2176 = fmul float %1664, %1664, !dbg !25 + %2177 = fmul float %1665, %1665, !dbg !25 + %2178 = fmul float %1666, %1666, !dbg !25 + %2179 = fmul float %1667, %1667, !dbg !25 + %2180 = fmul float %1668, %1668, !dbg !25 + %2181 = fmul float %1669, %1669, !dbg !25 + %2182 = fmul float %1670, %1670, !dbg !25 + %2183 = fmul float %1671, %1671, !dbg !25 + %2184 = fmul float %1672, %1672, !dbg !25 + %2185 = fmul float %1673, %1673, !dbg !25 + %2186 = fmul float %1674, %1674, !dbg !25 + %2187 = fmul float %1675, %1675, !dbg !25 + %2188 = fmul float %1676, %1676, !dbg !25 + %2189 = fmul float %1677, %1677, !dbg !25 + %2190 = fmul float %1678, %1678, !dbg !25 + %2191 = fmul float %1679, %1679, !dbg !25 + %2192 = fmul float %1680, %1680, !dbg !25 + %2193 = fmul float %1681, %1681, !dbg !25 + %2194 = fmul float %1682, %1682, !dbg !25 + %2195 = fmul float %1683, %1683, !dbg !25 + %2196 = fmul float %1684, %1684, !dbg !25 + %2197 = fmul float %1685, %1685, !dbg !25 + %2198 = fmul float %1686, %1686, !dbg !25 + %2199 = fmul float %1687, %1687, !dbg !25 + %2200 = fmul float %1688, %1688, !dbg !25 + %2201 = fmul float %1689, %1689, !dbg !25 + %2202 = fmul float %1690, %1690, !dbg !25 + %2203 = fmul float %1691, %1691, !dbg !25 + %2204 = fmul float %1692, %1692, !dbg !25 + %2205 = fmul float %1693, %1693, !dbg !25 + %2206 = fmul float %1694, %1694, !dbg !25 + %2207 = fmul float %1695, %1695, !dbg !25 + %2208 = fmul float %1696, %1696, !dbg !25 + %2209 = fmul float %1697, %1697, !dbg !25 + %2210 = fmul float %1698, %1698, !dbg !25 + %2211 = fmul float %1699, %1699, !dbg !25 + %2212 = fmul float %1700, %1700, !dbg !25 + %2213 = fmul float %1701, %1701, !dbg !25 + %2214 = fmul float %1702, %1702, !dbg !25 + %2215 = fmul float %1703, %1703, !dbg !25 + %2216 = fmul float %1704, %1704, !dbg !25 + %2217 = fmul float %1705, %1705, !dbg !25 + %2218 = fmul float %1706, %1706, !dbg !25 + %2219 = fmul float %1707, %1707, !dbg !25 + %2220 = fmul float %1708, %1708, !dbg !25 + %2221 = fmul float %1709, %1709, !dbg !25 + %2222 = fmul float %1710, %1710, !dbg !25 + %2223 = fmul float %1711, %1711, !dbg !25 + %2224 = fmul float %1712, %1712, !dbg !25 + %2225 = fmul float %1713, %1713, !dbg !25 + %2226 = fmul float %1714, %1714, !dbg !25 + %2227 = fmul float %1715, %1715, !dbg !25 + %2228 = fmul float %1716, %1716, !dbg !25 + %2229 = fmul float %1717, %1717, !dbg !25 + %2230 = fmul float %1718, %1718, !dbg !25 + %2231 = fmul float %1719, %1719, !dbg !25 + %2232 = fmul float %1720, %1720, !dbg !25 + %2233 = fmul float %1721, %1721, !dbg !25 + %2234 = fmul float %1722, %1722, !dbg !25 + %2235 = fmul float %1723, %1723, !dbg !25 + %2236 = fmul float %1724, %1724, !dbg !25 + %2237 = fmul float %1725, %1725, !dbg !25 + %2238 = fmul float %1726, %1726, !dbg !25 + %2239 = fmul float %1727, %1727, !dbg !25 + %2240 = fmul float %1728, %1728, !dbg !25 + %2241 = fmul float %1729, %1729, !dbg !25 + %2242 = fmul float %1730, %1730, !dbg !25 + %2243 = fmul float %1731, %1731, !dbg !25 + %2244 = fmul float %1732, %1732, !dbg !25 + %2245 = fmul float %1733, %1733, !dbg !25 + %2246 = fmul float %1734, %1734, !dbg !25 + %2247 = fmul float %1735, %1735, !dbg !25 + %2248 = fmul float %1736, %1736, !dbg !25 + %2249 = fmul float %1737, %1737, !dbg !25 + %2250 = fmul float %1738, %1738, !dbg !25 + %2251 = fmul float %1739, %1739, !dbg !25 + %2252 = fmul float %1740, %1740, !dbg !25 + %2253 = fmul float %1741, %1741, !dbg !25 + %2254 = fmul float %1742, %1742, !dbg !25 + %2255 = fmul float %1743, %1743, !dbg !25 + %2256 = fmul float %1744, %1744, !dbg !25 + %2257 = fmul float %1745, %1745, !dbg !25 + %2258 = fmul float %1746, %1746, !dbg !25 + %2259 = fmul float %1747, %1747, !dbg !25 + %2260 = fmul float %1748, %1748, !dbg !25 + %2261 = fmul float %1749, %1749, !dbg !25 + %2262 = fmul float %1750, %1750, !dbg !25 + %2263 = fmul float %1751, %1751, !dbg !25 + %2264 = fmul float %1752, %1752, !dbg !25 + %2265 = fmul float %1753, %1753, !dbg !25 + %2266 = fmul float %1754, %1754, !dbg !25 + %2267 = fmul float %1755, %1755, !dbg !25 + %2268 = fmul float %1756, %1756, !dbg !25 + %2269 = fmul float %1757, %1757, !dbg !25 + %2270 = fmul float %1758, %1758, !dbg !25 + %2271 = fmul float %1759, %1759, !dbg !25 + %2272 = fmul float %1760, %1760, !dbg !25 + %2273 = fmul float %1761, %1761, !dbg !25 + %2274 = fmul float %1762, %1762, !dbg !25 + %2275 = fmul float %1763, %1763, !dbg !25 + %2276 = fmul float %1764, %1764, !dbg !25 + %2277 = fmul float %1765, %1765, !dbg !25 + %2278 = fmul float %1766, %1766, !dbg !25 + %2279 = fmul float %1767, %1767, !dbg !25 + %2280 = fmul float %1768, %1768, !dbg !25 + %2281 = fmul float %1769, %1769, !dbg !25 + %2282 = fmul float %1770, %1770, !dbg !25 + %2283 = fmul float %1771, %1771, !dbg !25 + %2284 = fmul float %1772, %1772, !dbg !25 + %2285 = fmul float %1773, %1773, !dbg !25 + %2286 = fmul float %1774, %1774, !dbg !25 + %2287 = fmul float %1775, %1775, !dbg !25 + %2288 = fmul float %1776, %1776, !dbg !25 + %2289 = fmul float %1777, %1777, !dbg !25 + %2290 = fmul float %1778, %1778, !dbg !25 + %2291 = fmul float %1779, %1779, !dbg !25 + %2292 = fmul float %1780, %1780, !dbg !25 + %2293 = fmul float %1781, %1781, !dbg !25 + %2294 = fmul float %1782, %1782, !dbg !25 + %2295 = fmul float %1783, %1783, !dbg !25 + %2296 = fmul float %1784, %1784, !dbg !25 + %2297 = fmul float %1785, %1785, !dbg !25 + %2298 = fmul float %1786, %1786, !dbg !25 + %2299 = fmul float %1787, %1787, !dbg !25 + %2300 = fmul float %1788, %1788, !dbg !25 + %2301 = fmul float %1789, %1789, !dbg !25 + %2302 = fmul float %1790, %1790, !dbg !25 + %2303 = fmul float %1791, %1791, !dbg !25 + %2304 = fmul float %1792, %1792, !dbg !25 + %2305 = fmul float %1793, %1793, !dbg !25 + %2306 = fmul float %1794, %1794, !dbg !25 + %2307 = fmul float %1795, %1795, !dbg !25 + %2308 = fmul float %1796, %1796, !dbg !25 + %2309 = fmul float %1797, %1797, !dbg !25 + %2310 = fmul float %1798, %1798, !dbg !25 + %2311 = fmul float %1799, %1799, !dbg !25 + %2312 = fmul float %1800, %1800, !dbg !25 + %2313 = fmul float %1801, %1801, !dbg !25 + %2314 = fmul float %1802, %1802, !dbg !25 + %2315 = fmul float %1803, %1803, !dbg !25 + %2316 = fmul float %1804, %1804, !dbg !25 + %2317 = fmul float %1805, %1805, !dbg !25 + %2318 = fmul float %1806, %1806, !dbg !25 + %2319 = fmul float %1807, %1807, !dbg !25 + %2320 = fmul float %1808, %1808, !dbg !25 + %2321 = fmul float %1809, %1809, !dbg !25 + %2322 = fmul float %1810, %1810, !dbg !25 + %2323 = fmul float %1811, %1811, !dbg !25 + %2324 = fmul float %1812, %1812, !dbg !25 + %2325 = fmul float %1813, %1813, !dbg !25 + %2326 = fmul float %1814, %1814, !dbg !25 + %2327 = fmul float %1815, %1815, !dbg !25 + %2328 = fmul float %1816, %1816, !dbg !25 + %2329 = fmul float %1817, %1817, !dbg !25 + %2330 = fmul float %1818, %1818, !dbg !25 + %2331 = fmul float %1819, %1819, !dbg !25 + %2332 = fmul float %1820, %1820, !dbg !25 + %2333 = fmul float %1821, %1821, !dbg !25 + %2334 = fmul float %1822, %1822, !dbg !25 + %2335 = fmul float %1823, %1823, !dbg !25 + %2336 = fmul float %1824, %1824, !dbg !25 + %2337 = fmul float %1825, %1825, !dbg !25 + %2338 = fmul float %1826, %1826, !dbg !25 + %2339 = fmul float %1827, %1827, !dbg !25 + %2340 = fmul float %1828, %1828, !dbg !25 + %2341 = fmul float %1829, %1829, !dbg !25 + %2342 = fmul float %1830, %1830, !dbg !25 + %2343 = fmul float %1831, %1831, !dbg !25 + %2344 = fmul float %1832, %1832, !dbg !25 + %2345 = fmul float %1833, %1833, !dbg !25 + %2346 = fmul float %1834, %1834, !dbg !25 + %2347 = fmul float %1835, %1835, !dbg !25 + %2348 = fmul float %1836, %1836, !dbg !25 + %2349 = fmul float %1837, %1837, !dbg !25 + %2350 = fmul float %1838, %1838, !dbg !25 + %2351 = fmul float %1839, %1839, !dbg !25 + %2352 = fmul float %1840, %1840, !dbg !25 + %2353 = fmul float %1841, %1841, !dbg !25 + %2354 = fmul float %1842, %1842, !dbg !25 + %2355 = fmul float %1843, %1843, !dbg !25 + %2356 = fmul float %1844, %1844, !dbg !25 + %2357 = fmul float %1845, %1845, !dbg !25 + %2358 = fmul float %1846, %1846, !dbg !25 + %2359 = fmul float %1847, %1847, !dbg !25 + %2360 = fmul float %1848, %1848, !dbg !25 + %2361 = fmul float %1849, %1849, !dbg !25 + %2362 = fmul float %1850, %1850, !dbg !25 + %2363 = fmul float %1851, %1851, !dbg !25 + %2364 = fmul float %1852, %1852, !dbg !25 + %2365 = fmul float %1853, %1853, !dbg !25 + %2366 = fmul float %1854, %1854, !dbg !25 + %2367 = fmul float %1855, %1855, !dbg !25 + %2368 = fmul float %1856, %1856, !dbg !25 + %2369 = fmul float %1857, %1857, !dbg !25 + %2370 = fmul float %1858, %1858, !dbg !25 + %2371 = fmul float %1859, %1859, !dbg !25 + %2372 = fmul float %1860, %1860, !dbg !25 + %2373 = fmul float %1861, %1861, !dbg !25 + %2374 = fmul float %1862, %1862, !dbg !25 + %2375 = fmul float %1863, %1863, !dbg !25 + %2376 = fmul float %1864, %1864, !dbg !25 + %2377 = fmul float %1865, %1865, !dbg !25 + %2378 = fmul float %1866, %1866, !dbg !25 + %2379 = fmul float %1867, %1867, !dbg !25 + %2380 = fmul float %1868, %1868, !dbg !25 + %2381 = fmul float %1869, %1869, !dbg !25 + %2382 = fmul float %1870, %1870, !dbg !25 + %2383 = fmul float %1871, %1871, !dbg !25 + %2384 = fmul float %1872, %1872, !dbg !25 + %2385 = fmul float %1873, %1873, !dbg !25 + %2386 = fmul float %1874, %1874, !dbg !25 + %2387 = fmul float %1875, %1875, !dbg !25 + %2388 = fmul float %1876, %1876, !dbg !25 + %2389 = fmul float %1877, %1877, !dbg !25 + %2390 = fmul float %1878, %1878, !dbg !25 + %2391 = fmul float %1879, %1879, !dbg !25 + %2392 = fmul float %1880, %1880, !dbg !25 + %2393 = fmul float %1881, %1881, !dbg !25 + %2394 = fmul float %1882, %1882, !dbg !25 + %2395 = fmul float %1883, %1883, !dbg !25 + %2396 = fmul float %1884, %1884, !dbg !25 + %2397 = fmul float %1885, %1885, !dbg !25 + %2398 = fmul float %1886, %1886, !dbg !25 + %2399 = fmul float %1887, %1887, !dbg !25 + %2400 = fmul float %1888, %1888, !dbg !25 + %2401 = fmul float %1889, %1889, !dbg !25 + %2402 = fmul float %1890, %1890, !dbg !25 + %2403 = fmul float %1891, %1891, !dbg !25 + %2404 = fmul float %1892, %1892, !dbg !25 + %2405 = fmul float %1893, %1893, !dbg !25 + %2406 = fmul float %1894, %1894, !dbg !25 + %2407 = fmul float %1895, %1895, !dbg !25 + %2408 = fmul float %1896, %1896, !dbg !25 + %2409 = fmul float %1897, %1897, !dbg !25 + %2410 = fmul float %1898, %1898, !dbg !25 + %2411 = fmul float %1899, %1899, !dbg !25 + %2412 = fmul float %1900, %1900, !dbg !25 + %2413 = fmul float %1901, %1901, !dbg !25 + %2414 = fmul float %1902, %1902, !dbg !25 + %2415 = fmul float %1903, %1903, !dbg !25 + %2416 = fmul float %1904, %1904, !dbg !25 + %2417 = fmul float %1905, %1905, !dbg !25 + %2418 = fmul float %1906, %1906, !dbg !25 + %2419 = fmul float %1907, %1907, !dbg !25 + %2420 = fmul float %1908, %1908, !dbg !25 + %2421 = fmul float %1909, %1909, !dbg !25 + %2422 = fmul float %1910, %1910, !dbg !25 + %2423 = fmul float %1911, %1911, !dbg !25 + %2424 = fmul float %1912, %1912, !dbg !25 + %2425 = fmul float %1913, %1913, !dbg !25 + %2426 = fmul float %1914, %1914, !dbg !25 + %2427 = fmul float %1915, %1915, !dbg !25 + %2428 = fmul float %1916, %1916, !dbg !25 + %2429 = fmul float %1917, %1917, !dbg !25 + %2430 = fmul float %1918, %1918, !dbg !25 + %2431 = fmul float %1919, %1919, !dbg !25 + %2432 = fmul float %1920, %1920, !dbg !25 + %2433 = fmul float %1921, %1921, !dbg !25 + %2434 = fmul float %1922, %1922, !dbg !25 + %2435 = fmul float %1923, %1923, !dbg !25 + %2436 = fmul float %1924, %1924, !dbg !25 + %2437 = fmul float %1925, %1925, !dbg !25 + %2438 = fmul float %1926, %1926, !dbg !25 + %2439 = fmul float %1927, %1927, !dbg !25 + %2440 = fmul float %1928, %1928, !dbg !25 + %2441 = fmul float %1929, %1929, !dbg !25 + %2442 = fmul float %1930, %1930, !dbg !25 + %2443 = fmul float %1931, %1931, !dbg !25 + %2444 = fmul float %1932, %1932, !dbg !25 + %2445 = fmul float %1933, %1933, !dbg !25 + %2446 = fmul float %1934, %1934, !dbg !25 + %2447 = fmul float %1935, %1935, !dbg !25 + %2448 = fmul float %1936, %1936, !dbg !25 + %2449 = fmul float %1937, %1937, !dbg !25 + %2450 = fmul float %1938, %1938, !dbg !25 + %2451 = fmul float %1939, %1939, !dbg !25 + %2452 = fmul float %1940, %1940, !dbg !25 + %2453 = fmul float %1941, %1941, !dbg !25 + %2454 = fmul float %1942, %1942, !dbg !25 + %2455 = fmul float %1943, %1943, !dbg !25 + %2456 = fmul float %1944, %1944, !dbg !25 + %2457 = fmul float %1945, %1945, !dbg !25 + %2458 = fmul float %1946, %1946, !dbg !25 + %2459 = fmul float %1947, %1947, !dbg !25 + %2460 = fmul float %1948, %1948, !dbg !25 + %2461 = fmul float %1949, %1949, !dbg !25 + %2462 = fmul float %1950, %1950, !dbg !25 + %2463 = fmul float %1951, %1951, !dbg !25 + %2464 = fmul float %1952, %1952, !dbg !25 + %2465 = fmul float %1953, %1953, !dbg !25 + %2466 = fmul float %1954, %1954, !dbg !25 + %2467 = fmul float %1955, %1955, !dbg !25 + %2468 = fmul float %1956, %1956, !dbg !25 + %2469 = fmul float %1957, %1957, !dbg !25 + %2470 = fmul float %1958, %1958, !dbg !25 + %2471 = fmul float %1959, %1959, !dbg !25 + %2472 = fmul float %1960, %1960, !dbg !25 + %2473 = fmul float %1961, %1961, !dbg !25 + %2474 = fmul float %1962, %1962, !dbg !25 + %2475 = fmul float %1963, %1963, !dbg !25 + %2476 = fmul float %1964, %1964, !dbg !25 + %2477 = fmul float %1965, %1965, !dbg !25 + %2478 = fmul float %1966, %1966, !dbg !25 + %2479 = fadd float %1967, %1968, !dbg !26 + %2480 = fadd float %1969, %2479, !dbg !26 + %2481 = fadd float %1970, %2480, !dbg !26 + %2482 = fadd float %1971, %2481, !dbg !26 + %2483 = fadd float %1972, %2482, !dbg !26 + %2484 = fadd float %1973, %2483, !dbg !26 + %2485 = fadd float %1974, %2484, !dbg !26 + %2486 = fadd float %1975, %1976, !dbg !26 + %2487 = fadd float %1977, %2486, !dbg !26 + %2488 = fadd float %1978, %2487, !dbg !26 + %2489 = fadd float %1979, %2488, !dbg !26 + %2490 = fadd float %1980, %2489, !dbg !26 + %2491 = fadd float %1981, %2490, !dbg !26 + %2492 = fadd float %1982, %2491, !dbg !26 + %2493 = fadd float %1983, %1984, !dbg !26 + %2494 = fadd float %1985, %2493, !dbg !26 + %2495 = fadd float %1986, %2494, !dbg !26 + %2496 = fadd float %1987, %2495, !dbg !26 + %2497 = fadd float %1988, %2496, !dbg !26 + %2498 = fadd float %1989, %2497, !dbg !26 + %2499 = fadd float %1990, %2498, !dbg !26 + %2500 = fadd float %1991, %1992, !dbg !26 + %2501 = fadd float %1993, %2500, !dbg !26 + %2502 = fadd float %1994, %2501, !dbg !26 + %2503 = fadd float %1995, %2502, !dbg !26 + %2504 = fadd float %1996, %2503, !dbg !26 + %2505 = fadd float %1997, %2504, !dbg !26 + %2506 = fadd float %1998, %2505, !dbg !26 + %2507 = fadd float %1999, %2000, !dbg !26 + %2508 = fadd float %2001, %2507, !dbg !26 + %2509 = fadd float %2002, %2508, !dbg !26 + %2510 = fadd float %2003, %2509, !dbg !26 + %2511 = fadd float %2004, %2510, !dbg !26 + %2512 = fadd float %2005, %2511, !dbg !26 + %2513 = fadd float %2006, %2512, !dbg !26 + %2514 = fadd float %2007, %2008, !dbg !26 + %2515 = fadd float %2009, %2514, !dbg !26 + %2516 = fadd float %2010, %2515, !dbg !26 + %2517 = fadd float %2011, %2516, !dbg !26 + %2518 = fadd float %2012, %2517, !dbg !26 + %2519 = fadd float %2013, %2518, !dbg !26 + %2520 = fadd float %2014, %2519, !dbg !26 + %2521 = fadd float %2015, %2016, !dbg !26 + %2522 = fadd float %2017, %2521, !dbg !26 + %2523 = fadd float %2018, %2522, !dbg !26 + %2524 = fadd float %2019, %2523, !dbg !26 + %2525 = fadd float %2020, %2524, !dbg !26 + %2526 = fadd float %2021, %2525, !dbg !26 + %2527 = fadd float %2022, %2526, !dbg !26 + %2528 = fadd float %2023, %2024, !dbg !26 + %2529 = fadd float %2025, %2528, !dbg !26 + %2530 = fadd float %2026, %2529, !dbg !26 + %2531 = fadd float %2027, %2530, !dbg !26 + %2532 = fadd float %2028, %2531, !dbg !26 + %2533 = fadd float %2029, %2532, !dbg !26 + %2534 = fadd float %2030, %2533, !dbg !26 + %2535 = fadd float %2031, %2032, !dbg !26 + %2536 = fadd float %2033, %2535, !dbg !26 + %2537 = fadd float %2034, %2536, !dbg !26 + %2538 = fadd float %2035, %2537, !dbg !26 + %2539 = fadd float %2036, %2538, !dbg !26 + %2540 = fadd float %2037, %2539, !dbg !26 + %2541 = fadd float %2038, %2540, !dbg !26 + %2542 = fadd float %2039, %2040, !dbg !26 + %2543 = fadd float %2041, %2542, !dbg !26 + %2544 = fadd float %2042, %2543, !dbg !26 + %2545 = fadd float %2043, %2544, !dbg !26 + %2546 = fadd float %2044, %2545, !dbg !26 + %2547 = fadd float %2045, %2546, !dbg !26 + %2548 = fadd float %2046, %2547, !dbg !26 + %2549 = fadd float %2047, %2048, !dbg !26 + %2550 = fadd float %2049, %2549, !dbg !26 + %2551 = fadd float %2050, %2550, !dbg !26 + %2552 = fadd float %2051, %2551, !dbg !26 + %2553 = fadd float %2052, %2552, !dbg !26 + %2554 = fadd float %2053, %2553, !dbg !26 + %2555 = fadd float %2054, %2554, !dbg !26 + %2556 = fadd float %2055, %2056, !dbg !26 + %2557 = fadd float %2057, %2556, !dbg !26 + %2558 = fadd float %2058, %2557, !dbg !26 + %2559 = fadd float %2059, %2558, !dbg !26 + %2560 = fadd float %2060, %2559, !dbg !26 + %2561 = fadd float %2061, %2560, !dbg !26 + %2562 = fadd float %2062, %2561, !dbg !26 + %2563 = fadd float %2063, %2064, !dbg !26 + %2564 = fadd float %2065, %2563, !dbg !26 + %2565 = fadd float %2066, %2564, !dbg !26 + %2566 = fadd float %2067, %2565, !dbg !26 + %2567 = fadd float %2068, %2566, !dbg !26 + %2568 = fadd float %2069, %2567, !dbg !26 + %2569 = fadd float %2070, %2568, !dbg !26 + %2570 = fadd float %2071, %2072, !dbg !26 + %2571 = fadd float %2073, %2570, !dbg !26 + %2572 = fadd float %2074, %2571, !dbg !26 + %2573 = fadd float %2075, %2572, !dbg !26 + %2574 = fadd float %2076, %2573, !dbg !26 + %2575 = fadd float %2077, %2574, !dbg !26 + %2576 = fadd float %2078, %2575, !dbg !26 + %2577 = fadd float %2079, %2080, !dbg !26 + %2578 = fadd float %2081, %2577, !dbg !26 + %2579 = fadd float %2082, %2578, !dbg !26 + %2580 = fadd float %2083, %2579, !dbg !26 + %2581 = fadd float %2084, %2580, !dbg !26 + %2582 = fadd float %2085, %2581, !dbg !26 + %2583 = fadd float %2086, %2582, !dbg !26 + %2584 = fadd float %2087, %2088, !dbg !26 + %2585 = fadd float %2089, %2584, !dbg !26 + %2586 = fadd float %2090, %2585, !dbg !26 + %2587 = fadd float %2091, %2586, !dbg !26 + %2588 = fadd float %2092, %2587, !dbg !26 + %2589 = fadd float %2093, %2588, !dbg !26 + %2590 = fadd float %2094, %2589, !dbg !26 + %2591 = fadd float %2095, %2096, !dbg !26 + %2592 = fadd float %2097, %2591, !dbg !26 + %2593 = fadd float %2098, %2592, !dbg !26 + %2594 = fadd float %2099, %2593, !dbg !26 + %2595 = fadd float %2100, %2594, !dbg !26 + %2596 = fadd float %2101, %2595, !dbg !26 + %2597 = fadd float %2102, %2596, !dbg !26 + %2598 = fadd float %2103, %2104, !dbg !26 + %2599 = fadd float %2105, %2598, !dbg !26 + %2600 = fadd float %2106, %2599, !dbg !26 + %2601 = fadd float %2107, %2600, !dbg !26 + %2602 = fadd float %2108, %2601, !dbg !26 + %2603 = fadd float %2109, %2602, !dbg !26 + %2604 = fadd float %2110, %2603, !dbg !26 + %2605 = fadd float %2111, %2112, !dbg !26 + %2606 = fadd float %2113, %2605, !dbg !26 + %2607 = fadd float %2114, %2606, !dbg !26 + %2608 = fadd float %2115, %2607, !dbg !26 + %2609 = fadd float %2116, %2608, !dbg !26 + %2610 = fadd float %2117, %2609, !dbg !26 + %2611 = fadd float %2118, %2610, !dbg !26 + %2612 = fadd float %2119, %2120, !dbg !26 + %2613 = fadd float %2121, %2612, !dbg !26 + %2614 = fadd float %2122, %2613, !dbg !26 + %2615 = fadd float %2123, %2614, !dbg !26 + %2616 = fadd float %2124, %2615, !dbg !26 + %2617 = fadd float %2125, %2616, !dbg !26 + %2618 = fadd float %2126, %2617, !dbg !26 + %2619 = fadd float %2127, %2128, !dbg !26 + %2620 = fadd float %2129, %2619, !dbg !26 + %2621 = fadd float %2130, %2620, !dbg !26 + %2622 = fadd float %2131, %2621, !dbg !26 + %2623 = fadd float %2132, %2622, !dbg !26 + %2624 = fadd float %2133, %2623, !dbg !26 + %2625 = fadd float %2134, %2624, !dbg !26 + %2626 = fadd float %2135, %2136, !dbg !26 + %2627 = fadd float %2137, %2626, !dbg !26 + %2628 = fadd float %2138, %2627, !dbg !26 + %2629 = fadd float %2139, %2628, !dbg !26 + %2630 = fadd float %2140, %2629, !dbg !26 + %2631 = fadd float %2141, %2630, !dbg !26 + %2632 = fadd float %2142, %2631, !dbg !26 + %2633 = fadd float %2143, %2144, !dbg !26 + %2634 = fadd float %2145, %2633, !dbg !26 + %2635 = fadd float %2146, %2634, !dbg !26 + %2636 = fadd float %2147, %2635, !dbg !26 + %2637 = fadd float %2148, %2636, !dbg !26 + %2638 = fadd float %2149, %2637, !dbg !26 + %2639 = fadd float %2150, %2638, !dbg !26 + %2640 = fadd float %2151, %2152, !dbg !26 + %2641 = fadd float %2153, %2640, !dbg !26 + %2642 = fadd float %2154, %2641, !dbg !26 + %2643 = fadd float %2155, %2642, !dbg !26 + %2644 = fadd float %2156, %2643, !dbg !26 + %2645 = fadd float %2157, %2644, !dbg !26 + %2646 = fadd float %2158, %2645, !dbg !26 + %2647 = fadd float %2159, %2160, !dbg !26 + %2648 = fadd float %2161, %2647, !dbg !26 + %2649 = fadd float %2162, %2648, !dbg !26 + %2650 = fadd float %2163, %2649, !dbg !26 + %2651 = fadd float %2164, %2650, !dbg !26 + %2652 = fadd float %2165, %2651, !dbg !26 + %2653 = fadd float %2166, %2652, !dbg !26 + %2654 = fadd float %2167, %2168, !dbg !26 + %2655 = fadd float %2169, %2654, !dbg !26 + %2656 = fadd float %2170, %2655, !dbg !26 + %2657 = fadd float %2171, %2656, !dbg !26 + %2658 = fadd float %2172, %2657, !dbg !26 + %2659 = fadd float %2173, %2658, !dbg !26 + %2660 = fadd float %2174, %2659, !dbg !26 + %2661 = fadd float %2175, %2176, !dbg !26 + %2662 = fadd float %2177, %2661, !dbg !26 + %2663 = fadd float %2178, %2662, !dbg !26 + %2664 = fadd float %2179, %2663, !dbg !26 + %2665 = fadd float %2180, %2664, !dbg !26 + %2666 = fadd float %2181, %2665, !dbg !26 + %2667 = fadd float %2182, %2666, !dbg !26 + %2668 = fadd float %2183, %2184, !dbg !26 + %2669 = fadd float %2185, %2668, !dbg !26 + %2670 = fadd float %2186, %2669, !dbg !26 + %2671 = fadd float %2187, %2670, !dbg !26 + %2672 = fadd float %2188, %2671, !dbg !26 + %2673 = fadd float %2189, %2672, !dbg !26 + %2674 = fadd float %2190, %2673, !dbg !26 + %2675 = fadd float %2191, %2192, !dbg !26 + %2676 = fadd float %2193, %2675, !dbg !26 + %2677 = fadd float %2194, %2676, !dbg !26 + %2678 = fadd float %2195, %2677, !dbg !26 + %2679 = fadd float %2196, %2678, !dbg !26 + %2680 = fadd float %2197, %2679, !dbg !26 + %2681 = fadd float %2198, %2680, !dbg !26 + %2682 = fadd float %2199, %2200, !dbg !26 + %2683 = fadd float %2201, %2682, !dbg !26 + %2684 = fadd float %2202, %2683, !dbg !26 + %2685 = fadd float %2203, %2684, !dbg !26 + %2686 = fadd float %2204, %2685, !dbg !26 + %2687 = fadd float %2205, %2686, !dbg !26 + %2688 = fadd float %2206, %2687, !dbg !26 + %2689 = fadd float %2207, %2208, !dbg !26 + %2690 = fadd float %2209, %2689, !dbg !26 + %2691 = fadd float %2210, %2690, !dbg !26 + %2692 = fadd float %2211, %2691, !dbg !26 + %2693 = fadd float %2212, %2692, !dbg !26 + %2694 = fadd float %2213, %2693, !dbg !26 + %2695 = fadd float %2214, %2694, !dbg !26 + %2696 = fadd float %2215, %2216, !dbg !26 + %2697 = fadd float %2217, %2696, !dbg !26 + %2698 = fadd float %2218, %2697, !dbg !26 + %2699 = fadd float %2219, %2698, !dbg !26 + %2700 = fadd float %2220, %2699, !dbg !26 + %2701 = fadd float %2221, %2700, !dbg !26 + %2702 = fadd float %2222, %2701, !dbg !26 + %2703 = fadd float %2223, %2224, !dbg !26 + %2704 = fadd float %2225, %2703, !dbg !26 + %2705 = fadd float %2226, %2704, !dbg !26 + %2706 = fadd float %2227, %2705, !dbg !26 + %2707 = fadd float %2228, %2706, !dbg !26 + %2708 = fadd float %2229, %2707, !dbg !26 + %2709 = fadd float %2230, %2708, !dbg !26 + %2710 = fadd float %2231, %2232, !dbg !26 + %2711 = fadd float %2233, %2710, !dbg !26 + %2712 = fadd float %2234, %2711, !dbg !26 + %2713 = fadd float %2235, %2712, !dbg !26 + %2714 = fadd float %2236, %2713, !dbg !26 + %2715 = fadd float %2237, %2714, !dbg !26 + %2716 = fadd float %2238, %2715, !dbg !26 + %2717 = fadd float %2239, %2240, !dbg !26 + %2718 = fadd float %2241, %2717, !dbg !26 + %2719 = fadd float %2242, %2718, !dbg !26 + %2720 = fadd float %2243, %2719, !dbg !26 + %2721 = fadd float %2244, %2720, !dbg !26 + %2722 = fadd float %2245, %2721, !dbg !26 + %2723 = fadd float %2246, %2722, !dbg !26 + %2724 = fadd float %2247, %2248, !dbg !26 + %2725 = fadd float %2249, %2724, !dbg !26 + %2726 = fadd float %2250, %2725, !dbg !26 + %2727 = fadd float %2251, %2726, !dbg !26 + %2728 = fadd float %2252, %2727, !dbg !26 + %2729 = fadd float %2253, %2728, !dbg !26 + %2730 = fadd float %2254, %2729, !dbg !26 + %2731 = fadd float %2255, %2256, !dbg !26 + %2732 = fadd float %2257, %2731, !dbg !26 + %2733 = fadd float %2258, %2732, !dbg !26 + %2734 = fadd float %2259, %2733, !dbg !26 + %2735 = fadd float %2260, %2734, !dbg !26 + %2736 = fadd float %2261, %2735, !dbg !26 + %2737 = fadd float %2262, %2736, !dbg !26 + %2738 = fadd float %2263, %2264, !dbg !26 + %2739 = fadd float %2265, %2738, !dbg !26 + %2740 = fadd float %2266, %2739, !dbg !26 + %2741 = fadd float %2267, %2740, !dbg !26 + %2742 = fadd float %2268, %2741, !dbg !26 + %2743 = fadd float %2269, %2742, !dbg !26 + %2744 = fadd float %2270, %2743, !dbg !26 + %2745 = fadd float %2271, %2272, !dbg !26 + %2746 = fadd float %2273, %2745, !dbg !26 + %2747 = fadd float %2274, %2746, !dbg !26 + %2748 = fadd float %2275, %2747, !dbg !26 + %2749 = fadd float %2276, %2748, !dbg !26 + %2750 = fadd float %2277, %2749, !dbg !26 + %2751 = fadd float %2278, %2750, !dbg !26 + %2752 = fadd float %2279, %2280, !dbg !26 + %2753 = fadd float %2281, %2752, !dbg !26 + %2754 = fadd float %2282, %2753, !dbg !26 + %2755 = fadd float %2283, %2754, !dbg !26 + %2756 = fadd float %2284, %2755, !dbg !26 + %2757 = fadd float %2285, %2756, !dbg !26 + %2758 = fadd float %2286, %2757, !dbg !26 + %2759 = fadd float %2287, %2288, !dbg !26 + %2760 = fadd float %2289, %2759, !dbg !26 + %2761 = fadd float %2290, %2760, !dbg !26 + %2762 = fadd float %2291, %2761, !dbg !26 + %2763 = fadd float %2292, %2762, !dbg !26 + %2764 = fadd float %2293, %2763, !dbg !26 + %2765 = fadd float %2294, %2764, !dbg !26 + %2766 = fadd float %2295, %2296, !dbg !26 + %2767 = fadd float %2297, %2766, !dbg !26 + %2768 = fadd float %2298, %2767, !dbg !26 + %2769 = fadd float %2299, %2768, !dbg !26 + %2770 = fadd float %2300, %2769, !dbg !26 + %2771 = fadd float %2301, %2770, !dbg !26 + %2772 = fadd float %2302, %2771, !dbg !26 + %2773 = fadd float %2303, %2304, !dbg !26 + %2774 = fadd float %2305, %2773, !dbg !26 + %2775 = fadd float %2306, %2774, !dbg !26 + %2776 = fadd float %2307, %2775, !dbg !26 + %2777 = fadd float %2308, %2776, !dbg !26 + %2778 = fadd float %2309, %2777, !dbg !26 + %2779 = fadd float %2310, %2778, !dbg !26 + %2780 = fadd float %2311, %2312, !dbg !26 + %2781 = fadd float %2313, %2780, !dbg !26 + %2782 = fadd float %2314, %2781, !dbg !26 + %2783 = fadd float %2315, %2782, !dbg !26 + %2784 = fadd float %2316, %2783, !dbg !26 + %2785 = fadd float %2317, %2784, !dbg !26 + %2786 = fadd float %2318, %2785, !dbg !26 + %2787 = fadd float %2319, %2320, !dbg !26 + %2788 = fadd float %2321, %2787, !dbg !26 + %2789 = fadd float %2322, %2788, !dbg !26 + %2790 = fadd float %2323, %2789, !dbg !26 + %2791 = fadd float %2324, %2790, !dbg !26 + %2792 = fadd float %2325, %2791, !dbg !26 + %2793 = fadd float %2326, %2792, !dbg !26 + %2794 = fadd float %2327, %2328, !dbg !26 + %2795 = fadd float %2329, %2794, !dbg !26 + %2796 = fadd float %2330, %2795, !dbg !26 + %2797 = fadd float %2331, %2796, !dbg !26 + %2798 = fadd float %2332, %2797, !dbg !26 + %2799 = fadd float %2333, %2798, !dbg !26 + %2800 = fadd float %2334, %2799, !dbg !26 + %2801 = fadd float %2335, %2336, !dbg !26 + %2802 = fadd float %2337, %2801, !dbg !26 + %2803 = fadd float %2338, %2802, !dbg !26 + %2804 = fadd float %2339, %2803, !dbg !26 + %2805 = fadd float %2340, %2804, !dbg !26 + %2806 = fadd float %2341, %2805, !dbg !26 + %2807 = fadd float %2342, %2806, !dbg !26 + %2808 = fadd float %2343, %2344, !dbg !26 + %2809 = fadd float %2345, %2808, !dbg !26 + %2810 = fadd float %2346, %2809, !dbg !26 + %2811 = fadd float %2347, %2810, !dbg !26 + %2812 = fadd float %2348, %2811, !dbg !26 + %2813 = fadd float %2349, %2812, !dbg !26 + %2814 = fadd float %2350, %2813, !dbg !26 + %2815 = fadd float %2351, %2352, !dbg !26 + %2816 = fadd float %2353, %2815, !dbg !26 + %2817 = fadd float %2354, %2816, !dbg !26 + %2818 = fadd float %2355, %2817, !dbg !26 + %2819 = fadd float %2356, %2818, !dbg !26 + %2820 = fadd float %2357, %2819, !dbg !26 + %2821 = fadd float %2358, %2820, !dbg !26 + %2822 = fadd float %2359, %2360, !dbg !26 + %2823 = fadd float %2361, %2822, !dbg !26 + %2824 = fadd float %2362, %2823, !dbg !26 + %2825 = fadd float %2363, %2824, !dbg !26 + %2826 = fadd float %2364, %2825, !dbg !26 + %2827 = fadd float %2365, %2826, !dbg !26 + %2828 = fadd float %2366, %2827, !dbg !26 + %2829 = fadd float %2367, %2368, !dbg !26 + %2830 = fadd float %2369, %2829, !dbg !26 + %2831 = fadd float %2370, %2830, !dbg !26 + %2832 = fadd float %2371, %2831, !dbg !26 + %2833 = fadd float %2372, %2832, !dbg !26 + %2834 = fadd float %2373, %2833, !dbg !26 + %2835 = fadd float %2374, %2834, !dbg !26 + %2836 = fadd float %2375, %2376, !dbg !26 + %2837 = fadd float %2377, %2836, !dbg !26 + %2838 = fadd float %2378, %2837, !dbg !26 + %2839 = fadd float %2379, %2838, !dbg !26 + %2840 = fadd float %2380, %2839, !dbg !26 + %2841 = fadd float %2381, %2840, !dbg !26 + %2842 = fadd float %2382, %2841, !dbg !26 + %2843 = fadd float %2383, %2384, !dbg !26 + %2844 = fadd float %2385, %2843, !dbg !26 + %2845 = fadd float %2386, %2844, !dbg !26 + %2846 = fadd float %2387, %2845, !dbg !26 + %2847 = fadd float %2388, %2846, !dbg !26 + %2848 = fadd float %2389, %2847, !dbg !26 + %2849 = fadd float %2390, %2848, !dbg !26 + %2850 = fadd float %2391, %2392, !dbg !26 + %2851 = fadd float %2393, %2850, !dbg !26 + %2852 = fadd float %2394, %2851, !dbg !26 + %2853 = fadd float %2395, %2852, !dbg !26 + %2854 = fadd float %2396, %2853, !dbg !26 + %2855 = fadd float %2397, %2854, !dbg !26 + %2856 = fadd float %2398, %2855, !dbg !26 + %2857 = fadd float %2399, %2400, !dbg !26 + %2858 = fadd float %2401, %2857, !dbg !26 + %2859 = fadd float %2402, %2858, !dbg !26 + %2860 = fadd float %2403, %2859, !dbg !26 + %2861 = fadd float %2404, %2860, !dbg !26 + %2862 = fadd float %2405, %2861, !dbg !26 + %2863 = fadd float %2406, %2862, !dbg !26 + %2864 = fadd float %2407, %2408, !dbg !26 + %2865 = fadd float %2409, %2864, !dbg !26 + %2866 = fadd float %2410, %2865, !dbg !26 + %2867 = fadd float %2411, %2866, !dbg !26 + %2868 = fadd float %2412, %2867, !dbg !26 + %2869 = fadd float %2413, %2868, !dbg !26 + %2870 = fadd float %2414, %2869, !dbg !26 + %2871 = fadd float %2415, %2416, !dbg !26 + %2872 = fadd float %2417, %2871, !dbg !26 + %2873 = fadd float %2418, %2872, !dbg !26 + %2874 = fadd float %2419, %2873, !dbg !26 + %2875 = fadd float %2420, %2874, !dbg !26 + %2876 = fadd float %2421, %2875, !dbg !26 + %2877 = fadd float %2422, %2876, !dbg !26 + %2878 = fadd float %2423, %2424, !dbg !26 + %2879 = fadd float %2425, %2878, !dbg !26 + %2880 = fadd float %2426, %2879, !dbg !26 + %2881 = fadd float %2427, %2880, !dbg !26 + %2882 = fadd float %2428, %2881, !dbg !26 + %2883 = fadd float %2429, %2882, !dbg !26 + %2884 = fadd float %2430, %2883, !dbg !26 + %2885 = fadd float %2431, %2432, !dbg !26 + %2886 = fadd float %2433, %2885, !dbg !26 + %2887 = fadd float %2434, %2886, !dbg !26 + %2888 = fadd float %2435, %2887, !dbg !26 + %2889 = fadd float %2436, %2888, !dbg !26 + %2890 = fadd float %2437, %2889, !dbg !26 + %2891 = fadd float %2438, %2890, !dbg !26 + %2892 = fadd float %2439, %2440, !dbg !26 + %2893 = fadd float %2441, %2892, !dbg !26 + %2894 = fadd float %2442, %2893, !dbg !26 + %2895 = fadd float %2443, %2894, !dbg !26 + %2896 = fadd float %2444, %2895, !dbg !26 + %2897 = fadd float %2445, %2896, !dbg !26 + %2898 = fadd float %2446, %2897, !dbg !26 + %2899 = fadd float %2447, %2448, !dbg !26 + %2900 = fadd float %2449, %2899, !dbg !26 + %2901 = fadd float %2450, %2900, !dbg !26 + %2902 = fadd float %2451, %2901, !dbg !26 + %2903 = fadd float %2452, %2902, !dbg !26 + %2904 = fadd float %2453, %2903, !dbg !26 + %2905 = fadd float %2454, %2904, !dbg !26 + %2906 = fadd float %2455, %2456, !dbg !26 + %2907 = fadd float %2457, %2906, !dbg !26 + %2908 = fadd float %2458, %2907, !dbg !26 + %2909 = fadd float %2459, %2908, !dbg !26 + %2910 = fadd float %2460, %2909, !dbg !26 + %2911 = fadd float %2461, %2910, !dbg !26 + %2912 = fadd float %2462, %2911, !dbg !26 + %2913 = fadd float %2463, %2464, !dbg !26 + %2914 = fadd float %2465, %2913, !dbg !26 + %2915 = fadd float %2466, %2914, !dbg !26 + %2916 = fadd float %2467, %2915, !dbg !26 + %2917 = fadd float %2468, %2916, !dbg !26 + %2918 = fadd float %2469, %2917, !dbg !26 + %2919 = fadd float %2470, %2918, !dbg !26 + %2920 = fadd float %2471, %2472, !dbg !26 + %2921 = fadd float %2473, %2920, !dbg !26 + %2922 = fadd float %2474, %2921, !dbg !26 + %2923 = fadd float %2475, %2922, !dbg !26 + %2924 = fadd float %2476, %2923, !dbg !26 + %2925 = fadd float %2477, %2924, !dbg !26 + %2926 = fadd float %2478, %2925, !dbg !26 + %2927 = bitcast float %2485 to i32, !dbg !29 + %2928 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2927, i32 8, i32 31), !dbg !29 + %2929 = bitcast i32 %2928 to float, !dbg !29 + %2930 = fadd float %2485, %2929, !dbg !26 + %2931 = bitcast float %2930 to i32, !dbg !29 + %2932 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2931, i32 4, i32 31), !dbg !29 + %2933 = bitcast i32 %2932 to float, !dbg !29 + %2934 = fadd float %2930, %2933, !dbg !26 + %2935 = bitcast float %2934 to i32, !dbg !29 + %2936 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2935, i32 2, i32 31), !dbg !29 + %2937 = bitcast i32 %2936 to float, !dbg !29 + %2938 = fadd float %2934, %2937, !dbg !26 + %2939 = bitcast float %2938 to i32, !dbg !29 + %2940 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2939, i32 1, i32 31), !dbg !29 + %2941 = bitcast i32 %2940 to float, !dbg !29 + %2942 = fadd float %2938, %2941, !dbg !26 + %2943 = bitcast float %2492 to i32, !dbg !29 + %2944 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2943, i32 8, i32 31), !dbg !29 + %2945 = bitcast i32 %2944 to float, !dbg !29 + %2946 = fadd float %2492, %2945, !dbg !26 + %2947 = bitcast float %2946 to i32, !dbg !29 + %2948 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2947, i32 4, i32 31), !dbg !29 + %2949 = bitcast i32 %2948 to float, !dbg !29 + %2950 = fadd float %2946, %2949, !dbg !26 + %2951 = bitcast float %2950 to i32, !dbg !29 + %2952 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2951, i32 2, i32 31), !dbg !29 + %2953 = bitcast i32 %2952 to float, !dbg !29 + %2954 = fadd float %2950, %2953, !dbg !26 + %2955 = bitcast float %2954 to i32, !dbg !29 + %2956 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2955, i32 1, i32 31), !dbg !29 + %2957 = bitcast i32 %2956 to float, !dbg !29 + %2958 = fadd float %2954, %2957, !dbg !26 + %2959 = bitcast float %2499 to i32, !dbg !29 + %2960 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2959, i32 8, i32 31), !dbg !29 + %2961 = bitcast i32 %2960 to float, !dbg !29 + %2962 = fadd float %2499, %2961, !dbg !26 + %2963 = bitcast float %2962 to i32, !dbg !29 + %2964 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2963, i32 4, i32 31), !dbg !29 + %2965 = bitcast i32 %2964 to float, !dbg !29 + %2966 = fadd float %2962, %2965, !dbg !26 + %2967 = bitcast float %2966 to i32, !dbg !29 + %2968 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2967, i32 2, i32 31), !dbg !29 + %2969 = bitcast i32 %2968 to float, !dbg !29 + %2970 = fadd float %2966, %2969, !dbg !26 + %2971 = bitcast float %2970 to i32, !dbg !29 + %2972 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2971, i32 1, i32 31), !dbg !29 + %2973 = bitcast i32 %2972 to float, !dbg !29 + %2974 = fadd float %2970, %2973, !dbg !26 + %2975 = bitcast float %2506 to i32, !dbg !29 + %2976 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2975, i32 8, i32 31), !dbg !29 + %2977 = bitcast i32 %2976 to float, !dbg !29 + %2978 = fadd float %2506, %2977, !dbg !26 + %2979 = bitcast float %2978 to i32, !dbg !29 + %2980 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2979, i32 4, i32 31), !dbg !29 + %2981 = bitcast i32 %2980 to float, !dbg !29 + %2982 = fadd float %2978, %2981, !dbg !26 + %2983 = bitcast float %2982 to i32, !dbg !29 + %2984 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2983, i32 2, i32 31), !dbg !29 + %2985 = bitcast i32 %2984 to float, !dbg !29 + %2986 = fadd float %2982, %2985, !dbg !26 + %2987 = bitcast float %2986 to i32, !dbg !29 + %2988 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2987, i32 1, i32 31), !dbg !29 + %2989 = bitcast i32 %2988 to float, !dbg !29 + %2990 = fadd float %2986, %2989, !dbg !26 + %2991 = bitcast float %2513 to i32, !dbg !29 + %2992 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2991, i32 8, i32 31), !dbg !29 + %2993 = bitcast i32 %2992 to float, !dbg !29 + %2994 = fadd float %2513, %2993, !dbg !26 + %2995 = bitcast float %2994 to i32, !dbg !29 + %2996 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2995, i32 4, i32 31), !dbg !29 + %2997 = bitcast i32 %2996 to float, !dbg !29 + %2998 = fadd float %2994, %2997, !dbg !26 + %2999 = bitcast float %2998 to i32, !dbg !29 + %3000 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2999, i32 2, i32 31), !dbg !29 + %3001 = bitcast i32 %3000 to float, !dbg !29 + %3002 = fadd float %2998, %3001, !dbg !26 + %3003 = bitcast float %3002 to i32, !dbg !29 + %3004 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3003, i32 1, i32 31), !dbg !29 + %3005 = bitcast i32 %3004 to float, !dbg !29 + %3006 = fadd float %3002, %3005, !dbg !26 + %3007 = bitcast float %2520 to i32, !dbg !29 + %3008 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3007, i32 8, i32 31), !dbg !29 + %3009 = bitcast i32 %3008 to float, !dbg !29 + %3010 = fadd float %2520, %3009, !dbg !26 + %3011 = bitcast float %3010 to i32, !dbg !29 + %3012 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3011, i32 4, i32 31), !dbg !29 + %3013 = bitcast i32 %3012 to float, !dbg !29 + %3014 = fadd float %3010, %3013, !dbg !26 + %3015 = bitcast float %3014 to i32, !dbg !29 + %3016 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3015, i32 2, i32 31), !dbg !29 + %3017 = bitcast i32 %3016 to float, !dbg !29 + %3018 = fadd float %3014, %3017, !dbg !26 + %3019 = bitcast float %3018 to i32, !dbg !29 + %3020 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3019, i32 1, i32 31), !dbg !29 + %3021 = bitcast i32 %3020 to float, !dbg !29 + %3022 = fadd float %3018, %3021, !dbg !26 + %3023 = bitcast float %2527 to i32, !dbg !29 + %3024 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3023, i32 8, i32 31), !dbg !29 + %3025 = bitcast i32 %3024 to float, !dbg !29 + %3026 = fadd float %2527, %3025, !dbg !26 + %3027 = bitcast float %3026 to i32, !dbg !29 + %3028 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3027, i32 4, i32 31), !dbg !29 + %3029 = bitcast i32 %3028 to float, !dbg !29 + %3030 = fadd float %3026, %3029, !dbg !26 + %3031 = bitcast float %3030 to i32, !dbg !29 + %3032 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3031, i32 2, i32 31), !dbg !29 + %3033 = bitcast i32 %3032 to float, !dbg !29 + %3034 = fadd float %3030, %3033, !dbg !26 + %3035 = bitcast float %3034 to i32, !dbg !29 + %3036 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3035, i32 1, i32 31), !dbg !29 + %3037 = bitcast i32 %3036 to float, !dbg !29 + %3038 = fadd float %3034, %3037, !dbg !26 + %3039 = bitcast float %2534 to i32, !dbg !29 + %3040 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3039, i32 8, i32 31), !dbg !29 + %3041 = bitcast i32 %3040 to float, !dbg !29 + %3042 = fadd float %2534, %3041, !dbg !26 + %3043 = bitcast float %3042 to i32, !dbg !29 + %3044 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3043, i32 4, i32 31), !dbg !29 + %3045 = bitcast i32 %3044 to float, !dbg !29 + %3046 = fadd float %3042, %3045, !dbg !26 + %3047 = bitcast float %3046 to i32, !dbg !29 + %3048 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3047, i32 2, i32 31), !dbg !29 + %3049 = bitcast i32 %3048 to float, !dbg !29 + %3050 = fadd float %3046, %3049, !dbg !26 + %3051 = bitcast float %3050 to i32, !dbg !29 + %3052 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3051, i32 1, i32 31), !dbg !29 + %3053 = bitcast i32 %3052 to float, !dbg !29 + %3054 = fadd float %3050, %3053, !dbg !26 + %3055 = bitcast float %2541 to i32, !dbg !29 + %3056 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3055, i32 8, i32 31), !dbg !29 + %3057 = bitcast i32 %3056 to float, !dbg !29 + %3058 = fadd float %2541, %3057, !dbg !26 + %3059 = bitcast float %3058 to i32, !dbg !29 + %3060 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3059, i32 4, i32 31), !dbg !29 + %3061 = bitcast i32 %3060 to float, !dbg !29 + %3062 = fadd float %3058, %3061, !dbg !26 + %3063 = bitcast float %3062 to i32, !dbg !29 + %3064 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3063, i32 2, i32 31), !dbg !29 + %3065 = bitcast i32 %3064 to float, !dbg !29 + %3066 = fadd float %3062, %3065, !dbg !26 + %3067 = bitcast float %3066 to i32, !dbg !29 + %3068 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3067, i32 1, i32 31), !dbg !29 + %3069 = bitcast i32 %3068 to float, !dbg !29 + %3070 = fadd float %3066, %3069, !dbg !26 + %3071 = bitcast float %2548 to i32, !dbg !29 + %3072 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3071, i32 8, i32 31), !dbg !29 + %3073 = bitcast i32 %3072 to float, !dbg !29 + %3074 = fadd float %2548, %3073, !dbg !26 + %3075 = bitcast float %3074 to i32, !dbg !29 + %3076 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3075, i32 4, i32 31), !dbg !29 + %3077 = bitcast i32 %3076 to float, !dbg !29 + %3078 = fadd float %3074, %3077, !dbg !26 + %3079 = bitcast float %3078 to i32, !dbg !29 + %3080 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3079, i32 2, i32 31), !dbg !29 + %3081 = bitcast i32 %3080 to float, !dbg !29 + %3082 = fadd float %3078, %3081, !dbg !26 + %3083 = bitcast float %3082 to i32, !dbg !29 + %3084 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3083, i32 1, i32 31), !dbg !29 + %3085 = bitcast i32 %3084 to float, !dbg !29 + %3086 = fadd float %3082, %3085, !dbg !26 + %3087 = bitcast float %2555 to i32, !dbg !29 + %3088 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3087, i32 8, i32 31), !dbg !29 + %3089 = bitcast i32 %3088 to float, !dbg !29 + %3090 = fadd float %2555, %3089, !dbg !26 + %3091 = bitcast float %3090 to i32, !dbg !29 + %3092 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3091, i32 4, i32 31), !dbg !29 + %3093 = bitcast i32 %3092 to float, !dbg !29 + %3094 = fadd float %3090, %3093, !dbg !26 + %3095 = bitcast float %3094 to i32, !dbg !29 + %3096 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3095, i32 2, i32 31), !dbg !29 + %3097 = bitcast i32 %3096 to float, !dbg !29 + %3098 = fadd float %3094, %3097, !dbg !26 + %3099 = bitcast float %3098 to i32, !dbg !29 + %3100 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3099, i32 1, i32 31), !dbg !29 + %3101 = bitcast i32 %3100 to float, !dbg !29 + %3102 = fadd float %3098, %3101, !dbg !26 + %3103 = bitcast float %2562 to i32, !dbg !29 + %3104 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3103, i32 8, i32 31), !dbg !29 + %3105 = bitcast i32 %3104 to float, !dbg !29 + %3106 = fadd float %2562, %3105, !dbg !26 + %3107 = bitcast float %3106 to i32, !dbg !29 + %3108 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3107, i32 4, i32 31), !dbg !29 + %3109 = bitcast i32 %3108 to float, !dbg !29 + %3110 = fadd float %3106, %3109, !dbg !26 + %3111 = bitcast float %3110 to i32, !dbg !29 + %3112 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3111, i32 2, i32 31), !dbg !29 + %3113 = bitcast i32 %3112 to float, !dbg !29 + %3114 = fadd float %3110, %3113, !dbg !26 + %3115 = bitcast float %3114 to i32, !dbg !29 + %3116 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3115, i32 1, i32 31), !dbg !29 + %3117 = bitcast i32 %3116 to float, !dbg !29 + %3118 = fadd float %3114, %3117, !dbg !26 + %3119 = bitcast float %2569 to i32, !dbg !29 + %3120 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3119, i32 8, i32 31), !dbg !29 + %3121 = bitcast i32 %3120 to float, !dbg !29 + %3122 = fadd float %2569, %3121, !dbg !26 + %3123 = bitcast float %3122 to i32, !dbg !29 + %3124 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3123, i32 4, i32 31), !dbg !29 + %3125 = bitcast i32 %3124 to float, !dbg !29 + %3126 = fadd float %3122, %3125, !dbg !26 + %3127 = bitcast float %3126 to i32, !dbg !29 + %3128 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3127, i32 2, i32 31), !dbg !29 + %3129 = bitcast i32 %3128 to float, !dbg !29 + %3130 = fadd float %3126, %3129, !dbg !26 + %3131 = bitcast float %3130 to i32, !dbg !29 + %3132 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3131, i32 1, i32 31), !dbg !29 + %3133 = bitcast i32 %3132 to float, !dbg !29 + %3134 = fadd float %3130, %3133, !dbg !26 + %3135 = bitcast float %2576 to i32, !dbg !29 + %3136 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3135, i32 8, i32 31), !dbg !29 + %3137 = bitcast i32 %3136 to float, !dbg !29 + %3138 = fadd float %2576, %3137, !dbg !26 + %3139 = bitcast float %3138 to i32, !dbg !29 + %3140 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3139, i32 4, i32 31), !dbg !29 + %3141 = bitcast i32 %3140 to float, !dbg !29 + %3142 = fadd float %3138, %3141, !dbg !26 + %3143 = bitcast float %3142 to i32, !dbg !29 + %3144 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3143, i32 2, i32 31), !dbg !29 + %3145 = bitcast i32 %3144 to float, !dbg !29 + %3146 = fadd float %3142, %3145, !dbg !26 + %3147 = bitcast float %3146 to i32, !dbg !29 + %3148 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3147, i32 1, i32 31), !dbg !29 + %3149 = bitcast i32 %3148 to float, !dbg !29 + %3150 = fadd float %3146, %3149, !dbg !26 + %3151 = bitcast float %2583 to i32, !dbg !29 + %3152 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3151, i32 8, i32 31), !dbg !29 + %3153 = bitcast i32 %3152 to float, !dbg !29 + %3154 = fadd float %2583, %3153, !dbg !26 + %3155 = bitcast float %3154 to i32, !dbg !29 + %3156 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3155, i32 4, i32 31), !dbg !29 + %3157 = bitcast i32 %3156 to float, !dbg !29 + %3158 = fadd float %3154, %3157, !dbg !26 + %3159 = bitcast float %3158 to i32, !dbg !29 + %3160 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3159, i32 2, i32 31), !dbg !29 + %3161 = bitcast i32 %3160 to float, !dbg !29 + %3162 = fadd float %3158, %3161, !dbg !26 + %3163 = bitcast float %3162 to i32, !dbg !29 + %3164 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3163, i32 1, i32 31), !dbg !29 + %3165 = bitcast i32 %3164 to float, !dbg !29 + %3166 = fadd float %3162, %3165, !dbg !26 + %3167 = bitcast float %2590 to i32, !dbg !29 + %3168 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3167, i32 8, i32 31), !dbg !29 + %3169 = bitcast i32 %3168 to float, !dbg !29 + %3170 = fadd float %2590, %3169, !dbg !26 + %3171 = bitcast float %3170 to i32, !dbg !29 + %3172 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3171, i32 4, i32 31), !dbg !29 + %3173 = bitcast i32 %3172 to float, !dbg !29 + %3174 = fadd float %3170, %3173, !dbg !26 + %3175 = bitcast float %3174 to i32, !dbg !29 + %3176 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3175, i32 2, i32 31), !dbg !29 + %3177 = bitcast i32 %3176 to float, !dbg !29 + %3178 = fadd float %3174, %3177, !dbg !26 + %3179 = bitcast float %3178 to i32, !dbg !29 + %3180 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3179, i32 1, i32 31), !dbg !29 + %3181 = bitcast i32 %3180 to float, !dbg !29 + %3182 = fadd float %3178, %3181, !dbg !26 + %3183 = bitcast float %2597 to i32, !dbg !29 + %3184 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3183, i32 8, i32 31), !dbg !29 + %3185 = bitcast i32 %3184 to float, !dbg !29 + %3186 = fadd float %2597, %3185, !dbg !26 + %3187 = bitcast float %3186 to i32, !dbg !29 + %3188 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3187, i32 4, i32 31), !dbg !29 + %3189 = bitcast i32 %3188 to float, !dbg !29 + %3190 = fadd float %3186, %3189, !dbg !26 + %3191 = bitcast float %3190 to i32, !dbg !29 + %3192 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3191, i32 2, i32 31), !dbg !29 + %3193 = bitcast i32 %3192 to float, !dbg !29 + %3194 = fadd float %3190, %3193, !dbg !26 + %3195 = bitcast float %3194 to i32, !dbg !29 + %3196 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3195, i32 1, i32 31), !dbg !29 + %3197 = bitcast i32 %3196 to float, !dbg !29 + %3198 = fadd float %3194, %3197, !dbg !26 + %3199 = bitcast float %2604 to i32, !dbg !29 + %3200 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3199, i32 8, i32 31), !dbg !29 + %3201 = bitcast i32 %3200 to float, !dbg !29 + %3202 = fadd float %2604, %3201, !dbg !26 + %3203 = bitcast float %3202 to i32, !dbg !29 + %3204 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3203, i32 4, i32 31), !dbg !29 + %3205 = bitcast i32 %3204 to float, !dbg !29 + %3206 = fadd float %3202, %3205, !dbg !26 + %3207 = bitcast float %3206 to i32, !dbg !29 + %3208 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3207, i32 2, i32 31), !dbg !29 + %3209 = bitcast i32 %3208 to float, !dbg !29 + %3210 = fadd float %3206, %3209, !dbg !26 + %3211 = bitcast float %3210 to i32, !dbg !29 + %3212 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3211, i32 1, i32 31), !dbg !29 + %3213 = bitcast i32 %3212 to float, !dbg !29 + %3214 = fadd float %3210, %3213, !dbg !26 + %3215 = bitcast float %2611 to i32, !dbg !29 + %3216 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3215, i32 8, i32 31), !dbg !29 + %3217 = bitcast i32 %3216 to float, !dbg !29 + %3218 = fadd float %2611, %3217, !dbg !26 + %3219 = bitcast float %3218 to i32, !dbg !29 + %3220 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3219, i32 4, i32 31), !dbg !29 + %3221 = bitcast i32 %3220 to float, !dbg !29 + %3222 = fadd float %3218, %3221, !dbg !26 + %3223 = bitcast float %3222 to i32, !dbg !29 + %3224 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3223, i32 2, i32 31), !dbg !29 + %3225 = bitcast i32 %3224 to float, !dbg !29 + %3226 = fadd float %3222, %3225, !dbg !26 + %3227 = bitcast float %3226 to i32, !dbg !29 + %3228 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3227, i32 1, i32 31), !dbg !29 + %3229 = bitcast i32 %3228 to float, !dbg !29 + %3230 = fadd float %3226, %3229, !dbg !26 + %3231 = bitcast float %2618 to i32, !dbg !29 + %3232 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3231, i32 8, i32 31), !dbg !29 + %3233 = bitcast i32 %3232 to float, !dbg !29 + %3234 = fadd float %2618, %3233, !dbg !26 + %3235 = bitcast float %3234 to i32, !dbg !29 + %3236 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3235, i32 4, i32 31), !dbg !29 + %3237 = bitcast i32 %3236 to float, !dbg !29 + %3238 = fadd float %3234, %3237, !dbg !26 + %3239 = bitcast float %3238 to i32, !dbg !29 + %3240 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3239, i32 2, i32 31), !dbg !29 + %3241 = bitcast i32 %3240 to float, !dbg !29 + %3242 = fadd float %3238, %3241, !dbg !26 + %3243 = bitcast float %3242 to i32, !dbg !29 + %3244 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3243, i32 1, i32 31), !dbg !29 + %3245 = bitcast i32 %3244 to float, !dbg !29 + %3246 = fadd float %3242, %3245, !dbg !26 + %3247 = bitcast float %2625 to i32, !dbg !29 + %3248 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3247, i32 8, i32 31), !dbg !29 + %3249 = bitcast i32 %3248 to float, !dbg !29 + %3250 = fadd float %2625, %3249, !dbg !26 + %3251 = bitcast float %3250 to i32, !dbg !29 + %3252 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3251, i32 4, i32 31), !dbg !29 + %3253 = bitcast i32 %3252 to float, !dbg !29 + %3254 = fadd float %3250, %3253, !dbg !26 + %3255 = bitcast float %3254 to i32, !dbg !29 + %3256 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3255, i32 2, i32 31), !dbg !29 + %3257 = bitcast i32 %3256 to float, !dbg !29 + %3258 = fadd float %3254, %3257, !dbg !26 + %3259 = bitcast float %3258 to i32, !dbg !29 + %3260 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3259, i32 1, i32 31), !dbg !29 + %3261 = bitcast i32 %3260 to float, !dbg !29 + %3262 = fadd float %3258, %3261, !dbg !26 + %3263 = bitcast float %2632 to i32, !dbg !29 + %3264 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3263, i32 8, i32 31), !dbg !29 + %3265 = bitcast i32 %3264 to float, !dbg !29 + %3266 = fadd float %2632, %3265, !dbg !26 + %3267 = bitcast float %3266 to i32, !dbg !29 + %3268 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3267, i32 4, i32 31), !dbg !29 + %3269 = bitcast i32 %3268 to float, !dbg !29 + %3270 = fadd float %3266, %3269, !dbg !26 + %3271 = bitcast float %3270 to i32, !dbg !29 + %3272 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3271, i32 2, i32 31), !dbg !29 + %3273 = bitcast i32 %3272 to float, !dbg !29 + %3274 = fadd float %3270, %3273, !dbg !26 + %3275 = bitcast float %3274 to i32, !dbg !29 + %3276 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3275, i32 1, i32 31), !dbg !29 + %3277 = bitcast i32 %3276 to float, !dbg !29 + %3278 = fadd float %3274, %3277, !dbg !26 + %3279 = bitcast float %2639 to i32, !dbg !29 + %3280 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3279, i32 8, i32 31), !dbg !29 + %3281 = bitcast i32 %3280 to float, !dbg !29 + %3282 = fadd float %2639, %3281, !dbg !26 + %3283 = bitcast float %3282 to i32, !dbg !29 + %3284 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3283, i32 4, i32 31), !dbg !29 + %3285 = bitcast i32 %3284 to float, !dbg !29 + %3286 = fadd float %3282, %3285, !dbg !26 + %3287 = bitcast float %3286 to i32, !dbg !29 + %3288 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3287, i32 2, i32 31), !dbg !29 + %3289 = bitcast i32 %3288 to float, !dbg !29 + %3290 = fadd float %3286, %3289, !dbg !26 + %3291 = bitcast float %3290 to i32, !dbg !29 + %3292 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3291, i32 1, i32 31), !dbg !29 + %3293 = bitcast i32 %3292 to float, !dbg !29 + %3294 = fadd float %3290, %3293, !dbg !26 + %3295 = bitcast float %2646 to i32, !dbg !29 + %3296 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3295, i32 8, i32 31), !dbg !29 + %3297 = bitcast i32 %3296 to float, !dbg !29 + %3298 = fadd float %2646, %3297, !dbg !26 + %3299 = bitcast float %3298 to i32, !dbg !29 + %3300 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3299, i32 4, i32 31), !dbg !29 + %3301 = bitcast i32 %3300 to float, !dbg !29 + %3302 = fadd float %3298, %3301, !dbg !26 + %3303 = bitcast float %3302 to i32, !dbg !29 + %3304 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3303, i32 2, i32 31), !dbg !29 + %3305 = bitcast i32 %3304 to float, !dbg !29 + %3306 = fadd float %3302, %3305, !dbg !26 + %3307 = bitcast float %3306 to i32, !dbg !29 + %3308 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3307, i32 1, i32 31), !dbg !29 + %3309 = bitcast i32 %3308 to float, !dbg !29 + %3310 = fadd float %3306, %3309, !dbg !26 + %3311 = bitcast float %2653 to i32, !dbg !29 + %3312 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3311, i32 8, i32 31), !dbg !29 + %3313 = bitcast i32 %3312 to float, !dbg !29 + %3314 = fadd float %2653, %3313, !dbg !26 + %3315 = bitcast float %3314 to i32, !dbg !29 + %3316 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3315, i32 4, i32 31), !dbg !29 + %3317 = bitcast i32 %3316 to float, !dbg !29 + %3318 = fadd float %3314, %3317, !dbg !26 + %3319 = bitcast float %3318 to i32, !dbg !29 + %3320 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3319, i32 2, i32 31), !dbg !29 + %3321 = bitcast i32 %3320 to float, !dbg !29 + %3322 = fadd float %3318, %3321, !dbg !26 + %3323 = bitcast float %3322 to i32, !dbg !29 + %3324 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3323, i32 1, i32 31), !dbg !29 + %3325 = bitcast i32 %3324 to float, !dbg !29 + %3326 = fadd float %3322, %3325, !dbg !26 + %3327 = bitcast float %2660 to i32, !dbg !29 + %3328 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3327, i32 8, i32 31), !dbg !29 + %3329 = bitcast i32 %3328 to float, !dbg !29 + %3330 = fadd float %2660, %3329, !dbg !26 + %3331 = bitcast float %3330 to i32, !dbg !29 + %3332 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3331, i32 4, i32 31), !dbg !29 + %3333 = bitcast i32 %3332 to float, !dbg !29 + %3334 = fadd float %3330, %3333, !dbg !26 + %3335 = bitcast float %3334 to i32, !dbg !29 + %3336 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3335, i32 2, i32 31), !dbg !29 + %3337 = bitcast i32 %3336 to float, !dbg !29 + %3338 = fadd float %3334, %3337, !dbg !26 + %3339 = bitcast float %3338 to i32, !dbg !29 + %3340 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3339, i32 1, i32 31), !dbg !29 + %3341 = bitcast i32 %3340 to float, !dbg !29 + %3342 = fadd float %3338, %3341, !dbg !26 + %3343 = bitcast float %2667 to i32, !dbg !29 + %3344 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3343, i32 8, i32 31), !dbg !29 + %3345 = bitcast i32 %3344 to float, !dbg !29 + %3346 = fadd float %2667, %3345, !dbg !26 + %3347 = bitcast float %3346 to i32, !dbg !29 + %3348 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3347, i32 4, i32 31), !dbg !29 + %3349 = bitcast i32 %3348 to float, !dbg !29 + %3350 = fadd float %3346, %3349, !dbg !26 + %3351 = bitcast float %3350 to i32, !dbg !29 + %3352 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3351, i32 2, i32 31), !dbg !29 + %3353 = bitcast i32 %3352 to float, !dbg !29 + %3354 = fadd float %3350, %3353, !dbg !26 + %3355 = bitcast float %3354 to i32, !dbg !29 + %3356 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3355, i32 1, i32 31), !dbg !29 + %3357 = bitcast i32 %3356 to float, !dbg !29 + %3358 = fadd float %3354, %3357, !dbg !26 + %3359 = bitcast float %2674 to i32, !dbg !29 + %3360 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3359, i32 8, i32 31), !dbg !29 + %3361 = bitcast i32 %3360 to float, !dbg !29 + %3362 = fadd float %2674, %3361, !dbg !26 + %3363 = bitcast float %3362 to i32, !dbg !29 + %3364 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3363, i32 4, i32 31), !dbg !29 + %3365 = bitcast i32 %3364 to float, !dbg !29 + %3366 = fadd float %3362, %3365, !dbg !26 + %3367 = bitcast float %3366 to i32, !dbg !29 + %3368 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3367, i32 2, i32 31), !dbg !29 + %3369 = bitcast i32 %3368 to float, !dbg !29 + %3370 = fadd float %3366, %3369, !dbg !26 + %3371 = bitcast float %3370 to i32, !dbg !29 + %3372 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3371, i32 1, i32 31), !dbg !29 + %3373 = bitcast i32 %3372 to float, !dbg !29 + %3374 = fadd float %3370, %3373, !dbg !26 + %3375 = bitcast float %2681 to i32, !dbg !29 + %3376 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3375, i32 8, i32 31), !dbg !29 + %3377 = bitcast i32 %3376 to float, !dbg !29 + %3378 = fadd float %2681, %3377, !dbg !26 + %3379 = bitcast float %3378 to i32, !dbg !29 + %3380 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3379, i32 4, i32 31), !dbg !29 + %3381 = bitcast i32 %3380 to float, !dbg !29 + %3382 = fadd float %3378, %3381, !dbg !26 + %3383 = bitcast float %3382 to i32, !dbg !29 + %3384 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3383, i32 2, i32 31), !dbg !29 + %3385 = bitcast i32 %3384 to float, !dbg !29 + %3386 = fadd float %3382, %3385, !dbg !26 + %3387 = bitcast float %3386 to i32, !dbg !29 + %3388 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3387, i32 1, i32 31), !dbg !29 + %3389 = bitcast i32 %3388 to float, !dbg !29 + %3390 = fadd float %3386, %3389, !dbg !26 + %3391 = bitcast float %2688 to i32, !dbg !29 + %3392 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3391, i32 8, i32 31), !dbg !29 + %3393 = bitcast i32 %3392 to float, !dbg !29 + %3394 = fadd float %2688, %3393, !dbg !26 + %3395 = bitcast float %3394 to i32, !dbg !29 + %3396 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3395, i32 4, i32 31), !dbg !29 + %3397 = bitcast i32 %3396 to float, !dbg !29 + %3398 = fadd float %3394, %3397, !dbg !26 + %3399 = bitcast float %3398 to i32, !dbg !29 + %3400 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3399, i32 2, i32 31), !dbg !29 + %3401 = bitcast i32 %3400 to float, !dbg !29 + %3402 = fadd float %3398, %3401, !dbg !26 + %3403 = bitcast float %3402 to i32, !dbg !29 + %3404 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3403, i32 1, i32 31), !dbg !29 + %3405 = bitcast i32 %3404 to float, !dbg !29 + %3406 = fadd float %3402, %3405, !dbg !26 + %3407 = bitcast float %2695 to i32, !dbg !29 + %3408 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3407, i32 8, i32 31), !dbg !29 + %3409 = bitcast i32 %3408 to float, !dbg !29 + %3410 = fadd float %2695, %3409, !dbg !26 + %3411 = bitcast float %3410 to i32, !dbg !29 + %3412 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3411, i32 4, i32 31), !dbg !29 + %3413 = bitcast i32 %3412 to float, !dbg !29 + %3414 = fadd float %3410, %3413, !dbg !26 + %3415 = bitcast float %3414 to i32, !dbg !29 + %3416 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3415, i32 2, i32 31), !dbg !29 + %3417 = bitcast i32 %3416 to float, !dbg !29 + %3418 = fadd float %3414, %3417, !dbg !26 + %3419 = bitcast float %3418 to i32, !dbg !29 + %3420 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3419, i32 1, i32 31), !dbg !29 + %3421 = bitcast i32 %3420 to float, !dbg !29 + %3422 = fadd float %3418, %3421, !dbg !26 + %3423 = bitcast float %2702 to i32, !dbg !29 + %3424 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3423, i32 8, i32 31), !dbg !29 + %3425 = bitcast i32 %3424 to float, !dbg !29 + %3426 = fadd float %2702, %3425, !dbg !26 + %3427 = bitcast float %3426 to i32, !dbg !29 + %3428 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3427, i32 4, i32 31), !dbg !29 + %3429 = bitcast i32 %3428 to float, !dbg !29 + %3430 = fadd float %3426, %3429, !dbg !26 + %3431 = bitcast float %3430 to i32, !dbg !29 + %3432 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3431, i32 2, i32 31), !dbg !29 + %3433 = bitcast i32 %3432 to float, !dbg !29 + %3434 = fadd float %3430, %3433, !dbg !26 + %3435 = bitcast float %3434 to i32, !dbg !29 + %3436 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3435, i32 1, i32 31), !dbg !29 + %3437 = bitcast i32 %3436 to float, !dbg !29 + %3438 = fadd float %3434, %3437, !dbg !26 + %3439 = bitcast float %2709 to i32, !dbg !29 + %3440 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3439, i32 8, i32 31), !dbg !29 + %3441 = bitcast i32 %3440 to float, !dbg !29 + %3442 = fadd float %2709, %3441, !dbg !26 + %3443 = bitcast float %3442 to i32, !dbg !29 + %3444 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3443, i32 4, i32 31), !dbg !29 + %3445 = bitcast i32 %3444 to float, !dbg !29 + %3446 = fadd float %3442, %3445, !dbg !26 + %3447 = bitcast float %3446 to i32, !dbg !29 + %3448 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3447, i32 2, i32 31), !dbg !29 + %3449 = bitcast i32 %3448 to float, !dbg !29 + %3450 = fadd float %3446, %3449, !dbg !26 + %3451 = bitcast float %3450 to i32, !dbg !29 + %3452 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3451, i32 1, i32 31), !dbg !29 + %3453 = bitcast i32 %3452 to float, !dbg !29 + %3454 = fadd float %3450, %3453, !dbg !26 + %3455 = bitcast float %2716 to i32, !dbg !29 + %3456 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3455, i32 8, i32 31), !dbg !29 + %3457 = bitcast i32 %3456 to float, !dbg !29 + %3458 = fadd float %2716, %3457, !dbg !26 + %3459 = bitcast float %3458 to i32, !dbg !29 + %3460 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3459, i32 4, i32 31), !dbg !29 + %3461 = bitcast i32 %3460 to float, !dbg !29 + %3462 = fadd float %3458, %3461, !dbg !26 + %3463 = bitcast float %3462 to i32, !dbg !29 + %3464 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3463, i32 2, i32 31), !dbg !29 + %3465 = bitcast i32 %3464 to float, !dbg !29 + %3466 = fadd float %3462, %3465, !dbg !26 + %3467 = bitcast float %3466 to i32, !dbg !29 + %3468 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3467, i32 1, i32 31), !dbg !29 + %3469 = bitcast i32 %3468 to float, !dbg !29 + %3470 = fadd float %3466, %3469, !dbg !26 + %3471 = bitcast float %2723 to i32, !dbg !29 + %3472 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3471, i32 8, i32 31), !dbg !29 + %3473 = bitcast i32 %3472 to float, !dbg !29 + %3474 = fadd float %2723, %3473, !dbg !26 + %3475 = bitcast float %3474 to i32, !dbg !29 + %3476 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3475, i32 4, i32 31), !dbg !29 + %3477 = bitcast i32 %3476 to float, !dbg !29 + %3478 = fadd float %3474, %3477, !dbg !26 + %3479 = bitcast float %3478 to i32, !dbg !29 + %3480 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3479, i32 2, i32 31), !dbg !29 + %3481 = bitcast i32 %3480 to float, !dbg !29 + %3482 = fadd float %3478, %3481, !dbg !26 + %3483 = bitcast float %3482 to i32, !dbg !29 + %3484 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3483, i32 1, i32 31), !dbg !29 + %3485 = bitcast i32 %3484 to float, !dbg !29 + %3486 = fadd float %3482, %3485, !dbg !26 + %3487 = bitcast float %2730 to i32, !dbg !29 + %3488 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3487, i32 8, i32 31), !dbg !29 + %3489 = bitcast i32 %3488 to float, !dbg !29 + %3490 = fadd float %2730, %3489, !dbg !26 + %3491 = bitcast float %3490 to i32, !dbg !29 + %3492 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3491, i32 4, i32 31), !dbg !29 + %3493 = bitcast i32 %3492 to float, !dbg !29 + %3494 = fadd float %3490, %3493, !dbg !26 + %3495 = bitcast float %3494 to i32, !dbg !29 + %3496 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3495, i32 2, i32 31), !dbg !29 + %3497 = bitcast i32 %3496 to float, !dbg !29 + %3498 = fadd float %3494, %3497, !dbg !26 + %3499 = bitcast float %3498 to i32, !dbg !29 + %3500 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3499, i32 1, i32 31), !dbg !29 + %3501 = bitcast i32 %3500 to float, !dbg !29 + %3502 = fadd float %3498, %3501, !dbg !26 + %3503 = bitcast float %2737 to i32, !dbg !29 + %3504 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3503, i32 8, i32 31), !dbg !29 + %3505 = bitcast i32 %3504 to float, !dbg !29 + %3506 = fadd float %2737, %3505, !dbg !26 + %3507 = bitcast float %3506 to i32, !dbg !29 + %3508 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3507, i32 4, i32 31), !dbg !29 + %3509 = bitcast i32 %3508 to float, !dbg !29 + %3510 = fadd float %3506, %3509, !dbg !26 + %3511 = bitcast float %3510 to i32, !dbg !29 + %3512 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3511, i32 2, i32 31), !dbg !29 + %3513 = bitcast i32 %3512 to float, !dbg !29 + %3514 = fadd float %3510, %3513, !dbg !26 + %3515 = bitcast float %3514 to i32, !dbg !29 + %3516 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3515, i32 1, i32 31), !dbg !29 + %3517 = bitcast i32 %3516 to float, !dbg !29 + %3518 = fadd float %3514, %3517, !dbg !26 + %3519 = bitcast float %2744 to i32, !dbg !29 + %3520 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3519, i32 8, i32 31), !dbg !29 + %3521 = bitcast i32 %3520 to float, !dbg !29 + %3522 = fadd float %2744, %3521, !dbg !26 + %3523 = bitcast float %3522 to i32, !dbg !29 + %3524 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3523, i32 4, i32 31), !dbg !29 + %3525 = bitcast i32 %3524 to float, !dbg !29 + %3526 = fadd float %3522, %3525, !dbg !26 + %3527 = bitcast float %3526 to i32, !dbg !29 + %3528 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3527, i32 2, i32 31), !dbg !29 + %3529 = bitcast i32 %3528 to float, !dbg !29 + %3530 = fadd float %3526, %3529, !dbg !26 + %3531 = bitcast float %3530 to i32, !dbg !29 + %3532 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3531, i32 1, i32 31), !dbg !29 + %3533 = bitcast i32 %3532 to float, !dbg !29 + %3534 = fadd float %3530, %3533, !dbg !26 + %3535 = bitcast float %2751 to i32, !dbg !29 + %3536 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3535, i32 8, i32 31), !dbg !29 + %3537 = bitcast i32 %3536 to float, !dbg !29 + %3538 = fadd float %2751, %3537, !dbg !26 + %3539 = bitcast float %3538 to i32, !dbg !29 + %3540 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3539, i32 4, i32 31), !dbg !29 + %3541 = bitcast i32 %3540 to float, !dbg !29 + %3542 = fadd float %3538, %3541, !dbg !26 + %3543 = bitcast float %3542 to i32, !dbg !29 + %3544 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3543, i32 2, i32 31), !dbg !29 + %3545 = bitcast i32 %3544 to float, !dbg !29 + %3546 = fadd float %3542, %3545, !dbg !26 + %3547 = bitcast float %3546 to i32, !dbg !29 + %3548 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3547, i32 1, i32 31), !dbg !29 + %3549 = bitcast i32 %3548 to float, !dbg !29 + %3550 = fadd float %3546, %3549, !dbg !26 + %3551 = bitcast float %2758 to i32, !dbg !29 + %3552 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3551, i32 8, i32 31), !dbg !29 + %3553 = bitcast i32 %3552 to float, !dbg !29 + %3554 = fadd float %2758, %3553, !dbg !26 + %3555 = bitcast float %3554 to i32, !dbg !29 + %3556 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3555, i32 4, i32 31), !dbg !29 + %3557 = bitcast i32 %3556 to float, !dbg !29 + %3558 = fadd float %3554, %3557, !dbg !26 + %3559 = bitcast float %3558 to i32, !dbg !29 + %3560 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3559, i32 2, i32 31), !dbg !29 + %3561 = bitcast i32 %3560 to float, !dbg !29 + %3562 = fadd float %3558, %3561, !dbg !26 + %3563 = bitcast float %3562 to i32, !dbg !29 + %3564 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3563, i32 1, i32 31), !dbg !29 + %3565 = bitcast i32 %3564 to float, !dbg !29 + %3566 = fadd float %3562, %3565, !dbg !26 + %3567 = bitcast float %2765 to i32, !dbg !29 + %3568 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3567, i32 8, i32 31), !dbg !29 + %3569 = bitcast i32 %3568 to float, !dbg !29 + %3570 = fadd float %2765, %3569, !dbg !26 + %3571 = bitcast float %3570 to i32, !dbg !29 + %3572 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3571, i32 4, i32 31), !dbg !29 + %3573 = bitcast i32 %3572 to float, !dbg !29 + %3574 = fadd float %3570, %3573, !dbg !26 + %3575 = bitcast float %3574 to i32, !dbg !29 + %3576 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3575, i32 2, i32 31), !dbg !29 + %3577 = bitcast i32 %3576 to float, !dbg !29 + %3578 = fadd float %3574, %3577, !dbg !26 + %3579 = bitcast float %3578 to i32, !dbg !29 + %3580 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3579, i32 1, i32 31), !dbg !29 + %3581 = bitcast i32 %3580 to float, !dbg !29 + %3582 = fadd float %3578, %3581, !dbg !26 + %3583 = bitcast float %2772 to i32, !dbg !29 + %3584 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3583, i32 8, i32 31), !dbg !29 + %3585 = bitcast i32 %3584 to float, !dbg !29 + %3586 = fadd float %2772, %3585, !dbg !26 + %3587 = bitcast float %3586 to i32, !dbg !29 + %3588 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3587, i32 4, i32 31), !dbg !29 + %3589 = bitcast i32 %3588 to float, !dbg !29 + %3590 = fadd float %3586, %3589, !dbg !26 + %3591 = bitcast float %3590 to i32, !dbg !29 + %3592 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3591, i32 2, i32 31), !dbg !29 + %3593 = bitcast i32 %3592 to float, !dbg !29 + %3594 = fadd float %3590, %3593, !dbg !26 + %3595 = bitcast float %3594 to i32, !dbg !29 + %3596 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3595, i32 1, i32 31), !dbg !29 + %3597 = bitcast i32 %3596 to float, !dbg !29 + %3598 = fadd float %3594, %3597, !dbg !26 + %3599 = bitcast float %2779 to i32, !dbg !29 + %3600 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3599, i32 8, i32 31), !dbg !29 + %3601 = bitcast i32 %3600 to float, !dbg !29 + %3602 = fadd float %2779, %3601, !dbg !26 + %3603 = bitcast float %3602 to i32, !dbg !29 + %3604 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3603, i32 4, i32 31), !dbg !29 + %3605 = bitcast i32 %3604 to float, !dbg !29 + %3606 = fadd float %3602, %3605, !dbg !26 + %3607 = bitcast float %3606 to i32, !dbg !29 + %3608 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3607, i32 2, i32 31), !dbg !29 + %3609 = bitcast i32 %3608 to float, !dbg !29 + %3610 = fadd float %3606, %3609, !dbg !26 + %3611 = bitcast float %3610 to i32, !dbg !29 + %3612 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3611, i32 1, i32 31), !dbg !29 + %3613 = bitcast i32 %3612 to float, !dbg !29 + %3614 = fadd float %3610, %3613, !dbg !26 + %3615 = bitcast float %2786 to i32, !dbg !29 + %3616 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3615, i32 8, i32 31), !dbg !29 + %3617 = bitcast i32 %3616 to float, !dbg !29 + %3618 = fadd float %2786, %3617, !dbg !26 + %3619 = bitcast float %3618 to i32, !dbg !29 + %3620 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3619, i32 4, i32 31), !dbg !29 + %3621 = bitcast i32 %3620 to float, !dbg !29 + %3622 = fadd float %3618, %3621, !dbg !26 + %3623 = bitcast float %3622 to i32, !dbg !29 + %3624 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3623, i32 2, i32 31), !dbg !29 + %3625 = bitcast i32 %3624 to float, !dbg !29 + %3626 = fadd float %3622, %3625, !dbg !26 + %3627 = bitcast float %3626 to i32, !dbg !29 + %3628 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3627, i32 1, i32 31), !dbg !29 + %3629 = bitcast i32 %3628 to float, !dbg !29 + %3630 = fadd float %3626, %3629, !dbg !26 + %3631 = bitcast float %2793 to i32, !dbg !29 + %3632 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3631, i32 8, i32 31), !dbg !29 + %3633 = bitcast i32 %3632 to float, !dbg !29 + %3634 = fadd float %2793, %3633, !dbg !26 + %3635 = bitcast float %3634 to i32, !dbg !29 + %3636 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3635, i32 4, i32 31), !dbg !29 + %3637 = bitcast i32 %3636 to float, !dbg !29 + %3638 = fadd float %3634, %3637, !dbg !26 + %3639 = bitcast float %3638 to i32, !dbg !29 + %3640 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3639, i32 2, i32 31), !dbg !29 + %3641 = bitcast i32 %3640 to float, !dbg !29 + %3642 = fadd float %3638, %3641, !dbg !26 + %3643 = bitcast float %3642 to i32, !dbg !29 + %3644 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3643, i32 1, i32 31), !dbg !29 + %3645 = bitcast i32 %3644 to float, !dbg !29 + %3646 = fadd float %3642, %3645, !dbg !26 + %3647 = bitcast float %2800 to i32, !dbg !29 + %3648 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3647, i32 8, i32 31), !dbg !29 + %3649 = bitcast i32 %3648 to float, !dbg !29 + %3650 = fadd float %2800, %3649, !dbg !26 + %3651 = bitcast float %3650 to i32, !dbg !29 + %3652 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3651, i32 4, i32 31), !dbg !29 + %3653 = bitcast i32 %3652 to float, !dbg !29 + %3654 = fadd float %3650, %3653, !dbg !26 + %3655 = bitcast float %3654 to i32, !dbg !29 + %3656 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3655, i32 2, i32 31), !dbg !29 + %3657 = bitcast i32 %3656 to float, !dbg !29 + %3658 = fadd float %3654, %3657, !dbg !26 + %3659 = bitcast float %3658 to i32, !dbg !29 + %3660 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3659, i32 1, i32 31), !dbg !29 + %3661 = bitcast i32 %3660 to float, !dbg !29 + %3662 = fadd float %3658, %3661, !dbg !26 + %3663 = bitcast float %2807 to i32, !dbg !29 + %3664 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3663, i32 8, i32 31), !dbg !29 + %3665 = bitcast i32 %3664 to float, !dbg !29 + %3666 = fadd float %2807, %3665, !dbg !26 + %3667 = bitcast float %3666 to i32, !dbg !29 + %3668 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3667, i32 4, i32 31), !dbg !29 + %3669 = bitcast i32 %3668 to float, !dbg !29 + %3670 = fadd float %3666, %3669, !dbg !26 + %3671 = bitcast float %3670 to i32, !dbg !29 + %3672 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3671, i32 2, i32 31), !dbg !29 + %3673 = bitcast i32 %3672 to float, !dbg !29 + %3674 = fadd float %3670, %3673, !dbg !26 + %3675 = bitcast float %3674 to i32, !dbg !29 + %3676 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3675, i32 1, i32 31), !dbg !29 + %3677 = bitcast i32 %3676 to float, !dbg !29 + %3678 = fadd float %3674, %3677, !dbg !26 + %3679 = bitcast float %2814 to i32, !dbg !29 + %3680 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3679, i32 8, i32 31), !dbg !29 + %3681 = bitcast i32 %3680 to float, !dbg !29 + %3682 = fadd float %2814, %3681, !dbg !26 + %3683 = bitcast float %3682 to i32, !dbg !29 + %3684 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3683, i32 4, i32 31), !dbg !29 + %3685 = bitcast i32 %3684 to float, !dbg !29 + %3686 = fadd float %3682, %3685, !dbg !26 + %3687 = bitcast float %3686 to i32, !dbg !29 + %3688 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3687, i32 2, i32 31), !dbg !29 + %3689 = bitcast i32 %3688 to float, !dbg !29 + %3690 = fadd float %3686, %3689, !dbg !26 + %3691 = bitcast float %3690 to i32, !dbg !29 + %3692 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3691, i32 1, i32 31), !dbg !29 + %3693 = bitcast i32 %3692 to float, !dbg !29 + %3694 = fadd float %3690, %3693, !dbg !26 + %3695 = bitcast float %2821 to i32, !dbg !29 + %3696 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3695, i32 8, i32 31), !dbg !29 + %3697 = bitcast i32 %3696 to float, !dbg !29 + %3698 = fadd float %2821, %3697, !dbg !26 + %3699 = bitcast float %3698 to i32, !dbg !29 + %3700 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3699, i32 4, i32 31), !dbg !29 + %3701 = bitcast i32 %3700 to float, !dbg !29 + %3702 = fadd float %3698, %3701, !dbg !26 + %3703 = bitcast float %3702 to i32, !dbg !29 + %3704 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3703, i32 2, i32 31), !dbg !29 + %3705 = bitcast i32 %3704 to float, !dbg !29 + %3706 = fadd float %3702, %3705, !dbg !26 + %3707 = bitcast float %3706 to i32, !dbg !29 + %3708 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3707, i32 1, i32 31), !dbg !29 + %3709 = bitcast i32 %3708 to float, !dbg !29 + %3710 = fadd float %3706, %3709, !dbg !26 + %3711 = bitcast float %2828 to i32, !dbg !29 + %3712 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3711, i32 8, i32 31), !dbg !29 + %3713 = bitcast i32 %3712 to float, !dbg !29 + %3714 = fadd float %2828, %3713, !dbg !26 + %3715 = bitcast float %3714 to i32, !dbg !29 + %3716 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3715, i32 4, i32 31), !dbg !29 + %3717 = bitcast i32 %3716 to float, !dbg !29 + %3718 = fadd float %3714, %3717, !dbg !26 + %3719 = bitcast float %3718 to i32, !dbg !29 + %3720 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3719, i32 2, i32 31), !dbg !29 + %3721 = bitcast i32 %3720 to float, !dbg !29 + %3722 = fadd float %3718, %3721, !dbg !26 + %3723 = bitcast float %3722 to i32, !dbg !29 + %3724 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3723, i32 1, i32 31), !dbg !29 + %3725 = bitcast i32 %3724 to float, !dbg !29 + %3726 = fadd float %3722, %3725, !dbg !26 + %3727 = bitcast float %2835 to i32, !dbg !29 + %3728 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3727, i32 8, i32 31), !dbg !29 + %3729 = bitcast i32 %3728 to float, !dbg !29 + %3730 = fadd float %2835, %3729, !dbg !26 + %3731 = bitcast float %3730 to i32, !dbg !29 + %3732 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3731, i32 4, i32 31), !dbg !29 + %3733 = bitcast i32 %3732 to float, !dbg !29 + %3734 = fadd float %3730, %3733, !dbg !26 + %3735 = bitcast float %3734 to i32, !dbg !29 + %3736 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3735, i32 2, i32 31), !dbg !29 + %3737 = bitcast i32 %3736 to float, !dbg !29 + %3738 = fadd float %3734, %3737, !dbg !26 + %3739 = bitcast float %3738 to i32, !dbg !29 + %3740 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3739, i32 1, i32 31), !dbg !29 + %3741 = bitcast i32 %3740 to float, !dbg !29 + %3742 = fadd float %3738, %3741, !dbg !26 + %3743 = bitcast float %2842 to i32, !dbg !29 + %3744 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3743, i32 8, i32 31), !dbg !29 + %3745 = bitcast i32 %3744 to float, !dbg !29 + %3746 = fadd float %2842, %3745, !dbg !26 + %3747 = bitcast float %3746 to i32, !dbg !29 + %3748 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3747, i32 4, i32 31), !dbg !29 + %3749 = bitcast i32 %3748 to float, !dbg !29 + %3750 = fadd float %3746, %3749, !dbg !26 + %3751 = bitcast float %3750 to i32, !dbg !29 + %3752 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3751, i32 2, i32 31), !dbg !29 + %3753 = bitcast i32 %3752 to float, !dbg !29 + %3754 = fadd float %3750, %3753, !dbg !26 + %3755 = bitcast float %3754 to i32, !dbg !29 + %3756 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3755, i32 1, i32 31), !dbg !29 + %3757 = bitcast i32 %3756 to float, !dbg !29 + %3758 = fadd float %3754, %3757, !dbg !26 + %3759 = bitcast float %2849 to i32, !dbg !29 + %3760 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3759, i32 8, i32 31), !dbg !29 + %3761 = bitcast i32 %3760 to float, !dbg !29 + %3762 = fadd float %2849, %3761, !dbg !26 + %3763 = bitcast float %3762 to i32, !dbg !29 + %3764 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3763, i32 4, i32 31), !dbg !29 + %3765 = bitcast i32 %3764 to float, !dbg !29 + %3766 = fadd float %3762, %3765, !dbg !26 + %3767 = bitcast float %3766 to i32, !dbg !29 + %3768 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3767, i32 2, i32 31), !dbg !29 + %3769 = bitcast i32 %3768 to float, !dbg !29 + %3770 = fadd float %3766, %3769, !dbg !26 + %3771 = bitcast float %3770 to i32, !dbg !29 + %3772 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3771, i32 1, i32 31), !dbg !29 + %3773 = bitcast i32 %3772 to float, !dbg !29 + %3774 = fadd float %3770, %3773, !dbg !26 + %3775 = bitcast float %2856 to i32, !dbg !29 + %3776 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3775, i32 8, i32 31), !dbg !29 + %3777 = bitcast i32 %3776 to float, !dbg !29 + %3778 = fadd float %2856, %3777, !dbg !26 + %3779 = bitcast float %3778 to i32, !dbg !29 + %3780 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3779, i32 4, i32 31), !dbg !29 + %3781 = bitcast i32 %3780 to float, !dbg !29 + %3782 = fadd float %3778, %3781, !dbg !26 + %3783 = bitcast float %3782 to i32, !dbg !29 + %3784 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3783, i32 2, i32 31), !dbg !29 + %3785 = bitcast i32 %3784 to float, !dbg !29 + %3786 = fadd float %3782, %3785, !dbg !26 + %3787 = bitcast float %3786 to i32, !dbg !29 + %3788 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3787, i32 1, i32 31), !dbg !29 + %3789 = bitcast i32 %3788 to float, !dbg !29 + %3790 = fadd float %3786, %3789, !dbg !26 + %3791 = bitcast float %2863 to i32, !dbg !29 + %3792 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3791, i32 8, i32 31), !dbg !29 + %3793 = bitcast i32 %3792 to float, !dbg !29 + %3794 = fadd float %2863, %3793, !dbg !26 + %3795 = bitcast float %3794 to i32, !dbg !29 + %3796 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3795, i32 4, i32 31), !dbg !29 + %3797 = bitcast i32 %3796 to float, !dbg !29 + %3798 = fadd float %3794, %3797, !dbg !26 + %3799 = bitcast float %3798 to i32, !dbg !29 + %3800 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3799, i32 2, i32 31), !dbg !29 + %3801 = bitcast i32 %3800 to float, !dbg !29 + %3802 = fadd float %3798, %3801, !dbg !26 + %3803 = bitcast float %3802 to i32, !dbg !29 + %3804 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3803, i32 1, i32 31), !dbg !29 + %3805 = bitcast i32 %3804 to float, !dbg !29 + %3806 = fadd float %3802, %3805, !dbg !26 + %3807 = bitcast float %2870 to i32, !dbg !29 + %3808 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3807, i32 8, i32 31), !dbg !29 + %3809 = bitcast i32 %3808 to float, !dbg !29 + %3810 = fadd float %2870, %3809, !dbg !26 + %3811 = bitcast float %3810 to i32, !dbg !29 + %3812 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3811, i32 4, i32 31), !dbg !29 + %3813 = bitcast i32 %3812 to float, !dbg !29 + %3814 = fadd float %3810, %3813, !dbg !26 + %3815 = bitcast float %3814 to i32, !dbg !29 + %3816 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3815, i32 2, i32 31), !dbg !29 + %3817 = bitcast i32 %3816 to float, !dbg !29 + %3818 = fadd float %3814, %3817, !dbg !26 + %3819 = bitcast float %3818 to i32, !dbg !29 + %3820 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3819, i32 1, i32 31), !dbg !29 + %3821 = bitcast i32 %3820 to float, !dbg !29 + %3822 = fadd float %3818, %3821, !dbg !26 + %3823 = bitcast float %2877 to i32, !dbg !29 + %3824 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3823, i32 8, i32 31), !dbg !29 + %3825 = bitcast i32 %3824 to float, !dbg !29 + %3826 = fadd float %2877, %3825, !dbg !26 + %3827 = bitcast float %3826 to i32, !dbg !29 + %3828 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3827, i32 4, i32 31), !dbg !29 + %3829 = bitcast i32 %3828 to float, !dbg !29 + %3830 = fadd float %3826, %3829, !dbg !26 + %3831 = bitcast float %3830 to i32, !dbg !29 + %3832 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3831, i32 2, i32 31), !dbg !29 + %3833 = bitcast i32 %3832 to float, !dbg !29 + %3834 = fadd float %3830, %3833, !dbg !26 + %3835 = bitcast float %3834 to i32, !dbg !29 + %3836 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3835, i32 1, i32 31), !dbg !29 + %3837 = bitcast i32 %3836 to float, !dbg !29 + %3838 = fadd float %3834, %3837, !dbg !26 + %3839 = bitcast float %2884 to i32, !dbg !29 + %3840 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3839, i32 8, i32 31), !dbg !29 + %3841 = bitcast i32 %3840 to float, !dbg !29 + %3842 = fadd float %2884, %3841, !dbg !26 + %3843 = bitcast float %3842 to i32, !dbg !29 + %3844 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3843, i32 4, i32 31), !dbg !29 + %3845 = bitcast i32 %3844 to float, !dbg !29 + %3846 = fadd float %3842, %3845, !dbg !26 + %3847 = bitcast float %3846 to i32, !dbg !29 + %3848 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3847, i32 2, i32 31), !dbg !29 + %3849 = bitcast i32 %3848 to float, !dbg !29 + %3850 = fadd float %3846, %3849, !dbg !26 + %3851 = bitcast float %3850 to i32, !dbg !29 + %3852 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3851, i32 1, i32 31), !dbg !29 + %3853 = bitcast i32 %3852 to float, !dbg !29 + %3854 = fadd float %3850, %3853, !dbg !26 + %3855 = bitcast float %2891 to i32, !dbg !29 + %3856 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3855, i32 8, i32 31), !dbg !29 + %3857 = bitcast i32 %3856 to float, !dbg !29 + %3858 = fadd float %2891, %3857, !dbg !26 + %3859 = bitcast float %3858 to i32, !dbg !29 + %3860 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3859, i32 4, i32 31), !dbg !29 + %3861 = bitcast i32 %3860 to float, !dbg !29 + %3862 = fadd float %3858, %3861, !dbg !26 + %3863 = bitcast float %3862 to i32, !dbg !29 + %3864 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3863, i32 2, i32 31), !dbg !29 + %3865 = bitcast i32 %3864 to float, !dbg !29 + %3866 = fadd float %3862, %3865, !dbg !26 + %3867 = bitcast float %3866 to i32, !dbg !29 + %3868 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3867, i32 1, i32 31), !dbg !29 + %3869 = bitcast i32 %3868 to float, !dbg !29 + %3870 = fadd float %3866, %3869, !dbg !26 + %3871 = bitcast float %2898 to i32, !dbg !29 + %3872 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3871, i32 8, i32 31), !dbg !29 + %3873 = bitcast i32 %3872 to float, !dbg !29 + %3874 = fadd float %2898, %3873, !dbg !26 + %3875 = bitcast float %3874 to i32, !dbg !29 + %3876 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3875, i32 4, i32 31), !dbg !29 + %3877 = bitcast i32 %3876 to float, !dbg !29 + %3878 = fadd float %3874, %3877, !dbg !26 + %3879 = bitcast float %3878 to i32, !dbg !29 + %3880 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3879, i32 2, i32 31), !dbg !29 + %3881 = bitcast i32 %3880 to float, !dbg !29 + %3882 = fadd float %3878, %3881, !dbg !26 + %3883 = bitcast float %3882 to i32, !dbg !29 + %3884 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3883, i32 1, i32 31), !dbg !29 + %3885 = bitcast i32 %3884 to float, !dbg !29 + %3886 = fadd float %3882, %3885, !dbg !26 + %3887 = bitcast float %2905 to i32, !dbg !29 + %3888 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3887, i32 8, i32 31), !dbg !29 + %3889 = bitcast i32 %3888 to float, !dbg !29 + %3890 = fadd float %2905, %3889, !dbg !26 + %3891 = bitcast float %3890 to i32, !dbg !29 + %3892 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3891, i32 4, i32 31), !dbg !29 + %3893 = bitcast i32 %3892 to float, !dbg !29 + %3894 = fadd float %3890, %3893, !dbg !26 + %3895 = bitcast float %3894 to i32, !dbg !29 + %3896 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3895, i32 2, i32 31), !dbg !29 + %3897 = bitcast i32 %3896 to float, !dbg !29 + %3898 = fadd float %3894, %3897, !dbg !26 + %3899 = bitcast float %3898 to i32, !dbg !29 + %3900 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3899, i32 1, i32 31), !dbg !29 + %3901 = bitcast i32 %3900 to float, !dbg !29 + %3902 = fadd float %3898, %3901, !dbg !26 + %3903 = bitcast float %2912 to i32, !dbg !29 + %3904 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3903, i32 8, i32 31), !dbg !29 + %3905 = bitcast i32 %3904 to float, !dbg !29 + %3906 = fadd float %2912, %3905, !dbg !26 + %3907 = bitcast float %3906 to i32, !dbg !29 + %3908 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3907, i32 4, i32 31), !dbg !29 + %3909 = bitcast i32 %3908 to float, !dbg !29 + %3910 = fadd float %3906, %3909, !dbg !26 + %3911 = bitcast float %3910 to i32, !dbg !29 + %3912 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3911, i32 2, i32 31), !dbg !29 + %3913 = bitcast i32 %3912 to float, !dbg !29 + %3914 = fadd float %3910, %3913, !dbg !26 + %3915 = bitcast float %3914 to i32, !dbg !29 + %3916 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3915, i32 1, i32 31), !dbg !29 + %3917 = bitcast i32 %3916 to float, !dbg !29 + %3918 = fadd float %3914, %3917, !dbg !26 + %3919 = bitcast float %2919 to i32, !dbg !29 + %3920 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3919, i32 8, i32 31), !dbg !29 + %3921 = bitcast i32 %3920 to float, !dbg !29 + %3922 = fadd float %2919, %3921, !dbg !26 + %3923 = bitcast float %3922 to i32, !dbg !29 + %3924 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3923, i32 4, i32 31), !dbg !29 + %3925 = bitcast i32 %3924 to float, !dbg !29 + %3926 = fadd float %3922, %3925, !dbg !26 + %3927 = bitcast float %3926 to i32, !dbg !29 + %3928 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3927, i32 2, i32 31), !dbg !29 + %3929 = bitcast i32 %3928 to float, !dbg !29 + %3930 = fadd float %3926, %3929, !dbg !26 + %3931 = bitcast float %3930 to i32, !dbg !29 + %3932 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3931, i32 1, i32 31), !dbg !29 + %3933 = bitcast i32 %3932 to float, !dbg !29 + %3934 = fadd float %3930, %3933, !dbg !26 + %3935 = bitcast float %2926 to i32, !dbg !29 + %3936 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3935, i32 8, i32 31), !dbg !29 + %3937 = bitcast i32 %3936 to float, !dbg !29 + %3938 = fadd float %2926, %3937, !dbg !26 + %3939 = bitcast float %3938 to i32, !dbg !29 + %3940 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3939, i32 4, i32 31), !dbg !29 + %3941 = bitcast i32 %3940 to float, !dbg !29 + %3942 = fadd float %3938, %3941, !dbg !26 + %3943 = bitcast float %3942 to i32, !dbg !29 + %3944 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3943, i32 2, i32 31), !dbg !29 + %3945 = bitcast i32 %3944 to float, !dbg !29 + %3946 = fadd float %3942, %3945, !dbg !26 + %3947 = bitcast float %3946 to i32, !dbg !29 + %3948 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %3947, i32 1, i32 31), !dbg !29 + %3949 = bitcast i32 %3948 to float, !dbg !29 + %3950 = fadd float %3946, %3949, !dbg !26 + %3951 = fadd float %3, %2942, !dbg !32 + %3952 = fadd float %3, %2958, !dbg !32 + %3953 = fadd float %3, %2974, !dbg !32 + %3954 = fadd float %3, %2990, !dbg !32 + %3955 = fadd float %3, %3006, !dbg !32 + %3956 = fadd float %3, %3022, !dbg !32 + %3957 = fadd float %3, %3038, !dbg !32 + %3958 = fadd float %3, %3054, !dbg !32 + %3959 = fadd float %3, %3070, !dbg !32 + %3960 = fadd float %3, %3086, !dbg !32 + %3961 = fadd float %3, %3102, !dbg !32 + %3962 = fadd float %3, %3118, !dbg !32 + %3963 = fadd float %3, %3134, !dbg !32 + %3964 = fadd float %3, %3150, !dbg !32 + %3965 = fadd float %3, %3166, !dbg !32 + %3966 = fadd float %3, %3182, !dbg !32 + %3967 = fadd float %3, %3198, !dbg !32 + %3968 = fadd float %3, %3214, !dbg !32 + %3969 = fadd float %3, %3230, !dbg !32 + %3970 = fadd float %3, %3246, !dbg !32 + %3971 = fadd float %3, %3262, !dbg !32 + %3972 = fadd float %3, %3278, !dbg !32 + %3973 = fadd float %3, %3294, !dbg !32 + %3974 = fadd float %3, %3310, !dbg !32 + %3975 = fadd float %3, %3326, !dbg !32 + %3976 = fadd float %3, %3342, !dbg !32 + %3977 = fadd float %3, %3358, !dbg !32 + %3978 = fadd float %3, %3374, !dbg !32 + %3979 = fadd float %3, %3390, !dbg !32 + %3980 = fadd float %3, %3406, !dbg !32 + %3981 = fadd float %3, %3422, !dbg !32 + %3982 = fadd float %3, %3438, !dbg !32 + %3983 = fadd float %3, %3454, !dbg !32 + %3984 = fadd float %3, %3470, !dbg !32 + %3985 = fadd float %3, %3486, !dbg !32 + %3986 = fadd float %3, %3502, !dbg !32 + %3987 = fadd float %3, %3518, !dbg !32 + %3988 = fadd float %3, %3534, !dbg !32 + %3989 = fadd float %3, %3550, !dbg !32 + %3990 = fadd float %3, %3566, !dbg !32 + %3991 = fadd float %3, %3582, !dbg !32 + %3992 = fadd float %3, %3598, !dbg !32 + %3993 = fadd float %3, %3614, !dbg !32 + %3994 = fadd float %3, %3630, !dbg !32 + %3995 = fadd float %3, %3646, !dbg !32 + %3996 = fadd float %3, %3662, !dbg !32 + %3997 = fadd float %3, %3678, !dbg !32 + %3998 = fadd float %3, %3694, !dbg !32 + %3999 = fadd float %3, %3710, !dbg !32 + %4000 = fadd float %3, %3726, !dbg !32 + %4001 = fadd float %3, %3742, !dbg !32 + %4002 = fadd float %3, %3758, !dbg !32 + %4003 = fadd float %3, %3774, !dbg !32 + %4004 = fadd float %3, %3790, !dbg !32 + %4005 = fadd float %3, %3806, !dbg !32 + %4006 = fadd float %3, %3822, !dbg !32 + %4007 = fadd float %3, %3838, !dbg !32 + %4008 = fadd float %3, %3854, !dbg !32 + %4009 = fadd float %3, %3870, !dbg !32 + %4010 = fadd float %3, %3886, !dbg !32 + %4011 = fadd float %3, %3902, !dbg !32 + %4012 = fadd float %3, %3918, !dbg !32 + %4013 = fadd float %3, %3934, !dbg !32 + %4014 = fadd float %3, %3950, !dbg !32 + %4015 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i = icmp eq i32 %4015, 0, !dbg !33 + %4016 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i = icmp eq i32 %4016, 0, !dbg !33 + br i1 %.not.i, label %4022, label %4017, !dbg !33 + +4017: ; preds = %7 + br i1 %.not1.i, label %4020, label %4018, !dbg !33 + +4018: ; preds = %4017 + %4019 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3951) #7, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +4020: ; preds = %4017 + %4021 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3951) #7, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +4022: ; preds = %7 + br i1 %.not1.i, label %4025, label %4023, !dbg !33 + +4023: ; preds = %4022 + %4024 = tail call float @llvm.nvvm.sqrt.rn.f(float %3951) #7, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +4025: ; preds = %4022 + %4026 = tail call float @llvm.nvvm.sqrt.approx.f(float %3951) #7, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +__nv_sqrtf.exit: ; preds = %4018, %4020, %4023, %4025 + %.0.i = phi float [ %4019, %4018 ], [ %4021, %4020 ], [ %4024, %4023 ], [ %4026, %4025 ], !dbg !33 + %4027 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i1 = icmp eq i32 %4027, 0, !dbg !33 + %4028 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i4 = icmp eq i32 %4028, 0, !dbg !33 + br i1 %.not.i1, label %4034, label %4029, !dbg !33 + +4029: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %4032, label %4030, !dbg !33 + +4030: ; preds = %4029 + %4031 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3952) #7, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +4032: ; preds = %4029 + %4033 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3952) #7, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +4034: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %4037, label %4035, !dbg !33 + +4035: ; preds = %4034 + %4036 = tail call float @llvm.nvvm.sqrt.rn.f(float %3952) #7, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +4037: ; preds = %4034 + %4038 = tail call float @llvm.nvvm.sqrt.approx.f(float %3952) #7, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +__nv_sqrtf.exit5: ; preds = %4030, %4032, %4035, %4037 + %.0.i3 = phi float [ %4031, %4030 ], [ %4033, %4032 ], [ %4036, %4035 ], [ %4038, %4037 ], !dbg !33 + %4039 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i6 = icmp eq i32 %4039, 0, !dbg !33 + %4040 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i9 = icmp eq i32 %4040, 0, !dbg !33 + br i1 %.not.i6, label %4046, label %4041, !dbg !33 + +4041: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %4044, label %4042, !dbg !33 + +4042: ; preds = %4041 + %4043 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3953) #7, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +4044: ; preds = %4041 + %4045 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3953) #7, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +4046: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %4049, label %4047, !dbg !33 + +4047: ; preds = %4046 + %4048 = tail call float @llvm.nvvm.sqrt.rn.f(float %3953) #7, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +4049: ; preds = %4046 + %4050 = tail call float @llvm.nvvm.sqrt.approx.f(float %3953) #7, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +__nv_sqrtf.exit10: ; preds = %4042, %4044, %4047, %4049 + %.0.i8 = phi float [ %4043, %4042 ], [ %4045, %4044 ], [ %4048, %4047 ], [ %4050, %4049 ], !dbg !33 + %4051 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i11 = icmp eq i32 %4051, 0, !dbg !33 + %4052 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i14 = icmp eq i32 %4052, 0, !dbg !33 + br i1 %.not.i11, label %4058, label %4053, !dbg !33 + +4053: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %4056, label %4054, !dbg !33 + +4054: ; preds = %4053 + %4055 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3954) #7, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +4056: ; preds = %4053 + %4057 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3954) #7, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +4058: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %4061, label %4059, !dbg !33 + +4059: ; preds = %4058 + %4060 = tail call float @llvm.nvvm.sqrt.rn.f(float %3954) #7, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +4061: ; preds = %4058 + %4062 = tail call float @llvm.nvvm.sqrt.approx.f(float %3954) #7, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +__nv_sqrtf.exit15: ; preds = %4054, %4056, %4059, %4061 + %.0.i13 = phi float [ %4055, %4054 ], [ %4057, %4056 ], [ %4060, %4059 ], [ %4062, %4061 ], !dbg !33 + %4063 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i16 = icmp eq i32 %4063, 0, !dbg !33 + %4064 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i19 = icmp eq i32 %4064, 0, !dbg !33 + br i1 %.not.i16, label %4070, label %4065, !dbg !33 + +4065: ; preds = %__nv_sqrtf.exit15 + br i1 %.not1.i19, label %4068, label %4066, !dbg !33 + +4066: ; preds = %4065 + %4067 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3955) #7, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +4068: ; preds = %4065 + %4069 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3955) #7, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +4070: ; preds = %__nv_sqrtf.exit15 + br i1 %.not1.i19, label %4073, label %4071, !dbg !33 + +4071: ; preds = %4070 + %4072 = tail call float @llvm.nvvm.sqrt.rn.f(float %3955) #7, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +4073: ; preds = %4070 + %4074 = tail call float @llvm.nvvm.sqrt.approx.f(float %3955) #7, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +__nv_sqrtf.exit20: ; preds = %4066, %4068, %4071, %4073 + %.0.i18 = phi float [ %4067, %4066 ], [ %4069, %4068 ], [ %4072, %4071 ], [ %4074, %4073 ], !dbg !33 + %4075 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i21 = icmp eq i32 %4075, 0, !dbg !33 + %4076 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i24 = icmp eq i32 %4076, 0, !dbg !33 + br i1 %.not.i21, label %4082, label %4077, !dbg !33 + +4077: ; preds = %__nv_sqrtf.exit20 + br i1 %.not1.i24, label %4080, label %4078, !dbg !33 + +4078: ; preds = %4077 + %4079 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3956) #7, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +4080: ; preds = %4077 + %4081 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3956) #7, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +4082: ; preds = %__nv_sqrtf.exit20 + br i1 %.not1.i24, label %4085, label %4083, !dbg !33 + +4083: ; preds = %4082 + %4084 = tail call float @llvm.nvvm.sqrt.rn.f(float %3956) #7, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +4085: ; preds = %4082 + %4086 = tail call float @llvm.nvvm.sqrt.approx.f(float %3956) #7, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +__nv_sqrtf.exit25: ; preds = %4078, %4080, %4083, %4085 + %.0.i23 = phi float [ %4079, %4078 ], [ %4081, %4080 ], [ %4084, %4083 ], [ %4086, %4085 ], !dbg !33 + %4087 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i26 = icmp eq i32 %4087, 0, !dbg !33 + %4088 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i29 = icmp eq i32 %4088, 0, !dbg !33 + br i1 %.not.i26, label %4094, label %4089, !dbg !33 + +4089: ; preds = %__nv_sqrtf.exit25 + br i1 %.not1.i29, label %4092, label %4090, !dbg !33 + +4090: ; preds = %4089 + %4091 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3957) #7, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +4092: ; preds = %4089 + %4093 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3957) #7, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +4094: ; preds = %__nv_sqrtf.exit25 + br i1 %.not1.i29, label %4097, label %4095, !dbg !33 + +4095: ; preds = %4094 + %4096 = tail call float @llvm.nvvm.sqrt.rn.f(float %3957) #7, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +4097: ; preds = %4094 + %4098 = tail call float @llvm.nvvm.sqrt.approx.f(float %3957) #7, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +__nv_sqrtf.exit30: ; preds = %4090, %4092, %4095, %4097 + %.0.i28 = phi float [ %4091, %4090 ], [ %4093, %4092 ], [ %4096, %4095 ], [ %4098, %4097 ], !dbg !33 + %4099 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i31 = icmp eq i32 %4099, 0, !dbg !33 + %4100 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i34 = icmp eq i32 %4100, 0, !dbg !33 + br i1 %.not.i31, label %4106, label %4101, !dbg !33 + +4101: ; preds = %__nv_sqrtf.exit30 + br i1 %.not1.i34, label %4104, label %4102, !dbg !33 + +4102: ; preds = %4101 + %4103 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3958) #7, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +4104: ; preds = %4101 + %4105 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3958) #7, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +4106: ; preds = %__nv_sqrtf.exit30 + br i1 %.not1.i34, label %4109, label %4107, !dbg !33 + +4107: ; preds = %4106 + %4108 = tail call float @llvm.nvvm.sqrt.rn.f(float %3958) #7, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +4109: ; preds = %4106 + %4110 = tail call float @llvm.nvvm.sqrt.approx.f(float %3958) #7, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +__nv_sqrtf.exit35: ; preds = %4102, %4104, %4107, %4109 + %.0.i33 = phi float [ %4103, %4102 ], [ %4105, %4104 ], [ %4108, %4107 ], [ %4110, %4109 ], !dbg !33 + %4111 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i36 = icmp eq i32 %4111, 0, !dbg !33 + %4112 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i39 = icmp eq i32 %4112, 0, !dbg !33 + br i1 %.not.i36, label %4118, label %4113, !dbg !33 + +4113: ; preds = %__nv_sqrtf.exit35 + br i1 %.not1.i39, label %4116, label %4114, !dbg !33 + +4114: ; preds = %4113 + %4115 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3959) #7, !dbg !33 + br label %__nv_sqrtf.exit40, !dbg !33 + +4116: ; preds = %4113 + %4117 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3959) #7, !dbg !33 + br label %__nv_sqrtf.exit40, !dbg !33 + +4118: ; preds = %__nv_sqrtf.exit35 + br i1 %.not1.i39, label %4121, label %4119, !dbg !33 + +4119: ; preds = %4118 + %4120 = tail call float @llvm.nvvm.sqrt.rn.f(float %3959) #7, !dbg !33 + br label %__nv_sqrtf.exit40, !dbg !33 + +4121: ; preds = %4118 + %4122 = tail call float @llvm.nvvm.sqrt.approx.f(float %3959) #7, !dbg !33 + br label %__nv_sqrtf.exit40, !dbg !33 + +__nv_sqrtf.exit40: ; preds = %4114, %4116, %4119, %4121 + %.0.i38 = phi float [ %4115, %4114 ], [ %4117, %4116 ], [ %4120, %4119 ], [ %4122, %4121 ], !dbg !33 + %4123 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i41 = icmp eq i32 %4123, 0, !dbg !33 + %4124 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i44 = icmp eq i32 %4124, 0, !dbg !33 + br i1 %.not.i41, label %4130, label %4125, !dbg !33 + +4125: ; preds = %__nv_sqrtf.exit40 + br i1 %.not1.i44, label %4128, label %4126, !dbg !33 + +4126: ; preds = %4125 + %4127 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3960) #7, !dbg !33 + br label %__nv_sqrtf.exit45, !dbg !33 + +4128: ; preds = %4125 + %4129 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3960) #7, !dbg !33 + br label %__nv_sqrtf.exit45, !dbg !33 + +4130: ; preds = %__nv_sqrtf.exit40 + br i1 %.not1.i44, label %4133, label %4131, !dbg !33 + +4131: ; preds = %4130 + %4132 = tail call float @llvm.nvvm.sqrt.rn.f(float %3960) #7, !dbg !33 + br label %__nv_sqrtf.exit45, !dbg !33 + +4133: ; preds = %4130 + %4134 = tail call float @llvm.nvvm.sqrt.approx.f(float %3960) #7, !dbg !33 + br label %__nv_sqrtf.exit45, !dbg !33 + +__nv_sqrtf.exit45: ; preds = %4126, %4128, %4131, %4133 + %.0.i43 = phi float [ %4127, %4126 ], [ %4129, %4128 ], [ %4132, %4131 ], [ %4134, %4133 ], !dbg !33 + %4135 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i46 = icmp eq i32 %4135, 0, !dbg !33 + %4136 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i49 = icmp eq i32 %4136, 0, !dbg !33 + br i1 %.not.i46, label %4142, label %4137, !dbg !33 + +4137: ; preds = %__nv_sqrtf.exit45 + br i1 %.not1.i49, label %4140, label %4138, !dbg !33 + +4138: ; preds = %4137 + %4139 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3961) #7, !dbg !33 + br label %__nv_sqrtf.exit50, !dbg !33 + +4140: ; preds = %4137 + %4141 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3961) #7, !dbg !33 + br label %__nv_sqrtf.exit50, !dbg !33 + +4142: ; preds = %__nv_sqrtf.exit45 + br i1 %.not1.i49, label %4145, label %4143, !dbg !33 + +4143: ; preds = %4142 + %4144 = tail call float @llvm.nvvm.sqrt.rn.f(float %3961) #7, !dbg !33 + br label %__nv_sqrtf.exit50, !dbg !33 + +4145: ; preds = %4142 + %4146 = tail call float @llvm.nvvm.sqrt.approx.f(float %3961) #7, !dbg !33 + br label %__nv_sqrtf.exit50, !dbg !33 + +__nv_sqrtf.exit50: ; preds = %4138, %4140, %4143, %4145 + %.0.i48 = phi float [ %4139, %4138 ], [ %4141, %4140 ], [ %4144, %4143 ], [ %4146, %4145 ], !dbg !33 + %4147 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i51 = icmp eq i32 %4147, 0, !dbg !33 + %4148 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i54 = icmp eq i32 %4148, 0, !dbg !33 + br i1 %.not.i51, label %4154, label %4149, !dbg !33 + +4149: ; preds = %__nv_sqrtf.exit50 + br i1 %.not1.i54, label %4152, label %4150, !dbg !33 + +4150: ; preds = %4149 + %4151 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3962) #7, !dbg !33 + br label %__nv_sqrtf.exit55, !dbg !33 + +4152: ; preds = %4149 + %4153 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3962) #7, !dbg !33 + br label %__nv_sqrtf.exit55, !dbg !33 + +4154: ; preds = %__nv_sqrtf.exit50 + br i1 %.not1.i54, label %4157, label %4155, !dbg !33 + +4155: ; preds = %4154 + %4156 = tail call float @llvm.nvvm.sqrt.rn.f(float %3962) #7, !dbg !33 + br label %__nv_sqrtf.exit55, !dbg !33 + +4157: ; preds = %4154 + %4158 = tail call float @llvm.nvvm.sqrt.approx.f(float %3962) #7, !dbg !33 + br label %__nv_sqrtf.exit55, !dbg !33 + +__nv_sqrtf.exit55: ; preds = %4150, %4152, %4155, %4157 + %.0.i53 = phi float [ %4151, %4150 ], [ %4153, %4152 ], [ %4156, %4155 ], [ %4158, %4157 ], !dbg !33 + %4159 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i56 = icmp eq i32 %4159, 0, !dbg !33 + %4160 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i59 = icmp eq i32 %4160, 0, !dbg !33 + br i1 %.not.i56, label %4166, label %4161, !dbg !33 + +4161: ; preds = %__nv_sqrtf.exit55 + br i1 %.not1.i59, label %4164, label %4162, !dbg !33 + +4162: ; preds = %4161 + %4163 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3963) #7, !dbg !33 + br label %__nv_sqrtf.exit60, !dbg !33 + +4164: ; preds = %4161 + %4165 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3963) #7, !dbg !33 + br label %__nv_sqrtf.exit60, !dbg !33 + +4166: ; preds = %__nv_sqrtf.exit55 + br i1 %.not1.i59, label %4169, label %4167, !dbg !33 + +4167: ; preds = %4166 + %4168 = tail call float @llvm.nvvm.sqrt.rn.f(float %3963) #7, !dbg !33 + br label %__nv_sqrtf.exit60, !dbg !33 + +4169: ; preds = %4166 + %4170 = tail call float @llvm.nvvm.sqrt.approx.f(float %3963) #7, !dbg !33 + br label %__nv_sqrtf.exit60, !dbg !33 + +__nv_sqrtf.exit60: ; preds = %4162, %4164, %4167, %4169 + %.0.i58 = phi float [ %4163, %4162 ], [ %4165, %4164 ], [ %4168, %4167 ], [ %4170, %4169 ], !dbg !33 + %4171 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i61 = icmp eq i32 %4171, 0, !dbg !33 + %4172 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i64 = icmp eq i32 %4172, 0, !dbg !33 + br i1 %.not.i61, label %4178, label %4173, !dbg !33 + +4173: ; preds = %__nv_sqrtf.exit60 + br i1 %.not1.i64, label %4176, label %4174, !dbg !33 + +4174: ; preds = %4173 + %4175 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3964) #7, !dbg !33 + br label %__nv_sqrtf.exit65, !dbg !33 + +4176: ; preds = %4173 + %4177 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3964) #7, !dbg !33 + br label %__nv_sqrtf.exit65, !dbg !33 + +4178: ; preds = %__nv_sqrtf.exit60 + br i1 %.not1.i64, label %4181, label %4179, !dbg !33 + +4179: ; preds = %4178 + %4180 = tail call float @llvm.nvvm.sqrt.rn.f(float %3964) #7, !dbg !33 + br label %__nv_sqrtf.exit65, !dbg !33 + +4181: ; preds = %4178 + %4182 = tail call float @llvm.nvvm.sqrt.approx.f(float %3964) #7, !dbg !33 + br label %__nv_sqrtf.exit65, !dbg !33 + +__nv_sqrtf.exit65: ; preds = %4174, %4176, %4179, %4181 + %.0.i63 = phi float [ %4175, %4174 ], [ %4177, %4176 ], [ %4180, %4179 ], [ %4182, %4181 ], !dbg !33 + %4183 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i66 = icmp eq i32 %4183, 0, !dbg !33 + %4184 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i69 = icmp eq i32 %4184, 0, !dbg !33 + br i1 %.not.i66, label %4190, label %4185, !dbg !33 + +4185: ; preds = %__nv_sqrtf.exit65 + br i1 %.not1.i69, label %4188, label %4186, !dbg !33 + +4186: ; preds = %4185 + %4187 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3965) #7, !dbg !33 + br label %__nv_sqrtf.exit70, !dbg !33 + +4188: ; preds = %4185 + %4189 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3965) #7, !dbg !33 + br label %__nv_sqrtf.exit70, !dbg !33 + +4190: ; preds = %__nv_sqrtf.exit65 + br i1 %.not1.i69, label %4193, label %4191, !dbg !33 + +4191: ; preds = %4190 + %4192 = tail call float @llvm.nvvm.sqrt.rn.f(float %3965) #7, !dbg !33 + br label %__nv_sqrtf.exit70, !dbg !33 + +4193: ; preds = %4190 + %4194 = tail call float @llvm.nvvm.sqrt.approx.f(float %3965) #7, !dbg !33 + br label %__nv_sqrtf.exit70, !dbg !33 + +__nv_sqrtf.exit70: ; preds = %4186, %4188, %4191, %4193 + %.0.i68 = phi float [ %4187, %4186 ], [ %4189, %4188 ], [ %4192, %4191 ], [ %4194, %4193 ], !dbg !33 + %4195 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i71 = icmp eq i32 %4195, 0, !dbg !33 + %4196 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i74 = icmp eq i32 %4196, 0, !dbg !33 + br i1 %.not.i71, label %4202, label %4197, !dbg !33 + +4197: ; preds = %__nv_sqrtf.exit70 + br i1 %.not1.i74, label %4200, label %4198, !dbg !33 + +4198: ; preds = %4197 + %4199 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3966) #7, !dbg !33 + br label %__nv_sqrtf.exit75, !dbg !33 + +4200: ; preds = %4197 + %4201 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3966) #7, !dbg !33 + br label %__nv_sqrtf.exit75, !dbg !33 + +4202: ; preds = %__nv_sqrtf.exit70 + br i1 %.not1.i74, label %4205, label %4203, !dbg !33 + +4203: ; preds = %4202 + %4204 = tail call float @llvm.nvvm.sqrt.rn.f(float %3966) #7, !dbg !33 + br label %__nv_sqrtf.exit75, !dbg !33 + +4205: ; preds = %4202 + %4206 = tail call float @llvm.nvvm.sqrt.approx.f(float %3966) #7, !dbg !33 + br label %__nv_sqrtf.exit75, !dbg !33 + +__nv_sqrtf.exit75: ; preds = %4198, %4200, %4203, %4205 + %.0.i73 = phi float [ %4199, %4198 ], [ %4201, %4200 ], [ %4204, %4203 ], [ %4206, %4205 ], !dbg !33 + %4207 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i76 = icmp eq i32 %4207, 0, !dbg !33 + %4208 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i79 = icmp eq i32 %4208, 0, !dbg !33 + br i1 %.not.i76, label %4214, label %4209, !dbg !33 + +4209: ; preds = %__nv_sqrtf.exit75 + br i1 %.not1.i79, label %4212, label %4210, !dbg !33 + +4210: ; preds = %4209 + %4211 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3967) #7, !dbg !33 + br label %__nv_sqrtf.exit80, !dbg !33 + +4212: ; preds = %4209 + %4213 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3967) #7, !dbg !33 + br label %__nv_sqrtf.exit80, !dbg !33 + +4214: ; preds = %__nv_sqrtf.exit75 + br i1 %.not1.i79, label %4217, label %4215, !dbg !33 + +4215: ; preds = %4214 + %4216 = tail call float @llvm.nvvm.sqrt.rn.f(float %3967) #7, !dbg !33 + br label %__nv_sqrtf.exit80, !dbg !33 + +4217: ; preds = %4214 + %4218 = tail call float @llvm.nvvm.sqrt.approx.f(float %3967) #7, !dbg !33 + br label %__nv_sqrtf.exit80, !dbg !33 + +__nv_sqrtf.exit80: ; preds = %4210, %4212, %4215, %4217 + %.0.i78 = phi float [ %4211, %4210 ], [ %4213, %4212 ], [ %4216, %4215 ], [ %4218, %4217 ], !dbg !33 + %4219 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i81 = icmp eq i32 %4219, 0, !dbg !33 + %4220 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i84 = icmp eq i32 %4220, 0, !dbg !33 + br i1 %.not.i81, label %4226, label %4221, !dbg !33 + +4221: ; preds = %__nv_sqrtf.exit80 + br i1 %.not1.i84, label %4224, label %4222, !dbg !33 + +4222: ; preds = %4221 + %4223 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3968) #7, !dbg !33 + br label %__nv_sqrtf.exit85, !dbg !33 + +4224: ; preds = %4221 + %4225 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3968) #7, !dbg !33 + br label %__nv_sqrtf.exit85, !dbg !33 + +4226: ; preds = %__nv_sqrtf.exit80 + br i1 %.not1.i84, label %4229, label %4227, !dbg !33 + +4227: ; preds = %4226 + %4228 = tail call float @llvm.nvvm.sqrt.rn.f(float %3968) #7, !dbg !33 + br label %__nv_sqrtf.exit85, !dbg !33 + +4229: ; preds = %4226 + %4230 = tail call float @llvm.nvvm.sqrt.approx.f(float %3968) #7, !dbg !33 + br label %__nv_sqrtf.exit85, !dbg !33 + +__nv_sqrtf.exit85: ; preds = %4222, %4224, %4227, %4229 + %.0.i83 = phi float [ %4223, %4222 ], [ %4225, %4224 ], [ %4228, %4227 ], [ %4230, %4229 ], !dbg !33 + %4231 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i86 = icmp eq i32 %4231, 0, !dbg !33 + %4232 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i89 = icmp eq i32 %4232, 0, !dbg !33 + br i1 %.not.i86, label %4238, label %4233, !dbg !33 + +4233: ; preds = %__nv_sqrtf.exit85 + br i1 %.not1.i89, label %4236, label %4234, !dbg !33 + +4234: ; preds = %4233 + %4235 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3969) #7, !dbg !33 + br label %__nv_sqrtf.exit90, !dbg !33 + +4236: ; preds = %4233 + %4237 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3969) #7, !dbg !33 + br label %__nv_sqrtf.exit90, !dbg !33 + +4238: ; preds = %__nv_sqrtf.exit85 + br i1 %.not1.i89, label %4241, label %4239, !dbg !33 + +4239: ; preds = %4238 + %4240 = tail call float @llvm.nvvm.sqrt.rn.f(float %3969) #7, !dbg !33 + br label %__nv_sqrtf.exit90, !dbg !33 + +4241: ; preds = %4238 + %4242 = tail call float @llvm.nvvm.sqrt.approx.f(float %3969) #7, !dbg !33 + br label %__nv_sqrtf.exit90, !dbg !33 + +__nv_sqrtf.exit90: ; preds = %4234, %4236, %4239, %4241 + %.0.i88 = phi float [ %4235, %4234 ], [ %4237, %4236 ], [ %4240, %4239 ], [ %4242, %4241 ], !dbg !33 + %4243 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i91 = icmp eq i32 %4243, 0, !dbg !33 + %4244 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i94 = icmp eq i32 %4244, 0, !dbg !33 + br i1 %.not.i91, label %4250, label %4245, !dbg !33 + +4245: ; preds = %__nv_sqrtf.exit90 + br i1 %.not1.i94, label %4248, label %4246, !dbg !33 + +4246: ; preds = %4245 + %4247 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3970) #7, !dbg !33 + br label %__nv_sqrtf.exit95, !dbg !33 + +4248: ; preds = %4245 + %4249 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3970) #7, !dbg !33 + br label %__nv_sqrtf.exit95, !dbg !33 + +4250: ; preds = %__nv_sqrtf.exit90 + br i1 %.not1.i94, label %4253, label %4251, !dbg !33 + +4251: ; preds = %4250 + %4252 = tail call float @llvm.nvvm.sqrt.rn.f(float %3970) #7, !dbg !33 + br label %__nv_sqrtf.exit95, !dbg !33 + +4253: ; preds = %4250 + %4254 = tail call float @llvm.nvvm.sqrt.approx.f(float %3970) #7, !dbg !33 + br label %__nv_sqrtf.exit95, !dbg !33 + +__nv_sqrtf.exit95: ; preds = %4246, %4248, %4251, %4253 + %.0.i93 = phi float [ %4247, %4246 ], [ %4249, %4248 ], [ %4252, %4251 ], [ %4254, %4253 ], !dbg !33 + %4255 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i96 = icmp eq i32 %4255, 0, !dbg !33 + %4256 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i99 = icmp eq i32 %4256, 0, !dbg !33 + br i1 %.not.i96, label %4262, label %4257, !dbg !33 + +4257: ; preds = %__nv_sqrtf.exit95 + br i1 %.not1.i99, label %4260, label %4258, !dbg !33 + +4258: ; preds = %4257 + %4259 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3971) #7, !dbg !33 + br label %__nv_sqrtf.exit100, !dbg !33 + +4260: ; preds = %4257 + %4261 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3971) #7, !dbg !33 + br label %__nv_sqrtf.exit100, !dbg !33 + +4262: ; preds = %__nv_sqrtf.exit95 + br i1 %.not1.i99, label %4265, label %4263, !dbg !33 + +4263: ; preds = %4262 + %4264 = tail call float @llvm.nvvm.sqrt.rn.f(float %3971) #7, !dbg !33 + br label %__nv_sqrtf.exit100, !dbg !33 + +4265: ; preds = %4262 + %4266 = tail call float @llvm.nvvm.sqrt.approx.f(float %3971) #7, !dbg !33 + br label %__nv_sqrtf.exit100, !dbg !33 + +__nv_sqrtf.exit100: ; preds = %4258, %4260, %4263, %4265 + %.0.i98 = phi float [ %4259, %4258 ], [ %4261, %4260 ], [ %4264, %4263 ], [ %4266, %4265 ], !dbg !33 + %4267 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i101 = icmp eq i32 %4267, 0, !dbg !33 + %4268 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i104 = icmp eq i32 %4268, 0, !dbg !33 + br i1 %.not.i101, label %4274, label %4269, !dbg !33 + +4269: ; preds = %__nv_sqrtf.exit100 + br i1 %.not1.i104, label %4272, label %4270, !dbg !33 + +4270: ; preds = %4269 + %4271 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3972) #7, !dbg !33 + br label %__nv_sqrtf.exit105, !dbg !33 + +4272: ; preds = %4269 + %4273 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3972) #7, !dbg !33 + br label %__nv_sqrtf.exit105, !dbg !33 + +4274: ; preds = %__nv_sqrtf.exit100 + br i1 %.not1.i104, label %4277, label %4275, !dbg !33 + +4275: ; preds = %4274 + %4276 = tail call float @llvm.nvvm.sqrt.rn.f(float %3972) #7, !dbg !33 + br label %__nv_sqrtf.exit105, !dbg !33 + +4277: ; preds = %4274 + %4278 = tail call float @llvm.nvvm.sqrt.approx.f(float %3972) #7, !dbg !33 + br label %__nv_sqrtf.exit105, !dbg !33 + +__nv_sqrtf.exit105: ; preds = %4270, %4272, %4275, %4277 + %.0.i103 = phi float [ %4271, %4270 ], [ %4273, %4272 ], [ %4276, %4275 ], [ %4278, %4277 ], !dbg !33 + %4279 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i106 = icmp eq i32 %4279, 0, !dbg !33 + %4280 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i109 = icmp eq i32 %4280, 0, !dbg !33 + br i1 %.not.i106, label %4286, label %4281, !dbg !33 + +4281: ; preds = %__nv_sqrtf.exit105 + br i1 %.not1.i109, label %4284, label %4282, !dbg !33 + +4282: ; preds = %4281 + %4283 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3973) #7, !dbg !33 + br label %__nv_sqrtf.exit110, !dbg !33 + +4284: ; preds = %4281 + %4285 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3973) #7, !dbg !33 + br label %__nv_sqrtf.exit110, !dbg !33 + +4286: ; preds = %__nv_sqrtf.exit105 + br i1 %.not1.i109, label %4289, label %4287, !dbg !33 + +4287: ; preds = %4286 + %4288 = tail call float @llvm.nvvm.sqrt.rn.f(float %3973) #7, !dbg !33 + br label %__nv_sqrtf.exit110, !dbg !33 + +4289: ; preds = %4286 + %4290 = tail call float @llvm.nvvm.sqrt.approx.f(float %3973) #7, !dbg !33 + br label %__nv_sqrtf.exit110, !dbg !33 + +__nv_sqrtf.exit110: ; preds = %4282, %4284, %4287, %4289 + %.0.i108 = phi float [ %4283, %4282 ], [ %4285, %4284 ], [ %4288, %4287 ], [ %4290, %4289 ], !dbg !33 + %4291 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i111 = icmp eq i32 %4291, 0, !dbg !33 + %4292 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i114 = icmp eq i32 %4292, 0, !dbg !33 + br i1 %.not.i111, label %4298, label %4293, !dbg !33 + +4293: ; preds = %__nv_sqrtf.exit110 + br i1 %.not1.i114, label %4296, label %4294, !dbg !33 + +4294: ; preds = %4293 + %4295 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3974) #7, !dbg !33 + br label %__nv_sqrtf.exit115, !dbg !33 + +4296: ; preds = %4293 + %4297 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3974) #7, !dbg !33 + br label %__nv_sqrtf.exit115, !dbg !33 + +4298: ; preds = %__nv_sqrtf.exit110 + br i1 %.not1.i114, label %4301, label %4299, !dbg !33 + +4299: ; preds = %4298 + %4300 = tail call float @llvm.nvvm.sqrt.rn.f(float %3974) #7, !dbg !33 + br label %__nv_sqrtf.exit115, !dbg !33 + +4301: ; preds = %4298 + %4302 = tail call float @llvm.nvvm.sqrt.approx.f(float %3974) #7, !dbg !33 + br label %__nv_sqrtf.exit115, !dbg !33 + +__nv_sqrtf.exit115: ; preds = %4294, %4296, %4299, %4301 + %.0.i113 = phi float [ %4295, %4294 ], [ %4297, %4296 ], [ %4300, %4299 ], [ %4302, %4301 ], !dbg !33 + %4303 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i116 = icmp eq i32 %4303, 0, !dbg !33 + %4304 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i119 = icmp eq i32 %4304, 0, !dbg !33 + br i1 %.not.i116, label %4310, label %4305, !dbg !33 + +4305: ; preds = %__nv_sqrtf.exit115 + br i1 %.not1.i119, label %4308, label %4306, !dbg !33 + +4306: ; preds = %4305 + %4307 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3975) #7, !dbg !33 + br label %__nv_sqrtf.exit120, !dbg !33 + +4308: ; preds = %4305 + %4309 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3975) #7, !dbg !33 + br label %__nv_sqrtf.exit120, !dbg !33 + +4310: ; preds = %__nv_sqrtf.exit115 + br i1 %.not1.i119, label %4313, label %4311, !dbg !33 + +4311: ; preds = %4310 + %4312 = tail call float @llvm.nvvm.sqrt.rn.f(float %3975) #7, !dbg !33 + br label %__nv_sqrtf.exit120, !dbg !33 + +4313: ; preds = %4310 + %4314 = tail call float @llvm.nvvm.sqrt.approx.f(float %3975) #7, !dbg !33 + br label %__nv_sqrtf.exit120, !dbg !33 + +__nv_sqrtf.exit120: ; preds = %4306, %4308, %4311, %4313 + %.0.i118 = phi float [ %4307, %4306 ], [ %4309, %4308 ], [ %4312, %4311 ], [ %4314, %4313 ], !dbg !33 + %4315 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i121 = icmp eq i32 %4315, 0, !dbg !33 + %4316 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i124 = icmp eq i32 %4316, 0, !dbg !33 + br i1 %.not.i121, label %4322, label %4317, !dbg !33 + +4317: ; preds = %__nv_sqrtf.exit120 + br i1 %.not1.i124, label %4320, label %4318, !dbg !33 + +4318: ; preds = %4317 + %4319 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3976) #7, !dbg !33 + br label %__nv_sqrtf.exit125, !dbg !33 + +4320: ; preds = %4317 + %4321 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3976) #7, !dbg !33 + br label %__nv_sqrtf.exit125, !dbg !33 + +4322: ; preds = %__nv_sqrtf.exit120 + br i1 %.not1.i124, label %4325, label %4323, !dbg !33 + +4323: ; preds = %4322 + %4324 = tail call float @llvm.nvvm.sqrt.rn.f(float %3976) #7, !dbg !33 + br label %__nv_sqrtf.exit125, !dbg !33 + +4325: ; preds = %4322 + %4326 = tail call float @llvm.nvvm.sqrt.approx.f(float %3976) #7, !dbg !33 + br label %__nv_sqrtf.exit125, !dbg !33 + +__nv_sqrtf.exit125: ; preds = %4318, %4320, %4323, %4325 + %.0.i123 = phi float [ %4319, %4318 ], [ %4321, %4320 ], [ %4324, %4323 ], [ %4326, %4325 ], !dbg !33 + %4327 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i126 = icmp eq i32 %4327, 0, !dbg !33 + %4328 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i129 = icmp eq i32 %4328, 0, !dbg !33 + br i1 %.not.i126, label %4334, label %4329, !dbg !33 + +4329: ; preds = %__nv_sqrtf.exit125 + br i1 %.not1.i129, label %4332, label %4330, !dbg !33 + +4330: ; preds = %4329 + %4331 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3977) #7, !dbg !33 + br label %__nv_sqrtf.exit130, !dbg !33 + +4332: ; preds = %4329 + %4333 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3977) #7, !dbg !33 + br label %__nv_sqrtf.exit130, !dbg !33 + +4334: ; preds = %__nv_sqrtf.exit125 + br i1 %.not1.i129, label %4337, label %4335, !dbg !33 + +4335: ; preds = %4334 + %4336 = tail call float @llvm.nvvm.sqrt.rn.f(float %3977) #7, !dbg !33 + br label %__nv_sqrtf.exit130, !dbg !33 + +4337: ; preds = %4334 + %4338 = tail call float @llvm.nvvm.sqrt.approx.f(float %3977) #7, !dbg !33 + br label %__nv_sqrtf.exit130, !dbg !33 + +__nv_sqrtf.exit130: ; preds = %4330, %4332, %4335, %4337 + %.0.i128 = phi float [ %4331, %4330 ], [ %4333, %4332 ], [ %4336, %4335 ], [ %4338, %4337 ], !dbg !33 + %4339 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i131 = icmp eq i32 %4339, 0, !dbg !33 + %4340 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i134 = icmp eq i32 %4340, 0, !dbg !33 + br i1 %.not.i131, label %4346, label %4341, !dbg !33 + +4341: ; preds = %__nv_sqrtf.exit130 + br i1 %.not1.i134, label %4344, label %4342, !dbg !33 + +4342: ; preds = %4341 + %4343 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3978) #7, !dbg !33 + br label %__nv_sqrtf.exit135, !dbg !33 + +4344: ; preds = %4341 + %4345 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3978) #7, !dbg !33 + br label %__nv_sqrtf.exit135, !dbg !33 + +4346: ; preds = %__nv_sqrtf.exit130 + br i1 %.not1.i134, label %4349, label %4347, !dbg !33 + +4347: ; preds = %4346 + %4348 = tail call float @llvm.nvvm.sqrt.rn.f(float %3978) #7, !dbg !33 + br label %__nv_sqrtf.exit135, !dbg !33 + +4349: ; preds = %4346 + %4350 = tail call float @llvm.nvvm.sqrt.approx.f(float %3978) #7, !dbg !33 + br label %__nv_sqrtf.exit135, !dbg !33 + +__nv_sqrtf.exit135: ; preds = %4342, %4344, %4347, %4349 + %.0.i133 = phi float [ %4343, %4342 ], [ %4345, %4344 ], [ %4348, %4347 ], [ %4350, %4349 ], !dbg !33 + %4351 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i136 = icmp eq i32 %4351, 0, !dbg !33 + %4352 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i139 = icmp eq i32 %4352, 0, !dbg !33 + br i1 %.not.i136, label %4358, label %4353, !dbg !33 + +4353: ; preds = %__nv_sqrtf.exit135 + br i1 %.not1.i139, label %4356, label %4354, !dbg !33 + +4354: ; preds = %4353 + %4355 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3979) #7, !dbg !33 + br label %__nv_sqrtf.exit140, !dbg !33 + +4356: ; preds = %4353 + %4357 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3979) #7, !dbg !33 + br label %__nv_sqrtf.exit140, !dbg !33 + +4358: ; preds = %__nv_sqrtf.exit135 + br i1 %.not1.i139, label %4361, label %4359, !dbg !33 + +4359: ; preds = %4358 + %4360 = tail call float @llvm.nvvm.sqrt.rn.f(float %3979) #7, !dbg !33 + br label %__nv_sqrtf.exit140, !dbg !33 + +4361: ; preds = %4358 + %4362 = tail call float @llvm.nvvm.sqrt.approx.f(float %3979) #7, !dbg !33 + br label %__nv_sqrtf.exit140, !dbg !33 + +__nv_sqrtf.exit140: ; preds = %4354, %4356, %4359, %4361 + %.0.i138 = phi float [ %4355, %4354 ], [ %4357, %4356 ], [ %4360, %4359 ], [ %4362, %4361 ], !dbg !33 + %4363 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i141 = icmp eq i32 %4363, 0, !dbg !33 + %4364 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i144 = icmp eq i32 %4364, 0, !dbg !33 + br i1 %.not.i141, label %4370, label %4365, !dbg !33 + +4365: ; preds = %__nv_sqrtf.exit140 + br i1 %.not1.i144, label %4368, label %4366, !dbg !33 + +4366: ; preds = %4365 + %4367 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3980) #7, !dbg !33 + br label %__nv_sqrtf.exit145, !dbg !33 + +4368: ; preds = %4365 + %4369 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3980) #7, !dbg !33 + br label %__nv_sqrtf.exit145, !dbg !33 + +4370: ; preds = %__nv_sqrtf.exit140 + br i1 %.not1.i144, label %4373, label %4371, !dbg !33 + +4371: ; preds = %4370 + %4372 = tail call float @llvm.nvvm.sqrt.rn.f(float %3980) #7, !dbg !33 + br label %__nv_sqrtf.exit145, !dbg !33 + +4373: ; preds = %4370 + %4374 = tail call float @llvm.nvvm.sqrt.approx.f(float %3980) #7, !dbg !33 + br label %__nv_sqrtf.exit145, !dbg !33 + +__nv_sqrtf.exit145: ; preds = %4366, %4368, %4371, %4373 + %.0.i143 = phi float [ %4367, %4366 ], [ %4369, %4368 ], [ %4372, %4371 ], [ %4374, %4373 ], !dbg !33 + %4375 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i146 = icmp eq i32 %4375, 0, !dbg !33 + %4376 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i149 = icmp eq i32 %4376, 0, !dbg !33 + br i1 %.not.i146, label %4382, label %4377, !dbg !33 + +4377: ; preds = %__nv_sqrtf.exit145 + br i1 %.not1.i149, label %4380, label %4378, !dbg !33 + +4378: ; preds = %4377 + %4379 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3981) #7, !dbg !33 + br label %__nv_sqrtf.exit150, !dbg !33 + +4380: ; preds = %4377 + %4381 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3981) #7, !dbg !33 + br label %__nv_sqrtf.exit150, !dbg !33 + +4382: ; preds = %__nv_sqrtf.exit145 + br i1 %.not1.i149, label %4385, label %4383, !dbg !33 + +4383: ; preds = %4382 + %4384 = tail call float @llvm.nvvm.sqrt.rn.f(float %3981) #7, !dbg !33 + br label %__nv_sqrtf.exit150, !dbg !33 + +4385: ; preds = %4382 + %4386 = tail call float @llvm.nvvm.sqrt.approx.f(float %3981) #7, !dbg !33 + br label %__nv_sqrtf.exit150, !dbg !33 + +__nv_sqrtf.exit150: ; preds = %4378, %4380, %4383, %4385 + %.0.i148 = phi float [ %4379, %4378 ], [ %4381, %4380 ], [ %4384, %4383 ], [ %4386, %4385 ], !dbg !33 + %4387 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i151 = icmp eq i32 %4387, 0, !dbg !33 + %4388 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i154 = icmp eq i32 %4388, 0, !dbg !33 + br i1 %.not.i151, label %4394, label %4389, !dbg !33 + +4389: ; preds = %__nv_sqrtf.exit150 + br i1 %.not1.i154, label %4392, label %4390, !dbg !33 + +4390: ; preds = %4389 + %4391 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3982) #7, !dbg !33 + br label %__nv_sqrtf.exit155, !dbg !33 + +4392: ; preds = %4389 + %4393 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3982) #7, !dbg !33 + br label %__nv_sqrtf.exit155, !dbg !33 + +4394: ; preds = %__nv_sqrtf.exit150 + br i1 %.not1.i154, label %4397, label %4395, !dbg !33 + +4395: ; preds = %4394 + %4396 = tail call float @llvm.nvvm.sqrt.rn.f(float %3982) #7, !dbg !33 + br label %__nv_sqrtf.exit155, !dbg !33 + +4397: ; preds = %4394 + %4398 = tail call float @llvm.nvvm.sqrt.approx.f(float %3982) #7, !dbg !33 + br label %__nv_sqrtf.exit155, !dbg !33 + +__nv_sqrtf.exit155: ; preds = %4390, %4392, %4395, %4397 + %.0.i153 = phi float [ %4391, %4390 ], [ %4393, %4392 ], [ %4396, %4395 ], [ %4398, %4397 ], !dbg !33 + %4399 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i156 = icmp eq i32 %4399, 0, !dbg !33 + %4400 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i159 = icmp eq i32 %4400, 0, !dbg !33 + br i1 %.not.i156, label %4406, label %4401, !dbg !33 + +4401: ; preds = %__nv_sqrtf.exit155 + br i1 %.not1.i159, label %4404, label %4402, !dbg !33 + +4402: ; preds = %4401 + %4403 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3983) #7, !dbg !33 + br label %__nv_sqrtf.exit160, !dbg !33 + +4404: ; preds = %4401 + %4405 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3983) #7, !dbg !33 + br label %__nv_sqrtf.exit160, !dbg !33 + +4406: ; preds = %__nv_sqrtf.exit155 + br i1 %.not1.i159, label %4409, label %4407, !dbg !33 + +4407: ; preds = %4406 + %4408 = tail call float @llvm.nvvm.sqrt.rn.f(float %3983) #7, !dbg !33 + br label %__nv_sqrtf.exit160, !dbg !33 + +4409: ; preds = %4406 + %4410 = tail call float @llvm.nvvm.sqrt.approx.f(float %3983) #7, !dbg !33 + br label %__nv_sqrtf.exit160, !dbg !33 + +__nv_sqrtf.exit160: ; preds = %4402, %4404, %4407, %4409 + %.0.i158 = phi float [ %4403, %4402 ], [ %4405, %4404 ], [ %4408, %4407 ], [ %4410, %4409 ], !dbg !33 + %4411 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i161 = icmp eq i32 %4411, 0, !dbg !33 + %4412 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i164 = icmp eq i32 %4412, 0, !dbg !33 + br i1 %.not.i161, label %4418, label %4413, !dbg !33 + +4413: ; preds = %__nv_sqrtf.exit160 + br i1 %.not1.i164, label %4416, label %4414, !dbg !33 + +4414: ; preds = %4413 + %4415 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3984) #7, !dbg !33 + br label %__nv_sqrtf.exit165, !dbg !33 + +4416: ; preds = %4413 + %4417 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3984) #7, !dbg !33 + br label %__nv_sqrtf.exit165, !dbg !33 + +4418: ; preds = %__nv_sqrtf.exit160 + br i1 %.not1.i164, label %4421, label %4419, !dbg !33 + +4419: ; preds = %4418 + %4420 = tail call float @llvm.nvvm.sqrt.rn.f(float %3984) #7, !dbg !33 + br label %__nv_sqrtf.exit165, !dbg !33 + +4421: ; preds = %4418 + %4422 = tail call float @llvm.nvvm.sqrt.approx.f(float %3984) #7, !dbg !33 + br label %__nv_sqrtf.exit165, !dbg !33 + +__nv_sqrtf.exit165: ; preds = %4414, %4416, %4419, %4421 + %.0.i163 = phi float [ %4415, %4414 ], [ %4417, %4416 ], [ %4420, %4419 ], [ %4422, %4421 ], !dbg !33 + %4423 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i166 = icmp eq i32 %4423, 0, !dbg !33 + %4424 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i169 = icmp eq i32 %4424, 0, !dbg !33 + br i1 %.not.i166, label %4430, label %4425, !dbg !33 + +4425: ; preds = %__nv_sqrtf.exit165 + br i1 %.not1.i169, label %4428, label %4426, !dbg !33 + +4426: ; preds = %4425 + %4427 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3985) #7, !dbg !33 + br label %__nv_sqrtf.exit170, !dbg !33 + +4428: ; preds = %4425 + %4429 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3985) #7, !dbg !33 + br label %__nv_sqrtf.exit170, !dbg !33 + +4430: ; preds = %__nv_sqrtf.exit165 + br i1 %.not1.i169, label %4433, label %4431, !dbg !33 + +4431: ; preds = %4430 + %4432 = tail call float @llvm.nvvm.sqrt.rn.f(float %3985) #7, !dbg !33 + br label %__nv_sqrtf.exit170, !dbg !33 + +4433: ; preds = %4430 + %4434 = tail call float @llvm.nvvm.sqrt.approx.f(float %3985) #7, !dbg !33 + br label %__nv_sqrtf.exit170, !dbg !33 + +__nv_sqrtf.exit170: ; preds = %4426, %4428, %4431, %4433 + %.0.i168 = phi float [ %4427, %4426 ], [ %4429, %4428 ], [ %4432, %4431 ], [ %4434, %4433 ], !dbg !33 + %4435 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i171 = icmp eq i32 %4435, 0, !dbg !33 + %4436 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i174 = icmp eq i32 %4436, 0, !dbg !33 + br i1 %.not.i171, label %4442, label %4437, !dbg !33 + +4437: ; preds = %__nv_sqrtf.exit170 + br i1 %.not1.i174, label %4440, label %4438, !dbg !33 + +4438: ; preds = %4437 + %4439 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3986) #7, !dbg !33 + br label %__nv_sqrtf.exit175, !dbg !33 + +4440: ; preds = %4437 + %4441 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3986) #7, !dbg !33 + br label %__nv_sqrtf.exit175, !dbg !33 + +4442: ; preds = %__nv_sqrtf.exit170 + br i1 %.not1.i174, label %4445, label %4443, !dbg !33 + +4443: ; preds = %4442 + %4444 = tail call float @llvm.nvvm.sqrt.rn.f(float %3986) #7, !dbg !33 + br label %__nv_sqrtf.exit175, !dbg !33 + +4445: ; preds = %4442 + %4446 = tail call float @llvm.nvvm.sqrt.approx.f(float %3986) #7, !dbg !33 + br label %__nv_sqrtf.exit175, !dbg !33 + +__nv_sqrtf.exit175: ; preds = %4438, %4440, %4443, %4445 + %.0.i173 = phi float [ %4439, %4438 ], [ %4441, %4440 ], [ %4444, %4443 ], [ %4446, %4445 ], !dbg !33 + %4447 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i176 = icmp eq i32 %4447, 0, !dbg !33 + %4448 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i179 = icmp eq i32 %4448, 0, !dbg !33 + br i1 %.not.i176, label %4454, label %4449, !dbg !33 + +4449: ; preds = %__nv_sqrtf.exit175 + br i1 %.not1.i179, label %4452, label %4450, !dbg !33 + +4450: ; preds = %4449 + %4451 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3987) #7, !dbg !33 + br label %__nv_sqrtf.exit180, !dbg !33 + +4452: ; preds = %4449 + %4453 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3987) #7, !dbg !33 + br label %__nv_sqrtf.exit180, !dbg !33 + +4454: ; preds = %__nv_sqrtf.exit175 + br i1 %.not1.i179, label %4457, label %4455, !dbg !33 + +4455: ; preds = %4454 + %4456 = tail call float @llvm.nvvm.sqrt.rn.f(float %3987) #7, !dbg !33 + br label %__nv_sqrtf.exit180, !dbg !33 + +4457: ; preds = %4454 + %4458 = tail call float @llvm.nvvm.sqrt.approx.f(float %3987) #7, !dbg !33 + br label %__nv_sqrtf.exit180, !dbg !33 + +__nv_sqrtf.exit180: ; preds = %4450, %4452, %4455, %4457 + %.0.i178 = phi float [ %4451, %4450 ], [ %4453, %4452 ], [ %4456, %4455 ], [ %4458, %4457 ], !dbg !33 + %4459 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i181 = icmp eq i32 %4459, 0, !dbg !33 + %4460 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i184 = icmp eq i32 %4460, 0, !dbg !33 + br i1 %.not.i181, label %4466, label %4461, !dbg !33 + +4461: ; preds = %__nv_sqrtf.exit180 + br i1 %.not1.i184, label %4464, label %4462, !dbg !33 + +4462: ; preds = %4461 + %4463 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3988) #7, !dbg !33 + br label %__nv_sqrtf.exit185, !dbg !33 + +4464: ; preds = %4461 + %4465 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3988) #7, !dbg !33 + br label %__nv_sqrtf.exit185, !dbg !33 + +4466: ; preds = %__nv_sqrtf.exit180 + br i1 %.not1.i184, label %4469, label %4467, !dbg !33 + +4467: ; preds = %4466 + %4468 = tail call float @llvm.nvvm.sqrt.rn.f(float %3988) #7, !dbg !33 + br label %__nv_sqrtf.exit185, !dbg !33 + +4469: ; preds = %4466 + %4470 = tail call float @llvm.nvvm.sqrt.approx.f(float %3988) #7, !dbg !33 + br label %__nv_sqrtf.exit185, !dbg !33 + +__nv_sqrtf.exit185: ; preds = %4462, %4464, %4467, %4469 + %.0.i183 = phi float [ %4463, %4462 ], [ %4465, %4464 ], [ %4468, %4467 ], [ %4470, %4469 ], !dbg !33 + %4471 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i186 = icmp eq i32 %4471, 0, !dbg !33 + %4472 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i189 = icmp eq i32 %4472, 0, !dbg !33 + br i1 %.not.i186, label %4478, label %4473, !dbg !33 + +4473: ; preds = %__nv_sqrtf.exit185 + br i1 %.not1.i189, label %4476, label %4474, !dbg !33 + +4474: ; preds = %4473 + %4475 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3989) #7, !dbg !33 + br label %__nv_sqrtf.exit190, !dbg !33 + +4476: ; preds = %4473 + %4477 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3989) #7, !dbg !33 + br label %__nv_sqrtf.exit190, !dbg !33 + +4478: ; preds = %__nv_sqrtf.exit185 + br i1 %.not1.i189, label %4481, label %4479, !dbg !33 + +4479: ; preds = %4478 + %4480 = tail call float @llvm.nvvm.sqrt.rn.f(float %3989) #7, !dbg !33 + br label %__nv_sqrtf.exit190, !dbg !33 + +4481: ; preds = %4478 + %4482 = tail call float @llvm.nvvm.sqrt.approx.f(float %3989) #7, !dbg !33 + br label %__nv_sqrtf.exit190, !dbg !33 + +__nv_sqrtf.exit190: ; preds = %4474, %4476, %4479, %4481 + %.0.i188 = phi float [ %4475, %4474 ], [ %4477, %4476 ], [ %4480, %4479 ], [ %4482, %4481 ], !dbg !33 + %4483 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i191 = icmp eq i32 %4483, 0, !dbg !33 + %4484 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i194 = icmp eq i32 %4484, 0, !dbg !33 + br i1 %.not.i191, label %4490, label %4485, !dbg !33 + +4485: ; preds = %__nv_sqrtf.exit190 + br i1 %.not1.i194, label %4488, label %4486, !dbg !33 + +4486: ; preds = %4485 + %4487 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3990) #7, !dbg !33 + br label %__nv_sqrtf.exit195, !dbg !33 + +4488: ; preds = %4485 + %4489 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3990) #7, !dbg !33 + br label %__nv_sqrtf.exit195, !dbg !33 + +4490: ; preds = %__nv_sqrtf.exit190 + br i1 %.not1.i194, label %4493, label %4491, !dbg !33 + +4491: ; preds = %4490 + %4492 = tail call float @llvm.nvvm.sqrt.rn.f(float %3990) #7, !dbg !33 + br label %__nv_sqrtf.exit195, !dbg !33 + +4493: ; preds = %4490 + %4494 = tail call float @llvm.nvvm.sqrt.approx.f(float %3990) #7, !dbg !33 + br label %__nv_sqrtf.exit195, !dbg !33 + +__nv_sqrtf.exit195: ; preds = %4486, %4488, %4491, %4493 + %.0.i193 = phi float [ %4487, %4486 ], [ %4489, %4488 ], [ %4492, %4491 ], [ %4494, %4493 ], !dbg !33 + %4495 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i196 = icmp eq i32 %4495, 0, !dbg !33 + %4496 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i199 = icmp eq i32 %4496, 0, !dbg !33 + br i1 %.not.i196, label %4502, label %4497, !dbg !33 + +4497: ; preds = %__nv_sqrtf.exit195 + br i1 %.not1.i199, label %4500, label %4498, !dbg !33 + +4498: ; preds = %4497 + %4499 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3991) #7, !dbg !33 + br label %__nv_sqrtf.exit200, !dbg !33 + +4500: ; preds = %4497 + %4501 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3991) #7, !dbg !33 + br label %__nv_sqrtf.exit200, !dbg !33 + +4502: ; preds = %__nv_sqrtf.exit195 + br i1 %.not1.i199, label %4505, label %4503, !dbg !33 + +4503: ; preds = %4502 + %4504 = tail call float @llvm.nvvm.sqrt.rn.f(float %3991) #7, !dbg !33 + br label %__nv_sqrtf.exit200, !dbg !33 + +4505: ; preds = %4502 + %4506 = tail call float @llvm.nvvm.sqrt.approx.f(float %3991) #7, !dbg !33 + br label %__nv_sqrtf.exit200, !dbg !33 + +__nv_sqrtf.exit200: ; preds = %4498, %4500, %4503, %4505 + %.0.i198 = phi float [ %4499, %4498 ], [ %4501, %4500 ], [ %4504, %4503 ], [ %4506, %4505 ], !dbg !33 + %4507 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i201 = icmp eq i32 %4507, 0, !dbg !33 + %4508 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i204 = icmp eq i32 %4508, 0, !dbg !33 + br i1 %.not.i201, label %4514, label %4509, !dbg !33 + +4509: ; preds = %__nv_sqrtf.exit200 + br i1 %.not1.i204, label %4512, label %4510, !dbg !33 + +4510: ; preds = %4509 + %4511 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3992) #7, !dbg !33 + br label %__nv_sqrtf.exit205, !dbg !33 + +4512: ; preds = %4509 + %4513 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3992) #7, !dbg !33 + br label %__nv_sqrtf.exit205, !dbg !33 + +4514: ; preds = %__nv_sqrtf.exit200 + br i1 %.not1.i204, label %4517, label %4515, !dbg !33 + +4515: ; preds = %4514 + %4516 = tail call float @llvm.nvvm.sqrt.rn.f(float %3992) #7, !dbg !33 + br label %__nv_sqrtf.exit205, !dbg !33 + +4517: ; preds = %4514 + %4518 = tail call float @llvm.nvvm.sqrt.approx.f(float %3992) #7, !dbg !33 + br label %__nv_sqrtf.exit205, !dbg !33 + +__nv_sqrtf.exit205: ; preds = %4510, %4512, %4515, %4517 + %.0.i203 = phi float [ %4511, %4510 ], [ %4513, %4512 ], [ %4516, %4515 ], [ %4518, %4517 ], !dbg !33 + %4519 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i206 = icmp eq i32 %4519, 0, !dbg !33 + %4520 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i209 = icmp eq i32 %4520, 0, !dbg !33 + br i1 %.not.i206, label %4526, label %4521, !dbg !33 + +4521: ; preds = %__nv_sqrtf.exit205 + br i1 %.not1.i209, label %4524, label %4522, !dbg !33 + +4522: ; preds = %4521 + %4523 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3993) #7, !dbg !33 + br label %__nv_sqrtf.exit210, !dbg !33 + +4524: ; preds = %4521 + %4525 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3993) #7, !dbg !33 + br label %__nv_sqrtf.exit210, !dbg !33 + +4526: ; preds = %__nv_sqrtf.exit205 + br i1 %.not1.i209, label %4529, label %4527, !dbg !33 + +4527: ; preds = %4526 + %4528 = tail call float @llvm.nvvm.sqrt.rn.f(float %3993) #7, !dbg !33 + br label %__nv_sqrtf.exit210, !dbg !33 + +4529: ; preds = %4526 + %4530 = tail call float @llvm.nvvm.sqrt.approx.f(float %3993) #7, !dbg !33 + br label %__nv_sqrtf.exit210, !dbg !33 + +__nv_sqrtf.exit210: ; preds = %4522, %4524, %4527, %4529 + %.0.i208 = phi float [ %4523, %4522 ], [ %4525, %4524 ], [ %4528, %4527 ], [ %4530, %4529 ], !dbg !33 + %4531 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i211 = icmp eq i32 %4531, 0, !dbg !33 + %4532 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i214 = icmp eq i32 %4532, 0, !dbg !33 + br i1 %.not.i211, label %4538, label %4533, !dbg !33 + +4533: ; preds = %__nv_sqrtf.exit210 + br i1 %.not1.i214, label %4536, label %4534, !dbg !33 + +4534: ; preds = %4533 + %4535 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3994) #7, !dbg !33 + br label %__nv_sqrtf.exit215, !dbg !33 + +4536: ; preds = %4533 + %4537 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3994) #7, !dbg !33 + br label %__nv_sqrtf.exit215, !dbg !33 + +4538: ; preds = %__nv_sqrtf.exit210 + br i1 %.not1.i214, label %4541, label %4539, !dbg !33 + +4539: ; preds = %4538 + %4540 = tail call float @llvm.nvvm.sqrt.rn.f(float %3994) #7, !dbg !33 + br label %__nv_sqrtf.exit215, !dbg !33 + +4541: ; preds = %4538 + %4542 = tail call float @llvm.nvvm.sqrt.approx.f(float %3994) #7, !dbg !33 + br label %__nv_sqrtf.exit215, !dbg !33 + +__nv_sqrtf.exit215: ; preds = %4534, %4536, %4539, %4541 + %.0.i213 = phi float [ %4535, %4534 ], [ %4537, %4536 ], [ %4540, %4539 ], [ %4542, %4541 ], !dbg !33 + %4543 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i216 = icmp eq i32 %4543, 0, !dbg !33 + %4544 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i219 = icmp eq i32 %4544, 0, !dbg !33 + br i1 %.not.i216, label %4550, label %4545, !dbg !33 + +4545: ; preds = %__nv_sqrtf.exit215 + br i1 %.not1.i219, label %4548, label %4546, !dbg !33 + +4546: ; preds = %4545 + %4547 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3995) #7, !dbg !33 + br label %__nv_sqrtf.exit220, !dbg !33 + +4548: ; preds = %4545 + %4549 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3995) #7, !dbg !33 + br label %__nv_sqrtf.exit220, !dbg !33 + +4550: ; preds = %__nv_sqrtf.exit215 + br i1 %.not1.i219, label %4553, label %4551, !dbg !33 + +4551: ; preds = %4550 + %4552 = tail call float @llvm.nvvm.sqrt.rn.f(float %3995) #7, !dbg !33 + br label %__nv_sqrtf.exit220, !dbg !33 + +4553: ; preds = %4550 + %4554 = tail call float @llvm.nvvm.sqrt.approx.f(float %3995) #7, !dbg !33 + br label %__nv_sqrtf.exit220, !dbg !33 + +__nv_sqrtf.exit220: ; preds = %4546, %4548, %4551, %4553 + %.0.i218 = phi float [ %4547, %4546 ], [ %4549, %4548 ], [ %4552, %4551 ], [ %4554, %4553 ], !dbg !33 + %4555 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i221 = icmp eq i32 %4555, 0, !dbg !33 + %4556 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i224 = icmp eq i32 %4556, 0, !dbg !33 + br i1 %.not.i221, label %4562, label %4557, !dbg !33 + +4557: ; preds = %__nv_sqrtf.exit220 + br i1 %.not1.i224, label %4560, label %4558, !dbg !33 + +4558: ; preds = %4557 + %4559 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3996) #7, !dbg !33 + br label %__nv_sqrtf.exit225, !dbg !33 + +4560: ; preds = %4557 + %4561 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3996) #7, !dbg !33 + br label %__nv_sqrtf.exit225, !dbg !33 + +4562: ; preds = %__nv_sqrtf.exit220 + br i1 %.not1.i224, label %4565, label %4563, !dbg !33 + +4563: ; preds = %4562 + %4564 = tail call float @llvm.nvvm.sqrt.rn.f(float %3996) #7, !dbg !33 + br label %__nv_sqrtf.exit225, !dbg !33 + +4565: ; preds = %4562 + %4566 = tail call float @llvm.nvvm.sqrt.approx.f(float %3996) #7, !dbg !33 + br label %__nv_sqrtf.exit225, !dbg !33 + +__nv_sqrtf.exit225: ; preds = %4558, %4560, %4563, %4565 + %.0.i223 = phi float [ %4559, %4558 ], [ %4561, %4560 ], [ %4564, %4563 ], [ %4566, %4565 ], !dbg !33 + %4567 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i226 = icmp eq i32 %4567, 0, !dbg !33 + %4568 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i229 = icmp eq i32 %4568, 0, !dbg !33 + br i1 %.not.i226, label %4574, label %4569, !dbg !33 + +4569: ; preds = %__nv_sqrtf.exit225 + br i1 %.not1.i229, label %4572, label %4570, !dbg !33 + +4570: ; preds = %4569 + %4571 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3997) #7, !dbg !33 + br label %__nv_sqrtf.exit230, !dbg !33 + +4572: ; preds = %4569 + %4573 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3997) #7, !dbg !33 + br label %__nv_sqrtf.exit230, !dbg !33 + +4574: ; preds = %__nv_sqrtf.exit225 + br i1 %.not1.i229, label %4577, label %4575, !dbg !33 + +4575: ; preds = %4574 + %4576 = tail call float @llvm.nvvm.sqrt.rn.f(float %3997) #7, !dbg !33 + br label %__nv_sqrtf.exit230, !dbg !33 + +4577: ; preds = %4574 + %4578 = tail call float @llvm.nvvm.sqrt.approx.f(float %3997) #7, !dbg !33 + br label %__nv_sqrtf.exit230, !dbg !33 + +__nv_sqrtf.exit230: ; preds = %4570, %4572, %4575, %4577 + %.0.i228 = phi float [ %4571, %4570 ], [ %4573, %4572 ], [ %4576, %4575 ], [ %4578, %4577 ], !dbg !33 + %4579 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i231 = icmp eq i32 %4579, 0, !dbg !33 + %4580 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i234 = icmp eq i32 %4580, 0, !dbg !33 + br i1 %.not.i231, label %4586, label %4581, !dbg !33 + +4581: ; preds = %__nv_sqrtf.exit230 + br i1 %.not1.i234, label %4584, label %4582, !dbg !33 + +4582: ; preds = %4581 + %4583 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3998) #7, !dbg !33 + br label %__nv_sqrtf.exit235, !dbg !33 + +4584: ; preds = %4581 + %4585 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3998) #7, !dbg !33 + br label %__nv_sqrtf.exit235, !dbg !33 + +4586: ; preds = %__nv_sqrtf.exit230 + br i1 %.not1.i234, label %4589, label %4587, !dbg !33 + +4587: ; preds = %4586 + %4588 = tail call float @llvm.nvvm.sqrt.rn.f(float %3998) #7, !dbg !33 + br label %__nv_sqrtf.exit235, !dbg !33 + +4589: ; preds = %4586 + %4590 = tail call float @llvm.nvvm.sqrt.approx.f(float %3998) #7, !dbg !33 + br label %__nv_sqrtf.exit235, !dbg !33 + +__nv_sqrtf.exit235: ; preds = %4582, %4584, %4587, %4589 + %.0.i233 = phi float [ %4583, %4582 ], [ %4585, %4584 ], [ %4588, %4587 ], [ %4590, %4589 ], !dbg !33 + %4591 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i236 = icmp eq i32 %4591, 0, !dbg !33 + %4592 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i239 = icmp eq i32 %4592, 0, !dbg !33 + br i1 %.not.i236, label %4598, label %4593, !dbg !33 + +4593: ; preds = %__nv_sqrtf.exit235 + br i1 %.not1.i239, label %4596, label %4594, !dbg !33 + +4594: ; preds = %4593 + %4595 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %3999) #7, !dbg !33 + br label %__nv_sqrtf.exit240, !dbg !33 + +4596: ; preds = %4593 + %4597 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %3999) #7, !dbg !33 + br label %__nv_sqrtf.exit240, !dbg !33 + +4598: ; preds = %__nv_sqrtf.exit235 + br i1 %.not1.i239, label %4601, label %4599, !dbg !33 + +4599: ; preds = %4598 + %4600 = tail call float @llvm.nvvm.sqrt.rn.f(float %3999) #7, !dbg !33 + br label %__nv_sqrtf.exit240, !dbg !33 + +4601: ; preds = %4598 + %4602 = tail call float @llvm.nvvm.sqrt.approx.f(float %3999) #7, !dbg !33 + br label %__nv_sqrtf.exit240, !dbg !33 + +__nv_sqrtf.exit240: ; preds = %4594, %4596, %4599, %4601 + %.0.i238 = phi float [ %4595, %4594 ], [ %4597, %4596 ], [ %4600, %4599 ], [ %4602, %4601 ], !dbg !33 + %4603 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i241 = icmp eq i32 %4603, 0, !dbg !33 + %4604 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i244 = icmp eq i32 %4604, 0, !dbg !33 + br i1 %.not.i241, label %4610, label %4605, !dbg !33 + +4605: ; preds = %__nv_sqrtf.exit240 + br i1 %.not1.i244, label %4608, label %4606, !dbg !33 + +4606: ; preds = %4605 + %4607 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %4000) #7, !dbg !33 + br label %__nv_sqrtf.exit245, !dbg !33 + +4608: ; preds = %4605 + %4609 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %4000) #7, !dbg !33 + br label %__nv_sqrtf.exit245, !dbg !33 + +4610: ; preds = %__nv_sqrtf.exit240 + br i1 %.not1.i244, label %4613, label %4611, !dbg !33 + +4611: ; preds = %4610 + %4612 = tail call float @llvm.nvvm.sqrt.rn.f(float %4000) #7, !dbg !33 + br label %__nv_sqrtf.exit245, !dbg !33 + +4613: ; preds = %4610 + %4614 = tail call float @llvm.nvvm.sqrt.approx.f(float %4000) #7, !dbg !33 + br label %__nv_sqrtf.exit245, !dbg !33 + +__nv_sqrtf.exit245: ; preds = %4606, %4608, %4611, %4613 + %.0.i243 = phi float [ %4607, %4606 ], [ %4609, %4608 ], [ %4612, %4611 ], [ %4614, %4613 ], !dbg !33 + %4615 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i246 = icmp eq i32 %4615, 0, !dbg !33 + %4616 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i249 = icmp eq i32 %4616, 0, !dbg !33 + br i1 %.not.i246, label %4622, label %4617, !dbg !33 + +4617: ; preds = %__nv_sqrtf.exit245 + br i1 %.not1.i249, label %4620, label %4618, !dbg !33 + +4618: ; preds = %4617 + %4619 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %4001) #7, !dbg !33 + br label %__nv_sqrtf.exit250, !dbg !33 + +4620: ; preds = %4617 + %4621 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %4001) #7, !dbg !33 + br label %__nv_sqrtf.exit250, !dbg !33 + +4622: ; preds = %__nv_sqrtf.exit245 + br i1 %.not1.i249, label %4625, label %4623, !dbg !33 + +4623: ; preds = %4622 + %4624 = tail call float @llvm.nvvm.sqrt.rn.f(float %4001) #7, !dbg !33 + br label %__nv_sqrtf.exit250, !dbg !33 + +4625: ; preds = %4622 + %4626 = tail call float @llvm.nvvm.sqrt.approx.f(float %4001) #7, !dbg !33 + br label %__nv_sqrtf.exit250, !dbg !33 + +__nv_sqrtf.exit250: ; preds = %4618, %4620, %4623, %4625 + %.0.i248 = phi float [ %4619, %4618 ], [ %4621, %4620 ], [ %4624, %4623 ], [ %4626, %4625 ], !dbg !33 + %4627 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i251 = icmp eq i32 %4627, 0, !dbg !33 + %4628 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i254 = icmp eq i32 %4628, 0, !dbg !33 + br i1 %.not.i251, label %4634, label %4629, !dbg !33 + +4629: ; preds = %__nv_sqrtf.exit250 + br i1 %.not1.i254, label %4632, label %4630, !dbg !33 + +4630: ; preds = %4629 + %4631 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %4002) #7, !dbg !33 + br label %__nv_sqrtf.exit255, !dbg !33 + +4632: ; preds = %4629 + %4633 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %4002) #7, !dbg !33 + br label %__nv_sqrtf.exit255, !dbg !33 + +4634: ; preds = %__nv_sqrtf.exit250 + br i1 %.not1.i254, label %4637, label %4635, !dbg !33 + +4635: ; preds = %4634 + %4636 = tail call float @llvm.nvvm.sqrt.rn.f(float %4002) #7, !dbg !33 + br label %__nv_sqrtf.exit255, !dbg !33 + +4637: ; preds = %4634 + %4638 = tail call float @llvm.nvvm.sqrt.approx.f(float %4002) #7, !dbg !33 + br label %__nv_sqrtf.exit255, !dbg !33 + +__nv_sqrtf.exit255: ; preds = %4630, %4632, %4635, %4637 + %.0.i253 = phi float [ %4631, %4630 ], [ %4633, %4632 ], [ %4636, %4635 ], [ %4638, %4637 ], !dbg !33 + %4639 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i256 = icmp eq i32 %4639, 0, !dbg !33 + %4640 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i259 = icmp eq i32 %4640, 0, !dbg !33 + br i1 %.not.i256, label %4646, label %4641, !dbg !33 + +4641: ; preds = %__nv_sqrtf.exit255 + br i1 %.not1.i259, label %4644, label %4642, !dbg !33 + +4642: ; preds = %4641 + %4643 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %4003) #7, !dbg !33 + br label %__nv_sqrtf.exit260, !dbg !33 + +4644: ; preds = %4641 + %4645 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %4003) #7, !dbg !33 + br label %__nv_sqrtf.exit260, !dbg !33 + +4646: ; preds = %__nv_sqrtf.exit255 + br i1 %.not1.i259, label %4649, label %4647, !dbg !33 + +4647: ; preds = %4646 + %4648 = tail call float @llvm.nvvm.sqrt.rn.f(float %4003) #7, !dbg !33 + br label %__nv_sqrtf.exit260, !dbg !33 + +4649: ; preds = %4646 + %4650 = tail call float @llvm.nvvm.sqrt.approx.f(float %4003) #7, !dbg !33 + br label %__nv_sqrtf.exit260, !dbg !33 + +__nv_sqrtf.exit260: ; preds = %4642, %4644, %4647, %4649 + %.0.i258 = phi float [ %4643, %4642 ], [ %4645, %4644 ], [ %4648, %4647 ], [ %4650, %4649 ], !dbg !33 + %4651 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i261 = icmp eq i32 %4651, 0, !dbg !33 + %4652 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i264 = icmp eq i32 %4652, 0, !dbg !33 + br i1 %.not.i261, label %4658, label %4653, !dbg !33 + +4653: ; preds = %__nv_sqrtf.exit260 + br i1 %.not1.i264, label %4656, label %4654, !dbg !33 + +4654: ; preds = %4653 + %4655 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %4004) #7, !dbg !33 + br label %__nv_sqrtf.exit265, !dbg !33 + +4656: ; preds = %4653 + %4657 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %4004) #7, !dbg !33 + br label %__nv_sqrtf.exit265, !dbg !33 + +4658: ; preds = %__nv_sqrtf.exit260 + br i1 %.not1.i264, label %4661, label %4659, !dbg !33 + +4659: ; preds = %4658 + %4660 = tail call float @llvm.nvvm.sqrt.rn.f(float %4004) #7, !dbg !33 + br label %__nv_sqrtf.exit265, !dbg !33 + +4661: ; preds = %4658 + %4662 = tail call float @llvm.nvvm.sqrt.approx.f(float %4004) #7, !dbg !33 + br label %__nv_sqrtf.exit265, !dbg !33 + +__nv_sqrtf.exit265: ; preds = %4654, %4656, %4659, %4661 + %.0.i263 = phi float [ %4655, %4654 ], [ %4657, %4656 ], [ %4660, %4659 ], [ %4662, %4661 ], !dbg !33 + %4663 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i266 = icmp eq i32 %4663, 0, !dbg !33 + %4664 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i269 = icmp eq i32 %4664, 0, !dbg !33 + br i1 %.not.i266, label %4670, label %4665, !dbg !33 + +4665: ; preds = %__nv_sqrtf.exit265 + br i1 %.not1.i269, label %4668, label %4666, !dbg !33 + +4666: ; preds = %4665 + %4667 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %4005) #7, !dbg !33 + br label %__nv_sqrtf.exit270, !dbg !33 + +4668: ; preds = %4665 + %4669 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %4005) #7, !dbg !33 + br label %__nv_sqrtf.exit270, !dbg !33 + +4670: ; preds = %__nv_sqrtf.exit265 + br i1 %.not1.i269, label %4673, label %4671, !dbg !33 + +4671: ; preds = %4670 + %4672 = tail call float @llvm.nvvm.sqrt.rn.f(float %4005) #7, !dbg !33 + br label %__nv_sqrtf.exit270, !dbg !33 + +4673: ; preds = %4670 + %4674 = tail call float @llvm.nvvm.sqrt.approx.f(float %4005) #7, !dbg !33 + br label %__nv_sqrtf.exit270, !dbg !33 + +__nv_sqrtf.exit270: ; preds = %4666, %4668, %4671, %4673 + %.0.i268 = phi float [ %4667, %4666 ], [ %4669, %4668 ], [ %4672, %4671 ], [ %4674, %4673 ], !dbg !33 + %4675 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i271 = icmp eq i32 %4675, 0, !dbg !33 + %4676 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i274 = icmp eq i32 %4676, 0, !dbg !33 + br i1 %.not.i271, label %4682, label %4677, !dbg !33 + +4677: ; preds = %__nv_sqrtf.exit270 + br i1 %.not1.i274, label %4680, label %4678, !dbg !33 + +4678: ; preds = %4677 + %4679 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %4006) #7, !dbg !33 + br label %__nv_sqrtf.exit275, !dbg !33 + +4680: ; preds = %4677 + %4681 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %4006) #7, !dbg !33 + br label %__nv_sqrtf.exit275, !dbg !33 + +4682: ; preds = %__nv_sqrtf.exit270 + br i1 %.not1.i274, label %4685, label %4683, !dbg !33 + +4683: ; preds = %4682 + %4684 = tail call float @llvm.nvvm.sqrt.rn.f(float %4006) #7, !dbg !33 + br label %__nv_sqrtf.exit275, !dbg !33 + +4685: ; preds = %4682 + %4686 = tail call float @llvm.nvvm.sqrt.approx.f(float %4006) #7, !dbg !33 + br label %__nv_sqrtf.exit275, !dbg !33 + +__nv_sqrtf.exit275: ; preds = %4678, %4680, %4683, %4685 + %.0.i273 = phi float [ %4679, %4678 ], [ %4681, %4680 ], [ %4684, %4683 ], [ %4686, %4685 ], !dbg !33 + %4687 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i276 = icmp eq i32 %4687, 0, !dbg !33 + %4688 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i279 = icmp eq i32 %4688, 0, !dbg !33 + br i1 %.not.i276, label %4694, label %4689, !dbg !33 + +4689: ; preds = %__nv_sqrtf.exit275 + br i1 %.not1.i279, label %4692, label %4690, !dbg !33 + +4690: ; preds = %4689 + %4691 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %4007) #7, !dbg !33 + br label %__nv_sqrtf.exit280, !dbg !33 + +4692: ; preds = %4689 + %4693 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %4007) #7, !dbg !33 + br label %__nv_sqrtf.exit280, !dbg !33 + +4694: ; preds = %__nv_sqrtf.exit275 + br i1 %.not1.i279, label %4697, label %4695, !dbg !33 + +4695: ; preds = %4694 + %4696 = tail call float @llvm.nvvm.sqrt.rn.f(float %4007) #7, !dbg !33 + br label %__nv_sqrtf.exit280, !dbg !33 + +4697: ; preds = %4694 + %4698 = tail call float @llvm.nvvm.sqrt.approx.f(float %4007) #7, !dbg !33 + br label %__nv_sqrtf.exit280, !dbg !33 + +__nv_sqrtf.exit280: ; preds = %4690, %4692, %4695, %4697 + %.0.i278 = phi float [ %4691, %4690 ], [ %4693, %4692 ], [ %4696, %4695 ], [ %4698, %4697 ], !dbg !33 + %4699 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i281 = icmp eq i32 %4699, 0, !dbg !33 + %4700 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i284 = icmp eq i32 %4700, 0, !dbg !33 + br i1 %.not.i281, label %4706, label %4701, !dbg !33 + +4701: ; preds = %__nv_sqrtf.exit280 + br i1 %.not1.i284, label %4704, label %4702, !dbg !33 + +4702: ; preds = %4701 + %4703 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %4008) #7, !dbg !33 + br label %__nv_sqrtf.exit285, !dbg !33 + +4704: ; preds = %4701 + %4705 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %4008) #7, !dbg !33 + br label %__nv_sqrtf.exit285, !dbg !33 + +4706: ; preds = %__nv_sqrtf.exit280 + br i1 %.not1.i284, label %4709, label %4707, !dbg !33 + +4707: ; preds = %4706 + %4708 = tail call float @llvm.nvvm.sqrt.rn.f(float %4008) #7, !dbg !33 + br label %__nv_sqrtf.exit285, !dbg !33 + +4709: ; preds = %4706 + %4710 = tail call float @llvm.nvvm.sqrt.approx.f(float %4008) #7, !dbg !33 + br label %__nv_sqrtf.exit285, !dbg !33 + +__nv_sqrtf.exit285: ; preds = %4702, %4704, %4707, %4709 + %.0.i283 = phi float [ %4703, %4702 ], [ %4705, %4704 ], [ %4708, %4707 ], [ %4710, %4709 ], !dbg !33 + %4711 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i286 = icmp eq i32 %4711, 0, !dbg !33 + %4712 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i289 = icmp eq i32 %4712, 0, !dbg !33 + br i1 %.not.i286, label %4718, label %4713, !dbg !33 + +4713: ; preds = %__nv_sqrtf.exit285 + br i1 %.not1.i289, label %4716, label %4714, !dbg !33 + +4714: ; preds = %4713 + %4715 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %4009) #7, !dbg !33 + br label %__nv_sqrtf.exit290, !dbg !33 + +4716: ; preds = %4713 + %4717 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %4009) #7, !dbg !33 + br label %__nv_sqrtf.exit290, !dbg !33 + +4718: ; preds = %__nv_sqrtf.exit285 + br i1 %.not1.i289, label %4721, label %4719, !dbg !33 + +4719: ; preds = %4718 + %4720 = tail call float @llvm.nvvm.sqrt.rn.f(float %4009) #7, !dbg !33 + br label %__nv_sqrtf.exit290, !dbg !33 + +4721: ; preds = %4718 + %4722 = tail call float @llvm.nvvm.sqrt.approx.f(float %4009) #7, !dbg !33 + br label %__nv_sqrtf.exit290, !dbg !33 + +__nv_sqrtf.exit290: ; preds = %4714, %4716, %4719, %4721 + %.0.i288 = phi float [ %4715, %4714 ], [ %4717, %4716 ], [ %4720, %4719 ], [ %4722, %4721 ], !dbg !33 + %4723 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i291 = icmp eq i32 %4723, 0, !dbg !33 + %4724 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i294 = icmp eq i32 %4724, 0, !dbg !33 + br i1 %.not.i291, label %4730, label %4725, !dbg !33 + +4725: ; preds = %__nv_sqrtf.exit290 + br i1 %.not1.i294, label %4728, label %4726, !dbg !33 + +4726: ; preds = %4725 + %4727 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %4010) #7, !dbg !33 + br label %__nv_sqrtf.exit295, !dbg !33 + +4728: ; preds = %4725 + %4729 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %4010) #7, !dbg !33 + br label %__nv_sqrtf.exit295, !dbg !33 + +4730: ; preds = %__nv_sqrtf.exit290 + br i1 %.not1.i294, label %4733, label %4731, !dbg !33 + +4731: ; preds = %4730 + %4732 = tail call float @llvm.nvvm.sqrt.rn.f(float %4010) #7, !dbg !33 + br label %__nv_sqrtf.exit295, !dbg !33 + +4733: ; preds = %4730 + %4734 = tail call float @llvm.nvvm.sqrt.approx.f(float %4010) #7, !dbg !33 + br label %__nv_sqrtf.exit295, !dbg !33 + +__nv_sqrtf.exit295: ; preds = %4726, %4728, %4731, %4733 + %.0.i293 = phi float [ %4727, %4726 ], [ %4729, %4728 ], [ %4732, %4731 ], [ %4734, %4733 ], !dbg !33 + %4735 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i296 = icmp eq i32 %4735, 0, !dbg !33 + %4736 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i299 = icmp eq i32 %4736, 0, !dbg !33 + br i1 %.not.i296, label %4742, label %4737, !dbg !33 + +4737: ; preds = %__nv_sqrtf.exit295 + br i1 %.not1.i299, label %4740, label %4738, !dbg !33 + +4738: ; preds = %4737 + %4739 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %4011) #7, !dbg !33 + br label %__nv_sqrtf.exit300, !dbg !33 + +4740: ; preds = %4737 + %4741 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %4011) #7, !dbg !33 + br label %__nv_sqrtf.exit300, !dbg !33 + +4742: ; preds = %__nv_sqrtf.exit295 + br i1 %.not1.i299, label %4745, label %4743, !dbg !33 + +4743: ; preds = %4742 + %4744 = tail call float @llvm.nvvm.sqrt.rn.f(float %4011) #7, !dbg !33 + br label %__nv_sqrtf.exit300, !dbg !33 + +4745: ; preds = %4742 + %4746 = tail call float @llvm.nvvm.sqrt.approx.f(float %4011) #7, !dbg !33 + br label %__nv_sqrtf.exit300, !dbg !33 + +__nv_sqrtf.exit300: ; preds = %4738, %4740, %4743, %4745 + %.0.i298 = phi float [ %4739, %4738 ], [ %4741, %4740 ], [ %4744, %4743 ], [ %4746, %4745 ], !dbg !33 + %4747 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i301 = icmp eq i32 %4747, 0, !dbg !33 + %4748 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i304 = icmp eq i32 %4748, 0, !dbg !33 + br i1 %.not.i301, label %4754, label %4749, !dbg !33 + +4749: ; preds = %__nv_sqrtf.exit300 + br i1 %.not1.i304, label %4752, label %4750, !dbg !33 + +4750: ; preds = %4749 + %4751 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %4012) #7, !dbg !33 + br label %__nv_sqrtf.exit305, !dbg !33 + +4752: ; preds = %4749 + %4753 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %4012) #7, !dbg !33 + br label %__nv_sqrtf.exit305, !dbg !33 + +4754: ; preds = %__nv_sqrtf.exit300 + br i1 %.not1.i304, label %4757, label %4755, !dbg !33 + +4755: ; preds = %4754 + %4756 = tail call float @llvm.nvvm.sqrt.rn.f(float %4012) #7, !dbg !33 + br label %__nv_sqrtf.exit305, !dbg !33 + +4757: ; preds = %4754 + %4758 = tail call float @llvm.nvvm.sqrt.approx.f(float %4012) #7, !dbg !33 + br label %__nv_sqrtf.exit305, !dbg !33 + +__nv_sqrtf.exit305: ; preds = %4750, %4752, %4755, %4757 + %.0.i303 = phi float [ %4751, %4750 ], [ %4753, %4752 ], [ %4756, %4755 ], [ %4758, %4757 ], !dbg !33 + %4759 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i306 = icmp eq i32 %4759, 0, !dbg !33 + %4760 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i309 = icmp eq i32 %4760, 0, !dbg !33 + br i1 %.not.i306, label %4766, label %4761, !dbg !33 + +4761: ; preds = %__nv_sqrtf.exit305 + br i1 %.not1.i309, label %4764, label %4762, !dbg !33 + +4762: ; preds = %4761 + %4763 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %4013) #7, !dbg !33 + br label %__nv_sqrtf.exit310, !dbg !33 + +4764: ; preds = %4761 + %4765 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %4013) #7, !dbg !33 + br label %__nv_sqrtf.exit310, !dbg !33 + +4766: ; preds = %__nv_sqrtf.exit305 + br i1 %.not1.i309, label %4769, label %4767, !dbg !33 + +4767: ; preds = %4766 + %4768 = tail call float @llvm.nvvm.sqrt.rn.f(float %4013) #7, !dbg !33 + br label %__nv_sqrtf.exit310, !dbg !33 + +4769: ; preds = %4766 + %4770 = tail call float @llvm.nvvm.sqrt.approx.f(float %4013) #7, !dbg !33 + br label %__nv_sqrtf.exit310, !dbg !33 + +__nv_sqrtf.exit310: ; preds = %4762, %4764, %4767, %4769 + %.0.i308 = phi float [ %4763, %4762 ], [ %4765, %4764 ], [ %4768, %4767 ], [ %4770, %4769 ], !dbg !33 + %4771 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i311 = icmp eq i32 %4771, 0, !dbg !33 + %4772 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i314 = icmp eq i32 %4772, 0, !dbg !33 + br i1 %.not.i311, label %4778, label %4773, !dbg !33 + +4773: ; preds = %__nv_sqrtf.exit310 + br i1 %.not1.i314, label %4776, label %4774, !dbg !33 + +4774: ; preds = %4773 + %4775 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %4014) #7, !dbg !33 + br label %__nv_sqrtf.exit315, !dbg !33 + +4776: ; preds = %4773 + %4777 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %4014) #7, !dbg !33 + br label %__nv_sqrtf.exit315, !dbg !33 + +4778: ; preds = %__nv_sqrtf.exit310 + br i1 %.not1.i314, label %4781, label %4779, !dbg !33 + +4779: ; preds = %4778 + %4780 = tail call float @llvm.nvvm.sqrt.rn.f(float %4014) #7, !dbg !33 + br label %__nv_sqrtf.exit315, !dbg !33 + +4781: ; preds = %4778 + %4782 = tail call float @llvm.nvvm.sqrt.approx.f(float %4014) #7, !dbg !33 + br label %__nv_sqrtf.exit315, !dbg !33 + +__nv_sqrtf.exit315: ; preds = %4774, %4776, %4779, %4781 + %.0.i313 = phi float [ %4775, %4774 ], [ %4777, %4776 ], [ %4780, %4779 ], [ %4782, %4781 ], !dbg !33 + %4783 = sext i32 %4 to i64, !dbg !18 + %4784 = shl nuw nsw i32 %11, 2, !dbg !16 + %4785 = and i32 %4784, 124, !dbg !16 + %4786 = or disjoint i32 %4785, 3, !dbg !16 + %4787 = zext nneg i32 %4786 to i64, !dbg !17 + %4788 = or disjoint i64 %10, %4787, !dbg !17 + %4789 = getelementptr float, ptr addrspace(1) %2, i64 %4788, !dbg !34 + %4790 = or disjoint i32 %4785, 2, !dbg !16 + %4791 = zext nneg i32 %4790 to i64, !dbg !17 + %4792 = or disjoint i64 %10, %4791, !dbg !17 + %4793 = getelementptr float, ptr addrspace(1) %2, i64 %4792, !dbg !34 + %4794 = or disjoint i32 %4785, 1, !dbg !16 + %4795 = zext nneg i32 %4794 to i64, !dbg !17 + %4796 = or disjoint i64 %10, %4795, !dbg !17 + %4797 = getelementptr float, ptr addrspace(1) %2, i64 %4796, !dbg !34 + %4798 = zext nneg i32 %4785 to i64, !dbg !17 + %4799 = or disjoint i64 %10, %4798, !dbg !17 + %4800 = getelementptr float, ptr addrspace(1) %2, i64 %4799, !dbg !34 + %4801 = getelementptr bfloat, ptr addrspace(1) %1, i64 %173, !dbg !35 + %4802 = getelementptr bfloat, ptr addrspace(1) %4801, i64 %238, !dbg !36 + %4803 = getelementptr bfloat, ptr addrspace(1) %1, i64 %171, !dbg !35 + %4804 = getelementptr bfloat, ptr addrspace(1) %4803, i64 %238, !dbg !36 + %4805 = getelementptr bfloat, ptr addrspace(1) %1, i64 %169, !dbg !35 + %4806 = getelementptr bfloat, ptr addrspace(1) %4805, i64 %238, !dbg !36 + %4807 = getelementptr bfloat, ptr addrspace(1) %1, i64 %167, !dbg !35 + %4808 = getelementptr bfloat, ptr addrspace(1) %4807, i64 %238, !dbg !36 + %4809 = getelementptr bfloat, ptr addrspace(1) %1, i64 %165, !dbg !35 + %4810 = getelementptr bfloat, ptr addrspace(1) %4809, i64 %238, !dbg !36 + %4811 = getelementptr bfloat, ptr addrspace(1) %1, i64 %163, !dbg !35 + %4812 = getelementptr bfloat, ptr addrspace(1) %4811, i64 %238, !dbg !36 + %4813 = getelementptr bfloat, ptr addrspace(1) %1, i64 %161, !dbg !35 + %4814 = getelementptr bfloat, ptr addrspace(1) %4813, i64 %238, !dbg !36 + %4815 = getelementptr bfloat, ptr addrspace(1) %1, i64 %159, !dbg !35 + %4816 = getelementptr bfloat, ptr addrspace(1) %4815, i64 %238, !dbg !36 + %4817 = getelementptr bfloat, ptr addrspace(1) %1, i64 %157, !dbg !35 + %4818 = getelementptr bfloat, ptr addrspace(1) %4817, i64 %238, !dbg !36 + %4819 = getelementptr bfloat, ptr addrspace(1) %1, i64 %155, !dbg !35 + %4820 = getelementptr bfloat, ptr addrspace(1) %4819, i64 %238, !dbg !36 + %4821 = getelementptr bfloat, ptr addrspace(1) %1, i64 %153, !dbg !35 + %4822 = getelementptr bfloat, ptr addrspace(1) %4821, i64 %238, !dbg !36 + %4823 = getelementptr bfloat, ptr addrspace(1) %1, i64 %151, !dbg !35 + %4824 = getelementptr bfloat, ptr addrspace(1) %4823, i64 %238, !dbg !36 + %4825 = getelementptr bfloat, ptr addrspace(1) %1, i64 %149, !dbg !35 + %4826 = getelementptr bfloat, ptr addrspace(1) %4825, i64 %238, !dbg !36 + %4827 = getelementptr bfloat, ptr addrspace(1) %1, i64 %147, !dbg !35 + %4828 = getelementptr bfloat, ptr addrspace(1) %4827, i64 %238, !dbg !36 + %4829 = getelementptr bfloat, ptr addrspace(1) %1, i64 %145, !dbg !35 + %4830 = getelementptr bfloat, ptr addrspace(1) %4829, i64 %238, !dbg !36 + %4831 = getelementptr bfloat, ptr addrspace(1) %1, i64 %143, !dbg !35 + %4832 = getelementptr bfloat, ptr addrspace(1) %4831, i64 %238, !dbg !36 + %4833 = getelementptr bfloat, ptr addrspace(1) %1, i64 %141, !dbg !35 + %4834 = getelementptr bfloat, ptr addrspace(1) %4833, i64 %238, !dbg !36 + %4835 = getelementptr bfloat, ptr addrspace(1) %1, i64 %139, !dbg !35 + %4836 = getelementptr bfloat, ptr addrspace(1) %4835, i64 %238, !dbg !36 + %4837 = getelementptr bfloat, ptr addrspace(1) %1, i64 %137, !dbg !35 + %4838 = getelementptr bfloat, ptr addrspace(1) %4837, i64 %238, !dbg !36 + %4839 = getelementptr bfloat, ptr addrspace(1) %1, i64 %135, !dbg !35 + %4840 = getelementptr bfloat, ptr addrspace(1) %4839, i64 %238, !dbg !36 + %4841 = getelementptr bfloat, ptr addrspace(1) %1, i64 %133, !dbg !35 + %4842 = getelementptr bfloat, ptr addrspace(1) %4841, i64 %238, !dbg !36 + %4843 = getelementptr bfloat, ptr addrspace(1) %1, i64 %131, !dbg !35 + %4844 = getelementptr bfloat, ptr addrspace(1) %4843, i64 %238, !dbg !36 + %4845 = getelementptr bfloat, ptr addrspace(1) %1, i64 %129, !dbg !35 + %4846 = getelementptr bfloat, ptr addrspace(1) %4845, i64 %238, !dbg !36 + %4847 = getelementptr bfloat, ptr addrspace(1) %1, i64 %127, !dbg !35 + %4848 = getelementptr bfloat, ptr addrspace(1) %4847, i64 %238, !dbg !36 + %4849 = getelementptr bfloat, ptr addrspace(1) %1, i64 %125, !dbg !35 + %4850 = getelementptr bfloat, ptr addrspace(1) %4849, i64 %238, !dbg !36 + %4851 = getelementptr bfloat, ptr addrspace(1) %1, i64 %123, !dbg !35 + %4852 = getelementptr bfloat, ptr addrspace(1) %4851, i64 %238, !dbg !36 + %4853 = getelementptr bfloat, ptr addrspace(1) %1, i64 %121, !dbg !35 + %4854 = getelementptr bfloat, ptr addrspace(1) %4853, i64 %238, !dbg !36 + %4855 = getelementptr bfloat, ptr addrspace(1) %1, i64 %119, !dbg !35 + %4856 = getelementptr bfloat, ptr addrspace(1) %4855, i64 %238, !dbg !36 + %4857 = getelementptr bfloat, ptr addrspace(1) %1, i64 %117, !dbg !35 + %4858 = getelementptr bfloat, ptr addrspace(1) %4857, i64 %238, !dbg !36 + %4859 = getelementptr bfloat, ptr addrspace(1) %1, i64 %115, !dbg !35 + %4860 = getelementptr bfloat, ptr addrspace(1) %4859, i64 %238, !dbg !36 + %4861 = getelementptr bfloat, ptr addrspace(1) %1, i64 %113, !dbg !35 + %4862 = getelementptr bfloat, ptr addrspace(1) %4861, i64 %238, !dbg !36 + %4863 = getelementptr bfloat, ptr addrspace(1) %1, i64 %111, !dbg !35 + %4864 = getelementptr bfloat, ptr addrspace(1) %4863, i64 %238, !dbg !36 + %4865 = getelementptr bfloat, ptr addrspace(1) %1, i64 %109, !dbg !35 + %4866 = getelementptr bfloat, ptr addrspace(1) %4865, i64 %238, !dbg !36 + %4867 = getelementptr bfloat, ptr addrspace(1) %1, i64 %107, !dbg !35 + %4868 = getelementptr bfloat, ptr addrspace(1) %4867, i64 %238, !dbg !36 + %4869 = getelementptr bfloat, ptr addrspace(1) %1, i64 %105, !dbg !35 + %4870 = getelementptr bfloat, ptr addrspace(1) %4869, i64 %238, !dbg !36 + %4871 = getelementptr bfloat, ptr addrspace(1) %1, i64 %103, !dbg !35 + %4872 = getelementptr bfloat, ptr addrspace(1) %4871, i64 %238, !dbg !36 + %4873 = getelementptr bfloat, ptr addrspace(1) %1, i64 %101, !dbg !35 + %4874 = getelementptr bfloat, ptr addrspace(1) %4873, i64 %238, !dbg !36 + %4875 = getelementptr bfloat, ptr addrspace(1) %1, i64 %99, !dbg !35 + %4876 = getelementptr bfloat, ptr addrspace(1) %4875, i64 %238, !dbg !36 + %4877 = getelementptr bfloat, ptr addrspace(1) %1, i64 %97, !dbg !35 + %4878 = getelementptr bfloat, ptr addrspace(1) %4877, i64 %238, !dbg !36 + %4879 = getelementptr bfloat, ptr addrspace(1) %1, i64 %95, !dbg !35 + %4880 = getelementptr bfloat, ptr addrspace(1) %4879, i64 %238, !dbg !36 + %4881 = getelementptr bfloat, ptr addrspace(1) %1, i64 %93, !dbg !35 + %4882 = getelementptr bfloat, ptr addrspace(1) %4881, i64 %238, !dbg !36 + %4883 = getelementptr bfloat, ptr addrspace(1) %1, i64 %91, !dbg !35 + %4884 = getelementptr bfloat, ptr addrspace(1) %4883, i64 %238, !dbg !36 + %4885 = getelementptr bfloat, ptr addrspace(1) %1, i64 %89, !dbg !35 + %4886 = getelementptr bfloat, ptr addrspace(1) %4885, i64 %238, !dbg !36 + %4887 = getelementptr bfloat, ptr addrspace(1) %1, i64 %87, !dbg !35 + %4888 = getelementptr bfloat, ptr addrspace(1) %4887, i64 %238, !dbg !36 + %4889 = getelementptr bfloat, ptr addrspace(1) %1, i64 %85, !dbg !35 + %4890 = getelementptr bfloat, ptr addrspace(1) %4889, i64 %238, !dbg !36 + %4891 = getelementptr bfloat, ptr addrspace(1) %1, i64 %83, !dbg !35 + %4892 = getelementptr bfloat, ptr addrspace(1) %4891, i64 %238, !dbg !36 + %4893 = getelementptr bfloat, ptr addrspace(1) %1, i64 %81, !dbg !35 + %4894 = getelementptr bfloat, ptr addrspace(1) %4893, i64 %238, !dbg !36 + %4895 = getelementptr bfloat, ptr addrspace(1) %1, i64 %79, !dbg !35 + %4896 = getelementptr bfloat, ptr addrspace(1) %4895, i64 %238, !dbg !36 + %4897 = getelementptr bfloat, ptr addrspace(1) %1, i64 %77, !dbg !35 + %4898 = getelementptr bfloat, ptr addrspace(1) %4897, i64 %238, !dbg !36 + %4899 = getelementptr bfloat, ptr addrspace(1) %1, i64 %75, !dbg !35 + %4900 = getelementptr bfloat, ptr addrspace(1) %4899, i64 %238, !dbg !36 + %4901 = getelementptr bfloat, ptr addrspace(1) %1, i64 %73, !dbg !35 + %4902 = getelementptr bfloat, ptr addrspace(1) %4901, i64 %238, !dbg !36 + %4903 = getelementptr bfloat, ptr addrspace(1) %1, i64 %71, !dbg !35 + %4904 = getelementptr bfloat, ptr addrspace(1) %4903, i64 %238, !dbg !36 + %4905 = getelementptr bfloat, ptr addrspace(1) %1, i64 %69, !dbg !35 + %4906 = getelementptr bfloat, ptr addrspace(1) %4905, i64 %238, !dbg !36 + %4907 = getelementptr bfloat, ptr addrspace(1) %1, i64 %67, !dbg !35 + %4908 = getelementptr bfloat, ptr addrspace(1) %4907, i64 %238, !dbg !36 + %4909 = getelementptr bfloat, ptr addrspace(1) %1, i64 %65, !dbg !35 + %4910 = getelementptr bfloat, ptr addrspace(1) %4909, i64 %238, !dbg !36 + %4911 = getelementptr bfloat, ptr addrspace(1) %1, i64 %63, !dbg !35 + %4912 = getelementptr bfloat, ptr addrspace(1) %4911, i64 %238, !dbg !36 + %4913 = getelementptr bfloat, ptr addrspace(1) %1, i64 %61, !dbg !35 + %4914 = getelementptr bfloat, ptr addrspace(1) %4913, i64 %238, !dbg !36 + %4915 = getelementptr bfloat, ptr addrspace(1) %1, i64 %59, !dbg !35 + %4916 = getelementptr bfloat, ptr addrspace(1) %4915, i64 %238, !dbg !36 + %4917 = getelementptr bfloat, ptr addrspace(1) %1, i64 %57, !dbg !35 + %4918 = getelementptr bfloat, ptr addrspace(1) %4917, i64 %238, !dbg !36 + %4919 = getelementptr bfloat, ptr addrspace(1) %1, i64 %55, !dbg !35 + %4920 = getelementptr bfloat, ptr addrspace(1) %4919, i64 %238, !dbg !36 + %4921 = getelementptr bfloat, ptr addrspace(1) %1, i64 %53, !dbg !35 + %4922 = getelementptr bfloat, ptr addrspace(1) %4921, i64 %238, !dbg !36 + %4923 = getelementptr bfloat, ptr addrspace(1) %1, i64 %51, !dbg !35 + %4924 = getelementptr bfloat, ptr addrspace(1) %4923, i64 %238, !dbg !36 + %4925 = getelementptr bfloat, ptr addrspace(1) %1, i64 %49, !dbg !35 + %4926 = getelementptr bfloat, ptr addrspace(1) %4925, i64 %238, !dbg !36 + %4927 = getelementptr bfloat, ptr addrspace(1) %1, i64 %47, !dbg !35 + %4928 = getelementptr bfloat, ptr addrspace(1) %4927, i64 %238, !dbg !36 + %4929 = icmp slt i64 %4788, %4783, !dbg !18 + %4930 = icmp slt i64 %4792, %4783, !dbg !18 + %4931 = icmp slt i64 %4796, %4783, !dbg !18 + %4932 = icmp slt i64 %4799, %4783, !dbg !18 + %4933 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i), !dbg !37 + %4934 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i3), !dbg !37 + %4935 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i8), !dbg !37 + %4936 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i13), !dbg !37 + %4937 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i18), !dbg !37 + %4938 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i23), !dbg !37 + %4939 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i28), !dbg !37 + %4940 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i33), !dbg !37 + %4941 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i38), !dbg !37 + %4942 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i43), !dbg !37 + %4943 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i48), !dbg !37 + %4944 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i53), !dbg !37 + %4945 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i58), !dbg !37 + %4946 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i63), !dbg !37 + %4947 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i68), !dbg !37 + %4948 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i73), !dbg !37 + %4949 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i78), !dbg !37 + %4950 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i83), !dbg !37 + %4951 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i88), !dbg !37 + %4952 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i93), !dbg !37 + %4953 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i98), !dbg !37 + %4954 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i103), !dbg !37 + %4955 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i108), !dbg !37 + %4956 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i113), !dbg !37 + %4957 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i118), !dbg !37 + %4958 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i123), !dbg !37 + %4959 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i128), !dbg !37 + %4960 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i133), !dbg !37 + %4961 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i138), !dbg !37 + %4962 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i143), !dbg !37 + %4963 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i148), !dbg !37 + %4964 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i153), !dbg !37 + %4965 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i158), !dbg !37 + %4966 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i163), !dbg !37 + %4967 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i168), !dbg !37 + %4968 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i173), !dbg !37 + %4969 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i178), !dbg !37 + %4970 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i183), !dbg !37 + %4971 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i188), !dbg !37 + %4972 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i193), !dbg !37 + %4973 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i198), !dbg !37 + %4974 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i203), !dbg !37 + %4975 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i208), !dbg !37 + %4976 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i213), !dbg !37 + %4977 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i218), !dbg !37 + %4978 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i223), !dbg !37 + %4979 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i228), !dbg !37 + %4980 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i233), !dbg !37 + %4981 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i238), !dbg !37 + %4982 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i243), !dbg !37 + %4983 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i248), !dbg !37 + %4984 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i253), !dbg !37 + %4985 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i258), !dbg !37 + %4986 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i263), !dbg !37 + %4987 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i268), !dbg !37 + %4988 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i273), !dbg !37 + %4989 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i278), !dbg !37 + %4990 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i283), !dbg !37 + %4991 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i288), !dbg !37 + %4992 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i293), !dbg !37 + %4993 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i298), !dbg !37 + %4994 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i303), !dbg !37 + %4995 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i308), !dbg !37 + %4996 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i313), !dbg !37 + %4997 = fmul float %4933, %1455, !dbg !38 + %4998 = fmul float %4933, %1456, !dbg !38 + %4999 = fmul float %4933, %1457, !dbg !38 + %5000 = fmul float %4933, %1458, !dbg !38 + %5001 = fmul float %4933, %1459, !dbg !38 + %5002 = fmul float %4933, %1460, !dbg !38 + %5003 = fmul float %4933, %1461, !dbg !38 + %5004 = fmul float %4933, %1462, !dbg !38 + %5005 = fmul float %4934, %1463, !dbg !38 + %5006 = fmul float %4934, %1464, !dbg !38 + %5007 = fmul float %4934, %1465, !dbg !38 + %5008 = fmul float %4934, %1466, !dbg !38 + %5009 = fmul float %4934, %1467, !dbg !38 + %5010 = fmul float %4934, %1468, !dbg !38 + %5011 = fmul float %4934, %1469, !dbg !38 + %5012 = fmul float %4934, %1470, !dbg !38 + %5013 = fmul float %4935, %1471, !dbg !38 + %5014 = fmul float %4935, %1472, !dbg !38 + %5015 = fmul float %4935, %1473, !dbg !38 + %5016 = fmul float %4935, %1474, !dbg !38 + %5017 = fmul float %4935, %1475, !dbg !38 + %5018 = fmul float %4935, %1476, !dbg !38 + %5019 = fmul float %4935, %1477, !dbg !38 + %5020 = fmul float %4935, %1478, !dbg !38 + %5021 = fmul float %4936, %1479, !dbg !38 + %5022 = fmul float %4936, %1480, !dbg !38 + %5023 = fmul float %4936, %1481, !dbg !38 + %5024 = fmul float %4936, %1482, !dbg !38 + %5025 = fmul float %4936, %1483, !dbg !38 + %5026 = fmul float %4936, %1484, !dbg !38 + %5027 = fmul float %4936, %1485, !dbg !38 + %5028 = fmul float %4936, %1486, !dbg !38 + %5029 = fmul float %4937, %1487, !dbg !38 + %5030 = fmul float %4937, %1488, !dbg !38 + %5031 = fmul float %4937, %1489, !dbg !38 + %5032 = fmul float %4937, %1490, !dbg !38 + %5033 = fmul float %4937, %1491, !dbg !38 + %5034 = fmul float %4937, %1492, !dbg !38 + %5035 = fmul float %4937, %1493, !dbg !38 + %5036 = fmul float %4937, %1494, !dbg !38 + %5037 = fmul float %4938, %1495, !dbg !38 + %5038 = fmul float %4938, %1496, !dbg !38 + %5039 = fmul float %4938, %1497, !dbg !38 + %5040 = fmul float %4938, %1498, !dbg !38 + %5041 = fmul float %4938, %1499, !dbg !38 + %5042 = fmul float %4938, %1500, !dbg !38 + %5043 = fmul float %4938, %1501, !dbg !38 + %5044 = fmul float %4938, %1502, !dbg !38 + %5045 = fmul float %4939, %1503, !dbg !38 + %5046 = fmul float %4939, %1504, !dbg !38 + %5047 = fmul float %4939, %1505, !dbg !38 + %5048 = fmul float %4939, %1506, !dbg !38 + %5049 = fmul float %4939, %1507, !dbg !38 + %5050 = fmul float %4939, %1508, !dbg !38 + %5051 = fmul float %4939, %1509, !dbg !38 + %5052 = fmul float %4939, %1510, !dbg !38 + %5053 = fmul float %4940, %1511, !dbg !38 + %5054 = fmul float %4940, %1512, !dbg !38 + %5055 = fmul float %4940, %1513, !dbg !38 + %5056 = fmul float %4940, %1514, !dbg !38 + %5057 = fmul float %4940, %1515, !dbg !38 + %5058 = fmul float %4940, %1516, !dbg !38 + %5059 = fmul float %4940, %1517, !dbg !38 + %5060 = fmul float %4940, %1518, !dbg !38 + %5061 = fmul float %4941, %1519, !dbg !38 + %5062 = fmul float %4941, %1520, !dbg !38 + %5063 = fmul float %4941, %1521, !dbg !38 + %5064 = fmul float %4941, %1522, !dbg !38 + %5065 = fmul float %4941, %1523, !dbg !38 + %5066 = fmul float %4941, %1524, !dbg !38 + %5067 = fmul float %4941, %1525, !dbg !38 + %5068 = fmul float %4941, %1526, !dbg !38 + %5069 = fmul float %4942, %1527, !dbg !38 + %5070 = fmul float %4942, %1528, !dbg !38 + %5071 = fmul float %4942, %1529, !dbg !38 + %5072 = fmul float %4942, %1530, !dbg !38 + %5073 = fmul float %4942, %1531, !dbg !38 + %5074 = fmul float %4942, %1532, !dbg !38 + %5075 = fmul float %4942, %1533, !dbg !38 + %5076 = fmul float %4942, %1534, !dbg !38 + %5077 = fmul float %4943, %1535, !dbg !38 + %5078 = fmul float %4943, %1536, !dbg !38 + %5079 = fmul float %4943, %1537, !dbg !38 + %5080 = fmul float %4943, %1538, !dbg !38 + %5081 = fmul float %4943, %1539, !dbg !38 + %5082 = fmul float %4943, %1540, !dbg !38 + %5083 = fmul float %4943, %1541, !dbg !38 + %5084 = fmul float %4943, %1542, !dbg !38 + %5085 = fmul float %4944, %1543, !dbg !38 + %5086 = fmul float %4944, %1544, !dbg !38 + %5087 = fmul float %4944, %1545, !dbg !38 + %5088 = fmul float %4944, %1546, !dbg !38 + %5089 = fmul float %4944, %1547, !dbg !38 + %5090 = fmul float %4944, %1548, !dbg !38 + %5091 = fmul float %4944, %1549, !dbg !38 + %5092 = fmul float %4944, %1550, !dbg !38 + %5093 = fmul float %4945, %1551, !dbg !38 + %5094 = fmul float %4945, %1552, !dbg !38 + %5095 = fmul float %4945, %1553, !dbg !38 + %5096 = fmul float %4945, %1554, !dbg !38 + %5097 = fmul float %4945, %1555, !dbg !38 + %5098 = fmul float %4945, %1556, !dbg !38 + %5099 = fmul float %4945, %1557, !dbg !38 + %5100 = fmul float %4945, %1558, !dbg !38 + %5101 = fmul float %4946, %1559, !dbg !38 + %5102 = fmul float %4946, %1560, !dbg !38 + %5103 = fmul float %4946, %1561, !dbg !38 + %5104 = fmul float %4946, %1562, !dbg !38 + %5105 = fmul float %4946, %1563, !dbg !38 + %5106 = fmul float %4946, %1564, !dbg !38 + %5107 = fmul float %4946, %1565, !dbg !38 + %5108 = fmul float %4946, %1566, !dbg !38 + %5109 = fmul float %4947, %1567, !dbg !38 + %5110 = fmul float %4947, %1568, !dbg !38 + %5111 = fmul float %4947, %1569, !dbg !38 + %5112 = fmul float %4947, %1570, !dbg !38 + %5113 = fmul float %4947, %1571, !dbg !38 + %5114 = fmul float %4947, %1572, !dbg !38 + %5115 = fmul float %4947, %1573, !dbg !38 + %5116 = fmul float %4947, %1574, !dbg !38 + %5117 = fmul float %4948, %1575, !dbg !38 + %5118 = fmul float %4948, %1576, !dbg !38 + %5119 = fmul float %4948, %1577, !dbg !38 + %5120 = fmul float %4948, %1578, !dbg !38 + %5121 = fmul float %4948, %1579, !dbg !38 + %5122 = fmul float %4948, %1580, !dbg !38 + %5123 = fmul float %4948, %1581, !dbg !38 + %5124 = fmul float %4948, %1582, !dbg !38 + %5125 = fmul float %4949, %1583, !dbg !38 + %5126 = fmul float %4949, %1584, !dbg !38 + %5127 = fmul float %4949, %1585, !dbg !38 + %5128 = fmul float %4949, %1586, !dbg !38 + %5129 = fmul float %4949, %1587, !dbg !38 + %5130 = fmul float %4949, %1588, !dbg !38 + %5131 = fmul float %4949, %1589, !dbg !38 + %5132 = fmul float %4949, %1590, !dbg !38 + %5133 = fmul float %4950, %1591, !dbg !38 + %5134 = fmul float %4950, %1592, !dbg !38 + %5135 = fmul float %4950, %1593, !dbg !38 + %5136 = fmul float %4950, %1594, !dbg !38 + %5137 = fmul float %4950, %1595, !dbg !38 + %5138 = fmul float %4950, %1596, !dbg !38 + %5139 = fmul float %4950, %1597, !dbg !38 + %5140 = fmul float %4950, %1598, !dbg !38 + %5141 = fmul float %4951, %1599, !dbg !38 + %5142 = fmul float %4951, %1600, !dbg !38 + %5143 = fmul float %4951, %1601, !dbg !38 + %5144 = fmul float %4951, %1602, !dbg !38 + %5145 = fmul float %4951, %1603, !dbg !38 + %5146 = fmul float %4951, %1604, !dbg !38 + %5147 = fmul float %4951, %1605, !dbg !38 + %5148 = fmul float %4951, %1606, !dbg !38 + %5149 = fmul float %4952, %1607, !dbg !38 + %5150 = fmul float %4952, %1608, !dbg !38 + %5151 = fmul float %4952, %1609, !dbg !38 + %5152 = fmul float %4952, %1610, !dbg !38 + %5153 = fmul float %4952, %1611, !dbg !38 + %5154 = fmul float %4952, %1612, !dbg !38 + %5155 = fmul float %4952, %1613, !dbg !38 + %5156 = fmul float %4952, %1614, !dbg !38 + %5157 = fmul float %4953, %1615, !dbg !38 + %5158 = fmul float %4953, %1616, !dbg !38 + %5159 = fmul float %4953, %1617, !dbg !38 + %5160 = fmul float %4953, %1618, !dbg !38 + %5161 = fmul float %4953, %1619, !dbg !38 + %5162 = fmul float %4953, %1620, !dbg !38 + %5163 = fmul float %4953, %1621, !dbg !38 + %5164 = fmul float %4953, %1622, !dbg !38 + %5165 = fmul float %4954, %1623, !dbg !38 + %5166 = fmul float %4954, %1624, !dbg !38 + %5167 = fmul float %4954, %1625, !dbg !38 + %5168 = fmul float %4954, %1626, !dbg !38 + %5169 = fmul float %4954, %1627, !dbg !38 + %5170 = fmul float %4954, %1628, !dbg !38 + %5171 = fmul float %4954, %1629, !dbg !38 + %5172 = fmul float %4954, %1630, !dbg !38 + %5173 = fmul float %4955, %1631, !dbg !38 + %5174 = fmul float %4955, %1632, !dbg !38 + %5175 = fmul float %4955, %1633, !dbg !38 + %5176 = fmul float %4955, %1634, !dbg !38 + %5177 = fmul float %4955, %1635, !dbg !38 + %5178 = fmul float %4955, %1636, !dbg !38 + %5179 = fmul float %4955, %1637, !dbg !38 + %5180 = fmul float %4955, %1638, !dbg !38 + %5181 = fmul float %4956, %1639, !dbg !38 + %5182 = fmul float %4956, %1640, !dbg !38 + %5183 = fmul float %4956, %1641, !dbg !38 + %5184 = fmul float %4956, %1642, !dbg !38 + %5185 = fmul float %4956, %1643, !dbg !38 + %5186 = fmul float %4956, %1644, !dbg !38 + %5187 = fmul float %4956, %1645, !dbg !38 + %5188 = fmul float %4956, %1646, !dbg !38 + %5189 = fmul float %4957, %1647, !dbg !38 + %5190 = fmul float %4957, %1648, !dbg !38 + %5191 = fmul float %4957, %1649, !dbg !38 + %5192 = fmul float %4957, %1650, !dbg !38 + %5193 = fmul float %4957, %1651, !dbg !38 + %5194 = fmul float %4957, %1652, !dbg !38 + %5195 = fmul float %4957, %1653, !dbg !38 + %5196 = fmul float %4957, %1654, !dbg !38 + %5197 = fmul float %4958, %1655, !dbg !38 + %5198 = fmul float %4958, %1656, !dbg !38 + %5199 = fmul float %4958, %1657, !dbg !38 + %5200 = fmul float %4958, %1658, !dbg !38 + %5201 = fmul float %4958, %1659, !dbg !38 + %5202 = fmul float %4958, %1660, !dbg !38 + %5203 = fmul float %4958, %1661, !dbg !38 + %5204 = fmul float %4958, %1662, !dbg !38 + %5205 = fmul float %4959, %1663, !dbg !38 + %5206 = fmul float %4959, %1664, !dbg !38 + %5207 = fmul float %4959, %1665, !dbg !38 + %5208 = fmul float %4959, %1666, !dbg !38 + %5209 = fmul float %4959, %1667, !dbg !38 + %5210 = fmul float %4959, %1668, !dbg !38 + %5211 = fmul float %4959, %1669, !dbg !38 + %5212 = fmul float %4959, %1670, !dbg !38 + %5213 = fmul float %4960, %1671, !dbg !38 + %5214 = fmul float %4960, %1672, !dbg !38 + %5215 = fmul float %4960, %1673, !dbg !38 + %5216 = fmul float %4960, %1674, !dbg !38 + %5217 = fmul float %4960, %1675, !dbg !38 + %5218 = fmul float %4960, %1676, !dbg !38 + %5219 = fmul float %4960, %1677, !dbg !38 + %5220 = fmul float %4960, %1678, !dbg !38 + %5221 = fmul float %4961, %1679, !dbg !38 + %5222 = fmul float %4961, %1680, !dbg !38 + %5223 = fmul float %4961, %1681, !dbg !38 + %5224 = fmul float %4961, %1682, !dbg !38 + %5225 = fmul float %4961, %1683, !dbg !38 + %5226 = fmul float %4961, %1684, !dbg !38 + %5227 = fmul float %4961, %1685, !dbg !38 + %5228 = fmul float %4961, %1686, !dbg !38 + %5229 = fmul float %4962, %1687, !dbg !38 + %5230 = fmul float %4962, %1688, !dbg !38 + %5231 = fmul float %4962, %1689, !dbg !38 + %5232 = fmul float %4962, %1690, !dbg !38 + %5233 = fmul float %4962, %1691, !dbg !38 + %5234 = fmul float %4962, %1692, !dbg !38 + %5235 = fmul float %4962, %1693, !dbg !38 + %5236 = fmul float %4962, %1694, !dbg !38 + %5237 = fmul float %4963, %1695, !dbg !38 + %5238 = fmul float %4963, %1696, !dbg !38 + %5239 = fmul float %4963, %1697, !dbg !38 + %5240 = fmul float %4963, %1698, !dbg !38 + %5241 = fmul float %4963, %1699, !dbg !38 + %5242 = fmul float %4963, %1700, !dbg !38 + %5243 = fmul float %4963, %1701, !dbg !38 + %5244 = fmul float %4963, %1702, !dbg !38 + %5245 = fmul float %4964, %1703, !dbg !38 + %5246 = fmul float %4964, %1704, !dbg !38 + %5247 = fmul float %4964, %1705, !dbg !38 + %5248 = fmul float %4964, %1706, !dbg !38 + %5249 = fmul float %4964, %1707, !dbg !38 + %5250 = fmul float %4964, %1708, !dbg !38 + %5251 = fmul float %4964, %1709, !dbg !38 + %5252 = fmul float %4964, %1710, !dbg !38 + %5253 = fmul float %4965, %1711, !dbg !38 + %5254 = fmul float %4965, %1712, !dbg !38 + %5255 = fmul float %4965, %1713, !dbg !38 + %5256 = fmul float %4965, %1714, !dbg !38 + %5257 = fmul float %4965, %1715, !dbg !38 + %5258 = fmul float %4965, %1716, !dbg !38 + %5259 = fmul float %4965, %1717, !dbg !38 + %5260 = fmul float %4965, %1718, !dbg !38 + %5261 = fmul float %4966, %1719, !dbg !38 + %5262 = fmul float %4966, %1720, !dbg !38 + %5263 = fmul float %4966, %1721, !dbg !38 + %5264 = fmul float %4966, %1722, !dbg !38 + %5265 = fmul float %4966, %1723, !dbg !38 + %5266 = fmul float %4966, %1724, !dbg !38 + %5267 = fmul float %4966, %1725, !dbg !38 + %5268 = fmul float %4966, %1726, !dbg !38 + %5269 = fmul float %4967, %1727, !dbg !38 + %5270 = fmul float %4967, %1728, !dbg !38 + %5271 = fmul float %4967, %1729, !dbg !38 + %5272 = fmul float %4967, %1730, !dbg !38 + %5273 = fmul float %4967, %1731, !dbg !38 + %5274 = fmul float %4967, %1732, !dbg !38 + %5275 = fmul float %4967, %1733, !dbg !38 + %5276 = fmul float %4967, %1734, !dbg !38 + %5277 = fmul float %4968, %1735, !dbg !38 + %5278 = fmul float %4968, %1736, !dbg !38 + %5279 = fmul float %4968, %1737, !dbg !38 + %5280 = fmul float %4968, %1738, !dbg !38 + %5281 = fmul float %4968, %1739, !dbg !38 + %5282 = fmul float %4968, %1740, !dbg !38 + %5283 = fmul float %4968, %1741, !dbg !38 + %5284 = fmul float %4968, %1742, !dbg !38 + %5285 = fmul float %4969, %1743, !dbg !38 + %5286 = fmul float %4969, %1744, !dbg !38 + %5287 = fmul float %4969, %1745, !dbg !38 + %5288 = fmul float %4969, %1746, !dbg !38 + %5289 = fmul float %4969, %1747, !dbg !38 + %5290 = fmul float %4969, %1748, !dbg !38 + %5291 = fmul float %4969, %1749, !dbg !38 + %5292 = fmul float %4969, %1750, !dbg !38 + %5293 = fmul float %4970, %1751, !dbg !38 + %5294 = fmul float %4970, %1752, !dbg !38 + %5295 = fmul float %4970, %1753, !dbg !38 + %5296 = fmul float %4970, %1754, !dbg !38 + %5297 = fmul float %4970, %1755, !dbg !38 + %5298 = fmul float %4970, %1756, !dbg !38 + %5299 = fmul float %4970, %1757, !dbg !38 + %5300 = fmul float %4970, %1758, !dbg !38 + %5301 = fmul float %4971, %1759, !dbg !38 + %5302 = fmul float %4971, %1760, !dbg !38 + %5303 = fmul float %4971, %1761, !dbg !38 + %5304 = fmul float %4971, %1762, !dbg !38 + %5305 = fmul float %4971, %1763, !dbg !38 + %5306 = fmul float %4971, %1764, !dbg !38 + %5307 = fmul float %4971, %1765, !dbg !38 + %5308 = fmul float %4971, %1766, !dbg !38 + %5309 = fmul float %4972, %1767, !dbg !38 + %5310 = fmul float %4972, %1768, !dbg !38 + %5311 = fmul float %4972, %1769, !dbg !38 + %5312 = fmul float %4972, %1770, !dbg !38 + %5313 = fmul float %4972, %1771, !dbg !38 + %5314 = fmul float %4972, %1772, !dbg !38 + %5315 = fmul float %4972, %1773, !dbg !38 + %5316 = fmul float %4972, %1774, !dbg !38 + %5317 = fmul float %4973, %1775, !dbg !38 + %5318 = fmul float %4973, %1776, !dbg !38 + %5319 = fmul float %4973, %1777, !dbg !38 + %5320 = fmul float %4973, %1778, !dbg !38 + %5321 = fmul float %4973, %1779, !dbg !38 + %5322 = fmul float %4973, %1780, !dbg !38 + %5323 = fmul float %4973, %1781, !dbg !38 + %5324 = fmul float %4973, %1782, !dbg !38 + %5325 = fmul float %4974, %1783, !dbg !38 + %5326 = fmul float %4974, %1784, !dbg !38 + %5327 = fmul float %4974, %1785, !dbg !38 + %5328 = fmul float %4974, %1786, !dbg !38 + %5329 = fmul float %4974, %1787, !dbg !38 + %5330 = fmul float %4974, %1788, !dbg !38 + %5331 = fmul float %4974, %1789, !dbg !38 + %5332 = fmul float %4974, %1790, !dbg !38 + %5333 = fmul float %4975, %1791, !dbg !38 + %5334 = fmul float %4975, %1792, !dbg !38 + %5335 = fmul float %4975, %1793, !dbg !38 + %5336 = fmul float %4975, %1794, !dbg !38 + %5337 = fmul float %4975, %1795, !dbg !38 + %5338 = fmul float %4975, %1796, !dbg !38 + %5339 = fmul float %4975, %1797, !dbg !38 + %5340 = fmul float %4975, %1798, !dbg !38 + %5341 = fmul float %4976, %1799, !dbg !38 + %5342 = fmul float %4976, %1800, !dbg !38 + %5343 = fmul float %4976, %1801, !dbg !38 + %5344 = fmul float %4976, %1802, !dbg !38 + %5345 = fmul float %4976, %1803, !dbg !38 + %5346 = fmul float %4976, %1804, !dbg !38 + %5347 = fmul float %4976, %1805, !dbg !38 + %5348 = fmul float %4976, %1806, !dbg !38 + %5349 = fmul float %4977, %1807, !dbg !38 + %5350 = fmul float %4977, %1808, !dbg !38 + %5351 = fmul float %4977, %1809, !dbg !38 + %5352 = fmul float %4977, %1810, !dbg !38 + %5353 = fmul float %4977, %1811, !dbg !38 + %5354 = fmul float %4977, %1812, !dbg !38 + %5355 = fmul float %4977, %1813, !dbg !38 + %5356 = fmul float %4977, %1814, !dbg !38 + %5357 = fmul float %4978, %1815, !dbg !38 + %5358 = fmul float %4978, %1816, !dbg !38 + %5359 = fmul float %4978, %1817, !dbg !38 + %5360 = fmul float %4978, %1818, !dbg !38 + %5361 = fmul float %4978, %1819, !dbg !38 + %5362 = fmul float %4978, %1820, !dbg !38 + %5363 = fmul float %4978, %1821, !dbg !38 + %5364 = fmul float %4978, %1822, !dbg !38 + %5365 = fmul float %4979, %1823, !dbg !38 + %5366 = fmul float %4979, %1824, !dbg !38 + %5367 = fmul float %4979, %1825, !dbg !38 + %5368 = fmul float %4979, %1826, !dbg !38 + %5369 = fmul float %4979, %1827, !dbg !38 + %5370 = fmul float %4979, %1828, !dbg !38 + %5371 = fmul float %4979, %1829, !dbg !38 + %5372 = fmul float %4979, %1830, !dbg !38 + %5373 = fmul float %4980, %1831, !dbg !38 + %5374 = fmul float %4980, %1832, !dbg !38 + %5375 = fmul float %4980, %1833, !dbg !38 + %5376 = fmul float %4980, %1834, !dbg !38 + %5377 = fmul float %4980, %1835, !dbg !38 + %5378 = fmul float %4980, %1836, !dbg !38 + %5379 = fmul float %4980, %1837, !dbg !38 + %5380 = fmul float %4980, %1838, !dbg !38 + %5381 = fmul float %4981, %1839, !dbg !38 + %5382 = fmul float %4981, %1840, !dbg !38 + %5383 = fmul float %4981, %1841, !dbg !38 + %5384 = fmul float %4981, %1842, !dbg !38 + %5385 = fmul float %4981, %1843, !dbg !38 + %5386 = fmul float %4981, %1844, !dbg !38 + %5387 = fmul float %4981, %1845, !dbg !38 + %5388 = fmul float %4981, %1846, !dbg !38 + %5389 = fmul float %4982, %1847, !dbg !38 + %5390 = fmul float %4982, %1848, !dbg !38 + %5391 = fmul float %4982, %1849, !dbg !38 + %5392 = fmul float %4982, %1850, !dbg !38 + %5393 = fmul float %4982, %1851, !dbg !38 + %5394 = fmul float %4982, %1852, !dbg !38 + %5395 = fmul float %4982, %1853, !dbg !38 + %5396 = fmul float %4982, %1854, !dbg !38 + %5397 = fmul float %4983, %1855, !dbg !38 + %5398 = fmul float %4983, %1856, !dbg !38 + %5399 = fmul float %4983, %1857, !dbg !38 + %5400 = fmul float %4983, %1858, !dbg !38 + %5401 = fmul float %4983, %1859, !dbg !38 + %5402 = fmul float %4983, %1860, !dbg !38 + %5403 = fmul float %4983, %1861, !dbg !38 + %5404 = fmul float %4983, %1862, !dbg !38 + %5405 = fmul float %4984, %1863, !dbg !38 + %5406 = fmul float %4984, %1864, !dbg !38 + %5407 = fmul float %4984, %1865, !dbg !38 + %5408 = fmul float %4984, %1866, !dbg !38 + %5409 = fmul float %4984, %1867, !dbg !38 + %5410 = fmul float %4984, %1868, !dbg !38 + %5411 = fmul float %4984, %1869, !dbg !38 + %5412 = fmul float %4984, %1870, !dbg !38 + %5413 = fmul float %4985, %1871, !dbg !38 + %5414 = fmul float %4985, %1872, !dbg !38 + %5415 = fmul float %4985, %1873, !dbg !38 + %5416 = fmul float %4985, %1874, !dbg !38 + %5417 = fmul float %4985, %1875, !dbg !38 + %5418 = fmul float %4985, %1876, !dbg !38 + %5419 = fmul float %4985, %1877, !dbg !38 + %5420 = fmul float %4985, %1878, !dbg !38 + %5421 = fmul float %4986, %1879, !dbg !38 + %5422 = fmul float %4986, %1880, !dbg !38 + %5423 = fmul float %4986, %1881, !dbg !38 + %5424 = fmul float %4986, %1882, !dbg !38 + %5425 = fmul float %4986, %1883, !dbg !38 + %5426 = fmul float %4986, %1884, !dbg !38 + %5427 = fmul float %4986, %1885, !dbg !38 + %5428 = fmul float %4986, %1886, !dbg !38 + %5429 = fmul float %4987, %1887, !dbg !38 + %5430 = fmul float %4987, %1888, !dbg !38 + %5431 = fmul float %4987, %1889, !dbg !38 + %5432 = fmul float %4987, %1890, !dbg !38 + %5433 = fmul float %4987, %1891, !dbg !38 + %5434 = fmul float %4987, %1892, !dbg !38 + %5435 = fmul float %4987, %1893, !dbg !38 + %5436 = fmul float %4987, %1894, !dbg !38 + %5437 = fmul float %4988, %1895, !dbg !38 + %5438 = fmul float %4988, %1896, !dbg !38 + %5439 = fmul float %4988, %1897, !dbg !38 + %5440 = fmul float %4988, %1898, !dbg !38 + %5441 = fmul float %4988, %1899, !dbg !38 + %5442 = fmul float %4988, %1900, !dbg !38 + %5443 = fmul float %4988, %1901, !dbg !38 + %5444 = fmul float %4988, %1902, !dbg !38 + %5445 = fmul float %4989, %1903, !dbg !38 + %5446 = fmul float %4989, %1904, !dbg !38 + %5447 = fmul float %4989, %1905, !dbg !38 + %5448 = fmul float %4989, %1906, !dbg !38 + %5449 = fmul float %4989, %1907, !dbg !38 + %5450 = fmul float %4989, %1908, !dbg !38 + %5451 = fmul float %4989, %1909, !dbg !38 + %5452 = fmul float %4989, %1910, !dbg !38 + %5453 = fmul float %4990, %1911, !dbg !38 + %5454 = fmul float %4990, %1912, !dbg !38 + %5455 = fmul float %4990, %1913, !dbg !38 + %5456 = fmul float %4990, %1914, !dbg !38 + %5457 = fmul float %4990, %1915, !dbg !38 + %5458 = fmul float %4990, %1916, !dbg !38 + %5459 = fmul float %4990, %1917, !dbg !38 + %5460 = fmul float %4990, %1918, !dbg !38 + %5461 = fmul float %4991, %1919, !dbg !38 + %5462 = fmul float %4991, %1920, !dbg !38 + %5463 = fmul float %4991, %1921, !dbg !38 + %5464 = fmul float %4991, %1922, !dbg !38 + %5465 = fmul float %4991, %1923, !dbg !38 + %5466 = fmul float %4991, %1924, !dbg !38 + %5467 = fmul float %4991, %1925, !dbg !38 + %5468 = fmul float %4991, %1926, !dbg !38 + %5469 = fmul float %4992, %1927, !dbg !38 + %5470 = fmul float %4992, %1928, !dbg !38 + %5471 = fmul float %4992, %1929, !dbg !38 + %5472 = fmul float %4992, %1930, !dbg !38 + %5473 = fmul float %4992, %1931, !dbg !38 + %5474 = fmul float %4992, %1932, !dbg !38 + %5475 = fmul float %4992, %1933, !dbg !38 + %5476 = fmul float %4992, %1934, !dbg !38 + %5477 = fmul float %4993, %1935, !dbg !38 + %5478 = fmul float %4993, %1936, !dbg !38 + %5479 = fmul float %4993, %1937, !dbg !38 + %5480 = fmul float %4993, %1938, !dbg !38 + %5481 = fmul float %4993, %1939, !dbg !38 + %5482 = fmul float %4993, %1940, !dbg !38 + %5483 = fmul float %4993, %1941, !dbg !38 + %5484 = fmul float %4993, %1942, !dbg !38 + %5485 = fmul float %4994, %1943, !dbg !38 + %5486 = fmul float %4994, %1944, !dbg !38 + %5487 = fmul float %4994, %1945, !dbg !38 + %5488 = fmul float %4994, %1946, !dbg !38 + %5489 = fmul float %4994, %1947, !dbg !38 + %5490 = fmul float %4994, %1948, !dbg !38 + %5491 = fmul float %4994, %1949, !dbg !38 + %5492 = fmul float %4994, %1950, !dbg !38 + %5493 = fmul float %4995, %1951, !dbg !38 + %5494 = fmul float %4995, %1952, !dbg !38 + %5495 = fmul float %4995, %1953, !dbg !38 + %5496 = fmul float %4995, %1954, !dbg !38 + %5497 = fmul float %4995, %1955, !dbg !38 + %5498 = fmul float %4995, %1956, !dbg !38 + %5499 = fmul float %4995, %1957, !dbg !38 + %5500 = fmul float %4995, %1958, !dbg !38 + %5501 = fmul float %4996, %1959, !dbg !38 + %5502 = fmul float %4996, %1960, !dbg !38 + %5503 = fmul float %4996, %1961, !dbg !38 + %5504 = fmul float %4996, %1962, !dbg !38 + %5505 = fmul float %4996, %1963, !dbg !38 + %5506 = fmul float %4996, %1964, !dbg !38 + %5507 = fmul float %4996, %1965, !dbg !38 + %5508 = fmul float %4996, %1966, !dbg !38 + %5509 = fptrunc float %4997 to bfloat, !dbg !39 + %5510 = fptrunc float %4998 to bfloat, !dbg !39 + %5511 = fptrunc float %4999 to bfloat, !dbg !39 + %5512 = fptrunc float %5000 to bfloat, !dbg !39 + %5513 = fptrunc float %5001 to bfloat, !dbg !39 + %5514 = fptrunc float %5002 to bfloat, !dbg !39 + %5515 = fptrunc float %5003 to bfloat, !dbg !39 + %5516 = fptrunc float %5004 to bfloat, !dbg !39 + %5517 = fptrunc float %5005 to bfloat, !dbg !39 + %5518 = fptrunc float %5006 to bfloat, !dbg !39 + %5519 = fptrunc float %5007 to bfloat, !dbg !39 + %5520 = fptrunc float %5008 to bfloat, !dbg !39 + %5521 = fptrunc float %5009 to bfloat, !dbg !39 + %5522 = fptrunc float %5010 to bfloat, !dbg !39 + %5523 = fptrunc float %5011 to bfloat, !dbg !39 + %5524 = fptrunc float %5012 to bfloat, !dbg !39 + %5525 = fptrunc float %5013 to bfloat, !dbg !39 + %5526 = fptrunc float %5014 to bfloat, !dbg !39 + %5527 = fptrunc float %5015 to bfloat, !dbg !39 + %5528 = fptrunc float %5016 to bfloat, !dbg !39 + %5529 = fptrunc float %5017 to bfloat, !dbg !39 + %5530 = fptrunc float %5018 to bfloat, !dbg !39 + %5531 = fptrunc float %5019 to bfloat, !dbg !39 + %5532 = fptrunc float %5020 to bfloat, !dbg !39 + %5533 = fptrunc float %5021 to bfloat, !dbg !39 + %5534 = fptrunc float %5022 to bfloat, !dbg !39 + %5535 = fptrunc float %5023 to bfloat, !dbg !39 + %5536 = fptrunc float %5024 to bfloat, !dbg !39 + %5537 = fptrunc float %5025 to bfloat, !dbg !39 + %5538 = fptrunc float %5026 to bfloat, !dbg !39 + %5539 = fptrunc float %5027 to bfloat, !dbg !39 + %5540 = fptrunc float %5028 to bfloat, !dbg !39 + %5541 = fptrunc float %5029 to bfloat, !dbg !39 + %5542 = fptrunc float %5030 to bfloat, !dbg !39 + %5543 = fptrunc float %5031 to bfloat, !dbg !39 + %5544 = fptrunc float %5032 to bfloat, !dbg !39 + %5545 = fptrunc float %5033 to bfloat, !dbg !39 + %5546 = fptrunc float %5034 to bfloat, !dbg !39 + %5547 = fptrunc float %5035 to bfloat, !dbg !39 + %5548 = fptrunc float %5036 to bfloat, !dbg !39 + %5549 = fptrunc float %5037 to bfloat, !dbg !39 + %5550 = fptrunc float %5038 to bfloat, !dbg !39 + %5551 = fptrunc float %5039 to bfloat, !dbg !39 + %5552 = fptrunc float %5040 to bfloat, !dbg !39 + %5553 = fptrunc float %5041 to bfloat, !dbg !39 + %5554 = fptrunc float %5042 to bfloat, !dbg !39 + %5555 = fptrunc float %5043 to bfloat, !dbg !39 + %5556 = fptrunc float %5044 to bfloat, !dbg !39 + %5557 = fptrunc float %5045 to bfloat, !dbg !39 + %5558 = fptrunc float %5046 to bfloat, !dbg !39 + %5559 = fptrunc float %5047 to bfloat, !dbg !39 + %5560 = fptrunc float %5048 to bfloat, !dbg !39 + %5561 = fptrunc float %5049 to bfloat, !dbg !39 + %5562 = fptrunc float %5050 to bfloat, !dbg !39 + %5563 = fptrunc float %5051 to bfloat, !dbg !39 + %5564 = fptrunc float %5052 to bfloat, !dbg !39 + %5565 = fptrunc float %5053 to bfloat, !dbg !39 + %5566 = fptrunc float %5054 to bfloat, !dbg !39 + %5567 = fptrunc float %5055 to bfloat, !dbg !39 + %5568 = fptrunc float %5056 to bfloat, !dbg !39 + %5569 = fptrunc float %5057 to bfloat, !dbg !39 + %5570 = fptrunc float %5058 to bfloat, !dbg !39 + %5571 = fptrunc float %5059 to bfloat, !dbg !39 + %5572 = fptrunc float %5060 to bfloat, !dbg !39 + %5573 = fptrunc float %5061 to bfloat, !dbg !39 + %5574 = fptrunc float %5062 to bfloat, !dbg !39 + %5575 = fptrunc float %5063 to bfloat, !dbg !39 + %5576 = fptrunc float %5064 to bfloat, !dbg !39 + %5577 = fptrunc float %5065 to bfloat, !dbg !39 + %5578 = fptrunc float %5066 to bfloat, !dbg !39 + %5579 = fptrunc float %5067 to bfloat, !dbg !39 + %5580 = fptrunc float %5068 to bfloat, !dbg !39 + %5581 = fptrunc float %5069 to bfloat, !dbg !39 + %5582 = fptrunc float %5070 to bfloat, !dbg !39 + %5583 = fptrunc float %5071 to bfloat, !dbg !39 + %5584 = fptrunc float %5072 to bfloat, !dbg !39 + %5585 = fptrunc float %5073 to bfloat, !dbg !39 + %5586 = fptrunc float %5074 to bfloat, !dbg !39 + %5587 = fptrunc float %5075 to bfloat, !dbg !39 + %5588 = fptrunc float %5076 to bfloat, !dbg !39 + %5589 = fptrunc float %5077 to bfloat, !dbg !39 + %5590 = fptrunc float %5078 to bfloat, !dbg !39 + %5591 = fptrunc float %5079 to bfloat, !dbg !39 + %5592 = fptrunc float %5080 to bfloat, !dbg !39 + %5593 = fptrunc float %5081 to bfloat, !dbg !39 + %5594 = fptrunc float %5082 to bfloat, !dbg !39 + %5595 = fptrunc float %5083 to bfloat, !dbg !39 + %5596 = fptrunc float %5084 to bfloat, !dbg !39 + %5597 = fptrunc float %5085 to bfloat, !dbg !39 + %5598 = fptrunc float %5086 to bfloat, !dbg !39 + %5599 = fptrunc float %5087 to bfloat, !dbg !39 + %5600 = fptrunc float %5088 to bfloat, !dbg !39 + %5601 = fptrunc float %5089 to bfloat, !dbg !39 + %5602 = fptrunc float %5090 to bfloat, !dbg !39 + %5603 = fptrunc float %5091 to bfloat, !dbg !39 + %5604 = fptrunc float %5092 to bfloat, !dbg !39 + %5605 = fptrunc float %5093 to bfloat, !dbg !39 + %5606 = fptrunc float %5094 to bfloat, !dbg !39 + %5607 = fptrunc float %5095 to bfloat, !dbg !39 + %5608 = fptrunc float %5096 to bfloat, !dbg !39 + %5609 = fptrunc float %5097 to bfloat, !dbg !39 + %5610 = fptrunc float %5098 to bfloat, !dbg !39 + %5611 = fptrunc float %5099 to bfloat, !dbg !39 + %5612 = fptrunc float %5100 to bfloat, !dbg !39 + %5613 = fptrunc float %5101 to bfloat, !dbg !39 + %5614 = fptrunc float %5102 to bfloat, !dbg !39 + %5615 = fptrunc float %5103 to bfloat, !dbg !39 + %5616 = fptrunc float %5104 to bfloat, !dbg !39 + %5617 = fptrunc float %5105 to bfloat, !dbg !39 + %5618 = fptrunc float %5106 to bfloat, !dbg !39 + %5619 = fptrunc float %5107 to bfloat, !dbg !39 + %5620 = fptrunc float %5108 to bfloat, !dbg !39 + %5621 = fptrunc float %5109 to bfloat, !dbg !39 + %5622 = fptrunc float %5110 to bfloat, !dbg !39 + %5623 = fptrunc float %5111 to bfloat, !dbg !39 + %5624 = fptrunc float %5112 to bfloat, !dbg !39 + %5625 = fptrunc float %5113 to bfloat, !dbg !39 + %5626 = fptrunc float %5114 to bfloat, !dbg !39 + %5627 = fptrunc float %5115 to bfloat, !dbg !39 + %5628 = fptrunc float %5116 to bfloat, !dbg !39 + %5629 = fptrunc float %5117 to bfloat, !dbg !39 + %5630 = fptrunc float %5118 to bfloat, !dbg !39 + %5631 = fptrunc float %5119 to bfloat, !dbg !39 + %5632 = fptrunc float %5120 to bfloat, !dbg !39 + %5633 = fptrunc float %5121 to bfloat, !dbg !39 + %5634 = fptrunc float %5122 to bfloat, !dbg !39 + %5635 = fptrunc float %5123 to bfloat, !dbg !39 + %5636 = fptrunc float %5124 to bfloat, !dbg !39 + %5637 = fptrunc float %5125 to bfloat, !dbg !39 + %5638 = fptrunc float %5126 to bfloat, !dbg !39 + %5639 = fptrunc float %5127 to bfloat, !dbg !39 + %5640 = fptrunc float %5128 to bfloat, !dbg !39 + %5641 = fptrunc float %5129 to bfloat, !dbg !39 + %5642 = fptrunc float %5130 to bfloat, !dbg !39 + %5643 = fptrunc float %5131 to bfloat, !dbg !39 + %5644 = fptrunc float %5132 to bfloat, !dbg !39 + %5645 = fptrunc float %5133 to bfloat, !dbg !39 + %5646 = fptrunc float %5134 to bfloat, !dbg !39 + %5647 = fptrunc float %5135 to bfloat, !dbg !39 + %5648 = fptrunc float %5136 to bfloat, !dbg !39 + %5649 = fptrunc float %5137 to bfloat, !dbg !39 + %5650 = fptrunc float %5138 to bfloat, !dbg !39 + %5651 = fptrunc float %5139 to bfloat, !dbg !39 + %5652 = fptrunc float %5140 to bfloat, !dbg !39 + %5653 = fptrunc float %5141 to bfloat, !dbg !39 + %5654 = fptrunc float %5142 to bfloat, !dbg !39 + %5655 = fptrunc float %5143 to bfloat, !dbg !39 + %5656 = fptrunc float %5144 to bfloat, !dbg !39 + %5657 = fptrunc float %5145 to bfloat, !dbg !39 + %5658 = fptrunc float %5146 to bfloat, !dbg !39 + %5659 = fptrunc float %5147 to bfloat, !dbg !39 + %5660 = fptrunc float %5148 to bfloat, !dbg !39 + %5661 = fptrunc float %5149 to bfloat, !dbg !39 + %5662 = fptrunc float %5150 to bfloat, !dbg !39 + %5663 = fptrunc float %5151 to bfloat, !dbg !39 + %5664 = fptrunc float %5152 to bfloat, !dbg !39 + %5665 = fptrunc float %5153 to bfloat, !dbg !39 + %5666 = fptrunc float %5154 to bfloat, !dbg !39 + %5667 = fptrunc float %5155 to bfloat, !dbg !39 + %5668 = fptrunc float %5156 to bfloat, !dbg !39 + %5669 = fptrunc float %5157 to bfloat, !dbg !39 + %5670 = fptrunc float %5158 to bfloat, !dbg !39 + %5671 = fptrunc float %5159 to bfloat, !dbg !39 + %5672 = fptrunc float %5160 to bfloat, !dbg !39 + %5673 = fptrunc float %5161 to bfloat, !dbg !39 + %5674 = fptrunc float %5162 to bfloat, !dbg !39 + %5675 = fptrunc float %5163 to bfloat, !dbg !39 + %5676 = fptrunc float %5164 to bfloat, !dbg !39 + %5677 = fptrunc float %5165 to bfloat, !dbg !39 + %5678 = fptrunc float %5166 to bfloat, !dbg !39 + %5679 = fptrunc float %5167 to bfloat, !dbg !39 + %5680 = fptrunc float %5168 to bfloat, !dbg !39 + %5681 = fptrunc float %5169 to bfloat, !dbg !39 + %5682 = fptrunc float %5170 to bfloat, !dbg !39 + %5683 = fptrunc float %5171 to bfloat, !dbg !39 + %5684 = fptrunc float %5172 to bfloat, !dbg !39 + %5685 = fptrunc float %5173 to bfloat, !dbg !39 + %5686 = fptrunc float %5174 to bfloat, !dbg !39 + %5687 = fptrunc float %5175 to bfloat, !dbg !39 + %5688 = fptrunc float %5176 to bfloat, !dbg !39 + %5689 = fptrunc float %5177 to bfloat, !dbg !39 + %5690 = fptrunc float %5178 to bfloat, !dbg !39 + %5691 = fptrunc float %5179 to bfloat, !dbg !39 + %5692 = fptrunc float %5180 to bfloat, !dbg !39 + %5693 = fptrunc float %5181 to bfloat, !dbg !39 + %5694 = fptrunc float %5182 to bfloat, !dbg !39 + %5695 = fptrunc float %5183 to bfloat, !dbg !39 + %5696 = fptrunc float %5184 to bfloat, !dbg !39 + %5697 = fptrunc float %5185 to bfloat, !dbg !39 + %5698 = fptrunc float %5186 to bfloat, !dbg !39 + %5699 = fptrunc float %5187 to bfloat, !dbg !39 + %5700 = fptrunc float %5188 to bfloat, !dbg !39 + %5701 = fptrunc float %5189 to bfloat, !dbg !39 + %5702 = fptrunc float %5190 to bfloat, !dbg !39 + %5703 = fptrunc float %5191 to bfloat, !dbg !39 + %5704 = fptrunc float %5192 to bfloat, !dbg !39 + %5705 = fptrunc float %5193 to bfloat, !dbg !39 + %5706 = fptrunc float %5194 to bfloat, !dbg !39 + %5707 = fptrunc float %5195 to bfloat, !dbg !39 + %5708 = fptrunc float %5196 to bfloat, !dbg !39 + %5709 = fptrunc float %5197 to bfloat, !dbg !39 + %5710 = fptrunc float %5198 to bfloat, !dbg !39 + %5711 = fptrunc float %5199 to bfloat, !dbg !39 + %5712 = fptrunc float %5200 to bfloat, !dbg !39 + %5713 = fptrunc float %5201 to bfloat, !dbg !39 + %5714 = fptrunc float %5202 to bfloat, !dbg !39 + %5715 = fptrunc float %5203 to bfloat, !dbg !39 + %5716 = fptrunc float %5204 to bfloat, !dbg !39 + %5717 = fptrunc float %5205 to bfloat, !dbg !39 + %5718 = fptrunc float %5206 to bfloat, !dbg !39 + %5719 = fptrunc float %5207 to bfloat, !dbg !39 + %5720 = fptrunc float %5208 to bfloat, !dbg !39 + %5721 = fptrunc float %5209 to bfloat, !dbg !39 + %5722 = fptrunc float %5210 to bfloat, !dbg !39 + %5723 = fptrunc float %5211 to bfloat, !dbg !39 + %5724 = fptrunc float %5212 to bfloat, !dbg !39 + %5725 = fptrunc float %5213 to bfloat, !dbg !39 + %5726 = fptrunc float %5214 to bfloat, !dbg !39 + %5727 = fptrunc float %5215 to bfloat, !dbg !39 + %5728 = fptrunc float %5216 to bfloat, !dbg !39 + %5729 = fptrunc float %5217 to bfloat, !dbg !39 + %5730 = fptrunc float %5218 to bfloat, !dbg !39 + %5731 = fptrunc float %5219 to bfloat, !dbg !39 + %5732 = fptrunc float %5220 to bfloat, !dbg !39 + %5733 = fptrunc float %5221 to bfloat, !dbg !39 + %5734 = fptrunc float %5222 to bfloat, !dbg !39 + %5735 = fptrunc float %5223 to bfloat, !dbg !39 + %5736 = fptrunc float %5224 to bfloat, !dbg !39 + %5737 = fptrunc float %5225 to bfloat, !dbg !39 + %5738 = fptrunc float %5226 to bfloat, !dbg !39 + %5739 = fptrunc float %5227 to bfloat, !dbg !39 + %5740 = fptrunc float %5228 to bfloat, !dbg !39 + %5741 = fptrunc float %5229 to bfloat, !dbg !39 + %5742 = fptrunc float %5230 to bfloat, !dbg !39 + %5743 = fptrunc float %5231 to bfloat, !dbg !39 + %5744 = fptrunc float %5232 to bfloat, !dbg !39 + %5745 = fptrunc float %5233 to bfloat, !dbg !39 + %5746 = fptrunc float %5234 to bfloat, !dbg !39 + %5747 = fptrunc float %5235 to bfloat, !dbg !39 + %5748 = fptrunc float %5236 to bfloat, !dbg !39 + %5749 = fptrunc float %5237 to bfloat, !dbg !39 + %5750 = fptrunc float %5238 to bfloat, !dbg !39 + %5751 = fptrunc float %5239 to bfloat, !dbg !39 + %5752 = fptrunc float %5240 to bfloat, !dbg !39 + %5753 = fptrunc float %5241 to bfloat, !dbg !39 + %5754 = fptrunc float %5242 to bfloat, !dbg !39 + %5755 = fptrunc float %5243 to bfloat, !dbg !39 + %5756 = fptrunc float %5244 to bfloat, !dbg !39 + %5757 = fptrunc float %5245 to bfloat, !dbg !39 + %5758 = fptrunc float %5246 to bfloat, !dbg !39 + %5759 = fptrunc float %5247 to bfloat, !dbg !39 + %5760 = fptrunc float %5248 to bfloat, !dbg !39 + %5761 = fptrunc float %5249 to bfloat, !dbg !39 + %5762 = fptrunc float %5250 to bfloat, !dbg !39 + %5763 = fptrunc float %5251 to bfloat, !dbg !39 + %5764 = fptrunc float %5252 to bfloat, !dbg !39 + %5765 = fptrunc float %5253 to bfloat, !dbg !39 + %5766 = fptrunc float %5254 to bfloat, !dbg !39 + %5767 = fptrunc float %5255 to bfloat, !dbg !39 + %5768 = fptrunc float %5256 to bfloat, !dbg !39 + %5769 = fptrunc float %5257 to bfloat, !dbg !39 + %5770 = fptrunc float %5258 to bfloat, !dbg !39 + %5771 = fptrunc float %5259 to bfloat, !dbg !39 + %5772 = fptrunc float %5260 to bfloat, !dbg !39 + %5773 = fptrunc float %5261 to bfloat, !dbg !39 + %5774 = fptrunc float %5262 to bfloat, !dbg !39 + %5775 = fptrunc float %5263 to bfloat, !dbg !39 + %5776 = fptrunc float %5264 to bfloat, !dbg !39 + %5777 = fptrunc float %5265 to bfloat, !dbg !39 + %5778 = fptrunc float %5266 to bfloat, !dbg !39 + %5779 = fptrunc float %5267 to bfloat, !dbg !39 + %5780 = fptrunc float %5268 to bfloat, !dbg !39 + %5781 = fptrunc float %5269 to bfloat, !dbg !39 + %5782 = fptrunc float %5270 to bfloat, !dbg !39 + %5783 = fptrunc float %5271 to bfloat, !dbg !39 + %5784 = fptrunc float %5272 to bfloat, !dbg !39 + %5785 = fptrunc float %5273 to bfloat, !dbg !39 + %5786 = fptrunc float %5274 to bfloat, !dbg !39 + %5787 = fptrunc float %5275 to bfloat, !dbg !39 + %5788 = fptrunc float %5276 to bfloat, !dbg !39 + %5789 = fptrunc float %5277 to bfloat, !dbg !39 + %5790 = fptrunc float %5278 to bfloat, !dbg !39 + %5791 = fptrunc float %5279 to bfloat, !dbg !39 + %5792 = fptrunc float %5280 to bfloat, !dbg !39 + %5793 = fptrunc float %5281 to bfloat, !dbg !39 + %5794 = fptrunc float %5282 to bfloat, !dbg !39 + %5795 = fptrunc float %5283 to bfloat, !dbg !39 + %5796 = fptrunc float %5284 to bfloat, !dbg !39 + %5797 = fptrunc float %5285 to bfloat, !dbg !39 + %5798 = fptrunc float %5286 to bfloat, !dbg !39 + %5799 = fptrunc float %5287 to bfloat, !dbg !39 + %5800 = fptrunc float %5288 to bfloat, !dbg !39 + %5801 = fptrunc float %5289 to bfloat, !dbg !39 + %5802 = fptrunc float %5290 to bfloat, !dbg !39 + %5803 = fptrunc float %5291 to bfloat, !dbg !39 + %5804 = fptrunc float %5292 to bfloat, !dbg !39 + %5805 = fptrunc float %5293 to bfloat, !dbg !39 + %5806 = fptrunc float %5294 to bfloat, !dbg !39 + %5807 = fptrunc float %5295 to bfloat, !dbg !39 + %5808 = fptrunc float %5296 to bfloat, !dbg !39 + %5809 = fptrunc float %5297 to bfloat, !dbg !39 + %5810 = fptrunc float %5298 to bfloat, !dbg !39 + %5811 = fptrunc float %5299 to bfloat, !dbg !39 + %5812 = fptrunc float %5300 to bfloat, !dbg !39 + %5813 = fptrunc float %5301 to bfloat, !dbg !39 + %5814 = fptrunc float %5302 to bfloat, !dbg !39 + %5815 = fptrunc float %5303 to bfloat, !dbg !39 + %5816 = fptrunc float %5304 to bfloat, !dbg !39 + %5817 = fptrunc float %5305 to bfloat, !dbg !39 + %5818 = fptrunc float %5306 to bfloat, !dbg !39 + %5819 = fptrunc float %5307 to bfloat, !dbg !39 + %5820 = fptrunc float %5308 to bfloat, !dbg !39 + %5821 = fptrunc float %5309 to bfloat, !dbg !39 + %5822 = fptrunc float %5310 to bfloat, !dbg !39 + %5823 = fptrunc float %5311 to bfloat, !dbg !39 + %5824 = fptrunc float %5312 to bfloat, !dbg !39 + %5825 = fptrunc float %5313 to bfloat, !dbg !39 + %5826 = fptrunc float %5314 to bfloat, !dbg !39 + %5827 = fptrunc float %5315 to bfloat, !dbg !39 + %5828 = fptrunc float %5316 to bfloat, !dbg !39 + %5829 = fptrunc float %5317 to bfloat, !dbg !39 + %5830 = fptrunc float %5318 to bfloat, !dbg !39 + %5831 = fptrunc float %5319 to bfloat, !dbg !39 + %5832 = fptrunc float %5320 to bfloat, !dbg !39 + %5833 = fptrunc float %5321 to bfloat, !dbg !39 + %5834 = fptrunc float %5322 to bfloat, !dbg !39 + %5835 = fptrunc float %5323 to bfloat, !dbg !39 + %5836 = fptrunc float %5324 to bfloat, !dbg !39 + %5837 = fptrunc float %5325 to bfloat, !dbg !39 + %5838 = fptrunc float %5326 to bfloat, !dbg !39 + %5839 = fptrunc float %5327 to bfloat, !dbg !39 + %5840 = fptrunc float %5328 to bfloat, !dbg !39 + %5841 = fptrunc float %5329 to bfloat, !dbg !39 + %5842 = fptrunc float %5330 to bfloat, !dbg !39 + %5843 = fptrunc float %5331 to bfloat, !dbg !39 + %5844 = fptrunc float %5332 to bfloat, !dbg !39 + %5845 = fptrunc float %5333 to bfloat, !dbg !39 + %5846 = fptrunc float %5334 to bfloat, !dbg !39 + %5847 = fptrunc float %5335 to bfloat, !dbg !39 + %5848 = fptrunc float %5336 to bfloat, !dbg !39 + %5849 = fptrunc float %5337 to bfloat, !dbg !39 + %5850 = fptrunc float %5338 to bfloat, !dbg !39 + %5851 = fptrunc float %5339 to bfloat, !dbg !39 + %5852 = fptrunc float %5340 to bfloat, !dbg !39 + %5853 = fptrunc float %5341 to bfloat, !dbg !39 + %5854 = fptrunc float %5342 to bfloat, !dbg !39 + %5855 = fptrunc float %5343 to bfloat, !dbg !39 + %5856 = fptrunc float %5344 to bfloat, !dbg !39 + %5857 = fptrunc float %5345 to bfloat, !dbg !39 + %5858 = fptrunc float %5346 to bfloat, !dbg !39 + %5859 = fptrunc float %5347 to bfloat, !dbg !39 + %5860 = fptrunc float %5348 to bfloat, !dbg !39 + %5861 = fptrunc float %5349 to bfloat, !dbg !39 + %5862 = fptrunc float %5350 to bfloat, !dbg !39 + %5863 = fptrunc float %5351 to bfloat, !dbg !39 + %5864 = fptrunc float %5352 to bfloat, !dbg !39 + %5865 = fptrunc float %5353 to bfloat, !dbg !39 + %5866 = fptrunc float %5354 to bfloat, !dbg !39 + %5867 = fptrunc float %5355 to bfloat, !dbg !39 + %5868 = fptrunc float %5356 to bfloat, !dbg !39 + %5869 = fptrunc float %5357 to bfloat, !dbg !39 + %5870 = fptrunc float %5358 to bfloat, !dbg !39 + %5871 = fptrunc float %5359 to bfloat, !dbg !39 + %5872 = fptrunc float %5360 to bfloat, !dbg !39 + %5873 = fptrunc float %5361 to bfloat, !dbg !39 + %5874 = fptrunc float %5362 to bfloat, !dbg !39 + %5875 = fptrunc float %5363 to bfloat, !dbg !39 + %5876 = fptrunc float %5364 to bfloat, !dbg !39 + %5877 = fptrunc float %5365 to bfloat, !dbg !39 + %5878 = fptrunc float %5366 to bfloat, !dbg !39 + %5879 = fptrunc float %5367 to bfloat, !dbg !39 + %5880 = fptrunc float %5368 to bfloat, !dbg !39 + %5881 = fptrunc float %5369 to bfloat, !dbg !39 + %5882 = fptrunc float %5370 to bfloat, !dbg !39 + %5883 = fptrunc float %5371 to bfloat, !dbg !39 + %5884 = fptrunc float %5372 to bfloat, !dbg !39 + %5885 = fptrunc float %5373 to bfloat, !dbg !39 + %5886 = fptrunc float %5374 to bfloat, !dbg !39 + %5887 = fptrunc float %5375 to bfloat, !dbg !39 + %5888 = fptrunc float %5376 to bfloat, !dbg !39 + %5889 = fptrunc float %5377 to bfloat, !dbg !39 + %5890 = fptrunc float %5378 to bfloat, !dbg !39 + %5891 = fptrunc float %5379 to bfloat, !dbg !39 + %5892 = fptrunc float %5380 to bfloat, !dbg !39 + %5893 = insertelement <2 x float> poison, float %5381, i64 0, !dbg !39 + %5894 = insertelement <2 x float> %5893, float %5382, i64 1, !dbg !39 + %5895 = fptrunc <2 x float> %5894 to <2 x bfloat>, !dbg !39 + %5896 = fptrunc float %5383 to bfloat, !dbg !39 + %5897 = fptrunc float %5384 to bfloat, !dbg !39 + %5898 = fptrunc float %5385 to bfloat, !dbg !39 + %5899 = fptrunc float %5386 to bfloat, !dbg !39 + %5900 = fptrunc float %5387 to bfloat, !dbg !39 + %5901 = fptrunc float %5388 to bfloat, !dbg !39 + %5902 = insertelement <2 x float> poison, float %5389, i64 0, !dbg !39 + %5903 = insertelement <2 x float> %5902, float %5390, i64 1, !dbg !39 + %5904 = fptrunc <2 x float> %5903 to <2 x bfloat>, !dbg !39 + %5905 = insertelement <2 x float> poison, float %5391, i64 0, !dbg !39 + %5906 = insertelement <2 x float> %5905, float %5392, i64 1, !dbg !39 + %5907 = fptrunc <2 x float> %5906 to <2 x bfloat>, !dbg !39 + %5908 = insertelement <2 x float> poison, float %5393, i64 0, !dbg !39 + %5909 = insertelement <2 x float> %5908, float %5394, i64 1, !dbg !39 + %5910 = fptrunc <2 x float> %5909 to <2 x bfloat>, !dbg !39 + %5911 = insertelement <2 x float> poison, float %5395, i64 0, !dbg !39 + %5912 = insertelement <2 x float> %5911, float %5396, i64 1, !dbg !39 + %5913 = fptrunc <2 x float> %5912 to <2 x bfloat>, !dbg !39 + %5914 = insertelement <2 x float> poison, float %5397, i64 0, !dbg !39 + %5915 = insertelement <2 x float> %5914, float %5398, i64 1, !dbg !39 + %5916 = fptrunc <2 x float> %5915 to <2 x bfloat>, !dbg !39 + %5917 = insertelement <2 x float> poison, float %5399, i64 0, !dbg !39 + %5918 = insertelement <2 x float> %5917, float %5400, i64 1, !dbg !39 + %5919 = fptrunc <2 x float> %5918 to <2 x bfloat>, !dbg !39 + %5920 = insertelement <2 x float> poison, float %5401, i64 0, !dbg !39 + %5921 = insertelement <2 x float> %5920, float %5402, i64 1, !dbg !39 + %5922 = fptrunc <2 x float> %5921 to <2 x bfloat>, !dbg !39 + %5923 = insertelement <2 x float> poison, float %5403, i64 0, !dbg !39 + %5924 = insertelement <2 x float> %5923, float %5404, i64 1, !dbg !39 + %5925 = fptrunc <2 x float> %5924 to <2 x bfloat>, !dbg !39 + %5926 = insertelement <2 x float> poison, float %5405, i64 0, !dbg !39 + %5927 = insertelement <2 x float> %5926, float %5406, i64 1, !dbg !39 + %5928 = fptrunc <2 x float> %5927 to <2 x bfloat>, !dbg !39 + %5929 = insertelement <2 x float> poison, float %5407, i64 0, !dbg !39 + %5930 = insertelement <2 x float> %5929, float %5408, i64 1, !dbg !39 + %5931 = fptrunc <2 x float> %5930 to <2 x bfloat>, !dbg !39 + %5932 = insertelement <2 x float> poison, float %5409, i64 0, !dbg !39 + %5933 = insertelement <2 x float> %5932, float %5410, i64 1, !dbg !39 + %5934 = fptrunc <2 x float> %5933 to <2 x bfloat>, !dbg !39 + %5935 = insertelement <2 x float> poison, float %5411, i64 0, !dbg !39 + %5936 = insertelement <2 x float> %5935, float %5412, i64 1, !dbg !39 + %5937 = fptrunc <2 x float> %5936 to <2 x bfloat>, !dbg !39 + %5938 = insertelement <2 x float> poison, float %5413, i64 0, !dbg !39 + %5939 = insertelement <2 x float> %5938, float %5414, i64 1, !dbg !39 + %5940 = fptrunc <2 x float> %5939 to <2 x bfloat>, !dbg !39 + %5941 = insertelement <2 x float> poison, float %5415, i64 0, !dbg !39 + %5942 = insertelement <2 x float> %5941, float %5416, i64 1, !dbg !39 + %5943 = fptrunc <2 x float> %5942 to <2 x bfloat>, !dbg !39 + %5944 = insertelement <2 x float> poison, float %5417, i64 0, !dbg !39 + %5945 = insertelement <2 x float> %5944, float %5418, i64 1, !dbg !39 + %5946 = fptrunc <2 x float> %5945 to <2 x bfloat>, !dbg !39 + %5947 = insertelement <2 x float> poison, float %5419, i64 0, !dbg !39 + %5948 = insertelement <2 x float> %5947, float %5420, i64 1, !dbg !39 + %5949 = fptrunc <2 x float> %5948 to <2 x bfloat>, !dbg !39 + %5950 = insertelement <2 x float> poison, float %5421, i64 0, !dbg !39 + %5951 = insertelement <2 x float> %5950, float %5422, i64 1, !dbg !39 + %5952 = fptrunc <2 x float> %5951 to <2 x bfloat>, !dbg !39 + %5953 = insertelement <2 x float> poison, float %5423, i64 0, !dbg !39 + %5954 = insertelement <2 x float> %5953, float %5424, i64 1, !dbg !39 + %5955 = fptrunc <2 x float> %5954 to <2 x bfloat>, !dbg !39 + %5956 = insertelement <2 x float> poison, float %5425, i64 0, !dbg !39 + %5957 = insertelement <2 x float> %5956, float %5426, i64 1, !dbg !39 + %5958 = fptrunc <2 x float> %5957 to <2 x bfloat>, !dbg !39 + %5959 = insertelement <2 x float> poison, float %5427, i64 0, !dbg !39 + %5960 = insertelement <2 x float> %5959, float %5428, i64 1, !dbg !39 + %5961 = fptrunc <2 x float> %5960 to <2 x bfloat>, !dbg !39 + %5962 = insertelement <2 x float> poison, float %5429, i64 0, !dbg !39 + %5963 = insertelement <2 x float> %5962, float %5430, i64 1, !dbg !39 + %5964 = fptrunc <2 x float> %5963 to <2 x bfloat>, !dbg !39 + %5965 = insertelement <2 x float> poison, float %5431, i64 0, !dbg !39 + %5966 = insertelement <2 x float> %5965, float %5432, i64 1, !dbg !39 + %5967 = fptrunc <2 x float> %5966 to <2 x bfloat>, !dbg !39 + %5968 = insertelement <2 x float> poison, float %5433, i64 0, !dbg !39 + %5969 = insertelement <2 x float> %5968, float %5434, i64 1, !dbg !39 + %5970 = fptrunc <2 x float> %5969 to <2 x bfloat>, !dbg !39 + %5971 = insertelement <2 x float> poison, float %5435, i64 0, !dbg !39 + %5972 = insertelement <2 x float> %5971, float %5436, i64 1, !dbg !39 + %5973 = fptrunc <2 x float> %5972 to <2 x bfloat>, !dbg !39 + %5974 = insertelement <2 x float> poison, float %5437, i64 0, !dbg !39 + %5975 = insertelement <2 x float> %5974, float %5438, i64 1, !dbg !39 + %5976 = fptrunc <2 x float> %5975 to <2 x bfloat>, !dbg !39 + %5977 = insertelement <2 x float> poison, float %5439, i64 0, !dbg !39 + %5978 = insertelement <2 x float> %5977, float %5440, i64 1, !dbg !39 + %5979 = fptrunc <2 x float> %5978 to <2 x bfloat>, !dbg !39 + %5980 = insertelement <2 x float> poison, float %5441, i64 0, !dbg !39 + %5981 = insertelement <2 x float> %5980, float %5442, i64 1, !dbg !39 + %5982 = fptrunc <2 x float> %5981 to <2 x bfloat>, !dbg !39 + %5983 = insertelement <2 x float> poison, float %5443, i64 0, !dbg !39 + %5984 = insertelement <2 x float> %5983, float %5444, i64 1, !dbg !39 + %5985 = fptrunc <2 x float> %5984 to <2 x bfloat>, !dbg !39 + %5986 = insertelement <2 x float> poison, float %5445, i64 0, !dbg !39 + %5987 = insertelement <2 x float> %5986, float %5446, i64 1, !dbg !39 + %5988 = fptrunc <2 x float> %5987 to <2 x bfloat>, !dbg !39 + %5989 = insertelement <2 x float> poison, float %5447, i64 0, !dbg !39 + %5990 = insertelement <2 x float> %5989, float %5448, i64 1, !dbg !39 + %5991 = fptrunc <2 x float> %5990 to <2 x bfloat>, !dbg !39 + %5992 = insertelement <2 x float> poison, float %5449, i64 0, !dbg !39 + %5993 = insertelement <2 x float> %5992, float %5450, i64 1, !dbg !39 + %5994 = fptrunc <2 x float> %5993 to <2 x bfloat>, !dbg !39 + %5995 = insertelement <2 x float> poison, float %5451, i64 0, !dbg !39 + %5996 = insertelement <2 x float> %5995, float %5452, i64 1, !dbg !39 + %5997 = fptrunc <2 x float> %5996 to <2 x bfloat>, !dbg !39 + %5998 = insertelement <2 x float> poison, float %5453, i64 0, !dbg !39 + %5999 = insertelement <2 x float> %5998, float %5454, i64 1, !dbg !39 + %6000 = fptrunc <2 x float> %5999 to <2 x bfloat>, !dbg !39 + %6001 = insertelement <2 x float> poison, float %5455, i64 0, !dbg !39 + %6002 = insertelement <2 x float> %6001, float %5456, i64 1, !dbg !39 + %6003 = fptrunc <2 x float> %6002 to <2 x bfloat>, !dbg !39 + %6004 = insertelement <2 x float> poison, float %5457, i64 0, !dbg !39 + %6005 = insertelement <2 x float> %6004, float %5458, i64 1, !dbg !39 + %6006 = fptrunc <2 x float> %6005 to <2 x bfloat>, !dbg !39 + %6007 = insertelement <2 x float> poison, float %5459, i64 0, !dbg !39 + %6008 = insertelement <2 x float> %6007, float %5460, i64 1, !dbg !39 + %6009 = fptrunc <2 x float> %6008 to <2 x bfloat>, !dbg !39 + %6010 = insertelement <2 x float> poison, float %5461, i64 0, !dbg !39 + %6011 = insertelement <2 x float> %6010, float %5462, i64 1, !dbg !39 + %6012 = fptrunc <2 x float> %6011 to <2 x bfloat>, !dbg !39 + %6013 = insertelement <2 x float> poison, float %5463, i64 0, !dbg !39 + %6014 = insertelement <2 x float> %6013, float %5464, i64 1, !dbg !39 + %6015 = fptrunc <2 x float> %6014 to <2 x bfloat>, !dbg !39 + %6016 = insertelement <2 x float> poison, float %5465, i64 0, !dbg !39 + %6017 = insertelement <2 x float> %6016, float %5466, i64 1, !dbg !39 + %6018 = fptrunc <2 x float> %6017 to <2 x bfloat>, !dbg !39 + %6019 = insertelement <2 x float> poison, float %5467, i64 0, !dbg !39 + %6020 = insertelement <2 x float> %6019, float %5468, i64 1, !dbg !39 + %6021 = fptrunc <2 x float> %6020 to <2 x bfloat>, !dbg !39 + %6022 = insertelement <2 x float> poison, float %5469, i64 0, !dbg !39 + %6023 = insertelement <2 x float> %6022, float %5470, i64 1, !dbg !39 + %6024 = fptrunc <2 x float> %6023 to <2 x bfloat>, !dbg !39 + %6025 = insertelement <2 x float> poison, float %5471, i64 0, !dbg !39 + %6026 = insertelement <2 x float> %6025, float %5472, i64 1, !dbg !39 + %6027 = fptrunc <2 x float> %6026 to <2 x bfloat>, !dbg !39 + %6028 = insertelement <2 x float> poison, float %5473, i64 0, !dbg !39 + %6029 = insertelement <2 x float> %6028, float %5474, i64 1, !dbg !39 + %6030 = fptrunc <2 x float> %6029 to <2 x bfloat>, !dbg !39 + %6031 = insertelement <2 x float> poison, float %5475, i64 0, !dbg !39 + %6032 = insertelement <2 x float> %6031, float %5476, i64 1, !dbg !39 + %6033 = fptrunc <2 x float> %6032 to <2 x bfloat>, !dbg !39 + %6034 = insertelement <2 x float> poison, float %5477, i64 0, !dbg !39 + %6035 = insertelement <2 x float> %6034, float %5478, i64 1, !dbg !39 + %6036 = fptrunc <2 x float> %6035 to <2 x bfloat>, !dbg !39 + %6037 = insertelement <2 x float> poison, float %5479, i64 0, !dbg !39 + %6038 = insertelement <2 x float> %6037, float %5480, i64 1, !dbg !39 + %6039 = fptrunc <2 x float> %6038 to <2 x bfloat>, !dbg !39 + %6040 = insertelement <2 x float> poison, float %5481, i64 0, !dbg !39 + %6041 = insertelement <2 x float> %6040, float %5482, i64 1, !dbg !39 + %6042 = fptrunc <2 x float> %6041 to <2 x bfloat>, !dbg !39 + %6043 = insertelement <2 x float> poison, float %5483, i64 0, !dbg !39 + %6044 = insertelement <2 x float> %6043, float %5484, i64 1, !dbg !39 + %6045 = fptrunc <2 x float> %6044 to <2 x bfloat>, !dbg !39 + %6046 = insertelement <2 x float> poison, float %5485, i64 0, !dbg !39 + %6047 = insertelement <2 x float> %6046, float %5486, i64 1, !dbg !39 + %6048 = fptrunc <2 x float> %6047 to <2 x bfloat>, !dbg !39 + %6049 = insertelement <2 x float> poison, float %5487, i64 0, !dbg !39 + %6050 = insertelement <2 x float> %6049, float %5488, i64 1, !dbg !39 + %6051 = fptrunc <2 x float> %6050 to <2 x bfloat>, !dbg !39 + %6052 = insertelement <2 x float> poison, float %5489, i64 0, !dbg !39 + %6053 = insertelement <2 x float> %6052, float %5490, i64 1, !dbg !39 + %6054 = fptrunc <2 x float> %6053 to <2 x bfloat>, !dbg !39 + %6055 = insertelement <2 x float> poison, float %5491, i64 0, !dbg !39 + %6056 = insertelement <2 x float> %6055, float %5492, i64 1, !dbg !39 + %6057 = fptrunc <2 x float> %6056 to <2 x bfloat>, !dbg !39 + %6058 = insertelement <2 x float> poison, float %5493, i64 0, !dbg !39 + %6059 = insertelement <2 x float> %6058, float %5494, i64 1, !dbg !39 + %6060 = fptrunc <2 x float> %6059 to <2 x bfloat>, !dbg !39 + %6061 = insertelement <2 x float> poison, float %5495, i64 0, !dbg !39 + %6062 = insertelement <2 x float> %6061, float %5496, i64 1, !dbg !39 + %6063 = fptrunc <2 x float> %6062 to <2 x bfloat>, !dbg !39 + %6064 = insertelement <2 x float> poison, float %5497, i64 0, !dbg !39 + %6065 = insertelement <2 x float> %6064, float %5498, i64 1, !dbg !39 + %6066 = fptrunc <2 x float> %6065 to <2 x bfloat>, !dbg !39 + %6067 = insertelement <2 x float> poison, float %5499, i64 0, !dbg !39 + %6068 = insertelement <2 x float> %6067, float %5500, i64 1, !dbg !39 + %6069 = fptrunc <2 x float> %6068 to <2 x bfloat>, !dbg !39 + %6070 = insertelement <2 x float> poison, float %5501, i64 0, !dbg !39 + %6071 = insertelement <2 x float> %6070, float %5502, i64 1, !dbg !39 + %6072 = fptrunc <2 x float> %6071 to <2 x bfloat>, !dbg !39 + %6073 = insertelement <2 x float> poison, float %5503, i64 0, !dbg !39 + %6074 = insertelement <2 x float> %6073, float %5504, i64 1, !dbg !39 + %6075 = fptrunc <2 x float> %6074 to <2 x bfloat>, !dbg !39 + %6076 = insertelement <2 x float> poison, float %5505, i64 0, !dbg !39 + %6077 = insertelement <2 x float> %6076, float %5506, i64 1, !dbg !39 + %6078 = fptrunc <2 x float> %6077 to <2 x bfloat>, !dbg !39 + %6079 = insertelement <2 x float> poison, float %5507, i64 0, !dbg !39 + %6080 = insertelement <2 x float> %6079, float %5508, i64 1, !dbg !39 + %6081 = fptrunc <2 x float> %6080 to <2 x bfloat>, !dbg !39 + %6082 = insertelement <2 x bfloat> poison, bfloat %5509, i64 0, !dbg !40 + %6083 = insertelement <2 x bfloat> %6082, bfloat %5510, i64 1, !dbg !40 + %6084 = bitcast <2 x bfloat> %6083 to i32, !dbg !40 + %6085 = insertelement <2 x bfloat> poison, bfloat %5511, i64 0, !dbg !40 + %6086 = insertelement <2 x bfloat> %6085, bfloat %5512, i64 1, !dbg !40 + %6087 = bitcast <2 x bfloat> %6086 to i32, !dbg !40 + %6088 = insertelement <2 x bfloat> poison, bfloat %5513, i64 0, !dbg !40 + %6089 = insertelement <2 x bfloat> %6088, bfloat %5514, i64 1, !dbg !40 + %6090 = bitcast <2 x bfloat> %6089 to i32, !dbg !40 + %6091 = insertelement <2 x bfloat> poison, bfloat %5515, i64 0, !dbg !40 + %6092 = insertelement <2 x bfloat> %6091, bfloat %5516, i64 1, !dbg !40 + %6093 = bitcast <2 x bfloat> %6092 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6084, i32 %6087, i32 %6090, i32 %6093, ptr addrspace(1) %4928, i1 %303) #7, !dbg !40 + %6094 = insertelement <2 x bfloat> poison, bfloat %5517, i64 0, !dbg !40 + %6095 = insertelement <2 x bfloat> %6094, bfloat %5518, i64 1, !dbg !40 + %6096 = bitcast <2 x bfloat> %6095 to i32, !dbg !40 + %6097 = insertelement <2 x bfloat> poison, bfloat %5519, i64 0, !dbg !40 + %6098 = insertelement <2 x bfloat> %6097, bfloat %5520, i64 1, !dbg !40 + %6099 = bitcast <2 x bfloat> %6098 to i32, !dbg !40 + %6100 = insertelement <2 x bfloat> poison, bfloat %5521, i64 0, !dbg !40 + %6101 = insertelement <2 x bfloat> %6100, bfloat %5522, i64 1, !dbg !40 + %6102 = bitcast <2 x bfloat> %6101 to i32, !dbg !40 + %6103 = insertelement <2 x bfloat> poison, bfloat %5523, i64 0, !dbg !40 + %6104 = insertelement <2 x bfloat> %6103, bfloat %5524, i64 1, !dbg !40 + %6105 = bitcast <2 x bfloat> %6104 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6096, i32 %6099, i32 %6102, i32 %6105, ptr addrspace(1) %4926, i1 %321) #7, !dbg !40 + %6106 = insertelement <2 x bfloat> poison, bfloat %5525, i64 0, !dbg !40 + %6107 = insertelement <2 x bfloat> %6106, bfloat %5526, i64 1, !dbg !40 + %6108 = bitcast <2 x bfloat> %6107 to i32, !dbg !40 + %6109 = insertelement <2 x bfloat> poison, bfloat %5527, i64 0, !dbg !40 + %6110 = insertelement <2 x bfloat> %6109, bfloat %5528, i64 1, !dbg !40 + %6111 = bitcast <2 x bfloat> %6110 to i32, !dbg !40 + %6112 = insertelement <2 x bfloat> poison, bfloat %5529, i64 0, !dbg !40 + %6113 = insertelement <2 x bfloat> %6112, bfloat %5530, i64 1, !dbg !40 + %6114 = bitcast <2 x bfloat> %6113 to i32, !dbg !40 + %6115 = insertelement <2 x bfloat> poison, bfloat %5531, i64 0, !dbg !40 + %6116 = insertelement <2 x bfloat> %6115, bfloat %5532, i64 1, !dbg !40 + %6117 = bitcast <2 x bfloat> %6116 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6108, i32 %6111, i32 %6114, i32 %6117, ptr addrspace(1) %4924, i1 %339) #7, !dbg !40 + %6118 = insertelement <2 x bfloat> poison, bfloat %5533, i64 0, !dbg !40 + %6119 = insertelement <2 x bfloat> %6118, bfloat %5534, i64 1, !dbg !40 + %6120 = bitcast <2 x bfloat> %6119 to i32, !dbg !40 + %6121 = insertelement <2 x bfloat> poison, bfloat %5535, i64 0, !dbg !40 + %6122 = insertelement <2 x bfloat> %6121, bfloat %5536, i64 1, !dbg !40 + %6123 = bitcast <2 x bfloat> %6122 to i32, !dbg !40 + %6124 = insertelement <2 x bfloat> poison, bfloat %5537, i64 0, !dbg !40 + %6125 = insertelement <2 x bfloat> %6124, bfloat %5538, i64 1, !dbg !40 + %6126 = bitcast <2 x bfloat> %6125 to i32, !dbg !40 + %6127 = insertelement <2 x bfloat> poison, bfloat %5539, i64 0, !dbg !40 + %6128 = insertelement <2 x bfloat> %6127, bfloat %5540, i64 1, !dbg !40 + %6129 = bitcast <2 x bfloat> %6128 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6120, i32 %6123, i32 %6126, i32 %6129, ptr addrspace(1) %4922, i1 %357) #7, !dbg !40 + %6130 = insertelement <2 x bfloat> poison, bfloat %5541, i64 0, !dbg !40 + %6131 = insertelement <2 x bfloat> %6130, bfloat %5542, i64 1, !dbg !40 + %6132 = bitcast <2 x bfloat> %6131 to i32, !dbg !40 + %6133 = insertelement <2 x bfloat> poison, bfloat %5543, i64 0, !dbg !40 + %6134 = insertelement <2 x bfloat> %6133, bfloat %5544, i64 1, !dbg !40 + %6135 = bitcast <2 x bfloat> %6134 to i32, !dbg !40 + %6136 = insertelement <2 x bfloat> poison, bfloat %5545, i64 0, !dbg !40 + %6137 = insertelement <2 x bfloat> %6136, bfloat %5546, i64 1, !dbg !40 + %6138 = bitcast <2 x bfloat> %6137 to i32, !dbg !40 + %6139 = insertelement <2 x bfloat> poison, bfloat %5547, i64 0, !dbg !40 + %6140 = insertelement <2 x bfloat> %6139, bfloat %5548, i64 1, !dbg !40 + %6141 = bitcast <2 x bfloat> %6140 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6132, i32 %6135, i32 %6138, i32 %6141, ptr addrspace(1) %4920, i1 %375) #7, !dbg !40 + %6142 = insertelement <2 x bfloat> poison, bfloat %5549, i64 0, !dbg !40 + %6143 = insertelement <2 x bfloat> %6142, bfloat %5550, i64 1, !dbg !40 + %6144 = bitcast <2 x bfloat> %6143 to i32, !dbg !40 + %6145 = insertelement <2 x bfloat> poison, bfloat %5551, i64 0, !dbg !40 + %6146 = insertelement <2 x bfloat> %6145, bfloat %5552, i64 1, !dbg !40 + %6147 = bitcast <2 x bfloat> %6146 to i32, !dbg !40 + %6148 = insertelement <2 x bfloat> poison, bfloat %5553, i64 0, !dbg !40 + %6149 = insertelement <2 x bfloat> %6148, bfloat %5554, i64 1, !dbg !40 + %6150 = bitcast <2 x bfloat> %6149 to i32, !dbg !40 + %6151 = insertelement <2 x bfloat> poison, bfloat %5555, i64 0, !dbg !40 + %6152 = insertelement <2 x bfloat> %6151, bfloat %5556, i64 1, !dbg !40 + %6153 = bitcast <2 x bfloat> %6152 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6144, i32 %6147, i32 %6150, i32 %6153, ptr addrspace(1) %4918, i1 %393) #7, !dbg !40 + %6154 = insertelement <2 x bfloat> poison, bfloat %5557, i64 0, !dbg !40 + %6155 = insertelement <2 x bfloat> %6154, bfloat %5558, i64 1, !dbg !40 + %6156 = bitcast <2 x bfloat> %6155 to i32, !dbg !40 + %6157 = insertelement <2 x bfloat> poison, bfloat %5559, i64 0, !dbg !40 + %6158 = insertelement <2 x bfloat> %6157, bfloat %5560, i64 1, !dbg !40 + %6159 = bitcast <2 x bfloat> %6158 to i32, !dbg !40 + %6160 = insertelement <2 x bfloat> poison, bfloat %5561, i64 0, !dbg !40 + %6161 = insertelement <2 x bfloat> %6160, bfloat %5562, i64 1, !dbg !40 + %6162 = bitcast <2 x bfloat> %6161 to i32, !dbg !40 + %6163 = insertelement <2 x bfloat> poison, bfloat %5563, i64 0, !dbg !40 + %6164 = insertelement <2 x bfloat> %6163, bfloat %5564, i64 1, !dbg !40 + %6165 = bitcast <2 x bfloat> %6164 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6156, i32 %6159, i32 %6162, i32 %6165, ptr addrspace(1) %4916, i1 %411) #7, !dbg !40 + %6166 = insertelement <2 x bfloat> poison, bfloat %5565, i64 0, !dbg !40 + %6167 = insertelement <2 x bfloat> %6166, bfloat %5566, i64 1, !dbg !40 + %6168 = bitcast <2 x bfloat> %6167 to i32, !dbg !40 + %6169 = insertelement <2 x bfloat> poison, bfloat %5567, i64 0, !dbg !40 + %6170 = insertelement <2 x bfloat> %6169, bfloat %5568, i64 1, !dbg !40 + %6171 = bitcast <2 x bfloat> %6170 to i32, !dbg !40 + %6172 = insertelement <2 x bfloat> poison, bfloat %5569, i64 0, !dbg !40 + %6173 = insertelement <2 x bfloat> %6172, bfloat %5570, i64 1, !dbg !40 + %6174 = bitcast <2 x bfloat> %6173 to i32, !dbg !40 + %6175 = insertelement <2 x bfloat> poison, bfloat %5571, i64 0, !dbg !40 + %6176 = insertelement <2 x bfloat> %6175, bfloat %5572, i64 1, !dbg !40 + %6177 = bitcast <2 x bfloat> %6176 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6168, i32 %6171, i32 %6174, i32 %6177, ptr addrspace(1) %4914, i1 %429) #7, !dbg !40 + %6178 = insertelement <2 x bfloat> poison, bfloat %5573, i64 0, !dbg !40 + %6179 = insertelement <2 x bfloat> %6178, bfloat %5574, i64 1, !dbg !40 + %6180 = bitcast <2 x bfloat> %6179 to i32, !dbg !40 + %6181 = insertelement <2 x bfloat> poison, bfloat %5575, i64 0, !dbg !40 + %6182 = insertelement <2 x bfloat> %6181, bfloat %5576, i64 1, !dbg !40 + %6183 = bitcast <2 x bfloat> %6182 to i32, !dbg !40 + %6184 = insertelement <2 x bfloat> poison, bfloat %5577, i64 0, !dbg !40 + %6185 = insertelement <2 x bfloat> %6184, bfloat %5578, i64 1, !dbg !40 + %6186 = bitcast <2 x bfloat> %6185 to i32, !dbg !40 + %6187 = insertelement <2 x bfloat> poison, bfloat %5579, i64 0, !dbg !40 + %6188 = insertelement <2 x bfloat> %6187, bfloat %5580, i64 1, !dbg !40 + %6189 = bitcast <2 x bfloat> %6188 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6180, i32 %6183, i32 %6186, i32 %6189, ptr addrspace(1) %4912, i1 %447) #7, !dbg !40 + %6190 = insertelement <2 x bfloat> poison, bfloat %5581, i64 0, !dbg !40 + %6191 = insertelement <2 x bfloat> %6190, bfloat %5582, i64 1, !dbg !40 + %6192 = bitcast <2 x bfloat> %6191 to i32, !dbg !40 + %6193 = insertelement <2 x bfloat> poison, bfloat %5583, i64 0, !dbg !40 + %6194 = insertelement <2 x bfloat> %6193, bfloat %5584, i64 1, !dbg !40 + %6195 = bitcast <2 x bfloat> %6194 to i32, !dbg !40 + %6196 = insertelement <2 x bfloat> poison, bfloat %5585, i64 0, !dbg !40 + %6197 = insertelement <2 x bfloat> %6196, bfloat %5586, i64 1, !dbg !40 + %6198 = bitcast <2 x bfloat> %6197 to i32, !dbg !40 + %6199 = insertelement <2 x bfloat> poison, bfloat %5587, i64 0, !dbg !40 + %6200 = insertelement <2 x bfloat> %6199, bfloat %5588, i64 1, !dbg !40 + %6201 = bitcast <2 x bfloat> %6200 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6192, i32 %6195, i32 %6198, i32 %6201, ptr addrspace(1) %4910, i1 %465) #7, !dbg !40 + %6202 = insertelement <2 x bfloat> poison, bfloat %5589, i64 0, !dbg !40 + %6203 = insertelement <2 x bfloat> %6202, bfloat %5590, i64 1, !dbg !40 + %6204 = bitcast <2 x bfloat> %6203 to i32, !dbg !40 + %6205 = insertelement <2 x bfloat> poison, bfloat %5591, i64 0, !dbg !40 + %6206 = insertelement <2 x bfloat> %6205, bfloat %5592, i64 1, !dbg !40 + %6207 = bitcast <2 x bfloat> %6206 to i32, !dbg !40 + %6208 = insertelement <2 x bfloat> poison, bfloat %5593, i64 0, !dbg !40 + %6209 = insertelement <2 x bfloat> %6208, bfloat %5594, i64 1, !dbg !40 + %6210 = bitcast <2 x bfloat> %6209 to i32, !dbg !40 + %6211 = insertelement <2 x bfloat> poison, bfloat %5595, i64 0, !dbg !40 + %6212 = insertelement <2 x bfloat> %6211, bfloat %5596, i64 1, !dbg !40 + %6213 = bitcast <2 x bfloat> %6212 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6204, i32 %6207, i32 %6210, i32 %6213, ptr addrspace(1) %4908, i1 %483) #7, !dbg !40 + %6214 = insertelement <2 x bfloat> poison, bfloat %5597, i64 0, !dbg !40 + %6215 = insertelement <2 x bfloat> %6214, bfloat %5598, i64 1, !dbg !40 + %6216 = bitcast <2 x bfloat> %6215 to i32, !dbg !40 + %6217 = insertelement <2 x bfloat> poison, bfloat %5599, i64 0, !dbg !40 + %6218 = insertelement <2 x bfloat> %6217, bfloat %5600, i64 1, !dbg !40 + %6219 = bitcast <2 x bfloat> %6218 to i32, !dbg !40 + %6220 = insertelement <2 x bfloat> poison, bfloat %5601, i64 0, !dbg !40 + %6221 = insertelement <2 x bfloat> %6220, bfloat %5602, i64 1, !dbg !40 + %6222 = bitcast <2 x bfloat> %6221 to i32, !dbg !40 + %6223 = insertelement <2 x bfloat> poison, bfloat %5603, i64 0, !dbg !40 + %6224 = insertelement <2 x bfloat> %6223, bfloat %5604, i64 1, !dbg !40 + %6225 = bitcast <2 x bfloat> %6224 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6216, i32 %6219, i32 %6222, i32 %6225, ptr addrspace(1) %4906, i1 %501) #7, !dbg !40 + %6226 = insertelement <2 x bfloat> poison, bfloat %5605, i64 0, !dbg !40 + %6227 = insertelement <2 x bfloat> %6226, bfloat %5606, i64 1, !dbg !40 + %6228 = bitcast <2 x bfloat> %6227 to i32, !dbg !40 + %6229 = insertelement <2 x bfloat> poison, bfloat %5607, i64 0, !dbg !40 + %6230 = insertelement <2 x bfloat> %6229, bfloat %5608, i64 1, !dbg !40 + %6231 = bitcast <2 x bfloat> %6230 to i32, !dbg !40 + %6232 = insertelement <2 x bfloat> poison, bfloat %5609, i64 0, !dbg !40 + %6233 = insertelement <2 x bfloat> %6232, bfloat %5610, i64 1, !dbg !40 + %6234 = bitcast <2 x bfloat> %6233 to i32, !dbg !40 + %6235 = insertelement <2 x bfloat> poison, bfloat %5611, i64 0, !dbg !40 + %6236 = insertelement <2 x bfloat> %6235, bfloat %5612, i64 1, !dbg !40 + %6237 = bitcast <2 x bfloat> %6236 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6228, i32 %6231, i32 %6234, i32 %6237, ptr addrspace(1) %4904, i1 %519) #7, !dbg !40 + %6238 = insertelement <2 x bfloat> poison, bfloat %5613, i64 0, !dbg !40 + %6239 = insertelement <2 x bfloat> %6238, bfloat %5614, i64 1, !dbg !40 + %6240 = bitcast <2 x bfloat> %6239 to i32, !dbg !40 + %6241 = insertelement <2 x bfloat> poison, bfloat %5615, i64 0, !dbg !40 + %6242 = insertelement <2 x bfloat> %6241, bfloat %5616, i64 1, !dbg !40 + %6243 = bitcast <2 x bfloat> %6242 to i32, !dbg !40 + %6244 = insertelement <2 x bfloat> poison, bfloat %5617, i64 0, !dbg !40 + %6245 = insertelement <2 x bfloat> %6244, bfloat %5618, i64 1, !dbg !40 + %6246 = bitcast <2 x bfloat> %6245 to i32, !dbg !40 + %6247 = insertelement <2 x bfloat> poison, bfloat %5619, i64 0, !dbg !40 + %6248 = insertelement <2 x bfloat> %6247, bfloat %5620, i64 1, !dbg !40 + %6249 = bitcast <2 x bfloat> %6248 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6240, i32 %6243, i32 %6246, i32 %6249, ptr addrspace(1) %4902, i1 %537) #7, !dbg !40 + %6250 = insertelement <2 x bfloat> poison, bfloat %5621, i64 0, !dbg !40 + %6251 = insertelement <2 x bfloat> %6250, bfloat %5622, i64 1, !dbg !40 + %6252 = bitcast <2 x bfloat> %6251 to i32, !dbg !40 + %6253 = insertelement <2 x bfloat> poison, bfloat %5623, i64 0, !dbg !40 + %6254 = insertelement <2 x bfloat> %6253, bfloat %5624, i64 1, !dbg !40 + %6255 = bitcast <2 x bfloat> %6254 to i32, !dbg !40 + %6256 = insertelement <2 x bfloat> poison, bfloat %5625, i64 0, !dbg !40 + %6257 = insertelement <2 x bfloat> %6256, bfloat %5626, i64 1, !dbg !40 + %6258 = bitcast <2 x bfloat> %6257 to i32, !dbg !40 + %6259 = insertelement <2 x bfloat> poison, bfloat %5627, i64 0, !dbg !40 + %6260 = insertelement <2 x bfloat> %6259, bfloat %5628, i64 1, !dbg !40 + %6261 = bitcast <2 x bfloat> %6260 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6252, i32 %6255, i32 %6258, i32 %6261, ptr addrspace(1) %4900, i1 %555) #7, !dbg !40 + %6262 = insertelement <2 x bfloat> poison, bfloat %5629, i64 0, !dbg !40 + %6263 = insertelement <2 x bfloat> %6262, bfloat %5630, i64 1, !dbg !40 + %6264 = bitcast <2 x bfloat> %6263 to i32, !dbg !40 + %6265 = insertelement <2 x bfloat> poison, bfloat %5631, i64 0, !dbg !40 + %6266 = insertelement <2 x bfloat> %6265, bfloat %5632, i64 1, !dbg !40 + %6267 = bitcast <2 x bfloat> %6266 to i32, !dbg !40 + %6268 = insertelement <2 x bfloat> poison, bfloat %5633, i64 0, !dbg !40 + %6269 = insertelement <2 x bfloat> %6268, bfloat %5634, i64 1, !dbg !40 + %6270 = bitcast <2 x bfloat> %6269 to i32, !dbg !40 + %6271 = insertelement <2 x bfloat> poison, bfloat %5635, i64 0, !dbg !40 + %6272 = insertelement <2 x bfloat> %6271, bfloat %5636, i64 1, !dbg !40 + %6273 = bitcast <2 x bfloat> %6272 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6264, i32 %6267, i32 %6270, i32 %6273, ptr addrspace(1) %4898, i1 %573) #7, !dbg !40 + %6274 = insertelement <2 x bfloat> poison, bfloat %5637, i64 0, !dbg !40 + %6275 = insertelement <2 x bfloat> %6274, bfloat %5638, i64 1, !dbg !40 + %6276 = bitcast <2 x bfloat> %6275 to i32, !dbg !40 + %6277 = insertelement <2 x bfloat> poison, bfloat %5639, i64 0, !dbg !40 + %6278 = insertelement <2 x bfloat> %6277, bfloat %5640, i64 1, !dbg !40 + %6279 = bitcast <2 x bfloat> %6278 to i32, !dbg !40 + %6280 = insertelement <2 x bfloat> poison, bfloat %5641, i64 0, !dbg !40 + %6281 = insertelement <2 x bfloat> %6280, bfloat %5642, i64 1, !dbg !40 + %6282 = bitcast <2 x bfloat> %6281 to i32, !dbg !40 + %6283 = insertelement <2 x bfloat> poison, bfloat %5643, i64 0, !dbg !40 + %6284 = insertelement <2 x bfloat> %6283, bfloat %5644, i64 1, !dbg !40 + %6285 = bitcast <2 x bfloat> %6284 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6276, i32 %6279, i32 %6282, i32 %6285, ptr addrspace(1) %4896, i1 %591) #7, !dbg !40 + %6286 = insertelement <2 x bfloat> poison, bfloat %5645, i64 0, !dbg !40 + %6287 = insertelement <2 x bfloat> %6286, bfloat %5646, i64 1, !dbg !40 + %6288 = bitcast <2 x bfloat> %6287 to i32, !dbg !40 + %6289 = insertelement <2 x bfloat> poison, bfloat %5647, i64 0, !dbg !40 + %6290 = insertelement <2 x bfloat> %6289, bfloat %5648, i64 1, !dbg !40 + %6291 = bitcast <2 x bfloat> %6290 to i32, !dbg !40 + %6292 = insertelement <2 x bfloat> poison, bfloat %5649, i64 0, !dbg !40 + %6293 = insertelement <2 x bfloat> %6292, bfloat %5650, i64 1, !dbg !40 + %6294 = bitcast <2 x bfloat> %6293 to i32, !dbg !40 + %6295 = insertelement <2 x bfloat> poison, bfloat %5651, i64 0, !dbg !40 + %6296 = insertelement <2 x bfloat> %6295, bfloat %5652, i64 1, !dbg !40 + %6297 = bitcast <2 x bfloat> %6296 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6288, i32 %6291, i32 %6294, i32 %6297, ptr addrspace(1) %4894, i1 %609) #7, !dbg !40 + %6298 = insertelement <2 x bfloat> poison, bfloat %5653, i64 0, !dbg !40 + %6299 = insertelement <2 x bfloat> %6298, bfloat %5654, i64 1, !dbg !40 + %6300 = bitcast <2 x bfloat> %6299 to i32, !dbg !40 + %6301 = insertelement <2 x bfloat> poison, bfloat %5655, i64 0, !dbg !40 + %6302 = insertelement <2 x bfloat> %6301, bfloat %5656, i64 1, !dbg !40 + %6303 = bitcast <2 x bfloat> %6302 to i32, !dbg !40 + %6304 = insertelement <2 x bfloat> poison, bfloat %5657, i64 0, !dbg !40 + %6305 = insertelement <2 x bfloat> %6304, bfloat %5658, i64 1, !dbg !40 + %6306 = bitcast <2 x bfloat> %6305 to i32, !dbg !40 + %6307 = insertelement <2 x bfloat> poison, bfloat %5659, i64 0, !dbg !40 + %6308 = insertelement <2 x bfloat> %6307, bfloat %5660, i64 1, !dbg !40 + %6309 = bitcast <2 x bfloat> %6308 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6300, i32 %6303, i32 %6306, i32 %6309, ptr addrspace(1) %4892, i1 %627) #7, !dbg !40 + %6310 = insertelement <2 x bfloat> poison, bfloat %5661, i64 0, !dbg !40 + %6311 = insertelement <2 x bfloat> %6310, bfloat %5662, i64 1, !dbg !40 + %6312 = bitcast <2 x bfloat> %6311 to i32, !dbg !40 + %6313 = insertelement <2 x bfloat> poison, bfloat %5663, i64 0, !dbg !40 + %6314 = insertelement <2 x bfloat> %6313, bfloat %5664, i64 1, !dbg !40 + %6315 = bitcast <2 x bfloat> %6314 to i32, !dbg !40 + %6316 = insertelement <2 x bfloat> poison, bfloat %5665, i64 0, !dbg !40 + %6317 = insertelement <2 x bfloat> %6316, bfloat %5666, i64 1, !dbg !40 + %6318 = bitcast <2 x bfloat> %6317 to i32, !dbg !40 + %6319 = insertelement <2 x bfloat> poison, bfloat %5667, i64 0, !dbg !40 + %6320 = insertelement <2 x bfloat> %6319, bfloat %5668, i64 1, !dbg !40 + %6321 = bitcast <2 x bfloat> %6320 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6312, i32 %6315, i32 %6318, i32 %6321, ptr addrspace(1) %4890, i1 %645) #7, !dbg !40 + %6322 = insertelement <2 x bfloat> poison, bfloat %5669, i64 0, !dbg !40 + %6323 = insertelement <2 x bfloat> %6322, bfloat %5670, i64 1, !dbg !40 + %6324 = bitcast <2 x bfloat> %6323 to i32, !dbg !40 + %6325 = insertelement <2 x bfloat> poison, bfloat %5671, i64 0, !dbg !40 + %6326 = insertelement <2 x bfloat> %6325, bfloat %5672, i64 1, !dbg !40 + %6327 = bitcast <2 x bfloat> %6326 to i32, !dbg !40 + %6328 = insertelement <2 x bfloat> poison, bfloat %5673, i64 0, !dbg !40 + %6329 = insertelement <2 x bfloat> %6328, bfloat %5674, i64 1, !dbg !40 + %6330 = bitcast <2 x bfloat> %6329 to i32, !dbg !40 + %6331 = insertelement <2 x bfloat> poison, bfloat %5675, i64 0, !dbg !40 + %6332 = insertelement <2 x bfloat> %6331, bfloat %5676, i64 1, !dbg !40 + %6333 = bitcast <2 x bfloat> %6332 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6324, i32 %6327, i32 %6330, i32 %6333, ptr addrspace(1) %4888, i1 %663) #7, !dbg !40 + %6334 = insertelement <2 x bfloat> poison, bfloat %5677, i64 0, !dbg !40 + %6335 = insertelement <2 x bfloat> %6334, bfloat %5678, i64 1, !dbg !40 + %6336 = bitcast <2 x bfloat> %6335 to i32, !dbg !40 + %6337 = insertelement <2 x bfloat> poison, bfloat %5679, i64 0, !dbg !40 + %6338 = insertelement <2 x bfloat> %6337, bfloat %5680, i64 1, !dbg !40 + %6339 = bitcast <2 x bfloat> %6338 to i32, !dbg !40 + %6340 = insertelement <2 x bfloat> poison, bfloat %5681, i64 0, !dbg !40 + %6341 = insertelement <2 x bfloat> %6340, bfloat %5682, i64 1, !dbg !40 + %6342 = bitcast <2 x bfloat> %6341 to i32, !dbg !40 + %6343 = insertelement <2 x bfloat> poison, bfloat %5683, i64 0, !dbg !40 + %6344 = insertelement <2 x bfloat> %6343, bfloat %5684, i64 1, !dbg !40 + %6345 = bitcast <2 x bfloat> %6344 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6336, i32 %6339, i32 %6342, i32 %6345, ptr addrspace(1) %4886, i1 %681) #7, !dbg !40 + %6346 = insertelement <2 x bfloat> poison, bfloat %5685, i64 0, !dbg !40 + %6347 = insertelement <2 x bfloat> %6346, bfloat %5686, i64 1, !dbg !40 + %6348 = bitcast <2 x bfloat> %6347 to i32, !dbg !40 + %6349 = insertelement <2 x bfloat> poison, bfloat %5687, i64 0, !dbg !40 + %6350 = insertelement <2 x bfloat> %6349, bfloat %5688, i64 1, !dbg !40 + %6351 = bitcast <2 x bfloat> %6350 to i32, !dbg !40 + %6352 = insertelement <2 x bfloat> poison, bfloat %5689, i64 0, !dbg !40 + %6353 = insertelement <2 x bfloat> %6352, bfloat %5690, i64 1, !dbg !40 + %6354 = bitcast <2 x bfloat> %6353 to i32, !dbg !40 + %6355 = insertelement <2 x bfloat> poison, bfloat %5691, i64 0, !dbg !40 + %6356 = insertelement <2 x bfloat> %6355, bfloat %5692, i64 1, !dbg !40 + %6357 = bitcast <2 x bfloat> %6356 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6348, i32 %6351, i32 %6354, i32 %6357, ptr addrspace(1) %4884, i1 %699) #7, !dbg !40 + %6358 = insertelement <2 x bfloat> poison, bfloat %5693, i64 0, !dbg !40 + %6359 = insertelement <2 x bfloat> %6358, bfloat %5694, i64 1, !dbg !40 + %6360 = bitcast <2 x bfloat> %6359 to i32, !dbg !40 + %6361 = insertelement <2 x bfloat> poison, bfloat %5695, i64 0, !dbg !40 + %6362 = insertelement <2 x bfloat> %6361, bfloat %5696, i64 1, !dbg !40 + %6363 = bitcast <2 x bfloat> %6362 to i32, !dbg !40 + %6364 = insertelement <2 x bfloat> poison, bfloat %5697, i64 0, !dbg !40 + %6365 = insertelement <2 x bfloat> %6364, bfloat %5698, i64 1, !dbg !40 + %6366 = bitcast <2 x bfloat> %6365 to i32, !dbg !40 + %6367 = insertelement <2 x bfloat> poison, bfloat %5699, i64 0, !dbg !40 + %6368 = insertelement <2 x bfloat> %6367, bfloat %5700, i64 1, !dbg !40 + %6369 = bitcast <2 x bfloat> %6368 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6360, i32 %6363, i32 %6366, i32 %6369, ptr addrspace(1) %4882, i1 %717) #7, !dbg !40 + %6370 = insertelement <2 x bfloat> poison, bfloat %5701, i64 0, !dbg !40 + %6371 = insertelement <2 x bfloat> %6370, bfloat %5702, i64 1, !dbg !40 + %6372 = bitcast <2 x bfloat> %6371 to i32, !dbg !40 + %6373 = insertelement <2 x bfloat> poison, bfloat %5703, i64 0, !dbg !40 + %6374 = insertelement <2 x bfloat> %6373, bfloat %5704, i64 1, !dbg !40 + %6375 = bitcast <2 x bfloat> %6374 to i32, !dbg !40 + %6376 = insertelement <2 x bfloat> poison, bfloat %5705, i64 0, !dbg !40 + %6377 = insertelement <2 x bfloat> %6376, bfloat %5706, i64 1, !dbg !40 + %6378 = bitcast <2 x bfloat> %6377 to i32, !dbg !40 + %6379 = insertelement <2 x bfloat> poison, bfloat %5707, i64 0, !dbg !40 + %6380 = insertelement <2 x bfloat> %6379, bfloat %5708, i64 1, !dbg !40 + %6381 = bitcast <2 x bfloat> %6380 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6372, i32 %6375, i32 %6378, i32 %6381, ptr addrspace(1) %4880, i1 %735) #7, !dbg !40 + %6382 = insertelement <2 x bfloat> poison, bfloat %5709, i64 0, !dbg !40 + %6383 = insertelement <2 x bfloat> %6382, bfloat %5710, i64 1, !dbg !40 + %6384 = bitcast <2 x bfloat> %6383 to i32, !dbg !40 + %6385 = insertelement <2 x bfloat> poison, bfloat %5711, i64 0, !dbg !40 + %6386 = insertelement <2 x bfloat> %6385, bfloat %5712, i64 1, !dbg !40 + %6387 = bitcast <2 x bfloat> %6386 to i32, !dbg !40 + %6388 = insertelement <2 x bfloat> poison, bfloat %5713, i64 0, !dbg !40 + %6389 = insertelement <2 x bfloat> %6388, bfloat %5714, i64 1, !dbg !40 + %6390 = bitcast <2 x bfloat> %6389 to i32, !dbg !40 + %6391 = insertelement <2 x bfloat> poison, bfloat %5715, i64 0, !dbg !40 + %6392 = insertelement <2 x bfloat> %6391, bfloat %5716, i64 1, !dbg !40 + %6393 = bitcast <2 x bfloat> %6392 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6384, i32 %6387, i32 %6390, i32 %6393, ptr addrspace(1) %4878, i1 %753) #7, !dbg !40 + %6394 = insertelement <2 x bfloat> poison, bfloat %5717, i64 0, !dbg !40 + %6395 = insertelement <2 x bfloat> %6394, bfloat %5718, i64 1, !dbg !40 + %6396 = bitcast <2 x bfloat> %6395 to i32, !dbg !40 + %6397 = insertelement <2 x bfloat> poison, bfloat %5719, i64 0, !dbg !40 + %6398 = insertelement <2 x bfloat> %6397, bfloat %5720, i64 1, !dbg !40 + %6399 = bitcast <2 x bfloat> %6398 to i32, !dbg !40 + %6400 = insertelement <2 x bfloat> poison, bfloat %5721, i64 0, !dbg !40 + %6401 = insertelement <2 x bfloat> %6400, bfloat %5722, i64 1, !dbg !40 + %6402 = bitcast <2 x bfloat> %6401 to i32, !dbg !40 + %6403 = insertelement <2 x bfloat> poison, bfloat %5723, i64 0, !dbg !40 + %6404 = insertelement <2 x bfloat> %6403, bfloat %5724, i64 1, !dbg !40 + %6405 = bitcast <2 x bfloat> %6404 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6396, i32 %6399, i32 %6402, i32 %6405, ptr addrspace(1) %4876, i1 %771) #7, !dbg !40 + %6406 = insertelement <2 x bfloat> poison, bfloat %5725, i64 0, !dbg !40 + %6407 = insertelement <2 x bfloat> %6406, bfloat %5726, i64 1, !dbg !40 + %6408 = bitcast <2 x bfloat> %6407 to i32, !dbg !40 + %6409 = insertelement <2 x bfloat> poison, bfloat %5727, i64 0, !dbg !40 + %6410 = insertelement <2 x bfloat> %6409, bfloat %5728, i64 1, !dbg !40 + %6411 = bitcast <2 x bfloat> %6410 to i32, !dbg !40 + %6412 = insertelement <2 x bfloat> poison, bfloat %5729, i64 0, !dbg !40 + %6413 = insertelement <2 x bfloat> %6412, bfloat %5730, i64 1, !dbg !40 + %6414 = bitcast <2 x bfloat> %6413 to i32, !dbg !40 + %6415 = insertelement <2 x bfloat> poison, bfloat %5731, i64 0, !dbg !40 + %6416 = insertelement <2 x bfloat> %6415, bfloat %5732, i64 1, !dbg !40 + %6417 = bitcast <2 x bfloat> %6416 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6408, i32 %6411, i32 %6414, i32 %6417, ptr addrspace(1) %4874, i1 %789) #7, !dbg !40 + %6418 = insertelement <2 x bfloat> poison, bfloat %5733, i64 0, !dbg !40 + %6419 = insertelement <2 x bfloat> %6418, bfloat %5734, i64 1, !dbg !40 + %6420 = bitcast <2 x bfloat> %6419 to i32, !dbg !40 + %6421 = insertelement <2 x bfloat> poison, bfloat %5735, i64 0, !dbg !40 + %6422 = insertelement <2 x bfloat> %6421, bfloat %5736, i64 1, !dbg !40 + %6423 = bitcast <2 x bfloat> %6422 to i32, !dbg !40 + %6424 = insertelement <2 x bfloat> poison, bfloat %5737, i64 0, !dbg !40 + %6425 = insertelement <2 x bfloat> %6424, bfloat %5738, i64 1, !dbg !40 + %6426 = bitcast <2 x bfloat> %6425 to i32, !dbg !40 + %6427 = insertelement <2 x bfloat> poison, bfloat %5739, i64 0, !dbg !40 + %6428 = insertelement <2 x bfloat> %6427, bfloat %5740, i64 1, !dbg !40 + %6429 = bitcast <2 x bfloat> %6428 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6420, i32 %6423, i32 %6426, i32 %6429, ptr addrspace(1) %4872, i1 %807) #7, !dbg !40 + %6430 = insertelement <2 x bfloat> poison, bfloat %5741, i64 0, !dbg !40 + %6431 = insertelement <2 x bfloat> %6430, bfloat %5742, i64 1, !dbg !40 + %6432 = bitcast <2 x bfloat> %6431 to i32, !dbg !40 + %6433 = insertelement <2 x bfloat> poison, bfloat %5743, i64 0, !dbg !40 + %6434 = insertelement <2 x bfloat> %6433, bfloat %5744, i64 1, !dbg !40 + %6435 = bitcast <2 x bfloat> %6434 to i32, !dbg !40 + %6436 = insertelement <2 x bfloat> poison, bfloat %5745, i64 0, !dbg !40 + %6437 = insertelement <2 x bfloat> %6436, bfloat %5746, i64 1, !dbg !40 + %6438 = bitcast <2 x bfloat> %6437 to i32, !dbg !40 + %6439 = insertelement <2 x bfloat> poison, bfloat %5747, i64 0, !dbg !40 + %6440 = insertelement <2 x bfloat> %6439, bfloat %5748, i64 1, !dbg !40 + %6441 = bitcast <2 x bfloat> %6440 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6432, i32 %6435, i32 %6438, i32 %6441, ptr addrspace(1) %4870, i1 %825) #7, !dbg !40 + %6442 = insertelement <2 x bfloat> poison, bfloat %5749, i64 0, !dbg !40 + %6443 = insertelement <2 x bfloat> %6442, bfloat %5750, i64 1, !dbg !40 + %6444 = bitcast <2 x bfloat> %6443 to i32, !dbg !40 + %6445 = insertelement <2 x bfloat> poison, bfloat %5751, i64 0, !dbg !40 + %6446 = insertelement <2 x bfloat> %6445, bfloat %5752, i64 1, !dbg !40 + %6447 = bitcast <2 x bfloat> %6446 to i32, !dbg !40 + %6448 = insertelement <2 x bfloat> poison, bfloat %5753, i64 0, !dbg !40 + %6449 = insertelement <2 x bfloat> %6448, bfloat %5754, i64 1, !dbg !40 + %6450 = bitcast <2 x bfloat> %6449 to i32, !dbg !40 + %6451 = insertelement <2 x bfloat> poison, bfloat %5755, i64 0, !dbg !40 + %6452 = insertelement <2 x bfloat> %6451, bfloat %5756, i64 1, !dbg !40 + %6453 = bitcast <2 x bfloat> %6452 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6444, i32 %6447, i32 %6450, i32 %6453, ptr addrspace(1) %4868, i1 %843) #7, !dbg !40 + %6454 = insertelement <2 x bfloat> poison, bfloat %5757, i64 0, !dbg !40 + %6455 = insertelement <2 x bfloat> %6454, bfloat %5758, i64 1, !dbg !40 + %6456 = bitcast <2 x bfloat> %6455 to i32, !dbg !40 + %6457 = insertelement <2 x bfloat> poison, bfloat %5759, i64 0, !dbg !40 + %6458 = insertelement <2 x bfloat> %6457, bfloat %5760, i64 1, !dbg !40 + %6459 = bitcast <2 x bfloat> %6458 to i32, !dbg !40 + %6460 = insertelement <2 x bfloat> poison, bfloat %5761, i64 0, !dbg !40 + %6461 = insertelement <2 x bfloat> %6460, bfloat %5762, i64 1, !dbg !40 + %6462 = bitcast <2 x bfloat> %6461 to i32, !dbg !40 + %6463 = insertelement <2 x bfloat> poison, bfloat %5763, i64 0, !dbg !40 + %6464 = insertelement <2 x bfloat> %6463, bfloat %5764, i64 1, !dbg !40 + %6465 = bitcast <2 x bfloat> %6464 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6456, i32 %6459, i32 %6462, i32 %6465, ptr addrspace(1) %4866, i1 %861) #7, !dbg !40 + %6466 = insertelement <2 x bfloat> poison, bfloat %5765, i64 0, !dbg !40 + %6467 = insertelement <2 x bfloat> %6466, bfloat %5766, i64 1, !dbg !40 + %6468 = bitcast <2 x bfloat> %6467 to i32, !dbg !40 + %6469 = insertelement <2 x bfloat> poison, bfloat %5767, i64 0, !dbg !40 + %6470 = insertelement <2 x bfloat> %6469, bfloat %5768, i64 1, !dbg !40 + %6471 = bitcast <2 x bfloat> %6470 to i32, !dbg !40 + %6472 = insertelement <2 x bfloat> poison, bfloat %5769, i64 0, !dbg !40 + %6473 = insertelement <2 x bfloat> %6472, bfloat %5770, i64 1, !dbg !40 + %6474 = bitcast <2 x bfloat> %6473 to i32, !dbg !40 + %6475 = insertelement <2 x bfloat> poison, bfloat %5771, i64 0, !dbg !40 + %6476 = insertelement <2 x bfloat> %6475, bfloat %5772, i64 1, !dbg !40 + %6477 = bitcast <2 x bfloat> %6476 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6468, i32 %6471, i32 %6474, i32 %6477, ptr addrspace(1) %4864, i1 %879) #7, !dbg !40 + %6478 = insertelement <2 x bfloat> poison, bfloat %5773, i64 0, !dbg !40 + %6479 = insertelement <2 x bfloat> %6478, bfloat %5774, i64 1, !dbg !40 + %6480 = bitcast <2 x bfloat> %6479 to i32, !dbg !40 + %6481 = insertelement <2 x bfloat> poison, bfloat %5775, i64 0, !dbg !40 + %6482 = insertelement <2 x bfloat> %6481, bfloat %5776, i64 1, !dbg !40 + %6483 = bitcast <2 x bfloat> %6482 to i32, !dbg !40 + %6484 = insertelement <2 x bfloat> poison, bfloat %5777, i64 0, !dbg !40 + %6485 = insertelement <2 x bfloat> %6484, bfloat %5778, i64 1, !dbg !40 + %6486 = bitcast <2 x bfloat> %6485 to i32, !dbg !40 + %6487 = insertelement <2 x bfloat> poison, bfloat %5779, i64 0, !dbg !40 + %6488 = insertelement <2 x bfloat> %6487, bfloat %5780, i64 1, !dbg !40 + %6489 = bitcast <2 x bfloat> %6488 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6480, i32 %6483, i32 %6486, i32 %6489, ptr addrspace(1) %4862, i1 %897) #7, !dbg !40 + %6490 = insertelement <2 x bfloat> poison, bfloat %5781, i64 0, !dbg !40 + %6491 = insertelement <2 x bfloat> %6490, bfloat %5782, i64 1, !dbg !40 + %6492 = bitcast <2 x bfloat> %6491 to i32, !dbg !40 + %6493 = insertelement <2 x bfloat> poison, bfloat %5783, i64 0, !dbg !40 + %6494 = insertelement <2 x bfloat> %6493, bfloat %5784, i64 1, !dbg !40 + %6495 = bitcast <2 x bfloat> %6494 to i32, !dbg !40 + %6496 = insertelement <2 x bfloat> poison, bfloat %5785, i64 0, !dbg !40 + %6497 = insertelement <2 x bfloat> %6496, bfloat %5786, i64 1, !dbg !40 + %6498 = bitcast <2 x bfloat> %6497 to i32, !dbg !40 + %6499 = insertelement <2 x bfloat> poison, bfloat %5787, i64 0, !dbg !40 + %6500 = insertelement <2 x bfloat> %6499, bfloat %5788, i64 1, !dbg !40 + %6501 = bitcast <2 x bfloat> %6500 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6492, i32 %6495, i32 %6498, i32 %6501, ptr addrspace(1) %4860, i1 %915) #7, !dbg !40 + %6502 = insertelement <2 x bfloat> poison, bfloat %5789, i64 0, !dbg !40 + %6503 = insertelement <2 x bfloat> %6502, bfloat %5790, i64 1, !dbg !40 + %6504 = bitcast <2 x bfloat> %6503 to i32, !dbg !40 + %6505 = insertelement <2 x bfloat> poison, bfloat %5791, i64 0, !dbg !40 + %6506 = insertelement <2 x bfloat> %6505, bfloat %5792, i64 1, !dbg !40 + %6507 = bitcast <2 x bfloat> %6506 to i32, !dbg !40 + %6508 = insertelement <2 x bfloat> poison, bfloat %5793, i64 0, !dbg !40 + %6509 = insertelement <2 x bfloat> %6508, bfloat %5794, i64 1, !dbg !40 + %6510 = bitcast <2 x bfloat> %6509 to i32, !dbg !40 + %6511 = insertelement <2 x bfloat> poison, bfloat %5795, i64 0, !dbg !40 + %6512 = insertelement <2 x bfloat> %6511, bfloat %5796, i64 1, !dbg !40 + %6513 = bitcast <2 x bfloat> %6512 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6504, i32 %6507, i32 %6510, i32 %6513, ptr addrspace(1) %4858, i1 %933) #7, !dbg !40 + %6514 = insertelement <2 x bfloat> poison, bfloat %5797, i64 0, !dbg !40 + %6515 = insertelement <2 x bfloat> %6514, bfloat %5798, i64 1, !dbg !40 + %6516 = bitcast <2 x bfloat> %6515 to i32, !dbg !40 + %6517 = insertelement <2 x bfloat> poison, bfloat %5799, i64 0, !dbg !40 + %6518 = insertelement <2 x bfloat> %6517, bfloat %5800, i64 1, !dbg !40 + %6519 = bitcast <2 x bfloat> %6518 to i32, !dbg !40 + %6520 = insertelement <2 x bfloat> poison, bfloat %5801, i64 0, !dbg !40 + %6521 = insertelement <2 x bfloat> %6520, bfloat %5802, i64 1, !dbg !40 + %6522 = bitcast <2 x bfloat> %6521 to i32, !dbg !40 + %6523 = insertelement <2 x bfloat> poison, bfloat %5803, i64 0, !dbg !40 + %6524 = insertelement <2 x bfloat> %6523, bfloat %5804, i64 1, !dbg !40 + %6525 = bitcast <2 x bfloat> %6524 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6516, i32 %6519, i32 %6522, i32 %6525, ptr addrspace(1) %4856, i1 %951) #7, !dbg !40 + %6526 = insertelement <2 x bfloat> poison, bfloat %5805, i64 0, !dbg !40 + %6527 = insertelement <2 x bfloat> %6526, bfloat %5806, i64 1, !dbg !40 + %6528 = bitcast <2 x bfloat> %6527 to i32, !dbg !40 + %6529 = insertelement <2 x bfloat> poison, bfloat %5807, i64 0, !dbg !40 + %6530 = insertelement <2 x bfloat> %6529, bfloat %5808, i64 1, !dbg !40 + %6531 = bitcast <2 x bfloat> %6530 to i32, !dbg !40 + %6532 = insertelement <2 x bfloat> poison, bfloat %5809, i64 0, !dbg !40 + %6533 = insertelement <2 x bfloat> %6532, bfloat %5810, i64 1, !dbg !40 + %6534 = bitcast <2 x bfloat> %6533 to i32, !dbg !40 + %6535 = insertelement <2 x bfloat> poison, bfloat %5811, i64 0, !dbg !40 + %6536 = insertelement <2 x bfloat> %6535, bfloat %5812, i64 1, !dbg !40 + %6537 = bitcast <2 x bfloat> %6536 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6528, i32 %6531, i32 %6534, i32 %6537, ptr addrspace(1) %4854, i1 %969) #7, !dbg !40 + %6538 = insertelement <2 x bfloat> poison, bfloat %5813, i64 0, !dbg !40 + %6539 = insertelement <2 x bfloat> %6538, bfloat %5814, i64 1, !dbg !40 + %6540 = bitcast <2 x bfloat> %6539 to i32, !dbg !40 + %6541 = insertelement <2 x bfloat> poison, bfloat %5815, i64 0, !dbg !40 + %6542 = insertelement <2 x bfloat> %6541, bfloat %5816, i64 1, !dbg !40 + %6543 = bitcast <2 x bfloat> %6542 to i32, !dbg !40 + %6544 = insertelement <2 x bfloat> poison, bfloat %5817, i64 0, !dbg !40 + %6545 = insertelement <2 x bfloat> %6544, bfloat %5818, i64 1, !dbg !40 + %6546 = bitcast <2 x bfloat> %6545 to i32, !dbg !40 + %6547 = insertelement <2 x bfloat> poison, bfloat %5819, i64 0, !dbg !40 + %6548 = insertelement <2 x bfloat> %6547, bfloat %5820, i64 1, !dbg !40 + %6549 = bitcast <2 x bfloat> %6548 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6540, i32 %6543, i32 %6546, i32 %6549, ptr addrspace(1) %4852, i1 %987) #7, !dbg !40 + %6550 = insertelement <2 x bfloat> poison, bfloat %5821, i64 0, !dbg !40 + %6551 = insertelement <2 x bfloat> %6550, bfloat %5822, i64 1, !dbg !40 + %6552 = bitcast <2 x bfloat> %6551 to i32, !dbg !40 + %6553 = insertelement <2 x bfloat> poison, bfloat %5823, i64 0, !dbg !40 + %6554 = insertelement <2 x bfloat> %6553, bfloat %5824, i64 1, !dbg !40 + %6555 = bitcast <2 x bfloat> %6554 to i32, !dbg !40 + %6556 = insertelement <2 x bfloat> poison, bfloat %5825, i64 0, !dbg !40 + %6557 = insertelement <2 x bfloat> %6556, bfloat %5826, i64 1, !dbg !40 + %6558 = bitcast <2 x bfloat> %6557 to i32, !dbg !40 + %6559 = insertelement <2 x bfloat> poison, bfloat %5827, i64 0, !dbg !40 + %6560 = insertelement <2 x bfloat> %6559, bfloat %5828, i64 1, !dbg !40 + %6561 = bitcast <2 x bfloat> %6560 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6552, i32 %6555, i32 %6558, i32 %6561, ptr addrspace(1) %4850, i1 %1005) #7, !dbg !40 + %6562 = insertelement <2 x bfloat> poison, bfloat %5829, i64 0, !dbg !40 + %6563 = insertelement <2 x bfloat> %6562, bfloat %5830, i64 1, !dbg !40 + %6564 = bitcast <2 x bfloat> %6563 to i32, !dbg !40 + %6565 = insertelement <2 x bfloat> poison, bfloat %5831, i64 0, !dbg !40 + %6566 = insertelement <2 x bfloat> %6565, bfloat %5832, i64 1, !dbg !40 + %6567 = bitcast <2 x bfloat> %6566 to i32, !dbg !40 + %6568 = insertelement <2 x bfloat> poison, bfloat %5833, i64 0, !dbg !40 + %6569 = insertelement <2 x bfloat> %6568, bfloat %5834, i64 1, !dbg !40 + %6570 = bitcast <2 x bfloat> %6569 to i32, !dbg !40 + %6571 = insertelement <2 x bfloat> poison, bfloat %5835, i64 0, !dbg !40 + %6572 = insertelement <2 x bfloat> %6571, bfloat %5836, i64 1, !dbg !40 + %6573 = bitcast <2 x bfloat> %6572 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6564, i32 %6567, i32 %6570, i32 %6573, ptr addrspace(1) %4848, i1 %1023) #7, !dbg !40 + %6574 = insertelement <2 x bfloat> poison, bfloat %5837, i64 0, !dbg !40 + %6575 = insertelement <2 x bfloat> %6574, bfloat %5838, i64 1, !dbg !40 + %6576 = bitcast <2 x bfloat> %6575 to i32, !dbg !40 + %6577 = insertelement <2 x bfloat> poison, bfloat %5839, i64 0, !dbg !40 + %6578 = insertelement <2 x bfloat> %6577, bfloat %5840, i64 1, !dbg !40 + %6579 = bitcast <2 x bfloat> %6578 to i32, !dbg !40 + %6580 = insertelement <2 x bfloat> poison, bfloat %5841, i64 0, !dbg !40 + %6581 = insertelement <2 x bfloat> %6580, bfloat %5842, i64 1, !dbg !40 + %6582 = bitcast <2 x bfloat> %6581 to i32, !dbg !40 + %6583 = insertelement <2 x bfloat> poison, bfloat %5843, i64 0, !dbg !40 + %6584 = insertelement <2 x bfloat> %6583, bfloat %5844, i64 1, !dbg !40 + %6585 = bitcast <2 x bfloat> %6584 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6576, i32 %6579, i32 %6582, i32 %6585, ptr addrspace(1) %4846, i1 %1041) #7, !dbg !40 + %6586 = insertelement <2 x bfloat> poison, bfloat %5845, i64 0, !dbg !40 + %6587 = insertelement <2 x bfloat> %6586, bfloat %5846, i64 1, !dbg !40 + %6588 = bitcast <2 x bfloat> %6587 to i32, !dbg !40 + %6589 = insertelement <2 x bfloat> poison, bfloat %5847, i64 0, !dbg !40 + %6590 = insertelement <2 x bfloat> %6589, bfloat %5848, i64 1, !dbg !40 + %6591 = bitcast <2 x bfloat> %6590 to i32, !dbg !40 + %6592 = insertelement <2 x bfloat> poison, bfloat %5849, i64 0, !dbg !40 + %6593 = insertelement <2 x bfloat> %6592, bfloat %5850, i64 1, !dbg !40 + %6594 = bitcast <2 x bfloat> %6593 to i32, !dbg !40 + %6595 = insertelement <2 x bfloat> poison, bfloat %5851, i64 0, !dbg !40 + %6596 = insertelement <2 x bfloat> %6595, bfloat %5852, i64 1, !dbg !40 + %6597 = bitcast <2 x bfloat> %6596 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6588, i32 %6591, i32 %6594, i32 %6597, ptr addrspace(1) %4844, i1 %1059) #7, !dbg !40 + %6598 = insertelement <2 x bfloat> poison, bfloat %5853, i64 0, !dbg !40 + %6599 = insertelement <2 x bfloat> %6598, bfloat %5854, i64 1, !dbg !40 + %6600 = bitcast <2 x bfloat> %6599 to i32, !dbg !40 + %6601 = insertelement <2 x bfloat> poison, bfloat %5855, i64 0, !dbg !40 + %6602 = insertelement <2 x bfloat> %6601, bfloat %5856, i64 1, !dbg !40 + %6603 = bitcast <2 x bfloat> %6602 to i32, !dbg !40 + %6604 = insertelement <2 x bfloat> poison, bfloat %5857, i64 0, !dbg !40 + %6605 = insertelement <2 x bfloat> %6604, bfloat %5858, i64 1, !dbg !40 + %6606 = bitcast <2 x bfloat> %6605 to i32, !dbg !40 + %6607 = insertelement <2 x bfloat> poison, bfloat %5859, i64 0, !dbg !40 + %6608 = insertelement <2 x bfloat> %6607, bfloat %5860, i64 1, !dbg !40 + %6609 = bitcast <2 x bfloat> %6608 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6600, i32 %6603, i32 %6606, i32 %6609, ptr addrspace(1) %4842, i1 %1077) #7, !dbg !40 + %6610 = insertelement <2 x bfloat> poison, bfloat %5861, i64 0, !dbg !40 + %6611 = insertelement <2 x bfloat> %6610, bfloat %5862, i64 1, !dbg !40 + %6612 = bitcast <2 x bfloat> %6611 to i32, !dbg !40 + %6613 = insertelement <2 x bfloat> poison, bfloat %5863, i64 0, !dbg !40 + %6614 = insertelement <2 x bfloat> %6613, bfloat %5864, i64 1, !dbg !40 + %6615 = bitcast <2 x bfloat> %6614 to i32, !dbg !40 + %6616 = insertelement <2 x bfloat> poison, bfloat %5865, i64 0, !dbg !40 + %6617 = insertelement <2 x bfloat> %6616, bfloat %5866, i64 1, !dbg !40 + %6618 = bitcast <2 x bfloat> %6617 to i32, !dbg !40 + %6619 = insertelement <2 x bfloat> poison, bfloat %5867, i64 0, !dbg !40 + %6620 = insertelement <2 x bfloat> %6619, bfloat %5868, i64 1, !dbg !40 + %6621 = bitcast <2 x bfloat> %6620 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6612, i32 %6615, i32 %6618, i32 %6621, ptr addrspace(1) %4840, i1 %1095) #7, !dbg !40 + %6622 = insertelement <2 x bfloat> poison, bfloat %5869, i64 0, !dbg !40 + %6623 = insertelement <2 x bfloat> %6622, bfloat %5870, i64 1, !dbg !40 + %6624 = bitcast <2 x bfloat> %6623 to i32, !dbg !40 + %6625 = insertelement <2 x bfloat> poison, bfloat %5871, i64 0, !dbg !40 + %6626 = insertelement <2 x bfloat> %6625, bfloat %5872, i64 1, !dbg !40 + %6627 = bitcast <2 x bfloat> %6626 to i32, !dbg !40 + %6628 = insertelement <2 x bfloat> poison, bfloat %5873, i64 0, !dbg !40 + %6629 = insertelement <2 x bfloat> %6628, bfloat %5874, i64 1, !dbg !40 + %6630 = bitcast <2 x bfloat> %6629 to i32, !dbg !40 + %6631 = insertelement <2 x bfloat> poison, bfloat %5875, i64 0, !dbg !40 + %6632 = insertelement <2 x bfloat> %6631, bfloat %5876, i64 1, !dbg !40 + %6633 = bitcast <2 x bfloat> %6632 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6624, i32 %6627, i32 %6630, i32 %6633, ptr addrspace(1) %4838, i1 %1113) #7, !dbg !40 + %6634 = insertelement <2 x bfloat> poison, bfloat %5877, i64 0, !dbg !40 + %6635 = insertelement <2 x bfloat> %6634, bfloat %5878, i64 1, !dbg !40 + %6636 = bitcast <2 x bfloat> %6635 to i32, !dbg !40 + %6637 = insertelement <2 x bfloat> poison, bfloat %5879, i64 0, !dbg !40 + %6638 = insertelement <2 x bfloat> %6637, bfloat %5880, i64 1, !dbg !40 + %6639 = bitcast <2 x bfloat> %6638 to i32, !dbg !40 + %6640 = insertelement <2 x bfloat> poison, bfloat %5881, i64 0, !dbg !40 + %6641 = insertelement <2 x bfloat> %6640, bfloat %5882, i64 1, !dbg !40 + %6642 = bitcast <2 x bfloat> %6641 to i32, !dbg !40 + %6643 = insertelement <2 x bfloat> poison, bfloat %5883, i64 0, !dbg !40 + %6644 = insertelement <2 x bfloat> %6643, bfloat %5884, i64 1, !dbg !40 + %6645 = bitcast <2 x bfloat> %6644 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6636, i32 %6639, i32 %6642, i32 %6645, ptr addrspace(1) %4836, i1 %1131) #7, !dbg !40 + %6646 = insertelement <2 x bfloat> poison, bfloat %5885, i64 0, !dbg !40 + %6647 = insertelement <2 x bfloat> %6646, bfloat %5886, i64 1, !dbg !40 + %6648 = bitcast <2 x bfloat> %6647 to i32, !dbg !40 + %6649 = insertelement <2 x bfloat> poison, bfloat %5887, i64 0, !dbg !40 + %6650 = insertelement <2 x bfloat> %6649, bfloat %5888, i64 1, !dbg !40 + %6651 = bitcast <2 x bfloat> %6650 to i32, !dbg !40 + %6652 = insertelement <2 x bfloat> poison, bfloat %5889, i64 0, !dbg !40 + %6653 = insertelement <2 x bfloat> %6652, bfloat %5890, i64 1, !dbg !40 + %6654 = bitcast <2 x bfloat> %6653 to i32, !dbg !40 + %6655 = insertelement <2 x bfloat> poison, bfloat %5891, i64 0, !dbg !40 + %6656 = insertelement <2 x bfloat> %6655, bfloat %5892, i64 1, !dbg !40 + %6657 = bitcast <2 x bfloat> %6656 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6648, i32 %6651, i32 %6654, i32 %6657, ptr addrspace(1) %4834, i1 %1149) #7, !dbg !40 + %6658 = bitcast <2 x bfloat> %5895 to i32, !dbg !40 + %6659 = insertelement <2 x bfloat> poison, bfloat %5896, i64 0, !dbg !40 + %6660 = insertelement <2 x bfloat> %6659, bfloat %5897, i64 1, !dbg !40 + %6661 = bitcast <2 x bfloat> %6660 to i32, !dbg !40 + %6662 = insertelement <2 x bfloat> poison, bfloat %5898, i64 0, !dbg !40 + %6663 = insertelement <2 x bfloat> %6662, bfloat %5899, i64 1, !dbg !40 + %6664 = bitcast <2 x bfloat> %6663 to i32, !dbg !40 + %6665 = insertelement <2 x bfloat> poison, bfloat %5900, i64 0, !dbg !40 + %6666 = insertelement <2 x bfloat> %6665, bfloat %5901, i64 1, !dbg !40 + %6667 = bitcast <2 x bfloat> %6666 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6658, i32 %6661, i32 %6664, i32 %6667, ptr addrspace(1) %4832, i1 %1167) #7, !dbg !40 + %6668 = bitcast <2 x bfloat> %5904 to i32, !dbg !40 + %6669 = bitcast <2 x bfloat> %5907 to i32, !dbg !40 + %6670 = bitcast <2 x bfloat> %5910 to i32, !dbg !40 + %6671 = bitcast <2 x bfloat> %5913 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6668, i32 %6669, i32 %6670, i32 %6671, ptr addrspace(1) %4830, i1 %1185) #7, !dbg !40 + %6672 = bitcast <2 x bfloat> %5916 to i32, !dbg !40 + %6673 = bitcast <2 x bfloat> %5919 to i32, !dbg !40 + %6674 = bitcast <2 x bfloat> %5922 to i32, !dbg !40 + %6675 = bitcast <2 x bfloat> %5925 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6672, i32 %6673, i32 %6674, i32 %6675, ptr addrspace(1) %4828, i1 %1203) #7, !dbg !40 + %6676 = bitcast <2 x bfloat> %5928 to i32, !dbg !40 + %6677 = bitcast <2 x bfloat> %5931 to i32, !dbg !40 + %6678 = bitcast <2 x bfloat> %5934 to i32, !dbg !40 + %6679 = bitcast <2 x bfloat> %5937 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6676, i32 %6677, i32 %6678, i32 %6679, ptr addrspace(1) %4826, i1 %1221) #7, !dbg !40 + %6680 = bitcast <2 x bfloat> %5940 to i32, !dbg !40 + %6681 = bitcast <2 x bfloat> %5943 to i32, !dbg !40 + %6682 = bitcast <2 x bfloat> %5946 to i32, !dbg !40 + %6683 = bitcast <2 x bfloat> %5949 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6680, i32 %6681, i32 %6682, i32 %6683, ptr addrspace(1) %4824, i1 %1239) #7, !dbg !40 + %6684 = bitcast <2 x bfloat> %5952 to i32, !dbg !40 + %6685 = bitcast <2 x bfloat> %5955 to i32, !dbg !40 + %6686 = bitcast <2 x bfloat> %5958 to i32, !dbg !40 + %6687 = bitcast <2 x bfloat> %5961 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6684, i32 %6685, i32 %6686, i32 %6687, ptr addrspace(1) %4822, i1 %1257) #7, !dbg !40 + %6688 = bitcast <2 x bfloat> %5964 to i32, !dbg !40 + %6689 = bitcast <2 x bfloat> %5967 to i32, !dbg !40 + %6690 = bitcast <2 x bfloat> %5970 to i32, !dbg !40 + %6691 = bitcast <2 x bfloat> %5973 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6688, i32 %6689, i32 %6690, i32 %6691, ptr addrspace(1) %4820, i1 %1275) #7, !dbg !40 + %6692 = bitcast <2 x bfloat> %5976 to i32, !dbg !40 + %6693 = bitcast <2 x bfloat> %5979 to i32, !dbg !40 + %6694 = bitcast <2 x bfloat> %5982 to i32, !dbg !40 + %6695 = bitcast <2 x bfloat> %5985 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6692, i32 %6693, i32 %6694, i32 %6695, ptr addrspace(1) %4818, i1 %1293) #7, !dbg !40 + %6696 = bitcast <2 x bfloat> %5988 to i32, !dbg !40 + %6697 = bitcast <2 x bfloat> %5991 to i32, !dbg !40 + %6698 = bitcast <2 x bfloat> %5994 to i32, !dbg !40 + %6699 = bitcast <2 x bfloat> %5997 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6696, i32 %6697, i32 %6698, i32 %6699, ptr addrspace(1) %4816, i1 %1311) #7, !dbg !40 + %6700 = bitcast <2 x bfloat> %6000 to i32, !dbg !40 + %6701 = bitcast <2 x bfloat> %6003 to i32, !dbg !40 + %6702 = bitcast <2 x bfloat> %6006 to i32, !dbg !40 + %6703 = bitcast <2 x bfloat> %6009 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6700, i32 %6701, i32 %6702, i32 %6703, ptr addrspace(1) %4814, i1 %1329) #7, !dbg !40 + %6704 = bitcast <2 x bfloat> %6012 to i32, !dbg !40 + %6705 = bitcast <2 x bfloat> %6015 to i32, !dbg !40 + %6706 = bitcast <2 x bfloat> %6018 to i32, !dbg !40 + %6707 = bitcast <2 x bfloat> %6021 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6704, i32 %6705, i32 %6706, i32 %6707, ptr addrspace(1) %4812, i1 %1347) #7, !dbg !40 + %6708 = bitcast <2 x bfloat> %6024 to i32, !dbg !40 + %6709 = bitcast <2 x bfloat> %6027 to i32, !dbg !40 + %6710 = bitcast <2 x bfloat> %6030 to i32, !dbg !40 + %6711 = bitcast <2 x bfloat> %6033 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6708, i32 %6709, i32 %6710, i32 %6711, ptr addrspace(1) %4810, i1 %1365) #7, !dbg !40 + %6712 = bitcast <2 x bfloat> %6036 to i32, !dbg !40 + %6713 = bitcast <2 x bfloat> %6039 to i32, !dbg !40 + %6714 = bitcast <2 x bfloat> %6042 to i32, !dbg !40 + %6715 = bitcast <2 x bfloat> %6045 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6712, i32 %6713, i32 %6714, i32 %6715, ptr addrspace(1) %4808, i1 %1383) #7, !dbg !40 + %6716 = bitcast <2 x bfloat> %6048 to i32, !dbg !40 + %6717 = bitcast <2 x bfloat> %6051 to i32, !dbg !40 + %6718 = bitcast <2 x bfloat> %6054 to i32, !dbg !40 + %6719 = bitcast <2 x bfloat> %6057 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6716, i32 %6717, i32 %6718, i32 %6719, ptr addrspace(1) %4806, i1 %1401) #7, !dbg !40 + %6720 = bitcast <2 x bfloat> %6060 to i32, !dbg !40 + %6721 = bitcast <2 x bfloat> %6063 to i32, !dbg !40 + %6722 = bitcast <2 x bfloat> %6066 to i32, !dbg !40 + %6723 = bitcast <2 x bfloat> %6069 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6720, i32 %6721, i32 %6722, i32 %6723, ptr addrspace(1) %4804, i1 %1419) #7, !dbg !40 + %6724 = bitcast <2 x bfloat> %6072 to i32, !dbg !40 + %6725 = bitcast <2 x bfloat> %6075 to i32, !dbg !40 + %6726 = bitcast <2 x bfloat> %6078 to i32, !dbg !40 + %6727 = bitcast <2 x bfloat> %6081 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %6724, i32 %6725, i32 %6726, i32 %6727, ptr addrspace(1) %4802, i1 %1437) #7, !dbg !40 + %6728 = shl nuw nsw i32 %12, 3, !dbg !41 + %6729 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %6728, !dbg !41 + %6730 = bitcast float %4933 to i32, !dbg !41 + %6731 = bitcast float %4935 to i32, !dbg !41 + %6732 = bitcast float %4937 to i32, !dbg !41 + %6733 = bitcast float %4939 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %6729, i32 %6730, i32 %6731, i32 %6732, i32 %6733) #7, !dbg !41 + %6734 = getelementptr inbounds nuw i8, ptr addrspace(3) %6729, i32 16, !dbg !41 + %6735 = bitcast float %4941 to i32, !dbg !41 + %6736 = bitcast float %4943 to i32, !dbg !41 + %6737 = bitcast float %4945 to i32, !dbg !41 + %6738 = bitcast float %4947 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %6734, i32 %6735, i32 %6736, i32 %6737, i32 %6738) #7, !dbg !41 + %6739 = getelementptr inbounds nuw i8, ptr addrspace(3) %6729, i32 32, !dbg !41 + %6740 = bitcast float %4949 to i32, !dbg !41 + %6741 = bitcast float %4951 to i32, !dbg !41 + %6742 = bitcast float %4953 to i32, !dbg !41 + %6743 = bitcast float %4955 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %6739, i32 %6740, i32 %6741, i32 %6742, i32 %6743) #7, !dbg !41 + %6744 = getelementptr inbounds nuw i8, ptr addrspace(3) %6729, i32 48, !dbg !41 + %6745 = bitcast float %4957 to i32, !dbg !41 + %6746 = bitcast float %4959 to i32, !dbg !41 + %6747 = bitcast float %4961 to i32, !dbg !41 + %6748 = bitcast float %4963 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %6744, i32 %6745, i32 %6746, i32 %6747, i32 %6748) #7, !dbg !41 + %6749 = getelementptr inbounds nuw i8, ptr addrspace(3) %6729, i32 64, !dbg !41 + %6750 = bitcast float %4965 to i32, !dbg !41 + %6751 = bitcast float %4967 to i32, !dbg !41 + %6752 = bitcast float %4969 to i32, !dbg !41 + %6753 = bitcast float %4971 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %6749, i32 %6750, i32 %6751, i32 %6752, i32 %6753) #7, !dbg !41 + %6754 = getelementptr inbounds nuw i8, ptr addrspace(3) %6729, i32 80, !dbg !41 + %6755 = bitcast float %4973 to i32, !dbg !41 + %6756 = bitcast float %4975 to i32, !dbg !41 + %6757 = bitcast float %4977 to i32, !dbg !41 + %6758 = bitcast float %4979 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %6754, i32 %6755, i32 %6756, i32 %6757, i32 %6758) #7, !dbg !41 + %6759 = getelementptr inbounds nuw i8, ptr addrspace(3) %6729, i32 96, !dbg !41 + %6760 = bitcast float %4981 to i32, !dbg !41 + %6761 = bitcast float %4983 to i32, !dbg !41 + %6762 = bitcast float %4985 to i32, !dbg !41 + %6763 = bitcast float %4987 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %6759, i32 %6760, i32 %6761, i32 %6762, i32 %6763) #7, !dbg !41 + %6764 = getelementptr inbounds nuw i8, ptr addrspace(3) %6729, i32 112, !dbg !41 + %6765 = bitcast float %4989 to i32, !dbg !41 + %6766 = bitcast float %4991 to i32, !dbg !41 + %6767 = bitcast float %4993 to i32, !dbg !41 + %6768 = bitcast float %4995 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %6764, i32 %6765, i32 %6766, i32 %6767, i32 %6768) #7, !dbg !41 + tail call void @llvm.nvvm.bar.warp.sync(i32 -1), !dbg !41 + %6769 = shl nuw nsw i32 %44, 4, !dbg !41 + %6770 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %6769, !dbg !41 + %6771 = tail call { i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.b16.p3(ptr addrspace(3) %6770), !dbg !41 + %6772 = extractvalue { i32, i32 } %6771, 0, !dbg !41 + %6773 = extractvalue { i32, i32 } %6771, 1, !dbg !41 + tail call void @llvm.nvvm.bar.warp.sync(i32 -1), !dbg !41 + %6774 = bitcast float %4934 to i32, !dbg !41 + %6775 = bitcast float %4936 to i32, !dbg !41 + %6776 = bitcast float %4938 to i32, !dbg !41 + %6777 = bitcast float %4940 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %6729, i32 %6774, i32 %6775, i32 %6776, i32 %6777) #7, !dbg !41 + %6778 = bitcast float %4942 to i32, !dbg !41 + %6779 = bitcast float %4944 to i32, !dbg !41 + %6780 = bitcast float %4946 to i32, !dbg !41 + %6781 = bitcast float %4948 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %6734, i32 %6778, i32 %6779, i32 %6780, i32 %6781) #7, !dbg !41 + %6782 = bitcast float %4950 to i32, !dbg !41 + %6783 = bitcast float %4952 to i32, !dbg !41 + %6784 = bitcast float %4954 to i32, !dbg !41 + %6785 = bitcast float %4956 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %6739, i32 %6782, i32 %6783, i32 %6784, i32 %6785) #7, !dbg !41 + %6786 = bitcast float %4958 to i32, !dbg !41 + %6787 = bitcast float %4960 to i32, !dbg !41 + %6788 = bitcast float %4962 to i32, !dbg !41 + %6789 = bitcast float %4964 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %6744, i32 %6786, i32 %6787, i32 %6788, i32 %6789) #7, !dbg !41 + %6790 = bitcast float %4966 to i32, !dbg !41 + %6791 = bitcast float %4968 to i32, !dbg !41 + %6792 = bitcast float %4970 to i32, !dbg !41 + %6793 = bitcast float %4972 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %6749, i32 %6790, i32 %6791, i32 %6792, i32 %6793) #7, !dbg !41 + %6794 = bitcast float %4974 to i32, !dbg !41 + %6795 = bitcast float %4976 to i32, !dbg !41 + %6796 = bitcast float %4978 to i32, !dbg !41 + %6797 = bitcast float %4980 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %6754, i32 %6794, i32 %6795, i32 %6796, i32 %6797) #7, !dbg !41 + %6798 = bitcast float %4982 to i32, !dbg !41 + %6799 = bitcast float %4984 to i32, !dbg !41 + %6800 = bitcast float %4986 to i32, !dbg !41 + %6801 = bitcast float %4988 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %6759, i32 %6798, i32 %6799, i32 %6800, i32 %6801) #7, !dbg !41 + %6802 = bitcast float %4990 to i32, !dbg !41 + %6803 = bitcast float %4992 to i32, !dbg !41 + %6804 = bitcast float %4994 to i32, !dbg !41 + %6805 = bitcast float %4996 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %6764, i32 %6802, i32 %6803, i32 %6804, i32 %6805) #7, !dbg !41 + tail call void @llvm.nvvm.bar.warp.sync(i32 -1), !dbg !41 + %6806 = tail call { i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.b16.p3(ptr addrspace(3) %6770), !dbg !41 + %6807 = extractvalue { i32, i32 } %6806, 0, !dbg !41 + %6808 = extractvalue { i32, i32 } %6806, 1, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %6772, ptr addrspace(1) %4800, i1 %4932) #7, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %6773, ptr addrspace(1) %4797, i1 %4931) #7, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %6807, ptr addrspace(1) %4793, i1 %4930) #7, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %6808, ptr addrspace(1) %4789, i1 %4929) #7, !dbg !41 + ret void, !dbg !42 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.div.full(float, float) #3 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.bar.warp.sync(i32) #4 + +; Function Attrs: convergent nocallback nofree nounwind memory(argmem: read) +declare { i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.b16.p3(ptr addrspace(3) readonly captures(none)) #5 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #6 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.f(float) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="32" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) } +attributes #4 = { convergent nocallback nounwind } +attributes #5 = { convergent nocallback nofree nounwind memory(argmem: read) } +attributes #6 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #7 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "l2norm_fwd_kernel", linkageName: "l2norm_fwd_kernel", scope: !1, file: !1, line: 69, type: !6, scopeLine: 69, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !10, !11, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 70, column: 24, scope: !5) +!14 = !DILocation(line: 70, column: 30, scope: !5) +!15 = !DILocation(line: 71, column: 16, scope: !5) +!16 = !DILocation(line: 71, column: 34, scope: !5) +!17 = !DILocation(line: 71, column: 21, scope: !5) +!18 = !DILocation(line: 73, column: 16, scope: !5) +!19 = !DILocation(line: 74, column: 30, scope: !5) +!20 = !DILocation(line: 75, column: 29, scope: !5) +!21 = !DILocation(line: 75, column: 14, scope: !5) +!22 = !DILocation(line: 75, column: 33, scope: !5) +!23 = !DILocation(line: 79, column: 18, scope: !5) +!24 = !DILocation(line: 79, column: 47, scope: !5) +!25 = !DILocation(line: 80, column: 38, scope: !5) +!26 = !DILocation(line: 263, column: 15, scope: !27, inlinedAt: !29) +!27 = distinct !DILexicalBlockFile(scope: !5, file: !28, discriminator: 0) +!28 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!29 = !DILocation(line: 293, column: 36, scope: !27, inlinedAt: !30) +!30 = !DILocation(line: 80, column: 43, scope: !31) +!31 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!32 = !DILocation(line: 80, column: 48, scope: !5) +!33 = !DILocation(line: 80, column: 25, scope: !5) +!34 = !DILocation(line: 77, column: 20, scope: !5) +!35 = !DILocation(line: 76, column: 14, scope: !5) +!36 = !DILocation(line: 76, column: 33, scope: !5) +!37 = !DILocation(line: 80, column: 17, scope: !5) +!38 = !DILocation(line: 81, column: 16, scope: !5) +!39 = !DILocation(line: 83, column: 25, scope: !5) +!40 = !DILocation(line: 83, column: 18, scope: !5) +!41 = !DILocation(line: 84, column: 21, scope: !5) +!42 = !DILocation(line: 84, column: 4, scope: !5) diff --git a/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.ptx b/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..2b699ff3ee168b834fff7af7d2ea72e09e994f28 --- /dev/null +++ b/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.ptx @@ -0,0 +1,5826 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_fwd_kernel // -- Begin function l2norm_fwd_kernel +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; +.global .align 1 .b8 _$_str_$_2[17] = {95, 95, 67, 85, 68, 65, 95, 80, 82, 69, 67, 95, 83, 81, 82, 84}; + // @l2norm_fwd_kernel +.visible .entry l2norm_fwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_2, + .param .f32 l2norm_fwd_kernel_param_3, + .param .u32 l2norm_fwd_kernel_param_4, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_6 +) +.reqntid 32 +{ + .reg .pred %p<115>; + .reg .b16 %rs<672>; + .reg .b32 %r<2881>; + .reg .b64 %rd<531>; + .loc 1 69 0 // l2norm.py:69:0 +$L__func_begin0: + .loc 1 69 0 // l2norm.py:69:0 + +// %bb.0: // %__nv_sqrtf.exit + ld.param.b64 %rd133, [l2norm_fwd_kernel_param_0]; + ld.param.b64 %rd134, [l2norm_fwd_kernel_param_1]; +$L__tmp0: + .loc 1 70 24 // l2norm.py:70:24 + mov.u32 %r590, %ctaid.x; + ld.param.b64 %rd135, [l2norm_fwd_kernel_param_2]; + .loc 1 71 16 // l2norm.py:71:16 + mul.wide.u32 %rd136, %r590, 128; + ld.param.b32 %r591, [l2norm_fwd_kernel_param_3]; + ld.param.s32 %rd137, [l2norm_fwd_kernel_param_4]; + .loc 1 71 34 // l2norm.py:71:34 + mov.u32 %r592, %tid.x; + and.b32 %r593, %r592, 16; + bfe.u32 %r594, %r592, 4, 1; + or.b32 %r595, %r594, 2; + .loc 1 71 21 // l2norm.py:71:21 + or.b32 %r596, %r594, 92; + or.b32 %r597, %r594, 94; + or.b32 %r598, %r594, 88; + or.b32 %r599, %r594, 90; + or.b32 %r600, %r594, 84; + or.b32 %r601, %r594, 86; + or.b32 %r602, %r594, 80; + or.b32 %r603, %r594, 82; + or.b32 %r604, %r594, 76; + or.b32 %r605, %r594, 78; + or.b32 %r606, %r594, 72; + or.b32 %r607, %r594, 74; + or.b32 %r608, %r594, 68; + or.b32 %r609, %r594, 70; + or.b32 %r610, %r594, 64; + or.b32 %r611, %r594, 66; + or.b32 %r612, %r594, 124; + or.b32 %r613, %r594, 126; + or.b32 %r614, %r594, 120; + or.b32 %r615, %r594, 122; + or.b32 %r616, %r594, 116; + or.b32 %r617, %r594, 118; + or.b32 %r618, %r594, 112; + or.b32 %r619, %r594, 114; + or.b32 %r620, %r594, 108; + or.b32 %r621, %r594, 110; + or.b32 %r622, %r594, 104; + or.b32 %r623, %r594, 106; + or.b32 %r624, %r594, 100; + or.b32 %r625, %r594, 102; + or.b32 %r626, %r594, 96; + or.b32 %r627, %r594, 98; + or.b32 %r628, %r594, 60; + or.b32 %r629, %r594, 62; + or.b32 %r630, %r594, 56; + or.b32 %r631, %r594, 58; + or.b32 %r632, %r594, 52; + or.b32 %r633, %r594, 54; + or.b32 %r634, %r594, 48; + or.b32 %r635, %r594, 50; + or.b32 %r636, %r594, 44; + or.b32 %r637, %r594, 46; + or.b32 %r638, %r594, 40; + or.b32 %r639, %r594, 42; + or.b32 %r640, %r594, 36; + or.b32 %r641, %r594, 38; + or.b32 %r642, %r594, 32; + or.b32 %r643, %r594, 34; + or.b32 %r644, %r594, 16; + or.b32 %r645, %r594, 18; + or.b32 %r646, %r594, 20; + or.b32 %r647, %r594, 22; + or.b32 %r648, %r594, 24; + or.b32 %r649, %r594, 26; + or.b32 %r650, %r594, 28; + or.b32 %r651, %r594, 30; + or.b32 %r652, %r594, 8; + or.b32 %r653, %r594, 10; + or.b32 %r654, %r594, 12; + or.b32 %r655, %r594, 14; + or.b32 %r656, %r594, 4; + or.b32 %r657, %r594, 6; + cvt.u64.u32 %rd138, %r627; + cvt.u64.u32 %rd139, %r626; + cvt.u64.u32 %rd140, %r625; + cvt.u64.u32 %rd141, %r624; + cvt.u64.u32 %rd142, %r623; + cvt.u64.u32 %rd143, %r622; + cvt.u64.u32 %rd144, %r621; + cvt.u64.u32 %rd145, %r620; + cvt.u64.u32 %rd146, %r619; + cvt.u64.u32 %rd147, %r618; + cvt.u64.u32 %rd148, %r617; + cvt.u64.u32 %rd149, %r616; + cvt.u64.u32 %rd150, %r615; + cvt.u64.u32 %rd151, %r614; + cvt.u64.u32 %rd152, %r613; + cvt.u64.u32 %rd153, %r612; + cvt.u64.u32 %rd154, %r611; + cvt.u64.u32 %rd155, %r610; + cvt.u64.u32 %rd156, %r609; + cvt.u64.u32 %rd157, %r608; + cvt.u64.u32 %rd158, %r607; + cvt.u64.u32 %rd159, %r606; + cvt.u64.u32 %rd160, %r605; + cvt.u64.u32 %rd161, %r604; + cvt.u64.u32 %rd162, %r603; + cvt.u64.u32 %rd163, %r602; + cvt.u64.u32 %rd164, %r601; + cvt.u64.u32 %rd165, %r600; + cvt.u64.u32 %rd166, %r599; + cvt.u64.u32 %rd167, %r598; + cvt.u64.u32 %rd168, %r597; + cvt.u64.u32 %rd169, %r596; + cvt.u64.u32 %rd170, %r643; + cvt.u64.u32 %rd171, %r642; + cvt.u64.u32 %rd172, %r641; + cvt.u64.u32 %rd173, %r640; + cvt.u64.u32 %rd174, %r639; + cvt.u64.u32 %rd175, %r638; + cvt.u64.u32 %rd176, %r637; + cvt.u64.u32 %rd177, %r636; + cvt.u64.u32 %rd178, %r635; + cvt.u64.u32 %rd179, %r634; + cvt.u64.u32 %rd180, %r633; + cvt.u64.u32 %rd181, %r632; + cvt.u64.u32 %rd182, %r631; + cvt.u64.u32 %rd183, %r630; + cvt.u64.u32 %rd184, %r629; + cvt.u64.u32 %rd185, %r628; + cvt.u64.u32 %rd186, %r651; + cvt.u64.u32 %rd187, %r650; + cvt.u64.u32 %rd188, %r649; + cvt.u64.u32 %rd189, %r648; + cvt.u64.u32 %rd190, %r647; + cvt.u64.u32 %rd191, %r646; + cvt.u64.u32 %rd192, %r645; + cvt.u64.u32 %rd193, %r644; + cvt.u64.u32 %rd194, %r655; + cvt.u64.u32 %rd195, %r654; + cvt.u64.u32 %rd196, %r653; + cvt.u64.u32 %rd197, %r652; + cvt.u64.u32 %rd198, %r657; + cvt.u64.u32 %rd199, %r656; + cvt.u64.u32 %rd200, %r595; + cvt.u64.u32 %rd201, %r594; + or.b64 %rd202, %rd136, %rd201; + or.b64 %rd203, %rd136, %rd200; + or.b64 %rd204, %rd136, %rd199; + or.b64 %rd205, %rd136, %rd198; + or.b64 %rd206, %rd136, %rd197; + or.b64 %rd207, %rd136, %rd196; + or.b64 %rd208, %rd136, %rd195; + or.b64 %rd209, %rd136, %rd194; + or.b64 %rd210, %rd136, %rd193; + or.b64 %rd211, %rd136, %rd192; + or.b64 %rd212, %rd136, %rd191; + or.b64 %rd213, %rd136, %rd190; + or.b64 %rd214, %rd136, %rd189; + or.b64 %rd215, %rd136, %rd188; + or.b64 %rd216, %rd136, %rd187; + or.b64 %rd217, %rd136, %rd186; + or.b64 %rd218, %rd136, %rd185; + or.b64 %rd219, %rd136, %rd184; + or.b64 %rd220, %rd136, %rd183; + or.b64 %rd221, %rd136, %rd182; + or.b64 %rd222, %rd136, %rd181; + or.b64 %rd223, %rd136, %rd180; + or.b64 %rd224, %rd136, %rd179; + or.b64 %rd225, %rd136, %rd178; + or.b64 %rd226, %rd136, %rd177; + or.b64 %rd227, %rd136, %rd176; + or.b64 %rd228, %rd136, %rd175; + or.b64 %rd229, %rd136, %rd174; + or.b64 %rd230, %rd136, %rd173; + or.b64 %rd231, %rd136, %rd172; + or.b64 %rd232, %rd136, %rd171; + or.b64 %rd233, %rd136, %rd170; + or.b64 %rd234, %rd136, %rd169; + or.b64 %rd235, %rd136, %rd168; + or.b64 %rd236, %rd136, %rd167; + or.b64 %rd237, %rd136, %rd166; + or.b64 %rd238, %rd136, %rd165; + or.b64 %rd239, %rd136, %rd164; + or.b64 %rd240, %rd136, %rd163; + or.b64 %rd241, %rd136, %rd162; + or.b64 %rd242, %rd136, %rd161; + or.b64 %rd243, %rd136, %rd160; + or.b64 %rd244, %rd136, %rd159; + or.b64 %rd245, %rd136, %rd158; + or.b64 %rd246, %rd136, %rd157; + or.b64 %rd247, %rd136, %rd156; + or.b64 %rd248, %rd136, %rd155; + or.b64 %rd249, %rd136, %rd154; + or.b64 %rd250, %rd136, %rd153; + or.b64 %rd251, %rd136, %rd152; + or.b64 %rd252, %rd136, %rd151; + or.b64 %rd253, %rd136, %rd150; + or.b64 %rd254, %rd136, %rd149; + or.b64 %rd255, %rd136, %rd148; + or.b64 %rd256, %rd136, %rd147; + or.b64 %rd257, %rd136, %rd146; + or.b64 %rd258, %rd136, %rd145; + or.b64 %rd259, %rd136, %rd144; + or.b64 %rd260, %rd136, %rd143; + or.b64 %rd261, %rd136, %rd142; + or.b64 %rd262, %rd136, %rd141; + or.b64 %rd263, %rd136, %rd140; + or.b64 %rd264, %rd136, %rd139; + or.b64 %rd265, %rd136, %rd138; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p69, %rd265, %rd137; + selp.b16 %rs1, 1, 0, %p69; + shl.b16 %rs2, %rs1, 2; + setp.lt.s64 %p70, %rd264, %rd137; + selp.b16 %rs3, -1, 0, %p70; + shl.b16 %rs4, %rs3, 3; + or.b16 %rs5, %rs4, %rs2; + setp.lt.s64 %p71, %rd263, %rd137; + selp.b16 %rs6, 1, 0, %p71; + setp.lt.s64 %p72, %rd262, %rd137; + selp.b16 %rs7, -1, 0, %p72; + shl.b16 %rs8, %rs7, 1; + or.b16 %rs9, %rs6, %rs8; + and.b16 %rs10, %rs9, 3; + or.b16 %rs11, %rs10, %rs5; + shl.b16 %rs12, %rs11, 12; + setp.lt.s64 %p73, %rd261, %rd137; + selp.b16 %rs13, 1, 0, %p73; + shl.b16 %rs14, %rs13, 2; + setp.lt.s64 %p74, %rd260, %rd137; + selp.b16 %rs15, -1, 0, %p74; + shl.b16 %rs16, %rs15, 3; + or.b16 %rs17, %rs16, %rs14; + setp.lt.s64 %p75, %rd259, %rd137; + selp.b16 %rs18, 1, 0, %p75; + setp.lt.s64 %p76, %rd258, %rd137; + selp.b16 %rs19, -1, 0, %p76; + shl.b16 %rs20, %rs19, 1; + or.b16 %rs21, %rs18, %rs20; + and.b16 %rs22, %rs21, 3; + or.b16 %rs23, %rs22, %rs17; + and.b16 %rs24, %rs23, 15; + shl.b16 %rs25, %rs24, 8; + or.b16 %rs26, %rs12, %rs25; + setp.lt.s64 %p77, %rd257, %rd137; + selp.b16 %rs27, 1, 0, %p77; + shl.b16 %rs28, %rs27, 2; + setp.lt.s64 %p78, %rd256, %rd137; + selp.b16 %rs29, -1, 0, %p78; + shl.b16 %rs30, %rs29, 3; + or.b16 %rs31, %rs30, %rs28; + setp.lt.s64 %p79, %rd255, %rd137; + selp.b16 %rs32, 1, 0, %p79; + setp.lt.s64 %p80, %rd254, %rd137; + selp.b16 %rs33, -1, 0, %p80; + shl.b16 %rs34, %rs33, 1; + or.b16 %rs35, %rs32, %rs34; + and.b16 %rs36, %rs35, 3; + or.b16 %rs37, %rs36, %rs31; + shl.b16 %rs38, %rs37, 4; + setp.lt.s64 %p81, %rd253, %rd137; + selp.b16 %rs39, 1, 0, %p81; + shl.b16 %rs40, %rs39, 2; + setp.lt.s64 %p82, %rd252, %rd137; + selp.b16 %rs41, -1, 0, %p82; + shl.b16 %rs42, %rs41, 3; + or.b16 %rs43, %rs42, %rs40; + setp.lt.s64 %p64, %rd251, %rd137; + selp.b16 %rs44, 1, 0, %p64; + setp.lt.s64 %p83, %rd250, %rd137; + selp.b16 %rs45, -1, 0, %p83; + shl.b16 %rs46, %rs45, 1; + or.b16 %rs47, %rs44, %rs46; + and.b16 %rs48, %rs47, 3; + or.b16 %rs49, %rs48, %rs43; + and.b16 %rs50, %rs49, 15; + or.b16 %rs51, %rs50, %rs38; + and.b16 %rs52, %rs51, 255; + or.b16 %rs53, %rs52, %rs26; + cvt.u32.u16 %r658, %rs53; + setp.lt.s64 %p84, %rd249, %rd137; + selp.b16 %rs54, 1, 0, %p84; + shl.b16 %rs55, %rs54, 2; + setp.lt.s64 %p85, %rd248, %rd137; + selp.b16 %rs56, -1, 0, %p85; + shl.b16 %rs57, %rs56, 3; + or.b16 %rs58, %rs57, %rs55; + setp.lt.s64 %p86, %rd247, %rd137; + selp.b16 %rs59, 1, 0, %p86; + setp.lt.s64 %p87, %rd246, %rd137; + selp.b16 %rs60, -1, 0, %p87; + shl.b16 %rs61, %rs60, 1; + or.b16 %rs62, %rs59, %rs61; + and.b16 %rs63, %rs62, 3; + or.b16 %rs64, %rs63, %rs58; + shl.b16 %rs65, %rs64, 12; + setp.lt.s64 %p88, %rd245, %rd137; + selp.b16 %rs66, 1, 0, %p88; + shl.b16 %rs67, %rs66, 2; + setp.lt.s64 %p89, %rd244, %rd137; + selp.b16 %rs68, -1, 0, %p89; + shl.b16 %rs69, %rs68, 3; + or.b16 %rs70, %rs69, %rs67; + setp.lt.s64 %p90, %rd243, %rd137; + selp.b16 %rs71, 1, 0, %p90; + setp.lt.s64 %p91, %rd242, %rd137; + selp.b16 %rs72, -1, 0, %p91; + shl.b16 %rs73, %rs72, 1; + or.b16 %rs74, %rs71, %rs73; + and.b16 %rs75, %rs74, 3; + or.b16 %rs76, %rs75, %rs70; + and.b16 %rs77, %rs76, 15; + shl.b16 %rs78, %rs77, 8; + or.b16 %rs79, %rs65, %rs78; + setp.lt.s64 %p92, %rd241, %rd137; + selp.b16 %rs80, 1, 0, %p92; + shl.b16 %rs81, %rs80, 2; + setp.lt.s64 %p93, %rd240, %rd137; + selp.b16 %rs82, -1, 0, %p93; + shl.b16 %rs83, %rs82, 3; + or.b16 %rs84, %rs83, %rs81; + setp.lt.s64 %p94, %rd239, %rd137; + selp.b16 %rs85, 1, 0, %p94; + setp.lt.s64 %p95, %rd238, %rd137; + selp.b16 %rs86, -1, 0, %p95; + shl.b16 %rs87, %rs86, 1; + or.b16 %rs88, %rs85, %rs87; + and.b16 %rs89, %rs88, 3; + or.b16 %rs90, %rs89, %rs84; + shl.b16 %rs91, %rs90, 4; + setp.lt.s64 %p96, %rd237, %rd137; + selp.b16 %rs92, 1, 0, %p96; + shl.b16 %rs93, %rs92, 2; + setp.lt.s64 %p97, %rd236, %rd137; + selp.b16 %rs94, -1, 0, %p97; + shl.b16 %rs95, %rs94, 3; + or.b16 %rs96, %rs95, %rs93; + setp.lt.s64 %p98, %rd235, %rd137; + selp.b16 %rs97, 1, 0, %p98; + setp.lt.s64 %p99, %rd234, %rd137; + selp.b16 %rs98, -1, 0, %p99; + shl.b16 %rs99, %rs98, 1; + or.b16 %rs100, %rs97, %rs99; + and.b16 %rs101, %rs100, 3; + or.b16 %rs102, %rs101, %rs96; + and.b16 %rs103, %rs102, 15; + or.b16 %rs104, %rs103, %rs91; + and.b16 %rs105, %rs104, 255; + or.b16 %rs106, %rs105, %rs79; + cvt.u32.u16 %r659, %rs106; + shl.b32 %r660, %r659, 16; + or.b32 %r661, %r658, %r660; + cvt.u64.u32 %rd266, %r661; + setp.lt.s64 %p100, %rd233, %rd137; + selp.b16 %rs107, 1, 0, %p100; + shl.b16 %rs108, %rs107, 2; + setp.lt.s64 %p101, %rd232, %rd137; + selp.b16 %rs109, -1, 0, %p101; + shl.b16 %rs110, %rs109, 3; + or.b16 %rs111, %rs110, %rs108; + setp.lt.s64 %p102, %rd231, %rd137; + selp.b16 %rs112, 1, 0, %p102; + setp.lt.s64 %p103, %rd230, %rd137; + selp.b16 %rs113, -1, 0, %p103; + shl.b16 %rs114, %rs113, 1; + or.b16 %rs115, %rs112, %rs114; + and.b16 %rs116, %rs115, 3; + or.b16 %rs117, %rs116, %rs111; + shl.b16 %rs118, %rs117, 12; + setp.lt.s64 %p104, %rd229, %rd137; + selp.b16 %rs119, 1, 0, %p104; + shl.b16 %rs120, %rs119, 2; + setp.lt.s64 %p105, %rd228, %rd137; + selp.b16 %rs121, -1, 0, %p105; + shl.b16 %rs122, %rs121, 3; + or.b16 %rs123, %rs122, %rs120; + setp.lt.s64 %p106, %rd227, %rd137; + selp.b16 %rs124, 1, 0, %p106; + setp.lt.s64 %p107, %rd226, %rd137; + selp.b16 %rs125, -1, 0, %p107; + shl.b16 %rs126, %rs125, 1; + or.b16 %rs127, %rs124, %rs126; + and.b16 %rs128, %rs127, 3; + or.b16 %rs129, %rs128, %rs123; + and.b16 %rs130, %rs129, 15; + shl.b16 %rs131, %rs130, 8; + or.b16 %rs132, %rs118, %rs131; + setp.lt.s64 %p108, %rd225, %rd137; + selp.b16 %rs133, 1, 0, %p108; + shl.b16 %rs134, %rs133, 2; + setp.lt.s64 %p109, %rd224, %rd137; + selp.b16 %rs135, -1, 0, %p109; + shl.b16 %rs136, %rs135, 3; + or.b16 %rs137, %rs136, %rs134; + setp.lt.s64 %p110, %rd223, %rd137; + selp.b16 %rs138, 1, 0, %p110; + setp.lt.s64 %p111, %rd222, %rd137; + selp.b16 %rs139, -1, 0, %p111; + shl.b16 %rs140, %rs139, 1; + or.b16 %rs141, %rs138, %rs140; + and.b16 %rs142, %rs141, 3; + or.b16 %rs143, %rs142, %rs137; + shl.b16 %rs144, %rs143, 4; + setp.lt.s64 %p112, %rd221, %rd137; + selp.b16 %rs145, 1, 0, %p112; + shl.b16 %rs146, %rs145, 2; + setp.lt.s64 %p113, %rd220, %rd137; + selp.b16 %rs147, -1, 0, %p113; + shl.b16 %rs148, %rs147, 3; + or.b16 %rs149, %rs148, %rs146; + setp.lt.s64 %p30, %rd219, %rd137; + selp.b16 %rs150, 1, 0, %p30; + setp.lt.s64 %p114, %rd218, %rd137; + selp.b16 %rs151, -1, 0, %p114; + shl.b16 %rs152, %rs151, 1; + or.b16 %rs153, %rs150, %rs152; + and.b16 %rs154, %rs153, 3; + or.b16 %rs155, %rs154, %rs149; + and.b16 %rs156, %rs155, 15; + or.b16 %rs157, %rs156, %rs144; + and.b16 %rs158, %rs157, 255; + or.b16 %rs159, %rs158, %rs132; + cvt.u32.u16 %r662, %rs159; + shr.u32 %r663, %r662, 1; + shr.u32 %r664, %r662, 2; + shr.u32 %r665, %r662, 3; + shr.u32 %r666, %r662, 4; + shr.u32 %r667, %r662, 5; + shr.u32 %r668, %r662, 6; + shr.u32 %r669, %r662, 7; + shr.u32 %r670, %r662, 8; + shr.u32 %r671, %r662, 9; + shr.u32 %r672, %r662, 10; + shr.u32 %r673, %r662, 11; + shr.u32 %r674, %r662, 12; + shr.u32 %r675, %r662, 13; + shr.u32 %r676, %r662, 14; + shr.u32 %r677, %r662, 15; + setp.lt.s64 %p32, %rd217, %rd137; + setp.lt.s64 %p14, %rd216, %rd137; + setp.lt.s64 %p13, %rd215, %rd137; + setp.lt.s64 %p12, %rd214, %rd137; + setp.lt.s64 %p11, %rd213, %rd137; + setp.lt.s64 %p10, %rd212, %rd137; + setp.lt.s64 %p9, %rd211, %rd137; + setp.lt.s64 %p8, %rd210, %rd137; + setp.lt.s64 %p31, %rd209, %rd137; + setp.lt.s64 %p7, %rd208, %rd137; + setp.lt.s64 %p6, %rd207, %rd137; + setp.lt.s64 %p5, %rd206, %rd137; + setp.lt.s64 %p4, %rd205, %rd137; + setp.lt.s64 %p3, %rd204, %rd137; + setp.lt.s64 %p2, %rd203, %rd137; + setp.lt.s64 %p1, %rd202, %rd137; + .loc 1 74 30 // l2norm.py:74:30 + and.b32 %r678, %r592, 15; + shl.b32 %r679, %r678, 3; + .loc 1 75 14 // l2norm.py:75:14 + shl.b64 %rd267, %rd202, 8; + add.s64 %rd268, %rd133, %rd267; + shl.b64 %rd269, %rd203, 8; + add.s64 %rd270, %rd133, %rd269; + shl.b64 %rd271, %rd204, 8; + add.s64 %rd272, %rd133, %rd271; + shl.b64 %rd273, %rd205, 8; + add.s64 %rd274, %rd133, %rd273; + shl.b64 %rd275, %rd206, 8; + add.s64 %rd276, %rd133, %rd275; + shl.b64 %rd277, %rd207, 8; + add.s64 %rd278, %rd133, %rd277; + shl.b64 %rd279, %rd208, 8; + add.s64 %rd280, %rd133, %rd279; + shl.b64 %rd281, %rd209, 8; + add.s64 %rd282, %rd133, %rd281; + shl.b64 %rd283, %rd210, 8; + add.s64 %rd284, %rd133, %rd283; + shl.b64 %rd285, %rd211, 8; + add.s64 %rd286, %rd133, %rd285; + shl.b64 %rd287, %rd212, 8; + add.s64 %rd288, %rd133, %rd287; + shl.b64 %rd289, %rd213, 8; + add.s64 %rd290, %rd133, %rd289; + shl.b64 %rd291, %rd214, 8; + add.s64 %rd292, %rd133, %rd291; + shl.b64 %rd293, %rd215, 8; + add.s64 %rd294, %rd133, %rd293; + shl.b64 %rd295, %rd216, 8; + add.s64 %rd296, %rd133, %rd295; + shl.b64 %rd297, %rd217, 8; + add.s64 %rd298, %rd133, %rd297; + shl.b64 %rd299, %rd232, 8; + add.s64 %rd300, %rd133, %rd299; + shl.b64 %rd301, %rd233, 8; + add.s64 %rd302, %rd133, %rd301; + shl.b64 %rd303, %rd230, 8; + add.s64 %rd304, %rd133, %rd303; + shl.b64 %rd305, %rd231, 8; + add.s64 %rd306, %rd133, %rd305; + shl.b64 %rd307, %rd228, 8; + add.s64 %rd308, %rd133, %rd307; + shl.b64 %rd309, %rd229, 8; + add.s64 %rd310, %rd133, %rd309; + shl.b64 %rd311, %rd226, 8; + add.s64 %rd312, %rd133, %rd311; + shl.b64 %rd313, %rd227, 8; + add.s64 %rd314, %rd133, %rd313; + shl.b64 %rd315, %rd224, 8; + add.s64 %rd316, %rd133, %rd315; + shl.b64 %rd317, %rd225, 8; + add.s64 %rd318, %rd133, %rd317; + shl.b64 %rd319, %rd222, 8; + add.s64 %rd320, %rd133, %rd319; + shl.b64 %rd321, %rd223, 8; + add.s64 %rd322, %rd133, %rd321; + shl.b64 %rd323, %rd220, 8; + add.s64 %rd324, %rd133, %rd323; + shl.b64 %rd325, %rd221, 8; + add.s64 %rd326, %rd133, %rd325; + shl.b64 %rd327, %rd218, 8; + add.s64 %rd328, %rd133, %rd327; + shl.b64 %rd329, %rd219, 8; + add.s64 %rd330, %rd133, %rd329; + shl.b64 %rd331, %rd248, 8; + add.s64 %rd332, %rd133, %rd331; + shl.b64 %rd333, %rd249, 8; + add.s64 %rd334, %rd133, %rd333; + shl.b64 %rd335, %rd246, 8; + add.s64 %rd336, %rd133, %rd335; + shl.b64 %rd337, %rd247, 8; + add.s64 %rd338, %rd133, %rd337; + shl.b64 %rd339, %rd244, 8; + add.s64 %rd340, %rd133, %rd339; + shl.b64 %rd341, %rd245, 8; + add.s64 %rd342, %rd133, %rd341; + shl.b64 %rd343, %rd242, 8; + add.s64 %rd344, %rd133, %rd343; + shl.b64 %rd345, %rd243, 8; + add.s64 %rd346, %rd133, %rd345; + shl.b64 %rd347, %rd240, 8; + add.s64 %rd348, %rd133, %rd347; + shl.b64 %rd349, %rd241, 8; + add.s64 %rd350, %rd133, %rd349; + shl.b64 %rd351, %rd238, 8; + add.s64 %rd352, %rd133, %rd351; + shl.b64 %rd353, %rd239, 8; + add.s64 %rd354, %rd133, %rd353; + shl.b64 %rd355, %rd236, 8; + add.s64 %rd356, %rd133, %rd355; + shl.b64 %rd357, %rd237, 8; + add.s64 %rd358, %rd133, %rd357; + shl.b64 %rd359, %rd234, 8; + add.s64 %rd360, %rd133, %rd359; + shl.b64 %rd361, %rd235, 8; + add.s64 %rd362, %rd133, %rd361; + shl.b64 %rd363, %rd264, 8; + add.s64 %rd364, %rd133, %rd363; + shl.b64 %rd365, %rd265, 8; + add.s64 %rd366, %rd133, %rd365; + shl.b64 %rd367, %rd262, 8; + add.s64 %rd368, %rd133, %rd367; + shl.b64 %rd369, %rd263, 8; + add.s64 %rd370, %rd133, %rd369; + shl.b64 %rd371, %rd260, 8; + add.s64 %rd372, %rd133, %rd371; + shl.b64 %rd373, %rd261, 8; + add.s64 %rd374, %rd133, %rd373; + shl.b64 %rd375, %rd258, 8; + add.s64 %rd376, %rd133, %rd375; + shl.b64 %rd377, %rd259, 8; + add.s64 %rd378, %rd133, %rd377; + shl.b64 %rd379, %rd256, 8; + add.s64 %rd380, %rd133, %rd379; + shl.b64 %rd381, %rd257, 8; + add.s64 %rd382, %rd133, %rd381; + shl.b64 %rd383, %rd254, 8; + add.s64 %rd384, %rd133, %rd383; + shl.b64 %rd385, %rd255, 8; + add.s64 %rd386, %rd133, %rd385; + shl.b64 %rd387, %rd252, 8; + add.s64 %rd388, %rd133, %rd387; + shl.b64 %rd389, %rd253, 8; + add.s64 %rd390, %rd133, %rd389; + shl.b64 %rd391, %rd250, 8; + add.s64 %rd392, %rd133, %rd391; + shl.b64 %rd393, %rd251, 8; + add.s64 %rd394, %rd133, %rd393; + .loc 1 75 33 // l2norm.py:75:33 + mul.wide.u32 %rd395, %r679, 2; + add.s64 %rd1, %rd268, %rd395; + add.s64 %rd2, %rd270, %rd395; + add.s64 %rd3, %rd272, %rd395; + add.s64 %rd4, %rd274, %rd395; + add.s64 %rd5, %rd276, %rd395; + add.s64 %rd6, %rd278, %rd395; + add.s64 %rd7, %rd280, %rd395; + add.s64 %rd8, %rd282, %rd395; + add.s64 %rd9, %rd284, %rd395; + add.s64 %rd10, %rd286, %rd395; + add.s64 %rd11, %rd288, %rd395; + add.s64 %rd12, %rd290, %rd395; + add.s64 %rd13, %rd292, %rd395; + add.s64 %rd14, %rd294, %rd395; + add.s64 %rd15, %rd296, %rd395; + add.s64 %rd16, %rd298, %rd395; + add.s64 %rd17, %rd300, %rd395; + add.s64 %rd18, %rd302, %rd395; + add.s64 %rd19, %rd304, %rd395; + add.s64 %rd20, %rd306, %rd395; + add.s64 %rd21, %rd308, %rd395; + add.s64 %rd22, %rd310, %rd395; + add.s64 %rd23, %rd312, %rd395; + add.s64 %rd24, %rd314, %rd395; + add.s64 %rd25, %rd316, %rd395; + add.s64 %rd26, %rd318, %rd395; + add.s64 %rd27, %rd320, %rd395; + add.s64 %rd28, %rd322, %rd395; + add.s64 %rd29, %rd324, %rd395; + add.s64 %rd30, %rd326, %rd395; + add.s64 %rd31, %rd328, %rd395; + add.s64 %rd32, %rd330, %rd395; + add.s64 %rd33, %rd332, %rd395; + add.s64 %rd34, %rd334, %rd395; + add.s64 %rd35, %rd336, %rd395; + add.s64 %rd36, %rd338, %rd395; + add.s64 %rd37, %rd340, %rd395; + add.s64 %rd38, %rd342, %rd395; + add.s64 %rd39, %rd344, %rd395; + add.s64 %rd40, %rd346, %rd395; + add.s64 %rd41, %rd348, %rd395; + add.s64 %rd42, %rd350, %rd395; + add.s64 %rd43, %rd352, %rd395; + add.s64 %rd44, %rd354, %rd395; + add.s64 %rd45, %rd356, %rd395; + add.s64 %rd46, %rd358, %rd395; + add.s64 %rd47, %rd360, %rd395; + add.s64 %rd48, %rd362, %rd395; + add.s64 %rd49, %rd364, %rd395; + add.s64 %rd50, %rd366, %rd395; + add.s64 %rd51, %rd368, %rd395; + add.s64 %rd52, %rd370, %rd395; + add.s64 %rd53, %rd372, %rd395; + add.s64 %rd54, %rd374, %rd395; + add.s64 %rd55, %rd376, %rd395; + add.s64 %rd56, %rd378, %rd395; + add.s64 %rd57, %rd380, %rd395; + add.s64 %rd58, %rd382, %rd395; + add.s64 %rd59, %rd384, %rd395; + add.s64 %rd60, %rd386, %rd395; + add.s64 %rd61, %rd388, %rd395; + add.s64 %rd62, %rd390, %rd395; + add.s64 %rd63, %rd392, %rd395; + add.s64 %rd64, %rd394, %rd395; + mov.b32 %r5, 0; + .loc 1 79 18 // l2norm.py:79:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs160, %rs161}, %r1; + mov.b32 {%rs162, %rs163}, %r2; + mov.b32 {%rs164, %rs165}, %r3; + mov.b32 {%rs166, %rs167}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs168, %rs169}, %r6; + mov.b32 {%rs170, %rs171}, %r7; + mov.b32 {%rs172, %rs173}, %r8; + mov.b32 {%rs174, %rs175}, %r9; + // begin inline asm + mov.u32 %r10, %r5; + mov.u32 %r11, %r5; + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + @%p3 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs176, %rs177}, %r10; + mov.b32 {%rs178, %rs179}, %r11; + mov.b32 {%rs180, %rs181}, %r12; + mov.b32 {%rs182, %rs183}, %r13; + // begin inline asm + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + @%p4 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ]; + // end inline asm + mov.b32 {%rs184, %rs185}, %r14; + mov.b32 {%rs186, %rs187}, %r15; + mov.b32 {%rs188, %rs189}, %r16; + mov.b32 {%rs190, %rs191}, %r17; + // begin inline asm + mov.u32 %r18, %r5; + mov.u32 %r19, %r5; + mov.u32 %r20, %r5; + mov.u32 %r21, %r5; + @%p5 ld.global.v4.b32 { %r18, %r19, %r20, %r21 }, [ %rd5 + 0 ]; + // end inline asm + mov.b32 {%rs192, %rs193}, %r18; + mov.b32 {%rs194, %rs195}, %r19; + mov.b32 {%rs196, %rs197}, %r20; + mov.b32 {%rs198, %rs199}, %r21; + // begin inline asm + mov.u32 %r22, %r5; + mov.u32 %r23, %r5; + mov.u32 %r24, %r5; + mov.u32 %r25, %r5; + @%p6 ld.global.v4.b32 { %r22, %r23, %r24, %r25 }, [ %rd6 + 0 ]; + // end inline asm + mov.b32 {%rs200, %rs201}, %r22; + mov.b32 {%rs202, %rs203}, %r23; + mov.b32 {%rs204, %rs205}, %r24; + mov.b32 {%rs206, %rs207}, %r25; + // begin inline asm + mov.u32 %r26, %r5; + mov.u32 %r27, %r5; + mov.u32 %r28, %r5; + mov.u32 %r29, %r5; + @%p7 ld.global.v4.b32 { %r26, %r27, %r28, %r29 }, [ %rd7 + 0 ]; + // end inline asm + mov.b32 {%rs208, %rs209}, %r26; + mov.b32 {%rs210, %rs211}, %r27; + mov.b32 {%rs212, %rs213}, %r28; + mov.b32 {%rs214, %rs215}, %r29; + // begin inline asm + mov.u32 %r30, %r5; + mov.u32 %r31, %r5; + mov.u32 %r32, %r5; + mov.u32 %r33, %r5; + @%p31 ld.global.v4.b32 { %r30, %r31, %r32, %r33 }, [ %rd8 + 0 ]; + // end inline asm + mov.b32 {%rs216, %rs217}, %r30; + mov.b32 {%rs218, %rs219}, %r31; + mov.b32 {%rs220, %rs221}, %r32; + mov.b32 {%rs222, %rs223}, %r33; + // begin inline asm + mov.u32 %r34, %r5; + mov.u32 %r35, %r5; + mov.u32 %r36, %r5; + mov.u32 %r37, %r5; + @%p8 ld.global.v4.b32 { %r34, %r35, %r36, %r37 }, [ %rd9 + 0 ]; + // end inline asm + mov.b32 {%rs224, %rs225}, %r34; + mov.b32 {%rs226, %rs227}, %r35; + mov.b32 {%rs228, %rs229}, %r36; + mov.b32 {%rs230, %rs231}, %r37; + // begin inline asm + mov.u32 %r38, %r5; + mov.u32 %r39, %r5; + mov.u32 %r40, %r5; + mov.u32 %r41, %r5; + @%p9 ld.global.v4.b32 { %r38, %r39, %r40, %r41 }, [ %rd10 + 0 ]; + // end inline asm + mov.b32 {%rs232, %rs233}, %r38; + mov.b32 {%rs234, %rs235}, %r39; + mov.b32 {%rs236, %rs237}, %r40; + mov.b32 {%rs238, %rs239}, %r41; + // begin inline asm + mov.u32 %r42, %r5; + mov.u32 %r43, %r5; + mov.u32 %r44, %r5; + mov.u32 %r45, %r5; + @%p10 ld.global.v4.b32 { %r42, %r43, %r44, %r45 }, [ %rd11 + 0 ]; + // end inline asm + mov.b32 {%rs240, %rs241}, %r42; + mov.b32 {%rs242, %rs243}, %r43; + mov.b32 {%rs244, %rs245}, %r44; + mov.b32 {%rs246, %rs247}, %r45; + // begin inline asm + mov.u32 %r46, %r5; + mov.u32 %r47, %r5; + mov.u32 %r48, %r5; + mov.u32 %r49, %r5; + @%p11 ld.global.v4.b32 { %r46, %r47, %r48, %r49 }, [ %rd12 + 0 ]; + // end inline asm + mov.b32 {%rs248, %rs249}, %r46; + mov.b32 {%rs250, %rs251}, %r47; + mov.b32 {%rs252, %rs253}, %r48; + mov.b32 {%rs254, %rs255}, %r49; + // begin inline asm + mov.u32 %r50, %r5; + mov.u32 %r51, %r5; + mov.u32 %r52, %r5; + mov.u32 %r53, %r5; + @%p12 ld.global.v4.b32 { %r50, %r51, %r52, %r53 }, [ %rd13 + 0 ]; + // end inline asm + mov.b32 {%rs256, %rs257}, %r50; + mov.b32 {%rs258, %rs259}, %r51; + mov.b32 {%rs260, %rs261}, %r52; + mov.b32 {%rs262, %rs263}, %r53; + // begin inline asm + mov.u32 %r54, %r5; + mov.u32 %r55, %r5; + mov.u32 %r56, %r5; + mov.u32 %r57, %r5; + @%p13 ld.global.v4.b32 { %r54, %r55, %r56, %r57 }, [ %rd14 + 0 ]; + // end inline asm + mov.b32 {%rs264, %rs265}, %r54; + mov.b32 {%rs266, %rs267}, %r55; + mov.b32 {%rs268, %rs269}, %r56; + mov.b32 {%rs270, %rs271}, %r57; + // begin inline asm + mov.u32 %r58, %r5; + mov.u32 %r59, %r5; + mov.u32 %r60, %r5; + mov.u32 %r61, %r5; + @%p14 ld.global.v4.b32 { %r58, %r59, %r60, %r61 }, [ %rd15 + 0 ]; + // end inline asm + mov.b32 {%rs272, %rs273}, %r58; + mov.b32 {%rs274, %rs275}, %r59; + mov.b32 {%rs276, %rs277}, %r60; + mov.b32 {%rs278, %rs279}, %r61; + // begin inline asm + mov.u32 %r62, %r5; + mov.u32 %r63, %r5; + mov.u32 %r64, %r5; + mov.u32 %r65, %r5; + @%p32 ld.global.v4.b32 { %r62, %r63, %r64, %r65 }, [ %rd16 + 0 ]; + // end inline asm + mov.b32 {%rs280, %rs281}, %r62; + mov.b32 {%rs282, %rs283}, %r63; + mov.b32 {%rs284, %rs285}, %r64; + mov.b32 {%rs286, %rs287}, %r65; + and.b32 %r680, %r677, 1; + setp.ne.b32 %p15, %r680, 0; + // begin inline asm + mov.u32 %r66, %r5; + mov.u32 %r67, %r5; + mov.u32 %r68, %r5; + mov.u32 %r69, %r5; + @%p15 ld.global.v4.b32 { %r66, %r67, %r68, %r69 }, [ %rd17 + 0 ]; + // end inline asm + mov.b32 {%rs288, %rs289}, %r66; + mov.b32 {%rs290, %rs291}, %r67; + mov.b32 {%rs292, %rs293}, %r68; + mov.b32 {%rs294, %rs295}, %r69; + and.b32 %r681, %r676, 1; + setp.ne.b32 %p16, %r681, 0; + // begin inline asm + mov.u32 %r70, %r5; + mov.u32 %r71, %r5; + mov.u32 %r72, %r5; + mov.u32 %r73, %r5; + @%p16 ld.global.v4.b32 { %r70, %r71, %r72, %r73 }, [ %rd18 + 0 ]; + // end inline asm + mov.b32 {%rs296, %rs297}, %r70; + mov.b32 {%rs298, %rs299}, %r71; + mov.b32 {%rs300, %rs301}, %r72; + mov.b32 {%rs302, %rs303}, %r73; + and.b32 %r682, %r675, 1; + setp.ne.b32 %p17, %r682, 0; + // begin inline asm + mov.u32 %r74, %r5; + mov.u32 %r75, %r5; + mov.u32 %r76, %r5; + mov.u32 %r77, %r5; + @%p17 ld.global.v4.b32 { %r74, %r75, %r76, %r77 }, [ %rd19 + 0 ]; + // end inline asm + mov.b32 {%rs304, %rs305}, %r74; + mov.b32 {%rs306, %rs307}, %r75; + mov.b32 {%rs308, %rs309}, %r76; + mov.b32 {%rs310, %rs311}, %r77; + and.b32 %r683, %r674, 1; + setp.ne.b32 %p18, %r683, 0; + // begin inline asm + mov.u32 %r78, %r5; + mov.u32 %r79, %r5; + mov.u32 %r80, %r5; + mov.u32 %r81, %r5; + @%p18 ld.global.v4.b32 { %r78, %r79, %r80, %r81 }, [ %rd20 + 0 ]; + // end inline asm + mov.b32 {%rs312, %rs313}, %r78; + mov.b32 {%rs314, %rs315}, %r79; + mov.b32 {%rs316, %rs317}, %r80; + mov.b32 {%rs318, %rs319}, %r81; + and.b32 %r684, %r673, 1; + setp.ne.b32 %p19, %r684, 0; + // begin inline asm + mov.u32 %r82, %r5; + mov.u32 %r83, %r5; + mov.u32 %r84, %r5; + mov.u32 %r85, %r5; + @%p19 ld.global.v4.b32 { %r82, %r83, %r84, %r85 }, [ %rd21 + 0 ]; + // end inline asm + mov.b32 {%rs320, %rs321}, %r82; + mov.b32 {%rs322, %rs323}, %r83; + mov.b32 {%rs324, %rs325}, %r84; + mov.b32 {%rs326, %rs327}, %r85; + and.b32 %r685, %r672, 1; + setp.ne.b32 %p20, %r685, 0; + // begin inline asm + mov.u32 %r86, %r5; + mov.u32 %r87, %r5; + mov.u32 %r88, %r5; + mov.u32 %r89, %r5; + @%p20 ld.global.v4.b32 { %r86, %r87, %r88, %r89 }, [ %rd22 + 0 ]; + // end inline asm + mov.b32 {%rs328, %rs329}, %r86; + mov.b32 {%rs330, %rs331}, %r87; + mov.b32 {%rs332, %rs333}, %r88; + mov.b32 {%rs334, %rs335}, %r89; + and.b32 %r686, %r671, 1; + setp.ne.b32 %p21, %r686, 0; + // begin inline asm + mov.u32 %r90, %r5; + mov.u32 %r91, %r5; + mov.u32 %r92, %r5; + mov.u32 %r93, %r5; + @%p21 ld.global.v4.b32 { %r90, %r91, %r92, %r93 }, [ %rd23 + 0 ]; + // end inline asm + mov.b32 {%rs336, %rs337}, %r90; + mov.b32 {%rs338, %rs339}, %r91; + mov.b32 {%rs340, %rs341}, %r92; + mov.b32 {%rs342, %rs343}, %r93; + and.b32 %r687, %r670, 1; + setp.ne.b32 %p22, %r687, 0; + // begin inline asm + mov.u32 %r94, %r5; + mov.u32 %r95, %r5; + mov.u32 %r96, %r5; + mov.u32 %r97, %r5; + @%p22 ld.global.v4.b32 { %r94, %r95, %r96, %r97 }, [ %rd24 + 0 ]; + // end inline asm + mov.b32 {%rs344, %rs345}, %r94; + mov.b32 {%rs346, %rs347}, %r95; + mov.b32 {%rs348, %rs349}, %r96; + mov.b32 {%rs350, %rs351}, %r97; + and.b32 %r688, %r669, 1; + setp.ne.b32 %p23, %r688, 0; + // begin inline asm + mov.u32 %r98, %r5; + mov.u32 %r99, %r5; + mov.u32 %r100, %r5; + mov.u32 %r101, %r5; + @%p23 ld.global.v4.b32 { %r98, %r99, %r100, %r101 }, [ %rd25 + 0 ]; + // end inline asm + mov.b32 {%rs352, %rs353}, %r98; + mov.b32 {%rs354, %rs355}, %r99; + mov.b32 {%rs356, %rs357}, %r100; + mov.b32 {%rs358, %rs359}, %r101; + and.b32 %r689, %r668, 1; + setp.ne.b32 %p24, %r689, 0; + // begin inline asm + mov.u32 %r102, %r5; + mov.u32 %r103, %r5; + mov.u32 %r104, %r5; + mov.u32 %r105, %r5; + @%p24 ld.global.v4.b32 { %r102, %r103, %r104, %r105 }, [ %rd26 + 0 ]; + // end inline asm + mov.b32 {%rs360, %rs361}, %r102; + mov.b32 {%rs362, %rs363}, %r103; + mov.b32 {%rs364, %rs365}, %r104; + mov.b32 {%rs366, %rs367}, %r105; + and.b32 %r690, %r667, 1; + setp.ne.b32 %p25, %r690, 0; + // begin inline asm + mov.u32 %r106, %r5; + mov.u32 %r107, %r5; + mov.u32 %r108, %r5; + mov.u32 %r109, %r5; + @%p25 ld.global.v4.b32 { %r106, %r107, %r108, %r109 }, [ %rd27 + 0 ]; + // end inline asm + mov.b32 {%rs368, %rs369}, %r106; + mov.b32 {%rs370, %rs371}, %r107; + mov.b32 {%rs372, %rs373}, %r108; + mov.b32 {%rs374, %rs375}, %r109; + and.b32 %r691, %r666, 1; + setp.ne.b32 %p26, %r691, 0; + // begin inline asm + mov.u32 %r110, %r5; + mov.u32 %r111, %r5; + mov.u32 %r112, %r5; + mov.u32 %r113, %r5; + @%p26 ld.global.v4.b32 { %r110, %r111, %r112, %r113 }, [ %rd28 + 0 ]; + // end inline asm + mov.b32 {%rs376, %rs377}, %r110; + mov.b32 {%rs378, %rs379}, %r111; + mov.b32 {%rs380, %rs381}, %r112; + mov.b32 {%rs382, %rs383}, %r113; + and.b32 %r692, %r665, 1; + setp.ne.b32 %p27, %r692, 0; + // begin inline asm + mov.u32 %r114, %r5; + mov.u32 %r115, %r5; + mov.u32 %r116, %r5; + mov.u32 %r117, %r5; + @%p27 ld.global.v4.b32 { %r114, %r115, %r116, %r117 }, [ %rd29 + 0 ]; + // end inline asm + mov.b32 {%rs384, %rs385}, %r114; + mov.b32 {%rs386, %rs387}, %r115; + mov.b32 {%rs388, %rs389}, %r116; + mov.b32 {%rs390, %rs391}, %r117; + and.b32 %r693, %r664, 1; + setp.ne.b32 %p28, %r693, 0; + // begin inline asm + mov.u32 %r118, %r5; + mov.u32 %r119, %r5; + mov.u32 %r120, %r5; + mov.u32 %r121, %r5; + @%p28 ld.global.v4.b32 { %r118, %r119, %r120, %r121 }, [ %rd30 + 0 ]; + // end inline asm + mov.b32 {%rs392, %rs393}, %r118; + mov.b32 {%rs394, %rs395}, %r119; + mov.b32 {%rs396, %rs397}, %r120; + mov.b32 {%rs398, %rs399}, %r121; + and.b32 %r694, %r663, 1; + setp.ne.b32 %p29, %r694, 0; + // begin inline asm + mov.u32 %r122, %r5; + mov.u32 %r123, %r5; + mov.u32 %r124, %r5; + mov.u32 %r125, %r5; + @%p29 ld.global.v4.b32 { %r122, %r123, %r124, %r125 }, [ %rd31 + 0 ]; + // end inline asm + mov.b32 {%rs400, %rs401}, %r122; + mov.b32 {%rs402, %rs403}, %r123; + mov.b32 {%rs404, %rs405}, %r124; + mov.b32 {%rs406, %rs407}, %r125; + // begin inline asm + mov.u32 %r126, %r5; + mov.u32 %r127, %r5; + mov.u32 %r128, %r5; + mov.u32 %r129, %r5; + @%p30 ld.global.v4.b32 { %r126, %r127, %r128, %r129 }, [ %rd32 + 0 ]; + // end inline asm + mov.b32 {%rs408, %rs409}, %r126; + mov.b32 {%rs410, %rs411}, %r127; + mov.b32 {%rs412, %rs413}, %r128; + mov.b32 {%rs414, %rs415}, %r129; + shr.u64 %rd396, %rd266, 31; + and.b64 %rd397, %rd396, 1; + setp.ne.b64 %p33, %rd397, 0; + // begin inline asm + mov.u32 %r130, %r5; + mov.u32 %r131, %r5; + mov.u32 %r132, %r5; + mov.u32 %r133, %r5; + @%p33 ld.global.v4.b32 { %r130, %r131, %r132, %r133 }, [ %rd33 + 0 ]; + // end inline asm + mov.b32 {%rs416, %rs417}, %r130; + mov.b32 {%rs418, %rs419}, %r131; + mov.b32 {%rs420, %rs421}, %r132; + mov.b32 {%rs422, %rs423}, %r133; + shr.u64 %rd398, %rd266, 30; + and.b64 %rd399, %rd398, 1; + setp.ne.b64 %p34, %rd399, 0; + // begin inline asm + mov.u32 %r134, %r5; + mov.u32 %r135, %r5; + mov.u32 %r136, %r5; + mov.u32 %r137, %r5; + @%p34 ld.global.v4.b32 { %r134, %r135, %r136, %r137 }, [ %rd34 + 0 ]; + // end inline asm + mov.b32 {%rs424, %rs425}, %r134; + mov.b32 {%rs426, %rs427}, %r135; + mov.b32 {%rs428, %rs429}, %r136; + mov.b32 {%rs430, %rs431}, %r137; + shr.u64 %rd400, %rd266, 29; + and.b64 %rd401, %rd400, 1; + setp.ne.b64 %p35, %rd401, 0; + // begin inline asm + mov.u32 %r138, %r5; + mov.u32 %r139, %r5; + mov.u32 %r140, %r5; + mov.u32 %r141, %r5; + @%p35 ld.global.v4.b32 { %r138, %r139, %r140, %r141 }, [ %rd35 + 0 ]; + // end inline asm + mov.b32 {%rs432, %rs433}, %r138; + mov.b32 {%rs434, %rs435}, %r139; + mov.b32 {%rs436, %rs437}, %r140; + mov.b32 {%rs438, %rs439}, %r141; + shr.u64 %rd402, %rd266, 28; + and.b64 %rd403, %rd402, 1; + setp.ne.b64 %p36, %rd403, 0; + // begin inline asm + mov.u32 %r142, %r5; + mov.u32 %r143, %r5; + mov.u32 %r144, %r5; + mov.u32 %r145, %r5; + @%p36 ld.global.v4.b32 { %r142, %r143, %r144, %r145 }, [ %rd36 + 0 ]; + // end inline asm + mov.b32 {%rs440, %rs441}, %r142; + mov.b32 {%rs442, %rs443}, %r143; + mov.b32 {%rs444, %rs445}, %r144; + mov.b32 {%rs446, %rs447}, %r145; + shr.u64 %rd404, %rd266, 27; + and.b64 %rd405, %rd404, 1; + setp.ne.b64 %p37, %rd405, 0; + // begin inline asm + mov.u32 %r146, %r5; + mov.u32 %r147, %r5; + mov.u32 %r148, %r5; + mov.u32 %r149, %r5; + @%p37 ld.global.v4.b32 { %r146, %r147, %r148, %r149 }, [ %rd37 + 0 ]; + // end inline asm + mov.b32 {%rs448, %rs449}, %r146; + mov.b32 {%rs450, %rs451}, %r147; + mov.b32 {%rs452, %rs453}, %r148; + mov.b32 {%rs454, %rs455}, %r149; + shr.u64 %rd406, %rd266, 26; + and.b64 %rd407, %rd406, 1; + setp.ne.b64 %p38, %rd407, 0; + // begin inline asm + mov.u32 %r150, %r5; + mov.u32 %r151, %r5; + mov.u32 %r152, %r5; + mov.u32 %r153, %r5; + @%p38 ld.global.v4.b32 { %r150, %r151, %r152, %r153 }, [ %rd38 + 0 ]; + // end inline asm + mov.b32 {%rs456, %rs457}, %r150; + mov.b32 {%rs458, %rs459}, %r151; + mov.b32 {%rs460, %rs461}, %r152; + mov.b32 {%rs462, %rs463}, %r153; + shr.u64 %rd408, %rd266, 25; + and.b64 %rd409, %rd408, 1; + setp.ne.b64 %p39, %rd409, 0; + // begin inline asm + mov.u32 %r154, %r5; + mov.u32 %r155, %r5; + mov.u32 %r156, %r5; + mov.u32 %r157, %r5; + @%p39 ld.global.v4.b32 { %r154, %r155, %r156, %r157 }, [ %rd39 + 0 ]; + // end inline asm + mov.b32 {%rs464, %rs465}, %r154; + mov.b32 {%rs466, %rs467}, %r155; + mov.b32 {%rs468, %rs469}, %r156; + mov.b32 {%rs470, %rs471}, %r157; + shr.u64 %rd410, %rd266, 24; + and.b64 %rd411, %rd410, 1; + setp.ne.b64 %p40, %rd411, 0; + // begin inline asm + mov.u32 %r158, %r5; + mov.u32 %r159, %r5; + mov.u32 %r160, %r5; + mov.u32 %r161, %r5; + @%p40 ld.global.v4.b32 { %r158, %r159, %r160, %r161 }, [ %rd40 + 0 ]; + // end inline asm + mov.b32 {%rs472, %rs473}, %r158; + mov.b32 {%rs474, %rs475}, %r159; + mov.b32 {%rs476, %rs477}, %r160; + mov.b32 {%rs478, %rs479}, %r161; + shr.u64 %rd412, %rd266, 23; + and.b64 %rd413, %rd412, 1; + setp.ne.b64 %p41, %rd413, 0; + // begin inline asm + mov.u32 %r162, %r5; + mov.u32 %r163, %r5; + mov.u32 %r164, %r5; + mov.u32 %r165, %r5; + @%p41 ld.global.v4.b32 { %r162, %r163, %r164, %r165 }, [ %rd41 + 0 ]; + // end inline asm + mov.b32 {%rs480, %rs481}, %r162; + mov.b32 {%rs482, %rs483}, %r163; + mov.b32 {%rs484, %rs485}, %r164; + mov.b32 {%rs486, %rs487}, %r165; + shr.u64 %rd414, %rd266, 22; + and.b64 %rd415, %rd414, 1; + setp.ne.b64 %p42, %rd415, 0; + // begin inline asm + mov.u32 %r166, %r5; + mov.u32 %r167, %r5; + mov.u32 %r168, %r5; + mov.u32 %r169, %r5; + @%p42 ld.global.v4.b32 { %r166, %r167, %r168, %r169 }, [ %rd42 + 0 ]; + // end inline asm + mov.b32 {%rs488, %rs489}, %r166; + mov.b32 {%rs490, %rs491}, %r167; + mov.b32 {%rs492, %rs493}, %r168; + mov.b32 {%rs494, %rs495}, %r169; + shr.u64 %rd416, %rd266, 21; + and.b64 %rd417, %rd416, 1; + setp.ne.b64 %p43, %rd417, 0; + // begin inline asm + mov.u32 %r170, %r5; + mov.u32 %r171, %r5; + mov.u32 %r172, %r5; + mov.u32 %r173, %r5; + @%p43 ld.global.v4.b32 { %r170, %r171, %r172, %r173 }, [ %rd43 + 0 ]; + // end inline asm + mov.b32 {%rs496, %rs497}, %r170; + mov.b32 {%rs498, %rs499}, %r171; + mov.b32 {%rs500, %rs501}, %r172; + mov.b32 {%rs502, %rs503}, %r173; + shr.u64 %rd418, %rd266, 20; + and.b64 %rd419, %rd418, 1; + setp.ne.b64 %p44, %rd419, 0; + // begin inline asm + mov.u32 %r174, %r5; + mov.u32 %r175, %r5; + mov.u32 %r176, %r5; + mov.u32 %r177, %r5; + @%p44 ld.global.v4.b32 { %r174, %r175, %r176, %r177 }, [ %rd44 + 0 ]; + // end inline asm + mov.b32 {%rs504, %rs505}, %r174; + mov.b32 {%rs506, %rs507}, %r175; + mov.b32 {%rs508, %rs509}, %r176; + mov.b32 {%rs510, %rs511}, %r177; + shr.u64 %rd420, %rd266, 19; + and.b64 %rd421, %rd420, 1; + setp.ne.b64 %p45, %rd421, 0; + // begin inline asm + mov.u32 %r178, %r5; + mov.u32 %r179, %r5; + mov.u32 %r180, %r5; + mov.u32 %r181, %r5; + @%p45 ld.global.v4.b32 { %r178, %r179, %r180, %r181 }, [ %rd45 + 0 ]; + // end inline asm + mov.b32 {%rs512, %rs513}, %r178; + mov.b32 {%rs514, %rs515}, %r179; + mov.b32 {%rs516, %rs517}, %r180; + mov.b32 {%rs518, %rs519}, %r181; + shr.u64 %rd422, %rd266, 18; + and.b64 %rd423, %rd422, 1; + setp.ne.b64 %p46, %rd423, 0; + // begin inline asm + mov.u32 %r182, %r5; + mov.u32 %r183, %r5; + mov.u32 %r184, %r5; + mov.u32 %r185, %r5; + @%p46 ld.global.v4.b32 { %r182, %r183, %r184, %r185 }, [ %rd46 + 0 ]; + // end inline asm + mov.b32 {%rs520, %rs521}, %r182; + mov.b32 {%rs522, %rs523}, %r183; + mov.b32 {%rs524, %rs525}, %r184; + mov.b32 {%rs526, %rs527}, %r185; + shr.u64 %rd424, %rd266, 17; + and.b64 %rd425, %rd424, 1; + setp.ne.b64 %p47, %rd425, 0; + // begin inline asm + mov.u32 %r186, %r5; + mov.u32 %r187, %r5; + mov.u32 %r188, %r5; + mov.u32 %r189, %r5; + @%p47 ld.global.v4.b32 { %r186, %r187, %r188, %r189 }, [ %rd47 + 0 ]; + // end inline asm + mov.b32 {%rs528, %rs529}, %r186; + mov.b32 {%rs530, %rs531}, %r187; + mov.b32 {%rs532, %rs533}, %r188; + mov.b32 {%rs534, %rs535}, %r189; + shr.u64 %rd426, %rd266, 16; + and.b64 %rd427, %rd426, 1; + setp.ne.b64 %p48, %rd427, 0; + // begin inline asm + mov.u32 %r190, %r5; + mov.u32 %r191, %r5; + mov.u32 %r192, %r5; + mov.u32 %r193, %r5; + @%p48 ld.global.v4.b32 { %r190, %r191, %r192, %r193 }, [ %rd48 + 0 ]; + // end inline asm + mov.b32 {%rs536, %rs537}, %r190; + mov.b32 {%rs538, %rs539}, %r191; + mov.b32 {%rs540, %rs541}, %r192; + mov.b32 {%rs542, %rs543}, %r193; + shr.u64 %rd428, %rd266, 15; + and.b64 %rd429, %rd428, 1; + setp.ne.b64 %p49, %rd429, 0; + // begin inline asm + mov.u32 %r194, %r5; + mov.u32 %r195, %r5; + mov.u32 %r196, %r5; + mov.u32 %r197, %r5; + @%p49 ld.global.v4.b32 { %r194, %r195, %r196, %r197 }, [ %rd49 + 0 ]; + // end inline asm + mov.b32 {%rs544, %rs545}, %r194; + mov.b32 {%rs546, %rs547}, %r195; + mov.b32 {%rs548, %rs549}, %r196; + mov.b32 {%rs550, %rs551}, %r197; + shr.u64 %rd430, %rd266, 14; + and.b64 %rd431, %rd430, 1; + setp.ne.b64 %p50, %rd431, 0; + // begin inline asm + mov.u32 %r198, %r5; + mov.u32 %r199, %r5; + mov.u32 %r200, %r5; + mov.u32 %r201, %r5; + @%p50 ld.global.v4.b32 { %r198, %r199, %r200, %r201 }, [ %rd50 + 0 ]; + // end inline asm + mov.b32 {%rs552, %rs553}, %r198; + mov.b32 {%rs554, %rs555}, %r199; + mov.b32 {%rs556, %rs557}, %r200; + mov.b32 {%rs558, %rs559}, %r201; + shr.u64 %rd432, %rd266, 13; + and.b64 %rd433, %rd432, 1; + setp.ne.b64 %p51, %rd433, 0; + // begin inline asm + mov.u32 %r202, %r5; + mov.u32 %r203, %r5; + mov.u32 %r204, %r5; + mov.u32 %r205, %r5; + @%p51 ld.global.v4.b32 { %r202, %r203, %r204, %r205 }, [ %rd51 + 0 ]; + // end inline asm + mov.b32 {%rs560, %rs561}, %r202; + mov.b32 {%rs562, %rs563}, %r203; + mov.b32 {%rs564, %rs565}, %r204; + mov.b32 {%rs566, %rs567}, %r205; + shr.u64 %rd434, %rd266, 12; + and.b64 %rd435, %rd434, 1; + setp.ne.b64 %p52, %rd435, 0; + // begin inline asm + mov.u32 %r206, %r5; + mov.u32 %r207, %r5; + mov.u32 %r208, %r5; + mov.u32 %r209, %r5; + @%p52 ld.global.v4.b32 { %r206, %r207, %r208, %r209 }, [ %rd52 + 0 ]; + // end inline asm + mov.b32 {%rs568, %rs569}, %r206; + mov.b32 {%rs570, %rs571}, %r207; + mov.b32 {%rs572, %rs573}, %r208; + mov.b32 {%rs574, %rs575}, %r209; + shr.u64 %rd436, %rd266, 11; + and.b64 %rd437, %rd436, 1; + setp.ne.b64 %p53, %rd437, 0; + // begin inline asm + mov.u32 %r210, %r5; + mov.u32 %r211, %r5; + mov.u32 %r212, %r5; + mov.u32 %r213, %r5; + @%p53 ld.global.v4.b32 { %r210, %r211, %r212, %r213 }, [ %rd53 + 0 ]; + // end inline asm + mov.b32 {%rs576, %rs577}, %r210; + mov.b32 {%rs578, %rs579}, %r211; + mov.b32 {%rs580, %rs581}, %r212; + mov.b32 {%rs582, %rs583}, %r213; + shr.u64 %rd438, %rd266, 10; + and.b64 %rd439, %rd438, 1; + setp.ne.b64 %p54, %rd439, 0; + // begin inline asm + mov.u32 %r214, %r5; + mov.u32 %r215, %r5; + mov.u32 %r216, %r5; + mov.u32 %r217, %r5; + @%p54 ld.global.v4.b32 { %r214, %r215, %r216, %r217 }, [ %rd54 + 0 ]; + // end inline asm + mov.b32 {%rs584, %rs585}, %r214; + mov.b32 {%rs586, %rs587}, %r215; + mov.b32 {%rs588, %rs589}, %r216; + mov.b32 {%rs590, %rs591}, %r217; + shr.u64 %rd440, %rd266, 9; + and.b64 %rd441, %rd440, 1; + setp.ne.b64 %p55, %rd441, 0; + // begin inline asm + mov.u32 %r218, %r5; + mov.u32 %r219, %r5; + mov.u32 %r220, %r5; + mov.u32 %r221, %r5; + @%p55 ld.global.v4.b32 { %r218, %r219, %r220, %r221 }, [ %rd55 + 0 ]; + // end inline asm + mov.b32 {%rs592, %rs593}, %r218; + mov.b32 {%rs594, %rs595}, %r219; + mov.b32 {%rs596, %rs597}, %r220; + mov.b32 {%rs598, %rs599}, %r221; + shr.u64 %rd442, %rd266, 8; + and.b64 %rd443, %rd442, 1; + setp.ne.b64 %p56, %rd443, 0; + // begin inline asm + mov.u32 %r222, %r5; + mov.u32 %r223, %r5; + mov.u32 %r224, %r5; + mov.u32 %r225, %r5; + @%p56 ld.global.v4.b32 { %r222, %r223, %r224, %r225 }, [ %rd56 + 0 ]; + // end inline asm + mov.b32 {%rs600, %rs601}, %r222; + mov.b32 {%rs602, %rs603}, %r223; + mov.b32 {%rs604, %rs605}, %r224; + mov.b32 {%rs606, %rs607}, %r225; + shr.u64 %rd444, %rd266, 7; + and.b64 %rd445, %rd444, 1; + setp.ne.b64 %p57, %rd445, 0; + // begin inline asm + mov.u32 %r226, %r5; + mov.u32 %r227, %r5; + mov.u32 %r228, %r5; + mov.u32 %r229, %r5; + @%p57 ld.global.v4.b32 { %r226, %r227, %r228, %r229 }, [ %rd57 + 0 ]; + // end inline asm + mov.b32 {%rs608, %rs609}, %r226; + mov.b32 {%rs610, %rs611}, %r227; + mov.b32 {%rs612, %rs613}, %r228; + mov.b32 {%rs614, %rs615}, %r229; + shr.u64 %rd446, %rd266, 6; + and.b64 %rd447, %rd446, 1; + setp.ne.b64 %p58, %rd447, 0; + // begin inline asm + mov.u32 %r230, %r5; + mov.u32 %r231, %r5; + mov.u32 %r232, %r5; + mov.u32 %r233, %r5; + @%p58 ld.global.v4.b32 { %r230, %r231, %r232, %r233 }, [ %rd58 + 0 ]; + // end inline asm + mov.b32 {%rs616, %rs617}, %r230; + mov.b32 {%rs618, %rs619}, %r231; + mov.b32 {%rs620, %rs621}, %r232; + mov.b32 {%rs622, %rs623}, %r233; + shr.u64 %rd448, %rd266, 5; + and.b64 %rd449, %rd448, 1; + setp.ne.b64 %p59, %rd449, 0; + // begin inline asm + mov.u32 %r234, %r5; + mov.u32 %r235, %r5; + mov.u32 %r236, %r5; + mov.u32 %r237, %r5; + @%p59 ld.global.v4.b32 { %r234, %r235, %r236, %r237 }, [ %rd59 + 0 ]; + // end inline asm + mov.b32 {%rs624, %rs625}, %r234; + mov.b32 {%rs626, %rs627}, %r235; + mov.b32 {%rs628, %rs629}, %r236; + mov.b32 {%rs630, %rs631}, %r237; + shr.u64 %rd450, %rd266, 4; + and.b64 %rd451, %rd450, 1; + setp.ne.b64 %p60, %rd451, 0; + // begin inline asm + mov.u32 %r238, %r5; + mov.u32 %r239, %r5; + mov.u32 %r240, %r5; + mov.u32 %r241, %r5; + @%p60 ld.global.v4.b32 { %r238, %r239, %r240, %r241 }, [ %rd60 + 0 ]; + // end inline asm + mov.b32 {%rs632, %rs633}, %r238; + mov.b32 {%rs634, %rs635}, %r239; + mov.b32 {%rs636, %rs637}, %r240; + mov.b32 {%rs638, %rs639}, %r241; + shr.u64 %rd452, %rd266, 3; + and.b64 %rd453, %rd452, 1; + setp.ne.b64 %p61, %rd453, 0; + // begin inline asm + mov.u32 %r242, %r5; + mov.u32 %r243, %r5; + mov.u32 %r244, %r5; + mov.u32 %r245, %r5; + @%p61 ld.global.v4.b32 { %r242, %r243, %r244, %r245 }, [ %rd61 + 0 ]; + // end inline asm + mov.b32 {%rs640, %rs641}, %r242; + mov.b32 {%rs642, %rs643}, %r243; + mov.b32 {%rs644, %rs645}, %r244; + mov.b32 {%rs646, %rs647}, %r245; + shr.u64 %rd454, %rd266, 2; + and.b64 %rd455, %rd454, 1; + setp.ne.b64 %p62, %rd455, 0; + // begin inline asm + mov.u32 %r246, %r5; + mov.u32 %r247, %r5; + mov.u32 %r248, %r5; + mov.u32 %r249, %r5; + @%p62 ld.global.v4.b32 { %r246, %r247, %r248, %r249 }, [ %rd62 + 0 ]; + // end inline asm + mov.b32 {%rs648, %rs649}, %r246; + mov.b32 {%rs650, %rs651}, %r247; + mov.b32 {%rs652, %rs653}, %r248; + mov.b32 {%rs654, %rs655}, %r249; + shr.u64 %rd456, %rd266, 1; + and.b64 %rd457, %rd456, 1; + setp.ne.b64 %p63, %rd457, 0; + // begin inline asm + mov.u32 %r250, %r5; + mov.u32 %r251, %r5; + mov.u32 %r252, %r5; + mov.u32 %r253, %r5; + @%p63 ld.global.v4.b32 { %r250, %r251, %r252, %r253 }, [ %rd63 + 0 ]; + // end inline asm + mov.b32 {%rs656, %rs657}, %r250; + mov.b32 {%rs658, %rs659}, %r251; + mov.b32 {%rs660, %rs661}, %r252; + mov.b32 {%rs662, %rs663}, %r253; + // begin inline asm + mov.u32 %r254, %r5; + mov.u32 %r255, %r5; + mov.u32 %r256, %r5; + mov.u32 %r257, %r5; + @%p64 ld.global.v4.b32 { %r254, %r255, %r256, %r257 }, [ %rd64 + 0 ]; + // end inline asm + mov.b32 {%rs664, %rs665}, %r254; + mov.b32 {%rs666, %rs667}, %r255; + mov.b32 {%rs668, %rs669}, %r256; + mov.b32 {%rs670, %rs671}, %r257; + .loc 1 79 47 // l2norm.py:79:47 + cvt.f32.bf16 %r695, %rs160; + cvt.f32.bf16 %r696, %rs161; + cvt.f32.bf16 %r697, %rs162; + cvt.f32.bf16 %r698, %rs163; + cvt.f32.bf16 %r699, %rs164; + cvt.f32.bf16 %r700, %rs165; + cvt.f32.bf16 %r701, %rs166; + cvt.f32.bf16 %r702, %rs167; + cvt.f32.bf16 %r703, %rs168; + cvt.f32.bf16 %r704, %rs169; + cvt.f32.bf16 %r705, %rs170; + cvt.f32.bf16 %r706, %rs171; + cvt.f32.bf16 %r707, %rs172; + cvt.f32.bf16 %r708, %rs173; + cvt.f32.bf16 %r709, %rs174; + cvt.f32.bf16 %r710, %rs175; + cvt.f32.bf16 %r711, %rs176; + cvt.f32.bf16 %r712, %rs177; + cvt.f32.bf16 %r713, %rs178; + cvt.f32.bf16 %r714, %rs179; + cvt.f32.bf16 %r715, %rs180; + cvt.f32.bf16 %r716, %rs181; + cvt.f32.bf16 %r717, %rs182; + cvt.f32.bf16 %r718, %rs183; + cvt.f32.bf16 %r719, %rs184; + cvt.f32.bf16 %r720, %rs185; + cvt.f32.bf16 %r721, %rs186; + cvt.f32.bf16 %r722, %rs187; + cvt.f32.bf16 %r723, %rs188; + cvt.f32.bf16 %r724, %rs189; + cvt.f32.bf16 %r725, %rs190; + cvt.f32.bf16 %r726, %rs191; + cvt.f32.bf16 %r727, %rs192; + cvt.f32.bf16 %r728, %rs193; + cvt.f32.bf16 %r729, %rs194; + cvt.f32.bf16 %r730, %rs195; + cvt.f32.bf16 %r731, %rs196; + cvt.f32.bf16 %r732, %rs197; + cvt.f32.bf16 %r733, %rs198; + cvt.f32.bf16 %r734, %rs199; + cvt.f32.bf16 %r735, %rs200; + cvt.f32.bf16 %r736, %rs201; + cvt.f32.bf16 %r737, %rs202; + cvt.f32.bf16 %r738, %rs203; + cvt.f32.bf16 %r739, %rs204; + cvt.f32.bf16 %r740, %rs205; + cvt.f32.bf16 %r741, %rs206; + cvt.f32.bf16 %r742, %rs207; + cvt.f32.bf16 %r743, %rs208; + cvt.f32.bf16 %r744, %rs209; + cvt.f32.bf16 %r745, %rs210; + cvt.f32.bf16 %r746, %rs211; + cvt.f32.bf16 %r747, %rs212; + cvt.f32.bf16 %r748, %rs213; + cvt.f32.bf16 %r749, %rs214; + cvt.f32.bf16 %r750, %rs215; + cvt.f32.bf16 %r751, %rs216; + cvt.f32.bf16 %r752, %rs217; + cvt.f32.bf16 %r753, %rs218; + cvt.f32.bf16 %r754, %rs219; + cvt.f32.bf16 %r755, %rs220; + cvt.f32.bf16 %r756, %rs221; + cvt.f32.bf16 %r757, %rs222; + cvt.f32.bf16 %r758, %rs223; + cvt.f32.bf16 %r759, %rs224; + cvt.f32.bf16 %r760, %rs225; + cvt.f32.bf16 %r761, %rs226; + cvt.f32.bf16 %r762, %rs227; + cvt.f32.bf16 %r763, %rs228; + cvt.f32.bf16 %r764, %rs229; + cvt.f32.bf16 %r765, %rs230; + cvt.f32.bf16 %r766, %rs231; + cvt.f32.bf16 %r767, %rs232; + cvt.f32.bf16 %r768, %rs233; + cvt.f32.bf16 %r769, %rs234; + cvt.f32.bf16 %r770, %rs235; + cvt.f32.bf16 %r771, %rs236; + cvt.f32.bf16 %r772, %rs237; + cvt.f32.bf16 %r773, %rs238; + cvt.f32.bf16 %r774, %rs239; + cvt.f32.bf16 %r775, %rs240; + cvt.f32.bf16 %r776, %rs241; + cvt.f32.bf16 %r777, %rs242; + cvt.f32.bf16 %r778, %rs243; + cvt.f32.bf16 %r779, %rs244; + cvt.f32.bf16 %r780, %rs245; + cvt.f32.bf16 %r781, %rs246; + cvt.f32.bf16 %r782, %rs247; + cvt.f32.bf16 %r783, %rs248; + cvt.f32.bf16 %r784, %rs249; + cvt.f32.bf16 %r785, %rs250; + cvt.f32.bf16 %r786, %rs251; + cvt.f32.bf16 %r787, %rs252; + cvt.f32.bf16 %r788, %rs253; + cvt.f32.bf16 %r789, %rs254; + cvt.f32.bf16 %r790, %rs255; + cvt.f32.bf16 %r791, %rs256; + cvt.f32.bf16 %r792, %rs257; + cvt.f32.bf16 %r793, %rs258; + cvt.f32.bf16 %r794, %rs259; + cvt.f32.bf16 %r795, %rs260; + cvt.f32.bf16 %r796, %rs261; + cvt.f32.bf16 %r797, %rs262; + cvt.f32.bf16 %r798, %rs263; + cvt.f32.bf16 %r799, %rs264; + cvt.f32.bf16 %r800, %rs265; + cvt.f32.bf16 %r801, %rs266; + cvt.f32.bf16 %r802, %rs267; + cvt.f32.bf16 %r803, %rs268; + cvt.f32.bf16 %r804, %rs269; + cvt.f32.bf16 %r805, %rs270; + cvt.f32.bf16 %r806, %rs271; + cvt.f32.bf16 %r807, %rs272; + cvt.f32.bf16 %r808, %rs273; + cvt.f32.bf16 %r809, %rs274; + cvt.f32.bf16 %r810, %rs275; + cvt.f32.bf16 %r811, %rs276; + cvt.f32.bf16 %r812, %rs277; + cvt.f32.bf16 %r813, %rs278; + cvt.f32.bf16 %r814, %rs279; + cvt.f32.bf16 %r815, %rs280; + cvt.f32.bf16 %r816, %rs281; + cvt.f32.bf16 %r817, %rs282; + cvt.f32.bf16 %r818, %rs283; + cvt.f32.bf16 %r819, %rs284; + cvt.f32.bf16 %r820, %rs285; + cvt.f32.bf16 %r821, %rs286; + cvt.f32.bf16 %r822, %rs287; + cvt.f32.bf16 %r823, %rs288; + cvt.f32.bf16 %r824, %rs289; + cvt.f32.bf16 %r825, %rs290; + cvt.f32.bf16 %r826, %rs291; + cvt.f32.bf16 %r827, %rs292; + cvt.f32.bf16 %r828, %rs293; + cvt.f32.bf16 %r829, %rs294; + cvt.f32.bf16 %r830, %rs295; + cvt.f32.bf16 %r831, %rs296; + cvt.f32.bf16 %r832, %rs297; + cvt.f32.bf16 %r833, %rs298; + cvt.f32.bf16 %r834, %rs299; + cvt.f32.bf16 %r835, %rs300; + cvt.f32.bf16 %r836, %rs301; + cvt.f32.bf16 %r837, %rs302; + cvt.f32.bf16 %r838, %rs303; + cvt.f32.bf16 %r839, %rs304; + cvt.f32.bf16 %r840, %rs305; + cvt.f32.bf16 %r841, %rs306; + cvt.f32.bf16 %r842, %rs307; + cvt.f32.bf16 %r843, %rs308; + cvt.f32.bf16 %r844, %rs309; + cvt.f32.bf16 %r845, %rs310; + cvt.f32.bf16 %r846, %rs311; + cvt.f32.bf16 %r847, %rs312; + cvt.f32.bf16 %r848, %rs313; + cvt.f32.bf16 %r849, %rs314; + cvt.f32.bf16 %r850, %rs315; + cvt.f32.bf16 %r851, %rs316; + cvt.f32.bf16 %r852, %rs317; + cvt.f32.bf16 %r853, %rs318; + cvt.f32.bf16 %r854, %rs319; + cvt.f32.bf16 %r855, %rs320; + cvt.f32.bf16 %r856, %rs321; + cvt.f32.bf16 %r857, %rs322; + cvt.f32.bf16 %r858, %rs323; + cvt.f32.bf16 %r859, %rs324; + cvt.f32.bf16 %r860, %rs325; + cvt.f32.bf16 %r861, %rs326; + cvt.f32.bf16 %r862, %rs327; + cvt.f32.bf16 %r863, %rs328; + cvt.f32.bf16 %r864, %rs329; + cvt.f32.bf16 %r865, %rs330; + cvt.f32.bf16 %r866, %rs331; + cvt.f32.bf16 %r867, %rs332; + cvt.f32.bf16 %r868, %rs333; + cvt.f32.bf16 %r869, %rs334; + cvt.f32.bf16 %r870, %rs335; + cvt.f32.bf16 %r871, %rs336; + cvt.f32.bf16 %r872, %rs337; + cvt.f32.bf16 %r873, %rs338; + cvt.f32.bf16 %r874, %rs339; + cvt.f32.bf16 %r875, %rs340; + cvt.f32.bf16 %r876, %rs341; + cvt.f32.bf16 %r877, %rs342; + cvt.f32.bf16 %r878, %rs343; + cvt.f32.bf16 %r879, %rs344; + cvt.f32.bf16 %r880, %rs345; + cvt.f32.bf16 %r881, %rs346; + cvt.f32.bf16 %r882, %rs347; + cvt.f32.bf16 %r883, %rs348; + cvt.f32.bf16 %r884, %rs349; + cvt.f32.bf16 %r885, %rs350; + cvt.f32.bf16 %r886, %rs351; + cvt.f32.bf16 %r887, %rs352; + cvt.f32.bf16 %r888, %rs353; + cvt.f32.bf16 %r889, %rs354; + cvt.f32.bf16 %r890, %rs355; + cvt.f32.bf16 %r891, %rs356; + cvt.f32.bf16 %r892, %rs357; + cvt.f32.bf16 %r893, %rs358; + cvt.f32.bf16 %r894, %rs359; + cvt.f32.bf16 %r895, %rs360; + cvt.f32.bf16 %r896, %rs361; + cvt.f32.bf16 %r897, %rs362; + cvt.f32.bf16 %r898, %rs363; + cvt.f32.bf16 %r899, %rs364; + cvt.f32.bf16 %r900, %rs365; + cvt.f32.bf16 %r901, %rs366; + cvt.f32.bf16 %r902, %rs367; + cvt.f32.bf16 %r903, %rs368; + cvt.f32.bf16 %r904, %rs369; + cvt.f32.bf16 %r905, %rs370; + cvt.f32.bf16 %r906, %rs371; + cvt.f32.bf16 %r907, %rs372; + cvt.f32.bf16 %r908, %rs373; + cvt.f32.bf16 %r909, %rs374; + cvt.f32.bf16 %r910, %rs375; + cvt.f32.bf16 %r911, %rs376; + cvt.f32.bf16 %r912, %rs377; + cvt.f32.bf16 %r913, %rs378; + cvt.f32.bf16 %r914, %rs379; + cvt.f32.bf16 %r915, %rs380; + cvt.f32.bf16 %r916, %rs381; + cvt.f32.bf16 %r917, %rs382; + cvt.f32.bf16 %r918, %rs383; + cvt.f32.bf16 %r919, %rs384; + cvt.f32.bf16 %r920, %rs385; + cvt.f32.bf16 %r921, %rs386; + cvt.f32.bf16 %r922, %rs387; + cvt.f32.bf16 %r923, %rs388; + cvt.f32.bf16 %r924, %rs389; + cvt.f32.bf16 %r925, %rs390; + cvt.f32.bf16 %r926, %rs391; + cvt.f32.bf16 %r927, %rs392; + cvt.f32.bf16 %r928, %rs393; + cvt.f32.bf16 %r929, %rs394; + cvt.f32.bf16 %r930, %rs395; + cvt.f32.bf16 %r931, %rs396; + cvt.f32.bf16 %r932, %rs397; + cvt.f32.bf16 %r933, %rs398; + cvt.f32.bf16 %r934, %rs399; + cvt.f32.bf16 %r935, %rs400; + cvt.f32.bf16 %r936, %rs401; + cvt.f32.bf16 %r937, %rs402; + cvt.f32.bf16 %r938, %rs403; + cvt.f32.bf16 %r939, %rs404; + cvt.f32.bf16 %r940, %rs405; + cvt.f32.bf16 %r941, %rs406; + cvt.f32.bf16 %r942, %rs407; + cvt.f32.bf16 %r943, %rs408; + cvt.f32.bf16 %r944, %rs409; + cvt.f32.bf16 %r945, %rs410; + cvt.f32.bf16 %r946, %rs411; + cvt.f32.bf16 %r947, %rs412; + cvt.f32.bf16 %r948, %rs413; + cvt.f32.bf16 %r949, %rs414; + cvt.f32.bf16 %r950, %rs415; + cvt.f32.bf16 %r951, %rs416; + cvt.f32.bf16 %r952, %rs417; + cvt.f32.bf16 %r953, %rs418; + cvt.f32.bf16 %r954, %rs419; + cvt.f32.bf16 %r955, %rs420; + cvt.f32.bf16 %r956, %rs421; + cvt.f32.bf16 %r957, %rs422; + cvt.f32.bf16 %r958, %rs423; + cvt.f32.bf16 %r959, %rs424; + cvt.f32.bf16 %r960, %rs425; + cvt.f32.bf16 %r961, %rs426; + cvt.f32.bf16 %r962, %rs427; + cvt.f32.bf16 %r963, %rs428; + cvt.f32.bf16 %r964, %rs429; + cvt.f32.bf16 %r965, %rs430; + cvt.f32.bf16 %r966, %rs431; + cvt.f32.bf16 %r967, %rs432; + cvt.f32.bf16 %r968, %rs433; + cvt.f32.bf16 %r969, %rs434; + cvt.f32.bf16 %r970, %rs435; + cvt.f32.bf16 %r971, %rs436; + cvt.f32.bf16 %r972, %rs437; + cvt.f32.bf16 %r973, %rs438; + cvt.f32.bf16 %r974, %rs439; + cvt.f32.bf16 %r975, %rs440; + cvt.f32.bf16 %r976, %rs441; + cvt.f32.bf16 %r977, %rs442; + cvt.f32.bf16 %r978, %rs443; + cvt.f32.bf16 %r979, %rs444; + cvt.f32.bf16 %r980, %rs445; + cvt.f32.bf16 %r981, %rs446; + cvt.f32.bf16 %r982, %rs447; + cvt.f32.bf16 %r983, %rs448; + cvt.f32.bf16 %r984, %rs449; + cvt.f32.bf16 %r985, %rs450; + cvt.f32.bf16 %r986, %rs451; + cvt.f32.bf16 %r987, %rs452; + cvt.f32.bf16 %r988, %rs453; + cvt.f32.bf16 %r989, %rs454; + cvt.f32.bf16 %r990, %rs455; + cvt.f32.bf16 %r991, %rs456; + cvt.f32.bf16 %r992, %rs457; + cvt.f32.bf16 %r993, %rs458; + cvt.f32.bf16 %r994, %rs459; + cvt.f32.bf16 %r995, %rs460; + cvt.f32.bf16 %r996, %rs461; + cvt.f32.bf16 %r997, %rs462; + cvt.f32.bf16 %r998, %rs463; + cvt.f32.bf16 %r999, %rs464; + cvt.f32.bf16 %r1000, %rs465; + cvt.f32.bf16 %r1001, %rs466; + cvt.f32.bf16 %r1002, %rs467; + cvt.f32.bf16 %r1003, %rs468; + cvt.f32.bf16 %r1004, %rs469; + cvt.f32.bf16 %r1005, %rs470; + cvt.f32.bf16 %r1006, %rs471; + cvt.f32.bf16 %r1007, %rs472; + cvt.f32.bf16 %r1008, %rs473; + cvt.f32.bf16 %r1009, %rs474; + cvt.f32.bf16 %r1010, %rs475; + cvt.f32.bf16 %r1011, %rs476; + cvt.f32.bf16 %r1012, %rs477; + cvt.f32.bf16 %r1013, %rs478; + cvt.f32.bf16 %r1014, %rs479; + cvt.f32.bf16 %r1015, %rs480; + cvt.f32.bf16 %r1016, %rs481; + cvt.f32.bf16 %r1017, %rs482; + cvt.f32.bf16 %r1018, %rs483; + cvt.f32.bf16 %r1019, %rs484; + cvt.f32.bf16 %r1020, %rs485; + cvt.f32.bf16 %r1021, %rs486; + cvt.f32.bf16 %r1022, %rs487; + cvt.f32.bf16 %r1023, %rs488; + cvt.f32.bf16 %r1024, %rs489; + cvt.f32.bf16 %r1025, %rs490; + cvt.f32.bf16 %r1026, %rs491; + cvt.f32.bf16 %r1027, %rs492; + cvt.f32.bf16 %r1028, %rs493; + cvt.f32.bf16 %r1029, %rs494; + cvt.f32.bf16 %r1030, %rs495; + cvt.f32.bf16 %r1031, %rs496; + cvt.f32.bf16 %r1032, %rs497; + cvt.f32.bf16 %r1033, %rs498; + cvt.f32.bf16 %r1034, %rs499; + cvt.f32.bf16 %r1035, %rs500; + cvt.f32.bf16 %r1036, %rs501; + cvt.f32.bf16 %r1037, %rs502; + cvt.f32.bf16 %r1038, %rs503; + cvt.f32.bf16 %r1039, %rs504; + cvt.f32.bf16 %r1040, %rs505; + cvt.f32.bf16 %r1041, %rs506; + cvt.f32.bf16 %r1042, %rs507; + cvt.f32.bf16 %r1043, %rs508; + cvt.f32.bf16 %r1044, %rs509; + cvt.f32.bf16 %r1045, %rs510; + cvt.f32.bf16 %r1046, %rs511; + cvt.f32.bf16 %r1047, %rs512; + cvt.f32.bf16 %r1048, %rs513; + cvt.f32.bf16 %r1049, %rs514; + cvt.f32.bf16 %r1050, %rs515; + cvt.f32.bf16 %r1051, %rs516; + cvt.f32.bf16 %r1052, %rs517; + cvt.f32.bf16 %r1053, %rs518; + cvt.f32.bf16 %r1054, %rs519; + cvt.f32.bf16 %r1055, %rs520; + cvt.f32.bf16 %r1056, %rs521; + cvt.f32.bf16 %r1057, %rs522; + cvt.f32.bf16 %r1058, %rs523; + cvt.f32.bf16 %r1059, %rs524; + cvt.f32.bf16 %r1060, %rs525; + cvt.f32.bf16 %r1061, %rs526; + cvt.f32.bf16 %r1062, %rs527; + cvt.f32.bf16 %r1063, %rs528; + cvt.f32.bf16 %r1064, %rs529; + cvt.f32.bf16 %r1065, %rs530; + cvt.f32.bf16 %r1066, %rs531; + cvt.f32.bf16 %r1067, %rs532; + cvt.f32.bf16 %r1068, %rs533; + cvt.f32.bf16 %r1069, %rs534; + cvt.f32.bf16 %r1070, %rs535; + cvt.f32.bf16 %r1071, %rs536; + cvt.f32.bf16 %r1072, %rs537; + cvt.f32.bf16 %r1073, %rs538; + cvt.f32.bf16 %r1074, %rs539; + cvt.f32.bf16 %r1075, %rs540; + cvt.f32.bf16 %r1076, %rs541; + cvt.f32.bf16 %r1077, %rs542; + cvt.f32.bf16 %r1078, %rs543; + cvt.f32.bf16 %r1079, %rs544; + cvt.f32.bf16 %r1080, %rs545; + cvt.f32.bf16 %r1081, %rs546; + cvt.f32.bf16 %r1082, %rs547; + cvt.f32.bf16 %r1083, %rs548; + cvt.f32.bf16 %r1084, %rs549; + cvt.f32.bf16 %r1085, %rs550; + cvt.f32.bf16 %r1086, %rs551; + cvt.f32.bf16 %r1087, %rs552; + cvt.f32.bf16 %r1088, %rs553; + cvt.f32.bf16 %r1089, %rs554; + cvt.f32.bf16 %r1090, %rs555; + cvt.f32.bf16 %r1091, %rs556; + cvt.f32.bf16 %r1092, %rs557; + cvt.f32.bf16 %r1093, %rs558; + cvt.f32.bf16 %r1094, %rs559; + cvt.f32.bf16 %r1095, %rs560; + cvt.f32.bf16 %r1096, %rs561; + cvt.f32.bf16 %r1097, %rs562; + cvt.f32.bf16 %r1098, %rs563; + cvt.f32.bf16 %r1099, %rs564; + cvt.f32.bf16 %r1100, %rs565; + cvt.f32.bf16 %r1101, %rs566; + cvt.f32.bf16 %r1102, %rs567; + cvt.f32.bf16 %r1103, %rs568; + cvt.f32.bf16 %r1104, %rs569; + cvt.f32.bf16 %r1105, %rs570; + cvt.f32.bf16 %r1106, %rs571; + cvt.f32.bf16 %r1107, %rs572; + cvt.f32.bf16 %r1108, %rs573; + cvt.f32.bf16 %r1109, %rs574; + cvt.f32.bf16 %r1110, %rs575; + cvt.f32.bf16 %r1111, %rs576; + cvt.f32.bf16 %r1112, %rs577; + cvt.f32.bf16 %r1113, %rs578; + cvt.f32.bf16 %r1114, %rs579; + cvt.f32.bf16 %r1115, %rs580; + cvt.f32.bf16 %r1116, %rs581; + cvt.f32.bf16 %r1117, %rs582; + cvt.f32.bf16 %r1118, %rs583; + cvt.f32.bf16 %r1119, %rs584; + cvt.f32.bf16 %r1120, %rs585; + cvt.f32.bf16 %r1121, %rs586; + cvt.f32.bf16 %r1122, %rs587; + cvt.f32.bf16 %r1123, %rs588; + cvt.f32.bf16 %r1124, %rs589; + cvt.f32.bf16 %r1125, %rs590; + cvt.f32.bf16 %r1126, %rs591; + cvt.f32.bf16 %r1127, %rs592; + cvt.f32.bf16 %r1128, %rs593; + cvt.f32.bf16 %r1129, %rs594; + cvt.f32.bf16 %r1130, %rs595; + cvt.f32.bf16 %r1131, %rs596; + cvt.f32.bf16 %r1132, %rs597; + cvt.f32.bf16 %r1133, %rs598; + cvt.f32.bf16 %r1134, %rs599; + cvt.f32.bf16 %r1135, %rs600; + cvt.f32.bf16 %r1136, %rs601; + cvt.f32.bf16 %r1137, %rs602; + cvt.f32.bf16 %r1138, %rs603; + cvt.f32.bf16 %r1139, %rs604; + cvt.f32.bf16 %r1140, %rs605; + cvt.f32.bf16 %r1141, %rs606; + cvt.f32.bf16 %r1142, %rs607; + cvt.f32.bf16 %r1143, %rs608; + cvt.f32.bf16 %r1144, %rs609; + cvt.f32.bf16 %r1145, %rs610; + cvt.f32.bf16 %r1146, %rs611; + cvt.f32.bf16 %r1147, %rs612; + cvt.f32.bf16 %r1148, %rs613; + cvt.f32.bf16 %r1149, %rs614; + cvt.f32.bf16 %r1150, %rs615; + cvt.f32.bf16 %r1151, %rs616; + cvt.f32.bf16 %r1152, %rs617; + cvt.f32.bf16 %r1153, %rs618; + cvt.f32.bf16 %r1154, %rs619; + cvt.f32.bf16 %r1155, %rs620; + cvt.f32.bf16 %r1156, %rs621; + cvt.f32.bf16 %r1157, %rs622; + cvt.f32.bf16 %r1158, %rs623; + cvt.f32.bf16 %r1159, %rs624; + cvt.f32.bf16 %r1160, %rs625; + cvt.f32.bf16 %r1161, %rs626; + cvt.f32.bf16 %r1162, %rs627; + cvt.f32.bf16 %r1163, %rs628; + cvt.f32.bf16 %r1164, %rs629; + cvt.f32.bf16 %r1165, %rs630; + cvt.f32.bf16 %r1166, %rs631; + cvt.f32.bf16 %r1167, %rs632; + cvt.f32.bf16 %r1168, %rs633; + cvt.f32.bf16 %r1169, %rs634; + cvt.f32.bf16 %r1170, %rs635; + cvt.f32.bf16 %r1171, %rs636; + cvt.f32.bf16 %r1172, %rs637; + cvt.f32.bf16 %r1173, %rs638; + cvt.f32.bf16 %r1174, %rs639; + cvt.f32.bf16 %r1175, %rs640; + cvt.f32.bf16 %r1176, %rs641; + cvt.f32.bf16 %r1177, %rs642; + cvt.f32.bf16 %r1178, %rs643; + cvt.f32.bf16 %r1179, %rs644; + cvt.f32.bf16 %r1180, %rs645; + cvt.f32.bf16 %r1181, %rs646; + cvt.f32.bf16 %r1182, %rs647; + cvt.f32.bf16 %r1183, %rs648; + cvt.f32.bf16 %r1184, %rs649; + cvt.f32.bf16 %r1185, %rs650; + cvt.f32.bf16 %r1186, %rs651; + cvt.f32.bf16 %r1187, %rs652; + cvt.f32.bf16 %r1188, %rs653; + cvt.f32.bf16 %r1189, %rs654; + cvt.f32.bf16 %r1190, %rs655; + cvt.f32.bf16 %r1191, %rs656; + cvt.f32.bf16 %r1192, %rs657; + cvt.f32.bf16 %r1193, %rs658; + cvt.f32.bf16 %r1194, %rs659; + cvt.f32.bf16 %r1195, %rs660; + cvt.f32.bf16 %r1196, %rs661; + cvt.f32.bf16 %r1197, %rs662; + cvt.f32.bf16 %r1198, %rs663; + cvt.f32.bf16 %r1199, %rs664; + cvt.f32.bf16 %r1200, %rs665; + cvt.f32.bf16 %r1201, %rs666; + cvt.f32.bf16 %r1202, %rs667; + cvt.f32.bf16 %r1203, %rs668; + cvt.f32.bf16 %r1204, %rs669; + cvt.f32.bf16 %r1205, %rs670; + cvt.f32.bf16 %r1206, %rs671; + .loc 1 80 38 // l2norm.py:80:38 + mul.f32 %r1207, %r696, %r696; + mul.f32 %r1208, %r704, %r704; + mul.f32 %r1209, %r712, %r712; + mul.f32 %r1210, %r720, %r720; + mul.f32 %r1211, %r728, %r728; + mul.f32 %r1212, %r736, %r736; + mul.f32 %r1213, %r744, %r744; + mul.f32 %r1214, %r752, %r752; + mul.f32 %r1215, %r760, %r760; + mul.f32 %r1216, %r768, %r768; + mul.f32 %r1217, %r776, %r776; + mul.f32 %r1218, %r784, %r784; + mul.f32 %r1219, %r792, %r792; + mul.f32 %r1220, %r800, %r800; + mul.f32 %r1221, %r808, %r808; + mul.f32 %r1222, %r816, %r816; + mul.f32 %r1223, %r824, %r824; + mul.f32 %r1224, %r832, %r832; + mul.f32 %r1225, %r840, %r840; + mul.f32 %r1226, %r848, %r848; + mul.f32 %r1227, %r856, %r856; + mul.f32 %r1228, %r864, %r864; + mul.f32 %r1229, %r872, %r872; + mul.f32 %r1230, %r880, %r880; + mul.f32 %r1231, %r888, %r888; + mul.f32 %r1232, %r896, %r896; + mul.f32 %r1233, %r904, %r904; + mul.f32 %r1234, %r912, %r912; + mul.f32 %r1235, %r920, %r920; + mul.f32 %r1236, %r928, %r928; + mul.f32 %r1237, %r936, %r936; + mul.f32 %r1238, %r944, %r944; + mul.f32 %r1239, %r952, %r952; + mul.f32 %r1240, %r960, %r960; + mul.f32 %r1241, %r968, %r968; + mul.f32 %r1242, %r976, %r976; + mul.f32 %r1243, %r984, %r984; + mul.f32 %r1244, %r992, %r992; + mul.f32 %r1245, %r1000, %r1000; + mul.f32 %r1246, %r1008, %r1008; + mul.f32 %r1247, %r1016, %r1016; + mul.f32 %r1248, %r1024, %r1024; + mul.f32 %r1249, %r1032, %r1032; + mul.f32 %r1250, %r1040, %r1040; + mul.f32 %r1251, %r1048, %r1048; + mul.f32 %r1252, %r1056, %r1056; + mul.f32 %r1253, %r1064, %r1064; + mul.f32 %r1254, %r1072, %r1072; + mul.f32 %r1255, %r1080, %r1080; + mul.f32 %r1256, %r1088, %r1088; + mul.f32 %r1257, %r1096, %r1096; + mul.f32 %r1258, %r1104, %r1104; + mul.f32 %r1259, %r1112, %r1112; + mul.f32 %r1260, %r1120, %r1120; + mul.f32 %r1261, %r1128, %r1128; + mul.f32 %r1262, %r1136, %r1136; + mul.f32 %r1263, %r1144, %r1144; + mul.f32 %r1264, %r1152, %r1152; + mul.f32 %r1265, %r1160, %r1160; + mul.f32 %r1266, %r1168, %r1168; + mul.f32 %r1267, %r1176, %r1176; + mul.f32 %r1268, %r1184, %r1184; + mul.f32 %r1269, %r1192, %r1192; + mul.f32 %r1270, %r1200, %r1200; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + fma.rn.f32.bf16 %r1271, %rs160, %rs160, %r1207; + fma.rn.f32.bf16 %r1272, %rs162, %rs162, %r1271; + fma.rn.f32.bf16 %r1273, %rs163, %rs163, %r1272; + fma.rn.f32.bf16 %r1274, %rs164, %rs164, %r1273; + fma.rn.f32.bf16 %r1275, %rs165, %rs165, %r1274; + fma.rn.f32.bf16 %r1276, %rs166, %rs166, %r1275; + fma.rn.f32.bf16 %r1277, %rs167, %rs167, %r1276; + fma.rn.f32.bf16 %r1278, %rs168, %rs168, %r1208; + fma.rn.f32.bf16 %r1279, %rs170, %rs170, %r1278; + fma.rn.f32.bf16 %r1280, %rs171, %rs171, %r1279; + fma.rn.f32.bf16 %r1281, %rs172, %rs172, %r1280; + fma.rn.f32.bf16 %r1282, %rs173, %rs173, %r1281; + fma.rn.f32.bf16 %r1283, %rs174, %rs174, %r1282; + fma.rn.f32.bf16 %r1284, %rs175, %rs175, %r1283; + fma.rn.f32.bf16 %r1285, %rs176, %rs176, %r1209; + fma.rn.f32.bf16 %r1286, %rs178, %rs178, %r1285; + fma.rn.f32.bf16 %r1287, %rs179, %rs179, %r1286; + fma.rn.f32.bf16 %r1288, %rs180, %rs180, %r1287; + fma.rn.f32.bf16 %r1289, %rs181, %rs181, %r1288; + fma.rn.f32.bf16 %r1290, %rs182, %rs182, %r1289; + fma.rn.f32.bf16 %r1291, %rs183, %rs183, %r1290; + fma.rn.f32.bf16 %r1292, %rs184, %rs184, %r1210; + fma.rn.f32.bf16 %r1293, %rs186, %rs186, %r1292; + fma.rn.f32.bf16 %r1294, %rs187, %rs187, %r1293; + fma.rn.f32.bf16 %r1295, %rs188, %rs188, %r1294; + fma.rn.f32.bf16 %r1296, %rs189, %rs189, %r1295; + fma.rn.f32.bf16 %r1297, %rs190, %rs190, %r1296; + fma.rn.f32.bf16 %r1298, %rs191, %rs191, %r1297; + fma.rn.f32.bf16 %r1299, %rs192, %rs192, %r1211; + fma.rn.f32.bf16 %r1300, %rs194, %rs194, %r1299; + fma.rn.f32.bf16 %r1301, %rs195, %rs195, %r1300; + fma.rn.f32.bf16 %r1302, %rs196, %rs196, %r1301; + fma.rn.f32.bf16 %r1303, %rs197, %rs197, %r1302; + fma.rn.f32.bf16 %r1304, %rs198, %rs198, %r1303; + fma.rn.f32.bf16 %r1305, %rs199, %rs199, %r1304; + fma.rn.f32.bf16 %r1306, %rs200, %rs200, %r1212; + fma.rn.f32.bf16 %r1307, %rs202, %rs202, %r1306; + fma.rn.f32.bf16 %r1308, %rs203, %rs203, %r1307; + fma.rn.f32.bf16 %r1309, %rs204, %rs204, %r1308; + fma.rn.f32.bf16 %r1310, %rs205, %rs205, %r1309; + fma.rn.f32.bf16 %r1311, %rs206, %rs206, %r1310; + fma.rn.f32.bf16 %r1312, %rs207, %rs207, %r1311; + fma.rn.f32.bf16 %r1313, %rs208, %rs208, %r1213; + fma.rn.f32.bf16 %r1314, %rs210, %rs210, %r1313; + fma.rn.f32.bf16 %r1315, %rs211, %rs211, %r1314; + fma.rn.f32.bf16 %r1316, %rs212, %rs212, %r1315; + fma.rn.f32.bf16 %r1317, %rs213, %rs213, %r1316; + fma.rn.f32.bf16 %r1318, %rs214, %rs214, %r1317; + fma.rn.f32.bf16 %r1319, %rs215, %rs215, %r1318; + fma.rn.f32.bf16 %r1320, %rs216, %rs216, %r1214; + fma.rn.f32.bf16 %r1321, %rs218, %rs218, %r1320; + fma.rn.f32.bf16 %r1322, %rs219, %rs219, %r1321; + fma.rn.f32.bf16 %r1323, %rs220, %rs220, %r1322; + fma.rn.f32.bf16 %r1324, %rs221, %rs221, %r1323; + fma.rn.f32.bf16 %r1325, %rs222, %rs222, %r1324; + fma.rn.f32.bf16 %r1326, %rs223, %rs223, %r1325; + fma.rn.f32.bf16 %r1327, %rs224, %rs224, %r1215; + fma.rn.f32.bf16 %r1328, %rs226, %rs226, %r1327; + fma.rn.f32.bf16 %r1329, %rs227, %rs227, %r1328; + fma.rn.f32.bf16 %r1330, %rs228, %rs228, %r1329; + fma.rn.f32.bf16 %r1331, %rs229, %rs229, %r1330; + fma.rn.f32.bf16 %r1332, %rs230, %rs230, %r1331; + fma.rn.f32.bf16 %r1333, %rs231, %rs231, %r1332; + fma.rn.f32.bf16 %r1334, %rs232, %rs232, %r1216; + fma.rn.f32.bf16 %r1335, %rs234, %rs234, %r1334; + fma.rn.f32.bf16 %r1336, %rs235, %rs235, %r1335; + fma.rn.f32.bf16 %r1337, %rs236, %rs236, %r1336; + fma.rn.f32.bf16 %r1338, %rs237, %rs237, %r1337; + fma.rn.f32.bf16 %r1339, %rs238, %rs238, %r1338; + fma.rn.f32.bf16 %r1340, %rs239, %rs239, %r1339; + fma.rn.f32.bf16 %r1341, %rs240, %rs240, %r1217; + fma.rn.f32.bf16 %r1342, %rs242, %rs242, %r1341; + fma.rn.f32.bf16 %r1343, %rs243, %rs243, %r1342; + fma.rn.f32.bf16 %r1344, %rs244, %rs244, %r1343; + fma.rn.f32.bf16 %r1345, %rs245, %rs245, %r1344; + fma.rn.f32.bf16 %r1346, %rs246, %rs246, %r1345; + fma.rn.f32.bf16 %r1347, %rs247, %rs247, %r1346; + fma.rn.f32.bf16 %r1348, %rs248, %rs248, %r1218; + fma.rn.f32.bf16 %r1349, %rs250, %rs250, %r1348; + fma.rn.f32.bf16 %r1350, %rs251, %rs251, %r1349; + fma.rn.f32.bf16 %r1351, %rs252, %rs252, %r1350; + fma.rn.f32.bf16 %r1352, %rs253, %rs253, %r1351; + fma.rn.f32.bf16 %r1353, %rs254, %rs254, %r1352; + fma.rn.f32.bf16 %r1354, %rs255, %rs255, %r1353; + fma.rn.f32.bf16 %r1355, %rs256, %rs256, %r1219; + fma.rn.f32.bf16 %r1356, %rs258, %rs258, %r1355; + fma.rn.f32.bf16 %r1357, %rs259, %rs259, %r1356; + fma.rn.f32.bf16 %r1358, %rs260, %rs260, %r1357; + fma.rn.f32.bf16 %r1359, %rs261, %rs261, %r1358; + fma.rn.f32.bf16 %r1360, %rs262, %rs262, %r1359; + fma.rn.f32.bf16 %r1361, %rs263, %rs263, %r1360; + fma.rn.f32.bf16 %r1362, %rs264, %rs264, %r1220; + fma.rn.f32.bf16 %r1363, %rs266, %rs266, %r1362; + fma.rn.f32.bf16 %r1364, %rs267, %rs267, %r1363; + fma.rn.f32.bf16 %r1365, %rs268, %rs268, %r1364; + fma.rn.f32.bf16 %r1366, %rs269, %rs269, %r1365; + fma.rn.f32.bf16 %r1367, %rs270, %rs270, %r1366; + fma.rn.f32.bf16 %r1368, %rs271, %rs271, %r1367; + fma.rn.f32.bf16 %r1369, %rs272, %rs272, %r1221; + fma.rn.f32.bf16 %r1370, %rs274, %rs274, %r1369; + fma.rn.f32.bf16 %r1371, %rs275, %rs275, %r1370; + fma.rn.f32.bf16 %r1372, %rs276, %rs276, %r1371; + fma.rn.f32.bf16 %r1373, %rs277, %rs277, %r1372; + fma.rn.f32.bf16 %r1374, %rs278, %rs278, %r1373; + fma.rn.f32.bf16 %r1375, %rs279, %rs279, %r1374; + fma.rn.f32.bf16 %r1376, %rs280, %rs280, %r1222; + fma.rn.f32.bf16 %r1377, %rs282, %rs282, %r1376; + fma.rn.f32.bf16 %r1378, %rs283, %rs283, %r1377; + fma.rn.f32.bf16 %r1379, %rs284, %rs284, %r1378; + fma.rn.f32.bf16 %r1380, %rs285, %rs285, %r1379; + fma.rn.f32.bf16 %r1381, %rs286, %rs286, %r1380; + fma.rn.f32.bf16 %r1382, %rs287, %rs287, %r1381; + fma.rn.f32.bf16 %r1383, %rs288, %rs288, %r1223; + fma.rn.f32.bf16 %r1384, %rs290, %rs290, %r1383; + fma.rn.f32.bf16 %r1385, %rs291, %rs291, %r1384; + fma.rn.f32.bf16 %r1386, %rs292, %rs292, %r1385; + fma.rn.f32.bf16 %r1387, %rs293, %rs293, %r1386; + fma.rn.f32.bf16 %r1388, %rs294, %rs294, %r1387; + fma.rn.f32.bf16 %r1389, %rs295, %rs295, %r1388; + fma.rn.f32.bf16 %r1390, %rs296, %rs296, %r1224; + fma.rn.f32.bf16 %r1391, %rs298, %rs298, %r1390; + fma.rn.f32.bf16 %r1392, %rs299, %rs299, %r1391; + fma.rn.f32.bf16 %r1393, %rs300, %rs300, %r1392; + fma.rn.f32.bf16 %r1394, %rs301, %rs301, %r1393; + fma.rn.f32.bf16 %r1395, %rs302, %rs302, %r1394; + fma.rn.f32.bf16 %r1396, %rs303, %rs303, %r1395; + fma.rn.f32.bf16 %r1397, %rs304, %rs304, %r1225; + fma.rn.f32.bf16 %r1398, %rs306, %rs306, %r1397; + fma.rn.f32.bf16 %r1399, %rs307, %rs307, %r1398; + fma.rn.f32.bf16 %r1400, %rs308, %rs308, %r1399; + fma.rn.f32.bf16 %r1401, %rs309, %rs309, %r1400; + fma.rn.f32.bf16 %r1402, %rs310, %rs310, %r1401; + fma.rn.f32.bf16 %r1403, %rs311, %rs311, %r1402; + fma.rn.f32.bf16 %r1404, %rs312, %rs312, %r1226; + fma.rn.f32.bf16 %r1405, %rs314, %rs314, %r1404; + fma.rn.f32.bf16 %r1406, %rs315, %rs315, %r1405; + fma.rn.f32.bf16 %r1407, %rs316, %rs316, %r1406; + fma.rn.f32.bf16 %r1408, %rs317, %rs317, %r1407; + fma.rn.f32.bf16 %r1409, %rs318, %rs318, %r1408; + fma.rn.f32.bf16 %r1410, %rs319, %rs319, %r1409; + fma.rn.f32.bf16 %r1411, %rs320, %rs320, %r1227; + fma.rn.f32.bf16 %r1412, %rs322, %rs322, %r1411; + fma.rn.f32.bf16 %r1413, %rs323, %rs323, %r1412; + fma.rn.f32.bf16 %r1414, %rs324, %rs324, %r1413; + fma.rn.f32.bf16 %r1415, %rs325, %rs325, %r1414; + fma.rn.f32.bf16 %r1416, %rs326, %rs326, %r1415; + fma.rn.f32.bf16 %r1417, %rs327, %rs327, %r1416; + fma.rn.f32.bf16 %r1418, %rs328, %rs328, %r1228; + fma.rn.f32.bf16 %r1419, %rs330, %rs330, %r1418; + fma.rn.f32.bf16 %r1420, %rs331, %rs331, %r1419; + fma.rn.f32.bf16 %r1421, %rs332, %rs332, %r1420; + fma.rn.f32.bf16 %r1422, %rs333, %rs333, %r1421; + fma.rn.f32.bf16 %r1423, %rs334, %rs334, %r1422; + fma.rn.f32.bf16 %r1424, %rs335, %rs335, %r1423; + fma.rn.f32.bf16 %r1425, %rs336, %rs336, %r1229; + fma.rn.f32.bf16 %r1426, %rs338, %rs338, %r1425; + fma.rn.f32.bf16 %r1427, %rs339, %rs339, %r1426; + fma.rn.f32.bf16 %r1428, %rs340, %rs340, %r1427; + fma.rn.f32.bf16 %r1429, %rs341, %rs341, %r1428; + fma.rn.f32.bf16 %r1430, %rs342, %rs342, %r1429; + fma.rn.f32.bf16 %r1431, %rs343, %rs343, %r1430; + fma.rn.f32.bf16 %r1432, %rs344, %rs344, %r1230; + fma.rn.f32.bf16 %r1433, %rs346, %rs346, %r1432; + fma.rn.f32.bf16 %r1434, %rs347, %rs347, %r1433; + fma.rn.f32.bf16 %r1435, %rs348, %rs348, %r1434; + fma.rn.f32.bf16 %r1436, %rs349, %rs349, %r1435; + fma.rn.f32.bf16 %r1437, %rs350, %rs350, %r1436; + fma.rn.f32.bf16 %r1438, %rs351, %rs351, %r1437; + fma.rn.f32.bf16 %r1439, %rs352, %rs352, %r1231; + fma.rn.f32.bf16 %r1440, %rs354, %rs354, %r1439; + fma.rn.f32.bf16 %r1441, %rs355, %rs355, %r1440; + fma.rn.f32.bf16 %r1442, %rs356, %rs356, %r1441; + fma.rn.f32.bf16 %r1443, %rs357, %rs357, %r1442; + fma.rn.f32.bf16 %r1444, %rs358, %rs358, %r1443; + fma.rn.f32.bf16 %r1445, %rs359, %rs359, %r1444; + fma.rn.f32.bf16 %r1446, %rs360, %rs360, %r1232; + fma.rn.f32.bf16 %r1447, %rs362, %rs362, %r1446; + fma.rn.f32.bf16 %r1448, %rs363, %rs363, %r1447; + fma.rn.f32.bf16 %r1449, %rs364, %rs364, %r1448; + fma.rn.f32.bf16 %r1450, %rs365, %rs365, %r1449; + fma.rn.f32.bf16 %r1451, %rs366, %rs366, %r1450; + fma.rn.f32.bf16 %r1452, %rs367, %rs367, %r1451; + fma.rn.f32.bf16 %r1453, %rs368, %rs368, %r1233; + fma.rn.f32.bf16 %r1454, %rs370, %rs370, %r1453; + fma.rn.f32.bf16 %r1455, %rs371, %rs371, %r1454; + fma.rn.f32.bf16 %r1456, %rs372, %rs372, %r1455; + fma.rn.f32.bf16 %r1457, %rs373, %rs373, %r1456; + fma.rn.f32.bf16 %r1458, %rs374, %rs374, %r1457; + fma.rn.f32.bf16 %r1459, %rs375, %rs375, %r1458; + fma.rn.f32.bf16 %r1460, %rs376, %rs376, %r1234; + fma.rn.f32.bf16 %r1461, %rs378, %rs378, %r1460; + fma.rn.f32.bf16 %r1462, %rs379, %rs379, %r1461; + fma.rn.f32.bf16 %r1463, %rs380, %rs380, %r1462; + fma.rn.f32.bf16 %r1464, %rs381, %rs381, %r1463; + fma.rn.f32.bf16 %r1465, %rs382, %rs382, %r1464; + fma.rn.f32.bf16 %r1466, %rs383, %rs383, %r1465; + fma.rn.f32.bf16 %r1467, %rs384, %rs384, %r1235; + fma.rn.f32.bf16 %r1468, %rs386, %rs386, %r1467; + fma.rn.f32.bf16 %r1469, %rs387, %rs387, %r1468; + fma.rn.f32.bf16 %r1470, %rs388, %rs388, %r1469; + fma.rn.f32.bf16 %r1471, %rs389, %rs389, %r1470; + fma.rn.f32.bf16 %r1472, %rs390, %rs390, %r1471; + fma.rn.f32.bf16 %r1473, %rs391, %rs391, %r1472; + fma.rn.f32.bf16 %r1474, %rs392, %rs392, %r1236; + fma.rn.f32.bf16 %r1475, %rs394, %rs394, %r1474; + fma.rn.f32.bf16 %r1476, %rs395, %rs395, %r1475; + fma.rn.f32.bf16 %r1477, %rs396, %rs396, %r1476; + fma.rn.f32.bf16 %r1478, %rs397, %rs397, %r1477; + fma.rn.f32.bf16 %r1479, %rs398, %rs398, %r1478; + fma.rn.f32.bf16 %r1480, %rs399, %rs399, %r1479; + fma.rn.f32.bf16 %r1481, %rs400, %rs400, %r1237; + fma.rn.f32.bf16 %r1482, %rs402, %rs402, %r1481; + fma.rn.f32.bf16 %r1483, %rs403, %rs403, %r1482; + fma.rn.f32.bf16 %r1484, %rs404, %rs404, %r1483; + fma.rn.f32.bf16 %r1485, %rs405, %rs405, %r1484; + fma.rn.f32.bf16 %r1486, %rs406, %rs406, %r1485; + fma.rn.f32.bf16 %r1487, %rs407, %rs407, %r1486; + fma.rn.f32.bf16 %r1488, %rs408, %rs408, %r1238; + fma.rn.f32.bf16 %r1489, %rs410, %rs410, %r1488; + fma.rn.f32.bf16 %r1490, %rs411, %rs411, %r1489; + fma.rn.f32.bf16 %r1491, %rs412, %rs412, %r1490; + fma.rn.f32.bf16 %r1492, %rs413, %rs413, %r1491; + fma.rn.f32.bf16 %r1493, %rs414, %rs414, %r1492; + fma.rn.f32.bf16 %r1494, %rs415, %rs415, %r1493; + fma.rn.f32.bf16 %r1495, %rs416, %rs416, %r1239; + fma.rn.f32.bf16 %r1496, %rs418, %rs418, %r1495; + fma.rn.f32.bf16 %r1497, %rs419, %rs419, %r1496; + fma.rn.f32.bf16 %r1498, %rs420, %rs420, %r1497; + fma.rn.f32.bf16 %r1499, %rs421, %rs421, %r1498; + fma.rn.f32.bf16 %r1500, %rs422, %rs422, %r1499; + fma.rn.f32.bf16 %r1501, %rs423, %rs423, %r1500; + fma.rn.f32.bf16 %r1502, %rs424, %rs424, %r1240; + fma.rn.f32.bf16 %r1503, %rs426, %rs426, %r1502; + fma.rn.f32.bf16 %r1504, %rs427, %rs427, %r1503; + fma.rn.f32.bf16 %r1505, %rs428, %rs428, %r1504; + fma.rn.f32.bf16 %r1506, %rs429, %rs429, %r1505; + fma.rn.f32.bf16 %r1507, %rs430, %rs430, %r1506; + fma.rn.f32.bf16 %r1508, %rs431, %rs431, %r1507; + fma.rn.f32.bf16 %r1509, %rs432, %rs432, %r1241; + fma.rn.f32.bf16 %r1510, %rs434, %rs434, %r1509; + fma.rn.f32.bf16 %r1511, %rs435, %rs435, %r1510; + fma.rn.f32.bf16 %r1512, %rs436, %rs436, %r1511; + fma.rn.f32.bf16 %r1513, %rs437, %rs437, %r1512; + fma.rn.f32.bf16 %r1514, %rs438, %rs438, %r1513; + fma.rn.f32.bf16 %r1515, %rs439, %rs439, %r1514; + fma.rn.f32.bf16 %r1516, %rs440, %rs440, %r1242; + fma.rn.f32.bf16 %r1517, %rs442, %rs442, %r1516; + fma.rn.f32.bf16 %r1518, %rs443, %rs443, %r1517; + fma.rn.f32.bf16 %r1519, %rs444, %rs444, %r1518; + fma.rn.f32.bf16 %r1520, %rs445, %rs445, %r1519; + fma.rn.f32.bf16 %r1521, %rs446, %rs446, %r1520; + fma.rn.f32.bf16 %r1522, %rs447, %rs447, %r1521; + fma.rn.f32.bf16 %r1523, %rs448, %rs448, %r1243; + fma.rn.f32.bf16 %r1524, %rs450, %rs450, %r1523; + fma.rn.f32.bf16 %r1525, %rs451, %rs451, %r1524; + fma.rn.f32.bf16 %r1526, %rs452, %rs452, %r1525; + fma.rn.f32.bf16 %r1527, %rs453, %rs453, %r1526; + fma.rn.f32.bf16 %r1528, %rs454, %rs454, %r1527; + fma.rn.f32.bf16 %r1529, %rs455, %rs455, %r1528; + fma.rn.f32.bf16 %r1530, %rs456, %rs456, %r1244; + fma.rn.f32.bf16 %r1531, %rs458, %rs458, %r1530; + fma.rn.f32.bf16 %r1532, %rs459, %rs459, %r1531; + fma.rn.f32.bf16 %r1533, %rs460, %rs460, %r1532; + fma.rn.f32.bf16 %r1534, %rs461, %rs461, %r1533; + fma.rn.f32.bf16 %r1535, %rs462, %rs462, %r1534; + fma.rn.f32.bf16 %r1536, %rs463, %rs463, %r1535; + fma.rn.f32.bf16 %r1537, %rs464, %rs464, %r1245; + fma.rn.f32.bf16 %r1538, %rs466, %rs466, %r1537; + fma.rn.f32.bf16 %r1539, %rs467, %rs467, %r1538; + fma.rn.f32.bf16 %r1540, %rs468, %rs468, %r1539; + fma.rn.f32.bf16 %r1541, %rs469, %rs469, %r1540; + fma.rn.f32.bf16 %r1542, %rs470, %rs470, %r1541; + fma.rn.f32.bf16 %r1543, %rs471, %rs471, %r1542; + fma.rn.f32.bf16 %r1544, %rs472, %rs472, %r1246; + fma.rn.f32.bf16 %r1545, %rs474, %rs474, %r1544; + fma.rn.f32.bf16 %r1546, %rs475, %rs475, %r1545; + fma.rn.f32.bf16 %r1547, %rs476, %rs476, %r1546; + fma.rn.f32.bf16 %r1548, %rs477, %rs477, %r1547; + fma.rn.f32.bf16 %r1549, %rs478, %rs478, %r1548; + fma.rn.f32.bf16 %r1550, %rs479, %rs479, %r1549; + fma.rn.f32.bf16 %r1551, %rs480, %rs480, %r1247; + fma.rn.f32.bf16 %r1552, %rs482, %rs482, %r1551; + fma.rn.f32.bf16 %r1553, %rs483, %rs483, %r1552; + fma.rn.f32.bf16 %r1554, %rs484, %rs484, %r1553; + fma.rn.f32.bf16 %r1555, %rs485, %rs485, %r1554; + fma.rn.f32.bf16 %r1556, %rs486, %rs486, %r1555; + fma.rn.f32.bf16 %r1557, %rs487, %rs487, %r1556; + fma.rn.f32.bf16 %r1558, %rs488, %rs488, %r1248; + fma.rn.f32.bf16 %r1559, %rs490, %rs490, %r1558; + fma.rn.f32.bf16 %r1560, %rs491, %rs491, %r1559; + fma.rn.f32.bf16 %r1561, %rs492, %rs492, %r1560; + fma.rn.f32.bf16 %r1562, %rs493, %rs493, %r1561; + fma.rn.f32.bf16 %r1563, %rs494, %rs494, %r1562; + fma.rn.f32.bf16 %r1564, %rs495, %rs495, %r1563; + fma.rn.f32.bf16 %r1565, %rs496, %rs496, %r1249; + fma.rn.f32.bf16 %r1566, %rs498, %rs498, %r1565; + fma.rn.f32.bf16 %r1567, %rs499, %rs499, %r1566; + fma.rn.f32.bf16 %r1568, %rs500, %rs500, %r1567; + fma.rn.f32.bf16 %r1569, %rs501, %rs501, %r1568; + fma.rn.f32.bf16 %r1570, %rs502, %rs502, %r1569; + fma.rn.f32.bf16 %r1571, %rs503, %rs503, %r1570; + fma.rn.f32.bf16 %r1572, %rs504, %rs504, %r1250; + fma.rn.f32.bf16 %r1573, %rs506, %rs506, %r1572; + fma.rn.f32.bf16 %r1574, %rs507, %rs507, %r1573; + fma.rn.f32.bf16 %r1575, %rs508, %rs508, %r1574; + fma.rn.f32.bf16 %r1576, %rs509, %rs509, %r1575; + fma.rn.f32.bf16 %r1577, %rs510, %rs510, %r1576; + fma.rn.f32.bf16 %r1578, %rs511, %rs511, %r1577; + fma.rn.f32.bf16 %r1579, %rs512, %rs512, %r1251; + fma.rn.f32.bf16 %r1580, %rs514, %rs514, %r1579; + fma.rn.f32.bf16 %r1581, %rs515, %rs515, %r1580; + fma.rn.f32.bf16 %r1582, %rs516, %rs516, %r1581; + fma.rn.f32.bf16 %r1583, %rs517, %rs517, %r1582; + fma.rn.f32.bf16 %r1584, %rs518, %rs518, %r1583; + fma.rn.f32.bf16 %r1585, %rs519, %rs519, %r1584; + fma.rn.f32.bf16 %r1586, %rs520, %rs520, %r1252; + fma.rn.f32.bf16 %r1587, %rs522, %rs522, %r1586; + fma.rn.f32.bf16 %r1588, %rs523, %rs523, %r1587; + fma.rn.f32.bf16 %r1589, %rs524, %rs524, %r1588; + fma.rn.f32.bf16 %r1590, %rs525, %rs525, %r1589; + fma.rn.f32.bf16 %r1591, %rs526, %rs526, %r1590; + fma.rn.f32.bf16 %r1592, %rs527, %rs527, %r1591; + fma.rn.f32.bf16 %r1593, %rs528, %rs528, %r1253; + fma.rn.f32.bf16 %r1594, %rs530, %rs530, %r1593; + fma.rn.f32.bf16 %r1595, %rs531, %rs531, %r1594; + fma.rn.f32.bf16 %r1596, %rs532, %rs532, %r1595; + fma.rn.f32.bf16 %r1597, %rs533, %rs533, %r1596; + fma.rn.f32.bf16 %r1598, %rs534, %rs534, %r1597; + fma.rn.f32.bf16 %r1599, %rs535, %rs535, %r1598; + fma.rn.f32.bf16 %r1600, %rs536, %rs536, %r1254; + fma.rn.f32.bf16 %r1601, %rs538, %rs538, %r1600; + fma.rn.f32.bf16 %r1602, %rs539, %rs539, %r1601; + fma.rn.f32.bf16 %r1603, %rs540, %rs540, %r1602; + fma.rn.f32.bf16 %r1604, %rs541, %rs541, %r1603; + fma.rn.f32.bf16 %r1605, %rs542, %rs542, %r1604; + fma.rn.f32.bf16 %r1606, %rs543, %rs543, %r1605; + fma.rn.f32.bf16 %r1607, %rs544, %rs544, %r1255; + fma.rn.f32.bf16 %r1608, %rs546, %rs546, %r1607; + fma.rn.f32.bf16 %r1609, %rs547, %rs547, %r1608; + fma.rn.f32.bf16 %r1610, %rs548, %rs548, %r1609; + fma.rn.f32.bf16 %r1611, %rs549, %rs549, %r1610; + fma.rn.f32.bf16 %r1612, %rs550, %rs550, %r1611; + fma.rn.f32.bf16 %r1613, %rs551, %rs551, %r1612; + fma.rn.f32.bf16 %r1614, %rs552, %rs552, %r1256; + fma.rn.f32.bf16 %r1615, %rs554, %rs554, %r1614; + fma.rn.f32.bf16 %r1616, %rs555, %rs555, %r1615; + fma.rn.f32.bf16 %r1617, %rs556, %rs556, %r1616; + fma.rn.f32.bf16 %r1618, %rs557, %rs557, %r1617; + fma.rn.f32.bf16 %r1619, %rs558, %rs558, %r1618; + fma.rn.f32.bf16 %r1620, %rs559, %rs559, %r1619; + fma.rn.f32.bf16 %r1621, %rs560, %rs560, %r1257; + fma.rn.f32.bf16 %r1622, %rs562, %rs562, %r1621; + fma.rn.f32.bf16 %r1623, %rs563, %rs563, %r1622; + fma.rn.f32.bf16 %r1624, %rs564, %rs564, %r1623; + fma.rn.f32.bf16 %r1625, %rs565, %rs565, %r1624; + fma.rn.f32.bf16 %r1626, %rs566, %rs566, %r1625; + fma.rn.f32.bf16 %r1627, %rs567, %rs567, %r1626; + fma.rn.f32.bf16 %r1628, %rs568, %rs568, %r1258; + fma.rn.f32.bf16 %r1629, %rs570, %rs570, %r1628; + fma.rn.f32.bf16 %r1630, %rs571, %rs571, %r1629; + fma.rn.f32.bf16 %r1631, %rs572, %rs572, %r1630; + fma.rn.f32.bf16 %r1632, %rs573, %rs573, %r1631; + fma.rn.f32.bf16 %r1633, %rs574, %rs574, %r1632; + fma.rn.f32.bf16 %r1634, %rs575, %rs575, %r1633; + fma.rn.f32.bf16 %r1635, %rs576, %rs576, %r1259; + fma.rn.f32.bf16 %r1636, %rs578, %rs578, %r1635; + fma.rn.f32.bf16 %r1637, %rs579, %rs579, %r1636; + fma.rn.f32.bf16 %r1638, %rs580, %rs580, %r1637; + fma.rn.f32.bf16 %r1639, %rs581, %rs581, %r1638; + fma.rn.f32.bf16 %r1640, %rs582, %rs582, %r1639; + fma.rn.f32.bf16 %r1641, %rs583, %rs583, %r1640; + fma.rn.f32.bf16 %r1642, %rs584, %rs584, %r1260; + fma.rn.f32.bf16 %r1643, %rs586, %rs586, %r1642; + fma.rn.f32.bf16 %r1644, %rs587, %rs587, %r1643; + fma.rn.f32.bf16 %r1645, %rs588, %rs588, %r1644; + fma.rn.f32.bf16 %r1646, %rs589, %rs589, %r1645; + fma.rn.f32.bf16 %r1647, %rs590, %rs590, %r1646; + fma.rn.f32.bf16 %r1648, %rs591, %rs591, %r1647; + fma.rn.f32.bf16 %r1649, %rs592, %rs592, %r1261; + fma.rn.f32.bf16 %r1650, %rs594, %rs594, %r1649; + fma.rn.f32.bf16 %r1651, %rs595, %rs595, %r1650; + fma.rn.f32.bf16 %r1652, %rs596, %rs596, %r1651; + fma.rn.f32.bf16 %r1653, %rs597, %rs597, %r1652; + fma.rn.f32.bf16 %r1654, %rs598, %rs598, %r1653; + fma.rn.f32.bf16 %r1655, %rs599, %rs599, %r1654; + fma.rn.f32.bf16 %r1656, %rs600, %rs600, %r1262; + fma.rn.f32.bf16 %r1657, %rs602, %rs602, %r1656; + fma.rn.f32.bf16 %r1658, %rs603, %rs603, %r1657; + fma.rn.f32.bf16 %r1659, %rs604, %rs604, %r1658; + fma.rn.f32.bf16 %r1660, %rs605, %rs605, %r1659; + fma.rn.f32.bf16 %r1661, %rs606, %rs606, %r1660; + fma.rn.f32.bf16 %r1662, %rs607, %rs607, %r1661; + fma.rn.f32.bf16 %r1663, %rs608, %rs608, %r1263; + fma.rn.f32.bf16 %r1664, %rs610, %rs610, %r1663; + fma.rn.f32.bf16 %r1665, %rs611, %rs611, %r1664; + fma.rn.f32.bf16 %r1666, %rs612, %rs612, %r1665; + fma.rn.f32.bf16 %r1667, %rs613, %rs613, %r1666; + fma.rn.f32.bf16 %r1668, %rs614, %rs614, %r1667; + fma.rn.f32.bf16 %r1669, %rs615, %rs615, %r1668; + fma.rn.f32.bf16 %r1670, %rs616, %rs616, %r1264; + fma.rn.f32.bf16 %r1671, %rs618, %rs618, %r1670; + fma.rn.f32.bf16 %r1672, %rs619, %rs619, %r1671; + fma.rn.f32.bf16 %r1673, %rs620, %rs620, %r1672; + fma.rn.f32.bf16 %r1674, %rs621, %rs621, %r1673; + fma.rn.f32.bf16 %r1675, %rs622, %rs622, %r1674; + fma.rn.f32.bf16 %r1676, %rs623, %rs623, %r1675; + fma.rn.f32.bf16 %r1677, %rs624, %rs624, %r1265; + fma.rn.f32.bf16 %r1678, %rs626, %rs626, %r1677; + fma.rn.f32.bf16 %r1679, %rs627, %rs627, %r1678; + fma.rn.f32.bf16 %r1680, %rs628, %rs628, %r1679; + fma.rn.f32.bf16 %r1681, %rs629, %rs629, %r1680; + fma.rn.f32.bf16 %r1682, %rs630, %rs630, %r1681; + fma.rn.f32.bf16 %r1683, %rs631, %rs631, %r1682; + fma.rn.f32.bf16 %r1684, %rs632, %rs632, %r1266; + fma.rn.f32.bf16 %r1685, %rs634, %rs634, %r1684; + fma.rn.f32.bf16 %r1686, %rs635, %rs635, %r1685; + fma.rn.f32.bf16 %r1687, %rs636, %rs636, %r1686; + fma.rn.f32.bf16 %r1688, %rs637, %rs637, %r1687; + fma.rn.f32.bf16 %r1689, %rs638, %rs638, %r1688; + fma.rn.f32.bf16 %r1690, %rs639, %rs639, %r1689; + fma.rn.f32.bf16 %r1691, %rs640, %rs640, %r1267; + fma.rn.f32.bf16 %r1692, %rs642, %rs642, %r1691; + fma.rn.f32.bf16 %r1693, %rs643, %rs643, %r1692; + fma.rn.f32.bf16 %r1694, %rs644, %rs644, %r1693; + fma.rn.f32.bf16 %r1695, %rs645, %rs645, %r1694; + fma.rn.f32.bf16 %r1696, %rs646, %rs646, %r1695; + fma.rn.f32.bf16 %r1697, %rs647, %rs647, %r1696; + fma.rn.f32.bf16 %r1698, %rs648, %rs648, %r1268; + fma.rn.f32.bf16 %r1699, %rs650, %rs650, %r1698; + fma.rn.f32.bf16 %r1700, %rs651, %rs651, %r1699; + fma.rn.f32.bf16 %r1701, %rs652, %rs652, %r1700; + fma.rn.f32.bf16 %r1702, %rs653, %rs653, %r1701; + fma.rn.f32.bf16 %r1703, %rs654, %rs654, %r1702; + fma.rn.f32.bf16 %r1704, %rs655, %rs655, %r1703; + fma.rn.f32.bf16 %r1705, %rs656, %rs656, %r1269; + fma.rn.f32.bf16 %r1706, %rs658, %rs658, %r1705; + fma.rn.f32.bf16 %r1707, %rs659, %rs659, %r1706; + fma.rn.f32.bf16 %r1708, %rs660, %rs660, %r1707; + fma.rn.f32.bf16 %r1709, %rs661, %rs661, %r1708; + fma.rn.f32.bf16 %r1710, %rs662, %rs662, %r1709; + fma.rn.f32.bf16 %r1711, %rs663, %rs663, %r1710; + fma.rn.f32.bf16 %r1712, %rs664, %rs664, %r1270; + fma.rn.f32.bf16 %r1713, %rs666, %rs666, %r1712; + fma.rn.f32.bf16 %r1714, %rs667, %rs667, %r1713; + fma.rn.f32.bf16 %r1715, %rs668, %rs668, %r1714; + fma.rn.f32.bf16 %r1716, %rs669, %rs669, %r1715; + fma.rn.f32.bf16 %r1717, %rs670, %rs670, %r1716; + fma.rn.f32.bf16 %r1718, %rs671, %rs671, %r1717; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1719, %r1277, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1720, %r1277, %r1719; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1721, %r1720, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1722, %r1720, %r1721; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1723, %r1722, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1724, %r1722, %r1723; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1725, %r1724, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1726, %r1724, %r1725; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1727, %r1284, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1728, %r1284, %r1727; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1729, %r1728, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1730, %r1728, %r1729; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1731, %r1730, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1732, %r1730, %r1731; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1733, %r1732, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1734, %r1732, %r1733; +$L__tmp18: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1735, %r1291, 8, 31, -1; +$L__tmp19: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1736, %r1291, %r1735; +$L__tmp20: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1737, %r1736, 4, 31, -1; +$L__tmp21: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1738, %r1736, %r1737; +$L__tmp22: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1739, %r1738, 2, 31, -1; +$L__tmp23: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1740, %r1738, %r1739; +$L__tmp24: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1741, %r1740, 1, 31, -1; +$L__tmp25: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1742, %r1740, %r1741; +$L__tmp26: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1743, %r1298, 8, 31, -1; +$L__tmp27: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1744, %r1298, %r1743; +$L__tmp28: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1745, %r1744, 4, 31, -1; +$L__tmp29: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1746, %r1744, %r1745; +$L__tmp30: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1747, %r1746, 2, 31, -1; +$L__tmp31: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1748, %r1746, %r1747; +$L__tmp32: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1749, %r1748, 1, 31, -1; +$L__tmp33: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1750, %r1748, %r1749; +$L__tmp34: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1751, %r1305, 8, 31, -1; +$L__tmp35: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1752, %r1305, %r1751; +$L__tmp36: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1753, %r1752, 4, 31, -1; +$L__tmp37: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1754, %r1752, %r1753; +$L__tmp38: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1755, %r1754, 2, 31, -1; +$L__tmp39: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1756, %r1754, %r1755; +$L__tmp40: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1757, %r1756, 1, 31, -1; +$L__tmp41: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1758, %r1756, %r1757; +$L__tmp42: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1759, %r1312, 8, 31, -1; +$L__tmp43: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1760, %r1312, %r1759; +$L__tmp44: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1761, %r1760, 4, 31, -1; +$L__tmp45: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1762, %r1760, %r1761; +$L__tmp46: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1763, %r1762, 2, 31, -1; +$L__tmp47: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1764, %r1762, %r1763; +$L__tmp48: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1765, %r1764, 1, 31, -1; +$L__tmp49: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1766, %r1764, %r1765; +$L__tmp50: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1767, %r1319, 8, 31, -1; +$L__tmp51: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1768, %r1319, %r1767; +$L__tmp52: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1769, %r1768, 4, 31, -1; +$L__tmp53: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1770, %r1768, %r1769; +$L__tmp54: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1771, %r1770, 2, 31, -1; +$L__tmp55: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1772, %r1770, %r1771; +$L__tmp56: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1773, %r1772, 1, 31, -1; +$L__tmp57: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1774, %r1772, %r1773; +$L__tmp58: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1775, %r1326, 8, 31, -1; +$L__tmp59: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1776, %r1326, %r1775; +$L__tmp60: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1777, %r1776, 4, 31, -1; +$L__tmp61: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1778, %r1776, %r1777; +$L__tmp62: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1779, %r1778, 2, 31, -1; +$L__tmp63: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1780, %r1778, %r1779; +$L__tmp64: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1781, %r1780, 1, 31, -1; +$L__tmp65: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1782, %r1780, %r1781; +$L__tmp66: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1783, %r1333, 8, 31, -1; +$L__tmp67: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1784, %r1333, %r1783; +$L__tmp68: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1785, %r1784, 4, 31, -1; +$L__tmp69: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1786, %r1784, %r1785; +$L__tmp70: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1787, %r1786, 2, 31, -1; +$L__tmp71: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1788, %r1786, %r1787; +$L__tmp72: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1789, %r1788, 1, 31, -1; +$L__tmp73: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1790, %r1788, %r1789; +$L__tmp74: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1791, %r1340, 8, 31, -1; +$L__tmp75: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1792, %r1340, %r1791; +$L__tmp76: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1793, %r1792, 4, 31, -1; +$L__tmp77: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1794, %r1792, %r1793; +$L__tmp78: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1795, %r1794, 2, 31, -1; +$L__tmp79: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1796, %r1794, %r1795; +$L__tmp80: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1797, %r1796, 1, 31, -1; +$L__tmp81: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1798, %r1796, %r1797; +$L__tmp82: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1799, %r1347, 8, 31, -1; +$L__tmp83: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1800, %r1347, %r1799; +$L__tmp84: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1801, %r1800, 4, 31, -1; +$L__tmp85: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1802, %r1800, %r1801; +$L__tmp86: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1803, %r1802, 2, 31, -1; +$L__tmp87: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1804, %r1802, %r1803; +$L__tmp88: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1805, %r1804, 1, 31, -1; +$L__tmp89: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1806, %r1804, %r1805; +$L__tmp90: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1807, %r1354, 8, 31, -1; +$L__tmp91: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1808, %r1354, %r1807; +$L__tmp92: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1809, %r1808, 4, 31, -1; +$L__tmp93: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1810, %r1808, %r1809; +$L__tmp94: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1811, %r1810, 2, 31, -1; +$L__tmp95: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1812, %r1810, %r1811; +$L__tmp96: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1813, %r1812, 1, 31, -1; +$L__tmp97: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1814, %r1812, %r1813; +$L__tmp98: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1815, %r1361, 8, 31, -1; +$L__tmp99: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1816, %r1361, %r1815; +$L__tmp100: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1817, %r1816, 4, 31, -1; +$L__tmp101: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1818, %r1816, %r1817; +$L__tmp102: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1819, %r1818, 2, 31, -1; +$L__tmp103: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1820, %r1818, %r1819; +$L__tmp104: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1821, %r1820, 1, 31, -1; +$L__tmp105: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1822, %r1820, %r1821; +$L__tmp106: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1823, %r1368, 8, 31, -1; +$L__tmp107: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1824, %r1368, %r1823; +$L__tmp108: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1825, %r1824, 4, 31, -1; +$L__tmp109: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1826, %r1824, %r1825; +$L__tmp110: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1827, %r1826, 2, 31, -1; +$L__tmp111: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1828, %r1826, %r1827; +$L__tmp112: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1829, %r1828, 1, 31, -1; +$L__tmp113: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1830, %r1828, %r1829; +$L__tmp114: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1831, %r1375, 8, 31, -1; +$L__tmp115: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1832, %r1375, %r1831; +$L__tmp116: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1833, %r1832, 4, 31, -1; +$L__tmp117: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1834, %r1832, %r1833; +$L__tmp118: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1835, %r1834, 2, 31, -1; +$L__tmp119: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1836, %r1834, %r1835; +$L__tmp120: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1837, %r1836, 1, 31, -1; +$L__tmp121: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1838, %r1836, %r1837; +$L__tmp122: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1839, %r1382, 8, 31, -1; +$L__tmp123: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1840, %r1382, %r1839; +$L__tmp124: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1841, %r1840, 4, 31, -1; +$L__tmp125: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1842, %r1840, %r1841; +$L__tmp126: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1843, %r1842, 2, 31, -1; +$L__tmp127: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1844, %r1842, %r1843; +$L__tmp128: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1845, %r1844, 1, 31, -1; +$L__tmp129: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1846, %r1844, %r1845; +$L__tmp130: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1847, %r1389, 8, 31, -1; +$L__tmp131: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1848, %r1389, %r1847; +$L__tmp132: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1849, %r1848, 4, 31, -1; +$L__tmp133: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1850, %r1848, %r1849; +$L__tmp134: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1851, %r1850, 2, 31, -1; +$L__tmp135: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1852, %r1850, %r1851; +$L__tmp136: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1853, %r1852, 1, 31, -1; +$L__tmp137: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1854, %r1852, %r1853; +$L__tmp138: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1855, %r1396, 8, 31, -1; +$L__tmp139: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1856, %r1396, %r1855; +$L__tmp140: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1857, %r1856, 4, 31, -1; +$L__tmp141: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1858, %r1856, %r1857; +$L__tmp142: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1859, %r1858, 2, 31, -1; +$L__tmp143: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1860, %r1858, %r1859; +$L__tmp144: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1861, %r1860, 1, 31, -1; +$L__tmp145: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1862, %r1860, %r1861; +$L__tmp146: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1863, %r1403, 8, 31, -1; +$L__tmp147: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1864, %r1403, %r1863; +$L__tmp148: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1865, %r1864, 4, 31, -1; +$L__tmp149: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1866, %r1864, %r1865; +$L__tmp150: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1867, %r1866, 2, 31, -1; +$L__tmp151: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1868, %r1866, %r1867; +$L__tmp152: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1869, %r1868, 1, 31, -1; +$L__tmp153: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1870, %r1868, %r1869; +$L__tmp154: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1871, %r1410, 8, 31, -1; +$L__tmp155: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1872, %r1410, %r1871; +$L__tmp156: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1873, %r1872, 4, 31, -1; +$L__tmp157: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1874, %r1872, %r1873; +$L__tmp158: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1875, %r1874, 2, 31, -1; +$L__tmp159: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1876, %r1874, %r1875; +$L__tmp160: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1877, %r1876, 1, 31, -1; +$L__tmp161: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1878, %r1876, %r1877; +$L__tmp162: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1879, %r1417, 8, 31, -1; +$L__tmp163: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1880, %r1417, %r1879; +$L__tmp164: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1881, %r1880, 4, 31, -1; +$L__tmp165: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1882, %r1880, %r1881; +$L__tmp166: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1883, %r1882, 2, 31, -1; +$L__tmp167: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1884, %r1882, %r1883; +$L__tmp168: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1885, %r1884, 1, 31, -1; +$L__tmp169: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1886, %r1884, %r1885; +$L__tmp170: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1887, %r1424, 8, 31, -1; +$L__tmp171: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1888, %r1424, %r1887; +$L__tmp172: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1889, %r1888, 4, 31, -1; +$L__tmp173: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1890, %r1888, %r1889; +$L__tmp174: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1891, %r1890, 2, 31, -1; +$L__tmp175: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1892, %r1890, %r1891; +$L__tmp176: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1893, %r1892, 1, 31, -1; +$L__tmp177: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1894, %r1892, %r1893; +$L__tmp178: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1895, %r1431, 8, 31, -1; +$L__tmp179: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1896, %r1431, %r1895; +$L__tmp180: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1897, %r1896, 4, 31, -1; +$L__tmp181: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1898, %r1896, %r1897; +$L__tmp182: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1899, %r1898, 2, 31, -1; +$L__tmp183: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1900, %r1898, %r1899; +$L__tmp184: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1901, %r1900, 1, 31, -1; +$L__tmp185: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1902, %r1900, %r1901; +$L__tmp186: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1903, %r1438, 8, 31, -1; +$L__tmp187: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1904, %r1438, %r1903; +$L__tmp188: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1905, %r1904, 4, 31, -1; +$L__tmp189: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1906, %r1904, %r1905; +$L__tmp190: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1907, %r1906, 2, 31, -1; +$L__tmp191: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1908, %r1906, %r1907; +$L__tmp192: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1909, %r1908, 1, 31, -1; +$L__tmp193: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1910, %r1908, %r1909; +$L__tmp194: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1911, %r1445, 8, 31, -1; +$L__tmp195: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1912, %r1445, %r1911; +$L__tmp196: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1913, %r1912, 4, 31, -1; +$L__tmp197: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1914, %r1912, %r1913; +$L__tmp198: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1915, %r1914, 2, 31, -1; +$L__tmp199: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1916, %r1914, %r1915; +$L__tmp200: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1917, %r1916, 1, 31, -1; +$L__tmp201: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1918, %r1916, %r1917; +$L__tmp202: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1919, %r1452, 8, 31, -1; +$L__tmp203: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1920, %r1452, %r1919; +$L__tmp204: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1921, %r1920, 4, 31, -1; +$L__tmp205: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1922, %r1920, %r1921; +$L__tmp206: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1923, %r1922, 2, 31, -1; +$L__tmp207: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1924, %r1922, %r1923; +$L__tmp208: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1925, %r1924, 1, 31, -1; +$L__tmp209: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1926, %r1924, %r1925; +$L__tmp210: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1927, %r1459, 8, 31, -1; +$L__tmp211: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1928, %r1459, %r1927; +$L__tmp212: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1929, %r1928, 4, 31, -1; +$L__tmp213: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1930, %r1928, %r1929; +$L__tmp214: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1931, %r1930, 2, 31, -1; +$L__tmp215: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1932, %r1930, %r1931; +$L__tmp216: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1933, %r1932, 1, 31, -1; +$L__tmp217: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1934, %r1932, %r1933; +$L__tmp218: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1935, %r1466, 8, 31, -1; +$L__tmp219: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1936, %r1466, %r1935; +$L__tmp220: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1937, %r1936, 4, 31, -1; +$L__tmp221: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1938, %r1936, %r1937; +$L__tmp222: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1939, %r1938, 2, 31, -1; +$L__tmp223: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1940, %r1938, %r1939; +$L__tmp224: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1941, %r1940, 1, 31, -1; +$L__tmp225: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1942, %r1940, %r1941; +$L__tmp226: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1943, %r1473, 8, 31, -1; +$L__tmp227: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1944, %r1473, %r1943; +$L__tmp228: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1945, %r1944, 4, 31, -1; +$L__tmp229: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1946, %r1944, %r1945; +$L__tmp230: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1947, %r1946, 2, 31, -1; +$L__tmp231: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1948, %r1946, %r1947; +$L__tmp232: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1949, %r1948, 1, 31, -1; +$L__tmp233: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1950, %r1948, %r1949; +$L__tmp234: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1951, %r1480, 8, 31, -1; +$L__tmp235: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1952, %r1480, %r1951; +$L__tmp236: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1953, %r1952, 4, 31, -1; +$L__tmp237: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1954, %r1952, %r1953; +$L__tmp238: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1955, %r1954, 2, 31, -1; +$L__tmp239: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1956, %r1954, %r1955; +$L__tmp240: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1957, %r1956, 1, 31, -1; +$L__tmp241: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1958, %r1956, %r1957; +$L__tmp242: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1959, %r1487, 8, 31, -1; +$L__tmp243: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1960, %r1487, %r1959; +$L__tmp244: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1961, %r1960, 4, 31, -1; +$L__tmp245: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1962, %r1960, %r1961; +$L__tmp246: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1963, %r1962, 2, 31, -1; +$L__tmp247: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1964, %r1962, %r1963; +$L__tmp248: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1965, %r1964, 1, 31, -1; +$L__tmp249: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1966, %r1964, %r1965; +$L__tmp250: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1967, %r1494, 8, 31, -1; +$L__tmp251: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1968, %r1494, %r1967; +$L__tmp252: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1969, %r1968, 4, 31, -1; +$L__tmp253: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1970, %r1968, %r1969; +$L__tmp254: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1971, %r1970, 2, 31, -1; +$L__tmp255: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1972, %r1970, %r1971; +$L__tmp256: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1973, %r1972, 1, 31, -1; +$L__tmp257: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1974, %r1972, %r1973; +$L__tmp258: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1975, %r1501, 8, 31, -1; +$L__tmp259: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1976, %r1501, %r1975; +$L__tmp260: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1977, %r1976, 4, 31, -1; +$L__tmp261: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1978, %r1976, %r1977; +$L__tmp262: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1979, %r1978, 2, 31, -1; +$L__tmp263: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1980, %r1978, %r1979; +$L__tmp264: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1981, %r1980, 1, 31, -1; +$L__tmp265: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1982, %r1980, %r1981; +$L__tmp266: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1983, %r1508, 8, 31, -1; +$L__tmp267: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1984, %r1508, %r1983; +$L__tmp268: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1985, %r1984, 4, 31, -1; +$L__tmp269: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1986, %r1984, %r1985; +$L__tmp270: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1987, %r1986, 2, 31, -1; +$L__tmp271: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1988, %r1986, %r1987; +$L__tmp272: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1989, %r1988, 1, 31, -1; +$L__tmp273: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1990, %r1988, %r1989; +$L__tmp274: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1991, %r1515, 8, 31, -1; +$L__tmp275: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1992, %r1515, %r1991; +$L__tmp276: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1993, %r1992, 4, 31, -1; +$L__tmp277: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1994, %r1992, %r1993; +$L__tmp278: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1995, %r1994, 2, 31, -1; +$L__tmp279: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1996, %r1994, %r1995; +$L__tmp280: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1997, %r1996, 1, 31, -1; +$L__tmp281: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1998, %r1996, %r1997; +$L__tmp282: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1999, %r1522, 8, 31, -1; +$L__tmp283: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2000, %r1522, %r1999; +$L__tmp284: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2001, %r2000, 4, 31, -1; +$L__tmp285: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2002, %r2000, %r2001; +$L__tmp286: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2003, %r2002, 2, 31, -1; +$L__tmp287: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2004, %r2002, %r2003; +$L__tmp288: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2005, %r2004, 1, 31, -1; +$L__tmp289: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2006, %r2004, %r2005; +$L__tmp290: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2007, %r1529, 8, 31, -1; +$L__tmp291: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2008, %r1529, %r2007; +$L__tmp292: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2009, %r2008, 4, 31, -1; +$L__tmp293: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2010, %r2008, %r2009; +$L__tmp294: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2011, %r2010, 2, 31, -1; +$L__tmp295: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2012, %r2010, %r2011; +$L__tmp296: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2013, %r2012, 1, 31, -1; +$L__tmp297: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2014, %r2012, %r2013; +$L__tmp298: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2015, %r1536, 8, 31, -1; +$L__tmp299: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2016, %r1536, %r2015; +$L__tmp300: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2017, %r2016, 4, 31, -1; +$L__tmp301: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2018, %r2016, %r2017; +$L__tmp302: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2019, %r2018, 2, 31, -1; +$L__tmp303: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2020, %r2018, %r2019; +$L__tmp304: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2021, %r2020, 1, 31, -1; +$L__tmp305: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2022, %r2020, %r2021; +$L__tmp306: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2023, %r1543, 8, 31, -1; +$L__tmp307: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2024, %r1543, %r2023; +$L__tmp308: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2025, %r2024, 4, 31, -1; +$L__tmp309: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2026, %r2024, %r2025; +$L__tmp310: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2027, %r2026, 2, 31, -1; +$L__tmp311: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2028, %r2026, %r2027; +$L__tmp312: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2029, %r2028, 1, 31, -1; +$L__tmp313: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2030, %r2028, %r2029; +$L__tmp314: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2031, %r1550, 8, 31, -1; +$L__tmp315: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2032, %r1550, %r2031; +$L__tmp316: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2033, %r2032, 4, 31, -1; +$L__tmp317: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2034, %r2032, %r2033; +$L__tmp318: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2035, %r2034, 2, 31, -1; +$L__tmp319: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2036, %r2034, %r2035; +$L__tmp320: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2037, %r2036, 1, 31, -1; +$L__tmp321: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2038, %r2036, %r2037; +$L__tmp322: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2039, %r1557, 8, 31, -1; +$L__tmp323: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2040, %r1557, %r2039; +$L__tmp324: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2041, %r2040, 4, 31, -1; +$L__tmp325: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2042, %r2040, %r2041; +$L__tmp326: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2043, %r2042, 2, 31, -1; +$L__tmp327: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2044, %r2042, %r2043; +$L__tmp328: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2045, %r2044, 1, 31, -1; +$L__tmp329: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2046, %r2044, %r2045; +$L__tmp330: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2047, %r1564, 8, 31, -1; +$L__tmp331: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2048, %r1564, %r2047; +$L__tmp332: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2049, %r2048, 4, 31, -1; +$L__tmp333: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2050, %r2048, %r2049; +$L__tmp334: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2051, %r2050, 2, 31, -1; +$L__tmp335: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2052, %r2050, %r2051; +$L__tmp336: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2053, %r2052, 1, 31, -1; +$L__tmp337: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2054, %r2052, %r2053; +$L__tmp338: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2055, %r1571, 8, 31, -1; +$L__tmp339: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2056, %r1571, %r2055; +$L__tmp340: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2057, %r2056, 4, 31, -1; +$L__tmp341: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2058, %r2056, %r2057; +$L__tmp342: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2059, %r2058, 2, 31, -1; +$L__tmp343: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2060, %r2058, %r2059; +$L__tmp344: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2061, %r2060, 1, 31, -1; +$L__tmp345: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2062, %r2060, %r2061; +$L__tmp346: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2063, %r1578, 8, 31, -1; +$L__tmp347: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2064, %r1578, %r2063; +$L__tmp348: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2065, %r2064, 4, 31, -1; +$L__tmp349: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2066, %r2064, %r2065; +$L__tmp350: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2067, %r2066, 2, 31, -1; +$L__tmp351: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2068, %r2066, %r2067; +$L__tmp352: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2069, %r2068, 1, 31, -1; +$L__tmp353: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2070, %r2068, %r2069; +$L__tmp354: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2071, %r1585, 8, 31, -1; +$L__tmp355: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2072, %r1585, %r2071; +$L__tmp356: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2073, %r2072, 4, 31, -1; +$L__tmp357: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2074, %r2072, %r2073; +$L__tmp358: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2075, %r2074, 2, 31, -1; +$L__tmp359: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2076, %r2074, %r2075; +$L__tmp360: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2077, %r2076, 1, 31, -1; +$L__tmp361: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2078, %r2076, %r2077; +$L__tmp362: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2079, %r1592, 8, 31, -1; +$L__tmp363: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2080, %r1592, %r2079; +$L__tmp364: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2081, %r2080, 4, 31, -1; +$L__tmp365: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2082, %r2080, %r2081; +$L__tmp366: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2083, %r2082, 2, 31, -1; +$L__tmp367: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2084, %r2082, %r2083; +$L__tmp368: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2085, %r2084, 1, 31, -1; +$L__tmp369: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2086, %r2084, %r2085; +$L__tmp370: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2087, %r1599, 8, 31, -1; +$L__tmp371: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2088, %r1599, %r2087; +$L__tmp372: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2089, %r2088, 4, 31, -1; +$L__tmp373: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2090, %r2088, %r2089; +$L__tmp374: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2091, %r2090, 2, 31, -1; +$L__tmp375: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2092, %r2090, %r2091; +$L__tmp376: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2093, %r2092, 1, 31, -1; +$L__tmp377: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2094, %r2092, %r2093; +$L__tmp378: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2095, %r1606, 8, 31, -1; +$L__tmp379: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2096, %r1606, %r2095; +$L__tmp380: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2097, %r2096, 4, 31, -1; +$L__tmp381: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2098, %r2096, %r2097; +$L__tmp382: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2099, %r2098, 2, 31, -1; +$L__tmp383: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2100, %r2098, %r2099; +$L__tmp384: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2101, %r2100, 1, 31, -1; +$L__tmp385: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2102, %r2100, %r2101; +$L__tmp386: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2103, %r1613, 8, 31, -1; +$L__tmp387: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2104, %r1613, %r2103; +$L__tmp388: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2105, %r2104, 4, 31, -1; +$L__tmp389: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2106, %r2104, %r2105; +$L__tmp390: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2107, %r2106, 2, 31, -1; +$L__tmp391: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2108, %r2106, %r2107; +$L__tmp392: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2109, %r2108, 1, 31, -1; +$L__tmp393: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2110, %r2108, %r2109; +$L__tmp394: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2111, %r1620, 8, 31, -1; +$L__tmp395: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2112, %r1620, %r2111; +$L__tmp396: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2113, %r2112, 4, 31, -1; +$L__tmp397: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2114, %r2112, %r2113; +$L__tmp398: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2115, %r2114, 2, 31, -1; +$L__tmp399: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2116, %r2114, %r2115; +$L__tmp400: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2117, %r2116, 1, 31, -1; +$L__tmp401: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2118, %r2116, %r2117; +$L__tmp402: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2119, %r1627, 8, 31, -1; +$L__tmp403: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2120, %r1627, %r2119; +$L__tmp404: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2121, %r2120, 4, 31, -1; +$L__tmp405: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2122, %r2120, %r2121; +$L__tmp406: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2123, %r2122, 2, 31, -1; +$L__tmp407: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2124, %r2122, %r2123; +$L__tmp408: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2125, %r2124, 1, 31, -1; +$L__tmp409: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2126, %r2124, %r2125; +$L__tmp410: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2127, %r1634, 8, 31, -1; +$L__tmp411: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2128, %r1634, %r2127; +$L__tmp412: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2129, %r2128, 4, 31, -1; +$L__tmp413: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2130, %r2128, %r2129; +$L__tmp414: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2131, %r2130, 2, 31, -1; +$L__tmp415: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2132, %r2130, %r2131; +$L__tmp416: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2133, %r2132, 1, 31, -1; +$L__tmp417: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2134, %r2132, %r2133; +$L__tmp418: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2135, %r1641, 8, 31, -1; +$L__tmp419: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2136, %r1641, %r2135; +$L__tmp420: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2137, %r2136, 4, 31, -1; +$L__tmp421: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2138, %r2136, %r2137; +$L__tmp422: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2139, %r2138, 2, 31, -1; +$L__tmp423: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2140, %r2138, %r2139; +$L__tmp424: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2141, %r2140, 1, 31, -1; +$L__tmp425: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2142, %r2140, %r2141; +$L__tmp426: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2143, %r1648, 8, 31, -1; +$L__tmp427: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2144, %r1648, %r2143; +$L__tmp428: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2145, %r2144, 4, 31, -1; +$L__tmp429: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2146, %r2144, %r2145; +$L__tmp430: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2147, %r2146, 2, 31, -1; +$L__tmp431: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2148, %r2146, %r2147; +$L__tmp432: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2149, %r2148, 1, 31, -1; +$L__tmp433: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2150, %r2148, %r2149; +$L__tmp434: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2151, %r1655, 8, 31, -1; +$L__tmp435: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2152, %r1655, %r2151; +$L__tmp436: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2153, %r2152, 4, 31, -1; +$L__tmp437: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2154, %r2152, %r2153; +$L__tmp438: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2155, %r2154, 2, 31, -1; +$L__tmp439: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2156, %r2154, %r2155; +$L__tmp440: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2157, %r2156, 1, 31, -1; +$L__tmp441: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2158, %r2156, %r2157; +$L__tmp442: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2159, %r1662, 8, 31, -1; +$L__tmp443: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2160, %r1662, %r2159; +$L__tmp444: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2161, %r2160, 4, 31, -1; +$L__tmp445: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2162, %r2160, %r2161; +$L__tmp446: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2163, %r2162, 2, 31, -1; +$L__tmp447: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2164, %r2162, %r2163; +$L__tmp448: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2165, %r2164, 1, 31, -1; +$L__tmp449: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2166, %r2164, %r2165; +$L__tmp450: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2167, %r1669, 8, 31, -1; +$L__tmp451: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2168, %r1669, %r2167; +$L__tmp452: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2169, %r2168, 4, 31, -1; +$L__tmp453: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2170, %r2168, %r2169; +$L__tmp454: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2171, %r2170, 2, 31, -1; +$L__tmp455: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2172, %r2170, %r2171; +$L__tmp456: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2173, %r2172, 1, 31, -1; +$L__tmp457: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2174, %r2172, %r2173; +$L__tmp458: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2175, %r1676, 8, 31, -1; +$L__tmp459: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2176, %r1676, %r2175; +$L__tmp460: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2177, %r2176, 4, 31, -1; +$L__tmp461: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2178, %r2176, %r2177; +$L__tmp462: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2179, %r2178, 2, 31, -1; +$L__tmp463: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2180, %r2178, %r2179; +$L__tmp464: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2181, %r2180, 1, 31, -1; +$L__tmp465: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2182, %r2180, %r2181; +$L__tmp466: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2183, %r1683, 8, 31, -1; +$L__tmp467: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2184, %r1683, %r2183; +$L__tmp468: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2185, %r2184, 4, 31, -1; +$L__tmp469: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2186, %r2184, %r2185; +$L__tmp470: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2187, %r2186, 2, 31, -1; +$L__tmp471: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2188, %r2186, %r2187; +$L__tmp472: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2189, %r2188, 1, 31, -1; +$L__tmp473: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2190, %r2188, %r2189; +$L__tmp474: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2191, %r1690, 8, 31, -1; +$L__tmp475: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2192, %r1690, %r2191; +$L__tmp476: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2193, %r2192, 4, 31, -1; +$L__tmp477: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2194, %r2192, %r2193; +$L__tmp478: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2195, %r2194, 2, 31, -1; +$L__tmp479: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2196, %r2194, %r2195; +$L__tmp480: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2197, %r2196, 1, 31, -1; +$L__tmp481: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2198, %r2196, %r2197; +$L__tmp482: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2199, %r1697, 8, 31, -1; +$L__tmp483: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2200, %r1697, %r2199; +$L__tmp484: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2201, %r2200, 4, 31, -1; +$L__tmp485: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2202, %r2200, %r2201; +$L__tmp486: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2203, %r2202, 2, 31, -1; +$L__tmp487: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2204, %r2202, %r2203; +$L__tmp488: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2205, %r2204, 1, 31, -1; +$L__tmp489: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2206, %r2204, %r2205; +$L__tmp490: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2207, %r1704, 8, 31, -1; +$L__tmp491: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2208, %r1704, %r2207; +$L__tmp492: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2209, %r2208, 4, 31, -1; +$L__tmp493: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2210, %r2208, %r2209; +$L__tmp494: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2211, %r2210, 2, 31, -1; +$L__tmp495: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2212, %r2210, %r2211; +$L__tmp496: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2213, %r2212, 1, 31, -1; +$L__tmp497: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2214, %r2212, %r2213; +$L__tmp498: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2215, %r1711, 8, 31, -1; +$L__tmp499: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2216, %r1711, %r2215; +$L__tmp500: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2217, %r2216, 4, 31, -1; +$L__tmp501: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2218, %r2216, %r2217; +$L__tmp502: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2219, %r2218, 2, 31, -1; +$L__tmp503: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2220, %r2218, %r2219; +$L__tmp504: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2221, %r2220, 1, 31, -1; +$L__tmp505: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2222, %r2220, %r2221; +$L__tmp506: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2223, %r1718, 8, 31, -1; +$L__tmp507: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2224, %r1718, %r2223; +$L__tmp508: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2225, %r2224, 4, 31, -1; +$L__tmp509: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2226, %r2224, %r2225; +$L__tmp510: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2227, %r2226, 2, 31, -1; +$L__tmp511: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2228, %r2226, %r2227; +$L__tmp512: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r2229, %r2228, 1, 31, -1; +$L__tmp513: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r2230, %r2228, %r2229; +$L__tmp514: + .loc 1 80 48 // l2norm.py:80:48 + add.f32 %r2231, %r591, %r1726; + add.f32 %r2232, %r591, %r1734; + add.f32 %r2233, %r591, %r1742; + add.f32 %r2234, %r591, %r1750; + add.f32 %r2235, %r591, %r1758; + add.f32 %r2236, %r591, %r1766; + add.f32 %r2237, %r591, %r1774; + add.f32 %r2238, %r591, %r1782; + add.f32 %r2239, %r591, %r1790; + add.f32 %r2240, %r591, %r1798; + add.f32 %r2241, %r591, %r1806; + add.f32 %r2242, %r591, %r1814; + add.f32 %r2243, %r591, %r1822; + add.f32 %r2244, %r591, %r1830; + add.f32 %r2245, %r591, %r1838; + add.f32 %r2246, %r591, %r1846; + add.f32 %r2247, %r591, %r1854; + add.f32 %r2248, %r591, %r1862; + add.f32 %r2249, %r591, %r1870; + add.f32 %r2250, %r591, %r1878; + add.f32 %r2251, %r591, %r1886; + add.f32 %r2252, %r591, %r1894; + add.f32 %r2253, %r591, %r1902; + add.f32 %r2254, %r591, %r1910; + add.f32 %r2255, %r591, %r1918; + add.f32 %r2256, %r591, %r1926; + add.f32 %r2257, %r591, %r1934; + add.f32 %r2258, %r591, %r1942; + add.f32 %r2259, %r591, %r1950; + add.f32 %r2260, %r591, %r1958; + add.f32 %r2261, %r591, %r1966; + add.f32 %r2262, %r591, %r1974; + add.f32 %r2263, %r591, %r1982; + add.f32 %r2264, %r591, %r1990; + add.f32 %r2265, %r591, %r1998; + add.f32 %r2266, %r591, %r2006; + add.f32 %r2267, %r591, %r2014; + add.f32 %r2268, %r591, %r2022; + add.f32 %r2269, %r591, %r2030; + add.f32 %r2270, %r591, %r2038; + add.f32 %r2271, %r591, %r2046; + add.f32 %r2272, %r591, %r2054; + add.f32 %r2273, %r591, %r2062; + add.f32 %r2274, %r591, %r2070; + add.f32 %r2275, %r591, %r2078; + add.f32 %r2276, %r591, %r2086; + add.f32 %r2277, %r591, %r2094; + add.f32 %r2278, %r591, %r2102; + add.f32 %r2279, %r591, %r2110; + add.f32 %r2280, %r591, %r2118; + add.f32 %r2281, %r591, %r2126; + add.f32 %r2282, %r591, %r2134; + add.f32 %r2283, %r591, %r2142; + add.f32 %r2284, %r591, %r2150; + add.f32 %r2285, %r591, %r2158; + add.f32 %r2286, %r591, %r2166; + add.f32 %r2287, %r591, %r2174; + add.f32 %r2288, %r591, %r2182; + add.f32 %r2289, %r591, %r2190; + add.f32 %r2290, %r591, %r2198; + add.f32 %r2291, %r591, %r2206; + add.f32 %r2292, %r591, %r2214; + add.f32 %r2293, %r591, %r2222; + add.f32 %r2294, %r591, %r2230; + .loc 1 80 25 // l2norm.py:80:25 + sqrt.approx.ftz.f32 %r2295, %r2231; + sqrt.approx.ftz.f32 %r2296, %r2232; + sqrt.approx.ftz.f32 %r2297, %r2233; + sqrt.approx.ftz.f32 %r2298, %r2234; + sqrt.approx.ftz.f32 %r2299, %r2235; + sqrt.approx.ftz.f32 %r2300, %r2236; + sqrt.approx.ftz.f32 %r2301, %r2237; + sqrt.approx.ftz.f32 %r2302, %r2238; + sqrt.approx.ftz.f32 %r2303, %r2239; + sqrt.approx.ftz.f32 %r2304, %r2240; + sqrt.approx.ftz.f32 %r2305, %r2241; + sqrt.approx.ftz.f32 %r2306, %r2242; + sqrt.approx.ftz.f32 %r2307, %r2243; + sqrt.approx.ftz.f32 %r2308, %r2244; + sqrt.approx.ftz.f32 %r2309, %r2245; + sqrt.approx.ftz.f32 %r2310, %r2246; + sqrt.approx.ftz.f32 %r2311, %r2247; + sqrt.approx.ftz.f32 %r2312, %r2248; + sqrt.approx.ftz.f32 %r2313, %r2249; + sqrt.approx.ftz.f32 %r2314, %r2250; + sqrt.approx.ftz.f32 %r2315, %r2251; + sqrt.approx.ftz.f32 %r2316, %r2252; + sqrt.approx.ftz.f32 %r2317, %r2253; + sqrt.approx.ftz.f32 %r2318, %r2254; + sqrt.approx.ftz.f32 %r2319, %r2255; + sqrt.approx.ftz.f32 %r2320, %r2256; + sqrt.approx.ftz.f32 %r2321, %r2257; + sqrt.approx.ftz.f32 %r2322, %r2258; + sqrt.approx.ftz.f32 %r2323, %r2259; + sqrt.approx.ftz.f32 %r2324, %r2260; + sqrt.approx.ftz.f32 %r2325, %r2261; + sqrt.approx.ftz.f32 %r2326, %r2262; + sqrt.approx.ftz.f32 %r2327, %r2263; + sqrt.approx.ftz.f32 %r2328, %r2264; + sqrt.approx.ftz.f32 %r2329, %r2265; + sqrt.approx.ftz.f32 %r2330, %r2266; + sqrt.approx.ftz.f32 %r2331, %r2267; + sqrt.approx.ftz.f32 %r2332, %r2268; + sqrt.approx.ftz.f32 %r2333, %r2269; + sqrt.approx.ftz.f32 %r2334, %r2270; + sqrt.approx.ftz.f32 %r2335, %r2271; + sqrt.approx.ftz.f32 %r2336, %r2272; + sqrt.approx.ftz.f32 %r2337, %r2273; + sqrt.approx.ftz.f32 %r2338, %r2274; + sqrt.approx.ftz.f32 %r2339, %r2275; + sqrt.approx.ftz.f32 %r2340, %r2276; + sqrt.approx.ftz.f32 %r2341, %r2277; + sqrt.approx.ftz.f32 %r2342, %r2278; + sqrt.approx.ftz.f32 %r2343, %r2279; + sqrt.approx.ftz.f32 %r2344, %r2280; + sqrt.approx.ftz.f32 %r2345, %r2281; + sqrt.approx.ftz.f32 %r2346, %r2282; + sqrt.approx.ftz.f32 %r2347, %r2283; + sqrt.approx.ftz.f32 %r2348, %r2284; + sqrt.approx.ftz.f32 %r2349, %r2285; + sqrt.approx.ftz.f32 %r2350, %r2286; + sqrt.approx.ftz.f32 %r2351, %r2287; + sqrt.approx.ftz.f32 %r2352, %r2288; + sqrt.approx.ftz.f32 %r2353, %r2289; + sqrt.approx.ftz.f32 %r2354, %r2290; + sqrt.approx.ftz.f32 %r2355, %r2291; + sqrt.approx.ftz.f32 %r2356, %r2292; + sqrt.approx.ftz.f32 %r2357, %r2293; + sqrt.approx.ftz.f32 %r2358, %r2294; + .loc 1 71 34 // l2norm.py:71:34 + shl.b32 %r2359, %r592, 2; + and.b32 %r2360, %r2359, 124; + or.b32 %r2361, %r2360, 3; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd458, %r2361; + or.b64 %rd459, %rd136, %rd458; + cvt.u64.u32 %rd460, %r2360; + or.b64 %rd461, %rd136, %rd460; + .loc 1 77 20 // l2norm.py:77:20 + shl.b64 %rd462, %rd461, 2; + add.s64 %rd129, %rd135, %rd462; + add.s64 %rd132, %rd129, 12; + .loc 1 71 34 // l2norm.py:71:34 + or.b32 %r2362, %r2360, 2; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd463, %r2362; + or.b64 %rd464, %rd136, %rd463; + .loc 1 77 20 // l2norm.py:77:20 + add.s64 %rd131, %rd129, 8; + .loc 1 71 34 // l2norm.py:71:34 + or.b32 %r2363, %r2360, 1; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd465, %r2363; + or.b64 %rd466, %rd136, %rd465; + .loc 1 77 20 // l2norm.py:77:20 + add.s64 %rd130, %rd129, 4; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd467, %rd134, %rd393; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd128, %rd467, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd468, %rd134, %rd391; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd127, %rd468, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd469, %rd134, %rd389; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd126, %rd469, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd470, %rd134, %rd387; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd125, %rd470, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd471, %rd134, %rd385; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd124, %rd471, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd472, %rd134, %rd383; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd123, %rd472, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd473, %rd134, %rd381; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd122, %rd473, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd474, %rd134, %rd379; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd121, %rd474, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd475, %rd134, %rd377; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd120, %rd475, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd476, %rd134, %rd375; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd119, %rd476, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd477, %rd134, %rd373; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd118, %rd477, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd478, %rd134, %rd371; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd117, %rd478, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd479, %rd134, %rd369; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd116, %rd479, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd480, %rd134, %rd367; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd115, %rd480, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd481, %rd134, %rd365; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd114, %rd481, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd482, %rd134, %rd363; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd113, %rd482, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd483, %rd134, %rd361; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd112, %rd483, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd484, %rd134, %rd359; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd111, %rd484, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd485, %rd134, %rd357; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd110, %rd485, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd486, %rd134, %rd355; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd109, %rd486, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd487, %rd134, %rd353; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd108, %rd487, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd488, %rd134, %rd351; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd107, %rd488, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd489, %rd134, %rd349; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd106, %rd489, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd490, %rd134, %rd347; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd105, %rd490, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd491, %rd134, %rd345; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd104, %rd491, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd492, %rd134, %rd343; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd103, %rd492, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd493, %rd134, %rd341; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd102, %rd493, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd494, %rd134, %rd339; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd101, %rd494, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd495, %rd134, %rd337; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd100, %rd495, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd496, %rd134, %rd335; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd99, %rd496, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd497, %rd134, %rd333; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd98, %rd497, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd498, %rd134, %rd331; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd97, %rd498, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd499, %rd134, %rd329; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd96, %rd499, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd500, %rd134, %rd327; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd95, %rd500, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd501, %rd134, %rd325; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd94, %rd501, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd502, %rd134, %rd323; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd93, %rd502, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd503, %rd134, %rd321; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd92, %rd503, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd504, %rd134, %rd319; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd91, %rd504, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd505, %rd134, %rd317; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd90, %rd505, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd506, %rd134, %rd315; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd89, %rd506, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd507, %rd134, %rd313; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd88, %rd507, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd508, %rd134, %rd311; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd87, %rd508, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd509, %rd134, %rd309; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd86, %rd509, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd510, %rd134, %rd307; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd85, %rd510, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd511, %rd134, %rd305; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd84, %rd511, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd512, %rd134, %rd303; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd83, %rd512, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd513, %rd134, %rd301; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd82, %rd513, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd514, %rd134, %rd299; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd81, %rd514, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd515, %rd134, %rd297; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd80, %rd515, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd516, %rd134, %rd295; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd79, %rd516, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd517, %rd134, %rd293; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd78, %rd517, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd518, %rd134, %rd291; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd77, %rd518, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd519, %rd134, %rd289; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd76, %rd519, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd520, %rd134, %rd287; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd75, %rd520, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd521, %rd134, %rd285; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd74, %rd521, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd522, %rd134, %rd283; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd73, %rd522, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd523, %rd134, %rd281; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd72, %rd523, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd524, %rd134, %rd279; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd71, %rd524, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd525, %rd134, %rd277; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd70, %rd525, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd526, %rd134, %rd275; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd69, %rd526, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd527, %rd134, %rd273; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd68, %rd527, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd528, %rd134, %rd271; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd67, %rd528, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd529, %rd134, %rd269; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd66, %rd529, %rd395; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd530, %rd134, %rd267; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd65, %rd530, %rd395; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p68, %rd459, %rd137; + setp.lt.s64 %p67, %rd464, %rd137; + setp.lt.s64 %p66, %rd466, %rd137; + setp.lt.s64 %p65, %rd461, %rd137; + mov.b32 %r2364, 0f3F800000; + .loc 1 80 17 // l2norm.py:80:17 + div.full.f32 %r515, %r2364, %r2295; + div.full.f32 %r554, %r2364, %r2296; + div.full.f32 %r516, %r2364, %r2297; + div.full.f32 %r555, %r2364, %r2298; + div.full.f32 %r517, %r2364, %r2299; + div.full.f32 %r556, %r2364, %r2300; + div.full.f32 %r518, %r2364, %r2301; + div.full.f32 %r557, %r2364, %r2302; + div.full.f32 %r520, %r2364, %r2303; + div.full.f32 %r558, %r2364, %r2304; + div.full.f32 %r521, %r2364, %r2305; + div.full.f32 %r559, %r2364, %r2306; + div.full.f32 %r522, %r2364, %r2307; + div.full.f32 %r560, %r2364, %r2308; + div.full.f32 %r523, %r2364, %r2309; + div.full.f32 %r561, %r2364, %r2310; + div.full.f32 %r525, %r2364, %r2311; + div.full.f32 %r562, %r2364, %r2312; + div.full.f32 %r526, %r2364, %r2313; + div.full.f32 %r563, %r2364, %r2314; + div.full.f32 %r527, %r2364, %r2315; + div.full.f32 %r564, %r2364, %r2316; + div.full.f32 %r528, %r2364, %r2317; + div.full.f32 %r565, %r2364, %r2318; + div.full.f32 %r530, %r2364, %r2319; + div.full.f32 %r566, %r2364, %r2320; + div.full.f32 %r531, %r2364, %r2321; + div.full.f32 %r567, %r2364, %r2322; + div.full.f32 %r532, %r2364, %r2323; + div.full.f32 %r568, %r2364, %r2324; + div.full.f32 %r533, %r2364, %r2325; + div.full.f32 %r569, %r2364, %r2326; + div.full.f32 %r535, %r2364, %r2327; + div.full.f32 %r570, %r2364, %r2328; + div.full.f32 %r536, %r2364, %r2329; + div.full.f32 %r571, %r2364, %r2330; + div.full.f32 %r537, %r2364, %r2331; + div.full.f32 %r572, %r2364, %r2332; + div.full.f32 %r538, %r2364, %r2333; + div.full.f32 %r573, %r2364, %r2334; + div.full.f32 %r540, %r2364, %r2335; + div.full.f32 %r574, %r2364, %r2336; + div.full.f32 %r541, %r2364, %r2337; + div.full.f32 %r575, %r2364, %r2338; + div.full.f32 %r542, %r2364, %r2339; + div.full.f32 %r576, %r2364, %r2340; + div.full.f32 %r543, %r2364, %r2341; + div.full.f32 %r577, %r2364, %r2342; + div.full.f32 %r545, %r2364, %r2343; + div.full.f32 %r578, %r2364, %r2344; + div.full.f32 %r546, %r2364, %r2345; + div.full.f32 %r579, %r2364, %r2346; + div.full.f32 %r547, %r2364, %r2347; + div.full.f32 %r580, %r2364, %r2348; + div.full.f32 %r548, %r2364, %r2349; + div.full.f32 %r581, %r2364, %r2350; + div.full.f32 %r550, %r2364, %r2351; + div.full.f32 %r582, %r2364, %r2352; + div.full.f32 %r551, %r2364, %r2353; + div.full.f32 %r583, %r2364, %r2354; + div.full.f32 %r552, %r2364, %r2355; + div.full.f32 %r584, %r2364, %r2356; + div.full.f32 %r553, %r2364, %r2357; + div.full.f32 %r585, %r2364, %r2358; + .loc 1 81 16 // l2norm.py:81:16 + mul.f32 %r2365, %r515, %r695; + mul.f32 %r2366, %r515, %r696; + mul.f32 %r2367, %r515, %r697; + mul.f32 %r2368, %r515, %r698; + mul.f32 %r2369, %r515, %r699; + mul.f32 %r2370, %r515, %r700; + mul.f32 %r2371, %r515, %r701; + mul.f32 %r2372, %r515, %r702; + mul.f32 %r2373, %r554, %r703; + mul.f32 %r2374, %r554, %r704; + mul.f32 %r2375, %r554, %r705; + mul.f32 %r2376, %r554, %r706; + mul.f32 %r2377, %r554, %r707; + mul.f32 %r2378, %r554, %r708; + mul.f32 %r2379, %r554, %r709; + mul.f32 %r2380, %r554, %r710; + mul.f32 %r2381, %r516, %r711; + mul.f32 %r2382, %r516, %r712; + mul.f32 %r2383, %r516, %r713; + mul.f32 %r2384, %r516, %r714; + mul.f32 %r2385, %r516, %r715; + mul.f32 %r2386, %r516, %r716; + mul.f32 %r2387, %r516, %r717; + mul.f32 %r2388, %r516, %r718; + mul.f32 %r2389, %r555, %r719; + mul.f32 %r2390, %r555, %r720; + mul.f32 %r2391, %r555, %r721; + mul.f32 %r2392, %r555, %r722; + mul.f32 %r2393, %r555, %r723; + mul.f32 %r2394, %r555, %r724; + mul.f32 %r2395, %r555, %r725; + mul.f32 %r2396, %r555, %r726; + mul.f32 %r2397, %r517, %r727; + mul.f32 %r2398, %r517, %r728; + mul.f32 %r2399, %r517, %r729; + mul.f32 %r2400, %r517, %r730; + mul.f32 %r2401, %r517, %r731; + mul.f32 %r2402, %r517, %r732; + mul.f32 %r2403, %r517, %r733; + mul.f32 %r2404, %r517, %r734; + mul.f32 %r2405, %r556, %r735; + mul.f32 %r2406, %r556, %r736; + mul.f32 %r2407, %r556, %r737; + mul.f32 %r2408, %r556, %r738; + mul.f32 %r2409, %r556, %r739; + mul.f32 %r2410, %r556, %r740; + mul.f32 %r2411, %r556, %r741; + mul.f32 %r2412, %r556, %r742; + mul.f32 %r2413, %r518, %r743; + mul.f32 %r2414, %r518, %r744; + mul.f32 %r2415, %r518, %r745; + mul.f32 %r2416, %r518, %r746; + mul.f32 %r2417, %r518, %r747; + mul.f32 %r2418, %r518, %r748; + mul.f32 %r2419, %r518, %r749; + mul.f32 %r2420, %r518, %r750; + mul.f32 %r2421, %r557, %r751; + mul.f32 %r2422, %r557, %r752; + mul.f32 %r2423, %r557, %r753; + mul.f32 %r2424, %r557, %r754; + mul.f32 %r2425, %r557, %r755; + mul.f32 %r2426, %r557, %r756; + mul.f32 %r2427, %r557, %r757; + mul.f32 %r2428, %r557, %r758; + mul.f32 %r2429, %r520, %r759; + mul.f32 %r2430, %r520, %r760; + mul.f32 %r2431, %r520, %r761; + mul.f32 %r2432, %r520, %r762; + mul.f32 %r2433, %r520, %r763; + mul.f32 %r2434, %r520, %r764; + mul.f32 %r2435, %r520, %r765; + mul.f32 %r2436, %r520, %r766; + mul.f32 %r2437, %r558, %r767; + mul.f32 %r2438, %r558, %r768; + mul.f32 %r2439, %r558, %r769; + mul.f32 %r2440, %r558, %r770; + mul.f32 %r2441, %r558, %r771; + mul.f32 %r2442, %r558, %r772; + mul.f32 %r2443, %r558, %r773; + mul.f32 %r2444, %r558, %r774; + mul.f32 %r2445, %r521, %r775; + mul.f32 %r2446, %r521, %r776; + mul.f32 %r2447, %r521, %r777; + mul.f32 %r2448, %r521, %r778; + mul.f32 %r2449, %r521, %r779; + mul.f32 %r2450, %r521, %r780; + mul.f32 %r2451, %r521, %r781; + mul.f32 %r2452, %r521, %r782; + mul.f32 %r2453, %r559, %r783; + mul.f32 %r2454, %r559, %r784; + mul.f32 %r2455, %r559, %r785; + mul.f32 %r2456, %r559, %r786; + mul.f32 %r2457, %r559, %r787; + mul.f32 %r2458, %r559, %r788; + mul.f32 %r2459, %r559, %r789; + mul.f32 %r2460, %r559, %r790; + mul.f32 %r2461, %r522, %r791; + mul.f32 %r2462, %r522, %r792; + mul.f32 %r2463, %r522, %r793; + mul.f32 %r2464, %r522, %r794; + mul.f32 %r2465, %r522, %r795; + mul.f32 %r2466, %r522, %r796; + mul.f32 %r2467, %r522, %r797; + mul.f32 %r2468, %r522, %r798; + mul.f32 %r2469, %r560, %r799; + mul.f32 %r2470, %r560, %r800; + mul.f32 %r2471, %r560, %r801; + mul.f32 %r2472, %r560, %r802; + mul.f32 %r2473, %r560, %r803; + mul.f32 %r2474, %r560, %r804; + mul.f32 %r2475, %r560, %r805; + mul.f32 %r2476, %r560, %r806; + mul.f32 %r2477, %r523, %r807; + mul.f32 %r2478, %r523, %r808; + mul.f32 %r2479, %r523, %r809; + mul.f32 %r2480, %r523, %r810; + mul.f32 %r2481, %r523, %r811; + mul.f32 %r2482, %r523, %r812; + mul.f32 %r2483, %r523, %r813; + mul.f32 %r2484, %r523, %r814; + mul.f32 %r2485, %r561, %r815; + mul.f32 %r2486, %r561, %r816; + mul.f32 %r2487, %r561, %r817; + mul.f32 %r2488, %r561, %r818; + mul.f32 %r2489, %r561, %r819; + mul.f32 %r2490, %r561, %r820; + mul.f32 %r2491, %r561, %r821; + mul.f32 %r2492, %r561, %r822; + mul.f32 %r2493, %r525, %r823; + mul.f32 %r2494, %r525, %r824; + mul.f32 %r2495, %r525, %r825; + mul.f32 %r2496, %r525, %r826; + mul.f32 %r2497, %r525, %r827; + mul.f32 %r2498, %r525, %r828; + mul.f32 %r2499, %r525, %r829; + mul.f32 %r2500, %r525, %r830; + mul.f32 %r2501, %r562, %r831; + mul.f32 %r2502, %r562, %r832; + mul.f32 %r2503, %r562, %r833; + mul.f32 %r2504, %r562, %r834; + mul.f32 %r2505, %r562, %r835; + mul.f32 %r2506, %r562, %r836; + mul.f32 %r2507, %r562, %r837; + mul.f32 %r2508, %r562, %r838; + mul.f32 %r2509, %r526, %r839; + mul.f32 %r2510, %r526, %r840; + mul.f32 %r2511, %r526, %r841; + mul.f32 %r2512, %r526, %r842; + mul.f32 %r2513, %r526, %r843; + mul.f32 %r2514, %r526, %r844; + mul.f32 %r2515, %r526, %r845; + mul.f32 %r2516, %r526, %r846; + mul.f32 %r2517, %r563, %r847; + mul.f32 %r2518, %r563, %r848; + mul.f32 %r2519, %r563, %r849; + mul.f32 %r2520, %r563, %r850; + mul.f32 %r2521, %r563, %r851; + mul.f32 %r2522, %r563, %r852; + mul.f32 %r2523, %r563, %r853; + mul.f32 %r2524, %r563, %r854; + mul.f32 %r2525, %r527, %r855; + mul.f32 %r2526, %r527, %r856; + mul.f32 %r2527, %r527, %r857; + mul.f32 %r2528, %r527, %r858; + mul.f32 %r2529, %r527, %r859; + mul.f32 %r2530, %r527, %r860; + mul.f32 %r2531, %r527, %r861; + mul.f32 %r2532, %r527, %r862; + mul.f32 %r2533, %r564, %r863; + mul.f32 %r2534, %r564, %r864; + mul.f32 %r2535, %r564, %r865; + mul.f32 %r2536, %r564, %r866; + mul.f32 %r2537, %r564, %r867; + mul.f32 %r2538, %r564, %r868; + mul.f32 %r2539, %r564, %r869; + mul.f32 %r2540, %r564, %r870; + mul.f32 %r2541, %r528, %r871; + mul.f32 %r2542, %r528, %r872; + mul.f32 %r2543, %r528, %r873; + mul.f32 %r2544, %r528, %r874; + mul.f32 %r2545, %r528, %r875; + mul.f32 %r2546, %r528, %r876; + mul.f32 %r2547, %r528, %r877; + mul.f32 %r2548, %r528, %r878; + mul.f32 %r2549, %r565, %r879; + mul.f32 %r2550, %r565, %r880; + mul.f32 %r2551, %r565, %r881; + mul.f32 %r2552, %r565, %r882; + mul.f32 %r2553, %r565, %r883; + mul.f32 %r2554, %r565, %r884; + mul.f32 %r2555, %r565, %r885; + mul.f32 %r2556, %r565, %r886; + mul.f32 %r2557, %r530, %r887; + mul.f32 %r2558, %r530, %r888; + mul.f32 %r2559, %r530, %r889; + mul.f32 %r2560, %r530, %r890; + mul.f32 %r2561, %r530, %r891; + mul.f32 %r2562, %r530, %r892; + mul.f32 %r2563, %r530, %r893; + mul.f32 %r2564, %r530, %r894; + mul.f32 %r2565, %r566, %r895; + mul.f32 %r2566, %r566, %r896; + mul.f32 %r2567, %r566, %r897; + mul.f32 %r2568, %r566, %r898; + mul.f32 %r2569, %r566, %r899; + mul.f32 %r2570, %r566, %r900; + mul.f32 %r2571, %r566, %r901; + mul.f32 %r2572, %r566, %r902; + mul.f32 %r2573, %r531, %r903; + mul.f32 %r2574, %r531, %r904; + mul.f32 %r2575, %r531, %r905; + mul.f32 %r2576, %r531, %r906; + mul.f32 %r2577, %r531, %r907; + mul.f32 %r2578, %r531, %r908; + mul.f32 %r2579, %r531, %r909; + mul.f32 %r2580, %r531, %r910; + mul.f32 %r2581, %r567, %r911; + mul.f32 %r2582, %r567, %r912; + mul.f32 %r2583, %r567, %r913; + mul.f32 %r2584, %r567, %r914; + mul.f32 %r2585, %r567, %r915; + mul.f32 %r2586, %r567, %r916; + mul.f32 %r2587, %r567, %r917; + mul.f32 %r2588, %r567, %r918; + mul.f32 %r2589, %r532, %r919; + mul.f32 %r2590, %r532, %r920; + mul.f32 %r2591, %r532, %r921; + mul.f32 %r2592, %r532, %r922; + mul.f32 %r2593, %r532, %r923; + mul.f32 %r2594, %r532, %r924; + mul.f32 %r2595, %r532, %r925; + mul.f32 %r2596, %r532, %r926; + mul.f32 %r2597, %r568, %r927; + mul.f32 %r2598, %r568, %r928; + mul.f32 %r2599, %r568, %r929; + mul.f32 %r2600, %r568, %r930; + mul.f32 %r2601, %r568, %r931; + mul.f32 %r2602, %r568, %r932; + mul.f32 %r2603, %r568, %r933; + mul.f32 %r2604, %r568, %r934; + mul.f32 %r2605, %r533, %r935; + mul.f32 %r2606, %r533, %r936; + mul.f32 %r2607, %r533, %r937; + mul.f32 %r2608, %r533, %r938; + mul.f32 %r2609, %r533, %r939; + mul.f32 %r2610, %r533, %r940; + mul.f32 %r2611, %r533, %r941; + mul.f32 %r2612, %r533, %r942; + mul.f32 %r2613, %r569, %r943; + mul.f32 %r2614, %r569, %r944; + mul.f32 %r2615, %r569, %r945; + mul.f32 %r2616, %r569, %r946; + mul.f32 %r2617, %r569, %r947; + mul.f32 %r2618, %r569, %r948; + mul.f32 %r2619, %r569, %r949; + mul.f32 %r2620, %r569, %r950; + mul.f32 %r2621, %r535, %r951; + mul.f32 %r2622, %r535, %r952; + mul.f32 %r2623, %r535, %r953; + mul.f32 %r2624, %r535, %r954; + mul.f32 %r2625, %r535, %r955; + mul.f32 %r2626, %r535, %r956; + mul.f32 %r2627, %r535, %r957; + mul.f32 %r2628, %r535, %r958; + mul.f32 %r2629, %r570, %r959; + mul.f32 %r2630, %r570, %r960; + mul.f32 %r2631, %r570, %r961; + mul.f32 %r2632, %r570, %r962; + mul.f32 %r2633, %r570, %r963; + mul.f32 %r2634, %r570, %r964; + mul.f32 %r2635, %r570, %r965; + mul.f32 %r2636, %r570, %r966; + mul.f32 %r2637, %r536, %r967; + mul.f32 %r2638, %r536, %r968; + mul.f32 %r2639, %r536, %r969; + mul.f32 %r2640, %r536, %r970; + mul.f32 %r2641, %r536, %r971; + mul.f32 %r2642, %r536, %r972; + mul.f32 %r2643, %r536, %r973; + mul.f32 %r2644, %r536, %r974; + mul.f32 %r2645, %r571, %r975; + mul.f32 %r2646, %r571, %r976; + mul.f32 %r2647, %r571, %r977; + mul.f32 %r2648, %r571, %r978; + mul.f32 %r2649, %r571, %r979; + mul.f32 %r2650, %r571, %r980; + mul.f32 %r2651, %r571, %r981; + mul.f32 %r2652, %r571, %r982; + mul.f32 %r2653, %r537, %r983; + mul.f32 %r2654, %r537, %r984; + mul.f32 %r2655, %r537, %r985; + mul.f32 %r2656, %r537, %r986; + mul.f32 %r2657, %r537, %r987; + mul.f32 %r2658, %r537, %r988; + mul.f32 %r2659, %r537, %r989; + mul.f32 %r2660, %r537, %r990; + mul.f32 %r2661, %r572, %r991; + mul.f32 %r2662, %r572, %r992; + mul.f32 %r2663, %r572, %r993; + mul.f32 %r2664, %r572, %r994; + mul.f32 %r2665, %r572, %r995; + mul.f32 %r2666, %r572, %r996; + mul.f32 %r2667, %r572, %r997; + mul.f32 %r2668, %r572, %r998; + mul.f32 %r2669, %r538, %r999; + mul.f32 %r2670, %r538, %r1000; + mul.f32 %r2671, %r538, %r1001; + mul.f32 %r2672, %r538, %r1002; + mul.f32 %r2673, %r538, %r1003; + mul.f32 %r2674, %r538, %r1004; + mul.f32 %r2675, %r538, %r1005; + mul.f32 %r2676, %r538, %r1006; + mul.f32 %r2677, %r573, %r1007; + mul.f32 %r2678, %r573, %r1008; + mul.f32 %r2679, %r573, %r1009; + mul.f32 %r2680, %r573, %r1010; + mul.f32 %r2681, %r573, %r1011; + mul.f32 %r2682, %r573, %r1012; + mul.f32 %r2683, %r573, %r1013; + mul.f32 %r2684, %r573, %r1014; + mul.f32 %r2685, %r540, %r1015; + mul.f32 %r2686, %r540, %r1016; + mul.f32 %r2687, %r540, %r1017; + mul.f32 %r2688, %r540, %r1018; + mul.f32 %r2689, %r540, %r1019; + mul.f32 %r2690, %r540, %r1020; + mul.f32 %r2691, %r540, %r1021; + mul.f32 %r2692, %r540, %r1022; + mul.f32 %r2693, %r574, %r1023; + mul.f32 %r2694, %r574, %r1024; + mul.f32 %r2695, %r574, %r1025; + mul.f32 %r2696, %r574, %r1026; + mul.f32 %r2697, %r574, %r1027; + mul.f32 %r2698, %r574, %r1028; + mul.f32 %r2699, %r574, %r1029; + mul.f32 %r2700, %r574, %r1030; + mul.f32 %r2701, %r541, %r1031; + mul.f32 %r2702, %r541, %r1032; + mul.f32 %r2703, %r541, %r1033; + mul.f32 %r2704, %r541, %r1034; + mul.f32 %r2705, %r541, %r1035; + mul.f32 %r2706, %r541, %r1036; + mul.f32 %r2707, %r541, %r1037; + mul.f32 %r2708, %r541, %r1038; + mul.f32 %r2709, %r575, %r1039; + mul.f32 %r2710, %r575, %r1040; + mul.f32 %r2711, %r575, %r1041; + mul.f32 %r2712, %r575, %r1042; + mul.f32 %r2713, %r575, %r1043; + mul.f32 %r2714, %r575, %r1044; + mul.f32 %r2715, %r575, %r1045; + mul.f32 %r2716, %r575, %r1046; + mul.f32 %r2717, %r542, %r1047; + mul.f32 %r2718, %r542, %r1048; + mul.f32 %r2719, %r542, %r1049; + mul.f32 %r2720, %r542, %r1050; + mul.f32 %r2721, %r542, %r1051; + mul.f32 %r2722, %r542, %r1052; + mul.f32 %r2723, %r542, %r1053; + mul.f32 %r2724, %r542, %r1054; + mul.f32 %r2725, %r576, %r1055; + mul.f32 %r2726, %r576, %r1056; + mul.f32 %r2727, %r576, %r1057; + mul.f32 %r2728, %r576, %r1058; + mul.f32 %r2729, %r576, %r1059; + mul.f32 %r2730, %r576, %r1060; + mul.f32 %r2731, %r576, %r1061; + mul.f32 %r2732, %r576, %r1062; + mul.f32 %r2733, %r543, %r1063; + mul.f32 %r2734, %r543, %r1064; + mul.f32 %r2735, %r543, %r1065; + mul.f32 %r2736, %r543, %r1066; + mul.f32 %r2737, %r543, %r1067; + mul.f32 %r2738, %r543, %r1068; + mul.f32 %r2739, %r543, %r1069; + mul.f32 %r2740, %r543, %r1070; + mul.f32 %r2741, %r577, %r1071; + mul.f32 %r2742, %r577, %r1072; + mul.f32 %r2743, %r577, %r1073; + mul.f32 %r2744, %r577, %r1074; + mul.f32 %r2745, %r577, %r1075; + mul.f32 %r2746, %r577, %r1076; + mul.f32 %r2747, %r577, %r1077; + mul.f32 %r2748, %r577, %r1078; + mul.f32 %r2749, %r545, %r1079; + mul.f32 %r2750, %r545, %r1080; + mul.f32 %r2751, %r545, %r1081; + mul.f32 %r2752, %r545, %r1082; + mul.f32 %r2753, %r545, %r1083; + mul.f32 %r2754, %r545, %r1084; + mul.f32 %r2755, %r545, %r1085; + mul.f32 %r2756, %r545, %r1086; + mul.f32 %r2757, %r578, %r1087; + mul.f32 %r2758, %r578, %r1088; + mul.f32 %r2759, %r578, %r1089; + mul.f32 %r2760, %r578, %r1090; + mul.f32 %r2761, %r578, %r1091; + mul.f32 %r2762, %r578, %r1092; + mul.f32 %r2763, %r578, %r1093; + mul.f32 %r2764, %r578, %r1094; + mul.f32 %r2765, %r546, %r1095; + mul.f32 %r2766, %r546, %r1096; + mul.f32 %r2767, %r546, %r1097; + mul.f32 %r2768, %r546, %r1098; + mul.f32 %r2769, %r546, %r1099; + mul.f32 %r2770, %r546, %r1100; + mul.f32 %r2771, %r546, %r1101; + mul.f32 %r2772, %r546, %r1102; + mul.f32 %r2773, %r579, %r1103; + mul.f32 %r2774, %r579, %r1104; + mul.f32 %r2775, %r579, %r1105; + mul.f32 %r2776, %r579, %r1106; + mul.f32 %r2777, %r579, %r1107; + mul.f32 %r2778, %r579, %r1108; + mul.f32 %r2779, %r579, %r1109; + mul.f32 %r2780, %r579, %r1110; + mul.f32 %r2781, %r547, %r1111; + mul.f32 %r2782, %r547, %r1112; + mul.f32 %r2783, %r547, %r1113; + mul.f32 %r2784, %r547, %r1114; + mul.f32 %r2785, %r547, %r1115; + mul.f32 %r2786, %r547, %r1116; + mul.f32 %r2787, %r547, %r1117; + mul.f32 %r2788, %r547, %r1118; + mul.f32 %r2789, %r580, %r1119; + mul.f32 %r2790, %r580, %r1120; + mul.f32 %r2791, %r580, %r1121; + mul.f32 %r2792, %r580, %r1122; + mul.f32 %r2793, %r580, %r1123; + mul.f32 %r2794, %r580, %r1124; + mul.f32 %r2795, %r580, %r1125; + mul.f32 %r2796, %r580, %r1126; + mul.f32 %r2797, %r548, %r1127; + mul.f32 %r2798, %r548, %r1128; + mul.f32 %r2799, %r548, %r1129; + mul.f32 %r2800, %r548, %r1130; + mul.f32 %r2801, %r548, %r1131; + mul.f32 %r2802, %r548, %r1132; + mul.f32 %r2803, %r548, %r1133; + mul.f32 %r2804, %r548, %r1134; + mul.f32 %r2805, %r581, %r1135; + mul.f32 %r2806, %r581, %r1136; + mul.f32 %r2807, %r581, %r1137; + mul.f32 %r2808, %r581, %r1138; + mul.f32 %r2809, %r581, %r1139; + mul.f32 %r2810, %r581, %r1140; + mul.f32 %r2811, %r581, %r1141; + mul.f32 %r2812, %r581, %r1142; + mul.f32 %r2813, %r550, %r1143; + mul.f32 %r2814, %r550, %r1144; + mul.f32 %r2815, %r550, %r1145; + mul.f32 %r2816, %r550, %r1146; + mul.f32 %r2817, %r550, %r1147; + mul.f32 %r2818, %r550, %r1148; + mul.f32 %r2819, %r550, %r1149; + mul.f32 %r2820, %r550, %r1150; + mul.f32 %r2821, %r582, %r1151; + mul.f32 %r2822, %r582, %r1152; + mul.f32 %r2823, %r582, %r1153; + mul.f32 %r2824, %r582, %r1154; + mul.f32 %r2825, %r582, %r1155; + mul.f32 %r2826, %r582, %r1156; + mul.f32 %r2827, %r582, %r1157; + mul.f32 %r2828, %r582, %r1158; + mul.f32 %r2829, %r551, %r1159; + mul.f32 %r2830, %r551, %r1160; + mul.f32 %r2831, %r551, %r1161; + mul.f32 %r2832, %r551, %r1162; + mul.f32 %r2833, %r551, %r1163; + mul.f32 %r2834, %r551, %r1164; + mul.f32 %r2835, %r551, %r1165; + mul.f32 %r2836, %r551, %r1166; + mul.f32 %r2837, %r583, %r1167; + mul.f32 %r2838, %r583, %r1168; + mul.f32 %r2839, %r583, %r1169; + mul.f32 %r2840, %r583, %r1170; + mul.f32 %r2841, %r583, %r1171; + mul.f32 %r2842, %r583, %r1172; + mul.f32 %r2843, %r583, %r1173; + mul.f32 %r2844, %r583, %r1174; + mul.f32 %r2845, %r552, %r1175; + mul.f32 %r2846, %r552, %r1176; + mul.f32 %r2847, %r552, %r1177; + mul.f32 %r2848, %r552, %r1178; + mul.f32 %r2849, %r552, %r1179; + mul.f32 %r2850, %r552, %r1180; + mul.f32 %r2851, %r552, %r1181; + mul.f32 %r2852, %r552, %r1182; + mul.f32 %r2853, %r584, %r1183; + mul.f32 %r2854, %r584, %r1184; + mul.f32 %r2855, %r584, %r1185; + mul.f32 %r2856, %r584, %r1186; + mul.f32 %r2857, %r584, %r1187; + mul.f32 %r2858, %r584, %r1188; + mul.f32 %r2859, %r584, %r1189; + mul.f32 %r2860, %r584, %r1190; + mul.f32 %r2861, %r553, %r1191; + mul.f32 %r2862, %r553, %r1192; + mul.f32 %r2863, %r553, %r1193; + mul.f32 %r2864, %r553, %r1194; + mul.f32 %r2865, %r553, %r1195; + mul.f32 %r2866, %r553, %r1196; + mul.f32 %r2867, %r553, %r1197; + mul.f32 %r2868, %r553, %r1198; + mul.f32 %r2869, %r585, %r1199; + mul.f32 %r2870, %r585, %r1200; + mul.f32 %r2871, %r585, %r1201; + mul.f32 %r2872, %r585, %r1202; + mul.f32 %r2873, %r585, %r1203; + mul.f32 %r2874, %r585, %r1204; + mul.f32 %r2875, %r585, %r1205; + mul.f32 %r2876, %r585, %r1206; + .loc 1 83 25 // l2norm.py:83:25 + cvt.rn.bf16x2.f32 %r450, %r2750, %r2749; + cvt.rn.bf16x2.f32 %r454, %r2758, %r2757; + cvt.rn.bf16x2.f32 %r455, %r2760, %r2759; + cvt.rn.bf16x2.f32 %r456, %r2762, %r2761; + cvt.rn.bf16x2.f32 %r457, %r2764, %r2763; + cvt.rn.bf16x2.f32 %r458, %r2766, %r2765; + cvt.rn.bf16x2.f32 %r459, %r2768, %r2767; + cvt.rn.bf16x2.f32 %r460, %r2770, %r2769; + cvt.rn.bf16x2.f32 %r461, %r2772, %r2771; + cvt.rn.bf16x2.f32 %r462, %r2774, %r2773; + cvt.rn.bf16x2.f32 %r463, %r2776, %r2775; + cvt.rn.bf16x2.f32 %r464, %r2778, %r2777; + cvt.rn.bf16x2.f32 %r465, %r2780, %r2779; + cvt.rn.bf16x2.f32 %r466, %r2782, %r2781; + cvt.rn.bf16x2.f32 %r467, %r2784, %r2783; + cvt.rn.bf16x2.f32 %r468, %r2786, %r2785; + cvt.rn.bf16x2.f32 %r469, %r2788, %r2787; + cvt.rn.bf16x2.f32 %r470, %r2790, %r2789; + cvt.rn.bf16x2.f32 %r471, %r2792, %r2791; + cvt.rn.bf16x2.f32 %r472, %r2794, %r2793; + cvt.rn.bf16x2.f32 %r473, %r2796, %r2795; + cvt.rn.bf16x2.f32 %r474, %r2798, %r2797; + cvt.rn.bf16x2.f32 %r475, %r2800, %r2799; + cvt.rn.bf16x2.f32 %r476, %r2802, %r2801; + cvt.rn.bf16x2.f32 %r477, %r2804, %r2803; + cvt.rn.bf16x2.f32 %r478, %r2806, %r2805; + cvt.rn.bf16x2.f32 %r479, %r2808, %r2807; + cvt.rn.bf16x2.f32 %r480, %r2810, %r2809; + cvt.rn.bf16x2.f32 %r481, %r2812, %r2811; + cvt.rn.bf16x2.f32 %r482, %r2814, %r2813; + cvt.rn.bf16x2.f32 %r483, %r2816, %r2815; + cvt.rn.bf16x2.f32 %r484, %r2818, %r2817; + cvt.rn.bf16x2.f32 %r485, %r2820, %r2819; + cvt.rn.bf16x2.f32 %r486, %r2822, %r2821; + cvt.rn.bf16x2.f32 %r487, %r2824, %r2823; + cvt.rn.bf16x2.f32 %r488, %r2826, %r2825; + cvt.rn.bf16x2.f32 %r489, %r2828, %r2827; + cvt.rn.bf16x2.f32 %r490, %r2830, %r2829; + cvt.rn.bf16x2.f32 %r491, %r2832, %r2831; + cvt.rn.bf16x2.f32 %r492, %r2834, %r2833; + cvt.rn.bf16x2.f32 %r493, %r2836, %r2835; + cvt.rn.bf16x2.f32 %r494, %r2838, %r2837; + cvt.rn.bf16x2.f32 %r495, %r2840, %r2839; + cvt.rn.bf16x2.f32 %r496, %r2842, %r2841; + cvt.rn.bf16x2.f32 %r497, %r2844, %r2843; + cvt.rn.bf16x2.f32 %r498, %r2846, %r2845; + cvt.rn.bf16x2.f32 %r499, %r2848, %r2847; + cvt.rn.bf16x2.f32 %r500, %r2850, %r2849; + cvt.rn.bf16x2.f32 %r501, %r2852, %r2851; + cvt.rn.bf16x2.f32 %r502, %r2854, %r2853; + cvt.rn.bf16x2.f32 %r503, %r2856, %r2855; + cvt.rn.bf16x2.f32 %r504, %r2858, %r2857; + cvt.rn.bf16x2.f32 %r505, %r2860, %r2859; + cvt.rn.bf16x2.f32 %r506, %r2862, %r2861; + cvt.rn.bf16x2.f32 %r507, %r2864, %r2863; + cvt.rn.bf16x2.f32 %r508, %r2866, %r2865; + cvt.rn.bf16x2.f32 %r509, %r2868, %r2867; + cvt.rn.bf16x2.f32 %r510, %r2870, %r2869; + cvt.rn.bf16x2.f32 %r511, %r2872, %r2871; + cvt.rn.bf16x2.f32 %r512, %r2874, %r2873; + cvt.rn.bf16x2.f32 %r513, %r2876, %r2875; + .loc 1 83 18 // l2norm.py:83:18 + cvt.rn.bf16x2.f32 %r258, %r2366, %r2365; + cvt.rn.bf16x2.f32 %r259, %r2368, %r2367; + cvt.rn.bf16x2.f32 %r260, %r2370, %r2369; + cvt.rn.bf16x2.f32 %r261, %r2372, %r2371; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd65 + 0 ], { %r258, %r259, %r260, %r261 }; + // end inline asm + cvt.rn.bf16x2.f32 %r262, %r2374, %r2373; + cvt.rn.bf16x2.f32 %r263, %r2376, %r2375; + cvt.rn.bf16x2.f32 %r264, %r2378, %r2377; + cvt.rn.bf16x2.f32 %r265, %r2380, %r2379; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd66 + 0 ], { %r262, %r263, %r264, %r265 }; + // end inline asm + cvt.rn.bf16x2.f32 %r266, %r2382, %r2381; + cvt.rn.bf16x2.f32 %r267, %r2384, %r2383; + cvt.rn.bf16x2.f32 %r268, %r2386, %r2385; + cvt.rn.bf16x2.f32 %r269, %r2388, %r2387; + // begin inline asm + @%p3 st.global.v4.b32 [ %rd67 + 0 ], { %r266, %r267, %r268, %r269 }; + // end inline asm + cvt.rn.bf16x2.f32 %r270, %r2390, %r2389; + cvt.rn.bf16x2.f32 %r271, %r2392, %r2391; + cvt.rn.bf16x2.f32 %r272, %r2394, %r2393; + cvt.rn.bf16x2.f32 %r273, %r2396, %r2395; + // begin inline asm + @%p4 st.global.v4.b32 [ %rd68 + 0 ], { %r270, %r271, %r272, %r273 }; + // end inline asm + cvt.rn.bf16x2.f32 %r274, %r2398, %r2397; + cvt.rn.bf16x2.f32 %r275, %r2400, %r2399; + cvt.rn.bf16x2.f32 %r276, %r2402, %r2401; + cvt.rn.bf16x2.f32 %r277, %r2404, %r2403; + // begin inline asm + @%p5 st.global.v4.b32 [ %rd69 + 0 ], { %r274, %r275, %r276, %r277 }; + // end inline asm + cvt.rn.bf16x2.f32 %r278, %r2406, %r2405; + cvt.rn.bf16x2.f32 %r279, %r2408, %r2407; + cvt.rn.bf16x2.f32 %r280, %r2410, %r2409; + cvt.rn.bf16x2.f32 %r281, %r2412, %r2411; + // begin inline asm + @%p6 st.global.v4.b32 [ %rd70 + 0 ], { %r278, %r279, %r280, %r281 }; + // end inline asm + cvt.rn.bf16x2.f32 %r282, %r2414, %r2413; + cvt.rn.bf16x2.f32 %r283, %r2416, %r2415; + cvt.rn.bf16x2.f32 %r284, %r2418, %r2417; + cvt.rn.bf16x2.f32 %r285, %r2420, %r2419; + // begin inline asm + @%p7 st.global.v4.b32 [ %rd71 + 0 ], { %r282, %r283, %r284, %r285 }; + // end inline asm + cvt.rn.bf16x2.f32 %r286, %r2422, %r2421; + cvt.rn.bf16x2.f32 %r287, %r2424, %r2423; + cvt.rn.bf16x2.f32 %r288, %r2426, %r2425; + cvt.rn.bf16x2.f32 %r289, %r2428, %r2427; + // begin inline asm + @%p31 st.global.v4.b32 [ %rd72 + 0 ], { %r286, %r287, %r288, %r289 }; + // end inline asm + cvt.rn.bf16x2.f32 %r290, %r2430, %r2429; + cvt.rn.bf16x2.f32 %r291, %r2432, %r2431; + cvt.rn.bf16x2.f32 %r292, %r2434, %r2433; + cvt.rn.bf16x2.f32 %r293, %r2436, %r2435; + // begin inline asm + @%p8 st.global.v4.b32 [ %rd73 + 0 ], { %r290, %r291, %r292, %r293 }; + // end inline asm + cvt.rn.bf16x2.f32 %r294, %r2438, %r2437; + cvt.rn.bf16x2.f32 %r295, %r2440, %r2439; + cvt.rn.bf16x2.f32 %r296, %r2442, %r2441; + cvt.rn.bf16x2.f32 %r297, %r2444, %r2443; + // begin inline asm + @%p9 st.global.v4.b32 [ %rd74 + 0 ], { %r294, %r295, %r296, %r297 }; + // end inline asm + cvt.rn.bf16x2.f32 %r298, %r2446, %r2445; + cvt.rn.bf16x2.f32 %r299, %r2448, %r2447; + cvt.rn.bf16x2.f32 %r300, %r2450, %r2449; + cvt.rn.bf16x2.f32 %r301, %r2452, %r2451; + // begin inline asm + @%p10 st.global.v4.b32 [ %rd75 + 0 ], { %r298, %r299, %r300, %r301 }; + // end inline asm + cvt.rn.bf16x2.f32 %r302, %r2454, %r2453; + cvt.rn.bf16x2.f32 %r303, %r2456, %r2455; + cvt.rn.bf16x2.f32 %r304, %r2458, %r2457; + cvt.rn.bf16x2.f32 %r305, %r2460, %r2459; + // begin inline asm + @%p11 st.global.v4.b32 [ %rd76 + 0 ], { %r302, %r303, %r304, %r305 }; + // end inline asm + cvt.rn.bf16x2.f32 %r306, %r2462, %r2461; + cvt.rn.bf16x2.f32 %r307, %r2464, %r2463; + cvt.rn.bf16x2.f32 %r308, %r2466, %r2465; + cvt.rn.bf16x2.f32 %r309, %r2468, %r2467; + // begin inline asm + @%p12 st.global.v4.b32 [ %rd77 + 0 ], { %r306, %r307, %r308, %r309 }; + // end inline asm + cvt.rn.bf16x2.f32 %r310, %r2470, %r2469; + cvt.rn.bf16x2.f32 %r311, %r2472, %r2471; + cvt.rn.bf16x2.f32 %r312, %r2474, %r2473; + cvt.rn.bf16x2.f32 %r313, %r2476, %r2475; + // begin inline asm + @%p13 st.global.v4.b32 [ %rd78 + 0 ], { %r310, %r311, %r312, %r313 }; + // end inline asm + cvt.rn.bf16x2.f32 %r314, %r2478, %r2477; + cvt.rn.bf16x2.f32 %r315, %r2480, %r2479; + cvt.rn.bf16x2.f32 %r316, %r2482, %r2481; + cvt.rn.bf16x2.f32 %r317, %r2484, %r2483; + // begin inline asm + @%p14 st.global.v4.b32 [ %rd79 + 0 ], { %r314, %r315, %r316, %r317 }; + // end inline asm + cvt.rn.bf16x2.f32 %r318, %r2486, %r2485; + cvt.rn.bf16x2.f32 %r319, %r2488, %r2487; + cvt.rn.bf16x2.f32 %r320, %r2490, %r2489; + cvt.rn.bf16x2.f32 %r321, %r2492, %r2491; + // begin inline asm + @%p32 st.global.v4.b32 [ %rd80 + 0 ], { %r318, %r319, %r320, %r321 }; + // end inline asm + cvt.rn.bf16x2.f32 %r322, %r2494, %r2493; + cvt.rn.bf16x2.f32 %r323, %r2496, %r2495; + cvt.rn.bf16x2.f32 %r324, %r2498, %r2497; + cvt.rn.bf16x2.f32 %r325, %r2500, %r2499; + // begin inline asm + @%p15 st.global.v4.b32 [ %rd81 + 0 ], { %r322, %r323, %r324, %r325 }; + // end inline asm + cvt.rn.bf16x2.f32 %r326, %r2502, %r2501; + cvt.rn.bf16x2.f32 %r327, %r2504, %r2503; + cvt.rn.bf16x2.f32 %r328, %r2506, %r2505; + cvt.rn.bf16x2.f32 %r329, %r2508, %r2507; + // begin inline asm + @%p16 st.global.v4.b32 [ %rd82 + 0 ], { %r326, %r327, %r328, %r329 }; + // end inline asm + cvt.rn.bf16x2.f32 %r330, %r2510, %r2509; + cvt.rn.bf16x2.f32 %r331, %r2512, %r2511; + cvt.rn.bf16x2.f32 %r332, %r2514, %r2513; + cvt.rn.bf16x2.f32 %r333, %r2516, %r2515; + // begin inline asm + @%p17 st.global.v4.b32 [ %rd83 + 0 ], { %r330, %r331, %r332, %r333 }; + // end inline asm + cvt.rn.bf16x2.f32 %r334, %r2518, %r2517; + cvt.rn.bf16x2.f32 %r335, %r2520, %r2519; + cvt.rn.bf16x2.f32 %r336, %r2522, %r2521; + cvt.rn.bf16x2.f32 %r337, %r2524, %r2523; + // begin inline asm + @%p18 st.global.v4.b32 [ %rd84 + 0 ], { %r334, %r335, %r336, %r337 }; + // end inline asm + cvt.rn.bf16x2.f32 %r338, %r2526, %r2525; + cvt.rn.bf16x2.f32 %r339, %r2528, %r2527; + cvt.rn.bf16x2.f32 %r340, %r2530, %r2529; + cvt.rn.bf16x2.f32 %r341, %r2532, %r2531; + // begin inline asm + @%p19 st.global.v4.b32 [ %rd85 + 0 ], { %r338, %r339, %r340, %r341 }; + // end inline asm + cvt.rn.bf16x2.f32 %r342, %r2534, %r2533; + cvt.rn.bf16x2.f32 %r343, %r2536, %r2535; + cvt.rn.bf16x2.f32 %r344, %r2538, %r2537; + cvt.rn.bf16x2.f32 %r345, %r2540, %r2539; + // begin inline asm + @%p20 st.global.v4.b32 [ %rd86 + 0 ], { %r342, %r343, %r344, %r345 }; + // end inline asm + cvt.rn.bf16x2.f32 %r346, %r2542, %r2541; + cvt.rn.bf16x2.f32 %r347, %r2544, %r2543; + cvt.rn.bf16x2.f32 %r348, %r2546, %r2545; + cvt.rn.bf16x2.f32 %r349, %r2548, %r2547; + // begin inline asm + @%p21 st.global.v4.b32 [ %rd87 + 0 ], { %r346, %r347, %r348, %r349 }; + // end inline asm + cvt.rn.bf16x2.f32 %r350, %r2550, %r2549; + cvt.rn.bf16x2.f32 %r351, %r2552, %r2551; + cvt.rn.bf16x2.f32 %r352, %r2554, %r2553; + cvt.rn.bf16x2.f32 %r353, %r2556, %r2555; + // begin inline asm + @%p22 st.global.v4.b32 [ %rd88 + 0 ], { %r350, %r351, %r352, %r353 }; + // end inline asm + cvt.rn.bf16x2.f32 %r354, %r2558, %r2557; + cvt.rn.bf16x2.f32 %r355, %r2560, %r2559; + cvt.rn.bf16x2.f32 %r356, %r2562, %r2561; + cvt.rn.bf16x2.f32 %r357, %r2564, %r2563; + // begin inline asm + @%p23 st.global.v4.b32 [ %rd89 + 0 ], { %r354, %r355, %r356, %r357 }; + // end inline asm + cvt.rn.bf16x2.f32 %r358, %r2566, %r2565; + cvt.rn.bf16x2.f32 %r359, %r2568, %r2567; + cvt.rn.bf16x2.f32 %r360, %r2570, %r2569; + cvt.rn.bf16x2.f32 %r361, %r2572, %r2571; + // begin inline asm + @%p24 st.global.v4.b32 [ %rd90 + 0 ], { %r358, %r359, %r360, %r361 }; + // end inline asm + cvt.rn.bf16x2.f32 %r362, %r2574, %r2573; + cvt.rn.bf16x2.f32 %r363, %r2576, %r2575; + cvt.rn.bf16x2.f32 %r364, %r2578, %r2577; + cvt.rn.bf16x2.f32 %r365, %r2580, %r2579; + // begin inline asm + @%p25 st.global.v4.b32 [ %rd91 + 0 ], { %r362, %r363, %r364, %r365 }; + // end inline asm + cvt.rn.bf16x2.f32 %r366, %r2582, %r2581; + cvt.rn.bf16x2.f32 %r367, %r2584, %r2583; + cvt.rn.bf16x2.f32 %r368, %r2586, %r2585; + cvt.rn.bf16x2.f32 %r369, %r2588, %r2587; + // begin inline asm + @%p26 st.global.v4.b32 [ %rd92 + 0 ], { %r366, %r367, %r368, %r369 }; + // end inline asm + cvt.rn.bf16x2.f32 %r370, %r2590, %r2589; + cvt.rn.bf16x2.f32 %r371, %r2592, %r2591; + cvt.rn.bf16x2.f32 %r372, %r2594, %r2593; + cvt.rn.bf16x2.f32 %r373, %r2596, %r2595; + // begin inline asm + @%p27 st.global.v4.b32 [ %rd93 + 0 ], { %r370, %r371, %r372, %r373 }; + // end inline asm + cvt.rn.bf16x2.f32 %r374, %r2598, %r2597; + cvt.rn.bf16x2.f32 %r375, %r2600, %r2599; + cvt.rn.bf16x2.f32 %r376, %r2602, %r2601; + cvt.rn.bf16x2.f32 %r377, %r2604, %r2603; + // begin inline asm + @%p28 st.global.v4.b32 [ %rd94 + 0 ], { %r374, %r375, %r376, %r377 }; + // end inline asm + cvt.rn.bf16x2.f32 %r378, %r2606, %r2605; + cvt.rn.bf16x2.f32 %r379, %r2608, %r2607; + cvt.rn.bf16x2.f32 %r380, %r2610, %r2609; + cvt.rn.bf16x2.f32 %r381, %r2612, %r2611; + // begin inline asm + @%p29 st.global.v4.b32 [ %rd95 + 0 ], { %r378, %r379, %r380, %r381 }; + // end inline asm + cvt.rn.bf16x2.f32 %r382, %r2614, %r2613; + cvt.rn.bf16x2.f32 %r383, %r2616, %r2615; + cvt.rn.bf16x2.f32 %r384, %r2618, %r2617; + cvt.rn.bf16x2.f32 %r385, %r2620, %r2619; + // begin inline asm + @%p30 st.global.v4.b32 [ %rd96 + 0 ], { %r382, %r383, %r384, %r385 }; + // end inline asm + cvt.rn.bf16x2.f32 %r386, %r2622, %r2621; + cvt.rn.bf16x2.f32 %r387, %r2624, %r2623; + cvt.rn.bf16x2.f32 %r388, %r2626, %r2625; + cvt.rn.bf16x2.f32 %r389, %r2628, %r2627; + // begin inline asm + @%p33 st.global.v4.b32 [ %rd97 + 0 ], { %r386, %r387, %r388, %r389 }; + // end inline asm + cvt.rn.bf16x2.f32 %r390, %r2630, %r2629; + cvt.rn.bf16x2.f32 %r391, %r2632, %r2631; + cvt.rn.bf16x2.f32 %r392, %r2634, %r2633; + cvt.rn.bf16x2.f32 %r393, %r2636, %r2635; + // begin inline asm + @%p34 st.global.v4.b32 [ %rd98 + 0 ], { %r390, %r391, %r392, %r393 }; + // end inline asm + cvt.rn.bf16x2.f32 %r394, %r2638, %r2637; + cvt.rn.bf16x2.f32 %r395, %r2640, %r2639; + cvt.rn.bf16x2.f32 %r396, %r2642, %r2641; + cvt.rn.bf16x2.f32 %r397, %r2644, %r2643; + // begin inline asm + @%p35 st.global.v4.b32 [ %rd99 + 0 ], { %r394, %r395, %r396, %r397 }; + // end inline asm + cvt.rn.bf16x2.f32 %r398, %r2646, %r2645; + cvt.rn.bf16x2.f32 %r399, %r2648, %r2647; + cvt.rn.bf16x2.f32 %r400, %r2650, %r2649; + cvt.rn.bf16x2.f32 %r401, %r2652, %r2651; + // begin inline asm + @%p36 st.global.v4.b32 [ %rd100 + 0 ], { %r398, %r399, %r400, %r401 }; + // end inline asm + cvt.rn.bf16x2.f32 %r402, %r2654, %r2653; + cvt.rn.bf16x2.f32 %r403, %r2656, %r2655; + cvt.rn.bf16x2.f32 %r404, %r2658, %r2657; + cvt.rn.bf16x2.f32 %r405, %r2660, %r2659; + // begin inline asm + @%p37 st.global.v4.b32 [ %rd101 + 0 ], { %r402, %r403, %r404, %r405 }; + // end inline asm + cvt.rn.bf16x2.f32 %r406, %r2662, %r2661; + cvt.rn.bf16x2.f32 %r407, %r2664, %r2663; + cvt.rn.bf16x2.f32 %r408, %r2666, %r2665; + cvt.rn.bf16x2.f32 %r409, %r2668, %r2667; + // begin inline asm + @%p38 st.global.v4.b32 [ %rd102 + 0 ], { %r406, %r407, %r408, %r409 }; + // end inline asm + cvt.rn.bf16x2.f32 %r410, %r2670, %r2669; + cvt.rn.bf16x2.f32 %r411, %r2672, %r2671; + cvt.rn.bf16x2.f32 %r412, %r2674, %r2673; + cvt.rn.bf16x2.f32 %r413, %r2676, %r2675; + // begin inline asm + @%p39 st.global.v4.b32 [ %rd103 + 0 ], { %r410, %r411, %r412, %r413 }; + // end inline asm + cvt.rn.bf16x2.f32 %r414, %r2678, %r2677; + cvt.rn.bf16x2.f32 %r415, %r2680, %r2679; + cvt.rn.bf16x2.f32 %r416, %r2682, %r2681; + cvt.rn.bf16x2.f32 %r417, %r2684, %r2683; + // begin inline asm + @%p40 st.global.v4.b32 [ %rd104 + 0 ], { %r414, %r415, %r416, %r417 }; + // end inline asm + cvt.rn.bf16x2.f32 %r418, %r2686, %r2685; + cvt.rn.bf16x2.f32 %r419, %r2688, %r2687; + cvt.rn.bf16x2.f32 %r420, %r2690, %r2689; + cvt.rn.bf16x2.f32 %r421, %r2692, %r2691; + // begin inline asm + @%p41 st.global.v4.b32 [ %rd105 + 0 ], { %r418, %r419, %r420, %r421 }; + // end inline asm + cvt.rn.bf16x2.f32 %r422, %r2694, %r2693; + cvt.rn.bf16x2.f32 %r423, %r2696, %r2695; + cvt.rn.bf16x2.f32 %r424, %r2698, %r2697; + cvt.rn.bf16x2.f32 %r425, %r2700, %r2699; + // begin inline asm + @%p42 st.global.v4.b32 [ %rd106 + 0 ], { %r422, %r423, %r424, %r425 }; + // end inline asm + cvt.rn.bf16x2.f32 %r426, %r2702, %r2701; + cvt.rn.bf16x2.f32 %r427, %r2704, %r2703; + cvt.rn.bf16x2.f32 %r428, %r2706, %r2705; + cvt.rn.bf16x2.f32 %r429, %r2708, %r2707; + // begin inline asm + @%p43 st.global.v4.b32 [ %rd107 + 0 ], { %r426, %r427, %r428, %r429 }; + // end inline asm + cvt.rn.bf16x2.f32 %r430, %r2710, %r2709; + cvt.rn.bf16x2.f32 %r431, %r2712, %r2711; + cvt.rn.bf16x2.f32 %r432, %r2714, %r2713; + cvt.rn.bf16x2.f32 %r433, %r2716, %r2715; + // begin inline asm + @%p44 st.global.v4.b32 [ %rd108 + 0 ], { %r430, %r431, %r432, %r433 }; + // end inline asm + cvt.rn.bf16x2.f32 %r434, %r2718, %r2717; + cvt.rn.bf16x2.f32 %r435, %r2720, %r2719; + cvt.rn.bf16x2.f32 %r436, %r2722, %r2721; + cvt.rn.bf16x2.f32 %r437, %r2724, %r2723; + // begin inline asm + @%p45 st.global.v4.b32 [ %rd109 + 0 ], { %r434, %r435, %r436, %r437 }; + // end inline asm + cvt.rn.bf16x2.f32 %r438, %r2726, %r2725; + cvt.rn.bf16x2.f32 %r439, %r2728, %r2727; + cvt.rn.bf16x2.f32 %r440, %r2730, %r2729; + cvt.rn.bf16x2.f32 %r441, %r2732, %r2731; + // begin inline asm + @%p46 st.global.v4.b32 [ %rd110 + 0 ], { %r438, %r439, %r440, %r441 }; + // end inline asm + cvt.rn.bf16x2.f32 %r442, %r2734, %r2733; + cvt.rn.bf16x2.f32 %r443, %r2736, %r2735; + cvt.rn.bf16x2.f32 %r444, %r2738, %r2737; + cvt.rn.bf16x2.f32 %r445, %r2740, %r2739; + // begin inline asm + @%p47 st.global.v4.b32 [ %rd111 + 0 ], { %r442, %r443, %r444, %r445 }; + // end inline asm + cvt.rn.bf16x2.f32 %r446, %r2742, %r2741; + cvt.rn.bf16x2.f32 %r447, %r2744, %r2743; + cvt.rn.bf16x2.f32 %r448, %r2746, %r2745; + cvt.rn.bf16x2.f32 %r449, %r2748, %r2747; + // begin inline asm + @%p48 st.global.v4.b32 [ %rd112 + 0 ], { %r446, %r447, %r448, %r449 }; + // end inline asm + cvt.rn.bf16x2.f32 %r451, %r2752, %r2751; + cvt.rn.bf16x2.f32 %r452, %r2754, %r2753; + cvt.rn.bf16x2.f32 %r453, %r2756, %r2755; + // begin inline asm + @%p49 st.global.v4.b32 [ %rd113 + 0 ], { %r450, %r451, %r452, %r453 }; + // end inline asm + // begin inline asm + @%p50 st.global.v4.b32 [ %rd114 + 0 ], { %r454, %r455, %r456, %r457 }; + // end inline asm + // begin inline asm + @%p51 st.global.v4.b32 [ %rd115 + 0 ], { %r458, %r459, %r460, %r461 }; + // end inline asm + // begin inline asm + @%p52 st.global.v4.b32 [ %rd116 + 0 ], { %r462, %r463, %r464, %r465 }; + // end inline asm + // begin inline asm + @%p53 st.global.v4.b32 [ %rd117 + 0 ], { %r466, %r467, %r468, %r469 }; + // end inline asm + // begin inline asm + @%p54 st.global.v4.b32 [ %rd118 + 0 ], { %r470, %r471, %r472, %r473 }; + // end inline asm + // begin inline asm + @%p55 st.global.v4.b32 [ %rd119 + 0 ], { %r474, %r475, %r476, %r477 }; + // end inline asm + // begin inline asm + @%p56 st.global.v4.b32 [ %rd120 + 0 ], { %r478, %r479, %r480, %r481 }; + // end inline asm + // begin inline asm + @%p57 st.global.v4.b32 [ %rd121 + 0 ], { %r482, %r483, %r484, %r485 }; + // end inline asm + // begin inline asm + @%p58 st.global.v4.b32 [ %rd122 + 0 ], { %r486, %r487, %r488, %r489 }; + // end inline asm + // begin inline asm + @%p59 st.global.v4.b32 [ %rd123 + 0 ], { %r490, %r491, %r492, %r493 }; + // end inline asm + // begin inline asm + @%p60 st.global.v4.b32 [ %rd124 + 0 ], { %r494, %r495, %r496, %r497 }; + // end inline asm + // begin inline asm + @%p61 st.global.v4.b32 [ %rd125 + 0 ], { %r498, %r499, %r500, %r501 }; + // end inline asm + // begin inline asm + @%p62 st.global.v4.b32 [ %rd126 + 0 ], { %r502, %r503, %r504, %r505 }; + // end inline asm + // begin inline asm + @%p63 st.global.v4.b32 [ %rd127 + 0 ], { %r506, %r507, %r508, %r509 }; + // end inline asm + // begin inline asm + @%p64 st.global.v4.b32 [ %rd128 + 0 ], { %r510, %r511, %r512, %r513 }; + // end inline asm + .loc 1 84 21 // l2norm.py:84:21 + shl.b32 %r2877, %r593, 3; + mov.b32 %r2878, global_smem; + add.s32 %r514, %r2878, %r2877; + // begin inline asm + st.shared.v4.b32 [ %r514 + 0 ], { %r515, %r516, %r517, %r518 }; + // end inline asm + add.s32 %r519, %r514, 16; + // begin inline asm + st.shared.v4.b32 [ %r519 + 0 ], { %r520, %r521, %r522, %r523 }; + // end inline asm + add.s32 %r524, %r514, 32; + // begin inline asm + st.shared.v4.b32 [ %r524 + 0 ], { %r525, %r526, %r527, %r528 }; + // end inline asm + add.s32 %r529, %r514, 48; + // begin inline asm + st.shared.v4.b32 [ %r529 + 0 ], { %r530, %r531, %r532, %r533 }; + // end inline asm + add.s32 %r534, %r514, 64; + // begin inline asm + st.shared.v4.b32 [ %r534 + 0 ], { %r535, %r536, %r537, %r538 }; + // end inline asm + add.s32 %r539, %r514, 80; + // begin inline asm + st.shared.v4.b32 [ %r539 + 0 ], { %r540, %r541, %r542, %r543 }; + // end inline asm + add.s32 %r544, %r514, 96; + // begin inline asm + st.shared.v4.b32 [ %r544 + 0 ], { %r545, %r546, %r547, %r548 }; + // end inline asm + add.s32 %r549, %r514, 112; + // begin inline asm + st.shared.v4.b32 [ %r549 + 0 ], { %r550, %r551, %r552, %r553 }; + // end inline asm + bar.warp.sync -1; + shl.b32 %r2879, %r678, 4; + add.s32 %r2880, %r2878, %r2879; + ldmatrix.sync.aligned.m8n8.x2.shared.b16 {%r586, %r587}, [%r2880]; + bar.warp.sync -1; + // begin inline asm + st.shared.v4.b32 [ %r514 + 0 ], { %r554, %r555, %r556, %r557 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r519 + 0 ], { %r558, %r559, %r560, %r561 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r524 + 0 ], { %r562, %r563, %r564, %r565 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r529 + 0 ], { %r566, %r567, %r568, %r569 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r534 + 0 ], { %r570, %r571, %r572, %r573 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r539 + 0 ], { %r574, %r575, %r576, %r577 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r544 + 0 ], { %r578, %r579, %r580, %r581 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r549 + 0 ], { %r582, %r583, %r584, %r585 }; + // end inline asm + bar.warp.sync -1; + ldmatrix.sync.aligned.m8n8.x2.shared.b16 {%r588, %r589}, [%r2880]; + // begin inline asm + @%p65 st.global.b32 [ %rd129 + 0 ], { %r586 }; + // end inline asm + // begin inline asm + @%p66 st.global.b32 [ %rd130 + 0 ], { %r587 }; + // end inline asm + // begin inline asm + @%p67 st.global.b32 [ %rd131 + 0 ], { %r588 }; + // end inline asm + // begin inline asm + @%p68 st.global.b32 [ %rd132 + 0 ], { %r589 }; + // end inline asm + .loc 1 84 4 // l2norm.py:84:4 + ret; +$L__tmp515: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp514 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 80 // DW_AT_call_line +.b8 43 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp514 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.source b/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..65999ed34486c21231e2e10dd8e7d7c057966940 --- /dev/null +++ b/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.source @@ -0,0 +1,172 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc38 = loc(unknown) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc45 = loc("x"(#loc)) +#loc46 = loc("y"(#loc)) +#loc47 = loc("rstd"(#loc)) +#loc48 = loc("eps"(#loc)) +#loc49 = loc("T"(#loc)) +#loc81 = loc("input"(#loc36)) +#loc82 = loc("a"(#loc41)) +#loc83 = loc("b"(#loc41)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc50) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc51) + %o_t = arith.constant 128 : i32 loc(#loc52) + %o_t_1 = arith.constant 128 : i64 loc(#loc52) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc52) + %o_t_3 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc53) + %o_t_4 = arith.extsi %o_t_3 : tensor<128xi32> to tensor<128xi64> loc(#loc54) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<128xi64> loc(#loc54) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<128xi64> loc(#loc54) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc55) + %m_t = arith.extsi %T : i32 to i64 loc(#loc56) + %m_t_7 = tt.splat %m_t : i64 -> tensor<128xi64> loc(#loc56) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<128xi64> loc(#loc56) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<128xi1> -> tensor<128x1xi1> loc(#loc57) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc58) + %m_x_10 = arith.constant 128 : i32 loc(#loc59) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc59) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc59) + %m_x_13 = tt.broadcast %m_x : tensor<128x1xi1> -> tensor<128x128xi1> loc(#loc60) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<128x128xi1> loc(#loc60) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<128x128xi1> loc(#loc60) + %p_x = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc61) + %p_x_16 = arith.constant 128 : i32 loc(#loc62) + %p_x_17 = arith.constant 128 : i64 loc(#loc62) + %p_x_18 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc62) + %p_x_19 = arith.muli %p_x, %p_x_18 : tensor<128x1xi64> loc(#loc62) + %p_x_20 = tt.splat %x : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc63) + %p_x_21 = tt.addptr %p_x_20, %p_x_19 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc63) + %p_x_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc64) + %p_x_23 = tt.broadcast %p_x_21 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc65) + %p_x_24 = tt.broadcast %p_x_22 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc65) + %p_x_25 = tt.addptr %p_x_23, %p_x_24 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc65) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc66) + %p_y_26 = arith.constant 128 : i32 loc(#loc67) + %p_y_27 = arith.constant 128 : i64 loc(#loc67) + %p_y_28 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc67) + %p_y_29 = arith.muli %p_y, %p_y_28 : tensor<128x1xi64> loc(#loc67) + %p_y_30 = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc68) + %p_y_31 = tt.addptr %p_y_30, %p_y_29 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc68) + %p_y_32 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc69) + %p_y_33 = tt.broadcast %p_y_31 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc70) + %p_y_34 = tt.broadcast %p_y_32 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc70) + %p_y_35 = tt.addptr %p_y_33, %p_y_34 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc70) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr> loc(#loc71) + %p_rstd_36 = tt.addptr %p_rstd, %o_t_6 : tensor<128x!tt.ptr>, tensor<128xi64> loc(#loc71) + %b_x = arith.constant 0.000000e+00 : f32 loc(#loc72) + %b_x_37 = arith.constant dense<0.000000e+00> : tensor<128x128xf32> loc(#loc72) + %b_x_38 = arith.truncf %b_x_37 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc72) + %b_x_39 = tt.load %p_x_25, %m_x_15, %b_x_38 : tensor<128x128x!tt.ptr> loc(#loc72) + %b_x_40 = arith.extf %b_x_39 : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc73) + %b_rstd = arith.mulf %b_x_40, %b_x_40 : tensor<128x128xf32> loc(#loc74) + %b_rstd_41 = tt.call @triton.language.standard.sum__fp32S128_128S_c1_cFalse_cNone(%b_rstd) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc75) + %b_rstd_42 = tt.splat %eps : f32 -> tensor<128xf32> loc(#loc76) + %b_rstd_43 = arith.addf %b_rstd_41, %b_rstd_42 : tensor<128xf32> loc(#loc76) + %b_rstd_44 = math.sqrt %b_rstd_43 : tensor<128xf32> loc(#loc77) + %b_rstd_45 = arith.constant 1 : i32 loc(#loc78) + %b_rstd_46 = arith.constant 1.000000e+00 : f32 loc(#loc78) + %b_rstd_47 = arith.constant dense<1.000000e+00> : tensor<128xf32> loc(#loc78) + %b_rstd_48 = arith.divf %b_rstd_47, %b_rstd_44 : tensor<128xf32> loc(#loc78) + %b_y = tt.expand_dims %b_rstd_48 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc79) + %b_y_49 = tt.broadcast %b_y : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc80) + %b_y_50 = arith.mulf %b_x_40, %b_y_49 : tensor<128x128xf32> loc(#loc80) + %0 = arith.truncf %b_y_50 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc32) + tt.store %p_y_35, %0, %m_x_15 : tensor<128x128x!tt.ptr> loc(#loc33) + tt.store %p_rstd_36, %b_rstd_48, %m_t_8 : tensor<128x!tt.ptr> loc(#loc34) + tt.return loc(#loc35) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S128_128S_c1_cFalse_cNone(%input: tensor<128x128xf32> loc("input"(#loc36))) -> tensor<128xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc37) + tt.reduce.return %2 : f32 loc(#loc37) + }) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc37) + tt.return %0 : tensor<128xf32> loc(#loc39) + ^bb1: // no predecessors + %1 = ub.poison : tensor<128xf32> loc(#loc40) + tt.return %1 : tensor<128xf32> loc(#loc40) + } loc(#loc36) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc41)), %b: f32 loc("b"(#loc41))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc42) + tt.return %0 : f32 loc(#loc43) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc44) + tt.return %1 : f32 loc(#loc44) + } loc(#loc41) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:18) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:29) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:37) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc50 = loc("i_t"(#loc1)) +#loc51 = loc("i_t"(#loc2)) +#loc52 = loc("o_t"(#loc3)) +#loc53 = loc("o_t"(#loc4)) +#loc54 = loc("o_t"(#loc5)) +#loc55 = loc("o_d"(#loc6)) +#loc56 = loc("m_t"(#loc7)) +#loc57 = loc("m_x"(#loc8)) +#loc58 = loc("m_x"(#loc9)) +#loc59 = loc("m_x"(#loc10)) +#loc60 = loc("m_x"(#loc11)) +#loc61 = loc("p_x"(#loc12)) +#loc62 = loc("p_x"(#loc13)) +#loc63 = loc("p_x"(#loc14)) +#loc64 = loc("p_x"(#loc15)) +#loc65 = loc("p_x"(#loc16)) +#loc66 = loc("p_y"(#loc17)) +#loc67 = loc("p_y"(#loc18)) +#loc68 = loc("p_y"(#loc19)) +#loc69 = loc("p_y"(#loc20)) +#loc70 = loc("p_y"(#loc21)) +#loc71 = loc("p_rstd"(#loc22)) +#loc72 = loc("b_x"(#loc23)) +#loc73 = loc("b_x"(#loc24)) +#loc74 = loc("b_rstd"(#loc25)) +#loc75 = loc("b_rstd"(#loc26)) +#loc76 = loc("b_rstd"(#loc27)) +#loc77 = loc("b_rstd"(#loc28)) +#loc78 = loc("b_rstd"(#loc29)) +#loc79 = loc("b_y"(#loc30)) +#loc80 = loc("b_y"(#loc31)) diff --git a/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.ttgir b/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..6eb357769da8a34bbb4289962570dfabcd53be59 --- /dev/null +++ b/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.ttgir @@ -0,0 +1,135 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [1, 1], order = [1, 0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [4], threadsPerWarp = [32], warpsPerCTA = [1], order = [0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc1 = loc(unknown) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc34 = loc("x"(#loc)) +#loc35 = loc("y"(#loc)) +#loc36 = loc("rstd"(#loc)) +#loc37 = loc("eps"(#loc)) +#loc38 = loc("T"(#loc)) +#loc59 = loc("b_rstd"(#loc23)) +#loc66 = loc(callsite(#loc1 at #loc59)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 1 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<1.000000e+00> : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<128x128xbf16, #blocked> loc(#loc1) + %c128_i64 = arith.constant 128 : i64 loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<128x1xi64, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc39) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc40) + %o_t = arith.muli %i_t_3, %c128_i64 : i64 loc(#loc41) + %o_t_4 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc42) + %o_t_5 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #blocked1> loc(#loc42) + %o_t_6 = arith.extsi %o_t_4 : tensor<128xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_7 = arith.extsi %o_t_5 : tensor<128xi32, #blocked1> to tensor<128xi64, #blocked1> loc(#loc43) + %o_t_8 = tt.splat %o_t : i64 -> tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_9 = tt.splat %o_t : i64 -> tensor<128xi64, #blocked1> loc(#loc43) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<128xi64, #blocked1> loc(#loc43) + %m_t = arith.extsi %T : i32 to i64 loc(#loc44) + %m_t_12 = tt.splat %m_t : i64 -> tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_13 = tt.splat %m_t : i64 -> tensor<128xi64, #blocked1> loc(#loc44) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<128xi64, #blocked1> loc(#loc44) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<128xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128x1xi1, #blocked> loc(#loc45) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc46) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc46) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_1 : tensor<1x128xi32, #blocked> loc(#loc47) + %m_x_19 = tt.broadcast %m_x : tensor<128x1xi1, #blocked> -> tensor<128x128xi1, #blocked> loc(#loc48) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<128x128xi1, #blocked> loc(#loc48) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<128x128xi1, #blocked> loc(#loc48) + %p_x = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128x1xi64, #blocked> loc(#loc49) + %p_x_22 = arith.muli %p_x, %cst_2 : tensor<128x1xi64, #blocked> loc(#loc50) + %p_x_23 = tt.splat %x : !tt.ptr -> tensor<128x1x!tt.ptr, #blocked> loc(#loc51) + %p_x_24 = tt.addptr %p_x_23, %p_x_22 : tensor<128x1x!tt.ptr, #blocked>, tensor<128x1xi64, #blocked> loc(#loc51) + %p_x_25 = tt.broadcast %p_x_24 : tensor<128x1x!tt.ptr, #blocked> -> tensor<128x128x!tt.ptr, #blocked> loc(#loc52) + %p_x_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<128x128xi32, #blocked> loc(#loc52) + %p_x_27 = tt.addptr %p_x_25, %p_x_26 : tensor<128x128x!tt.ptr, #blocked>, tensor<128x128xi32, #blocked> loc(#loc52) + %p_y = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr, #blocked> loc(#loc53) + %p_y_28 = tt.addptr %p_y, %p_x_22 : tensor<128x1x!tt.ptr, #blocked>, tensor<128x1xi64, #blocked> loc(#loc53) + %p_y_29 = tt.broadcast %p_y_28 : tensor<128x1x!tt.ptr, #blocked> -> tensor<128x128x!tt.ptr, #blocked> loc(#loc54) + %p_y_30 = tt.addptr %p_y_29, %p_x_26 : tensor<128x128x!tt.ptr, #blocked>, tensor<128x128xi32, #blocked> loc(#loc54) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr, #blocked1> loc(#loc55) + %p_rstd_31 = tt.addptr %p_rstd, %o_t_11 : tensor<128x!tt.ptr, #blocked1>, tensor<128xi64, #blocked1> loc(#loc55) + %b_x = tt.load %p_x_27, %m_x_21, %cst_0 : tensor<128x128x!tt.ptr, #blocked> loc(#loc56) + %b_x_32 = arith.extf %b_x : tensor<128x128xbf16, #blocked> to tensor<128x128xf32, #blocked> loc(#loc57) + %b_rstd = arith.mulf %b_x_32, %b_x_32 : tensor<128x128xf32, #blocked> loc(#loc58) + %b_rstd_33 = "tt.reduce"(%b_rstd) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_40: f32 loc(callsite(#loc1 at #loc59)), %b_rstd_41: f32 loc(callsite(#loc1 at #loc59))): + %b_rstd_42 = arith.addf %b_rstd_40, %b_rstd_41 : f32 loc(#loc67) + tt.reduce.return %b_rstd_42 : f32 loc(#loc65) + }) : (tensor<128x128xf32, #blocked>) -> tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc65) + %b_rstd_34 = tt.splat %eps : f32 -> tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_36 = math.sqrt %b_rstd_35 : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc61) + %b_rstd_37 = arith.divf %cst, %b_rstd_36 : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc62) + %b_y = tt.expand_dims %b_rstd_37 {axis = 1 : i32} : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128x1xf32, #blocked> loc(#loc63) + %b_y_38 = tt.broadcast %b_y : tensor<128x1xf32, #blocked> -> tensor<128x128xf32, #blocked> loc(#loc64) + %b_y_39 = arith.mulf %b_x_32, %b_y_38 : tensor<128x128xf32, #blocked> loc(#loc64) + %0 = arith.truncf %b_y_39 : tensor<128x128xf32, #blocked> to tensor<128x128xbf16, #blocked> loc(#loc30) + tt.store %p_y_30, %0, %m_x_21 : tensor<128x128x!tt.ptr, #blocked> loc(#loc31) + %1 = ttg.convert_layout %b_rstd_37 : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128xf32, #blocked1> loc(#loc32) + tt.store %p_rstd_31, %1, %m_t_15 : tensor<128x!tt.ptr, #blocked1> loc(#loc32) + tt.return loc(#loc33) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc39 = loc("i_t"(#loc2)) +#loc40 = loc("i_t"(#loc3)) +#loc41 = loc("o_t"(#loc4)) +#loc42 = loc("o_t"(#loc5)) +#loc43 = loc("o_t"(#loc6)) +#loc44 = loc("m_t"(#loc7)) +#loc45 = loc("m_x"(#loc8)) +#loc46 = loc("m_x"(#loc9)) +#loc47 = loc("m_x"(#loc10)) +#loc48 = loc("m_x"(#loc11)) +#loc49 = loc("p_x"(#loc12)) +#loc50 = loc("p_x"(#loc13)) +#loc51 = loc("p_x"(#loc14)) +#loc52 = loc("p_x"(#loc15)) +#loc53 = loc("p_y"(#loc16)) +#loc54 = loc("p_y"(#loc17)) +#loc55 = loc("p_rstd"(#loc18)) +#loc56 = loc("b_x"(#loc19)) +#loc57 = loc("b_x"(#loc20)) +#loc58 = loc("b_rstd"(#loc21)) +#loc60 = loc("b_rstd"(#loc25)) +#loc61 = loc("b_rstd"(#loc26)) +#loc62 = loc("b_rstd"(#loc27)) +#loc63 = loc("b_y"(#loc28)) +#loc64 = loc("b_y"(#loc29)) +#loc65 = loc(callsite(#loc22 at #loc59)) +#loc67 = loc(callsite(#loc24 at #loc65)) diff --git a/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.ttir b/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..8f0dfe3371851abdc6b1c20ee94003c63d3c42a8 --- /dev/null +++ b/triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.ttir @@ -0,0 +1,125 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc3 = loc(unknown) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc34 = loc("x"(#loc)) +#loc35 = loc("y"(#loc)) +#loc36 = loc("rstd"(#loc)) +#loc37 = loc("eps"(#loc)) +#loc38 = loc("T"(#loc)) +#loc60 = loc("b_rstd"(#loc24)) +#loc66 = loc(callsite(#loc3 at #loc60)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %b_x = arith.constant dense<0.000000e+00> : tensor<128x128xbf16> loc(#loc39) + %b_rstd = arith.constant dense<1.000000e+00> : tensor<128xf32> loc(#loc40) + %cst = arith.constant dense<128> : tensor<128x1xi64> loc(#loc3) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc41) + %c128_i64 = arith.constant 128 : i64 loc(#loc3) + %i_t = tt.get_program_id x : i32 loc(#loc42) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc43) + %o_t = arith.muli %i_t_0, %c128_i64 : i64 loc(#loc44) + %o_t_1 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc45) + %o_t_2 = arith.extsi %o_t_1 : tensor<128xi32> to tensor<128xi64> loc(#loc46) + %o_t_3 = tt.splat %o_t : i64 -> tensor<128xi64> loc(#loc46) + %o_t_4 = arith.addi %o_t_3, %o_t_2 : tensor<128xi64> loc(#loc46) + %m_t = arith.extsi %T : i32 to i64 loc(#loc47) + %m_t_5 = tt.splat %m_t : i64 -> tensor<128xi64> loc(#loc47) + %m_t_6 = arith.cmpi slt, %o_t_4, %m_t_5 : tensor<128xi64> loc(#loc47) + %m_x_7 = tt.expand_dims %m_t_6 {axis = 1 : i32} : tensor<128xi1> -> tensor<128x1xi1> loc(#loc48) + %m_x_8 = tt.expand_dims %o_t_1 {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc49) + %m_x_9 = arith.cmpi slt, %m_x_8, %m_x : tensor<1x128xi32> loc(#loc41) + %m_x_10 = tt.broadcast %m_x_7 : tensor<128x1xi1> -> tensor<128x128xi1> loc(#loc50) + %m_x_11 = tt.broadcast %m_x_9 : tensor<1x128xi1> -> tensor<128x128xi1> loc(#loc50) + %m_x_12 = arith.andi %m_x_10, %m_x_11 : tensor<128x128xi1> loc(#loc50) + %p_x = tt.expand_dims %o_t_4 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc51) + %p_x_13 = arith.muli %p_x, %cst : tensor<128x1xi64> loc(#loc52) + %p_x_14 = tt.splat %x : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc53) + %p_x_15 = tt.addptr %p_x_14, %p_x_13 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc53) + %p_x_16 = tt.broadcast %p_x_15 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc54) + %p_x_17 = tt.broadcast %m_x_8 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc54) + %p_x_18 = tt.addptr %p_x_16, %p_x_17 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc54) + %p_y = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc55) + %p_y_19 = tt.addptr %p_y, %p_x_13 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc55) + %p_y_20 = tt.broadcast %p_y_19 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc56) + %p_y_21 = tt.addptr %p_y_20, %p_x_17 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc56) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr> loc(#loc57) + %p_rstd_22 = tt.addptr %p_rstd, %o_t_4 : tensor<128x!tt.ptr>, tensor<128xi64> loc(#loc57) + %b_x_23 = tt.load %p_x_18, %m_x_12, %b_x : tensor<128x128x!tt.ptr> loc(#loc39) + %b_x_24 = arith.extf %b_x_23 : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc58) + %b_rstd_25 = arith.mulf %b_x_24, %b_x_24 : tensor<128x128xf32> loc(#loc59) + %b_rstd_26 = "tt.reduce"(%b_rstd_25) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_33: f32 loc(callsite(#loc3 at #loc60)), %b_rstd_34: f32 loc(callsite(#loc3 at #loc60))): + %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : f32 loc(#loc67) + tt.reduce.return %b_rstd_35 : f32 loc(#loc65) + }) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc65) + %b_rstd_27 = tt.splat %eps : f32 -> tensor<128xf32> loc(#loc61) + %b_rstd_28 = arith.addf %b_rstd_26, %b_rstd_27 : tensor<128xf32> loc(#loc61) + %b_rstd_29 = math.sqrt %b_rstd_28 : tensor<128xf32> loc(#loc62) + %b_rstd_30 = arith.divf %b_rstd, %b_rstd_29 : tensor<128xf32> loc(#loc40) + %b_y = tt.expand_dims %b_rstd_30 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc63) + %b_y_31 = tt.broadcast %b_y : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc64) + %b_y_32 = arith.mulf %b_x_24, %b_y_31 : tensor<128x128xf32> loc(#loc64) + %0 = arith.truncf %b_y_32 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc30) + tt.store %p_y_21, %0, %m_x_12 : tensor<128x128x!tt.ptr> loc(#loc31) + tt.store %p_rstd_22, %b_rstd_30, %m_t_6 : tensor<128x!tt.ptr> loc(#loc32) + tt.return loc(#loc33) + } loc(#loc) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc39 = loc("b_x"(#loc1)) +#loc40 = loc("b_rstd"(#loc2)) +#loc41 = loc("m_x"(#loc4)) +#loc42 = loc("i_t"(#loc5)) +#loc43 = loc("i_t"(#loc6)) +#loc44 = loc("o_t"(#loc7)) +#loc45 = loc("o_t"(#loc8)) +#loc46 = loc("o_t"(#loc9)) +#loc47 = loc("m_t"(#loc10)) +#loc48 = loc("m_x"(#loc11)) +#loc49 = loc("m_x"(#loc12)) +#loc50 = loc("m_x"(#loc13)) +#loc51 = loc("p_x"(#loc14)) +#loc52 = loc("p_x"(#loc15)) +#loc53 = loc("p_x"(#loc16)) +#loc54 = loc("p_x"(#loc17)) +#loc55 = loc("p_y"(#loc18)) +#loc56 = loc("p_y"(#loc19)) +#loc57 = loc("p_rstd"(#loc20)) +#loc58 = loc("b_x"(#loc21)) +#loc59 = loc("b_rstd"(#loc22)) +#loc61 = loc("b_rstd"(#loc26)) +#loc62 = loc("b_rstd"(#loc27)) +#loc63 = loc("b_y"(#loc28)) +#loc64 = loc("b_y"(#loc29)) +#loc65 = loc(callsite(#loc23 at #loc60)) +#loc67 = loc(callsite(#loc25 at #loc65)) diff --git a/triton/gpu0/BWXLAG2FQUH6U5U77N4LFOPIOPFT7RH5XYFRGZSQTHBOXIBHSFJQ/l2norm_bwd_kernel.autotune.json b/triton/gpu0/BWXLAG2FQUH6U5U77N4LFOPIOPFT7RH5XYFRGZSQTHBOXIBHSFJQ/l2norm_bwd_kernel.autotune.json new file mode 100644 index 0000000000000000000000000000000000000000..67830368114bbb423e9bfd17096ccccea5986e5a --- /dev/null +++ b/triton/gpu0/BWXLAG2FQUH6U5U77N4LFOPIOPFT7RH5XYFRGZSQTHBOXIBHSFJQ/l2norm_bwd_kernel.autotune.json @@ -0,0 +1 @@ +{"key": [128, 16, "torch.bfloat16", "torch.float32", "torch.bfloat16", "torch.bfloat16"], "configs_timings": [[{"kwargs": {"BT": 8}, "num_warps": 1, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.12284799665212631, 0.12278400361537933, 0.12294399738311768]], [{"kwargs": {"BT": 16}, "num_warps": 1, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.12291199713945389, 0.12284799665212631, 0.12486399710178375]], [{"kwargs": {"BT": 32}, "num_warps": 1, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.13312000036239624, 0.13305599987506866, 0.133215993642807]], [{"kwargs": {"BT": 64}, "num_warps": 1, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.17817600071430206, 0.1780479997396469, 0.18012799322605133]], [{"kwargs": {"BT": 128}, "num_warps": 1, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.6339040100574493, 0.6327360272407532, 0.6349440217018127]], [{"kwargs": {"BT": 8}, "num_warps": 2, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.1228799968957901, 0.12275200337171555, 0.12297599762678146]], [{"kwargs": {"BT": 16}, "num_warps": 2, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.1228799968957901, 0.12278400361537933, 0.12294399738311768]], [{"kwargs": {"BT": 32}, "num_warps": 2, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.1228799968957901, 0.12281599640846252, 0.12297599762678146]], [{"kwargs": {"BT": 64}, "num_warps": 2, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.13516800105571747, 0.1350719928741455, 0.13523200154304504]], [{"kwargs": {"BT": 128}, "num_warps": 2, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.1884479969739914, 0.18835200369358063, 0.18953600525856018]], [{"kwargs": {"BT": 8}, "num_warps": 4, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.12278400361537933, 0.12083200365304947, 0.12291199713945389]], [{"kwargs": {"BT": 16}, "num_warps": 4, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.1228799968957901, 0.12185599654912949, 0.12297599762678146]], [{"kwargs": {"BT": 32}, "num_warps": 4, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.1228799968957901, 0.12278400361537933, 0.12300799787044525]], [{"kwargs": {"BT": 64}, "num_warps": 4, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.1228799968957901, 0.12281599640846252, 0.12297599762678146]], [{"kwargs": {"BT": 128}, "num_warps": 4, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.1372160017490387, 0.13715200126171112, 0.13728000223636627]], [{"kwargs": {"BT": 8}, "num_warps": 8, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.14947199821472168, 0.1493760049343109, 0.14953599870204926]], [{"kwargs": {"BT": 16}, "num_warps": 8, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.12086399644613266, 0.1207680031657219, 0.12284799665212631]], [{"kwargs": {"BT": 32}, "num_warps": 8, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.1228799968957901, 0.12095999717712402, 0.12297599762678146]], [{"kwargs": {"BT": 64}, "num_warps": 8, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.1228799968957901, 0.12281599640846252, 0.12297599762678146]], [{"kwargs": {"BT": 128}, "num_warps": 8, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.12697599828243256, 0.12691199779510498, 0.12703999876976013]], [{"kwargs": {"BT": 8}, "num_warps": 16, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.2826240062713623, 0.28252801299095154, 0.2826879918575287]], [{"kwargs": {"BT": 16}, "num_warps": 16, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.1576479971408844, 0.15756799280643463, 0.15772800147533417]], [{"kwargs": {"BT": 32}, "num_warps": 16, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.12284799665212631, 0.12278400361537933, 0.12371840327978134]], [{"kwargs": {"BT": 64}, "num_warps": 16, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.12281599640846252, 0.12086399644613266, 0.12291199713945389]], [{"kwargs": {"BT": 128}, "num_warps": 16, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.12492799758911133, 0.12483199685811996, 0.1249919980764389]]]} \ No newline at end of file diff --git a/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/__grp__l2norm_fwd_kernel.json b/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/__grp__l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..be53f1e289228f1dbbb2bb2205a0b7c416d1bea8 --- /dev/null +++ b/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/__grp__l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_fwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/l2norm_fwd_kernel.source", "l2norm_fwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/l2norm_fwd_kernel.ttir", "l2norm_fwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/l2norm_fwd_kernel.ttgir", "l2norm_fwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/l2norm_fwd_kernel.llir", "l2norm_fwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/l2norm_fwd_kernel.ptx", "l2norm_fwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/l2norm_fwd_kernel.cubin", "l2norm_fwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/l2norm_fwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/l2norm_fwd_kernel.cubin b/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/l2norm_fwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..0527965d4acfb01c0970128fc7e3ed49ba5b7997 Binary files /dev/null and b/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/l2norm_fwd_kernel.cubin differ diff --git a/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/l2norm_fwd_kernel.json b/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..d3322a2ed7afd96d06c9eff2933dee42dd64b713 --- /dev/null +++ b/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"hash": "0ddd7b4e2e6ba2be9f7856f7e4b3480e80c50c8662931549958b4bb801846a12", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 4, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 256, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_fwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/l2norm_fwd_kernel.llir b/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/l2norm_fwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..d74709cd72ee444d73eac08f055ded0f7bfbb8b5 --- /dev/null +++ b/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/l2norm_fwd_kernel.llir @@ -0,0 +1,1146 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 +@.str.2 = private unnamed_addr constant [17 x i8] c"__CUDA_PREC_SQRT\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_fwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, float %3, i32 %4, ptr addrspace(1) readnone captures(none) %5, ptr addrspace(1) readnone captures(none) %6) local_unnamed_addr #0 !dbg !5 { + %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %9 = zext nneg i32 %8 to i64, !dbg !14 + %10 = shl nuw nsw i64 %9, 6, !dbg !15 + %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !16 + %12 = and i32 %11, 112, !dbg !16 + %13 = lshr exact i32 %12, 4, !dbg !16 + %14 = or disjoint i32 %13, 8, !dbg !16 + %15 = or disjoint i32 %13, 16, !dbg !16 + %16 = or disjoint i32 %13, 24, !dbg !16 + %17 = or disjoint i32 %13, 32, !dbg !16 + %18 = or disjoint i32 %13, 40, !dbg !16 + %19 = or disjoint i32 %13, 48, !dbg !16 + %20 = or disjoint i32 %13, 56, !dbg !16 + %21 = zext nneg i32 %13 to i64, !dbg !17 + %22 = zext nneg i32 %14 to i64, !dbg !17 + %23 = zext nneg i32 %15 to i64, !dbg !17 + %24 = zext nneg i32 %16 to i64, !dbg !17 + %25 = zext nneg i32 %17 to i64, !dbg !17 + %26 = zext nneg i32 %18 to i64, !dbg !17 + %27 = zext nneg i32 %19 to i64, !dbg !17 + %28 = zext nneg i32 %20 to i64, !dbg !17 + %29 = or disjoint i64 %10, %21, !dbg !17 + %30 = or disjoint i64 %10, %22, !dbg !17 + %31 = or disjoint i64 %10, %23, !dbg !17 + %32 = or disjoint i64 %10, %24, !dbg !17 + %33 = or disjoint i64 %10, %25, !dbg !17 + %34 = or disjoint i64 %10, %26, !dbg !17 + %35 = or disjoint i64 %10, %27, !dbg !17 + %36 = or disjoint i64 %10, %28, !dbg !17 + %37 = sext i32 %4 to i64, !dbg !18 + %38 = icmp slt i64 %29, %37, !dbg !18 + %39 = icmp slt i64 %30, %37, !dbg !18 + %40 = icmp slt i64 %31, %37, !dbg !18 + %41 = icmp slt i64 %32, %37, !dbg !18 + %42 = icmp slt i64 %33, %37, !dbg !18 + %43 = icmp slt i64 %34, %37, !dbg !18 + %44 = icmp slt i64 %35, %37, !dbg !18 + %45 = icmp slt i64 %36, %37, !dbg !18 + %46 = shl nuw nsw i32 %11, 3, !dbg !19 + %47 = and i32 %46, 120, !dbg !19 + %48 = shl nuw nsw i64 %29, 7, !dbg !20 + %49 = shl nuw nsw i64 %30, 7, !dbg !20 + %50 = shl nuw nsw i64 %31, 7, !dbg !20 + %51 = shl nuw nsw i64 %32, 7, !dbg !20 + %52 = shl nuw nsw i64 %33, 7, !dbg !20 + %53 = shl nuw nsw i64 %34, 7, !dbg !20 + %54 = shl nuw nsw i64 %35, 7, !dbg !20 + %55 = shl nuw nsw i64 %36, 7, !dbg !20 + %56 = getelementptr bfloat, ptr addrspace(1) %0, i64 %48, !dbg !21 + %57 = getelementptr bfloat, ptr addrspace(1) %0, i64 %49, !dbg !21 + %58 = getelementptr bfloat, ptr addrspace(1) %0, i64 %50, !dbg !21 + %59 = getelementptr bfloat, ptr addrspace(1) %0, i64 %51, !dbg !21 + %60 = getelementptr bfloat, ptr addrspace(1) %0, i64 %52, !dbg !21 + %61 = getelementptr bfloat, ptr addrspace(1) %0, i64 %53, !dbg !21 + %62 = getelementptr bfloat, ptr addrspace(1) %0, i64 %54, !dbg !21 + %63 = getelementptr bfloat, ptr addrspace(1) %0, i64 %55, !dbg !21 + %64 = zext nneg i32 %47 to i64, !dbg !22 + %65 = getelementptr bfloat, ptr addrspace(1) %56, i64 %64, !dbg !22 + %66 = getelementptr bfloat, ptr addrspace(1) %57, i64 %64, !dbg !22 + %67 = getelementptr bfloat, ptr addrspace(1) %58, i64 %64, !dbg !22 + %68 = getelementptr bfloat, ptr addrspace(1) %59, i64 %64, !dbg !22 + %69 = getelementptr bfloat, ptr addrspace(1) %60, i64 %64, !dbg !22 + %70 = getelementptr bfloat, ptr addrspace(1) %61, i64 %64, !dbg !22 + %71 = getelementptr bfloat, ptr addrspace(1) %62, i64 %64, !dbg !22 + %72 = getelementptr bfloat, ptr addrspace(1) %63, i64 %64, !dbg !22 + %73 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %65, i1 %38) #6, !dbg !23 + %74 = extractvalue { i32, i32, i32, i32 } %73, 0, !dbg !23 + %75 = bitcast i32 %74 to <2 x bfloat>, !dbg !23 + %76 = extractvalue { i32, i32, i32, i32 } %73, 1, !dbg !23 + %77 = bitcast i32 %76 to <2 x bfloat>, !dbg !23 + %78 = extractvalue { i32, i32, i32, i32 } %73, 2, !dbg !23 + %79 = bitcast i32 %78 to <2 x bfloat>, !dbg !23 + %80 = extractvalue { i32, i32, i32, i32 } %73, 3, !dbg !23 + %81 = bitcast i32 %80 to <2 x bfloat>, !dbg !23 + %82 = extractelement <2 x bfloat> %75, i64 0, !dbg !23 + %83 = extractelement <2 x bfloat> %75, i64 1, !dbg !23 + %84 = extractelement <2 x bfloat> %77, i64 0, !dbg !23 + %85 = extractelement <2 x bfloat> %77, i64 1, !dbg !23 + %86 = extractelement <2 x bfloat> %79, i64 0, !dbg !23 + %87 = extractelement <2 x bfloat> %79, i64 1, !dbg !23 + %88 = extractelement <2 x bfloat> %81, i64 0, !dbg !23 + %89 = extractelement <2 x bfloat> %81, i64 1, !dbg !23 + %90 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %66, i1 %39) #6, !dbg !23 + %91 = extractvalue { i32, i32, i32, i32 } %90, 0, !dbg !23 + %92 = bitcast i32 %91 to <2 x bfloat>, !dbg !23 + %93 = extractvalue { i32, i32, i32, i32 } %90, 1, !dbg !23 + %94 = bitcast i32 %93 to <2 x bfloat>, !dbg !23 + %95 = extractvalue { i32, i32, i32, i32 } %90, 2, !dbg !23 + %96 = bitcast i32 %95 to <2 x bfloat>, !dbg !23 + %97 = extractvalue { i32, i32, i32, i32 } %90, 3, !dbg !23 + %98 = bitcast i32 %97 to <2 x bfloat>, !dbg !23 + %99 = extractelement <2 x bfloat> %92, i64 0, !dbg !23 + %100 = extractelement <2 x bfloat> %92, i64 1, !dbg !23 + %101 = extractelement <2 x bfloat> %94, i64 0, !dbg !23 + %102 = extractelement <2 x bfloat> %94, i64 1, !dbg !23 + %103 = extractelement <2 x bfloat> %96, i64 0, !dbg !23 + %104 = extractelement <2 x bfloat> %96, i64 1, !dbg !23 + %105 = extractelement <2 x bfloat> %98, i64 0, !dbg !23 + %106 = extractelement <2 x bfloat> %98, i64 1, !dbg !23 + %107 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %67, i1 %40) #6, !dbg !23 + %108 = extractvalue { i32, i32, i32, i32 } %107, 0, !dbg !23 + %109 = bitcast i32 %108 to <2 x bfloat>, !dbg !23 + %110 = extractvalue { i32, i32, i32, i32 } %107, 1, !dbg !23 + %111 = bitcast i32 %110 to <2 x bfloat>, !dbg !23 + %112 = extractvalue { i32, i32, i32, i32 } %107, 2, !dbg !23 + %113 = bitcast i32 %112 to <2 x bfloat>, !dbg !23 + %114 = extractvalue { i32, i32, i32, i32 } %107, 3, !dbg !23 + %115 = bitcast i32 %114 to <2 x bfloat>, !dbg !23 + %116 = extractelement <2 x bfloat> %109, i64 0, !dbg !23 + %117 = extractelement <2 x bfloat> %109, i64 1, !dbg !23 + %118 = extractelement <2 x bfloat> %111, i64 0, !dbg !23 + %119 = extractelement <2 x bfloat> %111, i64 1, !dbg !23 + %120 = extractelement <2 x bfloat> %113, i64 0, !dbg !23 + %121 = extractelement <2 x bfloat> %113, i64 1, !dbg !23 + %122 = extractelement <2 x bfloat> %115, i64 0, !dbg !23 + %123 = extractelement <2 x bfloat> %115, i64 1, !dbg !23 + %124 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %68, i1 %41) #6, !dbg !23 + %125 = extractvalue { i32, i32, i32, i32 } %124, 0, !dbg !23 + %126 = bitcast i32 %125 to <2 x bfloat>, !dbg !23 + %127 = extractvalue { i32, i32, i32, i32 } %124, 1, !dbg !23 + %128 = bitcast i32 %127 to <2 x bfloat>, !dbg !23 + %129 = extractvalue { i32, i32, i32, i32 } %124, 2, !dbg !23 + %130 = bitcast i32 %129 to <2 x bfloat>, !dbg !23 + %131 = extractvalue { i32, i32, i32, i32 } %124, 3, !dbg !23 + %132 = bitcast i32 %131 to <2 x bfloat>, !dbg !23 + %133 = extractelement <2 x bfloat> %126, i64 0, !dbg !23 + %134 = extractelement <2 x bfloat> %126, i64 1, !dbg !23 + %135 = extractelement <2 x bfloat> %128, i64 0, !dbg !23 + %136 = extractelement <2 x bfloat> %128, i64 1, !dbg !23 + %137 = extractelement <2 x bfloat> %130, i64 0, !dbg !23 + %138 = extractelement <2 x bfloat> %130, i64 1, !dbg !23 + %139 = extractelement <2 x bfloat> %132, i64 0, !dbg !23 + %140 = extractelement <2 x bfloat> %132, i64 1, !dbg !23 + %141 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %69, i1 %42) #6, !dbg !23 + %142 = extractvalue { i32, i32, i32, i32 } %141, 0, !dbg !23 + %143 = bitcast i32 %142 to <2 x bfloat>, !dbg !23 + %144 = extractvalue { i32, i32, i32, i32 } %141, 1, !dbg !23 + %145 = bitcast i32 %144 to <2 x bfloat>, !dbg !23 + %146 = extractvalue { i32, i32, i32, i32 } %141, 2, !dbg !23 + %147 = bitcast i32 %146 to <2 x bfloat>, !dbg !23 + %148 = extractvalue { i32, i32, i32, i32 } %141, 3, !dbg !23 + %149 = bitcast i32 %148 to <2 x bfloat>, !dbg !23 + %150 = extractelement <2 x bfloat> %143, i64 0, !dbg !23 + %151 = extractelement <2 x bfloat> %143, i64 1, !dbg !23 + %152 = extractelement <2 x bfloat> %145, i64 0, !dbg !23 + %153 = extractelement <2 x bfloat> %145, i64 1, !dbg !23 + %154 = extractelement <2 x bfloat> %147, i64 0, !dbg !23 + %155 = extractelement <2 x bfloat> %147, i64 1, !dbg !23 + %156 = extractelement <2 x bfloat> %149, i64 0, !dbg !23 + %157 = extractelement <2 x bfloat> %149, i64 1, !dbg !23 + %158 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %70, i1 %43) #6, !dbg !23 + %159 = extractvalue { i32, i32, i32, i32 } %158, 0, !dbg !23 + %160 = bitcast i32 %159 to <2 x bfloat>, !dbg !23 + %161 = extractvalue { i32, i32, i32, i32 } %158, 1, !dbg !23 + %162 = bitcast i32 %161 to <2 x bfloat>, !dbg !23 + %163 = extractvalue { i32, i32, i32, i32 } %158, 2, !dbg !23 + %164 = bitcast i32 %163 to <2 x bfloat>, !dbg !23 + %165 = extractvalue { i32, i32, i32, i32 } %158, 3, !dbg !23 + %166 = bitcast i32 %165 to <2 x bfloat>, !dbg !23 + %167 = extractelement <2 x bfloat> %160, i64 0, !dbg !23 + %168 = extractelement <2 x bfloat> %160, i64 1, !dbg !23 + %169 = extractelement <2 x bfloat> %162, i64 0, !dbg !23 + %170 = extractelement <2 x bfloat> %162, i64 1, !dbg !23 + %171 = extractelement <2 x bfloat> %164, i64 0, !dbg !23 + %172 = extractelement <2 x bfloat> %164, i64 1, !dbg !23 + %173 = extractelement <2 x bfloat> %166, i64 0, !dbg !23 + %174 = extractelement <2 x bfloat> %166, i64 1, !dbg !23 + %175 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %71, i1 %44) #6, !dbg !23 + %176 = extractvalue { i32, i32, i32, i32 } %175, 0, !dbg !23 + %177 = bitcast i32 %176 to <2 x bfloat>, !dbg !23 + %178 = extractvalue { i32, i32, i32, i32 } %175, 1, !dbg !23 + %179 = bitcast i32 %178 to <2 x bfloat>, !dbg !23 + %180 = extractvalue { i32, i32, i32, i32 } %175, 2, !dbg !23 + %181 = bitcast i32 %180 to <2 x bfloat>, !dbg !23 + %182 = extractvalue { i32, i32, i32, i32 } %175, 3, !dbg !23 + %183 = bitcast i32 %182 to <2 x bfloat>, !dbg !23 + %184 = extractelement <2 x bfloat> %177, i64 0, !dbg !23 + %185 = extractelement <2 x bfloat> %177, i64 1, !dbg !23 + %186 = extractelement <2 x bfloat> %179, i64 0, !dbg !23 + %187 = extractelement <2 x bfloat> %179, i64 1, !dbg !23 + %188 = extractelement <2 x bfloat> %181, i64 0, !dbg !23 + %189 = extractelement <2 x bfloat> %181, i64 1, !dbg !23 + %190 = extractelement <2 x bfloat> %183, i64 0, !dbg !23 + %191 = extractelement <2 x bfloat> %183, i64 1, !dbg !23 + %192 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %72, i1 %45) #6, !dbg !23 + %193 = extractvalue { i32, i32, i32, i32 } %192, 0, !dbg !23 + %194 = bitcast i32 %193 to <2 x bfloat>, !dbg !23 + %195 = extractvalue { i32, i32, i32, i32 } %192, 1, !dbg !23 + %196 = bitcast i32 %195 to <2 x bfloat>, !dbg !23 + %197 = extractvalue { i32, i32, i32, i32 } %192, 2, !dbg !23 + %198 = bitcast i32 %197 to <2 x bfloat>, !dbg !23 + %199 = extractvalue { i32, i32, i32, i32 } %192, 3, !dbg !23 + %200 = bitcast i32 %199 to <2 x bfloat>, !dbg !23 + %201 = extractelement <2 x bfloat> %194, i64 0, !dbg !23 + %202 = extractelement <2 x bfloat> %194, i64 1, !dbg !23 + %203 = extractelement <2 x bfloat> %196, i64 0, !dbg !23 + %204 = extractelement <2 x bfloat> %196, i64 1, !dbg !23 + %205 = extractelement <2 x bfloat> %198, i64 0, !dbg !23 + %206 = extractelement <2 x bfloat> %198, i64 1, !dbg !23 + %207 = extractelement <2 x bfloat> %200, i64 0, !dbg !23 + %208 = extractelement <2 x bfloat> %200, i64 1, !dbg !23 + %209 = fpext bfloat %82 to float, !dbg !24 + %210 = fpext bfloat %83 to float, !dbg !24 + %211 = fpext bfloat %84 to float, !dbg !24 + %212 = fpext bfloat %85 to float, !dbg !24 + %213 = fpext bfloat %86 to float, !dbg !24 + %214 = fpext bfloat %87 to float, !dbg !24 + %215 = fpext bfloat %88 to float, !dbg !24 + %216 = fpext bfloat %89 to float, !dbg !24 + %217 = fpext bfloat %99 to float, !dbg !24 + %218 = fpext bfloat %100 to float, !dbg !24 + %219 = fpext bfloat %101 to float, !dbg !24 + %220 = fpext bfloat %102 to float, !dbg !24 + %221 = fpext bfloat %103 to float, !dbg !24 + %222 = fpext bfloat %104 to float, !dbg !24 + %223 = fpext bfloat %105 to float, !dbg !24 + %224 = fpext bfloat %106 to float, !dbg !24 + %225 = fpext bfloat %116 to float, !dbg !24 + %226 = fpext bfloat %117 to float, !dbg !24 + %227 = fpext bfloat %118 to float, !dbg !24 + %228 = fpext bfloat %119 to float, !dbg !24 + %229 = fpext bfloat %120 to float, !dbg !24 + %230 = fpext bfloat %121 to float, !dbg !24 + %231 = fpext bfloat %122 to float, !dbg !24 + %232 = fpext bfloat %123 to float, !dbg !24 + %233 = fpext bfloat %133 to float, !dbg !24 + %234 = fpext bfloat %134 to float, !dbg !24 + %235 = fpext bfloat %135 to float, !dbg !24 + %236 = fpext bfloat %136 to float, !dbg !24 + %237 = fpext bfloat %137 to float, !dbg !24 + %238 = fpext bfloat %138 to float, !dbg !24 + %239 = fpext bfloat %139 to float, !dbg !24 + %240 = fpext bfloat %140 to float, !dbg !24 + %241 = fpext bfloat %150 to float, !dbg !24 + %242 = fpext bfloat %151 to float, !dbg !24 + %243 = fpext bfloat %152 to float, !dbg !24 + %244 = fpext bfloat %153 to float, !dbg !24 + %245 = fpext bfloat %154 to float, !dbg !24 + %246 = fpext bfloat %155 to float, !dbg !24 + %247 = fpext bfloat %156 to float, !dbg !24 + %248 = fpext bfloat %157 to float, !dbg !24 + %249 = fpext bfloat %167 to float, !dbg !24 + %250 = fpext bfloat %168 to float, !dbg !24 + %251 = fpext bfloat %169 to float, !dbg !24 + %252 = fpext bfloat %170 to float, !dbg !24 + %253 = fpext bfloat %171 to float, !dbg !24 + %254 = fpext bfloat %172 to float, !dbg !24 + %255 = fpext bfloat %173 to float, !dbg !24 + %256 = fpext bfloat %174 to float, !dbg !24 + %257 = fpext bfloat %184 to float, !dbg !24 + %258 = fpext bfloat %185 to float, !dbg !24 + %259 = fpext bfloat %186 to float, !dbg !24 + %260 = fpext bfloat %187 to float, !dbg !24 + %261 = fpext bfloat %188 to float, !dbg !24 + %262 = fpext bfloat %189 to float, !dbg !24 + %263 = fpext bfloat %190 to float, !dbg !24 + %264 = fpext bfloat %191 to float, !dbg !24 + %265 = fpext bfloat %201 to float, !dbg !24 + %266 = fpext bfloat %202 to float, !dbg !24 + %267 = fpext bfloat %203 to float, !dbg !24 + %268 = fpext bfloat %204 to float, !dbg !24 + %269 = fpext bfloat %205 to float, !dbg !24 + %270 = fpext bfloat %206 to float, !dbg !24 + %271 = fpext bfloat %207 to float, !dbg !24 + %272 = fpext bfloat %208 to float, !dbg !24 + %273 = fmul float %209, %209, !dbg !25 + %274 = fmul float %210, %210, !dbg !25 + %275 = fmul float %211, %211, !dbg !25 + %276 = fmul float %212, %212, !dbg !25 + %277 = fmul float %213, %213, !dbg !25 + %278 = fmul float %214, %214, !dbg !25 + %279 = fmul float %215, %215, !dbg !25 + %280 = fmul float %216, %216, !dbg !25 + %281 = fmul float %217, %217, !dbg !25 + %282 = fmul float %218, %218, !dbg !25 + %283 = fmul float %219, %219, !dbg !25 + %284 = fmul float %220, %220, !dbg !25 + %285 = fmul float %221, %221, !dbg !25 + %286 = fmul float %222, %222, !dbg !25 + %287 = fmul float %223, %223, !dbg !25 + %288 = fmul float %224, %224, !dbg !25 + %289 = fmul float %225, %225, !dbg !25 + %290 = fmul float %226, %226, !dbg !25 + %291 = fmul float %227, %227, !dbg !25 + %292 = fmul float %228, %228, !dbg !25 + %293 = fmul float %229, %229, !dbg !25 + %294 = fmul float %230, %230, !dbg !25 + %295 = fmul float %231, %231, !dbg !25 + %296 = fmul float %232, %232, !dbg !25 + %297 = fmul float %233, %233, !dbg !25 + %298 = fmul float %234, %234, !dbg !25 + %299 = fmul float %235, %235, !dbg !25 + %300 = fmul float %236, %236, !dbg !25 + %301 = fmul float %237, %237, !dbg !25 + %302 = fmul float %238, %238, !dbg !25 + %303 = fmul float %239, %239, !dbg !25 + %304 = fmul float %240, %240, !dbg !25 + %305 = fmul float %241, %241, !dbg !25 + %306 = fmul float %242, %242, !dbg !25 + %307 = fmul float %243, %243, !dbg !25 + %308 = fmul float %244, %244, !dbg !25 + %309 = fmul float %245, %245, !dbg !25 + %310 = fmul float %246, %246, !dbg !25 + %311 = fmul float %247, %247, !dbg !25 + %312 = fmul float %248, %248, !dbg !25 + %313 = fmul float %249, %249, !dbg !25 + %314 = fmul float %250, %250, !dbg !25 + %315 = fmul float %251, %251, !dbg !25 + %316 = fmul float %252, %252, !dbg !25 + %317 = fmul float %253, %253, !dbg !25 + %318 = fmul float %254, %254, !dbg !25 + %319 = fmul float %255, %255, !dbg !25 + %320 = fmul float %256, %256, !dbg !25 + %321 = fmul float %257, %257, !dbg !25 + %322 = fmul float %258, %258, !dbg !25 + %323 = fmul float %259, %259, !dbg !25 + %324 = fmul float %260, %260, !dbg !25 + %325 = fmul float %261, %261, !dbg !25 + %326 = fmul float %262, %262, !dbg !25 + %327 = fmul float %263, %263, !dbg !25 + %328 = fmul float %264, %264, !dbg !25 + %329 = fmul float %265, %265, !dbg !25 + %330 = fmul float %266, %266, !dbg !25 + %331 = fmul float %267, %267, !dbg !25 + %332 = fmul float %268, %268, !dbg !25 + %333 = fmul float %269, %269, !dbg !25 + %334 = fmul float %270, %270, !dbg !25 + %335 = fmul float %271, %271, !dbg !25 + %336 = fmul float %272, %272, !dbg !25 + %337 = fadd float %273, %274, !dbg !26 + %338 = fadd float %275, %337, !dbg !26 + %339 = fadd float %276, %338, !dbg !26 + %340 = fadd float %277, %339, !dbg !26 + %341 = fadd float %278, %340, !dbg !26 + %342 = fadd float %279, %341, !dbg !26 + %343 = fadd float %280, %342, !dbg !26 + %344 = fadd float %281, %282, !dbg !26 + %345 = fadd float %283, %344, !dbg !26 + %346 = fadd float %284, %345, !dbg !26 + %347 = fadd float %285, %346, !dbg !26 + %348 = fadd float %286, %347, !dbg !26 + %349 = fadd float %287, %348, !dbg !26 + %350 = fadd float %288, %349, !dbg !26 + %351 = fadd float %289, %290, !dbg !26 + %352 = fadd float %291, %351, !dbg !26 + %353 = fadd float %292, %352, !dbg !26 + %354 = fadd float %293, %353, !dbg !26 + %355 = fadd float %294, %354, !dbg !26 + %356 = fadd float %295, %355, !dbg !26 + %357 = fadd float %296, %356, !dbg !26 + %358 = fadd float %297, %298, !dbg !26 + %359 = fadd float %299, %358, !dbg !26 + %360 = fadd float %300, %359, !dbg !26 + %361 = fadd float %301, %360, !dbg !26 + %362 = fadd float %302, %361, !dbg !26 + %363 = fadd float %303, %362, !dbg !26 + %364 = fadd float %304, %363, !dbg !26 + %365 = fadd float %305, %306, !dbg !26 + %366 = fadd float %307, %365, !dbg !26 + %367 = fadd float %308, %366, !dbg !26 + %368 = fadd float %309, %367, !dbg !26 + %369 = fadd float %310, %368, !dbg !26 + %370 = fadd float %311, %369, !dbg !26 + %371 = fadd float %312, %370, !dbg !26 + %372 = fadd float %313, %314, !dbg !26 + %373 = fadd float %315, %372, !dbg !26 + %374 = fadd float %316, %373, !dbg !26 + %375 = fadd float %317, %374, !dbg !26 + %376 = fadd float %318, %375, !dbg !26 + %377 = fadd float %319, %376, !dbg !26 + %378 = fadd float %320, %377, !dbg !26 + %379 = fadd float %321, %322, !dbg !26 + %380 = fadd float %323, %379, !dbg !26 + %381 = fadd float %324, %380, !dbg !26 + %382 = fadd float %325, %381, !dbg !26 + %383 = fadd float %326, %382, !dbg !26 + %384 = fadd float %327, %383, !dbg !26 + %385 = fadd float %328, %384, !dbg !26 + %386 = fadd float %329, %330, !dbg !26 + %387 = fadd float %331, %386, !dbg !26 + %388 = fadd float %332, %387, !dbg !26 + %389 = fadd float %333, %388, !dbg !26 + %390 = fadd float %334, %389, !dbg !26 + %391 = fadd float %335, %390, !dbg !26 + %392 = fadd float %336, %391, !dbg !26 + %393 = bitcast float %343 to i32, !dbg !29 + %394 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %393, i32 8, i32 31), !dbg !29 + %395 = bitcast i32 %394 to float, !dbg !29 + %396 = fadd float %343, %395, !dbg !26 + %397 = bitcast float %396 to i32, !dbg !29 + %398 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %397, i32 4, i32 31), !dbg !29 + %399 = bitcast i32 %398 to float, !dbg !29 + %400 = fadd float %396, %399, !dbg !26 + %401 = bitcast float %400 to i32, !dbg !29 + %402 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %401, i32 2, i32 31), !dbg !29 + %403 = bitcast i32 %402 to float, !dbg !29 + %404 = fadd float %400, %403, !dbg !26 + %405 = bitcast float %404 to i32, !dbg !29 + %406 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %405, i32 1, i32 31), !dbg !29 + %407 = bitcast i32 %406 to float, !dbg !29 + %408 = fadd float %404, %407, !dbg !26 + %409 = bitcast float %350 to i32, !dbg !29 + %410 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %409, i32 8, i32 31), !dbg !29 + %411 = bitcast i32 %410 to float, !dbg !29 + %412 = fadd float %350, %411, !dbg !26 + %413 = bitcast float %412 to i32, !dbg !29 + %414 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %413, i32 4, i32 31), !dbg !29 + %415 = bitcast i32 %414 to float, !dbg !29 + %416 = fadd float %412, %415, !dbg !26 + %417 = bitcast float %416 to i32, !dbg !29 + %418 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %417, i32 2, i32 31), !dbg !29 + %419 = bitcast i32 %418 to float, !dbg !29 + %420 = fadd float %416, %419, !dbg !26 + %421 = bitcast float %420 to i32, !dbg !29 + %422 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %421, i32 1, i32 31), !dbg !29 + %423 = bitcast i32 %422 to float, !dbg !29 + %424 = fadd float %420, %423, !dbg !26 + %425 = bitcast float %357 to i32, !dbg !29 + %426 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %425, i32 8, i32 31), !dbg !29 + %427 = bitcast i32 %426 to float, !dbg !29 + %428 = fadd float %357, %427, !dbg !26 + %429 = bitcast float %428 to i32, !dbg !29 + %430 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %429, i32 4, i32 31), !dbg !29 + %431 = bitcast i32 %430 to float, !dbg !29 + %432 = fadd float %428, %431, !dbg !26 + %433 = bitcast float %432 to i32, !dbg !29 + %434 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %433, i32 2, i32 31), !dbg !29 + %435 = bitcast i32 %434 to float, !dbg !29 + %436 = fadd float %432, %435, !dbg !26 + %437 = bitcast float %436 to i32, !dbg !29 + %438 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %437, i32 1, i32 31), !dbg !29 + %439 = bitcast i32 %438 to float, !dbg !29 + %440 = fadd float %436, %439, !dbg !26 + %441 = bitcast float %364 to i32, !dbg !29 + %442 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %441, i32 8, i32 31), !dbg !29 + %443 = bitcast i32 %442 to float, !dbg !29 + %444 = fadd float %364, %443, !dbg !26 + %445 = bitcast float %444 to i32, !dbg !29 + %446 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %445, i32 4, i32 31), !dbg !29 + %447 = bitcast i32 %446 to float, !dbg !29 + %448 = fadd float %444, %447, !dbg !26 + %449 = bitcast float %448 to i32, !dbg !29 + %450 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %449, i32 2, i32 31), !dbg !29 + %451 = bitcast i32 %450 to float, !dbg !29 + %452 = fadd float %448, %451, !dbg !26 + %453 = bitcast float %452 to i32, !dbg !29 + %454 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %453, i32 1, i32 31), !dbg !29 + %455 = bitcast i32 %454 to float, !dbg !29 + %456 = fadd float %452, %455, !dbg !26 + %457 = bitcast float %371 to i32, !dbg !29 + %458 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %457, i32 8, i32 31), !dbg !29 + %459 = bitcast i32 %458 to float, !dbg !29 + %460 = fadd float %371, %459, !dbg !26 + %461 = bitcast float %460 to i32, !dbg !29 + %462 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %461, i32 4, i32 31), !dbg !29 + %463 = bitcast i32 %462 to float, !dbg !29 + %464 = fadd float %460, %463, !dbg !26 + %465 = bitcast float %464 to i32, !dbg !29 + %466 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %465, i32 2, i32 31), !dbg !29 + %467 = bitcast i32 %466 to float, !dbg !29 + %468 = fadd float %464, %467, !dbg !26 + %469 = bitcast float %468 to i32, !dbg !29 + %470 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %469, i32 1, i32 31), !dbg !29 + %471 = bitcast i32 %470 to float, !dbg !29 + %472 = fadd float %468, %471, !dbg !26 + %473 = bitcast float %378 to i32, !dbg !29 + %474 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %473, i32 8, i32 31), !dbg !29 + %475 = bitcast i32 %474 to float, !dbg !29 + %476 = fadd float %378, %475, !dbg !26 + %477 = bitcast float %476 to i32, !dbg !29 + %478 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %477, i32 4, i32 31), !dbg !29 + %479 = bitcast i32 %478 to float, !dbg !29 + %480 = fadd float %476, %479, !dbg !26 + %481 = bitcast float %480 to i32, !dbg !29 + %482 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %481, i32 2, i32 31), !dbg !29 + %483 = bitcast i32 %482 to float, !dbg !29 + %484 = fadd float %480, %483, !dbg !26 + %485 = bitcast float %484 to i32, !dbg !29 + %486 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %485, i32 1, i32 31), !dbg !29 + %487 = bitcast i32 %486 to float, !dbg !29 + %488 = fadd float %484, %487, !dbg !26 + %489 = bitcast float %385 to i32, !dbg !29 + %490 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %489, i32 8, i32 31), !dbg !29 + %491 = bitcast i32 %490 to float, !dbg !29 + %492 = fadd float %385, %491, !dbg !26 + %493 = bitcast float %492 to i32, !dbg !29 + %494 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %493, i32 4, i32 31), !dbg !29 + %495 = bitcast i32 %494 to float, !dbg !29 + %496 = fadd float %492, %495, !dbg !26 + %497 = bitcast float %496 to i32, !dbg !29 + %498 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %497, i32 2, i32 31), !dbg !29 + %499 = bitcast i32 %498 to float, !dbg !29 + %500 = fadd float %496, %499, !dbg !26 + %501 = bitcast float %500 to i32, !dbg !29 + %502 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %501, i32 1, i32 31), !dbg !29 + %503 = bitcast i32 %502 to float, !dbg !29 + %504 = fadd float %500, %503, !dbg !26 + %505 = bitcast float %392 to i32, !dbg !29 + %506 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %505, i32 8, i32 31), !dbg !29 + %507 = bitcast i32 %506 to float, !dbg !29 + %508 = fadd float %392, %507, !dbg !26 + %509 = bitcast float %508 to i32, !dbg !29 + %510 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %509, i32 4, i32 31), !dbg !29 + %511 = bitcast i32 %510 to float, !dbg !29 + %512 = fadd float %508, %511, !dbg !26 + %513 = bitcast float %512 to i32, !dbg !29 + %514 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %513, i32 2, i32 31), !dbg !29 + %515 = bitcast i32 %514 to float, !dbg !29 + %516 = fadd float %512, %515, !dbg !26 + %517 = bitcast float %516 to i32, !dbg !29 + %518 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %517, i32 1, i32 31), !dbg !29 + %519 = bitcast i32 %518 to float, !dbg !29 + %520 = fadd float %516, %519, !dbg !26 + %521 = fadd float %3, %408, !dbg !32 + %522 = fadd float %3, %424, !dbg !32 + %523 = fadd float %3, %440, !dbg !32 + %524 = fadd float %3, %456, !dbg !32 + %525 = fadd float %3, %472, !dbg !32 + %526 = fadd float %3, %488, !dbg !32 + %527 = fadd float %3, %504, !dbg !32 + %528 = fadd float %3, %520, !dbg !32 + %529 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i = icmp eq i32 %529, 0, !dbg !33 + %530 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i = icmp eq i32 %530, 0, !dbg !33 + br i1 %.not.i, label %536, label %531, !dbg !33 + +531: ; preds = %7 + br i1 %.not1.i, label %534, label %532, !dbg !33 + +532: ; preds = %531 + %533 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %521) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +534: ; preds = %531 + %535 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %521) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +536: ; preds = %7 + br i1 %.not1.i, label %539, label %537, !dbg !33 + +537: ; preds = %536 + %538 = tail call float @llvm.nvvm.sqrt.rn.f(float %521) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +539: ; preds = %536 + %540 = tail call float @llvm.nvvm.sqrt.approx.f(float %521) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +__nv_sqrtf.exit: ; preds = %532, %534, %537, %539 + %.0.i = phi float [ %533, %532 ], [ %535, %534 ], [ %538, %537 ], [ %540, %539 ], !dbg !33 + %541 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i1 = icmp eq i32 %541, 0, !dbg !33 + %542 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i4 = icmp eq i32 %542, 0, !dbg !33 + br i1 %.not.i1, label %548, label %543, !dbg !33 + +543: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %546, label %544, !dbg !33 + +544: ; preds = %543 + %545 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %522) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +546: ; preds = %543 + %547 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %522) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +548: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %551, label %549, !dbg !33 + +549: ; preds = %548 + %550 = tail call float @llvm.nvvm.sqrt.rn.f(float %522) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +551: ; preds = %548 + %552 = tail call float @llvm.nvvm.sqrt.approx.f(float %522) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +__nv_sqrtf.exit5: ; preds = %544, %546, %549, %551 + %.0.i3 = phi float [ %545, %544 ], [ %547, %546 ], [ %550, %549 ], [ %552, %551 ], !dbg !33 + %553 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i6 = icmp eq i32 %553, 0, !dbg !33 + %554 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i9 = icmp eq i32 %554, 0, !dbg !33 + br i1 %.not.i6, label %560, label %555, !dbg !33 + +555: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %558, label %556, !dbg !33 + +556: ; preds = %555 + %557 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %523) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +558: ; preds = %555 + %559 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %523) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +560: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %563, label %561, !dbg !33 + +561: ; preds = %560 + %562 = tail call float @llvm.nvvm.sqrt.rn.f(float %523) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +563: ; preds = %560 + %564 = tail call float @llvm.nvvm.sqrt.approx.f(float %523) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +__nv_sqrtf.exit10: ; preds = %556, %558, %561, %563 + %.0.i8 = phi float [ %557, %556 ], [ %559, %558 ], [ %562, %561 ], [ %564, %563 ], !dbg !33 + %565 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i11 = icmp eq i32 %565, 0, !dbg !33 + %566 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i14 = icmp eq i32 %566, 0, !dbg !33 + br i1 %.not.i11, label %572, label %567, !dbg !33 + +567: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %570, label %568, !dbg !33 + +568: ; preds = %567 + %569 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %524) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +570: ; preds = %567 + %571 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %524) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +572: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %575, label %573, !dbg !33 + +573: ; preds = %572 + %574 = tail call float @llvm.nvvm.sqrt.rn.f(float %524) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +575: ; preds = %572 + %576 = tail call float @llvm.nvvm.sqrt.approx.f(float %524) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +__nv_sqrtf.exit15: ; preds = %568, %570, %573, %575 + %.0.i13 = phi float [ %569, %568 ], [ %571, %570 ], [ %574, %573 ], [ %576, %575 ], !dbg !33 + %577 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i16 = icmp eq i32 %577, 0, !dbg !33 + %578 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i19 = icmp eq i32 %578, 0, !dbg !33 + br i1 %.not.i16, label %584, label %579, !dbg !33 + +579: ; preds = %__nv_sqrtf.exit15 + br i1 %.not1.i19, label %582, label %580, !dbg !33 + +580: ; preds = %579 + %581 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %525) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +582: ; preds = %579 + %583 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %525) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +584: ; preds = %__nv_sqrtf.exit15 + br i1 %.not1.i19, label %587, label %585, !dbg !33 + +585: ; preds = %584 + %586 = tail call float @llvm.nvvm.sqrt.rn.f(float %525) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +587: ; preds = %584 + %588 = tail call float @llvm.nvvm.sqrt.approx.f(float %525) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +__nv_sqrtf.exit20: ; preds = %580, %582, %585, %587 + %.0.i18 = phi float [ %581, %580 ], [ %583, %582 ], [ %586, %585 ], [ %588, %587 ], !dbg !33 + %589 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i21 = icmp eq i32 %589, 0, !dbg !33 + %590 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i24 = icmp eq i32 %590, 0, !dbg !33 + br i1 %.not.i21, label %596, label %591, !dbg !33 + +591: ; preds = %__nv_sqrtf.exit20 + br i1 %.not1.i24, label %594, label %592, !dbg !33 + +592: ; preds = %591 + %593 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %526) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +594: ; preds = %591 + %595 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %526) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +596: ; preds = %__nv_sqrtf.exit20 + br i1 %.not1.i24, label %599, label %597, !dbg !33 + +597: ; preds = %596 + %598 = tail call float @llvm.nvvm.sqrt.rn.f(float %526) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +599: ; preds = %596 + %600 = tail call float @llvm.nvvm.sqrt.approx.f(float %526) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +__nv_sqrtf.exit25: ; preds = %592, %594, %597, %599 + %.0.i23 = phi float [ %593, %592 ], [ %595, %594 ], [ %598, %597 ], [ %600, %599 ], !dbg !33 + %601 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i26 = icmp eq i32 %601, 0, !dbg !33 + %602 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i29 = icmp eq i32 %602, 0, !dbg !33 + br i1 %.not.i26, label %608, label %603, !dbg !33 + +603: ; preds = %__nv_sqrtf.exit25 + br i1 %.not1.i29, label %606, label %604, !dbg !33 + +604: ; preds = %603 + %605 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %527) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +606: ; preds = %603 + %607 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %527) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +608: ; preds = %__nv_sqrtf.exit25 + br i1 %.not1.i29, label %611, label %609, !dbg !33 + +609: ; preds = %608 + %610 = tail call float @llvm.nvvm.sqrt.rn.f(float %527) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +611: ; preds = %608 + %612 = tail call float @llvm.nvvm.sqrt.approx.f(float %527) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +__nv_sqrtf.exit30: ; preds = %604, %606, %609, %611 + %.0.i28 = phi float [ %605, %604 ], [ %607, %606 ], [ %610, %609 ], [ %612, %611 ], !dbg !33 + %613 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i31 = icmp eq i32 %613, 0, !dbg !33 + %614 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i34 = icmp eq i32 %614, 0, !dbg !33 + br i1 %.not.i31, label %620, label %615, !dbg !33 + +615: ; preds = %__nv_sqrtf.exit30 + br i1 %.not1.i34, label %618, label %616, !dbg !33 + +616: ; preds = %615 + %617 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %528) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +618: ; preds = %615 + %619 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %528) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +620: ; preds = %__nv_sqrtf.exit30 + br i1 %.not1.i34, label %623, label %621, !dbg !33 + +621: ; preds = %620 + %622 = tail call float @llvm.nvvm.sqrt.rn.f(float %528) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +623: ; preds = %620 + %624 = tail call float @llvm.nvvm.sqrt.approx.f(float %528) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +__nv_sqrtf.exit35: ; preds = %616, %618, %621, %623 + %.0.i33 = phi float [ %617, %616 ], [ %619, %618 ], [ %622, %621 ], [ %624, %623 ], !dbg !33 + %625 = and i32 %11, 63, !dbg !16 + %626 = zext nneg i32 %625 to i64, !dbg !17 + %627 = or disjoint i64 %10, %626, !dbg !17 + %628 = getelementptr float, ptr addrspace(1) %2, i64 %627, !dbg !34 + %629 = getelementptr bfloat, ptr addrspace(1) %1, i64 %55, !dbg !35 + %630 = getelementptr bfloat, ptr addrspace(1) %629, i64 %64, !dbg !36 + %631 = getelementptr bfloat, ptr addrspace(1) %1, i64 %54, !dbg !35 + %632 = getelementptr bfloat, ptr addrspace(1) %631, i64 %64, !dbg !36 + %633 = getelementptr bfloat, ptr addrspace(1) %1, i64 %53, !dbg !35 + %634 = getelementptr bfloat, ptr addrspace(1) %633, i64 %64, !dbg !36 + %635 = getelementptr bfloat, ptr addrspace(1) %1, i64 %52, !dbg !35 + %636 = getelementptr bfloat, ptr addrspace(1) %635, i64 %64, !dbg !36 + %637 = getelementptr bfloat, ptr addrspace(1) %1, i64 %51, !dbg !35 + %638 = getelementptr bfloat, ptr addrspace(1) %637, i64 %64, !dbg !36 + %639 = getelementptr bfloat, ptr addrspace(1) %1, i64 %50, !dbg !35 + %640 = getelementptr bfloat, ptr addrspace(1) %639, i64 %64, !dbg !36 + %641 = getelementptr bfloat, ptr addrspace(1) %1, i64 %49, !dbg !35 + %642 = getelementptr bfloat, ptr addrspace(1) %641, i64 %64, !dbg !36 + %643 = getelementptr bfloat, ptr addrspace(1) %1, i64 %48, !dbg !35 + %644 = getelementptr bfloat, ptr addrspace(1) %643, i64 %64, !dbg !36 + %645 = icmp slt i64 %627, %37, !dbg !18 + %646 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i), !dbg !37 + %647 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i3), !dbg !37 + %648 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i8), !dbg !37 + %649 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i13), !dbg !37 + %650 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i18), !dbg !37 + %651 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i23), !dbg !37 + %652 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i28), !dbg !37 + %653 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i33), !dbg !37 + %654 = fmul float %646, %209, !dbg !38 + %655 = fmul float %646, %210, !dbg !38 + %656 = fmul float %646, %211, !dbg !38 + %657 = fmul float %646, %212, !dbg !38 + %658 = fmul float %646, %213, !dbg !38 + %659 = fmul float %646, %214, !dbg !38 + %660 = fmul float %646, %215, !dbg !38 + %661 = fmul float %646, %216, !dbg !38 + %662 = fmul float %647, %217, !dbg !38 + %663 = fmul float %647, %218, !dbg !38 + %664 = fmul float %647, %219, !dbg !38 + %665 = fmul float %647, %220, !dbg !38 + %666 = fmul float %647, %221, !dbg !38 + %667 = fmul float %647, %222, !dbg !38 + %668 = fmul float %647, %223, !dbg !38 + %669 = fmul float %647, %224, !dbg !38 + %670 = fmul float %648, %225, !dbg !38 + %671 = fmul float %648, %226, !dbg !38 + %672 = fmul float %648, %227, !dbg !38 + %673 = fmul float %648, %228, !dbg !38 + %674 = fmul float %648, %229, !dbg !38 + %675 = fmul float %648, %230, !dbg !38 + %676 = fmul float %648, %231, !dbg !38 + %677 = fmul float %648, %232, !dbg !38 + %678 = fmul float %649, %233, !dbg !38 + %679 = fmul float %649, %234, !dbg !38 + %680 = fmul float %649, %235, !dbg !38 + %681 = fmul float %649, %236, !dbg !38 + %682 = fmul float %649, %237, !dbg !38 + %683 = fmul float %649, %238, !dbg !38 + %684 = fmul float %649, %239, !dbg !38 + %685 = fmul float %649, %240, !dbg !38 + %686 = fmul float %650, %241, !dbg !38 + %687 = fmul float %650, %242, !dbg !38 + %688 = fmul float %650, %243, !dbg !38 + %689 = fmul float %650, %244, !dbg !38 + %690 = fmul float %650, %245, !dbg !38 + %691 = fmul float %650, %246, !dbg !38 + %692 = fmul float %650, %247, !dbg !38 + %693 = fmul float %650, %248, !dbg !38 + %694 = fmul float %651, %249, !dbg !38 + %695 = fmul float %651, %250, !dbg !38 + %696 = fmul float %651, %251, !dbg !38 + %697 = fmul float %651, %252, !dbg !38 + %698 = fmul float %651, %253, !dbg !38 + %699 = fmul float %651, %254, !dbg !38 + %700 = fmul float %651, %255, !dbg !38 + %701 = fmul float %651, %256, !dbg !38 + %702 = fmul float %652, %257, !dbg !38 + %703 = fmul float %652, %258, !dbg !38 + %704 = fmul float %652, %259, !dbg !38 + %705 = fmul float %652, %260, !dbg !38 + %706 = fmul float %652, %261, !dbg !38 + %707 = fmul float %652, %262, !dbg !38 + %708 = fmul float %652, %263, !dbg !38 + %709 = fmul float %652, %264, !dbg !38 + %710 = fmul float %653, %265, !dbg !38 + %711 = fmul float %653, %266, !dbg !38 + %712 = fmul float %653, %267, !dbg !38 + %713 = fmul float %653, %268, !dbg !38 + %714 = fmul float %653, %269, !dbg !38 + %715 = fmul float %653, %270, !dbg !38 + %716 = fmul float %653, %271, !dbg !38 + %717 = fmul float %653, %272, !dbg !38 + %718 = fptrunc float %654 to bfloat, !dbg !39 + %719 = fptrunc float %655 to bfloat, !dbg !39 + %720 = fptrunc float %656 to bfloat, !dbg !39 + %721 = fptrunc float %657 to bfloat, !dbg !39 + %722 = fptrunc float %658 to bfloat, !dbg !39 + %723 = fptrunc float %659 to bfloat, !dbg !39 + %724 = fptrunc float %660 to bfloat, !dbg !39 + %725 = fptrunc float %661 to bfloat, !dbg !39 + %726 = fptrunc float %662 to bfloat, !dbg !39 + %727 = fptrunc float %663 to bfloat, !dbg !39 + %728 = fptrunc float %664 to bfloat, !dbg !39 + %729 = fptrunc float %665 to bfloat, !dbg !39 + %730 = fptrunc float %666 to bfloat, !dbg !39 + %731 = fptrunc float %667 to bfloat, !dbg !39 + %732 = fptrunc float %668 to bfloat, !dbg !39 + %733 = fptrunc float %669 to bfloat, !dbg !39 + %734 = fptrunc float %670 to bfloat, !dbg !39 + %735 = fptrunc float %671 to bfloat, !dbg !39 + %736 = fptrunc float %672 to bfloat, !dbg !39 + %737 = fptrunc float %673 to bfloat, !dbg !39 + %738 = fptrunc float %674 to bfloat, !dbg !39 + %739 = fptrunc float %675 to bfloat, !dbg !39 + %740 = fptrunc float %676 to bfloat, !dbg !39 + %741 = fptrunc float %677 to bfloat, !dbg !39 + %742 = fptrunc float %678 to bfloat, !dbg !39 + %743 = fptrunc float %679 to bfloat, !dbg !39 + %744 = fptrunc float %680 to bfloat, !dbg !39 + %745 = fptrunc float %681 to bfloat, !dbg !39 + %746 = fptrunc float %682 to bfloat, !dbg !39 + %747 = fptrunc float %683 to bfloat, !dbg !39 + %748 = fptrunc float %684 to bfloat, !dbg !39 + %749 = fptrunc float %685 to bfloat, !dbg !39 + %750 = fptrunc float %686 to bfloat, !dbg !39 + %751 = fptrunc float %687 to bfloat, !dbg !39 + %752 = fptrunc float %688 to bfloat, !dbg !39 + %753 = fptrunc float %689 to bfloat, !dbg !39 + %754 = fptrunc float %690 to bfloat, !dbg !39 + %755 = fptrunc float %691 to bfloat, !dbg !39 + %756 = fptrunc float %692 to bfloat, !dbg !39 + %757 = fptrunc float %693 to bfloat, !dbg !39 + %758 = fptrunc float %694 to bfloat, !dbg !39 + %759 = fptrunc float %695 to bfloat, !dbg !39 + %760 = fptrunc float %696 to bfloat, !dbg !39 + %761 = fptrunc float %697 to bfloat, !dbg !39 + %762 = fptrunc float %698 to bfloat, !dbg !39 + %763 = fptrunc float %699 to bfloat, !dbg !39 + %764 = fptrunc float %700 to bfloat, !dbg !39 + %765 = fptrunc float %701 to bfloat, !dbg !39 + %766 = fptrunc float %702 to bfloat, !dbg !39 + %767 = fptrunc float %703 to bfloat, !dbg !39 + %768 = fptrunc float %704 to bfloat, !dbg !39 + %769 = fptrunc float %705 to bfloat, !dbg !39 + %770 = fptrunc float %706 to bfloat, !dbg !39 + %771 = fptrunc float %707 to bfloat, !dbg !39 + %772 = fptrunc float %708 to bfloat, !dbg !39 + %773 = fptrunc float %709 to bfloat, !dbg !39 + %774 = fptrunc float %710 to bfloat, !dbg !39 + %775 = fptrunc float %711 to bfloat, !dbg !39 + %776 = fptrunc float %712 to bfloat, !dbg !39 + %777 = fptrunc float %713 to bfloat, !dbg !39 + %778 = fptrunc float %714 to bfloat, !dbg !39 + %779 = fptrunc float %715 to bfloat, !dbg !39 + %780 = fptrunc float %716 to bfloat, !dbg !39 + %781 = fptrunc float %717 to bfloat, !dbg !39 + %782 = insertelement <2 x bfloat> poison, bfloat %718, i64 0, !dbg !40 + %783 = insertelement <2 x bfloat> %782, bfloat %719, i64 1, !dbg !40 + %784 = bitcast <2 x bfloat> %783 to i32, !dbg !40 + %785 = insertelement <2 x bfloat> poison, bfloat %720, i64 0, !dbg !40 + %786 = insertelement <2 x bfloat> %785, bfloat %721, i64 1, !dbg !40 + %787 = bitcast <2 x bfloat> %786 to i32, !dbg !40 + %788 = insertelement <2 x bfloat> poison, bfloat %722, i64 0, !dbg !40 + %789 = insertelement <2 x bfloat> %788, bfloat %723, i64 1, !dbg !40 + %790 = bitcast <2 x bfloat> %789 to i32, !dbg !40 + %791 = insertelement <2 x bfloat> poison, bfloat %724, i64 0, !dbg !40 + %792 = insertelement <2 x bfloat> %791, bfloat %725, i64 1, !dbg !40 + %793 = bitcast <2 x bfloat> %792 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %784, i32 %787, i32 %790, i32 %793, ptr addrspace(1) %644, i1 %38) #6, !dbg !40 + %794 = insertelement <2 x bfloat> poison, bfloat %726, i64 0, !dbg !40 + %795 = insertelement <2 x bfloat> %794, bfloat %727, i64 1, !dbg !40 + %796 = bitcast <2 x bfloat> %795 to i32, !dbg !40 + %797 = insertelement <2 x bfloat> poison, bfloat %728, i64 0, !dbg !40 + %798 = insertelement <2 x bfloat> %797, bfloat %729, i64 1, !dbg !40 + %799 = bitcast <2 x bfloat> %798 to i32, !dbg !40 + %800 = insertelement <2 x bfloat> poison, bfloat %730, i64 0, !dbg !40 + %801 = insertelement <2 x bfloat> %800, bfloat %731, i64 1, !dbg !40 + %802 = bitcast <2 x bfloat> %801 to i32, !dbg !40 + %803 = insertelement <2 x bfloat> poison, bfloat %732, i64 0, !dbg !40 + %804 = insertelement <2 x bfloat> %803, bfloat %733, i64 1, !dbg !40 + %805 = bitcast <2 x bfloat> %804 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %796, i32 %799, i32 %802, i32 %805, ptr addrspace(1) %642, i1 %39) #6, !dbg !40 + %806 = insertelement <2 x bfloat> poison, bfloat %734, i64 0, !dbg !40 + %807 = insertelement <2 x bfloat> %806, bfloat %735, i64 1, !dbg !40 + %808 = bitcast <2 x bfloat> %807 to i32, !dbg !40 + %809 = insertelement <2 x bfloat> poison, bfloat %736, i64 0, !dbg !40 + %810 = insertelement <2 x bfloat> %809, bfloat %737, i64 1, !dbg !40 + %811 = bitcast <2 x bfloat> %810 to i32, !dbg !40 + %812 = insertelement <2 x bfloat> poison, bfloat %738, i64 0, !dbg !40 + %813 = insertelement <2 x bfloat> %812, bfloat %739, i64 1, !dbg !40 + %814 = bitcast <2 x bfloat> %813 to i32, !dbg !40 + %815 = insertelement <2 x bfloat> poison, bfloat %740, i64 0, !dbg !40 + %816 = insertelement <2 x bfloat> %815, bfloat %741, i64 1, !dbg !40 + %817 = bitcast <2 x bfloat> %816 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %808, i32 %811, i32 %814, i32 %817, ptr addrspace(1) %640, i1 %40) #6, !dbg !40 + %818 = insertelement <2 x bfloat> poison, bfloat %742, i64 0, !dbg !40 + %819 = insertelement <2 x bfloat> %818, bfloat %743, i64 1, !dbg !40 + %820 = bitcast <2 x bfloat> %819 to i32, !dbg !40 + %821 = insertelement <2 x bfloat> poison, bfloat %744, i64 0, !dbg !40 + %822 = insertelement <2 x bfloat> %821, bfloat %745, i64 1, !dbg !40 + %823 = bitcast <2 x bfloat> %822 to i32, !dbg !40 + %824 = insertelement <2 x bfloat> poison, bfloat %746, i64 0, !dbg !40 + %825 = insertelement <2 x bfloat> %824, bfloat %747, i64 1, !dbg !40 + %826 = bitcast <2 x bfloat> %825 to i32, !dbg !40 + %827 = insertelement <2 x bfloat> poison, bfloat %748, i64 0, !dbg !40 + %828 = insertelement <2 x bfloat> %827, bfloat %749, i64 1, !dbg !40 + %829 = bitcast <2 x bfloat> %828 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %820, i32 %823, i32 %826, i32 %829, ptr addrspace(1) %638, i1 %41) #6, !dbg !40 + %830 = insertelement <2 x bfloat> poison, bfloat %750, i64 0, !dbg !40 + %831 = insertelement <2 x bfloat> %830, bfloat %751, i64 1, !dbg !40 + %832 = bitcast <2 x bfloat> %831 to i32, !dbg !40 + %833 = insertelement <2 x bfloat> poison, bfloat %752, i64 0, !dbg !40 + %834 = insertelement <2 x bfloat> %833, bfloat %753, i64 1, !dbg !40 + %835 = bitcast <2 x bfloat> %834 to i32, !dbg !40 + %836 = insertelement <2 x bfloat> poison, bfloat %754, i64 0, !dbg !40 + %837 = insertelement <2 x bfloat> %836, bfloat %755, i64 1, !dbg !40 + %838 = bitcast <2 x bfloat> %837 to i32, !dbg !40 + %839 = insertelement <2 x bfloat> poison, bfloat %756, i64 0, !dbg !40 + %840 = insertelement <2 x bfloat> %839, bfloat %757, i64 1, !dbg !40 + %841 = bitcast <2 x bfloat> %840 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %832, i32 %835, i32 %838, i32 %841, ptr addrspace(1) %636, i1 %42) #6, !dbg !40 + %842 = insertelement <2 x bfloat> poison, bfloat %758, i64 0, !dbg !40 + %843 = insertelement <2 x bfloat> %842, bfloat %759, i64 1, !dbg !40 + %844 = bitcast <2 x bfloat> %843 to i32, !dbg !40 + %845 = insertelement <2 x bfloat> poison, bfloat %760, i64 0, !dbg !40 + %846 = insertelement <2 x bfloat> %845, bfloat %761, i64 1, !dbg !40 + %847 = bitcast <2 x bfloat> %846 to i32, !dbg !40 + %848 = insertelement <2 x bfloat> poison, bfloat %762, i64 0, !dbg !40 + %849 = insertelement <2 x bfloat> %848, bfloat %763, i64 1, !dbg !40 + %850 = bitcast <2 x bfloat> %849 to i32, !dbg !40 + %851 = insertelement <2 x bfloat> poison, bfloat %764, i64 0, !dbg !40 + %852 = insertelement <2 x bfloat> %851, bfloat %765, i64 1, !dbg !40 + %853 = bitcast <2 x bfloat> %852 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %844, i32 %847, i32 %850, i32 %853, ptr addrspace(1) %634, i1 %43) #6, !dbg !40 + %854 = insertelement <2 x bfloat> poison, bfloat %766, i64 0, !dbg !40 + %855 = insertelement <2 x bfloat> %854, bfloat %767, i64 1, !dbg !40 + %856 = bitcast <2 x bfloat> %855 to i32, !dbg !40 + %857 = insertelement <2 x bfloat> poison, bfloat %768, i64 0, !dbg !40 + %858 = insertelement <2 x bfloat> %857, bfloat %769, i64 1, !dbg !40 + %859 = bitcast <2 x bfloat> %858 to i32, !dbg !40 + %860 = insertelement <2 x bfloat> poison, bfloat %770, i64 0, !dbg !40 + %861 = insertelement <2 x bfloat> %860, bfloat %771, i64 1, !dbg !40 + %862 = bitcast <2 x bfloat> %861 to i32, !dbg !40 + %863 = insertelement <2 x bfloat> poison, bfloat %772, i64 0, !dbg !40 + %864 = insertelement <2 x bfloat> %863, bfloat %773, i64 1, !dbg !40 + %865 = bitcast <2 x bfloat> %864 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %856, i32 %859, i32 %862, i32 %865, ptr addrspace(1) %632, i1 %44) #6, !dbg !40 + %866 = insertelement <2 x bfloat> poison, bfloat %774, i64 0, !dbg !40 + %867 = insertelement <2 x bfloat> %866, bfloat %775, i64 1, !dbg !40 + %868 = bitcast <2 x bfloat> %867 to i32, !dbg !40 + %869 = insertelement <2 x bfloat> poison, bfloat %776, i64 0, !dbg !40 + %870 = insertelement <2 x bfloat> %869, bfloat %777, i64 1, !dbg !40 + %871 = bitcast <2 x bfloat> %870 to i32, !dbg !40 + %872 = insertelement <2 x bfloat> poison, bfloat %778, i64 0, !dbg !40 + %873 = insertelement <2 x bfloat> %872, bfloat %779, i64 1, !dbg !40 + %874 = bitcast <2 x bfloat> %873 to i32, !dbg !40 + %875 = insertelement <2 x bfloat> poison, bfloat %780, i64 0, !dbg !40 + %876 = insertelement <2 x bfloat> %875, bfloat %781, i64 1, !dbg !40 + %877 = bitcast <2 x bfloat> %876 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %868, i32 %871, i32 %874, i32 %877, ptr addrspace(1) %630, i1 %45) #6, !dbg !40 + %878 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %12, !dbg !41 + %879 = bitcast float %646 to i32, !dbg !41 + %880 = bitcast float %647 to i32, !dbg !41 + %881 = bitcast float %648 to i32, !dbg !41 + %882 = bitcast float %649 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %878, i32 %879, i32 %880, i32 %881, i32 %882) #6, !dbg !41 + %883 = getelementptr inbounds nuw i8, ptr addrspace(3) %878, i32 128, !dbg !41 + %884 = bitcast float %650 to i32, !dbg !41 + %885 = bitcast float %651 to i32, !dbg !41 + %886 = bitcast float %652 to i32, !dbg !41 + %887 = bitcast float %653 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %883, i32 %884, i32 %885, i32 %886, i32 %887) #6, !dbg !41 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !41 + %888 = shl nuw nsw i32 %11, 4, !dbg !41 + %889 = and i32 %888, 112, !dbg !41 + %890 = lshr i32 %11, 1, !dbg !41 + %891 = and i32 %890, 12, !dbg !41 + %892 = shl nuw nsw i32 %11, 2, !dbg !41 + %893 = and i32 %892, 128, !dbg !41 + %894 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %889, !dbg !41 + %895 = getelementptr inbounds nuw i8, ptr addrspace(3) %894, i32 %893, !dbg !41 + %896 = getelementptr inbounds nuw i8, ptr addrspace(3) %895, i32 %891, !dbg !41 + %897 = load i32, ptr addrspace(3) %896, align 4, !dbg !41 + %898 = and i32 %11, 64, !dbg !41 + %899 = icmp eq i32 %898, 0, !dbg !41 + %900 = and i1 %899, %645, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %897, ptr addrspace(1) %628, i1 %900) #6, !dbg !41 + ret void, !dbg !42 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.div.full(float, float) #3 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #4 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #5 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.f(float) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="128" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) } +attributes #4 = { convergent nocallback nounwind } +attributes #5 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #6 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "l2norm_fwd_kernel", linkageName: "l2norm_fwd_kernel", scope: !1, file: !1, line: 69, type: !6, scopeLine: 69, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !10, !11, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 70, column: 24, scope: !5) +!14 = !DILocation(line: 70, column: 30, scope: !5) +!15 = !DILocation(line: 71, column: 16, scope: !5) +!16 = !DILocation(line: 71, column: 34, scope: !5) +!17 = !DILocation(line: 71, column: 21, scope: !5) +!18 = !DILocation(line: 73, column: 16, scope: !5) +!19 = !DILocation(line: 74, column: 30, scope: !5) +!20 = !DILocation(line: 75, column: 29, scope: !5) +!21 = !DILocation(line: 75, column: 14, scope: !5) +!22 = !DILocation(line: 75, column: 33, scope: !5) +!23 = !DILocation(line: 79, column: 18, scope: !5) +!24 = !DILocation(line: 79, column: 47, scope: !5) +!25 = !DILocation(line: 80, column: 38, scope: !5) +!26 = !DILocation(line: 263, column: 15, scope: !27, inlinedAt: !29) +!27 = distinct !DILexicalBlockFile(scope: !5, file: !28, discriminator: 0) +!28 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!29 = !DILocation(line: 293, column: 36, scope: !27, inlinedAt: !30) +!30 = !DILocation(line: 80, column: 43, scope: !31) +!31 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!32 = !DILocation(line: 80, column: 48, scope: !5) +!33 = !DILocation(line: 80, column: 25, scope: !5) +!34 = !DILocation(line: 77, column: 20, scope: !5) +!35 = !DILocation(line: 76, column: 14, scope: !5) +!36 = !DILocation(line: 76, column: 33, scope: !5) +!37 = !DILocation(line: 80, column: 17, scope: !5) +!38 = !DILocation(line: 81, column: 16, scope: !5) +!39 = !DILocation(line: 83, column: 25, scope: !5) +!40 = !DILocation(line: 83, column: 18, scope: !5) +!41 = !DILocation(line: 84, column: 21, scope: !5) +!42 = !DILocation(line: 84, column: 4, scope: !5) diff --git a/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/l2norm_fwd_kernel.ptx b/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/l2norm_fwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..c88dbfe0df86b93f200094de1da5b6edee6687ce --- /dev/null +++ b/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/l2norm_fwd_kernel.ptx @@ -0,0 +1,969 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_fwd_kernel // -- Begin function l2norm_fwd_kernel +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; +.global .align 1 .b8 _$_str_$_2[17] = {95, 95, 67, 85, 68, 65, 95, 80, 82, 69, 67, 95, 83, 81, 82, 84}; + // @l2norm_fwd_kernel +.visible .entry l2norm_fwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_2, + .param .f32 l2norm_fwd_kernel_param_3, + .param .u32 l2norm_fwd_kernel_param_4, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_6 +) +.reqntid 128 +{ + .reg .pred %p<12>; + .reg .b16 %rs<65>; + .reg .b32 %r<376>; + .reg .b64 %rd<67>; + .loc 1 69 0 // l2norm.py:69:0 +$L__func_begin0: + .loc 1 69 0 // l2norm.py:69:0 + +// %bb.0: // %__nv_sqrtf.exit + ld.param.b64 %rd18, [l2norm_fwd_kernel_param_0]; + ld.param.b64 %rd19, [l2norm_fwd_kernel_param_1]; +$L__tmp0: + .loc 1 70 24 // l2norm.py:70:24 + mov.u32 %r77, %ctaid.x; + ld.param.b64 %rd20, [l2norm_fwd_kernel_param_2]; + .loc 1 71 16 // l2norm.py:71:16 + mul.wide.u32 %rd21, %r77, 64; + ld.param.b32 %r78, [l2norm_fwd_kernel_param_3]; + ld.param.s32 %rd22, [l2norm_fwd_kernel_param_4]; + .loc 1 71 34 // l2norm.py:71:34 + mov.u32 %r79, %tid.x; + and.b32 %r80, %r79, 112; + bfe.u32 %r81, %r79, 4, 3; + or.b32 %r82, %r81, 8; + or.b32 %r83, %r81, 16; + or.b32 %r84, %r81, 24; + or.b32 %r85, %r81, 32; + or.b32 %r86, %r81, 40; + or.b32 %r87, %r81, 48; + or.b32 %r88, %r81, 56; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd23, %r81; + cvt.u64.u32 %rd24, %r82; + cvt.u64.u32 %rd25, %r83; + cvt.u64.u32 %rd26, %r84; + cvt.u64.u32 %rd27, %r85; + cvt.u64.u32 %rd28, %r86; + cvt.u64.u32 %rd29, %r87; + cvt.u64.u32 %rd30, %r88; + or.b64 %rd31, %rd21, %rd23; + or.b64 %rd32, %rd21, %rd24; + or.b64 %rd33, %rd21, %rd25; + or.b64 %rd34, %rd21, %rd26; + or.b64 %rd35, %rd21, %rd27; + or.b64 %rd36, %rd21, %rd28; + or.b64 %rd37, %rd21, %rd29; + or.b64 %rd38, %rd21, %rd30; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p1, %rd31, %rd22; + setp.lt.s64 %p2, %rd32, %rd22; + setp.lt.s64 %p3, %rd33, %rd22; + setp.lt.s64 %p4, %rd34, %rd22; + setp.lt.s64 %p5, %rd35, %rd22; + setp.lt.s64 %p6, %rd36, %rd22; + setp.lt.s64 %p7, %rd37, %rd22; + setp.lt.s64 %p8, %rd38, %rd22; + .loc 1 74 30 // l2norm.py:74:30 + shl.b32 %r89, %r79, 3; + and.b32 %r90, %r89, 120; + .loc 1 75 14 // l2norm.py:75:14 + shl.b64 %rd39, %rd31, 8; + add.s64 %rd40, %rd18, %rd39; + shl.b64 %rd41, %rd32, 8; + add.s64 %rd42, %rd18, %rd41; + shl.b64 %rd43, %rd33, 8; + add.s64 %rd44, %rd18, %rd43; + shl.b64 %rd45, %rd34, 8; + add.s64 %rd46, %rd18, %rd45; + shl.b64 %rd47, %rd35, 8; + add.s64 %rd48, %rd18, %rd47; + shl.b64 %rd49, %rd36, 8; + add.s64 %rd50, %rd18, %rd49; + shl.b64 %rd51, %rd37, 8; + add.s64 %rd52, %rd18, %rd51; + shl.b64 %rd53, %rd38, 8; + add.s64 %rd54, %rd18, %rd53; + .loc 1 75 33 // l2norm.py:75:33 + mul.wide.u32 %rd55, %r90, 2; + add.s64 %rd1, %rd40, %rd55; + add.s64 %rd2, %rd42, %rd55; + add.s64 %rd3, %rd44, %rd55; + add.s64 %rd4, %rd46, %rd55; + add.s64 %rd5, %rd48, %rd55; + add.s64 %rd6, %rd50, %rd55; + add.s64 %rd7, %rd52, %rd55; + add.s64 %rd8, %rd54, %rd55; + mov.b32 %r5, 0; + .loc 1 79 18 // l2norm.py:79:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + mov.b32 {%rs5, %rs6}, %r3; + mov.b32 {%rs7, %rs8}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs9, %rs10}, %r6; + mov.b32 {%rs11, %rs12}, %r7; + mov.b32 {%rs13, %rs14}, %r8; + mov.b32 {%rs15, %rs16}, %r9; + // begin inline asm + mov.u32 %r10, %r5; + mov.u32 %r11, %r5; + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + @%p3 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs17, %rs18}, %r10; + mov.b32 {%rs19, %rs20}, %r11; + mov.b32 {%rs21, %rs22}, %r12; + mov.b32 {%rs23, %rs24}, %r13; + // begin inline asm + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + @%p4 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ]; + // end inline asm + mov.b32 {%rs25, %rs26}, %r14; + mov.b32 {%rs27, %rs28}, %r15; + mov.b32 {%rs29, %rs30}, %r16; + mov.b32 {%rs31, %rs32}, %r17; + // begin inline asm + mov.u32 %r18, %r5; + mov.u32 %r19, %r5; + mov.u32 %r20, %r5; + mov.u32 %r21, %r5; + @%p5 ld.global.v4.b32 { %r18, %r19, %r20, %r21 }, [ %rd5 + 0 ]; + // end inline asm + mov.b32 {%rs33, %rs34}, %r18; + mov.b32 {%rs35, %rs36}, %r19; + mov.b32 {%rs37, %rs38}, %r20; + mov.b32 {%rs39, %rs40}, %r21; + // begin inline asm + mov.u32 %r22, %r5; + mov.u32 %r23, %r5; + mov.u32 %r24, %r5; + mov.u32 %r25, %r5; + @%p6 ld.global.v4.b32 { %r22, %r23, %r24, %r25 }, [ %rd6 + 0 ]; + // end inline asm + mov.b32 {%rs41, %rs42}, %r22; + mov.b32 {%rs43, %rs44}, %r23; + mov.b32 {%rs45, %rs46}, %r24; + mov.b32 {%rs47, %rs48}, %r25; + // begin inline asm + mov.u32 %r26, %r5; + mov.u32 %r27, %r5; + mov.u32 %r28, %r5; + mov.u32 %r29, %r5; + @%p7 ld.global.v4.b32 { %r26, %r27, %r28, %r29 }, [ %rd7 + 0 ]; + // end inline asm + mov.b32 {%rs49, %rs50}, %r26; + mov.b32 {%rs51, %rs52}, %r27; + mov.b32 {%rs53, %rs54}, %r28; + mov.b32 {%rs55, %rs56}, %r29; + // begin inline asm + mov.u32 %r30, %r5; + mov.u32 %r31, %r5; + mov.u32 %r32, %r5; + mov.u32 %r33, %r5; + @%p8 ld.global.v4.b32 { %r30, %r31, %r32, %r33 }, [ %rd8 + 0 ]; + // end inline asm + mov.b32 {%rs57, %rs58}, %r30; + mov.b32 {%rs59, %rs60}, %r31; + mov.b32 {%rs61, %rs62}, %r32; + mov.b32 {%rs63, %rs64}, %r33; + .loc 1 79 47 // l2norm.py:79:47 + cvt.f32.bf16 %r91, %rs1; + cvt.f32.bf16 %r92, %rs2; + cvt.f32.bf16 %r93, %rs3; + cvt.f32.bf16 %r94, %rs4; + cvt.f32.bf16 %r95, %rs5; + cvt.f32.bf16 %r96, %rs6; + cvt.f32.bf16 %r97, %rs7; + cvt.f32.bf16 %r98, %rs8; + cvt.f32.bf16 %r99, %rs9; + cvt.f32.bf16 %r100, %rs10; + cvt.f32.bf16 %r101, %rs11; + cvt.f32.bf16 %r102, %rs12; + cvt.f32.bf16 %r103, %rs13; + cvt.f32.bf16 %r104, %rs14; + cvt.f32.bf16 %r105, %rs15; + cvt.f32.bf16 %r106, %rs16; + cvt.f32.bf16 %r107, %rs17; + cvt.f32.bf16 %r108, %rs18; + cvt.f32.bf16 %r109, %rs19; + cvt.f32.bf16 %r110, %rs20; + cvt.f32.bf16 %r111, %rs21; + cvt.f32.bf16 %r112, %rs22; + cvt.f32.bf16 %r113, %rs23; + cvt.f32.bf16 %r114, %rs24; + cvt.f32.bf16 %r115, %rs25; + cvt.f32.bf16 %r116, %rs26; + cvt.f32.bf16 %r117, %rs27; + cvt.f32.bf16 %r118, %rs28; + cvt.f32.bf16 %r119, %rs29; + cvt.f32.bf16 %r120, %rs30; + cvt.f32.bf16 %r121, %rs31; + cvt.f32.bf16 %r122, %rs32; + cvt.f32.bf16 %r123, %rs33; + cvt.f32.bf16 %r124, %rs34; + cvt.f32.bf16 %r125, %rs35; + cvt.f32.bf16 %r126, %rs36; + cvt.f32.bf16 %r127, %rs37; + cvt.f32.bf16 %r128, %rs38; + cvt.f32.bf16 %r129, %rs39; + cvt.f32.bf16 %r130, %rs40; + cvt.f32.bf16 %r131, %rs41; + cvt.f32.bf16 %r132, %rs42; + cvt.f32.bf16 %r133, %rs43; + cvt.f32.bf16 %r134, %rs44; + cvt.f32.bf16 %r135, %rs45; + cvt.f32.bf16 %r136, %rs46; + cvt.f32.bf16 %r137, %rs47; + cvt.f32.bf16 %r138, %rs48; + cvt.f32.bf16 %r139, %rs49; + cvt.f32.bf16 %r140, %rs50; + cvt.f32.bf16 %r141, %rs51; + cvt.f32.bf16 %r142, %rs52; + cvt.f32.bf16 %r143, %rs53; + cvt.f32.bf16 %r144, %rs54; + cvt.f32.bf16 %r145, %rs55; + cvt.f32.bf16 %r146, %rs56; + cvt.f32.bf16 %r147, %rs57; + cvt.f32.bf16 %r148, %rs58; + cvt.f32.bf16 %r149, %rs59; + cvt.f32.bf16 %r150, %rs60; + cvt.f32.bf16 %r151, %rs61; + cvt.f32.bf16 %r152, %rs62; + cvt.f32.bf16 %r153, %rs63; + cvt.f32.bf16 %r154, %rs64; + .loc 1 80 38 // l2norm.py:80:38 + mul.f32 %r155, %r92, %r92; + mul.f32 %r156, %r100, %r100; + mul.f32 %r157, %r108, %r108; + mul.f32 %r158, %r116, %r116; + mul.f32 %r159, %r124, %r124; + mul.f32 %r160, %r132, %r132; + mul.f32 %r161, %r140, %r140; + mul.f32 %r162, %r148, %r148; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + fma.rn.f32.bf16 %r163, %rs1, %rs1, %r155; + fma.rn.f32.bf16 %r164, %rs3, %rs3, %r163; + fma.rn.f32.bf16 %r165, %rs4, %rs4, %r164; + fma.rn.f32.bf16 %r166, %rs5, %rs5, %r165; + fma.rn.f32.bf16 %r167, %rs6, %rs6, %r166; + fma.rn.f32.bf16 %r168, %rs7, %rs7, %r167; + fma.rn.f32.bf16 %r169, %rs8, %rs8, %r168; + fma.rn.f32.bf16 %r170, %rs9, %rs9, %r156; + fma.rn.f32.bf16 %r171, %rs11, %rs11, %r170; + fma.rn.f32.bf16 %r172, %rs12, %rs12, %r171; + fma.rn.f32.bf16 %r173, %rs13, %rs13, %r172; + fma.rn.f32.bf16 %r174, %rs14, %rs14, %r173; + fma.rn.f32.bf16 %r175, %rs15, %rs15, %r174; + fma.rn.f32.bf16 %r176, %rs16, %rs16, %r175; + fma.rn.f32.bf16 %r177, %rs17, %rs17, %r157; + fma.rn.f32.bf16 %r178, %rs19, %rs19, %r177; + fma.rn.f32.bf16 %r179, %rs20, %rs20, %r178; + fma.rn.f32.bf16 %r180, %rs21, %rs21, %r179; + fma.rn.f32.bf16 %r181, %rs22, %rs22, %r180; + fma.rn.f32.bf16 %r182, %rs23, %rs23, %r181; + fma.rn.f32.bf16 %r183, %rs24, %rs24, %r182; + fma.rn.f32.bf16 %r184, %rs25, %rs25, %r158; + fma.rn.f32.bf16 %r185, %rs27, %rs27, %r184; + fma.rn.f32.bf16 %r186, %rs28, %rs28, %r185; + fma.rn.f32.bf16 %r187, %rs29, %rs29, %r186; + fma.rn.f32.bf16 %r188, %rs30, %rs30, %r187; + fma.rn.f32.bf16 %r189, %rs31, %rs31, %r188; + fma.rn.f32.bf16 %r190, %rs32, %rs32, %r189; + fma.rn.f32.bf16 %r191, %rs33, %rs33, %r159; + fma.rn.f32.bf16 %r192, %rs35, %rs35, %r191; + fma.rn.f32.bf16 %r193, %rs36, %rs36, %r192; + fma.rn.f32.bf16 %r194, %rs37, %rs37, %r193; + fma.rn.f32.bf16 %r195, %rs38, %rs38, %r194; + fma.rn.f32.bf16 %r196, %rs39, %rs39, %r195; + fma.rn.f32.bf16 %r197, %rs40, %rs40, %r196; + fma.rn.f32.bf16 %r198, %rs41, %rs41, %r160; + fma.rn.f32.bf16 %r199, %rs43, %rs43, %r198; + fma.rn.f32.bf16 %r200, %rs44, %rs44, %r199; + fma.rn.f32.bf16 %r201, %rs45, %rs45, %r200; + fma.rn.f32.bf16 %r202, %rs46, %rs46, %r201; + fma.rn.f32.bf16 %r203, %rs47, %rs47, %r202; + fma.rn.f32.bf16 %r204, %rs48, %rs48, %r203; + fma.rn.f32.bf16 %r205, %rs49, %rs49, %r161; + fma.rn.f32.bf16 %r206, %rs51, %rs51, %r205; + fma.rn.f32.bf16 %r207, %rs52, %rs52, %r206; + fma.rn.f32.bf16 %r208, %rs53, %rs53, %r207; + fma.rn.f32.bf16 %r209, %rs54, %rs54, %r208; + fma.rn.f32.bf16 %r210, %rs55, %rs55, %r209; + fma.rn.f32.bf16 %r211, %rs56, %rs56, %r210; + fma.rn.f32.bf16 %r212, %rs57, %rs57, %r162; + fma.rn.f32.bf16 %r213, %rs59, %rs59, %r212; + fma.rn.f32.bf16 %r214, %rs60, %rs60, %r213; + fma.rn.f32.bf16 %r215, %rs61, %rs61, %r214; + fma.rn.f32.bf16 %r216, %rs62, %rs62, %r215; + fma.rn.f32.bf16 %r217, %rs63, %rs63, %r216; + fma.rn.f32.bf16 %r218, %rs64, %rs64, %r217; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r219, %r169, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r220, %r169, %r219; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r221, %r220, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r222, %r220, %r221; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r223, %r222, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r224, %r222, %r223; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r225, %r224, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r226, %r224, %r225; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r227, %r176, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r228, %r176, %r227; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r229, %r228, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r230, %r228, %r229; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r231, %r230, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r232, %r230, %r231; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r233, %r232, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r234, %r232, %r233; +$L__tmp18: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r235, %r183, 8, 31, -1; +$L__tmp19: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r236, %r183, %r235; +$L__tmp20: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r237, %r236, 4, 31, -1; +$L__tmp21: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r238, %r236, %r237; +$L__tmp22: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r239, %r238, 2, 31, -1; +$L__tmp23: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r240, %r238, %r239; +$L__tmp24: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r241, %r240, 1, 31, -1; +$L__tmp25: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r242, %r240, %r241; +$L__tmp26: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r243, %r190, 8, 31, -1; +$L__tmp27: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r244, %r190, %r243; +$L__tmp28: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r245, %r244, 4, 31, -1; +$L__tmp29: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r246, %r244, %r245; +$L__tmp30: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r247, %r246, 2, 31, -1; +$L__tmp31: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r248, %r246, %r247; +$L__tmp32: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r249, %r248, 1, 31, -1; +$L__tmp33: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r250, %r248, %r249; +$L__tmp34: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r251, %r197, 8, 31, -1; +$L__tmp35: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r252, %r197, %r251; +$L__tmp36: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r253, %r252, 4, 31, -1; +$L__tmp37: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r254, %r252, %r253; +$L__tmp38: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r255, %r254, 2, 31, -1; +$L__tmp39: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r256, %r254, %r255; +$L__tmp40: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r257, %r256, 1, 31, -1; +$L__tmp41: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r258, %r256, %r257; +$L__tmp42: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r259, %r204, 8, 31, -1; +$L__tmp43: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r260, %r204, %r259; +$L__tmp44: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r261, %r260, 4, 31, -1; +$L__tmp45: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r262, %r260, %r261; +$L__tmp46: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r263, %r262, 2, 31, -1; +$L__tmp47: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r264, %r262, %r263; +$L__tmp48: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r265, %r264, 1, 31, -1; +$L__tmp49: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r266, %r264, %r265; +$L__tmp50: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r267, %r211, 8, 31, -1; +$L__tmp51: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r268, %r211, %r267; +$L__tmp52: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r269, %r268, 4, 31, -1; +$L__tmp53: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r270, %r268, %r269; +$L__tmp54: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r271, %r270, 2, 31, -1; +$L__tmp55: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r272, %r270, %r271; +$L__tmp56: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r273, %r272, 1, 31, -1; +$L__tmp57: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r274, %r272, %r273; +$L__tmp58: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r275, %r218, 8, 31, -1; +$L__tmp59: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r276, %r218, %r275; +$L__tmp60: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r277, %r276, 4, 31, -1; +$L__tmp61: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r278, %r276, %r277; +$L__tmp62: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r279, %r278, 2, 31, -1; +$L__tmp63: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r280, %r278, %r279; +$L__tmp64: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r281, %r280, 1, 31, -1; +$L__tmp65: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r282, %r280, %r281; +$L__tmp66: + .loc 1 80 48 // l2norm.py:80:48 + add.f32 %r283, %r78, %r226; + add.f32 %r284, %r78, %r234; + add.f32 %r285, %r78, %r242; + add.f32 %r286, %r78, %r250; + add.f32 %r287, %r78, %r258; + add.f32 %r288, %r78, %r266; + add.f32 %r289, %r78, %r274; + add.f32 %r290, %r78, %r282; + .loc 1 80 25 // l2norm.py:80:25 + sqrt.approx.ftz.f32 %r291, %r283; + sqrt.approx.ftz.f32 %r292, %r284; + sqrt.approx.ftz.f32 %r293, %r285; + sqrt.approx.ftz.f32 %r294, %r286; + sqrt.approx.ftz.f32 %r295, %r287; + sqrt.approx.ftz.f32 %r296, %r288; + sqrt.approx.ftz.f32 %r297, %r289; + sqrt.approx.ftz.f32 %r298, %r290; + .loc 1 71 34 // l2norm.py:71:34 + and.b32 %r299, %r79, 63; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd56, %r299; + or.b64 %rd57, %rd21, %rd56; + .loc 1 77 20 // l2norm.py:77:20 + shl.b64 %rd58, %rd57, 2; + add.s64 %rd17, %rd20, %rd58; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd59, %rd19, %rd53; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd16, %rd59, %rd55; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd60, %rd19, %rd51; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd15, %rd60, %rd55; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd61, %rd19, %rd49; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd14, %rd61, %rd55; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd62, %rd19, %rd47; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd13, %rd62, %rd55; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd63, %rd19, %rd45; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd12, %rd63, %rd55; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd64, %rd19, %rd43; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd11, %rd64, %rd55; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd65, %rd19, %rd41; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd10, %rd65, %rd55; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd66, %rd19, %rd39; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd9, %rd66, %rd55; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p10, %rd57, %rd22; + mov.b32 %r300, 0f3F800000; + .loc 1 80 17 // l2norm.py:80:17 + div.full.f32 %r67, %r300, %r291; + div.full.f32 %r68, %r300, %r292; + div.full.f32 %r69, %r300, %r293; + div.full.f32 %r70, %r300, %r294; + div.full.f32 %r72, %r300, %r295; + div.full.f32 %r73, %r300, %r296; + div.full.f32 %r74, %r300, %r297; + div.full.f32 %r75, %r300, %r298; + .loc 1 81 16 // l2norm.py:81:16 + mul.f32 %r301, %r67, %r91; + mul.f32 %r302, %r67, %r92; + mul.f32 %r303, %r67, %r93; + mul.f32 %r304, %r67, %r94; + mul.f32 %r305, %r67, %r95; + mul.f32 %r306, %r67, %r96; + mul.f32 %r307, %r67, %r97; + mul.f32 %r308, %r67, %r98; + mul.f32 %r309, %r68, %r99; + mul.f32 %r310, %r68, %r100; + mul.f32 %r311, %r68, %r101; + mul.f32 %r312, %r68, %r102; + mul.f32 %r313, %r68, %r103; + mul.f32 %r314, %r68, %r104; + mul.f32 %r315, %r68, %r105; + mul.f32 %r316, %r68, %r106; + mul.f32 %r317, %r69, %r107; + mul.f32 %r318, %r69, %r108; + mul.f32 %r319, %r69, %r109; + mul.f32 %r320, %r69, %r110; + mul.f32 %r321, %r69, %r111; + mul.f32 %r322, %r69, %r112; + mul.f32 %r323, %r69, %r113; + mul.f32 %r324, %r69, %r114; + mul.f32 %r325, %r70, %r115; + mul.f32 %r326, %r70, %r116; + mul.f32 %r327, %r70, %r117; + mul.f32 %r328, %r70, %r118; + mul.f32 %r329, %r70, %r119; + mul.f32 %r330, %r70, %r120; + mul.f32 %r331, %r70, %r121; + mul.f32 %r332, %r70, %r122; + mul.f32 %r333, %r72, %r123; + mul.f32 %r334, %r72, %r124; + mul.f32 %r335, %r72, %r125; + mul.f32 %r336, %r72, %r126; + mul.f32 %r337, %r72, %r127; + mul.f32 %r338, %r72, %r128; + mul.f32 %r339, %r72, %r129; + mul.f32 %r340, %r72, %r130; + mul.f32 %r341, %r73, %r131; + mul.f32 %r342, %r73, %r132; + mul.f32 %r343, %r73, %r133; + mul.f32 %r344, %r73, %r134; + mul.f32 %r345, %r73, %r135; + mul.f32 %r346, %r73, %r136; + mul.f32 %r347, %r73, %r137; + mul.f32 %r348, %r73, %r138; + mul.f32 %r349, %r74, %r139; + mul.f32 %r350, %r74, %r140; + mul.f32 %r351, %r74, %r141; + mul.f32 %r352, %r74, %r142; + mul.f32 %r353, %r74, %r143; + mul.f32 %r354, %r74, %r144; + mul.f32 %r355, %r74, %r145; + mul.f32 %r356, %r74, %r146; + mul.f32 %r357, %r75, %r147; + mul.f32 %r358, %r75, %r148; + mul.f32 %r359, %r75, %r149; + mul.f32 %r360, %r75, %r150; + mul.f32 %r361, %r75, %r151; + mul.f32 %r362, %r75, %r152; + mul.f32 %r363, %r75, %r153; + mul.f32 %r364, %r75, %r154; + .loc 1 83 18 // l2norm.py:83:18 + cvt.rn.bf16x2.f32 %r34, %r302, %r301; + cvt.rn.bf16x2.f32 %r35, %r304, %r303; + cvt.rn.bf16x2.f32 %r36, %r306, %r305; + cvt.rn.bf16x2.f32 %r37, %r308, %r307; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd9 + 0 ], { %r34, %r35, %r36, %r37 }; + // end inline asm + cvt.rn.bf16x2.f32 %r38, %r310, %r309; + cvt.rn.bf16x2.f32 %r39, %r312, %r311; + cvt.rn.bf16x2.f32 %r40, %r314, %r313; + cvt.rn.bf16x2.f32 %r41, %r316, %r315; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd10 + 0 ], { %r38, %r39, %r40, %r41 }; + // end inline asm + cvt.rn.bf16x2.f32 %r42, %r318, %r317; + cvt.rn.bf16x2.f32 %r43, %r320, %r319; + cvt.rn.bf16x2.f32 %r44, %r322, %r321; + cvt.rn.bf16x2.f32 %r45, %r324, %r323; + // begin inline asm + @%p3 st.global.v4.b32 [ %rd11 + 0 ], { %r42, %r43, %r44, %r45 }; + // end inline asm + cvt.rn.bf16x2.f32 %r46, %r326, %r325; + cvt.rn.bf16x2.f32 %r47, %r328, %r327; + cvt.rn.bf16x2.f32 %r48, %r330, %r329; + cvt.rn.bf16x2.f32 %r49, %r332, %r331; + // begin inline asm + @%p4 st.global.v4.b32 [ %rd12 + 0 ], { %r46, %r47, %r48, %r49 }; + // end inline asm + cvt.rn.bf16x2.f32 %r50, %r334, %r333; + cvt.rn.bf16x2.f32 %r51, %r336, %r335; + cvt.rn.bf16x2.f32 %r52, %r338, %r337; + cvt.rn.bf16x2.f32 %r53, %r340, %r339; + // begin inline asm + @%p5 st.global.v4.b32 [ %rd13 + 0 ], { %r50, %r51, %r52, %r53 }; + // end inline asm + cvt.rn.bf16x2.f32 %r54, %r342, %r341; + cvt.rn.bf16x2.f32 %r55, %r344, %r343; + cvt.rn.bf16x2.f32 %r56, %r346, %r345; + cvt.rn.bf16x2.f32 %r57, %r348, %r347; + // begin inline asm + @%p6 st.global.v4.b32 [ %rd14 + 0 ], { %r54, %r55, %r56, %r57 }; + // end inline asm + cvt.rn.bf16x2.f32 %r58, %r350, %r349; + cvt.rn.bf16x2.f32 %r59, %r352, %r351; + cvt.rn.bf16x2.f32 %r60, %r354, %r353; + cvt.rn.bf16x2.f32 %r61, %r356, %r355; + // begin inline asm + @%p7 st.global.v4.b32 [ %rd15 + 0 ], { %r58, %r59, %r60, %r61 }; + // end inline asm + cvt.rn.bf16x2.f32 %r62, %r358, %r357; + cvt.rn.bf16x2.f32 %r63, %r360, %r359; + cvt.rn.bf16x2.f32 %r64, %r362, %r361; + cvt.rn.bf16x2.f32 %r65, %r364, %r363; + // begin inline asm + @%p8 st.global.v4.b32 [ %rd16 + 0 ], { %r62, %r63, %r64, %r65 }; + // end inline asm + .loc 1 84 21 // l2norm.py:84:21 + mov.b32 %r365, global_smem; + add.s32 %r66, %r365, %r80; + // begin inline asm + st.shared.v4.b32 [ %r66 + 0 ], { %r67, %r68, %r69, %r70 }; + // end inline asm + add.s32 %r71, %r66, 128; + // begin inline asm + st.shared.v4.b32 [ %r71 + 0 ], { %r72, %r73, %r74, %r75 }; + // end inline asm + bar.sync 0; + shl.b32 %r366, %r79, 4; + and.b32 %r367, %r366, 112; + shr.u32 %r368, %r79, 1; + and.b32 %r369, %r368, 12; + shl.b32 %r370, %r79, 2; + and.b32 %r371, %r370, 128; + add.s32 %r372, %r365, %r367; + add.s32 %r373, %r372, %r371; + add.s32 %r374, %r373, %r369; + ld.shared.b32 %r76, [%r374]; + and.b32 %r375, %r79, 64; + setp.eq.b32 %p11, %r375, 0; + and.pred %p9, %p11, %p10; + // begin inline asm + @%p9 st.global.b32 [ %rd17 + 0 ], { %r76 }; + // end inline asm + .loc 1 84 4 // l2norm.py:84:4 + ret; +$L__tmp67: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp66 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 80 // DW_AT_call_line +.b8 43 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp66 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/l2norm_fwd_kernel.source b/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/l2norm_fwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..b5000c470f5e0683aea9e8045b2d0a8c059ea3d1 --- /dev/null +++ b/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/l2norm_fwd_kernel.source @@ -0,0 +1,172 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc38 = loc(unknown) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc45 = loc("x"(#loc)) +#loc46 = loc("y"(#loc)) +#loc47 = loc("rstd"(#loc)) +#loc48 = loc("eps"(#loc)) +#loc49 = loc("T"(#loc)) +#loc81 = loc("input"(#loc36)) +#loc82 = loc("a"(#loc41)) +#loc83 = loc("b"(#loc41)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc50) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc51) + %o_t = arith.constant 64 : i32 loc(#loc52) + %o_t_1 = arith.constant 64 : i64 loc(#loc52) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc52) + %o_t_3 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc53) + %o_t_4 = arith.extsi %o_t_3 : tensor<64xi32> to tensor<64xi64> loc(#loc54) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<64xi64> loc(#loc54) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<64xi64> loc(#loc54) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc55) + %m_t = arith.extsi %T : i32 to i64 loc(#loc56) + %m_t_7 = tt.splat %m_t : i64 -> tensor<64xi64> loc(#loc56) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<64xi64> loc(#loc56) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc57) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc58) + %m_x_10 = arith.constant 128 : i32 loc(#loc59) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc59) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc59) + %m_x_13 = tt.broadcast %m_x : tensor<64x1xi1> -> tensor<64x128xi1> loc(#loc60) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<64x128xi1> loc(#loc60) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<64x128xi1> loc(#loc60) + %p_x = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc61) + %p_x_16 = arith.constant 128 : i32 loc(#loc62) + %p_x_17 = arith.constant 128 : i64 loc(#loc62) + %p_x_18 = arith.constant dense<128> : tensor<64x1xi64> loc(#loc62) + %p_x_19 = arith.muli %p_x, %p_x_18 : tensor<64x1xi64> loc(#loc62) + %p_x_20 = tt.splat %x : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc63) + %p_x_21 = tt.addptr %p_x_20, %p_x_19 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc63) + %p_x_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc64) + %p_x_23 = tt.broadcast %p_x_21 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc65) + %p_x_24 = tt.broadcast %p_x_22 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc65) + %p_x_25 = tt.addptr %p_x_23, %p_x_24 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc65) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc66) + %p_y_26 = arith.constant 128 : i32 loc(#loc67) + %p_y_27 = arith.constant 128 : i64 loc(#loc67) + %p_y_28 = arith.constant dense<128> : tensor<64x1xi64> loc(#loc67) + %p_y_29 = arith.muli %p_y, %p_y_28 : tensor<64x1xi64> loc(#loc67) + %p_y_30 = tt.splat %y : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc68) + %p_y_31 = tt.addptr %p_y_30, %p_y_29 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc68) + %p_y_32 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc69) + %p_y_33 = tt.broadcast %p_y_31 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc70) + %p_y_34 = tt.broadcast %p_y_32 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc70) + %p_y_35 = tt.addptr %p_y_33, %p_y_34 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc70) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<64x!tt.ptr> loc(#loc71) + %p_rstd_36 = tt.addptr %p_rstd, %o_t_6 : tensor<64x!tt.ptr>, tensor<64xi64> loc(#loc71) + %b_x = arith.constant 0.000000e+00 : f32 loc(#loc72) + %b_x_37 = arith.constant dense<0.000000e+00> : tensor<64x128xf32> loc(#loc72) + %b_x_38 = arith.truncf %b_x_37 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc72) + %b_x_39 = tt.load %p_x_25, %m_x_15, %b_x_38 : tensor<64x128x!tt.ptr> loc(#loc72) + %b_x_40 = arith.extf %b_x_39 : tensor<64x128xbf16> to tensor<64x128xf32> loc(#loc73) + %b_rstd = arith.mulf %b_x_40, %b_x_40 : tensor<64x128xf32> loc(#loc74) + %b_rstd_41 = tt.call @triton.language.standard.sum__fp32S64_128S_c1_cFalse_cNone(%b_rstd) : (tensor<64x128xf32>) -> tensor<64xf32> loc(#loc75) + %b_rstd_42 = tt.splat %eps : f32 -> tensor<64xf32> loc(#loc76) + %b_rstd_43 = arith.addf %b_rstd_41, %b_rstd_42 : tensor<64xf32> loc(#loc76) + %b_rstd_44 = math.sqrt %b_rstd_43 : tensor<64xf32> loc(#loc77) + %b_rstd_45 = arith.constant 1 : i32 loc(#loc78) + %b_rstd_46 = arith.constant 1.000000e+00 : f32 loc(#loc78) + %b_rstd_47 = arith.constant dense<1.000000e+00> : tensor<64xf32> loc(#loc78) + %b_rstd_48 = arith.divf %b_rstd_47, %b_rstd_44 : tensor<64xf32> loc(#loc78) + %b_y = tt.expand_dims %b_rstd_48 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc79) + %b_y_49 = tt.broadcast %b_y : tensor<64x1xf32> -> tensor<64x128xf32> loc(#loc80) + %b_y_50 = arith.mulf %b_x_40, %b_y_49 : tensor<64x128xf32> loc(#loc80) + %0 = arith.truncf %b_y_50 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc32) + tt.store %p_y_35, %0, %m_x_15 : tensor<64x128x!tt.ptr> loc(#loc33) + tt.store %p_rstd_36, %b_rstd_48, %m_t_8 : tensor<64x!tt.ptr> loc(#loc34) + tt.return loc(#loc35) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S64_128S_c1_cFalse_cNone(%input: tensor<64x128xf32> loc("input"(#loc36))) -> tensor<64xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc37) + tt.reduce.return %2 : f32 loc(#loc37) + }) : (tensor<64x128xf32>) -> tensor<64xf32> loc(#loc37) + tt.return %0 : tensor<64xf32> loc(#loc39) + ^bb1: // no predecessors + %1 = ub.poison : tensor<64xf32> loc(#loc40) + tt.return %1 : tensor<64xf32> loc(#loc40) + } loc(#loc36) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc41)), %b: f32 loc("b"(#loc41))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc42) + tt.return %0 : f32 loc(#loc43) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc44) + tt.return %1 : f32 loc(#loc44) + } loc(#loc41) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:18) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:29) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:37) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc50 = loc("i_t"(#loc1)) +#loc51 = loc("i_t"(#loc2)) +#loc52 = loc("o_t"(#loc3)) +#loc53 = loc("o_t"(#loc4)) +#loc54 = loc("o_t"(#loc5)) +#loc55 = loc("o_d"(#loc6)) +#loc56 = loc("m_t"(#loc7)) +#loc57 = loc("m_x"(#loc8)) +#loc58 = loc("m_x"(#loc9)) +#loc59 = loc("m_x"(#loc10)) +#loc60 = loc("m_x"(#loc11)) +#loc61 = loc("p_x"(#loc12)) +#loc62 = loc("p_x"(#loc13)) +#loc63 = loc("p_x"(#loc14)) +#loc64 = loc("p_x"(#loc15)) +#loc65 = loc("p_x"(#loc16)) +#loc66 = loc("p_y"(#loc17)) +#loc67 = loc("p_y"(#loc18)) +#loc68 = loc("p_y"(#loc19)) +#loc69 = loc("p_y"(#loc20)) +#loc70 = loc("p_y"(#loc21)) +#loc71 = loc("p_rstd"(#loc22)) +#loc72 = loc("b_x"(#loc23)) +#loc73 = loc("b_x"(#loc24)) +#loc74 = loc("b_rstd"(#loc25)) +#loc75 = loc("b_rstd"(#loc26)) +#loc76 = loc("b_rstd"(#loc27)) +#loc77 = loc("b_rstd"(#loc28)) +#loc78 = loc("b_rstd"(#loc29)) +#loc79 = loc("b_y"(#loc30)) +#loc80 = loc("b_y"(#loc31)) diff --git a/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/l2norm_fwd_kernel.ttgir b/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/l2norm_fwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..f0e04b0670cefe7b77756edca5802772eb74aed9 --- /dev/null +++ b/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/l2norm_fwd_kernel.ttgir @@ -0,0 +1,135 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [4, 1], order = [1, 0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [4], order = [0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc1 = loc(unknown) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc34 = loc("x"(#loc)) +#loc35 = loc("y"(#loc)) +#loc36 = loc("rstd"(#loc)) +#loc37 = loc("eps"(#loc)) +#loc38 = loc("T"(#loc)) +#loc59 = loc("b_rstd"(#loc23)) +#loc66 = loc(callsite(#loc1 at #loc59)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 4 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<1.000000e+00> : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<64x128xbf16, #blocked> loc(#loc1) + %c64_i64 = arith.constant 64 : i64 loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<64x1xi64, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc39) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc40) + %o_t = arith.muli %i_t_3, %c64_i64 : i64 loc(#loc41) + %o_t_4 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc42) + %o_t_5 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #blocked1> loc(#loc42) + %o_t_6 = arith.extsi %o_t_4 : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_7 = arith.extsi %o_t_5 : tensor<64xi32, #blocked1> to tensor<64xi64, #blocked1> loc(#loc43) + %o_t_8 = tt.splat %o_t : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_9 = tt.splat %o_t : i64 -> tensor<64xi64, #blocked1> loc(#loc43) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<64xi64, #blocked1> loc(#loc43) + %m_t = arith.extsi %T : i32 to i64 loc(#loc44) + %m_t_12 = tt.splat %m_t : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_13 = tt.splat %m_t : i64 -> tensor<64xi64, #blocked1> loc(#loc44) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<64xi64, #blocked1> loc(#loc44) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<64xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64x1xi1, #blocked> loc(#loc45) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc46) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc46) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_1 : tensor<1x128xi32, #blocked> loc(#loc47) + %m_x_19 = tt.broadcast %m_x : tensor<64x1xi1, #blocked> -> tensor<64x128xi1, #blocked> loc(#loc48) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<64x128xi1, #blocked> loc(#loc48) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<64x128xi1, #blocked> loc(#loc48) + %p_x = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64x1xi64, #blocked> loc(#loc49) + %p_x_22 = arith.muli %p_x, %cst_2 : tensor<64x1xi64, #blocked> loc(#loc50) + %p_x_23 = tt.splat %x : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked> loc(#loc51) + %p_x_24 = tt.addptr %p_x_23, %p_x_22 : tensor<64x1x!tt.ptr, #blocked>, tensor<64x1xi64, #blocked> loc(#loc51) + %p_x_25 = tt.broadcast %p_x_24 : tensor<64x1x!tt.ptr, #blocked> -> tensor<64x128x!tt.ptr, #blocked> loc(#loc52) + %p_x_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<64x128xi32, #blocked> loc(#loc52) + %p_x_27 = tt.addptr %p_x_25, %p_x_26 : tensor<64x128x!tt.ptr, #blocked>, tensor<64x128xi32, #blocked> loc(#loc52) + %p_y = tt.splat %y : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked> loc(#loc53) + %p_y_28 = tt.addptr %p_y, %p_x_22 : tensor<64x1x!tt.ptr, #blocked>, tensor<64x1xi64, #blocked> loc(#loc53) + %p_y_29 = tt.broadcast %p_y_28 : tensor<64x1x!tt.ptr, #blocked> -> tensor<64x128x!tt.ptr, #blocked> loc(#loc54) + %p_y_30 = tt.addptr %p_y_29, %p_x_26 : tensor<64x128x!tt.ptr, #blocked>, tensor<64x128xi32, #blocked> loc(#loc54) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<64x!tt.ptr, #blocked1> loc(#loc55) + %p_rstd_31 = tt.addptr %p_rstd, %o_t_11 : tensor<64x!tt.ptr, #blocked1>, tensor<64xi64, #blocked1> loc(#loc55) + %b_x = tt.load %p_x_27, %m_x_21, %cst_0 : tensor<64x128x!tt.ptr, #blocked> loc(#loc56) + %b_x_32 = arith.extf %b_x : tensor<64x128xbf16, #blocked> to tensor<64x128xf32, #blocked> loc(#loc57) + %b_rstd = arith.mulf %b_x_32, %b_x_32 : tensor<64x128xf32, #blocked> loc(#loc58) + %b_rstd_33 = "tt.reduce"(%b_rstd) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_40: f32 loc(callsite(#loc1 at #loc59)), %b_rstd_41: f32 loc(callsite(#loc1 at #loc59))): + %b_rstd_42 = arith.addf %b_rstd_40, %b_rstd_41 : f32 loc(#loc67) + tt.reduce.return %b_rstd_42 : f32 loc(#loc65) + }) : (tensor<64x128xf32, #blocked>) -> tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc65) + %b_rstd_34 = tt.splat %eps : f32 -> tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_36 = math.sqrt %b_rstd_35 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc61) + %b_rstd_37 = arith.divf %cst, %b_rstd_36 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc62) + %b_y = tt.expand_dims %b_rstd_37 {axis = 1 : i32} : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64x1xf32, #blocked> loc(#loc63) + %b_y_38 = tt.broadcast %b_y : tensor<64x1xf32, #blocked> -> tensor<64x128xf32, #blocked> loc(#loc64) + %b_y_39 = arith.mulf %b_x_32, %b_y_38 : tensor<64x128xf32, #blocked> loc(#loc64) + %0 = arith.truncf %b_y_39 : tensor<64x128xf32, #blocked> to tensor<64x128xbf16, #blocked> loc(#loc30) + tt.store %p_y_30, %0, %m_x_21 : tensor<64x128x!tt.ptr, #blocked> loc(#loc31) + %1 = ttg.convert_layout %b_rstd_37 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64xf32, #blocked1> loc(#loc32) + tt.store %p_rstd_31, %1, %m_t_15 : tensor<64x!tt.ptr, #blocked1> loc(#loc32) + tt.return loc(#loc33) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc39 = loc("i_t"(#loc2)) +#loc40 = loc("i_t"(#loc3)) +#loc41 = loc("o_t"(#loc4)) +#loc42 = loc("o_t"(#loc5)) +#loc43 = loc("o_t"(#loc6)) +#loc44 = loc("m_t"(#loc7)) +#loc45 = loc("m_x"(#loc8)) +#loc46 = loc("m_x"(#loc9)) +#loc47 = loc("m_x"(#loc10)) +#loc48 = loc("m_x"(#loc11)) +#loc49 = loc("p_x"(#loc12)) +#loc50 = loc("p_x"(#loc13)) +#loc51 = loc("p_x"(#loc14)) +#loc52 = loc("p_x"(#loc15)) +#loc53 = loc("p_y"(#loc16)) +#loc54 = loc("p_y"(#loc17)) +#loc55 = loc("p_rstd"(#loc18)) +#loc56 = loc("b_x"(#loc19)) +#loc57 = loc("b_x"(#loc20)) +#loc58 = loc("b_rstd"(#loc21)) +#loc60 = loc("b_rstd"(#loc25)) +#loc61 = loc("b_rstd"(#loc26)) +#loc62 = loc("b_rstd"(#loc27)) +#loc63 = loc("b_y"(#loc28)) +#loc64 = loc("b_y"(#loc29)) +#loc65 = loc(callsite(#loc22 at #loc59)) +#loc67 = loc(callsite(#loc24 at #loc65)) diff --git a/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/l2norm_fwd_kernel.ttir b/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/l2norm_fwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..85ebb8b128ceed91f07ab82f5c0fe10a876f6335 --- /dev/null +++ b/triton/gpu0/BXOXWTRONORL5H3YK336JM2IB2AMKDEGMKJRKSMVRNF3QAMENIJA/l2norm_fwd_kernel.ttir @@ -0,0 +1,128 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc3 = loc(unknown) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc35 = loc("x"(#loc)) +#loc36 = loc("y"(#loc)) +#loc37 = loc("rstd"(#loc)) +#loc38 = loc("eps"(#loc)) +#loc39 = loc("T"(#loc)) +#loc62 = loc("b_rstd"(#loc25)) +#loc68 = loc(callsite(#loc3 at #loc62)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %b_x = arith.constant dense<0.000000e+00> : tensor<64x128xbf16> loc(#loc40) + %b_rstd = arith.constant dense<1.000000e+00> : tensor<64xf32> loc(#loc41) + %cst = arith.constant dense<128> : tensor<64x1xi64> loc(#loc3) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc42) + %o_t = arith.constant 64 : i64 loc(#loc43) + %i_t = tt.get_program_id x : i32 loc(#loc44) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc45) + %o_t_1 = arith.muli %i_t_0, %o_t : i64 loc(#loc43) + %o_t_2 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc46) + %o_t_3 = arith.extsi %o_t_2 : tensor<64xi32> to tensor<64xi64> loc(#loc47) + %o_t_4 = tt.splat %o_t_1 : i64 -> tensor<64xi64> loc(#loc47) + %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<64xi64> loc(#loc47) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc48) + %m_t = arith.extsi %T : i32 to i64 loc(#loc49) + %m_t_6 = tt.splat %m_t : i64 -> tensor<64xi64> loc(#loc49) + %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<64xi64> loc(#loc49) + %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc50) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc51) + %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc42) + %m_x_11 = tt.broadcast %m_x_8 : tensor<64x1xi1> -> tensor<64x128xi1> loc(#loc52) + %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<64x128xi1> loc(#loc52) + %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<64x128xi1> loc(#loc52) + %p_x = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc53) + %p_x_14 = arith.muli %p_x, %cst : tensor<64x1xi64> loc(#loc54) + %p_x_15 = tt.splat %x : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc55) + %p_x_16 = tt.addptr %p_x_15, %p_x_14 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc55) + %p_x_17 = tt.broadcast %p_x_16 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc56) + %p_x_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc56) + %p_x_19 = tt.addptr %p_x_17, %p_x_18 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc56) + %p_y = tt.splat %y : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc57) + %p_y_20 = tt.addptr %p_y, %p_x_14 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc57) + %p_y_21 = tt.broadcast %p_y_20 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc58) + %p_y_22 = tt.addptr %p_y_21, %p_x_18 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<64x!tt.ptr> loc(#loc59) + %p_rstd_23 = tt.addptr %p_rstd, %o_t_5 : tensor<64x!tt.ptr>, tensor<64xi64> loc(#loc59) + %b_x_24 = tt.load %p_x_19, %m_x_13, %b_x : tensor<64x128x!tt.ptr> loc(#loc40) + %b_x_25 = arith.extf %b_x_24 : tensor<64x128xbf16> to tensor<64x128xf32> loc(#loc60) + %b_rstd_26 = arith.mulf %b_x_25, %b_x_25 : tensor<64x128xf32> loc(#loc61) + %b_rstd_27 = "tt.reduce"(%b_rstd_26) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_34: f32 loc(callsite(#loc3 at #loc62)), %b_rstd_35: f32 loc(callsite(#loc3 at #loc62))): + %b_rstd_36 = arith.addf %b_rstd_34, %b_rstd_35 : f32 loc(#loc69) + tt.reduce.return %b_rstd_36 : f32 loc(#loc67) + }) : (tensor<64x128xf32>) -> tensor<64xf32> loc(#loc67) + %b_rstd_28 = tt.splat %eps : f32 -> tensor<64xf32> loc(#loc63) + %b_rstd_29 = arith.addf %b_rstd_27, %b_rstd_28 : tensor<64xf32> loc(#loc63) + %b_rstd_30 = math.sqrt %b_rstd_29 : tensor<64xf32> loc(#loc64) + %b_rstd_31 = arith.divf %b_rstd, %b_rstd_30 : tensor<64xf32> loc(#loc41) + %b_y = tt.expand_dims %b_rstd_31 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc65) + %b_y_32 = tt.broadcast %b_y : tensor<64x1xf32> -> tensor<64x128xf32> loc(#loc66) + %b_y_33 = arith.mulf %b_x_25, %b_y_32 : tensor<64x128xf32> loc(#loc66) + %0 = arith.truncf %b_y_33 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc31) + tt.store %p_y_22, %0, %m_x_13 : tensor<64x128x!tt.ptr> loc(#loc32) + tt.store %p_rstd_23, %b_rstd_31, %m_t_7 : tensor<64x!tt.ptr> loc(#loc33) + tt.return loc(#loc34) + } loc(#loc) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc40 = loc("b_x"(#loc1)) +#loc41 = loc("b_rstd"(#loc2)) +#loc42 = loc("m_x"(#loc4)) +#loc43 = loc("o_t"(#loc5)) +#loc44 = loc("i_t"(#loc6)) +#loc45 = loc("i_t"(#loc7)) +#loc46 = loc("o_t"(#loc8)) +#loc47 = loc("o_t"(#loc9)) +#loc48 = loc("o_d"(#loc10)) +#loc49 = loc("m_t"(#loc11)) +#loc50 = loc("m_x"(#loc12)) +#loc51 = loc("m_x"(#loc13)) +#loc52 = loc("m_x"(#loc14)) +#loc53 = loc("p_x"(#loc15)) +#loc54 = loc("p_x"(#loc16)) +#loc55 = loc("p_x"(#loc17)) +#loc56 = loc("p_x"(#loc18)) +#loc57 = loc("p_y"(#loc19)) +#loc58 = loc("p_y"(#loc20)) +#loc59 = loc("p_rstd"(#loc21)) +#loc60 = loc("b_x"(#loc22)) +#loc61 = loc("b_rstd"(#loc23)) +#loc63 = loc("b_rstd"(#loc27)) +#loc64 = loc("b_rstd"(#loc28)) +#loc65 = loc("b_y"(#loc29)) +#loc66 = loc("b_y"(#loc30)) +#loc67 = loc(callsite(#loc24 at #loc62)) +#loc69 = loc(callsite(#loc26 at #loc67)) diff --git a/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/__grp__l2norm_fwd_kernel.json b/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/__grp__l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..aa67cc350ded5028014ce76331db637412a453fd --- /dev/null +++ b/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/__grp__l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_fwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.source", "l2norm_fwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.ttir", "l2norm_fwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.ttgir", "l2norm_fwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.llir", "l2norm_fwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.ptx", "l2norm_fwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.cubin", "l2norm_fwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.cubin b/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..da161322c15171f8c62bc812d13ce80efa02139d --- /dev/null +++ b/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.cubin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:01f06397f0b30f222dafe7d0ff24bfb233bf0b57b9c1dc7883c733a1b6747267 +size 160768 diff --git a/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.json b/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..eede811871a2390cb8c147f34cabe63aafcd4f70 --- /dev/null +++ b/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"hash": "0de3d523b3c0ecc2a7296bf4788ab97dcf9e1b5e2c1b4f7b9440f78f9038b5a4", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 2, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 512, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_fwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.llir b/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..6b00d4f55664f17046aa8f1a8a24783f89b7a438 --- /dev/null +++ b/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.llir @@ -0,0 +1,4181 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 +@.str.2 = private unnamed_addr constant [17 x i8] c"__CUDA_PREC_SQRT\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_fwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, float %3, i32 %4, ptr addrspace(1) readnone captures(none) %5, ptr addrspace(1) readnone captures(none) %6) local_unnamed_addr #0 !dbg !5 { + %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %9 = zext nneg i32 %8 to i64, !dbg !14 + %10 = shl nuw nsw i64 %9, 7, !dbg !15 + %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !16 + %12 = lshr i32 %11, 4, !dbg !16 + %13 = and i32 %12, 3, !dbg !16 + %14 = or disjoint i32 %13, 4, !dbg !16 + %15 = insertelement <2 x i32> poison, i32 %13, i64 0, !dbg !16 + %16 = insertelement <4 x i32> poison, i32 %13, i64 0, !dbg !16 + %17 = insertelement <8 x i32> poison, i32 %12, i64 0, !dbg !16 + %18 = insertelement <8 x i32> %17, i32 %13, i64 1, !dbg !16 + %19 = insertelement <32 x i32> poison, i32 %14, i64 30, !dbg !17 + %20 = insertelement <32 x i32> %19, i32 %13, i64 31, !dbg !17 + %21 = shufflevector <8 x i32> %18, <8 x i32> poison, <32 x i32> , !dbg !17 + %22 = or <32 x i32> %21, , !dbg !17 + %23 = shufflevector <32 x i32> %22, <32 x i32> %20, <32 x i32> , !dbg !17 + %24 = shufflevector <8 x i32> %18, <8 x i32> poison, <32 x i32> , !dbg !17 + %25 = or <32 x i32> %24, , !dbg !17 + %26 = shufflevector <32 x i32> %23, <32 x i32> %25, <32 x i32> , !dbg !17 + %27 = shufflevector <4 x i32> %16, <4 x i32> poison, <32 x i32> , !dbg !17 + %28 = or disjoint <32 x i32> %27, , !dbg !17 + %29 = shufflevector <32 x i32> %26, <32 x i32> %28, <32 x i32> , !dbg !17 + %30 = shufflevector <2 x i32> %15, <2 x i32> poison, <32 x i32> , !dbg !17 + %31 = or disjoint <32 x i32> %30, , !dbg !17 + %32 = shufflevector <32 x i32> %29, <32 x i32> %31, <32 x i32> , !dbg !17 + %33 = zext <32 x i32> %32 to <32 x i64>, !dbg !17 + %34 = insertelement <32 x i64> poison, i64 %10, i64 0, !dbg !17 + %35 = shufflevector <32 x i64> %34, <32 x i64> poison, <32 x i32> zeroinitializer, !dbg !17 + %36 = or disjoint <32 x i64> %35, %33, !dbg !17 + %37 = insertelement <32 x i32> poison, i32 %4, i64 0, !dbg !18 + %38 = shufflevector <32 x i32> %37, <32 x i32> poison, <32 x i32> zeroinitializer, !dbg !18 + %39 = sext <32 x i32> %38 to <32 x i64>, !dbg !18 + %40 = icmp slt <32 x i64> %36, %39, !dbg !18 + %41 = shl nuw nsw i32 %11, 3, !dbg !19 + %42 = and i32 %41, 120, !dbg !19 + %43 = extractelement <32 x i64> %36, i64 31, !dbg !20 + %44 = shl nuw nsw i64 %43, 7, !dbg !20 + %45 = extractelement <32 x i64> %36, i64 30, !dbg !20 + %46 = shl nuw nsw i64 %45, 7, !dbg !20 + %47 = extractelement <32 x i64> %36, i64 29, !dbg !20 + %48 = shl nuw nsw i64 %47, 7, !dbg !20 + %49 = extractelement <32 x i64> %36, i64 28, !dbg !20 + %50 = shl nuw nsw i64 %49, 7, !dbg !20 + %51 = extractelement <32 x i64> %36, i64 27, !dbg !20 + %52 = shl nuw nsw i64 %51, 7, !dbg !20 + %53 = extractelement <32 x i64> %36, i64 26, !dbg !20 + %54 = shl nuw nsw i64 %53, 7, !dbg !20 + %55 = extractelement <32 x i64> %36, i64 25, !dbg !20 + %56 = shl nuw nsw i64 %55, 7, !dbg !20 + %57 = extractelement <32 x i64> %36, i64 24, !dbg !20 + %58 = shl nuw nsw i64 %57, 7, !dbg !20 + %59 = extractelement <32 x i64> %36, i64 23, !dbg !20 + %60 = shl nuw nsw i64 %59, 7, !dbg !20 + %61 = extractelement <32 x i64> %36, i64 22, !dbg !20 + %62 = shl nuw nsw i64 %61, 7, !dbg !20 + %63 = extractelement <32 x i64> %36, i64 21, !dbg !20 + %64 = shl nuw nsw i64 %63, 7, !dbg !20 + %65 = extractelement <32 x i64> %36, i64 20, !dbg !20 + %66 = shl nuw nsw i64 %65, 7, !dbg !20 + %67 = extractelement <32 x i64> %36, i64 19, !dbg !20 + %68 = shl nuw nsw i64 %67, 7, !dbg !20 + %69 = extractelement <32 x i64> %36, i64 18, !dbg !20 + %70 = shl nuw nsw i64 %69, 7, !dbg !20 + %71 = extractelement <32 x i64> %36, i64 17, !dbg !20 + %72 = shl nuw nsw i64 %71, 7, !dbg !20 + %73 = extractelement <32 x i64> %36, i64 16, !dbg !20 + %74 = shl nuw nsw i64 %73, 7, !dbg !20 + %75 = extractelement <32 x i64> %36, i64 15, !dbg !20 + %76 = shl nuw nsw i64 %75, 7, !dbg !20 + %77 = extractelement <32 x i64> %36, i64 14, !dbg !20 + %78 = shl nuw nsw i64 %77, 7, !dbg !20 + %79 = extractelement <32 x i64> %36, i64 13, !dbg !20 + %80 = shl nuw nsw i64 %79, 7, !dbg !20 + %81 = extractelement <32 x i64> %36, i64 12, !dbg !20 + %82 = shl nuw nsw i64 %81, 7, !dbg !20 + %83 = extractelement <32 x i64> %36, i64 11, !dbg !20 + %84 = shl nuw nsw i64 %83, 7, !dbg !20 + %85 = extractelement <32 x i64> %36, i64 10, !dbg !20 + %86 = shl nuw nsw i64 %85, 7, !dbg !20 + %87 = extractelement <32 x i64> %36, i64 9, !dbg !20 + %88 = shl nuw nsw i64 %87, 7, !dbg !20 + %89 = extractelement <32 x i64> %36, i64 8, !dbg !20 + %90 = shl nuw nsw i64 %89, 7, !dbg !20 + %91 = extractelement <32 x i64> %36, i64 7, !dbg !20 + %92 = shl nuw nsw i64 %91, 7, !dbg !20 + %93 = extractelement <32 x i64> %36, i64 6, !dbg !20 + %94 = shl nuw nsw i64 %93, 7, !dbg !20 + %95 = extractelement <32 x i64> %36, i64 5, !dbg !20 + %96 = shl nuw nsw i64 %95, 7, !dbg !20 + %97 = extractelement <32 x i64> %36, i64 4, !dbg !20 + %98 = shl nuw nsw i64 %97, 7, !dbg !20 + %99 = extractelement <32 x i64> %36, i64 3, !dbg !20 + %100 = shl nuw nsw i64 %99, 7, !dbg !20 + %101 = extractelement <32 x i64> %36, i64 2, !dbg !20 + %102 = shl nuw nsw i64 %101, 7, !dbg !20 + %103 = extractelement <32 x i64> %36, i64 1, !dbg !20 + %104 = shl nuw nsw i64 %103, 7, !dbg !20 + %105 = extractelement <32 x i64> %36, i64 0, !dbg !20 + %106 = shl nuw nsw i64 %105, 7, !dbg !20 + %107 = getelementptr bfloat, ptr addrspace(1) %0, i64 %44, !dbg !21 + %108 = getelementptr bfloat, ptr addrspace(1) %0, i64 %46, !dbg !21 + %109 = getelementptr bfloat, ptr addrspace(1) %0, i64 %48, !dbg !21 + %110 = getelementptr bfloat, ptr addrspace(1) %0, i64 %50, !dbg !21 + %111 = getelementptr bfloat, ptr addrspace(1) %0, i64 %52, !dbg !21 + %112 = getelementptr bfloat, ptr addrspace(1) %0, i64 %54, !dbg !21 + %113 = getelementptr bfloat, ptr addrspace(1) %0, i64 %56, !dbg !21 + %114 = getelementptr bfloat, ptr addrspace(1) %0, i64 %58, !dbg !21 + %115 = getelementptr bfloat, ptr addrspace(1) %0, i64 %60, !dbg !21 + %116 = getelementptr bfloat, ptr addrspace(1) %0, i64 %62, !dbg !21 + %117 = getelementptr bfloat, ptr addrspace(1) %0, i64 %64, !dbg !21 + %118 = getelementptr bfloat, ptr addrspace(1) %0, i64 %66, !dbg !21 + %119 = getelementptr bfloat, ptr addrspace(1) %0, i64 %68, !dbg !21 + %120 = getelementptr bfloat, ptr addrspace(1) %0, i64 %70, !dbg !21 + %121 = getelementptr bfloat, ptr addrspace(1) %0, i64 %72, !dbg !21 + %122 = getelementptr bfloat, ptr addrspace(1) %0, i64 %74, !dbg !21 + %123 = getelementptr bfloat, ptr addrspace(1) %0, i64 %76, !dbg !21 + %124 = getelementptr bfloat, ptr addrspace(1) %0, i64 %78, !dbg !21 + %125 = getelementptr bfloat, ptr addrspace(1) %0, i64 %80, !dbg !21 + %126 = getelementptr bfloat, ptr addrspace(1) %0, i64 %82, !dbg !21 + %127 = getelementptr bfloat, ptr addrspace(1) %0, i64 %84, !dbg !21 + %128 = getelementptr bfloat, ptr addrspace(1) %0, i64 %86, !dbg !21 + %129 = getelementptr bfloat, ptr addrspace(1) %0, i64 %88, !dbg !21 + %130 = getelementptr bfloat, ptr addrspace(1) %0, i64 %90, !dbg !21 + %131 = getelementptr bfloat, ptr addrspace(1) %0, i64 %92, !dbg !21 + %132 = getelementptr bfloat, ptr addrspace(1) %0, i64 %94, !dbg !21 + %133 = getelementptr bfloat, ptr addrspace(1) %0, i64 %96, !dbg !21 + %134 = getelementptr bfloat, ptr addrspace(1) %0, i64 %98, !dbg !21 + %135 = getelementptr bfloat, ptr addrspace(1) %0, i64 %100, !dbg !21 + %136 = getelementptr bfloat, ptr addrspace(1) %0, i64 %102, !dbg !21 + %137 = getelementptr bfloat, ptr addrspace(1) %0, i64 %104, !dbg !21 + %138 = getelementptr bfloat, ptr addrspace(1) %0, i64 %106, !dbg !21 + %139 = zext nneg i32 %42 to i64, !dbg !22 + %140 = getelementptr bfloat, ptr addrspace(1) %107, i64 %139, !dbg !22 + %141 = getelementptr bfloat, ptr addrspace(1) %108, i64 %139, !dbg !22 + %142 = getelementptr bfloat, ptr addrspace(1) %109, i64 %139, !dbg !22 + %143 = getelementptr bfloat, ptr addrspace(1) %110, i64 %139, !dbg !22 + %144 = getelementptr bfloat, ptr addrspace(1) %111, i64 %139, !dbg !22 + %145 = getelementptr bfloat, ptr addrspace(1) %112, i64 %139, !dbg !22 + %146 = getelementptr bfloat, ptr addrspace(1) %113, i64 %139, !dbg !22 + %147 = getelementptr bfloat, ptr addrspace(1) %114, i64 %139, !dbg !22 + %148 = getelementptr bfloat, ptr addrspace(1) %115, i64 %139, !dbg !22 + %149 = getelementptr bfloat, ptr addrspace(1) %116, i64 %139, !dbg !22 + %150 = getelementptr bfloat, ptr addrspace(1) %117, i64 %139, !dbg !22 + %151 = getelementptr bfloat, ptr addrspace(1) %118, i64 %139, !dbg !22 + %152 = getelementptr bfloat, ptr addrspace(1) %119, i64 %139, !dbg !22 + %153 = getelementptr bfloat, ptr addrspace(1) %120, i64 %139, !dbg !22 + %154 = getelementptr bfloat, ptr addrspace(1) %121, i64 %139, !dbg !22 + %155 = getelementptr bfloat, ptr addrspace(1) %122, i64 %139, !dbg !22 + %156 = getelementptr bfloat, ptr addrspace(1) %123, i64 %139, !dbg !22 + %157 = getelementptr bfloat, ptr addrspace(1) %124, i64 %139, !dbg !22 + %158 = getelementptr bfloat, ptr addrspace(1) %125, i64 %139, !dbg !22 + %159 = getelementptr bfloat, ptr addrspace(1) %126, i64 %139, !dbg !22 + %160 = getelementptr bfloat, ptr addrspace(1) %127, i64 %139, !dbg !22 + %161 = getelementptr bfloat, ptr addrspace(1) %128, i64 %139, !dbg !22 + %162 = getelementptr bfloat, ptr addrspace(1) %129, i64 %139, !dbg !22 + %163 = getelementptr bfloat, ptr addrspace(1) %130, i64 %139, !dbg !22 + %164 = getelementptr bfloat, ptr addrspace(1) %131, i64 %139, !dbg !22 + %165 = getelementptr bfloat, ptr addrspace(1) %132, i64 %139, !dbg !22 + %166 = getelementptr bfloat, ptr addrspace(1) %133, i64 %139, !dbg !22 + %167 = getelementptr bfloat, ptr addrspace(1) %134, i64 %139, !dbg !22 + %168 = getelementptr bfloat, ptr addrspace(1) %135, i64 %139, !dbg !22 + %169 = getelementptr bfloat, ptr addrspace(1) %136, i64 %139, !dbg !22 + %170 = getelementptr bfloat, ptr addrspace(1) %137, i64 %139, !dbg !22 + %171 = getelementptr bfloat, ptr addrspace(1) %138, i64 %139, !dbg !22 + %172 = extractelement <32 x i1> %40, i64 31, !dbg !23 + %173 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %140, i1 %172) #6, !dbg !23 + %174 = extractvalue { i32, i32, i32, i32 } %173, 0, !dbg !23 + %175 = bitcast i32 %174 to <2 x bfloat>, !dbg !23 + %176 = extractvalue { i32, i32, i32, i32 } %173, 1, !dbg !23 + %177 = bitcast i32 %176 to <2 x bfloat>, !dbg !23 + %178 = extractvalue { i32, i32, i32, i32 } %173, 2, !dbg !23 + %179 = bitcast i32 %178 to <2 x bfloat>, !dbg !23 + %180 = extractvalue { i32, i32, i32, i32 } %173, 3, !dbg !23 + %181 = bitcast i32 %180 to <2 x bfloat>, !dbg !23 + %182 = extractelement <2 x bfloat> %175, i64 0, !dbg !23 + %183 = extractelement <2 x bfloat> %175, i64 1, !dbg !23 + %184 = extractelement <2 x bfloat> %177, i64 0, !dbg !23 + %185 = extractelement <2 x bfloat> %177, i64 1, !dbg !23 + %186 = extractelement <2 x bfloat> %179, i64 0, !dbg !23 + %187 = extractelement <2 x bfloat> %179, i64 1, !dbg !23 + %188 = extractelement <2 x bfloat> %181, i64 0, !dbg !23 + %189 = extractelement <2 x bfloat> %181, i64 1, !dbg !23 + %190 = extractelement <32 x i1> %40, i64 30, !dbg !23 + %191 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %141, i1 %190) #6, !dbg !23 + %192 = extractvalue { i32, i32, i32, i32 } %191, 0, !dbg !23 + %193 = bitcast i32 %192 to <2 x bfloat>, !dbg !23 + %194 = extractvalue { i32, i32, i32, i32 } %191, 1, !dbg !23 + %195 = bitcast i32 %194 to <2 x bfloat>, !dbg !23 + %196 = extractvalue { i32, i32, i32, i32 } %191, 2, !dbg !23 + %197 = bitcast i32 %196 to <2 x bfloat>, !dbg !23 + %198 = extractvalue { i32, i32, i32, i32 } %191, 3, !dbg !23 + %199 = bitcast i32 %198 to <2 x bfloat>, !dbg !23 + %200 = extractelement <2 x bfloat> %193, i64 0, !dbg !23 + %201 = extractelement <2 x bfloat> %193, i64 1, !dbg !23 + %202 = extractelement <2 x bfloat> %195, i64 0, !dbg !23 + %203 = extractelement <2 x bfloat> %195, i64 1, !dbg !23 + %204 = extractelement <2 x bfloat> %197, i64 0, !dbg !23 + %205 = extractelement <2 x bfloat> %197, i64 1, !dbg !23 + %206 = extractelement <2 x bfloat> %199, i64 0, !dbg !23 + %207 = extractelement <2 x bfloat> %199, i64 1, !dbg !23 + %208 = extractelement <32 x i1> %40, i64 29, !dbg !23 + %209 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %142, i1 %208) #6, !dbg !23 + %210 = extractvalue { i32, i32, i32, i32 } %209, 0, !dbg !23 + %211 = bitcast i32 %210 to <2 x bfloat>, !dbg !23 + %212 = extractvalue { i32, i32, i32, i32 } %209, 1, !dbg !23 + %213 = bitcast i32 %212 to <2 x bfloat>, !dbg !23 + %214 = extractvalue { i32, i32, i32, i32 } %209, 2, !dbg !23 + %215 = bitcast i32 %214 to <2 x bfloat>, !dbg !23 + %216 = extractvalue { i32, i32, i32, i32 } %209, 3, !dbg !23 + %217 = bitcast i32 %216 to <2 x bfloat>, !dbg !23 + %218 = extractelement <2 x bfloat> %211, i64 0, !dbg !23 + %219 = extractelement <2 x bfloat> %211, i64 1, !dbg !23 + %220 = extractelement <2 x bfloat> %213, i64 0, !dbg !23 + %221 = extractelement <2 x bfloat> %213, i64 1, !dbg !23 + %222 = extractelement <2 x bfloat> %215, i64 0, !dbg !23 + %223 = extractelement <2 x bfloat> %215, i64 1, !dbg !23 + %224 = extractelement <2 x bfloat> %217, i64 0, !dbg !23 + %225 = extractelement <2 x bfloat> %217, i64 1, !dbg !23 + %226 = extractelement <32 x i1> %40, i64 28, !dbg !23 + %227 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %143, i1 %226) #6, !dbg !23 + %228 = extractvalue { i32, i32, i32, i32 } %227, 0, !dbg !23 + %229 = bitcast i32 %228 to <2 x bfloat>, !dbg !23 + %230 = extractvalue { i32, i32, i32, i32 } %227, 1, !dbg !23 + %231 = bitcast i32 %230 to <2 x bfloat>, !dbg !23 + %232 = extractvalue { i32, i32, i32, i32 } %227, 2, !dbg !23 + %233 = bitcast i32 %232 to <2 x bfloat>, !dbg !23 + %234 = extractvalue { i32, i32, i32, i32 } %227, 3, !dbg !23 + %235 = bitcast i32 %234 to <2 x bfloat>, !dbg !23 + %236 = extractelement <2 x bfloat> %229, i64 0, !dbg !23 + %237 = extractelement <2 x bfloat> %229, i64 1, !dbg !23 + %238 = extractelement <2 x bfloat> %231, i64 0, !dbg !23 + %239 = extractelement <2 x bfloat> %231, i64 1, !dbg !23 + %240 = extractelement <2 x bfloat> %233, i64 0, !dbg !23 + %241 = extractelement <2 x bfloat> %233, i64 1, !dbg !23 + %242 = extractelement <2 x bfloat> %235, i64 0, !dbg !23 + %243 = extractelement <2 x bfloat> %235, i64 1, !dbg !23 + %244 = extractelement <32 x i1> %40, i64 27, !dbg !23 + %245 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %144, i1 %244) #6, !dbg !23 + %246 = extractvalue { i32, i32, i32, i32 } %245, 0, !dbg !23 + %247 = bitcast i32 %246 to <2 x bfloat>, !dbg !23 + %248 = extractvalue { i32, i32, i32, i32 } %245, 1, !dbg !23 + %249 = bitcast i32 %248 to <2 x bfloat>, !dbg !23 + %250 = extractvalue { i32, i32, i32, i32 } %245, 2, !dbg !23 + %251 = bitcast i32 %250 to <2 x bfloat>, !dbg !23 + %252 = extractvalue { i32, i32, i32, i32 } %245, 3, !dbg !23 + %253 = bitcast i32 %252 to <2 x bfloat>, !dbg !23 + %254 = extractelement <2 x bfloat> %247, i64 0, !dbg !23 + %255 = extractelement <2 x bfloat> %247, i64 1, !dbg !23 + %256 = extractelement <2 x bfloat> %249, i64 0, !dbg !23 + %257 = extractelement <2 x bfloat> %249, i64 1, !dbg !23 + %258 = extractelement <2 x bfloat> %251, i64 0, !dbg !23 + %259 = extractelement <2 x bfloat> %251, i64 1, !dbg !23 + %260 = extractelement <2 x bfloat> %253, i64 0, !dbg !23 + %261 = extractelement <2 x bfloat> %253, i64 1, !dbg !23 + %262 = extractelement <32 x i1> %40, i64 26, !dbg !23 + %263 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %145, i1 %262) #6, !dbg !23 + %264 = extractvalue { i32, i32, i32, i32 } %263, 0, !dbg !23 + %265 = bitcast i32 %264 to <2 x bfloat>, !dbg !23 + %266 = extractvalue { i32, i32, i32, i32 } %263, 1, !dbg !23 + %267 = bitcast i32 %266 to <2 x bfloat>, !dbg !23 + %268 = extractvalue { i32, i32, i32, i32 } %263, 2, !dbg !23 + %269 = bitcast i32 %268 to <2 x bfloat>, !dbg !23 + %270 = extractvalue { i32, i32, i32, i32 } %263, 3, !dbg !23 + %271 = bitcast i32 %270 to <2 x bfloat>, !dbg !23 + %272 = extractelement <2 x bfloat> %265, i64 0, !dbg !23 + %273 = extractelement <2 x bfloat> %265, i64 1, !dbg !23 + %274 = extractelement <2 x bfloat> %267, i64 0, !dbg !23 + %275 = extractelement <2 x bfloat> %267, i64 1, !dbg !23 + %276 = extractelement <2 x bfloat> %269, i64 0, !dbg !23 + %277 = extractelement <2 x bfloat> %269, i64 1, !dbg !23 + %278 = extractelement <2 x bfloat> %271, i64 0, !dbg !23 + %279 = extractelement <2 x bfloat> %271, i64 1, !dbg !23 + %280 = extractelement <32 x i1> %40, i64 25, !dbg !23 + %281 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %146, i1 %280) #6, !dbg !23 + %282 = extractvalue { i32, i32, i32, i32 } %281, 0, !dbg !23 + %283 = bitcast i32 %282 to <2 x bfloat>, !dbg !23 + %284 = extractvalue { i32, i32, i32, i32 } %281, 1, !dbg !23 + %285 = bitcast i32 %284 to <2 x bfloat>, !dbg !23 + %286 = extractvalue { i32, i32, i32, i32 } %281, 2, !dbg !23 + %287 = bitcast i32 %286 to <2 x bfloat>, !dbg !23 + %288 = extractvalue { i32, i32, i32, i32 } %281, 3, !dbg !23 + %289 = bitcast i32 %288 to <2 x bfloat>, !dbg !23 + %290 = extractelement <2 x bfloat> %283, i64 0, !dbg !23 + %291 = extractelement <2 x bfloat> %283, i64 1, !dbg !23 + %292 = extractelement <2 x bfloat> %285, i64 0, !dbg !23 + %293 = extractelement <2 x bfloat> %285, i64 1, !dbg !23 + %294 = extractelement <2 x bfloat> %287, i64 0, !dbg !23 + %295 = extractelement <2 x bfloat> %287, i64 1, !dbg !23 + %296 = extractelement <2 x bfloat> %289, i64 0, !dbg !23 + %297 = extractelement <2 x bfloat> %289, i64 1, !dbg !23 + %298 = extractelement <32 x i1> %40, i64 24, !dbg !23 + %299 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %147, i1 %298) #6, !dbg !23 + %300 = extractvalue { i32, i32, i32, i32 } %299, 0, !dbg !23 + %301 = bitcast i32 %300 to <2 x bfloat>, !dbg !23 + %302 = extractvalue { i32, i32, i32, i32 } %299, 1, !dbg !23 + %303 = bitcast i32 %302 to <2 x bfloat>, !dbg !23 + %304 = extractvalue { i32, i32, i32, i32 } %299, 2, !dbg !23 + %305 = bitcast i32 %304 to <2 x bfloat>, !dbg !23 + %306 = extractvalue { i32, i32, i32, i32 } %299, 3, !dbg !23 + %307 = bitcast i32 %306 to <2 x bfloat>, !dbg !23 + %308 = extractelement <2 x bfloat> %301, i64 0, !dbg !23 + %309 = extractelement <2 x bfloat> %301, i64 1, !dbg !23 + %310 = extractelement <2 x bfloat> %303, i64 0, !dbg !23 + %311 = extractelement <2 x bfloat> %303, i64 1, !dbg !23 + %312 = extractelement <2 x bfloat> %305, i64 0, !dbg !23 + %313 = extractelement <2 x bfloat> %305, i64 1, !dbg !23 + %314 = extractelement <2 x bfloat> %307, i64 0, !dbg !23 + %315 = extractelement <2 x bfloat> %307, i64 1, !dbg !23 + %316 = extractelement <32 x i1> %40, i64 23, !dbg !23 + %317 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %148, i1 %316) #6, !dbg !23 + %318 = extractvalue { i32, i32, i32, i32 } %317, 0, !dbg !23 + %319 = bitcast i32 %318 to <2 x bfloat>, !dbg !23 + %320 = extractvalue { i32, i32, i32, i32 } %317, 1, !dbg !23 + %321 = bitcast i32 %320 to <2 x bfloat>, !dbg !23 + %322 = extractvalue { i32, i32, i32, i32 } %317, 2, !dbg !23 + %323 = bitcast i32 %322 to <2 x bfloat>, !dbg !23 + %324 = extractvalue { i32, i32, i32, i32 } %317, 3, !dbg !23 + %325 = bitcast i32 %324 to <2 x bfloat>, !dbg !23 + %326 = extractelement <2 x bfloat> %319, i64 0, !dbg !23 + %327 = extractelement <2 x bfloat> %319, i64 1, !dbg !23 + %328 = extractelement <2 x bfloat> %321, i64 0, !dbg !23 + %329 = extractelement <2 x bfloat> %321, i64 1, !dbg !23 + %330 = extractelement <2 x bfloat> %323, i64 0, !dbg !23 + %331 = extractelement <2 x bfloat> %323, i64 1, !dbg !23 + %332 = extractelement <2 x bfloat> %325, i64 0, !dbg !23 + %333 = extractelement <2 x bfloat> %325, i64 1, !dbg !23 + %334 = extractelement <32 x i1> %40, i64 22, !dbg !23 + %335 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %149, i1 %334) #6, !dbg !23 + %336 = extractvalue { i32, i32, i32, i32 } %335, 0, !dbg !23 + %337 = bitcast i32 %336 to <2 x bfloat>, !dbg !23 + %338 = extractvalue { i32, i32, i32, i32 } %335, 1, !dbg !23 + %339 = bitcast i32 %338 to <2 x bfloat>, !dbg !23 + %340 = extractvalue { i32, i32, i32, i32 } %335, 2, !dbg !23 + %341 = bitcast i32 %340 to <2 x bfloat>, !dbg !23 + %342 = extractvalue { i32, i32, i32, i32 } %335, 3, !dbg !23 + %343 = bitcast i32 %342 to <2 x bfloat>, !dbg !23 + %344 = extractelement <2 x bfloat> %337, i64 0, !dbg !23 + %345 = extractelement <2 x bfloat> %337, i64 1, !dbg !23 + %346 = extractelement <2 x bfloat> %339, i64 0, !dbg !23 + %347 = extractelement <2 x bfloat> %339, i64 1, !dbg !23 + %348 = extractelement <2 x bfloat> %341, i64 0, !dbg !23 + %349 = extractelement <2 x bfloat> %341, i64 1, !dbg !23 + %350 = extractelement <2 x bfloat> %343, i64 0, !dbg !23 + %351 = extractelement <2 x bfloat> %343, i64 1, !dbg !23 + %352 = extractelement <32 x i1> %40, i64 21, !dbg !23 + %353 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %150, i1 %352) #6, !dbg !23 + %354 = extractvalue { i32, i32, i32, i32 } %353, 0, !dbg !23 + %355 = bitcast i32 %354 to <2 x bfloat>, !dbg !23 + %356 = extractvalue { i32, i32, i32, i32 } %353, 1, !dbg !23 + %357 = bitcast i32 %356 to <2 x bfloat>, !dbg !23 + %358 = extractvalue { i32, i32, i32, i32 } %353, 2, !dbg !23 + %359 = bitcast i32 %358 to <2 x bfloat>, !dbg !23 + %360 = extractvalue { i32, i32, i32, i32 } %353, 3, !dbg !23 + %361 = bitcast i32 %360 to <2 x bfloat>, !dbg !23 + %362 = extractelement <2 x bfloat> %355, i64 0, !dbg !23 + %363 = extractelement <2 x bfloat> %355, i64 1, !dbg !23 + %364 = extractelement <2 x bfloat> %357, i64 0, !dbg !23 + %365 = extractelement <2 x bfloat> %357, i64 1, !dbg !23 + %366 = extractelement <2 x bfloat> %359, i64 0, !dbg !23 + %367 = extractelement <2 x bfloat> %359, i64 1, !dbg !23 + %368 = extractelement <2 x bfloat> %361, i64 0, !dbg !23 + %369 = extractelement <2 x bfloat> %361, i64 1, !dbg !23 + %370 = extractelement <32 x i1> %40, i64 20, !dbg !23 + %371 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %151, i1 %370) #6, !dbg !23 + %372 = extractvalue { i32, i32, i32, i32 } %371, 0, !dbg !23 + %373 = bitcast i32 %372 to <2 x bfloat>, !dbg !23 + %374 = extractvalue { i32, i32, i32, i32 } %371, 1, !dbg !23 + %375 = bitcast i32 %374 to <2 x bfloat>, !dbg !23 + %376 = extractvalue { i32, i32, i32, i32 } %371, 2, !dbg !23 + %377 = bitcast i32 %376 to <2 x bfloat>, !dbg !23 + %378 = extractvalue { i32, i32, i32, i32 } %371, 3, !dbg !23 + %379 = bitcast i32 %378 to <2 x bfloat>, !dbg !23 + %380 = extractelement <2 x bfloat> %373, i64 0, !dbg !23 + %381 = extractelement <2 x bfloat> %373, i64 1, !dbg !23 + %382 = extractelement <2 x bfloat> %375, i64 0, !dbg !23 + %383 = extractelement <2 x bfloat> %375, i64 1, !dbg !23 + %384 = extractelement <2 x bfloat> %377, i64 0, !dbg !23 + %385 = extractelement <2 x bfloat> %377, i64 1, !dbg !23 + %386 = extractelement <2 x bfloat> %379, i64 0, !dbg !23 + %387 = extractelement <2 x bfloat> %379, i64 1, !dbg !23 + %388 = extractelement <32 x i1> %40, i64 19, !dbg !23 + %389 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %152, i1 %388) #6, !dbg !23 + %390 = extractvalue { i32, i32, i32, i32 } %389, 0, !dbg !23 + %391 = bitcast i32 %390 to <2 x bfloat>, !dbg !23 + %392 = extractvalue { i32, i32, i32, i32 } %389, 1, !dbg !23 + %393 = bitcast i32 %392 to <2 x bfloat>, !dbg !23 + %394 = extractvalue { i32, i32, i32, i32 } %389, 2, !dbg !23 + %395 = bitcast i32 %394 to <2 x bfloat>, !dbg !23 + %396 = extractvalue { i32, i32, i32, i32 } %389, 3, !dbg !23 + %397 = bitcast i32 %396 to <2 x bfloat>, !dbg !23 + %398 = extractelement <2 x bfloat> %391, i64 0, !dbg !23 + %399 = extractelement <2 x bfloat> %391, i64 1, !dbg !23 + %400 = extractelement <2 x bfloat> %393, i64 0, !dbg !23 + %401 = extractelement <2 x bfloat> %393, i64 1, !dbg !23 + %402 = extractelement <2 x bfloat> %395, i64 0, !dbg !23 + %403 = extractelement <2 x bfloat> %395, i64 1, !dbg !23 + %404 = extractelement <2 x bfloat> %397, i64 0, !dbg !23 + %405 = extractelement <2 x bfloat> %397, i64 1, !dbg !23 + %406 = extractelement <32 x i1> %40, i64 18, !dbg !23 + %407 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %153, i1 %406) #6, !dbg !23 + %408 = extractvalue { i32, i32, i32, i32 } %407, 0, !dbg !23 + %409 = bitcast i32 %408 to <2 x bfloat>, !dbg !23 + %410 = extractvalue { i32, i32, i32, i32 } %407, 1, !dbg !23 + %411 = bitcast i32 %410 to <2 x bfloat>, !dbg !23 + %412 = extractvalue { i32, i32, i32, i32 } %407, 2, !dbg !23 + %413 = bitcast i32 %412 to <2 x bfloat>, !dbg !23 + %414 = extractvalue { i32, i32, i32, i32 } %407, 3, !dbg !23 + %415 = bitcast i32 %414 to <2 x bfloat>, !dbg !23 + %416 = extractelement <2 x bfloat> %409, i64 0, !dbg !23 + %417 = extractelement <2 x bfloat> %409, i64 1, !dbg !23 + %418 = extractelement <2 x bfloat> %411, i64 0, !dbg !23 + %419 = extractelement <2 x bfloat> %411, i64 1, !dbg !23 + %420 = extractelement <2 x bfloat> %413, i64 0, !dbg !23 + %421 = extractelement <2 x bfloat> %413, i64 1, !dbg !23 + %422 = extractelement <2 x bfloat> %415, i64 0, !dbg !23 + %423 = extractelement <2 x bfloat> %415, i64 1, !dbg !23 + %424 = extractelement <32 x i1> %40, i64 17, !dbg !23 + %425 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %154, i1 %424) #6, !dbg !23 + %426 = extractvalue { i32, i32, i32, i32 } %425, 0, !dbg !23 + %427 = bitcast i32 %426 to <2 x bfloat>, !dbg !23 + %428 = extractvalue { i32, i32, i32, i32 } %425, 1, !dbg !23 + %429 = bitcast i32 %428 to <2 x bfloat>, !dbg !23 + %430 = extractvalue { i32, i32, i32, i32 } %425, 2, !dbg !23 + %431 = bitcast i32 %430 to <2 x bfloat>, !dbg !23 + %432 = extractvalue { i32, i32, i32, i32 } %425, 3, !dbg !23 + %433 = bitcast i32 %432 to <2 x bfloat>, !dbg !23 + %434 = extractelement <2 x bfloat> %427, i64 0, !dbg !23 + %435 = extractelement <2 x bfloat> %427, i64 1, !dbg !23 + %436 = extractelement <2 x bfloat> %429, i64 0, !dbg !23 + %437 = extractelement <2 x bfloat> %429, i64 1, !dbg !23 + %438 = extractelement <2 x bfloat> %431, i64 0, !dbg !23 + %439 = extractelement <2 x bfloat> %431, i64 1, !dbg !23 + %440 = extractelement <2 x bfloat> %433, i64 0, !dbg !23 + %441 = extractelement <2 x bfloat> %433, i64 1, !dbg !23 + %442 = extractelement <32 x i1> %40, i64 16, !dbg !23 + %443 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %155, i1 %442) #6, !dbg !23 + %444 = extractvalue { i32, i32, i32, i32 } %443, 0, !dbg !23 + %445 = bitcast i32 %444 to <2 x bfloat>, !dbg !23 + %446 = extractvalue { i32, i32, i32, i32 } %443, 1, !dbg !23 + %447 = bitcast i32 %446 to <2 x bfloat>, !dbg !23 + %448 = extractvalue { i32, i32, i32, i32 } %443, 2, !dbg !23 + %449 = bitcast i32 %448 to <2 x bfloat>, !dbg !23 + %450 = extractvalue { i32, i32, i32, i32 } %443, 3, !dbg !23 + %451 = bitcast i32 %450 to <2 x bfloat>, !dbg !23 + %452 = extractelement <2 x bfloat> %445, i64 0, !dbg !23 + %453 = extractelement <2 x bfloat> %445, i64 1, !dbg !23 + %454 = extractelement <2 x bfloat> %447, i64 0, !dbg !23 + %455 = extractelement <2 x bfloat> %447, i64 1, !dbg !23 + %456 = extractelement <2 x bfloat> %449, i64 0, !dbg !23 + %457 = extractelement <2 x bfloat> %449, i64 1, !dbg !23 + %458 = extractelement <2 x bfloat> %451, i64 0, !dbg !23 + %459 = extractelement <2 x bfloat> %451, i64 1, !dbg !23 + %460 = extractelement <32 x i1> %40, i64 15, !dbg !23 + %461 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %156, i1 %460) #6, !dbg !23 + %462 = extractvalue { i32, i32, i32, i32 } %461, 0, !dbg !23 + %463 = bitcast i32 %462 to <2 x bfloat>, !dbg !23 + %464 = extractvalue { i32, i32, i32, i32 } %461, 1, !dbg !23 + %465 = bitcast i32 %464 to <2 x bfloat>, !dbg !23 + %466 = extractvalue { i32, i32, i32, i32 } %461, 2, !dbg !23 + %467 = bitcast i32 %466 to <2 x bfloat>, !dbg !23 + %468 = extractvalue { i32, i32, i32, i32 } %461, 3, !dbg !23 + %469 = bitcast i32 %468 to <2 x bfloat>, !dbg !23 + %470 = extractelement <2 x bfloat> %463, i64 0, !dbg !23 + %471 = extractelement <2 x bfloat> %463, i64 1, !dbg !23 + %472 = extractelement <2 x bfloat> %465, i64 0, !dbg !23 + %473 = extractelement <2 x bfloat> %465, i64 1, !dbg !23 + %474 = extractelement <2 x bfloat> %467, i64 0, !dbg !23 + %475 = extractelement <2 x bfloat> %467, i64 1, !dbg !23 + %476 = extractelement <2 x bfloat> %469, i64 0, !dbg !23 + %477 = extractelement <2 x bfloat> %469, i64 1, !dbg !23 + %478 = extractelement <32 x i1> %40, i64 14, !dbg !23 + %479 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %157, i1 %478) #6, !dbg !23 + %480 = extractvalue { i32, i32, i32, i32 } %479, 0, !dbg !23 + %481 = bitcast i32 %480 to <2 x bfloat>, !dbg !23 + %482 = extractvalue { i32, i32, i32, i32 } %479, 1, !dbg !23 + %483 = bitcast i32 %482 to <2 x bfloat>, !dbg !23 + %484 = extractvalue { i32, i32, i32, i32 } %479, 2, !dbg !23 + %485 = bitcast i32 %484 to <2 x bfloat>, !dbg !23 + %486 = extractvalue { i32, i32, i32, i32 } %479, 3, !dbg !23 + %487 = bitcast i32 %486 to <2 x bfloat>, !dbg !23 + %488 = extractelement <2 x bfloat> %481, i64 0, !dbg !23 + %489 = extractelement <2 x bfloat> %481, i64 1, !dbg !23 + %490 = extractelement <2 x bfloat> %483, i64 0, !dbg !23 + %491 = extractelement <2 x bfloat> %483, i64 1, !dbg !23 + %492 = extractelement <2 x bfloat> %485, i64 0, !dbg !23 + %493 = extractelement <2 x bfloat> %485, i64 1, !dbg !23 + %494 = extractelement <2 x bfloat> %487, i64 0, !dbg !23 + %495 = extractelement <2 x bfloat> %487, i64 1, !dbg !23 + %496 = extractelement <32 x i1> %40, i64 13, !dbg !23 + %497 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %158, i1 %496) #6, !dbg !23 + %498 = extractvalue { i32, i32, i32, i32 } %497, 0, !dbg !23 + %499 = bitcast i32 %498 to <2 x bfloat>, !dbg !23 + %500 = extractvalue { i32, i32, i32, i32 } %497, 1, !dbg !23 + %501 = bitcast i32 %500 to <2 x bfloat>, !dbg !23 + %502 = extractvalue { i32, i32, i32, i32 } %497, 2, !dbg !23 + %503 = bitcast i32 %502 to <2 x bfloat>, !dbg !23 + %504 = extractvalue { i32, i32, i32, i32 } %497, 3, !dbg !23 + %505 = bitcast i32 %504 to <2 x bfloat>, !dbg !23 + %506 = extractelement <2 x bfloat> %499, i64 0, !dbg !23 + %507 = extractelement <2 x bfloat> %499, i64 1, !dbg !23 + %508 = extractelement <2 x bfloat> %501, i64 0, !dbg !23 + %509 = extractelement <2 x bfloat> %501, i64 1, !dbg !23 + %510 = extractelement <2 x bfloat> %503, i64 0, !dbg !23 + %511 = extractelement <2 x bfloat> %503, i64 1, !dbg !23 + %512 = extractelement <2 x bfloat> %505, i64 0, !dbg !23 + %513 = extractelement <2 x bfloat> %505, i64 1, !dbg !23 + %514 = extractelement <32 x i1> %40, i64 12, !dbg !23 + %515 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %159, i1 %514) #6, !dbg !23 + %516 = extractvalue { i32, i32, i32, i32 } %515, 0, !dbg !23 + %517 = bitcast i32 %516 to <2 x bfloat>, !dbg !23 + %518 = extractvalue { i32, i32, i32, i32 } %515, 1, !dbg !23 + %519 = bitcast i32 %518 to <2 x bfloat>, !dbg !23 + %520 = extractvalue { i32, i32, i32, i32 } %515, 2, !dbg !23 + %521 = bitcast i32 %520 to <2 x bfloat>, !dbg !23 + %522 = extractvalue { i32, i32, i32, i32 } %515, 3, !dbg !23 + %523 = bitcast i32 %522 to <2 x bfloat>, !dbg !23 + %524 = extractelement <2 x bfloat> %517, i64 0, !dbg !23 + %525 = extractelement <2 x bfloat> %517, i64 1, !dbg !23 + %526 = extractelement <2 x bfloat> %519, i64 0, !dbg !23 + %527 = extractelement <2 x bfloat> %519, i64 1, !dbg !23 + %528 = extractelement <2 x bfloat> %521, i64 0, !dbg !23 + %529 = extractelement <2 x bfloat> %521, i64 1, !dbg !23 + %530 = extractelement <2 x bfloat> %523, i64 0, !dbg !23 + %531 = extractelement <2 x bfloat> %523, i64 1, !dbg !23 + %532 = extractelement <32 x i1> %40, i64 11, !dbg !23 + %533 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %160, i1 %532) #6, !dbg !23 + %534 = extractvalue { i32, i32, i32, i32 } %533, 0, !dbg !23 + %535 = bitcast i32 %534 to <2 x bfloat>, !dbg !23 + %536 = extractvalue { i32, i32, i32, i32 } %533, 1, !dbg !23 + %537 = bitcast i32 %536 to <2 x bfloat>, !dbg !23 + %538 = extractvalue { i32, i32, i32, i32 } %533, 2, !dbg !23 + %539 = bitcast i32 %538 to <2 x bfloat>, !dbg !23 + %540 = extractvalue { i32, i32, i32, i32 } %533, 3, !dbg !23 + %541 = bitcast i32 %540 to <2 x bfloat>, !dbg !23 + %542 = extractelement <2 x bfloat> %535, i64 0, !dbg !23 + %543 = extractelement <2 x bfloat> %535, i64 1, !dbg !23 + %544 = extractelement <2 x bfloat> %537, i64 0, !dbg !23 + %545 = extractelement <2 x bfloat> %537, i64 1, !dbg !23 + %546 = extractelement <2 x bfloat> %539, i64 0, !dbg !23 + %547 = extractelement <2 x bfloat> %539, i64 1, !dbg !23 + %548 = extractelement <2 x bfloat> %541, i64 0, !dbg !23 + %549 = extractelement <2 x bfloat> %541, i64 1, !dbg !23 + %550 = extractelement <32 x i1> %40, i64 10, !dbg !23 + %551 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %161, i1 %550) #6, !dbg !23 + %552 = extractvalue { i32, i32, i32, i32 } %551, 0, !dbg !23 + %553 = bitcast i32 %552 to <2 x bfloat>, !dbg !23 + %554 = extractvalue { i32, i32, i32, i32 } %551, 1, !dbg !23 + %555 = bitcast i32 %554 to <2 x bfloat>, !dbg !23 + %556 = extractvalue { i32, i32, i32, i32 } %551, 2, !dbg !23 + %557 = bitcast i32 %556 to <2 x bfloat>, !dbg !23 + %558 = extractvalue { i32, i32, i32, i32 } %551, 3, !dbg !23 + %559 = bitcast i32 %558 to <2 x bfloat>, !dbg !23 + %560 = extractelement <2 x bfloat> %553, i64 0, !dbg !23 + %561 = extractelement <2 x bfloat> %553, i64 1, !dbg !23 + %562 = extractelement <2 x bfloat> %555, i64 0, !dbg !23 + %563 = extractelement <2 x bfloat> %555, i64 1, !dbg !23 + %564 = extractelement <2 x bfloat> %557, i64 0, !dbg !23 + %565 = extractelement <2 x bfloat> %557, i64 1, !dbg !23 + %566 = extractelement <2 x bfloat> %559, i64 0, !dbg !23 + %567 = extractelement <2 x bfloat> %559, i64 1, !dbg !23 + %568 = extractelement <32 x i1> %40, i64 9, !dbg !23 + %569 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %162, i1 %568) #6, !dbg !23 + %570 = extractvalue { i32, i32, i32, i32 } %569, 0, !dbg !23 + %571 = bitcast i32 %570 to <2 x bfloat>, !dbg !23 + %572 = extractvalue { i32, i32, i32, i32 } %569, 1, !dbg !23 + %573 = bitcast i32 %572 to <2 x bfloat>, !dbg !23 + %574 = extractvalue { i32, i32, i32, i32 } %569, 2, !dbg !23 + %575 = bitcast i32 %574 to <2 x bfloat>, !dbg !23 + %576 = extractvalue { i32, i32, i32, i32 } %569, 3, !dbg !23 + %577 = bitcast i32 %576 to <2 x bfloat>, !dbg !23 + %578 = extractelement <2 x bfloat> %571, i64 0, !dbg !23 + %579 = extractelement <2 x bfloat> %571, i64 1, !dbg !23 + %580 = extractelement <2 x bfloat> %573, i64 0, !dbg !23 + %581 = extractelement <2 x bfloat> %573, i64 1, !dbg !23 + %582 = extractelement <2 x bfloat> %575, i64 0, !dbg !23 + %583 = extractelement <2 x bfloat> %575, i64 1, !dbg !23 + %584 = extractelement <2 x bfloat> %577, i64 0, !dbg !23 + %585 = extractelement <2 x bfloat> %577, i64 1, !dbg !23 + %586 = extractelement <32 x i1> %40, i64 8, !dbg !23 + %587 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %163, i1 %586) #6, !dbg !23 + %588 = extractvalue { i32, i32, i32, i32 } %587, 0, !dbg !23 + %589 = bitcast i32 %588 to <2 x bfloat>, !dbg !23 + %590 = extractvalue { i32, i32, i32, i32 } %587, 1, !dbg !23 + %591 = bitcast i32 %590 to <2 x bfloat>, !dbg !23 + %592 = extractvalue { i32, i32, i32, i32 } %587, 2, !dbg !23 + %593 = bitcast i32 %592 to <2 x bfloat>, !dbg !23 + %594 = extractvalue { i32, i32, i32, i32 } %587, 3, !dbg !23 + %595 = bitcast i32 %594 to <2 x bfloat>, !dbg !23 + %596 = extractelement <2 x bfloat> %589, i64 0, !dbg !23 + %597 = extractelement <2 x bfloat> %589, i64 1, !dbg !23 + %598 = extractelement <2 x bfloat> %591, i64 0, !dbg !23 + %599 = extractelement <2 x bfloat> %591, i64 1, !dbg !23 + %600 = extractelement <2 x bfloat> %593, i64 0, !dbg !23 + %601 = extractelement <2 x bfloat> %593, i64 1, !dbg !23 + %602 = extractelement <2 x bfloat> %595, i64 0, !dbg !23 + %603 = extractelement <2 x bfloat> %595, i64 1, !dbg !23 + %604 = extractelement <32 x i1> %40, i64 7, !dbg !23 + %605 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %164, i1 %604) #6, !dbg !23 + %606 = extractvalue { i32, i32, i32, i32 } %605, 0, !dbg !23 + %607 = bitcast i32 %606 to <2 x bfloat>, !dbg !23 + %608 = extractvalue { i32, i32, i32, i32 } %605, 1, !dbg !23 + %609 = bitcast i32 %608 to <2 x bfloat>, !dbg !23 + %610 = extractvalue { i32, i32, i32, i32 } %605, 2, !dbg !23 + %611 = bitcast i32 %610 to <2 x bfloat>, !dbg !23 + %612 = extractvalue { i32, i32, i32, i32 } %605, 3, !dbg !23 + %613 = bitcast i32 %612 to <2 x bfloat>, !dbg !23 + %614 = extractelement <2 x bfloat> %607, i64 0, !dbg !23 + %615 = extractelement <2 x bfloat> %607, i64 1, !dbg !23 + %616 = extractelement <2 x bfloat> %609, i64 0, !dbg !23 + %617 = extractelement <2 x bfloat> %609, i64 1, !dbg !23 + %618 = extractelement <2 x bfloat> %611, i64 0, !dbg !23 + %619 = extractelement <2 x bfloat> %611, i64 1, !dbg !23 + %620 = extractelement <2 x bfloat> %613, i64 0, !dbg !23 + %621 = extractelement <2 x bfloat> %613, i64 1, !dbg !23 + %622 = extractelement <32 x i1> %40, i64 6, !dbg !23 + %623 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %165, i1 %622) #6, !dbg !23 + %624 = extractvalue { i32, i32, i32, i32 } %623, 0, !dbg !23 + %625 = bitcast i32 %624 to <2 x bfloat>, !dbg !23 + %626 = extractvalue { i32, i32, i32, i32 } %623, 1, !dbg !23 + %627 = bitcast i32 %626 to <2 x bfloat>, !dbg !23 + %628 = extractvalue { i32, i32, i32, i32 } %623, 2, !dbg !23 + %629 = bitcast i32 %628 to <2 x bfloat>, !dbg !23 + %630 = extractvalue { i32, i32, i32, i32 } %623, 3, !dbg !23 + %631 = bitcast i32 %630 to <2 x bfloat>, !dbg !23 + %632 = extractelement <2 x bfloat> %625, i64 0, !dbg !23 + %633 = extractelement <2 x bfloat> %625, i64 1, !dbg !23 + %634 = extractelement <2 x bfloat> %627, i64 0, !dbg !23 + %635 = extractelement <2 x bfloat> %627, i64 1, !dbg !23 + %636 = extractelement <2 x bfloat> %629, i64 0, !dbg !23 + %637 = extractelement <2 x bfloat> %629, i64 1, !dbg !23 + %638 = extractelement <2 x bfloat> %631, i64 0, !dbg !23 + %639 = extractelement <2 x bfloat> %631, i64 1, !dbg !23 + %640 = extractelement <32 x i1> %40, i64 5, !dbg !23 + %641 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %166, i1 %640) #6, !dbg !23 + %642 = extractvalue { i32, i32, i32, i32 } %641, 0, !dbg !23 + %643 = bitcast i32 %642 to <2 x bfloat>, !dbg !23 + %644 = extractvalue { i32, i32, i32, i32 } %641, 1, !dbg !23 + %645 = bitcast i32 %644 to <2 x bfloat>, !dbg !23 + %646 = extractvalue { i32, i32, i32, i32 } %641, 2, !dbg !23 + %647 = bitcast i32 %646 to <2 x bfloat>, !dbg !23 + %648 = extractvalue { i32, i32, i32, i32 } %641, 3, !dbg !23 + %649 = bitcast i32 %648 to <2 x bfloat>, !dbg !23 + %650 = extractelement <2 x bfloat> %643, i64 0, !dbg !23 + %651 = extractelement <2 x bfloat> %643, i64 1, !dbg !23 + %652 = extractelement <2 x bfloat> %645, i64 0, !dbg !23 + %653 = extractelement <2 x bfloat> %645, i64 1, !dbg !23 + %654 = extractelement <2 x bfloat> %647, i64 0, !dbg !23 + %655 = extractelement <2 x bfloat> %647, i64 1, !dbg !23 + %656 = extractelement <2 x bfloat> %649, i64 0, !dbg !23 + %657 = extractelement <2 x bfloat> %649, i64 1, !dbg !23 + %658 = extractelement <32 x i1> %40, i64 4, !dbg !23 + %659 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %167, i1 %658) #6, !dbg !23 + %660 = extractvalue { i32, i32, i32, i32 } %659, 0, !dbg !23 + %661 = bitcast i32 %660 to <2 x bfloat>, !dbg !23 + %662 = extractvalue { i32, i32, i32, i32 } %659, 1, !dbg !23 + %663 = bitcast i32 %662 to <2 x bfloat>, !dbg !23 + %664 = extractvalue { i32, i32, i32, i32 } %659, 2, !dbg !23 + %665 = bitcast i32 %664 to <2 x bfloat>, !dbg !23 + %666 = extractvalue { i32, i32, i32, i32 } %659, 3, !dbg !23 + %667 = bitcast i32 %666 to <2 x bfloat>, !dbg !23 + %668 = extractelement <2 x bfloat> %661, i64 0, !dbg !23 + %669 = extractelement <2 x bfloat> %661, i64 1, !dbg !23 + %670 = extractelement <2 x bfloat> %663, i64 0, !dbg !23 + %671 = extractelement <2 x bfloat> %663, i64 1, !dbg !23 + %672 = extractelement <2 x bfloat> %665, i64 0, !dbg !23 + %673 = extractelement <2 x bfloat> %665, i64 1, !dbg !23 + %674 = extractelement <2 x bfloat> %667, i64 0, !dbg !23 + %675 = extractelement <2 x bfloat> %667, i64 1, !dbg !23 + %676 = extractelement <32 x i1> %40, i64 3, !dbg !23 + %677 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %168, i1 %676) #6, !dbg !23 + %678 = extractvalue { i32, i32, i32, i32 } %677, 0, !dbg !23 + %679 = bitcast i32 %678 to <2 x bfloat>, !dbg !23 + %680 = extractvalue { i32, i32, i32, i32 } %677, 1, !dbg !23 + %681 = bitcast i32 %680 to <2 x bfloat>, !dbg !23 + %682 = extractvalue { i32, i32, i32, i32 } %677, 2, !dbg !23 + %683 = bitcast i32 %682 to <2 x bfloat>, !dbg !23 + %684 = extractvalue { i32, i32, i32, i32 } %677, 3, !dbg !23 + %685 = bitcast i32 %684 to <2 x bfloat>, !dbg !23 + %686 = extractelement <2 x bfloat> %679, i64 0, !dbg !23 + %687 = extractelement <2 x bfloat> %679, i64 1, !dbg !23 + %688 = extractelement <2 x bfloat> %681, i64 0, !dbg !23 + %689 = extractelement <2 x bfloat> %681, i64 1, !dbg !23 + %690 = extractelement <2 x bfloat> %683, i64 0, !dbg !23 + %691 = extractelement <2 x bfloat> %683, i64 1, !dbg !23 + %692 = extractelement <2 x bfloat> %685, i64 0, !dbg !23 + %693 = extractelement <2 x bfloat> %685, i64 1, !dbg !23 + %694 = extractelement <32 x i1> %40, i64 2, !dbg !23 + %695 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %169, i1 %694) #6, !dbg !23 + %696 = extractvalue { i32, i32, i32, i32 } %695, 0, !dbg !23 + %697 = bitcast i32 %696 to <2 x bfloat>, !dbg !23 + %698 = extractvalue { i32, i32, i32, i32 } %695, 1, !dbg !23 + %699 = bitcast i32 %698 to <2 x bfloat>, !dbg !23 + %700 = extractvalue { i32, i32, i32, i32 } %695, 2, !dbg !23 + %701 = bitcast i32 %700 to <2 x bfloat>, !dbg !23 + %702 = extractvalue { i32, i32, i32, i32 } %695, 3, !dbg !23 + %703 = bitcast i32 %702 to <2 x bfloat>, !dbg !23 + %704 = extractelement <2 x bfloat> %697, i64 0, !dbg !23 + %705 = extractelement <2 x bfloat> %697, i64 1, !dbg !23 + %706 = extractelement <2 x bfloat> %699, i64 0, !dbg !23 + %707 = extractelement <2 x bfloat> %699, i64 1, !dbg !23 + %708 = extractelement <2 x bfloat> %701, i64 0, !dbg !23 + %709 = extractelement <2 x bfloat> %701, i64 1, !dbg !23 + %710 = extractelement <2 x bfloat> %703, i64 0, !dbg !23 + %711 = extractelement <2 x bfloat> %703, i64 1, !dbg !23 + %712 = extractelement <32 x i1> %40, i64 1, !dbg !23 + %713 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %170, i1 %712) #6, !dbg !23 + %714 = extractvalue { i32, i32, i32, i32 } %713, 0, !dbg !23 + %715 = bitcast i32 %714 to <2 x bfloat>, !dbg !23 + %716 = extractvalue { i32, i32, i32, i32 } %713, 1, !dbg !23 + %717 = bitcast i32 %716 to <2 x bfloat>, !dbg !23 + %718 = extractvalue { i32, i32, i32, i32 } %713, 2, !dbg !23 + %719 = bitcast i32 %718 to <2 x bfloat>, !dbg !23 + %720 = extractvalue { i32, i32, i32, i32 } %713, 3, !dbg !23 + %721 = bitcast i32 %720 to <2 x bfloat>, !dbg !23 + %722 = extractelement <2 x bfloat> %715, i64 0, !dbg !23 + %723 = extractelement <2 x bfloat> %715, i64 1, !dbg !23 + %724 = extractelement <2 x bfloat> %717, i64 0, !dbg !23 + %725 = extractelement <2 x bfloat> %717, i64 1, !dbg !23 + %726 = extractelement <2 x bfloat> %719, i64 0, !dbg !23 + %727 = extractelement <2 x bfloat> %719, i64 1, !dbg !23 + %728 = extractelement <2 x bfloat> %721, i64 0, !dbg !23 + %729 = extractelement <2 x bfloat> %721, i64 1, !dbg !23 + %730 = extractelement <32 x i1> %40, i64 0, !dbg !23 + %731 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %171, i1 %730) #6, !dbg !23 + %732 = extractvalue { i32, i32, i32, i32 } %731, 0, !dbg !23 + %733 = bitcast i32 %732 to <2 x bfloat>, !dbg !23 + %734 = extractvalue { i32, i32, i32, i32 } %731, 1, !dbg !23 + %735 = bitcast i32 %734 to <2 x bfloat>, !dbg !23 + %736 = extractvalue { i32, i32, i32, i32 } %731, 2, !dbg !23 + %737 = bitcast i32 %736 to <2 x bfloat>, !dbg !23 + %738 = extractvalue { i32, i32, i32, i32 } %731, 3, !dbg !23 + %739 = bitcast i32 %738 to <2 x bfloat>, !dbg !23 + %740 = extractelement <2 x bfloat> %733, i64 0, !dbg !23 + %741 = extractelement <2 x bfloat> %733, i64 1, !dbg !23 + %742 = extractelement <2 x bfloat> %735, i64 0, !dbg !23 + %743 = extractelement <2 x bfloat> %735, i64 1, !dbg !23 + %744 = extractelement <2 x bfloat> %737, i64 0, !dbg !23 + %745 = extractelement <2 x bfloat> %737, i64 1, !dbg !23 + %746 = extractelement <2 x bfloat> %739, i64 0, !dbg !23 + %747 = extractelement <2 x bfloat> %739, i64 1, !dbg !23 + %748 = fpext bfloat %182 to float, !dbg !24 + %749 = fpext bfloat %183 to float, !dbg !24 + %750 = fpext bfloat %184 to float, !dbg !24 + %751 = fpext bfloat %185 to float, !dbg !24 + %752 = fpext bfloat %186 to float, !dbg !24 + %753 = fpext bfloat %187 to float, !dbg !24 + %754 = fpext bfloat %188 to float, !dbg !24 + %755 = fpext bfloat %189 to float, !dbg !24 + %756 = fpext bfloat %200 to float, !dbg !24 + %757 = fpext bfloat %201 to float, !dbg !24 + %758 = fpext bfloat %202 to float, !dbg !24 + %759 = fpext bfloat %203 to float, !dbg !24 + %760 = fpext bfloat %204 to float, !dbg !24 + %761 = fpext bfloat %205 to float, !dbg !24 + %762 = fpext bfloat %206 to float, !dbg !24 + %763 = fpext bfloat %207 to float, !dbg !24 + %764 = fpext bfloat %218 to float, !dbg !24 + %765 = fpext bfloat %219 to float, !dbg !24 + %766 = fpext bfloat %220 to float, !dbg !24 + %767 = fpext bfloat %221 to float, !dbg !24 + %768 = fpext bfloat %222 to float, !dbg !24 + %769 = fpext bfloat %223 to float, !dbg !24 + %770 = fpext bfloat %224 to float, !dbg !24 + %771 = fpext bfloat %225 to float, !dbg !24 + %772 = fpext bfloat %236 to float, !dbg !24 + %773 = fpext bfloat %237 to float, !dbg !24 + %774 = fpext bfloat %238 to float, !dbg !24 + %775 = fpext bfloat %239 to float, !dbg !24 + %776 = fpext bfloat %240 to float, !dbg !24 + %777 = fpext bfloat %241 to float, !dbg !24 + %778 = fpext bfloat %242 to float, !dbg !24 + %779 = fpext bfloat %243 to float, !dbg !24 + %780 = fpext bfloat %254 to float, !dbg !24 + %781 = fpext bfloat %255 to float, !dbg !24 + %782 = fpext bfloat %256 to float, !dbg !24 + %783 = fpext bfloat %257 to float, !dbg !24 + %784 = fpext bfloat %258 to float, !dbg !24 + %785 = fpext bfloat %259 to float, !dbg !24 + %786 = fpext bfloat %260 to float, !dbg !24 + %787 = fpext bfloat %261 to float, !dbg !24 + %788 = fpext bfloat %272 to float, !dbg !24 + %789 = fpext bfloat %273 to float, !dbg !24 + %790 = fpext bfloat %274 to float, !dbg !24 + %791 = fpext bfloat %275 to float, !dbg !24 + %792 = fpext bfloat %276 to float, !dbg !24 + %793 = fpext bfloat %277 to float, !dbg !24 + %794 = fpext bfloat %278 to float, !dbg !24 + %795 = fpext bfloat %279 to float, !dbg !24 + %796 = fpext bfloat %290 to float, !dbg !24 + %797 = fpext bfloat %291 to float, !dbg !24 + %798 = fpext bfloat %292 to float, !dbg !24 + %799 = fpext bfloat %293 to float, !dbg !24 + %800 = fpext bfloat %294 to float, !dbg !24 + %801 = fpext bfloat %295 to float, !dbg !24 + %802 = fpext bfloat %296 to float, !dbg !24 + %803 = fpext bfloat %297 to float, !dbg !24 + %804 = fpext bfloat %308 to float, !dbg !24 + %805 = fpext bfloat %309 to float, !dbg !24 + %806 = fpext bfloat %310 to float, !dbg !24 + %807 = fpext bfloat %311 to float, !dbg !24 + %808 = fpext bfloat %312 to float, !dbg !24 + %809 = fpext bfloat %313 to float, !dbg !24 + %810 = fpext bfloat %314 to float, !dbg !24 + %811 = fpext bfloat %315 to float, !dbg !24 + %812 = fpext bfloat %326 to float, !dbg !24 + %813 = fpext bfloat %327 to float, !dbg !24 + %814 = fpext bfloat %328 to float, !dbg !24 + %815 = fpext bfloat %329 to float, !dbg !24 + %816 = fpext bfloat %330 to float, !dbg !24 + %817 = fpext bfloat %331 to float, !dbg !24 + %818 = fpext bfloat %332 to float, !dbg !24 + %819 = fpext bfloat %333 to float, !dbg !24 + %820 = fpext bfloat %344 to float, !dbg !24 + %821 = fpext bfloat %345 to float, !dbg !24 + %822 = fpext bfloat %346 to float, !dbg !24 + %823 = fpext bfloat %347 to float, !dbg !24 + %824 = fpext bfloat %348 to float, !dbg !24 + %825 = fpext bfloat %349 to float, !dbg !24 + %826 = fpext bfloat %350 to float, !dbg !24 + %827 = fpext bfloat %351 to float, !dbg !24 + %828 = fpext bfloat %362 to float, !dbg !24 + %829 = fpext bfloat %363 to float, !dbg !24 + %830 = fpext bfloat %364 to float, !dbg !24 + %831 = fpext bfloat %365 to float, !dbg !24 + %832 = fpext bfloat %366 to float, !dbg !24 + %833 = fpext bfloat %367 to float, !dbg !24 + %834 = fpext bfloat %368 to float, !dbg !24 + %835 = fpext bfloat %369 to float, !dbg !24 + %836 = fpext bfloat %380 to float, !dbg !24 + %837 = fpext bfloat %381 to float, !dbg !24 + %838 = fpext bfloat %382 to float, !dbg !24 + %839 = fpext bfloat %383 to float, !dbg !24 + %840 = fpext bfloat %384 to float, !dbg !24 + %841 = fpext bfloat %385 to float, !dbg !24 + %842 = fpext bfloat %386 to float, !dbg !24 + %843 = fpext bfloat %387 to float, !dbg !24 + %844 = fpext bfloat %398 to float, !dbg !24 + %845 = fpext bfloat %399 to float, !dbg !24 + %846 = fpext bfloat %400 to float, !dbg !24 + %847 = fpext bfloat %401 to float, !dbg !24 + %848 = fpext bfloat %402 to float, !dbg !24 + %849 = fpext bfloat %403 to float, !dbg !24 + %850 = fpext bfloat %404 to float, !dbg !24 + %851 = fpext bfloat %405 to float, !dbg !24 + %852 = fpext bfloat %416 to float, !dbg !24 + %853 = fpext bfloat %417 to float, !dbg !24 + %854 = fpext bfloat %418 to float, !dbg !24 + %855 = fpext bfloat %419 to float, !dbg !24 + %856 = fpext bfloat %420 to float, !dbg !24 + %857 = fpext bfloat %421 to float, !dbg !24 + %858 = fpext bfloat %422 to float, !dbg !24 + %859 = fpext bfloat %423 to float, !dbg !24 + %860 = fpext bfloat %434 to float, !dbg !24 + %861 = fpext bfloat %435 to float, !dbg !24 + %862 = fpext bfloat %436 to float, !dbg !24 + %863 = fpext bfloat %437 to float, !dbg !24 + %864 = fpext bfloat %438 to float, !dbg !24 + %865 = fpext bfloat %439 to float, !dbg !24 + %866 = fpext bfloat %440 to float, !dbg !24 + %867 = fpext bfloat %441 to float, !dbg !24 + %868 = fpext bfloat %452 to float, !dbg !24 + %869 = fpext bfloat %453 to float, !dbg !24 + %870 = fpext bfloat %454 to float, !dbg !24 + %871 = fpext bfloat %455 to float, !dbg !24 + %872 = fpext bfloat %456 to float, !dbg !24 + %873 = fpext bfloat %457 to float, !dbg !24 + %874 = fpext bfloat %458 to float, !dbg !24 + %875 = fpext bfloat %459 to float, !dbg !24 + %876 = fpext bfloat %470 to float, !dbg !24 + %877 = fpext bfloat %471 to float, !dbg !24 + %878 = fpext bfloat %472 to float, !dbg !24 + %879 = fpext bfloat %473 to float, !dbg !24 + %880 = fpext bfloat %474 to float, !dbg !24 + %881 = fpext bfloat %475 to float, !dbg !24 + %882 = fpext bfloat %476 to float, !dbg !24 + %883 = fpext bfloat %477 to float, !dbg !24 + %884 = fpext bfloat %488 to float, !dbg !24 + %885 = fpext bfloat %489 to float, !dbg !24 + %886 = fpext bfloat %490 to float, !dbg !24 + %887 = fpext bfloat %491 to float, !dbg !24 + %888 = fpext bfloat %492 to float, !dbg !24 + %889 = fpext bfloat %493 to float, !dbg !24 + %890 = fpext bfloat %494 to float, !dbg !24 + %891 = fpext bfloat %495 to float, !dbg !24 + %892 = fpext bfloat %506 to float, !dbg !24 + %893 = fpext bfloat %507 to float, !dbg !24 + %894 = fpext bfloat %508 to float, !dbg !24 + %895 = fpext bfloat %509 to float, !dbg !24 + %896 = fpext bfloat %510 to float, !dbg !24 + %897 = fpext bfloat %511 to float, !dbg !24 + %898 = fpext bfloat %512 to float, !dbg !24 + %899 = fpext bfloat %513 to float, !dbg !24 + %900 = fpext bfloat %524 to float, !dbg !24 + %901 = fpext bfloat %525 to float, !dbg !24 + %902 = fpext bfloat %526 to float, !dbg !24 + %903 = fpext bfloat %527 to float, !dbg !24 + %904 = fpext bfloat %528 to float, !dbg !24 + %905 = fpext bfloat %529 to float, !dbg !24 + %906 = fpext bfloat %530 to float, !dbg !24 + %907 = fpext bfloat %531 to float, !dbg !24 + %908 = fpext bfloat %542 to float, !dbg !24 + %909 = fpext bfloat %543 to float, !dbg !24 + %910 = fpext bfloat %544 to float, !dbg !24 + %911 = fpext bfloat %545 to float, !dbg !24 + %912 = fpext bfloat %546 to float, !dbg !24 + %913 = fpext bfloat %547 to float, !dbg !24 + %914 = fpext bfloat %548 to float, !dbg !24 + %915 = fpext bfloat %549 to float, !dbg !24 + %916 = fpext bfloat %560 to float, !dbg !24 + %917 = fpext bfloat %561 to float, !dbg !24 + %918 = fpext bfloat %562 to float, !dbg !24 + %919 = fpext bfloat %563 to float, !dbg !24 + %920 = fpext bfloat %564 to float, !dbg !24 + %921 = fpext bfloat %565 to float, !dbg !24 + %922 = fpext bfloat %566 to float, !dbg !24 + %923 = fpext bfloat %567 to float, !dbg !24 + %924 = fpext bfloat %578 to float, !dbg !24 + %925 = fpext bfloat %579 to float, !dbg !24 + %926 = fpext bfloat %580 to float, !dbg !24 + %927 = fpext bfloat %581 to float, !dbg !24 + %928 = fpext bfloat %582 to float, !dbg !24 + %929 = fpext bfloat %583 to float, !dbg !24 + %930 = fpext bfloat %584 to float, !dbg !24 + %931 = fpext bfloat %585 to float, !dbg !24 + %932 = fpext bfloat %596 to float, !dbg !24 + %933 = fpext bfloat %597 to float, !dbg !24 + %934 = fpext bfloat %598 to float, !dbg !24 + %935 = fpext bfloat %599 to float, !dbg !24 + %936 = fpext bfloat %600 to float, !dbg !24 + %937 = fpext bfloat %601 to float, !dbg !24 + %938 = fpext bfloat %602 to float, !dbg !24 + %939 = fpext bfloat %603 to float, !dbg !24 + %940 = fpext bfloat %614 to float, !dbg !24 + %941 = fpext bfloat %615 to float, !dbg !24 + %942 = fpext bfloat %616 to float, !dbg !24 + %943 = fpext bfloat %617 to float, !dbg !24 + %944 = fpext bfloat %618 to float, !dbg !24 + %945 = fpext bfloat %619 to float, !dbg !24 + %946 = fpext bfloat %620 to float, !dbg !24 + %947 = fpext bfloat %621 to float, !dbg !24 + %948 = fpext bfloat %632 to float, !dbg !24 + %949 = fpext bfloat %633 to float, !dbg !24 + %950 = fpext bfloat %634 to float, !dbg !24 + %951 = fpext bfloat %635 to float, !dbg !24 + %952 = fpext bfloat %636 to float, !dbg !24 + %953 = fpext bfloat %637 to float, !dbg !24 + %954 = fpext bfloat %638 to float, !dbg !24 + %955 = fpext bfloat %639 to float, !dbg !24 + %956 = fpext bfloat %650 to float, !dbg !24 + %957 = fpext bfloat %651 to float, !dbg !24 + %958 = fpext bfloat %652 to float, !dbg !24 + %959 = fpext bfloat %653 to float, !dbg !24 + %960 = fpext bfloat %654 to float, !dbg !24 + %961 = fpext bfloat %655 to float, !dbg !24 + %962 = fpext bfloat %656 to float, !dbg !24 + %963 = fpext bfloat %657 to float, !dbg !24 + %964 = fpext bfloat %668 to float, !dbg !24 + %965 = fpext bfloat %669 to float, !dbg !24 + %966 = fpext bfloat %670 to float, !dbg !24 + %967 = fpext bfloat %671 to float, !dbg !24 + %968 = fpext bfloat %672 to float, !dbg !24 + %969 = fpext bfloat %673 to float, !dbg !24 + %970 = fpext bfloat %674 to float, !dbg !24 + %971 = fpext bfloat %675 to float, !dbg !24 + %972 = fpext bfloat %686 to float, !dbg !24 + %973 = fpext bfloat %687 to float, !dbg !24 + %974 = fpext bfloat %688 to float, !dbg !24 + %975 = fpext bfloat %689 to float, !dbg !24 + %976 = fpext bfloat %690 to float, !dbg !24 + %977 = fpext bfloat %691 to float, !dbg !24 + %978 = fpext bfloat %692 to float, !dbg !24 + %979 = fpext bfloat %693 to float, !dbg !24 + %980 = fpext bfloat %704 to float, !dbg !24 + %981 = fpext bfloat %705 to float, !dbg !24 + %982 = fpext bfloat %706 to float, !dbg !24 + %983 = fpext bfloat %707 to float, !dbg !24 + %984 = fpext bfloat %708 to float, !dbg !24 + %985 = fpext bfloat %709 to float, !dbg !24 + %986 = fpext bfloat %710 to float, !dbg !24 + %987 = fpext bfloat %711 to float, !dbg !24 + %988 = fpext bfloat %722 to float, !dbg !24 + %989 = fpext bfloat %723 to float, !dbg !24 + %990 = fpext bfloat %724 to float, !dbg !24 + %991 = fpext bfloat %725 to float, !dbg !24 + %992 = fpext bfloat %726 to float, !dbg !24 + %993 = fpext bfloat %727 to float, !dbg !24 + %994 = fpext bfloat %728 to float, !dbg !24 + %995 = fpext bfloat %729 to float, !dbg !24 + %996 = fpext bfloat %740 to float, !dbg !24 + %997 = fpext bfloat %741 to float, !dbg !24 + %998 = fpext bfloat %742 to float, !dbg !24 + %999 = fpext bfloat %743 to float, !dbg !24 + %1000 = fpext bfloat %744 to float, !dbg !24 + %1001 = fpext bfloat %745 to float, !dbg !24 + %1002 = fpext bfloat %746 to float, !dbg !24 + %1003 = fpext bfloat %747 to float, !dbg !24 + %1004 = fmul float %748, %748, !dbg !25 + %1005 = fmul float %749, %749, !dbg !25 + %1006 = fmul float %750, %750, !dbg !25 + %1007 = fmul float %751, %751, !dbg !25 + %1008 = fmul float %752, %752, !dbg !25 + %1009 = fmul float %753, %753, !dbg !25 + %1010 = fmul float %754, %754, !dbg !25 + %1011 = fmul float %755, %755, !dbg !25 + %1012 = fmul float %756, %756, !dbg !25 + %1013 = fmul float %757, %757, !dbg !25 + %1014 = fmul float %758, %758, !dbg !25 + %1015 = fmul float %759, %759, !dbg !25 + %1016 = fmul float %760, %760, !dbg !25 + %1017 = fmul float %761, %761, !dbg !25 + %1018 = fmul float %762, %762, !dbg !25 + %1019 = fmul float %763, %763, !dbg !25 + %1020 = fmul float %764, %764, !dbg !25 + %1021 = fmul float %765, %765, !dbg !25 + %1022 = fmul float %766, %766, !dbg !25 + %1023 = fmul float %767, %767, !dbg !25 + %1024 = fmul float %768, %768, !dbg !25 + %1025 = fmul float %769, %769, !dbg !25 + %1026 = fmul float %770, %770, !dbg !25 + %1027 = fmul float %771, %771, !dbg !25 + %1028 = fmul float %772, %772, !dbg !25 + %1029 = fmul float %773, %773, !dbg !25 + %1030 = fmul float %774, %774, !dbg !25 + %1031 = fmul float %775, %775, !dbg !25 + %1032 = fmul float %776, %776, !dbg !25 + %1033 = fmul float %777, %777, !dbg !25 + %1034 = fmul float %778, %778, !dbg !25 + %1035 = fmul float %779, %779, !dbg !25 + %1036 = fmul float %780, %780, !dbg !25 + %1037 = fmul float %781, %781, !dbg !25 + %1038 = fmul float %782, %782, !dbg !25 + %1039 = fmul float %783, %783, !dbg !25 + %1040 = fmul float %784, %784, !dbg !25 + %1041 = fmul float %785, %785, !dbg !25 + %1042 = fmul float %786, %786, !dbg !25 + %1043 = fmul float %787, %787, !dbg !25 + %1044 = fmul float %788, %788, !dbg !25 + %1045 = fmul float %789, %789, !dbg !25 + %1046 = fmul float %790, %790, !dbg !25 + %1047 = fmul float %791, %791, !dbg !25 + %1048 = fmul float %792, %792, !dbg !25 + %1049 = fmul float %793, %793, !dbg !25 + %1050 = fmul float %794, %794, !dbg !25 + %1051 = fmul float %795, %795, !dbg !25 + %1052 = fmul float %796, %796, !dbg !25 + %1053 = fmul float %797, %797, !dbg !25 + %1054 = fmul float %798, %798, !dbg !25 + %1055 = fmul float %799, %799, !dbg !25 + %1056 = fmul float %800, %800, !dbg !25 + %1057 = fmul float %801, %801, !dbg !25 + %1058 = fmul float %802, %802, !dbg !25 + %1059 = fmul float %803, %803, !dbg !25 + %1060 = fmul float %804, %804, !dbg !25 + %1061 = fmul float %805, %805, !dbg !25 + %1062 = fmul float %806, %806, !dbg !25 + %1063 = fmul float %807, %807, !dbg !25 + %1064 = fmul float %808, %808, !dbg !25 + %1065 = fmul float %809, %809, !dbg !25 + %1066 = fmul float %810, %810, !dbg !25 + %1067 = fmul float %811, %811, !dbg !25 + %1068 = fmul float %812, %812, !dbg !25 + %1069 = fmul float %813, %813, !dbg !25 + %1070 = fmul float %814, %814, !dbg !25 + %1071 = fmul float %815, %815, !dbg !25 + %1072 = fmul float %816, %816, !dbg !25 + %1073 = fmul float %817, %817, !dbg !25 + %1074 = fmul float %818, %818, !dbg !25 + %1075 = fmul float %819, %819, !dbg !25 + %1076 = fmul float %820, %820, !dbg !25 + %1077 = fmul float %821, %821, !dbg !25 + %1078 = fmul float %822, %822, !dbg !25 + %1079 = fmul float %823, %823, !dbg !25 + %1080 = fmul float %824, %824, !dbg !25 + %1081 = fmul float %825, %825, !dbg !25 + %1082 = fmul float %826, %826, !dbg !25 + %1083 = fmul float %827, %827, !dbg !25 + %1084 = fmul float %828, %828, !dbg !25 + %1085 = fmul float %829, %829, !dbg !25 + %1086 = fmul float %830, %830, !dbg !25 + %1087 = fmul float %831, %831, !dbg !25 + %1088 = fmul float %832, %832, !dbg !25 + %1089 = fmul float %833, %833, !dbg !25 + %1090 = fmul float %834, %834, !dbg !25 + %1091 = fmul float %835, %835, !dbg !25 + %1092 = fmul float %836, %836, !dbg !25 + %1093 = fmul float %837, %837, !dbg !25 + %1094 = fmul float %838, %838, !dbg !25 + %1095 = fmul float %839, %839, !dbg !25 + %1096 = fmul float %840, %840, !dbg !25 + %1097 = fmul float %841, %841, !dbg !25 + %1098 = fmul float %842, %842, !dbg !25 + %1099 = fmul float %843, %843, !dbg !25 + %1100 = fmul float %844, %844, !dbg !25 + %1101 = fmul float %845, %845, !dbg !25 + %1102 = fmul float %846, %846, !dbg !25 + %1103 = fmul float %847, %847, !dbg !25 + %1104 = fmul float %848, %848, !dbg !25 + %1105 = fmul float %849, %849, !dbg !25 + %1106 = fmul float %850, %850, !dbg !25 + %1107 = fmul float %851, %851, !dbg !25 + %1108 = fmul float %852, %852, !dbg !25 + %1109 = fmul float %853, %853, !dbg !25 + %1110 = fmul float %854, %854, !dbg !25 + %1111 = fmul float %855, %855, !dbg !25 + %1112 = fmul float %856, %856, !dbg !25 + %1113 = fmul float %857, %857, !dbg !25 + %1114 = fmul float %858, %858, !dbg !25 + %1115 = fmul float %859, %859, !dbg !25 + %1116 = fmul float %860, %860, !dbg !25 + %1117 = fmul float %861, %861, !dbg !25 + %1118 = fmul float %862, %862, !dbg !25 + %1119 = fmul float %863, %863, !dbg !25 + %1120 = fmul float %864, %864, !dbg !25 + %1121 = fmul float %865, %865, !dbg !25 + %1122 = fmul float %866, %866, !dbg !25 + %1123 = fmul float %867, %867, !dbg !25 + %1124 = fmul float %868, %868, !dbg !25 + %1125 = fmul float %869, %869, !dbg !25 + %1126 = fmul float %870, %870, !dbg !25 + %1127 = fmul float %871, %871, !dbg !25 + %1128 = fmul float %872, %872, !dbg !25 + %1129 = fmul float %873, %873, !dbg !25 + %1130 = fmul float %874, %874, !dbg !25 + %1131 = fmul float %875, %875, !dbg !25 + %1132 = fmul float %876, %876, !dbg !25 + %1133 = fmul float %877, %877, !dbg !25 + %1134 = fmul float %878, %878, !dbg !25 + %1135 = fmul float %879, %879, !dbg !25 + %1136 = fmul float %880, %880, !dbg !25 + %1137 = fmul float %881, %881, !dbg !25 + %1138 = fmul float %882, %882, !dbg !25 + %1139 = fmul float %883, %883, !dbg !25 + %1140 = fmul float %884, %884, !dbg !25 + %1141 = fmul float %885, %885, !dbg !25 + %1142 = fmul float %886, %886, !dbg !25 + %1143 = fmul float %887, %887, !dbg !25 + %1144 = fmul float %888, %888, !dbg !25 + %1145 = fmul float %889, %889, !dbg !25 + %1146 = fmul float %890, %890, !dbg !25 + %1147 = fmul float %891, %891, !dbg !25 + %1148 = fmul float %892, %892, !dbg !25 + %1149 = fmul float %893, %893, !dbg !25 + %1150 = fmul float %894, %894, !dbg !25 + %1151 = fmul float %895, %895, !dbg !25 + %1152 = fmul float %896, %896, !dbg !25 + %1153 = fmul float %897, %897, !dbg !25 + %1154 = fmul float %898, %898, !dbg !25 + %1155 = fmul float %899, %899, !dbg !25 + %1156 = fmul float %900, %900, !dbg !25 + %1157 = fmul float %901, %901, !dbg !25 + %1158 = fmul float %902, %902, !dbg !25 + %1159 = fmul float %903, %903, !dbg !25 + %1160 = fmul float %904, %904, !dbg !25 + %1161 = fmul float %905, %905, !dbg !25 + %1162 = fmul float %906, %906, !dbg !25 + %1163 = fmul float %907, %907, !dbg !25 + %1164 = fmul float %908, %908, !dbg !25 + %1165 = fmul float %909, %909, !dbg !25 + %1166 = fmul float %910, %910, !dbg !25 + %1167 = fmul float %911, %911, !dbg !25 + %1168 = fmul float %912, %912, !dbg !25 + %1169 = fmul float %913, %913, !dbg !25 + %1170 = fmul float %914, %914, !dbg !25 + %1171 = fmul float %915, %915, !dbg !25 + %1172 = fmul float %916, %916, !dbg !25 + %1173 = fmul float %917, %917, !dbg !25 + %1174 = fmul float %918, %918, !dbg !25 + %1175 = fmul float %919, %919, !dbg !25 + %1176 = fmul float %920, %920, !dbg !25 + %1177 = fmul float %921, %921, !dbg !25 + %1178 = fmul float %922, %922, !dbg !25 + %1179 = fmul float %923, %923, !dbg !25 + %1180 = fmul float %924, %924, !dbg !25 + %1181 = fmul float %925, %925, !dbg !25 + %1182 = fmul float %926, %926, !dbg !25 + %1183 = fmul float %927, %927, !dbg !25 + %1184 = fmul float %928, %928, !dbg !25 + %1185 = fmul float %929, %929, !dbg !25 + %1186 = fmul float %930, %930, !dbg !25 + %1187 = fmul float %931, %931, !dbg !25 + %1188 = fmul float %932, %932, !dbg !25 + %1189 = fmul float %933, %933, !dbg !25 + %1190 = fmul float %934, %934, !dbg !25 + %1191 = fmul float %935, %935, !dbg !25 + %1192 = fmul float %936, %936, !dbg !25 + %1193 = fmul float %937, %937, !dbg !25 + %1194 = fmul float %938, %938, !dbg !25 + %1195 = fmul float %939, %939, !dbg !25 + %1196 = fmul float %940, %940, !dbg !25 + %1197 = fmul float %941, %941, !dbg !25 + %1198 = fmul float %942, %942, !dbg !25 + %1199 = fmul float %943, %943, !dbg !25 + %1200 = fmul float %944, %944, !dbg !25 + %1201 = fmul float %945, %945, !dbg !25 + %1202 = fmul float %946, %946, !dbg !25 + %1203 = fmul float %947, %947, !dbg !25 + %1204 = fmul float %948, %948, !dbg !25 + %1205 = fmul float %949, %949, !dbg !25 + %1206 = fmul float %950, %950, !dbg !25 + %1207 = fmul float %951, %951, !dbg !25 + %1208 = fmul float %952, %952, !dbg !25 + %1209 = fmul float %953, %953, !dbg !25 + %1210 = fmul float %954, %954, !dbg !25 + %1211 = fmul float %955, %955, !dbg !25 + %1212 = fmul float %956, %956, !dbg !25 + %1213 = fmul float %957, %957, !dbg !25 + %1214 = fmul float %958, %958, !dbg !25 + %1215 = fmul float %959, %959, !dbg !25 + %1216 = fmul float %960, %960, !dbg !25 + %1217 = fmul float %961, %961, !dbg !25 + %1218 = fmul float %962, %962, !dbg !25 + %1219 = fmul float %963, %963, !dbg !25 + %1220 = fmul float %964, %964, !dbg !25 + %1221 = fmul float %965, %965, !dbg !25 + %1222 = fmul float %966, %966, !dbg !25 + %1223 = fmul float %967, %967, !dbg !25 + %1224 = fmul float %968, %968, !dbg !25 + %1225 = fmul float %969, %969, !dbg !25 + %1226 = fmul float %970, %970, !dbg !25 + %1227 = fmul float %971, %971, !dbg !25 + %1228 = fmul float %972, %972, !dbg !25 + %1229 = fmul float %973, %973, !dbg !25 + %1230 = fmul float %974, %974, !dbg !25 + %1231 = fmul float %975, %975, !dbg !25 + %1232 = fmul float %976, %976, !dbg !25 + %1233 = fmul float %977, %977, !dbg !25 + %1234 = fmul float %978, %978, !dbg !25 + %1235 = fmul float %979, %979, !dbg !25 + %1236 = fmul float %980, %980, !dbg !25 + %1237 = fmul float %981, %981, !dbg !25 + %1238 = fmul float %982, %982, !dbg !25 + %1239 = fmul float %983, %983, !dbg !25 + %1240 = fmul float %984, %984, !dbg !25 + %1241 = fmul float %985, %985, !dbg !25 + %1242 = fmul float %986, %986, !dbg !25 + %1243 = fmul float %987, %987, !dbg !25 + %1244 = fmul float %988, %988, !dbg !25 + %1245 = fmul float %989, %989, !dbg !25 + %1246 = fmul float %990, %990, !dbg !25 + %1247 = fmul float %991, %991, !dbg !25 + %1248 = fmul float %992, %992, !dbg !25 + %1249 = fmul float %993, %993, !dbg !25 + %1250 = fmul float %994, %994, !dbg !25 + %1251 = fmul float %995, %995, !dbg !25 + %1252 = fmul float %996, %996, !dbg !25 + %1253 = fmul float %997, %997, !dbg !25 + %1254 = fmul float %998, %998, !dbg !25 + %1255 = fmul float %999, %999, !dbg !25 + %1256 = fmul float %1000, %1000, !dbg !25 + %1257 = fmul float %1001, %1001, !dbg !25 + %1258 = fmul float %1002, %1002, !dbg !25 + %1259 = fmul float %1003, %1003, !dbg !25 + %1260 = fadd float %1004, %1005, !dbg !26 + %1261 = fadd float %1006, %1260, !dbg !26 + %1262 = fadd float %1007, %1261, !dbg !26 + %1263 = fadd float %1008, %1262, !dbg !26 + %1264 = fadd float %1009, %1263, !dbg !26 + %1265 = fadd float %1010, %1264, !dbg !26 + %1266 = fadd float %1011, %1265, !dbg !26 + %1267 = fadd float %1012, %1013, !dbg !26 + %1268 = fadd float %1014, %1267, !dbg !26 + %1269 = fadd float %1015, %1268, !dbg !26 + %1270 = fadd float %1016, %1269, !dbg !26 + %1271 = fadd float %1017, %1270, !dbg !26 + %1272 = fadd float %1018, %1271, !dbg !26 + %1273 = fadd float %1019, %1272, !dbg !26 + %1274 = fadd float %1020, %1021, !dbg !26 + %1275 = fadd float %1022, %1274, !dbg !26 + %1276 = fadd float %1023, %1275, !dbg !26 + %1277 = fadd float %1024, %1276, !dbg !26 + %1278 = fadd float %1025, %1277, !dbg !26 + %1279 = fadd float %1026, %1278, !dbg !26 + %1280 = fadd float %1027, %1279, !dbg !26 + %1281 = fadd float %1028, %1029, !dbg !26 + %1282 = fadd float %1030, %1281, !dbg !26 + %1283 = fadd float %1031, %1282, !dbg !26 + %1284 = fadd float %1032, %1283, !dbg !26 + %1285 = fadd float %1033, %1284, !dbg !26 + %1286 = fadd float %1034, %1285, !dbg !26 + %1287 = fadd float %1035, %1286, !dbg !26 + %1288 = fadd float %1036, %1037, !dbg !26 + %1289 = fadd float %1038, %1288, !dbg !26 + %1290 = fadd float %1039, %1289, !dbg !26 + %1291 = fadd float %1040, %1290, !dbg !26 + %1292 = fadd float %1041, %1291, !dbg !26 + %1293 = fadd float %1042, %1292, !dbg !26 + %1294 = fadd float %1043, %1293, !dbg !26 + %1295 = fadd float %1044, %1045, !dbg !26 + %1296 = fadd float %1046, %1295, !dbg !26 + %1297 = fadd float %1047, %1296, !dbg !26 + %1298 = fadd float %1048, %1297, !dbg !26 + %1299 = fadd float %1049, %1298, !dbg !26 + %1300 = fadd float %1050, %1299, !dbg !26 + %1301 = fadd float %1051, %1300, !dbg !26 + %1302 = fadd float %1052, %1053, !dbg !26 + %1303 = fadd float %1054, %1302, !dbg !26 + %1304 = fadd float %1055, %1303, !dbg !26 + %1305 = fadd float %1056, %1304, !dbg !26 + %1306 = fadd float %1057, %1305, !dbg !26 + %1307 = fadd float %1058, %1306, !dbg !26 + %1308 = fadd float %1059, %1307, !dbg !26 + %1309 = fadd float %1060, %1061, !dbg !26 + %1310 = fadd float %1062, %1309, !dbg !26 + %1311 = fadd float %1063, %1310, !dbg !26 + %1312 = fadd float %1064, %1311, !dbg !26 + %1313 = fadd float %1065, %1312, !dbg !26 + %1314 = fadd float %1066, %1313, !dbg !26 + %1315 = fadd float %1067, %1314, !dbg !26 + %1316 = fadd float %1068, %1069, !dbg !26 + %1317 = fadd float %1070, %1316, !dbg !26 + %1318 = fadd float %1071, %1317, !dbg !26 + %1319 = fadd float %1072, %1318, !dbg !26 + %1320 = fadd float %1073, %1319, !dbg !26 + %1321 = fadd float %1074, %1320, !dbg !26 + %1322 = fadd float %1075, %1321, !dbg !26 + %1323 = fadd float %1076, %1077, !dbg !26 + %1324 = fadd float %1078, %1323, !dbg !26 + %1325 = fadd float %1079, %1324, !dbg !26 + %1326 = fadd float %1080, %1325, !dbg !26 + %1327 = fadd float %1081, %1326, !dbg !26 + %1328 = fadd float %1082, %1327, !dbg !26 + %1329 = fadd float %1083, %1328, !dbg !26 + %1330 = fadd float %1084, %1085, !dbg !26 + %1331 = fadd float %1086, %1330, !dbg !26 + %1332 = fadd float %1087, %1331, !dbg !26 + %1333 = fadd float %1088, %1332, !dbg !26 + %1334 = fadd float %1089, %1333, !dbg !26 + %1335 = fadd float %1090, %1334, !dbg !26 + %1336 = fadd float %1091, %1335, !dbg !26 + %1337 = fadd float %1092, %1093, !dbg !26 + %1338 = fadd float %1094, %1337, !dbg !26 + %1339 = fadd float %1095, %1338, !dbg !26 + %1340 = fadd float %1096, %1339, !dbg !26 + %1341 = fadd float %1097, %1340, !dbg !26 + %1342 = fadd float %1098, %1341, !dbg !26 + %1343 = fadd float %1099, %1342, !dbg !26 + %1344 = fadd float %1100, %1101, !dbg !26 + %1345 = fadd float %1102, %1344, !dbg !26 + %1346 = fadd float %1103, %1345, !dbg !26 + %1347 = fadd float %1104, %1346, !dbg !26 + %1348 = fadd float %1105, %1347, !dbg !26 + %1349 = fadd float %1106, %1348, !dbg !26 + %1350 = fadd float %1107, %1349, !dbg !26 + %1351 = fadd float %1108, %1109, !dbg !26 + %1352 = fadd float %1110, %1351, !dbg !26 + %1353 = fadd float %1111, %1352, !dbg !26 + %1354 = fadd float %1112, %1353, !dbg !26 + %1355 = fadd float %1113, %1354, !dbg !26 + %1356 = fadd float %1114, %1355, !dbg !26 + %1357 = fadd float %1115, %1356, !dbg !26 + %1358 = fadd float %1116, %1117, !dbg !26 + %1359 = fadd float %1118, %1358, !dbg !26 + %1360 = fadd float %1119, %1359, !dbg !26 + %1361 = fadd float %1120, %1360, !dbg !26 + %1362 = fadd float %1121, %1361, !dbg !26 + %1363 = fadd float %1122, %1362, !dbg !26 + %1364 = fadd float %1123, %1363, !dbg !26 + %1365 = fadd float %1124, %1125, !dbg !26 + %1366 = fadd float %1126, %1365, !dbg !26 + %1367 = fadd float %1127, %1366, !dbg !26 + %1368 = fadd float %1128, %1367, !dbg !26 + %1369 = fadd float %1129, %1368, !dbg !26 + %1370 = fadd float %1130, %1369, !dbg !26 + %1371 = fadd float %1131, %1370, !dbg !26 + %1372 = fadd float %1132, %1133, !dbg !26 + %1373 = fadd float %1134, %1372, !dbg !26 + %1374 = fadd float %1135, %1373, !dbg !26 + %1375 = fadd float %1136, %1374, !dbg !26 + %1376 = fadd float %1137, %1375, !dbg !26 + %1377 = fadd float %1138, %1376, !dbg !26 + %1378 = fadd float %1139, %1377, !dbg !26 + %1379 = fadd float %1140, %1141, !dbg !26 + %1380 = fadd float %1142, %1379, !dbg !26 + %1381 = fadd float %1143, %1380, !dbg !26 + %1382 = fadd float %1144, %1381, !dbg !26 + %1383 = fadd float %1145, %1382, !dbg !26 + %1384 = fadd float %1146, %1383, !dbg !26 + %1385 = fadd float %1147, %1384, !dbg !26 + %1386 = fadd float %1148, %1149, !dbg !26 + %1387 = fadd float %1150, %1386, !dbg !26 + %1388 = fadd float %1151, %1387, !dbg !26 + %1389 = fadd float %1152, %1388, !dbg !26 + %1390 = fadd float %1153, %1389, !dbg !26 + %1391 = fadd float %1154, %1390, !dbg !26 + %1392 = fadd float %1155, %1391, !dbg !26 + %1393 = fadd float %1156, %1157, !dbg !26 + %1394 = fadd float %1158, %1393, !dbg !26 + %1395 = fadd float %1159, %1394, !dbg !26 + %1396 = fadd float %1160, %1395, !dbg !26 + %1397 = fadd float %1161, %1396, !dbg !26 + %1398 = fadd float %1162, %1397, !dbg !26 + %1399 = fadd float %1163, %1398, !dbg !26 + %1400 = fadd float %1164, %1165, !dbg !26 + %1401 = fadd float %1166, %1400, !dbg !26 + %1402 = fadd float %1167, %1401, !dbg !26 + %1403 = fadd float %1168, %1402, !dbg !26 + %1404 = fadd float %1169, %1403, !dbg !26 + %1405 = fadd float %1170, %1404, !dbg !26 + %1406 = fadd float %1171, %1405, !dbg !26 + %1407 = fadd float %1172, %1173, !dbg !26 + %1408 = fadd float %1174, %1407, !dbg !26 + %1409 = fadd float %1175, %1408, !dbg !26 + %1410 = fadd float %1176, %1409, !dbg !26 + %1411 = fadd float %1177, %1410, !dbg !26 + %1412 = fadd float %1178, %1411, !dbg !26 + %1413 = fadd float %1179, %1412, !dbg !26 + %1414 = fadd float %1180, %1181, !dbg !26 + %1415 = fadd float %1182, %1414, !dbg !26 + %1416 = fadd float %1183, %1415, !dbg !26 + %1417 = fadd float %1184, %1416, !dbg !26 + %1418 = fadd float %1185, %1417, !dbg !26 + %1419 = fadd float %1186, %1418, !dbg !26 + %1420 = fadd float %1187, %1419, !dbg !26 + %1421 = fadd float %1188, %1189, !dbg !26 + %1422 = fadd float %1190, %1421, !dbg !26 + %1423 = fadd float %1191, %1422, !dbg !26 + %1424 = fadd float %1192, %1423, !dbg !26 + %1425 = fadd float %1193, %1424, !dbg !26 + %1426 = fadd float %1194, %1425, !dbg !26 + %1427 = fadd float %1195, %1426, !dbg !26 + %1428 = fadd float %1196, %1197, !dbg !26 + %1429 = fadd float %1198, %1428, !dbg !26 + %1430 = fadd float %1199, %1429, !dbg !26 + %1431 = fadd float %1200, %1430, !dbg !26 + %1432 = fadd float %1201, %1431, !dbg !26 + %1433 = fadd float %1202, %1432, !dbg !26 + %1434 = fadd float %1203, %1433, !dbg !26 + %1435 = fadd float %1204, %1205, !dbg !26 + %1436 = fadd float %1206, %1435, !dbg !26 + %1437 = fadd float %1207, %1436, !dbg !26 + %1438 = fadd float %1208, %1437, !dbg !26 + %1439 = fadd float %1209, %1438, !dbg !26 + %1440 = fadd float %1210, %1439, !dbg !26 + %1441 = fadd float %1211, %1440, !dbg !26 + %1442 = fadd float %1212, %1213, !dbg !26 + %1443 = fadd float %1214, %1442, !dbg !26 + %1444 = fadd float %1215, %1443, !dbg !26 + %1445 = fadd float %1216, %1444, !dbg !26 + %1446 = fadd float %1217, %1445, !dbg !26 + %1447 = fadd float %1218, %1446, !dbg !26 + %1448 = fadd float %1219, %1447, !dbg !26 + %1449 = fadd float %1220, %1221, !dbg !26 + %1450 = fadd float %1222, %1449, !dbg !26 + %1451 = fadd float %1223, %1450, !dbg !26 + %1452 = fadd float %1224, %1451, !dbg !26 + %1453 = fadd float %1225, %1452, !dbg !26 + %1454 = fadd float %1226, %1453, !dbg !26 + %1455 = fadd float %1227, %1454, !dbg !26 + %1456 = fadd float %1228, %1229, !dbg !26 + %1457 = fadd float %1230, %1456, !dbg !26 + %1458 = fadd float %1231, %1457, !dbg !26 + %1459 = fadd float %1232, %1458, !dbg !26 + %1460 = fadd float %1233, %1459, !dbg !26 + %1461 = fadd float %1234, %1460, !dbg !26 + %1462 = fadd float %1235, %1461, !dbg !26 + %1463 = fadd float %1236, %1237, !dbg !26 + %1464 = fadd float %1238, %1463, !dbg !26 + %1465 = fadd float %1239, %1464, !dbg !26 + %1466 = fadd float %1240, %1465, !dbg !26 + %1467 = fadd float %1241, %1466, !dbg !26 + %1468 = fadd float %1242, %1467, !dbg !26 + %1469 = fadd float %1243, %1468, !dbg !26 + %1470 = fadd float %1244, %1245, !dbg !26 + %1471 = fadd float %1246, %1470, !dbg !26 + %1472 = fadd float %1247, %1471, !dbg !26 + %1473 = fadd float %1248, %1472, !dbg !26 + %1474 = fadd float %1249, %1473, !dbg !26 + %1475 = fadd float %1250, %1474, !dbg !26 + %1476 = fadd float %1251, %1475, !dbg !26 + %1477 = fadd float %1252, %1253, !dbg !26 + %1478 = fadd float %1254, %1477, !dbg !26 + %1479 = fadd float %1255, %1478, !dbg !26 + %1480 = fadd float %1256, %1479, !dbg !26 + %1481 = fadd float %1257, %1480, !dbg !26 + %1482 = fadd float %1258, %1481, !dbg !26 + %1483 = fadd float %1259, %1482, !dbg !26 + %1484 = bitcast float %1266 to i32, !dbg !29 + %1485 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1484, i32 8, i32 31), !dbg !29 + %1486 = bitcast i32 %1485 to float, !dbg !29 + %1487 = fadd float %1266, %1486, !dbg !26 + %1488 = bitcast float %1487 to i32, !dbg !29 + %1489 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1488, i32 4, i32 31), !dbg !29 + %1490 = bitcast i32 %1489 to float, !dbg !29 + %1491 = fadd float %1487, %1490, !dbg !26 + %1492 = bitcast float %1491 to i32, !dbg !29 + %1493 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1492, i32 2, i32 31), !dbg !29 + %1494 = bitcast i32 %1493 to float, !dbg !29 + %1495 = fadd float %1491, %1494, !dbg !26 + %1496 = bitcast float %1495 to i32, !dbg !29 + %1497 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1496, i32 1, i32 31), !dbg !29 + %1498 = bitcast i32 %1497 to float, !dbg !29 + %1499 = fadd float %1495, %1498, !dbg !26 + %1500 = bitcast float %1273 to i32, !dbg !29 + %1501 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1500, i32 8, i32 31), !dbg !29 + %1502 = bitcast i32 %1501 to float, !dbg !29 + %1503 = fadd float %1273, %1502, !dbg !26 + %1504 = bitcast float %1503 to i32, !dbg !29 + %1505 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1504, i32 4, i32 31), !dbg !29 + %1506 = bitcast i32 %1505 to float, !dbg !29 + %1507 = fadd float %1503, %1506, !dbg !26 + %1508 = bitcast float %1507 to i32, !dbg !29 + %1509 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1508, i32 2, i32 31), !dbg !29 + %1510 = bitcast i32 %1509 to float, !dbg !29 + %1511 = fadd float %1507, %1510, !dbg !26 + %1512 = bitcast float %1511 to i32, !dbg !29 + %1513 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1512, i32 1, i32 31), !dbg !29 + %1514 = bitcast i32 %1513 to float, !dbg !29 + %1515 = fadd float %1511, %1514, !dbg !26 + %1516 = bitcast float %1280 to i32, !dbg !29 + %1517 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1516, i32 8, i32 31), !dbg !29 + %1518 = bitcast i32 %1517 to float, !dbg !29 + %1519 = fadd float %1280, %1518, !dbg !26 + %1520 = bitcast float %1519 to i32, !dbg !29 + %1521 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1520, i32 4, i32 31), !dbg !29 + %1522 = bitcast i32 %1521 to float, !dbg !29 + %1523 = fadd float %1519, %1522, !dbg !26 + %1524 = bitcast float %1523 to i32, !dbg !29 + %1525 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1524, i32 2, i32 31), !dbg !29 + %1526 = bitcast i32 %1525 to float, !dbg !29 + %1527 = fadd float %1523, %1526, !dbg !26 + %1528 = bitcast float %1527 to i32, !dbg !29 + %1529 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1528, i32 1, i32 31), !dbg !29 + %1530 = bitcast i32 %1529 to float, !dbg !29 + %1531 = fadd float %1527, %1530, !dbg !26 + %1532 = bitcast float %1287 to i32, !dbg !29 + %1533 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1532, i32 8, i32 31), !dbg !29 + %1534 = bitcast i32 %1533 to float, !dbg !29 + %1535 = fadd float %1287, %1534, !dbg !26 + %1536 = bitcast float %1535 to i32, !dbg !29 + %1537 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1536, i32 4, i32 31), !dbg !29 + %1538 = bitcast i32 %1537 to float, !dbg !29 + %1539 = fadd float %1535, %1538, !dbg !26 + %1540 = bitcast float %1539 to i32, !dbg !29 + %1541 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1540, i32 2, i32 31), !dbg !29 + %1542 = bitcast i32 %1541 to float, !dbg !29 + %1543 = fadd float %1539, %1542, !dbg !26 + %1544 = bitcast float %1543 to i32, !dbg !29 + %1545 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1544, i32 1, i32 31), !dbg !29 + %1546 = bitcast i32 %1545 to float, !dbg !29 + %1547 = fadd float %1543, %1546, !dbg !26 + %1548 = bitcast float %1294 to i32, !dbg !29 + %1549 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1548, i32 8, i32 31), !dbg !29 + %1550 = bitcast i32 %1549 to float, !dbg !29 + %1551 = fadd float %1294, %1550, !dbg !26 + %1552 = bitcast float %1551 to i32, !dbg !29 + %1553 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1552, i32 4, i32 31), !dbg !29 + %1554 = bitcast i32 %1553 to float, !dbg !29 + %1555 = fadd float %1551, %1554, !dbg !26 + %1556 = bitcast float %1555 to i32, !dbg !29 + %1557 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1556, i32 2, i32 31), !dbg !29 + %1558 = bitcast i32 %1557 to float, !dbg !29 + %1559 = fadd float %1555, %1558, !dbg !26 + %1560 = bitcast float %1559 to i32, !dbg !29 + %1561 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1560, i32 1, i32 31), !dbg !29 + %1562 = bitcast i32 %1561 to float, !dbg !29 + %1563 = fadd float %1559, %1562, !dbg !26 + %1564 = bitcast float %1301 to i32, !dbg !29 + %1565 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1564, i32 8, i32 31), !dbg !29 + %1566 = bitcast i32 %1565 to float, !dbg !29 + %1567 = fadd float %1301, %1566, !dbg !26 + %1568 = bitcast float %1567 to i32, !dbg !29 + %1569 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1568, i32 4, i32 31), !dbg !29 + %1570 = bitcast i32 %1569 to float, !dbg !29 + %1571 = fadd float %1567, %1570, !dbg !26 + %1572 = bitcast float %1571 to i32, !dbg !29 + %1573 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1572, i32 2, i32 31), !dbg !29 + %1574 = bitcast i32 %1573 to float, !dbg !29 + %1575 = fadd float %1571, %1574, !dbg !26 + %1576 = bitcast float %1575 to i32, !dbg !29 + %1577 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1576, i32 1, i32 31), !dbg !29 + %1578 = bitcast i32 %1577 to float, !dbg !29 + %1579 = fadd float %1575, %1578, !dbg !26 + %1580 = bitcast float %1308 to i32, !dbg !29 + %1581 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1580, i32 8, i32 31), !dbg !29 + %1582 = bitcast i32 %1581 to float, !dbg !29 + %1583 = fadd float %1308, %1582, !dbg !26 + %1584 = bitcast float %1583 to i32, !dbg !29 + %1585 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1584, i32 4, i32 31), !dbg !29 + %1586 = bitcast i32 %1585 to float, !dbg !29 + %1587 = fadd float %1583, %1586, !dbg !26 + %1588 = bitcast float %1587 to i32, !dbg !29 + %1589 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1588, i32 2, i32 31), !dbg !29 + %1590 = bitcast i32 %1589 to float, !dbg !29 + %1591 = fadd float %1587, %1590, !dbg !26 + %1592 = bitcast float %1591 to i32, !dbg !29 + %1593 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1592, i32 1, i32 31), !dbg !29 + %1594 = bitcast i32 %1593 to float, !dbg !29 + %1595 = fadd float %1591, %1594, !dbg !26 + %1596 = bitcast float %1315 to i32, !dbg !29 + %1597 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1596, i32 8, i32 31), !dbg !29 + %1598 = bitcast i32 %1597 to float, !dbg !29 + %1599 = fadd float %1315, %1598, !dbg !26 + %1600 = bitcast float %1599 to i32, !dbg !29 + %1601 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1600, i32 4, i32 31), !dbg !29 + %1602 = bitcast i32 %1601 to float, !dbg !29 + %1603 = fadd float %1599, %1602, !dbg !26 + %1604 = bitcast float %1603 to i32, !dbg !29 + %1605 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1604, i32 2, i32 31), !dbg !29 + %1606 = bitcast i32 %1605 to float, !dbg !29 + %1607 = fadd float %1603, %1606, !dbg !26 + %1608 = bitcast float %1607 to i32, !dbg !29 + %1609 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1608, i32 1, i32 31), !dbg !29 + %1610 = bitcast i32 %1609 to float, !dbg !29 + %1611 = fadd float %1607, %1610, !dbg !26 + %1612 = bitcast float %1322 to i32, !dbg !29 + %1613 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1612, i32 8, i32 31), !dbg !29 + %1614 = bitcast i32 %1613 to float, !dbg !29 + %1615 = fadd float %1322, %1614, !dbg !26 + %1616 = bitcast float %1615 to i32, !dbg !29 + %1617 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1616, i32 4, i32 31), !dbg !29 + %1618 = bitcast i32 %1617 to float, !dbg !29 + %1619 = fadd float %1615, %1618, !dbg !26 + %1620 = bitcast float %1619 to i32, !dbg !29 + %1621 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1620, i32 2, i32 31), !dbg !29 + %1622 = bitcast i32 %1621 to float, !dbg !29 + %1623 = fadd float %1619, %1622, !dbg !26 + %1624 = bitcast float %1623 to i32, !dbg !29 + %1625 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1624, i32 1, i32 31), !dbg !29 + %1626 = bitcast i32 %1625 to float, !dbg !29 + %1627 = fadd float %1623, %1626, !dbg !26 + %1628 = bitcast float %1329 to i32, !dbg !29 + %1629 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1628, i32 8, i32 31), !dbg !29 + %1630 = bitcast i32 %1629 to float, !dbg !29 + %1631 = fadd float %1329, %1630, !dbg !26 + %1632 = bitcast float %1631 to i32, !dbg !29 + %1633 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1632, i32 4, i32 31), !dbg !29 + %1634 = bitcast i32 %1633 to float, !dbg !29 + %1635 = fadd float %1631, %1634, !dbg !26 + %1636 = bitcast float %1635 to i32, !dbg !29 + %1637 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1636, i32 2, i32 31), !dbg !29 + %1638 = bitcast i32 %1637 to float, !dbg !29 + %1639 = fadd float %1635, %1638, !dbg !26 + %1640 = bitcast float %1639 to i32, !dbg !29 + %1641 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1640, i32 1, i32 31), !dbg !29 + %1642 = bitcast i32 %1641 to float, !dbg !29 + %1643 = fadd float %1639, %1642, !dbg !26 + %1644 = bitcast float %1336 to i32, !dbg !29 + %1645 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1644, i32 8, i32 31), !dbg !29 + %1646 = bitcast i32 %1645 to float, !dbg !29 + %1647 = fadd float %1336, %1646, !dbg !26 + %1648 = bitcast float %1647 to i32, !dbg !29 + %1649 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1648, i32 4, i32 31), !dbg !29 + %1650 = bitcast i32 %1649 to float, !dbg !29 + %1651 = fadd float %1647, %1650, !dbg !26 + %1652 = bitcast float %1651 to i32, !dbg !29 + %1653 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1652, i32 2, i32 31), !dbg !29 + %1654 = bitcast i32 %1653 to float, !dbg !29 + %1655 = fadd float %1651, %1654, !dbg !26 + %1656 = bitcast float %1655 to i32, !dbg !29 + %1657 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1656, i32 1, i32 31), !dbg !29 + %1658 = bitcast i32 %1657 to float, !dbg !29 + %1659 = fadd float %1655, %1658, !dbg !26 + %1660 = bitcast float %1343 to i32, !dbg !29 + %1661 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1660, i32 8, i32 31), !dbg !29 + %1662 = bitcast i32 %1661 to float, !dbg !29 + %1663 = fadd float %1343, %1662, !dbg !26 + %1664 = bitcast float %1663 to i32, !dbg !29 + %1665 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1664, i32 4, i32 31), !dbg !29 + %1666 = bitcast i32 %1665 to float, !dbg !29 + %1667 = fadd float %1663, %1666, !dbg !26 + %1668 = bitcast float %1667 to i32, !dbg !29 + %1669 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1668, i32 2, i32 31), !dbg !29 + %1670 = bitcast i32 %1669 to float, !dbg !29 + %1671 = fadd float %1667, %1670, !dbg !26 + %1672 = bitcast float %1671 to i32, !dbg !29 + %1673 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1672, i32 1, i32 31), !dbg !29 + %1674 = bitcast i32 %1673 to float, !dbg !29 + %1675 = fadd float %1671, %1674, !dbg !26 + %1676 = bitcast float %1350 to i32, !dbg !29 + %1677 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1676, i32 8, i32 31), !dbg !29 + %1678 = bitcast i32 %1677 to float, !dbg !29 + %1679 = fadd float %1350, %1678, !dbg !26 + %1680 = bitcast float %1679 to i32, !dbg !29 + %1681 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1680, i32 4, i32 31), !dbg !29 + %1682 = bitcast i32 %1681 to float, !dbg !29 + %1683 = fadd float %1679, %1682, !dbg !26 + %1684 = bitcast float %1683 to i32, !dbg !29 + %1685 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1684, i32 2, i32 31), !dbg !29 + %1686 = bitcast i32 %1685 to float, !dbg !29 + %1687 = fadd float %1683, %1686, !dbg !26 + %1688 = bitcast float %1687 to i32, !dbg !29 + %1689 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1688, i32 1, i32 31), !dbg !29 + %1690 = bitcast i32 %1689 to float, !dbg !29 + %1691 = fadd float %1687, %1690, !dbg !26 + %1692 = bitcast float %1357 to i32, !dbg !29 + %1693 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1692, i32 8, i32 31), !dbg !29 + %1694 = bitcast i32 %1693 to float, !dbg !29 + %1695 = fadd float %1357, %1694, !dbg !26 + %1696 = bitcast float %1695 to i32, !dbg !29 + %1697 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1696, i32 4, i32 31), !dbg !29 + %1698 = bitcast i32 %1697 to float, !dbg !29 + %1699 = fadd float %1695, %1698, !dbg !26 + %1700 = bitcast float %1699 to i32, !dbg !29 + %1701 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1700, i32 2, i32 31), !dbg !29 + %1702 = bitcast i32 %1701 to float, !dbg !29 + %1703 = fadd float %1699, %1702, !dbg !26 + %1704 = bitcast float %1703 to i32, !dbg !29 + %1705 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1704, i32 1, i32 31), !dbg !29 + %1706 = bitcast i32 %1705 to float, !dbg !29 + %1707 = fadd float %1703, %1706, !dbg !26 + %1708 = bitcast float %1364 to i32, !dbg !29 + %1709 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1708, i32 8, i32 31), !dbg !29 + %1710 = bitcast i32 %1709 to float, !dbg !29 + %1711 = fadd float %1364, %1710, !dbg !26 + %1712 = bitcast float %1711 to i32, !dbg !29 + %1713 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1712, i32 4, i32 31), !dbg !29 + %1714 = bitcast i32 %1713 to float, !dbg !29 + %1715 = fadd float %1711, %1714, !dbg !26 + %1716 = bitcast float %1715 to i32, !dbg !29 + %1717 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1716, i32 2, i32 31), !dbg !29 + %1718 = bitcast i32 %1717 to float, !dbg !29 + %1719 = fadd float %1715, %1718, !dbg !26 + %1720 = bitcast float %1719 to i32, !dbg !29 + %1721 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1720, i32 1, i32 31), !dbg !29 + %1722 = bitcast i32 %1721 to float, !dbg !29 + %1723 = fadd float %1719, %1722, !dbg !26 + %1724 = bitcast float %1371 to i32, !dbg !29 + %1725 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1724, i32 8, i32 31), !dbg !29 + %1726 = bitcast i32 %1725 to float, !dbg !29 + %1727 = fadd float %1371, %1726, !dbg !26 + %1728 = bitcast float %1727 to i32, !dbg !29 + %1729 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1728, i32 4, i32 31), !dbg !29 + %1730 = bitcast i32 %1729 to float, !dbg !29 + %1731 = fadd float %1727, %1730, !dbg !26 + %1732 = bitcast float %1731 to i32, !dbg !29 + %1733 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1732, i32 2, i32 31), !dbg !29 + %1734 = bitcast i32 %1733 to float, !dbg !29 + %1735 = fadd float %1731, %1734, !dbg !26 + %1736 = bitcast float %1735 to i32, !dbg !29 + %1737 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1736, i32 1, i32 31), !dbg !29 + %1738 = bitcast i32 %1737 to float, !dbg !29 + %1739 = fadd float %1735, %1738, !dbg !26 + %1740 = bitcast float %1378 to i32, !dbg !29 + %1741 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1740, i32 8, i32 31), !dbg !29 + %1742 = bitcast i32 %1741 to float, !dbg !29 + %1743 = fadd float %1378, %1742, !dbg !26 + %1744 = bitcast float %1743 to i32, !dbg !29 + %1745 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1744, i32 4, i32 31), !dbg !29 + %1746 = bitcast i32 %1745 to float, !dbg !29 + %1747 = fadd float %1743, %1746, !dbg !26 + %1748 = bitcast float %1747 to i32, !dbg !29 + %1749 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1748, i32 2, i32 31), !dbg !29 + %1750 = bitcast i32 %1749 to float, !dbg !29 + %1751 = fadd float %1747, %1750, !dbg !26 + %1752 = bitcast float %1751 to i32, !dbg !29 + %1753 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1752, i32 1, i32 31), !dbg !29 + %1754 = bitcast i32 %1753 to float, !dbg !29 + %1755 = fadd float %1751, %1754, !dbg !26 + %1756 = bitcast float %1385 to i32, !dbg !29 + %1757 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1756, i32 8, i32 31), !dbg !29 + %1758 = bitcast i32 %1757 to float, !dbg !29 + %1759 = fadd float %1385, %1758, !dbg !26 + %1760 = bitcast float %1759 to i32, !dbg !29 + %1761 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1760, i32 4, i32 31), !dbg !29 + %1762 = bitcast i32 %1761 to float, !dbg !29 + %1763 = fadd float %1759, %1762, !dbg !26 + %1764 = bitcast float %1763 to i32, !dbg !29 + %1765 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1764, i32 2, i32 31), !dbg !29 + %1766 = bitcast i32 %1765 to float, !dbg !29 + %1767 = fadd float %1763, %1766, !dbg !26 + %1768 = bitcast float %1767 to i32, !dbg !29 + %1769 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1768, i32 1, i32 31), !dbg !29 + %1770 = bitcast i32 %1769 to float, !dbg !29 + %1771 = fadd float %1767, %1770, !dbg !26 + %1772 = bitcast float %1392 to i32, !dbg !29 + %1773 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1772, i32 8, i32 31), !dbg !29 + %1774 = bitcast i32 %1773 to float, !dbg !29 + %1775 = fadd float %1392, %1774, !dbg !26 + %1776 = bitcast float %1775 to i32, !dbg !29 + %1777 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1776, i32 4, i32 31), !dbg !29 + %1778 = bitcast i32 %1777 to float, !dbg !29 + %1779 = fadd float %1775, %1778, !dbg !26 + %1780 = bitcast float %1779 to i32, !dbg !29 + %1781 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1780, i32 2, i32 31), !dbg !29 + %1782 = bitcast i32 %1781 to float, !dbg !29 + %1783 = fadd float %1779, %1782, !dbg !26 + %1784 = bitcast float %1783 to i32, !dbg !29 + %1785 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1784, i32 1, i32 31), !dbg !29 + %1786 = bitcast i32 %1785 to float, !dbg !29 + %1787 = fadd float %1783, %1786, !dbg !26 + %1788 = bitcast float %1399 to i32, !dbg !29 + %1789 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1788, i32 8, i32 31), !dbg !29 + %1790 = bitcast i32 %1789 to float, !dbg !29 + %1791 = fadd float %1399, %1790, !dbg !26 + %1792 = bitcast float %1791 to i32, !dbg !29 + %1793 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1792, i32 4, i32 31), !dbg !29 + %1794 = bitcast i32 %1793 to float, !dbg !29 + %1795 = fadd float %1791, %1794, !dbg !26 + %1796 = bitcast float %1795 to i32, !dbg !29 + %1797 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1796, i32 2, i32 31), !dbg !29 + %1798 = bitcast i32 %1797 to float, !dbg !29 + %1799 = fadd float %1795, %1798, !dbg !26 + %1800 = bitcast float %1799 to i32, !dbg !29 + %1801 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1800, i32 1, i32 31), !dbg !29 + %1802 = bitcast i32 %1801 to float, !dbg !29 + %1803 = fadd float %1799, %1802, !dbg !26 + %1804 = bitcast float %1406 to i32, !dbg !29 + %1805 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1804, i32 8, i32 31), !dbg !29 + %1806 = bitcast i32 %1805 to float, !dbg !29 + %1807 = fadd float %1406, %1806, !dbg !26 + %1808 = bitcast float %1807 to i32, !dbg !29 + %1809 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1808, i32 4, i32 31), !dbg !29 + %1810 = bitcast i32 %1809 to float, !dbg !29 + %1811 = fadd float %1807, %1810, !dbg !26 + %1812 = bitcast float %1811 to i32, !dbg !29 + %1813 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1812, i32 2, i32 31), !dbg !29 + %1814 = bitcast i32 %1813 to float, !dbg !29 + %1815 = fadd float %1811, %1814, !dbg !26 + %1816 = bitcast float %1815 to i32, !dbg !29 + %1817 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1816, i32 1, i32 31), !dbg !29 + %1818 = bitcast i32 %1817 to float, !dbg !29 + %1819 = fadd float %1815, %1818, !dbg !26 + %1820 = bitcast float %1413 to i32, !dbg !29 + %1821 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1820, i32 8, i32 31), !dbg !29 + %1822 = bitcast i32 %1821 to float, !dbg !29 + %1823 = fadd float %1413, %1822, !dbg !26 + %1824 = bitcast float %1823 to i32, !dbg !29 + %1825 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1824, i32 4, i32 31), !dbg !29 + %1826 = bitcast i32 %1825 to float, !dbg !29 + %1827 = fadd float %1823, %1826, !dbg !26 + %1828 = bitcast float %1827 to i32, !dbg !29 + %1829 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1828, i32 2, i32 31), !dbg !29 + %1830 = bitcast i32 %1829 to float, !dbg !29 + %1831 = fadd float %1827, %1830, !dbg !26 + %1832 = bitcast float %1831 to i32, !dbg !29 + %1833 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1832, i32 1, i32 31), !dbg !29 + %1834 = bitcast i32 %1833 to float, !dbg !29 + %1835 = fadd float %1831, %1834, !dbg !26 + %1836 = bitcast float %1420 to i32, !dbg !29 + %1837 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1836, i32 8, i32 31), !dbg !29 + %1838 = bitcast i32 %1837 to float, !dbg !29 + %1839 = fadd float %1420, %1838, !dbg !26 + %1840 = bitcast float %1839 to i32, !dbg !29 + %1841 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1840, i32 4, i32 31), !dbg !29 + %1842 = bitcast i32 %1841 to float, !dbg !29 + %1843 = fadd float %1839, %1842, !dbg !26 + %1844 = bitcast float %1843 to i32, !dbg !29 + %1845 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1844, i32 2, i32 31), !dbg !29 + %1846 = bitcast i32 %1845 to float, !dbg !29 + %1847 = fadd float %1843, %1846, !dbg !26 + %1848 = bitcast float %1847 to i32, !dbg !29 + %1849 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1848, i32 1, i32 31), !dbg !29 + %1850 = bitcast i32 %1849 to float, !dbg !29 + %1851 = fadd float %1847, %1850, !dbg !26 + %1852 = bitcast float %1427 to i32, !dbg !29 + %1853 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1852, i32 8, i32 31), !dbg !29 + %1854 = bitcast i32 %1853 to float, !dbg !29 + %1855 = fadd float %1427, %1854, !dbg !26 + %1856 = bitcast float %1855 to i32, !dbg !29 + %1857 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1856, i32 4, i32 31), !dbg !29 + %1858 = bitcast i32 %1857 to float, !dbg !29 + %1859 = fadd float %1855, %1858, !dbg !26 + %1860 = bitcast float %1859 to i32, !dbg !29 + %1861 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1860, i32 2, i32 31), !dbg !29 + %1862 = bitcast i32 %1861 to float, !dbg !29 + %1863 = fadd float %1859, %1862, !dbg !26 + %1864 = bitcast float %1863 to i32, !dbg !29 + %1865 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1864, i32 1, i32 31), !dbg !29 + %1866 = bitcast i32 %1865 to float, !dbg !29 + %1867 = fadd float %1863, %1866, !dbg !26 + %1868 = bitcast float %1434 to i32, !dbg !29 + %1869 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1868, i32 8, i32 31), !dbg !29 + %1870 = bitcast i32 %1869 to float, !dbg !29 + %1871 = fadd float %1434, %1870, !dbg !26 + %1872 = bitcast float %1871 to i32, !dbg !29 + %1873 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1872, i32 4, i32 31), !dbg !29 + %1874 = bitcast i32 %1873 to float, !dbg !29 + %1875 = fadd float %1871, %1874, !dbg !26 + %1876 = bitcast float %1875 to i32, !dbg !29 + %1877 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1876, i32 2, i32 31), !dbg !29 + %1878 = bitcast i32 %1877 to float, !dbg !29 + %1879 = fadd float %1875, %1878, !dbg !26 + %1880 = bitcast float %1879 to i32, !dbg !29 + %1881 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1880, i32 1, i32 31), !dbg !29 + %1882 = bitcast i32 %1881 to float, !dbg !29 + %1883 = fadd float %1879, %1882, !dbg !26 + %1884 = bitcast float %1441 to i32, !dbg !29 + %1885 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1884, i32 8, i32 31), !dbg !29 + %1886 = bitcast i32 %1885 to float, !dbg !29 + %1887 = fadd float %1441, %1886, !dbg !26 + %1888 = bitcast float %1887 to i32, !dbg !29 + %1889 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1888, i32 4, i32 31), !dbg !29 + %1890 = bitcast i32 %1889 to float, !dbg !29 + %1891 = fadd float %1887, %1890, !dbg !26 + %1892 = bitcast float %1891 to i32, !dbg !29 + %1893 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1892, i32 2, i32 31), !dbg !29 + %1894 = bitcast i32 %1893 to float, !dbg !29 + %1895 = fadd float %1891, %1894, !dbg !26 + %1896 = bitcast float %1895 to i32, !dbg !29 + %1897 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1896, i32 1, i32 31), !dbg !29 + %1898 = bitcast i32 %1897 to float, !dbg !29 + %1899 = fadd float %1895, %1898, !dbg !26 + %1900 = bitcast float %1448 to i32, !dbg !29 + %1901 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1900, i32 8, i32 31), !dbg !29 + %1902 = bitcast i32 %1901 to float, !dbg !29 + %1903 = fadd float %1448, %1902, !dbg !26 + %1904 = bitcast float %1903 to i32, !dbg !29 + %1905 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1904, i32 4, i32 31), !dbg !29 + %1906 = bitcast i32 %1905 to float, !dbg !29 + %1907 = fadd float %1903, %1906, !dbg !26 + %1908 = bitcast float %1907 to i32, !dbg !29 + %1909 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1908, i32 2, i32 31), !dbg !29 + %1910 = bitcast i32 %1909 to float, !dbg !29 + %1911 = fadd float %1907, %1910, !dbg !26 + %1912 = bitcast float %1911 to i32, !dbg !29 + %1913 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1912, i32 1, i32 31), !dbg !29 + %1914 = bitcast i32 %1913 to float, !dbg !29 + %1915 = fadd float %1911, %1914, !dbg !26 + %1916 = bitcast float %1455 to i32, !dbg !29 + %1917 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1916, i32 8, i32 31), !dbg !29 + %1918 = bitcast i32 %1917 to float, !dbg !29 + %1919 = fadd float %1455, %1918, !dbg !26 + %1920 = bitcast float %1919 to i32, !dbg !29 + %1921 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1920, i32 4, i32 31), !dbg !29 + %1922 = bitcast i32 %1921 to float, !dbg !29 + %1923 = fadd float %1919, %1922, !dbg !26 + %1924 = bitcast float %1923 to i32, !dbg !29 + %1925 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1924, i32 2, i32 31), !dbg !29 + %1926 = bitcast i32 %1925 to float, !dbg !29 + %1927 = fadd float %1923, %1926, !dbg !26 + %1928 = bitcast float %1927 to i32, !dbg !29 + %1929 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1928, i32 1, i32 31), !dbg !29 + %1930 = bitcast i32 %1929 to float, !dbg !29 + %1931 = fadd float %1927, %1930, !dbg !26 + %1932 = bitcast float %1462 to i32, !dbg !29 + %1933 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1932, i32 8, i32 31), !dbg !29 + %1934 = bitcast i32 %1933 to float, !dbg !29 + %1935 = fadd float %1462, %1934, !dbg !26 + %1936 = bitcast float %1935 to i32, !dbg !29 + %1937 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1936, i32 4, i32 31), !dbg !29 + %1938 = bitcast i32 %1937 to float, !dbg !29 + %1939 = fadd float %1935, %1938, !dbg !26 + %1940 = bitcast float %1939 to i32, !dbg !29 + %1941 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1940, i32 2, i32 31), !dbg !29 + %1942 = bitcast i32 %1941 to float, !dbg !29 + %1943 = fadd float %1939, %1942, !dbg !26 + %1944 = bitcast float %1943 to i32, !dbg !29 + %1945 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1944, i32 1, i32 31), !dbg !29 + %1946 = bitcast i32 %1945 to float, !dbg !29 + %1947 = fadd float %1943, %1946, !dbg !26 + %1948 = bitcast float %1469 to i32, !dbg !29 + %1949 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1948, i32 8, i32 31), !dbg !29 + %1950 = bitcast i32 %1949 to float, !dbg !29 + %1951 = fadd float %1469, %1950, !dbg !26 + %1952 = bitcast float %1951 to i32, !dbg !29 + %1953 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1952, i32 4, i32 31), !dbg !29 + %1954 = bitcast i32 %1953 to float, !dbg !29 + %1955 = fadd float %1951, %1954, !dbg !26 + %1956 = bitcast float %1955 to i32, !dbg !29 + %1957 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1956, i32 2, i32 31), !dbg !29 + %1958 = bitcast i32 %1957 to float, !dbg !29 + %1959 = fadd float %1955, %1958, !dbg !26 + %1960 = bitcast float %1959 to i32, !dbg !29 + %1961 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1960, i32 1, i32 31), !dbg !29 + %1962 = bitcast i32 %1961 to float, !dbg !29 + %1963 = fadd float %1959, %1962, !dbg !26 + %1964 = bitcast float %1476 to i32, !dbg !29 + %1965 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1964, i32 8, i32 31), !dbg !29 + %1966 = bitcast i32 %1965 to float, !dbg !29 + %1967 = fadd float %1476, %1966, !dbg !26 + %1968 = bitcast float %1967 to i32, !dbg !29 + %1969 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1968, i32 4, i32 31), !dbg !29 + %1970 = bitcast i32 %1969 to float, !dbg !29 + %1971 = fadd float %1967, %1970, !dbg !26 + %1972 = bitcast float %1971 to i32, !dbg !29 + %1973 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1972, i32 2, i32 31), !dbg !29 + %1974 = bitcast i32 %1973 to float, !dbg !29 + %1975 = fadd float %1971, %1974, !dbg !26 + %1976 = bitcast float %1975 to i32, !dbg !29 + %1977 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1976, i32 1, i32 31), !dbg !29 + %1978 = bitcast i32 %1977 to float, !dbg !29 + %1979 = fadd float %1975, %1978, !dbg !26 + %1980 = bitcast float %1483 to i32, !dbg !29 + %1981 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1980, i32 8, i32 31), !dbg !29 + %1982 = bitcast i32 %1981 to float, !dbg !29 + %1983 = fadd float %1483, %1982, !dbg !26 + %1984 = bitcast float %1983 to i32, !dbg !29 + %1985 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1984, i32 4, i32 31), !dbg !29 + %1986 = bitcast i32 %1985 to float, !dbg !29 + %1987 = fadd float %1983, %1986, !dbg !26 + %1988 = bitcast float %1987 to i32, !dbg !29 + %1989 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1988, i32 2, i32 31), !dbg !29 + %1990 = bitcast i32 %1989 to float, !dbg !29 + %1991 = fadd float %1987, %1990, !dbg !26 + %1992 = bitcast float %1991 to i32, !dbg !29 + %1993 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1992, i32 1, i32 31), !dbg !29 + %1994 = bitcast i32 %1993 to float, !dbg !29 + %1995 = fadd float %1991, %1994, !dbg !26 + %1996 = fadd float %3, %1499, !dbg !32 + %1997 = fadd float %3, %1515, !dbg !32 + %1998 = fadd float %3, %1531, !dbg !32 + %1999 = fadd float %3, %1547, !dbg !32 + %2000 = fadd float %3, %1563, !dbg !32 + %2001 = fadd float %3, %1579, !dbg !32 + %2002 = fadd float %3, %1595, !dbg !32 + %2003 = fadd float %3, %1611, !dbg !32 + %2004 = fadd float %3, %1627, !dbg !32 + %2005 = fadd float %3, %1643, !dbg !32 + %2006 = fadd float %3, %1659, !dbg !32 + %2007 = fadd float %3, %1675, !dbg !32 + %2008 = fadd float %3, %1691, !dbg !32 + %2009 = fadd float %3, %1707, !dbg !32 + %2010 = fadd float %3, %1723, !dbg !32 + %2011 = fadd float %3, %1739, !dbg !32 + %2012 = fadd float %3, %1755, !dbg !32 + %2013 = fadd float %3, %1771, !dbg !32 + %2014 = fadd float %3, %1787, !dbg !32 + %2015 = fadd float %3, %1803, !dbg !32 + %2016 = fadd float %3, %1819, !dbg !32 + %2017 = fadd float %3, %1835, !dbg !32 + %2018 = fadd float %3, %1851, !dbg !32 + %2019 = fadd float %3, %1867, !dbg !32 + %2020 = fadd float %3, %1883, !dbg !32 + %2021 = fadd float %3, %1899, !dbg !32 + %2022 = fadd float %3, %1915, !dbg !32 + %2023 = fadd float %3, %1931, !dbg !32 + %2024 = fadd float %3, %1947, !dbg !32 + %2025 = fadd float %3, %1963, !dbg !32 + %2026 = fadd float %3, %1979, !dbg !32 + %2027 = fadd float %3, %1995, !dbg !32 + %2028 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i = icmp eq i32 %2028, 0, !dbg !33 + %2029 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i = icmp eq i32 %2029, 0, !dbg !33 + br i1 %.not.i, label %2035, label %2030, !dbg !33 + +2030: ; preds = %7 + br i1 %.not1.i, label %2033, label %2031, !dbg !33 + +2031: ; preds = %2030 + %2032 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1996) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +2033: ; preds = %2030 + %2034 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1996) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +2035: ; preds = %7 + br i1 %.not1.i, label %2038, label %2036, !dbg !33 + +2036: ; preds = %2035 + %2037 = tail call float @llvm.nvvm.sqrt.rn.f(float %1996) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +2038: ; preds = %2035 + %2039 = tail call float @llvm.nvvm.sqrt.approx.f(float %1996) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +__nv_sqrtf.exit: ; preds = %2031, %2033, %2036, %2038 + %.0.i = phi float [ %2032, %2031 ], [ %2034, %2033 ], [ %2037, %2036 ], [ %2039, %2038 ], !dbg !33 + %2040 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i1 = icmp eq i32 %2040, 0, !dbg !33 + %2041 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i4 = icmp eq i32 %2041, 0, !dbg !33 + br i1 %.not.i1, label %2047, label %2042, !dbg !33 + +2042: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %2045, label %2043, !dbg !33 + +2043: ; preds = %2042 + %2044 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1997) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +2045: ; preds = %2042 + %2046 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1997) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +2047: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %2050, label %2048, !dbg !33 + +2048: ; preds = %2047 + %2049 = tail call float @llvm.nvvm.sqrt.rn.f(float %1997) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +2050: ; preds = %2047 + %2051 = tail call float @llvm.nvvm.sqrt.approx.f(float %1997) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +__nv_sqrtf.exit5: ; preds = %2043, %2045, %2048, %2050 + %.0.i3 = phi float [ %2044, %2043 ], [ %2046, %2045 ], [ %2049, %2048 ], [ %2051, %2050 ], !dbg !33 + %2052 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i6 = icmp eq i32 %2052, 0, !dbg !33 + %2053 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i9 = icmp eq i32 %2053, 0, !dbg !33 + br i1 %.not.i6, label %2059, label %2054, !dbg !33 + +2054: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %2057, label %2055, !dbg !33 + +2055: ; preds = %2054 + %2056 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1998) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +2057: ; preds = %2054 + %2058 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1998) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +2059: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %2062, label %2060, !dbg !33 + +2060: ; preds = %2059 + %2061 = tail call float @llvm.nvvm.sqrt.rn.f(float %1998) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +2062: ; preds = %2059 + %2063 = tail call float @llvm.nvvm.sqrt.approx.f(float %1998) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +__nv_sqrtf.exit10: ; preds = %2055, %2057, %2060, %2062 + %.0.i8 = phi float [ %2056, %2055 ], [ %2058, %2057 ], [ %2061, %2060 ], [ %2063, %2062 ], !dbg !33 + %2064 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i11 = icmp eq i32 %2064, 0, !dbg !33 + %2065 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i14 = icmp eq i32 %2065, 0, !dbg !33 + br i1 %.not.i11, label %2071, label %2066, !dbg !33 + +2066: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %2069, label %2067, !dbg !33 + +2067: ; preds = %2066 + %2068 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1999) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +2069: ; preds = %2066 + %2070 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1999) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +2071: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %2074, label %2072, !dbg !33 + +2072: ; preds = %2071 + %2073 = tail call float @llvm.nvvm.sqrt.rn.f(float %1999) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +2074: ; preds = %2071 + %2075 = tail call float @llvm.nvvm.sqrt.approx.f(float %1999) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +__nv_sqrtf.exit15: ; preds = %2067, %2069, %2072, %2074 + %.0.i13 = phi float [ %2068, %2067 ], [ %2070, %2069 ], [ %2073, %2072 ], [ %2075, %2074 ], !dbg !33 + %2076 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i16 = icmp eq i32 %2076, 0, !dbg !33 + %2077 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i19 = icmp eq i32 %2077, 0, !dbg !33 + br i1 %.not.i16, label %2083, label %2078, !dbg !33 + +2078: ; preds = %__nv_sqrtf.exit15 + br i1 %.not1.i19, label %2081, label %2079, !dbg !33 + +2079: ; preds = %2078 + %2080 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2000) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +2081: ; preds = %2078 + %2082 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2000) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +2083: ; preds = %__nv_sqrtf.exit15 + br i1 %.not1.i19, label %2086, label %2084, !dbg !33 + +2084: ; preds = %2083 + %2085 = tail call float @llvm.nvvm.sqrt.rn.f(float %2000) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +2086: ; preds = %2083 + %2087 = tail call float @llvm.nvvm.sqrt.approx.f(float %2000) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +__nv_sqrtf.exit20: ; preds = %2079, %2081, %2084, %2086 + %.0.i18 = phi float [ %2080, %2079 ], [ %2082, %2081 ], [ %2085, %2084 ], [ %2087, %2086 ], !dbg !33 + %2088 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i21 = icmp eq i32 %2088, 0, !dbg !33 + %2089 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i24 = icmp eq i32 %2089, 0, !dbg !33 + br i1 %.not.i21, label %2095, label %2090, !dbg !33 + +2090: ; preds = %__nv_sqrtf.exit20 + br i1 %.not1.i24, label %2093, label %2091, !dbg !33 + +2091: ; preds = %2090 + %2092 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2001) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +2093: ; preds = %2090 + %2094 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2001) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +2095: ; preds = %__nv_sqrtf.exit20 + br i1 %.not1.i24, label %2098, label %2096, !dbg !33 + +2096: ; preds = %2095 + %2097 = tail call float @llvm.nvvm.sqrt.rn.f(float %2001) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +2098: ; preds = %2095 + %2099 = tail call float @llvm.nvvm.sqrt.approx.f(float %2001) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +__nv_sqrtf.exit25: ; preds = %2091, %2093, %2096, %2098 + %.0.i23 = phi float [ %2092, %2091 ], [ %2094, %2093 ], [ %2097, %2096 ], [ %2099, %2098 ], !dbg !33 + %2100 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i26 = icmp eq i32 %2100, 0, !dbg !33 + %2101 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i29 = icmp eq i32 %2101, 0, !dbg !33 + br i1 %.not.i26, label %2107, label %2102, !dbg !33 + +2102: ; preds = %__nv_sqrtf.exit25 + br i1 %.not1.i29, label %2105, label %2103, !dbg !33 + +2103: ; preds = %2102 + %2104 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2002) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +2105: ; preds = %2102 + %2106 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2002) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +2107: ; preds = %__nv_sqrtf.exit25 + br i1 %.not1.i29, label %2110, label %2108, !dbg !33 + +2108: ; preds = %2107 + %2109 = tail call float @llvm.nvvm.sqrt.rn.f(float %2002) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +2110: ; preds = %2107 + %2111 = tail call float @llvm.nvvm.sqrt.approx.f(float %2002) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +__nv_sqrtf.exit30: ; preds = %2103, %2105, %2108, %2110 + %.0.i28 = phi float [ %2104, %2103 ], [ %2106, %2105 ], [ %2109, %2108 ], [ %2111, %2110 ], !dbg !33 + %2112 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i31 = icmp eq i32 %2112, 0, !dbg !33 + %2113 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i34 = icmp eq i32 %2113, 0, !dbg !33 + br i1 %.not.i31, label %2119, label %2114, !dbg !33 + +2114: ; preds = %__nv_sqrtf.exit30 + br i1 %.not1.i34, label %2117, label %2115, !dbg !33 + +2115: ; preds = %2114 + %2116 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2003) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +2117: ; preds = %2114 + %2118 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2003) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +2119: ; preds = %__nv_sqrtf.exit30 + br i1 %.not1.i34, label %2122, label %2120, !dbg !33 + +2120: ; preds = %2119 + %2121 = tail call float @llvm.nvvm.sqrt.rn.f(float %2003) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +2122: ; preds = %2119 + %2123 = tail call float @llvm.nvvm.sqrt.approx.f(float %2003) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +__nv_sqrtf.exit35: ; preds = %2115, %2117, %2120, %2122 + %.0.i33 = phi float [ %2116, %2115 ], [ %2118, %2117 ], [ %2121, %2120 ], [ %2123, %2122 ], !dbg !33 + %2124 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i36 = icmp eq i32 %2124, 0, !dbg !33 + %2125 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i39 = icmp eq i32 %2125, 0, !dbg !33 + br i1 %.not.i36, label %2131, label %2126, !dbg !33 + +2126: ; preds = %__nv_sqrtf.exit35 + br i1 %.not1.i39, label %2129, label %2127, !dbg !33 + +2127: ; preds = %2126 + %2128 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2004) #6, !dbg !33 + br label %__nv_sqrtf.exit40, !dbg !33 + +2129: ; preds = %2126 + %2130 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2004) #6, !dbg !33 + br label %__nv_sqrtf.exit40, !dbg !33 + +2131: ; preds = %__nv_sqrtf.exit35 + br i1 %.not1.i39, label %2134, label %2132, !dbg !33 + +2132: ; preds = %2131 + %2133 = tail call float @llvm.nvvm.sqrt.rn.f(float %2004) #6, !dbg !33 + br label %__nv_sqrtf.exit40, !dbg !33 + +2134: ; preds = %2131 + %2135 = tail call float @llvm.nvvm.sqrt.approx.f(float %2004) #6, !dbg !33 + br label %__nv_sqrtf.exit40, !dbg !33 + +__nv_sqrtf.exit40: ; preds = %2127, %2129, %2132, %2134 + %.0.i38 = phi float [ %2128, %2127 ], [ %2130, %2129 ], [ %2133, %2132 ], [ %2135, %2134 ], !dbg !33 + %2136 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i41 = icmp eq i32 %2136, 0, !dbg !33 + %2137 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i44 = icmp eq i32 %2137, 0, !dbg !33 + br i1 %.not.i41, label %2143, label %2138, !dbg !33 + +2138: ; preds = %__nv_sqrtf.exit40 + br i1 %.not1.i44, label %2141, label %2139, !dbg !33 + +2139: ; preds = %2138 + %2140 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2005) #6, !dbg !33 + br label %__nv_sqrtf.exit45, !dbg !33 + +2141: ; preds = %2138 + %2142 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2005) #6, !dbg !33 + br label %__nv_sqrtf.exit45, !dbg !33 + +2143: ; preds = %__nv_sqrtf.exit40 + br i1 %.not1.i44, label %2146, label %2144, !dbg !33 + +2144: ; preds = %2143 + %2145 = tail call float @llvm.nvvm.sqrt.rn.f(float %2005) #6, !dbg !33 + br label %__nv_sqrtf.exit45, !dbg !33 + +2146: ; preds = %2143 + %2147 = tail call float @llvm.nvvm.sqrt.approx.f(float %2005) #6, !dbg !33 + br label %__nv_sqrtf.exit45, !dbg !33 + +__nv_sqrtf.exit45: ; preds = %2139, %2141, %2144, %2146 + %.0.i43 = phi float [ %2140, %2139 ], [ %2142, %2141 ], [ %2145, %2144 ], [ %2147, %2146 ], !dbg !33 + %2148 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i46 = icmp eq i32 %2148, 0, !dbg !33 + %2149 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i49 = icmp eq i32 %2149, 0, !dbg !33 + br i1 %.not.i46, label %2155, label %2150, !dbg !33 + +2150: ; preds = %__nv_sqrtf.exit45 + br i1 %.not1.i49, label %2153, label %2151, !dbg !33 + +2151: ; preds = %2150 + %2152 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2006) #6, !dbg !33 + br label %__nv_sqrtf.exit50, !dbg !33 + +2153: ; preds = %2150 + %2154 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2006) #6, !dbg !33 + br label %__nv_sqrtf.exit50, !dbg !33 + +2155: ; preds = %__nv_sqrtf.exit45 + br i1 %.not1.i49, label %2158, label %2156, !dbg !33 + +2156: ; preds = %2155 + %2157 = tail call float @llvm.nvvm.sqrt.rn.f(float %2006) #6, !dbg !33 + br label %__nv_sqrtf.exit50, !dbg !33 + +2158: ; preds = %2155 + %2159 = tail call float @llvm.nvvm.sqrt.approx.f(float %2006) #6, !dbg !33 + br label %__nv_sqrtf.exit50, !dbg !33 + +__nv_sqrtf.exit50: ; preds = %2151, %2153, %2156, %2158 + %.0.i48 = phi float [ %2152, %2151 ], [ %2154, %2153 ], [ %2157, %2156 ], [ %2159, %2158 ], !dbg !33 + %2160 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i51 = icmp eq i32 %2160, 0, !dbg !33 + %2161 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i54 = icmp eq i32 %2161, 0, !dbg !33 + br i1 %.not.i51, label %2167, label %2162, !dbg !33 + +2162: ; preds = %__nv_sqrtf.exit50 + br i1 %.not1.i54, label %2165, label %2163, !dbg !33 + +2163: ; preds = %2162 + %2164 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2007) #6, !dbg !33 + br label %__nv_sqrtf.exit55, !dbg !33 + +2165: ; preds = %2162 + %2166 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2007) #6, !dbg !33 + br label %__nv_sqrtf.exit55, !dbg !33 + +2167: ; preds = %__nv_sqrtf.exit50 + br i1 %.not1.i54, label %2170, label %2168, !dbg !33 + +2168: ; preds = %2167 + %2169 = tail call float @llvm.nvvm.sqrt.rn.f(float %2007) #6, !dbg !33 + br label %__nv_sqrtf.exit55, !dbg !33 + +2170: ; preds = %2167 + %2171 = tail call float @llvm.nvvm.sqrt.approx.f(float %2007) #6, !dbg !33 + br label %__nv_sqrtf.exit55, !dbg !33 + +__nv_sqrtf.exit55: ; preds = %2163, %2165, %2168, %2170 + %.0.i53 = phi float [ %2164, %2163 ], [ %2166, %2165 ], [ %2169, %2168 ], [ %2171, %2170 ], !dbg !33 + %2172 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i56 = icmp eq i32 %2172, 0, !dbg !33 + %2173 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i59 = icmp eq i32 %2173, 0, !dbg !33 + br i1 %.not.i56, label %2179, label %2174, !dbg !33 + +2174: ; preds = %__nv_sqrtf.exit55 + br i1 %.not1.i59, label %2177, label %2175, !dbg !33 + +2175: ; preds = %2174 + %2176 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2008) #6, !dbg !33 + br label %__nv_sqrtf.exit60, !dbg !33 + +2177: ; preds = %2174 + %2178 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2008) #6, !dbg !33 + br label %__nv_sqrtf.exit60, !dbg !33 + +2179: ; preds = %__nv_sqrtf.exit55 + br i1 %.not1.i59, label %2182, label %2180, !dbg !33 + +2180: ; preds = %2179 + %2181 = tail call float @llvm.nvvm.sqrt.rn.f(float %2008) #6, !dbg !33 + br label %__nv_sqrtf.exit60, !dbg !33 + +2182: ; preds = %2179 + %2183 = tail call float @llvm.nvvm.sqrt.approx.f(float %2008) #6, !dbg !33 + br label %__nv_sqrtf.exit60, !dbg !33 + +__nv_sqrtf.exit60: ; preds = %2175, %2177, %2180, %2182 + %.0.i58 = phi float [ %2176, %2175 ], [ %2178, %2177 ], [ %2181, %2180 ], [ %2183, %2182 ], !dbg !33 + %2184 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i61 = icmp eq i32 %2184, 0, !dbg !33 + %2185 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i64 = icmp eq i32 %2185, 0, !dbg !33 + br i1 %.not.i61, label %2191, label %2186, !dbg !33 + +2186: ; preds = %__nv_sqrtf.exit60 + br i1 %.not1.i64, label %2189, label %2187, !dbg !33 + +2187: ; preds = %2186 + %2188 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2009) #6, !dbg !33 + br label %__nv_sqrtf.exit65, !dbg !33 + +2189: ; preds = %2186 + %2190 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2009) #6, !dbg !33 + br label %__nv_sqrtf.exit65, !dbg !33 + +2191: ; preds = %__nv_sqrtf.exit60 + br i1 %.not1.i64, label %2194, label %2192, !dbg !33 + +2192: ; preds = %2191 + %2193 = tail call float @llvm.nvvm.sqrt.rn.f(float %2009) #6, !dbg !33 + br label %__nv_sqrtf.exit65, !dbg !33 + +2194: ; preds = %2191 + %2195 = tail call float @llvm.nvvm.sqrt.approx.f(float %2009) #6, !dbg !33 + br label %__nv_sqrtf.exit65, !dbg !33 + +__nv_sqrtf.exit65: ; preds = %2187, %2189, %2192, %2194 + %.0.i63 = phi float [ %2188, %2187 ], [ %2190, %2189 ], [ %2193, %2192 ], [ %2195, %2194 ], !dbg !33 + %2196 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i66 = icmp eq i32 %2196, 0, !dbg !33 + %2197 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i69 = icmp eq i32 %2197, 0, !dbg !33 + br i1 %.not.i66, label %2203, label %2198, !dbg !33 + +2198: ; preds = %__nv_sqrtf.exit65 + br i1 %.not1.i69, label %2201, label %2199, !dbg !33 + +2199: ; preds = %2198 + %2200 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2010) #6, !dbg !33 + br label %__nv_sqrtf.exit70, !dbg !33 + +2201: ; preds = %2198 + %2202 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2010) #6, !dbg !33 + br label %__nv_sqrtf.exit70, !dbg !33 + +2203: ; preds = %__nv_sqrtf.exit65 + br i1 %.not1.i69, label %2206, label %2204, !dbg !33 + +2204: ; preds = %2203 + %2205 = tail call float @llvm.nvvm.sqrt.rn.f(float %2010) #6, !dbg !33 + br label %__nv_sqrtf.exit70, !dbg !33 + +2206: ; preds = %2203 + %2207 = tail call float @llvm.nvvm.sqrt.approx.f(float %2010) #6, !dbg !33 + br label %__nv_sqrtf.exit70, !dbg !33 + +__nv_sqrtf.exit70: ; preds = %2199, %2201, %2204, %2206 + %.0.i68 = phi float [ %2200, %2199 ], [ %2202, %2201 ], [ %2205, %2204 ], [ %2207, %2206 ], !dbg !33 + %2208 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i71 = icmp eq i32 %2208, 0, !dbg !33 + %2209 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i74 = icmp eq i32 %2209, 0, !dbg !33 + br i1 %.not.i71, label %2215, label %2210, !dbg !33 + +2210: ; preds = %__nv_sqrtf.exit70 + br i1 %.not1.i74, label %2213, label %2211, !dbg !33 + +2211: ; preds = %2210 + %2212 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2011) #6, !dbg !33 + br label %__nv_sqrtf.exit75, !dbg !33 + +2213: ; preds = %2210 + %2214 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2011) #6, !dbg !33 + br label %__nv_sqrtf.exit75, !dbg !33 + +2215: ; preds = %__nv_sqrtf.exit70 + br i1 %.not1.i74, label %2218, label %2216, !dbg !33 + +2216: ; preds = %2215 + %2217 = tail call float @llvm.nvvm.sqrt.rn.f(float %2011) #6, !dbg !33 + br label %__nv_sqrtf.exit75, !dbg !33 + +2218: ; preds = %2215 + %2219 = tail call float @llvm.nvvm.sqrt.approx.f(float %2011) #6, !dbg !33 + br label %__nv_sqrtf.exit75, !dbg !33 + +__nv_sqrtf.exit75: ; preds = %2211, %2213, %2216, %2218 + %.0.i73 = phi float [ %2212, %2211 ], [ %2214, %2213 ], [ %2217, %2216 ], [ %2219, %2218 ], !dbg !33 + %2220 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i76 = icmp eq i32 %2220, 0, !dbg !33 + %2221 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i79 = icmp eq i32 %2221, 0, !dbg !33 + br i1 %.not.i76, label %2227, label %2222, !dbg !33 + +2222: ; preds = %__nv_sqrtf.exit75 + br i1 %.not1.i79, label %2225, label %2223, !dbg !33 + +2223: ; preds = %2222 + %2224 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2012) #6, !dbg !33 + br label %__nv_sqrtf.exit80, !dbg !33 + +2225: ; preds = %2222 + %2226 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2012) #6, !dbg !33 + br label %__nv_sqrtf.exit80, !dbg !33 + +2227: ; preds = %__nv_sqrtf.exit75 + br i1 %.not1.i79, label %2230, label %2228, !dbg !33 + +2228: ; preds = %2227 + %2229 = tail call float @llvm.nvvm.sqrt.rn.f(float %2012) #6, !dbg !33 + br label %__nv_sqrtf.exit80, !dbg !33 + +2230: ; preds = %2227 + %2231 = tail call float @llvm.nvvm.sqrt.approx.f(float %2012) #6, !dbg !33 + br label %__nv_sqrtf.exit80, !dbg !33 + +__nv_sqrtf.exit80: ; preds = %2223, %2225, %2228, %2230 + %.0.i78 = phi float [ %2224, %2223 ], [ %2226, %2225 ], [ %2229, %2228 ], [ %2231, %2230 ], !dbg !33 + %2232 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i81 = icmp eq i32 %2232, 0, !dbg !33 + %2233 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i84 = icmp eq i32 %2233, 0, !dbg !33 + br i1 %.not.i81, label %2239, label %2234, !dbg !33 + +2234: ; preds = %__nv_sqrtf.exit80 + br i1 %.not1.i84, label %2237, label %2235, !dbg !33 + +2235: ; preds = %2234 + %2236 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2013) #6, !dbg !33 + br label %__nv_sqrtf.exit85, !dbg !33 + +2237: ; preds = %2234 + %2238 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2013) #6, !dbg !33 + br label %__nv_sqrtf.exit85, !dbg !33 + +2239: ; preds = %__nv_sqrtf.exit80 + br i1 %.not1.i84, label %2242, label %2240, !dbg !33 + +2240: ; preds = %2239 + %2241 = tail call float @llvm.nvvm.sqrt.rn.f(float %2013) #6, !dbg !33 + br label %__nv_sqrtf.exit85, !dbg !33 + +2242: ; preds = %2239 + %2243 = tail call float @llvm.nvvm.sqrt.approx.f(float %2013) #6, !dbg !33 + br label %__nv_sqrtf.exit85, !dbg !33 + +__nv_sqrtf.exit85: ; preds = %2235, %2237, %2240, %2242 + %.0.i83 = phi float [ %2236, %2235 ], [ %2238, %2237 ], [ %2241, %2240 ], [ %2243, %2242 ], !dbg !33 + %2244 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i86 = icmp eq i32 %2244, 0, !dbg !33 + %2245 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i89 = icmp eq i32 %2245, 0, !dbg !33 + br i1 %.not.i86, label %2251, label %2246, !dbg !33 + +2246: ; preds = %__nv_sqrtf.exit85 + br i1 %.not1.i89, label %2249, label %2247, !dbg !33 + +2247: ; preds = %2246 + %2248 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2014) #6, !dbg !33 + br label %__nv_sqrtf.exit90, !dbg !33 + +2249: ; preds = %2246 + %2250 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2014) #6, !dbg !33 + br label %__nv_sqrtf.exit90, !dbg !33 + +2251: ; preds = %__nv_sqrtf.exit85 + br i1 %.not1.i89, label %2254, label %2252, !dbg !33 + +2252: ; preds = %2251 + %2253 = tail call float @llvm.nvvm.sqrt.rn.f(float %2014) #6, !dbg !33 + br label %__nv_sqrtf.exit90, !dbg !33 + +2254: ; preds = %2251 + %2255 = tail call float @llvm.nvvm.sqrt.approx.f(float %2014) #6, !dbg !33 + br label %__nv_sqrtf.exit90, !dbg !33 + +__nv_sqrtf.exit90: ; preds = %2247, %2249, %2252, %2254 + %.0.i88 = phi float [ %2248, %2247 ], [ %2250, %2249 ], [ %2253, %2252 ], [ %2255, %2254 ], !dbg !33 + %2256 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i91 = icmp eq i32 %2256, 0, !dbg !33 + %2257 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i94 = icmp eq i32 %2257, 0, !dbg !33 + br i1 %.not.i91, label %2263, label %2258, !dbg !33 + +2258: ; preds = %__nv_sqrtf.exit90 + br i1 %.not1.i94, label %2261, label %2259, !dbg !33 + +2259: ; preds = %2258 + %2260 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2015) #6, !dbg !33 + br label %__nv_sqrtf.exit95, !dbg !33 + +2261: ; preds = %2258 + %2262 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2015) #6, !dbg !33 + br label %__nv_sqrtf.exit95, !dbg !33 + +2263: ; preds = %__nv_sqrtf.exit90 + br i1 %.not1.i94, label %2266, label %2264, !dbg !33 + +2264: ; preds = %2263 + %2265 = tail call float @llvm.nvvm.sqrt.rn.f(float %2015) #6, !dbg !33 + br label %__nv_sqrtf.exit95, !dbg !33 + +2266: ; preds = %2263 + %2267 = tail call float @llvm.nvvm.sqrt.approx.f(float %2015) #6, !dbg !33 + br label %__nv_sqrtf.exit95, !dbg !33 + +__nv_sqrtf.exit95: ; preds = %2259, %2261, %2264, %2266 + %.0.i93 = phi float [ %2260, %2259 ], [ %2262, %2261 ], [ %2265, %2264 ], [ %2267, %2266 ], !dbg !33 + %2268 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i96 = icmp eq i32 %2268, 0, !dbg !33 + %2269 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i99 = icmp eq i32 %2269, 0, !dbg !33 + br i1 %.not.i96, label %2275, label %2270, !dbg !33 + +2270: ; preds = %__nv_sqrtf.exit95 + br i1 %.not1.i99, label %2273, label %2271, !dbg !33 + +2271: ; preds = %2270 + %2272 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2016) #6, !dbg !33 + br label %__nv_sqrtf.exit100, !dbg !33 + +2273: ; preds = %2270 + %2274 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2016) #6, !dbg !33 + br label %__nv_sqrtf.exit100, !dbg !33 + +2275: ; preds = %__nv_sqrtf.exit95 + br i1 %.not1.i99, label %2278, label %2276, !dbg !33 + +2276: ; preds = %2275 + %2277 = tail call float @llvm.nvvm.sqrt.rn.f(float %2016) #6, !dbg !33 + br label %__nv_sqrtf.exit100, !dbg !33 + +2278: ; preds = %2275 + %2279 = tail call float @llvm.nvvm.sqrt.approx.f(float %2016) #6, !dbg !33 + br label %__nv_sqrtf.exit100, !dbg !33 + +__nv_sqrtf.exit100: ; preds = %2271, %2273, %2276, %2278 + %.0.i98 = phi float [ %2272, %2271 ], [ %2274, %2273 ], [ %2277, %2276 ], [ %2279, %2278 ], !dbg !33 + %2280 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i101 = icmp eq i32 %2280, 0, !dbg !33 + %2281 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i104 = icmp eq i32 %2281, 0, !dbg !33 + br i1 %.not.i101, label %2287, label %2282, !dbg !33 + +2282: ; preds = %__nv_sqrtf.exit100 + br i1 %.not1.i104, label %2285, label %2283, !dbg !33 + +2283: ; preds = %2282 + %2284 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2017) #6, !dbg !33 + br label %__nv_sqrtf.exit105, !dbg !33 + +2285: ; preds = %2282 + %2286 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2017) #6, !dbg !33 + br label %__nv_sqrtf.exit105, !dbg !33 + +2287: ; preds = %__nv_sqrtf.exit100 + br i1 %.not1.i104, label %2290, label %2288, !dbg !33 + +2288: ; preds = %2287 + %2289 = tail call float @llvm.nvvm.sqrt.rn.f(float %2017) #6, !dbg !33 + br label %__nv_sqrtf.exit105, !dbg !33 + +2290: ; preds = %2287 + %2291 = tail call float @llvm.nvvm.sqrt.approx.f(float %2017) #6, !dbg !33 + br label %__nv_sqrtf.exit105, !dbg !33 + +__nv_sqrtf.exit105: ; preds = %2283, %2285, %2288, %2290 + %.0.i103 = phi float [ %2284, %2283 ], [ %2286, %2285 ], [ %2289, %2288 ], [ %2291, %2290 ], !dbg !33 + %2292 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i106 = icmp eq i32 %2292, 0, !dbg !33 + %2293 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i109 = icmp eq i32 %2293, 0, !dbg !33 + br i1 %.not.i106, label %2299, label %2294, !dbg !33 + +2294: ; preds = %__nv_sqrtf.exit105 + br i1 %.not1.i109, label %2297, label %2295, !dbg !33 + +2295: ; preds = %2294 + %2296 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2018) #6, !dbg !33 + br label %__nv_sqrtf.exit110, !dbg !33 + +2297: ; preds = %2294 + %2298 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2018) #6, !dbg !33 + br label %__nv_sqrtf.exit110, !dbg !33 + +2299: ; preds = %__nv_sqrtf.exit105 + br i1 %.not1.i109, label %2302, label %2300, !dbg !33 + +2300: ; preds = %2299 + %2301 = tail call float @llvm.nvvm.sqrt.rn.f(float %2018) #6, !dbg !33 + br label %__nv_sqrtf.exit110, !dbg !33 + +2302: ; preds = %2299 + %2303 = tail call float @llvm.nvvm.sqrt.approx.f(float %2018) #6, !dbg !33 + br label %__nv_sqrtf.exit110, !dbg !33 + +__nv_sqrtf.exit110: ; preds = %2295, %2297, %2300, %2302 + %.0.i108 = phi float [ %2296, %2295 ], [ %2298, %2297 ], [ %2301, %2300 ], [ %2303, %2302 ], !dbg !33 + %2304 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i111 = icmp eq i32 %2304, 0, !dbg !33 + %2305 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i114 = icmp eq i32 %2305, 0, !dbg !33 + br i1 %.not.i111, label %2311, label %2306, !dbg !33 + +2306: ; preds = %__nv_sqrtf.exit110 + br i1 %.not1.i114, label %2309, label %2307, !dbg !33 + +2307: ; preds = %2306 + %2308 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2019) #6, !dbg !33 + br label %__nv_sqrtf.exit115, !dbg !33 + +2309: ; preds = %2306 + %2310 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2019) #6, !dbg !33 + br label %__nv_sqrtf.exit115, !dbg !33 + +2311: ; preds = %__nv_sqrtf.exit110 + br i1 %.not1.i114, label %2314, label %2312, !dbg !33 + +2312: ; preds = %2311 + %2313 = tail call float @llvm.nvvm.sqrt.rn.f(float %2019) #6, !dbg !33 + br label %__nv_sqrtf.exit115, !dbg !33 + +2314: ; preds = %2311 + %2315 = tail call float @llvm.nvvm.sqrt.approx.f(float %2019) #6, !dbg !33 + br label %__nv_sqrtf.exit115, !dbg !33 + +__nv_sqrtf.exit115: ; preds = %2307, %2309, %2312, %2314 + %.0.i113 = phi float [ %2308, %2307 ], [ %2310, %2309 ], [ %2313, %2312 ], [ %2315, %2314 ], !dbg !33 + %2316 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i116 = icmp eq i32 %2316, 0, !dbg !33 + %2317 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i119 = icmp eq i32 %2317, 0, !dbg !33 + br i1 %.not.i116, label %2323, label %2318, !dbg !33 + +2318: ; preds = %__nv_sqrtf.exit115 + br i1 %.not1.i119, label %2321, label %2319, !dbg !33 + +2319: ; preds = %2318 + %2320 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2020) #6, !dbg !33 + br label %__nv_sqrtf.exit120, !dbg !33 + +2321: ; preds = %2318 + %2322 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2020) #6, !dbg !33 + br label %__nv_sqrtf.exit120, !dbg !33 + +2323: ; preds = %__nv_sqrtf.exit115 + br i1 %.not1.i119, label %2326, label %2324, !dbg !33 + +2324: ; preds = %2323 + %2325 = tail call float @llvm.nvvm.sqrt.rn.f(float %2020) #6, !dbg !33 + br label %__nv_sqrtf.exit120, !dbg !33 + +2326: ; preds = %2323 + %2327 = tail call float @llvm.nvvm.sqrt.approx.f(float %2020) #6, !dbg !33 + br label %__nv_sqrtf.exit120, !dbg !33 + +__nv_sqrtf.exit120: ; preds = %2319, %2321, %2324, %2326 + %.0.i118 = phi float [ %2320, %2319 ], [ %2322, %2321 ], [ %2325, %2324 ], [ %2327, %2326 ], !dbg !33 + %2328 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i121 = icmp eq i32 %2328, 0, !dbg !33 + %2329 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i124 = icmp eq i32 %2329, 0, !dbg !33 + br i1 %.not.i121, label %2335, label %2330, !dbg !33 + +2330: ; preds = %__nv_sqrtf.exit120 + br i1 %.not1.i124, label %2333, label %2331, !dbg !33 + +2331: ; preds = %2330 + %2332 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2021) #6, !dbg !33 + br label %__nv_sqrtf.exit125, !dbg !33 + +2333: ; preds = %2330 + %2334 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2021) #6, !dbg !33 + br label %__nv_sqrtf.exit125, !dbg !33 + +2335: ; preds = %__nv_sqrtf.exit120 + br i1 %.not1.i124, label %2338, label %2336, !dbg !33 + +2336: ; preds = %2335 + %2337 = tail call float @llvm.nvvm.sqrt.rn.f(float %2021) #6, !dbg !33 + br label %__nv_sqrtf.exit125, !dbg !33 + +2338: ; preds = %2335 + %2339 = tail call float @llvm.nvvm.sqrt.approx.f(float %2021) #6, !dbg !33 + br label %__nv_sqrtf.exit125, !dbg !33 + +__nv_sqrtf.exit125: ; preds = %2331, %2333, %2336, %2338 + %.0.i123 = phi float [ %2332, %2331 ], [ %2334, %2333 ], [ %2337, %2336 ], [ %2339, %2338 ], !dbg !33 + %2340 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i126 = icmp eq i32 %2340, 0, !dbg !33 + %2341 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i129 = icmp eq i32 %2341, 0, !dbg !33 + br i1 %.not.i126, label %2347, label %2342, !dbg !33 + +2342: ; preds = %__nv_sqrtf.exit125 + br i1 %.not1.i129, label %2345, label %2343, !dbg !33 + +2343: ; preds = %2342 + %2344 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2022) #6, !dbg !33 + br label %__nv_sqrtf.exit130, !dbg !33 + +2345: ; preds = %2342 + %2346 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2022) #6, !dbg !33 + br label %__nv_sqrtf.exit130, !dbg !33 + +2347: ; preds = %__nv_sqrtf.exit125 + br i1 %.not1.i129, label %2350, label %2348, !dbg !33 + +2348: ; preds = %2347 + %2349 = tail call float @llvm.nvvm.sqrt.rn.f(float %2022) #6, !dbg !33 + br label %__nv_sqrtf.exit130, !dbg !33 + +2350: ; preds = %2347 + %2351 = tail call float @llvm.nvvm.sqrt.approx.f(float %2022) #6, !dbg !33 + br label %__nv_sqrtf.exit130, !dbg !33 + +__nv_sqrtf.exit130: ; preds = %2343, %2345, %2348, %2350 + %.0.i128 = phi float [ %2344, %2343 ], [ %2346, %2345 ], [ %2349, %2348 ], [ %2351, %2350 ], !dbg !33 + %2352 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i131 = icmp eq i32 %2352, 0, !dbg !33 + %2353 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i134 = icmp eq i32 %2353, 0, !dbg !33 + br i1 %.not.i131, label %2359, label %2354, !dbg !33 + +2354: ; preds = %__nv_sqrtf.exit130 + br i1 %.not1.i134, label %2357, label %2355, !dbg !33 + +2355: ; preds = %2354 + %2356 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2023) #6, !dbg !33 + br label %__nv_sqrtf.exit135, !dbg !33 + +2357: ; preds = %2354 + %2358 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2023) #6, !dbg !33 + br label %__nv_sqrtf.exit135, !dbg !33 + +2359: ; preds = %__nv_sqrtf.exit130 + br i1 %.not1.i134, label %2362, label %2360, !dbg !33 + +2360: ; preds = %2359 + %2361 = tail call float @llvm.nvvm.sqrt.rn.f(float %2023) #6, !dbg !33 + br label %__nv_sqrtf.exit135, !dbg !33 + +2362: ; preds = %2359 + %2363 = tail call float @llvm.nvvm.sqrt.approx.f(float %2023) #6, !dbg !33 + br label %__nv_sqrtf.exit135, !dbg !33 + +__nv_sqrtf.exit135: ; preds = %2355, %2357, %2360, %2362 + %.0.i133 = phi float [ %2356, %2355 ], [ %2358, %2357 ], [ %2361, %2360 ], [ %2363, %2362 ], !dbg !33 + %2364 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i136 = icmp eq i32 %2364, 0, !dbg !33 + %2365 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i139 = icmp eq i32 %2365, 0, !dbg !33 + br i1 %.not.i136, label %2371, label %2366, !dbg !33 + +2366: ; preds = %__nv_sqrtf.exit135 + br i1 %.not1.i139, label %2369, label %2367, !dbg !33 + +2367: ; preds = %2366 + %2368 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2024) #6, !dbg !33 + br label %__nv_sqrtf.exit140, !dbg !33 + +2369: ; preds = %2366 + %2370 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2024) #6, !dbg !33 + br label %__nv_sqrtf.exit140, !dbg !33 + +2371: ; preds = %__nv_sqrtf.exit135 + br i1 %.not1.i139, label %2374, label %2372, !dbg !33 + +2372: ; preds = %2371 + %2373 = tail call float @llvm.nvvm.sqrt.rn.f(float %2024) #6, !dbg !33 + br label %__nv_sqrtf.exit140, !dbg !33 + +2374: ; preds = %2371 + %2375 = tail call float @llvm.nvvm.sqrt.approx.f(float %2024) #6, !dbg !33 + br label %__nv_sqrtf.exit140, !dbg !33 + +__nv_sqrtf.exit140: ; preds = %2367, %2369, %2372, %2374 + %.0.i138 = phi float [ %2368, %2367 ], [ %2370, %2369 ], [ %2373, %2372 ], [ %2375, %2374 ], !dbg !33 + %2376 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i141 = icmp eq i32 %2376, 0, !dbg !33 + %2377 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i144 = icmp eq i32 %2377, 0, !dbg !33 + br i1 %.not.i141, label %2383, label %2378, !dbg !33 + +2378: ; preds = %__nv_sqrtf.exit140 + br i1 %.not1.i144, label %2381, label %2379, !dbg !33 + +2379: ; preds = %2378 + %2380 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2025) #6, !dbg !33 + br label %__nv_sqrtf.exit145, !dbg !33 + +2381: ; preds = %2378 + %2382 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2025) #6, !dbg !33 + br label %__nv_sqrtf.exit145, !dbg !33 + +2383: ; preds = %__nv_sqrtf.exit140 + br i1 %.not1.i144, label %2386, label %2384, !dbg !33 + +2384: ; preds = %2383 + %2385 = tail call float @llvm.nvvm.sqrt.rn.f(float %2025) #6, !dbg !33 + br label %__nv_sqrtf.exit145, !dbg !33 + +2386: ; preds = %2383 + %2387 = tail call float @llvm.nvvm.sqrt.approx.f(float %2025) #6, !dbg !33 + br label %__nv_sqrtf.exit145, !dbg !33 + +__nv_sqrtf.exit145: ; preds = %2379, %2381, %2384, %2386 + %.0.i143 = phi float [ %2380, %2379 ], [ %2382, %2381 ], [ %2385, %2384 ], [ %2387, %2386 ], !dbg !33 + %2388 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i146 = icmp eq i32 %2388, 0, !dbg !33 + %2389 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i149 = icmp eq i32 %2389, 0, !dbg !33 + br i1 %.not.i146, label %2395, label %2390, !dbg !33 + +2390: ; preds = %__nv_sqrtf.exit145 + br i1 %.not1.i149, label %2393, label %2391, !dbg !33 + +2391: ; preds = %2390 + %2392 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2026) #6, !dbg !33 + br label %__nv_sqrtf.exit150, !dbg !33 + +2393: ; preds = %2390 + %2394 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2026) #6, !dbg !33 + br label %__nv_sqrtf.exit150, !dbg !33 + +2395: ; preds = %__nv_sqrtf.exit145 + br i1 %.not1.i149, label %2398, label %2396, !dbg !33 + +2396: ; preds = %2395 + %2397 = tail call float @llvm.nvvm.sqrt.rn.f(float %2026) #6, !dbg !33 + br label %__nv_sqrtf.exit150, !dbg !33 + +2398: ; preds = %2395 + %2399 = tail call float @llvm.nvvm.sqrt.approx.f(float %2026) #6, !dbg !33 + br label %__nv_sqrtf.exit150, !dbg !33 + +__nv_sqrtf.exit150: ; preds = %2391, %2393, %2396, %2398 + %.0.i148 = phi float [ %2392, %2391 ], [ %2394, %2393 ], [ %2397, %2396 ], [ %2399, %2398 ], !dbg !33 + %2400 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i151 = icmp eq i32 %2400, 0, !dbg !33 + %2401 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i154 = icmp eq i32 %2401, 0, !dbg !33 + br i1 %.not.i151, label %2407, label %2402, !dbg !33 + +2402: ; preds = %__nv_sqrtf.exit150 + br i1 %.not1.i154, label %2405, label %2403, !dbg !33 + +2403: ; preds = %2402 + %2404 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2027) #6, !dbg !33 + br label %__nv_sqrtf.exit155, !dbg !33 + +2405: ; preds = %2402 + %2406 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2027) #6, !dbg !33 + br label %__nv_sqrtf.exit155, !dbg !33 + +2407: ; preds = %__nv_sqrtf.exit150 + br i1 %.not1.i154, label %2410, label %2408, !dbg !33 + +2408: ; preds = %2407 + %2409 = tail call float @llvm.nvvm.sqrt.rn.f(float %2027) #6, !dbg !33 + br label %__nv_sqrtf.exit155, !dbg !33 + +2410: ; preds = %2407 + %2411 = tail call float @llvm.nvvm.sqrt.approx.f(float %2027) #6, !dbg !33 + br label %__nv_sqrtf.exit155, !dbg !33 + +__nv_sqrtf.exit155: ; preds = %2403, %2405, %2408, %2410 + %.0.i153 = phi float [ %2404, %2403 ], [ %2406, %2405 ], [ %2409, %2408 ], [ %2411, %2410 ], !dbg !33 + %2412 = sext i32 %4 to i64, !dbg !18 + %2413 = shl nuw nsw i32 %11, 1, !dbg !16 + %2414 = and i32 %2413, 126, !dbg !16 + %2415 = or disjoint i32 %2414, 1, !dbg !16 + %2416 = zext nneg i32 %2415 to i64, !dbg !17 + %2417 = or disjoint i64 %10, %2416, !dbg !17 + %2418 = getelementptr float, ptr addrspace(1) %2, i64 %2417, !dbg !34 + %2419 = zext nneg i32 %2414 to i64, !dbg !17 + %2420 = or disjoint i64 %10, %2419, !dbg !17 + %2421 = getelementptr float, ptr addrspace(1) %2, i64 %2420, !dbg !34 + %2422 = getelementptr bfloat, ptr addrspace(1) %1, i64 %106, !dbg !35 + %2423 = getelementptr bfloat, ptr addrspace(1) %2422, i64 %139, !dbg !36 + %2424 = getelementptr bfloat, ptr addrspace(1) %1, i64 %104, !dbg !35 + %2425 = getelementptr bfloat, ptr addrspace(1) %2424, i64 %139, !dbg !36 + %2426 = getelementptr bfloat, ptr addrspace(1) %1, i64 %102, !dbg !35 + %2427 = getelementptr bfloat, ptr addrspace(1) %2426, i64 %139, !dbg !36 + %2428 = getelementptr bfloat, ptr addrspace(1) %1, i64 %100, !dbg !35 + %2429 = getelementptr bfloat, ptr addrspace(1) %2428, i64 %139, !dbg !36 + %2430 = getelementptr bfloat, ptr addrspace(1) %1, i64 %98, !dbg !35 + %2431 = getelementptr bfloat, ptr addrspace(1) %2430, i64 %139, !dbg !36 + %2432 = getelementptr bfloat, ptr addrspace(1) %1, i64 %96, !dbg !35 + %2433 = getelementptr bfloat, ptr addrspace(1) %2432, i64 %139, !dbg !36 + %2434 = getelementptr bfloat, ptr addrspace(1) %1, i64 %94, !dbg !35 + %2435 = getelementptr bfloat, ptr addrspace(1) %2434, i64 %139, !dbg !36 + %2436 = getelementptr bfloat, ptr addrspace(1) %1, i64 %92, !dbg !35 + %2437 = getelementptr bfloat, ptr addrspace(1) %2436, i64 %139, !dbg !36 + %2438 = getelementptr bfloat, ptr addrspace(1) %1, i64 %90, !dbg !35 + %2439 = getelementptr bfloat, ptr addrspace(1) %2438, i64 %139, !dbg !36 + %2440 = getelementptr bfloat, ptr addrspace(1) %1, i64 %88, !dbg !35 + %2441 = getelementptr bfloat, ptr addrspace(1) %2440, i64 %139, !dbg !36 + %2442 = getelementptr bfloat, ptr addrspace(1) %1, i64 %86, !dbg !35 + %2443 = getelementptr bfloat, ptr addrspace(1) %2442, i64 %139, !dbg !36 + %2444 = getelementptr bfloat, ptr addrspace(1) %1, i64 %84, !dbg !35 + %2445 = getelementptr bfloat, ptr addrspace(1) %2444, i64 %139, !dbg !36 + %2446 = getelementptr bfloat, ptr addrspace(1) %1, i64 %82, !dbg !35 + %2447 = getelementptr bfloat, ptr addrspace(1) %2446, i64 %139, !dbg !36 + %2448 = getelementptr bfloat, ptr addrspace(1) %1, i64 %80, !dbg !35 + %2449 = getelementptr bfloat, ptr addrspace(1) %2448, i64 %139, !dbg !36 + %2450 = getelementptr bfloat, ptr addrspace(1) %1, i64 %78, !dbg !35 + %2451 = getelementptr bfloat, ptr addrspace(1) %2450, i64 %139, !dbg !36 + %2452 = getelementptr bfloat, ptr addrspace(1) %1, i64 %76, !dbg !35 + %2453 = getelementptr bfloat, ptr addrspace(1) %2452, i64 %139, !dbg !36 + %2454 = getelementptr bfloat, ptr addrspace(1) %1, i64 %74, !dbg !35 + %2455 = getelementptr bfloat, ptr addrspace(1) %2454, i64 %139, !dbg !36 + %2456 = getelementptr bfloat, ptr addrspace(1) %1, i64 %72, !dbg !35 + %2457 = getelementptr bfloat, ptr addrspace(1) %2456, i64 %139, !dbg !36 + %2458 = getelementptr bfloat, ptr addrspace(1) %1, i64 %70, !dbg !35 + %2459 = getelementptr bfloat, ptr addrspace(1) %2458, i64 %139, !dbg !36 + %2460 = getelementptr bfloat, ptr addrspace(1) %1, i64 %68, !dbg !35 + %2461 = getelementptr bfloat, ptr addrspace(1) %2460, i64 %139, !dbg !36 + %2462 = getelementptr bfloat, ptr addrspace(1) %1, i64 %66, !dbg !35 + %2463 = getelementptr bfloat, ptr addrspace(1) %2462, i64 %139, !dbg !36 + %2464 = getelementptr bfloat, ptr addrspace(1) %1, i64 %64, !dbg !35 + %2465 = getelementptr bfloat, ptr addrspace(1) %2464, i64 %139, !dbg !36 + %2466 = getelementptr bfloat, ptr addrspace(1) %1, i64 %62, !dbg !35 + %2467 = getelementptr bfloat, ptr addrspace(1) %2466, i64 %139, !dbg !36 + %2468 = getelementptr bfloat, ptr addrspace(1) %1, i64 %60, !dbg !35 + %2469 = getelementptr bfloat, ptr addrspace(1) %2468, i64 %139, !dbg !36 + %2470 = getelementptr bfloat, ptr addrspace(1) %1, i64 %58, !dbg !35 + %2471 = getelementptr bfloat, ptr addrspace(1) %2470, i64 %139, !dbg !36 + %2472 = getelementptr bfloat, ptr addrspace(1) %1, i64 %56, !dbg !35 + %2473 = getelementptr bfloat, ptr addrspace(1) %2472, i64 %139, !dbg !36 + %2474 = getelementptr bfloat, ptr addrspace(1) %1, i64 %54, !dbg !35 + %2475 = getelementptr bfloat, ptr addrspace(1) %2474, i64 %139, !dbg !36 + %2476 = getelementptr bfloat, ptr addrspace(1) %1, i64 %52, !dbg !35 + %2477 = getelementptr bfloat, ptr addrspace(1) %2476, i64 %139, !dbg !36 + %2478 = getelementptr bfloat, ptr addrspace(1) %1, i64 %50, !dbg !35 + %2479 = getelementptr bfloat, ptr addrspace(1) %2478, i64 %139, !dbg !36 + %2480 = getelementptr bfloat, ptr addrspace(1) %1, i64 %48, !dbg !35 + %2481 = getelementptr bfloat, ptr addrspace(1) %2480, i64 %139, !dbg !36 + %2482 = getelementptr bfloat, ptr addrspace(1) %1, i64 %46, !dbg !35 + %2483 = getelementptr bfloat, ptr addrspace(1) %2482, i64 %139, !dbg !36 + %2484 = getelementptr bfloat, ptr addrspace(1) %1, i64 %44, !dbg !35 + %2485 = getelementptr bfloat, ptr addrspace(1) %2484, i64 %139, !dbg !36 + %2486 = icmp slt i64 %2417, %2412, !dbg !18 + %2487 = icmp slt i64 %2420, %2412, !dbg !18 + %2488 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i), !dbg !37 + %2489 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i3), !dbg !37 + %2490 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i8), !dbg !37 + %2491 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i13), !dbg !37 + %2492 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i18), !dbg !37 + %2493 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i23), !dbg !37 + %2494 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i28), !dbg !37 + %2495 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i33), !dbg !37 + %2496 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i38), !dbg !37 + %2497 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i43), !dbg !37 + %2498 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i48), !dbg !37 + %2499 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i53), !dbg !37 + %2500 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i58), !dbg !37 + %2501 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i63), !dbg !37 + %2502 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i68), !dbg !37 + %2503 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i73), !dbg !37 + %2504 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i78), !dbg !37 + %2505 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i83), !dbg !37 + %2506 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i88), !dbg !37 + %2507 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i93), !dbg !37 + %2508 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i98), !dbg !37 + %2509 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i103), !dbg !37 + %2510 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i108), !dbg !37 + %2511 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i113), !dbg !37 + %2512 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i118), !dbg !37 + %2513 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i123), !dbg !37 + %2514 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i128), !dbg !37 + %2515 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i133), !dbg !37 + %2516 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i138), !dbg !37 + %2517 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i143), !dbg !37 + %2518 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i148), !dbg !37 + %2519 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i153), !dbg !37 + %2520 = fmul float %2488, %748, !dbg !38 + %2521 = fmul float %2488, %749, !dbg !38 + %2522 = fmul float %2488, %750, !dbg !38 + %2523 = fmul float %2488, %751, !dbg !38 + %2524 = fmul float %2488, %752, !dbg !38 + %2525 = fmul float %2488, %753, !dbg !38 + %2526 = fmul float %2488, %754, !dbg !38 + %2527 = fmul float %2488, %755, !dbg !38 + %2528 = fmul float %2489, %756, !dbg !38 + %2529 = fmul float %2489, %757, !dbg !38 + %2530 = fmul float %2489, %758, !dbg !38 + %2531 = fmul float %2489, %759, !dbg !38 + %2532 = fmul float %2489, %760, !dbg !38 + %2533 = fmul float %2489, %761, !dbg !38 + %2534 = fmul float %2489, %762, !dbg !38 + %2535 = fmul float %2489, %763, !dbg !38 + %2536 = fmul float %2490, %764, !dbg !38 + %2537 = fmul float %2490, %765, !dbg !38 + %2538 = fmul float %2490, %766, !dbg !38 + %2539 = fmul float %2490, %767, !dbg !38 + %2540 = fmul float %2490, %768, !dbg !38 + %2541 = fmul float %2490, %769, !dbg !38 + %2542 = fmul float %2490, %770, !dbg !38 + %2543 = fmul float %2490, %771, !dbg !38 + %2544 = fmul float %2491, %772, !dbg !38 + %2545 = fmul float %2491, %773, !dbg !38 + %2546 = fmul float %2491, %774, !dbg !38 + %2547 = fmul float %2491, %775, !dbg !38 + %2548 = fmul float %2491, %776, !dbg !38 + %2549 = fmul float %2491, %777, !dbg !38 + %2550 = fmul float %2491, %778, !dbg !38 + %2551 = fmul float %2491, %779, !dbg !38 + %2552 = fmul float %2492, %780, !dbg !38 + %2553 = fmul float %2492, %781, !dbg !38 + %2554 = fmul float %2492, %782, !dbg !38 + %2555 = fmul float %2492, %783, !dbg !38 + %2556 = fmul float %2492, %784, !dbg !38 + %2557 = fmul float %2492, %785, !dbg !38 + %2558 = fmul float %2492, %786, !dbg !38 + %2559 = fmul float %2492, %787, !dbg !38 + %2560 = fmul float %2493, %788, !dbg !38 + %2561 = fmul float %2493, %789, !dbg !38 + %2562 = fmul float %2493, %790, !dbg !38 + %2563 = fmul float %2493, %791, !dbg !38 + %2564 = fmul float %2493, %792, !dbg !38 + %2565 = fmul float %2493, %793, !dbg !38 + %2566 = fmul float %2493, %794, !dbg !38 + %2567 = fmul float %2493, %795, !dbg !38 + %2568 = fmul float %2494, %796, !dbg !38 + %2569 = fmul float %2494, %797, !dbg !38 + %2570 = fmul float %2494, %798, !dbg !38 + %2571 = fmul float %2494, %799, !dbg !38 + %2572 = fmul float %2494, %800, !dbg !38 + %2573 = fmul float %2494, %801, !dbg !38 + %2574 = fmul float %2494, %802, !dbg !38 + %2575 = fmul float %2494, %803, !dbg !38 + %2576 = fmul float %2495, %804, !dbg !38 + %2577 = fmul float %2495, %805, !dbg !38 + %2578 = fmul float %2495, %806, !dbg !38 + %2579 = fmul float %2495, %807, !dbg !38 + %2580 = fmul float %2495, %808, !dbg !38 + %2581 = fmul float %2495, %809, !dbg !38 + %2582 = fmul float %2495, %810, !dbg !38 + %2583 = fmul float %2495, %811, !dbg !38 + %2584 = fmul float %2496, %812, !dbg !38 + %2585 = fmul float %2496, %813, !dbg !38 + %2586 = fmul float %2496, %814, !dbg !38 + %2587 = fmul float %2496, %815, !dbg !38 + %2588 = fmul float %2496, %816, !dbg !38 + %2589 = fmul float %2496, %817, !dbg !38 + %2590 = fmul float %2496, %818, !dbg !38 + %2591 = fmul float %2496, %819, !dbg !38 + %2592 = fmul float %2497, %820, !dbg !38 + %2593 = fmul float %2497, %821, !dbg !38 + %2594 = fmul float %2497, %822, !dbg !38 + %2595 = fmul float %2497, %823, !dbg !38 + %2596 = fmul float %2497, %824, !dbg !38 + %2597 = fmul float %2497, %825, !dbg !38 + %2598 = fmul float %2497, %826, !dbg !38 + %2599 = fmul float %2497, %827, !dbg !38 + %2600 = fmul float %2498, %828, !dbg !38 + %2601 = fmul float %2498, %829, !dbg !38 + %2602 = fmul float %2498, %830, !dbg !38 + %2603 = fmul float %2498, %831, !dbg !38 + %2604 = fmul float %2498, %832, !dbg !38 + %2605 = fmul float %2498, %833, !dbg !38 + %2606 = fmul float %2498, %834, !dbg !38 + %2607 = fmul float %2498, %835, !dbg !38 + %2608 = fmul float %2499, %836, !dbg !38 + %2609 = fmul float %2499, %837, !dbg !38 + %2610 = fmul float %2499, %838, !dbg !38 + %2611 = fmul float %2499, %839, !dbg !38 + %2612 = fmul float %2499, %840, !dbg !38 + %2613 = fmul float %2499, %841, !dbg !38 + %2614 = fmul float %2499, %842, !dbg !38 + %2615 = fmul float %2499, %843, !dbg !38 + %2616 = fmul float %2500, %844, !dbg !38 + %2617 = fmul float %2500, %845, !dbg !38 + %2618 = fmul float %2500, %846, !dbg !38 + %2619 = fmul float %2500, %847, !dbg !38 + %2620 = fmul float %2500, %848, !dbg !38 + %2621 = fmul float %2500, %849, !dbg !38 + %2622 = fmul float %2500, %850, !dbg !38 + %2623 = fmul float %2500, %851, !dbg !38 + %2624 = fmul float %2501, %852, !dbg !38 + %2625 = fmul float %2501, %853, !dbg !38 + %2626 = fmul float %2501, %854, !dbg !38 + %2627 = fmul float %2501, %855, !dbg !38 + %2628 = fmul float %2501, %856, !dbg !38 + %2629 = fmul float %2501, %857, !dbg !38 + %2630 = fmul float %2501, %858, !dbg !38 + %2631 = fmul float %2501, %859, !dbg !38 + %2632 = fmul float %2502, %860, !dbg !38 + %2633 = fmul float %2502, %861, !dbg !38 + %2634 = fmul float %2502, %862, !dbg !38 + %2635 = fmul float %2502, %863, !dbg !38 + %2636 = fmul float %2502, %864, !dbg !38 + %2637 = fmul float %2502, %865, !dbg !38 + %2638 = fmul float %2502, %866, !dbg !38 + %2639 = fmul float %2502, %867, !dbg !38 + %2640 = fmul float %2503, %868, !dbg !38 + %2641 = fmul float %2503, %869, !dbg !38 + %2642 = fmul float %2503, %870, !dbg !38 + %2643 = fmul float %2503, %871, !dbg !38 + %2644 = fmul float %2503, %872, !dbg !38 + %2645 = fmul float %2503, %873, !dbg !38 + %2646 = fmul float %2503, %874, !dbg !38 + %2647 = fmul float %2503, %875, !dbg !38 + %2648 = fmul float %2504, %876, !dbg !38 + %2649 = fmul float %2504, %877, !dbg !38 + %2650 = fmul float %2504, %878, !dbg !38 + %2651 = fmul float %2504, %879, !dbg !38 + %2652 = fmul float %2504, %880, !dbg !38 + %2653 = fmul float %2504, %881, !dbg !38 + %2654 = fmul float %2504, %882, !dbg !38 + %2655 = fmul float %2504, %883, !dbg !38 + %2656 = fmul float %2505, %884, !dbg !38 + %2657 = fmul float %2505, %885, !dbg !38 + %2658 = fmul float %2505, %886, !dbg !38 + %2659 = fmul float %2505, %887, !dbg !38 + %2660 = fmul float %2505, %888, !dbg !38 + %2661 = fmul float %2505, %889, !dbg !38 + %2662 = fmul float %2505, %890, !dbg !38 + %2663 = fmul float %2505, %891, !dbg !38 + %2664 = fmul float %2506, %892, !dbg !38 + %2665 = fmul float %2506, %893, !dbg !38 + %2666 = fmul float %2506, %894, !dbg !38 + %2667 = fmul float %2506, %895, !dbg !38 + %2668 = fmul float %2506, %896, !dbg !38 + %2669 = fmul float %2506, %897, !dbg !38 + %2670 = fmul float %2506, %898, !dbg !38 + %2671 = fmul float %2506, %899, !dbg !38 + %2672 = fmul float %2507, %900, !dbg !38 + %2673 = fmul float %2507, %901, !dbg !38 + %2674 = fmul float %2507, %902, !dbg !38 + %2675 = fmul float %2507, %903, !dbg !38 + %2676 = fmul float %2507, %904, !dbg !38 + %2677 = fmul float %2507, %905, !dbg !38 + %2678 = fmul float %2507, %906, !dbg !38 + %2679 = fmul float %2507, %907, !dbg !38 + %2680 = fmul float %2508, %908, !dbg !38 + %2681 = fmul float %2508, %909, !dbg !38 + %2682 = fmul float %2508, %910, !dbg !38 + %2683 = fmul float %2508, %911, !dbg !38 + %2684 = fmul float %2508, %912, !dbg !38 + %2685 = fmul float %2508, %913, !dbg !38 + %2686 = fmul float %2508, %914, !dbg !38 + %2687 = fmul float %2508, %915, !dbg !38 + %2688 = fmul float %2509, %916, !dbg !38 + %2689 = fmul float %2509, %917, !dbg !38 + %2690 = fmul float %2509, %918, !dbg !38 + %2691 = fmul float %2509, %919, !dbg !38 + %2692 = fmul float %2509, %920, !dbg !38 + %2693 = fmul float %2509, %921, !dbg !38 + %2694 = fmul float %2509, %922, !dbg !38 + %2695 = fmul float %2509, %923, !dbg !38 + %2696 = fmul float %2510, %924, !dbg !38 + %2697 = fmul float %2510, %925, !dbg !38 + %2698 = fmul float %2510, %926, !dbg !38 + %2699 = fmul float %2510, %927, !dbg !38 + %2700 = fmul float %2510, %928, !dbg !38 + %2701 = fmul float %2510, %929, !dbg !38 + %2702 = fmul float %2510, %930, !dbg !38 + %2703 = fmul float %2510, %931, !dbg !38 + %2704 = fmul float %2511, %932, !dbg !38 + %2705 = fmul float %2511, %933, !dbg !38 + %2706 = fmul float %2511, %934, !dbg !38 + %2707 = fmul float %2511, %935, !dbg !38 + %2708 = fmul float %2511, %936, !dbg !38 + %2709 = fmul float %2511, %937, !dbg !38 + %2710 = fmul float %2511, %938, !dbg !38 + %2711 = fmul float %2511, %939, !dbg !38 + %2712 = fmul float %2512, %940, !dbg !38 + %2713 = fmul float %2512, %941, !dbg !38 + %2714 = fmul float %2512, %942, !dbg !38 + %2715 = fmul float %2512, %943, !dbg !38 + %2716 = fmul float %2512, %944, !dbg !38 + %2717 = fmul float %2512, %945, !dbg !38 + %2718 = fmul float %2512, %946, !dbg !38 + %2719 = fmul float %2512, %947, !dbg !38 + %2720 = fmul float %2513, %948, !dbg !38 + %2721 = fmul float %2513, %949, !dbg !38 + %2722 = fmul float %2513, %950, !dbg !38 + %2723 = fmul float %2513, %951, !dbg !38 + %2724 = fmul float %2513, %952, !dbg !38 + %2725 = fmul float %2513, %953, !dbg !38 + %2726 = fmul float %2513, %954, !dbg !38 + %2727 = fmul float %2513, %955, !dbg !38 + %2728 = fmul float %2514, %956, !dbg !38 + %2729 = fmul float %2514, %957, !dbg !38 + %2730 = fmul float %2514, %958, !dbg !38 + %2731 = fmul float %2514, %959, !dbg !38 + %2732 = fmul float %2514, %960, !dbg !38 + %2733 = fmul float %2514, %961, !dbg !38 + %2734 = fmul float %2514, %962, !dbg !38 + %2735 = fmul float %2514, %963, !dbg !38 + %2736 = fmul float %2515, %964, !dbg !38 + %2737 = fmul float %2515, %965, !dbg !38 + %2738 = fmul float %2515, %966, !dbg !38 + %2739 = fmul float %2515, %967, !dbg !38 + %2740 = fmul float %2515, %968, !dbg !38 + %2741 = fmul float %2515, %969, !dbg !38 + %2742 = fmul float %2515, %970, !dbg !38 + %2743 = fmul float %2515, %971, !dbg !38 + %2744 = fmul float %2516, %972, !dbg !38 + %2745 = fmul float %2516, %973, !dbg !38 + %2746 = fmul float %2516, %974, !dbg !38 + %2747 = fmul float %2516, %975, !dbg !38 + %2748 = fmul float %2516, %976, !dbg !38 + %2749 = fmul float %2516, %977, !dbg !38 + %2750 = fmul float %2516, %978, !dbg !38 + %2751 = fmul float %2516, %979, !dbg !38 + %2752 = fmul float %2517, %980, !dbg !38 + %2753 = fmul float %2517, %981, !dbg !38 + %2754 = fmul float %2517, %982, !dbg !38 + %2755 = fmul float %2517, %983, !dbg !38 + %2756 = fmul float %2517, %984, !dbg !38 + %2757 = fmul float %2517, %985, !dbg !38 + %2758 = fmul float %2517, %986, !dbg !38 + %2759 = fmul float %2517, %987, !dbg !38 + %2760 = fmul float %2518, %988, !dbg !38 + %2761 = fmul float %2518, %989, !dbg !38 + %2762 = fmul float %2518, %990, !dbg !38 + %2763 = fmul float %2518, %991, !dbg !38 + %2764 = fmul float %2518, %992, !dbg !38 + %2765 = fmul float %2518, %993, !dbg !38 + %2766 = fmul float %2518, %994, !dbg !38 + %2767 = fmul float %2518, %995, !dbg !38 + %2768 = fmul float %2519, %996, !dbg !38 + %2769 = fmul float %2519, %997, !dbg !38 + %2770 = fmul float %2519, %998, !dbg !38 + %2771 = fmul float %2519, %999, !dbg !38 + %2772 = fmul float %2519, %1000, !dbg !38 + %2773 = fmul float %2519, %1001, !dbg !38 + %2774 = fmul float %2519, %1002, !dbg !38 + %2775 = fmul float %2519, %1003, !dbg !38 + %2776 = fptrunc float %2520 to bfloat, !dbg !39 + %2777 = fptrunc float %2521 to bfloat, !dbg !39 + %2778 = fptrunc float %2522 to bfloat, !dbg !39 + %2779 = fptrunc float %2523 to bfloat, !dbg !39 + %2780 = fptrunc float %2524 to bfloat, !dbg !39 + %2781 = fptrunc float %2525 to bfloat, !dbg !39 + %2782 = fptrunc float %2526 to bfloat, !dbg !39 + %2783 = fptrunc float %2527 to bfloat, !dbg !39 + %2784 = fptrunc float %2528 to bfloat, !dbg !39 + %2785 = fptrunc float %2529 to bfloat, !dbg !39 + %2786 = fptrunc float %2530 to bfloat, !dbg !39 + %2787 = fptrunc float %2531 to bfloat, !dbg !39 + %2788 = fptrunc float %2532 to bfloat, !dbg !39 + %2789 = fptrunc float %2533 to bfloat, !dbg !39 + %2790 = fptrunc float %2534 to bfloat, !dbg !39 + %2791 = fptrunc float %2535 to bfloat, !dbg !39 + %2792 = fptrunc float %2536 to bfloat, !dbg !39 + %2793 = fptrunc float %2537 to bfloat, !dbg !39 + %2794 = fptrunc float %2538 to bfloat, !dbg !39 + %2795 = fptrunc float %2539 to bfloat, !dbg !39 + %2796 = fptrunc float %2540 to bfloat, !dbg !39 + %2797 = fptrunc float %2541 to bfloat, !dbg !39 + %2798 = fptrunc float %2542 to bfloat, !dbg !39 + %2799 = fptrunc float %2543 to bfloat, !dbg !39 + %2800 = fptrunc float %2544 to bfloat, !dbg !39 + %2801 = fptrunc float %2545 to bfloat, !dbg !39 + %2802 = fptrunc float %2546 to bfloat, !dbg !39 + %2803 = fptrunc float %2547 to bfloat, !dbg !39 + %2804 = fptrunc float %2548 to bfloat, !dbg !39 + %2805 = fptrunc float %2549 to bfloat, !dbg !39 + %2806 = fptrunc float %2550 to bfloat, !dbg !39 + %2807 = fptrunc float %2551 to bfloat, !dbg !39 + %2808 = fptrunc float %2552 to bfloat, !dbg !39 + %2809 = fptrunc float %2553 to bfloat, !dbg !39 + %2810 = fptrunc float %2554 to bfloat, !dbg !39 + %2811 = fptrunc float %2555 to bfloat, !dbg !39 + %2812 = fptrunc float %2556 to bfloat, !dbg !39 + %2813 = fptrunc float %2557 to bfloat, !dbg !39 + %2814 = fptrunc float %2558 to bfloat, !dbg !39 + %2815 = fptrunc float %2559 to bfloat, !dbg !39 + %2816 = fptrunc float %2560 to bfloat, !dbg !39 + %2817 = fptrunc float %2561 to bfloat, !dbg !39 + %2818 = fptrunc float %2562 to bfloat, !dbg !39 + %2819 = fptrunc float %2563 to bfloat, !dbg !39 + %2820 = fptrunc float %2564 to bfloat, !dbg !39 + %2821 = fptrunc float %2565 to bfloat, !dbg !39 + %2822 = fptrunc float %2566 to bfloat, !dbg !39 + %2823 = fptrunc float %2567 to bfloat, !dbg !39 + %2824 = fptrunc float %2568 to bfloat, !dbg !39 + %2825 = fptrunc float %2569 to bfloat, !dbg !39 + %2826 = fptrunc float %2570 to bfloat, !dbg !39 + %2827 = fptrunc float %2571 to bfloat, !dbg !39 + %2828 = fptrunc float %2572 to bfloat, !dbg !39 + %2829 = fptrunc float %2573 to bfloat, !dbg !39 + %2830 = fptrunc float %2574 to bfloat, !dbg !39 + %2831 = fptrunc float %2575 to bfloat, !dbg !39 + %2832 = fptrunc float %2576 to bfloat, !dbg !39 + %2833 = fptrunc float %2577 to bfloat, !dbg !39 + %2834 = fptrunc float %2578 to bfloat, !dbg !39 + %2835 = fptrunc float %2579 to bfloat, !dbg !39 + %2836 = fptrunc float %2580 to bfloat, !dbg !39 + %2837 = fptrunc float %2581 to bfloat, !dbg !39 + %2838 = fptrunc float %2582 to bfloat, !dbg !39 + %2839 = fptrunc float %2583 to bfloat, !dbg !39 + %2840 = fptrunc float %2584 to bfloat, !dbg !39 + %2841 = fptrunc float %2585 to bfloat, !dbg !39 + %2842 = fptrunc float %2586 to bfloat, !dbg !39 + %2843 = fptrunc float %2587 to bfloat, !dbg !39 + %2844 = fptrunc float %2588 to bfloat, !dbg !39 + %2845 = fptrunc float %2589 to bfloat, !dbg !39 + %2846 = fptrunc float %2590 to bfloat, !dbg !39 + %2847 = fptrunc float %2591 to bfloat, !dbg !39 + %2848 = fptrunc float %2592 to bfloat, !dbg !39 + %2849 = fptrunc float %2593 to bfloat, !dbg !39 + %2850 = fptrunc float %2594 to bfloat, !dbg !39 + %2851 = fptrunc float %2595 to bfloat, !dbg !39 + %2852 = fptrunc float %2596 to bfloat, !dbg !39 + %2853 = fptrunc float %2597 to bfloat, !dbg !39 + %2854 = fptrunc float %2598 to bfloat, !dbg !39 + %2855 = fptrunc float %2599 to bfloat, !dbg !39 + %2856 = fptrunc float %2600 to bfloat, !dbg !39 + %2857 = fptrunc float %2601 to bfloat, !dbg !39 + %2858 = fptrunc float %2602 to bfloat, !dbg !39 + %2859 = fptrunc float %2603 to bfloat, !dbg !39 + %2860 = fptrunc float %2604 to bfloat, !dbg !39 + %2861 = fptrunc float %2605 to bfloat, !dbg !39 + %2862 = fptrunc float %2606 to bfloat, !dbg !39 + %2863 = fptrunc float %2607 to bfloat, !dbg !39 + %2864 = fptrunc float %2608 to bfloat, !dbg !39 + %2865 = fptrunc float %2609 to bfloat, !dbg !39 + %2866 = fptrunc float %2610 to bfloat, !dbg !39 + %2867 = fptrunc float %2611 to bfloat, !dbg !39 + %2868 = fptrunc float %2612 to bfloat, !dbg !39 + %2869 = fptrunc float %2613 to bfloat, !dbg !39 + %2870 = fptrunc float %2614 to bfloat, !dbg !39 + %2871 = fptrunc float %2615 to bfloat, !dbg !39 + %2872 = fptrunc float %2616 to bfloat, !dbg !39 + %2873 = fptrunc float %2617 to bfloat, !dbg !39 + %2874 = fptrunc float %2618 to bfloat, !dbg !39 + %2875 = fptrunc float %2619 to bfloat, !dbg !39 + %2876 = fptrunc float %2620 to bfloat, !dbg !39 + %2877 = fptrunc float %2621 to bfloat, !dbg !39 + %2878 = fptrunc float %2622 to bfloat, !dbg !39 + %2879 = fptrunc float %2623 to bfloat, !dbg !39 + %2880 = fptrunc float %2624 to bfloat, !dbg !39 + %2881 = fptrunc float %2625 to bfloat, !dbg !39 + %2882 = fptrunc float %2626 to bfloat, !dbg !39 + %2883 = fptrunc float %2627 to bfloat, !dbg !39 + %2884 = fptrunc float %2628 to bfloat, !dbg !39 + %2885 = fptrunc float %2629 to bfloat, !dbg !39 + %2886 = fptrunc float %2630 to bfloat, !dbg !39 + %2887 = fptrunc float %2631 to bfloat, !dbg !39 + %2888 = fptrunc float %2632 to bfloat, !dbg !39 + %2889 = fptrunc float %2633 to bfloat, !dbg !39 + %2890 = fptrunc float %2634 to bfloat, !dbg !39 + %2891 = fptrunc float %2635 to bfloat, !dbg !39 + %2892 = fptrunc float %2636 to bfloat, !dbg !39 + %2893 = fptrunc float %2637 to bfloat, !dbg !39 + %2894 = fptrunc float %2638 to bfloat, !dbg !39 + %2895 = fptrunc float %2639 to bfloat, !dbg !39 + %2896 = fptrunc float %2640 to bfloat, !dbg !39 + %2897 = fptrunc float %2641 to bfloat, !dbg !39 + %2898 = fptrunc float %2642 to bfloat, !dbg !39 + %2899 = fptrunc float %2643 to bfloat, !dbg !39 + %2900 = fptrunc float %2644 to bfloat, !dbg !39 + %2901 = fptrunc float %2645 to bfloat, !dbg !39 + %2902 = fptrunc float %2646 to bfloat, !dbg !39 + %2903 = fptrunc float %2647 to bfloat, !dbg !39 + %2904 = fptrunc float %2648 to bfloat, !dbg !39 + %2905 = fptrunc float %2649 to bfloat, !dbg !39 + %2906 = fptrunc float %2650 to bfloat, !dbg !39 + %2907 = fptrunc float %2651 to bfloat, !dbg !39 + %2908 = fptrunc float %2652 to bfloat, !dbg !39 + %2909 = fptrunc float %2653 to bfloat, !dbg !39 + %2910 = fptrunc float %2654 to bfloat, !dbg !39 + %2911 = fptrunc float %2655 to bfloat, !dbg !39 + %2912 = fptrunc float %2656 to bfloat, !dbg !39 + %2913 = fptrunc float %2657 to bfloat, !dbg !39 + %2914 = fptrunc float %2658 to bfloat, !dbg !39 + %2915 = fptrunc float %2659 to bfloat, !dbg !39 + %2916 = fptrunc float %2660 to bfloat, !dbg !39 + %2917 = fptrunc float %2661 to bfloat, !dbg !39 + %2918 = fptrunc float %2662 to bfloat, !dbg !39 + %2919 = fptrunc float %2663 to bfloat, !dbg !39 + %2920 = fptrunc float %2664 to bfloat, !dbg !39 + %2921 = fptrunc float %2665 to bfloat, !dbg !39 + %2922 = fptrunc float %2666 to bfloat, !dbg !39 + %2923 = fptrunc float %2667 to bfloat, !dbg !39 + %2924 = fptrunc float %2668 to bfloat, !dbg !39 + %2925 = fptrunc float %2669 to bfloat, !dbg !39 + %2926 = fptrunc float %2670 to bfloat, !dbg !39 + %2927 = fptrunc float %2671 to bfloat, !dbg !39 + %2928 = fptrunc float %2672 to bfloat, !dbg !39 + %2929 = fptrunc float %2673 to bfloat, !dbg !39 + %2930 = fptrunc float %2674 to bfloat, !dbg !39 + %2931 = fptrunc float %2675 to bfloat, !dbg !39 + %2932 = fptrunc float %2676 to bfloat, !dbg !39 + %2933 = fptrunc float %2677 to bfloat, !dbg !39 + %2934 = fptrunc float %2678 to bfloat, !dbg !39 + %2935 = fptrunc float %2679 to bfloat, !dbg !39 + %2936 = fptrunc float %2680 to bfloat, !dbg !39 + %2937 = fptrunc float %2681 to bfloat, !dbg !39 + %2938 = fptrunc float %2682 to bfloat, !dbg !39 + %2939 = fptrunc float %2683 to bfloat, !dbg !39 + %2940 = fptrunc float %2684 to bfloat, !dbg !39 + %2941 = fptrunc float %2685 to bfloat, !dbg !39 + %2942 = fptrunc float %2686 to bfloat, !dbg !39 + %2943 = fptrunc float %2687 to bfloat, !dbg !39 + %2944 = fptrunc float %2688 to bfloat, !dbg !39 + %2945 = fptrunc float %2689 to bfloat, !dbg !39 + %2946 = fptrunc float %2690 to bfloat, !dbg !39 + %2947 = fptrunc float %2691 to bfloat, !dbg !39 + %2948 = fptrunc float %2692 to bfloat, !dbg !39 + %2949 = fptrunc float %2693 to bfloat, !dbg !39 + %2950 = fptrunc float %2694 to bfloat, !dbg !39 + %2951 = fptrunc float %2695 to bfloat, !dbg !39 + %2952 = fptrunc float %2696 to bfloat, !dbg !39 + %2953 = fptrunc float %2697 to bfloat, !dbg !39 + %2954 = fptrunc float %2698 to bfloat, !dbg !39 + %2955 = fptrunc float %2699 to bfloat, !dbg !39 + %2956 = fptrunc float %2700 to bfloat, !dbg !39 + %2957 = fptrunc float %2701 to bfloat, !dbg !39 + %2958 = fptrunc float %2702 to bfloat, !dbg !39 + %2959 = fptrunc float %2703 to bfloat, !dbg !39 + %2960 = fptrunc float %2704 to bfloat, !dbg !39 + %2961 = fptrunc float %2705 to bfloat, !dbg !39 + %2962 = fptrunc float %2706 to bfloat, !dbg !39 + %2963 = fptrunc float %2707 to bfloat, !dbg !39 + %2964 = fptrunc float %2708 to bfloat, !dbg !39 + %2965 = fptrunc float %2709 to bfloat, !dbg !39 + %2966 = fptrunc float %2710 to bfloat, !dbg !39 + %2967 = fptrunc float %2711 to bfloat, !dbg !39 + %2968 = fptrunc float %2712 to bfloat, !dbg !39 + %2969 = fptrunc float %2713 to bfloat, !dbg !39 + %2970 = fptrunc float %2714 to bfloat, !dbg !39 + %2971 = fptrunc float %2715 to bfloat, !dbg !39 + %2972 = fptrunc float %2716 to bfloat, !dbg !39 + %2973 = fptrunc float %2717 to bfloat, !dbg !39 + %2974 = fptrunc float %2718 to bfloat, !dbg !39 + %2975 = fptrunc float %2719 to bfloat, !dbg !39 + %2976 = fptrunc float %2720 to bfloat, !dbg !39 + %2977 = fptrunc float %2721 to bfloat, !dbg !39 + %2978 = fptrunc float %2722 to bfloat, !dbg !39 + %2979 = fptrunc float %2723 to bfloat, !dbg !39 + %2980 = fptrunc float %2724 to bfloat, !dbg !39 + %2981 = fptrunc float %2725 to bfloat, !dbg !39 + %2982 = fptrunc float %2726 to bfloat, !dbg !39 + %2983 = fptrunc float %2727 to bfloat, !dbg !39 + %2984 = fptrunc float %2728 to bfloat, !dbg !39 + %2985 = fptrunc float %2729 to bfloat, !dbg !39 + %2986 = fptrunc float %2730 to bfloat, !dbg !39 + %2987 = fptrunc float %2731 to bfloat, !dbg !39 + %2988 = fptrunc float %2732 to bfloat, !dbg !39 + %2989 = fptrunc float %2733 to bfloat, !dbg !39 + %2990 = fptrunc float %2734 to bfloat, !dbg !39 + %2991 = fptrunc float %2735 to bfloat, !dbg !39 + %2992 = fptrunc float %2736 to bfloat, !dbg !39 + %2993 = fptrunc float %2737 to bfloat, !dbg !39 + %2994 = fptrunc float %2738 to bfloat, !dbg !39 + %2995 = fptrunc float %2739 to bfloat, !dbg !39 + %2996 = fptrunc float %2740 to bfloat, !dbg !39 + %2997 = fptrunc float %2741 to bfloat, !dbg !39 + %2998 = fptrunc float %2742 to bfloat, !dbg !39 + %2999 = fptrunc float %2743 to bfloat, !dbg !39 + %3000 = fptrunc float %2744 to bfloat, !dbg !39 + %3001 = fptrunc float %2745 to bfloat, !dbg !39 + %3002 = fptrunc float %2746 to bfloat, !dbg !39 + %3003 = fptrunc float %2747 to bfloat, !dbg !39 + %3004 = fptrunc float %2748 to bfloat, !dbg !39 + %3005 = fptrunc float %2749 to bfloat, !dbg !39 + %3006 = fptrunc float %2750 to bfloat, !dbg !39 + %3007 = fptrunc float %2751 to bfloat, !dbg !39 + %3008 = fptrunc float %2752 to bfloat, !dbg !39 + %3009 = fptrunc float %2753 to bfloat, !dbg !39 + %3010 = fptrunc float %2754 to bfloat, !dbg !39 + %3011 = fptrunc float %2755 to bfloat, !dbg !39 + %3012 = fptrunc float %2756 to bfloat, !dbg !39 + %3013 = fptrunc float %2757 to bfloat, !dbg !39 + %3014 = fptrunc float %2758 to bfloat, !dbg !39 + %3015 = fptrunc float %2759 to bfloat, !dbg !39 + %3016 = fptrunc float %2760 to bfloat, !dbg !39 + %3017 = fptrunc float %2761 to bfloat, !dbg !39 + %3018 = fptrunc float %2762 to bfloat, !dbg !39 + %3019 = fptrunc float %2763 to bfloat, !dbg !39 + %3020 = fptrunc float %2764 to bfloat, !dbg !39 + %3021 = fptrunc float %2765 to bfloat, !dbg !39 + %3022 = fptrunc float %2766 to bfloat, !dbg !39 + %3023 = fptrunc float %2767 to bfloat, !dbg !39 + %3024 = fptrunc float %2768 to bfloat, !dbg !39 + %3025 = fptrunc float %2769 to bfloat, !dbg !39 + %3026 = fptrunc float %2770 to bfloat, !dbg !39 + %3027 = fptrunc float %2771 to bfloat, !dbg !39 + %3028 = fptrunc float %2772 to bfloat, !dbg !39 + %3029 = fptrunc float %2773 to bfloat, !dbg !39 + %3030 = fptrunc float %2774 to bfloat, !dbg !39 + %3031 = fptrunc float %2775 to bfloat, !dbg !39 + %3032 = insertelement <2 x bfloat> poison, bfloat %2776, i64 0, !dbg !40 + %3033 = insertelement <2 x bfloat> %3032, bfloat %2777, i64 1, !dbg !40 + %3034 = bitcast <2 x bfloat> %3033 to i32, !dbg !40 + %3035 = insertelement <2 x bfloat> poison, bfloat %2778, i64 0, !dbg !40 + %3036 = insertelement <2 x bfloat> %3035, bfloat %2779, i64 1, !dbg !40 + %3037 = bitcast <2 x bfloat> %3036 to i32, !dbg !40 + %3038 = insertelement <2 x bfloat> poison, bfloat %2780, i64 0, !dbg !40 + %3039 = insertelement <2 x bfloat> %3038, bfloat %2781, i64 1, !dbg !40 + %3040 = bitcast <2 x bfloat> %3039 to i32, !dbg !40 + %3041 = insertelement <2 x bfloat> poison, bfloat %2782, i64 0, !dbg !40 + %3042 = insertelement <2 x bfloat> %3041, bfloat %2783, i64 1, !dbg !40 + %3043 = bitcast <2 x bfloat> %3042 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3034, i32 %3037, i32 %3040, i32 %3043, ptr addrspace(1) %2485, i1 %172) #6, !dbg !40 + %3044 = insertelement <2 x bfloat> poison, bfloat %2784, i64 0, !dbg !40 + %3045 = insertelement <2 x bfloat> %3044, bfloat %2785, i64 1, !dbg !40 + %3046 = bitcast <2 x bfloat> %3045 to i32, !dbg !40 + %3047 = insertelement <2 x bfloat> poison, bfloat %2786, i64 0, !dbg !40 + %3048 = insertelement <2 x bfloat> %3047, bfloat %2787, i64 1, !dbg !40 + %3049 = bitcast <2 x bfloat> %3048 to i32, !dbg !40 + %3050 = insertelement <2 x bfloat> poison, bfloat %2788, i64 0, !dbg !40 + %3051 = insertelement <2 x bfloat> %3050, bfloat %2789, i64 1, !dbg !40 + %3052 = bitcast <2 x bfloat> %3051 to i32, !dbg !40 + %3053 = insertelement <2 x bfloat> poison, bfloat %2790, i64 0, !dbg !40 + %3054 = insertelement <2 x bfloat> %3053, bfloat %2791, i64 1, !dbg !40 + %3055 = bitcast <2 x bfloat> %3054 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3046, i32 %3049, i32 %3052, i32 %3055, ptr addrspace(1) %2483, i1 %190) #6, !dbg !40 + %3056 = insertelement <2 x bfloat> poison, bfloat %2792, i64 0, !dbg !40 + %3057 = insertelement <2 x bfloat> %3056, bfloat %2793, i64 1, !dbg !40 + %3058 = bitcast <2 x bfloat> %3057 to i32, !dbg !40 + %3059 = insertelement <2 x bfloat> poison, bfloat %2794, i64 0, !dbg !40 + %3060 = insertelement <2 x bfloat> %3059, bfloat %2795, i64 1, !dbg !40 + %3061 = bitcast <2 x bfloat> %3060 to i32, !dbg !40 + %3062 = insertelement <2 x bfloat> poison, bfloat %2796, i64 0, !dbg !40 + %3063 = insertelement <2 x bfloat> %3062, bfloat %2797, i64 1, !dbg !40 + %3064 = bitcast <2 x bfloat> %3063 to i32, !dbg !40 + %3065 = insertelement <2 x bfloat> poison, bfloat %2798, i64 0, !dbg !40 + %3066 = insertelement <2 x bfloat> %3065, bfloat %2799, i64 1, !dbg !40 + %3067 = bitcast <2 x bfloat> %3066 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3058, i32 %3061, i32 %3064, i32 %3067, ptr addrspace(1) %2481, i1 %208) #6, !dbg !40 + %3068 = insertelement <2 x bfloat> poison, bfloat %2800, i64 0, !dbg !40 + %3069 = insertelement <2 x bfloat> %3068, bfloat %2801, i64 1, !dbg !40 + %3070 = bitcast <2 x bfloat> %3069 to i32, !dbg !40 + %3071 = insertelement <2 x bfloat> poison, bfloat %2802, i64 0, !dbg !40 + %3072 = insertelement <2 x bfloat> %3071, bfloat %2803, i64 1, !dbg !40 + %3073 = bitcast <2 x bfloat> %3072 to i32, !dbg !40 + %3074 = insertelement <2 x bfloat> poison, bfloat %2804, i64 0, !dbg !40 + %3075 = insertelement <2 x bfloat> %3074, bfloat %2805, i64 1, !dbg !40 + %3076 = bitcast <2 x bfloat> %3075 to i32, !dbg !40 + %3077 = insertelement <2 x bfloat> poison, bfloat %2806, i64 0, !dbg !40 + %3078 = insertelement <2 x bfloat> %3077, bfloat %2807, i64 1, !dbg !40 + %3079 = bitcast <2 x bfloat> %3078 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3070, i32 %3073, i32 %3076, i32 %3079, ptr addrspace(1) %2479, i1 %226) #6, !dbg !40 + %3080 = insertelement <2 x bfloat> poison, bfloat %2808, i64 0, !dbg !40 + %3081 = insertelement <2 x bfloat> %3080, bfloat %2809, i64 1, !dbg !40 + %3082 = bitcast <2 x bfloat> %3081 to i32, !dbg !40 + %3083 = insertelement <2 x bfloat> poison, bfloat %2810, i64 0, !dbg !40 + %3084 = insertelement <2 x bfloat> %3083, bfloat %2811, i64 1, !dbg !40 + %3085 = bitcast <2 x bfloat> %3084 to i32, !dbg !40 + %3086 = insertelement <2 x bfloat> poison, bfloat %2812, i64 0, !dbg !40 + %3087 = insertelement <2 x bfloat> %3086, bfloat %2813, i64 1, !dbg !40 + %3088 = bitcast <2 x bfloat> %3087 to i32, !dbg !40 + %3089 = insertelement <2 x bfloat> poison, bfloat %2814, i64 0, !dbg !40 + %3090 = insertelement <2 x bfloat> %3089, bfloat %2815, i64 1, !dbg !40 + %3091 = bitcast <2 x bfloat> %3090 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3082, i32 %3085, i32 %3088, i32 %3091, ptr addrspace(1) %2477, i1 %244) #6, !dbg !40 + %3092 = insertelement <2 x bfloat> poison, bfloat %2816, i64 0, !dbg !40 + %3093 = insertelement <2 x bfloat> %3092, bfloat %2817, i64 1, !dbg !40 + %3094 = bitcast <2 x bfloat> %3093 to i32, !dbg !40 + %3095 = insertelement <2 x bfloat> poison, bfloat %2818, i64 0, !dbg !40 + %3096 = insertelement <2 x bfloat> %3095, bfloat %2819, i64 1, !dbg !40 + %3097 = bitcast <2 x bfloat> %3096 to i32, !dbg !40 + %3098 = insertelement <2 x bfloat> poison, bfloat %2820, i64 0, !dbg !40 + %3099 = insertelement <2 x bfloat> %3098, bfloat %2821, i64 1, !dbg !40 + %3100 = bitcast <2 x bfloat> %3099 to i32, !dbg !40 + %3101 = insertelement <2 x bfloat> poison, bfloat %2822, i64 0, !dbg !40 + %3102 = insertelement <2 x bfloat> %3101, bfloat %2823, i64 1, !dbg !40 + %3103 = bitcast <2 x bfloat> %3102 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3094, i32 %3097, i32 %3100, i32 %3103, ptr addrspace(1) %2475, i1 %262) #6, !dbg !40 + %3104 = insertelement <2 x bfloat> poison, bfloat %2824, i64 0, !dbg !40 + %3105 = insertelement <2 x bfloat> %3104, bfloat %2825, i64 1, !dbg !40 + %3106 = bitcast <2 x bfloat> %3105 to i32, !dbg !40 + %3107 = insertelement <2 x bfloat> poison, bfloat %2826, i64 0, !dbg !40 + %3108 = insertelement <2 x bfloat> %3107, bfloat %2827, i64 1, !dbg !40 + %3109 = bitcast <2 x bfloat> %3108 to i32, !dbg !40 + %3110 = insertelement <2 x bfloat> poison, bfloat %2828, i64 0, !dbg !40 + %3111 = insertelement <2 x bfloat> %3110, bfloat %2829, i64 1, !dbg !40 + %3112 = bitcast <2 x bfloat> %3111 to i32, !dbg !40 + %3113 = insertelement <2 x bfloat> poison, bfloat %2830, i64 0, !dbg !40 + %3114 = insertelement <2 x bfloat> %3113, bfloat %2831, i64 1, !dbg !40 + %3115 = bitcast <2 x bfloat> %3114 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3106, i32 %3109, i32 %3112, i32 %3115, ptr addrspace(1) %2473, i1 %280) #6, !dbg !40 + %3116 = insertelement <2 x bfloat> poison, bfloat %2832, i64 0, !dbg !40 + %3117 = insertelement <2 x bfloat> %3116, bfloat %2833, i64 1, !dbg !40 + %3118 = bitcast <2 x bfloat> %3117 to i32, !dbg !40 + %3119 = insertelement <2 x bfloat> poison, bfloat %2834, i64 0, !dbg !40 + %3120 = insertelement <2 x bfloat> %3119, bfloat %2835, i64 1, !dbg !40 + %3121 = bitcast <2 x bfloat> %3120 to i32, !dbg !40 + %3122 = insertelement <2 x bfloat> poison, bfloat %2836, i64 0, !dbg !40 + %3123 = insertelement <2 x bfloat> %3122, bfloat %2837, i64 1, !dbg !40 + %3124 = bitcast <2 x bfloat> %3123 to i32, !dbg !40 + %3125 = insertelement <2 x bfloat> poison, bfloat %2838, i64 0, !dbg !40 + %3126 = insertelement <2 x bfloat> %3125, bfloat %2839, i64 1, !dbg !40 + %3127 = bitcast <2 x bfloat> %3126 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3118, i32 %3121, i32 %3124, i32 %3127, ptr addrspace(1) %2471, i1 %298) #6, !dbg !40 + %3128 = insertelement <2 x bfloat> poison, bfloat %2840, i64 0, !dbg !40 + %3129 = insertelement <2 x bfloat> %3128, bfloat %2841, i64 1, !dbg !40 + %3130 = bitcast <2 x bfloat> %3129 to i32, !dbg !40 + %3131 = insertelement <2 x bfloat> poison, bfloat %2842, i64 0, !dbg !40 + %3132 = insertelement <2 x bfloat> %3131, bfloat %2843, i64 1, !dbg !40 + %3133 = bitcast <2 x bfloat> %3132 to i32, !dbg !40 + %3134 = insertelement <2 x bfloat> poison, bfloat %2844, i64 0, !dbg !40 + %3135 = insertelement <2 x bfloat> %3134, bfloat %2845, i64 1, !dbg !40 + %3136 = bitcast <2 x bfloat> %3135 to i32, !dbg !40 + %3137 = insertelement <2 x bfloat> poison, bfloat %2846, i64 0, !dbg !40 + %3138 = insertelement <2 x bfloat> %3137, bfloat %2847, i64 1, !dbg !40 + %3139 = bitcast <2 x bfloat> %3138 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3130, i32 %3133, i32 %3136, i32 %3139, ptr addrspace(1) %2469, i1 %316) #6, !dbg !40 + %3140 = insertelement <2 x bfloat> poison, bfloat %2848, i64 0, !dbg !40 + %3141 = insertelement <2 x bfloat> %3140, bfloat %2849, i64 1, !dbg !40 + %3142 = bitcast <2 x bfloat> %3141 to i32, !dbg !40 + %3143 = insertelement <2 x bfloat> poison, bfloat %2850, i64 0, !dbg !40 + %3144 = insertelement <2 x bfloat> %3143, bfloat %2851, i64 1, !dbg !40 + %3145 = bitcast <2 x bfloat> %3144 to i32, !dbg !40 + %3146 = insertelement <2 x bfloat> poison, bfloat %2852, i64 0, !dbg !40 + %3147 = insertelement <2 x bfloat> %3146, bfloat %2853, i64 1, !dbg !40 + %3148 = bitcast <2 x bfloat> %3147 to i32, !dbg !40 + %3149 = insertelement <2 x bfloat> poison, bfloat %2854, i64 0, !dbg !40 + %3150 = insertelement <2 x bfloat> %3149, bfloat %2855, i64 1, !dbg !40 + %3151 = bitcast <2 x bfloat> %3150 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3142, i32 %3145, i32 %3148, i32 %3151, ptr addrspace(1) %2467, i1 %334) #6, !dbg !40 + %3152 = insertelement <2 x bfloat> poison, bfloat %2856, i64 0, !dbg !40 + %3153 = insertelement <2 x bfloat> %3152, bfloat %2857, i64 1, !dbg !40 + %3154 = bitcast <2 x bfloat> %3153 to i32, !dbg !40 + %3155 = insertelement <2 x bfloat> poison, bfloat %2858, i64 0, !dbg !40 + %3156 = insertelement <2 x bfloat> %3155, bfloat %2859, i64 1, !dbg !40 + %3157 = bitcast <2 x bfloat> %3156 to i32, !dbg !40 + %3158 = insertelement <2 x bfloat> poison, bfloat %2860, i64 0, !dbg !40 + %3159 = insertelement <2 x bfloat> %3158, bfloat %2861, i64 1, !dbg !40 + %3160 = bitcast <2 x bfloat> %3159 to i32, !dbg !40 + %3161 = insertelement <2 x bfloat> poison, bfloat %2862, i64 0, !dbg !40 + %3162 = insertelement <2 x bfloat> %3161, bfloat %2863, i64 1, !dbg !40 + %3163 = bitcast <2 x bfloat> %3162 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3154, i32 %3157, i32 %3160, i32 %3163, ptr addrspace(1) %2465, i1 %352) #6, !dbg !40 + %3164 = insertelement <2 x bfloat> poison, bfloat %2864, i64 0, !dbg !40 + %3165 = insertelement <2 x bfloat> %3164, bfloat %2865, i64 1, !dbg !40 + %3166 = bitcast <2 x bfloat> %3165 to i32, !dbg !40 + %3167 = insertelement <2 x bfloat> poison, bfloat %2866, i64 0, !dbg !40 + %3168 = insertelement <2 x bfloat> %3167, bfloat %2867, i64 1, !dbg !40 + %3169 = bitcast <2 x bfloat> %3168 to i32, !dbg !40 + %3170 = insertelement <2 x bfloat> poison, bfloat %2868, i64 0, !dbg !40 + %3171 = insertelement <2 x bfloat> %3170, bfloat %2869, i64 1, !dbg !40 + %3172 = bitcast <2 x bfloat> %3171 to i32, !dbg !40 + %3173 = insertelement <2 x bfloat> poison, bfloat %2870, i64 0, !dbg !40 + %3174 = insertelement <2 x bfloat> %3173, bfloat %2871, i64 1, !dbg !40 + %3175 = bitcast <2 x bfloat> %3174 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3166, i32 %3169, i32 %3172, i32 %3175, ptr addrspace(1) %2463, i1 %370) #6, !dbg !40 + %3176 = insertelement <2 x bfloat> poison, bfloat %2872, i64 0, !dbg !40 + %3177 = insertelement <2 x bfloat> %3176, bfloat %2873, i64 1, !dbg !40 + %3178 = bitcast <2 x bfloat> %3177 to i32, !dbg !40 + %3179 = insertelement <2 x bfloat> poison, bfloat %2874, i64 0, !dbg !40 + %3180 = insertelement <2 x bfloat> %3179, bfloat %2875, i64 1, !dbg !40 + %3181 = bitcast <2 x bfloat> %3180 to i32, !dbg !40 + %3182 = insertelement <2 x bfloat> poison, bfloat %2876, i64 0, !dbg !40 + %3183 = insertelement <2 x bfloat> %3182, bfloat %2877, i64 1, !dbg !40 + %3184 = bitcast <2 x bfloat> %3183 to i32, !dbg !40 + %3185 = insertelement <2 x bfloat> poison, bfloat %2878, i64 0, !dbg !40 + %3186 = insertelement <2 x bfloat> %3185, bfloat %2879, i64 1, !dbg !40 + %3187 = bitcast <2 x bfloat> %3186 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3178, i32 %3181, i32 %3184, i32 %3187, ptr addrspace(1) %2461, i1 %388) #6, !dbg !40 + %3188 = insertelement <2 x bfloat> poison, bfloat %2880, i64 0, !dbg !40 + %3189 = insertelement <2 x bfloat> %3188, bfloat %2881, i64 1, !dbg !40 + %3190 = bitcast <2 x bfloat> %3189 to i32, !dbg !40 + %3191 = insertelement <2 x bfloat> poison, bfloat %2882, i64 0, !dbg !40 + %3192 = insertelement <2 x bfloat> %3191, bfloat %2883, i64 1, !dbg !40 + %3193 = bitcast <2 x bfloat> %3192 to i32, !dbg !40 + %3194 = insertelement <2 x bfloat> poison, bfloat %2884, i64 0, !dbg !40 + %3195 = insertelement <2 x bfloat> %3194, bfloat %2885, i64 1, !dbg !40 + %3196 = bitcast <2 x bfloat> %3195 to i32, !dbg !40 + %3197 = insertelement <2 x bfloat> poison, bfloat %2886, i64 0, !dbg !40 + %3198 = insertelement <2 x bfloat> %3197, bfloat %2887, i64 1, !dbg !40 + %3199 = bitcast <2 x bfloat> %3198 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3190, i32 %3193, i32 %3196, i32 %3199, ptr addrspace(1) %2459, i1 %406) #6, !dbg !40 + %3200 = insertelement <2 x bfloat> poison, bfloat %2888, i64 0, !dbg !40 + %3201 = insertelement <2 x bfloat> %3200, bfloat %2889, i64 1, !dbg !40 + %3202 = bitcast <2 x bfloat> %3201 to i32, !dbg !40 + %3203 = insertelement <2 x bfloat> poison, bfloat %2890, i64 0, !dbg !40 + %3204 = insertelement <2 x bfloat> %3203, bfloat %2891, i64 1, !dbg !40 + %3205 = bitcast <2 x bfloat> %3204 to i32, !dbg !40 + %3206 = insertelement <2 x bfloat> poison, bfloat %2892, i64 0, !dbg !40 + %3207 = insertelement <2 x bfloat> %3206, bfloat %2893, i64 1, !dbg !40 + %3208 = bitcast <2 x bfloat> %3207 to i32, !dbg !40 + %3209 = insertelement <2 x bfloat> poison, bfloat %2894, i64 0, !dbg !40 + %3210 = insertelement <2 x bfloat> %3209, bfloat %2895, i64 1, !dbg !40 + %3211 = bitcast <2 x bfloat> %3210 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3202, i32 %3205, i32 %3208, i32 %3211, ptr addrspace(1) %2457, i1 %424) #6, !dbg !40 + %3212 = insertelement <2 x bfloat> poison, bfloat %2896, i64 0, !dbg !40 + %3213 = insertelement <2 x bfloat> %3212, bfloat %2897, i64 1, !dbg !40 + %3214 = bitcast <2 x bfloat> %3213 to i32, !dbg !40 + %3215 = insertelement <2 x bfloat> poison, bfloat %2898, i64 0, !dbg !40 + %3216 = insertelement <2 x bfloat> %3215, bfloat %2899, i64 1, !dbg !40 + %3217 = bitcast <2 x bfloat> %3216 to i32, !dbg !40 + %3218 = insertelement <2 x bfloat> poison, bfloat %2900, i64 0, !dbg !40 + %3219 = insertelement <2 x bfloat> %3218, bfloat %2901, i64 1, !dbg !40 + %3220 = bitcast <2 x bfloat> %3219 to i32, !dbg !40 + %3221 = insertelement <2 x bfloat> poison, bfloat %2902, i64 0, !dbg !40 + %3222 = insertelement <2 x bfloat> %3221, bfloat %2903, i64 1, !dbg !40 + %3223 = bitcast <2 x bfloat> %3222 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3214, i32 %3217, i32 %3220, i32 %3223, ptr addrspace(1) %2455, i1 %442) #6, !dbg !40 + %3224 = insertelement <2 x bfloat> poison, bfloat %2904, i64 0, !dbg !40 + %3225 = insertelement <2 x bfloat> %3224, bfloat %2905, i64 1, !dbg !40 + %3226 = bitcast <2 x bfloat> %3225 to i32, !dbg !40 + %3227 = insertelement <2 x bfloat> poison, bfloat %2906, i64 0, !dbg !40 + %3228 = insertelement <2 x bfloat> %3227, bfloat %2907, i64 1, !dbg !40 + %3229 = bitcast <2 x bfloat> %3228 to i32, !dbg !40 + %3230 = insertelement <2 x bfloat> poison, bfloat %2908, i64 0, !dbg !40 + %3231 = insertelement <2 x bfloat> %3230, bfloat %2909, i64 1, !dbg !40 + %3232 = bitcast <2 x bfloat> %3231 to i32, !dbg !40 + %3233 = insertelement <2 x bfloat> poison, bfloat %2910, i64 0, !dbg !40 + %3234 = insertelement <2 x bfloat> %3233, bfloat %2911, i64 1, !dbg !40 + %3235 = bitcast <2 x bfloat> %3234 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3226, i32 %3229, i32 %3232, i32 %3235, ptr addrspace(1) %2453, i1 %460) #6, !dbg !40 + %3236 = insertelement <2 x bfloat> poison, bfloat %2912, i64 0, !dbg !40 + %3237 = insertelement <2 x bfloat> %3236, bfloat %2913, i64 1, !dbg !40 + %3238 = bitcast <2 x bfloat> %3237 to i32, !dbg !40 + %3239 = insertelement <2 x bfloat> poison, bfloat %2914, i64 0, !dbg !40 + %3240 = insertelement <2 x bfloat> %3239, bfloat %2915, i64 1, !dbg !40 + %3241 = bitcast <2 x bfloat> %3240 to i32, !dbg !40 + %3242 = insertelement <2 x bfloat> poison, bfloat %2916, i64 0, !dbg !40 + %3243 = insertelement <2 x bfloat> %3242, bfloat %2917, i64 1, !dbg !40 + %3244 = bitcast <2 x bfloat> %3243 to i32, !dbg !40 + %3245 = insertelement <2 x bfloat> poison, bfloat %2918, i64 0, !dbg !40 + %3246 = insertelement <2 x bfloat> %3245, bfloat %2919, i64 1, !dbg !40 + %3247 = bitcast <2 x bfloat> %3246 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3238, i32 %3241, i32 %3244, i32 %3247, ptr addrspace(1) %2451, i1 %478) #6, !dbg !40 + %3248 = insertelement <2 x bfloat> poison, bfloat %2920, i64 0, !dbg !40 + %3249 = insertelement <2 x bfloat> %3248, bfloat %2921, i64 1, !dbg !40 + %3250 = bitcast <2 x bfloat> %3249 to i32, !dbg !40 + %3251 = insertelement <2 x bfloat> poison, bfloat %2922, i64 0, !dbg !40 + %3252 = insertelement <2 x bfloat> %3251, bfloat %2923, i64 1, !dbg !40 + %3253 = bitcast <2 x bfloat> %3252 to i32, !dbg !40 + %3254 = insertelement <2 x bfloat> poison, bfloat %2924, i64 0, !dbg !40 + %3255 = insertelement <2 x bfloat> %3254, bfloat %2925, i64 1, !dbg !40 + %3256 = bitcast <2 x bfloat> %3255 to i32, !dbg !40 + %3257 = insertelement <2 x bfloat> poison, bfloat %2926, i64 0, !dbg !40 + %3258 = insertelement <2 x bfloat> %3257, bfloat %2927, i64 1, !dbg !40 + %3259 = bitcast <2 x bfloat> %3258 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3250, i32 %3253, i32 %3256, i32 %3259, ptr addrspace(1) %2449, i1 %496) #6, !dbg !40 + %3260 = insertelement <2 x bfloat> poison, bfloat %2928, i64 0, !dbg !40 + %3261 = insertelement <2 x bfloat> %3260, bfloat %2929, i64 1, !dbg !40 + %3262 = bitcast <2 x bfloat> %3261 to i32, !dbg !40 + %3263 = insertelement <2 x bfloat> poison, bfloat %2930, i64 0, !dbg !40 + %3264 = insertelement <2 x bfloat> %3263, bfloat %2931, i64 1, !dbg !40 + %3265 = bitcast <2 x bfloat> %3264 to i32, !dbg !40 + %3266 = insertelement <2 x bfloat> poison, bfloat %2932, i64 0, !dbg !40 + %3267 = insertelement <2 x bfloat> %3266, bfloat %2933, i64 1, !dbg !40 + %3268 = bitcast <2 x bfloat> %3267 to i32, !dbg !40 + %3269 = insertelement <2 x bfloat> poison, bfloat %2934, i64 0, !dbg !40 + %3270 = insertelement <2 x bfloat> %3269, bfloat %2935, i64 1, !dbg !40 + %3271 = bitcast <2 x bfloat> %3270 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3262, i32 %3265, i32 %3268, i32 %3271, ptr addrspace(1) %2447, i1 %514) #6, !dbg !40 + %3272 = insertelement <2 x bfloat> poison, bfloat %2936, i64 0, !dbg !40 + %3273 = insertelement <2 x bfloat> %3272, bfloat %2937, i64 1, !dbg !40 + %3274 = bitcast <2 x bfloat> %3273 to i32, !dbg !40 + %3275 = insertelement <2 x bfloat> poison, bfloat %2938, i64 0, !dbg !40 + %3276 = insertelement <2 x bfloat> %3275, bfloat %2939, i64 1, !dbg !40 + %3277 = bitcast <2 x bfloat> %3276 to i32, !dbg !40 + %3278 = insertelement <2 x bfloat> poison, bfloat %2940, i64 0, !dbg !40 + %3279 = insertelement <2 x bfloat> %3278, bfloat %2941, i64 1, !dbg !40 + %3280 = bitcast <2 x bfloat> %3279 to i32, !dbg !40 + %3281 = insertelement <2 x bfloat> poison, bfloat %2942, i64 0, !dbg !40 + %3282 = insertelement <2 x bfloat> %3281, bfloat %2943, i64 1, !dbg !40 + %3283 = bitcast <2 x bfloat> %3282 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3274, i32 %3277, i32 %3280, i32 %3283, ptr addrspace(1) %2445, i1 %532) #6, !dbg !40 + %3284 = insertelement <2 x bfloat> poison, bfloat %2944, i64 0, !dbg !40 + %3285 = insertelement <2 x bfloat> %3284, bfloat %2945, i64 1, !dbg !40 + %3286 = bitcast <2 x bfloat> %3285 to i32, !dbg !40 + %3287 = insertelement <2 x bfloat> poison, bfloat %2946, i64 0, !dbg !40 + %3288 = insertelement <2 x bfloat> %3287, bfloat %2947, i64 1, !dbg !40 + %3289 = bitcast <2 x bfloat> %3288 to i32, !dbg !40 + %3290 = insertelement <2 x bfloat> poison, bfloat %2948, i64 0, !dbg !40 + %3291 = insertelement <2 x bfloat> %3290, bfloat %2949, i64 1, !dbg !40 + %3292 = bitcast <2 x bfloat> %3291 to i32, !dbg !40 + %3293 = insertelement <2 x bfloat> poison, bfloat %2950, i64 0, !dbg !40 + %3294 = insertelement <2 x bfloat> %3293, bfloat %2951, i64 1, !dbg !40 + %3295 = bitcast <2 x bfloat> %3294 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3286, i32 %3289, i32 %3292, i32 %3295, ptr addrspace(1) %2443, i1 %550) #6, !dbg !40 + %3296 = insertelement <2 x bfloat> poison, bfloat %2952, i64 0, !dbg !40 + %3297 = insertelement <2 x bfloat> %3296, bfloat %2953, i64 1, !dbg !40 + %3298 = bitcast <2 x bfloat> %3297 to i32, !dbg !40 + %3299 = insertelement <2 x bfloat> poison, bfloat %2954, i64 0, !dbg !40 + %3300 = insertelement <2 x bfloat> %3299, bfloat %2955, i64 1, !dbg !40 + %3301 = bitcast <2 x bfloat> %3300 to i32, !dbg !40 + %3302 = insertelement <2 x bfloat> poison, bfloat %2956, i64 0, !dbg !40 + %3303 = insertelement <2 x bfloat> %3302, bfloat %2957, i64 1, !dbg !40 + %3304 = bitcast <2 x bfloat> %3303 to i32, !dbg !40 + %3305 = insertelement <2 x bfloat> poison, bfloat %2958, i64 0, !dbg !40 + %3306 = insertelement <2 x bfloat> %3305, bfloat %2959, i64 1, !dbg !40 + %3307 = bitcast <2 x bfloat> %3306 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3298, i32 %3301, i32 %3304, i32 %3307, ptr addrspace(1) %2441, i1 %568) #6, !dbg !40 + %3308 = insertelement <2 x bfloat> poison, bfloat %2960, i64 0, !dbg !40 + %3309 = insertelement <2 x bfloat> %3308, bfloat %2961, i64 1, !dbg !40 + %3310 = bitcast <2 x bfloat> %3309 to i32, !dbg !40 + %3311 = insertelement <2 x bfloat> poison, bfloat %2962, i64 0, !dbg !40 + %3312 = insertelement <2 x bfloat> %3311, bfloat %2963, i64 1, !dbg !40 + %3313 = bitcast <2 x bfloat> %3312 to i32, !dbg !40 + %3314 = insertelement <2 x bfloat> poison, bfloat %2964, i64 0, !dbg !40 + %3315 = insertelement <2 x bfloat> %3314, bfloat %2965, i64 1, !dbg !40 + %3316 = bitcast <2 x bfloat> %3315 to i32, !dbg !40 + %3317 = insertelement <2 x bfloat> poison, bfloat %2966, i64 0, !dbg !40 + %3318 = insertelement <2 x bfloat> %3317, bfloat %2967, i64 1, !dbg !40 + %3319 = bitcast <2 x bfloat> %3318 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3310, i32 %3313, i32 %3316, i32 %3319, ptr addrspace(1) %2439, i1 %586) #6, !dbg !40 + %3320 = insertelement <2 x bfloat> poison, bfloat %2968, i64 0, !dbg !40 + %3321 = insertelement <2 x bfloat> %3320, bfloat %2969, i64 1, !dbg !40 + %3322 = bitcast <2 x bfloat> %3321 to i32, !dbg !40 + %3323 = insertelement <2 x bfloat> poison, bfloat %2970, i64 0, !dbg !40 + %3324 = insertelement <2 x bfloat> %3323, bfloat %2971, i64 1, !dbg !40 + %3325 = bitcast <2 x bfloat> %3324 to i32, !dbg !40 + %3326 = insertelement <2 x bfloat> poison, bfloat %2972, i64 0, !dbg !40 + %3327 = insertelement <2 x bfloat> %3326, bfloat %2973, i64 1, !dbg !40 + %3328 = bitcast <2 x bfloat> %3327 to i32, !dbg !40 + %3329 = insertelement <2 x bfloat> poison, bfloat %2974, i64 0, !dbg !40 + %3330 = insertelement <2 x bfloat> %3329, bfloat %2975, i64 1, !dbg !40 + %3331 = bitcast <2 x bfloat> %3330 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3322, i32 %3325, i32 %3328, i32 %3331, ptr addrspace(1) %2437, i1 %604) #6, !dbg !40 + %3332 = insertelement <2 x bfloat> poison, bfloat %2976, i64 0, !dbg !40 + %3333 = insertelement <2 x bfloat> %3332, bfloat %2977, i64 1, !dbg !40 + %3334 = bitcast <2 x bfloat> %3333 to i32, !dbg !40 + %3335 = insertelement <2 x bfloat> poison, bfloat %2978, i64 0, !dbg !40 + %3336 = insertelement <2 x bfloat> %3335, bfloat %2979, i64 1, !dbg !40 + %3337 = bitcast <2 x bfloat> %3336 to i32, !dbg !40 + %3338 = insertelement <2 x bfloat> poison, bfloat %2980, i64 0, !dbg !40 + %3339 = insertelement <2 x bfloat> %3338, bfloat %2981, i64 1, !dbg !40 + %3340 = bitcast <2 x bfloat> %3339 to i32, !dbg !40 + %3341 = insertelement <2 x bfloat> poison, bfloat %2982, i64 0, !dbg !40 + %3342 = insertelement <2 x bfloat> %3341, bfloat %2983, i64 1, !dbg !40 + %3343 = bitcast <2 x bfloat> %3342 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3334, i32 %3337, i32 %3340, i32 %3343, ptr addrspace(1) %2435, i1 %622) #6, !dbg !40 + %3344 = insertelement <2 x bfloat> poison, bfloat %2984, i64 0, !dbg !40 + %3345 = insertelement <2 x bfloat> %3344, bfloat %2985, i64 1, !dbg !40 + %3346 = bitcast <2 x bfloat> %3345 to i32, !dbg !40 + %3347 = insertelement <2 x bfloat> poison, bfloat %2986, i64 0, !dbg !40 + %3348 = insertelement <2 x bfloat> %3347, bfloat %2987, i64 1, !dbg !40 + %3349 = bitcast <2 x bfloat> %3348 to i32, !dbg !40 + %3350 = insertelement <2 x bfloat> poison, bfloat %2988, i64 0, !dbg !40 + %3351 = insertelement <2 x bfloat> %3350, bfloat %2989, i64 1, !dbg !40 + %3352 = bitcast <2 x bfloat> %3351 to i32, !dbg !40 + %3353 = insertelement <2 x bfloat> poison, bfloat %2990, i64 0, !dbg !40 + %3354 = insertelement <2 x bfloat> %3353, bfloat %2991, i64 1, !dbg !40 + %3355 = bitcast <2 x bfloat> %3354 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3346, i32 %3349, i32 %3352, i32 %3355, ptr addrspace(1) %2433, i1 %640) #6, !dbg !40 + %3356 = insertelement <2 x bfloat> poison, bfloat %2992, i64 0, !dbg !40 + %3357 = insertelement <2 x bfloat> %3356, bfloat %2993, i64 1, !dbg !40 + %3358 = bitcast <2 x bfloat> %3357 to i32, !dbg !40 + %3359 = insertelement <2 x bfloat> poison, bfloat %2994, i64 0, !dbg !40 + %3360 = insertelement <2 x bfloat> %3359, bfloat %2995, i64 1, !dbg !40 + %3361 = bitcast <2 x bfloat> %3360 to i32, !dbg !40 + %3362 = insertelement <2 x bfloat> poison, bfloat %2996, i64 0, !dbg !40 + %3363 = insertelement <2 x bfloat> %3362, bfloat %2997, i64 1, !dbg !40 + %3364 = bitcast <2 x bfloat> %3363 to i32, !dbg !40 + %3365 = insertelement <2 x bfloat> poison, bfloat %2998, i64 0, !dbg !40 + %3366 = insertelement <2 x bfloat> %3365, bfloat %2999, i64 1, !dbg !40 + %3367 = bitcast <2 x bfloat> %3366 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3358, i32 %3361, i32 %3364, i32 %3367, ptr addrspace(1) %2431, i1 %658) #6, !dbg !40 + %3368 = insertelement <2 x bfloat> poison, bfloat %3000, i64 0, !dbg !40 + %3369 = insertelement <2 x bfloat> %3368, bfloat %3001, i64 1, !dbg !40 + %3370 = bitcast <2 x bfloat> %3369 to i32, !dbg !40 + %3371 = insertelement <2 x bfloat> poison, bfloat %3002, i64 0, !dbg !40 + %3372 = insertelement <2 x bfloat> %3371, bfloat %3003, i64 1, !dbg !40 + %3373 = bitcast <2 x bfloat> %3372 to i32, !dbg !40 + %3374 = insertelement <2 x bfloat> poison, bfloat %3004, i64 0, !dbg !40 + %3375 = insertelement <2 x bfloat> %3374, bfloat %3005, i64 1, !dbg !40 + %3376 = bitcast <2 x bfloat> %3375 to i32, !dbg !40 + %3377 = insertelement <2 x bfloat> poison, bfloat %3006, i64 0, !dbg !40 + %3378 = insertelement <2 x bfloat> %3377, bfloat %3007, i64 1, !dbg !40 + %3379 = bitcast <2 x bfloat> %3378 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3370, i32 %3373, i32 %3376, i32 %3379, ptr addrspace(1) %2429, i1 %676) #6, !dbg !40 + %3380 = insertelement <2 x bfloat> poison, bfloat %3008, i64 0, !dbg !40 + %3381 = insertelement <2 x bfloat> %3380, bfloat %3009, i64 1, !dbg !40 + %3382 = bitcast <2 x bfloat> %3381 to i32, !dbg !40 + %3383 = insertelement <2 x bfloat> poison, bfloat %3010, i64 0, !dbg !40 + %3384 = insertelement <2 x bfloat> %3383, bfloat %3011, i64 1, !dbg !40 + %3385 = bitcast <2 x bfloat> %3384 to i32, !dbg !40 + %3386 = insertelement <2 x bfloat> poison, bfloat %3012, i64 0, !dbg !40 + %3387 = insertelement <2 x bfloat> %3386, bfloat %3013, i64 1, !dbg !40 + %3388 = bitcast <2 x bfloat> %3387 to i32, !dbg !40 + %3389 = insertelement <2 x bfloat> poison, bfloat %3014, i64 0, !dbg !40 + %3390 = insertelement <2 x bfloat> %3389, bfloat %3015, i64 1, !dbg !40 + %3391 = bitcast <2 x bfloat> %3390 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3382, i32 %3385, i32 %3388, i32 %3391, ptr addrspace(1) %2427, i1 %694) #6, !dbg !40 + %3392 = insertelement <2 x bfloat> poison, bfloat %3016, i64 0, !dbg !40 + %3393 = insertelement <2 x bfloat> %3392, bfloat %3017, i64 1, !dbg !40 + %3394 = bitcast <2 x bfloat> %3393 to i32, !dbg !40 + %3395 = insertelement <2 x bfloat> poison, bfloat %3018, i64 0, !dbg !40 + %3396 = insertelement <2 x bfloat> %3395, bfloat %3019, i64 1, !dbg !40 + %3397 = bitcast <2 x bfloat> %3396 to i32, !dbg !40 + %3398 = insertelement <2 x bfloat> poison, bfloat %3020, i64 0, !dbg !40 + %3399 = insertelement <2 x bfloat> %3398, bfloat %3021, i64 1, !dbg !40 + %3400 = bitcast <2 x bfloat> %3399 to i32, !dbg !40 + %3401 = insertelement <2 x bfloat> poison, bfloat %3022, i64 0, !dbg !40 + %3402 = insertelement <2 x bfloat> %3401, bfloat %3023, i64 1, !dbg !40 + %3403 = bitcast <2 x bfloat> %3402 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3394, i32 %3397, i32 %3400, i32 %3403, ptr addrspace(1) %2425, i1 %712) #6, !dbg !40 + %3404 = insertelement <2 x bfloat> poison, bfloat %3024, i64 0, !dbg !40 + %3405 = insertelement <2 x bfloat> %3404, bfloat %3025, i64 1, !dbg !40 + %3406 = bitcast <2 x bfloat> %3405 to i32, !dbg !40 + %3407 = insertelement <2 x bfloat> poison, bfloat %3026, i64 0, !dbg !40 + %3408 = insertelement <2 x bfloat> %3407, bfloat %3027, i64 1, !dbg !40 + %3409 = bitcast <2 x bfloat> %3408 to i32, !dbg !40 + %3410 = insertelement <2 x bfloat> poison, bfloat %3028, i64 0, !dbg !40 + %3411 = insertelement <2 x bfloat> %3410, bfloat %3029, i64 1, !dbg !40 + %3412 = bitcast <2 x bfloat> %3411 to i32, !dbg !40 + %3413 = insertelement <2 x bfloat> poison, bfloat %3030, i64 0, !dbg !40 + %3414 = insertelement <2 x bfloat> %3413, bfloat %3031, i64 1, !dbg !40 + %3415 = bitcast <2 x bfloat> %3414 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3406, i32 %3409, i32 %3412, i32 %3415, ptr addrspace(1) %2423, i1 %730) #6, !dbg !40 + %3416 = and i32 %11, 16, !dbg !41 + %3417 = icmp eq i32 %3416, 0, !dbg !41 + %3418 = select i1 %3417, i32 0, i32 320, !dbg !41 + %3419 = and i32 %11, 32, !dbg !41 + %3420 = shl nuw nsw i32 %3419, 1, !dbg !41 + %3421 = xor i32 %3418, %3420, !dbg !41 + %3422 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %3421, !dbg !41 + %3423 = bitcast float %2488 to i32, !dbg !41 + %3424 = bitcast float %2489 to i32, !dbg !41 + %3425 = bitcast float %2490 to i32, !dbg !41 + %3426 = bitcast float %2491 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %3422, i32 %3423, i32 %3424, i32 %3425, i32 %3426) #6, !dbg !41 + %3427 = getelementptr inbounds nuw i8, ptr addrspace(3) %3422, i32 16, !dbg !41 + %3428 = bitcast float %2492 to i32, !dbg !41 + %3429 = bitcast float %2493 to i32, !dbg !41 + %3430 = bitcast float %2494 to i32, !dbg !41 + %3431 = bitcast float %2495 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %3427, i32 %3428, i32 %3429, i32 %3430, i32 %3431) #6, !dbg !41 + %3432 = getelementptr inbounds nuw i8, ptr addrspace(3) %3422, i32 32, !dbg !41 + %3433 = bitcast float %2496 to i32, !dbg !41 + %3434 = bitcast float %2497 to i32, !dbg !41 + %3435 = bitcast float %2498 to i32, !dbg !41 + %3436 = bitcast float %2499 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %3432, i32 %3433, i32 %3434, i32 %3435, i32 %3436) #6, !dbg !41 + %3437 = getelementptr inbounds nuw i8, ptr addrspace(3) %3422, i32 48, !dbg !41 + %3438 = bitcast float %2500 to i32, !dbg !41 + %3439 = bitcast float %2501 to i32, !dbg !41 + %3440 = bitcast float %2502 to i32, !dbg !41 + %3441 = bitcast float %2503 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %3437, i32 %3438, i32 %3439, i32 %3440, i32 %3441) #6, !dbg !41 + %3442 = getelementptr inbounds nuw i8, ptr addrspace(3) %3422, i32 128, !dbg !41 + %3443 = bitcast float %2504 to i32, !dbg !41 + %3444 = bitcast float %2505 to i32, !dbg !41 + %3445 = bitcast float %2506 to i32, !dbg !41 + %3446 = bitcast float %2507 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %3442, i32 %3443, i32 %3444, i32 %3445, i32 %3446) #6, !dbg !41 + %3447 = getelementptr inbounds nuw i8, ptr addrspace(3) %3422, i32 144, !dbg !41 + %3448 = bitcast float %2508 to i32, !dbg !41 + %3449 = bitcast float %2509 to i32, !dbg !41 + %3450 = bitcast float %2510 to i32, !dbg !41 + %3451 = bitcast float %2511 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %3447, i32 %3448, i32 %3449, i32 %3450, i32 %3451) #6, !dbg !41 + %3452 = getelementptr inbounds nuw i8, ptr addrspace(3) %3422, i32 160, !dbg !41 + %3453 = bitcast float %2512 to i32, !dbg !41 + %3454 = bitcast float %2513 to i32, !dbg !41 + %3455 = bitcast float %2514 to i32, !dbg !41 + %3456 = bitcast float %2515 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %3452, i32 %3453, i32 %3454, i32 %3455, i32 %3456) #6, !dbg !41 + %3457 = getelementptr inbounds nuw i8, ptr addrspace(3) %3422, i32 176, !dbg !41 + %3458 = bitcast float %2516 to i32, !dbg !41 + %3459 = bitcast float %2517 to i32, !dbg !41 + %3460 = bitcast float %2518 to i32, !dbg !41 + %3461 = bitcast float %2519 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %3457, i32 %3458, i32 %3459, i32 %3460, i32 %3461) #6, !dbg !41 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !41 + %3462 = and i32 %2413, 60, !dbg !41 + %3463 = shl nuw nsw i32 %11, 6, !dbg !41 + %3464 = and i32 %3463, 64, !dbg !41 + %3465 = shl nuw nsw i32 %3419, 2, !dbg !41 + %3466 = or disjoint i32 %3464, %3465, !dbg !41 + %3467 = or disjoint i32 %3466, %3462, !dbg !41 + %3468 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %3467, !dbg !41 + %3469 = load i32, ptr addrspace(3) %3468, align 4, !dbg !41 + %3470 = xor i32 %3467, 320, !dbg !41 + %3471 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %3470, !dbg !41 + %3472 = load i32, ptr addrspace(3) %3471, align 4, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %3469, ptr addrspace(1) %2421, i1 %2487) #6, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %3472, ptr addrspace(1) %2418, i1 %2486) #6, !dbg !41 + ret void, !dbg !42 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.div.full(float, float) #3 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #4 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #5 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.f(float) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="64" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) } +attributes #4 = { convergent nocallback nounwind } +attributes #5 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #6 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "l2norm_fwd_kernel", linkageName: "l2norm_fwd_kernel", scope: !1, file: !1, line: 69, type: !6, scopeLine: 69, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !10, !11, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 70, column: 24, scope: !5) +!14 = !DILocation(line: 70, column: 30, scope: !5) +!15 = !DILocation(line: 71, column: 16, scope: !5) +!16 = !DILocation(line: 71, column: 34, scope: !5) +!17 = !DILocation(line: 71, column: 21, scope: !5) +!18 = !DILocation(line: 73, column: 16, scope: !5) +!19 = !DILocation(line: 74, column: 30, scope: !5) +!20 = !DILocation(line: 75, column: 29, scope: !5) +!21 = !DILocation(line: 75, column: 14, scope: !5) +!22 = !DILocation(line: 75, column: 33, scope: !5) +!23 = !DILocation(line: 79, column: 18, scope: !5) +!24 = !DILocation(line: 79, column: 47, scope: !5) +!25 = !DILocation(line: 80, column: 38, scope: !5) +!26 = !DILocation(line: 263, column: 15, scope: !27, inlinedAt: !29) +!27 = distinct !DILexicalBlockFile(scope: !5, file: !28, discriminator: 0) +!28 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!29 = !DILocation(line: 293, column: 36, scope: !27, inlinedAt: !30) +!30 = !DILocation(line: 80, column: 43, scope: !31) +!31 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!32 = !DILocation(line: 80, column: 48, scope: !5) +!33 = !DILocation(line: 80, column: 25, scope: !5) +!34 = !DILocation(line: 77, column: 20, scope: !5) +!35 = !DILocation(line: 76, column: 14, scope: !5) +!36 = !DILocation(line: 76, column: 33, scope: !5) +!37 = !DILocation(line: 80, column: 17, scope: !5) +!38 = !DILocation(line: 81, column: 16, scope: !5) +!39 = !DILocation(line: 83, column: 25, scope: !5) +!40 = !DILocation(line: 83, column: 18, scope: !5) +!41 = !DILocation(line: 84, column: 21, scope: !5) +!42 = !DILocation(line: 84, column: 4, scope: !5) diff --git a/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.ptx b/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..390dc6b6b24319ae356015ccef9e075b97a6f91d --- /dev/null +++ b/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.ptx @@ -0,0 +1,3024 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_fwd_kernel // -- Begin function l2norm_fwd_kernel +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; +.global .align 1 .b8 _$_str_$_2[17] = {95, 95, 67, 85, 68, 65, 95, 80, 82, 69, 67, 95, 83, 81, 82, 84}; + // @l2norm_fwd_kernel +.visible .entry l2norm_fwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_2, + .param .f32 l2norm_fwd_kernel_param_3, + .param .u32 l2norm_fwd_kernel_param_4, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_6 +) +.reqntid 64 +{ + .reg .pred %p<50>; + .reg .b16 %rs<310>; + .reg .b32 %r<1476>; + .reg .b64 %rd<238>; + .loc 1 69 0 // l2norm.py:69:0 +$L__func_begin0: + .loc 1 69 0 // l2norm.py:69:0 + +// %bb.0: // %__nv_sqrtf.exit + ld.param.b64 %rd67, [l2norm_fwd_kernel_param_0]; + ld.param.b64 %rd68, [l2norm_fwd_kernel_param_1]; +$L__tmp0: + .loc 1 70 24 // l2norm.py:70:24 + mov.u32 %r300, %ctaid.x; + ld.param.b64 %rd69, [l2norm_fwd_kernel_param_2]; + .loc 1 71 16 // l2norm.py:71:16 + mul.wide.u32 %rd70, %r300, 128; + ld.param.b32 %r301, [l2norm_fwd_kernel_param_3]; + ld.param.s32 %rd71, [l2norm_fwd_kernel_param_4]; + .loc 1 71 34 // l2norm.py:71:34 + mov.u32 %r302, %tid.x; + shr.u32 %r303, %r302, 4; + bfe.u32 %r304, %r302, 4, 2; + or.b32 %r305, %r304, 4; + .loc 1 71 21 // l2norm.py:71:21 + or.b32 %r306, %r304, 120; + or.b32 %r307, %r303, 124; + or.b32 %r308, %r304, 112; + or.b32 %r309, %r304, 116; + or.b32 %r310, %r304, 104; + or.b32 %r311, %r304, 108; + or.b32 %r312, %r304, 96; + or.b32 %r313, %r304, 100; + or.b32 %r314, %r304, 88; + or.b32 %r315, %r304, 92; + or.b32 %r316, %r304, 80; + or.b32 %r317, %r304, 84; + or.b32 %r318, %r304, 72; + or.b32 %r319, %r304, 76; + or.b32 %r320, %r304, 64; + or.b32 %r321, %r304, 68; + or.b32 %r322, %r304, 32; + or.b32 %r323, %r304, 36; + or.b32 %r324, %r304, 40; + or.b32 %r325, %r304, 44; + or.b32 %r326, %r304, 48; + or.b32 %r327, %r304, 52; + or.b32 %r328, %r304, 56; + or.b32 %r329, %r303, 60; + or.b32 %r330, %r304, 16; + or.b32 %r331, %r304, 20; + or.b32 %r332, %r304, 24; + or.b32 %r333, %r304, 28; + or.b32 %r334, %r304, 8; + or.b32 %r335, %r304, 12; + cvt.u64.u32 %rd72, %r321; + cvt.u64.u32 %rd73, %r320; + cvt.u64.u32 %rd74, %r319; + cvt.u64.u32 %rd75, %r318; + cvt.u64.u32 %rd76, %r317; + cvt.u64.u32 %rd77, %r316; + cvt.u64.u32 %rd78, %r315; + cvt.u64.u32 %rd79, %r314; + cvt.u64.u32 %rd80, %r313; + cvt.u64.u32 %rd81, %r312; + cvt.u64.u32 %rd82, %r311; + cvt.u64.u32 %rd83, %r310; + cvt.u64.u32 %rd84, %r309; + cvt.u64.u32 %rd85, %r308; + cvt.u64.u32 %rd86, %r307; + cvt.u64.u32 %rd87, %r306; + cvt.u64.u32 %rd88, %r329; + cvt.u64.u32 %rd89, %r328; + cvt.u64.u32 %rd90, %r327; + cvt.u64.u32 %rd91, %r326; + cvt.u64.u32 %rd92, %r325; + cvt.u64.u32 %rd93, %r324; + cvt.u64.u32 %rd94, %r323; + cvt.u64.u32 %rd95, %r322; + cvt.u64.u32 %rd96, %r333; + cvt.u64.u32 %rd97, %r332; + cvt.u64.u32 %rd98, %r331; + cvt.u64.u32 %rd99, %r330; + cvt.u64.u32 %rd100, %r335; + cvt.u64.u32 %rd101, %r334; + cvt.u64.u32 %rd102, %r305; + cvt.u64.u32 %rd103, %r304; + or.b64 %rd104, %rd70, %rd103; + or.b64 %rd105, %rd70, %rd102; + or.b64 %rd106, %rd70, %rd101; + or.b64 %rd107, %rd70, %rd100; + or.b64 %rd108, %rd70, %rd99; + or.b64 %rd109, %rd70, %rd98; + or.b64 %rd110, %rd70, %rd97; + or.b64 %rd111, %rd70, %rd96; + or.b64 %rd112, %rd70, %rd95; + or.b64 %rd113, %rd70, %rd94; + or.b64 %rd114, %rd70, %rd93; + or.b64 %rd115, %rd70, %rd92; + or.b64 %rd116, %rd70, %rd91; + or.b64 %rd117, %rd70, %rd90; + or.b64 %rd118, %rd70, %rd89; + or.b64 %rd119, %rd70, %rd88; + or.b64 %rd120, %rd70, %rd87; + or.b64 %rd121, %rd70, %rd86; + or.b64 %rd122, %rd70, %rd85; + or.b64 %rd123, %rd70, %rd84; + or.b64 %rd124, %rd70, %rd83; + or.b64 %rd125, %rd70, %rd82; + or.b64 %rd126, %rd70, %rd81; + or.b64 %rd127, %rd70, %rd80; + or.b64 %rd128, %rd70, %rd79; + or.b64 %rd129, %rd70, %rd78; + or.b64 %rd130, %rd70, %rd77; + or.b64 %rd131, %rd70, %rd76; + or.b64 %rd132, %rd70, %rd75; + or.b64 %rd133, %rd70, %rd74; + or.b64 %rd134, %rd70, %rd73; + or.b64 %rd135, %rd70, %rd72; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p35, %rd135, %rd71; + selp.b16 %rs1, 1, 0, %p35; + shl.b16 %rs2, %rs1, 2; + setp.lt.s64 %p36, %rd134, %rd71; + selp.b16 %rs3, -1, 0, %p36; + shl.b16 %rs4, %rs3, 3; + or.b16 %rs5, %rs4, %rs2; + setp.lt.s64 %p37, %rd133, %rd71; + selp.b16 %rs6, 1, 0, %p37; + setp.lt.s64 %p38, %rd132, %rd71; + selp.b16 %rs7, -1, 0, %p38; + shl.b16 %rs8, %rs7, 1; + or.b16 %rs9, %rs6, %rs8; + and.b16 %rs10, %rs9, 3; + or.b16 %rs11, %rs10, %rs5; + shl.b16 %rs12, %rs11, 12; + setp.lt.s64 %p39, %rd131, %rd71; + selp.b16 %rs13, 1, 0, %p39; + shl.b16 %rs14, %rs13, 2; + setp.lt.s64 %p40, %rd130, %rd71; + selp.b16 %rs15, -1, 0, %p40; + shl.b16 %rs16, %rs15, 3; + or.b16 %rs17, %rs16, %rs14; + setp.lt.s64 %p41, %rd129, %rd71; + selp.b16 %rs18, 1, 0, %p41; + setp.lt.s64 %p42, %rd128, %rd71; + selp.b16 %rs19, -1, 0, %p42; + shl.b16 %rs20, %rs19, 1; + or.b16 %rs21, %rs18, %rs20; + and.b16 %rs22, %rs21, 3; + or.b16 %rs23, %rs22, %rs17; + and.b16 %rs24, %rs23, 15; + shl.b16 %rs25, %rs24, 8; + or.b16 %rs26, %rs12, %rs25; + setp.lt.s64 %p43, %rd127, %rd71; + selp.b16 %rs27, 1, 0, %p43; + shl.b16 %rs28, %rs27, 2; + setp.lt.s64 %p44, %rd126, %rd71; + selp.b16 %rs29, -1, 0, %p44; + shl.b16 %rs30, %rs29, 3; + or.b16 %rs31, %rs30, %rs28; + setp.lt.s64 %p45, %rd125, %rd71; + selp.b16 %rs32, 1, 0, %p45; + setp.lt.s64 %p46, %rd124, %rd71; + selp.b16 %rs33, -1, 0, %p46; + shl.b16 %rs34, %rs33, 1; + or.b16 %rs35, %rs32, %rs34; + and.b16 %rs36, %rs35, 3; + or.b16 %rs37, %rs36, %rs31; + shl.b16 %rs38, %rs37, 4; + setp.lt.s64 %p47, %rd123, %rd71; + selp.b16 %rs39, 1, 0, %p47; + shl.b16 %rs40, %rs39, 2; + setp.lt.s64 %p48, %rd122, %rd71; + selp.b16 %rs41, -1, 0, %p48; + shl.b16 %rs42, %rs41, 3; + or.b16 %rs43, %rs42, %rs40; + setp.lt.s64 %p32, %rd121, %rd71; + selp.b16 %rs44, 1, 0, %p32; + setp.lt.s64 %p49, %rd120, %rd71; + selp.b16 %rs45, -1, 0, %p49; + shl.b16 %rs46, %rs45, 1; + or.b16 %rs47, %rs44, %rs46; + and.b16 %rs48, %rs47, 3; + or.b16 %rs49, %rs48, %rs43; + and.b16 %rs50, %rs49, 15; + or.b16 %rs51, %rs50, %rs38; + and.b16 %rs52, %rs51, 255; + or.b16 %rs53, %rs52, %rs26; + cvt.u32.u16 %r336, %rs53; + setp.lt.s64 %p16, %rd119, %rd71; + setp.lt.s64 %p15, %rd118, %rd71; + setp.lt.s64 %p14, %rd117, %rd71; + setp.lt.s64 %p13, %rd116, %rd71; + setp.lt.s64 %p12, %rd115, %rd71; + setp.lt.s64 %p11, %rd114, %rd71; + setp.lt.s64 %p10, %rd113, %rd71; + setp.lt.s64 %p9, %rd112, %rd71; + setp.lt.s64 %p1, %rd111, %rd71; + setp.lt.s64 %p8, %rd110, %rd71; + setp.lt.s64 %p7, %rd109, %rd71; + setp.lt.s64 %p6, %rd108, %rd71; + setp.lt.s64 %p5, %rd107, %rd71; + setp.lt.s64 %p4, %rd106, %rd71; + setp.lt.s64 %p3, %rd105, %rd71; + setp.lt.s64 %p2, %rd104, %rd71; + .loc 1 74 30 // l2norm.py:74:30 + shl.b32 %r337, %r302, 3; + and.b32 %r338, %r337, 120; + .loc 1 75 14 // l2norm.py:75:14 + shl.b64 %rd136, %rd104, 8; + add.s64 %rd137, %rd67, %rd136; + shl.b64 %rd138, %rd105, 8; + add.s64 %rd139, %rd67, %rd138; + shl.b64 %rd140, %rd106, 8; + add.s64 %rd141, %rd67, %rd140; + shl.b64 %rd142, %rd107, 8; + add.s64 %rd143, %rd67, %rd142; + shl.b64 %rd144, %rd108, 8; + add.s64 %rd145, %rd67, %rd144; + shl.b64 %rd146, %rd109, 8; + add.s64 %rd147, %rd67, %rd146; + shl.b64 %rd148, %rd110, 8; + add.s64 %rd149, %rd67, %rd148; + shl.b64 %rd150, %rd111, 8; + add.s64 %rd151, %rd67, %rd150; + shl.b64 %rd152, %rd112, 8; + add.s64 %rd153, %rd67, %rd152; + shl.b64 %rd154, %rd113, 8; + add.s64 %rd155, %rd67, %rd154; + shl.b64 %rd156, %rd114, 8; + add.s64 %rd157, %rd67, %rd156; + shl.b64 %rd158, %rd115, 8; + add.s64 %rd159, %rd67, %rd158; + shl.b64 %rd160, %rd116, 8; + add.s64 %rd161, %rd67, %rd160; + shl.b64 %rd162, %rd117, 8; + add.s64 %rd163, %rd67, %rd162; + shl.b64 %rd164, %rd118, 8; + add.s64 %rd165, %rd67, %rd164; + shl.b64 %rd166, %rd119, 8; + add.s64 %rd167, %rd67, %rd166; + shl.b64 %rd168, %rd134, 8; + add.s64 %rd169, %rd67, %rd168; + shl.b64 %rd170, %rd135, 8; + add.s64 %rd171, %rd67, %rd170; + shl.b64 %rd172, %rd132, 8; + add.s64 %rd173, %rd67, %rd172; + shl.b64 %rd174, %rd133, 8; + add.s64 %rd175, %rd67, %rd174; + shl.b64 %rd176, %rd130, 8; + add.s64 %rd177, %rd67, %rd176; + shl.b64 %rd178, %rd131, 8; + add.s64 %rd179, %rd67, %rd178; + shl.b64 %rd180, %rd128, 8; + add.s64 %rd181, %rd67, %rd180; + shl.b64 %rd182, %rd129, 8; + add.s64 %rd183, %rd67, %rd182; + shl.b64 %rd184, %rd126, 8; + add.s64 %rd185, %rd67, %rd184; + shl.b64 %rd186, %rd127, 8; + add.s64 %rd187, %rd67, %rd186; + shl.b64 %rd188, %rd124, 8; + add.s64 %rd189, %rd67, %rd188; + shl.b64 %rd190, %rd125, 8; + add.s64 %rd191, %rd67, %rd190; + shl.b64 %rd192, %rd122, 8; + add.s64 %rd193, %rd67, %rd192; + shl.b64 %rd194, %rd123, 8; + add.s64 %rd195, %rd67, %rd194; + shl.b64 %rd196, %rd120, 8; + add.s64 %rd197, %rd67, %rd196; + shl.b64 %rd198, %rd121, 8; + add.s64 %rd199, %rd67, %rd198; + .loc 1 75 33 // l2norm.py:75:33 + mul.wide.u32 %rd200, %r338, 2; + add.s64 %rd1, %rd137, %rd200; + add.s64 %rd2, %rd139, %rd200; + add.s64 %rd3, %rd141, %rd200; + add.s64 %rd4, %rd143, %rd200; + add.s64 %rd5, %rd145, %rd200; + add.s64 %rd6, %rd147, %rd200; + add.s64 %rd7, %rd149, %rd200; + add.s64 %rd8, %rd151, %rd200; + add.s64 %rd9, %rd153, %rd200; + add.s64 %rd10, %rd155, %rd200; + add.s64 %rd11, %rd157, %rd200; + add.s64 %rd12, %rd159, %rd200; + add.s64 %rd13, %rd161, %rd200; + add.s64 %rd14, %rd163, %rd200; + add.s64 %rd15, %rd165, %rd200; + add.s64 %rd16, %rd167, %rd200; + add.s64 %rd17, %rd169, %rd200; + add.s64 %rd18, %rd171, %rd200; + add.s64 %rd19, %rd173, %rd200; + add.s64 %rd20, %rd175, %rd200; + add.s64 %rd21, %rd177, %rd200; + add.s64 %rd22, %rd179, %rd200; + add.s64 %rd23, %rd181, %rd200; + add.s64 %rd24, %rd183, %rd200; + add.s64 %rd25, %rd185, %rd200; + add.s64 %rd26, %rd187, %rd200; + add.s64 %rd27, %rd189, %rd200; + add.s64 %rd28, %rd191, %rd200; + add.s64 %rd29, %rd193, %rd200; + add.s64 %rd30, %rd195, %rd200; + add.s64 %rd31, %rd197, %rd200; + add.s64 %rd32, %rd199, %rd200; + mov.b32 %r5, 0; + .loc 1 79 18 // l2norm.py:79:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p2 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs54, %rs55}, %r1; + mov.b32 {%rs56, %rs57}, %r2; + mov.b32 {%rs58, %rs59}, %r3; + mov.b32 {%rs60, %rs61}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p3 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs62, %rs63}, %r6; + mov.b32 {%rs64, %rs65}, %r7; + mov.b32 {%rs66, %rs67}, %r8; + mov.b32 {%rs68, %rs69}, %r9; + // begin inline asm + mov.u32 %r10, %r5; + mov.u32 %r11, %r5; + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + @%p4 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs70, %rs71}, %r10; + mov.b32 {%rs72, %rs73}, %r11; + mov.b32 {%rs74, %rs75}, %r12; + mov.b32 {%rs76, %rs77}, %r13; + // begin inline asm + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + @%p5 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ]; + // end inline asm + mov.b32 {%rs78, %rs79}, %r14; + mov.b32 {%rs80, %rs81}, %r15; + mov.b32 {%rs82, %rs83}, %r16; + mov.b32 {%rs84, %rs85}, %r17; + // begin inline asm + mov.u32 %r18, %r5; + mov.u32 %r19, %r5; + mov.u32 %r20, %r5; + mov.u32 %r21, %r5; + @%p6 ld.global.v4.b32 { %r18, %r19, %r20, %r21 }, [ %rd5 + 0 ]; + // end inline asm + mov.b32 {%rs86, %rs87}, %r18; + mov.b32 {%rs88, %rs89}, %r19; + mov.b32 {%rs90, %rs91}, %r20; + mov.b32 {%rs92, %rs93}, %r21; + // begin inline asm + mov.u32 %r22, %r5; + mov.u32 %r23, %r5; + mov.u32 %r24, %r5; + mov.u32 %r25, %r5; + @%p7 ld.global.v4.b32 { %r22, %r23, %r24, %r25 }, [ %rd6 + 0 ]; + // end inline asm + mov.b32 {%rs94, %rs95}, %r22; + mov.b32 {%rs96, %rs97}, %r23; + mov.b32 {%rs98, %rs99}, %r24; + mov.b32 {%rs100, %rs101}, %r25; + // begin inline asm + mov.u32 %r26, %r5; + mov.u32 %r27, %r5; + mov.u32 %r28, %r5; + mov.u32 %r29, %r5; + @%p8 ld.global.v4.b32 { %r26, %r27, %r28, %r29 }, [ %rd7 + 0 ]; + // end inline asm + mov.b32 {%rs102, %rs103}, %r26; + mov.b32 {%rs104, %rs105}, %r27; + mov.b32 {%rs106, %rs107}, %r28; + mov.b32 {%rs108, %rs109}, %r29; + // begin inline asm + mov.u32 %r30, %r5; + mov.u32 %r31, %r5; + mov.u32 %r32, %r5; + mov.u32 %r33, %r5; + @%p1 ld.global.v4.b32 { %r30, %r31, %r32, %r33 }, [ %rd8 + 0 ]; + // end inline asm + mov.b32 {%rs110, %rs111}, %r30; + mov.b32 {%rs112, %rs113}, %r31; + mov.b32 {%rs114, %rs115}, %r32; + mov.b32 {%rs116, %rs117}, %r33; + // begin inline asm + mov.u32 %r34, %r5; + mov.u32 %r35, %r5; + mov.u32 %r36, %r5; + mov.u32 %r37, %r5; + @%p9 ld.global.v4.b32 { %r34, %r35, %r36, %r37 }, [ %rd9 + 0 ]; + // end inline asm + mov.b32 {%rs118, %rs119}, %r34; + mov.b32 {%rs120, %rs121}, %r35; + mov.b32 {%rs122, %rs123}, %r36; + mov.b32 {%rs124, %rs125}, %r37; + // begin inline asm + mov.u32 %r38, %r5; + mov.u32 %r39, %r5; + mov.u32 %r40, %r5; + mov.u32 %r41, %r5; + @%p10 ld.global.v4.b32 { %r38, %r39, %r40, %r41 }, [ %rd10 + 0 ]; + // end inline asm + mov.b32 {%rs126, %rs127}, %r38; + mov.b32 {%rs128, %rs129}, %r39; + mov.b32 {%rs130, %rs131}, %r40; + mov.b32 {%rs132, %rs133}, %r41; + // begin inline asm + mov.u32 %r42, %r5; + mov.u32 %r43, %r5; + mov.u32 %r44, %r5; + mov.u32 %r45, %r5; + @%p11 ld.global.v4.b32 { %r42, %r43, %r44, %r45 }, [ %rd11 + 0 ]; + // end inline asm + mov.b32 {%rs134, %rs135}, %r42; + mov.b32 {%rs136, %rs137}, %r43; + mov.b32 {%rs138, %rs139}, %r44; + mov.b32 {%rs140, %rs141}, %r45; + // begin inline asm + mov.u32 %r46, %r5; + mov.u32 %r47, %r5; + mov.u32 %r48, %r5; + mov.u32 %r49, %r5; + @%p12 ld.global.v4.b32 { %r46, %r47, %r48, %r49 }, [ %rd12 + 0 ]; + // end inline asm + mov.b32 {%rs142, %rs143}, %r46; + mov.b32 {%rs144, %rs145}, %r47; + mov.b32 {%rs146, %rs147}, %r48; + mov.b32 {%rs148, %rs149}, %r49; + // begin inline asm + mov.u32 %r50, %r5; + mov.u32 %r51, %r5; + mov.u32 %r52, %r5; + mov.u32 %r53, %r5; + @%p13 ld.global.v4.b32 { %r50, %r51, %r52, %r53 }, [ %rd13 + 0 ]; + // end inline asm + mov.b32 {%rs150, %rs151}, %r50; + mov.b32 {%rs152, %rs153}, %r51; + mov.b32 {%rs154, %rs155}, %r52; + mov.b32 {%rs156, %rs157}, %r53; + // begin inline asm + mov.u32 %r54, %r5; + mov.u32 %r55, %r5; + mov.u32 %r56, %r5; + mov.u32 %r57, %r5; + @%p14 ld.global.v4.b32 { %r54, %r55, %r56, %r57 }, [ %rd14 + 0 ]; + // end inline asm + mov.b32 {%rs158, %rs159}, %r54; + mov.b32 {%rs160, %rs161}, %r55; + mov.b32 {%rs162, %rs163}, %r56; + mov.b32 {%rs164, %rs165}, %r57; + // begin inline asm + mov.u32 %r58, %r5; + mov.u32 %r59, %r5; + mov.u32 %r60, %r5; + mov.u32 %r61, %r5; + @%p15 ld.global.v4.b32 { %r58, %r59, %r60, %r61 }, [ %rd15 + 0 ]; + // end inline asm + mov.b32 {%rs166, %rs167}, %r58; + mov.b32 {%rs168, %rs169}, %r59; + mov.b32 {%rs170, %rs171}, %r60; + mov.b32 {%rs172, %rs173}, %r61; + // begin inline asm + mov.u32 %r62, %r5; + mov.u32 %r63, %r5; + mov.u32 %r64, %r5; + mov.u32 %r65, %r5; + @%p16 ld.global.v4.b32 { %r62, %r63, %r64, %r65 }, [ %rd16 + 0 ]; + // end inline asm + mov.b32 {%rs174, %rs175}, %r62; + mov.b32 {%rs176, %rs177}, %r63; + mov.b32 {%rs178, %rs179}, %r64; + mov.b32 {%rs180, %rs181}, %r65; + shr.u32 %r339, %r336, 15; + and.b32 %r340, %r339, 1; + setp.ne.b32 %p17, %r340, 0; + // begin inline asm + mov.u32 %r66, %r5; + mov.u32 %r67, %r5; + mov.u32 %r68, %r5; + mov.u32 %r69, %r5; + @%p17 ld.global.v4.b32 { %r66, %r67, %r68, %r69 }, [ %rd17 + 0 ]; + // end inline asm + mov.b32 {%rs182, %rs183}, %r66; + mov.b32 {%rs184, %rs185}, %r67; + mov.b32 {%rs186, %rs187}, %r68; + mov.b32 {%rs188, %rs189}, %r69; + shr.u32 %r341, %r336, 14; + and.b32 %r342, %r341, 1; + setp.ne.b32 %p18, %r342, 0; + // begin inline asm + mov.u32 %r70, %r5; + mov.u32 %r71, %r5; + mov.u32 %r72, %r5; + mov.u32 %r73, %r5; + @%p18 ld.global.v4.b32 { %r70, %r71, %r72, %r73 }, [ %rd18 + 0 ]; + // end inline asm + mov.b32 {%rs190, %rs191}, %r70; + mov.b32 {%rs192, %rs193}, %r71; + mov.b32 {%rs194, %rs195}, %r72; + mov.b32 {%rs196, %rs197}, %r73; + shr.u32 %r343, %r336, 13; + and.b32 %r344, %r343, 1; + setp.ne.b32 %p19, %r344, 0; + // begin inline asm + mov.u32 %r74, %r5; + mov.u32 %r75, %r5; + mov.u32 %r76, %r5; + mov.u32 %r77, %r5; + @%p19 ld.global.v4.b32 { %r74, %r75, %r76, %r77 }, [ %rd19 + 0 ]; + // end inline asm + mov.b32 {%rs198, %rs199}, %r74; + mov.b32 {%rs200, %rs201}, %r75; + mov.b32 {%rs202, %rs203}, %r76; + mov.b32 {%rs204, %rs205}, %r77; + shr.u32 %r345, %r336, 12; + and.b32 %r346, %r345, 1; + setp.ne.b32 %p20, %r346, 0; + // begin inline asm + mov.u32 %r78, %r5; + mov.u32 %r79, %r5; + mov.u32 %r80, %r5; + mov.u32 %r81, %r5; + @%p20 ld.global.v4.b32 { %r78, %r79, %r80, %r81 }, [ %rd20 + 0 ]; + // end inline asm + mov.b32 {%rs206, %rs207}, %r78; + mov.b32 {%rs208, %rs209}, %r79; + mov.b32 {%rs210, %rs211}, %r80; + mov.b32 {%rs212, %rs213}, %r81; + shr.u32 %r347, %r336, 11; + and.b32 %r348, %r347, 1; + setp.ne.b32 %p21, %r348, 0; + // begin inline asm + mov.u32 %r82, %r5; + mov.u32 %r83, %r5; + mov.u32 %r84, %r5; + mov.u32 %r85, %r5; + @%p21 ld.global.v4.b32 { %r82, %r83, %r84, %r85 }, [ %rd21 + 0 ]; + // end inline asm + mov.b32 {%rs214, %rs215}, %r82; + mov.b32 {%rs216, %rs217}, %r83; + mov.b32 {%rs218, %rs219}, %r84; + mov.b32 {%rs220, %rs221}, %r85; + shr.u32 %r349, %r336, 10; + and.b32 %r350, %r349, 1; + setp.ne.b32 %p22, %r350, 0; + // begin inline asm + mov.u32 %r86, %r5; + mov.u32 %r87, %r5; + mov.u32 %r88, %r5; + mov.u32 %r89, %r5; + @%p22 ld.global.v4.b32 { %r86, %r87, %r88, %r89 }, [ %rd22 + 0 ]; + // end inline asm + mov.b32 {%rs222, %rs223}, %r86; + mov.b32 {%rs224, %rs225}, %r87; + mov.b32 {%rs226, %rs227}, %r88; + mov.b32 {%rs228, %rs229}, %r89; + shr.u32 %r351, %r336, 9; + and.b32 %r352, %r351, 1; + setp.ne.b32 %p23, %r352, 0; + // begin inline asm + mov.u32 %r90, %r5; + mov.u32 %r91, %r5; + mov.u32 %r92, %r5; + mov.u32 %r93, %r5; + @%p23 ld.global.v4.b32 { %r90, %r91, %r92, %r93 }, [ %rd23 + 0 ]; + // end inline asm + mov.b32 {%rs230, %rs231}, %r90; + mov.b32 {%rs232, %rs233}, %r91; + mov.b32 {%rs234, %rs235}, %r92; + mov.b32 {%rs236, %rs237}, %r93; + shr.u32 %r353, %r336, 8; + and.b32 %r354, %r353, 1; + setp.ne.b32 %p24, %r354, 0; + // begin inline asm + mov.u32 %r94, %r5; + mov.u32 %r95, %r5; + mov.u32 %r96, %r5; + mov.u32 %r97, %r5; + @%p24 ld.global.v4.b32 { %r94, %r95, %r96, %r97 }, [ %rd24 + 0 ]; + // end inline asm + mov.b32 {%rs238, %rs239}, %r94; + mov.b32 {%rs240, %rs241}, %r95; + mov.b32 {%rs242, %rs243}, %r96; + mov.b32 {%rs244, %rs245}, %r97; + shr.u32 %r355, %r336, 7; + and.b32 %r356, %r355, 1; + setp.ne.b32 %p25, %r356, 0; + // begin inline asm + mov.u32 %r98, %r5; + mov.u32 %r99, %r5; + mov.u32 %r100, %r5; + mov.u32 %r101, %r5; + @%p25 ld.global.v4.b32 { %r98, %r99, %r100, %r101 }, [ %rd25 + 0 ]; + // end inline asm + mov.b32 {%rs246, %rs247}, %r98; + mov.b32 {%rs248, %rs249}, %r99; + mov.b32 {%rs250, %rs251}, %r100; + mov.b32 {%rs252, %rs253}, %r101; + shr.u32 %r357, %r336, 6; + and.b32 %r358, %r357, 1; + setp.ne.b32 %p26, %r358, 0; + // begin inline asm + mov.u32 %r102, %r5; + mov.u32 %r103, %r5; + mov.u32 %r104, %r5; + mov.u32 %r105, %r5; + @%p26 ld.global.v4.b32 { %r102, %r103, %r104, %r105 }, [ %rd26 + 0 ]; + // end inline asm + mov.b32 {%rs254, %rs255}, %r102; + mov.b32 {%rs256, %rs257}, %r103; + mov.b32 {%rs258, %rs259}, %r104; + mov.b32 {%rs260, %rs261}, %r105; + shr.u32 %r359, %r336, 5; + and.b32 %r360, %r359, 1; + setp.ne.b32 %p27, %r360, 0; + // begin inline asm + mov.u32 %r106, %r5; + mov.u32 %r107, %r5; + mov.u32 %r108, %r5; + mov.u32 %r109, %r5; + @%p27 ld.global.v4.b32 { %r106, %r107, %r108, %r109 }, [ %rd27 + 0 ]; + // end inline asm + mov.b32 {%rs262, %rs263}, %r106; + mov.b32 {%rs264, %rs265}, %r107; + mov.b32 {%rs266, %rs267}, %r108; + mov.b32 {%rs268, %rs269}, %r109; + shr.u32 %r361, %r336, 4; + and.b32 %r362, %r361, 1; + setp.ne.b32 %p28, %r362, 0; + // begin inline asm + mov.u32 %r110, %r5; + mov.u32 %r111, %r5; + mov.u32 %r112, %r5; + mov.u32 %r113, %r5; + @%p28 ld.global.v4.b32 { %r110, %r111, %r112, %r113 }, [ %rd28 + 0 ]; + // end inline asm + mov.b32 {%rs270, %rs271}, %r110; + mov.b32 {%rs272, %rs273}, %r111; + mov.b32 {%rs274, %rs275}, %r112; + mov.b32 {%rs276, %rs277}, %r113; + shr.u32 %r363, %r336, 3; + and.b32 %r364, %r363, 1; + setp.ne.b32 %p29, %r364, 0; + // begin inline asm + mov.u32 %r114, %r5; + mov.u32 %r115, %r5; + mov.u32 %r116, %r5; + mov.u32 %r117, %r5; + @%p29 ld.global.v4.b32 { %r114, %r115, %r116, %r117 }, [ %rd29 + 0 ]; + // end inline asm + mov.b32 {%rs278, %rs279}, %r114; + mov.b32 {%rs280, %rs281}, %r115; + mov.b32 {%rs282, %rs283}, %r116; + mov.b32 {%rs284, %rs285}, %r117; + shr.u32 %r365, %r336, 2; + and.b32 %r366, %r365, 1; + setp.ne.b32 %p30, %r366, 0; + // begin inline asm + mov.u32 %r118, %r5; + mov.u32 %r119, %r5; + mov.u32 %r120, %r5; + mov.u32 %r121, %r5; + @%p30 ld.global.v4.b32 { %r118, %r119, %r120, %r121 }, [ %rd30 + 0 ]; + // end inline asm + mov.b32 {%rs286, %rs287}, %r118; + mov.b32 {%rs288, %rs289}, %r119; + mov.b32 {%rs290, %rs291}, %r120; + mov.b32 {%rs292, %rs293}, %r121; + shr.u32 %r367, %r336, 1; + and.b32 %r368, %r367, 1; + setp.ne.b32 %p31, %r368, 0; + // begin inline asm + mov.u32 %r122, %r5; + mov.u32 %r123, %r5; + mov.u32 %r124, %r5; + mov.u32 %r125, %r5; + @%p31 ld.global.v4.b32 { %r122, %r123, %r124, %r125 }, [ %rd31 + 0 ]; + // end inline asm + mov.b32 {%rs294, %rs295}, %r122; + mov.b32 {%rs296, %rs297}, %r123; + mov.b32 {%rs298, %rs299}, %r124; + mov.b32 {%rs300, %rs301}, %r125; + // begin inline asm + mov.u32 %r126, %r5; + mov.u32 %r127, %r5; + mov.u32 %r128, %r5; + mov.u32 %r129, %r5; + @%p32 ld.global.v4.b32 { %r126, %r127, %r128, %r129 }, [ %rd32 + 0 ]; + // end inline asm + mov.b32 {%rs302, %rs303}, %r126; + mov.b32 {%rs304, %rs305}, %r127; + mov.b32 {%rs306, %rs307}, %r128; + mov.b32 {%rs308, %rs309}, %r129; + .loc 1 79 47 // l2norm.py:79:47 + cvt.f32.bf16 %r369, %rs54; + cvt.f32.bf16 %r370, %rs55; + cvt.f32.bf16 %r371, %rs56; + cvt.f32.bf16 %r372, %rs57; + cvt.f32.bf16 %r373, %rs58; + cvt.f32.bf16 %r374, %rs59; + cvt.f32.bf16 %r375, %rs60; + cvt.f32.bf16 %r376, %rs61; + cvt.f32.bf16 %r377, %rs62; + cvt.f32.bf16 %r378, %rs63; + cvt.f32.bf16 %r379, %rs64; + cvt.f32.bf16 %r380, %rs65; + cvt.f32.bf16 %r381, %rs66; + cvt.f32.bf16 %r382, %rs67; + cvt.f32.bf16 %r383, %rs68; + cvt.f32.bf16 %r384, %rs69; + cvt.f32.bf16 %r385, %rs70; + cvt.f32.bf16 %r386, %rs71; + cvt.f32.bf16 %r387, %rs72; + cvt.f32.bf16 %r388, %rs73; + cvt.f32.bf16 %r389, %rs74; + cvt.f32.bf16 %r390, %rs75; + cvt.f32.bf16 %r391, %rs76; + cvt.f32.bf16 %r392, %rs77; + cvt.f32.bf16 %r393, %rs78; + cvt.f32.bf16 %r394, %rs79; + cvt.f32.bf16 %r395, %rs80; + cvt.f32.bf16 %r396, %rs81; + cvt.f32.bf16 %r397, %rs82; + cvt.f32.bf16 %r398, %rs83; + cvt.f32.bf16 %r399, %rs84; + cvt.f32.bf16 %r400, %rs85; + cvt.f32.bf16 %r401, %rs86; + cvt.f32.bf16 %r402, %rs87; + cvt.f32.bf16 %r403, %rs88; + cvt.f32.bf16 %r404, %rs89; + cvt.f32.bf16 %r405, %rs90; + cvt.f32.bf16 %r406, %rs91; + cvt.f32.bf16 %r407, %rs92; + cvt.f32.bf16 %r408, %rs93; + cvt.f32.bf16 %r409, %rs94; + cvt.f32.bf16 %r410, %rs95; + cvt.f32.bf16 %r411, %rs96; + cvt.f32.bf16 %r412, %rs97; + cvt.f32.bf16 %r413, %rs98; + cvt.f32.bf16 %r414, %rs99; + cvt.f32.bf16 %r415, %rs100; + cvt.f32.bf16 %r416, %rs101; + cvt.f32.bf16 %r417, %rs102; + cvt.f32.bf16 %r418, %rs103; + cvt.f32.bf16 %r419, %rs104; + cvt.f32.bf16 %r420, %rs105; + cvt.f32.bf16 %r421, %rs106; + cvt.f32.bf16 %r422, %rs107; + cvt.f32.bf16 %r423, %rs108; + cvt.f32.bf16 %r424, %rs109; + cvt.f32.bf16 %r425, %rs110; + cvt.f32.bf16 %r426, %rs111; + cvt.f32.bf16 %r427, %rs112; + cvt.f32.bf16 %r428, %rs113; + cvt.f32.bf16 %r429, %rs114; + cvt.f32.bf16 %r430, %rs115; + cvt.f32.bf16 %r431, %rs116; + cvt.f32.bf16 %r432, %rs117; + cvt.f32.bf16 %r433, %rs118; + cvt.f32.bf16 %r434, %rs119; + cvt.f32.bf16 %r435, %rs120; + cvt.f32.bf16 %r436, %rs121; + cvt.f32.bf16 %r437, %rs122; + cvt.f32.bf16 %r438, %rs123; + cvt.f32.bf16 %r439, %rs124; + cvt.f32.bf16 %r440, %rs125; + cvt.f32.bf16 %r441, %rs126; + cvt.f32.bf16 %r442, %rs127; + cvt.f32.bf16 %r443, %rs128; + cvt.f32.bf16 %r444, %rs129; + cvt.f32.bf16 %r445, %rs130; + cvt.f32.bf16 %r446, %rs131; + cvt.f32.bf16 %r447, %rs132; + cvt.f32.bf16 %r448, %rs133; + cvt.f32.bf16 %r449, %rs134; + cvt.f32.bf16 %r450, %rs135; + cvt.f32.bf16 %r451, %rs136; + cvt.f32.bf16 %r452, %rs137; + cvt.f32.bf16 %r453, %rs138; + cvt.f32.bf16 %r454, %rs139; + cvt.f32.bf16 %r455, %rs140; + cvt.f32.bf16 %r456, %rs141; + cvt.f32.bf16 %r457, %rs142; + cvt.f32.bf16 %r458, %rs143; + cvt.f32.bf16 %r459, %rs144; + cvt.f32.bf16 %r460, %rs145; + cvt.f32.bf16 %r461, %rs146; + cvt.f32.bf16 %r462, %rs147; + cvt.f32.bf16 %r463, %rs148; + cvt.f32.bf16 %r464, %rs149; + cvt.f32.bf16 %r465, %rs150; + cvt.f32.bf16 %r466, %rs151; + cvt.f32.bf16 %r467, %rs152; + cvt.f32.bf16 %r468, %rs153; + cvt.f32.bf16 %r469, %rs154; + cvt.f32.bf16 %r470, %rs155; + cvt.f32.bf16 %r471, %rs156; + cvt.f32.bf16 %r472, %rs157; + cvt.f32.bf16 %r473, %rs158; + cvt.f32.bf16 %r474, %rs159; + cvt.f32.bf16 %r475, %rs160; + cvt.f32.bf16 %r476, %rs161; + cvt.f32.bf16 %r477, %rs162; + cvt.f32.bf16 %r478, %rs163; + cvt.f32.bf16 %r479, %rs164; + cvt.f32.bf16 %r480, %rs165; + cvt.f32.bf16 %r481, %rs166; + cvt.f32.bf16 %r482, %rs167; + cvt.f32.bf16 %r483, %rs168; + cvt.f32.bf16 %r484, %rs169; + cvt.f32.bf16 %r485, %rs170; + cvt.f32.bf16 %r486, %rs171; + cvt.f32.bf16 %r487, %rs172; + cvt.f32.bf16 %r488, %rs173; + cvt.f32.bf16 %r489, %rs174; + cvt.f32.bf16 %r490, %rs175; + cvt.f32.bf16 %r491, %rs176; + cvt.f32.bf16 %r492, %rs177; + cvt.f32.bf16 %r493, %rs178; + cvt.f32.bf16 %r494, %rs179; + cvt.f32.bf16 %r495, %rs180; + cvt.f32.bf16 %r496, %rs181; + cvt.f32.bf16 %r497, %rs182; + cvt.f32.bf16 %r498, %rs183; + cvt.f32.bf16 %r499, %rs184; + cvt.f32.bf16 %r500, %rs185; + cvt.f32.bf16 %r501, %rs186; + cvt.f32.bf16 %r502, %rs187; + cvt.f32.bf16 %r503, %rs188; + cvt.f32.bf16 %r504, %rs189; + cvt.f32.bf16 %r505, %rs190; + cvt.f32.bf16 %r506, %rs191; + cvt.f32.bf16 %r507, %rs192; + cvt.f32.bf16 %r508, %rs193; + cvt.f32.bf16 %r509, %rs194; + cvt.f32.bf16 %r510, %rs195; + cvt.f32.bf16 %r511, %rs196; + cvt.f32.bf16 %r512, %rs197; + cvt.f32.bf16 %r513, %rs198; + cvt.f32.bf16 %r514, %rs199; + cvt.f32.bf16 %r515, %rs200; + cvt.f32.bf16 %r516, %rs201; + cvt.f32.bf16 %r517, %rs202; + cvt.f32.bf16 %r518, %rs203; + cvt.f32.bf16 %r519, %rs204; + cvt.f32.bf16 %r520, %rs205; + cvt.f32.bf16 %r521, %rs206; + cvt.f32.bf16 %r522, %rs207; + cvt.f32.bf16 %r523, %rs208; + cvt.f32.bf16 %r524, %rs209; + cvt.f32.bf16 %r525, %rs210; + cvt.f32.bf16 %r526, %rs211; + cvt.f32.bf16 %r527, %rs212; + cvt.f32.bf16 %r528, %rs213; + cvt.f32.bf16 %r529, %rs214; + cvt.f32.bf16 %r530, %rs215; + cvt.f32.bf16 %r531, %rs216; + cvt.f32.bf16 %r532, %rs217; + cvt.f32.bf16 %r533, %rs218; + cvt.f32.bf16 %r534, %rs219; + cvt.f32.bf16 %r535, %rs220; + cvt.f32.bf16 %r536, %rs221; + cvt.f32.bf16 %r537, %rs222; + cvt.f32.bf16 %r538, %rs223; + cvt.f32.bf16 %r539, %rs224; + cvt.f32.bf16 %r540, %rs225; + cvt.f32.bf16 %r541, %rs226; + cvt.f32.bf16 %r542, %rs227; + cvt.f32.bf16 %r543, %rs228; + cvt.f32.bf16 %r544, %rs229; + cvt.f32.bf16 %r545, %rs230; + cvt.f32.bf16 %r546, %rs231; + cvt.f32.bf16 %r547, %rs232; + cvt.f32.bf16 %r548, %rs233; + cvt.f32.bf16 %r549, %rs234; + cvt.f32.bf16 %r550, %rs235; + cvt.f32.bf16 %r551, %rs236; + cvt.f32.bf16 %r552, %rs237; + cvt.f32.bf16 %r553, %rs238; + cvt.f32.bf16 %r554, %rs239; + cvt.f32.bf16 %r555, %rs240; + cvt.f32.bf16 %r556, %rs241; + cvt.f32.bf16 %r557, %rs242; + cvt.f32.bf16 %r558, %rs243; + cvt.f32.bf16 %r559, %rs244; + cvt.f32.bf16 %r560, %rs245; + cvt.f32.bf16 %r561, %rs246; + cvt.f32.bf16 %r562, %rs247; + cvt.f32.bf16 %r563, %rs248; + cvt.f32.bf16 %r564, %rs249; + cvt.f32.bf16 %r565, %rs250; + cvt.f32.bf16 %r566, %rs251; + cvt.f32.bf16 %r567, %rs252; + cvt.f32.bf16 %r568, %rs253; + cvt.f32.bf16 %r569, %rs254; + cvt.f32.bf16 %r570, %rs255; + cvt.f32.bf16 %r571, %rs256; + cvt.f32.bf16 %r572, %rs257; + cvt.f32.bf16 %r573, %rs258; + cvt.f32.bf16 %r574, %rs259; + cvt.f32.bf16 %r575, %rs260; + cvt.f32.bf16 %r576, %rs261; + cvt.f32.bf16 %r577, %rs262; + cvt.f32.bf16 %r578, %rs263; + cvt.f32.bf16 %r579, %rs264; + cvt.f32.bf16 %r580, %rs265; + cvt.f32.bf16 %r581, %rs266; + cvt.f32.bf16 %r582, %rs267; + cvt.f32.bf16 %r583, %rs268; + cvt.f32.bf16 %r584, %rs269; + cvt.f32.bf16 %r585, %rs270; + cvt.f32.bf16 %r586, %rs271; + cvt.f32.bf16 %r587, %rs272; + cvt.f32.bf16 %r588, %rs273; + cvt.f32.bf16 %r589, %rs274; + cvt.f32.bf16 %r590, %rs275; + cvt.f32.bf16 %r591, %rs276; + cvt.f32.bf16 %r592, %rs277; + cvt.f32.bf16 %r593, %rs278; + cvt.f32.bf16 %r594, %rs279; + cvt.f32.bf16 %r595, %rs280; + cvt.f32.bf16 %r596, %rs281; + cvt.f32.bf16 %r597, %rs282; + cvt.f32.bf16 %r598, %rs283; + cvt.f32.bf16 %r599, %rs284; + cvt.f32.bf16 %r600, %rs285; + cvt.f32.bf16 %r601, %rs286; + cvt.f32.bf16 %r602, %rs287; + cvt.f32.bf16 %r603, %rs288; + cvt.f32.bf16 %r604, %rs289; + cvt.f32.bf16 %r605, %rs290; + cvt.f32.bf16 %r606, %rs291; + cvt.f32.bf16 %r607, %rs292; + cvt.f32.bf16 %r608, %rs293; + cvt.f32.bf16 %r609, %rs294; + cvt.f32.bf16 %r610, %rs295; + cvt.f32.bf16 %r611, %rs296; + cvt.f32.bf16 %r612, %rs297; + cvt.f32.bf16 %r613, %rs298; + cvt.f32.bf16 %r614, %rs299; + cvt.f32.bf16 %r615, %rs300; + cvt.f32.bf16 %r616, %rs301; + cvt.f32.bf16 %r617, %rs302; + cvt.f32.bf16 %r618, %rs303; + cvt.f32.bf16 %r619, %rs304; + cvt.f32.bf16 %r620, %rs305; + cvt.f32.bf16 %r621, %rs306; + cvt.f32.bf16 %r622, %rs307; + cvt.f32.bf16 %r623, %rs308; + cvt.f32.bf16 %r624, %rs309; + .loc 1 80 38 // l2norm.py:80:38 + mul.f32 %r625, %r370, %r370; + mul.f32 %r626, %r378, %r378; + mul.f32 %r627, %r386, %r386; + mul.f32 %r628, %r394, %r394; + mul.f32 %r629, %r402, %r402; + mul.f32 %r630, %r410, %r410; + mul.f32 %r631, %r418, %r418; + mul.f32 %r632, %r426, %r426; + mul.f32 %r633, %r434, %r434; + mul.f32 %r634, %r442, %r442; + mul.f32 %r635, %r450, %r450; + mul.f32 %r636, %r458, %r458; + mul.f32 %r637, %r466, %r466; + mul.f32 %r638, %r474, %r474; + mul.f32 %r639, %r482, %r482; + mul.f32 %r640, %r490, %r490; + mul.f32 %r641, %r498, %r498; + mul.f32 %r642, %r506, %r506; + mul.f32 %r643, %r514, %r514; + mul.f32 %r644, %r522, %r522; + mul.f32 %r645, %r530, %r530; + mul.f32 %r646, %r538, %r538; + mul.f32 %r647, %r546, %r546; + mul.f32 %r648, %r554, %r554; + mul.f32 %r649, %r562, %r562; + mul.f32 %r650, %r570, %r570; + mul.f32 %r651, %r578, %r578; + mul.f32 %r652, %r586, %r586; + mul.f32 %r653, %r594, %r594; + mul.f32 %r654, %r602, %r602; + mul.f32 %r655, %r610, %r610; + mul.f32 %r656, %r618, %r618; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + fma.rn.f32.bf16 %r657, %rs54, %rs54, %r625; + fma.rn.f32.bf16 %r658, %rs56, %rs56, %r657; + fma.rn.f32.bf16 %r659, %rs57, %rs57, %r658; + fma.rn.f32.bf16 %r660, %rs58, %rs58, %r659; + fma.rn.f32.bf16 %r661, %rs59, %rs59, %r660; + fma.rn.f32.bf16 %r662, %rs60, %rs60, %r661; + fma.rn.f32.bf16 %r663, %rs61, %rs61, %r662; + fma.rn.f32.bf16 %r664, %rs62, %rs62, %r626; + fma.rn.f32.bf16 %r665, %rs64, %rs64, %r664; + fma.rn.f32.bf16 %r666, %rs65, %rs65, %r665; + fma.rn.f32.bf16 %r667, %rs66, %rs66, %r666; + fma.rn.f32.bf16 %r668, %rs67, %rs67, %r667; + fma.rn.f32.bf16 %r669, %rs68, %rs68, %r668; + fma.rn.f32.bf16 %r670, %rs69, %rs69, %r669; + fma.rn.f32.bf16 %r671, %rs70, %rs70, %r627; + fma.rn.f32.bf16 %r672, %rs72, %rs72, %r671; + fma.rn.f32.bf16 %r673, %rs73, %rs73, %r672; + fma.rn.f32.bf16 %r674, %rs74, %rs74, %r673; + fma.rn.f32.bf16 %r675, %rs75, %rs75, %r674; + fma.rn.f32.bf16 %r676, %rs76, %rs76, %r675; + fma.rn.f32.bf16 %r677, %rs77, %rs77, %r676; + fma.rn.f32.bf16 %r678, %rs78, %rs78, %r628; + fma.rn.f32.bf16 %r679, %rs80, %rs80, %r678; + fma.rn.f32.bf16 %r680, %rs81, %rs81, %r679; + fma.rn.f32.bf16 %r681, %rs82, %rs82, %r680; + fma.rn.f32.bf16 %r682, %rs83, %rs83, %r681; + fma.rn.f32.bf16 %r683, %rs84, %rs84, %r682; + fma.rn.f32.bf16 %r684, %rs85, %rs85, %r683; + fma.rn.f32.bf16 %r685, %rs86, %rs86, %r629; + fma.rn.f32.bf16 %r686, %rs88, %rs88, %r685; + fma.rn.f32.bf16 %r687, %rs89, %rs89, %r686; + fma.rn.f32.bf16 %r688, %rs90, %rs90, %r687; + fma.rn.f32.bf16 %r689, %rs91, %rs91, %r688; + fma.rn.f32.bf16 %r690, %rs92, %rs92, %r689; + fma.rn.f32.bf16 %r691, %rs93, %rs93, %r690; + fma.rn.f32.bf16 %r692, %rs94, %rs94, %r630; + fma.rn.f32.bf16 %r693, %rs96, %rs96, %r692; + fma.rn.f32.bf16 %r694, %rs97, %rs97, %r693; + fma.rn.f32.bf16 %r695, %rs98, %rs98, %r694; + fma.rn.f32.bf16 %r696, %rs99, %rs99, %r695; + fma.rn.f32.bf16 %r697, %rs100, %rs100, %r696; + fma.rn.f32.bf16 %r698, %rs101, %rs101, %r697; + fma.rn.f32.bf16 %r699, %rs102, %rs102, %r631; + fma.rn.f32.bf16 %r700, %rs104, %rs104, %r699; + fma.rn.f32.bf16 %r701, %rs105, %rs105, %r700; + fma.rn.f32.bf16 %r702, %rs106, %rs106, %r701; + fma.rn.f32.bf16 %r703, %rs107, %rs107, %r702; + fma.rn.f32.bf16 %r704, %rs108, %rs108, %r703; + fma.rn.f32.bf16 %r705, %rs109, %rs109, %r704; + fma.rn.f32.bf16 %r706, %rs110, %rs110, %r632; + fma.rn.f32.bf16 %r707, %rs112, %rs112, %r706; + fma.rn.f32.bf16 %r708, %rs113, %rs113, %r707; + fma.rn.f32.bf16 %r709, %rs114, %rs114, %r708; + fma.rn.f32.bf16 %r710, %rs115, %rs115, %r709; + fma.rn.f32.bf16 %r711, %rs116, %rs116, %r710; + fma.rn.f32.bf16 %r712, %rs117, %rs117, %r711; + fma.rn.f32.bf16 %r713, %rs118, %rs118, %r633; + fma.rn.f32.bf16 %r714, %rs120, %rs120, %r713; + fma.rn.f32.bf16 %r715, %rs121, %rs121, %r714; + fma.rn.f32.bf16 %r716, %rs122, %rs122, %r715; + fma.rn.f32.bf16 %r717, %rs123, %rs123, %r716; + fma.rn.f32.bf16 %r718, %rs124, %rs124, %r717; + fma.rn.f32.bf16 %r719, %rs125, %rs125, %r718; + fma.rn.f32.bf16 %r720, %rs126, %rs126, %r634; + fma.rn.f32.bf16 %r721, %rs128, %rs128, %r720; + fma.rn.f32.bf16 %r722, %rs129, %rs129, %r721; + fma.rn.f32.bf16 %r723, %rs130, %rs130, %r722; + fma.rn.f32.bf16 %r724, %rs131, %rs131, %r723; + fma.rn.f32.bf16 %r725, %rs132, %rs132, %r724; + fma.rn.f32.bf16 %r726, %rs133, %rs133, %r725; + fma.rn.f32.bf16 %r727, %rs134, %rs134, %r635; + fma.rn.f32.bf16 %r728, %rs136, %rs136, %r727; + fma.rn.f32.bf16 %r729, %rs137, %rs137, %r728; + fma.rn.f32.bf16 %r730, %rs138, %rs138, %r729; + fma.rn.f32.bf16 %r731, %rs139, %rs139, %r730; + fma.rn.f32.bf16 %r732, %rs140, %rs140, %r731; + fma.rn.f32.bf16 %r733, %rs141, %rs141, %r732; + fma.rn.f32.bf16 %r734, %rs142, %rs142, %r636; + fma.rn.f32.bf16 %r735, %rs144, %rs144, %r734; + fma.rn.f32.bf16 %r736, %rs145, %rs145, %r735; + fma.rn.f32.bf16 %r737, %rs146, %rs146, %r736; + fma.rn.f32.bf16 %r738, %rs147, %rs147, %r737; + fma.rn.f32.bf16 %r739, %rs148, %rs148, %r738; + fma.rn.f32.bf16 %r740, %rs149, %rs149, %r739; + fma.rn.f32.bf16 %r741, %rs150, %rs150, %r637; + fma.rn.f32.bf16 %r742, %rs152, %rs152, %r741; + fma.rn.f32.bf16 %r743, %rs153, %rs153, %r742; + fma.rn.f32.bf16 %r744, %rs154, %rs154, %r743; + fma.rn.f32.bf16 %r745, %rs155, %rs155, %r744; + fma.rn.f32.bf16 %r746, %rs156, %rs156, %r745; + fma.rn.f32.bf16 %r747, %rs157, %rs157, %r746; + fma.rn.f32.bf16 %r748, %rs158, %rs158, %r638; + fma.rn.f32.bf16 %r749, %rs160, %rs160, %r748; + fma.rn.f32.bf16 %r750, %rs161, %rs161, %r749; + fma.rn.f32.bf16 %r751, %rs162, %rs162, %r750; + fma.rn.f32.bf16 %r752, %rs163, %rs163, %r751; + fma.rn.f32.bf16 %r753, %rs164, %rs164, %r752; + fma.rn.f32.bf16 %r754, %rs165, %rs165, %r753; + fma.rn.f32.bf16 %r755, %rs166, %rs166, %r639; + fma.rn.f32.bf16 %r756, %rs168, %rs168, %r755; + fma.rn.f32.bf16 %r757, %rs169, %rs169, %r756; + fma.rn.f32.bf16 %r758, %rs170, %rs170, %r757; + fma.rn.f32.bf16 %r759, %rs171, %rs171, %r758; + fma.rn.f32.bf16 %r760, %rs172, %rs172, %r759; + fma.rn.f32.bf16 %r761, %rs173, %rs173, %r760; + fma.rn.f32.bf16 %r762, %rs174, %rs174, %r640; + fma.rn.f32.bf16 %r763, %rs176, %rs176, %r762; + fma.rn.f32.bf16 %r764, %rs177, %rs177, %r763; + fma.rn.f32.bf16 %r765, %rs178, %rs178, %r764; + fma.rn.f32.bf16 %r766, %rs179, %rs179, %r765; + fma.rn.f32.bf16 %r767, %rs180, %rs180, %r766; + fma.rn.f32.bf16 %r768, %rs181, %rs181, %r767; + fma.rn.f32.bf16 %r769, %rs182, %rs182, %r641; + fma.rn.f32.bf16 %r770, %rs184, %rs184, %r769; + fma.rn.f32.bf16 %r771, %rs185, %rs185, %r770; + fma.rn.f32.bf16 %r772, %rs186, %rs186, %r771; + fma.rn.f32.bf16 %r773, %rs187, %rs187, %r772; + fma.rn.f32.bf16 %r774, %rs188, %rs188, %r773; + fma.rn.f32.bf16 %r775, %rs189, %rs189, %r774; + fma.rn.f32.bf16 %r776, %rs190, %rs190, %r642; + fma.rn.f32.bf16 %r777, %rs192, %rs192, %r776; + fma.rn.f32.bf16 %r778, %rs193, %rs193, %r777; + fma.rn.f32.bf16 %r779, %rs194, %rs194, %r778; + fma.rn.f32.bf16 %r780, %rs195, %rs195, %r779; + fma.rn.f32.bf16 %r781, %rs196, %rs196, %r780; + fma.rn.f32.bf16 %r782, %rs197, %rs197, %r781; + fma.rn.f32.bf16 %r783, %rs198, %rs198, %r643; + fma.rn.f32.bf16 %r784, %rs200, %rs200, %r783; + fma.rn.f32.bf16 %r785, %rs201, %rs201, %r784; + fma.rn.f32.bf16 %r786, %rs202, %rs202, %r785; + fma.rn.f32.bf16 %r787, %rs203, %rs203, %r786; + fma.rn.f32.bf16 %r788, %rs204, %rs204, %r787; + fma.rn.f32.bf16 %r789, %rs205, %rs205, %r788; + fma.rn.f32.bf16 %r790, %rs206, %rs206, %r644; + fma.rn.f32.bf16 %r791, %rs208, %rs208, %r790; + fma.rn.f32.bf16 %r792, %rs209, %rs209, %r791; + fma.rn.f32.bf16 %r793, %rs210, %rs210, %r792; + fma.rn.f32.bf16 %r794, %rs211, %rs211, %r793; + fma.rn.f32.bf16 %r795, %rs212, %rs212, %r794; + fma.rn.f32.bf16 %r796, %rs213, %rs213, %r795; + fma.rn.f32.bf16 %r797, %rs214, %rs214, %r645; + fma.rn.f32.bf16 %r798, %rs216, %rs216, %r797; + fma.rn.f32.bf16 %r799, %rs217, %rs217, %r798; + fma.rn.f32.bf16 %r800, %rs218, %rs218, %r799; + fma.rn.f32.bf16 %r801, %rs219, %rs219, %r800; + fma.rn.f32.bf16 %r802, %rs220, %rs220, %r801; + fma.rn.f32.bf16 %r803, %rs221, %rs221, %r802; + fma.rn.f32.bf16 %r804, %rs222, %rs222, %r646; + fma.rn.f32.bf16 %r805, %rs224, %rs224, %r804; + fma.rn.f32.bf16 %r806, %rs225, %rs225, %r805; + fma.rn.f32.bf16 %r807, %rs226, %rs226, %r806; + fma.rn.f32.bf16 %r808, %rs227, %rs227, %r807; + fma.rn.f32.bf16 %r809, %rs228, %rs228, %r808; + fma.rn.f32.bf16 %r810, %rs229, %rs229, %r809; + fma.rn.f32.bf16 %r811, %rs230, %rs230, %r647; + fma.rn.f32.bf16 %r812, %rs232, %rs232, %r811; + fma.rn.f32.bf16 %r813, %rs233, %rs233, %r812; + fma.rn.f32.bf16 %r814, %rs234, %rs234, %r813; + fma.rn.f32.bf16 %r815, %rs235, %rs235, %r814; + fma.rn.f32.bf16 %r816, %rs236, %rs236, %r815; + fma.rn.f32.bf16 %r817, %rs237, %rs237, %r816; + fma.rn.f32.bf16 %r818, %rs238, %rs238, %r648; + fma.rn.f32.bf16 %r819, %rs240, %rs240, %r818; + fma.rn.f32.bf16 %r820, %rs241, %rs241, %r819; + fma.rn.f32.bf16 %r821, %rs242, %rs242, %r820; + fma.rn.f32.bf16 %r822, %rs243, %rs243, %r821; + fma.rn.f32.bf16 %r823, %rs244, %rs244, %r822; + fma.rn.f32.bf16 %r824, %rs245, %rs245, %r823; + fma.rn.f32.bf16 %r825, %rs246, %rs246, %r649; + fma.rn.f32.bf16 %r826, %rs248, %rs248, %r825; + fma.rn.f32.bf16 %r827, %rs249, %rs249, %r826; + fma.rn.f32.bf16 %r828, %rs250, %rs250, %r827; + fma.rn.f32.bf16 %r829, %rs251, %rs251, %r828; + fma.rn.f32.bf16 %r830, %rs252, %rs252, %r829; + fma.rn.f32.bf16 %r831, %rs253, %rs253, %r830; + fma.rn.f32.bf16 %r832, %rs254, %rs254, %r650; + fma.rn.f32.bf16 %r833, %rs256, %rs256, %r832; + fma.rn.f32.bf16 %r834, %rs257, %rs257, %r833; + fma.rn.f32.bf16 %r835, %rs258, %rs258, %r834; + fma.rn.f32.bf16 %r836, %rs259, %rs259, %r835; + fma.rn.f32.bf16 %r837, %rs260, %rs260, %r836; + fma.rn.f32.bf16 %r838, %rs261, %rs261, %r837; + fma.rn.f32.bf16 %r839, %rs262, %rs262, %r651; + fma.rn.f32.bf16 %r840, %rs264, %rs264, %r839; + fma.rn.f32.bf16 %r841, %rs265, %rs265, %r840; + fma.rn.f32.bf16 %r842, %rs266, %rs266, %r841; + fma.rn.f32.bf16 %r843, %rs267, %rs267, %r842; + fma.rn.f32.bf16 %r844, %rs268, %rs268, %r843; + fma.rn.f32.bf16 %r845, %rs269, %rs269, %r844; + fma.rn.f32.bf16 %r846, %rs270, %rs270, %r652; + fma.rn.f32.bf16 %r847, %rs272, %rs272, %r846; + fma.rn.f32.bf16 %r848, %rs273, %rs273, %r847; + fma.rn.f32.bf16 %r849, %rs274, %rs274, %r848; + fma.rn.f32.bf16 %r850, %rs275, %rs275, %r849; + fma.rn.f32.bf16 %r851, %rs276, %rs276, %r850; + fma.rn.f32.bf16 %r852, %rs277, %rs277, %r851; + fma.rn.f32.bf16 %r853, %rs278, %rs278, %r653; + fma.rn.f32.bf16 %r854, %rs280, %rs280, %r853; + fma.rn.f32.bf16 %r855, %rs281, %rs281, %r854; + fma.rn.f32.bf16 %r856, %rs282, %rs282, %r855; + fma.rn.f32.bf16 %r857, %rs283, %rs283, %r856; + fma.rn.f32.bf16 %r858, %rs284, %rs284, %r857; + fma.rn.f32.bf16 %r859, %rs285, %rs285, %r858; + fma.rn.f32.bf16 %r860, %rs286, %rs286, %r654; + fma.rn.f32.bf16 %r861, %rs288, %rs288, %r860; + fma.rn.f32.bf16 %r862, %rs289, %rs289, %r861; + fma.rn.f32.bf16 %r863, %rs290, %rs290, %r862; + fma.rn.f32.bf16 %r864, %rs291, %rs291, %r863; + fma.rn.f32.bf16 %r865, %rs292, %rs292, %r864; + fma.rn.f32.bf16 %r866, %rs293, %rs293, %r865; + fma.rn.f32.bf16 %r867, %rs294, %rs294, %r655; + fma.rn.f32.bf16 %r868, %rs296, %rs296, %r867; + fma.rn.f32.bf16 %r869, %rs297, %rs297, %r868; + fma.rn.f32.bf16 %r870, %rs298, %rs298, %r869; + fma.rn.f32.bf16 %r871, %rs299, %rs299, %r870; + fma.rn.f32.bf16 %r872, %rs300, %rs300, %r871; + fma.rn.f32.bf16 %r873, %rs301, %rs301, %r872; + fma.rn.f32.bf16 %r874, %rs302, %rs302, %r656; + fma.rn.f32.bf16 %r875, %rs304, %rs304, %r874; + fma.rn.f32.bf16 %r876, %rs305, %rs305, %r875; + fma.rn.f32.bf16 %r877, %rs306, %rs306, %r876; + fma.rn.f32.bf16 %r878, %rs307, %rs307, %r877; + fma.rn.f32.bf16 %r879, %rs308, %rs308, %r878; + fma.rn.f32.bf16 %r880, %rs309, %rs309, %r879; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r881, %r663, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r882, %r663, %r881; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r883, %r882, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r884, %r882, %r883; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r885, %r884, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r886, %r884, %r885; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r887, %r886, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r888, %r886, %r887; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r889, %r670, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r890, %r670, %r889; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r891, %r890, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r892, %r890, %r891; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r893, %r892, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r894, %r892, %r893; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r895, %r894, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r896, %r894, %r895; +$L__tmp18: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r897, %r677, 8, 31, -1; +$L__tmp19: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r898, %r677, %r897; +$L__tmp20: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r899, %r898, 4, 31, -1; +$L__tmp21: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r900, %r898, %r899; +$L__tmp22: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r901, %r900, 2, 31, -1; +$L__tmp23: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r902, %r900, %r901; +$L__tmp24: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r903, %r902, 1, 31, -1; +$L__tmp25: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r904, %r902, %r903; +$L__tmp26: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r905, %r684, 8, 31, -1; +$L__tmp27: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r906, %r684, %r905; +$L__tmp28: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r907, %r906, 4, 31, -1; +$L__tmp29: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r908, %r906, %r907; +$L__tmp30: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r909, %r908, 2, 31, -1; +$L__tmp31: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r910, %r908, %r909; +$L__tmp32: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r911, %r910, 1, 31, -1; +$L__tmp33: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r912, %r910, %r911; +$L__tmp34: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r913, %r691, 8, 31, -1; +$L__tmp35: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r914, %r691, %r913; +$L__tmp36: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r915, %r914, 4, 31, -1; +$L__tmp37: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r916, %r914, %r915; +$L__tmp38: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r917, %r916, 2, 31, -1; +$L__tmp39: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r918, %r916, %r917; +$L__tmp40: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r919, %r918, 1, 31, -1; +$L__tmp41: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r920, %r918, %r919; +$L__tmp42: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r921, %r698, 8, 31, -1; +$L__tmp43: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r922, %r698, %r921; +$L__tmp44: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r923, %r922, 4, 31, -1; +$L__tmp45: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r924, %r922, %r923; +$L__tmp46: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r925, %r924, 2, 31, -1; +$L__tmp47: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r926, %r924, %r925; +$L__tmp48: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r927, %r926, 1, 31, -1; +$L__tmp49: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r928, %r926, %r927; +$L__tmp50: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r929, %r705, 8, 31, -1; +$L__tmp51: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r930, %r705, %r929; +$L__tmp52: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r931, %r930, 4, 31, -1; +$L__tmp53: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r932, %r930, %r931; +$L__tmp54: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r933, %r932, 2, 31, -1; +$L__tmp55: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r934, %r932, %r933; +$L__tmp56: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r935, %r934, 1, 31, -1; +$L__tmp57: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r936, %r934, %r935; +$L__tmp58: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r937, %r712, 8, 31, -1; +$L__tmp59: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r938, %r712, %r937; +$L__tmp60: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r939, %r938, 4, 31, -1; +$L__tmp61: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r940, %r938, %r939; +$L__tmp62: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r941, %r940, 2, 31, -1; +$L__tmp63: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r942, %r940, %r941; +$L__tmp64: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r943, %r942, 1, 31, -1; +$L__tmp65: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r944, %r942, %r943; +$L__tmp66: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r945, %r719, 8, 31, -1; +$L__tmp67: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r946, %r719, %r945; +$L__tmp68: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r947, %r946, 4, 31, -1; +$L__tmp69: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r948, %r946, %r947; +$L__tmp70: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r949, %r948, 2, 31, -1; +$L__tmp71: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r950, %r948, %r949; +$L__tmp72: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r951, %r950, 1, 31, -1; +$L__tmp73: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r952, %r950, %r951; +$L__tmp74: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r953, %r726, 8, 31, -1; +$L__tmp75: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r954, %r726, %r953; +$L__tmp76: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r955, %r954, 4, 31, -1; +$L__tmp77: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r956, %r954, %r955; +$L__tmp78: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r957, %r956, 2, 31, -1; +$L__tmp79: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r958, %r956, %r957; +$L__tmp80: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r959, %r958, 1, 31, -1; +$L__tmp81: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r960, %r958, %r959; +$L__tmp82: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r961, %r733, 8, 31, -1; +$L__tmp83: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r962, %r733, %r961; +$L__tmp84: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r963, %r962, 4, 31, -1; +$L__tmp85: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r964, %r962, %r963; +$L__tmp86: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r965, %r964, 2, 31, -1; +$L__tmp87: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r966, %r964, %r965; +$L__tmp88: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r967, %r966, 1, 31, -1; +$L__tmp89: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r968, %r966, %r967; +$L__tmp90: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r969, %r740, 8, 31, -1; +$L__tmp91: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r970, %r740, %r969; +$L__tmp92: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r971, %r970, 4, 31, -1; +$L__tmp93: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r972, %r970, %r971; +$L__tmp94: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r973, %r972, 2, 31, -1; +$L__tmp95: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r974, %r972, %r973; +$L__tmp96: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r975, %r974, 1, 31, -1; +$L__tmp97: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r976, %r974, %r975; +$L__tmp98: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r977, %r747, 8, 31, -1; +$L__tmp99: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r978, %r747, %r977; +$L__tmp100: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r979, %r978, 4, 31, -1; +$L__tmp101: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r980, %r978, %r979; +$L__tmp102: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r981, %r980, 2, 31, -1; +$L__tmp103: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r982, %r980, %r981; +$L__tmp104: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r983, %r982, 1, 31, -1; +$L__tmp105: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r984, %r982, %r983; +$L__tmp106: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r985, %r754, 8, 31, -1; +$L__tmp107: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r986, %r754, %r985; +$L__tmp108: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r987, %r986, 4, 31, -1; +$L__tmp109: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r988, %r986, %r987; +$L__tmp110: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r989, %r988, 2, 31, -1; +$L__tmp111: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r990, %r988, %r989; +$L__tmp112: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r991, %r990, 1, 31, -1; +$L__tmp113: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r992, %r990, %r991; +$L__tmp114: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r993, %r761, 8, 31, -1; +$L__tmp115: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r994, %r761, %r993; +$L__tmp116: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r995, %r994, 4, 31, -1; +$L__tmp117: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r996, %r994, %r995; +$L__tmp118: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r997, %r996, 2, 31, -1; +$L__tmp119: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r998, %r996, %r997; +$L__tmp120: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r999, %r998, 1, 31, -1; +$L__tmp121: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1000, %r998, %r999; +$L__tmp122: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1001, %r768, 8, 31, -1; +$L__tmp123: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1002, %r768, %r1001; +$L__tmp124: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1003, %r1002, 4, 31, -1; +$L__tmp125: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1004, %r1002, %r1003; +$L__tmp126: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1005, %r1004, 2, 31, -1; +$L__tmp127: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1006, %r1004, %r1005; +$L__tmp128: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1007, %r1006, 1, 31, -1; +$L__tmp129: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1008, %r1006, %r1007; +$L__tmp130: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1009, %r775, 8, 31, -1; +$L__tmp131: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1010, %r775, %r1009; +$L__tmp132: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1011, %r1010, 4, 31, -1; +$L__tmp133: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1012, %r1010, %r1011; +$L__tmp134: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1013, %r1012, 2, 31, -1; +$L__tmp135: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1014, %r1012, %r1013; +$L__tmp136: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1015, %r1014, 1, 31, -1; +$L__tmp137: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1016, %r1014, %r1015; +$L__tmp138: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1017, %r782, 8, 31, -1; +$L__tmp139: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1018, %r782, %r1017; +$L__tmp140: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1019, %r1018, 4, 31, -1; +$L__tmp141: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1020, %r1018, %r1019; +$L__tmp142: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1021, %r1020, 2, 31, -1; +$L__tmp143: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1022, %r1020, %r1021; +$L__tmp144: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1023, %r1022, 1, 31, -1; +$L__tmp145: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1024, %r1022, %r1023; +$L__tmp146: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1025, %r789, 8, 31, -1; +$L__tmp147: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1026, %r789, %r1025; +$L__tmp148: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1027, %r1026, 4, 31, -1; +$L__tmp149: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1028, %r1026, %r1027; +$L__tmp150: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1029, %r1028, 2, 31, -1; +$L__tmp151: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1030, %r1028, %r1029; +$L__tmp152: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1031, %r1030, 1, 31, -1; +$L__tmp153: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1032, %r1030, %r1031; +$L__tmp154: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1033, %r796, 8, 31, -1; +$L__tmp155: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1034, %r796, %r1033; +$L__tmp156: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1035, %r1034, 4, 31, -1; +$L__tmp157: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1036, %r1034, %r1035; +$L__tmp158: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1037, %r1036, 2, 31, -1; +$L__tmp159: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1038, %r1036, %r1037; +$L__tmp160: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1039, %r1038, 1, 31, -1; +$L__tmp161: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1040, %r1038, %r1039; +$L__tmp162: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1041, %r803, 8, 31, -1; +$L__tmp163: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1042, %r803, %r1041; +$L__tmp164: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1043, %r1042, 4, 31, -1; +$L__tmp165: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1044, %r1042, %r1043; +$L__tmp166: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1045, %r1044, 2, 31, -1; +$L__tmp167: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1046, %r1044, %r1045; +$L__tmp168: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1047, %r1046, 1, 31, -1; +$L__tmp169: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1048, %r1046, %r1047; +$L__tmp170: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1049, %r810, 8, 31, -1; +$L__tmp171: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1050, %r810, %r1049; +$L__tmp172: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1051, %r1050, 4, 31, -1; +$L__tmp173: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1052, %r1050, %r1051; +$L__tmp174: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1053, %r1052, 2, 31, -1; +$L__tmp175: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1054, %r1052, %r1053; +$L__tmp176: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1055, %r1054, 1, 31, -1; +$L__tmp177: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1056, %r1054, %r1055; +$L__tmp178: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1057, %r817, 8, 31, -1; +$L__tmp179: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1058, %r817, %r1057; +$L__tmp180: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1059, %r1058, 4, 31, -1; +$L__tmp181: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1060, %r1058, %r1059; +$L__tmp182: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1061, %r1060, 2, 31, -1; +$L__tmp183: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1062, %r1060, %r1061; +$L__tmp184: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1063, %r1062, 1, 31, -1; +$L__tmp185: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1064, %r1062, %r1063; +$L__tmp186: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1065, %r824, 8, 31, -1; +$L__tmp187: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1066, %r824, %r1065; +$L__tmp188: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1067, %r1066, 4, 31, -1; +$L__tmp189: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1068, %r1066, %r1067; +$L__tmp190: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1069, %r1068, 2, 31, -1; +$L__tmp191: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1070, %r1068, %r1069; +$L__tmp192: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1071, %r1070, 1, 31, -1; +$L__tmp193: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1072, %r1070, %r1071; +$L__tmp194: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1073, %r831, 8, 31, -1; +$L__tmp195: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1074, %r831, %r1073; +$L__tmp196: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1075, %r1074, 4, 31, -1; +$L__tmp197: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1076, %r1074, %r1075; +$L__tmp198: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1077, %r1076, 2, 31, -1; +$L__tmp199: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1078, %r1076, %r1077; +$L__tmp200: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1079, %r1078, 1, 31, -1; +$L__tmp201: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1080, %r1078, %r1079; +$L__tmp202: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1081, %r838, 8, 31, -1; +$L__tmp203: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1082, %r838, %r1081; +$L__tmp204: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1083, %r1082, 4, 31, -1; +$L__tmp205: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1084, %r1082, %r1083; +$L__tmp206: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1085, %r1084, 2, 31, -1; +$L__tmp207: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1086, %r1084, %r1085; +$L__tmp208: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1087, %r1086, 1, 31, -1; +$L__tmp209: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1088, %r1086, %r1087; +$L__tmp210: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1089, %r845, 8, 31, -1; +$L__tmp211: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1090, %r845, %r1089; +$L__tmp212: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1091, %r1090, 4, 31, -1; +$L__tmp213: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1092, %r1090, %r1091; +$L__tmp214: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1093, %r1092, 2, 31, -1; +$L__tmp215: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1094, %r1092, %r1093; +$L__tmp216: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1095, %r1094, 1, 31, -1; +$L__tmp217: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1096, %r1094, %r1095; +$L__tmp218: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1097, %r852, 8, 31, -1; +$L__tmp219: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1098, %r852, %r1097; +$L__tmp220: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1099, %r1098, 4, 31, -1; +$L__tmp221: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1100, %r1098, %r1099; +$L__tmp222: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1101, %r1100, 2, 31, -1; +$L__tmp223: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1102, %r1100, %r1101; +$L__tmp224: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1103, %r1102, 1, 31, -1; +$L__tmp225: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1104, %r1102, %r1103; +$L__tmp226: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1105, %r859, 8, 31, -1; +$L__tmp227: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1106, %r859, %r1105; +$L__tmp228: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1107, %r1106, 4, 31, -1; +$L__tmp229: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1108, %r1106, %r1107; +$L__tmp230: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1109, %r1108, 2, 31, -1; +$L__tmp231: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1110, %r1108, %r1109; +$L__tmp232: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1111, %r1110, 1, 31, -1; +$L__tmp233: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1112, %r1110, %r1111; +$L__tmp234: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1113, %r866, 8, 31, -1; +$L__tmp235: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1114, %r866, %r1113; +$L__tmp236: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1115, %r1114, 4, 31, -1; +$L__tmp237: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1116, %r1114, %r1115; +$L__tmp238: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1117, %r1116, 2, 31, -1; +$L__tmp239: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1118, %r1116, %r1117; +$L__tmp240: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1119, %r1118, 1, 31, -1; +$L__tmp241: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1120, %r1118, %r1119; +$L__tmp242: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1121, %r873, 8, 31, -1; +$L__tmp243: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1122, %r873, %r1121; +$L__tmp244: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1123, %r1122, 4, 31, -1; +$L__tmp245: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1124, %r1122, %r1123; +$L__tmp246: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1125, %r1124, 2, 31, -1; +$L__tmp247: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1126, %r1124, %r1125; +$L__tmp248: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1127, %r1126, 1, 31, -1; +$L__tmp249: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1128, %r1126, %r1127; +$L__tmp250: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1129, %r880, 8, 31, -1; +$L__tmp251: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1130, %r880, %r1129; +$L__tmp252: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1131, %r1130, 4, 31, -1; +$L__tmp253: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1132, %r1130, %r1131; +$L__tmp254: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1133, %r1132, 2, 31, -1; +$L__tmp255: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1134, %r1132, %r1133; +$L__tmp256: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1135, %r1134, 1, 31, -1; +$L__tmp257: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1136, %r1134, %r1135; +$L__tmp258: + .loc 1 80 48 // l2norm.py:80:48 + add.f32 %r1137, %r301, %r888; + add.f32 %r1138, %r301, %r896; + add.f32 %r1139, %r301, %r904; + add.f32 %r1140, %r301, %r912; + add.f32 %r1141, %r301, %r920; + add.f32 %r1142, %r301, %r928; + add.f32 %r1143, %r301, %r936; + add.f32 %r1144, %r301, %r944; + add.f32 %r1145, %r301, %r952; + add.f32 %r1146, %r301, %r960; + add.f32 %r1147, %r301, %r968; + add.f32 %r1148, %r301, %r976; + add.f32 %r1149, %r301, %r984; + add.f32 %r1150, %r301, %r992; + add.f32 %r1151, %r301, %r1000; + add.f32 %r1152, %r301, %r1008; + add.f32 %r1153, %r301, %r1016; + add.f32 %r1154, %r301, %r1024; + add.f32 %r1155, %r301, %r1032; + add.f32 %r1156, %r301, %r1040; + add.f32 %r1157, %r301, %r1048; + add.f32 %r1158, %r301, %r1056; + add.f32 %r1159, %r301, %r1064; + add.f32 %r1160, %r301, %r1072; + add.f32 %r1161, %r301, %r1080; + add.f32 %r1162, %r301, %r1088; + add.f32 %r1163, %r301, %r1096; + add.f32 %r1164, %r301, %r1104; + add.f32 %r1165, %r301, %r1112; + add.f32 %r1166, %r301, %r1120; + add.f32 %r1167, %r301, %r1128; + add.f32 %r1168, %r301, %r1136; + .loc 1 80 25 // l2norm.py:80:25 + sqrt.approx.ftz.f32 %r1169, %r1137; + sqrt.approx.ftz.f32 %r1170, %r1138; + sqrt.approx.ftz.f32 %r1171, %r1139; + sqrt.approx.ftz.f32 %r1172, %r1140; + sqrt.approx.ftz.f32 %r1173, %r1141; + sqrt.approx.ftz.f32 %r1174, %r1142; + sqrt.approx.ftz.f32 %r1175, %r1143; + sqrt.approx.ftz.f32 %r1176, %r1144; + sqrt.approx.ftz.f32 %r1177, %r1145; + sqrt.approx.ftz.f32 %r1178, %r1146; + sqrt.approx.ftz.f32 %r1179, %r1147; + sqrt.approx.ftz.f32 %r1180, %r1148; + sqrt.approx.ftz.f32 %r1181, %r1149; + sqrt.approx.ftz.f32 %r1182, %r1150; + sqrt.approx.ftz.f32 %r1183, %r1151; + sqrt.approx.ftz.f32 %r1184, %r1152; + sqrt.approx.ftz.f32 %r1185, %r1153; + sqrt.approx.ftz.f32 %r1186, %r1154; + sqrt.approx.ftz.f32 %r1187, %r1155; + sqrt.approx.ftz.f32 %r1188, %r1156; + sqrt.approx.ftz.f32 %r1189, %r1157; + sqrt.approx.ftz.f32 %r1190, %r1158; + sqrt.approx.ftz.f32 %r1191, %r1159; + sqrt.approx.ftz.f32 %r1192, %r1160; + sqrt.approx.ftz.f32 %r1193, %r1161; + sqrt.approx.ftz.f32 %r1194, %r1162; + sqrt.approx.ftz.f32 %r1195, %r1163; + sqrt.approx.ftz.f32 %r1196, %r1164; + sqrt.approx.ftz.f32 %r1197, %r1165; + sqrt.approx.ftz.f32 %r1198, %r1166; + sqrt.approx.ftz.f32 %r1199, %r1167; + sqrt.approx.ftz.f32 %r1200, %r1168; + .loc 1 71 34 // l2norm.py:71:34 + shl.b32 %r1201, %r302, 1; + and.b32 %r1202, %r1201, 126; + or.b32 %r1203, %r1202, 1; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd201, %r1203; + or.b64 %rd202, %rd70, %rd201; + cvt.u64.u32 %rd203, %r1202; + or.b64 %rd204, %rd70, %rd203; + .loc 1 77 20 // l2norm.py:77:20 + shl.b64 %rd205, %rd204, 2; + add.s64 %rd65, %rd69, %rd205; + add.s64 %rd66, %rd65, 4; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd206, %rd68, %rd198; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd64, %rd206, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd207, %rd68, %rd196; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd63, %rd207, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd208, %rd68, %rd194; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd62, %rd208, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd209, %rd68, %rd192; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd61, %rd209, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd210, %rd68, %rd190; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd60, %rd210, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd211, %rd68, %rd188; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd59, %rd211, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd212, %rd68, %rd186; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd58, %rd212, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd213, %rd68, %rd184; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd57, %rd213, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd214, %rd68, %rd182; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd56, %rd214, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd215, %rd68, %rd180; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd55, %rd215, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd216, %rd68, %rd178; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd54, %rd216, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd217, %rd68, %rd176; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd53, %rd217, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd218, %rd68, %rd174; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd52, %rd218, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd219, %rd68, %rd172; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd51, %rd219, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd220, %rd68, %rd170; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd50, %rd220, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd221, %rd68, %rd168; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd49, %rd221, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd222, %rd68, %rd166; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd48, %rd222, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd223, %rd68, %rd164; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd47, %rd223, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd224, %rd68, %rd162; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd46, %rd224, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd225, %rd68, %rd160; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd45, %rd225, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd226, %rd68, %rd158; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd44, %rd226, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd227, %rd68, %rd156; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd43, %rd227, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd228, %rd68, %rd154; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd42, %rd228, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd229, %rd68, %rd152; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd41, %rd229, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd230, %rd68, %rd150; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd40, %rd230, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd231, %rd68, %rd148; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd39, %rd231, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd232, %rd68, %rd146; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd38, %rd232, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd233, %rd68, %rd144; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd37, %rd233, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd234, %rd68, %rd142; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd36, %rd234, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd235, %rd68, %rd140; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd35, %rd235, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd236, %rd68, %rd138; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd34, %rd236, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd237, %rd68, %rd136; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd33, %rd237, %rd200; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p34, %rd202, %rd71; + setp.lt.s64 %p33, %rd204, %rd71; + mov.b32 %r1204, 0f3F800000; + .loc 1 80 17 // l2norm.py:80:17 + div.full.f32 %r259, %r1204, %r1169; + div.full.f32 %r260, %r1204, %r1170; + div.full.f32 %r261, %r1204, %r1171; + div.full.f32 %r262, %r1204, %r1172; + div.full.f32 %r264, %r1204, %r1173; + div.full.f32 %r265, %r1204, %r1174; + div.full.f32 %r266, %r1204, %r1175; + div.full.f32 %r267, %r1204, %r1176; + div.full.f32 %r269, %r1204, %r1177; + div.full.f32 %r270, %r1204, %r1178; + div.full.f32 %r271, %r1204, %r1179; + div.full.f32 %r272, %r1204, %r1180; + div.full.f32 %r274, %r1204, %r1181; + div.full.f32 %r275, %r1204, %r1182; + div.full.f32 %r276, %r1204, %r1183; + div.full.f32 %r277, %r1204, %r1184; + div.full.f32 %r279, %r1204, %r1185; + div.full.f32 %r280, %r1204, %r1186; + div.full.f32 %r281, %r1204, %r1187; + div.full.f32 %r282, %r1204, %r1188; + div.full.f32 %r284, %r1204, %r1189; + div.full.f32 %r285, %r1204, %r1190; + div.full.f32 %r286, %r1204, %r1191; + div.full.f32 %r287, %r1204, %r1192; + div.full.f32 %r289, %r1204, %r1193; + div.full.f32 %r290, %r1204, %r1194; + div.full.f32 %r291, %r1204, %r1195; + div.full.f32 %r292, %r1204, %r1196; + div.full.f32 %r294, %r1204, %r1197; + div.full.f32 %r295, %r1204, %r1198; + div.full.f32 %r296, %r1204, %r1199; + div.full.f32 %r297, %r1204, %r1200; + .loc 1 81 16 // l2norm.py:81:16 + mul.f32 %r1205, %r259, %r369; + mul.f32 %r1206, %r259, %r370; + mul.f32 %r1207, %r259, %r371; + mul.f32 %r1208, %r259, %r372; + mul.f32 %r1209, %r259, %r373; + mul.f32 %r1210, %r259, %r374; + mul.f32 %r1211, %r259, %r375; + mul.f32 %r1212, %r259, %r376; + mul.f32 %r1213, %r260, %r377; + mul.f32 %r1214, %r260, %r378; + mul.f32 %r1215, %r260, %r379; + mul.f32 %r1216, %r260, %r380; + mul.f32 %r1217, %r260, %r381; + mul.f32 %r1218, %r260, %r382; + mul.f32 %r1219, %r260, %r383; + mul.f32 %r1220, %r260, %r384; + mul.f32 %r1221, %r261, %r385; + mul.f32 %r1222, %r261, %r386; + mul.f32 %r1223, %r261, %r387; + mul.f32 %r1224, %r261, %r388; + mul.f32 %r1225, %r261, %r389; + mul.f32 %r1226, %r261, %r390; + mul.f32 %r1227, %r261, %r391; + mul.f32 %r1228, %r261, %r392; + mul.f32 %r1229, %r262, %r393; + mul.f32 %r1230, %r262, %r394; + mul.f32 %r1231, %r262, %r395; + mul.f32 %r1232, %r262, %r396; + mul.f32 %r1233, %r262, %r397; + mul.f32 %r1234, %r262, %r398; + mul.f32 %r1235, %r262, %r399; + mul.f32 %r1236, %r262, %r400; + mul.f32 %r1237, %r264, %r401; + mul.f32 %r1238, %r264, %r402; + mul.f32 %r1239, %r264, %r403; + mul.f32 %r1240, %r264, %r404; + mul.f32 %r1241, %r264, %r405; + mul.f32 %r1242, %r264, %r406; + mul.f32 %r1243, %r264, %r407; + mul.f32 %r1244, %r264, %r408; + mul.f32 %r1245, %r265, %r409; + mul.f32 %r1246, %r265, %r410; + mul.f32 %r1247, %r265, %r411; + mul.f32 %r1248, %r265, %r412; + mul.f32 %r1249, %r265, %r413; + mul.f32 %r1250, %r265, %r414; + mul.f32 %r1251, %r265, %r415; + mul.f32 %r1252, %r265, %r416; + mul.f32 %r1253, %r266, %r417; + mul.f32 %r1254, %r266, %r418; + mul.f32 %r1255, %r266, %r419; + mul.f32 %r1256, %r266, %r420; + mul.f32 %r1257, %r266, %r421; + mul.f32 %r1258, %r266, %r422; + mul.f32 %r1259, %r266, %r423; + mul.f32 %r1260, %r266, %r424; + mul.f32 %r1261, %r267, %r425; + mul.f32 %r1262, %r267, %r426; + mul.f32 %r1263, %r267, %r427; + mul.f32 %r1264, %r267, %r428; + mul.f32 %r1265, %r267, %r429; + mul.f32 %r1266, %r267, %r430; + mul.f32 %r1267, %r267, %r431; + mul.f32 %r1268, %r267, %r432; + mul.f32 %r1269, %r269, %r433; + mul.f32 %r1270, %r269, %r434; + mul.f32 %r1271, %r269, %r435; + mul.f32 %r1272, %r269, %r436; + mul.f32 %r1273, %r269, %r437; + mul.f32 %r1274, %r269, %r438; + mul.f32 %r1275, %r269, %r439; + mul.f32 %r1276, %r269, %r440; + mul.f32 %r1277, %r270, %r441; + mul.f32 %r1278, %r270, %r442; + mul.f32 %r1279, %r270, %r443; + mul.f32 %r1280, %r270, %r444; + mul.f32 %r1281, %r270, %r445; + mul.f32 %r1282, %r270, %r446; + mul.f32 %r1283, %r270, %r447; + mul.f32 %r1284, %r270, %r448; + mul.f32 %r1285, %r271, %r449; + mul.f32 %r1286, %r271, %r450; + mul.f32 %r1287, %r271, %r451; + mul.f32 %r1288, %r271, %r452; + mul.f32 %r1289, %r271, %r453; + mul.f32 %r1290, %r271, %r454; + mul.f32 %r1291, %r271, %r455; + mul.f32 %r1292, %r271, %r456; + mul.f32 %r1293, %r272, %r457; + mul.f32 %r1294, %r272, %r458; + mul.f32 %r1295, %r272, %r459; + mul.f32 %r1296, %r272, %r460; + mul.f32 %r1297, %r272, %r461; + mul.f32 %r1298, %r272, %r462; + mul.f32 %r1299, %r272, %r463; + mul.f32 %r1300, %r272, %r464; + mul.f32 %r1301, %r274, %r465; + mul.f32 %r1302, %r274, %r466; + mul.f32 %r1303, %r274, %r467; + mul.f32 %r1304, %r274, %r468; + mul.f32 %r1305, %r274, %r469; + mul.f32 %r1306, %r274, %r470; + mul.f32 %r1307, %r274, %r471; + mul.f32 %r1308, %r274, %r472; + mul.f32 %r1309, %r275, %r473; + mul.f32 %r1310, %r275, %r474; + mul.f32 %r1311, %r275, %r475; + mul.f32 %r1312, %r275, %r476; + mul.f32 %r1313, %r275, %r477; + mul.f32 %r1314, %r275, %r478; + mul.f32 %r1315, %r275, %r479; + mul.f32 %r1316, %r275, %r480; + mul.f32 %r1317, %r276, %r481; + mul.f32 %r1318, %r276, %r482; + mul.f32 %r1319, %r276, %r483; + mul.f32 %r1320, %r276, %r484; + mul.f32 %r1321, %r276, %r485; + mul.f32 %r1322, %r276, %r486; + mul.f32 %r1323, %r276, %r487; + mul.f32 %r1324, %r276, %r488; + mul.f32 %r1325, %r277, %r489; + mul.f32 %r1326, %r277, %r490; + mul.f32 %r1327, %r277, %r491; + mul.f32 %r1328, %r277, %r492; + mul.f32 %r1329, %r277, %r493; + mul.f32 %r1330, %r277, %r494; + mul.f32 %r1331, %r277, %r495; + mul.f32 %r1332, %r277, %r496; + mul.f32 %r1333, %r279, %r497; + mul.f32 %r1334, %r279, %r498; + mul.f32 %r1335, %r279, %r499; + mul.f32 %r1336, %r279, %r500; + mul.f32 %r1337, %r279, %r501; + mul.f32 %r1338, %r279, %r502; + mul.f32 %r1339, %r279, %r503; + mul.f32 %r1340, %r279, %r504; + mul.f32 %r1341, %r280, %r505; + mul.f32 %r1342, %r280, %r506; + mul.f32 %r1343, %r280, %r507; + mul.f32 %r1344, %r280, %r508; + mul.f32 %r1345, %r280, %r509; + mul.f32 %r1346, %r280, %r510; + mul.f32 %r1347, %r280, %r511; + mul.f32 %r1348, %r280, %r512; + mul.f32 %r1349, %r281, %r513; + mul.f32 %r1350, %r281, %r514; + mul.f32 %r1351, %r281, %r515; + mul.f32 %r1352, %r281, %r516; + mul.f32 %r1353, %r281, %r517; + mul.f32 %r1354, %r281, %r518; + mul.f32 %r1355, %r281, %r519; + mul.f32 %r1356, %r281, %r520; + mul.f32 %r1357, %r282, %r521; + mul.f32 %r1358, %r282, %r522; + mul.f32 %r1359, %r282, %r523; + mul.f32 %r1360, %r282, %r524; + mul.f32 %r1361, %r282, %r525; + mul.f32 %r1362, %r282, %r526; + mul.f32 %r1363, %r282, %r527; + mul.f32 %r1364, %r282, %r528; + mul.f32 %r1365, %r284, %r529; + mul.f32 %r1366, %r284, %r530; + mul.f32 %r1367, %r284, %r531; + mul.f32 %r1368, %r284, %r532; + mul.f32 %r1369, %r284, %r533; + mul.f32 %r1370, %r284, %r534; + mul.f32 %r1371, %r284, %r535; + mul.f32 %r1372, %r284, %r536; + mul.f32 %r1373, %r285, %r537; + mul.f32 %r1374, %r285, %r538; + mul.f32 %r1375, %r285, %r539; + mul.f32 %r1376, %r285, %r540; + mul.f32 %r1377, %r285, %r541; + mul.f32 %r1378, %r285, %r542; + mul.f32 %r1379, %r285, %r543; + mul.f32 %r1380, %r285, %r544; + mul.f32 %r1381, %r286, %r545; + mul.f32 %r1382, %r286, %r546; + mul.f32 %r1383, %r286, %r547; + mul.f32 %r1384, %r286, %r548; + mul.f32 %r1385, %r286, %r549; + mul.f32 %r1386, %r286, %r550; + mul.f32 %r1387, %r286, %r551; + mul.f32 %r1388, %r286, %r552; + mul.f32 %r1389, %r287, %r553; + mul.f32 %r1390, %r287, %r554; + mul.f32 %r1391, %r287, %r555; + mul.f32 %r1392, %r287, %r556; + mul.f32 %r1393, %r287, %r557; + mul.f32 %r1394, %r287, %r558; + mul.f32 %r1395, %r287, %r559; + mul.f32 %r1396, %r287, %r560; + mul.f32 %r1397, %r289, %r561; + mul.f32 %r1398, %r289, %r562; + mul.f32 %r1399, %r289, %r563; + mul.f32 %r1400, %r289, %r564; + mul.f32 %r1401, %r289, %r565; + mul.f32 %r1402, %r289, %r566; + mul.f32 %r1403, %r289, %r567; + mul.f32 %r1404, %r289, %r568; + mul.f32 %r1405, %r290, %r569; + mul.f32 %r1406, %r290, %r570; + mul.f32 %r1407, %r290, %r571; + mul.f32 %r1408, %r290, %r572; + mul.f32 %r1409, %r290, %r573; + mul.f32 %r1410, %r290, %r574; + mul.f32 %r1411, %r290, %r575; + mul.f32 %r1412, %r290, %r576; + mul.f32 %r1413, %r291, %r577; + mul.f32 %r1414, %r291, %r578; + mul.f32 %r1415, %r291, %r579; + mul.f32 %r1416, %r291, %r580; + mul.f32 %r1417, %r291, %r581; + mul.f32 %r1418, %r291, %r582; + mul.f32 %r1419, %r291, %r583; + mul.f32 %r1420, %r291, %r584; + mul.f32 %r1421, %r292, %r585; + mul.f32 %r1422, %r292, %r586; + mul.f32 %r1423, %r292, %r587; + mul.f32 %r1424, %r292, %r588; + mul.f32 %r1425, %r292, %r589; + mul.f32 %r1426, %r292, %r590; + mul.f32 %r1427, %r292, %r591; + mul.f32 %r1428, %r292, %r592; + mul.f32 %r1429, %r294, %r593; + mul.f32 %r1430, %r294, %r594; + mul.f32 %r1431, %r294, %r595; + mul.f32 %r1432, %r294, %r596; + mul.f32 %r1433, %r294, %r597; + mul.f32 %r1434, %r294, %r598; + mul.f32 %r1435, %r294, %r599; + mul.f32 %r1436, %r294, %r600; + mul.f32 %r1437, %r295, %r601; + mul.f32 %r1438, %r295, %r602; + mul.f32 %r1439, %r295, %r603; + mul.f32 %r1440, %r295, %r604; + mul.f32 %r1441, %r295, %r605; + mul.f32 %r1442, %r295, %r606; + mul.f32 %r1443, %r295, %r607; + mul.f32 %r1444, %r295, %r608; + mul.f32 %r1445, %r296, %r609; + mul.f32 %r1446, %r296, %r610; + mul.f32 %r1447, %r296, %r611; + mul.f32 %r1448, %r296, %r612; + mul.f32 %r1449, %r296, %r613; + mul.f32 %r1450, %r296, %r614; + mul.f32 %r1451, %r296, %r615; + mul.f32 %r1452, %r296, %r616; + mul.f32 %r1453, %r297, %r617; + mul.f32 %r1454, %r297, %r618; + mul.f32 %r1455, %r297, %r619; + mul.f32 %r1456, %r297, %r620; + mul.f32 %r1457, %r297, %r621; + mul.f32 %r1458, %r297, %r622; + mul.f32 %r1459, %r297, %r623; + mul.f32 %r1460, %r297, %r624; + .loc 1 83 18 // l2norm.py:83:18 + cvt.rn.bf16x2.f32 %r130, %r1206, %r1205; + cvt.rn.bf16x2.f32 %r131, %r1208, %r1207; + cvt.rn.bf16x2.f32 %r132, %r1210, %r1209; + cvt.rn.bf16x2.f32 %r133, %r1212, %r1211; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd33 + 0 ], { %r130, %r131, %r132, %r133 }; + // end inline asm + cvt.rn.bf16x2.f32 %r134, %r1214, %r1213; + cvt.rn.bf16x2.f32 %r135, %r1216, %r1215; + cvt.rn.bf16x2.f32 %r136, %r1218, %r1217; + cvt.rn.bf16x2.f32 %r137, %r1220, %r1219; + // begin inline asm + @%p3 st.global.v4.b32 [ %rd34 + 0 ], { %r134, %r135, %r136, %r137 }; + // end inline asm + cvt.rn.bf16x2.f32 %r138, %r1222, %r1221; + cvt.rn.bf16x2.f32 %r139, %r1224, %r1223; + cvt.rn.bf16x2.f32 %r140, %r1226, %r1225; + cvt.rn.bf16x2.f32 %r141, %r1228, %r1227; + // begin inline asm + @%p4 st.global.v4.b32 [ %rd35 + 0 ], { %r138, %r139, %r140, %r141 }; + // end inline asm + cvt.rn.bf16x2.f32 %r142, %r1230, %r1229; + cvt.rn.bf16x2.f32 %r143, %r1232, %r1231; + cvt.rn.bf16x2.f32 %r144, %r1234, %r1233; + cvt.rn.bf16x2.f32 %r145, %r1236, %r1235; + // begin inline asm + @%p5 st.global.v4.b32 [ %rd36 + 0 ], { %r142, %r143, %r144, %r145 }; + // end inline asm + cvt.rn.bf16x2.f32 %r146, %r1238, %r1237; + cvt.rn.bf16x2.f32 %r147, %r1240, %r1239; + cvt.rn.bf16x2.f32 %r148, %r1242, %r1241; + cvt.rn.bf16x2.f32 %r149, %r1244, %r1243; + // begin inline asm + @%p6 st.global.v4.b32 [ %rd37 + 0 ], { %r146, %r147, %r148, %r149 }; + // end inline asm + cvt.rn.bf16x2.f32 %r150, %r1246, %r1245; + cvt.rn.bf16x2.f32 %r151, %r1248, %r1247; + cvt.rn.bf16x2.f32 %r152, %r1250, %r1249; + cvt.rn.bf16x2.f32 %r153, %r1252, %r1251; + // begin inline asm + @%p7 st.global.v4.b32 [ %rd38 + 0 ], { %r150, %r151, %r152, %r153 }; + // end inline asm + cvt.rn.bf16x2.f32 %r154, %r1254, %r1253; + cvt.rn.bf16x2.f32 %r155, %r1256, %r1255; + cvt.rn.bf16x2.f32 %r156, %r1258, %r1257; + cvt.rn.bf16x2.f32 %r157, %r1260, %r1259; + // begin inline asm + @%p8 st.global.v4.b32 [ %rd39 + 0 ], { %r154, %r155, %r156, %r157 }; + // end inline asm + cvt.rn.bf16x2.f32 %r158, %r1262, %r1261; + cvt.rn.bf16x2.f32 %r159, %r1264, %r1263; + cvt.rn.bf16x2.f32 %r160, %r1266, %r1265; + cvt.rn.bf16x2.f32 %r161, %r1268, %r1267; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd40 + 0 ], { %r158, %r159, %r160, %r161 }; + // end inline asm + cvt.rn.bf16x2.f32 %r162, %r1270, %r1269; + cvt.rn.bf16x2.f32 %r163, %r1272, %r1271; + cvt.rn.bf16x2.f32 %r164, %r1274, %r1273; + cvt.rn.bf16x2.f32 %r165, %r1276, %r1275; + // begin inline asm + @%p9 st.global.v4.b32 [ %rd41 + 0 ], { %r162, %r163, %r164, %r165 }; + // end inline asm + cvt.rn.bf16x2.f32 %r166, %r1278, %r1277; + cvt.rn.bf16x2.f32 %r167, %r1280, %r1279; + cvt.rn.bf16x2.f32 %r168, %r1282, %r1281; + cvt.rn.bf16x2.f32 %r169, %r1284, %r1283; + // begin inline asm + @%p10 st.global.v4.b32 [ %rd42 + 0 ], { %r166, %r167, %r168, %r169 }; + // end inline asm + cvt.rn.bf16x2.f32 %r170, %r1286, %r1285; + cvt.rn.bf16x2.f32 %r171, %r1288, %r1287; + cvt.rn.bf16x2.f32 %r172, %r1290, %r1289; + cvt.rn.bf16x2.f32 %r173, %r1292, %r1291; + // begin inline asm + @%p11 st.global.v4.b32 [ %rd43 + 0 ], { %r170, %r171, %r172, %r173 }; + // end inline asm + cvt.rn.bf16x2.f32 %r174, %r1294, %r1293; + cvt.rn.bf16x2.f32 %r175, %r1296, %r1295; + cvt.rn.bf16x2.f32 %r176, %r1298, %r1297; + cvt.rn.bf16x2.f32 %r177, %r1300, %r1299; + // begin inline asm + @%p12 st.global.v4.b32 [ %rd44 + 0 ], { %r174, %r175, %r176, %r177 }; + // end inline asm + cvt.rn.bf16x2.f32 %r178, %r1302, %r1301; + cvt.rn.bf16x2.f32 %r179, %r1304, %r1303; + cvt.rn.bf16x2.f32 %r180, %r1306, %r1305; + cvt.rn.bf16x2.f32 %r181, %r1308, %r1307; + // begin inline asm + @%p13 st.global.v4.b32 [ %rd45 + 0 ], { %r178, %r179, %r180, %r181 }; + // end inline asm + cvt.rn.bf16x2.f32 %r182, %r1310, %r1309; + cvt.rn.bf16x2.f32 %r183, %r1312, %r1311; + cvt.rn.bf16x2.f32 %r184, %r1314, %r1313; + cvt.rn.bf16x2.f32 %r185, %r1316, %r1315; + // begin inline asm + @%p14 st.global.v4.b32 [ %rd46 + 0 ], { %r182, %r183, %r184, %r185 }; + // end inline asm + cvt.rn.bf16x2.f32 %r186, %r1318, %r1317; + cvt.rn.bf16x2.f32 %r187, %r1320, %r1319; + cvt.rn.bf16x2.f32 %r188, %r1322, %r1321; + cvt.rn.bf16x2.f32 %r189, %r1324, %r1323; + // begin inline asm + @%p15 st.global.v4.b32 [ %rd47 + 0 ], { %r186, %r187, %r188, %r189 }; + // end inline asm + cvt.rn.bf16x2.f32 %r190, %r1326, %r1325; + cvt.rn.bf16x2.f32 %r191, %r1328, %r1327; + cvt.rn.bf16x2.f32 %r192, %r1330, %r1329; + cvt.rn.bf16x2.f32 %r193, %r1332, %r1331; + // begin inline asm + @%p16 st.global.v4.b32 [ %rd48 + 0 ], { %r190, %r191, %r192, %r193 }; + // end inline asm + cvt.rn.bf16x2.f32 %r194, %r1334, %r1333; + cvt.rn.bf16x2.f32 %r195, %r1336, %r1335; + cvt.rn.bf16x2.f32 %r196, %r1338, %r1337; + cvt.rn.bf16x2.f32 %r197, %r1340, %r1339; + // begin inline asm + @%p17 st.global.v4.b32 [ %rd49 + 0 ], { %r194, %r195, %r196, %r197 }; + // end inline asm + cvt.rn.bf16x2.f32 %r198, %r1342, %r1341; + cvt.rn.bf16x2.f32 %r199, %r1344, %r1343; + cvt.rn.bf16x2.f32 %r200, %r1346, %r1345; + cvt.rn.bf16x2.f32 %r201, %r1348, %r1347; + // begin inline asm + @%p18 st.global.v4.b32 [ %rd50 + 0 ], { %r198, %r199, %r200, %r201 }; + // end inline asm + cvt.rn.bf16x2.f32 %r202, %r1350, %r1349; + cvt.rn.bf16x2.f32 %r203, %r1352, %r1351; + cvt.rn.bf16x2.f32 %r204, %r1354, %r1353; + cvt.rn.bf16x2.f32 %r205, %r1356, %r1355; + // begin inline asm + @%p19 st.global.v4.b32 [ %rd51 + 0 ], { %r202, %r203, %r204, %r205 }; + // end inline asm + cvt.rn.bf16x2.f32 %r206, %r1358, %r1357; + cvt.rn.bf16x2.f32 %r207, %r1360, %r1359; + cvt.rn.bf16x2.f32 %r208, %r1362, %r1361; + cvt.rn.bf16x2.f32 %r209, %r1364, %r1363; + // begin inline asm + @%p20 st.global.v4.b32 [ %rd52 + 0 ], { %r206, %r207, %r208, %r209 }; + // end inline asm + cvt.rn.bf16x2.f32 %r210, %r1366, %r1365; + cvt.rn.bf16x2.f32 %r211, %r1368, %r1367; + cvt.rn.bf16x2.f32 %r212, %r1370, %r1369; + cvt.rn.bf16x2.f32 %r213, %r1372, %r1371; + // begin inline asm + @%p21 st.global.v4.b32 [ %rd53 + 0 ], { %r210, %r211, %r212, %r213 }; + // end inline asm + cvt.rn.bf16x2.f32 %r214, %r1374, %r1373; + cvt.rn.bf16x2.f32 %r215, %r1376, %r1375; + cvt.rn.bf16x2.f32 %r216, %r1378, %r1377; + cvt.rn.bf16x2.f32 %r217, %r1380, %r1379; + // begin inline asm + @%p22 st.global.v4.b32 [ %rd54 + 0 ], { %r214, %r215, %r216, %r217 }; + // end inline asm + cvt.rn.bf16x2.f32 %r218, %r1382, %r1381; + cvt.rn.bf16x2.f32 %r219, %r1384, %r1383; + cvt.rn.bf16x2.f32 %r220, %r1386, %r1385; + cvt.rn.bf16x2.f32 %r221, %r1388, %r1387; + // begin inline asm + @%p23 st.global.v4.b32 [ %rd55 + 0 ], { %r218, %r219, %r220, %r221 }; + // end inline asm + cvt.rn.bf16x2.f32 %r222, %r1390, %r1389; + cvt.rn.bf16x2.f32 %r223, %r1392, %r1391; + cvt.rn.bf16x2.f32 %r224, %r1394, %r1393; + cvt.rn.bf16x2.f32 %r225, %r1396, %r1395; + // begin inline asm + @%p24 st.global.v4.b32 [ %rd56 + 0 ], { %r222, %r223, %r224, %r225 }; + // end inline asm + cvt.rn.bf16x2.f32 %r226, %r1398, %r1397; + cvt.rn.bf16x2.f32 %r227, %r1400, %r1399; + cvt.rn.bf16x2.f32 %r228, %r1402, %r1401; + cvt.rn.bf16x2.f32 %r229, %r1404, %r1403; + // begin inline asm + @%p25 st.global.v4.b32 [ %rd57 + 0 ], { %r226, %r227, %r228, %r229 }; + // end inline asm + cvt.rn.bf16x2.f32 %r230, %r1406, %r1405; + cvt.rn.bf16x2.f32 %r231, %r1408, %r1407; + cvt.rn.bf16x2.f32 %r232, %r1410, %r1409; + cvt.rn.bf16x2.f32 %r233, %r1412, %r1411; + // begin inline asm + @%p26 st.global.v4.b32 [ %rd58 + 0 ], { %r230, %r231, %r232, %r233 }; + // end inline asm + cvt.rn.bf16x2.f32 %r234, %r1414, %r1413; + cvt.rn.bf16x2.f32 %r235, %r1416, %r1415; + cvt.rn.bf16x2.f32 %r236, %r1418, %r1417; + cvt.rn.bf16x2.f32 %r237, %r1420, %r1419; + // begin inline asm + @%p27 st.global.v4.b32 [ %rd59 + 0 ], { %r234, %r235, %r236, %r237 }; + // end inline asm + cvt.rn.bf16x2.f32 %r238, %r1422, %r1421; + cvt.rn.bf16x2.f32 %r239, %r1424, %r1423; + cvt.rn.bf16x2.f32 %r240, %r1426, %r1425; + cvt.rn.bf16x2.f32 %r241, %r1428, %r1427; + // begin inline asm + @%p28 st.global.v4.b32 [ %rd60 + 0 ], { %r238, %r239, %r240, %r241 }; + // end inline asm + cvt.rn.bf16x2.f32 %r242, %r1430, %r1429; + cvt.rn.bf16x2.f32 %r243, %r1432, %r1431; + cvt.rn.bf16x2.f32 %r244, %r1434, %r1433; + cvt.rn.bf16x2.f32 %r245, %r1436, %r1435; + // begin inline asm + @%p29 st.global.v4.b32 [ %rd61 + 0 ], { %r242, %r243, %r244, %r245 }; + // end inline asm + cvt.rn.bf16x2.f32 %r246, %r1438, %r1437; + cvt.rn.bf16x2.f32 %r247, %r1440, %r1439; + cvt.rn.bf16x2.f32 %r248, %r1442, %r1441; + cvt.rn.bf16x2.f32 %r249, %r1444, %r1443; + // begin inline asm + @%p30 st.global.v4.b32 [ %rd62 + 0 ], { %r246, %r247, %r248, %r249 }; + // end inline asm + cvt.rn.bf16x2.f32 %r250, %r1446, %r1445; + cvt.rn.bf16x2.f32 %r251, %r1448, %r1447; + cvt.rn.bf16x2.f32 %r252, %r1450, %r1449; + cvt.rn.bf16x2.f32 %r253, %r1452, %r1451; + // begin inline asm + @%p31 st.global.v4.b32 [ %rd63 + 0 ], { %r250, %r251, %r252, %r253 }; + // end inline asm + cvt.rn.bf16x2.f32 %r254, %r1454, %r1453; + cvt.rn.bf16x2.f32 %r255, %r1456, %r1455; + cvt.rn.bf16x2.f32 %r256, %r1458, %r1457; + cvt.rn.bf16x2.f32 %r257, %r1460, %r1459; + // begin inline asm + @%p32 st.global.v4.b32 [ %rd64 + 0 ], { %r254, %r255, %r256, %r257 }; + // end inline asm + .loc 1 84 21 // l2norm.py:84:21 + bfe.s32 %r1461, %r302, 4, 1; + and.b32 %r1462, %r1461, 320; + and.b32 %r1463, %r302, 32; + shl.b32 %r1464, %r1463, 1; + xor.b32 %r1465, %r1462, %r1464; + mov.b32 %r1466, global_smem; + add.s32 %r258, %r1466, %r1465; + // begin inline asm + st.shared.v4.b32 [ %r258 + 0 ], { %r259, %r260, %r261, %r262 }; + // end inline asm + add.s32 %r263, %r258, 16; + // begin inline asm + st.shared.v4.b32 [ %r263 + 0 ], { %r264, %r265, %r266, %r267 }; + // end inline asm + add.s32 %r268, %r258, 32; + // begin inline asm + st.shared.v4.b32 [ %r268 + 0 ], { %r269, %r270, %r271, %r272 }; + // end inline asm + add.s32 %r273, %r258, 48; + // begin inline asm + st.shared.v4.b32 [ %r273 + 0 ], { %r274, %r275, %r276, %r277 }; + // end inline asm + add.s32 %r278, %r258, 128; + // begin inline asm + st.shared.v4.b32 [ %r278 + 0 ], { %r279, %r280, %r281, %r282 }; + // end inline asm + add.s32 %r283, %r258, 144; + // begin inline asm + st.shared.v4.b32 [ %r283 + 0 ], { %r284, %r285, %r286, %r287 }; + // end inline asm + add.s32 %r288, %r258, 160; + // begin inline asm + st.shared.v4.b32 [ %r288 + 0 ], { %r289, %r290, %r291, %r292 }; + // end inline asm + add.s32 %r293, %r258, 176; + // begin inline asm + st.shared.v4.b32 [ %r293 + 0 ], { %r294, %r295, %r296, %r297 }; + // end inline asm + bar.sync 0; + and.b32 %r1467, %r1201, 60; + shl.b32 %r1468, %r302, 6; + and.b32 %r1469, %r1468, 64; + shl.b32 %r1470, %r1463, 2; + or.b32 %r1471, %r1469, %r1470; + or.b32 %r1472, %r1471, %r1467; + add.s32 %r1473, %r1466, %r1472; + ld.shared.b32 %r298, [%r1473]; + xor.b32 %r1474, %r1472, 64; + add.s32 %r1475, %r1466, %r1474; + ld.shared.b32 %r299, [%r1475+256]; + // begin inline asm + @%p33 st.global.b32 [ %rd65 + 0 ], { %r298 }; + // end inline asm + // begin inline asm + @%p34 st.global.b32 [ %rd66 + 0 ], { %r299 }; + // end inline asm + .loc 1 84 4 // l2norm.py:84:4 + ret; +$L__tmp259: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp258 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 80 // DW_AT_call_line +.b8 43 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp258 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.source b/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..65999ed34486c21231e2e10dd8e7d7c057966940 --- /dev/null +++ b/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.source @@ -0,0 +1,172 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc38 = loc(unknown) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc45 = loc("x"(#loc)) +#loc46 = loc("y"(#loc)) +#loc47 = loc("rstd"(#loc)) +#loc48 = loc("eps"(#loc)) +#loc49 = loc("T"(#loc)) +#loc81 = loc("input"(#loc36)) +#loc82 = loc("a"(#loc41)) +#loc83 = loc("b"(#loc41)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc50) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc51) + %o_t = arith.constant 128 : i32 loc(#loc52) + %o_t_1 = arith.constant 128 : i64 loc(#loc52) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc52) + %o_t_3 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc53) + %o_t_4 = arith.extsi %o_t_3 : tensor<128xi32> to tensor<128xi64> loc(#loc54) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<128xi64> loc(#loc54) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<128xi64> loc(#loc54) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc55) + %m_t = arith.extsi %T : i32 to i64 loc(#loc56) + %m_t_7 = tt.splat %m_t : i64 -> tensor<128xi64> loc(#loc56) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<128xi64> loc(#loc56) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<128xi1> -> tensor<128x1xi1> loc(#loc57) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc58) + %m_x_10 = arith.constant 128 : i32 loc(#loc59) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc59) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc59) + %m_x_13 = tt.broadcast %m_x : tensor<128x1xi1> -> tensor<128x128xi1> loc(#loc60) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<128x128xi1> loc(#loc60) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<128x128xi1> loc(#loc60) + %p_x = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc61) + %p_x_16 = arith.constant 128 : i32 loc(#loc62) + %p_x_17 = arith.constant 128 : i64 loc(#loc62) + %p_x_18 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc62) + %p_x_19 = arith.muli %p_x, %p_x_18 : tensor<128x1xi64> loc(#loc62) + %p_x_20 = tt.splat %x : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc63) + %p_x_21 = tt.addptr %p_x_20, %p_x_19 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc63) + %p_x_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc64) + %p_x_23 = tt.broadcast %p_x_21 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc65) + %p_x_24 = tt.broadcast %p_x_22 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc65) + %p_x_25 = tt.addptr %p_x_23, %p_x_24 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc65) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc66) + %p_y_26 = arith.constant 128 : i32 loc(#loc67) + %p_y_27 = arith.constant 128 : i64 loc(#loc67) + %p_y_28 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc67) + %p_y_29 = arith.muli %p_y, %p_y_28 : tensor<128x1xi64> loc(#loc67) + %p_y_30 = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc68) + %p_y_31 = tt.addptr %p_y_30, %p_y_29 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc68) + %p_y_32 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc69) + %p_y_33 = tt.broadcast %p_y_31 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc70) + %p_y_34 = tt.broadcast %p_y_32 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc70) + %p_y_35 = tt.addptr %p_y_33, %p_y_34 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc70) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr> loc(#loc71) + %p_rstd_36 = tt.addptr %p_rstd, %o_t_6 : tensor<128x!tt.ptr>, tensor<128xi64> loc(#loc71) + %b_x = arith.constant 0.000000e+00 : f32 loc(#loc72) + %b_x_37 = arith.constant dense<0.000000e+00> : tensor<128x128xf32> loc(#loc72) + %b_x_38 = arith.truncf %b_x_37 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc72) + %b_x_39 = tt.load %p_x_25, %m_x_15, %b_x_38 : tensor<128x128x!tt.ptr> loc(#loc72) + %b_x_40 = arith.extf %b_x_39 : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc73) + %b_rstd = arith.mulf %b_x_40, %b_x_40 : tensor<128x128xf32> loc(#loc74) + %b_rstd_41 = tt.call @triton.language.standard.sum__fp32S128_128S_c1_cFalse_cNone(%b_rstd) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc75) + %b_rstd_42 = tt.splat %eps : f32 -> tensor<128xf32> loc(#loc76) + %b_rstd_43 = arith.addf %b_rstd_41, %b_rstd_42 : tensor<128xf32> loc(#loc76) + %b_rstd_44 = math.sqrt %b_rstd_43 : tensor<128xf32> loc(#loc77) + %b_rstd_45 = arith.constant 1 : i32 loc(#loc78) + %b_rstd_46 = arith.constant 1.000000e+00 : f32 loc(#loc78) + %b_rstd_47 = arith.constant dense<1.000000e+00> : tensor<128xf32> loc(#loc78) + %b_rstd_48 = arith.divf %b_rstd_47, %b_rstd_44 : tensor<128xf32> loc(#loc78) + %b_y = tt.expand_dims %b_rstd_48 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc79) + %b_y_49 = tt.broadcast %b_y : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc80) + %b_y_50 = arith.mulf %b_x_40, %b_y_49 : tensor<128x128xf32> loc(#loc80) + %0 = arith.truncf %b_y_50 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc32) + tt.store %p_y_35, %0, %m_x_15 : tensor<128x128x!tt.ptr> loc(#loc33) + tt.store %p_rstd_36, %b_rstd_48, %m_t_8 : tensor<128x!tt.ptr> loc(#loc34) + tt.return loc(#loc35) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S128_128S_c1_cFalse_cNone(%input: tensor<128x128xf32> loc("input"(#loc36))) -> tensor<128xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc37) + tt.reduce.return %2 : f32 loc(#loc37) + }) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc37) + tt.return %0 : tensor<128xf32> loc(#loc39) + ^bb1: // no predecessors + %1 = ub.poison : tensor<128xf32> loc(#loc40) + tt.return %1 : tensor<128xf32> loc(#loc40) + } loc(#loc36) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc41)), %b: f32 loc("b"(#loc41))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc42) + tt.return %0 : f32 loc(#loc43) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc44) + tt.return %1 : f32 loc(#loc44) + } loc(#loc41) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:18) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:29) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:37) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc50 = loc("i_t"(#loc1)) +#loc51 = loc("i_t"(#loc2)) +#loc52 = loc("o_t"(#loc3)) +#loc53 = loc("o_t"(#loc4)) +#loc54 = loc("o_t"(#loc5)) +#loc55 = loc("o_d"(#loc6)) +#loc56 = loc("m_t"(#loc7)) +#loc57 = loc("m_x"(#loc8)) +#loc58 = loc("m_x"(#loc9)) +#loc59 = loc("m_x"(#loc10)) +#loc60 = loc("m_x"(#loc11)) +#loc61 = loc("p_x"(#loc12)) +#loc62 = loc("p_x"(#loc13)) +#loc63 = loc("p_x"(#loc14)) +#loc64 = loc("p_x"(#loc15)) +#loc65 = loc("p_x"(#loc16)) +#loc66 = loc("p_y"(#loc17)) +#loc67 = loc("p_y"(#loc18)) +#loc68 = loc("p_y"(#loc19)) +#loc69 = loc("p_y"(#loc20)) +#loc70 = loc("p_y"(#loc21)) +#loc71 = loc("p_rstd"(#loc22)) +#loc72 = loc("b_x"(#loc23)) +#loc73 = loc("b_x"(#loc24)) +#loc74 = loc("b_rstd"(#loc25)) +#loc75 = loc("b_rstd"(#loc26)) +#loc76 = loc("b_rstd"(#loc27)) +#loc77 = loc("b_rstd"(#loc28)) +#loc78 = loc("b_rstd"(#loc29)) +#loc79 = loc("b_y"(#loc30)) +#loc80 = loc("b_y"(#loc31)) diff --git a/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.ttgir b/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..bf833fcff3eaf2220658ed40775f48f108ef9417 --- /dev/null +++ b/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.ttgir @@ -0,0 +1,135 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [2, 1], order = [1, 0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [2], threadsPerWarp = [32], warpsPerCTA = [2], order = [0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc1 = loc(unknown) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc34 = loc("x"(#loc)) +#loc35 = loc("y"(#loc)) +#loc36 = loc("rstd"(#loc)) +#loc37 = loc("eps"(#loc)) +#loc38 = loc("T"(#loc)) +#loc59 = loc("b_rstd"(#loc23)) +#loc66 = loc(callsite(#loc1 at #loc59)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 2 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<1.000000e+00> : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<128x128xbf16, #blocked> loc(#loc1) + %c128_i64 = arith.constant 128 : i64 loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<128x1xi64, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc39) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc40) + %o_t = arith.muli %i_t_3, %c128_i64 : i64 loc(#loc41) + %o_t_4 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc42) + %o_t_5 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #blocked1> loc(#loc42) + %o_t_6 = arith.extsi %o_t_4 : tensor<128xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_7 = arith.extsi %o_t_5 : tensor<128xi32, #blocked1> to tensor<128xi64, #blocked1> loc(#loc43) + %o_t_8 = tt.splat %o_t : i64 -> tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_9 = tt.splat %o_t : i64 -> tensor<128xi64, #blocked1> loc(#loc43) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<128xi64, #blocked1> loc(#loc43) + %m_t = arith.extsi %T : i32 to i64 loc(#loc44) + %m_t_12 = tt.splat %m_t : i64 -> tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_13 = tt.splat %m_t : i64 -> tensor<128xi64, #blocked1> loc(#loc44) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<128xi64, #blocked1> loc(#loc44) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<128xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128x1xi1, #blocked> loc(#loc45) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc46) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc46) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_1 : tensor<1x128xi32, #blocked> loc(#loc47) + %m_x_19 = tt.broadcast %m_x : tensor<128x1xi1, #blocked> -> tensor<128x128xi1, #blocked> loc(#loc48) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<128x128xi1, #blocked> loc(#loc48) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<128x128xi1, #blocked> loc(#loc48) + %p_x = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128x1xi64, #blocked> loc(#loc49) + %p_x_22 = arith.muli %p_x, %cst_2 : tensor<128x1xi64, #blocked> loc(#loc50) + %p_x_23 = tt.splat %x : !tt.ptr -> tensor<128x1x!tt.ptr, #blocked> loc(#loc51) + %p_x_24 = tt.addptr %p_x_23, %p_x_22 : tensor<128x1x!tt.ptr, #blocked>, tensor<128x1xi64, #blocked> loc(#loc51) + %p_x_25 = tt.broadcast %p_x_24 : tensor<128x1x!tt.ptr, #blocked> -> tensor<128x128x!tt.ptr, #blocked> loc(#loc52) + %p_x_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<128x128xi32, #blocked> loc(#loc52) + %p_x_27 = tt.addptr %p_x_25, %p_x_26 : tensor<128x128x!tt.ptr, #blocked>, tensor<128x128xi32, #blocked> loc(#loc52) + %p_y = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr, #blocked> loc(#loc53) + %p_y_28 = tt.addptr %p_y, %p_x_22 : tensor<128x1x!tt.ptr, #blocked>, tensor<128x1xi64, #blocked> loc(#loc53) + %p_y_29 = tt.broadcast %p_y_28 : tensor<128x1x!tt.ptr, #blocked> -> tensor<128x128x!tt.ptr, #blocked> loc(#loc54) + %p_y_30 = tt.addptr %p_y_29, %p_x_26 : tensor<128x128x!tt.ptr, #blocked>, tensor<128x128xi32, #blocked> loc(#loc54) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr, #blocked1> loc(#loc55) + %p_rstd_31 = tt.addptr %p_rstd, %o_t_11 : tensor<128x!tt.ptr, #blocked1>, tensor<128xi64, #blocked1> loc(#loc55) + %b_x = tt.load %p_x_27, %m_x_21, %cst_0 : tensor<128x128x!tt.ptr, #blocked> loc(#loc56) + %b_x_32 = arith.extf %b_x : tensor<128x128xbf16, #blocked> to tensor<128x128xf32, #blocked> loc(#loc57) + %b_rstd = arith.mulf %b_x_32, %b_x_32 : tensor<128x128xf32, #blocked> loc(#loc58) + %b_rstd_33 = "tt.reduce"(%b_rstd) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_40: f32 loc(callsite(#loc1 at #loc59)), %b_rstd_41: f32 loc(callsite(#loc1 at #loc59))): + %b_rstd_42 = arith.addf %b_rstd_40, %b_rstd_41 : f32 loc(#loc67) + tt.reduce.return %b_rstd_42 : f32 loc(#loc65) + }) : (tensor<128x128xf32, #blocked>) -> tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc65) + %b_rstd_34 = tt.splat %eps : f32 -> tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_36 = math.sqrt %b_rstd_35 : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc61) + %b_rstd_37 = arith.divf %cst, %b_rstd_36 : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc62) + %b_y = tt.expand_dims %b_rstd_37 {axis = 1 : i32} : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128x1xf32, #blocked> loc(#loc63) + %b_y_38 = tt.broadcast %b_y : tensor<128x1xf32, #blocked> -> tensor<128x128xf32, #blocked> loc(#loc64) + %b_y_39 = arith.mulf %b_x_32, %b_y_38 : tensor<128x128xf32, #blocked> loc(#loc64) + %0 = arith.truncf %b_y_39 : tensor<128x128xf32, #blocked> to tensor<128x128xbf16, #blocked> loc(#loc30) + tt.store %p_y_30, %0, %m_x_21 : tensor<128x128x!tt.ptr, #blocked> loc(#loc31) + %1 = ttg.convert_layout %b_rstd_37 : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128xf32, #blocked1> loc(#loc32) + tt.store %p_rstd_31, %1, %m_t_15 : tensor<128x!tt.ptr, #blocked1> loc(#loc32) + tt.return loc(#loc33) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc39 = loc("i_t"(#loc2)) +#loc40 = loc("i_t"(#loc3)) +#loc41 = loc("o_t"(#loc4)) +#loc42 = loc("o_t"(#loc5)) +#loc43 = loc("o_t"(#loc6)) +#loc44 = loc("m_t"(#loc7)) +#loc45 = loc("m_x"(#loc8)) +#loc46 = loc("m_x"(#loc9)) +#loc47 = loc("m_x"(#loc10)) +#loc48 = loc("m_x"(#loc11)) +#loc49 = loc("p_x"(#loc12)) +#loc50 = loc("p_x"(#loc13)) +#loc51 = loc("p_x"(#loc14)) +#loc52 = loc("p_x"(#loc15)) +#loc53 = loc("p_y"(#loc16)) +#loc54 = loc("p_y"(#loc17)) +#loc55 = loc("p_rstd"(#loc18)) +#loc56 = loc("b_x"(#loc19)) +#loc57 = loc("b_x"(#loc20)) +#loc58 = loc("b_rstd"(#loc21)) +#loc60 = loc("b_rstd"(#loc25)) +#loc61 = loc("b_rstd"(#loc26)) +#loc62 = loc("b_rstd"(#loc27)) +#loc63 = loc("b_y"(#loc28)) +#loc64 = loc("b_y"(#loc29)) +#loc65 = loc(callsite(#loc22 at #loc59)) +#loc67 = loc(callsite(#loc24 at #loc65)) diff --git a/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.ttir b/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..8f0dfe3371851abdc6b1c20ee94003c63d3c42a8 --- /dev/null +++ b/triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.ttir @@ -0,0 +1,125 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc3 = loc(unknown) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc34 = loc("x"(#loc)) +#loc35 = loc("y"(#loc)) +#loc36 = loc("rstd"(#loc)) +#loc37 = loc("eps"(#loc)) +#loc38 = loc("T"(#loc)) +#loc60 = loc("b_rstd"(#loc24)) +#loc66 = loc(callsite(#loc3 at #loc60)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %b_x = arith.constant dense<0.000000e+00> : tensor<128x128xbf16> loc(#loc39) + %b_rstd = arith.constant dense<1.000000e+00> : tensor<128xf32> loc(#loc40) + %cst = arith.constant dense<128> : tensor<128x1xi64> loc(#loc3) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc41) + %c128_i64 = arith.constant 128 : i64 loc(#loc3) + %i_t = tt.get_program_id x : i32 loc(#loc42) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc43) + %o_t = arith.muli %i_t_0, %c128_i64 : i64 loc(#loc44) + %o_t_1 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc45) + %o_t_2 = arith.extsi %o_t_1 : tensor<128xi32> to tensor<128xi64> loc(#loc46) + %o_t_3 = tt.splat %o_t : i64 -> tensor<128xi64> loc(#loc46) + %o_t_4 = arith.addi %o_t_3, %o_t_2 : tensor<128xi64> loc(#loc46) + %m_t = arith.extsi %T : i32 to i64 loc(#loc47) + %m_t_5 = tt.splat %m_t : i64 -> tensor<128xi64> loc(#loc47) + %m_t_6 = arith.cmpi slt, %o_t_4, %m_t_5 : tensor<128xi64> loc(#loc47) + %m_x_7 = tt.expand_dims %m_t_6 {axis = 1 : i32} : tensor<128xi1> -> tensor<128x1xi1> loc(#loc48) + %m_x_8 = tt.expand_dims %o_t_1 {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc49) + %m_x_9 = arith.cmpi slt, %m_x_8, %m_x : tensor<1x128xi32> loc(#loc41) + %m_x_10 = tt.broadcast %m_x_7 : tensor<128x1xi1> -> tensor<128x128xi1> loc(#loc50) + %m_x_11 = tt.broadcast %m_x_9 : tensor<1x128xi1> -> tensor<128x128xi1> loc(#loc50) + %m_x_12 = arith.andi %m_x_10, %m_x_11 : tensor<128x128xi1> loc(#loc50) + %p_x = tt.expand_dims %o_t_4 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc51) + %p_x_13 = arith.muli %p_x, %cst : tensor<128x1xi64> loc(#loc52) + %p_x_14 = tt.splat %x : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc53) + %p_x_15 = tt.addptr %p_x_14, %p_x_13 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc53) + %p_x_16 = tt.broadcast %p_x_15 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc54) + %p_x_17 = tt.broadcast %m_x_8 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc54) + %p_x_18 = tt.addptr %p_x_16, %p_x_17 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc54) + %p_y = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc55) + %p_y_19 = tt.addptr %p_y, %p_x_13 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc55) + %p_y_20 = tt.broadcast %p_y_19 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc56) + %p_y_21 = tt.addptr %p_y_20, %p_x_17 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc56) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr> loc(#loc57) + %p_rstd_22 = tt.addptr %p_rstd, %o_t_4 : tensor<128x!tt.ptr>, tensor<128xi64> loc(#loc57) + %b_x_23 = tt.load %p_x_18, %m_x_12, %b_x : tensor<128x128x!tt.ptr> loc(#loc39) + %b_x_24 = arith.extf %b_x_23 : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc58) + %b_rstd_25 = arith.mulf %b_x_24, %b_x_24 : tensor<128x128xf32> loc(#loc59) + %b_rstd_26 = "tt.reduce"(%b_rstd_25) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_33: f32 loc(callsite(#loc3 at #loc60)), %b_rstd_34: f32 loc(callsite(#loc3 at #loc60))): + %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : f32 loc(#loc67) + tt.reduce.return %b_rstd_35 : f32 loc(#loc65) + }) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc65) + %b_rstd_27 = tt.splat %eps : f32 -> tensor<128xf32> loc(#loc61) + %b_rstd_28 = arith.addf %b_rstd_26, %b_rstd_27 : tensor<128xf32> loc(#loc61) + %b_rstd_29 = math.sqrt %b_rstd_28 : tensor<128xf32> loc(#loc62) + %b_rstd_30 = arith.divf %b_rstd, %b_rstd_29 : tensor<128xf32> loc(#loc40) + %b_y = tt.expand_dims %b_rstd_30 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc63) + %b_y_31 = tt.broadcast %b_y : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc64) + %b_y_32 = arith.mulf %b_x_24, %b_y_31 : tensor<128x128xf32> loc(#loc64) + %0 = arith.truncf %b_y_32 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc30) + tt.store %p_y_21, %0, %m_x_12 : tensor<128x128x!tt.ptr> loc(#loc31) + tt.store %p_rstd_22, %b_rstd_30, %m_t_6 : tensor<128x!tt.ptr> loc(#loc32) + tt.return loc(#loc33) + } loc(#loc) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc39 = loc("b_x"(#loc1)) +#loc40 = loc("b_rstd"(#loc2)) +#loc41 = loc("m_x"(#loc4)) +#loc42 = loc("i_t"(#loc5)) +#loc43 = loc("i_t"(#loc6)) +#loc44 = loc("o_t"(#loc7)) +#loc45 = loc("o_t"(#loc8)) +#loc46 = loc("o_t"(#loc9)) +#loc47 = loc("m_t"(#loc10)) +#loc48 = loc("m_x"(#loc11)) +#loc49 = loc("m_x"(#loc12)) +#loc50 = loc("m_x"(#loc13)) +#loc51 = loc("p_x"(#loc14)) +#loc52 = loc("p_x"(#loc15)) +#loc53 = loc("p_x"(#loc16)) +#loc54 = loc("p_x"(#loc17)) +#loc55 = loc("p_y"(#loc18)) +#loc56 = loc("p_y"(#loc19)) +#loc57 = loc("p_rstd"(#loc20)) +#loc58 = loc("b_x"(#loc21)) +#loc59 = loc("b_rstd"(#loc22)) +#loc61 = loc("b_rstd"(#loc26)) +#loc62 = loc("b_rstd"(#loc27)) +#loc63 = loc("b_y"(#loc28)) +#loc64 = loc("b_y"(#loc29)) +#loc65 = loc(callsite(#loc23 at #loc60)) +#loc67 = loc(callsite(#loc25 at #loc65)) diff --git a/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/__grp__l2norm_fwd_kernel.json b/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/__grp__l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..57f3cdb20b8c6728a867e65d5a7fc1bb8a7f14fa --- /dev/null +++ b/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/__grp__l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_fwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/l2norm_fwd_kernel.source", "l2norm_fwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/l2norm_fwd_kernel.ttir", "l2norm_fwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/l2norm_fwd_kernel.ttgir", "l2norm_fwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/l2norm_fwd_kernel.llir", "l2norm_fwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/l2norm_fwd_kernel.ptx", "l2norm_fwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/l2norm_fwd_kernel.cubin", "l2norm_fwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/l2norm_fwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/l2norm_fwd_kernel.cubin b/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/l2norm_fwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..406d90e7df79f722999ea5b13a6aea3e9b984bb6 Binary files /dev/null and b/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/l2norm_fwd_kernel.cubin differ diff --git a/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/l2norm_fwd_kernel.json b/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..0e7e5acb7ec2e4449cb25446033c9f93fafd5aaf --- /dev/null +++ b/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"hash": "0e36d53a790da9361bcf318feb26c9bb0f270c04764b394376307c54c6b39124", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 2, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 128, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_fwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/l2norm_fwd_kernel.llir b/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/l2norm_fwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..627c0508dc94ce9f7595b0f6ff9a2f82a7e68083 --- /dev/null +++ b/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/l2norm_fwd_kernel.llir @@ -0,0 +1,1143 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 +@.str.2 = private unnamed_addr constant [17 x i8] c"__CUDA_PREC_SQRT\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_fwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, float %3, i32 %4, ptr addrspace(1) readnone captures(none) %5, ptr addrspace(1) readnone captures(none) %6) local_unnamed_addr #0 !dbg !5 { + %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %9 = zext nneg i32 %8 to i64, !dbg !14 + %10 = shl nuw nsw i64 %9, 5, !dbg !15 + %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !16 + %12 = and i32 %11, 48, !dbg !16 + %13 = lshr exact i32 %12, 4, !dbg !16 + %14 = or disjoint i32 %13, 4, !dbg !16 + %15 = or disjoint i32 %13, 8, !dbg !16 + %16 = or disjoint i32 %13, 12, !dbg !16 + %17 = or disjoint i32 %13, 16, !dbg !16 + %18 = or disjoint i32 %13, 20, !dbg !16 + %19 = or disjoint i32 %13, 24, !dbg !16 + %20 = or disjoint i32 %13, 28, !dbg !16 + %21 = zext nneg i32 %13 to i64, !dbg !17 + %22 = zext nneg i32 %14 to i64, !dbg !17 + %23 = zext nneg i32 %15 to i64, !dbg !17 + %24 = zext nneg i32 %16 to i64, !dbg !17 + %25 = zext nneg i32 %17 to i64, !dbg !17 + %26 = zext nneg i32 %18 to i64, !dbg !17 + %27 = zext nneg i32 %19 to i64, !dbg !17 + %28 = zext nneg i32 %20 to i64, !dbg !17 + %29 = or disjoint i64 %10, %21, !dbg !17 + %30 = or disjoint i64 %10, %22, !dbg !17 + %31 = or disjoint i64 %10, %23, !dbg !17 + %32 = or disjoint i64 %10, %24, !dbg !17 + %33 = or disjoint i64 %10, %25, !dbg !17 + %34 = or disjoint i64 %10, %26, !dbg !17 + %35 = or disjoint i64 %10, %27, !dbg !17 + %36 = or disjoint i64 %10, %28, !dbg !17 + %37 = sext i32 %4 to i64, !dbg !18 + %38 = icmp slt i64 %29, %37, !dbg !18 + %39 = icmp slt i64 %30, %37, !dbg !18 + %40 = icmp slt i64 %31, %37, !dbg !18 + %41 = icmp slt i64 %32, %37, !dbg !18 + %42 = icmp slt i64 %33, %37, !dbg !18 + %43 = icmp slt i64 %34, %37, !dbg !18 + %44 = icmp slt i64 %35, %37, !dbg !18 + %45 = icmp slt i64 %36, %37, !dbg !18 + %46 = shl nuw nsw i32 %11, 3, !dbg !19 + %47 = and i32 %46, 120, !dbg !19 + %48 = shl nuw nsw i64 %29, 7, !dbg !20 + %49 = shl nuw nsw i64 %30, 7, !dbg !20 + %50 = shl nuw nsw i64 %31, 7, !dbg !20 + %51 = shl nuw nsw i64 %32, 7, !dbg !20 + %52 = shl nuw nsw i64 %33, 7, !dbg !20 + %53 = shl nuw nsw i64 %34, 7, !dbg !20 + %54 = shl nuw nsw i64 %35, 7, !dbg !20 + %55 = shl nuw nsw i64 %36, 7, !dbg !20 + %56 = getelementptr bfloat, ptr addrspace(1) %0, i64 %48, !dbg !21 + %57 = getelementptr bfloat, ptr addrspace(1) %0, i64 %49, !dbg !21 + %58 = getelementptr bfloat, ptr addrspace(1) %0, i64 %50, !dbg !21 + %59 = getelementptr bfloat, ptr addrspace(1) %0, i64 %51, !dbg !21 + %60 = getelementptr bfloat, ptr addrspace(1) %0, i64 %52, !dbg !21 + %61 = getelementptr bfloat, ptr addrspace(1) %0, i64 %53, !dbg !21 + %62 = getelementptr bfloat, ptr addrspace(1) %0, i64 %54, !dbg !21 + %63 = getelementptr bfloat, ptr addrspace(1) %0, i64 %55, !dbg !21 + %64 = zext nneg i32 %47 to i64, !dbg !22 + %65 = getelementptr bfloat, ptr addrspace(1) %56, i64 %64, !dbg !22 + %66 = getelementptr bfloat, ptr addrspace(1) %57, i64 %64, !dbg !22 + %67 = getelementptr bfloat, ptr addrspace(1) %58, i64 %64, !dbg !22 + %68 = getelementptr bfloat, ptr addrspace(1) %59, i64 %64, !dbg !22 + %69 = getelementptr bfloat, ptr addrspace(1) %60, i64 %64, !dbg !22 + %70 = getelementptr bfloat, ptr addrspace(1) %61, i64 %64, !dbg !22 + %71 = getelementptr bfloat, ptr addrspace(1) %62, i64 %64, !dbg !22 + %72 = getelementptr bfloat, ptr addrspace(1) %63, i64 %64, !dbg !22 + %73 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %65, i1 %38) #6, !dbg !23 + %74 = extractvalue { i32, i32, i32, i32 } %73, 0, !dbg !23 + %75 = bitcast i32 %74 to <2 x bfloat>, !dbg !23 + %76 = extractvalue { i32, i32, i32, i32 } %73, 1, !dbg !23 + %77 = bitcast i32 %76 to <2 x bfloat>, !dbg !23 + %78 = extractvalue { i32, i32, i32, i32 } %73, 2, !dbg !23 + %79 = bitcast i32 %78 to <2 x bfloat>, !dbg !23 + %80 = extractvalue { i32, i32, i32, i32 } %73, 3, !dbg !23 + %81 = bitcast i32 %80 to <2 x bfloat>, !dbg !23 + %82 = extractelement <2 x bfloat> %75, i64 0, !dbg !23 + %83 = extractelement <2 x bfloat> %75, i64 1, !dbg !23 + %84 = extractelement <2 x bfloat> %77, i64 0, !dbg !23 + %85 = extractelement <2 x bfloat> %77, i64 1, !dbg !23 + %86 = extractelement <2 x bfloat> %79, i64 0, !dbg !23 + %87 = extractelement <2 x bfloat> %79, i64 1, !dbg !23 + %88 = extractelement <2 x bfloat> %81, i64 0, !dbg !23 + %89 = extractelement <2 x bfloat> %81, i64 1, !dbg !23 + %90 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %66, i1 %39) #6, !dbg !23 + %91 = extractvalue { i32, i32, i32, i32 } %90, 0, !dbg !23 + %92 = bitcast i32 %91 to <2 x bfloat>, !dbg !23 + %93 = extractvalue { i32, i32, i32, i32 } %90, 1, !dbg !23 + %94 = bitcast i32 %93 to <2 x bfloat>, !dbg !23 + %95 = extractvalue { i32, i32, i32, i32 } %90, 2, !dbg !23 + %96 = bitcast i32 %95 to <2 x bfloat>, !dbg !23 + %97 = extractvalue { i32, i32, i32, i32 } %90, 3, !dbg !23 + %98 = bitcast i32 %97 to <2 x bfloat>, !dbg !23 + %99 = extractelement <2 x bfloat> %92, i64 0, !dbg !23 + %100 = extractelement <2 x bfloat> %92, i64 1, !dbg !23 + %101 = extractelement <2 x bfloat> %94, i64 0, !dbg !23 + %102 = extractelement <2 x bfloat> %94, i64 1, !dbg !23 + %103 = extractelement <2 x bfloat> %96, i64 0, !dbg !23 + %104 = extractelement <2 x bfloat> %96, i64 1, !dbg !23 + %105 = extractelement <2 x bfloat> %98, i64 0, !dbg !23 + %106 = extractelement <2 x bfloat> %98, i64 1, !dbg !23 + %107 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %67, i1 %40) #6, !dbg !23 + %108 = extractvalue { i32, i32, i32, i32 } %107, 0, !dbg !23 + %109 = bitcast i32 %108 to <2 x bfloat>, !dbg !23 + %110 = extractvalue { i32, i32, i32, i32 } %107, 1, !dbg !23 + %111 = bitcast i32 %110 to <2 x bfloat>, !dbg !23 + %112 = extractvalue { i32, i32, i32, i32 } %107, 2, !dbg !23 + %113 = bitcast i32 %112 to <2 x bfloat>, !dbg !23 + %114 = extractvalue { i32, i32, i32, i32 } %107, 3, !dbg !23 + %115 = bitcast i32 %114 to <2 x bfloat>, !dbg !23 + %116 = extractelement <2 x bfloat> %109, i64 0, !dbg !23 + %117 = extractelement <2 x bfloat> %109, i64 1, !dbg !23 + %118 = extractelement <2 x bfloat> %111, i64 0, !dbg !23 + %119 = extractelement <2 x bfloat> %111, i64 1, !dbg !23 + %120 = extractelement <2 x bfloat> %113, i64 0, !dbg !23 + %121 = extractelement <2 x bfloat> %113, i64 1, !dbg !23 + %122 = extractelement <2 x bfloat> %115, i64 0, !dbg !23 + %123 = extractelement <2 x bfloat> %115, i64 1, !dbg !23 + %124 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %68, i1 %41) #6, !dbg !23 + %125 = extractvalue { i32, i32, i32, i32 } %124, 0, !dbg !23 + %126 = bitcast i32 %125 to <2 x bfloat>, !dbg !23 + %127 = extractvalue { i32, i32, i32, i32 } %124, 1, !dbg !23 + %128 = bitcast i32 %127 to <2 x bfloat>, !dbg !23 + %129 = extractvalue { i32, i32, i32, i32 } %124, 2, !dbg !23 + %130 = bitcast i32 %129 to <2 x bfloat>, !dbg !23 + %131 = extractvalue { i32, i32, i32, i32 } %124, 3, !dbg !23 + %132 = bitcast i32 %131 to <2 x bfloat>, !dbg !23 + %133 = extractelement <2 x bfloat> %126, i64 0, !dbg !23 + %134 = extractelement <2 x bfloat> %126, i64 1, !dbg !23 + %135 = extractelement <2 x bfloat> %128, i64 0, !dbg !23 + %136 = extractelement <2 x bfloat> %128, i64 1, !dbg !23 + %137 = extractelement <2 x bfloat> %130, i64 0, !dbg !23 + %138 = extractelement <2 x bfloat> %130, i64 1, !dbg !23 + %139 = extractelement <2 x bfloat> %132, i64 0, !dbg !23 + %140 = extractelement <2 x bfloat> %132, i64 1, !dbg !23 + %141 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %69, i1 %42) #6, !dbg !23 + %142 = extractvalue { i32, i32, i32, i32 } %141, 0, !dbg !23 + %143 = bitcast i32 %142 to <2 x bfloat>, !dbg !23 + %144 = extractvalue { i32, i32, i32, i32 } %141, 1, !dbg !23 + %145 = bitcast i32 %144 to <2 x bfloat>, !dbg !23 + %146 = extractvalue { i32, i32, i32, i32 } %141, 2, !dbg !23 + %147 = bitcast i32 %146 to <2 x bfloat>, !dbg !23 + %148 = extractvalue { i32, i32, i32, i32 } %141, 3, !dbg !23 + %149 = bitcast i32 %148 to <2 x bfloat>, !dbg !23 + %150 = extractelement <2 x bfloat> %143, i64 0, !dbg !23 + %151 = extractelement <2 x bfloat> %143, i64 1, !dbg !23 + %152 = extractelement <2 x bfloat> %145, i64 0, !dbg !23 + %153 = extractelement <2 x bfloat> %145, i64 1, !dbg !23 + %154 = extractelement <2 x bfloat> %147, i64 0, !dbg !23 + %155 = extractelement <2 x bfloat> %147, i64 1, !dbg !23 + %156 = extractelement <2 x bfloat> %149, i64 0, !dbg !23 + %157 = extractelement <2 x bfloat> %149, i64 1, !dbg !23 + %158 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %70, i1 %43) #6, !dbg !23 + %159 = extractvalue { i32, i32, i32, i32 } %158, 0, !dbg !23 + %160 = bitcast i32 %159 to <2 x bfloat>, !dbg !23 + %161 = extractvalue { i32, i32, i32, i32 } %158, 1, !dbg !23 + %162 = bitcast i32 %161 to <2 x bfloat>, !dbg !23 + %163 = extractvalue { i32, i32, i32, i32 } %158, 2, !dbg !23 + %164 = bitcast i32 %163 to <2 x bfloat>, !dbg !23 + %165 = extractvalue { i32, i32, i32, i32 } %158, 3, !dbg !23 + %166 = bitcast i32 %165 to <2 x bfloat>, !dbg !23 + %167 = extractelement <2 x bfloat> %160, i64 0, !dbg !23 + %168 = extractelement <2 x bfloat> %160, i64 1, !dbg !23 + %169 = extractelement <2 x bfloat> %162, i64 0, !dbg !23 + %170 = extractelement <2 x bfloat> %162, i64 1, !dbg !23 + %171 = extractelement <2 x bfloat> %164, i64 0, !dbg !23 + %172 = extractelement <2 x bfloat> %164, i64 1, !dbg !23 + %173 = extractelement <2 x bfloat> %166, i64 0, !dbg !23 + %174 = extractelement <2 x bfloat> %166, i64 1, !dbg !23 + %175 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %71, i1 %44) #6, !dbg !23 + %176 = extractvalue { i32, i32, i32, i32 } %175, 0, !dbg !23 + %177 = bitcast i32 %176 to <2 x bfloat>, !dbg !23 + %178 = extractvalue { i32, i32, i32, i32 } %175, 1, !dbg !23 + %179 = bitcast i32 %178 to <2 x bfloat>, !dbg !23 + %180 = extractvalue { i32, i32, i32, i32 } %175, 2, !dbg !23 + %181 = bitcast i32 %180 to <2 x bfloat>, !dbg !23 + %182 = extractvalue { i32, i32, i32, i32 } %175, 3, !dbg !23 + %183 = bitcast i32 %182 to <2 x bfloat>, !dbg !23 + %184 = extractelement <2 x bfloat> %177, i64 0, !dbg !23 + %185 = extractelement <2 x bfloat> %177, i64 1, !dbg !23 + %186 = extractelement <2 x bfloat> %179, i64 0, !dbg !23 + %187 = extractelement <2 x bfloat> %179, i64 1, !dbg !23 + %188 = extractelement <2 x bfloat> %181, i64 0, !dbg !23 + %189 = extractelement <2 x bfloat> %181, i64 1, !dbg !23 + %190 = extractelement <2 x bfloat> %183, i64 0, !dbg !23 + %191 = extractelement <2 x bfloat> %183, i64 1, !dbg !23 + %192 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %72, i1 %45) #6, !dbg !23 + %193 = extractvalue { i32, i32, i32, i32 } %192, 0, !dbg !23 + %194 = bitcast i32 %193 to <2 x bfloat>, !dbg !23 + %195 = extractvalue { i32, i32, i32, i32 } %192, 1, !dbg !23 + %196 = bitcast i32 %195 to <2 x bfloat>, !dbg !23 + %197 = extractvalue { i32, i32, i32, i32 } %192, 2, !dbg !23 + %198 = bitcast i32 %197 to <2 x bfloat>, !dbg !23 + %199 = extractvalue { i32, i32, i32, i32 } %192, 3, !dbg !23 + %200 = bitcast i32 %199 to <2 x bfloat>, !dbg !23 + %201 = extractelement <2 x bfloat> %194, i64 0, !dbg !23 + %202 = extractelement <2 x bfloat> %194, i64 1, !dbg !23 + %203 = extractelement <2 x bfloat> %196, i64 0, !dbg !23 + %204 = extractelement <2 x bfloat> %196, i64 1, !dbg !23 + %205 = extractelement <2 x bfloat> %198, i64 0, !dbg !23 + %206 = extractelement <2 x bfloat> %198, i64 1, !dbg !23 + %207 = extractelement <2 x bfloat> %200, i64 0, !dbg !23 + %208 = extractelement <2 x bfloat> %200, i64 1, !dbg !23 + %209 = fpext bfloat %82 to float, !dbg !24 + %210 = fpext bfloat %83 to float, !dbg !24 + %211 = fpext bfloat %84 to float, !dbg !24 + %212 = fpext bfloat %85 to float, !dbg !24 + %213 = fpext bfloat %86 to float, !dbg !24 + %214 = fpext bfloat %87 to float, !dbg !24 + %215 = fpext bfloat %88 to float, !dbg !24 + %216 = fpext bfloat %89 to float, !dbg !24 + %217 = fpext bfloat %99 to float, !dbg !24 + %218 = fpext bfloat %100 to float, !dbg !24 + %219 = fpext bfloat %101 to float, !dbg !24 + %220 = fpext bfloat %102 to float, !dbg !24 + %221 = fpext bfloat %103 to float, !dbg !24 + %222 = fpext bfloat %104 to float, !dbg !24 + %223 = fpext bfloat %105 to float, !dbg !24 + %224 = fpext bfloat %106 to float, !dbg !24 + %225 = fpext bfloat %116 to float, !dbg !24 + %226 = fpext bfloat %117 to float, !dbg !24 + %227 = fpext bfloat %118 to float, !dbg !24 + %228 = fpext bfloat %119 to float, !dbg !24 + %229 = fpext bfloat %120 to float, !dbg !24 + %230 = fpext bfloat %121 to float, !dbg !24 + %231 = fpext bfloat %122 to float, !dbg !24 + %232 = fpext bfloat %123 to float, !dbg !24 + %233 = fpext bfloat %133 to float, !dbg !24 + %234 = fpext bfloat %134 to float, !dbg !24 + %235 = fpext bfloat %135 to float, !dbg !24 + %236 = fpext bfloat %136 to float, !dbg !24 + %237 = fpext bfloat %137 to float, !dbg !24 + %238 = fpext bfloat %138 to float, !dbg !24 + %239 = fpext bfloat %139 to float, !dbg !24 + %240 = fpext bfloat %140 to float, !dbg !24 + %241 = fpext bfloat %150 to float, !dbg !24 + %242 = fpext bfloat %151 to float, !dbg !24 + %243 = fpext bfloat %152 to float, !dbg !24 + %244 = fpext bfloat %153 to float, !dbg !24 + %245 = fpext bfloat %154 to float, !dbg !24 + %246 = fpext bfloat %155 to float, !dbg !24 + %247 = fpext bfloat %156 to float, !dbg !24 + %248 = fpext bfloat %157 to float, !dbg !24 + %249 = fpext bfloat %167 to float, !dbg !24 + %250 = fpext bfloat %168 to float, !dbg !24 + %251 = fpext bfloat %169 to float, !dbg !24 + %252 = fpext bfloat %170 to float, !dbg !24 + %253 = fpext bfloat %171 to float, !dbg !24 + %254 = fpext bfloat %172 to float, !dbg !24 + %255 = fpext bfloat %173 to float, !dbg !24 + %256 = fpext bfloat %174 to float, !dbg !24 + %257 = fpext bfloat %184 to float, !dbg !24 + %258 = fpext bfloat %185 to float, !dbg !24 + %259 = fpext bfloat %186 to float, !dbg !24 + %260 = fpext bfloat %187 to float, !dbg !24 + %261 = fpext bfloat %188 to float, !dbg !24 + %262 = fpext bfloat %189 to float, !dbg !24 + %263 = fpext bfloat %190 to float, !dbg !24 + %264 = fpext bfloat %191 to float, !dbg !24 + %265 = fpext bfloat %201 to float, !dbg !24 + %266 = fpext bfloat %202 to float, !dbg !24 + %267 = fpext bfloat %203 to float, !dbg !24 + %268 = fpext bfloat %204 to float, !dbg !24 + %269 = fpext bfloat %205 to float, !dbg !24 + %270 = fpext bfloat %206 to float, !dbg !24 + %271 = fpext bfloat %207 to float, !dbg !24 + %272 = fpext bfloat %208 to float, !dbg !24 + %273 = fmul float %209, %209, !dbg !25 + %274 = fmul float %210, %210, !dbg !25 + %275 = fmul float %211, %211, !dbg !25 + %276 = fmul float %212, %212, !dbg !25 + %277 = fmul float %213, %213, !dbg !25 + %278 = fmul float %214, %214, !dbg !25 + %279 = fmul float %215, %215, !dbg !25 + %280 = fmul float %216, %216, !dbg !25 + %281 = fmul float %217, %217, !dbg !25 + %282 = fmul float %218, %218, !dbg !25 + %283 = fmul float %219, %219, !dbg !25 + %284 = fmul float %220, %220, !dbg !25 + %285 = fmul float %221, %221, !dbg !25 + %286 = fmul float %222, %222, !dbg !25 + %287 = fmul float %223, %223, !dbg !25 + %288 = fmul float %224, %224, !dbg !25 + %289 = fmul float %225, %225, !dbg !25 + %290 = fmul float %226, %226, !dbg !25 + %291 = fmul float %227, %227, !dbg !25 + %292 = fmul float %228, %228, !dbg !25 + %293 = fmul float %229, %229, !dbg !25 + %294 = fmul float %230, %230, !dbg !25 + %295 = fmul float %231, %231, !dbg !25 + %296 = fmul float %232, %232, !dbg !25 + %297 = fmul float %233, %233, !dbg !25 + %298 = fmul float %234, %234, !dbg !25 + %299 = fmul float %235, %235, !dbg !25 + %300 = fmul float %236, %236, !dbg !25 + %301 = fmul float %237, %237, !dbg !25 + %302 = fmul float %238, %238, !dbg !25 + %303 = fmul float %239, %239, !dbg !25 + %304 = fmul float %240, %240, !dbg !25 + %305 = fmul float %241, %241, !dbg !25 + %306 = fmul float %242, %242, !dbg !25 + %307 = fmul float %243, %243, !dbg !25 + %308 = fmul float %244, %244, !dbg !25 + %309 = fmul float %245, %245, !dbg !25 + %310 = fmul float %246, %246, !dbg !25 + %311 = fmul float %247, %247, !dbg !25 + %312 = fmul float %248, %248, !dbg !25 + %313 = fmul float %249, %249, !dbg !25 + %314 = fmul float %250, %250, !dbg !25 + %315 = fmul float %251, %251, !dbg !25 + %316 = fmul float %252, %252, !dbg !25 + %317 = fmul float %253, %253, !dbg !25 + %318 = fmul float %254, %254, !dbg !25 + %319 = fmul float %255, %255, !dbg !25 + %320 = fmul float %256, %256, !dbg !25 + %321 = fmul float %257, %257, !dbg !25 + %322 = fmul float %258, %258, !dbg !25 + %323 = fmul float %259, %259, !dbg !25 + %324 = fmul float %260, %260, !dbg !25 + %325 = fmul float %261, %261, !dbg !25 + %326 = fmul float %262, %262, !dbg !25 + %327 = fmul float %263, %263, !dbg !25 + %328 = fmul float %264, %264, !dbg !25 + %329 = fmul float %265, %265, !dbg !25 + %330 = fmul float %266, %266, !dbg !25 + %331 = fmul float %267, %267, !dbg !25 + %332 = fmul float %268, %268, !dbg !25 + %333 = fmul float %269, %269, !dbg !25 + %334 = fmul float %270, %270, !dbg !25 + %335 = fmul float %271, %271, !dbg !25 + %336 = fmul float %272, %272, !dbg !25 + %337 = fadd float %273, %274, !dbg !26 + %338 = fadd float %275, %337, !dbg !26 + %339 = fadd float %276, %338, !dbg !26 + %340 = fadd float %277, %339, !dbg !26 + %341 = fadd float %278, %340, !dbg !26 + %342 = fadd float %279, %341, !dbg !26 + %343 = fadd float %280, %342, !dbg !26 + %344 = fadd float %281, %282, !dbg !26 + %345 = fadd float %283, %344, !dbg !26 + %346 = fadd float %284, %345, !dbg !26 + %347 = fadd float %285, %346, !dbg !26 + %348 = fadd float %286, %347, !dbg !26 + %349 = fadd float %287, %348, !dbg !26 + %350 = fadd float %288, %349, !dbg !26 + %351 = fadd float %289, %290, !dbg !26 + %352 = fadd float %291, %351, !dbg !26 + %353 = fadd float %292, %352, !dbg !26 + %354 = fadd float %293, %353, !dbg !26 + %355 = fadd float %294, %354, !dbg !26 + %356 = fadd float %295, %355, !dbg !26 + %357 = fadd float %296, %356, !dbg !26 + %358 = fadd float %297, %298, !dbg !26 + %359 = fadd float %299, %358, !dbg !26 + %360 = fadd float %300, %359, !dbg !26 + %361 = fadd float %301, %360, !dbg !26 + %362 = fadd float %302, %361, !dbg !26 + %363 = fadd float %303, %362, !dbg !26 + %364 = fadd float %304, %363, !dbg !26 + %365 = fadd float %305, %306, !dbg !26 + %366 = fadd float %307, %365, !dbg !26 + %367 = fadd float %308, %366, !dbg !26 + %368 = fadd float %309, %367, !dbg !26 + %369 = fadd float %310, %368, !dbg !26 + %370 = fadd float %311, %369, !dbg !26 + %371 = fadd float %312, %370, !dbg !26 + %372 = fadd float %313, %314, !dbg !26 + %373 = fadd float %315, %372, !dbg !26 + %374 = fadd float %316, %373, !dbg !26 + %375 = fadd float %317, %374, !dbg !26 + %376 = fadd float %318, %375, !dbg !26 + %377 = fadd float %319, %376, !dbg !26 + %378 = fadd float %320, %377, !dbg !26 + %379 = fadd float %321, %322, !dbg !26 + %380 = fadd float %323, %379, !dbg !26 + %381 = fadd float %324, %380, !dbg !26 + %382 = fadd float %325, %381, !dbg !26 + %383 = fadd float %326, %382, !dbg !26 + %384 = fadd float %327, %383, !dbg !26 + %385 = fadd float %328, %384, !dbg !26 + %386 = fadd float %329, %330, !dbg !26 + %387 = fadd float %331, %386, !dbg !26 + %388 = fadd float %332, %387, !dbg !26 + %389 = fadd float %333, %388, !dbg !26 + %390 = fadd float %334, %389, !dbg !26 + %391 = fadd float %335, %390, !dbg !26 + %392 = fadd float %336, %391, !dbg !26 + %393 = bitcast float %343 to i32, !dbg !29 + %394 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %393, i32 8, i32 31), !dbg !29 + %395 = bitcast i32 %394 to float, !dbg !29 + %396 = fadd float %343, %395, !dbg !26 + %397 = bitcast float %396 to i32, !dbg !29 + %398 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %397, i32 4, i32 31), !dbg !29 + %399 = bitcast i32 %398 to float, !dbg !29 + %400 = fadd float %396, %399, !dbg !26 + %401 = bitcast float %400 to i32, !dbg !29 + %402 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %401, i32 2, i32 31), !dbg !29 + %403 = bitcast i32 %402 to float, !dbg !29 + %404 = fadd float %400, %403, !dbg !26 + %405 = bitcast float %404 to i32, !dbg !29 + %406 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %405, i32 1, i32 31), !dbg !29 + %407 = bitcast i32 %406 to float, !dbg !29 + %408 = fadd float %404, %407, !dbg !26 + %409 = bitcast float %350 to i32, !dbg !29 + %410 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %409, i32 8, i32 31), !dbg !29 + %411 = bitcast i32 %410 to float, !dbg !29 + %412 = fadd float %350, %411, !dbg !26 + %413 = bitcast float %412 to i32, !dbg !29 + %414 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %413, i32 4, i32 31), !dbg !29 + %415 = bitcast i32 %414 to float, !dbg !29 + %416 = fadd float %412, %415, !dbg !26 + %417 = bitcast float %416 to i32, !dbg !29 + %418 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %417, i32 2, i32 31), !dbg !29 + %419 = bitcast i32 %418 to float, !dbg !29 + %420 = fadd float %416, %419, !dbg !26 + %421 = bitcast float %420 to i32, !dbg !29 + %422 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %421, i32 1, i32 31), !dbg !29 + %423 = bitcast i32 %422 to float, !dbg !29 + %424 = fadd float %420, %423, !dbg !26 + %425 = bitcast float %357 to i32, !dbg !29 + %426 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %425, i32 8, i32 31), !dbg !29 + %427 = bitcast i32 %426 to float, !dbg !29 + %428 = fadd float %357, %427, !dbg !26 + %429 = bitcast float %428 to i32, !dbg !29 + %430 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %429, i32 4, i32 31), !dbg !29 + %431 = bitcast i32 %430 to float, !dbg !29 + %432 = fadd float %428, %431, !dbg !26 + %433 = bitcast float %432 to i32, !dbg !29 + %434 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %433, i32 2, i32 31), !dbg !29 + %435 = bitcast i32 %434 to float, !dbg !29 + %436 = fadd float %432, %435, !dbg !26 + %437 = bitcast float %436 to i32, !dbg !29 + %438 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %437, i32 1, i32 31), !dbg !29 + %439 = bitcast i32 %438 to float, !dbg !29 + %440 = fadd float %436, %439, !dbg !26 + %441 = bitcast float %364 to i32, !dbg !29 + %442 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %441, i32 8, i32 31), !dbg !29 + %443 = bitcast i32 %442 to float, !dbg !29 + %444 = fadd float %364, %443, !dbg !26 + %445 = bitcast float %444 to i32, !dbg !29 + %446 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %445, i32 4, i32 31), !dbg !29 + %447 = bitcast i32 %446 to float, !dbg !29 + %448 = fadd float %444, %447, !dbg !26 + %449 = bitcast float %448 to i32, !dbg !29 + %450 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %449, i32 2, i32 31), !dbg !29 + %451 = bitcast i32 %450 to float, !dbg !29 + %452 = fadd float %448, %451, !dbg !26 + %453 = bitcast float %452 to i32, !dbg !29 + %454 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %453, i32 1, i32 31), !dbg !29 + %455 = bitcast i32 %454 to float, !dbg !29 + %456 = fadd float %452, %455, !dbg !26 + %457 = bitcast float %371 to i32, !dbg !29 + %458 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %457, i32 8, i32 31), !dbg !29 + %459 = bitcast i32 %458 to float, !dbg !29 + %460 = fadd float %371, %459, !dbg !26 + %461 = bitcast float %460 to i32, !dbg !29 + %462 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %461, i32 4, i32 31), !dbg !29 + %463 = bitcast i32 %462 to float, !dbg !29 + %464 = fadd float %460, %463, !dbg !26 + %465 = bitcast float %464 to i32, !dbg !29 + %466 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %465, i32 2, i32 31), !dbg !29 + %467 = bitcast i32 %466 to float, !dbg !29 + %468 = fadd float %464, %467, !dbg !26 + %469 = bitcast float %468 to i32, !dbg !29 + %470 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %469, i32 1, i32 31), !dbg !29 + %471 = bitcast i32 %470 to float, !dbg !29 + %472 = fadd float %468, %471, !dbg !26 + %473 = bitcast float %378 to i32, !dbg !29 + %474 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %473, i32 8, i32 31), !dbg !29 + %475 = bitcast i32 %474 to float, !dbg !29 + %476 = fadd float %378, %475, !dbg !26 + %477 = bitcast float %476 to i32, !dbg !29 + %478 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %477, i32 4, i32 31), !dbg !29 + %479 = bitcast i32 %478 to float, !dbg !29 + %480 = fadd float %476, %479, !dbg !26 + %481 = bitcast float %480 to i32, !dbg !29 + %482 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %481, i32 2, i32 31), !dbg !29 + %483 = bitcast i32 %482 to float, !dbg !29 + %484 = fadd float %480, %483, !dbg !26 + %485 = bitcast float %484 to i32, !dbg !29 + %486 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %485, i32 1, i32 31), !dbg !29 + %487 = bitcast i32 %486 to float, !dbg !29 + %488 = fadd float %484, %487, !dbg !26 + %489 = bitcast float %385 to i32, !dbg !29 + %490 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %489, i32 8, i32 31), !dbg !29 + %491 = bitcast i32 %490 to float, !dbg !29 + %492 = fadd float %385, %491, !dbg !26 + %493 = bitcast float %492 to i32, !dbg !29 + %494 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %493, i32 4, i32 31), !dbg !29 + %495 = bitcast i32 %494 to float, !dbg !29 + %496 = fadd float %492, %495, !dbg !26 + %497 = bitcast float %496 to i32, !dbg !29 + %498 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %497, i32 2, i32 31), !dbg !29 + %499 = bitcast i32 %498 to float, !dbg !29 + %500 = fadd float %496, %499, !dbg !26 + %501 = bitcast float %500 to i32, !dbg !29 + %502 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %501, i32 1, i32 31), !dbg !29 + %503 = bitcast i32 %502 to float, !dbg !29 + %504 = fadd float %500, %503, !dbg !26 + %505 = bitcast float %392 to i32, !dbg !29 + %506 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %505, i32 8, i32 31), !dbg !29 + %507 = bitcast i32 %506 to float, !dbg !29 + %508 = fadd float %392, %507, !dbg !26 + %509 = bitcast float %508 to i32, !dbg !29 + %510 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %509, i32 4, i32 31), !dbg !29 + %511 = bitcast i32 %510 to float, !dbg !29 + %512 = fadd float %508, %511, !dbg !26 + %513 = bitcast float %512 to i32, !dbg !29 + %514 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %513, i32 2, i32 31), !dbg !29 + %515 = bitcast i32 %514 to float, !dbg !29 + %516 = fadd float %512, %515, !dbg !26 + %517 = bitcast float %516 to i32, !dbg !29 + %518 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %517, i32 1, i32 31), !dbg !29 + %519 = bitcast i32 %518 to float, !dbg !29 + %520 = fadd float %516, %519, !dbg !26 + %521 = fadd float %3, %408, !dbg !32 + %522 = fadd float %3, %424, !dbg !32 + %523 = fadd float %3, %440, !dbg !32 + %524 = fadd float %3, %456, !dbg !32 + %525 = fadd float %3, %472, !dbg !32 + %526 = fadd float %3, %488, !dbg !32 + %527 = fadd float %3, %504, !dbg !32 + %528 = fadd float %3, %520, !dbg !32 + %529 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i = icmp eq i32 %529, 0, !dbg !33 + %530 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i = icmp eq i32 %530, 0, !dbg !33 + br i1 %.not.i, label %536, label %531, !dbg !33 + +531: ; preds = %7 + br i1 %.not1.i, label %534, label %532, !dbg !33 + +532: ; preds = %531 + %533 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %521) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +534: ; preds = %531 + %535 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %521) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +536: ; preds = %7 + br i1 %.not1.i, label %539, label %537, !dbg !33 + +537: ; preds = %536 + %538 = tail call float @llvm.nvvm.sqrt.rn.f(float %521) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +539: ; preds = %536 + %540 = tail call float @llvm.nvvm.sqrt.approx.f(float %521) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +__nv_sqrtf.exit: ; preds = %532, %534, %537, %539 + %.0.i = phi float [ %533, %532 ], [ %535, %534 ], [ %538, %537 ], [ %540, %539 ], !dbg !33 + %541 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i1 = icmp eq i32 %541, 0, !dbg !33 + %542 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i4 = icmp eq i32 %542, 0, !dbg !33 + br i1 %.not.i1, label %548, label %543, !dbg !33 + +543: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %546, label %544, !dbg !33 + +544: ; preds = %543 + %545 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %522) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +546: ; preds = %543 + %547 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %522) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +548: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %551, label %549, !dbg !33 + +549: ; preds = %548 + %550 = tail call float @llvm.nvvm.sqrt.rn.f(float %522) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +551: ; preds = %548 + %552 = tail call float @llvm.nvvm.sqrt.approx.f(float %522) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +__nv_sqrtf.exit5: ; preds = %544, %546, %549, %551 + %.0.i3 = phi float [ %545, %544 ], [ %547, %546 ], [ %550, %549 ], [ %552, %551 ], !dbg !33 + %553 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i6 = icmp eq i32 %553, 0, !dbg !33 + %554 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i9 = icmp eq i32 %554, 0, !dbg !33 + br i1 %.not.i6, label %560, label %555, !dbg !33 + +555: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %558, label %556, !dbg !33 + +556: ; preds = %555 + %557 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %523) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +558: ; preds = %555 + %559 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %523) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +560: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %563, label %561, !dbg !33 + +561: ; preds = %560 + %562 = tail call float @llvm.nvvm.sqrt.rn.f(float %523) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +563: ; preds = %560 + %564 = tail call float @llvm.nvvm.sqrt.approx.f(float %523) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +__nv_sqrtf.exit10: ; preds = %556, %558, %561, %563 + %.0.i8 = phi float [ %557, %556 ], [ %559, %558 ], [ %562, %561 ], [ %564, %563 ], !dbg !33 + %565 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i11 = icmp eq i32 %565, 0, !dbg !33 + %566 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i14 = icmp eq i32 %566, 0, !dbg !33 + br i1 %.not.i11, label %572, label %567, !dbg !33 + +567: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %570, label %568, !dbg !33 + +568: ; preds = %567 + %569 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %524) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +570: ; preds = %567 + %571 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %524) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +572: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %575, label %573, !dbg !33 + +573: ; preds = %572 + %574 = tail call float @llvm.nvvm.sqrt.rn.f(float %524) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +575: ; preds = %572 + %576 = tail call float @llvm.nvvm.sqrt.approx.f(float %524) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +__nv_sqrtf.exit15: ; preds = %568, %570, %573, %575 + %.0.i13 = phi float [ %569, %568 ], [ %571, %570 ], [ %574, %573 ], [ %576, %575 ], !dbg !33 + %577 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i16 = icmp eq i32 %577, 0, !dbg !33 + %578 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i19 = icmp eq i32 %578, 0, !dbg !33 + br i1 %.not.i16, label %584, label %579, !dbg !33 + +579: ; preds = %__nv_sqrtf.exit15 + br i1 %.not1.i19, label %582, label %580, !dbg !33 + +580: ; preds = %579 + %581 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %525) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +582: ; preds = %579 + %583 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %525) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +584: ; preds = %__nv_sqrtf.exit15 + br i1 %.not1.i19, label %587, label %585, !dbg !33 + +585: ; preds = %584 + %586 = tail call float @llvm.nvvm.sqrt.rn.f(float %525) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +587: ; preds = %584 + %588 = tail call float @llvm.nvvm.sqrt.approx.f(float %525) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +__nv_sqrtf.exit20: ; preds = %580, %582, %585, %587 + %.0.i18 = phi float [ %581, %580 ], [ %583, %582 ], [ %586, %585 ], [ %588, %587 ], !dbg !33 + %589 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i21 = icmp eq i32 %589, 0, !dbg !33 + %590 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i24 = icmp eq i32 %590, 0, !dbg !33 + br i1 %.not.i21, label %596, label %591, !dbg !33 + +591: ; preds = %__nv_sqrtf.exit20 + br i1 %.not1.i24, label %594, label %592, !dbg !33 + +592: ; preds = %591 + %593 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %526) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +594: ; preds = %591 + %595 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %526) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +596: ; preds = %__nv_sqrtf.exit20 + br i1 %.not1.i24, label %599, label %597, !dbg !33 + +597: ; preds = %596 + %598 = tail call float @llvm.nvvm.sqrt.rn.f(float %526) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +599: ; preds = %596 + %600 = tail call float @llvm.nvvm.sqrt.approx.f(float %526) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +__nv_sqrtf.exit25: ; preds = %592, %594, %597, %599 + %.0.i23 = phi float [ %593, %592 ], [ %595, %594 ], [ %598, %597 ], [ %600, %599 ], !dbg !33 + %601 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i26 = icmp eq i32 %601, 0, !dbg !33 + %602 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i29 = icmp eq i32 %602, 0, !dbg !33 + br i1 %.not.i26, label %608, label %603, !dbg !33 + +603: ; preds = %__nv_sqrtf.exit25 + br i1 %.not1.i29, label %606, label %604, !dbg !33 + +604: ; preds = %603 + %605 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %527) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +606: ; preds = %603 + %607 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %527) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +608: ; preds = %__nv_sqrtf.exit25 + br i1 %.not1.i29, label %611, label %609, !dbg !33 + +609: ; preds = %608 + %610 = tail call float @llvm.nvvm.sqrt.rn.f(float %527) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +611: ; preds = %608 + %612 = tail call float @llvm.nvvm.sqrt.approx.f(float %527) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +__nv_sqrtf.exit30: ; preds = %604, %606, %609, %611 + %.0.i28 = phi float [ %605, %604 ], [ %607, %606 ], [ %610, %609 ], [ %612, %611 ], !dbg !33 + %613 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i31 = icmp eq i32 %613, 0, !dbg !33 + %614 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i34 = icmp eq i32 %614, 0, !dbg !33 + br i1 %.not.i31, label %620, label %615, !dbg !33 + +615: ; preds = %__nv_sqrtf.exit30 + br i1 %.not1.i34, label %618, label %616, !dbg !33 + +616: ; preds = %615 + %617 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %528) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +618: ; preds = %615 + %619 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %528) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +620: ; preds = %__nv_sqrtf.exit30 + br i1 %.not1.i34, label %623, label %621, !dbg !33 + +621: ; preds = %620 + %622 = tail call float @llvm.nvvm.sqrt.rn.f(float %528) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +623: ; preds = %620 + %624 = tail call float @llvm.nvvm.sqrt.approx.f(float %528) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +__nv_sqrtf.exit35: ; preds = %616, %618, %621, %623 + %.0.i33 = phi float [ %617, %616 ], [ %619, %618 ], [ %622, %621 ], [ %624, %623 ], !dbg !33 + %625 = and i32 %11, 31, !dbg !16 + %626 = zext nneg i32 %625 to i64, !dbg !17 + %627 = or disjoint i64 %10, %626, !dbg !17 + %628 = getelementptr float, ptr addrspace(1) %2, i64 %627, !dbg !34 + %629 = getelementptr bfloat, ptr addrspace(1) %1, i64 %55, !dbg !35 + %630 = getelementptr bfloat, ptr addrspace(1) %629, i64 %64, !dbg !36 + %631 = getelementptr bfloat, ptr addrspace(1) %1, i64 %54, !dbg !35 + %632 = getelementptr bfloat, ptr addrspace(1) %631, i64 %64, !dbg !36 + %633 = getelementptr bfloat, ptr addrspace(1) %1, i64 %53, !dbg !35 + %634 = getelementptr bfloat, ptr addrspace(1) %633, i64 %64, !dbg !36 + %635 = getelementptr bfloat, ptr addrspace(1) %1, i64 %52, !dbg !35 + %636 = getelementptr bfloat, ptr addrspace(1) %635, i64 %64, !dbg !36 + %637 = getelementptr bfloat, ptr addrspace(1) %1, i64 %51, !dbg !35 + %638 = getelementptr bfloat, ptr addrspace(1) %637, i64 %64, !dbg !36 + %639 = getelementptr bfloat, ptr addrspace(1) %1, i64 %50, !dbg !35 + %640 = getelementptr bfloat, ptr addrspace(1) %639, i64 %64, !dbg !36 + %641 = getelementptr bfloat, ptr addrspace(1) %1, i64 %49, !dbg !35 + %642 = getelementptr bfloat, ptr addrspace(1) %641, i64 %64, !dbg !36 + %643 = getelementptr bfloat, ptr addrspace(1) %1, i64 %48, !dbg !35 + %644 = getelementptr bfloat, ptr addrspace(1) %643, i64 %64, !dbg !36 + %645 = icmp slt i64 %627, %37, !dbg !18 + %646 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i), !dbg !37 + %647 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i3), !dbg !37 + %648 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i8), !dbg !37 + %649 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i13), !dbg !37 + %650 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i18), !dbg !37 + %651 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i23), !dbg !37 + %652 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i28), !dbg !37 + %653 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i33), !dbg !37 + %654 = fmul float %646, %209, !dbg !38 + %655 = fmul float %646, %210, !dbg !38 + %656 = fmul float %646, %211, !dbg !38 + %657 = fmul float %646, %212, !dbg !38 + %658 = fmul float %646, %213, !dbg !38 + %659 = fmul float %646, %214, !dbg !38 + %660 = fmul float %646, %215, !dbg !38 + %661 = fmul float %646, %216, !dbg !38 + %662 = fmul float %647, %217, !dbg !38 + %663 = fmul float %647, %218, !dbg !38 + %664 = fmul float %647, %219, !dbg !38 + %665 = fmul float %647, %220, !dbg !38 + %666 = fmul float %647, %221, !dbg !38 + %667 = fmul float %647, %222, !dbg !38 + %668 = fmul float %647, %223, !dbg !38 + %669 = fmul float %647, %224, !dbg !38 + %670 = fmul float %648, %225, !dbg !38 + %671 = fmul float %648, %226, !dbg !38 + %672 = fmul float %648, %227, !dbg !38 + %673 = fmul float %648, %228, !dbg !38 + %674 = fmul float %648, %229, !dbg !38 + %675 = fmul float %648, %230, !dbg !38 + %676 = fmul float %648, %231, !dbg !38 + %677 = fmul float %648, %232, !dbg !38 + %678 = fmul float %649, %233, !dbg !38 + %679 = fmul float %649, %234, !dbg !38 + %680 = fmul float %649, %235, !dbg !38 + %681 = fmul float %649, %236, !dbg !38 + %682 = fmul float %649, %237, !dbg !38 + %683 = fmul float %649, %238, !dbg !38 + %684 = fmul float %649, %239, !dbg !38 + %685 = fmul float %649, %240, !dbg !38 + %686 = fmul float %650, %241, !dbg !38 + %687 = fmul float %650, %242, !dbg !38 + %688 = fmul float %650, %243, !dbg !38 + %689 = fmul float %650, %244, !dbg !38 + %690 = fmul float %650, %245, !dbg !38 + %691 = fmul float %650, %246, !dbg !38 + %692 = fmul float %650, %247, !dbg !38 + %693 = fmul float %650, %248, !dbg !38 + %694 = fmul float %651, %249, !dbg !38 + %695 = fmul float %651, %250, !dbg !38 + %696 = fmul float %651, %251, !dbg !38 + %697 = fmul float %651, %252, !dbg !38 + %698 = fmul float %651, %253, !dbg !38 + %699 = fmul float %651, %254, !dbg !38 + %700 = fmul float %651, %255, !dbg !38 + %701 = fmul float %651, %256, !dbg !38 + %702 = fmul float %652, %257, !dbg !38 + %703 = fmul float %652, %258, !dbg !38 + %704 = fmul float %652, %259, !dbg !38 + %705 = fmul float %652, %260, !dbg !38 + %706 = fmul float %652, %261, !dbg !38 + %707 = fmul float %652, %262, !dbg !38 + %708 = fmul float %652, %263, !dbg !38 + %709 = fmul float %652, %264, !dbg !38 + %710 = fmul float %653, %265, !dbg !38 + %711 = fmul float %653, %266, !dbg !38 + %712 = fmul float %653, %267, !dbg !38 + %713 = fmul float %653, %268, !dbg !38 + %714 = fmul float %653, %269, !dbg !38 + %715 = fmul float %653, %270, !dbg !38 + %716 = fmul float %653, %271, !dbg !38 + %717 = fmul float %653, %272, !dbg !38 + %718 = fptrunc float %654 to bfloat, !dbg !39 + %719 = fptrunc float %655 to bfloat, !dbg !39 + %720 = fptrunc float %656 to bfloat, !dbg !39 + %721 = fptrunc float %657 to bfloat, !dbg !39 + %722 = fptrunc float %658 to bfloat, !dbg !39 + %723 = fptrunc float %659 to bfloat, !dbg !39 + %724 = fptrunc float %660 to bfloat, !dbg !39 + %725 = fptrunc float %661 to bfloat, !dbg !39 + %726 = fptrunc float %662 to bfloat, !dbg !39 + %727 = fptrunc float %663 to bfloat, !dbg !39 + %728 = fptrunc float %664 to bfloat, !dbg !39 + %729 = fptrunc float %665 to bfloat, !dbg !39 + %730 = fptrunc float %666 to bfloat, !dbg !39 + %731 = fptrunc float %667 to bfloat, !dbg !39 + %732 = fptrunc float %668 to bfloat, !dbg !39 + %733 = fptrunc float %669 to bfloat, !dbg !39 + %734 = fptrunc float %670 to bfloat, !dbg !39 + %735 = fptrunc float %671 to bfloat, !dbg !39 + %736 = fptrunc float %672 to bfloat, !dbg !39 + %737 = fptrunc float %673 to bfloat, !dbg !39 + %738 = fptrunc float %674 to bfloat, !dbg !39 + %739 = fptrunc float %675 to bfloat, !dbg !39 + %740 = fptrunc float %676 to bfloat, !dbg !39 + %741 = fptrunc float %677 to bfloat, !dbg !39 + %742 = fptrunc float %678 to bfloat, !dbg !39 + %743 = fptrunc float %679 to bfloat, !dbg !39 + %744 = fptrunc float %680 to bfloat, !dbg !39 + %745 = fptrunc float %681 to bfloat, !dbg !39 + %746 = fptrunc float %682 to bfloat, !dbg !39 + %747 = fptrunc float %683 to bfloat, !dbg !39 + %748 = fptrunc float %684 to bfloat, !dbg !39 + %749 = fptrunc float %685 to bfloat, !dbg !39 + %750 = fptrunc float %686 to bfloat, !dbg !39 + %751 = fptrunc float %687 to bfloat, !dbg !39 + %752 = fptrunc float %688 to bfloat, !dbg !39 + %753 = fptrunc float %689 to bfloat, !dbg !39 + %754 = fptrunc float %690 to bfloat, !dbg !39 + %755 = fptrunc float %691 to bfloat, !dbg !39 + %756 = fptrunc float %692 to bfloat, !dbg !39 + %757 = fptrunc float %693 to bfloat, !dbg !39 + %758 = fptrunc float %694 to bfloat, !dbg !39 + %759 = fptrunc float %695 to bfloat, !dbg !39 + %760 = fptrunc float %696 to bfloat, !dbg !39 + %761 = fptrunc float %697 to bfloat, !dbg !39 + %762 = fptrunc float %698 to bfloat, !dbg !39 + %763 = fptrunc float %699 to bfloat, !dbg !39 + %764 = fptrunc float %700 to bfloat, !dbg !39 + %765 = fptrunc float %701 to bfloat, !dbg !39 + %766 = fptrunc float %702 to bfloat, !dbg !39 + %767 = fptrunc float %703 to bfloat, !dbg !39 + %768 = fptrunc float %704 to bfloat, !dbg !39 + %769 = fptrunc float %705 to bfloat, !dbg !39 + %770 = fptrunc float %706 to bfloat, !dbg !39 + %771 = fptrunc float %707 to bfloat, !dbg !39 + %772 = fptrunc float %708 to bfloat, !dbg !39 + %773 = fptrunc float %709 to bfloat, !dbg !39 + %774 = fptrunc float %710 to bfloat, !dbg !39 + %775 = fptrunc float %711 to bfloat, !dbg !39 + %776 = fptrunc float %712 to bfloat, !dbg !39 + %777 = fptrunc float %713 to bfloat, !dbg !39 + %778 = fptrunc float %714 to bfloat, !dbg !39 + %779 = fptrunc float %715 to bfloat, !dbg !39 + %780 = fptrunc float %716 to bfloat, !dbg !39 + %781 = fptrunc float %717 to bfloat, !dbg !39 + %782 = insertelement <2 x bfloat> poison, bfloat %718, i64 0, !dbg !40 + %783 = insertelement <2 x bfloat> %782, bfloat %719, i64 1, !dbg !40 + %784 = bitcast <2 x bfloat> %783 to i32, !dbg !40 + %785 = insertelement <2 x bfloat> poison, bfloat %720, i64 0, !dbg !40 + %786 = insertelement <2 x bfloat> %785, bfloat %721, i64 1, !dbg !40 + %787 = bitcast <2 x bfloat> %786 to i32, !dbg !40 + %788 = insertelement <2 x bfloat> poison, bfloat %722, i64 0, !dbg !40 + %789 = insertelement <2 x bfloat> %788, bfloat %723, i64 1, !dbg !40 + %790 = bitcast <2 x bfloat> %789 to i32, !dbg !40 + %791 = insertelement <2 x bfloat> poison, bfloat %724, i64 0, !dbg !40 + %792 = insertelement <2 x bfloat> %791, bfloat %725, i64 1, !dbg !40 + %793 = bitcast <2 x bfloat> %792 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %784, i32 %787, i32 %790, i32 %793, ptr addrspace(1) %644, i1 %38) #6, !dbg !40 + %794 = insertelement <2 x bfloat> poison, bfloat %726, i64 0, !dbg !40 + %795 = insertelement <2 x bfloat> %794, bfloat %727, i64 1, !dbg !40 + %796 = bitcast <2 x bfloat> %795 to i32, !dbg !40 + %797 = insertelement <2 x bfloat> poison, bfloat %728, i64 0, !dbg !40 + %798 = insertelement <2 x bfloat> %797, bfloat %729, i64 1, !dbg !40 + %799 = bitcast <2 x bfloat> %798 to i32, !dbg !40 + %800 = insertelement <2 x bfloat> poison, bfloat %730, i64 0, !dbg !40 + %801 = insertelement <2 x bfloat> %800, bfloat %731, i64 1, !dbg !40 + %802 = bitcast <2 x bfloat> %801 to i32, !dbg !40 + %803 = insertelement <2 x bfloat> poison, bfloat %732, i64 0, !dbg !40 + %804 = insertelement <2 x bfloat> %803, bfloat %733, i64 1, !dbg !40 + %805 = bitcast <2 x bfloat> %804 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %796, i32 %799, i32 %802, i32 %805, ptr addrspace(1) %642, i1 %39) #6, !dbg !40 + %806 = insertelement <2 x bfloat> poison, bfloat %734, i64 0, !dbg !40 + %807 = insertelement <2 x bfloat> %806, bfloat %735, i64 1, !dbg !40 + %808 = bitcast <2 x bfloat> %807 to i32, !dbg !40 + %809 = insertelement <2 x bfloat> poison, bfloat %736, i64 0, !dbg !40 + %810 = insertelement <2 x bfloat> %809, bfloat %737, i64 1, !dbg !40 + %811 = bitcast <2 x bfloat> %810 to i32, !dbg !40 + %812 = insertelement <2 x bfloat> poison, bfloat %738, i64 0, !dbg !40 + %813 = insertelement <2 x bfloat> %812, bfloat %739, i64 1, !dbg !40 + %814 = bitcast <2 x bfloat> %813 to i32, !dbg !40 + %815 = insertelement <2 x bfloat> poison, bfloat %740, i64 0, !dbg !40 + %816 = insertelement <2 x bfloat> %815, bfloat %741, i64 1, !dbg !40 + %817 = bitcast <2 x bfloat> %816 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %808, i32 %811, i32 %814, i32 %817, ptr addrspace(1) %640, i1 %40) #6, !dbg !40 + %818 = insertelement <2 x bfloat> poison, bfloat %742, i64 0, !dbg !40 + %819 = insertelement <2 x bfloat> %818, bfloat %743, i64 1, !dbg !40 + %820 = bitcast <2 x bfloat> %819 to i32, !dbg !40 + %821 = insertelement <2 x bfloat> poison, bfloat %744, i64 0, !dbg !40 + %822 = insertelement <2 x bfloat> %821, bfloat %745, i64 1, !dbg !40 + %823 = bitcast <2 x bfloat> %822 to i32, !dbg !40 + %824 = insertelement <2 x bfloat> poison, bfloat %746, i64 0, !dbg !40 + %825 = insertelement <2 x bfloat> %824, bfloat %747, i64 1, !dbg !40 + %826 = bitcast <2 x bfloat> %825 to i32, !dbg !40 + %827 = insertelement <2 x bfloat> poison, bfloat %748, i64 0, !dbg !40 + %828 = insertelement <2 x bfloat> %827, bfloat %749, i64 1, !dbg !40 + %829 = bitcast <2 x bfloat> %828 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %820, i32 %823, i32 %826, i32 %829, ptr addrspace(1) %638, i1 %41) #6, !dbg !40 + %830 = insertelement <2 x bfloat> poison, bfloat %750, i64 0, !dbg !40 + %831 = insertelement <2 x bfloat> %830, bfloat %751, i64 1, !dbg !40 + %832 = bitcast <2 x bfloat> %831 to i32, !dbg !40 + %833 = insertelement <2 x bfloat> poison, bfloat %752, i64 0, !dbg !40 + %834 = insertelement <2 x bfloat> %833, bfloat %753, i64 1, !dbg !40 + %835 = bitcast <2 x bfloat> %834 to i32, !dbg !40 + %836 = insertelement <2 x bfloat> poison, bfloat %754, i64 0, !dbg !40 + %837 = insertelement <2 x bfloat> %836, bfloat %755, i64 1, !dbg !40 + %838 = bitcast <2 x bfloat> %837 to i32, !dbg !40 + %839 = insertelement <2 x bfloat> poison, bfloat %756, i64 0, !dbg !40 + %840 = insertelement <2 x bfloat> %839, bfloat %757, i64 1, !dbg !40 + %841 = bitcast <2 x bfloat> %840 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %832, i32 %835, i32 %838, i32 %841, ptr addrspace(1) %636, i1 %42) #6, !dbg !40 + %842 = insertelement <2 x bfloat> poison, bfloat %758, i64 0, !dbg !40 + %843 = insertelement <2 x bfloat> %842, bfloat %759, i64 1, !dbg !40 + %844 = bitcast <2 x bfloat> %843 to i32, !dbg !40 + %845 = insertelement <2 x bfloat> poison, bfloat %760, i64 0, !dbg !40 + %846 = insertelement <2 x bfloat> %845, bfloat %761, i64 1, !dbg !40 + %847 = bitcast <2 x bfloat> %846 to i32, !dbg !40 + %848 = insertelement <2 x bfloat> poison, bfloat %762, i64 0, !dbg !40 + %849 = insertelement <2 x bfloat> %848, bfloat %763, i64 1, !dbg !40 + %850 = bitcast <2 x bfloat> %849 to i32, !dbg !40 + %851 = insertelement <2 x bfloat> poison, bfloat %764, i64 0, !dbg !40 + %852 = insertelement <2 x bfloat> %851, bfloat %765, i64 1, !dbg !40 + %853 = bitcast <2 x bfloat> %852 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %844, i32 %847, i32 %850, i32 %853, ptr addrspace(1) %634, i1 %43) #6, !dbg !40 + %854 = insertelement <2 x bfloat> poison, bfloat %766, i64 0, !dbg !40 + %855 = insertelement <2 x bfloat> %854, bfloat %767, i64 1, !dbg !40 + %856 = bitcast <2 x bfloat> %855 to i32, !dbg !40 + %857 = insertelement <2 x bfloat> poison, bfloat %768, i64 0, !dbg !40 + %858 = insertelement <2 x bfloat> %857, bfloat %769, i64 1, !dbg !40 + %859 = bitcast <2 x bfloat> %858 to i32, !dbg !40 + %860 = insertelement <2 x bfloat> poison, bfloat %770, i64 0, !dbg !40 + %861 = insertelement <2 x bfloat> %860, bfloat %771, i64 1, !dbg !40 + %862 = bitcast <2 x bfloat> %861 to i32, !dbg !40 + %863 = insertelement <2 x bfloat> poison, bfloat %772, i64 0, !dbg !40 + %864 = insertelement <2 x bfloat> %863, bfloat %773, i64 1, !dbg !40 + %865 = bitcast <2 x bfloat> %864 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %856, i32 %859, i32 %862, i32 %865, ptr addrspace(1) %632, i1 %44) #6, !dbg !40 + %866 = insertelement <2 x bfloat> poison, bfloat %774, i64 0, !dbg !40 + %867 = insertelement <2 x bfloat> %866, bfloat %775, i64 1, !dbg !40 + %868 = bitcast <2 x bfloat> %867 to i32, !dbg !40 + %869 = insertelement <2 x bfloat> poison, bfloat %776, i64 0, !dbg !40 + %870 = insertelement <2 x bfloat> %869, bfloat %777, i64 1, !dbg !40 + %871 = bitcast <2 x bfloat> %870 to i32, !dbg !40 + %872 = insertelement <2 x bfloat> poison, bfloat %778, i64 0, !dbg !40 + %873 = insertelement <2 x bfloat> %872, bfloat %779, i64 1, !dbg !40 + %874 = bitcast <2 x bfloat> %873 to i32, !dbg !40 + %875 = insertelement <2 x bfloat> poison, bfloat %780, i64 0, !dbg !40 + %876 = insertelement <2 x bfloat> %875, bfloat %781, i64 1, !dbg !40 + %877 = bitcast <2 x bfloat> %876 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %868, i32 %871, i32 %874, i32 %877, ptr addrspace(1) %630, i1 %45) #6, !dbg !40 + %878 = shl nuw nsw i32 %12, 1, !dbg !41 + %879 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %878, !dbg !41 + %880 = bitcast float %646 to i32, !dbg !41 + %881 = bitcast float %647 to i32, !dbg !41 + %882 = bitcast float %648 to i32, !dbg !41 + %883 = bitcast float %649 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %879, i32 %880, i32 %881, i32 %882, i32 %883) #6, !dbg !41 + %884 = getelementptr inbounds nuw i8, ptr addrspace(3) %879, i32 16, !dbg !41 + %885 = bitcast float %650 to i32, !dbg !41 + %886 = bitcast float %651 to i32, !dbg !41 + %887 = bitcast float %652 to i32, !dbg !41 + %888 = bitcast float %653 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %884, i32 %885, i32 %886, i32 %887, i32 %888) #6, !dbg !41 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !41 + %889 = shl nuw nsw i32 %11, 5, !dbg !41 + %890 = and i32 %889, 96, !dbg !41 + %891 = and i32 %11, 28, !dbg !41 + %892 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %890, !dbg !41 + %893 = getelementptr inbounds nuw i8, ptr addrspace(3) %892, i32 %891, !dbg !41 + %894 = load i32, ptr addrspace(3) %893, align 4, !dbg !41 + %895 = and i32 %11, 32, !dbg !41 + %896 = icmp eq i32 %895, 0, !dbg !41 + %897 = and i1 %896, %645, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %894, ptr addrspace(1) %628, i1 %897) #6, !dbg !41 + ret void, !dbg !42 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.div.full(float, float) #3 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #4 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #5 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.f(float) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="64" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) } +attributes #4 = { convergent nocallback nounwind } +attributes #5 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #6 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "l2norm_fwd_kernel", linkageName: "l2norm_fwd_kernel", scope: !1, file: !1, line: 69, type: !6, scopeLine: 69, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !10, !11, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 70, column: 24, scope: !5) +!14 = !DILocation(line: 70, column: 30, scope: !5) +!15 = !DILocation(line: 71, column: 16, scope: !5) +!16 = !DILocation(line: 71, column: 34, scope: !5) +!17 = !DILocation(line: 71, column: 21, scope: !5) +!18 = !DILocation(line: 73, column: 16, scope: !5) +!19 = !DILocation(line: 74, column: 30, scope: !5) +!20 = !DILocation(line: 75, column: 29, scope: !5) +!21 = !DILocation(line: 75, column: 14, scope: !5) +!22 = !DILocation(line: 75, column: 33, scope: !5) +!23 = !DILocation(line: 79, column: 18, scope: !5) +!24 = !DILocation(line: 79, column: 47, scope: !5) +!25 = !DILocation(line: 80, column: 38, scope: !5) +!26 = !DILocation(line: 263, column: 15, scope: !27, inlinedAt: !29) +!27 = distinct !DILexicalBlockFile(scope: !5, file: !28, discriminator: 0) +!28 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!29 = !DILocation(line: 293, column: 36, scope: !27, inlinedAt: !30) +!30 = !DILocation(line: 80, column: 43, scope: !31) +!31 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!32 = !DILocation(line: 80, column: 48, scope: !5) +!33 = !DILocation(line: 80, column: 25, scope: !5) +!34 = !DILocation(line: 77, column: 20, scope: !5) +!35 = !DILocation(line: 76, column: 14, scope: !5) +!36 = !DILocation(line: 76, column: 33, scope: !5) +!37 = !DILocation(line: 80, column: 17, scope: !5) +!38 = !DILocation(line: 81, column: 16, scope: !5) +!39 = !DILocation(line: 83, column: 25, scope: !5) +!40 = !DILocation(line: 83, column: 18, scope: !5) +!41 = !DILocation(line: 84, column: 21, scope: !5) +!42 = !DILocation(line: 84, column: 4, scope: !5) diff --git a/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/l2norm_fwd_kernel.ptx b/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/l2norm_fwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..bf9f670baccbc9ea04a140b39905f33c68561db6 --- /dev/null +++ b/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/l2norm_fwd_kernel.ptx @@ -0,0 +1,966 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_fwd_kernel // -- Begin function l2norm_fwd_kernel +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; +.global .align 1 .b8 _$_str_$_2[17] = {95, 95, 67, 85, 68, 65, 95, 80, 82, 69, 67, 95, 83, 81, 82, 84}; + // @l2norm_fwd_kernel +.visible .entry l2norm_fwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_2, + .param .f32 l2norm_fwd_kernel_param_3, + .param .u32 l2norm_fwd_kernel_param_4, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_6 +) +.reqntid 64 +{ + .reg .pred %p<12>; + .reg .b16 %rs<65>; + .reg .b32 %r<373>; + .reg .b64 %rd<67>; + .loc 1 69 0 // l2norm.py:69:0 +$L__func_begin0: + .loc 1 69 0 // l2norm.py:69:0 + +// %bb.0: // %__nv_sqrtf.exit + ld.param.b64 %rd18, [l2norm_fwd_kernel_param_0]; + ld.param.b64 %rd19, [l2norm_fwd_kernel_param_1]; +$L__tmp0: + .loc 1 70 24 // l2norm.py:70:24 + mov.u32 %r77, %ctaid.x; + ld.param.b64 %rd20, [l2norm_fwd_kernel_param_2]; + .loc 1 71 16 // l2norm.py:71:16 + mul.wide.u32 %rd21, %r77, 32; + ld.param.b32 %r78, [l2norm_fwd_kernel_param_3]; + ld.param.s32 %rd22, [l2norm_fwd_kernel_param_4]; + .loc 1 71 34 // l2norm.py:71:34 + mov.u32 %r79, %tid.x; + and.b32 %r80, %r79, 48; + bfe.u32 %r81, %r79, 4, 2; + or.b32 %r82, %r81, 4; + or.b32 %r83, %r81, 8; + or.b32 %r84, %r81, 12; + or.b32 %r85, %r81, 16; + or.b32 %r86, %r81, 20; + or.b32 %r87, %r81, 24; + or.b32 %r88, %r81, 28; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd23, %r81; + cvt.u64.u32 %rd24, %r82; + cvt.u64.u32 %rd25, %r83; + cvt.u64.u32 %rd26, %r84; + cvt.u64.u32 %rd27, %r85; + cvt.u64.u32 %rd28, %r86; + cvt.u64.u32 %rd29, %r87; + cvt.u64.u32 %rd30, %r88; + or.b64 %rd31, %rd21, %rd23; + or.b64 %rd32, %rd21, %rd24; + or.b64 %rd33, %rd21, %rd25; + or.b64 %rd34, %rd21, %rd26; + or.b64 %rd35, %rd21, %rd27; + or.b64 %rd36, %rd21, %rd28; + or.b64 %rd37, %rd21, %rd29; + or.b64 %rd38, %rd21, %rd30; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p1, %rd31, %rd22; + setp.lt.s64 %p2, %rd32, %rd22; + setp.lt.s64 %p3, %rd33, %rd22; + setp.lt.s64 %p4, %rd34, %rd22; + setp.lt.s64 %p5, %rd35, %rd22; + setp.lt.s64 %p6, %rd36, %rd22; + setp.lt.s64 %p7, %rd37, %rd22; + setp.lt.s64 %p8, %rd38, %rd22; + .loc 1 74 30 // l2norm.py:74:30 + shl.b32 %r89, %r79, 3; + and.b32 %r90, %r89, 120; + .loc 1 75 14 // l2norm.py:75:14 + shl.b64 %rd39, %rd31, 8; + add.s64 %rd40, %rd18, %rd39; + shl.b64 %rd41, %rd32, 8; + add.s64 %rd42, %rd18, %rd41; + shl.b64 %rd43, %rd33, 8; + add.s64 %rd44, %rd18, %rd43; + shl.b64 %rd45, %rd34, 8; + add.s64 %rd46, %rd18, %rd45; + shl.b64 %rd47, %rd35, 8; + add.s64 %rd48, %rd18, %rd47; + shl.b64 %rd49, %rd36, 8; + add.s64 %rd50, %rd18, %rd49; + shl.b64 %rd51, %rd37, 8; + add.s64 %rd52, %rd18, %rd51; + shl.b64 %rd53, %rd38, 8; + add.s64 %rd54, %rd18, %rd53; + .loc 1 75 33 // l2norm.py:75:33 + mul.wide.u32 %rd55, %r90, 2; + add.s64 %rd1, %rd40, %rd55; + add.s64 %rd2, %rd42, %rd55; + add.s64 %rd3, %rd44, %rd55; + add.s64 %rd4, %rd46, %rd55; + add.s64 %rd5, %rd48, %rd55; + add.s64 %rd6, %rd50, %rd55; + add.s64 %rd7, %rd52, %rd55; + add.s64 %rd8, %rd54, %rd55; + mov.b32 %r5, 0; + .loc 1 79 18 // l2norm.py:79:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + mov.b32 {%rs5, %rs6}, %r3; + mov.b32 {%rs7, %rs8}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs9, %rs10}, %r6; + mov.b32 {%rs11, %rs12}, %r7; + mov.b32 {%rs13, %rs14}, %r8; + mov.b32 {%rs15, %rs16}, %r9; + // begin inline asm + mov.u32 %r10, %r5; + mov.u32 %r11, %r5; + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + @%p3 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs17, %rs18}, %r10; + mov.b32 {%rs19, %rs20}, %r11; + mov.b32 {%rs21, %rs22}, %r12; + mov.b32 {%rs23, %rs24}, %r13; + // begin inline asm + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + @%p4 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ]; + // end inline asm + mov.b32 {%rs25, %rs26}, %r14; + mov.b32 {%rs27, %rs28}, %r15; + mov.b32 {%rs29, %rs30}, %r16; + mov.b32 {%rs31, %rs32}, %r17; + // begin inline asm + mov.u32 %r18, %r5; + mov.u32 %r19, %r5; + mov.u32 %r20, %r5; + mov.u32 %r21, %r5; + @%p5 ld.global.v4.b32 { %r18, %r19, %r20, %r21 }, [ %rd5 + 0 ]; + // end inline asm + mov.b32 {%rs33, %rs34}, %r18; + mov.b32 {%rs35, %rs36}, %r19; + mov.b32 {%rs37, %rs38}, %r20; + mov.b32 {%rs39, %rs40}, %r21; + // begin inline asm + mov.u32 %r22, %r5; + mov.u32 %r23, %r5; + mov.u32 %r24, %r5; + mov.u32 %r25, %r5; + @%p6 ld.global.v4.b32 { %r22, %r23, %r24, %r25 }, [ %rd6 + 0 ]; + // end inline asm + mov.b32 {%rs41, %rs42}, %r22; + mov.b32 {%rs43, %rs44}, %r23; + mov.b32 {%rs45, %rs46}, %r24; + mov.b32 {%rs47, %rs48}, %r25; + // begin inline asm + mov.u32 %r26, %r5; + mov.u32 %r27, %r5; + mov.u32 %r28, %r5; + mov.u32 %r29, %r5; + @%p7 ld.global.v4.b32 { %r26, %r27, %r28, %r29 }, [ %rd7 + 0 ]; + // end inline asm + mov.b32 {%rs49, %rs50}, %r26; + mov.b32 {%rs51, %rs52}, %r27; + mov.b32 {%rs53, %rs54}, %r28; + mov.b32 {%rs55, %rs56}, %r29; + // begin inline asm + mov.u32 %r30, %r5; + mov.u32 %r31, %r5; + mov.u32 %r32, %r5; + mov.u32 %r33, %r5; + @%p8 ld.global.v4.b32 { %r30, %r31, %r32, %r33 }, [ %rd8 + 0 ]; + // end inline asm + mov.b32 {%rs57, %rs58}, %r30; + mov.b32 {%rs59, %rs60}, %r31; + mov.b32 {%rs61, %rs62}, %r32; + mov.b32 {%rs63, %rs64}, %r33; + .loc 1 79 47 // l2norm.py:79:47 + cvt.f32.bf16 %r91, %rs1; + cvt.f32.bf16 %r92, %rs2; + cvt.f32.bf16 %r93, %rs3; + cvt.f32.bf16 %r94, %rs4; + cvt.f32.bf16 %r95, %rs5; + cvt.f32.bf16 %r96, %rs6; + cvt.f32.bf16 %r97, %rs7; + cvt.f32.bf16 %r98, %rs8; + cvt.f32.bf16 %r99, %rs9; + cvt.f32.bf16 %r100, %rs10; + cvt.f32.bf16 %r101, %rs11; + cvt.f32.bf16 %r102, %rs12; + cvt.f32.bf16 %r103, %rs13; + cvt.f32.bf16 %r104, %rs14; + cvt.f32.bf16 %r105, %rs15; + cvt.f32.bf16 %r106, %rs16; + cvt.f32.bf16 %r107, %rs17; + cvt.f32.bf16 %r108, %rs18; + cvt.f32.bf16 %r109, %rs19; + cvt.f32.bf16 %r110, %rs20; + cvt.f32.bf16 %r111, %rs21; + cvt.f32.bf16 %r112, %rs22; + cvt.f32.bf16 %r113, %rs23; + cvt.f32.bf16 %r114, %rs24; + cvt.f32.bf16 %r115, %rs25; + cvt.f32.bf16 %r116, %rs26; + cvt.f32.bf16 %r117, %rs27; + cvt.f32.bf16 %r118, %rs28; + cvt.f32.bf16 %r119, %rs29; + cvt.f32.bf16 %r120, %rs30; + cvt.f32.bf16 %r121, %rs31; + cvt.f32.bf16 %r122, %rs32; + cvt.f32.bf16 %r123, %rs33; + cvt.f32.bf16 %r124, %rs34; + cvt.f32.bf16 %r125, %rs35; + cvt.f32.bf16 %r126, %rs36; + cvt.f32.bf16 %r127, %rs37; + cvt.f32.bf16 %r128, %rs38; + cvt.f32.bf16 %r129, %rs39; + cvt.f32.bf16 %r130, %rs40; + cvt.f32.bf16 %r131, %rs41; + cvt.f32.bf16 %r132, %rs42; + cvt.f32.bf16 %r133, %rs43; + cvt.f32.bf16 %r134, %rs44; + cvt.f32.bf16 %r135, %rs45; + cvt.f32.bf16 %r136, %rs46; + cvt.f32.bf16 %r137, %rs47; + cvt.f32.bf16 %r138, %rs48; + cvt.f32.bf16 %r139, %rs49; + cvt.f32.bf16 %r140, %rs50; + cvt.f32.bf16 %r141, %rs51; + cvt.f32.bf16 %r142, %rs52; + cvt.f32.bf16 %r143, %rs53; + cvt.f32.bf16 %r144, %rs54; + cvt.f32.bf16 %r145, %rs55; + cvt.f32.bf16 %r146, %rs56; + cvt.f32.bf16 %r147, %rs57; + cvt.f32.bf16 %r148, %rs58; + cvt.f32.bf16 %r149, %rs59; + cvt.f32.bf16 %r150, %rs60; + cvt.f32.bf16 %r151, %rs61; + cvt.f32.bf16 %r152, %rs62; + cvt.f32.bf16 %r153, %rs63; + cvt.f32.bf16 %r154, %rs64; + .loc 1 80 38 // l2norm.py:80:38 + mul.f32 %r155, %r92, %r92; + mul.f32 %r156, %r100, %r100; + mul.f32 %r157, %r108, %r108; + mul.f32 %r158, %r116, %r116; + mul.f32 %r159, %r124, %r124; + mul.f32 %r160, %r132, %r132; + mul.f32 %r161, %r140, %r140; + mul.f32 %r162, %r148, %r148; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + fma.rn.f32.bf16 %r163, %rs1, %rs1, %r155; + fma.rn.f32.bf16 %r164, %rs3, %rs3, %r163; + fma.rn.f32.bf16 %r165, %rs4, %rs4, %r164; + fma.rn.f32.bf16 %r166, %rs5, %rs5, %r165; + fma.rn.f32.bf16 %r167, %rs6, %rs6, %r166; + fma.rn.f32.bf16 %r168, %rs7, %rs7, %r167; + fma.rn.f32.bf16 %r169, %rs8, %rs8, %r168; + fma.rn.f32.bf16 %r170, %rs9, %rs9, %r156; + fma.rn.f32.bf16 %r171, %rs11, %rs11, %r170; + fma.rn.f32.bf16 %r172, %rs12, %rs12, %r171; + fma.rn.f32.bf16 %r173, %rs13, %rs13, %r172; + fma.rn.f32.bf16 %r174, %rs14, %rs14, %r173; + fma.rn.f32.bf16 %r175, %rs15, %rs15, %r174; + fma.rn.f32.bf16 %r176, %rs16, %rs16, %r175; + fma.rn.f32.bf16 %r177, %rs17, %rs17, %r157; + fma.rn.f32.bf16 %r178, %rs19, %rs19, %r177; + fma.rn.f32.bf16 %r179, %rs20, %rs20, %r178; + fma.rn.f32.bf16 %r180, %rs21, %rs21, %r179; + fma.rn.f32.bf16 %r181, %rs22, %rs22, %r180; + fma.rn.f32.bf16 %r182, %rs23, %rs23, %r181; + fma.rn.f32.bf16 %r183, %rs24, %rs24, %r182; + fma.rn.f32.bf16 %r184, %rs25, %rs25, %r158; + fma.rn.f32.bf16 %r185, %rs27, %rs27, %r184; + fma.rn.f32.bf16 %r186, %rs28, %rs28, %r185; + fma.rn.f32.bf16 %r187, %rs29, %rs29, %r186; + fma.rn.f32.bf16 %r188, %rs30, %rs30, %r187; + fma.rn.f32.bf16 %r189, %rs31, %rs31, %r188; + fma.rn.f32.bf16 %r190, %rs32, %rs32, %r189; + fma.rn.f32.bf16 %r191, %rs33, %rs33, %r159; + fma.rn.f32.bf16 %r192, %rs35, %rs35, %r191; + fma.rn.f32.bf16 %r193, %rs36, %rs36, %r192; + fma.rn.f32.bf16 %r194, %rs37, %rs37, %r193; + fma.rn.f32.bf16 %r195, %rs38, %rs38, %r194; + fma.rn.f32.bf16 %r196, %rs39, %rs39, %r195; + fma.rn.f32.bf16 %r197, %rs40, %rs40, %r196; + fma.rn.f32.bf16 %r198, %rs41, %rs41, %r160; + fma.rn.f32.bf16 %r199, %rs43, %rs43, %r198; + fma.rn.f32.bf16 %r200, %rs44, %rs44, %r199; + fma.rn.f32.bf16 %r201, %rs45, %rs45, %r200; + fma.rn.f32.bf16 %r202, %rs46, %rs46, %r201; + fma.rn.f32.bf16 %r203, %rs47, %rs47, %r202; + fma.rn.f32.bf16 %r204, %rs48, %rs48, %r203; + fma.rn.f32.bf16 %r205, %rs49, %rs49, %r161; + fma.rn.f32.bf16 %r206, %rs51, %rs51, %r205; + fma.rn.f32.bf16 %r207, %rs52, %rs52, %r206; + fma.rn.f32.bf16 %r208, %rs53, %rs53, %r207; + fma.rn.f32.bf16 %r209, %rs54, %rs54, %r208; + fma.rn.f32.bf16 %r210, %rs55, %rs55, %r209; + fma.rn.f32.bf16 %r211, %rs56, %rs56, %r210; + fma.rn.f32.bf16 %r212, %rs57, %rs57, %r162; + fma.rn.f32.bf16 %r213, %rs59, %rs59, %r212; + fma.rn.f32.bf16 %r214, %rs60, %rs60, %r213; + fma.rn.f32.bf16 %r215, %rs61, %rs61, %r214; + fma.rn.f32.bf16 %r216, %rs62, %rs62, %r215; + fma.rn.f32.bf16 %r217, %rs63, %rs63, %r216; + fma.rn.f32.bf16 %r218, %rs64, %rs64, %r217; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r219, %r169, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r220, %r169, %r219; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r221, %r220, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r222, %r220, %r221; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r223, %r222, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r224, %r222, %r223; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r225, %r224, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r226, %r224, %r225; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r227, %r176, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r228, %r176, %r227; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r229, %r228, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r230, %r228, %r229; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r231, %r230, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r232, %r230, %r231; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r233, %r232, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r234, %r232, %r233; +$L__tmp18: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r235, %r183, 8, 31, -1; +$L__tmp19: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r236, %r183, %r235; +$L__tmp20: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r237, %r236, 4, 31, -1; +$L__tmp21: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r238, %r236, %r237; +$L__tmp22: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r239, %r238, 2, 31, -1; +$L__tmp23: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r240, %r238, %r239; +$L__tmp24: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r241, %r240, 1, 31, -1; +$L__tmp25: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r242, %r240, %r241; +$L__tmp26: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r243, %r190, 8, 31, -1; +$L__tmp27: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r244, %r190, %r243; +$L__tmp28: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r245, %r244, 4, 31, -1; +$L__tmp29: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r246, %r244, %r245; +$L__tmp30: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r247, %r246, 2, 31, -1; +$L__tmp31: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r248, %r246, %r247; +$L__tmp32: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r249, %r248, 1, 31, -1; +$L__tmp33: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r250, %r248, %r249; +$L__tmp34: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r251, %r197, 8, 31, -1; +$L__tmp35: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r252, %r197, %r251; +$L__tmp36: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r253, %r252, 4, 31, -1; +$L__tmp37: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r254, %r252, %r253; +$L__tmp38: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r255, %r254, 2, 31, -1; +$L__tmp39: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r256, %r254, %r255; +$L__tmp40: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r257, %r256, 1, 31, -1; +$L__tmp41: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r258, %r256, %r257; +$L__tmp42: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r259, %r204, 8, 31, -1; +$L__tmp43: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r260, %r204, %r259; +$L__tmp44: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r261, %r260, 4, 31, -1; +$L__tmp45: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r262, %r260, %r261; +$L__tmp46: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r263, %r262, 2, 31, -1; +$L__tmp47: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r264, %r262, %r263; +$L__tmp48: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r265, %r264, 1, 31, -1; +$L__tmp49: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r266, %r264, %r265; +$L__tmp50: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r267, %r211, 8, 31, -1; +$L__tmp51: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r268, %r211, %r267; +$L__tmp52: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r269, %r268, 4, 31, -1; +$L__tmp53: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r270, %r268, %r269; +$L__tmp54: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r271, %r270, 2, 31, -1; +$L__tmp55: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r272, %r270, %r271; +$L__tmp56: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r273, %r272, 1, 31, -1; +$L__tmp57: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r274, %r272, %r273; +$L__tmp58: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r275, %r218, 8, 31, -1; +$L__tmp59: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r276, %r218, %r275; +$L__tmp60: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r277, %r276, 4, 31, -1; +$L__tmp61: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r278, %r276, %r277; +$L__tmp62: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r279, %r278, 2, 31, -1; +$L__tmp63: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r280, %r278, %r279; +$L__tmp64: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r281, %r280, 1, 31, -1; +$L__tmp65: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r282, %r280, %r281; +$L__tmp66: + .loc 1 80 48 // l2norm.py:80:48 + add.f32 %r283, %r78, %r226; + add.f32 %r284, %r78, %r234; + add.f32 %r285, %r78, %r242; + add.f32 %r286, %r78, %r250; + add.f32 %r287, %r78, %r258; + add.f32 %r288, %r78, %r266; + add.f32 %r289, %r78, %r274; + add.f32 %r290, %r78, %r282; + .loc 1 80 25 // l2norm.py:80:25 + sqrt.approx.ftz.f32 %r291, %r283; + sqrt.approx.ftz.f32 %r292, %r284; + sqrt.approx.ftz.f32 %r293, %r285; + sqrt.approx.ftz.f32 %r294, %r286; + sqrt.approx.ftz.f32 %r295, %r287; + sqrt.approx.ftz.f32 %r296, %r288; + sqrt.approx.ftz.f32 %r297, %r289; + sqrt.approx.ftz.f32 %r298, %r290; + .loc 1 71 34 // l2norm.py:71:34 + and.b32 %r299, %r79, 31; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd56, %r299; + or.b64 %rd57, %rd21, %rd56; + .loc 1 77 20 // l2norm.py:77:20 + shl.b64 %rd58, %rd57, 2; + add.s64 %rd17, %rd20, %rd58; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd59, %rd19, %rd53; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd16, %rd59, %rd55; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd60, %rd19, %rd51; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd15, %rd60, %rd55; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd61, %rd19, %rd49; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd14, %rd61, %rd55; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd62, %rd19, %rd47; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd13, %rd62, %rd55; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd63, %rd19, %rd45; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd12, %rd63, %rd55; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd64, %rd19, %rd43; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd11, %rd64, %rd55; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd65, %rd19, %rd41; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd10, %rd65, %rd55; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd66, %rd19, %rd39; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd9, %rd66, %rd55; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p10, %rd57, %rd22; + mov.b32 %r300, 0f3F800000; + .loc 1 80 17 // l2norm.py:80:17 + div.full.f32 %r67, %r300, %r291; + div.full.f32 %r68, %r300, %r292; + div.full.f32 %r69, %r300, %r293; + div.full.f32 %r70, %r300, %r294; + div.full.f32 %r72, %r300, %r295; + div.full.f32 %r73, %r300, %r296; + div.full.f32 %r74, %r300, %r297; + div.full.f32 %r75, %r300, %r298; + .loc 1 81 16 // l2norm.py:81:16 + mul.f32 %r301, %r67, %r91; + mul.f32 %r302, %r67, %r92; + mul.f32 %r303, %r67, %r93; + mul.f32 %r304, %r67, %r94; + mul.f32 %r305, %r67, %r95; + mul.f32 %r306, %r67, %r96; + mul.f32 %r307, %r67, %r97; + mul.f32 %r308, %r67, %r98; + mul.f32 %r309, %r68, %r99; + mul.f32 %r310, %r68, %r100; + mul.f32 %r311, %r68, %r101; + mul.f32 %r312, %r68, %r102; + mul.f32 %r313, %r68, %r103; + mul.f32 %r314, %r68, %r104; + mul.f32 %r315, %r68, %r105; + mul.f32 %r316, %r68, %r106; + mul.f32 %r317, %r69, %r107; + mul.f32 %r318, %r69, %r108; + mul.f32 %r319, %r69, %r109; + mul.f32 %r320, %r69, %r110; + mul.f32 %r321, %r69, %r111; + mul.f32 %r322, %r69, %r112; + mul.f32 %r323, %r69, %r113; + mul.f32 %r324, %r69, %r114; + mul.f32 %r325, %r70, %r115; + mul.f32 %r326, %r70, %r116; + mul.f32 %r327, %r70, %r117; + mul.f32 %r328, %r70, %r118; + mul.f32 %r329, %r70, %r119; + mul.f32 %r330, %r70, %r120; + mul.f32 %r331, %r70, %r121; + mul.f32 %r332, %r70, %r122; + mul.f32 %r333, %r72, %r123; + mul.f32 %r334, %r72, %r124; + mul.f32 %r335, %r72, %r125; + mul.f32 %r336, %r72, %r126; + mul.f32 %r337, %r72, %r127; + mul.f32 %r338, %r72, %r128; + mul.f32 %r339, %r72, %r129; + mul.f32 %r340, %r72, %r130; + mul.f32 %r341, %r73, %r131; + mul.f32 %r342, %r73, %r132; + mul.f32 %r343, %r73, %r133; + mul.f32 %r344, %r73, %r134; + mul.f32 %r345, %r73, %r135; + mul.f32 %r346, %r73, %r136; + mul.f32 %r347, %r73, %r137; + mul.f32 %r348, %r73, %r138; + mul.f32 %r349, %r74, %r139; + mul.f32 %r350, %r74, %r140; + mul.f32 %r351, %r74, %r141; + mul.f32 %r352, %r74, %r142; + mul.f32 %r353, %r74, %r143; + mul.f32 %r354, %r74, %r144; + mul.f32 %r355, %r74, %r145; + mul.f32 %r356, %r74, %r146; + mul.f32 %r357, %r75, %r147; + mul.f32 %r358, %r75, %r148; + mul.f32 %r359, %r75, %r149; + mul.f32 %r360, %r75, %r150; + mul.f32 %r361, %r75, %r151; + mul.f32 %r362, %r75, %r152; + mul.f32 %r363, %r75, %r153; + mul.f32 %r364, %r75, %r154; + .loc 1 83 18 // l2norm.py:83:18 + cvt.rn.bf16x2.f32 %r34, %r302, %r301; + cvt.rn.bf16x2.f32 %r35, %r304, %r303; + cvt.rn.bf16x2.f32 %r36, %r306, %r305; + cvt.rn.bf16x2.f32 %r37, %r308, %r307; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd9 + 0 ], { %r34, %r35, %r36, %r37 }; + // end inline asm + cvt.rn.bf16x2.f32 %r38, %r310, %r309; + cvt.rn.bf16x2.f32 %r39, %r312, %r311; + cvt.rn.bf16x2.f32 %r40, %r314, %r313; + cvt.rn.bf16x2.f32 %r41, %r316, %r315; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd10 + 0 ], { %r38, %r39, %r40, %r41 }; + // end inline asm + cvt.rn.bf16x2.f32 %r42, %r318, %r317; + cvt.rn.bf16x2.f32 %r43, %r320, %r319; + cvt.rn.bf16x2.f32 %r44, %r322, %r321; + cvt.rn.bf16x2.f32 %r45, %r324, %r323; + // begin inline asm + @%p3 st.global.v4.b32 [ %rd11 + 0 ], { %r42, %r43, %r44, %r45 }; + // end inline asm + cvt.rn.bf16x2.f32 %r46, %r326, %r325; + cvt.rn.bf16x2.f32 %r47, %r328, %r327; + cvt.rn.bf16x2.f32 %r48, %r330, %r329; + cvt.rn.bf16x2.f32 %r49, %r332, %r331; + // begin inline asm + @%p4 st.global.v4.b32 [ %rd12 + 0 ], { %r46, %r47, %r48, %r49 }; + // end inline asm + cvt.rn.bf16x2.f32 %r50, %r334, %r333; + cvt.rn.bf16x2.f32 %r51, %r336, %r335; + cvt.rn.bf16x2.f32 %r52, %r338, %r337; + cvt.rn.bf16x2.f32 %r53, %r340, %r339; + // begin inline asm + @%p5 st.global.v4.b32 [ %rd13 + 0 ], { %r50, %r51, %r52, %r53 }; + // end inline asm + cvt.rn.bf16x2.f32 %r54, %r342, %r341; + cvt.rn.bf16x2.f32 %r55, %r344, %r343; + cvt.rn.bf16x2.f32 %r56, %r346, %r345; + cvt.rn.bf16x2.f32 %r57, %r348, %r347; + // begin inline asm + @%p6 st.global.v4.b32 [ %rd14 + 0 ], { %r54, %r55, %r56, %r57 }; + // end inline asm + cvt.rn.bf16x2.f32 %r58, %r350, %r349; + cvt.rn.bf16x2.f32 %r59, %r352, %r351; + cvt.rn.bf16x2.f32 %r60, %r354, %r353; + cvt.rn.bf16x2.f32 %r61, %r356, %r355; + // begin inline asm + @%p7 st.global.v4.b32 [ %rd15 + 0 ], { %r58, %r59, %r60, %r61 }; + // end inline asm + cvt.rn.bf16x2.f32 %r62, %r358, %r357; + cvt.rn.bf16x2.f32 %r63, %r360, %r359; + cvt.rn.bf16x2.f32 %r64, %r362, %r361; + cvt.rn.bf16x2.f32 %r65, %r364, %r363; + // begin inline asm + @%p8 st.global.v4.b32 [ %rd16 + 0 ], { %r62, %r63, %r64, %r65 }; + // end inline asm + .loc 1 84 21 // l2norm.py:84:21 + shl.b32 %r365, %r80, 1; + mov.b32 %r366, global_smem; + add.s32 %r66, %r366, %r365; + // begin inline asm + st.shared.v4.b32 [ %r66 + 0 ], { %r67, %r68, %r69, %r70 }; + // end inline asm + add.s32 %r71, %r66, 16; + // begin inline asm + st.shared.v4.b32 [ %r71 + 0 ], { %r72, %r73, %r74, %r75 }; + // end inline asm + bar.sync 0; + shl.b32 %r367, %r79, 5; + and.b32 %r368, %r367, 96; + and.b32 %r369, %r79, 28; + add.s32 %r370, %r366, %r368; + add.s32 %r371, %r370, %r369; + ld.shared.b32 %r76, [%r371]; + and.b32 %r372, %r79, 32; + setp.eq.b32 %p11, %r372, 0; + and.pred %p9, %p11, %p10; + // begin inline asm + @%p9 st.global.b32 [ %rd17 + 0 ], { %r76 }; + // end inline asm + .loc 1 84 4 // l2norm.py:84:4 + ret; +$L__tmp67: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp66 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 80 // DW_AT_call_line +.b8 43 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp66 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/l2norm_fwd_kernel.source b/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/l2norm_fwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..d9fa83aac0600b1e5d35d95ed8aaad5c5636e577 --- /dev/null +++ b/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/l2norm_fwd_kernel.source @@ -0,0 +1,172 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc38 = loc(unknown) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc45 = loc("x"(#loc)) +#loc46 = loc("y"(#loc)) +#loc47 = loc("rstd"(#loc)) +#loc48 = loc("eps"(#loc)) +#loc49 = loc("T"(#loc)) +#loc81 = loc("input"(#loc36)) +#loc82 = loc("a"(#loc41)) +#loc83 = loc("b"(#loc41)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc50) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc51) + %o_t = arith.constant 32 : i32 loc(#loc52) + %o_t_1 = arith.constant 32 : i64 loc(#loc52) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc52) + %o_t_3 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32> loc(#loc53) + %o_t_4 = arith.extsi %o_t_3 : tensor<32xi32> to tensor<32xi64> loc(#loc54) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<32xi64> loc(#loc54) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<32xi64> loc(#loc54) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc55) + %m_t = arith.extsi %T : i32 to i64 loc(#loc56) + %m_t_7 = tt.splat %m_t : i64 -> tensor<32xi64> loc(#loc56) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<32xi64> loc(#loc56) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<32xi1> -> tensor<32x1xi1> loc(#loc57) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc58) + %m_x_10 = arith.constant 128 : i32 loc(#loc59) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc59) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc59) + %m_x_13 = tt.broadcast %m_x : tensor<32x1xi1> -> tensor<32x128xi1> loc(#loc60) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<32x128xi1> loc(#loc60) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<32x128xi1> loc(#loc60) + %p_x = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<32xi64> -> tensor<32x1xi64> loc(#loc61) + %p_x_16 = arith.constant 128 : i32 loc(#loc62) + %p_x_17 = arith.constant 128 : i64 loc(#loc62) + %p_x_18 = arith.constant dense<128> : tensor<32x1xi64> loc(#loc62) + %p_x_19 = arith.muli %p_x, %p_x_18 : tensor<32x1xi64> loc(#loc62) + %p_x_20 = tt.splat %x : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc63) + %p_x_21 = tt.addptr %p_x_20, %p_x_19 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc63) + %p_x_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc64) + %p_x_23 = tt.broadcast %p_x_21 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc65) + %p_x_24 = tt.broadcast %p_x_22 : tensor<1x128xi32> -> tensor<32x128xi32> loc(#loc65) + %p_x_25 = tt.addptr %p_x_23, %p_x_24 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc65) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<32xi64> -> tensor<32x1xi64> loc(#loc66) + %p_y_26 = arith.constant 128 : i32 loc(#loc67) + %p_y_27 = arith.constant 128 : i64 loc(#loc67) + %p_y_28 = arith.constant dense<128> : tensor<32x1xi64> loc(#loc67) + %p_y_29 = arith.muli %p_y, %p_y_28 : tensor<32x1xi64> loc(#loc67) + %p_y_30 = tt.splat %y : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc68) + %p_y_31 = tt.addptr %p_y_30, %p_y_29 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc68) + %p_y_32 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc69) + %p_y_33 = tt.broadcast %p_y_31 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc70) + %p_y_34 = tt.broadcast %p_y_32 : tensor<1x128xi32> -> tensor<32x128xi32> loc(#loc70) + %p_y_35 = tt.addptr %p_y_33, %p_y_34 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc70) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<32x!tt.ptr> loc(#loc71) + %p_rstd_36 = tt.addptr %p_rstd, %o_t_6 : tensor<32x!tt.ptr>, tensor<32xi64> loc(#loc71) + %b_x = arith.constant 0.000000e+00 : f32 loc(#loc72) + %b_x_37 = arith.constant dense<0.000000e+00> : tensor<32x128xf32> loc(#loc72) + %b_x_38 = arith.truncf %b_x_37 : tensor<32x128xf32> to tensor<32x128xbf16> loc(#loc72) + %b_x_39 = tt.load %p_x_25, %m_x_15, %b_x_38 : tensor<32x128x!tt.ptr> loc(#loc72) + %b_x_40 = arith.extf %b_x_39 : tensor<32x128xbf16> to tensor<32x128xf32> loc(#loc73) + %b_rstd = arith.mulf %b_x_40, %b_x_40 : tensor<32x128xf32> loc(#loc74) + %b_rstd_41 = tt.call @triton.language.standard.sum__fp32S32_128S_c1_cFalse_cNone(%b_rstd) : (tensor<32x128xf32>) -> tensor<32xf32> loc(#loc75) + %b_rstd_42 = tt.splat %eps : f32 -> tensor<32xf32> loc(#loc76) + %b_rstd_43 = arith.addf %b_rstd_41, %b_rstd_42 : tensor<32xf32> loc(#loc76) + %b_rstd_44 = math.sqrt %b_rstd_43 : tensor<32xf32> loc(#loc77) + %b_rstd_45 = arith.constant 1 : i32 loc(#loc78) + %b_rstd_46 = arith.constant 1.000000e+00 : f32 loc(#loc78) + %b_rstd_47 = arith.constant dense<1.000000e+00> : tensor<32xf32> loc(#loc78) + %b_rstd_48 = arith.divf %b_rstd_47, %b_rstd_44 : tensor<32xf32> loc(#loc78) + %b_y = tt.expand_dims %b_rstd_48 {axis = 1 : i32} : tensor<32xf32> -> tensor<32x1xf32> loc(#loc79) + %b_y_49 = tt.broadcast %b_y : tensor<32x1xf32> -> tensor<32x128xf32> loc(#loc80) + %b_y_50 = arith.mulf %b_x_40, %b_y_49 : tensor<32x128xf32> loc(#loc80) + %0 = arith.truncf %b_y_50 : tensor<32x128xf32> to tensor<32x128xbf16> loc(#loc32) + tt.store %p_y_35, %0, %m_x_15 : tensor<32x128x!tt.ptr> loc(#loc33) + tt.store %p_rstd_36, %b_rstd_48, %m_t_8 : tensor<32x!tt.ptr> loc(#loc34) + tt.return loc(#loc35) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S32_128S_c1_cFalse_cNone(%input: tensor<32x128xf32> loc("input"(#loc36))) -> tensor<32xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc37) + tt.reduce.return %2 : f32 loc(#loc37) + }) : (tensor<32x128xf32>) -> tensor<32xf32> loc(#loc37) + tt.return %0 : tensor<32xf32> loc(#loc39) + ^bb1: // no predecessors + %1 = ub.poison : tensor<32xf32> loc(#loc40) + tt.return %1 : tensor<32xf32> loc(#loc40) + } loc(#loc36) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc41)), %b: f32 loc("b"(#loc41))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc42) + tt.return %0 : f32 loc(#loc43) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc44) + tt.return %1 : f32 loc(#loc44) + } loc(#loc41) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:18) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:29) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:37) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc50 = loc("i_t"(#loc1)) +#loc51 = loc("i_t"(#loc2)) +#loc52 = loc("o_t"(#loc3)) +#loc53 = loc("o_t"(#loc4)) +#loc54 = loc("o_t"(#loc5)) +#loc55 = loc("o_d"(#loc6)) +#loc56 = loc("m_t"(#loc7)) +#loc57 = loc("m_x"(#loc8)) +#loc58 = loc("m_x"(#loc9)) +#loc59 = loc("m_x"(#loc10)) +#loc60 = loc("m_x"(#loc11)) +#loc61 = loc("p_x"(#loc12)) +#loc62 = loc("p_x"(#loc13)) +#loc63 = loc("p_x"(#loc14)) +#loc64 = loc("p_x"(#loc15)) +#loc65 = loc("p_x"(#loc16)) +#loc66 = loc("p_y"(#loc17)) +#loc67 = loc("p_y"(#loc18)) +#loc68 = loc("p_y"(#loc19)) +#loc69 = loc("p_y"(#loc20)) +#loc70 = loc("p_y"(#loc21)) +#loc71 = loc("p_rstd"(#loc22)) +#loc72 = loc("b_x"(#loc23)) +#loc73 = loc("b_x"(#loc24)) +#loc74 = loc("b_rstd"(#loc25)) +#loc75 = loc("b_rstd"(#loc26)) +#loc76 = loc("b_rstd"(#loc27)) +#loc77 = loc("b_rstd"(#loc28)) +#loc78 = loc("b_rstd"(#loc29)) +#loc79 = loc("b_y"(#loc30)) +#loc80 = loc("b_y"(#loc31)) diff --git a/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/l2norm_fwd_kernel.ttgir b/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/l2norm_fwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..b137a7870e377a1a99f2fc1d8c75cb728717eea2 --- /dev/null +++ b/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/l2norm_fwd_kernel.ttgir @@ -0,0 +1,135 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [2, 1], order = [1, 0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [2], order = [0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc1 = loc(unknown) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc34 = loc("x"(#loc)) +#loc35 = loc("y"(#loc)) +#loc36 = loc("rstd"(#loc)) +#loc37 = loc("eps"(#loc)) +#loc38 = loc("T"(#loc)) +#loc59 = loc("b_rstd"(#loc23)) +#loc66 = loc(callsite(#loc1 at #loc59)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 2 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<1.000000e+00> : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<32x128xbf16, #blocked> loc(#loc1) + %c32_i64 = arith.constant 32 : i64 loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<32x1xi64, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc39) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc40) + %o_t = arith.muli %i_t_3, %c32_i64 : i64 loc(#loc41) + %o_t_4 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc42) + %o_t_5 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32, #blocked1> loc(#loc42) + %o_t_6 = arith.extsi %o_t_4 : tensor<32xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_7 = arith.extsi %o_t_5 : tensor<32xi32, #blocked1> to tensor<32xi64, #blocked1> loc(#loc43) + %o_t_8 = tt.splat %o_t : i64 -> tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_9 = tt.splat %o_t : i64 -> tensor<32xi64, #blocked1> loc(#loc43) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<32xi64, #blocked1> loc(#loc43) + %m_t = arith.extsi %T : i32 to i64 loc(#loc44) + %m_t_12 = tt.splat %m_t : i64 -> tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_13 = tt.splat %m_t : i64 -> tensor<32xi64, #blocked1> loc(#loc44) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<32xi64, #blocked1> loc(#loc44) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<32xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32x1xi1, #blocked> loc(#loc45) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc46) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc46) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_1 : tensor<1x128xi32, #blocked> loc(#loc47) + %m_x_19 = tt.broadcast %m_x : tensor<32x1xi1, #blocked> -> tensor<32x128xi1, #blocked> loc(#loc48) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<32x128xi1, #blocked> loc(#loc48) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<32x128xi1, #blocked> loc(#loc48) + %p_x = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32x1xi64, #blocked> loc(#loc49) + %p_x_22 = arith.muli %p_x, %cst_2 : tensor<32x1xi64, #blocked> loc(#loc50) + %p_x_23 = tt.splat %x : !tt.ptr -> tensor<32x1x!tt.ptr, #blocked> loc(#loc51) + %p_x_24 = tt.addptr %p_x_23, %p_x_22 : tensor<32x1x!tt.ptr, #blocked>, tensor<32x1xi64, #blocked> loc(#loc51) + %p_x_25 = tt.broadcast %p_x_24 : tensor<32x1x!tt.ptr, #blocked> -> tensor<32x128x!tt.ptr, #blocked> loc(#loc52) + %p_x_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<32x128xi32, #blocked> loc(#loc52) + %p_x_27 = tt.addptr %p_x_25, %p_x_26 : tensor<32x128x!tt.ptr, #blocked>, tensor<32x128xi32, #blocked> loc(#loc52) + %p_y = tt.splat %y : !tt.ptr -> tensor<32x1x!tt.ptr, #blocked> loc(#loc53) + %p_y_28 = tt.addptr %p_y, %p_x_22 : tensor<32x1x!tt.ptr, #blocked>, tensor<32x1xi64, #blocked> loc(#loc53) + %p_y_29 = tt.broadcast %p_y_28 : tensor<32x1x!tt.ptr, #blocked> -> tensor<32x128x!tt.ptr, #blocked> loc(#loc54) + %p_y_30 = tt.addptr %p_y_29, %p_x_26 : tensor<32x128x!tt.ptr, #blocked>, tensor<32x128xi32, #blocked> loc(#loc54) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<32x!tt.ptr, #blocked1> loc(#loc55) + %p_rstd_31 = tt.addptr %p_rstd, %o_t_11 : tensor<32x!tt.ptr, #blocked1>, tensor<32xi64, #blocked1> loc(#loc55) + %b_x = tt.load %p_x_27, %m_x_21, %cst_0 : tensor<32x128x!tt.ptr, #blocked> loc(#loc56) + %b_x_32 = arith.extf %b_x : tensor<32x128xbf16, #blocked> to tensor<32x128xf32, #blocked> loc(#loc57) + %b_rstd = arith.mulf %b_x_32, %b_x_32 : tensor<32x128xf32, #blocked> loc(#loc58) + %b_rstd_33 = "tt.reduce"(%b_rstd) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_40: f32 loc(callsite(#loc1 at #loc59)), %b_rstd_41: f32 loc(callsite(#loc1 at #loc59))): + %b_rstd_42 = arith.addf %b_rstd_40, %b_rstd_41 : f32 loc(#loc67) + tt.reduce.return %b_rstd_42 : f32 loc(#loc65) + }) : (tensor<32x128xf32, #blocked>) -> tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc65) + %b_rstd_34 = tt.splat %eps : f32 -> tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_36 = math.sqrt %b_rstd_35 : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc61) + %b_rstd_37 = arith.divf %cst, %b_rstd_36 : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc62) + %b_y = tt.expand_dims %b_rstd_37 {axis = 1 : i32} : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32x1xf32, #blocked> loc(#loc63) + %b_y_38 = tt.broadcast %b_y : tensor<32x1xf32, #blocked> -> tensor<32x128xf32, #blocked> loc(#loc64) + %b_y_39 = arith.mulf %b_x_32, %b_y_38 : tensor<32x128xf32, #blocked> loc(#loc64) + %0 = arith.truncf %b_y_39 : tensor<32x128xf32, #blocked> to tensor<32x128xbf16, #blocked> loc(#loc30) + tt.store %p_y_30, %0, %m_x_21 : tensor<32x128x!tt.ptr, #blocked> loc(#loc31) + %1 = ttg.convert_layout %b_rstd_37 : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32xf32, #blocked1> loc(#loc32) + tt.store %p_rstd_31, %1, %m_t_15 : tensor<32x!tt.ptr, #blocked1> loc(#loc32) + tt.return loc(#loc33) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc39 = loc("i_t"(#loc2)) +#loc40 = loc("i_t"(#loc3)) +#loc41 = loc("o_t"(#loc4)) +#loc42 = loc("o_t"(#loc5)) +#loc43 = loc("o_t"(#loc6)) +#loc44 = loc("m_t"(#loc7)) +#loc45 = loc("m_x"(#loc8)) +#loc46 = loc("m_x"(#loc9)) +#loc47 = loc("m_x"(#loc10)) +#loc48 = loc("m_x"(#loc11)) +#loc49 = loc("p_x"(#loc12)) +#loc50 = loc("p_x"(#loc13)) +#loc51 = loc("p_x"(#loc14)) +#loc52 = loc("p_x"(#loc15)) +#loc53 = loc("p_y"(#loc16)) +#loc54 = loc("p_y"(#loc17)) +#loc55 = loc("p_rstd"(#loc18)) +#loc56 = loc("b_x"(#loc19)) +#loc57 = loc("b_x"(#loc20)) +#loc58 = loc("b_rstd"(#loc21)) +#loc60 = loc("b_rstd"(#loc25)) +#loc61 = loc("b_rstd"(#loc26)) +#loc62 = loc("b_rstd"(#loc27)) +#loc63 = loc("b_y"(#loc28)) +#loc64 = loc("b_y"(#loc29)) +#loc65 = loc(callsite(#loc22 at #loc59)) +#loc67 = loc(callsite(#loc24 at #loc65)) diff --git a/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/l2norm_fwd_kernel.ttir b/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/l2norm_fwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..3688281bceaf7f8abdf8568104090be5e0170141 --- /dev/null +++ b/triton/gpu0/BY3NKOTZBWUTMG6PGGH6WJWJXMHSODAEOZFTSQ3WGB6FJRVTSESA/l2norm_fwd_kernel.ttir @@ -0,0 +1,128 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc3 = loc(unknown) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc35 = loc("x"(#loc)) +#loc36 = loc("y"(#loc)) +#loc37 = loc("rstd"(#loc)) +#loc38 = loc("eps"(#loc)) +#loc39 = loc("T"(#loc)) +#loc62 = loc("b_rstd"(#loc25)) +#loc68 = loc(callsite(#loc3 at #loc62)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %b_x = arith.constant dense<0.000000e+00> : tensor<32x128xbf16> loc(#loc40) + %b_rstd = arith.constant dense<1.000000e+00> : tensor<32xf32> loc(#loc41) + %cst = arith.constant dense<128> : tensor<32x1xi64> loc(#loc3) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc42) + %o_t = arith.constant 32 : i64 loc(#loc43) + %i_t = tt.get_program_id x : i32 loc(#loc44) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc45) + %o_t_1 = arith.muli %i_t_0, %o_t : i64 loc(#loc43) + %o_t_2 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32> loc(#loc46) + %o_t_3 = arith.extsi %o_t_2 : tensor<32xi32> to tensor<32xi64> loc(#loc47) + %o_t_4 = tt.splat %o_t_1 : i64 -> tensor<32xi64> loc(#loc47) + %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<32xi64> loc(#loc47) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc48) + %m_t = arith.extsi %T : i32 to i64 loc(#loc49) + %m_t_6 = tt.splat %m_t : i64 -> tensor<32xi64> loc(#loc49) + %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<32xi64> loc(#loc49) + %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<32xi1> -> tensor<32x1xi1> loc(#loc50) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc51) + %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc42) + %m_x_11 = tt.broadcast %m_x_8 : tensor<32x1xi1> -> tensor<32x128xi1> loc(#loc52) + %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<32x128xi1> loc(#loc52) + %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<32x128xi1> loc(#loc52) + %p_x = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<32xi64> -> tensor<32x1xi64> loc(#loc53) + %p_x_14 = arith.muli %p_x, %cst : tensor<32x1xi64> loc(#loc54) + %p_x_15 = tt.splat %x : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc55) + %p_x_16 = tt.addptr %p_x_15, %p_x_14 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc55) + %p_x_17 = tt.broadcast %p_x_16 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc56) + %p_x_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<32x128xi32> loc(#loc56) + %p_x_19 = tt.addptr %p_x_17, %p_x_18 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc56) + %p_y = tt.splat %y : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc57) + %p_y_20 = tt.addptr %p_y, %p_x_14 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc57) + %p_y_21 = tt.broadcast %p_y_20 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc58) + %p_y_22 = tt.addptr %p_y_21, %p_x_18 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<32x!tt.ptr> loc(#loc59) + %p_rstd_23 = tt.addptr %p_rstd, %o_t_5 : tensor<32x!tt.ptr>, tensor<32xi64> loc(#loc59) + %b_x_24 = tt.load %p_x_19, %m_x_13, %b_x : tensor<32x128x!tt.ptr> loc(#loc40) + %b_x_25 = arith.extf %b_x_24 : tensor<32x128xbf16> to tensor<32x128xf32> loc(#loc60) + %b_rstd_26 = arith.mulf %b_x_25, %b_x_25 : tensor<32x128xf32> loc(#loc61) + %b_rstd_27 = "tt.reduce"(%b_rstd_26) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_34: f32 loc(callsite(#loc3 at #loc62)), %b_rstd_35: f32 loc(callsite(#loc3 at #loc62))): + %b_rstd_36 = arith.addf %b_rstd_34, %b_rstd_35 : f32 loc(#loc69) + tt.reduce.return %b_rstd_36 : f32 loc(#loc67) + }) : (tensor<32x128xf32>) -> tensor<32xf32> loc(#loc67) + %b_rstd_28 = tt.splat %eps : f32 -> tensor<32xf32> loc(#loc63) + %b_rstd_29 = arith.addf %b_rstd_27, %b_rstd_28 : tensor<32xf32> loc(#loc63) + %b_rstd_30 = math.sqrt %b_rstd_29 : tensor<32xf32> loc(#loc64) + %b_rstd_31 = arith.divf %b_rstd, %b_rstd_30 : tensor<32xf32> loc(#loc41) + %b_y = tt.expand_dims %b_rstd_31 {axis = 1 : i32} : tensor<32xf32> -> tensor<32x1xf32> loc(#loc65) + %b_y_32 = tt.broadcast %b_y : tensor<32x1xf32> -> tensor<32x128xf32> loc(#loc66) + %b_y_33 = arith.mulf %b_x_25, %b_y_32 : tensor<32x128xf32> loc(#loc66) + %0 = arith.truncf %b_y_33 : tensor<32x128xf32> to tensor<32x128xbf16> loc(#loc31) + tt.store %p_y_22, %0, %m_x_13 : tensor<32x128x!tt.ptr> loc(#loc32) + tt.store %p_rstd_23, %b_rstd_31, %m_t_7 : tensor<32x!tt.ptr> loc(#loc33) + tt.return loc(#loc34) + } loc(#loc) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc40 = loc("b_x"(#loc1)) +#loc41 = loc("b_rstd"(#loc2)) +#loc42 = loc("m_x"(#loc4)) +#loc43 = loc("o_t"(#loc5)) +#loc44 = loc("i_t"(#loc6)) +#loc45 = loc("i_t"(#loc7)) +#loc46 = loc("o_t"(#loc8)) +#loc47 = loc("o_t"(#loc9)) +#loc48 = loc("o_d"(#loc10)) +#loc49 = loc("m_t"(#loc11)) +#loc50 = loc("m_x"(#loc12)) +#loc51 = loc("m_x"(#loc13)) +#loc52 = loc("m_x"(#loc14)) +#loc53 = loc("p_x"(#loc15)) +#loc54 = loc("p_x"(#loc16)) +#loc55 = loc("p_x"(#loc17)) +#loc56 = loc("p_x"(#loc18)) +#loc57 = loc("p_y"(#loc19)) +#loc58 = loc("p_y"(#loc20)) +#loc59 = loc("p_rstd"(#loc21)) +#loc60 = loc("b_x"(#loc22)) +#loc61 = loc("b_rstd"(#loc23)) +#loc63 = loc("b_rstd"(#loc27)) +#loc64 = loc("b_rstd"(#loc28)) +#loc65 = loc("b_y"(#loc29)) +#loc66 = loc("b_y"(#loc30)) +#loc67 = loc(callsite(#loc24 at #loc62)) +#loc69 = loc(callsite(#loc26 at #loc67)) diff --git a/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/__grp__l2norm_bwd_kernel.json b/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/__grp__l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..50b665004ed32a609516752145d6c40e44840f14 --- /dev/null +++ b/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/__grp__l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_bwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.source", "l2norm_bwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.ttir", "l2norm_bwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.ttgir", "l2norm_bwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.llir", "l2norm_bwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.ptx", "l2norm_bwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.cubin", "l2norm_bwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.cubin b/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..0dc25ef00885563c9c838f5517d5cd348fff0efe --- /dev/null +++ b/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.cubin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:adbcec56fcf2fe81419abda0e6425a1b21d050b9d8904323a3278ad6155f8006 +size 118080 diff --git a/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.json b/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..2c508422849b8c4a1057bbd769977cb66d719f3f --- /dev/null +++ b/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"hash": "16c2136ac0e79cb2a6c2799e839d54e753c761124bf8775ac477cdb9681428d2", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 2, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 256, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_bwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.llir b/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..40a357066fef3dccfa9b6131fe01883141a0340f --- /dev/null +++ b/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.llir @@ -0,0 +1,2439 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_bwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, ptr addrspace(1) %3, float %4, i32 %5, ptr addrspace(1) readnone captures(none) %6, ptr addrspace(1) readnone captures(none) %7) local_unnamed_addr #0 !dbg !4 { + %9 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !12 + %10 = zext nneg i32 %9 to i64, !dbg !13 + %11 = shl nuw nsw i64 %10, 6, !dbg !14 + %12 = sext i32 %5 to i64, !dbg !15 + %13 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !16 + %14 = and i32 %13, 48, !dbg !16 + %15 = lshr exact i32 %14, 4, !dbg !16 + %16 = insertelement <16 x i32> poison, i32 %13, i64 0, !dbg !16 + %17 = insertelement <16 x i32> %16, i32 %15, i64 1, !dbg !16 + %18 = shufflevector <16 x i32> %17, <16 x i32> poison, <16 x i32> , !dbg !16 + %19 = and <16 x i32> %18, , !dbg !16 + %20 = or disjoint <16 x i32> %18, , !dbg !16 + %21 = shufflevector <16 x i32> %19, <16 x i32> %20, <16 x i32> , !dbg !16 + %22 = zext nneg i32 %15 to i64, !dbg !17 + %23 = zext nneg <16 x i32> %21 to <16 x i64>, !dbg !17 + %24 = or disjoint i64 %11, %22, !dbg !17 + %25 = insertelement <16 x i64> poison, i64 %11, i64 0, !dbg !17 + %26 = shufflevector <16 x i64> %25, <16 x i64> poison, <16 x i32> zeroinitializer, !dbg !17 + %27 = or disjoint <16 x i64> %26, %23, !dbg !17 + %28 = icmp slt i64 %24, %12, !dbg !15 + %29 = insertelement <16 x i32> poison, i32 %5, i64 0, !dbg !15 + %30 = shufflevector <16 x i32> %29, <16 x i32> poison, <16 x i32> zeroinitializer, !dbg !15 + %31 = sext <16 x i32> %30 to <16 x i64>, !dbg !15 + %32 = icmp slt <16 x i64> %27, %31, !dbg !15 + %33 = shl nuw nsw i32 %13, 3, !dbg !18 + %34 = and i32 %33, 120, !dbg !18 + %35 = shl nuw nsw i64 %24, 7, !dbg !19 + %36 = extractelement <16 x i64> %27, i64 15, !dbg !19 + %37 = shl nuw nsw i64 %36, 7, !dbg !19 + %38 = extractelement <16 x i64> %27, i64 14, !dbg !19 + %39 = shl nuw nsw i64 %38, 7, !dbg !19 + %40 = extractelement <16 x i64> %27, i64 13, !dbg !19 + %41 = shl nuw nsw i64 %40, 7, !dbg !19 + %42 = extractelement <16 x i64> %27, i64 12, !dbg !19 + %43 = shl nuw nsw i64 %42, 7, !dbg !19 + %44 = extractelement <16 x i64> %27, i64 11, !dbg !19 + %45 = shl nuw nsw i64 %44, 7, !dbg !19 + %46 = extractelement <16 x i64> %27, i64 10, !dbg !19 + %47 = shl nuw nsw i64 %46, 7, !dbg !19 + %48 = extractelement <16 x i64> %27, i64 9, !dbg !19 + %49 = shl nuw nsw i64 %48, 7, !dbg !19 + %50 = extractelement <16 x i64> %27, i64 8, !dbg !19 + %51 = shl nuw nsw i64 %50, 7, !dbg !19 + %52 = extractelement <16 x i64> %27, i64 7, !dbg !19 + %53 = shl nuw nsw i64 %52, 7, !dbg !19 + %54 = extractelement <16 x i64> %27, i64 6, !dbg !19 + %55 = shl nuw nsw i64 %54, 7, !dbg !19 + %56 = extractelement <16 x i64> %27, i64 5, !dbg !19 + %57 = shl nuw nsw i64 %56, 7, !dbg !19 + %58 = extractelement <16 x i64> %27, i64 4, !dbg !19 + %59 = shl nuw nsw i64 %58, 7, !dbg !19 + %60 = extractelement <16 x i64> %27, i64 3, !dbg !19 + %61 = shl nuw nsw i64 %60, 7, !dbg !19 + %62 = extractelement <16 x i64> %27, i64 2, !dbg !19 + %63 = shl nuw nsw i64 %62, 7, !dbg !19 + %64 = extractelement <16 x i64> %27, i64 1, !dbg !19 + %65 = shl nuw nsw i64 %64, 7, !dbg !19 + %66 = getelementptr bfloat, ptr addrspace(1) %0, i64 %35, !dbg !20 + %67 = getelementptr bfloat, ptr addrspace(1) %0, i64 %37, !dbg !20 + %68 = getelementptr bfloat, ptr addrspace(1) %0, i64 %39, !dbg !20 + %69 = getelementptr bfloat, ptr addrspace(1) %0, i64 %41, !dbg !20 + %70 = getelementptr bfloat, ptr addrspace(1) %0, i64 %43, !dbg !20 + %71 = getelementptr bfloat, ptr addrspace(1) %0, i64 %45, !dbg !20 + %72 = getelementptr bfloat, ptr addrspace(1) %0, i64 %47, !dbg !20 + %73 = getelementptr bfloat, ptr addrspace(1) %0, i64 %49, !dbg !20 + %74 = getelementptr bfloat, ptr addrspace(1) %0, i64 %51, !dbg !20 + %75 = getelementptr bfloat, ptr addrspace(1) %0, i64 %53, !dbg !20 + %76 = getelementptr bfloat, ptr addrspace(1) %0, i64 %55, !dbg !20 + %77 = getelementptr bfloat, ptr addrspace(1) %0, i64 %57, !dbg !20 + %78 = getelementptr bfloat, ptr addrspace(1) %0, i64 %59, !dbg !20 + %79 = getelementptr bfloat, ptr addrspace(1) %0, i64 %61, !dbg !20 + %80 = getelementptr bfloat, ptr addrspace(1) %0, i64 %63, !dbg !20 + %81 = getelementptr bfloat, ptr addrspace(1) %0, i64 %65, !dbg !20 + %82 = zext nneg i32 %34 to i64, !dbg !21 + %83 = getelementptr bfloat, ptr addrspace(1) %66, i64 %82, !dbg !21 + %84 = getelementptr bfloat, ptr addrspace(1) %67, i64 %82, !dbg !21 + %85 = getelementptr bfloat, ptr addrspace(1) %68, i64 %82, !dbg !21 + %86 = getelementptr bfloat, ptr addrspace(1) %69, i64 %82, !dbg !21 + %87 = getelementptr bfloat, ptr addrspace(1) %70, i64 %82, !dbg !21 + %88 = getelementptr bfloat, ptr addrspace(1) %71, i64 %82, !dbg !21 + %89 = getelementptr bfloat, ptr addrspace(1) %72, i64 %82, !dbg !21 + %90 = getelementptr bfloat, ptr addrspace(1) %73, i64 %82, !dbg !21 + %91 = getelementptr bfloat, ptr addrspace(1) %74, i64 %82, !dbg !21 + %92 = getelementptr bfloat, ptr addrspace(1) %75, i64 %82, !dbg !21 + %93 = getelementptr bfloat, ptr addrspace(1) %76, i64 %82, !dbg !21 + %94 = getelementptr bfloat, ptr addrspace(1) %77, i64 %82, !dbg !21 + %95 = getelementptr bfloat, ptr addrspace(1) %78, i64 %82, !dbg !21 + %96 = getelementptr bfloat, ptr addrspace(1) %79, i64 %82, !dbg !21 + %97 = getelementptr bfloat, ptr addrspace(1) %80, i64 %82, !dbg !21 + %98 = getelementptr bfloat, ptr addrspace(1) %81, i64 %82, !dbg !21 + %99 = extractelement <16 x i64> %27, i64 0, !dbg !22 + %100 = getelementptr float, ptr addrspace(1) %1, i64 %99, !dbg !22 + %101 = getelementptr bfloat, ptr addrspace(1) %2, i64 %35, !dbg !23 + %102 = getelementptr bfloat, ptr addrspace(1) %2, i64 %37, !dbg !23 + %103 = getelementptr bfloat, ptr addrspace(1) %2, i64 %39, !dbg !23 + %104 = getelementptr bfloat, ptr addrspace(1) %2, i64 %41, !dbg !23 + %105 = getelementptr bfloat, ptr addrspace(1) %2, i64 %43, !dbg !23 + %106 = getelementptr bfloat, ptr addrspace(1) %2, i64 %45, !dbg !23 + %107 = getelementptr bfloat, ptr addrspace(1) %2, i64 %47, !dbg !23 + %108 = getelementptr bfloat, ptr addrspace(1) %2, i64 %49, !dbg !23 + %109 = getelementptr bfloat, ptr addrspace(1) %2, i64 %51, !dbg !23 + %110 = getelementptr bfloat, ptr addrspace(1) %2, i64 %53, !dbg !23 + %111 = getelementptr bfloat, ptr addrspace(1) %2, i64 %55, !dbg !23 + %112 = getelementptr bfloat, ptr addrspace(1) %2, i64 %57, !dbg !23 + %113 = getelementptr bfloat, ptr addrspace(1) %2, i64 %59, !dbg !23 + %114 = getelementptr bfloat, ptr addrspace(1) %2, i64 %61, !dbg !23 + %115 = getelementptr bfloat, ptr addrspace(1) %2, i64 %63, !dbg !23 + %116 = getelementptr bfloat, ptr addrspace(1) %2, i64 %65, !dbg !23 + %117 = getelementptr bfloat, ptr addrspace(1) %101, i64 %82, !dbg !24 + %118 = getelementptr bfloat, ptr addrspace(1) %102, i64 %82, !dbg !24 + %119 = getelementptr bfloat, ptr addrspace(1) %103, i64 %82, !dbg !24 + %120 = getelementptr bfloat, ptr addrspace(1) %104, i64 %82, !dbg !24 + %121 = getelementptr bfloat, ptr addrspace(1) %105, i64 %82, !dbg !24 + %122 = getelementptr bfloat, ptr addrspace(1) %106, i64 %82, !dbg !24 + %123 = getelementptr bfloat, ptr addrspace(1) %107, i64 %82, !dbg !24 + %124 = getelementptr bfloat, ptr addrspace(1) %108, i64 %82, !dbg !24 + %125 = getelementptr bfloat, ptr addrspace(1) %109, i64 %82, !dbg !24 + %126 = getelementptr bfloat, ptr addrspace(1) %110, i64 %82, !dbg !24 + %127 = getelementptr bfloat, ptr addrspace(1) %111, i64 %82, !dbg !24 + %128 = getelementptr bfloat, ptr addrspace(1) %112, i64 %82, !dbg !24 + %129 = getelementptr bfloat, ptr addrspace(1) %113, i64 %82, !dbg !24 + %130 = getelementptr bfloat, ptr addrspace(1) %114, i64 %82, !dbg !24 + %131 = getelementptr bfloat, ptr addrspace(1) %115, i64 %82, !dbg !24 + %132 = getelementptr bfloat, ptr addrspace(1) %116, i64 %82, !dbg !24 + %133 = getelementptr bfloat, ptr addrspace(1) %3, i64 %35, !dbg !25 + %134 = getelementptr bfloat, ptr addrspace(1) %3, i64 %37, !dbg !25 + %135 = getelementptr bfloat, ptr addrspace(1) %3, i64 %39, !dbg !25 + %136 = getelementptr bfloat, ptr addrspace(1) %3, i64 %41, !dbg !25 + %137 = getelementptr bfloat, ptr addrspace(1) %3, i64 %43, !dbg !25 + %138 = getelementptr bfloat, ptr addrspace(1) %3, i64 %45, !dbg !25 + %139 = getelementptr bfloat, ptr addrspace(1) %3, i64 %47, !dbg !25 + %140 = getelementptr bfloat, ptr addrspace(1) %3, i64 %49, !dbg !25 + %141 = getelementptr bfloat, ptr addrspace(1) %3, i64 %51, !dbg !25 + %142 = getelementptr bfloat, ptr addrspace(1) %3, i64 %53, !dbg !25 + %143 = getelementptr bfloat, ptr addrspace(1) %3, i64 %55, !dbg !25 + %144 = getelementptr bfloat, ptr addrspace(1) %3, i64 %57, !dbg !25 + %145 = getelementptr bfloat, ptr addrspace(1) %3, i64 %59, !dbg !25 + %146 = getelementptr bfloat, ptr addrspace(1) %3, i64 %61, !dbg !25 + %147 = getelementptr bfloat, ptr addrspace(1) %3, i64 %63, !dbg !25 + %148 = getelementptr bfloat, ptr addrspace(1) %3, i64 %65, !dbg !25 + %149 = getelementptr bfloat, ptr addrspace(1) %133, i64 %82, !dbg !26 + %150 = getelementptr bfloat, ptr addrspace(1) %134, i64 %82, !dbg !26 + %151 = getelementptr bfloat, ptr addrspace(1) %135, i64 %82, !dbg !26 + %152 = getelementptr bfloat, ptr addrspace(1) %136, i64 %82, !dbg !26 + %153 = getelementptr bfloat, ptr addrspace(1) %137, i64 %82, !dbg !26 + %154 = getelementptr bfloat, ptr addrspace(1) %138, i64 %82, !dbg !26 + %155 = getelementptr bfloat, ptr addrspace(1) %139, i64 %82, !dbg !26 + %156 = getelementptr bfloat, ptr addrspace(1) %140, i64 %82, !dbg !26 + %157 = getelementptr bfloat, ptr addrspace(1) %141, i64 %82, !dbg !26 + %158 = getelementptr bfloat, ptr addrspace(1) %142, i64 %82, !dbg !26 + %159 = getelementptr bfloat, ptr addrspace(1) %143, i64 %82, !dbg !26 + %160 = getelementptr bfloat, ptr addrspace(1) %144, i64 %82, !dbg !26 + %161 = getelementptr bfloat, ptr addrspace(1) %145, i64 %82, !dbg !26 + %162 = getelementptr bfloat, ptr addrspace(1) %146, i64 %82, !dbg !26 + %163 = getelementptr bfloat, ptr addrspace(1) %147, i64 %82, !dbg !26 + %164 = getelementptr bfloat, ptr addrspace(1) %148, i64 %82, !dbg !26 + %165 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %83, i1 %28) #4, !dbg !27 + %166 = extractvalue { i32, i32, i32, i32 } %165, 0, !dbg !27 + %167 = bitcast i32 %166 to <2 x bfloat>, !dbg !27 + %168 = extractvalue { i32, i32, i32, i32 } %165, 1, !dbg !27 + %169 = bitcast i32 %168 to <2 x bfloat>, !dbg !27 + %170 = extractvalue { i32, i32, i32, i32 } %165, 2, !dbg !27 + %171 = bitcast i32 %170 to <2 x bfloat>, !dbg !27 + %172 = extractvalue { i32, i32, i32, i32 } %165, 3, !dbg !27 + %173 = bitcast i32 %172 to <2 x bfloat>, !dbg !27 + %174 = extractelement <2 x bfloat> %167, i64 0, !dbg !27 + %175 = extractelement <2 x bfloat> %167, i64 1, !dbg !27 + %176 = extractelement <2 x bfloat> %169, i64 0, !dbg !27 + %177 = extractelement <2 x bfloat> %169, i64 1, !dbg !27 + %178 = extractelement <2 x bfloat> %171, i64 0, !dbg !27 + %179 = extractelement <2 x bfloat> %171, i64 1, !dbg !27 + %180 = extractelement <2 x bfloat> %173, i64 0, !dbg !27 + %181 = extractelement <2 x bfloat> %173, i64 1, !dbg !27 + %182 = extractelement <16 x i1> %32, i64 15, !dbg !27 + %183 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %84, i1 %182) #4, !dbg !27 + %184 = extractvalue { i32, i32, i32, i32 } %183, 0, !dbg !27 + %185 = bitcast i32 %184 to <2 x bfloat>, !dbg !27 + %186 = extractvalue { i32, i32, i32, i32 } %183, 1, !dbg !27 + %187 = bitcast i32 %186 to <2 x bfloat>, !dbg !27 + %188 = extractvalue { i32, i32, i32, i32 } %183, 2, !dbg !27 + %189 = bitcast i32 %188 to <2 x bfloat>, !dbg !27 + %190 = extractvalue { i32, i32, i32, i32 } %183, 3, !dbg !27 + %191 = bitcast i32 %190 to <2 x bfloat>, !dbg !27 + %192 = extractelement <2 x bfloat> %185, i64 0, !dbg !27 + %193 = extractelement <2 x bfloat> %185, i64 1, !dbg !27 + %194 = extractelement <2 x bfloat> %187, i64 0, !dbg !27 + %195 = extractelement <2 x bfloat> %187, i64 1, !dbg !27 + %196 = extractelement <2 x bfloat> %189, i64 0, !dbg !27 + %197 = extractelement <2 x bfloat> %189, i64 1, !dbg !27 + %198 = extractelement <2 x bfloat> %191, i64 0, !dbg !27 + %199 = extractelement <2 x bfloat> %191, i64 1, !dbg !27 + %200 = extractelement <16 x i1> %32, i64 14, !dbg !27 + %201 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %85, i1 %200) #4, !dbg !27 + %202 = extractvalue { i32, i32, i32, i32 } %201, 0, !dbg !27 + %203 = bitcast i32 %202 to <2 x bfloat>, !dbg !27 + %204 = extractvalue { i32, i32, i32, i32 } %201, 1, !dbg !27 + %205 = bitcast i32 %204 to <2 x bfloat>, !dbg !27 + %206 = extractvalue { i32, i32, i32, i32 } %201, 2, !dbg !27 + %207 = bitcast i32 %206 to <2 x bfloat>, !dbg !27 + %208 = extractvalue { i32, i32, i32, i32 } %201, 3, !dbg !27 + %209 = bitcast i32 %208 to <2 x bfloat>, !dbg !27 + %210 = extractelement <2 x bfloat> %203, i64 0, !dbg !27 + %211 = extractelement <2 x bfloat> %203, i64 1, !dbg !27 + %212 = extractelement <2 x bfloat> %205, i64 0, !dbg !27 + %213 = extractelement <2 x bfloat> %205, i64 1, !dbg !27 + %214 = extractelement <2 x bfloat> %207, i64 0, !dbg !27 + %215 = extractelement <2 x bfloat> %207, i64 1, !dbg !27 + %216 = extractelement <2 x bfloat> %209, i64 0, !dbg !27 + %217 = extractelement <2 x bfloat> %209, i64 1, !dbg !27 + %218 = extractelement <16 x i1> %32, i64 13, !dbg !27 + %219 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %86, i1 %218) #4, !dbg !27 + %220 = extractvalue { i32, i32, i32, i32 } %219, 0, !dbg !27 + %221 = bitcast i32 %220 to <2 x bfloat>, !dbg !27 + %222 = extractvalue { i32, i32, i32, i32 } %219, 1, !dbg !27 + %223 = bitcast i32 %222 to <2 x bfloat>, !dbg !27 + %224 = extractvalue { i32, i32, i32, i32 } %219, 2, !dbg !27 + %225 = bitcast i32 %224 to <2 x bfloat>, !dbg !27 + %226 = extractvalue { i32, i32, i32, i32 } %219, 3, !dbg !27 + %227 = bitcast i32 %226 to <2 x bfloat>, !dbg !27 + %228 = extractelement <2 x bfloat> %221, i64 0, !dbg !27 + %229 = extractelement <2 x bfloat> %221, i64 1, !dbg !27 + %230 = extractelement <2 x bfloat> %223, i64 0, !dbg !27 + %231 = extractelement <2 x bfloat> %223, i64 1, !dbg !27 + %232 = extractelement <2 x bfloat> %225, i64 0, !dbg !27 + %233 = extractelement <2 x bfloat> %225, i64 1, !dbg !27 + %234 = extractelement <2 x bfloat> %227, i64 0, !dbg !27 + %235 = extractelement <2 x bfloat> %227, i64 1, !dbg !27 + %236 = extractelement <16 x i1> %32, i64 12, !dbg !27 + %237 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %87, i1 %236) #4, !dbg !27 + %238 = extractvalue { i32, i32, i32, i32 } %237, 0, !dbg !27 + %239 = bitcast i32 %238 to <2 x bfloat>, !dbg !27 + %240 = extractvalue { i32, i32, i32, i32 } %237, 1, !dbg !27 + %241 = bitcast i32 %240 to <2 x bfloat>, !dbg !27 + %242 = extractvalue { i32, i32, i32, i32 } %237, 2, !dbg !27 + %243 = bitcast i32 %242 to <2 x bfloat>, !dbg !27 + %244 = extractvalue { i32, i32, i32, i32 } %237, 3, !dbg !27 + %245 = bitcast i32 %244 to <2 x bfloat>, !dbg !27 + %246 = extractelement <2 x bfloat> %239, i64 0, !dbg !27 + %247 = extractelement <2 x bfloat> %239, i64 1, !dbg !27 + %248 = extractelement <2 x bfloat> %241, i64 0, !dbg !27 + %249 = extractelement <2 x bfloat> %241, i64 1, !dbg !27 + %250 = extractelement <2 x bfloat> %243, i64 0, !dbg !27 + %251 = extractelement <2 x bfloat> %243, i64 1, !dbg !27 + %252 = extractelement <2 x bfloat> %245, i64 0, !dbg !27 + %253 = extractelement <2 x bfloat> %245, i64 1, !dbg !27 + %254 = extractelement <16 x i1> %32, i64 11, !dbg !27 + %255 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %88, i1 %254) #4, !dbg !27 + %256 = extractvalue { i32, i32, i32, i32 } %255, 0, !dbg !27 + %257 = bitcast i32 %256 to <2 x bfloat>, !dbg !27 + %258 = extractvalue { i32, i32, i32, i32 } %255, 1, !dbg !27 + %259 = bitcast i32 %258 to <2 x bfloat>, !dbg !27 + %260 = extractvalue { i32, i32, i32, i32 } %255, 2, !dbg !27 + %261 = bitcast i32 %260 to <2 x bfloat>, !dbg !27 + %262 = extractvalue { i32, i32, i32, i32 } %255, 3, !dbg !27 + %263 = bitcast i32 %262 to <2 x bfloat>, !dbg !27 + %264 = extractelement <2 x bfloat> %257, i64 0, !dbg !27 + %265 = extractelement <2 x bfloat> %257, i64 1, !dbg !27 + %266 = extractelement <2 x bfloat> %259, i64 0, !dbg !27 + %267 = extractelement <2 x bfloat> %259, i64 1, !dbg !27 + %268 = extractelement <2 x bfloat> %261, i64 0, !dbg !27 + %269 = extractelement <2 x bfloat> %261, i64 1, !dbg !27 + %270 = extractelement <2 x bfloat> %263, i64 0, !dbg !27 + %271 = extractelement <2 x bfloat> %263, i64 1, !dbg !27 + %272 = extractelement <16 x i1> %32, i64 10, !dbg !27 + %273 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %89, i1 %272) #4, !dbg !27 + %274 = extractvalue { i32, i32, i32, i32 } %273, 0, !dbg !27 + %275 = bitcast i32 %274 to <2 x bfloat>, !dbg !27 + %276 = extractvalue { i32, i32, i32, i32 } %273, 1, !dbg !27 + %277 = bitcast i32 %276 to <2 x bfloat>, !dbg !27 + %278 = extractvalue { i32, i32, i32, i32 } %273, 2, !dbg !27 + %279 = bitcast i32 %278 to <2 x bfloat>, !dbg !27 + %280 = extractvalue { i32, i32, i32, i32 } %273, 3, !dbg !27 + %281 = bitcast i32 %280 to <2 x bfloat>, !dbg !27 + %282 = extractelement <2 x bfloat> %275, i64 0, !dbg !27 + %283 = extractelement <2 x bfloat> %275, i64 1, !dbg !27 + %284 = extractelement <2 x bfloat> %277, i64 0, !dbg !27 + %285 = extractelement <2 x bfloat> %277, i64 1, !dbg !27 + %286 = extractelement <2 x bfloat> %279, i64 0, !dbg !27 + %287 = extractelement <2 x bfloat> %279, i64 1, !dbg !27 + %288 = extractelement <2 x bfloat> %281, i64 0, !dbg !27 + %289 = extractelement <2 x bfloat> %281, i64 1, !dbg !27 + %290 = extractelement <16 x i1> %32, i64 9, !dbg !27 + %291 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %90, i1 %290) #4, !dbg !27 + %292 = extractvalue { i32, i32, i32, i32 } %291, 0, !dbg !27 + %293 = bitcast i32 %292 to <2 x bfloat>, !dbg !27 + %294 = extractvalue { i32, i32, i32, i32 } %291, 1, !dbg !27 + %295 = bitcast i32 %294 to <2 x bfloat>, !dbg !27 + %296 = extractvalue { i32, i32, i32, i32 } %291, 2, !dbg !27 + %297 = bitcast i32 %296 to <2 x bfloat>, !dbg !27 + %298 = extractvalue { i32, i32, i32, i32 } %291, 3, !dbg !27 + %299 = bitcast i32 %298 to <2 x bfloat>, !dbg !27 + %300 = extractelement <2 x bfloat> %293, i64 0, !dbg !27 + %301 = extractelement <2 x bfloat> %293, i64 1, !dbg !27 + %302 = extractelement <2 x bfloat> %295, i64 0, !dbg !27 + %303 = extractelement <2 x bfloat> %295, i64 1, !dbg !27 + %304 = extractelement <2 x bfloat> %297, i64 0, !dbg !27 + %305 = extractelement <2 x bfloat> %297, i64 1, !dbg !27 + %306 = extractelement <2 x bfloat> %299, i64 0, !dbg !27 + %307 = extractelement <2 x bfloat> %299, i64 1, !dbg !27 + %308 = extractelement <16 x i1> %32, i64 8, !dbg !27 + %309 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %91, i1 %308) #4, !dbg !27 + %310 = extractvalue { i32, i32, i32, i32 } %309, 0, !dbg !27 + %311 = bitcast i32 %310 to <2 x bfloat>, !dbg !27 + %312 = extractvalue { i32, i32, i32, i32 } %309, 1, !dbg !27 + %313 = bitcast i32 %312 to <2 x bfloat>, !dbg !27 + %314 = extractvalue { i32, i32, i32, i32 } %309, 2, !dbg !27 + %315 = bitcast i32 %314 to <2 x bfloat>, !dbg !27 + %316 = extractvalue { i32, i32, i32, i32 } %309, 3, !dbg !27 + %317 = bitcast i32 %316 to <2 x bfloat>, !dbg !27 + %318 = extractelement <2 x bfloat> %311, i64 0, !dbg !27 + %319 = extractelement <2 x bfloat> %311, i64 1, !dbg !27 + %320 = extractelement <2 x bfloat> %313, i64 0, !dbg !27 + %321 = extractelement <2 x bfloat> %313, i64 1, !dbg !27 + %322 = extractelement <2 x bfloat> %315, i64 0, !dbg !27 + %323 = extractelement <2 x bfloat> %315, i64 1, !dbg !27 + %324 = extractelement <2 x bfloat> %317, i64 0, !dbg !27 + %325 = extractelement <2 x bfloat> %317, i64 1, !dbg !27 + %326 = extractelement <16 x i1> %32, i64 7, !dbg !27 + %327 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %92, i1 %326) #4, !dbg !27 + %328 = extractvalue { i32, i32, i32, i32 } %327, 0, !dbg !27 + %329 = bitcast i32 %328 to <2 x bfloat>, !dbg !27 + %330 = extractvalue { i32, i32, i32, i32 } %327, 1, !dbg !27 + %331 = bitcast i32 %330 to <2 x bfloat>, !dbg !27 + %332 = extractvalue { i32, i32, i32, i32 } %327, 2, !dbg !27 + %333 = bitcast i32 %332 to <2 x bfloat>, !dbg !27 + %334 = extractvalue { i32, i32, i32, i32 } %327, 3, !dbg !27 + %335 = bitcast i32 %334 to <2 x bfloat>, !dbg !27 + %336 = extractelement <2 x bfloat> %329, i64 0, !dbg !27 + %337 = extractelement <2 x bfloat> %329, i64 1, !dbg !27 + %338 = extractelement <2 x bfloat> %331, i64 0, !dbg !27 + %339 = extractelement <2 x bfloat> %331, i64 1, !dbg !27 + %340 = extractelement <2 x bfloat> %333, i64 0, !dbg !27 + %341 = extractelement <2 x bfloat> %333, i64 1, !dbg !27 + %342 = extractelement <2 x bfloat> %335, i64 0, !dbg !27 + %343 = extractelement <2 x bfloat> %335, i64 1, !dbg !27 + %344 = extractelement <16 x i1> %32, i64 6, !dbg !27 + %345 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %93, i1 %344) #4, !dbg !27 + %346 = extractvalue { i32, i32, i32, i32 } %345, 0, !dbg !27 + %347 = bitcast i32 %346 to <2 x bfloat>, !dbg !27 + %348 = extractvalue { i32, i32, i32, i32 } %345, 1, !dbg !27 + %349 = bitcast i32 %348 to <2 x bfloat>, !dbg !27 + %350 = extractvalue { i32, i32, i32, i32 } %345, 2, !dbg !27 + %351 = bitcast i32 %350 to <2 x bfloat>, !dbg !27 + %352 = extractvalue { i32, i32, i32, i32 } %345, 3, !dbg !27 + %353 = bitcast i32 %352 to <2 x bfloat>, !dbg !27 + %354 = extractelement <2 x bfloat> %347, i64 0, !dbg !27 + %355 = extractelement <2 x bfloat> %347, i64 1, !dbg !27 + %356 = extractelement <2 x bfloat> %349, i64 0, !dbg !27 + %357 = extractelement <2 x bfloat> %349, i64 1, !dbg !27 + %358 = extractelement <2 x bfloat> %351, i64 0, !dbg !27 + %359 = extractelement <2 x bfloat> %351, i64 1, !dbg !27 + %360 = extractelement <2 x bfloat> %353, i64 0, !dbg !27 + %361 = extractelement <2 x bfloat> %353, i64 1, !dbg !27 + %362 = extractelement <16 x i1> %32, i64 5, !dbg !27 + %363 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %94, i1 %362) #4, !dbg !27 + %364 = extractvalue { i32, i32, i32, i32 } %363, 0, !dbg !27 + %365 = bitcast i32 %364 to <2 x bfloat>, !dbg !27 + %366 = extractvalue { i32, i32, i32, i32 } %363, 1, !dbg !27 + %367 = bitcast i32 %366 to <2 x bfloat>, !dbg !27 + %368 = extractvalue { i32, i32, i32, i32 } %363, 2, !dbg !27 + %369 = bitcast i32 %368 to <2 x bfloat>, !dbg !27 + %370 = extractvalue { i32, i32, i32, i32 } %363, 3, !dbg !27 + %371 = bitcast i32 %370 to <2 x bfloat>, !dbg !27 + %372 = extractelement <2 x bfloat> %365, i64 0, !dbg !27 + %373 = extractelement <2 x bfloat> %365, i64 1, !dbg !27 + %374 = extractelement <2 x bfloat> %367, i64 0, !dbg !27 + %375 = extractelement <2 x bfloat> %367, i64 1, !dbg !27 + %376 = extractelement <2 x bfloat> %369, i64 0, !dbg !27 + %377 = extractelement <2 x bfloat> %369, i64 1, !dbg !27 + %378 = extractelement <2 x bfloat> %371, i64 0, !dbg !27 + %379 = extractelement <2 x bfloat> %371, i64 1, !dbg !27 + %380 = extractelement <16 x i1> %32, i64 4, !dbg !27 + %381 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %95, i1 %380) #4, !dbg !27 + %382 = extractvalue { i32, i32, i32, i32 } %381, 0, !dbg !27 + %383 = bitcast i32 %382 to <2 x bfloat>, !dbg !27 + %384 = extractvalue { i32, i32, i32, i32 } %381, 1, !dbg !27 + %385 = bitcast i32 %384 to <2 x bfloat>, !dbg !27 + %386 = extractvalue { i32, i32, i32, i32 } %381, 2, !dbg !27 + %387 = bitcast i32 %386 to <2 x bfloat>, !dbg !27 + %388 = extractvalue { i32, i32, i32, i32 } %381, 3, !dbg !27 + %389 = bitcast i32 %388 to <2 x bfloat>, !dbg !27 + %390 = extractelement <2 x bfloat> %383, i64 0, !dbg !27 + %391 = extractelement <2 x bfloat> %383, i64 1, !dbg !27 + %392 = extractelement <2 x bfloat> %385, i64 0, !dbg !27 + %393 = extractelement <2 x bfloat> %385, i64 1, !dbg !27 + %394 = extractelement <2 x bfloat> %387, i64 0, !dbg !27 + %395 = extractelement <2 x bfloat> %387, i64 1, !dbg !27 + %396 = extractelement <2 x bfloat> %389, i64 0, !dbg !27 + %397 = extractelement <2 x bfloat> %389, i64 1, !dbg !27 + %398 = extractelement <16 x i1> %32, i64 3, !dbg !27 + %399 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %96, i1 %398) #4, !dbg !27 + %400 = extractvalue { i32, i32, i32, i32 } %399, 0, !dbg !27 + %401 = bitcast i32 %400 to <2 x bfloat>, !dbg !27 + %402 = extractvalue { i32, i32, i32, i32 } %399, 1, !dbg !27 + %403 = bitcast i32 %402 to <2 x bfloat>, !dbg !27 + %404 = extractvalue { i32, i32, i32, i32 } %399, 2, !dbg !27 + %405 = bitcast i32 %404 to <2 x bfloat>, !dbg !27 + %406 = extractvalue { i32, i32, i32, i32 } %399, 3, !dbg !27 + %407 = bitcast i32 %406 to <2 x bfloat>, !dbg !27 + %408 = extractelement <2 x bfloat> %401, i64 0, !dbg !27 + %409 = extractelement <2 x bfloat> %401, i64 1, !dbg !27 + %410 = extractelement <2 x bfloat> %403, i64 0, !dbg !27 + %411 = extractelement <2 x bfloat> %403, i64 1, !dbg !27 + %412 = extractelement <2 x bfloat> %405, i64 0, !dbg !27 + %413 = extractelement <2 x bfloat> %405, i64 1, !dbg !27 + %414 = extractelement <2 x bfloat> %407, i64 0, !dbg !27 + %415 = extractelement <2 x bfloat> %407, i64 1, !dbg !27 + %416 = extractelement <16 x i1> %32, i64 2, !dbg !27 + %417 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %97, i1 %416) #4, !dbg !27 + %418 = extractvalue { i32, i32, i32, i32 } %417, 0, !dbg !27 + %419 = bitcast i32 %418 to <2 x bfloat>, !dbg !27 + %420 = extractvalue { i32, i32, i32, i32 } %417, 1, !dbg !27 + %421 = bitcast i32 %420 to <2 x bfloat>, !dbg !27 + %422 = extractvalue { i32, i32, i32, i32 } %417, 2, !dbg !27 + %423 = bitcast i32 %422 to <2 x bfloat>, !dbg !27 + %424 = extractvalue { i32, i32, i32, i32 } %417, 3, !dbg !27 + %425 = bitcast i32 %424 to <2 x bfloat>, !dbg !27 + %426 = extractelement <2 x bfloat> %419, i64 0, !dbg !27 + %427 = extractelement <2 x bfloat> %419, i64 1, !dbg !27 + %428 = extractelement <2 x bfloat> %421, i64 0, !dbg !27 + %429 = extractelement <2 x bfloat> %421, i64 1, !dbg !27 + %430 = extractelement <2 x bfloat> %423, i64 0, !dbg !27 + %431 = extractelement <2 x bfloat> %423, i64 1, !dbg !27 + %432 = extractelement <2 x bfloat> %425, i64 0, !dbg !27 + %433 = extractelement <2 x bfloat> %425, i64 1, !dbg !27 + %434 = extractelement <16 x i1> %32, i64 1, !dbg !27 + %435 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %98, i1 %434) #4, !dbg !27 + %436 = extractvalue { i32, i32, i32, i32 } %435, 0, !dbg !27 + %437 = bitcast i32 %436 to <2 x bfloat>, !dbg !27 + %438 = extractvalue { i32, i32, i32, i32 } %435, 1, !dbg !27 + %439 = bitcast i32 %438 to <2 x bfloat>, !dbg !27 + %440 = extractvalue { i32, i32, i32, i32 } %435, 2, !dbg !27 + %441 = bitcast i32 %440 to <2 x bfloat>, !dbg !27 + %442 = extractvalue { i32, i32, i32, i32 } %435, 3, !dbg !27 + %443 = bitcast i32 %442 to <2 x bfloat>, !dbg !27 + %444 = extractelement <2 x bfloat> %437, i64 0, !dbg !27 + %445 = extractelement <2 x bfloat> %437, i64 1, !dbg !27 + %446 = extractelement <2 x bfloat> %439, i64 0, !dbg !27 + %447 = extractelement <2 x bfloat> %439, i64 1, !dbg !27 + %448 = extractelement <2 x bfloat> %441, i64 0, !dbg !27 + %449 = extractelement <2 x bfloat> %441, i64 1, !dbg !27 + %450 = extractelement <2 x bfloat> %443, i64 0, !dbg !27 + %451 = extractelement <2 x bfloat> %443, i64 1, !dbg !27 + %452 = fpext bfloat %174 to float, !dbg !28 + %453 = fpext bfloat %175 to float, !dbg !28 + %454 = fpext bfloat %176 to float, !dbg !28 + %455 = fpext bfloat %177 to float, !dbg !28 + %456 = fpext bfloat %178 to float, !dbg !28 + %457 = fpext bfloat %179 to float, !dbg !28 + %458 = fpext bfloat %180 to float, !dbg !28 + %459 = fpext bfloat %181 to float, !dbg !28 + %460 = fpext bfloat %192 to float, !dbg !28 + %461 = fpext bfloat %193 to float, !dbg !28 + %462 = fpext bfloat %194 to float, !dbg !28 + %463 = fpext bfloat %195 to float, !dbg !28 + %464 = fpext bfloat %196 to float, !dbg !28 + %465 = fpext bfloat %197 to float, !dbg !28 + %466 = fpext bfloat %198 to float, !dbg !28 + %467 = fpext bfloat %199 to float, !dbg !28 + %468 = fpext bfloat %210 to float, !dbg !28 + %469 = fpext bfloat %211 to float, !dbg !28 + %470 = fpext bfloat %212 to float, !dbg !28 + %471 = fpext bfloat %213 to float, !dbg !28 + %472 = fpext bfloat %214 to float, !dbg !28 + %473 = fpext bfloat %215 to float, !dbg !28 + %474 = fpext bfloat %216 to float, !dbg !28 + %475 = fpext bfloat %217 to float, !dbg !28 + %476 = fpext bfloat %228 to float, !dbg !28 + %477 = fpext bfloat %229 to float, !dbg !28 + %478 = fpext bfloat %230 to float, !dbg !28 + %479 = fpext bfloat %231 to float, !dbg !28 + %480 = fpext bfloat %232 to float, !dbg !28 + %481 = fpext bfloat %233 to float, !dbg !28 + %482 = fpext bfloat %234 to float, !dbg !28 + %483 = fpext bfloat %235 to float, !dbg !28 + %484 = fpext bfloat %246 to float, !dbg !28 + %485 = fpext bfloat %247 to float, !dbg !28 + %486 = fpext bfloat %248 to float, !dbg !28 + %487 = fpext bfloat %249 to float, !dbg !28 + %488 = fpext bfloat %250 to float, !dbg !28 + %489 = fpext bfloat %251 to float, !dbg !28 + %490 = fpext bfloat %252 to float, !dbg !28 + %491 = fpext bfloat %253 to float, !dbg !28 + %492 = fpext bfloat %264 to float, !dbg !28 + %493 = fpext bfloat %265 to float, !dbg !28 + %494 = fpext bfloat %266 to float, !dbg !28 + %495 = fpext bfloat %267 to float, !dbg !28 + %496 = fpext bfloat %268 to float, !dbg !28 + %497 = fpext bfloat %269 to float, !dbg !28 + %498 = fpext bfloat %270 to float, !dbg !28 + %499 = fpext bfloat %271 to float, !dbg !28 + %500 = fpext bfloat %282 to float, !dbg !28 + %501 = fpext bfloat %283 to float, !dbg !28 + %502 = fpext bfloat %284 to float, !dbg !28 + %503 = fpext bfloat %285 to float, !dbg !28 + %504 = fpext bfloat %286 to float, !dbg !28 + %505 = fpext bfloat %287 to float, !dbg !28 + %506 = fpext bfloat %288 to float, !dbg !28 + %507 = fpext bfloat %289 to float, !dbg !28 + %508 = fpext bfloat %300 to float, !dbg !28 + %509 = fpext bfloat %301 to float, !dbg !28 + %510 = fpext bfloat %302 to float, !dbg !28 + %511 = fpext bfloat %303 to float, !dbg !28 + %512 = fpext bfloat %304 to float, !dbg !28 + %513 = fpext bfloat %305 to float, !dbg !28 + %514 = fpext bfloat %306 to float, !dbg !28 + %515 = fpext bfloat %307 to float, !dbg !28 + %516 = fpext bfloat %318 to float, !dbg !28 + %517 = fpext bfloat %319 to float, !dbg !28 + %518 = fpext bfloat %320 to float, !dbg !28 + %519 = fpext bfloat %321 to float, !dbg !28 + %520 = fpext bfloat %322 to float, !dbg !28 + %521 = fpext bfloat %323 to float, !dbg !28 + %522 = fpext bfloat %324 to float, !dbg !28 + %523 = fpext bfloat %325 to float, !dbg !28 + %524 = fpext bfloat %336 to float, !dbg !28 + %525 = fpext bfloat %337 to float, !dbg !28 + %526 = fpext bfloat %338 to float, !dbg !28 + %527 = fpext bfloat %339 to float, !dbg !28 + %528 = fpext bfloat %340 to float, !dbg !28 + %529 = fpext bfloat %341 to float, !dbg !28 + %530 = fpext bfloat %342 to float, !dbg !28 + %531 = fpext bfloat %343 to float, !dbg !28 + %532 = fpext bfloat %354 to float, !dbg !28 + %533 = fpext bfloat %355 to float, !dbg !28 + %534 = fpext bfloat %356 to float, !dbg !28 + %535 = fpext bfloat %357 to float, !dbg !28 + %536 = fpext bfloat %358 to float, !dbg !28 + %537 = fpext bfloat %359 to float, !dbg !28 + %538 = fpext bfloat %360 to float, !dbg !28 + %539 = fpext bfloat %361 to float, !dbg !28 + %540 = fpext bfloat %372 to float, !dbg !28 + %541 = fpext bfloat %373 to float, !dbg !28 + %542 = fpext bfloat %374 to float, !dbg !28 + %543 = fpext bfloat %375 to float, !dbg !28 + %544 = fpext bfloat %376 to float, !dbg !28 + %545 = fpext bfloat %377 to float, !dbg !28 + %546 = fpext bfloat %378 to float, !dbg !28 + %547 = fpext bfloat %379 to float, !dbg !28 + %548 = fpext bfloat %390 to float, !dbg !28 + %549 = fpext bfloat %391 to float, !dbg !28 + %550 = fpext bfloat %392 to float, !dbg !28 + %551 = fpext bfloat %393 to float, !dbg !28 + %552 = fpext bfloat %394 to float, !dbg !28 + %553 = fpext bfloat %395 to float, !dbg !28 + %554 = fpext bfloat %396 to float, !dbg !28 + %555 = fpext bfloat %397 to float, !dbg !28 + %556 = fpext bfloat %408 to float, !dbg !28 + %557 = fpext bfloat %409 to float, !dbg !28 + %558 = fpext bfloat %410 to float, !dbg !28 + %559 = fpext bfloat %411 to float, !dbg !28 + %560 = fpext bfloat %412 to float, !dbg !28 + %561 = fpext bfloat %413 to float, !dbg !28 + %562 = fpext bfloat %414 to float, !dbg !28 + %563 = fpext bfloat %415 to float, !dbg !28 + %564 = fpext bfloat %426 to float, !dbg !28 + %565 = fpext bfloat %427 to float, !dbg !28 + %566 = fpext bfloat %428 to float, !dbg !28 + %567 = fpext bfloat %429 to float, !dbg !28 + %568 = fpext bfloat %430 to float, !dbg !28 + %569 = fpext bfloat %431 to float, !dbg !28 + %570 = fpext bfloat %432 to float, !dbg !28 + %571 = fpext bfloat %433 to float, !dbg !28 + %572 = fpext bfloat %444 to float, !dbg !28 + %573 = fpext bfloat %445 to float, !dbg !28 + %574 = fpext bfloat %446 to float, !dbg !28 + %575 = fpext bfloat %447 to float, !dbg !28 + %576 = fpext bfloat %448 to float, !dbg !28 + %577 = fpext bfloat %449 to float, !dbg !28 + %578 = fpext bfloat %450 to float, !dbg !28 + %579 = fpext bfloat %451 to float, !dbg !28 + %580 = extractelement <16 x i1> %32, i64 0, !dbg !29 + %581 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %100, i1 %580) #4, !dbg !29 + %582 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %117, i1 %28) #4, !dbg !30 + %583 = extractvalue { i32, i32, i32, i32 } %582, 0, !dbg !30 + %584 = bitcast i32 %583 to <2 x bfloat>, !dbg !30 + %585 = extractvalue { i32, i32, i32, i32 } %582, 1, !dbg !30 + %586 = bitcast i32 %585 to <2 x bfloat>, !dbg !30 + %587 = extractvalue { i32, i32, i32, i32 } %582, 2, !dbg !30 + %588 = bitcast i32 %587 to <2 x bfloat>, !dbg !30 + %589 = extractvalue { i32, i32, i32, i32 } %582, 3, !dbg !30 + %590 = bitcast i32 %589 to <2 x bfloat>, !dbg !30 + %591 = extractelement <2 x bfloat> %584, i64 0, !dbg !30 + %592 = extractelement <2 x bfloat> %584, i64 1, !dbg !30 + %593 = extractelement <2 x bfloat> %586, i64 0, !dbg !30 + %594 = extractelement <2 x bfloat> %586, i64 1, !dbg !30 + %595 = extractelement <2 x bfloat> %588, i64 0, !dbg !30 + %596 = extractelement <2 x bfloat> %588, i64 1, !dbg !30 + %597 = extractelement <2 x bfloat> %590, i64 0, !dbg !30 + %598 = extractelement <2 x bfloat> %590, i64 1, !dbg !30 + %599 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %118, i1 %182) #4, !dbg !30 + %600 = extractvalue { i32, i32, i32, i32 } %599, 0, !dbg !30 + %601 = bitcast i32 %600 to <2 x bfloat>, !dbg !30 + %602 = extractvalue { i32, i32, i32, i32 } %599, 1, !dbg !30 + %603 = bitcast i32 %602 to <2 x bfloat>, !dbg !30 + %604 = extractvalue { i32, i32, i32, i32 } %599, 2, !dbg !30 + %605 = bitcast i32 %604 to <2 x bfloat>, !dbg !30 + %606 = extractvalue { i32, i32, i32, i32 } %599, 3, !dbg !30 + %607 = bitcast i32 %606 to <2 x bfloat>, !dbg !30 + %608 = extractelement <2 x bfloat> %601, i64 0, !dbg !30 + %609 = extractelement <2 x bfloat> %601, i64 1, !dbg !30 + %610 = extractelement <2 x bfloat> %603, i64 0, !dbg !30 + %611 = extractelement <2 x bfloat> %603, i64 1, !dbg !30 + %612 = extractelement <2 x bfloat> %605, i64 0, !dbg !30 + %613 = extractelement <2 x bfloat> %605, i64 1, !dbg !30 + %614 = extractelement <2 x bfloat> %607, i64 0, !dbg !30 + %615 = extractelement <2 x bfloat> %607, i64 1, !dbg !30 + %616 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %119, i1 %200) #4, !dbg !30 + %617 = extractvalue { i32, i32, i32, i32 } %616, 0, !dbg !30 + %618 = bitcast i32 %617 to <2 x bfloat>, !dbg !30 + %619 = extractvalue { i32, i32, i32, i32 } %616, 1, !dbg !30 + %620 = bitcast i32 %619 to <2 x bfloat>, !dbg !30 + %621 = extractvalue { i32, i32, i32, i32 } %616, 2, !dbg !30 + %622 = bitcast i32 %621 to <2 x bfloat>, !dbg !30 + %623 = extractvalue { i32, i32, i32, i32 } %616, 3, !dbg !30 + %624 = bitcast i32 %623 to <2 x bfloat>, !dbg !30 + %625 = extractelement <2 x bfloat> %618, i64 0, !dbg !30 + %626 = extractelement <2 x bfloat> %618, i64 1, !dbg !30 + %627 = extractelement <2 x bfloat> %620, i64 0, !dbg !30 + %628 = extractelement <2 x bfloat> %620, i64 1, !dbg !30 + %629 = extractelement <2 x bfloat> %622, i64 0, !dbg !30 + %630 = extractelement <2 x bfloat> %622, i64 1, !dbg !30 + %631 = extractelement <2 x bfloat> %624, i64 0, !dbg !30 + %632 = extractelement <2 x bfloat> %624, i64 1, !dbg !30 + %633 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %120, i1 %218) #4, !dbg !30 + %634 = extractvalue { i32, i32, i32, i32 } %633, 0, !dbg !30 + %635 = bitcast i32 %634 to <2 x bfloat>, !dbg !30 + %636 = extractvalue { i32, i32, i32, i32 } %633, 1, !dbg !30 + %637 = bitcast i32 %636 to <2 x bfloat>, !dbg !30 + %638 = extractvalue { i32, i32, i32, i32 } %633, 2, !dbg !30 + %639 = bitcast i32 %638 to <2 x bfloat>, !dbg !30 + %640 = extractvalue { i32, i32, i32, i32 } %633, 3, !dbg !30 + %641 = bitcast i32 %640 to <2 x bfloat>, !dbg !30 + %642 = extractelement <2 x bfloat> %635, i64 0, !dbg !30 + %643 = extractelement <2 x bfloat> %635, i64 1, !dbg !30 + %644 = extractelement <2 x bfloat> %637, i64 0, !dbg !30 + %645 = extractelement <2 x bfloat> %637, i64 1, !dbg !30 + %646 = extractelement <2 x bfloat> %639, i64 0, !dbg !30 + %647 = extractelement <2 x bfloat> %639, i64 1, !dbg !30 + %648 = extractelement <2 x bfloat> %641, i64 0, !dbg !30 + %649 = extractelement <2 x bfloat> %641, i64 1, !dbg !30 + %650 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %121, i1 %236) #4, !dbg !30 + %651 = extractvalue { i32, i32, i32, i32 } %650, 0, !dbg !30 + %652 = bitcast i32 %651 to <2 x bfloat>, !dbg !30 + %653 = extractvalue { i32, i32, i32, i32 } %650, 1, !dbg !30 + %654 = bitcast i32 %653 to <2 x bfloat>, !dbg !30 + %655 = extractvalue { i32, i32, i32, i32 } %650, 2, !dbg !30 + %656 = bitcast i32 %655 to <2 x bfloat>, !dbg !30 + %657 = extractvalue { i32, i32, i32, i32 } %650, 3, !dbg !30 + %658 = bitcast i32 %657 to <2 x bfloat>, !dbg !30 + %659 = extractelement <2 x bfloat> %652, i64 0, !dbg !30 + %660 = extractelement <2 x bfloat> %652, i64 1, !dbg !30 + %661 = extractelement <2 x bfloat> %654, i64 0, !dbg !30 + %662 = extractelement <2 x bfloat> %654, i64 1, !dbg !30 + %663 = extractelement <2 x bfloat> %656, i64 0, !dbg !30 + %664 = extractelement <2 x bfloat> %656, i64 1, !dbg !30 + %665 = extractelement <2 x bfloat> %658, i64 0, !dbg !30 + %666 = extractelement <2 x bfloat> %658, i64 1, !dbg !30 + %667 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %122, i1 %254) #4, !dbg !30 + %668 = extractvalue { i32, i32, i32, i32 } %667, 0, !dbg !30 + %669 = bitcast i32 %668 to <2 x bfloat>, !dbg !30 + %670 = extractvalue { i32, i32, i32, i32 } %667, 1, !dbg !30 + %671 = bitcast i32 %670 to <2 x bfloat>, !dbg !30 + %672 = extractvalue { i32, i32, i32, i32 } %667, 2, !dbg !30 + %673 = bitcast i32 %672 to <2 x bfloat>, !dbg !30 + %674 = extractvalue { i32, i32, i32, i32 } %667, 3, !dbg !30 + %675 = bitcast i32 %674 to <2 x bfloat>, !dbg !30 + %676 = extractelement <2 x bfloat> %669, i64 0, !dbg !30 + %677 = extractelement <2 x bfloat> %669, i64 1, !dbg !30 + %678 = extractelement <2 x bfloat> %671, i64 0, !dbg !30 + %679 = extractelement <2 x bfloat> %671, i64 1, !dbg !30 + %680 = extractelement <2 x bfloat> %673, i64 0, !dbg !30 + %681 = extractelement <2 x bfloat> %673, i64 1, !dbg !30 + %682 = extractelement <2 x bfloat> %675, i64 0, !dbg !30 + %683 = extractelement <2 x bfloat> %675, i64 1, !dbg !30 + %684 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %123, i1 %272) #4, !dbg !30 + %685 = extractvalue { i32, i32, i32, i32 } %684, 0, !dbg !30 + %686 = bitcast i32 %685 to <2 x bfloat>, !dbg !30 + %687 = extractvalue { i32, i32, i32, i32 } %684, 1, !dbg !30 + %688 = bitcast i32 %687 to <2 x bfloat>, !dbg !30 + %689 = extractvalue { i32, i32, i32, i32 } %684, 2, !dbg !30 + %690 = bitcast i32 %689 to <2 x bfloat>, !dbg !30 + %691 = extractvalue { i32, i32, i32, i32 } %684, 3, !dbg !30 + %692 = bitcast i32 %691 to <2 x bfloat>, !dbg !30 + %693 = extractelement <2 x bfloat> %686, i64 0, !dbg !30 + %694 = extractelement <2 x bfloat> %686, i64 1, !dbg !30 + %695 = extractelement <2 x bfloat> %688, i64 0, !dbg !30 + %696 = extractelement <2 x bfloat> %688, i64 1, !dbg !30 + %697 = extractelement <2 x bfloat> %690, i64 0, !dbg !30 + %698 = extractelement <2 x bfloat> %690, i64 1, !dbg !30 + %699 = extractelement <2 x bfloat> %692, i64 0, !dbg !30 + %700 = extractelement <2 x bfloat> %692, i64 1, !dbg !30 + %701 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %124, i1 %290) #4, !dbg !30 + %702 = extractvalue { i32, i32, i32, i32 } %701, 0, !dbg !30 + %703 = bitcast i32 %702 to <2 x bfloat>, !dbg !30 + %704 = extractvalue { i32, i32, i32, i32 } %701, 1, !dbg !30 + %705 = bitcast i32 %704 to <2 x bfloat>, !dbg !30 + %706 = extractvalue { i32, i32, i32, i32 } %701, 2, !dbg !30 + %707 = bitcast i32 %706 to <2 x bfloat>, !dbg !30 + %708 = extractvalue { i32, i32, i32, i32 } %701, 3, !dbg !30 + %709 = bitcast i32 %708 to <2 x bfloat>, !dbg !30 + %710 = extractelement <2 x bfloat> %703, i64 0, !dbg !30 + %711 = extractelement <2 x bfloat> %703, i64 1, !dbg !30 + %712 = extractelement <2 x bfloat> %705, i64 0, !dbg !30 + %713 = extractelement <2 x bfloat> %705, i64 1, !dbg !30 + %714 = extractelement <2 x bfloat> %707, i64 0, !dbg !30 + %715 = extractelement <2 x bfloat> %707, i64 1, !dbg !30 + %716 = extractelement <2 x bfloat> %709, i64 0, !dbg !30 + %717 = extractelement <2 x bfloat> %709, i64 1, !dbg !30 + %718 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %125, i1 %308) #4, !dbg !30 + %719 = extractvalue { i32, i32, i32, i32 } %718, 0, !dbg !30 + %720 = bitcast i32 %719 to <2 x bfloat>, !dbg !30 + %721 = extractvalue { i32, i32, i32, i32 } %718, 1, !dbg !30 + %722 = bitcast i32 %721 to <2 x bfloat>, !dbg !30 + %723 = extractvalue { i32, i32, i32, i32 } %718, 2, !dbg !30 + %724 = bitcast i32 %723 to <2 x bfloat>, !dbg !30 + %725 = extractvalue { i32, i32, i32, i32 } %718, 3, !dbg !30 + %726 = bitcast i32 %725 to <2 x bfloat>, !dbg !30 + %727 = extractelement <2 x bfloat> %720, i64 0, !dbg !30 + %728 = extractelement <2 x bfloat> %720, i64 1, !dbg !30 + %729 = extractelement <2 x bfloat> %722, i64 0, !dbg !30 + %730 = extractelement <2 x bfloat> %722, i64 1, !dbg !30 + %731 = extractelement <2 x bfloat> %724, i64 0, !dbg !30 + %732 = extractelement <2 x bfloat> %724, i64 1, !dbg !30 + %733 = extractelement <2 x bfloat> %726, i64 0, !dbg !30 + %734 = extractelement <2 x bfloat> %726, i64 1, !dbg !30 + %735 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %126, i1 %326) #4, !dbg !30 + %736 = extractvalue { i32, i32, i32, i32 } %735, 0, !dbg !30 + %737 = bitcast i32 %736 to <2 x bfloat>, !dbg !30 + %738 = extractvalue { i32, i32, i32, i32 } %735, 1, !dbg !30 + %739 = bitcast i32 %738 to <2 x bfloat>, !dbg !30 + %740 = extractvalue { i32, i32, i32, i32 } %735, 2, !dbg !30 + %741 = bitcast i32 %740 to <2 x bfloat>, !dbg !30 + %742 = extractvalue { i32, i32, i32, i32 } %735, 3, !dbg !30 + %743 = bitcast i32 %742 to <2 x bfloat>, !dbg !30 + %744 = extractelement <2 x bfloat> %737, i64 0, !dbg !30 + %745 = extractelement <2 x bfloat> %737, i64 1, !dbg !30 + %746 = extractelement <2 x bfloat> %739, i64 0, !dbg !30 + %747 = extractelement <2 x bfloat> %739, i64 1, !dbg !30 + %748 = extractelement <2 x bfloat> %741, i64 0, !dbg !30 + %749 = extractelement <2 x bfloat> %741, i64 1, !dbg !30 + %750 = extractelement <2 x bfloat> %743, i64 0, !dbg !30 + %751 = extractelement <2 x bfloat> %743, i64 1, !dbg !30 + %752 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %127, i1 %344) #4, !dbg !30 + %753 = extractvalue { i32, i32, i32, i32 } %752, 0, !dbg !30 + %754 = bitcast i32 %753 to <2 x bfloat>, !dbg !30 + %755 = extractvalue { i32, i32, i32, i32 } %752, 1, !dbg !30 + %756 = bitcast i32 %755 to <2 x bfloat>, !dbg !30 + %757 = extractvalue { i32, i32, i32, i32 } %752, 2, !dbg !30 + %758 = bitcast i32 %757 to <2 x bfloat>, !dbg !30 + %759 = extractvalue { i32, i32, i32, i32 } %752, 3, !dbg !30 + %760 = bitcast i32 %759 to <2 x bfloat>, !dbg !30 + %761 = extractelement <2 x bfloat> %754, i64 0, !dbg !30 + %762 = extractelement <2 x bfloat> %754, i64 1, !dbg !30 + %763 = extractelement <2 x bfloat> %756, i64 0, !dbg !30 + %764 = extractelement <2 x bfloat> %756, i64 1, !dbg !30 + %765 = extractelement <2 x bfloat> %758, i64 0, !dbg !30 + %766 = extractelement <2 x bfloat> %758, i64 1, !dbg !30 + %767 = extractelement <2 x bfloat> %760, i64 0, !dbg !30 + %768 = extractelement <2 x bfloat> %760, i64 1, !dbg !30 + %769 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %128, i1 %362) #4, !dbg !30 + %770 = extractvalue { i32, i32, i32, i32 } %769, 0, !dbg !30 + %771 = bitcast i32 %770 to <2 x bfloat>, !dbg !30 + %772 = extractvalue { i32, i32, i32, i32 } %769, 1, !dbg !30 + %773 = bitcast i32 %772 to <2 x bfloat>, !dbg !30 + %774 = extractvalue { i32, i32, i32, i32 } %769, 2, !dbg !30 + %775 = bitcast i32 %774 to <2 x bfloat>, !dbg !30 + %776 = extractvalue { i32, i32, i32, i32 } %769, 3, !dbg !30 + %777 = bitcast i32 %776 to <2 x bfloat>, !dbg !30 + %778 = extractelement <2 x bfloat> %771, i64 0, !dbg !30 + %779 = extractelement <2 x bfloat> %771, i64 1, !dbg !30 + %780 = extractelement <2 x bfloat> %773, i64 0, !dbg !30 + %781 = extractelement <2 x bfloat> %773, i64 1, !dbg !30 + %782 = extractelement <2 x bfloat> %775, i64 0, !dbg !30 + %783 = extractelement <2 x bfloat> %775, i64 1, !dbg !30 + %784 = extractelement <2 x bfloat> %777, i64 0, !dbg !30 + %785 = extractelement <2 x bfloat> %777, i64 1, !dbg !30 + %786 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %129, i1 %380) #4, !dbg !30 + %787 = extractvalue { i32, i32, i32, i32 } %786, 0, !dbg !30 + %788 = bitcast i32 %787 to <2 x bfloat>, !dbg !30 + %789 = extractvalue { i32, i32, i32, i32 } %786, 1, !dbg !30 + %790 = bitcast i32 %789 to <2 x bfloat>, !dbg !30 + %791 = extractvalue { i32, i32, i32, i32 } %786, 2, !dbg !30 + %792 = bitcast i32 %791 to <2 x bfloat>, !dbg !30 + %793 = extractvalue { i32, i32, i32, i32 } %786, 3, !dbg !30 + %794 = bitcast i32 %793 to <2 x bfloat>, !dbg !30 + %795 = extractelement <2 x bfloat> %788, i64 0, !dbg !30 + %796 = extractelement <2 x bfloat> %788, i64 1, !dbg !30 + %797 = extractelement <2 x bfloat> %790, i64 0, !dbg !30 + %798 = extractelement <2 x bfloat> %790, i64 1, !dbg !30 + %799 = extractelement <2 x bfloat> %792, i64 0, !dbg !30 + %800 = extractelement <2 x bfloat> %792, i64 1, !dbg !30 + %801 = extractelement <2 x bfloat> %794, i64 0, !dbg !30 + %802 = extractelement <2 x bfloat> %794, i64 1, !dbg !30 + %803 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %130, i1 %398) #4, !dbg !30 + %804 = extractvalue { i32, i32, i32, i32 } %803, 0, !dbg !30 + %805 = bitcast i32 %804 to <2 x bfloat>, !dbg !30 + %806 = extractvalue { i32, i32, i32, i32 } %803, 1, !dbg !30 + %807 = bitcast i32 %806 to <2 x bfloat>, !dbg !30 + %808 = extractvalue { i32, i32, i32, i32 } %803, 2, !dbg !30 + %809 = bitcast i32 %808 to <2 x bfloat>, !dbg !30 + %810 = extractvalue { i32, i32, i32, i32 } %803, 3, !dbg !30 + %811 = bitcast i32 %810 to <2 x bfloat>, !dbg !30 + %812 = extractelement <2 x bfloat> %805, i64 0, !dbg !30 + %813 = extractelement <2 x bfloat> %805, i64 1, !dbg !30 + %814 = extractelement <2 x bfloat> %807, i64 0, !dbg !30 + %815 = extractelement <2 x bfloat> %807, i64 1, !dbg !30 + %816 = extractelement <2 x bfloat> %809, i64 0, !dbg !30 + %817 = extractelement <2 x bfloat> %809, i64 1, !dbg !30 + %818 = extractelement <2 x bfloat> %811, i64 0, !dbg !30 + %819 = extractelement <2 x bfloat> %811, i64 1, !dbg !30 + %820 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %131, i1 %416) #4, !dbg !30 + %821 = extractvalue { i32, i32, i32, i32 } %820, 0, !dbg !30 + %822 = bitcast i32 %821 to <2 x bfloat>, !dbg !30 + %823 = extractvalue { i32, i32, i32, i32 } %820, 1, !dbg !30 + %824 = bitcast i32 %823 to <2 x bfloat>, !dbg !30 + %825 = extractvalue { i32, i32, i32, i32 } %820, 2, !dbg !30 + %826 = bitcast i32 %825 to <2 x bfloat>, !dbg !30 + %827 = extractvalue { i32, i32, i32, i32 } %820, 3, !dbg !30 + %828 = bitcast i32 %827 to <2 x bfloat>, !dbg !30 + %829 = extractelement <2 x bfloat> %822, i64 0, !dbg !30 + %830 = extractelement <2 x bfloat> %822, i64 1, !dbg !30 + %831 = extractelement <2 x bfloat> %824, i64 0, !dbg !30 + %832 = extractelement <2 x bfloat> %824, i64 1, !dbg !30 + %833 = extractelement <2 x bfloat> %826, i64 0, !dbg !30 + %834 = extractelement <2 x bfloat> %826, i64 1, !dbg !30 + %835 = extractelement <2 x bfloat> %828, i64 0, !dbg !30 + %836 = extractelement <2 x bfloat> %828, i64 1, !dbg !30 + %837 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %132, i1 %434) #4, !dbg !30 + %838 = extractvalue { i32, i32, i32, i32 } %837, 0, !dbg !30 + %839 = bitcast i32 %838 to <2 x bfloat>, !dbg !30 + %840 = extractvalue { i32, i32, i32, i32 } %837, 1, !dbg !30 + %841 = bitcast i32 %840 to <2 x bfloat>, !dbg !30 + %842 = extractvalue { i32, i32, i32, i32 } %837, 2, !dbg !30 + %843 = bitcast i32 %842 to <2 x bfloat>, !dbg !30 + %844 = extractvalue { i32, i32, i32, i32 } %837, 3, !dbg !30 + %845 = bitcast i32 %844 to <2 x bfloat>, !dbg !30 + %846 = extractelement <2 x bfloat> %839, i64 0, !dbg !30 + %847 = extractelement <2 x bfloat> %839, i64 1, !dbg !30 + %848 = extractelement <2 x bfloat> %841, i64 0, !dbg !30 + %849 = extractelement <2 x bfloat> %841, i64 1, !dbg !30 + %850 = extractelement <2 x bfloat> %843, i64 0, !dbg !30 + %851 = extractelement <2 x bfloat> %843, i64 1, !dbg !30 + %852 = extractelement <2 x bfloat> %845, i64 0, !dbg !30 + %853 = extractelement <2 x bfloat> %845, i64 1, !dbg !30 + %854 = fpext bfloat %591 to float, !dbg !31 + %855 = fpext bfloat %592 to float, !dbg !31 + %856 = fpext bfloat %593 to float, !dbg !31 + %857 = fpext bfloat %594 to float, !dbg !31 + %858 = fpext bfloat %595 to float, !dbg !31 + %859 = fpext bfloat %596 to float, !dbg !31 + %860 = fpext bfloat %597 to float, !dbg !31 + %861 = fpext bfloat %598 to float, !dbg !31 + %862 = fpext bfloat %608 to float, !dbg !31 + %863 = fpext bfloat %609 to float, !dbg !31 + %864 = fpext bfloat %610 to float, !dbg !31 + %865 = fpext bfloat %611 to float, !dbg !31 + %866 = fpext bfloat %612 to float, !dbg !31 + %867 = fpext bfloat %613 to float, !dbg !31 + %868 = fpext bfloat %614 to float, !dbg !31 + %869 = fpext bfloat %615 to float, !dbg !31 + %870 = fpext bfloat %625 to float, !dbg !31 + %871 = fpext bfloat %626 to float, !dbg !31 + %872 = fpext bfloat %627 to float, !dbg !31 + %873 = fpext bfloat %628 to float, !dbg !31 + %874 = fpext bfloat %629 to float, !dbg !31 + %875 = fpext bfloat %630 to float, !dbg !31 + %876 = fpext bfloat %631 to float, !dbg !31 + %877 = fpext bfloat %632 to float, !dbg !31 + %878 = fpext bfloat %642 to float, !dbg !31 + %879 = fpext bfloat %643 to float, !dbg !31 + %880 = fpext bfloat %644 to float, !dbg !31 + %881 = fpext bfloat %645 to float, !dbg !31 + %882 = fpext bfloat %646 to float, !dbg !31 + %883 = fpext bfloat %647 to float, !dbg !31 + %884 = fpext bfloat %648 to float, !dbg !31 + %885 = fpext bfloat %649 to float, !dbg !31 + %886 = fpext bfloat %659 to float, !dbg !31 + %887 = fpext bfloat %660 to float, !dbg !31 + %888 = fpext bfloat %661 to float, !dbg !31 + %889 = fpext bfloat %662 to float, !dbg !31 + %890 = fpext bfloat %663 to float, !dbg !31 + %891 = fpext bfloat %664 to float, !dbg !31 + %892 = fpext bfloat %665 to float, !dbg !31 + %893 = fpext bfloat %666 to float, !dbg !31 + %894 = fpext bfloat %676 to float, !dbg !31 + %895 = fpext bfloat %677 to float, !dbg !31 + %896 = fpext bfloat %678 to float, !dbg !31 + %897 = fpext bfloat %679 to float, !dbg !31 + %898 = fpext bfloat %680 to float, !dbg !31 + %899 = fpext bfloat %681 to float, !dbg !31 + %900 = fpext bfloat %682 to float, !dbg !31 + %901 = fpext bfloat %683 to float, !dbg !31 + %902 = fpext bfloat %693 to float, !dbg !31 + %903 = fpext bfloat %694 to float, !dbg !31 + %904 = fpext bfloat %695 to float, !dbg !31 + %905 = fpext bfloat %696 to float, !dbg !31 + %906 = fpext bfloat %697 to float, !dbg !31 + %907 = fpext bfloat %698 to float, !dbg !31 + %908 = fpext bfloat %699 to float, !dbg !31 + %909 = fpext bfloat %700 to float, !dbg !31 + %910 = fpext bfloat %710 to float, !dbg !31 + %911 = fpext bfloat %711 to float, !dbg !31 + %912 = fpext bfloat %712 to float, !dbg !31 + %913 = fpext bfloat %713 to float, !dbg !31 + %914 = fpext bfloat %714 to float, !dbg !31 + %915 = fpext bfloat %715 to float, !dbg !31 + %916 = fpext bfloat %716 to float, !dbg !31 + %917 = fpext bfloat %717 to float, !dbg !31 + %918 = fpext bfloat %727 to float, !dbg !31 + %919 = fpext bfloat %728 to float, !dbg !31 + %920 = fpext bfloat %729 to float, !dbg !31 + %921 = fpext bfloat %730 to float, !dbg !31 + %922 = fpext bfloat %731 to float, !dbg !31 + %923 = fpext bfloat %732 to float, !dbg !31 + %924 = fpext bfloat %733 to float, !dbg !31 + %925 = fpext bfloat %734 to float, !dbg !31 + %926 = fpext bfloat %744 to float, !dbg !31 + %927 = fpext bfloat %745 to float, !dbg !31 + %928 = fpext bfloat %746 to float, !dbg !31 + %929 = fpext bfloat %747 to float, !dbg !31 + %930 = fpext bfloat %748 to float, !dbg !31 + %931 = fpext bfloat %749 to float, !dbg !31 + %932 = fpext bfloat %750 to float, !dbg !31 + %933 = fpext bfloat %751 to float, !dbg !31 + %934 = fpext bfloat %761 to float, !dbg !31 + %935 = fpext bfloat %762 to float, !dbg !31 + %936 = fpext bfloat %763 to float, !dbg !31 + %937 = fpext bfloat %764 to float, !dbg !31 + %938 = fpext bfloat %765 to float, !dbg !31 + %939 = fpext bfloat %766 to float, !dbg !31 + %940 = fpext bfloat %767 to float, !dbg !31 + %941 = fpext bfloat %768 to float, !dbg !31 + %942 = fpext bfloat %778 to float, !dbg !31 + %943 = fpext bfloat %779 to float, !dbg !31 + %944 = fpext bfloat %780 to float, !dbg !31 + %945 = fpext bfloat %781 to float, !dbg !31 + %946 = fpext bfloat %782 to float, !dbg !31 + %947 = fpext bfloat %783 to float, !dbg !31 + %948 = fpext bfloat %784 to float, !dbg !31 + %949 = fpext bfloat %785 to float, !dbg !31 + %950 = fpext bfloat %795 to float, !dbg !31 + %951 = fpext bfloat %796 to float, !dbg !31 + %952 = fpext bfloat %797 to float, !dbg !31 + %953 = fpext bfloat %798 to float, !dbg !31 + %954 = fpext bfloat %799 to float, !dbg !31 + %955 = fpext bfloat %800 to float, !dbg !31 + %956 = fpext bfloat %801 to float, !dbg !31 + %957 = fpext bfloat %802 to float, !dbg !31 + %958 = fpext bfloat %812 to float, !dbg !31 + %959 = fpext bfloat %813 to float, !dbg !31 + %960 = fpext bfloat %814 to float, !dbg !31 + %961 = fpext bfloat %815 to float, !dbg !31 + %962 = fpext bfloat %816 to float, !dbg !31 + %963 = fpext bfloat %817 to float, !dbg !31 + %964 = fpext bfloat %818 to float, !dbg !31 + %965 = fpext bfloat %819 to float, !dbg !31 + %966 = fpext bfloat %829 to float, !dbg !31 + %967 = fpext bfloat %830 to float, !dbg !31 + %968 = fpext bfloat %831 to float, !dbg !31 + %969 = fpext bfloat %832 to float, !dbg !31 + %970 = fpext bfloat %833 to float, !dbg !31 + %971 = fpext bfloat %834 to float, !dbg !31 + %972 = fpext bfloat %835 to float, !dbg !31 + %973 = fpext bfloat %836 to float, !dbg !31 + %974 = fpext bfloat %846 to float, !dbg !31 + %975 = fpext bfloat %847 to float, !dbg !31 + %976 = fpext bfloat %848 to float, !dbg !31 + %977 = fpext bfloat %849 to float, !dbg !31 + %978 = fpext bfloat %850 to float, !dbg !31 + %979 = fpext bfloat %851 to float, !dbg !31 + %980 = fpext bfloat %852 to float, !dbg !31 + %981 = fpext bfloat %853 to float, !dbg !31 + %982 = shl nuw nsw i32 %13, 5, !dbg !32 + %983 = and i32 %982, 96, !dbg !32 + %984 = and i32 %13, 28, !dbg !32 + %985 = shl nuw nsw i32 %13, 2, !dbg !32 + %986 = and i32 %985, 128, !dbg !32 + %987 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %983, !dbg !32 + %988 = getelementptr inbounds nuw i8, ptr addrspace(3) %987, i32 %986, !dbg !32 + %989 = getelementptr inbounds nuw i8, ptr addrspace(3) %988, i32 %984, !dbg !32 + %990 = insertelement <1 x i32> poison, i32 %581, i64 0, !dbg !32 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) %989, <1 x i32> %990) #4, !dbg !32 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !32 + %991 = shl nuw nsw i32 %14, 1, !dbg !32 + %992 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %991, !dbg !32 + %993 = load float, ptr addrspace(3) %992, align 16, !dbg !32 + %994 = getelementptr inbounds nuw i8, ptr addrspace(3) %992, i32 4, !dbg !32 + %995 = load float, ptr addrspace(3) %994, align 4, !dbg !32 + %996 = getelementptr inbounds nuw i8, ptr addrspace(3) %992, i32 8, !dbg !32 + %997 = load float, ptr addrspace(3) %996, align 8, !dbg !32 + %998 = getelementptr inbounds nuw i8, ptr addrspace(3) %992, i32 12, !dbg !32 + %999 = load float, ptr addrspace(3) %998, align 4, !dbg !32 + %1000 = getelementptr inbounds nuw i8, ptr addrspace(3) %992, i32 16, !dbg !32 + %1001 = load float, ptr addrspace(3) %1000, align 16, !dbg !32 + %1002 = getelementptr inbounds nuw i8, ptr addrspace(3) %992, i32 20, !dbg !32 + %1003 = load float, ptr addrspace(3) %1002, align 4, !dbg !32 + %1004 = getelementptr inbounds nuw i8, ptr addrspace(3) %992, i32 24, !dbg !32 + %1005 = load float, ptr addrspace(3) %1004, align 8, !dbg !32 + %1006 = getelementptr inbounds nuw i8, ptr addrspace(3) %992, i32 28, !dbg !32 + %1007 = load float, ptr addrspace(3) %1006, align 4, !dbg !32 + %1008 = getelementptr inbounds nuw i8, ptr addrspace(3) %992, i32 128, !dbg !32 + %1009 = load float, ptr addrspace(3) %1008, align 16, !dbg !32 + %1010 = getelementptr inbounds nuw i8, ptr addrspace(3) %992, i32 132, !dbg !32 + %1011 = load float, ptr addrspace(3) %1010, align 4, !dbg !32 + %1012 = getelementptr inbounds nuw i8, ptr addrspace(3) %992, i32 136, !dbg !32 + %1013 = load float, ptr addrspace(3) %1012, align 8, !dbg !32 + %1014 = getelementptr inbounds nuw i8, ptr addrspace(3) %992, i32 140, !dbg !32 + %1015 = load float, ptr addrspace(3) %1014, align 4, !dbg !32 + %1016 = getelementptr inbounds nuw i8, ptr addrspace(3) %992, i32 144, !dbg !32 + %1017 = load float, ptr addrspace(3) %1016, align 16, !dbg !32 + %1018 = getelementptr inbounds nuw i8, ptr addrspace(3) %992, i32 148, !dbg !32 + %1019 = load float, ptr addrspace(3) %1018, align 4, !dbg !32 + %1020 = getelementptr inbounds nuw i8, ptr addrspace(3) %992, i32 152, !dbg !32 + %1021 = load float, ptr addrspace(3) %1020, align 8, !dbg !32 + %1022 = getelementptr inbounds nuw i8, ptr addrspace(3) %992, i32 156, !dbg !32 + %1023 = load float, ptr addrspace(3) %1022, align 4, !dbg !32 + %1024 = fmul float %993, %854, !dbg !32 + %1025 = fmul float %993, %855, !dbg !32 + %1026 = fmul float %993, %856, !dbg !32 + %1027 = fmul float %993, %857, !dbg !32 + %1028 = fmul float %993, %858, !dbg !32 + %1029 = fmul float %993, %859, !dbg !32 + %1030 = fmul float %993, %860, !dbg !32 + %1031 = fmul float %993, %861, !dbg !32 + %1032 = fmul float %995, %862, !dbg !32 + %1033 = fmul float %995, %863, !dbg !32 + %1034 = fmul float %995, %864, !dbg !32 + %1035 = fmul float %995, %865, !dbg !32 + %1036 = fmul float %995, %866, !dbg !32 + %1037 = fmul float %995, %867, !dbg !32 + %1038 = fmul float %995, %868, !dbg !32 + %1039 = fmul float %995, %869, !dbg !32 + %1040 = fmul float %997, %870, !dbg !32 + %1041 = fmul float %997, %871, !dbg !32 + %1042 = fmul float %997, %872, !dbg !32 + %1043 = fmul float %997, %873, !dbg !32 + %1044 = fmul float %997, %874, !dbg !32 + %1045 = fmul float %997, %875, !dbg !32 + %1046 = fmul float %997, %876, !dbg !32 + %1047 = fmul float %997, %877, !dbg !32 + %1048 = fmul float %999, %878, !dbg !32 + %1049 = fmul float %999, %879, !dbg !32 + %1050 = fmul float %999, %880, !dbg !32 + %1051 = fmul float %999, %881, !dbg !32 + %1052 = fmul float %999, %882, !dbg !32 + %1053 = fmul float %999, %883, !dbg !32 + %1054 = fmul float %999, %884, !dbg !32 + %1055 = fmul float %999, %885, !dbg !32 + %1056 = fmul float %1001, %886, !dbg !32 + %1057 = fmul float %1001, %887, !dbg !32 + %1058 = fmul float %1001, %888, !dbg !32 + %1059 = fmul float %1001, %889, !dbg !32 + %1060 = fmul float %1001, %890, !dbg !32 + %1061 = fmul float %1001, %891, !dbg !32 + %1062 = fmul float %1001, %892, !dbg !32 + %1063 = fmul float %1001, %893, !dbg !32 + %1064 = fmul float %1003, %894, !dbg !32 + %1065 = fmul float %1003, %895, !dbg !32 + %1066 = fmul float %1003, %896, !dbg !32 + %1067 = fmul float %1003, %897, !dbg !32 + %1068 = fmul float %1003, %898, !dbg !32 + %1069 = fmul float %1003, %899, !dbg !32 + %1070 = fmul float %1003, %900, !dbg !32 + %1071 = fmul float %1003, %901, !dbg !32 + %1072 = fmul float %1005, %902, !dbg !32 + %1073 = fmul float %1005, %903, !dbg !32 + %1074 = fmul float %1005, %904, !dbg !32 + %1075 = fmul float %1005, %905, !dbg !32 + %1076 = fmul float %1005, %906, !dbg !32 + %1077 = fmul float %1005, %907, !dbg !32 + %1078 = fmul float %1005, %908, !dbg !32 + %1079 = fmul float %1005, %909, !dbg !32 + %1080 = fmul float %1007, %910, !dbg !32 + %1081 = fmul float %1007, %911, !dbg !32 + %1082 = fmul float %1007, %912, !dbg !32 + %1083 = fmul float %1007, %913, !dbg !32 + %1084 = fmul float %1007, %914, !dbg !32 + %1085 = fmul float %1007, %915, !dbg !32 + %1086 = fmul float %1007, %916, !dbg !32 + %1087 = fmul float %1007, %917, !dbg !32 + %1088 = fmul float %1009, %918, !dbg !32 + %1089 = fmul float %1009, %919, !dbg !32 + %1090 = fmul float %1009, %920, !dbg !32 + %1091 = fmul float %1009, %921, !dbg !32 + %1092 = fmul float %1009, %922, !dbg !32 + %1093 = fmul float %1009, %923, !dbg !32 + %1094 = fmul float %1009, %924, !dbg !32 + %1095 = fmul float %1009, %925, !dbg !32 + %1096 = fmul float %1011, %926, !dbg !32 + %1097 = fmul float %1011, %927, !dbg !32 + %1098 = fmul float %1011, %928, !dbg !32 + %1099 = fmul float %1011, %929, !dbg !32 + %1100 = fmul float %1011, %930, !dbg !32 + %1101 = fmul float %1011, %931, !dbg !32 + %1102 = fmul float %1011, %932, !dbg !32 + %1103 = fmul float %1011, %933, !dbg !32 + %1104 = fmul float %1013, %934, !dbg !32 + %1105 = fmul float %1013, %935, !dbg !32 + %1106 = fmul float %1013, %936, !dbg !32 + %1107 = fmul float %1013, %937, !dbg !32 + %1108 = fmul float %1013, %938, !dbg !32 + %1109 = fmul float %1013, %939, !dbg !32 + %1110 = fmul float %1013, %940, !dbg !32 + %1111 = fmul float %1013, %941, !dbg !32 + %1112 = fmul float %1015, %942, !dbg !32 + %1113 = fmul float %1015, %943, !dbg !32 + %1114 = fmul float %1015, %944, !dbg !32 + %1115 = fmul float %1015, %945, !dbg !32 + %1116 = fmul float %1015, %946, !dbg !32 + %1117 = fmul float %1015, %947, !dbg !32 + %1118 = fmul float %1015, %948, !dbg !32 + %1119 = fmul float %1015, %949, !dbg !32 + %1120 = fmul float %1017, %950, !dbg !32 + %1121 = fmul float %1017, %951, !dbg !32 + %1122 = fmul float %1017, %952, !dbg !32 + %1123 = fmul float %1017, %953, !dbg !32 + %1124 = fmul float %1017, %954, !dbg !32 + %1125 = fmul float %1017, %955, !dbg !32 + %1126 = fmul float %1017, %956, !dbg !32 + %1127 = fmul float %1017, %957, !dbg !32 + %1128 = fmul float %1019, %958, !dbg !32 + %1129 = fmul float %1019, %959, !dbg !32 + %1130 = fmul float %1019, %960, !dbg !32 + %1131 = fmul float %1019, %961, !dbg !32 + %1132 = fmul float %1019, %962, !dbg !32 + %1133 = fmul float %1019, %963, !dbg !32 + %1134 = fmul float %1019, %964, !dbg !32 + %1135 = fmul float %1019, %965, !dbg !32 + %1136 = fmul float %1021, %966, !dbg !32 + %1137 = fmul float %1021, %967, !dbg !32 + %1138 = fmul float %1021, %968, !dbg !32 + %1139 = fmul float %1021, %969, !dbg !32 + %1140 = fmul float %1021, %970, !dbg !32 + %1141 = fmul float %1021, %971, !dbg !32 + %1142 = fmul float %1021, %972, !dbg !32 + %1143 = fmul float %1021, %973, !dbg !32 + %1144 = fmul float %1023, %974, !dbg !32 + %1145 = fmul float %1023, %975, !dbg !32 + %1146 = fmul float %1023, %976, !dbg !32 + %1147 = fmul float %1023, %977, !dbg !32 + %1148 = fmul float %1023, %978, !dbg !32 + %1149 = fmul float %1023, %979, !dbg !32 + %1150 = fmul float %1023, %980, !dbg !32 + %1151 = fmul float %1023, %981, !dbg !32 + %1152 = fmul float %452, %854, !dbg !33 + %1153 = fmul float %453, %855, !dbg !33 + %1154 = fmul float %454, %856, !dbg !33 + %1155 = fmul float %455, %857, !dbg !33 + %1156 = fmul float %456, %858, !dbg !33 + %1157 = fmul float %457, %859, !dbg !33 + %1158 = fmul float %458, %860, !dbg !33 + %1159 = fmul float %459, %861, !dbg !33 + %1160 = fmul float %460, %862, !dbg !33 + %1161 = fmul float %461, %863, !dbg !33 + %1162 = fmul float %462, %864, !dbg !33 + %1163 = fmul float %463, %865, !dbg !33 + %1164 = fmul float %464, %866, !dbg !33 + %1165 = fmul float %465, %867, !dbg !33 + %1166 = fmul float %466, %868, !dbg !33 + %1167 = fmul float %467, %869, !dbg !33 + %1168 = fmul float %468, %870, !dbg !33 + %1169 = fmul float %469, %871, !dbg !33 + %1170 = fmul float %470, %872, !dbg !33 + %1171 = fmul float %471, %873, !dbg !33 + %1172 = fmul float %472, %874, !dbg !33 + %1173 = fmul float %473, %875, !dbg !33 + %1174 = fmul float %474, %876, !dbg !33 + %1175 = fmul float %475, %877, !dbg !33 + %1176 = fmul float %476, %878, !dbg !33 + %1177 = fmul float %477, %879, !dbg !33 + %1178 = fmul float %478, %880, !dbg !33 + %1179 = fmul float %479, %881, !dbg !33 + %1180 = fmul float %480, %882, !dbg !33 + %1181 = fmul float %481, %883, !dbg !33 + %1182 = fmul float %482, %884, !dbg !33 + %1183 = fmul float %483, %885, !dbg !33 + %1184 = fmul float %484, %886, !dbg !33 + %1185 = fmul float %485, %887, !dbg !33 + %1186 = fmul float %486, %888, !dbg !33 + %1187 = fmul float %487, %889, !dbg !33 + %1188 = fmul float %488, %890, !dbg !33 + %1189 = fmul float %489, %891, !dbg !33 + %1190 = fmul float %490, %892, !dbg !33 + %1191 = fmul float %491, %893, !dbg !33 + %1192 = fmul float %492, %894, !dbg !33 + %1193 = fmul float %493, %895, !dbg !33 + %1194 = fmul float %494, %896, !dbg !33 + %1195 = fmul float %495, %897, !dbg !33 + %1196 = fmul float %496, %898, !dbg !33 + %1197 = fmul float %497, %899, !dbg !33 + %1198 = fmul float %498, %900, !dbg !33 + %1199 = fmul float %499, %901, !dbg !33 + %1200 = fmul float %500, %902, !dbg !33 + %1201 = fmul float %501, %903, !dbg !33 + %1202 = fmul float %502, %904, !dbg !33 + %1203 = fmul float %503, %905, !dbg !33 + %1204 = fmul float %504, %906, !dbg !33 + %1205 = fmul float %505, %907, !dbg !33 + %1206 = fmul float %506, %908, !dbg !33 + %1207 = fmul float %507, %909, !dbg !33 + %1208 = fmul float %508, %910, !dbg !33 + %1209 = fmul float %509, %911, !dbg !33 + %1210 = fmul float %510, %912, !dbg !33 + %1211 = fmul float %511, %913, !dbg !33 + %1212 = fmul float %512, %914, !dbg !33 + %1213 = fmul float %513, %915, !dbg !33 + %1214 = fmul float %514, %916, !dbg !33 + %1215 = fmul float %515, %917, !dbg !33 + %1216 = fmul float %516, %918, !dbg !33 + %1217 = fmul float %517, %919, !dbg !33 + %1218 = fmul float %518, %920, !dbg !33 + %1219 = fmul float %519, %921, !dbg !33 + %1220 = fmul float %520, %922, !dbg !33 + %1221 = fmul float %521, %923, !dbg !33 + %1222 = fmul float %522, %924, !dbg !33 + %1223 = fmul float %523, %925, !dbg !33 + %1224 = fmul float %524, %926, !dbg !33 + %1225 = fmul float %525, %927, !dbg !33 + %1226 = fmul float %526, %928, !dbg !33 + %1227 = fmul float %527, %929, !dbg !33 + %1228 = fmul float %528, %930, !dbg !33 + %1229 = fmul float %529, %931, !dbg !33 + %1230 = fmul float %530, %932, !dbg !33 + %1231 = fmul float %531, %933, !dbg !33 + %1232 = fmul float %532, %934, !dbg !33 + %1233 = fmul float %533, %935, !dbg !33 + %1234 = fmul float %534, %936, !dbg !33 + %1235 = fmul float %535, %937, !dbg !33 + %1236 = fmul float %536, %938, !dbg !33 + %1237 = fmul float %537, %939, !dbg !33 + %1238 = fmul float %538, %940, !dbg !33 + %1239 = fmul float %539, %941, !dbg !33 + %1240 = fmul float %540, %942, !dbg !33 + %1241 = fmul float %541, %943, !dbg !33 + %1242 = fmul float %542, %944, !dbg !33 + %1243 = fmul float %543, %945, !dbg !33 + %1244 = fmul float %544, %946, !dbg !33 + %1245 = fmul float %545, %947, !dbg !33 + %1246 = fmul float %546, %948, !dbg !33 + %1247 = fmul float %547, %949, !dbg !33 + %1248 = fmul float %548, %950, !dbg !33 + %1249 = fmul float %549, %951, !dbg !33 + %1250 = fmul float %550, %952, !dbg !33 + %1251 = fmul float %551, %953, !dbg !33 + %1252 = fmul float %552, %954, !dbg !33 + %1253 = fmul float %553, %955, !dbg !33 + %1254 = fmul float %554, %956, !dbg !33 + %1255 = fmul float %555, %957, !dbg !33 + %1256 = fmul float %556, %958, !dbg !33 + %1257 = fmul float %557, %959, !dbg !33 + %1258 = fmul float %558, %960, !dbg !33 + %1259 = fmul float %559, %961, !dbg !33 + %1260 = fmul float %560, %962, !dbg !33 + %1261 = fmul float %561, %963, !dbg !33 + %1262 = fmul float %562, %964, !dbg !33 + %1263 = fmul float %563, %965, !dbg !33 + %1264 = fmul float %564, %966, !dbg !33 + %1265 = fmul float %565, %967, !dbg !33 + %1266 = fmul float %566, %968, !dbg !33 + %1267 = fmul float %567, %969, !dbg !33 + %1268 = fmul float %568, %970, !dbg !33 + %1269 = fmul float %569, %971, !dbg !33 + %1270 = fmul float %570, %972, !dbg !33 + %1271 = fmul float %571, %973, !dbg !33 + %1272 = fmul float %572, %974, !dbg !33 + %1273 = fmul float %573, %975, !dbg !33 + %1274 = fmul float %574, %976, !dbg !33 + %1275 = fmul float %575, %977, !dbg !33 + %1276 = fmul float %576, %978, !dbg !33 + %1277 = fmul float %577, %979, !dbg !33 + %1278 = fmul float %578, %980, !dbg !33 + %1279 = fmul float %579, %981, !dbg !33 + %1280 = fadd float %1152, %1153, !dbg !34 + %1281 = fadd float %1154, %1280, !dbg !34 + %1282 = fadd float %1155, %1281, !dbg !34 + %1283 = fadd float %1156, %1282, !dbg !34 + %1284 = fadd float %1157, %1283, !dbg !34 + %1285 = fadd float %1158, %1284, !dbg !34 + %1286 = fadd float %1159, %1285, !dbg !34 + %1287 = fadd float %1160, %1161, !dbg !34 + %1288 = fadd float %1162, %1287, !dbg !34 + %1289 = fadd float %1163, %1288, !dbg !34 + %1290 = fadd float %1164, %1289, !dbg !34 + %1291 = fadd float %1165, %1290, !dbg !34 + %1292 = fadd float %1166, %1291, !dbg !34 + %1293 = fadd float %1167, %1292, !dbg !34 + %1294 = fadd float %1168, %1169, !dbg !34 + %1295 = fadd float %1170, %1294, !dbg !34 + %1296 = fadd float %1171, %1295, !dbg !34 + %1297 = fadd float %1172, %1296, !dbg !34 + %1298 = fadd float %1173, %1297, !dbg !34 + %1299 = fadd float %1174, %1298, !dbg !34 + %1300 = fadd float %1175, %1299, !dbg !34 + %1301 = fadd float %1176, %1177, !dbg !34 + %1302 = fadd float %1178, %1301, !dbg !34 + %1303 = fadd float %1179, %1302, !dbg !34 + %1304 = fadd float %1180, %1303, !dbg !34 + %1305 = fadd float %1181, %1304, !dbg !34 + %1306 = fadd float %1182, %1305, !dbg !34 + %1307 = fadd float %1183, %1306, !dbg !34 + %1308 = fadd float %1184, %1185, !dbg !34 + %1309 = fadd float %1186, %1308, !dbg !34 + %1310 = fadd float %1187, %1309, !dbg !34 + %1311 = fadd float %1188, %1310, !dbg !34 + %1312 = fadd float %1189, %1311, !dbg !34 + %1313 = fadd float %1190, %1312, !dbg !34 + %1314 = fadd float %1191, %1313, !dbg !34 + %1315 = fadd float %1192, %1193, !dbg !34 + %1316 = fadd float %1194, %1315, !dbg !34 + %1317 = fadd float %1195, %1316, !dbg !34 + %1318 = fadd float %1196, %1317, !dbg !34 + %1319 = fadd float %1197, %1318, !dbg !34 + %1320 = fadd float %1198, %1319, !dbg !34 + %1321 = fadd float %1199, %1320, !dbg !34 + %1322 = fadd float %1200, %1201, !dbg !34 + %1323 = fadd float %1202, %1322, !dbg !34 + %1324 = fadd float %1203, %1323, !dbg !34 + %1325 = fadd float %1204, %1324, !dbg !34 + %1326 = fadd float %1205, %1325, !dbg !34 + %1327 = fadd float %1206, %1326, !dbg !34 + %1328 = fadd float %1207, %1327, !dbg !34 + %1329 = fadd float %1208, %1209, !dbg !34 + %1330 = fadd float %1210, %1329, !dbg !34 + %1331 = fadd float %1211, %1330, !dbg !34 + %1332 = fadd float %1212, %1331, !dbg !34 + %1333 = fadd float %1213, %1332, !dbg !34 + %1334 = fadd float %1214, %1333, !dbg !34 + %1335 = fadd float %1215, %1334, !dbg !34 + %1336 = fadd float %1216, %1217, !dbg !34 + %1337 = fadd float %1218, %1336, !dbg !34 + %1338 = fadd float %1219, %1337, !dbg !34 + %1339 = fadd float %1220, %1338, !dbg !34 + %1340 = fadd float %1221, %1339, !dbg !34 + %1341 = fadd float %1222, %1340, !dbg !34 + %1342 = fadd float %1223, %1341, !dbg !34 + %1343 = fadd float %1224, %1225, !dbg !34 + %1344 = fadd float %1226, %1343, !dbg !34 + %1345 = fadd float %1227, %1344, !dbg !34 + %1346 = fadd float %1228, %1345, !dbg !34 + %1347 = fadd float %1229, %1346, !dbg !34 + %1348 = fadd float %1230, %1347, !dbg !34 + %1349 = fadd float %1231, %1348, !dbg !34 + %1350 = fadd float %1232, %1233, !dbg !34 + %1351 = fadd float %1234, %1350, !dbg !34 + %1352 = fadd float %1235, %1351, !dbg !34 + %1353 = fadd float %1236, %1352, !dbg !34 + %1354 = fadd float %1237, %1353, !dbg !34 + %1355 = fadd float %1238, %1354, !dbg !34 + %1356 = fadd float %1239, %1355, !dbg !34 + %1357 = fadd float %1240, %1241, !dbg !34 + %1358 = fadd float %1242, %1357, !dbg !34 + %1359 = fadd float %1243, %1358, !dbg !34 + %1360 = fadd float %1244, %1359, !dbg !34 + %1361 = fadd float %1245, %1360, !dbg !34 + %1362 = fadd float %1246, %1361, !dbg !34 + %1363 = fadd float %1247, %1362, !dbg !34 + %1364 = fadd float %1248, %1249, !dbg !34 + %1365 = fadd float %1250, %1364, !dbg !34 + %1366 = fadd float %1251, %1365, !dbg !34 + %1367 = fadd float %1252, %1366, !dbg !34 + %1368 = fadd float %1253, %1367, !dbg !34 + %1369 = fadd float %1254, %1368, !dbg !34 + %1370 = fadd float %1255, %1369, !dbg !34 + %1371 = fadd float %1256, %1257, !dbg !34 + %1372 = fadd float %1258, %1371, !dbg !34 + %1373 = fadd float %1259, %1372, !dbg !34 + %1374 = fadd float %1260, %1373, !dbg !34 + %1375 = fadd float %1261, %1374, !dbg !34 + %1376 = fadd float %1262, %1375, !dbg !34 + %1377 = fadd float %1263, %1376, !dbg !34 + %1378 = fadd float %1264, %1265, !dbg !34 + %1379 = fadd float %1266, %1378, !dbg !34 + %1380 = fadd float %1267, %1379, !dbg !34 + %1381 = fadd float %1268, %1380, !dbg !34 + %1382 = fadd float %1269, %1381, !dbg !34 + %1383 = fadd float %1270, %1382, !dbg !34 + %1384 = fadd float %1271, %1383, !dbg !34 + %1385 = fadd float %1272, %1273, !dbg !34 + %1386 = fadd float %1274, %1385, !dbg !34 + %1387 = fadd float %1275, %1386, !dbg !34 + %1388 = fadd float %1276, %1387, !dbg !34 + %1389 = fadd float %1277, %1388, !dbg !34 + %1390 = fadd float %1278, %1389, !dbg !34 + %1391 = fadd float %1279, %1390, !dbg !34 + %1392 = bitcast float %1286 to i32, !dbg !37 + %1393 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1392, i32 8, i32 31), !dbg !37 + %1394 = bitcast i32 %1393 to float, !dbg !37 + %1395 = fadd float %1286, %1394, !dbg !34 + %1396 = bitcast float %1395 to i32, !dbg !37 + %1397 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1396, i32 4, i32 31), !dbg !37 + %1398 = bitcast i32 %1397 to float, !dbg !37 + %1399 = fadd float %1395, %1398, !dbg !34 + %1400 = bitcast float %1399 to i32, !dbg !37 + %1401 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1400, i32 2, i32 31), !dbg !37 + %1402 = bitcast i32 %1401 to float, !dbg !37 + %1403 = fadd float %1399, %1402, !dbg !34 + %1404 = bitcast float %1403 to i32, !dbg !37 + %1405 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1404, i32 1, i32 31), !dbg !37 + %1406 = bitcast i32 %1405 to float, !dbg !37 + %1407 = fadd float %1403, %1406, !dbg !34 + %1408 = bitcast float %1293 to i32, !dbg !37 + %1409 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1408, i32 8, i32 31), !dbg !37 + %1410 = bitcast i32 %1409 to float, !dbg !37 + %1411 = fadd float %1293, %1410, !dbg !34 + %1412 = bitcast float %1411 to i32, !dbg !37 + %1413 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1412, i32 4, i32 31), !dbg !37 + %1414 = bitcast i32 %1413 to float, !dbg !37 + %1415 = fadd float %1411, %1414, !dbg !34 + %1416 = bitcast float %1415 to i32, !dbg !37 + %1417 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1416, i32 2, i32 31), !dbg !37 + %1418 = bitcast i32 %1417 to float, !dbg !37 + %1419 = fadd float %1415, %1418, !dbg !34 + %1420 = bitcast float %1419 to i32, !dbg !37 + %1421 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1420, i32 1, i32 31), !dbg !37 + %1422 = bitcast i32 %1421 to float, !dbg !37 + %1423 = fadd float %1419, %1422, !dbg !34 + %1424 = bitcast float %1300 to i32, !dbg !37 + %1425 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1424, i32 8, i32 31), !dbg !37 + %1426 = bitcast i32 %1425 to float, !dbg !37 + %1427 = fadd float %1300, %1426, !dbg !34 + %1428 = bitcast float %1427 to i32, !dbg !37 + %1429 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1428, i32 4, i32 31), !dbg !37 + %1430 = bitcast i32 %1429 to float, !dbg !37 + %1431 = fadd float %1427, %1430, !dbg !34 + %1432 = bitcast float %1431 to i32, !dbg !37 + %1433 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1432, i32 2, i32 31), !dbg !37 + %1434 = bitcast i32 %1433 to float, !dbg !37 + %1435 = fadd float %1431, %1434, !dbg !34 + %1436 = bitcast float %1435 to i32, !dbg !37 + %1437 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1436, i32 1, i32 31), !dbg !37 + %1438 = bitcast i32 %1437 to float, !dbg !37 + %1439 = fadd float %1435, %1438, !dbg !34 + %1440 = bitcast float %1307 to i32, !dbg !37 + %1441 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1440, i32 8, i32 31), !dbg !37 + %1442 = bitcast i32 %1441 to float, !dbg !37 + %1443 = fadd float %1307, %1442, !dbg !34 + %1444 = bitcast float %1443 to i32, !dbg !37 + %1445 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1444, i32 4, i32 31), !dbg !37 + %1446 = bitcast i32 %1445 to float, !dbg !37 + %1447 = fadd float %1443, %1446, !dbg !34 + %1448 = bitcast float %1447 to i32, !dbg !37 + %1449 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1448, i32 2, i32 31), !dbg !37 + %1450 = bitcast i32 %1449 to float, !dbg !37 + %1451 = fadd float %1447, %1450, !dbg !34 + %1452 = bitcast float %1451 to i32, !dbg !37 + %1453 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1452, i32 1, i32 31), !dbg !37 + %1454 = bitcast i32 %1453 to float, !dbg !37 + %1455 = fadd float %1451, %1454, !dbg !34 + %1456 = bitcast float %1314 to i32, !dbg !37 + %1457 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1456, i32 8, i32 31), !dbg !37 + %1458 = bitcast i32 %1457 to float, !dbg !37 + %1459 = fadd float %1314, %1458, !dbg !34 + %1460 = bitcast float %1459 to i32, !dbg !37 + %1461 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1460, i32 4, i32 31), !dbg !37 + %1462 = bitcast i32 %1461 to float, !dbg !37 + %1463 = fadd float %1459, %1462, !dbg !34 + %1464 = bitcast float %1463 to i32, !dbg !37 + %1465 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1464, i32 2, i32 31), !dbg !37 + %1466 = bitcast i32 %1465 to float, !dbg !37 + %1467 = fadd float %1463, %1466, !dbg !34 + %1468 = bitcast float %1467 to i32, !dbg !37 + %1469 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1468, i32 1, i32 31), !dbg !37 + %1470 = bitcast i32 %1469 to float, !dbg !37 + %1471 = fadd float %1467, %1470, !dbg !34 + %1472 = bitcast float %1321 to i32, !dbg !37 + %1473 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1472, i32 8, i32 31), !dbg !37 + %1474 = bitcast i32 %1473 to float, !dbg !37 + %1475 = fadd float %1321, %1474, !dbg !34 + %1476 = bitcast float %1475 to i32, !dbg !37 + %1477 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1476, i32 4, i32 31), !dbg !37 + %1478 = bitcast i32 %1477 to float, !dbg !37 + %1479 = fadd float %1475, %1478, !dbg !34 + %1480 = bitcast float %1479 to i32, !dbg !37 + %1481 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1480, i32 2, i32 31), !dbg !37 + %1482 = bitcast i32 %1481 to float, !dbg !37 + %1483 = fadd float %1479, %1482, !dbg !34 + %1484 = bitcast float %1483 to i32, !dbg !37 + %1485 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1484, i32 1, i32 31), !dbg !37 + %1486 = bitcast i32 %1485 to float, !dbg !37 + %1487 = fadd float %1483, %1486, !dbg !34 + %1488 = bitcast float %1328 to i32, !dbg !37 + %1489 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1488, i32 8, i32 31), !dbg !37 + %1490 = bitcast i32 %1489 to float, !dbg !37 + %1491 = fadd float %1328, %1490, !dbg !34 + %1492 = bitcast float %1491 to i32, !dbg !37 + %1493 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1492, i32 4, i32 31), !dbg !37 + %1494 = bitcast i32 %1493 to float, !dbg !37 + %1495 = fadd float %1491, %1494, !dbg !34 + %1496 = bitcast float %1495 to i32, !dbg !37 + %1497 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1496, i32 2, i32 31), !dbg !37 + %1498 = bitcast i32 %1497 to float, !dbg !37 + %1499 = fadd float %1495, %1498, !dbg !34 + %1500 = bitcast float %1499 to i32, !dbg !37 + %1501 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1500, i32 1, i32 31), !dbg !37 + %1502 = bitcast i32 %1501 to float, !dbg !37 + %1503 = fadd float %1499, %1502, !dbg !34 + %1504 = bitcast float %1335 to i32, !dbg !37 + %1505 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1504, i32 8, i32 31), !dbg !37 + %1506 = bitcast i32 %1505 to float, !dbg !37 + %1507 = fadd float %1335, %1506, !dbg !34 + %1508 = bitcast float %1507 to i32, !dbg !37 + %1509 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1508, i32 4, i32 31), !dbg !37 + %1510 = bitcast i32 %1509 to float, !dbg !37 + %1511 = fadd float %1507, %1510, !dbg !34 + %1512 = bitcast float %1511 to i32, !dbg !37 + %1513 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1512, i32 2, i32 31), !dbg !37 + %1514 = bitcast i32 %1513 to float, !dbg !37 + %1515 = fadd float %1511, %1514, !dbg !34 + %1516 = bitcast float %1515 to i32, !dbg !37 + %1517 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1516, i32 1, i32 31), !dbg !37 + %1518 = bitcast i32 %1517 to float, !dbg !37 + %1519 = fadd float %1515, %1518, !dbg !34 + %1520 = bitcast float %1342 to i32, !dbg !37 + %1521 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1520, i32 8, i32 31), !dbg !37 + %1522 = bitcast i32 %1521 to float, !dbg !37 + %1523 = fadd float %1342, %1522, !dbg !34 + %1524 = bitcast float %1523 to i32, !dbg !37 + %1525 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1524, i32 4, i32 31), !dbg !37 + %1526 = bitcast i32 %1525 to float, !dbg !37 + %1527 = fadd float %1523, %1526, !dbg !34 + %1528 = bitcast float %1527 to i32, !dbg !37 + %1529 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1528, i32 2, i32 31), !dbg !37 + %1530 = bitcast i32 %1529 to float, !dbg !37 + %1531 = fadd float %1527, %1530, !dbg !34 + %1532 = bitcast float %1531 to i32, !dbg !37 + %1533 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1532, i32 1, i32 31), !dbg !37 + %1534 = bitcast i32 %1533 to float, !dbg !37 + %1535 = fadd float %1531, %1534, !dbg !34 + %1536 = bitcast float %1349 to i32, !dbg !37 + %1537 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1536, i32 8, i32 31), !dbg !37 + %1538 = bitcast i32 %1537 to float, !dbg !37 + %1539 = fadd float %1349, %1538, !dbg !34 + %1540 = bitcast float %1539 to i32, !dbg !37 + %1541 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1540, i32 4, i32 31), !dbg !37 + %1542 = bitcast i32 %1541 to float, !dbg !37 + %1543 = fadd float %1539, %1542, !dbg !34 + %1544 = bitcast float %1543 to i32, !dbg !37 + %1545 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1544, i32 2, i32 31), !dbg !37 + %1546 = bitcast i32 %1545 to float, !dbg !37 + %1547 = fadd float %1543, %1546, !dbg !34 + %1548 = bitcast float %1547 to i32, !dbg !37 + %1549 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1548, i32 1, i32 31), !dbg !37 + %1550 = bitcast i32 %1549 to float, !dbg !37 + %1551 = fadd float %1547, %1550, !dbg !34 + %1552 = bitcast float %1356 to i32, !dbg !37 + %1553 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1552, i32 8, i32 31), !dbg !37 + %1554 = bitcast i32 %1553 to float, !dbg !37 + %1555 = fadd float %1356, %1554, !dbg !34 + %1556 = bitcast float %1555 to i32, !dbg !37 + %1557 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1556, i32 4, i32 31), !dbg !37 + %1558 = bitcast i32 %1557 to float, !dbg !37 + %1559 = fadd float %1555, %1558, !dbg !34 + %1560 = bitcast float %1559 to i32, !dbg !37 + %1561 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1560, i32 2, i32 31), !dbg !37 + %1562 = bitcast i32 %1561 to float, !dbg !37 + %1563 = fadd float %1559, %1562, !dbg !34 + %1564 = bitcast float %1563 to i32, !dbg !37 + %1565 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1564, i32 1, i32 31), !dbg !37 + %1566 = bitcast i32 %1565 to float, !dbg !37 + %1567 = fadd float %1563, %1566, !dbg !34 + %1568 = bitcast float %1363 to i32, !dbg !37 + %1569 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1568, i32 8, i32 31), !dbg !37 + %1570 = bitcast i32 %1569 to float, !dbg !37 + %1571 = fadd float %1363, %1570, !dbg !34 + %1572 = bitcast float %1571 to i32, !dbg !37 + %1573 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1572, i32 4, i32 31), !dbg !37 + %1574 = bitcast i32 %1573 to float, !dbg !37 + %1575 = fadd float %1571, %1574, !dbg !34 + %1576 = bitcast float %1575 to i32, !dbg !37 + %1577 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1576, i32 2, i32 31), !dbg !37 + %1578 = bitcast i32 %1577 to float, !dbg !37 + %1579 = fadd float %1575, %1578, !dbg !34 + %1580 = bitcast float %1579 to i32, !dbg !37 + %1581 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1580, i32 1, i32 31), !dbg !37 + %1582 = bitcast i32 %1581 to float, !dbg !37 + %1583 = fadd float %1579, %1582, !dbg !34 + %1584 = bitcast float %1370 to i32, !dbg !37 + %1585 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1584, i32 8, i32 31), !dbg !37 + %1586 = bitcast i32 %1585 to float, !dbg !37 + %1587 = fadd float %1370, %1586, !dbg !34 + %1588 = bitcast float %1587 to i32, !dbg !37 + %1589 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1588, i32 4, i32 31), !dbg !37 + %1590 = bitcast i32 %1589 to float, !dbg !37 + %1591 = fadd float %1587, %1590, !dbg !34 + %1592 = bitcast float %1591 to i32, !dbg !37 + %1593 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1592, i32 2, i32 31), !dbg !37 + %1594 = bitcast i32 %1593 to float, !dbg !37 + %1595 = fadd float %1591, %1594, !dbg !34 + %1596 = bitcast float %1595 to i32, !dbg !37 + %1597 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1596, i32 1, i32 31), !dbg !37 + %1598 = bitcast i32 %1597 to float, !dbg !37 + %1599 = fadd float %1595, %1598, !dbg !34 + %1600 = bitcast float %1377 to i32, !dbg !37 + %1601 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1600, i32 8, i32 31), !dbg !37 + %1602 = bitcast i32 %1601 to float, !dbg !37 + %1603 = fadd float %1377, %1602, !dbg !34 + %1604 = bitcast float %1603 to i32, !dbg !37 + %1605 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1604, i32 4, i32 31), !dbg !37 + %1606 = bitcast i32 %1605 to float, !dbg !37 + %1607 = fadd float %1603, %1606, !dbg !34 + %1608 = bitcast float %1607 to i32, !dbg !37 + %1609 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1608, i32 2, i32 31), !dbg !37 + %1610 = bitcast i32 %1609 to float, !dbg !37 + %1611 = fadd float %1607, %1610, !dbg !34 + %1612 = bitcast float %1611 to i32, !dbg !37 + %1613 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1612, i32 1, i32 31), !dbg !37 + %1614 = bitcast i32 %1613 to float, !dbg !37 + %1615 = fadd float %1611, %1614, !dbg !34 + %1616 = bitcast float %1384 to i32, !dbg !37 + %1617 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1616, i32 8, i32 31), !dbg !37 + %1618 = bitcast i32 %1617 to float, !dbg !37 + %1619 = fadd float %1384, %1618, !dbg !34 + %1620 = bitcast float %1619 to i32, !dbg !37 + %1621 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1620, i32 4, i32 31), !dbg !37 + %1622 = bitcast i32 %1621 to float, !dbg !37 + %1623 = fadd float %1619, %1622, !dbg !34 + %1624 = bitcast float %1623 to i32, !dbg !37 + %1625 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1624, i32 2, i32 31), !dbg !37 + %1626 = bitcast i32 %1625 to float, !dbg !37 + %1627 = fadd float %1623, %1626, !dbg !34 + %1628 = bitcast float %1627 to i32, !dbg !37 + %1629 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1628, i32 1, i32 31), !dbg !37 + %1630 = bitcast i32 %1629 to float, !dbg !37 + %1631 = fadd float %1627, %1630, !dbg !34 + %1632 = bitcast float %1391 to i32, !dbg !37 + %1633 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1632, i32 8, i32 31), !dbg !37 + %1634 = bitcast i32 %1633 to float, !dbg !37 + %1635 = fadd float %1391, %1634, !dbg !34 + %1636 = bitcast float %1635 to i32, !dbg !37 + %1637 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1636, i32 4, i32 31), !dbg !37 + %1638 = bitcast i32 %1637 to float, !dbg !37 + %1639 = fadd float %1635, %1638, !dbg !34 + %1640 = bitcast float %1639 to i32, !dbg !37 + %1641 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1640, i32 2, i32 31), !dbg !37 + %1642 = bitcast i32 %1641 to float, !dbg !37 + %1643 = fadd float %1639, %1642, !dbg !34 + %1644 = bitcast float %1643 to i32, !dbg !37 + %1645 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1644, i32 1, i32 31), !dbg !37 + %1646 = bitcast i32 %1645 to float, !dbg !37 + %1647 = fadd float %1643, %1646, !dbg !34 + %1648 = fmul float %1407, %452, !dbg !40 + %1649 = fmul float %1407, %453, !dbg !40 + %1650 = fmul float %1407, %454, !dbg !40 + %1651 = fmul float %1407, %455, !dbg !40 + %1652 = fmul float %1407, %456, !dbg !40 + %1653 = fmul float %1407, %457, !dbg !40 + %1654 = fmul float %1407, %458, !dbg !40 + %1655 = fmul float %1407, %459, !dbg !40 + %1656 = fmul float %1423, %460, !dbg !40 + %1657 = fmul float %1423, %461, !dbg !40 + %1658 = fmul float %1423, %462, !dbg !40 + %1659 = fmul float %1423, %463, !dbg !40 + %1660 = fmul float %1423, %464, !dbg !40 + %1661 = fmul float %1423, %465, !dbg !40 + %1662 = fmul float %1423, %466, !dbg !40 + %1663 = fmul float %1423, %467, !dbg !40 + %1664 = fmul float %1439, %468, !dbg !40 + %1665 = fmul float %1439, %469, !dbg !40 + %1666 = fmul float %1439, %470, !dbg !40 + %1667 = fmul float %1439, %471, !dbg !40 + %1668 = fmul float %1439, %472, !dbg !40 + %1669 = fmul float %1439, %473, !dbg !40 + %1670 = fmul float %1439, %474, !dbg !40 + %1671 = fmul float %1439, %475, !dbg !40 + %1672 = fmul float %1455, %476, !dbg !40 + %1673 = fmul float %1455, %477, !dbg !40 + %1674 = fmul float %1455, %478, !dbg !40 + %1675 = fmul float %1455, %479, !dbg !40 + %1676 = fmul float %1455, %480, !dbg !40 + %1677 = fmul float %1455, %481, !dbg !40 + %1678 = fmul float %1455, %482, !dbg !40 + %1679 = fmul float %1455, %483, !dbg !40 + %1680 = fmul float %1471, %484, !dbg !40 + %1681 = fmul float %1471, %485, !dbg !40 + %1682 = fmul float %1471, %486, !dbg !40 + %1683 = fmul float %1471, %487, !dbg !40 + %1684 = fmul float %1471, %488, !dbg !40 + %1685 = fmul float %1471, %489, !dbg !40 + %1686 = fmul float %1471, %490, !dbg !40 + %1687 = fmul float %1471, %491, !dbg !40 + %1688 = fmul float %1487, %492, !dbg !40 + %1689 = fmul float %1487, %493, !dbg !40 + %1690 = fmul float %1487, %494, !dbg !40 + %1691 = fmul float %1487, %495, !dbg !40 + %1692 = fmul float %1487, %496, !dbg !40 + %1693 = fmul float %1487, %497, !dbg !40 + %1694 = fmul float %1487, %498, !dbg !40 + %1695 = fmul float %1487, %499, !dbg !40 + %1696 = fmul float %1503, %500, !dbg !40 + %1697 = fmul float %1503, %501, !dbg !40 + %1698 = fmul float %1503, %502, !dbg !40 + %1699 = fmul float %1503, %503, !dbg !40 + %1700 = fmul float %1503, %504, !dbg !40 + %1701 = fmul float %1503, %505, !dbg !40 + %1702 = fmul float %1503, %506, !dbg !40 + %1703 = fmul float %1503, %507, !dbg !40 + %1704 = fmul float %1519, %508, !dbg !40 + %1705 = fmul float %1519, %509, !dbg !40 + %1706 = fmul float %1519, %510, !dbg !40 + %1707 = fmul float %1519, %511, !dbg !40 + %1708 = fmul float %1519, %512, !dbg !40 + %1709 = fmul float %1519, %513, !dbg !40 + %1710 = fmul float %1519, %514, !dbg !40 + %1711 = fmul float %1519, %515, !dbg !40 + %1712 = fmul float %1535, %516, !dbg !40 + %1713 = fmul float %1535, %517, !dbg !40 + %1714 = fmul float %1535, %518, !dbg !40 + %1715 = fmul float %1535, %519, !dbg !40 + %1716 = fmul float %1535, %520, !dbg !40 + %1717 = fmul float %1535, %521, !dbg !40 + %1718 = fmul float %1535, %522, !dbg !40 + %1719 = fmul float %1535, %523, !dbg !40 + %1720 = fmul float %1551, %524, !dbg !40 + %1721 = fmul float %1551, %525, !dbg !40 + %1722 = fmul float %1551, %526, !dbg !40 + %1723 = fmul float %1551, %527, !dbg !40 + %1724 = fmul float %1551, %528, !dbg !40 + %1725 = fmul float %1551, %529, !dbg !40 + %1726 = fmul float %1551, %530, !dbg !40 + %1727 = fmul float %1551, %531, !dbg !40 + %1728 = fmul float %1567, %532, !dbg !40 + %1729 = fmul float %1567, %533, !dbg !40 + %1730 = fmul float %1567, %534, !dbg !40 + %1731 = fmul float %1567, %535, !dbg !40 + %1732 = fmul float %1567, %536, !dbg !40 + %1733 = fmul float %1567, %537, !dbg !40 + %1734 = fmul float %1567, %538, !dbg !40 + %1735 = fmul float %1567, %539, !dbg !40 + %1736 = fmul float %1583, %540, !dbg !40 + %1737 = fmul float %1583, %541, !dbg !40 + %1738 = fmul float %1583, %542, !dbg !40 + %1739 = fmul float %1583, %543, !dbg !40 + %1740 = fmul float %1583, %544, !dbg !40 + %1741 = fmul float %1583, %545, !dbg !40 + %1742 = fmul float %1583, %546, !dbg !40 + %1743 = fmul float %1583, %547, !dbg !40 + %1744 = fmul float %1599, %548, !dbg !40 + %1745 = fmul float %1599, %549, !dbg !40 + %1746 = fmul float %1599, %550, !dbg !40 + %1747 = fmul float %1599, %551, !dbg !40 + %1748 = fmul float %1599, %552, !dbg !40 + %1749 = fmul float %1599, %553, !dbg !40 + %1750 = fmul float %1599, %554, !dbg !40 + %1751 = fmul float %1599, %555, !dbg !40 + %1752 = fmul float %1615, %556, !dbg !40 + %1753 = fmul float %1615, %557, !dbg !40 + %1754 = fmul float %1615, %558, !dbg !40 + %1755 = fmul float %1615, %559, !dbg !40 + %1756 = fmul float %1615, %560, !dbg !40 + %1757 = fmul float %1615, %561, !dbg !40 + %1758 = fmul float %1615, %562, !dbg !40 + %1759 = fmul float %1615, %563, !dbg !40 + %1760 = fmul float %1631, %564, !dbg !40 + %1761 = fmul float %1631, %565, !dbg !40 + %1762 = fmul float %1631, %566, !dbg !40 + %1763 = fmul float %1631, %567, !dbg !40 + %1764 = fmul float %1631, %568, !dbg !40 + %1765 = fmul float %1631, %569, !dbg !40 + %1766 = fmul float %1631, %570, !dbg !40 + %1767 = fmul float %1631, %571, !dbg !40 + %1768 = fmul float %1647, %572, !dbg !40 + %1769 = fmul float %1647, %573, !dbg !40 + %1770 = fmul float %1647, %574, !dbg !40 + %1771 = fmul float %1647, %575, !dbg !40 + %1772 = fmul float %1647, %576, !dbg !40 + %1773 = fmul float %1647, %577, !dbg !40 + %1774 = fmul float %1647, %578, !dbg !40 + %1775 = fmul float %1647, %579, !dbg !40 + %1776 = fmul float %993, %1648, !dbg !41 + %1777 = fmul float %993, %1649, !dbg !41 + %1778 = fmul float %993, %1650, !dbg !41 + %1779 = fmul float %993, %1651, !dbg !41 + %1780 = fmul float %993, %1652, !dbg !41 + %1781 = fmul float %993, %1653, !dbg !41 + %1782 = fmul float %993, %1654, !dbg !41 + %1783 = fmul float %993, %1655, !dbg !41 + %1784 = fmul float %995, %1656, !dbg !41 + %1785 = fmul float %995, %1657, !dbg !41 + %1786 = fmul float %995, %1658, !dbg !41 + %1787 = fmul float %995, %1659, !dbg !41 + %1788 = fmul float %995, %1660, !dbg !41 + %1789 = fmul float %995, %1661, !dbg !41 + %1790 = fmul float %995, %1662, !dbg !41 + %1791 = fmul float %995, %1663, !dbg !41 + %1792 = fmul float %997, %1664, !dbg !41 + %1793 = fmul float %997, %1665, !dbg !41 + %1794 = fmul float %997, %1666, !dbg !41 + %1795 = fmul float %997, %1667, !dbg !41 + %1796 = fmul float %997, %1668, !dbg !41 + %1797 = fmul float %997, %1669, !dbg !41 + %1798 = fmul float %997, %1670, !dbg !41 + %1799 = fmul float %997, %1671, !dbg !41 + %1800 = fmul float %999, %1672, !dbg !41 + %1801 = fmul float %999, %1673, !dbg !41 + %1802 = fmul float %999, %1674, !dbg !41 + %1803 = fmul float %999, %1675, !dbg !41 + %1804 = fmul float %999, %1676, !dbg !41 + %1805 = fmul float %999, %1677, !dbg !41 + %1806 = fmul float %999, %1678, !dbg !41 + %1807 = fmul float %999, %1679, !dbg !41 + %1808 = fmul float %1001, %1680, !dbg !41 + %1809 = fmul float %1001, %1681, !dbg !41 + %1810 = fmul float %1001, %1682, !dbg !41 + %1811 = fmul float %1001, %1683, !dbg !41 + %1812 = fmul float %1001, %1684, !dbg !41 + %1813 = fmul float %1001, %1685, !dbg !41 + %1814 = fmul float %1001, %1686, !dbg !41 + %1815 = fmul float %1001, %1687, !dbg !41 + %1816 = fmul float %1003, %1688, !dbg !41 + %1817 = fmul float %1003, %1689, !dbg !41 + %1818 = fmul float %1003, %1690, !dbg !41 + %1819 = fmul float %1003, %1691, !dbg !41 + %1820 = fmul float %1003, %1692, !dbg !41 + %1821 = fmul float %1003, %1693, !dbg !41 + %1822 = fmul float %1003, %1694, !dbg !41 + %1823 = fmul float %1003, %1695, !dbg !41 + %1824 = fmul float %1005, %1696, !dbg !41 + %1825 = fmul float %1005, %1697, !dbg !41 + %1826 = fmul float %1005, %1698, !dbg !41 + %1827 = fmul float %1005, %1699, !dbg !41 + %1828 = fmul float %1005, %1700, !dbg !41 + %1829 = fmul float %1005, %1701, !dbg !41 + %1830 = fmul float %1005, %1702, !dbg !41 + %1831 = fmul float %1005, %1703, !dbg !41 + %1832 = fmul float %1007, %1704, !dbg !41 + %1833 = fmul float %1007, %1705, !dbg !41 + %1834 = fmul float %1007, %1706, !dbg !41 + %1835 = fmul float %1007, %1707, !dbg !41 + %1836 = fmul float %1007, %1708, !dbg !41 + %1837 = fmul float %1007, %1709, !dbg !41 + %1838 = fmul float %1007, %1710, !dbg !41 + %1839 = fmul float %1007, %1711, !dbg !41 + %1840 = fmul float %1009, %1712, !dbg !41 + %1841 = fmul float %1009, %1713, !dbg !41 + %1842 = fmul float %1009, %1714, !dbg !41 + %1843 = fmul float %1009, %1715, !dbg !41 + %1844 = fmul float %1009, %1716, !dbg !41 + %1845 = fmul float %1009, %1717, !dbg !41 + %1846 = fmul float %1009, %1718, !dbg !41 + %1847 = fmul float %1009, %1719, !dbg !41 + %1848 = fmul float %1011, %1720, !dbg !41 + %1849 = fmul float %1011, %1721, !dbg !41 + %1850 = fmul float %1011, %1722, !dbg !41 + %1851 = fmul float %1011, %1723, !dbg !41 + %1852 = fmul float %1011, %1724, !dbg !41 + %1853 = fmul float %1011, %1725, !dbg !41 + %1854 = fmul float %1011, %1726, !dbg !41 + %1855 = fmul float %1011, %1727, !dbg !41 + %1856 = fmul float %1013, %1728, !dbg !41 + %1857 = fmul float %1013, %1729, !dbg !41 + %1858 = fmul float %1013, %1730, !dbg !41 + %1859 = fmul float %1013, %1731, !dbg !41 + %1860 = fmul float %1013, %1732, !dbg !41 + %1861 = fmul float %1013, %1733, !dbg !41 + %1862 = fmul float %1013, %1734, !dbg !41 + %1863 = fmul float %1013, %1735, !dbg !41 + %1864 = fmul float %1015, %1736, !dbg !41 + %1865 = fmul float %1015, %1737, !dbg !41 + %1866 = fmul float %1015, %1738, !dbg !41 + %1867 = fmul float %1015, %1739, !dbg !41 + %1868 = fmul float %1015, %1740, !dbg !41 + %1869 = fmul float %1015, %1741, !dbg !41 + %1870 = fmul float %1015, %1742, !dbg !41 + %1871 = fmul float %1015, %1743, !dbg !41 + %1872 = fmul float %1017, %1744, !dbg !41 + %1873 = fmul float %1017, %1745, !dbg !41 + %1874 = fmul float %1017, %1746, !dbg !41 + %1875 = fmul float %1017, %1747, !dbg !41 + %1876 = fmul float %1017, %1748, !dbg !41 + %1877 = fmul float %1017, %1749, !dbg !41 + %1878 = fmul float %1017, %1750, !dbg !41 + %1879 = fmul float %1017, %1751, !dbg !41 + %1880 = fmul float %1019, %1752, !dbg !41 + %1881 = fmul float %1019, %1753, !dbg !41 + %1882 = fmul float %1019, %1754, !dbg !41 + %1883 = fmul float %1019, %1755, !dbg !41 + %1884 = fmul float %1019, %1756, !dbg !41 + %1885 = fmul float %1019, %1757, !dbg !41 + %1886 = fmul float %1019, %1758, !dbg !41 + %1887 = fmul float %1019, %1759, !dbg !41 + %1888 = fmul float %1021, %1760, !dbg !41 + %1889 = fmul float %1021, %1761, !dbg !41 + %1890 = fmul float %1021, %1762, !dbg !41 + %1891 = fmul float %1021, %1763, !dbg !41 + %1892 = fmul float %1021, %1764, !dbg !41 + %1893 = fmul float %1021, %1765, !dbg !41 + %1894 = fmul float %1021, %1766, !dbg !41 + %1895 = fmul float %1021, %1767, !dbg !41 + %1896 = fmul float %1023, %1768, !dbg !41 + %1897 = fmul float %1023, %1769, !dbg !41 + %1898 = fmul float %1023, %1770, !dbg !41 + %1899 = fmul float %1023, %1771, !dbg !41 + %1900 = fmul float %1023, %1772, !dbg !41 + %1901 = fmul float %1023, %1773, !dbg !41 + %1902 = fmul float %1023, %1774, !dbg !41 + %1903 = fmul float %1023, %1775, !dbg !41 + %1904 = fsub float %1024, %1776, !dbg !42 + %1905 = fsub float %1025, %1777, !dbg !42 + %1906 = fsub float %1026, %1778, !dbg !42 + %1907 = fsub float %1027, %1779, !dbg !42 + %1908 = fsub float %1028, %1780, !dbg !42 + %1909 = fsub float %1029, %1781, !dbg !42 + %1910 = fsub float %1030, %1782, !dbg !42 + %1911 = fsub float %1031, %1783, !dbg !42 + %1912 = fsub float %1032, %1784, !dbg !42 + %1913 = fsub float %1033, %1785, !dbg !42 + %1914 = fsub float %1034, %1786, !dbg !42 + %1915 = fsub float %1035, %1787, !dbg !42 + %1916 = fsub float %1036, %1788, !dbg !42 + %1917 = fsub float %1037, %1789, !dbg !42 + %1918 = fsub float %1038, %1790, !dbg !42 + %1919 = fsub float %1039, %1791, !dbg !42 + %1920 = fsub float %1040, %1792, !dbg !42 + %1921 = fsub float %1041, %1793, !dbg !42 + %1922 = fsub float %1042, %1794, !dbg !42 + %1923 = fsub float %1043, %1795, !dbg !42 + %1924 = fsub float %1044, %1796, !dbg !42 + %1925 = fsub float %1045, %1797, !dbg !42 + %1926 = fsub float %1046, %1798, !dbg !42 + %1927 = fsub float %1047, %1799, !dbg !42 + %1928 = insertelement <2 x float> poison, float %1048, i64 0, !dbg !42 + %1929 = insertelement <2 x float> %1928, float %1049, i64 1, !dbg !42 + %1930 = insertelement <2 x float> poison, float %1800, i64 0, !dbg !42 + %1931 = insertelement <2 x float> %1930, float %1801, i64 1, !dbg !42 + %1932 = fsub <2 x float> %1929, %1931, !dbg !42 + %1933 = insertelement <2 x float> poison, float %1050, i64 0, !dbg !42 + %1934 = insertelement <2 x float> %1933, float %1051, i64 1, !dbg !42 + %1935 = insertelement <2 x float> poison, float %1802, i64 0, !dbg !42 + %1936 = insertelement <2 x float> %1935, float %1803, i64 1, !dbg !42 + %1937 = fsub <2 x float> %1934, %1936, !dbg !42 + %1938 = insertelement <2 x float> poison, float %1052, i64 0, !dbg !42 + %1939 = insertelement <2 x float> %1938, float %1053, i64 1, !dbg !42 + %1940 = insertelement <2 x float> poison, float %1804, i64 0, !dbg !42 + %1941 = insertelement <2 x float> %1940, float %1805, i64 1, !dbg !42 + %1942 = fsub <2 x float> %1939, %1941, !dbg !42 + %1943 = insertelement <2 x float> poison, float %1054, i64 0, !dbg !42 + %1944 = insertelement <2 x float> %1943, float %1055, i64 1, !dbg !42 + %1945 = insertelement <2 x float> poison, float %1806, i64 0, !dbg !42 + %1946 = insertelement <2 x float> %1945, float %1807, i64 1, !dbg !42 + %1947 = fsub <2 x float> %1944, %1946, !dbg !42 + %1948 = insertelement <2 x float> poison, float %1056, i64 0, !dbg !42 + %1949 = insertelement <2 x float> %1948, float %1057, i64 1, !dbg !42 + %1950 = insertelement <2 x float> poison, float %1808, i64 0, !dbg !42 + %1951 = insertelement <2 x float> %1950, float %1809, i64 1, !dbg !42 + %1952 = fsub <2 x float> %1949, %1951, !dbg !42 + %1953 = insertelement <2 x float> poison, float %1058, i64 0, !dbg !42 + %1954 = insertelement <2 x float> %1953, float %1059, i64 1, !dbg !42 + %1955 = insertelement <2 x float> poison, float %1810, i64 0, !dbg !42 + %1956 = insertelement <2 x float> %1955, float %1811, i64 1, !dbg !42 + %1957 = fsub <2 x float> %1954, %1956, !dbg !42 + %1958 = insertelement <2 x float> poison, float %1060, i64 0, !dbg !42 + %1959 = insertelement <2 x float> %1958, float %1061, i64 1, !dbg !42 + %1960 = insertelement <2 x float> poison, float %1812, i64 0, !dbg !42 + %1961 = insertelement <2 x float> %1960, float %1813, i64 1, !dbg !42 + %1962 = fsub <2 x float> %1959, %1961, !dbg !42 + %1963 = insertelement <2 x float> poison, float %1062, i64 0, !dbg !42 + %1964 = insertelement <2 x float> %1963, float %1063, i64 1, !dbg !42 + %1965 = insertelement <2 x float> poison, float %1814, i64 0, !dbg !42 + %1966 = insertelement <2 x float> %1965, float %1815, i64 1, !dbg !42 + %1967 = fsub <2 x float> %1964, %1966, !dbg !42 + %1968 = insertelement <2 x float> poison, float %1064, i64 0, !dbg !42 + %1969 = insertelement <2 x float> %1968, float %1065, i64 1, !dbg !42 + %1970 = insertelement <2 x float> poison, float %1816, i64 0, !dbg !42 + %1971 = insertelement <2 x float> %1970, float %1817, i64 1, !dbg !42 + %1972 = fsub <2 x float> %1969, %1971, !dbg !42 + %1973 = insertelement <2 x float> poison, float %1066, i64 0, !dbg !42 + %1974 = insertelement <2 x float> %1973, float %1067, i64 1, !dbg !42 + %1975 = insertelement <2 x float> poison, float %1818, i64 0, !dbg !42 + %1976 = insertelement <2 x float> %1975, float %1819, i64 1, !dbg !42 + %1977 = fsub <2 x float> %1974, %1976, !dbg !42 + %1978 = insertelement <2 x float> poison, float %1068, i64 0, !dbg !42 + %1979 = insertelement <2 x float> %1978, float %1069, i64 1, !dbg !42 + %1980 = insertelement <2 x float> poison, float %1820, i64 0, !dbg !42 + %1981 = insertelement <2 x float> %1980, float %1821, i64 1, !dbg !42 + %1982 = fsub <2 x float> %1979, %1981, !dbg !42 + %1983 = insertelement <2 x float> poison, float %1070, i64 0, !dbg !42 + %1984 = insertelement <2 x float> %1983, float %1071, i64 1, !dbg !42 + %1985 = insertelement <2 x float> poison, float %1822, i64 0, !dbg !42 + %1986 = insertelement <2 x float> %1985, float %1823, i64 1, !dbg !42 + %1987 = fsub <2 x float> %1984, %1986, !dbg !42 + %1988 = insertelement <2 x float> poison, float %1072, i64 0, !dbg !42 + %1989 = insertelement <2 x float> %1988, float %1073, i64 1, !dbg !42 + %1990 = insertelement <2 x float> poison, float %1824, i64 0, !dbg !42 + %1991 = insertelement <2 x float> %1990, float %1825, i64 1, !dbg !42 + %1992 = fsub <2 x float> %1989, %1991, !dbg !42 + %1993 = insertelement <2 x float> poison, float %1074, i64 0, !dbg !42 + %1994 = insertelement <2 x float> %1993, float %1075, i64 1, !dbg !42 + %1995 = insertelement <2 x float> poison, float %1826, i64 0, !dbg !42 + %1996 = insertelement <2 x float> %1995, float %1827, i64 1, !dbg !42 + %1997 = fsub <2 x float> %1994, %1996, !dbg !42 + %1998 = insertelement <2 x float> poison, float %1076, i64 0, !dbg !42 + %1999 = insertelement <2 x float> %1998, float %1077, i64 1, !dbg !42 + %2000 = insertelement <2 x float> poison, float %1828, i64 0, !dbg !42 + %2001 = insertelement <2 x float> %2000, float %1829, i64 1, !dbg !42 + %2002 = fsub <2 x float> %1999, %2001, !dbg !42 + %2003 = insertelement <2 x float> poison, float %1078, i64 0, !dbg !42 + %2004 = insertelement <2 x float> %2003, float %1079, i64 1, !dbg !42 + %2005 = insertelement <2 x float> poison, float %1830, i64 0, !dbg !42 + %2006 = insertelement <2 x float> %2005, float %1831, i64 1, !dbg !42 + %2007 = fsub <2 x float> %2004, %2006, !dbg !42 + %2008 = insertelement <2 x float> poison, float %1080, i64 0, !dbg !42 + %2009 = insertelement <2 x float> %2008, float %1081, i64 1, !dbg !42 + %2010 = insertelement <2 x float> poison, float %1832, i64 0, !dbg !42 + %2011 = insertelement <2 x float> %2010, float %1833, i64 1, !dbg !42 + %2012 = fsub <2 x float> %2009, %2011, !dbg !42 + %2013 = insertelement <2 x float> poison, float %1082, i64 0, !dbg !42 + %2014 = insertelement <2 x float> %2013, float %1083, i64 1, !dbg !42 + %2015 = insertelement <2 x float> poison, float %1834, i64 0, !dbg !42 + %2016 = insertelement <2 x float> %2015, float %1835, i64 1, !dbg !42 + %2017 = fsub <2 x float> %2014, %2016, !dbg !42 + %2018 = insertelement <2 x float> poison, float %1084, i64 0, !dbg !42 + %2019 = insertelement <2 x float> %2018, float %1085, i64 1, !dbg !42 + %2020 = insertelement <2 x float> poison, float %1836, i64 0, !dbg !42 + %2021 = insertelement <2 x float> %2020, float %1837, i64 1, !dbg !42 + %2022 = fsub <2 x float> %2019, %2021, !dbg !42 + %2023 = insertelement <2 x float> poison, float %1086, i64 0, !dbg !42 + %2024 = insertelement <2 x float> %2023, float %1087, i64 1, !dbg !42 + %2025 = insertelement <2 x float> poison, float %1838, i64 0, !dbg !42 + %2026 = insertelement <2 x float> %2025, float %1839, i64 1, !dbg !42 + %2027 = fsub <2 x float> %2024, %2026, !dbg !42 + %2028 = insertelement <2 x float> poison, float %1088, i64 0, !dbg !42 + %2029 = insertelement <2 x float> %2028, float %1089, i64 1, !dbg !42 + %2030 = insertelement <2 x float> poison, float %1840, i64 0, !dbg !42 + %2031 = insertelement <2 x float> %2030, float %1841, i64 1, !dbg !42 + %2032 = fsub <2 x float> %2029, %2031, !dbg !42 + %2033 = insertelement <2 x float> poison, float %1090, i64 0, !dbg !42 + %2034 = insertelement <2 x float> %2033, float %1091, i64 1, !dbg !42 + %2035 = insertelement <2 x float> poison, float %1842, i64 0, !dbg !42 + %2036 = insertelement <2 x float> %2035, float %1843, i64 1, !dbg !42 + %2037 = fsub <2 x float> %2034, %2036, !dbg !42 + %2038 = insertelement <2 x float> poison, float %1092, i64 0, !dbg !42 + %2039 = insertelement <2 x float> %2038, float %1093, i64 1, !dbg !42 + %2040 = insertelement <2 x float> poison, float %1844, i64 0, !dbg !42 + %2041 = insertelement <2 x float> %2040, float %1845, i64 1, !dbg !42 + %2042 = fsub <2 x float> %2039, %2041, !dbg !42 + %2043 = insertelement <2 x float> poison, float %1094, i64 0, !dbg !42 + %2044 = insertelement <2 x float> %2043, float %1095, i64 1, !dbg !42 + %2045 = insertelement <2 x float> poison, float %1846, i64 0, !dbg !42 + %2046 = insertelement <2 x float> %2045, float %1847, i64 1, !dbg !42 + %2047 = fsub <2 x float> %2044, %2046, !dbg !42 + %2048 = insertelement <2 x float> poison, float %1096, i64 0, !dbg !42 + %2049 = insertelement <2 x float> %2048, float %1097, i64 1, !dbg !42 + %2050 = insertelement <2 x float> poison, float %1848, i64 0, !dbg !42 + %2051 = insertelement <2 x float> %2050, float %1849, i64 1, !dbg !42 + %2052 = fsub <2 x float> %2049, %2051, !dbg !42 + %2053 = insertelement <2 x float> poison, float %1098, i64 0, !dbg !42 + %2054 = insertelement <2 x float> %2053, float %1099, i64 1, !dbg !42 + %2055 = insertelement <2 x float> poison, float %1850, i64 0, !dbg !42 + %2056 = insertelement <2 x float> %2055, float %1851, i64 1, !dbg !42 + %2057 = fsub <2 x float> %2054, %2056, !dbg !42 + %2058 = insertelement <2 x float> poison, float %1100, i64 0, !dbg !42 + %2059 = insertelement <2 x float> %2058, float %1101, i64 1, !dbg !42 + %2060 = insertelement <2 x float> poison, float %1852, i64 0, !dbg !42 + %2061 = insertelement <2 x float> %2060, float %1853, i64 1, !dbg !42 + %2062 = fsub <2 x float> %2059, %2061, !dbg !42 + %2063 = insertelement <2 x float> poison, float %1102, i64 0, !dbg !42 + %2064 = insertelement <2 x float> %2063, float %1103, i64 1, !dbg !42 + %2065 = insertelement <2 x float> poison, float %1854, i64 0, !dbg !42 + %2066 = insertelement <2 x float> %2065, float %1855, i64 1, !dbg !42 + %2067 = fsub <2 x float> %2064, %2066, !dbg !42 + %2068 = insertelement <2 x float> poison, float %1104, i64 0, !dbg !42 + %2069 = insertelement <2 x float> %2068, float %1105, i64 1, !dbg !42 + %2070 = insertelement <2 x float> poison, float %1856, i64 0, !dbg !42 + %2071 = insertelement <2 x float> %2070, float %1857, i64 1, !dbg !42 + %2072 = fsub <2 x float> %2069, %2071, !dbg !42 + %2073 = insertelement <2 x float> poison, float %1106, i64 0, !dbg !42 + %2074 = insertelement <2 x float> %2073, float %1107, i64 1, !dbg !42 + %2075 = insertelement <2 x float> poison, float %1858, i64 0, !dbg !42 + %2076 = insertelement <2 x float> %2075, float %1859, i64 1, !dbg !42 + %2077 = fsub <2 x float> %2074, %2076, !dbg !42 + %2078 = insertelement <2 x float> poison, float %1108, i64 0, !dbg !42 + %2079 = insertelement <2 x float> %2078, float %1109, i64 1, !dbg !42 + %2080 = insertelement <2 x float> poison, float %1860, i64 0, !dbg !42 + %2081 = insertelement <2 x float> %2080, float %1861, i64 1, !dbg !42 + %2082 = fsub <2 x float> %2079, %2081, !dbg !42 + %2083 = insertelement <2 x float> poison, float %1110, i64 0, !dbg !42 + %2084 = insertelement <2 x float> %2083, float %1111, i64 1, !dbg !42 + %2085 = insertelement <2 x float> poison, float %1862, i64 0, !dbg !42 + %2086 = insertelement <2 x float> %2085, float %1863, i64 1, !dbg !42 + %2087 = fsub <2 x float> %2084, %2086, !dbg !42 + %2088 = insertelement <2 x float> poison, float %1112, i64 0, !dbg !42 + %2089 = insertelement <2 x float> %2088, float %1113, i64 1, !dbg !42 + %2090 = insertelement <2 x float> poison, float %1864, i64 0, !dbg !42 + %2091 = insertelement <2 x float> %2090, float %1865, i64 1, !dbg !42 + %2092 = fsub <2 x float> %2089, %2091, !dbg !42 + %2093 = insertelement <2 x float> poison, float %1114, i64 0, !dbg !42 + %2094 = insertelement <2 x float> %2093, float %1115, i64 1, !dbg !42 + %2095 = insertelement <2 x float> poison, float %1866, i64 0, !dbg !42 + %2096 = insertelement <2 x float> %2095, float %1867, i64 1, !dbg !42 + %2097 = fsub <2 x float> %2094, %2096, !dbg !42 + %2098 = insertelement <2 x float> poison, float %1116, i64 0, !dbg !42 + %2099 = insertelement <2 x float> %2098, float %1117, i64 1, !dbg !42 + %2100 = insertelement <2 x float> poison, float %1868, i64 0, !dbg !42 + %2101 = insertelement <2 x float> %2100, float %1869, i64 1, !dbg !42 + %2102 = fsub <2 x float> %2099, %2101, !dbg !42 + %2103 = insertelement <2 x float> poison, float %1118, i64 0, !dbg !42 + %2104 = insertelement <2 x float> %2103, float %1119, i64 1, !dbg !42 + %2105 = insertelement <2 x float> poison, float %1870, i64 0, !dbg !42 + %2106 = insertelement <2 x float> %2105, float %1871, i64 1, !dbg !42 + %2107 = fsub <2 x float> %2104, %2106, !dbg !42 + %2108 = insertelement <2 x float> poison, float %1120, i64 0, !dbg !42 + %2109 = insertelement <2 x float> %2108, float %1121, i64 1, !dbg !42 + %2110 = insertelement <2 x float> poison, float %1872, i64 0, !dbg !42 + %2111 = insertelement <2 x float> %2110, float %1873, i64 1, !dbg !42 + %2112 = fsub <2 x float> %2109, %2111, !dbg !42 + %2113 = insertelement <2 x float> poison, float %1122, i64 0, !dbg !42 + %2114 = insertelement <2 x float> %2113, float %1123, i64 1, !dbg !42 + %2115 = insertelement <2 x float> poison, float %1874, i64 0, !dbg !42 + %2116 = insertelement <2 x float> %2115, float %1875, i64 1, !dbg !42 + %2117 = fsub <2 x float> %2114, %2116, !dbg !42 + %2118 = insertelement <2 x float> poison, float %1124, i64 0, !dbg !42 + %2119 = insertelement <2 x float> %2118, float %1125, i64 1, !dbg !42 + %2120 = insertelement <2 x float> poison, float %1876, i64 0, !dbg !42 + %2121 = insertelement <2 x float> %2120, float %1877, i64 1, !dbg !42 + %2122 = fsub <2 x float> %2119, %2121, !dbg !42 + %2123 = insertelement <2 x float> poison, float %1126, i64 0, !dbg !42 + %2124 = insertelement <2 x float> %2123, float %1127, i64 1, !dbg !42 + %2125 = insertelement <2 x float> poison, float %1878, i64 0, !dbg !42 + %2126 = insertelement <2 x float> %2125, float %1879, i64 1, !dbg !42 + %2127 = fsub <2 x float> %2124, %2126, !dbg !42 + %2128 = insertelement <2 x float> poison, float %1128, i64 0, !dbg !42 + %2129 = insertelement <2 x float> %2128, float %1129, i64 1, !dbg !42 + %2130 = insertelement <2 x float> poison, float %1880, i64 0, !dbg !42 + %2131 = insertelement <2 x float> %2130, float %1881, i64 1, !dbg !42 + %2132 = fsub <2 x float> %2129, %2131, !dbg !42 + %2133 = insertelement <2 x float> poison, float %1130, i64 0, !dbg !42 + %2134 = insertelement <2 x float> %2133, float %1131, i64 1, !dbg !42 + %2135 = insertelement <2 x float> poison, float %1882, i64 0, !dbg !42 + %2136 = insertelement <2 x float> %2135, float %1883, i64 1, !dbg !42 + %2137 = fsub <2 x float> %2134, %2136, !dbg !42 + %2138 = insertelement <2 x float> poison, float %1132, i64 0, !dbg !42 + %2139 = insertelement <2 x float> %2138, float %1133, i64 1, !dbg !42 + %2140 = insertelement <2 x float> poison, float %1884, i64 0, !dbg !42 + %2141 = insertelement <2 x float> %2140, float %1885, i64 1, !dbg !42 + %2142 = fsub <2 x float> %2139, %2141, !dbg !42 + %2143 = insertelement <2 x float> poison, float %1134, i64 0, !dbg !42 + %2144 = insertelement <2 x float> %2143, float %1135, i64 1, !dbg !42 + %2145 = insertelement <2 x float> poison, float %1886, i64 0, !dbg !42 + %2146 = insertelement <2 x float> %2145, float %1887, i64 1, !dbg !42 + %2147 = fsub <2 x float> %2144, %2146, !dbg !42 + %2148 = insertelement <2 x float> poison, float %1136, i64 0, !dbg !42 + %2149 = insertelement <2 x float> %2148, float %1137, i64 1, !dbg !42 + %2150 = insertelement <2 x float> poison, float %1888, i64 0, !dbg !42 + %2151 = insertelement <2 x float> %2150, float %1889, i64 1, !dbg !42 + %2152 = fsub <2 x float> %2149, %2151, !dbg !42 + %2153 = insertelement <2 x float> poison, float %1138, i64 0, !dbg !42 + %2154 = insertelement <2 x float> %2153, float %1139, i64 1, !dbg !42 + %2155 = insertelement <2 x float> poison, float %1890, i64 0, !dbg !42 + %2156 = insertelement <2 x float> %2155, float %1891, i64 1, !dbg !42 + %2157 = fsub <2 x float> %2154, %2156, !dbg !42 + %2158 = insertelement <2 x float> poison, float %1140, i64 0, !dbg !42 + %2159 = insertelement <2 x float> %2158, float %1141, i64 1, !dbg !42 + %2160 = insertelement <2 x float> poison, float %1892, i64 0, !dbg !42 + %2161 = insertelement <2 x float> %2160, float %1893, i64 1, !dbg !42 + %2162 = fsub <2 x float> %2159, %2161, !dbg !42 + %2163 = insertelement <2 x float> poison, float %1142, i64 0, !dbg !42 + %2164 = insertelement <2 x float> %2163, float %1143, i64 1, !dbg !42 + %2165 = insertelement <2 x float> poison, float %1894, i64 0, !dbg !42 + %2166 = insertelement <2 x float> %2165, float %1895, i64 1, !dbg !42 + %2167 = fsub <2 x float> %2164, %2166, !dbg !42 + %2168 = insertelement <2 x float> poison, float %1144, i64 0, !dbg !42 + %2169 = insertelement <2 x float> %2168, float %1145, i64 1, !dbg !42 + %2170 = insertelement <2 x float> poison, float %1896, i64 0, !dbg !42 + %2171 = insertelement <2 x float> %2170, float %1897, i64 1, !dbg !42 + %2172 = fsub <2 x float> %2169, %2171, !dbg !42 + %2173 = insertelement <2 x float> poison, float %1146, i64 0, !dbg !42 + %2174 = insertelement <2 x float> %2173, float %1147, i64 1, !dbg !42 + %2175 = insertelement <2 x float> poison, float %1898, i64 0, !dbg !42 + %2176 = insertelement <2 x float> %2175, float %1899, i64 1, !dbg !42 + %2177 = fsub <2 x float> %2174, %2176, !dbg !42 + %2178 = insertelement <2 x float> poison, float %1148, i64 0, !dbg !42 + %2179 = insertelement <2 x float> %2178, float %1149, i64 1, !dbg !42 + %2180 = insertelement <2 x float> poison, float %1900, i64 0, !dbg !42 + %2181 = insertelement <2 x float> %2180, float %1901, i64 1, !dbg !42 + %2182 = fsub <2 x float> %2179, %2181, !dbg !42 + %2183 = insertelement <2 x float> poison, float %1150, i64 0, !dbg !42 + %2184 = insertelement <2 x float> %2183, float %1151, i64 1, !dbg !42 + %2185 = insertelement <2 x float> poison, float %1902, i64 0, !dbg !42 + %2186 = insertelement <2 x float> %2185, float %1903, i64 1, !dbg !42 + %2187 = fsub <2 x float> %2184, %2186, !dbg !42 + %2188 = fptrunc float %1904 to bfloat, !dbg !43 + %2189 = fptrunc float %1905 to bfloat, !dbg !43 + %2190 = fptrunc float %1906 to bfloat, !dbg !43 + %2191 = fptrunc float %1907 to bfloat, !dbg !43 + %2192 = fptrunc float %1908 to bfloat, !dbg !43 + %2193 = fptrunc float %1909 to bfloat, !dbg !43 + %2194 = fptrunc float %1910 to bfloat, !dbg !43 + %2195 = fptrunc float %1911 to bfloat, !dbg !43 + %2196 = fptrunc float %1912 to bfloat, !dbg !43 + %2197 = fptrunc float %1913 to bfloat, !dbg !43 + %2198 = fptrunc float %1914 to bfloat, !dbg !43 + %2199 = fptrunc float %1915 to bfloat, !dbg !43 + %2200 = fptrunc float %1916 to bfloat, !dbg !43 + %2201 = fptrunc float %1917 to bfloat, !dbg !43 + %2202 = fptrunc float %1918 to bfloat, !dbg !43 + %2203 = fptrunc float %1919 to bfloat, !dbg !43 + %2204 = fptrunc float %1920 to bfloat, !dbg !43 + %2205 = fptrunc float %1921 to bfloat, !dbg !43 + %2206 = fptrunc float %1922 to bfloat, !dbg !43 + %2207 = fptrunc float %1923 to bfloat, !dbg !43 + %2208 = fptrunc float %1924 to bfloat, !dbg !43 + %2209 = fptrunc float %1925 to bfloat, !dbg !43 + %2210 = fptrunc float %1926 to bfloat, !dbg !43 + %2211 = fptrunc float %1927 to bfloat, !dbg !43 + %2212 = fptrunc <2 x float> %1932 to <2 x bfloat>, !dbg !43 + %2213 = fptrunc <2 x float> %1937 to <2 x bfloat>, !dbg !43 + %2214 = fptrunc <2 x float> %1942 to <2 x bfloat>, !dbg !43 + %2215 = fptrunc <2 x float> %1947 to <2 x bfloat>, !dbg !43 + %2216 = fptrunc <2 x float> %1952 to <2 x bfloat>, !dbg !43 + %2217 = fptrunc <2 x float> %1957 to <2 x bfloat>, !dbg !43 + %2218 = fptrunc <2 x float> %1962 to <2 x bfloat>, !dbg !43 + %2219 = fptrunc <2 x float> %1967 to <2 x bfloat>, !dbg !43 + %2220 = fptrunc <2 x float> %1972 to <2 x bfloat>, !dbg !43 + %2221 = fptrunc <2 x float> %1977 to <2 x bfloat>, !dbg !43 + %2222 = fptrunc <2 x float> %1982 to <2 x bfloat>, !dbg !43 + %2223 = fptrunc <2 x float> %1987 to <2 x bfloat>, !dbg !43 + %2224 = fptrunc <2 x float> %1992 to <2 x bfloat>, !dbg !43 + %2225 = fptrunc <2 x float> %1997 to <2 x bfloat>, !dbg !43 + %2226 = fptrunc <2 x float> %2002 to <2 x bfloat>, !dbg !43 + %2227 = fptrunc <2 x float> %2007 to <2 x bfloat>, !dbg !43 + %2228 = fptrunc <2 x float> %2012 to <2 x bfloat>, !dbg !43 + %2229 = fptrunc <2 x float> %2017 to <2 x bfloat>, !dbg !43 + %2230 = fptrunc <2 x float> %2022 to <2 x bfloat>, !dbg !43 + %2231 = fptrunc <2 x float> %2027 to <2 x bfloat>, !dbg !43 + %2232 = fptrunc <2 x float> %2032 to <2 x bfloat>, !dbg !43 + %2233 = fptrunc <2 x float> %2037 to <2 x bfloat>, !dbg !43 + %2234 = fptrunc <2 x float> %2042 to <2 x bfloat>, !dbg !43 + %2235 = fptrunc <2 x float> %2047 to <2 x bfloat>, !dbg !43 + %2236 = fptrunc <2 x float> %2052 to <2 x bfloat>, !dbg !43 + %2237 = fptrunc <2 x float> %2057 to <2 x bfloat>, !dbg !43 + %2238 = fptrunc <2 x float> %2062 to <2 x bfloat>, !dbg !43 + %2239 = fptrunc <2 x float> %2067 to <2 x bfloat>, !dbg !43 + %2240 = fptrunc <2 x float> %2072 to <2 x bfloat>, !dbg !43 + %2241 = fptrunc <2 x float> %2077 to <2 x bfloat>, !dbg !43 + %2242 = fptrunc <2 x float> %2082 to <2 x bfloat>, !dbg !43 + %2243 = fptrunc <2 x float> %2087 to <2 x bfloat>, !dbg !43 + %2244 = fptrunc <2 x float> %2092 to <2 x bfloat>, !dbg !43 + %2245 = fptrunc <2 x float> %2097 to <2 x bfloat>, !dbg !43 + %2246 = fptrunc <2 x float> %2102 to <2 x bfloat>, !dbg !43 + %2247 = fptrunc <2 x float> %2107 to <2 x bfloat>, !dbg !43 + %2248 = fptrunc <2 x float> %2112 to <2 x bfloat>, !dbg !43 + %2249 = fptrunc <2 x float> %2117 to <2 x bfloat>, !dbg !43 + %2250 = fptrunc <2 x float> %2122 to <2 x bfloat>, !dbg !43 + %2251 = fptrunc <2 x float> %2127 to <2 x bfloat>, !dbg !43 + %2252 = fptrunc <2 x float> %2132 to <2 x bfloat>, !dbg !43 + %2253 = fptrunc <2 x float> %2137 to <2 x bfloat>, !dbg !43 + %2254 = fptrunc <2 x float> %2142 to <2 x bfloat>, !dbg !43 + %2255 = fptrunc <2 x float> %2147 to <2 x bfloat>, !dbg !43 + %2256 = fptrunc <2 x float> %2152 to <2 x bfloat>, !dbg !43 + %2257 = fptrunc <2 x float> %2157 to <2 x bfloat>, !dbg !43 + %2258 = fptrunc <2 x float> %2162 to <2 x bfloat>, !dbg !43 + %2259 = fptrunc <2 x float> %2167 to <2 x bfloat>, !dbg !43 + %2260 = fptrunc <2 x float> %2172 to <2 x bfloat>, !dbg !43 + %2261 = fptrunc <2 x float> %2177 to <2 x bfloat>, !dbg !43 + %2262 = fptrunc <2 x float> %2182 to <2 x bfloat>, !dbg !43 + %2263 = fptrunc <2 x float> %2187 to <2 x bfloat>, !dbg !43 + %2264 = insertelement <2 x bfloat> poison, bfloat %2188, i64 0, !dbg !44 + %2265 = insertelement <2 x bfloat> %2264, bfloat %2189, i64 1, !dbg !44 + %2266 = bitcast <2 x bfloat> %2265 to i32, !dbg !44 + %2267 = insertelement <2 x bfloat> poison, bfloat %2190, i64 0, !dbg !44 + %2268 = insertelement <2 x bfloat> %2267, bfloat %2191, i64 1, !dbg !44 + %2269 = bitcast <2 x bfloat> %2268 to i32, !dbg !44 + %2270 = insertelement <2 x bfloat> poison, bfloat %2192, i64 0, !dbg !44 + %2271 = insertelement <2 x bfloat> %2270, bfloat %2193, i64 1, !dbg !44 + %2272 = bitcast <2 x bfloat> %2271 to i32, !dbg !44 + %2273 = insertelement <2 x bfloat> poison, bfloat %2194, i64 0, !dbg !44 + %2274 = insertelement <2 x bfloat> %2273, bfloat %2195, i64 1, !dbg !44 + %2275 = bitcast <2 x bfloat> %2274 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %2266, i32 %2269, i32 %2272, i32 %2275, ptr addrspace(1) %149, i1 %28) #4, !dbg !44 + %2276 = insertelement <2 x bfloat> poison, bfloat %2196, i64 0, !dbg !44 + %2277 = insertelement <2 x bfloat> %2276, bfloat %2197, i64 1, !dbg !44 + %2278 = bitcast <2 x bfloat> %2277 to i32, !dbg !44 + %2279 = insertelement <2 x bfloat> poison, bfloat %2198, i64 0, !dbg !44 + %2280 = insertelement <2 x bfloat> %2279, bfloat %2199, i64 1, !dbg !44 + %2281 = bitcast <2 x bfloat> %2280 to i32, !dbg !44 + %2282 = insertelement <2 x bfloat> poison, bfloat %2200, i64 0, !dbg !44 + %2283 = insertelement <2 x bfloat> %2282, bfloat %2201, i64 1, !dbg !44 + %2284 = bitcast <2 x bfloat> %2283 to i32, !dbg !44 + %2285 = insertelement <2 x bfloat> poison, bfloat %2202, i64 0, !dbg !44 + %2286 = insertelement <2 x bfloat> %2285, bfloat %2203, i64 1, !dbg !44 + %2287 = bitcast <2 x bfloat> %2286 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %2278, i32 %2281, i32 %2284, i32 %2287, ptr addrspace(1) %150, i1 %182) #4, !dbg !44 + %2288 = insertelement <2 x bfloat> poison, bfloat %2204, i64 0, !dbg !44 + %2289 = insertelement <2 x bfloat> %2288, bfloat %2205, i64 1, !dbg !44 + %2290 = bitcast <2 x bfloat> %2289 to i32, !dbg !44 + %2291 = insertelement <2 x bfloat> poison, bfloat %2206, i64 0, !dbg !44 + %2292 = insertelement <2 x bfloat> %2291, bfloat %2207, i64 1, !dbg !44 + %2293 = bitcast <2 x bfloat> %2292 to i32, !dbg !44 + %2294 = insertelement <2 x bfloat> poison, bfloat %2208, i64 0, !dbg !44 + %2295 = insertelement <2 x bfloat> %2294, bfloat %2209, i64 1, !dbg !44 + %2296 = bitcast <2 x bfloat> %2295 to i32, !dbg !44 + %2297 = insertelement <2 x bfloat> poison, bfloat %2210, i64 0, !dbg !44 + %2298 = insertelement <2 x bfloat> %2297, bfloat %2211, i64 1, !dbg !44 + %2299 = bitcast <2 x bfloat> %2298 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %2290, i32 %2293, i32 %2296, i32 %2299, ptr addrspace(1) %151, i1 %200) #4, !dbg !44 + %2300 = bitcast <2 x bfloat> %2212 to i32, !dbg !44 + %2301 = bitcast <2 x bfloat> %2213 to i32, !dbg !44 + %2302 = bitcast <2 x bfloat> %2214 to i32, !dbg !44 + %2303 = bitcast <2 x bfloat> %2215 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %2300, i32 %2301, i32 %2302, i32 %2303, ptr addrspace(1) %152, i1 %218) #4, !dbg !44 + %2304 = bitcast <2 x bfloat> %2216 to i32, !dbg !44 + %2305 = bitcast <2 x bfloat> %2217 to i32, !dbg !44 + %2306 = bitcast <2 x bfloat> %2218 to i32, !dbg !44 + %2307 = bitcast <2 x bfloat> %2219 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %2304, i32 %2305, i32 %2306, i32 %2307, ptr addrspace(1) %153, i1 %236) #4, !dbg !44 + %2308 = bitcast <2 x bfloat> %2220 to i32, !dbg !44 + %2309 = bitcast <2 x bfloat> %2221 to i32, !dbg !44 + %2310 = bitcast <2 x bfloat> %2222 to i32, !dbg !44 + %2311 = bitcast <2 x bfloat> %2223 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %2308, i32 %2309, i32 %2310, i32 %2311, ptr addrspace(1) %154, i1 %254) #4, !dbg !44 + %2312 = bitcast <2 x bfloat> %2224 to i32, !dbg !44 + %2313 = bitcast <2 x bfloat> %2225 to i32, !dbg !44 + %2314 = bitcast <2 x bfloat> %2226 to i32, !dbg !44 + %2315 = bitcast <2 x bfloat> %2227 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %2312, i32 %2313, i32 %2314, i32 %2315, ptr addrspace(1) %155, i1 %272) #4, !dbg !44 + %2316 = bitcast <2 x bfloat> %2228 to i32, !dbg !44 + %2317 = bitcast <2 x bfloat> %2229 to i32, !dbg !44 + %2318 = bitcast <2 x bfloat> %2230 to i32, !dbg !44 + %2319 = bitcast <2 x bfloat> %2231 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %2316, i32 %2317, i32 %2318, i32 %2319, ptr addrspace(1) %156, i1 %290) #4, !dbg !44 + %2320 = bitcast <2 x bfloat> %2232 to i32, !dbg !44 + %2321 = bitcast <2 x bfloat> %2233 to i32, !dbg !44 + %2322 = bitcast <2 x bfloat> %2234 to i32, !dbg !44 + %2323 = bitcast <2 x bfloat> %2235 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %2320, i32 %2321, i32 %2322, i32 %2323, ptr addrspace(1) %157, i1 %308) #4, !dbg !44 + %2324 = bitcast <2 x bfloat> %2236 to i32, !dbg !44 + %2325 = bitcast <2 x bfloat> %2237 to i32, !dbg !44 + %2326 = bitcast <2 x bfloat> %2238 to i32, !dbg !44 + %2327 = bitcast <2 x bfloat> %2239 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %2324, i32 %2325, i32 %2326, i32 %2327, ptr addrspace(1) %158, i1 %326) #4, !dbg !44 + %2328 = bitcast <2 x bfloat> %2240 to i32, !dbg !44 + %2329 = bitcast <2 x bfloat> %2241 to i32, !dbg !44 + %2330 = bitcast <2 x bfloat> %2242 to i32, !dbg !44 + %2331 = bitcast <2 x bfloat> %2243 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %2328, i32 %2329, i32 %2330, i32 %2331, ptr addrspace(1) %159, i1 %344) #4, !dbg !44 + %2332 = bitcast <2 x bfloat> %2244 to i32, !dbg !44 + %2333 = bitcast <2 x bfloat> %2245 to i32, !dbg !44 + %2334 = bitcast <2 x bfloat> %2246 to i32, !dbg !44 + %2335 = bitcast <2 x bfloat> %2247 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %2332, i32 %2333, i32 %2334, i32 %2335, ptr addrspace(1) %160, i1 %362) #4, !dbg !44 + %2336 = bitcast <2 x bfloat> %2248 to i32, !dbg !44 + %2337 = bitcast <2 x bfloat> %2249 to i32, !dbg !44 + %2338 = bitcast <2 x bfloat> %2250 to i32, !dbg !44 + %2339 = bitcast <2 x bfloat> %2251 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %2336, i32 %2337, i32 %2338, i32 %2339, ptr addrspace(1) %161, i1 %380) #4, !dbg !44 + %2340 = bitcast <2 x bfloat> %2252 to i32, !dbg !44 + %2341 = bitcast <2 x bfloat> %2253 to i32, !dbg !44 + %2342 = bitcast <2 x bfloat> %2254 to i32, !dbg !44 + %2343 = bitcast <2 x bfloat> %2255 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %2340, i32 %2341, i32 %2342, i32 %2343, ptr addrspace(1) %162, i1 %398) #4, !dbg !44 + %2344 = bitcast <2 x bfloat> %2256 to i32, !dbg !44 + %2345 = bitcast <2 x bfloat> %2257 to i32, !dbg !44 + %2346 = bitcast <2 x bfloat> %2258 to i32, !dbg !44 + %2347 = bitcast <2 x bfloat> %2259 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %2344, i32 %2345, i32 %2346, i32 %2347, ptr addrspace(1) %163, i1 %416) #4, !dbg !44 + %2348 = bitcast <2 x bfloat> %2260 to i32, !dbg !44 + %2349 = bitcast <2 x bfloat> %2261 to i32, !dbg !44 + %2350 = bitcast <2 x bfloat> %2262 to i32, !dbg !44 + %2351 = bitcast <2 x bfloat> %2263 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %2348, i32 %2349, i32 %2350, i32 %2351, ptr addrspace(1) %164, i1 %434) #4, !dbg !44 + ret void, !dbg !45 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #2 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="64" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind } +attributes #3 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #4 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = distinct !DISubprogram(name: "l2norm_bwd_kernel", linkageName: "l2norm_bwd_kernel", scope: !1, file: !1, line: 93, type: !5, scopeLine: 93, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!5 = !DISubroutineType(cc: DW_CC_normal, types: !6) +!6 = !{null, !7, !9, !7, !7, !10, !11, !7, !7} +!7 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !8, size: 64, dwarfAddressSpace: 1) +!8 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!9 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !10, size: 64, dwarfAddressSpace: 1) +!10 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!11 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!12 = !DILocation(line: 94, column: 24, scope: !4) +!13 = !DILocation(line: 94, column: 30, scope: !4) +!14 = !DILocation(line: 95, column: 16, scope: !4) +!15 = !DILocation(line: 97, column: 16, scope: !4) +!16 = !DILocation(line: 95, column: 34, scope: !4) +!17 = !DILocation(line: 95, column: 21, scope: !4) +!18 = !DILocation(line: 98, column: 30, scope: !4) +!19 = !DILocation(line: 99, column: 29, scope: !4) +!20 = !DILocation(line: 99, column: 14, scope: !4) +!21 = !DILocation(line: 99, column: 33, scope: !4) +!22 = !DILocation(line: 100, column: 20, scope: !4) +!23 = !DILocation(line: 101, column: 16, scope: !4) +!24 = !DILocation(line: 101, column: 35, scope: !4) +!25 = !DILocation(line: 102, column: 16, scope: !4) +!26 = !DILocation(line: 102, column: 35, scope: !4) +!27 = !DILocation(line: 104, column: 18, scope: !4) +!28 = !DILocation(line: 104, column: 47, scope: !4) +!29 = !DILocation(line: 105, column: 21, scope: !4) +!30 = !DILocation(line: 106, column: 19, scope: !4) +!31 = !DILocation(line: 106, column: 49, scope: !4) +!32 = !DILocation(line: 107, column: 18, scope: !4) +!33 = !DILocation(line: 107, column: 50, scope: !4) +!34 = !DILocation(line: 263, column: 15, scope: !35, inlinedAt: !37) +!35 = distinct !DILexicalBlockFile(scope: !4, file: !36, discriminator: 0) +!36 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!37 = !DILocation(line: 293, column: 36, scope: !35, inlinedAt: !38) +!38 = !DILocation(line: 107, column: 55, scope: !39) +!39 = distinct !DILexicalBlockFile(scope: !4, file: !1, discriminator: 0) +!40 = !DILocation(line: 107, column: 69, scope: !4) +!41 = !DILocation(line: 107, column: 75, scope: !4) +!42 = !DILocation(line: 107, column: 36, scope: !4) +!43 = !DILocation(line: 108, column: 27, scope: !4) +!44 = !DILocation(line: 108, column: 19, scope: !4) +!45 = !DILocation(line: 108, column: 4, scope: !4) diff --git a/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.ptx b/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..5226c9b8269548d7abb202df7075e563d072c646 --- /dev/null +++ b/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.ptx @@ -0,0 +1,2352 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_bwd_kernel // -- Begin function l2norm_bwd_kernel +.extern .shared .align 16 .b8 global_smem[]; + // @l2norm_bwd_kernel +.visible .entry l2norm_bwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_2, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_3, + .param .f32 l2norm_bwd_kernel_param_4, + .param .u32 l2norm_bwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_6, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_7 +) +.reqntid 64 +{ + .reg .pred %p<18>; + .reg .b16 %rs<257>; + .reg .b32 %r<1268>; + .reg .b64 %rd<312>; + .loc 1 93 0 // l2norm.py:93:0 +$L__func_begin0: + .loc 1 93 0 // l2norm.py:93:0 + +// %bb.0: + ld.param.b64 %rd50, [l2norm_bwd_kernel_param_0]; + ld.param.b64 %rd51, [l2norm_bwd_kernel_param_1]; +$L__tmp0: + .loc 1 94 24 // l2norm.py:94:24 + mov.u32 %r196, %ctaid.x; + ld.param.b64 %rd52, [l2norm_bwd_kernel_param_2]; + .loc 1 95 16 // l2norm.py:95:16 + mul.wide.u32 %rd53, %r196, 64; + ld.param.b64 %rd54, [l2norm_bwd_kernel_param_3]; + ld.param.s32 %rd55, [l2norm_bwd_kernel_param_5]; + .loc 1 95 34 // l2norm.py:95:34 + mov.u32 %r197, %tid.x; + and.b32 %r198, %r197, 48; + bfe.u32 %r199, %r197, 4, 2; + and.b32 %r200, %r197, 63; + or.b32 %r201, %r199, 4; + or.b32 %r202, %r199, 8; + or.b32 %r203, %r199, 12; + or.b32 %r204, %r199, 16; + or.b32 %r205, %r199, 20; + or.b32 %r206, %r199, 24; + or.b32 %r207, %r199, 28; + or.b32 %r208, %r199, 32; + or.b32 %r209, %r199, 36; + or.b32 %r210, %r199, 40; + or.b32 %r211, %r199, 44; + or.b32 %r212, %r199, 48; + or.b32 %r213, %r199, 52; + or.b32 %r214, %r199, 56; + or.b32 %r215, %r199, 60; + .loc 1 95 21 // l2norm.py:95:21 + cvt.u64.u32 %rd56, %r199; + cvt.u64.u32 %rd57, %r200; + cvt.u64.u32 %rd58, %r215; + cvt.u64.u32 %rd59, %r214; + cvt.u64.u32 %rd60, %r213; + cvt.u64.u32 %rd61, %r212; + cvt.u64.u32 %rd62, %r211; + cvt.u64.u32 %rd63, %r210; + cvt.u64.u32 %rd64, %r209; + cvt.u64.u32 %rd65, %r208; + cvt.u64.u32 %rd66, %r207; + cvt.u64.u32 %rd67, %r206; + cvt.u64.u32 %rd68, %r205; + cvt.u64.u32 %rd69, %r204; + cvt.u64.u32 %rd70, %r203; + cvt.u64.u32 %rd71, %r202; + cvt.u64.u32 %rd72, %r201; + or.b64 %rd73, %rd53, %rd56; + or.b64 %rd74, %rd53, %rd72; + or.b64 %rd75, %rd53, %rd71; + or.b64 %rd76, %rd53, %rd70; + or.b64 %rd77, %rd53, %rd69; + or.b64 %rd78, %rd53, %rd68; + or.b64 %rd79, %rd53, %rd67; + or.b64 %rd80, %rd53, %rd66; + or.b64 %rd81, %rd53, %rd65; + or.b64 %rd82, %rd53, %rd64; + or.b64 %rd83, %rd53, %rd63; + or.b64 %rd84, %rd53, %rd62; + or.b64 %rd85, %rd53, %rd61; + or.b64 %rd86, %rd53, %rd60; + or.b64 %rd87, %rd53, %rd59; + or.b64 %rd88, %rd53, %rd58; + or.b64 %rd89, %rd53, %rd57; + .loc 1 97 16 // l2norm.py:97:16 + setp.lt.s64 %p1, %rd73, %rd55; + setp.lt.s64 %p16, %rd89, %rd55; + setp.lt.s64 %p15, %rd88, %rd55; + setp.lt.s64 %p14, %rd87, %rd55; + setp.lt.s64 %p13, %rd86, %rd55; + setp.lt.s64 %p12, %rd85, %rd55; + setp.lt.s64 %p11, %rd84, %rd55; + setp.lt.s64 %p10, %rd83, %rd55; + setp.lt.s64 %p9, %rd82, %rd55; + setp.lt.s64 %p17, %rd81, %rd55; + setp.lt.s64 %p8, %rd80, %rd55; + setp.lt.s64 %p7, %rd79, %rd55; + setp.lt.s64 %p6, %rd78, %rd55; + setp.lt.s64 %p5, %rd77, %rd55; + setp.lt.s64 %p4, %rd76, %rd55; + setp.lt.s64 %p3, %rd75, %rd55; + setp.lt.s64 %p2, %rd74, %rd55; + .loc 1 98 30 // l2norm.py:98:30 + shl.b32 %r216, %r197, 3; + and.b32 %r217, %r216, 120; + .loc 1 99 14 // l2norm.py:99:14 + shl.b64 %rd90, %rd73, 8; + add.s64 %rd91, %rd50, %rd90; + shl.b64 %rd92, %rd74, 8; + add.s64 %rd93, %rd50, %rd92; + shl.b64 %rd94, %rd75, 8; + add.s64 %rd95, %rd50, %rd94; + shl.b64 %rd96, %rd76, 8; + add.s64 %rd97, %rd50, %rd96; + shl.b64 %rd98, %rd77, 8; + add.s64 %rd99, %rd50, %rd98; + shl.b64 %rd100, %rd78, 8; + add.s64 %rd101, %rd50, %rd100; + shl.b64 %rd102, %rd79, 8; + add.s64 %rd103, %rd50, %rd102; + shl.b64 %rd104, %rd80, 8; + add.s64 %rd105, %rd50, %rd104; + shl.b64 %rd106, %rd81, 8; + add.s64 %rd107, %rd50, %rd106; + shl.b64 %rd108, %rd82, 8; + add.s64 %rd109, %rd50, %rd108; + shl.b64 %rd110, %rd83, 8; + add.s64 %rd111, %rd50, %rd110; + shl.b64 %rd112, %rd84, 8; + add.s64 %rd113, %rd50, %rd112; + shl.b64 %rd114, %rd85, 8; + add.s64 %rd115, %rd50, %rd114; + shl.b64 %rd116, %rd86, 8; + add.s64 %rd117, %rd50, %rd116; + shl.b64 %rd118, %rd87, 8; + add.s64 %rd119, %rd50, %rd118; + shl.b64 %rd120, %rd88, 8; + add.s64 %rd121, %rd50, %rd120; + .loc 1 99 33 // l2norm.py:99:33 + mul.wide.u32 %rd122, %r217, 2; + add.s64 %rd1, %rd91, %rd122; + add.s64 %rd2, %rd93, %rd122; + add.s64 %rd3, %rd95, %rd122; + add.s64 %rd4, %rd97, %rd122; + add.s64 %rd5, %rd99, %rd122; + add.s64 %rd6, %rd101, %rd122; + add.s64 %rd7, %rd103, %rd122; + add.s64 %rd8, %rd105, %rd122; + add.s64 %rd9, %rd107, %rd122; + add.s64 %rd10, %rd109, %rd122; + add.s64 %rd11, %rd111, %rd122; + add.s64 %rd12, %rd113, %rd122; + add.s64 %rd13, %rd115, %rd122; + add.s64 %rd14, %rd117, %rd122; + add.s64 %rd15, %rd119, %rd122; + add.s64 %rd16, %rd121, %rd122; + .loc 1 100 20 // l2norm.py:100:20 + shl.b64 %rd123, %rd89, 2; + add.s64 %rd17, %rd51, %rd123; + .loc 1 101 16 // l2norm.py:101:16 + add.s64 %rd124, %rd52, %rd90; + add.s64 %rd125, %rd52, %rd92; + add.s64 %rd126, %rd52, %rd94; + add.s64 %rd127, %rd52, %rd96; + add.s64 %rd128, %rd52, %rd98; + add.s64 %rd129, %rd52, %rd100; + add.s64 %rd130, %rd52, %rd102; + add.s64 %rd131, %rd52, %rd104; + add.s64 %rd132, %rd52, %rd106; + add.s64 %rd133, %rd52, %rd108; + add.s64 %rd134, %rd52, %rd110; + add.s64 %rd135, %rd52, %rd112; + add.s64 %rd136, %rd52, %rd114; + add.s64 %rd137, %rd52, %rd116; + add.s64 %rd138, %rd52, %rd118; + add.s64 %rd139, %rd52, %rd120; + .loc 1 101 35 // l2norm.py:101:35 + add.s64 %rd18, %rd124, %rd122; + add.s64 %rd19, %rd125, %rd122; + add.s64 %rd20, %rd126, %rd122; + add.s64 %rd21, %rd127, %rd122; + add.s64 %rd22, %rd128, %rd122; + add.s64 %rd23, %rd129, %rd122; + add.s64 %rd24, %rd130, %rd122; + add.s64 %rd25, %rd131, %rd122; + add.s64 %rd26, %rd132, %rd122; + add.s64 %rd27, %rd133, %rd122; + add.s64 %rd28, %rd134, %rd122; + add.s64 %rd29, %rd135, %rd122; + add.s64 %rd30, %rd136, %rd122; + add.s64 %rd31, %rd137, %rd122; + add.s64 %rd32, %rd138, %rd122; + add.s64 %rd33, %rd139, %rd122; + .loc 1 102 16 // l2norm.py:102:16 + add.s64 %rd140, %rd54, %rd90; + add.s64 %rd141, %rd54, %rd92; + add.s64 %rd142, %rd54, %rd94; + add.s64 %rd143, %rd54, %rd96; + add.s64 %rd144, %rd54, %rd98; + add.s64 %rd145, %rd54, %rd100; + add.s64 %rd146, %rd54, %rd102; + add.s64 %rd147, %rd54, %rd104; + add.s64 %rd148, %rd54, %rd106; + add.s64 %rd149, %rd54, %rd108; + add.s64 %rd150, %rd54, %rd110; + add.s64 %rd151, %rd54, %rd112; + add.s64 %rd152, %rd54, %rd114; + add.s64 %rd153, %rd54, %rd116; + add.s64 %rd154, %rd54, %rd118; + add.s64 %rd155, %rd54, %rd120; + .loc 1 102 35 // l2norm.py:102:35 + add.s64 %rd34, %rd140, %rd122; + add.s64 %rd35, %rd141, %rd122; + add.s64 %rd36, %rd142, %rd122; + add.s64 %rd37, %rd143, %rd122; + add.s64 %rd38, %rd144, %rd122; + add.s64 %rd39, %rd145, %rd122; + add.s64 %rd40, %rd146, %rd122; + add.s64 %rd41, %rd147, %rd122; + add.s64 %rd42, %rd148, %rd122; + add.s64 %rd43, %rd149, %rd122; + add.s64 %rd44, %rd150, %rd122; + add.s64 %rd45, %rd151, %rd122; + add.s64 %rd46, %rd152, %rd122; + add.s64 %rd47, %rd153, %rd122; + add.s64 %rd48, %rd154, %rd122; + add.s64 %rd49, %rd155, %rd122; + mov.b32 %r5, 0; + .loc 1 104 18 // l2norm.py:104:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + mov.b32 {%rs5, %rs6}, %r3; + mov.b32 {%rs7, %rs8}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs9, %rs10}, %r6; + mov.b32 {%rs11, %rs12}, %r7; + mov.b32 {%rs13, %rs14}, %r8; + mov.b32 {%rs15, %rs16}, %r9; + // begin inline asm + mov.u32 %r10, %r5; + mov.u32 %r11, %r5; + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + @%p3 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs17, %rs18}, %r10; + mov.b32 {%rs19, %rs20}, %r11; + mov.b32 {%rs21, %rs22}, %r12; + mov.b32 {%rs23, %rs24}, %r13; + // begin inline asm + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + @%p4 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ]; + // end inline asm + mov.b32 {%rs25, %rs26}, %r14; + mov.b32 {%rs27, %rs28}, %r15; + mov.b32 {%rs29, %rs30}, %r16; + mov.b32 {%rs31, %rs32}, %r17; + // begin inline asm + mov.u32 %r18, %r5; + mov.u32 %r19, %r5; + mov.u32 %r20, %r5; + mov.u32 %r21, %r5; + @%p5 ld.global.v4.b32 { %r18, %r19, %r20, %r21 }, [ %rd5 + 0 ]; + // end inline asm + mov.b32 {%rs33, %rs34}, %r18; + mov.b32 {%rs35, %rs36}, %r19; + mov.b32 {%rs37, %rs38}, %r20; + mov.b32 {%rs39, %rs40}, %r21; + // begin inline asm + mov.u32 %r22, %r5; + mov.u32 %r23, %r5; + mov.u32 %r24, %r5; + mov.u32 %r25, %r5; + @%p6 ld.global.v4.b32 { %r22, %r23, %r24, %r25 }, [ %rd6 + 0 ]; + // end inline asm + mov.b32 {%rs41, %rs42}, %r22; + mov.b32 {%rs43, %rs44}, %r23; + mov.b32 {%rs45, %rs46}, %r24; + mov.b32 {%rs47, %rs48}, %r25; + // begin inline asm + mov.u32 %r26, %r5; + mov.u32 %r27, %r5; + mov.u32 %r28, %r5; + mov.u32 %r29, %r5; + @%p7 ld.global.v4.b32 { %r26, %r27, %r28, %r29 }, [ %rd7 + 0 ]; + // end inline asm + mov.b32 {%rs49, %rs50}, %r26; + mov.b32 {%rs51, %rs52}, %r27; + mov.b32 {%rs53, %rs54}, %r28; + mov.b32 {%rs55, %rs56}, %r29; + // begin inline asm + mov.u32 %r30, %r5; + mov.u32 %r31, %r5; + mov.u32 %r32, %r5; + mov.u32 %r33, %r5; + @%p8 ld.global.v4.b32 { %r30, %r31, %r32, %r33 }, [ %rd8 + 0 ]; + // end inline asm + mov.b32 {%rs57, %rs58}, %r30; + mov.b32 {%rs59, %rs60}, %r31; + mov.b32 {%rs61, %rs62}, %r32; + mov.b32 {%rs63, %rs64}, %r33; + // begin inline asm + mov.u32 %r34, %r5; + mov.u32 %r35, %r5; + mov.u32 %r36, %r5; + mov.u32 %r37, %r5; + @%p17 ld.global.v4.b32 { %r34, %r35, %r36, %r37 }, [ %rd9 + 0 ]; + // end inline asm + mov.b32 {%rs65, %rs66}, %r34; + mov.b32 {%rs67, %rs68}, %r35; + mov.b32 {%rs69, %rs70}, %r36; + mov.b32 {%rs71, %rs72}, %r37; + // begin inline asm + mov.u32 %r38, %r5; + mov.u32 %r39, %r5; + mov.u32 %r40, %r5; + mov.u32 %r41, %r5; + @%p9 ld.global.v4.b32 { %r38, %r39, %r40, %r41 }, [ %rd10 + 0 ]; + // end inline asm + mov.b32 {%rs73, %rs74}, %r38; + mov.b32 {%rs75, %rs76}, %r39; + mov.b32 {%rs77, %rs78}, %r40; + mov.b32 {%rs79, %rs80}, %r41; + // begin inline asm + mov.u32 %r42, %r5; + mov.u32 %r43, %r5; + mov.u32 %r44, %r5; + mov.u32 %r45, %r5; + @%p10 ld.global.v4.b32 { %r42, %r43, %r44, %r45 }, [ %rd11 + 0 ]; + // end inline asm + mov.b32 {%rs81, %rs82}, %r42; + mov.b32 {%rs83, %rs84}, %r43; + mov.b32 {%rs85, %rs86}, %r44; + mov.b32 {%rs87, %rs88}, %r45; + // begin inline asm + mov.u32 %r46, %r5; + mov.u32 %r47, %r5; + mov.u32 %r48, %r5; + mov.u32 %r49, %r5; + @%p11 ld.global.v4.b32 { %r46, %r47, %r48, %r49 }, [ %rd12 + 0 ]; + // end inline asm + mov.b32 {%rs89, %rs90}, %r46; + mov.b32 {%rs91, %rs92}, %r47; + mov.b32 {%rs93, %rs94}, %r48; + mov.b32 {%rs95, %rs96}, %r49; + // begin inline asm + mov.u32 %r50, %r5; + mov.u32 %r51, %r5; + mov.u32 %r52, %r5; + mov.u32 %r53, %r5; + @%p12 ld.global.v4.b32 { %r50, %r51, %r52, %r53 }, [ %rd13 + 0 ]; + // end inline asm + mov.b32 {%rs97, %rs98}, %r50; + mov.b32 {%rs99, %rs100}, %r51; + mov.b32 {%rs101, %rs102}, %r52; + mov.b32 {%rs103, %rs104}, %r53; + // begin inline asm + mov.u32 %r54, %r5; + mov.u32 %r55, %r5; + mov.u32 %r56, %r5; + mov.u32 %r57, %r5; + @%p13 ld.global.v4.b32 { %r54, %r55, %r56, %r57 }, [ %rd14 + 0 ]; + // end inline asm + mov.b32 {%rs105, %rs106}, %r54; + mov.b32 {%rs107, %rs108}, %r55; + mov.b32 {%rs109, %rs110}, %r56; + mov.b32 {%rs111, %rs112}, %r57; + // begin inline asm + mov.u32 %r58, %r5; + mov.u32 %r59, %r5; + mov.u32 %r60, %r5; + mov.u32 %r61, %r5; + @%p14 ld.global.v4.b32 { %r58, %r59, %r60, %r61 }, [ %rd15 + 0 ]; + // end inline asm + mov.b32 {%rs113, %rs114}, %r58; + mov.b32 {%rs115, %rs116}, %r59; + mov.b32 {%rs117, %rs118}, %r60; + mov.b32 {%rs119, %rs120}, %r61; + // begin inline asm + mov.u32 %r62, %r5; + mov.u32 %r63, %r5; + mov.u32 %r64, %r5; + mov.u32 %r65, %r5; + @%p15 ld.global.v4.b32 { %r62, %r63, %r64, %r65 }, [ %rd16 + 0 ]; + // end inline asm + mov.b32 {%rs121, %rs122}, %r62; + mov.b32 {%rs123, %rs124}, %r63; + mov.b32 {%rs125, %rs126}, %r64; + mov.b32 {%rs127, %rs128}, %r65; + .loc 1 104 47 // l2norm.py:104:47 + cvt.f32.bf16 %r218, %rs1; + cvt.f32.bf16 %r219, %rs2; + cvt.f32.bf16 %r220, %rs3; + cvt.f32.bf16 %r221, %rs4; + cvt.f32.bf16 %r222, %rs5; + cvt.f32.bf16 %r223, %rs6; + cvt.f32.bf16 %r224, %rs7; + cvt.f32.bf16 %r225, %rs8; + cvt.f32.bf16 %r226, %rs9; + cvt.f32.bf16 %r227, %rs10; + cvt.f32.bf16 %r228, %rs11; + cvt.f32.bf16 %r229, %rs12; + cvt.f32.bf16 %r230, %rs13; + cvt.f32.bf16 %r231, %rs14; + cvt.f32.bf16 %r232, %rs15; + cvt.f32.bf16 %r233, %rs16; + cvt.f32.bf16 %r234, %rs17; + cvt.f32.bf16 %r235, %rs18; + cvt.f32.bf16 %r236, %rs19; + cvt.f32.bf16 %r237, %rs20; + cvt.f32.bf16 %r238, %rs21; + cvt.f32.bf16 %r239, %rs22; + cvt.f32.bf16 %r240, %rs23; + cvt.f32.bf16 %r241, %rs24; + cvt.f32.bf16 %r242, %rs25; + cvt.f32.bf16 %r243, %rs26; + cvt.f32.bf16 %r244, %rs27; + cvt.f32.bf16 %r245, %rs28; + cvt.f32.bf16 %r246, %rs29; + cvt.f32.bf16 %r247, %rs30; + cvt.f32.bf16 %r248, %rs31; + cvt.f32.bf16 %r249, %rs32; + cvt.f32.bf16 %r250, %rs33; + cvt.f32.bf16 %r251, %rs34; + cvt.f32.bf16 %r252, %rs35; + cvt.f32.bf16 %r253, %rs36; + cvt.f32.bf16 %r254, %rs37; + cvt.f32.bf16 %r255, %rs38; + cvt.f32.bf16 %r256, %rs39; + cvt.f32.bf16 %r257, %rs40; + cvt.f32.bf16 %r258, %rs41; + cvt.f32.bf16 %r259, %rs42; + cvt.f32.bf16 %r260, %rs43; + cvt.f32.bf16 %r261, %rs44; + cvt.f32.bf16 %r262, %rs45; + cvt.f32.bf16 %r263, %rs46; + cvt.f32.bf16 %r264, %rs47; + cvt.f32.bf16 %r265, %rs48; + cvt.f32.bf16 %r266, %rs49; + cvt.f32.bf16 %r267, %rs50; + cvt.f32.bf16 %r268, %rs51; + cvt.f32.bf16 %r269, %rs52; + cvt.f32.bf16 %r270, %rs53; + cvt.f32.bf16 %r271, %rs54; + cvt.f32.bf16 %r272, %rs55; + cvt.f32.bf16 %r273, %rs56; + cvt.f32.bf16 %r274, %rs57; + cvt.f32.bf16 %r275, %rs58; + cvt.f32.bf16 %r276, %rs59; + cvt.f32.bf16 %r277, %rs60; + cvt.f32.bf16 %r278, %rs61; + cvt.f32.bf16 %r279, %rs62; + cvt.f32.bf16 %r280, %rs63; + cvt.f32.bf16 %r281, %rs64; + cvt.f32.bf16 %r282, %rs65; + cvt.f32.bf16 %r283, %rs66; + cvt.f32.bf16 %r284, %rs67; + cvt.f32.bf16 %r285, %rs68; + cvt.f32.bf16 %r286, %rs69; + cvt.f32.bf16 %r287, %rs70; + cvt.f32.bf16 %r288, %rs71; + cvt.f32.bf16 %r289, %rs72; + cvt.f32.bf16 %r290, %rs73; + cvt.f32.bf16 %r291, %rs74; + cvt.f32.bf16 %r292, %rs75; + cvt.f32.bf16 %r293, %rs76; + cvt.f32.bf16 %r294, %rs77; + cvt.f32.bf16 %r295, %rs78; + cvt.f32.bf16 %r296, %rs79; + cvt.f32.bf16 %r297, %rs80; + cvt.f32.bf16 %r298, %rs81; + cvt.f32.bf16 %r299, %rs82; + cvt.f32.bf16 %r300, %rs83; + cvt.f32.bf16 %r301, %rs84; + cvt.f32.bf16 %r302, %rs85; + cvt.f32.bf16 %r303, %rs86; + cvt.f32.bf16 %r304, %rs87; + cvt.f32.bf16 %r305, %rs88; + cvt.f32.bf16 %r306, %rs89; + cvt.f32.bf16 %r307, %rs90; + cvt.f32.bf16 %r308, %rs91; + cvt.f32.bf16 %r309, %rs92; + cvt.f32.bf16 %r310, %rs93; + cvt.f32.bf16 %r311, %rs94; + cvt.f32.bf16 %r312, %rs95; + cvt.f32.bf16 %r313, %rs96; + cvt.f32.bf16 %r314, %rs97; + cvt.f32.bf16 %r315, %rs98; + cvt.f32.bf16 %r316, %rs99; + cvt.f32.bf16 %r317, %rs100; + cvt.f32.bf16 %r318, %rs101; + cvt.f32.bf16 %r319, %rs102; + cvt.f32.bf16 %r320, %rs103; + cvt.f32.bf16 %r321, %rs104; + cvt.f32.bf16 %r322, %rs105; + cvt.f32.bf16 %r323, %rs106; + cvt.f32.bf16 %r324, %rs107; + cvt.f32.bf16 %r325, %rs108; + cvt.f32.bf16 %r326, %rs109; + cvt.f32.bf16 %r327, %rs110; + cvt.f32.bf16 %r328, %rs111; + cvt.f32.bf16 %r329, %rs112; + cvt.f32.bf16 %r330, %rs113; + cvt.f32.bf16 %r331, %rs114; + cvt.f32.bf16 %r332, %rs115; + cvt.f32.bf16 %r333, %rs116; + cvt.f32.bf16 %r334, %rs117; + cvt.f32.bf16 %r335, %rs118; + cvt.f32.bf16 %r336, %rs119; + cvt.f32.bf16 %r337, %rs120; + cvt.f32.bf16 %r338, %rs121; + cvt.f32.bf16 %r339, %rs122; + cvt.f32.bf16 %r340, %rs123; + cvt.f32.bf16 %r341, %rs124; + cvt.f32.bf16 %r342, %rs125; + cvt.f32.bf16 %r343, %rs126; + cvt.f32.bf16 %r344, %rs127; + cvt.f32.bf16 %r345, %rs128; + .loc 1 105 21 // l2norm.py:105:21 + // begin inline asm + mov.u32 %r131, %r5; + @%p16 ld.global.b32 { %r131 }, [ %rd17 + 0 ]; + // end inline asm + .loc 1 106 19 // l2norm.py:106:19 + // begin inline asm + mov.u32 %r66, %r5; + mov.u32 %r67, %r5; + mov.u32 %r68, %r5; + mov.u32 %r69, %r5; + @%p1 ld.global.v4.b32 { %r66, %r67, %r68, %r69 }, [ %rd18 + 0 ]; + // end inline asm + mov.b32 {%rs129, %rs130}, %r66; + mov.b32 {%rs131, %rs132}, %r67; + mov.b32 {%rs133, %rs134}, %r68; + mov.b32 {%rs135, %rs136}, %r69; + // begin inline asm + mov.u32 %r70, %r5; + mov.u32 %r71, %r5; + mov.u32 %r72, %r5; + mov.u32 %r73, %r5; + @%p2 ld.global.v4.b32 { %r70, %r71, %r72, %r73 }, [ %rd19 + 0 ]; + // end inline asm + mov.b32 {%rs137, %rs138}, %r70; + mov.b32 {%rs139, %rs140}, %r71; + mov.b32 {%rs141, %rs142}, %r72; + mov.b32 {%rs143, %rs144}, %r73; + // begin inline asm + mov.u32 %r74, %r5; + mov.u32 %r75, %r5; + mov.u32 %r76, %r5; + mov.u32 %r77, %r5; + @%p3 ld.global.v4.b32 { %r74, %r75, %r76, %r77 }, [ %rd20 + 0 ]; + // end inline asm + mov.b32 {%rs145, %rs146}, %r74; + mov.b32 {%rs147, %rs148}, %r75; + mov.b32 {%rs149, %rs150}, %r76; + mov.b32 {%rs151, %rs152}, %r77; + // begin inline asm + mov.u32 %r78, %r5; + mov.u32 %r79, %r5; + mov.u32 %r80, %r5; + mov.u32 %r81, %r5; + @%p4 ld.global.v4.b32 { %r78, %r79, %r80, %r81 }, [ %rd21 + 0 ]; + // end inline asm + mov.b32 {%rs153, %rs154}, %r78; + mov.b32 {%rs155, %rs156}, %r79; + mov.b32 {%rs157, %rs158}, %r80; + mov.b32 {%rs159, %rs160}, %r81; + // begin inline asm + mov.u32 %r82, %r5; + mov.u32 %r83, %r5; + mov.u32 %r84, %r5; + mov.u32 %r85, %r5; + @%p5 ld.global.v4.b32 { %r82, %r83, %r84, %r85 }, [ %rd22 + 0 ]; + // end inline asm + mov.b32 {%rs161, %rs162}, %r82; + mov.b32 {%rs163, %rs164}, %r83; + mov.b32 {%rs165, %rs166}, %r84; + mov.b32 {%rs167, %rs168}, %r85; + // begin inline asm + mov.u32 %r86, %r5; + mov.u32 %r87, %r5; + mov.u32 %r88, %r5; + mov.u32 %r89, %r5; + @%p6 ld.global.v4.b32 { %r86, %r87, %r88, %r89 }, [ %rd23 + 0 ]; + // end inline asm + mov.b32 {%rs169, %rs170}, %r86; + mov.b32 {%rs171, %rs172}, %r87; + mov.b32 {%rs173, %rs174}, %r88; + mov.b32 {%rs175, %rs176}, %r89; + // begin inline asm + mov.u32 %r90, %r5; + mov.u32 %r91, %r5; + mov.u32 %r92, %r5; + mov.u32 %r93, %r5; + @%p7 ld.global.v4.b32 { %r90, %r91, %r92, %r93 }, [ %rd24 + 0 ]; + // end inline asm + mov.b32 {%rs177, %rs178}, %r90; + mov.b32 {%rs179, %rs180}, %r91; + mov.b32 {%rs181, %rs182}, %r92; + mov.b32 {%rs183, %rs184}, %r93; + // begin inline asm + mov.u32 %r94, %r5; + mov.u32 %r95, %r5; + mov.u32 %r96, %r5; + mov.u32 %r97, %r5; + @%p8 ld.global.v4.b32 { %r94, %r95, %r96, %r97 }, [ %rd25 + 0 ]; + // end inline asm + mov.b32 {%rs185, %rs186}, %r94; + mov.b32 {%rs187, %rs188}, %r95; + mov.b32 {%rs189, %rs190}, %r96; + mov.b32 {%rs191, %rs192}, %r97; + // begin inline asm + mov.u32 %r98, %r5; + mov.u32 %r99, %r5; + mov.u32 %r100, %r5; + mov.u32 %r101, %r5; + @%p17 ld.global.v4.b32 { %r98, %r99, %r100, %r101 }, [ %rd26 + 0 ]; + // end inline asm + mov.b32 {%rs193, %rs194}, %r98; + mov.b32 {%rs195, %rs196}, %r99; + mov.b32 {%rs197, %rs198}, %r100; + mov.b32 {%rs199, %rs200}, %r101; + // begin inline asm + mov.u32 %r102, %r5; + mov.u32 %r103, %r5; + mov.u32 %r104, %r5; + mov.u32 %r105, %r5; + @%p9 ld.global.v4.b32 { %r102, %r103, %r104, %r105 }, [ %rd27 + 0 ]; + // end inline asm + mov.b32 {%rs201, %rs202}, %r102; + mov.b32 {%rs203, %rs204}, %r103; + mov.b32 {%rs205, %rs206}, %r104; + mov.b32 {%rs207, %rs208}, %r105; + // begin inline asm + mov.u32 %r106, %r5; + mov.u32 %r107, %r5; + mov.u32 %r108, %r5; + mov.u32 %r109, %r5; + @%p10 ld.global.v4.b32 { %r106, %r107, %r108, %r109 }, [ %rd28 + 0 ]; + // end inline asm + mov.b32 {%rs209, %rs210}, %r106; + mov.b32 {%rs211, %rs212}, %r107; + mov.b32 {%rs213, %rs214}, %r108; + mov.b32 {%rs215, %rs216}, %r109; + // begin inline asm + mov.u32 %r110, %r5; + mov.u32 %r111, %r5; + mov.u32 %r112, %r5; + mov.u32 %r113, %r5; + @%p11 ld.global.v4.b32 { %r110, %r111, %r112, %r113 }, [ %rd29 + 0 ]; + // end inline asm + mov.b32 {%rs217, %rs218}, %r110; + mov.b32 {%rs219, %rs220}, %r111; + mov.b32 {%rs221, %rs222}, %r112; + mov.b32 {%rs223, %rs224}, %r113; + // begin inline asm + mov.u32 %r114, %r5; + mov.u32 %r115, %r5; + mov.u32 %r116, %r5; + mov.u32 %r117, %r5; + @%p12 ld.global.v4.b32 { %r114, %r115, %r116, %r117 }, [ %rd30 + 0 ]; + // end inline asm + mov.b32 {%rs225, %rs226}, %r114; + mov.b32 {%rs227, %rs228}, %r115; + mov.b32 {%rs229, %rs230}, %r116; + mov.b32 {%rs231, %rs232}, %r117; + // begin inline asm + mov.u32 %r118, %r5; + mov.u32 %r119, %r5; + mov.u32 %r120, %r5; + mov.u32 %r121, %r5; + @%p13 ld.global.v4.b32 { %r118, %r119, %r120, %r121 }, [ %rd31 + 0 ]; + // end inline asm + mov.b32 {%rs233, %rs234}, %r118; + mov.b32 {%rs235, %rs236}, %r119; + mov.b32 {%rs237, %rs238}, %r120; + mov.b32 {%rs239, %rs240}, %r121; + // begin inline asm + mov.u32 %r122, %r5; + mov.u32 %r123, %r5; + mov.u32 %r124, %r5; + mov.u32 %r125, %r5; + @%p14 ld.global.v4.b32 { %r122, %r123, %r124, %r125 }, [ %rd32 + 0 ]; + // end inline asm + mov.b32 {%rs241, %rs242}, %r122; + mov.b32 {%rs243, %rs244}, %r123; + mov.b32 {%rs245, %rs246}, %r124; + mov.b32 {%rs247, %rs248}, %r125; + // begin inline asm + mov.u32 %r126, %r5; + mov.u32 %r127, %r5; + mov.u32 %r128, %r5; + mov.u32 %r129, %r5; + @%p15 ld.global.v4.b32 { %r126, %r127, %r128, %r129 }, [ %rd33 + 0 ]; + // end inline asm + mov.b32 {%rs249, %rs250}, %r126; + mov.b32 {%rs251, %rs252}, %r127; + mov.b32 {%rs253, %rs254}, %r128; + mov.b32 {%rs255, %rs256}, %r129; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r346, %rs129; + cvt.f32.bf16 %r347, %rs130; + cvt.f32.bf16 %r348, %rs131; + cvt.f32.bf16 %r349, %rs132; + cvt.f32.bf16 %r350, %rs133; + cvt.f32.bf16 %r351, %rs134; + cvt.f32.bf16 %r352, %rs135; + cvt.f32.bf16 %r353, %rs136; + cvt.f32.bf16 %r354, %rs137; + cvt.f32.bf16 %r355, %rs138; + cvt.f32.bf16 %r356, %rs139; + cvt.f32.bf16 %r357, %rs140; + cvt.f32.bf16 %r358, %rs141; + cvt.f32.bf16 %r359, %rs142; + cvt.f32.bf16 %r360, %rs143; + cvt.f32.bf16 %r361, %rs144; + cvt.f32.bf16 %r362, %rs145; + cvt.f32.bf16 %r363, %rs146; + cvt.f32.bf16 %r364, %rs147; + cvt.f32.bf16 %r365, %rs148; + cvt.f32.bf16 %r366, %rs149; + cvt.f32.bf16 %r367, %rs150; + cvt.f32.bf16 %r368, %rs151; + cvt.f32.bf16 %r369, %rs152; + cvt.f32.bf16 %r370, %rs153; + cvt.f32.bf16 %r371, %rs154; + cvt.f32.bf16 %r372, %rs155; + cvt.f32.bf16 %r373, %rs156; + cvt.f32.bf16 %r374, %rs157; + cvt.f32.bf16 %r375, %rs158; + cvt.f32.bf16 %r376, %rs159; + cvt.f32.bf16 %r377, %rs160; + cvt.f32.bf16 %r378, %rs161; + cvt.f32.bf16 %r379, %rs162; + cvt.f32.bf16 %r380, %rs163; + cvt.f32.bf16 %r381, %rs164; + cvt.f32.bf16 %r382, %rs165; + cvt.f32.bf16 %r383, %rs166; + cvt.f32.bf16 %r384, %rs167; + cvt.f32.bf16 %r385, %rs168; + cvt.f32.bf16 %r386, %rs169; + cvt.f32.bf16 %r387, %rs170; + cvt.f32.bf16 %r388, %rs171; + cvt.f32.bf16 %r389, %rs172; + cvt.f32.bf16 %r390, %rs173; + cvt.f32.bf16 %r391, %rs174; + cvt.f32.bf16 %r392, %rs175; + cvt.f32.bf16 %r393, %rs176; + cvt.f32.bf16 %r394, %rs177; + cvt.f32.bf16 %r395, %rs178; + cvt.f32.bf16 %r396, %rs179; + cvt.f32.bf16 %r397, %rs180; + cvt.f32.bf16 %r398, %rs181; + cvt.f32.bf16 %r399, %rs182; + cvt.f32.bf16 %r400, %rs183; + cvt.f32.bf16 %r401, %rs184; + cvt.f32.bf16 %r402, %rs185; + cvt.f32.bf16 %r403, %rs186; + cvt.f32.bf16 %r404, %rs187; + cvt.f32.bf16 %r405, %rs188; + cvt.f32.bf16 %r406, %rs189; + cvt.f32.bf16 %r407, %rs190; + cvt.f32.bf16 %r408, %rs191; + cvt.f32.bf16 %r409, %rs192; + cvt.f32.bf16 %r410, %rs193; + cvt.f32.bf16 %r411, %rs194; + cvt.f32.bf16 %r412, %rs195; + cvt.f32.bf16 %r413, %rs196; + cvt.f32.bf16 %r414, %rs197; + cvt.f32.bf16 %r415, %rs198; + cvt.f32.bf16 %r416, %rs199; + cvt.f32.bf16 %r417, %rs200; + cvt.f32.bf16 %r418, %rs201; + cvt.f32.bf16 %r419, %rs202; + cvt.f32.bf16 %r420, %rs203; + cvt.f32.bf16 %r421, %rs204; + cvt.f32.bf16 %r422, %rs205; + cvt.f32.bf16 %r423, %rs206; + cvt.f32.bf16 %r424, %rs207; + cvt.f32.bf16 %r425, %rs208; + cvt.f32.bf16 %r426, %rs209; + cvt.f32.bf16 %r427, %rs210; + cvt.f32.bf16 %r428, %rs211; + cvt.f32.bf16 %r429, %rs212; + cvt.f32.bf16 %r430, %rs213; + cvt.f32.bf16 %r431, %rs214; + cvt.f32.bf16 %r432, %rs215; + cvt.f32.bf16 %r433, %rs216; + cvt.f32.bf16 %r434, %rs217; + cvt.f32.bf16 %r435, %rs218; + cvt.f32.bf16 %r436, %rs219; + cvt.f32.bf16 %r437, %rs220; + cvt.f32.bf16 %r438, %rs221; + cvt.f32.bf16 %r439, %rs222; + cvt.f32.bf16 %r440, %rs223; + cvt.f32.bf16 %r441, %rs224; + cvt.f32.bf16 %r442, %rs225; + cvt.f32.bf16 %r443, %rs226; + cvt.f32.bf16 %r444, %rs227; + cvt.f32.bf16 %r445, %rs228; + cvt.f32.bf16 %r446, %rs229; + cvt.f32.bf16 %r447, %rs230; + cvt.f32.bf16 %r448, %rs231; + cvt.f32.bf16 %r449, %rs232; + cvt.f32.bf16 %r450, %rs233; + cvt.f32.bf16 %r451, %rs234; + cvt.f32.bf16 %r452, %rs235; + cvt.f32.bf16 %r453, %rs236; + cvt.f32.bf16 %r454, %rs237; + cvt.f32.bf16 %r455, %rs238; + cvt.f32.bf16 %r456, %rs239; + cvt.f32.bf16 %r457, %rs240; + cvt.f32.bf16 %r458, %rs241; + cvt.f32.bf16 %r459, %rs242; + cvt.f32.bf16 %r460, %rs243; + cvt.f32.bf16 %r461, %rs244; + cvt.f32.bf16 %r462, %rs245; + cvt.f32.bf16 %r463, %rs246; + cvt.f32.bf16 %r464, %rs247; + cvt.f32.bf16 %r465, %rs248; + cvt.f32.bf16 %r466, %rs249; + cvt.f32.bf16 %r467, %rs250; + cvt.f32.bf16 %r468, %rs251; + cvt.f32.bf16 %r469, %rs252; + cvt.f32.bf16 %r470, %rs253; + cvt.f32.bf16 %r471, %rs254; + cvt.f32.bf16 %r472, %rs255; + cvt.f32.bf16 %r473, %rs256; + .loc 1 107 18 // l2norm.py:107:18 + shl.b32 %r474, %r197, 5; + and.b32 %r475, %r474, 96; + and.b32 %r476, %r197, 28; + shl.b32 %r477, %r197, 2; + and.b32 %r478, %r477, 128; + mov.b32 %r479, global_smem; + add.s32 %r480, %r479, %r475; + add.s32 %r481, %r480, %r478; + add.s32 %r130, %r481, %r476; + // begin inline asm + st.shared.b32 [ %r130 + 0 ], %r131; + // end inline asm + bar.sync 0; + shl.b32 %r482, %r198, 1; + add.s32 %r483, %r479, %r482; + ld.shared.v4.b32 {%r484, %r485, %r486, %r487}, [%r483]; + ld.shared.v4.b32 {%r488, %r489, %r490, %r491}, [%r483+16]; + ld.shared.v4.b32 {%r492, %r493, %r494, %r495}, [%r483+128]; + ld.shared.v4.b32 {%r496, %r497, %r498, %r499}, [%r483+144]; + mul.f32 %r500, %r487, %r370; + mul.f32 %r501, %r487, %r371; + mul.f32 %r502, %r487, %r372; + mul.f32 %r503, %r487, %r373; + mul.f32 %r504, %r487, %r374; + mul.f32 %r505, %r487, %r375; + mul.f32 %r506, %r487, %r376; + mul.f32 %r507, %r487, %r377; + mul.f32 %r508, %r488, %r378; + mul.f32 %r509, %r488, %r379; + mul.f32 %r510, %r488, %r380; + mul.f32 %r511, %r488, %r381; + mul.f32 %r512, %r488, %r382; + mul.f32 %r513, %r488, %r383; + mul.f32 %r514, %r488, %r384; + mul.f32 %r515, %r488, %r385; + mul.f32 %r516, %r489, %r386; + mul.f32 %r517, %r489, %r387; + mul.f32 %r518, %r489, %r388; + mul.f32 %r519, %r489, %r389; + mul.f32 %r520, %r489, %r390; + mul.f32 %r521, %r489, %r391; + mul.f32 %r522, %r489, %r392; + mul.f32 %r523, %r489, %r393; + mul.f32 %r524, %r490, %r394; + mul.f32 %r525, %r490, %r395; + mul.f32 %r526, %r490, %r396; + mul.f32 %r527, %r490, %r397; + mul.f32 %r528, %r490, %r398; + mul.f32 %r529, %r490, %r399; + mul.f32 %r530, %r490, %r400; + mul.f32 %r531, %r490, %r401; + mul.f32 %r532, %r491, %r402; + mul.f32 %r533, %r491, %r403; + mul.f32 %r534, %r491, %r404; + mul.f32 %r535, %r491, %r405; + mul.f32 %r536, %r491, %r406; + mul.f32 %r537, %r491, %r407; + mul.f32 %r538, %r491, %r408; + mul.f32 %r539, %r491, %r409; + mul.f32 %r540, %r492, %r410; + mul.f32 %r541, %r492, %r411; + mul.f32 %r542, %r492, %r412; + mul.f32 %r543, %r492, %r413; + mul.f32 %r544, %r492, %r414; + mul.f32 %r545, %r492, %r415; + mul.f32 %r546, %r492, %r416; + mul.f32 %r547, %r492, %r417; + mul.f32 %r548, %r493, %r418; + mul.f32 %r549, %r493, %r419; + mul.f32 %r550, %r493, %r420; + mul.f32 %r551, %r493, %r421; + mul.f32 %r552, %r493, %r422; + mul.f32 %r553, %r493, %r423; + mul.f32 %r554, %r493, %r424; + mul.f32 %r555, %r493, %r425; + mul.f32 %r556, %r494, %r426; + mul.f32 %r557, %r494, %r427; + mul.f32 %r558, %r494, %r428; + mul.f32 %r559, %r494, %r429; + mul.f32 %r560, %r494, %r430; + mul.f32 %r561, %r494, %r431; + mul.f32 %r562, %r494, %r432; + mul.f32 %r563, %r494, %r433; + mul.f32 %r564, %r495, %r434; + mul.f32 %r565, %r495, %r435; + mul.f32 %r566, %r495, %r436; + mul.f32 %r567, %r495, %r437; + mul.f32 %r568, %r495, %r438; + mul.f32 %r569, %r495, %r439; + mul.f32 %r570, %r495, %r440; + mul.f32 %r571, %r495, %r441; + mul.f32 %r572, %r496, %r442; + mul.f32 %r573, %r496, %r443; + mul.f32 %r574, %r496, %r444; + mul.f32 %r575, %r496, %r445; + mul.f32 %r576, %r496, %r446; + mul.f32 %r577, %r496, %r447; + mul.f32 %r578, %r496, %r448; + mul.f32 %r579, %r496, %r449; + mul.f32 %r580, %r497, %r450; + mul.f32 %r581, %r497, %r451; + mul.f32 %r582, %r497, %r452; + mul.f32 %r583, %r497, %r453; + mul.f32 %r584, %r497, %r454; + mul.f32 %r585, %r497, %r455; + mul.f32 %r586, %r497, %r456; + mul.f32 %r587, %r497, %r457; + mul.f32 %r588, %r498, %r458; + mul.f32 %r589, %r498, %r459; + mul.f32 %r590, %r498, %r460; + mul.f32 %r591, %r498, %r461; + mul.f32 %r592, %r498, %r462; + mul.f32 %r593, %r498, %r463; + mul.f32 %r594, %r498, %r464; + mul.f32 %r595, %r498, %r465; + mul.f32 %r596, %r499, %r466; + mul.f32 %r597, %r499, %r467; + mul.f32 %r598, %r499, %r468; + mul.f32 %r599, %r499, %r469; + mul.f32 %r600, %r499, %r470; + mul.f32 %r601, %r499, %r471; + mul.f32 %r602, %r499, %r472; + mul.f32 %r603, %r499, %r473; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r604, %r219, %r347; + mul.f32 %r605, %r227, %r355; + mul.f32 %r606, %r235, %r363; + mul.f32 %r607, %r243, %r371; + mul.f32 %r608, %r251, %r379; + mul.f32 %r609, %r259, %r387; + mul.f32 %r610, %r267, %r395; + mul.f32 %r611, %r275, %r403; + mul.f32 %r612, %r283, %r411; + mul.f32 %r613, %r291, %r419; + mul.f32 %r614, %r299, %r427; + mul.f32 %r615, %r307, %r435; + mul.f32 %r616, %r315, %r443; + mul.f32 %r617, %r323, %r451; + mul.f32 %r618, %r331, %r459; + mul.f32 %r619, %r339, %r467; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r620, %rs1, %rs129, %r604; + fma.rn.f32.bf16 %r621, %rs3, %rs131, %r620; + fma.rn.f32.bf16 %r622, %rs4, %rs132, %r621; + fma.rn.f32.bf16 %r623, %rs5, %rs133, %r622; + fma.rn.f32.bf16 %r624, %rs6, %rs134, %r623; + fma.rn.f32.bf16 %r625, %rs7, %rs135, %r624; + fma.rn.f32.bf16 %r626, %rs8, %rs136, %r625; + fma.rn.f32.bf16 %r627, %rs9, %rs137, %r605; + fma.rn.f32.bf16 %r628, %rs11, %rs139, %r627; + fma.rn.f32.bf16 %r629, %rs12, %rs140, %r628; + fma.rn.f32.bf16 %r630, %rs13, %rs141, %r629; + fma.rn.f32.bf16 %r631, %rs14, %rs142, %r630; + fma.rn.f32.bf16 %r632, %rs15, %rs143, %r631; + fma.rn.f32.bf16 %r633, %rs16, %rs144, %r632; + fma.rn.f32.bf16 %r634, %rs17, %rs145, %r606; + fma.rn.f32.bf16 %r635, %rs19, %rs147, %r634; + fma.rn.f32.bf16 %r636, %rs20, %rs148, %r635; + fma.rn.f32.bf16 %r637, %rs21, %rs149, %r636; + fma.rn.f32.bf16 %r638, %rs22, %rs150, %r637; + fma.rn.f32.bf16 %r639, %rs23, %rs151, %r638; + fma.rn.f32.bf16 %r640, %rs24, %rs152, %r639; + fma.rn.f32.bf16 %r641, %rs25, %rs153, %r607; + fma.rn.f32.bf16 %r642, %rs27, %rs155, %r641; + fma.rn.f32.bf16 %r643, %rs28, %rs156, %r642; + fma.rn.f32.bf16 %r644, %rs29, %rs157, %r643; + fma.rn.f32.bf16 %r645, %rs30, %rs158, %r644; + fma.rn.f32.bf16 %r646, %rs31, %rs159, %r645; + fma.rn.f32.bf16 %r647, %rs32, %rs160, %r646; + fma.rn.f32.bf16 %r648, %rs33, %rs161, %r608; + fma.rn.f32.bf16 %r649, %rs35, %rs163, %r648; + fma.rn.f32.bf16 %r650, %rs36, %rs164, %r649; + fma.rn.f32.bf16 %r651, %rs37, %rs165, %r650; + fma.rn.f32.bf16 %r652, %rs38, %rs166, %r651; + fma.rn.f32.bf16 %r653, %rs39, %rs167, %r652; + fma.rn.f32.bf16 %r654, %rs40, %rs168, %r653; + fma.rn.f32.bf16 %r655, %rs41, %rs169, %r609; + fma.rn.f32.bf16 %r656, %rs43, %rs171, %r655; + fma.rn.f32.bf16 %r657, %rs44, %rs172, %r656; + fma.rn.f32.bf16 %r658, %rs45, %rs173, %r657; + fma.rn.f32.bf16 %r659, %rs46, %rs174, %r658; + fma.rn.f32.bf16 %r660, %rs47, %rs175, %r659; + fma.rn.f32.bf16 %r661, %rs48, %rs176, %r660; + fma.rn.f32.bf16 %r662, %rs49, %rs177, %r610; + fma.rn.f32.bf16 %r663, %rs51, %rs179, %r662; + fma.rn.f32.bf16 %r664, %rs52, %rs180, %r663; + fma.rn.f32.bf16 %r665, %rs53, %rs181, %r664; + fma.rn.f32.bf16 %r666, %rs54, %rs182, %r665; + fma.rn.f32.bf16 %r667, %rs55, %rs183, %r666; + fma.rn.f32.bf16 %r668, %rs56, %rs184, %r667; + fma.rn.f32.bf16 %r669, %rs57, %rs185, %r611; + fma.rn.f32.bf16 %r670, %rs59, %rs187, %r669; + fma.rn.f32.bf16 %r671, %rs60, %rs188, %r670; + fma.rn.f32.bf16 %r672, %rs61, %rs189, %r671; + fma.rn.f32.bf16 %r673, %rs62, %rs190, %r672; + fma.rn.f32.bf16 %r674, %rs63, %rs191, %r673; + fma.rn.f32.bf16 %r675, %rs64, %rs192, %r674; + fma.rn.f32.bf16 %r676, %rs65, %rs193, %r612; + fma.rn.f32.bf16 %r677, %rs67, %rs195, %r676; + fma.rn.f32.bf16 %r678, %rs68, %rs196, %r677; + fma.rn.f32.bf16 %r679, %rs69, %rs197, %r678; + fma.rn.f32.bf16 %r680, %rs70, %rs198, %r679; + fma.rn.f32.bf16 %r681, %rs71, %rs199, %r680; + fma.rn.f32.bf16 %r682, %rs72, %rs200, %r681; + fma.rn.f32.bf16 %r683, %rs73, %rs201, %r613; + fma.rn.f32.bf16 %r684, %rs75, %rs203, %r683; + fma.rn.f32.bf16 %r685, %rs76, %rs204, %r684; + fma.rn.f32.bf16 %r686, %rs77, %rs205, %r685; + fma.rn.f32.bf16 %r687, %rs78, %rs206, %r686; + fma.rn.f32.bf16 %r688, %rs79, %rs207, %r687; + fma.rn.f32.bf16 %r689, %rs80, %rs208, %r688; + fma.rn.f32.bf16 %r690, %rs81, %rs209, %r614; + fma.rn.f32.bf16 %r691, %rs83, %rs211, %r690; + fma.rn.f32.bf16 %r692, %rs84, %rs212, %r691; + fma.rn.f32.bf16 %r693, %rs85, %rs213, %r692; + fma.rn.f32.bf16 %r694, %rs86, %rs214, %r693; + fma.rn.f32.bf16 %r695, %rs87, %rs215, %r694; + fma.rn.f32.bf16 %r696, %rs88, %rs216, %r695; + fma.rn.f32.bf16 %r697, %rs89, %rs217, %r615; + fma.rn.f32.bf16 %r698, %rs91, %rs219, %r697; + fma.rn.f32.bf16 %r699, %rs92, %rs220, %r698; + fma.rn.f32.bf16 %r700, %rs93, %rs221, %r699; + fma.rn.f32.bf16 %r701, %rs94, %rs222, %r700; + fma.rn.f32.bf16 %r702, %rs95, %rs223, %r701; + fma.rn.f32.bf16 %r703, %rs96, %rs224, %r702; + fma.rn.f32.bf16 %r704, %rs97, %rs225, %r616; + fma.rn.f32.bf16 %r705, %rs99, %rs227, %r704; + fma.rn.f32.bf16 %r706, %rs100, %rs228, %r705; + fma.rn.f32.bf16 %r707, %rs101, %rs229, %r706; + fma.rn.f32.bf16 %r708, %rs102, %rs230, %r707; + fma.rn.f32.bf16 %r709, %rs103, %rs231, %r708; + fma.rn.f32.bf16 %r710, %rs104, %rs232, %r709; + fma.rn.f32.bf16 %r711, %rs105, %rs233, %r617; + fma.rn.f32.bf16 %r712, %rs107, %rs235, %r711; + fma.rn.f32.bf16 %r713, %rs108, %rs236, %r712; + fma.rn.f32.bf16 %r714, %rs109, %rs237, %r713; + fma.rn.f32.bf16 %r715, %rs110, %rs238, %r714; + fma.rn.f32.bf16 %r716, %rs111, %rs239, %r715; + fma.rn.f32.bf16 %r717, %rs112, %rs240, %r716; + fma.rn.f32.bf16 %r718, %rs113, %rs241, %r618; + fma.rn.f32.bf16 %r719, %rs115, %rs243, %r718; + fma.rn.f32.bf16 %r720, %rs116, %rs244, %r719; + fma.rn.f32.bf16 %r721, %rs117, %rs245, %r720; + fma.rn.f32.bf16 %r722, %rs118, %rs246, %r721; + fma.rn.f32.bf16 %r723, %rs119, %rs247, %r722; + fma.rn.f32.bf16 %r724, %rs120, %rs248, %r723; + fma.rn.f32.bf16 %r725, %rs121, %rs249, %r619; + fma.rn.f32.bf16 %r726, %rs123, %rs251, %r725; + fma.rn.f32.bf16 %r727, %rs124, %rs252, %r726; + fma.rn.f32.bf16 %r728, %rs125, %rs253, %r727; + fma.rn.f32.bf16 %r729, %rs126, %rs254, %r728; + fma.rn.f32.bf16 %r730, %rs127, %rs255, %r729; + fma.rn.f32.bf16 %r731, %rs128, %rs256, %r730; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r732, %r626, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r733, %r626, %r732; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r734, %r733, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r735, %r733, %r734; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r736, %r735, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r737, %r735, %r736; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r738, %r737, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r739, %r737, %r738; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r740, %r633, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r741, %r633, %r740; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r742, %r741, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r743, %r741, %r742; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r744, %r743, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r745, %r743, %r744; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r746, %r745, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r747, %r745, %r746; +$L__tmp18: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r748, %r640, 8, 31, -1; +$L__tmp19: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r749, %r640, %r748; +$L__tmp20: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r750, %r749, 4, 31, -1; +$L__tmp21: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r751, %r749, %r750; +$L__tmp22: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r752, %r751, 2, 31, -1; +$L__tmp23: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r753, %r751, %r752; +$L__tmp24: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r754, %r753, 1, 31, -1; +$L__tmp25: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r755, %r753, %r754; +$L__tmp26: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r756, %r647, 8, 31, -1; +$L__tmp27: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r757, %r647, %r756; +$L__tmp28: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r758, %r757, 4, 31, -1; +$L__tmp29: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r759, %r757, %r758; +$L__tmp30: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r760, %r759, 2, 31, -1; +$L__tmp31: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r761, %r759, %r760; +$L__tmp32: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r762, %r761, 1, 31, -1; +$L__tmp33: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r763, %r761, %r762; +$L__tmp34: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r764, %r654, 8, 31, -1; +$L__tmp35: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r765, %r654, %r764; +$L__tmp36: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r766, %r765, 4, 31, -1; +$L__tmp37: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r767, %r765, %r766; +$L__tmp38: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r768, %r767, 2, 31, -1; +$L__tmp39: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r769, %r767, %r768; +$L__tmp40: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r770, %r769, 1, 31, -1; +$L__tmp41: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r771, %r769, %r770; +$L__tmp42: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r772, %r661, 8, 31, -1; +$L__tmp43: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r773, %r661, %r772; +$L__tmp44: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r774, %r773, 4, 31, -1; +$L__tmp45: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r775, %r773, %r774; +$L__tmp46: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r776, %r775, 2, 31, -1; +$L__tmp47: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r777, %r775, %r776; +$L__tmp48: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r778, %r777, 1, 31, -1; +$L__tmp49: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r779, %r777, %r778; +$L__tmp50: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r780, %r668, 8, 31, -1; +$L__tmp51: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r781, %r668, %r780; +$L__tmp52: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r782, %r781, 4, 31, -1; +$L__tmp53: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r783, %r781, %r782; +$L__tmp54: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r784, %r783, 2, 31, -1; +$L__tmp55: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r785, %r783, %r784; +$L__tmp56: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r786, %r785, 1, 31, -1; +$L__tmp57: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r787, %r785, %r786; +$L__tmp58: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r788, %r675, 8, 31, -1; +$L__tmp59: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r789, %r675, %r788; +$L__tmp60: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r790, %r789, 4, 31, -1; +$L__tmp61: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r791, %r789, %r790; +$L__tmp62: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r792, %r791, 2, 31, -1; +$L__tmp63: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r793, %r791, %r792; +$L__tmp64: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r794, %r793, 1, 31, -1; +$L__tmp65: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r795, %r793, %r794; +$L__tmp66: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r796, %r682, 8, 31, -1; +$L__tmp67: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r797, %r682, %r796; +$L__tmp68: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r798, %r797, 4, 31, -1; +$L__tmp69: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r799, %r797, %r798; +$L__tmp70: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r800, %r799, 2, 31, -1; +$L__tmp71: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r801, %r799, %r800; +$L__tmp72: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r802, %r801, 1, 31, -1; +$L__tmp73: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r803, %r801, %r802; +$L__tmp74: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r804, %r689, 8, 31, -1; +$L__tmp75: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r805, %r689, %r804; +$L__tmp76: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r806, %r805, 4, 31, -1; +$L__tmp77: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r807, %r805, %r806; +$L__tmp78: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r808, %r807, 2, 31, -1; +$L__tmp79: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r809, %r807, %r808; +$L__tmp80: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r810, %r809, 1, 31, -1; +$L__tmp81: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r811, %r809, %r810; +$L__tmp82: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r812, %r696, 8, 31, -1; +$L__tmp83: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r813, %r696, %r812; +$L__tmp84: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r814, %r813, 4, 31, -1; +$L__tmp85: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r815, %r813, %r814; +$L__tmp86: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r816, %r815, 2, 31, -1; +$L__tmp87: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r817, %r815, %r816; +$L__tmp88: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r818, %r817, 1, 31, -1; +$L__tmp89: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r819, %r817, %r818; +$L__tmp90: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r820, %r703, 8, 31, -1; +$L__tmp91: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r821, %r703, %r820; +$L__tmp92: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r822, %r821, 4, 31, -1; +$L__tmp93: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r823, %r821, %r822; +$L__tmp94: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r824, %r823, 2, 31, -1; +$L__tmp95: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r825, %r823, %r824; +$L__tmp96: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r826, %r825, 1, 31, -1; +$L__tmp97: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r827, %r825, %r826; +$L__tmp98: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r828, %r710, 8, 31, -1; +$L__tmp99: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r829, %r710, %r828; +$L__tmp100: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r830, %r829, 4, 31, -1; +$L__tmp101: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r831, %r829, %r830; +$L__tmp102: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r832, %r831, 2, 31, -1; +$L__tmp103: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r833, %r831, %r832; +$L__tmp104: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r834, %r833, 1, 31, -1; +$L__tmp105: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r835, %r833, %r834; +$L__tmp106: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r836, %r717, 8, 31, -1; +$L__tmp107: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r837, %r717, %r836; +$L__tmp108: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r838, %r837, 4, 31, -1; +$L__tmp109: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r839, %r837, %r838; +$L__tmp110: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r840, %r839, 2, 31, -1; +$L__tmp111: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r841, %r839, %r840; +$L__tmp112: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r842, %r841, 1, 31, -1; +$L__tmp113: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r843, %r841, %r842; +$L__tmp114: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r844, %r724, 8, 31, -1; +$L__tmp115: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r845, %r724, %r844; +$L__tmp116: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r846, %r845, 4, 31, -1; +$L__tmp117: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r847, %r845, %r846; +$L__tmp118: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r848, %r847, 2, 31, -1; +$L__tmp119: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r849, %r847, %r848; +$L__tmp120: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r850, %r849, 1, 31, -1; +$L__tmp121: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r851, %r849, %r850; +$L__tmp122: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r852, %r731, 8, 31, -1; +$L__tmp123: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r853, %r731, %r852; +$L__tmp124: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r854, %r853, 4, 31, -1; +$L__tmp125: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r855, %r853, %r854; +$L__tmp126: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r856, %r855, 2, 31, -1; +$L__tmp127: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r857, %r855, %r856; +$L__tmp128: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r858, %r857, 1, 31, -1; +$L__tmp129: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r859, %r857, %r858; +$L__tmp130: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r860, %r739, %r218; + mul.f32 %r861, %r739, %r219; + mul.f32 %r862, %r739, %r220; + mul.f32 %r863, %r739, %r221; + mul.f32 %r864, %r739, %r222; + mul.f32 %r865, %r739, %r223; + mul.f32 %r866, %r739, %r224; + mul.f32 %r867, %r739, %r225; + mul.f32 %r868, %r747, %r226; + mul.f32 %r869, %r747, %r227; + mul.f32 %r870, %r747, %r228; + mul.f32 %r871, %r747, %r229; + mul.f32 %r872, %r747, %r230; + mul.f32 %r873, %r747, %r231; + mul.f32 %r874, %r747, %r232; + mul.f32 %r875, %r747, %r233; + mul.f32 %r876, %r755, %r234; + mul.f32 %r877, %r755, %r235; + mul.f32 %r878, %r755, %r236; + mul.f32 %r879, %r755, %r237; + mul.f32 %r880, %r755, %r238; + mul.f32 %r881, %r755, %r239; + mul.f32 %r882, %r755, %r240; + mul.f32 %r883, %r755, %r241; + mul.f32 %r884, %r763, %r242; + mul.f32 %r885, %r763, %r243; + mul.f32 %r886, %r763, %r244; + mul.f32 %r887, %r763, %r245; + mul.f32 %r888, %r763, %r246; + mul.f32 %r889, %r763, %r247; + mul.f32 %r890, %r763, %r248; + mul.f32 %r891, %r763, %r249; + mul.f32 %r892, %r771, %r250; + mul.f32 %r893, %r771, %r251; + mul.f32 %r894, %r771, %r252; + mul.f32 %r895, %r771, %r253; + mul.f32 %r896, %r771, %r254; + mul.f32 %r897, %r771, %r255; + mul.f32 %r898, %r771, %r256; + mul.f32 %r899, %r771, %r257; + mul.f32 %r900, %r779, %r258; + mul.f32 %r901, %r779, %r259; + mul.f32 %r902, %r779, %r260; + mul.f32 %r903, %r779, %r261; + mul.f32 %r904, %r779, %r262; + mul.f32 %r905, %r779, %r263; + mul.f32 %r906, %r779, %r264; + mul.f32 %r907, %r779, %r265; + mul.f32 %r908, %r787, %r266; + mul.f32 %r909, %r787, %r267; + mul.f32 %r910, %r787, %r268; + mul.f32 %r911, %r787, %r269; + mul.f32 %r912, %r787, %r270; + mul.f32 %r913, %r787, %r271; + mul.f32 %r914, %r787, %r272; + mul.f32 %r915, %r787, %r273; + mul.f32 %r916, %r795, %r274; + mul.f32 %r917, %r795, %r275; + mul.f32 %r918, %r795, %r276; + mul.f32 %r919, %r795, %r277; + mul.f32 %r920, %r795, %r278; + mul.f32 %r921, %r795, %r279; + mul.f32 %r922, %r795, %r280; + mul.f32 %r923, %r795, %r281; + mul.f32 %r924, %r803, %r282; + mul.f32 %r925, %r803, %r283; + mul.f32 %r926, %r803, %r284; + mul.f32 %r927, %r803, %r285; + mul.f32 %r928, %r803, %r286; + mul.f32 %r929, %r803, %r287; + mul.f32 %r930, %r803, %r288; + mul.f32 %r931, %r803, %r289; + mul.f32 %r932, %r811, %r290; + mul.f32 %r933, %r811, %r291; + mul.f32 %r934, %r811, %r292; + mul.f32 %r935, %r811, %r293; + mul.f32 %r936, %r811, %r294; + mul.f32 %r937, %r811, %r295; + mul.f32 %r938, %r811, %r296; + mul.f32 %r939, %r811, %r297; + mul.f32 %r940, %r819, %r298; + mul.f32 %r941, %r819, %r299; + mul.f32 %r942, %r819, %r300; + mul.f32 %r943, %r819, %r301; + mul.f32 %r944, %r819, %r302; + mul.f32 %r945, %r819, %r303; + mul.f32 %r946, %r819, %r304; + mul.f32 %r947, %r819, %r305; + mul.f32 %r948, %r827, %r306; + mul.f32 %r949, %r827, %r307; + mul.f32 %r950, %r827, %r308; + mul.f32 %r951, %r827, %r309; + mul.f32 %r952, %r827, %r310; + mul.f32 %r953, %r827, %r311; + mul.f32 %r954, %r827, %r312; + mul.f32 %r955, %r827, %r313; + mul.f32 %r956, %r835, %r314; + mul.f32 %r957, %r835, %r315; + mul.f32 %r958, %r835, %r316; + mul.f32 %r959, %r835, %r317; + mul.f32 %r960, %r835, %r318; + mul.f32 %r961, %r835, %r319; + mul.f32 %r962, %r835, %r320; + mul.f32 %r963, %r835, %r321; + mul.f32 %r964, %r843, %r322; + mul.f32 %r965, %r843, %r323; + mul.f32 %r966, %r843, %r324; + mul.f32 %r967, %r843, %r325; + mul.f32 %r968, %r843, %r326; + mul.f32 %r969, %r843, %r327; + mul.f32 %r970, %r843, %r328; + mul.f32 %r971, %r843, %r329; + mul.f32 %r972, %r851, %r330; + mul.f32 %r973, %r851, %r331; + mul.f32 %r974, %r851, %r332; + mul.f32 %r975, %r851, %r333; + mul.f32 %r976, %r851, %r334; + mul.f32 %r977, %r851, %r335; + mul.f32 %r978, %r851, %r336; + mul.f32 %r979, %r851, %r337; + mul.f32 %r980, %r859, %r338; + mul.f32 %r981, %r859, %r339; + mul.f32 %r982, %r859, %r340; + mul.f32 %r983, %r859, %r341; + mul.f32 %r984, %r859, %r342; + mul.f32 %r985, %r859, %r343; + mul.f32 %r986, %r859, %r344; + mul.f32 %r987, %r859, %r345; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r988, %r484, %r860; + mul.f32 %r989, %r484, %r861; + mul.f32 %r990, %r484, %r862; + mul.f32 %r991, %r484, %r863; + mul.f32 %r992, %r484, %r864; + mul.f32 %r993, %r484, %r865; + mul.f32 %r994, %r484, %r866; + mul.f32 %r995, %r484, %r867; + mul.f32 %r996, %r485, %r868; + mul.f32 %r997, %r485, %r869; + mul.f32 %r998, %r485, %r870; + mul.f32 %r999, %r485, %r871; + mul.f32 %r1000, %r485, %r872; + mul.f32 %r1001, %r485, %r873; + mul.f32 %r1002, %r485, %r874; + mul.f32 %r1003, %r485, %r875; + mul.f32 %r1004, %r486, %r876; + mul.f32 %r1005, %r486, %r877; + mul.f32 %r1006, %r486, %r878; + mul.f32 %r1007, %r486, %r879; + mul.f32 %r1008, %r486, %r880; + mul.f32 %r1009, %r486, %r881; + mul.f32 %r1010, %r486, %r882; + mul.f32 %r1011, %r486, %r883; + mul.f32 %r1012, %r487, %r884; + mul.f32 %r1013, %r487, %r885; + mul.f32 %r1014, %r487, %r886; + mul.f32 %r1015, %r487, %r887; + mul.f32 %r1016, %r487, %r888; + mul.f32 %r1017, %r487, %r889; + mul.f32 %r1018, %r487, %r890; + mul.f32 %r1019, %r487, %r891; + mul.f32 %r1020, %r488, %r892; + mul.f32 %r1021, %r488, %r893; + mul.f32 %r1022, %r488, %r894; + mul.f32 %r1023, %r488, %r895; + mul.f32 %r1024, %r488, %r896; + mul.f32 %r1025, %r488, %r897; + mul.f32 %r1026, %r488, %r898; + mul.f32 %r1027, %r488, %r899; + mul.f32 %r1028, %r489, %r900; + mul.f32 %r1029, %r489, %r901; + mul.f32 %r1030, %r489, %r902; + mul.f32 %r1031, %r489, %r903; + mul.f32 %r1032, %r489, %r904; + mul.f32 %r1033, %r489, %r905; + mul.f32 %r1034, %r489, %r906; + mul.f32 %r1035, %r489, %r907; + mul.f32 %r1036, %r490, %r908; + mul.f32 %r1037, %r490, %r909; + mul.f32 %r1038, %r490, %r910; + mul.f32 %r1039, %r490, %r911; + mul.f32 %r1040, %r490, %r912; + mul.f32 %r1041, %r490, %r913; + mul.f32 %r1042, %r490, %r914; + mul.f32 %r1043, %r490, %r915; + mul.f32 %r1044, %r491, %r916; + mul.f32 %r1045, %r491, %r917; + mul.f32 %r1046, %r491, %r918; + mul.f32 %r1047, %r491, %r919; + mul.f32 %r1048, %r491, %r920; + mul.f32 %r1049, %r491, %r921; + mul.f32 %r1050, %r491, %r922; + mul.f32 %r1051, %r491, %r923; + mul.f32 %r1052, %r492, %r924; + mul.f32 %r1053, %r492, %r925; + mul.f32 %r1054, %r492, %r926; + mul.f32 %r1055, %r492, %r927; + mul.f32 %r1056, %r492, %r928; + mul.f32 %r1057, %r492, %r929; + mul.f32 %r1058, %r492, %r930; + mul.f32 %r1059, %r492, %r931; + mul.f32 %r1060, %r493, %r932; + mul.f32 %r1061, %r493, %r933; + mul.f32 %r1062, %r493, %r934; + mul.f32 %r1063, %r493, %r935; + mul.f32 %r1064, %r493, %r936; + mul.f32 %r1065, %r493, %r937; + mul.f32 %r1066, %r493, %r938; + mul.f32 %r1067, %r493, %r939; + mul.f32 %r1068, %r494, %r940; + mul.f32 %r1069, %r494, %r941; + mul.f32 %r1070, %r494, %r942; + mul.f32 %r1071, %r494, %r943; + mul.f32 %r1072, %r494, %r944; + mul.f32 %r1073, %r494, %r945; + mul.f32 %r1074, %r494, %r946; + mul.f32 %r1075, %r494, %r947; + mul.f32 %r1076, %r495, %r948; + mul.f32 %r1077, %r495, %r949; + mul.f32 %r1078, %r495, %r950; + mul.f32 %r1079, %r495, %r951; + mul.f32 %r1080, %r495, %r952; + mul.f32 %r1081, %r495, %r953; + mul.f32 %r1082, %r495, %r954; + mul.f32 %r1083, %r495, %r955; + mul.f32 %r1084, %r496, %r956; + mul.f32 %r1085, %r496, %r957; + mul.f32 %r1086, %r496, %r958; + mul.f32 %r1087, %r496, %r959; + mul.f32 %r1088, %r496, %r960; + mul.f32 %r1089, %r496, %r961; + mul.f32 %r1090, %r496, %r962; + mul.f32 %r1091, %r496, %r963; + mul.f32 %r1092, %r497, %r964; + mul.f32 %r1093, %r497, %r965; + mul.f32 %r1094, %r497, %r966; + mul.f32 %r1095, %r497, %r967; + mul.f32 %r1096, %r497, %r968; + mul.f32 %r1097, %r497, %r969; + mul.f32 %r1098, %r497, %r970; + mul.f32 %r1099, %r497, %r971; + mul.f32 %r1100, %r498, %r972; + mul.f32 %r1101, %r498, %r973; + mul.f32 %r1102, %r498, %r974; + mul.f32 %r1103, %r498, %r975; + mul.f32 %r1104, %r498, %r976; + mul.f32 %r1105, %r498, %r977; + mul.f32 %r1106, %r498, %r978; + mul.f32 %r1107, %r498, %r979; + mul.f32 %r1108, %r499, %r980; + mul.f32 %r1109, %r499, %r981; + mul.f32 %r1110, %r499, %r982; + mul.f32 %r1111, %r499, %r983; + mul.f32 %r1112, %r499, %r984; + mul.f32 %r1113, %r499, %r985; + mul.f32 %r1114, %r499, %r986; + mul.f32 %r1115, %r499, %r987; + .loc 1 107 36 // l2norm.py:107:36 + neg.f32 %r1116, %r988; + fma.rn.f32 %r1117, %r484, %r346, %r1116; + neg.f32 %r1118, %r989; + fma.rn.f32 %r1119, %r484, %r347, %r1118; + neg.f32 %r1120, %r990; + fma.rn.f32 %r1121, %r484, %r348, %r1120; + neg.f32 %r1122, %r991; + fma.rn.f32 %r1123, %r484, %r349, %r1122; + neg.f32 %r1124, %r992; + fma.rn.f32 %r1125, %r484, %r350, %r1124; + neg.f32 %r1126, %r993; + fma.rn.f32 %r1127, %r484, %r351, %r1126; + neg.f32 %r1128, %r994; + fma.rn.f32 %r1129, %r484, %r352, %r1128; + neg.f32 %r1130, %r995; + fma.rn.f32 %r1131, %r484, %r353, %r1130; + neg.f32 %r1132, %r996; + fma.rn.f32 %r1133, %r485, %r354, %r1132; + neg.f32 %r1134, %r997; + fma.rn.f32 %r1135, %r485, %r355, %r1134; + neg.f32 %r1136, %r998; + fma.rn.f32 %r1137, %r485, %r356, %r1136; + neg.f32 %r1138, %r999; + fma.rn.f32 %r1139, %r485, %r357, %r1138; + neg.f32 %r1140, %r1000; + fma.rn.f32 %r1141, %r485, %r358, %r1140; + neg.f32 %r1142, %r1001; + fma.rn.f32 %r1143, %r485, %r359, %r1142; + neg.f32 %r1144, %r1002; + fma.rn.f32 %r1145, %r485, %r360, %r1144; + neg.f32 %r1146, %r1003; + fma.rn.f32 %r1147, %r485, %r361, %r1146; + neg.f32 %r1148, %r1004; + fma.rn.f32 %r1149, %r486, %r362, %r1148; + neg.f32 %r1150, %r1005; + fma.rn.f32 %r1151, %r486, %r363, %r1150; + neg.f32 %r1152, %r1006; + fma.rn.f32 %r1153, %r486, %r364, %r1152; + neg.f32 %r1154, %r1007; + fma.rn.f32 %r1155, %r486, %r365, %r1154; + neg.f32 %r1156, %r1008; + fma.rn.f32 %r1157, %r486, %r366, %r1156; + neg.f32 %r1158, %r1009; + fma.rn.f32 %r1159, %r486, %r367, %r1158; + neg.f32 %r1160, %r1010; + fma.rn.f32 %r1161, %r486, %r368, %r1160; + neg.f32 %r1162, %r1011; + fma.rn.f32 %r1163, %r486, %r369, %r1162; + mov.b64 %rd156, {%r500, %r501}; + mov.b64 %rd157, {%r1012, %r1013}; + sub.f32x2 %rd158, %rd156, %rd157; + mov.b64 %rd159, {%r502, %r503}; + mov.b64 %rd160, {%r1014, %r1015}; + sub.f32x2 %rd161, %rd159, %rd160; + mov.b64 %rd162, {%r504, %r505}; + mov.b64 %rd163, {%r1016, %r1017}; + sub.f32x2 %rd164, %rd162, %rd163; + mov.b64 %rd165, {%r506, %r507}; + mov.b64 %rd166, {%r1018, %r1019}; + sub.f32x2 %rd167, %rd165, %rd166; + mov.b64 %rd168, {%r508, %r509}; + mov.b64 %rd169, {%r1020, %r1021}; + sub.f32x2 %rd170, %rd168, %rd169; + mov.b64 %rd171, {%r510, %r511}; + mov.b64 %rd172, {%r1022, %r1023}; + sub.f32x2 %rd173, %rd171, %rd172; + mov.b64 %rd174, {%r512, %r513}; + mov.b64 %rd175, {%r1024, %r1025}; + sub.f32x2 %rd176, %rd174, %rd175; + mov.b64 %rd177, {%r514, %r515}; + mov.b64 %rd178, {%r1026, %r1027}; + sub.f32x2 %rd179, %rd177, %rd178; + mov.b64 %rd180, {%r516, %r517}; + mov.b64 %rd181, {%r1028, %r1029}; + sub.f32x2 %rd182, %rd180, %rd181; + mov.b64 %rd183, {%r518, %r519}; + mov.b64 %rd184, {%r1030, %r1031}; + sub.f32x2 %rd185, %rd183, %rd184; + mov.b64 %rd186, {%r520, %r521}; + mov.b64 %rd187, {%r1032, %r1033}; + sub.f32x2 %rd188, %rd186, %rd187; + mov.b64 %rd189, {%r522, %r523}; + mov.b64 %rd190, {%r1034, %r1035}; + sub.f32x2 %rd191, %rd189, %rd190; + mov.b64 %rd192, {%r524, %r525}; + mov.b64 %rd193, {%r1036, %r1037}; + sub.f32x2 %rd194, %rd192, %rd193; + mov.b64 %rd195, {%r526, %r527}; + mov.b64 %rd196, {%r1038, %r1039}; + sub.f32x2 %rd197, %rd195, %rd196; + mov.b64 %rd198, {%r528, %r529}; + mov.b64 %rd199, {%r1040, %r1041}; + sub.f32x2 %rd200, %rd198, %rd199; + mov.b64 %rd201, {%r530, %r531}; + mov.b64 %rd202, {%r1042, %r1043}; + sub.f32x2 %rd203, %rd201, %rd202; + mov.b64 %rd204, {%r532, %r533}; + mov.b64 %rd205, {%r1044, %r1045}; + sub.f32x2 %rd206, %rd204, %rd205; + mov.b64 %rd207, {%r534, %r535}; + mov.b64 %rd208, {%r1046, %r1047}; + sub.f32x2 %rd209, %rd207, %rd208; + mov.b64 %rd210, {%r536, %r537}; + mov.b64 %rd211, {%r1048, %r1049}; + sub.f32x2 %rd212, %rd210, %rd211; + mov.b64 %rd213, {%r538, %r539}; + mov.b64 %rd214, {%r1050, %r1051}; + sub.f32x2 %rd215, %rd213, %rd214; + mov.b64 %rd216, {%r540, %r541}; + mov.b64 %rd217, {%r1052, %r1053}; + sub.f32x2 %rd218, %rd216, %rd217; + mov.b64 %rd219, {%r542, %r543}; + mov.b64 %rd220, {%r1054, %r1055}; + sub.f32x2 %rd221, %rd219, %rd220; + mov.b64 %rd222, {%r544, %r545}; + mov.b64 %rd223, {%r1056, %r1057}; + sub.f32x2 %rd224, %rd222, %rd223; + mov.b64 %rd225, {%r546, %r547}; + mov.b64 %rd226, {%r1058, %r1059}; + sub.f32x2 %rd227, %rd225, %rd226; + mov.b64 %rd228, {%r548, %r549}; + mov.b64 %rd229, {%r1060, %r1061}; + sub.f32x2 %rd230, %rd228, %rd229; + mov.b64 %rd231, {%r550, %r551}; + mov.b64 %rd232, {%r1062, %r1063}; + sub.f32x2 %rd233, %rd231, %rd232; + mov.b64 %rd234, {%r552, %r553}; + mov.b64 %rd235, {%r1064, %r1065}; + sub.f32x2 %rd236, %rd234, %rd235; + mov.b64 %rd237, {%r554, %r555}; + mov.b64 %rd238, {%r1066, %r1067}; + sub.f32x2 %rd239, %rd237, %rd238; + mov.b64 %rd240, {%r556, %r557}; + mov.b64 %rd241, {%r1068, %r1069}; + sub.f32x2 %rd242, %rd240, %rd241; + mov.b64 %rd243, {%r558, %r559}; + mov.b64 %rd244, {%r1070, %r1071}; + sub.f32x2 %rd245, %rd243, %rd244; + mov.b64 %rd246, {%r560, %r561}; + mov.b64 %rd247, {%r1072, %r1073}; + sub.f32x2 %rd248, %rd246, %rd247; + mov.b64 %rd249, {%r562, %r563}; + mov.b64 %rd250, {%r1074, %r1075}; + sub.f32x2 %rd251, %rd249, %rd250; + mov.b64 %rd252, {%r564, %r565}; + mov.b64 %rd253, {%r1076, %r1077}; + sub.f32x2 %rd254, %rd252, %rd253; + mov.b64 %rd255, {%r566, %r567}; + mov.b64 %rd256, {%r1078, %r1079}; + sub.f32x2 %rd257, %rd255, %rd256; + mov.b64 %rd258, {%r568, %r569}; + mov.b64 %rd259, {%r1080, %r1081}; + sub.f32x2 %rd260, %rd258, %rd259; + mov.b64 %rd261, {%r570, %r571}; + mov.b64 %rd262, {%r1082, %r1083}; + sub.f32x2 %rd263, %rd261, %rd262; + mov.b64 %rd264, {%r572, %r573}; + mov.b64 %rd265, {%r1084, %r1085}; + sub.f32x2 %rd266, %rd264, %rd265; + mov.b64 %rd267, {%r574, %r575}; + mov.b64 %rd268, {%r1086, %r1087}; + sub.f32x2 %rd269, %rd267, %rd268; + mov.b64 %rd270, {%r576, %r577}; + mov.b64 %rd271, {%r1088, %r1089}; + sub.f32x2 %rd272, %rd270, %rd271; + mov.b64 %rd273, {%r578, %r579}; + mov.b64 %rd274, {%r1090, %r1091}; + sub.f32x2 %rd275, %rd273, %rd274; + mov.b64 %rd276, {%r580, %r581}; + mov.b64 %rd277, {%r1092, %r1093}; + sub.f32x2 %rd278, %rd276, %rd277; + mov.b64 %rd279, {%r582, %r583}; + mov.b64 %rd280, {%r1094, %r1095}; + sub.f32x2 %rd281, %rd279, %rd280; + mov.b64 %rd282, {%r584, %r585}; + mov.b64 %rd283, {%r1096, %r1097}; + sub.f32x2 %rd284, %rd282, %rd283; + mov.b64 %rd285, {%r586, %r587}; + mov.b64 %rd286, {%r1098, %r1099}; + sub.f32x2 %rd287, %rd285, %rd286; + mov.b64 %rd288, {%r588, %r589}; + mov.b64 %rd289, {%r1100, %r1101}; + sub.f32x2 %rd290, %rd288, %rd289; + mov.b64 %rd291, {%r590, %r591}; + mov.b64 %rd292, {%r1102, %r1103}; + sub.f32x2 %rd293, %rd291, %rd292; + mov.b64 %rd294, {%r592, %r593}; + mov.b64 %rd295, {%r1104, %r1105}; + sub.f32x2 %rd296, %rd294, %rd295; + mov.b64 %rd297, {%r594, %r595}; + mov.b64 %rd298, {%r1106, %r1107}; + sub.f32x2 %rd299, %rd297, %rd298; + mov.b64 %rd300, {%r596, %r597}; + mov.b64 %rd301, {%r1108, %r1109}; + sub.f32x2 %rd302, %rd300, %rd301; + mov.b64 %rd303, {%r598, %r599}; + mov.b64 %rd304, {%r1110, %r1111}; + sub.f32x2 %rd305, %rd303, %rd304; + mov.b64 %rd306, {%r600, %r601}; + mov.b64 %rd307, {%r1112, %r1113}; + sub.f32x2 %rd308, %rd306, %rd307; + mov.b64 %rd309, {%r602, %r603}; + mov.b64 %rd310, {%r1114, %r1115}; + sub.f32x2 %rd311, %rd309, %rd310; + .loc 1 108 27 // l2norm.py:108:27 + mov.b64 {%r1164, %r1165}, %rd158; + cvt.rn.bf16x2.f32 %r144, %r1165, %r1164; + mov.b64 {%r1166, %r1167}, %rd161; + cvt.rn.bf16x2.f32 %r145, %r1167, %r1166; + mov.b64 {%r1168, %r1169}, %rd164; + cvt.rn.bf16x2.f32 %r146, %r1169, %r1168; + mov.b64 {%r1170, %r1171}, %rd167; + cvt.rn.bf16x2.f32 %r147, %r1171, %r1170; + mov.b64 {%r1172, %r1173}, %rd170; + cvt.rn.bf16x2.f32 %r148, %r1173, %r1172; + mov.b64 {%r1174, %r1175}, %rd173; + cvt.rn.bf16x2.f32 %r149, %r1175, %r1174; + mov.b64 {%r1176, %r1177}, %rd176; + cvt.rn.bf16x2.f32 %r150, %r1177, %r1176; + mov.b64 {%r1178, %r1179}, %rd179; + cvt.rn.bf16x2.f32 %r151, %r1179, %r1178; + mov.b64 {%r1180, %r1181}, %rd182; + cvt.rn.bf16x2.f32 %r152, %r1181, %r1180; + mov.b64 {%r1182, %r1183}, %rd185; + cvt.rn.bf16x2.f32 %r153, %r1183, %r1182; + mov.b64 {%r1184, %r1185}, %rd188; + cvt.rn.bf16x2.f32 %r154, %r1185, %r1184; + mov.b64 {%r1186, %r1187}, %rd191; + cvt.rn.bf16x2.f32 %r155, %r1187, %r1186; + mov.b64 {%r1188, %r1189}, %rd194; + cvt.rn.bf16x2.f32 %r156, %r1189, %r1188; + mov.b64 {%r1190, %r1191}, %rd197; + cvt.rn.bf16x2.f32 %r157, %r1191, %r1190; + mov.b64 {%r1192, %r1193}, %rd200; + cvt.rn.bf16x2.f32 %r158, %r1193, %r1192; + mov.b64 {%r1194, %r1195}, %rd203; + cvt.rn.bf16x2.f32 %r159, %r1195, %r1194; + mov.b64 {%r1196, %r1197}, %rd206; + cvt.rn.bf16x2.f32 %r160, %r1197, %r1196; + mov.b64 {%r1198, %r1199}, %rd209; + cvt.rn.bf16x2.f32 %r161, %r1199, %r1198; + mov.b64 {%r1200, %r1201}, %rd212; + cvt.rn.bf16x2.f32 %r162, %r1201, %r1200; + mov.b64 {%r1202, %r1203}, %rd215; + cvt.rn.bf16x2.f32 %r163, %r1203, %r1202; + mov.b64 {%r1204, %r1205}, %rd218; + cvt.rn.bf16x2.f32 %r164, %r1205, %r1204; + mov.b64 {%r1206, %r1207}, %rd221; + cvt.rn.bf16x2.f32 %r165, %r1207, %r1206; + mov.b64 {%r1208, %r1209}, %rd224; + cvt.rn.bf16x2.f32 %r166, %r1209, %r1208; + mov.b64 {%r1210, %r1211}, %rd227; + cvt.rn.bf16x2.f32 %r167, %r1211, %r1210; + mov.b64 {%r1212, %r1213}, %rd230; + cvt.rn.bf16x2.f32 %r168, %r1213, %r1212; + mov.b64 {%r1214, %r1215}, %rd233; + cvt.rn.bf16x2.f32 %r169, %r1215, %r1214; + mov.b64 {%r1216, %r1217}, %rd236; + cvt.rn.bf16x2.f32 %r170, %r1217, %r1216; + mov.b64 {%r1218, %r1219}, %rd239; + cvt.rn.bf16x2.f32 %r171, %r1219, %r1218; + mov.b64 {%r1220, %r1221}, %rd242; + cvt.rn.bf16x2.f32 %r172, %r1221, %r1220; + mov.b64 {%r1222, %r1223}, %rd245; + cvt.rn.bf16x2.f32 %r173, %r1223, %r1222; + mov.b64 {%r1224, %r1225}, %rd248; + cvt.rn.bf16x2.f32 %r174, %r1225, %r1224; + mov.b64 {%r1226, %r1227}, %rd251; + cvt.rn.bf16x2.f32 %r175, %r1227, %r1226; + mov.b64 {%r1228, %r1229}, %rd254; + cvt.rn.bf16x2.f32 %r176, %r1229, %r1228; + mov.b64 {%r1230, %r1231}, %rd257; + cvt.rn.bf16x2.f32 %r177, %r1231, %r1230; + mov.b64 {%r1232, %r1233}, %rd260; + cvt.rn.bf16x2.f32 %r178, %r1233, %r1232; + mov.b64 {%r1234, %r1235}, %rd263; + cvt.rn.bf16x2.f32 %r179, %r1235, %r1234; + mov.b64 {%r1236, %r1237}, %rd266; + cvt.rn.bf16x2.f32 %r180, %r1237, %r1236; + mov.b64 {%r1238, %r1239}, %rd269; + cvt.rn.bf16x2.f32 %r181, %r1239, %r1238; + mov.b64 {%r1240, %r1241}, %rd272; + cvt.rn.bf16x2.f32 %r182, %r1241, %r1240; + mov.b64 {%r1242, %r1243}, %rd275; + cvt.rn.bf16x2.f32 %r183, %r1243, %r1242; + mov.b64 {%r1244, %r1245}, %rd278; + cvt.rn.bf16x2.f32 %r184, %r1245, %r1244; + mov.b64 {%r1246, %r1247}, %rd281; + cvt.rn.bf16x2.f32 %r185, %r1247, %r1246; + mov.b64 {%r1248, %r1249}, %rd284; + cvt.rn.bf16x2.f32 %r186, %r1249, %r1248; + mov.b64 {%r1250, %r1251}, %rd287; + cvt.rn.bf16x2.f32 %r187, %r1251, %r1250; + mov.b64 {%r1252, %r1253}, %rd290; + cvt.rn.bf16x2.f32 %r188, %r1253, %r1252; + mov.b64 {%r1254, %r1255}, %rd293; + cvt.rn.bf16x2.f32 %r189, %r1255, %r1254; + mov.b64 {%r1256, %r1257}, %rd296; + cvt.rn.bf16x2.f32 %r190, %r1257, %r1256; + mov.b64 {%r1258, %r1259}, %rd299; + cvt.rn.bf16x2.f32 %r191, %r1259, %r1258; + mov.b64 {%r1260, %r1261}, %rd302; + cvt.rn.bf16x2.f32 %r192, %r1261, %r1260; + mov.b64 {%r1262, %r1263}, %rd305; + cvt.rn.bf16x2.f32 %r193, %r1263, %r1262; + mov.b64 {%r1264, %r1265}, %rd308; + cvt.rn.bf16x2.f32 %r194, %r1265, %r1264; + mov.b64 {%r1266, %r1267}, %rd311; + cvt.rn.bf16x2.f32 %r195, %r1267, %r1266; + .loc 1 108 19 // l2norm.py:108:19 + cvt.rn.bf16x2.f32 %r132, %r1119, %r1117; + cvt.rn.bf16x2.f32 %r133, %r1123, %r1121; + cvt.rn.bf16x2.f32 %r134, %r1127, %r1125; + cvt.rn.bf16x2.f32 %r135, %r1131, %r1129; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd34 + 0 ], { %r132, %r133, %r134, %r135 }; + // end inline asm + cvt.rn.bf16x2.f32 %r136, %r1135, %r1133; + cvt.rn.bf16x2.f32 %r137, %r1139, %r1137; + cvt.rn.bf16x2.f32 %r138, %r1143, %r1141; + cvt.rn.bf16x2.f32 %r139, %r1147, %r1145; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd35 + 0 ], { %r136, %r137, %r138, %r139 }; + // end inline asm + cvt.rn.bf16x2.f32 %r140, %r1151, %r1149; + cvt.rn.bf16x2.f32 %r141, %r1155, %r1153; + cvt.rn.bf16x2.f32 %r142, %r1159, %r1157; + cvt.rn.bf16x2.f32 %r143, %r1163, %r1161; + // begin inline asm + @%p3 st.global.v4.b32 [ %rd36 + 0 ], { %r140, %r141, %r142, %r143 }; + // end inline asm + // begin inline asm + @%p4 st.global.v4.b32 [ %rd37 + 0 ], { %r144, %r145, %r146, %r147 }; + // end inline asm + // begin inline asm + @%p5 st.global.v4.b32 [ %rd38 + 0 ], { %r148, %r149, %r150, %r151 }; + // end inline asm + // begin inline asm + @%p6 st.global.v4.b32 [ %rd39 + 0 ], { %r152, %r153, %r154, %r155 }; + // end inline asm + // begin inline asm + @%p7 st.global.v4.b32 [ %rd40 + 0 ], { %r156, %r157, %r158, %r159 }; + // end inline asm + // begin inline asm + @%p8 st.global.v4.b32 [ %rd41 + 0 ], { %r160, %r161, %r162, %r163 }; + // end inline asm + // begin inline asm + @%p17 st.global.v4.b32 [ %rd42 + 0 ], { %r164, %r165, %r166, %r167 }; + // end inline asm + // begin inline asm + @%p9 st.global.v4.b32 [ %rd43 + 0 ], { %r168, %r169, %r170, %r171 }; + // end inline asm + // begin inline asm + @%p10 st.global.v4.b32 [ %rd44 + 0 ], { %r172, %r173, %r174, %r175 }; + // end inline asm + // begin inline asm + @%p11 st.global.v4.b32 [ %rd45 + 0 ], { %r176, %r177, %r178, %r179 }; + // end inline asm + // begin inline asm + @%p12 st.global.v4.b32 [ %rd46 + 0 ], { %r180, %r181, %r182, %r183 }; + // end inline asm + // begin inline asm + @%p13 st.global.v4.b32 [ %rd47 + 0 ], { %r184, %r185, %r186, %r187 }; + // end inline asm + // begin inline asm + @%p14 st.global.v4.b32 [ %rd48 + 0 ], { %r188, %r189, %r190, %r191 }; + // end inline asm + // begin inline asm + @%p15 st.global.v4.b32 [ %rd49 + 0 ], { %r192, %r193, %r194, %r195 }; + // end inline asm + .loc 1 108 4 // l2norm.py:108:4 + ret; +$L__tmp131: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 98 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp130 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 107 // DW_AT_call_line +.b8 55 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp130 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.source b/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..e5004915e0ed656b5a11e815a4f48ddf52d64251 --- /dev/null +++ b/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.source @@ -0,0 +1,210 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc47 = loc(unknown) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc54 = loc("y"(#loc)) +#loc55 = loc("rstd"(#loc)) +#loc56 = loc("dy"(#loc)) +#loc57 = loc("dx"(#loc)) +#loc58 = loc("eps"(#loc)) +#loc59 = loc("T"(#loc)) +#loc101 = loc("input"(#loc45)) +#loc102 = loc("a"(#loc50)) +#loc103 = loc("b"(#loc50)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc60) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc61) + %o_t = arith.constant 64 : i32 loc(#loc62) + %o_t_1 = arith.constant 64 : i64 loc(#loc62) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc62) + %o_t_3 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc63) + %o_t_4 = arith.extsi %o_t_3 : tensor<64xi32> to tensor<64xi64> loc(#loc64) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<64xi64> loc(#loc64) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<64xi64> loc(#loc64) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc65) + %m_t = arith.extsi %T : i32 to i64 loc(#loc66) + %m_t_7 = tt.splat %m_t : i64 -> tensor<64xi64> loc(#loc66) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<64xi64> loc(#loc66) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc67) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc68) + %m_x_10 = arith.constant 128 : i32 loc(#loc69) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc69) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc69) + %m_x_13 = tt.broadcast %m_x : tensor<64x1xi1> -> tensor<64x128xi1> loc(#loc70) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<64x128xi1> loc(#loc70) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<64x128xi1> loc(#loc70) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc71) + %p_y_16 = arith.constant 128 : i32 loc(#loc72) + %p_y_17 = arith.constant 128 : i64 loc(#loc72) + %p_y_18 = arith.constant dense<128> : tensor<64x1xi64> loc(#loc72) + %p_y_19 = arith.muli %p_y, %p_y_18 : tensor<64x1xi64> loc(#loc72) + %p_y_20 = tt.splat %y : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc73) + %p_y_21 = tt.addptr %p_y_20, %p_y_19 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc73) + %p_y_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc74) + %p_y_23 = tt.broadcast %p_y_21 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc75) + %p_y_24 = tt.broadcast %p_y_22 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc75) + %p_y_25 = tt.addptr %p_y_23, %p_y_24 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc75) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<64x!tt.ptr> loc(#loc76) + %p_rstd_26 = tt.addptr %p_rstd, %o_t_6 : tensor<64x!tt.ptr>, tensor<64xi64> loc(#loc76) + %p_dy = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc77) + %p_dy_27 = arith.constant 128 : i32 loc(#loc78) + %p_dy_28 = arith.constant 128 : i64 loc(#loc78) + %p_dy_29 = arith.constant dense<128> : tensor<64x1xi64> loc(#loc78) + %p_dy_30 = arith.muli %p_dy, %p_dy_29 : tensor<64x1xi64> loc(#loc78) + %p_dy_31 = tt.splat %dy : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc79) + %p_dy_32 = tt.addptr %p_dy_31, %p_dy_30 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc79) + %p_dy_33 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc80) + %p_dy_34 = tt.broadcast %p_dy_32 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc81) + %p_dy_35 = tt.broadcast %p_dy_33 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc81) + %p_dy_36 = tt.addptr %p_dy_34, %p_dy_35 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc81) + %p_dx = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc82) + %p_dx_37 = arith.constant 128 : i32 loc(#loc83) + %p_dx_38 = arith.constant 128 : i64 loc(#loc83) + %p_dx_39 = arith.constant dense<128> : tensor<64x1xi64> loc(#loc83) + %p_dx_40 = arith.muli %p_dx, %p_dx_39 : tensor<64x1xi64> loc(#loc83) + %p_dx_41 = tt.splat %dx : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc84) + %p_dx_42 = tt.addptr %p_dx_41, %p_dx_40 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc84) + %p_dx_43 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc85) + %p_dx_44 = tt.broadcast %p_dx_42 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc86) + %p_dx_45 = tt.broadcast %p_dx_43 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc86) + %p_dx_46 = tt.addptr %p_dx_44, %p_dx_45 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc86) + %b_y = arith.constant 0.000000e+00 : f32 loc(#loc87) + %b_y_47 = arith.constant dense<0.000000e+00> : tensor<64x128xf32> loc(#loc87) + %b_y_48 = arith.truncf %b_y_47 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc87) + %b_y_49 = tt.load %p_y_25, %m_x_15, %b_y_48 : tensor<64x128x!tt.ptr> loc(#loc87) + %b_y_50 = arith.extf %b_y_49 : tensor<64x128xbf16> to tensor<64x128xf32> loc(#loc88) + %b_rstd = arith.constant 0.000000e+00 : f32 loc(#loc89) + %b_rstd_51 = arith.constant dense<0.000000e+00> : tensor<64xf32> loc(#loc89) + %b_rstd_52 = tt.load %p_rstd_26, %m_t_8, %b_rstd_51 : tensor<64x!tt.ptr> loc(#loc89) + %b_dy = arith.constant 0.000000e+00 : f32 loc(#loc90) + %b_dy_53 = arith.constant dense<0.000000e+00> : tensor<64x128xf32> loc(#loc90) + %b_dy_54 = arith.truncf %b_dy_53 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc90) + %b_dy_55 = tt.load %p_dy_36, %m_x_15, %b_dy_54 : tensor<64x128x!tt.ptr> loc(#loc90) + %b_dy_56 = arith.extf %b_dy_55 : tensor<64x128xbf16> to tensor<64x128xf32> loc(#loc91) + %b_dx = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc92) + %b_dx_57 = tt.broadcast %b_dx : tensor<64x1xf32> -> tensor<64x128xf32> loc(#loc93) + %b_dx_58 = arith.mulf %b_dy_56, %b_dx_57 : tensor<64x128xf32> loc(#loc93) + %b_dx_59 = arith.mulf %b_dy_56, %b_y_50 : tensor<64x128xf32> loc(#loc94) + %b_dx_60 = tt.call @triton.language.standard.sum__fp32S64_128S_c1_cFalse_cNone(%b_dx_59) : (tensor<64x128xf32>) -> tensor<64xf32> loc(#loc95) + %b_dx_61 = tt.expand_dims %b_dx_60 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc96) + %b_dx_62 = tt.broadcast %b_dx_61 : tensor<64x1xf32> -> tensor<64x128xf32> loc(#loc97) + %b_dx_63 = arith.mulf %b_dx_62, %b_y_50 : tensor<64x128xf32> loc(#loc97) + %b_dx_64 = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc98) + %b_dx_65 = tt.broadcast %b_dx_64 : tensor<64x1xf32> -> tensor<64x128xf32> loc(#loc99) + %b_dx_66 = arith.mulf %b_dx_63, %b_dx_65 : tensor<64x128xf32> loc(#loc99) + %b_dx_67 = arith.subf %b_dx_58, %b_dx_66 : tensor<64x128xf32> loc(#loc100) + %0 = arith.truncf %b_dx_67 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc42) + tt.store %p_dx_46, %0, %m_x_15 : tensor<64x128x!tt.ptr> loc(#loc43) + tt.return loc(#loc44) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S64_128S_c1_cFalse_cNone(%input: tensor<64x128xf32> loc("input"(#loc45))) -> tensor<64xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc46) + tt.reduce.return %2 : f32 loc(#loc46) + }) : (tensor<64x128xf32>) -> tensor<64xf32> loc(#loc46) + tt.return %0 : tensor<64xf32> loc(#loc48) + ^bb1: // no predecessors + %1 = ub.poison : tensor<64xf32> loc(#loc49) + tt.return %1 : tensor<64xf32> loc(#loc49) + } loc(#loc45) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc50)), %b: f32 loc("b"(#loc50))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc51) + tt.return %0 : f32 loc(#loc52) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc53) + tt.return %1 : f32 loc(#loc53) + } loc(#loc50) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:31) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:39) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:20) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:31) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:39) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:82) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc60 = loc("i_t"(#loc1)) +#loc61 = loc("i_t"(#loc2)) +#loc62 = loc("o_t"(#loc3)) +#loc63 = loc("o_t"(#loc4)) +#loc64 = loc("o_t"(#loc5)) +#loc65 = loc("o_d"(#loc6)) +#loc66 = loc("m_t"(#loc7)) +#loc67 = loc("m_x"(#loc8)) +#loc68 = loc("m_x"(#loc9)) +#loc69 = loc("m_x"(#loc10)) +#loc70 = loc("m_x"(#loc11)) +#loc71 = loc("p_y"(#loc12)) +#loc72 = loc("p_y"(#loc13)) +#loc73 = loc("p_y"(#loc14)) +#loc74 = loc("p_y"(#loc15)) +#loc75 = loc("p_y"(#loc16)) +#loc76 = loc("p_rstd"(#loc17)) +#loc77 = loc("p_dy"(#loc18)) +#loc78 = loc("p_dy"(#loc19)) +#loc79 = loc("p_dy"(#loc20)) +#loc80 = loc("p_dy"(#loc21)) +#loc81 = loc("p_dy"(#loc22)) +#loc82 = loc("p_dx"(#loc23)) +#loc83 = loc("p_dx"(#loc24)) +#loc84 = loc("p_dx"(#loc25)) +#loc85 = loc("p_dx"(#loc26)) +#loc86 = loc("p_dx"(#loc27)) +#loc87 = loc("b_y"(#loc28)) +#loc88 = loc("b_y"(#loc29)) +#loc89 = loc("b_rstd"(#loc30)) +#loc90 = loc("b_dy"(#loc31)) +#loc91 = loc("b_dy"(#loc32)) +#loc92 = loc("b_dx"(#loc33)) +#loc93 = loc("b_dx"(#loc34)) +#loc94 = loc("b_dx"(#loc35)) +#loc95 = loc("b_dx"(#loc36)) +#loc96 = loc("b_dx"(#loc37)) +#loc97 = loc("b_dx"(#loc38)) +#loc98 = loc("b_dx"(#loc39)) +#loc99 = loc("b_dx"(#loc40)) +#loc100 = loc("b_dx"(#loc41)) diff --git a/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.ttgir b/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..2b4fb9dc8053dafa1c1a4ad11b668f7294e4719f --- /dev/null +++ b/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.ttgir @@ -0,0 +1,154 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [2, 1], order = [1, 0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [2], order = [0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc39 = loc("y"(#loc)) +#loc40 = loc("rstd"(#loc)) +#loc41 = loc("dy"(#loc)) +#loc42 = loc("dx"(#loc)) +#loc43 = loc("eps"(#loc)) +#loc44 = loc("T"(#loc)) +#loc72 = loc("b_dx"(#loc30)) +#loc78 = loc(callsite(#loc1 at #loc72)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 2 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %c64_i64 = arith.constant 64 : i64 loc(#loc1) + %cst = arith.constant dense<0.000000e+00> : tensor<64x128xbf16, #blocked> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<64xf32, #blocked1> loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<64x1xi64, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc45) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc46) + %o_t = arith.muli %i_t_3, %c64_i64 : i64 loc(#loc47) + %o_t_4 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc48) + %o_t_5 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #blocked1> loc(#loc48) + %o_t_6 = arith.extsi %o_t_4 : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %o_t_7 = arith.extsi %o_t_5 : tensor<64xi32, #blocked1> to tensor<64xi64, #blocked1> loc(#loc49) + %o_t_8 = tt.splat %o_t : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %o_t_9 = tt.splat %o_t : i64 -> tensor<64xi64, #blocked1> loc(#loc49) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<64xi64, #blocked1> loc(#loc49) + %m_t = arith.extsi %T : i32 to i64 loc(#loc50) + %m_t_12 = tt.splat %m_t : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc50) + %m_t_13 = tt.splat %m_t : i64 -> tensor<64xi64, #blocked1> loc(#loc50) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc50) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<64xi64, #blocked1> loc(#loc50) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<64xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64x1xi1, #blocked> loc(#loc51) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc52) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc52) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_2 : tensor<1x128xi32, #blocked> loc(#loc53) + %m_x_19 = tt.broadcast %m_x : tensor<64x1xi1, #blocked> -> tensor<64x128xi1, #blocked> loc(#loc54) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<64x128xi1, #blocked> loc(#loc54) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<64x128xi1, #blocked> loc(#loc54) + %p_y = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64x1xi64, #blocked> loc(#loc55) + %p_y_22 = arith.muli %p_y, %cst_1 : tensor<64x1xi64, #blocked> loc(#loc56) + %p_y_23 = tt.splat %y : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked> loc(#loc57) + %p_y_24 = tt.addptr %p_y_23, %p_y_22 : tensor<64x1x!tt.ptr, #blocked>, tensor<64x1xi64, #blocked> loc(#loc57) + %p_y_25 = tt.broadcast %p_y_24 : tensor<64x1x!tt.ptr, #blocked> -> tensor<64x128x!tt.ptr, #blocked> loc(#loc58) + %p_y_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<64x128xi32, #blocked> loc(#loc58) + %p_y_27 = tt.addptr %p_y_25, %p_y_26 : tensor<64x128x!tt.ptr, #blocked>, tensor<64x128xi32, #blocked> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<64x!tt.ptr, #blocked1> loc(#loc59) + %p_rstd_28 = tt.addptr %p_rstd, %o_t_11 : tensor<64x!tt.ptr, #blocked1>, tensor<64xi64, #blocked1> loc(#loc59) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked> loc(#loc60) + %p_dy_29 = tt.addptr %p_dy, %p_y_22 : tensor<64x1x!tt.ptr, #blocked>, tensor<64x1xi64, #blocked> loc(#loc60) + %p_dy_30 = tt.broadcast %p_dy_29 : tensor<64x1x!tt.ptr, #blocked> -> tensor<64x128x!tt.ptr, #blocked> loc(#loc61) + %p_dy_31 = tt.addptr %p_dy_30, %p_y_26 : tensor<64x128x!tt.ptr, #blocked>, tensor<64x128xi32, #blocked> loc(#loc61) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked> loc(#loc62) + %p_dx_32 = tt.addptr %p_dx, %p_y_22 : tensor<64x1x!tt.ptr, #blocked>, tensor<64x1xi64, #blocked> loc(#loc62) + %p_dx_33 = tt.broadcast %p_dx_32 : tensor<64x1x!tt.ptr, #blocked> -> tensor<64x128x!tt.ptr, #blocked> loc(#loc63) + %p_dx_34 = tt.addptr %p_dx_33, %p_y_26 : tensor<64x128x!tt.ptr, #blocked>, tensor<64x128xi32, #blocked> loc(#loc63) + %b_y = tt.load %p_y_27, %m_x_21, %cst : tensor<64x128x!tt.ptr, #blocked> loc(#loc64) + %b_y_35 = arith.extf %b_y : tensor<64x128xbf16, #blocked> to tensor<64x128xf32, #blocked> loc(#loc65) + %b_rstd = tt.load %p_rstd_28, %m_t_15, %cst_0 : tensor<64x!tt.ptr, #blocked1> loc(#loc66) + %b_dy = tt.load %p_dy_31, %m_x_21, %cst : tensor<64x128x!tt.ptr, #blocked> loc(#loc67) + %b_dy_36 = arith.extf %b_dy : tensor<64x128xbf16, #blocked> to tensor<64x128xf32, #blocked> loc(#loc68) + %b_dx = ttg.convert_layout %b_rstd : tensor<64xf32, #blocked1> -> tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc69) + %b_dx_37 = tt.expand_dims %b_dx {axis = 1 : i32} : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64x1xf32, #blocked> loc(#loc70) + %b_dx_38 = tt.broadcast %b_dx_37 : tensor<64x1xf32, #blocked> -> tensor<64x128xf32, #blocked> loc(#loc69) + %b_dx_39 = arith.mulf %b_dy_36, %b_dx_38 : tensor<64x128xf32, #blocked> loc(#loc69) + %b_dx_40 = arith.mulf %b_dy_36, %b_y_35 : tensor<64x128xf32, #blocked> loc(#loc71) + %b_dx_41 = "tt.reduce"(%b_dx_40) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_47: f32 loc(callsite(#loc1 at #loc72)), %b_dx_48: f32 loc(callsite(#loc1 at #loc72))): + %b_dx_49 = arith.addf %b_dx_47, %b_dx_48 : f32 loc(#loc79) + tt.reduce.return %b_dx_49 : f32 loc(#loc77) + }) : (tensor<64x128xf32, #blocked>) -> tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc77) + %b_dx_42 = tt.expand_dims %b_dx_41 {axis = 1 : i32} : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64x1xf32, #blocked> loc(#loc73) + %b_dx_43 = tt.broadcast %b_dx_42 : tensor<64x1xf32, #blocked> -> tensor<64x128xf32, #blocked> loc(#loc74) + %b_dx_44 = arith.mulf %b_dx_43, %b_y_35 : tensor<64x128xf32, #blocked> loc(#loc74) + %b_dx_45 = arith.mulf %b_dx_44, %b_dx_38 : tensor<64x128xf32, #blocked> loc(#loc75) + %b_dx_46 = arith.subf %b_dx_39, %b_dx_45 : tensor<64x128xf32, #blocked> loc(#loc76) + %0 = arith.truncf %b_dx_46 : tensor<64x128xf32, #blocked> to tensor<64x128xbf16, #blocked> loc(#loc36) + tt.store %p_dx_34, %0, %m_x_21 : tensor<64x128x!tt.ptr, #blocked> loc(#loc37) + tt.return loc(#loc38) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc45 = loc("i_t"(#loc2)) +#loc46 = loc("i_t"(#loc3)) +#loc47 = loc("o_t"(#loc4)) +#loc48 = loc("o_t"(#loc5)) +#loc49 = loc("o_t"(#loc6)) +#loc50 = loc("m_t"(#loc7)) +#loc51 = loc("m_x"(#loc8)) +#loc52 = loc("m_x"(#loc9)) +#loc53 = loc("m_x"(#loc10)) +#loc54 = loc("m_x"(#loc11)) +#loc55 = loc("p_y"(#loc12)) +#loc56 = loc("p_y"(#loc13)) +#loc57 = loc("p_y"(#loc14)) +#loc58 = loc("p_y"(#loc15)) +#loc59 = loc("p_rstd"(#loc16)) +#loc60 = loc("p_dy"(#loc17)) +#loc61 = loc("p_dy"(#loc18)) +#loc62 = loc("p_dx"(#loc19)) +#loc63 = loc("p_dx"(#loc20)) +#loc64 = loc("b_y"(#loc21)) +#loc65 = loc("b_y"(#loc22)) +#loc66 = loc("b_rstd"(#loc23)) +#loc67 = loc("b_dy"(#loc24)) +#loc68 = loc("b_dy"(#loc25)) +#loc69 = loc("b_dx"(#loc26)) +#loc70 = loc("b_dx"(#loc27)) +#loc71 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc32)) +#loc74 = loc("b_dx"(#loc33)) +#loc75 = loc("b_dx"(#loc34)) +#loc76 = loc("b_dx"(#loc35)) +#loc77 = loc(callsite(#loc29 at #loc72)) +#loc79 = loc(callsite(#loc31 at #loc77)) diff --git a/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.ttir b/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..3b703edcc6d40b9b8fa5aa712507941f59d4cd5b --- /dev/null +++ b/triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.ttir @@ -0,0 +1,147 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc40 = loc("y"(#loc)) +#loc41 = loc("rstd"(#loc)) +#loc42 = loc("dy"(#loc)) +#loc43 = loc("dx"(#loc)) +#loc44 = loc("eps"(#loc)) +#loc45 = loc("T"(#loc)) +#loc74 = loc("b_dx"(#loc31)) +#loc80 = loc(callsite(#loc1 at #loc74)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<0.000000e+00> : tensor<64x128xbf16> loc(#loc1) + %b_rstd = arith.constant dense<0.000000e+00> : tensor<64xf32> loc(#loc46) + %cst_0 = arith.constant dense<128> : tensor<64x1xi64> loc(#loc1) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc47) + %o_t = arith.constant 64 : i64 loc(#loc48) + %i_t = tt.get_program_id x : i32 loc(#loc49) + %i_t_1 = arith.extsi %i_t : i32 to i64 loc(#loc50) + %o_t_2 = arith.muli %i_t_1, %o_t : i64 loc(#loc48) + %o_t_3 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc51) + %o_t_4 = arith.extsi %o_t_3 : tensor<64xi32> to tensor<64xi64> loc(#loc52) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<64xi64> loc(#loc52) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<64xi64> loc(#loc52) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc53) + %m_t = arith.extsi %T : i32 to i64 loc(#loc54) + %m_t_7 = tt.splat %m_t : i64 -> tensor<64xi64> loc(#loc54) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<64xi64> loc(#loc54) + %m_x_9 = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc55) + %m_x_10 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc56) + %m_x_11 = arith.cmpi slt, %m_x_10, %m_x : tensor<1x128xi32> loc(#loc47) + %m_x_12 = tt.broadcast %m_x_9 : tensor<64x1xi1> -> tensor<64x128xi1> loc(#loc57) + %m_x_13 = tt.broadcast %m_x_11 : tensor<1x128xi1> -> tensor<64x128xi1> loc(#loc57) + %m_x_14 = arith.andi %m_x_12, %m_x_13 : tensor<64x128xi1> loc(#loc57) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc58) + %p_y_15 = arith.muli %p_y, %cst_0 : tensor<64x1xi64> loc(#loc59) + %p_y_16 = tt.splat %y : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc60) + %p_y_17 = tt.addptr %p_y_16, %p_y_15 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc60) + %p_y_18 = tt.broadcast %p_y_17 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc61) + %p_y_19 = tt.broadcast %m_x_10 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc61) + %p_y_20 = tt.addptr %p_y_18, %p_y_19 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc61) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<64x!tt.ptr> loc(#loc62) + %p_rstd_21 = tt.addptr %p_rstd, %o_t_6 : tensor<64x!tt.ptr>, tensor<64xi64> loc(#loc62) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc63) + %p_dy_22 = tt.addptr %p_dy, %p_y_15 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc63) + %p_dy_23 = tt.broadcast %p_dy_22 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc64) + %p_dy_24 = tt.addptr %p_dy_23, %p_y_19 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc64) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc65) + %p_dx_25 = tt.addptr %p_dx, %p_y_15 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc65) + %p_dx_26 = tt.broadcast %p_dx_25 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc66) + %p_dx_27 = tt.addptr %p_dx_26, %p_y_19 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc66) + %b_y = tt.load %p_y_20, %m_x_14, %cst : tensor<64x128x!tt.ptr> loc(#loc67) + %b_y_28 = arith.extf %b_y : tensor<64x128xbf16> to tensor<64x128xf32> loc(#loc68) + %b_rstd_29 = tt.load %p_rstd_21, %m_t_8, %b_rstd : tensor<64x!tt.ptr> loc(#loc46) + %b_dy = tt.load %p_dy_24, %m_x_14, %cst : tensor<64x128x!tt.ptr> loc(#loc69) + %b_dy_30 = arith.extf %b_dy : tensor<64x128xbf16> to tensor<64x128xf32> loc(#loc70) + %b_dx = tt.expand_dims %b_rstd_29 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc71) + %b_dx_31 = tt.broadcast %b_dx : tensor<64x1xf32> -> tensor<64x128xf32> loc(#loc72) + %b_dx_32 = arith.mulf %b_dy_30, %b_dx_31 : tensor<64x128xf32> loc(#loc72) + %b_dx_33 = arith.mulf %b_dy_30, %b_y_28 : tensor<64x128xf32> loc(#loc73) + %b_dx_34 = "tt.reduce"(%b_dx_33) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_40: f32 loc(callsite(#loc1 at #loc74)), %b_dx_41: f32 loc(callsite(#loc1 at #loc74))): + %b_dx_42 = arith.addf %b_dx_40, %b_dx_41 : f32 loc(#loc81) + tt.reduce.return %b_dx_42 : f32 loc(#loc79) + }) : (tensor<64x128xf32>) -> tensor<64xf32> loc(#loc79) + %b_dx_35 = tt.expand_dims %b_dx_34 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc75) + %b_dx_36 = tt.broadcast %b_dx_35 : tensor<64x1xf32> -> tensor<64x128xf32> loc(#loc76) + %b_dx_37 = arith.mulf %b_dx_36, %b_y_28 : tensor<64x128xf32> loc(#loc76) + %b_dx_38 = arith.mulf %b_dx_37, %b_dx_31 : tensor<64x128xf32> loc(#loc77) + %b_dx_39 = arith.subf %b_dx_32, %b_dx_38 : tensor<64x128xf32> loc(#loc78) + %0 = arith.truncf %b_dx_39 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc37) + tt.store %p_dx_27, %0, %m_x_14 : tensor<64x128x!tt.ptr> loc(#loc38) + tt.return loc(#loc39) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("b_rstd"(#loc2)) +#loc47 = loc("m_x"(#loc3)) +#loc48 = loc("o_t"(#loc4)) +#loc49 = loc("i_t"(#loc5)) +#loc50 = loc("i_t"(#loc6)) +#loc51 = loc("o_t"(#loc7)) +#loc52 = loc("o_t"(#loc8)) +#loc53 = loc("o_d"(#loc9)) +#loc54 = loc("m_t"(#loc10)) +#loc55 = loc("m_x"(#loc11)) +#loc56 = loc("m_x"(#loc12)) +#loc57 = loc("m_x"(#loc13)) +#loc58 = loc("p_y"(#loc14)) +#loc59 = loc("p_y"(#loc15)) +#loc60 = loc("p_y"(#loc16)) +#loc61 = loc("p_y"(#loc17)) +#loc62 = loc("p_rstd"(#loc18)) +#loc63 = loc("p_dy"(#loc19)) +#loc64 = loc("p_dy"(#loc20)) +#loc65 = loc("p_dx"(#loc21)) +#loc66 = loc("p_dx"(#loc22)) +#loc67 = loc("b_y"(#loc23)) +#loc68 = loc("b_y"(#loc24)) +#loc69 = loc("b_dy"(#loc25)) +#loc70 = loc("b_dy"(#loc26)) +#loc71 = loc("b_dx"(#loc27)) +#loc72 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc29)) +#loc75 = loc("b_dx"(#loc33)) +#loc76 = loc("b_dx"(#loc34)) +#loc77 = loc("b_dx"(#loc35)) +#loc78 = loc("b_dx"(#loc36)) +#loc79 = loc(callsite(#loc30 at #loc74)) +#loc81 = loc(callsite(#loc32 at #loc79)) diff --git a/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/__grp__l2norm_fwd_kernel.json b/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/__grp__l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..9a558a984ed0c9e1ffad82bf811f337219deebdb --- /dev/null +++ b/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/__grp__l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_fwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.source", "l2norm_fwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.ttir", "l2norm_fwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.ttgir", "l2norm_fwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.llir", "l2norm_fwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.ptx", "l2norm_fwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.cubin", "l2norm_fwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.cubin b/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..e617d532bfd0aa8e273513ca942044878259763a --- /dev/null +++ b/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.cubin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:90dbfe30b76c691acb0425888b284030758d8f8ba0f1e207b84590a8b84ef881 +size 160296 diff --git a/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.json b/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..5feb933214e25b674766fd39a35416b876d857c7 --- /dev/null +++ b/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"hash": "17133313d17370126a6806eb0694fd24f687dcbd091aec5b7417861444f2bfc1", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 1, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 256, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_fwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.llir b/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..e58d03c0890ee8ee7e1bc22729785de0e4699ee4 --- /dev/null +++ b/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.llir @@ -0,0 +1,4174 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 +@.str.2 = private unnamed_addr constant [17 x i8] c"__CUDA_PREC_SQRT\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_fwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, float %3, i32 %4, ptr addrspace(1) readnone captures(none) %5, ptr addrspace(1) readnone captures(none) %6) local_unnamed_addr #0 !dbg !5 { + %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %9 = zext nneg i32 %8 to i64, !dbg !14 + %10 = shl nuw nsw i64 %9, 6, !dbg !15 + %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !16 + %12 = and i32 %11, 16, !dbg !16 + %.lobit = lshr exact i32 %12, 4, !dbg !16 + %13 = or disjoint i32 %.lobit, 2, !dbg !16 + %14 = insertelement <2 x i32> poison, i32 %.lobit, i64 0, !dbg !16 + %15 = insertelement <4 x i32> poison, i32 %.lobit, i64 0, !dbg !16 + %16 = insertelement <8 x i32> poison, i32 %.lobit, i64 0, !dbg !16 + %17 = insertelement <16 x i32> poison, i32 %.lobit, i64 0, !dbg !16 + %18 = insertelement <32 x i32> poison, i32 %13, i64 30, !dbg !17 + %19 = insertelement <32 x i32> %18, i32 %.lobit, i64 31, !dbg !17 + %20 = shufflevector <16 x i32> %17, <16 x i32> poison, <32 x i32> , !dbg !17 + %21 = or disjoint <32 x i32> %20, , !dbg !17 + %22 = shufflevector <32 x i32> %21, <32 x i32> %19, <32 x i32> , !dbg !17 + %23 = shufflevector <8 x i32> %16, <8 x i32> poison, <32 x i32> , !dbg !17 + %24 = or disjoint <32 x i32> %23, , !dbg !17 + %25 = shufflevector <32 x i32> %22, <32 x i32> %24, <32 x i32> , !dbg !17 + %26 = shufflevector <4 x i32> %15, <4 x i32> poison, <32 x i32> , !dbg !17 + %27 = or disjoint <32 x i32> %26, , !dbg !17 + %28 = shufflevector <32 x i32> %25, <32 x i32> %27, <32 x i32> , !dbg !17 + %29 = shufflevector <2 x i32> %14, <2 x i32> poison, <32 x i32> , !dbg !17 + %30 = or disjoint <32 x i32> %29, , !dbg !17 + %31 = shufflevector <32 x i32> %28, <32 x i32> %30, <32 x i32> , !dbg !17 + %32 = zext <32 x i32> %31 to <32 x i64>, !dbg !17 + %33 = insertelement <32 x i64> poison, i64 %10, i64 0, !dbg !17 + %34 = shufflevector <32 x i64> %33, <32 x i64> poison, <32 x i32> zeroinitializer, !dbg !17 + %35 = or disjoint <32 x i64> %34, %32, !dbg !17 + %36 = insertelement <32 x i32> poison, i32 %4, i64 0, !dbg !18 + %37 = shufflevector <32 x i32> %36, <32 x i32> poison, <32 x i32> zeroinitializer, !dbg !18 + %38 = sext <32 x i32> %37 to <32 x i64>, !dbg !18 + %39 = icmp slt <32 x i64> %35, %38, !dbg !18 + %40 = and i32 %11, 15, !dbg !19 + %41 = shl nuw nsw i32 %40, 3, !dbg !19 + %42 = extractelement <32 x i64> %35, i64 31, !dbg !20 + %43 = shl nuw nsw i64 %42, 7, !dbg !20 + %44 = extractelement <32 x i64> %35, i64 30, !dbg !20 + %45 = shl nuw nsw i64 %44, 7, !dbg !20 + %46 = extractelement <32 x i64> %35, i64 29, !dbg !20 + %47 = shl nuw nsw i64 %46, 7, !dbg !20 + %48 = extractelement <32 x i64> %35, i64 28, !dbg !20 + %49 = shl nuw nsw i64 %48, 7, !dbg !20 + %50 = extractelement <32 x i64> %35, i64 27, !dbg !20 + %51 = shl nuw nsw i64 %50, 7, !dbg !20 + %52 = extractelement <32 x i64> %35, i64 26, !dbg !20 + %53 = shl nuw nsw i64 %52, 7, !dbg !20 + %54 = extractelement <32 x i64> %35, i64 25, !dbg !20 + %55 = shl nuw nsw i64 %54, 7, !dbg !20 + %56 = extractelement <32 x i64> %35, i64 24, !dbg !20 + %57 = shl nuw nsw i64 %56, 7, !dbg !20 + %58 = extractelement <32 x i64> %35, i64 23, !dbg !20 + %59 = shl nuw nsw i64 %58, 7, !dbg !20 + %60 = extractelement <32 x i64> %35, i64 22, !dbg !20 + %61 = shl nuw nsw i64 %60, 7, !dbg !20 + %62 = extractelement <32 x i64> %35, i64 21, !dbg !20 + %63 = shl nuw nsw i64 %62, 7, !dbg !20 + %64 = extractelement <32 x i64> %35, i64 20, !dbg !20 + %65 = shl nuw nsw i64 %64, 7, !dbg !20 + %66 = extractelement <32 x i64> %35, i64 19, !dbg !20 + %67 = shl nuw nsw i64 %66, 7, !dbg !20 + %68 = extractelement <32 x i64> %35, i64 18, !dbg !20 + %69 = shl nuw nsw i64 %68, 7, !dbg !20 + %70 = extractelement <32 x i64> %35, i64 17, !dbg !20 + %71 = shl nuw nsw i64 %70, 7, !dbg !20 + %72 = extractelement <32 x i64> %35, i64 16, !dbg !20 + %73 = shl nuw nsw i64 %72, 7, !dbg !20 + %74 = extractelement <32 x i64> %35, i64 15, !dbg !20 + %75 = shl nuw nsw i64 %74, 7, !dbg !20 + %76 = extractelement <32 x i64> %35, i64 14, !dbg !20 + %77 = shl nuw nsw i64 %76, 7, !dbg !20 + %78 = extractelement <32 x i64> %35, i64 13, !dbg !20 + %79 = shl nuw nsw i64 %78, 7, !dbg !20 + %80 = extractelement <32 x i64> %35, i64 12, !dbg !20 + %81 = shl nuw nsw i64 %80, 7, !dbg !20 + %82 = extractelement <32 x i64> %35, i64 11, !dbg !20 + %83 = shl nuw nsw i64 %82, 7, !dbg !20 + %84 = extractelement <32 x i64> %35, i64 10, !dbg !20 + %85 = shl nuw nsw i64 %84, 7, !dbg !20 + %86 = extractelement <32 x i64> %35, i64 9, !dbg !20 + %87 = shl nuw nsw i64 %86, 7, !dbg !20 + %88 = extractelement <32 x i64> %35, i64 8, !dbg !20 + %89 = shl nuw nsw i64 %88, 7, !dbg !20 + %90 = extractelement <32 x i64> %35, i64 7, !dbg !20 + %91 = shl nuw nsw i64 %90, 7, !dbg !20 + %92 = extractelement <32 x i64> %35, i64 6, !dbg !20 + %93 = shl nuw nsw i64 %92, 7, !dbg !20 + %94 = extractelement <32 x i64> %35, i64 5, !dbg !20 + %95 = shl nuw nsw i64 %94, 7, !dbg !20 + %96 = extractelement <32 x i64> %35, i64 4, !dbg !20 + %97 = shl nuw nsw i64 %96, 7, !dbg !20 + %98 = extractelement <32 x i64> %35, i64 3, !dbg !20 + %99 = shl nuw nsw i64 %98, 7, !dbg !20 + %100 = extractelement <32 x i64> %35, i64 2, !dbg !20 + %101 = shl nuw nsw i64 %100, 7, !dbg !20 + %102 = extractelement <32 x i64> %35, i64 1, !dbg !20 + %103 = shl nuw nsw i64 %102, 7, !dbg !20 + %104 = extractelement <32 x i64> %35, i64 0, !dbg !20 + %105 = shl nuw nsw i64 %104, 7, !dbg !20 + %106 = getelementptr bfloat, ptr addrspace(1) %0, i64 %43, !dbg !21 + %107 = getelementptr bfloat, ptr addrspace(1) %0, i64 %45, !dbg !21 + %108 = getelementptr bfloat, ptr addrspace(1) %0, i64 %47, !dbg !21 + %109 = getelementptr bfloat, ptr addrspace(1) %0, i64 %49, !dbg !21 + %110 = getelementptr bfloat, ptr addrspace(1) %0, i64 %51, !dbg !21 + %111 = getelementptr bfloat, ptr addrspace(1) %0, i64 %53, !dbg !21 + %112 = getelementptr bfloat, ptr addrspace(1) %0, i64 %55, !dbg !21 + %113 = getelementptr bfloat, ptr addrspace(1) %0, i64 %57, !dbg !21 + %114 = getelementptr bfloat, ptr addrspace(1) %0, i64 %59, !dbg !21 + %115 = getelementptr bfloat, ptr addrspace(1) %0, i64 %61, !dbg !21 + %116 = getelementptr bfloat, ptr addrspace(1) %0, i64 %63, !dbg !21 + %117 = getelementptr bfloat, ptr addrspace(1) %0, i64 %65, !dbg !21 + %118 = getelementptr bfloat, ptr addrspace(1) %0, i64 %67, !dbg !21 + %119 = getelementptr bfloat, ptr addrspace(1) %0, i64 %69, !dbg !21 + %120 = getelementptr bfloat, ptr addrspace(1) %0, i64 %71, !dbg !21 + %121 = getelementptr bfloat, ptr addrspace(1) %0, i64 %73, !dbg !21 + %122 = getelementptr bfloat, ptr addrspace(1) %0, i64 %75, !dbg !21 + %123 = getelementptr bfloat, ptr addrspace(1) %0, i64 %77, !dbg !21 + %124 = getelementptr bfloat, ptr addrspace(1) %0, i64 %79, !dbg !21 + %125 = getelementptr bfloat, ptr addrspace(1) %0, i64 %81, !dbg !21 + %126 = getelementptr bfloat, ptr addrspace(1) %0, i64 %83, !dbg !21 + %127 = getelementptr bfloat, ptr addrspace(1) %0, i64 %85, !dbg !21 + %128 = getelementptr bfloat, ptr addrspace(1) %0, i64 %87, !dbg !21 + %129 = getelementptr bfloat, ptr addrspace(1) %0, i64 %89, !dbg !21 + %130 = getelementptr bfloat, ptr addrspace(1) %0, i64 %91, !dbg !21 + %131 = getelementptr bfloat, ptr addrspace(1) %0, i64 %93, !dbg !21 + %132 = getelementptr bfloat, ptr addrspace(1) %0, i64 %95, !dbg !21 + %133 = getelementptr bfloat, ptr addrspace(1) %0, i64 %97, !dbg !21 + %134 = getelementptr bfloat, ptr addrspace(1) %0, i64 %99, !dbg !21 + %135 = getelementptr bfloat, ptr addrspace(1) %0, i64 %101, !dbg !21 + %136 = getelementptr bfloat, ptr addrspace(1) %0, i64 %103, !dbg !21 + %137 = getelementptr bfloat, ptr addrspace(1) %0, i64 %105, !dbg !21 + %138 = zext nneg i32 %41 to i64, !dbg !22 + %139 = getelementptr bfloat, ptr addrspace(1) %106, i64 %138, !dbg !22 + %140 = getelementptr bfloat, ptr addrspace(1) %107, i64 %138, !dbg !22 + %141 = getelementptr bfloat, ptr addrspace(1) %108, i64 %138, !dbg !22 + %142 = getelementptr bfloat, ptr addrspace(1) %109, i64 %138, !dbg !22 + %143 = getelementptr bfloat, ptr addrspace(1) %110, i64 %138, !dbg !22 + %144 = getelementptr bfloat, ptr addrspace(1) %111, i64 %138, !dbg !22 + %145 = getelementptr bfloat, ptr addrspace(1) %112, i64 %138, !dbg !22 + %146 = getelementptr bfloat, ptr addrspace(1) %113, i64 %138, !dbg !22 + %147 = getelementptr bfloat, ptr addrspace(1) %114, i64 %138, !dbg !22 + %148 = getelementptr bfloat, ptr addrspace(1) %115, i64 %138, !dbg !22 + %149 = getelementptr bfloat, ptr addrspace(1) %116, i64 %138, !dbg !22 + %150 = getelementptr bfloat, ptr addrspace(1) %117, i64 %138, !dbg !22 + %151 = getelementptr bfloat, ptr addrspace(1) %118, i64 %138, !dbg !22 + %152 = getelementptr bfloat, ptr addrspace(1) %119, i64 %138, !dbg !22 + %153 = getelementptr bfloat, ptr addrspace(1) %120, i64 %138, !dbg !22 + %154 = getelementptr bfloat, ptr addrspace(1) %121, i64 %138, !dbg !22 + %155 = getelementptr bfloat, ptr addrspace(1) %122, i64 %138, !dbg !22 + %156 = getelementptr bfloat, ptr addrspace(1) %123, i64 %138, !dbg !22 + %157 = getelementptr bfloat, ptr addrspace(1) %124, i64 %138, !dbg !22 + %158 = getelementptr bfloat, ptr addrspace(1) %125, i64 %138, !dbg !22 + %159 = getelementptr bfloat, ptr addrspace(1) %126, i64 %138, !dbg !22 + %160 = getelementptr bfloat, ptr addrspace(1) %127, i64 %138, !dbg !22 + %161 = getelementptr bfloat, ptr addrspace(1) %128, i64 %138, !dbg !22 + %162 = getelementptr bfloat, ptr addrspace(1) %129, i64 %138, !dbg !22 + %163 = getelementptr bfloat, ptr addrspace(1) %130, i64 %138, !dbg !22 + %164 = getelementptr bfloat, ptr addrspace(1) %131, i64 %138, !dbg !22 + %165 = getelementptr bfloat, ptr addrspace(1) %132, i64 %138, !dbg !22 + %166 = getelementptr bfloat, ptr addrspace(1) %133, i64 %138, !dbg !22 + %167 = getelementptr bfloat, ptr addrspace(1) %134, i64 %138, !dbg !22 + %168 = getelementptr bfloat, ptr addrspace(1) %135, i64 %138, !dbg !22 + %169 = getelementptr bfloat, ptr addrspace(1) %136, i64 %138, !dbg !22 + %170 = getelementptr bfloat, ptr addrspace(1) %137, i64 %138, !dbg !22 + %171 = extractelement <32 x i1> %39, i64 31, !dbg !23 + %172 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %139, i1 %171) #7, !dbg !23 + %173 = extractvalue { i32, i32, i32, i32 } %172, 0, !dbg !23 + %174 = bitcast i32 %173 to <2 x bfloat>, !dbg !23 + %175 = extractvalue { i32, i32, i32, i32 } %172, 1, !dbg !23 + %176 = bitcast i32 %175 to <2 x bfloat>, !dbg !23 + %177 = extractvalue { i32, i32, i32, i32 } %172, 2, !dbg !23 + %178 = bitcast i32 %177 to <2 x bfloat>, !dbg !23 + %179 = extractvalue { i32, i32, i32, i32 } %172, 3, !dbg !23 + %180 = bitcast i32 %179 to <2 x bfloat>, !dbg !23 + %181 = extractelement <2 x bfloat> %174, i64 0, !dbg !23 + %182 = extractelement <2 x bfloat> %174, i64 1, !dbg !23 + %183 = extractelement <2 x bfloat> %176, i64 0, !dbg !23 + %184 = extractelement <2 x bfloat> %176, i64 1, !dbg !23 + %185 = extractelement <2 x bfloat> %178, i64 0, !dbg !23 + %186 = extractelement <2 x bfloat> %178, i64 1, !dbg !23 + %187 = extractelement <2 x bfloat> %180, i64 0, !dbg !23 + %188 = extractelement <2 x bfloat> %180, i64 1, !dbg !23 + %189 = extractelement <32 x i1> %39, i64 30, !dbg !23 + %190 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %140, i1 %189) #7, !dbg !23 + %191 = extractvalue { i32, i32, i32, i32 } %190, 0, !dbg !23 + %192 = bitcast i32 %191 to <2 x bfloat>, !dbg !23 + %193 = extractvalue { i32, i32, i32, i32 } %190, 1, !dbg !23 + %194 = bitcast i32 %193 to <2 x bfloat>, !dbg !23 + %195 = extractvalue { i32, i32, i32, i32 } %190, 2, !dbg !23 + %196 = bitcast i32 %195 to <2 x bfloat>, !dbg !23 + %197 = extractvalue { i32, i32, i32, i32 } %190, 3, !dbg !23 + %198 = bitcast i32 %197 to <2 x bfloat>, !dbg !23 + %199 = extractelement <2 x bfloat> %192, i64 0, !dbg !23 + %200 = extractelement <2 x bfloat> %192, i64 1, !dbg !23 + %201 = extractelement <2 x bfloat> %194, i64 0, !dbg !23 + %202 = extractelement <2 x bfloat> %194, i64 1, !dbg !23 + %203 = extractelement <2 x bfloat> %196, i64 0, !dbg !23 + %204 = extractelement <2 x bfloat> %196, i64 1, !dbg !23 + %205 = extractelement <2 x bfloat> %198, i64 0, !dbg !23 + %206 = extractelement <2 x bfloat> %198, i64 1, !dbg !23 + %207 = extractelement <32 x i1> %39, i64 29, !dbg !23 + %208 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %141, i1 %207) #7, !dbg !23 + %209 = extractvalue { i32, i32, i32, i32 } %208, 0, !dbg !23 + %210 = bitcast i32 %209 to <2 x bfloat>, !dbg !23 + %211 = extractvalue { i32, i32, i32, i32 } %208, 1, !dbg !23 + %212 = bitcast i32 %211 to <2 x bfloat>, !dbg !23 + %213 = extractvalue { i32, i32, i32, i32 } %208, 2, !dbg !23 + %214 = bitcast i32 %213 to <2 x bfloat>, !dbg !23 + %215 = extractvalue { i32, i32, i32, i32 } %208, 3, !dbg !23 + %216 = bitcast i32 %215 to <2 x bfloat>, !dbg !23 + %217 = extractelement <2 x bfloat> %210, i64 0, !dbg !23 + %218 = extractelement <2 x bfloat> %210, i64 1, !dbg !23 + %219 = extractelement <2 x bfloat> %212, i64 0, !dbg !23 + %220 = extractelement <2 x bfloat> %212, i64 1, !dbg !23 + %221 = extractelement <2 x bfloat> %214, i64 0, !dbg !23 + %222 = extractelement <2 x bfloat> %214, i64 1, !dbg !23 + %223 = extractelement <2 x bfloat> %216, i64 0, !dbg !23 + %224 = extractelement <2 x bfloat> %216, i64 1, !dbg !23 + %225 = extractelement <32 x i1> %39, i64 28, !dbg !23 + %226 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %142, i1 %225) #7, !dbg !23 + %227 = extractvalue { i32, i32, i32, i32 } %226, 0, !dbg !23 + %228 = bitcast i32 %227 to <2 x bfloat>, !dbg !23 + %229 = extractvalue { i32, i32, i32, i32 } %226, 1, !dbg !23 + %230 = bitcast i32 %229 to <2 x bfloat>, !dbg !23 + %231 = extractvalue { i32, i32, i32, i32 } %226, 2, !dbg !23 + %232 = bitcast i32 %231 to <2 x bfloat>, !dbg !23 + %233 = extractvalue { i32, i32, i32, i32 } %226, 3, !dbg !23 + %234 = bitcast i32 %233 to <2 x bfloat>, !dbg !23 + %235 = extractelement <2 x bfloat> %228, i64 0, !dbg !23 + %236 = extractelement <2 x bfloat> %228, i64 1, !dbg !23 + %237 = extractelement <2 x bfloat> %230, i64 0, !dbg !23 + %238 = extractelement <2 x bfloat> %230, i64 1, !dbg !23 + %239 = extractelement <2 x bfloat> %232, i64 0, !dbg !23 + %240 = extractelement <2 x bfloat> %232, i64 1, !dbg !23 + %241 = extractelement <2 x bfloat> %234, i64 0, !dbg !23 + %242 = extractelement <2 x bfloat> %234, i64 1, !dbg !23 + %243 = extractelement <32 x i1> %39, i64 27, !dbg !23 + %244 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %143, i1 %243) #7, !dbg !23 + %245 = extractvalue { i32, i32, i32, i32 } %244, 0, !dbg !23 + %246 = bitcast i32 %245 to <2 x bfloat>, !dbg !23 + %247 = extractvalue { i32, i32, i32, i32 } %244, 1, !dbg !23 + %248 = bitcast i32 %247 to <2 x bfloat>, !dbg !23 + %249 = extractvalue { i32, i32, i32, i32 } %244, 2, !dbg !23 + %250 = bitcast i32 %249 to <2 x bfloat>, !dbg !23 + %251 = extractvalue { i32, i32, i32, i32 } %244, 3, !dbg !23 + %252 = bitcast i32 %251 to <2 x bfloat>, !dbg !23 + %253 = extractelement <2 x bfloat> %246, i64 0, !dbg !23 + %254 = extractelement <2 x bfloat> %246, i64 1, !dbg !23 + %255 = extractelement <2 x bfloat> %248, i64 0, !dbg !23 + %256 = extractelement <2 x bfloat> %248, i64 1, !dbg !23 + %257 = extractelement <2 x bfloat> %250, i64 0, !dbg !23 + %258 = extractelement <2 x bfloat> %250, i64 1, !dbg !23 + %259 = extractelement <2 x bfloat> %252, i64 0, !dbg !23 + %260 = extractelement <2 x bfloat> %252, i64 1, !dbg !23 + %261 = extractelement <32 x i1> %39, i64 26, !dbg !23 + %262 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %144, i1 %261) #7, !dbg !23 + %263 = extractvalue { i32, i32, i32, i32 } %262, 0, !dbg !23 + %264 = bitcast i32 %263 to <2 x bfloat>, !dbg !23 + %265 = extractvalue { i32, i32, i32, i32 } %262, 1, !dbg !23 + %266 = bitcast i32 %265 to <2 x bfloat>, !dbg !23 + %267 = extractvalue { i32, i32, i32, i32 } %262, 2, !dbg !23 + %268 = bitcast i32 %267 to <2 x bfloat>, !dbg !23 + %269 = extractvalue { i32, i32, i32, i32 } %262, 3, !dbg !23 + %270 = bitcast i32 %269 to <2 x bfloat>, !dbg !23 + %271 = extractelement <2 x bfloat> %264, i64 0, !dbg !23 + %272 = extractelement <2 x bfloat> %264, i64 1, !dbg !23 + %273 = extractelement <2 x bfloat> %266, i64 0, !dbg !23 + %274 = extractelement <2 x bfloat> %266, i64 1, !dbg !23 + %275 = extractelement <2 x bfloat> %268, i64 0, !dbg !23 + %276 = extractelement <2 x bfloat> %268, i64 1, !dbg !23 + %277 = extractelement <2 x bfloat> %270, i64 0, !dbg !23 + %278 = extractelement <2 x bfloat> %270, i64 1, !dbg !23 + %279 = extractelement <32 x i1> %39, i64 25, !dbg !23 + %280 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %145, i1 %279) #7, !dbg !23 + %281 = extractvalue { i32, i32, i32, i32 } %280, 0, !dbg !23 + %282 = bitcast i32 %281 to <2 x bfloat>, !dbg !23 + %283 = extractvalue { i32, i32, i32, i32 } %280, 1, !dbg !23 + %284 = bitcast i32 %283 to <2 x bfloat>, !dbg !23 + %285 = extractvalue { i32, i32, i32, i32 } %280, 2, !dbg !23 + %286 = bitcast i32 %285 to <2 x bfloat>, !dbg !23 + %287 = extractvalue { i32, i32, i32, i32 } %280, 3, !dbg !23 + %288 = bitcast i32 %287 to <2 x bfloat>, !dbg !23 + %289 = extractelement <2 x bfloat> %282, i64 0, !dbg !23 + %290 = extractelement <2 x bfloat> %282, i64 1, !dbg !23 + %291 = extractelement <2 x bfloat> %284, i64 0, !dbg !23 + %292 = extractelement <2 x bfloat> %284, i64 1, !dbg !23 + %293 = extractelement <2 x bfloat> %286, i64 0, !dbg !23 + %294 = extractelement <2 x bfloat> %286, i64 1, !dbg !23 + %295 = extractelement <2 x bfloat> %288, i64 0, !dbg !23 + %296 = extractelement <2 x bfloat> %288, i64 1, !dbg !23 + %297 = extractelement <32 x i1> %39, i64 24, !dbg !23 + %298 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %146, i1 %297) #7, !dbg !23 + %299 = extractvalue { i32, i32, i32, i32 } %298, 0, !dbg !23 + %300 = bitcast i32 %299 to <2 x bfloat>, !dbg !23 + %301 = extractvalue { i32, i32, i32, i32 } %298, 1, !dbg !23 + %302 = bitcast i32 %301 to <2 x bfloat>, !dbg !23 + %303 = extractvalue { i32, i32, i32, i32 } %298, 2, !dbg !23 + %304 = bitcast i32 %303 to <2 x bfloat>, !dbg !23 + %305 = extractvalue { i32, i32, i32, i32 } %298, 3, !dbg !23 + %306 = bitcast i32 %305 to <2 x bfloat>, !dbg !23 + %307 = extractelement <2 x bfloat> %300, i64 0, !dbg !23 + %308 = extractelement <2 x bfloat> %300, i64 1, !dbg !23 + %309 = extractelement <2 x bfloat> %302, i64 0, !dbg !23 + %310 = extractelement <2 x bfloat> %302, i64 1, !dbg !23 + %311 = extractelement <2 x bfloat> %304, i64 0, !dbg !23 + %312 = extractelement <2 x bfloat> %304, i64 1, !dbg !23 + %313 = extractelement <2 x bfloat> %306, i64 0, !dbg !23 + %314 = extractelement <2 x bfloat> %306, i64 1, !dbg !23 + %315 = extractelement <32 x i1> %39, i64 23, !dbg !23 + %316 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %147, i1 %315) #7, !dbg !23 + %317 = extractvalue { i32, i32, i32, i32 } %316, 0, !dbg !23 + %318 = bitcast i32 %317 to <2 x bfloat>, !dbg !23 + %319 = extractvalue { i32, i32, i32, i32 } %316, 1, !dbg !23 + %320 = bitcast i32 %319 to <2 x bfloat>, !dbg !23 + %321 = extractvalue { i32, i32, i32, i32 } %316, 2, !dbg !23 + %322 = bitcast i32 %321 to <2 x bfloat>, !dbg !23 + %323 = extractvalue { i32, i32, i32, i32 } %316, 3, !dbg !23 + %324 = bitcast i32 %323 to <2 x bfloat>, !dbg !23 + %325 = extractelement <2 x bfloat> %318, i64 0, !dbg !23 + %326 = extractelement <2 x bfloat> %318, i64 1, !dbg !23 + %327 = extractelement <2 x bfloat> %320, i64 0, !dbg !23 + %328 = extractelement <2 x bfloat> %320, i64 1, !dbg !23 + %329 = extractelement <2 x bfloat> %322, i64 0, !dbg !23 + %330 = extractelement <2 x bfloat> %322, i64 1, !dbg !23 + %331 = extractelement <2 x bfloat> %324, i64 0, !dbg !23 + %332 = extractelement <2 x bfloat> %324, i64 1, !dbg !23 + %333 = extractelement <32 x i1> %39, i64 22, !dbg !23 + %334 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %148, i1 %333) #7, !dbg !23 + %335 = extractvalue { i32, i32, i32, i32 } %334, 0, !dbg !23 + %336 = bitcast i32 %335 to <2 x bfloat>, !dbg !23 + %337 = extractvalue { i32, i32, i32, i32 } %334, 1, !dbg !23 + %338 = bitcast i32 %337 to <2 x bfloat>, !dbg !23 + %339 = extractvalue { i32, i32, i32, i32 } %334, 2, !dbg !23 + %340 = bitcast i32 %339 to <2 x bfloat>, !dbg !23 + %341 = extractvalue { i32, i32, i32, i32 } %334, 3, !dbg !23 + %342 = bitcast i32 %341 to <2 x bfloat>, !dbg !23 + %343 = extractelement <2 x bfloat> %336, i64 0, !dbg !23 + %344 = extractelement <2 x bfloat> %336, i64 1, !dbg !23 + %345 = extractelement <2 x bfloat> %338, i64 0, !dbg !23 + %346 = extractelement <2 x bfloat> %338, i64 1, !dbg !23 + %347 = extractelement <2 x bfloat> %340, i64 0, !dbg !23 + %348 = extractelement <2 x bfloat> %340, i64 1, !dbg !23 + %349 = extractelement <2 x bfloat> %342, i64 0, !dbg !23 + %350 = extractelement <2 x bfloat> %342, i64 1, !dbg !23 + %351 = extractelement <32 x i1> %39, i64 21, !dbg !23 + %352 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %149, i1 %351) #7, !dbg !23 + %353 = extractvalue { i32, i32, i32, i32 } %352, 0, !dbg !23 + %354 = bitcast i32 %353 to <2 x bfloat>, !dbg !23 + %355 = extractvalue { i32, i32, i32, i32 } %352, 1, !dbg !23 + %356 = bitcast i32 %355 to <2 x bfloat>, !dbg !23 + %357 = extractvalue { i32, i32, i32, i32 } %352, 2, !dbg !23 + %358 = bitcast i32 %357 to <2 x bfloat>, !dbg !23 + %359 = extractvalue { i32, i32, i32, i32 } %352, 3, !dbg !23 + %360 = bitcast i32 %359 to <2 x bfloat>, !dbg !23 + %361 = extractelement <2 x bfloat> %354, i64 0, !dbg !23 + %362 = extractelement <2 x bfloat> %354, i64 1, !dbg !23 + %363 = extractelement <2 x bfloat> %356, i64 0, !dbg !23 + %364 = extractelement <2 x bfloat> %356, i64 1, !dbg !23 + %365 = extractelement <2 x bfloat> %358, i64 0, !dbg !23 + %366 = extractelement <2 x bfloat> %358, i64 1, !dbg !23 + %367 = extractelement <2 x bfloat> %360, i64 0, !dbg !23 + %368 = extractelement <2 x bfloat> %360, i64 1, !dbg !23 + %369 = extractelement <32 x i1> %39, i64 20, !dbg !23 + %370 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %150, i1 %369) #7, !dbg !23 + %371 = extractvalue { i32, i32, i32, i32 } %370, 0, !dbg !23 + %372 = bitcast i32 %371 to <2 x bfloat>, !dbg !23 + %373 = extractvalue { i32, i32, i32, i32 } %370, 1, !dbg !23 + %374 = bitcast i32 %373 to <2 x bfloat>, !dbg !23 + %375 = extractvalue { i32, i32, i32, i32 } %370, 2, !dbg !23 + %376 = bitcast i32 %375 to <2 x bfloat>, !dbg !23 + %377 = extractvalue { i32, i32, i32, i32 } %370, 3, !dbg !23 + %378 = bitcast i32 %377 to <2 x bfloat>, !dbg !23 + %379 = extractelement <2 x bfloat> %372, i64 0, !dbg !23 + %380 = extractelement <2 x bfloat> %372, i64 1, !dbg !23 + %381 = extractelement <2 x bfloat> %374, i64 0, !dbg !23 + %382 = extractelement <2 x bfloat> %374, i64 1, !dbg !23 + %383 = extractelement <2 x bfloat> %376, i64 0, !dbg !23 + %384 = extractelement <2 x bfloat> %376, i64 1, !dbg !23 + %385 = extractelement <2 x bfloat> %378, i64 0, !dbg !23 + %386 = extractelement <2 x bfloat> %378, i64 1, !dbg !23 + %387 = extractelement <32 x i1> %39, i64 19, !dbg !23 + %388 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %151, i1 %387) #7, !dbg !23 + %389 = extractvalue { i32, i32, i32, i32 } %388, 0, !dbg !23 + %390 = bitcast i32 %389 to <2 x bfloat>, !dbg !23 + %391 = extractvalue { i32, i32, i32, i32 } %388, 1, !dbg !23 + %392 = bitcast i32 %391 to <2 x bfloat>, !dbg !23 + %393 = extractvalue { i32, i32, i32, i32 } %388, 2, !dbg !23 + %394 = bitcast i32 %393 to <2 x bfloat>, !dbg !23 + %395 = extractvalue { i32, i32, i32, i32 } %388, 3, !dbg !23 + %396 = bitcast i32 %395 to <2 x bfloat>, !dbg !23 + %397 = extractelement <2 x bfloat> %390, i64 0, !dbg !23 + %398 = extractelement <2 x bfloat> %390, i64 1, !dbg !23 + %399 = extractelement <2 x bfloat> %392, i64 0, !dbg !23 + %400 = extractelement <2 x bfloat> %392, i64 1, !dbg !23 + %401 = extractelement <2 x bfloat> %394, i64 0, !dbg !23 + %402 = extractelement <2 x bfloat> %394, i64 1, !dbg !23 + %403 = extractelement <2 x bfloat> %396, i64 0, !dbg !23 + %404 = extractelement <2 x bfloat> %396, i64 1, !dbg !23 + %405 = extractelement <32 x i1> %39, i64 18, !dbg !23 + %406 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %152, i1 %405) #7, !dbg !23 + %407 = extractvalue { i32, i32, i32, i32 } %406, 0, !dbg !23 + %408 = bitcast i32 %407 to <2 x bfloat>, !dbg !23 + %409 = extractvalue { i32, i32, i32, i32 } %406, 1, !dbg !23 + %410 = bitcast i32 %409 to <2 x bfloat>, !dbg !23 + %411 = extractvalue { i32, i32, i32, i32 } %406, 2, !dbg !23 + %412 = bitcast i32 %411 to <2 x bfloat>, !dbg !23 + %413 = extractvalue { i32, i32, i32, i32 } %406, 3, !dbg !23 + %414 = bitcast i32 %413 to <2 x bfloat>, !dbg !23 + %415 = extractelement <2 x bfloat> %408, i64 0, !dbg !23 + %416 = extractelement <2 x bfloat> %408, i64 1, !dbg !23 + %417 = extractelement <2 x bfloat> %410, i64 0, !dbg !23 + %418 = extractelement <2 x bfloat> %410, i64 1, !dbg !23 + %419 = extractelement <2 x bfloat> %412, i64 0, !dbg !23 + %420 = extractelement <2 x bfloat> %412, i64 1, !dbg !23 + %421 = extractelement <2 x bfloat> %414, i64 0, !dbg !23 + %422 = extractelement <2 x bfloat> %414, i64 1, !dbg !23 + %423 = extractelement <32 x i1> %39, i64 17, !dbg !23 + %424 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %153, i1 %423) #7, !dbg !23 + %425 = extractvalue { i32, i32, i32, i32 } %424, 0, !dbg !23 + %426 = bitcast i32 %425 to <2 x bfloat>, !dbg !23 + %427 = extractvalue { i32, i32, i32, i32 } %424, 1, !dbg !23 + %428 = bitcast i32 %427 to <2 x bfloat>, !dbg !23 + %429 = extractvalue { i32, i32, i32, i32 } %424, 2, !dbg !23 + %430 = bitcast i32 %429 to <2 x bfloat>, !dbg !23 + %431 = extractvalue { i32, i32, i32, i32 } %424, 3, !dbg !23 + %432 = bitcast i32 %431 to <2 x bfloat>, !dbg !23 + %433 = extractelement <2 x bfloat> %426, i64 0, !dbg !23 + %434 = extractelement <2 x bfloat> %426, i64 1, !dbg !23 + %435 = extractelement <2 x bfloat> %428, i64 0, !dbg !23 + %436 = extractelement <2 x bfloat> %428, i64 1, !dbg !23 + %437 = extractelement <2 x bfloat> %430, i64 0, !dbg !23 + %438 = extractelement <2 x bfloat> %430, i64 1, !dbg !23 + %439 = extractelement <2 x bfloat> %432, i64 0, !dbg !23 + %440 = extractelement <2 x bfloat> %432, i64 1, !dbg !23 + %441 = extractelement <32 x i1> %39, i64 16, !dbg !23 + %442 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %154, i1 %441) #7, !dbg !23 + %443 = extractvalue { i32, i32, i32, i32 } %442, 0, !dbg !23 + %444 = bitcast i32 %443 to <2 x bfloat>, !dbg !23 + %445 = extractvalue { i32, i32, i32, i32 } %442, 1, !dbg !23 + %446 = bitcast i32 %445 to <2 x bfloat>, !dbg !23 + %447 = extractvalue { i32, i32, i32, i32 } %442, 2, !dbg !23 + %448 = bitcast i32 %447 to <2 x bfloat>, !dbg !23 + %449 = extractvalue { i32, i32, i32, i32 } %442, 3, !dbg !23 + %450 = bitcast i32 %449 to <2 x bfloat>, !dbg !23 + %451 = extractelement <2 x bfloat> %444, i64 0, !dbg !23 + %452 = extractelement <2 x bfloat> %444, i64 1, !dbg !23 + %453 = extractelement <2 x bfloat> %446, i64 0, !dbg !23 + %454 = extractelement <2 x bfloat> %446, i64 1, !dbg !23 + %455 = extractelement <2 x bfloat> %448, i64 0, !dbg !23 + %456 = extractelement <2 x bfloat> %448, i64 1, !dbg !23 + %457 = extractelement <2 x bfloat> %450, i64 0, !dbg !23 + %458 = extractelement <2 x bfloat> %450, i64 1, !dbg !23 + %459 = extractelement <32 x i1> %39, i64 15, !dbg !23 + %460 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %155, i1 %459) #7, !dbg !23 + %461 = extractvalue { i32, i32, i32, i32 } %460, 0, !dbg !23 + %462 = bitcast i32 %461 to <2 x bfloat>, !dbg !23 + %463 = extractvalue { i32, i32, i32, i32 } %460, 1, !dbg !23 + %464 = bitcast i32 %463 to <2 x bfloat>, !dbg !23 + %465 = extractvalue { i32, i32, i32, i32 } %460, 2, !dbg !23 + %466 = bitcast i32 %465 to <2 x bfloat>, !dbg !23 + %467 = extractvalue { i32, i32, i32, i32 } %460, 3, !dbg !23 + %468 = bitcast i32 %467 to <2 x bfloat>, !dbg !23 + %469 = extractelement <2 x bfloat> %462, i64 0, !dbg !23 + %470 = extractelement <2 x bfloat> %462, i64 1, !dbg !23 + %471 = extractelement <2 x bfloat> %464, i64 0, !dbg !23 + %472 = extractelement <2 x bfloat> %464, i64 1, !dbg !23 + %473 = extractelement <2 x bfloat> %466, i64 0, !dbg !23 + %474 = extractelement <2 x bfloat> %466, i64 1, !dbg !23 + %475 = extractelement <2 x bfloat> %468, i64 0, !dbg !23 + %476 = extractelement <2 x bfloat> %468, i64 1, !dbg !23 + %477 = extractelement <32 x i1> %39, i64 14, !dbg !23 + %478 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %156, i1 %477) #7, !dbg !23 + %479 = extractvalue { i32, i32, i32, i32 } %478, 0, !dbg !23 + %480 = bitcast i32 %479 to <2 x bfloat>, !dbg !23 + %481 = extractvalue { i32, i32, i32, i32 } %478, 1, !dbg !23 + %482 = bitcast i32 %481 to <2 x bfloat>, !dbg !23 + %483 = extractvalue { i32, i32, i32, i32 } %478, 2, !dbg !23 + %484 = bitcast i32 %483 to <2 x bfloat>, !dbg !23 + %485 = extractvalue { i32, i32, i32, i32 } %478, 3, !dbg !23 + %486 = bitcast i32 %485 to <2 x bfloat>, !dbg !23 + %487 = extractelement <2 x bfloat> %480, i64 0, !dbg !23 + %488 = extractelement <2 x bfloat> %480, i64 1, !dbg !23 + %489 = extractelement <2 x bfloat> %482, i64 0, !dbg !23 + %490 = extractelement <2 x bfloat> %482, i64 1, !dbg !23 + %491 = extractelement <2 x bfloat> %484, i64 0, !dbg !23 + %492 = extractelement <2 x bfloat> %484, i64 1, !dbg !23 + %493 = extractelement <2 x bfloat> %486, i64 0, !dbg !23 + %494 = extractelement <2 x bfloat> %486, i64 1, !dbg !23 + %495 = extractelement <32 x i1> %39, i64 13, !dbg !23 + %496 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %157, i1 %495) #7, !dbg !23 + %497 = extractvalue { i32, i32, i32, i32 } %496, 0, !dbg !23 + %498 = bitcast i32 %497 to <2 x bfloat>, !dbg !23 + %499 = extractvalue { i32, i32, i32, i32 } %496, 1, !dbg !23 + %500 = bitcast i32 %499 to <2 x bfloat>, !dbg !23 + %501 = extractvalue { i32, i32, i32, i32 } %496, 2, !dbg !23 + %502 = bitcast i32 %501 to <2 x bfloat>, !dbg !23 + %503 = extractvalue { i32, i32, i32, i32 } %496, 3, !dbg !23 + %504 = bitcast i32 %503 to <2 x bfloat>, !dbg !23 + %505 = extractelement <2 x bfloat> %498, i64 0, !dbg !23 + %506 = extractelement <2 x bfloat> %498, i64 1, !dbg !23 + %507 = extractelement <2 x bfloat> %500, i64 0, !dbg !23 + %508 = extractelement <2 x bfloat> %500, i64 1, !dbg !23 + %509 = extractelement <2 x bfloat> %502, i64 0, !dbg !23 + %510 = extractelement <2 x bfloat> %502, i64 1, !dbg !23 + %511 = extractelement <2 x bfloat> %504, i64 0, !dbg !23 + %512 = extractelement <2 x bfloat> %504, i64 1, !dbg !23 + %513 = extractelement <32 x i1> %39, i64 12, !dbg !23 + %514 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %158, i1 %513) #7, !dbg !23 + %515 = extractvalue { i32, i32, i32, i32 } %514, 0, !dbg !23 + %516 = bitcast i32 %515 to <2 x bfloat>, !dbg !23 + %517 = extractvalue { i32, i32, i32, i32 } %514, 1, !dbg !23 + %518 = bitcast i32 %517 to <2 x bfloat>, !dbg !23 + %519 = extractvalue { i32, i32, i32, i32 } %514, 2, !dbg !23 + %520 = bitcast i32 %519 to <2 x bfloat>, !dbg !23 + %521 = extractvalue { i32, i32, i32, i32 } %514, 3, !dbg !23 + %522 = bitcast i32 %521 to <2 x bfloat>, !dbg !23 + %523 = extractelement <2 x bfloat> %516, i64 0, !dbg !23 + %524 = extractelement <2 x bfloat> %516, i64 1, !dbg !23 + %525 = extractelement <2 x bfloat> %518, i64 0, !dbg !23 + %526 = extractelement <2 x bfloat> %518, i64 1, !dbg !23 + %527 = extractelement <2 x bfloat> %520, i64 0, !dbg !23 + %528 = extractelement <2 x bfloat> %520, i64 1, !dbg !23 + %529 = extractelement <2 x bfloat> %522, i64 0, !dbg !23 + %530 = extractelement <2 x bfloat> %522, i64 1, !dbg !23 + %531 = extractelement <32 x i1> %39, i64 11, !dbg !23 + %532 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %159, i1 %531) #7, !dbg !23 + %533 = extractvalue { i32, i32, i32, i32 } %532, 0, !dbg !23 + %534 = bitcast i32 %533 to <2 x bfloat>, !dbg !23 + %535 = extractvalue { i32, i32, i32, i32 } %532, 1, !dbg !23 + %536 = bitcast i32 %535 to <2 x bfloat>, !dbg !23 + %537 = extractvalue { i32, i32, i32, i32 } %532, 2, !dbg !23 + %538 = bitcast i32 %537 to <2 x bfloat>, !dbg !23 + %539 = extractvalue { i32, i32, i32, i32 } %532, 3, !dbg !23 + %540 = bitcast i32 %539 to <2 x bfloat>, !dbg !23 + %541 = extractelement <2 x bfloat> %534, i64 0, !dbg !23 + %542 = extractelement <2 x bfloat> %534, i64 1, !dbg !23 + %543 = extractelement <2 x bfloat> %536, i64 0, !dbg !23 + %544 = extractelement <2 x bfloat> %536, i64 1, !dbg !23 + %545 = extractelement <2 x bfloat> %538, i64 0, !dbg !23 + %546 = extractelement <2 x bfloat> %538, i64 1, !dbg !23 + %547 = extractelement <2 x bfloat> %540, i64 0, !dbg !23 + %548 = extractelement <2 x bfloat> %540, i64 1, !dbg !23 + %549 = extractelement <32 x i1> %39, i64 10, !dbg !23 + %550 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %160, i1 %549) #7, !dbg !23 + %551 = extractvalue { i32, i32, i32, i32 } %550, 0, !dbg !23 + %552 = bitcast i32 %551 to <2 x bfloat>, !dbg !23 + %553 = extractvalue { i32, i32, i32, i32 } %550, 1, !dbg !23 + %554 = bitcast i32 %553 to <2 x bfloat>, !dbg !23 + %555 = extractvalue { i32, i32, i32, i32 } %550, 2, !dbg !23 + %556 = bitcast i32 %555 to <2 x bfloat>, !dbg !23 + %557 = extractvalue { i32, i32, i32, i32 } %550, 3, !dbg !23 + %558 = bitcast i32 %557 to <2 x bfloat>, !dbg !23 + %559 = extractelement <2 x bfloat> %552, i64 0, !dbg !23 + %560 = extractelement <2 x bfloat> %552, i64 1, !dbg !23 + %561 = extractelement <2 x bfloat> %554, i64 0, !dbg !23 + %562 = extractelement <2 x bfloat> %554, i64 1, !dbg !23 + %563 = extractelement <2 x bfloat> %556, i64 0, !dbg !23 + %564 = extractelement <2 x bfloat> %556, i64 1, !dbg !23 + %565 = extractelement <2 x bfloat> %558, i64 0, !dbg !23 + %566 = extractelement <2 x bfloat> %558, i64 1, !dbg !23 + %567 = extractelement <32 x i1> %39, i64 9, !dbg !23 + %568 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %161, i1 %567) #7, !dbg !23 + %569 = extractvalue { i32, i32, i32, i32 } %568, 0, !dbg !23 + %570 = bitcast i32 %569 to <2 x bfloat>, !dbg !23 + %571 = extractvalue { i32, i32, i32, i32 } %568, 1, !dbg !23 + %572 = bitcast i32 %571 to <2 x bfloat>, !dbg !23 + %573 = extractvalue { i32, i32, i32, i32 } %568, 2, !dbg !23 + %574 = bitcast i32 %573 to <2 x bfloat>, !dbg !23 + %575 = extractvalue { i32, i32, i32, i32 } %568, 3, !dbg !23 + %576 = bitcast i32 %575 to <2 x bfloat>, !dbg !23 + %577 = extractelement <2 x bfloat> %570, i64 0, !dbg !23 + %578 = extractelement <2 x bfloat> %570, i64 1, !dbg !23 + %579 = extractelement <2 x bfloat> %572, i64 0, !dbg !23 + %580 = extractelement <2 x bfloat> %572, i64 1, !dbg !23 + %581 = extractelement <2 x bfloat> %574, i64 0, !dbg !23 + %582 = extractelement <2 x bfloat> %574, i64 1, !dbg !23 + %583 = extractelement <2 x bfloat> %576, i64 0, !dbg !23 + %584 = extractelement <2 x bfloat> %576, i64 1, !dbg !23 + %585 = extractelement <32 x i1> %39, i64 8, !dbg !23 + %586 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %162, i1 %585) #7, !dbg !23 + %587 = extractvalue { i32, i32, i32, i32 } %586, 0, !dbg !23 + %588 = bitcast i32 %587 to <2 x bfloat>, !dbg !23 + %589 = extractvalue { i32, i32, i32, i32 } %586, 1, !dbg !23 + %590 = bitcast i32 %589 to <2 x bfloat>, !dbg !23 + %591 = extractvalue { i32, i32, i32, i32 } %586, 2, !dbg !23 + %592 = bitcast i32 %591 to <2 x bfloat>, !dbg !23 + %593 = extractvalue { i32, i32, i32, i32 } %586, 3, !dbg !23 + %594 = bitcast i32 %593 to <2 x bfloat>, !dbg !23 + %595 = extractelement <2 x bfloat> %588, i64 0, !dbg !23 + %596 = extractelement <2 x bfloat> %588, i64 1, !dbg !23 + %597 = extractelement <2 x bfloat> %590, i64 0, !dbg !23 + %598 = extractelement <2 x bfloat> %590, i64 1, !dbg !23 + %599 = extractelement <2 x bfloat> %592, i64 0, !dbg !23 + %600 = extractelement <2 x bfloat> %592, i64 1, !dbg !23 + %601 = extractelement <2 x bfloat> %594, i64 0, !dbg !23 + %602 = extractelement <2 x bfloat> %594, i64 1, !dbg !23 + %603 = extractelement <32 x i1> %39, i64 7, !dbg !23 + %604 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %163, i1 %603) #7, !dbg !23 + %605 = extractvalue { i32, i32, i32, i32 } %604, 0, !dbg !23 + %606 = bitcast i32 %605 to <2 x bfloat>, !dbg !23 + %607 = extractvalue { i32, i32, i32, i32 } %604, 1, !dbg !23 + %608 = bitcast i32 %607 to <2 x bfloat>, !dbg !23 + %609 = extractvalue { i32, i32, i32, i32 } %604, 2, !dbg !23 + %610 = bitcast i32 %609 to <2 x bfloat>, !dbg !23 + %611 = extractvalue { i32, i32, i32, i32 } %604, 3, !dbg !23 + %612 = bitcast i32 %611 to <2 x bfloat>, !dbg !23 + %613 = extractelement <2 x bfloat> %606, i64 0, !dbg !23 + %614 = extractelement <2 x bfloat> %606, i64 1, !dbg !23 + %615 = extractelement <2 x bfloat> %608, i64 0, !dbg !23 + %616 = extractelement <2 x bfloat> %608, i64 1, !dbg !23 + %617 = extractelement <2 x bfloat> %610, i64 0, !dbg !23 + %618 = extractelement <2 x bfloat> %610, i64 1, !dbg !23 + %619 = extractelement <2 x bfloat> %612, i64 0, !dbg !23 + %620 = extractelement <2 x bfloat> %612, i64 1, !dbg !23 + %621 = extractelement <32 x i1> %39, i64 6, !dbg !23 + %622 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %164, i1 %621) #7, !dbg !23 + %623 = extractvalue { i32, i32, i32, i32 } %622, 0, !dbg !23 + %624 = bitcast i32 %623 to <2 x bfloat>, !dbg !23 + %625 = extractvalue { i32, i32, i32, i32 } %622, 1, !dbg !23 + %626 = bitcast i32 %625 to <2 x bfloat>, !dbg !23 + %627 = extractvalue { i32, i32, i32, i32 } %622, 2, !dbg !23 + %628 = bitcast i32 %627 to <2 x bfloat>, !dbg !23 + %629 = extractvalue { i32, i32, i32, i32 } %622, 3, !dbg !23 + %630 = bitcast i32 %629 to <2 x bfloat>, !dbg !23 + %631 = extractelement <2 x bfloat> %624, i64 0, !dbg !23 + %632 = extractelement <2 x bfloat> %624, i64 1, !dbg !23 + %633 = extractelement <2 x bfloat> %626, i64 0, !dbg !23 + %634 = extractelement <2 x bfloat> %626, i64 1, !dbg !23 + %635 = extractelement <2 x bfloat> %628, i64 0, !dbg !23 + %636 = extractelement <2 x bfloat> %628, i64 1, !dbg !23 + %637 = extractelement <2 x bfloat> %630, i64 0, !dbg !23 + %638 = extractelement <2 x bfloat> %630, i64 1, !dbg !23 + %639 = extractelement <32 x i1> %39, i64 5, !dbg !23 + %640 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %165, i1 %639) #7, !dbg !23 + %641 = extractvalue { i32, i32, i32, i32 } %640, 0, !dbg !23 + %642 = bitcast i32 %641 to <2 x bfloat>, !dbg !23 + %643 = extractvalue { i32, i32, i32, i32 } %640, 1, !dbg !23 + %644 = bitcast i32 %643 to <2 x bfloat>, !dbg !23 + %645 = extractvalue { i32, i32, i32, i32 } %640, 2, !dbg !23 + %646 = bitcast i32 %645 to <2 x bfloat>, !dbg !23 + %647 = extractvalue { i32, i32, i32, i32 } %640, 3, !dbg !23 + %648 = bitcast i32 %647 to <2 x bfloat>, !dbg !23 + %649 = extractelement <2 x bfloat> %642, i64 0, !dbg !23 + %650 = extractelement <2 x bfloat> %642, i64 1, !dbg !23 + %651 = extractelement <2 x bfloat> %644, i64 0, !dbg !23 + %652 = extractelement <2 x bfloat> %644, i64 1, !dbg !23 + %653 = extractelement <2 x bfloat> %646, i64 0, !dbg !23 + %654 = extractelement <2 x bfloat> %646, i64 1, !dbg !23 + %655 = extractelement <2 x bfloat> %648, i64 0, !dbg !23 + %656 = extractelement <2 x bfloat> %648, i64 1, !dbg !23 + %657 = extractelement <32 x i1> %39, i64 4, !dbg !23 + %658 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %166, i1 %657) #7, !dbg !23 + %659 = extractvalue { i32, i32, i32, i32 } %658, 0, !dbg !23 + %660 = bitcast i32 %659 to <2 x bfloat>, !dbg !23 + %661 = extractvalue { i32, i32, i32, i32 } %658, 1, !dbg !23 + %662 = bitcast i32 %661 to <2 x bfloat>, !dbg !23 + %663 = extractvalue { i32, i32, i32, i32 } %658, 2, !dbg !23 + %664 = bitcast i32 %663 to <2 x bfloat>, !dbg !23 + %665 = extractvalue { i32, i32, i32, i32 } %658, 3, !dbg !23 + %666 = bitcast i32 %665 to <2 x bfloat>, !dbg !23 + %667 = extractelement <2 x bfloat> %660, i64 0, !dbg !23 + %668 = extractelement <2 x bfloat> %660, i64 1, !dbg !23 + %669 = extractelement <2 x bfloat> %662, i64 0, !dbg !23 + %670 = extractelement <2 x bfloat> %662, i64 1, !dbg !23 + %671 = extractelement <2 x bfloat> %664, i64 0, !dbg !23 + %672 = extractelement <2 x bfloat> %664, i64 1, !dbg !23 + %673 = extractelement <2 x bfloat> %666, i64 0, !dbg !23 + %674 = extractelement <2 x bfloat> %666, i64 1, !dbg !23 + %675 = extractelement <32 x i1> %39, i64 3, !dbg !23 + %676 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %167, i1 %675) #7, !dbg !23 + %677 = extractvalue { i32, i32, i32, i32 } %676, 0, !dbg !23 + %678 = bitcast i32 %677 to <2 x bfloat>, !dbg !23 + %679 = extractvalue { i32, i32, i32, i32 } %676, 1, !dbg !23 + %680 = bitcast i32 %679 to <2 x bfloat>, !dbg !23 + %681 = extractvalue { i32, i32, i32, i32 } %676, 2, !dbg !23 + %682 = bitcast i32 %681 to <2 x bfloat>, !dbg !23 + %683 = extractvalue { i32, i32, i32, i32 } %676, 3, !dbg !23 + %684 = bitcast i32 %683 to <2 x bfloat>, !dbg !23 + %685 = extractelement <2 x bfloat> %678, i64 0, !dbg !23 + %686 = extractelement <2 x bfloat> %678, i64 1, !dbg !23 + %687 = extractelement <2 x bfloat> %680, i64 0, !dbg !23 + %688 = extractelement <2 x bfloat> %680, i64 1, !dbg !23 + %689 = extractelement <2 x bfloat> %682, i64 0, !dbg !23 + %690 = extractelement <2 x bfloat> %682, i64 1, !dbg !23 + %691 = extractelement <2 x bfloat> %684, i64 0, !dbg !23 + %692 = extractelement <2 x bfloat> %684, i64 1, !dbg !23 + %693 = extractelement <32 x i1> %39, i64 2, !dbg !23 + %694 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %168, i1 %693) #7, !dbg !23 + %695 = extractvalue { i32, i32, i32, i32 } %694, 0, !dbg !23 + %696 = bitcast i32 %695 to <2 x bfloat>, !dbg !23 + %697 = extractvalue { i32, i32, i32, i32 } %694, 1, !dbg !23 + %698 = bitcast i32 %697 to <2 x bfloat>, !dbg !23 + %699 = extractvalue { i32, i32, i32, i32 } %694, 2, !dbg !23 + %700 = bitcast i32 %699 to <2 x bfloat>, !dbg !23 + %701 = extractvalue { i32, i32, i32, i32 } %694, 3, !dbg !23 + %702 = bitcast i32 %701 to <2 x bfloat>, !dbg !23 + %703 = extractelement <2 x bfloat> %696, i64 0, !dbg !23 + %704 = extractelement <2 x bfloat> %696, i64 1, !dbg !23 + %705 = extractelement <2 x bfloat> %698, i64 0, !dbg !23 + %706 = extractelement <2 x bfloat> %698, i64 1, !dbg !23 + %707 = extractelement <2 x bfloat> %700, i64 0, !dbg !23 + %708 = extractelement <2 x bfloat> %700, i64 1, !dbg !23 + %709 = extractelement <2 x bfloat> %702, i64 0, !dbg !23 + %710 = extractelement <2 x bfloat> %702, i64 1, !dbg !23 + %711 = extractelement <32 x i1> %39, i64 1, !dbg !23 + %712 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %169, i1 %711) #7, !dbg !23 + %713 = extractvalue { i32, i32, i32, i32 } %712, 0, !dbg !23 + %714 = bitcast i32 %713 to <2 x bfloat>, !dbg !23 + %715 = extractvalue { i32, i32, i32, i32 } %712, 1, !dbg !23 + %716 = bitcast i32 %715 to <2 x bfloat>, !dbg !23 + %717 = extractvalue { i32, i32, i32, i32 } %712, 2, !dbg !23 + %718 = bitcast i32 %717 to <2 x bfloat>, !dbg !23 + %719 = extractvalue { i32, i32, i32, i32 } %712, 3, !dbg !23 + %720 = bitcast i32 %719 to <2 x bfloat>, !dbg !23 + %721 = extractelement <2 x bfloat> %714, i64 0, !dbg !23 + %722 = extractelement <2 x bfloat> %714, i64 1, !dbg !23 + %723 = extractelement <2 x bfloat> %716, i64 0, !dbg !23 + %724 = extractelement <2 x bfloat> %716, i64 1, !dbg !23 + %725 = extractelement <2 x bfloat> %718, i64 0, !dbg !23 + %726 = extractelement <2 x bfloat> %718, i64 1, !dbg !23 + %727 = extractelement <2 x bfloat> %720, i64 0, !dbg !23 + %728 = extractelement <2 x bfloat> %720, i64 1, !dbg !23 + %729 = extractelement <32 x i1> %39, i64 0, !dbg !23 + %730 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %170, i1 %729) #7, !dbg !23 + %731 = extractvalue { i32, i32, i32, i32 } %730, 0, !dbg !23 + %732 = bitcast i32 %731 to <2 x bfloat>, !dbg !23 + %733 = extractvalue { i32, i32, i32, i32 } %730, 1, !dbg !23 + %734 = bitcast i32 %733 to <2 x bfloat>, !dbg !23 + %735 = extractvalue { i32, i32, i32, i32 } %730, 2, !dbg !23 + %736 = bitcast i32 %735 to <2 x bfloat>, !dbg !23 + %737 = extractvalue { i32, i32, i32, i32 } %730, 3, !dbg !23 + %738 = bitcast i32 %737 to <2 x bfloat>, !dbg !23 + %739 = extractelement <2 x bfloat> %732, i64 0, !dbg !23 + %740 = extractelement <2 x bfloat> %732, i64 1, !dbg !23 + %741 = extractelement <2 x bfloat> %734, i64 0, !dbg !23 + %742 = extractelement <2 x bfloat> %734, i64 1, !dbg !23 + %743 = extractelement <2 x bfloat> %736, i64 0, !dbg !23 + %744 = extractelement <2 x bfloat> %736, i64 1, !dbg !23 + %745 = extractelement <2 x bfloat> %738, i64 0, !dbg !23 + %746 = extractelement <2 x bfloat> %738, i64 1, !dbg !23 + %747 = fpext bfloat %181 to float, !dbg !24 + %748 = fpext bfloat %182 to float, !dbg !24 + %749 = fpext bfloat %183 to float, !dbg !24 + %750 = fpext bfloat %184 to float, !dbg !24 + %751 = fpext bfloat %185 to float, !dbg !24 + %752 = fpext bfloat %186 to float, !dbg !24 + %753 = fpext bfloat %187 to float, !dbg !24 + %754 = fpext bfloat %188 to float, !dbg !24 + %755 = fpext bfloat %199 to float, !dbg !24 + %756 = fpext bfloat %200 to float, !dbg !24 + %757 = fpext bfloat %201 to float, !dbg !24 + %758 = fpext bfloat %202 to float, !dbg !24 + %759 = fpext bfloat %203 to float, !dbg !24 + %760 = fpext bfloat %204 to float, !dbg !24 + %761 = fpext bfloat %205 to float, !dbg !24 + %762 = fpext bfloat %206 to float, !dbg !24 + %763 = fpext bfloat %217 to float, !dbg !24 + %764 = fpext bfloat %218 to float, !dbg !24 + %765 = fpext bfloat %219 to float, !dbg !24 + %766 = fpext bfloat %220 to float, !dbg !24 + %767 = fpext bfloat %221 to float, !dbg !24 + %768 = fpext bfloat %222 to float, !dbg !24 + %769 = fpext bfloat %223 to float, !dbg !24 + %770 = fpext bfloat %224 to float, !dbg !24 + %771 = fpext bfloat %235 to float, !dbg !24 + %772 = fpext bfloat %236 to float, !dbg !24 + %773 = fpext bfloat %237 to float, !dbg !24 + %774 = fpext bfloat %238 to float, !dbg !24 + %775 = fpext bfloat %239 to float, !dbg !24 + %776 = fpext bfloat %240 to float, !dbg !24 + %777 = fpext bfloat %241 to float, !dbg !24 + %778 = fpext bfloat %242 to float, !dbg !24 + %779 = fpext bfloat %253 to float, !dbg !24 + %780 = fpext bfloat %254 to float, !dbg !24 + %781 = fpext bfloat %255 to float, !dbg !24 + %782 = fpext bfloat %256 to float, !dbg !24 + %783 = fpext bfloat %257 to float, !dbg !24 + %784 = fpext bfloat %258 to float, !dbg !24 + %785 = fpext bfloat %259 to float, !dbg !24 + %786 = fpext bfloat %260 to float, !dbg !24 + %787 = fpext bfloat %271 to float, !dbg !24 + %788 = fpext bfloat %272 to float, !dbg !24 + %789 = fpext bfloat %273 to float, !dbg !24 + %790 = fpext bfloat %274 to float, !dbg !24 + %791 = fpext bfloat %275 to float, !dbg !24 + %792 = fpext bfloat %276 to float, !dbg !24 + %793 = fpext bfloat %277 to float, !dbg !24 + %794 = fpext bfloat %278 to float, !dbg !24 + %795 = fpext bfloat %289 to float, !dbg !24 + %796 = fpext bfloat %290 to float, !dbg !24 + %797 = fpext bfloat %291 to float, !dbg !24 + %798 = fpext bfloat %292 to float, !dbg !24 + %799 = fpext bfloat %293 to float, !dbg !24 + %800 = fpext bfloat %294 to float, !dbg !24 + %801 = fpext bfloat %295 to float, !dbg !24 + %802 = fpext bfloat %296 to float, !dbg !24 + %803 = fpext bfloat %307 to float, !dbg !24 + %804 = fpext bfloat %308 to float, !dbg !24 + %805 = fpext bfloat %309 to float, !dbg !24 + %806 = fpext bfloat %310 to float, !dbg !24 + %807 = fpext bfloat %311 to float, !dbg !24 + %808 = fpext bfloat %312 to float, !dbg !24 + %809 = fpext bfloat %313 to float, !dbg !24 + %810 = fpext bfloat %314 to float, !dbg !24 + %811 = fpext bfloat %325 to float, !dbg !24 + %812 = fpext bfloat %326 to float, !dbg !24 + %813 = fpext bfloat %327 to float, !dbg !24 + %814 = fpext bfloat %328 to float, !dbg !24 + %815 = fpext bfloat %329 to float, !dbg !24 + %816 = fpext bfloat %330 to float, !dbg !24 + %817 = fpext bfloat %331 to float, !dbg !24 + %818 = fpext bfloat %332 to float, !dbg !24 + %819 = fpext bfloat %343 to float, !dbg !24 + %820 = fpext bfloat %344 to float, !dbg !24 + %821 = fpext bfloat %345 to float, !dbg !24 + %822 = fpext bfloat %346 to float, !dbg !24 + %823 = fpext bfloat %347 to float, !dbg !24 + %824 = fpext bfloat %348 to float, !dbg !24 + %825 = fpext bfloat %349 to float, !dbg !24 + %826 = fpext bfloat %350 to float, !dbg !24 + %827 = fpext bfloat %361 to float, !dbg !24 + %828 = fpext bfloat %362 to float, !dbg !24 + %829 = fpext bfloat %363 to float, !dbg !24 + %830 = fpext bfloat %364 to float, !dbg !24 + %831 = fpext bfloat %365 to float, !dbg !24 + %832 = fpext bfloat %366 to float, !dbg !24 + %833 = fpext bfloat %367 to float, !dbg !24 + %834 = fpext bfloat %368 to float, !dbg !24 + %835 = fpext bfloat %379 to float, !dbg !24 + %836 = fpext bfloat %380 to float, !dbg !24 + %837 = fpext bfloat %381 to float, !dbg !24 + %838 = fpext bfloat %382 to float, !dbg !24 + %839 = fpext bfloat %383 to float, !dbg !24 + %840 = fpext bfloat %384 to float, !dbg !24 + %841 = fpext bfloat %385 to float, !dbg !24 + %842 = fpext bfloat %386 to float, !dbg !24 + %843 = fpext bfloat %397 to float, !dbg !24 + %844 = fpext bfloat %398 to float, !dbg !24 + %845 = fpext bfloat %399 to float, !dbg !24 + %846 = fpext bfloat %400 to float, !dbg !24 + %847 = fpext bfloat %401 to float, !dbg !24 + %848 = fpext bfloat %402 to float, !dbg !24 + %849 = fpext bfloat %403 to float, !dbg !24 + %850 = fpext bfloat %404 to float, !dbg !24 + %851 = fpext bfloat %415 to float, !dbg !24 + %852 = fpext bfloat %416 to float, !dbg !24 + %853 = fpext bfloat %417 to float, !dbg !24 + %854 = fpext bfloat %418 to float, !dbg !24 + %855 = fpext bfloat %419 to float, !dbg !24 + %856 = fpext bfloat %420 to float, !dbg !24 + %857 = fpext bfloat %421 to float, !dbg !24 + %858 = fpext bfloat %422 to float, !dbg !24 + %859 = fpext bfloat %433 to float, !dbg !24 + %860 = fpext bfloat %434 to float, !dbg !24 + %861 = fpext bfloat %435 to float, !dbg !24 + %862 = fpext bfloat %436 to float, !dbg !24 + %863 = fpext bfloat %437 to float, !dbg !24 + %864 = fpext bfloat %438 to float, !dbg !24 + %865 = fpext bfloat %439 to float, !dbg !24 + %866 = fpext bfloat %440 to float, !dbg !24 + %867 = fpext bfloat %451 to float, !dbg !24 + %868 = fpext bfloat %452 to float, !dbg !24 + %869 = fpext bfloat %453 to float, !dbg !24 + %870 = fpext bfloat %454 to float, !dbg !24 + %871 = fpext bfloat %455 to float, !dbg !24 + %872 = fpext bfloat %456 to float, !dbg !24 + %873 = fpext bfloat %457 to float, !dbg !24 + %874 = fpext bfloat %458 to float, !dbg !24 + %875 = fpext bfloat %469 to float, !dbg !24 + %876 = fpext bfloat %470 to float, !dbg !24 + %877 = fpext bfloat %471 to float, !dbg !24 + %878 = fpext bfloat %472 to float, !dbg !24 + %879 = fpext bfloat %473 to float, !dbg !24 + %880 = fpext bfloat %474 to float, !dbg !24 + %881 = fpext bfloat %475 to float, !dbg !24 + %882 = fpext bfloat %476 to float, !dbg !24 + %883 = fpext bfloat %487 to float, !dbg !24 + %884 = fpext bfloat %488 to float, !dbg !24 + %885 = fpext bfloat %489 to float, !dbg !24 + %886 = fpext bfloat %490 to float, !dbg !24 + %887 = fpext bfloat %491 to float, !dbg !24 + %888 = fpext bfloat %492 to float, !dbg !24 + %889 = fpext bfloat %493 to float, !dbg !24 + %890 = fpext bfloat %494 to float, !dbg !24 + %891 = fpext bfloat %505 to float, !dbg !24 + %892 = fpext bfloat %506 to float, !dbg !24 + %893 = fpext bfloat %507 to float, !dbg !24 + %894 = fpext bfloat %508 to float, !dbg !24 + %895 = fpext bfloat %509 to float, !dbg !24 + %896 = fpext bfloat %510 to float, !dbg !24 + %897 = fpext bfloat %511 to float, !dbg !24 + %898 = fpext bfloat %512 to float, !dbg !24 + %899 = fpext bfloat %523 to float, !dbg !24 + %900 = fpext bfloat %524 to float, !dbg !24 + %901 = fpext bfloat %525 to float, !dbg !24 + %902 = fpext bfloat %526 to float, !dbg !24 + %903 = fpext bfloat %527 to float, !dbg !24 + %904 = fpext bfloat %528 to float, !dbg !24 + %905 = fpext bfloat %529 to float, !dbg !24 + %906 = fpext bfloat %530 to float, !dbg !24 + %907 = fpext bfloat %541 to float, !dbg !24 + %908 = fpext bfloat %542 to float, !dbg !24 + %909 = fpext bfloat %543 to float, !dbg !24 + %910 = fpext bfloat %544 to float, !dbg !24 + %911 = fpext bfloat %545 to float, !dbg !24 + %912 = fpext bfloat %546 to float, !dbg !24 + %913 = fpext bfloat %547 to float, !dbg !24 + %914 = fpext bfloat %548 to float, !dbg !24 + %915 = fpext bfloat %559 to float, !dbg !24 + %916 = fpext bfloat %560 to float, !dbg !24 + %917 = fpext bfloat %561 to float, !dbg !24 + %918 = fpext bfloat %562 to float, !dbg !24 + %919 = fpext bfloat %563 to float, !dbg !24 + %920 = fpext bfloat %564 to float, !dbg !24 + %921 = fpext bfloat %565 to float, !dbg !24 + %922 = fpext bfloat %566 to float, !dbg !24 + %923 = fpext bfloat %577 to float, !dbg !24 + %924 = fpext bfloat %578 to float, !dbg !24 + %925 = fpext bfloat %579 to float, !dbg !24 + %926 = fpext bfloat %580 to float, !dbg !24 + %927 = fpext bfloat %581 to float, !dbg !24 + %928 = fpext bfloat %582 to float, !dbg !24 + %929 = fpext bfloat %583 to float, !dbg !24 + %930 = fpext bfloat %584 to float, !dbg !24 + %931 = fpext bfloat %595 to float, !dbg !24 + %932 = fpext bfloat %596 to float, !dbg !24 + %933 = fpext bfloat %597 to float, !dbg !24 + %934 = fpext bfloat %598 to float, !dbg !24 + %935 = fpext bfloat %599 to float, !dbg !24 + %936 = fpext bfloat %600 to float, !dbg !24 + %937 = fpext bfloat %601 to float, !dbg !24 + %938 = fpext bfloat %602 to float, !dbg !24 + %939 = fpext bfloat %613 to float, !dbg !24 + %940 = fpext bfloat %614 to float, !dbg !24 + %941 = fpext bfloat %615 to float, !dbg !24 + %942 = fpext bfloat %616 to float, !dbg !24 + %943 = fpext bfloat %617 to float, !dbg !24 + %944 = fpext bfloat %618 to float, !dbg !24 + %945 = fpext bfloat %619 to float, !dbg !24 + %946 = fpext bfloat %620 to float, !dbg !24 + %947 = fpext bfloat %631 to float, !dbg !24 + %948 = fpext bfloat %632 to float, !dbg !24 + %949 = fpext bfloat %633 to float, !dbg !24 + %950 = fpext bfloat %634 to float, !dbg !24 + %951 = fpext bfloat %635 to float, !dbg !24 + %952 = fpext bfloat %636 to float, !dbg !24 + %953 = fpext bfloat %637 to float, !dbg !24 + %954 = fpext bfloat %638 to float, !dbg !24 + %955 = fpext bfloat %649 to float, !dbg !24 + %956 = fpext bfloat %650 to float, !dbg !24 + %957 = fpext bfloat %651 to float, !dbg !24 + %958 = fpext bfloat %652 to float, !dbg !24 + %959 = fpext bfloat %653 to float, !dbg !24 + %960 = fpext bfloat %654 to float, !dbg !24 + %961 = fpext bfloat %655 to float, !dbg !24 + %962 = fpext bfloat %656 to float, !dbg !24 + %963 = fpext bfloat %667 to float, !dbg !24 + %964 = fpext bfloat %668 to float, !dbg !24 + %965 = fpext bfloat %669 to float, !dbg !24 + %966 = fpext bfloat %670 to float, !dbg !24 + %967 = fpext bfloat %671 to float, !dbg !24 + %968 = fpext bfloat %672 to float, !dbg !24 + %969 = fpext bfloat %673 to float, !dbg !24 + %970 = fpext bfloat %674 to float, !dbg !24 + %971 = fpext bfloat %685 to float, !dbg !24 + %972 = fpext bfloat %686 to float, !dbg !24 + %973 = fpext bfloat %687 to float, !dbg !24 + %974 = fpext bfloat %688 to float, !dbg !24 + %975 = fpext bfloat %689 to float, !dbg !24 + %976 = fpext bfloat %690 to float, !dbg !24 + %977 = fpext bfloat %691 to float, !dbg !24 + %978 = fpext bfloat %692 to float, !dbg !24 + %979 = fpext bfloat %703 to float, !dbg !24 + %980 = fpext bfloat %704 to float, !dbg !24 + %981 = fpext bfloat %705 to float, !dbg !24 + %982 = fpext bfloat %706 to float, !dbg !24 + %983 = fpext bfloat %707 to float, !dbg !24 + %984 = fpext bfloat %708 to float, !dbg !24 + %985 = fpext bfloat %709 to float, !dbg !24 + %986 = fpext bfloat %710 to float, !dbg !24 + %987 = fpext bfloat %721 to float, !dbg !24 + %988 = fpext bfloat %722 to float, !dbg !24 + %989 = fpext bfloat %723 to float, !dbg !24 + %990 = fpext bfloat %724 to float, !dbg !24 + %991 = fpext bfloat %725 to float, !dbg !24 + %992 = fpext bfloat %726 to float, !dbg !24 + %993 = fpext bfloat %727 to float, !dbg !24 + %994 = fpext bfloat %728 to float, !dbg !24 + %995 = fpext bfloat %739 to float, !dbg !24 + %996 = fpext bfloat %740 to float, !dbg !24 + %997 = fpext bfloat %741 to float, !dbg !24 + %998 = fpext bfloat %742 to float, !dbg !24 + %999 = fpext bfloat %743 to float, !dbg !24 + %1000 = fpext bfloat %744 to float, !dbg !24 + %1001 = fpext bfloat %745 to float, !dbg !24 + %1002 = fpext bfloat %746 to float, !dbg !24 + %1003 = fmul float %747, %747, !dbg !25 + %1004 = fmul float %748, %748, !dbg !25 + %1005 = fmul float %749, %749, !dbg !25 + %1006 = fmul float %750, %750, !dbg !25 + %1007 = fmul float %751, %751, !dbg !25 + %1008 = fmul float %752, %752, !dbg !25 + %1009 = fmul float %753, %753, !dbg !25 + %1010 = fmul float %754, %754, !dbg !25 + %1011 = fmul float %755, %755, !dbg !25 + %1012 = fmul float %756, %756, !dbg !25 + %1013 = fmul float %757, %757, !dbg !25 + %1014 = fmul float %758, %758, !dbg !25 + %1015 = fmul float %759, %759, !dbg !25 + %1016 = fmul float %760, %760, !dbg !25 + %1017 = fmul float %761, %761, !dbg !25 + %1018 = fmul float %762, %762, !dbg !25 + %1019 = fmul float %763, %763, !dbg !25 + %1020 = fmul float %764, %764, !dbg !25 + %1021 = fmul float %765, %765, !dbg !25 + %1022 = fmul float %766, %766, !dbg !25 + %1023 = fmul float %767, %767, !dbg !25 + %1024 = fmul float %768, %768, !dbg !25 + %1025 = fmul float %769, %769, !dbg !25 + %1026 = fmul float %770, %770, !dbg !25 + %1027 = fmul float %771, %771, !dbg !25 + %1028 = fmul float %772, %772, !dbg !25 + %1029 = fmul float %773, %773, !dbg !25 + %1030 = fmul float %774, %774, !dbg !25 + %1031 = fmul float %775, %775, !dbg !25 + %1032 = fmul float %776, %776, !dbg !25 + %1033 = fmul float %777, %777, !dbg !25 + %1034 = fmul float %778, %778, !dbg !25 + %1035 = fmul float %779, %779, !dbg !25 + %1036 = fmul float %780, %780, !dbg !25 + %1037 = fmul float %781, %781, !dbg !25 + %1038 = fmul float %782, %782, !dbg !25 + %1039 = fmul float %783, %783, !dbg !25 + %1040 = fmul float %784, %784, !dbg !25 + %1041 = fmul float %785, %785, !dbg !25 + %1042 = fmul float %786, %786, !dbg !25 + %1043 = fmul float %787, %787, !dbg !25 + %1044 = fmul float %788, %788, !dbg !25 + %1045 = fmul float %789, %789, !dbg !25 + %1046 = fmul float %790, %790, !dbg !25 + %1047 = fmul float %791, %791, !dbg !25 + %1048 = fmul float %792, %792, !dbg !25 + %1049 = fmul float %793, %793, !dbg !25 + %1050 = fmul float %794, %794, !dbg !25 + %1051 = fmul float %795, %795, !dbg !25 + %1052 = fmul float %796, %796, !dbg !25 + %1053 = fmul float %797, %797, !dbg !25 + %1054 = fmul float %798, %798, !dbg !25 + %1055 = fmul float %799, %799, !dbg !25 + %1056 = fmul float %800, %800, !dbg !25 + %1057 = fmul float %801, %801, !dbg !25 + %1058 = fmul float %802, %802, !dbg !25 + %1059 = fmul float %803, %803, !dbg !25 + %1060 = fmul float %804, %804, !dbg !25 + %1061 = fmul float %805, %805, !dbg !25 + %1062 = fmul float %806, %806, !dbg !25 + %1063 = fmul float %807, %807, !dbg !25 + %1064 = fmul float %808, %808, !dbg !25 + %1065 = fmul float %809, %809, !dbg !25 + %1066 = fmul float %810, %810, !dbg !25 + %1067 = fmul float %811, %811, !dbg !25 + %1068 = fmul float %812, %812, !dbg !25 + %1069 = fmul float %813, %813, !dbg !25 + %1070 = fmul float %814, %814, !dbg !25 + %1071 = fmul float %815, %815, !dbg !25 + %1072 = fmul float %816, %816, !dbg !25 + %1073 = fmul float %817, %817, !dbg !25 + %1074 = fmul float %818, %818, !dbg !25 + %1075 = fmul float %819, %819, !dbg !25 + %1076 = fmul float %820, %820, !dbg !25 + %1077 = fmul float %821, %821, !dbg !25 + %1078 = fmul float %822, %822, !dbg !25 + %1079 = fmul float %823, %823, !dbg !25 + %1080 = fmul float %824, %824, !dbg !25 + %1081 = fmul float %825, %825, !dbg !25 + %1082 = fmul float %826, %826, !dbg !25 + %1083 = fmul float %827, %827, !dbg !25 + %1084 = fmul float %828, %828, !dbg !25 + %1085 = fmul float %829, %829, !dbg !25 + %1086 = fmul float %830, %830, !dbg !25 + %1087 = fmul float %831, %831, !dbg !25 + %1088 = fmul float %832, %832, !dbg !25 + %1089 = fmul float %833, %833, !dbg !25 + %1090 = fmul float %834, %834, !dbg !25 + %1091 = fmul float %835, %835, !dbg !25 + %1092 = fmul float %836, %836, !dbg !25 + %1093 = fmul float %837, %837, !dbg !25 + %1094 = fmul float %838, %838, !dbg !25 + %1095 = fmul float %839, %839, !dbg !25 + %1096 = fmul float %840, %840, !dbg !25 + %1097 = fmul float %841, %841, !dbg !25 + %1098 = fmul float %842, %842, !dbg !25 + %1099 = fmul float %843, %843, !dbg !25 + %1100 = fmul float %844, %844, !dbg !25 + %1101 = fmul float %845, %845, !dbg !25 + %1102 = fmul float %846, %846, !dbg !25 + %1103 = fmul float %847, %847, !dbg !25 + %1104 = fmul float %848, %848, !dbg !25 + %1105 = fmul float %849, %849, !dbg !25 + %1106 = fmul float %850, %850, !dbg !25 + %1107 = fmul float %851, %851, !dbg !25 + %1108 = fmul float %852, %852, !dbg !25 + %1109 = fmul float %853, %853, !dbg !25 + %1110 = fmul float %854, %854, !dbg !25 + %1111 = fmul float %855, %855, !dbg !25 + %1112 = fmul float %856, %856, !dbg !25 + %1113 = fmul float %857, %857, !dbg !25 + %1114 = fmul float %858, %858, !dbg !25 + %1115 = fmul float %859, %859, !dbg !25 + %1116 = fmul float %860, %860, !dbg !25 + %1117 = fmul float %861, %861, !dbg !25 + %1118 = fmul float %862, %862, !dbg !25 + %1119 = fmul float %863, %863, !dbg !25 + %1120 = fmul float %864, %864, !dbg !25 + %1121 = fmul float %865, %865, !dbg !25 + %1122 = fmul float %866, %866, !dbg !25 + %1123 = fmul float %867, %867, !dbg !25 + %1124 = fmul float %868, %868, !dbg !25 + %1125 = fmul float %869, %869, !dbg !25 + %1126 = fmul float %870, %870, !dbg !25 + %1127 = fmul float %871, %871, !dbg !25 + %1128 = fmul float %872, %872, !dbg !25 + %1129 = fmul float %873, %873, !dbg !25 + %1130 = fmul float %874, %874, !dbg !25 + %1131 = fmul float %875, %875, !dbg !25 + %1132 = fmul float %876, %876, !dbg !25 + %1133 = fmul float %877, %877, !dbg !25 + %1134 = fmul float %878, %878, !dbg !25 + %1135 = fmul float %879, %879, !dbg !25 + %1136 = fmul float %880, %880, !dbg !25 + %1137 = fmul float %881, %881, !dbg !25 + %1138 = fmul float %882, %882, !dbg !25 + %1139 = fmul float %883, %883, !dbg !25 + %1140 = fmul float %884, %884, !dbg !25 + %1141 = fmul float %885, %885, !dbg !25 + %1142 = fmul float %886, %886, !dbg !25 + %1143 = fmul float %887, %887, !dbg !25 + %1144 = fmul float %888, %888, !dbg !25 + %1145 = fmul float %889, %889, !dbg !25 + %1146 = fmul float %890, %890, !dbg !25 + %1147 = fmul float %891, %891, !dbg !25 + %1148 = fmul float %892, %892, !dbg !25 + %1149 = fmul float %893, %893, !dbg !25 + %1150 = fmul float %894, %894, !dbg !25 + %1151 = fmul float %895, %895, !dbg !25 + %1152 = fmul float %896, %896, !dbg !25 + %1153 = fmul float %897, %897, !dbg !25 + %1154 = fmul float %898, %898, !dbg !25 + %1155 = fmul float %899, %899, !dbg !25 + %1156 = fmul float %900, %900, !dbg !25 + %1157 = fmul float %901, %901, !dbg !25 + %1158 = fmul float %902, %902, !dbg !25 + %1159 = fmul float %903, %903, !dbg !25 + %1160 = fmul float %904, %904, !dbg !25 + %1161 = fmul float %905, %905, !dbg !25 + %1162 = fmul float %906, %906, !dbg !25 + %1163 = fmul float %907, %907, !dbg !25 + %1164 = fmul float %908, %908, !dbg !25 + %1165 = fmul float %909, %909, !dbg !25 + %1166 = fmul float %910, %910, !dbg !25 + %1167 = fmul float %911, %911, !dbg !25 + %1168 = fmul float %912, %912, !dbg !25 + %1169 = fmul float %913, %913, !dbg !25 + %1170 = fmul float %914, %914, !dbg !25 + %1171 = fmul float %915, %915, !dbg !25 + %1172 = fmul float %916, %916, !dbg !25 + %1173 = fmul float %917, %917, !dbg !25 + %1174 = fmul float %918, %918, !dbg !25 + %1175 = fmul float %919, %919, !dbg !25 + %1176 = fmul float %920, %920, !dbg !25 + %1177 = fmul float %921, %921, !dbg !25 + %1178 = fmul float %922, %922, !dbg !25 + %1179 = fmul float %923, %923, !dbg !25 + %1180 = fmul float %924, %924, !dbg !25 + %1181 = fmul float %925, %925, !dbg !25 + %1182 = fmul float %926, %926, !dbg !25 + %1183 = fmul float %927, %927, !dbg !25 + %1184 = fmul float %928, %928, !dbg !25 + %1185 = fmul float %929, %929, !dbg !25 + %1186 = fmul float %930, %930, !dbg !25 + %1187 = fmul float %931, %931, !dbg !25 + %1188 = fmul float %932, %932, !dbg !25 + %1189 = fmul float %933, %933, !dbg !25 + %1190 = fmul float %934, %934, !dbg !25 + %1191 = fmul float %935, %935, !dbg !25 + %1192 = fmul float %936, %936, !dbg !25 + %1193 = fmul float %937, %937, !dbg !25 + %1194 = fmul float %938, %938, !dbg !25 + %1195 = fmul float %939, %939, !dbg !25 + %1196 = fmul float %940, %940, !dbg !25 + %1197 = fmul float %941, %941, !dbg !25 + %1198 = fmul float %942, %942, !dbg !25 + %1199 = fmul float %943, %943, !dbg !25 + %1200 = fmul float %944, %944, !dbg !25 + %1201 = fmul float %945, %945, !dbg !25 + %1202 = fmul float %946, %946, !dbg !25 + %1203 = fmul float %947, %947, !dbg !25 + %1204 = fmul float %948, %948, !dbg !25 + %1205 = fmul float %949, %949, !dbg !25 + %1206 = fmul float %950, %950, !dbg !25 + %1207 = fmul float %951, %951, !dbg !25 + %1208 = fmul float %952, %952, !dbg !25 + %1209 = fmul float %953, %953, !dbg !25 + %1210 = fmul float %954, %954, !dbg !25 + %1211 = fmul float %955, %955, !dbg !25 + %1212 = fmul float %956, %956, !dbg !25 + %1213 = fmul float %957, %957, !dbg !25 + %1214 = fmul float %958, %958, !dbg !25 + %1215 = fmul float %959, %959, !dbg !25 + %1216 = fmul float %960, %960, !dbg !25 + %1217 = fmul float %961, %961, !dbg !25 + %1218 = fmul float %962, %962, !dbg !25 + %1219 = fmul float %963, %963, !dbg !25 + %1220 = fmul float %964, %964, !dbg !25 + %1221 = fmul float %965, %965, !dbg !25 + %1222 = fmul float %966, %966, !dbg !25 + %1223 = fmul float %967, %967, !dbg !25 + %1224 = fmul float %968, %968, !dbg !25 + %1225 = fmul float %969, %969, !dbg !25 + %1226 = fmul float %970, %970, !dbg !25 + %1227 = fmul float %971, %971, !dbg !25 + %1228 = fmul float %972, %972, !dbg !25 + %1229 = fmul float %973, %973, !dbg !25 + %1230 = fmul float %974, %974, !dbg !25 + %1231 = fmul float %975, %975, !dbg !25 + %1232 = fmul float %976, %976, !dbg !25 + %1233 = fmul float %977, %977, !dbg !25 + %1234 = fmul float %978, %978, !dbg !25 + %1235 = fmul float %979, %979, !dbg !25 + %1236 = fmul float %980, %980, !dbg !25 + %1237 = fmul float %981, %981, !dbg !25 + %1238 = fmul float %982, %982, !dbg !25 + %1239 = fmul float %983, %983, !dbg !25 + %1240 = fmul float %984, %984, !dbg !25 + %1241 = fmul float %985, %985, !dbg !25 + %1242 = fmul float %986, %986, !dbg !25 + %1243 = fmul float %987, %987, !dbg !25 + %1244 = fmul float %988, %988, !dbg !25 + %1245 = fmul float %989, %989, !dbg !25 + %1246 = fmul float %990, %990, !dbg !25 + %1247 = fmul float %991, %991, !dbg !25 + %1248 = fmul float %992, %992, !dbg !25 + %1249 = fmul float %993, %993, !dbg !25 + %1250 = fmul float %994, %994, !dbg !25 + %1251 = fmul float %995, %995, !dbg !25 + %1252 = fmul float %996, %996, !dbg !25 + %1253 = fmul float %997, %997, !dbg !25 + %1254 = fmul float %998, %998, !dbg !25 + %1255 = fmul float %999, %999, !dbg !25 + %1256 = fmul float %1000, %1000, !dbg !25 + %1257 = fmul float %1001, %1001, !dbg !25 + %1258 = fmul float %1002, %1002, !dbg !25 + %1259 = fadd float %1003, %1004, !dbg !26 + %1260 = fadd float %1005, %1259, !dbg !26 + %1261 = fadd float %1006, %1260, !dbg !26 + %1262 = fadd float %1007, %1261, !dbg !26 + %1263 = fadd float %1008, %1262, !dbg !26 + %1264 = fadd float %1009, %1263, !dbg !26 + %1265 = fadd float %1010, %1264, !dbg !26 + %1266 = fadd float %1011, %1012, !dbg !26 + %1267 = fadd float %1013, %1266, !dbg !26 + %1268 = fadd float %1014, %1267, !dbg !26 + %1269 = fadd float %1015, %1268, !dbg !26 + %1270 = fadd float %1016, %1269, !dbg !26 + %1271 = fadd float %1017, %1270, !dbg !26 + %1272 = fadd float %1018, %1271, !dbg !26 + %1273 = fadd float %1019, %1020, !dbg !26 + %1274 = fadd float %1021, %1273, !dbg !26 + %1275 = fadd float %1022, %1274, !dbg !26 + %1276 = fadd float %1023, %1275, !dbg !26 + %1277 = fadd float %1024, %1276, !dbg !26 + %1278 = fadd float %1025, %1277, !dbg !26 + %1279 = fadd float %1026, %1278, !dbg !26 + %1280 = fadd float %1027, %1028, !dbg !26 + %1281 = fadd float %1029, %1280, !dbg !26 + %1282 = fadd float %1030, %1281, !dbg !26 + %1283 = fadd float %1031, %1282, !dbg !26 + %1284 = fadd float %1032, %1283, !dbg !26 + %1285 = fadd float %1033, %1284, !dbg !26 + %1286 = fadd float %1034, %1285, !dbg !26 + %1287 = fadd float %1035, %1036, !dbg !26 + %1288 = fadd float %1037, %1287, !dbg !26 + %1289 = fadd float %1038, %1288, !dbg !26 + %1290 = fadd float %1039, %1289, !dbg !26 + %1291 = fadd float %1040, %1290, !dbg !26 + %1292 = fadd float %1041, %1291, !dbg !26 + %1293 = fadd float %1042, %1292, !dbg !26 + %1294 = fadd float %1043, %1044, !dbg !26 + %1295 = fadd float %1045, %1294, !dbg !26 + %1296 = fadd float %1046, %1295, !dbg !26 + %1297 = fadd float %1047, %1296, !dbg !26 + %1298 = fadd float %1048, %1297, !dbg !26 + %1299 = fadd float %1049, %1298, !dbg !26 + %1300 = fadd float %1050, %1299, !dbg !26 + %1301 = fadd float %1051, %1052, !dbg !26 + %1302 = fadd float %1053, %1301, !dbg !26 + %1303 = fadd float %1054, %1302, !dbg !26 + %1304 = fadd float %1055, %1303, !dbg !26 + %1305 = fadd float %1056, %1304, !dbg !26 + %1306 = fadd float %1057, %1305, !dbg !26 + %1307 = fadd float %1058, %1306, !dbg !26 + %1308 = fadd float %1059, %1060, !dbg !26 + %1309 = fadd float %1061, %1308, !dbg !26 + %1310 = fadd float %1062, %1309, !dbg !26 + %1311 = fadd float %1063, %1310, !dbg !26 + %1312 = fadd float %1064, %1311, !dbg !26 + %1313 = fadd float %1065, %1312, !dbg !26 + %1314 = fadd float %1066, %1313, !dbg !26 + %1315 = fadd float %1067, %1068, !dbg !26 + %1316 = fadd float %1069, %1315, !dbg !26 + %1317 = fadd float %1070, %1316, !dbg !26 + %1318 = fadd float %1071, %1317, !dbg !26 + %1319 = fadd float %1072, %1318, !dbg !26 + %1320 = fadd float %1073, %1319, !dbg !26 + %1321 = fadd float %1074, %1320, !dbg !26 + %1322 = fadd float %1075, %1076, !dbg !26 + %1323 = fadd float %1077, %1322, !dbg !26 + %1324 = fadd float %1078, %1323, !dbg !26 + %1325 = fadd float %1079, %1324, !dbg !26 + %1326 = fadd float %1080, %1325, !dbg !26 + %1327 = fadd float %1081, %1326, !dbg !26 + %1328 = fadd float %1082, %1327, !dbg !26 + %1329 = fadd float %1083, %1084, !dbg !26 + %1330 = fadd float %1085, %1329, !dbg !26 + %1331 = fadd float %1086, %1330, !dbg !26 + %1332 = fadd float %1087, %1331, !dbg !26 + %1333 = fadd float %1088, %1332, !dbg !26 + %1334 = fadd float %1089, %1333, !dbg !26 + %1335 = fadd float %1090, %1334, !dbg !26 + %1336 = fadd float %1091, %1092, !dbg !26 + %1337 = fadd float %1093, %1336, !dbg !26 + %1338 = fadd float %1094, %1337, !dbg !26 + %1339 = fadd float %1095, %1338, !dbg !26 + %1340 = fadd float %1096, %1339, !dbg !26 + %1341 = fadd float %1097, %1340, !dbg !26 + %1342 = fadd float %1098, %1341, !dbg !26 + %1343 = fadd float %1099, %1100, !dbg !26 + %1344 = fadd float %1101, %1343, !dbg !26 + %1345 = fadd float %1102, %1344, !dbg !26 + %1346 = fadd float %1103, %1345, !dbg !26 + %1347 = fadd float %1104, %1346, !dbg !26 + %1348 = fadd float %1105, %1347, !dbg !26 + %1349 = fadd float %1106, %1348, !dbg !26 + %1350 = fadd float %1107, %1108, !dbg !26 + %1351 = fadd float %1109, %1350, !dbg !26 + %1352 = fadd float %1110, %1351, !dbg !26 + %1353 = fadd float %1111, %1352, !dbg !26 + %1354 = fadd float %1112, %1353, !dbg !26 + %1355 = fadd float %1113, %1354, !dbg !26 + %1356 = fadd float %1114, %1355, !dbg !26 + %1357 = fadd float %1115, %1116, !dbg !26 + %1358 = fadd float %1117, %1357, !dbg !26 + %1359 = fadd float %1118, %1358, !dbg !26 + %1360 = fadd float %1119, %1359, !dbg !26 + %1361 = fadd float %1120, %1360, !dbg !26 + %1362 = fadd float %1121, %1361, !dbg !26 + %1363 = fadd float %1122, %1362, !dbg !26 + %1364 = fadd float %1123, %1124, !dbg !26 + %1365 = fadd float %1125, %1364, !dbg !26 + %1366 = fadd float %1126, %1365, !dbg !26 + %1367 = fadd float %1127, %1366, !dbg !26 + %1368 = fadd float %1128, %1367, !dbg !26 + %1369 = fadd float %1129, %1368, !dbg !26 + %1370 = fadd float %1130, %1369, !dbg !26 + %1371 = fadd float %1131, %1132, !dbg !26 + %1372 = fadd float %1133, %1371, !dbg !26 + %1373 = fadd float %1134, %1372, !dbg !26 + %1374 = fadd float %1135, %1373, !dbg !26 + %1375 = fadd float %1136, %1374, !dbg !26 + %1376 = fadd float %1137, %1375, !dbg !26 + %1377 = fadd float %1138, %1376, !dbg !26 + %1378 = fadd float %1139, %1140, !dbg !26 + %1379 = fadd float %1141, %1378, !dbg !26 + %1380 = fadd float %1142, %1379, !dbg !26 + %1381 = fadd float %1143, %1380, !dbg !26 + %1382 = fadd float %1144, %1381, !dbg !26 + %1383 = fadd float %1145, %1382, !dbg !26 + %1384 = fadd float %1146, %1383, !dbg !26 + %1385 = fadd float %1147, %1148, !dbg !26 + %1386 = fadd float %1149, %1385, !dbg !26 + %1387 = fadd float %1150, %1386, !dbg !26 + %1388 = fadd float %1151, %1387, !dbg !26 + %1389 = fadd float %1152, %1388, !dbg !26 + %1390 = fadd float %1153, %1389, !dbg !26 + %1391 = fadd float %1154, %1390, !dbg !26 + %1392 = fadd float %1155, %1156, !dbg !26 + %1393 = fadd float %1157, %1392, !dbg !26 + %1394 = fadd float %1158, %1393, !dbg !26 + %1395 = fadd float %1159, %1394, !dbg !26 + %1396 = fadd float %1160, %1395, !dbg !26 + %1397 = fadd float %1161, %1396, !dbg !26 + %1398 = fadd float %1162, %1397, !dbg !26 + %1399 = fadd float %1163, %1164, !dbg !26 + %1400 = fadd float %1165, %1399, !dbg !26 + %1401 = fadd float %1166, %1400, !dbg !26 + %1402 = fadd float %1167, %1401, !dbg !26 + %1403 = fadd float %1168, %1402, !dbg !26 + %1404 = fadd float %1169, %1403, !dbg !26 + %1405 = fadd float %1170, %1404, !dbg !26 + %1406 = fadd float %1171, %1172, !dbg !26 + %1407 = fadd float %1173, %1406, !dbg !26 + %1408 = fadd float %1174, %1407, !dbg !26 + %1409 = fadd float %1175, %1408, !dbg !26 + %1410 = fadd float %1176, %1409, !dbg !26 + %1411 = fadd float %1177, %1410, !dbg !26 + %1412 = fadd float %1178, %1411, !dbg !26 + %1413 = fadd float %1179, %1180, !dbg !26 + %1414 = fadd float %1181, %1413, !dbg !26 + %1415 = fadd float %1182, %1414, !dbg !26 + %1416 = fadd float %1183, %1415, !dbg !26 + %1417 = fadd float %1184, %1416, !dbg !26 + %1418 = fadd float %1185, %1417, !dbg !26 + %1419 = fadd float %1186, %1418, !dbg !26 + %1420 = fadd float %1187, %1188, !dbg !26 + %1421 = fadd float %1189, %1420, !dbg !26 + %1422 = fadd float %1190, %1421, !dbg !26 + %1423 = fadd float %1191, %1422, !dbg !26 + %1424 = fadd float %1192, %1423, !dbg !26 + %1425 = fadd float %1193, %1424, !dbg !26 + %1426 = fadd float %1194, %1425, !dbg !26 + %1427 = fadd float %1195, %1196, !dbg !26 + %1428 = fadd float %1197, %1427, !dbg !26 + %1429 = fadd float %1198, %1428, !dbg !26 + %1430 = fadd float %1199, %1429, !dbg !26 + %1431 = fadd float %1200, %1430, !dbg !26 + %1432 = fadd float %1201, %1431, !dbg !26 + %1433 = fadd float %1202, %1432, !dbg !26 + %1434 = fadd float %1203, %1204, !dbg !26 + %1435 = fadd float %1205, %1434, !dbg !26 + %1436 = fadd float %1206, %1435, !dbg !26 + %1437 = fadd float %1207, %1436, !dbg !26 + %1438 = fadd float %1208, %1437, !dbg !26 + %1439 = fadd float %1209, %1438, !dbg !26 + %1440 = fadd float %1210, %1439, !dbg !26 + %1441 = fadd float %1211, %1212, !dbg !26 + %1442 = fadd float %1213, %1441, !dbg !26 + %1443 = fadd float %1214, %1442, !dbg !26 + %1444 = fadd float %1215, %1443, !dbg !26 + %1445 = fadd float %1216, %1444, !dbg !26 + %1446 = fadd float %1217, %1445, !dbg !26 + %1447 = fadd float %1218, %1446, !dbg !26 + %1448 = fadd float %1219, %1220, !dbg !26 + %1449 = fadd float %1221, %1448, !dbg !26 + %1450 = fadd float %1222, %1449, !dbg !26 + %1451 = fadd float %1223, %1450, !dbg !26 + %1452 = fadd float %1224, %1451, !dbg !26 + %1453 = fadd float %1225, %1452, !dbg !26 + %1454 = fadd float %1226, %1453, !dbg !26 + %1455 = fadd float %1227, %1228, !dbg !26 + %1456 = fadd float %1229, %1455, !dbg !26 + %1457 = fadd float %1230, %1456, !dbg !26 + %1458 = fadd float %1231, %1457, !dbg !26 + %1459 = fadd float %1232, %1458, !dbg !26 + %1460 = fadd float %1233, %1459, !dbg !26 + %1461 = fadd float %1234, %1460, !dbg !26 + %1462 = fadd float %1235, %1236, !dbg !26 + %1463 = fadd float %1237, %1462, !dbg !26 + %1464 = fadd float %1238, %1463, !dbg !26 + %1465 = fadd float %1239, %1464, !dbg !26 + %1466 = fadd float %1240, %1465, !dbg !26 + %1467 = fadd float %1241, %1466, !dbg !26 + %1468 = fadd float %1242, %1467, !dbg !26 + %1469 = fadd float %1243, %1244, !dbg !26 + %1470 = fadd float %1245, %1469, !dbg !26 + %1471 = fadd float %1246, %1470, !dbg !26 + %1472 = fadd float %1247, %1471, !dbg !26 + %1473 = fadd float %1248, %1472, !dbg !26 + %1474 = fadd float %1249, %1473, !dbg !26 + %1475 = fadd float %1250, %1474, !dbg !26 + %1476 = fadd float %1251, %1252, !dbg !26 + %1477 = fadd float %1253, %1476, !dbg !26 + %1478 = fadd float %1254, %1477, !dbg !26 + %1479 = fadd float %1255, %1478, !dbg !26 + %1480 = fadd float %1256, %1479, !dbg !26 + %1481 = fadd float %1257, %1480, !dbg !26 + %1482 = fadd float %1258, %1481, !dbg !26 + %1483 = bitcast float %1265 to i32, !dbg !29 + %1484 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1483, i32 8, i32 31), !dbg !29 + %1485 = bitcast i32 %1484 to float, !dbg !29 + %1486 = fadd float %1265, %1485, !dbg !26 + %1487 = bitcast float %1486 to i32, !dbg !29 + %1488 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1487, i32 4, i32 31), !dbg !29 + %1489 = bitcast i32 %1488 to float, !dbg !29 + %1490 = fadd float %1486, %1489, !dbg !26 + %1491 = bitcast float %1490 to i32, !dbg !29 + %1492 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1491, i32 2, i32 31), !dbg !29 + %1493 = bitcast i32 %1492 to float, !dbg !29 + %1494 = fadd float %1490, %1493, !dbg !26 + %1495 = bitcast float %1494 to i32, !dbg !29 + %1496 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1495, i32 1, i32 31), !dbg !29 + %1497 = bitcast i32 %1496 to float, !dbg !29 + %1498 = fadd float %1494, %1497, !dbg !26 + %1499 = bitcast float %1272 to i32, !dbg !29 + %1500 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1499, i32 8, i32 31), !dbg !29 + %1501 = bitcast i32 %1500 to float, !dbg !29 + %1502 = fadd float %1272, %1501, !dbg !26 + %1503 = bitcast float %1502 to i32, !dbg !29 + %1504 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1503, i32 4, i32 31), !dbg !29 + %1505 = bitcast i32 %1504 to float, !dbg !29 + %1506 = fadd float %1502, %1505, !dbg !26 + %1507 = bitcast float %1506 to i32, !dbg !29 + %1508 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1507, i32 2, i32 31), !dbg !29 + %1509 = bitcast i32 %1508 to float, !dbg !29 + %1510 = fadd float %1506, %1509, !dbg !26 + %1511 = bitcast float %1510 to i32, !dbg !29 + %1512 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1511, i32 1, i32 31), !dbg !29 + %1513 = bitcast i32 %1512 to float, !dbg !29 + %1514 = fadd float %1510, %1513, !dbg !26 + %1515 = bitcast float %1279 to i32, !dbg !29 + %1516 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1515, i32 8, i32 31), !dbg !29 + %1517 = bitcast i32 %1516 to float, !dbg !29 + %1518 = fadd float %1279, %1517, !dbg !26 + %1519 = bitcast float %1518 to i32, !dbg !29 + %1520 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1519, i32 4, i32 31), !dbg !29 + %1521 = bitcast i32 %1520 to float, !dbg !29 + %1522 = fadd float %1518, %1521, !dbg !26 + %1523 = bitcast float %1522 to i32, !dbg !29 + %1524 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1523, i32 2, i32 31), !dbg !29 + %1525 = bitcast i32 %1524 to float, !dbg !29 + %1526 = fadd float %1522, %1525, !dbg !26 + %1527 = bitcast float %1526 to i32, !dbg !29 + %1528 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1527, i32 1, i32 31), !dbg !29 + %1529 = bitcast i32 %1528 to float, !dbg !29 + %1530 = fadd float %1526, %1529, !dbg !26 + %1531 = bitcast float %1286 to i32, !dbg !29 + %1532 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1531, i32 8, i32 31), !dbg !29 + %1533 = bitcast i32 %1532 to float, !dbg !29 + %1534 = fadd float %1286, %1533, !dbg !26 + %1535 = bitcast float %1534 to i32, !dbg !29 + %1536 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1535, i32 4, i32 31), !dbg !29 + %1537 = bitcast i32 %1536 to float, !dbg !29 + %1538 = fadd float %1534, %1537, !dbg !26 + %1539 = bitcast float %1538 to i32, !dbg !29 + %1540 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1539, i32 2, i32 31), !dbg !29 + %1541 = bitcast i32 %1540 to float, !dbg !29 + %1542 = fadd float %1538, %1541, !dbg !26 + %1543 = bitcast float %1542 to i32, !dbg !29 + %1544 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1543, i32 1, i32 31), !dbg !29 + %1545 = bitcast i32 %1544 to float, !dbg !29 + %1546 = fadd float %1542, %1545, !dbg !26 + %1547 = bitcast float %1293 to i32, !dbg !29 + %1548 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1547, i32 8, i32 31), !dbg !29 + %1549 = bitcast i32 %1548 to float, !dbg !29 + %1550 = fadd float %1293, %1549, !dbg !26 + %1551 = bitcast float %1550 to i32, !dbg !29 + %1552 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1551, i32 4, i32 31), !dbg !29 + %1553 = bitcast i32 %1552 to float, !dbg !29 + %1554 = fadd float %1550, %1553, !dbg !26 + %1555 = bitcast float %1554 to i32, !dbg !29 + %1556 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1555, i32 2, i32 31), !dbg !29 + %1557 = bitcast i32 %1556 to float, !dbg !29 + %1558 = fadd float %1554, %1557, !dbg !26 + %1559 = bitcast float %1558 to i32, !dbg !29 + %1560 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1559, i32 1, i32 31), !dbg !29 + %1561 = bitcast i32 %1560 to float, !dbg !29 + %1562 = fadd float %1558, %1561, !dbg !26 + %1563 = bitcast float %1300 to i32, !dbg !29 + %1564 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1563, i32 8, i32 31), !dbg !29 + %1565 = bitcast i32 %1564 to float, !dbg !29 + %1566 = fadd float %1300, %1565, !dbg !26 + %1567 = bitcast float %1566 to i32, !dbg !29 + %1568 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1567, i32 4, i32 31), !dbg !29 + %1569 = bitcast i32 %1568 to float, !dbg !29 + %1570 = fadd float %1566, %1569, !dbg !26 + %1571 = bitcast float %1570 to i32, !dbg !29 + %1572 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1571, i32 2, i32 31), !dbg !29 + %1573 = bitcast i32 %1572 to float, !dbg !29 + %1574 = fadd float %1570, %1573, !dbg !26 + %1575 = bitcast float %1574 to i32, !dbg !29 + %1576 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1575, i32 1, i32 31), !dbg !29 + %1577 = bitcast i32 %1576 to float, !dbg !29 + %1578 = fadd float %1574, %1577, !dbg !26 + %1579 = bitcast float %1307 to i32, !dbg !29 + %1580 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1579, i32 8, i32 31), !dbg !29 + %1581 = bitcast i32 %1580 to float, !dbg !29 + %1582 = fadd float %1307, %1581, !dbg !26 + %1583 = bitcast float %1582 to i32, !dbg !29 + %1584 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1583, i32 4, i32 31), !dbg !29 + %1585 = bitcast i32 %1584 to float, !dbg !29 + %1586 = fadd float %1582, %1585, !dbg !26 + %1587 = bitcast float %1586 to i32, !dbg !29 + %1588 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1587, i32 2, i32 31), !dbg !29 + %1589 = bitcast i32 %1588 to float, !dbg !29 + %1590 = fadd float %1586, %1589, !dbg !26 + %1591 = bitcast float %1590 to i32, !dbg !29 + %1592 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1591, i32 1, i32 31), !dbg !29 + %1593 = bitcast i32 %1592 to float, !dbg !29 + %1594 = fadd float %1590, %1593, !dbg !26 + %1595 = bitcast float %1314 to i32, !dbg !29 + %1596 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1595, i32 8, i32 31), !dbg !29 + %1597 = bitcast i32 %1596 to float, !dbg !29 + %1598 = fadd float %1314, %1597, !dbg !26 + %1599 = bitcast float %1598 to i32, !dbg !29 + %1600 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1599, i32 4, i32 31), !dbg !29 + %1601 = bitcast i32 %1600 to float, !dbg !29 + %1602 = fadd float %1598, %1601, !dbg !26 + %1603 = bitcast float %1602 to i32, !dbg !29 + %1604 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1603, i32 2, i32 31), !dbg !29 + %1605 = bitcast i32 %1604 to float, !dbg !29 + %1606 = fadd float %1602, %1605, !dbg !26 + %1607 = bitcast float %1606 to i32, !dbg !29 + %1608 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1607, i32 1, i32 31), !dbg !29 + %1609 = bitcast i32 %1608 to float, !dbg !29 + %1610 = fadd float %1606, %1609, !dbg !26 + %1611 = bitcast float %1321 to i32, !dbg !29 + %1612 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1611, i32 8, i32 31), !dbg !29 + %1613 = bitcast i32 %1612 to float, !dbg !29 + %1614 = fadd float %1321, %1613, !dbg !26 + %1615 = bitcast float %1614 to i32, !dbg !29 + %1616 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1615, i32 4, i32 31), !dbg !29 + %1617 = bitcast i32 %1616 to float, !dbg !29 + %1618 = fadd float %1614, %1617, !dbg !26 + %1619 = bitcast float %1618 to i32, !dbg !29 + %1620 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1619, i32 2, i32 31), !dbg !29 + %1621 = bitcast i32 %1620 to float, !dbg !29 + %1622 = fadd float %1618, %1621, !dbg !26 + %1623 = bitcast float %1622 to i32, !dbg !29 + %1624 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1623, i32 1, i32 31), !dbg !29 + %1625 = bitcast i32 %1624 to float, !dbg !29 + %1626 = fadd float %1622, %1625, !dbg !26 + %1627 = bitcast float %1328 to i32, !dbg !29 + %1628 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1627, i32 8, i32 31), !dbg !29 + %1629 = bitcast i32 %1628 to float, !dbg !29 + %1630 = fadd float %1328, %1629, !dbg !26 + %1631 = bitcast float %1630 to i32, !dbg !29 + %1632 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1631, i32 4, i32 31), !dbg !29 + %1633 = bitcast i32 %1632 to float, !dbg !29 + %1634 = fadd float %1630, %1633, !dbg !26 + %1635 = bitcast float %1634 to i32, !dbg !29 + %1636 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1635, i32 2, i32 31), !dbg !29 + %1637 = bitcast i32 %1636 to float, !dbg !29 + %1638 = fadd float %1634, %1637, !dbg !26 + %1639 = bitcast float %1638 to i32, !dbg !29 + %1640 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1639, i32 1, i32 31), !dbg !29 + %1641 = bitcast i32 %1640 to float, !dbg !29 + %1642 = fadd float %1638, %1641, !dbg !26 + %1643 = bitcast float %1335 to i32, !dbg !29 + %1644 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1643, i32 8, i32 31), !dbg !29 + %1645 = bitcast i32 %1644 to float, !dbg !29 + %1646 = fadd float %1335, %1645, !dbg !26 + %1647 = bitcast float %1646 to i32, !dbg !29 + %1648 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1647, i32 4, i32 31), !dbg !29 + %1649 = bitcast i32 %1648 to float, !dbg !29 + %1650 = fadd float %1646, %1649, !dbg !26 + %1651 = bitcast float %1650 to i32, !dbg !29 + %1652 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1651, i32 2, i32 31), !dbg !29 + %1653 = bitcast i32 %1652 to float, !dbg !29 + %1654 = fadd float %1650, %1653, !dbg !26 + %1655 = bitcast float %1654 to i32, !dbg !29 + %1656 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1655, i32 1, i32 31), !dbg !29 + %1657 = bitcast i32 %1656 to float, !dbg !29 + %1658 = fadd float %1654, %1657, !dbg !26 + %1659 = bitcast float %1342 to i32, !dbg !29 + %1660 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1659, i32 8, i32 31), !dbg !29 + %1661 = bitcast i32 %1660 to float, !dbg !29 + %1662 = fadd float %1342, %1661, !dbg !26 + %1663 = bitcast float %1662 to i32, !dbg !29 + %1664 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1663, i32 4, i32 31), !dbg !29 + %1665 = bitcast i32 %1664 to float, !dbg !29 + %1666 = fadd float %1662, %1665, !dbg !26 + %1667 = bitcast float %1666 to i32, !dbg !29 + %1668 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1667, i32 2, i32 31), !dbg !29 + %1669 = bitcast i32 %1668 to float, !dbg !29 + %1670 = fadd float %1666, %1669, !dbg !26 + %1671 = bitcast float %1670 to i32, !dbg !29 + %1672 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1671, i32 1, i32 31), !dbg !29 + %1673 = bitcast i32 %1672 to float, !dbg !29 + %1674 = fadd float %1670, %1673, !dbg !26 + %1675 = bitcast float %1349 to i32, !dbg !29 + %1676 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1675, i32 8, i32 31), !dbg !29 + %1677 = bitcast i32 %1676 to float, !dbg !29 + %1678 = fadd float %1349, %1677, !dbg !26 + %1679 = bitcast float %1678 to i32, !dbg !29 + %1680 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1679, i32 4, i32 31), !dbg !29 + %1681 = bitcast i32 %1680 to float, !dbg !29 + %1682 = fadd float %1678, %1681, !dbg !26 + %1683 = bitcast float %1682 to i32, !dbg !29 + %1684 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1683, i32 2, i32 31), !dbg !29 + %1685 = bitcast i32 %1684 to float, !dbg !29 + %1686 = fadd float %1682, %1685, !dbg !26 + %1687 = bitcast float %1686 to i32, !dbg !29 + %1688 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1687, i32 1, i32 31), !dbg !29 + %1689 = bitcast i32 %1688 to float, !dbg !29 + %1690 = fadd float %1686, %1689, !dbg !26 + %1691 = bitcast float %1356 to i32, !dbg !29 + %1692 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1691, i32 8, i32 31), !dbg !29 + %1693 = bitcast i32 %1692 to float, !dbg !29 + %1694 = fadd float %1356, %1693, !dbg !26 + %1695 = bitcast float %1694 to i32, !dbg !29 + %1696 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1695, i32 4, i32 31), !dbg !29 + %1697 = bitcast i32 %1696 to float, !dbg !29 + %1698 = fadd float %1694, %1697, !dbg !26 + %1699 = bitcast float %1698 to i32, !dbg !29 + %1700 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1699, i32 2, i32 31), !dbg !29 + %1701 = bitcast i32 %1700 to float, !dbg !29 + %1702 = fadd float %1698, %1701, !dbg !26 + %1703 = bitcast float %1702 to i32, !dbg !29 + %1704 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1703, i32 1, i32 31), !dbg !29 + %1705 = bitcast i32 %1704 to float, !dbg !29 + %1706 = fadd float %1702, %1705, !dbg !26 + %1707 = bitcast float %1363 to i32, !dbg !29 + %1708 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1707, i32 8, i32 31), !dbg !29 + %1709 = bitcast i32 %1708 to float, !dbg !29 + %1710 = fadd float %1363, %1709, !dbg !26 + %1711 = bitcast float %1710 to i32, !dbg !29 + %1712 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1711, i32 4, i32 31), !dbg !29 + %1713 = bitcast i32 %1712 to float, !dbg !29 + %1714 = fadd float %1710, %1713, !dbg !26 + %1715 = bitcast float %1714 to i32, !dbg !29 + %1716 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1715, i32 2, i32 31), !dbg !29 + %1717 = bitcast i32 %1716 to float, !dbg !29 + %1718 = fadd float %1714, %1717, !dbg !26 + %1719 = bitcast float %1718 to i32, !dbg !29 + %1720 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1719, i32 1, i32 31), !dbg !29 + %1721 = bitcast i32 %1720 to float, !dbg !29 + %1722 = fadd float %1718, %1721, !dbg !26 + %1723 = bitcast float %1370 to i32, !dbg !29 + %1724 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1723, i32 8, i32 31), !dbg !29 + %1725 = bitcast i32 %1724 to float, !dbg !29 + %1726 = fadd float %1370, %1725, !dbg !26 + %1727 = bitcast float %1726 to i32, !dbg !29 + %1728 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1727, i32 4, i32 31), !dbg !29 + %1729 = bitcast i32 %1728 to float, !dbg !29 + %1730 = fadd float %1726, %1729, !dbg !26 + %1731 = bitcast float %1730 to i32, !dbg !29 + %1732 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1731, i32 2, i32 31), !dbg !29 + %1733 = bitcast i32 %1732 to float, !dbg !29 + %1734 = fadd float %1730, %1733, !dbg !26 + %1735 = bitcast float %1734 to i32, !dbg !29 + %1736 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1735, i32 1, i32 31), !dbg !29 + %1737 = bitcast i32 %1736 to float, !dbg !29 + %1738 = fadd float %1734, %1737, !dbg !26 + %1739 = bitcast float %1377 to i32, !dbg !29 + %1740 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1739, i32 8, i32 31), !dbg !29 + %1741 = bitcast i32 %1740 to float, !dbg !29 + %1742 = fadd float %1377, %1741, !dbg !26 + %1743 = bitcast float %1742 to i32, !dbg !29 + %1744 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1743, i32 4, i32 31), !dbg !29 + %1745 = bitcast i32 %1744 to float, !dbg !29 + %1746 = fadd float %1742, %1745, !dbg !26 + %1747 = bitcast float %1746 to i32, !dbg !29 + %1748 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1747, i32 2, i32 31), !dbg !29 + %1749 = bitcast i32 %1748 to float, !dbg !29 + %1750 = fadd float %1746, %1749, !dbg !26 + %1751 = bitcast float %1750 to i32, !dbg !29 + %1752 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1751, i32 1, i32 31), !dbg !29 + %1753 = bitcast i32 %1752 to float, !dbg !29 + %1754 = fadd float %1750, %1753, !dbg !26 + %1755 = bitcast float %1384 to i32, !dbg !29 + %1756 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1755, i32 8, i32 31), !dbg !29 + %1757 = bitcast i32 %1756 to float, !dbg !29 + %1758 = fadd float %1384, %1757, !dbg !26 + %1759 = bitcast float %1758 to i32, !dbg !29 + %1760 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1759, i32 4, i32 31), !dbg !29 + %1761 = bitcast i32 %1760 to float, !dbg !29 + %1762 = fadd float %1758, %1761, !dbg !26 + %1763 = bitcast float %1762 to i32, !dbg !29 + %1764 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1763, i32 2, i32 31), !dbg !29 + %1765 = bitcast i32 %1764 to float, !dbg !29 + %1766 = fadd float %1762, %1765, !dbg !26 + %1767 = bitcast float %1766 to i32, !dbg !29 + %1768 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1767, i32 1, i32 31), !dbg !29 + %1769 = bitcast i32 %1768 to float, !dbg !29 + %1770 = fadd float %1766, %1769, !dbg !26 + %1771 = bitcast float %1391 to i32, !dbg !29 + %1772 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1771, i32 8, i32 31), !dbg !29 + %1773 = bitcast i32 %1772 to float, !dbg !29 + %1774 = fadd float %1391, %1773, !dbg !26 + %1775 = bitcast float %1774 to i32, !dbg !29 + %1776 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1775, i32 4, i32 31), !dbg !29 + %1777 = bitcast i32 %1776 to float, !dbg !29 + %1778 = fadd float %1774, %1777, !dbg !26 + %1779 = bitcast float %1778 to i32, !dbg !29 + %1780 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1779, i32 2, i32 31), !dbg !29 + %1781 = bitcast i32 %1780 to float, !dbg !29 + %1782 = fadd float %1778, %1781, !dbg !26 + %1783 = bitcast float %1782 to i32, !dbg !29 + %1784 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1783, i32 1, i32 31), !dbg !29 + %1785 = bitcast i32 %1784 to float, !dbg !29 + %1786 = fadd float %1782, %1785, !dbg !26 + %1787 = bitcast float %1398 to i32, !dbg !29 + %1788 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1787, i32 8, i32 31), !dbg !29 + %1789 = bitcast i32 %1788 to float, !dbg !29 + %1790 = fadd float %1398, %1789, !dbg !26 + %1791 = bitcast float %1790 to i32, !dbg !29 + %1792 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1791, i32 4, i32 31), !dbg !29 + %1793 = bitcast i32 %1792 to float, !dbg !29 + %1794 = fadd float %1790, %1793, !dbg !26 + %1795 = bitcast float %1794 to i32, !dbg !29 + %1796 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1795, i32 2, i32 31), !dbg !29 + %1797 = bitcast i32 %1796 to float, !dbg !29 + %1798 = fadd float %1794, %1797, !dbg !26 + %1799 = bitcast float %1798 to i32, !dbg !29 + %1800 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1799, i32 1, i32 31), !dbg !29 + %1801 = bitcast i32 %1800 to float, !dbg !29 + %1802 = fadd float %1798, %1801, !dbg !26 + %1803 = bitcast float %1405 to i32, !dbg !29 + %1804 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1803, i32 8, i32 31), !dbg !29 + %1805 = bitcast i32 %1804 to float, !dbg !29 + %1806 = fadd float %1405, %1805, !dbg !26 + %1807 = bitcast float %1806 to i32, !dbg !29 + %1808 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1807, i32 4, i32 31), !dbg !29 + %1809 = bitcast i32 %1808 to float, !dbg !29 + %1810 = fadd float %1806, %1809, !dbg !26 + %1811 = bitcast float %1810 to i32, !dbg !29 + %1812 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1811, i32 2, i32 31), !dbg !29 + %1813 = bitcast i32 %1812 to float, !dbg !29 + %1814 = fadd float %1810, %1813, !dbg !26 + %1815 = bitcast float %1814 to i32, !dbg !29 + %1816 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1815, i32 1, i32 31), !dbg !29 + %1817 = bitcast i32 %1816 to float, !dbg !29 + %1818 = fadd float %1814, %1817, !dbg !26 + %1819 = bitcast float %1412 to i32, !dbg !29 + %1820 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1819, i32 8, i32 31), !dbg !29 + %1821 = bitcast i32 %1820 to float, !dbg !29 + %1822 = fadd float %1412, %1821, !dbg !26 + %1823 = bitcast float %1822 to i32, !dbg !29 + %1824 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1823, i32 4, i32 31), !dbg !29 + %1825 = bitcast i32 %1824 to float, !dbg !29 + %1826 = fadd float %1822, %1825, !dbg !26 + %1827 = bitcast float %1826 to i32, !dbg !29 + %1828 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1827, i32 2, i32 31), !dbg !29 + %1829 = bitcast i32 %1828 to float, !dbg !29 + %1830 = fadd float %1826, %1829, !dbg !26 + %1831 = bitcast float %1830 to i32, !dbg !29 + %1832 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1831, i32 1, i32 31), !dbg !29 + %1833 = bitcast i32 %1832 to float, !dbg !29 + %1834 = fadd float %1830, %1833, !dbg !26 + %1835 = bitcast float %1419 to i32, !dbg !29 + %1836 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1835, i32 8, i32 31), !dbg !29 + %1837 = bitcast i32 %1836 to float, !dbg !29 + %1838 = fadd float %1419, %1837, !dbg !26 + %1839 = bitcast float %1838 to i32, !dbg !29 + %1840 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1839, i32 4, i32 31), !dbg !29 + %1841 = bitcast i32 %1840 to float, !dbg !29 + %1842 = fadd float %1838, %1841, !dbg !26 + %1843 = bitcast float %1842 to i32, !dbg !29 + %1844 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1843, i32 2, i32 31), !dbg !29 + %1845 = bitcast i32 %1844 to float, !dbg !29 + %1846 = fadd float %1842, %1845, !dbg !26 + %1847 = bitcast float %1846 to i32, !dbg !29 + %1848 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1847, i32 1, i32 31), !dbg !29 + %1849 = bitcast i32 %1848 to float, !dbg !29 + %1850 = fadd float %1846, %1849, !dbg !26 + %1851 = bitcast float %1426 to i32, !dbg !29 + %1852 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1851, i32 8, i32 31), !dbg !29 + %1853 = bitcast i32 %1852 to float, !dbg !29 + %1854 = fadd float %1426, %1853, !dbg !26 + %1855 = bitcast float %1854 to i32, !dbg !29 + %1856 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1855, i32 4, i32 31), !dbg !29 + %1857 = bitcast i32 %1856 to float, !dbg !29 + %1858 = fadd float %1854, %1857, !dbg !26 + %1859 = bitcast float %1858 to i32, !dbg !29 + %1860 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1859, i32 2, i32 31), !dbg !29 + %1861 = bitcast i32 %1860 to float, !dbg !29 + %1862 = fadd float %1858, %1861, !dbg !26 + %1863 = bitcast float %1862 to i32, !dbg !29 + %1864 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1863, i32 1, i32 31), !dbg !29 + %1865 = bitcast i32 %1864 to float, !dbg !29 + %1866 = fadd float %1862, %1865, !dbg !26 + %1867 = bitcast float %1433 to i32, !dbg !29 + %1868 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1867, i32 8, i32 31), !dbg !29 + %1869 = bitcast i32 %1868 to float, !dbg !29 + %1870 = fadd float %1433, %1869, !dbg !26 + %1871 = bitcast float %1870 to i32, !dbg !29 + %1872 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1871, i32 4, i32 31), !dbg !29 + %1873 = bitcast i32 %1872 to float, !dbg !29 + %1874 = fadd float %1870, %1873, !dbg !26 + %1875 = bitcast float %1874 to i32, !dbg !29 + %1876 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1875, i32 2, i32 31), !dbg !29 + %1877 = bitcast i32 %1876 to float, !dbg !29 + %1878 = fadd float %1874, %1877, !dbg !26 + %1879 = bitcast float %1878 to i32, !dbg !29 + %1880 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1879, i32 1, i32 31), !dbg !29 + %1881 = bitcast i32 %1880 to float, !dbg !29 + %1882 = fadd float %1878, %1881, !dbg !26 + %1883 = bitcast float %1440 to i32, !dbg !29 + %1884 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1883, i32 8, i32 31), !dbg !29 + %1885 = bitcast i32 %1884 to float, !dbg !29 + %1886 = fadd float %1440, %1885, !dbg !26 + %1887 = bitcast float %1886 to i32, !dbg !29 + %1888 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1887, i32 4, i32 31), !dbg !29 + %1889 = bitcast i32 %1888 to float, !dbg !29 + %1890 = fadd float %1886, %1889, !dbg !26 + %1891 = bitcast float %1890 to i32, !dbg !29 + %1892 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1891, i32 2, i32 31), !dbg !29 + %1893 = bitcast i32 %1892 to float, !dbg !29 + %1894 = fadd float %1890, %1893, !dbg !26 + %1895 = bitcast float %1894 to i32, !dbg !29 + %1896 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1895, i32 1, i32 31), !dbg !29 + %1897 = bitcast i32 %1896 to float, !dbg !29 + %1898 = fadd float %1894, %1897, !dbg !26 + %1899 = bitcast float %1447 to i32, !dbg !29 + %1900 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1899, i32 8, i32 31), !dbg !29 + %1901 = bitcast i32 %1900 to float, !dbg !29 + %1902 = fadd float %1447, %1901, !dbg !26 + %1903 = bitcast float %1902 to i32, !dbg !29 + %1904 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1903, i32 4, i32 31), !dbg !29 + %1905 = bitcast i32 %1904 to float, !dbg !29 + %1906 = fadd float %1902, %1905, !dbg !26 + %1907 = bitcast float %1906 to i32, !dbg !29 + %1908 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1907, i32 2, i32 31), !dbg !29 + %1909 = bitcast i32 %1908 to float, !dbg !29 + %1910 = fadd float %1906, %1909, !dbg !26 + %1911 = bitcast float %1910 to i32, !dbg !29 + %1912 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1911, i32 1, i32 31), !dbg !29 + %1913 = bitcast i32 %1912 to float, !dbg !29 + %1914 = fadd float %1910, %1913, !dbg !26 + %1915 = bitcast float %1454 to i32, !dbg !29 + %1916 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1915, i32 8, i32 31), !dbg !29 + %1917 = bitcast i32 %1916 to float, !dbg !29 + %1918 = fadd float %1454, %1917, !dbg !26 + %1919 = bitcast float %1918 to i32, !dbg !29 + %1920 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1919, i32 4, i32 31), !dbg !29 + %1921 = bitcast i32 %1920 to float, !dbg !29 + %1922 = fadd float %1918, %1921, !dbg !26 + %1923 = bitcast float %1922 to i32, !dbg !29 + %1924 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1923, i32 2, i32 31), !dbg !29 + %1925 = bitcast i32 %1924 to float, !dbg !29 + %1926 = fadd float %1922, %1925, !dbg !26 + %1927 = bitcast float %1926 to i32, !dbg !29 + %1928 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1927, i32 1, i32 31), !dbg !29 + %1929 = bitcast i32 %1928 to float, !dbg !29 + %1930 = fadd float %1926, %1929, !dbg !26 + %1931 = bitcast float %1461 to i32, !dbg !29 + %1932 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1931, i32 8, i32 31), !dbg !29 + %1933 = bitcast i32 %1932 to float, !dbg !29 + %1934 = fadd float %1461, %1933, !dbg !26 + %1935 = bitcast float %1934 to i32, !dbg !29 + %1936 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1935, i32 4, i32 31), !dbg !29 + %1937 = bitcast i32 %1936 to float, !dbg !29 + %1938 = fadd float %1934, %1937, !dbg !26 + %1939 = bitcast float %1938 to i32, !dbg !29 + %1940 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1939, i32 2, i32 31), !dbg !29 + %1941 = bitcast i32 %1940 to float, !dbg !29 + %1942 = fadd float %1938, %1941, !dbg !26 + %1943 = bitcast float %1942 to i32, !dbg !29 + %1944 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1943, i32 1, i32 31), !dbg !29 + %1945 = bitcast i32 %1944 to float, !dbg !29 + %1946 = fadd float %1942, %1945, !dbg !26 + %1947 = bitcast float %1468 to i32, !dbg !29 + %1948 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1947, i32 8, i32 31), !dbg !29 + %1949 = bitcast i32 %1948 to float, !dbg !29 + %1950 = fadd float %1468, %1949, !dbg !26 + %1951 = bitcast float %1950 to i32, !dbg !29 + %1952 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1951, i32 4, i32 31), !dbg !29 + %1953 = bitcast i32 %1952 to float, !dbg !29 + %1954 = fadd float %1950, %1953, !dbg !26 + %1955 = bitcast float %1954 to i32, !dbg !29 + %1956 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1955, i32 2, i32 31), !dbg !29 + %1957 = bitcast i32 %1956 to float, !dbg !29 + %1958 = fadd float %1954, %1957, !dbg !26 + %1959 = bitcast float %1958 to i32, !dbg !29 + %1960 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1959, i32 1, i32 31), !dbg !29 + %1961 = bitcast i32 %1960 to float, !dbg !29 + %1962 = fadd float %1958, %1961, !dbg !26 + %1963 = bitcast float %1475 to i32, !dbg !29 + %1964 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1963, i32 8, i32 31), !dbg !29 + %1965 = bitcast i32 %1964 to float, !dbg !29 + %1966 = fadd float %1475, %1965, !dbg !26 + %1967 = bitcast float %1966 to i32, !dbg !29 + %1968 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1967, i32 4, i32 31), !dbg !29 + %1969 = bitcast i32 %1968 to float, !dbg !29 + %1970 = fadd float %1966, %1969, !dbg !26 + %1971 = bitcast float %1970 to i32, !dbg !29 + %1972 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1971, i32 2, i32 31), !dbg !29 + %1973 = bitcast i32 %1972 to float, !dbg !29 + %1974 = fadd float %1970, %1973, !dbg !26 + %1975 = bitcast float %1974 to i32, !dbg !29 + %1976 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1975, i32 1, i32 31), !dbg !29 + %1977 = bitcast i32 %1976 to float, !dbg !29 + %1978 = fadd float %1974, %1977, !dbg !26 + %1979 = bitcast float %1482 to i32, !dbg !29 + %1980 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1979, i32 8, i32 31), !dbg !29 + %1981 = bitcast i32 %1980 to float, !dbg !29 + %1982 = fadd float %1482, %1981, !dbg !26 + %1983 = bitcast float %1982 to i32, !dbg !29 + %1984 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1983, i32 4, i32 31), !dbg !29 + %1985 = bitcast i32 %1984 to float, !dbg !29 + %1986 = fadd float %1982, %1985, !dbg !26 + %1987 = bitcast float %1986 to i32, !dbg !29 + %1988 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1987, i32 2, i32 31), !dbg !29 + %1989 = bitcast i32 %1988 to float, !dbg !29 + %1990 = fadd float %1986, %1989, !dbg !26 + %1991 = bitcast float %1990 to i32, !dbg !29 + %1992 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1991, i32 1, i32 31), !dbg !29 + %1993 = bitcast i32 %1992 to float, !dbg !29 + %1994 = fadd float %1990, %1993, !dbg !26 + %1995 = fadd float %3, %1498, !dbg !32 + %1996 = fadd float %3, %1514, !dbg !32 + %1997 = fadd float %3, %1530, !dbg !32 + %1998 = fadd float %3, %1546, !dbg !32 + %1999 = fadd float %3, %1562, !dbg !32 + %2000 = fadd float %3, %1578, !dbg !32 + %2001 = fadd float %3, %1594, !dbg !32 + %2002 = fadd float %3, %1610, !dbg !32 + %2003 = fadd float %3, %1626, !dbg !32 + %2004 = fadd float %3, %1642, !dbg !32 + %2005 = fadd float %3, %1658, !dbg !32 + %2006 = fadd float %3, %1674, !dbg !32 + %2007 = fadd float %3, %1690, !dbg !32 + %2008 = fadd float %3, %1706, !dbg !32 + %2009 = fadd float %3, %1722, !dbg !32 + %2010 = fadd float %3, %1738, !dbg !32 + %2011 = fadd float %3, %1754, !dbg !32 + %2012 = fadd float %3, %1770, !dbg !32 + %2013 = fadd float %3, %1786, !dbg !32 + %2014 = fadd float %3, %1802, !dbg !32 + %2015 = fadd float %3, %1818, !dbg !32 + %2016 = fadd float %3, %1834, !dbg !32 + %2017 = fadd float %3, %1850, !dbg !32 + %2018 = fadd float %3, %1866, !dbg !32 + %2019 = fadd float %3, %1882, !dbg !32 + %2020 = fadd float %3, %1898, !dbg !32 + %2021 = fadd float %3, %1914, !dbg !32 + %2022 = fadd float %3, %1930, !dbg !32 + %2023 = fadd float %3, %1946, !dbg !32 + %2024 = fadd float %3, %1962, !dbg !32 + %2025 = fadd float %3, %1978, !dbg !32 + %2026 = fadd float %3, %1994, !dbg !32 + %2027 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i = icmp eq i32 %2027, 0, !dbg !33 + %2028 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i = icmp eq i32 %2028, 0, !dbg !33 + br i1 %.not.i, label %2034, label %2029, !dbg !33 + +2029: ; preds = %7 + br i1 %.not1.i, label %2032, label %2030, !dbg !33 + +2030: ; preds = %2029 + %2031 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1995) #7, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +2032: ; preds = %2029 + %2033 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1995) #7, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +2034: ; preds = %7 + br i1 %.not1.i, label %2037, label %2035, !dbg !33 + +2035: ; preds = %2034 + %2036 = tail call float @llvm.nvvm.sqrt.rn.f(float %1995) #7, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +2037: ; preds = %2034 + %2038 = tail call float @llvm.nvvm.sqrt.approx.f(float %1995) #7, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +__nv_sqrtf.exit: ; preds = %2030, %2032, %2035, %2037 + %.0.i = phi float [ %2031, %2030 ], [ %2033, %2032 ], [ %2036, %2035 ], [ %2038, %2037 ], !dbg !33 + %2039 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i1 = icmp eq i32 %2039, 0, !dbg !33 + %2040 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i4 = icmp eq i32 %2040, 0, !dbg !33 + br i1 %.not.i1, label %2046, label %2041, !dbg !33 + +2041: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %2044, label %2042, !dbg !33 + +2042: ; preds = %2041 + %2043 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1996) #7, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +2044: ; preds = %2041 + %2045 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1996) #7, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +2046: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %2049, label %2047, !dbg !33 + +2047: ; preds = %2046 + %2048 = tail call float @llvm.nvvm.sqrt.rn.f(float %1996) #7, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +2049: ; preds = %2046 + %2050 = tail call float @llvm.nvvm.sqrt.approx.f(float %1996) #7, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +__nv_sqrtf.exit5: ; preds = %2042, %2044, %2047, %2049 + %.0.i3 = phi float [ %2043, %2042 ], [ %2045, %2044 ], [ %2048, %2047 ], [ %2050, %2049 ], !dbg !33 + %2051 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i6 = icmp eq i32 %2051, 0, !dbg !33 + %2052 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i9 = icmp eq i32 %2052, 0, !dbg !33 + br i1 %.not.i6, label %2058, label %2053, !dbg !33 + +2053: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %2056, label %2054, !dbg !33 + +2054: ; preds = %2053 + %2055 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1997) #7, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +2056: ; preds = %2053 + %2057 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1997) #7, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +2058: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %2061, label %2059, !dbg !33 + +2059: ; preds = %2058 + %2060 = tail call float @llvm.nvvm.sqrt.rn.f(float %1997) #7, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +2061: ; preds = %2058 + %2062 = tail call float @llvm.nvvm.sqrt.approx.f(float %1997) #7, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +__nv_sqrtf.exit10: ; preds = %2054, %2056, %2059, %2061 + %.0.i8 = phi float [ %2055, %2054 ], [ %2057, %2056 ], [ %2060, %2059 ], [ %2062, %2061 ], !dbg !33 + %2063 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i11 = icmp eq i32 %2063, 0, !dbg !33 + %2064 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i14 = icmp eq i32 %2064, 0, !dbg !33 + br i1 %.not.i11, label %2070, label %2065, !dbg !33 + +2065: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %2068, label %2066, !dbg !33 + +2066: ; preds = %2065 + %2067 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1998) #7, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +2068: ; preds = %2065 + %2069 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1998) #7, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +2070: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %2073, label %2071, !dbg !33 + +2071: ; preds = %2070 + %2072 = tail call float @llvm.nvvm.sqrt.rn.f(float %1998) #7, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +2073: ; preds = %2070 + %2074 = tail call float @llvm.nvvm.sqrt.approx.f(float %1998) #7, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +__nv_sqrtf.exit15: ; preds = %2066, %2068, %2071, %2073 + %.0.i13 = phi float [ %2067, %2066 ], [ %2069, %2068 ], [ %2072, %2071 ], [ %2074, %2073 ], !dbg !33 + %2075 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i16 = icmp eq i32 %2075, 0, !dbg !33 + %2076 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i19 = icmp eq i32 %2076, 0, !dbg !33 + br i1 %.not.i16, label %2082, label %2077, !dbg !33 + +2077: ; preds = %__nv_sqrtf.exit15 + br i1 %.not1.i19, label %2080, label %2078, !dbg !33 + +2078: ; preds = %2077 + %2079 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1999) #7, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +2080: ; preds = %2077 + %2081 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1999) #7, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +2082: ; preds = %__nv_sqrtf.exit15 + br i1 %.not1.i19, label %2085, label %2083, !dbg !33 + +2083: ; preds = %2082 + %2084 = tail call float @llvm.nvvm.sqrt.rn.f(float %1999) #7, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +2085: ; preds = %2082 + %2086 = tail call float @llvm.nvvm.sqrt.approx.f(float %1999) #7, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +__nv_sqrtf.exit20: ; preds = %2078, %2080, %2083, %2085 + %.0.i18 = phi float [ %2079, %2078 ], [ %2081, %2080 ], [ %2084, %2083 ], [ %2086, %2085 ], !dbg !33 + %2087 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i21 = icmp eq i32 %2087, 0, !dbg !33 + %2088 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i24 = icmp eq i32 %2088, 0, !dbg !33 + br i1 %.not.i21, label %2094, label %2089, !dbg !33 + +2089: ; preds = %__nv_sqrtf.exit20 + br i1 %.not1.i24, label %2092, label %2090, !dbg !33 + +2090: ; preds = %2089 + %2091 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2000) #7, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +2092: ; preds = %2089 + %2093 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2000) #7, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +2094: ; preds = %__nv_sqrtf.exit20 + br i1 %.not1.i24, label %2097, label %2095, !dbg !33 + +2095: ; preds = %2094 + %2096 = tail call float @llvm.nvvm.sqrt.rn.f(float %2000) #7, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +2097: ; preds = %2094 + %2098 = tail call float @llvm.nvvm.sqrt.approx.f(float %2000) #7, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +__nv_sqrtf.exit25: ; preds = %2090, %2092, %2095, %2097 + %.0.i23 = phi float [ %2091, %2090 ], [ %2093, %2092 ], [ %2096, %2095 ], [ %2098, %2097 ], !dbg !33 + %2099 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i26 = icmp eq i32 %2099, 0, !dbg !33 + %2100 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i29 = icmp eq i32 %2100, 0, !dbg !33 + br i1 %.not.i26, label %2106, label %2101, !dbg !33 + +2101: ; preds = %__nv_sqrtf.exit25 + br i1 %.not1.i29, label %2104, label %2102, !dbg !33 + +2102: ; preds = %2101 + %2103 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2001) #7, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +2104: ; preds = %2101 + %2105 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2001) #7, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +2106: ; preds = %__nv_sqrtf.exit25 + br i1 %.not1.i29, label %2109, label %2107, !dbg !33 + +2107: ; preds = %2106 + %2108 = tail call float @llvm.nvvm.sqrt.rn.f(float %2001) #7, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +2109: ; preds = %2106 + %2110 = tail call float @llvm.nvvm.sqrt.approx.f(float %2001) #7, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +__nv_sqrtf.exit30: ; preds = %2102, %2104, %2107, %2109 + %.0.i28 = phi float [ %2103, %2102 ], [ %2105, %2104 ], [ %2108, %2107 ], [ %2110, %2109 ], !dbg !33 + %2111 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i31 = icmp eq i32 %2111, 0, !dbg !33 + %2112 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i34 = icmp eq i32 %2112, 0, !dbg !33 + br i1 %.not.i31, label %2118, label %2113, !dbg !33 + +2113: ; preds = %__nv_sqrtf.exit30 + br i1 %.not1.i34, label %2116, label %2114, !dbg !33 + +2114: ; preds = %2113 + %2115 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2002) #7, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +2116: ; preds = %2113 + %2117 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2002) #7, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +2118: ; preds = %__nv_sqrtf.exit30 + br i1 %.not1.i34, label %2121, label %2119, !dbg !33 + +2119: ; preds = %2118 + %2120 = tail call float @llvm.nvvm.sqrt.rn.f(float %2002) #7, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +2121: ; preds = %2118 + %2122 = tail call float @llvm.nvvm.sqrt.approx.f(float %2002) #7, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +__nv_sqrtf.exit35: ; preds = %2114, %2116, %2119, %2121 + %.0.i33 = phi float [ %2115, %2114 ], [ %2117, %2116 ], [ %2120, %2119 ], [ %2122, %2121 ], !dbg !33 + %2123 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i36 = icmp eq i32 %2123, 0, !dbg !33 + %2124 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i39 = icmp eq i32 %2124, 0, !dbg !33 + br i1 %.not.i36, label %2130, label %2125, !dbg !33 + +2125: ; preds = %__nv_sqrtf.exit35 + br i1 %.not1.i39, label %2128, label %2126, !dbg !33 + +2126: ; preds = %2125 + %2127 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2003) #7, !dbg !33 + br label %__nv_sqrtf.exit40, !dbg !33 + +2128: ; preds = %2125 + %2129 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2003) #7, !dbg !33 + br label %__nv_sqrtf.exit40, !dbg !33 + +2130: ; preds = %__nv_sqrtf.exit35 + br i1 %.not1.i39, label %2133, label %2131, !dbg !33 + +2131: ; preds = %2130 + %2132 = tail call float @llvm.nvvm.sqrt.rn.f(float %2003) #7, !dbg !33 + br label %__nv_sqrtf.exit40, !dbg !33 + +2133: ; preds = %2130 + %2134 = tail call float @llvm.nvvm.sqrt.approx.f(float %2003) #7, !dbg !33 + br label %__nv_sqrtf.exit40, !dbg !33 + +__nv_sqrtf.exit40: ; preds = %2126, %2128, %2131, %2133 + %.0.i38 = phi float [ %2127, %2126 ], [ %2129, %2128 ], [ %2132, %2131 ], [ %2134, %2133 ], !dbg !33 + %2135 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i41 = icmp eq i32 %2135, 0, !dbg !33 + %2136 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i44 = icmp eq i32 %2136, 0, !dbg !33 + br i1 %.not.i41, label %2142, label %2137, !dbg !33 + +2137: ; preds = %__nv_sqrtf.exit40 + br i1 %.not1.i44, label %2140, label %2138, !dbg !33 + +2138: ; preds = %2137 + %2139 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2004) #7, !dbg !33 + br label %__nv_sqrtf.exit45, !dbg !33 + +2140: ; preds = %2137 + %2141 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2004) #7, !dbg !33 + br label %__nv_sqrtf.exit45, !dbg !33 + +2142: ; preds = %__nv_sqrtf.exit40 + br i1 %.not1.i44, label %2145, label %2143, !dbg !33 + +2143: ; preds = %2142 + %2144 = tail call float @llvm.nvvm.sqrt.rn.f(float %2004) #7, !dbg !33 + br label %__nv_sqrtf.exit45, !dbg !33 + +2145: ; preds = %2142 + %2146 = tail call float @llvm.nvvm.sqrt.approx.f(float %2004) #7, !dbg !33 + br label %__nv_sqrtf.exit45, !dbg !33 + +__nv_sqrtf.exit45: ; preds = %2138, %2140, %2143, %2145 + %.0.i43 = phi float [ %2139, %2138 ], [ %2141, %2140 ], [ %2144, %2143 ], [ %2146, %2145 ], !dbg !33 + %2147 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i46 = icmp eq i32 %2147, 0, !dbg !33 + %2148 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i49 = icmp eq i32 %2148, 0, !dbg !33 + br i1 %.not.i46, label %2154, label %2149, !dbg !33 + +2149: ; preds = %__nv_sqrtf.exit45 + br i1 %.not1.i49, label %2152, label %2150, !dbg !33 + +2150: ; preds = %2149 + %2151 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2005) #7, !dbg !33 + br label %__nv_sqrtf.exit50, !dbg !33 + +2152: ; preds = %2149 + %2153 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2005) #7, !dbg !33 + br label %__nv_sqrtf.exit50, !dbg !33 + +2154: ; preds = %__nv_sqrtf.exit45 + br i1 %.not1.i49, label %2157, label %2155, !dbg !33 + +2155: ; preds = %2154 + %2156 = tail call float @llvm.nvvm.sqrt.rn.f(float %2005) #7, !dbg !33 + br label %__nv_sqrtf.exit50, !dbg !33 + +2157: ; preds = %2154 + %2158 = tail call float @llvm.nvvm.sqrt.approx.f(float %2005) #7, !dbg !33 + br label %__nv_sqrtf.exit50, !dbg !33 + +__nv_sqrtf.exit50: ; preds = %2150, %2152, %2155, %2157 + %.0.i48 = phi float [ %2151, %2150 ], [ %2153, %2152 ], [ %2156, %2155 ], [ %2158, %2157 ], !dbg !33 + %2159 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i51 = icmp eq i32 %2159, 0, !dbg !33 + %2160 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i54 = icmp eq i32 %2160, 0, !dbg !33 + br i1 %.not.i51, label %2166, label %2161, !dbg !33 + +2161: ; preds = %__nv_sqrtf.exit50 + br i1 %.not1.i54, label %2164, label %2162, !dbg !33 + +2162: ; preds = %2161 + %2163 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2006) #7, !dbg !33 + br label %__nv_sqrtf.exit55, !dbg !33 + +2164: ; preds = %2161 + %2165 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2006) #7, !dbg !33 + br label %__nv_sqrtf.exit55, !dbg !33 + +2166: ; preds = %__nv_sqrtf.exit50 + br i1 %.not1.i54, label %2169, label %2167, !dbg !33 + +2167: ; preds = %2166 + %2168 = tail call float @llvm.nvvm.sqrt.rn.f(float %2006) #7, !dbg !33 + br label %__nv_sqrtf.exit55, !dbg !33 + +2169: ; preds = %2166 + %2170 = tail call float @llvm.nvvm.sqrt.approx.f(float %2006) #7, !dbg !33 + br label %__nv_sqrtf.exit55, !dbg !33 + +__nv_sqrtf.exit55: ; preds = %2162, %2164, %2167, %2169 + %.0.i53 = phi float [ %2163, %2162 ], [ %2165, %2164 ], [ %2168, %2167 ], [ %2170, %2169 ], !dbg !33 + %2171 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i56 = icmp eq i32 %2171, 0, !dbg !33 + %2172 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i59 = icmp eq i32 %2172, 0, !dbg !33 + br i1 %.not.i56, label %2178, label %2173, !dbg !33 + +2173: ; preds = %__nv_sqrtf.exit55 + br i1 %.not1.i59, label %2176, label %2174, !dbg !33 + +2174: ; preds = %2173 + %2175 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2007) #7, !dbg !33 + br label %__nv_sqrtf.exit60, !dbg !33 + +2176: ; preds = %2173 + %2177 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2007) #7, !dbg !33 + br label %__nv_sqrtf.exit60, !dbg !33 + +2178: ; preds = %__nv_sqrtf.exit55 + br i1 %.not1.i59, label %2181, label %2179, !dbg !33 + +2179: ; preds = %2178 + %2180 = tail call float @llvm.nvvm.sqrt.rn.f(float %2007) #7, !dbg !33 + br label %__nv_sqrtf.exit60, !dbg !33 + +2181: ; preds = %2178 + %2182 = tail call float @llvm.nvvm.sqrt.approx.f(float %2007) #7, !dbg !33 + br label %__nv_sqrtf.exit60, !dbg !33 + +__nv_sqrtf.exit60: ; preds = %2174, %2176, %2179, %2181 + %.0.i58 = phi float [ %2175, %2174 ], [ %2177, %2176 ], [ %2180, %2179 ], [ %2182, %2181 ], !dbg !33 + %2183 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i61 = icmp eq i32 %2183, 0, !dbg !33 + %2184 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i64 = icmp eq i32 %2184, 0, !dbg !33 + br i1 %.not.i61, label %2190, label %2185, !dbg !33 + +2185: ; preds = %__nv_sqrtf.exit60 + br i1 %.not1.i64, label %2188, label %2186, !dbg !33 + +2186: ; preds = %2185 + %2187 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2008) #7, !dbg !33 + br label %__nv_sqrtf.exit65, !dbg !33 + +2188: ; preds = %2185 + %2189 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2008) #7, !dbg !33 + br label %__nv_sqrtf.exit65, !dbg !33 + +2190: ; preds = %__nv_sqrtf.exit60 + br i1 %.not1.i64, label %2193, label %2191, !dbg !33 + +2191: ; preds = %2190 + %2192 = tail call float @llvm.nvvm.sqrt.rn.f(float %2008) #7, !dbg !33 + br label %__nv_sqrtf.exit65, !dbg !33 + +2193: ; preds = %2190 + %2194 = tail call float @llvm.nvvm.sqrt.approx.f(float %2008) #7, !dbg !33 + br label %__nv_sqrtf.exit65, !dbg !33 + +__nv_sqrtf.exit65: ; preds = %2186, %2188, %2191, %2193 + %.0.i63 = phi float [ %2187, %2186 ], [ %2189, %2188 ], [ %2192, %2191 ], [ %2194, %2193 ], !dbg !33 + %2195 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i66 = icmp eq i32 %2195, 0, !dbg !33 + %2196 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i69 = icmp eq i32 %2196, 0, !dbg !33 + br i1 %.not.i66, label %2202, label %2197, !dbg !33 + +2197: ; preds = %__nv_sqrtf.exit65 + br i1 %.not1.i69, label %2200, label %2198, !dbg !33 + +2198: ; preds = %2197 + %2199 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2009) #7, !dbg !33 + br label %__nv_sqrtf.exit70, !dbg !33 + +2200: ; preds = %2197 + %2201 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2009) #7, !dbg !33 + br label %__nv_sqrtf.exit70, !dbg !33 + +2202: ; preds = %__nv_sqrtf.exit65 + br i1 %.not1.i69, label %2205, label %2203, !dbg !33 + +2203: ; preds = %2202 + %2204 = tail call float @llvm.nvvm.sqrt.rn.f(float %2009) #7, !dbg !33 + br label %__nv_sqrtf.exit70, !dbg !33 + +2205: ; preds = %2202 + %2206 = tail call float @llvm.nvvm.sqrt.approx.f(float %2009) #7, !dbg !33 + br label %__nv_sqrtf.exit70, !dbg !33 + +__nv_sqrtf.exit70: ; preds = %2198, %2200, %2203, %2205 + %.0.i68 = phi float [ %2199, %2198 ], [ %2201, %2200 ], [ %2204, %2203 ], [ %2206, %2205 ], !dbg !33 + %2207 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i71 = icmp eq i32 %2207, 0, !dbg !33 + %2208 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i74 = icmp eq i32 %2208, 0, !dbg !33 + br i1 %.not.i71, label %2214, label %2209, !dbg !33 + +2209: ; preds = %__nv_sqrtf.exit70 + br i1 %.not1.i74, label %2212, label %2210, !dbg !33 + +2210: ; preds = %2209 + %2211 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2010) #7, !dbg !33 + br label %__nv_sqrtf.exit75, !dbg !33 + +2212: ; preds = %2209 + %2213 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2010) #7, !dbg !33 + br label %__nv_sqrtf.exit75, !dbg !33 + +2214: ; preds = %__nv_sqrtf.exit70 + br i1 %.not1.i74, label %2217, label %2215, !dbg !33 + +2215: ; preds = %2214 + %2216 = tail call float @llvm.nvvm.sqrt.rn.f(float %2010) #7, !dbg !33 + br label %__nv_sqrtf.exit75, !dbg !33 + +2217: ; preds = %2214 + %2218 = tail call float @llvm.nvvm.sqrt.approx.f(float %2010) #7, !dbg !33 + br label %__nv_sqrtf.exit75, !dbg !33 + +__nv_sqrtf.exit75: ; preds = %2210, %2212, %2215, %2217 + %.0.i73 = phi float [ %2211, %2210 ], [ %2213, %2212 ], [ %2216, %2215 ], [ %2218, %2217 ], !dbg !33 + %2219 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i76 = icmp eq i32 %2219, 0, !dbg !33 + %2220 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i79 = icmp eq i32 %2220, 0, !dbg !33 + br i1 %.not.i76, label %2226, label %2221, !dbg !33 + +2221: ; preds = %__nv_sqrtf.exit75 + br i1 %.not1.i79, label %2224, label %2222, !dbg !33 + +2222: ; preds = %2221 + %2223 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2011) #7, !dbg !33 + br label %__nv_sqrtf.exit80, !dbg !33 + +2224: ; preds = %2221 + %2225 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2011) #7, !dbg !33 + br label %__nv_sqrtf.exit80, !dbg !33 + +2226: ; preds = %__nv_sqrtf.exit75 + br i1 %.not1.i79, label %2229, label %2227, !dbg !33 + +2227: ; preds = %2226 + %2228 = tail call float @llvm.nvvm.sqrt.rn.f(float %2011) #7, !dbg !33 + br label %__nv_sqrtf.exit80, !dbg !33 + +2229: ; preds = %2226 + %2230 = tail call float @llvm.nvvm.sqrt.approx.f(float %2011) #7, !dbg !33 + br label %__nv_sqrtf.exit80, !dbg !33 + +__nv_sqrtf.exit80: ; preds = %2222, %2224, %2227, %2229 + %.0.i78 = phi float [ %2223, %2222 ], [ %2225, %2224 ], [ %2228, %2227 ], [ %2230, %2229 ], !dbg !33 + %2231 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i81 = icmp eq i32 %2231, 0, !dbg !33 + %2232 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i84 = icmp eq i32 %2232, 0, !dbg !33 + br i1 %.not.i81, label %2238, label %2233, !dbg !33 + +2233: ; preds = %__nv_sqrtf.exit80 + br i1 %.not1.i84, label %2236, label %2234, !dbg !33 + +2234: ; preds = %2233 + %2235 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2012) #7, !dbg !33 + br label %__nv_sqrtf.exit85, !dbg !33 + +2236: ; preds = %2233 + %2237 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2012) #7, !dbg !33 + br label %__nv_sqrtf.exit85, !dbg !33 + +2238: ; preds = %__nv_sqrtf.exit80 + br i1 %.not1.i84, label %2241, label %2239, !dbg !33 + +2239: ; preds = %2238 + %2240 = tail call float @llvm.nvvm.sqrt.rn.f(float %2012) #7, !dbg !33 + br label %__nv_sqrtf.exit85, !dbg !33 + +2241: ; preds = %2238 + %2242 = tail call float @llvm.nvvm.sqrt.approx.f(float %2012) #7, !dbg !33 + br label %__nv_sqrtf.exit85, !dbg !33 + +__nv_sqrtf.exit85: ; preds = %2234, %2236, %2239, %2241 + %.0.i83 = phi float [ %2235, %2234 ], [ %2237, %2236 ], [ %2240, %2239 ], [ %2242, %2241 ], !dbg !33 + %2243 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i86 = icmp eq i32 %2243, 0, !dbg !33 + %2244 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i89 = icmp eq i32 %2244, 0, !dbg !33 + br i1 %.not.i86, label %2250, label %2245, !dbg !33 + +2245: ; preds = %__nv_sqrtf.exit85 + br i1 %.not1.i89, label %2248, label %2246, !dbg !33 + +2246: ; preds = %2245 + %2247 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2013) #7, !dbg !33 + br label %__nv_sqrtf.exit90, !dbg !33 + +2248: ; preds = %2245 + %2249 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2013) #7, !dbg !33 + br label %__nv_sqrtf.exit90, !dbg !33 + +2250: ; preds = %__nv_sqrtf.exit85 + br i1 %.not1.i89, label %2253, label %2251, !dbg !33 + +2251: ; preds = %2250 + %2252 = tail call float @llvm.nvvm.sqrt.rn.f(float %2013) #7, !dbg !33 + br label %__nv_sqrtf.exit90, !dbg !33 + +2253: ; preds = %2250 + %2254 = tail call float @llvm.nvvm.sqrt.approx.f(float %2013) #7, !dbg !33 + br label %__nv_sqrtf.exit90, !dbg !33 + +__nv_sqrtf.exit90: ; preds = %2246, %2248, %2251, %2253 + %.0.i88 = phi float [ %2247, %2246 ], [ %2249, %2248 ], [ %2252, %2251 ], [ %2254, %2253 ], !dbg !33 + %2255 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i91 = icmp eq i32 %2255, 0, !dbg !33 + %2256 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i94 = icmp eq i32 %2256, 0, !dbg !33 + br i1 %.not.i91, label %2262, label %2257, !dbg !33 + +2257: ; preds = %__nv_sqrtf.exit90 + br i1 %.not1.i94, label %2260, label %2258, !dbg !33 + +2258: ; preds = %2257 + %2259 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2014) #7, !dbg !33 + br label %__nv_sqrtf.exit95, !dbg !33 + +2260: ; preds = %2257 + %2261 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2014) #7, !dbg !33 + br label %__nv_sqrtf.exit95, !dbg !33 + +2262: ; preds = %__nv_sqrtf.exit90 + br i1 %.not1.i94, label %2265, label %2263, !dbg !33 + +2263: ; preds = %2262 + %2264 = tail call float @llvm.nvvm.sqrt.rn.f(float %2014) #7, !dbg !33 + br label %__nv_sqrtf.exit95, !dbg !33 + +2265: ; preds = %2262 + %2266 = tail call float @llvm.nvvm.sqrt.approx.f(float %2014) #7, !dbg !33 + br label %__nv_sqrtf.exit95, !dbg !33 + +__nv_sqrtf.exit95: ; preds = %2258, %2260, %2263, %2265 + %.0.i93 = phi float [ %2259, %2258 ], [ %2261, %2260 ], [ %2264, %2263 ], [ %2266, %2265 ], !dbg !33 + %2267 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i96 = icmp eq i32 %2267, 0, !dbg !33 + %2268 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i99 = icmp eq i32 %2268, 0, !dbg !33 + br i1 %.not.i96, label %2274, label %2269, !dbg !33 + +2269: ; preds = %__nv_sqrtf.exit95 + br i1 %.not1.i99, label %2272, label %2270, !dbg !33 + +2270: ; preds = %2269 + %2271 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2015) #7, !dbg !33 + br label %__nv_sqrtf.exit100, !dbg !33 + +2272: ; preds = %2269 + %2273 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2015) #7, !dbg !33 + br label %__nv_sqrtf.exit100, !dbg !33 + +2274: ; preds = %__nv_sqrtf.exit95 + br i1 %.not1.i99, label %2277, label %2275, !dbg !33 + +2275: ; preds = %2274 + %2276 = tail call float @llvm.nvvm.sqrt.rn.f(float %2015) #7, !dbg !33 + br label %__nv_sqrtf.exit100, !dbg !33 + +2277: ; preds = %2274 + %2278 = tail call float @llvm.nvvm.sqrt.approx.f(float %2015) #7, !dbg !33 + br label %__nv_sqrtf.exit100, !dbg !33 + +__nv_sqrtf.exit100: ; preds = %2270, %2272, %2275, %2277 + %.0.i98 = phi float [ %2271, %2270 ], [ %2273, %2272 ], [ %2276, %2275 ], [ %2278, %2277 ], !dbg !33 + %2279 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i101 = icmp eq i32 %2279, 0, !dbg !33 + %2280 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i104 = icmp eq i32 %2280, 0, !dbg !33 + br i1 %.not.i101, label %2286, label %2281, !dbg !33 + +2281: ; preds = %__nv_sqrtf.exit100 + br i1 %.not1.i104, label %2284, label %2282, !dbg !33 + +2282: ; preds = %2281 + %2283 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2016) #7, !dbg !33 + br label %__nv_sqrtf.exit105, !dbg !33 + +2284: ; preds = %2281 + %2285 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2016) #7, !dbg !33 + br label %__nv_sqrtf.exit105, !dbg !33 + +2286: ; preds = %__nv_sqrtf.exit100 + br i1 %.not1.i104, label %2289, label %2287, !dbg !33 + +2287: ; preds = %2286 + %2288 = tail call float @llvm.nvvm.sqrt.rn.f(float %2016) #7, !dbg !33 + br label %__nv_sqrtf.exit105, !dbg !33 + +2289: ; preds = %2286 + %2290 = tail call float @llvm.nvvm.sqrt.approx.f(float %2016) #7, !dbg !33 + br label %__nv_sqrtf.exit105, !dbg !33 + +__nv_sqrtf.exit105: ; preds = %2282, %2284, %2287, %2289 + %.0.i103 = phi float [ %2283, %2282 ], [ %2285, %2284 ], [ %2288, %2287 ], [ %2290, %2289 ], !dbg !33 + %2291 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i106 = icmp eq i32 %2291, 0, !dbg !33 + %2292 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i109 = icmp eq i32 %2292, 0, !dbg !33 + br i1 %.not.i106, label %2298, label %2293, !dbg !33 + +2293: ; preds = %__nv_sqrtf.exit105 + br i1 %.not1.i109, label %2296, label %2294, !dbg !33 + +2294: ; preds = %2293 + %2295 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2017) #7, !dbg !33 + br label %__nv_sqrtf.exit110, !dbg !33 + +2296: ; preds = %2293 + %2297 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2017) #7, !dbg !33 + br label %__nv_sqrtf.exit110, !dbg !33 + +2298: ; preds = %__nv_sqrtf.exit105 + br i1 %.not1.i109, label %2301, label %2299, !dbg !33 + +2299: ; preds = %2298 + %2300 = tail call float @llvm.nvvm.sqrt.rn.f(float %2017) #7, !dbg !33 + br label %__nv_sqrtf.exit110, !dbg !33 + +2301: ; preds = %2298 + %2302 = tail call float @llvm.nvvm.sqrt.approx.f(float %2017) #7, !dbg !33 + br label %__nv_sqrtf.exit110, !dbg !33 + +__nv_sqrtf.exit110: ; preds = %2294, %2296, %2299, %2301 + %.0.i108 = phi float [ %2295, %2294 ], [ %2297, %2296 ], [ %2300, %2299 ], [ %2302, %2301 ], !dbg !33 + %2303 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i111 = icmp eq i32 %2303, 0, !dbg !33 + %2304 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i114 = icmp eq i32 %2304, 0, !dbg !33 + br i1 %.not.i111, label %2310, label %2305, !dbg !33 + +2305: ; preds = %__nv_sqrtf.exit110 + br i1 %.not1.i114, label %2308, label %2306, !dbg !33 + +2306: ; preds = %2305 + %2307 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2018) #7, !dbg !33 + br label %__nv_sqrtf.exit115, !dbg !33 + +2308: ; preds = %2305 + %2309 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2018) #7, !dbg !33 + br label %__nv_sqrtf.exit115, !dbg !33 + +2310: ; preds = %__nv_sqrtf.exit110 + br i1 %.not1.i114, label %2313, label %2311, !dbg !33 + +2311: ; preds = %2310 + %2312 = tail call float @llvm.nvvm.sqrt.rn.f(float %2018) #7, !dbg !33 + br label %__nv_sqrtf.exit115, !dbg !33 + +2313: ; preds = %2310 + %2314 = tail call float @llvm.nvvm.sqrt.approx.f(float %2018) #7, !dbg !33 + br label %__nv_sqrtf.exit115, !dbg !33 + +__nv_sqrtf.exit115: ; preds = %2306, %2308, %2311, %2313 + %.0.i113 = phi float [ %2307, %2306 ], [ %2309, %2308 ], [ %2312, %2311 ], [ %2314, %2313 ], !dbg !33 + %2315 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i116 = icmp eq i32 %2315, 0, !dbg !33 + %2316 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i119 = icmp eq i32 %2316, 0, !dbg !33 + br i1 %.not.i116, label %2322, label %2317, !dbg !33 + +2317: ; preds = %__nv_sqrtf.exit115 + br i1 %.not1.i119, label %2320, label %2318, !dbg !33 + +2318: ; preds = %2317 + %2319 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2019) #7, !dbg !33 + br label %__nv_sqrtf.exit120, !dbg !33 + +2320: ; preds = %2317 + %2321 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2019) #7, !dbg !33 + br label %__nv_sqrtf.exit120, !dbg !33 + +2322: ; preds = %__nv_sqrtf.exit115 + br i1 %.not1.i119, label %2325, label %2323, !dbg !33 + +2323: ; preds = %2322 + %2324 = tail call float @llvm.nvvm.sqrt.rn.f(float %2019) #7, !dbg !33 + br label %__nv_sqrtf.exit120, !dbg !33 + +2325: ; preds = %2322 + %2326 = tail call float @llvm.nvvm.sqrt.approx.f(float %2019) #7, !dbg !33 + br label %__nv_sqrtf.exit120, !dbg !33 + +__nv_sqrtf.exit120: ; preds = %2318, %2320, %2323, %2325 + %.0.i118 = phi float [ %2319, %2318 ], [ %2321, %2320 ], [ %2324, %2323 ], [ %2326, %2325 ], !dbg !33 + %2327 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i121 = icmp eq i32 %2327, 0, !dbg !33 + %2328 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i124 = icmp eq i32 %2328, 0, !dbg !33 + br i1 %.not.i121, label %2334, label %2329, !dbg !33 + +2329: ; preds = %__nv_sqrtf.exit120 + br i1 %.not1.i124, label %2332, label %2330, !dbg !33 + +2330: ; preds = %2329 + %2331 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2020) #7, !dbg !33 + br label %__nv_sqrtf.exit125, !dbg !33 + +2332: ; preds = %2329 + %2333 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2020) #7, !dbg !33 + br label %__nv_sqrtf.exit125, !dbg !33 + +2334: ; preds = %__nv_sqrtf.exit120 + br i1 %.not1.i124, label %2337, label %2335, !dbg !33 + +2335: ; preds = %2334 + %2336 = tail call float @llvm.nvvm.sqrt.rn.f(float %2020) #7, !dbg !33 + br label %__nv_sqrtf.exit125, !dbg !33 + +2337: ; preds = %2334 + %2338 = tail call float @llvm.nvvm.sqrt.approx.f(float %2020) #7, !dbg !33 + br label %__nv_sqrtf.exit125, !dbg !33 + +__nv_sqrtf.exit125: ; preds = %2330, %2332, %2335, %2337 + %.0.i123 = phi float [ %2331, %2330 ], [ %2333, %2332 ], [ %2336, %2335 ], [ %2338, %2337 ], !dbg !33 + %2339 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i126 = icmp eq i32 %2339, 0, !dbg !33 + %2340 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i129 = icmp eq i32 %2340, 0, !dbg !33 + br i1 %.not.i126, label %2346, label %2341, !dbg !33 + +2341: ; preds = %__nv_sqrtf.exit125 + br i1 %.not1.i129, label %2344, label %2342, !dbg !33 + +2342: ; preds = %2341 + %2343 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2021) #7, !dbg !33 + br label %__nv_sqrtf.exit130, !dbg !33 + +2344: ; preds = %2341 + %2345 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2021) #7, !dbg !33 + br label %__nv_sqrtf.exit130, !dbg !33 + +2346: ; preds = %__nv_sqrtf.exit125 + br i1 %.not1.i129, label %2349, label %2347, !dbg !33 + +2347: ; preds = %2346 + %2348 = tail call float @llvm.nvvm.sqrt.rn.f(float %2021) #7, !dbg !33 + br label %__nv_sqrtf.exit130, !dbg !33 + +2349: ; preds = %2346 + %2350 = tail call float @llvm.nvvm.sqrt.approx.f(float %2021) #7, !dbg !33 + br label %__nv_sqrtf.exit130, !dbg !33 + +__nv_sqrtf.exit130: ; preds = %2342, %2344, %2347, %2349 + %.0.i128 = phi float [ %2343, %2342 ], [ %2345, %2344 ], [ %2348, %2347 ], [ %2350, %2349 ], !dbg !33 + %2351 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i131 = icmp eq i32 %2351, 0, !dbg !33 + %2352 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i134 = icmp eq i32 %2352, 0, !dbg !33 + br i1 %.not.i131, label %2358, label %2353, !dbg !33 + +2353: ; preds = %__nv_sqrtf.exit130 + br i1 %.not1.i134, label %2356, label %2354, !dbg !33 + +2354: ; preds = %2353 + %2355 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2022) #7, !dbg !33 + br label %__nv_sqrtf.exit135, !dbg !33 + +2356: ; preds = %2353 + %2357 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2022) #7, !dbg !33 + br label %__nv_sqrtf.exit135, !dbg !33 + +2358: ; preds = %__nv_sqrtf.exit130 + br i1 %.not1.i134, label %2361, label %2359, !dbg !33 + +2359: ; preds = %2358 + %2360 = tail call float @llvm.nvvm.sqrt.rn.f(float %2022) #7, !dbg !33 + br label %__nv_sqrtf.exit135, !dbg !33 + +2361: ; preds = %2358 + %2362 = tail call float @llvm.nvvm.sqrt.approx.f(float %2022) #7, !dbg !33 + br label %__nv_sqrtf.exit135, !dbg !33 + +__nv_sqrtf.exit135: ; preds = %2354, %2356, %2359, %2361 + %.0.i133 = phi float [ %2355, %2354 ], [ %2357, %2356 ], [ %2360, %2359 ], [ %2362, %2361 ], !dbg !33 + %2363 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i136 = icmp eq i32 %2363, 0, !dbg !33 + %2364 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i139 = icmp eq i32 %2364, 0, !dbg !33 + br i1 %.not.i136, label %2370, label %2365, !dbg !33 + +2365: ; preds = %__nv_sqrtf.exit135 + br i1 %.not1.i139, label %2368, label %2366, !dbg !33 + +2366: ; preds = %2365 + %2367 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2023) #7, !dbg !33 + br label %__nv_sqrtf.exit140, !dbg !33 + +2368: ; preds = %2365 + %2369 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2023) #7, !dbg !33 + br label %__nv_sqrtf.exit140, !dbg !33 + +2370: ; preds = %__nv_sqrtf.exit135 + br i1 %.not1.i139, label %2373, label %2371, !dbg !33 + +2371: ; preds = %2370 + %2372 = tail call float @llvm.nvvm.sqrt.rn.f(float %2023) #7, !dbg !33 + br label %__nv_sqrtf.exit140, !dbg !33 + +2373: ; preds = %2370 + %2374 = tail call float @llvm.nvvm.sqrt.approx.f(float %2023) #7, !dbg !33 + br label %__nv_sqrtf.exit140, !dbg !33 + +__nv_sqrtf.exit140: ; preds = %2366, %2368, %2371, %2373 + %.0.i138 = phi float [ %2367, %2366 ], [ %2369, %2368 ], [ %2372, %2371 ], [ %2374, %2373 ], !dbg !33 + %2375 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i141 = icmp eq i32 %2375, 0, !dbg !33 + %2376 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i144 = icmp eq i32 %2376, 0, !dbg !33 + br i1 %.not.i141, label %2382, label %2377, !dbg !33 + +2377: ; preds = %__nv_sqrtf.exit140 + br i1 %.not1.i144, label %2380, label %2378, !dbg !33 + +2378: ; preds = %2377 + %2379 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2024) #7, !dbg !33 + br label %__nv_sqrtf.exit145, !dbg !33 + +2380: ; preds = %2377 + %2381 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2024) #7, !dbg !33 + br label %__nv_sqrtf.exit145, !dbg !33 + +2382: ; preds = %__nv_sqrtf.exit140 + br i1 %.not1.i144, label %2385, label %2383, !dbg !33 + +2383: ; preds = %2382 + %2384 = tail call float @llvm.nvvm.sqrt.rn.f(float %2024) #7, !dbg !33 + br label %__nv_sqrtf.exit145, !dbg !33 + +2385: ; preds = %2382 + %2386 = tail call float @llvm.nvvm.sqrt.approx.f(float %2024) #7, !dbg !33 + br label %__nv_sqrtf.exit145, !dbg !33 + +__nv_sqrtf.exit145: ; preds = %2378, %2380, %2383, %2385 + %.0.i143 = phi float [ %2379, %2378 ], [ %2381, %2380 ], [ %2384, %2383 ], [ %2386, %2385 ], !dbg !33 + %2387 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i146 = icmp eq i32 %2387, 0, !dbg !33 + %2388 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i149 = icmp eq i32 %2388, 0, !dbg !33 + br i1 %.not.i146, label %2394, label %2389, !dbg !33 + +2389: ; preds = %__nv_sqrtf.exit145 + br i1 %.not1.i149, label %2392, label %2390, !dbg !33 + +2390: ; preds = %2389 + %2391 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2025) #7, !dbg !33 + br label %__nv_sqrtf.exit150, !dbg !33 + +2392: ; preds = %2389 + %2393 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2025) #7, !dbg !33 + br label %__nv_sqrtf.exit150, !dbg !33 + +2394: ; preds = %__nv_sqrtf.exit145 + br i1 %.not1.i149, label %2397, label %2395, !dbg !33 + +2395: ; preds = %2394 + %2396 = tail call float @llvm.nvvm.sqrt.rn.f(float %2025) #7, !dbg !33 + br label %__nv_sqrtf.exit150, !dbg !33 + +2397: ; preds = %2394 + %2398 = tail call float @llvm.nvvm.sqrt.approx.f(float %2025) #7, !dbg !33 + br label %__nv_sqrtf.exit150, !dbg !33 + +__nv_sqrtf.exit150: ; preds = %2390, %2392, %2395, %2397 + %.0.i148 = phi float [ %2391, %2390 ], [ %2393, %2392 ], [ %2396, %2395 ], [ %2398, %2397 ], !dbg !33 + %2399 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i151 = icmp eq i32 %2399, 0, !dbg !33 + %2400 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i154 = icmp eq i32 %2400, 0, !dbg !33 + br i1 %.not.i151, label %2406, label %2401, !dbg !33 + +2401: ; preds = %__nv_sqrtf.exit150 + br i1 %.not1.i154, label %2404, label %2402, !dbg !33 + +2402: ; preds = %2401 + %2403 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2026) #7, !dbg !33 + br label %__nv_sqrtf.exit155, !dbg !33 + +2404: ; preds = %2401 + %2405 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2026) #7, !dbg !33 + br label %__nv_sqrtf.exit155, !dbg !33 + +2406: ; preds = %__nv_sqrtf.exit150 + br i1 %.not1.i154, label %2409, label %2407, !dbg !33 + +2407: ; preds = %2406 + %2408 = tail call float @llvm.nvvm.sqrt.rn.f(float %2026) #7, !dbg !33 + br label %__nv_sqrtf.exit155, !dbg !33 + +2409: ; preds = %2406 + %2410 = tail call float @llvm.nvvm.sqrt.approx.f(float %2026) #7, !dbg !33 + br label %__nv_sqrtf.exit155, !dbg !33 + +__nv_sqrtf.exit155: ; preds = %2402, %2404, %2407, %2409 + %.0.i153 = phi float [ %2403, %2402 ], [ %2405, %2404 ], [ %2408, %2407 ], [ %2410, %2409 ], !dbg !33 + %2411 = sext i32 %4 to i64, !dbg !18 + %2412 = shl nuw nsw i32 %11, 1, !dbg !16 + %2413 = and i32 %2412, 62, !dbg !16 + %2414 = or disjoint i32 %2413, 1, !dbg !16 + %2415 = zext nneg i32 %2414 to i64, !dbg !17 + %2416 = or disjoint i64 %10, %2415, !dbg !17 + %2417 = getelementptr float, ptr addrspace(1) %2, i64 %2416, !dbg !34 + %2418 = zext nneg i32 %2413 to i64, !dbg !17 + %2419 = or disjoint i64 %10, %2418, !dbg !17 + %2420 = getelementptr float, ptr addrspace(1) %2, i64 %2419, !dbg !34 + %2421 = getelementptr bfloat, ptr addrspace(1) %1, i64 %105, !dbg !35 + %2422 = getelementptr bfloat, ptr addrspace(1) %2421, i64 %138, !dbg !36 + %2423 = getelementptr bfloat, ptr addrspace(1) %1, i64 %103, !dbg !35 + %2424 = getelementptr bfloat, ptr addrspace(1) %2423, i64 %138, !dbg !36 + %2425 = getelementptr bfloat, ptr addrspace(1) %1, i64 %101, !dbg !35 + %2426 = getelementptr bfloat, ptr addrspace(1) %2425, i64 %138, !dbg !36 + %2427 = getelementptr bfloat, ptr addrspace(1) %1, i64 %99, !dbg !35 + %2428 = getelementptr bfloat, ptr addrspace(1) %2427, i64 %138, !dbg !36 + %2429 = getelementptr bfloat, ptr addrspace(1) %1, i64 %97, !dbg !35 + %2430 = getelementptr bfloat, ptr addrspace(1) %2429, i64 %138, !dbg !36 + %2431 = getelementptr bfloat, ptr addrspace(1) %1, i64 %95, !dbg !35 + %2432 = getelementptr bfloat, ptr addrspace(1) %2431, i64 %138, !dbg !36 + %2433 = getelementptr bfloat, ptr addrspace(1) %1, i64 %93, !dbg !35 + %2434 = getelementptr bfloat, ptr addrspace(1) %2433, i64 %138, !dbg !36 + %2435 = getelementptr bfloat, ptr addrspace(1) %1, i64 %91, !dbg !35 + %2436 = getelementptr bfloat, ptr addrspace(1) %2435, i64 %138, !dbg !36 + %2437 = getelementptr bfloat, ptr addrspace(1) %1, i64 %89, !dbg !35 + %2438 = getelementptr bfloat, ptr addrspace(1) %2437, i64 %138, !dbg !36 + %2439 = getelementptr bfloat, ptr addrspace(1) %1, i64 %87, !dbg !35 + %2440 = getelementptr bfloat, ptr addrspace(1) %2439, i64 %138, !dbg !36 + %2441 = getelementptr bfloat, ptr addrspace(1) %1, i64 %85, !dbg !35 + %2442 = getelementptr bfloat, ptr addrspace(1) %2441, i64 %138, !dbg !36 + %2443 = getelementptr bfloat, ptr addrspace(1) %1, i64 %83, !dbg !35 + %2444 = getelementptr bfloat, ptr addrspace(1) %2443, i64 %138, !dbg !36 + %2445 = getelementptr bfloat, ptr addrspace(1) %1, i64 %81, !dbg !35 + %2446 = getelementptr bfloat, ptr addrspace(1) %2445, i64 %138, !dbg !36 + %2447 = getelementptr bfloat, ptr addrspace(1) %1, i64 %79, !dbg !35 + %2448 = getelementptr bfloat, ptr addrspace(1) %2447, i64 %138, !dbg !36 + %2449 = getelementptr bfloat, ptr addrspace(1) %1, i64 %77, !dbg !35 + %2450 = getelementptr bfloat, ptr addrspace(1) %2449, i64 %138, !dbg !36 + %2451 = getelementptr bfloat, ptr addrspace(1) %1, i64 %75, !dbg !35 + %2452 = getelementptr bfloat, ptr addrspace(1) %2451, i64 %138, !dbg !36 + %2453 = getelementptr bfloat, ptr addrspace(1) %1, i64 %73, !dbg !35 + %2454 = getelementptr bfloat, ptr addrspace(1) %2453, i64 %138, !dbg !36 + %2455 = getelementptr bfloat, ptr addrspace(1) %1, i64 %71, !dbg !35 + %2456 = getelementptr bfloat, ptr addrspace(1) %2455, i64 %138, !dbg !36 + %2457 = getelementptr bfloat, ptr addrspace(1) %1, i64 %69, !dbg !35 + %2458 = getelementptr bfloat, ptr addrspace(1) %2457, i64 %138, !dbg !36 + %2459 = getelementptr bfloat, ptr addrspace(1) %1, i64 %67, !dbg !35 + %2460 = getelementptr bfloat, ptr addrspace(1) %2459, i64 %138, !dbg !36 + %2461 = getelementptr bfloat, ptr addrspace(1) %1, i64 %65, !dbg !35 + %2462 = getelementptr bfloat, ptr addrspace(1) %2461, i64 %138, !dbg !36 + %2463 = getelementptr bfloat, ptr addrspace(1) %1, i64 %63, !dbg !35 + %2464 = getelementptr bfloat, ptr addrspace(1) %2463, i64 %138, !dbg !36 + %2465 = getelementptr bfloat, ptr addrspace(1) %1, i64 %61, !dbg !35 + %2466 = getelementptr bfloat, ptr addrspace(1) %2465, i64 %138, !dbg !36 + %2467 = getelementptr bfloat, ptr addrspace(1) %1, i64 %59, !dbg !35 + %2468 = getelementptr bfloat, ptr addrspace(1) %2467, i64 %138, !dbg !36 + %2469 = getelementptr bfloat, ptr addrspace(1) %1, i64 %57, !dbg !35 + %2470 = getelementptr bfloat, ptr addrspace(1) %2469, i64 %138, !dbg !36 + %2471 = getelementptr bfloat, ptr addrspace(1) %1, i64 %55, !dbg !35 + %2472 = getelementptr bfloat, ptr addrspace(1) %2471, i64 %138, !dbg !36 + %2473 = getelementptr bfloat, ptr addrspace(1) %1, i64 %53, !dbg !35 + %2474 = getelementptr bfloat, ptr addrspace(1) %2473, i64 %138, !dbg !36 + %2475 = getelementptr bfloat, ptr addrspace(1) %1, i64 %51, !dbg !35 + %2476 = getelementptr bfloat, ptr addrspace(1) %2475, i64 %138, !dbg !36 + %2477 = getelementptr bfloat, ptr addrspace(1) %1, i64 %49, !dbg !35 + %2478 = getelementptr bfloat, ptr addrspace(1) %2477, i64 %138, !dbg !36 + %2479 = getelementptr bfloat, ptr addrspace(1) %1, i64 %47, !dbg !35 + %2480 = getelementptr bfloat, ptr addrspace(1) %2479, i64 %138, !dbg !36 + %2481 = getelementptr bfloat, ptr addrspace(1) %1, i64 %45, !dbg !35 + %2482 = getelementptr bfloat, ptr addrspace(1) %2481, i64 %138, !dbg !36 + %2483 = getelementptr bfloat, ptr addrspace(1) %1, i64 %43, !dbg !35 + %2484 = getelementptr bfloat, ptr addrspace(1) %2483, i64 %138, !dbg !36 + %2485 = icmp slt i64 %2416, %2411, !dbg !18 + %2486 = icmp slt i64 %2419, %2411, !dbg !18 + %2487 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i), !dbg !37 + %2488 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i3), !dbg !37 + %2489 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i8), !dbg !37 + %2490 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i13), !dbg !37 + %2491 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i18), !dbg !37 + %2492 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i23), !dbg !37 + %2493 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i28), !dbg !37 + %2494 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i33), !dbg !37 + %2495 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i38), !dbg !37 + %2496 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i43), !dbg !37 + %2497 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i48), !dbg !37 + %2498 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i53), !dbg !37 + %2499 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i58), !dbg !37 + %2500 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i63), !dbg !37 + %2501 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i68), !dbg !37 + %2502 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i73), !dbg !37 + %2503 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i78), !dbg !37 + %2504 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i83), !dbg !37 + %2505 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i88), !dbg !37 + %2506 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i93), !dbg !37 + %2507 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i98), !dbg !37 + %2508 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i103), !dbg !37 + %2509 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i108), !dbg !37 + %2510 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i113), !dbg !37 + %2511 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i118), !dbg !37 + %2512 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i123), !dbg !37 + %2513 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i128), !dbg !37 + %2514 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i133), !dbg !37 + %2515 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i138), !dbg !37 + %2516 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i143), !dbg !37 + %2517 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i148), !dbg !37 + %2518 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i153), !dbg !37 + %2519 = fmul float %2487, %747, !dbg !38 + %2520 = fmul float %2487, %748, !dbg !38 + %2521 = fmul float %2487, %749, !dbg !38 + %2522 = fmul float %2487, %750, !dbg !38 + %2523 = fmul float %2487, %751, !dbg !38 + %2524 = fmul float %2487, %752, !dbg !38 + %2525 = fmul float %2487, %753, !dbg !38 + %2526 = fmul float %2487, %754, !dbg !38 + %2527 = fmul float %2488, %755, !dbg !38 + %2528 = fmul float %2488, %756, !dbg !38 + %2529 = fmul float %2488, %757, !dbg !38 + %2530 = fmul float %2488, %758, !dbg !38 + %2531 = fmul float %2488, %759, !dbg !38 + %2532 = fmul float %2488, %760, !dbg !38 + %2533 = fmul float %2488, %761, !dbg !38 + %2534 = fmul float %2488, %762, !dbg !38 + %2535 = fmul float %2489, %763, !dbg !38 + %2536 = fmul float %2489, %764, !dbg !38 + %2537 = fmul float %2489, %765, !dbg !38 + %2538 = fmul float %2489, %766, !dbg !38 + %2539 = fmul float %2489, %767, !dbg !38 + %2540 = fmul float %2489, %768, !dbg !38 + %2541 = fmul float %2489, %769, !dbg !38 + %2542 = fmul float %2489, %770, !dbg !38 + %2543 = fmul float %2490, %771, !dbg !38 + %2544 = fmul float %2490, %772, !dbg !38 + %2545 = fmul float %2490, %773, !dbg !38 + %2546 = fmul float %2490, %774, !dbg !38 + %2547 = fmul float %2490, %775, !dbg !38 + %2548 = fmul float %2490, %776, !dbg !38 + %2549 = fmul float %2490, %777, !dbg !38 + %2550 = fmul float %2490, %778, !dbg !38 + %2551 = fmul float %2491, %779, !dbg !38 + %2552 = fmul float %2491, %780, !dbg !38 + %2553 = fmul float %2491, %781, !dbg !38 + %2554 = fmul float %2491, %782, !dbg !38 + %2555 = fmul float %2491, %783, !dbg !38 + %2556 = fmul float %2491, %784, !dbg !38 + %2557 = fmul float %2491, %785, !dbg !38 + %2558 = fmul float %2491, %786, !dbg !38 + %2559 = fmul float %2492, %787, !dbg !38 + %2560 = fmul float %2492, %788, !dbg !38 + %2561 = fmul float %2492, %789, !dbg !38 + %2562 = fmul float %2492, %790, !dbg !38 + %2563 = fmul float %2492, %791, !dbg !38 + %2564 = fmul float %2492, %792, !dbg !38 + %2565 = fmul float %2492, %793, !dbg !38 + %2566 = fmul float %2492, %794, !dbg !38 + %2567 = fmul float %2493, %795, !dbg !38 + %2568 = fmul float %2493, %796, !dbg !38 + %2569 = fmul float %2493, %797, !dbg !38 + %2570 = fmul float %2493, %798, !dbg !38 + %2571 = fmul float %2493, %799, !dbg !38 + %2572 = fmul float %2493, %800, !dbg !38 + %2573 = fmul float %2493, %801, !dbg !38 + %2574 = fmul float %2493, %802, !dbg !38 + %2575 = fmul float %2494, %803, !dbg !38 + %2576 = fmul float %2494, %804, !dbg !38 + %2577 = fmul float %2494, %805, !dbg !38 + %2578 = fmul float %2494, %806, !dbg !38 + %2579 = fmul float %2494, %807, !dbg !38 + %2580 = fmul float %2494, %808, !dbg !38 + %2581 = fmul float %2494, %809, !dbg !38 + %2582 = fmul float %2494, %810, !dbg !38 + %2583 = fmul float %2495, %811, !dbg !38 + %2584 = fmul float %2495, %812, !dbg !38 + %2585 = fmul float %2495, %813, !dbg !38 + %2586 = fmul float %2495, %814, !dbg !38 + %2587 = fmul float %2495, %815, !dbg !38 + %2588 = fmul float %2495, %816, !dbg !38 + %2589 = fmul float %2495, %817, !dbg !38 + %2590 = fmul float %2495, %818, !dbg !38 + %2591 = fmul float %2496, %819, !dbg !38 + %2592 = fmul float %2496, %820, !dbg !38 + %2593 = fmul float %2496, %821, !dbg !38 + %2594 = fmul float %2496, %822, !dbg !38 + %2595 = fmul float %2496, %823, !dbg !38 + %2596 = fmul float %2496, %824, !dbg !38 + %2597 = fmul float %2496, %825, !dbg !38 + %2598 = fmul float %2496, %826, !dbg !38 + %2599 = fmul float %2497, %827, !dbg !38 + %2600 = fmul float %2497, %828, !dbg !38 + %2601 = fmul float %2497, %829, !dbg !38 + %2602 = fmul float %2497, %830, !dbg !38 + %2603 = fmul float %2497, %831, !dbg !38 + %2604 = fmul float %2497, %832, !dbg !38 + %2605 = fmul float %2497, %833, !dbg !38 + %2606 = fmul float %2497, %834, !dbg !38 + %2607 = fmul float %2498, %835, !dbg !38 + %2608 = fmul float %2498, %836, !dbg !38 + %2609 = fmul float %2498, %837, !dbg !38 + %2610 = fmul float %2498, %838, !dbg !38 + %2611 = fmul float %2498, %839, !dbg !38 + %2612 = fmul float %2498, %840, !dbg !38 + %2613 = fmul float %2498, %841, !dbg !38 + %2614 = fmul float %2498, %842, !dbg !38 + %2615 = fmul float %2499, %843, !dbg !38 + %2616 = fmul float %2499, %844, !dbg !38 + %2617 = fmul float %2499, %845, !dbg !38 + %2618 = fmul float %2499, %846, !dbg !38 + %2619 = fmul float %2499, %847, !dbg !38 + %2620 = fmul float %2499, %848, !dbg !38 + %2621 = fmul float %2499, %849, !dbg !38 + %2622 = fmul float %2499, %850, !dbg !38 + %2623 = fmul float %2500, %851, !dbg !38 + %2624 = fmul float %2500, %852, !dbg !38 + %2625 = fmul float %2500, %853, !dbg !38 + %2626 = fmul float %2500, %854, !dbg !38 + %2627 = fmul float %2500, %855, !dbg !38 + %2628 = fmul float %2500, %856, !dbg !38 + %2629 = fmul float %2500, %857, !dbg !38 + %2630 = fmul float %2500, %858, !dbg !38 + %2631 = fmul float %2501, %859, !dbg !38 + %2632 = fmul float %2501, %860, !dbg !38 + %2633 = fmul float %2501, %861, !dbg !38 + %2634 = fmul float %2501, %862, !dbg !38 + %2635 = fmul float %2501, %863, !dbg !38 + %2636 = fmul float %2501, %864, !dbg !38 + %2637 = fmul float %2501, %865, !dbg !38 + %2638 = fmul float %2501, %866, !dbg !38 + %2639 = fmul float %2502, %867, !dbg !38 + %2640 = fmul float %2502, %868, !dbg !38 + %2641 = fmul float %2502, %869, !dbg !38 + %2642 = fmul float %2502, %870, !dbg !38 + %2643 = fmul float %2502, %871, !dbg !38 + %2644 = fmul float %2502, %872, !dbg !38 + %2645 = fmul float %2502, %873, !dbg !38 + %2646 = fmul float %2502, %874, !dbg !38 + %2647 = fmul float %2503, %875, !dbg !38 + %2648 = fmul float %2503, %876, !dbg !38 + %2649 = fmul float %2503, %877, !dbg !38 + %2650 = fmul float %2503, %878, !dbg !38 + %2651 = fmul float %2503, %879, !dbg !38 + %2652 = fmul float %2503, %880, !dbg !38 + %2653 = fmul float %2503, %881, !dbg !38 + %2654 = fmul float %2503, %882, !dbg !38 + %2655 = fmul float %2504, %883, !dbg !38 + %2656 = fmul float %2504, %884, !dbg !38 + %2657 = fmul float %2504, %885, !dbg !38 + %2658 = fmul float %2504, %886, !dbg !38 + %2659 = fmul float %2504, %887, !dbg !38 + %2660 = fmul float %2504, %888, !dbg !38 + %2661 = fmul float %2504, %889, !dbg !38 + %2662 = fmul float %2504, %890, !dbg !38 + %2663 = fmul float %2505, %891, !dbg !38 + %2664 = fmul float %2505, %892, !dbg !38 + %2665 = fmul float %2505, %893, !dbg !38 + %2666 = fmul float %2505, %894, !dbg !38 + %2667 = fmul float %2505, %895, !dbg !38 + %2668 = fmul float %2505, %896, !dbg !38 + %2669 = fmul float %2505, %897, !dbg !38 + %2670 = fmul float %2505, %898, !dbg !38 + %2671 = fmul float %2506, %899, !dbg !38 + %2672 = fmul float %2506, %900, !dbg !38 + %2673 = fmul float %2506, %901, !dbg !38 + %2674 = fmul float %2506, %902, !dbg !38 + %2675 = fmul float %2506, %903, !dbg !38 + %2676 = fmul float %2506, %904, !dbg !38 + %2677 = fmul float %2506, %905, !dbg !38 + %2678 = fmul float %2506, %906, !dbg !38 + %2679 = fmul float %2507, %907, !dbg !38 + %2680 = fmul float %2507, %908, !dbg !38 + %2681 = fmul float %2507, %909, !dbg !38 + %2682 = fmul float %2507, %910, !dbg !38 + %2683 = fmul float %2507, %911, !dbg !38 + %2684 = fmul float %2507, %912, !dbg !38 + %2685 = fmul float %2507, %913, !dbg !38 + %2686 = fmul float %2507, %914, !dbg !38 + %2687 = fmul float %2508, %915, !dbg !38 + %2688 = fmul float %2508, %916, !dbg !38 + %2689 = fmul float %2508, %917, !dbg !38 + %2690 = fmul float %2508, %918, !dbg !38 + %2691 = fmul float %2508, %919, !dbg !38 + %2692 = fmul float %2508, %920, !dbg !38 + %2693 = fmul float %2508, %921, !dbg !38 + %2694 = fmul float %2508, %922, !dbg !38 + %2695 = fmul float %2509, %923, !dbg !38 + %2696 = fmul float %2509, %924, !dbg !38 + %2697 = fmul float %2509, %925, !dbg !38 + %2698 = fmul float %2509, %926, !dbg !38 + %2699 = fmul float %2509, %927, !dbg !38 + %2700 = fmul float %2509, %928, !dbg !38 + %2701 = fmul float %2509, %929, !dbg !38 + %2702 = fmul float %2509, %930, !dbg !38 + %2703 = fmul float %2510, %931, !dbg !38 + %2704 = fmul float %2510, %932, !dbg !38 + %2705 = fmul float %2510, %933, !dbg !38 + %2706 = fmul float %2510, %934, !dbg !38 + %2707 = fmul float %2510, %935, !dbg !38 + %2708 = fmul float %2510, %936, !dbg !38 + %2709 = fmul float %2510, %937, !dbg !38 + %2710 = fmul float %2510, %938, !dbg !38 + %2711 = fmul float %2511, %939, !dbg !38 + %2712 = fmul float %2511, %940, !dbg !38 + %2713 = fmul float %2511, %941, !dbg !38 + %2714 = fmul float %2511, %942, !dbg !38 + %2715 = fmul float %2511, %943, !dbg !38 + %2716 = fmul float %2511, %944, !dbg !38 + %2717 = fmul float %2511, %945, !dbg !38 + %2718 = fmul float %2511, %946, !dbg !38 + %2719 = fmul float %2512, %947, !dbg !38 + %2720 = fmul float %2512, %948, !dbg !38 + %2721 = fmul float %2512, %949, !dbg !38 + %2722 = fmul float %2512, %950, !dbg !38 + %2723 = fmul float %2512, %951, !dbg !38 + %2724 = fmul float %2512, %952, !dbg !38 + %2725 = fmul float %2512, %953, !dbg !38 + %2726 = fmul float %2512, %954, !dbg !38 + %2727 = fmul float %2513, %955, !dbg !38 + %2728 = fmul float %2513, %956, !dbg !38 + %2729 = fmul float %2513, %957, !dbg !38 + %2730 = fmul float %2513, %958, !dbg !38 + %2731 = fmul float %2513, %959, !dbg !38 + %2732 = fmul float %2513, %960, !dbg !38 + %2733 = fmul float %2513, %961, !dbg !38 + %2734 = fmul float %2513, %962, !dbg !38 + %2735 = fmul float %2514, %963, !dbg !38 + %2736 = fmul float %2514, %964, !dbg !38 + %2737 = fmul float %2514, %965, !dbg !38 + %2738 = fmul float %2514, %966, !dbg !38 + %2739 = fmul float %2514, %967, !dbg !38 + %2740 = fmul float %2514, %968, !dbg !38 + %2741 = fmul float %2514, %969, !dbg !38 + %2742 = fmul float %2514, %970, !dbg !38 + %2743 = fmul float %2515, %971, !dbg !38 + %2744 = fmul float %2515, %972, !dbg !38 + %2745 = fmul float %2515, %973, !dbg !38 + %2746 = fmul float %2515, %974, !dbg !38 + %2747 = fmul float %2515, %975, !dbg !38 + %2748 = fmul float %2515, %976, !dbg !38 + %2749 = fmul float %2515, %977, !dbg !38 + %2750 = fmul float %2515, %978, !dbg !38 + %2751 = fmul float %2516, %979, !dbg !38 + %2752 = fmul float %2516, %980, !dbg !38 + %2753 = fmul float %2516, %981, !dbg !38 + %2754 = fmul float %2516, %982, !dbg !38 + %2755 = fmul float %2516, %983, !dbg !38 + %2756 = fmul float %2516, %984, !dbg !38 + %2757 = fmul float %2516, %985, !dbg !38 + %2758 = fmul float %2516, %986, !dbg !38 + %2759 = fmul float %2517, %987, !dbg !38 + %2760 = fmul float %2517, %988, !dbg !38 + %2761 = fmul float %2517, %989, !dbg !38 + %2762 = fmul float %2517, %990, !dbg !38 + %2763 = fmul float %2517, %991, !dbg !38 + %2764 = fmul float %2517, %992, !dbg !38 + %2765 = fmul float %2517, %993, !dbg !38 + %2766 = fmul float %2517, %994, !dbg !38 + %2767 = fmul float %2518, %995, !dbg !38 + %2768 = fmul float %2518, %996, !dbg !38 + %2769 = fmul float %2518, %997, !dbg !38 + %2770 = fmul float %2518, %998, !dbg !38 + %2771 = fmul float %2518, %999, !dbg !38 + %2772 = fmul float %2518, %1000, !dbg !38 + %2773 = fmul float %2518, %1001, !dbg !38 + %2774 = fmul float %2518, %1002, !dbg !38 + %2775 = fptrunc float %2519 to bfloat, !dbg !39 + %2776 = fptrunc float %2520 to bfloat, !dbg !39 + %2777 = fptrunc float %2521 to bfloat, !dbg !39 + %2778 = fptrunc float %2522 to bfloat, !dbg !39 + %2779 = fptrunc float %2523 to bfloat, !dbg !39 + %2780 = fptrunc float %2524 to bfloat, !dbg !39 + %2781 = fptrunc float %2525 to bfloat, !dbg !39 + %2782 = fptrunc float %2526 to bfloat, !dbg !39 + %2783 = fptrunc float %2527 to bfloat, !dbg !39 + %2784 = fptrunc float %2528 to bfloat, !dbg !39 + %2785 = fptrunc float %2529 to bfloat, !dbg !39 + %2786 = fptrunc float %2530 to bfloat, !dbg !39 + %2787 = fptrunc float %2531 to bfloat, !dbg !39 + %2788 = fptrunc float %2532 to bfloat, !dbg !39 + %2789 = fptrunc float %2533 to bfloat, !dbg !39 + %2790 = fptrunc float %2534 to bfloat, !dbg !39 + %2791 = fptrunc float %2535 to bfloat, !dbg !39 + %2792 = fptrunc float %2536 to bfloat, !dbg !39 + %2793 = fptrunc float %2537 to bfloat, !dbg !39 + %2794 = fptrunc float %2538 to bfloat, !dbg !39 + %2795 = fptrunc float %2539 to bfloat, !dbg !39 + %2796 = fptrunc float %2540 to bfloat, !dbg !39 + %2797 = fptrunc float %2541 to bfloat, !dbg !39 + %2798 = fptrunc float %2542 to bfloat, !dbg !39 + %2799 = fptrunc float %2543 to bfloat, !dbg !39 + %2800 = fptrunc float %2544 to bfloat, !dbg !39 + %2801 = fptrunc float %2545 to bfloat, !dbg !39 + %2802 = fptrunc float %2546 to bfloat, !dbg !39 + %2803 = fptrunc float %2547 to bfloat, !dbg !39 + %2804 = fptrunc float %2548 to bfloat, !dbg !39 + %2805 = fptrunc float %2549 to bfloat, !dbg !39 + %2806 = fptrunc float %2550 to bfloat, !dbg !39 + %2807 = fptrunc float %2551 to bfloat, !dbg !39 + %2808 = fptrunc float %2552 to bfloat, !dbg !39 + %2809 = fptrunc float %2553 to bfloat, !dbg !39 + %2810 = fptrunc float %2554 to bfloat, !dbg !39 + %2811 = fptrunc float %2555 to bfloat, !dbg !39 + %2812 = fptrunc float %2556 to bfloat, !dbg !39 + %2813 = fptrunc float %2557 to bfloat, !dbg !39 + %2814 = fptrunc float %2558 to bfloat, !dbg !39 + %2815 = fptrunc float %2559 to bfloat, !dbg !39 + %2816 = fptrunc float %2560 to bfloat, !dbg !39 + %2817 = fptrunc float %2561 to bfloat, !dbg !39 + %2818 = fptrunc float %2562 to bfloat, !dbg !39 + %2819 = fptrunc float %2563 to bfloat, !dbg !39 + %2820 = fptrunc float %2564 to bfloat, !dbg !39 + %2821 = fptrunc float %2565 to bfloat, !dbg !39 + %2822 = fptrunc float %2566 to bfloat, !dbg !39 + %2823 = fptrunc float %2567 to bfloat, !dbg !39 + %2824 = fptrunc float %2568 to bfloat, !dbg !39 + %2825 = fptrunc float %2569 to bfloat, !dbg !39 + %2826 = fptrunc float %2570 to bfloat, !dbg !39 + %2827 = fptrunc float %2571 to bfloat, !dbg !39 + %2828 = fptrunc float %2572 to bfloat, !dbg !39 + %2829 = fptrunc float %2573 to bfloat, !dbg !39 + %2830 = fptrunc float %2574 to bfloat, !dbg !39 + %2831 = fptrunc float %2575 to bfloat, !dbg !39 + %2832 = fptrunc float %2576 to bfloat, !dbg !39 + %2833 = fptrunc float %2577 to bfloat, !dbg !39 + %2834 = fptrunc float %2578 to bfloat, !dbg !39 + %2835 = fptrunc float %2579 to bfloat, !dbg !39 + %2836 = fptrunc float %2580 to bfloat, !dbg !39 + %2837 = fptrunc float %2581 to bfloat, !dbg !39 + %2838 = fptrunc float %2582 to bfloat, !dbg !39 + %2839 = fptrunc float %2583 to bfloat, !dbg !39 + %2840 = fptrunc float %2584 to bfloat, !dbg !39 + %2841 = fptrunc float %2585 to bfloat, !dbg !39 + %2842 = fptrunc float %2586 to bfloat, !dbg !39 + %2843 = fptrunc float %2587 to bfloat, !dbg !39 + %2844 = fptrunc float %2588 to bfloat, !dbg !39 + %2845 = fptrunc float %2589 to bfloat, !dbg !39 + %2846 = fptrunc float %2590 to bfloat, !dbg !39 + %2847 = fptrunc float %2591 to bfloat, !dbg !39 + %2848 = fptrunc float %2592 to bfloat, !dbg !39 + %2849 = fptrunc float %2593 to bfloat, !dbg !39 + %2850 = fptrunc float %2594 to bfloat, !dbg !39 + %2851 = fptrunc float %2595 to bfloat, !dbg !39 + %2852 = fptrunc float %2596 to bfloat, !dbg !39 + %2853 = fptrunc float %2597 to bfloat, !dbg !39 + %2854 = fptrunc float %2598 to bfloat, !dbg !39 + %2855 = fptrunc float %2599 to bfloat, !dbg !39 + %2856 = fptrunc float %2600 to bfloat, !dbg !39 + %2857 = fptrunc float %2601 to bfloat, !dbg !39 + %2858 = fptrunc float %2602 to bfloat, !dbg !39 + %2859 = fptrunc float %2603 to bfloat, !dbg !39 + %2860 = fptrunc float %2604 to bfloat, !dbg !39 + %2861 = fptrunc float %2605 to bfloat, !dbg !39 + %2862 = fptrunc float %2606 to bfloat, !dbg !39 + %2863 = fptrunc float %2607 to bfloat, !dbg !39 + %2864 = fptrunc float %2608 to bfloat, !dbg !39 + %2865 = fptrunc float %2609 to bfloat, !dbg !39 + %2866 = fptrunc float %2610 to bfloat, !dbg !39 + %2867 = fptrunc float %2611 to bfloat, !dbg !39 + %2868 = fptrunc float %2612 to bfloat, !dbg !39 + %2869 = fptrunc float %2613 to bfloat, !dbg !39 + %2870 = fptrunc float %2614 to bfloat, !dbg !39 + %2871 = fptrunc float %2615 to bfloat, !dbg !39 + %2872 = fptrunc float %2616 to bfloat, !dbg !39 + %2873 = fptrunc float %2617 to bfloat, !dbg !39 + %2874 = fptrunc float %2618 to bfloat, !dbg !39 + %2875 = fptrunc float %2619 to bfloat, !dbg !39 + %2876 = fptrunc float %2620 to bfloat, !dbg !39 + %2877 = fptrunc float %2621 to bfloat, !dbg !39 + %2878 = fptrunc float %2622 to bfloat, !dbg !39 + %2879 = fptrunc float %2623 to bfloat, !dbg !39 + %2880 = fptrunc float %2624 to bfloat, !dbg !39 + %2881 = fptrunc float %2625 to bfloat, !dbg !39 + %2882 = fptrunc float %2626 to bfloat, !dbg !39 + %2883 = fptrunc float %2627 to bfloat, !dbg !39 + %2884 = fptrunc float %2628 to bfloat, !dbg !39 + %2885 = fptrunc float %2629 to bfloat, !dbg !39 + %2886 = fptrunc float %2630 to bfloat, !dbg !39 + %2887 = fptrunc float %2631 to bfloat, !dbg !39 + %2888 = fptrunc float %2632 to bfloat, !dbg !39 + %2889 = fptrunc float %2633 to bfloat, !dbg !39 + %2890 = fptrunc float %2634 to bfloat, !dbg !39 + %2891 = fptrunc float %2635 to bfloat, !dbg !39 + %2892 = fptrunc float %2636 to bfloat, !dbg !39 + %2893 = fptrunc float %2637 to bfloat, !dbg !39 + %2894 = fptrunc float %2638 to bfloat, !dbg !39 + %2895 = fptrunc float %2639 to bfloat, !dbg !39 + %2896 = fptrunc float %2640 to bfloat, !dbg !39 + %2897 = fptrunc float %2641 to bfloat, !dbg !39 + %2898 = fptrunc float %2642 to bfloat, !dbg !39 + %2899 = fptrunc float %2643 to bfloat, !dbg !39 + %2900 = fptrunc float %2644 to bfloat, !dbg !39 + %2901 = fptrunc float %2645 to bfloat, !dbg !39 + %2902 = fptrunc float %2646 to bfloat, !dbg !39 + %2903 = fptrunc float %2647 to bfloat, !dbg !39 + %2904 = fptrunc float %2648 to bfloat, !dbg !39 + %2905 = fptrunc float %2649 to bfloat, !dbg !39 + %2906 = fptrunc float %2650 to bfloat, !dbg !39 + %2907 = fptrunc float %2651 to bfloat, !dbg !39 + %2908 = fptrunc float %2652 to bfloat, !dbg !39 + %2909 = fptrunc float %2653 to bfloat, !dbg !39 + %2910 = fptrunc float %2654 to bfloat, !dbg !39 + %2911 = fptrunc float %2655 to bfloat, !dbg !39 + %2912 = fptrunc float %2656 to bfloat, !dbg !39 + %2913 = fptrunc float %2657 to bfloat, !dbg !39 + %2914 = fptrunc float %2658 to bfloat, !dbg !39 + %2915 = fptrunc float %2659 to bfloat, !dbg !39 + %2916 = fptrunc float %2660 to bfloat, !dbg !39 + %2917 = fptrunc float %2661 to bfloat, !dbg !39 + %2918 = fptrunc float %2662 to bfloat, !dbg !39 + %2919 = fptrunc float %2663 to bfloat, !dbg !39 + %2920 = fptrunc float %2664 to bfloat, !dbg !39 + %2921 = fptrunc float %2665 to bfloat, !dbg !39 + %2922 = fptrunc float %2666 to bfloat, !dbg !39 + %2923 = fptrunc float %2667 to bfloat, !dbg !39 + %2924 = fptrunc float %2668 to bfloat, !dbg !39 + %2925 = fptrunc float %2669 to bfloat, !dbg !39 + %2926 = fptrunc float %2670 to bfloat, !dbg !39 + %2927 = fptrunc float %2671 to bfloat, !dbg !39 + %2928 = fptrunc float %2672 to bfloat, !dbg !39 + %2929 = fptrunc float %2673 to bfloat, !dbg !39 + %2930 = fptrunc float %2674 to bfloat, !dbg !39 + %2931 = fptrunc float %2675 to bfloat, !dbg !39 + %2932 = fptrunc float %2676 to bfloat, !dbg !39 + %2933 = fptrunc float %2677 to bfloat, !dbg !39 + %2934 = fptrunc float %2678 to bfloat, !dbg !39 + %2935 = fptrunc float %2679 to bfloat, !dbg !39 + %2936 = fptrunc float %2680 to bfloat, !dbg !39 + %2937 = fptrunc float %2681 to bfloat, !dbg !39 + %2938 = fptrunc float %2682 to bfloat, !dbg !39 + %2939 = fptrunc float %2683 to bfloat, !dbg !39 + %2940 = fptrunc float %2684 to bfloat, !dbg !39 + %2941 = fptrunc float %2685 to bfloat, !dbg !39 + %2942 = fptrunc float %2686 to bfloat, !dbg !39 + %2943 = fptrunc float %2687 to bfloat, !dbg !39 + %2944 = fptrunc float %2688 to bfloat, !dbg !39 + %2945 = fptrunc float %2689 to bfloat, !dbg !39 + %2946 = fptrunc float %2690 to bfloat, !dbg !39 + %2947 = fptrunc float %2691 to bfloat, !dbg !39 + %2948 = fptrunc float %2692 to bfloat, !dbg !39 + %2949 = fptrunc float %2693 to bfloat, !dbg !39 + %2950 = fptrunc float %2694 to bfloat, !dbg !39 + %2951 = fptrunc float %2695 to bfloat, !dbg !39 + %2952 = fptrunc float %2696 to bfloat, !dbg !39 + %2953 = fptrunc float %2697 to bfloat, !dbg !39 + %2954 = fptrunc float %2698 to bfloat, !dbg !39 + %2955 = fptrunc float %2699 to bfloat, !dbg !39 + %2956 = fptrunc float %2700 to bfloat, !dbg !39 + %2957 = fptrunc float %2701 to bfloat, !dbg !39 + %2958 = fptrunc float %2702 to bfloat, !dbg !39 + %2959 = fptrunc float %2703 to bfloat, !dbg !39 + %2960 = fptrunc float %2704 to bfloat, !dbg !39 + %2961 = fptrunc float %2705 to bfloat, !dbg !39 + %2962 = fptrunc float %2706 to bfloat, !dbg !39 + %2963 = fptrunc float %2707 to bfloat, !dbg !39 + %2964 = fptrunc float %2708 to bfloat, !dbg !39 + %2965 = fptrunc float %2709 to bfloat, !dbg !39 + %2966 = fptrunc float %2710 to bfloat, !dbg !39 + %2967 = fptrunc float %2711 to bfloat, !dbg !39 + %2968 = fptrunc float %2712 to bfloat, !dbg !39 + %2969 = fptrunc float %2713 to bfloat, !dbg !39 + %2970 = fptrunc float %2714 to bfloat, !dbg !39 + %2971 = fptrunc float %2715 to bfloat, !dbg !39 + %2972 = fptrunc float %2716 to bfloat, !dbg !39 + %2973 = fptrunc float %2717 to bfloat, !dbg !39 + %2974 = fptrunc float %2718 to bfloat, !dbg !39 + %2975 = fptrunc float %2719 to bfloat, !dbg !39 + %2976 = fptrunc float %2720 to bfloat, !dbg !39 + %2977 = fptrunc float %2721 to bfloat, !dbg !39 + %2978 = fptrunc float %2722 to bfloat, !dbg !39 + %2979 = fptrunc float %2723 to bfloat, !dbg !39 + %2980 = fptrunc float %2724 to bfloat, !dbg !39 + %2981 = fptrunc float %2725 to bfloat, !dbg !39 + %2982 = fptrunc float %2726 to bfloat, !dbg !39 + %2983 = fptrunc float %2727 to bfloat, !dbg !39 + %2984 = fptrunc float %2728 to bfloat, !dbg !39 + %2985 = fptrunc float %2729 to bfloat, !dbg !39 + %2986 = fptrunc float %2730 to bfloat, !dbg !39 + %2987 = fptrunc float %2731 to bfloat, !dbg !39 + %2988 = fptrunc float %2732 to bfloat, !dbg !39 + %2989 = fptrunc float %2733 to bfloat, !dbg !39 + %2990 = fptrunc float %2734 to bfloat, !dbg !39 + %2991 = fptrunc float %2735 to bfloat, !dbg !39 + %2992 = fptrunc float %2736 to bfloat, !dbg !39 + %2993 = fptrunc float %2737 to bfloat, !dbg !39 + %2994 = fptrunc float %2738 to bfloat, !dbg !39 + %2995 = fptrunc float %2739 to bfloat, !dbg !39 + %2996 = fptrunc float %2740 to bfloat, !dbg !39 + %2997 = fptrunc float %2741 to bfloat, !dbg !39 + %2998 = fptrunc float %2742 to bfloat, !dbg !39 + %2999 = fptrunc float %2743 to bfloat, !dbg !39 + %3000 = fptrunc float %2744 to bfloat, !dbg !39 + %3001 = fptrunc float %2745 to bfloat, !dbg !39 + %3002 = fptrunc float %2746 to bfloat, !dbg !39 + %3003 = fptrunc float %2747 to bfloat, !dbg !39 + %3004 = fptrunc float %2748 to bfloat, !dbg !39 + %3005 = fptrunc float %2749 to bfloat, !dbg !39 + %3006 = fptrunc float %2750 to bfloat, !dbg !39 + %3007 = fptrunc float %2751 to bfloat, !dbg !39 + %3008 = fptrunc float %2752 to bfloat, !dbg !39 + %3009 = fptrunc float %2753 to bfloat, !dbg !39 + %3010 = fptrunc float %2754 to bfloat, !dbg !39 + %3011 = fptrunc float %2755 to bfloat, !dbg !39 + %3012 = fptrunc float %2756 to bfloat, !dbg !39 + %3013 = fptrunc float %2757 to bfloat, !dbg !39 + %3014 = fptrunc float %2758 to bfloat, !dbg !39 + %3015 = fptrunc float %2759 to bfloat, !dbg !39 + %3016 = fptrunc float %2760 to bfloat, !dbg !39 + %3017 = fptrunc float %2761 to bfloat, !dbg !39 + %3018 = fptrunc float %2762 to bfloat, !dbg !39 + %3019 = fptrunc float %2763 to bfloat, !dbg !39 + %3020 = fptrunc float %2764 to bfloat, !dbg !39 + %3021 = fptrunc float %2765 to bfloat, !dbg !39 + %3022 = fptrunc float %2766 to bfloat, !dbg !39 + %3023 = fptrunc float %2767 to bfloat, !dbg !39 + %3024 = fptrunc float %2768 to bfloat, !dbg !39 + %3025 = fptrunc float %2769 to bfloat, !dbg !39 + %3026 = fptrunc float %2770 to bfloat, !dbg !39 + %3027 = fptrunc float %2771 to bfloat, !dbg !39 + %3028 = fptrunc float %2772 to bfloat, !dbg !39 + %3029 = fptrunc float %2773 to bfloat, !dbg !39 + %3030 = fptrunc float %2774 to bfloat, !dbg !39 + %3031 = insertelement <2 x bfloat> poison, bfloat %2775, i64 0, !dbg !40 + %3032 = insertelement <2 x bfloat> %3031, bfloat %2776, i64 1, !dbg !40 + %3033 = bitcast <2 x bfloat> %3032 to i32, !dbg !40 + %3034 = insertelement <2 x bfloat> poison, bfloat %2777, i64 0, !dbg !40 + %3035 = insertelement <2 x bfloat> %3034, bfloat %2778, i64 1, !dbg !40 + %3036 = bitcast <2 x bfloat> %3035 to i32, !dbg !40 + %3037 = insertelement <2 x bfloat> poison, bfloat %2779, i64 0, !dbg !40 + %3038 = insertelement <2 x bfloat> %3037, bfloat %2780, i64 1, !dbg !40 + %3039 = bitcast <2 x bfloat> %3038 to i32, !dbg !40 + %3040 = insertelement <2 x bfloat> poison, bfloat %2781, i64 0, !dbg !40 + %3041 = insertelement <2 x bfloat> %3040, bfloat %2782, i64 1, !dbg !40 + %3042 = bitcast <2 x bfloat> %3041 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3033, i32 %3036, i32 %3039, i32 %3042, ptr addrspace(1) %2484, i1 %171) #7, !dbg !40 + %3043 = insertelement <2 x bfloat> poison, bfloat %2783, i64 0, !dbg !40 + %3044 = insertelement <2 x bfloat> %3043, bfloat %2784, i64 1, !dbg !40 + %3045 = bitcast <2 x bfloat> %3044 to i32, !dbg !40 + %3046 = insertelement <2 x bfloat> poison, bfloat %2785, i64 0, !dbg !40 + %3047 = insertelement <2 x bfloat> %3046, bfloat %2786, i64 1, !dbg !40 + %3048 = bitcast <2 x bfloat> %3047 to i32, !dbg !40 + %3049 = insertelement <2 x bfloat> poison, bfloat %2787, i64 0, !dbg !40 + %3050 = insertelement <2 x bfloat> %3049, bfloat %2788, i64 1, !dbg !40 + %3051 = bitcast <2 x bfloat> %3050 to i32, !dbg !40 + %3052 = insertelement <2 x bfloat> poison, bfloat %2789, i64 0, !dbg !40 + %3053 = insertelement <2 x bfloat> %3052, bfloat %2790, i64 1, !dbg !40 + %3054 = bitcast <2 x bfloat> %3053 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3045, i32 %3048, i32 %3051, i32 %3054, ptr addrspace(1) %2482, i1 %189) #7, !dbg !40 + %3055 = insertelement <2 x bfloat> poison, bfloat %2791, i64 0, !dbg !40 + %3056 = insertelement <2 x bfloat> %3055, bfloat %2792, i64 1, !dbg !40 + %3057 = bitcast <2 x bfloat> %3056 to i32, !dbg !40 + %3058 = insertelement <2 x bfloat> poison, bfloat %2793, i64 0, !dbg !40 + %3059 = insertelement <2 x bfloat> %3058, bfloat %2794, i64 1, !dbg !40 + %3060 = bitcast <2 x bfloat> %3059 to i32, !dbg !40 + %3061 = insertelement <2 x bfloat> poison, bfloat %2795, i64 0, !dbg !40 + %3062 = insertelement <2 x bfloat> %3061, bfloat %2796, i64 1, !dbg !40 + %3063 = bitcast <2 x bfloat> %3062 to i32, !dbg !40 + %3064 = insertelement <2 x bfloat> poison, bfloat %2797, i64 0, !dbg !40 + %3065 = insertelement <2 x bfloat> %3064, bfloat %2798, i64 1, !dbg !40 + %3066 = bitcast <2 x bfloat> %3065 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3057, i32 %3060, i32 %3063, i32 %3066, ptr addrspace(1) %2480, i1 %207) #7, !dbg !40 + %3067 = insertelement <2 x bfloat> poison, bfloat %2799, i64 0, !dbg !40 + %3068 = insertelement <2 x bfloat> %3067, bfloat %2800, i64 1, !dbg !40 + %3069 = bitcast <2 x bfloat> %3068 to i32, !dbg !40 + %3070 = insertelement <2 x bfloat> poison, bfloat %2801, i64 0, !dbg !40 + %3071 = insertelement <2 x bfloat> %3070, bfloat %2802, i64 1, !dbg !40 + %3072 = bitcast <2 x bfloat> %3071 to i32, !dbg !40 + %3073 = insertelement <2 x bfloat> poison, bfloat %2803, i64 0, !dbg !40 + %3074 = insertelement <2 x bfloat> %3073, bfloat %2804, i64 1, !dbg !40 + %3075 = bitcast <2 x bfloat> %3074 to i32, !dbg !40 + %3076 = insertelement <2 x bfloat> poison, bfloat %2805, i64 0, !dbg !40 + %3077 = insertelement <2 x bfloat> %3076, bfloat %2806, i64 1, !dbg !40 + %3078 = bitcast <2 x bfloat> %3077 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3069, i32 %3072, i32 %3075, i32 %3078, ptr addrspace(1) %2478, i1 %225) #7, !dbg !40 + %3079 = insertelement <2 x bfloat> poison, bfloat %2807, i64 0, !dbg !40 + %3080 = insertelement <2 x bfloat> %3079, bfloat %2808, i64 1, !dbg !40 + %3081 = bitcast <2 x bfloat> %3080 to i32, !dbg !40 + %3082 = insertelement <2 x bfloat> poison, bfloat %2809, i64 0, !dbg !40 + %3083 = insertelement <2 x bfloat> %3082, bfloat %2810, i64 1, !dbg !40 + %3084 = bitcast <2 x bfloat> %3083 to i32, !dbg !40 + %3085 = insertelement <2 x bfloat> poison, bfloat %2811, i64 0, !dbg !40 + %3086 = insertelement <2 x bfloat> %3085, bfloat %2812, i64 1, !dbg !40 + %3087 = bitcast <2 x bfloat> %3086 to i32, !dbg !40 + %3088 = insertelement <2 x bfloat> poison, bfloat %2813, i64 0, !dbg !40 + %3089 = insertelement <2 x bfloat> %3088, bfloat %2814, i64 1, !dbg !40 + %3090 = bitcast <2 x bfloat> %3089 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3081, i32 %3084, i32 %3087, i32 %3090, ptr addrspace(1) %2476, i1 %243) #7, !dbg !40 + %3091 = insertelement <2 x bfloat> poison, bfloat %2815, i64 0, !dbg !40 + %3092 = insertelement <2 x bfloat> %3091, bfloat %2816, i64 1, !dbg !40 + %3093 = bitcast <2 x bfloat> %3092 to i32, !dbg !40 + %3094 = insertelement <2 x bfloat> poison, bfloat %2817, i64 0, !dbg !40 + %3095 = insertelement <2 x bfloat> %3094, bfloat %2818, i64 1, !dbg !40 + %3096 = bitcast <2 x bfloat> %3095 to i32, !dbg !40 + %3097 = insertelement <2 x bfloat> poison, bfloat %2819, i64 0, !dbg !40 + %3098 = insertelement <2 x bfloat> %3097, bfloat %2820, i64 1, !dbg !40 + %3099 = bitcast <2 x bfloat> %3098 to i32, !dbg !40 + %3100 = insertelement <2 x bfloat> poison, bfloat %2821, i64 0, !dbg !40 + %3101 = insertelement <2 x bfloat> %3100, bfloat %2822, i64 1, !dbg !40 + %3102 = bitcast <2 x bfloat> %3101 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3093, i32 %3096, i32 %3099, i32 %3102, ptr addrspace(1) %2474, i1 %261) #7, !dbg !40 + %3103 = insertelement <2 x bfloat> poison, bfloat %2823, i64 0, !dbg !40 + %3104 = insertelement <2 x bfloat> %3103, bfloat %2824, i64 1, !dbg !40 + %3105 = bitcast <2 x bfloat> %3104 to i32, !dbg !40 + %3106 = insertelement <2 x bfloat> poison, bfloat %2825, i64 0, !dbg !40 + %3107 = insertelement <2 x bfloat> %3106, bfloat %2826, i64 1, !dbg !40 + %3108 = bitcast <2 x bfloat> %3107 to i32, !dbg !40 + %3109 = insertelement <2 x bfloat> poison, bfloat %2827, i64 0, !dbg !40 + %3110 = insertelement <2 x bfloat> %3109, bfloat %2828, i64 1, !dbg !40 + %3111 = bitcast <2 x bfloat> %3110 to i32, !dbg !40 + %3112 = insertelement <2 x bfloat> poison, bfloat %2829, i64 0, !dbg !40 + %3113 = insertelement <2 x bfloat> %3112, bfloat %2830, i64 1, !dbg !40 + %3114 = bitcast <2 x bfloat> %3113 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3105, i32 %3108, i32 %3111, i32 %3114, ptr addrspace(1) %2472, i1 %279) #7, !dbg !40 + %3115 = insertelement <2 x bfloat> poison, bfloat %2831, i64 0, !dbg !40 + %3116 = insertelement <2 x bfloat> %3115, bfloat %2832, i64 1, !dbg !40 + %3117 = bitcast <2 x bfloat> %3116 to i32, !dbg !40 + %3118 = insertelement <2 x bfloat> poison, bfloat %2833, i64 0, !dbg !40 + %3119 = insertelement <2 x bfloat> %3118, bfloat %2834, i64 1, !dbg !40 + %3120 = bitcast <2 x bfloat> %3119 to i32, !dbg !40 + %3121 = insertelement <2 x bfloat> poison, bfloat %2835, i64 0, !dbg !40 + %3122 = insertelement <2 x bfloat> %3121, bfloat %2836, i64 1, !dbg !40 + %3123 = bitcast <2 x bfloat> %3122 to i32, !dbg !40 + %3124 = insertelement <2 x bfloat> poison, bfloat %2837, i64 0, !dbg !40 + %3125 = insertelement <2 x bfloat> %3124, bfloat %2838, i64 1, !dbg !40 + %3126 = bitcast <2 x bfloat> %3125 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3117, i32 %3120, i32 %3123, i32 %3126, ptr addrspace(1) %2470, i1 %297) #7, !dbg !40 + %3127 = insertelement <2 x bfloat> poison, bfloat %2839, i64 0, !dbg !40 + %3128 = insertelement <2 x bfloat> %3127, bfloat %2840, i64 1, !dbg !40 + %3129 = bitcast <2 x bfloat> %3128 to i32, !dbg !40 + %3130 = insertelement <2 x bfloat> poison, bfloat %2841, i64 0, !dbg !40 + %3131 = insertelement <2 x bfloat> %3130, bfloat %2842, i64 1, !dbg !40 + %3132 = bitcast <2 x bfloat> %3131 to i32, !dbg !40 + %3133 = insertelement <2 x bfloat> poison, bfloat %2843, i64 0, !dbg !40 + %3134 = insertelement <2 x bfloat> %3133, bfloat %2844, i64 1, !dbg !40 + %3135 = bitcast <2 x bfloat> %3134 to i32, !dbg !40 + %3136 = insertelement <2 x bfloat> poison, bfloat %2845, i64 0, !dbg !40 + %3137 = insertelement <2 x bfloat> %3136, bfloat %2846, i64 1, !dbg !40 + %3138 = bitcast <2 x bfloat> %3137 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3129, i32 %3132, i32 %3135, i32 %3138, ptr addrspace(1) %2468, i1 %315) #7, !dbg !40 + %3139 = insertelement <2 x bfloat> poison, bfloat %2847, i64 0, !dbg !40 + %3140 = insertelement <2 x bfloat> %3139, bfloat %2848, i64 1, !dbg !40 + %3141 = bitcast <2 x bfloat> %3140 to i32, !dbg !40 + %3142 = insertelement <2 x bfloat> poison, bfloat %2849, i64 0, !dbg !40 + %3143 = insertelement <2 x bfloat> %3142, bfloat %2850, i64 1, !dbg !40 + %3144 = bitcast <2 x bfloat> %3143 to i32, !dbg !40 + %3145 = insertelement <2 x bfloat> poison, bfloat %2851, i64 0, !dbg !40 + %3146 = insertelement <2 x bfloat> %3145, bfloat %2852, i64 1, !dbg !40 + %3147 = bitcast <2 x bfloat> %3146 to i32, !dbg !40 + %3148 = insertelement <2 x bfloat> poison, bfloat %2853, i64 0, !dbg !40 + %3149 = insertelement <2 x bfloat> %3148, bfloat %2854, i64 1, !dbg !40 + %3150 = bitcast <2 x bfloat> %3149 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3141, i32 %3144, i32 %3147, i32 %3150, ptr addrspace(1) %2466, i1 %333) #7, !dbg !40 + %3151 = insertelement <2 x bfloat> poison, bfloat %2855, i64 0, !dbg !40 + %3152 = insertelement <2 x bfloat> %3151, bfloat %2856, i64 1, !dbg !40 + %3153 = bitcast <2 x bfloat> %3152 to i32, !dbg !40 + %3154 = insertelement <2 x bfloat> poison, bfloat %2857, i64 0, !dbg !40 + %3155 = insertelement <2 x bfloat> %3154, bfloat %2858, i64 1, !dbg !40 + %3156 = bitcast <2 x bfloat> %3155 to i32, !dbg !40 + %3157 = insertelement <2 x bfloat> poison, bfloat %2859, i64 0, !dbg !40 + %3158 = insertelement <2 x bfloat> %3157, bfloat %2860, i64 1, !dbg !40 + %3159 = bitcast <2 x bfloat> %3158 to i32, !dbg !40 + %3160 = insertelement <2 x bfloat> poison, bfloat %2861, i64 0, !dbg !40 + %3161 = insertelement <2 x bfloat> %3160, bfloat %2862, i64 1, !dbg !40 + %3162 = bitcast <2 x bfloat> %3161 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3153, i32 %3156, i32 %3159, i32 %3162, ptr addrspace(1) %2464, i1 %351) #7, !dbg !40 + %3163 = insertelement <2 x bfloat> poison, bfloat %2863, i64 0, !dbg !40 + %3164 = insertelement <2 x bfloat> %3163, bfloat %2864, i64 1, !dbg !40 + %3165 = bitcast <2 x bfloat> %3164 to i32, !dbg !40 + %3166 = insertelement <2 x bfloat> poison, bfloat %2865, i64 0, !dbg !40 + %3167 = insertelement <2 x bfloat> %3166, bfloat %2866, i64 1, !dbg !40 + %3168 = bitcast <2 x bfloat> %3167 to i32, !dbg !40 + %3169 = insertelement <2 x bfloat> poison, bfloat %2867, i64 0, !dbg !40 + %3170 = insertelement <2 x bfloat> %3169, bfloat %2868, i64 1, !dbg !40 + %3171 = bitcast <2 x bfloat> %3170 to i32, !dbg !40 + %3172 = insertelement <2 x bfloat> poison, bfloat %2869, i64 0, !dbg !40 + %3173 = insertelement <2 x bfloat> %3172, bfloat %2870, i64 1, !dbg !40 + %3174 = bitcast <2 x bfloat> %3173 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3165, i32 %3168, i32 %3171, i32 %3174, ptr addrspace(1) %2462, i1 %369) #7, !dbg !40 + %3175 = insertelement <2 x bfloat> poison, bfloat %2871, i64 0, !dbg !40 + %3176 = insertelement <2 x bfloat> %3175, bfloat %2872, i64 1, !dbg !40 + %3177 = bitcast <2 x bfloat> %3176 to i32, !dbg !40 + %3178 = insertelement <2 x bfloat> poison, bfloat %2873, i64 0, !dbg !40 + %3179 = insertelement <2 x bfloat> %3178, bfloat %2874, i64 1, !dbg !40 + %3180 = bitcast <2 x bfloat> %3179 to i32, !dbg !40 + %3181 = insertelement <2 x bfloat> poison, bfloat %2875, i64 0, !dbg !40 + %3182 = insertelement <2 x bfloat> %3181, bfloat %2876, i64 1, !dbg !40 + %3183 = bitcast <2 x bfloat> %3182 to i32, !dbg !40 + %3184 = insertelement <2 x bfloat> poison, bfloat %2877, i64 0, !dbg !40 + %3185 = insertelement <2 x bfloat> %3184, bfloat %2878, i64 1, !dbg !40 + %3186 = bitcast <2 x bfloat> %3185 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3177, i32 %3180, i32 %3183, i32 %3186, ptr addrspace(1) %2460, i1 %387) #7, !dbg !40 + %3187 = insertelement <2 x bfloat> poison, bfloat %2879, i64 0, !dbg !40 + %3188 = insertelement <2 x bfloat> %3187, bfloat %2880, i64 1, !dbg !40 + %3189 = bitcast <2 x bfloat> %3188 to i32, !dbg !40 + %3190 = insertelement <2 x bfloat> poison, bfloat %2881, i64 0, !dbg !40 + %3191 = insertelement <2 x bfloat> %3190, bfloat %2882, i64 1, !dbg !40 + %3192 = bitcast <2 x bfloat> %3191 to i32, !dbg !40 + %3193 = insertelement <2 x bfloat> poison, bfloat %2883, i64 0, !dbg !40 + %3194 = insertelement <2 x bfloat> %3193, bfloat %2884, i64 1, !dbg !40 + %3195 = bitcast <2 x bfloat> %3194 to i32, !dbg !40 + %3196 = insertelement <2 x bfloat> poison, bfloat %2885, i64 0, !dbg !40 + %3197 = insertelement <2 x bfloat> %3196, bfloat %2886, i64 1, !dbg !40 + %3198 = bitcast <2 x bfloat> %3197 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3189, i32 %3192, i32 %3195, i32 %3198, ptr addrspace(1) %2458, i1 %405) #7, !dbg !40 + %3199 = insertelement <2 x bfloat> poison, bfloat %2887, i64 0, !dbg !40 + %3200 = insertelement <2 x bfloat> %3199, bfloat %2888, i64 1, !dbg !40 + %3201 = bitcast <2 x bfloat> %3200 to i32, !dbg !40 + %3202 = insertelement <2 x bfloat> poison, bfloat %2889, i64 0, !dbg !40 + %3203 = insertelement <2 x bfloat> %3202, bfloat %2890, i64 1, !dbg !40 + %3204 = bitcast <2 x bfloat> %3203 to i32, !dbg !40 + %3205 = insertelement <2 x bfloat> poison, bfloat %2891, i64 0, !dbg !40 + %3206 = insertelement <2 x bfloat> %3205, bfloat %2892, i64 1, !dbg !40 + %3207 = bitcast <2 x bfloat> %3206 to i32, !dbg !40 + %3208 = insertelement <2 x bfloat> poison, bfloat %2893, i64 0, !dbg !40 + %3209 = insertelement <2 x bfloat> %3208, bfloat %2894, i64 1, !dbg !40 + %3210 = bitcast <2 x bfloat> %3209 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3201, i32 %3204, i32 %3207, i32 %3210, ptr addrspace(1) %2456, i1 %423) #7, !dbg !40 + %3211 = insertelement <2 x bfloat> poison, bfloat %2895, i64 0, !dbg !40 + %3212 = insertelement <2 x bfloat> %3211, bfloat %2896, i64 1, !dbg !40 + %3213 = bitcast <2 x bfloat> %3212 to i32, !dbg !40 + %3214 = insertelement <2 x bfloat> poison, bfloat %2897, i64 0, !dbg !40 + %3215 = insertelement <2 x bfloat> %3214, bfloat %2898, i64 1, !dbg !40 + %3216 = bitcast <2 x bfloat> %3215 to i32, !dbg !40 + %3217 = insertelement <2 x bfloat> poison, bfloat %2899, i64 0, !dbg !40 + %3218 = insertelement <2 x bfloat> %3217, bfloat %2900, i64 1, !dbg !40 + %3219 = bitcast <2 x bfloat> %3218 to i32, !dbg !40 + %3220 = insertelement <2 x bfloat> poison, bfloat %2901, i64 0, !dbg !40 + %3221 = insertelement <2 x bfloat> %3220, bfloat %2902, i64 1, !dbg !40 + %3222 = bitcast <2 x bfloat> %3221 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3213, i32 %3216, i32 %3219, i32 %3222, ptr addrspace(1) %2454, i1 %441) #7, !dbg !40 + %3223 = insertelement <2 x bfloat> poison, bfloat %2903, i64 0, !dbg !40 + %3224 = insertelement <2 x bfloat> %3223, bfloat %2904, i64 1, !dbg !40 + %3225 = bitcast <2 x bfloat> %3224 to i32, !dbg !40 + %3226 = insertelement <2 x bfloat> poison, bfloat %2905, i64 0, !dbg !40 + %3227 = insertelement <2 x bfloat> %3226, bfloat %2906, i64 1, !dbg !40 + %3228 = bitcast <2 x bfloat> %3227 to i32, !dbg !40 + %3229 = insertelement <2 x bfloat> poison, bfloat %2907, i64 0, !dbg !40 + %3230 = insertelement <2 x bfloat> %3229, bfloat %2908, i64 1, !dbg !40 + %3231 = bitcast <2 x bfloat> %3230 to i32, !dbg !40 + %3232 = insertelement <2 x bfloat> poison, bfloat %2909, i64 0, !dbg !40 + %3233 = insertelement <2 x bfloat> %3232, bfloat %2910, i64 1, !dbg !40 + %3234 = bitcast <2 x bfloat> %3233 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3225, i32 %3228, i32 %3231, i32 %3234, ptr addrspace(1) %2452, i1 %459) #7, !dbg !40 + %3235 = insertelement <2 x bfloat> poison, bfloat %2911, i64 0, !dbg !40 + %3236 = insertelement <2 x bfloat> %3235, bfloat %2912, i64 1, !dbg !40 + %3237 = bitcast <2 x bfloat> %3236 to i32, !dbg !40 + %3238 = insertelement <2 x bfloat> poison, bfloat %2913, i64 0, !dbg !40 + %3239 = insertelement <2 x bfloat> %3238, bfloat %2914, i64 1, !dbg !40 + %3240 = bitcast <2 x bfloat> %3239 to i32, !dbg !40 + %3241 = insertelement <2 x bfloat> poison, bfloat %2915, i64 0, !dbg !40 + %3242 = insertelement <2 x bfloat> %3241, bfloat %2916, i64 1, !dbg !40 + %3243 = bitcast <2 x bfloat> %3242 to i32, !dbg !40 + %3244 = insertelement <2 x bfloat> poison, bfloat %2917, i64 0, !dbg !40 + %3245 = insertelement <2 x bfloat> %3244, bfloat %2918, i64 1, !dbg !40 + %3246 = bitcast <2 x bfloat> %3245 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3237, i32 %3240, i32 %3243, i32 %3246, ptr addrspace(1) %2450, i1 %477) #7, !dbg !40 + %3247 = insertelement <2 x bfloat> poison, bfloat %2919, i64 0, !dbg !40 + %3248 = insertelement <2 x bfloat> %3247, bfloat %2920, i64 1, !dbg !40 + %3249 = bitcast <2 x bfloat> %3248 to i32, !dbg !40 + %3250 = insertelement <2 x bfloat> poison, bfloat %2921, i64 0, !dbg !40 + %3251 = insertelement <2 x bfloat> %3250, bfloat %2922, i64 1, !dbg !40 + %3252 = bitcast <2 x bfloat> %3251 to i32, !dbg !40 + %3253 = insertelement <2 x bfloat> poison, bfloat %2923, i64 0, !dbg !40 + %3254 = insertelement <2 x bfloat> %3253, bfloat %2924, i64 1, !dbg !40 + %3255 = bitcast <2 x bfloat> %3254 to i32, !dbg !40 + %3256 = insertelement <2 x bfloat> poison, bfloat %2925, i64 0, !dbg !40 + %3257 = insertelement <2 x bfloat> %3256, bfloat %2926, i64 1, !dbg !40 + %3258 = bitcast <2 x bfloat> %3257 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3249, i32 %3252, i32 %3255, i32 %3258, ptr addrspace(1) %2448, i1 %495) #7, !dbg !40 + %3259 = insertelement <2 x bfloat> poison, bfloat %2927, i64 0, !dbg !40 + %3260 = insertelement <2 x bfloat> %3259, bfloat %2928, i64 1, !dbg !40 + %3261 = bitcast <2 x bfloat> %3260 to i32, !dbg !40 + %3262 = insertelement <2 x bfloat> poison, bfloat %2929, i64 0, !dbg !40 + %3263 = insertelement <2 x bfloat> %3262, bfloat %2930, i64 1, !dbg !40 + %3264 = bitcast <2 x bfloat> %3263 to i32, !dbg !40 + %3265 = insertelement <2 x bfloat> poison, bfloat %2931, i64 0, !dbg !40 + %3266 = insertelement <2 x bfloat> %3265, bfloat %2932, i64 1, !dbg !40 + %3267 = bitcast <2 x bfloat> %3266 to i32, !dbg !40 + %3268 = insertelement <2 x bfloat> poison, bfloat %2933, i64 0, !dbg !40 + %3269 = insertelement <2 x bfloat> %3268, bfloat %2934, i64 1, !dbg !40 + %3270 = bitcast <2 x bfloat> %3269 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3261, i32 %3264, i32 %3267, i32 %3270, ptr addrspace(1) %2446, i1 %513) #7, !dbg !40 + %3271 = insertelement <2 x bfloat> poison, bfloat %2935, i64 0, !dbg !40 + %3272 = insertelement <2 x bfloat> %3271, bfloat %2936, i64 1, !dbg !40 + %3273 = bitcast <2 x bfloat> %3272 to i32, !dbg !40 + %3274 = insertelement <2 x bfloat> poison, bfloat %2937, i64 0, !dbg !40 + %3275 = insertelement <2 x bfloat> %3274, bfloat %2938, i64 1, !dbg !40 + %3276 = bitcast <2 x bfloat> %3275 to i32, !dbg !40 + %3277 = insertelement <2 x bfloat> poison, bfloat %2939, i64 0, !dbg !40 + %3278 = insertelement <2 x bfloat> %3277, bfloat %2940, i64 1, !dbg !40 + %3279 = bitcast <2 x bfloat> %3278 to i32, !dbg !40 + %3280 = insertelement <2 x bfloat> poison, bfloat %2941, i64 0, !dbg !40 + %3281 = insertelement <2 x bfloat> %3280, bfloat %2942, i64 1, !dbg !40 + %3282 = bitcast <2 x bfloat> %3281 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3273, i32 %3276, i32 %3279, i32 %3282, ptr addrspace(1) %2444, i1 %531) #7, !dbg !40 + %3283 = insertelement <2 x bfloat> poison, bfloat %2943, i64 0, !dbg !40 + %3284 = insertelement <2 x bfloat> %3283, bfloat %2944, i64 1, !dbg !40 + %3285 = bitcast <2 x bfloat> %3284 to i32, !dbg !40 + %3286 = insertelement <2 x bfloat> poison, bfloat %2945, i64 0, !dbg !40 + %3287 = insertelement <2 x bfloat> %3286, bfloat %2946, i64 1, !dbg !40 + %3288 = bitcast <2 x bfloat> %3287 to i32, !dbg !40 + %3289 = insertelement <2 x bfloat> poison, bfloat %2947, i64 0, !dbg !40 + %3290 = insertelement <2 x bfloat> %3289, bfloat %2948, i64 1, !dbg !40 + %3291 = bitcast <2 x bfloat> %3290 to i32, !dbg !40 + %3292 = insertelement <2 x bfloat> poison, bfloat %2949, i64 0, !dbg !40 + %3293 = insertelement <2 x bfloat> %3292, bfloat %2950, i64 1, !dbg !40 + %3294 = bitcast <2 x bfloat> %3293 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3285, i32 %3288, i32 %3291, i32 %3294, ptr addrspace(1) %2442, i1 %549) #7, !dbg !40 + %3295 = insertelement <2 x bfloat> poison, bfloat %2951, i64 0, !dbg !40 + %3296 = insertelement <2 x bfloat> %3295, bfloat %2952, i64 1, !dbg !40 + %3297 = bitcast <2 x bfloat> %3296 to i32, !dbg !40 + %3298 = insertelement <2 x bfloat> poison, bfloat %2953, i64 0, !dbg !40 + %3299 = insertelement <2 x bfloat> %3298, bfloat %2954, i64 1, !dbg !40 + %3300 = bitcast <2 x bfloat> %3299 to i32, !dbg !40 + %3301 = insertelement <2 x bfloat> poison, bfloat %2955, i64 0, !dbg !40 + %3302 = insertelement <2 x bfloat> %3301, bfloat %2956, i64 1, !dbg !40 + %3303 = bitcast <2 x bfloat> %3302 to i32, !dbg !40 + %3304 = insertelement <2 x bfloat> poison, bfloat %2957, i64 0, !dbg !40 + %3305 = insertelement <2 x bfloat> %3304, bfloat %2958, i64 1, !dbg !40 + %3306 = bitcast <2 x bfloat> %3305 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3297, i32 %3300, i32 %3303, i32 %3306, ptr addrspace(1) %2440, i1 %567) #7, !dbg !40 + %3307 = insertelement <2 x bfloat> poison, bfloat %2959, i64 0, !dbg !40 + %3308 = insertelement <2 x bfloat> %3307, bfloat %2960, i64 1, !dbg !40 + %3309 = bitcast <2 x bfloat> %3308 to i32, !dbg !40 + %3310 = insertelement <2 x bfloat> poison, bfloat %2961, i64 0, !dbg !40 + %3311 = insertelement <2 x bfloat> %3310, bfloat %2962, i64 1, !dbg !40 + %3312 = bitcast <2 x bfloat> %3311 to i32, !dbg !40 + %3313 = insertelement <2 x bfloat> poison, bfloat %2963, i64 0, !dbg !40 + %3314 = insertelement <2 x bfloat> %3313, bfloat %2964, i64 1, !dbg !40 + %3315 = bitcast <2 x bfloat> %3314 to i32, !dbg !40 + %3316 = insertelement <2 x bfloat> poison, bfloat %2965, i64 0, !dbg !40 + %3317 = insertelement <2 x bfloat> %3316, bfloat %2966, i64 1, !dbg !40 + %3318 = bitcast <2 x bfloat> %3317 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3309, i32 %3312, i32 %3315, i32 %3318, ptr addrspace(1) %2438, i1 %585) #7, !dbg !40 + %3319 = insertelement <2 x bfloat> poison, bfloat %2967, i64 0, !dbg !40 + %3320 = insertelement <2 x bfloat> %3319, bfloat %2968, i64 1, !dbg !40 + %3321 = bitcast <2 x bfloat> %3320 to i32, !dbg !40 + %3322 = insertelement <2 x bfloat> poison, bfloat %2969, i64 0, !dbg !40 + %3323 = insertelement <2 x bfloat> %3322, bfloat %2970, i64 1, !dbg !40 + %3324 = bitcast <2 x bfloat> %3323 to i32, !dbg !40 + %3325 = insertelement <2 x bfloat> poison, bfloat %2971, i64 0, !dbg !40 + %3326 = insertelement <2 x bfloat> %3325, bfloat %2972, i64 1, !dbg !40 + %3327 = bitcast <2 x bfloat> %3326 to i32, !dbg !40 + %3328 = insertelement <2 x bfloat> poison, bfloat %2973, i64 0, !dbg !40 + %3329 = insertelement <2 x bfloat> %3328, bfloat %2974, i64 1, !dbg !40 + %3330 = bitcast <2 x bfloat> %3329 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3321, i32 %3324, i32 %3327, i32 %3330, ptr addrspace(1) %2436, i1 %603) #7, !dbg !40 + %3331 = insertelement <2 x bfloat> poison, bfloat %2975, i64 0, !dbg !40 + %3332 = insertelement <2 x bfloat> %3331, bfloat %2976, i64 1, !dbg !40 + %3333 = bitcast <2 x bfloat> %3332 to i32, !dbg !40 + %3334 = insertelement <2 x bfloat> poison, bfloat %2977, i64 0, !dbg !40 + %3335 = insertelement <2 x bfloat> %3334, bfloat %2978, i64 1, !dbg !40 + %3336 = bitcast <2 x bfloat> %3335 to i32, !dbg !40 + %3337 = insertelement <2 x bfloat> poison, bfloat %2979, i64 0, !dbg !40 + %3338 = insertelement <2 x bfloat> %3337, bfloat %2980, i64 1, !dbg !40 + %3339 = bitcast <2 x bfloat> %3338 to i32, !dbg !40 + %3340 = insertelement <2 x bfloat> poison, bfloat %2981, i64 0, !dbg !40 + %3341 = insertelement <2 x bfloat> %3340, bfloat %2982, i64 1, !dbg !40 + %3342 = bitcast <2 x bfloat> %3341 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3333, i32 %3336, i32 %3339, i32 %3342, ptr addrspace(1) %2434, i1 %621) #7, !dbg !40 + %3343 = insertelement <2 x bfloat> poison, bfloat %2983, i64 0, !dbg !40 + %3344 = insertelement <2 x bfloat> %3343, bfloat %2984, i64 1, !dbg !40 + %3345 = bitcast <2 x bfloat> %3344 to i32, !dbg !40 + %3346 = insertelement <2 x bfloat> poison, bfloat %2985, i64 0, !dbg !40 + %3347 = insertelement <2 x bfloat> %3346, bfloat %2986, i64 1, !dbg !40 + %3348 = bitcast <2 x bfloat> %3347 to i32, !dbg !40 + %3349 = insertelement <2 x bfloat> poison, bfloat %2987, i64 0, !dbg !40 + %3350 = insertelement <2 x bfloat> %3349, bfloat %2988, i64 1, !dbg !40 + %3351 = bitcast <2 x bfloat> %3350 to i32, !dbg !40 + %3352 = insertelement <2 x bfloat> poison, bfloat %2989, i64 0, !dbg !40 + %3353 = insertelement <2 x bfloat> %3352, bfloat %2990, i64 1, !dbg !40 + %3354 = bitcast <2 x bfloat> %3353 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3345, i32 %3348, i32 %3351, i32 %3354, ptr addrspace(1) %2432, i1 %639) #7, !dbg !40 + %3355 = insertelement <2 x bfloat> poison, bfloat %2991, i64 0, !dbg !40 + %3356 = insertelement <2 x bfloat> %3355, bfloat %2992, i64 1, !dbg !40 + %3357 = bitcast <2 x bfloat> %3356 to i32, !dbg !40 + %3358 = insertelement <2 x bfloat> poison, bfloat %2993, i64 0, !dbg !40 + %3359 = insertelement <2 x bfloat> %3358, bfloat %2994, i64 1, !dbg !40 + %3360 = bitcast <2 x bfloat> %3359 to i32, !dbg !40 + %3361 = insertelement <2 x bfloat> poison, bfloat %2995, i64 0, !dbg !40 + %3362 = insertelement <2 x bfloat> %3361, bfloat %2996, i64 1, !dbg !40 + %3363 = bitcast <2 x bfloat> %3362 to i32, !dbg !40 + %3364 = insertelement <2 x bfloat> poison, bfloat %2997, i64 0, !dbg !40 + %3365 = insertelement <2 x bfloat> %3364, bfloat %2998, i64 1, !dbg !40 + %3366 = bitcast <2 x bfloat> %3365 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3357, i32 %3360, i32 %3363, i32 %3366, ptr addrspace(1) %2430, i1 %657) #7, !dbg !40 + %3367 = insertelement <2 x bfloat> poison, bfloat %2999, i64 0, !dbg !40 + %3368 = insertelement <2 x bfloat> %3367, bfloat %3000, i64 1, !dbg !40 + %3369 = bitcast <2 x bfloat> %3368 to i32, !dbg !40 + %3370 = insertelement <2 x bfloat> poison, bfloat %3001, i64 0, !dbg !40 + %3371 = insertelement <2 x bfloat> %3370, bfloat %3002, i64 1, !dbg !40 + %3372 = bitcast <2 x bfloat> %3371 to i32, !dbg !40 + %3373 = insertelement <2 x bfloat> poison, bfloat %3003, i64 0, !dbg !40 + %3374 = insertelement <2 x bfloat> %3373, bfloat %3004, i64 1, !dbg !40 + %3375 = bitcast <2 x bfloat> %3374 to i32, !dbg !40 + %3376 = insertelement <2 x bfloat> poison, bfloat %3005, i64 0, !dbg !40 + %3377 = insertelement <2 x bfloat> %3376, bfloat %3006, i64 1, !dbg !40 + %3378 = bitcast <2 x bfloat> %3377 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3369, i32 %3372, i32 %3375, i32 %3378, ptr addrspace(1) %2428, i1 %675) #7, !dbg !40 + %3379 = insertelement <2 x bfloat> poison, bfloat %3007, i64 0, !dbg !40 + %3380 = insertelement <2 x bfloat> %3379, bfloat %3008, i64 1, !dbg !40 + %3381 = bitcast <2 x bfloat> %3380 to i32, !dbg !40 + %3382 = insertelement <2 x bfloat> poison, bfloat %3009, i64 0, !dbg !40 + %3383 = insertelement <2 x bfloat> %3382, bfloat %3010, i64 1, !dbg !40 + %3384 = bitcast <2 x bfloat> %3383 to i32, !dbg !40 + %3385 = insertelement <2 x bfloat> poison, bfloat %3011, i64 0, !dbg !40 + %3386 = insertelement <2 x bfloat> %3385, bfloat %3012, i64 1, !dbg !40 + %3387 = bitcast <2 x bfloat> %3386 to i32, !dbg !40 + %3388 = insertelement <2 x bfloat> poison, bfloat %3013, i64 0, !dbg !40 + %3389 = insertelement <2 x bfloat> %3388, bfloat %3014, i64 1, !dbg !40 + %3390 = bitcast <2 x bfloat> %3389 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3381, i32 %3384, i32 %3387, i32 %3390, ptr addrspace(1) %2426, i1 %693) #7, !dbg !40 + %3391 = insertelement <2 x bfloat> poison, bfloat %3015, i64 0, !dbg !40 + %3392 = insertelement <2 x bfloat> %3391, bfloat %3016, i64 1, !dbg !40 + %3393 = bitcast <2 x bfloat> %3392 to i32, !dbg !40 + %3394 = insertelement <2 x bfloat> poison, bfloat %3017, i64 0, !dbg !40 + %3395 = insertelement <2 x bfloat> %3394, bfloat %3018, i64 1, !dbg !40 + %3396 = bitcast <2 x bfloat> %3395 to i32, !dbg !40 + %3397 = insertelement <2 x bfloat> poison, bfloat %3019, i64 0, !dbg !40 + %3398 = insertelement <2 x bfloat> %3397, bfloat %3020, i64 1, !dbg !40 + %3399 = bitcast <2 x bfloat> %3398 to i32, !dbg !40 + %3400 = insertelement <2 x bfloat> poison, bfloat %3021, i64 0, !dbg !40 + %3401 = insertelement <2 x bfloat> %3400, bfloat %3022, i64 1, !dbg !40 + %3402 = bitcast <2 x bfloat> %3401 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3393, i32 %3396, i32 %3399, i32 %3402, ptr addrspace(1) %2424, i1 %711) #7, !dbg !40 + %3403 = insertelement <2 x bfloat> poison, bfloat %3023, i64 0, !dbg !40 + %3404 = insertelement <2 x bfloat> %3403, bfloat %3024, i64 1, !dbg !40 + %3405 = bitcast <2 x bfloat> %3404 to i32, !dbg !40 + %3406 = insertelement <2 x bfloat> poison, bfloat %3025, i64 0, !dbg !40 + %3407 = insertelement <2 x bfloat> %3406, bfloat %3026, i64 1, !dbg !40 + %3408 = bitcast <2 x bfloat> %3407 to i32, !dbg !40 + %3409 = insertelement <2 x bfloat> poison, bfloat %3027, i64 0, !dbg !40 + %3410 = insertelement <2 x bfloat> %3409, bfloat %3028, i64 1, !dbg !40 + %3411 = bitcast <2 x bfloat> %3410 to i32, !dbg !40 + %3412 = insertelement <2 x bfloat> poison, bfloat %3029, i64 0, !dbg !40 + %3413 = insertelement <2 x bfloat> %3412, bfloat %3030, i64 1, !dbg !40 + %3414 = bitcast <2 x bfloat> %3413 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3405, i32 %3408, i32 %3411, i32 %3414, ptr addrspace(1) %2422, i1 %729) #7, !dbg !40 + %3415 = shl nuw nsw i32 %12, 3, !dbg !41 + %3416 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %3415, !dbg !41 + %3417 = bitcast float %2487 to i32, !dbg !41 + %3418 = bitcast float %2488 to i32, !dbg !41 + %3419 = bitcast float %2489 to i32, !dbg !41 + %3420 = bitcast float %2490 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %3416, i32 %3417, i32 %3418, i32 %3419, i32 %3420) #7, !dbg !41 + %3421 = getelementptr inbounds nuw i8, ptr addrspace(3) %3416, i32 16, !dbg !41 + %3422 = bitcast float %2491 to i32, !dbg !41 + %3423 = bitcast float %2492 to i32, !dbg !41 + %3424 = bitcast float %2493 to i32, !dbg !41 + %3425 = bitcast float %2494 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %3421, i32 %3422, i32 %3423, i32 %3424, i32 %3425) #7, !dbg !41 + %3426 = getelementptr inbounds nuw i8, ptr addrspace(3) %3416, i32 32, !dbg !41 + %3427 = bitcast float %2495 to i32, !dbg !41 + %3428 = bitcast float %2496 to i32, !dbg !41 + %3429 = bitcast float %2497 to i32, !dbg !41 + %3430 = bitcast float %2498 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %3426, i32 %3427, i32 %3428, i32 %3429, i32 %3430) #7, !dbg !41 + %3431 = getelementptr inbounds nuw i8, ptr addrspace(3) %3416, i32 48, !dbg !41 + %3432 = bitcast float %2499 to i32, !dbg !41 + %3433 = bitcast float %2500 to i32, !dbg !41 + %3434 = bitcast float %2501 to i32, !dbg !41 + %3435 = bitcast float %2502 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %3431, i32 %3432, i32 %3433, i32 %3434, i32 %3435) #7, !dbg !41 + %3436 = getelementptr inbounds nuw i8, ptr addrspace(3) %3416, i32 64, !dbg !41 + %3437 = bitcast float %2503 to i32, !dbg !41 + %3438 = bitcast float %2504 to i32, !dbg !41 + %3439 = bitcast float %2505 to i32, !dbg !41 + %3440 = bitcast float %2506 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %3436, i32 %3437, i32 %3438, i32 %3439, i32 %3440) #7, !dbg !41 + %3441 = getelementptr inbounds nuw i8, ptr addrspace(3) %3416, i32 80, !dbg !41 + %3442 = bitcast float %2507 to i32, !dbg !41 + %3443 = bitcast float %2508 to i32, !dbg !41 + %3444 = bitcast float %2509 to i32, !dbg !41 + %3445 = bitcast float %2510 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %3441, i32 %3442, i32 %3443, i32 %3444, i32 %3445) #7, !dbg !41 + %3446 = getelementptr inbounds nuw i8, ptr addrspace(3) %3416, i32 96, !dbg !41 + %3447 = bitcast float %2511 to i32, !dbg !41 + %3448 = bitcast float %2512 to i32, !dbg !41 + %3449 = bitcast float %2513 to i32, !dbg !41 + %3450 = bitcast float %2514 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %3446, i32 %3447, i32 %3448, i32 %3449, i32 %3450) #7, !dbg !41 + %3451 = getelementptr inbounds nuw i8, ptr addrspace(3) %3416, i32 112, !dbg !41 + %3452 = bitcast float %2515 to i32, !dbg !41 + %3453 = bitcast float %2516 to i32, !dbg !41 + %3454 = bitcast float %2517 to i32, !dbg !41 + %3455 = bitcast float %2518 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %3451, i32 %3452, i32 %3453, i32 %3454, i32 %3455) #7, !dbg !41 + tail call void @llvm.nvvm.bar.warp.sync(i32 -1), !dbg !41 + %3456 = shl nuw nsw i32 %40, 4, !dbg !41 + %3457 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %3456, !dbg !41 + %3458 = tail call { i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.b16.p3(ptr addrspace(3) %3457), !dbg !41 + %3459 = extractvalue { i32, i32 } %3458, 0, !dbg !41 + %3460 = extractvalue { i32, i32 } %3458, 1, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %3459, ptr addrspace(1) %2420, i1 %2486) #7, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %3460, ptr addrspace(1) %2417, i1 %2485) #7, !dbg !41 + ret void, !dbg !42 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.div.full(float, float) #3 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.bar.warp.sync(i32) #4 + +; Function Attrs: convergent nocallback nofree nounwind memory(argmem: read) +declare { i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.b16.p3(ptr addrspace(3) readonly captures(none)) #5 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #6 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.f(float) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="32" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) } +attributes #4 = { convergent nocallback nounwind } +attributes #5 = { convergent nocallback nofree nounwind memory(argmem: read) } +attributes #6 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #7 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "l2norm_fwd_kernel", linkageName: "l2norm_fwd_kernel", scope: !1, file: !1, line: 69, type: !6, scopeLine: 69, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !10, !11, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 70, column: 24, scope: !5) +!14 = !DILocation(line: 70, column: 30, scope: !5) +!15 = !DILocation(line: 71, column: 16, scope: !5) +!16 = !DILocation(line: 71, column: 34, scope: !5) +!17 = !DILocation(line: 71, column: 21, scope: !5) +!18 = !DILocation(line: 73, column: 16, scope: !5) +!19 = !DILocation(line: 74, column: 30, scope: !5) +!20 = !DILocation(line: 75, column: 29, scope: !5) +!21 = !DILocation(line: 75, column: 14, scope: !5) +!22 = !DILocation(line: 75, column: 33, scope: !5) +!23 = !DILocation(line: 79, column: 18, scope: !5) +!24 = !DILocation(line: 79, column: 47, scope: !5) +!25 = !DILocation(line: 80, column: 38, scope: !5) +!26 = !DILocation(line: 263, column: 15, scope: !27, inlinedAt: !29) +!27 = distinct !DILexicalBlockFile(scope: !5, file: !28, discriminator: 0) +!28 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!29 = !DILocation(line: 293, column: 36, scope: !27, inlinedAt: !30) +!30 = !DILocation(line: 80, column: 43, scope: !31) +!31 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!32 = !DILocation(line: 80, column: 48, scope: !5) +!33 = !DILocation(line: 80, column: 25, scope: !5) +!34 = !DILocation(line: 77, column: 20, scope: !5) +!35 = !DILocation(line: 76, column: 14, scope: !5) +!36 = !DILocation(line: 76, column: 33, scope: !5) +!37 = !DILocation(line: 80, column: 17, scope: !5) +!38 = !DILocation(line: 81, column: 16, scope: !5) +!39 = !DILocation(line: 83, column: 25, scope: !5) +!40 = !DILocation(line: 83, column: 18, scope: !5) +!41 = !DILocation(line: 84, column: 21, scope: !5) +!42 = !DILocation(line: 84, column: 4, scope: !5) diff --git a/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.ptx b/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..0bc04fbb0005fa7b3bcfc7cbd415407a1aba40a9 --- /dev/null +++ b/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.ptx @@ -0,0 +1,3012 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_fwd_kernel // -- Begin function l2norm_fwd_kernel +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; +.global .align 1 .b8 _$_str_$_2[17] = {95, 95, 67, 85, 68, 65, 95, 80, 82, 69, 67, 95, 83, 81, 82, 84}; + // @l2norm_fwd_kernel +.visible .entry l2norm_fwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_2, + .param .f32 l2norm_fwd_kernel_param_3, + .param .u32 l2norm_fwd_kernel_param_4, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_6 +) +.reqntid 32 +{ + .reg .pred %p<50>; + .reg .b16 %rs<310>; + .reg .b32 %r<1465>; + .reg .b64 %rd<238>; + .loc 1 69 0 // l2norm.py:69:0 +$L__func_begin0: + .loc 1 69 0 // l2norm.py:69:0 + +// %bb.0: // %__nv_sqrtf.exit + ld.param.b64 %rd67, [l2norm_fwd_kernel_param_0]; + ld.param.b64 %rd68, [l2norm_fwd_kernel_param_1]; +$L__tmp0: + .loc 1 70 24 // l2norm.py:70:24 + mov.u32 %r300, %ctaid.x; + ld.param.b64 %rd69, [l2norm_fwd_kernel_param_2]; + .loc 1 71 16 // l2norm.py:71:16 + mul.wide.u32 %rd70, %r300, 64; + ld.param.b32 %r301, [l2norm_fwd_kernel_param_3]; + ld.param.s32 %rd71, [l2norm_fwd_kernel_param_4]; + .loc 1 71 34 // l2norm.py:71:34 + mov.u32 %r302, %tid.x; + and.b32 %r303, %r302, 16; + bfe.u32 %r304, %r302, 4, 1; + or.b32 %r305, %r304, 2; + .loc 1 71 21 // l2norm.py:71:21 + or.b32 %r306, %r304, 60; + or.b32 %r307, %r304, 62; + or.b32 %r308, %r304, 56; + or.b32 %r309, %r304, 58; + or.b32 %r310, %r304, 52; + or.b32 %r311, %r304, 54; + or.b32 %r312, %r304, 48; + or.b32 %r313, %r304, 50; + or.b32 %r314, %r304, 44; + or.b32 %r315, %r304, 46; + or.b32 %r316, %r304, 40; + or.b32 %r317, %r304, 42; + or.b32 %r318, %r304, 36; + or.b32 %r319, %r304, 38; + or.b32 %r320, %r304, 32; + or.b32 %r321, %r304, 34; + or.b32 %r322, %r304, 16; + or.b32 %r323, %r304, 18; + or.b32 %r324, %r304, 20; + or.b32 %r325, %r304, 22; + or.b32 %r326, %r304, 24; + or.b32 %r327, %r304, 26; + or.b32 %r328, %r304, 28; + or.b32 %r329, %r304, 30; + or.b32 %r330, %r304, 8; + or.b32 %r331, %r304, 10; + or.b32 %r332, %r304, 12; + or.b32 %r333, %r304, 14; + or.b32 %r334, %r304, 4; + or.b32 %r335, %r304, 6; + cvt.u64.u32 %rd72, %r321; + cvt.u64.u32 %rd73, %r320; + cvt.u64.u32 %rd74, %r319; + cvt.u64.u32 %rd75, %r318; + cvt.u64.u32 %rd76, %r317; + cvt.u64.u32 %rd77, %r316; + cvt.u64.u32 %rd78, %r315; + cvt.u64.u32 %rd79, %r314; + cvt.u64.u32 %rd80, %r313; + cvt.u64.u32 %rd81, %r312; + cvt.u64.u32 %rd82, %r311; + cvt.u64.u32 %rd83, %r310; + cvt.u64.u32 %rd84, %r309; + cvt.u64.u32 %rd85, %r308; + cvt.u64.u32 %rd86, %r307; + cvt.u64.u32 %rd87, %r306; + cvt.u64.u32 %rd88, %r329; + cvt.u64.u32 %rd89, %r328; + cvt.u64.u32 %rd90, %r327; + cvt.u64.u32 %rd91, %r326; + cvt.u64.u32 %rd92, %r325; + cvt.u64.u32 %rd93, %r324; + cvt.u64.u32 %rd94, %r323; + cvt.u64.u32 %rd95, %r322; + cvt.u64.u32 %rd96, %r333; + cvt.u64.u32 %rd97, %r332; + cvt.u64.u32 %rd98, %r331; + cvt.u64.u32 %rd99, %r330; + cvt.u64.u32 %rd100, %r335; + cvt.u64.u32 %rd101, %r334; + cvt.u64.u32 %rd102, %r305; + cvt.u64.u32 %rd103, %r304; + or.b64 %rd104, %rd70, %rd103; + or.b64 %rd105, %rd70, %rd102; + or.b64 %rd106, %rd70, %rd101; + or.b64 %rd107, %rd70, %rd100; + or.b64 %rd108, %rd70, %rd99; + or.b64 %rd109, %rd70, %rd98; + or.b64 %rd110, %rd70, %rd97; + or.b64 %rd111, %rd70, %rd96; + or.b64 %rd112, %rd70, %rd95; + or.b64 %rd113, %rd70, %rd94; + or.b64 %rd114, %rd70, %rd93; + or.b64 %rd115, %rd70, %rd92; + or.b64 %rd116, %rd70, %rd91; + or.b64 %rd117, %rd70, %rd90; + or.b64 %rd118, %rd70, %rd89; + or.b64 %rd119, %rd70, %rd88; + or.b64 %rd120, %rd70, %rd87; + or.b64 %rd121, %rd70, %rd86; + or.b64 %rd122, %rd70, %rd85; + or.b64 %rd123, %rd70, %rd84; + or.b64 %rd124, %rd70, %rd83; + or.b64 %rd125, %rd70, %rd82; + or.b64 %rd126, %rd70, %rd81; + or.b64 %rd127, %rd70, %rd80; + or.b64 %rd128, %rd70, %rd79; + or.b64 %rd129, %rd70, %rd78; + or.b64 %rd130, %rd70, %rd77; + or.b64 %rd131, %rd70, %rd76; + or.b64 %rd132, %rd70, %rd75; + or.b64 %rd133, %rd70, %rd74; + or.b64 %rd134, %rd70, %rd73; + or.b64 %rd135, %rd70, %rd72; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p35, %rd135, %rd71; + selp.b16 %rs1, 1, 0, %p35; + shl.b16 %rs2, %rs1, 2; + setp.lt.s64 %p36, %rd134, %rd71; + selp.b16 %rs3, -1, 0, %p36; + shl.b16 %rs4, %rs3, 3; + or.b16 %rs5, %rs4, %rs2; + setp.lt.s64 %p37, %rd133, %rd71; + selp.b16 %rs6, 1, 0, %p37; + setp.lt.s64 %p38, %rd132, %rd71; + selp.b16 %rs7, -1, 0, %p38; + shl.b16 %rs8, %rs7, 1; + or.b16 %rs9, %rs6, %rs8; + and.b16 %rs10, %rs9, 3; + or.b16 %rs11, %rs10, %rs5; + shl.b16 %rs12, %rs11, 12; + setp.lt.s64 %p39, %rd131, %rd71; + selp.b16 %rs13, 1, 0, %p39; + shl.b16 %rs14, %rs13, 2; + setp.lt.s64 %p40, %rd130, %rd71; + selp.b16 %rs15, -1, 0, %p40; + shl.b16 %rs16, %rs15, 3; + or.b16 %rs17, %rs16, %rs14; + setp.lt.s64 %p41, %rd129, %rd71; + selp.b16 %rs18, 1, 0, %p41; + setp.lt.s64 %p42, %rd128, %rd71; + selp.b16 %rs19, -1, 0, %p42; + shl.b16 %rs20, %rs19, 1; + or.b16 %rs21, %rs18, %rs20; + and.b16 %rs22, %rs21, 3; + or.b16 %rs23, %rs22, %rs17; + and.b16 %rs24, %rs23, 15; + shl.b16 %rs25, %rs24, 8; + or.b16 %rs26, %rs12, %rs25; + setp.lt.s64 %p43, %rd127, %rd71; + selp.b16 %rs27, 1, 0, %p43; + shl.b16 %rs28, %rs27, 2; + setp.lt.s64 %p44, %rd126, %rd71; + selp.b16 %rs29, -1, 0, %p44; + shl.b16 %rs30, %rs29, 3; + or.b16 %rs31, %rs30, %rs28; + setp.lt.s64 %p45, %rd125, %rd71; + selp.b16 %rs32, 1, 0, %p45; + setp.lt.s64 %p46, %rd124, %rd71; + selp.b16 %rs33, -1, 0, %p46; + shl.b16 %rs34, %rs33, 1; + or.b16 %rs35, %rs32, %rs34; + and.b16 %rs36, %rs35, 3; + or.b16 %rs37, %rs36, %rs31; + shl.b16 %rs38, %rs37, 4; + setp.lt.s64 %p47, %rd123, %rd71; + selp.b16 %rs39, 1, 0, %p47; + shl.b16 %rs40, %rs39, 2; + setp.lt.s64 %p48, %rd122, %rd71; + selp.b16 %rs41, -1, 0, %p48; + shl.b16 %rs42, %rs41, 3; + or.b16 %rs43, %rs42, %rs40; + setp.lt.s64 %p32, %rd121, %rd71; + selp.b16 %rs44, 1, 0, %p32; + setp.lt.s64 %p49, %rd120, %rd71; + selp.b16 %rs45, -1, 0, %p49; + shl.b16 %rs46, %rs45, 1; + or.b16 %rs47, %rs44, %rs46; + and.b16 %rs48, %rs47, 3; + or.b16 %rs49, %rs48, %rs43; + and.b16 %rs50, %rs49, 15; + or.b16 %rs51, %rs50, %rs38; + and.b16 %rs52, %rs51, 255; + or.b16 %rs53, %rs52, %rs26; + cvt.u32.u16 %r336, %rs53; + setp.lt.s64 %p16, %rd119, %rd71; + setp.lt.s64 %p15, %rd118, %rd71; + setp.lt.s64 %p14, %rd117, %rd71; + setp.lt.s64 %p13, %rd116, %rd71; + setp.lt.s64 %p12, %rd115, %rd71; + setp.lt.s64 %p11, %rd114, %rd71; + setp.lt.s64 %p10, %rd113, %rd71; + setp.lt.s64 %p9, %rd112, %rd71; + setp.lt.s64 %p1, %rd111, %rd71; + setp.lt.s64 %p8, %rd110, %rd71; + setp.lt.s64 %p7, %rd109, %rd71; + setp.lt.s64 %p6, %rd108, %rd71; + setp.lt.s64 %p5, %rd107, %rd71; + setp.lt.s64 %p4, %rd106, %rd71; + setp.lt.s64 %p3, %rd105, %rd71; + setp.lt.s64 %p2, %rd104, %rd71; + .loc 1 74 30 // l2norm.py:74:30 + and.b32 %r337, %r302, 15; + shl.b32 %r338, %r337, 3; + .loc 1 75 14 // l2norm.py:75:14 + shl.b64 %rd136, %rd104, 8; + add.s64 %rd137, %rd67, %rd136; + shl.b64 %rd138, %rd105, 8; + add.s64 %rd139, %rd67, %rd138; + shl.b64 %rd140, %rd106, 8; + add.s64 %rd141, %rd67, %rd140; + shl.b64 %rd142, %rd107, 8; + add.s64 %rd143, %rd67, %rd142; + shl.b64 %rd144, %rd108, 8; + add.s64 %rd145, %rd67, %rd144; + shl.b64 %rd146, %rd109, 8; + add.s64 %rd147, %rd67, %rd146; + shl.b64 %rd148, %rd110, 8; + add.s64 %rd149, %rd67, %rd148; + shl.b64 %rd150, %rd111, 8; + add.s64 %rd151, %rd67, %rd150; + shl.b64 %rd152, %rd112, 8; + add.s64 %rd153, %rd67, %rd152; + shl.b64 %rd154, %rd113, 8; + add.s64 %rd155, %rd67, %rd154; + shl.b64 %rd156, %rd114, 8; + add.s64 %rd157, %rd67, %rd156; + shl.b64 %rd158, %rd115, 8; + add.s64 %rd159, %rd67, %rd158; + shl.b64 %rd160, %rd116, 8; + add.s64 %rd161, %rd67, %rd160; + shl.b64 %rd162, %rd117, 8; + add.s64 %rd163, %rd67, %rd162; + shl.b64 %rd164, %rd118, 8; + add.s64 %rd165, %rd67, %rd164; + shl.b64 %rd166, %rd119, 8; + add.s64 %rd167, %rd67, %rd166; + shl.b64 %rd168, %rd134, 8; + add.s64 %rd169, %rd67, %rd168; + shl.b64 %rd170, %rd135, 8; + add.s64 %rd171, %rd67, %rd170; + shl.b64 %rd172, %rd132, 8; + add.s64 %rd173, %rd67, %rd172; + shl.b64 %rd174, %rd133, 8; + add.s64 %rd175, %rd67, %rd174; + shl.b64 %rd176, %rd130, 8; + add.s64 %rd177, %rd67, %rd176; + shl.b64 %rd178, %rd131, 8; + add.s64 %rd179, %rd67, %rd178; + shl.b64 %rd180, %rd128, 8; + add.s64 %rd181, %rd67, %rd180; + shl.b64 %rd182, %rd129, 8; + add.s64 %rd183, %rd67, %rd182; + shl.b64 %rd184, %rd126, 8; + add.s64 %rd185, %rd67, %rd184; + shl.b64 %rd186, %rd127, 8; + add.s64 %rd187, %rd67, %rd186; + shl.b64 %rd188, %rd124, 8; + add.s64 %rd189, %rd67, %rd188; + shl.b64 %rd190, %rd125, 8; + add.s64 %rd191, %rd67, %rd190; + shl.b64 %rd192, %rd122, 8; + add.s64 %rd193, %rd67, %rd192; + shl.b64 %rd194, %rd123, 8; + add.s64 %rd195, %rd67, %rd194; + shl.b64 %rd196, %rd120, 8; + add.s64 %rd197, %rd67, %rd196; + shl.b64 %rd198, %rd121, 8; + add.s64 %rd199, %rd67, %rd198; + .loc 1 75 33 // l2norm.py:75:33 + mul.wide.u32 %rd200, %r338, 2; + add.s64 %rd1, %rd137, %rd200; + add.s64 %rd2, %rd139, %rd200; + add.s64 %rd3, %rd141, %rd200; + add.s64 %rd4, %rd143, %rd200; + add.s64 %rd5, %rd145, %rd200; + add.s64 %rd6, %rd147, %rd200; + add.s64 %rd7, %rd149, %rd200; + add.s64 %rd8, %rd151, %rd200; + add.s64 %rd9, %rd153, %rd200; + add.s64 %rd10, %rd155, %rd200; + add.s64 %rd11, %rd157, %rd200; + add.s64 %rd12, %rd159, %rd200; + add.s64 %rd13, %rd161, %rd200; + add.s64 %rd14, %rd163, %rd200; + add.s64 %rd15, %rd165, %rd200; + add.s64 %rd16, %rd167, %rd200; + add.s64 %rd17, %rd169, %rd200; + add.s64 %rd18, %rd171, %rd200; + add.s64 %rd19, %rd173, %rd200; + add.s64 %rd20, %rd175, %rd200; + add.s64 %rd21, %rd177, %rd200; + add.s64 %rd22, %rd179, %rd200; + add.s64 %rd23, %rd181, %rd200; + add.s64 %rd24, %rd183, %rd200; + add.s64 %rd25, %rd185, %rd200; + add.s64 %rd26, %rd187, %rd200; + add.s64 %rd27, %rd189, %rd200; + add.s64 %rd28, %rd191, %rd200; + add.s64 %rd29, %rd193, %rd200; + add.s64 %rd30, %rd195, %rd200; + add.s64 %rd31, %rd197, %rd200; + add.s64 %rd32, %rd199, %rd200; + mov.b32 %r5, 0; + .loc 1 79 18 // l2norm.py:79:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p2 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs54, %rs55}, %r1; + mov.b32 {%rs56, %rs57}, %r2; + mov.b32 {%rs58, %rs59}, %r3; + mov.b32 {%rs60, %rs61}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p3 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs62, %rs63}, %r6; + mov.b32 {%rs64, %rs65}, %r7; + mov.b32 {%rs66, %rs67}, %r8; + mov.b32 {%rs68, %rs69}, %r9; + // begin inline asm + mov.u32 %r10, %r5; + mov.u32 %r11, %r5; + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + @%p4 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs70, %rs71}, %r10; + mov.b32 {%rs72, %rs73}, %r11; + mov.b32 {%rs74, %rs75}, %r12; + mov.b32 {%rs76, %rs77}, %r13; + // begin inline asm + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + @%p5 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ]; + // end inline asm + mov.b32 {%rs78, %rs79}, %r14; + mov.b32 {%rs80, %rs81}, %r15; + mov.b32 {%rs82, %rs83}, %r16; + mov.b32 {%rs84, %rs85}, %r17; + // begin inline asm + mov.u32 %r18, %r5; + mov.u32 %r19, %r5; + mov.u32 %r20, %r5; + mov.u32 %r21, %r5; + @%p6 ld.global.v4.b32 { %r18, %r19, %r20, %r21 }, [ %rd5 + 0 ]; + // end inline asm + mov.b32 {%rs86, %rs87}, %r18; + mov.b32 {%rs88, %rs89}, %r19; + mov.b32 {%rs90, %rs91}, %r20; + mov.b32 {%rs92, %rs93}, %r21; + // begin inline asm + mov.u32 %r22, %r5; + mov.u32 %r23, %r5; + mov.u32 %r24, %r5; + mov.u32 %r25, %r5; + @%p7 ld.global.v4.b32 { %r22, %r23, %r24, %r25 }, [ %rd6 + 0 ]; + // end inline asm + mov.b32 {%rs94, %rs95}, %r22; + mov.b32 {%rs96, %rs97}, %r23; + mov.b32 {%rs98, %rs99}, %r24; + mov.b32 {%rs100, %rs101}, %r25; + // begin inline asm + mov.u32 %r26, %r5; + mov.u32 %r27, %r5; + mov.u32 %r28, %r5; + mov.u32 %r29, %r5; + @%p8 ld.global.v4.b32 { %r26, %r27, %r28, %r29 }, [ %rd7 + 0 ]; + // end inline asm + mov.b32 {%rs102, %rs103}, %r26; + mov.b32 {%rs104, %rs105}, %r27; + mov.b32 {%rs106, %rs107}, %r28; + mov.b32 {%rs108, %rs109}, %r29; + // begin inline asm + mov.u32 %r30, %r5; + mov.u32 %r31, %r5; + mov.u32 %r32, %r5; + mov.u32 %r33, %r5; + @%p1 ld.global.v4.b32 { %r30, %r31, %r32, %r33 }, [ %rd8 + 0 ]; + // end inline asm + mov.b32 {%rs110, %rs111}, %r30; + mov.b32 {%rs112, %rs113}, %r31; + mov.b32 {%rs114, %rs115}, %r32; + mov.b32 {%rs116, %rs117}, %r33; + // begin inline asm + mov.u32 %r34, %r5; + mov.u32 %r35, %r5; + mov.u32 %r36, %r5; + mov.u32 %r37, %r5; + @%p9 ld.global.v4.b32 { %r34, %r35, %r36, %r37 }, [ %rd9 + 0 ]; + // end inline asm + mov.b32 {%rs118, %rs119}, %r34; + mov.b32 {%rs120, %rs121}, %r35; + mov.b32 {%rs122, %rs123}, %r36; + mov.b32 {%rs124, %rs125}, %r37; + // begin inline asm + mov.u32 %r38, %r5; + mov.u32 %r39, %r5; + mov.u32 %r40, %r5; + mov.u32 %r41, %r5; + @%p10 ld.global.v4.b32 { %r38, %r39, %r40, %r41 }, [ %rd10 + 0 ]; + // end inline asm + mov.b32 {%rs126, %rs127}, %r38; + mov.b32 {%rs128, %rs129}, %r39; + mov.b32 {%rs130, %rs131}, %r40; + mov.b32 {%rs132, %rs133}, %r41; + // begin inline asm + mov.u32 %r42, %r5; + mov.u32 %r43, %r5; + mov.u32 %r44, %r5; + mov.u32 %r45, %r5; + @%p11 ld.global.v4.b32 { %r42, %r43, %r44, %r45 }, [ %rd11 + 0 ]; + // end inline asm + mov.b32 {%rs134, %rs135}, %r42; + mov.b32 {%rs136, %rs137}, %r43; + mov.b32 {%rs138, %rs139}, %r44; + mov.b32 {%rs140, %rs141}, %r45; + // begin inline asm + mov.u32 %r46, %r5; + mov.u32 %r47, %r5; + mov.u32 %r48, %r5; + mov.u32 %r49, %r5; + @%p12 ld.global.v4.b32 { %r46, %r47, %r48, %r49 }, [ %rd12 + 0 ]; + // end inline asm + mov.b32 {%rs142, %rs143}, %r46; + mov.b32 {%rs144, %rs145}, %r47; + mov.b32 {%rs146, %rs147}, %r48; + mov.b32 {%rs148, %rs149}, %r49; + // begin inline asm + mov.u32 %r50, %r5; + mov.u32 %r51, %r5; + mov.u32 %r52, %r5; + mov.u32 %r53, %r5; + @%p13 ld.global.v4.b32 { %r50, %r51, %r52, %r53 }, [ %rd13 + 0 ]; + // end inline asm + mov.b32 {%rs150, %rs151}, %r50; + mov.b32 {%rs152, %rs153}, %r51; + mov.b32 {%rs154, %rs155}, %r52; + mov.b32 {%rs156, %rs157}, %r53; + // begin inline asm + mov.u32 %r54, %r5; + mov.u32 %r55, %r5; + mov.u32 %r56, %r5; + mov.u32 %r57, %r5; + @%p14 ld.global.v4.b32 { %r54, %r55, %r56, %r57 }, [ %rd14 + 0 ]; + // end inline asm + mov.b32 {%rs158, %rs159}, %r54; + mov.b32 {%rs160, %rs161}, %r55; + mov.b32 {%rs162, %rs163}, %r56; + mov.b32 {%rs164, %rs165}, %r57; + // begin inline asm + mov.u32 %r58, %r5; + mov.u32 %r59, %r5; + mov.u32 %r60, %r5; + mov.u32 %r61, %r5; + @%p15 ld.global.v4.b32 { %r58, %r59, %r60, %r61 }, [ %rd15 + 0 ]; + // end inline asm + mov.b32 {%rs166, %rs167}, %r58; + mov.b32 {%rs168, %rs169}, %r59; + mov.b32 {%rs170, %rs171}, %r60; + mov.b32 {%rs172, %rs173}, %r61; + // begin inline asm + mov.u32 %r62, %r5; + mov.u32 %r63, %r5; + mov.u32 %r64, %r5; + mov.u32 %r65, %r5; + @%p16 ld.global.v4.b32 { %r62, %r63, %r64, %r65 }, [ %rd16 + 0 ]; + // end inline asm + mov.b32 {%rs174, %rs175}, %r62; + mov.b32 {%rs176, %rs177}, %r63; + mov.b32 {%rs178, %rs179}, %r64; + mov.b32 {%rs180, %rs181}, %r65; + shr.u32 %r339, %r336, 15; + and.b32 %r340, %r339, 1; + setp.ne.b32 %p17, %r340, 0; + // begin inline asm + mov.u32 %r66, %r5; + mov.u32 %r67, %r5; + mov.u32 %r68, %r5; + mov.u32 %r69, %r5; + @%p17 ld.global.v4.b32 { %r66, %r67, %r68, %r69 }, [ %rd17 + 0 ]; + // end inline asm + mov.b32 {%rs182, %rs183}, %r66; + mov.b32 {%rs184, %rs185}, %r67; + mov.b32 {%rs186, %rs187}, %r68; + mov.b32 {%rs188, %rs189}, %r69; + shr.u32 %r341, %r336, 14; + and.b32 %r342, %r341, 1; + setp.ne.b32 %p18, %r342, 0; + // begin inline asm + mov.u32 %r70, %r5; + mov.u32 %r71, %r5; + mov.u32 %r72, %r5; + mov.u32 %r73, %r5; + @%p18 ld.global.v4.b32 { %r70, %r71, %r72, %r73 }, [ %rd18 + 0 ]; + // end inline asm + mov.b32 {%rs190, %rs191}, %r70; + mov.b32 {%rs192, %rs193}, %r71; + mov.b32 {%rs194, %rs195}, %r72; + mov.b32 {%rs196, %rs197}, %r73; + shr.u32 %r343, %r336, 13; + and.b32 %r344, %r343, 1; + setp.ne.b32 %p19, %r344, 0; + // begin inline asm + mov.u32 %r74, %r5; + mov.u32 %r75, %r5; + mov.u32 %r76, %r5; + mov.u32 %r77, %r5; + @%p19 ld.global.v4.b32 { %r74, %r75, %r76, %r77 }, [ %rd19 + 0 ]; + // end inline asm + mov.b32 {%rs198, %rs199}, %r74; + mov.b32 {%rs200, %rs201}, %r75; + mov.b32 {%rs202, %rs203}, %r76; + mov.b32 {%rs204, %rs205}, %r77; + shr.u32 %r345, %r336, 12; + and.b32 %r346, %r345, 1; + setp.ne.b32 %p20, %r346, 0; + // begin inline asm + mov.u32 %r78, %r5; + mov.u32 %r79, %r5; + mov.u32 %r80, %r5; + mov.u32 %r81, %r5; + @%p20 ld.global.v4.b32 { %r78, %r79, %r80, %r81 }, [ %rd20 + 0 ]; + // end inline asm + mov.b32 {%rs206, %rs207}, %r78; + mov.b32 {%rs208, %rs209}, %r79; + mov.b32 {%rs210, %rs211}, %r80; + mov.b32 {%rs212, %rs213}, %r81; + shr.u32 %r347, %r336, 11; + and.b32 %r348, %r347, 1; + setp.ne.b32 %p21, %r348, 0; + // begin inline asm + mov.u32 %r82, %r5; + mov.u32 %r83, %r5; + mov.u32 %r84, %r5; + mov.u32 %r85, %r5; + @%p21 ld.global.v4.b32 { %r82, %r83, %r84, %r85 }, [ %rd21 + 0 ]; + // end inline asm + mov.b32 {%rs214, %rs215}, %r82; + mov.b32 {%rs216, %rs217}, %r83; + mov.b32 {%rs218, %rs219}, %r84; + mov.b32 {%rs220, %rs221}, %r85; + shr.u32 %r349, %r336, 10; + and.b32 %r350, %r349, 1; + setp.ne.b32 %p22, %r350, 0; + // begin inline asm + mov.u32 %r86, %r5; + mov.u32 %r87, %r5; + mov.u32 %r88, %r5; + mov.u32 %r89, %r5; + @%p22 ld.global.v4.b32 { %r86, %r87, %r88, %r89 }, [ %rd22 + 0 ]; + // end inline asm + mov.b32 {%rs222, %rs223}, %r86; + mov.b32 {%rs224, %rs225}, %r87; + mov.b32 {%rs226, %rs227}, %r88; + mov.b32 {%rs228, %rs229}, %r89; + shr.u32 %r351, %r336, 9; + and.b32 %r352, %r351, 1; + setp.ne.b32 %p23, %r352, 0; + // begin inline asm + mov.u32 %r90, %r5; + mov.u32 %r91, %r5; + mov.u32 %r92, %r5; + mov.u32 %r93, %r5; + @%p23 ld.global.v4.b32 { %r90, %r91, %r92, %r93 }, [ %rd23 + 0 ]; + // end inline asm + mov.b32 {%rs230, %rs231}, %r90; + mov.b32 {%rs232, %rs233}, %r91; + mov.b32 {%rs234, %rs235}, %r92; + mov.b32 {%rs236, %rs237}, %r93; + shr.u32 %r353, %r336, 8; + and.b32 %r354, %r353, 1; + setp.ne.b32 %p24, %r354, 0; + // begin inline asm + mov.u32 %r94, %r5; + mov.u32 %r95, %r5; + mov.u32 %r96, %r5; + mov.u32 %r97, %r5; + @%p24 ld.global.v4.b32 { %r94, %r95, %r96, %r97 }, [ %rd24 + 0 ]; + // end inline asm + mov.b32 {%rs238, %rs239}, %r94; + mov.b32 {%rs240, %rs241}, %r95; + mov.b32 {%rs242, %rs243}, %r96; + mov.b32 {%rs244, %rs245}, %r97; + shr.u32 %r355, %r336, 7; + and.b32 %r356, %r355, 1; + setp.ne.b32 %p25, %r356, 0; + // begin inline asm + mov.u32 %r98, %r5; + mov.u32 %r99, %r5; + mov.u32 %r100, %r5; + mov.u32 %r101, %r5; + @%p25 ld.global.v4.b32 { %r98, %r99, %r100, %r101 }, [ %rd25 + 0 ]; + // end inline asm + mov.b32 {%rs246, %rs247}, %r98; + mov.b32 {%rs248, %rs249}, %r99; + mov.b32 {%rs250, %rs251}, %r100; + mov.b32 {%rs252, %rs253}, %r101; + shr.u32 %r357, %r336, 6; + and.b32 %r358, %r357, 1; + setp.ne.b32 %p26, %r358, 0; + // begin inline asm + mov.u32 %r102, %r5; + mov.u32 %r103, %r5; + mov.u32 %r104, %r5; + mov.u32 %r105, %r5; + @%p26 ld.global.v4.b32 { %r102, %r103, %r104, %r105 }, [ %rd26 + 0 ]; + // end inline asm + mov.b32 {%rs254, %rs255}, %r102; + mov.b32 {%rs256, %rs257}, %r103; + mov.b32 {%rs258, %rs259}, %r104; + mov.b32 {%rs260, %rs261}, %r105; + shr.u32 %r359, %r336, 5; + and.b32 %r360, %r359, 1; + setp.ne.b32 %p27, %r360, 0; + // begin inline asm + mov.u32 %r106, %r5; + mov.u32 %r107, %r5; + mov.u32 %r108, %r5; + mov.u32 %r109, %r5; + @%p27 ld.global.v4.b32 { %r106, %r107, %r108, %r109 }, [ %rd27 + 0 ]; + // end inline asm + mov.b32 {%rs262, %rs263}, %r106; + mov.b32 {%rs264, %rs265}, %r107; + mov.b32 {%rs266, %rs267}, %r108; + mov.b32 {%rs268, %rs269}, %r109; + shr.u32 %r361, %r336, 4; + and.b32 %r362, %r361, 1; + setp.ne.b32 %p28, %r362, 0; + // begin inline asm + mov.u32 %r110, %r5; + mov.u32 %r111, %r5; + mov.u32 %r112, %r5; + mov.u32 %r113, %r5; + @%p28 ld.global.v4.b32 { %r110, %r111, %r112, %r113 }, [ %rd28 + 0 ]; + // end inline asm + mov.b32 {%rs270, %rs271}, %r110; + mov.b32 {%rs272, %rs273}, %r111; + mov.b32 {%rs274, %rs275}, %r112; + mov.b32 {%rs276, %rs277}, %r113; + shr.u32 %r363, %r336, 3; + and.b32 %r364, %r363, 1; + setp.ne.b32 %p29, %r364, 0; + // begin inline asm + mov.u32 %r114, %r5; + mov.u32 %r115, %r5; + mov.u32 %r116, %r5; + mov.u32 %r117, %r5; + @%p29 ld.global.v4.b32 { %r114, %r115, %r116, %r117 }, [ %rd29 + 0 ]; + // end inline asm + mov.b32 {%rs278, %rs279}, %r114; + mov.b32 {%rs280, %rs281}, %r115; + mov.b32 {%rs282, %rs283}, %r116; + mov.b32 {%rs284, %rs285}, %r117; + shr.u32 %r365, %r336, 2; + and.b32 %r366, %r365, 1; + setp.ne.b32 %p30, %r366, 0; + // begin inline asm + mov.u32 %r118, %r5; + mov.u32 %r119, %r5; + mov.u32 %r120, %r5; + mov.u32 %r121, %r5; + @%p30 ld.global.v4.b32 { %r118, %r119, %r120, %r121 }, [ %rd30 + 0 ]; + // end inline asm + mov.b32 {%rs286, %rs287}, %r118; + mov.b32 {%rs288, %rs289}, %r119; + mov.b32 {%rs290, %rs291}, %r120; + mov.b32 {%rs292, %rs293}, %r121; + shr.u32 %r367, %r336, 1; + and.b32 %r368, %r367, 1; + setp.ne.b32 %p31, %r368, 0; + // begin inline asm + mov.u32 %r122, %r5; + mov.u32 %r123, %r5; + mov.u32 %r124, %r5; + mov.u32 %r125, %r5; + @%p31 ld.global.v4.b32 { %r122, %r123, %r124, %r125 }, [ %rd31 + 0 ]; + // end inline asm + mov.b32 {%rs294, %rs295}, %r122; + mov.b32 {%rs296, %rs297}, %r123; + mov.b32 {%rs298, %rs299}, %r124; + mov.b32 {%rs300, %rs301}, %r125; + // begin inline asm + mov.u32 %r126, %r5; + mov.u32 %r127, %r5; + mov.u32 %r128, %r5; + mov.u32 %r129, %r5; + @%p32 ld.global.v4.b32 { %r126, %r127, %r128, %r129 }, [ %rd32 + 0 ]; + // end inline asm + mov.b32 {%rs302, %rs303}, %r126; + mov.b32 {%rs304, %rs305}, %r127; + mov.b32 {%rs306, %rs307}, %r128; + mov.b32 {%rs308, %rs309}, %r129; + .loc 1 79 47 // l2norm.py:79:47 + cvt.f32.bf16 %r369, %rs54; + cvt.f32.bf16 %r370, %rs55; + cvt.f32.bf16 %r371, %rs56; + cvt.f32.bf16 %r372, %rs57; + cvt.f32.bf16 %r373, %rs58; + cvt.f32.bf16 %r374, %rs59; + cvt.f32.bf16 %r375, %rs60; + cvt.f32.bf16 %r376, %rs61; + cvt.f32.bf16 %r377, %rs62; + cvt.f32.bf16 %r378, %rs63; + cvt.f32.bf16 %r379, %rs64; + cvt.f32.bf16 %r380, %rs65; + cvt.f32.bf16 %r381, %rs66; + cvt.f32.bf16 %r382, %rs67; + cvt.f32.bf16 %r383, %rs68; + cvt.f32.bf16 %r384, %rs69; + cvt.f32.bf16 %r385, %rs70; + cvt.f32.bf16 %r386, %rs71; + cvt.f32.bf16 %r387, %rs72; + cvt.f32.bf16 %r388, %rs73; + cvt.f32.bf16 %r389, %rs74; + cvt.f32.bf16 %r390, %rs75; + cvt.f32.bf16 %r391, %rs76; + cvt.f32.bf16 %r392, %rs77; + cvt.f32.bf16 %r393, %rs78; + cvt.f32.bf16 %r394, %rs79; + cvt.f32.bf16 %r395, %rs80; + cvt.f32.bf16 %r396, %rs81; + cvt.f32.bf16 %r397, %rs82; + cvt.f32.bf16 %r398, %rs83; + cvt.f32.bf16 %r399, %rs84; + cvt.f32.bf16 %r400, %rs85; + cvt.f32.bf16 %r401, %rs86; + cvt.f32.bf16 %r402, %rs87; + cvt.f32.bf16 %r403, %rs88; + cvt.f32.bf16 %r404, %rs89; + cvt.f32.bf16 %r405, %rs90; + cvt.f32.bf16 %r406, %rs91; + cvt.f32.bf16 %r407, %rs92; + cvt.f32.bf16 %r408, %rs93; + cvt.f32.bf16 %r409, %rs94; + cvt.f32.bf16 %r410, %rs95; + cvt.f32.bf16 %r411, %rs96; + cvt.f32.bf16 %r412, %rs97; + cvt.f32.bf16 %r413, %rs98; + cvt.f32.bf16 %r414, %rs99; + cvt.f32.bf16 %r415, %rs100; + cvt.f32.bf16 %r416, %rs101; + cvt.f32.bf16 %r417, %rs102; + cvt.f32.bf16 %r418, %rs103; + cvt.f32.bf16 %r419, %rs104; + cvt.f32.bf16 %r420, %rs105; + cvt.f32.bf16 %r421, %rs106; + cvt.f32.bf16 %r422, %rs107; + cvt.f32.bf16 %r423, %rs108; + cvt.f32.bf16 %r424, %rs109; + cvt.f32.bf16 %r425, %rs110; + cvt.f32.bf16 %r426, %rs111; + cvt.f32.bf16 %r427, %rs112; + cvt.f32.bf16 %r428, %rs113; + cvt.f32.bf16 %r429, %rs114; + cvt.f32.bf16 %r430, %rs115; + cvt.f32.bf16 %r431, %rs116; + cvt.f32.bf16 %r432, %rs117; + cvt.f32.bf16 %r433, %rs118; + cvt.f32.bf16 %r434, %rs119; + cvt.f32.bf16 %r435, %rs120; + cvt.f32.bf16 %r436, %rs121; + cvt.f32.bf16 %r437, %rs122; + cvt.f32.bf16 %r438, %rs123; + cvt.f32.bf16 %r439, %rs124; + cvt.f32.bf16 %r440, %rs125; + cvt.f32.bf16 %r441, %rs126; + cvt.f32.bf16 %r442, %rs127; + cvt.f32.bf16 %r443, %rs128; + cvt.f32.bf16 %r444, %rs129; + cvt.f32.bf16 %r445, %rs130; + cvt.f32.bf16 %r446, %rs131; + cvt.f32.bf16 %r447, %rs132; + cvt.f32.bf16 %r448, %rs133; + cvt.f32.bf16 %r449, %rs134; + cvt.f32.bf16 %r450, %rs135; + cvt.f32.bf16 %r451, %rs136; + cvt.f32.bf16 %r452, %rs137; + cvt.f32.bf16 %r453, %rs138; + cvt.f32.bf16 %r454, %rs139; + cvt.f32.bf16 %r455, %rs140; + cvt.f32.bf16 %r456, %rs141; + cvt.f32.bf16 %r457, %rs142; + cvt.f32.bf16 %r458, %rs143; + cvt.f32.bf16 %r459, %rs144; + cvt.f32.bf16 %r460, %rs145; + cvt.f32.bf16 %r461, %rs146; + cvt.f32.bf16 %r462, %rs147; + cvt.f32.bf16 %r463, %rs148; + cvt.f32.bf16 %r464, %rs149; + cvt.f32.bf16 %r465, %rs150; + cvt.f32.bf16 %r466, %rs151; + cvt.f32.bf16 %r467, %rs152; + cvt.f32.bf16 %r468, %rs153; + cvt.f32.bf16 %r469, %rs154; + cvt.f32.bf16 %r470, %rs155; + cvt.f32.bf16 %r471, %rs156; + cvt.f32.bf16 %r472, %rs157; + cvt.f32.bf16 %r473, %rs158; + cvt.f32.bf16 %r474, %rs159; + cvt.f32.bf16 %r475, %rs160; + cvt.f32.bf16 %r476, %rs161; + cvt.f32.bf16 %r477, %rs162; + cvt.f32.bf16 %r478, %rs163; + cvt.f32.bf16 %r479, %rs164; + cvt.f32.bf16 %r480, %rs165; + cvt.f32.bf16 %r481, %rs166; + cvt.f32.bf16 %r482, %rs167; + cvt.f32.bf16 %r483, %rs168; + cvt.f32.bf16 %r484, %rs169; + cvt.f32.bf16 %r485, %rs170; + cvt.f32.bf16 %r486, %rs171; + cvt.f32.bf16 %r487, %rs172; + cvt.f32.bf16 %r488, %rs173; + cvt.f32.bf16 %r489, %rs174; + cvt.f32.bf16 %r490, %rs175; + cvt.f32.bf16 %r491, %rs176; + cvt.f32.bf16 %r492, %rs177; + cvt.f32.bf16 %r493, %rs178; + cvt.f32.bf16 %r494, %rs179; + cvt.f32.bf16 %r495, %rs180; + cvt.f32.bf16 %r496, %rs181; + cvt.f32.bf16 %r497, %rs182; + cvt.f32.bf16 %r498, %rs183; + cvt.f32.bf16 %r499, %rs184; + cvt.f32.bf16 %r500, %rs185; + cvt.f32.bf16 %r501, %rs186; + cvt.f32.bf16 %r502, %rs187; + cvt.f32.bf16 %r503, %rs188; + cvt.f32.bf16 %r504, %rs189; + cvt.f32.bf16 %r505, %rs190; + cvt.f32.bf16 %r506, %rs191; + cvt.f32.bf16 %r507, %rs192; + cvt.f32.bf16 %r508, %rs193; + cvt.f32.bf16 %r509, %rs194; + cvt.f32.bf16 %r510, %rs195; + cvt.f32.bf16 %r511, %rs196; + cvt.f32.bf16 %r512, %rs197; + cvt.f32.bf16 %r513, %rs198; + cvt.f32.bf16 %r514, %rs199; + cvt.f32.bf16 %r515, %rs200; + cvt.f32.bf16 %r516, %rs201; + cvt.f32.bf16 %r517, %rs202; + cvt.f32.bf16 %r518, %rs203; + cvt.f32.bf16 %r519, %rs204; + cvt.f32.bf16 %r520, %rs205; + cvt.f32.bf16 %r521, %rs206; + cvt.f32.bf16 %r522, %rs207; + cvt.f32.bf16 %r523, %rs208; + cvt.f32.bf16 %r524, %rs209; + cvt.f32.bf16 %r525, %rs210; + cvt.f32.bf16 %r526, %rs211; + cvt.f32.bf16 %r527, %rs212; + cvt.f32.bf16 %r528, %rs213; + cvt.f32.bf16 %r529, %rs214; + cvt.f32.bf16 %r530, %rs215; + cvt.f32.bf16 %r531, %rs216; + cvt.f32.bf16 %r532, %rs217; + cvt.f32.bf16 %r533, %rs218; + cvt.f32.bf16 %r534, %rs219; + cvt.f32.bf16 %r535, %rs220; + cvt.f32.bf16 %r536, %rs221; + cvt.f32.bf16 %r537, %rs222; + cvt.f32.bf16 %r538, %rs223; + cvt.f32.bf16 %r539, %rs224; + cvt.f32.bf16 %r540, %rs225; + cvt.f32.bf16 %r541, %rs226; + cvt.f32.bf16 %r542, %rs227; + cvt.f32.bf16 %r543, %rs228; + cvt.f32.bf16 %r544, %rs229; + cvt.f32.bf16 %r545, %rs230; + cvt.f32.bf16 %r546, %rs231; + cvt.f32.bf16 %r547, %rs232; + cvt.f32.bf16 %r548, %rs233; + cvt.f32.bf16 %r549, %rs234; + cvt.f32.bf16 %r550, %rs235; + cvt.f32.bf16 %r551, %rs236; + cvt.f32.bf16 %r552, %rs237; + cvt.f32.bf16 %r553, %rs238; + cvt.f32.bf16 %r554, %rs239; + cvt.f32.bf16 %r555, %rs240; + cvt.f32.bf16 %r556, %rs241; + cvt.f32.bf16 %r557, %rs242; + cvt.f32.bf16 %r558, %rs243; + cvt.f32.bf16 %r559, %rs244; + cvt.f32.bf16 %r560, %rs245; + cvt.f32.bf16 %r561, %rs246; + cvt.f32.bf16 %r562, %rs247; + cvt.f32.bf16 %r563, %rs248; + cvt.f32.bf16 %r564, %rs249; + cvt.f32.bf16 %r565, %rs250; + cvt.f32.bf16 %r566, %rs251; + cvt.f32.bf16 %r567, %rs252; + cvt.f32.bf16 %r568, %rs253; + cvt.f32.bf16 %r569, %rs254; + cvt.f32.bf16 %r570, %rs255; + cvt.f32.bf16 %r571, %rs256; + cvt.f32.bf16 %r572, %rs257; + cvt.f32.bf16 %r573, %rs258; + cvt.f32.bf16 %r574, %rs259; + cvt.f32.bf16 %r575, %rs260; + cvt.f32.bf16 %r576, %rs261; + cvt.f32.bf16 %r577, %rs262; + cvt.f32.bf16 %r578, %rs263; + cvt.f32.bf16 %r579, %rs264; + cvt.f32.bf16 %r580, %rs265; + cvt.f32.bf16 %r581, %rs266; + cvt.f32.bf16 %r582, %rs267; + cvt.f32.bf16 %r583, %rs268; + cvt.f32.bf16 %r584, %rs269; + cvt.f32.bf16 %r585, %rs270; + cvt.f32.bf16 %r586, %rs271; + cvt.f32.bf16 %r587, %rs272; + cvt.f32.bf16 %r588, %rs273; + cvt.f32.bf16 %r589, %rs274; + cvt.f32.bf16 %r590, %rs275; + cvt.f32.bf16 %r591, %rs276; + cvt.f32.bf16 %r592, %rs277; + cvt.f32.bf16 %r593, %rs278; + cvt.f32.bf16 %r594, %rs279; + cvt.f32.bf16 %r595, %rs280; + cvt.f32.bf16 %r596, %rs281; + cvt.f32.bf16 %r597, %rs282; + cvt.f32.bf16 %r598, %rs283; + cvt.f32.bf16 %r599, %rs284; + cvt.f32.bf16 %r600, %rs285; + cvt.f32.bf16 %r601, %rs286; + cvt.f32.bf16 %r602, %rs287; + cvt.f32.bf16 %r603, %rs288; + cvt.f32.bf16 %r604, %rs289; + cvt.f32.bf16 %r605, %rs290; + cvt.f32.bf16 %r606, %rs291; + cvt.f32.bf16 %r607, %rs292; + cvt.f32.bf16 %r608, %rs293; + cvt.f32.bf16 %r609, %rs294; + cvt.f32.bf16 %r610, %rs295; + cvt.f32.bf16 %r611, %rs296; + cvt.f32.bf16 %r612, %rs297; + cvt.f32.bf16 %r613, %rs298; + cvt.f32.bf16 %r614, %rs299; + cvt.f32.bf16 %r615, %rs300; + cvt.f32.bf16 %r616, %rs301; + cvt.f32.bf16 %r617, %rs302; + cvt.f32.bf16 %r618, %rs303; + cvt.f32.bf16 %r619, %rs304; + cvt.f32.bf16 %r620, %rs305; + cvt.f32.bf16 %r621, %rs306; + cvt.f32.bf16 %r622, %rs307; + cvt.f32.bf16 %r623, %rs308; + cvt.f32.bf16 %r624, %rs309; + .loc 1 80 38 // l2norm.py:80:38 + mul.f32 %r625, %r370, %r370; + mul.f32 %r626, %r378, %r378; + mul.f32 %r627, %r386, %r386; + mul.f32 %r628, %r394, %r394; + mul.f32 %r629, %r402, %r402; + mul.f32 %r630, %r410, %r410; + mul.f32 %r631, %r418, %r418; + mul.f32 %r632, %r426, %r426; + mul.f32 %r633, %r434, %r434; + mul.f32 %r634, %r442, %r442; + mul.f32 %r635, %r450, %r450; + mul.f32 %r636, %r458, %r458; + mul.f32 %r637, %r466, %r466; + mul.f32 %r638, %r474, %r474; + mul.f32 %r639, %r482, %r482; + mul.f32 %r640, %r490, %r490; + mul.f32 %r641, %r498, %r498; + mul.f32 %r642, %r506, %r506; + mul.f32 %r643, %r514, %r514; + mul.f32 %r644, %r522, %r522; + mul.f32 %r645, %r530, %r530; + mul.f32 %r646, %r538, %r538; + mul.f32 %r647, %r546, %r546; + mul.f32 %r648, %r554, %r554; + mul.f32 %r649, %r562, %r562; + mul.f32 %r650, %r570, %r570; + mul.f32 %r651, %r578, %r578; + mul.f32 %r652, %r586, %r586; + mul.f32 %r653, %r594, %r594; + mul.f32 %r654, %r602, %r602; + mul.f32 %r655, %r610, %r610; + mul.f32 %r656, %r618, %r618; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + fma.rn.f32.bf16 %r657, %rs54, %rs54, %r625; + fma.rn.f32.bf16 %r658, %rs56, %rs56, %r657; + fma.rn.f32.bf16 %r659, %rs57, %rs57, %r658; + fma.rn.f32.bf16 %r660, %rs58, %rs58, %r659; + fma.rn.f32.bf16 %r661, %rs59, %rs59, %r660; + fma.rn.f32.bf16 %r662, %rs60, %rs60, %r661; + fma.rn.f32.bf16 %r663, %rs61, %rs61, %r662; + fma.rn.f32.bf16 %r664, %rs62, %rs62, %r626; + fma.rn.f32.bf16 %r665, %rs64, %rs64, %r664; + fma.rn.f32.bf16 %r666, %rs65, %rs65, %r665; + fma.rn.f32.bf16 %r667, %rs66, %rs66, %r666; + fma.rn.f32.bf16 %r668, %rs67, %rs67, %r667; + fma.rn.f32.bf16 %r669, %rs68, %rs68, %r668; + fma.rn.f32.bf16 %r670, %rs69, %rs69, %r669; + fma.rn.f32.bf16 %r671, %rs70, %rs70, %r627; + fma.rn.f32.bf16 %r672, %rs72, %rs72, %r671; + fma.rn.f32.bf16 %r673, %rs73, %rs73, %r672; + fma.rn.f32.bf16 %r674, %rs74, %rs74, %r673; + fma.rn.f32.bf16 %r675, %rs75, %rs75, %r674; + fma.rn.f32.bf16 %r676, %rs76, %rs76, %r675; + fma.rn.f32.bf16 %r677, %rs77, %rs77, %r676; + fma.rn.f32.bf16 %r678, %rs78, %rs78, %r628; + fma.rn.f32.bf16 %r679, %rs80, %rs80, %r678; + fma.rn.f32.bf16 %r680, %rs81, %rs81, %r679; + fma.rn.f32.bf16 %r681, %rs82, %rs82, %r680; + fma.rn.f32.bf16 %r682, %rs83, %rs83, %r681; + fma.rn.f32.bf16 %r683, %rs84, %rs84, %r682; + fma.rn.f32.bf16 %r684, %rs85, %rs85, %r683; + fma.rn.f32.bf16 %r685, %rs86, %rs86, %r629; + fma.rn.f32.bf16 %r686, %rs88, %rs88, %r685; + fma.rn.f32.bf16 %r687, %rs89, %rs89, %r686; + fma.rn.f32.bf16 %r688, %rs90, %rs90, %r687; + fma.rn.f32.bf16 %r689, %rs91, %rs91, %r688; + fma.rn.f32.bf16 %r690, %rs92, %rs92, %r689; + fma.rn.f32.bf16 %r691, %rs93, %rs93, %r690; + fma.rn.f32.bf16 %r692, %rs94, %rs94, %r630; + fma.rn.f32.bf16 %r693, %rs96, %rs96, %r692; + fma.rn.f32.bf16 %r694, %rs97, %rs97, %r693; + fma.rn.f32.bf16 %r695, %rs98, %rs98, %r694; + fma.rn.f32.bf16 %r696, %rs99, %rs99, %r695; + fma.rn.f32.bf16 %r697, %rs100, %rs100, %r696; + fma.rn.f32.bf16 %r698, %rs101, %rs101, %r697; + fma.rn.f32.bf16 %r699, %rs102, %rs102, %r631; + fma.rn.f32.bf16 %r700, %rs104, %rs104, %r699; + fma.rn.f32.bf16 %r701, %rs105, %rs105, %r700; + fma.rn.f32.bf16 %r702, %rs106, %rs106, %r701; + fma.rn.f32.bf16 %r703, %rs107, %rs107, %r702; + fma.rn.f32.bf16 %r704, %rs108, %rs108, %r703; + fma.rn.f32.bf16 %r705, %rs109, %rs109, %r704; + fma.rn.f32.bf16 %r706, %rs110, %rs110, %r632; + fma.rn.f32.bf16 %r707, %rs112, %rs112, %r706; + fma.rn.f32.bf16 %r708, %rs113, %rs113, %r707; + fma.rn.f32.bf16 %r709, %rs114, %rs114, %r708; + fma.rn.f32.bf16 %r710, %rs115, %rs115, %r709; + fma.rn.f32.bf16 %r711, %rs116, %rs116, %r710; + fma.rn.f32.bf16 %r712, %rs117, %rs117, %r711; + fma.rn.f32.bf16 %r713, %rs118, %rs118, %r633; + fma.rn.f32.bf16 %r714, %rs120, %rs120, %r713; + fma.rn.f32.bf16 %r715, %rs121, %rs121, %r714; + fma.rn.f32.bf16 %r716, %rs122, %rs122, %r715; + fma.rn.f32.bf16 %r717, %rs123, %rs123, %r716; + fma.rn.f32.bf16 %r718, %rs124, %rs124, %r717; + fma.rn.f32.bf16 %r719, %rs125, %rs125, %r718; + fma.rn.f32.bf16 %r720, %rs126, %rs126, %r634; + fma.rn.f32.bf16 %r721, %rs128, %rs128, %r720; + fma.rn.f32.bf16 %r722, %rs129, %rs129, %r721; + fma.rn.f32.bf16 %r723, %rs130, %rs130, %r722; + fma.rn.f32.bf16 %r724, %rs131, %rs131, %r723; + fma.rn.f32.bf16 %r725, %rs132, %rs132, %r724; + fma.rn.f32.bf16 %r726, %rs133, %rs133, %r725; + fma.rn.f32.bf16 %r727, %rs134, %rs134, %r635; + fma.rn.f32.bf16 %r728, %rs136, %rs136, %r727; + fma.rn.f32.bf16 %r729, %rs137, %rs137, %r728; + fma.rn.f32.bf16 %r730, %rs138, %rs138, %r729; + fma.rn.f32.bf16 %r731, %rs139, %rs139, %r730; + fma.rn.f32.bf16 %r732, %rs140, %rs140, %r731; + fma.rn.f32.bf16 %r733, %rs141, %rs141, %r732; + fma.rn.f32.bf16 %r734, %rs142, %rs142, %r636; + fma.rn.f32.bf16 %r735, %rs144, %rs144, %r734; + fma.rn.f32.bf16 %r736, %rs145, %rs145, %r735; + fma.rn.f32.bf16 %r737, %rs146, %rs146, %r736; + fma.rn.f32.bf16 %r738, %rs147, %rs147, %r737; + fma.rn.f32.bf16 %r739, %rs148, %rs148, %r738; + fma.rn.f32.bf16 %r740, %rs149, %rs149, %r739; + fma.rn.f32.bf16 %r741, %rs150, %rs150, %r637; + fma.rn.f32.bf16 %r742, %rs152, %rs152, %r741; + fma.rn.f32.bf16 %r743, %rs153, %rs153, %r742; + fma.rn.f32.bf16 %r744, %rs154, %rs154, %r743; + fma.rn.f32.bf16 %r745, %rs155, %rs155, %r744; + fma.rn.f32.bf16 %r746, %rs156, %rs156, %r745; + fma.rn.f32.bf16 %r747, %rs157, %rs157, %r746; + fma.rn.f32.bf16 %r748, %rs158, %rs158, %r638; + fma.rn.f32.bf16 %r749, %rs160, %rs160, %r748; + fma.rn.f32.bf16 %r750, %rs161, %rs161, %r749; + fma.rn.f32.bf16 %r751, %rs162, %rs162, %r750; + fma.rn.f32.bf16 %r752, %rs163, %rs163, %r751; + fma.rn.f32.bf16 %r753, %rs164, %rs164, %r752; + fma.rn.f32.bf16 %r754, %rs165, %rs165, %r753; + fma.rn.f32.bf16 %r755, %rs166, %rs166, %r639; + fma.rn.f32.bf16 %r756, %rs168, %rs168, %r755; + fma.rn.f32.bf16 %r757, %rs169, %rs169, %r756; + fma.rn.f32.bf16 %r758, %rs170, %rs170, %r757; + fma.rn.f32.bf16 %r759, %rs171, %rs171, %r758; + fma.rn.f32.bf16 %r760, %rs172, %rs172, %r759; + fma.rn.f32.bf16 %r761, %rs173, %rs173, %r760; + fma.rn.f32.bf16 %r762, %rs174, %rs174, %r640; + fma.rn.f32.bf16 %r763, %rs176, %rs176, %r762; + fma.rn.f32.bf16 %r764, %rs177, %rs177, %r763; + fma.rn.f32.bf16 %r765, %rs178, %rs178, %r764; + fma.rn.f32.bf16 %r766, %rs179, %rs179, %r765; + fma.rn.f32.bf16 %r767, %rs180, %rs180, %r766; + fma.rn.f32.bf16 %r768, %rs181, %rs181, %r767; + fma.rn.f32.bf16 %r769, %rs182, %rs182, %r641; + fma.rn.f32.bf16 %r770, %rs184, %rs184, %r769; + fma.rn.f32.bf16 %r771, %rs185, %rs185, %r770; + fma.rn.f32.bf16 %r772, %rs186, %rs186, %r771; + fma.rn.f32.bf16 %r773, %rs187, %rs187, %r772; + fma.rn.f32.bf16 %r774, %rs188, %rs188, %r773; + fma.rn.f32.bf16 %r775, %rs189, %rs189, %r774; + fma.rn.f32.bf16 %r776, %rs190, %rs190, %r642; + fma.rn.f32.bf16 %r777, %rs192, %rs192, %r776; + fma.rn.f32.bf16 %r778, %rs193, %rs193, %r777; + fma.rn.f32.bf16 %r779, %rs194, %rs194, %r778; + fma.rn.f32.bf16 %r780, %rs195, %rs195, %r779; + fma.rn.f32.bf16 %r781, %rs196, %rs196, %r780; + fma.rn.f32.bf16 %r782, %rs197, %rs197, %r781; + fma.rn.f32.bf16 %r783, %rs198, %rs198, %r643; + fma.rn.f32.bf16 %r784, %rs200, %rs200, %r783; + fma.rn.f32.bf16 %r785, %rs201, %rs201, %r784; + fma.rn.f32.bf16 %r786, %rs202, %rs202, %r785; + fma.rn.f32.bf16 %r787, %rs203, %rs203, %r786; + fma.rn.f32.bf16 %r788, %rs204, %rs204, %r787; + fma.rn.f32.bf16 %r789, %rs205, %rs205, %r788; + fma.rn.f32.bf16 %r790, %rs206, %rs206, %r644; + fma.rn.f32.bf16 %r791, %rs208, %rs208, %r790; + fma.rn.f32.bf16 %r792, %rs209, %rs209, %r791; + fma.rn.f32.bf16 %r793, %rs210, %rs210, %r792; + fma.rn.f32.bf16 %r794, %rs211, %rs211, %r793; + fma.rn.f32.bf16 %r795, %rs212, %rs212, %r794; + fma.rn.f32.bf16 %r796, %rs213, %rs213, %r795; + fma.rn.f32.bf16 %r797, %rs214, %rs214, %r645; + fma.rn.f32.bf16 %r798, %rs216, %rs216, %r797; + fma.rn.f32.bf16 %r799, %rs217, %rs217, %r798; + fma.rn.f32.bf16 %r800, %rs218, %rs218, %r799; + fma.rn.f32.bf16 %r801, %rs219, %rs219, %r800; + fma.rn.f32.bf16 %r802, %rs220, %rs220, %r801; + fma.rn.f32.bf16 %r803, %rs221, %rs221, %r802; + fma.rn.f32.bf16 %r804, %rs222, %rs222, %r646; + fma.rn.f32.bf16 %r805, %rs224, %rs224, %r804; + fma.rn.f32.bf16 %r806, %rs225, %rs225, %r805; + fma.rn.f32.bf16 %r807, %rs226, %rs226, %r806; + fma.rn.f32.bf16 %r808, %rs227, %rs227, %r807; + fma.rn.f32.bf16 %r809, %rs228, %rs228, %r808; + fma.rn.f32.bf16 %r810, %rs229, %rs229, %r809; + fma.rn.f32.bf16 %r811, %rs230, %rs230, %r647; + fma.rn.f32.bf16 %r812, %rs232, %rs232, %r811; + fma.rn.f32.bf16 %r813, %rs233, %rs233, %r812; + fma.rn.f32.bf16 %r814, %rs234, %rs234, %r813; + fma.rn.f32.bf16 %r815, %rs235, %rs235, %r814; + fma.rn.f32.bf16 %r816, %rs236, %rs236, %r815; + fma.rn.f32.bf16 %r817, %rs237, %rs237, %r816; + fma.rn.f32.bf16 %r818, %rs238, %rs238, %r648; + fma.rn.f32.bf16 %r819, %rs240, %rs240, %r818; + fma.rn.f32.bf16 %r820, %rs241, %rs241, %r819; + fma.rn.f32.bf16 %r821, %rs242, %rs242, %r820; + fma.rn.f32.bf16 %r822, %rs243, %rs243, %r821; + fma.rn.f32.bf16 %r823, %rs244, %rs244, %r822; + fma.rn.f32.bf16 %r824, %rs245, %rs245, %r823; + fma.rn.f32.bf16 %r825, %rs246, %rs246, %r649; + fma.rn.f32.bf16 %r826, %rs248, %rs248, %r825; + fma.rn.f32.bf16 %r827, %rs249, %rs249, %r826; + fma.rn.f32.bf16 %r828, %rs250, %rs250, %r827; + fma.rn.f32.bf16 %r829, %rs251, %rs251, %r828; + fma.rn.f32.bf16 %r830, %rs252, %rs252, %r829; + fma.rn.f32.bf16 %r831, %rs253, %rs253, %r830; + fma.rn.f32.bf16 %r832, %rs254, %rs254, %r650; + fma.rn.f32.bf16 %r833, %rs256, %rs256, %r832; + fma.rn.f32.bf16 %r834, %rs257, %rs257, %r833; + fma.rn.f32.bf16 %r835, %rs258, %rs258, %r834; + fma.rn.f32.bf16 %r836, %rs259, %rs259, %r835; + fma.rn.f32.bf16 %r837, %rs260, %rs260, %r836; + fma.rn.f32.bf16 %r838, %rs261, %rs261, %r837; + fma.rn.f32.bf16 %r839, %rs262, %rs262, %r651; + fma.rn.f32.bf16 %r840, %rs264, %rs264, %r839; + fma.rn.f32.bf16 %r841, %rs265, %rs265, %r840; + fma.rn.f32.bf16 %r842, %rs266, %rs266, %r841; + fma.rn.f32.bf16 %r843, %rs267, %rs267, %r842; + fma.rn.f32.bf16 %r844, %rs268, %rs268, %r843; + fma.rn.f32.bf16 %r845, %rs269, %rs269, %r844; + fma.rn.f32.bf16 %r846, %rs270, %rs270, %r652; + fma.rn.f32.bf16 %r847, %rs272, %rs272, %r846; + fma.rn.f32.bf16 %r848, %rs273, %rs273, %r847; + fma.rn.f32.bf16 %r849, %rs274, %rs274, %r848; + fma.rn.f32.bf16 %r850, %rs275, %rs275, %r849; + fma.rn.f32.bf16 %r851, %rs276, %rs276, %r850; + fma.rn.f32.bf16 %r852, %rs277, %rs277, %r851; + fma.rn.f32.bf16 %r853, %rs278, %rs278, %r653; + fma.rn.f32.bf16 %r854, %rs280, %rs280, %r853; + fma.rn.f32.bf16 %r855, %rs281, %rs281, %r854; + fma.rn.f32.bf16 %r856, %rs282, %rs282, %r855; + fma.rn.f32.bf16 %r857, %rs283, %rs283, %r856; + fma.rn.f32.bf16 %r858, %rs284, %rs284, %r857; + fma.rn.f32.bf16 %r859, %rs285, %rs285, %r858; + fma.rn.f32.bf16 %r860, %rs286, %rs286, %r654; + fma.rn.f32.bf16 %r861, %rs288, %rs288, %r860; + fma.rn.f32.bf16 %r862, %rs289, %rs289, %r861; + fma.rn.f32.bf16 %r863, %rs290, %rs290, %r862; + fma.rn.f32.bf16 %r864, %rs291, %rs291, %r863; + fma.rn.f32.bf16 %r865, %rs292, %rs292, %r864; + fma.rn.f32.bf16 %r866, %rs293, %rs293, %r865; + fma.rn.f32.bf16 %r867, %rs294, %rs294, %r655; + fma.rn.f32.bf16 %r868, %rs296, %rs296, %r867; + fma.rn.f32.bf16 %r869, %rs297, %rs297, %r868; + fma.rn.f32.bf16 %r870, %rs298, %rs298, %r869; + fma.rn.f32.bf16 %r871, %rs299, %rs299, %r870; + fma.rn.f32.bf16 %r872, %rs300, %rs300, %r871; + fma.rn.f32.bf16 %r873, %rs301, %rs301, %r872; + fma.rn.f32.bf16 %r874, %rs302, %rs302, %r656; + fma.rn.f32.bf16 %r875, %rs304, %rs304, %r874; + fma.rn.f32.bf16 %r876, %rs305, %rs305, %r875; + fma.rn.f32.bf16 %r877, %rs306, %rs306, %r876; + fma.rn.f32.bf16 %r878, %rs307, %rs307, %r877; + fma.rn.f32.bf16 %r879, %rs308, %rs308, %r878; + fma.rn.f32.bf16 %r880, %rs309, %rs309, %r879; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r881, %r663, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r882, %r663, %r881; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r883, %r882, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r884, %r882, %r883; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r885, %r884, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r886, %r884, %r885; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r887, %r886, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r888, %r886, %r887; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r889, %r670, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r890, %r670, %r889; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r891, %r890, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r892, %r890, %r891; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r893, %r892, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r894, %r892, %r893; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r895, %r894, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r896, %r894, %r895; +$L__tmp18: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r897, %r677, 8, 31, -1; +$L__tmp19: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r898, %r677, %r897; +$L__tmp20: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r899, %r898, 4, 31, -1; +$L__tmp21: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r900, %r898, %r899; +$L__tmp22: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r901, %r900, 2, 31, -1; +$L__tmp23: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r902, %r900, %r901; +$L__tmp24: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r903, %r902, 1, 31, -1; +$L__tmp25: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r904, %r902, %r903; +$L__tmp26: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r905, %r684, 8, 31, -1; +$L__tmp27: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r906, %r684, %r905; +$L__tmp28: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r907, %r906, 4, 31, -1; +$L__tmp29: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r908, %r906, %r907; +$L__tmp30: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r909, %r908, 2, 31, -1; +$L__tmp31: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r910, %r908, %r909; +$L__tmp32: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r911, %r910, 1, 31, -1; +$L__tmp33: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r912, %r910, %r911; +$L__tmp34: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r913, %r691, 8, 31, -1; +$L__tmp35: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r914, %r691, %r913; +$L__tmp36: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r915, %r914, 4, 31, -1; +$L__tmp37: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r916, %r914, %r915; +$L__tmp38: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r917, %r916, 2, 31, -1; +$L__tmp39: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r918, %r916, %r917; +$L__tmp40: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r919, %r918, 1, 31, -1; +$L__tmp41: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r920, %r918, %r919; +$L__tmp42: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r921, %r698, 8, 31, -1; +$L__tmp43: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r922, %r698, %r921; +$L__tmp44: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r923, %r922, 4, 31, -1; +$L__tmp45: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r924, %r922, %r923; +$L__tmp46: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r925, %r924, 2, 31, -1; +$L__tmp47: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r926, %r924, %r925; +$L__tmp48: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r927, %r926, 1, 31, -1; +$L__tmp49: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r928, %r926, %r927; +$L__tmp50: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r929, %r705, 8, 31, -1; +$L__tmp51: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r930, %r705, %r929; +$L__tmp52: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r931, %r930, 4, 31, -1; +$L__tmp53: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r932, %r930, %r931; +$L__tmp54: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r933, %r932, 2, 31, -1; +$L__tmp55: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r934, %r932, %r933; +$L__tmp56: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r935, %r934, 1, 31, -1; +$L__tmp57: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r936, %r934, %r935; +$L__tmp58: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r937, %r712, 8, 31, -1; +$L__tmp59: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r938, %r712, %r937; +$L__tmp60: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r939, %r938, 4, 31, -1; +$L__tmp61: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r940, %r938, %r939; +$L__tmp62: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r941, %r940, 2, 31, -1; +$L__tmp63: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r942, %r940, %r941; +$L__tmp64: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r943, %r942, 1, 31, -1; +$L__tmp65: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r944, %r942, %r943; +$L__tmp66: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r945, %r719, 8, 31, -1; +$L__tmp67: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r946, %r719, %r945; +$L__tmp68: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r947, %r946, 4, 31, -1; +$L__tmp69: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r948, %r946, %r947; +$L__tmp70: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r949, %r948, 2, 31, -1; +$L__tmp71: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r950, %r948, %r949; +$L__tmp72: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r951, %r950, 1, 31, -1; +$L__tmp73: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r952, %r950, %r951; +$L__tmp74: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r953, %r726, 8, 31, -1; +$L__tmp75: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r954, %r726, %r953; +$L__tmp76: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r955, %r954, 4, 31, -1; +$L__tmp77: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r956, %r954, %r955; +$L__tmp78: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r957, %r956, 2, 31, -1; +$L__tmp79: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r958, %r956, %r957; +$L__tmp80: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r959, %r958, 1, 31, -1; +$L__tmp81: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r960, %r958, %r959; +$L__tmp82: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r961, %r733, 8, 31, -1; +$L__tmp83: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r962, %r733, %r961; +$L__tmp84: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r963, %r962, 4, 31, -1; +$L__tmp85: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r964, %r962, %r963; +$L__tmp86: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r965, %r964, 2, 31, -1; +$L__tmp87: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r966, %r964, %r965; +$L__tmp88: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r967, %r966, 1, 31, -1; +$L__tmp89: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r968, %r966, %r967; +$L__tmp90: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r969, %r740, 8, 31, -1; +$L__tmp91: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r970, %r740, %r969; +$L__tmp92: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r971, %r970, 4, 31, -1; +$L__tmp93: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r972, %r970, %r971; +$L__tmp94: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r973, %r972, 2, 31, -1; +$L__tmp95: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r974, %r972, %r973; +$L__tmp96: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r975, %r974, 1, 31, -1; +$L__tmp97: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r976, %r974, %r975; +$L__tmp98: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r977, %r747, 8, 31, -1; +$L__tmp99: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r978, %r747, %r977; +$L__tmp100: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r979, %r978, 4, 31, -1; +$L__tmp101: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r980, %r978, %r979; +$L__tmp102: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r981, %r980, 2, 31, -1; +$L__tmp103: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r982, %r980, %r981; +$L__tmp104: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r983, %r982, 1, 31, -1; +$L__tmp105: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r984, %r982, %r983; +$L__tmp106: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r985, %r754, 8, 31, -1; +$L__tmp107: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r986, %r754, %r985; +$L__tmp108: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r987, %r986, 4, 31, -1; +$L__tmp109: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r988, %r986, %r987; +$L__tmp110: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r989, %r988, 2, 31, -1; +$L__tmp111: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r990, %r988, %r989; +$L__tmp112: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r991, %r990, 1, 31, -1; +$L__tmp113: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r992, %r990, %r991; +$L__tmp114: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r993, %r761, 8, 31, -1; +$L__tmp115: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r994, %r761, %r993; +$L__tmp116: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r995, %r994, 4, 31, -1; +$L__tmp117: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r996, %r994, %r995; +$L__tmp118: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r997, %r996, 2, 31, -1; +$L__tmp119: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r998, %r996, %r997; +$L__tmp120: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r999, %r998, 1, 31, -1; +$L__tmp121: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1000, %r998, %r999; +$L__tmp122: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1001, %r768, 8, 31, -1; +$L__tmp123: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1002, %r768, %r1001; +$L__tmp124: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1003, %r1002, 4, 31, -1; +$L__tmp125: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1004, %r1002, %r1003; +$L__tmp126: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1005, %r1004, 2, 31, -1; +$L__tmp127: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1006, %r1004, %r1005; +$L__tmp128: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1007, %r1006, 1, 31, -1; +$L__tmp129: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1008, %r1006, %r1007; +$L__tmp130: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1009, %r775, 8, 31, -1; +$L__tmp131: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1010, %r775, %r1009; +$L__tmp132: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1011, %r1010, 4, 31, -1; +$L__tmp133: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1012, %r1010, %r1011; +$L__tmp134: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1013, %r1012, 2, 31, -1; +$L__tmp135: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1014, %r1012, %r1013; +$L__tmp136: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1015, %r1014, 1, 31, -1; +$L__tmp137: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1016, %r1014, %r1015; +$L__tmp138: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1017, %r782, 8, 31, -1; +$L__tmp139: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1018, %r782, %r1017; +$L__tmp140: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1019, %r1018, 4, 31, -1; +$L__tmp141: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1020, %r1018, %r1019; +$L__tmp142: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1021, %r1020, 2, 31, -1; +$L__tmp143: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1022, %r1020, %r1021; +$L__tmp144: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1023, %r1022, 1, 31, -1; +$L__tmp145: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1024, %r1022, %r1023; +$L__tmp146: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1025, %r789, 8, 31, -1; +$L__tmp147: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1026, %r789, %r1025; +$L__tmp148: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1027, %r1026, 4, 31, -1; +$L__tmp149: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1028, %r1026, %r1027; +$L__tmp150: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1029, %r1028, 2, 31, -1; +$L__tmp151: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1030, %r1028, %r1029; +$L__tmp152: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1031, %r1030, 1, 31, -1; +$L__tmp153: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1032, %r1030, %r1031; +$L__tmp154: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1033, %r796, 8, 31, -1; +$L__tmp155: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1034, %r796, %r1033; +$L__tmp156: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1035, %r1034, 4, 31, -1; +$L__tmp157: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1036, %r1034, %r1035; +$L__tmp158: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1037, %r1036, 2, 31, -1; +$L__tmp159: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1038, %r1036, %r1037; +$L__tmp160: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1039, %r1038, 1, 31, -1; +$L__tmp161: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1040, %r1038, %r1039; +$L__tmp162: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1041, %r803, 8, 31, -1; +$L__tmp163: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1042, %r803, %r1041; +$L__tmp164: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1043, %r1042, 4, 31, -1; +$L__tmp165: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1044, %r1042, %r1043; +$L__tmp166: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1045, %r1044, 2, 31, -1; +$L__tmp167: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1046, %r1044, %r1045; +$L__tmp168: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1047, %r1046, 1, 31, -1; +$L__tmp169: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1048, %r1046, %r1047; +$L__tmp170: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1049, %r810, 8, 31, -1; +$L__tmp171: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1050, %r810, %r1049; +$L__tmp172: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1051, %r1050, 4, 31, -1; +$L__tmp173: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1052, %r1050, %r1051; +$L__tmp174: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1053, %r1052, 2, 31, -1; +$L__tmp175: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1054, %r1052, %r1053; +$L__tmp176: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1055, %r1054, 1, 31, -1; +$L__tmp177: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1056, %r1054, %r1055; +$L__tmp178: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1057, %r817, 8, 31, -1; +$L__tmp179: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1058, %r817, %r1057; +$L__tmp180: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1059, %r1058, 4, 31, -1; +$L__tmp181: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1060, %r1058, %r1059; +$L__tmp182: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1061, %r1060, 2, 31, -1; +$L__tmp183: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1062, %r1060, %r1061; +$L__tmp184: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1063, %r1062, 1, 31, -1; +$L__tmp185: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1064, %r1062, %r1063; +$L__tmp186: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1065, %r824, 8, 31, -1; +$L__tmp187: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1066, %r824, %r1065; +$L__tmp188: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1067, %r1066, 4, 31, -1; +$L__tmp189: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1068, %r1066, %r1067; +$L__tmp190: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1069, %r1068, 2, 31, -1; +$L__tmp191: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1070, %r1068, %r1069; +$L__tmp192: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1071, %r1070, 1, 31, -1; +$L__tmp193: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1072, %r1070, %r1071; +$L__tmp194: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1073, %r831, 8, 31, -1; +$L__tmp195: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1074, %r831, %r1073; +$L__tmp196: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1075, %r1074, 4, 31, -1; +$L__tmp197: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1076, %r1074, %r1075; +$L__tmp198: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1077, %r1076, 2, 31, -1; +$L__tmp199: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1078, %r1076, %r1077; +$L__tmp200: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1079, %r1078, 1, 31, -1; +$L__tmp201: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1080, %r1078, %r1079; +$L__tmp202: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1081, %r838, 8, 31, -1; +$L__tmp203: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1082, %r838, %r1081; +$L__tmp204: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1083, %r1082, 4, 31, -1; +$L__tmp205: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1084, %r1082, %r1083; +$L__tmp206: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1085, %r1084, 2, 31, -1; +$L__tmp207: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1086, %r1084, %r1085; +$L__tmp208: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1087, %r1086, 1, 31, -1; +$L__tmp209: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1088, %r1086, %r1087; +$L__tmp210: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1089, %r845, 8, 31, -1; +$L__tmp211: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1090, %r845, %r1089; +$L__tmp212: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1091, %r1090, 4, 31, -1; +$L__tmp213: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1092, %r1090, %r1091; +$L__tmp214: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1093, %r1092, 2, 31, -1; +$L__tmp215: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1094, %r1092, %r1093; +$L__tmp216: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1095, %r1094, 1, 31, -1; +$L__tmp217: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1096, %r1094, %r1095; +$L__tmp218: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1097, %r852, 8, 31, -1; +$L__tmp219: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1098, %r852, %r1097; +$L__tmp220: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1099, %r1098, 4, 31, -1; +$L__tmp221: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1100, %r1098, %r1099; +$L__tmp222: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1101, %r1100, 2, 31, -1; +$L__tmp223: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1102, %r1100, %r1101; +$L__tmp224: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1103, %r1102, 1, 31, -1; +$L__tmp225: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1104, %r1102, %r1103; +$L__tmp226: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1105, %r859, 8, 31, -1; +$L__tmp227: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1106, %r859, %r1105; +$L__tmp228: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1107, %r1106, 4, 31, -1; +$L__tmp229: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1108, %r1106, %r1107; +$L__tmp230: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1109, %r1108, 2, 31, -1; +$L__tmp231: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1110, %r1108, %r1109; +$L__tmp232: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1111, %r1110, 1, 31, -1; +$L__tmp233: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1112, %r1110, %r1111; +$L__tmp234: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1113, %r866, 8, 31, -1; +$L__tmp235: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1114, %r866, %r1113; +$L__tmp236: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1115, %r1114, 4, 31, -1; +$L__tmp237: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1116, %r1114, %r1115; +$L__tmp238: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1117, %r1116, 2, 31, -1; +$L__tmp239: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1118, %r1116, %r1117; +$L__tmp240: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1119, %r1118, 1, 31, -1; +$L__tmp241: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1120, %r1118, %r1119; +$L__tmp242: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1121, %r873, 8, 31, -1; +$L__tmp243: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1122, %r873, %r1121; +$L__tmp244: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1123, %r1122, 4, 31, -1; +$L__tmp245: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1124, %r1122, %r1123; +$L__tmp246: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1125, %r1124, 2, 31, -1; +$L__tmp247: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1126, %r1124, %r1125; +$L__tmp248: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1127, %r1126, 1, 31, -1; +$L__tmp249: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1128, %r1126, %r1127; +$L__tmp250: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1129, %r880, 8, 31, -1; +$L__tmp251: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1130, %r880, %r1129; +$L__tmp252: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1131, %r1130, 4, 31, -1; +$L__tmp253: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1132, %r1130, %r1131; +$L__tmp254: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1133, %r1132, 2, 31, -1; +$L__tmp255: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1134, %r1132, %r1133; +$L__tmp256: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1135, %r1134, 1, 31, -1; +$L__tmp257: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1136, %r1134, %r1135; +$L__tmp258: + .loc 1 80 48 // l2norm.py:80:48 + add.f32 %r1137, %r301, %r888; + add.f32 %r1138, %r301, %r896; + add.f32 %r1139, %r301, %r904; + add.f32 %r1140, %r301, %r912; + add.f32 %r1141, %r301, %r920; + add.f32 %r1142, %r301, %r928; + add.f32 %r1143, %r301, %r936; + add.f32 %r1144, %r301, %r944; + add.f32 %r1145, %r301, %r952; + add.f32 %r1146, %r301, %r960; + add.f32 %r1147, %r301, %r968; + add.f32 %r1148, %r301, %r976; + add.f32 %r1149, %r301, %r984; + add.f32 %r1150, %r301, %r992; + add.f32 %r1151, %r301, %r1000; + add.f32 %r1152, %r301, %r1008; + add.f32 %r1153, %r301, %r1016; + add.f32 %r1154, %r301, %r1024; + add.f32 %r1155, %r301, %r1032; + add.f32 %r1156, %r301, %r1040; + add.f32 %r1157, %r301, %r1048; + add.f32 %r1158, %r301, %r1056; + add.f32 %r1159, %r301, %r1064; + add.f32 %r1160, %r301, %r1072; + add.f32 %r1161, %r301, %r1080; + add.f32 %r1162, %r301, %r1088; + add.f32 %r1163, %r301, %r1096; + add.f32 %r1164, %r301, %r1104; + add.f32 %r1165, %r301, %r1112; + add.f32 %r1166, %r301, %r1120; + add.f32 %r1167, %r301, %r1128; + add.f32 %r1168, %r301, %r1136; + .loc 1 80 25 // l2norm.py:80:25 + sqrt.approx.ftz.f32 %r1169, %r1137; + sqrt.approx.ftz.f32 %r1170, %r1138; + sqrt.approx.ftz.f32 %r1171, %r1139; + sqrt.approx.ftz.f32 %r1172, %r1140; + sqrt.approx.ftz.f32 %r1173, %r1141; + sqrt.approx.ftz.f32 %r1174, %r1142; + sqrt.approx.ftz.f32 %r1175, %r1143; + sqrt.approx.ftz.f32 %r1176, %r1144; + sqrt.approx.ftz.f32 %r1177, %r1145; + sqrt.approx.ftz.f32 %r1178, %r1146; + sqrt.approx.ftz.f32 %r1179, %r1147; + sqrt.approx.ftz.f32 %r1180, %r1148; + sqrt.approx.ftz.f32 %r1181, %r1149; + sqrt.approx.ftz.f32 %r1182, %r1150; + sqrt.approx.ftz.f32 %r1183, %r1151; + sqrt.approx.ftz.f32 %r1184, %r1152; + sqrt.approx.ftz.f32 %r1185, %r1153; + sqrt.approx.ftz.f32 %r1186, %r1154; + sqrt.approx.ftz.f32 %r1187, %r1155; + sqrt.approx.ftz.f32 %r1188, %r1156; + sqrt.approx.ftz.f32 %r1189, %r1157; + sqrt.approx.ftz.f32 %r1190, %r1158; + sqrt.approx.ftz.f32 %r1191, %r1159; + sqrt.approx.ftz.f32 %r1192, %r1160; + sqrt.approx.ftz.f32 %r1193, %r1161; + sqrt.approx.ftz.f32 %r1194, %r1162; + sqrt.approx.ftz.f32 %r1195, %r1163; + sqrt.approx.ftz.f32 %r1196, %r1164; + sqrt.approx.ftz.f32 %r1197, %r1165; + sqrt.approx.ftz.f32 %r1198, %r1166; + sqrt.approx.ftz.f32 %r1199, %r1167; + sqrt.approx.ftz.f32 %r1200, %r1168; + .loc 1 71 34 // l2norm.py:71:34 + shl.b32 %r1201, %r302, 1; + and.b32 %r1202, %r1201, 62; + or.b32 %r1203, %r1202, 1; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd201, %r1203; + or.b64 %rd202, %rd70, %rd201; + cvt.u64.u32 %rd203, %r1202; + or.b64 %rd204, %rd70, %rd203; + .loc 1 77 20 // l2norm.py:77:20 + shl.b64 %rd205, %rd204, 2; + add.s64 %rd65, %rd69, %rd205; + add.s64 %rd66, %rd65, 4; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd206, %rd68, %rd198; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd64, %rd206, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd207, %rd68, %rd196; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd63, %rd207, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd208, %rd68, %rd194; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd62, %rd208, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd209, %rd68, %rd192; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd61, %rd209, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd210, %rd68, %rd190; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd60, %rd210, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd211, %rd68, %rd188; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd59, %rd211, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd212, %rd68, %rd186; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd58, %rd212, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd213, %rd68, %rd184; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd57, %rd213, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd214, %rd68, %rd182; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd56, %rd214, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd215, %rd68, %rd180; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd55, %rd215, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd216, %rd68, %rd178; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd54, %rd216, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd217, %rd68, %rd176; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd53, %rd217, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd218, %rd68, %rd174; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd52, %rd218, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd219, %rd68, %rd172; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd51, %rd219, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd220, %rd68, %rd170; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd50, %rd220, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd221, %rd68, %rd168; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd49, %rd221, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd222, %rd68, %rd166; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd48, %rd222, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd223, %rd68, %rd164; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd47, %rd223, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd224, %rd68, %rd162; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd46, %rd224, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd225, %rd68, %rd160; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd45, %rd225, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd226, %rd68, %rd158; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd44, %rd226, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd227, %rd68, %rd156; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd43, %rd227, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd228, %rd68, %rd154; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd42, %rd228, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd229, %rd68, %rd152; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd41, %rd229, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd230, %rd68, %rd150; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd40, %rd230, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd231, %rd68, %rd148; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd39, %rd231, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd232, %rd68, %rd146; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd38, %rd232, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd233, %rd68, %rd144; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd37, %rd233, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd234, %rd68, %rd142; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd36, %rd234, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd235, %rd68, %rd140; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd35, %rd235, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd236, %rd68, %rd138; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd34, %rd236, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd237, %rd68, %rd136; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd33, %rd237, %rd200; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p34, %rd202, %rd71; + setp.lt.s64 %p33, %rd204, %rd71; + mov.b32 %r1204, 0f3F800000; + .loc 1 80 17 // l2norm.py:80:17 + div.full.f32 %r259, %r1204, %r1169; + div.full.f32 %r260, %r1204, %r1170; + div.full.f32 %r261, %r1204, %r1171; + div.full.f32 %r262, %r1204, %r1172; + div.full.f32 %r264, %r1204, %r1173; + div.full.f32 %r265, %r1204, %r1174; + div.full.f32 %r266, %r1204, %r1175; + div.full.f32 %r267, %r1204, %r1176; + div.full.f32 %r269, %r1204, %r1177; + div.full.f32 %r270, %r1204, %r1178; + div.full.f32 %r271, %r1204, %r1179; + div.full.f32 %r272, %r1204, %r1180; + div.full.f32 %r274, %r1204, %r1181; + div.full.f32 %r275, %r1204, %r1182; + div.full.f32 %r276, %r1204, %r1183; + div.full.f32 %r277, %r1204, %r1184; + div.full.f32 %r279, %r1204, %r1185; + div.full.f32 %r280, %r1204, %r1186; + div.full.f32 %r281, %r1204, %r1187; + div.full.f32 %r282, %r1204, %r1188; + div.full.f32 %r284, %r1204, %r1189; + div.full.f32 %r285, %r1204, %r1190; + div.full.f32 %r286, %r1204, %r1191; + div.full.f32 %r287, %r1204, %r1192; + div.full.f32 %r289, %r1204, %r1193; + div.full.f32 %r290, %r1204, %r1194; + div.full.f32 %r291, %r1204, %r1195; + div.full.f32 %r292, %r1204, %r1196; + div.full.f32 %r294, %r1204, %r1197; + div.full.f32 %r295, %r1204, %r1198; + div.full.f32 %r296, %r1204, %r1199; + div.full.f32 %r297, %r1204, %r1200; + .loc 1 81 16 // l2norm.py:81:16 + mul.f32 %r1205, %r259, %r369; + mul.f32 %r1206, %r259, %r370; + mul.f32 %r1207, %r259, %r371; + mul.f32 %r1208, %r259, %r372; + mul.f32 %r1209, %r259, %r373; + mul.f32 %r1210, %r259, %r374; + mul.f32 %r1211, %r259, %r375; + mul.f32 %r1212, %r259, %r376; + mul.f32 %r1213, %r260, %r377; + mul.f32 %r1214, %r260, %r378; + mul.f32 %r1215, %r260, %r379; + mul.f32 %r1216, %r260, %r380; + mul.f32 %r1217, %r260, %r381; + mul.f32 %r1218, %r260, %r382; + mul.f32 %r1219, %r260, %r383; + mul.f32 %r1220, %r260, %r384; + mul.f32 %r1221, %r261, %r385; + mul.f32 %r1222, %r261, %r386; + mul.f32 %r1223, %r261, %r387; + mul.f32 %r1224, %r261, %r388; + mul.f32 %r1225, %r261, %r389; + mul.f32 %r1226, %r261, %r390; + mul.f32 %r1227, %r261, %r391; + mul.f32 %r1228, %r261, %r392; + mul.f32 %r1229, %r262, %r393; + mul.f32 %r1230, %r262, %r394; + mul.f32 %r1231, %r262, %r395; + mul.f32 %r1232, %r262, %r396; + mul.f32 %r1233, %r262, %r397; + mul.f32 %r1234, %r262, %r398; + mul.f32 %r1235, %r262, %r399; + mul.f32 %r1236, %r262, %r400; + mul.f32 %r1237, %r264, %r401; + mul.f32 %r1238, %r264, %r402; + mul.f32 %r1239, %r264, %r403; + mul.f32 %r1240, %r264, %r404; + mul.f32 %r1241, %r264, %r405; + mul.f32 %r1242, %r264, %r406; + mul.f32 %r1243, %r264, %r407; + mul.f32 %r1244, %r264, %r408; + mul.f32 %r1245, %r265, %r409; + mul.f32 %r1246, %r265, %r410; + mul.f32 %r1247, %r265, %r411; + mul.f32 %r1248, %r265, %r412; + mul.f32 %r1249, %r265, %r413; + mul.f32 %r1250, %r265, %r414; + mul.f32 %r1251, %r265, %r415; + mul.f32 %r1252, %r265, %r416; + mul.f32 %r1253, %r266, %r417; + mul.f32 %r1254, %r266, %r418; + mul.f32 %r1255, %r266, %r419; + mul.f32 %r1256, %r266, %r420; + mul.f32 %r1257, %r266, %r421; + mul.f32 %r1258, %r266, %r422; + mul.f32 %r1259, %r266, %r423; + mul.f32 %r1260, %r266, %r424; + mul.f32 %r1261, %r267, %r425; + mul.f32 %r1262, %r267, %r426; + mul.f32 %r1263, %r267, %r427; + mul.f32 %r1264, %r267, %r428; + mul.f32 %r1265, %r267, %r429; + mul.f32 %r1266, %r267, %r430; + mul.f32 %r1267, %r267, %r431; + mul.f32 %r1268, %r267, %r432; + mul.f32 %r1269, %r269, %r433; + mul.f32 %r1270, %r269, %r434; + mul.f32 %r1271, %r269, %r435; + mul.f32 %r1272, %r269, %r436; + mul.f32 %r1273, %r269, %r437; + mul.f32 %r1274, %r269, %r438; + mul.f32 %r1275, %r269, %r439; + mul.f32 %r1276, %r269, %r440; + mul.f32 %r1277, %r270, %r441; + mul.f32 %r1278, %r270, %r442; + mul.f32 %r1279, %r270, %r443; + mul.f32 %r1280, %r270, %r444; + mul.f32 %r1281, %r270, %r445; + mul.f32 %r1282, %r270, %r446; + mul.f32 %r1283, %r270, %r447; + mul.f32 %r1284, %r270, %r448; + mul.f32 %r1285, %r271, %r449; + mul.f32 %r1286, %r271, %r450; + mul.f32 %r1287, %r271, %r451; + mul.f32 %r1288, %r271, %r452; + mul.f32 %r1289, %r271, %r453; + mul.f32 %r1290, %r271, %r454; + mul.f32 %r1291, %r271, %r455; + mul.f32 %r1292, %r271, %r456; + mul.f32 %r1293, %r272, %r457; + mul.f32 %r1294, %r272, %r458; + mul.f32 %r1295, %r272, %r459; + mul.f32 %r1296, %r272, %r460; + mul.f32 %r1297, %r272, %r461; + mul.f32 %r1298, %r272, %r462; + mul.f32 %r1299, %r272, %r463; + mul.f32 %r1300, %r272, %r464; + mul.f32 %r1301, %r274, %r465; + mul.f32 %r1302, %r274, %r466; + mul.f32 %r1303, %r274, %r467; + mul.f32 %r1304, %r274, %r468; + mul.f32 %r1305, %r274, %r469; + mul.f32 %r1306, %r274, %r470; + mul.f32 %r1307, %r274, %r471; + mul.f32 %r1308, %r274, %r472; + mul.f32 %r1309, %r275, %r473; + mul.f32 %r1310, %r275, %r474; + mul.f32 %r1311, %r275, %r475; + mul.f32 %r1312, %r275, %r476; + mul.f32 %r1313, %r275, %r477; + mul.f32 %r1314, %r275, %r478; + mul.f32 %r1315, %r275, %r479; + mul.f32 %r1316, %r275, %r480; + mul.f32 %r1317, %r276, %r481; + mul.f32 %r1318, %r276, %r482; + mul.f32 %r1319, %r276, %r483; + mul.f32 %r1320, %r276, %r484; + mul.f32 %r1321, %r276, %r485; + mul.f32 %r1322, %r276, %r486; + mul.f32 %r1323, %r276, %r487; + mul.f32 %r1324, %r276, %r488; + mul.f32 %r1325, %r277, %r489; + mul.f32 %r1326, %r277, %r490; + mul.f32 %r1327, %r277, %r491; + mul.f32 %r1328, %r277, %r492; + mul.f32 %r1329, %r277, %r493; + mul.f32 %r1330, %r277, %r494; + mul.f32 %r1331, %r277, %r495; + mul.f32 %r1332, %r277, %r496; + mul.f32 %r1333, %r279, %r497; + mul.f32 %r1334, %r279, %r498; + mul.f32 %r1335, %r279, %r499; + mul.f32 %r1336, %r279, %r500; + mul.f32 %r1337, %r279, %r501; + mul.f32 %r1338, %r279, %r502; + mul.f32 %r1339, %r279, %r503; + mul.f32 %r1340, %r279, %r504; + mul.f32 %r1341, %r280, %r505; + mul.f32 %r1342, %r280, %r506; + mul.f32 %r1343, %r280, %r507; + mul.f32 %r1344, %r280, %r508; + mul.f32 %r1345, %r280, %r509; + mul.f32 %r1346, %r280, %r510; + mul.f32 %r1347, %r280, %r511; + mul.f32 %r1348, %r280, %r512; + mul.f32 %r1349, %r281, %r513; + mul.f32 %r1350, %r281, %r514; + mul.f32 %r1351, %r281, %r515; + mul.f32 %r1352, %r281, %r516; + mul.f32 %r1353, %r281, %r517; + mul.f32 %r1354, %r281, %r518; + mul.f32 %r1355, %r281, %r519; + mul.f32 %r1356, %r281, %r520; + mul.f32 %r1357, %r282, %r521; + mul.f32 %r1358, %r282, %r522; + mul.f32 %r1359, %r282, %r523; + mul.f32 %r1360, %r282, %r524; + mul.f32 %r1361, %r282, %r525; + mul.f32 %r1362, %r282, %r526; + mul.f32 %r1363, %r282, %r527; + mul.f32 %r1364, %r282, %r528; + mul.f32 %r1365, %r284, %r529; + mul.f32 %r1366, %r284, %r530; + mul.f32 %r1367, %r284, %r531; + mul.f32 %r1368, %r284, %r532; + mul.f32 %r1369, %r284, %r533; + mul.f32 %r1370, %r284, %r534; + mul.f32 %r1371, %r284, %r535; + mul.f32 %r1372, %r284, %r536; + mul.f32 %r1373, %r285, %r537; + mul.f32 %r1374, %r285, %r538; + mul.f32 %r1375, %r285, %r539; + mul.f32 %r1376, %r285, %r540; + mul.f32 %r1377, %r285, %r541; + mul.f32 %r1378, %r285, %r542; + mul.f32 %r1379, %r285, %r543; + mul.f32 %r1380, %r285, %r544; + mul.f32 %r1381, %r286, %r545; + mul.f32 %r1382, %r286, %r546; + mul.f32 %r1383, %r286, %r547; + mul.f32 %r1384, %r286, %r548; + mul.f32 %r1385, %r286, %r549; + mul.f32 %r1386, %r286, %r550; + mul.f32 %r1387, %r286, %r551; + mul.f32 %r1388, %r286, %r552; + mul.f32 %r1389, %r287, %r553; + mul.f32 %r1390, %r287, %r554; + mul.f32 %r1391, %r287, %r555; + mul.f32 %r1392, %r287, %r556; + mul.f32 %r1393, %r287, %r557; + mul.f32 %r1394, %r287, %r558; + mul.f32 %r1395, %r287, %r559; + mul.f32 %r1396, %r287, %r560; + mul.f32 %r1397, %r289, %r561; + mul.f32 %r1398, %r289, %r562; + mul.f32 %r1399, %r289, %r563; + mul.f32 %r1400, %r289, %r564; + mul.f32 %r1401, %r289, %r565; + mul.f32 %r1402, %r289, %r566; + mul.f32 %r1403, %r289, %r567; + mul.f32 %r1404, %r289, %r568; + mul.f32 %r1405, %r290, %r569; + mul.f32 %r1406, %r290, %r570; + mul.f32 %r1407, %r290, %r571; + mul.f32 %r1408, %r290, %r572; + mul.f32 %r1409, %r290, %r573; + mul.f32 %r1410, %r290, %r574; + mul.f32 %r1411, %r290, %r575; + mul.f32 %r1412, %r290, %r576; + mul.f32 %r1413, %r291, %r577; + mul.f32 %r1414, %r291, %r578; + mul.f32 %r1415, %r291, %r579; + mul.f32 %r1416, %r291, %r580; + mul.f32 %r1417, %r291, %r581; + mul.f32 %r1418, %r291, %r582; + mul.f32 %r1419, %r291, %r583; + mul.f32 %r1420, %r291, %r584; + mul.f32 %r1421, %r292, %r585; + mul.f32 %r1422, %r292, %r586; + mul.f32 %r1423, %r292, %r587; + mul.f32 %r1424, %r292, %r588; + mul.f32 %r1425, %r292, %r589; + mul.f32 %r1426, %r292, %r590; + mul.f32 %r1427, %r292, %r591; + mul.f32 %r1428, %r292, %r592; + mul.f32 %r1429, %r294, %r593; + mul.f32 %r1430, %r294, %r594; + mul.f32 %r1431, %r294, %r595; + mul.f32 %r1432, %r294, %r596; + mul.f32 %r1433, %r294, %r597; + mul.f32 %r1434, %r294, %r598; + mul.f32 %r1435, %r294, %r599; + mul.f32 %r1436, %r294, %r600; + mul.f32 %r1437, %r295, %r601; + mul.f32 %r1438, %r295, %r602; + mul.f32 %r1439, %r295, %r603; + mul.f32 %r1440, %r295, %r604; + mul.f32 %r1441, %r295, %r605; + mul.f32 %r1442, %r295, %r606; + mul.f32 %r1443, %r295, %r607; + mul.f32 %r1444, %r295, %r608; + mul.f32 %r1445, %r296, %r609; + mul.f32 %r1446, %r296, %r610; + mul.f32 %r1447, %r296, %r611; + mul.f32 %r1448, %r296, %r612; + mul.f32 %r1449, %r296, %r613; + mul.f32 %r1450, %r296, %r614; + mul.f32 %r1451, %r296, %r615; + mul.f32 %r1452, %r296, %r616; + mul.f32 %r1453, %r297, %r617; + mul.f32 %r1454, %r297, %r618; + mul.f32 %r1455, %r297, %r619; + mul.f32 %r1456, %r297, %r620; + mul.f32 %r1457, %r297, %r621; + mul.f32 %r1458, %r297, %r622; + mul.f32 %r1459, %r297, %r623; + mul.f32 %r1460, %r297, %r624; + .loc 1 83 18 // l2norm.py:83:18 + cvt.rn.bf16x2.f32 %r130, %r1206, %r1205; + cvt.rn.bf16x2.f32 %r131, %r1208, %r1207; + cvt.rn.bf16x2.f32 %r132, %r1210, %r1209; + cvt.rn.bf16x2.f32 %r133, %r1212, %r1211; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd33 + 0 ], { %r130, %r131, %r132, %r133 }; + // end inline asm + cvt.rn.bf16x2.f32 %r134, %r1214, %r1213; + cvt.rn.bf16x2.f32 %r135, %r1216, %r1215; + cvt.rn.bf16x2.f32 %r136, %r1218, %r1217; + cvt.rn.bf16x2.f32 %r137, %r1220, %r1219; + // begin inline asm + @%p3 st.global.v4.b32 [ %rd34 + 0 ], { %r134, %r135, %r136, %r137 }; + // end inline asm + cvt.rn.bf16x2.f32 %r138, %r1222, %r1221; + cvt.rn.bf16x2.f32 %r139, %r1224, %r1223; + cvt.rn.bf16x2.f32 %r140, %r1226, %r1225; + cvt.rn.bf16x2.f32 %r141, %r1228, %r1227; + // begin inline asm + @%p4 st.global.v4.b32 [ %rd35 + 0 ], { %r138, %r139, %r140, %r141 }; + // end inline asm + cvt.rn.bf16x2.f32 %r142, %r1230, %r1229; + cvt.rn.bf16x2.f32 %r143, %r1232, %r1231; + cvt.rn.bf16x2.f32 %r144, %r1234, %r1233; + cvt.rn.bf16x2.f32 %r145, %r1236, %r1235; + // begin inline asm + @%p5 st.global.v4.b32 [ %rd36 + 0 ], { %r142, %r143, %r144, %r145 }; + // end inline asm + cvt.rn.bf16x2.f32 %r146, %r1238, %r1237; + cvt.rn.bf16x2.f32 %r147, %r1240, %r1239; + cvt.rn.bf16x2.f32 %r148, %r1242, %r1241; + cvt.rn.bf16x2.f32 %r149, %r1244, %r1243; + // begin inline asm + @%p6 st.global.v4.b32 [ %rd37 + 0 ], { %r146, %r147, %r148, %r149 }; + // end inline asm + cvt.rn.bf16x2.f32 %r150, %r1246, %r1245; + cvt.rn.bf16x2.f32 %r151, %r1248, %r1247; + cvt.rn.bf16x2.f32 %r152, %r1250, %r1249; + cvt.rn.bf16x2.f32 %r153, %r1252, %r1251; + // begin inline asm + @%p7 st.global.v4.b32 [ %rd38 + 0 ], { %r150, %r151, %r152, %r153 }; + // end inline asm + cvt.rn.bf16x2.f32 %r154, %r1254, %r1253; + cvt.rn.bf16x2.f32 %r155, %r1256, %r1255; + cvt.rn.bf16x2.f32 %r156, %r1258, %r1257; + cvt.rn.bf16x2.f32 %r157, %r1260, %r1259; + // begin inline asm + @%p8 st.global.v4.b32 [ %rd39 + 0 ], { %r154, %r155, %r156, %r157 }; + // end inline asm + cvt.rn.bf16x2.f32 %r158, %r1262, %r1261; + cvt.rn.bf16x2.f32 %r159, %r1264, %r1263; + cvt.rn.bf16x2.f32 %r160, %r1266, %r1265; + cvt.rn.bf16x2.f32 %r161, %r1268, %r1267; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd40 + 0 ], { %r158, %r159, %r160, %r161 }; + // end inline asm + cvt.rn.bf16x2.f32 %r162, %r1270, %r1269; + cvt.rn.bf16x2.f32 %r163, %r1272, %r1271; + cvt.rn.bf16x2.f32 %r164, %r1274, %r1273; + cvt.rn.bf16x2.f32 %r165, %r1276, %r1275; + // begin inline asm + @%p9 st.global.v4.b32 [ %rd41 + 0 ], { %r162, %r163, %r164, %r165 }; + // end inline asm + cvt.rn.bf16x2.f32 %r166, %r1278, %r1277; + cvt.rn.bf16x2.f32 %r167, %r1280, %r1279; + cvt.rn.bf16x2.f32 %r168, %r1282, %r1281; + cvt.rn.bf16x2.f32 %r169, %r1284, %r1283; + // begin inline asm + @%p10 st.global.v4.b32 [ %rd42 + 0 ], { %r166, %r167, %r168, %r169 }; + // end inline asm + cvt.rn.bf16x2.f32 %r170, %r1286, %r1285; + cvt.rn.bf16x2.f32 %r171, %r1288, %r1287; + cvt.rn.bf16x2.f32 %r172, %r1290, %r1289; + cvt.rn.bf16x2.f32 %r173, %r1292, %r1291; + // begin inline asm + @%p11 st.global.v4.b32 [ %rd43 + 0 ], { %r170, %r171, %r172, %r173 }; + // end inline asm + cvt.rn.bf16x2.f32 %r174, %r1294, %r1293; + cvt.rn.bf16x2.f32 %r175, %r1296, %r1295; + cvt.rn.bf16x2.f32 %r176, %r1298, %r1297; + cvt.rn.bf16x2.f32 %r177, %r1300, %r1299; + // begin inline asm + @%p12 st.global.v4.b32 [ %rd44 + 0 ], { %r174, %r175, %r176, %r177 }; + // end inline asm + cvt.rn.bf16x2.f32 %r178, %r1302, %r1301; + cvt.rn.bf16x2.f32 %r179, %r1304, %r1303; + cvt.rn.bf16x2.f32 %r180, %r1306, %r1305; + cvt.rn.bf16x2.f32 %r181, %r1308, %r1307; + // begin inline asm + @%p13 st.global.v4.b32 [ %rd45 + 0 ], { %r178, %r179, %r180, %r181 }; + // end inline asm + cvt.rn.bf16x2.f32 %r182, %r1310, %r1309; + cvt.rn.bf16x2.f32 %r183, %r1312, %r1311; + cvt.rn.bf16x2.f32 %r184, %r1314, %r1313; + cvt.rn.bf16x2.f32 %r185, %r1316, %r1315; + // begin inline asm + @%p14 st.global.v4.b32 [ %rd46 + 0 ], { %r182, %r183, %r184, %r185 }; + // end inline asm + cvt.rn.bf16x2.f32 %r186, %r1318, %r1317; + cvt.rn.bf16x2.f32 %r187, %r1320, %r1319; + cvt.rn.bf16x2.f32 %r188, %r1322, %r1321; + cvt.rn.bf16x2.f32 %r189, %r1324, %r1323; + // begin inline asm + @%p15 st.global.v4.b32 [ %rd47 + 0 ], { %r186, %r187, %r188, %r189 }; + // end inline asm + cvt.rn.bf16x2.f32 %r190, %r1326, %r1325; + cvt.rn.bf16x2.f32 %r191, %r1328, %r1327; + cvt.rn.bf16x2.f32 %r192, %r1330, %r1329; + cvt.rn.bf16x2.f32 %r193, %r1332, %r1331; + // begin inline asm + @%p16 st.global.v4.b32 [ %rd48 + 0 ], { %r190, %r191, %r192, %r193 }; + // end inline asm + cvt.rn.bf16x2.f32 %r194, %r1334, %r1333; + cvt.rn.bf16x2.f32 %r195, %r1336, %r1335; + cvt.rn.bf16x2.f32 %r196, %r1338, %r1337; + cvt.rn.bf16x2.f32 %r197, %r1340, %r1339; + // begin inline asm + @%p17 st.global.v4.b32 [ %rd49 + 0 ], { %r194, %r195, %r196, %r197 }; + // end inline asm + cvt.rn.bf16x2.f32 %r198, %r1342, %r1341; + cvt.rn.bf16x2.f32 %r199, %r1344, %r1343; + cvt.rn.bf16x2.f32 %r200, %r1346, %r1345; + cvt.rn.bf16x2.f32 %r201, %r1348, %r1347; + // begin inline asm + @%p18 st.global.v4.b32 [ %rd50 + 0 ], { %r198, %r199, %r200, %r201 }; + // end inline asm + cvt.rn.bf16x2.f32 %r202, %r1350, %r1349; + cvt.rn.bf16x2.f32 %r203, %r1352, %r1351; + cvt.rn.bf16x2.f32 %r204, %r1354, %r1353; + cvt.rn.bf16x2.f32 %r205, %r1356, %r1355; + // begin inline asm + @%p19 st.global.v4.b32 [ %rd51 + 0 ], { %r202, %r203, %r204, %r205 }; + // end inline asm + cvt.rn.bf16x2.f32 %r206, %r1358, %r1357; + cvt.rn.bf16x2.f32 %r207, %r1360, %r1359; + cvt.rn.bf16x2.f32 %r208, %r1362, %r1361; + cvt.rn.bf16x2.f32 %r209, %r1364, %r1363; + // begin inline asm + @%p20 st.global.v4.b32 [ %rd52 + 0 ], { %r206, %r207, %r208, %r209 }; + // end inline asm + cvt.rn.bf16x2.f32 %r210, %r1366, %r1365; + cvt.rn.bf16x2.f32 %r211, %r1368, %r1367; + cvt.rn.bf16x2.f32 %r212, %r1370, %r1369; + cvt.rn.bf16x2.f32 %r213, %r1372, %r1371; + // begin inline asm + @%p21 st.global.v4.b32 [ %rd53 + 0 ], { %r210, %r211, %r212, %r213 }; + // end inline asm + cvt.rn.bf16x2.f32 %r214, %r1374, %r1373; + cvt.rn.bf16x2.f32 %r215, %r1376, %r1375; + cvt.rn.bf16x2.f32 %r216, %r1378, %r1377; + cvt.rn.bf16x2.f32 %r217, %r1380, %r1379; + // begin inline asm + @%p22 st.global.v4.b32 [ %rd54 + 0 ], { %r214, %r215, %r216, %r217 }; + // end inline asm + cvt.rn.bf16x2.f32 %r218, %r1382, %r1381; + cvt.rn.bf16x2.f32 %r219, %r1384, %r1383; + cvt.rn.bf16x2.f32 %r220, %r1386, %r1385; + cvt.rn.bf16x2.f32 %r221, %r1388, %r1387; + // begin inline asm + @%p23 st.global.v4.b32 [ %rd55 + 0 ], { %r218, %r219, %r220, %r221 }; + // end inline asm + cvt.rn.bf16x2.f32 %r222, %r1390, %r1389; + cvt.rn.bf16x2.f32 %r223, %r1392, %r1391; + cvt.rn.bf16x2.f32 %r224, %r1394, %r1393; + cvt.rn.bf16x2.f32 %r225, %r1396, %r1395; + // begin inline asm + @%p24 st.global.v4.b32 [ %rd56 + 0 ], { %r222, %r223, %r224, %r225 }; + // end inline asm + cvt.rn.bf16x2.f32 %r226, %r1398, %r1397; + cvt.rn.bf16x2.f32 %r227, %r1400, %r1399; + cvt.rn.bf16x2.f32 %r228, %r1402, %r1401; + cvt.rn.bf16x2.f32 %r229, %r1404, %r1403; + // begin inline asm + @%p25 st.global.v4.b32 [ %rd57 + 0 ], { %r226, %r227, %r228, %r229 }; + // end inline asm + cvt.rn.bf16x2.f32 %r230, %r1406, %r1405; + cvt.rn.bf16x2.f32 %r231, %r1408, %r1407; + cvt.rn.bf16x2.f32 %r232, %r1410, %r1409; + cvt.rn.bf16x2.f32 %r233, %r1412, %r1411; + // begin inline asm + @%p26 st.global.v4.b32 [ %rd58 + 0 ], { %r230, %r231, %r232, %r233 }; + // end inline asm + cvt.rn.bf16x2.f32 %r234, %r1414, %r1413; + cvt.rn.bf16x2.f32 %r235, %r1416, %r1415; + cvt.rn.bf16x2.f32 %r236, %r1418, %r1417; + cvt.rn.bf16x2.f32 %r237, %r1420, %r1419; + // begin inline asm + @%p27 st.global.v4.b32 [ %rd59 + 0 ], { %r234, %r235, %r236, %r237 }; + // end inline asm + cvt.rn.bf16x2.f32 %r238, %r1422, %r1421; + cvt.rn.bf16x2.f32 %r239, %r1424, %r1423; + cvt.rn.bf16x2.f32 %r240, %r1426, %r1425; + cvt.rn.bf16x2.f32 %r241, %r1428, %r1427; + // begin inline asm + @%p28 st.global.v4.b32 [ %rd60 + 0 ], { %r238, %r239, %r240, %r241 }; + // end inline asm + cvt.rn.bf16x2.f32 %r242, %r1430, %r1429; + cvt.rn.bf16x2.f32 %r243, %r1432, %r1431; + cvt.rn.bf16x2.f32 %r244, %r1434, %r1433; + cvt.rn.bf16x2.f32 %r245, %r1436, %r1435; + // begin inline asm + @%p29 st.global.v4.b32 [ %rd61 + 0 ], { %r242, %r243, %r244, %r245 }; + // end inline asm + cvt.rn.bf16x2.f32 %r246, %r1438, %r1437; + cvt.rn.bf16x2.f32 %r247, %r1440, %r1439; + cvt.rn.bf16x2.f32 %r248, %r1442, %r1441; + cvt.rn.bf16x2.f32 %r249, %r1444, %r1443; + // begin inline asm + @%p30 st.global.v4.b32 [ %rd62 + 0 ], { %r246, %r247, %r248, %r249 }; + // end inline asm + cvt.rn.bf16x2.f32 %r250, %r1446, %r1445; + cvt.rn.bf16x2.f32 %r251, %r1448, %r1447; + cvt.rn.bf16x2.f32 %r252, %r1450, %r1449; + cvt.rn.bf16x2.f32 %r253, %r1452, %r1451; + // begin inline asm + @%p31 st.global.v4.b32 [ %rd63 + 0 ], { %r250, %r251, %r252, %r253 }; + // end inline asm + cvt.rn.bf16x2.f32 %r254, %r1454, %r1453; + cvt.rn.bf16x2.f32 %r255, %r1456, %r1455; + cvt.rn.bf16x2.f32 %r256, %r1458, %r1457; + cvt.rn.bf16x2.f32 %r257, %r1460, %r1459; + // begin inline asm + @%p32 st.global.v4.b32 [ %rd64 + 0 ], { %r254, %r255, %r256, %r257 }; + // end inline asm + .loc 1 84 21 // l2norm.py:84:21 + shl.b32 %r1461, %r303, 3; + mov.b32 %r1462, global_smem; + add.s32 %r258, %r1462, %r1461; + // begin inline asm + st.shared.v4.b32 [ %r258 + 0 ], { %r259, %r260, %r261, %r262 }; + // end inline asm + add.s32 %r263, %r258, 16; + // begin inline asm + st.shared.v4.b32 [ %r263 + 0 ], { %r264, %r265, %r266, %r267 }; + // end inline asm + add.s32 %r268, %r258, 32; + // begin inline asm + st.shared.v4.b32 [ %r268 + 0 ], { %r269, %r270, %r271, %r272 }; + // end inline asm + add.s32 %r273, %r258, 48; + // begin inline asm + st.shared.v4.b32 [ %r273 + 0 ], { %r274, %r275, %r276, %r277 }; + // end inline asm + add.s32 %r278, %r258, 64; + // begin inline asm + st.shared.v4.b32 [ %r278 + 0 ], { %r279, %r280, %r281, %r282 }; + // end inline asm + add.s32 %r283, %r258, 80; + // begin inline asm + st.shared.v4.b32 [ %r283 + 0 ], { %r284, %r285, %r286, %r287 }; + // end inline asm + add.s32 %r288, %r258, 96; + // begin inline asm + st.shared.v4.b32 [ %r288 + 0 ], { %r289, %r290, %r291, %r292 }; + // end inline asm + add.s32 %r293, %r258, 112; + // begin inline asm + st.shared.v4.b32 [ %r293 + 0 ], { %r294, %r295, %r296, %r297 }; + // end inline asm + bar.warp.sync -1; + shl.b32 %r1463, %r337, 4; + add.s32 %r1464, %r1462, %r1463; + ldmatrix.sync.aligned.m8n8.x2.shared.b16 {%r298, %r299}, [%r1464]; + // begin inline asm + @%p33 st.global.b32 [ %rd65 + 0 ], { %r298 }; + // end inline asm + // begin inline asm + @%p34 st.global.b32 [ %rd66 + 0 ], { %r299 }; + // end inline asm + .loc 1 84 4 // l2norm.py:84:4 + ret; +$L__tmp259: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp258 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 80 // DW_AT_call_line +.b8 43 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp258 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.source b/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..b5000c470f5e0683aea9e8045b2d0a8c059ea3d1 --- /dev/null +++ b/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.source @@ -0,0 +1,172 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc38 = loc(unknown) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc45 = loc("x"(#loc)) +#loc46 = loc("y"(#loc)) +#loc47 = loc("rstd"(#loc)) +#loc48 = loc("eps"(#loc)) +#loc49 = loc("T"(#loc)) +#loc81 = loc("input"(#loc36)) +#loc82 = loc("a"(#loc41)) +#loc83 = loc("b"(#loc41)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc50) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc51) + %o_t = arith.constant 64 : i32 loc(#loc52) + %o_t_1 = arith.constant 64 : i64 loc(#loc52) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc52) + %o_t_3 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc53) + %o_t_4 = arith.extsi %o_t_3 : tensor<64xi32> to tensor<64xi64> loc(#loc54) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<64xi64> loc(#loc54) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<64xi64> loc(#loc54) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc55) + %m_t = arith.extsi %T : i32 to i64 loc(#loc56) + %m_t_7 = tt.splat %m_t : i64 -> tensor<64xi64> loc(#loc56) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<64xi64> loc(#loc56) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc57) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc58) + %m_x_10 = arith.constant 128 : i32 loc(#loc59) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc59) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc59) + %m_x_13 = tt.broadcast %m_x : tensor<64x1xi1> -> tensor<64x128xi1> loc(#loc60) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<64x128xi1> loc(#loc60) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<64x128xi1> loc(#loc60) + %p_x = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc61) + %p_x_16 = arith.constant 128 : i32 loc(#loc62) + %p_x_17 = arith.constant 128 : i64 loc(#loc62) + %p_x_18 = arith.constant dense<128> : tensor<64x1xi64> loc(#loc62) + %p_x_19 = arith.muli %p_x, %p_x_18 : tensor<64x1xi64> loc(#loc62) + %p_x_20 = tt.splat %x : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc63) + %p_x_21 = tt.addptr %p_x_20, %p_x_19 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc63) + %p_x_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc64) + %p_x_23 = tt.broadcast %p_x_21 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc65) + %p_x_24 = tt.broadcast %p_x_22 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc65) + %p_x_25 = tt.addptr %p_x_23, %p_x_24 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc65) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc66) + %p_y_26 = arith.constant 128 : i32 loc(#loc67) + %p_y_27 = arith.constant 128 : i64 loc(#loc67) + %p_y_28 = arith.constant dense<128> : tensor<64x1xi64> loc(#loc67) + %p_y_29 = arith.muli %p_y, %p_y_28 : tensor<64x1xi64> loc(#loc67) + %p_y_30 = tt.splat %y : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc68) + %p_y_31 = tt.addptr %p_y_30, %p_y_29 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc68) + %p_y_32 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc69) + %p_y_33 = tt.broadcast %p_y_31 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc70) + %p_y_34 = tt.broadcast %p_y_32 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc70) + %p_y_35 = tt.addptr %p_y_33, %p_y_34 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc70) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<64x!tt.ptr> loc(#loc71) + %p_rstd_36 = tt.addptr %p_rstd, %o_t_6 : tensor<64x!tt.ptr>, tensor<64xi64> loc(#loc71) + %b_x = arith.constant 0.000000e+00 : f32 loc(#loc72) + %b_x_37 = arith.constant dense<0.000000e+00> : tensor<64x128xf32> loc(#loc72) + %b_x_38 = arith.truncf %b_x_37 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc72) + %b_x_39 = tt.load %p_x_25, %m_x_15, %b_x_38 : tensor<64x128x!tt.ptr> loc(#loc72) + %b_x_40 = arith.extf %b_x_39 : tensor<64x128xbf16> to tensor<64x128xf32> loc(#loc73) + %b_rstd = arith.mulf %b_x_40, %b_x_40 : tensor<64x128xf32> loc(#loc74) + %b_rstd_41 = tt.call @triton.language.standard.sum__fp32S64_128S_c1_cFalse_cNone(%b_rstd) : (tensor<64x128xf32>) -> tensor<64xf32> loc(#loc75) + %b_rstd_42 = tt.splat %eps : f32 -> tensor<64xf32> loc(#loc76) + %b_rstd_43 = arith.addf %b_rstd_41, %b_rstd_42 : tensor<64xf32> loc(#loc76) + %b_rstd_44 = math.sqrt %b_rstd_43 : tensor<64xf32> loc(#loc77) + %b_rstd_45 = arith.constant 1 : i32 loc(#loc78) + %b_rstd_46 = arith.constant 1.000000e+00 : f32 loc(#loc78) + %b_rstd_47 = arith.constant dense<1.000000e+00> : tensor<64xf32> loc(#loc78) + %b_rstd_48 = arith.divf %b_rstd_47, %b_rstd_44 : tensor<64xf32> loc(#loc78) + %b_y = tt.expand_dims %b_rstd_48 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc79) + %b_y_49 = tt.broadcast %b_y : tensor<64x1xf32> -> tensor<64x128xf32> loc(#loc80) + %b_y_50 = arith.mulf %b_x_40, %b_y_49 : tensor<64x128xf32> loc(#loc80) + %0 = arith.truncf %b_y_50 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc32) + tt.store %p_y_35, %0, %m_x_15 : tensor<64x128x!tt.ptr> loc(#loc33) + tt.store %p_rstd_36, %b_rstd_48, %m_t_8 : tensor<64x!tt.ptr> loc(#loc34) + tt.return loc(#loc35) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S64_128S_c1_cFalse_cNone(%input: tensor<64x128xf32> loc("input"(#loc36))) -> tensor<64xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc37) + tt.reduce.return %2 : f32 loc(#loc37) + }) : (tensor<64x128xf32>) -> tensor<64xf32> loc(#loc37) + tt.return %0 : tensor<64xf32> loc(#loc39) + ^bb1: // no predecessors + %1 = ub.poison : tensor<64xf32> loc(#loc40) + tt.return %1 : tensor<64xf32> loc(#loc40) + } loc(#loc36) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc41)), %b: f32 loc("b"(#loc41))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc42) + tt.return %0 : f32 loc(#loc43) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc44) + tt.return %1 : f32 loc(#loc44) + } loc(#loc41) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:18) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:29) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:37) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc50 = loc("i_t"(#loc1)) +#loc51 = loc("i_t"(#loc2)) +#loc52 = loc("o_t"(#loc3)) +#loc53 = loc("o_t"(#loc4)) +#loc54 = loc("o_t"(#loc5)) +#loc55 = loc("o_d"(#loc6)) +#loc56 = loc("m_t"(#loc7)) +#loc57 = loc("m_x"(#loc8)) +#loc58 = loc("m_x"(#loc9)) +#loc59 = loc("m_x"(#loc10)) +#loc60 = loc("m_x"(#loc11)) +#loc61 = loc("p_x"(#loc12)) +#loc62 = loc("p_x"(#loc13)) +#loc63 = loc("p_x"(#loc14)) +#loc64 = loc("p_x"(#loc15)) +#loc65 = loc("p_x"(#loc16)) +#loc66 = loc("p_y"(#loc17)) +#loc67 = loc("p_y"(#loc18)) +#loc68 = loc("p_y"(#loc19)) +#loc69 = loc("p_y"(#loc20)) +#loc70 = loc("p_y"(#loc21)) +#loc71 = loc("p_rstd"(#loc22)) +#loc72 = loc("b_x"(#loc23)) +#loc73 = loc("b_x"(#loc24)) +#loc74 = loc("b_rstd"(#loc25)) +#loc75 = loc("b_rstd"(#loc26)) +#loc76 = loc("b_rstd"(#loc27)) +#loc77 = loc("b_rstd"(#loc28)) +#loc78 = loc("b_rstd"(#loc29)) +#loc79 = loc("b_y"(#loc30)) +#loc80 = loc("b_y"(#loc31)) diff --git a/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.ttgir b/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..f27aa354afea36a6da1b03d28871bee46654b0dd --- /dev/null +++ b/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.ttgir @@ -0,0 +1,135 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [1, 1], order = [1, 0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [2], threadsPerWarp = [32], warpsPerCTA = [1], order = [0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc1 = loc(unknown) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc34 = loc("x"(#loc)) +#loc35 = loc("y"(#loc)) +#loc36 = loc("rstd"(#loc)) +#loc37 = loc("eps"(#loc)) +#loc38 = loc("T"(#loc)) +#loc59 = loc("b_rstd"(#loc23)) +#loc66 = loc(callsite(#loc1 at #loc59)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 1 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<1.000000e+00> : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<64x128xbf16, #blocked> loc(#loc1) + %c64_i64 = arith.constant 64 : i64 loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<64x1xi64, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc39) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc40) + %o_t = arith.muli %i_t_3, %c64_i64 : i64 loc(#loc41) + %o_t_4 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc42) + %o_t_5 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #blocked1> loc(#loc42) + %o_t_6 = arith.extsi %o_t_4 : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_7 = arith.extsi %o_t_5 : tensor<64xi32, #blocked1> to tensor<64xi64, #blocked1> loc(#loc43) + %o_t_8 = tt.splat %o_t : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_9 = tt.splat %o_t : i64 -> tensor<64xi64, #blocked1> loc(#loc43) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<64xi64, #blocked1> loc(#loc43) + %m_t = arith.extsi %T : i32 to i64 loc(#loc44) + %m_t_12 = tt.splat %m_t : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_13 = tt.splat %m_t : i64 -> tensor<64xi64, #blocked1> loc(#loc44) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<64xi64, #blocked1> loc(#loc44) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<64xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64x1xi1, #blocked> loc(#loc45) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc46) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc46) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_1 : tensor<1x128xi32, #blocked> loc(#loc47) + %m_x_19 = tt.broadcast %m_x : tensor<64x1xi1, #blocked> -> tensor<64x128xi1, #blocked> loc(#loc48) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<64x128xi1, #blocked> loc(#loc48) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<64x128xi1, #blocked> loc(#loc48) + %p_x = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64x1xi64, #blocked> loc(#loc49) + %p_x_22 = arith.muli %p_x, %cst_2 : tensor<64x1xi64, #blocked> loc(#loc50) + %p_x_23 = tt.splat %x : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked> loc(#loc51) + %p_x_24 = tt.addptr %p_x_23, %p_x_22 : tensor<64x1x!tt.ptr, #blocked>, tensor<64x1xi64, #blocked> loc(#loc51) + %p_x_25 = tt.broadcast %p_x_24 : tensor<64x1x!tt.ptr, #blocked> -> tensor<64x128x!tt.ptr, #blocked> loc(#loc52) + %p_x_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<64x128xi32, #blocked> loc(#loc52) + %p_x_27 = tt.addptr %p_x_25, %p_x_26 : tensor<64x128x!tt.ptr, #blocked>, tensor<64x128xi32, #blocked> loc(#loc52) + %p_y = tt.splat %y : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked> loc(#loc53) + %p_y_28 = tt.addptr %p_y, %p_x_22 : tensor<64x1x!tt.ptr, #blocked>, tensor<64x1xi64, #blocked> loc(#loc53) + %p_y_29 = tt.broadcast %p_y_28 : tensor<64x1x!tt.ptr, #blocked> -> tensor<64x128x!tt.ptr, #blocked> loc(#loc54) + %p_y_30 = tt.addptr %p_y_29, %p_x_26 : tensor<64x128x!tt.ptr, #blocked>, tensor<64x128xi32, #blocked> loc(#loc54) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<64x!tt.ptr, #blocked1> loc(#loc55) + %p_rstd_31 = tt.addptr %p_rstd, %o_t_11 : tensor<64x!tt.ptr, #blocked1>, tensor<64xi64, #blocked1> loc(#loc55) + %b_x = tt.load %p_x_27, %m_x_21, %cst_0 : tensor<64x128x!tt.ptr, #blocked> loc(#loc56) + %b_x_32 = arith.extf %b_x : tensor<64x128xbf16, #blocked> to tensor<64x128xf32, #blocked> loc(#loc57) + %b_rstd = arith.mulf %b_x_32, %b_x_32 : tensor<64x128xf32, #blocked> loc(#loc58) + %b_rstd_33 = "tt.reduce"(%b_rstd) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_40: f32 loc(callsite(#loc1 at #loc59)), %b_rstd_41: f32 loc(callsite(#loc1 at #loc59))): + %b_rstd_42 = arith.addf %b_rstd_40, %b_rstd_41 : f32 loc(#loc67) + tt.reduce.return %b_rstd_42 : f32 loc(#loc65) + }) : (tensor<64x128xf32, #blocked>) -> tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc65) + %b_rstd_34 = tt.splat %eps : f32 -> tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_36 = math.sqrt %b_rstd_35 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc61) + %b_rstd_37 = arith.divf %cst, %b_rstd_36 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc62) + %b_y = tt.expand_dims %b_rstd_37 {axis = 1 : i32} : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64x1xf32, #blocked> loc(#loc63) + %b_y_38 = tt.broadcast %b_y : tensor<64x1xf32, #blocked> -> tensor<64x128xf32, #blocked> loc(#loc64) + %b_y_39 = arith.mulf %b_x_32, %b_y_38 : tensor<64x128xf32, #blocked> loc(#loc64) + %0 = arith.truncf %b_y_39 : tensor<64x128xf32, #blocked> to tensor<64x128xbf16, #blocked> loc(#loc30) + tt.store %p_y_30, %0, %m_x_21 : tensor<64x128x!tt.ptr, #blocked> loc(#loc31) + %1 = ttg.convert_layout %b_rstd_37 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64xf32, #blocked1> loc(#loc32) + tt.store %p_rstd_31, %1, %m_t_15 : tensor<64x!tt.ptr, #blocked1> loc(#loc32) + tt.return loc(#loc33) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc39 = loc("i_t"(#loc2)) +#loc40 = loc("i_t"(#loc3)) +#loc41 = loc("o_t"(#loc4)) +#loc42 = loc("o_t"(#loc5)) +#loc43 = loc("o_t"(#loc6)) +#loc44 = loc("m_t"(#loc7)) +#loc45 = loc("m_x"(#loc8)) +#loc46 = loc("m_x"(#loc9)) +#loc47 = loc("m_x"(#loc10)) +#loc48 = loc("m_x"(#loc11)) +#loc49 = loc("p_x"(#loc12)) +#loc50 = loc("p_x"(#loc13)) +#loc51 = loc("p_x"(#loc14)) +#loc52 = loc("p_x"(#loc15)) +#loc53 = loc("p_y"(#loc16)) +#loc54 = loc("p_y"(#loc17)) +#loc55 = loc("p_rstd"(#loc18)) +#loc56 = loc("b_x"(#loc19)) +#loc57 = loc("b_x"(#loc20)) +#loc58 = loc("b_rstd"(#loc21)) +#loc60 = loc("b_rstd"(#loc25)) +#loc61 = loc("b_rstd"(#loc26)) +#loc62 = loc("b_rstd"(#loc27)) +#loc63 = loc("b_y"(#loc28)) +#loc64 = loc("b_y"(#loc29)) +#loc65 = loc(callsite(#loc22 at #loc59)) +#loc67 = loc(callsite(#loc24 at #loc65)) diff --git a/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.ttir b/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..85ebb8b128ceed91f07ab82f5c0fe10a876f6335 --- /dev/null +++ b/triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.ttir @@ -0,0 +1,128 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc3 = loc(unknown) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc35 = loc("x"(#loc)) +#loc36 = loc("y"(#loc)) +#loc37 = loc("rstd"(#loc)) +#loc38 = loc("eps"(#loc)) +#loc39 = loc("T"(#loc)) +#loc62 = loc("b_rstd"(#loc25)) +#loc68 = loc(callsite(#loc3 at #loc62)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %b_x = arith.constant dense<0.000000e+00> : tensor<64x128xbf16> loc(#loc40) + %b_rstd = arith.constant dense<1.000000e+00> : tensor<64xf32> loc(#loc41) + %cst = arith.constant dense<128> : tensor<64x1xi64> loc(#loc3) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc42) + %o_t = arith.constant 64 : i64 loc(#loc43) + %i_t = tt.get_program_id x : i32 loc(#loc44) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc45) + %o_t_1 = arith.muli %i_t_0, %o_t : i64 loc(#loc43) + %o_t_2 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc46) + %o_t_3 = arith.extsi %o_t_2 : tensor<64xi32> to tensor<64xi64> loc(#loc47) + %o_t_4 = tt.splat %o_t_1 : i64 -> tensor<64xi64> loc(#loc47) + %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<64xi64> loc(#loc47) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc48) + %m_t = arith.extsi %T : i32 to i64 loc(#loc49) + %m_t_6 = tt.splat %m_t : i64 -> tensor<64xi64> loc(#loc49) + %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<64xi64> loc(#loc49) + %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc50) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc51) + %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc42) + %m_x_11 = tt.broadcast %m_x_8 : tensor<64x1xi1> -> tensor<64x128xi1> loc(#loc52) + %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<64x128xi1> loc(#loc52) + %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<64x128xi1> loc(#loc52) + %p_x = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc53) + %p_x_14 = arith.muli %p_x, %cst : tensor<64x1xi64> loc(#loc54) + %p_x_15 = tt.splat %x : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc55) + %p_x_16 = tt.addptr %p_x_15, %p_x_14 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc55) + %p_x_17 = tt.broadcast %p_x_16 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc56) + %p_x_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc56) + %p_x_19 = tt.addptr %p_x_17, %p_x_18 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc56) + %p_y = tt.splat %y : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc57) + %p_y_20 = tt.addptr %p_y, %p_x_14 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc57) + %p_y_21 = tt.broadcast %p_y_20 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc58) + %p_y_22 = tt.addptr %p_y_21, %p_x_18 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<64x!tt.ptr> loc(#loc59) + %p_rstd_23 = tt.addptr %p_rstd, %o_t_5 : tensor<64x!tt.ptr>, tensor<64xi64> loc(#loc59) + %b_x_24 = tt.load %p_x_19, %m_x_13, %b_x : tensor<64x128x!tt.ptr> loc(#loc40) + %b_x_25 = arith.extf %b_x_24 : tensor<64x128xbf16> to tensor<64x128xf32> loc(#loc60) + %b_rstd_26 = arith.mulf %b_x_25, %b_x_25 : tensor<64x128xf32> loc(#loc61) + %b_rstd_27 = "tt.reduce"(%b_rstd_26) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_34: f32 loc(callsite(#loc3 at #loc62)), %b_rstd_35: f32 loc(callsite(#loc3 at #loc62))): + %b_rstd_36 = arith.addf %b_rstd_34, %b_rstd_35 : f32 loc(#loc69) + tt.reduce.return %b_rstd_36 : f32 loc(#loc67) + }) : (tensor<64x128xf32>) -> tensor<64xf32> loc(#loc67) + %b_rstd_28 = tt.splat %eps : f32 -> tensor<64xf32> loc(#loc63) + %b_rstd_29 = arith.addf %b_rstd_27, %b_rstd_28 : tensor<64xf32> loc(#loc63) + %b_rstd_30 = math.sqrt %b_rstd_29 : tensor<64xf32> loc(#loc64) + %b_rstd_31 = arith.divf %b_rstd, %b_rstd_30 : tensor<64xf32> loc(#loc41) + %b_y = tt.expand_dims %b_rstd_31 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc65) + %b_y_32 = tt.broadcast %b_y : tensor<64x1xf32> -> tensor<64x128xf32> loc(#loc66) + %b_y_33 = arith.mulf %b_x_25, %b_y_32 : tensor<64x128xf32> loc(#loc66) + %0 = arith.truncf %b_y_33 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc31) + tt.store %p_y_22, %0, %m_x_13 : tensor<64x128x!tt.ptr> loc(#loc32) + tt.store %p_rstd_23, %b_rstd_31, %m_t_7 : tensor<64x!tt.ptr> loc(#loc33) + tt.return loc(#loc34) + } loc(#loc) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc40 = loc("b_x"(#loc1)) +#loc41 = loc("b_rstd"(#loc2)) +#loc42 = loc("m_x"(#loc4)) +#loc43 = loc("o_t"(#loc5)) +#loc44 = loc("i_t"(#loc6)) +#loc45 = loc("i_t"(#loc7)) +#loc46 = loc("o_t"(#loc8)) +#loc47 = loc("o_t"(#loc9)) +#loc48 = loc("o_d"(#loc10)) +#loc49 = loc("m_t"(#loc11)) +#loc50 = loc("m_x"(#loc12)) +#loc51 = loc("m_x"(#loc13)) +#loc52 = loc("m_x"(#loc14)) +#loc53 = loc("p_x"(#loc15)) +#loc54 = loc("p_x"(#loc16)) +#loc55 = loc("p_x"(#loc17)) +#loc56 = loc("p_x"(#loc18)) +#loc57 = loc("p_y"(#loc19)) +#loc58 = loc("p_y"(#loc20)) +#loc59 = loc("p_rstd"(#loc21)) +#loc60 = loc("b_x"(#loc22)) +#loc61 = loc("b_rstd"(#loc23)) +#loc63 = loc("b_rstd"(#loc27)) +#loc64 = loc("b_rstd"(#loc28)) +#loc65 = loc("b_y"(#loc29)) +#loc66 = loc("b_y"(#loc30)) +#loc67 = loc(callsite(#loc24 at #loc62)) +#loc69 = loc(callsite(#loc26 at #loc67)) diff --git a/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/__grp__chunk_bwd_kernel_dv_local.json b/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/__grp__chunk_bwd_kernel_dv_local.json new file mode 100644 index 0000000000000000000000000000000000000000..24d34f3966aea27fbe6dd473f9fd88c2684ce428 --- /dev/null +++ b/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/__grp__chunk_bwd_kernel_dv_local.json @@ -0,0 +1 @@ +{"child_paths": {"chunk_bwd_kernel_dv_local.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.source", "chunk_bwd_kernel_dv_local.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.ttir", "chunk_bwd_kernel_dv_local.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.ttgir", "chunk_bwd_kernel_dv_local.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.llir", "chunk_bwd_kernel_dv_local.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.ptx", "chunk_bwd_kernel_dv_local.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.cubin", "chunk_bwd_kernel_dv_local.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.json"}} \ No newline at end of file diff --git a/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.cubin b/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.cubin new file mode 100644 index 0000000000000000000000000000000000000000..c9ac5eed15e2c3ad8cb8357bae0a523771dffe3c --- /dev/null +++ b/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.cubin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1cceb66099d8d655c2aea0a8fe118bd5bd3148f3a9aaa409b66aa3ecaa69afc4 +size 160256 diff --git a/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.json b/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.json new file mode 100644 index 0000000000000000000000000000000000000000..7317882ffa22497725789659fd5e4dae5030ccf2 --- /dev/null +++ b/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.json @@ -0,0 +1 @@ +{"hash": "1733707d5bdc2c13194a054e4b31e82e36bfc151a9c8b58be2f97129aef4babd", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 2, "num_ctas": 1, "num_stages": 4, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 32768, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "chunk_bwd_kernel_dv_local"} \ No newline at end of file diff --git a/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.llir b/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.llir new file mode 100644 index 0000000000000000000000000000000000000000..7ef244eee71fad93f233656aa656ae697d3e326f --- /dev/null +++ b/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.llir @@ -0,0 +1,3893 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @chunk_bwd_kernel_dv_local(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, ptr addrspace(1) %3, ptr addrspace(1) %4, float %5, i32 %6, ptr addrspace(1) readnone captures(none) %7, ptr addrspace(1) readnone captures(none) %8) local_unnamed_addr #0 !dbg !5 { + %10 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %11 = zext nneg i32 %10 to i64, !dbg !14 + %12 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.y(), !dbg !15 + %13 = zext nneg i32 %12 to i64, !dbg !16 + %14 = and i64 %13, 65504, !dbg !17 + %15 = and i64 %13, 31, !dbg !18 + %16 = sext i32 %6 to i64, !dbg !17 + %17 = mul nsw i64 %14, %16, !dbg !19 + %18 = or disjoint i64 %17, %15, !dbg !20 + %19 = shl nsw i64 %18, 7, !dbg !21 + %20 = getelementptr bfloat, ptr addrspace(1) %0, i64 %19, !dbg !22 + %21 = getelementptr bfloat, ptr addrspace(1) %1, i64 %19, !dbg !23 + %22 = shl nuw nsw i64 %11, 6, !dbg !24 + %23 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !25 + %24 = and i32 %23, 48, !dbg !25 + %25 = lshr exact i32 %24, 4, !dbg !25 + %26 = and i32 %23, 28, !dbg !25 + %27 = and i32 %23, 3, !dbg !25 + %28 = and i32 %23, 32, !dbg !25 + %29 = insertelement <16 x i32> poison, i32 %23, i64 0, !dbg !25 + %30 = insertelement <16 x i32> %29, i32 %25, i64 1, !dbg !25 + %31 = shufflevector <16 x i32> %30, <16 x i32> poison, <16 x i32> , !dbg !25 + %32 = and <16 x i32> %31, , !dbg !25 + %33 = or disjoint <16 x i32> %31, , !dbg !25 + %34 = shufflevector <16 x i32> %32, <16 x i32> %33, <16 x i32> , !dbg !25 + %35 = zext nneg i32 %25 to i64, !dbg !26 + %36 = zext nneg <16 x i32> %34 to <16 x i64>, !dbg !26 + %37 = or disjoint i64 %22, %35, !dbg !26 + %38 = insertelement <16 x i64> poison, i64 %22, i64 0, !dbg !26 + %39 = shufflevector <16 x i64> %38, <16 x i64> poison, <16 x i32> zeroinitializer, !dbg !26 + %40 = or disjoint <16 x i64> %39, %36, !dbg !26 + %41 = icmp slt i64 %37, %16, !dbg !27 + %42 = insertelement <16 x i32> poison, i32 %6, i64 0, !dbg !27 + %43 = shufflevector <16 x i32> %42, <16 x i32> poison, <16 x i32> zeroinitializer, !dbg !27 + %44 = sext <16 x i32> %43 to <16 x i64>, !dbg !27 + %45 = icmp slt <16 x i64> %40, %44, !dbg !27 + %46 = getelementptr float, ptr addrspace(1) %2, i64 %18, !dbg !28 + %47 = extractelement <16 x i64> %40, i64 0, !dbg !29 + %.idx = shl nuw nsw i64 %47, 7, !dbg !29 + %48 = getelementptr i8, ptr addrspace(1) %46, i64 %.idx, !dbg !29 + %49 = extractelement <16 x i1> %45, i64 0, !dbg !30 + %50 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %48, i1 %49) #7, !dbg !30 + %51 = and i32 %23, 7, !dbg !31 + %52 = shl nuw nsw i32 %51, 4, !dbg !31 + %53 = and i32 %23, 24, !dbg !31 + %54 = lshr exact i32 %53, 1, !dbg !31 + %55 = shl nuw nsw i32 %28, 2, !dbg !31 + %56 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %55, !dbg !31 + %57 = getelementptr inbounds nuw i8, ptr addrspace(3) %56, i32 %54, !dbg !31 + %58 = getelementptr inbounds nuw i8, ptr addrspace(3) %57, i32 %52, !dbg !31 + %59 = insertelement <1 x i32> poison, i32 %50, i64 0, !dbg !31 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) %58, <1 x i32> %59) #7, !dbg !31 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !31 + %60 = shl nuw nsw i32 %26, 2, !dbg !31 + %61 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %60, !dbg !31 + %62 = load float, ptr addrspace(3) %61, align 16, !dbg !31 + %63 = getelementptr inbounds nuw i8, ptr addrspace(3) %61, i32 4, !dbg !31 + %64 = load float, ptr addrspace(3) %63, align 4, !dbg !31 + %65 = getelementptr inbounds nuw i8, ptr addrspace(3) %61, i32 8, !dbg !31 + %66 = load float, ptr addrspace(3) %65, align 8, !dbg !31 + %67 = getelementptr inbounds nuw i8, ptr addrspace(3) %61, i32 12, !dbg !31 + %68 = load float, ptr addrspace(3) %67, align 4, !dbg !31 + %69 = getelementptr inbounds nuw i8, ptr addrspace(3) %61, i32 128, !dbg !31 + %70 = load float, ptr addrspace(3) %69, align 16, !dbg !31 + %71 = getelementptr inbounds nuw i8, ptr addrspace(3) %61, i32 132, !dbg !31 + %72 = load float, ptr addrspace(3) %71, align 4, !dbg !31 + %73 = getelementptr inbounds nuw i8, ptr addrspace(3) %61, i32 136, !dbg !31 + %74 = load float, ptr addrspace(3) %73, align 8, !dbg !31 + %75 = getelementptr inbounds nuw i8, ptr addrspace(3) %61, i32 140, !dbg !31 + %76 = load float, ptr addrspace(3) %75, align 4, !dbg !31 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !32 + %77 = shl nuw nsw i32 %23, 3, !dbg !32 + %78 = and i32 %77, 112, !dbg !32 + %79 = shl nuw nsw i32 %23, 2, !dbg !32 + %80 = and i32 %79, 132, !dbg !32 + %81 = and i32 %23, 16, !dbg !32 + %82 = lshr exact i32 %81, 1, !dbg !32 + %83 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %78, !dbg !32 + %84 = getelementptr inbounds nuw i8, ptr addrspace(3) %83, i32 %80, !dbg !32 + %85 = getelementptr inbounds nuw i8, ptr addrspace(3) %84, i32 %82, !dbg !32 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) %85, <1 x i32> %59) #7, !dbg !32 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !32 + %86 = shl nuw nsw i32 %27, 4, !dbg !32 + %87 = shl nuw nsw i32 %28, 1, !dbg !32 + %88 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %87, !dbg !32 + %89 = getelementptr inbounds nuw i8, ptr addrspace(3) %88, i32 %86, !dbg !32 + %90 = load float, ptr addrspace(3) %89, align 16, !dbg !32 + %91 = getelementptr inbounds nuw i8, ptr addrspace(3) %89, i32 4, !dbg !32 + %92 = load float, ptr addrspace(3) %91, align 4, !dbg !32 + %93 = getelementptr inbounds nuw i8, ptr addrspace(3) %89, i32 8, !dbg !32 + %94 = load float, ptr addrspace(3) %93, align 8, !dbg !32 + %95 = getelementptr inbounds nuw i8, ptr addrspace(3) %89, i32 12, !dbg !32 + %96 = load float, ptr addrspace(3) %95, align 4, !dbg !32 + %97 = getelementptr inbounds nuw i8, ptr addrspace(3) %89, i32 128, !dbg !32 + %98 = load float, ptr addrspace(3) %97, align 16, !dbg !32 + %99 = getelementptr inbounds nuw i8, ptr addrspace(3) %89, i32 132, !dbg !32 + %100 = load float, ptr addrspace(3) %99, align 4, !dbg !32 + %101 = getelementptr inbounds nuw i8, ptr addrspace(3) %89, i32 136, !dbg !32 + %102 = load float, ptr addrspace(3) %101, align 8, !dbg !32 + %103 = getelementptr inbounds nuw i8, ptr addrspace(3) %89, i32 140, !dbg !32 + %104 = load float, ptr addrspace(3) %103, align 4, !dbg !32 + %105 = and i32 %23, 15, !dbg !37 + %106 = shl nuw nsw i32 %105, 3, !dbg !37 + %107 = shl nuw nsw i64 %37, 12, !dbg !38 + %108 = extractelement <16 x i64> %40, i64 15, !dbg !38 + %109 = shl nuw nsw i64 %108, 12, !dbg !38 + %110 = extractelement <16 x i64> %40, i64 14, !dbg !38 + %111 = shl nuw nsw i64 %110, 12, !dbg !38 + %112 = extractelement <16 x i64> %40, i64 13, !dbg !38 + %113 = shl nuw nsw i64 %112, 12, !dbg !38 + %114 = extractelement <16 x i64> %40, i64 12, !dbg !38 + %115 = shl nuw nsw i64 %114, 12, !dbg !38 + %116 = extractelement <16 x i64> %40, i64 11, !dbg !38 + %117 = shl nuw nsw i64 %116, 12, !dbg !38 + %118 = extractelement <16 x i64> %40, i64 10, !dbg !38 + %119 = shl nuw nsw i64 %118, 12, !dbg !38 + %120 = extractelement <16 x i64> %40, i64 9, !dbg !38 + %121 = shl nuw nsw i64 %120, 12, !dbg !38 + %122 = extractelement <16 x i64> %40, i64 8, !dbg !38 + %123 = shl nuw nsw i64 %122, 12, !dbg !38 + %124 = extractelement <16 x i64> %40, i64 7, !dbg !38 + %125 = shl nuw nsw i64 %124, 12, !dbg !38 + %126 = extractelement <16 x i64> %40, i64 6, !dbg !38 + %127 = shl nuw nsw i64 %126, 12, !dbg !38 + %128 = extractelement <16 x i64> %40, i64 5, !dbg !38 + %129 = shl nuw nsw i64 %128, 12, !dbg !38 + %130 = extractelement <16 x i64> %40, i64 4, !dbg !38 + %131 = shl nuw nsw i64 %130, 12, !dbg !38 + %132 = extractelement <16 x i64> %40, i64 3, !dbg !38 + %133 = shl nuw nsw i64 %132, 12, !dbg !38 + %134 = extractelement <16 x i64> %40, i64 2, !dbg !38 + %135 = shl nuw nsw i64 %134, 12, !dbg !38 + %136 = extractelement <16 x i64> %40, i64 1, !dbg !38 + %137 = shl nuw nsw i64 %136, 12, !dbg !38 + %138 = getelementptr bfloat, ptr addrspace(1) %21, i64 %107, !dbg !39 + %139 = getelementptr bfloat, ptr addrspace(1) %21, i64 %109, !dbg !39 + %140 = getelementptr bfloat, ptr addrspace(1) %21, i64 %111, !dbg !39 + %141 = getelementptr bfloat, ptr addrspace(1) %21, i64 %113, !dbg !39 + %142 = getelementptr bfloat, ptr addrspace(1) %21, i64 %115, !dbg !39 + %143 = getelementptr bfloat, ptr addrspace(1) %21, i64 %117, !dbg !39 + %144 = getelementptr bfloat, ptr addrspace(1) %21, i64 %119, !dbg !39 + %145 = getelementptr bfloat, ptr addrspace(1) %21, i64 %121, !dbg !39 + %146 = getelementptr bfloat, ptr addrspace(1) %21, i64 %123, !dbg !39 + %147 = getelementptr bfloat, ptr addrspace(1) %21, i64 %125, !dbg !39 + %148 = getelementptr bfloat, ptr addrspace(1) %21, i64 %127, !dbg !39 + %149 = getelementptr bfloat, ptr addrspace(1) %21, i64 %129, !dbg !39 + %150 = getelementptr bfloat, ptr addrspace(1) %21, i64 %131, !dbg !39 + %151 = getelementptr bfloat, ptr addrspace(1) %21, i64 %133, !dbg !39 + %152 = getelementptr bfloat, ptr addrspace(1) %21, i64 %135, !dbg !39 + %153 = getelementptr bfloat, ptr addrspace(1) %21, i64 %137, !dbg !39 + %154 = zext nneg i32 %106 to i64, !dbg !40 + %155 = getelementptr bfloat, ptr addrspace(1) %138, i64 %154, !dbg !40 + %156 = getelementptr bfloat, ptr addrspace(1) %139, i64 %154, !dbg !40 + %157 = getelementptr bfloat, ptr addrspace(1) %140, i64 %154, !dbg !40 + %158 = getelementptr bfloat, ptr addrspace(1) %141, i64 %154, !dbg !40 + %159 = getelementptr bfloat, ptr addrspace(1) %142, i64 %154, !dbg !40 + %160 = getelementptr bfloat, ptr addrspace(1) %143, i64 %154, !dbg !40 + %161 = getelementptr bfloat, ptr addrspace(1) %144, i64 %154, !dbg !40 + %162 = getelementptr bfloat, ptr addrspace(1) %145, i64 %154, !dbg !40 + %163 = getelementptr bfloat, ptr addrspace(1) %146, i64 %154, !dbg !40 + %164 = getelementptr bfloat, ptr addrspace(1) %147, i64 %154, !dbg !40 + %165 = getelementptr bfloat, ptr addrspace(1) %148, i64 %154, !dbg !40 + %166 = getelementptr bfloat, ptr addrspace(1) %149, i64 %154, !dbg !40 + %167 = getelementptr bfloat, ptr addrspace(1) %150, i64 %154, !dbg !40 + %168 = getelementptr bfloat, ptr addrspace(1) %151, i64 %154, !dbg !40 + %169 = getelementptr bfloat, ptr addrspace(1) %152, i64 %154, !dbg !40 + %170 = getelementptr bfloat, ptr addrspace(1) %153, i64 %154, !dbg !40 + %171 = getelementptr bfloat, ptr addrspace(1) %20, i64 %154, !dbg !41 + %172 = getelementptr bfloat, ptr addrspace(1) %171, i64 %107, !dbg !42 + %173 = getelementptr bfloat, ptr addrspace(1) %171, i64 %109, !dbg !42 + %174 = getelementptr bfloat, ptr addrspace(1) %171, i64 %111, !dbg !42 + %175 = getelementptr bfloat, ptr addrspace(1) %171, i64 %113, !dbg !42 + %176 = getelementptr bfloat, ptr addrspace(1) %171, i64 %115, !dbg !42 + %177 = getelementptr bfloat, ptr addrspace(1) %171, i64 %117, !dbg !42 + %178 = getelementptr bfloat, ptr addrspace(1) %171, i64 %119, !dbg !42 + %179 = getelementptr bfloat, ptr addrspace(1) %171, i64 %121, !dbg !42 + %180 = getelementptr bfloat, ptr addrspace(1) %171, i64 %123, !dbg !42 + %181 = getelementptr bfloat, ptr addrspace(1) %171, i64 %125, !dbg !42 + %182 = getelementptr bfloat, ptr addrspace(1) %171, i64 %127, !dbg !42 + %183 = getelementptr bfloat, ptr addrspace(1) %171, i64 %129, !dbg !42 + %184 = getelementptr bfloat, ptr addrspace(1) %171, i64 %131, !dbg !42 + %185 = getelementptr bfloat, ptr addrspace(1) %171, i64 %133, !dbg !42 + %186 = getelementptr bfloat, ptr addrspace(1) %171, i64 %135, !dbg !42 + %187 = getelementptr bfloat, ptr addrspace(1) %171, i64 %137, !dbg !42 + %188 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %155, i1 %41) #7, !dbg !43 + %189 = extractvalue { i32, i32, i32, i32 } %188, 0, !dbg !43 + %190 = extractvalue { i32, i32, i32, i32 } %188, 1, !dbg !43 + %191 = extractvalue { i32, i32, i32, i32 } %188, 2, !dbg !43 + %192 = extractvalue { i32, i32, i32, i32 } %188, 3, !dbg !43 + %193 = extractelement <16 x i1> %45, i64 15, !dbg !43 + %194 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %156, i1 %193) #7, !dbg !43 + %195 = extractvalue { i32, i32, i32, i32 } %194, 0, !dbg !43 + %196 = extractvalue { i32, i32, i32, i32 } %194, 1, !dbg !43 + %197 = extractvalue { i32, i32, i32, i32 } %194, 2, !dbg !43 + %198 = extractvalue { i32, i32, i32, i32 } %194, 3, !dbg !43 + %199 = extractelement <16 x i1> %45, i64 14, !dbg !43 + %200 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %157, i1 %199) #7, !dbg !43 + %201 = extractvalue { i32, i32, i32, i32 } %200, 0, !dbg !43 + %202 = extractvalue { i32, i32, i32, i32 } %200, 1, !dbg !43 + %203 = extractvalue { i32, i32, i32, i32 } %200, 2, !dbg !43 + %204 = extractvalue { i32, i32, i32, i32 } %200, 3, !dbg !43 + %205 = extractelement <16 x i1> %45, i64 13, !dbg !43 + %206 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %158, i1 %205) #7, !dbg !43 + %207 = extractvalue { i32, i32, i32, i32 } %206, 0, !dbg !43 + %208 = extractvalue { i32, i32, i32, i32 } %206, 1, !dbg !43 + %209 = extractvalue { i32, i32, i32, i32 } %206, 2, !dbg !43 + %210 = extractvalue { i32, i32, i32, i32 } %206, 3, !dbg !43 + %211 = extractelement <16 x i1> %45, i64 12, !dbg !43 + %212 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %159, i1 %211) #7, !dbg !43 + %213 = extractvalue { i32, i32, i32, i32 } %212, 0, !dbg !43 + %214 = extractvalue { i32, i32, i32, i32 } %212, 1, !dbg !43 + %215 = extractvalue { i32, i32, i32, i32 } %212, 2, !dbg !43 + %216 = extractvalue { i32, i32, i32, i32 } %212, 3, !dbg !43 + %217 = extractelement <16 x i1> %45, i64 11, !dbg !43 + %218 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %160, i1 %217) #7, !dbg !43 + %219 = extractvalue { i32, i32, i32, i32 } %218, 0, !dbg !43 + %220 = extractvalue { i32, i32, i32, i32 } %218, 1, !dbg !43 + %221 = extractvalue { i32, i32, i32, i32 } %218, 2, !dbg !43 + %222 = extractvalue { i32, i32, i32, i32 } %218, 3, !dbg !43 + %223 = extractelement <16 x i1> %45, i64 10, !dbg !43 + %224 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %161, i1 %223) #7, !dbg !43 + %225 = extractvalue { i32, i32, i32, i32 } %224, 0, !dbg !43 + %226 = extractvalue { i32, i32, i32, i32 } %224, 1, !dbg !43 + %227 = extractvalue { i32, i32, i32, i32 } %224, 2, !dbg !43 + %228 = extractvalue { i32, i32, i32, i32 } %224, 3, !dbg !43 + %229 = extractelement <16 x i1> %45, i64 9, !dbg !43 + %230 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %162, i1 %229) #7, !dbg !43 + %231 = extractvalue { i32, i32, i32, i32 } %230, 0, !dbg !43 + %232 = extractvalue { i32, i32, i32, i32 } %230, 1, !dbg !43 + %233 = extractvalue { i32, i32, i32, i32 } %230, 2, !dbg !43 + %234 = extractvalue { i32, i32, i32, i32 } %230, 3, !dbg !43 + %235 = extractelement <16 x i1> %45, i64 8, !dbg !43 + %236 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %163, i1 %235) #7, !dbg !43 + %237 = extractvalue { i32, i32, i32, i32 } %236, 0, !dbg !43 + %238 = extractvalue { i32, i32, i32, i32 } %236, 1, !dbg !43 + %239 = extractvalue { i32, i32, i32, i32 } %236, 2, !dbg !43 + %240 = extractvalue { i32, i32, i32, i32 } %236, 3, !dbg !43 + %241 = extractelement <16 x i1> %45, i64 7, !dbg !43 + %242 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %164, i1 %241) #7, !dbg !43 + %243 = extractvalue { i32, i32, i32, i32 } %242, 0, !dbg !43 + %244 = extractvalue { i32, i32, i32, i32 } %242, 1, !dbg !43 + %245 = extractvalue { i32, i32, i32, i32 } %242, 2, !dbg !43 + %246 = extractvalue { i32, i32, i32, i32 } %242, 3, !dbg !43 + %247 = extractelement <16 x i1> %45, i64 6, !dbg !43 + %248 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %165, i1 %247) #7, !dbg !43 + %249 = extractvalue { i32, i32, i32, i32 } %248, 0, !dbg !43 + %250 = extractvalue { i32, i32, i32, i32 } %248, 1, !dbg !43 + %251 = extractvalue { i32, i32, i32, i32 } %248, 2, !dbg !43 + %252 = extractvalue { i32, i32, i32, i32 } %248, 3, !dbg !43 + %253 = extractelement <16 x i1> %45, i64 5, !dbg !43 + %254 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %166, i1 %253) #7, !dbg !43 + %255 = extractvalue { i32, i32, i32, i32 } %254, 0, !dbg !43 + %256 = extractvalue { i32, i32, i32, i32 } %254, 1, !dbg !43 + %257 = extractvalue { i32, i32, i32, i32 } %254, 2, !dbg !43 + %258 = extractvalue { i32, i32, i32, i32 } %254, 3, !dbg !43 + %259 = extractelement <16 x i1> %45, i64 4, !dbg !43 + %260 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %167, i1 %259) #7, !dbg !43 + %261 = extractvalue { i32, i32, i32, i32 } %260, 0, !dbg !43 + %262 = extractvalue { i32, i32, i32, i32 } %260, 1, !dbg !43 + %263 = extractvalue { i32, i32, i32, i32 } %260, 2, !dbg !43 + %264 = extractvalue { i32, i32, i32, i32 } %260, 3, !dbg !43 + %265 = extractelement <16 x i1> %45, i64 3, !dbg !43 + %266 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %168, i1 %265) #7, !dbg !43 + %267 = extractvalue { i32, i32, i32, i32 } %266, 0, !dbg !43 + %268 = extractvalue { i32, i32, i32, i32 } %266, 1, !dbg !43 + %269 = extractvalue { i32, i32, i32, i32 } %266, 2, !dbg !43 + %270 = extractvalue { i32, i32, i32, i32 } %266, 3, !dbg !43 + %271 = extractelement <16 x i1> %45, i64 2, !dbg !43 + %272 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %169, i1 %271) #7, !dbg !43 + %273 = extractvalue { i32, i32, i32, i32 } %272, 0, !dbg !43 + %274 = extractvalue { i32, i32, i32, i32 } %272, 1, !dbg !43 + %275 = extractvalue { i32, i32, i32, i32 } %272, 2, !dbg !43 + %276 = extractvalue { i32, i32, i32, i32 } %272, 3, !dbg !43 + %277 = extractelement <16 x i1> %45, i64 1, !dbg !43 + %278 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %170, i1 %277) #7, !dbg !43 + %279 = extractvalue { i32, i32, i32, i32 } %278, 0, !dbg !43 + %280 = extractvalue { i32, i32, i32, i32 } %278, 1, !dbg !43 + %281 = extractvalue { i32, i32, i32, i32 } %278, 2, !dbg !43 + %282 = extractvalue { i32, i32, i32, i32 } %278, 3, !dbg !43 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !43 + %283 = shl nuw nsw i32 %23, 4, !dbg !43 + %284 = and i32 %283, 1008, !dbg !43 + %285 = xor i32 %284, %24, !dbg !43 + %286 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %285, !dbg !43 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %286, i32 %189, i32 %190, i32 %191, i32 %192) #7, !dbg !43 + %287 = or disjoint i32 %285, 2048, !dbg !43 + %288 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %287, !dbg !43 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %288, i32 %201, i32 %202, i32 %203, i32 %204) #7, !dbg !43 + %289 = or disjoint i32 %285, 4096, !dbg !43 + %290 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %289, !dbg !43 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %290, i32 %213, i32 %214, i32 %215, i32 %216) #7, !dbg !43 + %291 = or disjoint i32 %285, 6144, !dbg !43 + %292 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %291, !dbg !43 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %292, i32 %225, i32 %226, i32 %227, i32 %228) #7, !dbg !43 + %293 = or disjoint i32 %285, 8192, !dbg !43 + %294 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %293, !dbg !43 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %294, i32 %237, i32 %238, i32 %239, i32 %240) #7, !dbg !43 + %295 = or disjoint i32 %285, 10240, !dbg !43 + %296 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %295, !dbg !43 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %296, i32 %249, i32 %250, i32 %251, i32 %252) #7, !dbg !43 + %297 = or disjoint i32 %285, 12288, !dbg !43 + %298 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %297, !dbg !43 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %298, i32 %261, i32 %262, i32 %263, i32 %264) #7, !dbg !43 + %299 = or disjoint i32 %285, 14336, !dbg !43 + %300 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %299, !dbg !43 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %300, i32 %273, i32 %274, i32 %275, i32 %276) #7, !dbg !43 + %301 = xor i32 %285, 1088, !dbg !43 + %302 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %301, !dbg !43 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %302, i32 %195, i32 %196, i32 %197, i32 %198) #7, !dbg !43 + %303 = or disjoint i32 %301, 2048, !dbg !43 + %304 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %303, !dbg !43 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %304, i32 %207, i32 %208, i32 %209, i32 %210) #7, !dbg !43 + %305 = or disjoint i32 %301, 4096, !dbg !43 + %306 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %305, !dbg !43 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %306, i32 %219, i32 %220, i32 %221, i32 %222) #7, !dbg !43 + %307 = or disjoint i32 %301, 6144, !dbg !43 + %308 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %307, !dbg !43 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %308, i32 %231, i32 %232, i32 %233, i32 %234) #7, !dbg !43 + %309 = or disjoint i32 %301, 8192, !dbg !43 + %310 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %309, !dbg !43 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %310, i32 %243, i32 %244, i32 %245, i32 %246) #7, !dbg !43 + %311 = or disjoint i32 %301, 10240, !dbg !43 + %312 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %311, !dbg !43 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %312, i32 %255, i32 %256, i32 %257, i32 %258) #7, !dbg !43 + %313 = or disjoint i32 %301, 12288, !dbg !43 + %314 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %313, !dbg !43 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %314, i32 %267, i32 %268, i32 %269, i32 %270) #7, !dbg !43 + %315 = or disjoint i32 %301, 14336, !dbg !43 + %316 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %315, !dbg !43 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %316, i32 %279, i32 %280, i32 %281, i32 %282) #7, !dbg !43 + %317 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %172, i1 %41) #7, !dbg !44 + %318 = extractvalue { i32, i32, i32, i32 } %317, 0, !dbg !44 + %319 = extractvalue { i32, i32, i32, i32 } %317, 1, !dbg !44 + %320 = extractvalue { i32, i32, i32, i32 } %317, 2, !dbg !44 + %321 = extractvalue { i32, i32, i32, i32 } %317, 3, !dbg !44 + %322 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %173, i1 %193) #7, !dbg !44 + %323 = extractvalue { i32, i32, i32, i32 } %322, 0, !dbg !44 + %324 = extractvalue { i32, i32, i32, i32 } %322, 1, !dbg !44 + %325 = extractvalue { i32, i32, i32, i32 } %322, 2, !dbg !44 + %326 = extractvalue { i32, i32, i32, i32 } %322, 3, !dbg !44 + %327 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %174, i1 %199) #7, !dbg !44 + %328 = extractvalue { i32, i32, i32, i32 } %327, 0, !dbg !44 + %329 = extractvalue { i32, i32, i32, i32 } %327, 1, !dbg !44 + %330 = extractvalue { i32, i32, i32, i32 } %327, 2, !dbg !44 + %331 = extractvalue { i32, i32, i32, i32 } %327, 3, !dbg !44 + %332 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %175, i1 %205) #7, !dbg !44 + %333 = extractvalue { i32, i32, i32, i32 } %332, 0, !dbg !44 + %334 = extractvalue { i32, i32, i32, i32 } %332, 1, !dbg !44 + %335 = extractvalue { i32, i32, i32, i32 } %332, 2, !dbg !44 + %336 = extractvalue { i32, i32, i32, i32 } %332, 3, !dbg !44 + %337 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %176, i1 %211) #7, !dbg !44 + %338 = extractvalue { i32, i32, i32, i32 } %337, 0, !dbg !44 + %339 = extractvalue { i32, i32, i32, i32 } %337, 1, !dbg !44 + %340 = extractvalue { i32, i32, i32, i32 } %337, 2, !dbg !44 + %341 = extractvalue { i32, i32, i32, i32 } %337, 3, !dbg !44 + %342 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %177, i1 %217) #7, !dbg !44 + %343 = extractvalue { i32, i32, i32, i32 } %342, 0, !dbg !44 + %344 = extractvalue { i32, i32, i32, i32 } %342, 1, !dbg !44 + %345 = extractvalue { i32, i32, i32, i32 } %342, 2, !dbg !44 + %346 = extractvalue { i32, i32, i32, i32 } %342, 3, !dbg !44 + %347 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %178, i1 %223) #7, !dbg !44 + %348 = extractvalue { i32, i32, i32, i32 } %347, 0, !dbg !44 + %349 = extractvalue { i32, i32, i32, i32 } %347, 1, !dbg !44 + %350 = extractvalue { i32, i32, i32, i32 } %347, 2, !dbg !44 + %351 = extractvalue { i32, i32, i32, i32 } %347, 3, !dbg !44 + %352 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %179, i1 %229) #7, !dbg !44 + %353 = extractvalue { i32, i32, i32, i32 } %352, 0, !dbg !44 + %354 = extractvalue { i32, i32, i32, i32 } %352, 1, !dbg !44 + %355 = extractvalue { i32, i32, i32, i32 } %352, 2, !dbg !44 + %356 = extractvalue { i32, i32, i32, i32 } %352, 3, !dbg !44 + %357 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %180, i1 %235) #7, !dbg !44 + %358 = extractvalue { i32, i32, i32, i32 } %357, 0, !dbg !44 + %359 = extractvalue { i32, i32, i32, i32 } %357, 1, !dbg !44 + %360 = extractvalue { i32, i32, i32, i32 } %357, 2, !dbg !44 + %361 = extractvalue { i32, i32, i32, i32 } %357, 3, !dbg !44 + %362 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %181, i1 %241) #7, !dbg !44 + %363 = extractvalue { i32, i32, i32, i32 } %362, 0, !dbg !44 + %364 = extractvalue { i32, i32, i32, i32 } %362, 1, !dbg !44 + %365 = extractvalue { i32, i32, i32, i32 } %362, 2, !dbg !44 + %366 = extractvalue { i32, i32, i32, i32 } %362, 3, !dbg !44 + %367 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %182, i1 %247) #7, !dbg !44 + %368 = extractvalue { i32, i32, i32, i32 } %367, 0, !dbg !44 + %369 = extractvalue { i32, i32, i32, i32 } %367, 1, !dbg !44 + %370 = extractvalue { i32, i32, i32, i32 } %367, 2, !dbg !44 + %371 = extractvalue { i32, i32, i32, i32 } %367, 3, !dbg !44 + %372 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %183, i1 %253) #7, !dbg !44 + %373 = extractvalue { i32, i32, i32, i32 } %372, 0, !dbg !44 + %374 = extractvalue { i32, i32, i32, i32 } %372, 1, !dbg !44 + %375 = extractvalue { i32, i32, i32, i32 } %372, 2, !dbg !44 + %376 = extractvalue { i32, i32, i32, i32 } %372, 3, !dbg !44 + %377 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %184, i1 %259) #7, !dbg !44 + %378 = extractvalue { i32, i32, i32, i32 } %377, 0, !dbg !44 + %379 = extractvalue { i32, i32, i32, i32 } %377, 1, !dbg !44 + %380 = extractvalue { i32, i32, i32, i32 } %377, 2, !dbg !44 + %381 = extractvalue { i32, i32, i32, i32 } %377, 3, !dbg !44 + %382 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %185, i1 %265) #7, !dbg !44 + %383 = extractvalue { i32, i32, i32, i32 } %382, 0, !dbg !44 + %384 = extractvalue { i32, i32, i32, i32 } %382, 1, !dbg !44 + %385 = extractvalue { i32, i32, i32, i32 } %382, 2, !dbg !44 + %386 = extractvalue { i32, i32, i32, i32 } %382, 3, !dbg !44 + %387 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %186, i1 %271) #7, !dbg !44 + %388 = extractvalue { i32, i32, i32, i32 } %387, 0, !dbg !44 + %389 = extractvalue { i32, i32, i32, i32 } %387, 1, !dbg !44 + %390 = extractvalue { i32, i32, i32, i32 } %387, 2, !dbg !44 + %391 = extractvalue { i32, i32, i32, i32 } %387, 3, !dbg !44 + %392 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %187, i1 %277) #7, !dbg !44 + %393 = extractvalue { i32, i32, i32, i32 } %392, 0, !dbg !44 + %394 = extractvalue { i32, i32, i32, i32 } %392, 1, !dbg !44 + %395 = extractvalue { i32, i32, i32, i32 } %392, 2, !dbg !44 + %396 = extractvalue { i32, i32, i32, i32 } %392, 3, !dbg !44 + %397 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %285, !dbg !44 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %397, i32 %318, i32 %319, i32 %320, i32 %321) #7, !dbg !44 + %398 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %287, !dbg !44 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %398, i32 %328, i32 %329, i32 %330, i32 %331) #7, !dbg !44 + %399 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %289, !dbg !44 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %399, i32 %338, i32 %339, i32 %340, i32 %341) #7, !dbg !44 + %400 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %291, !dbg !44 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %400, i32 %348, i32 %349, i32 %350, i32 %351) #7, !dbg !44 + %401 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %293, !dbg !44 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %401, i32 %358, i32 %359, i32 %360, i32 %361) #7, !dbg !44 + %402 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %295, !dbg !44 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %402, i32 %368, i32 %369, i32 %370, i32 %371) #7, !dbg !44 + %403 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %297, !dbg !44 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %403, i32 %378, i32 %379, i32 %380, i32 %381) #7, !dbg !44 + %404 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %299, !dbg !44 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %404, i32 %388, i32 %389, i32 %390, i32 %391) #7, !dbg !44 + %405 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %301, !dbg !44 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %405, i32 %323, i32 %324, i32 %325, i32 %326) #7, !dbg !44 + %406 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %303, !dbg !44 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %406, i32 %333, i32 %334, i32 %335, i32 %336) #7, !dbg !44 + %407 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %305, !dbg !44 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %407, i32 %343, i32 %344, i32 %345, i32 %346) #7, !dbg !44 + %408 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %307, !dbg !44 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %408, i32 %353, i32 %354, i32 %355, i32 %356) #7, !dbg !44 + %409 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %309, !dbg !44 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %409, i32 %363, i32 %364, i32 %365, i32 %366) #7, !dbg !44 + %410 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %311, !dbg !44 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %410, i32 %373, i32 %374, i32 %375, i32 %376) #7, !dbg !44 + %411 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %313, !dbg !44 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %411, i32 %383, i32 %384, i32 %385, i32 %386) #7, !dbg !44 + %412 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %315, !dbg !44 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %412, i32 %393, i32 %394, i32 %395, i32 %396) #7, !dbg !44 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !43 + %413 = shl nuw nsw i32 %105, 8, !dbg !43 + %414 = or disjoint i32 %413, %52, !dbg !43 + %415 = xor i32 %414, %81, !dbg !43 + %416 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %415, !dbg !43 + %417 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %416), !dbg !43 + %418 = extractvalue { i32, i32, i32, i32 } %417, 0, !dbg !43 + %419 = extractvalue { i32, i32, i32, i32 } %417, 1, !dbg !43 + %420 = extractvalue { i32, i32, i32, i32 } %417, 2, !dbg !43 + %421 = extractvalue { i32, i32, i32, i32 } %417, 3, !dbg !43 + %422 = getelementptr inbounds nuw i8, ptr addrspace(3) %416, i32 128, !dbg !43 + %423 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %422), !dbg !43 + %424 = extractvalue { i32, i32, i32, i32 } %423, 0, !dbg !43 + %425 = extractvalue { i32, i32, i32, i32 } %423, 1, !dbg !43 + %426 = extractvalue { i32, i32, i32, i32 } %423, 2, !dbg !43 + %427 = extractvalue { i32, i32, i32, i32 } %423, 3, !dbg !43 + %428 = getelementptr inbounds nuw i8, ptr addrspace(3) %416, i32 4096, !dbg !43 + %429 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %428), !dbg !43 + %430 = extractvalue { i32, i32, i32, i32 } %429, 0, !dbg !43 + %431 = extractvalue { i32, i32, i32, i32 } %429, 1, !dbg !43 + %432 = extractvalue { i32, i32, i32, i32 } %429, 2, !dbg !43 + %433 = extractvalue { i32, i32, i32, i32 } %429, 3, !dbg !43 + %434 = getelementptr inbounds nuw i8, ptr addrspace(3) %416, i32 4224, !dbg !43 + %435 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %434), !dbg !43 + %436 = extractvalue { i32, i32, i32, i32 } %435, 0, !dbg !43 + %437 = extractvalue { i32, i32, i32, i32 } %435, 1, !dbg !43 + %438 = extractvalue { i32, i32, i32, i32 } %435, 2, !dbg !43 + %439 = extractvalue { i32, i32, i32, i32 } %435, 3, !dbg !43 + %440 = getelementptr inbounds nuw i8, ptr addrspace(3) %416, i32 8192, !dbg !43 + %441 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %440), !dbg !43 + %442 = extractvalue { i32, i32, i32, i32 } %441, 0, !dbg !43 + %443 = extractvalue { i32, i32, i32, i32 } %441, 1, !dbg !43 + %444 = extractvalue { i32, i32, i32, i32 } %441, 2, !dbg !43 + %445 = extractvalue { i32, i32, i32, i32 } %441, 3, !dbg !43 + %446 = getelementptr inbounds nuw i8, ptr addrspace(3) %416, i32 8320, !dbg !43 + %447 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %446), !dbg !43 + %448 = extractvalue { i32, i32, i32, i32 } %447, 0, !dbg !43 + %449 = extractvalue { i32, i32, i32, i32 } %447, 1, !dbg !43 + %450 = extractvalue { i32, i32, i32, i32 } %447, 2, !dbg !43 + %451 = extractvalue { i32, i32, i32, i32 } %447, 3, !dbg !43 + %452 = getelementptr inbounds nuw i8, ptr addrspace(3) %416, i32 12288, !dbg !43 + %453 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %452), !dbg !43 + %454 = extractvalue { i32, i32, i32, i32 } %453, 0, !dbg !43 + %455 = extractvalue { i32, i32, i32, i32 } %453, 1, !dbg !43 + %456 = extractvalue { i32, i32, i32, i32 } %453, 2, !dbg !43 + %457 = extractvalue { i32, i32, i32, i32 } %453, 3, !dbg !43 + %458 = getelementptr inbounds nuw i8, ptr addrspace(3) %416, i32 12416, !dbg !43 + %459 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %458), !dbg !43 + %460 = extractvalue { i32, i32, i32, i32 } %459, 0, !dbg !43 + %461 = extractvalue { i32, i32, i32, i32 } %459, 1, !dbg !43 + %462 = extractvalue { i32, i32, i32, i32 } %459, 2, !dbg !43 + %463 = extractvalue { i32, i32, i32, i32 } %459, 3, !dbg !43 + %464 = xor i32 %415, 32, !dbg !43 + %465 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %464, !dbg !43 + %466 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %465), !dbg !43 + %467 = extractvalue { i32, i32, i32, i32 } %466, 0, !dbg !43 + %468 = extractvalue { i32, i32, i32, i32 } %466, 1, !dbg !43 + %469 = extractvalue { i32, i32, i32, i32 } %466, 2, !dbg !43 + %470 = extractvalue { i32, i32, i32, i32 } %466, 3, !dbg !43 + %471 = getelementptr inbounds nuw i8, ptr addrspace(3) %465, i32 128, !dbg !43 + %472 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %471), !dbg !43 + %473 = extractvalue { i32, i32, i32, i32 } %472, 0, !dbg !43 + %474 = extractvalue { i32, i32, i32, i32 } %472, 1, !dbg !43 + %475 = extractvalue { i32, i32, i32, i32 } %472, 2, !dbg !43 + %476 = extractvalue { i32, i32, i32, i32 } %472, 3, !dbg !43 + %477 = getelementptr inbounds nuw i8, ptr addrspace(3) %465, i32 4096, !dbg !43 + %478 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %477), !dbg !43 + %479 = extractvalue { i32, i32, i32, i32 } %478, 0, !dbg !43 + %480 = extractvalue { i32, i32, i32, i32 } %478, 1, !dbg !43 + %481 = extractvalue { i32, i32, i32, i32 } %478, 2, !dbg !43 + %482 = extractvalue { i32, i32, i32, i32 } %478, 3, !dbg !43 + %483 = getelementptr inbounds nuw i8, ptr addrspace(3) %465, i32 4224, !dbg !43 + %484 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %483), !dbg !43 + %485 = extractvalue { i32, i32, i32, i32 } %484, 0, !dbg !43 + %486 = extractvalue { i32, i32, i32, i32 } %484, 1, !dbg !43 + %487 = extractvalue { i32, i32, i32, i32 } %484, 2, !dbg !43 + %488 = extractvalue { i32, i32, i32, i32 } %484, 3, !dbg !43 + %489 = getelementptr inbounds nuw i8, ptr addrspace(3) %465, i32 8192, !dbg !43 + %490 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %489), !dbg !43 + %491 = extractvalue { i32, i32, i32, i32 } %490, 0, !dbg !43 + %492 = extractvalue { i32, i32, i32, i32 } %490, 1, !dbg !43 + %493 = extractvalue { i32, i32, i32, i32 } %490, 2, !dbg !43 + %494 = extractvalue { i32, i32, i32, i32 } %490, 3, !dbg !43 + %495 = getelementptr inbounds nuw i8, ptr addrspace(3) %465, i32 8320, !dbg !43 + %496 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %495), !dbg !43 + %497 = extractvalue { i32, i32, i32, i32 } %496, 0, !dbg !43 + %498 = extractvalue { i32, i32, i32, i32 } %496, 1, !dbg !43 + %499 = extractvalue { i32, i32, i32, i32 } %496, 2, !dbg !43 + %500 = extractvalue { i32, i32, i32, i32 } %496, 3, !dbg !43 + %501 = getelementptr inbounds nuw i8, ptr addrspace(3) %465, i32 12288, !dbg !43 + %502 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %501), !dbg !43 + %503 = extractvalue { i32, i32, i32, i32 } %502, 0, !dbg !43 + %504 = extractvalue { i32, i32, i32, i32 } %502, 1, !dbg !43 + %505 = extractvalue { i32, i32, i32, i32 } %502, 2, !dbg !43 + %506 = extractvalue { i32, i32, i32, i32 } %502, 3, !dbg !43 + %507 = getelementptr inbounds nuw i8, ptr addrspace(3) %465, i32 12416, !dbg !43 + %508 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %507), !dbg !43 + %509 = extractvalue { i32, i32, i32, i32 } %508, 0, !dbg !43 + %510 = extractvalue { i32, i32, i32, i32 } %508, 1, !dbg !43 + %511 = extractvalue { i32, i32, i32, i32 } %508, 2, !dbg !43 + %512 = extractvalue { i32, i32, i32, i32 } %508, 3, !dbg !43 + %513 = xor i32 %415, 64, !dbg !43 + %514 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %513, !dbg !43 + %515 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %514), !dbg !43 + %516 = extractvalue { i32, i32, i32, i32 } %515, 0, !dbg !43 + %517 = extractvalue { i32, i32, i32, i32 } %515, 1, !dbg !43 + %518 = extractvalue { i32, i32, i32, i32 } %515, 2, !dbg !43 + %519 = extractvalue { i32, i32, i32, i32 } %515, 3, !dbg !43 + %520 = getelementptr inbounds nuw i8, ptr addrspace(3) %514, i32 128, !dbg !43 + %521 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %520), !dbg !43 + %522 = extractvalue { i32, i32, i32, i32 } %521, 0, !dbg !43 + %523 = extractvalue { i32, i32, i32, i32 } %521, 1, !dbg !43 + %524 = extractvalue { i32, i32, i32, i32 } %521, 2, !dbg !43 + %525 = extractvalue { i32, i32, i32, i32 } %521, 3, !dbg !43 + %526 = getelementptr inbounds nuw i8, ptr addrspace(3) %514, i32 4096, !dbg !43 + %527 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %526), !dbg !43 + %528 = extractvalue { i32, i32, i32, i32 } %527, 0, !dbg !43 + %529 = extractvalue { i32, i32, i32, i32 } %527, 1, !dbg !43 + %530 = extractvalue { i32, i32, i32, i32 } %527, 2, !dbg !43 + %531 = extractvalue { i32, i32, i32, i32 } %527, 3, !dbg !43 + %532 = getelementptr inbounds nuw i8, ptr addrspace(3) %514, i32 4224, !dbg !43 + %533 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %532), !dbg !43 + %534 = extractvalue { i32, i32, i32, i32 } %533, 0, !dbg !43 + %535 = extractvalue { i32, i32, i32, i32 } %533, 1, !dbg !43 + %536 = extractvalue { i32, i32, i32, i32 } %533, 2, !dbg !43 + %537 = extractvalue { i32, i32, i32, i32 } %533, 3, !dbg !43 + %538 = getelementptr inbounds nuw i8, ptr addrspace(3) %514, i32 8192, !dbg !43 + %539 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %538), !dbg !43 + %540 = extractvalue { i32, i32, i32, i32 } %539, 0, !dbg !43 + %541 = extractvalue { i32, i32, i32, i32 } %539, 1, !dbg !43 + %542 = extractvalue { i32, i32, i32, i32 } %539, 2, !dbg !43 + %543 = extractvalue { i32, i32, i32, i32 } %539, 3, !dbg !43 + %544 = getelementptr inbounds nuw i8, ptr addrspace(3) %514, i32 8320, !dbg !43 + %545 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %544), !dbg !43 + %546 = extractvalue { i32, i32, i32, i32 } %545, 0, !dbg !43 + %547 = extractvalue { i32, i32, i32, i32 } %545, 1, !dbg !43 + %548 = extractvalue { i32, i32, i32, i32 } %545, 2, !dbg !43 + %549 = extractvalue { i32, i32, i32, i32 } %545, 3, !dbg !43 + %550 = getelementptr inbounds nuw i8, ptr addrspace(3) %514, i32 12288, !dbg !43 + %551 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %550), !dbg !43 + %552 = extractvalue { i32, i32, i32, i32 } %551, 0, !dbg !43 + %553 = extractvalue { i32, i32, i32, i32 } %551, 1, !dbg !43 + %554 = extractvalue { i32, i32, i32, i32 } %551, 2, !dbg !43 + %555 = extractvalue { i32, i32, i32, i32 } %551, 3, !dbg !43 + %556 = getelementptr inbounds nuw i8, ptr addrspace(3) %514, i32 12416, !dbg !43 + %557 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %556), !dbg !43 + %558 = extractvalue { i32, i32, i32, i32 } %557, 0, !dbg !43 + %559 = extractvalue { i32, i32, i32, i32 } %557, 1, !dbg !43 + %560 = extractvalue { i32, i32, i32, i32 } %557, 2, !dbg !43 + %561 = extractvalue { i32, i32, i32, i32 } %557, 3, !dbg !43 + %562 = xor i32 %415, 96, !dbg !43 + %563 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %562, !dbg !43 + %564 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %563), !dbg !43 + %565 = extractvalue { i32, i32, i32, i32 } %564, 0, !dbg !43 + %566 = extractvalue { i32, i32, i32, i32 } %564, 1, !dbg !43 + %567 = extractvalue { i32, i32, i32, i32 } %564, 2, !dbg !43 + %568 = extractvalue { i32, i32, i32, i32 } %564, 3, !dbg !43 + %569 = getelementptr inbounds nuw i8, ptr addrspace(3) %563, i32 128, !dbg !43 + %570 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %569), !dbg !43 + %571 = extractvalue { i32, i32, i32, i32 } %570, 0, !dbg !43 + %572 = extractvalue { i32, i32, i32, i32 } %570, 1, !dbg !43 + %573 = extractvalue { i32, i32, i32, i32 } %570, 2, !dbg !43 + %574 = extractvalue { i32, i32, i32, i32 } %570, 3, !dbg !43 + %575 = getelementptr inbounds nuw i8, ptr addrspace(3) %563, i32 4096, !dbg !43 + %576 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %575), !dbg !43 + %577 = extractvalue { i32, i32, i32, i32 } %576, 0, !dbg !43 + %578 = extractvalue { i32, i32, i32, i32 } %576, 1, !dbg !43 + %579 = extractvalue { i32, i32, i32, i32 } %576, 2, !dbg !43 + %580 = extractvalue { i32, i32, i32, i32 } %576, 3, !dbg !43 + %581 = getelementptr inbounds nuw i8, ptr addrspace(3) %563, i32 4224, !dbg !43 + %582 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %581), !dbg !43 + %583 = extractvalue { i32, i32, i32, i32 } %582, 0, !dbg !43 + %584 = extractvalue { i32, i32, i32, i32 } %582, 1, !dbg !43 + %585 = extractvalue { i32, i32, i32, i32 } %582, 2, !dbg !43 + %586 = extractvalue { i32, i32, i32, i32 } %582, 3, !dbg !43 + %587 = getelementptr inbounds nuw i8, ptr addrspace(3) %563, i32 8192, !dbg !43 + %588 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %587), !dbg !43 + %589 = extractvalue { i32, i32, i32, i32 } %588, 0, !dbg !43 + %590 = extractvalue { i32, i32, i32, i32 } %588, 1, !dbg !43 + %591 = extractvalue { i32, i32, i32, i32 } %588, 2, !dbg !43 + %592 = extractvalue { i32, i32, i32, i32 } %588, 3, !dbg !43 + %593 = getelementptr inbounds nuw i8, ptr addrspace(3) %563, i32 8320, !dbg !43 + %594 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %593), !dbg !43 + %595 = extractvalue { i32, i32, i32, i32 } %594, 0, !dbg !43 + %596 = extractvalue { i32, i32, i32, i32 } %594, 1, !dbg !43 + %597 = extractvalue { i32, i32, i32, i32 } %594, 2, !dbg !43 + %598 = extractvalue { i32, i32, i32, i32 } %594, 3, !dbg !43 + %599 = getelementptr inbounds nuw i8, ptr addrspace(3) %563, i32 12288, !dbg !43 + %600 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %599), !dbg !43 + %601 = extractvalue { i32, i32, i32, i32 } %600, 0, !dbg !43 + %602 = extractvalue { i32, i32, i32, i32 } %600, 1, !dbg !43 + %603 = extractvalue { i32, i32, i32, i32 } %600, 2, !dbg !43 + %604 = extractvalue { i32, i32, i32, i32 } %600, 3, !dbg !43 + %605 = getelementptr inbounds nuw i8, ptr addrspace(3) %563, i32 12416, !dbg !43 + %606 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %605), !dbg !43 + %607 = extractvalue { i32, i32, i32, i32 } %606, 0, !dbg !43 + %608 = extractvalue { i32, i32, i32, i32 } %606, 1, !dbg !43 + %609 = extractvalue { i32, i32, i32, i32 } %606, 2, !dbg !43 + %610 = extractvalue { i32, i32, i32, i32 } %606, 3, !dbg !43 + %611 = shl nuw nsw i32 %51, 8, !dbg !44 + %612 = shl nuw nsw i32 %53, 1, !dbg !44 + %613 = shl nuw nsw i32 %28, 6, !dbg !44 + %614 = or disjoint i32 %611, %52, !dbg !44 + %615 = xor i32 %614, %612, !dbg !44 + %616 = or disjoint i32 %615, %613, !dbg !44 + %617 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %616, !dbg !44 + %618 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %617), !dbg !44 + %619 = extractvalue { i32, i32, i32, i32 } %618, 0, !dbg !44 + %620 = extractvalue { i32, i32, i32, i32 } %618, 1, !dbg !44 + %621 = extractvalue { i32, i32, i32, i32 } %618, 2, !dbg !44 + %622 = extractvalue { i32, i32, i32, i32 } %618, 3, !dbg !44 + %623 = getelementptr inbounds nuw i8, ptr addrspace(3) %617, i32 128, !dbg !44 + %624 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %623), !dbg !44 + %625 = extractvalue { i32, i32, i32, i32 } %624, 0, !dbg !44 + %626 = extractvalue { i32, i32, i32, i32 } %624, 1, !dbg !44 + %627 = extractvalue { i32, i32, i32, i32 } %624, 2, !dbg !44 + %628 = extractvalue { i32, i32, i32, i32 } %624, 3, !dbg !44 + %629 = getelementptr inbounds nuw i8, ptr addrspace(3) %617, i32 4096, !dbg !44 + %630 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %629), !dbg !44 + %631 = extractvalue { i32, i32, i32, i32 } %630, 0, !dbg !44 + %632 = extractvalue { i32, i32, i32, i32 } %630, 1, !dbg !44 + %633 = extractvalue { i32, i32, i32, i32 } %630, 2, !dbg !44 + %634 = extractvalue { i32, i32, i32, i32 } %630, 3, !dbg !44 + %635 = getelementptr inbounds nuw i8, ptr addrspace(3) %617, i32 4224, !dbg !44 + %636 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %635), !dbg !44 + %637 = extractvalue { i32, i32, i32, i32 } %636, 0, !dbg !44 + %638 = extractvalue { i32, i32, i32, i32 } %636, 1, !dbg !44 + %639 = extractvalue { i32, i32, i32, i32 } %636, 2, !dbg !44 + %640 = extractvalue { i32, i32, i32, i32 } %636, 3, !dbg !44 + %641 = getelementptr inbounds nuw i8, ptr addrspace(3) %617, i32 8192, !dbg !44 + %642 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %641), !dbg !44 + %643 = extractvalue { i32, i32, i32, i32 } %642, 0, !dbg !44 + %644 = extractvalue { i32, i32, i32, i32 } %642, 1, !dbg !44 + %645 = extractvalue { i32, i32, i32, i32 } %642, 2, !dbg !44 + %646 = extractvalue { i32, i32, i32, i32 } %642, 3, !dbg !44 + %647 = getelementptr inbounds nuw i8, ptr addrspace(3) %617, i32 8320, !dbg !44 + %648 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %647), !dbg !44 + %649 = extractvalue { i32, i32, i32, i32 } %648, 0, !dbg !44 + %650 = extractvalue { i32, i32, i32, i32 } %648, 1, !dbg !44 + %651 = extractvalue { i32, i32, i32, i32 } %648, 2, !dbg !44 + %652 = extractvalue { i32, i32, i32, i32 } %648, 3, !dbg !44 + %653 = getelementptr inbounds nuw i8, ptr addrspace(3) %617, i32 12288, !dbg !44 + %654 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %653), !dbg !44 + %655 = extractvalue { i32, i32, i32, i32 } %654, 0, !dbg !44 + %656 = extractvalue { i32, i32, i32, i32 } %654, 1, !dbg !44 + %657 = extractvalue { i32, i32, i32, i32 } %654, 2, !dbg !44 + %658 = extractvalue { i32, i32, i32, i32 } %654, 3, !dbg !44 + %659 = getelementptr inbounds nuw i8, ptr addrspace(3) %617, i32 12416, !dbg !44 + %660 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %659), !dbg !44 + %661 = extractvalue { i32, i32, i32, i32 } %660, 0, !dbg !44 + %662 = extractvalue { i32, i32, i32, i32 } %660, 1, !dbg !44 + %663 = extractvalue { i32, i32, i32, i32 } %660, 2, !dbg !44 + %664 = extractvalue { i32, i32, i32, i32 } %660, 3, !dbg !44 + %665 = xor i32 %616, 64, !dbg !44 + %666 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %665, !dbg !44 + %667 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %666), !dbg !44 + %668 = extractvalue { i32, i32, i32, i32 } %667, 0, !dbg !44 + %669 = extractvalue { i32, i32, i32, i32 } %667, 1, !dbg !44 + %670 = extractvalue { i32, i32, i32, i32 } %667, 2, !dbg !44 + %671 = extractvalue { i32, i32, i32, i32 } %667, 3, !dbg !44 + %672 = getelementptr inbounds nuw i8, ptr addrspace(3) %666, i32 128, !dbg !44 + %673 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %672), !dbg !44 + %674 = extractvalue { i32, i32, i32, i32 } %673, 0, !dbg !44 + %675 = extractvalue { i32, i32, i32, i32 } %673, 1, !dbg !44 + %676 = extractvalue { i32, i32, i32, i32 } %673, 2, !dbg !44 + %677 = extractvalue { i32, i32, i32, i32 } %673, 3, !dbg !44 + %678 = getelementptr inbounds nuw i8, ptr addrspace(3) %666, i32 4096, !dbg !44 + %679 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %678), !dbg !44 + %680 = extractvalue { i32, i32, i32, i32 } %679, 0, !dbg !44 + %681 = extractvalue { i32, i32, i32, i32 } %679, 1, !dbg !44 + %682 = extractvalue { i32, i32, i32, i32 } %679, 2, !dbg !44 + %683 = extractvalue { i32, i32, i32, i32 } %679, 3, !dbg !44 + %684 = getelementptr inbounds nuw i8, ptr addrspace(3) %666, i32 4224, !dbg !44 + %685 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %684), !dbg !44 + %686 = extractvalue { i32, i32, i32, i32 } %685, 0, !dbg !44 + %687 = extractvalue { i32, i32, i32, i32 } %685, 1, !dbg !44 + %688 = extractvalue { i32, i32, i32, i32 } %685, 2, !dbg !44 + %689 = extractvalue { i32, i32, i32, i32 } %685, 3, !dbg !44 + %690 = getelementptr inbounds nuw i8, ptr addrspace(3) %666, i32 8192, !dbg !44 + %691 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %690), !dbg !44 + %692 = extractvalue { i32, i32, i32, i32 } %691, 0, !dbg !44 + %693 = extractvalue { i32, i32, i32, i32 } %691, 1, !dbg !44 + %694 = extractvalue { i32, i32, i32, i32 } %691, 2, !dbg !44 + %695 = extractvalue { i32, i32, i32, i32 } %691, 3, !dbg !44 + %696 = getelementptr inbounds nuw i8, ptr addrspace(3) %666, i32 8320, !dbg !44 + %697 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %696), !dbg !44 + %698 = extractvalue { i32, i32, i32, i32 } %697, 0, !dbg !44 + %699 = extractvalue { i32, i32, i32, i32 } %697, 1, !dbg !44 + %700 = extractvalue { i32, i32, i32, i32 } %697, 2, !dbg !44 + %701 = extractvalue { i32, i32, i32, i32 } %697, 3, !dbg !44 + %702 = getelementptr inbounds nuw i8, ptr addrspace(3) %666, i32 12288, !dbg !44 + %703 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %702), !dbg !44 + %704 = extractvalue { i32, i32, i32, i32 } %703, 0, !dbg !44 + %705 = extractvalue { i32, i32, i32, i32 } %703, 1, !dbg !44 + %706 = extractvalue { i32, i32, i32, i32 } %703, 2, !dbg !44 + %707 = extractvalue { i32, i32, i32, i32 } %703, 3, !dbg !44 + %708 = getelementptr inbounds nuw i8, ptr addrspace(3) %666, i32 12416, !dbg !44 + %709 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %708), !dbg !44 + %710 = extractvalue { i32, i32, i32, i32 } %709, 0, !dbg !44 + %711 = extractvalue { i32, i32, i32, i32 } %709, 1, !dbg !44 + %712 = extractvalue { i32, i32, i32, i32 } %709, 2, !dbg !44 + %713 = extractvalue { i32, i32, i32, i32 } %709, 3, !dbg !44 + %714 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %418, i32 %419, i32 %420, i32 %421, i32 %619, i32 %620) #7, !dbg !45 + %715 = extractvalue { float, float, float, float } %714, 0, !dbg !45 + %716 = extractvalue { float, float, float, float } %714, 1, !dbg !45 + %717 = extractvalue { float, float, float, float } %714, 2, !dbg !45 + %718 = extractvalue { float, float, float, float } %714, 3, !dbg !45 + %719 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %418, i32 %419, i32 %420, i32 %421, i32 %631, i32 %632) #7, !dbg !45 + %720 = extractvalue { float, float, float, float } %719, 0, !dbg !45 + %721 = extractvalue { float, float, float, float } %719, 1, !dbg !45 + %722 = extractvalue { float, float, float, float } %719, 2, !dbg !45 + %723 = extractvalue { float, float, float, float } %719, 3, !dbg !45 + %724 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %418, i32 %419, i32 %420, i32 %421, i32 %643, i32 %644) #7, !dbg !45 + %725 = extractvalue { float, float, float, float } %724, 0, !dbg !45 + %726 = extractvalue { float, float, float, float } %724, 1, !dbg !45 + %727 = extractvalue { float, float, float, float } %724, 2, !dbg !45 + %728 = extractvalue { float, float, float, float } %724, 3, !dbg !45 + %729 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %418, i32 %419, i32 %420, i32 %421, i32 %655, i32 %656) #7, !dbg !45 + %730 = extractvalue { float, float, float, float } %729, 0, !dbg !45 + %731 = extractvalue { float, float, float, float } %729, 1, !dbg !45 + %732 = extractvalue { float, float, float, float } %729, 2, !dbg !45 + %733 = extractvalue { float, float, float, float } %729, 3, !dbg !45 + %734 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %430, i32 %431, i32 %432, i32 %433, i32 %619, i32 %620) #7, !dbg !45 + %735 = extractvalue { float, float, float, float } %734, 0, !dbg !45 + %736 = extractvalue { float, float, float, float } %734, 1, !dbg !45 + %737 = extractvalue { float, float, float, float } %734, 2, !dbg !45 + %738 = extractvalue { float, float, float, float } %734, 3, !dbg !45 + %739 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %430, i32 %431, i32 %432, i32 %433, i32 %631, i32 %632) #7, !dbg !45 + %740 = extractvalue { float, float, float, float } %739, 0, !dbg !45 + %741 = extractvalue { float, float, float, float } %739, 1, !dbg !45 + %742 = extractvalue { float, float, float, float } %739, 2, !dbg !45 + %743 = extractvalue { float, float, float, float } %739, 3, !dbg !45 + %744 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %430, i32 %431, i32 %432, i32 %433, i32 %643, i32 %644) #7, !dbg !45 + %745 = extractvalue { float, float, float, float } %744, 0, !dbg !45 + %746 = extractvalue { float, float, float, float } %744, 1, !dbg !45 + %747 = extractvalue { float, float, float, float } %744, 2, !dbg !45 + %748 = extractvalue { float, float, float, float } %744, 3, !dbg !45 + %749 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %430, i32 %431, i32 %432, i32 %433, i32 %655, i32 %656) #7, !dbg !45 + %750 = extractvalue { float, float, float, float } %749, 0, !dbg !45 + %751 = extractvalue { float, float, float, float } %749, 1, !dbg !45 + %752 = extractvalue { float, float, float, float } %749, 2, !dbg !45 + %753 = extractvalue { float, float, float, float } %749, 3, !dbg !45 + %754 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %442, i32 %443, i32 %444, i32 %445, i32 %619, i32 %620) #7, !dbg !45 + %755 = extractvalue { float, float, float, float } %754, 0, !dbg !45 + %756 = extractvalue { float, float, float, float } %754, 1, !dbg !45 + %757 = extractvalue { float, float, float, float } %754, 2, !dbg !45 + %758 = extractvalue { float, float, float, float } %754, 3, !dbg !45 + %759 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %442, i32 %443, i32 %444, i32 %445, i32 %631, i32 %632) #7, !dbg !45 + %760 = extractvalue { float, float, float, float } %759, 0, !dbg !45 + %761 = extractvalue { float, float, float, float } %759, 1, !dbg !45 + %762 = extractvalue { float, float, float, float } %759, 2, !dbg !45 + %763 = extractvalue { float, float, float, float } %759, 3, !dbg !45 + %764 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %442, i32 %443, i32 %444, i32 %445, i32 %643, i32 %644) #7, !dbg !45 + %765 = extractvalue { float, float, float, float } %764, 0, !dbg !45 + %766 = extractvalue { float, float, float, float } %764, 1, !dbg !45 + %767 = extractvalue { float, float, float, float } %764, 2, !dbg !45 + %768 = extractvalue { float, float, float, float } %764, 3, !dbg !45 + %769 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %442, i32 %443, i32 %444, i32 %445, i32 %655, i32 %656) #7, !dbg !45 + %770 = extractvalue { float, float, float, float } %769, 0, !dbg !45 + %771 = extractvalue { float, float, float, float } %769, 1, !dbg !45 + %772 = extractvalue { float, float, float, float } %769, 2, !dbg !45 + %773 = extractvalue { float, float, float, float } %769, 3, !dbg !45 + %774 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %454, i32 %455, i32 %456, i32 %457, i32 %619, i32 %620) #7, !dbg !45 + %775 = extractvalue { float, float, float, float } %774, 0, !dbg !45 + %776 = extractvalue { float, float, float, float } %774, 1, !dbg !45 + %777 = extractvalue { float, float, float, float } %774, 2, !dbg !45 + %778 = extractvalue { float, float, float, float } %774, 3, !dbg !45 + %779 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %454, i32 %455, i32 %456, i32 %457, i32 %631, i32 %632) #7, !dbg !45 + %780 = extractvalue { float, float, float, float } %779, 0, !dbg !45 + %781 = extractvalue { float, float, float, float } %779, 1, !dbg !45 + %782 = extractvalue { float, float, float, float } %779, 2, !dbg !45 + %783 = extractvalue { float, float, float, float } %779, 3, !dbg !45 + %784 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %454, i32 %455, i32 %456, i32 %457, i32 %643, i32 %644) #7, !dbg !45 + %785 = extractvalue { float, float, float, float } %784, 0, !dbg !45 + %786 = extractvalue { float, float, float, float } %784, 1, !dbg !45 + %787 = extractvalue { float, float, float, float } %784, 2, !dbg !45 + %788 = extractvalue { float, float, float, float } %784, 3, !dbg !45 + %789 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %454, i32 %455, i32 %456, i32 %457, i32 %655, i32 %656) #7, !dbg !45 + %790 = extractvalue { float, float, float, float } %789, 0, !dbg !45 + %791 = extractvalue { float, float, float, float } %789, 1, !dbg !45 + %792 = extractvalue { float, float, float, float } %789, 2, !dbg !45 + %793 = extractvalue { float, float, float, float } %789, 3, !dbg !45 + %794 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %715, float %716, float %717, float %718, i32 %467, i32 %468, i32 %469, i32 %470, i32 %621, i32 %622) #7, !dbg !45 + %795 = extractvalue { float, float, float, float } %794, 0, !dbg !45 + %796 = extractvalue { float, float, float, float } %794, 1, !dbg !45 + %797 = extractvalue { float, float, float, float } %794, 2, !dbg !45 + %798 = extractvalue { float, float, float, float } %794, 3, !dbg !45 + %799 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %720, float %721, float %722, float %723, i32 %467, i32 %468, i32 %469, i32 %470, i32 %633, i32 %634) #7, !dbg !45 + %800 = extractvalue { float, float, float, float } %799, 0, !dbg !45 + %801 = extractvalue { float, float, float, float } %799, 1, !dbg !45 + %802 = extractvalue { float, float, float, float } %799, 2, !dbg !45 + %803 = extractvalue { float, float, float, float } %799, 3, !dbg !45 + %804 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %725, float %726, float %727, float %728, i32 %467, i32 %468, i32 %469, i32 %470, i32 %645, i32 %646) #7, !dbg !45 + %805 = extractvalue { float, float, float, float } %804, 0, !dbg !45 + %806 = extractvalue { float, float, float, float } %804, 1, !dbg !45 + %807 = extractvalue { float, float, float, float } %804, 2, !dbg !45 + %808 = extractvalue { float, float, float, float } %804, 3, !dbg !45 + %809 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %730, float %731, float %732, float %733, i32 %467, i32 %468, i32 %469, i32 %470, i32 %657, i32 %658) #7, !dbg !45 + %810 = extractvalue { float, float, float, float } %809, 0, !dbg !45 + %811 = extractvalue { float, float, float, float } %809, 1, !dbg !45 + %812 = extractvalue { float, float, float, float } %809, 2, !dbg !45 + %813 = extractvalue { float, float, float, float } %809, 3, !dbg !45 + %814 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %735, float %736, float %737, float %738, i32 %479, i32 %480, i32 %481, i32 %482, i32 %621, i32 %622) #7, !dbg !45 + %815 = extractvalue { float, float, float, float } %814, 0, !dbg !45 + %816 = extractvalue { float, float, float, float } %814, 1, !dbg !45 + %817 = extractvalue { float, float, float, float } %814, 2, !dbg !45 + %818 = extractvalue { float, float, float, float } %814, 3, !dbg !45 + %819 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %740, float %741, float %742, float %743, i32 %479, i32 %480, i32 %481, i32 %482, i32 %633, i32 %634) #7, !dbg !45 + %820 = extractvalue { float, float, float, float } %819, 0, !dbg !45 + %821 = extractvalue { float, float, float, float } %819, 1, !dbg !45 + %822 = extractvalue { float, float, float, float } %819, 2, !dbg !45 + %823 = extractvalue { float, float, float, float } %819, 3, !dbg !45 + %824 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %745, float %746, float %747, float %748, i32 %479, i32 %480, i32 %481, i32 %482, i32 %645, i32 %646) #7, !dbg !45 + %825 = extractvalue { float, float, float, float } %824, 0, !dbg !45 + %826 = extractvalue { float, float, float, float } %824, 1, !dbg !45 + %827 = extractvalue { float, float, float, float } %824, 2, !dbg !45 + %828 = extractvalue { float, float, float, float } %824, 3, !dbg !45 + %829 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %750, float %751, float %752, float %753, i32 %479, i32 %480, i32 %481, i32 %482, i32 %657, i32 %658) #7, !dbg !45 + %830 = extractvalue { float, float, float, float } %829, 0, !dbg !45 + %831 = extractvalue { float, float, float, float } %829, 1, !dbg !45 + %832 = extractvalue { float, float, float, float } %829, 2, !dbg !45 + %833 = extractvalue { float, float, float, float } %829, 3, !dbg !45 + %834 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %755, float %756, float %757, float %758, i32 %491, i32 %492, i32 %493, i32 %494, i32 %621, i32 %622) #7, !dbg !45 + %835 = extractvalue { float, float, float, float } %834, 0, !dbg !45 + %836 = extractvalue { float, float, float, float } %834, 1, !dbg !45 + %837 = extractvalue { float, float, float, float } %834, 2, !dbg !45 + %838 = extractvalue { float, float, float, float } %834, 3, !dbg !45 + %839 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %760, float %761, float %762, float %763, i32 %491, i32 %492, i32 %493, i32 %494, i32 %633, i32 %634) #7, !dbg !45 + %840 = extractvalue { float, float, float, float } %839, 0, !dbg !45 + %841 = extractvalue { float, float, float, float } %839, 1, !dbg !45 + %842 = extractvalue { float, float, float, float } %839, 2, !dbg !45 + %843 = extractvalue { float, float, float, float } %839, 3, !dbg !45 + %844 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %765, float %766, float %767, float %768, i32 %491, i32 %492, i32 %493, i32 %494, i32 %645, i32 %646) #7, !dbg !45 + %845 = extractvalue { float, float, float, float } %844, 0, !dbg !45 + %846 = extractvalue { float, float, float, float } %844, 1, !dbg !45 + %847 = extractvalue { float, float, float, float } %844, 2, !dbg !45 + %848 = extractvalue { float, float, float, float } %844, 3, !dbg !45 + %849 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %770, float %771, float %772, float %773, i32 %491, i32 %492, i32 %493, i32 %494, i32 %657, i32 %658) #7, !dbg !45 + %850 = extractvalue { float, float, float, float } %849, 0, !dbg !45 + %851 = extractvalue { float, float, float, float } %849, 1, !dbg !45 + %852 = extractvalue { float, float, float, float } %849, 2, !dbg !45 + %853 = extractvalue { float, float, float, float } %849, 3, !dbg !45 + %854 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %775, float %776, float %777, float %778, i32 %503, i32 %504, i32 %505, i32 %506, i32 %621, i32 %622) #7, !dbg !45 + %855 = extractvalue { float, float, float, float } %854, 0, !dbg !45 + %856 = extractvalue { float, float, float, float } %854, 1, !dbg !45 + %857 = extractvalue { float, float, float, float } %854, 2, !dbg !45 + %858 = extractvalue { float, float, float, float } %854, 3, !dbg !45 + %859 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %780, float %781, float %782, float %783, i32 %503, i32 %504, i32 %505, i32 %506, i32 %633, i32 %634) #7, !dbg !45 + %860 = extractvalue { float, float, float, float } %859, 0, !dbg !45 + %861 = extractvalue { float, float, float, float } %859, 1, !dbg !45 + %862 = extractvalue { float, float, float, float } %859, 2, !dbg !45 + %863 = extractvalue { float, float, float, float } %859, 3, !dbg !45 + %864 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %785, float %786, float %787, float %788, i32 %503, i32 %504, i32 %505, i32 %506, i32 %645, i32 %646) #7, !dbg !45 + %865 = extractvalue { float, float, float, float } %864, 0, !dbg !45 + %866 = extractvalue { float, float, float, float } %864, 1, !dbg !45 + %867 = extractvalue { float, float, float, float } %864, 2, !dbg !45 + %868 = extractvalue { float, float, float, float } %864, 3, !dbg !45 + %869 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %790, float %791, float %792, float %793, i32 %503, i32 %504, i32 %505, i32 %506, i32 %657, i32 %658) #7, !dbg !45 + %870 = extractvalue { float, float, float, float } %869, 0, !dbg !45 + %871 = extractvalue { float, float, float, float } %869, 1, !dbg !45 + %872 = extractvalue { float, float, float, float } %869, 2, !dbg !45 + %873 = extractvalue { float, float, float, float } %869, 3, !dbg !45 + %874 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %795, float %796, float %797, float %798, i32 %516, i32 %517, i32 %518, i32 %519, i32 %668, i32 %669) #7, !dbg !45 + %875 = extractvalue { float, float, float, float } %874, 0, !dbg !45 + %876 = extractvalue { float, float, float, float } %874, 1, !dbg !45 + %877 = extractvalue { float, float, float, float } %874, 2, !dbg !45 + %878 = extractvalue { float, float, float, float } %874, 3, !dbg !45 + %879 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %800, float %801, float %802, float %803, i32 %516, i32 %517, i32 %518, i32 %519, i32 %680, i32 %681) #7, !dbg !45 + %880 = extractvalue { float, float, float, float } %879, 0, !dbg !45 + %881 = extractvalue { float, float, float, float } %879, 1, !dbg !45 + %882 = extractvalue { float, float, float, float } %879, 2, !dbg !45 + %883 = extractvalue { float, float, float, float } %879, 3, !dbg !45 + %884 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %805, float %806, float %807, float %808, i32 %516, i32 %517, i32 %518, i32 %519, i32 %692, i32 %693) #7, !dbg !45 + %885 = extractvalue { float, float, float, float } %884, 0, !dbg !45 + %886 = extractvalue { float, float, float, float } %884, 1, !dbg !45 + %887 = extractvalue { float, float, float, float } %884, 2, !dbg !45 + %888 = extractvalue { float, float, float, float } %884, 3, !dbg !45 + %889 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %810, float %811, float %812, float %813, i32 %516, i32 %517, i32 %518, i32 %519, i32 %704, i32 %705) #7, !dbg !45 + %890 = extractvalue { float, float, float, float } %889, 0, !dbg !45 + %891 = extractvalue { float, float, float, float } %889, 1, !dbg !45 + %892 = extractvalue { float, float, float, float } %889, 2, !dbg !45 + %893 = extractvalue { float, float, float, float } %889, 3, !dbg !45 + %894 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %815, float %816, float %817, float %818, i32 %528, i32 %529, i32 %530, i32 %531, i32 %668, i32 %669) #7, !dbg !45 + %895 = extractvalue { float, float, float, float } %894, 0, !dbg !45 + %896 = extractvalue { float, float, float, float } %894, 1, !dbg !45 + %897 = extractvalue { float, float, float, float } %894, 2, !dbg !45 + %898 = extractvalue { float, float, float, float } %894, 3, !dbg !45 + %899 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %820, float %821, float %822, float %823, i32 %528, i32 %529, i32 %530, i32 %531, i32 %680, i32 %681) #7, !dbg !45 + %900 = extractvalue { float, float, float, float } %899, 0, !dbg !45 + %901 = extractvalue { float, float, float, float } %899, 1, !dbg !45 + %902 = extractvalue { float, float, float, float } %899, 2, !dbg !45 + %903 = extractvalue { float, float, float, float } %899, 3, !dbg !45 + %904 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %825, float %826, float %827, float %828, i32 %528, i32 %529, i32 %530, i32 %531, i32 %692, i32 %693) #7, !dbg !45 + %905 = extractvalue { float, float, float, float } %904, 0, !dbg !45 + %906 = extractvalue { float, float, float, float } %904, 1, !dbg !45 + %907 = extractvalue { float, float, float, float } %904, 2, !dbg !45 + %908 = extractvalue { float, float, float, float } %904, 3, !dbg !45 + %909 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %830, float %831, float %832, float %833, i32 %528, i32 %529, i32 %530, i32 %531, i32 %704, i32 %705) #7, !dbg !45 + %910 = extractvalue { float, float, float, float } %909, 0, !dbg !45 + %911 = extractvalue { float, float, float, float } %909, 1, !dbg !45 + %912 = extractvalue { float, float, float, float } %909, 2, !dbg !45 + %913 = extractvalue { float, float, float, float } %909, 3, !dbg !45 + %914 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %835, float %836, float %837, float %838, i32 %540, i32 %541, i32 %542, i32 %543, i32 %668, i32 %669) #7, !dbg !45 + %915 = extractvalue { float, float, float, float } %914, 0, !dbg !45 + %916 = extractvalue { float, float, float, float } %914, 1, !dbg !45 + %917 = extractvalue { float, float, float, float } %914, 2, !dbg !45 + %918 = extractvalue { float, float, float, float } %914, 3, !dbg !45 + %919 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %840, float %841, float %842, float %843, i32 %540, i32 %541, i32 %542, i32 %543, i32 %680, i32 %681) #7, !dbg !45 + %920 = extractvalue { float, float, float, float } %919, 0, !dbg !45 + %921 = extractvalue { float, float, float, float } %919, 1, !dbg !45 + %922 = extractvalue { float, float, float, float } %919, 2, !dbg !45 + %923 = extractvalue { float, float, float, float } %919, 3, !dbg !45 + %924 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %845, float %846, float %847, float %848, i32 %540, i32 %541, i32 %542, i32 %543, i32 %692, i32 %693) #7, !dbg !45 + %925 = extractvalue { float, float, float, float } %924, 0, !dbg !45 + %926 = extractvalue { float, float, float, float } %924, 1, !dbg !45 + %927 = extractvalue { float, float, float, float } %924, 2, !dbg !45 + %928 = extractvalue { float, float, float, float } %924, 3, !dbg !45 + %929 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %850, float %851, float %852, float %853, i32 %540, i32 %541, i32 %542, i32 %543, i32 %704, i32 %705) #7, !dbg !45 + %930 = extractvalue { float, float, float, float } %929, 0, !dbg !45 + %931 = extractvalue { float, float, float, float } %929, 1, !dbg !45 + %932 = extractvalue { float, float, float, float } %929, 2, !dbg !45 + %933 = extractvalue { float, float, float, float } %929, 3, !dbg !45 + %934 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %855, float %856, float %857, float %858, i32 %552, i32 %553, i32 %554, i32 %555, i32 %668, i32 %669) #7, !dbg !45 + %935 = extractvalue { float, float, float, float } %934, 0, !dbg !45 + %936 = extractvalue { float, float, float, float } %934, 1, !dbg !45 + %937 = extractvalue { float, float, float, float } %934, 2, !dbg !45 + %938 = extractvalue { float, float, float, float } %934, 3, !dbg !45 + %939 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %860, float %861, float %862, float %863, i32 %552, i32 %553, i32 %554, i32 %555, i32 %680, i32 %681) #7, !dbg !45 + %940 = extractvalue { float, float, float, float } %939, 0, !dbg !45 + %941 = extractvalue { float, float, float, float } %939, 1, !dbg !45 + %942 = extractvalue { float, float, float, float } %939, 2, !dbg !45 + %943 = extractvalue { float, float, float, float } %939, 3, !dbg !45 + %944 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %865, float %866, float %867, float %868, i32 %552, i32 %553, i32 %554, i32 %555, i32 %692, i32 %693) #7, !dbg !45 + %945 = extractvalue { float, float, float, float } %944, 0, !dbg !45 + %946 = extractvalue { float, float, float, float } %944, 1, !dbg !45 + %947 = extractvalue { float, float, float, float } %944, 2, !dbg !45 + %948 = extractvalue { float, float, float, float } %944, 3, !dbg !45 + %949 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %870, float %871, float %872, float %873, i32 %552, i32 %553, i32 %554, i32 %555, i32 %704, i32 %705) #7, !dbg !45 + %950 = extractvalue { float, float, float, float } %949, 0, !dbg !45 + %951 = extractvalue { float, float, float, float } %949, 1, !dbg !45 + %952 = extractvalue { float, float, float, float } %949, 2, !dbg !45 + %953 = extractvalue { float, float, float, float } %949, 3, !dbg !45 + %954 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %875, float %876, float %877, float %878, i32 %565, i32 %566, i32 %567, i32 %568, i32 %670, i32 %671) #7, !dbg !45 + %955 = extractvalue { float, float, float, float } %954, 0, !dbg !45 + %956 = extractvalue { float, float, float, float } %954, 1, !dbg !45 + %957 = extractvalue { float, float, float, float } %954, 2, !dbg !45 + %958 = extractvalue { float, float, float, float } %954, 3, !dbg !45 + %959 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %880, float %881, float %882, float %883, i32 %565, i32 %566, i32 %567, i32 %568, i32 %682, i32 %683) #7, !dbg !45 + %960 = extractvalue { float, float, float, float } %959, 0, !dbg !45 + %961 = extractvalue { float, float, float, float } %959, 1, !dbg !45 + %962 = extractvalue { float, float, float, float } %959, 2, !dbg !45 + %963 = extractvalue { float, float, float, float } %959, 3, !dbg !45 + %964 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %885, float %886, float %887, float %888, i32 %565, i32 %566, i32 %567, i32 %568, i32 %694, i32 %695) #7, !dbg !45 + %965 = extractvalue { float, float, float, float } %964, 0, !dbg !45 + %966 = extractvalue { float, float, float, float } %964, 1, !dbg !45 + %967 = extractvalue { float, float, float, float } %964, 2, !dbg !45 + %968 = extractvalue { float, float, float, float } %964, 3, !dbg !45 + %969 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %890, float %891, float %892, float %893, i32 %565, i32 %566, i32 %567, i32 %568, i32 %706, i32 %707) #7, !dbg !45 + %970 = extractvalue { float, float, float, float } %969, 0, !dbg !45 + %971 = extractvalue { float, float, float, float } %969, 1, !dbg !45 + %972 = extractvalue { float, float, float, float } %969, 2, !dbg !45 + %973 = extractvalue { float, float, float, float } %969, 3, !dbg !45 + %974 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %895, float %896, float %897, float %898, i32 %577, i32 %578, i32 %579, i32 %580, i32 %670, i32 %671) #7, !dbg !45 + %975 = extractvalue { float, float, float, float } %974, 0, !dbg !45 + %976 = extractvalue { float, float, float, float } %974, 1, !dbg !45 + %977 = extractvalue { float, float, float, float } %974, 2, !dbg !45 + %978 = extractvalue { float, float, float, float } %974, 3, !dbg !45 + %979 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %900, float %901, float %902, float %903, i32 %577, i32 %578, i32 %579, i32 %580, i32 %682, i32 %683) #7, !dbg !45 + %980 = extractvalue { float, float, float, float } %979, 0, !dbg !45 + %981 = extractvalue { float, float, float, float } %979, 1, !dbg !45 + %982 = extractvalue { float, float, float, float } %979, 2, !dbg !45 + %983 = extractvalue { float, float, float, float } %979, 3, !dbg !45 + %984 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %905, float %906, float %907, float %908, i32 %577, i32 %578, i32 %579, i32 %580, i32 %694, i32 %695) #7, !dbg !45 + %985 = extractvalue { float, float, float, float } %984, 0, !dbg !45 + %986 = extractvalue { float, float, float, float } %984, 1, !dbg !45 + %987 = extractvalue { float, float, float, float } %984, 2, !dbg !45 + %988 = extractvalue { float, float, float, float } %984, 3, !dbg !45 + %989 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %910, float %911, float %912, float %913, i32 %577, i32 %578, i32 %579, i32 %580, i32 %706, i32 %707) #7, !dbg !45 + %990 = extractvalue { float, float, float, float } %989, 0, !dbg !45 + %991 = extractvalue { float, float, float, float } %989, 1, !dbg !45 + %992 = extractvalue { float, float, float, float } %989, 2, !dbg !45 + %993 = extractvalue { float, float, float, float } %989, 3, !dbg !45 + %994 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %915, float %916, float %917, float %918, i32 %589, i32 %590, i32 %591, i32 %592, i32 %670, i32 %671) #7, !dbg !45 + %995 = extractvalue { float, float, float, float } %994, 0, !dbg !45 + %996 = extractvalue { float, float, float, float } %994, 1, !dbg !45 + %997 = extractvalue { float, float, float, float } %994, 2, !dbg !45 + %998 = extractvalue { float, float, float, float } %994, 3, !dbg !45 + %999 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %920, float %921, float %922, float %923, i32 %589, i32 %590, i32 %591, i32 %592, i32 %682, i32 %683) #7, !dbg !45 + %1000 = extractvalue { float, float, float, float } %999, 0, !dbg !45 + %1001 = extractvalue { float, float, float, float } %999, 1, !dbg !45 + %1002 = extractvalue { float, float, float, float } %999, 2, !dbg !45 + %1003 = extractvalue { float, float, float, float } %999, 3, !dbg !45 + %1004 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %925, float %926, float %927, float %928, i32 %589, i32 %590, i32 %591, i32 %592, i32 %694, i32 %695) #7, !dbg !45 + %1005 = extractvalue { float, float, float, float } %1004, 0, !dbg !45 + %1006 = extractvalue { float, float, float, float } %1004, 1, !dbg !45 + %1007 = extractvalue { float, float, float, float } %1004, 2, !dbg !45 + %1008 = extractvalue { float, float, float, float } %1004, 3, !dbg !45 + %1009 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %930, float %931, float %932, float %933, i32 %589, i32 %590, i32 %591, i32 %592, i32 %706, i32 %707) #7, !dbg !45 + %1010 = extractvalue { float, float, float, float } %1009, 0, !dbg !45 + %1011 = extractvalue { float, float, float, float } %1009, 1, !dbg !45 + %1012 = extractvalue { float, float, float, float } %1009, 2, !dbg !45 + %1013 = extractvalue { float, float, float, float } %1009, 3, !dbg !45 + %1014 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %935, float %936, float %937, float %938, i32 %601, i32 %602, i32 %603, i32 %604, i32 %670, i32 %671) #7, !dbg !45 + %1015 = extractvalue { float, float, float, float } %1014, 0, !dbg !45 + %1016 = extractvalue { float, float, float, float } %1014, 1, !dbg !45 + %1017 = extractvalue { float, float, float, float } %1014, 2, !dbg !45 + %1018 = extractvalue { float, float, float, float } %1014, 3, !dbg !45 + %1019 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %940, float %941, float %942, float %943, i32 %601, i32 %602, i32 %603, i32 %604, i32 %682, i32 %683) #7, !dbg !45 + %1020 = extractvalue { float, float, float, float } %1019, 0, !dbg !45 + %1021 = extractvalue { float, float, float, float } %1019, 1, !dbg !45 + %1022 = extractvalue { float, float, float, float } %1019, 2, !dbg !45 + %1023 = extractvalue { float, float, float, float } %1019, 3, !dbg !45 + %1024 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %945, float %946, float %947, float %948, i32 %601, i32 %602, i32 %603, i32 %604, i32 %694, i32 %695) #7, !dbg !45 + %1025 = extractvalue { float, float, float, float } %1024, 0, !dbg !45 + %1026 = extractvalue { float, float, float, float } %1024, 1, !dbg !45 + %1027 = extractvalue { float, float, float, float } %1024, 2, !dbg !45 + %1028 = extractvalue { float, float, float, float } %1024, 3, !dbg !45 + %1029 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %950, float %951, float %952, float %953, i32 %601, i32 %602, i32 %603, i32 %604, i32 %706, i32 %707) #7, !dbg !45 + %1030 = extractvalue { float, float, float, float } %1029, 0, !dbg !45 + %1031 = extractvalue { float, float, float, float } %1029, 1, !dbg !45 + %1032 = extractvalue { float, float, float, float } %1029, 2, !dbg !45 + %1033 = extractvalue { float, float, float, float } %1029, 3, !dbg !45 + %1034 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %955, float %956, float %957, float %958, i32 %424, i32 %425, i32 %426, i32 %427, i32 %625, i32 %626) #7, !dbg !45 + %1035 = extractvalue { float, float, float, float } %1034, 0, !dbg !45 + %1036 = extractvalue { float, float, float, float } %1034, 1, !dbg !45 + %1037 = extractvalue { float, float, float, float } %1034, 2, !dbg !45 + %1038 = extractvalue { float, float, float, float } %1034, 3, !dbg !45 + %1039 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %960, float %961, float %962, float %963, i32 %424, i32 %425, i32 %426, i32 %427, i32 %637, i32 %638) #7, !dbg !45 + %1040 = extractvalue { float, float, float, float } %1039, 0, !dbg !45 + %1041 = extractvalue { float, float, float, float } %1039, 1, !dbg !45 + %1042 = extractvalue { float, float, float, float } %1039, 2, !dbg !45 + %1043 = extractvalue { float, float, float, float } %1039, 3, !dbg !45 + %1044 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %965, float %966, float %967, float %968, i32 %424, i32 %425, i32 %426, i32 %427, i32 %649, i32 %650) #7, !dbg !45 + %1045 = extractvalue { float, float, float, float } %1044, 0, !dbg !45 + %1046 = extractvalue { float, float, float, float } %1044, 1, !dbg !45 + %1047 = extractvalue { float, float, float, float } %1044, 2, !dbg !45 + %1048 = extractvalue { float, float, float, float } %1044, 3, !dbg !45 + %1049 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %970, float %971, float %972, float %973, i32 %424, i32 %425, i32 %426, i32 %427, i32 %661, i32 %662) #7, !dbg !45 + %1050 = extractvalue { float, float, float, float } %1049, 0, !dbg !45 + %1051 = extractvalue { float, float, float, float } %1049, 1, !dbg !45 + %1052 = extractvalue { float, float, float, float } %1049, 2, !dbg !45 + %1053 = extractvalue { float, float, float, float } %1049, 3, !dbg !45 + %1054 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %975, float %976, float %977, float %978, i32 %436, i32 %437, i32 %438, i32 %439, i32 %625, i32 %626) #7, !dbg !45 + %1055 = extractvalue { float, float, float, float } %1054, 0, !dbg !45 + %1056 = extractvalue { float, float, float, float } %1054, 1, !dbg !45 + %1057 = extractvalue { float, float, float, float } %1054, 2, !dbg !45 + %1058 = extractvalue { float, float, float, float } %1054, 3, !dbg !45 + %1059 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %980, float %981, float %982, float %983, i32 %436, i32 %437, i32 %438, i32 %439, i32 %637, i32 %638) #7, !dbg !45 + %1060 = extractvalue { float, float, float, float } %1059, 0, !dbg !45 + %1061 = extractvalue { float, float, float, float } %1059, 1, !dbg !45 + %1062 = extractvalue { float, float, float, float } %1059, 2, !dbg !45 + %1063 = extractvalue { float, float, float, float } %1059, 3, !dbg !45 + %1064 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %985, float %986, float %987, float %988, i32 %436, i32 %437, i32 %438, i32 %439, i32 %649, i32 %650) #7, !dbg !45 + %1065 = extractvalue { float, float, float, float } %1064, 0, !dbg !45 + %1066 = extractvalue { float, float, float, float } %1064, 1, !dbg !45 + %1067 = extractvalue { float, float, float, float } %1064, 2, !dbg !45 + %1068 = extractvalue { float, float, float, float } %1064, 3, !dbg !45 + %1069 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %990, float %991, float %992, float %993, i32 %436, i32 %437, i32 %438, i32 %439, i32 %661, i32 %662) #7, !dbg !45 + %1070 = extractvalue { float, float, float, float } %1069, 0, !dbg !45 + %1071 = extractvalue { float, float, float, float } %1069, 1, !dbg !45 + %1072 = extractvalue { float, float, float, float } %1069, 2, !dbg !45 + %1073 = extractvalue { float, float, float, float } %1069, 3, !dbg !45 + %1074 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %995, float %996, float %997, float %998, i32 %448, i32 %449, i32 %450, i32 %451, i32 %625, i32 %626) #7, !dbg !45 + %1075 = extractvalue { float, float, float, float } %1074, 0, !dbg !45 + %1076 = extractvalue { float, float, float, float } %1074, 1, !dbg !45 + %1077 = extractvalue { float, float, float, float } %1074, 2, !dbg !45 + %1078 = extractvalue { float, float, float, float } %1074, 3, !dbg !45 + %1079 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1000, float %1001, float %1002, float %1003, i32 %448, i32 %449, i32 %450, i32 %451, i32 %637, i32 %638) #7, !dbg !45 + %1080 = extractvalue { float, float, float, float } %1079, 0, !dbg !45 + %1081 = extractvalue { float, float, float, float } %1079, 1, !dbg !45 + %1082 = extractvalue { float, float, float, float } %1079, 2, !dbg !45 + %1083 = extractvalue { float, float, float, float } %1079, 3, !dbg !45 + %1084 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1005, float %1006, float %1007, float %1008, i32 %448, i32 %449, i32 %450, i32 %451, i32 %649, i32 %650) #7, !dbg !45 + %1085 = extractvalue { float, float, float, float } %1084, 0, !dbg !45 + %1086 = extractvalue { float, float, float, float } %1084, 1, !dbg !45 + %1087 = extractvalue { float, float, float, float } %1084, 2, !dbg !45 + %1088 = extractvalue { float, float, float, float } %1084, 3, !dbg !45 + %1089 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1010, float %1011, float %1012, float %1013, i32 %448, i32 %449, i32 %450, i32 %451, i32 %661, i32 %662) #7, !dbg !45 + %1090 = extractvalue { float, float, float, float } %1089, 0, !dbg !45 + %1091 = extractvalue { float, float, float, float } %1089, 1, !dbg !45 + %1092 = extractvalue { float, float, float, float } %1089, 2, !dbg !45 + %1093 = extractvalue { float, float, float, float } %1089, 3, !dbg !45 + %1094 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1015, float %1016, float %1017, float %1018, i32 %460, i32 %461, i32 %462, i32 %463, i32 %625, i32 %626) #7, !dbg !45 + %1095 = extractvalue { float, float, float, float } %1094, 0, !dbg !45 + %1096 = extractvalue { float, float, float, float } %1094, 1, !dbg !45 + %1097 = extractvalue { float, float, float, float } %1094, 2, !dbg !45 + %1098 = extractvalue { float, float, float, float } %1094, 3, !dbg !45 + %1099 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1020, float %1021, float %1022, float %1023, i32 %460, i32 %461, i32 %462, i32 %463, i32 %637, i32 %638) #7, !dbg !45 + %1100 = extractvalue { float, float, float, float } %1099, 0, !dbg !45 + %1101 = extractvalue { float, float, float, float } %1099, 1, !dbg !45 + %1102 = extractvalue { float, float, float, float } %1099, 2, !dbg !45 + %1103 = extractvalue { float, float, float, float } %1099, 3, !dbg !45 + %1104 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1025, float %1026, float %1027, float %1028, i32 %460, i32 %461, i32 %462, i32 %463, i32 %649, i32 %650) #7, !dbg !45 + %1105 = extractvalue { float, float, float, float } %1104, 0, !dbg !45 + %1106 = extractvalue { float, float, float, float } %1104, 1, !dbg !45 + %1107 = extractvalue { float, float, float, float } %1104, 2, !dbg !45 + %1108 = extractvalue { float, float, float, float } %1104, 3, !dbg !45 + %1109 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1030, float %1031, float %1032, float %1033, i32 %460, i32 %461, i32 %462, i32 %463, i32 %661, i32 %662) #7, !dbg !45 + %1110 = extractvalue { float, float, float, float } %1109, 0, !dbg !45 + %1111 = extractvalue { float, float, float, float } %1109, 1, !dbg !45 + %1112 = extractvalue { float, float, float, float } %1109, 2, !dbg !45 + %1113 = extractvalue { float, float, float, float } %1109, 3, !dbg !45 + %1114 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1035, float %1036, float %1037, float %1038, i32 %473, i32 %474, i32 %475, i32 %476, i32 %627, i32 %628) #7, !dbg !45 + %1115 = extractvalue { float, float, float, float } %1114, 0, !dbg !45 + %1116 = extractvalue { float, float, float, float } %1114, 1, !dbg !45 + %1117 = extractvalue { float, float, float, float } %1114, 2, !dbg !45 + %1118 = extractvalue { float, float, float, float } %1114, 3, !dbg !45 + %1119 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1040, float %1041, float %1042, float %1043, i32 %473, i32 %474, i32 %475, i32 %476, i32 %639, i32 %640) #7, !dbg !45 + %1120 = extractvalue { float, float, float, float } %1119, 0, !dbg !45 + %1121 = extractvalue { float, float, float, float } %1119, 1, !dbg !45 + %1122 = extractvalue { float, float, float, float } %1119, 2, !dbg !45 + %1123 = extractvalue { float, float, float, float } %1119, 3, !dbg !45 + %1124 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1045, float %1046, float %1047, float %1048, i32 %473, i32 %474, i32 %475, i32 %476, i32 %651, i32 %652) #7, !dbg !45 + %1125 = extractvalue { float, float, float, float } %1124, 0, !dbg !45 + %1126 = extractvalue { float, float, float, float } %1124, 1, !dbg !45 + %1127 = extractvalue { float, float, float, float } %1124, 2, !dbg !45 + %1128 = extractvalue { float, float, float, float } %1124, 3, !dbg !45 + %1129 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1050, float %1051, float %1052, float %1053, i32 %473, i32 %474, i32 %475, i32 %476, i32 %663, i32 %664) #7, !dbg !45 + %1130 = extractvalue { float, float, float, float } %1129, 0, !dbg !45 + %1131 = extractvalue { float, float, float, float } %1129, 1, !dbg !45 + %1132 = extractvalue { float, float, float, float } %1129, 2, !dbg !45 + %1133 = extractvalue { float, float, float, float } %1129, 3, !dbg !45 + %1134 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1055, float %1056, float %1057, float %1058, i32 %485, i32 %486, i32 %487, i32 %488, i32 %627, i32 %628) #7, !dbg !45 + %1135 = extractvalue { float, float, float, float } %1134, 0, !dbg !45 + %1136 = extractvalue { float, float, float, float } %1134, 1, !dbg !45 + %1137 = extractvalue { float, float, float, float } %1134, 2, !dbg !45 + %1138 = extractvalue { float, float, float, float } %1134, 3, !dbg !45 + %1139 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1060, float %1061, float %1062, float %1063, i32 %485, i32 %486, i32 %487, i32 %488, i32 %639, i32 %640) #7, !dbg !45 + %1140 = extractvalue { float, float, float, float } %1139, 0, !dbg !45 + %1141 = extractvalue { float, float, float, float } %1139, 1, !dbg !45 + %1142 = extractvalue { float, float, float, float } %1139, 2, !dbg !45 + %1143 = extractvalue { float, float, float, float } %1139, 3, !dbg !45 + %1144 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1065, float %1066, float %1067, float %1068, i32 %485, i32 %486, i32 %487, i32 %488, i32 %651, i32 %652) #7, !dbg !45 + %1145 = extractvalue { float, float, float, float } %1144, 0, !dbg !45 + %1146 = extractvalue { float, float, float, float } %1144, 1, !dbg !45 + %1147 = extractvalue { float, float, float, float } %1144, 2, !dbg !45 + %1148 = extractvalue { float, float, float, float } %1144, 3, !dbg !45 + %1149 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1070, float %1071, float %1072, float %1073, i32 %485, i32 %486, i32 %487, i32 %488, i32 %663, i32 %664) #7, !dbg !45 + %1150 = extractvalue { float, float, float, float } %1149, 0, !dbg !45 + %1151 = extractvalue { float, float, float, float } %1149, 1, !dbg !45 + %1152 = extractvalue { float, float, float, float } %1149, 2, !dbg !45 + %1153 = extractvalue { float, float, float, float } %1149, 3, !dbg !45 + %1154 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1075, float %1076, float %1077, float %1078, i32 %497, i32 %498, i32 %499, i32 %500, i32 %627, i32 %628) #7, !dbg !45 + %1155 = extractvalue { float, float, float, float } %1154, 0, !dbg !45 + %1156 = extractvalue { float, float, float, float } %1154, 1, !dbg !45 + %1157 = extractvalue { float, float, float, float } %1154, 2, !dbg !45 + %1158 = extractvalue { float, float, float, float } %1154, 3, !dbg !45 + %1159 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1080, float %1081, float %1082, float %1083, i32 %497, i32 %498, i32 %499, i32 %500, i32 %639, i32 %640) #7, !dbg !45 + %1160 = extractvalue { float, float, float, float } %1159, 0, !dbg !45 + %1161 = extractvalue { float, float, float, float } %1159, 1, !dbg !45 + %1162 = extractvalue { float, float, float, float } %1159, 2, !dbg !45 + %1163 = extractvalue { float, float, float, float } %1159, 3, !dbg !45 + %1164 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1085, float %1086, float %1087, float %1088, i32 %497, i32 %498, i32 %499, i32 %500, i32 %651, i32 %652) #7, !dbg !45 + %1165 = extractvalue { float, float, float, float } %1164, 0, !dbg !45 + %1166 = extractvalue { float, float, float, float } %1164, 1, !dbg !45 + %1167 = extractvalue { float, float, float, float } %1164, 2, !dbg !45 + %1168 = extractvalue { float, float, float, float } %1164, 3, !dbg !45 + %1169 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1090, float %1091, float %1092, float %1093, i32 %497, i32 %498, i32 %499, i32 %500, i32 %663, i32 %664) #7, !dbg !45 + %1170 = extractvalue { float, float, float, float } %1169, 0, !dbg !45 + %1171 = extractvalue { float, float, float, float } %1169, 1, !dbg !45 + %1172 = extractvalue { float, float, float, float } %1169, 2, !dbg !45 + %1173 = extractvalue { float, float, float, float } %1169, 3, !dbg !45 + %1174 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1095, float %1096, float %1097, float %1098, i32 %509, i32 %510, i32 %511, i32 %512, i32 %627, i32 %628) #7, !dbg !45 + %1175 = extractvalue { float, float, float, float } %1174, 0, !dbg !45 + %1176 = extractvalue { float, float, float, float } %1174, 1, !dbg !45 + %1177 = extractvalue { float, float, float, float } %1174, 2, !dbg !45 + %1178 = extractvalue { float, float, float, float } %1174, 3, !dbg !45 + %1179 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1100, float %1101, float %1102, float %1103, i32 %509, i32 %510, i32 %511, i32 %512, i32 %639, i32 %640) #7, !dbg !45 + %1180 = extractvalue { float, float, float, float } %1179, 0, !dbg !45 + %1181 = extractvalue { float, float, float, float } %1179, 1, !dbg !45 + %1182 = extractvalue { float, float, float, float } %1179, 2, !dbg !45 + %1183 = extractvalue { float, float, float, float } %1179, 3, !dbg !45 + %1184 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1105, float %1106, float %1107, float %1108, i32 %509, i32 %510, i32 %511, i32 %512, i32 %651, i32 %652) #7, !dbg !45 + %1185 = extractvalue { float, float, float, float } %1184, 0, !dbg !45 + %1186 = extractvalue { float, float, float, float } %1184, 1, !dbg !45 + %1187 = extractvalue { float, float, float, float } %1184, 2, !dbg !45 + %1188 = extractvalue { float, float, float, float } %1184, 3, !dbg !45 + %1189 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1110, float %1111, float %1112, float %1113, i32 %509, i32 %510, i32 %511, i32 %512, i32 %663, i32 %664) #7, !dbg !45 + %1190 = extractvalue { float, float, float, float } %1189, 0, !dbg !45 + %1191 = extractvalue { float, float, float, float } %1189, 1, !dbg !45 + %1192 = extractvalue { float, float, float, float } %1189, 2, !dbg !45 + %1193 = extractvalue { float, float, float, float } %1189, 3, !dbg !45 + %1194 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1115, float %1116, float %1117, float %1118, i32 %522, i32 %523, i32 %524, i32 %525, i32 %674, i32 %675) #7, !dbg !45 + %1195 = extractvalue { float, float, float, float } %1194, 0, !dbg !45 + %1196 = extractvalue { float, float, float, float } %1194, 1, !dbg !45 + %1197 = extractvalue { float, float, float, float } %1194, 2, !dbg !45 + %1198 = extractvalue { float, float, float, float } %1194, 3, !dbg !45 + %1199 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1120, float %1121, float %1122, float %1123, i32 %522, i32 %523, i32 %524, i32 %525, i32 %686, i32 %687) #7, !dbg !45 + %1200 = extractvalue { float, float, float, float } %1199, 0, !dbg !45 + %1201 = extractvalue { float, float, float, float } %1199, 1, !dbg !45 + %1202 = extractvalue { float, float, float, float } %1199, 2, !dbg !45 + %1203 = extractvalue { float, float, float, float } %1199, 3, !dbg !45 + %1204 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1125, float %1126, float %1127, float %1128, i32 %522, i32 %523, i32 %524, i32 %525, i32 %698, i32 %699) #7, !dbg !45 + %1205 = extractvalue { float, float, float, float } %1204, 0, !dbg !45 + %1206 = extractvalue { float, float, float, float } %1204, 1, !dbg !45 + %1207 = extractvalue { float, float, float, float } %1204, 2, !dbg !45 + %1208 = extractvalue { float, float, float, float } %1204, 3, !dbg !45 + %1209 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1130, float %1131, float %1132, float %1133, i32 %522, i32 %523, i32 %524, i32 %525, i32 %710, i32 %711) #7, !dbg !45 + %1210 = extractvalue { float, float, float, float } %1209, 0, !dbg !45 + %1211 = extractvalue { float, float, float, float } %1209, 1, !dbg !45 + %1212 = extractvalue { float, float, float, float } %1209, 2, !dbg !45 + %1213 = extractvalue { float, float, float, float } %1209, 3, !dbg !45 + %1214 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1135, float %1136, float %1137, float %1138, i32 %534, i32 %535, i32 %536, i32 %537, i32 %674, i32 %675) #7, !dbg !45 + %1215 = extractvalue { float, float, float, float } %1214, 0, !dbg !45 + %1216 = extractvalue { float, float, float, float } %1214, 1, !dbg !45 + %1217 = extractvalue { float, float, float, float } %1214, 2, !dbg !45 + %1218 = extractvalue { float, float, float, float } %1214, 3, !dbg !45 + %1219 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1140, float %1141, float %1142, float %1143, i32 %534, i32 %535, i32 %536, i32 %537, i32 %686, i32 %687) #7, !dbg !45 + %1220 = extractvalue { float, float, float, float } %1219, 0, !dbg !45 + %1221 = extractvalue { float, float, float, float } %1219, 1, !dbg !45 + %1222 = extractvalue { float, float, float, float } %1219, 2, !dbg !45 + %1223 = extractvalue { float, float, float, float } %1219, 3, !dbg !45 + %1224 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1145, float %1146, float %1147, float %1148, i32 %534, i32 %535, i32 %536, i32 %537, i32 %698, i32 %699) #7, !dbg !45 + %1225 = extractvalue { float, float, float, float } %1224, 0, !dbg !45 + %1226 = extractvalue { float, float, float, float } %1224, 1, !dbg !45 + %1227 = extractvalue { float, float, float, float } %1224, 2, !dbg !45 + %1228 = extractvalue { float, float, float, float } %1224, 3, !dbg !45 + %1229 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1150, float %1151, float %1152, float %1153, i32 %534, i32 %535, i32 %536, i32 %537, i32 %710, i32 %711) #7, !dbg !45 + %1230 = extractvalue { float, float, float, float } %1229, 0, !dbg !45 + %1231 = extractvalue { float, float, float, float } %1229, 1, !dbg !45 + %1232 = extractvalue { float, float, float, float } %1229, 2, !dbg !45 + %1233 = extractvalue { float, float, float, float } %1229, 3, !dbg !45 + %1234 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1155, float %1156, float %1157, float %1158, i32 %546, i32 %547, i32 %548, i32 %549, i32 %674, i32 %675) #7, !dbg !45 + %1235 = extractvalue { float, float, float, float } %1234, 0, !dbg !45 + %1236 = extractvalue { float, float, float, float } %1234, 1, !dbg !45 + %1237 = extractvalue { float, float, float, float } %1234, 2, !dbg !45 + %1238 = extractvalue { float, float, float, float } %1234, 3, !dbg !45 + %1239 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1160, float %1161, float %1162, float %1163, i32 %546, i32 %547, i32 %548, i32 %549, i32 %686, i32 %687) #7, !dbg !45 + %1240 = extractvalue { float, float, float, float } %1239, 0, !dbg !45 + %1241 = extractvalue { float, float, float, float } %1239, 1, !dbg !45 + %1242 = extractvalue { float, float, float, float } %1239, 2, !dbg !45 + %1243 = extractvalue { float, float, float, float } %1239, 3, !dbg !45 + %1244 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1165, float %1166, float %1167, float %1168, i32 %546, i32 %547, i32 %548, i32 %549, i32 %698, i32 %699) #7, !dbg !45 + %1245 = extractvalue { float, float, float, float } %1244, 0, !dbg !45 + %1246 = extractvalue { float, float, float, float } %1244, 1, !dbg !45 + %1247 = extractvalue { float, float, float, float } %1244, 2, !dbg !45 + %1248 = extractvalue { float, float, float, float } %1244, 3, !dbg !45 + %1249 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1170, float %1171, float %1172, float %1173, i32 %546, i32 %547, i32 %548, i32 %549, i32 %710, i32 %711) #7, !dbg !45 + %1250 = extractvalue { float, float, float, float } %1249, 0, !dbg !45 + %1251 = extractvalue { float, float, float, float } %1249, 1, !dbg !45 + %1252 = extractvalue { float, float, float, float } %1249, 2, !dbg !45 + %1253 = extractvalue { float, float, float, float } %1249, 3, !dbg !45 + %1254 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1175, float %1176, float %1177, float %1178, i32 %558, i32 %559, i32 %560, i32 %561, i32 %674, i32 %675) #7, !dbg !45 + %1255 = extractvalue { float, float, float, float } %1254, 0, !dbg !45 + %1256 = extractvalue { float, float, float, float } %1254, 1, !dbg !45 + %1257 = extractvalue { float, float, float, float } %1254, 2, !dbg !45 + %1258 = extractvalue { float, float, float, float } %1254, 3, !dbg !45 + %1259 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1180, float %1181, float %1182, float %1183, i32 %558, i32 %559, i32 %560, i32 %561, i32 %686, i32 %687) #7, !dbg !45 + %1260 = extractvalue { float, float, float, float } %1259, 0, !dbg !45 + %1261 = extractvalue { float, float, float, float } %1259, 1, !dbg !45 + %1262 = extractvalue { float, float, float, float } %1259, 2, !dbg !45 + %1263 = extractvalue { float, float, float, float } %1259, 3, !dbg !45 + %1264 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1185, float %1186, float %1187, float %1188, i32 %558, i32 %559, i32 %560, i32 %561, i32 %698, i32 %699) #7, !dbg !45 + %1265 = extractvalue { float, float, float, float } %1264, 0, !dbg !45 + %1266 = extractvalue { float, float, float, float } %1264, 1, !dbg !45 + %1267 = extractvalue { float, float, float, float } %1264, 2, !dbg !45 + %1268 = extractvalue { float, float, float, float } %1264, 3, !dbg !45 + %1269 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1190, float %1191, float %1192, float %1193, i32 %558, i32 %559, i32 %560, i32 %561, i32 %710, i32 %711) #7, !dbg !45 + %1270 = extractvalue { float, float, float, float } %1269, 0, !dbg !45 + %1271 = extractvalue { float, float, float, float } %1269, 1, !dbg !45 + %1272 = extractvalue { float, float, float, float } %1269, 2, !dbg !45 + %1273 = extractvalue { float, float, float, float } %1269, 3, !dbg !45 + %1274 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1195, float %1196, float %1197, float %1198, i32 %571, i32 %572, i32 %573, i32 %574, i32 %676, i32 %677) #7, !dbg !45 + %1275 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1200, float %1201, float %1202, float %1203, i32 %571, i32 %572, i32 %573, i32 %574, i32 %688, i32 %689) #7, !dbg !45 + %1276 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1205, float %1206, float %1207, float %1208, i32 %571, i32 %572, i32 %573, i32 %574, i32 %700, i32 %701) #7, !dbg !45 + %1277 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1210, float %1211, float %1212, float %1213, i32 %571, i32 %572, i32 %573, i32 %574, i32 %712, i32 %713) #7, !dbg !45 + %1278 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1215, float %1216, float %1217, float %1218, i32 %583, i32 %584, i32 %585, i32 %586, i32 %676, i32 %677) #7, !dbg !45 + %1279 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1220, float %1221, float %1222, float %1223, i32 %583, i32 %584, i32 %585, i32 %586, i32 %688, i32 %689) #7, !dbg !45 + %1280 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1225, float %1226, float %1227, float %1228, i32 %583, i32 %584, i32 %585, i32 %586, i32 %700, i32 %701) #7, !dbg !45 + %1281 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1230, float %1231, float %1232, float %1233, i32 %583, i32 %584, i32 %585, i32 %586, i32 %712, i32 %713) #7, !dbg !45 + %1282 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1235, float %1236, float %1237, float %1238, i32 %595, i32 %596, i32 %597, i32 %598, i32 %676, i32 %677) #7, !dbg !45 + %1283 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1240, float %1241, float %1242, float %1243, i32 %595, i32 %596, i32 %597, i32 %598, i32 %688, i32 %689) #7, !dbg !45 + %1284 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1245, float %1246, float %1247, float %1248, i32 %595, i32 %596, i32 %597, i32 %598, i32 %700, i32 %701) #7, !dbg !45 + %1285 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1250, float %1251, float %1252, float %1253, i32 %595, i32 %596, i32 %597, i32 %598, i32 %712, i32 %713) #7, !dbg !45 + %1286 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1255, float %1256, float %1257, float %1258, i32 %607, i32 %608, i32 %609, i32 %610, i32 %676, i32 %677) #7, !dbg !45 + %1287 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1260, float %1261, float %1262, float %1263, i32 %607, i32 %608, i32 %609, i32 %610, i32 %688, i32 %689) #7, !dbg !45 + %1288 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1265, float %1266, float %1267, float %1268, i32 %607, i32 %608, i32 %609, i32 %610, i32 %700, i32 %701) #7, !dbg !45 + %1289 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1270, float %1271, float %1272, float %1273, i32 %607, i32 %608, i32 %609, i32 %610, i32 %712, i32 %713) #7, !dbg !45 + %1290 = fsub float %90, %62, !dbg !31 + %1291 = fsub float %92, %62, !dbg !31 + %1292 = fsub float %90, %64, !dbg !31 + %1293 = fsub float %92, %64, !dbg !31 + %1294 = fsub float %94, %62, !dbg !31 + %1295 = fsub float %96, %62, !dbg !31 + %1296 = fsub float %94, %64, !dbg !31 + %1297 = fsub float %96, %64, !dbg !31 + %1298 = fsub float %98, %62, !dbg !31 + %1299 = fsub float %100, %62, !dbg !31 + %1300 = fsub float %98, %64, !dbg !31 + %1301 = fsub float %100, %64, !dbg !31 + %1302 = fsub float %102, %62, !dbg !31 + %1303 = fsub float %104, %62, !dbg !31 + %1304 = fsub float %102, %64, !dbg !31 + %1305 = fsub float %104, %64, !dbg !31 + %1306 = fsub float %94, %66, !dbg !31 + %1307 = fsub float %96, %66, !dbg !31 + %1308 = fsub float %94, %68, !dbg !31 + %1309 = fsub float %96, %68, !dbg !31 + %1310 = fsub float %98, %66, !dbg !31 + %1311 = fsub float %100, %66, !dbg !31 + %1312 = fsub float %98, %68, !dbg !31 + %1313 = fsub float %100, %68, !dbg !31 + %1314 = fsub float %102, %66, !dbg !31 + %1315 = fsub float %104, %66, !dbg !31 + %1316 = fsub float %102, %68, !dbg !31 + %1317 = fsub float %104, %68, !dbg !31 + %1318 = fsub float %98, %70, !dbg !31 + %1319 = fsub float %100, %70, !dbg !31 + %1320 = fsub float %98, %72, !dbg !31 + %1321 = fsub float %100, %72, !dbg !31 + %1322 = fsub float %102, %70, !dbg !31 + %1323 = fsub float %104, %70, !dbg !31 + %1324 = fsub float %102, %72, !dbg !31 + %1325 = fsub float %104, %72, !dbg !31 + %1326 = fsub float %102, %74, !dbg !31 + %1327 = fsub float %104, %74, !dbg !31 + %1328 = fsub float %102, %76, !dbg !31 + %1329 = fsub float %104, %76, !dbg !31 + %1330 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i = icmp eq i32 %1330, 0, !dbg !32 + br i1 %.not.i, label %1333, label %1331, !dbg !32 + +1331: ; preds = %9 + %1332 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1290), !dbg !32 + br label %__nv_exp2f.exit, !dbg !32 + +1333: ; preds = %9 + %1334 = tail call float @llvm.nvvm.ex2.approx.f32(float %1290), !dbg !32 + br label %__nv_exp2f.exit, !dbg !32 + +__nv_exp2f.exit: ; preds = %1331, %1333 + %.0.i = phi float [ %1332, %1331 ], [ %1334, %1333 ], !dbg !32 + %1335 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i544 = icmp eq i32 %1335, 0, !dbg !32 + br i1 %.not.i544, label %1338, label %1336, !dbg !32 + +1336: ; preds = %__nv_exp2f.exit + %1337 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1291), !dbg !32 + br label %__nv_exp2f.exit546, !dbg !32 + +1338: ; preds = %__nv_exp2f.exit + %1339 = tail call float @llvm.nvvm.ex2.approx.f32(float %1291), !dbg !32 + br label %__nv_exp2f.exit546, !dbg !32 + +__nv_exp2f.exit546: ; preds = %1336, %1338 + %.0.i545 = phi float [ %1337, %1336 ], [ %1339, %1338 ], !dbg !32 + %1340 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i547 = icmp eq i32 %1340, 0, !dbg !32 + br i1 %.not.i547, label %1343, label %1341, !dbg !32 + +1341: ; preds = %__nv_exp2f.exit546 + %1342 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1292), !dbg !32 + br label %__nv_exp2f.exit549, !dbg !32 + +1343: ; preds = %__nv_exp2f.exit546 + %1344 = tail call float @llvm.nvvm.ex2.approx.f32(float %1292), !dbg !32 + br label %__nv_exp2f.exit549, !dbg !32 + +__nv_exp2f.exit549: ; preds = %1341, %1343 + %.0.i548 = phi float [ %1342, %1341 ], [ %1344, %1343 ], !dbg !32 + %1345 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i550 = icmp eq i32 %1345, 0, !dbg !32 + br i1 %.not.i550, label %1348, label %1346, !dbg !32 + +1346: ; preds = %__nv_exp2f.exit549 + %1347 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1293), !dbg !32 + br label %__nv_exp2f.exit552, !dbg !32 + +1348: ; preds = %__nv_exp2f.exit549 + %1349 = tail call float @llvm.nvvm.ex2.approx.f32(float %1293), !dbg !32 + br label %__nv_exp2f.exit552, !dbg !32 + +__nv_exp2f.exit552: ; preds = %1346, %1348 + %.0.i551 = phi float [ %1347, %1346 ], [ %1349, %1348 ], !dbg !32 + %1350 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i553 = icmp eq i32 %1350, 0, !dbg !32 + br i1 %.not.i553, label %1353, label %1351, !dbg !32 + +1351: ; preds = %__nv_exp2f.exit552 + %1352 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1294), !dbg !32 + br label %__nv_exp2f.exit555, !dbg !32 + +1353: ; preds = %__nv_exp2f.exit552 + %1354 = tail call float @llvm.nvvm.ex2.approx.f32(float %1294), !dbg !32 + br label %__nv_exp2f.exit555, !dbg !32 + +__nv_exp2f.exit555: ; preds = %1351, %1353 + %.0.i554 = phi float [ %1352, %1351 ], [ %1354, %1353 ], !dbg !32 + %1355 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i556 = icmp eq i32 %1355, 0, !dbg !32 + br i1 %.not.i556, label %1358, label %1356, !dbg !32 + +1356: ; preds = %__nv_exp2f.exit555 + %1357 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1295), !dbg !32 + br label %__nv_exp2f.exit558, !dbg !32 + +1358: ; preds = %__nv_exp2f.exit555 + %1359 = tail call float @llvm.nvvm.ex2.approx.f32(float %1295), !dbg !32 + br label %__nv_exp2f.exit558, !dbg !32 + +__nv_exp2f.exit558: ; preds = %1356, %1358 + %.0.i557 = phi float [ %1357, %1356 ], [ %1359, %1358 ], !dbg !32 + %1360 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i559 = icmp eq i32 %1360, 0, !dbg !32 + br i1 %.not.i559, label %1363, label %1361, !dbg !32 + +1361: ; preds = %__nv_exp2f.exit558 + %1362 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1296), !dbg !32 + br label %__nv_exp2f.exit561, !dbg !32 + +1363: ; preds = %__nv_exp2f.exit558 + %1364 = tail call float @llvm.nvvm.ex2.approx.f32(float %1296), !dbg !32 + br label %__nv_exp2f.exit561, !dbg !32 + +__nv_exp2f.exit561: ; preds = %1361, %1363 + %.0.i560 = phi float [ %1362, %1361 ], [ %1364, %1363 ], !dbg !32 + %1365 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i562 = icmp eq i32 %1365, 0, !dbg !32 + br i1 %.not.i562, label %1368, label %1366, !dbg !32 + +1366: ; preds = %__nv_exp2f.exit561 + %1367 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1297), !dbg !32 + br label %__nv_exp2f.exit564, !dbg !32 + +1368: ; preds = %__nv_exp2f.exit561 + %1369 = tail call float @llvm.nvvm.ex2.approx.f32(float %1297), !dbg !32 + br label %__nv_exp2f.exit564, !dbg !32 + +__nv_exp2f.exit564: ; preds = %1366, %1368 + %.0.i563 = phi float [ %1367, %1366 ], [ %1369, %1368 ], !dbg !32 + %1370 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i565 = icmp eq i32 %1370, 0, !dbg !32 + br i1 %.not.i565, label %1373, label %1371, !dbg !32 + +1371: ; preds = %__nv_exp2f.exit564 + %1372 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1298), !dbg !32 + br label %__nv_exp2f.exit567, !dbg !32 + +1373: ; preds = %__nv_exp2f.exit564 + %1374 = tail call float @llvm.nvvm.ex2.approx.f32(float %1298), !dbg !32 + br label %__nv_exp2f.exit567, !dbg !32 + +__nv_exp2f.exit567: ; preds = %1371, %1373 + %.0.i566 = phi float [ %1372, %1371 ], [ %1374, %1373 ], !dbg !32 + %1375 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i568 = icmp eq i32 %1375, 0, !dbg !32 + br i1 %.not.i568, label %1378, label %1376, !dbg !32 + +1376: ; preds = %__nv_exp2f.exit567 + %1377 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1299), !dbg !32 + br label %__nv_exp2f.exit570, !dbg !32 + +1378: ; preds = %__nv_exp2f.exit567 + %1379 = tail call float @llvm.nvvm.ex2.approx.f32(float %1299), !dbg !32 + br label %__nv_exp2f.exit570, !dbg !32 + +__nv_exp2f.exit570: ; preds = %1376, %1378 + %.0.i569 = phi float [ %1377, %1376 ], [ %1379, %1378 ], !dbg !32 + %1380 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i571 = icmp eq i32 %1380, 0, !dbg !32 + br i1 %.not.i571, label %1383, label %1381, !dbg !32 + +1381: ; preds = %__nv_exp2f.exit570 + %1382 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1300), !dbg !32 + br label %__nv_exp2f.exit573, !dbg !32 + +1383: ; preds = %__nv_exp2f.exit570 + %1384 = tail call float @llvm.nvvm.ex2.approx.f32(float %1300), !dbg !32 + br label %__nv_exp2f.exit573, !dbg !32 + +__nv_exp2f.exit573: ; preds = %1381, %1383 + %.0.i572 = phi float [ %1382, %1381 ], [ %1384, %1383 ], !dbg !32 + %1385 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i574 = icmp eq i32 %1385, 0, !dbg !32 + br i1 %.not.i574, label %1388, label %1386, !dbg !32 + +1386: ; preds = %__nv_exp2f.exit573 + %1387 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1301), !dbg !32 + br label %__nv_exp2f.exit576, !dbg !32 + +1388: ; preds = %__nv_exp2f.exit573 + %1389 = tail call float @llvm.nvvm.ex2.approx.f32(float %1301), !dbg !32 + br label %__nv_exp2f.exit576, !dbg !32 + +__nv_exp2f.exit576: ; preds = %1386, %1388 + %.0.i575 = phi float [ %1387, %1386 ], [ %1389, %1388 ], !dbg !32 + %1390 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i577 = icmp eq i32 %1390, 0, !dbg !32 + br i1 %.not.i577, label %1393, label %1391, !dbg !32 + +1391: ; preds = %__nv_exp2f.exit576 + %1392 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1302), !dbg !32 + br label %__nv_exp2f.exit579, !dbg !32 + +1393: ; preds = %__nv_exp2f.exit576 + %1394 = tail call float @llvm.nvvm.ex2.approx.f32(float %1302), !dbg !32 + br label %__nv_exp2f.exit579, !dbg !32 + +__nv_exp2f.exit579: ; preds = %1391, %1393 + %.0.i578 = phi float [ %1392, %1391 ], [ %1394, %1393 ], !dbg !32 + %1395 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i580 = icmp eq i32 %1395, 0, !dbg !32 + br i1 %.not.i580, label %1398, label %1396, !dbg !32 + +1396: ; preds = %__nv_exp2f.exit579 + %1397 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1303), !dbg !32 + br label %__nv_exp2f.exit582, !dbg !32 + +1398: ; preds = %__nv_exp2f.exit579 + %1399 = tail call float @llvm.nvvm.ex2.approx.f32(float %1303), !dbg !32 + br label %__nv_exp2f.exit582, !dbg !32 + +__nv_exp2f.exit582: ; preds = %1396, %1398 + %.0.i581 = phi float [ %1397, %1396 ], [ %1399, %1398 ], !dbg !32 + %1400 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i583 = icmp eq i32 %1400, 0, !dbg !32 + br i1 %.not.i583, label %1403, label %1401, !dbg !32 + +1401: ; preds = %__nv_exp2f.exit582 + %1402 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1304), !dbg !32 + br label %__nv_exp2f.exit585, !dbg !32 + +1403: ; preds = %__nv_exp2f.exit582 + %1404 = tail call float @llvm.nvvm.ex2.approx.f32(float %1304), !dbg !32 + br label %__nv_exp2f.exit585, !dbg !32 + +__nv_exp2f.exit585: ; preds = %1401, %1403 + %.0.i584 = phi float [ %1402, %1401 ], [ %1404, %1403 ], !dbg !32 + %1405 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i586 = icmp eq i32 %1405, 0, !dbg !32 + br i1 %.not.i586, label %1408, label %1406, !dbg !32 + +1406: ; preds = %__nv_exp2f.exit585 + %1407 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1305), !dbg !32 + br label %__nv_exp2f.exit588, !dbg !32 + +1408: ; preds = %__nv_exp2f.exit585 + %1409 = tail call float @llvm.nvvm.ex2.approx.f32(float %1305), !dbg !32 + br label %__nv_exp2f.exit588, !dbg !32 + +__nv_exp2f.exit588: ; preds = %1406, %1408 + %.0.i587 = phi float [ %1407, %1406 ], [ %1409, %1408 ], !dbg !32 + %1410 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %1411 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %1412 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %1413 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %1414 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i601 = icmp eq i32 %1414, 0, !dbg !32 + br i1 %.not.i601, label %1417, label %1415, !dbg !32 + +1415: ; preds = %__nv_exp2f.exit588 + %1416 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1306), !dbg !32 + br label %__nv_exp2f.exit603, !dbg !32 + +1417: ; preds = %__nv_exp2f.exit588 + %1418 = tail call float @llvm.nvvm.ex2.approx.f32(float %1306), !dbg !32 + br label %__nv_exp2f.exit603, !dbg !32 + +__nv_exp2f.exit603: ; preds = %1415, %1417 + %.0.i602 = phi float [ %1416, %1415 ], [ %1418, %1417 ], !dbg !32 + %1419 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i604 = icmp eq i32 %1419, 0, !dbg !32 + br i1 %.not.i604, label %1422, label %1420, !dbg !32 + +1420: ; preds = %__nv_exp2f.exit603 + %1421 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1307), !dbg !32 + br label %__nv_exp2f.exit606, !dbg !32 + +1422: ; preds = %__nv_exp2f.exit603 + %1423 = tail call float @llvm.nvvm.ex2.approx.f32(float %1307), !dbg !32 + br label %__nv_exp2f.exit606, !dbg !32 + +__nv_exp2f.exit606: ; preds = %1420, %1422 + %.0.i605 = phi float [ %1421, %1420 ], [ %1423, %1422 ], !dbg !32 + %1424 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i607 = icmp eq i32 %1424, 0, !dbg !32 + br i1 %.not.i607, label %1427, label %1425, !dbg !32 + +1425: ; preds = %__nv_exp2f.exit606 + %1426 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1308), !dbg !32 + br label %__nv_exp2f.exit609, !dbg !32 + +1427: ; preds = %__nv_exp2f.exit606 + %1428 = tail call float @llvm.nvvm.ex2.approx.f32(float %1308), !dbg !32 + br label %__nv_exp2f.exit609, !dbg !32 + +__nv_exp2f.exit609: ; preds = %1425, %1427 + %.0.i608 = phi float [ %1426, %1425 ], [ %1428, %1427 ], !dbg !32 + %1429 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i610 = icmp eq i32 %1429, 0, !dbg !32 + br i1 %.not.i610, label %1432, label %1430, !dbg !32 + +1430: ; preds = %__nv_exp2f.exit609 + %1431 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1309), !dbg !32 + br label %__nv_exp2f.exit612, !dbg !32 + +1432: ; preds = %__nv_exp2f.exit609 + %1433 = tail call float @llvm.nvvm.ex2.approx.f32(float %1309), !dbg !32 + br label %__nv_exp2f.exit612, !dbg !32 + +__nv_exp2f.exit612: ; preds = %1430, %1432 + %.0.i611 = phi float [ %1431, %1430 ], [ %1433, %1432 ], !dbg !32 + %1434 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i613 = icmp eq i32 %1434, 0, !dbg !32 + br i1 %.not.i613, label %1437, label %1435, !dbg !32 + +1435: ; preds = %__nv_exp2f.exit612 + %1436 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1310), !dbg !32 + br label %__nv_exp2f.exit615, !dbg !32 + +1437: ; preds = %__nv_exp2f.exit612 + %1438 = tail call float @llvm.nvvm.ex2.approx.f32(float %1310), !dbg !32 + br label %__nv_exp2f.exit615, !dbg !32 + +__nv_exp2f.exit615: ; preds = %1435, %1437 + %.0.i614 = phi float [ %1436, %1435 ], [ %1438, %1437 ], !dbg !32 + %1439 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i616 = icmp eq i32 %1439, 0, !dbg !32 + br i1 %.not.i616, label %1442, label %1440, !dbg !32 + +1440: ; preds = %__nv_exp2f.exit615 + %1441 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1311), !dbg !32 + br label %__nv_exp2f.exit618, !dbg !32 + +1442: ; preds = %__nv_exp2f.exit615 + %1443 = tail call float @llvm.nvvm.ex2.approx.f32(float %1311), !dbg !32 + br label %__nv_exp2f.exit618, !dbg !32 + +__nv_exp2f.exit618: ; preds = %1440, %1442 + %.0.i617 = phi float [ %1441, %1440 ], [ %1443, %1442 ], !dbg !32 + %1444 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i619 = icmp eq i32 %1444, 0, !dbg !32 + br i1 %.not.i619, label %1447, label %1445, !dbg !32 + +1445: ; preds = %__nv_exp2f.exit618 + %1446 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1312), !dbg !32 + br label %__nv_exp2f.exit621, !dbg !32 + +1447: ; preds = %__nv_exp2f.exit618 + %1448 = tail call float @llvm.nvvm.ex2.approx.f32(float %1312), !dbg !32 + br label %__nv_exp2f.exit621, !dbg !32 + +__nv_exp2f.exit621: ; preds = %1445, %1447 + %.0.i620 = phi float [ %1446, %1445 ], [ %1448, %1447 ], !dbg !32 + %1449 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i622 = icmp eq i32 %1449, 0, !dbg !32 + br i1 %.not.i622, label %1452, label %1450, !dbg !32 + +1450: ; preds = %__nv_exp2f.exit621 + %1451 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1313), !dbg !32 + br label %__nv_exp2f.exit624, !dbg !32 + +1452: ; preds = %__nv_exp2f.exit621 + %1453 = tail call float @llvm.nvvm.ex2.approx.f32(float %1313), !dbg !32 + br label %__nv_exp2f.exit624, !dbg !32 + +__nv_exp2f.exit624: ; preds = %1450, %1452 + %.0.i623 = phi float [ %1451, %1450 ], [ %1453, %1452 ], !dbg !32 + %1454 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i625 = icmp eq i32 %1454, 0, !dbg !32 + br i1 %.not.i625, label %1457, label %1455, !dbg !32 + +1455: ; preds = %__nv_exp2f.exit624 + %1456 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1314), !dbg !32 + br label %__nv_exp2f.exit627, !dbg !32 + +1457: ; preds = %__nv_exp2f.exit624 + %1458 = tail call float @llvm.nvvm.ex2.approx.f32(float %1314), !dbg !32 + br label %__nv_exp2f.exit627, !dbg !32 + +__nv_exp2f.exit627: ; preds = %1455, %1457 + %.0.i626 = phi float [ %1456, %1455 ], [ %1458, %1457 ], !dbg !32 + %1459 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i628 = icmp eq i32 %1459, 0, !dbg !32 + br i1 %.not.i628, label %1462, label %1460, !dbg !32 + +1460: ; preds = %__nv_exp2f.exit627 + %1461 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1315), !dbg !32 + br label %__nv_exp2f.exit630, !dbg !32 + +1462: ; preds = %__nv_exp2f.exit627 + %1463 = tail call float @llvm.nvvm.ex2.approx.f32(float %1315), !dbg !32 + br label %__nv_exp2f.exit630, !dbg !32 + +__nv_exp2f.exit630: ; preds = %1460, %1462 + %.0.i629 = phi float [ %1461, %1460 ], [ %1463, %1462 ], !dbg !32 + %1464 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i631 = icmp eq i32 %1464, 0, !dbg !32 + br i1 %.not.i631, label %1467, label %1465, !dbg !32 + +1465: ; preds = %__nv_exp2f.exit630 + %1466 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1316), !dbg !32 + br label %__nv_exp2f.exit633, !dbg !32 + +1467: ; preds = %__nv_exp2f.exit630 + %1468 = tail call float @llvm.nvvm.ex2.approx.f32(float %1316), !dbg !32 + br label %__nv_exp2f.exit633, !dbg !32 + +__nv_exp2f.exit633: ; preds = %1465, %1467 + %.0.i632 = phi float [ %1466, %1465 ], [ %1468, %1467 ], !dbg !32 + %1469 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i634 = icmp eq i32 %1469, 0, !dbg !32 + br i1 %.not.i634, label %1472, label %1470, !dbg !32 + +1470: ; preds = %__nv_exp2f.exit633 + %1471 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1317), !dbg !32 + br label %__nv_exp2f.exit636, !dbg !32 + +1472: ; preds = %__nv_exp2f.exit633 + %1473 = tail call float @llvm.nvvm.ex2.approx.f32(float %1317), !dbg !32 + br label %__nv_exp2f.exit636, !dbg !32 + +__nv_exp2f.exit636: ; preds = %1470, %1472 + %.0.i635 = phi float [ %1471, %1470 ], [ %1473, %1472 ], !dbg !32 + %1474 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %1475 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %1476 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %1477 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %1478 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %1479 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %1480 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %1481 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %1482 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i661 = icmp eq i32 %1482, 0, !dbg !32 + br i1 %.not.i661, label %1485, label %1483, !dbg !32 + +1483: ; preds = %__nv_exp2f.exit636 + %1484 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1318), !dbg !32 + br label %__nv_exp2f.exit663, !dbg !32 + +1485: ; preds = %__nv_exp2f.exit636 + %1486 = tail call float @llvm.nvvm.ex2.approx.f32(float %1318), !dbg !32 + br label %__nv_exp2f.exit663, !dbg !32 + +__nv_exp2f.exit663: ; preds = %1483, %1485 + %.0.i662 = phi float [ %1484, %1483 ], [ %1486, %1485 ], !dbg !32 + %1487 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i664 = icmp eq i32 %1487, 0, !dbg !32 + br i1 %.not.i664, label %1490, label %1488, !dbg !32 + +1488: ; preds = %__nv_exp2f.exit663 + %1489 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1319), !dbg !32 + br label %__nv_exp2f.exit666, !dbg !32 + +1490: ; preds = %__nv_exp2f.exit663 + %1491 = tail call float @llvm.nvvm.ex2.approx.f32(float %1319), !dbg !32 + br label %__nv_exp2f.exit666, !dbg !32 + +__nv_exp2f.exit666: ; preds = %1488, %1490 + %.0.i665 = phi float [ %1489, %1488 ], [ %1491, %1490 ], !dbg !32 + %1492 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i667 = icmp eq i32 %1492, 0, !dbg !32 + br i1 %.not.i667, label %1495, label %1493, !dbg !32 + +1493: ; preds = %__nv_exp2f.exit666 + %1494 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1320), !dbg !32 + br label %__nv_exp2f.exit669, !dbg !32 + +1495: ; preds = %__nv_exp2f.exit666 + %1496 = tail call float @llvm.nvvm.ex2.approx.f32(float %1320), !dbg !32 + br label %__nv_exp2f.exit669, !dbg !32 + +__nv_exp2f.exit669: ; preds = %1493, %1495 + %.0.i668 = phi float [ %1494, %1493 ], [ %1496, %1495 ], !dbg !32 + %1497 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i670 = icmp eq i32 %1497, 0, !dbg !32 + br i1 %.not.i670, label %1500, label %1498, !dbg !32 + +1498: ; preds = %__nv_exp2f.exit669 + %1499 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1321), !dbg !32 + br label %__nv_exp2f.exit672, !dbg !32 + +1500: ; preds = %__nv_exp2f.exit669 + %1501 = tail call float @llvm.nvvm.ex2.approx.f32(float %1321), !dbg !32 + br label %__nv_exp2f.exit672, !dbg !32 + +__nv_exp2f.exit672: ; preds = %1498, %1500 + %.0.i671 = phi float [ %1499, %1498 ], [ %1501, %1500 ], !dbg !32 + %1502 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i673 = icmp eq i32 %1502, 0, !dbg !32 + br i1 %.not.i673, label %1505, label %1503, !dbg !32 + +1503: ; preds = %__nv_exp2f.exit672 + %1504 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1322), !dbg !32 + br label %__nv_exp2f.exit675, !dbg !32 + +1505: ; preds = %__nv_exp2f.exit672 + %1506 = tail call float @llvm.nvvm.ex2.approx.f32(float %1322), !dbg !32 + br label %__nv_exp2f.exit675, !dbg !32 + +__nv_exp2f.exit675: ; preds = %1503, %1505 + %.0.i674 = phi float [ %1504, %1503 ], [ %1506, %1505 ], !dbg !32 + %1507 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i676 = icmp eq i32 %1507, 0, !dbg !32 + br i1 %.not.i676, label %1510, label %1508, !dbg !32 + +1508: ; preds = %__nv_exp2f.exit675 + %1509 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1323), !dbg !32 + br label %__nv_exp2f.exit678, !dbg !32 + +1510: ; preds = %__nv_exp2f.exit675 + %1511 = tail call float @llvm.nvvm.ex2.approx.f32(float %1323), !dbg !32 + br label %__nv_exp2f.exit678, !dbg !32 + +__nv_exp2f.exit678: ; preds = %1508, %1510 + %.0.i677 = phi float [ %1509, %1508 ], [ %1511, %1510 ], !dbg !32 + %1512 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i679 = icmp eq i32 %1512, 0, !dbg !32 + br i1 %.not.i679, label %1515, label %1513, !dbg !32 + +1513: ; preds = %__nv_exp2f.exit678 + %1514 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1324), !dbg !32 + br label %__nv_exp2f.exit681, !dbg !32 + +1515: ; preds = %__nv_exp2f.exit678 + %1516 = tail call float @llvm.nvvm.ex2.approx.f32(float %1324), !dbg !32 + br label %__nv_exp2f.exit681, !dbg !32 + +__nv_exp2f.exit681: ; preds = %1513, %1515 + %.0.i680 = phi float [ %1514, %1513 ], [ %1516, %1515 ], !dbg !32 + %1517 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i682 = icmp eq i32 %1517, 0, !dbg !32 + br i1 %.not.i682, label %1520, label %1518, !dbg !32 + +1518: ; preds = %__nv_exp2f.exit681 + %1519 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1325), !dbg !32 + br label %__nv_exp2f.exit684, !dbg !32 + +1520: ; preds = %__nv_exp2f.exit681 + %1521 = tail call float @llvm.nvvm.ex2.approx.f32(float %1325), !dbg !32 + br label %__nv_exp2f.exit684, !dbg !32 + +__nv_exp2f.exit684: ; preds = %1518, %1520 + %.0.i683 = phi float [ %1519, %1518 ], [ %1521, %1520 ], !dbg !32 + %1522 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %1523 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %1524 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %1525 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %1526 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %1527 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %1528 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %1529 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %1530 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %1531 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %1532 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %1533 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %1534 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i721 = icmp eq i32 %1534, 0, !dbg !32 + br i1 %.not.i721, label %1537, label %1535, !dbg !32 + +1535: ; preds = %__nv_exp2f.exit684 + %1536 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1326), !dbg !32 + br label %__nv_exp2f.exit723, !dbg !32 + +1537: ; preds = %__nv_exp2f.exit684 + %1538 = tail call float @llvm.nvvm.ex2.approx.f32(float %1326), !dbg !32 + br label %__nv_exp2f.exit723, !dbg !32 + +__nv_exp2f.exit723: ; preds = %1535, %1537 + %.0.i722 = phi float [ %1536, %1535 ], [ %1538, %1537 ], !dbg !32 + %1539 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i724 = icmp eq i32 %1539, 0, !dbg !32 + br i1 %.not.i724, label %1542, label %1540, !dbg !32 + +1540: ; preds = %__nv_exp2f.exit723 + %1541 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1327), !dbg !32 + br label %__nv_exp2f.exit726, !dbg !32 + +1542: ; preds = %__nv_exp2f.exit723 + %1543 = tail call float @llvm.nvvm.ex2.approx.f32(float %1327), !dbg !32 + br label %__nv_exp2f.exit726, !dbg !32 + +__nv_exp2f.exit726: ; preds = %1540, %1542 + %.0.i725 = phi float [ %1541, %1540 ], [ %1543, %1542 ], !dbg !32 + %1544 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i727 = icmp eq i32 %1544, 0, !dbg !32 + br i1 %.not.i727, label %1547, label %1545, !dbg !32 + +1545: ; preds = %__nv_exp2f.exit726 + %1546 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1328), !dbg !32 + br label %__nv_exp2f.exit729, !dbg !32 + +1547: ; preds = %__nv_exp2f.exit726 + %1548 = tail call float @llvm.nvvm.ex2.approx.f32(float %1328), !dbg !32 + br label %__nv_exp2f.exit729, !dbg !32 + +__nv_exp2f.exit729: ; preds = %1545, %1547 + %.0.i728 = phi float [ %1546, %1545 ], [ %1548, %1547 ], !dbg !32 + %1549 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !32 + %.not.i730 = icmp eq i32 %1549, 0, !dbg !32 + br i1 %.not.i730, label %1552, label %1550, !dbg !32 + +1550: ; preds = %__nv_exp2f.exit729 + %1551 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1329), !dbg !32 + br label %__nv_exp2f.exit732, !dbg !32 + +1552: ; preds = %__nv_exp2f.exit729 + %1553 = tail call float @llvm.nvvm.ex2.approx.f32(float %1329), !dbg !32 + br label %__nv_exp2f.exit732, !dbg !32 + +__nv_exp2f.exit732: ; preds = %1550, %1552 + %.0.i731 = phi float [ %1551, %1550 ], [ %1553, %1552 ], !dbg !32 + %1554 = extractvalue { float, float, float, float } %1289, 3, !dbg !45 + %1555 = extractvalue { float, float, float, float } %1289, 2, !dbg !45 + %1556 = extractvalue { float, float, float, float } %1289, 1, !dbg !45 + %1557 = extractvalue { float, float, float, float } %1289, 0, !dbg !45 + %1558 = extractvalue { float, float, float, float } %1285, 3, !dbg !45 + %1559 = extractvalue { float, float, float, float } %1285, 2, !dbg !45 + %1560 = extractvalue { float, float, float, float } %1285, 1, !dbg !45 + %1561 = extractvalue { float, float, float, float } %1285, 0, !dbg !45 + %1562 = extractvalue { float, float, float, float } %1284, 3, !dbg !45 + %1563 = extractvalue { float, float, float, float } %1284, 2, !dbg !45 + %1564 = extractvalue { float, float, float, float } %1284, 1, !dbg !45 + %1565 = extractvalue { float, float, float, float } %1284, 0, !dbg !45 + %1566 = extractvalue { float, float, float, float } %1281, 3, !dbg !45 + %1567 = extractvalue { float, float, float, float } %1281, 2, !dbg !45 + %1568 = extractvalue { float, float, float, float } %1281, 1, !dbg !45 + %1569 = extractvalue { float, float, float, float } %1281, 0, !dbg !45 + %1570 = extractvalue { float, float, float, float } %1280, 3, !dbg !45 + %1571 = extractvalue { float, float, float, float } %1280, 2, !dbg !45 + %1572 = extractvalue { float, float, float, float } %1280, 1, !dbg !45 + %1573 = extractvalue { float, float, float, float } %1280, 0, !dbg !45 + %1574 = extractvalue { float, float, float, float } %1279, 3, !dbg !45 + %1575 = extractvalue { float, float, float, float } %1279, 2, !dbg !45 + %1576 = extractvalue { float, float, float, float } %1279, 1, !dbg !45 + %1577 = extractvalue { float, float, float, float } %1279, 0, !dbg !45 + %1578 = extractvalue { float, float, float, float } %1277, 3, !dbg !45 + %1579 = extractvalue { float, float, float, float } %1277, 2, !dbg !45 + %1580 = extractvalue { float, float, float, float } %1277, 1, !dbg !45 + %1581 = extractvalue { float, float, float, float } %1277, 0, !dbg !45 + %1582 = extractvalue { float, float, float, float } %1276, 3, !dbg !45 + %1583 = extractvalue { float, float, float, float } %1276, 2, !dbg !45 + %1584 = extractvalue { float, float, float, float } %1276, 1, !dbg !45 + %1585 = extractvalue { float, float, float, float } %1276, 0, !dbg !45 + %1586 = extractvalue { float, float, float, float } %1275, 3, !dbg !45 + %1587 = extractvalue { float, float, float, float } %1275, 2, !dbg !45 + %1588 = extractvalue { float, float, float, float } %1275, 1, !dbg !45 + %1589 = extractvalue { float, float, float, float } %1275, 0, !dbg !45 + %1590 = extractvalue { float, float, float, float } %1274, 3, !dbg !45 + %1591 = extractvalue { float, float, float, float } %1274, 2, !dbg !45 + %1592 = extractvalue { float, float, float, float } %1274, 1, !dbg !45 + %1593 = extractvalue { float, float, float, float } %1274, 0, !dbg !45 + %1594 = lshr exact i32 %28, 2, !dbg !25 + %1595 = shl nuw nsw i32 %27, 1, !dbg !25 + %1596 = getelementptr bfloat, ptr addrspace(1) %4, i64 %19, !dbg !46 + %1597 = getelementptr bfloat, ptr addrspace(1) %3, i64 %19, !dbg !47 + %1598 = insertelement <2 x float> poison, float %5, i64 0, !dbg !48 + %1599 = shufflevector <2 x float> %1598, <2 x float> poison, <2 x i32> zeroinitializer, !dbg !48 + %1600 = insertelement <2 x float> poison, float %1591, i64 0, !dbg !48 + %1601 = insertelement <2 x float> %1600, float %1590, i64 1, !dbg !48 + %1602 = fmul <2 x float> %1599, %1601, !dbg !48 + %1603 = fadd <2 x float> %1602, zeroinitializer, !dbg !49 + %1604 = insertelement <2 x float> poison, float %.0.i548, i64 0, !dbg !50 + %1605 = insertelement <2 x float> %1604, float %.0.i551, i64 1, !dbg !50 + %1606 = fmul <2 x float> %1603, %1605, !dbg !50 + %1607 = insertelement <2 x float> poison, float %1593, i64 0, !dbg !48 + %1608 = insertelement <2 x float> %1607, float %1592, i64 1, !dbg !48 + %1609 = fmul <2 x float> %1599, %1608, !dbg !48 + %1610 = fadd <2 x float> %1609, zeroinitializer, !dbg !49 + %1611 = insertelement <2 x float> poison, float %.0.i, i64 0, !dbg !50 + %1612 = insertelement <2 x float> %1611, float %.0.i545, i64 1, !dbg !50 + %1613 = fmul <2 x float> %1610, %1612, !dbg !50 + %1614 = fptrunc <2 x float> %1613 to <2 x bfloat>, !dbg !51 + %1615 = fptrunc <2 x float> %1606 to <2 x bfloat>, !dbg !51 + %1616 = insertelement <2 x float> poison, float %1587, i64 0, !dbg !48 + %1617 = insertelement <2 x float> %1616, float %1586, i64 1, !dbg !48 + %1618 = fmul <2 x float> %1599, %1617, !dbg !48 + %1619 = fadd <2 x float> %1618, zeroinitializer, !dbg !49 + %1620 = insertelement <2 x float> poison, float %.0.i560, i64 0, !dbg !50 + %1621 = insertelement <2 x float> %1620, float %.0.i563, i64 1, !dbg !50 + %1622 = fmul <2 x float> %1619, %1621, !dbg !50 + %1623 = insertelement <2 x float> poison, float %1589, i64 0, !dbg !48 + %1624 = insertelement <2 x float> %1623, float %1588, i64 1, !dbg !48 + %1625 = fmul <2 x float> %1599, %1624, !dbg !48 + %1626 = fadd <2 x float> %1625, zeroinitializer, !dbg !49 + %1627 = insertelement <2 x float> poison, float %.0.i554, i64 0, !dbg !50 + %1628 = insertelement <2 x float> %1627, float %.0.i557, i64 1, !dbg !50 + %1629 = fmul <2 x float> %1626, %1628, !dbg !50 + %1630 = fptrunc <2 x float> %1629 to <2 x bfloat>, !dbg !51 + %1631 = fptrunc <2 x float> %1622 to <2 x bfloat>, !dbg !51 + %1632 = insertelement <2 x float> poison, float %1575, i64 0, !dbg !48 + %1633 = insertelement <2 x float> %1632, float %1574, i64 1, !dbg !48 + %1634 = fmul <2 x float> %1599, %1633, !dbg !48 + %1635 = fadd <2 x float> %1634, zeroinitializer, !dbg !49 + %1636 = insertelement <2 x float> poison, float %.0.i608, i64 0, !dbg !50 + %1637 = insertelement <2 x float> %1636, float %.0.i611, i64 1, !dbg !50 + %1638 = fmul <2 x float> %1635, %1637, !dbg !50 + %1639 = insertelement <2 x float> poison, float %1577, i64 0, !dbg !48 + %1640 = insertelement <2 x float> %1639, float %1576, i64 1, !dbg !48 + %1641 = fmul <2 x float> %1599, %1640, !dbg !48 + %1642 = fadd <2 x float> %1641, zeroinitializer, !dbg !49 + %1643 = insertelement <2 x float> poison, float %.0.i602, i64 0, !dbg !50 + %1644 = insertelement <2 x float> %1643, float %.0.i605, i64 1, !dbg !50 + %1645 = fmul <2 x float> %1642, %1644, !dbg !50 + %1646 = insertelement <2 x float> poison, float %1583, i64 0, !dbg !48 + %1647 = insertelement <2 x float> %1646, float %1582, i64 1, !dbg !48 + %1648 = fmul <2 x float> %1599, %1647, !dbg !48 + %1649 = fadd <2 x float> %1648, zeroinitializer, !dbg !49 + %1650 = insertelement <2 x float> poison, float %.0.i572, i64 0, !dbg !50 + %1651 = insertelement <2 x float> %1650, float %.0.i575, i64 1, !dbg !50 + %1652 = fmul <2 x float> %1649, %1651, !dbg !50 + %1653 = fptrunc <2 x float> %1652 to <2 x bfloat>, !dbg !51 + %1654 = insertelement <2 x float> poison, float %1585, i64 0, !dbg !48 + %1655 = insertelement <2 x float> %1654, float %1584, i64 1, !dbg !48 + %1656 = fmul <2 x float> %1599, %1655, !dbg !48 + %1657 = fadd <2 x float> %1656, zeroinitializer, !dbg !49 + %1658 = insertelement <2 x float> poison, float %.0.i566, i64 0, !dbg !50 + %1659 = insertelement <2 x float> %1658, float %.0.i569, i64 1, !dbg !50 + %1660 = fmul <2 x float> %1657, %1659, !dbg !50 + %1661 = fptrunc <2 x float> %1660 to <2 x bfloat>, !dbg !51 + %1662 = insertelement <2 x float> poison, float %1579, i64 0, !dbg !48 + %1663 = insertelement <2 x float> %1662, float %1578, i64 1, !dbg !48 + %1664 = fmul <2 x float> %1599, %1663, !dbg !48 + %1665 = fadd <2 x float> %1664, zeroinitializer, !dbg !49 + %1666 = insertelement <2 x float> poison, float %.0.i584, i64 0, !dbg !50 + %1667 = insertelement <2 x float> %1666, float %.0.i587, i64 1, !dbg !50 + %1668 = fmul <2 x float> %1665, %1667, !dbg !50 + %1669 = fptrunc <2 x float> %1668 to <2 x bfloat>, !dbg !51 + %1670 = insertelement <2 x float> poison, float %1581, i64 0, !dbg !48 + %1671 = insertelement <2 x float> %1670, float %1580, i64 1, !dbg !48 + %1672 = fmul <2 x float> %1599, %1671, !dbg !48 + %1673 = fadd <2 x float> %1672, zeroinitializer, !dbg !49 + %1674 = insertelement <2 x float> poison, float %.0.i578, i64 0, !dbg !50 + %1675 = insertelement <2 x float> %1674, float %.0.i581, i64 1, !dbg !50 + %1676 = fmul <2 x float> %1673, %1675, !dbg !50 + %1677 = fptrunc <2 x float> %1676 to <2 x bfloat>, !dbg !51 + %1678 = fptrunc <2 x float> %1645 to <2 x bfloat>, !dbg !51 + %1679 = fptrunc <2 x float> %1638 to <2 x bfloat>, !dbg !51 + %1680 = insertelement <2 x float> poison, float %1571, i64 0, !dbg !48 + %1681 = insertelement <2 x float> %1680, float %1570, i64 1, !dbg !48 + %1682 = fmul <2 x float> %1599, %1681, !dbg !48 + %1683 = fadd <2 x float> %1682, zeroinitializer, !dbg !49 + %1684 = insertelement <2 x float> poison, float %.0.i620, i64 0, !dbg !50 + %1685 = insertelement <2 x float> %1684, float %.0.i623, i64 1, !dbg !50 + %1686 = fmul <2 x float> %1683, %1685, !dbg !50 + %1687 = insertelement <2 x float> poison, float %1573, i64 0, !dbg !48 + %1688 = insertelement <2 x float> %1687, float %1572, i64 1, !dbg !48 + %1689 = fmul <2 x float> %1599, %1688, !dbg !48 + %1690 = fadd <2 x float> %1689, zeroinitializer, !dbg !49 + %1691 = insertelement <2 x float> poison, float %.0.i614, i64 0, !dbg !50 + %1692 = insertelement <2 x float> %1691, float %.0.i617, i64 1, !dbg !50 + %1693 = fmul <2 x float> %1690, %1692, !dbg !50 + %1694 = fptrunc <2 x float> %1693 to <2 x bfloat>, !dbg !51 + %1695 = fptrunc <2 x float> %1686 to <2 x bfloat>, !dbg !51 + %1696 = insertelement <2 x float> poison, float %1567, i64 0, !dbg !48 + %1697 = insertelement <2 x float> %1696, float %1566, i64 1, !dbg !48 + %1698 = fmul <2 x float> %1599, %1697, !dbg !48 + %1699 = fadd <2 x float> %1698, zeroinitializer, !dbg !49 + %1700 = insertelement <2 x float> poison, float %.0.i632, i64 0, !dbg !50 + %1701 = insertelement <2 x float> %1700, float %.0.i635, i64 1, !dbg !50 + %1702 = fmul <2 x float> %1699, %1701, !dbg !50 + %1703 = insertelement <2 x float> poison, float %1569, i64 0, !dbg !48 + %1704 = insertelement <2 x float> %1703, float %1568, i64 1, !dbg !48 + %1705 = fmul <2 x float> %1599, %1704, !dbg !48 + %1706 = fadd <2 x float> %1705, zeroinitializer, !dbg !49 + %1707 = insertelement <2 x float> poison, float %.0.i626, i64 0, !dbg !50 + %1708 = insertelement <2 x float> %1707, float %.0.i629, i64 1, !dbg !50 + %1709 = fmul <2 x float> %1706, %1708, !dbg !50 + %1710 = fptrunc <2 x float> %1709 to <2 x bfloat>, !dbg !51 + %1711 = fptrunc <2 x float> %1702 to <2 x bfloat>, !dbg !51 + %1712 = insertelement <2 x float> poison, float %1563, i64 0, !dbg !48 + %1713 = insertelement <2 x float> %1712, float %1562, i64 1, !dbg !48 + %1714 = fmul <2 x float> %1599, %1713, !dbg !48 + %1715 = fadd <2 x float> %1714, zeroinitializer, !dbg !49 + %1716 = insertelement <2 x float> poison, float %.0.i668, i64 0, !dbg !50 + %1717 = insertelement <2 x float> %1716, float %.0.i671, i64 1, !dbg !50 + %1718 = fmul <2 x float> %1715, %1717, !dbg !50 + %1719 = insertelement <2 x float> poison, float %1565, i64 0, !dbg !48 + %1720 = insertelement <2 x float> %1719, float %1564, i64 1, !dbg !48 + %1721 = fmul <2 x float> %1599, %1720, !dbg !48 + %1722 = fadd <2 x float> %1721, zeroinitializer, !dbg !49 + %1723 = insertelement <2 x float> poison, float %.0.i662, i64 0, !dbg !50 + %1724 = insertelement <2 x float> %1723, float %.0.i665, i64 1, !dbg !50 + %1725 = fmul <2 x float> %1722, %1724, !dbg !50 + %1726 = fptrunc <2 x float> %1725 to <2 x bfloat>, !dbg !51 + %1727 = fptrunc <2 x float> %1718 to <2 x bfloat>, !dbg !51 + %1728 = insertelement <2 x float> poison, float %1559, i64 0, !dbg !48 + %1729 = insertelement <2 x float> %1728, float %1558, i64 1, !dbg !48 + %1730 = fmul <2 x float> %1599, %1729, !dbg !48 + %1731 = fadd <2 x float> %1730, zeroinitializer, !dbg !49 + %1732 = insertelement <2 x float> poison, float %.0.i680, i64 0, !dbg !50 + %1733 = insertelement <2 x float> %1732, float %.0.i683, i64 1, !dbg !50 + %1734 = fmul <2 x float> %1731, %1733, !dbg !50 + %1735 = insertelement <2 x float> poison, float %1561, i64 0, !dbg !48 + %1736 = insertelement <2 x float> %1735, float %1560, i64 1, !dbg !48 + %1737 = fmul <2 x float> %1599, %1736, !dbg !48 + %1738 = fadd <2 x float> %1737, zeroinitializer, !dbg !49 + %1739 = insertelement <2 x float> poison, float %.0.i674, i64 0, !dbg !50 + %1740 = insertelement <2 x float> %1739, float %.0.i677, i64 1, !dbg !50 + %1741 = fmul <2 x float> %1738, %1740, !dbg !50 + %1742 = fptrunc <2 x float> %1741 to <2 x bfloat>, !dbg !51 + %1743 = fptrunc <2 x float> %1734 to <2 x bfloat>, !dbg !51 + %1744 = insertelement <2 x float> poison, float %1555, i64 0, !dbg !48 + %1745 = insertelement <2 x float> %1744, float %1554, i64 1, !dbg !48 + %1746 = fmul <2 x float> %1599, %1745, !dbg !48 + %1747 = fadd <2 x float> %1746, zeroinitializer, !dbg !49 + %1748 = insertelement <2 x float> poison, float %.0.i728, i64 0, !dbg !50 + %1749 = insertelement <2 x float> %1748, float %.0.i731, i64 1, !dbg !50 + %1750 = fmul <2 x float> %1747, %1749, !dbg !50 + %1751 = insertelement <2 x float> poison, float %1557, i64 0, !dbg !48 + %1752 = insertelement <2 x float> %1751, float %1556, i64 1, !dbg !48 + %1753 = fmul <2 x float> %1599, %1752, !dbg !48 + %1754 = fadd <2 x float> %1753, zeroinitializer, !dbg !49 + %1755 = insertelement <2 x float> poison, float %.0.i722, i64 0, !dbg !50 + %1756 = insertelement <2 x float> %1755, float %.0.i725, i64 1, !dbg !50 + %1757 = fmul <2 x float> %1754, %1756, !dbg !50 + %1758 = fptrunc <2 x float> %1757 to <2 x bfloat>, !dbg !51 + %1759 = lshr exact i32 %26, 2, !dbg !25 + %1760 = insertelement <4 x i32> poison, i32 %1759, i64 0, !dbg !25 + %1761 = shufflevector <4 x i32> %1760, <4 x i32> poison, <4 x i32> zeroinitializer, !dbg !25 + %1762 = or disjoint <4 x i32> %1761, , !dbg !25 + %1763 = or disjoint i32 %1759, 24, !dbg !25 + %1764 = or disjoint i32 %1759, 16, !dbg !25 + %1765 = or disjoint i32 %1759, 8, !dbg !25 + %1766 = insertelement <2 x i32> poison, i32 %1759, i64 0, !dbg !52 + %1767 = insertelement <2 x i32> %1766, i32 %1764, i64 1, !dbg !52 + %1768 = or disjoint i32 %1594, %1595, !dbg !25 + %1769 = or disjoint i32 %1768, 1, !dbg !25 + %1770 = or disjoint i32 %1768, 17, !dbg !25 + %1771 = or disjoint i32 %1768, 16, !dbg !25 + %1772 = insertelement <4 x i32> poison, i32 %1768, i64 0, !dbg !25 + %1773 = shufflevector <4 x i32> %1772, <4 x i32> poison, <4 x i32> zeroinitializer, !dbg !25 + %1774 = or disjoint <4 x i32> %1773, , !dbg !25 + %1775 = insertelement <16 x i32> poison, i32 %1759, i64 0, !dbg !26 + %1776 = insertelement <16 x i32> %1775, i32 %1765, i64 1, !dbg !26 + %1777 = insertelement <16 x i32> %1776, i32 %1764, i64 2, !dbg !26 + %1778 = insertelement <16 x i32> %1777, i32 %1763, i64 3, !dbg !26 + %1779 = insertelement <16 x i32> %1778, i32 %1768, i64 8, !dbg !26 + %1780 = insertelement <16 x i32> %1779, i32 %1769, i64 9, !dbg !26 + %1781 = insertelement <16 x i32> %1780, i32 %1771, i64 10, !dbg !26 + %1782 = insertelement <16 x i32> %1781, i32 %1770, i64 11, !dbg !26 + %1783 = shufflevector <4 x i32> %1762, <4 x i32> poison, <16 x i32> , !dbg !26 + %1784 = shufflevector <16 x i32> %1782, <16 x i32> %1783, <16 x i32> , !dbg !26 + %1785 = shufflevector <4 x i32> %1774, <4 x i32> poison, <16 x i32> , !dbg !26 + %1786 = shufflevector <16 x i32> %1784, <16 x i32> %1785, <16 x i32> , !dbg !26 + %1787 = zext <16 x i32> %1786 to <16 x i64>, !dbg !26 + %1788 = or disjoint <16 x i64> %39, %1787, !dbg !26 + %1789 = sext <16 x i32> %43 to <16 x i64>, !dbg !27 + %1790 = icmp slt <16 x i64> %1788, %1789, !dbg !27 + %1791 = shufflevector <16 x i1> %1790, <16 x i1> poison, <2 x i32> , !dbg !53 + %1792 = shufflevector <4 x i32> %1762, <4 x i32> poison, <2 x i32> , !dbg !52 + %1793 = shufflevector <2 x i32> %1766, <2 x i32> poison, <4 x i32> , !dbg !52 + %1794 = shufflevector <4 x i32> %1793, <4 x i32> %1762, <2 x i32> , !dbg !52 + %1795 = shufflevector <16 x i1> %1790, <16 x i1> poison, <2 x i32> , !dbg !53 + %1796 = icmp samesign ule i32 %1759, %1769, !dbg !52 + %1797 = icmp samesign ule i32 %1765, %1769, !dbg !52 + %1798 = icmp samesign ule i32 %1765, %1768, !dbg !52 + %1799 = extractelement <16 x i1> %1790, i64 9, !dbg !53 + %1800 = and i1 %1796, %1799, !dbg !53 + %1801 = insertelement <2 x i32> poison, i32 %1768, i64 0, !dbg !52 + %1802 = insertelement <2 x i32> %1801, i32 %1770, i64 1, !dbg !52 + %1803 = icmp samesign ule <2 x i32> %1767, %1802, !dbg !52 + %1804 = shufflevector <2 x i32> %1801, <2 x i32> poison, <4 x i32> , !dbg !52 + %1805 = shufflevector <4 x i32> %1804, <4 x i32> %1774, <2 x i32> , !dbg !52 + %1806 = icmp samesign ule <2 x i32> %1794, %1805, !dbg !52 + %1807 = insertelement <2 x i32> %1792, i32 %1759, i64 0, !dbg !52 + %1808 = shufflevector <4 x i32> %1774, <4 x i32> poison, <2 x i32> , !dbg !52 + %1809 = insertelement <2 x i32> %1808, i32 %1768, i64 0, !dbg !52 + %1810 = icmp samesign ule <2 x i32> %1807, %1809, !dbg !52 + %1811 = icmp samesign ule i32 %1759, %1768, !dbg !52 + %1812 = extractelement <16 x i1> %1790, i64 8, !dbg !53 + %1813 = and i1 %1811, %1812, !dbg !53 + %1814 = extractelement <16 x i1> %1790, i64 0, !dbg !53 + %1815 = and i1 %1814, %1800, !dbg !53 + %1816 = and i1 %1814, %1813, !dbg !53 + %1817 = and i1 %1797, %1799, !dbg !53 + %1818 = and i1 %1798, %1812, !dbg !53 + %1819 = extractelement <16 x i1> %1790, i64 1, !dbg !53 + %1820 = and i1 %1819, %1817, !dbg !53 + %1821 = and i1 %1819, %1818, !dbg !53 + %1822 = insertelement <2 x i1> poison, i1 %1816, i64 0, !dbg !54 + %1823 = insertelement <2 x i1> %1822, i1 %1815, i64 1, !dbg !54 + %1824 = select <2 x i1> %1823, <2 x bfloat> %1614, <2 x bfloat> zeroinitializer, !dbg !54 + %1825 = insertelement <2 x i1> poison, i1 %1821, i64 0, !dbg !54 + %1826 = insertelement <2 x i1> %1825, i1 %1820, i64 1, !dbg !54 + %1827 = select <2 x i1> %1826, <2 x bfloat> %1615, <2 x bfloat> zeroinitializer, !dbg !54 + %1828 = icmp samesign ule i32 %1763, %1770, !dbg !52 + %1829 = icmp samesign ule i32 %1763, %1771, !dbg !52 + %1830 = extractelement <16 x i1> %1790, i64 11, !dbg !53 + %1831 = and i1 %1814, %1830, !dbg !55 + %1832 = extractelement <16 x i1> %1790, i64 10, !dbg !53 + %1833 = and i1 %1814, %1832, !dbg !55 + %1834 = and i1 %1819, %1830, !dbg !55 + %1835 = and i1 %1819, %1832, !dbg !55 + %1836 = shufflevector <16 x i1> %1790, <16 x i1> poison, <2 x i32> , !dbg !53 + %1837 = and <2 x i1> %1803, %1836, !dbg !53 + %1838 = and <2 x i1> %1791, %1837, !dbg !53 + %1839 = and i1 %1828, %1830, !dbg !53 + %1840 = and i1 %1829, %1832, !dbg !53 + %1841 = extractelement <16 x i1> %1790, i64 3, !dbg !53 + %1842 = and i1 %1841, %1839, !dbg !53 + %1843 = and i1 %1841, %1840, !dbg !53 + %1844 = insertelement <2 x i1> poison, i1 %1833, i64 0, !dbg !54 + %1845 = insertelement <2 x i1> %1844, i1 %1831, i64 1, !dbg !54 + %1846 = select <2 x i1> %1845, <2 x bfloat> %1630, <2 x bfloat> zeroinitializer, !dbg !54 + %1847 = insertelement <2 x i1> poison, i1 %1835, i64 0, !dbg !54 + %1848 = insertelement <2 x i1> %1847, i1 %1834, i64 1, !dbg !54 + %1849 = select <2 x i1> %1848, <2 x bfloat> %1631, <2 x bfloat> zeroinitializer, !dbg !54 + %1850 = select <2 x i1> %1838, <2 x bfloat> %1678, <2 x bfloat> zeroinitializer, !dbg !54 + %1851 = insertelement <2 x i1> poison, i1 %1843, i64 0, !dbg !54 + %1852 = insertelement <2 x i1> %1851, i1 %1842, i64 1, !dbg !54 + %1853 = select <2 x i1> %1852, <2 x bfloat> %1679, <2 x bfloat> zeroinitializer, !dbg !54 + %1854 = icmp samesign ule <4 x i32> %1762, %1774, !dbg !52 + %1855 = extractelement <4 x i1> %1854, i64 1, !dbg !52 + %shift = shufflevector <4 x i32> %1762, <4 x i32> poison, <4 x i32> , !dbg !52 + %1856 = icmp ule <4 x i32> %shift, %1774, !dbg !52 + %1857 = extractelement <4 x i1> %1856, i64 0, !dbg !52 + %1858 = extractelement <16 x i1> %1790, i64 13, !dbg !53 + %1859 = and i1 %1814, %1858, !dbg !55 + %1860 = extractelement <16 x i1> %1790, i64 12, !dbg !53 + %1861 = and i1 %1814, %1860, !dbg !55 + %1862 = and i1 %1819, %1858, !dbg !55 + %1863 = and i1 %1819, %1860, !dbg !55 + %1864 = extractelement <16 x i1> %1790, i64 2, !dbg !55 + %1865 = and i1 %1864, %1858, !dbg !55 + %1866 = and i1 %1864, %1860, !dbg !55 + %1867 = and i1 %1841, %1858, !dbg !55 + %1868 = and i1 %1841, %1860, !dbg !55 + %1869 = shufflevector <16 x i1> %1790, <16 x i1> poison, <2 x i32> , !dbg !53 + %1870 = and <2 x i1> %1806, %1869, !dbg !53 + %1871 = and <2 x i1> %1795, %1870, !dbg !53 + %1872 = and i1 %1855, %1858, !dbg !53 + %1873 = and i1 %1857, %1860, !dbg !53 + %1874 = extractelement <16 x i1> %1790, i64 5, !dbg !53 + %1875 = and i1 %1874, %1872, !dbg !53 + %1876 = and i1 %1874, %1873, !dbg !53 + %1877 = insertelement <2 x i1> poison, i1 %1861, i64 0, !dbg !54 + %1878 = insertelement <2 x i1> %1877, i1 %1859, i64 1, !dbg !54 + %1879 = select <2 x i1> %1878, <2 x bfloat> %1661, <2 x bfloat> zeroinitializer, !dbg !54 + %1880 = insertelement <2 x i1> poison, i1 %1863, i64 0, !dbg !54 + %1881 = insertelement <2 x i1> %1880, i1 %1862, i64 1, !dbg !54 + %1882 = select <2 x i1> %1881, <2 x bfloat> %1653, <2 x bfloat> zeroinitializer, !dbg !54 + %1883 = insertelement <2 x i1> poison, i1 %1866, i64 0, !dbg !54 + %1884 = insertelement <2 x i1> %1883, i1 %1865, i64 1, !dbg !54 + %1885 = select <2 x i1> %1884, <2 x bfloat> %1694, <2 x bfloat> zeroinitializer, !dbg !54 + %1886 = insertelement <2 x i1> poison, i1 %1868, i64 0, !dbg !54 + %1887 = insertelement <2 x i1> %1886, i1 %1867, i64 1, !dbg !54 + %1888 = select <2 x i1> %1887, <2 x bfloat> %1695, <2 x bfloat> zeroinitializer, !dbg !54 + %1889 = select <2 x i1> %1871, <2 x bfloat> %1726, <2 x bfloat> zeroinitializer, !dbg !54 + %1890 = insertelement <2 x i1> poison, i1 %1876, i64 0, !dbg !54 + %1891 = insertelement <2 x i1> %1890, i1 %1875, i64 1, !dbg !54 + %1892 = select <2 x i1> %1891, <2 x bfloat> %1727, <2 x bfloat> zeroinitializer, !dbg !54 + %1893 = icmp samesign ule <4 x i32> %1762, %1774, !dbg !52 + %1894 = extractelement <4 x i1> %1893, i64 3, !dbg !52 + %shift859 = shufflevector <4 x i32> %1762, <4 x i32> poison, <4 x i32> , !dbg !52 + %1895 = icmp ule <4 x i32> %shift859, %1774, !dbg !52 + %1896 = extractelement <4 x i1> %1895, i64 2, !dbg !52 + %1897 = extractelement <16 x i1> %1790, i64 15, !dbg !53 + %1898 = and i1 %1814, %1897, !dbg !55 + %1899 = extractelement <16 x i1> %1790, i64 14, !dbg !53 + %1900 = and i1 %1814, %1899, !dbg !55 + %1901 = and i1 %1819, %1897, !dbg !55 + %1902 = and i1 %1819, %1899, !dbg !55 + %1903 = and i1 %1864, %1897, !dbg !55 + %1904 = and i1 %1864, %1899, !dbg !55 + %1905 = and i1 %1841, %1897, !dbg !55 + %1906 = and i1 %1841, %1899, !dbg !55 + %1907 = extractelement <16 x i1> %1790, i64 4, !dbg !55 + %1908 = and i1 %1907, %1897, !dbg !55 + %1909 = and i1 %1907, %1899, !dbg !55 + %1910 = and i1 %1874, %1897, !dbg !55 + %1911 = and i1 %1874, %1899, !dbg !55 + %1912 = shufflevector <16 x i1> %1790, <16 x i1> poison, <2 x i32> , !dbg !53 + %1913 = and <2 x i1> %1810, %1912, !dbg !53 + %1914 = shufflevector <16 x i1> %1790, <16 x i1> poison, <2 x i32> , !dbg !53 + %1915 = and <2 x i1> %1914, %1913, !dbg !53 + %1916 = and i1 %1894, %1897, !dbg !53 + %1917 = and i1 %1896, %1899, !dbg !53 + %1918 = extractelement <16 x i1> %1790, i64 7, !dbg !53 + %1919 = and i1 %1918, %1916, !dbg !53 + %1920 = and i1 %1918, %1917, !dbg !53 + %1921 = insertelement <2 x i1> poison, i1 %1900, i64 0, !dbg !54 + %1922 = insertelement <2 x i1> %1921, i1 %1898, i64 1, !dbg !54 + %1923 = select <2 x i1> %1922, <2 x bfloat> %1677, <2 x bfloat> zeroinitializer, !dbg !54 + %1924 = insertelement <2 x i1> poison, i1 %1902, i64 0, !dbg !54 + %1925 = insertelement <2 x i1> %1924, i1 %1901, i64 1, !dbg !54 + %1926 = select <2 x i1> %1925, <2 x bfloat> %1669, <2 x bfloat> zeroinitializer, !dbg !54 + %1927 = insertelement <2 x i1> poison, i1 %1904, i64 0, !dbg !54 + %1928 = insertelement <2 x i1> %1927, i1 %1903, i64 1, !dbg !54 + %1929 = select <2 x i1> %1928, <2 x bfloat> %1710, <2 x bfloat> zeroinitializer, !dbg !54 + %1930 = insertelement <2 x i1> poison, i1 %1906, i64 0, !dbg !54 + %1931 = insertelement <2 x i1> %1930, i1 %1905, i64 1, !dbg !54 + %1932 = select <2 x i1> %1931, <2 x bfloat> %1711, <2 x bfloat> zeroinitializer, !dbg !54 + %1933 = insertelement <2 x i1> poison, i1 %1909, i64 0, !dbg !54 + %1934 = insertelement <2 x i1> %1933, i1 %1908, i64 1, !dbg !54 + %1935 = select <2 x i1> %1934, <2 x bfloat> %1742, <2 x bfloat> zeroinitializer, !dbg !54 + %1936 = insertelement <2 x i1> poison, i1 %1911, i64 0, !dbg !54 + %1937 = insertelement <2 x i1> %1936, i1 %1910, i64 1, !dbg !54 + %1938 = select <2 x i1> %1937, <2 x bfloat> %1743, <2 x bfloat> zeroinitializer, !dbg !54 + %1939 = select <2 x i1> %1915, <2 x bfloat> %1758, <2 x bfloat> zeroinitializer, !dbg !54 + %1940 = fptrunc <2 x float> %1750 to <2 x bfloat>, !dbg !51 + %1941 = insertelement <2 x i1> poison, i1 %1920, i64 0, !dbg !54 + %1942 = insertelement <2 x i1> %1941, i1 %1919, i64 1, !dbg !54 + %1943 = select <2 x i1> %1942, <2 x bfloat> %1940, <2 x bfloat> zeroinitializer, !dbg !54 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !51 + %1944 = shl nuw nsw i32 %105, 7, !dbg !51 + %1945 = shl nuw nsw i32 %81, 1, !dbg !51 + %1946 = lshr exact i32 %28, 1, !dbg !51 + %1947 = or disjoint i32 %1944, %52, !dbg !51 + %1948 = or disjoint i32 %1945, %1946, !dbg !51 + %1949 = xor i32 %1947, %1948, !dbg !51 + %1950 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %1949, !dbg !51 + %1951 = bitcast <2 x bfloat> %1824 to i32, !dbg !51 + %1952 = bitcast <2 x bfloat> %1827 to i32, !dbg !51 + %1953 = bitcast <2 x bfloat> %1846 to i32, !dbg !51 + %1954 = bitcast <2 x bfloat> %1849 to i32, !dbg !51 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %1950, i32 %1951, i32 %1952, i32 %1953, i32 %1954), !dbg !51 + %1955 = getelementptr inbounds nuw i8, ptr addrspace(3) %1950, i32 2048, !dbg !51 + %1956 = bitcast <2 x bfloat> %1850 to i32, !dbg !51 + %1957 = bitcast <2 x bfloat> %1853 to i32, !dbg !51 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %1955, i32 0, i32 0, i32 %1956, i32 %1957), !dbg !51 + %1958 = getelementptr inbounds nuw i8, ptr addrspace(3) %1950, i32 4096, !dbg !51 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %1958, i32 0, i32 0, i32 0, i32 0), !dbg !51 + %1959 = getelementptr inbounds nuw i8, ptr addrspace(3) %1950, i32 6144, !dbg !51 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %1959, i32 0, i32 0, i32 0, i32 0), !dbg !51 + %1960 = xor i32 %1949, 64, !dbg !51 + %1961 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %1960, !dbg !51 + %1962 = bitcast <2 x bfloat> %1879 to i32, !dbg !51 + %1963 = bitcast <2 x bfloat> %1882 to i32, !dbg !51 + %1964 = bitcast <2 x bfloat> %1923 to i32, !dbg !51 + %1965 = bitcast <2 x bfloat> %1926 to i32, !dbg !51 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %1961, i32 %1962, i32 %1963, i32 %1964, i32 %1965), !dbg !51 + %1966 = getelementptr inbounds nuw i8, ptr addrspace(3) %1961, i32 2048, !dbg !51 + %1967 = bitcast <2 x bfloat> %1885 to i32, !dbg !51 + %1968 = bitcast <2 x bfloat> %1888 to i32, !dbg !51 + %1969 = bitcast <2 x bfloat> %1929 to i32, !dbg !51 + %1970 = bitcast <2 x bfloat> %1932 to i32, !dbg !51 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %1966, i32 %1967, i32 %1968, i32 %1969, i32 %1970), !dbg !51 + %1971 = getelementptr inbounds nuw i8, ptr addrspace(3) %1961, i32 4096, !dbg !51 + %1972 = bitcast <2 x bfloat> %1889 to i32, !dbg !51 + %1973 = bitcast <2 x bfloat> %1892 to i32, !dbg !51 + %1974 = bitcast <2 x bfloat> %1935 to i32, !dbg !51 + %1975 = bitcast <2 x bfloat> %1938 to i32, !dbg !51 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %1971, i32 %1972, i32 %1973, i32 %1974, i32 %1975), !dbg !51 + %1976 = getelementptr inbounds nuw i8, ptr addrspace(3) %1961, i32 6144, !dbg !51 + %1977 = bitcast <2 x bfloat> %1939 to i32, !dbg !51 + %1978 = bitcast <2 x bfloat> %1943 to i32, !dbg !51 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %1976, i32 0, i32 0, i32 %1977, i32 %1978), !dbg !51 + %1979 = getelementptr bfloat, ptr addrspace(1) %1597, i64 %107, !dbg !56 + %1980 = getelementptr bfloat, ptr addrspace(1) %1597, i64 %109, !dbg !56 + %1981 = getelementptr bfloat, ptr addrspace(1) %1597, i64 %111, !dbg !56 + %1982 = getelementptr bfloat, ptr addrspace(1) %1597, i64 %113, !dbg !56 + %1983 = getelementptr bfloat, ptr addrspace(1) %1597, i64 %115, !dbg !56 + %1984 = getelementptr bfloat, ptr addrspace(1) %1597, i64 %117, !dbg !56 + %1985 = getelementptr bfloat, ptr addrspace(1) %1597, i64 %119, !dbg !56 + %1986 = getelementptr bfloat, ptr addrspace(1) %1597, i64 %121, !dbg !56 + %1987 = getelementptr bfloat, ptr addrspace(1) %1597, i64 %123, !dbg !56 + %1988 = getelementptr bfloat, ptr addrspace(1) %1597, i64 %125, !dbg !56 + %1989 = getelementptr bfloat, ptr addrspace(1) %1597, i64 %127, !dbg !56 + %1990 = getelementptr bfloat, ptr addrspace(1) %1597, i64 %129, !dbg !56 + %1991 = getelementptr bfloat, ptr addrspace(1) %1597, i64 %131, !dbg !56 + %1992 = getelementptr bfloat, ptr addrspace(1) %1597, i64 %133, !dbg !56 + %1993 = getelementptr bfloat, ptr addrspace(1) %1597, i64 %135, !dbg !56 + %1994 = getelementptr bfloat, ptr addrspace(1) %1597, i64 %137, !dbg !56 + %1995 = getelementptr bfloat, ptr addrspace(1) %1979, i64 %154, !dbg !57 + %1996 = getelementptr bfloat, ptr addrspace(1) %1980, i64 %154, !dbg !57 + %1997 = getelementptr bfloat, ptr addrspace(1) %1981, i64 %154, !dbg !57 + %1998 = getelementptr bfloat, ptr addrspace(1) %1982, i64 %154, !dbg !57 + %1999 = getelementptr bfloat, ptr addrspace(1) %1983, i64 %154, !dbg !57 + %2000 = getelementptr bfloat, ptr addrspace(1) %1984, i64 %154, !dbg !57 + %2001 = getelementptr bfloat, ptr addrspace(1) %1985, i64 %154, !dbg !57 + %2002 = getelementptr bfloat, ptr addrspace(1) %1986, i64 %154, !dbg !57 + %2003 = getelementptr bfloat, ptr addrspace(1) %1987, i64 %154, !dbg !57 + %2004 = getelementptr bfloat, ptr addrspace(1) %1988, i64 %154, !dbg !57 + %2005 = getelementptr bfloat, ptr addrspace(1) %1989, i64 %154, !dbg !57 + %2006 = getelementptr bfloat, ptr addrspace(1) %1990, i64 %154, !dbg !57 + %2007 = getelementptr bfloat, ptr addrspace(1) %1991, i64 %154, !dbg !57 + %2008 = getelementptr bfloat, ptr addrspace(1) %1992, i64 %154, !dbg !57 + %2009 = getelementptr bfloat, ptr addrspace(1) %1993, i64 %154, !dbg !57 + %2010 = getelementptr bfloat, ptr addrspace(1) %1994, i64 %154, !dbg !57 + %2011 = getelementptr bfloat, ptr addrspace(1) %1596, i64 %107, !dbg !58 + %2012 = getelementptr bfloat, ptr addrspace(1) %1596, i64 %109, !dbg !58 + %2013 = getelementptr bfloat, ptr addrspace(1) %1596, i64 %111, !dbg !58 + %2014 = getelementptr bfloat, ptr addrspace(1) %1596, i64 %113, !dbg !58 + %2015 = getelementptr bfloat, ptr addrspace(1) %1596, i64 %115, !dbg !58 + %2016 = getelementptr bfloat, ptr addrspace(1) %1596, i64 %117, !dbg !58 + %2017 = getelementptr bfloat, ptr addrspace(1) %1596, i64 %119, !dbg !58 + %2018 = getelementptr bfloat, ptr addrspace(1) %1596, i64 %121, !dbg !58 + %2019 = getelementptr bfloat, ptr addrspace(1) %1596, i64 %123, !dbg !58 + %2020 = getelementptr bfloat, ptr addrspace(1) %1596, i64 %125, !dbg !58 + %2021 = getelementptr bfloat, ptr addrspace(1) %1596, i64 %127, !dbg !58 + %2022 = getelementptr bfloat, ptr addrspace(1) %1596, i64 %129, !dbg !58 + %2023 = getelementptr bfloat, ptr addrspace(1) %1596, i64 %131, !dbg !58 + %2024 = getelementptr bfloat, ptr addrspace(1) %1596, i64 %133, !dbg !58 + %2025 = getelementptr bfloat, ptr addrspace(1) %1596, i64 %135, !dbg !58 + %2026 = getelementptr bfloat, ptr addrspace(1) %1596, i64 %137, !dbg !58 + %2027 = getelementptr bfloat, ptr addrspace(1) %2011, i64 %154, !dbg !59 + %2028 = getelementptr bfloat, ptr addrspace(1) %2012, i64 %154, !dbg !59 + %2029 = getelementptr bfloat, ptr addrspace(1) %2013, i64 %154, !dbg !59 + %2030 = getelementptr bfloat, ptr addrspace(1) %2014, i64 %154, !dbg !59 + %2031 = getelementptr bfloat, ptr addrspace(1) %2015, i64 %154, !dbg !59 + %2032 = getelementptr bfloat, ptr addrspace(1) %2016, i64 %154, !dbg !59 + %2033 = getelementptr bfloat, ptr addrspace(1) %2017, i64 %154, !dbg !59 + %2034 = getelementptr bfloat, ptr addrspace(1) %2018, i64 %154, !dbg !59 + %2035 = getelementptr bfloat, ptr addrspace(1) %2019, i64 %154, !dbg !59 + %2036 = getelementptr bfloat, ptr addrspace(1) %2020, i64 %154, !dbg !59 + %2037 = getelementptr bfloat, ptr addrspace(1) %2021, i64 %154, !dbg !59 + %2038 = getelementptr bfloat, ptr addrspace(1) %2022, i64 %154, !dbg !59 + %2039 = getelementptr bfloat, ptr addrspace(1) %2023, i64 %154, !dbg !59 + %2040 = getelementptr bfloat, ptr addrspace(1) %2024, i64 %154, !dbg !59 + %2041 = getelementptr bfloat, ptr addrspace(1) %2025, i64 %154, !dbg !59 + %2042 = getelementptr bfloat, ptr addrspace(1) %2026, i64 %154, !dbg !59 + %2043 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %1995, i1 %41) #7, !dbg !60 + %2044 = extractvalue { i32, i32, i32, i32 } %2043, 0, !dbg !60 + %2045 = extractvalue { i32, i32, i32, i32 } %2043, 1, !dbg !60 + %2046 = extractvalue { i32, i32, i32, i32 } %2043, 2, !dbg !60 + %2047 = extractvalue { i32, i32, i32, i32 } %2043, 3, !dbg !60 + %2048 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %1996, i1 %193) #7, !dbg !60 + %2049 = extractvalue { i32, i32, i32, i32 } %2048, 0, !dbg !60 + %2050 = extractvalue { i32, i32, i32, i32 } %2048, 1, !dbg !60 + %2051 = extractvalue { i32, i32, i32, i32 } %2048, 2, !dbg !60 + %2052 = extractvalue { i32, i32, i32, i32 } %2048, 3, !dbg !60 + %2053 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %1997, i1 %199) #7, !dbg !60 + %2054 = extractvalue { i32, i32, i32, i32 } %2053, 0, !dbg !60 + %2055 = extractvalue { i32, i32, i32, i32 } %2053, 1, !dbg !60 + %2056 = extractvalue { i32, i32, i32, i32 } %2053, 2, !dbg !60 + %2057 = extractvalue { i32, i32, i32, i32 } %2053, 3, !dbg !60 + %2058 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %1998, i1 %205) #7, !dbg !60 + %2059 = extractvalue { i32, i32, i32, i32 } %2058, 0, !dbg !60 + %2060 = extractvalue { i32, i32, i32, i32 } %2058, 1, !dbg !60 + %2061 = extractvalue { i32, i32, i32, i32 } %2058, 2, !dbg !60 + %2062 = extractvalue { i32, i32, i32, i32 } %2058, 3, !dbg !60 + %2063 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %1999, i1 %211) #7, !dbg !60 + %2064 = extractvalue { i32, i32, i32, i32 } %2063, 0, !dbg !60 + %2065 = extractvalue { i32, i32, i32, i32 } %2063, 1, !dbg !60 + %2066 = extractvalue { i32, i32, i32, i32 } %2063, 2, !dbg !60 + %2067 = extractvalue { i32, i32, i32, i32 } %2063, 3, !dbg !60 + %2068 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %2000, i1 %217) #7, !dbg !60 + %2069 = extractvalue { i32, i32, i32, i32 } %2068, 0, !dbg !60 + %2070 = extractvalue { i32, i32, i32, i32 } %2068, 1, !dbg !60 + %2071 = extractvalue { i32, i32, i32, i32 } %2068, 2, !dbg !60 + %2072 = extractvalue { i32, i32, i32, i32 } %2068, 3, !dbg !60 + %2073 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %2001, i1 %223) #7, !dbg !60 + %2074 = extractvalue { i32, i32, i32, i32 } %2073, 0, !dbg !60 + %2075 = extractvalue { i32, i32, i32, i32 } %2073, 1, !dbg !60 + %2076 = extractvalue { i32, i32, i32, i32 } %2073, 2, !dbg !60 + %2077 = extractvalue { i32, i32, i32, i32 } %2073, 3, !dbg !60 + %2078 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %2002, i1 %229) #7, !dbg !60 + %2079 = extractvalue { i32, i32, i32, i32 } %2078, 0, !dbg !60 + %2080 = extractvalue { i32, i32, i32, i32 } %2078, 1, !dbg !60 + %2081 = extractvalue { i32, i32, i32, i32 } %2078, 2, !dbg !60 + %2082 = extractvalue { i32, i32, i32, i32 } %2078, 3, !dbg !60 + %2083 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %2003, i1 %235) #7, !dbg !60 + %2084 = extractvalue { i32, i32, i32, i32 } %2083, 0, !dbg !60 + %2085 = extractvalue { i32, i32, i32, i32 } %2083, 1, !dbg !60 + %2086 = extractvalue { i32, i32, i32, i32 } %2083, 2, !dbg !60 + %2087 = extractvalue { i32, i32, i32, i32 } %2083, 3, !dbg !60 + %2088 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %2004, i1 %241) #7, !dbg !60 + %2089 = extractvalue { i32, i32, i32, i32 } %2088, 0, !dbg !60 + %2090 = extractvalue { i32, i32, i32, i32 } %2088, 1, !dbg !60 + %2091 = extractvalue { i32, i32, i32, i32 } %2088, 2, !dbg !60 + %2092 = extractvalue { i32, i32, i32, i32 } %2088, 3, !dbg !60 + %2093 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %2005, i1 %247) #7, !dbg !60 + %2094 = extractvalue { i32, i32, i32, i32 } %2093, 0, !dbg !60 + %2095 = extractvalue { i32, i32, i32, i32 } %2093, 1, !dbg !60 + %2096 = extractvalue { i32, i32, i32, i32 } %2093, 2, !dbg !60 + %2097 = extractvalue { i32, i32, i32, i32 } %2093, 3, !dbg !60 + %2098 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %2006, i1 %253) #7, !dbg !60 + %2099 = extractvalue { i32, i32, i32, i32 } %2098, 0, !dbg !60 + %2100 = extractvalue { i32, i32, i32, i32 } %2098, 1, !dbg !60 + %2101 = extractvalue { i32, i32, i32, i32 } %2098, 2, !dbg !60 + %2102 = extractvalue { i32, i32, i32, i32 } %2098, 3, !dbg !60 + %2103 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %2007, i1 %259) #7, !dbg !60 + %2104 = extractvalue { i32, i32, i32, i32 } %2103, 0, !dbg !60 + %2105 = extractvalue { i32, i32, i32, i32 } %2103, 1, !dbg !60 + %2106 = extractvalue { i32, i32, i32, i32 } %2103, 2, !dbg !60 + %2107 = extractvalue { i32, i32, i32, i32 } %2103, 3, !dbg !60 + %2108 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %2008, i1 %265) #7, !dbg !60 + %2109 = extractvalue { i32, i32, i32, i32 } %2108, 0, !dbg !60 + %2110 = extractvalue { i32, i32, i32, i32 } %2108, 1, !dbg !60 + %2111 = extractvalue { i32, i32, i32, i32 } %2108, 2, !dbg !60 + %2112 = extractvalue { i32, i32, i32, i32 } %2108, 3, !dbg !60 + %2113 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %2009, i1 %271) #7, !dbg !60 + %2114 = extractvalue { i32, i32, i32, i32 } %2113, 0, !dbg !60 + %2115 = extractvalue { i32, i32, i32, i32 } %2113, 1, !dbg !60 + %2116 = extractvalue { i32, i32, i32, i32 } %2113, 2, !dbg !60 + %2117 = extractvalue { i32, i32, i32, i32 } %2113, 3, !dbg !60 + %2118 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %2010, i1 %277) #7, !dbg !60 + %2119 = extractvalue { i32, i32, i32, i32 } %2118, 0, !dbg !60 + %2120 = extractvalue { i32, i32, i32, i32 } %2118, 1, !dbg !60 + %2121 = extractvalue { i32, i32, i32, i32 } %2118, 2, !dbg !60 + %2122 = extractvalue { i32, i32, i32, i32 } %2118, 3, !dbg !60 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %286, i32 %2044, i32 %2045, i32 %2046, i32 %2047) #7, !dbg !60 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %288, i32 %2054, i32 %2055, i32 %2056, i32 %2057) #7, !dbg !60 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %290, i32 %2064, i32 %2065, i32 %2066, i32 %2067) #7, !dbg !60 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %292, i32 %2074, i32 %2075, i32 %2076, i32 %2077) #7, !dbg !60 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %294, i32 %2084, i32 %2085, i32 %2086, i32 %2087) #7, !dbg !60 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %296, i32 %2094, i32 %2095, i32 %2096, i32 %2097) #7, !dbg !60 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %298, i32 %2104, i32 %2105, i32 %2106, i32 %2107) #7, !dbg !60 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %300, i32 %2114, i32 %2115, i32 %2116, i32 %2117) #7, !dbg !60 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %302, i32 %2049, i32 %2050, i32 %2051, i32 %2052) #7, !dbg !60 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %304, i32 %2059, i32 %2060, i32 %2061, i32 %2062) #7, !dbg !60 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %306, i32 %2069, i32 %2070, i32 %2071, i32 %2072) #7, !dbg !60 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %308, i32 %2079, i32 %2080, i32 %2081, i32 %2082) #7, !dbg !60 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %310, i32 %2089, i32 %2090, i32 %2091, i32 %2092) #7, !dbg !60 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %312, i32 %2099, i32 %2100, i32 %2101, i32 %2102) #7, !dbg !60 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %314, i32 %2109, i32 %2110, i32 %2111, i32 %2112) #7, !dbg !60 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %316, i32 %2119, i32 %2120, i32 %2121, i32 %2122) #7, !dbg !60 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !51 + %2123 = xor i32 %1947, %81, !dbg !51 + %2124 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %2123, !dbg !51 + %2125 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %2124), !dbg !51 + %2126 = extractvalue { i32, i32, i32, i32 } %2125, 0, !dbg !51 + %2127 = extractvalue { i32, i32, i32, i32 } %2125, 1, !dbg !51 + %2128 = extractvalue { i32, i32, i32, i32 } %2125, 2, !dbg !51 + %2129 = extractvalue { i32, i32, i32, i32 } %2125, 3, !dbg !51 + %2130 = getelementptr inbounds nuw i8, ptr addrspace(3) %2124, i32 2048, !dbg !51 + %2131 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %2130), !dbg !51 + %2132 = extractvalue { i32, i32, i32, i32 } %2131, 0, !dbg !51 + %2133 = extractvalue { i32, i32, i32, i32 } %2131, 1, !dbg !51 + %2134 = extractvalue { i32, i32, i32, i32 } %2131, 2, !dbg !51 + %2135 = extractvalue { i32, i32, i32, i32 } %2131, 3, !dbg !51 + %2136 = getelementptr inbounds nuw i8, ptr addrspace(3) %2124, i32 4096, !dbg !51 + %2137 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %2136), !dbg !51 + %2138 = extractvalue { i32, i32, i32, i32 } %2137, 0, !dbg !51 + %2139 = extractvalue { i32, i32, i32, i32 } %2137, 1, !dbg !51 + %2140 = extractvalue { i32, i32, i32, i32 } %2137, 2, !dbg !51 + %2141 = extractvalue { i32, i32, i32, i32 } %2137, 3, !dbg !51 + %2142 = getelementptr inbounds nuw i8, ptr addrspace(3) %2124, i32 6144, !dbg !51 + %2143 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %2142), !dbg !51 + %2144 = extractvalue { i32, i32, i32, i32 } %2143, 0, !dbg !51 + %2145 = extractvalue { i32, i32, i32, i32 } %2143, 1, !dbg !51 + %2146 = extractvalue { i32, i32, i32, i32 } %2143, 2, !dbg !51 + %2147 = extractvalue { i32, i32, i32, i32 } %2143, 3, !dbg !51 + %2148 = xor i32 %2123, 32, !dbg !51 + %2149 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %2148, !dbg !51 + %2150 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %2149), !dbg !51 + %2151 = extractvalue { i32, i32, i32, i32 } %2150, 0, !dbg !51 + %2152 = extractvalue { i32, i32, i32, i32 } %2150, 1, !dbg !51 + %2153 = extractvalue { i32, i32, i32, i32 } %2150, 2, !dbg !51 + %2154 = extractvalue { i32, i32, i32, i32 } %2150, 3, !dbg !51 + %2155 = getelementptr inbounds nuw i8, ptr addrspace(3) %2149, i32 2048, !dbg !51 + %2156 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %2155), !dbg !51 + %2157 = extractvalue { i32, i32, i32, i32 } %2156, 0, !dbg !51 + %2158 = extractvalue { i32, i32, i32, i32 } %2156, 1, !dbg !51 + %2159 = extractvalue { i32, i32, i32, i32 } %2156, 2, !dbg !51 + %2160 = extractvalue { i32, i32, i32, i32 } %2156, 3, !dbg !51 + %2161 = getelementptr inbounds nuw i8, ptr addrspace(3) %2149, i32 4096, !dbg !51 + %2162 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %2161), !dbg !51 + %2163 = extractvalue { i32, i32, i32, i32 } %2162, 0, !dbg !51 + %2164 = extractvalue { i32, i32, i32, i32 } %2162, 1, !dbg !51 + %2165 = extractvalue { i32, i32, i32, i32 } %2162, 2, !dbg !51 + %2166 = extractvalue { i32, i32, i32, i32 } %2162, 3, !dbg !51 + %2167 = getelementptr inbounds nuw i8, ptr addrspace(3) %2149, i32 6144, !dbg !51 + %2168 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %2167), !dbg !51 + %2169 = extractvalue { i32, i32, i32, i32 } %2168, 0, !dbg !51 + %2170 = extractvalue { i32, i32, i32, i32 } %2168, 1, !dbg !51 + %2171 = extractvalue { i32, i32, i32, i32 } %2168, 2, !dbg !51 + %2172 = extractvalue { i32, i32, i32, i32 } %2168, 3, !dbg !51 + %2173 = xor i32 %2123, 64, !dbg !51 + %2174 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %2173, !dbg !51 + %2175 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %2174), !dbg !51 + %2176 = extractvalue { i32, i32, i32, i32 } %2175, 0, !dbg !51 + %2177 = extractvalue { i32, i32, i32, i32 } %2175, 1, !dbg !51 + %2178 = extractvalue { i32, i32, i32, i32 } %2175, 2, !dbg !51 + %2179 = extractvalue { i32, i32, i32, i32 } %2175, 3, !dbg !51 + %2180 = getelementptr inbounds nuw i8, ptr addrspace(3) %2174, i32 2048, !dbg !51 + %2181 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %2180), !dbg !51 + %2182 = extractvalue { i32, i32, i32, i32 } %2181, 0, !dbg !51 + %2183 = extractvalue { i32, i32, i32, i32 } %2181, 1, !dbg !51 + %2184 = extractvalue { i32, i32, i32, i32 } %2181, 2, !dbg !51 + %2185 = extractvalue { i32, i32, i32, i32 } %2181, 3, !dbg !51 + %2186 = getelementptr inbounds nuw i8, ptr addrspace(3) %2174, i32 4096, !dbg !51 + %2187 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %2186), !dbg !51 + %2188 = extractvalue { i32, i32, i32, i32 } %2187, 0, !dbg !51 + %2189 = extractvalue { i32, i32, i32, i32 } %2187, 1, !dbg !51 + %2190 = extractvalue { i32, i32, i32, i32 } %2187, 2, !dbg !51 + %2191 = extractvalue { i32, i32, i32, i32 } %2187, 3, !dbg !51 + %2192 = getelementptr inbounds nuw i8, ptr addrspace(3) %2174, i32 6144, !dbg !51 + %2193 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %2192), !dbg !51 + %2194 = extractvalue { i32, i32, i32, i32 } %2193, 0, !dbg !51 + %2195 = extractvalue { i32, i32, i32, i32 } %2193, 1, !dbg !51 + %2196 = extractvalue { i32, i32, i32, i32 } %2193, 2, !dbg !51 + %2197 = extractvalue { i32, i32, i32, i32 } %2193, 3, !dbg !51 + %2198 = xor i32 %2123, 96, !dbg !51 + %2199 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %2198, !dbg !51 + %2200 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %2199), !dbg !51 + %2201 = extractvalue { i32, i32, i32, i32 } %2200, 0, !dbg !51 + %2202 = extractvalue { i32, i32, i32, i32 } %2200, 1, !dbg !51 + %2203 = extractvalue { i32, i32, i32, i32 } %2200, 2, !dbg !51 + %2204 = extractvalue { i32, i32, i32, i32 } %2200, 3, !dbg !51 + %2205 = getelementptr inbounds nuw i8, ptr addrspace(3) %2199, i32 2048, !dbg !51 + %2206 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %2205), !dbg !51 + %2207 = extractvalue { i32, i32, i32, i32 } %2206, 0, !dbg !51 + %2208 = extractvalue { i32, i32, i32, i32 } %2206, 1, !dbg !51 + %2209 = extractvalue { i32, i32, i32, i32 } %2206, 2, !dbg !51 + %2210 = extractvalue { i32, i32, i32, i32 } %2206, 3, !dbg !51 + %2211 = getelementptr inbounds nuw i8, ptr addrspace(3) %2199, i32 4096, !dbg !51 + %2212 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %2211), !dbg !51 + %2213 = extractvalue { i32, i32, i32, i32 } %2212, 0, !dbg !51 + %2214 = extractvalue { i32, i32, i32, i32 } %2212, 1, !dbg !51 + %2215 = extractvalue { i32, i32, i32, i32 } %2212, 2, !dbg !51 + %2216 = extractvalue { i32, i32, i32, i32 } %2212, 3, !dbg !51 + %2217 = getelementptr inbounds nuw i8, ptr addrspace(3) %2199, i32 6144, !dbg !51 + %2218 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %2217), !dbg !51 + %2219 = extractvalue { i32, i32, i32, i32 } %2218, 0, !dbg !51 + %2220 = extractvalue { i32, i32, i32, i32 } %2218, 1, !dbg !51 + %2221 = extractvalue { i32, i32, i32, i32 } %2218, 2, !dbg !51 + %2222 = extractvalue { i32, i32, i32, i32 } %2218, 3, !dbg !51 + %2223 = shl nuw nsw i32 %23, 8, !dbg !60 + %2224 = and i32 %2223, 7936, !dbg !60 + %2225 = or disjoint i32 %2224, %52, !dbg !60 + %2226 = xor i32 %2225, %1946, !dbg !60 + %2227 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %2226, !dbg !60 + %2228 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %2227), !dbg !60 + %2229 = extractvalue { i32, i32, i32, i32 } %2228, 0, !dbg !60 + %2230 = extractvalue { i32, i32, i32, i32 } %2228, 1, !dbg !60 + %2231 = extractvalue { i32, i32, i32, i32 } %2228, 2, !dbg !60 + %2232 = extractvalue { i32, i32, i32, i32 } %2228, 3, !dbg !60 + %2233 = getelementptr inbounds nuw i8, ptr addrspace(3) %2227, i32 8192, !dbg !60 + %2234 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %2233), !dbg !60 + %2235 = extractvalue { i32, i32, i32, i32 } %2234, 0, !dbg !60 + %2236 = extractvalue { i32, i32, i32, i32 } %2234, 1, !dbg !60 + %2237 = extractvalue { i32, i32, i32, i32 } %2234, 2, !dbg !60 + %2238 = extractvalue { i32, i32, i32, i32 } %2234, 3, !dbg !60 + %2239 = getelementptr inbounds nuw i8, ptr addrspace(3) %2227, i32 128, !dbg !60 + %2240 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %2239), !dbg !60 + %2241 = extractvalue { i32, i32, i32, i32 } %2240, 0, !dbg !60 + %2242 = extractvalue { i32, i32, i32, i32 } %2240, 1, !dbg !60 + %2243 = extractvalue { i32, i32, i32, i32 } %2240, 2, !dbg !60 + %2244 = extractvalue { i32, i32, i32, i32 } %2240, 3, !dbg !60 + %2245 = getelementptr inbounds nuw i8, ptr addrspace(3) %2227, i32 8320, !dbg !60 + %2246 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %2245), !dbg !60 + %2247 = extractvalue { i32, i32, i32, i32 } %2246, 0, !dbg !60 + %2248 = extractvalue { i32, i32, i32, i32 } %2246, 1, !dbg !60 + %2249 = extractvalue { i32, i32, i32, i32 } %2246, 2, !dbg !60 + %2250 = extractvalue { i32, i32, i32, i32 } %2246, 3, !dbg !60 + %2251 = xor i32 %2226, 32, !dbg !60 + %2252 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %2251, !dbg !60 + %2253 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %2252), !dbg !60 + %2254 = extractvalue { i32, i32, i32, i32 } %2253, 0, !dbg !60 + %2255 = extractvalue { i32, i32, i32, i32 } %2253, 1, !dbg !60 + %2256 = extractvalue { i32, i32, i32, i32 } %2253, 2, !dbg !60 + %2257 = extractvalue { i32, i32, i32, i32 } %2253, 3, !dbg !60 + %2258 = getelementptr inbounds nuw i8, ptr addrspace(3) %2252, i32 8192, !dbg !60 + %2259 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %2258), !dbg !60 + %2260 = extractvalue { i32, i32, i32, i32 } %2259, 0, !dbg !60 + %2261 = extractvalue { i32, i32, i32, i32 } %2259, 1, !dbg !60 + %2262 = extractvalue { i32, i32, i32, i32 } %2259, 2, !dbg !60 + %2263 = extractvalue { i32, i32, i32, i32 } %2259, 3, !dbg !60 + %2264 = getelementptr inbounds nuw i8, ptr addrspace(3) %2252, i32 128, !dbg !60 + %2265 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %2264), !dbg !60 + %2266 = extractvalue { i32, i32, i32, i32 } %2265, 0, !dbg !60 + %2267 = extractvalue { i32, i32, i32, i32 } %2265, 1, !dbg !60 + %2268 = extractvalue { i32, i32, i32, i32 } %2265, 2, !dbg !60 + %2269 = extractvalue { i32, i32, i32, i32 } %2265, 3, !dbg !60 + %2270 = getelementptr inbounds nuw i8, ptr addrspace(3) %2252, i32 8320, !dbg !60 + %2271 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %2270), !dbg !60 + %2272 = extractvalue { i32, i32, i32, i32 } %2271, 0, !dbg !60 + %2273 = extractvalue { i32, i32, i32, i32 } %2271, 1, !dbg !60 + %2274 = extractvalue { i32, i32, i32, i32 } %2271, 2, !dbg !60 + %2275 = extractvalue { i32, i32, i32, i32 } %2271, 3, !dbg !60 + %2276 = xor i32 %2226, 64, !dbg !60 + %2277 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %2276, !dbg !60 + %2278 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %2277), !dbg !60 + %2279 = extractvalue { i32, i32, i32, i32 } %2278, 0, !dbg !60 + %2280 = extractvalue { i32, i32, i32, i32 } %2278, 1, !dbg !60 + %2281 = extractvalue { i32, i32, i32, i32 } %2278, 2, !dbg !60 + %2282 = extractvalue { i32, i32, i32, i32 } %2278, 3, !dbg !60 + %2283 = getelementptr inbounds nuw i8, ptr addrspace(3) %2277, i32 8192, !dbg !60 + %2284 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %2283), !dbg !60 + %2285 = extractvalue { i32, i32, i32, i32 } %2284, 0, !dbg !60 + %2286 = extractvalue { i32, i32, i32, i32 } %2284, 1, !dbg !60 + %2287 = extractvalue { i32, i32, i32, i32 } %2284, 2, !dbg !60 + %2288 = extractvalue { i32, i32, i32, i32 } %2284, 3, !dbg !60 + %2289 = getelementptr inbounds nuw i8, ptr addrspace(3) %2277, i32 128, !dbg !60 + %2290 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %2289), !dbg !60 + %2291 = extractvalue { i32, i32, i32, i32 } %2290, 0, !dbg !60 + %2292 = extractvalue { i32, i32, i32, i32 } %2290, 1, !dbg !60 + %2293 = extractvalue { i32, i32, i32, i32 } %2290, 2, !dbg !60 + %2294 = extractvalue { i32, i32, i32, i32 } %2290, 3, !dbg !60 + %2295 = getelementptr inbounds nuw i8, ptr addrspace(3) %2277, i32 8320, !dbg !60 + %2296 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %2295), !dbg !60 + %2297 = extractvalue { i32, i32, i32, i32 } %2296, 0, !dbg !60 + %2298 = extractvalue { i32, i32, i32, i32 } %2296, 1, !dbg !60 + %2299 = extractvalue { i32, i32, i32, i32 } %2296, 2, !dbg !60 + %2300 = extractvalue { i32, i32, i32, i32 } %2296, 3, !dbg !60 + %2301 = xor i32 %2226, 96, !dbg !60 + %2302 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %2301, !dbg !60 + %2303 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %2302), !dbg !60 + %2304 = extractvalue { i32, i32, i32, i32 } %2303, 0, !dbg !60 + %2305 = extractvalue { i32, i32, i32, i32 } %2303, 1, !dbg !60 + %2306 = extractvalue { i32, i32, i32, i32 } %2303, 2, !dbg !60 + %2307 = extractvalue { i32, i32, i32, i32 } %2303, 3, !dbg !60 + %2308 = getelementptr inbounds nuw i8, ptr addrspace(3) %2302, i32 8192, !dbg !60 + %2309 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %2308), !dbg !60 + %2310 = extractvalue { i32, i32, i32, i32 } %2309, 0, !dbg !60 + %2311 = extractvalue { i32, i32, i32, i32 } %2309, 1, !dbg !60 + %2312 = extractvalue { i32, i32, i32, i32 } %2309, 2, !dbg !60 + %2313 = extractvalue { i32, i32, i32, i32 } %2309, 3, !dbg !60 + %2314 = getelementptr inbounds nuw i8, ptr addrspace(3) %2302, i32 128, !dbg !60 + %2315 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %2314), !dbg !60 + %2316 = extractvalue { i32, i32, i32, i32 } %2315, 0, !dbg !60 + %2317 = extractvalue { i32, i32, i32, i32 } %2315, 1, !dbg !60 + %2318 = extractvalue { i32, i32, i32, i32 } %2315, 2, !dbg !60 + %2319 = extractvalue { i32, i32, i32, i32 } %2315, 3, !dbg !60 + %2320 = getelementptr inbounds nuw i8, ptr addrspace(3) %2302, i32 8320, !dbg !60 + %2321 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %2320), !dbg !60 + %2322 = extractvalue { i32, i32, i32, i32 } %2321, 0, !dbg !60 + %2323 = extractvalue { i32, i32, i32, i32 } %2321, 1, !dbg !60 + %2324 = extractvalue { i32, i32, i32, i32 } %2321, 2, !dbg !60 + %2325 = extractvalue { i32, i32, i32, i32 } %2321, 3, !dbg !60 + %2326 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2126, i32 %2127, i32 %2128, i32 %2129, i32 %2229, i32 %2230) #7, !dbg !61 + %2327 = extractvalue { float, float, float, float } %2326, 0, !dbg !61 + %2328 = extractvalue { float, float, float, float } %2326, 1, !dbg !61 + %2329 = extractvalue { float, float, float, float } %2326, 2, !dbg !61 + %2330 = extractvalue { float, float, float, float } %2326, 3, !dbg !61 + %2331 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2126, i32 %2127, i32 %2128, i32 %2129, i32 %2254, i32 %2255) #7, !dbg !61 + %2332 = extractvalue { float, float, float, float } %2331, 0, !dbg !61 + %2333 = extractvalue { float, float, float, float } %2331, 1, !dbg !61 + %2334 = extractvalue { float, float, float, float } %2331, 2, !dbg !61 + %2335 = extractvalue { float, float, float, float } %2331, 3, !dbg !61 + %2336 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2126, i32 %2127, i32 %2128, i32 %2129, i32 %2279, i32 %2280) #7, !dbg !61 + %2337 = extractvalue { float, float, float, float } %2336, 0, !dbg !61 + %2338 = extractvalue { float, float, float, float } %2336, 1, !dbg !61 + %2339 = extractvalue { float, float, float, float } %2336, 2, !dbg !61 + %2340 = extractvalue { float, float, float, float } %2336, 3, !dbg !61 + %2341 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2126, i32 %2127, i32 %2128, i32 %2129, i32 %2304, i32 %2305) #7, !dbg !61 + %2342 = extractvalue { float, float, float, float } %2341, 0, !dbg !61 + %2343 = extractvalue { float, float, float, float } %2341, 1, !dbg !61 + %2344 = extractvalue { float, float, float, float } %2341, 2, !dbg !61 + %2345 = extractvalue { float, float, float, float } %2341, 3, !dbg !61 + %2346 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2126, i32 %2127, i32 %2128, i32 %2129, i32 %2241, i32 %2242) #7, !dbg !61 + %2347 = extractvalue { float, float, float, float } %2346, 0, !dbg !61 + %2348 = extractvalue { float, float, float, float } %2346, 1, !dbg !61 + %2349 = extractvalue { float, float, float, float } %2346, 2, !dbg !61 + %2350 = extractvalue { float, float, float, float } %2346, 3, !dbg !61 + %2351 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2126, i32 %2127, i32 %2128, i32 %2129, i32 %2266, i32 %2267) #7, !dbg !61 + %2352 = extractvalue { float, float, float, float } %2351, 0, !dbg !61 + %2353 = extractvalue { float, float, float, float } %2351, 1, !dbg !61 + %2354 = extractvalue { float, float, float, float } %2351, 2, !dbg !61 + %2355 = extractvalue { float, float, float, float } %2351, 3, !dbg !61 + %2356 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2126, i32 %2127, i32 %2128, i32 %2129, i32 %2291, i32 %2292) #7, !dbg !61 + %2357 = extractvalue { float, float, float, float } %2356, 0, !dbg !61 + %2358 = extractvalue { float, float, float, float } %2356, 1, !dbg !61 + %2359 = extractvalue { float, float, float, float } %2356, 2, !dbg !61 + %2360 = extractvalue { float, float, float, float } %2356, 3, !dbg !61 + %2361 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2126, i32 %2127, i32 %2128, i32 %2129, i32 %2316, i32 %2317) #7, !dbg !61 + %2362 = extractvalue { float, float, float, float } %2361, 0, !dbg !61 + %2363 = extractvalue { float, float, float, float } %2361, 1, !dbg !61 + %2364 = extractvalue { float, float, float, float } %2361, 2, !dbg !61 + %2365 = extractvalue { float, float, float, float } %2361, 3, !dbg !61 + %2366 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2132, i32 %2133, i32 %2134, i32 %2135, i32 %2229, i32 %2230) #7, !dbg !61 + %2367 = extractvalue { float, float, float, float } %2366, 0, !dbg !61 + %2368 = extractvalue { float, float, float, float } %2366, 1, !dbg !61 + %2369 = extractvalue { float, float, float, float } %2366, 2, !dbg !61 + %2370 = extractvalue { float, float, float, float } %2366, 3, !dbg !61 + %2371 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2132, i32 %2133, i32 %2134, i32 %2135, i32 %2254, i32 %2255) #7, !dbg !61 + %2372 = extractvalue { float, float, float, float } %2371, 0, !dbg !61 + %2373 = extractvalue { float, float, float, float } %2371, 1, !dbg !61 + %2374 = extractvalue { float, float, float, float } %2371, 2, !dbg !61 + %2375 = extractvalue { float, float, float, float } %2371, 3, !dbg !61 + %2376 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2132, i32 %2133, i32 %2134, i32 %2135, i32 %2279, i32 %2280) #7, !dbg !61 + %2377 = extractvalue { float, float, float, float } %2376, 0, !dbg !61 + %2378 = extractvalue { float, float, float, float } %2376, 1, !dbg !61 + %2379 = extractvalue { float, float, float, float } %2376, 2, !dbg !61 + %2380 = extractvalue { float, float, float, float } %2376, 3, !dbg !61 + %2381 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2132, i32 %2133, i32 %2134, i32 %2135, i32 %2304, i32 %2305) #7, !dbg !61 + %2382 = extractvalue { float, float, float, float } %2381, 0, !dbg !61 + %2383 = extractvalue { float, float, float, float } %2381, 1, !dbg !61 + %2384 = extractvalue { float, float, float, float } %2381, 2, !dbg !61 + %2385 = extractvalue { float, float, float, float } %2381, 3, !dbg !61 + %2386 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2132, i32 %2133, i32 %2134, i32 %2135, i32 %2241, i32 %2242) #7, !dbg !61 + %2387 = extractvalue { float, float, float, float } %2386, 0, !dbg !61 + %2388 = extractvalue { float, float, float, float } %2386, 1, !dbg !61 + %2389 = extractvalue { float, float, float, float } %2386, 2, !dbg !61 + %2390 = extractvalue { float, float, float, float } %2386, 3, !dbg !61 + %2391 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2132, i32 %2133, i32 %2134, i32 %2135, i32 %2266, i32 %2267) #7, !dbg !61 + %2392 = extractvalue { float, float, float, float } %2391, 0, !dbg !61 + %2393 = extractvalue { float, float, float, float } %2391, 1, !dbg !61 + %2394 = extractvalue { float, float, float, float } %2391, 2, !dbg !61 + %2395 = extractvalue { float, float, float, float } %2391, 3, !dbg !61 + %2396 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2132, i32 %2133, i32 %2134, i32 %2135, i32 %2291, i32 %2292) #7, !dbg !61 + %2397 = extractvalue { float, float, float, float } %2396, 0, !dbg !61 + %2398 = extractvalue { float, float, float, float } %2396, 1, !dbg !61 + %2399 = extractvalue { float, float, float, float } %2396, 2, !dbg !61 + %2400 = extractvalue { float, float, float, float } %2396, 3, !dbg !61 + %2401 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2132, i32 %2133, i32 %2134, i32 %2135, i32 %2316, i32 %2317) #7, !dbg !61 + %2402 = extractvalue { float, float, float, float } %2401, 0, !dbg !61 + %2403 = extractvalue { float, float, float, float } %2401, 1, !dbg !61 + %2404 = extractvalue { float, float, float, float } %2401, 2, !dbg !61 + %2405 = extractvalue { float, float, float, float } %2401, 3, !dbg !61 + %2406 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2138, i32 %2139, i32 %2140, i32 %2141, i32 %2229, i32 %2230) #7, !dbg !61 + %2407 = extractvalue { float, float, float, float } %2406, 0, !dbg !61 + %2408 = extractvalue { float, float, float, float } %2406, 1, !dbg !61 + %2409 = extractvalue { float, float, float, float } %2406, 2, !dbg !61 + %2410 = extractvalue { float, float, float, float } %2406, 3, !dbg !61 + %2411 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2138, i32 %2139, i32 %2140, i32 %2141, i32 %2254, i32 %2255) #7, !dbg !61 + %2412 = extractvalue { float, float, float, float } %2411, 0, !dbg !61 + %2413 = extractvalue { float, float, float, float } %2411, 1, !dbg !61 + %2414 = extractvalue { float, float, float, float } %2411, 2, !dbg !61 + %2415 = extractvalue { float, float, float, float } %2411, 3, !dbg !61 + %2416 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2138, i32 %2139, i32 %2140, i32 %2141, i32 %2279, i32 %2280) #7, !dbg !61 + %2417 = extractvalue { float, float, float, float } %2416, 0, !dbg !61 + %2418 = extractvalue { float, float, float, float } %2416, 1, !dbg !61 + %2419 = extractvalue { float, float, float, float } %2416, 2, !dbg !61 + %2420 = extractvalue { float, float, float, float } %2416, 3, !dbg !61 + %2421 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2138, i32 %2139, i32 %2140, i32 %2141, i32 %2304, i32 %2305) #7, !dbg !61 + %2422 = extractvalue { float, float, float, float } %2421, 0, !dbg !61 + %2423 = extractvalue { float, float, float, float } %2421, 1, !dbg !61 + %2424 = extractvalue { float, float, float, float } %2421, 2, !dbg !61 + %2425 = extractvalue { float, float, float, float } %2421, 3, !dbg !61 + %2426 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2138, i32 %2139, i32 %2140, i32 %2141, i32 %2241, i32 %2242) #7, !dbg !61 + %2427 = extractvalue { float, float, float, float } %2426, 0, !dbg !61 + %2428 = extractvalue { float, float, float, float } %2426, 1, !dbg !61 + %2429 = extractvalue { float, float, float, float } %2426, 2, !dbg !61 + %2430 = extractvalue { float, float, float, float } %2426, 3, !dbg !61 + %2431 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2138, i32 %2139, i32 %2140, i32 %2141, i32 %2266, i32 %2267) #7, !dbg !61 + %2432 = extractvalue { float, float, float, float } %2431, 0, !dbg !61 + %2433 = extractvalue { float, float, float, float } %2431, 1, !dbg !61 + %2434 = extractvalue { float, float, float, float } %2431, 2, !dbg !61 + %2435 = extractvalue { float, float, float, float } %2431, 3, !dbg !61 + %2436 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2138, i32 %2139, i32 %2140, i32 %2141, i32 %2291, i32 %2292) #7, !dbg !61 + %2437 = extractvalue { float, float, float, float } %2436, 0, !dbg !61 + %2438 = extractvalue { float, float, float, float } %2436, 1, !dbg !61 + %2439 = extractvalue { float, float, float, float } %2436, 2, !dbg !61 + %2440 = extractvalue { float, float, float, float } %2436, 3, !dbg !61 + %2441 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2138, i32 %2139, i32 %2140, i32 %2141, i32 %2316, i32 %2317) #7, !dbg !61 + %2442 = extractvalue { float, float, float, float } %2441, 0, !dbg !61 + %2443 = extractvalue { float, float, float, float } %2441, 1, !dbg !61 + %2444 = extractvalue { float, float, float, float } %2441, 2, !dbg !61 + %2445 = extractvalue { float, float, float, float } %2441, 3, !dbg !61 + %2446 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2144, i32 %2145, i32 %2146, i32 %2147, i32 %2229, i32 %2230) #7, !dbg !61 + %2447 = extractvalue { float, float, float, float } %2446, 0, !dbg !61 + %2448 = extractvalue { float, float, float, float } %2446, 1, !dbg !61 + %2449 = extractvalue { float, float, float, float } %2446, 2, !dbg !61 + %2450 = extractvalue { float, float, float, float } %2446, 3, !dbg !61 + %2451 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2144, i32 %2145, i32 %2146, i32 %2147, i32 %2254, i32 %2255) #7, !dbg !61 + %2452 = extractvalue { float, float, float, float } %2451, 0, !dbg !61 + %2453 = extractvalue { float, float, float, float } %2451, 1, !dbg !61 + %2454 = extractvalue { float, float, float, float } %2451, 2, !dbg !61 + %2455 = extractvalue { float, float, float, float } %2451, 3, !dbg !61 + %2456 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2144, i32 %2145, i32 %2146, i32 %2147, i32 %2279, i32 %2280) #7, !dbg !61 + %2457 = extractvalue { float, float, float, float } %2456, 0, !dbg !61 + %2458 = extractvalue { float, float, float, float } %2456, 1, !dbg !61 + %2459 = extractvalue { float, float, float, float } %2456, 2, !dbg !61 + %2460 = extractvalue { float, float, float, float } %2456, 3, !dbg !61 + %2461 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2144, i32 %2145, i32 %2146, i32 %2147, i32 %2304, i32 %2305) #7, !dbg !61 + %2462 = extractvalue { float, float, float, float } %2461, 0, !dbg !61 + %2463 = extractvalue { float, float, float, float } %2461, 1, !dbg !61 + %2464 = extractvalue { float, float, float, float } %2461, 2, !dbg !61 + %2465 = extractvalue { float, float, float, float } %2461, 3, !dbg !61 + %2466 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2144, i32 %2145, i32 %2146, i32 %2147, i32 %2241, i32 %2242) #7, !dbg !61 + %2467 = extractvalue { float, float, float, float } %2466, 0, !dbg !61 + %2468 = extractvalue { float, float, float, float } %2466, 1, !dbg !61 + %2469 = extractvalue { float, float, float, float } %2466, 2, !dbg !61 + %2470 = extractvalue { float, float, float, float } %2466, 3, !dbg !61 + %2471 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2144, i32 %2145, i32 %2146, i32 %2147, i32 %2266, i32 %2267) #7, !dbg !61 + %2472 = extractvalue { float, float, float, float } %2471, 0, !dbg !61 + %2473 = extractvalue { float, float, float, float } %2471, 1, !dbg !61 + %2474 = extractvalue { float, float, float, float } %2471, 2, !dbg !61 + %2475 = extractvalue { float, float, float, float } %2471, 3, !dbg !61 + %2476 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2144, i32 %2145, i32 %2146, i32 %2147, i32 %2291, i32 %2292) #7, !dbg !61 + %2477 = extractvalue { float, float, float, float } %2476, 0, !dbg !61 + %2478 = extractvalue { float, float, float, float } %2476, 1, !dbg !61 + %2479 = extractvalue { float, float, float, float } %2476, 2, !dbg !61 + %2480 = extractvalue { float, float, float, float } %2476, 3, !dbg !61 + %2481 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2144, i32 %2145, i32 %2146, i32 %2147, i32 %2316, i32 %2317) #7, !dbg !61 + %2482 = extractvalue { float, float, float, float } %2481, 0, !dbg !61 + %2483 = extractvalue { float, float, float, float } %2481, 1, !dbg !61 + %2484 = extractvalue { float, float, float, float } %2481, 2, !dbg !61 + %2485 = extractvalue { float, float, float, float } %2481, 3, !dbg !61 + %2486 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2327, float %2328, float %2329, float %2330, i32 %2151, i32 %2152, i32 %2153, i32 %2154, i32 %2231, i32 %2232) #7, !dbg !61 + %2487 = extractvalue { float, float, float, float } %2486, 0, !dbg !61 + %2488 = extractvalue { float, float, float, float } %2486, 1, !dbg !61 + %2489 = extractvalue { float, float, float, float } %2486, 2, !dbg !61 + %2490 = extractvalue { float, float, float, float } %2486, 3, !dbg !61 + %2491 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2332, float %2333, float %2334, float %2335, i32 %2151, i32 %2152, i32 %2153, i32 %2154, i32 %2256, i32 %2257) #7, !dbg !61 + %2492 = extractvalue { float, float, float, float } %2491, 0, !dbg !61 + %2493 = extractvalue { float, float, float, float } %2491, 1, !dbg !61 + %2494 = extractvalue { float, float, float, float } %2491, 2, !dbg !61 + %2495 = extractvalue { float, float, float, float } %2491, 3, !dbg !61 + %2496 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2337, float %2338, float %2339, float %2340, i32 %2151, i32 %2152, i32 %2153, i32 %2154, i32 %2281, i32 %2282) #7, !dbg !61 + %2497 = extractvalue { float, float, float, float } %2496, 0, !dbg !61 + %2498 = extractvalue { float, float, float, float } %2496, 1, !dbg !61 + %2499 = extractvalue { float, float, float, float } %2496, 2, !dbg !61 + %2500 = extractvalue { float, float, float, float } %2496, 3, !dbg !61 + %2501 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2342, float %2343, float %2344, float %2345, i32 %2151, i32 %2152, i32 %2153, i32 %2154, i32 %2306, i32 %2307) #7, !dbg !61 + %2502 = extractvalue { float, float, float, float } %2501, 0, !dbg !61 + %2503 = extractvalue { float, float, float, float } %2501, 1, !dbg !61 + %2504 = extractvalue { float, float, float, float } %2501, 2, !dbg !61 + %2505 = extractvalue { float, float, float, float } %2501, 3, !dbg !61 + %2506 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2347, float %2348, float %2349, float %2350, i32 %2151, i32 %2152, i32 %2153, i32 %2154, i32 %2243, i32 %2244) #7, !dbg !61 + %2507 = extractvalue { float, float, float, float } %2506, 0, !dbg !61 + %2508 = extractvalue { float, float, float, float } %2506, 1, !dbg !61 + %2509 = extractvalue { float, float, float, float } %2506, 2, !dbg !61 + %2510 = extractvalue { float, float, float, float } %2506, 3, !dbg !61 + %2511 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2352, float %2353, float %2354, float %2355, i32 %2151, i32 %2152, i32 %2153, i32 %2154, i32 %2268, i32 %2269) #7, !dbg !61 + %2512 = extractvalue { float, float, float, float } %2511, 0, !dbg !61 + %2513 = extractvalue { float, float, float, float } %2511, 1, !dbg !61 + %2514 = extractvalue { float, float, float, float } %2511, 2, !dbg !61 + %2515 = extractvalue { float, float, float, float } %2511, 3, !dbg !61 + %2516 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2357, float %2358, float %2359, float %2360, i32 %2151, i32 %2152, i32 %2153, i32 %2154, i32 %2293, i32 %2294) #7, !dbg !61 + %2517 = extractvalue { float, float, float, float } %2516, 0, !dbg !61 + %2518 = extractvalue { float, float, float, float } %2516, 1, !dbg !61 + %2519 = extractvalue { float, float, float, float } %2516, 2, !dbg !61 + %2520 = extractvalue { float, float, float, float } %2516, 3, !dbg !61 + %2521 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2362, float %2363, float %2364, float %2365, i32 %2151, i32 %2152, i32 %2153, i32 %2154, i32 %2318, i32 %2319) #7, !dbg !61 + %2522 = extractvalue { float, float, float, float } %2521, 0, !dbg !61 + %2523 = extractvalue { float, float, float, float } %2521, 1, !dbg !61 + %2524 = extractvalue { float, float, float, float } %2521, 2, !dbg !61 + %2525 = extractvalue { float, float, float, float } %2521, 3, !dbg !61 + %2526 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2367, float %2368, float %2369, float %2370, i32 %2157, i32 %2158, i32 %2159, i32 %2160, i32 %2231, i32 %2232) #7, !dbg !61 + %2527 = extractvalue { float, float, float, float } %2526, 0, !dbg !61 + %2528 = extractvalue { float, float, float, float } %2526, 1, !dbg !61 + %2529 = extractvalue { float, float, float, float } %2526, 2, !dbg !61 + %2530 = extractvalue { float, float, float, float } %2526, 3, !dbg !61 + %2531 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2372, float %2373, float %2374, float %2375, i32 %2157, i32 %2158, i32 %2159, i32 %2160, i32 %2256, i32 %2257) #7, !dbg !61 + %2532 = extractvalue { float, float, float, float } %2531, 0, !dbg !61 + %2533 = extractvalue { float, float, float, float } %2531, 1, !dbg !61 + %2534 = extractvalue { float, float, float, float } %2531, 2, !dbg !61 + %2535 = extractvalue { float, float, float, float } %2531, 3, !dbg !61 + %2536 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2377, float %2378, float %2379, float %2380, i32 %2157, i32 %2158, i32 %2159, i32 %2160, i32 %2281, i32 %2282) #7, !dbg !61 + %2537 = extractvalue { float, float, float, float } %2536, 0, !dbg !61 + %2538 = extractvalue { float, float, float, float } %2536, 1, !dbg !61 + %2539 = extractvalue { float, float, float, float } %2536, 2, !dbg !61 + %2540 = extractvalue { float, float, float, float } %2536, 3, !dbg !61 + %2541 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2382, float %2383, float %2384, float %2385, i32 %2157, i32 %2158, i32 %2159, i32 %2160, i32 %2306, i32 %2307) #7, !dbg !61 + %2542 = extractvalue { float, float, float, float } %2541, 0, !dbg !61 + %2543 = extractvalue { float, float, float, float } %2541, 1, !dbg !61 + %2544 = extractvalue { float, float, float, float } %2541, 2, !dbg !61 + %2545 = extractvalue { float, float, float, float } %2541, 3, !dbg !61 + %2546 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2387, float %2388, float %2389, float %2390, i32 %2157, i32 %2158, i32 %2159, i32 %2160, i32 %2243, i32 %2244) #7, !dbg !61 + %2547 = extractvalue { float, float, float, float } %2546, 0, !dbg !61 + %2548 = extractvalue { float, float, float, float } %2546, 1, !dbg !61 + %2549 = extractvalue { float, float, float, float } %2546, 2, !dbg !61 + %2550 = extractvalue { float, float, float, float } %2546, 3, !dbg !61 + %2551 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2392, float %2393, float %2394, float %2395, i32 %2157, i32 %2158, i32 %2159, i32 %2160, i32 %2268, i32 %2269) #7, !dbg !61 + %2552 = extractvalue { float, float, float, float } %2551, 0, !dbg !61 + %2553 = extractvalue { float, float, float, float } %2551, 1, !dbg !61 + %2554 = extractvalue { float, float, float, float } %2551, 2, !dbg !61 + %2555 = extractvalue { float, float, float, float } %2551, 3, !dbg !61 + %2556 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2397, float %2398, float %2399, float %2400, i32 %2157, i32 %2158, i32 %2159, i32 %2160, i32 %2293, i32 %2294) #7, !dbg !61 + %2557 = extractvalue { float, float, float, float } %2556, 0, !dbg !61 + %2558 = extractvalue { float, float, float, float } %2556, 1, !dbg !61 + %2559 = extractvalue { float, float, float, float } %2556, 2, !dbg !61 + %2560 = extractvalue { float, float, float, float } %2556, 3, !dbg !61 + %2561 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2402, float %2403, float %2404, float %2405, i32 %2157, i32 %2158, i32 %2159, i32 %2160, i32 %2318, i32 %2319) #7, !dbg !61 + %2562 = extractvalue { float, float, float, float } %2561, 0, !dbg !61 + %2563 = extractvalue { float, float, float, float } %2561, 1, !dbg !61 + %2564 = extractvalue { float, float, float, float } %2561, 2, !dbg !61 + %2565 = extractvalue { float, float, float, float } %2561, 3, !dbg !61 + %2566 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2407, float %2408, float %2409, float %2410, i32 %2163, i32 %2164, i32 %2165, i32 %2166, i32 %2231, i32 %2232) #7, !dbg !61 + %2567 = extractvalue { float, float, float, float } %2566, 0, !dbg !61 + %2568 = extractvalue { float, float, float, float } %2566, 1, !dbg !61 + %2569 = extractvalue { float, float, float, float } %2566, 2, !dbg !61 + %2570 = extractvalue { float, float, float, float } %2566, 3, !dbg !61 + %2571 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2412, float %2413, float %2414, float %2415, i32 %2163, i32 %2164, i32 %2165, i32 %2166, i32 %2256, i32 %2257) #7, !dbg !61 + %2572 = extractvalue { float, float, float, float } %2571, 0, !dbg !61 + %2573 = extractvalue { float, float, float, float } %2571, 1, !dbg !61 + %2574 = extractvalue { float, float, float, float } %2571, 2, !dbg !61 + %2575 = extractvalue { float, float, float, float } %2571, 3, !dbg !61 + %2576 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2417, float %2418, float %2419, float %2420, i32 %2163, i32 %2164, i32 %2165, i32 %2166, i32 %2281, i32 %2282) #7, !dbg !61 + %2577 = extractvalue { float, float, float, float } %2576, 0, !dbg !61 + %2578 = extractvalue { float, float, float, float } %2576, 1, !dbg !61 + %2579 = extractvalue { float, float, float, float } %2576, 2, !dbg !61 + %2580 = extractvalue { float, float, float, float } %2576, 3, !dbg !61 + %2581 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2422, float %2423, float %2424, float %2425, i32 %2163, i32 %2164, i32 %2165, i32 %2166, i32 %2306, i32 %2307) #7, !dbg !61 + %2582 = extractvalue { float, float, float, float } %2581, 0, !dbg !61 + %2583 = extractvalue { float, float, float, float } %2581, 1, !dbg !61 + %2584 = extractvalue { float, float, float, float } %2581, 2, !dbg !61 + %2585 = extractvalue { float, float, float, float } %2581, 3, !dbg !61 + %2586 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2427, float %2428, float %2429, float %2430, i32 %2163, i32 %2164, i32 %2165, i32 %2166, i32 %2243, i32 %2244) #7, !dbg !61 + %2587 = extractvalue { float, float, float, float } %2586, 0, !dbg !61 + %2588 = extractvalue { float, float, float, float } %2586, 1, !dbg !61 + %2589 = extractvalue { float, float, float, float } %2586, 2, !dbg !61 + %2590 = extractvalue { float, float, float, float } %2586, 3, !dbg !61 + %2591 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2432, float %2433, float %2434, float %2435, i32 %2163, i32 %2164, i32 %2165, i32 %2166, i32 %2268, i32 %2269) #7, !dbg !61 + %2592 = extractvalue { float, float, float, float } %2591, 0, !dbg !61 + %2593 = extractvalue { float, float, float, float } %2591, 1, !dbg !61 + %2594 = extractvalue { float, float, float, float } %2591, 2, !dbg !61 + %2595 = extractvalue { float, float, float, float } %2591, 3, !dbg !61 + %2596 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2437, float %2438, float %2439, float %2440, i32 %2163, i32 %2164, i32 %2165, i32 %2166, i32 %2293, i32 %2294) #7, !dbg !61 + %2597 = extractvalue { float, float, float, float } %2596, 0, !dbg !61 + %2598 = extractvalue { float, float, float, float } %2596, 1, !dbg !61 + %2599 = extractvalue { float, float, float, float } %2596, 2, !dbg !61 + %2600 = extractvalue { float, float, float, float } %2596, 3, !dbg !61 + %2601 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2442, float %2443, float %2444, float %2445, i32 %2163, i32 %2164, i32 %2165, i32 %2166, i32 %2318, i32 %2319) #7, !dbg !61 + %2602 = extractvalue { float, float, float, float } %2601, 0, !dbg !61 + %2603 = extractvalue { float, float, float, float } %2601, 1, !dbg !61 + %2604 = extractvalue { float, float, float, float } %2601, 2, !dbg !61 + %2605 = extractvalue { float, float, float, float } %2601, 3, !dbg !61 + %2606 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2447, float %2448, float %2449, float %2450, i32 %2169, i32 %2170, i32 %2171, i32 %2172, i32 %2231, i32 %2232) #7, !dbg !61 + %2607 = extractvalue { float, float, float, float } %2606, 0, !dbg !61 + %2608 = extractvalue { float, float, float, float } %2606, 1, !dbg !61 + %2609 = extractvalue { float, float, float, float } %2606, 2, !dbg !61 + %2610 = extractvalue { float, float, float, float } %2606, 3, !dbg !61 + %2611 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2452, float %2453, float %2454, float %2455, i32 %2169, i32 %2170, i32 %2171, i32 %2172, i32 %2256, i32 %2257) #7, !dbg !61 + %2612 = extractvalue { float, float, float, float } %2611, 0, !dbg !61 + %2613 = extractvalue { float, float, float, float } %2611, 1, !dbg !61 + %2614 = extractvalue { float, float, float, float } %2611, 2, !dbg !61 + %2615 = extractvalue { float, float, float, float } %2611, 3, !dbg !61 + %2616 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2457, float %2458, float %2459, float %2460, i32 %2169, i32 %2170, i32 %2171, i32 %2172, i32 %2281, i32 %2282) #7, !dbg !61 + %2617 = extractvalue { float, float, float, float } %2616, 0, !dbg !61 + %2618 = extractvalue { float, float, float, float } %2616, 1, !dbg !61 + %2619 = extractvalue { float, float, float, float } %2616, 2, !dbg !61 + %2620 = extractvalue { float, float, float, float } %2616, 3, !dbg !61 + %2621 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2462, float %2463, float %2464, float %2465, i32 %2169, i32 %2170, i32 %2171, i32 %2172, i32 %2306, i32 %2307) #7, !dbg !61 + %2622 = extractvalue { float, float, float, float } %2621, 0, !dbg !61 + %2623 = extractvalue { float, float, float, float } %2621, 1, !dbg !61 + %2624 = extractvalue { float, float, float, float } %2621, 2, !dbg !61 + %2625 = extractvalue { float, float, float, float } %2621, 3, !dbg !61 + %2626 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2467, float %2468, float %2469, float %2470, i32 %2169, i32 %2170, i32 %2171, i32 %2172, i32 %2243, i32 %2244) #7, !dbg !61 + %2627 = extractvalue { float, float, float, float } %2626, 0, !dbg !61 + %2628 = extractvalue { float, float, float, float } %2626, 1, !dbg !61 + %2629 = extractvalue { float, float, float, float } %2626, 2, !dbg !61 + %2630 = extractvalue { float, float, float, float } %2626, 3, !dbg !61 + %2631 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2472, float %2473, float %2474, float %2475, i32 %2169, i32 %2170, i32 %2171, i32 %2172, i32 %2268, i32 %2269) #7, !dbg !61 + %2632 = extractvalue { float, float, float, float } %2631, 0, !dbg !61 + %2633 = extractvalue { float, float, float, float } %2631, 1, !dbg !61 + %2634 = extractvalue { float, float, float, float } %2631, 2, !dbg !61 + %2635 = extractvalue { float, float, float, float } %2631, 3, !dbg !61 + %2636 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2477, float %2478, float %2479, float %2480, i32 %2169, i32 %2170, i32 %2171, i32 %2172, i32 %2293, i32 %2294) #7, !dbg !61 + %2637 = extractvalue { float, float, float, float } %2636, 0, !dbg !61 + %2638 = extractvalue { float, float, float, float } %2636, 1, !dbg !61 + %2639 = extractvalue { float, float, float, float } %2636, 2, !dbg !61 + %2640 = extractvalue { float, float, float, float } %2636, 3, !dbg !61 + %2641 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2482, float %2483, float %2484, float %2485, i32 %2169, i32 %2170, i32 %2171, i32 %2172, i32 %2318, i32 %2319) #7, !dbg !61 + %2642 = extractvalue { float, float, float, float } %2641, 0, !dbg !61 + %2643 = extractvalue { float, float, float, float } %2641, 1, !dbg !61 + %2644 = extractvalue { float, float, float, float } %2641, 2, !dbg !61 + %2645 = extractvalue { float, float, float, float } %2641, 3, !dbg !61 + %2646 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2487, float %2488, float %2489, float %2490, i32 %2176, i32 %2177, i32 %2178, i32 %2179, i32 %2235, i32 %2236) #7, !dbg !61 + %2647 = extractvalue { float, float, float, float } %2646, 0, !dbg !61 + %2648 = extractvalue { float, float, float, float } %2646, 1, !dbg !61 + %2649 = extractvalue { float, float, float, float } %2646, 2, !dbg !61 + %2650 = extractvalue { float, float, float, float } %2646, 3, !dbg !61 + %2651 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2492, float %2493, float %2494, float %2495, i32 %2176, i32 %2177, i32 %2178, i32 %2179, i32 %2260, i32 %2261) #7, !dbg !61 + %2652 = extractvalue { float, float, float, float } %2651, 0, !dbg !61 + %2653 = extractvalue { float, float, float, float } %2651, 1, !dbg !61 + %2654 = extractvalue { float, float, float, float } %2651, 2, !dbg !61 + %2655 = extractvalue { float, float, float, float } %2651, 3, !dbg !61 + %2656 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2497, float %2498, float %2499, float %2500, i32 %2176, i32 %2177, i32 %2178, i32 %2179, i32 %2285, i32 %2286) #7, !dbg !61 + %2657 = extractvalue { float, float, float, float } %2656, 0, !dbg !61 + %2658 = extractvalue { float, float, float, float } %2656, 1, !dbg !61 + %2659 = extractvalue { float, float, float, float } %2656, 2, !dbg !61 + %2660 = extractvalue { float, float, float, float } %2656, 3, !dbg !61 + %2661 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2502, float %2503, float %2504, float %2505, i32 %2176, i32 %2177, i32 %2178, i32 %2179, i32 %2310, i32 %2311) #7, !dbg !61 + %2662 = extractvalue { float, float, float, float } %2661, 0, !dbg !61 + %2663 = extractvalue { float, float, float, float } %2661, 1, !dbg !61 + %2664 = extractvalue { float, float, float, float } %2661, 2, !dbg !61 + %2665 = extractvalue { float, float, float, float } %2661, 3, !dbg !61 + %2666 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2507, float %2508, float %2509, float %2510, i32 %2176, i32 %2177, i32 %2178, i32 %2179, i32 %2247, i32 %2248) #7, !dbg !61 + %2667 = extractvalue { float, float, float, float } %2666, 0, !dbg !61 + %2668 = extractvalue { float, float, float, float } %2666, 1, !dbg !61 + %2669 = extractvalue { float, float, float, float } %2666, 2, !dbg !61 + %2670 = extractvalue { float, float, float, float } %2666, 3, !dbg !61 + %2671 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2512, float %2513, float %2514, float %2515, i32 %2176, i32 %2177, i32 %2178, i32 %2179, i32 %2272, i32 %2273) #7, !dbg !61 + %2672 = extractvalue { float, float, float, float } %2671, 0, !dbg !61 + %2673 = extractvalue { float, float, float, float } %2671, 1, !dbg !61 + %2674 = extractvalue { float, float, float, float } %2671, 2, !dbg !61 + %2675 = extractvalue { float, float, float, float } %2671, 3, !dbg !61 + %2676 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2517, float %2518, float %2519, float %2520, i32 %2176, i32 %2177, i32 %2178, i32 %2179, i32 %2297, i32 %2298) #7, !dbg !61 + %2677 = extractvalue { float, float, float, float } %2676, 0, !dbg !61 + %2678 = extractvalue { float, float, float, float } %2676, 1, !dbg !61 + %2679 = extractvalue { float, float, float, float } %2676, 2, !dbg !61 + %2680 = extractvalue { float, float, float, float } %2676, 3, !dbg !61 + %2681 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2522, float %2523, float %2524, float %2525, i32 %2176, i32 %2177, i32 %2178, i32 %2179, i32 %2322, i32 %2323) #7, !dbg !61 + %2682 = extractvalue { float, float, float, float } %2681, 0, !dbg !61 + %2683 = extractvalue { float, float, float, float } %2681, 1, !dbg !61 + %2684 = extractvalue { float, float, float, float } %2681, 2, !dbg !61 + %2685 = extractvalue { float, float, float, float } %2681, 3, !dbg !61 + %2686 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2527, float %2528, float %2529, float %2530, i32 %2182, i32 %2183, i32 %2184, i32 %2185, i32 %2235, i32 %2236) #7, !dbg !61 + %2687 = extractvalue { float, float, float, float } %2686, 0, !dbg !61 + %2688 = extractvalue { float, float, float, float } %2686, 1, !dbg !61 + %2689 = extractvalue { float, float, float, float } %2686, 2, !dbg !61 + %2690 = extractvalue { float, float, float, float } %2686, 3, !dbg !61 + %2691 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2532, float %2533, float %2534, float %2535, i32 %2182, i32 %2183, i32 %2184, i32 %2185, i32 %2260, i32 %2261) #7, !dbg !61 + %2692 = extractvalue { float, float, float, float } %2691, 0, !dbg !61 + %2693 = extractvalue { float, float, float, float } %2691, 1, !dbg !61 + %2694 = extractvalue { float, float, float, float } %2691, 2, !dbg !61 + %2695 = extractvalue { float, float, float, float } %2691, 3, !dbg !61 + %2696 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2537, float %2538, float %2539, float %2540, i32 %2182, i32 %2183, i32 %2184, i32 %2185, i32 %2285, i32 %2286) #7, !dbg !61 + %2697 = extractvalue { float, float, float, float } %2696, 0, !dbg !61 + %2698 = extractvalue { float, float, float, float } %2696, 1, !dbg !61 + %2699 = extractvalue { float, float, float, float } %2696, 2, !dbg !61 + %2700 = extractvalue { float, float, float, float } %2696, 3, !dbg !61 + %2701 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2542, float %2543, float %2544, float %2545, i32 %2182, i32 %2183, i32 %2184, i32 %2185, i32 %2310, i32 %2311) #7, !dbg !61 + %2702 = extractvalue { float, float, float, float } %2701, 0, !dbg !61 + %2703 = extractvalue { float, float, float, float } %2701, 1, !dbg !61 + %2704 = extractvalue { float, float, float, float } %2701, 2, !dbg !61 + %2705 = extractvalue { float, float, float, float } %2701, 3, !dbg !61 + %2706 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2547, float %2548, float %2549, float %2550, i32 %2182, i32 %2183, i32 %2184, i32 %2185, i32 %2247, i32 %2248) #7, !dbg !61 + %2707 = extractvalue { float, float, float, float } %2706, 0, !dbg !61 + %2708 = extractvalue { float, float, float, float } %2706, 1, !dbg !61 + %2709 = extractvalue { float, float, float, float } %2706, 2, !dbg !61 + %2710 = extractvalue { float, float, float, float } %2706, 3, !dbg !61 + %2711 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2552, float %2553, float %2554, float %2555, i32 %2182, i32 %2183, i32 %2184, i32 %2185, i32 %2272, i32 %2273) #7, !dbg !61 + %2712 = extractvalue { float, float, float, float } %2711, 0, !dbg !61 + %2713 = extractvalue { float, float, float, float } %2711, 1, !dbg !61 + %2714 = extractvalue { float, float, float, float } %2711, 2, !dbg !61 + %2715 = extractvalue { float, float, float, float } %2711, 3, !dbg !61 + %2716 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2557, float %2558, float %2559, float %2560, i32 %2182, i32 %2183, i32 %2184, i32 %2185, i32 %2297, i32 %2298) #7, !dbg !61 + %2717 = extractvalue { float, float, float, float } %2716, 0, !dbg !61 + %2718 = extractvalue { float, float, float, float } %2716, 1, !dbg !61 + %2719 = extractvalue { float, float, float, float } %2716, 2, !dbg !61 + %2720 = extractvalue { float, float, float, float } %2716, 3, !dbg !61 + %2721 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2562, float %2563, float %2564, float %2565, i32 %2182, i32 %2183, i32 %2184, i32 %2185, i32 %2322, i32 %2323) #7, !dbg !61 + %2722 = extractvalue { float, float, float, float } %2721, 0, !dbg !61 + %2723 = extractvalue { float, float, float, float } %2721, 1, !dbg !61 + %2724 = extractvalue { float, float, float, float } %2721, 2, !dbg !61 + %2725 = extractvalue { float, float, float, float } %2721, 3, !dbg !61 + %2726 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2567, float %2568, float %2569, float %2570, i32 %2188, i32 %2189, i32 %2190, i32 %2191, i32 %2235, i32 %2236) #7, !dbg !61 + %2727 = extractvalue { float, float, float, float } %2726, 0, !dbg !61 + %2728 = extractvalue { float, float, float, float } %2726, 1, !dbg !61 + %2729 = extractvalue { float, float, float, float } %2726, 2, !dbg !61 + %2730 = extractvalue { float, float, float, float } %2726, 3, !dbg !61 + %2731 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2572, float %2573, float %2574, float %2575, i32 %2188, i32 %2189, i32 %2190, i32 %2191, i32 %2260, i32 %2261) #7, !dbg !61 + %2732 = extractvalue { float, float, float, float } %2731, 0, !dbg !61 + %2733 = extractvalue { float, float, float, float } %2731, 1, !dbg !61 + %2734 = extractvalue { float, float, float, float } %2731, 2, !dbg !61 + %2735 = extractvalue { float, float, float, float } %2731, 3, !dbg !61 + %2736 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2577, float %2578, float %2579, float %2580, i32 %2188, i32 %2189, i32 %2190, i32 %2191, i32 %2285, i32 %2286) #7, !dbg !61 + %2737 = extractvalue { float, float, float, float } %2736, 0, !dbg !61 + %2738 = extractvalue { float, float, float, float } %2736, 1, !dbg !61 + %2739 = extractvalue { float, float, float, float } %2736, 2, !dbg !61 + %2740 = extractvalue { float, float, float, float } %2736, 3, !dbg !61 + %2741 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2582, float %2583, float %2584, float %2585, i32 %2188, i32 %2189, i32 %2190, i32 %2191, i32 %2310, i32 %2311) #7, !dbg !61 + %2742 = extractvalue { float, float, float, float } %2741, 0, !dbg !61 + %2743 = extractvalue { float, float, float, float } %2741, 1, !dbg !61 + %2744 = extractvalue { float, float, float, float } %2741, 2, !dbg !61 + %2745 = extractvalue { float, float, float, float } %2741, 3, !dbg !61 + %2746 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2587, float %2588, float %2589, float %2590, i32 %2188, i32 %2189, i32 %2190, i32 %2191, i32 %2247, i32 %2248) #7, !dbg !61 + %2747 = extractvalue { float, float, float, float } %2746, 0, !dbg !61 + %2748 = extractvalue { float, float, float, float } %2746, 1, !dbg !61 + %2749 = extractvalue { float, float, float, float } %2746, 2, !dbg !61 + %2750 = extractvalue { float, float, float, float } %2746, 3, !dbg !61 + %2751 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2592, float %2593, float %2594, float %2595, i32 %2188, i32 %2189, i32 %2190, i32 %2191, i32 %2272, i32 %2273) #7, !dbg !61 + %2752 = extractvalue { float, float, float, float } %2751, 0, !dbg !61 + %2753 = extractvalue { float, float, float, float } %2751, 1, !dbg !61 + %2754 = extractvalue { float, float, float, float } %2751, 2, !dbg !61 + %2755 = extractvalue { float, float, float, float } %2751, 3, !dbg !61 + %2756 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2597, float %2598, float %2599, float %2600, i32 %2188, i32 %2189, i32 %2190, i32 %2191, i32 %2297, i32 %2298) #7, !dbg !61 + %2757 = extractvalue { float, float, float, float } %2756, 0, !dbg !61 + %2758 = extractvalue { float, float, float, float } %2756, 1, !dbg !61 + %2759 = extractvalue { float, float, float, float } %2756, 2, !dbg !61 + %2760 = extractvalue { float, float, float, float } %2756, 3, !dbg !61 + %2761 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2602, float %2603, float %2604, float %2605, i32 %2188, i32 %2189, i32 %2190, i32 %2191, i32 %2322, i32 %2323) #7, !dbg !61 + %2762 = extractvalue { float, float, float, float } %2761, 0, !dbg !61 + %2763 = extractvalue { float, float, float, float } %2761, 1, !dbg !61 + %2764 = extractvalue { float, float, float, float } %2761, 2, !dbg !61 + %2765 = extractvalue { float, float, float, float } %2761, 3, !dbg !61 + %2766 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2607, float %2608, float %2609, float %2610, i32 %2194, i32 %2195, i32 %2196, i32 %2197, i32 %2235, i32 %2236) #7, !dbg !61 + %2767 = extractvalue { float, float, float, float } %2766, 0, !dbg !61 + %2768 = extractvalue { float, float, float, float } %2766, 1, !dbg !61 + %2769 = extractvalue { float, float, float, float } %2766, 2, !dbg !61 + %2770 = extractvalue { float, float, float, float } %2766, 3, !dbg !61 + %2771 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2612, float %2613, float %2614, float %2615, i32 %2194, i32 %2195, i32 %2196, i32 %2197, i32 %2260, i32 %2261) #7, !dbg !61 + %2772 = extractvalue { float, float, float, float } %2771, 0, !dbg !61 + %2773 = extractvalue { float, float, float, float } %2771, 1, !dbg !61 + %2774 = extractvalue { float, float, float, float } %2771, 2, !dbg !61 + %2775 = extractvalue { float, float, float, float } %2771, 3, !dbg !61 + %2776 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2617, float %2618, float %2619, float %2620, i32 %2194, i32 %2195, i32 %2196, i32 %2197, i32 %2285, i32 %2286) #7, !dbg !61 + %2777 = extractvalue { float, float, float, float } %2776, 0, !dbg !61 + %2778 = extractvalue { float, float, float, float } %2776, 1, !dbg !61 + %2779 = extractvalue { float, float, float, float } %2776, 2, !dbg !61 + %2780 = extractvalue { float, float, float, float } %2776, 3, !dbg !61 + %2781 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2622, float %2623, float %2624, float %2625, i32 %2194, i32 %2195, i32 %2196, i32 %2197, i32 %2310, i32 %2311) #7, !dbg !61 + %2782 = extractvalue { float, float, float, float } %2781, 0, !dbg !61 + %2783 = extractvalue { float, float, float, float } %2781, 1, !dbg !61 + %2784 = extractvalue { float, float, float, float } %2781, 2, !dbg !61 + %2785 = extractvalue { float, float, float, float } %2781, 3, !dbg !61 + %2786 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2627, float %2628, float %2629, float %2630, i32 %2194, i32 %2195, i32 %2196, i32 %2197, i32 %2247, i32 %2248) #7, !dbg !61 + %2787 = extractvalue { float, float, float, float } %2786, 0, !dbg !61 + %2788 = extractvalue { float, float, float, float } %2786, 1, !dbg !61 + %2789 = extractvalue { float, float, float, float } %2786, 2, !dbg !61 + %2790 = extractvalue { float, float, float, float } %2786, 3, !dbg !61 + %2791 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2632, float %2633, float %2634, float %2635, i32 %2194, i32 %2195, i32 %2196, i32 %2197, i32 %2272, i32 %2273) #7, !dbg !61 + %2792 = extractvalue { float, float, float, float } %2791, 0, !dbg !61 + %2793 = extractvalue { float, float, float, float } %2791, 1, !dbg !61 + %2794 = extractvalue { float, float, float, float } %2791, 2, !dbg !61 + %2795 = extractvalue { float, float, float, float } %2791, 3, !dbg !61 + %2796 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2637, float %2638, float %2639, float %2640, i32 %2194, i32 %2195, i32 %2196, i32 %2197, i32 %2297, i32 %2298) #7, !dbg !61 + %2797 = extractvalue { float, float, float, float } %2796, 0, !dbg !61 + %2798 = extractvalue { float, float, float, float } %2796, 1, !dbg !61 + %2799 = extractvalue { float, float, float, float } %2796, 2, !dbg !61 + %2800 = extractvalue { float, float, float, float } %2796, 3, !dbg !61 + %2801 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2642, float %2643, float %2644, float %2645, i32 %2194, i32 %2195, i32 %2196, i32 %2197, i32 %2322, i32 %2323) #7, !dbg !61 + %2802 = extractvalue { float, float, float, float } %2801, 0, !dbg !61 + %2803 = extractvalue { float, float, float, float } %2801, 1, !dbg !61 + %2804 = extractvalue { float, float, float, float } %2801, 2, !dbg !61 + %2805 = extractvalue { float, float, float, float } %2801, 3, !dbg !61 + %2806 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2647, float %2648, float %2649, float %2650, i32 %2201, i32 %2202, i32 %2203, i32 %2204, i32 %2237, i32 %2238) #7, !dbg !61 + %2807 = extractvalue { float, float, float, float } %2806, 0, !dbg !61 + %2808 = extractvalue { float, float, float, float } %2806, 1, !dbg !61 + %2809 = extractvalue { float, float, float, float } %2806, 2, !dbg !61 + %2810 = extractvalue { float, float, float, float } %2806, 3, !dbg !61 + %2811 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2652, float %2653, float %2654, float %2655, i32 %2201, i32 %2202, i32 %2203, i32 %2204, i32 %2262, i32 %2263) #7, !dbg !61 + %2812 = extractvalue { float, float, float, float } %2811, 0, !dbg !61 + %2813 = extractvalue { float, float, float, float } %2811, 1, !dbg !61 + %2814 = extractvalue { float, float, float, float } %2811, 2, !dbg !61 + %2815 = extractvalue { float, float, float, float } %2811, 3, !dbg !61 + %2816 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2657, float %2658, float %2659, float %2660, i32 %2201, i32 %2202, i32 %2203, i32 %2204, i32 %2287, i32 %2288) #7, !dbg !61 + %2817 = extractvalue { float, float, float, float } %2816, 0, !dbg !61 + %2818 = extractvalue { float, float, float, float } %2816, 1, !dbg !61 + %2819 = extractvalue { float, float, float, float } %2816, 2, !dbg !61 + %2820 = extractvalue { float, float, float, float } %2816, 3, !dbg !61 + %2821 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2662, float %2663, float %2664, float %2665, i32 %2201, i32 %2202, i32 %2203, i32 %2204, i32 %2312, i32 %2313) #7, !dbg !61 + %2822 = extractvalue { float, float, float, float } %2821, 0, !dbg !61 + %2823 = extractvalue { float, float, float, float } %2821, 1, !dbg !61 + %2824 = extractvalue { float, float, float, float } %2821, 2, !dbg !61 + %2825 = extractvalue { float, float, float, float } %2821, 3, !dbg !61 + %2826 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2667, float %2668, float %2669, float %2670, i32 %2201, i32 %2202, i32 %2203, i32 %2204, i32 %2249, i32 %2250) #7, !dbg !61 + %2827 = extractvalue { float, float, float, float } %2826, 0, !dbg !61 + %2828 = extractvalue { float, float, float, float } %2826, 1, !dbg !61 + %2829 = extractvalue { float, float, float, float } %2826, 2, !dbg !61 + %2830 = extractvalue { float, float, float, float } %2826, 3, !dbg !61 + %2831 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2672, float %2673, float %2674, float %2675, i32 %2201, i32 %2202, i32 %2203, i32 %2204, i32 %2274, i32 %2275) #7, !dbg !61 + %2832 = extractvalue { float, float, float, float } %2831, 0, !dbg !61 + %2833 = extractvalue { float, float, float, float } %2831, 1, !dbg !61 + %2834 = extractvalue { float, float, float, float } %2831, 2, !dbg !61 + %2835 = extractvalue { float, float, float, float } %2831, 3, !dbg !61 + %2836 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2677, float %2678, float %2679, float %2680, i32 %2201, i32 %2202, i32 %2203, i32 %2204, i32 %2299, i32 %2300) #7, !dbg !61 + %2837 = extractvalue { float, float, float, float } %2836, 0, !dbg !61 + %2838 = extractvalue { float, float, float, float } %2836, 1, !dbg !61 + %2839 = extractvalue { float, float, float, float } %2836, 2, !dbg !61 + %2840 = extractvalue { float, float, float, float } %2836, 3, !dbg !61 + %2841 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2682, float %2683, float %2684, float %2685, i32 %2201, i32 %2202, i32 %2203, i32 %2204, i32 %2324, i32 %2325) #7, !dbg !61 + %2842 = extractvalue { float, float, float, float } %2841, 0, !dbg !61 + %2843 = extractvalue { float, float, float, float } %2841, 1, !dbg !61 + %2844 = extractvalue { float, float, float, float } %2841, 2, !dbg !61 + %2845 = extractvalue { float, float, float, float } %2841, 3, !dbg !61 + %2846 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2687, float %2688, float %2689, float %2690, i32 %2207, i32 %2208, i32 %2209, i32 %2210, i32 %2237, i32 %2238) #7, !dbg !61 + %2847 = extractvalue { float, float, float, float } %2846, 0, !dbg !61 + %2848 = extractvalue { float, float, float, float } %2846, 1, !dbg !61 + %2849 = extractvalue { float, float, float, float } %2846, 2, !dbg !61 + %2850 = extractvalue { float, float, float, float } %2846, 3, !dbg !61 + %2851 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2692, float %2693, float %2694, float %2695, i32 %2207, i32 %2208, i32 %2209, i32 %2210, i32 %2262, i32 %2263) #7, !dbg !61 + %2852 = extractvalue { float, float, float, float } %2851, 0, !dbg !61 + %2853 = extractvalue { float, float, float, float } %2851, 1, !dbg !61 + %2854 = extractvalue { float, float, float, float } %2851, 2, !dbg !61 + %2855 = extractvalue { float, float, float, float } %2851, 3, !dbg !61 + %2856 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2697, float %2698, float %2699, float %2700, i32 %2207, i32 %2208, i32 %2209, i32 %2210, i32 %2287, i32 %2288) #7, !dbg !61 + %2857 = extractvalue { float, float, float, float } %2856, 0, !dbg !61 + %2858 = extractvalue { float, float, float, float } %2856, 1, !dbg !61 + %2859 = extractvalue { float, float, float, float } %2856, 2, !dbg !61 + %2860 = extractvalue { float, float, float, float } %2856, 3, !dbg !61 + %2861 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2702, float %2703, float %2704, float %2705, i32 %2207, i32 %2208, i32 %2209, i32 %2210, i32 %2312, i32 %2313) #7, !dbg !61 + %2862 = extractvalue { float, float, float, float } %2861, 0, !dbg !61 + %2863 = extractvalue { float, float, float, float } %2861, 1, !dbg !61 + %2864 = extractvalue { float, float, float, float } %2861, 2, !dbg !61 + %2865 = extractvalue { float, float, float, float } %2861, 3, !dbg !61 + %2866 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2707, float %2708, float %2709, float %2710, i32 %2207, i32 %2208, i32 %2209, i32 %2210, i32 %2249, i32 %2250) #7, !dbg !61 + %2867 = extractvalue { float, float, float, float } %2866, 0, !dbg !61 + %2868 = extractvalue { float, float, float, float } %2866, 1, !dbg !61 + %2869 = extractvalue { float, float, float, float } %2866, 2, !dbg !61 + %2870 = extractvalue { float, float, float, float } %2866, 3, !dbg !61 + %2871 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2712, float %2713, float %2714, float %2715, i32 %2207, i32 %2208, i32 %2209, i32 %2210, i32 %2274, i32 %2275) #7, !dbg !61 + %2872 = extractvalue { float, float, float, float } %2871, 0, !dbg !61 + %2873 = extractvalue { float, float, float, float } %2871, 1, !dbg !61 + %2874 = extractvalue { float, float, float, float } %2871, 2, !dbg !61 + %2875 = extractvalue { float, float, float, float } %2871, 3, !dbg !61 + %2876 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2717, float %2718, float %2719, float %2720, i32 %2207, i32 %2208, i32 %2209, i32 %2210, i32 %2299, i32 %2300) #7, !dbg !61 + %2877 = extractvalue { float, float, float, float } %2876, 0, !dbg !61 + %2878 = extractvalue { float, float, float, float } %2876, 1, !dbg !61 + %2879 = extractvalue { float, float, float, float } %2876, 2, !dbg !61 + %2880 = extractvalue { float, float, float, float } %2876, 3, !dbg !61 + %2881 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2722, float %2723, float %2724, float %2725, i32 %2207, i32 %2208, i32 %2209, i32 %2210, i32 %2324, i32 %2325) #7, !dbg !61 + %2882 = extractvalue { float, float, float, float } %2881, 0, !dbg !61 + %2883 = extractvalue { float, float, float, float } %2881, 1, !dbg !61 + %2884 = extractvalue { float, float, float, float } %2881, 2, !dbg !61 + %2885 = extractvalue { float, float, float, float } %2881, 3, !dbg !61 + %2886 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2727, float %2728, float %2729, float %2730, i32 %2213, i32 %2214, i32 %2215, i32 %2216, i32 %2237, i32 %2238) #7, !dbg !61 + %2887 = extractvalue { float, float, float, float } %2886, 0, !dbg !61 + %2888 = extractvalue { float, float, float, float } %2886, 1, !dbg !61 + %2889 = extractvalue { float, float, float, float } %2886, 2, !dbg !61 + %2890 = extractvalue { float, float, float, float } %2886, 3, !dbg !61 + %2891 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2732, float %2733, float %2734, float %2735, i32 %2213, i32 %2214, i32 %2215, i32 %2216, i32 %2262, i32 %2263) #7, !dbg !61 + %2892 = extractvalue { float, float, float, float } %2891, 0, !dbg !61 + %2893 = extractvalue { float, float, float, float } %2891, 1, !dbg !61 + %2894 = extractvalue { float, float, float, float } %2891, 2, !dbg !61 + %2895 = extractvalue { float, float, float, float } %2891, 3, !dbg !61 + %2896 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2737, float %2738, float %2739, float %2740, i32 %2213, i32 %2214, i32 %2215, i32 %2216, i32 %2287, i32 %2288) #7, !dbg !61 + %2897 = extractvalue { float, float, float, float } %2896, 0, !dbg !61 + %2898 = extractvalue { float, float, float, float } %2896, 1, !dbg !61 + %2899 = extractvalue { float, float, float, float } %2896, 2, !dbg !61 + %2900 = extractvalue { float, float, float, float } %2896, 3, !dbg !61 + %2901 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2742, float %2743, float %2744, float %2745, i32 %2213, i32 %2214, i32 %2215, i32 %2216, i32 %2312, i32 %2313) #7, !dbg !61 + %2902 = extractvalue { float, float, float, float } %2901, 0, !dbg !61 + %2903 = extractvalue { float, float, float, float } %2901, 1, !dbg !61 + %2904 = extractvalue { float, float, float, float } %2901, 2, !dbg !61 + %2905 = extractvalue { float, float, float, float } %2901, 3, !dbg !61 + %2906 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2747, float %2748, float %2749, float %2750, i32 %2213, i32 %2214, i32 %2215, i32 %2216, i32 %2249, i32 %2250) #7, !dbg !61 + %2907 = extractvalue { float, float, float, float } %2906, 0, !dbg !61 + %2908 = extractvalue { float, float, float, float } %2906, 1, !dbg !61 + %2909 = extractvalue { float, float, float, float } %2906, 2, !dbg !61 + %2910 = extractvalue { float, float, float, float } %2906, 3, !dbg !61 + %2911 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2752, float %2753, float %2754, float %2755, i32 %2213, i32 %2214, i32 %2215, i32 %2216, i32 %2274, i32 %2275) #7, !dbg !61 + %2912 = extractvalue { float, float, float, float } %2911, 0, !dbg !61 + %2913 = extractvalue { float, float, float, float } %2911, 1, !dbg !61 + %2914 = extractvalue { float, float, float, float } %2911, 2, !dbg !61 + %2915 = extractvalue { float, float, float, float } %2911, 3, !dbg !61 + %2916 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2757, float %2758, float %2759, float %2760, i32 %2213, i32 %2214, i32 %2215, i32 %2216, i32 %2299, i32 %2300) #7, !dbg !61 + %2917 = extractvalue { float, float, float, float } %2916, 0, !dbg !61 + %2918 = extractvalue { float, float, float, float } %2916, 1, !dbg !61 + %2919 = extractvalue { float, float, float, float } %2916, 2, !dbg !61 + %2920 = extractvalue { float, float, float, float } %2916, 3, !dbg !61 + %2921 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2762, float %2763, float %2764, float %2765, i32 %2213, i32 %2214, i32 %2215, i32 %2216, i32 %2324, i32 %2325) #7, !dbg !61 + %2922 = extractvalue { float, float, float, float } %2921, 0, !dbg !61 + %2923 = extractvalue { float, float, float, float } %2921, 1, !dbg !61 + %2924 = extractvalue { float, float, float, float } %2921, 2, !dbg !61 + %2925 = extractvalue { float, float, float, float } %2921, 3, !dbg !61 + %2926 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2767, float %2768, float %2769, float %2770, i32 %2219, i32 %2220, i32 %2221, i32 %2222, i32 %2237, i32 %2238) #7, !dbg !61 + %2927 = extractvalue { float, float, float, float } %2926, 0, !dbg !61 + %2928 = extractvalue { float, float, float, float } %2926, 1, !dbg !61 + %2929 = extractvalue { float, float, float, float } %2926, 2, !dbg !61 + %2930 = extractvalue { float, float, float, float } %2926, 3, !dbg !61 + %2931 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2772, float %2773, float %2774, float %2775, i32 %2219, i32 %2220, i32 %2221, i32 %2222, i32 %2262, i32 %2263) #7, !dbg !61 + %2932 = extractvalue { float, float, float, float } %2931, 0, !dbg !61 + %2933 = extractvalue { float, float, float, float } %2931, 1, !dbg !61 + %2934 = extractvalue { float, float, float, float } %2931, 2, !dbg !61 + %2935 = extractvalue { float, float, float, float } %2931, 3, !dbg !61 + %2936 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2777, float %2778, float %2779, float %2780, i32 %2219, i32 %2220, i32 %2221, i32 %2222, i32 %2287, i32 %2288) #7, !dbg !61 + %2937 = extractvalue { float, float, float, float } %2936, 0, !dbg !61 + %2938 = extractvalue { float, float, float, float } %2936, 1, !dbg !61 + %2939 = extractvalue { float, float, float, float } %2936, 2, !dbg !61 + %2940 = extractvalue { float, float, float, float } %2936, 3, !dbg !61 + %2941 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2782, float %2783, float %2784, float %2785, i32 %2219, i32 %2220, i32 %2221, i32 %2222, i32 %2312, i32 %2313) #7, !dbg !61 + %2942 = extractvalue { float, float, float, float } %2941, 0, !dbg !61 + %2943 = extractvalue { float, float, float, float } %2941, 1, !dbg !61 + %2944 = extractvalue { float, float, float, float } %2941, 2, !dbg !61 + %2945 = extractvalue { float, float, float, float } %2941, 3, !dbg !61 + %2946 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2787, float %2788, float %2789, float %2790, i32 %2219, i32 %2220, i32 %2221, i32 %2222, i32 %2249, i32 %2250) #7, !dbg !61 + %2947 = extractvalue { float, float, float, float } %2946, 0, !dbg !61 + %2948 = extractvalue { float, float, float, float } %2946, 1, !dbg !61 + %2949 = extractvalue { float, float, float, float } %2946, 2, !dbg !61 + %2950 = extractvalue { float, float, float, float } %2946, 3, !dbg !61 + %2951 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2792, float %2793, float %2794, float %2795, i32 %2219, i32 %2220, i32 %2221, i32 %2222, i32 %2274, i32 %2275) #7, !dbg !61 + %2952 = extractvalue { float, float, float, float } %2951, 0, !dbg !61 + %2953 = extractvalue { float, float, float, float } %2951, 1, !dbg !61 + %2954 = extractvalue { float, float, float, float } %2951, 2, !dbg !61 + %2955 = extractvalue { float, float, float, float } %2951, 3, !dbg !61 + %2956 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2797, float %2798, float %2799, float %2800, i32 %2219, i32 %2220, i32 %2221, i32 %2222, i32 %2299, i32 %2300) #7, !dbg !61 + %2957 = extractvalue { float, float, float, float } %2956, 0, !dbg !61 + %2958 = extractvalue { float, float, float, float } %2956, 1, !dbg !61 + %2959 = extractvalue { float, float, float, float } %2956, 2, !dbg !61 + %2960 = extractvalue { float, float, float, float } %2956, 3, !dbg !61 + %2961 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2802, float %2803, float %2804, float %2805, i32 %2219, i32 %2220, i32 %2221, i32 %2222, i32 %2324, i32 %2325) #7, !dbg !61 + %2962 = extractvalue { float, float, float, float } %2961, 0, !dbg !61 + %2963 = extractvalue { float, float, float, float } %2961, 1, !dbg !61 + %2964 = extractvalue { float, float, float, float } %2961, 2, !dbg !61 + %2965 = extractvalue { float, float, float, float } %2961, 3, !dbg !61 + %2966 = insertelement <2 x float> poison, float %2807, i64 0, !dbg !62 + %2967 = insertelement <2 x float> %2966, float %2808, i64 1, !dbg !62 + %2968 = fptrunc <2 x float> %2967 to <2 x bfloat>, !dbg !62 + %2969 = insertelement <2 x float> poison, float %2809, i64 0, !dbg !62 + %2970 = insertelement <2 x float> %2969, float %2810, i64 1, !dbg !62 + %2971 = fptrunc <2 x float> %2970 to <2 x bfloat>, !dbg !62 + %2972 = insertelement <2 x float> poison, float %2812, i64 0, !dbg !62 + %2973 = insertelement <2 x float> %2972, float %2813, i64 1, !dbg !62 + %2974 = fptrunc <2 x float> %2973 to <2 x bfloat>, !dbg !62 + %2975 = insertelement <2 x float> poison, float %2814, i64 0, !dbg !62 + %2976 = insertelement <2 x float> %2975, float %2815, i64 1, !dbg !62 + %2977 = fptrunc <2 x float> %2976 to <2 x bfloat>, !dbg !62 + %2978 = insertelement <2 x float> poison, float %2817, i64 0, !dbg !62 + %2979 = insertelement <2 x float> %2978, float %2818, i64 1, !dbg !62 + %2980 = fptrunc <2 x float> %2979 to <2 x bfloat>, !dbg !62 + %2981 = insertelement <2 x float> poison, float %2819, i64 0, !dbg !62 + %2982 = insertelement <2 x float> %2981, float %2820, i64 1, !dbg !62 + %2983 = fptrunc <2 x float> %2982 to <2 x bfloat>, !dbg !62 + %2984 = insertelement <2 x float> poison, float %2822, i64 0, !dbg !62 + %2985 = insertelement <2 x float> %2984, float %2823, i64 1, !dbg !62 + %2986 = fptrunc <2 x float> %2985 to <2 x bfloat>, !dbg !62 + %2987 = insertelement <2 x float> poison, float %2824, i64 0, !dbg !62 + %2988 = insertelement <2 x float> %2987, float %2825, i64 1, !dbg !62 + %2989 = fptrunc <2 x float> %2988 to <2 x bfloat>, !dbg !62 + %2990 = insertelement <2 x float> poison, float %2827, i64 0, !dbg !62 + %2991 = insertelement <2 x float> %2990, float %2828, i64 1, !dbg !62 + %2992 = fptrunc <2 x float> %2991 to <2 x bfloat>, !dbg !62 + %2993 = insertelement <2 x float> poison, float %2829, i64 0, !dbg !62 + %2994 = insertelement <2 x float> %2993, float %2830, i64 1, !dbg !62 + %2995 = fptrunc <2 x float> %2994 to <2 x bfloat>, !dbg !62 + %2996 = insertelement <2 x float> poison, float %2832, i64 0, !dbg !62 + %2997 = insertelement <2 x float> %2996, float %2833, i64 1, !dbg !62 + %2998 = fptrunc <2 x float> %2997 to <2 x bfloat>, !dbg !62 + %2999 = insertelement <2 x float> poison, float %2834, i64 0, !dbg !62 + %3000 = insertelement <2 x float> %2999, float %2835, i64 1, !dbg !62 + %3001 = fptrunc <2 x float> %3000 to <2 x bfloat>, !dbg !62 + %3002 = insertelement <2 x float> poison, float %2837, i64 0, !dbg !62 + %3003 = insertelement <2 x float> %3002, float %2838, i64 1, !dbg !62 + %3004 = fptrunc <2 x float> %3003 to <2 x bfloat>, !dbg !62 + %3005 = insertelement <2 x float> poison, float %2839, i64 0, !dbg !62 + %3006 = insertelement <2 x float> %3005, float %2840, i64 1, !dbg !62 + %3007 = fptrunc <2 x float> %3006 to <2 x bfloat>, !dbg !62 + %3008 = insertelement <2 x float> poison, float %2842, i64 0, !dbg !62 + %3009 = insertelement <2 x float> %3008, float %2843, i64 1, !dbg !62 + %3010 = fptrunc <2 x float> %3009 to <2 x bfloat>, !dbg !62 + %3011 = insertelement <2 x float> poison, float %2844, i64 0, !dbg !62 + %3012 = insertelement <2 x float> %3011, float %2845, i64 1, !dbg !62 + %3013 = fptrunc <2 x float> %3012 to <2 x bfloat>, !dbg !62 + %3014 = insertelement <2 x float> poison, float %2847, i64 0, !dbg !62 + %3015 = insertelement <2 x float> %3014, float %2848, i64 1, !dbg !62 + %3016 = fptrunc <2 x float> %3015 to <2 x bfloat>, !dbg !62 + %3017 = insertelement <2 x float> poison, float %2849, i64 0, !dbg !62 + %3018 = insertelement <2 x float> %3017, float %2850, i64 1, !dbg !62 + %3019 = fptrunc <2 x float> %3018 to <2 x bfloat>, !dbg !62 + %3020 = insertelement <2 x float> poison, float %2852, i64 0, !dbg !62 + %3021 = insertelement <2 x float> %3020, float %2853, i64 1, !dbg !62 + %3022 = fptrunc <2 x float> %3021 to <2 x bfloat>, !dbg !62 + %3023 = insertelement <2 x float> poison, float %2854, i64 0, !dbg !62 + %3024 = insertelement <2 x float> %3023, float %2855, i64 1, !dbg !62 + %3025 = fptrunc <2 x float> %3024 to <2 x bfloat>, !dbg !62 + %3026 = insertelement <2 x float> poison, float %2857, i64 0, !dbg !62 + %3027 = insertelement <2 x float> %3026, float %2858, i64 1, !dbg !62 + %3028 = fptrunc <2 x float> %3027 to <2 x bfloat>, !dbg !62 + %3029 = insertelement <2 x float> poison, float %2859, i64 0, !dbg !62 + %3030 = insertelement <2 x float> %3029, float %2860, i64 1, !dbg !62 + %3031 = fptrunc <2 x float> %3030 to <2 x bfloat>, !dbg !62 + %3032 = insertelement <2 x float> poison, float %2862, i64 0, !dbg !62 + %3033 = insertelement <2 x float> %3032, float %2863, i64 1, !dbg !62 + %3034 = fptrunc <2 x float> %3033 to <2 x bfloat>, !dbg !62 + %3035 = insertelement <2 x float> poison, float %2864, i64 0, !dbg !62 + %3036 = insertelement <2 x float> %3035, float %2865, i64 1, !dbg !62 + %3037 = fptrunc <2 x float> %3036 to <2 x bfloat>, !dbg !62 + %3038 = insertelement <2 x float> poison, float %2867, i64 0, !dbg !62 + %3039 = insertelement <2 x float> %3038, float %2868, i64 1, !dbg !62 + %3040 = fptrunc <2 x float> %3039 to <2 x bfloat>, !dbg !62 + %3041 = insertelement <2 x float> poison, float %2869, i64 0, !dbg !62 + %3042 = insertelement <2 x float> %3041, float %2870, i64 1, !dbg !62 + %3043 = fptrunc <2 x float> %3042 to <2 x bfloat>, !dbg !62 + %3044 = insertelement <2 x float> poison, float %2872, i64 0, !dbg !62 + %3045 = insertelement <2 x float> %3044, float %2873, i64 1, !dbg !62 + %3046 = fptrunc <2 x float> %3045 to <2 x bfloat>, !dbg !62 + %3047 = insertelement <2 x float> poison, float %2874, i64 0, !dbg !62 + %3048 = insertelement <2 x float> %3047, float %2875, i64 1, !dbg !62 + %3049 = fptrunc <2 x float> %3048 to <2 x bfloat>, !dbg !62 + %3050 = insertelement <2 x float> poison, float %2877, i64 0, !dbg !62 + %3051 = insertelement <2 x float> %3050, float %2878, i64 1, !dbg !62 + %3052 = fptrunc <2 x float> %3051 to <2 x bfloat>, !dbg !62 + %3053 = insertelement <2 x float> poison, float %2879, i64 0, !dbg !62 + %3054 = insertelement <2 x float> %3053, float %2880, i64 1, !dbg !62 + %3055 = fptrunc <2 x float> %3054 to <2 x bfloat>, !dbg !62 + %3056 = insertelement <2 x float> poison, float %2882, i64 0, !dbg !62 + %3057 = insertelement <2 x float> %3056, float %2883, i64 1, !dbg !62 + %3058 = fptrunc <2 x float> %3057 to <2 x bfloat>, !dbg !62 + %3059 = insertelement <2 x float> poison, float %2884, i64 0, !dbg !62 + %3060 = insertelement <2 x float> %3059, float %2885, i64 1, !dbg !62 + %3061 = fptrunc <2 x float> %3060 to <2 x bfloat>, !dbg !62 + %3062 = insertelement <2 x float> poison, float %2887, i64 0, !dbg !62 + %3063 = insertelement <2 x float> %3062, float %2888, i64 1, !dbg !62 + %3064 = fptrunc <2 x float> %3063 to <2 x bfloat>, !dbg !62 + %3065 = insertelement <2 x float> poison, float %2889, i64 0, !dbg !62 + %3066 = insertelement <2 x float> %3065, float %2890, i64 1, !dbg !62 + %3067 = fptrunc <2 x float> %3066 to <2 x bfloat>, !dbg !62 + %3068 = insertelement <2 x float> poison, float %2892, i64 0, !dbg !62 + %3069 = insertelement <2 x float> %3068, float %2893, i64 1, !dbg !62 + %3070 = fptrunc <2 x float> %3069 to <2 x bfloat>, !dbg !62 + %3071 = insertelement <2 x float> poison, float %2894, i64 0, !dbg !62 + %3072 = insertelement <2 x float> %3071, float %2895, i64 1, !dbg !62 + %3073 = fptrunc <2 x float> %3072 to <2 x bfloat>, !dbg !62 + %3074 = insertelement <2 x float> poison, float %2897, i64 0, !dbg !62 + %3075 = insertelement <2 x float> %3074, float %2898, i64 1, !dbg !62 + %3076 = fptrunc <2 x float> %3075 to <2 x bfloat>, !dbg !62 + %3077 = insertelement <2 x float> poison, float %2899, i64 0, !dbg !62 + %3078 = insertelement <2 x float> %3077, float %2900, i64 1, !dbg !62 + %3079 = fptrunc <2 x float> %3078 to <2 x bfloat>, !dbg !62 + %3080 = insertelement <2 x float> poison, float %2902, i64 0, !dbg !62 + %3081 = insertelement <2 x float> %3080, float %2903, i64 1, !dbg !62 + %3082 = fptrunc <2 x float> %3081 to <2 x bfloat>, !dbg !62 + %3083 = insertelement <2 x float> poison, float %2904, i64 0, !dbg !62 + %3084 = insertelement <2 x float> %3083, float %2905, i64 1, !dbg !62 + %3085 = fptrunc <2 x float> %3084 to <2 x bfloat>, !dbg !62 + %3086 = insertelement <2 x float> poison, float %2907, i64 0, !dbg !62 + %3087 = insertelement <2 x float> %3086, float %2908, i64 1, !dbg !62 + %3088 = fptrunc <2 x float> %3087 to <2 x bfloat>, !dbg !62 + %3089 = insertelement <2 x float> poison, float %2909, i64 0, !dbg !62 + %3090 = insertelement <2 x float> %3089, float %2910, i64 1, !dbg !62 + %3091 = fptrunc <2 x float> %3090 to <2 x bfloat>, !dbg !62 + %3092 = insertelement <2 x float> poison, float %2912, i64 0, !dbg !62 + %3093 = insertelement <2 x float> %3092, float %2913, i64 1, !dbg !62 + %3094 = fptrunc <2 x float> %3093 to <2 x bfloat>, !dbg !62 + %3095 = insertelement <2 x float> poison, float %2914, i64 0, !dbg !62 + %3096 = insertelement <2 x float> %3095, float %2915, i64 1, !dbg !62 + %3097 = fptrunc <2 x float> %3096 to <2 x bfloat>, !dbg !62 + %3098 = insertelement <2 x float> poison, float %2917, i64 0, !dbg !62 + %3099 = insertelement <2 x float> %3098, float %2918, i64 1, !dbg !62 + %3100 = fptrunc <2 x float> %3099 to <2 x bfloat>, !dbg !62 + %3101 = insertelement <2 x float> poison, float %2919, i64 0, !dbg !62 + %3102 = insertelement <2 x float> %3101, float %2920, i64 1, !dbg !62 + %3103 = fptrunc <2 x float> %3102 to <2 x bfloat>, !dbg !62 + %3104 = insertelement <2 x float> poison, float %2922, i64 0, !dbg !62 + %3105 = insertelement <2 x float> %3104, float %2923, i64 1, !dbg !62 + %3106 = fptrunc <2 x float> %3105 to <2 x bfloat>, !dbg !62 + %3107 = insertelement <2 x float> poison, float %2924, i64 0, !dbg !62 + %3108 = insertelement <2 x float> %3107, float %2925, i64 1, !dbg !62 + %3109 = fptrunc <2 x float> %3108 to <2 x bfloat>, !dbg !62 + %3110 = insertelement <2 x float> poison, float %2927, i64 0, !dbg !62 + %3111 = insertelement <2 x float> %3110, float %2928, i64 1, !dbg !62 + %3112 = fptrunc <2 x float> %3111 to <2 x bfloat>, !dbg !62 + %3113 = insertelement <2 x float> poison, float %2929, i64 0, !dbg !62 + %3114 = insertelement <2 x float> %3113, float %2930, i64 1, !dbg !62 + %3115 = fptrunc <2 x float> %3114 to <2 x bfloat>, !dbg !62 + %3116 = insertelement <2 x float> poison, float %2932, i64 0, !dbg !62 + %3117 = insertelement <2 x float> %3116, float %2933, i64 1, !dbg !62 + %3118 = fptrunc <2 x float> %3117 to <2 x bfloat>, !dbg !62 + %3119 = insertelement <2 x float> poison, float %2934, i64 0, !dbg !62 + %3120 = insertelement <2 x float> %3119, float %2935, i64 1, !dbg !62 + %3121 = fptrunc <2 x float> %3120 to <2 x bfloat>, !dbg !62 + %3122 = insertelement <2 x float> poison, float %2937, i64 0, !dbg !62 + %3123 = insertelement <2 x float> %3122, float %2938, i64 1, !dbg !62 + %3124 = fptrunc <2 x float> %3123 to <2 x bfloat>, !dbg !62 + %3125 = insertelement <2 x float> poison, float %2939, i64 0, !dbg !62 + %3126 = insertelement <2 x float> %3125, float %2940, i64 1, !dbg !62 + %3127 = fptrunc <2 x float> %3126 to <2 x bfloat>, !dbg !62 + %3128 = insertelement <2 x float> poison, float %2942, i64 0, !dbg !62 + %3129 = insertelement <2 x float> %3128, float %2943, i64 1, !dbg !62 + %3130 = fptrunc <2 x float> %3129 to <2 x bfloat>, !dbg !62 + %3131 = insertelement <2 x float> poison, float %2944, i64 0, !dbg !62 + %3132 = insertelement <2 x float> %3131, float %2945, i64 1, !dbg !62 + %3133 = fptrunc <2 x float> %3132 to <2 x bfloat>, !dbg !62 + %3134 = insertelement <2 x float> poison, float %2947, i64 0, !dbg !62 + %3135 = insertelement <2 x float> %3134, float %2948, i64 1, !dbg !62 + %3136 = fptrunc <2 x float> %3135 to <2 x bfloat>, !dbg !62 + %3137 = insertelement <2 x float> poison, float %2949, i64 0, !dbg !62 + %3138 = insertelement <2 x float> %3137, float %2950, i64 1, !dbg !62 + %3139 = fptrunc <2 x float> %3138 to <2 x bfloat>, !dbg !62 + %3140 = insertelement <2 x float> poison, float %2952, i64 0, !dbg !62 + %3141 = insertelement <2 x float> %3140, float %2953, i64 1, !dbg !62 + %3142 = fptrunc <2 x float> %3141 to <2 x bfloat>, !dbg !62 + %3143 = insertelement <2 x float> poison, float %2954, i64 0, !dbg !62 + %3144 = insertelement <2 x float> %3143, float %2955, i64 1, !dbg !62 + %3145 = fptrunc <2 x float> %3144 to <2 x bfloat>, !dbg !62 + %3146 = insertelement <2 x float> poison, float %2957, i64 0, !dbg !62 + %3147 = insertelement <2 x float> %3146, float %2958, i64 1, !dbg !62 + %3148 = fptrunc <2 x float> %3147 to <2 x bfloat>, !dbg !62 + %3149 = insertelement <2 x float> poison, float %2959, i64 0, !dbg !62 + %3150 = insertelement <2 x float> %3149, float %2960, i64 1, !dbg !62 + %3151 = fptrunc <2 x float> %3150 to <2 x bfloat>, !dbg !62 + %3152 = insertelement <2 x float> poison, float %2962, i64 0, !dbg !62 + %3153 = insertelement <2 x float> %3152, float %2963, i64 1, !dbg !62 + %3154 = fptrunc <2 x float> %3153 to <2 x bfloat>, !dbg !62 + %3155 = insertelement <2 x float> poison, float %2964, i64 0, !dbg !62 + %3156 = insertelement <2 x float> %3155, float %2965, i64 1, !dbg !62 + %3157 = fptrunc <2 x float> %3156 to <2 x bfloat>, !dbg !62 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !63 + %3158 = shl nuw nsw i32 %53, 2, !dbg !63 + %3159 = xor i32 %52, %3158, !dbg !63 + %3160 = xor i32 %3159, %1946, !dbg !63 + %3161 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %611, !dbg !63 + %3162 = getelementptr inbounds nuw i8, ptr addrspace(3) %3161, i32 %3160, !dbg !63 + %3163 = bitcast <2 x bfloat> %2968 to i32, !dbg !63 + %3164 = bitcast <2 x bfloat> %2974 to i32, !dbg !63 + %3165 = bitcast <2 x bfloat> %2980 to i32, !dbg !63 + %3166 = bitcast <2 x bfloat> %2986 to i32, !dbg !63 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %3162, i32 %3163, i32 %3164, i32 %3165, i32 %3166), !dbg !63 + %3167 = getelementptr inbounds nuw i8, ptr addrspace(3) %3162, i32 128, !dbg !63 + %3168 = bitcast <2 x bfloat> %2992 to i32, !dbg !63 + %3169 = bitcast <2 x bfloat> %2998 to i32, !dbg !63 + %3170 = bitcast <2 x bfloat> %3004 to i32, !dbg !63 + %3171 = bitcast <2 x bfloat> %3010 to i32, !dbg !63 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %3167, i32 %3168, i32 %3169, i32 %3170, i32 %3171), !dbg !63 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !63 + %3172 = load <4 x i32>, ptr addrspace(3) %286, align 16, !dbg !63 + %3173 = load <4 x i32>, ptr addrspace(3) %302, align 16, !dbg !63 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !63 + %3174 = bitcast <2 x bfloat> %2971 to i32, !dbg !63 + %3175 = bitcast <2 x bfloat> %2977 to i32, !dbg !63 + %3176 = bitcast <2 x bfloat> %2983 to i32, !dbg !63 + %3177 = bitcast <2 x bfloat> %2989 to i32, !dbg !63 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %3162, i32 %3174, i32 %3175, i32 %3176, i32 %3177), !dbg !63 + %3178 = bitcast <2 x bfloat> %2995 to i32, !dbg !63 + %3179 = bitcast <2 x bfloat> %3001 to i32, !dbg !63 + %3180 = bitcast <2 x bfloat> %3007 to i32, !dbg !63 + %3181 = bitcast <2 x bfloat> %3013 to i32, !dbg !63 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %3167, i32 %3178, i32 %3179, i32 %3180, i32 %3181), !dbg !63 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !63 + %3182 = load <4 x i32>, ptr addrspace(3) %286, align 16, !dbg !63 + %3183 = load <4 x i32>, ptr addrspace(3) %302, align 16, !dbg !63 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !63 + %3184 = bitcast <2 x bfloat> %3016 to i32, !dbg !63 + %3185 = bitcast <2 x bfloat> %3022 to i32, !dbg !63 + %3186 = bitcast <2 x bfloat> %3028 to i32, !dbg !63 + %3187 = bitcast <2 x bfloat> %3034 to i32, !dbg !63 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %3162, i32 %3184, i32 %3185, i32 %3186, i32 %3187), !dbg !63 + %3188 = bitcast <2 x bfloat> %3040 to i32, !dbg !63 + %3189 = bitcast <2 x bfloat> %3046 to i32, !dbg !63 + %3190 = bitcast <2 x bfloat> %3052 to i32, !dbg !63 + %3191 = bitcast <2 x bfloat> %3058 to i32, !dbg !63 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %3167, i32 %3188, i32 %3189, i32 %3190, i32 %3191), !dbg !63 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !63 + %3192 = load <4 x i32>, ptr addrspace(3) %286, align 16, !dbg !63 + %3193 = load <4 x i32>, ptr addrspace(3) %302, align 16, !dbg !63 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !63 + %3194 = bitcast <2 x bfloat> %3019 to i32, !dbg !63 + %3195 = bitcast <2 x bfloat> %3025 to i32, !dbg !63 + %3196 = bitcast <2 x bfloat> %3031 to i32, !dbg !63 + %3197 = bitcast <2 x bfloat> %3037 to i32, !dbg !63 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %3162, i32 %3194, i32 %3195, i32 %3196, i32 %3197), !dbg !63 + %3198 = bitcast <2 x bfloat> %3043 to i32, !dbg !63 + %3199 = bitcast <2 x bfloat> %3049 to i32, !dbg !63 + %3200 = bitcast <2 x bfloat> %3055 to i32, !dbg !63 + %3201 = bitcast <2 x bfloat> %3061 to i32, !dbg !63 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %3167, i32 %3198, i32 %3199, i32 %3200, i32 %3201), !dbg !63 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !63 + %3202 = load <4 x i32>, ptr addrspace(3) %286, align 16, !dbg !63 + %3203 = load <4 x i32>, ptr addrspace(3) %302, align 16, !dbg !63 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !63 + %3204 = bitcast <2 x bfloat> %3064 to i32, !dbg !63 + %3205 = bitcast <2 x bfloat> %3070 to i32, !dbg !63 + %3206 = bitcast <2 x bfloat> %3076 to i32, !dbg !63 + %3207 = bitcast <2 x bfloat> %3082 to i32, !dbg !63 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %3162, i32 %3204, i32 %3205, i32 %3206, i32 %3207), !dbg !63 + %3208 = bitcast <2 x bfloat> %3088 to i32, !dbg !63 + %3209 = bitcast <2 x bfloat> %3094 to i32, !dbg !63 + %3210 = bitcast <2 x bfloat> %3100 to i32, !dbg !63 + %3211 = bitcast <2 x bfloat> %3106 to i32, !dbg !63 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %3167, i32 %3208, i32 %3209, i32 %3210, i32 %3211), !dbg !63 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !63 + %3212 = load <4 x i32>, ptr addrspace(3) %286, align 16, !dbg !63 + %3213 = load <4 x i32>, ptr addrspace(3) %302, align 16, !dbg !63 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !63 + %3214 = bitcast <2 x bfloat> %3067 to i32, !dbg !63 + %3215 = bitcast <2 x bfloat> %3073 to i32, !dbg !63 + %3216 = bitcast <2 x bfloat> %3079 to i32, !dbg !63 + %3217 = bitcast <2 x bfloat> %3085 to i32, !dbg !63 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %3162, i32 %3214, i32 %3215, i32 %3216, i32 %3217), !dbg !63 + %3218 = bitcast <2 x bfloat> %3091 to i32, !dbg !63 + %3219 = bitcast <2 x bfloat> %3097 to i32, !dbg !63 + %3220 = bitcast <2 x bfloat> %3103 to i32, !dbg !63 + %3221 = bitcast <2 x bfloat> %3109 to i32, !dbg !63 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %3167, i32 %3218, i32 %3219, i32 %3220, i32 %3221), !dbg !63 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !63 + %3222 = load <4 x i32>, ptr addrspace(3) %286, align 16, !dbg !63 + %3223 = load <4 x i32>, ptr addrspace(3) %302, align 16, !dbg !63 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !63 + %3224 = bitcast <2 x bfloat> %3112 to i32, !dbg !63 + %3225 = bitcast <2 x bfloat> %3118 to i32, !dbg !63 + %3226 = bitcast <2 x bfloat> %3124 to i32, !dbg !63 + %3227 = bitcast <2 x bfloat> %3130 to i32, !dbg !63 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %3162, i32 %3224, i32 %3225, i32 %3226, i32 %3227), !dbg !63 + %3228 = bitcast <2 x bfloat> %3136 to i32, !dbg !63 + %3229 = bitcast <2 x bfloat> %3142 to i32, !dbg !63 + %3230 = bitcast <2 x bfloat> %3148 to i32, !dbg !63 + %3231 = bitcast <2 x bfloat> %3154 to i32, !dbg !63 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %3167, i32 %3228, i32 %3229, i32 %3230, i32 %3231), !dbg !63 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !63 + %3232 = load <4 x i32>, ptr addrspace(3) %286, align 16, !dbg !63 + %3233 = load <4 x i32>, ptr addrspace(3) %302, align 16, !dbg !63 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !63 + %3234 = bitcast <2 x bfloat> %3115 to i32, !dbg !63 + %3235 = bitcast <2 x bfloat> %3121 to i32, !dbg !63 + %3236 = bitcast <2 x bfloat> %3127 to i32, !dbg !63 + %3237 = bitcast <2 x bfloat> %3133 to i32, !dbg !63 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %3162, i32 %3234, i32 %3235, i32 %3236, i32 %3237), !dbg !63 + %3238 = bitcast <2 x bfloat> %3139 to i32, !dbg !63 + %3239 = bitcast <2 x bfloat> %3145 to i32, !dbg !63 + %3240 = bitcast <2 x bfloat> %3151 to i32, !dbg !63 + %3241 = bitcast <2 x bfloat> %3157 to i32, !dbg !63 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %3167, i32 %3238, i32 %3239, i32 %3240, i32 %3241), !dbg !63 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !63 + %3242 = load <4 x i32>, ptr addrspace(3) %286, align 16, !dbg !63 + %3243 = load <4 x i32>, ptr addrspace(3) %302, align 16, !dbg !63 + %.extract = extractelement <4 x i32> %3172, i64 0, !dbg !63 + %.extract734 = extractelement <4 x i32> %3172, i64 1, !dbg !63 + %.extract736 = extractelement <4 x i32> %3172, i64 2, !dbg !63 + %.extract738 = extractelement <4 x i32> %3172, i64 3, !dbg !63 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract, i32 %.extract734, i32 %.extract736, i32 %.extract738, ptr addrspace(1) %2027, i1 %41) #7, !dbg !63 + %.extract740 = extractelement <4 x i32> %3173, i64 0, !dbg !63 + %.extract742 = extractelement <4 x i32> %3173, i64 1, !dbg !63 + %.extract744 = extractelement <4 x i32> %3173, i64 2, !dbg !63 + %.extract746 = extractelement <4 x i32> %3173, i64 3, !dbg !63 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract740, i32 %.extract742, i32 %.extract744, i32 %.extract746, ptr addrspace(1) %2028, i1 %193) #7, !dbg !63 + %.extract748 = extractelement <4 x i32> %3182, i64 0, !dbg !63 + %.extract750 = extractelement <4 x i32> %3182, i64 1, !dbg !63 + %.extract752 = extractelement <4 x i32> %3182, i64 2, !dbg !63 + %.extract754 = extractelement <4 x i32> %3182, i64 3, !dbg !63 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract748, i32 %.extract750, i32 %.extract752, i32 %.extract754, ptr addrspace(1) %2029, i1 %199) #7, !dbg !63 + %.extract756 = extractelement <4 x i32> %3183, i64 0, !dbg !63 + %.extract758 = extractelement <4 x i32> %3183, i64 1, !dbg !63 + %.extract760 = extractelement <4 x i32> %3183, i64 2, !dbg !63 + %.extract762 = extractelement <4 x i32> %3183, i64 3, !dbg !63 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract756, i32 %.extract758, i32 %.extract760, i32 %.extract762, ptr addrspace(1) %2030, i1 %205) #7, !dbg !63 + %.extract764 = extractelement <4 x i32> %3192, i64 0, !dbg !63 + %.extract766 = extractelement <4 x i32> %3192, i64 1, !dbg !63 + %.extract768 = extractelement <4 x i32> %3192, i64 2, !dbg !63 + %.extract770 = extractelement <4 x i32> %3192, i64 3, !dbg !63 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract764, i32 %.extract766, i32 %.extract768, i32 %.extract770, ptr addrspace(1) %2031, i1 %211) #7, !dbg !63 + %.extract772 = extractelement <4 x i32> %3193, i64 0, !dbg !63 + %.extract774 = extractelement <4 x i32> %3193, i64 1, !dbg !63 + %.extract776 = extractelement <4 x i32> %3193, i64 2, !dbg !63 + %.extract778 = extractelement <4 x i32> %3193, i64 3, !dbg !63 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract772, i32 %.extract774, i32 %.extract776, i32 %.extract778, ptr addrspace(1) %2032, i1 %217) #7, !dbg !63 + %.extract780 = extractelement <4 x i32> %3202, i64 0, !dbg !63 + %.extract782 = extractelement <4 x i32> %3202, i64 1, !dbg !63 + %.extract784 = extractelement <4 x i32> %3202, i64 2, !dbg !63 + %.extract786 = extractelement <4 x i32> %3202, i64 3, !dbg !63 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract780, i32 %.extract782, i32 %.extract784, i32 %.extract786, ptr addrspace(1) %2033, i1 %223) #7, !dbg !63 + %.extract788 = extractelement <4 x i32> %3203, i64 0, !dbg !63 + %.extract790 = extractelement <4 x i32> %3203, i64 1, !dbg !63 + %.extract792 = extractelement <4 x i32> %3203, i64 2, !dbg !63 + %.extract794 = extractelement <4 x i32> %3203, i64 3, !dbg !63 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract788, i32 %.extract790, i32 %.extract792, i32 %.extract794, ptr addrspace(1) %2034, i1 %229) #7, !dbg !63 + %.extract796 = extractelement <4 x i32> %3212, i64 0, !dbg !63 + %.extract798 = extractelement <4 x i32> %3212, i64 1, !dbg !63 + %.extract800 = extractelement <4 x i32> %3212, i64 2, !dbg !63 + %.extract802 = extractelement <4 x i32> %3212, i64 3, !dbg !63 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract796, i32 %.extract798, i32 %.extract800, i32 %.extract802, ptr addrspace(1) %2035, i1 %235) #7, !dbg !63 + %.extract804 = extractelement <4 x i32> %3213, i64 0, !dbg !63 + %.extract806 = extractelement <4 x i32> %3213, i64 1, !dbg !63 + %.extract808 = extractelement <4 x i32> %3213, i64 2, !dbg !63 + %.extract810 = extractelement <4 x i32> %3213, i64 3, !dbg !63 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract804, i32 %.extract806, i32 %.extract808, i32 %.extract810, ptr addrspace(1) %2036, i1 %241) #7, !dbg !63 + %.extract812 = extractelement <4 x i32> %3222, i64 0, !dbg !63 + %.extract814 = extractelement <4 x i32> %3222, i64 1, !dbg !63 + %.extract816 = extractelement <4 x i32> %3222, i64 2, !dbg !63 + %.extract818 = extractelement <4 x i32> %3222, i64 3, !dbg !63 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract812, i32 %.extract814, i32 %.extract816, i32 %.extract818, ptr addrspace(1) %2037, i1 %247) #7, !dbg !63 + %.extract820 = extractelement <4 x i32> %3223, i64 0, !dbg !63 + %.extract822 = extractelement <4 x i32> %3223, i64 1, !dbg !63 + %.extract824 = extractelement <4 x i32> %3223, i64 2, !dbg !63 + %.extract826 = extractelement <4 x i32> %3223, i64 3, !dbg !63 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract820, i32 %.extract822, i32 %.extract824, i32 %.extract826, ptr addrspace(1) %2038, i1 %253) #7, !dbg !63 + %.extract828 = extractelement <4 x i32> %3232, i64 0, !dbg !63 + %.extract830 = extractelement <4 x i32> %3232, i64 1, !dbg !63 + %.extract832 = extractelement <4 x i32> %3232, i64 2, !dbg !63 + %.extract834 = extractelement <4 x i32> %3232, i64 3, !dbg !63 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract828, i32 %.extract830, i32 %.extract832, i32 %.extract834, ptr addrspace(1) %2039, i1 %259) #7, !dbg !63 + %.extract836 = extractelement <4 x i32> %3233, i64 0, !dbg !63 + %.extract838 = extractelement <4 x i32> %3233, i64 1, !dbg !63 + %.extract840 = extractelement <4 x i32> %3233, i64 2, !dbg !63 + %.extract842 = extractelement <4 x i32> %3233, i64 3, !dbg !63 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract836, i32 %.extract838, i32 %.extract840, i32 %.extract842, ptr addrspace(1) %2040, i1 %265) #7, !dbg !63 + %.extract844 = extractelement <4 x i32> %3242, i64 0, !dbg !63 + %.extract846 = extractelement <4 x i32> %3242, i64 1, !dbg !63 + %.extract848 = extractelement <4 x i32> %3242, i64 2, !dbg !63 + %.extract850 = extractelement <4 x i32> %3242, i64 3, !dbg !63 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract844, i32 %.extract846, i32 %.extract848, i32 %.extract850, ptr addrspace(1) %2041, i1 %271) #7, !dbg !63 + %.extract852 = extractelement <4 x i32> %3243, i64 0, !dbg !63 + %.extract854 = extractelement <4 x i32> %3243, i64 1, !dbg !63 + %.extract856 = extractelement <4 x i32> %3243, i64 2, !dbg !63 + %.extract858 = extractelement <4 x i32> %3243, i64 3, !dbg !63 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %.extract852, i32 %.extract854, i32 %.extract856, i32 %.extract858, ptr addrspace(1) %2042, i1 %277) #7, !dbg !63 + ret void, !dbg !64 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 65535) i32 @llvm.nvvm.read.ptx.sreg.ctaid.y() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #2 + +; Function Attrs: convergent nocallback nofree nounwind memory(argmem: read) +declare { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) readonly captures(none)) #3 + +; Function Attrs: convergent nocallback nounwind memory(argmem: write) +declare void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) writeonly captures(none), i32, i32, i32, i32) #4 + +; Function Attrs: convergent nocallback nofree nounwind memory(argmem: read) +declare { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) readonly captures(none)) #3 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #5 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.ex2.approx.ftz.f32(float) #6 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.ex2.approx.f32(float) #6 + +attributes #0 = { nounwind "nvvm.reqntid"="64" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind } +attributes #3 = { convergent nocallback nofree nounwind memory(argmem: read) } +attributes #4 = { convergent nocallback nounwind memory(argmem: write) } +attributes #5 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #6 = { mustprogress nocallback nofree nosync nounwind willreturn memory(none) } +attributes #7 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "chunk_o.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "chunk_bwd_kernel_dv_local", linkageName: "chunk_bwd_kernel_dv_local", scope: !1, file: !1, line: 466, type: !6, scopeLine: 466, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !10, !8, !8, !11, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 490, column: 30, scope: !5) +!14 = !DILocation(line: 490, column: 36, scope: !5) +!15 = !DILocation(line: 490, column: 61, scope: !5) +!16 = !DILocation(line: 490, column: 67, scope: !5) +!17 = !DILocation(line: 497, column: 25, scope: !5) +!18 = !DILocation(line: 491, column: 34, scope: !5) +!19 = !DILocation(line: 499, column: 16, scope: !5) +!20 = !DILocation(line: 499, column: 20, scope: !5) +!21 = !DILocation(line: 499, column: 40, scope: !5) +!22 = !DILocation(line: 499, column: 9, scope: !5) +!23 = !DILocation(line: 500, column: 9, scope: !5) +!24 = !DILocation(line: 504, column: 16, scope: !5) +!25 = !DILocation(line: 504, column: 34, scope: !5) +!26 = !DILocation(line: 504, column: 21, scope: !5) +!27 = !DILocation(line: 505, column: 16, scope: !5) +!28 = !DILocation(line: 511, column: 17, scope: !5) +!29 = !DILocation(line: 512, column: 22, scope: !5) +!30 = !DILocation(line: 513, column: 26, scope: !5) +!31 = !DILocation(line: 529, column: 39, scope: !5) +!32 = !DILocation(line: 47, column: 33, scope: !33, inlinedAt: !35) +!33 = distinct !DILexicalBlockFile(scope: !5, file: !34, discriminator: 0) +!34 = !DIFile(filename: "op.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils") +!35 = !DILocation(line: 529, column: 24, scope: !36) +!36 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!37 = !DILocation(line: 520, column: 42, scope: !5) +!38 = !DILocation(line: 522, column: 38, scope: !5) +!39 = !DILocation(line: 522, column: 22, scope: !5) +!40 = !DILocation(line: 522, column: 45, scope: !5) +!41 = !DILocation(line: 523, column: 22, scope: !5) +!42 = !DILocation(line: 523, column: 37, scope: !5) +!43 = !DILocation(line: 525, column: 26, scope: !5) +!44 = !DILocation(line: 526, column: 26, scope: !5) +!45 = !DILocation(line: 527, column: 31, scope: !5) +!46 = !DILocation(line: 502, column: 10, scope: !5) +!47 = !DILocation(line: 501, column: 10, scope: !5) +!48 = !DILocation(line: 527, column: 38, scope: !5) +!49 = !DILocation(line: 527, column: 19, scope: !5) +!50 = !DILocation(line: 529, column: 19, scope: !5) +!51 = !DILocation(line: 531, column: 35, scope: !5) +!52 = !DILocation(line: 530, column: 27, scope: !5) +!53 = !DILocation(line: 530, column: 44, scope: !5) +!54 = !DILocation(line: 531, column: 29, scope: !5) +!55 = !DILocation(line: 530, column: 59, scope: !5) +!56 = !DILocation(line: 536, column: 20, scope: !5) +!57 = !DILocation(line: 536, column: 44, scope: !5) +!58 = !DILocation(line: 537, column: 20, scope: !5) +!59 = !DILocation(line: 537, column: 44, scope: !5) +!60 = !DILocation(line: 538, column: 23, scope: !5) +!61 = !DILocation(line: 539, column: 42, scope: !5) +!62 = !DILocation(line: 540, column: 31, scope: !5) +!63 = !DILocation(line: 540, column: 23, scope: !5) +!64 = !DILocation(line: 533, column: 4, scope: !5) diff --git a/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.ptx b/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.ptx new file mode 100644 index 0000000000000000000000000000000000000000..7edaa5cccdebc5c0c6f8e67fcb44aa7f5ec32a9d --- /dev/null +++ b/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.ptx @@ -0,0 +1,2843 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl chunk_bwd_kernel_dv_local // -- Begin function chunk_bwd_kernel_dv_local +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; + // @chunk_bwd_kernel_dv_local +.visible .entry chunk_bwd_kernel_dv_local( + .param .u64 .ptr .global .align 1 chunk_bwd_kernel_dv_local_param_0, + .param .u64 .ptr .global .align 1 chunk_bwd_kernel_dv_local_param_1, + .param .u64 .ptr .global .align 1 chunk_bwd_kernel_dv_local_param_2, + .param .u64 .ptr .global .align 1 chunk_bwd_kernel_dv_local_param_3, + .param .u64 .ptr .global .align 1 chunk_bwd_kernel_dv_local_param_4, + .param .f32 chunk_bwd_kernel_dv_local_param_5, + .param .u32 chunk_bwd_kernel_dv_local_param_6, + .param .u64 .ptr .global .align 1 chunk_bwd_kernel_dv_local_param_7, + .param .u64 .ptr .global .align 1 chunk_bwd_kernel_dv_local_param_8 +) +.reqntid 64 +{ + .reg .pred %p<47>; + .reg .b16 %rs<137>; + .reg .b32 %r<1149>; + .reg .b64 %rd<300>; + .loc 1 466 0 // chunk_o.py:466:0 +$L__func_begin0: + .loc 1 466 0 // chunk_o.py:466:0 + +// %bb.0: // %__nv_exp2f.exit + ld.param.b64 %rd66, [chunk_bwd_kernel_dv_local_param_0]; +$L__tmp0: + .loc 1 490 30 // chunk_o.py:490:30 + mov.u32 %r804, %ctaid.x; + ld.param.b64 %rd67, [chunk_bwd_kernel_dv_local_param_1]; + ld.param.b64 %rd68, [chunk_bwd_kernel_dv_local_param_2]; + ld.param.b64 %rd69, [chunk_bwd_kernel_dv_local_param_3]; + .loc 1 490 61 // chunk_o.py:490:61 + mov.u32 %r805, %ctaid.y; + .loc 1 490 67 // chunk_o.py:490:67 + cvt.u64.u32 %rd70, %r805; + ld.param.b64 %rd71, [chunk_bwd_kernel_dv_local_param_4]; + .loc 1 497 25 // chunk_o.py:497:25 + and.b64 %rd72, %rd70, 65504; + ld.param.b32 %r806, [chunk_bwd_kernel_dv_local_param_5]; + .loc 1 491 34 // chunk_o.py:491:34 + and.b64 %rd73, %rd70, 31; + ld.param.s32 %rd74, [chunk_bwd_kernel_dv_local_param_6]; + .loc 1 499 16 // chunk_o.py:499:16 + mul.lo.s64 %rd75, %rd72, %rd74; + .loc 1 499 20 // chunk_o.py:499:20 + or.b64 %rd76, %rd75, %rd73; + .loc 1 499 9 // chunk_o.py:499:9 + shl.b64 %rd77, %rd76, 8; + add.s64 %rd78, %rd66, %rd77; + .loc 1 500 9 // chunk_o.py:500:9 + add.s64 %rd79, %rd67, %rd77; + .loc 1 504 16 // chunk_o.py:504:16 + mul.wide.u32 %rd80, %r804, 64; + .loc 1 504 34 // chunk_o.py:504:34 + mov.u32 %r807, %tid.x; + and.b32 %r808, %r807, 48; + bfe.u32 %r809, %r807, 4, 2; + and.b32 %r810, %r807, 28; + and.b32 %r811, %r807, 3; + and.b32 %r812, %r807, 32; + and.b32 %r813, %r807, 63; + or.b32 %r814, %r809, 4; + or.b32 %r815, %r809, 8; + or.b32 %r816, %r809, 12; + or.b32 %r817, %r809, 16; + or.b32 %r818, %r809, 20; + or.b32 %r819, %r809, 24; + or.b32 %r820, %r809, 28; + or.b32 %r821, %r809, 32; + or.b32 %r822, %r809, 36; + or.b32 %r823, %r809, 40; + or.b32 %r824, %r809, 44; + or.b32 %r825, %r809, 48; + or.b32 %r826, %r809, 52; + or.b32 %r827, %r809, 56; + or.b32 %r828, %r809, 60; + .loc 1 504 21 // chunk_o.py:504:21 + cvt.u64.u32 %rd81, %r809; + cvt.u64.u32 %rd82, %r828; + cvt.u64.u32 %rd83, %r827; + cvt.u64.u32 %rd84, %r826; + cvt.u64.u32 %rd85, %r825; + cvt.u64.u32 %rd86, %r824; + cvt.u64.u32 %rd87, %r823; + cvt.u64.u32 %rd88, %r822; + cvt.u64.u32 %rd89, %r821; + cvt.u64.u32 %rd90, %r820; + cvt.u64.u32 %rd91, %r819; + cvt.u64.u32 %rd92, %r818; + cvt.u64.u32 %rd93, %r817; + cvt.u64.u32 %rd94, %r816; + cvt.u64.u32 %rd95, %r815; + cvt.u64.u32 %rd96, %r814; + cvt.u64.u32 %rd97, %r813; + or.b64 %rd98, %rd80, %rd81; + or.b64 %rd99, %rd80, %rd97; + or.b64 %rd100, %rd80, %rd96; + or.b64 %rd101, %rd80, %rd95; + or.b64 %rd102, %rd80, %rd94; + or.b64 %rd103, %rd80, %rd93; + or.b64 %rd104, %rd80, %rd92; + or.b64 %rd105, %rd80, %rd91; + or.b64 %rd106, %rd80, %rd90; + or.b64 %rd107, %rd80, %rd89; + or.b64 %rd108, %rd80, %rd88; + or.b64 %rd109, %rd80, %rd87; + or.b64 %rd110, %rd80, %rd86; + or.b64 %rd111, %rd80, %rd85; + or.b64 %rd112, %rd80, %rd84; + or.b64 %rd113, %rd80, %rd83; + or.b64 %rd114, %rd80, %rd82; + .loc 1 505 16 // chunk_o.py:505:16 + setp.lt.s64 %p2, %rd98, %rd74; + setp.lt.s64 %p16, %rd114, %rd74; + setp.lt.s64 %p15, %rd113, %rd74; + setp.lt.s64 %p14, %rd112, %rd74; + setp.lt.s64 %p13, %rd111, %rd74; + setp.lt.s64 %p12, %rd110, %rd74; + setp.lt.s64 %p11, %rd109, %rd74; + setp.lt.s64 %p10, %rd108, %rd74; + setp.lt.s64 %p17, %rd107, %rd74; + setp.lt.s64 %p9, %rd106, %rd74; + setp.lt.s64 %p8, %rd105, %rd74; + setp.lt.s64 %p7, %rd104, %rd74; + setp.lt.s64 %p6, %rd103, %rd74; + setp.lt.s64 %p5, %rd102, %rd74; + setp.lt.s64 %p4, %rd101, %rd74; + setp.lt.s64 %p3, %rd100, %rd74; + setp.lt.s64 %p1, %rd99, %rd74; + .loc 1 511 17 // chunk_o.py:511:17 + shl.b64 %rd115, %rd76, 2; + add.s64 %rd116, %rd68, %rd115; + .loc 1 512 22 // chunk_o.py:512:22 + shl.b64 %rd117, %rd99, 7; + add.s64 %rd1, %rd116, %rd117; + mov.b32 %r672, 0; + .loc 1 513 26 // chunk_o.py:513:26 + // begin inline asm + mov.u32 %r2, %r672; + @%p1 ld.global.b32 { %r2 }, [ %rd1 + 0 ]; + // end inline asm + .loc 1 529 39 // chunk_o.py:529:39 + and.b32 %r829, %r807, 7; + shl.b32 %r830, %r829, 4; + and.b32 %r831, %r807, 24; + shr.u32 %r832, %r831, 1; + shl.b32 %r833, %r812, 2; + mov.b32 %r834, global_smem; + add.s32 %r835, %r834, %r833; + add.s32 %r836, %r835, %r832; + add.s32 %r1, %r836, %r830; + // begin inline asm + st.shared.b32 [ %r1 + 0 ], %r2; + // end inline asm + bar.sync 0; + shl.b32 %r837, %r810, 2; + add.s32 %r838, %r834, %r837; + ld.shared.v4.b32 {%r839, %r840, %r841, %r842}, [%r838]; + ld.shared.v4.b32 {%r843, %r844, %r845, %r846}, [%r838+128]; +$L__tmp1: + .loc 2 47 33 // op.py:47:33 @[ chunk_o.py:529:24 ] + bar.sync 0; + shl.b32 %r847, %r807, 3; + and.b32 %r848, %r847, 112; + shl.b32 %r849, %r807, 2; + and.b32 %r850, %r849, 132; + and.b32 %r851, %r807, 16; + shr.u32 %r852, %r851, 1; + add.s32 %r853, %r834, %r848; + add.s32 %r854, %r853, %r850; + add.s32 %r3, %r854, %r852; + // begin inline asm + st.shared.b32 [ %r3 + 0 ], %r2; + // end inline asm + bar.sync 0; + shl.b32 %r855, %r811, 4; + shl.b32 %r856, %r812, 1; + add.s32 %r857, %r834, %r856; + add.s32 %r858, %r857, %r855; + ld.shared.v4.b32 {%r859, %r860, %r861, %r862}, [%r858]; + ld.shared.v4.b32 {%r863, %r864, %r865, %r866}, [%r858+128]; +$L__tmp2: + .loc 1 520 42 // chunk_o.py:520:42 + and.b32 %r867, %r807, 15; + shl.b32 %r868, %r867, 3; + .loc 1 522 22 // chunk_o.py:522:22 + shl.b64 %rd118, %rd98, 13; + add.s64 %rd119, %rd79, %rd118; + shl.b64 %rd120, %rd100, 13; + add.s64 %rd121, %rd79, %rd120; + shl.b64 %rd122, %rd101, 13; + add.s64 %rd123, %rd79, %rd122; + shl.b64 %rd124, %rd102, 13; + add.s64 %rd125, %rd79, %rd124; + shl.b64 %rd126, %rd103, 13; + add.s64 %rd127, %rd79, %rd126; + shl.b64 %rd128, %rd104, 13; + add.s64 %rd129, %rd79, %rd128; + shl.b64 %rd130, %rd105, 13; + add.s64 %rd131, %rd79, %rd130; + shl.b64 %rd132, %rd106, 13; + add.s64 %rd133, %rd79, %rd132; + shl.b64 %rd134, %rd107, 13; + add.s64 %rd135, %rd79, %rd134; + shl.b64 %rd136, %rd108, 13; + add.s64 %rd137, %rd79, %rd136; + shl.b64 %rd138, %rd109, 13; + add.s64 %rd139, %rd79, %rd138; + shl.b64 %rd140, %rd110, 13; + add.s64 %rd141, %rd79, %rd140; + shl.b64 %rd142, %rd111, 13; + add.s64 %rd143, %rd79, %rd142; + shl.b64 %rd144, %rd112, 13; + add.s64 %rd145, %rd79, %rd144; + shl.b64 %rd146, %rd113, 13; + add.s64 %rd147, %rd79, %rd146; + shl.b64 %rd148, %rd114, 13; + add.s64 %rd149, %rd79, %rd148; + .loc 1 522 45 // chunk_o.py:522:45 + mul.wide.u32 %rd150, %r868, 2; + add.s64 %rd2, %rd119, %rd150; + add.s64 %rd3, %rd121, %rd150; + add.s64 %rd4, %rd123, %rd150; + add.s64 %rd5, %rd125, %rd150; + add.s64 %rd6, %rd127, %rd150; + add.s64 %rd7, %rd129, %rd150; + add.s64 %rd8, %rd131, %rd150; + add.s64 %rd9, %rd133, %rd150; + add.s64 %rd10, %rd135, %rd150; + add.s64 %rd11, %rd137, %rd150; + add.s64 %rd12, %rd139, %rd150; + add.s64 %rd13, %rd141, %rd150; + add.s64 %rd14, %rd143, %rd150; + add.s64 %rd15, %rd145, %rd150; + add.s64 %rd16, %rd147, %rd150; + add.s64 %rd17, %rd149, %rd150; + .loc 1 523 22 // chunk_o.py:523:22 + add.s64 %rd151, %rd78, %rd150; + .loc 1 523 37 // chunk_o.py:523:37 + add.s64 %rd18, %rd151, %rd118; + add.s64 %rd19, %rd151, %rd120; + add.s64 %rd20, %rd151, %rd122; + add.s64 %rd21, %rd151, %rd124; + add.s64 %rd22, %rd151, %rd126; + add.s64 %rd23, %rd151, %rd128; + add.s64 %rd24, %rd151, %rd130; + add.s64 %rd25, %rd151, %rd132; + add.s64 %rd26, %rd151, %rd134; + add.s64 %rd27, %rd151, %rd136; + add.s64 %rd28, %rd151, %rd138; + add.s64 %rd29, %rd151, %rd140; + add.s64 %rd30, %rd151, %rd142; + add.s64 %rd31, %rd151, %rd144; + add.s64 %rd32, %rd151, %rd146; + add.s64 %rd33, %rd151, %rd148; + .loc 1 525 26 // chunk_o.py:525:26 + // begin inline asm + mov.u32 %r5, %r672; + mov.u32 %r6, %r672; + mov.u32 %r7, %r672; + mov.u32 %r8, %r672; + @%p2 ld.global.v4.b32 { %r5, %r6, %r7, %r8 }, [ %rd2 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r45, %r672; + mov.u32 %r46, %r672; + mov.u32 %r47, %r672; + mov.u32 %r48, %r672; + @%p3 ld.global.v4.b32 { %r45, %r46, %r47, %r48 }, [ %rd3 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r10, %r672; + mov.u32 %r11, %r672; + mov.u32 %r12, %r672; + mov.u32 %r13, %r672; + @%p4 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd4 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r49, %r672; + mov.u32 %r50, %r672; + mov.u32 %r51, %r672; + mov.u32 %r52, %r672; + @%p5 ld.global.v4.b32 { %r49, %r50, %r51, %r52 }, [ %rd5 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r15, %r672; + mov.u32 %r16, %r672; + mov.u32 %r17, %r672; + mov.u32 %r18, %r672; + @%p6 ld.global.v4.b32 { %r15, %r16, %r17, %r18 }, [ %rd6 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r53, %r672; + mov.u32 %r54, %r672; + mov.u32 %r55, %r672; + mov.u32 %r56, %r672; + @%p7 ld.global.v4.b32 { %r53, %r54, %r55, %r56 }, [ %rd7 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r20, %r672; + mov.u32 %r21, %r672; + mov.u32 %r22, %r672; + mov.u32 %r23, %r672; + @%p8 ld.global.v4.b32 { %r20, %r21, %r22, %r23 }, [ %rd8 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r57, %r672; + mov.u32 %r58, %r672; + mov.u32 %r59, %r672; + mov.u32 %r60, %r672; + @%p9 ld.global.v4.b32 { %r57, %r58, %r59, %r60 }, [ %rd9 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r25, %r672; + mov.u32 %r26, %r672; + mov.u32 %r27, %r672; + mov.u32 %r28, %r672; + @%p17 ld.global.v4.b32 { %r25, %r26, %r27, %r28 }, [ %rd10 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r61, %r672; + mov.u32 %r62, %r672; + mov.u32 %r63, %r672; + mov.u32 %r64, %r672; + @%p10 ld.global.v4.b32 { %r61, %r62, %r63, %r64 }, [ %rd11 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r30, %r672; + mov.u32 %r31, %r672; + mov.u32 %r32, %r672; + mov.u32 %r33, %r672; + @%p11 ld.global.v4.b32 { %r30, %r31, %r32, %r33 }, [ %rd12 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r65, %r672; + mov.u32 %r66, %r672; + mov.u32 %r67, %r672; + mov.u32 %r68, %r672; + @%p12 ld.global.v4.b32 { %r65, %r66, %r67, %r68 }, [ %rd13 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r35, %r672; + mov.u32 %r36, %r672; + mov.u32 %r37, %r672; + mov.u32 %r38, %r672; + @%p13 ld.global.v4.b32 { %r35, %r36, %r37, %r38 }, [ %rd14 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r69, %r672; + mov.u32 %r70, %r672; + mov.u32 %r71, %r672; + mov.u32 %r72, %r672; + @%p14 ld.global.v4.b32 { %r69, %r70, %r71, %r72 }, [ %rd15 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r40, %r672; + mov.u32 %r41, %r672; + mov.u32 %r42, %r672; + mov.u32 %r43, %r672; + @%p15 ld.global.v4.b32 { %r40, %r41, %r42, %r43 }, [ %rd16 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r73, %r672; + mov.u32 %r74, %r672; + mov.u32 %r75, %r672; + mov.u32 %r76, %r672; + @%p16 ld.global.v4.b32 { %r73, %r74, %r75, %r76 }, [ %rd17 + 0 ]; + // end inline asm + bar.sync 0; + shl.b32 %r869, %r807, 4; + and.b32 %r870, %r869, 1008; + xor.b32 %r871, %r870, %r808; + add.s32 %r4, %r834, %r871; + // begin inline asm + st.shared.v4.b32 [ %r4 + 0 ], { %r5, %r6, %r7, %r8 }; + // end inline asm + add.s32 %r9, %r4, 2048; + // begin inline asm + st.shared.v4.b32 [ %r9 + 0 ], { %r10, %r11, %r12, %r13 }; + // end inline asm + add.s32 %r14, %r4, 4096; + // begin inline asm + st.shared.v4.b32 [ %r14 + 0 ], { %r15, %r16, %r17, %r18 }; + // end inline asm + add.s32 %r19, %r4, 6144; + // begin inline asm + st.shared.v4.b32 [ %r19 + 0 ], { %r20, %r21, %r22, %r23 }; + // end inline asm + add.s32 %r24, %r4, 8192; + // begin inline asm + st.shared.v4.b32 [ %r24 + 0 ], { %r25, %r26, %r27, %r28 }; + // end inline asm + add.s32 %r29, %r4, 10240; + // begin inline asm + st.shared.v4.b32 [ %r29 + 0 ], { %r30, %r31, %r32, %r33 }; + // end inline asm + add.s32 %r34, %r4, 12288; + // begin inline asm + st.shared.v4.b32 [ %r34 + 0 ], { %r35, %r36, %r37, %r38 }; + // end inline asm + add.s32 %r39, %r4, 14336; + // begin inline asm + st.shared.v4.b32 [ %r39 + 0 ], { %r40, %r41, %r42, %r43 }; + // end inline asm + xor.b32 %r872, %r871, 64; + add.s32 %r873, %r834, %r872; + add.s32 %r44, %r873, 1024; + // begin inline asm + st.shared.v4.b32 [ %r44 + 0 ], { %r45, %r46, %r47, %r48 }; + // end inline asm + add.s32 %r449, %r873, 3072; + // begin inline asm + st.shared.v4.b32 [ %r449 + 0 ], { %r49, %r50, %r51, %r52 }; + // end inline asm + add.s32 %r454, %r873, 5120; + // begin inline asm + st.shared.v4.b32 [ %r454 + 0 ], { %r53, %r54, %r55, %r56 }; + // end inline asm + add.s32 %r459, %r873, 7168; + // begin inline asm + st.shared.v4.b32 [ %r459 + 0 ], { %r57, %r58, %r59, %r60 }; + // end inline asm + add.s32 %r464, %r873, 9216; + // begin inline asm + st.shared.v4.b32 [ %r464 + 0 ], { %r61, %r62, %r63, %r64 }; + // end inline asm + add.s32 %r469, %r873, 11264; + // begin inline asm + st.shared.v4.b32 [ %r469 + 0 ], { %r65, %r66, %r67, %r68 }; + // end inline asm + add.s32 %r474, %r873, 13312; + // begin inline asm + st.shared.v4.b32 [ %r474 + 0 ], { %r69, %r70, %r71, %r72 }; + // end inline asm + add.s32 %r479, %r873, 15360; + // begin inline asm + st.shared.v4.b32 [ %r479 + 0 ], { %r73, %r74, %r75, %r76 }; + // end inline asm + .loc 1 526 26 // chunk_o.py:526:26 + // begin inline asm + mov.u32 %r78, %r672; + mov.u32 %r79, %r672; + mov.u32 %r80, %r672; + mov.u32 %r81, %r672; + @%p2 ld.global.v4.b32 { %r78, %r79, %r80, %r81 }, [ %rd18 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r118, %r672; + mov.u32 %r119, %r672; + mov.u32 %r120, %r672; + mov.u32 %r121, %r672; + @%p3 ld.global.v4.b32 { %r118, %r119, %r120, %r121 }, [ %rd19 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r83, %r672; + mov.u32 %r84, %r672; + mov.u32 %r85, %r672; + mov.u32 %r86, %r672; + @%p4 ld.global.v4.b32 { %r83, %r84, %r85, %r86 }, [ %rd20 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r123, %r672; + mov.u32 %r124, %r672; + mov.u32 %r125, %r672; + mov.u32 %r126, %r672; + @%p5 ld.global.v4.b32 { %r123, %r124, %r125, %r126 }, [ %rd21 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r88, %r672; + mov.u32 %r89, %r672; + mov.u32 %r90, %r672; + mov.u32 %r91, %r672; + @%p6 ld.global.v4.b32 { %r88, %r89, %r90, %r91 }, [ %rd22 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r128, %r672; + mov.u32 %r129, %r672; + mov.u32 %r130, %r672; + mov.u32 %r131, %r672; + @%p7 ld.global.v4.b32 { %r128, %r129, %r130, %r131 }, [ %rd23 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r93, %r672; + mov.u32 %r94, %r672; + mov.u32 %r95, %r672; + mov.u32 %r96, %r672; + @%p8 ld.global.v4.b32 { %r93, %r94, %r95, %r96 }, [ %rd24 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r133, %r672; + mov.u32 %r134, %r672; + mov.u32 %r135, %r672; + mov.u32 %r136, %r672; + @%p9 ld.global.v4.b32 { %r133, %r134, %r135, %r136 }, [ %rd25 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r98, %r672; + mov.u32 %r99, %r672; + mov.u32 %r100, %r672; + mov.u32 %r101, %r672; + @%p17 ld.global.v4.b32 { %r98, %r99, %r100, %r101 }, [ %rd26 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r138, %r672; + mov.u32 %r139, %r672; + mov.u32 %r140, %r672; + mov.u32 %r141, %r672; + @%p10 ld.global.v4.b32 { %r138, %r139, %r140, %r141 }, [ %rd27 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r103, %r672; + mov.u32 %r104, %r672; + mov.u32 %r105, %r672; + mov.u32 %r106, %r672; + @%p11 ld.global.v4.b32 { %r103, %r104, %r105, %r106 }, [ %rd28 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r143, %r672; + mov.u32 %r144, %r672; + mov.u32 %r145, %r672; + mov.u32 %r146, %r672; + @%p12 ld.global.v4.b32 { %r143, %r144, %r145, %r146 }, [ %rd29 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r108, %r672; + mov.u32 %r109, %r672; + mov.u32 %r110, %r672; + mov.u32 %r111, %r672; + @%p13 ld.global.v4.b32 { %r108, %r109, %r110, %r111 }, [ %rd30 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r148, %r672; + mov.u32 %r149, %r672; + mov.u32 %r150, %r672; + mov.u32 %r151, %r672; + @%p14 ld.global.v4.b32 { %r148, %r149, %r150, %r151 }, [ %rd31 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r113, %r672; + mov.u32 %r114, %r672; + mov.u32 %r115, %r672; + mov.u32 %r116, %r672; + @%p15 ld.global.v4.b32 { %r113, %r114, %r115, %r116 }, [ %rd32 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r153, %r672; + mov.u32 %r154, %r672; + mov.u32 %r155, %r672; + mov.u32 %r156, %r672; + @%p16 ld.global.v4.b32 { %r153, %r154, %r155, %r156 }, [ %rd33 + 0 ]; + // end inline asm + add.s32 %r77, %r4, 16384; + // begin inline asm + st.shared.v4.b32 [ %r77 + 0 ], { %r78, %r79, %r80, %r81 }; + // end inline asm + add.s32 %r82, %r4, 18432; + // begin inline asm + st.shared.v4.b32 [ %r82 + 0 ], { %r83, %r84, %r85, %r86 }; + // end inline asm + add.s32 %r87, %r4, 20480; + // begin inline asm + st.shared.v4.b32 [ %r87 + 0 ], { %r88, %r89, %r90, %r91 }; + // end inline asm + add.s32 %r92, %r4, 22528; + // begin inline asm + st.shared.v4.b32 [ %r92 + 0 ], { %r93, %r94, %r95, %r96 }; + // end inline asm + add.s32 %r97, %r4, 24576; + // begin inline asm + st.shared.v4.b32 [ %r97 + 0 ], { %r98, %r99, %r100, %r101 }; + // end inline asm + add.s32 %r102, %r4, 26624; + // begin inline asm + st.shared.v4.b32 [ %r102 + 0 ], { %r103, %r104, %r105, %r106 }; + // end inline asm + add.s32 %r107, %r4, 28672; + // begin inline asm + st.shared.v4.b32 [ %r107 + 0 ], { %r108, %r109, %r110, %r111 }; + // end inline asm + add.s32 %r112, %r4, 30720; + // begin inline asm + st.shared.v4.b32 [ %r112 + 0 ], { %r113, %r114, %r115, %r116 }; + // end inline asm + add.s32 %r117, %r873, 17408; + // begin inline asm + st.shared.v4.b32 [ %r117 + 0 ], { %r118, %r119, %r120, %r121 }; + // end inline asm + add.s32 %r122, %r873, 19456; + // begin inline asm + st.shared.v4.b32 [ %r122 + 0 ], { %r123, %r124, %r125, %r126 }; + // end inline asm + add.s32 %r127, %r873, 21504; + // begin inline asm + st.shared.v4.b32 [ %r127 + 0 ], { %r128, %r129, %r130, %r131 }; + // end inline asm + add.s32 %r132, %r873, 23552; + // begin inline asm + st.shared.v4.b32 [ %r132 + 0 ], { %r133, %r134, %r135, %r136 }; + // end inline asm + add.s32 %r137, %r873, 25600; + // begin inline asm + st.shared.v4.b32 [ %r137 + 0 ], { %r138, %r139, %r140, %r141 }; + // end inline asm + add.s32 %r142, %r873, 27648; + // begin inline asm + st.shared.v4.b32 [ %r142 + 0 ], { %r143, %r144, %r145, %r146 }; + // end inline asm + add.s32 %r147, %r873, 29696; + // begin inline asm + st.shared.v4.b32 [ %r147 + 0 ], { %r148, %r149, %r150, %r151 }; + // end inline asm + add.s32 %r152, %r873, 31744; + // begin inline asm + st.shared.v4.b32 [ %r152 + 0 ], { %r153, %r154, %r155, %r156 }; + // end inline asm + .loc 1 525 26 // chunk_o.py:525:26 + bar.sync 0; + shl.b32 %r874, %r867, 8; + or.b32 %r875, %r874, %r830; + xor.b32 %r876, %r875, %r851; + add.s32 %r877, %r834, %r876; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r157, %r158, %r159, %r160}, [%r877]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r317, %r318, %r319, %r320}, [%r877+128]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r169, %r170, %r171, %r172}, [%r877+4096]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r329, %r330, %r331, %r332}, [%r877+4224]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r173, %r174, %r175, %r176}, [%r877+8192]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r333, %r334, %r335, %r336}, [%r877+8320]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r177, %r178, %r179, %r180}, [%r877+12288]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r337, %r338, %r339, %r340}, [%r877+12416]; + xor.b32 %r878, %r876, 32; + add.s32 %r879, %r834, %r878; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r185, %r186, %r187, %r188}, [%r879]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r341, %r342, %r343, %r344}, [%r879+128]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r213, %r214, %r215, %r216}, [%r879+4096]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r353, %r354, %r355, %r356}, [%r879+4224]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r233, %r234, %r235, %r236}, [%r879+8192]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r357, %r358, %r359, %r360}, [%r879+8320]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r253, %r254, %r255, %r256}, [%r879+12288]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r361, %r362, %r363, %r364}, [%r879+12416]; + xor.b32 %r880, %r876, 64; + add.s32 %r881, %r834, %r880; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r269, %r270, %r271, %r272}, [%r881]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r365, %r366, %r367, %r368}, [%r881+128]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r281, %r282, %r283, %r284}, [%r881+4096]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r377, %r378, %r379, %r380}, [%r881+4224]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r285, %r286, %r287, %r288}, [%r881+8192]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r381, %r382, %r383, %r384}, [%r881+8320]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r289, %r290, %r291, %r292}, [%r881+12288]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r385, %r386, %r387, %r388}, [%r881+12416]; + xor.b32 %r882, %r876, 96; + add.s32 %r883, %r834, %r882; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r293, %r294, %r295, %r296}, [%r883]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r389, %r390, %r391, %r392}, [%r883+128]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r305, %r306, %r307, %r308}, [%r883+4096]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r401, %r402, %r403, %r404}, [%r883+4224]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r309, %r310, %r311, %r312}, [%r883+8192]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r405, %r406, %r407, %r408}, [%r883+8320]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r313, %r314, %r315, %r316}, [%r883+12288]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r409, %r410, %r411, %r412}, [%r883+12416]; + .loc 1 526 26 // chunk_o.py:526:26 + shl.b32 %r884, %r829, 8; + shl.b32 %r885, %r831, 1; + shl.b32 %r886, %r812, 6; + or.b32 %r887, %r884, %r830; + xor.b32 %r888, %r887, %r885; + or.b32 %r889, %r888, %r886; + add.s32 %r890, %r834, %r889; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r161, %r162, %r189, %r190}, [%r890+16384]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r321, %r322, %r345, %r346}, [%r890+16512]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r163, %r164, %r195, %r196}, [%r890+20480]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r323, %r324, %r347, %r348}, [%r890+20608]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r165, %r166, %r201, %r202}, [%r890+24576]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r325, %r326, %r349, %r350}, [%r890+24704]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r167, %r168, %r207, %r208}, [%r890+28672]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r327, %r328, %r351, %r352}, [%r890+28800]; + xor.b32 %r891, %r889, 64; + add.s32 %r892, %r834, %r891; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r273, %r274, %r297, %r298}, [%r892+16384]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r369, %r370, %r393, %r394}, [%r892+16512]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r275, %r276, %r299, %r300}, [%r892+20480]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r371, %r372, %r395, %r396}, [%r892+20608]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r277, %r278, %r301, %r302}, [%r892+24576]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r373, %r374, %r397, %r398}, [%r892+24704]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r279, %r280, %r303, %r304}, [%r892+28672]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r375, %r376, %r399, %r400}, [%r892+28800]; + .loc 1 527 31 // chunk_o.py:527:31 + mov.b32 %r181, %r672; + mov.b32 %r182, %r672; + mov.b32 %r183, %r672; + mov.b32 %r184, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r181, %r182, %r183, %r184 }, { %r157, %r158, %r159, %r160 }, { %r161, %r162 }, { %r181, %r182, %r183, %r184 }; + // end inline asm + mov.b32 %r191, %r672; + mov.b32 %r192, %r672; + mov.b32 %r193, %r672; + mov.b32 %r194, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r191, %r192, %r193, %r194 }, { %r157, %r158, %r159, %r160 }, { %r163, %r164 }, { %r191, %r192, %r193, %r194 }; + // end inline asm + mov.b32 %r197, %r672; + mov.b32 %r198, %r672; + mov.b32 %r199, %r672; + mov.b32 %r200, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r197, %r198, %r199, %r200 }, { %r157, %r158, %r159, %r160 }, { %r165, %r166 }, { %r197, %r198, %r199, %r200 }; + // end inline asm + mov.b32 %r203, %r672; + mov.b32 %r204, %r672; + mov.b32 %r205, %r672; + mov.b32 %r206, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r203, %r204, %r205, %r206 }, { %r157, %r158, %r159, %r160 }, { %r167, %r168 }, { %r203, %r204, %r205, %r206 }; + // end inline asm + mov.b32 %r209, %r672; + mov.b32 %r210, %r672; + mov.b32 %r211, %r672; + mov.b32 %r212, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r209, %r210, %r211, %r212 }, { %r169, %r170, %r171, %r172 }, { %r161, %r162 }, { %r209, %r210, %r211, %r212 }; + // end inline asm + mov.b32 %r217, %r672; + mov.b32 %r218, %r672; + mov.b32 %r219, %r672; + mov.b32 %r220, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r217, %r218, %r219, %r220 }, { %r169, %r170, %r171, %r172 }, { %r163, %r164 }, { %r217, %r218, %r219, %r220 }; + // end inline asm + mov.b32 %r221, %r672; + mov.b32 %r222, %r672; + mov.b32 %r223, %r672; + mov.b32 %r224, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r221, %r222, %r223, %r224 }, { %r169, %r170, %r171, %r172 }, { %r165, %r166 }, { %r221, %r222, %r223, %r224 }; + // end inline asm + mov.b32 %r225, %r672; + mov.b32 %r226, %r672; + mov.b32 %r227, %r672; + mov.b32 %r228, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r225, %r226, %r227, %r228 }, { %r169, %r170, %r171, %r172 }, { %r167, %r168 }, { %r225, %r226, %r227, %r228 }; + // end inline asm + mov.b32 %r229, %r672; + mov.b32 %r230, %r672; + mov.b32 %r231, %r672; + mov.b32 %r232, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r229, %r230, %r231, %r232 }, { %r173, %r174, %r175, %r176 }, { %r161, %r162 }, { %r229, %r230, %r231, %r232 }; + // end inline asm + mov.b32 %r237, %r672; + mov.b32 %r238, %r672; + mov.b32 %r239, %r672; + mov.b32 %r240, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r237, %r238, %r239, %r240 }, { %r173, %r174, %r175, %r176 }, { %r163, %r164 }, { %r237, %r238, %r239, %r240 }; + // end inline asm + mov.b32 %r241, %r672; + mov.b32 %r242, %r672; + mov.b32 %r243, %r672; + mov.b32 %r244, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r241, %r242, %r243, %r244 }, { %r173, %r174, %r175, %r176 }, { %r165, %r166 }, { %r241, %r242, %r243, %r244 }; + // end inline asm + mov.b32 %r245, %r672; + mov.b32 %r246, %r672; + mov.b32 %r247, %r672; + mov.b32 %r248, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r245, %r246, %r247, %r248 }, { %r173, %r174, %r175, %r176 }, { %r167, %r168 }, { %r245, %r246, %r247, %r248 }; + // end inline asm + mov.b32 %r249, %r672; + mov.b32 %r250, %r672; + mov.b32 %r251, %r672; + mov.b32 %r252, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r249, %r250, %r251, %r252 }, { %r177, %r178, %r179, %r180 }, { %r161, %r162 }, { %r249, %r250, %r251, %r252 }; + // end inline asm + mov.b32 %r257, %r672; + mov.b32 %r258, %r672; + mov.b32 %r259, %r672; + mov.b32 %r260, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r257, %r258, %r259, %r260 }, { %r177, %r178, %r179, %r180 }, { %r163, %r164 }, { %r257, %r258, %r259, %r260 }; + // end inline asm + mov.b32 %r261, %r672; + mov.b32 %r262, %r672; + mov.b32 %r263, %r672; + mov.b32 %r264, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r261, %r262, %r263, %r264 }, { %r177, %r178, %r179, %r180 }, { %r165, %r166 }, { %r261, %r262, %r263, %r264 }; + // end inline asm + mov.b32 %r265, %r672; + mov.b32 %r266, %r672; + mov.b32 %r267, %r672; + mov.b32 %r268, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r265, %r266, %r267, %r268 }, { %r177, %r178, %r179, %r180 }, { %r167, %r168 }, { %r265, %r266, %r267, %r268 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r181, %r182, %r183, %r184 }, { %r185, %r186, %r187, %r188 }, { %r189, %r190 }, { %r181, %r182, %r183, %r184 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r191, %r192, %r193, %r194 }, { %r185, %r186, %r187, %r188 }, { %r195, %r196 }, { %r191, %r192, %r193, %r194 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r197, %r198, %r199, %r200 }, { %r185, %r186, %r187, %r188 }, { %r201, %r202 }, { %r197, %r198, %r199, %r200 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r203, %r204, %r205, %r206 }, { %r185, %r186, %r187, %r188 }, { %r207, %r208 }, { %r203, %r204, %r205, %r206 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r209, %r210, %r211, %r212 }, { %r213, %r214, %r215, %r216 }, { %r189, %r190 }, { %r209, %r210, %r211, %r212 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r217, %r218, %r219, %r220 }, { %r213, %r214, %r215, %r216 }, { %r195, %r196 }, { %r217, %r218, %r219, %r220 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r221, %r222, %r223, %r224 }, { %r213, %r214, %r215, %r216 }, { %r201, %r202 }, { %r221, %r222, %r223, %r224 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r225, %r226, %r227, %r228 }, { %r213, %r214, %r215, %r216 }, { %r207, %r208 }, { %r225, %r226, %r227, %r228 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r229, %r230, %r231, %r232 }, { %r233, %r234, %r235, %r236 }, { %r189, %r190 }, { %r229, %r230, %r231, %r232 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r237, %r238, %r239, %r240 }, { %r233, %r234, %r235, %r236 }, { %r195, %r196 }, { %r237, %r238, %r239, %r240 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r241, %r242, %r243, %r244 }, { %r233, %r234, %r235, %r236 }, { %r201, %r202 }, { %r241, %r242, %r243, %r244 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r245, %r246, %r247, %r248 }, { %r233, %r234, %r235, %r236 }, { %r207, %r208 }, { %r245, %r246, %r247, %r248 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r249, %r250, %r251, %r252 }, { %r253, %r254, %r255, %r256 }, { %r189, %r190 }, { %r249, %r250, %r251, %r252 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r257, %r258, %r259, %r260 }, { %r253, %r254, %r255, %r256 }, { %r195, %r196 }, { %r257, %r258, %r259, %r260 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r261, %r262, %r263, %r264 }, { %r253, %r254, %r255, %r256 }, { %r201, %r202 }, { %r261, %r262, %r263, %r264 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r265, %r266, %r267, %r268 }, { %r253, %r254, %r255, %r256 }, { %r207, %r208 }, { %r265, %r266, %r267, %r268 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r181, %r182, %r183, %r184 }, { %r269, %r270, %r271, %r272 }, { %r273, %r274 }, { %r181, %r182, %r183, %r184 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r191, %r192, %r193, %r194 }, { %r269, %r270, %r271, %r272 }, { %r275, %r276 }, { %r191, %r192, %r193, %r194 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r197, %r198, %r199, %r200 }, { %r269, %r270, %r271, %r272 }, { %r277, %r278 }, { %r197, %r198, %r199, %r200 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r203, %r204, %r205, %r206 }, { %r269, %r270, %r271, %r272 }, { %r279, %r280 }, { %r203, %r204, %r205, %r206 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r209, %r210, %r211, %r212 }, { %r281, %r282, %r283, %r284 }, { %r273, %r274 }, { %r209, %r210, %r211, %r212 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r217, %r218, %r219, %r220 }, { %r281, %r282, %r283, %r284 }, { %r275, %r276 }, { %r217, %r218, %r219, %r220 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r221, %r222, %r223, %r224 }, { %r281, %r282, %r283, %r284 }, { %r277, %r278 }, { %r221, %r222, %r223, %r224 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r225, %r226, %r227, %r228 }, { %r281, %r282, %r283, %r284 }, { %r279, %r280 }, { %r225, %r226, %r227, %r228 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r229, %r230, %r231, %r232 }, { %r285, %r286, %r287, %r288 }, { %r273, %r274 }, { %r229, %r230, %r231, %r232 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r237, %r238, %r239, %r240 }, { %r285, %r286, %r287, %r288 }, { %r275, %r276 }, { %r237, %r238, %r239, %r240 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r241, %r242, %r243, %r244 }, { %r285, %r286, %r287, %r288 }, { %r277, %r278 }, { %r241, %r242, %r243, %r244 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r245, %r246, %r247, %r248 }, { %r285, %r286, %r287, %r288 }, { %r279, %r280 }, { %r245, %r246, %r247, %r248 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r249, %r250, %r251, %r252 }, { %r289, %r290, %r291, %r292 }, { %r273, %r274 }, { %r249, %r250, %r251, %r252 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r257, %r258, %r259, %r260 }, { %r289, %r290, %r291, %r292 }, { %r275, %r276 }, { %r257, %r258, %r259, %r260 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r261, %r262, %r263, %r264 }, { %r289, %r290, %r291, %r292 }, { %r277, %r278 }, { %r261, %r262, %r263, %r264 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r265, %r266, %r267, %r268 }, { %r289, %r290, %r291, %r292 }, { %r279, %r280 }, { %r265, %r266, %r267, %r268 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r181, %r182, %r183, %r184 }, { %r293, %r294, %r295, %r296 }, { %r297, %r298 }, { %r181, %r182, %r183, %r184 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r191, %r192, %r193, %r194 }, { %r293, %r294, %r295, %r296 }, { %r299, %r300 }, { %r191, %r192, %r193, %r194 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r197, %r198, %r199, %r200 }, { %r293, %r294, %r295, %r296 }, { %r301, %r302 }, { %r197, %r198, %r199, %r200 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r203, %r204, %r205, %r206 }, { %r293, %r294, %r295, %r296 }, { %r303, %r304 }, { %r203, %r204, %r205, %r206 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r209, %r210, %r211, %r212 }, { %r305, %r306, %r307, %r308 }, { %r297, %r298 }, { %r209, %r210, %r211, %r212 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r217, %r218, %r219, %r220 }, { %r305, %r306, %r307, %r308 }, { %r299, %r300 }, { %r217, %r218, %r219, %r220 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r221, %r222, %r223, %r224 }, { %r305, %r306, %r307, %r308 }, { %r301, %r302 }, { %r221, %r222, %r223, %r224 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r225, %r226, %r227, %r228 }, { %r305, %r306, %r307, %r308 }, { %r303, %r304 }, { %r225, %r226, %r227, %r228 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r229, %r230, %r231, %r232 }, { %r309, %r310, %r311, %r312 }, { %r297, %r298 }, { %r229, %r230, %r231, %r232 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r237, %r238, %r239, %r240 }, { %r309, %r310, %r311, %r312 }, { %r299, %r300 }, { %r237, %r238, %r239, %r240 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r241, %r242, %r243, %r244 }, { %r309, %r310, %r311, %r312 }, { %r301, %r302 }, { %r241, %r242, %r243, %r244 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r245, %r246, %r247, %r248 }, { %r309, %r310, %r311, %r312 }, { %r303, %r304 }, { %r245, %r246, %r247, %r248 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r249, %r250, %r251, %r252 }, { %r313, %r314, %r315, %r316 }, { %r297, %r298 }, { %r249, %r250, %r251, %r252 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r257, %r258, %r259, %r260 }, { %r313, %r314, %r315, %r316 }, { %r299, %r300 }, { %r257, %r258, %r259, %r260 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r261, %r262, %r263, %r264 }, { %r313, %r314, %r315, %r316 }, { %r301, %r302 }, { %r261, %r262, %r263, %r264 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r265, %r266, %r267, %r268 }, { %r313, %r314, %r315, %r316 }, { %r303, %r304 }, { %r265, %r266, %r267, %r268 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r181, %r182, %r183, %r184 }, { %r317, %r318, %r319, %r320 }, { %r321, %r322 }, { %r181, %r182, %r183, %r184 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r191, %r192, %r193, %r194 }, { %r317, %r318, %r319, %r320 }, { %r323, %r324 }, { %r191, %r192, %r193, %r194 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r197, %r198, %r199, %r200 }, { %r317, %r318, %r319, %r320 }, { %r325, %r326 }, { %r197, %r198, %r199, %r200 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r203, %r204, %r205, %r206 }, { %r317, %r318, %r319, %r320 }, { %r327, %r328 }, { %r203, %r204, %r205, %r206 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r209, %r210, %r211, %r212 }, { %r329, %r330, %r331, %r332 }, { %r321, %r322 }, { %r209, %r210, %r211, %r212 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r217, %r218, %r219, %r220 }, { %r329, %r330, %r331, %r332 }, { %r323, %r324 }, { %r217, %r218, %r219, %r220 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r221, %r222, %r223, %r224 }, { %r329, %r330, %r331, %r332 }, { %r325, %r326 }, { %r221, %r222, %r223, %r224 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r225, %r226, %r227, %r228 }, { %r329, %r330, %r331, %r332 }, { %r327, %r328 }, { %r225, %r226, %r227, %r228 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r229, %r230, %r231, %r232 }, { %r333, %r334, %r335, %r336 }, { %r321, %r322 }, { %r229, %r230, %r231, %r232 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r237, %r238, %r239, %r240 }, { %r333, %r334, %r335, %r336 }, { %r323, %r324 }, { %r237, %r238, %r239, %r240 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r241, %r242, %r243, %r244 }, { %r333, %r334, %r335, %r336 }, { %r325, %r326 }, { %r241, %r242, %r243, %r244 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r245, %r246, %r247, %r248 }, { %r333, %r334, %r335, %r336 }, { %r327, %r328 }, { %r245, %r246, %r247, %r248 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r249, %r250, %r251, %r252 }, { %r337, %r338, %r339, %r340 }, { %r321, %r322 }, { %r249, %r250, %r251, %r252 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r257, %r258, %r259, %r260 }, { %r337, %r338, %r339, %r340 }, { %r323, %r324 }, { %r257, %r258, %r259, %r260 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r261, %r262, %r263, %r264 }, { %r337, %r338, %r339, %r340 }, { %r325, %r326 }, { %r261, %r262, %r263, %r264 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r265, %r266, %r267, %r268 }, { %r337, %r338, %r339, %r340 }, { %r327, %r328 }, { %r265, %r266, %r267, %r268 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r181, %r182, %r183, %r184 }, { %r341, %r342, %r343, %r344 }, { %r345, %r346 }, { %r181, %r182, %r183, %r184 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r191, %r192, %r193, %r194 }, { %r341, %r342, %r343, %r344 }, { %r347, %r348 }, { %r191, %r192, %r193, %r194 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r197, %r198, %r199, %r200 }, { %r341, %r342, %r343, %r344 }, { %r349, %r350 }, { %r197, %r198, %r199, %r200 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r203, %r204, %r205, %r206 }, { %r341, %r342, %r343, %r344 }, { %r351, %r352 }, { %r203, %r204, %r205, %r206 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r209, %r210, %r211, %r212 }, { %r353, %r354, %r355, %r356 }, { %r345, %r346 }, { %r209, %r210, %r211, %r212 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r217, %r218, %r219, %r220 }, { %r353, %r354, %r355, %r356 }, { %r347, %r348 }, { %r217, %r218, %r219, %r220 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r221, %r222, %r223, %r224 }, { %r353, %r354, %r355, %r356 }, { %r349, %r350 }, { %r221, %r222, %r223, %r224 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r225, %r226, %r227, %r228 }, { %r353, %r354, %r355, %r356 }, { %r351, %r352 }, { %r225, %r226, %r227, %r228 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r229, %r230, %r231, %r232 }, { %r357, %r358, %r359, %r360 }, { %r345, %r346 }, { %r229, %r230, %r231, %r232 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r237, %r238, %r239, %r240 }, { %r357, %r358, %r359, %r360 }, { %r347, %r348 }, { %r237, %r238, %r239, %r240 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r241, %r242, %r243, %r244 }, { %r357, %r358, %r359, %r360 }, { %r349, %r350 }, { %r241, %r242, %r243, %r244 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r245, %r246, %r247, %r248 }, { %r357, %r358, %r359, %r360 }, { %r351, %r352 }, { %r245, %r246, %r247, %r248 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r249, %r250, %r251, %r252 }, { %r361, %r362, %r363, %r364 }, { %r345, %r346 }, { %r249, %r250, %r251, %r252 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r257, %r258, %r259, %r260 }, { %r361, %r362, %r363, %r364 }, { %r347, %r348 }, { %r257, %r258, %r259, %r260 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r261, %r262, %r263, %r264 }, { %r361, %r362, %r363, %r364 }, { %r349, %r350 }, { %r261, %r262, %r263, %r264 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r265, %r266, %r267, %r268 }, { %r361, %r362, %r363, %r364 }, { %r351, %r352 }, { %r265, %r266, %r267, %r268 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r181, %r182, %r183, %r184 }, { %r365, %r366, %r367, %r368 }, { %r369, %r370 }, { %r181, %r182, %r183, %r184 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r191, %r192, %r193, %r194 }, { %r365, %r366, %r367, %r368 }, { %r371, %r372 }, { %r191, %r192, %r193, %r194 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r197, %r198, %r199, %r200 }, { %r365, %r366, %r367, %r368 }, { %r373, %r374 }, { %r197, %r198, %r199, %r200 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r203, %r204, %r205, %r206 }, { %r365, %r366, %r367, %r368 }, { %r375, %r376 }, { %r203, %r204, %r205, %r206 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r209, %r210, %r211, %r212 }, { %r377, %r378, %r379, %r380 }, { %r369, %r370 }, { %r209, %r210, %r211, %r212 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r217, %r218, %r219, %r220 }, { %r377, %r378, %r379, %r380 }, { %r371, %r372 }, { %r217, %r218, %r219, %r220 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r221, %r222, %r223, %r224 }, { %r377, %r378, %r379, %r380 }, { %r373, %r374 }, { %r221, %r222, %r223, %r224 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r225, %r226, %r227, %r228 }, { %r377, %r378, %r379, %r380 }, { %r375, %r376 }, { %r225, %r226, %r227, %r228 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r229, %r230, %r231, %r232 }, { %r381, %r382, %r383, %r384 }, { %r369, %r370 }, { %r229, %r230, %r231, %r232 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r237, %r238, %r239, %r240 }, { %r381, %r382, %r383, %r384 }, { %r371, %r372 }, { %r237, %r238, %r239, %r240 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r241, %r242, %r243, %r244 }, { %r381, %r382, %r383, %r384 }, { %r373, %r374 }, { %r241, %r242, %r243, %r244 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r245, %r246, %r247, %r248 }, { %r381, %r382, %r383, %r384 }, { %r375, %r376 }, { %r245, %r246, %r247, %r248 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r249, %r250, %r251, %r252 }, { %r385, %r386, %r387, %r388 }, { %r369, %r370 }, { %r249, %r250, %r251, %r252 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r257, %r258, %r259, %r260 }, { %r385, %r386, %r387, %r388 }, { %r371, %r372 }, { %r257, %r258, %r259, %r260 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r261, %r262, %r263, %r264 }, { %r385, %r386, %r387, %r388 }, { %r373, %r374 }, { %r261, %r262, %r263, %r264 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r265, %r266, %r267, %r268 }, { %r385, %r386, %r387, %r388 }, { %r375, %r376 }, { %r265, %r266, %r267, %r268 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r181, %r182, %r183, %r184 }, { %r389, %r390, %r391, %r392 }, { %r393, %r394 }, { %r181, %r182, %r183, %r184 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r191, %r192, %r193, %r194 }, { %r389, %r390, %r391, %r392 }, { %r395, %r396 }, { %r191, %r192, %r193, %r194 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r197, %r198, %r199, %r200 }, { %r389, %r390, %r391, %r392 }, { %r397, %r398 }, { %r197, %r198, %r199, %r200 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r203, %r204, %r205, %r206 }, { %r389, %r390, %r391, %r392 }, { %r399, %r400 }, { %r203, %r204, %r205, %r206 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r209, %r210, %r211, %r212 }, { %r401, %r402, %r403, %r404 }, { %r393, %r394 }, { %r209, %r210, %r211, %r212 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r217, %r218, %r219, %r220 }, { %r401, %r402, %r403, %r404 }, { %r395, %r396 }, { %r217, %r218, %r219, %r220 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r221, %r222, %r223, %r224 }, { %r401, %r402, %r403, %r404 }, { %r397, %r398 }, { %r221, %r222, %r223, %r224 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r225, %r226, %r227, %r228 }, { %r401, %r402, %r403, %r404 }, { %r399, %r400 }, { %r225, %r226, %r227, %r228 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r229, %r230, %r231, %r232 }, { %r405, %r406, %r407, %r408 }, { %r393, %r394 }, { %r229, %r230, %r231, %r232 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r237, %r238, %r239, %r240 }, { %r405, %r406, %r407, %r408 }, { %r395, %r396 }, { %r237, %r238, %r239, %r240 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r241, %r242, %r243, %r244 }, { %r405, %r406, %r407, %r408 }, { %r397, %r398 }, { %r241, %r242, %r243, %r244 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r245, %r246, %r247, %r248 }, { %r405, %r406, %r407, %r408 }, { %r399, %r400 }, { %r245, %r246, %r247, %r248 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r249, %r250, %r251, %r252 }, { %r409, %r410, %r411, %r412 }, { %r393, %r394 }, { %r249, %r250, %r251, %r252 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r257, %r258, %r259, %r260 }, { %r409, %r410, %r411, %r412 }, { %r395, %r396 }, { %r257, %r258, %r259, %r260 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r261, %r262, %r263, %r264 }, { %r409, %r410, %r411, %r412 }, { %r397, %r398 }, { %r261, %r262, %r263, %r264 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r265, %r266, %r267, %r268 }, { %r409, %r410, %r411, %r412 }, { %r399, %r400 }, { %r265, %r266, %r267, %r268 }; + // end inline asm + .loc 1 529 39 // chunk_o.py:529:39 + sub.f32 %r893, %r859, %r839; + sub.f32 %r894, %r860, %r839; + sub.f32 %r895, %r859, %r840; + sub.f32 %r896, %r860, %r840; + sub.f32 %r897, %r861, %r839; + sub.f32 %r898, %r862, %r839; + sub.f32 %r899, %r861, %r840; + sub.f32 %r900, %r862, %r840; + sub.f32 %r901, %r863, %r839; + sub.f32 %r902, %r864, %r839; + sub.f32 %r903, %r863, %r840; + sub.f32 %r904, %r864, %r840; + sub.f32 %r905, %r865, %r839; + sub.f32 %r906, %r866, %r839; + sub.f32 %r907, %r865, %r840; + sub.f32 %r908, %r866, %r840; + sub.f32 %r909, %r861, %r841; + sub.f32 %r910, %r862, %r841; + sub.f32 %r911, %r861, %r842; + sub.f32 %r912, %r862, %r842; + sub.f32 %r913, %r863, %r841; + sub.f32 %r914, %r864, %r841; + sub.f32 %r915, %r863, %r842; + sub.f32 %r916, %r864, %r842; + sub.f32 %r917, %r865, %r841; + sub.f32 %r918, %r866, %r841; + sub.f32 %r919, %r865, %r842; + sub.f32 %r920, %r866, %r842; + sub.f32 %r921, %r863, %r843; + sub.f32 %r922, %r864, %r843; + sub.f32 %r923, %r863, %r844; + sub.f32 %r924, %r864, %r844; + sub.f32 %r925, %r865, %r843; + sub.f32 %r926, %r866, %r843; + sub.f32 %r927, %r865, %r844; + sub.f32 %r928, %r866, %r844; + sub.f32 %r929, %r865, %r845; + sub.f32 %r930, %r866, %r845; + sub.f32 %r931, %r865, %r846; + sub.f32 %r932, %r866, %r846; +$L__tmp3: + .loc 2 47 33 // op.py:47:33 @[ chunk_o.py:529:24 ] + ex2.approx.ftz.f32 %r933, %r893; + ex2.approx.ftz.f32 %r934, %r894; + ex2.approx.ftz.f32 %r935, %r895; + ex2.approx.ftz.f32 %r936, %r896; + ex2.approx.ftz.f32 %r937, %r897; + ex2.approx.ftz.f32 %r938, %r898; + ex2.approx.ftz.f32 %r939, %r899; + ex2.approx.ftz.f32 %r940, %r900; + ex2.approx.ftz.f32 %r941, %r901; + ex2.approx.ftz.f32 %r942, %r902; + ex2.approx.ftz.f32 %r943, %r903; + ex2.approx.ftz.f32 %r944, %r904; + ex2.approx.ftz.f32 %r945, %r905; + ex2.approx.ftz.f32 %r946, %r906; + ex2.approx.ftz.f32 %r947, %r907; + ex2.approx.ftz.f32 %r948, %r908; + ex2.approx.ftz.f32 %r949, %r909; + ex2.approx.ftz.f32 %r950, %r910; + ex2.approx.ftz.f32 %r951, %r911; + ex2.approx.ftz.f32 %r952, %r912; + ex2.approx.ftz.f32 %r953, %r913; + ex2.approx.ftz.f32 %r954, %r914; + ex2.approx.ftz.f32 %r955, %r915; + ex2.approx.ftz.f32 %r956, %r916; + ex2.approx.ftz.f32 %r957, %r917; + ex2.approx.ftz.f32 %r958, %r918; + ex2.approx.ftz.f32 %r959, %r919; + ex2.approx.ftz.f32 %r960, %r920; + ex2.approx.ftz.f32 %r961, %r921; + ex2.approx.ftz.f32 %r962, %r922; + ex2.approx.ftz.f32 %r963, %r923; + ex2.approx.ftz.f32 %r964, %r924; + ex2.approx.ftz.f32 %r965, %r925; + ex2.approx.ftz.f32 %r966, %r926; + ex2.approx.ftz.f32 %r967, %r927; + ex2.approx.ftz.f32 %r968, %r928; + ex2.approx.ftz.f32 %r969, %r929; + ex2.approx.ftz.f32 %r970, %r930; + ex2.approx.ftz.f32 %r971, %r931; + ex2.approx.ftz.f32 %r972, %r932; +$L__tmp4: + .loc 1 504 34 // chunk_o.py:504:34 + shr.u32 %r973, %r812, 2; + shl.b32 %r974, %r811, 1; + .loc 1 502 10 // chunk_o.py:502:10 + add.s64 %rd152, %rd71, %rd77; + .loc 1 501 10 // chunk_o.py:501:10 + add.s64 %rd153, %rd69, %rd77; + .loc 1 527 38 // chunk_o.py:527:38 + mov.b64 %rd154, {%r806, %r806}; + mov.b64 %rd155, {%r183, %r184}; + mov.b32 %r975, 0f00000000; + .loc 1 527 19 // chunk_o.py:527:19 + mov.b64 %rd156, {%r975, %r975}; + fma.rn.f32x2 %rd157, %rd154, %rd155, %rd156; + .loc 1 529 19 // chunk_o.py:529:19 + mov.b64 %rd158, {%r935, %r936}; + mul.f32x2 %rd159, %rd157, %rd158; + .loc 1 527 38 // chunk_o.py:527:38 + mov.b64 %rd160, {%r181, %r182}; + .loc 1 527 19 // chunk_o.py:527:19 + fma.rn.f32x2 %rd161, %rd154, %rd160, %rd156; + .loc 1 529 19 // chunk_o.py:529:19 + mov.b64 %rd162, {%r933, %r934}; + mul.f32x2 %rd163, %rd161, %rd162; + .loc 1 531 35 // chunk_o.py:531:35 + mov.b64 {%r976, %r977}, %rd163; + cvt.rn.bf16.f32 %rs1, %r976; + cvt.rn.bf16.f32 %rs2, %r977; + mov.b64 {%r978, %r979}, %rd159; + cvt.rn.bf16.f32 %rs3, %r978; + cvt.rn.bf16.f32 %rs4, %r979; + .loc 1 527 38 // chunk_o.py:527:38 + mov.b64 %rd164, {%r193, %r194}; + .loc 1 527 19 // chunk_o.py:527:19 + fma.rn.f32x2 %rd165, %rd154, %rd164, %rd156; + .loc 1 529 19 // chunk_o.py:529:19 + mov.b64 %rd166, {%r939, %r940}; + mul.f32x2 %rd167, %rd165, %rd166; + .loc 1 527 38 // chunk_o.py:527:38 + mov.b64 %rd168, {%r191, %r192}; + .loc 1 527 19 // chunk_o.py:527:19 + fma.rn.f32x2 %rd169, %rd154, %rd168, %rd156; + .loc 1 529 19 // chunk_o.py:529:19 + mov.b64 %rd170, {%r937, %r938}; + mul.f32x2 %rd171, %rd169, %rd170; + .loc 1 531 35 // chunk_o.py:531:35 + mov.b64 {%r980, %r981}, %rd171; + cvt.rn.bf16.f32 %rs5, %r980; + cvt.rn.bf16.f32 %rs6, %r981; + mov.b64 {%r982, %r983}, %rd167; + cvt.rn.bf16.f32 %rs7, %r982; + cvt.rn.bf16.f32 %rs8, %r983; + .loc 1 527 38 // chunk_o.py:527:38 + mov.b64 %rd172, {%r219, %r220}; + .loc 1 527 19 // chunk_o.py:527:19 + fma.rn.f32x2 %rd173, %rd154, %rd172, %rd156; + .loc 1 529 19 // chunk_o.py:529:19 + mov.b64 %rd174, {%r951, %r952}; + mul.f32x2 %rd175, %rd173, %rd174; + .loc 1 527 38 // chunk_o.py:527:38 + mov.b64 %rd176, {%r217, %r218}; + .loc 1 527 19 // chunk_o.py:527:19 + fma.rn.f32x2 %rd177, %rd154, %rd176, %rd156; + .loc 1 529 19 // chunk_o.py:529:19 + mov.b64 %rd178, {%r949, %r950}; + mul.f32x2 %rd179, %rd177, %rd178; + .loc 1 527 38 // chunk_o.py:527:38 + mov.b64 %rd180, {%r199, %r200}; + .loc 1 527 19 // chunk_o.py:527:19 + fma.rn.f32x2 %rd181, %rd154, %rd180, %rd156; + .loc 1 529 19 // chunk_o.py:529:19 + mov.b64 %rd182, {%r943, %r944}; + mul.f32x2 %rd183, %rd181, %rd182; + .loc 1 531 35 // chunk_o.py:531:35 + mov.b64 {%r984, %r985}, %rd183; + cvt.rn.bf16.f32 %rs9, %r984; + cvt.rn.bf16.f32 %rs10, %r985; + .loc 1 527 38 // chunk_o.py:527:38 + mov.b64 %rd184, {%r197, %r198}; + .loc 1 527 19 // chunk_o.py:527:19 + fma.rn.f32x2 %rd185, %rd154, %rd184, %rd156; + .loc 1 529 19 // chunk_o.py:529:19 + mov.b64 %rd186, {%r941, %r942}; + mul.f32x2 %rd187, %rd185, %rd186; + .loc 1 531 35 // chunk_o.py:531:35 + mov.b64 {%r986, %r987}, %rd187; + cvt.rn.bf16.f32 %rs11, %r986; + cvt.rn.bf16.f32 %rs12, %r987; + .loc 1 527 38 // chunk_o.py:527:38 + mov.b64 %rd188, {%r205, %r206}; + .loc 1 527 19 // chunk_o.py:527:19 + fma.rn.f32x2 %rd189, %rd154, %rd188, %rd156; + .loc 1 529 19 // chunk_o.py:529:19 + mov.b64 %rd190, {%r947, %r948}; + mul.f32x2 %rd191, %rd189, %rd190; + .loc 1 531 35 // chunk_o.py:531:35 + mov.b64 {%r988, %r989}, %rd191; + cvt.rn.bf16.f32 %rs13, %r988; + cvt.rn.bf16.f32 %rs14, %r989; + .loc 1 527 38 // chunk_o.py:527:38 + mov.b64 %rd192, {%r203, %r204}; + .loc 1 527 19 // chunk_o.py:527:19 + fma.rn.f32x2 %rd193, %rd154, %rd192, %rd156; + .loc 1 529 19 // chunk_o.py:529:19 + mov.b64 %rd194, {%r945, %r946}; + mul.f32x2 %rd195, %rd193, %rd194; + .loc 1 531 35 // chunk_o.py:531:35 + mov.b64 {%r990, %r991}, %rd195; + cvt.rn.bf16.f32 %rs15, %r990; + cvt.rn.bf16.f32 %rs16, %r991; + mov.b64 {%r992, %r993}, %rd179; + cvt.rn.bf16.f32 %rs17, %r992; + cvt.rn.bf16.f32 %rs18, %r993; + mov.b64 {%r994, %r995}, %rd175; + cvt.rn.bf16.f32 %rs19, %r994; + cvt.rn.bf16.f32 %rs20, %r995; + .loc 1 527 38 // chunk_o.py:527:38 + mov.b64 %rd196, {%r223, %r224}; + .loc 1 527 19 // chunk_o.py:527:19 + fma.rn.f32x2 %rd197, %rd154, %rd196, %rd156; + .loc 1 529 19 // chunk_o.py:529:19 + mov.b64 %rd198, {%r955, %r956}; + mul.f32x2 %rd199, %rd197, %rd198; + .loc 1 527 38 // chunk_o.py:527:38 + mov.b64 %rd200, {%r221, %r222}; + .loc 1 527 19 // chunk_o.py:527:19 + fma.rn.f32x2 %rd201, %rd154, %rd200, %rd156; + .loc 1 529 19 // chunk_o.py:529:19 + mov.b64 %rd202, {%r953, %r954}; + mul.f32x2 %rd203, %rd201, %rd202; + .loc 1 531 35 // chunk_o.py:531:35 + mov.b64 {%r996, %r997}, %rd203; + cvt.rn.bf16.f32 %rs21, %r996; + cvt.rn.bf16.f32 %rs22, %r997; + mov.b64 {%r998, %r999}, %rd199; + cvt.rn.bf16.f32 %rs23, %r998; + cvt.rn.bf16.f32 %rs24, %r999; + .loc 1 527 38 // chunk_o.py:527:38 + mov.b64 %rd204, {%r227, %r228}; + .loc 1 527 19 // chunk_o.py:527:19 + fma.rn.f32x2 %rd205, %rd154, %rd204, %rd156; + .loc 1 529 19 // chunk_o.py:529:19 + mov.b64 %rd206, {%r959, %r960}; + mul.f32x2 %rd207, %rd205, %rd206; + .loc 1 527 38 // chunk_o.py:527:38 + mov.b64 %rd208, {%r225, %r226}; + .loc 1 527 19 // chunk_o.py:527:19 + fma.rn.f32x2 %rd209, %rd154, %rd208, %rd156; + .loc 1 529 19 // chunk_o.py:529:19 + mov.b64 %rd210, {%r957, %r958}; + mul.f32x2 %rd211, %rd209, %rd210; + .loc 1 531 35 // chunk_o.py:531:35 + mov.b64 {%r1000, %r1001}, %rd211; + cvt.rn.bf16.f32 %rs25, %r1000; + cvt.rn.bf16.f32 %rs26, %r1001; + mov.b64 {%r1002, %r1003}, %rd207; + cvt.rn.bf16.f32 %rs27, %r1002; + cvt.rn.bf16.f32 %rs28, %r1003; + .loc 1 527 38 // chunk_o.py:527:38 + mov.b64 %rd212, {%r243, %r244}; + .loc 1 527 19 // chunk_o.py:527:19 + fma.rn.f32x2 %rd213, %rd154, %rd212, %rd156; + .loc 1 529 19 // chunk_o.py:529:19 + mov.b64 %rd214, {%r963, %r964}; + mul.f32x2 %rd215, %rd213, %rd214; + .loc 1 527 38 // chunk_o.py:527:38 + mov.b64 %rd216, {%r241, %r242}; + .loc 1 527 19 // chunk_o.py:527:19 + fma.rn.f32x2 %rd217, %rd154, %rd216, %rd156; + .loc 1 529 19 // chunk_o.py:529:19 + mov.b64 %rd218, {%r961, %r962}; + mul.f32x2 %rd219, %rd217, %rd218; + .loc 1 531 35 // chunk_o.py:531:35 + mov.b64 {%r1004, %r1005}, %rd219; + cvt.rn.bf16.f32 %rs29, %r1004; + cvt.rn.bf16.f32 %rs30, %r1005; + mov.b64 {%r1006, %r1007}, %rd215; + cvt.rn.bf16.f32 %rs31, %r1006; + cvt.rn.bf16.f32 %rs32, %r1007; + .loc 1 527 38 // chunk_o.py:527:38 + mov.b64 %rd220, {%r247, %r248}; + .loc 1 527 19 // chunk_o.py:527:19 + fma.rn.f32x2 %rd221, %rd154, %rd220, %rd156; + .loc 1 529 19 // chunk_o.py:529:19 + mov.b64 %rd222, {%r967, %r968}; + mul.f32x2 %rd223, %rd221, %rd222; + .loc 1 527 38 // chunk_o.py:527:38 + mov.b64 %rd224, {%r245, %r246}; + .loc 1 527 19 // chunk_o.py:527:19 + fma.rn.f32x2 %rd225, %rd154, %rd224, %rd156; + .loc 1 529 19 // chunk_o.py:529:19 + mov.b64 %rd226, {%r965, %r966}; + mul.f32x2 %rd227, %rd225, %rd226; + .loc 1 531 35 // chunk_o.py:531:35 + mov.b64 {%r1008, %r1009}, %rd227; + cvt.rn.bf16.f32 %rs33, %r1008; + cvt.rn.bf16.f32 %rs34, %r1009; + mov.b64 {%r1010, %r1011}, %rd223; + cvt.rn.bf16.f32 %rs35, %r1010; + cvt.rn.bf16.f32 %rs36, %r1011; + .loc 1 527 38 // chunk_o.py:527:38 + mov.b64 %rd228, {%r267, %r268}; + .loc 1 527 19 // chunk_o.py:527:19 + fma.rn.f32x2 %rd229, %rd154, %rd228, %rd156; + .loc 1 529 19 // chunk_o.py:529:19 + mov.b64 %rd230, {%r971, %r972}; + mul.f32x2 %rd231, %rd229, %rd230; + .loc 1 527 38 // chunk_o.py:527:38 + mov.b64 %rd232, {%r265, %r266}; + .loc 1 527 19 // chunk_o.py:527:19 + fma.rn.f32x2 %rd233, %rd154, %rd232, %rd156; + .loc 1 529 19 // chunk_o.py:529:19 + mov.b64 %rd234, {%r969, %r970}; + mul.f32x2 %rd235, %rd233, %rd234; + .loc 1 531 35 // chunk_o.py:531:35 + mov.b64 {%r1012, %r1013}, %rd235; + cvt.rn.bf16.f32 %rs37, %r1012; + cvt.rn.bf16.f32 %rs38, %r1013; + .loc 1 504 34 // chunk_o.py:504:34 + bfe.u32 %r1014, %r807, 2, 3; + or.b32 %r1015, %r1014, 32; + or.b32 %r1016, %r1014, 40; + or.b32 %r1017, %r1014, 48; + or.b32 %r1018, %r1014, 56; + or.b32 %r1019, %r1014, 24; + or.b32 %r1020, %r1014, 16; + or.b32 %r1021, %r1014, 8; + or.b32 %r1022, %r973, %r974; + or.b32 %r1023, %r1022, 1; + or.b32 %r1024, %r1022, 17; + or.b32 %r1025, %r1022, 16; + or.b32 %r1026, %r1022, 33; + or.b32 %r1027, %r1022, 32; + or.b32 %r1028, %r1022, 49; + or.b32 %r1029, %r1022, 48; + .loc 1 504 21 // chunk_o.py:504:21 + cvt.u64.u32 %rd236, %r1029; + cvt.u64.u32 %rd237, %r1028; + cvt.u64.u32 %rd238, %r1027; + cvt.u64.u32 %rd239, %r1026; + cvt.u64.u32 %rd240, %r1025; + cvt.u64.u32 %rd241, %r1024; + cvt.u64.u32 %rd242, %r1022; + cvt.u64.u32 %rd243, %r1023; + cvt.u64.u32 %rd244, %r1018; + cvt.u64.u32 %rd245, %r1017; + cvt.u64.u32 %rd246, %r1016; + cvt.u64.u32 %rd247, %r1015; + cvt.u64.u32 %rd248, %r1019; + cvt.u64.u32 %rd249, %r1020; + cvt.u64.u32 %rd250, %r1021; + cvt.u64.u32 %rd251, %r1014; + or.b64 %rd252, %rd80, %rd243; + or.b64 %rd253, %rd80, %rd251; + or.b64 %rd254, %rd80, %rd242; + or.b64 %rd255, %rd80, %rd250; + or.b64 %rd256, %rd80, %rd241; + or.b64 %rd257, %rd80, %rd240; + or.b64 %rd258, %rd80, %rd249; + or.b64 %rd259, %rd80, %rd248; + or.b64 %rd260, %rd80, %rd239; + or.b64 %rd261, %rd80, %rd238; + or.b64 %rd262, %rd80, %rd247; + or.b64 %rd263, %rd80, %rd246; + or.b64 %rd264, %rd80, %rd237; + or.b64 %rd265, %rd80, %rd236; + or.b64 %rd266, %rd80, %rd245; + or.b64 %rd267, %rd80, %rd244; + .loc 1 505 16 // chunk_o.py:505:16 + setp.lt.s64 %p18, %rd267, %rd74; + setp.lt.s64 %p19, %rd266, %rd74; + setp.lt.s64 %p20, %rd265, %rd74; + setp.lt.s64 %p21, %rd264, %rd74; + setp.lt.s64 %p22, %rd263, %rd74; + setp.lt.s64 %p23, %rd262, %rd74; + setp.lt.s64 %p24, %rd261, %rd74; + setp.lt.s64 %p25, %rd260, %rd74; + setp.lt.s64 %p26, %rd259, %rd74; + setp.lt.s64 %p27, %rd258, %rd74; + setp.lt.s64 %p28, %rd257, %rd74; + setp.lt.s64 %p29, %rd256, %rd74; + setp.lt.s64 %p30, %rd255, %rd74; + setp.lt.s64 %p31, %rd254, %rd74; + setp.lt.s64 %p32, %rd253, %rd74; + setp.lt.s64 %p33, %rd252, %rd74; + .loc 1 530 27 // chunk_o.py:530:27 + setp.le.u32 %p34, %r1014, %r1023; + setp.le.u32 %p35, %r1021, %r1023; + setp.le.u32 %p36, %r1021, %r1022; + setp.le.u32 %p37, %r1020, %r1024; + setp.le.u32 %p38, %r1015, %r1026; + setp.le.u32 %p39, %r1017, %r1028; + setp.le.u32 %p40, %r1014, %r1022; + .loc 1 531 29 // chunk_o.py:531:29 + selp.b16 %rs39, %rs2, 0x0000, %p33; + selp.b16 %rs40, %rs39, 0x0000, %p34; + selp.b16 %rs41, %rs40, 0x0000, %p32; + selp.b16 %rs42, %rs1, 0x0000, %p31; + selp.b16 %rs43, %rs42, 0x0000, %p40; + selp.b16 %rs44, %rs43, 0x0000, %p32; + mov.b32 %r1030, {%rs44, %rs41}; + selp.b16 %rs45, %rs4, 0x0000, %p33; + selp.b16 %rs46, %rs45, 0x0000, %p35; + selp.b16 %rs47, %rs46, 0x0000, %p30; + selp.b16 %rs48, %rs3, 0x0000, %p31; + selp.b16 %rs49, %rs48, 0x0000, %p36; + selp.b16 %rs50, %rs49, 0x0000, %p30; + mov.b32 %r1031, {%rs50, %rs47}; + .loc 1 530 27 // chunk_o.py:530:27 + setp.le.u32 %p41, %r1019, %r1024; + setp.le.u32 %p42, %r1019, %r1025; + .loc 1 531 29 // chunk_o.py:531:29 + selp.b16 %rs51, %rs6, 0x0000, %p29; + selp.b16 %rs52, %rs51, 0x0000, %p32; + selp.b16 %rs53, %rs5, 0x0000, %p28; + selp.b16 %rs54, %rs53, 0x0000, %p32; + mov.b32 %r1032, {%rs54, %rs52}; + selp.b16 %rs55, %rs8, 0x0000, %p29; + selp.b16 %rs56, %rs55, 0x0000, %p30; + selp.b16 %rs57, %rs7, 0x0000, %p28; + selp.b16 %rs58, %rs57, 0x0000, %p30; + mov.b32 %r1033, {%rs58, %rs56}; + selp.b16 %rs59, %rs18, 0x0000, %p29; + selp.b16 %rs60, %rs59, 0x0000, %p37; + selp.b16 %rs61, %rs60, 0x0000, %p27; + selp.b16 %rs62, %rs17, 0x0000, %p28; + selp.b16 %rs63, %rs62, 0x0000, %p40; + selp.b16 %rs64, %rs63, 0x0000, %p27; + mov.b32 %r1034, {%rs64, %rs61}; + selp.b16 %rs65, %rs20, 0x0000, %p29; + selp.b16 %rs66, %rs65, 0x0000, %p41; + selp.b16 %rs67, %rs66, 0x0000, %p26; + selp.b16 %rs68, %rs19, 0x0000, %p28; + selp.b16 %rs69, %rs68, 0x0000, %p42; + selp.b16 %rs70, %rs69, 0x0000, %p26; + mov.b32 %r1035, {%rs70, %rs67}; + .loc 1 530 27 // chunk_o.py:530:27 + setp.le.u32 %p43, %r1018, %r1028; + setp.le.u32 %p44, %r1016, %r1026; + setp.le.u32 %p45, %r1016, %r1027; + .loc 1 531 29 // chunk_o.py:531:29 + selp.b16 %rs71, %rs12, 0x0000, %p25; + selp.b16 %rs72, %rs71, 0x0000, %p32; + selp.b16 %rs73, %rs11, 0x0000, %p24; + selp.b16 %rs74, %rs73, 0x0000, %p32; + mov.b32 %r1036, {%rs74, %rs72}; + selp.b16 %rs75, %rs10, 0x0000, %p25; + selp.b16 %rs76, %rs75, 0x0000, %p30; + selp.b16 %rs77, %rs9, 0x0000, %p24; + selp.b16 %rs78, %rs77, 0x0000, %p30; + mov.b32 %r1037, {%rs78, %rs76}; + selp.b16 %rs79, %rs22, 0x0000, %p25; + selp.b16 %rs80, %rs79, 0x0000, %p27; + selp.b16 %rs81, %rs21, 0x0000, %p24; + selp.b16 %rs82, %rs81, 0x0000, %p27; + mov.b32 %r1038, {%rs82, %rs80}; + selp.b16 %rs83, %rs24, 0x0000, %p25; + selp.b16 %rs84, %rs83, 0x0000, %p26; + selp.b16 %rs85, %rs23, 0x0000, %p24; + selp.b16 %rs86, %rs85, 0x0000, %p26; + mov.b32 %r1039, {%rs86, %rs84}; + selp.b16 %rs87, %rs30, 0x0000, %p25; + selp.b16 %rs88, %rs87, 0x0000, %p38; + selp.b16 %rs89, %rs88, 0x0000, %p23; + selp.b16 %rs90, %rs29, 0x0000, %p24; + selp.b16 %rs91, %rs90, 0x0000, %p40; + selp.b16 %rs92, %rs91, 0x0000, %p23; + mov.b32 %r1040, {%rs92, %rs89}; + selp.b16 %rs93, %rs32, 0x0000, %p25; + selp.b16 %rs94, %rs93, 0x0000, %p44; + selp.b16 %rs95, %rs94, 0x0000, %p22; + selp.b16 %rs96, %rs31, 0x0000, %p24; + selp.b16 %rs97, %rs96, 0x0000, %p45; + selp.b16 %rs98, %rs97, 0x0000, %p22; + mov.b32 %r1041, {%rs98, %rs95}; + .loc 1 530 27 // chunk_o.py:530:27 + setp.le.u32 %p46, %r1018, %r1029; + .loc 1 531 29 // chunk_o.py:531:29 + selp.b16 %rs99, %rs16, 0x0000, %p21; + selp.b16 %rs100, %rs99, 0x0000, %p32; + selp.b16 %rs101, %rs15, 0x0000, %p20; + selp.b16 %rs102, %rs101, 0x0000, %p32; + mov.b32 %r1042, {%rs102, %rs100}; + selp.b16 %rs103, %rs14, 0x0000, %p21; + selp.b16 %rs104, %rs103, 0x0000, %p30; + selp.b16 %rs105, %rs13, 0x0000, %p20; + selp.b16 %rs106, %rs105, 0x0000, %p30; + mov.b32 %r1043, {%rs106, %rs104}; + selp.b16 %rs107, %rs26, 0x0000, %p21; + selp.b16 %rs108, %rs107, 0x0000, %p27; + selp.b16 %rs109, %rs25, 0x0000, %p20; + selp.b16 %rs110, %rs109, 0x0000, %p27; + mov.b32 %r1044, {%rs110, %rs108}; + selp.b16 %rs111, %rs28, 0x0000, %p21; + selp.b16 %rs112, %rs111, 0x0000, %p26; + selp.b16 %rs113, %rs27, 0x0000, %p20; + selp.b16 %rs114, %rs113, 0x0000, %p26; + mov.b32 %r1045, {%rs114, %rs112}; + selp.b16 %rs115, %rs34, 0x0000, %p21; + selp.b16 %rs116, %rs115, 0x0000, %p23; + selp.b16 %rs117, %rs33, 0x0000, %p20; + selp.b16 %rs118, %rs117, 0x0000, %p23; + mov.b32 %r1046, {%rs118, %rs116}; + selp.b16 %rs119, %rs36, 0x0000, %p21; + selp.b16 %rs120, %rs119, 0x0000, %p22; + selp.b16 %rs121, %rs35, 0x0000, %p20; + selp.b16 %rs122, %rs121, 0x0000, %p22; + mov.b32 %r1047, {%rs122, %rs120}; + selp.b16 %rs123, %rs38, 0x0000, %p21; + selp.b16 %rs124, %rs123, 0x0000, %p39; + selp.b16 %rs125, %rs124, 0x0000, %p19; + selp.b16 %rs126, %rs37, 0x0000, %p20; + selp.b16 %rs127, %rs126, 0x0000, %p40; + selp.b16 %rs128, %rs127, 0x0000, %p19; + mov.b32 %r1048, {%rs128, %rs125}; + .loc 1 531 35 // chunk_o.py:531:35 + mov.b64 {%r1049, %r1050}, %rd231; + cvt.rn.bf16.f32 %rs129, %r1049; + cvt.rn.bf16.f32 %rs130, %r1050; + .loc 1 531 29 // chunk_o.py:531:29 + selp.b16 %rs131, %rs130, 0x0000, %p21; + selp.b16 %rs132, %rs131, 0x0000, %p43; + selp.b16 %rs133, %rs132, 0x0000, %p18; + selp.b16 %rs134, %rs129, 0x0000, %p20; + selp.b16 %rs135, %rs134, 0x0000, %p46; + selp.b16 %rs136, %rs135, 0x0000, %p18; + mov.b32 %r1051, {%rs136, %rs133}; + .loc 1 531 35 // chunk_o.py:531:35 + bar.sync 0; + shl.b32 %r1052, %r867, 7; + shl.b32 %r1053, %r851, 1; + shr.u32 %r1054, %r812, 1; + or.b32 %r1055, %r1052, %r830; + or.b32 %r1056, %r1053, %r1054; + xor.b32 %r1057, %r1055, %r1056; + add.s32 %r1058, %r834, %r1057; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1058+16384], {%r1030, %r1031, %r1032, %r1033}; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1058+18432], {%r672, %r672, %r1034, %r1035}; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1058+20480], {%r672, %r672, %r672, %r672}; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1058+22528], {%r672, %r672, %r672, %r672}; + xor.b32 %r1059, %r1057, 64; + add.s32 %r1060, %r834, %r1059; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1060+16384], {%r1036, %r1037, %r1042, %r1043}; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1060+18432], {%r1038, %r1039, %r1044, %r1045}; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1060+20480], {%r1040, %r1041, %r1046, %r1047}; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1060+22528], {%r672, %r672, %r1048, %r1051}; + .loc 1 536 20 // chunk_o.py:536:20 + add.s64 %rd268, %rd153, %rd118; + add.s64 %rd269, %rd153, %rd120; + add.s64 %rd270, %rd153, %rd122; + add.s64 %rd271, %rd153, %rd124; + add.s64 %rd272, %rd153, %rd126; + add.s64 %rd273, %rd153, %rd128; + add.s64 %rd274, %rd153, %rd130; + add.s64 %rd275, %rd153, %rd132; + add.s64 %rd276, %rd153, %rd134; + add.s64 %rd277, %rd153, %rd136; + add.s64 %rd278, %rd153, %rd138; + add.s64 %rd279, %rd153, %rd140; + add.s64 %rd280, %rd153, %rd142; + add.s64 %rd281, %rd153, %rd144; + add.s64 %rd282, %rd153, %rd146; + add.s64 %rd283, %rd153, %rd148; + .loc 1 536 44 // chunk_o.py:536:44 + add.s64 %rd34, %rd268, %rd150; + add.s64 %rd35, %rd269, %rd150; + add.s64 %rd36, %rd270, %rd150; + add.s64 %rd37, %rd271, %rd150; + add.s64 %rd38, %rd272, %rd150; + add.s64 %rd39, %rd273, %rd150; + add.s64 %rd40, %rd274, %rd150; + add.s64 %rd41, %rd275, %rd150; + add.s64 %rd42, %rd276, %rd150; + add.s64 %rd43, %rd277, %rd150; + add.s64 %rd44, %rd278, %rd150; + add.s64 %rd45, %rd279, %rd150; + add.s64 %rd46, %rd280, %rd150; + add.s64 %rd47, %rd281, %rd150; + add.s64 %rd48, %rd282, %rd150; + add.s64 %rd49, %rd283, %rd150; + .loc 1 537 20 // chunk_o.py:537:20 + add.s64 %rd284, %rd152, %rd118; + add.s64 %rd285, %rd152, %rd120; + add.s64 %rd286, %rd152, %rd122; + add.s64 %rd287, %rd152, %rd124; + add.s64 %rd288, %rd152, %rd126; + add.s64 %rd289, %rd152, %rd128; + add.s64 %rd290, %rd152, %rd130; + add.s64 %rd291, %rd152, %rd132; + add.s64 %rd292, %rd152, %rd134; + add.s64 %rd293, %rd152, %rd136; + add.s64 %rd294, %rd152, %rd138; + add.s64 %rd295, %rd152, %rd140; + add.s64 %rd296, %rd152, %rd142; + add.s64 %rd297, %rd152, %rd144; + add.s64 %rd298, %rd152, %rd146; + add.s64 %rd299, %rd152, %rd148; + .loc 1 537 44 // chunk_o.py:537:44 + add.s64 %rd50, %rd284, %rd150; + add.s64 %rd51, %rd285, %rd150; + add.s64 %rd52, %rd286, %rd150; + add.s64 %rd53, %rd287, %rd150; + add.s64 %rd54, %rd288, %rd150; + add.s64 %rd55, %rd289, %rd150; + add.s64 %rd56, %rd290, %rd150; + add.s64 %rd57, %rd291, %rd150; + add.s64 %rd58, %rd292, %rd150; + add.s64 %rd59, %rd293, %rd150; + add.s64 %rd60, %rd294, %rd150; + add.s64 %rd61, %rd295, %rd150; + add.s64 %rd62, %rd296, %rd150; + add.s64 %rd63, %rd297, %rd150; + add.s64 %rd64, %rd298, %rd150; + add.s64 %rd65, %rd299, %rd150; + .loc 1 538 23 // chunk_o.py:538:23 + // begin inline asm + mov.u32 %r413, %r672; + mov.u32 %r414, %r672; + mov.u32 %r415, %r672; + mov.u32 %r416, %r672; + @%p2 ld.global.v4.b32 { %r413, %r414, %r415, %r416 }, [ %rd34 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r445, %r672; + mov.u32 %r446, %r672; + mov.u32 %r447, %r672; + mov.u32 %r448, %r672; + @%p3 ld.global.v4.b32 { %r445, %r446, %r447, %r448 }, [ %rd35 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r417, %r672; + mov.u32 %r418, %r672; + mov.u32 %r419, %r672; + mov.u32 %r420, %r672; + @%p4 ld.global.v4.b32 { %r417, %r418, %r419, %r420 }, [ %rd36 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r450, %r672; + mov.u32 %r451, %r672; + mov.u32 %r452, %r672; + mov.u32 %r453, %r672; + @%p5 ld.global.v4.b32 { %r450, %r451, %r452, %r453 }, [ %rd37 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r421, %r672; + mov.u32 %r422, %r672; + mov.u32 %r423, %r672; + mov.u32 %r424, %r672; + @%p6 ld.global.v4.b32 { %r421, %r422, %r423, %r424 }, [ %rd38 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r455, %r672; + mov.u32 %r456, %r672; + mov.u32 %r457, %r672; + mov.u32 %r458, %r672; + @%p7 ld.global.v4.b32 { %r455, %r456, %r457, %r458 }, [ %rd39 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r425, %r672; + mov.u32 %r426, %r672; + mov.u32 %r427, %r672; + mov.u32 %r428, %r672; + @%p8 ld.global.v4.b32 { %r425, %r426, %r427, %r428 }, [ %rd40 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r460, %r672; + mov.u32 %r461, %r672; + mov.u32 %r462, %r672; + mov.u32 %r463, %r672; + @%p9 ld.global.v4.b32 { %r460, %r461, %r462, %r463 }, [ %rd41 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r429, %r672; + mov.u32 %r430, %r672; + mov.u32 %r431, %r672; + mov.u32 %r432, %r672; + @%p17 ld.global.v4.b32 { %r429, %r430, %r431, %r432 }, [ %rd42 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r465, %r672; + mov.u32 %r466, %r672; + mov.u32 %r467, %r672; + mov.u32 %r468, %r672; + @%p10 ld.global.v4.b32 { %r465, %r466, %r467, %r468 }, [ %rd43 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r433, %r672; + mov.u32 %r434, %r672; + mov.u32 %r435, %r672; + mov.u32 %r436, %r672; + @%p11 ld.global.v4.b32 { %r433, %r434, %r435, %r436 }, [ %rd44 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r470, %r672; + mov.u32 %r471, %r672; + mov.u32 %r472, %r672; + mov.u32 %r473, %r672; + @%p12 ld.global.v4.b32 { %r470, %r471, %r472, %r473 }, [ %rd45 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r437, %r672; + mov.u32 %r438, %r672; + mov.u32 %r439, %r672; + mov.u32 %r440, %r672; + @%p13 ld.global.v4.b32 { %r437, %r438, %r439, %r440 }, [ %rd46 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r475, %r672; + mov.u32 %r476, %r672; + mov.u32 %r477, %r672; + mov.u32 %r478, %r672; + @%p14 ld.global.v4.b32 { %r475, %r476, %r477, %r478 }, [ %rd47 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r441, %r672; + mov.u32 %r442, %r672; + mov.u32 %r443, %r672; + mov.u32 %r444, %r672; + @%p15 ld.global.v4.b32 { %r441, %r442, %r443, %r444 }, [ %rd48 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r480, %r672; + mov.u32 %r481, %r672; + mov.u32 %r482, %r672; + mov.u32 %r483, %r672; + @%p16 ld.global.v4.b32 { %r480, %r481, %r482, %r483 }, [ %rd49 + 0 ]; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r4 + 0 ], { %r413, %r414, %r415, %r416 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r9 + 0 ], { %r417, %r418, %r419, %r420 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r14 + 0 ], { %r421, %r422, %r423, %r424 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r19 + 0 ], { %r425, %r426, %r427, %r428 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r24 + 0 ], { %r429, %r430, %r431, %r432 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r29 + 0 ], { %r433, %r434, %r435, %r436 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r34 + 0 ], { %r437, %r438, %r439, %r440 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r39 + 0 ], { %r441, %r442, %r443, %r444 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r44 + 0 ], { %r445, %r446, %r447, %r448 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r449 + 0 ], { %r450, %r451, %r452, %r453 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r454 + 0 ], { %r455, %r456, %r457, %r458 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r459 + 0 ], { %r460, %r461, %r462, %r463 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r464 + 0 ], { %r465, %r466, %r467, %r468 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r469 + 0 ], { %r470, %r471, %r472, %r473 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r474 + 0 ], { %r475, %r476, %r477, %r478 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r479 + 0 ], { %r480, %r481, %r482, %r483 }; + // end inline asm + .loc 1 531 35 // chunk_o.py:531:35 + bar.sync 0; + xor.b32 %r1061, %r1055, %r851; + add.s32 %r1062, %r834, %r1061; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r484, %r485, %r486, %r487}, [%r1062+16384]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r504, %r505, %r506, %r507}, [%r1062+18432]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r508, %r509, %r510, %r511}, [%r1062+20480]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r512, %r513, %r514, %r515}, [%r1062+22528]; + xor.b32 %r1063, %r1061, 32; + add.s32 %r1064, %r834, %r1063; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r520, %r521, %r522, %r523}, [%r1064+16384]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r572, %r573, %r574, %r575}, [%r1064+18432]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r608, %r609, %r610, %r611}, [%r1064+20480]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r644, %r645, %r646, %r647}, [%r1064+22528]; + xor.b32 %r1065, %r1061, 64; + add.s32 %r1066, %r834, %r1065; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r676, %r677, %r678, %r679}, [%r1066+16384]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r696, %r697, %r698, %r699}, [%r1066+18432]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r700, %r701, %r702, %r703}, [%r1066+20480]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r704, %r705, %r706, %r707}, [%r1066+22528]; + xor.b32 %r1067, %r1061, 96; + add.s32 %r1068, %r834, %r1067; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r708, %r709, %r710, %r711}, [%r1068+16384]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r728, %r729, %r730, %r731}, [%r1068+18432]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r732, %r733, %r734, %r735}, [%r1068+20480]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r736, %r737, %r738, %r739}, [%r1068+22528]; + .loc 1 538 23 // chunk_o.py:538:23 + shl.b32 %r1069, %r807, 8; + and.b32 %r1070, %r1069, 7936; + or.b32 %r1071, %r1070, %r830; + xor.b32 %r1072, %r1071, %r1054; + add.s32 %r1073, %r834, %r1072; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r488, %r489, %r524, %r525}, [%r1073]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r680, %r681, %r712, %r713}, [%r1073+8192]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r496, %r497, %r548, %r549}, [%r1073+128]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r688, %r689, %r720, %r721}, [%r1073+8320]; + xor.b32 %r1074, %r1072, 32; + add.s32 %r1075, %r834, %r1074; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r490, %r491, %r530, %r531}, [%r1075]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r682, %r683, %r714, %r715}, [%r1075+8192]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r498, %r499, %r554, %r555}, [%r1075+128]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r690, %r691, %r722, %r723}, [%r1075+8320]; + xor.b32 %r1076, %r1072, 64; + add.s32 %r1077, %r834, %r1076; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r492, %r493, %r536, %r537}, [%r1077]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r684, %r685, %r716, %r717}, [%r1077+8192]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r500, %r501, %r560, %r561}, [%r1077+128]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r692, %r693, %r724, %r725}, [%r1077+8320]; + xor.b32 %r1078, %r1072, 96; + add.s32 %r1079, %r834, %r1078; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r494, %r495, %r542, %r543}, [%r1079]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r686, %r687, %r718, %r719}, [%r1079+8192]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r502, %r503, %r566, %r567}, [%r1079+128]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r694, %r695, %r726, %r727}, [%r1079+8320]; + .loc 1 539 42 // chunk_o.py:539:42 + mov.b32 %r516, %r672; + mov.b32 %r517, %r672; + mov.b32 %r518, %r672; + mov.b32 %r519, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r516, %r517, %r518, %r519 }, { %r484, %r485, %r486, %r487 }, { %r488, %r489 }, { %r516, %r517, %r518, %r519 }; + // end inline asm + mov.b32 %r526, %r672; + mov.b32 %r527, %r672; + mov.b32 %r528, %r672; + mov.b32 %r529, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r526, %r527, %r528, %r529 }, { %r484, %r485, %r486, %r487 }, { %r490, %r491 }, { %r526, %r527, %r528, %r529 }; + // end inline asm + mov.b32 %r532, %r672; + mov.b32 %r533, %r672; + mov.b32 %r534, %r672; + mov.b32 %r535, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r532, %r533, %r534, %r535 }, { %r484, %r485, %r486, %r487 }, { %r492, %r493 }, { %r532, %r533, %r534, %r535 }; + // end inline asm + mov.b32 %r538, %r672; + mov.b32 %r539, %r672; + mov.b32 %r540, %r672; + mov.b32 %r541, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r538, %r539, %r540, %r541 }, { %r484, %r485, %r486, %r487 }, { %r494, %r495 }, { %r538, %r539, %r540, %r541 }; + // end inline asm + mov.b32 %r544, %r672; + mov.b32 %r545, %r672; + mov.b32 %r546, %r672; + mov.b32 %r547, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r544, %r545, %r546, %r547 }, { %r484, %r485, %r486, %r487 }, { %r496, %r497 }, { %r544, %r545, %r546, %r547 }; + // end inline asm + mov.b32 %r550, %r672; + mov.b32 %r551, %r672; + mov.b32 %r552, %r672; + mov.b32 %r553, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r550, %r551, %r552, %r553 }, { %r484, %r485, %r486, %r487 }, { %r498, %r499 }, { %r550, %r551, %r552, %r553 }; + // end inline asm + mov.b32 %r556, %r672; + mov.b32 %r557, %r672; + mov.b32 %r558, %r672; + mov.b32 %r559, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r556, %r557, %r558, %r559 }, { %r484, %r485, %r486, %r487 }, { %r500, %r501 }, { %r556, %r557, %r558, %r559 }; + // end inline asm + mov.b32 %r562, %r672; + mov.b32 %r563, %r672; + mov.b32 %r564, %r672; + mov.b32 %r565, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r562, %r563, %r564, %r565 }, { %r484, %r485, %r486, %r487 }, { %r502, %r503 }, { %r562, %r563, %r564, %r565 }; + // end inline asm + mov.b32 %r568, %r672; + mov.b32 %r569, %r672; + mov.b32 %r570, %r672; + mov.b32 %r571, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r568, %r569, %r570, %r571 }, { %r504, %r505, %r506, %r507 }, { %r488, %r489 }, { %r568, %r569, %r570, %r571 }; + // end inline asm + mov.b32 %r576, %r672; + mov.b32 %r577, %r672; + mov.b32 %r578, %r672; + mov.b32 %r579, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r576, %r577, %r578, %r579 }, { %r504, %r505, %r506, %r507 }, { %r490, %r491 }, { %r576, %r577, %r578, %r579 }; + // end inline asm + mov.b32 %r580, %r672; + mov.b32 %r581, %r672; + mov.b32 %r582, %r672; + mov.b32 %r583, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r580, %r581, %r582, %r583 }, { %r504, %r505, %r506, %r507 }, { %r492, %r493 }, { %r580, %r581, %r582, %r583 }; + // end inline asm + mov.b32 %r584, %r672; + mov.b32 %r585, %r672; + mov.b32 %r586, %r672; + mov.b32 %r587, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r584, %r585, %r586, %r587 }, { %r504, %r505, %r506, %r507 }, { %r494, %r495 }, { %r584, %r585, %r586, %r587 }; + // end inline asm + mov.b32 %r588, %r672; + mov.b32 %r589, %r672; + mov.b32 %r590, %r672; + mov.b32 %r591, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r588, %r589, %r590, %r591 }, { %r504, %r505, %r506, %r507 }, { %r496, %r497 }, { %r588, %r589, %r590, %r591 }; + // end inline asm + mov.b32 %r592, %r672; + mov.b32 %r593, %r672; + mov.b32 %r594, %r672; + mov.b32 %r595, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r592, %r593, %r594, %r595 }, { %r504, %r505, %r506, %r507 }, { %r498, %r499 }, { %r592, %r593, %r594, %r595 }; + // end inline asm + mov.b32 %r596, %r672; + mov.b32 %r597, %r672; + mov.b32 %r598, %r672; + mov.b32 %r599, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r596, %r597, %r598, %r599 }, { %r504, %r505, %r506, %r507 }, { %r500, %r501 }, { %r596, %r597, %r598, %r599 }; + // end inline asm + mov.b32 %r600, %r672; + mov.b32 %r601, %r672; + mov.b32 %r602, %r672; + mov.b32 %r603, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r600, %r601, %r602, %r603 }, { %r504, %r505, %r506, %r507 }, { %r502, %r503 }, { %r600, %r601, %r602, %r603 }; + // end inline asm + mov.b32 %r604, %r672; + mov.b32 %r605, %r672; + mov.b32 %r606, %r672; + mov.b32 %r607, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r604, %r605, %r606, %r607 }, { %r508, %r509, %r510, %r511 }, { %r488, %r489 }, { %r604, %r605, %r606, %r607 }; + // end inline asm + mov.b32 %r612, %r672; + mov.b32 %r613, %r672; + mov.b32 %r614, %r672; + mov.b32 %r615, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r612, %r613, %r614, %r615 }, { %r508, %r509, %r510, %r511 }, { %r490, %r491 }, { %r612, %r613, %r614, %r615 }; + // end inline asm + mov.b32 %r616, %r672; + mov.b32 %r617, %r672; + mov.b32 %r618, %r672; + mov.b32 %r619, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r616, %r617, %r618, %r619 }, { %r508, %r509, %r510, %r511 }, { %r492, %r493 }, { %r616, %r617, %r618, %r619 }; + // end inline asm + mov.b32 %r620, %r672; + mov.b32 %r621, %r672; + mov.b32 %r622, %r672; + mov.b32 %r623, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r620, %r621, %r622, %r623 }, { %r508, %r509, %r510, %r511 }, { %r494, %r495 }, { %r620, %r621, %r622, %r623 }; + // end inline asm + mov.b32 %r624, %r672; + mov.b32 %r625, %r672; + mov.b32 %r626, %r672; + mov.b32 %r627, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r624, %r625, %r626, %r627 }, { %r508, %r509, %r510, %r511 }, { %r496, %r497 }, { %r624, %r625, %r626, %r627 }; + // end inline asm + mov.b32 %r628, %r672; + mov.b32 %r629, %r672; + mov.b32 %r630, %r672; + mov.b32 %r631, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r628, %r629, %r630, %r631 }, { %r508, %r509, %r510, %r511 }, { %r498, %r499 }, { %r628, %r629, %r630, %r631 }; + // end inline asm + mov.b32 %r632, %r672; + mov.b32 %r633, %r672; + mov.b32 %r634, %r672; + mov.b32 %r635, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r632, %r633, %r634, %r635 }, { %r508, %r509, %r510, %r511 }, { %r500, %r501 }, { %r632, %r633, %r634, %r635 }; + // end inline asm + mov.b32 %r636, %r672; + mov.b32 %r637, %r672; + mov.b32 %r638, %r672; + mov.b32 %r639, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r636, %r637, %r638, %r639 }, { %r508, %r509, %r510, %r511 }, { %r502, %r503 }, { %r636, %r637, %r638, %r639 }; + // end inline asm + mov.b32 %r640, %r672; + mov.b32 %r641, %r672; + mov.b32 %r642, %r672; + mov.b32 %r643, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r640, %r641, %r642, %r643 }, { %r512, %r513, %r514, %r515 }, { %r488, %r489 }, { %r640, %r641, %r642, %r643 }; + // end inline asm + mov.b32 %r648, %r672; + mov.b32 %r649, %r672; + mov.b32 %r650, %r672; + mov.b32 %r651, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r648, %r649, %r650, %r651 }, { %r512, %r513, %r514, %r515 }, { %r490, %r491 }, { %r648, %r649, %r650, %r651 }; + // end inline asm + mov.b32 %r652, %r672; + mov.b32 %r653, %r672; + mov.b32 %r654, %r672; + mov.b32 %r655, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r652, %r653, %r654, %r655 }, { %r512, %r513, %r514, %r515 }, { %r492, %r493 }, { %r652, %r653, %r654, %r655 }; + // end inline asm + mov.b32 %r656, %r672; + mov.b32 %r657, %r672; + mov.b32 %r658, %r672; + mov.b32 %r659, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r656, %r657, %r658, %r659 }, { %r512, %r513, %r514, %r515 }, { %r494, %r495 }, { %r656, %r657, %r658, %r659 }; + // end inline asm + mov.b32 %r660, %r672; + mov.b32 %r661, %r672; + mov.b32 %r662, %r672; + mov.b32 %r663, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r660, %r661, %r662, %r663 }, { %r512, %r513, %r514, %r515 }, { %r496, %r497 }, { %r660, %r661, %r662, %r663 }; + // end inline asm + mov.b32 %r664, %r672; + mov.b32 %r665, %r672; + mov.b32 %r666, %r672; + mov.b32 %r667, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r664, %r665, %r666, %r667 }, { %r512, %r513, %r514, %r515 }, { %r498, %r499 }, { %r664, %r665, %r666, %r667 }; + // end inline asm + mov.b32 %r668, %r672; + mov.b32 %r669, %r672; + mov.b32 %r670, %r672; + mov.b32 %r671, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r668, %r669, %r670, %r671 }, { %r512, %r513, %r514, %r515 }, { %r500, %r501 }, { %r668, %r669, %r670, %r671 }; + // end inline asm + mov.b32 %r673, %r672; + mov.b32 %r674, %r672; + mov.b32 %r675, %r672; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r672, %r673, %r674, %r675 }, { %r512, %r513, %r514, %r515 }, { %r502, %r503 }, { %r672, %r673, %r674, %r675 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r516, %r517, %r518, %r519 }, { %r520, %r521, %r522, %r523 }, { %r524, %r525 }, { %r516, %r517, %r518, %r519 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r526, %r527, %r528, %r529 }, { %r520, %r521, %r522, %r523 }, { %r530, %r531 }, { %r526, %r527, %r528, %r529 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r532, %r533, %r534, %r535 }, { %r520, %r521, %r522, %r523 }, { %r536, %r537 }, { %r532, %r533, %r534, %r535 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r538, %r539, %r540, %r541 }, { %r520, %r521, %r522, %r523 }, { %r542, %r543 }, { %r538, %r539, %r540, %r541 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r544, %r545, %r546, %r547 }, { %r520, %r521, %r522, %r523 }, { %r548, %r549 }, { %r544, %r545, %r546, %r547 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r550, %r551, %r552, %r553 }, { %r520, %r521, %r522, %r523 }, { %r554, %r555 }, { %r550, %r551, %r552, %r553 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r556, %r557, %r558, %r559 }, { %r520, %r521, %r522, %r523 }, { %r560, %r561 }, { %r556, %r557, %r558, %r559 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r562, %r563, %r564, %r565 }, { %r520, %r521, %r522, %r523 }, { %r566, %r567 }, { %r562, %r563, %r564, %r565 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r568, %r569, %r570, %r571 }, { %r572, %r573, %r574, %r575 }, { %r524, %r525 }, { %r568, %r569, %r570, %r571 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r576, %r577, %r578, %r579 }, { %r572, %r573, %r574, %r575 }, { %r530, %r531 }, { %r576, %r577, %r578, %r579 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r580, %r581, %r582, %r583 }, { %r572, %r573, %r574, %r575 }, { %r536, %r537 }, { %r580, %r581, %r582, %r583 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r584, %r585, %r586, %r587 }, { %r572, %r573, %r574, %r575 }, { %r542, %r543 }, { %r584, %r585, %r586, %r587 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r588, %r589, %r590, %r591 }, { %r572, %r573, %r574, %r575 }, { %r548, %r549 }, { %r588, %r589, %r590, %r591 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r592, %r593, %r594, %r595 }, { %r572, %r573, %r574, %r575 }, { %r554, %r555 }, { %r592, %r593, %r594, %r595 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r596, %r597, %r598, %r599 }, { %r572, %r573, %r574, %r575 }, { %r560, %r561 }, { %r596, %r597, %r598, %r599 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r600, %r601, %r602, %r603 }, { %r572, %r573, %r574, %r575 }, { %r566, %r567 }, { %r600, %r601, %r602, %r603 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r604, %r605, %r606, %r607 }, { %r608, %r609, %r610, %r611 }, { %r524, %r525 }, { %r604, %r605, %r606, %r607 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r612, %r613, %r614, %r615 }, { %r608, %r609, %r610, %r611 }, { %r530, %r531 }, { %r612, %r613, %r614, %r615 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r616, %r617, %r618, %r619 }, { %r608, %r609, %r610, %r611 }, { %r536, %r537 }, { %r616, %r617, %r618, %r619 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r620, %r621, %r622, %r623 }, { %r608, %r609, %r610, %r611 }, { %r542, %r543 }, { %r620, %r621, %r622, %r623 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r624, %r625, %r626, %r627 }, { %r608, %r609, %r610, %r611 }, { %r548, %r549 }, { %r624, %r625, %r626, %r627 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r628, %r629, %r630, %r631 }, { %r608, %r609, %r610, %r611 }, { %r554, %r555 }, { %r628, %r629, %r630, %r631 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r632, %r633, %r634, %r635 }, { %r608, %r609, %r610, %r611 }, { %r560, %r561 }, { %r632, %r633, %r634, %r635 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r636, %r637, %r638, %r639 }, { %r608, %r609, %r610, %r611 }, { %r566, %r567 }, { %r636, %r637, %r638, %r639 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r640, %r641, %r642, %r643 }, { %r644, %r645, %r646, %r647 }, { %r524, %r525 }, { %r640, %r641, %r642, %r643 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r648, %r649, %r650, %r651 }, { %r644, %r645, %r646, %r647 }, { %r530, %r531 }, { %r648, %r649, %r650, %r651 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r652, %r653, %r654, %r655 }, { %r644, %r645, %r646, %r647 }, { %r536, %r537 }, { %r652, %r653, %r654, %r655 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r656, %r657, %r658, %r659 }, { %r644, %r645, %r646, %r647 }, { %r542, %r543 }, { %r656, %r657, %r658, %r659 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r660, %r661, %r662, %r663 }, { %r644, %r645, %r646, %r647 }, { %r548, %r549 }, { %r660, %r661, %r662, %r663 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r664, %r665, %r666, %r667 }, { %r644, %r645, %r646, %r647 }, { %r554, %r555 }, { %r664, %r665, %r666, %r667 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r668, %r669, %r670, %r671 }, { %r644, %r645, %r646, %r647 }, { %r560, %r561 }, { %r668, %r669, %r670, %r671 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r672, %r673, %r674, %r675 }, { %r644, %r645, %r646, %r647 }, { %r566, %r567 }, { %r672, %r673, %r674, %r675 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r516, %r517, %r518, %r519 }, { %r676, %r677, %r678, %r679 }, { %r680, %r681 }, { %r516, %r517, %r518, %r519 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r526, %r527, %r528, %r529 }, { %r676, %r677, %r678, %r679 }, { %r682, %r683 }, { %r526, %r527, %r528, %r529 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r532, %r533, %r534, %r535 }, { %r676, %r677, %r678, %r679 }, { %r684, %r685 }, { %r532, %r533, %r534, %r535 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r538, %r539, %r540, %r541 }, { %r676, %r677, %r678, %r679 }, { %r686, %r687 }, { %r538, %r539, %r540, %r541 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r544, %r545, %r546, %r547 }, { %r676, %r677, %r678, %r679 }, { %r688, %r689 }, { %r544, %r545, %r546, %r547 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r550, %r551, %r552, %r553 }, { %r676, %r677, %r678, %r679 }, { %r690, %r691 }, { %r550, %r551, %r552, %r553 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r556, %r557, %r558, %r559 }, { %r676, %r677, %r678, %r679 }, { %r692, %r693 }, { %r556, %r557, %r558, %r559 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r562, %r563, %r564, %r565 }, { %r676, %r677, %r678, %r679 }, { %r694, %r695 }, { %r562, %r563, %r564, %r565 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r568, %r569, %r570, %r571 }, { %r696, %r697, %r698, %r699 }, { %r680, %r681 }, { %r568, %r569, %r570, %r571 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r576, %r577, %r578, %r579 }, { %r696, %r697, %r698, %r699 }, { %r682, %r683 }, { %r576, %r577, %r578, %r579 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r580, %r581, %r582, %r583 }, { %r696, %r697, %r698, %r699 }, { %r684, %r685 }, { %r580, %r581, %r582, %r583 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r584, %r585, %r586, %r587 }, { %r696, %r697, %r698, %r699 }, { %r686, %r687 }, { %r584, %r585, %r586, %r587 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r588, %r589, %r590, %r591 }, { %r696, %r697, %r698, %r699 }, { %r688, %r689 }, { %r588, %r589, %r590, %r591 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r592, %r593, %r594, %r595 }, { %r696, %r697, %r698, %r699 }, { %r690, %r691 }, { %r592, %r593, %r594, %r595 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r596, %r597, %r598, %r599 }, { %r696, %r697, %r698, %r699 }, { %r692, %r693 }, { %r596, %r597, %r598, %r599 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r600, %r601, %r602, %r603 }, { %r696, %r697, %r698, %r699 }, { %r694, %r695 }, { %r600, %r601, %r602, %r603 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r604, %r605, %r606, %r607 }, { %r700, %r701, %r702, %r703 }, { %r680, %r681 }, { %r604, %r605, %r606, %r607 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r612, %r613, %r614, %r615 }, { %r700, %r701, %r702, %r703 }, { %r682, %r683 }, { %r612, %r613, %r614, %r615 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r616, %r617, %r618, %r619 }, { %r700, %r701, %r702, %r703 }, { %r684, %r685 }, { %r616, %r617, %r618, %r619 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r620, %r621, %r622, %r623 }, { %r700, %r701, %r702, %r703 }, { %r686, %r687 }, { %r620, %r621, %r622, %r623 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r624, %r625, %r626, %r627 }, { %r700, %r701, %r702, %r703 }, { %r688, %r689 }, { %r624, %r625, %r626, %r627 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r628, %r629, %r630, %r631 }, { %r700, %r701, %r702, %r703 }, { %r690, %r691 }, { %r628, %r629, %r630, %r631 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r632, %r633, %r634, %r635 }, { %r700, %r701, %r702, %r703 }, { %r692, %r693 }, { %r632, %r633, %r634, %r635 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r636, %r637, %r638, %r639 }, { %r700, %r701, %r702, %r703 }, { %r694, %r695 }, { %r636, %r637, %r638, %r639 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r640, %r641, %r642, %r643 }, { %r704, %r705, %r706, %r707 }, { %r680, %r681 }, { %r640, %r641, %r642, %r643 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r648, %r649, %r650, %r651 }, { %r704, %r705, %r706, %r707 }, { %r682, %r683 }, { %r648, %r649, %r650, %r651 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r652, %r653, %r654, %r655 }, { %r704, %r705, %r706, %r707 }, { %r684, %r685 }, { %r652, %r653, %r654, %r655 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r656, %r657, %r658, %r659 }, { %r704, %r705, %r706, %r707 }, { %r686, %r687 }, { %r656, %r657, %r658, %r659 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r660, %r661, %r662, %r663 }, { %r704, %r705, %r706, %r707 }, { %r688, %r689 }, { %r660, %r661, %r662, %r663 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r664, %r665, %r666, %r667 }, { %r704, %r705, %r706, %r707 }, { %r690, %r691 }, { %r664, %r665, %r666, %r667 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r668, %r669, %r670, %r671 }, { %r704, %r705, %r706, %r707 }, { %r692, %r693 }, { %r668, %r669, %r670, %r671 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r672, %r673, %r674, %r675 }, { %r704, %r705, %r706, %r707 }, { %r694, %r695 }, { %r672, %r673, %r674, %r675 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r516, %r517, %r518, %r519 }, { %r708, %r709, %r710, %r711 }, { %r712, %r713 }, { %r516, %r517, %r518, %r519 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r526, %r527, %r528, %r529 }, { %r708, %r709, %r710, %r711 }, { %r714, %r715 }, { %r526, %r527, %r528, %r529 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r532, %r533, %r534, %r535 }, { %r708, %r709, %r710, %r711 }, { %r716, %r717 }, { %r532, %r533, %r534, %r535 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r538, %r539, %r540, %r541 }, { %r708, %r709, %r710, %r711 }, { %r718, %r719 }, { %r538, %r539, %r540, %r541 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r544, %r545, %r546, %r547 }, { %r708, %r709, %r710, %r711 }, { %r720, %r721 }, { %r544, %r545, %r546, %r547 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r550, %r551, %r552, %r553 }, { %r708, %r709, %r710, %r711 }, { %r722, %r723 }, { %r550, %r551, %r552, %r553 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r556, %r557, %r558, %r559 }, { %r708, %r709, %r710, %r711 }, { %r724, %r725 }, { %r556, %r557, %r558, %r559 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r562, %r563, %r564, %r565 }, { %r708, %r709, %r710, %r711 }, { %r726, %r727 }, { %r562, %r563, %r564, %r565 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r568, %r569, %r570, %r571 }, { %r728, %r729, %r730, %r731 }, { %r712, %r713 }, { %r568, %r569, %r570, %r571 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r576, %r577, %r578, %r579 }, { %r728, %r729, %r730, %r731 }, { %r714, %r715 }, { %r576, %r577, %r578, %r579 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r580, %r581, %r582, %r583 }, { %r728, %r729, %r730, %r731 }, { %r716, %r717 }, { %r580, %r581, %r582, %r583 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r584, %r585, %r586, %r587 }, { %r728, %r729, %r730, %r731 }, { %r718, %r719 }, { %r584, %r585, %r586, %r587 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r588, %r589, %r590, %r591 }, { %r728, %r729, %r730, %r731 }, { %r720, %r721 }, { %r588, %r589, %r590, %r591 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r592, %r593, %r594, %r595 }, { %r728, %r729, %r730, %r731 }, { %r722, %r723 }, { %r592, %r593, %r594, %r595 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r596, %r597, %r598, %r599 }, { %r728, %r729, %r730, %r731 }, { %r724, %r725 }, { %r596, %r597, %r598, %r599 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r600, %r601, %r602, %r603 }, { %r728, %r729, %r730, %r731 }, { %r726, %r727 }, { %r600, %r601, %r602, %r603 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r604, %r605, %r606, %r607 }, { %r732, %r733, %r734, %r735 }, { %r712, %r713 }, { %r604, %r605, %r606, %r607 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r612, %r613, %r614, %r615 }, { %r732, %r733, %r734, %r735 }, { %r714, %r715 }, { %r612, %r613, %r614, %r615 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r616, %r617, %r618, %r619 }, { %r732, %r733, %r734, %r735 }, { %r716, %r717 }, { %r616, %r617, %r618, %r619 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r620, %r621, %r622, %r623 }, { %r732, %r733, %r734, %r735 }, { %r718, %r719 }, { %r620, %r621, %r622, %r623 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r624, %r625, %r626, %r627 }, { %r732, %r733, %r734, %r735 }, { %r720, %r721 }, { %r624, %r625, %r626, %r627 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r628, %r629, %r630, %r631 }, { %r732, %r733, %r734, %r735 }, { %r722, %r723 }, { %r628, %r629, %r630, %r631 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r632, %r633, %r634, %r635 }, { %r732, %r733, %r734, %r735 }, { %r724, %r725 }, { %r632, %r633, %r634, %r635 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r636, %r637, %r638, %r639 }, { %r732, %r733, %r734, %r735 }, { %r726, %r727 }, { %r636, %r637, %r638, %r639 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r640, %r641, %r642, %r643 }, { %r736, %r737, %r738, %r739 }, { %r712, %r713 }, { %r640, %r641, %r642, %r643 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r648, %r649, %r650, %r651 }, { %r736, %r737, %r738, %r739 }, { %r714, %r715 }, { %r648, %r649, %r650, %r651 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r652, %r653, %r654, %r655 }, { %r736, %r737, %r738, %r739 }, { %r716, %r717 }, { %r652, %r653, %r654, %r655 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r656, %r657, %r658, %r659 }, { %r736, %r737, %r738, %r739 }, { %r718, %r719 }, { %r656, %r657, %r658, %r659 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r660, %r661, %r662, %r663 }, { %r736, %r737, %r738, %r739 }, { %r720, %r721 }, { %r660, %r661, %r662, %r663 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r664, %r665, %r666, %r667 }, { %r736, %r737, %r738, %r739 }, { %r722, %r723 }, { %r664, %r665, %r666, %r667 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r668, %r669, %r670, %r671 }, { %r736, %r737, %r738, %r739 }, { %r724, %r725 }, { %r668, %r669, %r670, %r671 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r672, %r673, %r674, %r675 }, { %r736, %r737, %r738, %r739 }, { %r726, %r727 }, { %r672, %r673, %r674, %r675 }; + // end inline asm + .loc 1 540 31 // chunk_o.py:540:31 + cvt.rn.bf16x2.f32 %r1080, %r517, %r516; + cvt.rn.bf16x2.f32 %r1081, %r519, %r518; + cvt.rn.bf16x2.f32 %r1082, %r527, %r526; + cvt.rn.bf16x2.f32 %r1083, %r529, %r528; + cvt.rn.bf16x2.f32 %r1084, %r533, %r532; + cvt.rn.bf16x2.f32 %r1085, %r535, %r534; + cvt.rn.bf16x2.f32 %r1086, %r539, %r538; + cvt.rn.bf16x2.f32 %r1087, %r541, %r540; + cvt.rn.bf16x2.f32 %r1088, %r545, %r544; + cvt.rn.bf16x2.f32 %r1089, %r547, %r546; + cvt.rn.bf16x2.f32 %r1090, %r551, %r550; + cvt.rn.bf16x2.f32 %r1091, %r553, %r552; + cvt.rn.bf16x2.f32 %r1092, %r557, %r556; + cvt.rn.bf16x2.f32 %r1093, %r559, %r558; + cvt.rn.bf16x2.f32 %r1094, %r563, %r562; + cvt.rn.bf16x2.f32 %r1095, %r565, %r564; + cvt.rn.bf16x2.f32 %r1096, %r569, %r568; + cvt.rn.bf16x2.f32 %r1097, %r571, %r570; + cvt.rn.bf16x2.f32 %r1098, %r577, %r576; + cvt.rn.bf16x2.f32 %r1099, %r579, %r578; + cvt.rn.bf16x2.f32 %r1100, %r581, %r580; + cvt.rn.bf16x2.f32 %r1101, %r583, %r582; + cvt.rn.bf16x2.f32 %r1102, %r585, %r584; + cvt.rn.bf16x2.f32 %r1103, %r587, %r586; + cvt.rn.bf16x2.f32 %r1104, %r589, %r588; + cvt.rn.bf16x2.f32 %r1105, %r591, %r590; + cvt.rn.bf16x2.f32 %r1106, %r593, %r592; + cvt.rn.bf16x2.f32 %r1107, %r595, %r594; + cvt.rn.bf16x2.f32 %r1108, %r597, %r596; + cvt.rn.bf16x2.f32 %r1109, %r599, %r598; + cvt.rn.bf16x2.f32 %r1110, %r601, %r600; + cvt.rn.bf16x2.f32 %r1111, %r603, %r602; + cvt.rn.bf16x2.f32 %r1112, %r605, %r604; + cvt.rn.bf16x2.f32 %r1113, %r607, %r606; + cvt.rn.bf16x2.f32 %r1114, %r613, %r612; + cvt.rn.bf16x2.f32 %r1115, %r615, %r614; + cvt.rn.bf16x2.f32 %r1116, %r617, %r616; + cvt.rn.bf16x2.f32 %r1117, %r619, %r618; + cvt.rn.bf16x2.f32 %r1118, %r621, %r620; + cvt.rn.bf16x2.f32 %r1119, %r623, %r622; + cvt.rn.bf16x2.f32 %r1120, %r625, %r624; + cvt.rn.bf16x2.f32 %r1121, %r627, %r626; + cvt.rn.bf16x2.f32 %r1122, %r629, %r628; + cvt.rn.bf16x2.f32 %r1123, %r631, %r630; + cvt.rn.bf16x2.f32 %r1124, %r633, %r632; + cvt.rn.bf16x2.f32 %r1125, %r635, %r634; + cvt.rn.bf16x2.f32 %r1126, %r637, %r636; + cvt.rn.bf16x2.f32 %r1127, %r639, %r638; + cvt.rn.bf16x2.f32 %r1128, %r641, %r640; + cvt.rn.bf16x2.f32 %r1129, %r643, %r642; + cvt.rn.bf16x2.f32 %r1130, %r649, %r648; + cvt.rn.bf16x2.f32 %r1131, %r651, %r650; + cvt.rn.bf16x2.f32 %r1132, %r653, %r652; + cvt.rn.bf16x2.f32 %r1133, %r655, %r654; + cvt.rn.bf16x2.f32 %r1134, %r657, %r656; + cvt.rn.bf16x2.f32 %r1135, %r659, %r658; + cvt.rn.bf16x2.f32 %r1136, %r661, %r660; + cvt.rn.bf16x2.f32 %r1137, %r663, %r662; + cvt.rn.bf16x2.f32 %r1138, %r665, %r664; + cvt.rn.bf16x2.f32 %r1139, %r667, %r666; + cvt.rn.bf16x2.f32 %r1140, %r669, %r668; + cvt.rn.bf16x2.f32 %r1141, %r671, %r670; + cvt.rn.bf16x2.f32 %r1142, %r673, %r672; + cvt.rn.bf16x2.f32 %r1143, %r675, %r674; + .loc 1 540 23 // chunk_o.py:540:23 + bar.sync 0; + shl.b32 %r1144, %r831, 2; + xor.b32 %r1145, %r830, %r1144; + xor.b32 %r1146, %r1145, %r1054; + add.s32 %r1147, %r834, %r884; + add.s32 %r1148, %r1147, %r1146; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1148], {%r1080, %r1082, %r1084, %r1086}; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1148+128], {%r1088, %r1090, %r1092, %r1094}; + bar.sync 0; + ld.shared.v4.b32 {%r740, %r741, %r742, %r743}, [%r4]; + ld.shared.v4.b32 {%r744, %r745, %r746, %r747}, [%r873+1024]; + bar.sync 0; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1148], {%r1081, %r1083, %r1085, %r1087}; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1148+128], {%r1089, %r1091, %r1093, %r1095}; + bar.sync 0; + ld.shared.v4.b32 {%r748, %r749, %r750, %r751}, [%r4]; + ld.shared.v4.b32 {%r752, %r753, %r754, %r755}, [%r873+1024]; + bar.sync 0; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1148], {%r1096, %r1098, %r1100, %r1102}; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1148+128], {%r1104, %r1106, %r1108, %r1110}; + bar.sync 0; + ld.shared.v4.b32 {%r756, %r757, %r758, %r759}, [%r4]; + ld.shared.v4.b32 {%r760, %r761, %r762, %r763}, [%r873+1024]; + bar.sync 0; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1148], {%r1097, %r1099, %r1101, %r1103}; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1148+128], {%r1105, %r1107, %r1109, %r1111}; + bar.sync 0; + ld.shared.v4.b32 {%r764, %r765, %r766, %r767}, [%r4]; + ld.shared.v4.b32 {%r768, %r769, %r770, %r771}, [%r873+1024]; + bar.sync 0; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1148], {%r1112, %r1114, %r1116, %r1118}; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1148+128], {%r1120, %r1122, %r1124, %r1126}; + bar.sync 0; + ld.shared.v4.b32 {%r772, %r773, %r774, %r775}, [%r4]; + ld.shared.v4.b32 {%r776, %r777, %r778, %r779}, [%r873+1024]; + bar.sync 0; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1148], {%r1113, %r1115, %r1117, %r1119}; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1148+128], {%r1121, %r1123, %r1125, %r1127}; + bar.sync 0; + ld.shared.v4.b32 {%r780, %r781, %r782, %r783}, [%r4]; + ld.shared.v4.b32 {%r784, %r785, %r786, %r787}, [%r873+1024]; + bar.sync 0; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1148], {%r1128, %r1130, %r1132, %r1134}; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1148+128], {%r1136, %r1138, %r1140, %r1142}; + bar.sync 0; + ld.shared.v4.b32 {%r788, %r789, %r790, %r791}, [%r4]; + ld.shared.v4.b32 {%r792, %r793, %r794, %r795}, [%r873+1024]; + bar.sync 0; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1148], {%r1129, %r1131, %r1133, %r1135}; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1148+128], {%r1137, %r1139, %r1141, %r1143}; + bar.sync 0; + ld.shared.v4.b32 {%r796, %r797, %r798, %r799}, [%r4]; + ld.shared.v4.b32 {%r800, %r801, %r802, %r803}, [%r873+1024]; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd50 + 0 ], { %r740, %r741, %r742, %r743 }; + // end inline asm + // begin inline asm + @%p3 st.global.v4.b32 [ %rd51 + 0 ], { %r744, %r745, %r746, %r747 }; + // end inline asm + // begin inline asm + @%p4 st.global.v4.b32 [ %rd52 + 0 ], { %r748, %r749, %r750, %r751 }; + // end inline asm + // begin inline asm + @%p5 st.global.v4.b32 [ %rd53 + 0 ], { %r752, %r753, %r754, %r755 }; + // end inline asm + // begin inline asm + @%p6 st.global.v4.b32 [ %rd54 + 0 ], { %r756, %r757, %r758, %r759 }; + // end inline asm + // begin inline asm + @%p7 st.global.v4.b32 [ %rd55 + 0 ], { %r760, %r761, %r762, %r763 }; + // end inline asm + // begin inline asm + @%p8 st.global.v4.b32 [ %rd56 + 0 ], { %r764, %r765, %r766, %r767 }; + // end inline asm + // begin inline asm + @%p9 st.global.v4.b32 [ %rd57 + 0 ], { %r768, %r769, %r770, %r771 }; + // end inline asm + // begin inline asm + @%p17 st.global.v4.b32 [ %rd58 + 0 ], { %r772, %r773, %r774, %r775 }; + // end inline asm + // begin inline asm + @%p10 st.global.v4.b32 [ %rd59 + 0 ], { %r776, %r777, %r778, %r779 }; + // end inline asm + // begin inline asm + @%p11 st.global.v4.b32 [ %rd60 + 0 ], { %r780, %r781, %r782, %r783 }; + // end inline asm + // begin inline asm + @%p12 st.global.v4.b32 [ %rd61 + 0 ], { %r784, %r785, %r786, %r787 }; + // end inline asm + // begin inline asm + @%p13 st.global.v4.b32 [ %rd62 + 0 ], { %r788, %r789, %r790, %r791 }; + // end inline asm + // begin inline asm + @%p14 st.global.v4.b32 [ %rd63 + 0 ], { %r792, %r793, %r794, %r795 }; + // end inline asm + // begin inline asm + @%p15 st.global.v4.b32 [ %rd64 + 0 ], { %r796, %r797, %r798, %r799 }; + // end inline asm + // begin inline asm + @%p16 st.global.v4.b32 [ %rd65 + 0 ], { %r800, %r801, %r802, %r803 }; + // end inline asm + .loc 1 533 4 // chunk_o.py:533:4 + ret; +$L__tmp5: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 178 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xab DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 99 // DW_AT_name +.b8 104 +.b8 117 +.b8 110 +.b8 107 +.b8 95 +.b8 111 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 111 +.b8 112 +.b8 115 +.b8 47 +.b8 99 +.b8 111 +.b8 109 +.b8 109 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // Abbrev [2] 0x6a:0x1c DW_TAG_subprogram +.b8 99 // DW_AT_name +.b8 104 +.b8 117 +.b8 110 +.b8 107 +.b8 95 +.b8 98 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 95 +.b8 100 +.b8 118 +.b8 95 +.b8 108 +.b8 111 +.b8 99 +.b8 97 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x86:0x2f DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 106 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x9b:0x19 DW_TAG_inlined_subroutine +.b32 106 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp4 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 17 // DW_AT_call_line +.b8 2 +.b8 24 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.source b/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.source new file mode 100644 index 0000000000000000000000000000000000000000..36a53942f374e9b526eb7f27bdc28ca064f17ed8 --- /dev/null +++ b/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.source @@ -0,0 +1,504 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":466:0) +#loc114 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:0) +#loc118 = loc("q"(#loc)) +#loc119 = loc("k"(#loc)) +#loc120 = loc("g"(#loc)) +#loc121 = loc("do"(#loc)) +#loc122 = loc("dv"(#loc)) +#loc123 = loc("scale"(#loc)) +#loc124 = loc("T"(#loc)) +#loc212 = loc("x"(#loc114)) +module { + tt.func public @chunk_bwd_kernel_dv_local(%q: !tt.ptr {tt.divisibility = 16 : i32} loc("q"(#loc)), %k: !tt.ptr {tt.divisibility = 16 : i32} loc("k"(#loc)), %g: !tt.ptr {tt.divisibility = 16 : i32} loc("g"(#loc)), %do: !tt.ptr {tt.divisibility = 16 : i32} loc("do"(#loc)), %dv: !tt.ptr {tt.divisibility = 16 : i32} loc("dv"(#loc)), %scale: f32 loc("scale"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %0 = tt.get_program_id x : i32 loc(#loc1) + %1 = arith.extsi %0 : i32 to i64 loc(#loc2) + %2 = tt.get_program_id y : i32 loc(#loc3) + %3 = arith.extsi %2 : i32 to i64 loc(#loc4) + %c32_i32 = arith.constant 32 : i32 loc(#loc5) + %c32_i64 = arith.constant 32 : i64 loc(#loc5) + %4 = arith.divsi %3, %c32_i64 : i64 loc(#loc5) + %c32_i32_0 = arith.constant 32 : i32 loc(#loc6) + %c32_i64_1 = arith.constant 32 : i64 loc(#loc6) + %5 = arith.remsi %3, %c32_i64_1 : i64 loc(#loc6) + %6 = arith.extsi %T : i32 to i64 loc(#loc7) + %7 = arith.muli %4, %6 : i64 loc(#loc7) + %8 = arith.extsi %T : i32 to i64 loc(#loc8) + %9 = arith.muli %4, %8 : i64 loc(#loc8) + %10 = arith.extsi %T : i32 to i64 loc(#loc9) + %11 = arith.addi %9, %10 : i64 loc(#loc9) + %q_2 = arith.constant 32 : i32 loc(#loc125) + %q_3 = arith.constant 32 : i64 loc(#loc125) + %q_4 = arith.muli %7, %q_3 : i64 loc(#loc125) + %q_5 = arith.constant 1 : i32 loc(#loc126) + %q_6 = arith.constant 1 : i64 loc(#loc126) + %q_7 = arith.divsi %5, %q_6 : i64 loc(#loc126) + %q_8 = arith.addi %q_4, %q_7 : i64 loc(#loc127) + %q_9 = arith.constant 128 : i32 loc(#loc128) + %q_10 = arith.constant 128 : i64 loc(#loc128) + %q_11 = arith.muli %q_8, %q_10 : i64 loc(#loc128) + %q_12 = tt.addptr %q, %q_11 : !tt.ptr, i64 loc(#loc129) + %k_13 = arith.constant 32 : i32 loc(#loc130) + %k_14 = arith.constant 32 : i64 loc(#loc130) + %k_15 = arith.muli %7, %k_14 : i64 loc(#loc130) + %k_16 = arith.constant 1 : i32 loc(#loc131) + %k_17 = arith.constant 1 : i64 loc(#loc131) + %k_18 = arith.divsi %5, %k_17 : i64 loc(#loc131) + %k_19 = arith.addi %k_15, %k_18 : i64 loc(#loc132) + %k_20 = arith.constant 128 : i32 loc(#loc133) + %k_21 = arith.constant 128 : i64 loc(#loc133) + %k_22 = arith.muli %k_19, %k_21 : i64 loc(#loc133) + %k_23 = tt.addptr %k, %k_22 : !tt.ptr, i64 loc(#loc134) + %do_24 = arith.constant 32 : i32 loc(#loc135) + %do_25 = arith.constant 32 : i64 loc(#loc135) + %do_26 = arith.muli %7, %do_25 : i64 loc(#loc135) + %do_27 = arith.addi %do_26, %5 : i64 loc(#loc136) + %do_28 = arith.constant 128 : i32 loc(#loc137) + %do_29 = arith.constant 128 : i64 loc(#loc137) + %do_30 = arith.muli %do_27, %do_29 : i64 loc(#loc137) + %do_31 = tt.addptr %do, %do_30 : !tt.ptr, i64 loc(#loc138) + %dv_32 = arith.constant 32 : i32 loc(#loc139) + %dv_33 = arith.constant 32 : i64 loc(#loc139) + %dv_34 = arith.muli %7, %dv_33 : i64 loc(#loc139) + %dv_35 = arith.addi %dv_34, %5 : i64 loc(#loc140) + %dv_36 = arith.constant 128 : i32 loc(#loc141) + %dv_37 = arith.constant 128 : i64 loc(#loc141) + %dv_38 = arith.muli %dv_35, %dv_37 : i64 loc(#loc141) + %dv_39 = tt.addptr %dv, %dv_38 : !tt.ptr, i64 loc(#loc142) + %o_t = arith.constant 64 : i32 loc(#loc143) + %o_t_40 = arith.constant 64 : i64 loc(#loc143) + %o_t_41 = arith.muli %1, %o_t_40 : i64 loc(#loc143) + %o_t_42 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc144) + %o_t_43 = arith.extsi %o_t_42 : tensor<64xi32> to tensor<64xi64> loc(#loc145) + %o_t_44 = tt.splat %o_t_41 : i64 -> tensor<64xi64> loc(#loc145) + %o_t_45 = arith.addi %o_t_44, %o_t_43 : tensor<64xi64> loc(#loc145) + %m_t = arith.extsi %T : i32 to i64 loc(#loc146) + %m_t_46 = tt.splat %m_t : i64 -> tensor<64xi64> loc(#loc146) + %m_t_47 = arith.cmpi slt, %o_t_45, %m_t_46 : tensor<64xi64> loc(#loc146) + %g_48 = arith.constant 32 : i32 loc(#loc147) + %g_49 = arith.constant 32 : i64 loc(#loc147) + %g_50 = arith.muli %7, %g_49 : i64 loc(#loc147) + %g_51 = arith.addi %g_50, %5 : i64 loc(#loc148) + %g_52 = tt.addptr %g, %g_51 : !tt.ptr, i64 loc(#loc149) + %p_g = arith.constant 32 : i32 loc(#loc150) + %p_g_53 = arith.constant 32 : i64 loc(#loc150) + %p_g_54 = arith.constant dense<32> : tensor<64xi64> loc(#loc150) + %p_g_55 = arith.muli %o_t_45, %p_g_54 : tensor<64xi64> loc(#loc150) + %p_g_56 = tt.splat %g_52 : !tt.ptr -> tensor<64x!tt.ptr> loc(#loc151) + %p_g_57 = tt.addptr %p_g_56, %p_g_55 : tensor<64x!tt.ptr>, tensor<64xi64> loc(#loc151) + %b_g = arith.constant 0.000000e+00 : f32 loc(#loc152) + %b_g_58 = arith.constant dense<0.000000e+00> : tensor<64xf32> loc(#loc152) + %b_g_59 = tt.load %p_g_57, %m_t_47, %b_g_58 : tensor<64x!tt.ptr> loc(#loc152) + %b_A = tt.call @triton.language.standard.zeros__Tc64_c64T_cfp32() : () -> tensor<64x64xf32> loc(#loc153) + tt.call @triton.language.standard.cdiv__c128_c128() : () -> () loc(#loc39) + tt.call @triton.language.standard.cdiv__c128_c128() : () -> () loc(#loc39) + %c0_i32 = arith.constant 0 : i32 loc(#loc40) + %c1_i32 = arith.constant 1 : i32 loc(#loc40) + %c1_i32_60 = arith.constant 1 : i32 loc(#loc40) + %12 = arith.bitcast %c0_i32 : i32 to i32 loc(#loc40) + %13 = arith.bitcast %c1_i32 : i32 to i32 loc(#loc40) + %14 = arith.bitcast %c1_i32_60 : i32 to i32 loc(#loc40) + %15 = ub.poison : i32 loc(#loc40) + %b_A_61 = scf.for %i_k = %12 to %13 step %14 iter_args(%b_A_87 = %b_A) -> (tensor<64x64xf32>) : i32 { + %o_k = arith.constant 128 : i32 loc(#loc155) + %o_k_88 = arith.constant 128 : i32 loc(#loc155) + %o_k_89 = arith.extsi %i_k : i32 to i64 loc(#loc155) + %o_k_90 = arith.extsi %o_k_88 : i32 to i64 loc(#loc155) + %o_k_91 = arith.muli %o_k_89, %o_k_90 : i64 loc(#loc155) + %o_k_92 = arith.constant 2147483647 : i64 loc(#loc155) + %o_k_93 = arith.constant -2147483648 : i64 loc(#loc155) + %o_k_94 = arith.cmpi sle, %o_k_91, %o_k_92 : i64 loc(#loc155) + %o_k_95 = arith.cmpi sge, %o_k_91, %o_k_93 : i64 loc(#loc155) + %o_k_96 = arith.andi %o_k_94, %o_k_95 : i1 loc(#loc155) + %o_k_97 = arith.muli %i_k, %o_k_88 : i32 loc(#loc155) + %o_k_98 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc156) + %o_k_99 = tt.splat %o_k_97 : i32 -> tensor<128xi32> loc(#loc157) + %o_k_100 = arith.extsi %o_k_99 : tensor<128xi32> to tensor<128xi64> loc(#loc157) + %o_k_101 = arith.extsi %o_k_98 : tensor<128xi32> to tensor<128xi64> loc(#loc157) + %o_k_102 = arith.addi %o_k_100, %o_k_101 : tensor<128xi64> loc(#loc157) + %o_k_103 = arith.constant 2147483647 : i64 loc(#loc157) + %o_k_104 = arith.constant -2147483648 : i64 loc(#loc157) + %o_k_105 = arith.constant dense<2147483647> : tensor<128xi64> loc(#loc157) + %o_k_106 = arith.cmpi sle, %o_k_102, %o_k_105 : tensor<128xi64> loc(#loc157) + %o_k_107 = arith.constant dense<-2147483648> : tensor<128xi64> loc(#loc157) + %o_k_108 = arith.cmpi sge, %o_k_102, %o_k_107 : tensor<128xi64> loc(#loc157) + %o_k_109 = arith.andi %o_k_106, %o_k_108 : tensor<128xi1> loc(#loc157) + %o_k_110 = arith.addi %o_k_99, %o_k_98 : tensor<128xi32> loc(#loc157) + %m_k = arith.constant 128 : i32 loc(#loc158) + %m_k_111 = arith.constant dense<128> : tensor<128xi32> loc(#loc158) + %m_k_112 = arith.cmpi slt, %o_k_110, %m_k_111 : tensor<128xi32> loc(#loc158) + %p_k = tt.expand_dims %o_t_45 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc159) + %p_k_113 = arith.constant 4096 : i32 loc(#loc160) + %p_k_114 = arith.constant 4096 : i64 loc(#loc160) + %p_k_115 = arith.constant dense<4096> : tensor<64x1xi64> loc(#loc160) + %p_k_116 = arith.muli %p_k, %p_k_115 : tensor<64x1xi64> loc(#loc160) + %p_k_117 = tt.splat %k_23 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc161) + %p_k_118 = tt.addptr %p_k_117, %p_k_116 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc161) + %p_k_119 = tt.expand_dims %o_k_110 {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc162) + %p_k_120 = tt.broadcast %p_k_118 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc163) + %p_k_121 = tt.broadcast %p_k_119 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc163) + %p_k_122 = tt.addptr %p_k_120, %p_k_121 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc163) + %p_q = tt.expand_dims %o_k_110 {axis = 1 : i32} : tensor<128xi32> -> tensor<128x1xi32> loc(#loc164) + %p_q_123 = tt.splat %q_12 : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc165) + %p_q_124 = tt.addptr %p_q_123, %p_q : tensor<128x1x!tt.ptr>, tensor<128x1xi32> loc(#loc165) + %p_q_125 = tt.expand_dims %o_t_45 {axis = 0 : i32} : tensor<64xi64> -> tensor<1x64xi64> loc(#loc166) + %p_q_126 = arith.constant 4096 : i32 loc(#loc167) + %p_q_127 = arith.constant 4096 : i64 loc(#loc167) + %p_q_128 = arith.constant dense<4096> : tensor<1x64xi64> loc(#loc167) + %p_q_129 = arith.muli %p_q_125, %p_q_128 : tensor<1x64xi64> loc(#loc167) + %p_q_130 = tt.broadcast %p_q_124 : tensor<128x1x!tt.ptr> -> tensor<128x64x!tt.ptr> loc(#loc168) + %p_q_131 = tt.broadcast %p_q_129 : tensor<1x64xi64> -> tensor<128x64xi64> loc(#loc168) + %p_q_132 = tt.addptr %p_q_130, %p_q_131 : tensor<128x64x!tt.ptr>, tensor<128x64xi64> loc(#loc168) + %b_k = tt.expand_dims %m_t_47 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc169) + %b_k_133 = tt.expand_dims %m_k_112 {axis = 0 : i32} : tensor<128xi1> -> tensor<1x128xi1> loc(#loc170) + %b_k_134 = tt.broadcast %b_k : tensor<64x1xi1> -> tensor<64x128xi1> loc(#loc171) + %b_k_135 = tt.broadcast %b_k_133 : tensor<1x128xi1> -> tensor<64x128xi1> loc(#loc171) + %b_k_136 = arith.andi %b_k_134, %b_k_135 : tensor<64x128xi1> loc(#loc171) + %b_k_137 = arith.constant 0.000000e+00 : f32 loc(#loc172) + %b_k_138 = arith.constant dense<0.000000e+00> : tensor<64x128xf32> loc(#loc172) + %b_k_139 = arith.truncf %b_k_138 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc172) + %b_k_140 = tt.load %p_k_122, %b_k_136, %b_k_139 : tensor<64x128x!tt.ptr> loc(#loc172) + %b_q = tt.expand_dims %m_k_112 {axis = 1 : i32} : tensor<128xi1> -> tensor<128x1xi1> loc(#loc173) + %b_q_141 = tt.expand_dims %m_t_47 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc174) + %b_q_142 = tt.broadcast %b_q : tensor<128x1xi1> -> tensor<128x64xi1> loc(#loc175) + %b_q_143 = tt.broadcast %b_q_141 : tensor<1x64xi1> -> tensor<128x64xi1> loc(#loc175) + %b_q_144 = arith.andi %b_q_142, %b_q_143 : tensor<128x64xi1> loc(#loc175) + %b_q_145 = arith.constant 0.000000e+00 : f32 loc(#loc176) + %b_q_146 = arith.constant dense<0.000000e+00> : tensor<128x64xf32> loc(#loc176) + %b_q_147 = arith.truncf %b_q_146 : tensor<128x64xf32> to tensor<128x64xbf16> loc(#loc176) + %b_q_148 = tt.load %p_q_132, %b_q_144, %b_q_147 : tensor<128x64x!tt.ptr> loc(#loc176) + %b_A_149 = arith.constant 0.000000e+00 : f32 loc(#loc177) + %b_A_150 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc177) + %b_A_151 = tt.dot %b_k_140, %b_q_148, %b_A_150, inputPrecision = tf32 : tensor<64x128xbf16> * tensor<128x64xbf16> -> tensor<64x64xf32> loc(#loc177) + %b_A_152 = tt.splat %scale : f32 -> tensor<64x64xf32> loc(#loc178) + %b_A_153 = arith.mulf %b_A_151, %b_A_152 : tensor<64x64xf32> loc(#loc178) + %b_A_154 = arith.addf %b_A_87, %b_A_153 : tensor<64x64xf32> loc(#loc179) + scf.yield %b_A_154 : tensor<64x64xf32> loc(#loc66) + } loc(#loc154) + %b_A_62 = tt.expand_dims %b_g_59 {axis = 0 : i32} : tensor<64xf32> -> tensor<1x64xf32> loc(#loc180) + %b_A_63 = tt.expand_dims %b_g_59 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc181) + %b_A_64 = tt.broadcast %b_A_62 : tensor<1x64xf32> -> tensor<64x64xf32> loc(#loc182) + %b_A_65 = tt.broadcast %b_A_63 : tensor<64x1xf32> -> tensor<64x64xf32> loc(#loc182) + %b_A_66 = arith.subf %b_A_64, %b_A_65 : tensor<64x64xf32> loc(#loc182) + %b_A_67 = tt.call @fla.ops.utils.op.exp2__fp32S64_64S(%b_A_66) : (tensor<64x64xf32>) -> tensor<64x64xf32> loc(#loc183) + %b_A_68 = arith.mulf %b_A_61, %b_A_67 : tensor<64x64xf32> loc(#loc184) + %m_A = tt.expand_dims %o_t_45 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc185) + %m_A_69 = tt.expand_dims %o_t_45 {axis = 0 : i32} : tensor<64xi64> -> tensor<1x64xi64> loc(#loc186) + %m_A_70 = tt.broadcast %m_A : tensor<64x1xi64> -> tensor<64x64xi64> loc(#loc187) + %m_A_71 = tt.broadcast %m_A_69 : tensor<1x64xi64> -> tensor<64x64xi64> loc(#loc187) + %m_A_72 = arith.cmpi sle, %m_A_70, %m_A_71 : tensor<64x64xi64> loc(#loc187) + %m_A_73 = tt.expand_dims %m_t_47 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc188) + %m_A_74 = tt.expand_dims %m_t_47 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc189) + %m_A_75 = tt.broadcast %m_A_73 : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc189) + %m_A_76 = tt.broadcast %m_A_74 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc189) + %m_A_77 = arith.andi %m_A_75, %m_A_76 : tensor<64x64xi1> loc(#loc189) + %m_A_78 = arith.andi %m_A_72, %m_A_77 : tensor<64x64xi1> loc(#loc190) + %b_A_79 = arith.constant 0 : i32 loc(#loc191) + %b_A_80 = arith.constant 0.000000e+00 : f32 loc(#loc191) + %b_A_81 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc191) + %b_A_82 = arith.select %m_A_78, %b_A_68, %b_A_81 : tensor<64x64xi1>, tensor<64x64xf32> loc(#loc191) + %b_A_83 = arith.truncf %b_A_82 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc192) + tt.call @triton.language.standard.cdiv__c128_c128() : () -> () loc(#loc80) + tt.call @triton.language.standard.cdiv__c128_c128() : () -> () loc(#loc80) + %c0_i32_84 = arith.constant 0 : i32 loc(#loc81) + %c1_i32_85 = arith.constant 1 : i32 loc(#loc81) + %c1_i32_86 = arith.constant 1 : i32 loc(#loc81) + %16 = arith.bitcast %c0_i32_84 : i32 to i32 loc(#loc81) + %17 = arith.bitcast %c1_i32_85 : i32 to i32 loc(#loc81) + %18 = arith.bitcast %c1_i32_86 : i32 to i32 loc(#loc81) + %19 = ub.poison : i32 loc(#loc81) + scf.for %i_v = %16 to %17 step %18 : i32 { + %o_v = arith.constant 128 : i32 loc(#loc193) + %o_v_87 = arith.constant 128 : i32 loc(#loc193) + %o_v_88 = arith.extsi %i_v : i32 to i64 loc(#loc193) + %o_v_89 = arith.extsi %o_v_87 : i32 to i64 loc(#loc193) + %o_v_90 = arith.muli %o_v_88, %o_v_89 : i64 loc(#loc193) + %o_v_91 = arith.constant 2147483647 : i64 loc(#loc193) + %o_v_92 = arith.constant -2147483648 : i64 loc(#loc193) + %o_v_93 = arith.cmpi sle, %o_v_90, %o_v_91 : i64 loc(#loc193) + %o_v_94 = arith.cmpi sge, %o_v_90, %o_v_92 : i64 loc(#loc193) + %o_v_95 = arith.andi %o_v_93, %o_v_94 : i1 loc(#loc193) + %o_v_96 = arith.muli %i_v, %o_v_87 : i32 loc(#loc193) + %o_v_97 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc194) + %o_v_98 = tt.splat %o_v_96 : i32 -> tensor<128xi32> loc(#loc195) + %o_v_99 = arith.extsi %o_v_98 : tensor<128xi32> to tensor<128xi64> loc(#loc195) + %o_v_100 = arith.extsi %o_v_97 : tensor<128xi32> to tensor<128xi64> loc(#loc195) + %o_v_101 = arith.addi %o_v_99, %o_v_100 : tensor<128xi64> loc(#loc195) + %o_v_102 = arith.constant 2147483647 : i64 loc(#loc195) + %o_v_103 = arith.constant -2147483648 : i64 loc(#loc195) + %o_v_104 = arith.constant dense<2147483647> : tensor<128xi64> loc(#loc195) + %o_v_105 = arith.cmpi sle, %o_v_101, %o_v_104 : tensor<128xi64> loc(#loc195) + %o_v_106 = arith.constant dense<-2147483648> : tensor<128xi64> loc(#loc195) + %o_v_107 = arith.cmpi sge, %o_v_101, %o_v_106 : tensor<128xi64> loc(#loc195) + %o_v_108 = arith.andi %o_v_105, %o_v_107 : tensor<128xi1> loc(#loc195) + %o_v_109 = arith.addi %o_v_98, %o_v_97 : tensor<128xi32> loc(#loc195) + %m_v = arith.constant 128 : i32 loc(#loc196) + %m_v_110 = arith.constant dense<128> : tensor<128xi32> loc(#loc196) + %m_v_111 = arith.cmpi slt, %o_v_109, %m_v_110 : tensor<128xi32> loc(#loc196) + %p_do = tt.expand_dims %o_t_45 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc197) + %p_do_112 = arith.constant 4096 : i32 loc(#loc198) + %p_do_113 = arith.constant 4096 : i64 loc(#loc198) + %p_do_114 = arith.constant dense<4096> : tensor<64x1xi64> loc(#loc198) + %p_do_115 = arith.muli %p_do, %p_do_114 : tensor<64x1xi64> loc(#loc198) + %p_do_116 = tt.splat %do_31 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc199) + %p_do_117 = tt.addptr %p_do_116, %p_do_115 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc199) + %p_do_118 = tt.expand_dims %o_v_109 {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc200) + %p_do_119 = tt.broadcast %p_do_117 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc201) + %p_do_120 = tt.broadcast %p_do_118 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc201) + %p_do_121 = tt.addptr %p_do_119, %p_do_120 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc201) + %p_dv = tt.expand_dims %o_t_45 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc202) + %p_dv_122 = arith.constant 4096 : i32 loc(#loc203) + %p_dv_123 = arith.constant 4096 : i64 loc(#loc203) + %p_dv_124 = arith.constant dense<4096> : tensor<64x1xi64> loc(#loc203) + %p_dv_125 = arith.muli %p_dv, %p_dv_124 : tensor<64x1xi64> loc(#loc203) + %p_dv_126 = tt.splat %dv_39 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc204) + %p_dv_127 = tt.addptr %p_dv_126, %p_dv_125 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc204) + %p_dv_128 = tt.expand_dims %o_v_109 {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc205) + %p_dv_129 = tt.broadcast %p_dv_127 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc206) + %p_dv_130 = tt.broadcast %p_dv_128 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc206) + %p_dv_131 = tt.addptr %p_dv_129, %p_dv_130 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc206) + %b_do = tt.expand_dims %m_t_47 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc207) + %b_do_132 = tt.expand_dims %m_v_111 {axis = 0 : i32} : tensor<128xi1> -> tensor<1x128xi1> loc(#loc208) + %b_do_133 = tt.broadcast %b_do : tensor<64x1xi1> -> tensor<64x128xi1> loc(#loc209) + %b_do_134 = tt.broadcast %b_do_132 : tensor<1x128xi1> -> tensor<64x128xi1> loc(#loc209) + %b_do_135 = arith.andi %b_do_133, %b_do_134 : tensor<64x128xi1> loc(#loc209) + %b_do_136 = arith.constant 0.000000e+00 : f32 loc(#loc210) + %b_do_137 = arith.constant dense<0.000000e+00> : tensor<64x128xf32> loc(#loc210) + %b_do_138 = arith.truncf %b_do_137 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc210) + %b_do_139 = tt.load %p_do_121, %b_do_135, %b_do_138 : tensor<64x128x!tt.ptr> loc(#loc210) + %b_dv = arith.constant 0.000000e+00 : f32 loc(#loc211) + %b_dv_140 = arith.constant dense<0.000000e+00> : tensor<64x128xf32> loc(#loc211) + %b_dv_141 = tt.dot %b_A_83, %b_do_139, %b_dv_140, inputPrecision = tf32 : tensor<64x64xbf16> * tensor<64x128xbf16> -> tensor<64x128xf32> loc(#loc211) + %20 = tt.expand_dims %m_t_47 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc101) + %21 = tt.expand_dims %m_v_111 {axis = 0 : i32} : tensor<128xi1> -> tensor<1x128xi1> loc(#loc102) + %22 = tt.broadcast %20 : tensor<64x1xi1> -> tensor<64x128xi1> loc(#loc103) + %23 = tt.broadcast %21 : tensor<1x128xi1> -> tensor<64x128xi1> loc(#loc103) + %24 = arith.andi %22, %23 : tensor<64x128xi1> loc(#loc103) + %25 = arith.truncf %b_dv_141 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc104) + tt.store %p_dv_131, %25, %24 : tensor<64x128x!tt.ptr> loc(#loc105) + } loc(#loc81) + tt.return loc(#loc106) + } loc(#loc) + tt.func private @triton.language.standard.zeros__Tc64_c64T_cfp32() -> tensor<64x64xf32> attributes {noinline = false} { + %cst = arith.constant 0.000000e+00 : f32 loc(#loc108) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc108) + tt.return %cst_0 : tensor<64x64xf32> loc(#loc109) + ^bb1: // no predecessors + %0 = ub.poison : tensor<64x64xf32> loc(#loc110) + tt.return %0 : tensor<64x64xf32> loc(#loc110) + } loc(#loc107) + tt.func private @triton.language.standard.cdiv__c128_c128() attributes {noinline = false} { + tt.return loc(#loc112) + ^bb1: // no predecessors + tt.return loc(#loc113) + } loc(#loc111) + tt.func private @fla.ops.utils.op.exp2__fp32S64_64S(%x: tensor<64x64xf32> loc("x"(#loc114))) -> tensor<64x64xf32> attributes {noinline = false} { + %0 = math.exp2 %x : tensor<64x64xf32> loc(#loc115) + tt.return %0 : tensor<64x64xf32> loc(#loc116) + ^bb1: // no predecessors + %1 = ub.poison : tensor<64x64xf32> loc(#loc117) + tt.return %1 : tensor<64x64xf32> loc(#loc117) + } loc(#loc114) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":490:30) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":490:36) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":490:61) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":490:67) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":491:23) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":491:34) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":497:25) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":497:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":497:38) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":499:16) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":499:28) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":499:20) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":499:40) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":499:9) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":500:16) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":500:28) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":500:20) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":500:40) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":500:9) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":501:17) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":501:22) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":501:29) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":501:10) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":502:17) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":502:22) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":502:29) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":502:10) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":504:16) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":504:34) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":504:21) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":505:16) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":511:23) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":511:28) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":511:17) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":512:28) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":512:22) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":513:26) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":518:23) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":519:36) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":519:25) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":520:24) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":520:42) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":520:29) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":521:24) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":522:26) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":522:38) +#loc47 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":522:22) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":522:49) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":522:45) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":523:26) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":523:22) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":523:41) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":523:53) +#loc54 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":523:37) +#loc55 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":525:40) +#loc56 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":525:55) +#loc57 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":525:51) +#loc58 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":525:26) +#loc59 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":526:40) +#loc60 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":526:55) +#loc61 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":526:51) +#loc62 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":526:26) +#loc63 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":527:31) +#loc64 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":527:38) +#loc65 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":527:19) +#loc66 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":527:12) +#loc67 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":529:28) +#loc68 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":529:43) +#loc69 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":529:39) +#loc70 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":529:24) +#loc71 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":529:19) +#loc72 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":530:15) +#loc73 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":530:31) +#loc74 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":530:27) +#loc75 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":530:48) +#loc76 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":530:59) +#loc77 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":530:44) +#loc78 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":531:29) +#loc79 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":531:35) +#loc80 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":533:32) +#loc81 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":533:21) +#loc82 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":534:20) +#loc83 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":534:38) +#loc84 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":534:25) +#loc85 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":535:20) +#loc86 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":536:24) +#loc87 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":536:36) +#loc88 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":536:20) +#loc89 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":536:48) +#loc90 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":536:44) +#loc91 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":537:24) +#loc92 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":537:36) +#loc93 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":537:20) +#loc94 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":537:48) +#loc95 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":537:44) +#loc96 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":538:38) +#loc97 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":538:53) +#loc98 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":538:49) +#loc99 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":538:23) +#loc100 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":539:42) +#loc101 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":540:64) +#loc102 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":540:79) +#loc103 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":540:75) +#loc104 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":540:31) +#loc105 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":540:23) +#loc106 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":533:4) +#loc107 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":120:0) +#loc108 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":129:31) +#loc109 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":129:11) +#loc110 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":129:4) +#loc111 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":34:0) +#loc112 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":43:11) +#loc113 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":43:4) +#loc115 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:33) +#loc116 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:20) +#loc117 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:13) +#loc125 = loc("q"(#loc10)) +#loc126 = loc("q"(#loc11)) +#loc127 = loc("q"(#loc12)) +#loc128 = loc("q"(#loc13)) +#loc129 = loc("q"(#loc14)) +#loc130 = loc("k"(#loc15)) +#loc131 = loc("k"(#loc16)) +#loc132 = loc("k"(#loc17)) +#loc133 = loc("k"(#loc18)) +#loc134 = loc("k"(#loc19)) +#loc135 = loc("do"(#loc20)) +#loc136 = loc("do"(#loc21)) +#loc137 = loc("do"(#loc22)) +#loc138 = loc("do"(#loc23)) +#loc139 = loc("dv"(#loc24)) +#loc140 = loc("dv"(#loc25)) +#loc141 = loc("dv"(#loc26)) +#loc142 = loc("dv"(#loc27)) +#loc143 = loc("o_t"(#loc28)) +#loc144 = loc("o_t"(#loc29)) +#loc145 = loc("o_t"(#loc30)) +#loc146 = loc("m_t"(#loc31)) +#loc147 = loc("g"(#loc32)) +#loc148 = loc("g"(#loc33)) +#loc149 = loc("g"(#loc34)) +#loc150 = loc("p_g"(#loc35)) +#loc151 = loc("p_g"(#loc36)) +#loc152 = loc("b_g"(#loc37)) +#loc153 = loc("b_A"(#loc38)) +#loc154 = loc("b_A"(#loc40)) +#loc155 = loc("o_k"(#loc41)) +#loc156 = loc("o_k"(#loc42)) +#loc157 = loc("o_k"(#loc43)) +#loc158 = loc("m_k"(#loc44)) +#loc159 = loc("p_k"(#loc45)) +#loc160 = loc("p_k"(#loc46)) +#loc161 = loc("p_k"(#loc47)) +#loc162 = loc("p_k"(#loc48)) +#loc163 = loc("p_k"(#loc49)) +#loc164 = loc("p_q"(#loc50)) +#loc165 = loc("p_q"(#loc51)) +#loc166 = loc("p_q"(#loc52)) +#loc167 = loc("p_q"(#loc53)) +#loc168 = loc("p_q"(#loc54)) +#loc169 = loc("b_k"(#loc55)) +#loc170 = loc("b_k"(#loc56)) +#loc171 = loc("b_k"(#loc57)) +#loc172 = loc("b_k"(#loc58)) +#loc173 = loc("b_q"(#loc59)) +#loc174 = loc("b_q"(#loc60)) +#loc175 = loc("b_q"(#loc61)) +#loc176 = loc("b_q"(#loc62)) +#loc177 = loc("b_A"(#loc63)) +#loc178 = loc("b_A"(#loc64)) +#loc179 = loc("b_A"(#loc65)) +#loc180 = loc("b_A"(#loc67)) +#loc181 = loc("b_A"(#loc68)) +#loc182 = loc("b_A"(#loc69)) +#loc183 = loc("b_A"(#loc70)) +#loc184 = loc("b_A"(#loc71)) +#loc185 = loc("m_A"(#loc72)) +#loc186 = loc("m_A"(#loc73)) +#loc187 = loc("m_A"(#loc74)) +#loc188 = loc("m_A"(#loc75)) +#loc189 = loc("m_A"(#loc76)) +#loc190 = loc("m_A"(#loc77)) +#loc191 = loc("b_A"(#loc78)) +#loc192 = loc("b_A"(#loc79)) +#loc193 = loc("o_v"(#loc82)) +#loc194 = loc("o_v"(#loc83)) +#loc195 = loc("o_v"(#loc84)) +#loc196 = loc("m_v"(#loc85)) +#loc197 = loc("p_do"(#loc86)) +#loc198 = loc("p_do"(#loc87)) +#loc199 = loc("p_do"(#loc88)) +#loc200 = loc("p_do"(#loc89)) +#loc201 = loc("p_do"(#loc90)) +#loc202 = loc("p_dv"(#loc91)) +#loc203 = loc("p_dv"(#loc92)) +#loc204 = loc("p_dv"(#loc93)) +#loc205 = loc("p_dv"(#loc94)) +#loc206 = loc("p_dv"(#loc95)) +#loc207 = loc("b_do"(#loc96)) +#loc208 = loc("b_do"(#loc97)) +#loc209 = loc("b_do"(#loc98)) +#loc210 = loc("b_do"(#loc99)) +#loc211 = loc("b_dv"(#loc100)) diff --git a/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.ttgir b/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..b2e0c725e5596979173a3c27165ad7e90bfbfa82 --- /dev/null +++ b/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.ttgir @@ -0,0 +1,284 @@ +#blocked = #ttg.blocked<{sizePerThread = [8, 1], threadsPerWarp = [16, 2], warpsPerCTA = [1, 2], order = [0, 1]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [2, 1], order = [1, 0]}> +#blocked2 = #ttg.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [2], order = [0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":466:0) +#mma = #ttg.nvidia_mma<{versionMajor = 2, versionMinor = 0, warpsPerCTA = [1, 2], instrShape = [16, 8]}> +#shared = #ttg.swizzled_shared<{vec = 8, perPhase = 1, maxPhase = 8, order = [1, 0]}> +#shared1 = #ttg.swizzled_shared<{vec = 8, perPhase = 1, maxPhase = 8, order = [0, 1]}> +#smem = #ttg.shared_memory +#loc67 = loc("q"(#loc)) +#loc68 = loc("k"(#loc)) +#loc69 = loc("g"(#loc)) +#loc70 = loc("do"(#loc)) +#loc71 = loc("dv"(#loc)) +#loc72 = loc("scale"(#loc)) +#loc73 = loc("T"(#loc)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 2 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @chunk_bwd_kernel_dv_local(%q: !tt.ptr {tt.divisibility = 16 : i32} loc("q"(#loc)), %k: !tt.ptr {tt.divisibility = 16 : i32} loc("k"(#loc)), %g: !tt.ptr {tt.divisibility = 16 : i32} loc("g"(#loc)), %do: !tt.ptr {tt.divisibility = 16 : i32} loc("do"(#loc)), %dv: !tt.ptr {tt.divisibility = 16 : i32} loc("dv"(#loc)), %scale: f32 loc("scale"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<0.000000e+00> : tensor<64x128xf32, #mma> loc(#loc1) + %cst_0 = arith.constant dense<4096> : tensor<1x64xi64, #blocked> loc(#loc1) + %cst_1 = arith.constant dense<4096> : tensor<64x1xi64, #blocked1> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<128xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_3 = arith.constant dense<128> : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc1) + %cst_4 = arith.constant dense<0.000000e+00> : tensor<64xf32, #blocked2> loc(#loc1) + %cst_5 = arith.constant dense<32> : tensor<64xi64, #blocked2> loc(#loc1) + %c64_i64 = arith.constant 64 : i64 loc(#loc1) + %c128_i64 = arith.constant 128 : i64 loc(#loc1) + %c32_i64 = arith.constant 32 : i64 loc(#loc1) + %cst_6 = arith.constant dense<0.000000e+00> : tensor<64x128xbf16, #blocked1> loc(#loc1) + %cst_7 = arith.constant dense<0.000000e+00> : tensor<128x64xbf16, #blocked> loc(#loc1) + %cst_8 = arith.constant dense<0.000000e+00> : tensor<64x64xf32, #mma> loc(#loc1) + %0 = tt.get_program_id x : i32 loc(#loc2) + %1 = arith.extsi %0 : i32 to i64 loc(#loc3) + %2 = tt.get_program_id y : i32 loc(#loc4) + %3 = arith.extsi %2 : i32 to i64 loc(#loc5) + %4 = arith.divsi %3, %c32_i64 : i64 loc(#loc6) + %5 = arith.remsi %3, %c32_i64 : i64 loc(#loc7) + %6 = arith.extsi %T : i32 to i64 loc(#loc8) + %7 = arith.muli %4, %6 : i64 loc(#loc8) + %q_9 = arith.muli %7, %c32_i64 : i64 loc(#loc74) + %q_10 = arith.addi %q_9, %5 : i64 loc(#loc75) + %q_11 = arith.muli %q_10, %c128_i64 : i64 loc(#loc76) + %q_12 = tt.addptr %q, %q_11 : !tt.ptr, i64 loc(#loc77) + %k_13 = tt.addptr %k, %q_11 : !tt.ptr, i64 loc(#loc78) + %do_14 = tt.addptr %do, %q_11 : !tt.ptr, i64 loc(#loc79) + %dv_15 = tt.addptr %dv, %q_11 : !tt.ptr, i64 loc(#loc80) + %o_t = arith.muli %1, %c64_i64 : i64 loc(#loc81) + %o_t_16 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc82) + %o_t_17 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc82) + %o_t_18 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc82) + %o_t_19 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc82) + %o_t_20 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #blocked2> loc(#loc82) + %o_t_21 = arith.extsi %o_t_16 : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> to tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc83) + %o_t_22 = arith.extsi %o_t_17 : tensor<64xi32, #ttg.slice<{dim = 1, parent = #mma}>> to tensor<64xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc83) + %o_t_23 = arith.extsi %o_t_18 : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked}>> to tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc83) + %o_t_24 = arith.extsi %o_t_19 : tensor<64xi32, #ttg.slice<{dim = 0, parent = #mma}>> to tensor<64xi64, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc83) + %o_t_25 = arith.extsi %o_t_20 : tensor<64xi32, #blocked2> to tensor<64xi64, #blocked2> loc(#loc83) + %o_t_26 = tt.splat %o_t : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc83) + %o_t_27 = tt.splat %o_t : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc83) + %o_t_28 = tt.splat %o_t : i64 -> tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc83) + %o_t_29 = tt.splat %o_t : i64 -> tensor<64xi64, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc83) + %o_t_30 = tt.splat %o_t : i64 -> tensor<64xi64, #blocked2> loc(#loc83) + %o_t_31 = arith.addi %o_t_26, %o_t_21 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc83) + %o_t_32 = arith.addi %o_t_27, %o_t_22 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc83) + %o_t_33 = arith.addi %o_t_28, %o_t_23 : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc83) + %o_t_34 = arith.addi %o_t_29, %o_t_24 : tensor<64xi64, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc83) + %o_t_35 = arith.addi %o_t_30, %o_t_25 : tensor<64xi64, #blocked2> loc(#loc83) + %m_t = tt.splat %6 : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc84) + %m_t_36 = tt.splat %6 : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc84) + %m_t_37 = tt.splat %6 : i64 -> tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc84) + %m_t_38 = tt.splat %6 : i64 -> tensor<64xi64, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc84) + %m_t_39 = tt.splat %6 : i64 -> tensor<64xi64, #blocked2> loc(#loc84) + %m_t_40 = arith.cmpi slt, %o_t_31, %m_t : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc84) + %m_t_41 = arith.cmpi slt, %o_t_32, %m_t_36 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc84) + %m_t_42 = arith.cmpi slt, %o_t_33, %m_t_37 : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc84) + %m_t_43 = arith.cmpi slt, %o_t_34, %m_t_38 : tensor<64xi64, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc84) + %m_t_44 = arith.cmpi slt, %o_t_35, %m_t_39 : tensor<64xi64, #blocked2> loc(#loc84) + %g_45 = tt.addptr %g, %q_10 : !tt.ptr, i64 loc(#loc85) + %p_g = arith.muli %o_t_35, %cst_5 : tensor<64xi64, #blocked2> loc(#loc86) + %p_g_46 = tt.splat %g_45 : !tt.ptr -> tensor<64x!tt.ptr, #blocked2> loc(#loc87) + %p_g_47 = tt.addptr %p_g_46, %p_g : tensor<64x!tt.ptr, #blocked2>, tensor<64xi64, #blocked2> loc(#loc87) + %b_g = tt.load %p_g_47, %m_t_44, %cst_4 : tensor<64x!tt.ptr, #blocked2> loc(#loc88) + %b_A = ttg.convert_layout %b_g : tensor<64xf32, #blocked2> -> tensor<64xf32, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc89) + %b_A_48 = ttg.convert_layout %b_g : tensor<64xf32, #blocked2> -> tensor<64xf32, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc128) + %o_k = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc91) + %o_k_49 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc91) + %m_k = arith.cmpi slt, %o_k, %cst_3 : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc92) + %m_k_50 = arith.cmpi slt, %o_k_49, %cst_2 : tensor<128xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc92) + %p_k = tt.expand_dims %o_t_31 {axis = 1 : i32} : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<64x1xi64, #blocked1> loc(#loc93) + %p_k_51 = tt.expand_dims %o_t_32 {axis = 1 : i32} : tensor<64xi64, #ttg.slice<{dim = 1, parent = #mma}>> -> tensor<64x1xi64, #mma> loc(#loc93) + %p_k_52 = arith.muli %p_k, %cst_1 : tensor<64x1xi64, #blocked1> loc(#loc94) + %p_k_53 = tt.splat %k_13 : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked1> loc(#loc95) + %p_k_54 = tt.addptr %p_k_53, %p_k_52 : tensor<64x1x!tt.ptr, #blocked1>, tensor<64x1xi64, #blocked1> loc(#loc95) + %p_k_55 = tt.expand_dims %o_k {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x128xi32, #blocked1> loc(#loc96) + %p_k_56 = tt.broadcast %p_k_54 : tensor<64x1x!tt.ptr, #blocked1> -> tensor<64x128x!tt.ptr, #blocked1> loc(#loc97) + %p_k_57 = tt.broadcast %p_k_55 : tensor<1x128xi32, #blocked1> -> tensor<64x128xi32, #blocked1> loc(#loc97) + %p_k_58 = tt.addptr %p_k_56, %p_k_57 : tensor<64x128x!tt.ptr, #blocked1>, tensor<64x128xi32, #blocked1> loc(#loc97) + %p_q = tt.expand_dims %o_k_49 {axis = 1 : i32} : tensor<128xi32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128x1xi32, #blocked> loc(#loc98) + %p_q_59 = tt.splat %q_12 : !tt.ptr -> tensor<128x1x!tt.ptr, #blocked> loc(#loc99) + %p_q_60 = tt.addptr %p_q_59, %p_q : tensor<128x1x!tt.ptr, #blocked>, tensor<128x1xi32, #blocked> loc(#loc99) + %p_q_61 = tt.expand_dims %o_t_33 {axis = 0 : i32} : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x64xi64, #blocked> loc(#loc100) + %p_q_62 = tt.expand_dims %o_t_34 {axis = 0 : i32} : tensor<64xi64, #ttg.slice<{dim = 0, parent = #mma}>> -> tensor<1x64xi64, #mma> loc(#loc100) + %p_q_63 = arith.muli %p_q_61, %cst_0 : tensor<1x64xi64, #blocked> loc(#loc101) + %p_q_64 = tt.broadcast %p_q_60 : tensor<128x1x!tt.ptr, #blocked> -> tensor<128x64x!tt.ptr, #blocked> loc(#loc102) + %p_q_65 = tt.broadcast %p_q_63 : tensor<1x64xi64, #blocked> -> tensor<128x64xi64, #blocked> loc(#loc102) + %p_q_66 = tt.addptr %p_q_64, %p_q_65 : tensor<128x64x!tt.ptr, #blocked>, tensor<128x64xi64, #blocked> loc(#loc102) + %b_k = tt.expand_dims %m_t_40 {axis = 1 : i32} : tensor<64xi1, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<64x1xi1, #blocked1> loc(#loc103) + %b_k_67 = tt.expand_dims %m_t_41 {axis = 1 : i32} : tensor<64xi1, #ttg.slice<{dim = 1, parent = #mma}>> -> tensor<64x1xi1, #mma> loc(#loc103) + %b_k_68 = tt.expand_dims %m_k {axis = 0 : i32} : tensor<128xi1, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x128xi1, #blocked1> loc(#loc104) + %b_k_69 = tt.broadcast %b_k : tensor<64x1xi1, #blocked1> -> tensor<64x128xi1, #blocked1> loc(#loc105) + %b_k_70 = tt.broadcast %b_k_68 : tensor<1x128xi1, #blocked1> -> tensor<64x128xi1, #blocked1> loc(#loc105) + %b_k_71 = arith.andi %b_k_69, %b_k_70 : tensor<64x128xi1, #blocked1> loc(#loc105) + %b_k_72 = tt.load %p_k_58, %b_k_71, %cst_6 : tensor<64x128x!tt.ptr, #blocked1> loc(#loc106) + %b_k_73 = ttg.local_alloc %b_k_72 : (tensor<64x128xbf16, #blocked1>) -> !ttg.memdesc<64x128xbf16, #shared, #smem> loc(#loc106) + %b_q = tt.expand_dims %m_k_50 {axis = 1 : i32} : tensor<128xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128x1xi1, #blocked> loc(#loc107) + %b_q_74 = tt.expand_dims %m_t_42 {axis = 0 : i32} : tensor<64xi1, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x64xi1, #blocked> loc(#loc108) + %b_q_75 = tt.expand_dims %m_t_43 {axis = 0 : i32} : tensor<64xi1, #ttg.slice<{dim = 0, parent = #mma}>> -> tensor<1x64xi1, #mma> loc(#loc108) + %b_q_76 = tt.broadcast %b_q : tensor<128x1xi1, #blocked> -> tensor<128x64xi1, #blocked> loc(#loc109) + %b_q_77 = tt.broadcast %b_q_74 : tensor<1x64xi1, #blocked> -> tensor<128x64xi1, #blocked> loc(#loc109) + %b_q_78 = arith.andi %b_q_76, %b_q_77 : tensor<128x64xi1, #blocked> loc(#loc109) + %b_q_79 = tt.load %p_q_66, %b_q_78, %cst_7 : tensor<128x64x!tt.ptr, #blocked> loc(#loc110) + %b_q_80 = ttg.local_alloc %b_q_79 : (tensor<128x64xbf16, #blocked>) -> !ttg.memdesc<128x64xbf16, #shared1, #smem> loc(#loc110) + %b_k_81 = ttg.local_load %b_k_73 : !ttg.memdesc<64x128xbf16, #shared, #smem> -> tensor<64x128xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma, kWidth = 2}>> loc(#loc106) + %b_q_82 = ttg.local_load %b_q_80 : !ttg.memdesc<128x64xbf16, #shared1, #smem> -> tensor<128x64xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> loc(#loc110) + %b_A_83 = tt.dot %b_k_81, %b_q_82, %cst_8, inputPrecision = tf32 : tensor<64x128xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma, kWidth = 2}>> * tensor<128x64xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> -> tensor<64x64xf32, #mma> loc(#loc111) + %b_A_84 = tt.splat %scale : f32 -> tensor<64x64xf32, #mma> loc(#loc112) + %b_A_85 = arith.mulf %b_A_83, %b_A_84 : tensor<64x64xf32, #mma> loc(#loc112) + %b_A_86 = arith.addf %b_A_85, %cst_8 : tensor<64x64xf32, #mma> loc(#loc113) + %b_A_87 = tt.expand_dims %b_A_48 {axis = 0 : i32} : tensor<64xf32, #ttg.slice<{dim = 0, parent = #mma}>> -> tensor<1x64xf32, #mma> loc(#loc114) + %b_A_88 = tt.expand_dims %b_A {axis = 1 : i32} : tensor<64xf32, #ttg.slice<{dim = 1, parent = #mma}>> -> tensor<64x1xf32, #mma> loc(#loc115) + %b_A_89 = tt.broadcast %b_A_87 : tensor<1x64xf32, #mma> -> tensor<64x64xf32, #mma> loc(#loc89) + %b_A_90 = tt.broadcast %b_A_88 : tensor<64x1xf32, #mma> -> tensor<64x64xf32, #mma> loc(#loc89) + %b_A_91 = arith.subf %b_A_89, %b_A_90 : tensor<64x64xf32, #mma> loc(#loc89) + %b_A_92 = math.exp2 %b_A_91 : tensor<64x64xf32, #mma> loc(#loc128) + %b_A_93 = arith.mulf %b_A_86, %b_A_92 : tensor<64x64xf32, #mma> loc(#loc116) + %m_A = tt.broadcast %p_k_51 : tensor<64x1xi64, #mma> -> tensor<64x64xi64, #mma> loc(#loc117) + %m_A_94 = tt.broadcast %p_q_62 : tensor<1x64xi64, #mma> -> tensor<64x64xi64, #mma> loc(#loc117) + %m_A_95 = arith.cmpi sle, %m_A, %m_A_94 : tensor<64x64xi64, #mma> loc(#loc117) + %m_A_96 = tt.broadcast %b_k_67 : tensor<64x1xi1, #mma> -> tensor<64x64xi1, #mma> loc(#loc118) + %m_A_97 = tt.broadcast %b_q_75 : tensor<1x64xi1, #mma> -> tensor<64x64xi1, #mma> loc(#loc118) + %m_A_98 = arith.andi %m_A_96, %m_A_97 : tensor<64x64xi1, #mma> loc(#loc118) + %m_A_99 = arith.andi %m_A_95, %m_A_98 : tensor<64x64xi1, #mma> loc(#loc119) + %b_A_100 = arith.select %m_A_99, %b_A_93, %cst_8 : tensor<64x64xi1, #mma>, tensor<64x64xf32, #mma> loc(#loc120) + %b_A_101 = arith.truncf %b_A_100 : tensor<64x64xf32, #mma> to tensor<64x64xbf16, #mma> loc(#loc121) + %b_A_102 = ttg.local_alloc %b_A_101 : (tensor<64x64xbf16, #mma>) -> !ttg.memdesc<64x64xbf16, #shared, #smem> loc(#loc121) + %p_do = tt.splat %do_14 : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked1> loc(#loc122) + %p_do_103 = tt.addptr %p_do, %p_k_52 : tensor<64x1x!tt.ptr, #blocked1>, tensor<64x1xi64, #blocked1> loc(#loc122) + %p_do_104 = tt.broadcast %p_do_103 : tensor<64x1x!tt.ptr, #blocked1> -> tensor<64x128x!tt.ptr, #blocked1> loc(#loc123) + %p_do_105 = tt.addptr %p_do_104, %p_k_57 : tensor<64x128x!tt.ptr, #blocked1>, tensor<64x128xi32, #blocked1> loc(#loc123) + %p_dv = tt.splat %dv_15 : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked1> loc(#loc124) + %p_dv_106 = tt.addptr %p_dv, %p_k_52 : tensor<64x1x!tt.ptr, #blocked1>, tensor<64x1xi64, #blocked1> loc(#loc124) + %p_dv_107 = tt.broadcast %p_dv_106 : tensor<64x1x!tt.ptr, #blocked1> -> tensor<64x128x!tt.ptr, #blocked1> loc(#loc125) + %p_dv_108 = tt.addptr %p_dv_107, %p_k_57 : tensor<64x128x!tt.ptr, #blocked1>, tensor<64x128xi32, #blocked1> loc(#loc125) + %b_do = tt.load %p_do_105, %b_k_71, %cst_6 : tensor<64x128x!tt.ptr, #blocked1> loc(#loc126) + %b_do_109 = ttg.local_alloc %b_do : (tensor<64x128xbf16, #blocked1>) -> !ttg.memdesc<64x128xbf16, #shared, #smem> loc(#loc126) + %b_A_110 = ttg.local_load %b_A_102 : !ttg.memdesc<64x64xbf16, #shared, #smem> -> tensor<64x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma, kWidth = 2}>> loc(#loc121) + %b_do_111 = ttg.local_load %b_do_109 : !ttg.memdesc<64x128xbf16, #shared, #smem> -> tensor<64x128xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> loc(#loc126) + %b_dv = tt.dot %b_A_110, %b_do_111, %cst, inputPrecision = tf32 : tensor<64x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma, kWidth = 2}>> * tensor<64x128xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> -> tensor<64x128xf32, #mma> loc(#loc127) + %8 = arith.truncf %b_dv : tensor<64x128xf32, #mma> to tensor<64x128xbf16, #mma> loc(#loc64) + %9 = ttg.convert_layout %8 : tensor<64x128xbf16, #mma> -> tensor<64x128xbf16, #blocked1> loc(#loc65) + tt.store %p_dv_108, %9, %b_k_71 : tensor<64x128x!tt.ptr, #blocked1> loc(#loc65) + tt.return loc(#loc66) + } loc(#loc) +} loc(#loc) +#loc1 = loc(unknown) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":490:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":490:36) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":490:61) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":490:67) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":491:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":491:34) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":497:25) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":499:16) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":499:20) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":499:40) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":499:9) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":500:9) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":501:10) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":502:10) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":504:16) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":504:34) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":504:21) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":505:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":511:17) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":512:28) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":512:22) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":513:26) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":529:39) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:33) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":529:24) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":520:42) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":521:24) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":522:26) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":522:38) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":522:22) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":522:49) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":522:45) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":523:26) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":523:22) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":523:41) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":523:53) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":523:37) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":525:40) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":525:55) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":525:51) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":525:26) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":526:40) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":526:55) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":526:51) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":526:26) +#loc47 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":527:31) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":527:38) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":527:19) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":529:28) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":529:43) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":529:19) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":530:27) +#loc54 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":530:59) +#loc55 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":530:44) +#loc56 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":531:29) +#loc57 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":531:35) +#loc58 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":536:20) +#loc59 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":536:44) +#loc60 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":537:20) +#loc61 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":537:44) +#loc62 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":538:23) +#loc63 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":539:42) +#loc64 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":540:31) +#loc65 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":540:23) +#loc66 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":533:4) +#loc74 = loc("q"(#loc9)) +#loc75 = loc("q"(#loc10)) +#loc76 = loc("q"(#loc11)) +#loc77 = loc("q"(#loc12)) +#loc78 = loc("k"(#loc13)) +#loc79 = loc("do"(#loc14)) +#loc80 = loc("dv"(#loc15)) +#loc81 = loc("o_t"(#loc16)) +#loc82 = loc("o_t"(#loc17)) +#loc83 = loc("o_t"(#loc18)) +#loc84 = loc("m_t"(#loc19)) +#loc85 = loc("g"(#loc20)) +#loc86 = loc("p_g"(#loc21)) +#loc87 = loc("p_g"(#loc22)) +#loc88 = loc("b_g"(#loc23)) +#loc89 = loc("b_A"(#loc24)) +#loc90 = loc("b_A"(#loc26)) +#loc91 = loc("o_k"(#loc27)) +#loc92 = loc("m_k"(#loc28)) +#loc93 = loc("p_k"(#loc29)) +#loc94 = loc("p_k"(#loc30)) +#loc95 = loc("p_k"(#loc31)) +#loc96 = loc("p_k"(#loc32)) +#loc97 = loc("p_k"(#loc33)) +#loc98 = loc("p_q"(#loc34)) +#loc99 = loc("p_q"(#loc35)) +#loc100 = loc("p_q"(#loc36)) +#loc101 = loc("p_q"(#loc37)) +#loc102 = loc("p_q"(#loc38)) +#loc103 = loc("b_k"(#loc39)) +#loc104 = loc("b_k"(#loc40)) +#loc105 = loc("b_k"(#loc41)) +#loc106 = loc("b_k"(#loc42)) +#loc107 = loc("b_q"(#loc43)) +#loc108 = loc("b_q"(#loc44)) +#loc109 = loc("b_q"(#loc45)) +#loc110 = loc("b_q"(#loc46)) +#loc111 = loc("b_A"(#loc47)) +#loc112 = loc("b_A"(#loc48)) +#loc113 = loc("b_A"(#loc49)) +#loc114 = loc("b_A"(#loc50)) +#loc115 = loc("b_A"(#loc51)) +#loc116 = loc("b_A"(#loc52)) +#loc117 = loc("m_A"(#loc53)) +#loc118 = loc("m_A"(#loc54)) +#loc119 = loc("m_A"(#loc55)) +#loc120 = loc("b_A"(#loc56)) +#loc121 = loc("b_A"(#loc57)) +#loc122 = loc("p_do"(#loc58)) +#loc123 = loc("p_do"(#loc59)) +#loc124 = loc("p_dv"(#loc60)) +#loc125 = loc("p_dv"(#loc61)) +#loc126 = loc("b_do"(#loc62)) +#loc127 = loc("b_dv"(#loc63)) +#loc128 = loc(callsite(#loc25 at #loc90)) diff --git a/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.ttir b/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.ttir new file mode 100644 index 0000000000000000000000000000000000000000..10884b7a82587a348aea4a3a3be4bfca19a7180b --- /dev/null +++ b/triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.ttir @@ -0,0 +1,235 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":466:0) +#loc67 = loc("q"(#loc)) +#loc68 = loc("k"(#loc)) +#loc69 = loc("g"(#loc)) +#loc70 = loc("do"(#loc)) +#loc71 = loc("dv"(#loc)) +#loc72 = loc("scale"(#loc)) +#loc73 = loc("T"(#loc)) +module { + tt.func public @chunk_bwd_kernel_dv_local(%q: !tt.ptr {tt.divisibility = 16 : i32} loc("q"(#loc)), %k: !tt.ptr {tt.divisibility = 16 : i32} loc("k"(#loc)), %g: !tt.ptr {tt.divisibility = 16 : i32} loc("g"(#loc)), %do: !tt.ptr {tt.divisibility = 16 : i32} loc("do"(#loc)), %dv: !tt.ptr {tt.divisibility = 16 : i32} loc("dv"(#loc)), %scale: f32 loc("scale"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %b_q = arith.constant dense<0.000000e+00> : tensor<128x64xbf16> loc(#loc74) + %cst = arith.constant dense<0.000000e+00> : tensor<64x128xbf16> loc(#loc2) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc2) + %cst_1 = arith.constant dense<0.000000e+00> : tensor<64x128xf32> loc(#loc2) + %cst_2 = arith.constant dense<4096> : tensor<1x64xi64> loc(#loc2) + %cst_3 = arith.constant dense<4096> : tensor<64x1xi64> loc(#loc2) + %cst_4 = arith.constant dense<128> : tensor<128xi32> loc(#loc2) + %b_g = arith.constant dense<0.000000e+00> : tensor<64xf32> loc(#loc75) + %p_g = arith.constant dense<32> : tensor<64xi64> loc(#loc76) + %o_t = arith.constant 64 : i64 loc(#loc77) + %c128_i64 = arith.constant 128 : i64 loc(#loc2) + %c32_i64 = arith.constant 32 : i64 loc(#loc2) + %0 = tt.get_program_id x : i32 loc(#loc6) + %1 = arith.extsi %0 : i32 to i64 loc(#loc7) + %2 = tt.get_program_id y : i32 loc(#loc8) + %3 = arith.extsi %2 : i32 to i64 loc(#loc9) + %4 = arith.divsi %3, %c32_i64 : i64 loc(#loc10) + %5 = arith.remsi %3, %c32_i64 : i64 loc(#loc11) + %6 = arith.extsi %T : i32 to i64 loc(#loc12) + %7 = arith.muli %4, %6 : i64 loc(#loc12) + %q_5 = arith.muli %7, %c32_i64 : i64 loc(#loc78) + %q_6 = arith.addi %q_5, %5 : i64 loc(#loc79) + %q_7 = arith.muli %q_6, %c128_i64 : i64 loc(#loc80) + %q_8 = tt.addptr %q, %q_7 : !tt.ptr, i64 loc(#loc81) + %k_9 = tt.addptr %k, %q_7 : !tt.ptr, i64 loc(#loc82) + %do_10 = tt.addptr %do, %q_7 : !tt.ptr, i64 loc(#loc83) + %dv_11 = tt.addptr %dv, %q_7 : !tt.ptr, i64 loc(#loc84) + %o_t_12 = arith.muli %1, %o_t : i64 loc(#loc77) + %o_t_13 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc85) + %o_t_14 = arith.extsi %o_t_13 : tensor<64xi32> to tensor<64xi64> loc(#loc86) + %o_t_15 = tt.splat %o_t_12 : i64 -> tensor<64xi64> loc(#loc86) + %o_t_16 = arith.addi %o_t_15, %o_t_14 : tensor<64xi64> loc(#loc86) + %m_t = tt.splat %6 : i64 -> tensor<64xi64> loc(#loc87) + %m_t_17 = arith.cmpi slt, %o_t_16, %m_t : tensor<64xi64> loc(#loc87) + %g_18 = tt.addptr %g, %q_6 : !tt.ptr, i64 loc(#loc88) + %p_g_19 = arith.muli %o_t_16, %p_g : tensor<64xi64> loc(#loc76) + %p_g_20 = tt.splat %g_18 : !tt.ptr -> tensor<64x!tt.ptr> loc(#loc89) + %p_g_21 = tt.addptr %p_g_20, %p_g_19 : tensor<64x!tt.ptr>, tensor<64xi64> loc(#loc89) + %b_g_22 = tt.load %p_g_21, %m_t_17, %b_g : tensor<64x!tt.ptr> loc(#loc75) + %o_k = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc90) + %m_k = arith.cmpi slt, %o_k, %cst_4 : tensor<128xi32> loc(#loc91) + %p_k = tt.expand_dims %o_t_16 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc92) + %p_k_23 = arith.muli %p_k, %cst_3 : tensor<64x1xi64> loc(#loc93) + %p_k_24 = tt.splat %k_9 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc94) + %p_k_25 = tt.addptr %p_k_24, %p_k_23 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc94) + %p_k_26 = tt.expand_dims %o_k {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc95) + %p_k_27 = tt.broadcast %p_k_25 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc96) + %p_k_28 = tt.broadcast %p_k_26 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc96) + %p_k_29 = tt.addptr %p_k_27, %p_k_28 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc96) + %p_q = tt.expand_dims %o_k {axis = 1 : i32} : tensor<128xi32> -> tensor<128x1xi32> loc(#loc97) + %p_q_30 = tt.splat %q_8 : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc98) + %p_q_31 = tt.addptr %p_q_30, %p_q : tensor<128x1x!tt.ptr>, tensor<128x1xi32> loc(#loc98) + %p_q_32 = tt.expand_dims %o_t_16 {axis = 0 : i32} : tensor<64xi64> -> tensor<1x64xi64> loc(#loc99) + %p_q_33 = arith.muli %p_q_32, %cst_2 : tensor<1x64xi64> loc(#loc100) + %p_q_34 = tt.broadcast %p_q_31 : tensor<128x1x!tt.ptr> -> tensor<128x64x!tt.ptr> loc(#loc101) + %p_q_35 = tt.broadcast %p_q_33 : tensor<1x64xi64> -> tensor<128x64xi64> loc(#loc101) + %p_q_36 = tt.addptr %p_q_34, %p_q_35 : tensor<128x64x!tt.ptr>, tensor<128x64xi64> loc(#loc101) + %b_k = tt.expand_dims %m_t_17 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc102) + %b_k_37 = tt.expand_dims %m_k {axis = 0 : i32} : tensor<128xi1> -> tensor<1x128xi1> loc(#loc103) + %b_k_38 = tt.broadcast %b_k : tensor<64x1xi1> -> tensor<64x128xi1> loc(#loc104) + %b_k_39 = tt.broadcast %b_k_37 : tensor<1x128xi1> -> tensor<64x128xi1> loc(#loc104) + %b_k_40 = arith.andi %b_k_38, %b_k_39 : tensor<64x128xi1> loc(#loc104) + %b_k_41 = tt.load %p_k_29, %b_k_40, %cst : tensor<64x128x!tt.ptr> loc(#loc105) + %b_q_42 = tt.expand_dims %m_k {axis = 1 : i32} : tensor<128xi1> -> tensor<128x1xi1> loc(#loc106) + %b_q_43 = tt.expand_dims %m_t_17 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc107) + %b_q_44 = tt.broadcast %b_q_42 : tensor<128x1xi1> -> tensor<128x64xi1> loc(#loc108) + %b_q_45 = tt.broadcast %b_q_43 : tensor<1x64xi1> -> tensor<128x64xi1> loc(#loc108) + %b_q_46 = arith.andi %b_q_44, %b_q_45 : tensor<128x64xi1> loc(#loc108) + %b_q_47 = tt.load %p_q_36, %b_q_46, %b_q : tensor<128x64x!tt.ptr> loc(#loc74) + %b_A = tt.dot %b_k_41, %b_q_47, %cst_0, inputPrecision = tf32 : tensor<64x128xbf16> * tensor<128x64xbf16> -> tensor<64x64xf32> loc(#loc109) + %b_A_48 = tt.splat %scale : f32 -> tensor<64x64xf32> loc(#loc110) + %b_A_49 = arith.mulf %b_A, %b_A_48 : tensor<64x64xf32> loc(#loc110) + %b_A_50 = arith.addf %b_A_49, %cst_0 : tensor<64x64xf32> loc(#loc111) + %b_A_51 = tt.expand_dims %b_g_22 {axis = 0 : i32} : tensor<64xf32> -> tensor<1x64xf32> loc(#loc112) + %b_A_52 = tt.expand_dims %b_g_22 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc113) + %b_A_53 = tt.broadcast %b_A_51 : tensor<1x64xf32> -> tensor<64x64xf32> loc(#loc114) + %b_A_54 = tt.broadcast %b_A_52 : tensor<64x1xf32> -> tensor<64x64xf32> loc(#loc114) + %b_A_55 = arith.subf %b_A_53, %b_A_54 : tensor<64x64xf32> loc(#loc114) + %b_A_56 = math.exp2 %b_A_55 : tensor<64x64xf32> loc(#loc128) + %b_A_57 = arith.mulf %b_A_50, %b_A_56 : tensor<64x64xf32> loc(#loc116) + %m_A = tt.broadcast %p_k : tensor<64x1xi64> -> tensor<64x64xi64> loc(#loc117) + %m_A_58 = tt.broadcast %p_q_32 : tensor<1x64xi64> -> tensor<64x64xi64> loc(#loc117) + %m_A_59 = arith.cmpi sle, %m_A, %m_A_58 : tensor<64x64xi64> loc(#loc117) + %m_A_60 = tt.broadcast %b_k : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc118) + %m_A_61 = tt.broadcast %b_q_43 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc118) + %m_A_62 = arith.andi %m_A_60, %m_A_61 : tensor<64x64xi1> loc(#loc118) + %m_A_63 = arith.andi %m_A_59, %m_A_62 : tensor<64x64xi1> loc(#loc119) + %b_A_64 = arith.select %m_A_63, %b_A_57, %cst_0 : tensor<64x64xi1>, tensor<64x64xf32> loc(#loc120) + %b_A_65 = arith.truncf %b_A_64 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc121) + %p_do = tt.splat %do_10 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc122) + %p_do_66 = tt.addptr %p_do, %p_k_23 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc122) + %p_do_67 = tt.broadcast %p_do_66 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc123) + %p_do_68 = tt.addptr %p_do_67, %p_k_28 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc123) + %p_dv = tt.splat %dv_11 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc124) + %p_dv_69 = tt.addptr %p_dv, %p_k_23 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc124) + %p_dv_70 = tt.broadcast %p_dv_69 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc125) + %p_dv_71 = tt.addptr %p_dv_70, %p_k_28 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc125) + %b_do = tt.load %p_do_68, %b_k_40, %cst : tensor<64x128x!tt.ptr> loc(#loc126) + %b_dv = tt.dot %b_A_65, %b_do, %cst_1, inputPrecision = tf32 : tensor<64x64xbf16> * tensor<64x128xbf16> -> tensor<64x128xf32> loc(#loc127) + %8 = arith.truncf %b_dv : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc64) + tt.store %p_dv_71, %8, %b_k_40 : tensor<64x128x!tt.ptr> loc(#loc65) + tt.return loc(#loc66) + } loc(#loc) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":526:26) +#loc2 = loc(unknown) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":513:26) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":512:28) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":504:16) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":490:30) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":490:36) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":490:61) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":490:67) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":491:23) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":491:34) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":497:25) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":499:16) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":499:20) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":499:40) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":499:9) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":500:9) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":501:10) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":502:10) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":504:34) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":504:21) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":505:16) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":511:17) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":512:22) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":520:42) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":521:24) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":522:26) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":522:38) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":522:22) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":522:49) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":522:45) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":523:26) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":523:22) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":523:41) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":523:53) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":523:37) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":525:40) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":525:55) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":525:51) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":525:26) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":526:40) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":526:55) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":526:51) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":527:31) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":527:38) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":527:19) +#loc47 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":529:28) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":529:43) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":529:39) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:33) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":529:24) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":529:19) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":530:27) +#loc54 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":530:59) +#loc55 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":530:44) +#loc56 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":531:29) +#loc57 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":531:35) +#loc58 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":536:20) +#loc59 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":536:44) +#loc60 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":537:20) +#loc61 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":537:44) +#loc62 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":538:23) +#loc63 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":539:42) +#loc64 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":540:31) +#loc65 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":540:23) +#loc66 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":533:4) +#loc74 = loc("b_q"(#loc1)) +#loc75 = loc("b_g"(#loc3)) +#loc76 = loc("p_g"(#loc4)) +#loc77 = loc("o_t"(#loc5)) +#loc78 = loc("q"(#loc13)) +#loc79 = loc("q"(#loc14)) +#loc80 = loc("q"(#loc15)) +#loc81 = loc("q"(#loc16)) +#loc82 = loc("k"(#loc17)) +#loc83 = loc("do"(#loc18)) +#loc84 = loc("dv"(#loc19)) +#loc85 = loc("o_t"(#loc20)) +#loc86 = loc("o_t"(#loc21)) +#loc87 = loc("m_t"(#loc22)) +#loc88 = loc("g"(#loc23)) +#loc89 = loc("p_g"(#loc24)) +#loc90 = loc("o_k"(#loc25)) +#loc91 = loc("m_k"(#loc26)) +#loc92 = loc("p_k"(#loc27)) +#loc93 = loc("p_k"(#loc28)) +#loc94 = loc("p_k"(#loc29)) +#loc95 = loc("p_k"(#loc30)) +#loc96 = loc("p_k"(#loc31)) +#loc97 = loc("p_q"(#loc32)) +#loc98 = loc("p_q"(#loc33)) +#loc99 = loc("p_q"(#loc34)) +#loc100 = loc("p_q"(#loc35)) +#loc101 = loc("p_q"(#loc36)) +#loc102 = loc("b_k"(#loc37)) +#loc103 = loc("b_k"(#loc38)) +#loc104 = loc("b_k"(#loc39)) +#loc105 = loc("b_k"(#loc40)) +#loc106 = loc("b_q"(#loc41)) +#loc107 = loc("b_q"(#loc42)) +#loc108 = loc("b_q"(#loc43)) +#loc109 = loc("b_A"(#loc44)) +#loc110 = loc("b_A"(#loc45)) +#loc111 = loc("b_A"(#loc46)) +#loc112 = loc("b_A"(#loc47)) +#loc113 = loc("b_A"(#loc48)) +#loc114 = loc("b_A"(#loc49)) +#loc115 = loc("b_A"(#loc51)) +#loc116 = loc("b_A"(#loc52)) +#loc117 = loc("m_A"(#loc53)) +#loc118 = loc("m_A"(#loc54)) +#loc119 = loc("m_A"(#loc55)) +#loc120 = loc("b_A"(#loc56)) +#loc121 = loc("b_A"(#loc57)) +#loc122 = loc("p_do"(#loc58)) +#loc123 = loc("p_do"(#loc59)) +#loc124 = loc("p_dv"(#loc60)) +#loc125 = loc("p_dv"(#loc61)) +#loc126 = loc("b_do"(#loc62)) +#loc127 = loc("b_dv"(#loc63)) +#loc128 = loc(callsite(#loc50 at #loc115)) diff --git a/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/__grp__l2norm_fwd_kernel.json b/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/__grp__l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..29fb4110012399ec9cdd7cf242caf3831cdc7c95 --- /dev/null +++ b/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/__grp__l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_fwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/l2norm_fwd_kernel.source", "l2norm_fwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/l2norm_fwd_kernel.ttir", "l2norm_fwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/l2norm_fwd_kernel.ttgir", "l2norm_fwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/l2norm_fwd_kernel.llir", "l2norm_fwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/l2norm_fwd_kernel.ptx", "l2norm_fwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/l2norm_fwd_kernel.cubin", "l2norm_fwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/l2norm_fwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/l2norm_fwd_kernel.cubin b/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/l2norm_fwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..9abef151dae160ade70af4a66e9b2ae99e0c9dd7 Binary files /dev/null and b/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/l2norm_fwd_kernel.cubin differ diff --git a/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/l2norm_fwd_kernel.json b/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..362c8ef14c6805e5c15489f3e2a1f2b3a5b5ca4f --- /dev/null +++ b/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"hash": "17997274280bfe68bdfae51960f7522c78abbfc85f306b13bf849f15bb7af142", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 8, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 256, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_fwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/l2norm_fwd_kernel.llir b/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/l2norm_fwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..a9eb789100099aee9135e15092e32ed65fbb1fea --- /dev/null +++ b/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/l2norm_fwd_kernel.llir @@ -0,0 +1,637 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 +@.str.2 = private unnamed_addr constant [17 x i8] c"__CUDA_PREC_SQRT\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_fwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, float %3, i32 %4, ptr addrspace(1) readnone captures(none) %5, ptr addrspace(1) readnone captures(none) %6) local_unnamed_addr #0 !dbg !5 { + %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %9 = zext nneg i32 %8 to i64, !dbg !14 + %10 = shl nuw nsw i64 %9, 6, !dbg !15 + %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !16 + %12 = and i32 %11, 240, !dbg !16 + %13 = lshr exact i32 %12, 4, !dbg !16 + %14 = or disjoint i32 %13, 16, !dbg !16 + %15 = or disjoint i32 %13, 32, !dbg !16 + %16 = or disjoint i32 %13, 48, !dbg !16 + %17 = zext nneg i32 %13 to i64, !dbg !17 + %18 = zext nneg i32 %14 to i64, !dbg !17 + %19 = zext nneg i32 %15 to i64, !dbg !17 + %20 = zext nneg i32 %16 to i64, !dbg !17 + %21 = or disjoint i64 %10, %17, !dbg !17 + %22 = or disjoint i64 %10, %18, !dbg !17 + %23 = or disjoint i64 %10, %19, !dbg !17 + %24 = or disjoint i64 %10, %20, !dbg !17 + %25 = sext i32 %4 to i64, !dbg !18 + %26 = icmp slt i64 %21, %25, !dbg !18 + %27 = icmp slt i64 %22, %25, !dbg !18 + %28 = icmp slt i64 %23, %25, !dbg !18 + %29 = icmp slt i64 %24, %25, !dbg !18 + %30 = shl nuw nsw i32 %11, 3, !dbg !19 + %31 = and i32 %30, 120, !dbg !19 + %32 = shl nuw nsw i64 %21, 7, !dbg !20 + %33 = shl nuw nsw i64 %22, 7, !dbg !20 + %34 = shl nuw nsw i64 %23, 7, !dbg !20 + %35 = shl nuw nsw i64 %24, 7, !dbg !20 + %36 = getelementptr bfloat, ptr addrspace(1) %0, i64 %32, !dbg !21 + %37 = getelementptr bfloat, ptr addrspace(1) %0, i64 %33, !dbg !21 + %38 = getelementptr bfloat, ptr addrspace(1) %0, i64 %34, !dbg !21 + %39 = getelementptr bfloat, ptr addrspace(1) %0, i64 %35, !dbg !21 + %40 = zext nneg i32 %31 to i64, !dbg !22 + %41 = getelementptr bfloat, ptr addrspace(1) %36, i64 %40, !dbg !22 + %42 = getelementptr bfloat, ptr addrspace(1) %37, i64 %40, !dbg !22 + %43 = getelementptr bfloat, ptr addrspace(1) %38, i64 %40, !dbg !22 + %44 = getelementptr bfloat, ptr addrspace(1) %39, i64 %40, !dbg !22 + %45 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %41, i1 %26) #6, !dbg !23 + %46 = extractvalue { i32, i32, i32, i32 } %45, 0, !dbg !23 + %47 = bitcast i32 %46 to <2 x bfloat>, !dbg !23 + %48 = extractvalue { i32, i32, i32, i32 } %45, 1, !dbg !23 + %49 = bitcast i32 %48 to <2 x bfloat>, !dbg !23 + %50 = extractvalue { i32, i32, i32, i32 } %45, 2, !dbg !23 + %51 = bitcast i32 %50 to <2 x bfloat>, !dbg !23 + %52 = extractvalue { i32, i32, i32, i32 } %45, 3, !dbg !23 + %53 = bitcast i32 %52 to <2 x bfloat>, !dbg !23 + %54 = extractelement <2 x bfloat> %47, i64 0, !dbg !23 + %55 = extractelement <2 x bfloat> %47, i64 1, !dbg !23 + %56 = extractelement <2 x bfloat> %49, i64 0, !dbg !23 + %57 = extractelement <2 x bfloat> %49, i64 1, !dbg !23 + %58 = extractelement <2 x bfloat> %51, i64 0, !dbg !23 + %59 = extractelement <2 x bfloat> %51, i64 1, !dbg !23 + %60 = extractelement <2 x bfloat> %53, i64 0, !dbg !23 + %61 = extractelement <2 x bfloat> %53, i64 1, !dbg !23 + %62 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %42, i1 %27) #6, !dbg !23 + %63 = extractvalue { i32, i32, i32, i32 } %62, 0, !dbg !23 + %64 = bitcast i32 %63 to <2 x bfloat>, !dbg !23 + %65 = extractvalue { i32, i32, i32, i32 } %62, 1, !dbg !23 + %66 = bitcast i32 %65 to <2 x bfloat>, !dbg !23 + %67 = extractvalue { i32, i32, i32, i32 } %62, 2, !dbg !23 + %68 = bitcast i32 %67 to <2 x bfloat>, !dbg !23 + %69 = extractvalue { i32, i32, i32, i32 } %62, 3, !dbg !23 + %70 = bitcast i32 %69 to <2 x bfloat>, !dbg !23 + %71 = extractelement <2 x bfloat> %64, i64 0, !dbg !23 + %72 = extractelement <2 x bfloat> %64, i64 1, !dbg !23 + %73 = extractelement <2 x bfloat> %66, i64 0, !dbg !23 + %74 = extractelement <2 x bfloat> %66, i64 1, !dbg !23 + %75 = extractelement <2 x bfloat> %68, i64 0, !dbg !23 + %76 = extractelement <2 x bfloat> %68, i64 1, !dbg !23 + %77 = extractelement <2 x bfloat> %70, i64 0, !dbg !23 + %78 = extractelement <2 x bfloat> %70, i64 1, !dbg !23 + %79 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %43, i1 %28) #6, !dbg !23 + %80 = extractvalue { i32, i32, i32, i32 } %79, 0, !dbg !23 + %81 = bitcast i32 %80 to <2 x bfloat>, !dbg !23 + %82 = extractvalue { i32, i32, i32, i32 } %79, 1, !dbg !23 + %83 = bitcast i32 %82 to <2 x bfloat>, !dbg !23 + %84 = extractvalue { i32, i32, i32, i32 } %79, 2, !dbg !23 + %85 = bitcast i32 %84 to <2 x bfloat>, !dbg !23 + %86 = extractvalue { i32, i32, i32, i32 } %79, 3, !dbg !23 + %87 = bitcast i32 %86 to <2 x bfloat>, !dbg !23 + %88 = extractelement <2 x bfloat> %81, i64 0, !dbg !23 + %89 = extractelement <2 x bfloat> %81, i64 1, !dbg !23 + %90 = extractelement <2 x bfloat> %83, i64 0, !dbg !23 + %91 = extractelement <2 x bfloat> %83, i64 1, !dbg !23 + %92 = extractelement <2 x bfloat> %85, i64 0, !dbg !23 + %93 = extractelement <2 x bfloat> %85, i64 1, !dbg !23 + %94 = extractelement <2 x bfloat> %87, i64 0, !dbg !23 + %95 = extractelement <2 x bfloat> %87, i64 1, !dbg !23 + %96 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %44, i1 %29) #6, !dbg !23 + %97 = extractvalue { i32, i32, i32, i32 } %96, 0, !dbg !23 + %98 = bitcast i32 %97 to <2 x bfloat>, !dbg !23 + %99 = extractvalue { i32, i32, i32, i32 } %96, 1, !dbg !23 + %100 = bitcast i32 %99 to <2 x bfloat>, !dbg !23 + %101 = extractvalue { i32, i32, i32, i32 } %96, 2, !dbg !23 + %102 = bitcast i32 %101 to <2 x bfloat>, !dbg !23 + %103 = extractvalue { i32, i32, i32, i32 } %96, 3, !dbg !23 + %104 = bitcast i32 %103 to <2 x bfloat>, !dbg !23 + %105 = extractelement <2 x bfloat> %98, i64 0, !dbg !23 + %106 = extractelement <2 x bfloat> %98, i64 1, !dbg !23 + %107 = extractelement <2 x bfloat> %100, i64 0, !dbg !23 + %108 = extractelement <2 x bfloat> %100, i64 1, !dbg !23 + %109 = extractelement <2 x bfloat> %102, i64 0, !dbg !23 + %110 = extractelement <2 x bfloat> %102, i64 1, !dbg !23 + %111 = extractelement <2 x bfloat> %104, i64 0, !dbg !23 + %112 = extractelement <2 x bfloat> %104, i64 1, !dbg !23 + %113 = fpext bfloat %54 to float, !dbg !24 + %114 = fpext bfloat %55 to float, !dbg !24 + %115 = fpext bfloat %56 to float, !dbg !24 + %116 = fpext bfloat %57 to float, !dbg !24 + %117 = fpext bfloat %58 to float, !dbg !24 + %118 = fpext bfloat %59 to float, !dbg !24 + %119 = fpext bfloat %60 to float, !dbg !24 + %120 = fpext bfloat %61 to float, !dbg !24 + %121 = fpext bfloat %71 to float, !dbg !24 + %122 = fpext bfloat %72 to float, !dbg !24 + %123 = fpext bfloat %73 to float, !dbg !24 + %124 = fpext bfloat %74 to float, !dbg !24 + %125 = fpext bfloat %75 to float, !dbg !24 + %126 = fpext bfloat %76 to float, !dbg !24 + %127 = fpext bfloat %77 to float, !dbg !24 + %128 = fpext bfloat %78 to float, !dbg !24 + %129 = fpext bfloat %88 to float, !dbg !24 + %130 = fpext bfloat %89 to float, !dbg !24 + %131 = fpext bfloat %90 to float, !dbg !24 + %132 = fpext bfloat %91 to float, !dbg !24 + %133 = fpext bfloat %92 to float, !dbg !24 + %134 = fpext bfloat %93 to float, !dbg !24 + %135 = fpext bfloat %94 to float, !dbg !24 + %136 = fpext bfloat %95 to float, !dbg !24 + %137 = fpext bfloat %105 to float, !dbg !24 + %138 = fpext bfloat %106 to float, !dbg !24 + %139 = fpext bfloat %107 to float, !dbg !24 + %140 = fpext bfloat %108 to float, !dbg !24 + %141 = fpext bfloat %109 to float, !dbg !24 + %142 = fpext bfloat %110 to float, !dbg !24 + %143 = fpext bfloat %111 to float, !dbg !24 + %144 = fpext bfloat %112 to float, !dbg !24 + %145 = fmul float %113, %113, !dbg !25 + %146 = fmul float %114, %114, !dbg !25 + %147 = fmul float %115, %115, !dbg !25 + %148 = fmul float %116, %116, !dbg !25 + %149 = fmul float %117, %117, !dbg !25 + %150 = fmul float %118, %118, !dbg !25 + %151 = fmul float %119, %119, !dbg !25 + %152 = fmul float %120, %120, !dbg !25 + %153 = fmul float %121, %121, !dbg !25 + %154 = fmul float %122, %122, !dbg !25 + %155 = fmul float %123, %123, !dbg !25 + %156 = fmul float %124, %124, !dbg !25 + %157 = fmul float %125, %125, !dbg !25 + %158 = fmul float %126, %126, !dbg !25 + %159 = fmul float %127, %127, !dbg !25 + %160 = fmul float %128, %128, !dbg !25 + %161 = fmul float %129, %129, !dbg !25 + %162 = fmul float %130, %130, !dbg !25 + %163 = fmul float %131, %131, !dbg !25 + %164 = fmul float %132, %132, !dbg !25 + %165 = fmul float %133, %133, !dbg !25 + %166 = fmul float %134, %134, !dbg !25 + %167 = fmul float %135, %135, !dbg !25 + %168 = fmul float %136, %136, !dbg !25 + %169 = fmul float %137, %137, !dbg !25 + %170 = fmul float %138, %138, !dbg !25 + %171 = fmul float %139, %139, !dbg !25 + %172 = fmul float %140, %140, !dbg !25 + %173 = fmul float %141, %141, !dbg !25 + %174 = fmul float %142, %142, !dbg !25 + %175 = fmul float %143, %143, !dbg !25 + %176 = fmul float %144, %144, !dbg !25 + %177 = fadd float %145, %146, !dbg !26 + %178 = fadd float %147, %177, !dbg !26 + %179 = fadd float %148, %178, !dbg !26 + %180 = fadd float %149, %179, !dbg !26 + %181 = fadd float %150, %180, !dbg !26 + %182 = fadd float %151, %181, !dbg !26 + %183 = fadd float %152, %182, !dbg !26 + %184 = fadd float %153, %154, !dbg !26 + %185 = fadd float %155, %184, !dbg !26 + %186 = fadd float %156, %185, !dbg !26 + %187 = fadd float %157, %186, !dbg !26 + %188 = fadd float %158, %187, !dbg !26 + %189 = fadd float %159, %188, !dbg !26 + %190 = fadd float %160, %189, !dbg !26 + %191 = fadd float %161, %162, !dbg !26 + %192 = fadd float %163, %191, !dbg !26 + %193 = fadd float %164, %192, !dbg !26 + %194 = fadd float %165, %193, !dbg !26 + %195 = fadd float %166, %194, !dbg !26 + %196 = fadd float %167, %195, !dbg !26 + %197 = fadd float %168, %196, !dbg !26 + %198 = fadd float %169, %170, !dbg !26 + %199 = fadd float %171, %198, !dbg !26 + %200 = fadd float %172, %199, !dbg !26 + %201 = fadd float %173, %200, !dbg !26 + %202 = fadd float %174, %201, !dbg !26 + %203 = fadd float %175, %202, !dbg !26 + %204 = fadd float %176, %203, !dbg !26 + %205 = bitcast float %183 to i32, !dbg !29 + %206 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %205, i32 8, i32 31), !dbg !29 + %207 = bitcast i32 %206 to float, !dbg !29 + %208 = fadd float %183, %207, !dbg !26 + %209 = bitcast float %208 to i32, !dbg !29 + %210 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %209, i32 4, i32 31), !dbg !29 + %211 = bitcast i32 %210 to float, !dbg !29 + %212 = fadd float %208, %211, !dbg !26 + %213 = bitcast float %212 to i32, !dbg !29 + %214 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %213, i32 2, i32 31), !dbg !29 + %215 = bitcast i32 %214 to float, !dbg !29 + %216 = fadd float %212, %215, !dbg !26 + %217 = bitcast float %216 to i32, !dbg !29 + %218 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %217, i32 1, i32 31), !dbg !29 + %219 = bitcast i32 %218 to float, !dbg !29 + %220 = fadd float %216, %219, !dbg !26 + %221 = bitcast float %190 to i32, !dbg !29 + %222 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %221, i32 8, i32 31), !dbg !29 + %223 = bitcast i32 %222 to float, !dbg !29 + %224 = fadd float %190, %223, !dbg !26 + %225 = bitcast float %224 to i32, !dbg !29 + %226 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %225, i32 4, i32 31), !dbg !29 + %227 = bitcast i32 %226 to float, !dbg !29 + %228 = fadd float %224, %227, !dbg !26 + %229 = bitcast float %228 to i32, !dbg !29 + %230 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %229, i32 2, i32 31), !dbg !29 + %231 = bitcast i32 %230 to float, !dbg !29 + %232 = fadd float %228, %231, !dbg !26 + %233 = bitcast float %232 to i32, !dbg !29 + %234 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %233, i32 1, i32 31), !dbg !29 + %235 = bitcast i32 %234 to float, !dbg !29 + %236 = fadd float %232, %235, !dbg !26 + %237 = bitcast float %197 to i32, !dbg !29 + %238 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %237, i32 8, i32 31), !dbg !29 + %239 = bitcast i32 %238 to float, !dbg !29 + %240 = fadd float %197, %239, !dbg !26 + %241 = bitcast float %240 to i32, !dbg !29 + %242 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %241, i32 4, i32 31), !dbg !29 + %243 = bitcast i32 %242 to float, !dbg !29 + %244 = fadd float %240, %243, !dbg !26 + %245 = bitcast float %244 to i32, !dbg !29 + %246 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %245, i32 2, i32 31), !dbg !29 + %247 = bitcast i32 %246 to float, !dbg !29 + %248 = fadd float %244, %247, !dbg !26 + %249 = bitcast float %248 to i32, !dbg !29 + %250 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %249, i32 1, i32 31), !dbg !29 + %251 = bitcast i32 %250 to float, !dbg !29 + %252 = fadd float %248, %251, !dbg !26 + %253 = bitcast float %204 to i32, !dbg !29 + %254 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %253, i32 8, i32 31), !dbg !29 + %255 = bitcast i32 %254 to float, !dbg !29 + %256 = fadd float %204, %255, !dbg !26 + %257 = bitcast float %256 to i32, !dbg !29 + %258 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %257, i32 4, i32 31), !dbg !29 + %259 = bitcast i32 %258 to float, !dbg !29 + %260 = fadd float %256, %259, !dbg !26 + %261 = bitcast float %260 to i32, !dbg !29 + %262 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %261, i32 2, i32 31), !dbg !29 + %263 = bitcast i32 %262 to float, !dbg !29 + %264 = fadd float %260, %263, !dbg !26 + %265 = bitcast float %264 to i32, !dbg !29 + %266 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %265, i32 1, i32 31), !dbg !29 + %267 = bitcast i32 %266 to float, !dbg !29 + %268 = fadd float %264, %267, !dbg !26 + %269 = fadd float %3, %220, !dbg !32 + %270 = fadd float %3, %236, !dbg !32 + %271 = fadd float %3, %252, !dbg !32 + %272 = fadd float %3, %268, !dbg !32 + %273 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i = icmp eq i32 %273, 0, !dbg !33 + %274 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i = icmp eq i32 %274, 0, !dbg !33 + br i1 %.not.i, label %280, label %275, !dbg !33 + +275: ; preds = %7 + br i1 %.not1.i, label %278, label %276, !dbg !33 + +276: ; preds = %275 + %277 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %269) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +278: ; preds = %275 + %279 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %269) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +280: ; preds = %7 + br i1 %.not1.i, label %283, label %281, !dbg !33 + +281: ; preds = %280 + %282 = tail call float @llvm.nvvm.sqrt.rn.f(float %269) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +283: ; preds = %280 + %284 = tail call float @llvm.nvvm.sqrt.approx.f(float %269) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +__nv_sqrtf.exit: ; preds = %276, %278, %281, %283 + %.0.i = phi float [ %277, %276 ], [ %279, %278 ], [ %282, %281 ], [ %284, %283 ], !dbg !33 + %285 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i1 = icmp eq i32 %285, 0, !dbg !33 + %286 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i4 = icmp eq i32 %286, 0, !dbg !33 + br i1 %.not.i1, label %292, label %287, !dbg !33 + +287: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %290, label %288, !dbg !33 + +288: ; preds = %287 + %289 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %270) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +290: ; preds = %287 + %291 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %270) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +292: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %295, label %293, !dbg !33 + +293: ; preds = %292 + %294 = tail call float @llvm.nvvm.sqrt.rn.f(float %270) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +295: ; preds = %292 + %296 = tail call float @llvm.nvvm.sqrt.approx.f(float %270) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +__nv_sqrtf.exit5: ; preds = %288, %290, %293, %295 + %.0.i3 = phi float [ %289, %288 ], [ %291, %290 ], [ %294, %293 ], [ %296, %295 ], !dbg !33 + %297 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i6 = icmp eq i32 %297, 0, !dbg !33 + %298 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i9 = icmp eq i32 %298, 0, !dbg !33 + br i1 %.not.i6, label %304, label %299, !dbg !33 + +299: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %302, label %300, !dbg !33 + +300: ; preds = %299 + %301 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %271) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +302: ; preds = %299 + %303 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %271) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +304: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %307, label %305, !dbg !33 + +305: ; preds = %304 + %306 = tail call float @llvm.nvvm.sqrt.rn.f(float %271) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +307: ; preds = %304 + %308 = tail call float @llvm.nvvm.sqrt.approx.f(float %271) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +__nv_sqrtf.exit10: ; preds = %300, %302, %305, %307 + %.0.i8 = phi float [ %301, %300 ], [ %303, %302 ], [ %306, %305 ], [ %308, %307 ], !dbg !33 + %309 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i11 = icmp eq i32 %309, 0, !dbg !33 + %310 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i14 = icmp eq i32 %310, 0, !dbg !33 + br i1 %.not.i11, label %316, label %311, !dbg !33 + +311: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %314, label %312, !dbg !33 + +312: ; preds = %311 + %313 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %272) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +314: ; preds = %311 + %315 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %272) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +316: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %319, label %317, !dbg !33 + +317: ; preds = %316 + %318 = tail call float @llvm.nvvm.sqrt.rn.f(float %272) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +319: ; preds = %316 + %320 = tail call float @llvm.nvvm.sqrt.approx.f(float %272) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +__nv_sqrtf.exit15: ; preds = %312, %314, %317, %319 + %.0.i13 = phi float [ %313, %312 ], [ %315, %314 ], [ %318, %317 ], [ %320, %319 ], !dbg !33 + %321 = and i32 %11, 63, !dbg !16 + %322 = zext nneg i32 %321 to i64, !dbg !17 + %323 = or disjoint i64 %10, %322, !dbg !17 + %324 = getelementptr float, ptr addrspace(1) %2, i64 %323, !dbg !34 + %325 = getelementptr bfloat, ptr addrspace(1) %1, i64 %35, !dbg !35 + %326 = getelementptr bfloat, ptr addrspace(1) %325, i64 %40, !dbg !36 + %327 = getelementptr bfloat, ptr addrspace(1) %1, i64 %34, !dbg !35 + %328 = getelementptr bfloat, ptr addrspace(1) %327, i64 %40, !dbg !36 + %329 = getelementptr bfloat, ptr addrspace(1) %1, i64 %33, !dbg !35 + %330 = getelementptr bfloat, ptr addrspace(1) %329, i64 %40, !dbg !36 + %331 = getelementptr bfloat, ptr addrspace(1) %1, i64 %32, !dbg !35 + %332 = getelementptr bfloat, ptr addrspace(1) %331, i64 %40, !dbg !36 + %333 = icmp slt i64 %323, %25, !dbg !18 + %334 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i), !dbg !37 + %335 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i3), !dbg !37 + %336 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i8), !dbg !37 + %337 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i13), !dbg !37 + %338 = fmul float %334, %113, !dbg !38 + %339 = fmul float %334, %114, !dbg !38 + %340 = fmul float %334, %115, !dbg !38 + %341 = fmul float %334, %116, !dbg !38 + %342 = fmul float %334, %117, !dbg !38 + %343 = fmul float %334, %118, !dbg !38 + %344 = fmul float %334, %119, !dbg !38 + %345 = fmul float %334, %120, !dbg !38 + %346 = fmul float %335, %121, !dbg !38 + %347 = fmul float %335, %122, !dbg !38 + %348 = fmul float %335, %123, !dbg !38 + %349 = fmul float %335, %124, !dbg !38 + %350 = fmul float %335, %125, !dbg !38 + %351 = fmul float %335, %126, !dbg !38 + %352 = fmul float %335, %127, !dbg !38 + %353 = fmul float %335, %128, !dbg !38 + %354 = fmul float %336, %129, !dbg !38 + %355 = fmul float %336, %130, !dbg !38 + %356 = fmul float %336, %131, !dbg !38 + %357 = fmul float %336, %132, !dbg !38 + %358 = fmul float %336, %133, !dbg !38 + %359 = fmul float %336, %134, !dbg !38 + %360 = fmul float %336, %135, !dbg !38 + %361 = fmul float %336, %136, !dbg !38 + %362 = fmul float %337, %137, !dbg !38 + %363 = fmul float %337, %138, !dbg !38 + %364 = fmul float %337, %139, !dbg !38 + %365 = fmul float %337, %140, !dbg !38 + %366 = fmul float %337, %141, !dbg !38 + %367 = fmul float %337, %142, !dbg !38 + %368 = fmul float %337, %143, !dbg !38 + %369 = fmul float %337, %144, !dbg !38 + %370 = fptrunc float %338 to bfloat, !dbg !39 + %371 = fptrunc float %339 to bfloat, !dbg !39 + %372 = fptrunc float %340 to bfloat, !dbg !39 + %373 = fptrunc float %341 to bfloat, !dbg !39 + %374 = fptrunc float %342 to bfloat, !dbg !39 + %375 = fptrunc float %343 to bfloat, !dbg !39 + %376 = fptrunc float %344 to bfloat, !dbg !39 + %377 = fptrunc float %345 to bfloat, !dbg !39 + %378 = fptrunc float %346 to bfloat, !dbg !39 + %379 = fptrunc float %347 to bfloat, !dbg !39 + %380 = fptrunc float %348 to bfloat, !dbg !39 + %381 = fptrunc float %349 to bfloat, !dbg !39 + %382 = fptrunc float %350 to bfloat, !dbg !39 + %383 = fptrunc float %351 to bfloat, !dbg !39 + %384 = fptrunc float %352 to bfloat, !dbg !39 + %385 = fptrunc float %353 to bfloat, !dbg !39 + %386 = fptrunc float %354 to bfloat, !dbg !39 + %387 = fptrunc float %355 to bfloat, !dbg !39 + %388 = fptrunc float %356 to bfloat, !dbg !39 + %389 = fptrunc float %357 to bfloat, !dbg !39 + %390 = fptrunc float %358 to bfloat, !dbg !39 + %391 = fptrunc float %359 to bfloat, !dbg !39 + %392 = fptrunc float %360 to bfloat, !dbg !39 + %393 = fptrunc float %361 to bfloat, !dbg !39 + %394 = fptrunc float %362 to bfloat, !dbg !39 + %395 = fptrunc float %363 to bfloat, !dbg !39 + %396 = fptrunc float %364 to bfloat, !dbg !39 + %397 = fptrunc float %365 to bfloat, !dbg !39 + %398 = fptrunc float %366 to bfloat, !dbg !39 + %399 = fptrunc float %367 to bfloat, !dbg !39 + %400 = fptrunc float %368 to bfloat, !dbg !39 + %401 = fptrunc float %369 to bfloat, !dbg !39 + %402 = insertelement <2 x bfloat> poison, bfloat %370, i64 0, !dbg !40 + %403 = insertelement <2 x bfloat> %402, bfloat %371, i64 1, !dbg !40 + %404 = bitcast <2 x bfloat> %403 to i32, !dbg !40 + %405 = insertelement <2 x bfloat> poison, bfloat %372, i64 0, !dbg !40 + %406 = insertelement <2 x bfloat> %405, bfloat %373, i64 1, !dbg !40 + %407 = bitcast <2 x bfloat> %406 to i32, !dbg !40 + %408 = insertelement <2 x bfloat> poison, bfloat %374, i64 0, !dbg !40 + %409 = insertelement <2 x bfloat> %408, bfloat %375, i64 1, !dbg !40 + %410 = bitcast <2 x bfloat> %409 to i32, !dbg !40 + %411 = insertelement <2 x bfloat> poison, bfloat %376, i64 0, !dbg !40 + %412 = insertelement <2 x bfloat> %411, bfloat %377, i64 1, !dbg !40 + %413 = bitcast <2 x bfloat> %412 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %404, i32 %407, i32 %410, i32 %413, ptr addrspace(1) %332, i1 %26) #6, !dbg !40 + %414 = insertelement <2 x bfloat> poison, bfloat %378, i64 0, !dbg !40 + %415 = insertelement <2 x bfloat> %414, bfloat %379, i64 1, !dbg !40 + %416 = bitcast <2 x bfloat> %415 to i32, !dbg !40 + %417 = insertelement <2 x bfloat> poison, bfloat %380, i64 0, !dbg !40 + %418 = insertelement <2 x bfloat> %417, bfloat %381, i64 1, !dbg !40 + %419 = bitcast <2 x bfloat> %418 to i32, !dbg !40 + %420 = insertelement <2 x bfloat> poison, bfloat %382, i64 0, !dbg !40 + %421 = insertelement <2 x bfloat> %420, bfloat %383, i64 1, !dbg !40 + %422 = bitcast <2 x bfloat> %421 to i32, !dbg !40 + %423 = insertelement <2 x bfloat> poison, bfloat %384, i64 0, !dbg !40 + %424 = insertelement <2 x bfloat> %423, bfloat %385, i64 1, !dbg !40 + %425 = bitcast <2 x bfloat> %424 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %416, i32 %419, i32 %422, i32 %425, ptr addrspace(1) %330, i1 %27) #6, !dbg !40 + %426 = insertelement <2 x bfloat> poison, bfloat %386, i64 0, !dbg !40 + %427 = insertelement <2 x bfloat> %426, bfloat %387, i64 1, !dbg !40 + %428 = bitcast <2 x bfloat> %427 to i32, !dbg !40 + %429 = insertelement <2 x bfloat> poison, bfloat %388, i64 0, !dbg !40 + %430 = insertelement <2 x bfloat> %429, bfloat %389, i64 1, !dbg !40 + %431 = bitcast <2 x bfloat> %430 to i32, !dbg !40 + %432 = insertelement <2 x bfloat> poison, bfloat %390, i64 0, !dbg !40 + %433 = insertelement <2 x bfloat> %432, bfloat %391, i64 1, !dbg !40 + %434 = bitcast <2 x bfloat> %433 to i32, !dbg !40 + %435 = insertelement <2 x bfloat> poison, bfloat %392, i64 0, !dbg !40 + %436 = insertelement <2 x bfloat> %435, bfloat %393, i64 1, !dbg !40 + %437 = bitcast <2 x bfloat> %436 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %428, i32 %431, i32 %434, i32 %437, ptr addrspace(1) %328, i1 %28) #6, !dbg !40 + %438 = insertelement <2 x bfloat> poison, bfloat %394, i64 0, !dbg !40 + %439 = insertelement <2 x bfloat> %438, bfloat %395, i64 1, !dbg !40 + %440 = bitcast <2 x bfloat> %439 to i32, !dbg !40 + %441 = insertelement <2 x bfloat> poison, bfloat %396, i64 0, !dbg !40 + %442 = insertelement <2 x bfloat> %441, bfloat %397, i64 1, !dbg !40 + %443 = bitcast <2 x bfloat> %442 to i32, !dbg !40 + %444 = insertelement <2 x bfloat> poison, bfloat %398, i64 0, !dbg !40 + %445 = insertelement <2 x bfloat> %444, bfloat %399, i64 1, !dbg !40 + %446 = bitcast <2 x bfloat> %445 to i32, !dbg !40 + %447 = insertelement <2 x bfloat> poison, bfloat %400, i64 0, !dbg !40 + %448 = insertelement <2 x bfloat> %447, bfloat %401, i64 1, !dbg !40 + %449 = bitcast <2 x bfloat> %448 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %440, i32 %443, i32 %446, i32 %449, ptr addrspace(1) %326, i1 %29) #6, !dbg !40 + %450 = lshr exact i32 %12, 1, !dbg !41 + %451 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %450, !dbg !41 + %452 = bitcast float %334 to i32, !dbg !41 + %453 = bitcast float %335 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v2.b32 [ $0 + 0 ], { $1, $2 };", "r,r,r"(ptr addrspace(3) %451, i32 %452, i32 %453) #6, !dbg !41 + %454 = getelementptr inbounds nuw i8, ptr addrspace(3) %451, i32 128, !dbg !41 + %455 = bitcast float %336 to i32, !dbg !41 + %456 = bitcast float %337 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v2.b32 [ $0 + 0 ], { $1, $2 };", "r,r,r"(ptr addrspace(3) nonnull %454, i32 %455, i32 %456) #6, !dbg !41 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !41 + %457 = lshr i32 %11, 2, !dbg !41 + %458 = and i32 %457, 4, !dbg !41 + %459 = shl nuw nsw i32 %11, 2, !dbg !41 + %460 = and i32 %459, 128, !dbg !41 + %461 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %458, !dbg !41 + %462 = getelementptr inbounds nuw i8, ptr addrspace(3) %461, i32 %460, !dbg !41 + %463 = getelementptr inbounds nuw i8, ptr addrspace(3) %462, i32 %31, !dbg !41 + %464 = load i32, ptr addrspace(3) %463, align 4, !dbg !41 + %465 = and i32 %11, 192, !dbg !41 + %466 = icmp eq i32 %465, 0, !dbg !41 + %467 = and i1 %466, %333, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %464, ptr addrspace(1) %324, i1 %467) #6, !dbg !41 + ret void, !dbg !42 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.div.full(float, float) #3 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #4 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #5 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.f(float) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="256" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) } +attributes #4 = { convergent nocallback nounwind } +attributes #5 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #6 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "l2norm_fwd_kernel", linkageName: "l2norm_fwd_kernel", scope: !1, file: !1, line: 69, type: !6, scopeLine: 69, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !10, !11, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 70, column: 24, scope: !5) +!14 = !DILocation(line: 70, column: 30, scope: !5) +!15 = !DILocation(line: 71, column: 16, scope: !5) +!16 = !DILocation(line: 71, column: 34, scope: !5) +!17 = !DILocation(line: 71, column: 21, scope: !5) +!18 = !DILocation(line: 73, column: 16, scope: !5) +!19 = !DILocation(line: 74, column: 30, scope: !5) +!20 = !DILocation(line: 75, column: 29, scope: !5) +!21 = !DILocation(line: 75, column: 14, scope: !5) +!22 = !DILocation(line: 75, column: 33, scope: !5) +!23 = !DILocation(line: 79, column: 18, scope: !5) +!24 = !DILocation(line: 79, column: 47, scope: !5) +!25 = !DILocation(line: 80, column: 38, scope: !5) +!26 = !DILocation(line: 263, column: 15, scope: !27, inlinedAt: !29) +!27 = distinct !DILexicalBlockFile(scope: !5, file: !28, discriminator: 0) +!28 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!29 = !DILocation(line: 293, column: 36, scope: !27, inlinedAt: !30) +!30 = !DILocation(line: 80, column: 43, scope: !31) +!31 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!32 = !DILocation(line: 80, column: 48, scope: !5) +!33 = !DILocation(line: 80, column: 25, scope: !5) +!34 = !DILocation(line: 77, column: 20, scope: !5) +!35 = !DILocation(line: 76, column: 14, scope: !5) +!36 = !DILocation(line: 76, column: 33, scope: !5) +!37 = !DILocation(line: 80, column: 17, scope: !5) +!38 = !DILocation(line: 81, column: 16, scope: !5) +!39 = !DILocation(line: 83, column: 25, scope: !5) +!40 = !DILocation(line: 83, column: 18, scope: !5) +!41 = !DILocation(line: 84, column: 21, scope: !5) +!42 = !DILocation(line: 84, column: 4, scope: !5) diff --git a/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/l2norm_fwd_kernel.ptx b/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/l2norm_fwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..fecbe610272e5b85fe448594c8a7eecfa8c24ddb --- /dev/null +++ b/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/l2norm_fwd_kernel.ptx @@ -0,0 +1,648 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_fwd_kernel // -- Begin function l2norm_fwd_kernel +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; +.global .align 1 .b8 _$_str_$_2[17] = {95, 95, 67, 85, 68, 65, 95, 80, 82, 69, 67, 95, 83, 81, 82, 84}; + // @l2norm_fwd_kernel +.visible .entry l2norm_fwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_2, + .param .f32 l2norm_fwd_kernel_param_3, + .param .u32 l2norm_fwd_kernel_param_4, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_6 +) +.reqntid 256 +{ + .reg .pred %p<8>; + .reg .b16 %rs<33>; + .reg .b32 %r<199>; + .reg .b64 %rd<39>; + .loc 1 69 0 // l2norm.py:69:0 +$L__func_begin0: + .loc 1 69 0 // l2norm.py:69:0 + +// %bb.0: // %__nv_sqrtf.exit + ld.param.b64 %rd10, [l2norm_fwd_kernel_param_0]; + ld.param.b64 %rd11, [l2norm_fwd_kernel_param_1]; +$L__tmp0: + .loc 1 70 24 // l2norm.py:70:24 + mov.u32 %r41, %ctaid.x; + ld.param.b64 %rd12, [l2norm_fwd_kernel_param_2]; + .loc 1 71 16 // l2norm.py:71:16 + mul.wide.u32 %rd13, %r41, 64; + ld.param.b32 %r42, [l2norm_fwd_kernel_param_3]; + ld.param.s32 %rd14, [l2norm_fwd_kernel_param_4]; + .loc 1 71 34 // l2norm.py:71:34 + mov.u32 %r43, %tid.x; + and.b32 %r44, %r43, 240; + bfe.u32 %r45, %r43, 4, 4; + or.b32 %r46, %r45, 16; + or.b32 %r47, %r45, 32; + or.b32 %r48, %r45, 48; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd15, %r45; + cvt.u64.u32 %rd16, %r46; + cvt.u64.u32 %rd17, %r47; + cvt.u64.u32 %rd18, %r48; + or.b64 %rd19, %rd13, %rd15; + or.b64 %rd20, %rd13, %rd16; + or.b64 %rd21, %rd13, %rd17; + or.b64 %rd22, %rd13, %rd18; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p1, %rd19, %rd14; + setp.lt.s64 %p2, %rd20, %rd14; + setp.lt.s64 %p3, %rd21, %rd14; + setp.lt.s64 %p4, %rd22, %rd14; + .loc 1 74 30 // l2norm.py:74:30 + shl.b32 %r49, %r43, 3; + and.b32 %r50, %r49, 120; + .loc 1 75 14 // l2norm.py:75:14 + shl.b64 %rd23, %rd19, 8; + add.s64 %rd24, %rd10, %rd23; + shl.b64 %rd25, %rd20, 8; + add.s64 %rd26, %rd10, %rd25; + shl.b64 %rd27, %rd21, 8; + add.s64 %rd28, %rd10, %rd27; + shl.b64 %rd29, %rd22, 8; + add.s64 %rd30, %rd10, %rd29; + .loc 1 75 33 // l2norm.py:75:33 + mul.wide.u32 %rd31, %r50, 2; + add.s64 %rd1, %rd24, %rd31; + add.s64 %rd2, %rd26, %rd31; + add.s64 %rd3, %rd28, %rd31; + add.s64 %rd4, %rd30, %rd31; + mov.b32 %r5, 0; + .loc 1 79 18 // l2norm.py:79:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + mov.b32 {%rs5, %rs6}, %r3; + mov.b32 {%rs7, %rs8}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs9, %rs10}, %r6; + mov.b32 {%rs11, %rs12}, %r7; + mov.b32 {%rs13, %rs14}, %r8; + mov.b32 {%rs15, %rs16}, %r9; + // begin inline asm + mov.u32 %r10, %r5; + mov.u32 %r11, %r5; + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + @%p3 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs17, %rs18}, %r10; + mov.b32 {%rs19, %rs20}, %r11; + mov.b32 {%rs21, %rs22}, %r12; + mov.b32 {%rs23, %rs24}, %r13; + // begin inline asm + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + @%p4 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ]; + // end inline asm + mov.b32 {%rs25, %rs26}, %r14; + mov.b32 {%rs27, %rs28}, %r15; + mov.b32 {%rs29, %rs30}, %r16; + mov.b32 {%rs31, %rs32}, %r17; + .loc 1 79 47 // l2norm.py:79:47 + cvt.f32.bf16 %r51, %rs1; + cvt.f32.bf16 %r52, %rs2; + cvt.f32.bf16 %r53, %rs3; + cvt.f32.bf16 %r54, %rs4; + cvt.f32.bf16 %r55, %rs5; + cvt.f32.bf16 %r56, %rs6; + cvt.f32.bf16 %r57, %rs7; + cvt.f32.bf16 %r58, %rs8; + cvt.f32.bf16 %r59, %rs9; + cvt.f32.bf16 %r60, %rs10; + cvt.f32.bf16 %r61, %rs11; + cvt.f32.bf16 %r62, %rs12; + cvt.f32.bf16 %r63, %rs13; + cvt.f32.bf16 %r64, %rs14; + cvt.f32.bf16 %r65, %rs15; + cvt.f32.bf16 %r66, %rs16; + cvt.f32.bf16 %r67, %rs17; + cvt.f32.bf16 %r68, %rs18; + cvt.f32.bf16 %r69, %rs19; + cvt.f32.bf16 %r70, %rs20; + cvt.f32.bf16 %r71, %rs21; + cvt.f32.bf16 %r72, %rs22; + cvt.f32.bf16 %r73, %rs23; + cvt.f32.bf16 %r74, %rs24; + cvt.f32.bf16 %r75, %rs25; + cvt.f32.bf16 %r76, %rs26; + cvt.f32.bf16 %r77, %rs27; + cvt.f32.bf16 %r78, %rs28; + cvt.f32.bf16 %r79, %rs29; + cvt.f32.bf16 %r80, %rs30; + cvt.f32.bf16 %r81, %rs31; + cvt.f32.bf16 %r82, %rs32; + .loc 1 80 38 // l2norm.py:80:38 + mul.f32 %r83, %r52, %r52; + mul.f32 %r84, %r60, %r60; + mul.f32 %r85, %r68, %r68; + mul.f32 %r86, %r76, %r76; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + fma.rn.f32.bf16 %r87, %rs1, %rs1, %r83; + fma.rn.f32.bf16 %r88, %rs3, %rs3, %r87; + fma.rn.f32.bf16 %r89, %rs4, %rs4, %r88; + fma.rn.f32.bf16 %r90, %rs5, %rs5, %r89; + fma.rn.f32.bf16 %r91, %rs6, %rs6, %r90; + fma.rn.f32.bf16 %r92, %rs7, %rs7, %r91; + fma.rn.f32.bf16 %r93, %rs8, %rs8, %r92; + fma.rn.f32.bf16 %r94, %rs9, %rs9, %r84; + fma.rn.f32.bf16 %r95, %rs11, %rs11, %r94; + fma.rn.f32.bf16 %r96, %rs12, %rs12, %r95; + fma.rn.f32.bf16 %r97, %rs13, %rs13, %r96; + fma.rn.f32.bf16 %r98, %rs14, %rs14, %r97; + fma.rn.f32.bf16 %r99, %rs15, %rs15, %r98; + fma.rn.f32.bf16 %r100, %rs16, %rs16, %r99; + fma.rn.f32.bf16 %r101, %rs17, %rs17, %r85; + fma.rn.f32.bf16 %r102, %rs19, %rs19, %r101; + fma.rn.f32.bf16 %r103, %rs20, %rs20, %r102; + fma.rn.f32.bf16 %r104, %rs21, %rs21, %r103; + fma.rn.f32.bf16 %r105, %rs22, %rs22, %r104; + fma.rn.f32.bf16 %r106, %rs23, %rs23, %r105; + fma.rn.f32.bf16 %r107, %rs24, %rs24, %r106; + fma.rn.f32.bf16 %r108, %rs25, %rs25, %r86; + fma.rn.f32.bf16 %r109, %rs27, %rs27, %r108; + fma.rn.f32.bf16 %r110, %rs28, %rs28, %r109; + fma.rn.f32.bf16 %r111, %rs29, %rs29, %r110; + fma.rn.f32.bf16 %r112, %rs30, %rs30, %r111; + fma.rn.f32.bf16 %r113, %rs31, %rs31, %r112; + fma.rn.f32.bf16 %r114, %rs32, %rs32, %r113; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r115, %r93, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r116, %r93, %r115; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r117, %r116, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r118, %r116, %r117; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r119, %r118, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r120, %r118, %r119; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r121, %r120, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r122, %r120, %r121; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r123, %r100, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r124, %r100, %r123; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r125, %r124, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r126, %r124, %r125; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r127, %r126, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r128, %r126, %r127; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r129, %r128, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r130, %r128, %r129; +$L__tmp18: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r131, %r107, 8, 31, -1; +$L__tmp19: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r132, %r107, %r131; +$L__tmp20: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r133, %r132, 4, 31, -1; +$L__tmp21: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r134, %r132, %r133; +$L__tmp22: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r135, %r134, 2, 31, -1; +$L__tmp23: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r136, %r134, %r135; +$L__tmp24: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r137, %r136, 1, 31, -1; +$L__tmp25: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r138, %r136, %r137; +$L__tmp26: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r139, %r114, 8, 31, -1; +$L__tmp27: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r140, %r114, %r139; +$L__tmp28: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r141, %r140, 4, 31, -1; +$L__tmp29: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r142, %r140, %r141; +$L__tmp30: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r143, %r142, 2, 31, -1; +$L__tmp31: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r144, %r142, %r143; +$L__tmp32: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r145, %r144, 1, 31, -1; +$L__tmp33: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r146, %r144, %r145; +$L__tmp34: + .loc 1 80 48 // l2norm.py:80:48 + add.f32 %r147, %r42, %r122; + add.f32 %r148, %r42, %r130; + add.f32 %r149, %r42, %r138; + add.f32 %r150, %r42, %r146; + .loc 1 80 25 // l2norm.py:80:25 + sqrt.approx.ftz.f32 %r151, %r147; + sqrt.approx.ftz.f32 %r152, %r148; + sqrt.approx.ftz.f32 %r153, %r149; + sqrt.approx.ftz.f32 %r154, %r150; + .loc 1 71 34 // l2norm.py:71:34 + and.b32 %r155, %r43, 63; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd32, %r155; + or.b64 %rd33, %rd13, %rd32; + .loc 1 77 20 // l2norm.py:77:20 + shl.b64 %rd34, %rd33, 2; + add.s64 %rd9, %rd12, %rd34; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd35, %rd11, %rd29; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd8, %rd35, %rd31; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd36, %rd11, %rd27; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd7, %rd36, %rd31; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd37, %rd11, %rd25; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd6, %rd37, %rd31; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd38, %rd11, %rd23; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd5, %rd38, %rd31; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p6, %rd33, %rd14; + mov.b32 %r156, 0f3F800000; + .loc 1 80 17 // l2norm.py:80:17 + div.full.f32 %r35, %r156, %r151; + div.full.f32 %r36, %r156, %r152; + div.full.f32 %r38, %r156, %r153; + div.full.f32 %r39, %r156, %r154; + .loc 1 81 16 // l2norm.py:81:16 + mul.f32 %r157, %r35, %r51; + mul.f32 %r158, %r35, %r52; + mul.f32 %r159, %r35, %r53; + mul.f32 %r160, %r35, %r54; + mul.f32 %r161, %r35, %r55; + mul.f32 %r162, %r35, %r56; + mul.f32 %r163, %r35, %r57; + mul.f32 %r164, %r35, %r58; + mul.f32 %r165, %r36, %r59; + mul.f32 %r166, %r36, %r60; + mul.f32 %r167, %r36, %r61; + mul.f32 %r168, %r36, %r62; + mul.f32 %r169, %r36, %r63; + mul.f32 %r170, %r36, %r64; + mul.f32 %r171, %r36, %r65; + mul.f32 %r172, %r36, %r66; + mul.f32 %r173, %r38, %r67; + mul.f32 %r174, %r38, %r68; + mul.f32 %r175, %r38, %r69; + mul.f32 %r176, %r38, %r70; + mul.f32 %r177, %r38, %r71; + mul.f32 %r178, %r38, %r72; + mul.f32 %r179, %r38, %r73; + mul.f32 %r180, %r38, %r74; + mul.f32 %r181, %r39, %r75; + mul.f32 %r182, %r39, %r76; + mul.f32 %r183, %r39, %r77; + mul.f32 %r184, %r39, %r78; + mul.f32 %r185, %r39, %r79; + mul.f32 %r186, %r39, %r80; + mul.f32 %r187, %r39, %r81; + mul.f32 %r188, %r39, %r82; + .loc 1 83 18 // l2norm.py:83:18 + cvt.rn.bf16x2.f32 %r18, %r158, %r157; + cvt.rn.bf16x2.f32 %r19, %r160, %r159; + cvt.rn.bf16x2.f32 %r20, %r162, %r161; + cvt.rn.bf16x2.f32 %r21, %r164, %r163; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd5 + 0 ], { %r18, %r19, %r20, %r21 }; + // end inline asm + cvt.rn.bf16x2.f32 %r22, %r166, %r165; + cvt.rn.bf16x2.f32 %r23, %r168, %r167; + cvt.rn.bf16x2.f32 %r24, %r170, %r169; + cvt.rn.bf16x2.f32 %r25, %r172, %r171; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd6 + 0 ], { %r22, %r23, %r24, %r25 }; + // end inline asm + cvt.rn.bf16x2.f32 %r26, %r174, %r173; + cvt.rn.bf16x2.f32 %r27, %r176, %r175; + cvt.rn.bf16x2.f32 %r28, %r178, %r177; + cvt.rn.bf16x2.f32 %r29, %r180, %r179; + // begin inline asm + @%p3 st.global.v4.b32 [ %rd7 + 0 ], { %r26, %r27, %r28, %r29 }; + // end inline asm + cvt.rn.bf16x2.f32 %r30, %r182, %r181; + cvt.rn.bf16x2.f32 %r31, %r184, %r183; + cvt.rn.bf16x2.f32 %r32, %r186, %r185; + cvt.rn.bf16x2.f32 %r33, %r188, %r187; + // begin inline asm + @%p4 st.global.v4.b32 [ %rd8 + 0 ], { %r30, %r31, %r32, %r33 }; + // end inline asm + .loc 1 84 21 // l2norm.py:84:21 + shr.u32 %r189, %r44, 1; + mov.b32 %r190, global_smem; + add.s32 %r34, %r190, %r189; + // begin inline asm + st.shared.v2.b32 [ %r34 + 0 ], { %r35, %r36 }; + // end inline asm + add.s32 %r37, %r34, 128; + // begin inline asm + st.shared.v2.b32 [ %r37 + 0 ], { %r38, %r39 }; + // end inline asm + bar.sync 0; + shr.u32 %r191, %r43, 2; + and.b32 %r192, %r191, 4; + shl.b32 %r193, %r43, 2; + and.b32 %r194, %r193, 128; + add.s32 %r195, %r190, %r192; + add.s32 %r196, %r195, %r194; + add.s32 %r197, %r196, %r50; + ld.shared.b32 %r40, [%r197]; + and.b32 %r198, %r43, 192; + setp.eq.b32 %p7, %r198, 0; + and.pred %p5, %p7, %p6; + // begin inline asm + @%p5 st.global.b32 [ %rd9 + 0 ], { %r40 }; + // end inline asm + .loc 1 84 4 // l2norm.py:84:4 + ret; +$L__tmp35: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp34 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 80 // DW_AT_call_line +.b8 43 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp34 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/l2norm_fwd_kernel.source b/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/l2norm_fwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..b5000c470f5e0683aea9e8045b2d0a8c059ea3d1 --- /dev/null +++ b/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/l2norm_fwd_kernel.source @@ -0,0 +1,172 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc38 = loc(unknown) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc45 = loc("x"(#loc)) +#loc46 = loc("y"(#loc)) +#loc47 = loc("rstd"(#loc)) +#loc48 = loc("eps"(#loc)) +#loc49 = loc("T"(#loc)) +#loc81 = loc("input"(#loc36)) +#loc82 = loc("a"(#loc41)) +#loc83 = loc("b"(#loc41)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc50) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc51) + %o_t = arith.constant 64 : i32 loc(#loc52) + %o_t_1 = arith.constant 64 : i64 loc(#loc52) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc52) + %o_t_3 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc53) + %o_t_4 = arith.extsi %o_t_3 : tensor<64xi32> to tensor<64xi64> loc(#loc54) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<64xi64> loc(#loc54) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<64xi64> loc(#loc54) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc55) + %m_t = arith.extsi %T : i32 to i64 loc(#loc56) + %m_t_7 = tt.splat %m_t : i64 -> tensor<64xi64> loc(#loc56) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<64xi64> loc(#loc56) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc57) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc58) + %m_x_10 = arith.constant 128 : i32 loc(#loc59) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc59) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc59) + %m_x_13 = tt.broadcast %m_x : tensor<64x1xi1> -> tensor<64x128xi1> loc(#loc60) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<64x128xi1> loc(#loc60) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<64x128xi1> loc(#loc60) + %p_x = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc61) + %p_x_16 = arith.constant 128 : i32 loc(#loc62) + %p_x_17 = arith.constant 128 : i64 loc(#loc62) + %p_x_18 = arith.constant dense<128> : tensor<64x1xi64> loc(#loc62) + %p_x_19 = arith.muli %p_x, %p_x_18 : tensor<64x1xi64> loc(#loc62) + %p_x_20 = tt.splat %x : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc63) + %p_x_21 = tt.addptr %p_x_20, %p_x_19 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc63) + %p_x_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc64) + %p_x_23 = tt.broadcast %p_x_21 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc65) + %p_x_24 = tt.broadcast %p_x_22 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc65) + %p_x_25 = tt.addptr %p_x_23, %p_x_24 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc65) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc66) + %p_y_26 = arith.constant 128 : i32 loc(#loc67) + %p_y_27 = arith.constant 128 : i64 loc(#loc67) + %p_y_28 = arith.constant dense<128> : tensor<64x1xi64> loc(#loc67) + %p_y_29 = arith.muli %p_y, %p_y_28 : tensor<64x1xi64> loc(#loc67) + %p_y_30 = tt.splat %y : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc68) + %p_y_31 = tt.addptr %p_y_30, %p_y_29 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc68) + %p_y_32 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc69) + %p_y_33 = tt.broadcast %p_y_31 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc70) + %p_y_34 = tt.broadcast %p_y_32 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc70) + %p_y_35 = tt.addptr %p_y_33, %p_y_34 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc70) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<64x!tt.ptr> loc(#loc71) + %p_rstd_36 = tt.addptr %p_rstd, %o_t_6 : tensor<64x!tt.ptr>, tensor<64xi64> loc(#loc71) + %b_x = arith.constant 0.000000e+00 : f32 loc(#loc72) + %b_x_37 = arith.constant dense<0.000000e+00> : tensor<64x128xf32> loc(#loc72) + %b_x_38 = arith.truncf %b_x_37 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc72) + %b_x_39 = tt.load %p_x_25, %m_x_15, %b_x_38 : tensor<64x128x!tt.ptr> loc(#loc72) + %b_x_40 = arith.extf %b_x_39 : tensor<64x128xbf16> to tensor<64x128xf32> loc(#loc73) + %b_rstd = arith.mulf %b_x_40, %b_x_40 : tensor<64x128xf32> loc(#loc74) + %b_rstd_41 = tt.call @triton.language.standard.sum__fp32S64_128S_c1_cFalse_cNone(%b_rstd) : (tensor<64x128xf32>) -> tensor<64xf32> loc(#loc75) + %b_rstd_42 = tt.splat %eps : f32 -> tensor<64xf32> loc(#loc76) + %b_rstd_43 = arith.addf %b_rstd_41, %b_rstd_42 : tensor<64xf32> loc(#loc76) + %b_rstd_44 = math.sqrt %b_rstd_43 : tensor<64xf32> loc(#loc77) + %b_rstd_45 = arith.constant 1 : i32 loc(#loc78) + %b_rstd_46 = arith.constant 1.000000e+00 : f32 loc(#loc78) + %b_rstd_47 = arith.constant dense<1.000000e+00> : tensor<64xf32> loc(#loc78) + %b_rstd_48 = arith.divf %b_rstd_47, %b_rstd_44 : tensor<64xf32> loc(#loc78) + %b_y = tt.expand_dims %b_rstd_48 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc79) + %b_y_49 = tt.broadcast %b_y : tensor<64x1xf32> -> tensor<64x128xf32> loc(#loc80) + %b_y_50 = arith.mulf %b_x_40, %b_y_49 : tensor<64x128xf32> loc(#loc80) + %0 = arith.truncf %b_y_50 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc32) + tt.store %p_y_35, %0, %m_x_15 : tensor<64x128x!tt.ptr> loc(#loc33) + tt.store %p_rstd_36, %b_rstd_48, %m_t_8 : tensor<64x!tt.ptr> loc(#loc34) + tt.return loc(#loc35) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S64_128S_c1_cFalse_cNone(%input: tensor<64x128xf32> loc("input"(#loc36))) -> tensor<64xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc37) + tt.reduce.return %2 : f32 loc(#loc37) + }) : (tensor<64x128xf32>) -> tensor<64xf32> loc(#loc37) + tt.return %0 : tensor<64xf32> loc(#loc39) + ^bb1: // no predecessors + %1 = ub.poison : tensor<64xf32> loc(#loc40) + tt.return %1 : tensor<64xf32> loc(#loc40) + } loc(#loc36) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc41)), %b: f32 loc("b"(#loc41))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc42) + tt.return %0 : f32 loc(#loc43) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc44) + tt.return %1 : f32 loc(#loc44) + } loc(#loc41) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:18) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:29) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:37) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc50 = loc("i_t"(#loc1)) +#loc51 = loc("i_t"(#loc2)) +#loc52 = loc("o_t"(#loc3)) +#loc53 = loc("o_t"(#loc4)) +#loc54 = loc("o_t"(#loc5)) +#loc55 = loc("o_d"(#loc6)) +#loc56 = loc("m_t"(#loc7)) +#loc57 = loc("m_x"(#loc8)) +#loc58 = loc("m_x"(#loc9)) +#loc59 = loc("m_x"(#loc10)) +#loc60 = loc("m_x"(#loc11)) +#loc61 = loc("p_x"(#loc12)) +#loc62 = loc("p_x"(#loc13)) +#loc63 = loc("p_x"(#loc14)) +#loc64 = loc("p_x"(#loc15)) +#loc65 = loc("p_x"(#loc16)) +#loc66 = loc("p_y"(#loc17)) +#loc67 = loc("p_y"(#loc18)) +#loc68 = loc("p_y"(#loc19)) +#loc69 = loc("p_y"(#loc20)) +#loc70 = loc("p_y"(#loc21)) +#loc71 = loc("p_rstd"(#loc22)) +#loc72 = loc("b_x"(#loc23)) +#loc73 = loc("b_x"(#loc24)) +#loc74 = loc("b_rstd"(#loc25)) +#loc75 = loc("b_rstd"(#loc26)) +#loc76 = loc("b_rstd"(#loc27)) +#loc77 = loc("b_rstd"(#loc28)) +#loc78 = loc("b_rstd"(#loc29)) +#loc79 = loc("b_y"(#loc30)) +#loc80 = loc("b_y"(#loc31)) diff --git a/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/l2norm_fwd_kernel.ttgir b/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/l2norm_fwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..33748d09ca551567f81a51bf937094e585232745 --- /dev/null +++ b/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/l2norm_fwd_kernel.ttgir @@ -0,0 +1,135 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [8, 1], order = [1, 0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [8], order = [0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc1 = loc(unknown) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc34 = loc("x"(#loc)) +#loc35 = loc("y"(#loc)) +#loc36 = loc("rstd"(#loc)) +#loc37 = loc("eps"(#loc)) +#loc38 = loc("T"(#loc)) +#loc59 = loc("b_rstd"(#loc23)) +#loc66 = loc(callsite(#loc1 at #loc59)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 8 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<1.000000e+00> : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<64x128xbf16, #blocked> loc(#loc1) + %c64_i64 = arith.constant 64 : i64 loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<64x1xi64, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc39) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc40) + %o_t = arith.muli %i_t_3, %c64_i64 : i64 loc(#loc41) + %o_t_4 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc42) + %o_t_5 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #blocked1> loc(#loc42) + %o_t_6 = arith.extsi %o_t_4 : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_7 = arith.extsi %o_t_5 : tensor<64xi32, #blocked1> to tensor<64xi64, #blocked1> loc(#loc43) + %o_t_8 = tt.splat %o_t : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_9 = tt.splat %o_t : i64 -> tensor<64xi64, #blocked1> loc(#loc43) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<64xi64, #blocked1> loc(#loc43) + %m_t = arith.extsi %T : i32 to i64 loc(#loc44) + %m_t_12 = tt.splat %m_t : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_13 = tt.splat %m_t : i64 -> tensor<64xi64, #blocked1> loc(#loc44) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<64xi64, #blocked1> loc(#loc44) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<64xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64x1xi1, #blocked> loc(#loc45) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc46) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc46) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_1 : tensor<1x128xi32, #blocked> loc(#loc47) + %m_x_19 = tt.broadcast %m_x : tensor<64x1xi1, #blocked> -> tensor<64x128xi1, #blocked> loc(#loc48) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<64x128xi1, #blocked> loc(#loc48) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<64x128xi1, #blocked> loc(#loc48) + %p_x = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64x1xi64, #blocked> loc(#loc49) + %p_x_22 = arith.muli %p_x, %cst_2 : tensor<64x1xi64, #blocked> loc(#loc50) + %p_x_23 = tt.splat %x : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked> loc(#loc51) + %p_x_24 = tt.addptr %p_x_23, %p_x_22 : tensor<64x1x!tt.ptr, #blocked>, tensor<64x1xi64, #blocked> loc(#loc51) + %p_x_25 = tt.broadcast %p_x_24 : tensor<64x1x!tt.ptr, #blocked> -> tensor<64x128x!tt.ptr, #blocked> loc(#loc52) + %p_x_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<64x128xi32, #blocked> loc(#loc52) + %p_x_27 = tt.addptr %p_x_25, %p_x_26 : tensor<64x128x!tt.ptr, #blocked>, tensor<64x128xi32, #blocked> loc(#loc52) + %p_y = tt.splat %y : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked> loc(#loc53) + %p_y_28 = tt.addptr %p_y, %p_x_22 : tensor<64x1x!tt.ptr, #blocked>, tensor<64x1xi64, #blocked> loc(#loc53) + %p_y_29 = tt.broadcast %p_y_28 : tensor<64x1x!tt.ptr, #blocked> -> tensor<64x128x!tt.ptr, #blocked> loc(#loc54) + %p_y_30 = tt.addptr %p_y_29, %p_x_26 : tensor<64x128x!tt.ptr, #blocked>, tensor<64x128xi32, #blocked> loc(#loc54) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<64x!tt.ptr, #blocked1> loc(#loc55) + %p_rstd_31 = tt.addptr %p_rstd, %o_t_11 : tensor<64x!tt.ptr, #blocked1>, tensor<64xi64, #blocked1> loc(#loc55) + %b_x = tt.load %p_x_27, %m_x_21, %cst_0 : tensor<64x128x!tt.ptr, #blocked> loc(#loc56) + %b_x_32 = arith.extf %b_x : tensor<64x128xbf16, #blocked> to tensor<64x128xf32, #blocked> loc(#loc57) + %b_rstd = arith.mulf %b_x_32, %b_x_32 : tensor<64x128xf32, #blocked> loc(#loc58) + %b_rstd_33 = "tt.reduce"(%b_rstd) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_40: f32 loc(callsite(#loc1 at #loc59)), %b_rstd_41: f32 loc(callsite(#loc1 at #loc59))): + %b_rstd_42 = arith.addf %b_rstd_40, %b_rstd_41 : f32 loc(#loc67) + tt.reduce.return %b_rstd_42 : f32 loc(#loc65) + }) : (tensor<64x128xf32, #blocked>) -> tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc65) + %b_rstd_34 = tt.splat %eps : f32 -> tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_36 = math.sqrt %b_rstd_35 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc61) + %b_rstd_37 = arith.divf %cst, %b_rstd_36 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc62) + %b_y = tt.expand_dims %b_rstd_37 {axis = 1 : i32} : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64x1xf32, #blocked> loc(#loc63) + %b_y_38 = tt.broadcast %b_y : tensor<64x1xf32, #blocked> -> tensor<64x128xf32, #blocked> loc(#loc64) + %b_y_39 = arith.mulf %b_x_32, %b_y_38 : tensor<64x128xf32, #blocked> loc(#loc64) + %0 = arith.truncf %b_y_39 : tensor<64x128xf32, #blocked> to tensor<64x128xbf16, #blocked> loc(#loc30) + tt.store %p_y_30, %0, %m_x_21 : tensor<64x128x!tt.ptr, #blocked> loc(#loc31) + %1 = ttg.convert_layout %b_rstd_37 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64xf32, #blocked1> loc(#loc32) + tt.store %p_rstd_31, %1, %m_t_15 : tensor<64x!tt.ptr, #blocked1> loc(#loc32) + tt.return loc(#loc33) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc39 = loc("i_t"(#loc2)) +#loc40 = loc("i_t"(#loc3)) +#loc41 = loc("o_t"(#loc4)) +#loc42 = loc("o_t"(#loc5)) +#loc43 = loc("o_t"(#loc6)) +#loc44 = loc("m_t"(#loc7)) +#loc45 = loc("m_x"(#loc8)) +#loc46 = loc("m_x"(#loc9)) +#loc47 = loc("m_x"(#loc10)) +#loc48 = loc("m_x"(#loc11)) +#loc49 = loc("p_x"(#loc12)) +#loc50 = loc("p_x"(#loc13)) +#loc51 = loc("p_x"(#loc14)) +#loc52 = loc("p_x"(#loc15)) +#loc53 = loc("p_y"(#loc16)) +#loc54 = loc("p_y"(#loc17)) +#loc55 = loc("p_rstd"(#loc18)) +#loc56 = loc("b_x"(#loc19)) +#loc57 = loc("b_x"(#loc20)) +#loc58 = loc("b_rstd"(#loc21)) +#loc60 = loc("b_rstd"(#loc25)) +#loc61 = loc("b_rstd"(#loc26)) +#loc62 = loc("b_rstd"(#loc27)) +#loc63 = loc("b_y"(#loc28)) +#loc64 = loc("b_y"(#loc29)) +#loc65 = loc(callsite(#loc22 at #loc59)) +#loc67 = loc(callsite(#loc24 at #loc65)) diff --git a/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/l2norm_fwd_kernel.ttir b/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/l2norm_fwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..85ebb8b128ceed91f07ab82f5c0fe10a876f6335 --- /dev/null +++ b/triton/gpu0/C6MXE5BIBP7GRPP24UMWB52SFR4KXP6IL4YGWE57QSPRLO326FBA/l2norm_fwd_kernel.ttir @@ -0,0 +1,128 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc3 = loc(unknown) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc35 = loc("x"(#loc)) +#loc36 = loc("y"(#loc)) +#loc37 = loc("rstd"(#loc)) +#loc38 = loc("eps"(#loc)) +#loc39 = loc("T"(#loc)) +#loc62 = loc("b_rstd"(#loc25)) +#loc68 = loc(callsite(#loc3 at #loc62)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %b_x = arith.constant dense<0.000000e+00> : tensor<64x128xbf16> loc(#loc40) + %b_rstd = arith.constant dense<1.000000e+00> : tensor<64xf32> loc(#loc41) + %cst = arith.constant dense<128> : tensor<64x1xi64> loc(#loc3) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc42) + %o_t = arith.constant 64 : i64 loc(#loc43) + %i_t = tt.get_program_id x : i32 loc(#loc44) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc45) + %o_t_1 = arith.muli %i_t_0, %o_t : i64 loc(#loc43) + %o_t_2 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc46) + %o_t_3 = arith.extsi %o_t_2 : tensor<64xi32> to tensor<64xi64> loc(#loc47) + %o_t_4 = tt.splat %o_t_1 : i64 -> tensor<64xi64> loc(#loc47) + %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<64xi64> loc(#loc47) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc48) + %m_t = arith.extsi %T : i32 to i64 loc(#loc49) + %m_t_6 = tt.splat %m_t : i64 -> tensor<64xi64> loc(#loc49) + %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<64xi64> loc(#loc49) + %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc50) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc51) + %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc42) + %m_x_11 = tt.broadcast %m_x_8 : tensor<64x1xi1> -> tensor<64x128xi1> loc(#loc52) + %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<64x128xi1> loc(#loc52) + %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<64x128xi1> loc(#loc52) + %p_x = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc53) + %p_x_14 = arith.muli %p_x, %cst : tensor<64x1xi64> loc(#loc54) + %p_x_15 = tt.splat %x : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc55) + %p_x_16 = tt.addptr %p_x_15, %p_x_14 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc55) + %p_x_17 = tt.broadcast %p_x_16 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc56) + %p_x_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc56) + %p_x_19 = tt.addptr %p_x_17, %p_x_18 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc56) + %p_y = tt.splat %y : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc57) + %p_y_20 = tt.addptr %p_y, %p_x_14 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc57) + %p_y_21 = tt.broadcast %p_y_20 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc58) + %p_y_22 = tt.addptr %p_y_21, %p_x_18 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<64x!tt.ptr> loc(#loc59) + %p_rstd_23 = tt.addptr %p_rstd, %o_t_5 : tensor<64x!tt.ptr>, tensor<64xi64> loc(#loc59) + %b_x_24 = tt.load %p_x_19, %m_x_13, %b_x : tensor<64x128x!tt.ptr> loc(#loc40) + %b_x_25 = arith.extf %b_x_24 : tensor<64x128xbf16> to tensor<64x128xf32> loc(#loc60) + %b_rstd_26 = arith.mulf %b_x_25, %b_x_25 : tensor<64x128xf32> loc(#loc61) + %b_rstd_27 = "tt.reduce"(%b_rstd_26) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_34: f32 loc(callsite(#loc3 at #loc62)), %b_rstd_35: f32 loc(callsite(#loc3 at #loc62))): + %b_rstd_36 = arith.addf %b_rstd_34, %b_rstd_35 : f32 loc(#loc69) + tt.reduce.return %b_rstd_36 : f32 loc(#loc67) + }) : (tensor<64x128xf32>) -> tensor<64xf32> loc(#loc67) + %b_rstd_28 = tt.splat %eps : f32 -> tensor<64xf32> loc(#loc63) + %b_rstd_29 = arith.addf %b_rstd_27, %b_rstd_28 : tensor<64xf32> loc(#loc63) + %b_rstd_30 = math.sqrt %b_rstd_29 : tensor<64xf32> loc(#loc64) + %b_rstd_31 = arith.divf %b_rstd, %b_rstd_30 : tensor<64xf32> loc(#loc41) + %b_y = tt.expand_dims %b_rstd_31 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc65) + %b_y_32 = tt.broadcast %b_y : tensor<64x1xf32> -> tensor<64x128xf32> loc(#loc66) + %b_y_33 = arith.mulf %b_x_25, %b_y_32 : tensor<64x128xf32> loc(#loc66) + %0 = arith.truncf %b_y_33 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc31) + tt.store %p_y_22, %0, %m_x_13 : tensor<64x128x!tt.ptr> loc(#loc32) + tt.store %p_rstd_23, %b_rstd_31, %m_t_7 : tensor<64x!tt.ptr> loc(#loc33) + tt.return loc(#loc34) + } loc(#loc) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc40 = loc("b_x"(#loc1)) +#loc41 = loc("b_rstd"(#loc2)) +#loc42 = loc("m_x"(#loc4)) +#loc43 = loc("o_t"(#loc5)) +#loc44 = loc("i_t"(#loc6)) +#loc45 = loc("i_t"(#loc7)) +#loc46 = loc("o_t"(#loc8)) +#loc47 = loc("o_t"(#loc9)) +#loc48 = loc("o_d"(#loc10)) +#loc49 = loc("m_t"(#loc11)) +#loc50 = loc("m_x"(#loc12)) +#loc51 = loc("m_x"(#loc13)) +#loc52 = loc("m_x"(#loc14)) +#loc53 = loc("p_x"(#loc15)) +#loc54 = loc("p_x"(#loc16)) +#loc55 = loc("p_x"(#loc17)) +#loc56 = loc("p_x"(#loc18)) +#loc57 = loc("p_y"(#loc19)) +#loc58 = loc("p_y"(#loc20)) +#loc59 = loc("p_rstd"(#loc21)) +#loc60 = loc("b_x"(#loc22)) +#loc61 = loc("b_rstd"(#loc23)) +#loc63 = loc("b_rstd"(#loc27)) +#loc64 = loc("b_rstd"(#loc28)) +#loc65 = loc("b_y"(#loc29)) +#loc66 = loc("b_y"(#loc30)) +#loc67 = loc(callsite(#loc24 at #loc62)) +#loc69 = loc(callsite(#loc26 at #loc67)) diff --git a/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/__grp__l2norm_bwd_kernel.json b/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/__grp__l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..e252c986ea85ac9a245cab1d62a28cdd19ac2f29 --- /dev/null +++ b/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/__grp__l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_bwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/l2norm_bwd_kernel.source", "l2norm_bwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/l2norm_bwd_kernel.ttir", "l2norm_bwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/l2norm_bwd_kernel.ttgir", "l2norm_bwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/l2norm_bwd_kernel.llir", "l2norm_bwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/l2norm_bwd_kernel.ptx", "l2norm_bwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/l2norm_bwd_kernel.cubin", "l2norm_bwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/l2norm_bwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/l2norm_bwd_kernel.cubin b/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/l2norm_bwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..5abac1700bc3a50d7ca178992f23ef07b3874b1e Binary files /dev/null and b/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/l2norm_bwd_kernel.cubin differ diff --git a/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/l2norm_bwd_kernel.json b/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..098af457167699c398f82b7f61b77106858beeb2 --- /dev/null +++ b/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"hash": "17d1f94560f28b98b87fa1221cb503b7b5d82056ef85d6910fd0b2720a15f6d2", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 4, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 0, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_bwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/l2norm_bwd_kernel.llir b/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/l2norm_bwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..6e7fd33243f5e28cf843a40240bbe02e10d29cf8 --- /dev/null +++ b/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/l2norm_bwd_kernel.llir @@ -0,0 +1,377 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_bwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, ptr addrspace(1) %3, float %4, i32 %5, ptr addrspace(1) readnone captures(none) %6, ptr addrspace(1) readnone captures(none) %7) local_unnamed_addr #0 !dbg !4 { + %9 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !12 + %10 = zext nneg i32 %9 to i64, !dbg !13 + %11 = shl nuw nsw i64 %10, 4, !dbg !14 + %12 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !15 + %13 = lshr i32 %12, 4, !dbg !15 + %14 = and i32 %13, 7, !dbg !15 + %15 = or disjoint i32 %14, 8, !dbg !15 + %16 = zext nneg i32 %14 to i64, !dbg !16 + %17 = zext nneg i32 %15 to i64, !dbg !16 + %18 = or disjoint i64 %11, %16, !dbg !16 + %19 = or disjoint i64 %11, %17, !dbg !16 + %20 = sext i32 %5 to i64, !dbg !17 + %21 = icmp slt i64 %18, %20, !dbg !17 + %22 = icmp slt i64 %19, %20, !dbg !17 + %23 = shl nuw nsw i32 %12, 3, !dbg !18 + %24 = and i32 %23, 120, !dbg !18 + %25 = shl nuw nsw i64 %18, 7, !dbg !19 + %26 = shl nuw nsw i64 %19, 7, !dbg !19 + %27 = getelementptr bfloat, ptr addrspace(1) %0, i64 %25, !dbg !20 + %28 = getelementptr bfloat, ptr addrspace(1) %0, i64 %26, !dbg !20 + %29 = zext nneg i32 %24 to i64, !dbg !21 + %30 = getelementptr bfloat, ptr addrspace(1) %27, i64 %29, !dbg !21 + %31 = getelementptr bfloat, ptr addrspace(1) %28, i64 %29, !dbg !21 + %32 = getelementptr float, ptr addrspace(1) %1, i64 %18, !dbg !22 + %33 = getelementptr float, ptr addrspace(1) %1, i64 %19, !dbg !22 + %34 = getelementptr bfloat, ptr addrspace(1) %2, i64 %25, !dbg !23 + %35 = getelementptr bfloat, ptr addrspace(1) %2, i64 %26, !dbg !23 + %36 = getelementptr bfloat, ptr addrspace(1) %34, i64 %29, !dbg !24 + %37 = getelementptr bfloat, ptr addrspace(1) %35, i64 %29, !dbg !24 + %38 = getelementptr bfloat, ptr addrspace(1) %3, i64 %25, !dbg !25 + %39 = getelementptr bfloat, ptr addrspace(1) %3, i64 %26, !dbg !25 + %40 = getelementptr bfloat, ptr addrspace(1) %38, i64 %29, !dbg !26 + %41 = getelementptr bfloat, ptr addrspace(1) %39, i64 %29, !dbg !26 + %42 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %30, i1 %21) #3, !dbg !27 + %43 = extractvalue { i32, i32, i32, i32 } %42, 0, !dbg !27 + %44 = bitcast i32 %43 to <2 x bfloat>, !dbg !27 + %45 = extractvalue { i32, i32, i32, i32 } %42, 1, !dbg !27 + %46 = bitcast i32 %45 to <2 x bfloat>, !dbg !27 + %47 = extractvalue { i32, i32, i32, i32 } %42, 2, !dbg !27 + %48 = bitcast i32 %47 to <2 x bfloat>, !dbg !27 + %49 = extractvalue { i32, i32, i32, i32 } %42, 3, !dbg !27 + %50 = bitcast i32 %49 to <2 x bfloat>, !dbg !27 + %51 = extractelement <2 x bfloat> %44, i64 0, !dbg !27 + %52 = extractelement <2 x bfloat> %44, i64 1, !dbg !27 + %53 = extractelement <2 x bfloat> %46, i64 0, !dbg !27 + %54 = extractelement <2 x bfloat> %46, i64 1, !dbg !27 + %55 = extractelement <2 x bfloat> %48, i64 0, !dbg !27 + %56 = extractelement <2 x bfloat> %48, i64 1, !dbg !27 + %57 = extractelement <2 x bfloat> %50, i64 0, !dbg !27 + %58 = extractelement <2 x bfloat> %50, i64 1, !dbg !27 + %59 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %31, i1 %22) #3, !dbg !27 + %60 = extractvalue { i32, i32, i32, i32 } %59, 0, !dbg !27 + %61 = bitcast i32 %60 to <2 x bfloat>, !dbg !27 + %62 = extractvalue { i32, i32, i32, i32 } %59, 1, !dbg !27 + %63 = bitcast i32 %62 to <2 x bfloat>, !dbg !27 + %64 = extractvalue { i32, i32, i32, i32 } %59, 2, !dbg !27 + %65 = bitcast i32 %64 to <2 x bfloat>, !dbg !27 + %66 = extractvalue { i32, i32, i32, i32 } %59, 3, !dbg !27 + %67 = bitcast i32 %66 to <2 x bfloat>, !dbg !27 + %68 = extractelement <2 x bfloat> %61, i64 0, !dbg !27 + %69 = extractelement <2 x bfloat> %61, i64 1, !dbg !27 + %70 = extractelement <2 x bfloat> %63, i64 0, !dbg !27 + %71 = extractelement <2 x bfloat> %63, i64 1, !dbg !27 + %72 = extractelement <2 x bfloat> %65, i64 0, !dbg !27 + %73 = extractelement <2 x bfloat> %65, i64 1, !dbg !27 + %74 = extractelement <2 x bfloat> %67, i64 0, !dbg !27 + %75 = extractelement <2 x bfloat> %67, i64 1, !dbg !27 + %76 = fpext bfloat %51 to float, !dbg !28 + %77 = fpext bfloat %52 to float, !dbg !28 + %78 = fpext bfloat %53 to float, !dbg !28 + %79 = fpext bfloat %54 to float, !dbg !28 + %80 = fpext bfloat %55 to float, !dbg !28 + %81 = fpext bfloat %56 to float, !dbg !28 + %82 = fpext bfloat %57 to float, !dbg !28 + %83 = fpext bfloat %58 to float, !dbg !28 + %84 = fpext bfloat %68 to float, !dbg !28 + %85 = fpext bfloat %69 to float, !dbg !28 + %86 = fpext bfloat %70 to float, !dbg !28 + %87 = fpext bfloat %71 to float, !dbg !28 + %88 = fpext bfloat %72 to float, !dbg !28 + %89 = fpext bfloat %73 to float, !dbg !28 + %90 = fpext bfloat %74 to float, !dbg !28 + %91 = fpext bfloat %75 to float, !dbg !28 + %92 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %32, i1 %21) #3, !dbg !29 + %93 = bitcast i32 %92 to float, !dbg !29 + %94 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %33, i1 %22) #3, !dbg !29 + %95 = bitcast i32 %94 to float, !dbg !29 + %96 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %36, i1 %21) #3, !dbg !30 + %97 = extractvalue { i32, i32, i32, i32 } %96, 0, !dbg !30 + %98 = bitcast i32 %97 to <2 x bfloat>, !dbg !30 + %99 = extractvalue { i32, i32, i32, i32 } %96, 1, !dbg !30 + %100 = bitcast i32 %99 to <2 x bfloat>, !dbg !30 + %101 = extractvalue { i32, i32, i32, i32 } %96, 2, !dbg !30 + %102 = bitcast i32 %101 to <2 x bfloat>, !dbg !30 + %103 = extractvalue { i32, i32, i32, i32 } %96, 3, !dbg !30 + %104 = bitcast i32 %103 to <2 x bfloat>, !dbg !30 + %105 = extractelement <2 x bfloat> %98, i64 0, !dbg !30 + %106 = extractelement <2 x bfloat> %98, i64 1, !dbg !30 + %107 = extractelement <2 x bfloat> %100, i64 0, !dbg !30 + %108 = extractelement <2 x bfloat> %100, i64 1, !dbg !30 + %109 = extractelement <2 x bfloat> %102, i64 0, !dbg !30 + %110 = extractelement <2 x bfloat> %102, i64 1, !dbg !30 + %111 = extractelement <2 x bfloat> %104, i64 0, !dbg !30 + %112 = extractelement <2 x bfloat> %104, i64 1, !dbg !30 + %113 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %37, i1 %22) #3, !dbg !30 + %114 = extractvalue { i32, i32, i32, i32 } %113, 0, !dbg !30 + %115 = bitcast i32 %114 to <2 x bfloat>, !dbg !30 + %116 = extractvalue { i32, i32, i32, i32 } %113, 1, !dbg !30 + %117 = bitcast i32 %116 to <2 x bfloat>, !dbg !30 + %118 = extractvalue { i32, i32, i32, i32 } %113, 2, !dbg !30 + %119 = bitcast i32 %118 to <2 x bfloat>, !dbg !30 + %120 = extractvalue { i32, i32, i32, i32 } %113, 3, !dbg !30 + %121 = bitcast i32 %120 to <2 x bfloat>, !dbg !30 + %122 = extractelement <2 x bfloat> %115, i64 0, !dbg !30 + %123 = extractelement <2 x bfloat> %115, i64 1, !dbg !30 + %124 = extractelement <2 x bfloat> %117, i64 0, !dbg !30 + %125 = extractelement <2 x bfloat> %117, i64 1, !dbg !30 + %126 = extractelement <2 x bfloat> %119, i64 0, !dbg !30 + %127 = extractelement <2 x bfloat> %119, i64 1, !dbg !30 + %128 = extractelement <2 x bfloat> %121, i64 0, !dbg !30 + %129 = extractelement <2 x bfloat> %121, i64 1, !dbg !30 + %130 = fpext bfloat %105 to float, !dbg !31 + %131 = fpext bfloat %106 to float, !dbg !31 + %132 = fpext bfloat %107 to float, !dbg !31 + %133 = fpext bfloat %108 to float, !dbg !31 + %134 = fpext bfloat %109 to float, !dbg !31 + %135 = fpext bfloat %110 to float, !dbg !31 + %136 = fpext bfloat %111 to float, !dbg !31 + %137 = fpext bfloat %112 to float, !dbg !31 + %138 = fpext bfloat %122 to float, !dbg !31 + %139 = fpext bfloat %123 to float, !dbg !31 + %140 = fpext bfloat %124 to float, !dbg !31 + %141 = fpext bfloat %125 to float, !dbg !31 + %142 = fpext bfloat %126 to float, !dbg !31 + %143 = fpext bfloat %127 to float, !dbg !31 + %144 = fpext bfloat %128 to float, !dbg !31 + %145 = fpext bfloat %129 to float, !dbg !31 + %146 = fmul float %93, %130, !dbg !32 + %147 = fmul float %93, %131, !dbg !32 + %148 = fmul float %93, %132, !dbg !32 + %149 = fmul float %93, %133, !dbg !32 + %150 = fmul float %93, %134, !dbg !32 + %151 = fmul float %93, %135, !dbg !32 + %152 = fmul float %93, %136, !dbg !32 + %153 = fmul float %93, %137, !dbg !32 + %154 = fmul float %95, %138, !dbg !32 + %155 = fmul float %95, %139, !dbg !32 + %156 = fmul float %95, %140, !dbg !32 + %157 = fmul float %95, %141, !dbg !32 + %158 = fmul float %95, %142, !dbg !32 + %159 = fmul float %95, %143, !dbg !32 + %160 = fmul float %95, %144, !dbg !32 + %161 = fmul float %95, %145, !dbg !32 + %162 = fmul float %76, %130, !dbg !33 + %163 = fmul float %77, %131, !dbg !33 + %164 = fmul float %78, %132, !dbg !33 + %165 = fmul float %79, %133, !dbg !33 + %166 = fmul float %80, %134, !dbg !33 + %167 = fmul float %81, %135, !dbg !33 + %168 = fmul float %82, %136, !dbg !33 + %169 = fmul float %83, %137, !dbg !33 + %170 = fmul float %84, %138, !dbg !33 + %171 = fmul float %85, %139, !dbg !33 + %172 = fmul float %86, %140, !dbg !33 + %173 = fmul float %87, %141, !dbg !33 + %174 = fmul float %88, %142, !dbg !33 + %175 = fmul float %89, %143, !dbg !33 + %176 = fmul float %90, %144, !dbg !33 + %177 = fmul float %91, %145, !dbg !33 + %178 = fadd float %162, %163, !dbg !34 + %179 = fadd float %164, %178, !dbg !34 + %180 = fadd float %165, %179, !dbg !34 + %181 = fadd float %166, %180, !dbg !34 + %182 = fadd float %167, %181, !dbg !34 + %183 = fadd float %168, %182, !dbg !34 + %184 = fadd float %169, %183, !dbg !34 + %185 = fadd float %170, %171, !dbg !34 + %186 = fadd float %172, %185, !dbg !34 + %187 = fadd float %173, %186, !dbg !34 + %188 = fadd float %174, %187, !dbg !34 + %189 = fadd float %175, %188, !dbg !34 + %190 = fadd float %176, %189, !dbg !34 + %191 = fadd float %177, %190, !dbg !34 + %192 = bitcast float %184 to i32, !dbg !37 + %193 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %192, i32 8, i32 31), !dbg !37 + %194 = bitcast i32 %193 to float, !dbg !37 + %195 = fadd float %184, %194, !dbg !34 + %196 = bitcast float %195 to i32, !dbg !37 + %197 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %196, i32 4, i32 31), !dbg !37 + %198 = bitcast i32 %197 to float, !dbg !37 + %199 = fadd float %195, %198, !dbg !34 + %200 = bitcast float %199 to i32, !dbg !37 + %201 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %200, i32 2, i32 31), !dbg !37 + %202 = bitcast i32 %201 to float, !dbg !37 + %203 = fadd float %199, %202, !dbg !34 + %204 = bitcast float %203 to i32, !dbg !37 + %205 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %204, i32 1, i32 31), !dbg !37 + %206 = bitcast i32 %205 to float, !dbg !37 + %207 = fadd float %203, %206, !dbg !34 + %208 = bitcast float %191 to i32, !dbg !37 + %209 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %208, i32 8, i32 31), !dbg !37 + %210 = bitcast i32 %209 to float, !dbg !37 + %211 = fadd float %191, %210, !dbg !34 + %212 = bitcast float %211 to i32, !dbg !37 + %213 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %212, i32 4, i32 31), !dbg !37 + %214 = bitcast i32 %213 to float, !dbg !37 + %215 = fadd float %211, %214, !dbg !34 + %216 = bitcast float %215 to i32, !dbg !37 + %217 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %216, i32 2, i32 31), !dbg !37 + %218 = bitcast i32 %217 to float, !dbg !37 + %219 = fadd float %215, %218, !dbg !34 + %220 = bitcast float %219 to i32, !dbg !37 + %221 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %220, i32 1, i32 31), !dbg !37 + %222 = bitcast i32 %221 to float, !dbg !37 + %223 = fadd float %219, %222, !dbg !34 + %224 = fmul float %207, %76, !dbg !40 + %225 = fmul float %207, %77, !dbg !40 + %226 = fmul float %207, %78, !dbg !40 + %227 = fmul float %207, %79, !dbg !40 + %228 = fmul float %207, %80, !dbg !40 + %229 = fmul float %207, %81, !dbg !40 + %230 = fmul float %207, %82, !dbg !40 + %231 = fmul float %207, %83, !dbg !40 + %232 = fmul float %223, %84, !dbg !40 + %233 = fmul float %223, %85, !dbg !40 + %234 = fmul float %223, %86, !dbg !40 + %235 = fmul float %223, %87, !dbg !40 + %236 = fmul float %223, %88, !dbg !40 + %237 = fmul float %223, %89, !dbg !40 + %238 = fmul float %223, %90, !dbg !40 + %239 = fmul float %223, %91, !dbg !40 + %240 = fmul float %224, %93, !dbg !41 + %241 = fmul float %225, %93, !dbg !41 + %242 = fmul float %226, %93, !dbg !41 + %243 = fmul float %227, %93, !dbg !41 + %244 = fmul float %228, %93, !dbg !41 + %245 = fmul float %229, %93, !dbg !41 + %246 = fmul float %230, %93, !dbg !41 + %247 = fmul float %231, %93, !dbg !41 + %248 = fmul float %232, %95, !dbg !41 + %249 = fmul float %233, %95, !dbg !41 + %250 = fmul float %234, %95, !dbg !41 + %251 = fmul float %235, %95, !dbg !41 + %252 = fmul float %236, %95, !dbg !41 + %253 = fmul float %237, %95, !dbg !41 + %254 = fmul float %238, %95, !dbg !41 + %255 = fmul float %239, %95, !dbg !41 + %256 = fsub float %146, %240, !dbg !42 + %257 = fsub float %147, %241, !dbg !42 + %258 = fsub float %148, %242, !dbg !42 + %259 = fsub float %149, %243, !dbg !42 + %260 = fsub float %150, %244, !dbg !42 + %261 = fsub float %151, %245, !dbg !42 + %262 = fsub float %152, %246, !dbg !42 + %263 = fsub float %153, %247, !dbg !42 + %264 = fsub float %154, %248, !dbg !42 + %265 = fsub float %155, %249, !dbg !42 + %266 = fsub float %156, %250, !dbg !42 + %267 = fsub float %157, %251, !dbg !42 + %268 = fsub float %158, %252, !dbg !42 + %269 = fsub float %159, %253, !dbg !42 + %270 = fsub float %160, %254, !dbg !42 + %271 = fsub float %161, %255, !dbg !42 + %272 = fptrunc float %256 to bfloat, !dbg !43 + %273 = fptrunc float %257 to bfloat, !dbg !43 + %274 = fptrunc float %258 to bfloat, !dbg !43 + %275 = fptrunc float %259 to bfloat, !dbg !43 + %276 = fptrunc float %260 to bfloat, !dbg !43 + %277 = fptrunc float %261 to bfloat, !dbg !43 + %278 = fptrunc float %262 to bfloat, !dbg !43 + %279 = fptrunc float %263 to bfloat, !dbg !43 + %280 = fptrunc float %264 to bfloat, !dbg !43 + %281 = fptrunc float %265 to bfloat, !dbg !43 + %282 = fptrunc float %266 to bfloat, !dbg !43 + %283 = fptrunc float %267 to bfloat, !dbg !43 + %284 = fptrunc float %268 to bfloat, !dbg !43 + %285 = fptrunc float %269 to bfloat, !dbg !43 + %286 = fptrunc float %270 to bfloat, !dbg !43 + %287 = fptrunc float %271 to bfloat, !dbg !43 + %288 = insertelement <2 x bfloat> poison, bfloat %272, i64 0, !dbg !44 + %289 = insertelement <2 x bfloat> %288, bfloat %273, i64 1, !dbg !44 + %290 = bitcast <2 x bfloat> %289 to i32, !dbg !44 + %291 = insertelement <2 x bfloat> poison, bfloat %274, i64 0, !dbg !44 + %292 = insertelement <2 x bfloat> %291, bfloat %275, i64 1, !dbg !44 + %293 = bitcast <2 x bfloat> %292 to i32, !dbg !44 + %294 = insertelement <2 x bfloat> poison, bfloat %276, i64 0, !dbg !44 + %295 = insertelement <2 x bfloat> %294, bfloat %277, i64 1, !dbg !44 + %296 = bitcast <2 x bfloat> %295 to i32, !dbg !44 + %297 = insertelement <2 x bfloat> poison, bfloat %278, i64 0, !dbg !44 + %298 = insertelement <2 x bfloat> %297, bfloat %279, i64 1, !dbg !44 + %299 = bitcast <2 x bfloat> %298 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %290, i32 %293, i32 %296, i32 %299, ptr addrspace(1) %40, i1 %21) #3, !dbg !44 + %300 = insertelement <2 x bfloat> poison, bfloat %280, i64 0, !dbg !44 + %301 = insertelement <2 x bfloat> %300, bfloat %281, i64 1, !dbg !44 + %302 = bitcast <2 x bfloat> %301 to i32, !dbg !44 + %303 = insertelement <2 x bfloat> poison, bfloat %282, i64 0, !dbg !44 + %304 = insertelement <2 x bfloat> %303, bfloat %283, i64 1, !dbg !44 + %305 = bitcast <2 x bfloat> %304 to i32, !dbg !44 + %306 = insertelement <2 x bfloat> poison, bfloat %284, i64 0, !dbg !44 + %307 = insertelement <2 x bfloat> %306, bfloat %285, i64 1, !dbg !44 + %308 = bitcast <2 x bfloat> %307 to i32, !dbg !44 + %309 = insertelement <2 x bfloat> poison, bfloat %286, i64 0, !dbg !44 + %310 = insertelement <2 x bfloat> %309, bfloat %287, i64 1, !dbg !44 + %311 = bitcast <2 x bfloat> %310 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %302, i32 %305, i32 %308, i32 %311, ptr addrspace(1) %41, i1 %22) #3, !dbg !44 + ret void, !dbg !45 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +attributes #0 = { nounwind "nvvm.reqntid"="128" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = distinct !DISubprogram(name: "l2norm_bwd_kernel", linkageName: "l2norm_bwd_kernel", scope: !1, file: !1, line: 93, type: !5, scopeLine: 93, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!5 = !DISubroutineType(cc: DW_CC_normal, types: !6) +!6 = !{null, !7, !9, !7, !7, !10, !11, !7, !7} +!7 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !8, size: 64, dwarfAddressSpace: 1) +!8 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!9 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !10, size: 64, dwarfAddressSpace: 1) +!10 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!11 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!12 = !DILocation(line: 94, column: 24, scope: !4) +!13 = !DILocation(line: 94, column: 30, scope: !4) +!14 = !DILocation(line: 95, column: 16, scope: !4) +!15 = !DILocation(line: 95, column: 34, scope: !4) +!16 = !DILocation(line: 95, column: 21, scope: !4) +!17 = !DILocation(line: 97, column: 16, scope: !4) +!18 = !DILocation(line: 98, column: 30, scope: !4) +!19 = !DILocation(line: 99, column: 29, scope: !4) +!20 = !DILocation(line: 99, column: 14, scope: !4) +!21 = !DILocation(line: 99, column: 33, scope: !4) +!22 = !DILocation(line: 100, column: 20, scope: !4) +!23 = !DILocation(line: 101, column: 16, scope: !4) +!24 = !DILocation(line: 101, column: 35, scope: !4) +!25 = !DILocation(line: 102, column: 16, scope: !4) +!26 = !DILocation(line: 102, column: 35, scope: !4) +!27 = !DILocation(line: 104, column: 18, scope: !4) +!28 = !DILocation(line: 104, column: 47, scope: !4) +!29 = !DILocation(line: 105, column: 21, scope: !4) +!30 = !DILocation(line: 106, column: 19, scope: !4) +!31 = !DILocation(line: 106, column: 49, scope: !4) +!32 = !DILocation(line: 107, column: 18, scope: !4) +!33 = !DILocation(line: 107, column: 50, scope: !4) +!34 = !DILocation(line: 263, column: 15, scope: !35, inlinedAt: !37) +!35 = distinct !DILexicalBlockFile(scope: !4, file: !36, discriminator: 0) +!36 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!37 = !DILocation(line: 293, column: 36, scope: !35, inlinedAt: !38) +!38 = !DILocation(line: 107, column: 55, scope: !39) +!39 = distinct !DILexicalBlockFile(scope: !4, file: !1, discriminator: 0) +!40 = !DILocation(line: 107, column: 69, scope: !4) +!41 = !DILocation(line: 107, column: 75, scope: !4) +!42 = !DILocation(line: 107, column: 36, scope: !4) +!43 = !DILocation(line: 108, column: 27, scope: !4) +!44 = !DILocation(line: 108, column: 19, scope: !4) +!45 = !DILocation(line: 108, column: 4, scope: !4) diff --git a/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/l2norm_bwd_kernel.ptx b/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/l2norm_bwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..74726982c6d29274b33b25142bd47cc89c6182a5 --- /dev/null +++ b/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/l2norm_bwd_kernel.ptx @@ -0,0 +1,546 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_bwd_kernel // -- Begin function l2norm_bwd_kernel + // @l2norm_bwd_kernel +.visible .entry l2norm_bwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_2, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_3, + .param .f32 l2norm_bwd_kernel_param_4, + .param .u32 l2norm_bwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_6, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_7 +) +.reqntid 128 +{ + .reg .pred %p<3>; + .reg .b16 %rs<33>; + .reg .b32 %r<162>; + .reg .b64 %rd<29>; + .loc 1 93 0 // l2norm.py:93:0 +$L__func_begin0: + .loc 1 93 0 // l2norm.py:93:0 + +// %bb.0: + ld.param.b64 %rd9, [l2norm_bwd_kernel_param_0]; + ld.param.b64 %rd10, [l2norm_bwd_kernel_param_1]; +$L__tmp0: + .loc 1 94 24 // l2norm.py:94:24 + mov.u32 %r28, %ctaid.x; + ld.param.b64 %rd11, [l2norm_bwd_kernel_param_2]; + .loc 1 95 16 // l2norm.py:95:16 + mul.wide.u32 %rd12, %r28, 16; + ld.param.b64 %rd13, [l2norm_bwd_kernel_param_3]; + .loc 1 95 34 // l2norm.py:95:34 + mov.u32 %r29, %tid.x; + ld.param.s32 %rd14, [l2norm_bwd_kernel_param_5]; + bfe.u32 %r30, %r29, 4, 3; + or.b32 %r31, %r30, 8; + .loc 1 95 21 // l2norm.py:95:21 + cvt.u64.u32 %rd15, %r30; + cvt.u64.u32 %rd16, %r31; + or.b64 %rd17, %rd12, %rd15; + or.b64 %rd18, %rd12, %rd16; + .loc 1 97 16 // l2norm.py:97:16 + setp.lt.s64 %p1, %rd17, %rd14; + setp.lt.s64 %p2, %rd18, %rd14; + .loc 1 98 30 // l2norm.py:98:30 + shl.b32 %r32, %r29, 3; + and.b32 %r33, %r32, 120; + .loc 1 99 14 // l2norm.py:99:14 + shl.b64 %rd19, %rd17, 8; + add.s64 %rd20, %rd9, %rd19; + shl.b64 %rd21, %rd18, 8; + add.s64 %rd22, %rd9, %rd21; + .loc 1 99 33 // l2norm.py:99:33 + mul.wide.u32 %rd23, %r33, 2; + add.s64 %rd1, %rd20, %rd23; + add.s64 %rd2, %rd22, %rd23; + .loc 1 100 20 // l2norm.py:100:20 + shl.b64 %rd24, %rd17, 2; + add.s64 %rd3, %rd10, %rd24; + add.s64 %rd4, %rd3, 32; + .loc 1 101 16 // l2norm.py:101:16 + add.s64 %rd25, %rd11, %rd19; + add.s64 %rd26, %rd11, %rd21; + .loc 1 101 35 // l2norm.py:101:35 + add.s64 %rd5, %rd25, %rd23; + add.s64 %rd6, %rd26, %rd23; + .loc 1 102 16 // l2norm.py:102:16 + add.s64 %rd27, %rd13, %rd19; + add.s64 %rd28, %rd13, %rd21; + .loc 1 102 35 // l2norm.py:102:35 + add.s64 %rd7, %rd27, %rd23; + add.s64 %rd8, %rd28, %rd23; + mov.b32 %r5, 0; + .loc 1 104 18 // l2norm.py:104:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + mov.b32 {%rs5, %rs6}, %r3; + mov.b32 {%rs7, %rs8}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs9, %rs10}, %r6; + mov.b32 {%rs11, %rs12}, %r7; + mov.b32 {%rs13, %rs14}, %r8; + mov.b32 {%rs15, %rs16}, %r9; + .loc 1 104 47 // l2norm.py:104:47 + cvt.f32.bf16 %r34, %rs1; + cvt.f32.bf16 %r35, %rs2; + cvt.f32.bf16 %r36, %rs3; + cvt.f32.bf16 %r37, %rs4; + cvt.f32.bf16 %r38, %rs5; + cvt.f32.bf16 %r39, %rs6; + cvt.f32.bf16 %r40, %rs7; + cvt.f32.bf16 %r41, %rs8; + cvt.f32.bf16 %r42, %rs9; + cvt.f32.bf16 %r43, %rs10; + cvt.f32.bf16 %r44, %rs11; + cvt.f32.bf16 %r45, %rs12; + cvt.f32.bf16 %r46, %rs13; + cvt.f32.bf16 %r47, %rs14; + cvt.f32.bf16 %r48, %rs15; + cvt.f32.bf16 %r49, %rs16; + .loc 1 105 21 // l2norm.py:105:21 + // begin inline asm + mov.u32 %r10, %r5; + @%p1 ld.global.b32 { %r10 }, [ %rd3 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r11, %r5; + @%p2 ld.global.b32 { %r11 }, [ %rd4 + 0 ]; + // end inline asm + .loc 1 106 19 // l2norm.py:106:19 + // begin inline asm + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + @%p1 ld.global.v4.b32 { %r12, %r13, %r14, %r15 }, [ %rd5 + 0 ]; + // end inline asm + mov.b32 {%rs17, %rs18}, %r12; + mov.b32 {%rs19, %rs20}, %r13; + mov.b32 {%rs21, %rs22}, %r14; + mov.b32 {%rs23, %rs24}, %r15; + // begin inline asm + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + mov.u32 %r18, %r5; + mov.u32 %r19, %r5; + @%p2 ld.global.v4.b32 { %r16, %r17, %r18, %r19 }, [ %rd6 + 0 ]; + // end inline asm + mov.b32 {%rs25, %rs26}, %r16; + mov.b32 {%rs27, %rs28}, %r17; + mov.b32 {%rs29, %rs30}, %r18; + mov.b32 {%rs31, %rs32}, %r19; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r50, %rs17; + cvt.f32.bf16 %r51, %rs18; + cvt.f32.bf16 %r52, %rs19; + cvt.f32.bf16 %r53, %rs20; + cvt.f32.bf16 %r54, %rs21; + cvt.f32.bf16 %r55, %rs22; + cvt.f32.bf16 %r56, %rs23; + cvt.f32.bf16 %r57, %rs24; + cvt.f32.bf16 %r58, %rs25; + cvt.f32.bf16 %r59, %rs26; + cvt.f32.bf16 %r60, %rs27; + cvt.f32.bf16 %r61, %rs28; + cvt.f32.bf16 %r62, %rs29; + cvt.f32.bf16 %r63, %rs30; + cvt.f32.bf16 %r64, %rs31; + cvt.f32.bf16 %r65, %rs32; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r66, %r35, %r51; + mul.f32 %r67, %r43, %r59; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r68, %rs1, %rs17, %r66; + fma.rn.f32.bf16 %r69, %rs3, %rs19, %r68; + fma.rn.f32.bf16 %r70, %rs4, %rs20, %r69; + fma.rn.f32.bf16 %r71, %rs5, %rs21, %r70; + fma.rn.f32.bf16 %r72, %rs6, %rs22, %r71; + fma.rn.f32.bf16 %r73, %rs7, %rs23, %r72; + fma.rn.f32.bf16 %r74, %rs8, %rs24, %r73; + fma.rn.f32.bf16 %r75, %rs9, %rs25, %r67; + fma.rn.f32.bf16 %r76, %rs11, %rs27, %r75; + fma.rn.f32.bf16 %r77, %rs12, %rs28, %r76; + fma.rn.f32.bf16 %r78, %rs13, %rs29, %r77; + fma.rn.f32.bf16 %r79, %rs14, %rs30, %r78; + fma.rn.f32.bf16 %r80, %rs15, %rs31, %r79; + fma.rn.f32.bf16 %r81, %rs16, %rs32, %r80; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r82, %r74, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r83, %r74, %r82; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r84, %r83, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r85, %r83, %r84; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r86, %r85, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r87, %r85, %r86; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r88, %r87, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r89, %r87, %r88; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r90, %r81, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r91, %r81, %r90; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r92, %r91, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r93, %r91, %r92; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r94, %r93, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r95, %r93, %r94; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r96, %r95, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r97, %r95, %r96; +$L__tmp18: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r98, %r89, %r34; + mul.f32 %r99, %r89, %r35; + mul.f32 %r100, %r89, %r36; + mul.f32 %r101, %r89, %r37; + mul.f32 %r102, %r89, %r38; + mul.f32 %r103, %r89, %r39; + mul.f32 %r104, %r89, %r40; + mul.f32 %r105, %r89, %r41; + mul.f32 %r106, %r97, %r42; + mul.f32 %r107, %r97, %r43; + mul.f32 %r108, %r97, %r44; + mul.f32 %r109, %r97, %r45; + mul.f32 %r110, %r97, %r46; + mul.f32 %r111, %r97, %r47; + mul.f32 %r112, %r97, %r48; + mul.f32 %r113, %r97, %r49; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r114, %r98, %r10; + mul.f32 %r115, %r99, %r10; + mul.f32 %r116, %r100, %r10; + mul.f32 %r117, %r101, %r10; + mul.f32 %r118, %r102, %r10; + mul.f32 %r119, %r103, %r10; + mul.f32 %r120, %r104, %r10; + mul.f32 %r121, %r105, %r10; + mul.f32 %r122, %r106, %r11; + mul.f32 %r123, %r107, %r11; + mul.f32 %r124, %r108, %r11; + mul.f32 %r125, %r109, %r11; + mul.f32 %r126, %r110, %r11; + mul.f32 %r127, %r111, %r11; + mul.f32 %r128, %r112, %r11; + mul.f32 %r129, %r113, %r11; + .loc 1 107 36 // l2norm.py:107:36 + neg.f32 %r130, %r114; + fma.rn.f32 %r131, %r10, %r50, %r130; + neg.f32 %r132, %r115; + fma.rn.f32 %r133, %r10, %r51, %r132; + neg.f32 %r134, %r116; + fma.rn.f32 %r135, %r10, %r52, %r134; + neg.f32 %r136, %r117; + fma.rn.f32 %r137, %r10, %r53, %r136; + neg.f32 %r138, %r118; + fma.rn.f32 %r139, %r10, %r54, %r138; + neg.f32 %r140, %r119; + fma.rn.f32 %r141, %r10, %r55, %r140; + neg.f32 %r142, %r120; + fma.rn.f32 %r143, %r10, %r56, %r142; + neg.f32 %r144, %r121; + fma.rn.f32 %r145, %r10, %r57, %r144; + neg.f32 %r146, %r122; + fma.rn.f32 %r147, %r11, %r58, %r146; + neg.f32 %r148, %r123; + fma.rn.f32 %r149, %r11, %r59, %r148; + neg.f32 %r150, %r124; + fma.rn.f32 %r151, %r11, %r60, %r150; + neg.f32 %r152, %r125; + fma.rn.f32 %r153, %r11, %r61, %r152; + neg.f32 %r154, %r126; + fma.rn.f32 %r155, %r11, %r62, %r154; + neg.f32 %r156, %r127; + fma.rn.f32 %r157, %r11, %r63, %r156; + neg.f32 %r158, %r128; + fma.rn.f32 %r159, %r11, %r64, %r158; + neg.f32 %r160, %r129; + fma.rn.f32 %r161, %r11, %r65, %r160; + .loc 1 108 19 // l2norm.py:108:19 + cvt.rn.bf16x2.f32 %r20, %r133, %r131; + cvt.rn.bf16x2.f32 %r21, %r137, %r135; + cvt.rn.bf16x2.f32 %r22, %r141, %r139; + cvt.rn.bf16x2.f32 %r23, %r145, %r143; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd7 + 0 ], { %r20, %r21, %r22, %r23 }; + // end inline asm + cvt.rn.bf16x2.f32 %r24, %r149, %r147; + cvt.rn.bf16x2.f32 %r25, %r153, %r151; + cvt.rn.bf16x2.f32 %r26, %r157, %r155; + cvt.rn.bf16x2.f32 %r27, %r161, %r159; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd8 + 0 ], { %r24, %r25, %r26, %r27 }; + // end inline asm + .loc 1 108 4 // l2norm.py:108:4 + ret; +$L__tmp19: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 98 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp18 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 107 // DW_AT_call_line +.b8 55 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp18 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/l2norm_bwd_kernel.source b/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/l2norm_bwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..25926967fd8ce8ce5392a0037c1d84708c2d91fa --- /dev/null +++ b/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/l2norm_bwd_kernel.source @@ -0,0 +1,210 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc47 = loc(unknown) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc54 = loc("y"(#loc)) +#loc55 = loc("rstd"(#loc)) +#loc56 = loc("dy"(#loc)) +#loc57 = loc("dx"(#loc)) +#loc58 = loc("eps"(#loc)) +#loc59 = loc("T"(#loc)) +#loc101 = loc("input"(#loc45)) +#loc102 = loc("a"(#loc50)) +#loc103 = loc("b"(#loc50)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc60) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc61) + %o_t = arith.constant 16 : i32 loc(#loc62) + %o_t_1 = arith.constant 16 : i64 loc(#loc62) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc62) + %o_t_3 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32> loc(#loc63) + %o_t_4 = arith.extsi %o_t_3 : tensor<16xi32> to tensor<16xi64> loc(#loc64) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<16xi64> loc(#loc64) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<16xi64> loc(#loc64) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc65) + %m_t = arith.extsi %T : i32 to i64 loc(#loc66) + %m_t_7 = tt.splat %m_t : i64 -> tensor<16xi64> loc(#loc66) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<16xi64> loc(#loc66) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc67) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc68) + %m_x_10 = arith.constant 128 : i32 loc(#loc69) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc69) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc69) + %m_x_13 = tt.broadcast %m_x : tensor<16x1xi1> -> tensor<16x128xi1> loc(#loc70) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<16x128xi1> loc(#loc70) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<16x128xi1> loc(#loc70) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc71) + %p_y_16 = arith.constant 128 : i32 loc(#loc72) + %p_y_17 = arith.constant 128 : i64 loc(#loc72) + %p_y_18 = arith.constant dense<128> : tensor<16x1xi64> loc(#loc72) + %p_y_19 = arith.muli %p_y, %p_y_18 : tensor<16x1xi64> loc(#loc72) + %p_y_20 = tt.splat %y : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc73) + %p_y_21 = tt.addptr %p_y_20, %p_y_19 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc73) + %p_y_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc74) + %p_y_23 = tt.broadcast %p_y_21 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc75) + %p_y_24 = tt.broadcast %p_y_22 : tensor<1x128xi32> -> tensor<16x128xi32> loc(#loc75) + %p_y_25 = tt.addptr %p_y_23, %p_y_24 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc75) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc76) + %p_rstd_26 = tt.addptr %p_rstd, %o_t_6 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc76) + %p_dy = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc77) + %p_dy_27 = arith.constant 128 : i32 loc(#loc78) + %p_dy_28 = arith.constant 128 : i64 loc(#loc78) + %p_dy_29 = arith.constant dense<128> : tensor<16x1xi64> loc(#loc78) + %p_dy_30 = arith.muli %p_dy, %p_dy_29 : tensor<16x1xi64> loc(#loc78) + %p_dy_31 = tt.splat %dy : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc79) + %p_dy_32 = tt.addptr %p_dy_31, %p_dy_30 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc79) + %p_dy_33 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc80) + %p_dy_34 = tt.broadcast %p_dy_32 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc81) + %p_dy_35 = tt.broadcast %p_dy_33 : tensor<1x128xi32> -> tensor<16x128xi32> loc(#loc81) + %p_dy_36 = tt.addptr %p_dy_34, %p_dy_35 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc81) + %p_dx = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc82) + %p_dx_37 = arith.constant 128 : i32 loc(#loc83) + %p_dx_38 = arith.constant 128 : i64 loc(#loc83) + %p_dx_39 = arith.constant dense<128> : tensor<16x1xi64> loc(#loc83) + %p_dx_40 = arith.muli %p_dx, %p_dx_39 : tensor<16x1xi64> loc(#loc83) + %p_dx_41 = tt.splat %dx : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc84) + %p_dx_42 = tt.addptr %p_dx_41, %p_dx_40 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc84) + %p_dx_43 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc85) + %p_dx_44 = tt.broadcast %p_dx_42 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc86) + %p_dx_45 = tt.broadcast %p_dx_43 : tensor<1x128xi32> -> tensor<16x128xi32> loc(#loc86) + %p_dx_46 = tt.addptr %p_dx_44, %p_dx_45 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc86) + %b_y = arith.constant 0.000000e+00 : f32 loc(#loc87) + %b_y_47 = arith.constant dense<0.000000e+00> : tensor<16x128xf32> loc(#loc87) + %b_y_48 = arith.truncf %b_y_47 : tensor<16x128xf32> to tensor<16x128xbf16> loc(#loc87) + %b_y_49 = tt.load %p_y_25, %m_x_15, %b_y_48 : tensor<16x128x!tt.ptr> loc(#loc87) + %b_y_50 = arith.extf %b_y_49 : tensor<16x128xbf16> to tensor<16x128xf32> loc(#loc88) + %b_rstd = arith.constant 0.000000e+00 : f32 loc(#loc89) + %b_rstd_51 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc89) + %b_rstd_52 = tt.load %p_rstd_26, %m_t_8, %b_rstd_51 : tensor<16x!tt.ptr> loc(#loc89) + %b_dy = arith.constant 0.000000e+00 : f32 loc(#loc90) + %b_dy_53 = arith.constant dense<0.000000e+00> : tensor<16x128xf32> loc(#loc90) + %b_dy_54 = arith.truncf %b_dy_53 : tensor<16x128xf32> to tensor<16x128xbf16> loc(#loc90) + %b_dy_55 = tt.load %p_dy_36, %m_x_15, %b_dy_54 : tensor<16x128x!tt.ptr> loc(#loc90) + %b_dy_56 = arith.extf %b_dy_55 : tensor<16x128xbf16> to tensor<16x128xf32> loc(#loc91) + %b_dx = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc92) + %b_dx_57 = tt.broadcast %b_dx : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc93) + %b_dx_58 = arith.mulf %b_dy_56, %b_dx_57 : tensor<16x128xf32> loc(#loc93) + %b_dx_59 = arith.mulf %b_dy_56, %b_y_50 : tensor<16x128xf32> loc(#loc94) + %b_dx_60 = tt.call @triton.language.standard.sum__fp32S16_128S_c1_cFalse_cNone(%b_dx_59) : (tensor<16x128xf32>) -> tensor<16xf32> loc(#loc95) + %b_dx_61 = tt.expand_dims %b_dx_60 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc96) + %b_dx_62 = tt.broadcast %b_dx_61 : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc97) + %b_dx_63 = arith.mulf %b_dx_62, %b_y_50 : tensor<16x128xf32> loc(#loc97) + %b_dx_64 = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc98) + %b_dx_65 = tt.broadcast %b_dx_64 : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc99) + %b_dx_66 = arith.mulf %b_dx_63, %b_dx_65 : tensor<16x128xf32> loc(#loc99) + %b_dx_67 = arith.subf %b_dx_58, %b_dx_66 : tensor<16x128xf32> loc(#loc100) + %0 = arith.truncf %b_dx_67 : tensor<16x128xf32> to tensor<16x128xbf16> loc(#loc42) + tt.store %p_dx_46, %0, %m_x_15 : tensor<16x128x!tt.ptr> loc(#loc43) + tt.return loc(#loc44) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S16_128S_c1_cFalse_cNone(%input: tensor<16x128xf32> loc("input"(#loc45))) -> tensor<16xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc46) + tt.reduce.return %2 : f32 loc(#loc46) + }) : (tensor<16x128xf32>) -> tensor<16xf32> loc(#loc46) + tt.return %0 : tensor<16xf32> loc(#loc48) + ^bb1: // no predecessors + %1 = ub.poison : tensor<16xf32> loc(#loc49) + tt.return %1 : tensor<16xf32> loc(#loc49) + } loc(#loc45) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc50)), %b: f32 loc("b"(#loc50))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc51) + tt.return %0 : f32 loc(#loc52) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc53) + tt.return %1 : f32 loc(#loc53) + } loc(#loc50) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:31) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:39) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:20) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:31) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:39) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:82) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc60 = loc("i_t"(#loc1)) +#loc61 = loc("i_t"(#loc2)) +#loc62 = loc("o_t"(#loc3)) +#loc63 = loc("o_t"(#loc4)) +#loc64 = loc("o_t"(#loc5)) +#loc65 = loc("o_d"(#loc6)) +#loc66 = loc("m_t"(#loc7)) +#loc67 = loc("m_x"(#loc8)) +#loc68 = loc("m_x"(#loc9)) +#loc69 = loc("m_x"(#loc10)) +#loc70 = loc("m_x"(#loc11)) +#loc71 = loc("p_y"(#loc12)) +#loc72 = loc("p_y"(#loc13)) +#loc73 = loc("p_y"(#loc14)) +#loc74 = loc("p_y"(#loc15)) +#loc75 = loc("p_y"(#loc16)) +#loc76 = loc("p_rstd"(#loc17)) +#loc77 = loc("p_dy"(#loc18)) +#loc78 = loc("p_dy"(#loc19)) +#loc79 = loc("p_dy"(#loc20)) +#loc80 = loc("p_dy"(#loc21)) +#loc81 = loc("p_dy"(#loc22)) +#loc82 = loc("p_dx"(#loc23)) +#loc83 = loc("p_dx"(#loc24)) +#loc84 = loc("p_dx"(#loc25)) +#loc85 = loc("p_dx"(#loc26)) +#loc86 = loc("p_dx"(#loc27)) +#loc87 = loc("b_y"(#loc28)) +#loc88 = loc("b_y"(#loc29)) +#loc89 = loc("b_rstd"(#loc30)) +#loc90 = loc("b_dy"(#loc31)) +#loc91 = loc("b_dy"(#loc32)) +#loc92 = loc("b_dx"(#loc33)) +#loc93 = loc("b_dx"(#loc34)) +#loc94 = loc("b_dx"(#loc35)) +#loc95 = loc("b_dx"(#loc36)) +#loc96 = loc("b_dx"(#loc37)) +#loc97 = loc("b_dx"(#loc38)) +#loc98 = loc("b_dx"(#loc39)) +#loc99 = loc("b_dx"(#loc40)) +#loc100 = loc("b_dx"(#loc41)) diff --git a/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/l2norm_bwd_kernel.ttgir b/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/l2norm_bwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..8d1c3d0bada8744cf5dc4409499b5f4aef5b1e01 --- /dev/null +++ b/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/l2norm_bwd_kernel.ttgir @@ -0,0 +1,146 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [4, 1], order = [1, 0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc39 = loc("y"(#loc)) +#loc40 = loc("rstd"(#loc)) +#loc41 = loc("dy"(#loc)) +#loc42 = loc("dx"(#loc)) +#loc43 = loc("eps"(#loc)) +#loc44 = loc("T"(#loc)) +#loc72 = loc("b_dx"(#loc30)) +#loc78 = loc(callsite(#loc1 at #loc72)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 4 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %c16_i64 = arith.constant 16 : i64 loc(#loc1) + %cst = arith.constant dense<0.000000e+00> : tensor<16x128xbf16, #blocked> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<16x1xi64, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc45) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc46) + %o_t = arith.muli %i_t_3, %c16_i64 : i64 loc(#loc47) + %o_t_4 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc48) + %o_t_5 = arith.extsi %o_t_4 : tensor<16xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %o_t_6 = tt.splat %o_t : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %o_t_7 = arith.addi %o_t_6, %o_t_5 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %m_t = arith.extsi %T : i32 to i64 loc(#loc50) + %m_t_8 = tt.splat %m_t : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc50) + %m_t_9 = arith.cmpi slt, %o_t_7, %m_t_8 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc50) + %m_x = tt.expand_dims %m_t_9 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi1, #blocked> loc(#loc51) + %m_x_10 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc52) + %m_x_11 = tt.expand_dims %m_x_10 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc52) + %m_x_12 = arith.cmpi slt, %m_x_11, %cst_2 : tensor<1x128xi32, #blocked> loc(#loc53) + %m_x_13 = tt.broadcast %m_x : tensor<16x1xi1, #blocked> -> tensor<16x128xi1, #blocked> loc(#loc54) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1, #blocked> -> tensor<16x128xi1, #blocked> loc(#loc54) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<16x128xi1, #blocked> loc(#loc54) + %p_y = tt.expand_dims %o_t_7 {axis = 1 : i32} : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi64, #blocked> loc(#loc55) + %p_y_16 = arith.muli %p_y, %cst_1 : tensor<16x1xi64, #blocked> loc(#loc56) + %p_y_17 = tt.splat %y : !tt.ptr -> tensor<16x1x!tt.ptr, #blocked> loc(#loc57) + %p_y_18 = tt.addptr %p_y_17, %p_y_16 : tensor<16x1x!tt.ptr, #blocked>, tensor<16x1xi64, #blocked> loc(#loc57) + %p_y_19 = tt.broadcast %p_y_18 : tensor<16x1x!tt.ptr, #blocked> -> tensor<16x128x!tt.ptr, #blocked> loc(#loc58) + %p_y_20 = tt.broadcast %m_x_11 : tensor<1x128xi32, #blocked> -> tensor<16x128xi32, #blocked> loc(#loc58) + %p_y_21 = tt.addptr %p_y_19, %p_y_20 : tensor<16x128x!tt.ptr, #blocked>, tensor<16x128xi32, #blocked> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc59) + %p_rstd_22 = tt.addptr %p_rstd, %o_t_7 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>>, tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc59) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<16x1x!tt.ptr, #blocked> loc(#loc60) + %p_dy_23 = tt.addptr %p_dy, %p_y_16 : tensor<16x1x!tt.ptr, #blocked>, tensor<16x1xi64, #blocked> loc(#loc60) + %p_dy_24 = tt.broadcast %p_dy_23 : tensor<16x1x!tt.ptr, #blocked> -> tensor<16x128x!tt.ptr, #blocked> loc(#loc61) + %p_dy_25 = tt.addptr %p_dy_24, %p_y_20 : tensor<16x128x!tt.ptr, #blocked>, tensor<16x128xi32, #blocked> loc(#loc61) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<16x1x!tt.ptr, #blocked> loc(#loc62) + %p_dx_26 = tt.addptr %p_dx, %p_y_16 : tensor<16x1x!tt.ptr, #blocked>, tensor<16x1xi64, #blocked> loc(#loc62) + %p_dx_27 = tt.broadcast %p_dx_26 : tensor<16x1x!tt.ptr, #blocked> -> tensor<16x128x!tt.ptr, #blocked> loc(#loc63) + %p_dx_28 = tt.addptr %p_dx_27, %p_y_20 : tensor<16x128x!tt.ptr, #blocked>, tensor<16x128xi32, #blocked> loc(#loc63) + %b_y = tt.load %p_y_21, %m_x_15, %cst : tensor<16x128x!tt.ptr, #blocked> loc(#loc64) + %b_y_29 = arith.extf %b_y : tensor<16x128xbf16, #blocked> to tensor<16x128xf32, #blocked> loc(#loc65) + %b_rstd = tt.load %p_rstd_22, %m_t_9, %cst_0 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc66) + %b_dy = tt.load %p_dy_25, %m_x_15, %cst : tensor<16x128x!tt.ptr, #blocked> loc(#loc67) + %b_dy_30 = arith.extf %b_dy : tensor<16x128xbf16, #blocked> to tensor<16x128xf32, #blocked> loc(#loc68) + %b_dx = tt.expand_dims %b_rstd {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xf32, #blocked> loc(#loc69) + %b_dx_31 = tt.broadcast %b_dx : tensor<16x1xf32, #blocked> -> tensor<16x128xf32, #blocked> loc(#loc70) + %b_dx_32 = arith.mulf %b_dy_30, %b_dx_31 : tensor<16x128xf32, #blocked> loc(#loc70) + %b_dx_33 = arith.mulf %b_dy_30, %b_y_29 : tensor<16x128xf32, #blocked> loc(#loc71) + %b_dx_34 = "tt.reduce"(%b_dx_33) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_40: f32 loc(callsite(#loc1 at #loc72)), %b_dx_41: f32 loc(callsite(#loc1 at #loc72))): + %b_dx_42 = arith.addf %b_dx_40, %b_dx_41 : f32 loc(#loc79) + tt.reduce.return %b_dx_42 : f32 loc(#loc77) + }) : (tensor<16x128xf32, #blocked>) -> tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc77) + %b_dx_35 = tt.expand_dims %b_dx_34 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xf32, #blocked> loc(#loc73) + %b_dx_36 = tt.broadcast %b_dx_35 : tensor<16x1xf32, #blocked> -> tensor<16x128xf32, #blocked> loc(#loc74) + %b_dx_37 = arith.mulf %b_dx_36, %b_y_29 : tensor<16x128xf32, #blocked> loc(#loc74) + %b_dx_38 = arith.mulf %b_dx_37, %b_dx_31 : tensor<16x128xf32, #blocked> loc(#loc75) + %b_dx_39 = arith.subf %b_dx_32, %b_dx_38 : tensor<16x128xf32, #blocked> loc(#loc76) + %0 = arith.truncf %b_dx_39 : tensor<16x128xf32, #blocked> to tensor<16x128xbf16, #blocked> loc(#loc36) + tt.store %p_dx_28, %0, %m_x_15 : tensor<16x128x!tt.ptr, #blocked> loc(#loc37) + tt.return loc(#loc38) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc45 = loc("i_t"(#loc2)) +#loc46 = loc("i_t"(#loc3)) +#loc47 = loc("o_t"(#loc4)) +#loc48 = loc("o_t"(#loc5)) +#loc49 = loc("o_t"(#loc6)) +#loc50 = loc("m_t"(#loc7)) +#loc51 = loc("m_x"(#loc8)) +#loc52 = loc("m_x"(#loc9)) +#loc53 = loc("m_x"(#loc10)) +#loc54 = loc("m_x"(#loc11)) +#loc55 = loc("p_y"(#loc12)) +#loc56 = loc("p_y"(#loc13)) +#loc57 = loc("p_y"(#loc14)) +#loc58 = loc("p_y"(#loc15)) +#loc59 = loc("p_rstd"(#loc16)) +#loc60 = loc("p_dy"(#loc17)) +#loc61 = loc("p_dy"(#loc18)) +#loc62 = loc("p_dx"(#loc19)) +#loc63 = loc("p_dx"(#loc20)) +#loc64 = loc("b_y"(#loc21)) +#loc65 = loc("b_y"(#loc22)) +#loc66 = loc("b_rstd"(#loc23)) +#loc67 = loc("b_dy"(#loc24)) +#loc68 = loc("b_dy"(#loc25)) +#loc69 = loc("b_dx"(#loc26)) +#loc70 = loc("b_dx"(#loc27)) +#loc71 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc32)) +#loc74 = loc("b_dx"(#loc33)) +#loc75 = loc("b_dx"(#loc34)) +#loc76 = loc("b_dx"(#loc35)) +#loc77 = loc(callsite(#loc29 at #loc72)) +#loc79 = loc(callsite(#loc31 at #loc77)) diff --git a/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/l2norm_bwd_kernel.ttir b/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/l2norm_bwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..6c29416e2d7c989d3b096ee1e98cbabbdbd07e07 --- /dev/null +++ b/triton/gpu0/C7I7SRLA6KFZROD7UERBZNIDW625QICW56C5NEIP2CZHECQV63JA/l2norm_bwd_kernel.ttir @@ -0,0 +1,147 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc40 = loc("y"(#loc)) +#loc41 = loc("rstd"(#loc)) +#loc42 = loc("dy"(#loc)) +#loc43 = loc("dx"(#loc)) +#loc44 = loc("eps"(#loc)) +#loc45 = loc("T"(#loc)) +#loc74 = loc("b_dx"(#loc31)) +#loc80 = loc(callsite(#loc1 at #loc74)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<0.000000e+00> : tensor<16x128xbf16> loc(#loc1) + %b_rstd = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc46) + %cst_0 = arith.constant dense<128> : tensor<16x1xi64> loc(#loc1) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc47) + %o_t = arith.constant 16 : i64 loc(#loc48) + %i_t = tt.get_program_id x : i32 loc(#loc49) + %i_t_1 = arith.extsi %i_t : i32 to i64 loc(#loc50) + %o_t_2 = arith.muli %i_t_1, %o_t : i64 loc(#loc48) + %o_t_3 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32> loc(#loc51) + %o_t_4 = arith.extsi %o_t_3 : tensor<16xi32> to tensor<16xi64> loc(#loc52) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<16xi64> loc(#loc52) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<16xi64> loc(#loc52) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc53) + %m_t = arith.extsi %T : i32 to i64 loc(#loc54) + %m_t_7 = tt.splat %m_t : i64 -> tensor<16xi64> loc(#loc54) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<16xi64> loc(#loc54) + %m_x_9 = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc55) + %m_x_10 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc56) + %m_x_11 = arith.cmpi slt, %m_x_10, %m_x : tensor<1x128xi32> loc(#loc47) + %m_x_12 = tt.broadcast %m_x_9 : tensor<16x1xi1> -> tensor<16x128xi1> loc(#loc57) + %m_x_13 = tt.broadcast %m_x_11 : tensor<1x128xi1> -> tensor<16x128xi1> loc(#loc57) + %m_x_14 = arith.andi %m_x_12, %m_x_13 : tensor<16x128xi1> loc(#loc57) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc58) + %p_y_15 = arith.muli %p_y, %cst_0 : tensor<16x1xi64> loc(#loc59) + %p_y_16 = tt.splat %y : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc60) + %p_y_17 = tt.addptr %p_y_16, %p_y_15 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc60) + %p_y_18 = tt.broadcast %p_y_17 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc61) + %p_y_19 = tt.broadcast %m_x_10 : tensor<1x128xi32> -> tensor<16x128xi32> loc(#loc61) + %p_y_20 = tt.addptr %p_y_18, %p_y_19 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc61) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc62) + %p_rstd_21 = tt.addptr %p_rstd, %o_t_6 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc62) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc63) + %p_dy_22 = tt.addptr %p_dy, %p_y_15 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc63) + %p_dy_23 = tt.broadcast %p_dy_22 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc64) + %p_dy_24 = tt.addptr %p_dy_23, %p_y_19 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc64) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc65) + %p_dx_25 = tt.addptr %p_dx, %p_y_15 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc65) + %p_dx_26 = tt.broadcast %p_dx_25 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc66) + %p_dx_27 = tt.addptr %p_dx_26, %p_y_19 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc66) + %b_y = tt.load %p_y_20, %m_x_14, %cst : tensor<16x128x!tt.ptr> loc(#loc67) + %b_y_28 = arith.extf %b_y : tensor<16x128xbf16> to tensor<16x128xf32> loc(#loc68) + %b_rstd_29 = tt.load %p_rstd_21, %m_t_8, %b_rstd : tensor<16x!tt.ptr> loc(#loc46) + %b_dy = tt.load %p_dy_24, %m_x_14, %cst : tensor<16x128x!tt.ptr> loc(#loc69) + %b_dy_30 = arith.extf %b_dy : tensor<16x128xbf16> to tensor<16x128xf32> loc(#loc70) + %b_dx = tt.expand_dims %b_rstd_29 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc71) + %b_dx_31 = tt.broadcast %b_dx : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc72) + %b_dx_32 = arith.mulf %b_dy_30, %b_dx_31 : tensor<16x128xf32> loc(#loc72) + %b_dx_33 = arith.mulf %b_dy_30, %b_y_28 : tensor<16x128xf32> loc(#loc73) + %b_dx_34 = "tt.reduce"(%b_dx_33) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_40: f32 loc(callsite(#loc1 at #loc74)), %b_dx_41: f32 loc(callsite(#loc1 at #loc74))): + %b_dx_42 = arith.addf %b_dx_40, %b_dx_41 : f32 loc(#loc81) + tt.reduce.return %b_dx_42 : f32 loc(#loc79) + }) : (tensor<16x128xf32>) -> tensor<16xf32> loc(#loc79) + %b_dx_35 = tt.expand_dims %b_dx_34 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc75) + %b_dx_36 = tt.broadcast %b_dx_35 : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc76) + %b_dx_37 = arith.mulf %b_dx_36, %b_y_28 : tensor<16x128xf32> loc(#loc76) + %b_dx_38 = arith.mulf %b_dx_37, %b_dx_31 : tensor<16x128xf32> loc(#loc77) + %b_dx_39 = arith.subf %b_dx_32, %b_dx_38 : tensor<16x128xf32> loc(#loc78) + %0 = arith.truncf %b_dx_39 : tensor<16x128xf32> to tensor<16x128xbf16> loc(#loc37) + tt.store %p_dx_27, %0, %m_x_14 : tensor<16x128x!tt.ptr> loc(#loc38) + tt.return loc(#loc39) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("b_rstd"(#loc2)) +#loc47 = loc("m_x"(#loc3)) +#loc48 = loc("o_t"(#loc4)) +#loc49 = loc("i_t"(#loc5)) +#loc50 = loc("i_t"(#loc6)) +#loc51 = loc("o_t"(#loc7)) +#loc52 = loc("o_t"(#loc8)) +#loc53 = loc("o_d"(#loc9)) +#loc54 = loc("m_t"(#loc10)) +#loc55 = loc("m_x"(#loc11)) +#loc56 = loc("m_x"(#loc12)) +#loc57 = loc("m_x"(#loc13)) +#loc58 = loc("p_y"(#loc14)) +#loc59 = loc("p_y"(#loc15)) +#loc60 = loc("p_y"(#loc16)) +#loc61 = loc("p_y"(#loc17)) +#loc62 = loc("p_rstd"(#loc18)) +#loc63 = loc("p_dy"(#loc19)) +#loc64 = loc("p_dy"(#loc20)) +#loc65 = loc("p_dx"(#loc21)) +#loc66 = loc("p_dx"(#loc22)) +#loc67 = loc("b_y"(#loc23)) +#loc68 = loc("b_y"(#loc24)) +#loc69 = loc("b_dy"(#loc25)) +#loc70 = loc("b_dy"(#loc26)) +#loc71 = loc("b_dx"(#loc27)) +#loc72 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc29)) +#loc75 = loc("b_dx"(#loc33)) +#loc76 = loc("b_dx"(#loc34)) +#loc77 = loc("b_dx"(#loc35)) +#loc78 = loc("b_dx"(#loc36)) +#loc79 = loc(callsite(#loc30 at #loc74)) +#loc81 = loc(callsite(#loc32 at #loc79)) diff --git a/triton/gpu0/CCMWG5G4FFG3XC5VSCVXJFAYDG2BZ2DPCNUVEN4FNJGTHQ35E6PQ/l2norm_fwd_kernel.autotune.json b/triton/gpu0/CCMWG5G4FFG3XC5VSCVXJFAYDG2BZ2DPCNUVEN4FNJGTHQ35E6PQ/l2norm_fwd_kernel.autotune.json new file mode 100644 index 0000000000000000000000000000000000000000..f49e15d5b569aa16bebb20ed93fec8b8131b92c2 --- /dev/null +++ b/triton/gpu0/CCMWG5G4FFG3XC5VSCVXJFAYDG2BZ2DPCNUVEN4FNJGTHQ35E6PQ/l2norm_fwd_kernel.autotune.json @@ -0,0 +1 @@ +{"key": [128, 7, "torch.bfloat16", "torch.bfloat16", "torch.float32"], "configs_timings": [[{"kwargs": {"BT": 8}, "num_warps": 1, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.04095999896526337, 0.040832001715898514, 0.04102399945259094]], [{"kwargs": {"BT": 16}, "num_warps": 1, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.04092799872159958, 0.040863998234272, 0.04108799993991852]], [{"kwargs": {"BT": 32}, "num_warps": 1, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.04297599941492081, 0.04287999868392944, 0.04303999990224838]], [{"kwargs": {"BT": 64}, "num_warps": 1, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.05116799846291542, 0.05110400170087814, 0.051263999193906784]], [{"kwargs": {"BT": 128}, "num_warps": 1, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.08806400001049042, 0.08796799927949905, 0.09004800021648407]], [{"kwargs": {"BT": 8}, "num_warps": 2, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.04095999896526337, 0.040863998234272, 0.04102399945259094]], [{"kwargs": {"BT": 16}, "num_warps": 2, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.04095999896526337, 0.04089599847793579, 0.04102399945259094]], [{"kwargs": {"BT": 32}, "num_warps": 2, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.04095999896526337, 0.040863998234272, 0.04105599969625473]], [{"kwargs": {"BT": 64}, "num_warps": 2, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.04294399917125702, 0.04102399945259094, 0.04303999990224838]], [{"kwargs": {"BT": 128}, "num_warps": 2, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.05116799846291542, 0.05110400170087814, 0.051231998950242996]], [{"kwargs": {"BT": 8}, "num_warps": 4, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.043007999658584595, 0.04291199892759323, 0.04499199986457825]], [{"kwargs": {"BT": 16}, "num_warps": 4, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.04095999896526337, 0.040832001715898514, 0.04102399945259094]], [{"kwargs": {"BT": 32}, "num_warps": 4, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.04095999896526337, 0.040863998234272, 0.04102399945259094]], [{"kwargs": {"BT": 64}, "num_warps": 4, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.04095999896526337, 0.040863998234272, 0.04294399917125702]], [{"kwargs": {"BT": 128}, "num_warps": 4, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.04294399917125702, 0.042847998440265656, 0.04303999990224838]], [{"kwargs": {"BT": 8}, "num_warps": 8, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.0634239986538887, 0.06335999816656113, 0.06351999938488007]], [{"kwargs": {"BT": 16}, "num_warps": 8, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.045024000108242035, 0.04492799937725067, 0.04508800059556961]], [{"kwargs": {"BT": 32}, "num_warps": 8, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.04092799872159958, 0.040832001715898514, 0.040991999208927155]], [{"kwargs": {"BT": 64}, "num_warps": 8, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.04095999896526337, 0.040863998234272, 0.04102399945259094]], [{"kwargs": {"BT": 128}, "num_warps": 8, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.04092799872159958, 0.040863998234272, 0.04102399945259094]], [{"kwargs": {"BT": 8}, "num_warps": 16, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.12691199779510498, 0.1268479973077774, 0.12700800597667694]], [{"kwargs": {"BT": 16}, "num_warps": 16, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.0655680000782013, 0.06543999910354614, 0.06752000004053116]], [{"kwargs": {"BT": 32}, "num_warps": 16, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.04707200080156326, 0.0469760000705719, 0.04713600128889084]], [{"kwargs": {"BT": 64}, "num_warps": 16, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.04089599847793579, 0.04076800122857094, 0.040991999208927155]], [{"kwargs": {"BT": 128}, "num_warps": 16, "num_ctas": 1, "num_stages": 3, "maxnreg": null, "pre_hook": null, "ir_override": null}, [0.04092799872159958, 0.040863998234272, 0.04102399945259094]]]} \ No newline at end of file diff --git a/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/__grp__l2norm_fwd_kernel.json b/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/__grp__l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..5c4462c5249cc05943d84cb26f29da815f591aed --- /dev/null +++ b/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/__grp__l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_fwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/l2norm_fwd_kernel.source", "l2norm_fwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/l2norm_fwd_kernel.ttir", "l2norm_fwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/l2norm_fwd_kernel.ttgir", "l2norm_fwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/l2norm_fwd_kernel.llir", "l2norm_fwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/l2norm_fwd_kernel.ptx", "l2norm_fwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/l2norm_fwd_kernel.cubin", "l2norm_fwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/l2norm_fwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/l2norm_fwd_kernel.cubin b/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/l2norm_fwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..406d90e7df79f722999ea5b13a6aea3e9b984bb6 Binary files /dev/null and b/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/l2norm_fwd_kernel.cubin differ diff --git a/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/l2norm_fwd_kernel.json b/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..446948fbacca4e42492963ebe82494cb54bf8d25 --- /dev/null +++ b/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"hash": "10c4b816affb2c4a36665611118a31457d85045c69c360a15b0f2fc3949537af", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 2, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 128, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_fwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/l2norm_fwd_kernel.llir b/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/l2norm_fwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..627c0508dc94ce9f7595b0f6ff9a2f82a7e68083 --- /dev/null +++ b/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/l2norm_fwd_kernel.llir @@ -0,0 +1,1143 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 +@.str.2 = private unnamed_addr constant [17 x i8] c"__CUDA_PREC_SQRT\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_fwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, float %3, i32 %4, ptr addrspace(1) readnone captures(none) %5, ptr addrspace(1) readnone captures(none) %6) local_unnamed_addr #0 !dbg !5 { + %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %9 = zext nneg i32 %8 to i64, !dbg !14 + %10 = shl nuw nsw i64 %9, 5, !dbg !15 + %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !16 + %12 = and i32 %11, 48, !dbg !16 + %13 = lshr exact i32 %12, 4, !dbg !16 + %14 = or disjoint i32 %13, 4, !dbg !16 + %15 = or disjoint i32 %13, 8, !dbg !16 + %16 = or disjoint i32 %13, 12, !dbg !16 + %17 = or disjoint i32 %13, 16, !dbg !16 + %18 = or disjoint i32 %13, 20, !dbg !16 + %19 = or disjoint i32 %13, 24, !dbg !16 + %20 = or disjoint i32 %13, 28, !dbg !16 + %21 = zext nneg i32 %13 to i64, !dbg !17 + %22 = zext nneg i32 %14 to i64, !dbg !17 + %23 = zext nneg i32 %15 to i64, !dbg !17 + %24 = zext nneg i32 %16 to i64, !dbg !17 + %25 = zext nneg i32 %17 to i64, !dbg !17 + %26 = zext nneg i32 %18 to i64, !dbg !17 + %27 = zext nneg i32 %19 to i64, !dbg !17 + %28 = zext nneg i32 %20 to i64, !dbg !17 + %29 = or disjoint i64 %10, %21, !dbg !17 + %30 = or disjoint i64 %10, %22, !dbg !17 + %31 = or disjoint i64 %10, %23, !dbg !17 + %32 = or disjoint i64 %10, %24, !dbg !17 + %33 = or disjoint i64 %10, %25, !dbg !17 + %34 = or disjoint i64 %10, %26, !dbg !17 + %35 = or disjoint i64 %10, %27, !dbg !17 + %36 = or disjoint i64 %10, %28, !dbg !17 + %37 = sext i32 %4 to i64, !dbg !18 + %38 = icmp slt i64 %29, %37, !dbg !18 + %39 = icmp slt i64 %30, %37, !dbg !18 + %40 = icmp slt i64 %31, %37, !dbg !18 + %41 = icmp slt i64 %32, %37, !dbg !18 + %42 = icmp slt i64 %33, %37, !dbg !18 + %43 = icmp slt i64 %34, %37, !dbg !18 + %44 = icmp slt i64 %35, %37, !dbg !18 + %45 = icmp slt i64 %36, %37, !dbg !18 + %46 = shl nuw nsw i32 %11, 3, !dbg !19 + %47 = and i32 %46, 120, !dbg !19 + %48 = shl nuw nsw i64 %29, 7, !dbg !20 + %49 = shl nuw nsw i64 %30, 7, !dbg !20 + %50 = shl nuw nsw i64 %31, 7, !dbg !20 + %51 = shl nuw nsw i64 %32, 7, !dbg !20 + %52 = shl nuw nsw i64 %33, 7, !dbg !20 + %53 = shl nuw nsw i64 %34, 7, !dbg !20 + %54 = shl nuw nsw i64 %35, 7, !dbg !20 + %55 = shl nuw nsw i64 %36, 7, !dbg !20 + %56 = getelementptr bfloat, ptr addrspace(1) %0, i64 %48, !dbg !21 + %57 = getelementptr bfloat, ptr addrspace(1) %0, i64 %49, !dbg !21 + %58 = getelementptr bfloat, ptr addrspace(1) %0, i64 %50, !dbg !21 + %59 = getelementptr bfloat, ptr addrspace(1) %0, i64 %51, !dbg !21 + %60 = getelementptr bfloat, ptr addrspace(1) %0, i64 %52, !dbg !21 + %61 = getelementptr bfloat, ptr addrspace(1) %0, i64 %53, !dbg !21 + %62 = getelementptr bfloat, ptr addrspace(1) %0, i64 %54, !dbg !21 + %63 = getelementptr bfloat, ptr addrspace(1) %0, i64 %55, !dbg !21 + %64 = zext nneg i32 %47 to i64, !dbg !22 + %65 = getelementptr bfloat, ptr addrspace(1) %56, i64 %64, !dbg !22 + %66 = getelementptr bfloat, ptr addrspace(1) %57, i64 %64, !dbg !22 + %67 = getelementptr bfloat, ptr addrspace(1) %58, i64 %64, !dbg !22 + %68 = getelementptr bfloat, ptr addrspace(1) %59, i64 %64, !dbg !22 + %69 = getelementptr bfloat, ptr addrspace(1) %60, i64 %64, !dbg !22 + %70 = getelementptr bfloat, ptr addrspace(1) %61, i64 %64, !dbg !22 + %71 = getelementptr bfloat, ptr addrspace(1) %62, i64 %64, !dbg !22 + %72 = getelementptr bfloat, ptr addrspace(1) %63, i64 %64, !dbg !22 + %73 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %65, i1 %38) #6, !dbg !23 + %74 = extractvalue { i32, i32, i32, i32 } %73, 0, !dbg !23 + %75 = bitcast i32 %74 to <2 x bfloat>, !dbg !23 + %76 = extractvalue { i32, i32, i32, i32 } %73, 1, !dbg !23 + %77 = bitcast i32 %76 to <2 x bfloat>, !dbg !23 + %78 = extractvalue { i32, i32, i32, i32 } %73, 2, !dbg !23 + %79 = bitcast i32 %78 to <2 x bfloat>, !dbg !23 + %80 = extractvalue { i32, i32, i32, i32 } %73, 3, !dbg !23 + %81 = bitcast i32 %80 to <2 x bfloat>, !dbg !23 + %82 = extractelement <2 x bfloat> %75, i64 0, !dbg !23 + %83 = extractelement <2 x bfloat> %75, i64 1, !dbg !23 + %84 = extractelement <2 x bfloat> %77, i64 0, !dbg !23 + %85 = extractelement <2 x bfloat> %77, i64 1, !dbg !23 + %86 = extractelement <2 x bfloat> %79, i64 0, !dbg !23 + %87 = extractelement <2 x bfloat> %79, i64 1, !dbg !23 + %88 = extractelement <2 x bfloat> %81, i64 0, !dbg !23 + %89 = extractelement <2 x bfloat> %81, i64 1, !dbg !23 + %90 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %66, i1 %39) #6, !dbg !23 + %91 = extractvalue { i32, i32, i32, i32 } %90, 0, !dbg !23 + %92 = bitcast i32 %91 to <2 x bfloat>, !dbg !23 + %93 = extractvalue { i32, i32, i32, i32 } %90, 1, !dbg !23 + %94 = bitcast i32 %93 to <2 x bfloat>, !dbg !23 + %95 = extractvalue { i32, i32, i32, i32 } %90, 2, !dbg !23 + %96 = bitcast i32 %95 to <2 x bfloat>, !dbg !23 + %97 = extractvalue { i32, i32, i32, i32 } %90, 3, !dbg !23 + %98 = bitcast i32 %97 to <2 x bfloat>, !dbg !23 + %99 = extractelement <2 x bfloat> %92, i64 0, !dbg !23 + %100 = extractelement <2 x bfloat> %92, i64 1, !dbg !23 + %101 = extractelement <2 x bfloat> %94, i64 0, !dbg !23 + %102 = extractelement <2 x bfloat> %94, i64 1, !dbg !23 + %103 = extractelement <2 x bfloat> %96, i64 0, !dbg !23 + %104 = extractelement <2 x bfloat> %96, i64 1, !dbg !23 + %105 = extractelement <2 x bfloat> %98, i64 0, !dbg !23 + %106 = extractelement <2 x bfloat> %98, i64 1, !dbg !23 + %107 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %67, i1 %40) #6, !dbg !23 + %108 = extractvalue { i32, i32, i32, i32 } %107, 0, !dbg !23 + %109 = bitcast i32 %108 to <2 x bfloat>, !dbg !23 + %110 = extractvalue { i32, i32, i32, i32 } %107, 1, !dbg !23 + %111 = bitcast i32 %110 to <2 x bfloat>, !dbg !23 + %112 = extractvalue { i32, i32, i32, i32 } %107, 2, !dbg !23 + %113 = bitcast i32 %112 to <2 x bfloat>, !dbg !23 + %114 = extractvalue { i32, i32, i32, i32 } %107, 3, !dbg !23 + %115 = bitcast i32 %114 to <2 x bfloat>, !dbg !23 + %116 = extractelement <2 x bfloat> %109, i64 0, !dbg !23 + %117 = extractelement <2 x bfloat> %109, i64 1, !dbg !23 + %118 = extractelement <2 x bfloat> %111, i64 0, !dbg !23 + %119 = extractelement <2 x bfloat> %111, i64 1, !dbg !23 + %120 = extractelement <2 x bfloat> %113, i64 0, !dbg !23 + %121 = extractelement <2 x bfloat> %113, i64 1, !dbg !23 + %122 = extractelement <2 x bfloat> %115, i64 0, !dbg !23 + %123 = extractelement <2 x bfloat> %115, i64 1, !dbg !23 + %124 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %68, i1 %41) #6, !dbg !23 + %125 = extractvalue { i32, i32, i32, i32 } %124, 0, !dbg !23 + %126 = bitcast i32 %125 to <2 x bfloat>, !dbg !23 + %127 = extractvalue { i32, i32, i32, i32 } %124, 1, !dbg !23 + %128 = bitcast i32 %127 to <2 x bfloat>, !dbg !23 + %129 = extractvalue { i32, i32, i32, i32 } %124, 2, !dbg !23 + %130 = bitcast i32 %129 to <2 x bfloat>, !dbg !23 + %131 = extractvalue { i32, i32, i32, i32 } %124, 3, !dbg !23 + %132 = bitcast i32 %131 to <2 x bfloat>, !dbg !23 + %133 = extractelement <2 x bfloat> %126, i64 0, !dbg !23 + %134 = extractelement <2 x bfloat> %126, i64 1, !dbg !23 + %135 = extractelement <2 x bfloat> %128, i64 0, !dbg !23 + %136 = extractelement <2 x bfloat> %128, i64 1, !dbg !23 + %137 = extractelement <2 x bfloat> %130, i64 0, !dbg !23 + %138 = extractelement <2 x bfloat> %130, i64 1, !dbg !23 + %139 = extractelement <2 x bfloat> %132, i64 0, !dbg !23 + %140 = extractelement <2 x bfloat> %132, i64 1, !dbg !23 + %141 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %69, i1 %42) #6, !dbg !23 + %142 = extractvalue { i32, i32, i32, i32 } %141, 0, !dbg !23 + %143 = bitcast i32 %142 to <2 x bfloat>, !dbg !23 + %144 = extractvalue { i32, i32, i32, i32 } %141, 1, !dbg !23 + %145 = bitcast i32 %144 to <2 x bfloat>, !dbg !23 + %146 = extractvalue { i32, i32, i32, i32 } %141, 2, !dbg !23 + %147 = bitcast i32 %146 to <2 x bfloat>, !dbg !23 + %148 = extractvalue { i32, i32, i32, i32 } %141, 3, !dbg !23 + %149 = bitcast i32 %148 to <2 x bfloat>, !dbg !23 + %150 = extractelement <2 x bfloat> %143, i64 0, !dbg !23 + %151 = extractelement <2 x bfloat> %143, i64 1, !dbg !23 + %152 = extractelement <2 x bfloat> %145, i64 0, !dbg !23 + %153 = extractelement <2 x bfloat> %145, i64 1, !dbg !23 + %154 = extractelement <2 x bfloat> %147, i64 0, !dbg !23 + %155 = extractelement <2 x bfloat> %147, i64 1, !dbg !23 + %156 = extractelement <2 x bfloat> %149, i64 0, !dbg !23 + %157 = extractelement <2 x bfloat> %149, i64 1, !dbg !23 + %158 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %70, i1 %43) #6, !dbg !23 + %159 = extractvalue { i32, i32, i32, i32 } %158, 0, !dbg !23 + %160 = bitcast i32 %159 to <2 x bfloat>, !dbg !23 + %161 = extractvalue { i32, i32, i32, i32 } %158, 1, !dbg !23 + %162 = bitcast i32 %161 to <2 x bfloat>, !dbg !23 + %163 = extractvalue { i32, i32, i32, i32 } %158, 2, !dbg !23 + %164 = bitcast i32 %163 to <2 x bfloat>, !dbg !23 + %165 = extractvalue { i32, i32, i32, i32 } %158, 3, !dbg !23 + %166 = bitcast i32 %165 to <2 x bfloat>, !dbg !23 + %167 = extractelement <2 x bfloat> %160, i64 0, !dbg !23 + %168 = extractelement <2 x bfloat> %160, i64 1, !dbg !23 + %169 = extractelement <2 x bfloat> %162, i64 0, !dbg !23 + %170 = extractelement <2 x bfloat> %162, i64 1, !dbg !23 + %171 = extractelement <2 x bfloat> %164, i64 0, !dbg !23 + %172 = extractelement <2 x bfloat> %164, i64 1, !dbg !23 + %173 = extractelement <2 x bfloat> %166, i64 0, !dbg !23 + %174 = extractelement <2 x bfloat> %166, i64 1, !dbg !23 + %175 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %71, i1 %44) #6, !dbg !23 + %176 = extractvalue { i32, i32, i32, i32 } %175, 0, !dbg !23 + %177 = bitcast i32 %176 to <2 x bfloat>, !dbg !23 + %178 = extractvalue { i32, i32, i32, i32 } %175, 1, !dbg !23 + %179 = bitcast i32 %178 to <2 x bfloat>, !dbg !23 + %180 = extractvalue { i32, i32, i32, i32 } %175, 2, !dbg !23 + %181 = bitcast i32 %180 to <2 x bfloat>, !dbg !23 + %182 = extractvalue { i32, i32, i32, i32 } %175, 3, !dbg !23 + %183 = bitcast i32 %182 to <2 x bfloat>, !dbg !23 + %184 = extractelement <2 x bfloat> %177, i64 0, !dbg !23 + %185 = extractelement <2 x bfloat> %177, i64 1, !dbg !23 + %186 = extractelement <2 x bfloat> %179, i64 0, !dbg !23 + %187 = extractelement <2 x bfloat> %179, i64 1, !dbg !23 + %188 = extractelement <2 x bfloat> %181, i64 0, !dbg !23 + %189 = extractelement <2 x bfloat> %181, i64 1, !dbg !23 + %190 = extractelement <2 x bfloat> %183, i64 0, !dbg !23 + %191 = extractelement <2 x bfloat> %183, i64 1, !dbg !23 + %192 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %72, i1 %45) #6, !dbg !23 + %193 = extractvalue { i32, i32, i32, i32 } %192, 0, !dbg !23 + %194 = bitcast i32 %193 to <2 x bfloat>, !dbg !23 + %195 = extractvalue { i32, i32, i32, i32 } %192, 1, !dbg !23 + %196 = bitcast i32 %195 to <2 x bfloat>, !dbg !23 + %197 = extractvalue { i32, i32, i32, i32 } %192, 2, !dbg !23 + %198 = bitcast i32 %197 to <2 x bfloat>, !dbg !23 + %199 = extractvalue { i32, i32, i32, i32 } %192, 3, !dbg !23 + %200 = bitcast i32 %199 to <2 x bfloat>, !dbg !23 + %201 = extractelement <2 x bfloat> %194, i64 0, !dbg !23 + %202 = extractelement <2 x bfloat> %194, i64 1, !dbg !23 + %203 = extractelement <2 x bfloat> %196, i64 0, !dbg !23 + %204 = extractelement <2 x bfloat> %196, i64 1, !dbg !23 + %205 = extractelement <2 x bfloat> %198, i64 0, !dbg !23 + %206 = extractelement <2 x bfloat> %198, i64 1, !dbg !23 + %207 = extractelement <2 x bfloat> %200, i64 0, !dbg !23 + %208 = extractelement <2 x bfloat> %200, i64 1, !dbg !23 + %209 = fpext bfloat %82 to float, !dbg !24 + %210 = fpext bfloat %83 to float, !dbg !24 + %211 = fpext bfloat %84 to float, !dbg !24 + %212 = fpext bfloat %85 to float, !dbg !24 + %213 = fpext bfloat %86 to float, !dbg !24 + %214 = fpext bfloat %87 to float, !dbg !24 + %215 = fpext bfloat %88 to float, !dbg !24 + %216 = fpext bfloat %89 to float, !dbg !24 + %217 = fpext bfloat %99 to float, !dbg !24 + %218 = fpext bfloat %100 to float, !dbg !24 + %219 = fpext bfloat %101 to float, !dbg !24 + %220 = fpext bfloat %102 to float, !dbg !24 + %221 = fpext bfloat %103 to float, !dbg !24 + %222 = fpext bfloat %104 to float, !dbg !24 + %223 = fpext bfloat %105 to float, !dbg !24 + %224 = fpext bfloat %106 to float, !dbg !24 + %225 = fpext bfloat %116 to float, !dbg !24 + %226 = fpext bfloat %117 to float, !dbg !24 + %227 = fpext bfloat %118 to float, !dbg !24 + %228 = fpext bfloat %119 to float, !dbg !24 + %229 = fpext bfloat %120 to float, !dbg !24 + %230 = fpext bfloat %121 to float, !dbg !24 + %231 = fpext bfloat %122 to float, !dbg !24 + %232 = fpext bfloat %123 to float, !dbg !24 + %233 = fpext bfloat %133 to float, !dbg !24 + %234 = fpext bfloat %134 to float, !dbg !24 + %235 = fpext bfloat %135 to float, !dbg !24 + %236 = fpext bfloat %136 to float, !dbg !24 + %237 = fpext bfloat %137 to float, !dbg !24 + %238 = fpext bfloat %138 to float, !dbg !24 + %239 = fpext bfloat %139 to float, !dbg !24 + %240 = fpext bfloat %140 to float, !dbg !24 + %241 = fpext bfloat %150 to float, !dbg !24 + %242 = fpext bfloat %151 to float, !dbg !24 + %243 = fpext bfloat %152 to float, !dbg !24 + %244 = fpext bfloat %153 to float, !dbg !24 + %245 = fpext bfloat %154 to float, !dbg !24 + %246 = fpext bfloat %155 to float, !dbg !24 + %247 = fpext bfloat %156 to float, !dbg !24 + %248 = fpext bfloat %157 to float, !dbg !24 + %249 = fpext bfloat %167 to float, !dbg !24 + %250 = fpext bfloat %168 to float, !dbg !24 + %251 = fpext bfloat %169 to float, !dbg !24 + %252 = fpext bfloat %170 to float, !dbg !24 + %253 = fpext bfloat %171 to float, !dbg !24 + %254 = fpext bfloat %172 to float, !dbg !24 + %255 = fpext bfloat %173 to float, !dbg !24 + %256 = fpext bfloat %174 to float, !dbg !24 + %257 = fpext bfloat %184 to float, !dbg !24 + %258 = fpext bfloat %185 to float, !dbg !24 + %259 = fpext bfloat %186 to float, !dbg !24 + %260 = fpext bfloat %187 to float, !dbg !24 + %261 = fpext bfloat %188 to float, !dbg !24 + %262 = fpext bfloat %189 to float, !dbg !24 + %263 = fpext bfloat %190 to float, !dbg !24 + %264 = fpext bfloat %191 to float, !dbg !24 + %265 = fpext bfloat %201 to float, !dbg !24 + %266 = fpext bfloat %202 to float, !dbg !24 + %267 = fpext bfloat %203 to float, !dbg !24 + %268 = fpext bfloat %204 to float, !dbg !24 + %269 = fpext bfloat %205 to float, !dbg !24 + %270 = fpext bfloat %206 to float, !dbg !24 + %271 = fpext bfloat %207 to float, !dbg !24 + %272 = fpext bfloat %208 to float, !dbg !24 + %273 = fmul float %209, %209, !dbg !25 + %274 = fmul float %210, %210, !dbg !25 + %275 = fmul float %211, %211, !dbg !25 + %276 = fmul float %212, %212, !dbg !25 + %277 = fmul float %213, %213, !dbg !25 + %278 = fmul float %214, %214, !dbg !25 + %279 = fmul float %215, %215, !dbg !25 + %280 = fmul float %216, %216, !dbg !25 + %281 = fmul float %217, %217, !dbg !25 + %282 = fmul float %218, %218, !dbg !25 + %283 = fmul float %219, %219, !dbg !25 + %284 = fmul float %220, %220, !dbg !25 + %285 = fmul float %221, %221, !dbg !25 + %286 = fmul float %222, %222, !dbg !25 + %287 = fmul float %223, %223, !dbg !25 + %288 = fmul float %224, %224, !dbg !25 + %289 = fmul float %225, %225, !dbg !25 + %290 = fmul float %226, %226, !dbg !25 + %291 = fmul float %227, %227, !dbg !25 + %292 = fmul float %228, %228, !dbg !25 + %293 = fmul float %229, %229, !dbg !25 + %294 = fmul float %230, %230, !dbg !25 + %295 = fmul float %231, %231, !dbg !25 + %296 = fmul float %232, %232, !dbg !25 + %297 = fmul float %233, %233, !dbg !25 + %298 = fmul float %234, %234, !dbg !25 + %299 = fmul float %235, %235, !dbg !25 + %300 = fmul float %236, %236, !dbg !25 + %301 = fmul float %237, %237, !dbg !25 + %302 = fmul float %238, %238, !dbg !25 + %303 = fmul float %239, %239, !dbg !25 + %304 = fmul float %240, %240, !dbg !25 + %305 = fmul float %241, %241, !dbg !25 + %306 = fmul float %242, %242, !dbg !25 + %307 = fmul float %243, %243, !dbg !25 + %308 = fmul float %244, %244, !dbg !25 + %309 = fmul float %245, %245, !dbg !25 + %310 = fmul float %246, %246, !dbg !25 + %311 = fmul float %247, %247, !dbg !25 + %312 = fmul float %248, %248, !dbg !25 + %313 = fmul float %249, %249, !dbg !25 + %314 = fmul float %250, %250, !dbg !25 + %315 = fmul float %251, %251, !dbg !25 + %316 = fmul float %252, %252, !dbg !25 + %317 = fmul float %253, %253, !dbg !25 + %318 = fmul float %254, %254, !dbg !25 + %319 = fmul float %255, %255, !dbg !25 + %320 = fmul float %256, %256, !dbg !25 + %321 = fmul float %257, %257, !dbg !25 + %322 = fmul float %258, %258, !dbg !25 + %323 = fmul float %259, %259, !dbg !25 + %324 = fmul float %260, %260, !dbg !25 + %325 = fmul float %261, %261, !dbg !25 + %326 = fmul float %262, %262, !dbg !25 + %327 = fmul float %263, %263, !dbg !25 + %328 = fmul float %264, %264, !dbg !25 + %329 = fmul float %265, %265, !dbg !25 + %330 = fmul float %266, %266, !dbg !25 + %331 = fmul float %267, %267, !dbg !25 + %332 = fmul float %268, %268, !dbg !25 + %333 = fmul float %269, %269, !dbg !25 + %334 = fmul float %270, %270, !dbg !25 + %335 = fmul float %271, %271, !dbg !25 + %336 = fmul float %272, %272, !dbg !25 + %337 = fadd float %273, %274, !dbg !26 + %338 = fadd float %275, %337, !dbg !26 + %339 = fadd float %276, %338, !dbg !26 + %340 = fadd float %277, %339, !dbg !26 + %341 = fadd float %278, %340, !dbg !26 + %342 = fadd float %279, %341, !dbg !26 + %343 = fadd float %280, %342, !dbg !26 + %344 = fadd float %281, %282, !dbg !26 + %345 = fadd float %283, %344, !dbg !26 + %346 = fadd float %284, %345, !dbg !26 + %347 = fadd float %285, %346, !dbg !26 + %348 = fadd float %286, %347, !dbg !26 + %349 = fadd float %287, %348, !dbg !26 + %350 = fadd float %288, %349, !dbg !26 + %351 = fadd float %289, %290, !dbg !26 + %352 = fadd float %291, %351, !dbg !26 + %353 = fadd float %292, %352, !dbg !26 + %354 = fadd float %293, %353, !dbg !26 + %355 = fadd float %294, %354, !dbg !26 + %356 = fadd float %295, %355, !dbg !26 + %357 = fadd float %296, %356, !dbg !26 + %358 = fadd float %297, %298, !dbg !26 + %359 = fadd float %299, %358, !dbg !26 + %360 = fadd float %300, %359, !dbg !26 + %361 = fadd float %301, %360, !dbg !26 + %362 = fadd float %302, %361, !dbg !26 + %363 = fadd float %303, %362, !dbg !26 + %364 = fadd float %304, %363, !dbg !26 + %365 = fadd float %305, %306, !dbg !26 + %366 = fadd float %307, %365, !dbg !26 + %367 = fadd float %308, %366, !dbg !26 + %368 = fadd float %309, %367, !dbg !26 + %369 = fadd float %310, %368, !dbg !26 + %370 = fadd float %311, %369, !dbg !26 + %371 = fadd float %312, %370, !dbg !26 + %372 = fadd float %313, %314, !dbg !26 + %373 = fadd float %315, %372, !dbg !26 + %374 = fadd float %316, %373, !dbg !26 + %375 = fadd float %317, %374, !dbg !26 + %376 = fadd float %318, %375, !dbg !26 + %377 = fadd float %319, %376, !dbg !26 + %378 = fadd float %320, %377, !dbg !26 + %379 = fadd float %321, %322, !dbg !26 + %380 = fadd float %323, %379, !dbg !26 + %381 = fadd float %324, %380, !dbg !26 + %382 = fadd float %325, %381, !dbg !26 + %383 = fadd float %326, %382, !dbg !26 + %384 = fadd float %327, %383, !dbg !26 + %385 = fadd float %328, %384, !dbg !26 + %386 = fadd float %329, %330, !dbg !26 + %387 = fadd float %331, %386, !dbg !26 + %388 = fadd float %332, %387, !dbg !26 + %389 = fadd float %333, %388, !dbg !26 + %390 = fadd float %334, %389, !dbg !26 + %391 = fadd float %335, %390, !dbg !26 + %392 = fadd float %336, %391, !dbg !26 + %393 = bitcast float %343 to i32, !dbg !29 + %394 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %393, i32 8, i32 31), !dbg !29 + %395 = bitcast i32 %394 to float, !dbg !29 + %396 = fadd float %343, %395, !dbg !26 + %397 = bitcast float %396 to i32, !dbg !29 + %398 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %397, i32 4, i32 31), !dbg !29 + %399 = bitcast i32 %398 to float, !dbg !29 + %400 = fadd float %396, %399, !dbg !26 + %401 = bitcast float %400 to i32, !dbg !29 + %402 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %401, i32 2, i32 31), !dbg !29 + %403 = bitcast i32 %402 to float, !dbg !29 + %404 = fadd float %400, %403, !dbg !26 + %405 = bitcast float %404 to i32, !dbg !29 + %406 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %405, i32 1, i32 31), !dbg !29 + %407 = bitcast i32 %406 to float, !dbg !29 + %408 = fadd float %404, %407, !dbg !26 + %409 = bitcast float %350 to i32, !dbg !29 + %410 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %409, i32 8, i32 31), !dbg !29 + %411 = bitcast i32 %410 to float, !dbg !29 + %412 = fadd float %350, %411, !dbg !26 + %413 = bitcast float %412 to i32, !dbg !29 + %414 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %413, i32 4, i32 31), !dbg !29 + %415 = bitcast i32 %414 to float, !dbg !29 + %416 = fadd float %412, %415, !dbg !26 + %417 = bitcast float %416 to i32, !dbg !29 + %418 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %417, i32 2, i32 31), !dbg !29 + %419 = bitcast i32 %418 to float, !dbg !29 + %420 = fadd float %416, %419, !dbg !26 + %421 = bitcast float %420 to i32, !dbg !29 + %422 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %421, i32 1, i32 31), !dbg !29 + %423 = bitcast i32 %422 to float, !dbg !29 + %424 = fadd float %420, %423, !dbg !26 + %425 = bitcast float %357 to i32, !dbg !29 + %426 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %425, i32 8, i32 31), !dbg !29 + %427 = bitcast i32 %426 to float, !dbg !29 + %428 = fadd float %357, %427, !dbg !26 + %429 = bitcast float %428 to i32, !dbg !29 + %430 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %429, i32 4, i32 31), !dbg !29 + %431 = bitcast i32 %430 to float, !dbg !29 + %432 = fadd float %428, %431, !dbg !26 + %433 = bitcast float %432 to i32, !dbg !29 + %434 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %433, i32 2, i32 31), !dbg !29 + %435 = bitcast i32 %434 to float, !dbg !29 + %436 = fadd float %432, %435, !dbg !26 + %437 = bitcast float %436 to i32, !dbg !29 + %438 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %437, i32 1, i32 31), !dbg !29 + %439 = bitcast i32 %438 to float, !dbg !29 + %440 = fadd float %436, %439, !dbg !26 + %441 = bitcast float %364 to i32, !dbg !29 + %442 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %441, i32 8, i32 31), !dbg !29 + %443 = bitcast i32 %442 to float, !dbg !29 + %444 = fadd float %364, %443, !dbg !26 + %445 = bitcast float %444 to i32, !dbg !29 + %446 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %445, i32 4, i32 31), !dbg !29 + %447 = bitcast i32 %446 to float, !dbg !29 + %448 = fadd float %444, %447, !dbg !26 + %449 = bitcast float %448 to i32, !dbg !29 + %450 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %449, i32 2, i32 31), !dbg !29 + %451 = bitcast i32 %450 to float, !dbg !29 + %452 = fadd float %448, %451, !dbg !26 + %453 = bitcast float %452 to i32, !dbg !29 + %454 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %453, i32 1, i32 31), !dbg !29 + %455 = bitcast i32 %454 to float, !dbg !29 + %456 = fadd float %452, %455, !dbg !26 + %457 = bitcast float %371 to i32, !dbg !29 + %458 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %457, i32 8, i32 31), !dbg !29 + %459 = bitcast i32 %458 to float, !dbg !29 + %460 = fadd float %371, %459, !dbg !26 + %461 = bitcast float %460 to i32, !dbg !29 + %462 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %461, i32 4, i32 31), !dbg !29 + %463 = bitcast i32 %462 to float, !dbg !29 + %464 = fadd float %460, %463, !dbg !26 + %465 = bitcast float %464 to i32, !dbg !29 + %466 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %465, i32 2, i32 31), !dbg !29 + %467 = bitcast i32 %466 to float, !dbg !29 + %468 = fadd float %464, %467, !dbg !26 + %469 = bitcast float %468 to i32, !dbg !29 + %470 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %469, i32 1, i32 31), !dbg !29 + %471 = bitcast i32 %470 to float, !dbg !29 + %472 = fadd float %468, %471, !dbg !26 + %473 = bitcast float %378 to i32, !dbg !29 + %474 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %473, i32 8, i32 31), !dbg !29 + %475 = bitcast i32 %474 to float, !dbg !29 + %476 = fadd float %378, %475, !dbg !26 + %477 = bitcast float %476 to i32, !dbg !29 + %478 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %477, i32 4, i32 31), !dbg !29 + %479 = bitcast i32 %478 to float, !dbg !29 + %480 = fadd float %476, %479, !dbg !26 + %481 = bitcast float %480 to i32, !dbg !29 + %482 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %481, i32 2, i32 31), !dbg !29 + %483 = bitcast i32 %482 to float, !dbg !29 + %484 = fadd float %480, %483, !dbg !26 + %485 = bitcast float %484 to i32, !dbg !29 + %486 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %485, i32 1, i32 31), !dbg !29 + %487 = bitcast i32 %486 to float, !dbg !29 + %488 = fadd float %484, %487, !dbg !26 + %489 = bitcast float %385 to i32, !dbg !29 + %490 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %489, i32 8, i32 31), !dbg !29 + %491 = bitcast i32 %490 to float, !dbg !29 + %492 = fadd float %385, %491, !dbg !26 + %493 = bitcast float %492 to i32, !dbg !29 + %494 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %493, i32 4, i32 31), !dbg !29 + %495 = bitcast i32 %494 to float, !dbg !29 + %496 = fadd float %492, %495, !dbg !26 + %497 = bitcast float %496 to i32, !dbg !29 + %498 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %497, i32 2, i32 31), !dbg !29 + %499 = bitcast i32 %498 to float, !dbg !29 + %500 = fadd float %496, %499, !dbg !26 + %501 = bitcast float %500 to i32, !dbg !29 + %502 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %501, i32 1, i32 31), !dbg !29 + %503 = bitcast i32 %502 to float, !dbg !29 + %504 = fadd float %500, %503, !dbg !26 + %505 = bitcast float %392 to i32, !dbg !29 + %506 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %505, i32 8, i32 31), !dbg !29 + %507 = bitcast i32 %506 to float, !dbg !29 + %508 = fadd float %392, %507, !dbg !26 + %509 = bitcast float %508 to i32, !dbg !29 + %510 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %509, i32 4, i32 31), !dbg !29 + %511 = bitcast i32 %510 to float, !dbg !29 + %512 = fadd float %508, %511, !dbg !26 + %513 = bitcast float %512 to i32, !dbg !29 + %514 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %513, i32 2, i32 31), !dbg !29 + %515 = bitcast i32 %514 to float, !dbg !29 + %516 = fadd float %512, %515, !dbg !26 + %517 = bitcast float %516 to i32, !dbg !29 + %518 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %517, i32 1, i32 31), !dbg !29 + %519 = bitcast i32 %518 to float, !dbg !29 + %520 = fadd float %516, %519, !dbg !26 + %521 = fadd float %3, %408, !dbg !32 + %522 = fadd float %3, %424, !dbg !32 + %523 = fadd float %3, %440, !dbg !32 + %524 = fadd float %3, %456, !dbg !32 + %525 = fadd float %3, %472, !dbg !32 + %526 = fadd float %3, %488, !dbg !32 + %527 = fadd float %3, %504, !dbg !32 + %528 = fadd float %3, %520, !dbg !32 + %529 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i = icmp eq i32 %529, 0, !dbg !33 + %530 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i = icmp eq i32 %530, 0, !dbg !33 + br i1 %.not.i, label %536, label %531, !dbg !33 + +531: ; preds = %7 + br i1 %.not1.i, label %534, label %532, !dbg !33 + +532: ; preds = %531 + %533 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %521) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +534: ; preds = %531 + %535 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %521) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +536: ; preds = %7 + br i1 %.not1.i, label %539, label %537, !dbg !33 + +537: ; preds = %536 + %538 = tail call float @llvm.nvvm.sqrt.rn.f(float %521) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +539: ; preds = %536 + %540 = tail call float @llvm.nvvm.sqrt.approx.f(float %521) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +__nv_sqrtf.exit: ; preds = %532, %534, %537, %539 + %.0.i = phi float [ %533, %532 ], [ %535, %534 ], [ %538, %537 ], [ %540, %539 ], !dbg !33 + %541 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i1 = icmp eq i32 %541, 0, !dbg !33 + %542 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i4 = icmp eq i32 %542, 0, !dbg !33 + br i1 %.not.i1, label %548, label %543, !dbg !33 + +543: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %546, label %544, !dbg !33 + +544: ; preds = %543 + %545 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %522) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +546: ; preds = %543 + %547 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %522) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +548: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %551, label %549, !dbg !33 + +549: ; preds = %548 + %550 = tail call float @llvm.nvvm.sqrt.rn.f(float %522) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +551: ; preds = %548 + %552 = tail call float @llvm.nvvm.sqrt.approx.f(float %522) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +__nv_sqrtf.exit5: ; preds = %544, %546, %549, %551 + %.0.i3 = phi float [ %545, %544 ], [ %547, %546 ], [ %550, %549 ], [ %552, %551 ], !dbg !33 + %553 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i6 = icmp eq i32 %553, 0, !dbg !33 + %554 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i9 = icmp eq i32 %554, 0, !dbg !33 + br i1 %.not.i6, label %560, label %555, !dbg !33 + +555: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %558, label %556, !dbg !33 + +556: ; preds = %555 + %557 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %523) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +558: ; preds = %555 + %559 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %523) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +560: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %563, label %561, !dbg !33 + +561: ; preds = %560 + %562 = tail call float @llvm.nvvm.sqrt.rn.f(float %523) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +563: ; preds = %560 + %564 = tail call float @llvm.nvvm.sqrt.approx.f(float %523) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +__nv_sqrtf.exit10: ; preds = %556, %558, %561, %563 + %.0.i8 = phi float [ %557, %556 ], [ %559, %558 ], [ %562, %561 ], [ %564, %563 ], !dbg !33 + %565 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i11 = icmp eq i32 %565, 0, !dbg !33 + %566 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i14 = icmp eq i32 %566, 0, !dbg !33 + br i1 %.not.i11, label %572, label %567, !dbg !33 + +567: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %570, label %568, !dbg !33 + +568: ; preds = %567 + %569 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %524) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +570: ; preds = %567 + %571 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %524) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +572: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %575, label %573, !dbg !33 + +573: ; preds = %572 + %574 = tail call float @llvm.nvvm.sqrt.rn.f(float %524) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +575: ; preds = %572 + %576 = tail call float @llvm.nvvm.sqrt.approx.f(float %524) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +__nv_sqrtf.exit15: ; preds = %568, %570, %573, %575 + %.0.i13 = phi float [ %569, %568 ], [ %571, %570 ], [ %574, %573 ], [ %576, %575 ], !dbg !33 + %577 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i16 = icmp eq i32 %577, 0, !dbg !33 + %578 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i19 = icmp eq i32 %578, 0, !dbg !33 + br i1 %.not.i16, label %584, label %579, !dbg !33 + +579: ; preds = %__nv_sqrtf.exit15 + br i1 %.not1.i19, label %582, label %580, !dbg !33 + +580: ; preds = %579 + %581 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %525) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +582: ; preds = %579 + %583 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %525) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +584: ; preds = %__nv_sqrtf.exit15 + br i1 %.not1.i19, label %587, label %585, !dbg !33 + +585: ; preds = %584 + %586 = tail call float @llvm.nvvm.sqrt.rn.f(float %525) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +587: ; preds = %584 + %588 = tail call float @llvm.nvvm.sqrt.approx.f(float %525) #6, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +__nv_sqrtf.exit20: ; preds = %580, %582, %585, %587 + %.0.i18 = phi float [ %581, %580 ], [ %583, %582 ], [ %586, %585 ], [ %588, %587 ], !dbg !33 + %589 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i21 = icmp eq i32 %589, 0, !dbg !33 + %590 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i24 = icmp eq i32 %590, 0, !dbg !33 + br i1 %.not.i21, label %596, label %591, !dbg !33 + +591: ; preds = %__nv_sqrtf.exit20 + br i1 %.not1.i24, label %594, label %592, !dbg !33 + +592: ; preds = %591 + %593 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %526) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +594: ; preds = %591 + %595 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %526) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +596: ; preds = %__nv_sqrtf.exit20 + br i1 %.not1.i24, label %599, label %597, !dbg !33 + +597: ; preds = %596 + %598 = tail call float @llvm.nvvm.sqrt.rn.f(float %526) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +599: ; preds = %596 + %600 = tail call float @llvm.nvvm.sqrt.approx.f(float %526) #6, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +__nv_sqrtf.exit25: ; preds = %592, %594, %597, %599 + %.0.i23 = phi float [ %593, %592 ], [ %595, %594 ], [ %598, %597 ], [ %600, %599 ], !dbg !33 + %601 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i26 = icmp eq i32 %601, 0, !dbg !33 + %602 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i29 = icmp eq i32 %602, 0, !dbg !33 + br i1 %.not.i26, label %608, label %603, !dbg !33 + +603: ; preds = %__nv_sqrtf.exit25 + br i1 %.not1.i29, label %606, label %604, !dbg !33 + +604: ; preds = %603 + %605 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %527) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +606: ; preds = %603 + %607 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %527) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +608: ; preds = %__nv_sqrtf.exit25 + br i1 %.not1.i29, label %611, label %609, !dbg !33 + +609: ; preds = %608 + %610 = tail call float @llvm.nvvm.sqrt.rn.f(float %527) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +611: ; preds = %608 + %612 = tail call float @llvm.nvvm.sqrt.approx.f(float %527) #6, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +__nv_sqrtf.exit30: ; preds = %604, %606, %609, %611 + %.0.i28 = phi float [ %605, %604 ], [ %607, %606 ], [ %610, %609 ], [ %612, %611 ], !dbg !33 + %613 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i31 = icmp eq i32 %613, 0, !dbg !33 + %614 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i34 = icmp eq i32 %614, 0, !dbg !33 + br i1 %.not.i31, label %620, label %615, !dbg !33 + +615: ; preds = %__nv_sqrtf.exit30 + br i1 %.not1.i34, label %618, label %616, !dbg !33 + +616: ; preds = %615 + %617 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %528) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +618: ; preds = %615 + %619 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %528) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +620: ; preds = %__nv_sqrtf.exit30 + br i1 %.not1.i34, label %623, label %621, !dbg !33 + +621: ; preds = %620 + %622 = tail call float @llvm.nvvm.sqrt.rn.f(float %528) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +623: ; preds = %620 + %624 = tail call float @llvm.nvvm.sqrt.approx.f(float %528) #6, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +__nv_sqrtf.exit35: ; preds = %616, %618, %621, %623 + %.0.i33 = phi float [ %617, %616 ], [ %619, %618 ], [ %622, %621 ], [ %624, %623 ], !dbg !33 + %625 = and i32 %11, 31, !dbg !16 + %626 = zext nneg i32 %625 to i64, !dbg !17 + %627 = or disjoint i64 %10, %626, !dbg !17 + %628 = getelementptr float, ptr addrspace(1) %2, i64 %627, !dbg !34 + %629 = getelementptr bfloat, ptr addrspace(1) %1, i64 %55, !dbg !35 + %630 = getelementptr bfloat, ptr addrspace(1) %629, i64 %64, !dbg !36 + %631 = getelementptr bfloat, ptr addrspace(1) %1, i64 %54, !dbg !35 + %632 = getelementptr bfloat, ptr addrspace(1) %631, i64 %64, !dbg !36 + %633 = getelementptr bfloat, ptr addrspace(1) %1, i64 %53, !dbg !35 + %634 = getelementptr bfloat, ptr addrspace(1) %633, i64 %64, !dbg !36 + %635 = getelementptr bfloat, ptr addrspace(1) %1, i64 %52, !dbg !35 + %636 = getelementptr bfloat, ptr addrspace(1) %635, i64 %64, !dbg !36 + %637 = getelementptr bfloat, ptr addrspace(1) %1, i64 %51, !dbg !35 + %638 = getelementptr bfloat, ptr addrspace(1) %637, i64 %64, !dbg !36 + %639 = getelementptr bfloat, ptr addrspace(1) %1, i64 %50, !dbg !35 + %640 = getelementptr bfloat, ptr addrspace(1) %639, i64 %64, !dbg !36 + %641 = getelementptr bfloat, ptr addrspace(1) %1, i64 %49, !dbg !35 + %642 = getelementptr bfloat, ptr addrspace(1) %641, i64 %64, !dbg !36 + %643 = getelementptr bfloat, ptr addrspace(1) %1, i64 %48, !dbg !35 + %644 = getelementptr bfloat, ptr addrspace(1) %643, i64 %64, !dbg !36 + %645 = icmp slt i64 %627, %37, !dbg !18 + %646 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i), !dbg !37 + %647 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i3), !dbg !37 + %648 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i8), !dbg !37 + %649 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i13), !dbg !37 + %650 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i18), !dbg !37 + %651 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i23), !dbg !37 + %652 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i28), !dbg !37 + %653 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i33), !dbg !37 + %654 = fmul float %646, %209, !dbg !38 + %655 = fmul float %646, %210, !dbg !38 + %656 = fmul float %646, %211, !dbg !38 + %657 = fmul float %646, %212, !dbg !38 + %658 = fmul float %646, %213, !dbg !38 + %659 = fmul float %646, %214, !dbg !38 + %660 = fmul float %646, %215, !dbg !38 + %661 = fmul float %646, %216, !dbg !38 + %662 = fmul float %647, %217, !dbg !38 + %663 = fmul float %647, %218, !dbg !38 + %664 = fmul float %647, %219, !dbg !38 + %665 = fmul float %647, %220, !dbg !38 + %666 = fmul float %647, %221, !dbg !38 + %667 = fmul float %647, %222, !dbg !38 + %668 = fmul float %647, %223, !dbg !38 + %669 = fmul float %647, %224, !dbg !38 + %670 = fmul float %648, %225, !dbg !38 + %671 = fmul float %648, %226, !dbg !38 + %672 = fmul float %648, %227, !dbg !38 + %673 = fmul float %648, %228, !dbg !38 + %674 = fmul float %648, %229, !dbg !38 + %675 = fmul float %648, %230, !dbg !38 + %676 = fmul float %648, %231, !dbg !38 + %677 = fmul float %648, %232, !dbg !38 + %678 = fmul float %649, %233, !dbg !38 + %679 = fmul float %649, %234, !dbg !38 + %680 = fmul float %649, %235, !dbg !38 + %681 = fmul float %649, %236, !dbg !38 + %682 = fmul float %649, %237, !dbg !38 + %683 = fmul float %649, %238, !dbg !38 + %684 = fmul float %649, %239, !dbg !38 + %685 = fmul float %649, %240, !dbg !38 + %686 = fmul float %650, %241, !dbg !38 + %687 = fmul float %650, %242, !dbg !38 + %688 = fmul float %650, %243, !dbg !38 + %689 = fmul float %650, %244, !dbg !38 + %690 = fmul float %650, %245, !dbg !38 + %691 = fmul float %650, %246, !dbg !38 + %692 = fmul float %650, %247, !dbg !38 + %693 = fmul float %650, %248, !dbg !38 + %694 = fmul float %651, %249, !dbg !38 + %695 = fmul float %651, %250, !dbg !38 + %696 = fmul float %651, %251, !dbg !38 + %697 = fmul float %651, %252, !dbg !38 + %698 = fmul float %651, %253, !dbg !38 + %699 = fmul float %651, %254, !dbg !38 + %700 = fmul float %651, %255, !dbg !38 + %701 = fmul float %651, %256, !dbg !38 + %702 = fmul float %652, %257, !dbg !38 + %703 = fmul float %652, %258, !dbg !38 + %704 = fmul float %652, %259, !dbg !38 + %705 = fmul float %652, %260, !dbg !38 + %706 = fmul float %652, %261, !dbg !38 + %707 = fmul float %652, %262, !dbg !38 + %708 = fmul float %652, %263, !dbg !38 + %709 = fmul float %652, %264, !dbg !38 + %710 = fmul float %653, %265, !dbg !38 + %711 = fmul float %653, %266, !dbg !38 + %712 = fmul float %653, %267, !dbg !38 + %713 = fmul float %653, %268, !dbg !38 + %714 = fmul float %653, %269, !dbg !38 + %715 = fmul float %653, %270, !dbg !38 + %716 = fmul float %653, %271, !dbg !38 + %717 = fmul float %653, %272, !dbg !38 + %718 = fptrunc float %654 to bfloat, !dbg !39 + %719 = fptrunc float %655 to bfloat, !dbg !39 + %720 = fptrunc float %656 to bfloat, !dbg !39 + %721 = fptrunc float %657 to bfloat, !dbg !39 + %722 = fptrunc float %658 to bfloat, !dbg !39 + %723 = fptrunc float %659 to bfloat, !dbg !39 + %724 = fptrunc float %660 to bfloat, !dbg !39 + %725 = fptrunc float %661 to bfloat, !dbg !39 + %726 = fptrunc float %662 to bfloat, !dbg !39 + %727 = fptrunc float %663 to bfloat, !dbg !39 + %728 = fptrunc float %664 to bfloat, !dbg !39 + %729 = fptrunc float %665 to bfloat, !dbg !39 + %730 = fptrunc float %666 to bfloat, !dbg !39 + %731 = fptrunc float %667 to bfloat, !dbg !39 + %732 = fptrunc float %668 to bfloat, !dbg !39 + %733 = fptrunc float %669 to bfloat, !dbg !39 + %734 = fptrunc float %670 to bfloat, !dbg !39 + %735 = fptrunc float %671 to bfloat, !dbg !39 + %736 = fptrunc float %672 to bfloat, !dbg !39 + %737 = fptrunc float %673 to bfloat, !dbg !39 + %738 = fptrunc float %674 to bfloat, !dbg !39 + %739 = fptrunc float %675 to bfloat, !dbg !39 + %740 = fptrunc float %676 to bfloat, !dbg !39 + %741 = fptrunc float %677 to bfloat, !dbg !39 + %742 = fptrunc float %678 to bfloat, !dbg !39 + %743 = fptrunc float %679 to bfloat, !dbg !39 + %744 = fptrunc float %680 to bfloat, !dbg !39 + %745 = fptrunc float %681 to bfloat, !dbg !39 + %746 = fptrunc float %682 to bfloat, !dbg !39 + %747 = fptrunc float %683 to bfloat, !dbg !39 + %748 = fptrunc float %684 to bfloat, !dbg !39 + %749 = fptrunc float %685 to bfloat, !dbg !39 + %750 = fptrunc float %686 to bfloat, !dbg !39 + %751 = fptrunc float %687 to bfloat, !dbg !39 + %752 = fptrunc float %688 to bfloat, !dbg !39 + %753 = fptrunc float %689 to bfloat, !dbg !39 + %754 = fptrunc float %690 to bfloat, !dbg !39 + %755 = fptrunc float %691 to bfloat, !dbg !39 + %756 = fptrunc float %692 to bfloat, !dbg !39 + %757 = fptrunc float %693 to bfloat, !dbg !39 + %758 = fptrunc float %694 to bfloat, !dbg !39 + %759 = fptrunc float %695 to bfloat, !dbg !39 + %760 = fptrunc float %696 to bfloat, !dbg !39 + %761 = fptrunc float %697 to bfloat, !dbg !39 + %762 = fptrunc float %698 to bfloat, !dbg !39 + %763 = fptrunc float %699 to bfloat, !dbg !39 + %764 = fptrunc float %700 to bfloat, !dbg !39 + %765 = fptrunc float %701 to bfloat, !dbg !39 + %766 = fptrunc float %702 to bfloat, !dbg !39 + %767 = fptrunc float %703 to bfloat, !dbg !39 + %768 = fptrunc float %704 to bfloat, !dbg !39 + %769 = fptrunc float %705 to bfloat, !dbg !39 + %770 = fptrunc float %706 to bfloat, !dbg !39 + %771 = fptrunc float %707 to bfloat, !dbg !39 + %772 = fptrunc float %708 to bfloat, !dbg !39 + %773 = fptrunc float %709 to bfloat, !dbg !39 + %774 = fptrunc float %710 to bfloat, !dbg !39 + %775 = fptrunc float %711 to bfloat, !dbg !39 + %776 = fptrunc float %712 to bfloat, !dbg !39 + %777 = fptrunc float %713 to bfloat, !dbg !39 + %778 = fptrunc float %714 to bfloat, !dbg !39 + %779 = fptrunc float %715 to bfloat, !dbg !39 + %780 = fptrunc float %716 to bfloat, !dbg !39 + %781 = fptrunc float %717 to bfloat, !dbg !39 + %782 = insertelement <2 x bfloat> poison, bfloat %718, i64 0, !dbg !40 + %783 = insertelement <2 x bfloat> %782, bfloat %719, i64 1, !dbg !40 + %784 = bitcast <2 x bfloat> %783 to i32, !dbg !40 + %785 = insertelement <2 x bfloat> poison, bfloat %720, i64 0, !dbg !40 + %786 = insertelement <2 x bfloat> %785, bfloat %721, i64 1, !dbg !40 + %787 = bitcast <2 x bfloat> %786 to i32, !dbg !40 + %788 = insertelement <2 x bfloat> poison, bfloat %722, i64 0, !dbg !40 + %789 = insertelement <2 x bfloat> %788, bfloat %723, i64 1, !dbg !40 + %790 = bitcast <2 x bfloat> %789 to i32, !dbg !40 + %791 = insertelement <2 x bfloat> poison, bfloat %724, i64 0, !dbg !40 + %792 = insertelement <2 x bfloat> %791, bfloat %725, i64 1, !dbg !40 + %793 = bitcast <2 x bfloat> %792 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %784, i32 %787, i32 %790, i32 %793, ptr addrspace(1) %644, i1 %38) #6, !dbg !40 + %794 = insertelement <2 x bfloat> poison, bfloat %726, i64 0, !dbg !40 + %795 = insertelement <2 x bfloat> %794, bfloat %727, i64 1, !dbg !40 + %796 = bitcast <2 x bfloat> %795 to i32, !dbg !40 + %797 = insertelement <2 x bfloat> poison, bfloat %728, i64 0, !dbg !40 + %798 = insertelement <2 x bfloat> %797, bfloat %729, i64 1, !dbg !40 + %799 = bitcast <2 x bfloat> %798 to i32, !dbg !40 + %800 = insertelement <2 x bfloat> poison, bfloat %730, i64 0, !dbg !40 + %801 = insertelement <2 x bfloat> %800, bfloat %731, i64 1, !dbg !40 + %802 = bitcast <2 x bfloat> %801 to i32, !dbg !40 + %803 = insertelement <2 x bfloat> poison, bfloat %732, i64 0, !dbg !40 + %804 = insertelement <2 x bfloat> %803, bfloat %733, i64 1, !dbg !40 + %805 = bitcast <2 x bfloat> %804 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %796, i32 %799, i32 %802, i32 %805, ptr addrspace(1) %642, i1 %39) #6, !dbg !40 + %806 = insertelement <2 x bfloat> poison, bfloat %734, i64 0, !dbg !40 + %807 = insertelement <2 x bfloat> %806, bfloat %735, i64 1, !dbg !40 + %808 = bitcast <2 x bfloat> %807 to i32, !dbg !40 + %809 = insertelement <2 x bfloat> poison, bfloat %736, i64 0, !dbg !40 + %810 = insertelement <2 x bfloat> %809, bfloat %737, i64 1, !dbg !40 + %811 = bitcast <2 x bfloat> %810 to i32, !dbg !40 + %812 = insertelement <2 x bfloat> poison, bfloat %738, i64 0, !dbg !40 + %813 = insertelement <2 x bfloat> %812, bfloat %739, i64 1, !dbg !40 + %814 = bitcast <2 x bfloat> %813 to i32, !dbg !40 + %815 = insertelement <2 x bfloat> poison, bfloat %740, i64 0, !dbg !40 + %816 = insertelement <2 x bfloat> %815, bfloat %741, i64 1, !dbg !40 + %817 = bitcast <2 x bfloat> %816 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %808, i32 %811, i32 %814, i32 %817, ptr addrspace(1) %640, i1 %40) #6, !dbg !40 + %818 = insertelement <2 x bfloat> poison, bfloat %742, i64 0, !dbg !40 + %819 = insertelement <2 x bfloat> %818, bfloat %743, i64 1, !dbg !40 + %820 = bitcast <2 x bfloat> %819 to i32, !dbg !40 + %821 = insertelement <2 x bfloat> poison, bfloat %744, i64 0, !dbg !40 + %822 = insertelement <2 x bfloat> %821, bfloat %745, i64 1, !dbg !40 + %823 = bitcast <2 x bfloat> %822 to i32, !dbg !40 + %824 = insertelement <2 x bfloat> poison, bfloat %746, i64 0, !dbg !40 + %825 = insertelement <2 x bfloat> %824, bfloat %747, i64 1, !dbg !40 + %826 = bitcast <2 x bfloat> %825 to i32, !dbg !40 + %827 = insertelement <2 x bfloat> poison, bfloat %748, i64 0, !dbg !40 + %828 = insertelement <2 x bfloat> %827, bfloat %749, i64 1, !dbg !40 + %829 = bitcast <2 x bfloat> %828 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %820, i32 %823, i32 %826, i32 %829, ptr addrspace(1) %638, i1 %41) #6, !dbg !40 + %830 = insertelement <2 x bfloat> poison, bfloat %750, i64 0, !dbg !40 + %831 = insertelement <2 x bfloat> %830, bfloat %751, i64 1, !dbg !40 + %832 = bitcast <2 x bfloat> %831 to i32, !dbg !40 + %833 = insertelement <2 x bfloat> poison, bfloat %752, i64 0, !dbg !40 + %834 = insertelement <2 x bfloat> %833, bfloat %753, i64 1, !dbg !40 + %835 = bitcast <2 x bfloat> %834 to i32, !dbg !40 + %836 = insertelement <2 x bfloat> poison, bfloat %754, i64 0, !dbg !40 + %837 = insertelement <2 x bfloat> %836, bfloat %755, i64 1, !dbg !40 + %838 = bitcast <2 x bfloat> %837 to i32, !dbg !40 + %839 = insertelement <2 x bfloat> poison, bfloat %756, i64 0, !dbg !40 + %840 = insertelement <2 x bfloat> %839, bfloat %757, i64 1, !dbg !40 + %841 = bitcast <2 x bfloat> %840 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %832, i32 %835, i32 %838, i32 %841, ptr addrspace(1) %636, i1 %42) #6, !dbg !40 + %842 = insertelement <2 x bfloat> poison, bfloat %758, i64 0, !dbg !40 + %843 = insertelement <2 x bfloat> %842, bfloat %759, i64 1, !dbg !40 + %844 = bitcast <2 x bfloat> %843 to i32, !dbg !40 + %845 = insertelement <2 x bfloat> poison, bfloat %760, i64 0, !dbg !40 + %846 = insertelement <2 x bfloat> %845, bfloat %761, i64 1, !dbg !40 + %847 = bitcast <2 x bfloat> %846 to i32, !dbg !40 + %848 = insertelement <2 x bfloat> poison, bfloat %762, i64 0, !dbg !40 + %849 = insertelement <2 x bfloat> %848, bfloat %763, i64 1, !dbg !40 + %850 = bitcast <2 x bfloat> %849 to i32, !dbg !40 + %851 = insertelement <2 x bfloat> poison, bfloat %764, i64 0, !dbg !40 + %852 = insertelement <2 x bfloat> %851, bfloat %765, i64 1, !dbg !40 + %853 = bitcast <2 x bfloat> %852 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %844, i32 %847, i32 %850, i32 %853, ptr addrspace(1) %634, i1 %43) #6, !dbg !40 + %854 = insertelement <2 x bfloat> poison, bfloat %766, i64 0, !dbg !40 + %855 = insertelement <2 x bfloat> %854, bfloat %767, i64 1, !dbg !40 + %856 = bitcast <2 x bfloat> %855 to i32, !dbg !40 + %857 = insertelement <2 x bfloat> poison, bfloat %768, i64 0, !dbg !40 + %858 = insertelement <2 x bfloat> %857, bfloat %769, i64 1, !dbg !40 + %859 = bitcast <2 x bfloat> %858 to i32, !dbg !40 + %860 = insertelement <2 x bfloat> poison, bfloat %770, i64 0, !dbg !40 + %861 = insertelement <2 x bfloat> %860, bfloat %771, i64 1, !dbg !40 + %862 = bitcast <2 x bfloat> %861 to i32, !dbg !40 + %863 = insertelement <2 x bfloat> poison, bfloat %772, i64 0, !dbg !40 + %864 = insertelement <2 x bfloat> %863, bfloat %773, i64 1, !dbg !40 + %865 = bitcast <2 x bfloat> %864 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %856, i32 %859, i32 %862, i32 %865, ptr addrspace(1) %632, i1 %44) #6, !dbg !40 + %866 = insertelement <2 x bfloat> poison, bfloat %774, i64 0, !dbg !40 + %867 = insertelement <2 x bfloat> %866, bfloat %775, i64 1, !dbg !40 + %868 = bitcast <2 x bfloat> %867 to i32, !dbg !40 + %869 = insertelement <2 x bfloat> poison, bfloat %776, i64 0, !dbg !40 + %870 = insertelement <2 x bfloat> %869, bfloat %777, i64 1, !dbg !40 + %871 = bitcast <2 x bfloat> %870 to i32, !dbg !40 + %872 = insertelement <2 x bfloat> poison, bfloat %778, i64 0, !dbg !40 + %873 = insertelement <2 x bfloat> %872, bfloat %779, i64 1, !dbg !40 + %874 = bitcast <2 x bfloat> %873 to i32, !dbg !40 + %875 = insertelement <2 x bfloat> poison, bfloat %780, i64 0, !dbg !40 + %876 = insertelement <2 x bfloat> %875, bfloat %781, i64 1, !dbg !40 + %877 = bitcast <2 x bfloat> %876 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %868, i32 %871, i32 %874, i32 %877, ptr addrspace(1) %630, i1 %45) #6, !dbg !40 + %878 = shl nuw nsw i32 %12, 1, !dbg !41 + %879 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %878, !dbg !41 + %880 = bitcast float %646 to i32, !dbg !41 + %881 = bitcast float %647 to i32, !dbg !41 + %882 = bitcast float %648 to i32, !dbg !41 + %883 = bitcast float %649 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %879, i32 %880, i32 %881, i32 %882, i32 %883) #6, !dbg !41 + %884 = getelementptr inbounds nuw i8, ptr addrspace(3) %879, i32 16, !dbg !41 + %885 = bitcast float %650 to i32, !dbg !41 + %886 = bitcast float %651 to i32, !dbg !41 + %887 = bitcast float %652 to i32, !dbg !41 + %888 = bitcast float %653 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %884, i32 %885, i32 %886, i32 %887, i32 %888) #6, !dbg !41 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !41 + %889 = shl nuw nsw i32 %11, 5, !dbg !41 + %890 = and i32 %889, 96, !dbg !41 + %891 = and i32 %11, 28, !dbg !41 + %892 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %890, !dbg !41 + %893 = getelementptr inbounds nuw i8, ptr addrspace(3) %892, i32 %891, !dbg !41 + %894 = load i32, ptr addrspace(3) %893, align 4, !dbg !41 + %895 = and i32 %11, 32, !dbg !41 + %896 = icmp eq i32 %895, 0, !dbg !41 + %897 = and i1 %896, %645, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %894, ptr addrspace(1) %628, i1 %897) #6, !dbg !41 + ret void, !dbg !42 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.div.full(float, float) #3 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #4 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #5 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.f(float) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="64" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) } +attributes #4 = { convergent nocallback nounwind } +attributes #5 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #6 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "l2norm_fwd_kernel", linkageName: "l2norm_fwd_kernel", scope: !1, file: !1, line: 69, type: !6, scopeLine: 69, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !10, !11, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 70, column: 24, scope: !5) +!14 = !DILocation(line: 70, column: 30, scope: !5) +!15 = !DILocation(line: 71, column: 16, scope: !5) +!16 = !DILocation(line: 71, column: 34, scope: !5) +!17 = !DILocation(line: 71, column: 21, scope: !5) +!18 = !DILocation(line: 73, column: 16, scope: !5) +!19 = !DILocation(line: 74, column: 30, scope: !5) +!20 = !DILocation(line: 75, column: 29, scope: !5) +!21 = !DILocation(line: 75, column: 14, scope: !5) +!22 = !DILocation(line: 75, column: 33, scope: !5) +!23 = !DILocation(line: 79, column: 18, scope: !5) +!24 = !DILocation(line: 79, column: 47, scope: !5) +!25 = !DILocation(line: 80, column: 38, scope: !5) +!26 = !DILocation(line: 263, column: 15, scope: !27, inlinedAt: !29) +!27 = distinct !DILexicalBlockFile(scope: !5, file: !28, discriminator: 0) +!28 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!29 = !DILocation(line: 293, column: 36, scope: !27, inlinedAt: !30) +!30 = !DILocation(line: 80, column: 43, scope: !31) +!31 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!32 = !DILocation(line: 80, column: 48, scope: !5) +!33 = !DILocation(line: 80, column: 25, scope: !5) +!34 = !DILocation(line: 77, column: 20, scope: !5) +!35 = !DILocation(line: 76, column: 14, scope: !5) +!36 = !DILocation(line: 76, column: 33, scope: !5) +!37 = !DILocation(line: 80, column: 17, scope: !5) +!38 = !DILocation(line: 81, column: 16, scope: !5) +!39 = !DILocation(line: 83, column: 25, scope: !5) +!40 = !DILocation(line: 83, column: 18, scope: !5) +!41 = !DILocation(line: 84, column: 21, scope: !5) +!42 = !DILocation(line: 84, column: 4, scope: !5) diff --git a/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/l2norm_fwd_kernel.ptx b/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/l2norm_fwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..bf9f670baccbc9ea04a140b39905f33c68561db6 --- /dev/null +++ b/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/l2norm_fwd_kernel.ptx @@ -0,0 +1,966 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_fwd_kernel // -- Begin function l2norm_fwd_kernel +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; +.global .align 1 .b8 _$_str_$_2[17] = {95, 95, 67, 85, 68, 65, 95, 80, 82, 69, 67, 95, 83, 81, 82, 84}; + // @l2norm_fwd_kernel +.visible .entry l2norm_fwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_2, + .param .f32 l2norm_fwd_kernel_param_3, + .param .u32 l2norm_fwd_kernel_param_4, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_6 +) +.reqntid 64 +{ + .reg .pred %p<12>; + .reg .b16 %rs<65>; + .reg .b32 %r<373>; + .reg .b64 %rd<67>; + .loc 1 69 0 // l2norm.py:69:0 +$L__func_begin0: + .loc 1 69 0 // l2norm.py:69:0 + +// %bb.0: // %__nv_sqrtf.exit + ld.param.b64 %rd18, [l2norm_fwd_kernel_param_0]; + ld.param.b64 %rd19, [l2norm_fwd_kernel_param_1]; +$L__tmp0: + .loc 1 70 24 // l2norm.py:70:24 + mov.u32 %r77, %ctaid.x; + ld.param.b64 %rd20, [l2norm_fwd_kernel_param_2]; + .loc 1 71 16 // l2norm.py:71:16 + mul.wide.u32 %rd21, %r77, 32; + ld.param.b32 %r78, [l2norm_fwd_kernel_param_3]; + ld.param.s32 %rd22, [l2norm_fwd_kernel_param_4]; + .loc 1 71 34 // l2norm.py:71:34 + mov.u32 %r79, %tid.x; + and.b32 %r80, %r79, 48; + bfe.u32 %r81, %r79, 4, 2; + or.b32 %r82, %r81, 4; + or.b32 %r83, %r81, 8; + or.b32 %r84, %r81, 12; + or.b32 %r85, %r81, 16; + or.b32 %r86, %r81, 20; + or.b32 %r87, %r81, 24; + or.b32 %r88, %r81, 28; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd23, %r81; + cvt.u64.u32 %rd24, %r82; + cvt.u64.u32 %rd25, %r83; + cvt.u64.u32 %rd26, %r84; + cvt.u64.u32 %rd27, %r85; + cvt.u64.u32 %rd28, %r86; + cvt.u64.u32 %rd29, %r87; + cvt.u64.u32 %rd30, %r88; + or.b64 %rd31, %rd21, %rd23; + or.b64 %rd32, %rd21, %rd24; + or.b64 %rd33, %rd21, %rd25; + or.b64 %rd34, %rd21, %rd26; + or.b64 %rd35, %rd21, %rd27; + or.b64 %rd36, %rd21, %rd28; + or.b64 %rd37, %rd21, %rd29; + or.b64 %rd38, %rd21, %rd30; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p1, %rd31, %rd22; + setp.lt.s64 %p2, %rd32, %rd22; + setp.lt.s64 %p3, %rd33, %rd22; + setp.lt.s64 %p4, %rd34, %rd22; + setp.lt.s64 %p5, %rd35, %rd22; + setp.lt.s64 %p6, %rd36, %rd22; + setp.lt.s64 %p7, %rd37, %rd22; + setp.lt.s64 %p8, %rd38, %rd22; + .loc 1 74 30 // l2norm.py:74:30 + shl.b32 %r89, %r79, 3; + and.b32 %r90, %r89, 120; + .loc 1 75 14 // l2norm.py:75:14 + shl.b64 %rd39, %rd31, 8; + add.s64 %rd40, %rd18, %rd39; + shl.b64 %rd41, %rd32, 8; + add.s64 %rd42, %rd18, %rd41; + shl.b64 %rd43, %rd33, 8; + add.s64 %rd44, %rd18, %rd43; + shl.b64 %rd45, %rd34, 8; + add.s64 %rd46, %rd18, %rd45; + shl.b64 %rd47, %rd35, 8; + add.s64 %rd48, %rd18, %rd47; + shl.b64 %rd49, %rd36, 8; + add.s64 %rd50, %rd18, %rd49; + shl.b64 %rd51, %rd37, 8; + add.s64 %rd52, %rd18, %rd51; + shl.b64 %rd53, %rd38, 8; + add.s64 %rd54, %rd18, %rd53; + .loc 1 75 33 // l2norm.py:75:33 + mul.wide.u32 %rd55, %r90, 2; + add.s64 %rd1, %rd40, %rd55; + add.s64 %rd2, %rd42, %rd55; + add.s64 %rd3, %rd44, %rd55; + add.s64 %rd4, %rd46, %rd55; + add.s64 %rd5, %rd48, %rd55; + add.s64 %rd6, %rd50, %rd55; + add.s64 %rd7, %rd52, %rd55; + add.s64 %rd8, %rd54, %rd55; + mov.b32 %r5, 0; + .loc 1 79 18 // l2norm.py:79:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + mov.b32 {%rs5, %rs6}, %r3; + mov.b32 {%rs7, %rs8}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs9, %rs10}, %r6; + mov.b32 {%rs11, %rs12}, %r7; + mov.b32 {%rs13, %rs14}, %r8; + mov.b32 {%rs15, %rs16}, %r9; + // begin inline asm + mov.u32 %r10, %r5; + mov.u32 %r11, %r5; + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + @%p3 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs17, %rs18}, %r10; + mov.b32 {%rs19, %rs20}, %r11; + mov.b32 {%rs21, %rs22}, %r12; + mov.b32 {%rs23, %rs24}, %r13; + // begin inline asm + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + @%p4 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ]; + // end inline asm + mov.b32 {%rs25, %rs26}, %r14; + mov.b32 {%rs27, %rs28}, %r15; + mov.b32 {%rs29, %rs30}, %r16; + mov.b32 {%rs31, %rs32}, %r17; + // begin inline asm + mov.u32 %r18, %r5; + mov.u32 %r19, %r5; + mov.u32 %r20, %r5; + mov.u32 %r21, %r5; + @%p5 ld.global.v4.b32 { %r18, %r19, %r20, %r21 }, [ %rd5 + 0 ]; + // end inline asm + mov.b32 {%rs33, %rs34}, %r18; + mov.b32 {%rs35, %rs36}, %r19; + mov.b32 {%rs37, %rs38}, %r20; + mov.b32 {%rs39, %rs40}, %r21; + // begin inline asm + mov.u32 %r22, %r5; + mov.u32 %r23, %r5; + mov.u32 %r24, %r5; + mov.u32 %r25, %r5; + @%p6 ld.global.v4.b32 { %r22, %r23, %r24, %r25 }, [ %rd6 + 0 ]; + // end inline asm + mov.b32 {%rs41, %rs42}, %r22; + mov.b32 {%rs43, %rs44}, %r23; + mov.b32 {%rs45, %rs46}, %r24; + mov.b32 {%rs47, %rs48}, %r25; + // begin inline asm + mov.u32 %r26, %r5; + mov.u32 %r27, %r5; + mov.u32 %r28, %r5; + mov.u32 %r29, %r5; + @%p7 ld.global.v4.b32 { %r26, %r27, %r28, %r29 }, [ %rd7 + 0 ]; + // end inline asm + mov.b32 {%rs49, %rs50}, %r26; + mov.b32 {%rs51, %rs52}, %r27; + mov.b32 {%rs53, %rs54}, %r28; + mov.b32 {%rs55, %rs56}, %r29; + // begin inline asm + mov.u32 %r30, %r5; + mov.u32 %r31, %r5; + mov.u32 %r32, %r5; + mov.u32 %r33, %r5; + @%p8 ld.global.v4.b32 { %r30, %r31, %r32, %r33 }, [ %rd8 + 0 ]; + // end inline asm + mov.b32 {%rs57, %rs58}, %r30; + mov.b32 {%rs59, %rs60}, %r31; + mov.b32 {%rs61, %rs62}, %r32; + mov.b32 {%rs63, %rs64}, %r33; + .loc 1 79 47 // l2norm.py:79:47 + cvt.f32.bf16 %r91, %rs1; + cvt.f32.bf16 %r92, %rs2; + cvt.f32.bf16 %r93, %rs3; + cvt.f32.bf16 %r94, %rs4; + cvt.f32.bf16 %r95, %rs5; + cvt.f32.bf16 %r96, %rs6; + cvt.f32.bf16 %r97, %rs7; + cvt.f32.bf16 %r98, %rs8; + cvt.f32.bf16 %r99, %rs9; + cvt.f32.bf16 %r100, %rs10; + cvt.f32.bf16 %r101, %rs11; + cvt.f32.bf16 %r102, %rs12; + cvt.f32.bf16 %r103, %rs13; + cvt.f32.bf16 %r104, %rs14; + cvt.f32.bf16 %r105, %rs15; + cvt.f32.bf16 %r106, %rs16; + cvt.f32.bf16 %r107, %rs17; + cvt.f32.bf16 %r108, %rs18; + cvt.f32.bf16 %r109, %rs19; + cvt.f32.bf16 %r110, %rs20; + cvt.f32.bf16 %r111, %rs21; + cvt.f32.bf16 %r112, %rs22; + cvt.f32.bf16 %r113, %rs23; + cvt.f32.bf16 %r114, %rs24; + cvt.f32.bf16 %r115, %rs25; + cvt.f32.bf16 %r116, %rs26; + cvt.f32.bf16 %r117, %rs27; + cvt.f32.bf16 %r118, %rs28; + cvt.f32.bf16 %r119, %rs29; + cvt.f32.bf16 %r120, %rs30; + cvt.f32.bf16 %r121, %rs31; + cvt.f32.bf16 %r122, %rs32; + cvt.f32.bf16 %r123, %rs33; + cvt.f32.bf16 %r124, %rs34; + cvt.f32.bf16 %r125, %rs35; + cvt.f32.bf16 %r126, %rs36; + cvt.f32.bf16 %r127, %rs37; + cvt.f32.bf16 %r128, %rs38; + cvt.f32.bf16 %r129, %rs39; + cvt.f32.bf16 %r130, %rs40; + cvt.f32.bf16 %r131, %rs41; + cvt.f32.bf16 %r132, %rs42; + cvt.f32.bf16 %r133, %rs43; + cvt.f32.bf16 %r134, %rs44; + cvt.f32.bf16 %r135, %rs45; + cvt.f32.bf16 %r136, %rs46; + cvt.f32.bf16 %r137, %rs47; + cvt.f32.bf16 %r138, %rs48; + cvt.f32.bf16 %r139, %rs49; + cvt.f32.bf16 %r140, %rs50; + cvt.f32.bf16 %r141, %rs51; + cvt.f32.bf16 %r142, %rs52; + cvt.f32.bf16 %r143, %rs53; + cvt.f32.bf16 %r144, %rs54; + cvt.f32.bf16 %r145, %rs55; + cvt.f32.bf16 %r146, %rs56; + cvt.f32.bf16 %r147, %rs57; + cvt.f32.bf16 %r148, %rs58; + cvt.f32.bf16 %r149, %rs59; + cvt.f32.bf16 %r150, %rs60; + cvt.f32.bf16 %r151, %rs61; + cvt.f32.bf16 %r152, %rs62; + cvt.f32.bf16 %r153, %rs63; + cvt.f32.bf16 %r154, %rs64; + .loc 1 80 38 // l2norm.py:80:38 + mul.f32 %r155, %r92, %r92; + mul.f32 %r156, %r100, %r100; + mul.f32 %r157, %r108, %r108; + mul.f32 %r158, %r116, %r116; + mul.f32 %r159, %r124, %r124; + mul.f32 %r160, %r132, %r132; + mul.f32 %r161, %r140, %r140; + mul.f32 %r162, %r148, %r148; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + fma.rn.f32.bf16 %r163, %rs1, %rs1, %r155; + fma.rn.f32.bf16 %r164, %rs3, %rs3, %r163; + fma.rn.f32.bf16 %r165, %rs4, %rs4, %r164; + fma.rn.f32.bf16 %r166, %rs5, %rs5, %r165; + fma.rn.f32.bf16 %r167, %rs6, %rs6, %r166; + fma.rn.f32.bf16 %r168, %rs7, %rs7, %r167; + fma.rn.f32.bf16 %r169, %rs8, %rs8, %r168; + fma.rn.f32.bf16 %r170, %rs9, %rs9, %r156; + fma.rn.f32.bf16 %r171, %rs11, %rs11, %r170; + fma.rn.f32.bf16 %r172, %rs12, %rs12, %r171; + fma.rn.f32.bf16 %r173, %rs13, %rs13, %r172; + fma.rn.f32.bf16 %r174, %rs14, %rs14, %r173; + fma.rn.f32.bf16 %r175, %rs15, %rs15, %r174; + fma.rn.f32.bf16 %r176, %rs16, %rs16, %r175; + fma.rn.f32.bf16 %r177, %rs17, %rs17, %r157; + fma.rn.f32.bf16 %r178, %rs19, %rs19, %r177; + fma.rn.f32.bf16 %r179, %rs20, %rs20, %r178; + fma.rn.f32.bf16 %r180, %rs21, %rs21, %r179; + fma.rn.f32.bf16 %r181, %rs22, %rs22, %r180; + fma.rn.f32.bf16 %r182, %rs23, %rs23, %r181; + fma.rn.f32.bf16 %r183, %rs24, %rs24, %r182; + fma.rn.f32.bf16 %r184, %rs25, %rs25, %r158; + fma.rn.f32.bf16 %r185, %rs27, %rs27, %r184; + fma.rn.f32.bf16 %r186, %rs28, %rs28, %r185; + fma.rn.f32.bf16 %r187, %rs29, %rs29, %r186; + fma.rn.f32.bf16 %r188, %rs30, %rs30, %r187; + fma.rn.f32.bf16 %r189, %rs31, %rs31, %r188; + fma.rn.f32.bf16 %r190, %rs32, %rs32, %r189; + fma.rn.f32.bf16 %r191, %rs33, %rs33, %r159; + fma.rn.f32.bf16 %r192, %rs35, %rs35, %r191; + fma.rn.f32.bf16 %r193, %rs36, %rs36, %r192; + fma.rn.f32.bf16 %r194, %rs37, %rs37, %r193; + fma.rn.f32.bf16 %r195, %rs38, %rs38, %r194; + fma.rn.f32.bf16 %r196, %rs39, %rs39, %r195; + fma.rn.f32.bf16 %r197, %rs40, %rs40, %r196; + fma.rn.f32.bf16 %r198, %rs41, %rs41, %r160; + fma.rn.f32.bf16 %r199, %rs43, %rs43, %r198; + fma.rn.f32.bf16 %r200, %rs44, %rs44, %r199; + fma.rn.f32.bf16 %r201, %rs45, %rs45, %r200; + fma.rn.f32.bf16 %r202, %rs46, %rs46, %r201; + fma.rn.f32.bf16 %r203, %rs47, %rs47, %r202; + fma.rn.f32.bf16 %r204, %rs48, %rs48, %r203; + fma.rn.f32.bf16 %r205, %rs49, %rs49, %r161; + fma.rn.f32.bf16 %r206, %rs51, %rs51, %r205; + fma.rn.f32.bf16 %r207, %rs52, %rs52, %r206; + fma.rn.f32.bf16 %r208, %rs53, %rs53, %r207; + fma.rn.f32.bf16 %r209, %rs54, %rs54, %r208; + fma.rn.f32.bf16 %r210, %rs55, %rs55, %r209; + fma.rn.f32.bf16 %r211, %rs56, %rs56, %r210; + fma.rn.f32.bf16 %r212, %rs57, %rs57, %r162; + fma.rn.f32.bf16 %r213, %rs59, %rs59, %r212; + fma.rn.f32.bf16 %r214, %rs60, %rs60, %r213; + fma.rn.f32.bf16 %r215, %rs61, %rs61, %r214; + fma.rn.f32.bf16 %r216, %rs62, %rs62, %r215; + fma.rn.f32.bf16 %r217, %rs63, %rs63, %r216; + fma.rn.f32.bf16 %r218, %rs64, %rs64, %r217; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r219, %r169, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r220, %r169, %r219; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r221, %r220, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r222, %r220, %r221; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r223, %r222, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r224, %r222, %r223; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r225, %r224, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r226, %r224, %r225; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r227, %r176, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r228, %r176, %r227; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r229, %r228, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r230, %r228, %r229; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r231, %r230, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r232, %r230, %r231; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r233, %r232, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r234, %r232, %r233; +$L__tmp18: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r235, %r183, 8, 31, -1; +$L__tmp19: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r236, %r183, %r235; +$L__tmp20: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r237, %r236, 4, 31, -1; +$L__tmp21: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r238, %r236, %r237; +$L__tmp22: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r239, %r238, 2, 31, -1; +$L__tmp23: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r240, %r238, %r239; +$L__tmp24: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r241, %r240, 1, 31, -1; +$L__tmp25: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r242, %r240, %r241; +$L__tmp26: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r243, %r190, 8, 31, -1; +$L__tmp27: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r244, %r190, %r243; +$L__tmp28: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r245, %r244, 4, 31, -1; +$L__tmp29: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r246, %r244, %r245; +$L__tmp30: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r247, %r246, 2, 31, -1; +$L__tmp31: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r248, %r246, %r247; +$L__tmp32: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r249, %r248, 1, 31, -1; +$L__tmp33: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r250, %r248, %r249; +$L__tmp34: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r251, %r197, 8, 31, -1; +$L__tmp35: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r252, %r197, %r251; +$L__tmp36: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r253, %r252, 4, 31, -1; +$L__tmp37: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r254, %r252, %r253; +$L__tmp38: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r255, %r254, 2, 31, -1; +$L__tmp39: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r256, %r254, %r255; +$L__tmp40: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r257, %r256, 1, 31, -1; +$L__tmp41: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r258, %r256, %r257; +$L__tmp42: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r259, %r204, 8, 31, -1; +$L__tmp43: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r260, %r204, %r259; +$L__tmp44: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r261, %r260, 4, 31, -1; +$L__tmp45: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r262, %r260, %r261; +$L__tmp46: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r263, %r262, 2, 31, -1; +$L__tmp47: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r264, %r262, %r263; +$L__tmp48: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r265, %r264, 1, 31, -1; +$L__tmp49: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r266, %r264, %r265; +$L__tmp50: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r267, %r211, 8, 31, -1; +$L__tmp51: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r268, %r211, %r267; +$L__tmp52: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r269, %r268, 4, 31, -1; +$L__tmp53: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r270, %r268, %r269; +$L__tmp54: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r271, %r270, 2, 31, -1; +$L__tmp55: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r272, %r270, %r271; +$L__tmp56: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r273, %r272, 1, 31, -1; +$L__tmp57: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r274, %r272, %r273; +$L__tmp58: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r275, %r218, 8, 31, -1; +$L__tmp59: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r276, %r218, %r275; +$L__tmp60: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r277, %r276, 4, 31, -1; +$L__tmp61: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r278, %r276, %r277; +$L__tmp62: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r279, %r278, 2, 31, -1; +$L__tmp63: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r280, %r278, %r279; +$L__tmp64: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r281, %r280, 1, 31, -1; +$L__tmp65: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r282, %r280, %r281; +$L__tmp66: + .loc 1 80 48 // l2norm.py:80:48 + add.f32 %r283, %r78, %r226; + add.f32 %r284, %r78, %r234; + add.f32 %r285, %r78, %r242; + add.f32 %r286, %r78, %r250; + add.f32 %r287, %r78, %r258; + add.f32 %r288, %r78, %r266; + add.f32 %r289, %r78, %r274; + add.f32 %r290, %r78, %r282; + .loc 1 80 25 // l2norm.py:80:25 + sqrt.approx.ftz.f32 %r291, %r283; + sqrt.approx.ftz.f32 %r292, %r284; + sqrt.approx.ftz.f32 %r293, %r285; + sqrt.approx.ftz.f32 %r294, %r286; + sqrt.approx.ftz.f32 %r295, %r287; + sqrt.approx.ftz.f32 %r296, %r288; + sqrt.approx.ftz.f32 %r297, %r289; + sqrt.approx.ftz.f32 %r298, %r290; + .loc 1 71 34 // l2norm.py:71:34 + and.b32 %r299, %r79, 31; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd56, %r299; + or.b64 %rd57, %rd21, %rd56; + .loc 1 77 20 // l2norm.py:77:20 + shl.b64 %rd58, %rd57, 2; + add.s64 %rd17, %rd20, %rd58; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd59, %rd19, %rd53; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd16, %rd59, %rd55; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd60, %rd19, %rd51; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd15, %rd60, %rd55; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd61, %rd19, %rd49; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd14, %rd61, %rd55; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd62, %rd19, %rd47; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd13, %rd62, %rd55; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd63, %rd19, %rd45; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd12, %rd63, %rd55; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd64, %rd19, %rd43; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd11, %rd64, %rd55; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd65, %rd19, %rd41; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd10, %rd65, %rd55; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd66, %rd19, %rd39; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd9, %rd66, %rd55; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p10, %rd57, %rd22; + mov.b32 %r300, 0f3F800000; + .loc 1 80 17 // l2norm.py:80:17 + div.full.f32 %r67, %r300, %r291; + div.full.f32 %r68, %r300, %r292; + div.full.f32 %r69, %r300, %r293; + div.full.f32 %r70, %r300, %r294; + div.full.f32 %r72, %r300, %r295; + div.full.f32 %r73, %r300, %r296; + div.full.f32 %r74, %r300, %r297; + div.full.f32 %r75, %r300, %r298; + .loc 1 81 16 // l2norm.py:81:16 + mul.f32 %r301, %r67, %r91; + mul.f32 %r302, %r67, %r92; + mul.f32 %r303, %r67, %r93; + mul.f32 %r304, %r67, %r94; + mul.f32 %r305, %r67, %r95; + mul.f32 %r306, %r67, %r96; + mul.f32 %r307, %r67, %r97; + mul.f32 %r308, %r67, %r98; + mul.f32 %r309, %r68, %r99; + mul.f32 %r310, %r68, %r100; + mul.f32 %r311, %r68, %r101; + mul.f32 %r312, %r68, %r102; + mul.f32 %r313, %r68, %r103; + mul.f32 %r314, %r68, %r104; + mul.f32 %r315, %r68, %r105; + mul.f32 %r316, %r68, %r106; + mul.f32 %r317, %r69, %r107; + mul.f32 %r318, %r69, %r108; + mul.f32 %r319, %r69, %r109; + mul.f32 %r320, %r69, %r110; + mul.f32 %r321, %r69, %r111; + mul.f32 %r322, %r69, %r112; + mul.f32 %r323, %r69, %r113; + mul.f32 %r324, %r69, %r114; + mul.f32 %r325, %r70, %r115; + mul.f32 %r326, %r70, %r116; + mul.f32 %r327, %r70, %r117; + mul.f32 %r328, %r70, %r118; + mul.f32 %r329, %r70, %r119; + mul.f32 %r330, %r70, %r120; + mul.f32 %r331, %r70, %r121; + mul.f32 %r332, %r70, %r122; + mul.f32 %r333, %r72, %r123; + mul.f32 %r334, %r72, %r124; + mul.f32 %r335, %r72, %r125; + mul.f32 %r336, %r72, %r126; + mul.f32 %r337, %r72, %r127; + mul.f32 %r338, %r72, %r128; + mul.f32 %r339, %r72, %r129; + mul.f32 %r340, %r72, %r130; + mul.f32 %r341, %r73, %r131; + mul.f32 %r342, %r73, %r132; + mul.f32 %r343, %r73, %r133; + mul.f32 %r344, %r73, %r134; + mul.f32 %r345, %r73, %r135; + mul.f32 %r346, %r73, %r136; + mul.f32 %r347, %r73, %r137; + mul.f32 %r348, %r73, %r138; + mul.f32 %r349, %r74, %r139; + mul.f32 %r350, %r74, %r140; + mul.f32 %r351, %r74, %r141; + mul.f32 %r352, %r74, %r142; + mul.f32 %r353, %r74, %r143; + mul.f32 %r354, %r74, %r144; + mul.f32 %r355, %r74, %r145; + mul.f32 %r356, %r74, %r146; + mul.f32 %r357, %r75, %r147; + mul.f32 %r358, %r75, %r148; + mul.f32 %r359, %r75, %r149; + mul.f32 %r360, %r75, %r150; + mul.f32 %r361, %r75, %r151; + mul.f32 %r362, %r75, %r152; + mul.f32 %r363, %r75, %r153; + mul.f32 %r364, %r75, %r154; + .loc 1 83 18 // l2norm.py:83:18 + cvt.rn.bf16x2.f32 %r34, %r302, %r301; + cvt.rn.bf16x2.f32 %r35, %r304, %r303; + cvt.rn.bf16x2.f32 %r36, %r306, %r305; + cvt.rn.bf16x2.f32 %r37, %r308, %r307; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd9 + 0 ], { %r34, %r35, %r36, %r37 }; + // end inline asm + cvt.rn.bf16x2.f32 %r38, %r310, %r309; + cvt.rn.bf16x2.f32 %r39, %r312, %r311; + cvt.rn.bf16x2.f32 %r40, %r314, %r313; + cvt.rn.bf16x2.f32 %r41, %r316, %r315; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd10 + 0 ], { %r38, %r39, %r40, %r41 }; + // end inline asm + cvt.rn.bf16x2.f32 %r42, %r318, %r317; + cvt.rn.bf16x2.f32 %r43, %r320, %r319; + cvt.rn.bf16x2.f32 %r44, %r322, %r321; + cvt.rn.bf16x2.f32 %r45, %r324, %r323; + // begin inline asm + @%p3 st.global.v4.b32 [ %rd11 + 0 ], { %r42, %r43, %r44, %r45 }; + // end inline asm + cvt.rn.bf16x2.f32 %r46, %r326, %r325; + cvt.rn.bf16x2.f32 %r47, %r328, %r327; + cvt.rn.bf16x2.f32 %r48, %r330, %r329; + cvt.rn.bf16x2.f32 %r49, %r332, %r331; + // begin inline asm + @%p4 st.global.v4.b32 [ %rd12 + 0 ], { %r46, %r47, %r48, %r49 }; + // end inline asm + cvt.rn.bf16x2.f32 %r50, %r334, %r333; + cvt.rn.bf16x2.f32 %r51, %r336, %r335; + cvt.rn.bf16x2.f32 %r52, %r338, %r337; + cvt.rn.bf16x2.f32 %r53, %r340, %r339; + // begin inline asm + @%p5 st.global.v4.b32 [ %rd13 + 0 ], { %r50, %r51, %r52, %r53 }; + // end inline asm + cvt.rn.bf16x2.f32 %r54, %r342, %r341; + cvt.rn.bf16x2.f32 %r55, %r344, %r343; + cvt.rn.bf16x2.f32 %r56, %r346, %r345; + cvt.rn.bf16x2.f32 %r57, %r348, %r347; + // begin inline asm + @%p6 st.global.v4.b32 [ %rd14 + 0 ], { %r54, %r55, %r56, %r57 }; + // end inline asm + cvt.rn.bf16x2.f32 %r58, %r350, %r349; + cvt.rn.bf16x2.f32 %r59, %r352, %r351; + cvt.rn.bf16x2.f32 %r60, %r354, %r353; + cvt.rn.bf16x2.f32 %r61, %r356, %r355; + // begin inline asm + @%p7 st.global.v4.b32 [ %rd15 + 0 ], { %r58, %r59, %r60, %r61 }; + // end inline asm + cvt.rn.bf16x2.f32 %r62, %r358, %r357; + cvt.rn.bf16x2.f32 %r63, %r360, %r359; + cvt.rn.bf16x2.f32 %r64, %r362, %r361; + cvt.rn.bf16x2.f32 %r65, %r364, %r363; + // begin inline asm + @%p8 st.global.v4.b32 [ %rd16 + 0 ], { %r62, %r63, %r64, %r65 }; + // end inline asm + .loc 1 84 21 // l2norm.py:84:21 + shl.b32 %r365, %r80, 1; + mov.b32 %r366, global_smem; + add.s32 %r66, %r366, %r365; + // begin inline asm + st.shared.v4.b32 [ %r66 + 0 ], { %r67, %r68, %r69, %r70 }; + // end inline asm + add.s32 %r71, %r66, 16; + // begin inline asm + st.shared.v4.b32 [ %r71 + 0 ], { %r72, %r73, %r74, %r75 }; + // end inline asm + bar.sync 0; + shl.b32 %r367, %r79, 5; + and.b32 %r368, %r367, 96; + and.b32 %r369, %r79, 28; + add.s32 %r370, %r366, %r368; + add.s32 %r371, %r370, %r369; + ld.shared.b32 %r76, [%r371]; + and.b32 %r372, %r79, 32; + setp.eq.b32 %p11, %r372, 0; + and.pred %p9, %p11, %p10; + // begin inline asm + @%p9 st.global.b32 [ %rd17 + 0 ], { %r76 }; + // end inline asm + .loc 1 84 4 // l2norm.py:84:4 + ret; +$L__tmp67: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp66 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 80 // DW_AT_call_line +.b8 43 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp66 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/l2norm_fwd_kernel.source b/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/l2norm_fwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..d9fa83aac0600b1e5d35d95ed8aaad5c5636e577 --- /dev/null +++ b/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/l2norm_fwd_kernel.source @@ -0,0 +1,172 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc38 = loc(unknown) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc45 = loc("x"(#loc)) +#loc46 = loc("y"(#loc)) +#loc47 = loc("rstd"(#loc)) +#loc48 = loc("eps"(#loc)) +#loc49 = loc("T"(#loc)) +#loc81 = loc("input"(#loc36)) +#loc82 = loc("a"(#loc41)) +#loc83 = loc("b"(#loc41)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc50) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc51) + %o_t = arith.constant 32 : i32 loc(#loc52) + %o_t_1 = arith.constant 32 : i64 loc(#loc52) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc52) + %o_t_3 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32> loc(#loc53) + %o_t_4 = arith.extsi %o_t_3 : tensor<32xi32> to tensor<32xi64> loc(#loc54) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<32xi64> loc(#loc54) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<32xi64> loc(#loc54) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc55) + %m_t = arith.extsi %T : i32 to i64 loc(#loc56) + %m_t_7 = tt.splat %m_t : i64 -> tensor<32xi64> loc(#loc56) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<32xi64> loc(#loc56) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<32xi1> -> tensor<32x1xi1> loc(#loc57) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc58) + %m_x_10 = arith.constant 128 : i32 loc(#loc59) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc59) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc59) + %m_x_13 = tt.broadcast %m_x : tensor<32x1xi1> -> tensor<32x128xi1> loc(#loc60) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<32x128xi1> loc(#loc60) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<32x128xi1> loc(#loc60) + %p_x = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<32xi64> -> tensor<32x1xi64> loc(#loc61) + %p_x_16 = arith.constant 128 : i32 loc(#loc62) + %p_x_17 = arith.constant 128 : i64 loc(#loc62) + %p_x_18 = arith.constant dense<128> : tensor<32x1xi64> loc(#loc62) + %p_x_19 = arith.muli %p_x, %p_x_18 : tensor<32x1xi64> loc(#loc62) + %p_x_20 = tt.splat %x : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc63) + %p_x_21 = tt.addptr %p_x_20, %p_x_19 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc63) + %p_x_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc64) + %p_x_23 = tt.broadcast %p_x_21 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc65) + %p_x_24 = tt.broadcast %p_x_22 : tensor<1x128xi32> -> tensor<32x128xi32> loc(#loc65) + %p_x_25 = tt.addptr %p_x_23, %p_x_24 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc65) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<32xi64> -> tensor<32x1xi64> loc(#loc66) + %p_y_26 = arith.constant 128 : i32 loc(#loc67) + %p_y_27 = arith.constant 128 : i64 loc(#loc67) + %p_y_28 = arith.constant dense<128> : tensor<32x1xi64> loc(#loc67) + %p_y_29 = arith.muli %p_y, %p_y_28 : tensor<32x1xi64> loc(#loc67) + %p_y_30 = tt.splat %y : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc68) + %p_y_31 = tt.addptr %p_y_30, %p_y_29 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc68) + %p_y_32 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc69) + %p_y_33 = tt.broadcast %p_y_31 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc70) + %p_y_34 = tt.broadcast %p_y_32 : tensor<1x128xi32> -> tensor<32x128xi32> loc(#loc70) + %p_y_35 = tt.addptr %p_y_33, %p_y_34 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc70) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<32x!tt.ptr> loc(#loc71) + %p_rstd_36 = tt.addptr %p_rstd, %o_t_6 : tensor<32x!tt.ptr>, tensor<32xi64> loc(#loc71) + %b_x = arith.constant 0.000000e+00 : f32 loc(#loc72) + %b_x_37 = arith.constant dense<0.000000e+00> : tensor<32x128xf32> loc(#loc72) + %b_x_38 = arith.truncf %b_x_37 : tensor<32x128xf32> to tensor<32x128xbf16> loc(#loc72) + %b_x_39 = tt.load %p_x_25, %m_x_15, %b_x_38 : tensor<32x128x!tt.ptr> loc(#loc72) + %b_x_40 = arith.extf %b_x_39 : tensor<32x128xbf16> to tensor<32x128xf32> loc(#loc73) + %b_rstd = arith.mulf %b_x_40, %b_x_40 : tensor<32x128xf32> loc(#loc74) + %b_rstd_41 = tt.call @triton.language.standard.sum__fp32S32_128S_c1_cFalse_cNone(%b_rstd) : (tensor<32x128xf32>) -> tensor<32xf32> loc(#loc75) + %b_rstd_42 = tt.splat %eps : f32 -> tensor<32xf32> loc(#loc76) + %b_rstd_43 = arith.addf %b_rstd_41, %b_rstd_42 : tensor<32xf32> loc(#loc76) + %b_rstd_44 = math.sqrt %b_rstd_43 : tensor<32xf32> loc(#loc77) + %b_rstd_45 = arith.constant 1 : i32 loc(#loc78) + %b_rstd_46 = arith.constant 1.000000e+00 : f32 loc(#loc78) + %b_rstd_47 = arith.constant dense<1.000000e+00> : tensor<32xf32> loc(#loc78) + %b_rstd_48 = arith.divf %b_rstd_47, %b_rstd_44 : tensor<32xf32> loc(#loc78) + %b_y = tt.expand_dims %b_rstd_48 {axis = 1 : i32} : tensor<32xf32> -> tensor<32x1xf32> loc(#loc79) + %b_y_49 = tt.broadcast %b_y : tensor<32x1xf32> -> tensor<32x128xf32> loc(#loc80) + %b_y_50 = arith.mulf %b_x_40, %b_y_49 : tensor<32x128xf32> loc(#loc80) + %0 = arith.truncf %b_y_50 : tensor<32x128xf32> to tensor<32x128xbf16> loc(#loc32) + tt.store %p_y_35, %0, %m_x_15 : tensor<32x128x!tt.ptr> loc(#loc33) + tt.store %p_rstd_36, %b_rstd_48, %m_t_8 : tensor<32x!tt.ptr> loc(#loc34) + tt.return loc(#loc35) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S32_128S_c1_cFalse_cNone(%input: tensor<32x128xf32> loc("input"(#loc36))) -> tensor<32xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc37) + tt.reduce.return %2 : f32 loc(#loc37) + }) : (tensor<32x128xf32>) -> tensor<32xf32> loc(#loc37) + tt.return %0 : tensor<32xf32> loc(#loc39) + ^bb1: // no predecessors + %1 = ub.poison : tensor<32xf32> loc(#loc40) + tt.return %1 : tensor<32xf32> loc(#loc40) + } loc(#loc36) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc41)), %b: f32 loc("b"(#loc41))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc42) + tt.return %0 : f32 loc(#loc43) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc44) + tt.return %1 : f32 loc(#loc44) + } loc(#loc41) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:18) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:29) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:37) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc50 = loc("i_t"(#loc1)) +#loc51 = loc("i_t"(#loc2)) +#loc52 = loc("o_t"(#loc3)) +#loc53 = loc("o_t"(#loc4)) +#loc54 = loc("o_t"(#loc5)) +#loc55 = loc("o_d"(#loc6)) +#loc56 = loc("m_t"(#loc7)) +#loc57 = loc("m_x"(#loc8)) +#loc58 = loc("m_x"(#loc9)) +#loc59 = loc("m_x"(#loc10)) +#loc60 = loc("m_x"(#loc11)) +#loc61 = loc("p_x"(#loc12)) +#loc62 = loc("p_x"(#loc13)) +#loc63 = loc("p_x"(#loc14)) +#loc64 = loc("p_x"(#loc15)) +#loc65 = loc("p_x"(#loc16)) +#loc66 = loc("p_y"(#loc17)) +#loc67 = loc("p_y"(#loc18)) +#loc68 = loc("p_y"(#loc19)) +#loc69 = loc("p_y"(#loc20)) +#loc70 = loc("p_y"(#loc21)) +#loc71 = loc("p_rstd"(#loc22)) +#loc72 = loc("b_x"(#loc23)) +#loc73 = loc("b_x"(#loc24)) +#loc74 = loc("b_rstd"(#loc25)) +#loc75 = loc("b_rstd"(#loc26)) +#loc76 = loc("b_rstd"(#loc27)) +#loc77 = loc("b_rstd"(#loc28)) +#loc78 = loc("b_rstd"(#loc29)) +#loc79 = loc("b_y"(#loc30)) +#loc80 = loc("b_y"(#loc31)) diff --git a/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/l2norm_fwd_kernel.ttgir b/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/l2norm_fwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..b137a7870e377a1a99f2fc1d8c75cb728717eea2 --- /dev/null +++ b/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/l2norm_fwd_kernel.ttgir @@ -0,0 +1,135 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [2, 1], order = [1, 0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [2], order = [0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc1 = loc(unknown) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc34 = loc("x"(#loc)) +#loc35 = loc("y"(#loc)) +#loc36 = loc("rstd"(#loc)) +#loc37 = loc("eps"(#loc)) +#loc38 = loc("T"(#loc)) +#loc59 = loc("b_rstd"(#loc23)) +#loc66 = loc(callsite(#loc1 at #loc59)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 2 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<1.000000e+00> : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<32x128xbf16, #blocked> loc(#loc1) + %c32_i64 = arith.constant 32 : i64 loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<32x1xi64, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc39) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc40) + %o_t = arith.muli %i_t_3, %c32_i64 : i64 loc(#loc41) + %o_t_4 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc42) + %o_t_5 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32, #blocked1> loc(#loc42) + %o_t_6 = arith.extsi %o_t_4 : tensor<32xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_7 = arith.extsi %o_t_5 : tensor<32xi32, #blocked1> to tensor<32xi64, #blocked1> loc(#loc43) + %o_t_8 = tt.splat %o_t : i64 -> tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_9 = tt.splat %o_t : i64 -> tensor<32xi64, #blocked1> loc(#loc43) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<32xi64, #blocked1> loc(#loc43) + %m_t = arith.extsi %T : i32 to i64 loc(#loc44) + %m_t_12 = tt.splat %m_t : i64 -> tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_13 = tt.splat %m_t : i64 -> tensor<32xi64, #blocked1> loc(#loc44) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<32xi64, #blocked1> loc(#loc44) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<32xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32x1xi1, #blocked> loc(#loc45) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc46) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc46) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_1 : tensor<1x128xi32, #blocked> loc(#loc47) + %m_x_19 = tt.broadcast %m_x : tensor<32x1xi1, #blocked> -> tensor<32x128xi1, #blocked> loc(#loc48) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<32x128xi1, #blocked> loc(#loc48) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<32x128xi1, #blocked> loc(#loc48) + %p_x = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32x1xi64, #blocked> loc(#loc49) + %p_x_22 = arith.muli %p_x, %cst_2 : tensor<32x1xi64, #blocked> loc(#loc50) + %p_x_23 = tt.splat %x : !tt.ptr -> tensor<32x1x!tt.ptr, #blocked> loc(#loc51) + %p_x_24 = tt.addptr %p_x_23, %p_x_22 : tensor<32x1x!tt.ptr, #blocked>, tensor<32x1xi64, #blocked> loc(#loc51) + %p_x_25 = tt.broadcast %p_x_24 : tensor<32x1x!tt.ptr, #blocked> -> tensor<32x128x!tt.ptr, #blocked> loc(#loc52) + %p_x_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<32x128xi32, #blocked> loc(#loc52) + %p_x_27 = tt.addptr %p_x_25, %p_x_26 : tensor<32x128x!tt.ptr, #blocked>, tensor<32x128xi32, #blocked> loc(#loc52) + %p_y = tt.splat %y : !tt.ptr -> tensor<32x1x!tt.ptr, #blocked> loc(#loc53) + %p_y_28 = tt.addptr %p_y, %p_x_22 : tensor<32x1x!tt.ptr, #blocked>, tensor<32x1xi64, #blocked> loc(#loc53) + %p_y_29 = tt.broadcast %p_y_28 : tensor<32x1x!tt.ptr, #blocked> -> tensor<32x128x!tt.ptr, #blocked> loc(#loc54) + %p_y_30 = tt.addptr %p_y_29, %p_x_26 : tensor<32x128x!tt.ptr, #blocked>, tensor<32x128xi32, #blocked> loc(#loc54) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<32x!tt.ptr, #blocked1> loc(#loc55) + %p_rstd_31 = tt.addptr %p_rstd, %o_t_11 : tensor<32x!tt.ptr, #blocked1>, tensor<32xi64, #blocked1> loc(#loc55) + %b_x = tt.load %p_x_27, %m_x_21, %cst_0 : tensor<32x128x!tt.ptr, #blocked> loc(#loc56) + %b_x_32 = arith.extf %b_x : tensor<32x128xbf16, #blocked> to tensor<32x128xf32, #blocked> loc(#loc57) + %b_rstd = arith.mulf %b_x_32, %b_x_32 : tensor<32x128xf32, #blocked> loc(#loc58) + %b_rstd_33 = "tt.reduce"(%b_rstd) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_40: f32 loc(callsite(#loc1 at #loc59)), %b_rstd_41: f32 loc(callsite(#loc1 at #loc59))): + %b_rstd_42 = arith.addf %b_rstd_40, %b_rstd_41 : f32 loc(#loc67) + tt.reduce.return %b_rstd_42 : f32 loc(#loc65) + }) : (tensor<32x128xf32, #blocked>) -> tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc65) + %b_rstd_34 = tt.splat %eps : f32 -> tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_36 = math.sqrt %b_rstd_35 : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc61) + %b_rstd_37 = arith.divf %cst, %b_rstd_36 : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc62) + %b_y = tt.expand_dims %b_rstd_37 {axis = 1 : i32} : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32x1xf32, #blocked> loc(#loc63) + %b_y_38 = tt.broadcast %b_y : tensor<32x1xf32, #blocked> -> tensor<32x128xf32, #blocked> loc(#loc64) + %b_y_39 = arith.mulf %b_x_32, %b_y_38 : tensor<32x128xf32, #blocked> loc(#loc64) + %0 = arith.truncf %b_y_39 : tensor<32x128xf32, #blocked> to tensor<32x128xbf16, #blocked> loc(#loc30) + tt.store %p_y_30, %0, %m_x_21 : tensor<32x128x!tt.ptr, #blocked> loc(#loc31) + %1 = ttg.convert_layout %b_rstd_37 : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32xf32, #blocked1> loc(#loc32) + tt.store %p_rstd_31, %1, %m_t_15 : tensor<32x!tt.ptr, #blocked1> loc(#loc32) + tt.return loc(#loc33) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc39 = loc("i_t"(#loc2)) +#loc40 = loc("i_t"(#loc3)) +#loc41 = loc("o_t"(#loc4)) +#loc42 = loc("o_t"(#loc5)) +#loc43 = loc("o_t"(#loc6)) +#loc44 = loc("m_t"(#loc7)) +#loc45 = loc("m_x"(#loc8)) +#loc46 = loc("m_x"(#loc9)) +#loc47 = loc("m_x"(#loc10)) +#loc48 = loc("m_x"(#loc11)) +#loc49 = loc("p_x"(#loc12)) +#loc50 = loc("p_x"(#loc13)) +#loc51 = loc("p_x"(#loc14)) +#loc52 = loc("p_x"(#loc15)) +#loc53 = loc("p_y"(#loc16)) +#loc54 = loc("p_y"(#loc17)) +#loc55 = loc("p_rstd"(#loc18)) +#loc56 = loc("b_x"(#loc19)) +#loc57 = loc("b_x"(#loc20)) +#loc58 = loc("b_rstd"(#loc21)) +#loc60 = loc("b_rstd"(#loc25)) +#loc61 = loc("b_rstd"(#loc26)) +#loc62 = loc("b_rstd"(#loc27)) +#loc63 = loc("b_y"(#loc28)) +#loc64 = loc("b_y"(#loc29)) +#loc65 = loc(callsite(#loc22 at #loc59)) +#loc67 = loc(callsite(#loc24 at #loc65)) diff --git a/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/l2norm_fwd_kernel.ttir b/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/l2norm_fwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..3688281bceaf7f8abdf8568104090be5e0170141 --- /dev/null +++ b/triton/gpu0/CDCLQFVP7MWEUNTGKYIRDCRRIV6YKBC4NHBWBIK3B4X4HFEVG6XQ/l2norm_fwd_kernel.ttir @@ -0,0 +1,128 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc3 = loc(unknown) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc35 = loc("x"(#loc)) +#loc36 = loc("y"(#loc)) +#loc37 = loc("rstd"(#loc)) +#loc38 = loc("eps"(#loc)) +#loc39 = loc("T"(#loc)) +#loc62 = loc("b_rstd"(#loc25)) +#loc68 = loc(callsite(#loc3 at #loc62)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %b_x = arith.constant dense<0.000000e+00> : tensor<32x128xbf16> loc(#loc40) + %b_rstd = arith.constant dense<1.000000e+00> : tensor<32xf32> loc(#loc41) + %cst = arith.constant dense<128> : tensor<32x1xi64> loc(#loc3) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc42) + %o_t = arith.constant 32 : i64 loc(#loc43) + %i_t = tt.get_program_id x : i32 loc(#loc44) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc45) + %o_t_1 = arith.muli %i_t_0, %o_t : i64 loc(#loc43) + %o_t_2 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32> loc(#loc46) + %o_t_3 = arith.extsi %o_t_2 : tensor<32xi32> to tensor<32xi64> loc(#loc47) + %o_t_4 = tt.splat %o_t_1 : i64 -> tensor<32xi64> loc(#loc47) + %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<32xi64> loc(#loc47) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc48) + %m_t = arith.extsi %T : i32 to i64 loc(#loc49) + %m_t_6 = tt.splat %m_t : i64 -> tensor<32xi64> loc(#loc49) + %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<32xi64> loc(#loc49) + %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<32xi1> -> tensor<32x1xi1> loc(#loc50) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc51) + %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc42) + %m_x_11 = tt.broadcast %m_x_8 : tensor<32x1xi1> -> tensor<32x128xi1> loc(#loc52) + %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<32x128xi1> loc(#loc52) + %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<32x128xi1> loc(#loc52) + %p_x = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<32xi64> -> tensor<32x1xi64> loc(#loc53) + %p_x_14 = arith.muli %p_x, %cst : tensor<32x1xi64> loc(#loc54) + %p_x_15 = tt.splat %x : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc55) + %p_x_16 = tt.addptr %p_x_15, %p_x_14 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc55) + %p_x_17 = tt.broadcast %p_x_16 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc56) + %p_x_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<32x128xi32> loc(#loc56) + %p_x_19 = tt.addptr %p_x_17, %p_x_18 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc56) + %p_y = tt.splat %y : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc57) + %p_y_20 = tt.addptr %p_y, %p_x_14 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc57) + %p_y_21 = tt.broadcast %p_y_20 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc58) + %p_y_22 = tt.addptr %p_y_21, %p_x_18 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<32x!tt.ptr> loc(#loc59) + %p_rstd_23 = tt.addptr %p_rstd, %o_t_5 : tensor<32x!tt.ptr>, tensor<32xi64> loc(#loc59) + %b_x_24 = tt.load %p_x_19, %m_x_13, %b_x : tensor<32x128x!tt.ptr> loc(#loc40) + %b_x_25 = arith.extf %b_x_24 : tensor<32x128xbf16> to tensor<32x128xf32> loc(#loc60) + %b_rstd_26 = arith.mulf %b_x_25, %b_x_25 : tensor<32x128xf32> loc(#loc61) + %b_rstd_27 = "tt.reduce"(%b_rstd_26) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_34: f32 loc(callsite(#loc3 at #loc62)), %b_rstd_35: f32 loc(callsite(#loc3 at #loc62))): + %b_rstd_36 = arith.addf %b_rstd_34, %b_rstd_35 : f32 loc(#loc69) + tt.reduce.return %b_rstd_36 : f32 loc(#loc67) + }) : (tensor<32x128xf32>) -> tensor<32xf32> loc(#loc67) + %b_rstd_28 = tt.splat %eps : f32 -> tensor<32xf32> loc(#loc63) + %b_rstd_29 = arith.addf %b_rstd_27, %b_rstd_28 : tensor<32xf32> loc(#loc63) + %b_rstd_30 = math.sqrt %b_rstd_29 : tensor<32xf32> loc(#loc64) + %b_rstd_31 = arith.divf %b_rstd, %b_rstd_30 : tensor<32xf32> loc(#loc41) + %b_y = tt.expand_dims %b_rstd_31 {axis = 1 : i32} : tensor<32xf32> -> tensor<32x1xf32> loc(#loc65) + %b_y_32 = tt.broadcast %b_y : tensor<32x1xf32> -> tensor<32x128xf32> loc(#loc66) + %b_y_33 = arith.mulf %b_x_25, %b_y_32 : tensor<32x128xf32> loc(#loc66) + %0 = arith.truncf %b_y_33 : tensor<32x128xf32> to tensor<32x128xbf16> loc(#loc31) + tt.store %p_y_22, %0, %m_x_13 : tensor<32x128x!tt.ptr> loc(#loc32) + tt.store %p_rstd_23, %b_rstd_31, %m_t_7 : tensor<32x!tt.ptr> loc(#loc33) + tt.return loc(#loc34) + } loc(#loc) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc40 = loc("b_x"(#loc1)) +#loc41 = loc("b_rstd"(#loc2)) +#loc42 = loc("m_x"(#loc4)) +#loc43 = loc("o_t"(#loc5)) +#loc44 = loc("i_t"(#loc6)) +#loc45 = loc("i_t"(#loc7)) +#loc46 = loc("o_t"(#loc8)) +#loc47 = loc("o_t"(#loc9)) +#loc48 = loc("o_d"(#loc10)) +#loc49 = loc("m_t"(#loc11)) +#loc50 = loc("m_x"(#loc12)) +#loc51 = loc("m_x"(#loc13)) +#loc52 = loc("m_x"(#loc14)) +#loc53 = loc("p_x"(#loc15)) +#loc54 = loc("p_x"(#loc16)) +#loc55 = loc("p_x"(#loc17)) +#loc56 = loc("p_x"(#loc18)) +#loc57 = loc("p_y"(#loc19)) +#loc58 = loc("p_y"(#loc20)) +#loc59 = loc("p_rstd"(#loc21)) +#loc60 = loc("b_x"(#loc22)) +#loc61 = loc("b_rstd"(#loc23)) +#loc63 = loc("b_rstd"(#loc27)) +#loc64 = loc("b_rstd"(#loc28)) +#loc65 = loc("b_y"(#loc29)) +#loc66 = loc("b_y"(#loc30)) +#loc67 = loc(callsite(#loc24 at #loc62)) +#loc69 = loc(callsite(#loc26 at #loc67)) diff --git a/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/__grp__l2norm_fwd_kernel.json b/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/__grp__l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..4d7227d02bb797413cb261308d5b823d7e49e44e --- /dev/null +++ b/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/__grp__l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_fwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/l2norm_fwd_kernel.source", "l2norm_fwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/l2norm_fwd_kernel.ttir", "l2norm_fwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/l2norm_fwd_kernel.ttgir", "l2norm_fwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/l2norm_fwd_kernel.llir", "l2norm_fwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/l2norm_fwd_kernel.ptx", "l2norm_fwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/l2norm_fwd_kernel.cubin", "l2norm_fwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/l2norm_fwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/l2norm_fwd_kernel.cubin b/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/l2norm_fwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..d21943cbb588cffc6c90d5a9a5b8738c44506d0d Binary files /dev/null and b/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/l2norm_fwd_kernel.cubin differ diff --git a/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/l2norm_fwd_kernel.json b/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..0b96734873a0d0a03449d136c729278b07d2b7ac --- /dev/null +++ b/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"hash": "10dff723476ef7b2d16c390b01e73872a4fb70e2490b387de0d693639e5c52f4", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 4, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 32, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_fwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/l2norm_fwd_kernel.llir b/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/l2norm_fwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..639e9082bdf936e023ff352182cf149819527738 --- /dev/null +++ b/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/l2norm_fwd_kernel.llir @@ -0,0 +1,249 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 +@.str.2 = private unnamed_addr constant [17 x i8] c"__CUDA_PREC_SQRT\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_fwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, float %3, i32 %4, ptr addrspace(1) readnone captures(none) %5, ptr addrspace(1) readnone captures(none) %6) local_unnamed_addr #0 !dbg !5 { + %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %9 = zext nneg i32 %8 to i64, !dbg !14 + %10 = shl nuw nsw i64 %9, 3, !dbg !15 + %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !16 + %12 = and i32 %11, 112, !dbg !16 + %13 = lshr exact i32 %12, 4, !dbg !16 + %14 = zext nneg i32 %13 to i64, !dbg !17 + %15 = or disjoint i64 %10, %14, !dbg !17 + %16 = sext i32 %4 to i64, !dbg !18 + %17 = icmp slt i64 %15, %16, !dbg !18 + %18 = shl nuw nsw i32 %11, 3, !dbg !19 + %19 = and i32 %18, 120, !dbg !19 + %20 = shl nuw nsw i64 %15, 7, !dbg !20 + %21 = getelementptr bfloat, ptr addrspace(1) %0, i64 %20, !dbg !21 + %22 = zext nneg i32 %19 to i64, !dbg !22 + %23 = getelementptr bfloat, ptr addrspace(1) %21, i64 %22, !dbg !22 + %24 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %23, i1 %17) #6, !dbg !23 + %25 = extractvalue { i32, i32, i32, i32 } %24, 0, !dbg !23 + %26 = bitcast i32 %25 to <2 x bfloat>, !dbg !23 + %27 = extractvalue { i32, i32, i32, i32 } %24, 1, !dbg !23 + %28 = bitcast i32 %27 to <2 x bfloat>, !dbg !23 + %29 = extractvalue { i32, i32, i32, i32 } %24, 2, !dbg !23 + %30 = bitcast i32 %29 to <2 x bfloat>, !dbg !23 + %31 = extractvalue { i32, i32, i32, i32 } %24, 3, !dbg !23 + %32 = bitcast i32 %31 to <2 x bfloat>, !dbg !23 + %33 = extractelement <2 x bfloat> %26, i64 0, !dbg !23 + %34 = extractelement <2 x bfloat> %26, i64 1, !dbg !23 + %35 = extractelement <2 x bfloat> %28, i64 0, !dbg !23 + %36 = extractelement <2 x bfloat> %28, i64 1, !dbg !23 + %37 = extractelement <2 x bfloat> %30, i64 0, !dbg !23 + %38 = extractelement <2 x bfloat> %30, i64 1, !dbg !23 + %39 = extractelement <2 x bfloat> %32, i64 0, !dbg !23 + %40 = extractelement <2 x bfloat> %32, i64 1, !dbg !23 + %41 = fpext bfloat %33 to float, !dbg !24 + %42 = fpext bfloat %34 to float, !dbg !24 + %43 = fpext bfloat %35 to float, !dbg !24 + %44 = fpext bfloat %36 to float, !dbg !24 + %45 = fpext bfloat %37 to float, !dbg !24 + %46 = fpext bfloat %38 to float, !dbg !24 + %47 = fpext bfloat %39 to float, !dbg !24 + %48 = fpext bfloat %40 to float, !dbg !24 + %49 = fmul float %41, %41, !dbg !25 + %50 = fmul float %42, %42, !dbg !25 + %51 = fmul float %43, %43, !dbg !25 + %52 = fmul float %44, %44, !dbg !25 + %53 = fmul float %45, %45, !dbg !25 + %54 = fmul float %46, %46, !dbg !25 + %55 = fmul float %47, %47, !dbg !25 + %56 = fmul float %48, %48, !dbg !25 + %57 = fadd float %49, %50, !dbg !26 + %58 = fadd float %51, %57, !dbg !26 + %59 = fadd float %52, %58, !dbg !26 + %60 = fadd float %53, %59, !dbg !26 + %61 = fadd float %54, %60, !dbg !26 + %62 = fadd float %55, %61, !dbg !26 + %63 = fadd float %56, %62, !dbg !26 + %64 = bitcast float %63 to i32, !dbg !29 + %65 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %64, i32 8, i32 31), !dbg !29 + %66 = bitcast i32 %65 to float, !dbg !29 + %67 = fadd float %63, %66, !dbg !26 + %68 = bitcast float %67 to i32, !dbg !29 + %69 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %68, i32 4, i32 31), !dbg !29 + %70 = bitcast i32 %69 to float, !dbg !29 + %71 = fadd float %67, %70, !dbg !26 + %72 = bitcast float %71 to i32, !dbg !29 + %73 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %72, i32 2, i32 31), !dbg !29 + %74 = bitcast i32 %73 to float, !dbg !29 + %75 = fadd float %71, %74, !dbg !26 + %76 = bitcast float %75 to i32, !dbg !29 + %77 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %76, i32 1, i32 31), !dbg !29 + %78 = bitcast i32 %77 to float, !dbg !29 + %79 = fadd float %75, %78, !dbg !26 + %80 = fadd float %3, %79, !dbg !32 + %81 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i = icmp eq i32 %81, 0, !dbg !33 + %82 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i = icmp eq i32 %82, 0, !dbg !33 + br i1 %.not.i, label %88, label %83, !dbg !33 + +83: ; preds = %7 + br i1 %.not1.i, label %86, label %84, !dbg !33 + +84: ; preds = %83 + %85 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %80) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +86: ; preds = %83 + %87 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %80) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +88: ; preds = %7 + br i1 %.not1.i, label %91, label %89, !dbg !33 + +89: ; preds = %88 + %90 = tail call float @llvm.nvvm.sqrt.rn.f(float %80) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +91: ; preds = %88 + %92 = tail call float @llvm.nvvm.sqrt.approx.f(float %80) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +__nv_sqrtf.exit: ; preds = %84, %86, %89, %91 + %.0.i = phi float [ %85, %84 ], [ %87, %86 ], [ %90, %89 ], [ %92, %91 ], !dbg !33 + %93 = and i32 %11, 7, !dbg !16 + %94 = zext nneg i32 %93 to i64, !dbg !17 + %95 = or disjoint i64 %10, %94, !dbg !17 + %96 = getelementptr float, ptr addrspace(1) %2, i64 %95, !dbg !34 + %97 = getelementptr bfloat, ptr addrspace(1) %1, i64 %20, !dbg !35 + %98 = getelementptr bfloat, ptr addrspace(1) %97, i64 %22, !dbg !36 + %99 = icmp slt i64 %95, %16, !dbg !18 + %100 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i), !dbg !37 + %101 = fmul float %100, %41, !dbg !38 + %102 = fmul float %100, %42, !dbg !38 + %103 = fmul float %100, %43, !dbg !38 + %104 = fmul float %100, %44, !dbg !38 + %105 = fmul float %100, %45, !dbg !38 + %106 = fmul float %100, %46, !dbg !38 + %107 = fmul float %100, %47, !dbg !38 + %108 = fmul float %100, %48, !dbg !38 + %109 = fptrunc float %101 to bfloat, !dbg !39 + %110 = fptrunc float %102 to bfloat, !dbg !39 + %111 = fptrunc float %103 to bfloat, !dbg !39 + %112 = fptrunc float %104 to bfloat, !dbg !39 + %113 = fptrunc float %105 to bfloat, !dbg !39 + %114 = fptrunc float %106 to bfloat, !dbg !39 + %115 = fptrunc float %107 to bfloat, !dbg !39 + %116 = fptrunc float %108 to bfloat, !dbg !39 + %117 = insertelement <2 x bfloat> poison, bfloat %109, i64 0, !dbg !40 + %118 = insertelement <2 x bfloat> %117, bfloat %110, i64 1, !dbg !40 + %119 = bitcast <2 x bfloat> %118 to i32, !dbg !40 + %120 = insertelement <2 x bfloat> poison, bfloat %111, i64 0, !dbg !40 + %121 = insertelement <2 x bfloat> %120, bfloat %112, i64 1, !dbg !40 + %122 = bitcast <2 x bfloat> %121 to i32, !dbg !40 + %123 = insertelement <2 x bfloat> poison, bfloat %113, i64 0, !dbg !40 + %124 = insertelement <2 x bfloat> %123, bfloat %114, i64 1, !dbg !40 + %125 = bitcast <2 x bfloat> %124 to i32, !dbg !40 + %126 = insertelement <2 x bfloat> poison, bfloat %115, i64 0, !dbg !40 + %127 = insertelement <2 x bfloat> %126, bfloat %116, i64 1, !dbg !40 + %128 = bitcast <2 x bfloat> %127 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %119, i32 %122, i32 %125, i32 %128, ptr addrspace(1) %98, i1 %17) #6, !dbg !40 + %129 = lshr exact i32 %12, 2, !dbg !41 + %130 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %129, !dbg !41 + %131 = bitcast float %100 to <1 x i32>, !dbg !41 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) %130, <1 x i32> %131) #6, !dbg !41 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !41 + %132 = shl nuw nsw i32 %93, 2, !dbg !41 + %133 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %132, !dbg !41 + %134 = load i32, ptr addrspace(3) %133, align 4, !dbg !41 + %135 = and i32 %11, 120, !dbg !41 + %136 = icmp eq i32 %135, 0, !dbg !41 + %137 = and i1 %136, %99, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %134, ptr addrspace(1) %96, i1 %137) #6, !dbg !41 + ret void, !dbg !42 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.div.full(float, float) #3 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #4 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #5 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.f(float) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="128" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) } +attributes #4 = { convergent nocallback nounwind } +attributes #5 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #6 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "l2norm_fwd_kernel", linkageName: "l2norm_fwd_kernel", scope: !1, file: !1, line: 69, type: !6, scopeLine: 69, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !10, !11, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 70, column: 24, scope: !5) +!14 = !DILocation(line: 70, column: 30, scope: !5) +!15 = !DILocation(line: 71, column: 16, scope: !5) +!16 = !DILocation(line: 71, column: 34, scope: !5) +!17 = !DILocation(line: 71, column: 21, scope: !5) +!18 = !DILocation(line: 73, column: 16, scope: !5) +!19 = !DILocation(line: 74, column: 30, scope: !5) +!20 = !DILocation(line: 75, column: 29, scope: !5) +!21 = !DILocation(line: 75, column: 14, scope: !5) +!22 = !DILocation(line: 75, column: 33, scope: !5) +!23 = !DILocation(line: 79, column: 18, scope: !5) +!24 = !DILocation(line: 79, column: 47, scope: !5) +!25 = !DILocation(line: 80, column: 38, scope: !5) +!26 = !DILocation(line: 263, column: 15, scope: !27, inlinedAt: !29) +!27 = distinct !DILexicalBlockFile(scope: !5, file: !28, discriminator: 0) +!28 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!29 = !DILocation(line: 293, column: 36, scope: !27, inlinedAt: !30) +!30 = !DILocation(line: 80, column: 43, scope: !31) +!31 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!32 = !DILocation(line: 80, column: 48, scope: !5) +!33 = !DILocation(line: 80, column: 25, scope: !5) +!34 = !DILocation(line: 77, column: 20, scope: !5) +!35 = !DILocation(line: 76, column: 14, scope: !5) +!36 = !DILocation(line: 76, column: 33, scope: !5) +!37 = !DILocation(line: 80, column: 17, scope: !5) +!38 = !DILocation(line: 81, column: 16, scope: !5) +!39 = !DILocation(line: 83, column: 25, scope: !5) +!40 = !DILocation(line: 83, column: 18, scope: !5) +!41 = !DILocation(line: 84, column: 21, scope: !5) +!42 = !DILocation(line: 84, column: 4, scope: !5) diff --git a/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/l2norm_fwd_kernel.ptx b/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/l2norm_fwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..4aa67ae72e3c97c8320f79c6dfc50a50840500f4 --- /dev/null +++ b/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/l2norm_fwd_kernel.ptx @@ -0,0 +1,399 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_fwd_kernel // -- Begin function l2norm_fwd_kernel +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; +.global .align 1 .b8 _$_str_$_2[17] = {95, 95, 67, 85, 68, 65, 95, 80, 82, 69, 67, 95, 83, 81, 82, 84}; + // @l2norm_fwd_kernel +.visible .entry l2norm_fwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_2, + .param .f32 l2norm_fwd_kernel_param_3, + .param .u32 l2norm_fwd_kernel_param_4, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_6 +) +.reqntid 128 +{ + .reg .pred %p<5>; + .reg .b16 %rs<9>; + .reg .b32 %r<61>; + .reg .b64 %rd<18>; + .loc 1 69 0 // l2norm.py:69:0 +$L__func_begin0: + .loc 1 69 0 // l2norm.py:69:0 + +// %bb.0: // %__nv_sqrtf.exit + ld.param.b64 %rd4, [l2norm_fwd_kernel_param_0]; + ld.param.b64 %rd5, [l2norm_fwd_kernel_param_1]; +$L__tmp0: + .loc 1 70 24 // l2norm.py:70:24 + mov.u32 %r13, %ctaid.x; + ld.param.b64 %rd6, [l2norm_fwd_kernel_param_2]; + .loc 1 71 16 // l2norm.py:71:16 + mul.wide.u32 %rd7, %r13, 8; + ld.param.b32 %r14, [l2norm_fwd_kernel_param_3]; + ld.param.s32 %rd8, [l2norm_fwd_kernel_param_4]; + .loc 1 71 34 // l2norm.py:71:34 + mov.u32 %r15, %tid.x; + and.b32 %r16, %r15, 112; + bfe.u32 %r17, %r15, 4, 3; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd9, %r17; + or.b64 %rd10, %rd7, %rd9; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p1, %rd10, %rd8; + .loc 1 74 30 // l2norm.py:74:30 + shl.b32 %r18, %r15, 3; + and.b32 %r19, %r18, 120; + .loc 1 75 14 // l2norm.py:75:14 + shl.b64 %rd11, %rd10, 8; + add.s64 %rd12, %rd4, %rd11; + .loc 1 75 33 // l2norm.py:75:33 + mul.wide.u32 %rd13, %r19, 2; + add.s64 %rd1, %rd12, %rd13; + mov.b32 %r5, 0; + .loc 1 79 18 // l2norm.py:79:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + mov.b32 {%rs5, %rs6}, %r3; + mov.b32 {%rs7, %rs8}, %r4; + .loc 1 79 47 // l2norm.py:79:47 + cvt.f32.bf16 %r20, %rs1; + cvt.f32.bf16 %r21, %rs2; + cvt.f32.bf16 %r22, %rs3; + cvt.f32.bf16 %r23, %rs4; + cvt.f32.bf16 %r24, %rs5; + cvt.f32.bf16 %r25, %rs6; + cvt.f32.bf16 %r26, %rs7; + cvt.f32.bf16 %r27, %rs8; + .loc 1 80 38 // l2norm.py:80:38 + mul.f32 %r28, %r21, %r21; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + fma.rn.f32.bf16 %r29, %rs1, %rs1, %r28; + fma.rn.f32.bf16 %r30, %rs3, %rs3, %r29; + fma.rn.f32.bf16 %r31, %rs4, %rs4, %r30; + fma.rn.f32.bf16 %r32, %rs5, %rs5, %r31; + fma.rn.f32.bf16 %r33, %rs6, %rs6, %r32; + fma.rn.f32.bf16 %r34, %rs7, %rs7, %r33; + fma.rn.f32.bf16 %r35, %rs8, %rs8, %r34; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r36, %r35, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r37, %r35, %r36; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r38, %r37, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r39, %r37, %r38; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r40, %r39, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r41, %r39, %r40; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r42, %r41, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r43, %r41, %r42; +$L__tmp10: + .loc 1 80 48 // l2norm.py:80:48 + add.f32 %r44, %r14, %r43; + .loc 1 80 25 // l2norm.py:80:25 + sqrt.approx.ftz.f32 %r45, %r44; + .loc 1 71 34 // l2norm.py:71:34 + and.b32 %r46, %r15, 7; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd14, %r46; + or.b64 %rd15, %rd7, %rd14; + .loc 1 77 20 // l2norm.py:77:20 + shl.b64 %rd16, %rd15, 2; + add.s64 %rd3, %rd6, %rd16; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd17, %rd5, %rd11; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd2, %rd17, %rd13; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p3, %rd15, %rd8; + mov.b32 %r47, 0f3F800000; + .loc 1 80 17 // l2norm.py:80:17 + div.full.f32 %r11, %r47, %r45; + .loc 1 81 16 // l2norm.py:81:16 + mul.f32 %r48, %r11, %r20; + mul.f32 %r49, %r11, %r21; + mul.f32 %r50, %r11, %r22; + mul.f32 %r51, %r11, %r23; + mul.f32 %r52, %r11, %r24; + mul.f32 %r53, %r11, %r25; + mul.f32 %r54, %r11, %r26; + mul.f32 %r55, %r11, %r27; + .loc 1 83 18 // l2norm.py:83:18 + cvt.rn.bf16x2.f32 %r6, %r49, %r48; + cvt.rn.bf16x2.f32 %r7, %r51, %r50; + cvt.rn.bf16x2.f32 %r8, %r53, %r52; + cvt.rn.bf16x2.f32 %r9, %r55, %r54; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd2 + 0 ], { %r6, %r7, %r8, %r9 }; + // end inline asm + .loc 1 84 21 // l2norm.py:84:21 + shr.u32 %r56, %r16, 2; + mov.b32 %r57, global_smem; + add.s32 %r10, %r57, %r56; + // begin inline asm + st.shared.b32 [ %r10 + 0 ], %r11; + // end inline asm + bar.sync 0; + shl.b32 %r58, %r46, 2; + add.s32 %r59, %r57, %r58; + ld.shared.b32 %r12, [%r59]; + and.b32 %r60, %r15, 120; + setp.eq.b32 %p4, %r60, 0; + and.pred %p2, %p4, %p3; + // begin inline asm + @%p2 st.global.b32 [ %rd3 + 0 ], { %r12 }; + // end inline asm + .loc 1 84 4 // l2norm.py:84:4 + ret; +$L__tmp11: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp10 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 80 // DW_AT_call_line +.b8 43 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp10 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/l2norm_fwd_kernel.source b/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/l2norm_fwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..c7c5a735c81381ccb0efd0d2a8f8d86240068e11 --- /dev/null +++ b/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/l2norm_fwd_kernel.source @@ -0,0 +1,172 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc38 = loc(unknown) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc45 = loc("x"(#loc)) +#loc46 = loc("y"(#loc)) +#loc47 = loc("rstd"(#loc)) +#loc48 = loc("eps"(#loc)) +#loc49 = loc("T"(#loc)) +#loc81 = loc("input"(#loc36)) +#loc82 = loc("a"(#loc41)) +#loc83 = loc("b"(#loc41)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc50) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc51) + %o_t = arith.constant 8 : i32 loc(#loc52) + %o_t_1 = arith.constant 8 : i64 loc(#loc52) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc52) + %o_t_3 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32> loc(#loc53) + %o_t_4 = arith.extsi %o_t_3 : tensor<8xi32> to tensor<8xi64> loc(#loc54) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<8xi64> loc(#loc54) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<8xi64> loc(#loc54) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc55) + %m_t = arith.extsi %T : i32 to i64 loc(#loc56) + %m_t_7 = tt.splat %m_t : i64 -> tensor<8xi64> loc(#loc56) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<8xi64> loc(#loc56) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<8xi1> -> tensor<8x1xi1> loc(#loc57) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc58) + %m_x_10 = arith.constant 128 : i32 loc(#loc59) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc59) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc59) + %m_x_13 = tt.broadcast %m_x : tensor<8x1xi1> -> tensor<8x128xi1> loc(#loc60) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<8x128xi1> loc(#loc60) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<8x128xi1> loc(#loc60) + %p_x = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc61) + %p_x_16 = arith.constant 128 : i32 loc(#loc62) + %p_x_17 = arith.constant 128 : i64 loc(#loc62) + %p_x_18 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc62) + %p_x_19 = arith.muli %p_x, %p_x_18 : tensor<8x1xi64> loc(#loc62) + %p_x_20 = tt.splat %x : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc63) + %p_x_21 = tt.addptr %p_x_20, %p_x_19 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc63) + %p_x_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc64) + %p_x_23 = tt.broadcast %p_x_21 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc65) + %p_x_24 = tt.broadcast %p_x_22 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc65) + %p_x_25 = tt.addptr %p_x_23, %p_x_24 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc65) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc66) + %p_y_26 = arith.constant 128 : i32 loc(#loc67) + %p_y_27 = arith.constant 128 : i64 loc(#loc67) + %p_y_28 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc67) + %p_y_29 = arith.muli %p_y, %p_y_28 : tensor<8x1xi64> loc(#loc67) + %p_y_30 = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc68) + %p_y_31 = tt.addptr %p_y_30, %p_y_29 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc68) + %p_y_32 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc69) + %p_y_33 = tt.broadcast %p_y_31 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc70) + %p_y_34 = tt.broadcast %p_y_32 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc70) + %p_y_35 = tt.addptr %p_y_33, %p_y_34 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc70) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr> loc(#loc71) + %p_rstd_36 = tt.addptr %p_rstd, %o_t_6 : tensor<8x!tt.ptr>, tensor<8xi64> loc(#loc71) + %b_x = arith.constant 0.000000e+00 : f32 loc(#loc72) + %b_x_37 = arith.constant dense<0.000000e+00> : tensor<8x128xf32> loc(#loc72) + %b_x_38 = arith.truncf %b_x_37 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc72) + %b_x_39 = tt.load %p_x_25, %m_x_15, %b_x_38 : tensor<8x128x!tt.ptr> loc(#loc72) + %b_x_40 = arith.extf %b_x_39 : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc73) + %b_rstd = arith.mulf %b_x_40, %b_x_40 : tensor<8x128xf32> loc(#loc74) + %b_rstd_41 = tt.call @triton.language.standard.sum__fp32S8_128S_c1_cFalse_cNone(%b_rstd) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc75) + %b_rstd_42 = tt.splat %eps : f32 -> tensor<8xf32> loc(#loc76) + %b_rstd_43 = arith.addf %b_rstd_41, %b_rstd_42 : tensor<8xf32> loc(#loc76) + %b_rstd_44 = math.sqrt %b_rstd_43 : tensor<8xf32> loc(#loc77) + %b_rstd_45 = arith.constant 1 : i32 loc(#loc78) + %b_rstd_46 = arith.constant 1.000000e+00 : f32 loc(#loc78) + %b_rstd_47 = arith.constant dense<1.000000e+00> : tensor<8xf32> loc(#loc78) + %b_rstd_48 = arith.divf %b_rstd_47, %b_rstd_44 : tensor<8xf32> loc(#loc78) + %b_y = tt.expand_dims %b_rstd_48 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc79) + %b_y_49 = tt.broadcast %b_y : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc80) + %b_y_50 = arith.mulf %b_x_40, %b_y_49 : tensor<8x128xf32> loc(#loc80) + %0 = arith.truncf %b_y_50 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc32) + tt.store %p_y_35, %0, %m_x_15 : tensor<8x128x!tt.ptr> loc(#loc33) + tt.store %p_rstd_36, %b_rstd_48, %m_t_8 : tensor<8x!tt.ptr> loc(#loc34) + tt.return loc(#loc35) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S8_128S_c1_cFalse_cNone(%input: tensor<8x128xf32> loc("input"(#loc36))) -> tensor<8xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc37) + tt.reduce.return %2 : f32 loc(#loc37) + }) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc37) + tt.return %0 : tensor<8xf32> loc(#loc39) + ^bb1: // no predecessors + %1 = ub.poison : tensor<8xf32> loc(#loc40) + tt.return %1 : tensor<8xf32> loc(#loc40) + } loc(#loc36) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc41)), %b: f32 loc("b"(#loc41))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc42) + tt.return %0 : f32 loc(#loc43) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc44) + tt.return %1 : f32 loc(#loc44) + } loc(#loc41) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:18) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:29) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:37) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc50 = loc("i_t"(#loc1)) +#loc51 = loc("i_t"(#loc2)) +#loc52 = loc("o_t"(#loc3)) +#loc53 = loc("o_t"(#loc4)) +#loc54 = loc("o_t"(#loc5)) +#loc55 = loc("o_d"(#loc6)) +#loc56 = loc("m_t"(#loc7)) +#loc57 = loc("m_x"(#loc8)) +#loc58 = loc("m_x"(#loc9)) +#loc59 = loc("m_x"(#loc10)) +#loc60 = loc("m_x"(#loc11)) +#loc61 = loc("p_x"(#loc12)) +#loc62 = loc("p_x"(#loc13)) +#loc63 = loc("p_x"(#loc14)) +#loc64 = loc("p_x"(#loc15)) +#loc65 = loc("p_x"(#loc16)) +#loc66 = loc("p_y"(#loc17)) +#loc67 = loc("p_y"(#loc18)) +#loc68 = loc("p_y"(#loc19)) +#loc69 = loc("p_y"(#loc20)) +#loc70 = loc("p_y"(#loc21)) +#loc71 = loc("p_rstd"(#loc22)) +#loc72 = loc("b_x"(#loc23)) +#loc73 = loc("b_x"(#loc24)) +#loc74 = loc("b_rstd"(#loc25)) +#loc75 = loc("b_rstd"(#loc26)) +#loc76 = loc("b_rstd"(#loc27)) +#loc77 = loc("b_rstd"(#loc28)) +#loc78 = loc("b_rstd"(#loc29)) +#loc79 = loc("b_y"(#loc30)) +#loc80 = loc("b_y"(#loc31)) diff --git a/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/l2norm_fwd_kernel.ttgir b/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/l2norm_fwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..72b0f913da168c2c233286b0953ca7e41e1d4463 --- /dev/null +++ b/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/l2norm_fwd_kernel.ttgir @@ -0,0 +1,135 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [4, 1], order = [1, 0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [4], order = [0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc1 = loc(unknown) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc34 = loc("x"(#loc)) +#loc35 = loc("y"(#loc)) +#loc36 = loc("rstd"(#loc)) +#loc37 = loc("eps"(#loc)) +#loc38 = loc("T"(#loc)) +#loc59 = loc("b_rstd"(#loc23)) +#loc66 = loc(callsite(#loc1 at #loc59)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 4 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<1.000000e+00> : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<8x128xbf16, #blocked> loc(#loc1) + %c8_i64 = arith.constant 8 : i64 loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<8x1xi64, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc39) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc40) + %o_t = arith.muli %i_t_3, %c8_i64 : i64 loc(#loc41) + %o_t_4 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc42) + %o_t_5 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32, #blocked1> loc(#loc42) + %o_t_6 = arith.extsi %o_t_4 : tensor<8xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_7 = arith.extsi %o_t_5 : tensor<8xi32, #blocked1> to tensor<8xi64, #blocked1> loc(#loc43) + %o_t_8 = tt.splat %o_t : i64 -> tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_9 = tt.splat %o_t : i64 -> tensor<8xi64, #blocked1> loc(#loc43) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<8xi64, #blocked1> loc(#loc43) + %m_t = arith.extsi %T : i32 to i64 loc(#loc44) + %m_t_12 = tt.splat %m_t : i64 -> tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_13 = tt.splat %m_t : i64 -> tensor<8xi64, #blocked1> loc(#loc44) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<8xi64, #blocked1> loc(#loc44) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<8xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xi1, #blocked> loc(#loc45) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc46) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc46) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_1 : tensor<1x128xi32, #blocked> loc(#loc47) + %m_x_19 = tt.broadcast %m_x : tensor<8x1xi1, #blocked> -> tensor<8x128xi1, #blocked> loc(#loc48) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<8x128xi1, #blocked> loc(#loc48) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<8x128xi1, #blocked> loc(#loc48) + %p_x = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xi64, #blocked> loc(#loc49) + %p_x_22 = arith.muli %p_x, %cst_2 : tensor<8x1xi64, #blocked> loc(#loc50) + %p_x_23 = tt.splat %x : !tt.ptr -> tensor<8x1x!tt.ptr, #blocked> loc(#loc51) + %p_x_24 = tt.addptr %p_x_23, %p_x_22 : tensor<8x1x!tt.ptr, #blocked>, tensor<8x1xi64, #blocked> loc(#loc51) + %p_x_25 = tt.broadcast %p_x_24 : tensor<8x1x!tt.ptr, #blocked> -> tensor<8x128x!tt.ptr, #blocked> loc(#loc52) + %p_x_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<8x128xi32, #blocked> loc(#loc52) + %p_x_27 = tt.addptr %p_x_25, %p_x_26 : tensor<8x128x!tt.ptr, #blocked>, tensor<8x128xi32, #blocked> loc(#loc52) + %p_y = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr, #blocked> loc(#loc53) + %p_y_28 = tt.addptr %p_y, %p_x_22 : tensor<8x1x!tt.ptr, #blocked>, tensor<8x1xi64, #blocked> loc(#loc53) + %p_y_29 = tt.broadcast %p_y_28 : tensor<8x1x!tt.ptr, #blocked> -> tensor<8x128x!tt.ptr, #blocked> loc(#loc54) + %p_y_30 = tt.addptr %p_y_29, %p_x_26 : tensor<8x128x!tt.ptr, #blocked>, tensor<8x128xi32, #blocked> loc(#loc54) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr, #blocked1> loc(#loc55) + %p_rstd_31 = tt.addptr %p_rstd, %o_t_11 : tensor<8x!tt.ptr, #blocked1>, tensor<8xi64, #blocked1> loc(#loc55) + %b_x = tt.load %p_x_27, %m_x_21, %cst_0 : tensor<8x128x!tt.ptr, #blocked> loc(#loc56) + %b_x_32 = arith.extf %b_x : tensor<8x128xbf16, #blocked> to tensor<8x128xf32, #blocked> loc(#loc57) + %b_rstd = arith.mulf %b_x_32, %b_x_32 : tensor<8x128xf32, #blocked> loc(#loc58) + %b_rstd_33 = "tt.reduce"(%b_rstd) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_40: f32 loc(callsite(#loc1 at #loc59)), %b_rstd_41: f32 loc(callsite(#loc1 at #loc59))): + %b_rstd_42 = arith.addf %b_rstd_40, %b_rstd_41 : f32 loc(#loc67) + tt.reduce.return %b_rstd_42 : f32 loc(#loc65) + }) : (tensor<8x128xf32, #blocked>) -> tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc65) + %b_rstd_34 = tt.splat %eps : f32 -> tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_36 = math.sqrt %b_rstd_35 : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc61) + %b_rstd_37 = arith.divf %cst, %b_rstd_36 : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc62) + %b_y = tt.expand_dims %b_rstd_37 {axis = 1 : i32} : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xf32, #blocked> loc(#loc63) + %b_y_38 = tt.broadcast %b_y : tensor<8x1xf32, #blocked> -> tensor<8x128xf32, #blocked> loc(#loc64) + %b_y_39 = arith.mulf %b_x_32, %b_y_38 : tensor<8x128xf32, #blocked> loc(#loc64) + %0 = arith.truncf %b_y_39 : tensor<8x128xf32, #blocked> to tensor<8x128xbf16, #blocked> loc(#loc30) + tt.store %p_y_30, %0, %m_x_21 : tensor<8x128x!tt.ptr, #blocked> loc(#loc31) + %1 = ttg.convert_layout %b_rstd_37 : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8xf32, #blocked1> loc(#loc32) + tt.store %p_rstd_31, %1, %m_t_15 : tensor<8x!tt.ptr, #blocked1> loc(#loc32) + tt.return loc(#loc33) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc39 = loc("i_t"(#loc2)) +#loc40 = loc("i_t"(#loc3)) +#loc41 = loc("o_t"(#loc4)) +#loc42 = loc("o_t"(#loc5)) +#loc43 = loc("o_t"(#loc6)) +#loc44 = loc("m_t"(#loc7)) +#loc45 = loc("m_x"(#loc8)) +#loc46 = loc("m_x"(#loc9)) +#loc47 = loc("m_x"(#loc10)) +#loc48 = loc("m_x"(#loc11)) +#loc49 = loc("p_x"(#loc12)) +#loc50 = loc("p_x"(#loc13)) +#loc51 = loc("p_x"(#loc14)) +#loc52 = loc("p_x"(#loc15)) +#loc53 = loc("p_y"(#loc16)) +#loc54 = loc("p_y"(#loc17)) +#loc55 = loc("p_rstd"(#loc18)) +#loc56 = loc("b_x"(#loc19)) +#loc57 = loc("b_x"(#loc20)) +#loc58 = loc("b_rstd"(#loc21)) +#loc60 = loc("b_rstd"(#loc25)) +#loc61 = loc("b_rstd"(#loc26)) +#loc62 = loc("b_rstd"(#loc27)) +#loc63 = loc("b_y"(#loc28)) +#loc64 = loc("b_y"(#loc29)) +#loc65 = loc(callsite(#loc22 at #loc59)) +#loc67 = loc(callsite(#loc24 at #loc65)) diff --git a/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/l2norm_fwd_kernel.ttir b/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/l2norm_fwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..8bbf17da02489a0d195ed0bd020180f9e4ac2496 --- /dev/null +++ b/triton/gpu0/CDP7OI2HN333FULMHEFQDZZYOKSPW4HCJEFTQ7PA22JWHHS4KL2A/l2norm_fwd_kernel.ttir @@ -0,0 +1,128 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc3 = loc(unknown) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc35 = loc("x"(#loc)) +#loc36 = loc("y"(#loc)) +#loc37 = loc("rstd"(#loc)) +#loc38 = loc("eps"(#loc)) +#loc39 = loc("T"(#loc)) +#loc62 = loc("b_rstd"(#loc25)) +#loc68 = loc(callsite(#loc3 at #loc62)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %b_x = arith.constant dense<0.000000e+00> : tensor<8x128xbf16> loc(#loc40) + %b_rstd = arith.constant dense<1.000000e+00> : tensor<8xf32> loc(#loc41) + %cst = arith.constant dense<128> : tensor<8x1xi64> loc(#loc3) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc42) + %o_t = arith.constant 8 : i64 loc(#loc43) + %i_t = tt.get_program_id x : i32 loc(#loc44) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc45) + %o_t_1 = arith.muli %i_t_0, %o_t : i64 loc(#loc43) + %o_t_2 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32> loc(#loc46) + %o_t_3 = arith.extsi %o_t_2 : tensor<8xi32> to tensor<8xi64> loc(#loc47) + %o_t_4 = tt.splat %o_t_1 : i64 -> tensor<8xi64> loc(#loc47) + %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<8xi64> loc(#loc47) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc48) + %m_t = arith.extsi %T : i32 to i64 loc(#loc49) + %m_t_6 = tt.splat %m_t : i64 -> tensor<8xi64> loc(#loc49) + %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<8xi64> loc(#loc49) + %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<8xi1> -> tensor<8x1xi1> loc(#loc50) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc51) + %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc42) + %m_x_11 = tt.broadcast %m_x_8 : tensor<8x1xi1> -> tensor<8x128xi1> loc(#loc52) + %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<8x128xi1> loc(#loc52) + %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<8x128xi1> loc(#loc52) + %p_x = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc53) + %p_x_14 = arith.muli %p_x, %cst : tensor<8x1xi64> loc(#loc54) + %p_x_15 = tt.splat %x : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc55) + %p_x_16 = tt.addptr %p_x_15, %p_x_14 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc55) + %p_x_17 = tt.broadcast %p_x_16 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc56) + %p_x_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc56) + %p_x_19 = tt.addptr %p_x_17, %p_x_18 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc56) + %p_y = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc57) + %p_y_20 = tt.addptr %p_y, %p_x_14 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc57) + %p_y_21 = tt.broadcast %p_y_20 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc58) + %p_y_22 = tt.addptr %p_y_21, %p_x_18 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr> loc(#loc59) + %p_rstd_23 = tt.addptr %p_rstd, %o_t_5 : tensor<8x!tt.ptr>, tensor<8xi64> loc(#loc59) + %b_x_24 = tt.load %p_x_19, %m_x_13, %b_x : tensor<8x128x!tt.ptr> loc(#loc40) + %b_x_25 = arith.extf %b_x_24 : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc60) + %b_rstd_26 = arith.mulf %b_x_25, %b_x_25 : tensor<8x128xf32> loc(#loc61) + %b_rstd_27 = "tt.reduce"(%b_rstd_26) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_34: f32 loc(callsite(#loc3 at #loc62)), %b_rstd_35: f32 loc(callsite(#loc3 at #loc62))): + %b_rstd_36 = arith.addf %b_rstd_34, %b_rstd_35 : f32 loc(#loc69) + tt.reduce.return %b_rstd_36 : f32 loc(#loc67) + }) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc67) + %b_rstd_28 = tt.splat %eps : f32 -> tensor<8xf32> loc(#loc63) + %b_rstd_29 = arith.addf %b_rstd_27, %b_rstd_28 : tensor<8xf32> loc(#loc63) + %b_rstd_30 = math.sqrt %b_rstd_29 : tensor<8xf32> loc(#loc64) + %b_rstd_31 = arith.divf %b_rstd, %b_rstd_30 : tensor<8xf32> loc(#loc41) + %b_y = tt.expand_dims %b_rstd_31 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc65) + %b_y_32 = tt.broadcast %b_y : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc66) + %b_y_33 = arith.mulf %b_x_25, %b_y_32 : tensor<8x128xf32> loc(#loc66) + %0 = arith.truncf %b_y_33 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc31) + tt.store %p_y_22, %0, %m_x_13 : tensor<8x128x!tt.ptr> loc(#loc32) + tt.store %p_rstd_23, %b_rstd_31, %m_t_7 : tensor<8x!tt.ptr> loc(#loc33) + tt.return loc(#loc34) + } loc(#loc) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc40 = loc("b_x"(#loc1)) +#loc41 = loc("b_rstd"(#loc2)) +#loc42 = loc("m_x"(#loc4)) +#loc43 = loc("o_t"(#loc5)) +#loc44 = loc("i_t"(#loc6)) +#loc45 = loc("i_t"(#loc7)) +#loc46 = loc("o_t"(#loc8)) +#loc47 = loc("o_t"(#loc9)) +#loc48 = loc("o_d"(#loc10)) +#loc49 = loc("m_t"(#loc11)) +#loc50 = loc("m_x"(#loc12)) +#loc51 = loc("m_x"(#loc13)) +#loc52 = loc("m_x"(#loc14)) +#loc53 = loc("p_x"(#loc15)) +#loc54 = loc("p_x"(#loc16)) +#loc55 = loc("p_x"(#loc17)) +#loc56 = loc("p_x"(#loc18)) +#loc57 = loc("p_y"(#loc19)) +#loc58 = loc("p_y"(#loc20)) +#loc59 = loc("p_rstd"(#loc21)) +#loc60 = loc("b_x"(#loc22)) +#loc61 = loc("b_rstd"(#loc23)) +#loc63 = loc("b_rstd"(#loc27)) +#loc64 = loc("b_rstd"(#loc28)) +#loc65 = loc("b_y"(#loc29)) +#loc66 = loc("b_y"(#loc30)) +#loc67 = loc(callsite(#loc24 at #loc62)) +#loc69 = loc(callsite(#loc26 at #loc67)) diff --git a/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/__grp__l2norm_bwd_kernel.json b/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/__grp__l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..60e498fe4938af4d6b011a4267d79121f8c6e194 --- /dev/null +++ b/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/__grp__l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_bwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/l2norm_bwd_kernel.source", "l2norm_bwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/l2norm_bwd_kernel.ttir", "l2norm_bwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/l2norm_bwd_kernel.ttgir", "l2norm_bwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/l2norm_bwd_kernel.llir", "l2norm_bwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/l2norm_bwd_kernel.ptx", "l2norm_bwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/l2norm_bwd_kernel.cubin", "l2norm_bwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/l2norm_bwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/l2norm_bwd_kernel.cubin b/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/l2norm_bwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..498032cb3c3c433335094bccd52803a8da45db27 Binary files /dev/null and b/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/l2norm_bwd_kernel.cubin differ diff --git a/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/l2norm_bwd_kernel.json b/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..a102fc03bc0ef3a5d0ef86f0cb56db5a42a81e34 --- /dev/null +++ b/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"hash": "1110096f4990c250aeb8664f97b7a68df450327099fa16da515e11afe2689be2", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 16, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 0, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_bwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/l2norm_bwd_kernel.llir b/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/l2norm_bwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..ca8dee858015e7fc1cb948b8ca9720fa7a8871d6 --- /dev/null +++ b/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/l2norm_bwd_kernel.llir @@ -0,0 +1,175 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_bwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, ptr addrspace(1) %3, float %4, i32 %5, ptr addrspace(1) readnone captures(none) %6, ptr addrspace(1) readnone captures(none) %7) local_unnamed_addr #0 !dbg !4 { + %9 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !12 + %10 = zext nneg i32 %9 to i64, !dbg !13 + %11 = shl nuw nsw i64 %10, 4, !dbg !14 + %12 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !15 + %13 = lshr i32 %12, 5, !dbg !15 + %14 = and i32 %13, 15, !dbg !15 + %15 = zext nneg i32 %14 to i64, !dbg !16 + %16 = or disjoint i64 %11, %15, !dbg !16 + %17 = sext i32 %5 to i64, !dbg !17 + %18 = icmp slt i64 %16, %17, !dbg !17 + %19 = shl nuw nsw i32 %12, 2, !dbg !18 + %20 = and i32 %19, 124, !dbg !18 + %21 = shl nuw nsw i64 %16, 7, !dbg !19 + %22 = getelementptr bfloat, ptr addrspace(1) %0, i64 %21, !dbg !20 + %23 = zext nneg i32 %20 to i64, !dbg !21 + %24 = getelementptr bfloat, ptr addrspace(1) %22, i64 %23, !dbg !21 + %25 = getelementptr float, ptr addrspace(1) %1, i64 %16, !dbg !22 + %26 = getelementptr bfloat, ptr addrspace(1) %2, i64 %21, !dbg !23 + %27 = getelementptr bfloat, ptr addrspace(1) %26, i64 %23, !dbg !24 + %28 = getelementptr bfloat, ptr addrspace(1) %3, i64 %21, !dbg !25 + %29 = getelementptr bfloat, ptr addrspace(1) %28, i64 %23, !dbg !26 + %30 = tail call { i32, i32 } asm sideeffect "mov.u32 $0, $2;\0A\09mov.u32 $1, $3;\0A\09@$5 ld.global.v2.b32 { $0, $1 }, [ $4 + 0 ];", "=r,=r,r,r,l,b"(i32 0, i32 0, ptr addrspace(1) %24, i1 %18) #3, !dbg !27 + %31 = extractvalue { i32, i32 } %30, 0, !dbg !27 + %32 = bitcast i32 %31 to <2 x bfloat>, !dbg !27 + %33 = extractvalue { i32, i32 } %30, 1, !dbg !27 + %34 = bitcast i32 %33 to <2 x bfloat>, !dbg !27 + %35 = extractelement <2 x bfloat> %32, i64 0, !dbg !27 + %36 = extractelement <2 x bfloat> %32, i64 1, !dbg !27 + %37 = extractelement <2 x bfloat> %34, i64 0, !dbg !27 + %38 = extractelement <2 x bfloat> %34, i64 1, !dbg !27 + %39 = fpext bfloat %35 to float, !dbg !28 + %40 = fpext bfloat %36 to float, !dbg !28 + %41 = fpext bfloat %37 to float, !dbg !28 + %42 = fpext bfloat %38 to float, !dbg !28 + %43 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %25, i1 %18) #3, !dbg !29 + %44 = bitcast i32 %43 to float, !dbg !29 + %45 = tail call { i32, i32 } asm sideeffect "mov.u32 $0, $2;\0A\09mov.u32 $1, $3;\0A\09@$5 ld.global.v2.b32 { $0, $1 }, [ $4 + 0 ];", "=r,=r,r,r,l,b"(i32 0, i32 0, ptr addrspace(1) %27, i1 %18) #3, !dbg !30 + %46 = extractvalue { i32, i32 } %45, 0, !dbg !30 + %47 = bitcast i32 %46 to <2 x bfloat>, !dbg !30 + %48 = extractvalue { i32, i32 } %45, 1, !dbg !30 + %49 = bitcast i32 %48 to <2 x bfloat>, !dbg !30 + %50 = extractelement <2 x bfloat> %47, i64 0, !dbg !30 + %51 = extractelement <2 x bfloat> %47, i64 1, !dbg !30 + %52 = extractelement <2 x bfloat> %49, i64 0, !dbg !30 + %53 = extractelement <2 x bfloat> %49, i64 1, !dbg !30 + %54 = fpext bfloat %50 to float, !dbg !31 + %55 = fpext bfloat %51 to float, !dbg !31 + %56 = fpext bfloat %52 to float, !dbg !31 + %57 = fpext bfloat %53 to float, !dbg !31 + %58 = fmul float %44, %54, !dbg !32 + %59 = fmul float %44, %55, !dbg !32 + %60 = fmul float %44, %56, !dbg !32 + %61 = fmul float %44, %57, !dbg !32 + %62 = fmul float %39, %54, !dbg !33 + %63 = fmul float %40, %55, !dbg !33 + %64 = fmul float %41, %56, !dbg !33 + %65 = fmul float %42, %57, !dbg !33 + %66 = fadd float %62, %63, !dbg !34 + %67 = fadd float %64, %66, !dbg !34 + %68 = fadd float %65, %67, !dbg !34 + %69 = bitcast float %68 to i32, !dbg !37 + %70 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %69, i32 16, i32 31), !dbg !37 + %71 = bitcast i32 %70 to float, !dbg !37 + %72 = fadd float %68, %71, !dbg !34 + %73 = bitcast float %72 to i32, !dbg !37 + %74 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %73, i32 8, i32 31), !dbg !37 + %75 = bitcast i32 %74 to float, !dbg !37 + %76 = fadd float %72, %75, !dbg !34 + %77 = bitcast float %76 to i32, !dbg !37 + %78 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %77, i32 4, i32 31), !dbg !37 + %79 = bitcast i32 %78 to float, !dbg !37 + %80 = fadd float %76, %79, !dbg !34 + %81 = bitcast float %80 to i32, !dbg !37 + %82 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %81, i32 2, i32 31), !dbg !37 + %83 = bitcast i32 %82 to float, !dbg !37 + %84 = fadd float %80, %83, !dbg !34 + %85 = bitcast float %84 to i32, !dbg !37 + %86 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %85, i32 1, i32 31), !dbg !37 + %87 = bitcast i32 %86 to float, !dbg !37 + %88 = fadd float %84, %87, !dbg !34 + %89 = fmul float %88, %39, !dbg !40 + %90 = fmul float %88, %40, !dbg !40 + %91 = fmul float %88, %41, !dbg !40 + %92 = fmul float %88, %42, !dbg !40 + %93 = fmul float %89, %44, !dbg !41 + %94 = fmul float %90, %44, !dbg !41 + %95 = fmul float %91, %44, !dbg !41 + %96 = fmul float %92, %44, !dbg !41 + %97 = fsub float %58, %93, !dbg !42 + %98 = fsub float %59, %94, !dbg !42 + %99 = fsub float %60, %95, !dbg !42 + %100 = fsub float %61, %96, !dbg !42 + %101 = fptrunc float %97 to bfloat, !dbg !43 + %102 = fptrunc float %98 to bfloat, !dbg !43 + %103 = fptrunc float %99 to bfloat, !dbg !43 + %104 = fptrunc float %100 to bfloat, !dbg !43 + %105 = insertelement <2 x bfloat> poison, bfloat %101, i64 0, !dbg !44 + %106 = insertelement <2 x bfloat> %105, bfloat %102, i64 1, !dbg !44 + %107 = bitcast <2 x bfloat> %106 to i32, !dbg !44 + %108 = insertelement <2 x bfloat> poison, bfloat %103, i64 0, !dbg !44 + %109 = insertelement <2 x bfloat> %108, bfloat %104, i64 1, !dbg !44 + %110 = bitcast <2 x bfloat> %109 to i32, !dbg !44 + tail call void asm sideeffect "@$3 st.global.v2.b32 [ $2 + 0 ], { $0, $1 };", "r,r,l,b"(i32 %107, i32 %110, ptr addrspace(1) %29, i1 %18) #3, !dbg !44 + ret void, !dbg !45 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +attributes #0 = { nounwind "nvvm.reqntid"="512" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = distinct !DISubprogram(name: "l2norm_bwd_kernel", linkageName: "l2norm_bwd_kernel", scope: !1, file: !1, line: 93, type: !5, scopeLine: 93, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!5 = !DISubroutineType(cc: DW_CC_normal, types: !6) +!6 = !{null, !7, !9, !7, !7, !10, !11, !7, !7} +!7 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !8, size: 64, dwarfAddressSpace: 1) +!8 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!9 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !10, size: 64, dwarfAddressSpace: 1) +!10 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!11 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!12 = !DILocation(line: 94, column: 24, scope: !4) +!13 = !DILocation(line: 94, column: 30, scope: !4) +!14 = !DILocation(line: 95, column: 16, scope: !4) +!15 = !DILocation(line: 95, column: 34, scope: !4) +!16 = !DILocation(line: 95, column: 21, scope: !4) +!17 = !DILocation(line: 97, column: 16, scope: !4) +!18 = !DILocation(line: 98, column: 30, scope: !4) +!19 = !DILocation(line: 99, column: 29, scope: !4) +!20 = !DILocation(line: 99, column: 14, scope: !4) +!21 = !DILocation(line: 99, column: 33, scope: !4) +!22 = !DILocation(line: 100, column: 20, scope: !4) +!23 = !DILocation(line: 101, column: 16, scope: !4) +!24 = !DILocation(line: 101, column: 35, scope: !4) +!25 = !DILocation(line: 102, column: 16, scope: !4) +!26 = !DILocation(line: 102, column: 35, scope: !4) +!27 = !DILocation(line: 104, column: 18, scope: !4) +!28 = !DILocation(line: 104, column: 47, scope: !4) +!29 = !DILocation(line: 105, column: 21, scope: !4) +!30 = !DILocation(line: 106, column: 19, scope: !4) +!31 = !DILocation(line: 106, column: 49, scope: !4) +!32 = !DILocation(line: 107, column: 18, scope: !4) +!33 = !DILocation(line: 107, column: 50, scope: !4) +!34 = !DILocation(line: 263, column: 15, scope: !35, inlinedAt: !37) +!35 = distinct !DILexicalBlockFile(scope: !4, file: !36, discriminator: 0) +!36 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!37 = !DILocation(line: 293, column: 36, scope: !35, inlinedAt: !38) +!38 = !DILocation(line: 107, column: 55, scope: !39) +!39 = distinct !DILexicalBlockFile(scope: !4, file: !1, discriminator: 0) +!40 = !DILocation(line: 107, column: 69, scope: !4) +!41 = !DILocation(line: 107, column: 75, scope: !4) +!42 = !DILocation(line: 107, column: 36, scope: !4) +!43 = !DILocation(line: 108, column: 27, scope: !4) +!44 = !DILocation(line: 108, column: 19, scope: !4) +!45 = !DILocation(line: 108, column: 4, scope: !4) diff --git a/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/l2norm_bwd_kernel.ptx b/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/l2norm_bwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..60c7aa90e1115de6bec0dfcc60d7ec964e0a7059 --- /dev/null +++ b/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/l2norm_bwd_kernel.ptx @@ -0,0 +1,389 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_bwd_kernel // -- Begin function l2norm_bwd_kernel + // @l2norm_bwd_kernel +.visible .entry l2norm_bwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_2, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_3, + .param .f32 l2norm_bwd_kernel_param_4, + .param .u32 l2norm_bwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_6, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_7 +) +.reqntid 512 +{ + .reg .pred %p<2>; + .reg .b16 %rs<9>; + .reg .b32 %r<52>; + .reg .b64 %rd<19>; + .loc 1 93 0 // l2norm.py:93:0 +$L__func_begin0: + .loc 1 93 0 // l2norm.py:93:0 + +// %bb.0: + ld.param.b64 %rd5, [l2norm_bwd_kernel_param_0]; + ld.param.b64 %rd6, [l2norm_bwd_kernel_param_1]; +$L__tmp0: + .loc 1 94 24 // l2norm.py:94:24 + mov.u32 %r9, %ctaid.x; + ld.param.b64 %rd7, [l2norm_bwd_kernel_param_2]; + .loc 1 95 16 // l2norm.py:95:16 + mul.wide.u32 %rd8, %r9, 16; + ld.param.b64 %rd9, [l2norm_bwd_kernel_param_3]; + .loc 1 95 34 // l2norm.py:95:34 + mov.u32 %r10, %tid.x; + ld.param.s32 %rd10, [l2norm_bwd_kernel_param_5]; + bfe.u32 %r11, %r10, 5, 4; + .loc 1 95 21 // l2norm.py:95:21 + cvt.u64.u32 %rd11, %r11; + or.b64 %rd12, %rd8, %rd11; + .loc 1 97 16 // l2norm.py:97:16 + setp.lt.s64 %p1, %rd12, %rd10; + .loc 1 98 30 // l2norm.py:98:30 + shl.b32 %r12, %r10, 2; + and.b32 %r13, %r12, 124; + .loc 1 99 14 // l2norm.py:99:14 + shl.b64 %rd13, %rd12, 8; + add.s64 %rd14, %rd5, %rd13; + .loc 1 99 33 // l2norm.py:99:33 + mul.wide.u32 %rd15, %r13, 2; + add.s64 %rd1, %rd14, %rd15; + .loc 1 100 20 // l2norm.py:100:20 + shl.b64 %rd16, %rd12, 2; + add.s64 %rd2, %rd6, %rd16; + .loc 1 101 16 // l2norm.py:101:16 + add.s64 %rd17, %rd7, %rd13; + .loc 1 101 35 // l2norm.py:101:35 + add.s64 %rd3, %rd17, %rd15; + .loc 1 102 16 // l2norm.py:102:16 + add.s64 %rd18, %rd9, %rd13; + .loc 1 102 35 // l2norm.py:102:35 + add.s64 %rd4, %rd18, %rd15; + mov.b32 %r3, 0; + .loc 1 104 18 // l2norm.py:104:18 + // begin inline asm + mov.u32 %r1, %r3; + mov.u32 %r2, %r3; + @%p1 ld.global.v2.b32 { %r1, %r2 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + .loc 1 104 47 // l2norm.py:104:47 + cvt.f32.bf16 %r14, %rs1; + cvt.f32.bf16 %r15, %rs2; + cvt.f32.bf16 %r16, %rs3; + cvt.f32.bf16 %r17, %rs4; + .loc 1 105 21 // l2norm.py:105:21 + // begin inline asm + mov.u32 %r4, %r3; + @%p1 ld.global.b32 { %r4 }, [ %rd2 + 0 ]; + // end inline asm + .loc 1 106 19 // l2norm.py:106:19 + // begin inline asm + mov.u32 %r5, %r3; + mov.u32 %r6, %r3; + @%p1 ld.global.v2.b32 { %r5, %r6 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs5, %rs6}, %r5; + mov.b32 {%rs7, %rs8}, %r6; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r18, %rs5; + cvt.f32.bf16 %r19, %rs6; + cvt.f32.bf16 %r20, %rs7; + cvt.f32.bf16 %r21, %rs8; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r22, %r15, %r19; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r23, %rs1, %rs5, %r22; + fma.rn.f32.bf16 %r24, %rs3, %rs7, %r23; + fma.rn.f32.bf16 %r25, %rs4, %rs8, %r24; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r26, %r25, 16, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r27, %r25, %r26; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r28, %r27, 8, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r29, %r27, %r28; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r30, %r29, 4, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r31, %r29, %r30; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r32, %r31, 2, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r33, %r31, %r32; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r34, %r33, 1, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r35, %r33, %r34; +$L__tmp12: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r36, %r35, %r14; + mul.f32 %r37, %r35, %r15; + mul.f32 %r38, %r35, %r16; + mul.f32 %r39, %r35, %r17; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r40, %r36, %r4; + mul.f32 %r41, %r37, %r4; + mul.f32 %r42, %r38, %r4; + mul.f32 %r43, %r39, %r4; + .loc 1 107 36 // l2norm.py:107:36 + neg.f32 %r44, %r40; + fma.rn.f32 %r45, %r4, %r18, %r44; + neg.f32 %r46, %r41; + fma.rn.f32 %r47, %r4, %r19, %r46; + neg.f32 %r48, %r42; + fma.rn.f32 %r49, %r4, %r20, %r48; + neg.f32 %r50, %r43; + fma.rn.f32 %r51, %r4, %r21, %r50; + .loc 1 108 19 // l2norm.py:108:19 + cvt.rn.bf16x2.f32 %r7, %r47, %r45; + cvt.rn.bf16x2.f32 %r8, %r51, %r49; + // begin inline asm + @%p1 st.global.v2.b32 [ %rd4 + 0 ], { %r7, %r8 }; + // end inline asm + .loc 1 108 4 // l2norm.py:108:4 + ret; +$L__tmp13: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 98 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp12 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 107 // DW_AT_call_line +.b8 55 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp12 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/l2norm_bwd_kernel.source b/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/l2norm_bwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..25926967fd8ce8ce5392a0037c1d84708c2d91fa --- /dev/null +++ b/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/l2norm_bwd_kernel.source @@ -0,0 +1,210 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc47 = loc(unknown) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc54 = loc("y"(#loc)) +#loc55 = loc("rstd"(#loc)) +#loc56 = loc("dy"(#loc)) +#loc57 = loc("dx"(#loc)) +#loc58 = loc("eps"(#loc)) +#loc59 = loc("T"(#loc)) +#loc101 = loc("input"(#loc45)) +#loc102 = loc("a"(#loc50)) +#loc103 = loc("b"(#loc50)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc60) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc61) + %o_t = arith.constant 16 : i32 loc(#loc62) + %o_t_1 = arith.constant 16 : i64 loc(#loc62) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc62) + %o_t_3 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32> loc(#loc63) + %o_t_4 = arith.extsi %o_t_3 : tensor<16xi32> to tensor<16xi64> loc(#loc64) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<16xi64> loc(#loc64) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<16xi64> loc(#loc64) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc65) + %m_t = arith.extsi %T : i32 to i64 loc(#loc66) + %m_t_7 = tt.splat %m_t : i64 -> tensor<16xi64> loc(#loc66) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<16xi64> loc(#loc66) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc67) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc68) + %m_x_10 = arith.constant 128 : i32 loc(#loc69) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc69) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc69) + %m_x_13 = tt.broadcast %m_x : tensor<16x1xi1> -> tensor<16x128xi1> loc(#loc70) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<16x128xi1> loc(#loc70) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<16x128xi1> loc(#loc70) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc71) + %p_y_16 = arith.constant 128 : i32 loc(#loc72) + %p_y_17 = arith.constant 128 : i64 loc(#loc72) + %p_y_18 = arith.constant dense<128> : tensor<16x1xi64> loc(#loc72) + %p_y_19 = arith.muli %p_y, %p_y_18 : tensor<16x1xi64> loc(#loc72) + %p_y_20 = tt.splat %y : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc73) + %p_y_21 = tt.addptr %p_y_20, %p_y_19 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc73) + %p_y_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc74) + %p_y_23 = tt.broadcast %p_y_21 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc75) + %p_y_24 = tt.broadcast %p_y_22 : tensor<1x128xi32> -> tensor<16x128xi32> loc(#loc75) + %p_y_25 = tt.addptr %p_y_23, %p_y_24 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc75) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc76) + %p_rstd_26 = tt.addptr %p_rstd, %o_t_6 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc76) + %p_dy = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc77) + %p_dy_27 = arith.constant 128 : i32 loc(#loc78) + %p_dy_28 = arith.constant 128 : i64 loc(#loc78) + %p_dy_29 = arith.constant dense<128> : tensor<16x1xi64> loc(#loc78) + %p_dy_30 = arith.muli %p_dy, %p_dy_29 : tensor<16x1xi64> loc(#loc78) + %p_dy_31 = tt.splat %dy : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc79) + %p_dy_32 = tt.addptr %p_dy_31, %p_dy_30 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc79) + %p_dy_33 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc80) + %p_dy_34 = tt.broadcast %p_dy_32 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc81) + %p_dy_35 = tt.broadcast %p_dy_33 : tensor<1x128xi32> -> tensor<16x128xi32> loc(#loc81) + %p_dy_36 = tt.addptr %p_dy_34, %p_dy_35 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc81) + %p_dx = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc82) + %p_dx_37 = arith.constant 128 : i32 loc(#loc83) + %p_dx_38 = arith.constant 128 : i64 loc(#loc83) + %p_dx_39 = arith.constant dense<128> : tensor<16x1xi64> loc(#loc83) + %p_dx_40 = arith.muli %p_dx, %p_dx_39 : tensor<16x1xi64> loc(#loc83) + %p_dx_41 = tt.splat %dx : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc84) + %p_dx_42 = tt.addptr %p_dx_41, %p_dx_40 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc84) + %p_dx_43 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc85) + %p_dx_44 = tt.broadcast %p_dx_42 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc86) + %p_dx_45 = tt.broadcast %p_dx_43 : tensor<1x128xi32> -> tensor<16x128xi32> loc(#loc86) + %p_dx_46 = tt.addptr %p_dx_44, %p_dx_45 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc86) + %b_y = arith.constant 0.000000e+00 : f32 loc(#loc87) + %b_y_47 = arith.constant dense<0.000000e+00> : tensor<16x128xf32> loc(#loc87) + %b_y_48 = arith.truncf %b_y_47 : tensor<16x128xf32> to tensor<16x128xbf16> loc(#loc87) + %b_y_49 = tt.load %p_y_25, %m_x_15, %b_y_48 : tensor<16x128x!tt.ptr> loc(#loc87) + %b_y_50 = arith.extf %b_y_49 : tensor<16x128xbf16> to tensor<16x128xf32> loc(#loc88) + %b_rstd = arith.constant 0.000000e+00 : f32 loc(#loc89) + %b_rstd_51 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc89) + %b_rstd_52 = tt.load %p_rstd_26, %m_t_8, %b_rstd_51 : tensor<16x!tt.ptr> loc(#loc89) + %b_dy = arith.constant 0.000000e+00 : f32 loc(#loc90) + %b_dy_53 = arith.constant dense<0.000000e+00> : tensor<16x128xf32> loc(#loc90) + %b_dy_54 = arith.truncf %b_dy_53 : tensor<16x128xf32> to tensor<16x128xbf16> loc(#loc90) + %b_dy_55 = tt.load %p_dy_36, %m_x_15, %b_dy_54 : tensor<16x128x!tt.ptr> loc(#loc90) + %b_dy_56 = arith.extf %b_dy_55 : tensor<16x128xbf16> to tensor<16x128xf32> loc(#loc91) + %b_dx = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc92) + %b_dx_57 = tt.broadcast %b_dx : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc93) + %b_dx_58 = arith.mulf %b_dy_56, %b_dx_57 : tensor<16x128xf32> loc(#loc93) + %b_dx_59 = arith.mulf %b_dy_56, %b_y_50 : tensor<16x128xf32> loc(#loc94) + %b_dx_60 = tt.call @triton.language.standard.sum__fp32S16_128S_c1_cFalse_cNone(%b_dx_59) : (tensor<16x128xf32>) -> tensor<16xf32> loc(#loc95) + %b_dx_61 = tt.expand_dims %b_dx_60 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc96) + %b_dx_62 = tt.broadcast %b_dx_61 : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc97) + %b_dx_63 = arith.mulf %b_dx_62, %b_y_50 : tensor<16x128xf32> loc(#loc97) + %b_dx_64 = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc98) + %b_dx_65 = tt.broadcast %b_dx_64 : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc99) + %b_dx_66 = arith.mulf %b_dx_63, %b_dx_65 : tensor<16x128xf32> loc(#loc99) + %b_dx_67 = arith.subf %b_dx_58, %b_dx_66 : tensor<16x128xf32> loc(#loc100) + %0 = arith.truncf %b_dx_67 : tensor<16x128xf32> to tensor<16x128xbf16> loc(#loc42) + tt.store %p_dx_46, %0, %m_x_15 : tensor<16x128x!tt.ptr> loc(#loc43) + tt.return loc(#loc44) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S16_128S_c1_cFalse_cNone(%input: tensor<16x128xf32> loc("input"(#loc45))) -> tensor<16xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc46) + tt.reduce.return %2 : f32 loc(#loc46) + }) : (tensor<16x128xf32>) -> tensor<16xf32> loc(#loc46) + tt.return %0 : tensor<16xf32> loc(#loc48) + ^bb1: // no predecessors + %1 = ub.poison : tensor<16xf32> loc(#loc49) + tt.return %1 : tensor<16xf32> loc(#loc49) + } loc(#loc45) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc50)), %b: f32 loc("b"(#loc50))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc51) + tt.return %0 : f32 loc(#loc52) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc53) + tt.return %1 : f32 loc(#loc53) + } loc(#loc50) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:31) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:39) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:20) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:31) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:39) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:82) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc60 = loc("i_t"(#loc1)) +#loc61 = loc("i_t"(#loc2)) +#loc62 = loc("o_t"(#loc3)) +#loc63 = loc("o_t"(#loc4)) +#loc64 = loc("o_t"(#loc5)) +#loc65 = loc("o_d"(#loc6)) +#loc66 = loc("m_t"(#loc7)) +#loc67 = loc("m_x"(#loc8)) +#loc68 = loc("m_x"(#loc9)) +#loc69 = loc("m_x"(#loc10)) +#loc70 = loc("m_x"(#loc11)) +#loc71 = loc("p_y"(#loc12)) +#loc72 = loc("p_y"(#loc13)) +#loc73 = loc("p_y"(#loc14)) +#loc74 = loc("p_y"(#loc15)) +#loc75 = loc("p_y"(#loc16)) +#loc76 = loc("p_rstd"(#loc17)) +#loc77 = loc("p_dy"(#loc18)) +#loc78 = loc("p_dy"(#loc19)) +#loc79 = loc("p_dy"(#loc20)) +#loc80 = loc("p_dy"(#loc21)) +#loc81 = loc("p_dy"(#loc22)) +#loc82 = loc("p_dx"(#loc23)) +#loc83 = loc("p_dx"(#loc24)) +#loc84 = loc("p_dx"(#loc25)) +#loc85 = loc("p_dx"(#loc26)) +#loc86 = loc("p_dx"(#loc27)) +#loc87 = loc("b_y"(#loc28)) +#loc88 = loc("b_y"(#loc29)) +#loc89 = loc("b_rstd"(#loc30)) +#loc90 = loc("b_dy"(#loc31)) +#loc91 = loc("b_dy"(#loc32)) +#loc92 = loc("b_dx"(#loc33)) +#loc93 = loc("b_dx"(#loc34)) +#loc94 = loc("b_dx"(#loc35)) +#loc95 = loc("b_dx"(#loc36)) +#loc96 = loc("b_dx"(#loc37)) +#loc97 = loc("b_dx"(#loc38)) +#loc98 = loc("b_dx"(#loc39)) +#loc99 = loc("b_dx"(#loc40)) +#loc100 = loc("b_dx"(#loc41)) diff --git a/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/l2norm_bwd_kernel.ttgir b/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/l2norm_bwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..13ffa3aae314cc1769ddb853593df6d5124b57fd --- /dev/null +++ b/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/l2norm_bwd_kernel.ttgir @@ -0,0 +1,146 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 4], threadsPerWarp = [1, 32], warpsPerCTA = [16, 1], order = [1, 0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc39 = loc("y"(#loc)) +#loc40 = loc("rstd"(#loc)) +#loc41 = loc("dy"(#loc)) +#loc42 = loc("dx"(#loc)) +#loc43 = loc("eps"(#loc)) +#loc44 = loc("T"(#loc)) +#loc72 = loc("b_dx"(#loc30)) +#loc78 = loc(callsite(#loc1 at #loc72)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 16 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %c16_i64 = arith.constant 16 : i64 loc(#loc1) + %cst = arith.constant dense<0.000000e+00> : tensor<16x128xbf16, #blocked> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<16x1xi64, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc45) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc46) + %o_t = arith.muli %i_t_3, %c16_i64 : i64 loc(#loc47) + %o_t_4 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc48) + %o_t_5 = arith.extsi %o_t_4 : tensor<16xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %o_t_6 = tt.splat %o_t : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %o_t_7 = arith.addi %o_t_6, %o_t_5 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %m_t = arith.extsi %T : i32 to i64 loc(#loc50) + %m_t_8 = tt.splat %m_t : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc50) + %m_t_9 = arith.cmpi slt, %o_t_7, %m_t_8 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc50) + %m_x = tt.expand_dims %m_t_9 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi1, #blocked> loc(#loc51) + %m_x_10 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc52) + %m_x_11 = tt.expand_dims %m_x_10 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc52) + %m_x_12 = arith.cmpi slt, %m_x_11, %cst_2 : tensor<1x128xi32, #blocked> loc(#loc53) + %m_x_13 = tt.broadcast %m_x : tensor<16x1xi1, #blocked> -> tensor<16x128xi1, #blocked> loc(#loc54) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1, #blocked> -> tensor<16x128xi1, #blocked> loc(#loc54) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<16x128xi1, #blocked> loc(#loc54) + %p_y = tt.expand_dims %o_t_7 {axis = 1 : i32} : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi64, #blocked> loc(#loc55) + %p_y_16 = arith.muli %p_y, %cst_1 : tensor<16x1xi64, #blocked> loc(#loc56) + %p_y_17 = tt.splat %y : !tt.ptr -> tensor<16x1x!tt.ptr, #blocked> loc(#loc57) + %p_y_18 = tt.addptr %p_y_17, %p_y_16 : tensor<16x1x!tt.ptr, #blocked>, tensor<16x1xi64, #blocked> loc(#loc57) + %p_y_19 = tt.broadcast %p_y_18 : tensor<16x1x!tt.ptr, #blocked> -> tensor<16x128x!tt.ptr, #blocked> loc(#loc58) + %p_y_20 = tt.broadcast %m_x_11 : tensor<1x128xi32, #blocked> -> tensor<16x128xi32, #blocked> loc(#loc58) + %p_y_21 = tt.addptr %p_y_19, %p_y_20 : tensor<16x128x!tt.ptr, #blocked>, tensor<16x128xi32, #blocked> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc59) + %p_rstd_22 = tt.addptr %p_rstd, %o_t_7 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>>, tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc59) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<16x1x!tt.ptr, #blocked> loc(#loc60) + %p_dy_23 = tt.addptr %p_dy, %p_y_16 : tensor<16x1x!tt.ptr, #blocked>, tensor<16x1xi64, #blocked> loc(#loc60) + %p_dy_24 = tt.broadcast %p_dy_23 : tensor<16x1x!tt.ptr, #blocked> -> tensor<16x128x!tt.ptr, #blocked> loc(#loc61) + %p_dy_25 = tt.addptr %p_dy_24, %p_y_20 : tensor<16x128x!tt.ptr, #blocked>, tensor<16x128xi32, #blocked> loc(#loc61) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<16x1x!tt.ptr, #blocked> loc(#loc62) + %p_dx_26 = tt.addptr %p_dx, %p_y_16 : tensor<16x1x!tt.ptr, #blocked>, tensor<16x1xi64, #blocked> loc(#loc62) + %p_dx_27 = tt.broadcast %p_dx_26 : tensor<16x1x!tt.ptr, #blocked> -> tensor<16x128x!tt.ptr, #blocked> loc(#loc63) + %p_dx_28 = tt.addptr %p_dx_27, %p_y_20 : tensor<16x128x!tt.ptr, #blocked>, tensor<16x128xi32, #blocked> loc(#loc63) + %b_y = tt.load %p_y_21, %m_x_15, %cst : tensor<16x128x!tt.ptr, #blocked> loc(#loc64) + %b_y_29 = arith.extf %b_y : tensor<16x128xbf16, #blocked> to tensor<16x128xf32, #blocked> loc(#loc65) + %b_rstd = tt.load %p_rstd_22, %m_t_9, %cst_0 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc66) + %b_dy = tt.load %p_dy_25, %m_x_15, %cst : tensor<16x128x!tt.ptr, #blocked> loc(#loc67) + %b_dy_30 = arith.extf %b_dy : tensor<16x128xbf16, #blocked> to tensor<16x128xf32, #blocked> loc(#loc68) + %b_dx = tt.expand_dims %b_rstd {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xf32, #blocked> loc(#loc69) + %b_dx_31 = tt.broadcast %b_dx : tensor<16x1xf32, #blocked> -> tensor<16x128xf32, #blocked> loc(#loc70) + %b_dx_32 = arith.mulf %b_dy_30, %b_dx_31 : tensor<16x128xf32, #blocked> loc(#loc70) + %b_dx_33 = arith.mulf %b_dy_30, %b_y_29 : tensor<16x128xf32, #blocked> loc(#loc71) + %b_dx_34 = "tt.reduce"(%b_dx_33) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_40: f32 loc(callsite(#loc1 at #loc72)), %b_dx_41: f32 loc(callsite(#loc1 at #loc72))): + %b_dx_42 = arith.addf %b_dx_40, %b_dx_41 : f32 loc(#loc79) + tt.reduce.return %b_dx_42 : f32 loc(#loc77) + }) : (tensor<16x128xf32, #blocked>) -> tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc77) + %b_dx_35 = tt.expand_dims %b_dx_34 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xf32, #blocked> loc(#loc73) + %b_dx_36 = tt.broadcast %b_dx_35 : tensor<16x1xf32, #blocked> -> tensor<16x128xf32, #blocked> loc(#loc74) + %b_dx_37 = arith.mulf %b_dx_36, %b_y_29 : tensor<16x128xf32, #blocked> loc(#loc74) + %b_dx_38 = arith.mulf %b_dx_37, %b_dx_31 : tensor<16x128xf32, #blocked> loc(#loc75) + %b_dx_39 = arith.subf %b_dx_32, %b_dx_38 : tensor<16x128xf32, #blocked> loc(#loc76) + %0 = arith.truncf %b_dx_39 : tensor<16x128xf32, #blocked> to tensor<16x128xbf16, #blocked> loc(#loc36) + tt.store %p_dx_28, %0, %m_x_15 : tensor<16x128x!tt.ptr, #blocked> loc(#loc37) + tt.return loc(#loc38) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc45 = loc("i_t"(#loc2)) +#loc46 = loc("i_t"(#loc3)) +#loc47 = loc("o_t"(#loc4)) +#loc48 = loc("o_t"(#loc5)) +#loc49 = loc("o_t"(#loc6)) +#loc50 = loc("m_t"(#loc7)) +#loc51 = loc("m_x"(#loc8)) +#loc52 = loc("m_x"(#loc9)) +#loc53 = loc("m_x"(#loc10)) +#loc54 = loc("m_x"(#loc11)) +#loc55 = loc("p_y"(#loc12)) +#loc56 = loc("p_y"(#loc13)) +#loc57 = loc("p_y"(#loc14)) +#loc58 = loc("p_y"(#loc15)) +#loc59 = loc("p_rstd"(#loc16)) +#loc60 = loc("p_dy"(#loc17)) +#loc61 = loc("p_dy"(#loc18)) +#loc62 = loc("p_dx"(#loc19)) +#loc63 = loc("p_dx"(#loc20)) +#loc64 = loc("b_y"(#loc21)) +#loc65 = loc("b_y"(#loc22)) +#loc66 = loc("b_rstd"(#loc23)) +#loc67 = loc("b_dy"(#loc24)) +#loc68 = loc("b_dy"(#loc25)) +#loc69 = loc("b_dx"(#loc26)) +#loc70 = loc("b_dx"(#loc27)) +#loc71 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc32)) +#loc74 = loc("b_dx"(#loc33)) +#loc75 = loc("b_dx"(#loc34)) +#loc76 = loc("b_dx"(#loc35)) +#loc77 = loc(callsite(#loc29 at #loc72)) +#loc79 = loc(callsite(#loc31 at #loc77)) diff --git a/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/l2norm_bwd_kernel.ttir b/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/l2norm_bwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..6c29416e2d7c989d3b096ee1e98cbabbdbd07e07 --- /dev/null +++ b/triton/gpu0/CEIAS32JSDBFBLVYMZHZPN5GRX2FAMTQTH5BNWSRLYI27YTITPRA/l2norm_bwd_kernel.ttir @@ -0,0 +1,147 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc40 = loc("y"(#loc)) +#loc41 = loc("rstd"(#loc)) +#loc42 = loc("dy"(#loc)) +#loc43 = loc("dx"(#loc)) +#loc44 = loc("eps"(#loc)) +#loc45 = loc("T"(#loc)) +#loc74 = loc("b_dx"(#loc31)) +#loc80 = loc(callsite(#loc1 at #loc74)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<0.000000e+00> : tensor<16x128xbf16> loc(#loc1) + %b_rstd = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc46) + %cst_0 = arith.constant dense<128> : tensor<16x1xi64> loc(#loc1) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc47) + %o_t = arith.constant 16 : i64 loc(#loc48) + %i_t = tt.get_program_id x : i32 loc(#loc49) + %i_t_1 = arith.extsi %i_t : i32 to i64 loc(#loc50) + %o_t_2 = arith.muli %i_t_1, %o_t : i64 loc(#loc48) + %o_t_3 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32> loc(#loc51) + %o_t_4 = arith.extsi %o_t_3 : tensor<16xi32> to tensor<16xi64> loc(#loc52) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<16xi64> loc(#loc52) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<16xi64> loc(#loc52) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc53) + %m_t = arith.extsi %T : i32 to i64 loc(#loc54) + %m_t_7 = tt.splat %m_t : i64 -> tensor<16xi64> loc(#loc54) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<16xi64> loc(#loc54) + %m_x_9 = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc55) + %m_x_10 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc56) + %m_x_11 = arith.cmpi slt, %m_x_10, %m_x : tensor<1x128xi32> loc(#loc47) + %m_x_12 = tt.broadcast %m_x_9 : tensor<16x1xi1> -> tensor<16x128xi1> loc(#loc57) + %m_x_13 = tt.broadcast %m_x_11 : tensor<1x128xi1> -> tensor<16x128xi1> loc(#loc57) + %m_x_14 = arith.andi %m_x_12, %m_x_13 : tensor<16x128xi1> loc(#loc57) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc58) + %p_y_15 = arith.muli %p_y, %cst_0 : tensor<16x1xi64> loc(#loc59) + %p_y_16 = tt.splat %y : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc60) + %p_y_17 = tt.addptr %p_y_16, %p_y_15 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc60) + %p_y_18 = tt.broadcast %p_y_17 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc61) + %p_y_19 = tt.broadcast %m_x_10 : tensor<1x128xi32> -> tensor<16x128xi32> loc(#loc61) + %p_y_20 = tt.addptr %p_y_18, %p_y_19 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc61) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc62) + %p_rstd_21 = tt.addptr %p_rstd, %o_t_6 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc62) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc63) + %p_dy_22 = tt.addptr %p_dy, %p_y_15 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc63) + %p_dy_23 = tt.broadcast %p_dy_22 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc64) + %p_dy_24 = tt.addptr %p_dy_23, %p_y_19 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc64) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc65) + %p_dx_25 = tt.addptr %p_dx, %p_y_15 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc65) + %p_dx_26 = tt.broadcast %p_dx_25 : tensor<16x1x!tt.ptr> -> tensor<16x128x!tt.ptr> loc(#loc66) + %p_dx_27 = tt.addptr %p_dx_26, %p_y_19 : tensor<16x128x!tt.ptr>, tensor<16x128xi32> loc(#loc66) + %b_y = tt.load %p_y_20, %m_x_14, %cst : tensor<16x128x!tt.ptr> loc(#loc67) + %b_y_28 = arith.extf %b_y : tensor<16x128xbf16> to tensor<16x128xf32> loc(#loc68) + %b_rstd_29 = tt.load %p_rstd_21, %m_t_8, %b_rstd : tensor<16x!tt.ptr> loc(#loc46) + %b_dy = tt.load %p_dy_24, %m_x_14, %cst : tensor<16x128x!tt.ptr> loc(#loc69) + %b_dy_30 = arith.extf %b_dy : tensor<16x128xbf16> to tensor<16x128xf32> loc(#loc70) + %b_dx = tt.expand_dims %b_rstd_29 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc71) + %b_dx_31 = tt.broadcast %b_dx : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc72) + %b_dx_32 = arith.mulf %b_dy_30, %b_dx_31 : tensor<16x128xf32> loc(#loc72) + %b_dx_33 = arith.mulf %b_dy_30, %b_y_28 : tensor<16x128xf32> loc(#loc73) + %b_dx_34 = "tt.reduce"(%b_dx_33) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_40: f32 loc(callsite(#loc1 at #loc74)), %b_dx_41: f32 loc(callsite(#loc1 at #loc74))): + %b_dx_42 = arith.addf %b_dx_40, %b_dx_41 : f32 loc(#loc81) + tt.reduce.return %b_dx_42 : f32 loc(#loc79) + }) : (tensor<16x128xf32>) -> tensor<16xf32> loc(#loc79) + %b_dx_35 = tt.expand_dims %b_dx_34 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc75) + %b_dx_36 = tt.broadcast %b_dx_35 : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc76) + %b_dx_37 = arith.mulf %b_dx_36, %b_y_28 : tensor<16x128xf32> loc(#loc76) + %b_dx_38 = arith.mulf %b_dx_37, %b_dx_31 : tensor<16x128xf32> loc(#loc77) + %b_dx_39 = arith.subf %b_dx_32, %b_dx_38 : tensor<16x128xf32> loc(#loc78) + %0 = arith.truncf %b_dx_39 : tensor<16x128xf32> to tensor<16x128xbf16> loc(#loc37) + tt.store %p_dx_27, %0, %m_x_14 : tensor<16x128x!tt.ptr> loc(#loc38) + tt.return loc(#loc39) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("b_rstd"(#loc2)) +#loc47 = loc("m_x"(#loc3)) +#loc48 = loc("o_t"(#loc4)) +#loc49 = loc("i_t"(#loc5)) +#loc50 = loc("i_t"(#loc6)) +#loc51 = loc("o_t"(#loc7)) +#loc52 = loc("o_t"(#loc8)) +#loc53 = loc("o_d"(#loc9)) +#loc54 = loc("m_t"(#loc10)) +#loc55 = loc("m_x"(#loc11)) +#loc56 = loc("m_x"(#loc12)) +#loc57 = loc("m_x"(#loc13)) +#loc58 = loc("p_y"(#loc14)) +#loc59 = loc("p_y"(#loc15)) +#loc60 = loc("p_y"(#loc16)) +#loc61 = loc("p_y"(#loc17)) +#loc62 = loc("p_rstd"(#loc18)) +#loc63 = loc("p_dy"(#loc19)) +#loc64 = loc("p_dy"(#loc20)) +#loc65 = loc("p_dx"(#loc21)) +#loc66 = loc("p_dx"(#loc22)) +#loc67 = loc("b_y"(#loc23)) +#loc68 = loc("b_y"(#loc24)) +#loc69 = loc("b_dy"(#loc25)) +#loc70 = loc("b_dy"(#loc26)) +#loc71 = loc("b_dx"(#loc27)) +#loc72 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc29)) +#loc75 = loc("b_dx"(#loc33)) +#loc76 = loc("b_dx"(#loc34)) +#loc77 = loc("b_dx"(#loc35)) +#loc78 = loc("b_dx"(#loc36)) +#loc79 = loc(callsite(#loc30 at #loc74)) +#loc81 = loc(callsite(#loc32 at #loc79)) diff --git a/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/__grp__chunk_gated_delta_rule_fwd_kkt_solve_kernel.json b/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/__grp__chunk_gated_delta_rule_fwd_kkt_solve_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..ced94f5b1bda0dc1e822ffd386c63fc61fd3a154 --- /dev/null +++ b/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/__grp__chunk_gated_delta_rule_fwd_kkt_solve_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"chunk_gated_delta_rule_fwd_kkt_solve_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.source", "chunk_gated_delta_rule_fwd_kkt_solve_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.ttir", "chunk_gated_delta_rule_fwd_kkt_solve_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.ttgir", "chunk_gated_delta_rule_fwd_kkt_solve_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.llir", "chunk_gated_delta_rule_fwd_kkt_solve_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.ptx", "chunk_gated_delta_rule_fwd_kkt_solve_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.cubin", "chunk_gated_delta_rule_fwd_kkt_solve_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.cubin b/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..a85e0d3010f0e8e99d2cb0c2ddc1e404954c0ace --- /dev/null +++ b/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.cubin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a9b8fb35cca4bce465247592e389f15d98adb9ad33bdb273eaf8bb6b3945d31f +size 205944 diff --git a/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.json b/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..dd3dccfaaacf115b90dacace8b2761e999223aae --- /dev/null +++ b/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.json @@ -0,0 +1 @@ +{"hash": "114a3b24b9e7d98d62ceb9608b0079684ee72ba626e0f66db13013f58b93d270", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 4, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 8192, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "chunk_gated_delta_rule_fwd_kkt_solve_kernel"} \ No newline at end of file diff --git a/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.llir b/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..de961a1e65875d24459f57d68c1a09a5f9f22678 --- /dev/null +++ b/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.llir @@ -0,0 +1,5028 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @chunk_gated_delta_rule_fwd_kkt_solve_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, ptr addrspace(1) %3, i32 %4, ptr addrspace(1) readnone captures(none) %5, ptr addrspace(1) readnone captures(none) %6) local_unnamed_addr #0 !dbg !5 { + %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %9 = zext nneg i32 %8 to i64, !dbg !14 + %10 = sext i32 %4 to i64, !dbg !15 + %11 = shl nuw nsw i64 %9, 6, !dbg !16 + %.not = icmp slt i64 %11, %10, !dbg !17 + br i1 %.not, label %12, label %common.ret, !dbg !17 + +common.ret: ; preds = %7, %._crit_edge493 + ret void, !dbg !18 + +12: ; preds = %7 + %13 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.y(), !dbg !19 + %14 = zext nneg i32 %13 to i64, !dbg !20 + %15 = and i64 %14, 65504, !dbg !15 + %16 = and i64 %14, 31, !dbg !21 + %17 = or disjoint i64 %11, 16, !dbg !22 + %18 = or disjoint i64 %11, 32, !dbg !23 + %19 = or disjoint i64 %11, 48, !dbg !24 + %20 = mul nuw nsw i64 %15, %10, !dbg !25 + %21 = or disjoint i64 %20, %16, !dbg !26 + %.idx = shl nuw nsw i64 %21, 8, !dbg !27 + %22 = getelementptr i8, ptr addrspace(1) %0, i64 %.idx, !dbg !27 + %23 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !28 + %24 = and i32 %23, 120, !dbg !28 + %25 = lshr exact i32 %24, 3, !dbg !28 + %26 = and i32 %23, 28, !dbg !28 + %27 = and i32 %23, 7, !dbg !28 + %28 = and i32 %23, 3, !dbg !28 + %29 = and i32 %23, 32, !dbg !28 + %30 = lshr exact i32 %29, 2, !dbg !28 + %31 = zext nneg i32 %25 to i64, !dbg !29 + %32 = or disjoint i64 %11, %31, !dbg !29 + %33 = icmp samesign ult i64 %32, %10, !dbg !30 + %34 = or disjoint i64 %17, %31, !dbg !31 + %35 = icmp samesign ult i64 %34, %10, !dbg !32 + %36 = or disjoint i64 %18, %31, !dbg !33 + %37 = icmp samesign ult i64 %36, %10, !dbg !34 + %38 = or disjoint i64 %19, %31, !dbg !35 + %39 = icmp samesign ult i64 %38, %10, !dbg !36 + %40 = shl nuw nsw i32 %28, 1, !dbg !28 + %41 = or disjoint i32 %40, %30, !dbg !28 + %42 = or disjoint i32 %41, 1, !dbg !28 + %43 = or disjoint i32 %40, 1, !dbg !28 + %44 = or disjoint i32 %40, 8, !dbg !28 + %45 = or disjoint i32 %40, 9, !dbg !28 + %46 = insertelement <4 x i32> poison, i32 %40, i64 0, !dbg !29 + %47 = insertelement <4 x i32> %46, i32 %43, i64 1, !dbg !29 + %48 = insertelement <4 x i32> %47, i32 %44, i64 2, !dbg !29 + %49 = insertelement <4 x i32> %48, i32 %45, i64 3, !dbg !29 + %.mask = and <4 x i32> %49, splat (i32 255), !dbg !33 + %50 = zext nneg <4 x i32> %.mask to <4 x i64>, !dbg !33 + %51 = extractelement <4 x i64> %50, i64 0, !dbg !29 + %52 = or disjoint i64 %11, %51, !dbg !29 + %53 = extractelement <4 x i64> %50, i64 1, !dbg !29 + %54 = or disjoint i64 %11, %53, !dbg !29 + %55 = extractelement <4 x i64> %50, i64 2, !dbg !29 + %56 = or disjoint i64 %11, %55, !dbg !29 + %57 = extractelement <4 x i64> %50, i64 3, !dbg !29 + %58 = or disjoint i64 %11, %57, !dbg !29 + %59 = icmp samesign ult i64 %52, %10, !dbg !30 + %60 = icmp samesign ult i64 %54, %10, !dbg !30 + %61 = icmp samesign ult i64 %56, %10, !dbg !30 + %62 = icmp samesign ult i64 %58, %10, !dbg !30 + %63 = insertelement <2 x i32> poison, i32 %41, i64 0, !dbg !29 + %64 = insertelement <2 x i32> %63, i32 %42, i64 1, !dbg !29 + %65 = insertelement <2 x i64> poison, i64 %11, i64 0, !dbg !29 + %66 = shufflevector <2 x i64> %65, <2 x i64> poison, <2 x i32> zeroinitializer, !dbg !29 + %67 = zext nneg <2 x i32> %64 to <2 x i64>, !dbg !29 + %68 = or disjoint <2 x i64> %66, %67, !dbg !29 + %69 = insertelement <2 x i64> poison, i64 %10, i64 0, !dbg !30 + %70 = shufflevector <2 x i64> %69, <2 x i64> poison, <2 x i32> zeroinitializer, !dbg !30 + %71 = icmp ult <2 x i64> %68, %70, !dbg !30 + %72 = shufflevector <2 x i1> %71, <2 x i1> poison, <4 x i32> , !dbg !30 + %73 = lshr exact i32 %26, 2, !dbg !28 + %74 = or disjoint i32 %73, 8, !dbg !28 + %75 = insertelement <2 x i32> poison, i32 %73, i64 0, !dbg !29 + %76 = insertelement <2 x i32> %75, i32 %74, i64 1, !dbg !29 + %77 = zext nneg <2 x i32> %76 to <2 x i64>, !dbg !29 + %78 = or disjoint <2 x i64> %66, %77, !dbg !29 + %79 = shl <2 x i64> %70, splat (i64 32), !dbg !30 + %80 = ashr exact <2 x i64> %79, splat (i64 32), !dbg !30 + %81 = icmp ult <2 x i64> %78, %80, !dbg !30 + %82 = shufflevector <2 x i1> %81, <2 x i1> poison, <4 x i32> , !dbg !30 + %83 = extractelement <2 x i64> %77, i64 1, !dbg !35 + %84 = or disjoint i64 %17, %83, !dbg !31 + %85 = extractelement <2 x i64> %77, i64 0, !dbg !35 + %86 = or disjoint i64 %17, %85, !dbg !31 + %87 = insertelement <4 x i64> poison, i64 %17, i64 0, !dbg !31 + %88 = shufflevector <4 x i64> %87, <4 x i64> poison, <4 x i32> zeroinitializer, !dbg !31 + %89 = or disjoint <4 x i64> %88, %50, !dbg !31 + %90 = icmp samesign ult i64 %84, %10, !dbg !32 + %91 = icmp samesign ult i64 %86, %10, !dbg !32 + %92 = insertelement <4 x i32> poison, i32 %4, i64 0, !dbg !32 + %93 = shufflevector <4 x i32> %92, <4 x i32> poison, <4 x i32> zeroinitializer, !dbg !32 + %94 = sext <4 x i32> %93 to <4 x i64>, !dbg !32 + %95 = icmp ult <4 x i64> %89, %94, !dbg !32 + %96 = or disjoint i64 %18, %83, !dbg !33 + %97 = or disjoint i64 %18, %85, !dbg !33 + %98 = insertelement <4 x i64> poison, i64 %18, i64 0, !dbg !33 + %99 = shufflevector <4 x i64> %98, <4 x i64> poison, <4 x i32> zeroinitializer, !dbg !33 + %100 = or disjoint <4 x i64> %99, %50, !dbg !33 + %101 = icmp samesign ult i64 %96, %10, !dbg !34 + %102 = icmp samesign ult i64 %97, %10, !dbg !34 + %103 = sext <4 x i32> %93 to <4 x i64>, !dbg !34 + %104 = icmp ult <4 x i64> %100, %103, !dbg !34 + %105 = or disjoint i64 %19, %83, !dbg !35 + %106 = or disjoint i64 %19, %85, !dbg !35 + %107 = insertelement <4 x i64> poison, i64 %19, i64 0, !dbg !35 + %108 = shufflevector <4 x i64> %107, <4 x i64> poison, <4 x i32> zeroinitializer, !dbg !35 + %.mask708 = and <4 x i32> %49, splat (i32 255), !dbg !35 + %109 = zext nneg <4 x i32> %.mask708 to <4 x i64>, !dbg !35 + %110 = or disjoint <4 x i64> %108, %109, !dbg !35 + %111 = icmp samesign ult i64 %105, %10, !dbg !36 + %112 = icmp samesign ult i64 %106, %10, !dbg !36 + %113 = sext <4 x i32> %93 to <4 x i64>, !dbg !36 + %114 = icmp ult <4 x i64> %110, %113, !dbg !36 + %115 = getelementptr bfloat, ptr addrspace(1) %2, i64 %21, !dbg !37 + %116 = extractelement <2 x i64> %78, i64 0, !dbg !38 + %117 = shl nuw nsw i64 %116, 5, !dbg !38 + %118 = extractelement <2 x i64> %78, i64 1, !dbg !38 + %119 = shl nuw nsw i64 %118, 5, !dbg !38 + %120 = getelementptr bfloat, ptr addrspace(1) %115, i64 %117, !dbg !39 + %121 = getelementptr bfloat, ptr addrspace(1) %115, i64 %119, !dbg !39 + %122 = shl nuw nsw i64 %86, 5, !dbg !40 + %123 = shl nuw nsw i64 %84, 5, !dbg !40 + %124 = getelementptr bfloat, ptr addrspace(1) %115, i64 %122, !dbg !41 + %125 = getelementptr bfloat, ptr addrspace(1) %115, i64 %123, !dbg !41 + %126 = shl nuw nsw i64 %97, 5, !dbg !42 + %127 = shl nuw nsw i64 %96, 5, !dbg !42 + %128 = getelementptr bfloat, ptr addrspace(1) %115, i64 %126, !dbg !43 + %129 = getelementptr bfloat, ptr addrspace(1) %115, i64 %127, !dbg !43 + %130 = shl nuw nsw i64 %106, 5, !dbg !44 + %131 = shl nuw nsw i64 %105, 5, !dbg !44 + %132 = getelementptr bfloat, ptr addrspace(1) %115, i64 %130, !dbg !45 + %133 = getelementptr bfloat, ptr addrspace(1) %115, i64 %131, !dbg !45 + %134 = extractelement <2 x i1> %81, i64 0, !dbg !46 + %135 = tail call i16 asm sideeffect "mov.u16 $0, $1;\0A\09@$3 ld.global.b16 { $0 }, [ $2 + 0 ];", "=c,c,l,b"(i16 0, ptr addrspace(1) %120, i1 %134) #2, !dbg !47 + %136 = extractelement <2 x i1> %81, i64 1, !dbg !46 + %137 = tail call i16 asm sideeffect "mov.u16 $0, $1;\0A\09@$3 ld.global.b16 { $0 }, [ $2 + 0 ];", "=c,c,l,b"(i16 0, ptr addrspace(1) %121, i1 %136) #2, !dbg !47 + %138 = tail call i16 asm sideeffect "mov.u16 $0, $1;\0A\09@$3 ld.global.b16 { $0 }, [ $2 + 0 ];", "=c,c,l,b"(i16 0, ptr addrspace(1) %124, i1 %91) #2, !dbg !48 + %139 = tail call i16 asm sideeffect "mov.u16 $0, $1;\0A\09@$3 ld.global.b16 { $0 }, [ $2 + 0 ];", "=c,c,l,b"(i16 0, ptr addrspace(1) %125, i1 %90) #2, !dbg !48 + %140 = tail call i16 asm sideeffect "mov.u16 $0, $1;\0A\09@$3 ld.global.b16 { $0 }, [ $2 + 0 ];", "=c,c,l,b"(i16 0, ptr addrspace(1) %128, i1 %102) #2, !dbg !49 + %141 = tail call i16 asm sideeffect "mov.u16 $0, $1;\0A\09@$3 ld.global.b16 { $0 }, [ $2 + 0 ];", "=c,c,l,b"(i16 0, ptr addrspace(1) %129, i1 %101) #2, !dbg !49 + %142 = tail call i16 asm sideeffect "mov.u16 $0, $1;\0A\09@$3 ld.global.b16 { $0 }, [ $2 + 0 ];", "=c,c,l,b"(i16 0, ptr addrspace(1) %132, i1 %112) #2, !dbg !50 + %143 = tail call i16 asm sideeffect "mov.u16 $0, $1;\0A\09@$3 ld.global.b16 { $0 }, [ $2 + 0 ];", "=c,c,l,b"(i16 0, ptr addrspace(1) %133, i1 %111) #2, !dbg !50 + %144 = getelementptr float, ptr addrspace(1) %1, i64 %21, !dbg !51 + %145 = getelementptr float, ptr addrspace(1) %144, i64 %117, !dbg !52 + %146 = getelementptr float, ptr addrspace(1) %144, i64 %119, !dbg !52 + %147 = extractelement <2 x i64> %68, i64 0, !dbg !52 + %.idx8 = shl nuw nsw i64 %147, 7, !dbg !52 + %148 = getelementptr i8, ptr addrspace(1) %144, i64 %.idx8, !dbg !52 + %149 = extractelement <2 x i64> %68, i64 1, !dbg !52 + %.idx9 = shl nuw nsw i64 %149, 7, !dbg !52 + %150 = getelementptr i8, ptr addrspace(1) %144, i64 %.idx9, !dbg !52 + %.idx10 = shl nuw nsw i64 %52, 7, !dbg !52 + %151 = getelementptr i8, ptr addrspace(1) %144, i64 %.idx10, !dbg !52 + %.idx11 = shl nuw nsw i64 %54, 7, !dbg !52 + %152 = getelementptr i8, ptr addrspace(1) %144, i64 %.idx11, !dbg !52 + %.idx12 = shl nuw nsw i64 %56, 7, !dbg !52 + %153 = getelementptr i8, ptr addrspace(1) %144, i64 %.idx12, !dbg !52 + %.idx13 = shl nuw nsw i64 %58, 7, !dbg !52 + %154 = getelementptr i8, ptr addrspace(1) %144, i64 %.idx13, !dbg !52 + %155 = getelementptr float, ptr addrspace(1) %144, i64 %122, !dbg !53 + %156 = getelementptr float, ptr addrspace(1) %144, i64 %123, !dbg !53 + %157 = extractelement <4 x i64> %89, i64 0, !dbg !53 + %.idx14 = shl nuw nsw i64 %157, 7, !dbg !53 + %158 = getelementptr i8, ptr addrspace(1) %144, i64 %.idx14, !dbg !53 + %159 = extractelement <4 x i64> %89, i64 1, !dbg !53 + %.idx15 = shl nuw nsw i64 %159, 7, !dbg !53 + %160 = getelementptr i8, ptr addrspace(1) %144, i64 %.idx15, !dbg !53 + %161 = extractelement <4 x i64> %89, i64 2, !dbg !53 + %.idx16 = shl nuw nsw i64 %161, 7, !dbg !53 + %162 = getelementptr i8, ptr addrspace(1) %144, i64 %.idx16, !dbg !53 + %163 = extractelement <4 x i64> %89, i64 3, !dbg !53 + %.idx17 = shl nuw nsw i64 %163, 7, !dbg !53 + %164 = getelementptr i8, ptr addrspace(1) %144, i64 %.idx17, !dbg !53 + %165 = getelementptr float, ptr addrspace(1) %144, i64 %126, !dbg !54 + %166 = getelementptr float, ptr addrspace(1) %144, i64 %127, !dbg !54 + %167 = extractelement <4 x i64> %100, i64 0, !dbg !54 + %.idx18 = shl nuw nsw i64 %167, 7, !dbg !54 + %168 = getelementptr i8, ptr addrspace(1) %144, i64 %.idx18, !dbg !54 + %169 = extractelement <4 x i64> %100, i64 1, !dbg !54 + %.idx19 = shl nuw nsw i64 %169, 7, !dbg !54 + %170 = getelementptr i8, ptr addrspace(1) %144, i64 %.idx19, !dbg !54 + %171 = extractelement <4 x i64> %100, i64 2, !dbg !54 + %.idx20 = shl nuw nsw i64 %171, 7, !dbg !54 + %172 = getelementptr i8, ptr addrspace(1) %144, i64 %.idx20, !dbg !54 + %173 = extractelement <4 x i64> %100, i64 3, !dbg !54 + %.idx21 = shl nuw nsw i64 %173, 7, !dbg !54 + %174 = getelementptr i8, ptr addrspace(1) %144, i64 %.idx21, !dbg !54 + %175 = getelementptr float, ptr addrspace(1) %144, i64 %130, !dbg !55 + %176 = getelementptr float, ptr addrspace(1) %144, i64 %131, !dbg !55 + %177 = extractelement <4 x i64> %110, i64 0, !dbg !55 + %.idx22 = shl nuw nsw i64 %177, 7, !dbg !55 + %178 = getelementptr i8, ptr addrspace(1) %144, i64 %.idx22, !dbg !55 + %179 = extractelement <4 x i64> %110, i64 1, !dbg !55 + %.idx23 = shl nuw nsw i64 %179, 7, !dbg !55 + %180 = getelementptr i8, ptr addrspace(1) %144, i64 %.idx23, !dbg !55 + %181 = extractelement <4 x i64> %110, i64 2, !dbg !55 + %.idx24 = shl nuw nsw i64 %181, 7, !dbg !55 + %182 = getelementptr i8, ptr addrspace(1) %144, i64 %.idx24, !dbg !55 + %183 = extractelement <4 x i64> %110, i64 3, !dbg !55 + %.idx25 = shl nuw nsw i64 %183, 7, !dbg !55 + %184 = getelementptr i8, ptr addrspace(1) %144, i64 %.idx25, !dbg !55 + %185 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %145, i1 %134) #2, !dbg !46 + %186 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %146, i1 %136) #2, !dbg !46 + %187 = extractelement <2 x i1> %71, i64 0, !dbg !46 + %188 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %148, i1 %187) #2, !dbg !46 + %189 = extractelement <2 x i1> %71, i64 1, !dbg !46 + %190 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %150, i1 %189) #2, !dbg !46 + %191 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %151, i1 %59) #2, !dbg !46 + %192 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %152, i1 %60) #2, !dbg !46 + %193 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %153, i1 %61) #2, !dbg !46 + %194 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %154, i1 %62) #2, !dbg !46 + %195 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %155, i1 %91) #2, !dbg !56 + %196 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %156, i1 %90) #2, !dbg !56 + %197 = extractelement <4 x i1> %95, i64 0, !dbg !56 + %198 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %158, i1 %197) #2, !dbg !56 + %199 = extractelement <4 x i1> %95, i64 1, !dbg !56 + %200 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %160, i1 %199) #2, !dbg !56 + %201 = extractelement <4 x i1> %95, i64 2, !dbg !56 + %202 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %162, i1 %201) #2, !dbg !56 + %203 = extractelement <4 x i1> %95, i64 3, !dbg !56 + %204 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %164, i1 %203) #2, !dbg !56 + %205 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %165, i1 %102) #2, !dbg !57 + %206 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %166, i1 %101) #2, !dbg !57 + %207 = extractelement <4 x i1> %104, i64 0, !dbg !57 + %208 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %168, i1 %207) #2, !dbg !57 + %209 = extractelement <4 x i1> %104, i64 1, !dbg !57 + %210 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %170, i1 %209) #2, !dbg !57 + %211 = extractelement <4 x i1> %104, i64 2, !dbg !57 + %212 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %172, i1 %211) #2, !dbg !57 + %213 = extractelement <4 x i1> %104, i64 3, !dbg !57 + %214 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %174, i1 %213) #2, !dbg !57 + %215 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %175, i1 %112) #2, !dbg !58 + %216 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %176, i1 %111) #2, !dbg !58 + %217 = extractelement <4 x i1> %114, i64 0, !dbg !58 + %218 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %178, i1 %217) #2, !dbg !58 + %219 = extractelement <4 x i1> %114, i64 1, !dbg !58 + %220 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %180, i1 %219) #2, !dbg !58 + %221 = extractelement <4 x i1> %114, i64 2, !dbg !58 + %222 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %182, i1 %221) #2, !dbg !58 + %223 = extractelement <4 x i1> %114, i64 3, !dbg !58 + %224 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %184, i1 %223) #2, !dbg !58 + %225 = shl nuw nsw i32 %27, 3, !dbg !59 + %.idx26 = shl nuw nsw i64 %32, 13, !dbg !60 + %226 = getelementptr i8, ptr addrspace(1) %22, i64 %.idx26, !dbg !60 + %227 = icmp samesign ult i64 %17, %10, !dbg !61 + %228 = zext nneg i32 %225 to i64, !dbg !62 + %229 = getelementptr bfloat, ptr addrspace(1) %226, i64 %228, !dbg !62 + %230 = shl nuw nsw i32 %23, 4, !dbg !63 + %231 = and i32 %230, 2032, !dbg !63 + %232 = and i32 %23, 56, !dbg !63 + %233 = shl nuw nsw i32 %232, 1, !dbg !63 + %234 = xor i32 %231, %233, !dbg !63 + %235 = getelementptr i8, ptr addrspace(3) @global_smem, i32 %234, !dbg !63 + %236 = select i1 %33, i32 16, i32 0, !dbg !63 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %235, ptr addrspace(1) %229, i32 %236) #2, !dbg !63 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !63 + %237 = zext nneg i32 %225 to i64, !dbg !62 + %238 = getelementptr bfloat, ptr addrspace(1) %226, i64 %237, !dbg !62 + %239 = getelementptr i8, ptr addrspace(1) %238, i64 128, !dbg !62 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !63 + %240 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 2048), i32 %234, !dbg !63 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %240, ptr addrspace(1) %239, i32 %236) #2, !dbg !63 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !63 + %241 = shl nuw nsw i32 %27, 7 + %242 = shl nuw nsw i32 %27, 4 + %243 = shl nuw nsw i32 %23, 1 + %244 = and i32 %243, 48 + %245 = or disjoint i32 %241, %242 + %246 = xor i32 %245, %244 + %247 = xor i32 %246, 64 + %248 = shl nuw nsw i32 %29, 5 + %249 = or disjoint i32 %241, %248 + %250 = xor i32 %242, %244 + %251 = or disjoint i32 %249, %250 + %252 = xor i32 %251, 64 + %253 = shl nuw nsw i32 %23, 7 + %254 = and i32 %253, 1920 + %255 = and i32 %23, 16 + %256 = xor i32 %242, %255 + %257 = or disjoint i32 %256, %254 + %258 = xor i32 %257, 32 + %259 = xor i32 %257, 64 + %260 = xor i32 %257, 96 + %.idx178 = shl nuw nsw i64 %34, 13 + %261 = getelementptr i8, ptr addrspace(1) %22, i64 %.idx178 + %262 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 4096), i32 %234 + %263 = shl nuw nsw i32 %232, 5 + %264 = shl nuw nsw i32 %24, 1 + %265 = or disjoint i32 %263, %242 + %266 = xor i32 %265, %264 + %267 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 6144), i32 %266 + %268 = shl nuw nsw i32 %27, 8 + %269 = or disjoint i32 %250, %268 + %270 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 6144), i32 %269 + %271 = getelementptr inbounds nuw i8, ptr addrspace(3) %270, i32 128 + %272 = xor i32 %269, 64 + %273 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 6144), i32 %272 + %274 = getelementptr inbounds nuw i8, ptr addrspace(3) %273, i32 128 + %275 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 4096), i32 %257 + %276 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 4096), i32 %258 + %277 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 4096), i32 %259 + %278 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 4096), i32 %260 + %279 = icmp samesign ult i64 %18, %10 + %.idx186 = shl nuw nsw i64 %36, 13 + %280 = getelementptr i8, ptr addrspace(1) %22, i64 %.idx186 + %281 = icmp samesign ult i64 %19, %10 + %.idx194 = shl nuw nsw i64 %38, 13 + %282 = getelementptr i8, ptr addrspace(1) %22, i64 %.idx194 + %283 = shl nuw nsw i32 %23, 3, !dbg !64 + %284 = and i32 %283, 56, !dbg !64 + %285 = zext nneg i32 %284 to i64, !dbg !64 + %286 = zext nneg i32 %225 to i64 + %invariant.gep = getelementptr bfloat, ptr addrspace(1) %226, i64 %286, !dbg !64 + br label %287, !dbg !64 + +287: ; preds = %12, %971 + %indvars.iv544 = phi i64 [ 0, %12 ], [ 128, %971 ] + %indvars.iv542 = phi i64 [ %285, %12 ], [ %indvars.iv.next543, %971 ] + %288 = phi i1 [ true, %12 ], [ false, %971 ] + %289 = phi i32 [ -1, %12 ], [ %360, %971 ] + %290 = phi i32 [ 1, %12 ], [ %428, %971 ] + %291 = phi float [ 0.000000e+00, %12 ], [ %1024, %971 ] + %292 = phi float [ 0.000000e+00, %12 ], [ %1025, %971 ] + %293 = phi float [ 0.000000e+00, %12 ], [ %1026, %971 ] + %294 = phi float [ 0.000000e+00, %12 ], [ %1027, %971 ] + %295 = phi float [ 0.000000e+00, %12 ], [ %1028, %971 ] + %296 = phi float [ 0.000000e+00, %12 ], [ %1029, %971 ] + %297 = phi float [ 0.000000e+00, %12 ], [ %1030, %971 ] + %298 = phi float [ 0.000000e+00, %12 ], [ %1031, %971 ] + %299 = phi float [ 0.000000e+00, %12 ], [ %1016, %971 ] + %300 = phi float [ 0.000000e+00, %12 ], [ %1017, %971 ] + %301 = phi float [ 0.000000e+00, %12 ], [ %1018, %971 ] + %302 = phi float [ 0.000000e+00, %12 ], [ %1019, %971 ] + %303 = phi float [ 0.000000e+00, %12 ], [ %1020, %971 ] + %304 = phi float [ 0.000000e+00, %12 ], [ %1021, %971 ] + %305 = phi float [ 0.000000e+00, %12 ], [ %1022, %971 ] + %306 = phi float [ 0.000000e+00, %12 ], [ %1023, %971 ] + %307 = phi float [ 0.000000e+00, %12 ], [ %1008, %971 ] + %308 = phi float [ 0.000000e+00, %12 ], [ %1009, %971 ] + %309 = phi float [ 0.000000e+00, %12 ], [ %1010, %971 ] + %310 = phi float [ 0.000000e+00, %12 ], [ %1011, %971 ] + %311 = phi float [ 0.000000e+00, %12 ], [ %1012, %971 ] + %312 = phi float [ 0.000000e+00, %12 ], [ %1013, %971 ] + %313 = phi float [ 0.000000e+00, %12 ], [ %1014, %971 ] + %314 = phi float [ 0.000000e+00, %12 ], [ %1015, %971 ] + %315 = phi float [ 0.000000e+00, %12 ], [ %1000, %971 ] + %316 = phi float [ 0.000000e+00, %12 ], [ %1001, %971 ] + %317 = phi float [ 0.000000e+00, %12 ], [ %1002, %971 ] + %318 = phi float [ 0.000000e+00, %12 ], [ %1003, %971 ] + %319 = phi float [ 0.000000e+00, %12 ], [ %1004, %971 ] + %320 = phi float [ 0.000000e+00, %12 ], [ %1005, %971 ] + %321 = phi float [ 0.000000e+00, %12 ], [ %1006, %971 ] + %322 = phi float [ 0.000000e+00, %12 ], [ %1007, %971 ] + %323 = phi float [ 0.000000e+00, %12 ], [ %992, %971 ] + %324 = phi float [ 0.000000e+00, %12 ], [ %993, %971 ] + %325 = phi float [ 0.000000e+00, %12 ], [ %994, %971 ] + %326 = phi float [ 0.000000e+00, %12 ], [ %995, %971 ] + %327 = phi float [ 0.000000e+00, %12 ], [ %996, %971 ] + %328 = phi float [ 0.000000e+00, %12 ], [ %997, %971 ] + %329 = phi float [ 0.000000e+00, %12 ], [ %998, %971 ] + %330 = phi float [ 0.000000e+00, %12 ], [ %999, %971 ] + %331 = phi float [ 0.000000e+00, %12 ], [ %984, %971 ] + %332 = phi float [ 0.000000e+00, %12 ], [ %985, %971 ] + %333 = phi float [ 0.000000e+00, %12 ], [ %986, %971 ] + %334 = phi float [ 0.000000e+00, %12 ], [ %987, %971 ] + %335 = phi float [ 0.000000e+00, %12 ], [ %988, %971 ] + %336 = phi float [ 0.000000e+00, %12 ], [ %989, %971 ] + %337 = phi float [ 0.000000e+00, %12 ], [ %990, %971 ] + %338 = phi float [ 0.000000e+00, %12 ], [ %991, %971 ] + %339 = phi float [ 0.000000e+00, %12 ], [ %980, %971 ] + %340 = phi float [ 0.000000e+00, %12 ], [ %981, %971 ] + %341 = phi float [ 0.000000e+00, %12 ], [ %982, %971 ] + %342 = phi float [ 0.000000e+00, %12 ], [ %983, %971 ] + %343 = phi float [ 0.000000e+00, %12 ], [ %976, %971 ] + %344 = phi float [ 0.000000e+00, %12 ], [ %977, %971 ] + %345 = phi float [ 0.000000e+00, %12 ], [ %978, %971 ] + %346 = phi float [ 0.000000e+00, %12 ], [ %979, %971 ] + %347 = phi float [ 0.000000e+00, %12 ], [ %972, %971 ] + %348 = phi float [ 0.000000e+00, %12 ], [ %973, %971 ] + %349 = phi float [ 0.000000e+00, %12 ], [ %974, %971 ] + %350 = phi float [ 0.000000e+00, %12 ], [ %975, %971 ] + %351 = phi float [ 0.000000e+00, %12 ], [ %422, %971 ] + %352 = phi float [ 0.000000e+00, %12 ], [ %423, %971 ] + %353 = phi float [ 0.000000e+00, %12 ], [ %424, %971 ] + %354 = phi float [ 0.000000e+00, %12 ], [ %425, %971 ] + %355 = phi <4 x float> [ zeroinitializer, %12 ], [ %1032, %971 ] + %356 = phi <4 x float> [ zeroinitializer, %12 ], [ %1033, %971 ] + %357 = phi <4 x float> [ zeroinitializer, %12 ], [ %1034, %971 ] + %358 = add i32 %289, 1, !dbg !64 + %359 = icmp sgt i32 %358, 1, !dbg !64 + %360 = select i1 %359, i32 0, i32 %358, !dbg !64 + tail call void @llvm.nvvm.cp.async.wait.group(i32 1), !dbg !63 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !63 + %.idx176 = shl i32 %360, 11, !dbg !63 + %361 = getelementptr i8, ptr addrspace(3) @global_smem, i32 %.idx176, !dbg !63 + %362 = getelementptr inbounds nuw i8, ptr addrspace(3) %361, i32 %246, !dbg !65 + %363 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %362), !dbg !65 + %364 = getelementptr inbounds nuw i8, ptr addrspace(3) %362, i32 1024, !dbg !65 + %365 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %364), !dbg !65 + %366 = getelementptr inbounds nuw i8, ptr addrspace(3) %361, i32 %247, !dbg !65 + %367 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %366), !dbg !65 + %368 = getelementptr inbounds nuw i8, ptr addrspace(3) %366, i32 1024, !dbg !65 + %369 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %368), !dbg !65 + %370 = getelementptr inbounds nuw i8, ptr addrspace(3) %361, i32 %251, !dbg !65 + %371 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %370), !dbg !65 + %372 = extractvalue { i32, i32, i32, i32 } %371, 0, !dbg !65 + %373 = extractvalue { i32, i32, i32, i32 } %371, 1, !dbg !65 + %374 = extractvalue { i32, i32, i32, i32 } %371, 2, !dbg !65 + %375 = extractvalue { i32, i32, i32, i32 } %371, 3, !dbg !65 + %376 = getelementptr inbounds nuw i8, ptr addrspace(3) %361, i32 %252, !dbg !65 + %377 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %376), !dbg !65 + %378 = extractvalue { i32, i32, i32, i32 } %377, 0, !dbg !65 + %379 = extractvalue { i32, i32, i32, i32 } %377, 1, !dbg !65 + %380 = extractvalue { i32, i32, i32, i32 } %377, 2, !dbg !65 + %381 = extractvalue { i32, i32, i32, i32 } %377, 3, !dbg !65 + %382 = getelementptr inbounds nuw i8, ptr addrspace(3) %361, i32 %257, !dbg !63 + %383 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %382), !dbg !63 + %384 = extractvalue { i32, i32, i32, i32 } %383, 0, !dbg !63 + %385 = extractvalue { i32, i32, i32, i32 } %383, 1, !dbg !63 + %386 = extractvalue { i32, i32, i32, i32 } %383, 2, !dbg !63 + %387 = extractvalue { i32, i32, i32, i32 } %383, 3, !dbg !63 + %388 = getelementptr inbounds nuw i8, ptr addrspace(3) %361, i32 %258, !dbg !63 + %389 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %388), !dbg !63 + %390 = extractvalue { i32, i32, i32, i32 } %389, 0, !dbg !63 + %391 = extractvalue { i32, i32, i32, i32 } %389, 1, !dbg !63 + %392 = extractvalue { i32, i32, i32, i32 } %389, 2, !dbg !63 + %393 = extractvalue { i32, i32, i32, i32 } %389, 3, !dbg !63 + %394 = getelementptr inbounds nuw i8, ptr addrspace(3) %361, i32 %259, !dbg !63 + %395 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %394), !dbg !63 + %396 = extractvalue { i32, i32, i32, i32 } %395, 0, !dbg !63 + %397 = extractvalue { i32, i32, i32, i32 } %395, 1, !dbg !63 + %398 = extractvalue { i32, i32, i32, i32 } %395, 2, !dbg !63 + %399 = extractvalue { i32, i32, i32, i32 } %395, 3, !dbg !63 + %400 = getelementptr inbounds nuw i8, ptr addrspace(3) %361, i32 %260, !dbg !63 + %401 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %400), !dbg !63 + %402 = extractvalue { i32, i32, i32, i32 } %401, 0, !dbg !63 + %403 = extractvalue { i32, i32, i32, i32 } %401, 1, !dbg !63 + %404 = extractvalue { i32, i32, i32, i32 } %401, 2, !dbg !63 + %405 = extractvalue { i32, i32, i32, i32 } %401, 3, !dbg !63 + %406 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %351, float %352, float %353, float %354, i32 %384, i32 %385, i32 %386, i32 %387, i32 %372, i32 %373) #2, !dbg !65 + %407 = extractvalue { float, float, float, float } %406, 0, !dbg !65 + %408 = extractvalue { float, float, float, float } %406, 1, !dbg !65 + %409 = extractvalue { float, float, float, float } %406, 2, !dbg !65 + %410 = extractvalue { float, float, float, float } %406, 3, !dbg !65 + %411 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %407, float %408, float %409, float %410, i32 %390, i32 %391, i32 %392, i32 %393, i32 %374, i32 %375) #2, !dbg !65 + %412 = extractvalue { float, float, float, float } %411, 0, !dbg !65 + %413 = extractvalue { float, float, float, float } %411, 1, !dbg !65 + %414 = extractvalue { float, float, float, float } %411, 2, !dbg !65 + %415 = extractvalue { float, float, float, float } %411, 3, !dbg !65 + %416 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %412, float %413, float %414, float %415, i32 %396, i32 %397, i32 %398, i32 %399, i32 %378, i32 %379) #2, !dbg !65 + %417 = extractvalue { float, float, float, float } %416, 0, !dbg !65 + %418 = extractvalue { float, float, float, float } %416, 1, !dbg !65 + %419 = extractvalue { float, float, float, float } %416, 2, !dbg !65 + %420 = extractvalue { float, float, float, float } %416, 3, !dbg !65 + %421 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %417, float %418, float %419, float %420, i32 %402, i32 %403, i32 %404, i32 %405, i32 %380, i32 %381) #2, !dbg !65 + %422 = extractvalue { float, float, float, float } %421, 0, !dbg !65 + %423 = extractvalue { float, float, float, float } %421, 1, !dbg !65 + %424 = extractvalue { float, float, float, float } %421, 2, !dbg !65 + %425 = extractvalue { float, float, float, float } %421, 3, !dbg !65 + %426 = add i32 %290, 1, !dbg !64 + %427 = icmp sgt i32 %426, 1, !dbg !64 + %428 = select i1 %427, i32 0, i32 %426, !dbg !64 + %gep603 = getelementptr i8, ptr addrspace(1) %invariant.gep, i64 %indvars.iv544, !dbg !62 + %429 = getelementptr i8, ptr addrspace(1) %gep603, i64 256, !dbg !62 + %.idx177 = shl i32 %428, 11, !dbg !63 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !63 + %gep = getelementptr i8, ptr addrspace(3) %235, i32 %.idx177, !dbg !63 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %gep, ptr addrspace(1) %429, i32 0) #2, !dbg !63 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !63 + br i1 %227, label %430, label %971, !dbg !66 + +430: ; preds = %287 + %431 = extractvalue { i32, i32, i32, i32 } %369, 3, !dbg !65 + %432 = extractvalue { i32, i32, i32, i32 } %369, 2, !dbg !65 + %433 = extractvalue { i32, i32, i32, i32 } %369, 1, !dbg !65 + %434 = extractvalue { i32, i32, i32, i32 } %369, 0, !dbg !65 + %435 = extractvalue { i32, i32, i32, i32 } %367, 3, !dbg !65 + %436 = extractvalue { i32, i32, i32, i32 } %367, 2, !dbg !65 + %437 = extractvalue { i32, i32, i32, i32 } %367, 1, !dbg !65 + %438 = extractvalue { i32, i32, i32, i32 } %367, 0, !dbg !65 + %439 = extractvalue { i32, i32, i32, i32 } %365, 3, !dbg !65 + %440 = extractvalue { i32, i32, i32, i32 } %365, 2, !dbg !65 + %441 = extractvalue { i32, i32, i32, i32 } %365, 1, !dbg !65 + %442 = extractvalue { i32, i32, i32, i32 } %365, 0, !dbg !65 + %443 = extractvalue { i32, i32, i32, i32 } %363, 3, !dbg !65 + %444 = extractvalue { i32, i32, i32, i32 } %363, 2, !dbg !65 + %445 = extractvalue { i32, i32, i32, i32 } %363, 1, !dbg !65 + %446 = extractvalue { i32, i32, i32, i32 } %363, 0, !dbg !65 + %447 = getelementptr bfloat, ptr addrspace(1) %261, i64 %indvars.iv542, !dbg !67 + %448 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %447, i1 %35) #2, !dbg !68 + %449 = extractvalue { i32, i32, i32, i32 } %448, 0, !dbg !68 + %450 = extractvalue { i32, i32, i32, i32 } %448, 1, !dbg !68 + %451 = extractvalue { i32, i32, i32, i32 } %448, 2, !dbg !68 + %452 = extractvalue { i32, i32, i32, i32 } %448, 3, !dbg !68 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %262, i32 %449, i32 %450, i32 %451, i32 %452) #2, !dbg !68 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %267, i32 %449, i32 %450, i32 %451, i32 %452) #2, !dbg !69 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !69 + %453 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %270), !dbg !69 + %454 = extractvalue { i32, i32, i32, i32 } %453, 0, !dbg !69 + %455 = extractvalue { i32, i32, i32, i32 } %453, 1, !dbg !69 + %456 = extractvalue { i32, i32, i32, i32 } %453, 2, !dbg !69 + %457 = extractvalue { i32, i32, i32, i32 } %453, 3, !dbg !69 + %458 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %271), !dbg !69 + %459 = extractvalue { i32, i32, i32, i32 } %458, 0, !dbg !69 + %460 = extractvalue { i32, i32, i32, i32 } %458, 1, !dbg !69 + %461 = extractvalue { i32, i32, i32, i32 } %458, 2, !dbg !69 + %462 = extractvalue { i32, i32, i32, i32 } %458, 3, !dbg !69 + %463 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %273), !dbg !69 + %464 = extractvalue { i32, i32, i32, i32 } %463, 0, !dbg !69 + %465 = extractvalue { i32, i32, i32, i32 } %463, 1, !dbg !69 + %466 = extractvalue { i32, i32, i32, i32 } %463, 2, !dbg !69 + %467 = extractvalue { i32, i32, i32, i32 } %463, 3, !dbg !69 + %468 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %274), !dbg !69 + %469 = extractvalue { i32, i32, i32, i32 } %468, 0, !dbg !69 + %470 = extractvalue { i32, i32, i32, i32 } %468, 1, !dbg !69 + %471 = extractvalue { i32, i32, i32, i32 } %468, 2, !dbg !69 + %472 = extractvalue { i32, i32, i32, i32 } %468, 3, !dbg !69 + %473 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %275), !dbg !68 + %474 = extractvalue { i32, i32, i32, i32 } %473, 0, !dbg !68 + %475 = extractvalue { i32, i32, i32, i32 } %473, 1, !dbg !68 + %476 = extractvalue { i32, i32, i32, i32 } %473, 2, !dbg !68 + %477 = extractvalue { i32, i32, i32, i32 } %473, 3, !dbg !68 + %478 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %276), !dbg !68 + %479 = extractvalue { i32, i32, i32, i32 } %478, 0, !dbg !68 + %480 = extractvalue { i32, i32, i32, i32 } %478, 1, !dbg !68 + %481 = extractvalue { i32, i32, i32, i32 } %478, 2, !dbg !68 + %482 = extractvalue { i32, i32, i32, i32 } %478, 3, !dbg !68 + %483 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %277), !dbg !68 + %484 = extractvalue { i32, i32, i32, i32 } %483, 0, !dbg !68 + %485 = extractvalue { i32, i32, i32, i32 } %483, 1, !dbg !68 + %486 = extractvalue { i32, i32, i32, i32 } %483, 2, !dbg !68 + %487 = extractvalue { i32, i32, i32, i32 } %483, 3, !dbg !68 + %488 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %278), !dbg !68 + %489 = extractvalue { i32, i32, i32, i32 } %488, 0, !dbg !68 + %490 = extractvalue { i32, i32, i32, i32 } %488, 1, !dbg !68 + %491 = extractvalue { i32, i32, i32, i32 } %488, 2, !dbg !68 + %492 = extractvalue { i32, i32, i32, i32 } %488, 3, !dbg !68 + %493 = extractelement <4 x float> %357, i64 0, !dbg !70 + %494 = extractelement <4 x float> %357, i64 1, !dbg !70 + %495 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %493, float %494, float %347, float %348, i32 %474, i32 %475, i32 %476, i32 %477, i32 %454, i32 %455) #2, !dbg !70 + %496 = extractvalue { float, float, float, float } %495, 0, !dbg !70 + %497 = extractvalue { float, float, float, float } %495, 1, !dbg !70 + %498 = extractvalue { float, float, float, float } %495, 2, !dbg !70 + %499 = extractvalue { float, float, float, float } %495, 3, !dbg !70 + %500 = extractelement <4 x float> %357, i64 2, !dbg !70 + %501 = extractelement <4 x float> %357, i64 3, !dbg !70 + %502 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %349, float %350, float %500, float %501, i32 %474, i32 %475, i32 %476, i32 %477, i32 %459, i32 %460) #2, !dbg !70 + %503 = extractvalue { float, float, float, float } %502, 0, !dbg !70 + %504 = extractvalue { float, float, float, float } %502, 1, !dbg !70 + %505 = extractvalue { float, float, float, float } %502, 2, !dbg !70 + %506 = extractvalue { float, float, float, float } %502, 3, !dbg !70 + %507 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %496, float %497, float %498, float %499, i32 %479, i32 %480, i32 %481, i32 %482, i32 %456, i32 %457) #2, !dbg !70 + %508 = extractvalue { float, float, float, float } %507, 0, !dbg !70 + %509 = extractvalue { float, float, float, float } %507, 1, !dbg !70 + %510 = extractvalue { float, float, float, float } %507, 2, !dbg !70 + %511 = extractvalue { float, float, float, float } %507, 3, !dbg !70 + %512 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %503, float %504, float %505, float %506, i32 %479, i32 %480, i32 %481, i32 %482, i32 %461, i32 %462) #2, !dbg !70 + %513 = extractvalue { float, float, float, float } %512, 0, !dbg !70 + %514 = extractvalue { float, float, float, float } %512, 1, !dbg !70 + %515 = extractvalue { float, float, float, float } %512, 2, !dbg !70 + %516 = extractvalue { float, float, float, float } %512, 3, !dbg !70 + %517 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %508, float %509, float %510, float %511, i32 %484, i32 %485, i32 %486, i32 %487, i32 %464, i32 %465) #2, !dbg !70 + %518 = extractvalue { float, float, float, float } %517, 0, !dbg !70 + %519 = extractvalue { float, float, float, float } %517, 1, !dbg !70 + %520 = extractvalue { float, float, float, float } %517, 2, !dbg !70 + %521 = extractvalue { float, float, float, float } %517, 3, !dbg !70 + %522 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %513, float %514, float %515, float %516, i32 %484, i32 %485, i32 %486, i32 %487, i32 %469, i32 %470) #2, !dbg !70 + %523 = extractvalue { float, float, float, float } %522, 0, !dbg !70 + %524 = extractvalue { float, float, float, float } %522, 1, !dbg !70 + %525 = extractvalue { float, float, float, float } %522, 2, !dbg !70 + %526 = extractvalue { float, float, float, float } %522, 3, !dbg !70 + %527 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %518, float %519, float %520, float %521, i32 %489, i32 %490, i32 %491, i32 %492, i32 %466, i32 %467) #2, !dbg !70 + %528 = extractvalue { float, float, float, float } %527, 0, !dbg !70 + %529 = extractvalue { float, float, float, float } %527, 1, !dbg !70 + %530 = extractvalue { float, float, float, float } %527, 2, !dbg !70 + %531 = extractvalue { float, float, float, float } %527, 3, !dbg !70 + %532 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %523, float %524, float %525, float %526, i32 %489, i32 %490, i32 %491, i32 %492, i32 %471, i32 %472) #2, !dbg !70 + %533 = extractvalue { float, float, float, float } %532, 0, !dbg !70 + %534 = extractvalue { float, float, float, float } %532, 1, !dbg !70 + %535 = extractvalue { float, float, float, float } %532, 2, !dbg !70 + %536 = extractvalue { float, float, float, float } %532, 3, !dbg !70 + %537 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %331, float %332, float %333, float %334, i32 %474, i32 %475, i32 %476, i32 %477, i32 %446, i32 %445) #2, !dbg !71 + %538 = extractvalue { float, float, float, float } %537, 0, !dbg !71 + %539 = extractvalue { float, float, float, float } %537, 1, !dbg !71 + %540 = extractvalue { float, float, float, float } %537, 2, !dbg !71 + %541 = extractvalue { float, float, float, float } %537, 3, !dbg !71 + %542 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %335, float %336, float %337, float %338, i32 %474, i32 %475, i32 %476, i32 %477, i32 %442, i32 %441) #2, !dbg !71 + %543 = extractvalue { float, float, float, float } %542, 0, !dbg !71 + %544 = extractvalue { float, float, float, float } %542, 1, !dbg !71 + %545 = extractvalue { float, float, float, float } %542, 2, !dbg !71 + %546 = extractvalue { float, float, float, float } %542, 3, !dbg !71 + %547 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %538, float %539, float %540, float %541, i32 %479, i32 %480, i32 %481, i32 %482, i32 %444, i32 %443) #2, !dbg !71 + %548 = extractvalue { float, float, float, float } %547, 0, !dbg !71 + %549 = extractvalue { float, float, float, float } %547, 1, !dbg !71 + %550 = extractvalue { float, float, float, float } %547, 2, !dbg !71 + %551 = extractvalue { float, float, float, float } %547, 3, !dbg !71 + %552 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %543, float %544, float %545, float %546, i32 %479, i32 %480, i32 %481, i32 %482, i32 %440, i32 %439) #2, !dbg !71 + %553 = extractvalue { float, float, float, float } %552, 0, !dbg !71 + %554 = extractvalue { float, float, float, float } %552, 1, !dbg !71 + %555 = extractvalue { float, float, float, float } %552, 2, !dbg !71 + %556 = extractvalue { float, float, float, float } %552, 3, !dbg !71 + %557 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %548, float %549, float %550, float %551, i32 %484, i32 %485, i32 %486, i32 %487, i32 %438, i32 %437) #2, !dbg !71 + %558 = extractvalue { float, float, float, float } %557, 0, !dbg !71 + %559 = extractvalue { float, float, float, float } %557, 1, !dbg !71 + %560 = extractvalue { float, float, float, float } %557, 2, !dbg !71 + %561 = extractvalue { float, float, float, float } %557, 3, !dbg !71 + %562 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %553, float %554, float %555, float %556, i32 %484, i32 %485, i32 %486, i32 %487, i32 %434, i32 %433) #2, !dbg !71 + %563 = extractvalue { float, float, float, float } %562, 0, !dbg !71 + %564 = extractvalue { float, float, float, float } %562, 1, !dbg !71 + %565 = extractvalue { float, float, float, float } %562, 2, !dbg !71 + %566 = extractvalue { float, float, float, float } %562, 3, !dbg !71 + %567 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %558, float %559, float %560, float %561, i32 %489, i32 %490, i32 %491, i32 %492, i32 %436, i32 %435) #2, !dbg !71 + %568 = extractvalue { float, float, float, float } %567, 0, !dbg !71 + %569 = extractvalue { float, float, float, float } %567, 1, !dbg !71 + %570 = extractvalue { float, float, float, float } %567, 2, !dbg !71 + %571 = extractvalue { float, float, float, float } %567, 3, !dbg !71 + %572 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %563, float %564, float %565, float %566, i32 %489, i32 %490, i32 %491, i32 %492, i32 %432, i32 %431) #2, !dbg !71 + %573 = extractvalue { float, float, float, float } %572, 0, !dbg !71 + %574 = extractvalue { float, float, float, float } %572, 1, !dbg !71 + %575 = extractvalue { float, float, float, float } %572, 2, !dbg !71 + %576 = extractvalue { float, float, float, float } %572, 3, !dbg !71 + %577 = insertelement <4 x float> poison, float %528, i64 0, !dbg !70 + %578 = insertelement <4 x float> %577, float %529, i64 1, !dbg !70 + %579 = insertelement <4 x float> %578, float %535, i64 2, !dbg !70 + %580 = insertelement <4 x float> %579, float %536, i64 3, !dbg !70 + br i1 %279, label %581, label %971, !dbg !72 + +581: ; preds = %430 + %582 = getelementptr bfloat, ptr addrspace(1) %280, i64 %indvars.iv542, !dbg !73 + %583 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %582, i1 %37) #2, !dbg !74 + %584 = extractvalue { i32, i32, i32, i32 } %583, 0, !dbg !74 + %585 = extractvalue { i32, i32, i32, i32 } %583, 1, !dbg !74 + %586 = extractvalue { i32, i32, i32, i32 } %583, 2, !dbg !74 + %587 = extractvalue { i32, i32, i32, i32 } %583, 3, !dbg !74 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !74 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %262, i32 %584, i32 %585, i32 %586, i32 %587) #2, !dbg !74 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %267, i32 %584, i32 %585, i32 %586, i32 %587) #2, !dbg !75 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !75 + %588 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %270), !dbg !75 + %589 = extractvalue { i32, i32, i32, i32 } %588, 0, !dbg !75 + %590 = extractvalue { i32, i32, i32, i32 } %588, 1, !dbg !75 + %591 = extractvalue { i32, i32, i32, i32 } %588, 2, !dbg !75 + %592 = extractvalue { i32, i32, i32, i32 } %588, 3, !dbg !75 + %593 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %271), !dbg !75 + %594 = extractvalue { i32, i32, i32, i32 } %593, 0, !dbg !75 + %595 = extractvalue { i32, i32, i32, i32 } %593, 1, !dbg !75 + %596 = extractvalue { i32, i32, i32, i32 } %593, 2, !dbg !75 + %597 = extractvalue { i32, i32, i32, i32 } %593, 3, !dbg !75 + %598 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %273), !dbg !75 + %599 = extractvalue { i32, i32, i32, i32 } %598, 0, !dbg !75 + %600 = extractvalue { i32, i32, i32, i32 } %598, 1, !dbg !75 + %601 = extractvalue { i32, i32, i32, i32 } %598, 2, !dbg !75 + %602 = extractvalue { i32, i32, i32, i32 } %598, 3, !dbg !75 + %603 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %274), !dbg !75 + %604 = extractvalue { i32, i32, i32, i32 } %603, 0, !dbg !75 + %605 = extractvalue { i32, i32, i32, i32 } %603, 1, !dbg !75 + %606 = extractvalue { i32, i32, i32, i32 } %603, 2, !dbg !75 + %607 = extractvalue { i32, i32, i32, i32 } %603, 3, !dbg !75 + %608 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %275), !dbg !74 + %609 = extractvalue { i32, i32, i32, i32 } %608, 0, !dbg !74 + %610 = extractvalue { i32, i32, i32, i32 } %608, 1, !dbg !74 + %611 = extractvalue { i32, i32, i32, i32 } %608, 2, !dbg !74 + %612 = extractvalue { i32, i32, i32, i32 } %608, 3, !dbg !74 + %613 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %276), !dbg !74 + %614 = extractvalue { i32, i32, i32, i32 } %613, 0, !dbg !74 + %615 = extractvalue { i32, i32, i32, i32 } %613, 1, !dbg !74 + %616 = extractvalue { i32, i32, i32, i32 } %613, 2, !dbg !74 + %617 = extractvalue { i32, i32, i32, i32 } %613, 3, !dbg !74 + %618 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %277), !dbg !74 + %619 = extractvalue { i32, i32, i32, i32 } %618, 0, !dbg !74 + %620 = extractvalue { i32, i32, i32, i32 } %618, 1, !dbg !74 + %621 = extractvalue { i32, i32, i32, i32 } %618, 2, !dbg !74 + %622 = extractvalue { i32, i32, i32, i32 } %618, 3, !dbg !74 + %623 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %278), !dbg !74 + %624 = extractvalue { i32, i32, i32, i32 } %623, 0, !dbg !74 + %625 = extractvalue { i32, i32, i32, i32 } %623, 1, !dbg !74 + %626 = extractvalue { i32, i32, i32, i32 } %623, 2, !dbg !74 + %627 = extractvalue { i32, i32, i32, i32 } %623, 3, !dbg !74 + %628 = extractelement <4 x float> %356, i64 0, !dbg !76 + %629 = extractelement <4 x float> %356, i64 1, !dbg !76 + %630 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %628, float %629, float %343, float %344, i32 %609, i32 %610, i32 %611, i32 %612, i32 %589, i32 %590) #2, !dbg !76 + %631 = extractvalue { float, float, float, float } %630, 0, !dbg !76 + %632 = extractvalue { float, float, float, float } %630, 1, !dbg !76 + %633 = extractvalue { float, float, float, float } %630, 2, !dbg !76 + %634 = extractvalue { float, float, float, float } %630, 3, !dbg !76 + %635 = extractelement <4 x float> %356, i64 2, !dbg !76 + %636 = extractelement <4 x float> %356, i64 3, !dbg !76 + %637 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %345, float %346, float %635, float %636, i32 %609, i32 %610, i32 %611, i32 %612, i32 %594, i32 %595) #2, !dbg !76 + %638 = extractvalue { float, float, float, float } %637, 0, !dbg !76 + %639 = extractvalue { float, float, float, float } %637, 1, !dbg !76 + %640 = extractvalue { float, float, float, float } %637, 2, !dbg !76 + %641 = extractvalue { float, float, float, float } %637, 3, !dbg !76 + %642 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %631, float %632, float %633, float %634, i32 %614, i32 %615, i32 %616, i32 %617, i32 %591, i32 %592) #2, !dbg !76 + %643 = extractvalue { float, float, float, float } %642, 0, !dbg !76 + %644 = extractvalue { float, float, float, float } %642, 1, !dbg !76 + %645 = extractvalue { float, float, float, float } %642, 2, !dbg !76 + %646 = extractvalue { float, float, float, float } %642, 3, !dbg !76 + %647 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %638, float %639, float %640, float %641, i32 %614, i32 %615, i32 %616, i32 %617, i32 %596, i32 %597) #2, !dbg !76 + %648 = extractvalue { float, float, float, float } %647, 0, !dbg !76 + %649 = extractvalue { float, float, float, float } %647, 1, !dbg !76 + %650 = extractvalue { float, float, float, float } %647, 2, !dbg !76 + %651 = extractvalue { float, float, float, float } %647, 3, !dbg !76 + %652 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %643, float %644, float %645, float %646, i32 %619, i32 %620, i32 %621, i32 %622, i32 %599, i32 %600) #2, !dbg !76 + %653 = extractvalue { float, float, float, float } %652, 0, !dbg !76 + %654 = extractvalue { float, float, float, float } %652, 1, !dbg !76 + %655 = extractvalue { float, float, float, float } %652, 2, !dbg !76 + %656 = extractvalue { float, float, float, float } %652, 3, !dbg !76 + %657 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %648, float %649, float %650, float %651, i32 %619, i32 %620, i32 %621, i32 %622, i32 %604, i32 %605) #2, !dbg !76 + %658 = extractvalue { float, float, float, float } %657, 0, !dbg !76 + %659 = extractvalue { float, float, float, float } %657, 1, !dbg !76 + %660 = extractvalue { float, float, float, float } %657, 2, !dbg !76 + %661 = extractvalue { float, float, float, float } %657, 3, !dbg !76 + %662 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %653, float %654, float %655, float %656, i32 %624, i32 %625, i32 %626, i32 %627, i32 %601, i32 %602) #2, !dbg !76 + %663 = extractvalue { float, float, float, float } %662, 0, !dbg !76 + %664 = extractvalue { float, float, float, float } %662, 1, !dbg !76 + %665 = extractvalue { float, float, float, float } %662, 2, !dbg !76 + %666 = extractvalue { float, float, float, float } %662, 3, !dbg !76 + %667 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %658, float %659, float %660, float %661, i32 %624, i32 %625, i32 %626, i32 %627, i32 %606, i32 %607) #2, !dbg !76 + %668 = extractvalue { float, float, float, float } %667, 0, !dbg !76 + %669 = extractvalue { float, float, float, float } %667, 1, !dbg !76 + %670 = extractvalue { float, float, float, float } %667, 2, !dbg !76 + %671 = extractvalue { float, float, float, float } %667, 3, !dbg !76 + %672 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %323, float %324, float %325, float %326, i32 %609, i32 %610, i32 %611, i32 %612, i32 %446, i32 %445) #2, !dbg !77 + %673 = extractvalue { float, float, float, float } %672, 0, !dbg !77 + %674 = extractvalue { float, float, float, float } %672, 1, !dbg !77 + %675 = extractvalue { float, float, float, float } %672, 2, !dbg !77 + %676 = extractvalue { float, float, float, float } %672, 3, !dbg !77 + %677 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %327, float %328, float %329, float %330, i32 %609, i32 %610, i32 %611, i32 %612, i32 %442, i32 %441) #2, !dbg !77 + %678 = extractvalue { float, float, float, float } %677, 0, !dbg !77 + %679 = extractvalue { float, float, float, float } %677, 1, !dbg !77 + %680 = extractvalue { float, float, float, float } %677, 2, !dbg !77 + %681 = extractvalue { float, float, float, float } %677, 3, !dbg !77 + %682 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %673, float %674, float %675, float %676, i32 %614, i32 %615, i32 %616, i32 %617, i32 %444, i32 %443) #2, !dbg !77 + %683 = extractvalue { float, float, float, float } %682, 0, !dbg !77 + %684 = extractvalue { float, float, float, float } %682, 1, !dbg !77 + %685 = extractvalue { float, float, float, float } %682, 2, !dbg !77 + %686 = extractvalue { float, float, float, float } %682, 3, !dbg !77 + %687 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %678, float %679, float %680, float %681, i32 %614, i32 %615, i32 %616, i32 %617, i32 %440, i32 %439) #2, !dbg !77 + %688 = extractvalue { float, float, float, float } %687, 0, !dbg !77 + %689 = extractvalue { float, float, float, float } %687, 1, !dbg !77 + %690 = extractvalue { float, float, float, float } %687, 2, !dbg !77 + %691 = extractvalue { float, float, float, float } %687, 3, !dbg !77 + %692 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %683, float %684, float %685, float %686, i32 %619, i32 %620, i32 %621, i32 %622, i32 %438, i32 %437) #2, !dbg !77 + %693 = extractvalue { float, float, float, float } %692, 0, !dbg !77 + %694 = extractvalue { float, float, float, float } %692, 1, !dbg !77 + %695 = extractvalue { float, float, float, float } %692, 2, !dbg !77 + %696 = extractvalue { float, float, float, float } %692, 3, !dbg !77 + %697 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %688, float %689, float %690, float %691, i32 %619, i32 %620, i32 %621, i32 %622, i32 %434, i32 %433) #2, !dbg !77 + %698 = extractvalue { float, float, float, float } %697, 0, !dbg !77 + %699 = extractvalue { float, float, float, float } %697, 1, !dbg !77 + %700 = extractvalue { float, float, float, float } %697, 2, !dbg !77 + %701 = extractvalue { float, float, float, float } %697, 3, !dbg !77 + %702 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %693, float %694, float %695, float %696, i32 %624, i32 %625, i32 %626, i32 %627, i32 %436, i32 %435) #2, !dbg !77 + %703 = extractvalue { float, float, float, float } %702, 0, !dbg !77 + %704 = extractvalue { float, float, float, float } %702, 1, !dbg !77 + %705 = extractvalue { float, float, float, float } %702, 2, !dbg !77 + %706 = extractvalue { float, float, float, float } %702, 3, !dbg !77 + %707 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %698, float %699, float %700, float %701, i32 %624, i32 %625, i32 %626, i32 %627, i32 %432, i32 %431) #2, !dbg !77 + %708 = extractvalue { float, float, float, float } %707, 0, !dbg !77 + %709 = extractvalue { float, float, float, float } %707, 1, !dbg !77 + %710 = extractvalue { float, float, float, float } %707, 2, !dbg !77 + %711 = extractvalue { float, float, float, float } %707, 3, !dbg !77 + %712 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %315, float %316, float %317, float %318, i32 %609, i32 %610, i32 %611, i32 %612, i32 %454, i32 %455) #2, !dbg !78 + %713 = extractvalue { float, float, float, float } %712, 0, !dbg !78 + %714 = extractvalue { float, float, float, float } %712, 1, !dbg !78 + %715 = extractvalue { float, float, float, float } %712, 2, !dbg !78 + %716 = extractvalue { float, float, float, float } %712, 3, !dbg !78 + %717 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %319, float %320, float %321, float %322, i32 %609, i32 %610, i32 %611, i32 %612, i32 %459, i32 %460) #2, !dbg !78 + %718 = extractvalue { float, float, float, float } %717, 0, !dbg !78 + %719 = extractvalue { float, float, float, float } %717, 1, !dbg !78 + %720 = extractvalue { float, float, float, float } %717, 2, !dbg !78 + %721 = extractvalue { float, float, float, float } %717, 3, !dbg !78 + %722 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %713, float %714, float %715, float %716, i32 %614, i32 %615, i32 %616, i32 %617, i32 %456, i32 %457) #2, !dbg !78 + %723 = extractvalue { float, float, float, float } %722, 0, !dbg !78 + %724 = extractvalue { float, float, float, float } %722, 1, !dbg !78 + %725 = extractvalue { float, float, float, float } %722, 2, !dbg !78 + %726 = extractvalue { float, float, float, float } %722, 3, !dbg !78 + %727 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %718, float %719, float %720, float %721, i32 %614, i32 %615, i32 %616, i32 %617, i32 %461, i32 %462) #2, !dbg !78 + %728 = extractvalue { float, float, float, float } %727, 0, !dbg !78 + %729 = extractvalue { float, float, float, float } %727, 1, !dbg !78 + %730 = extractvalue { float, float, float, float } %727, 2, !dbg !78 + %731 = extractvalue { float, float, float, float } %727, 3, !dbg !78 + %732 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %723, float %724, float %725, float %726, i32 %619, i32 %620, i32 %621, i32 %622, i32 %464, i32 %465) #2, !dbg !78 + %733 = extractvalue { float, float, float, float } %732, 0, !dbg !78 + %734 = extractvalue { float, float, float, float } %732, 1, !dbg !78 + %735 = extractvalue { float, float, float, float } %732, 2, !dbg !78 + %736 = extractvalue { float, float, float, float } %732, 3, !dbg !78 + %737 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %728, float %729, float %730, float %731, i32 %619, i32 %620, i32 %621, i32 %622, i32 %469, i32 %470) #2, !dbg !78 + %738 = extractvalue { float, float, float, float } %737, 0, !dbg !78 + %739 = extractvalue { float, float, float, float } %737, 1, !dbg !78 + %740 = extractvalue { float, float, float, float } %737, 2, !dbg !78 + %741 = extractvalue { float, float, float, float } %737, 3, !dbg !78 + %742 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %733, float %734, float %735, float %736, i32 %624, i32 %625, i32 %626, i32 %627, i32 %466, i32 %467) #2, !dbg !78 + %743 = extractvalue { float, float, float, float } %742, 0, !dbg !78 + %744 = extractvalue { float, float, float, float } %742, 1, !dbg !78 + %745 = extractvalue { float, float, float, float } %742, 2, !dbg !78 + %746 = extractvalue { float, float, float, float } %742, 3, !dbg !78 + %747 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %738, float %739, float %740, float %741, i32 %624, i32 %625, i32 %626, i32 %627, i32 %471, i32 %472) #2, !dbg !78 + %748 = extractvalue { float, float, float, float } %747, 0, !dbg !78 + %749 = extractvalue { float, float, float, float } %747, 1, !dbg !78 + %750 = extractvalue { float, float, float, float } %747, 2, !dbg !78 + %751 = extractvalue { float, float, float, float } %747, 3, !dbg !78 + %752 = insertelement <4 x float> poison, float %663, i64 0, !dbg !72 + %753 = insertelement <4 x float> %752, float %664, i64 1, !dbg !72 + %754 = insertelement <4 x float> %753, float %670, i64 2, !dbg !72 + %755 = insertelement <4 x float> %754, float %671, i64 3, !dbg !72 + br i1 %281, label %756, label %971, !dbg !79 + +756: ; preds = %581 + %757 = getelementptr bfloat, ptr addrspace(1) %282, i64 %indvars.iv542, !dbg !80 + %758 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %757, i1 %39) #2, !dbg !81 + %759 = extractvalue { i32, i32, i32, i32 } %758, 0, !dbg !81 + %760 = extractvalue { i32, i32, i32, i32 } %758, 1, !dbg !81 + %761 = extractvalue { i32, i32, i32, i32 } %758, 2, !dbg !81 + %762 = extractvalue { i32, i32, i32, i32 } %758, 3, !dbg !81 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !81 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %262, i32 %759, i32 %760, i32 %761, i32 %762) #2, !dbg !81 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %267, i32 %759, i32 %760, i32 %761, i32 %762) #2, !dbg !82 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !82 + %763 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %270), !dbg !82 + %764 = extractvalue { i32, i32, i32, i32 } %763, 0, !dbg !82 + %765 = extractvalue { i32, i32, i32, i32 } %763, 1, !dbg !82 + %766 = extractvalue { i32, i32, i32, i32 } %763, 2, !dbg !82 + %767 = extractvalue { i32, i32, i32, i32 } %763, 3, !dbg !82 + %768 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %271), !dbg !82 + %769 = extractvalue { i32, i32, i32, i32 } %768, 0, !dbg !82 + %770 = extractvalue { i32, i32, i32, i32 } %768, 1, !dbg !82 + %771 = extractvalue { i32, i32, i32, i32 } %768, 2, !dbg !82 + %772 = extractvalue { i32, i32, i32, i32 } %768, 3, !dbg !82 + %773 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %273), !dbg !82 + %774 = extractvalue { i32, i32, i32, i32 } %773, 0, !dbg !82 + %775 = extractvalue { i32, i32, i32, i32 } %773, 1, !dbg !82 + %776 = extractvalue { i32, i32, i32, i32 } %773, 2, !dbg !82 + %777 = extractvalue { i32, i32, i32, i32 } %773, 3, !dbg !82 + %778 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %274), !dbg !82 + %779 = extractvalue { i32, i32, i32, i32 } %778, 0, !dbg !82 + %780 = extractvalue { i32, i32, i32, i32 } %778, 1, !dbg !82 + %781 = extractvalue { i32, i32, i32, i32 } %778, 2, !dbg !82 + %782 = extractvalue { i32, i32, i32, i32 } %778, 3, !dbg !82 + %783 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %275), !dbg !81 + %784 = extractvalue { i32, i32, i32, i32 } %783, 0, !dbg !81 + %785 = extractvalue { i32, i32, i32, i32 } %783, 1, !dbg !81 + %786 = extractvalue { i32, i32, i32, i32 } %783, 2, !dbg !81 + %787 = extractvalue { i32, i32, i32, i32 } %783, 3, !dbg !81 + %788 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %276), !dbg !81 + %789 = extractvalue { i32, i32, i32, i32 } %788, 0, !dbg !81 + %790 = extractvalue { i32, i32, i32, i32 } %788, 1, !dbg !81 + %791 = extractvalue { i32, i32, i32, i32 } %788, 2, !dbg !81 + %792 = extractvalue { i32, i32, i32, i32 } %788, 3, !dbg !81 + %793 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %277), !dbg !81 + %794 = extractvalue { i32, i32, i32, i32 } %793, 0, !dbg !81 + %795 = extractvalue { i32, i32, i32, i32 } %793, 1, !dbg !81 + %796 = extractvalue { i32, i32, i32, i32 } %793, 2, !dbg !81 + %797 = extractvalue { i32, i32, i32, i32 } %793, 3, !dbg !81 + %798 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %278), !dbg !81 + %799 = extractvalue { i32, i32, i32, i32 } %798, 0, !dbg !81 + %800 = extractvalue { i32, i32, i32, i32 } %798, 1, !dbg !81 + %801 = extractvalue { i32, i32, i32, i32 } %798, 2, !dbg !81 + %802 = extractvalue { i32, i32, i32, i32 } %798, 3, !dbg !81 + %803 = extractelement <4 x float> %355, i64 0, !dbg !83 + %804 = extractelement <4 x float> %355, i64 1, !dbg !83 + %805 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %803, float %804, float %339, float %340, i32 %784, i32 %785, i32 %786, i32 %787, i32 %764, i32 %765) #2, !dbg !83 + %806 = extractvalue { float, float, float, float } %805, 0, !dbg !83 + %807 = extractvalue { float, float, float, float } %805, 1, !dbg !83 + %808 = extractvalue { float, float, float, float } %805, 2, !dbg !83 + %809 = extractvalue { float, float, float, float } %805, 3, !dbg !83 + %810 = extractelement <4 x float> %355, i64 2, !dbg !83 + %811 = extractelement <4 x float> %355, i64 3, !dbg !83 + %812 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %341, float %342, float %810, float %811, i32 %784, i32 %785, i32 %786, i32 %787, i32 %769, i32 %770) #2, !dbg !83 + %813 = extractvalue { float, float, float, float } %812, 0, !dbg !83 + %814 = extractvalue { float, float, float, float } %812, 1, !dbg !83 + %815 = extractvalue { float, float, float, float } %812, 2, !dbg !83 + %816 = extractvalue { float, float, float, float } %812, 3, !dbg !83 + %817 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %806, float %807, float %808, float %809, i32 %789, i32 %790, i32 %791, i32 %792, i32 %766, i32 %767) #2, !dbg !83 + %818 = extractvalue { float, float, float, float } %817, 0, !dbg !83 + %819 = extractvalue { float, float, float, float } %817, 1, !dbg !83 + %820 = extractvalue { float, float, float, float } %817, 2, !dbg !83 + %821 = extractvalue { float, float, float, float } %817, 3, !dbg !83 + %822 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %813, float %814, float %815, float %816, i32 %789, i32 %790, i32 %791, i32 %792, i32 %771, i32 %772) #2, !dbg !83 + %823 = extractvalue { float, float, float, float } %822, 0, !dbg !83 + %824 = extractvalue { float, float, float, float } %822, 1, !dbg !83 + %825 = extractvalue { float, float, float, float } %822, 2, !dbg !83 + %826 = extractvalue { float, float, float, float } %822, 3, !dbg !83 + %827 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %818, float %819, float %820, float %821, i32 %794, i32 %795, i32 %796, i32 %797, i32 %774, i32 %775) #2, !dbg !83 + %828 = extractvalue { float, float, float, float } %827, 0, !dbg !83 + %829 = extractvalue { float, float, float, float } %827, 1, !dbg !83 + %830 = extractvalue { float, float, float, float } %827, 2, !dbg !83 + %831 = extractvalue { float, float, float, float } %827, 3, !dbg !83 + %832 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %823, float %824, float %825, float %826, i32 %794, i32 %795, i32 %796, i32 %797, i32 %779, i32 %780) #2, !dbg !83 + %833 = extractvalue { float, float, float, float } %832, 0, !dbg !83 + %834 = extractvalue { float, float, float, float } %832, 1, !dbg !83 + %835 = extractvalue { float, float, float, float } %832, 2, !dbg !83 + %836 = extractvalue { float, float, float, float } %832, 3, !dbg !83 + %837 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %828, float %829, float %830, float %831, i32 %799, i32 %800, i32 %801, i32 %802, i32 %776, i32 %777) #2, !dbg !83 + %838 = extractvalue { float, float, float, float } %837, 0, !dbg !83 + %839 = extractvalue { float, float, float, float } %837, 1, !dbg !83 + %840 = extractvalue { float, float, float, float } %837, 2, !dbg !83 + %841 = extractvalue { float, float, float, float } %837, 3, !dbg !83 + %842 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %833, float %834, float %835, float %836, i32 %799, i32 %800, i32 %801, i32 %802, i32 %781, i32 %782) #2, !dbg !83 + %843 = extractvalue { float, float, float, float } %842, 0, !dbg !83 + %844 = extractvalue { float, float, float, float } %842, 1, !dbg !83 + %845 = extractvalue { float, float, float, float } %842, 2, !dbg !83 + %846 = extractvalue { float, float, float, float } %842, 3, !dbg !83 + %847 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %307, float %308, float %309, float %310, i32 %784, i32 %785, i32 %786, i32 %787, i32 %446, i32 %445) #2, !dbg !84 + %848 = extractvalue { float, float, float, float } %847, 0, !dbg !84 + %849 = extractvalue { float, float, float, float } %847, 1, !dbg !84 + %850 = extractvalue { float, float, float, float } %847, 2, !dbg !84 + %851 = extractvalue { float, float, float, float } %847, 3, !dbg !84 + %852 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %311, float %312, float %313, float %314, i32 %784, i32 %785, i32 %786, i32 %787, i32 %442, i32 %441) #2, !dbg !84 + %853 = extractvalue { float, float, float, float } %852, 0, !dbg !84 + %854 = extractvalue { float, float, float, float } %852, 1, !dbg !84 + %855 = extractvalue { float, float, float, float } %852, 2, !dbg !84 + %856 = extractvalue { float, float, float, float } %852, 3, !dbg !84 + %857 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %848, float %849, float %850, float %851, i32 %789, i32 %790, i32 %791, i32 %792, i32 %444, i32 %443) #2, !dbg !84 + %858 = extractvalue { float, float, float, float } %857, 0, !dbg !84 + %859 = extractvalue { float, float, float, float } %857, 1, !dbg !84 + %860 = extractvalue { float, float, float, float } %857, 2, !dbg !84 + %861 = extractvalue { float, float, float, float } %857, 3, !dbg !84 + %862 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %853, float %854, float %855, float %856, i32 %789, i32 %790, i32 %791, i32 %792, i32 %440, i32 %439) #2, !dbg !84 + %863 = extractvalue { float, float, float, float } %862, 0, !dbg !84 + %864 = extractvalue { float, float, float, float } %862, 1, !dbg !84 + %865 = extractvalue { float, float, float, float } %862, 2, !dbg !84 + %866 = extractvalue { float, float, float, float } %862, 3, !dbg !84 + %867 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %858, float %859, float %860, float %861, i32 %794, i32 %795, i32 %796, i32 %797, i32 %438, i32 %437) #2, !dbg !84 + %868 = extractvalue { float, float, float, float } %867, 0, !dbg !84 + %869 = extractvalue { float, float, float, float } %867, 1, !dbg !84 + %870 = extractvalue { float, float, float, float } %867, 2, !dbg !84 + %871 = extractvalue { float, float, float, float } %867, 3, !dbg !84 + %872 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %863, float %864, float %865, float %866, i32 %794, i32 %795, i32 %796, i32 %797, i32 %434, i32 %433) #2, !dbg !84 + %873 = extractvalue { float, float, float, float } %872, 0, !dbg !84 + %874 = extractvalue { float, float, float, float } %872, 1, !dbg !84 + %875 = extractvalue { float, float, float, float } %872, 2, !dbg !84 + %876 = extractvalue { float, float, float, float } %872, 3, !dbg !84 + %877 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %868, float %869, float %870, float %871, i32 %799, i32 %800, i32 %801, i32 %802, i32 %436, i32 %435) #2, !dbg !84 + %878 = extractvalue { float, float, float, float } %877, 0, !dbg !84 + %879 = extractvalue { float, float, float, float } %877, 1, !dbg !84 + %880 = extractvalue { float, float, float, float } %877, 2, !dbg !84 + %881 = extractvalue { float, float, float, float } %877, 3, !dbg !84 + %882 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %873, float %874, float %875, float %876, i32 %799, i32 %800, i32 %801, i32 %802, i32 %432, i32 %431) #2, !dbg !84 + %883 = extractvalue { float, float, float, float } %882, 0, !dbg !84 + %884 = extractvalue { float, float, float, float } %882, 1, !dbg !84 + %885 = extractvalue { float, float, float, float } %882, 2, !dbg !84 + %886 = extractvalue { float, float, float, float } %882, 3, !dbg !84 + %887 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %299, float %300, float %301, float %302, i32 %784, i32 %785, i32 %786, i32 %787, i32 %454, i32 %455) #2, !dbg !85 + %888 = extractvalue { float, float, float, float } %887, 0, !dbg !85 + %889 = extractvalue { float, float, float, float } %887, 1, !dbg !85 + %890 = extractvalue { float, float, float, float } %887, 2, !dbg !85 + %891 = extractvalue { float, float, float, float } %887, 3, !dbg !85 + %892 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %303, float %304, float %305, float %306, i32 %784, i32 %785, i32 %786, i32 %787, i32 %459, i32 %460) #2, !dbg !85 + %893 = extractvalue { float, float, float, float } %892, 0, !dbg !85 + %894 = extractvalue { float, float, float, float } %892, 1, !dbg !85 + %895 = extractvalue { float, float, float, float } %892, 2, !dbg !85 + %896 = extractvalue { float, float, float, float } %892, 3, !dbg !85 + %897 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %888, float %889, float %890, float %891, i32 %789, i32 %790, i32 %791, i32 %792, i32 %456, i32 %457) #2, !dbg !85 + %898 = extractvalue { float, float, float, float } %897, 0, !dbg !85 + %899 = extractvalue { float, float, float, float } %897, 1, !dbg !85 + %900 = extractvalue { float, float, float, float } %897, 2, !dbg !85 + %901 = extractvalue { float, float, float, float } %897, 3, !dbg !85 + %902 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %893, float %894, float %895, float %896, i32 %789, i32 %790, i32 %791, i32 %792, i32 %461, i32 %462) #2, !dbg !85 + %903 = extractvalue { float, float, float, float } %902, 0, !dbg !85 + %904 = extractvalue { float, float, float, float } %902, 1, !dbg !85 + %905 = extractvalue { float, float, float, float } %902, 2, !dbg !85 + %906 = extractvalue { float, float, float, float } %902, 3, !dbg !85 + %907 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %898, float %899, float %900, float %901, i32 %794, i32 %795, i32 %796, i32 %797, i32 %464, i32 %465) #2, !dbg !85 + %908 = extractvalue { float, float, float, float } %907, 0, !dbg !85 + %909 = extractvalue { float, float, float, float } %907, 1, !dbg !85 + %910 = extractvalue { float, float, float, float } %907, 2, !dbg !85 + %911 = extractvalue { float, float, float, float } %907, 3, !dbg !85 + %912 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %903, float %904, float %905, float %906, i32 %794, i32 %795, i32 %796, i32 %797, i32 %469, i32 %470) #2, !dbg !85 + %913 = extractvalue { float, float, float, float } %912, 0, !dbg !85 + %914 = extractvalue { float, float, float, float } %912, 1, !dbg !85 + %915 = extractvalue { float, float, float, float } %912, 2, !dbg !85 + %916 = extractvalue { float, float, float, float } %912, 3, !dbg !85 + %917 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %908, float %909, float %910, float %911, i32 %799, i32 %800, i32 %801, i32 %802, i32 %466, i32 %467) #2, !dbg !85 + %918 = extractvalue { float, float, float, float } %917, 0, !dbg !85 + %919 = extractvalue { float, float, float, float } %917, 1, !dbg !85 + %920 = extractvalue { float, float, float, float } %917, 2, !dbg !85 + %921 = extractvalue { float, float, float, float } %917, 3, !dbg !85 + %922 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %913, float %914, float %915, float %916, i32 %799, i32 %800, i32 %801, i32 %802, i32 %471, i32 %472) #2, !dbg !85 + %923 = extractvalue { float, float, float, float } %922, 0, !dbg !85 + %924 = extractvalue { float, float, float, float } %922, 1, !dbg !85 + %925 = extractvalue { float, float, float, float } %922, 2, !dbg !85 + %926 = extractvalue { float, float, float, float } %922, 3, !dbg !85 + %927 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %291, float %292, float %293, float %294, i32 %784, i32 %785, i32 %786, i32 %787, i32 %589, i32 %590) #2, !dbg !86 + %928 = extractvalue { float, float, float, float } %927, 0, !dbg !86 + %929 = extractvalue { float, float, float, float } %927, 1, !dbg !86 + %930 = extractvalue { float, float, float, float } %927, 2, !dbg !86 + %931 = extractvalue { float, float, float, float } %927, 3, !dbg !86 + %932 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %295, float %296, float %297, float %298, i32 %784, i32 %785, i32 %786, i32 %787, i32 %594, i32 %595) #2, !dbg !86 + %933 = extractvalue { float, float, float, float } %932, 0, !dbg !86 + %934 = extractvalue { float, float, float, float } %932, 1, !dbg !86 + %935 = extractvalue { float, float, float, float } %932, 2, !dbg !86 + %936 = extractvalue { float, float, float, float } %932, 3, !dbg !86 + %937 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %928, float %929, float %930, float %931, i32 %789, i32 %790, i32 %791, i32 %792, i32 %591, i32 %592) #2, !dbg !86 + %938 = extractvalue { float, float, float, float } %937, 0, !dbg !86 + %939 = extractvalue { float, float, float, float } %937, 1, !dbg !86 + %940 = extractvalue { float, float, float, float } %937, 2, !dbg !86 + %941 = extractvalue { float, float, float, float } %937, 3, !dbg !86 + %942 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %933, float %934, float %935, float %936, i32 %789, i32 %790, i32 %791, i32 %792, i32 %596, i32 %597) #2, !dbg !86 + %943 = extractvalue { float, float, float, float } %942, 0, !dbg !86 + %944 = extractvalue { float, float, float, float } %942, 1, !dbg !86 + %945 = extractvalue { float, float, float, float } %942, 2, !dbg !86 + %946 = extractvalue { float, float, float, float } %942, 3, !dbg !86 + %947 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %938, float %939, float %940, float %941, i32 %794, i32 %795, i32 %796, i32 %797, i32 %599, i32 %600) #2, !dbg !86 + %948 = extractvalue { float, float, float, float } %947, 0, !dbg !86 + %949 = extractvalue { float, float, float, float } %947, 1, !dbg !86 + %950 = extractvalue { float, float, float, float } %947, 2, !dbg !86 + %951 = extractvalue { float, float, float, float } %947, 3, !dbg !86 + %952 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %943, float %944, float %945, float %946, i32 %794, i32 %795, i32 %796, i32 %797, i32 %604, i32 %605) #2, !dbg !86 + %953 = extractvalue { float, float, float, float } %952, 0, !dbg !86 + %954 = extractvalue { float, float, float, float } %952, 1, !dbg !86 + %955 = extractvalue { float, float, float, float } %952, 2, !dbg !86 + %956 = extractvalue { float, float, float, float } %952, 3, !dbg !86 + %957 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %948, float %949, float %950, float %951, i32 %799, i32 %800, i32 %801, i32 %802, i32 %601, i32 %602) #2, !dbg !86 + %958 = extractvalue { float, float, float, float } %957, 0, !dbg !86 + %959 = extractvalue { float, float, float, float } %957, 1, !dbg !86 + %960 = extractvalue { float, float, float, float } %957, 2, !dbg !86 + %961 = extractvalue { float, float, float, float } %957, 3, !dbg !86 + %962 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %953, float %954, float %955, float %956, i32 %799, i32 %800, i32 %801, i32 %802, i32 %606, i32 %607) #2, !dbg !86 + %963 = extractvalue { float, float, float, float } %962, 0, !dbg !86 + %964 = extractvalue { float, float, float, float } %962, 1, !dbg !86 + %965 = extractvalue { float, float, float, float } %962, 2, !dbg !86 + %966 = extractvalue { float, float, float, float } %962, 3, !dbg !86 + %967 = insertelement <4 x float> poison, float %838, i64 0, !dbg !72 + %968 = insertelement <4 x float> %967, float %839, i64 1, !dbg !72 + %969 = insertelement <4 x float> %968, float %845, i64 2, !dbg !72 + %970 = insertelement <4 x float> %969, float %846, i64 3, !dbg !72 + br label %971, !dbg !79 + +971: ; preds = %756, %581, %430, %287 + %972 = phi float [ %347, %287 ], [ %530, %430 ], [ %530, %581 ], [ %530, %756 ], !dbg !70 + %973 = phi float [ %348, %287 ], [ %531, %430 ], [ %531, %581 ], [ %531, %756 ], !dbg !70 + %974 = phi float [ %349, %287 ], [ %533, %430 ], [ %533, %581 ], [ %533, %756 ], !dbg !70 + %975 = phi float [ %350, %287 ], [ %534, %430 ], [ %534, %581 ], [ %534, %756 ], !dbg !70 + %976 = phi float [ %343, %287 ], [ %343, %430 ], [ %665, %581 ], [ %665, %756 ], !dbg !72 + %977 = phi float [ %344, %287 ], [ %344, %430 ], [ %666, %581 ], [ %666, %756 ], !dbg !72 + %978 = phi float [ %345, %287 ], [ %345, %430 ], [ %668, %581 ], [ %668, %756 ], !dbg !72 + %979 = phi float [ %346, %287 ], [ %346, %430 ], [ %669, %581 ], [ %669, %756 ], !dbg !72 + %980 = phi float [ %339, %287 ], [ %339, %430 ], [ %339, %581 ], [ %840, %756 ], !dbg !72 + %981 = phi float [ %340, %287 ], [ %340, %430 ], [ %340, %581 ], [ %841, %756 ], !dbg !72 + %982 = phi float [ %341, %287 ], [ %341, %430 ], [ %341, %581 ], [ %843, %756 ], !dbg !72 + %983 = phi float [ %342, %287 ], [ %342, %430 ], [ %342, %581 ], [ %844, %756 ], !dbg !72 + %984 = phi float [ %331, %287 ], [ %568, %430 ], [ %568, %581 ], [ %568, %756 ], !dbg !71 + %985 = phi float [ %332, %287 ], [ %569, %430 ], [ %569, %581 ], [ %569, %756 ], !dbg !71 + %986 = phi float [ %333, %287 ], [ %570, %430 ], [ %570, %581 ], [ %570, %756 ], !dbg !71 + %987 = phi float [ %334, %287 ], [ %571, %430 ], [ %571, %581 ], [ %571, %756 ], !dbg !71 + %988 = phi float [ %335, %287 ], [ %573, %430 ], [ %573, %581 ], [ %573, %756 ], !dbg !71 + %989 = phi float [ %336, %287 ], [ %574, %430 ], [ %574, %581 ], [ %574, %756 ], !dbg !71 + %990 = phi float [ %337, %287 ], [ %575, %430 ], [ %575, %581 ], [ %575, %756 ], !dbg !71 + %991 = phi float [ %338, %287 ], [ %576, %430 ], [ %576, %581 ], [ %576, %756 ], !dbg !71 + %992 = phi float [ %323, %287 ], [ %323, %430 ], [ %703, %581 ], [ %703, %756 ], !dbg !72 + %993 = phi float [ %324, %287 ], [ %324, %430 ], [ %704, %581 ], [ %704, %756 ], !dbg !72 + %994 = phi float [ %325, %287 ], [ %325, %430 ], [ %705, %581 ], [ %705, %756 ], !dbg !72 + %995 = phi float [ %326, %287 ], [ %326, %430 ], [ %706, %581 ], [ %706, %756 ], !dbg !72 + %996 = phi float [ %327, %287 ], [ %327, %430 ], [ %708, %581 ], [ %708, %756 ], !dbg !72 + %997 = phi float [ %328, %287 ], [ %328, %430 ], [ %709, %581 ], [ %709, %756 ], !dbg !72 + %998 = phi float [ %329, %287 ], [ %329, %430 ], [ %710, %581 ], [ %710, %756 ], !dbg !72 + %999 = phi float [ %330, %287 ], [ %330, %430 ], [ %711, %581 ], [ %711, %756 ], !dbg !72 + %1000 = phi float [ %315, %287 ], [ %315, %430 ], [ %743, %581 ], [ %743, %756 ], !dbg !72 + %1001 = phi float [ %316, %287 ], [ %316, %430 ], [ %744, %581 ], [ %744, %756 ], !dbg !72 + %1002 = phi float [ %317, %287 ], [ %317, %430 ], [ %745, %581 ], [ %745, %756 ], !dbg !72 + %1003 = phi float [ %318, %287 ], [ %318, %430 ], [ %746, %581 ], [ %746, %756 ], !dbg !72 + %1004 = phi float [ %319, %287 ], [ %319, %430 ], [ %748, %581 ], [ %748, %756 ], !dbg !72 + %1005 = phi float [ %320, %287 ], [ %320, %430 ], [ %749, %581 ], [ %749, %756 ], !dbg !72 + %1006 = phi float [ %321, %287 ], [ %321, %430 ], [ %750, %581 ], [ %750, %756 ], !dbg !72 + %1007 = phi float [ %322, %287 ], [ %322, %430 ], [ %751, %581 ], [ %751, %756 ], !dbg !72 + %1008 = phi float [ %307, %287 ], [ %307, %430 ], [ %307, %581 ], [ %878, %756 ], !dbg !72 + %1009 = phi float [ %308, %287 ], [ %308, %430 ], [ %308, %581 ], [ %879, %756 ], !dbg !72 + %1010 = phi float [ %309, %287 ], [ %309, %430 ], [ %309, %581 ], [ %880, %756 ], !dbg !72 + %1011 = phi float [ %310, %287 ], [ %310, %430 ], [ %310, %581 ], [ %881, %756 ], !dbg !72 + %1012 = phi float [ %311, %287 ], [ %311, %430 ], [ %311, %581 ], [ %883, %756 ], !dbg !72 + %1013 = phi float [ %312, %287 ], [ %312, %430 ], [ %312, %581 ], [ %884, %756 ], !dbg !72 + %1014 = phi float [ %313, %287 ], [ %313, %430 ], [ %313, %581 ], [ %885, %756 ], !dbg !72 + %1015 = phi float [ %314, %287 ], [ %314, %430 ], [ %314, %581 ], [ %886, %756 ], !dbg !72 + %1016 = phi float [ %299, %287 ], [ %299, %430 ], [ %299, %581 ], [ %918, %756 ], !dbg !72 + %1017 = phi float [ %300, %287 ], [ %300, %430 ], [ %300, %581 ], [ %919, %756 ], !dbg !72 + %1018 = phi float [ %301, %287 ], [ %301, %430 ], [ %301, %581 ], [ %920, %756 ], !dbg !72 + %1019 = phi float [ %302, %287 ], [ %302, %430 ], [ %302, %581 ], [ %921, %756 ], !dbg !72 + %1020 = phi float [ %303, %287 ], [ %303, %430 ], [ %303, %581 ], [ %923, %756 ], !dbg !72 + %1021 = phi float [ %304, %287 ], [ %304, %430 ], [ %304, %581 ], [ %924, %756 ], !dbg !72 + %1022 = phi float [ %305, %287 ], [ %305, %430 ], [ %305, %581 ], [ %925, %756 ], !dbg !72 + %1023 = phi float [ %306, %287 ], [ %306, %430 ], [ %306, %581 ], [ %926, %756 ], !dbg !72 + %1024 = phi float [ %291, %287 ], [ %291, %430 ], [ %291, %581 ], [ %958, %756 ], !dbg !72 + %1025 = phi float [ %292, %287 ], [ %292, %430 ], [ %292, %581 ], [ %959, %756 ], !dbg !72 + %1026 = phi float [ %293, %287 ], [ %293, %430 ], [ %293, %581 ], [ %960, %756 ], !dbg !72 + %1027 = phi float [ %294, %287 ], [ %294, %430 ], [ %294, %581 ], [ %961, %756 ], !dbg !72 + %1028 = phi float [ %295, %287 ], [ %295, %430 ], [ %295, %581 ], [ %963, %756 ], !dbg !72 + %1029 = phi float [ %296, %287 ], [ %296, %430 ], [ %296, %581 ], [ %964, %756 ], !dbg !72 + %1030 = phi float [ %297, %287 ], [ %297, %430 ], [ %297, %581 ], [ %965, %756 ], !dbg !72 + %1031 = phi float [ %298, %287 ], [ %298, %430 ], [ %298, %581 ], [ %966, %756 ], !dbg !72 + %1032 = phi <4 x float> [ %355, %287 ], [ %355, %430 ], [ %355, %581 ], [ %970, %756 ], !dbg !72 + %1033 = phi <4 x float> [ %356, %287 ], [ %356, %430 ], [ %755, %581 ], [ %755, %756 ], !dbg !72 + %1034 = phi <4 x float> [ %357, %287 ], [ %580, %430 ], [ %580, %581 ], [ %580, %756 ], !dbg !70 + %indvars.iv.next543 = add nuw nsw i64 %indvars.iv542, 64, !dbg !64 + br i1 %288, label %287, label %1035, !dbg !64 + +1035: ; preds = %971 + %.idx7 = shl nuw nsw i64 %21, 7, !dbg !87 + %1036 = getelementptr i8, ptr addrspace(1) %3, i64 %.idx7, !dbg !87 + %1037 = and i32 %23, 31, !dbg !28 + %1038 = lshr i32 %23, 5, !dbg !28 + %1039 = shl nuw nsw i32 %27, 1, !dbg !28 + %1040 = insertelement <2 x i16> poison, i16 %135, i64 0, !dbg !47 + %1041 = insertelement <2 x i16> %1040, i16 %137, i64 1, !dbg !47 + %1042 = bitcast <2 x i16> %1041 to <2 x bfloat>, !dbg !47 + %1043 = shufflevector <2 x bfloat> %1042, <2 x bfloat> poison, <4 x i32> , !dbg !88 + %1044 = fpext <4 x bfloat> %1043 to <4 x float>, !dbg !88 + %1045 = insertelement <2 x i16> poison, i16 %138, i64 0, !dbg !48 + %1046 = insertelement <2 x i16> %1045, i16 %139, i64 1, !dbg !48 + %1047 = bitcast <2 x i16> %1046 to <2 x bfloat>, !dbg !48 + %1048 = shufflevector <2 x bfloat> %1047, <2 x bfloat> poison, <4 x i32> , !dbg !89 + %1049 = fpext <4 x bfloat> %1048 to <4 x float>, !dbg !89 + %1050 = insertelement <2 x i16> poison, i16 %140, i64 0, !dbg !49 + %1051 = insertelement <2 x i16> %1050, i16 %141, i64 1, !dbg !49 + %1052 = bitcast <2 x i16> %1051 to <2 x bfloat>, !dbg !49 + %1053 = shufflevector <2 x bfloat> %1052, <2 x bfloat> poison, <4 x i32> , !dbg !90 + %1054 = fpext <4 x bfloat> %1053 to <4 x float>, !dbg !90 + %1055 = insertelement <2 x i16> poison, i16 %142, i64 0, !dbg !50 + %1056 = insertelement <2 x i16> %1055, i16 %143, i64 1, !dbg !50 + %1057 = bitcast <2 x i16> %1056 to <2 x bfloat>, !dbg !50 + %1058 = shufflevector <2 x bfloat> %1057, <2 x bfloat> poison, <4 x i32> , !dbg !91 + %1059 = fpext <4 x bfloat> %1058 to <4 x float>, !dbg !91 + %1060 = bitcast i32 %185 to float, !dbg !46 + %1061 = bitcast i32 %186 to float, !dbg !46 + %1062 = bitcast i32 %188 to float, !dbg !46 + %1063 = bitcast i32 %190 to float, !dbg !46 + %1064 = bitcast i32 %191 to float, !dbg !46 + %1065 = bitcast i32 %192 to float, !dbg !46 + %1066 = bitcast i32 %193 to float, !dbg !46 + %1067 = bitcast i32 %194 to float, !dbg !46 + %1068 = bitcast i32 %195 to float, !dbg !56 + %1069 = bitcast i32 %196 to float, !dbg !56 + %1070 = bitcast i32 %198 to float, !dbg !56 + %1071 = bitcast i32 %200 to float, !dbg !56 + %1072 = bitcast i32 %202 to float, !dbg !56 + %1073 = bitcast i32 %204 to float, !dbg !56 + %1074 = bitcast i32 %205 to float, !dbg !57 + %1075 = bitcast i32 %206 to float, !dbg !57 + %1076 = bitcast i32 %208 to float, !dbg !57 + %1077 = bitcast i32 %210 to float, !dbg !57 + %1078 = bitcast i32 %212 to float, !dbg !57 + %1079 = bitcast i32 %214 to float, !dbg !57 + %1080 = bitcast i32 %215 to float, !dbg !58 + %1081 = bitcast i32 %216 to float, !dbg !58 + %1082 = bitcast i32 %218 to float, !dbg !58 + %1083 = bitcast i32 %220 to float, !dbg !58 + %1084 = bitcast i32 %222 to float, !dbg !58 + %1085 = bitcast i32 %224 to float, !dbg !58 + tail call void @llvm.nvvm.cp.async.wait.group(i32 0), !dbg !64 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !64 + %1086 = shufflevector <2 x i32> %76, <2 x i32> poison, <4 x i32> , !dbg !92 + %1087 = shufflevector <2 x i32> %64, <2 x i32> poison, <4 x i32> , !dbg !92 + %1088 = icmp samesign ugt <4 x i32> %1086, %1087, !dbg !92 + %1089 = insertelement <4 x i32> poison, i32 %73, i64 0, !dbg !92 + %1090 = insertelement <4 x i32> %1089, i32 %74, i64 1, !dbg !92 + %1091 = shufflevector <4 x i32> %1090, <4 x i32> poison, <4 x i32> , !dbg !92 + %1092 = insertelement <4 x i32> %47, i32 %45, i64 3, !dbg !92 + %1093 = shufflevector <4 x i32> %1092, <4 x i32> poison, <4 x i32> , !dbg !92 + %1094 = icmp samesign ugt <4 x i32> %1091, %1093, !dbg !92 + %1095 = icmp eq i32 %73, %40, !dbg !93 + %1096 = icmp eq i32 %73, %43, !dbg !93 + %1097 = icmp eq i32 %74, %45, !dbg !93 + %1098 = icmp eq i32 %73, %41, !dbg !93 + %1099 = icmp eq i32 %73, %42, !dbg !93 + %1100 = icmp eq i32 %74, %41, !dbg !93 + %1101 = icmp eq i32 %74, %42, !dbg !93 + %1102 = and <4 x i1> %1088, %82, !dbg !94 + %1103 = and <4 x i1> %72, %1102, !dbg !95 + %1104 = fsub float %1060, %1062, !dbg !96 + %1105 = fsub float %1060, %1063, !dbg !96 + %1106 = fsub float %1061, %1062, !dbg !96 + %1107 = fsub float %1061, %1063, !dbg !96 + %1108 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !97 + %.not.i = icmp eq i32 %1108, 0, !dbg !97 + br i1 %.not.i, label %1111, label %1109, !dbg !97 + +1109: ; preds = %1035 + %1110 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1104), !dbg !97 + br label %__nv_exp2f.exit, !dbg !97 + +1111: ; preds = %1035 + %1112 = tail call float @llvm.nvvm.ex2.approx.f32(float %1104), !dbg !97 + br label %__nv_exp2f.exit, !dbg !97 + +__nv_exp2f.exit: ; preds = %1109, %1111 + %.0.i = phi float [ %1110, %1109 ], [ %1112, %1111 ], !dbg !97 + %1113 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !97 + %.not.i202 = icmp eq i32 %1113, 0, !dbg !97 + br i1 %.not.i202, label %1116, label %1114, !dbg !97 + +1114: ; preds = %__nv_exp2f.exit + %1115 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1105), !dbg !97 + br label %__nv_exp2f.exit204, !dbg !97 + +1116: ; preds = %__nv_exp2f.exit + %1117 = tail call float @llvm.nvvm.ex2.approx.f32(float %1105), !dbg !97 + br label %__nv_exp2f.exit204, !dbg !97 + +__nv_exp2f.exit204: ; preds = %1114, %1116 + %.0.i203 = phi float [ %1115, %1114 ], [ %1117, %1116 ], !dbg !97 + %1118 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !97 + %.not.i205 = icmp eq i32 %1118, 0, !dbg !97 + br i1 %.not.i205, label %1121, label %1119, !dbg !97 + +1119: ; preds = %__nv_exp2f.exit204 + %1120 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1106), !dbg !97 + br label %__nv_exp2f.exit207, !dbg !97 + +1121: ; preds = %__nv_exp2f.exit204 + %1122 = tail call float @llvm.nvvm.ex2.approx.f32(float %1106), !dbg !97 + br label %__nv_exp2f.exit207, !dbg !97 + +__nv_exp2f.exit207: ; preds = %1119, %1121 + %.0.i206 = phi float [ %1120, %1119 ], [ %1122, %1121 ], !dbg !97 + %1123 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !97 + %.not.i208 = icmp eq i32 %1123, 0, !dbg !97 + br i1 %.not.i208, label %1126, label %1124, !dbg !97 + +1124: ; preds = %__nv_exp2f.exit207 + %1125 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1107), !dbg !97 + br label %__nv_exp2f.exit210, !dbg !97 + +1126: ; preds = %__nv_exp2f.exit207 + %1127 = tail call float @llvm.nvvm.ex2.approx.f32(float %1107), !dbg !97 + br label %__nv_exp2f.exit210, !dbg !97 + +__nv_exp2f.exit210: ; preds = %1124, %1126 + %.0.i209 = phi float [ %1125, %1124 ], [ %1127, %1126 ], !dbg !97 + %1128 = insertelement <4 x float> poison, float %.0.i, i64 0, !dbg !102 + %1129 = insertelement <4 x float> %1128, float %.0.i203, i64 1, !dbg !102 + %1130 = insertelement <4 x float> %1129, float %.0.i206, i64 2, !dbg !102 + %1131 = insertelement <4 x float> %1130, float %.0.i209, i64 3, !dbg !102 + %1132 = select <4 x i1> %1103, <4 x float> %1131, <4 x float> zeroinitializer, !dbg !102 + %1133 = insertelement <4 x float> poison, float %422, i64 0, !dbg !103 + %1134 = insertelement <4 x float> %1133, float %423, i64 1, !dbg !103 + %1135 = insertelement <4 x float> %1134, float %424, i64 2, !dbg !103 + %1136 = insertelement <4 x float> %1135, float %425, i64 3, !dbg !103 + %1137 = fmul <4 x float> %1136, %1132, !dbg !103 + %1138 = shufflevector <4 x i1> %1094, <4 x i1> poison, <4 x i32> , !dbg !104 + %1139 = insertelement <4 x i1> poison, i1 %91, i64 0, !dbg !104 + %1140 = insertelement <4 x i1> %1139, i1 %90, i64 1, !dbg !104 + %1141 = shufflevector <4 x i1> %1140, <4 x i1> poison, <4 x i32> , !dbg !104 + %1142 = and <4 x i1> %1138, %1141, !dbg !104 + %1143 = and i1 %90, %197, !dbg !105 + %1144 = and i1 %90, %199, !dbg !105 + %1145 = and <4 x i1> %95, %1142, !dbg !105 + %1146 = fsub float %1068, %1070, !dbg !106 + %1147 = fsub float %1068, %1071, !dbg !106 + %1148 = fsub float %1069, %1070, !dbg !106 + %1149 = fsub float %1069, %1071, !dbg !106 + %1150 = fsub float %1069, %1072, !dbg !106 + %1151 = fsub float %1069, %1073, !dbg !106 + %1152 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !107 + %.not.i211 = icmp eq i32 %1152, 0, !dbg !107 + br i1 %.not.i211, label %1155, label %1153, !dbg !107 + +1153: ; preds = %__nv_exp2f.exit210 + %1154 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1146), !dbg !107 + br label %__nv_exp2f.exit213, !dbg !107 + +1155: ; preds = %__nv_exp2f.exit210 + %1156 = tail call float @llvm.nvvm.ex2.approx.f32(float %1146), !dbg !107 + br label %__nv_exp2f.exit213, !dbg !107 + +__nv_exp2f.exit213: ; preds = %1153, %1155 + %.0.i212 = phi float [ %1154, %1153 ], [ %1156, %1155 ], !dbg !107 + %1157 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !107 + %.not.i214 = icmp eq i32 %1157, 0, !dbg !107 + br i1 %.not.i214, label %1160, label %1158, !dbg !107 + +1158: ; preds = %__nv_exp2f.exit213 + %1159 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1147), !dbg !107 + br label %__nv_exp2f.exit216, !dbg !107 + +1160: ; preds = %__nv_exp2f.exit213 + %1161 = tail call float @llvm.nvvm.ex2.approx.f32(float %1147), !dbg !107 + br label %__nv_exp2f.exit216, !dbg !107 + +__nv_exp2f.exit216: ; preds = %1158, %1160 + %.0.i215 = phi float [ %1159, %1158 ], [ %1161, %1160 ], !dbg !107 + %1162 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !107 + %.not.i217 = icmp eq i32 %1162, 0, !dbg !107 + br i1 %.not.i217, label %1165, label %1163, !dbg !107 + +1163: ; preds = %__nv_exp2f.exit216 + %1164 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1148), !dbg !107 + br label %__nv_exp2f.exit219, !dbg !107 + +1165: ; preds = %__nv_exp2f.exit216 + %1166 = tail call float @llvm.nvvm.ex2.approx.f32(float %1148), !dbg !107 + br label %__nv_exp2f.exit219, !dbg !107 + +__nv_exp2f.exit219: ; preds = %1163, %1165 + %.0.i218 = phi float [ %1164, %1163 ], [ %1166, %1165 ], !dbg !107 + %1167 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !107 + %.not.i220 = icmp eq i32 %1167, 0, !dbg !107 + br i1 %.not.i220, label %1170, label %1168, !dbg !107 + +1168: ; preds = %__nv_exp2f.exit219 + %1169 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1149), !dbg !107 + br label %__nv_exp2f.exit222, !dbg !107 + +1170: ; preds = %__nv_exp2f.exit219 + %1171 = tail call float @llvm.nvvm.ex2.approx.f32(float %1149), !dbg !107 + br label %__nv_exp2f.exit222, !dbg !107 + +__nv_exp2f.exit222: ; preds = %1168, %1170 + %.0.i221 = phi float [ %1169, %1168 ], [ %1171, %1170 ], !dbg !107 + %1172 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !107 + %1173 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !107 + %1174 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !107 + %.not.i229 = icmp eq i32 %1174, 0, !dbg !107 + br i1 %.not.i229, label %1177, label %1175, !dbg !107 + +1175: ; preds = %__nv_exp2f.exit222 + %1176 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1150), !dbg !107 + br label %__nv_exp2f.exit231, !dbg !107 + +1177: ; preds = %__nv_exp2f.exit222 + %1178 = tail call float @llvm.nvvm.ex2.approx.f32(float %1150), !dbg !107 + br label %__nv_exp2f.exit231, !dbg !107 + +__nv_exp2f.exit231: ; preds = %1175, %1177 + %.0.i230 = phi float [ %1176, %1175 ], [ %1178, %1177 ], !dbg !107 + %1179 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !107 + %.not.i232 = icmp eq i32 %1179, 0, !dbg !107 + br i1 %.not.i232, label %1182, label %1180, !dbg !107 + +1180: ; preds = %__nv_exp2f.exit231 + %1181 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1151), !dbg !107 + br label %__nv_exp2f.exit234, !dbg !107 + +1182: ; preds = %__nv_exp2f.exit231 + %1183 = tail call float @llvm.nvvm.ex2.approx.f32(float %1151), !dbg !107 + br label %__nv_exp2f.exit234, !dbg !107 + +__nv_exp2f.exit234: ; preds = %1180, %1182 + %.0.i233 = phi float [ %1181, %1180 ], [ %1183, %1182 ], !dbg !107 + %1184 = insertelement <4 x float> poison, float %.0.i212, i64 0, !dbg !109 + %1185 = insertelement <4 x float> %1184, float %.0.i215, i64 1, !dbg !109 + %1186 = insertelement <4 x float> %1185, float %.0.i230, i64 2, !dbg !109 + %1187 = insertelement <4 x float> %1186, float %.0.i233, i64 3, !dbg !109 + %1188 = select <4 x i1> %1145, <4 x float> %1187, <4 x float> zeroinitializer, !dbg !109 + %1189 = select i1 %1143, float %.0.i218, float 0.000000e+00, !dbg !109 + %1190 = select i1 %1144, float %.0.i221, float 0.000000e+00, !dbg !109 + %1191 = fmul float %972, %1189, !dbg !110 + %1192 = fmul float %973, %1190, !dbg !110 + %1193 = fmul float %974, 0.000000e+00, !dbg !110 + %1194 = fmul float %975, 0.000000e+00, !dbg !110 + %1195 = fmul <4 x float> %1034, %1188, !dbg !110 + %1196 = insertelement <4 x i1> poison, i1 %102, i64 0, !dbg !111 + %1197 = insertelement <4 x i1> %1196, i1 %101, i64 1, !dbg !111 + %1198 = shufflevector <4 x i1> %1197, <4 x i1> poison, <4 x i32> , !dbg !111 + %1199 = and <4 x i1> %1138, %1198, !dbg !111 + %1200 = and i1 %101, %207, !dbg !112 + %1201 = and i1 %101, %209, !dbg !112 + %1202 = and <4 x i1> %104, %1199, !dbg !112 + %1203 = fsub float %1074, %1076, !dbg !113 + %1204 = fsub float %1074, %1077, !dbg !113 + %1205 = fsub float %1075, %1076, !dbg !113 + %1206 = fsub float %1075, %1077, !dbg !113 + %1207 = fsub float %1075, %1078, !dbg !113 + %1208 = fsub float %1075, %1079, !dbg !113 + %1209 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !114 + %.not.i235 = icmp eq i32 %1209, 0, !dbg !114 + br i1 %.not.i235, label %1212, label %1210, !dbg !114 + +1210: ; preds = %__nv_exp2f.exit234 + %1211 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1203), !dbg !114 + br label %__nv_exp2f.exit237, !dbg !114 + +1212: ; preds = %__nv_exp2f.exit234 + %1213 = tail call float @llvm.nvvm.ex2.approx.f32(float %1203), !dbg !114 + br label %__nv_exp2f.exit237, !dbg !114 + +__nv_exp2f.exit237: ; preds = %1210, %1212 + %.0.i236 = phi float [ %1211, %1210 ], [ %1213, %1212 ], !dbg !114 + %1214 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !114 + %.not.i238 = icmp eq i32 %1214, 0, !dbg !114 + br i1 %.not.i238, label %1217, label %1215, !dbg !114 + +1215: ; preds = %__nv_exp2f.exit237 + %1216 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1204), !dbg !114 + br label %__nv_exp2f.exit240, !dbg !114 + +1217: ; preds = %__nv_exp2f.exit237 + %1218 = tail call float @llvm.nvvm.ex2.approx.f32(float %1204), !dbg !114 + br label %__nv_exp2f.exit240, !dbg !114 + +__nv_exp2f.exit240: ; preds = %1215, %1217 + %.0.i239 = phi float [ %1216, %1215 ], [ %1218, %1217 ], !dbg !114 + %1219 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !114 + %.not.i241 = icmp eq i32 %1219, 0, !dbg !114 + br i1 %.not.i241, label %1222, label %1220, !dbg !114 + +1220: ; preds = %__nv_exp2f.exit240 + %1221 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1205), !dbg !114 + br label %__nv_exp2f.exit243, !dbg !114 + +1222: ; preds = %__nv_exp2f.exit240 + %1223 = tail call float @llvm.nvvm.ex2.approx.f32(float %1205), !dbg !114 + br label %__nv_exp2f.exit243, !dbg !114 + +__nv_exp2f.exit243: ; preds = %1220, %1222 + %.0.i242 = phi float [ %1221, %1220 ], [ %1223, %1222 ], !dbg !114 + %1224 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !114 + %.not.i244 = icmp eq i32 %1224, 0, !dbg !114 + br i1 %.not.i244, label %1227, label %1225, !dbg !114 + +1225: ; preds = %__nv_exp2f.exit243 + %1226 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1206), !dbg !114 + br label %__nv_exp2f.exit246, !dbg !114 + +1227: ; preds = %__nv_exp2f.exit243 + %1228 = tail call float @llvm.nvvm.ex2.approx.f32(float %1206), !dbg !114 + br label %__nv_exp2f.exit246, !dbg !114 + +__nv_exp2f.exit246: ; preds = %1225, %1227 + %.0.i245 = phi float [ %1226, %1225 ], [ %1228, %1227 ], !dbg !114 + %1229 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !114 + %1230 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !114 + %1231 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !114 + %.not.i253 = icmp eq i32 %1231, 0, !dbg !114 + br i1 %.not.i253, label %1234, label %1232, !dbg !114 + +1232: ; preds = %__nv_exp2f.exit246 + %1233 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1207), !dbg !114 + br label %__nv_exp2f.exit255, !dbg !114 + +1234: ; preds = %__nv_exp2f.exit246 + %1235 = tail call float @llvm.nvvm.ex2.approx.f32(float %1207), !dbg !114 + br label %__nv_exp2f.exit255, !dbg !114 + +__nv_exp2f.exit255: ; preds = %1232, %1234 + %.0.i254 = phi float [ %1233, %1232 ], [ %1235, %1234 ], !dbg !114 + %1236 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !114 + %.not.i256 = icmp eq i32 %1236, 0, !dbg !114 + br i1 %.not.i256, label %1239, label %1237, !dbg !114 + +1237: ; preds = %__nv_exp2f.exit255 + %1238 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1208), !dbg !114 + br label %__nv_exp2f.exit258, !dbg !114 + +1239: ; preds = %__nv_exp2f.exit255 + %1240 = tail call float @llvm.nvvm.ex2.approx.f32(float %1208), !dbg !114 + br label %__nv_exp2f.exit258, !dbg !114 + +__nv_exp2f.exit258: ; preds = %1237, %1239 + %.0.i257 = phi float [ %1238, %1237 ], [ %1240, %1239 ], !dbg !114 + %1241 = insertelement <4 x float> poison, float %.0.i236, i64 0, !dbg !116 + %1242 = insertelement <4 x float> %1241, float %.0.i239, i64 1, !dbg !116 + %1243 = insertelement <4 x float> %1242, float %.0.i254, i64 2, !dbg !116 + %1244 = insertelement <4 x float> %1243, float %.0.i257, i64 3, !dbg !116 + %1245 = select <4 x i1> %1202, <4 x float> %1244, <4 x float> zeroinitializer, !dbg !116 + %1246 = select i1 %1200, float %.0.i242, float 0.000000e+00, !dbg !116 + %1247 = select i1 %1201, float %.0.i245, float 0.000000e+00, !dbg !116 + %1248 = fmul float %976, %1246, !dbg !117 + %1249 = fmul float %977, %1247, !dbg !117 + %1250 = fmul float %978, 0.000000e+00, !dbg !117 + %1251 = fmul float %979, 0.000000e+00, !dbg !117 + %1252 = fmul <4 x float> %1033, %1245, !dbg !117 + %1253 = insertelement <4 x i1> poison, i1 %112, i64 0, !dbg !118 + %1254 = insertelement <4 x i1> %1253, i1 %111, i64 1, !dbg !118 + %1255 = shufflevector <4 x i1> %1254, <4 x i1> poison, <4 x i32> , !dbg !118 + %1256 = and <4 x i1> %1094, %1255, !dbg !118 + %1257 = and i1 %111, %217, !dbg !119 + %1258 = and i1 %111, %219, !dbg !119 + %1259 = and <4 x i1> %114, %1256, !dbg !119 + %1260 = fsub float %1080, %1082, !dbg !120 + %1261 = fsub float %1080, %1083, !dbg !120 + %1262 = fsub float %1081, %1082, !dbg !120 + %1263 = fsub float %1081, %1083, !dbg !120 + %1264 = fsub float %1081, %1084, !dbg !120 + %1265 = fsub float %1081, %1085, !dbg !120 + %1266 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !121 + %.not.i259 = icmp eq i32 %1266, 0, !dbg !121 + br i1 %.not.i259, label %1269, label %1267, !dbg !121 + +1267: ; preds = %__nv_exp2f.exit258 + %1268 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1260), !dbg !121 + br label %__nv_exp2f.exit261, !dbg !121 + +1269: ; preds = %__nv_exp2f.exit258 + %1270 = tail call float @llvm.nvvm.ex2.approx.f32(float %1260), !dbg !121 + br label %__nv_exp2f.exit261, !dbg !121 + +__nv_exp2f.exit261: ; preds = %1267, %1269 + %.0.i260 = phi float [ %1268, %1267 ], [ %1270, %1269 ], !dbg !121 + %1271 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !121 + %.not.i262 = icmp eq i32 %1271, 0, !dbg !121 + br i1 %.not.i262, label %1274, label %1272, !dbg !121 + +1272: ; preds = %__nv_exp2f.exit261 + %1273 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1261), !dbg !121 + br label %__nv_exp2f.exit264, !dbg !121 + +1274: ; preds = %__nv_exp2f.exit261 + %1275 = tail call float @llvm.nvvm.ex2.approx.f32(float %1261), !dbg !121 + br label %__nv_exp2f.exit264, !dbg !121 + +__nv_exp2f.exit264: ; preds = %1272, %1274 + %.0.i263 = phi float [ %1273, %1272 ], [ %1275, %1274 ], !dbg !121 + %1276 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !121 + %.not.i265 = icmp eq i32 %1276, 0, !dbg !121 + br i1 %.not.i265, label %1279, label %1277, !dbg !121 + +1277: ; preds = %__nv_exp2f.exit264 + %1278 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1262), !dbg !121 + br label %__nv_exp2f.exit267, !dbg !121 + +1279: ; preds = %__nv_exp2f.exit264 + %1280 = tail call float @llvm.nvvm.ex2.approx.f32(float %1262), !dbg !121 + br label %__nv_exp2f.exit267, !dbg !121 + +__nv_exp2f.exit267: ; preds = %1277, %1279 + %.0.i266 = phi float [ %1278, %1277 ], [ %1280, %1279 ], !dbg !121 + %1281 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !121 + %.not.i268 = icmp eq i32 %1281, 0, !dbg !121 + br i1 %.not.i268, label %1284, label %1282, !dbg !121 + +1282: ; preds = %__nv_exp2f.exit267 + %1283 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1263), !dbg !121 + br label %__nv_exp2f.exit270, !dbg !121 + +1284: ; preds = %__nv_exp2f.exit267 + %1285 = tail call float @llvm.nvvm.ex2.approx.f32(float %1263), !dbg !121 + br label %__nv_exp2f.exit270, !dbg !121 + +__nv_exp2f.exit270: ; preds = %1282, %1284 + %.0.i269 = phi float [ %1283, %1282 ], [ %1285, %1284 ], !dbg !121 + %1286 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !121 + %1287 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !121 + %1288 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !121 + %.not.i277 = icmp eq i32 %1288, 0, !dbg !121 + br i1 %.not.i277, label %1291, label %1289, !dbg !121 + +1289: ; preds = %__nv_exp2f.exit270 + %1290 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1264), !dbg !121 + br label %__nv_exp2f.exit279, !dbg !121 + +1291: ; preds = %__nv_exp2f.exit270 + %1292 = tail call float @llvm.nvvm.ex2.approx.f32(float %1264), !dbg !121 + br label %__nv_exp2f.exit279, !dbg !121 + +__nv_exp2f.exit279: ; preds = %1289, %1291 + %.0.i278 = phi float [ %1290, %1289 ], [ %1292, %1291 ], !dbg !121 + %1293 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !121 + %.not.i280 = icmp eq i32 %1293, 0, !dbg !121 + br i1 %.not.i280, label %1296, label %1294, !dbg !121 + +1294: ; preds = %__nv_exp2f.exit279 + %1295 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1265), !dbg !121 + br label %__nv_exp2f.exit282, !dbg !121 + +1296: ; preds = %__nv_exp2f.exit279 + %1297 = tail call float @llvm.nvvm.ex2.approx.f32(float %1265), !dbg !121 + br label %__nv_exp2f.exit282, !dbg !121 + +__nv_exp2f.exit282: ; preds = %1294, %1296 + %.0.i281 = phi float [ %1295, %1294 ], [ %1297, %1296 ], !dbg !121 + %1298 = insertelement <4 x float> poison, float %.0.i260, i64 0, !dbg !123 + %1299 = insertelement <4 x float> %1298, float %.0.i263, i64 1, !dbg !123 + %1300 = insertelement <4 x float> %1299, float %.0.i278, i64 2, !dbg !123 + %1301 = insertelement <4 x float> %1300, float %.0.i281, i64 3, !dbg !123 + %1302 = select <4 x i1> %1259, <4 x float> %1301, <4 x float> zeroinitializer, !dbg !123 + %1303 = select i1 %1257, float %.0.i266, float 0.000000e+00, !dbg !123 + %1304 = select i1 %1258, float %.0.i269, float 0.000000e+00, !dbg !123 + %1305 = fmul float %980, %1303, !dbg !124 + %1306 = fmul float %981, %1304, !dbg !124 + %1307 = fmul float %982, 0.000000e+00, !dbg !124 + %1308 = fmul float %983, 0.000000e+00, !dbg !124 + %1309 = fmul <4 x float> %1032, %1302, !dbg !124 + %1310 = and i1 %91, %59, !dbg !125 + %1311 = and i1 %91, %60, !dbg !125 + %1312 = and i1 %90, %59, !dbg !125 + %1313 = and i1 %90, %60, !dbg !125 + %1314 = and i1 %91, %61, !dbg !125 + %1315 = and i1 %91, %62, !dbg !125 + %1316 = and i1 %90, %61, !dbg !125 + %1317 = and i1 %90, %62, !dbg !125 + %1318 = fsub float %1068, %1064, !dbg !126 + %1319 = fsub float %1068, %1065, !dbg !126 + %1320 = fsub float %1069, %1064, !dbg !126 + %1321 = fsub float %1069, %1065, !dbg !126 + %1322 = fsub float %1068, %1066, !dbg !126 + %1323 = fsub float %1068, %1067, !dbg !126 + %1324 = fsub float %1069, %1066, !dbg !126 + %1325 = fsub float %1069, %1067, !dbg !126 + %1326 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !127 + %.not.i283 = icmp eq i32 %1326, 0, !dbg !127 + br i1 %.not.i283, label %1329, label %1327, !dbg !127 + +1327: ; preds = %__nv_exp2f.exit282 + %1328 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1318), !dbg !127 + br label %__nv_exp2f.exit285, !dbg !127 + +1329: ; preds = %__nv_exp2f.exit282 + %1330 = tail call float @llvm.nvvm.ex2.approx.f32(float %1318), !dbg !127 + br label %__nv_exp2f.exit285, !dbg !127 + +__nv_exp2f.exit285: ; preds = %1327, %1329 + %.0.i284 = phi float [ %1328, %1327 ], [ %1330, %1329 ], !dbg !127 + %1331 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !127 + %.not.i286 = icmp eq i32 %1331, 0, !dbg !127 + br i1 %.not.i286, label %1334, label %1332, !dbg !127 + +1332: ; preds = %__nv_exp2f.exit285 + %1333 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1319), !dbg !127 + br label %__nv_exp2f.exit288, !dbg !127 + +1334: ; preds = %__nv_exp2f.exit285 + %1335 = tail call float @llvm.nvvm.ex2.approx.f32(float %1319), !dbg !127 + br label %__nv_exp2f.exit288, !dbg !127 + +__nv_exp2f.exit288: ; preds = %1332, %1334 + %.0.i287 = phi float [ %1333, %1332 ], [ %1335, %1334 ], !dbg !127 + %1336 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !127 + %.not.i289 = icmp eq i32 %1336, 0, !dbg !127 + br i1 %.not.i289, label %1339, label %1337, !dbg !127 + +1337: ; preds = %__nv_exp2f.exit288 + %1338 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1320), !dbg !127 + br label %__nv_exp2f.exit291, !dbg !127 + +1339: ; preds = %__nv_exp2f.exit288 + %1340 = tail call float @llvm.nvvm.ex2.approx.f32(float %1320), !dbg !127 + br label %__nv_exp2f.exit291, !dbg !127 + +__nv_exp2f.exit291: ; preds = %1337, %1339 + %.0.i290 = phi float [ %1338, %1337 ], [ %1340, %1339 ], !dbg !127 + %1341 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !127 + %.not.i292 = icmp eq i32 %1341, 0, !dbg !127 + br i1 %.not.i292, label %1344, label %1342, !dbg !127 + +1342: ; preds = %__nv_exp2f.exit291 + %1343 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1321), !dbg !127 + br label %__nv_exp2f.exit294, !dbg !127 + +1344: ; preds = %__nv_exp2f.exit291 + %1345 = tail call float @llvm.nvvm.ex2.approx.f32(float %1321), !dbg !127 + br label %__nv_exp2f.exit294, !dbg !127 + +__nv_exp2f.exit294: ; preds = %1342, %1344 + %.0.i293 = phi float [ %1343, %1342 ], [ %1345, %1344 ], !dbg !127 + %1346 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !127 + %.not.i295 = icmp eq i32 %1346, 0, !dbg !127 + br i1 %.not.i295, label %1349, label %1347, !dbg !127 + +1347: ; preds = %__nv_exp2f.exit294 + %1348 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1322), !dbg !127 + br label %__nv_exp2f.exit297, !dbg !127 + +1349: ; preds = %__nv_exp2f.exit294 + %1350 = tail call float @llvm.nvvm.ex2.approx.f32(float %1322), !dbg !127 + br label %__nv_exp2f.exit297, !dbg !127 + +__nv_exp2f.exit297: ; preds = %1347, %1349 + %.0.i296 = phi float [ %1348, %1347 ], [ %1350, %1349 ], !dbg !127 + %1351 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !127 + %.not.i298 = icmp eq i32 %1351, 0, !dbg !127 + br i1 %.not.i298, label %1354, label %1352, !dbg !127 + +1352: ; preds = %__nv_exp2f.exit297 + %1353 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1323), !dbg !127 + br label %__nv_exp2f.exit300, !dbg !127 + +1354: ; preds = %__nv_exp2f.exit297 + %1355 = tail call float @llvm.nvvm.ex2.approx.f32(float %1323), !dbg !127 + br label %__nv_exp2f.exit300, !dbg !127 + +__nv_exp2f.exit300: ; preds = %1352, %1354 + %.0.i299 = phi float [ %1353, %1352 ], [ %1355, %1354 ], !dbg !127 + %1356 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !127 + %.not.i301 = icmp eq i32 %1356, 0, !dbg !127 + br i1 %.not.i301, label %1359, label %1357, !dbg !127 + +1357: ; preds = %__nv_exp2f.exit300 + %1358 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1324), !dbg !127 + br label %__nv_exp2f.exit303, !dbg !127 + +1359: ; preds = %__nv_exp2f.exit300 + %1360 = tail call float @llvm.nvvm.ex2.approx.f32(float %1324), !dbg !127 + br label %__nv_exp2f.exit303, !dbg !127 + +__nv_exp2f.exit303: ; preds = %1357, %1359 + %.0.i302 = phi float [ %1358, %1357 ], [ %1360, %1359 ], !dbg !127 + %1361 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !127 + %.not.i304 = icmp eq i32 %1361, 0, !dbg !127 + br i1 %.not.i304, label %1364, label %1362, !dbg !127 + +1362: ; preds = %__nv_exp2f.exit303 + %1363 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1325), !dbg !127 + br label %__nv_exp2f.exit306, !dbg !127 + +1364: ; preds = %__nv_exp2f.exit303 + %1365 = tail call float @llvm.nvvm.ex2.approx.f32(float %1325), !dbg !127 + br label %__nv_exp2f.exit306, !dbg !127 + +__nv_exp2f.exit306: ; preds = %1362, %1364 + %.0.i305 = phi float [ %1363, %1362 ], [ %1365, %1364 ], !dbg !127 + %1366 = select i1 %1310, float %.0.i284, float 0.000000e+00, !dbg !129 + %1367 = select i1 %1311, float %.0.i287, float 0.000000e+00, !dbg !129 + %1368 = select i1 %1312, float %.0.i290, float 0.000000e+00, !dbg !129 + %1369 = select i1 %1313, float %.0.i293, float 0.000000e+00, !dbg !129 + %1370 = select i1 %1314, float %.0.i296, float 0.000000e+00, !dbg !129 + %1371 = select i1 %1315, float %.0.i299, float 0.000000e+00, !dbg !129 + %1372 = select i1 %1316, float %.0.i302, float 0.000000e+00, !dbg !129 + %1373 = select i1 %1317, float %.0.i305, float 0.000000e+00, !dbg !129 + %1374 = fmul float %984, %1366, !dbg !130 + %1375 = fmul float %985, %1367, !dbg !130 + %1376 = fmul float %986, %1368, !dbg !130 + %1377 = fmul float %987, %1369, !dbg !130 + %1378 = fmul float %988, %1370, !dbg !130 + %1379 = fmul float %989, %1371, !dbg !130 + %1380 = fmul float %990, %1372, !dbg !130 + %1381 = fmul float %991, %1373, !dbg !130 + %1382 = and i1 %102, %59, !dbg !131 + %1383 = and i1 %102, %60, !dbg !131 + %1384 = and i1 %101, %59, !dbg !131 + %1385 = and i1 %101, %60, !dbg !131 + %1386 = and i1 %102, %61, !dbg !131 + %1387 = and i1 %102, %62, !dbg !131 + %1388 = and i1 %101, %61, !dbg !131 + %1389 = and i1 %101, %62, !dbg !131 + %1390 = fsub float %1074, %1064, !dbg !132 + %1391 = fsub float %1074, %1065, !dbg !132 + %1392 = fsub float %1075, %1064, !dbg !132 + %1393 = fsub float %1075, %1065, !dbg !132 + %1394 = fsub float %1074, %1066, !dbg !132 + %1395 = fsub float %1074, %1067, !dbg !132 + %1396 = fsub float %1075, %1066, !dbg !132 + %1397 = fsub float %1075, %1067, !dbg !132 + %1398 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !133 + %.not.i307 = icmp eq i32 %1398, 0, !dbg !133 + br i1 %.not.i307, label %1401, label %1399, !dbg !133 + +1399: ; preds = %__nv_exp2f.exit306 + %1400 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1390), !dbg !133 + br label %__nv_exp2f.exit309, !dbg !133 + +1401: ; preds = %__nv_exp2f.exit306 + %1402 = tail call float @llvm.nvvm.ex2.approx.f32(float %1390), !dbg !133 + br label %__nv_exp2f.exit309, !dbg !133 + +__nv_exp2f.exit309: ; preds = %1399, %1401 + %.0.i308 = phi float [ %1400, %1399 ], [ %1402, %1401 ], !dbg !133 + %1403 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !133 + %.not.i310 = icmp eq i32 %1403, 0, !dbg !133 + br i1 %.not.i310, label %1406, label %1404, !dbg !133 + +1404: ; preds = %__nv_exp2f.exit309 + %1405 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1391), !dbg !133 + br label %__nv_exp2f.exit312, !dbg !133 + +1406: ; preds = %__nv_exp2f.exit309 + %1407 = tail call float @llvm.nvvm.ex2.approx.f32(float %1391), !dbg !133 + br label %__nv_exp2f.exit312, !dbg !133 + +__nv_exp2f.exit312: ; preds = %1404, %1406 + %.0.i311 = phi float [ %1405, %1404 ], [ %1407, %1406 ], !dbg !133 + %1408 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !133 + %.not.i313 = icmp eq i32 %1408, 0, !dbg !133 + br i1 %.not.i313, label %1411, label %1409, !dbg !133 + +1409: ; preds = %__nv_exp2f.exit312 + %1410 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1392), !dbg !133 + br label %__nv_exp2f.exit315, !dbg !133 + +1411: ; preds = %__nv_exp2f.exit312 + %1412 = tail call float @llvm.nvvm.ex2.approx.f32(float %1392), !dbg !133 + br label %__nv_exp2f.exit315, !dbg !133 + +__nv_exp2f.exit315: ; preds = %1409, %1411 + %.0.i314 = phi float [ %1410, %1409 ], [ %1412, %1411 ], !dbg !133 + %1413 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !133 + %.not.i316 = icmp eq i32 %1413, 0, !dbg !133 + br i1 %.not.i316, label %1416, label %1414, !dbg !133 + +1414: ; preds = %__nv_exp2f.exit315 + %1415 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1393), !dbg !133 + br label %__nv_exp2f.exit318, !dbg !133 + +1416: ; preds = %__nv_exp2f.exit315 + %1417 = tail call float @llvm.nvvm.ex2.approx.f32(float %1393), !dbg !133 + br label %__nv_exp2f.exit318, !dbg !133 + +__nv_exp2f.exit318: ; preds = %1414, %1416 + %.0.i317 = phi float [ %1415, %1414 ], [ %1417, %1416 ], !dbg !133 + %1418 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !133 + %.not.i319 = icmp eq i32 %1418, 0, !dbg !133 + br i1 %.not.i319, label %1421, label %1419, !dbg !133 + +1419: ; preds = %__nv_exp2f.exit318 + %1420 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1394), !dbg !133 + br label %__nv_exp2f.exit321, !dbg !133 + +1421: ; preds = %__nv_exp2f.exit318 + %1422 = tail call float @llvm.nvvm.ex2.approx.f32(float %1394), !dbg !133 + br label %__nv_exp2f.exit321, !dbg !133 + +__nv_exp2f.exit321: ; preds = %1419, %1421 + %.0.i320 = phi float [ %1420, %1419 ], [ %1422, %1421 ], !dbg !133 + %1423 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !133 + %.not.i322 = icmp eq i32 %1423, 0, !dbg !133 + br i1 %.not.i322, label %1426, label %1424, !dbg !133 + +1424: ; preds = %__nv_exp2f.exit321 + %1425 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1395), !dbg !133 + br label %__nv_exp2f.exit324, !dbg !133 + +1426: ; preds = %__nv_exp2f.exit321 + %1427 = tail call float @llvm.nvvm.ex2.approx.f32(float %1395), !dbg !133 + br label %__nv_exp2f.exit324, !dbg !133 + +__nv_exp2f.exit324: ; preds = %1424, %1426 + %.0.i323 = phi float [ %1425, %1424 ], [ %1427, %1426 ], !dbg !133 + %1428 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !133 + %.not.i325 = icmp eq i32 %1428, 0, !dbg !133 + br i1 %.not.i325, label %1431, label %1429, !dbg !133 + +1429: ; preds = %__nv_exp2f.exit324 + %1430 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1396), !dbg !133 + br label %__nv_exp2f.exit327, !dbg !133 + +1431: ; preds = %__nv_exp2f.exit324 + %1432 = tail call float @llvm.nvvm.ex2.approx.f32(float %1396), !dbg !133 + br label %__nv_exp2f.exit327, !dbg !133 + +__nv_exp2f.exit327: ; preds = %1429, %1431 + %.0.i326 = phi float [ %1430, %1429 ], [ %1432, %1431 ], !dbg !133 + %1433 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !133 + %.not.i328 = icmp eq i32 %1433, 0, !dbg !133 + br i1 %.not.i328, label %1436, label %1434, !dbg !133 + +1434: ; preds = %__nv_exp2f.exit327 + %1435 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1397), !dbg !133 + br label %__nv_exp2f.exit330, !dbg !133 + +1436: ; preds = %__nv_exp2f.exit327 + %1437 = tail call float @llvm.nvvm.ex2.approx.f32(float %1397), !dbg !133 + br label %__nv_exp2f.exit330, !dbg !133 + +__nv_exp2f.exit330: ; preds = %1434, %1436 + %.0.i329 = phi float [ %1435, %1434 ], [ %1437, %1436 ], !dbg !133 + %1438 = select i1 %1382, float %.0.i308, float 0.000000e+00, !dbg !135 + %1439 = select i1 %1383, float %.0.i311, float 0.000000e+00, !dbg !135 + %1440 = select i1 %1384, float %.0.i314, float 0.000000e+00, !dbg !135 + %1441 = select i1 %1385, float %.0.i317, float 0.000000e+00, !dbg !135 + %1442 = select i1 %1386, float %.0.i320, float 0.000000e+00, !dbg !135 + %1443 = select i1 %1387, float %.0.i323, float 0.000000e+00, !dbg !135 + %1444 = select i1 %1388, float %.0.i326, float 0.000000e+00, !dbg !135 + %1445 = select i1 %1389, float %.0.i329, float 0.000000e+00, !dbg !135 + %1446 = fmul float %992, %1438, !dbg !136 + %1447 = fmul float %993, %1439, !dbg !136 + %1448 = fmul float %994, %1440, !dbg !136 + %1449 = fmul float %995, %1441, !dbg !136 + %1450 = fmul float %996, %1442, !dbg !136 + %1451 = fmul float %997, %1443, !dbg !136 + %1452 = fmul float %998, %1444, !dbg !136 + %1453 = fmul float %999, %1445, !dbg !136 + %1454 = and i1 %102, %197, !dbg !137 + %1455 = and i1 %102, %199, !dbg !137 + %1456 = and i1 %101, %197, !dbg !137 + %1457 = and i1 %101, %199, !dbg !137 + %1458 = and i1 %102, %201, !dbg !137 + %1459 = and i1 %102, %203, !dbg !137 + %1460 = and i1 %101, %201, !dbg !137 + %1461 = and i1 %101, %203, !dbg !137 + %1462 = fsub float %1074, %1070, !dbg !138 + %1463 = fsub float %1074, %1071, !dbg !138 + %1464 = fsub float %1075, %1070, !dbg !138 + %1465 = fsub float %1075, %1071, !dbg !138 + %1466 = fsub float %1074, %1072, !dbg !138 + %1467 = fsub float %1074, %1073, !dbg !138 + %1468 = fsub float %1075, %1072, !dbg !138 + %1469 = fsub float %1075, %1073, !dbg !138 + %1470 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !139 + %.not.i331 = icmp eq i32 %1470, 0, !dbg !139 + br i1 %.not.i331, label %1473, label %1471, !dbg !139 + +1471: ; preds = %__nv_exp2f.exit330 + %1472 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1462), !dbg !139 + br label %__nv_exp2f.exit333, !dbg !139 + +1473: ; preds = %__nv_exp2f.exit330 + %1474 = tail call float @llvm.nvvm.ex2.approx.f32(float %1462), !dbg !139 + br label %__nv_exp2f.exit333, !dbg !139 + +__nv_exp2f.exit333: ; preds = %1471, %1473 + %.0.i332 = phi float [ %1472, %1471 ], [ %1474, %1473 ], !dbg !139 + %1475 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !139 + %.not.i334 = icmp eq i32 %1475, 0, !dbg !139 + br i1 %.not.i334, label %1478, label %1476, !dbg !139 + +1476: ; preds = %__nv_exp2f.exit333 + %1477 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1463), !dbg !139 + br label %__nv_exp2f.exit336, !dbg !139 + +1478: ; preds = %__nv_exp2f.exit333 + %1479 = tail call float @llvm.nvvm.ex2.approx.f32(float %1463), !dbg !139 + br label %__nv_exp2f.exit336, !dbg !139 + +__nv_exp2f.exit336: ; preds = %1476, %1478 + %.0.i335 = phi float [ %1477, %1476 ], [ %1479, %1478 ], !dbg !139 + %1480 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !139 + %.not.i337 = icmp eq i32 %1480, 0, !dbg !139 + br i1 %.not.i337, label %1483, label %1481, !dbg !139 + +1481: ; preds = %__nv_exp2f.exit336 + %1482 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1464), !dbg !139 + br label %__nv_exp2f.exit339, !dbg !139 + +1483: ; preds = %__nv_exp2f.exit336 + %1484 = tail call float @llvm.nvvm.ex2.approx.f32(float %1464), !dbg !139 + br label %__nv_exp2f.exit339, !dbg !139 + +__nv_exp2f.exit339: ; preds = %1481, %1483 + %.0.i338 = phi float [ %1482, %1481 ], [ %1484, %1483 ], !dbg !139 + %1485 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !139 + %.not.i340 = icmp eq i32 %1485, 0, !dbg !139 + br i1 %.not.i340, label %1488, label %1486, !dbg !139 + +1486: ; preds = %__nv_exp2f.exit339 + %1487 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1465), !dbg !139 + br label %__nv_exp2f.exit342, !dbg !139 + +1488: ; preds = %__nv_exp2f.exit339 + %1489 = tail call float @llvm.nvvm.ex2.approx.f32(float %1465), !dbg !139 + br label %__nv_exp2f.exit342, !dbg !139 + +__nv_exp2f.exit342: ; preds = %1486, %1488 + %.0.i341 = phi float [ %1487, %1486 ], [ %1489, %1488 ], !dbg !139 + %1490 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !139 + %.not.i343 = icmp eq i32 %1490, 0, !dbg !139 + br i1 %.not.i343, label %1493, label %1491, !dbg !139 + +1491: ; preds = %__nv_exp2f.exit342 + %1492 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1466), !dbg !139 + br label %__nv_exp2f.exit345, !dbg !139 + +1493: ; preds = %__nv_exp2f.exit342 + %1494 = tail call float @llvm.nvvm.ex2.approx.f32(float %1466), !dbg !139 + br label %__nv_exp2f.exit345, !dbg !139 + +__nv_exp2f.exit345: ; preds = %1491, %1493 + %.0.i344 = phi float [ %1492, %1491 ], [ %1494, %1493 ], !dbg !139 + %1495 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !139 + %.not.i346 = icmp eq i32 %1495, 0, !dbg !139 + br i1 %.not.i346, label %1498, label %1496, !dbg !139 + +1496: ; preds = %__nv_exp2f.exit345 + %1497 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1467), !dbg !139 + br label %__nv_exp2f.exit348, !dbg !139 + +1498: ; preds = %__nv_exp2f.exit345 + %1499 = tail call float @llvm.nvvm.ex2.approx.f32(float %1467), !dbg !139 + br label %__nv_exp2f.exit348, !dbg !139 + +__nv_exp2f.exit348: ; preds = %1496, %1498 + %.0.i347 = phi float [ %1497, %1496 ], [ %1499, %1498 ], !dbg !139 + %1500 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !139 + %.not.i349 = icmp eq i32 %1500, 0, !dbg !139 + br i1 %.not.i349, label %1503, label %1501, !dbg !139 + +1501: ; preds = %__nv_exp2f.exit348 + %1502 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1468), !dbg !139 + br label %__nv_exp2f.exit351, !dbg !139 + +1503: ; preds = %__nv_exp2f.exit348 + %1504 = tail call float @llvm.nvvm.ex2.approx.f32(float %1468), !dbg !139 + br label %__nv_exp2f.exit351, !dbg !139 + +__nv_exp2f.exit351: ; preds = %1501, %1503 + %.0.i350 = phi float [ %1502, %1501 ], [ %1504, %1503 ], !dbg !139 + %1505 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !139 + %.not.i352 = icmp eq i32 %1505, 0, !dbg !139 + br i1 %.not.i352, label %1508, label %1506, !dbg !139 + +1506: ; preds = %__nv_exp2f.exit351 + %1507 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1469), !dbg !139 + br label %__nv_exp2f.exit354, !dbg !139 + +1508: ; preds = %__nv_exp2f.exit351 + %1509 = tail call float @llvm.nvvm.ex2.approx.f32(float %1469), !dbg !139 + br label %__nv_exp2f.exit354, !dbg !139 + +__nv_exp2f.exit354: ; preds = %1506, %1508 + %.0.i353 = phi float [ %1507, %1506 ], [ %1509, %1508 ], !dbg !139 + %1510 = select i1 %1454, float %.0.i332, float 0.000000e+00, !dbg !141 + %1511 = select i1 %1455, float %.0.i335, float 0.000000e+00, !dbg !141 + %1512 = select i1 %1456, float %.0.i338, float 0.000000e+00, !dbg !141 + %1513 = select i1 %1457, float %.0.i341, float 0.000000e+00, !dbg !141 + %1514 = select i1 %1458, float %.0.i344, float 0.000000e+00, !dbg !141 + %1515 = select i1 %1459, float %.0.i347, float 0.000000e+00, !dbg !141 + %1516 = select i1 %1460, float %.0.i350, float 0.000000e+00, !dbg !141 + %1517 = select i1 %1461, float %.0.i353, float 0.000000e+00, !dbg !141 + %1518 = fmul float %1000, %1510, !dbg !142 + %1519 = fmul float %1001, %1511, !dbg !142 + %1520 = fmul float %1002, %1512, !dbg !142 + %1521 = fmul float %1003, %1513, !dbg !142 + %1522 = fmul float %1004, %1514, !dbg !142 + %1523 = fmul float %1005, %1515, !dbg !142 + %1524 = fmul float %1006, %1516, !dbg !142 + %1525 = fmul float %1007, %1517, !dbg !142 + %1526 = and i1 %112, %59, !dbg !143 + %1527 = and i1 %112, %60, !dbg !143 + %1528 = and i1 %111, %59, !dbg !143 + %1529 = and i1 %111, %60, !dbg !143 + %1530 = and i1 %112, %61, !dbg !143 + %1531 = and i1 %112, %62, !dbg !143 + %1532 = and i1 %111, %61, !dbg !143 + %1533 = and i1 %111, %62, !dbg !143 + %1534 = fsub float %1080, %1064, !dbg !144 + %1535 = fsub float %1080, %1065, !dbg !144 + %1536 = fsub float %1081, %1064, !dbg !144 + %1537 = fsub float %1081, %1065, !dbg !144 + %1538 = fsub float %1080, %1066, !dbg !144 + %1539 = fsub float %1080, %1067, !dbg !144 + %1540 = fsub float %1081, %1066, !dbg !144 + %1541 = fsub float %1081, %1067, !dbg !144 + %1542 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !145 + %.not.i355 = icmp eq i32 %1542, 0, !dbg !145 + br i1 %.not.i355, label %1545, label %1543, !dbg !145 + +1543: ; preds = %__nv_exp2f.exit354 + %1544 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1534), !dbg !145 + br label %__nv_exp2f.exit357, !dbg !145 + +1545: ; preds = %__nv_exp2f.exit354 + %1546 = tail call float @llvm.nvvm.ex2.approx.f32(float %1534), !dbg !145 + br label %__nv_exp2f.exit357, !dbg !145 + +__nv_exp2f.exit357: ; preds = %1543, %1545 + %.0.i356 = phi float [ %1544, %1543 ], [ %1546, %1545 ], !dbg !145 + %1547 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !145 + %.not.i358 = icmp eq i32 %1547, 0, !dbg !145 + br i1 %.not.i358, label %1550, label %1548, !dbg !145 + +1548: ; preds = %__nv_exp2f.exit357 + %1549 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1535), !dbg !145 + br label %__nv_exp2f.exit360, !dbg !145 + +1550: ; preds = %__nv_exp2f.exit357 + %1551 = tail call float @llvm.nvvm.ex2.approx.f32(float %1535), !dbg !145 + br label %__nv_exp2f.exit360, !dbg !145 + +__nv_exp2f.exit360: ; preds = %1548, %1550 + %.0.i359 = phi float [ %1549, %1548 ], [ %1551, %1550 ], !dbg !145 + %1552 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !145 + %.not.i361 = icmp eq i32 %1552, 0, !dbg !145 + br i1 %.not.i361, label %1555, label %1553, !dbg !145 + +1553: ; preds = %__nv_exp2f.exit360 + %1554 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1536), !dbg !145 + br label %__nv_exp2f.exit363, !dbg !145 + +1555: ; preds = %__nv_exp2f.exit360 + %1556 = tail call float @llvm.nvvm.ex2.approx.f32(float %1536), !dbg !145 + br label %__nv_exp2f.exit363, !dbg !145 + +__nv_exp2f.exit363: ; preds = %1553, %1555 + %.0.i362 = phi float [ %1554, %1553 ], [ %1556, %1555 ], !dbg !145 + %1557 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !145 + %.not.i364 = icmp eq i32 %1557, 0, !dbg !145 + br i1 %.not.i364, label %1560, label %1558, !dbg !145 + +1558: ; preds = %__nv_exp2f.exit363 + %1559 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1537), !dbg !145 + br label %__nv_exp2f.exit366, !dbg !145 + +1560: ; preds = %__nv_exp2f.exit363 + %1561 = tail call float @llvm.nvvm.ex2.approx.f32(float %1537), !dbg !145 + br label %__nv_exp2f.exit366, !dbg !145 + +__nv_exp2f.exit366: ; preds = %1558, %1560 + %.0.i365 = phi float [ %1559, %1558 ], [ %1561, %1560 ], !dbg !145 + %1562 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !145 + %.not.i367 = icmp eq i32 %1562, 0, !dbg !145 + br i1 %.not.i367, label %1565, label %1563, !dbg !145 + +1563: ; preds = %__nv_exp2f.exit366 + %1564 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1538), !dbg !145 + br label %__nv_exp2f.exit369, !dbg !145 + +1565: ; preds = %__nv_exp2f.exit366 + %1566 = tail call float @llvm.nvvm.ex2.approx.f32(float %1538), !dbg !145 + br label %__nv_exp2f.exit369, !dbg !145 + +__nv_exp2f.exit369: ; preds = %1563, %1565 + %.0.i368 = phi float [ %1564, %1563 ], [ %1566, %1565 ], !dbg !145 + %1567 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !145 + %.not.i370 = icmp eq i32 %1567, 0, !dbg !145 + br i1 %.not.i370, label %1570, label %1568, !dbg !145 + +1568: ; preds = %__nv_exp2f.exit369 + %1569 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1539), !dbg !145 + br label %__nv_exp2f.exit372, !dbg !145 + +1570: ; preds = %__nv_exp2f.exit369 + %1571 = tail call float @llvm.nvvm.ex2.approx.f32(float %1539), !dbg !145 + br label %__nv_exp2f.exit372, !dbg !145 + +__nv_exp2f.exit372: ; preds = %1568, %1570 + %.0.i371 = phi float [ %1569, %1568 ], [ %1571, %1570 ], !dbg !145 + %1572 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !145 + %.not.i373 = icmp eq i32 %1572, 0, !dbg !145 + br i1 %.not.i373, label %1575, label %1573, !dbg !145 + +1573: ; preds = %__nv_exp2f.exit372 + %1574 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1540), !dbg !145 + br label %__nv_exp2f.exit375, !dbg !145 + +1575: ; preds = %__nv_exp2f.exit372 + %1576 = tail call float @llvm.nvvm.ex2.approx.f32(float %1540), !dbg !145 + br label %__nv_exp2f.exit375, !dbg !145 + +__nv_exp2f.exit375: ; preds = %1573, %1575 + %.0.i374 = phi float [ %1574, %1573 ], [ %1576, %1575 ], !dbg !145 + %1577 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !145 + %.not.i376 = icmp eq i32 %1577, 0, !dbg !145 + br i1 %.not.i376, label %1580, label %1578, !dbg !145 + +1578: ; preds = %__nv_exp2f.exit375 + %1579 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1541), !dbg !145 + br label %__nv_exp2f.exit378, !dbg !145 + +1580: ; preds = %__nv_exp2f.exit375 + %1581 = tail call float @llvm.nvvm.ex2.approx.f32(float %1541), !dbg !145 + br label %__nv_exp2f.exit378, !dbg !145 + +__nv_exp2f.exit378: ; preds = %1578, %1580 + %.0.i377 = phi float [ %1579, %1578 ], [ %1581, %1580 ], !dbg !145 + %1582 = select i1 %1526, float %.0.i356, float 0.000000e+00, !dbg !147 + %1583 = select i1 %1527, float %.0.i359, float 0.000000e+00, !dbg !147 + %1584 = select i1 %1528, float %.0.i362, float 0.000000e+00, !dbg !147 + %1585 = select i1 %1529, float %.0.i365, float 0.000000e+00, !dbg !147 + %1586 = select i1 %1530, float %.0.i368, float 0.000000e+00, !dbg !147 + %1587 = select i1 %1531, float %.0.i371, float 0.000000e+00, !dbg !147 + %1588 = select i1 %1532, float %.0.i374, float 0.000000e+00, !dbg !147 + %1589 = select i1 %1533, float %.0.i377, float 0.000000e+00, !dbg !147 + %1590 = fmul float %1008, %1582, !dbg !148 + %1591 = fmul float %1009, %1583, !dbg !148 + %1592 = fmul float %1010, %1584, !dbg !148 + %1593 = fmul float %1011, %1585, !dbg !148 + %1594 = fmul float %1012, %1586, !dbg !148 + %1595 = fmul float %1013, %1587, !dbg !148 + %1596 = fmul float %1014, %1588, !dbg !148 + %1597 = fmul float %1015, %1589, !dbg !148 + %1598 = and i1 %112, %197, !dbg !149 + %1599 = and i1 %112, %199, !dbg !149 + %1600 = and i1 %111, %197, !dbg !149 + %1601 = and i1 %111, %199, !dbg !149 + %1602 = and i1 %112, %201, !dbg !149 + %1603 = and i1 %112, %203, !dbg !149 + %1604 = and i1 %111, %201, !dbg !149 + %1605 = and i1 %111, %203, !dbg !149 + %1606 = fsub float %1080, %1070, !dbg !150 + %1607 = fsub float %1080, %1071, !dbg !150 + %1608 = fsub float %1081, %1070, !dbg !150 + %1609 = fsub float %1081, %1071, !dbg !150 + %1610 = fsub float %1080, %1072, !dbg !150 + %1611 = fsub float %1080, %1073, !dbg !150 + %1612 = fsub float %1081, %1072, !dbg !150 + %1613 = fsub float %1081, %1073, !dbg !150 + %1614 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !151 + %.not.i379 = icmp eq i32 %1614, 0, !dbg !151 + br i1 %.not.i379, label %1617, label %1615, !dbg !151 + +1615: ; preds = %__nv_exp2f.exit378 + %1616 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1606), !dbg !151 + br label %__nv_exp2f.exit381, !dbg !151 + +1617: ; preds = %__nv_exp2f.exit378 + %1618 = tail call float @llvm.nvvm.ex2.approx.f32(float %1606), !dbg !151 + br label %__nv_exp2f.exit381, !dbg !151 + +__nv_exp2f.exit381: ; preds = %1615, %1617 + %.0.i380 = phi float [ %1616, %1615 ], [ %1618, %1617 ], !dbg !151 + %1619 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !151 + %.not.i382 = icmp eq i32 %1619, 0, !dbg !151 + br i1 %.not.i382, label %1622, label %1620, !dbg !151 + +1620: ; preds = %__nv_exp2f.exit381 + %1621 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1607), !dbg !151 + br label %__nv_exp2f.exit384, !dbg !151 + +1622: ; preds = %__nv_exp2f.exit381 + %1623 = tail call float @llvm.nvvm.ex2.approx.f32(float %1607), !dbg !151 + br label %__nv_exp2f.exit384, !dbg !151 + +__nv_exp2f.exit384: ; preds = %1620, %1622 + %.0.i383 = phi float [ %1621, %1620 ], [ %1623, %1622 ], !dbg !151 + %1624 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !151 + %.not.i385 = icmp eq i32 %1624, 0, !dbg !151 + br i1 %.not.i385, label %1627, label %1625, !dbg !151 + +1625: ; preds = %__nv_exp2f.exit384 + %1626 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1608), !dbg !151 + br label %__nv_exp2f.exit387, !dbg !151 + +1627: ; preds = %__nv_exp2f.exit384 + %1628 = tail call float @llvm.nvvm.ex2.approx.f32(float %1608), !dbg !151 + br label %__nv_exp2f.exit387, !dbg !151 + +__nv_exp2f.exit387: ; preds = %1625, %1627 + %.0.i386 = phi float [ %1626, %1625 ], [ %1628, %1627 ], !dbg !151 + %1629 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !151 + %.not.i388 = icmp eq i32 %1629, 0, !dbg !151 + br i1 %.not.i388, label %1632, label %1630, !dbg !151 + +1630: ; preds = %__nv_exp2f.exit387 + %1631 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1609), !dbg !151 + br label %__nv_exp2f.exit390, !dbg !151 + +1632: ; preds = %__nv_exp2f.exit387 + %1633 = tail call float @llvm.nvvm.ex2.approx.f32(float %1609), !dbg !151 + br label %__nv_exp2f.exit390, !dbg !151 + +__nv_exp2f.exit390: ; preds = %1630, %1632 + %.0.i389 = phi float [ %1631, %1630 ], [ %1633, %1632 ], !dbg !151 + %1634 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !151 + %.not.i391 = icmp eq i32 %1634, 0, !dbg !151 + br i1 %.not.i391, label %1637, label %1635, !dbg !151 + +1635: ; preds = %__nv_exp2f.exit390 + %1636 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1610), !dbg !151 + br label %__nv_exp2f.exit393, !dbg !151 + +1637: ; preds = %__nv_exp2f.exit390 + %1638 = tail call float @llvm.nvvm.ex2.approx.f32(float %1610), !dbg !151 + br label %__nv_exp2f.exit393, !dbg !151 + +__nv_exp2f.exit393: ; preds = %1635, %1637 + %.0.i392 = phi float [ %1636, %1635 ], [ %1638, %1637 ], !dbg !151 + %1639 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !151 + %.not.i394 = icmp eq i32 %1639, 0, !dbg !151 + br i1 %.not.i394, label %1642, label %1640, !dbg !151 + +1640: ; preds = %__nv_exp2f.exit393 + %1641 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1611), !dbg !151 + br label %__nv_exp2f.exit396, !dbg !151 + +1642: ; preds = %__nv_exp2f.exit393 + %1643 = tail call float @llvm.nvvm.ex2.approx.f32(float %1611), !dbg !151 + br label %__nv_exp2f.exit396, !dbg !151 + +__nv_exp2f.exit396: ; preds = %1640, %1642 + %.0.i395 = phi float [ %1641, %1640 ], [ %1643, %1642 ], !dbg !151 + %1644 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !151 + %.not.i397 = icmp eq i32 %1644, 0, !dbg !151 + br i1 %.not.i397, label %1647, label %1645, !dbg !151 + +1645: ; preds = %__nv_exp2f.exit396 + %1646 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1612), !dbg !151 + br label %__nv_exp2f.exit399, !dbg !151 + +1647: ; preds = %__nv_exp2f.exit396 + %1648 = tail call float @llvm.nvvm.ex2.approx.f32(float %1612), !dbg !151 + br label %__nv_exp2f.exit399, !dbg !151 + +__nv_exp2f.exit399: ; preds = %1645, %1647 + %.0.i398 = phi float [ %1646, %1645 ], [ %1648, %1647 ], !dbg !151 + %1649 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !151 + %.not.i400 = icmp eq i32 %1649, 0, !dbg !151 + br i1 %.not.i400, label %1652, label %1650, !dbg !151 + +1650: ; preds = %__nv_exp2f.exit399 + %1651 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1613), !dbg !151 + br label %__nv_exp2f.exit402, !dbg !151 + +1652: ; preds = %__nv_exp2f.exit399 + %1653 = tail call float @llvm.nvvm.ex2.approx.f32(float %1613), !dbg !151 + br label %__nv_exp2f.exit402, !dbg !151 + +__nv_exp2f.exit402: ; preds = %1650, %1652 + %.0.i401 = phi float [ %1651, %1650 ], [ %1653, %1652 ], !dbg !151 + %1654 = select i1 %1598, float %.0.i380, float 0.000000e+00, !dbg !153 + %1655 = select i1 %1599, float %.0.i383, float 0.000000e+00, !dbg !153 + %1656 = select i1 %1600, float %.0.i386, float 0.000000e+00, !dbg !153 + %1657 = select i1 %1601, float %.0.i389, float 0.000000e+00, !dbg !153 + %1658 = select i1 %1602, float %.0.i392, float 0.000000e+00, !dbg !153 + %1659 = select i1 %1603, float %.0.i395, float 0.000000e+00, !dbg !153 + %1660 = select i1 %1604, float %.0.i398, float 0.000000e+00, !dbg !153 + %1661 = select i1 %1605, float %.0.i401, float 0.000000e+00, !dbg !153 + %1662 = fmul float %1016, %1654, !dbg !154 + %1663 = fmul float %1017, %1655, !dbg !154 + %1664 = fmul float %1018, %1656, !dbg !154 + %1665 = fmul float %1019, %1657, !dbg !154 + %1666 = fmul float %1020, %1658, !dbg !154 + %1667 = fmul float %1021, %1659, !dbg !154 + %1668 = fmul float %1022, %1660, !dbg !154 + %1669 = fmul float %1023, %1661, !dbg !154 + %1670 = and i1 %112, %207, !dbg !155 + %1671 = and i1 %112, %209, !dbg !155 + %1672 = and i1 %111, %207, !dbg !155 + %1673 = and i1 %111, %209, !dbg !155 + %1674 = and i1 %112, %211, !dbg !155 + %1675 = and i1 %112, %213, !dbg !155 + %1676 = and i1 %111, %211, !dbg !155 + %1677 = and i1 %111, %213, !dbg !155 + %1678 = fsub float %1080, %1076, !dbg !156 + %1679 = fsub float %1080, %1077, !dbg !156 + %1680 = fsub float %1081, %1076, !dbg !156 + %1681 = fsub float %1081, %1077, !dbg !156 + %1682 = fsub float %1080, %1078, !dbg !156 + %1683 = fsub float %1080, %1079, !dbg !156 + %1684 = fsub float %1081, %1078, !dbg !156 + %1685 = fsub float %1081, %1079, !dbg !156 + %1686 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !157 + %.not.i403 = icmp eq i32 %1686, 0, !dbg !157 + br i1 %.not.i403, label %1689, label %1687, !dbg !157 + +1687: ; preds = %__nv_exp2f.exit402 + %1688 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1678), !dbg !157 + br label %__nv_exp2f.exit405, !dbg !157 + +1689: ; preds = %__nv_exp2f.exit402 + %1690 = tail call float @llvm.nvvm.ex2.approx.f32(float %1678), !dbg !157 + br label %__nv_exp2f.exit405, !dbg !157 + +__nv_exp2f.exit405: ; preds = %1687, %1689 + %.0.i404 = phi float [ %1688, %1687 ], [ %1690, %1689 ], !dbg !157 + %1691 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !157 + %.not.i406 = icmp eq i32 %1691, 0, !dbg !157 + br i1 %.not.i406, label %1694, label %1692, !dbg !157 + +1692: ; preds = %__nv_exp2f.exit405 + %1693 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1679), !dbg !157 + br label %__nv_exp2f.exit408, !dbg !157 + +1694: ; preds = %__nv_exp2f.exit405 + %1695 = tail call float @llvm.nvvm.ex2.approx.f32(float %1679), !dbg !157 + br label %__nv_exp2f.exit408, !dbg !157 + +__nv_exp2f.exit408: ; preds = %1692, %1694 + %.0.i407 = phi float [ %1693, %1692 ], [ %1695, %1694 ], !dbg !157 + %1696 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !157 + %.not.i409 = icmp eq i32 %1696, 0, !dbg !157 + br i1 %.not.i409, label %1699, label %1697, !dbg !157 + +1697: ; preds = %__nv_exp2f.exit408 + %1698 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1680), !dbg !157 + br label %__nv_exp2f.exit411, !dbg !157 + +1699: ; preds = %__nv_exp2f.exit408 + %1700 = tail call float @llvm.nvvm.ex2.approx.f32(float %1680), !dbg !157 + br label %__nv_exp2f.exit411, !dbg !157 + +__nv_exp2f.exit411: ; preds = %1697, %1699 + %.0.i410 = phi float [ %1698, %1697 ], [ %1700, %1699 ], !dbg !157 + %1701 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !157 + %.not.i412 = icmp eq i32 %1701, 0, !dbg !157 + br i1 %.not.i412, label %1704, label %1702, !dbg !157 + +1702: ; preds = %__nv_exp2f.exit411 + %1703 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1681), !dbg !157 + br label %__nv_exp2f.exit414, !dbg !157 + +1704: ; preds = %__nv_exp2f.exit411 + %1705 = tail call float @llvm.nvvm.ex2.approx.f32(float %1681), !dbg !157 + br label %__nv_exp2f.exit414, !dbg !157 + +__nv_exp2f.exit414: ; preds = %1702, %1704 + %.0.i413 = phi float [ %1703, %1702 ], [ %1705, %1704 ], !dbg !157 + %1706 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !157 + %.not.i415 = icmp eq i32 %1706, 0, !dbg !157 + br i1 %.not.i415, label %1709, label %1707, !dbg !157 + +1707: ; preds = %__nv_exp2f.exit414 + %1708 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1682), !dbg !157 + br label %__nv_exp2f.exit417, !dbg !157 + +1709: ; preds = %__nv_exp2f.exit414 + %1710 = tail call float @llvm.nvvm.ex2.approx.f32(float %1682), !dbg !157 + br label %__nv_exp2f.exit417, !dbg !157 + +__nv_exp2f.exit417: ; preds = %1707, %1709 + %.0.i416 = phi float [ %1708, %1707 ], [ %1710, %1709 ], !dbg !157 + %1711 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !157 + %.not.i418 = icmp eq i32 %1711, 0, !dbg !157 + br i1 %.not.i418, label %1714, label %1712, !dbg !157 + +1712: ; preds = %__nv_exp2f.exit417 + %1713 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1683), !dbg !157 + br label %__nv_exp2f.exit420, !dbg !157 + +1714: ; preds = %__nv_exp2f.exit417 + %1715 = tail call float @llvm.nvvm.ex2.approx.f32(float %1683), !dbg !157 + br label %__nv_exp2f.exit420, !dbg !157 + +__nv_exp2f.exit420: ; preds = %1712, %1714 + %.0.i419 = phi float [ %1713, %1712 ], [ %1715, %1714 ], !dbg !157 + %1716 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !157 + %.not.i421 = icmp eq i32 %1716, 0, !dbg !157 + br i1 %.not.i421, label %1719, label %1717, !dbg !157 + +1717: ; preds = %__nv_exp2f.exit420 + %1718 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1684), !dbg !157 + br label %__nv_exp2f.exit423, !dbg !157 + +1719: ; preds = %__nv_exp2f.exit420 + %1720 = tail call float @llvm.nvvm.ex2.approx.f32(float %1684), !dbg !157 + br label %__nv_exp2f.exit423, !dbg !157 + +__nv_exp2f.exit423: ; preds = %1717, %1719 + %.0.i422 = phi float [ %1718, %1717 ], [ %1720, %1719 ], !dbg !157 + %1721 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !157 + %.not.i424 = icmp eq i32 %1721, 0, !dbg !157 + br i1 %.not.i424, label %1724, label %1722, !dbg !157 + +1722: ; preds = %__nv_exp2f.exit423 + %1723 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %1685), !dbg !157 + br label %__nv_exp2f.exit426, !dbg !157 + +1724: ; preds = %__nv_exp2f.exit423 + %1725 = tail call float @llvm.nvvm.ex2.approx.f32(float %1685), !dbg !157 + br label %__nv_exp2f.exit426, !dbg !157 + +__nv_exp2f.exit426: ; preds = %1722, %1724 + %.0.i425 = phi float [ %1723, %1722 ], [ %1725, %1724 ], !dbg !157 + %1726 = select i1 %1670, float %.0.i404, float 0.000000e+00, !dbg !159 + %1727 = select i1 %1671, float %.0.i407, float 0.000000e+00, !dbg !159 + %1728 = select i1 %1672, float %.0.i410, float 0.000000e+00, !dbg !159 + %1729 = select i1 %1673, float %.0.i413, float 0.000000e+00, !dbg !159 + %1730 = select i1 %1674, float %.0.i416, float 0.000000e+00, !dbg !159 + %1731 = select i1 %1675, float %.0.i419, float 0.000000e+00, !dbg !159 + %1732 = select i1 %1676, float %.0.i422, float 0.000000e+00, !dbg !159 + %1733 = select i1 %1677, float %.0.i425, float 0.000000e+00, !dbg !159 + %1734 = fmul float %1024, %1726, !dbg !160 + %1735 = fmul float %1025, %1727, !dbg !160 + %1736 = fmul float %1026, %1728, !dbg !160 + %1737 = fmul float %1027, %1729, !dbg !160 + %1738 = fmul float %1028, %1730, !dbg !160 + %1739 = fmul float %1029, %1731, !dbg !160 + %1740 = fmul float %1030, %1732, !dbg !160 + %1741 = fmul float %1031, %1733, !dbg !160 + %1742 = fmul <4 x float> %1137, %1044, !dbg !161 + %1743 = fmul <4 x float> %1195, %1049, !dbg !162 + %1744 = extractelement <4 x float> %1049, i64 2, !dbg !163 + %1745 = fmul float %1191, %1744, !dbg !162 + %1746 = fmul float %1192, %1744, !dbg !162 + %1747 = extractelement <4 x float> %1049, i64 0, !dbg !163 + %1748 = fmul float %1193, %1747, !dbg !162 + %1749 = fmul float %1194, %1747, !dbg !162 + %1750 = fmul <4 x float> %1252, %1054, !dbg !164 + %1751 = extractelement <4 x float> %1054, i64 2, !dbg !165 + %1752 = fmul float %1248, %1751, !dbg !164 + %1753 = fmul float %1249, %1751, !dbg !164 + %1754 = extractelement <4 x float> %1054, i64 0, !dbg !165 + %1755 = fmul float %1250, %1754, !dbg !164 + %1756 = fmul float %1251, %1754, !dbg !164 + %1757 = fmul <4 x float> %1309, %1059, !dbg !166 + %1758 = extractelement <4 x float> %1059, i64 2, !dbg !167 + %1759 = fmul float %1305, %1758, !dbg !166 + %1760 = fmul float %1306, %1758, !dbg !166 + %1761 = extractelement <4 x float> %1059, i64 0, !dbg !167 + %1762 = fmul float %1307, %1761, !dbg !166 + %1763 = fmul float %1308, %1761, !dbg !166 + %1764 = fmul float %1374, %1747, !dbg !163 + %1765 = fmul float %1375, %1747, !dbg !163 + %1766 = fmul float %1376, %1744, !dbg !163 + %1767 = fmul float %1377, %1744, !dbg !163 + %1768 = fmul float %1378, %1747, !dbg !163 + %1769 = fmul float %1379, %1747, !dbg !163 + %1770 = fmul float %1380, %1744, !dbg !163 + %1771 = fmul float %1381, %1744, !dbg !163 + %1772 = fmul float %1446, %1754, !dbg !168 + %1773 = fmul float %1447, %1754, !dbg !168 + %1774 = fmul float %1448, %1751, !dbg !168 + %1775 = fmul float %1449, %1751, !dbg !168 + %1776 = fmul float %1450, %1754, !dbg !168 + %1777 = fmul float %1451, %1754, !dbg !168 + %1778 = fmul float %1452, %1751, !dbg !168 + %1779 = fmul float %1453, %1751, !dbg !168 + %1780 = fmul float %1518, %1754, !dbg !165 + %1781 = fmul float %1519, %1754, !dbg !165 + %1782 = fmul float %1520, %1751, !dbg !165 + %1783 = fmul float %1521, %1751, !dbg !165 + %1784 = fmul float %1522, %1754, !dbg !165 + %1785 = fmul float %1523, %1754, !dbg !165 + %1786 = fmul float %1524, %1751, !dbg !165 + %1787 = fmul float %1525, %1751, !dbg !165 + %1788 = fmul float %1590, %1761, !dbg !169 + %1789 = fmul float %1591, %1761, !dbg !169 + %1790 = fmul float %1592, %1758, !dbg !169 + %1791 = fmul float %1593, %1758, !dbg !169 + %1792 = fmul float %1594, %1761, !dbg !169 + %1793 = fmul float %1595, %1761, !dbg !169 + %1794 = fmul float %1596, %1758, !dbg !169 + %1795 = fmul float %1597, %1758, !dbg !169 + %1796 = fmul float %1662, %1761, !dbg !170 + %1797 = fmul float %1663, %1761, !dbg !170 + %1798 = fmul float %1664, %1758, !dbg !170 + %1799 = fmul float %1665, %1758, !dbg !170 + %1800 = fmul float %1666, %1761, !dbg !170 + %1801 = fmul float %1667, %1761, !dbg !170 + %1802 = fmul float %1668, %1758, !dbg !170 + %1803 = fmul float %1669, %1758, !dbg !170 + %1804 = fmul float %1734, %1761, !dbg !167 + %1805 = fmul float %1735, %1761, !dbg !167 + %1806 = fmul float %1736, %1758, !dbg !167 + %1807 = fmul float %1737, %1758, !dbg !167 + %1808 = fmul float %1738, %1761, !dbg !167 + %1809 = fmul float %1739, %1761, !dbg !167 + %1810 = fmul float %1740, %1758, !dbg !167 + %1811 = fmul float %1741, %1758, !dbg !167 + %1812 = fsub <4 x float> zeroinitializer, %1742, !dbg !171 + %1813 = fsub float 0.000000e+00, %1745, !dbg !172 + %1814 = fsub float 0.000000e+00, %1746, !dbg !172 + %1815 = fsub float 0.000000e+00, %1748, !dbg !172 + %1816 = fsub float 0.000000e+00, %1749, !dbg !172 + %1817 = fsub <4 x float> zeroinitializer, %1743, !dbg !172 + %1818 = fsub float 0.000000e+00, %1752, !dbg !173 + %1819 = fsub float 0.000000e+00, %1753, !dbg !173 + %1820 = fsub float 0.000000e+00, %1755, !dbg !173 + %1821 = fsub float 0.000000e+00, %1756, !dbg !173 + %1822 = fsub <4 x float> zeroinitializer, %1750, !dbg !173 + %1823 = fsub float 0.000000e+00, %1759, !dbg !174 + %1824 = fsub float 0.000000e+00, %1760, !dbg !174 + %1825 = fsub float 0.000000e+00, %1762, !dbg !174 + %1826 = fsub float 0.000000e+00, %1763, !dbg !174 + %1827 = fsub <4 x float> zeroinitializer, %1757, !dbg !174 + %1828 = sub nsw i64 %10, %11, !dbg !175 + %1829 = icmp sgt i64 %1828, 2, !dbg !176 + br i1 %1829, label %.lr.ph, label %._crit_edge, !dbg !176 + +.lr.ph: ; preds = %__nv_exp2f.exit426 + %1830 = shl nuw nsw i32 %28, 4 + %1831 = lshr exact i32 %29, 3 + %1832 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %1830 + %1833 = getelementptr inbounds nuw i8, ptr addrspace(3) %1832, i32 %1831 + %1834 = getelementptr inbounds nuw i8, ptr addrspace(3) %1833, i32 8 + %1835 = shl nuw nsw i32 %26, 1 + %1836 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %1835 + %1837 = tail call i64 @llvm.umin.i64(i64 %1828, i64 16), !dbg !176 + %1838 = getelementptr inbounds nuw i8, ptr addrspace(3) %1836, i32 4 + %1839 = extractelement <4 x float> %1812, i64 0, !dbg !177 + %1840 = extractelement <4 x float> %1812, i64 1, !dbg !177 + %1841 = extractelement <4 x float> %1812, i64 2, !dbg !177 + %1842 = extractelement <4 x float> %1812, i64 3, !dbg !177 + %1843 = zext nneg <2 x i32> %64 to <2 x i64> + br label %1844, !dbg !176 + +1844: ; preds = %.lr.ph, %1844 + %1845 = phi i64 [ 2, %.lr.ph ], [ %1933, %1844 ] + %1846 = phi <4 x float> [ %1812, %.lr.ph ], [ %1932, %1844 ] + %1847 = icmp eq i64 %1845, %85, !dbg !178 + %1848 = icmp eq i64 %1845, %83, !dbg !178 + %1849 = select i1 %1847, float %1839, float 0.000000e+00, !dbg !177 + %1850 = select i1 %1847, float %1840, float 0.000000e+00, !dbg !177 + %1851 = select i1 %1848, float %1841, float 0.000000e+00, !dbg !177 + %1852 = select i1 %1848, float %1842, float 0.000000e+00, !dbg !177 + %1853 = fadd float %1849, %1851, !dbg !179 + %1854 = fadd float %1850, %1852, !dbg !179 + %1855 = bitcast float %1853 to i32, !dbg !182 + %1856 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1855, i32 16, i32 31), !dbg !182 + %1857 = bitcast i32 %1856 to float, !dbg !182 + %1858 = fadd float %1853, %1857, !dbg !179 + %1859 = bitcast float %1858 to i32, !dbg !182 + %1860 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1859, i32 8, i32 31), !dbg !182 + %1861 = bitcast i32 %1860 to float, !dbg !182 + %1862 = bitcast float %1854 to i32, !dbg !182 + %1863 = insertelement <2 x i64> poison, i64 %1845, i64 0, !dbg !184 + %1864 = shufflevector <2 x i64> %1863, <2 x i64> poison, <2 x i32> zeroinitializer, !dbg !184 + %1865 = icmp samesign ugt <2 x i64> %1864, %1843, !dbg !184 + %1866 = fadd float %1858, %1861, !dbg !179 + %1867 = bitcast float %1866 to i32, !dbg !182 + %1868 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1867, i32 4, i32 31), !dbg !182 + %1869 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1862, i32 16, i32 31), !dbg !182 + %1870 = bitcast i32 %1869 to float, !dbg !182 + %1871 = fadd float %1854, %1870, !dbg !179 + %1872 = bitcast float %1871 to i32, !dbg !182 + %1873 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1872, i32 8, i32 31), !dbg !182 + %1874 = bitcast i32 %1873 to float, !dbg !182 + %1875 = fadd float %1871, %1874, !dbg !179 + %1876 = bitcast float %1875 to i32, !dbg !182 + %1877 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1876, i32 4, i32 31), !dbg !182 + %1878 = insertelement <2 x i32> poison, i32 %1868, i64 0, !dbg !182 + %1879 = insertelement <2 x i32> %1878, i32 %1877, i64 1, !dbg !182 + %1880 = bitcast <2 x i32> %1879 to <2 x float>, !dbg !182 + %1881 = insertelement <2 x float> poison, float %1866, i64 0, !dbg !179 + %1882 = insertelement <2 x float> %1881, float %1875, i64 1, !dbg !179 + %1883 = fadd <2 x float> %1882, %1880, !dbg !179 + %1884 = select <2 x i1> %1865, <2 x float> %1883, <2 x float> zeroinitializer, !dbg !185 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !186 + %1885 = extractelement <2 x float> %1884, i64 0, !dbg !186 + %1886 = bitcast float %1885 to <1 x i32>, !dbg !186 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) %1833, <1 x i32> %1886) #2, !dbg !186 + %1887 = extractelement <2 x float> %1884, i64 1, !dbg !186 + %1888 = bitcast float %1887 to <1 x i32>, !dbg !186 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) nonnull %1834, <1 x i32> %1888) #2, !dbg !186 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !186 + %1889 = load float, ptr addrspace(3) %1836, align 8, !dbg !186 + %1890 = load float, ptr addrspace(3) %1838, align 4, !dbg !186 + %1891 = extractelement <4 x float> %1846, i64 0, !dbg !186 + %1892 = fmul float %1891, %1889, !dbg !186 + %1893 = extractelement <4 x float> %1846, i64 1, !dbg !186 + %1894 = fmul float %1893, %1889, !dbg !186 + %1895 = extractelement <4 x float> %1846, i64 2, !dbg !186 + %1896 = fmul float %1895, %1890, !dbg !186 + %1897 = extractelement <4 x float> %1846, i64 3, !dbg !186 + %1898 = fmul float %1897, %1890, !dbg !186 + %1899 = fadd float %1892, %1896, !dbg !187 + %1900 = fadd float %1894, %1898, !dbg !187 + %1901 = bitcast float %1899 to i32, !dbg !188 + %1902 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1901, i32 16, i32 31), !dbg !188 + %1903 = bitcast i32 %1902 to float, !dbg !188 + %1904 = fadd float %1899, %1903, !dbg !187 + %1905 = bitcast float %1904 to i32, !dbg !188 + %1906 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1905, i32 8, i32 31), !dbg !188 + %1907 = bitcast i32 %1906 to float, !dbg !188 + %1908 = fadd float %1904, %1907, !dbg !187 + %1909 = bitcast float %1908 to i32, !dbg !188 + %1910 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1909, i32 4, i32 31), !dbg !188 + %1911 = bitcast float %1900 to i32, !dbg !188 + %1912 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1911, i32 16, i32 31), !dbg !188 + %1913 = bitcast i32 %1912 to float, !dbg !188 + %1914 = fadd float %1900, %1913, !dbg !187 + %1915 = bitcast float %1914 to i32, !dbg !188 + %1916 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1915, i32 8, i32 31), !dbg !188 + %1917 = bitcast i32 %1916 to float, !dbg !188 + %1918 = fadd float %1914, %1917, !dbg !187 + %1919 = bitcast float %1918 to i32, !dbg !188 + %1920 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1919, i32 4, i32 31), !dbg !188 + %1921 = insertelement <2 x i32> poison, i32 %1910, i64 0, !dbg !188 + %1922 = insertelement <2 x i32> %1921, i32 %1920, i64 1, !dbg !188 + %1923 = bitcast <2 x i32> %1922 to <2 x float>, !dbg !188 + %1924 = insertelement <2 x float> poison, float %1908, i64 0, !dbg !187 + %1925 = insertelement <2 x float> %1924, float %1918, i64 1, !dbg !187 + %1926 = fadd <2 x float> %1925, %1923, !dbg !187 + %1927 = fadd <2 x float> %1884, %1926, !dbg !190 + %1928 = shufflevector <2 x float> %1927, <2 x float> poison, <4 x i32> , !dbg !190 + %1929 = insertelement <4 x i1> poison, i1 %1847, i64 0, !dbg !191 + %1930 = insertelement <4 x i1> %1929, i1 %1848, i64 1, !dbg !191 + %1931 = shufflevector <4 x i1> %1930, <4 x i1> poison, <4 x i32> , !dbg !191 + %1932 = select <4 x i1> %1931, <4 x float> %1928, <4 x float> %1846, !dbg !191 + %1933 = add nuw nsw i64 %1845, 1, !dbg !176 + %exitcond.not = icmp eq i64 %1933, %1837, !dbg !176 + br i1 %exitcond.not, label %._crit_edge, label %1844, !dbg !176 + +._crit_edge: ; preds = %1844, %__nv_exp2f.exit426 + %1934 = phi <4 x float> [ %1812, %__nv_exp2f.exit426 ], [ %1932, %1844 ] + %1935 = sub nsw i64 %10, %17, !dbg !192 + %1936 = icmp sgt i64 %1935, 2, !dbg !193 + br i1 %1936, label %.lr.ph454, label %._crit_edge455, !dbg !193 + +.lr.ph454: ; preds = %._crit_edge + %1937 = and i32 %23, 4 + %1938 = icmp eq i32 %1937, 0 + %1939 = shl nuw nsw i32 %23, 3 + %1940 = and i32 %1939, 24 + %1941 = lshr i32 %23, 3 + %1942 = and i32 %1941, 3 + %1943 = or disjoint i32 %1942, %1940 + %1944 = or disjoint i32 %1943, %1937 + %1945 = xor i32 %1944, 4 + %1946 = tail call i64 @llvm.umin.i64(i64 %1935, i64 16), !dbg !193 + %1947 = extractelement <4 x float> %1817, i64 0, !dbg !194 + %1948 = extractelement <4 x float> %1817, i64 1, !dbg !194 + %1949 = extractelement <4 x float> %1817, i64 2, !dbg !194 + %1950 = extractelement <4 x float> %1817, i64 3, !dbg !194 + %.mask710 = and <4 x i32> %49, splat (i32 255) + %1951 = zext nneg <4 x i32> %.mask710 to <4 x i64> + br label %1952, !dbg !193 + +1952: ; preds = %.lr.ph454, %1952 + %.pn54450 = phi float [ %1816, %.lr.ph454 ], [ %2115, %1952 ] + %.pn56449 = phi float [ %1815, %.lr.ph454 ], [ %2113, %1952 ] + %.pn58448 = phi float [ %1814, %.lr.ph454 ], [ %2111, %1952 ] + %.pn60447 = phi float [ %1813, %.lr.ph454 ], [ %2109, %1952 ] + %1953 = phi i64 [ 2, %.lr.ph454 ], [ %2120, %1952 ] + %1954 = phi <4 x float> [ %1817, %.lr.ph454 ], [ %2119, %1952 ] + %1955 = icmp eq i64 %1953, %85, !dbg !195 + %1956 = icmp eq i64 %1953, %83, !dbg !195 + %1957 = select i1 %1955, float %1947, float 0.000000e+00, !dbg !194 + %1958 = select i1 %1955, float %1948, float 0.000000e+00, !dbg !194 + %1959 = select i1 %1956, float %1813, float 0.000000e+00, !dbg !194 + %1960 = select i1 %1956, float %1814, float 0.000000e+00, !dbg !194 + %1961 = select i1 %1955, float %1815, float 0.000000e+00, !dbg !194 + %1962 = select i1 %1955, float %1816, float 0.000000e+00, !dbg !194 + %1963 = select i1 %1956, float %1949, float 0.000000e+00, !dbg !194 + %1964 = select i1 %1956, float %1950, float 0.000000e+00, !dbg !194 + %1965 = fadd float %1957, %1959, !dbg !196 + %1966 = fadd float %1958, %1960, !dbg !196 + %1967 = fadd float %1961, %1963, !dbg !196 + %1968 = fadd float %1962, %1964, !dbg !196 + %1969 = bitcast float %1965 to i32, !dbg !197 + %1970 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1969, i32 16, i32 31), !dbg !197 + %1971 = bitcast i32 %1970 to float, !dbg !197 + %1972 = fadd float %1965, %1971, !dbg !196 + %1973 = bitcast float %1972 to i32, !dbg !197 + %1974 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1973, i32 8, i32 31), !dbg !197 + %1975 = bitcast i32 %1974 to float, !dbg !197 + %1976 = bitcast float %1966 to i32, !dbg !197 + %1977 = bitcast float %1967 to i32, !dbg !197 + %1978 = bitcast float %1968 to i32, !dbg !197 + %1979 = fadd float %1972, %1975, !dbg !196 + %1980 = bitcast float %1979 to i32, !dbg !197 + %1981 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1980, i32 4, i32 31), !dbg !197 + %1982 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1976, i32 16, i32 31), !dbg !197 + %1983 = bitcast i32 %1982 to float, !dbg !197 + %1984 = fadd float %1966, %1983, !dbg !196 + %1985 = bitcast float %1984 to i32, !dbg !197 + %1986 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1985, i32 8, i32 31), !dbg !197 + %1987 = bitcast i32 %1986 to float, !dbg !197 + %1988 = fadd float %1984, %1987, !dbg !196 + %1989 = bitcast float %1988 to i32, !dbg !197 + %1990 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1989, i32 4, i32 31), !dbg !197 + %1991 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1977, i32 16, i32 31), !dbg !197 + %1992 = bitcast i32 %1991 to float, !dbg !197 + %1993 = fadd float %1967, %1992, !dbg !196 + %1994 = bitcast float %1993 to i32, !dbg !197 + %1995 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1994, i32 8, i32 31), !dbg !197 + %1996 = bitcast i32 %1995 to float, !dbg !197 + %1997 = fadd float %1993, %1996, !dbg !196 + %1998 = bitcast float %1997 to i32, !dbg !197 + %1999 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1998, i32 4, i32 31), !dbg !197 + %2000 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1978, i32 16, i32 31), !dbg !197 + %2001 = bitcast i32 %2000 to float, !dbg !197 + %2002 = fadd float %1968, %2001, !dbg !196 + %2003 = bitcast float %2002 to i32, !dbg !197 + %2004 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2003, i32 8, i32 31), !dbg !197 + %2005 = bitcast i32 %2004 to float, !dbg !197 + %2006 = fadd float %2002, %2005, !dbg !196 + %2007 = bitcast float %2006 to i32, !dbg !197 + %2008 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2007, i32 4, i32 31), !dbg !197 + %2009 = insertelement <4 x i32> poison, i32 %1981, i64 0, !dbg !197 + %2010 = insertelement <4 x i32> %2009, i32 %1990, i64 1, !dbg !197 + %2011 = insertelement <4 x i32> %2010, i32 %1999, i64 2, !dbg !197 + %2012 = insertelement <4 x i32> %2011, i32 %2008, i64 3, !dbg !197 + %2013 = bitcast <4 x i32> %2012 to <4 x float>, !dbg !197 + %2014 = insertelement <4 x float> poison, float %1979, i64 0, !dbg !196 + %2015 = insertelement <4 x float> %2014, float %1988, i64 1, !dbg !196 + %2016 = insertelement <4 x float> %2015, float %1997, i64 2, !dbg !196 + %2017 = insertelement <4 x float> %2016, float %2006, i64 3, !dbg !196 + %2018 = fadd <4 x float> %2017, %2013, !dbg !196 + %2019 = insertelement <4 x i64> poison, i64 %1953, i64 0, !dbg !199 + %2020 = shufflevector <4 x i64> %2019, <4 x i64> poison, <4 x i32> zeroinitializer, !dbg !199 + %2021 = icmp samesign ugt <4 x i64> %2020, %1951, !dbg !199 + %2022 = select <4 x i1> %2021, <4 x float> %2018, <4 x float> zeroinitializer, !dbg !200 + %2023 = extractelement <4 x float> %2022, i64 0, !dbg !201 + %2024 = extractelement <4 x float> %2022, i64 1, !dbg !201 + %2025 = select i1 %1938, float %2023, float %2024, !dbg !201 + %2026 = select i1 %1938, float %2024, float %2023, !dbg !201 + %2027 = extractelement <4 x float> %2022, i64 2, !dbg !201 + %2028 = extractelement <4 x float> %2022, i64 3, !dbg !201 + %2029 = select i1 %1938, float %2027, float %2028, !dbg !201 + %2030 = select i1 %1938, float %2028, float %2027, !dbg !201 + %2031 = bitcast float %2025 to i32, !dbg !201 + %2032 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2031, i32 %1944, i32 31), !dbg !201 + %2033 = bitcast float %2029 to i32, !dbg !201 + %2034 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2033, i32 %1944, i32 31), !dbg !201 + %2035 = bitcast float %2026 to i32, !dbg !201 + %2036 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2035, i32 %1945, i32 31), !dbg !201 + %2037 = bitcast float %2030 to i32, !dbg !201 + %2038 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2037, i32 %1945, i32 31), !dbg !201 + %.v173 = select i1 %1938, i32 %2032, i32 %2036, !dbg !201 + %2039 = bitcast i32 %.v173 to float, !dbg !201 + %.v174 = select i1 %1938, i32 %2034, i32 %2038, !dbg !201 + %2040 = bitcast i32 %.v174 to float, !dbg !201 + %2041 = extractelement <4 x float> %1954, i64 0, !dbg !201 + %2042 = fmul float %2041, %2039, !dbg !201 + %2043 = extractelement <4 x float> %1954, i64 1, !dbg !201 + %2044 = fmul float %2043, %2039, !dbg !201 + %2045 = fmul float %.pn60447, %2040, !dbg !201 + %2046 = fmul float %.pn58448, %2040, !dbg !201 + %2047 = fmul float %.pn56449, %2039, !dbg !201 + %2048 = fmul float %.pn54450, %2039, !dbg !201 + %2049 = extractelement <4 x float> %1954, i64 2, !dbg !201 + %2050 = fmul float %2049, %2040, !dbg !201 + %2051 = extractelement <4 x float> %1954, i64 3, !dbg !201 + %2052 = fmul float %2051, %2040, !dbg !201 + %2053 = fadd float %2042, %2045, !dbg !202 + %2054 = fadd float %2044, %2046, !dbg !202 + %2055 = fadd float %2047, %2050, !dbg !202 + %2056 = fadd float %2048, %2052, !dbg !202 + %2057 = bitcast float %2053 to i32, !dbg !203 + %2058 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2057, i32 16, i32 31), !dbg !203 + %2059 = bitcast i32 %2058 to float, !dbg !203 + %2060 = fadd float %2053, %2059, !dbg !202 + %2061 = bitcast float %2060 to i32, !dbg !203 + %2062 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2061, i32 8, i32 31), !dbg !203 + %2063 = bitcast i32 %2062 to float, !dbg !203 + %2064 = fadd float %2060, %2063, !dbg !202 + %2065 = bitcast float %2064 to i32, !dbg !203 + %2066 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2065, i32 4, i32 31), !dbg !203 + %2067 = bitcast float %2054 to i32, !dbg !203 + %2068 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2067, i32 16, i32 31), !dbg !203 + %2069 = bitcast i32 %2068 to float, !dbg !203 + %2070 = fadd float %2054, %2069, !dbg !202 + %2071 = bitcast float %2070 to i32, !dbg !203 + %2072 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2071, i32 8, i32 31), !dbg !203 + %2073 = bitcast i32 %2072 to float, !dbg !203 + %2074 = fadd float %2070, %2073, !dbg !202 + %2075 = bitcast float %2074 to i32, !dbg !203 + %2076 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2075, i32 4, i32 31), !dbg !203 + %2077 = bitcast float %2055 to i32, !dbg !203 + %2078 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2077, i32 16, i32 31), !dbg !203 + %2079 = bitcast i32 %2078 to float, !dbg !203 + %2080 = fadd float %2055, %2079, !dbg !202 + %2081 = bitcast float %2080 to i32, !dbg !203 + %2082 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2081, i32 8, i32 31), !dbg !203 + %2083 = bitcast i32 %2082 to float, !dbg !203 + %2084 = fadd float %2080, %2083, !dbg !202 + %2085 = bitcast float %2084 to i32, !dbg !203 + %2086 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2085, i32 4, i32 31), !dbg !203 + %2087 = bitcast float %2056 to i32, !dbg !203 + %2088 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2087, i32 16, i32 31), !dbg !203 + %2089 = bitcast i32 %2088 to float, !dbg !203 + %2090 = fadd float %2056, %2089, !dbg !202 + %2091 = bitcast float %2090 to i32, !dbg !203 + %2092 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2091, i32 8, i32 31), !dbg !203 + %2093 = bitcast i32 %2092 to float, !dbg !203 + %2094 = fadd float %2090, %2093, !dbg !202 + %2095 = bitcast float %2094 to i32, !dbg !203 + %2096 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2095, i32 4, i32 31), !dbg !203 + %2097 = insertelement <4 x i32> poison, i32 %2066, i64 0, !dbg !203 + %2098 = insertelement <4 x i32> %2097, i32 %2076, i64 1, !dbg !203 + %2099 = insertelement <4 x i32> %2098, i32 %2086, i64 2, !dbg !203 + %2100 = insertelement <4 x i32> %2099, i32 %2096, i64 3, !dbg !203 + %2101 = bitcast <4 x i32> %2100 to <4 x float>, !dbg !203 + %2102 = insertelement <4 x float> poison, float %2064, i64 0, !dbg !202 + %2103 = insertelement <4 x float> %2102, float %2074, i64 1, !dbg !202 + %2104 = insertelement <4 x float> %2103, float %2084, i64 2, !dbg !202 + %2105 = insertelement <4 x float> %2104, float %2094, i64 3, !dbg !202 + %2106 = fadd <4 x float> %2105, %2101, !dbg !202 + %2107 = fadd <4 x float> %2022, %2106, !dbg !205 + %2108 = extractelement <4 x float> %2107, i64 0, !dbg !206 + %2109 = select i1 %1956, float %2108, float %.pn60447, !dbg !206 + %2110 = extractelement <4 x float> %2107, i64 1, !dbg !206 + %2111 = select i1 %1956, float %2110, float %.pn58448, !dbg !206 + %2112 = extractelement <4 x float> %2107, i64 2, !dbg !206 + %2113 = select i1 %1955, float %2112, float %.pn56449, !dbg !206 + %2114 = extractelement <4 x float> %2107, i64 3, !dbg !206 + %2115 = select i1 %1955, float %2114, float %.pn54450, !dbg !206 + %2116 = insertelement <4 x i1> poison, i1 %1955, i64 0, !dbg !206 + %2117 = insertelement <4 x i1> %2116, i1 %1956, i64 1, !dbg !206 + %2118 = shufflevector <4 x i1> %2117, <4 x i1> poison, <4 x i32> , !dbg !206 + %2119 = select <4 x i1> %2118, <4 x float> %2107, <4 x float> %1954, !dbg !206 + %2120 = add nuw nsw i64 %1953, 1, !dbg !193 + %exitcond548.not = icmp eq i64 %2120, %1946, !dbg !193 + br i1 %exitcond548.not, label %._crit_edge455.loopexit, label %1952, !dbg !193 + +._crit_edge455.loopexit: ; preds = %1952 + %2121 = fadd float %2109, 0.000000e+00, !dbg !207 + %2122 = fadd float %2111, 0.000000e+00, !dbg !207 + %2123 = fadd float %2113, 0.000000e+00, !dbg !207 + %2124 = fadd float %2115, 0.000000e+00, !dbg !207 + br label %._crit_edge455, !dbg !208 + +._crit_edge455: ; preds = %._crit_edge455.loopexit, %._crit_edge + %.pn60.lcssa = phi float [ %1813, %._crit_edge ], [ %2121, %._crit_edge455.loopexit ] + %.pn58.lcssa = phi float [ %1814, %._crit_edge ], [ %2122, %._crit_edge455.loopexit ] + %.pn56.lcssa = phi float [ %1815, %._crit_edge ], [ %2123, %._crit_edge455.loopexit ] + %.pn54.lcssa = phi float [ %1816, %._crit_edge ], [ %2124, %._crit_edge455.loopexit ] + %2125 = phi <4 x float> [ %1817, %._crit_edge ], [ %2119, %._crit_edge455.loopexit ] + %2126 = sub nsw i64 %10, %18, !dbg !208 + %2127 = icmp sgt i64 %2126, 2, !dbg !209 + br i1 %2127, label %.lr.ph473, label %._crit_edge474, !dbg !209 + +.lr.ph473: ; preds = %._crit_edge455 + %2128 = and i32 %23, 4 + %2129 = icmp eq i32 %2128, 0 + %2130 = shl nuw nsw i32 %23, 3 + %2131 = and i32 %2130, 24 + %2132 = lshr i32 %23, 3 + %2133 = and i32 %2132, 3 + %2134 = or disjoint i32 %2133, %2131 + %2135 = or disjoint i32 %2134, %2128 + %2136 = xor i32 %2135, 4 + %2137 = tail call i64 @llvm.umin.i64(i64 %2126, i64 16), !dbg !209 + %2138 = extractelement <4 x float> %1822, i64 0, !dbg !210 + %2139 = extractelement <4 x float> %1822, i64 1, !dbg !210 + %2140 = extractelement <4 x float> %1822, i64 2, !dbg !210 + %2141 = extractelement <4 x float> %1822, i64 3, !dbg !210 + %.mask711 = and <4 x i32> %49, splat (i32 255) + %2142 = zext nneg <4 x i32> %.mask711 to <4 x i64> + br label %2143, !dbg !209 + +2143: ; preds = %.lr.ph473, %2143 + %.pn70469 = phi float [ %1821, %.lr.ph473 ], [ %2306, %2143 ] + %.pn72468 = phi float [ %1820, %.lr.ph473 ], [ %2304, %2143 ] + %.pn74467 = phi float [ %1819, %.lr.ph473 ], [ %2302, %2143 ] + %.pn76466 = phi float [ %1818, %.lr.ph473 ], [ %2300, %2143 ] + %2144 = phi i64 [ 2, %.lr.ph473 ], [ %2311, %2143 ] + %2145 = phi <4 x float> [ %1822, %.lr.ph473 ], [ %2310, %2143 ] + %2146 = icmp eq i64 %2144, %85, !dbg !211 + %2147 = icmp eq i64 %2144, %83, !dbg !211 + %2148 = select i1 %2146, float %2138, float 0.000000e+00, !dbg !210 + %2149 = select i1 %2146, float %2139, float 0.000000e+00, !dbg !210 + %2150 = select i1 %2147, float %1818, float 0.000000e+00, !dbg !210 + %2151 = select i1 %2147, float %1819, float 0.000000e+00, !dbg !210 + %2152 = select i1 %2146, float %1820, float 0.000000e+00, !dbg !210 + %2153 = select i1 %2146, float %1821, float 0.000000e+00, !dbg !210 + %2154 = select i1 %2147, float %2140, float 0.000000e+00, !dbg !210 + %2155 = select i1 %2147, float %2141, float 0.000000e+00, !dbg !210 + %2156 = fadd float %2148, %2150, !dbg !212 + %2157 = fadd float %2149, %2151, !dbg !212 + %2158 = fadd float %2152, %2154, !dbg !212 + %2159 = fadd float %2153, %2155, !dbg !212 + %2160 = bitcast float %2156 to i32, !dbg !213 + %2161 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2160, i32 16, i32 31), !dbg !213 + %2162 = bitcast i32 %2161 to float, !dbg !213 + %2163 = fadd float %2156, %2162, !dbg !212 + %2164 = bitcast float %2163 to i32, !dbg !213 + %2165 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2164, i32 8, i32 31), !dbg !213 + %2166 = bitcast i32 %2165 to float, !dbg !213 + %2167 = bitcast float %2157 to i32, !dbg !213 + %2168 = bitcast float %2158 to i32, !dbg !213 + %2169 = bitcast float %2159 to i32, !dbg !213 + %2170 = fadd float %2163, %2166, !dbg !212 + %2171 = bitcast float %2170 to i32, !dbg !213 + %2172 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2171, i32 4, i32 31), !dbg !213 + %2173 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2167, i32 16, i32 31), !dbg !213 + %2174 = bitcast i32 %2173 to float, !dbg !213 + %2175 = fadd float %2157, %2174, !dbg !212 + %2176 = bitcast float %2175 to i32, !dbg !213 + %2177 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2176, i32 8, i32 31), !dbg !213 + %2178 = bitcast i32 %2177 to float, !dbg !213 + %2179 = fadd float %2175, %2178, !dbg !212 + %2180 = bitcast float %2179 to i32, !dbg !213 + %2181 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2180, i32 4, i32 31), !dbg !213 + %2182 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2168, i32 16, i32 31), !dbg !213 + %2183 = bitcast i32 %2182 to float, !dbg !213 + %2184 = fadd float %2158, %2183, !dbg !212 + %2185 = bitcast float %2184 to i32, !dbg !213 + %2186 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2185, i32 8, i32 31), !dbg !213 + %2187 = bitcast i32 %2186 to float, !dbg !213 + %2188 = fadd float %2184, %2187, !dbg !212 + %2189 = bitcast float %2188 to i32, !dbg !213 + %2190 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2189, i32 4, i32 31), !dbg !213 + %2191 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2169, i32 16, i32 31), !dbg !213 + %2192 = bitcast i32 %2191 to float, !dbg !213 + %2193 = fadd float %2159, %2192, !dbg !212 + %2194 = bitcast float %2193 to i32, !dbg !213 + %2195 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2194, i32 8, i32 31), !dbg !213 + %2196 = bitcast i32 %2195 to float, !dbg !213 + %2197 = fadd float %2193, %2196, !dbg !212 + %2198 = bitcast float %2197 to i32, !dbg !213 + %2199 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2198, i32 4, i32 31), !dbg !213 + %2200 = insertelement <4 x i32> poison, i32 %2172, i64 0, !dbg !213 + %2201 = insertelement <4 x i32> %2200, i32 %2181, i64 1, !dbg !213 + %2202 = insertelement <4 x i32> %2201, i32 %2190, i64 2, !dbg !213 + %2203 = insertelement <4 x i32> %2202, i32 %2199, i64 3, !dbg !213 + %2204 = bitcast <4 x i32> %2203 to <4 x float>, !dbg !213 + %2205 = insertelement <4 x float> poison, float %2170, i64 0, !dbg !212 + %2206 = insertelement <4 x float> %2205, float %2179, i64 1, !dbg !212 + %2207 = insertelement <4 x float> %2206, float %2188, i64 2, !dbg !212 + %2208 = insertelement <4 x float> %2207, float %2197, i64 3, !dbg !212 + %2209 = fadd <4 x float> %2208, %2204, !dbg !212 + %2210 = insertelement <4 x i64> poison, i64 %2144, i64 0, !dbg !215 + %2211 = shufflevector <4 x i64> %2210, <4 x i64> poison, <4 x i32> zeroinitializer, !dbg !215 + %2212 = icmp samesign ugt <4 x i64> %2211, %2142, !dbg !215 + %2213 = select <4 x i1> %2212, <4 x float> %2209, <4 x float> zeroinitializer, !dbg !216 + %2214 = extractelement <4 x float> %2213, i64 0, !dbg !217 + %2215 = extractelement <4 x float> %2213, i64 1, !dbg !217 + %2216 = select i1 %2129, float %2214, float %2215, !dbg !217 + %2217 = select i1 %2129, float %2215, float %2214, !dbg !217 + %2218 = extractelement <4 x float> %2213, i64 2, !dbg !217 + %2219 = extractelement <4 x float> %2213, i64 3, !dbg !217 + %2220 = select i1 %2129, float %2218, float %2219, !dbg !217 + %2221 = select i1 %2129, float %2219, float %2218, !dbg !217 + %2222 = bitcast float %2216 to i32, !dbg !217 + %2223 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2222, i32 %2135, i32 31), !dbg !217 + %2224 = bitcast float %2220 to i32, !dbg !217 + %2225 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2224, i32 %2135, i32 31), !dbg !217 + %2226 = bitcast float %2217 to i32, !dbg !217 + %2227 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2226, i32 %2136, i32 31), !dbg !217 + %2228 = bitcast float %2221 to i32, !dbg !217 + %2229 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2228, i32 %2136, i32 31), !dbg !217 + %.v171 = select i1 %2129, i32 %2223, i32 %2227, !dbg !217 + %2230 = bitcast i32 %.v171 to float, !dbg !217 + %.v172 = select i1 %2129, i32 %2225, i32 %2229, !dbg !217 + %2231 = bitcast i32 %.v172 to float, !dbg !217 + %2232 = extractelement <4 x float> %2145, i64 0, !dbg !217 + %2233 = fmul float %2232, %2230, !dbg !217 + %2234 = extractelement <4 x float> %2145, i64 1, !dbg !217 + %2235 = fmul float %2234, %2230, !dbg !217 + %2236 = fmul float %.pn76466, %2231, !dbg !217 + %2237 = fmul float %.pn74467, %2231, !dbg !217 + %2238 = fmul float %.pn72468, %2230, !dbg !217 + %2239 = fmul float %.pn70469, %2230, !dbg !217 + %2240 = extractelement <4 x float> %2145, i64 2, !dbg !217 + %2241 = fmul float %2240, %2231, !dbg !217 + %2242 = extractelement <4 x float> %2145, i64 3, !dbg !217 + %2243 = fmul float %2242, %2231, !dbg !217 + %2244 = fadd float %2233, %2236, !dbg !218 + %2245 = fadd float %2235, %2237, !dbg !218 + %2246 = fadd float %2238, %2241, !dbg !218 + %2247 = fadd float %2239, %2243, !dbg !218 + %2248 = bitcast float %2244 to i32, !dbg !219 + %2249 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2248, i32 16, i32 31), !dbg !219 + %2250 = bitcast i32 %2249 to float, !dbg !219 + %2251 = fadd float %2244, %2250, !dbg !218 + %2252 = bitcast float %2251 to i32, !dbg !219 + %2253 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2252, i32 8, i32 31), !dbg !219 + %2254 = bitcast i32 %2253 to float, !dbg !219 + %2255 = fadd float %2251, %2254, !dbg !218 + %2256 = bitcast float %2255 to i32, !dbg !219 + %2257 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2256, i32 4, i32 31), !dbg !219 + %2258 = bitcast float %2245 to i32, !dbg !219 + %2259 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2258, i32 16, i32 31), !dbg !219 + %2260 = bitcast i32 %2259 to float, !dbg !219 + %2261 = fadd float %2245, %2260, !dbg !218 + %2262 = bitcast float %2261 to i32, !dbg !219 + %2263 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2262, i32 8, i32 31), !dbg !219 + %2264 = bitcast i32 %2263 to float, !dbg !219 + %2265 = fadd float %2261, %2264, !dbg !218 + %2266 = bitcast float %2265 to i32, !dbg !219 + %2267 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2266, i32 4, i32 31), !dbg !219 + %2268 = bitcast float %2246 to i32, !dbg !219 + %2269 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2268, i32 16, i32 31), !dbg !219 + %2270 = bitcast i32 %2269 to float, !dbg !219 + %2271 = fadd float %2246, %2270, !dbg !218 + %2272 = bitcast float %2271 to i32, !dbg !219 + %2273 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2272, i32 8, i32 31), !dbg !219 + %2274 = bitcast i32 %2273 to float, !dbg !219 + %2275 = fadd float %2271, %2274, !dbg !218 + %2276 = bitcast float %2275 to i32, !dbg !219 + %2277 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2276, i32 4, i32 31), !dbg !219 + %2278 = bitcast float %2247 to i32, !dbg !219 + %2279 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2278, i32 16, i32 31), !dbg !219 + %2280 = bitcast i32 %2279 to float, !dbg !219 + %2281 = fadd float %2247, %2280, !dbg !218 + %2282 = bitcast float %2281 to i32, !dbg !219 + %2283 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2282, i32 8, i32 31), !dbg !219 + %2284 = bitcast i32 %2283 to float, !dbg !219 + %2285 = fadd float %2281, %2284, !dbg !218 + %2286 = bitcast float %2285 to i32, !dbg !219 + %2287 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2286, i32 4, i32 31), !dbg !219 + %2288 = insertelement <4 x i32> poison, i32 %2257, i64 0, !dbg !219 + %2289 = insertelement <4 x i32> %2288, i32 %2267, i64 1, !dbg !219 + %2290 = insertelement <4 x i32> %2289, i32 %2277, i64 2, !dbg !219 + %2291 = insertelement <4 x i32> %2290, i32 %2287, i64 3, !dbg !219 + %2292 = bitcast <4 x i32> %2291 to <4 x float>, !dbg !219 + %2293 = insertelement <4 x float> poison, float %2255, i64 0, !dbg !218 + %2294 = insertelement <4 x float> %2293, float %2265, i64 1, !dbg !218 + %2295 = insertelement <4 x float> %2294, float %2275, i64 2, !dbg !218 + %2296 = insertelement <4 x float> %2295, float %2285, i64 3, !dbg !218 + %2297 = fadd <4 x float> %2296, %2292, !dbg !218 + %2298 = fadd <4 x float> %2213, %2297, !dbg !221 + %2299 = extractelement <4 x float> %2298, i64 0, !dbg !222 + %2300 = select i1 %2147, float %2299, float %.pn76466, !dbg !222 + %2301 = extractelement <4 x float> %2298, i64 1, !dbg !222 + %2302 = select i1 %2147, float %2301, float %.pn74467, !dbg !222 + %2303 = extractelement <4 x float> %2298, i64 2, !dbg !222 + %2304 = select i1 %2146, float %2303, float %.pn72468, !dbg !222 + %2305 = extractelement <4 x float> %2298, i64 3, !dbg !222 + %2306 = select i1 %2146, float %2305, float %.pn70469, !dbg !222 + %2307 = insertelement <4 x i1> poison, i1 %2146, i64 0, !dbg !222 + %2308 = insertelement <4 x i1> %2307, i1 %2147, i64 1, !dbg !222 + %2309 = shufflevector <4 x i1> %2308, <4 x i1> poison, <4 x i32> , !dbg !222 + %2310 = select <4 x i1> %2309, <4 x float> %2298, <4 x float> %2145, !dbg !222 + %2311 = add nuw nsw i64 %2144, 1, !dbg !209 + %exitcond550.not = icmp eq i64 %2311, %2137, !dbg !209 + br i1 %exitcond550.not, label %._crit_edge474.loopexit, label %2143, !dbg !209 + +._crit_edge474.loopexit: ; preds = %2143 + %2312 = fadd float %2300, 0.000000e+00, !dbg !223 + %2313 = fadd float %2302, 0.000000e+00, !dbg !223 + %2314 = fadd float %2304, 0.000000e+00, !dbg !223 + %2315 = fadd float %2306, 0.000000e+00, !dbg !223 + br label %._crit_edge474, !dbg !224 + +._crit_edge474: ; preds = %._crit_edge474.loopexit, %._crit_edge455 + %.pn76.lcssa = phi float [ %1818, %._crit_edge455 ], [ %2312, %._crit_edge474.loopexit ] + %.pn74.lcssa = phi float [ %1819, %._crit_edge455 ], [ %2313, %._crit_edge474.loopexit ] + %.pn72.lcssa = phi float [ %1820, %._crit_edge455 ], [ %2314, %._crit_edge474.loopexit ] + %.pn70.lcssa = phi float [ %1821, %._crit_edge455 ], [ %2315, %._crit_edge474.loopexit ] + %2316 = phi <4 x float> [ %1822, %._crit_edge455 ], [ %2310, %._crit_edge474.loopexit ] + %2317 = sub nsw i64 %10, %19, !dbg !224 + %2318 = icmp sgt i64 %2317, 2, !dbg !225 + %2319 = and i32 %23, 4 + br i1 %2318, label %.lr.ph492, label %._crit_edge474.._crit_edge493_crit_edge, !dbg !225 + +._crit_edge474.._crit_edge493_crit_edge: ; preds = %._crit_edge474 + %.pre553 = shl nuw nsw i32 %23, 3, !dbg !163 + %.pre555 = and i32 %.pre553, 24, !dbg !163 + %.pre557 = lshr i32 %23, 3, !dbg !163 + %.pre559 = and i32 %.pre557, 3, !dbg !163 + br label %._crit_edge493, !dbg !225 + +.lr.ph492: ; preds = %._crit_edge474 + %2320 = icmp eq i32 %2319, 0 + %2321 = shl nuw nsw i32 %23, 3 + %2322 = and i32 %2321, 24 + %2323 = lshr i32 %23, 3 + %2324 = and i32 %2323, 3 + %2325 = or disjoint i32 %2324, %2322 + %2326 = or disjoint i32 %2325, %2319 + %2327 = xor i32 %2326, 4 + %2328 = tail call i64 @llvm.umin.i64(i64 %2317, i64 16), !dbg !225 + %2329 = extractelement <4 x float> %1827, i64 0, !dbg !226 + %2330 = extractelement <4 x float> %1827, i64 1, !dbg !226 + %2331 = extractelement <4 x float> %1827, i64 2, !dbg !226 + %2332 = extractelement <4 x float> %1827, i64 3, !dbg !226 + %.mask712 = and <4 x i32> %49, splat (i32 255) + %2333 = zext nneg <4 x i32> %.mask712 to <4 x i64> + br label %2334, !dbg !225 + +2334: ; preds = %.lr.ph492, %2334 + %.pn86488 = phi float [ %1826, %.lr.ph492 ], [ %2497, %2334 ] + %.pn88487 = phi float [ %1825, %.lr.ph492 ], [ %2495, %2334 ] + %.pn90486 = phi float [ %1824, %.lr.ph492 ], [ %2493, %2334 ] + %.pn92485 = phi float [ %1823, %.lr.ph492 ], [ %2491, %2334 ] + %2335 = phi i64 [ 2, %.lr.ph492 ], [ %2502, %2334 ] + %2336 = phi <4 x float> [ %1827, %.lr.ph492 ], [ %2501, %2334 ] + %2337 = icmp eq i64 %2335, %85, !dbg !227 + %2338 = icmp eq i64 %2335, %83, !dbg !227 + %2339 = select i1 %2337, float %2329, float 0.000000e+00, !dbg !226 + %2340 = select i1 %2337, float %2330, float 0.000000e+00, !dbg !226 + %2341 = select i1 %2338, float %1823, float 0.000000e+00, !dbg !226 + %2342 = select i1 %2338, float %1824, float 0.000000e+00, !dbg !226 + %2343 = select i1 %2337, float %1825, float 0.000000e+00, !dbg !226 + %2344 = select i1 %2337, float %1826, float 0.000000e+00, !dbg !226 + %2345 = select i1 %2338, float %2331, float 0.000000e+00, !dbg !226 + %2346 = select i1 %2338, float %2332, float 0.000000e+00, !dbg !226 + %2347 = fadd float %2339, %2341, !dbg !228 + %2348 = fadd float %2340, %2342, !dbg !228 + %2349 = fadd float %2343, %2345, !dbg !228 + %2350 = fadd float %2344, %2346, !dbg !228 + %2351 = bitcast float %2347 to i32, !dbg !229 + %2352 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2351, i32 16, i32 31), !dbg !229 + %2353 = bitcast i32 %2352 to float, !dbg !229 + %2354 = fadd float %2347, %2353, !dbg !228 + %2355 = bitcast float %2354 to i32, !dbg !229 + %2356 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2355, i32 8, i32 31), !dbg !229 + %2357 = bitcast i32 %2356 to float, !dbg !229 + %2358 = bitcast float %2348 to i32, !dbg !229 + %2359 = bitcast float %2349 to i32, !dbg !229 + %2360 = bitcast float %2350 to i32, !dbg !229 + %2361 = insertelement <4 x i64> poison, i64 %2335, i64 0, !dbg !231 + %2362 = shufflevector <4 x i64> %2361, <4 x i64> poison, <4 x i32> zeroinitializer, !dbg !231 + %2363 = icmp samesign ugt <4 x i64> %2362, %2333, !dbg !231 + %2364 = fadd float %2354, %2357, !dbg !228 + %2365 = bitcast float %2364 to i32, !dbg !229 + %2366 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2365, i32 4, i32 31), !dbg !229 + %2367 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2358, i32 16, i32 31), !dbg !229 + %2368 = bitcast i32 %2367 to float, !dbg !229 + %2369 = fadd float %2348, %2368, !dbg !228 + %2370 = bitcast float %2369 to i32, !dbg !229 + %2371 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2370, i32 8, i32 31), !dbg !229 + %2372 = bitcast i32 %2371 to float, !dbg !229 + %2373 = fadd float %2369, %2372, !dbg !228 + %2374 = bitcast float %2373 to i32, !dbg !229 + %2375 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2374, i32 4, i32 31), !dbg !229 + %2376 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2359, i32 16, i32 31), !dbg !229 + %2377 = bitcast i32 %2376 to float, !dbg !229 + %2378 = fadd float %2349, %2377, !dbg !228 + %2379 = bitcast float %2378 to i32, !dbg !229 + %2380 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2379, i32 8, i32 31), !dbg !229 + %2381 = bitcast i32 %2380 to float, !dbg !229 + %2382 = fadd float %2378, %2381, !dbg !228 + %2383 = bitcast float %2382 to i32, !dbg !229 + %2384 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2383, i32 4, i32 31), !dbg !229 + %2385 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2360, i32 16, i32 31), !dbg !229 + %2386 = bitcast i32 %2385 to float, !dbg !229 + %2387 = fadd float %2350, %2386, !dbg !228 + %2388 = bitcast float %2387 to i32, !dbg !229 + %2389 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2388, i32 8, i32 31), !dbg !229 + %2390 = bitcast i32 %2389 to float, !dbg !229 + %2391 = fadd float %2387, %2390, !dbg !228 + %2392 = bitcast float %2391 to i32, !dbg !229 + %2393 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2392, i32 4, i32 31), !dbg !229 + %2394 = insertelement <4 x i32> poison, i32 %2366, i64 0, !dbg !229 + %2395 = insertelement <4 x i32> %2394, i32 %2375, i64 1, !dbg !229 + %2396 = insertelement <4 x i32> %2395, i32 %2384, i64 2, !dbg !229 + %2397 = insertelement <4 x i32> %2396, i32 %2393, i64 3, !dbg !229 + %2398 = bitcast <4 x i32> %2397 to <4 x float>, !dbg !229 + %2399 = insertelement <4 x float> poison, float %2364, i64 0, !dbg !228 + %2400 = insertelement <4 x float> %2399, float %2373, i64 1, !dbg !228 + %2401 = insertelement <4 x float> %2400, float %2382, i64 2, !dbg !228 + %2402 = insertelement <4 x float> %2401, float %2391, i64 3, !dbg !228 + %2403 = fadd <4 x float> %2402, %2398, !dbg !228 + %2404 = select <4 x i1> %2363, <4 x float> %2403, <4 x float> zeroinitializer, !dbg !232 + %2405 = extractelement <4 x float> %2404, i64 0, !dbg !233 + %2406 = extractelement <4 x float> %2404, i64 1, !dbg !233 + %2407 = select i1 %2320, float %2405, float %2406, !dbg !233 + %2408 = select i1 %2320, float %2406, float %2405, !dbg !233 + %2409 = extractelement <4 x float> %2404, i64 2, !dbg !233 + %2410 = extractelement <4 x float> %2404, i64 3, !dbg !233 + %2411 = select i1 %2320, float %2409, float %2410, !dbg !233 + %2412 = select i1 %2320, float %2410, float %2409, !dbg !233 + %2413 = bitcast float %2407 to i32, !dbg !233 + %2414 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2413, i32 %2326, i32 31), !dbg !233 + %2415 = bitcast float %2411 to i32, !dbg !233 + %2416 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2415, i32 %2326, i32 31), !dbg !233 + %2417 = bitcast float %2408 to i32, !dbg !233 + %2418 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2417, i32 %2327, i32 31), !dbg !233 + %2419 = bitcast float %2412 to i32, !dbg !233 + %2420 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2419, i32 %2327, i32 31), !dbg !233 + %.v169 = select i1 %2320, i32 %2414, i32 %2418, !dbg !233 + %2421 = bitcast i32 %.v169 to float, !dbg !233 + %.v170 = select i1 %2320, i32 %2416, i32 %2420, !dbg !233 + %2422 = bitcast i32 %.v170 to float, !dbg !233 + %2423 = extractelement <4 x float> %2336, i64 0, !dbg !233 + %2424 = fmul float %2423, %2421, !dbg !233 + %2425 = extractelement <4 x float> %2336, i64 1, !dbg !233 + %2426 = fmul float %2425, %2421, !dbg !233 + %2427 = fmul float %.pn92485, %2422, !dbg !233 + %2428 = fmul float %.pn90486, %2422, !dbg !233 + %2429 = fmul float %.pn88487, %2421, !dbg !233 + %2430 = fmul float %.pn86488, %2421, !dbg !233 + %2431 = extractelement <4 x float> %2336, i64 2, !dbg !233 + %2432 = fmul float %2431, %2422, !dbg !233 + %2433 = extractelement <4 x float> %2336, i64 3, !dbg !233 + %2434 = fmul float %2433, %2422, !dbg !233 + %2435 = fadd float %2424, %2427, !dbg !234 + %2436 = fadd float %2426, %2428, !dbg !234 + %2437 = fadd float %2429, %2432, !dbg !234 + %2438 = fadd float %2430, %2434, !dbg !234 + %2439 = bitcast float %2435 to i32, !dbg !235 + %2440 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2439, i32 16, i32 31), !dbg !235 + %2441 = bitcast i32 %2440 to float, !dbg !235 + %2442 = fadd float %2435, %2441, !dbg !234 + %2443 = bitcast float %2442 to i32, !dbg !235 + %2444 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2443, i32 8, i32 31), !dbg !235 + %2445 = bitcast i32 %2444 to float, !dbg !235 + %2446 = fadd float %2442, %2445, !dbg !234 + %2447 = bitcast float %2446 to i32, !dbg !235 + %2448 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2447, i32 4, i32 31), !dbg !235 + %2449 = bitcast float %2436 to i32, !dbg !235 + %2450 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2449, i32 16, i32 31), !dbg !235 + %2451 = bitcast i32 %2450 to float, !dbg !235 + %2452 = fadd float %2436, %2451, !dbg !234 + %2453 = bitcast float %2452 to i32, !dbg !235 + %2454 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2453, i32 8, i32 31), !dbg !235 + %2455 = bitcast i32 %2454 to float, !dbg !235 + %2456 = fadd float %2452, %2455, !dbg !234 + %2457 = bitcast float %2456 to i32, !dbg !235 + %2458 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2457, i32 4, i32 31), !dbg !235 + %2459 = bitcast float %2437 to i32, !dbg !235 + %2460 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2459, i32 16, i32 31), !dbg !235 + %2461 = bitcast i32 %2460 to float, !dbg !235 + %2462 = fadd float %2437, %2461, !dbg !234 + %2463 = bitcast float %2462 to i32, !dbg !235 + %2464 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2463, i32 8, i32 31), !dbg !235 + %2465 = bitcast i32 %2464 to float, !dbg !235 + %2466 = fadd float %2462, %2465, !dbg !234 + %2467 = bitcast float %2466 to i32, !dbg !235 + %2468 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2467, i32 4, i32 31), !dbg !235 + %2469 = bitcast float %2438 to i32, !dbg !235 + %2470 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2469, i32 16, i32 31), !dbg !235 + %2471 = bitcast i32 %2470 to float, !dbg !235 + %2472 = fadd float %2438, %2471, !dbg !234 + %2473 = bitcast float %2472 to i32, !dbg !235 + %2474 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2473, i32 8, i32 31), !dbg !235 + %2475 = bitcast i32 %2474 to float, !dbg !235 + %2476 = fadd float %2472, %2475, !dbg !234 + %2477 = bitcast float %2476 to i32, !dbg !235 + %2478 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %2477, i32 4, i32 31), !dbg !235 + %2479 = insertelement <4 x i32> poison, i32 %2448, i64 0, !dbg !235 + %2480 = insertelement <4 x i32> %2479, i32 %2458, i64 1, !dbg !235 + %2481 = insertelement <4 x i32> %2480, i32 %2468, i64 2, !dbg !235 + %2482 = insertelement <4 x i32> %2481, i32 %2478, i64 3, !dbg !235 + %2483 = bitcast <4 x i32> %2482 to <4 x float>, !dbg !235 + %2484 = insertelement <4 x float> poison, float %2446, i64 0, !dbg !234 + %2485 = insertelement <4 x float> %2484, float %2456, i64 1, !dbg !234 + %2486 = insertelement <4 x float> %2485, float %2466, i64 2, !dbg !234 + %2487 = insertelement <4 x float> %2486, float %2476, i64 3, !dbg !234 + %2488 = fadd <4 x float> %2487, %2483, !dbg !234 + %2489 = fadd <4 x float> %2404, %2488, !dbg !237 + %2490 = extractelement <4 x float> %2489, i64 0, !dbg !238 + %2491 = select i1 %2338, float %2490, float %.pn92485, !dbg !238 + %2492 = extractelement <4 x float> %2489, i64 1, !dbg !238 + %2493 = select i1 %2338, float %2492, float %.pn90486, !dbg !238 + %2494 = extractelement <4 x float> %2489, i64 2, !dbg !238 + %2495 = select i1 %2337, float %2494, float %.pn88487, !dbg !238 + %2496 = extractelement <4 x float> %2489, i64 3, !dbg !238 + %2497 = select i1 %2337, float %2496, float %.pn86488, !dbg !238 + %2498 = insertelement <4 x i1> poison, i1 %2337, i64 0, !dbg !238 + %2499 = insertelement <4 x i1> %2498, i1 %2338, i64 1, !dbg !238 + %2500 = shufflevector <4 x i1> %2499, <4 x i1> poison, <4 x i32> , !dbg !238 + %2501 = select <4 x i1> %2500, <4 x float> %2489, <4 x float> %2336, !dbg !238 + %2502 = add nuw nsw i64 %2335, 1, !dbg !225 + %exitcond552.not = icmp eq i64 %2502, %2328, !dbg !225 + br i1 %exitcond552.not, label %._crit_edge493.loopexit, label %2334, !dbg !225 + +._crit_edge493.loopexit: ; preds = %2334 + %2503 = fadd float %2491, 0.000000e+00, !dbg !239 + %2504 = fadd float %2493, 0.000000e+00, !dbg !239 + %2505 = fadd float %2495, 0.000000e+00, !dbg !239 + %2506 = fadd float %2497, 0.000000e+00, !dbg !239 + br label %._crit_edge493, !dbg !240 + +._crit_edge493: ; preds = %._crit_edge474.._crit_edge493_crit_edge, %._crit_edge493.loopexit + %.pre-phi560 = phi i32 [ %.pre559, %._crit_edge474.._crit_edge493_crit_edge ], [ %2324, %._crit_edge493.loopexit ], !dbg !163 + %.pre-phi556 = phi i32 [ %.pre555, %._crit_edge474.._crit_edge493_crit_edge ], [ %2322, %._crit_edge493.loopexit ], !dbg !163 + %.pn92.lcssa = phi float [ %1823, %._crit_edge474.._crit_edge493_crit_edge ], [ %2503, %._crit_edge493.loopexit ] + %.pn90.lcssa = phi float [ %1824, %._crit_edge474.._crit_edge493_crit_edge ], [ %2504, %._crit_edge493.loopexit ] + %.pn88.lcssa = phi float [ %1825, %._crit_edge474.._crit_edge493_crit_edge ], [ %2505, %._crit_edge493.loopexit ] + %.pn86.lcssa = phi float [ %1826, %._crit_edge474.._crit_edge493_crit_edge ], [ %2506, %._crit_edge493.loopexit ] + %2507 = phi <4 x float> [ %1827, %._crit_edge474.._crit_edge493_crit_edge ], [ %2501, %._crit_edge493.loopexit ] + %2508 = uitofp i1 %1095 to float, !dbg !240 + %2509 = uitofp i1 %1096 to float, !dbg !240 + %2510 = uitofp i1 %1097 to float, !dbg !240 + %2511 = uitofp i1 %1098 to float, !dbg !240 + %2512 = uitofp i1 %1099 to float, !dbg !240 + %2513 = uitofp i1 %1100 to float, !dbg !240 + %2514 = uitofp i1 %1101 to float, !dbg !240 + %2515 = extractelement <4 x float> %1934, i64 0, !dbg !240 + %2516 = fadd float %2515, %2511, !dbg !240 + %2517 = extractelement <4 x float> %1934, i64 1, !dbg !240 + %2518 = fadd float %2517, %2512, !dbg !240 + %2519 = extractelement <4 x float> %1934, i64 2, !dbg !240 + %2520 = fadd float %2519, %2513, !dbg !240 + %2521 = extractelement <4 x float> %1934, i64 3, !dbg !240 + %2522 = fadd float %2521, %2514, !dbg !240 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !240 + %2523 = and i32 %230, 320, !dbg !240 + %2524 = shl nuw nsw i32 %28, 3, !dbg !240 + %2525 = and i32 %23, 8, !dbg !240 + %2526 = icmp eq i32 %2525, 0, !dbg !240 + %2527 = select i1 %2526, i32 0, i32 160, !dbg !240 + %2528 = xor i32 %2527, %29, !dbg !240 + %2529 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %2523, !dbg !240 + %2530 = getelementptr inbounds nuw i8, ptr addrspace(3) %2529, i32 %2524, !dbg !240 + %2531 = getelementptr inbounds nuw i8, ptr addrspace(3) %2530, i32 %2528, !dbg !240 + %2532 = bitcast float %2516 to i32, !dbg !240 + %2533 = bitcast float %2518 to i32, !dbg !240 + tail call void asm sideeffect "st.shared.v2.b32 [ $0 + 0 ], { $1, $2 };", "r,r,r"(ptr addrspace(3) %2531, i32 %2532, i32 %2533) #2, !dbg !240 + %2534 = getelementptr inbounds nuw i8, ptr addrspace(3) %2531, i32 512, !dbg !240 + %2535 = bitcast float %2520 to i32, !dbg !240 + %2536 = bitcast float %2522 to i32, !dbg !240 + tail call void asm sideeffect "st.shared.v2.b32 [ $0 + 0 ], { $1, $2 };", "r,r,r"(ptr addrspace(3) nonnull %2534, i32 %2535, i32 %2536) #2, !dbg !240 + %2537 = extractelement <4 x float> %2125, i64 0, !dbg !207 + %2538 = fadd float %2537, %2508, !dbg !207 + %2539 = extractelement <4 x float> %2125, i64 1, !dbg !207 + %2540 = fadd float %2539, %2509, !dbg !207 + %2541 = extractelement <4 x float> %2125, i64 2, !dbg !207 + %2542 = fadd float %2541, %2508, !dbg !207 + %2543 = extractelement <4 x float> %2125, i64 3, !dbg !207 + %2544 = fadd float %2543, %2510, !dbg !207 + %2545 = extractelement <4 x float> %2316, i64 0, !dbg !223 + %2546 = fadd float %2545, %2508, !dbg !223 + %2547 = extractelement <4 x float> %2316, i64 1, !dbg !223 + %2548 = fadd float %2547, %2509, !dbg !223 + %2549 = extractelement <4 x float> %2316, i64 2, !dbg !223 + %2550 = fadd float %2549, %2508, !dbg !223 + %2551 = extractelement <4 x float> %2316, i64 3, !dbg !223 + %2552 = fadd float %2551, %2510, !dbg !223 + %2553 = extractelement <4 x float> %2507, i64 0, !dbg !239 + %2554 = fadd float %2553, %2508, !dbg !239 + %2555 = extractelement <4 x float> %2507, i64 1, !dbg !239 + %2556 = fadd float %2555, %2509, !dbg !239 + %2557 = extractelement <4 x float> %2507, i64 2, !dbg !239 + %2558 = fadd float %2557, %2508, !dbg !239 + %2559 = extractelement <4 x float> %2507, i64 3, !dbg !239 + %2560 = fadd float %2559, %2510, !dbg !239 + %2561 = icmp eq i32 %2319, 0, !dbg !163 + %2562 = select i1 %2561, float %1764, float %1765, !dbg !163 + %2563 = select i1 %2561, float %1765, float %1764, !dbg !163 + %2564 = select i1 %2561, float %1766, float %1767, !dbg !163 + %2565 = select i1 %2561, float %1767, float %1766, !dbg !163 + %2566 = select i1 %2561, float %1768, float %1769, !dbg !163 + %2567 = select i1 %2561, float %1769, float %1768, !dbg !163 + %2568 = select i1 %2561, float %1770, float %1771, !dbg !163 + %2569 = select i1 %2561, float %1771, float %1770, !dbg !163 + %2570 = or disjoint i32 %2319, %.pre-phi556, !dbg !163 + %2571 = or disjoint i32 %2570, %.pre-phi560, !dbg !163 + %2572 = bitcast float %2562 to i32, !dbg !163 + %2573 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2572, i32 %2571, i32 31), !dbg !163 + %2574 = xor i32 %2571, 4, !dbg !163 + %2575 = bitcast float %2563 to i32, !dbg !163 + %2576 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2575, i32 %2574, i32 31), !dbg !163 + %2577 = bitcast float %2564 to i32, !dbg !163 + %2578 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2577, i32 %2571, i32 31), !dbg !163 + %2579 = bitcast float %2565 to i32, !dbg !163 + %2580 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2579, i32 %2574, i32 31), !dbg !163 + %2581 = bitcast float %2566 to i32, !dbg !163 + %2582 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2581, i32 %2571, i32 31), !dbg !163 + %2583 = bitcast float %2567 to i32, !dbg !163 + %2584 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2583, i32 %2574, i32 31), !dbg !163 + %2585 = bitcast float %2568 to i32, !dbg !163 + %2586 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2585, i32 %2571, i32 31), !dbg !163 + %2587 = bitcast float %2569 to i32, !dbg !163 + %2588 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2587, i32 %2574, i32 31), !dbg !163 + %.v = select i1 %2561, i32 %2576, i32 %2573, !dbg !163 + %.v97 = select i1 %2561, i32 %2580, i32 %2578, !dbg !163 + %.v98 = select i1 %2561, i32 %2584, i32 %2582, !dbg !163 + %.v99 = select i1 %2561, i32 %2588, i32 %2586, !dbg !163 + %2589 = bitcast float %2538 to i32, !dbg !241 + %2590 = bitcast float %.pn60.lcssa to i32, !dbg !241 + %2591 = bitcast float %.pn56.lcssa to i32, !dbg !241 + %2592 = bitcast float %2542 to i32, !dbg !241 + %2593 = bitcast float %2540 to i32, !dbg !241 + %2594 = bitcast float %.pn58.lcssa to i32, !dbg !241 + %2595 = bitcast float %.pn54.lcssa to i32, !dbg !241 + %2596 = bitcast float %2544 to i32, !dbg !241 + %2597 = select i1 %2561, i32 %2573, i32 %2576, !dbg !163 + %2598 = select i1 %2561, i32 %2578, i32 %2580, !dbg !163 + %2599 = select i1 %2561, i32 %2582, i32 %2584, !dbg !163 + %2600 = select i1 %2561, i32 %2586, i32 %2588, !dbg !163 + %2601 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2589, i32 %2590, i32 %2591, i32 %2592, i32 %2597, i32 %2598) #2, !dbg !241 + %2602 = extractvalue { float, float, float, float } %2601, 0, !dbg !241 + %2603 = extractvalue { float, float, float, float } %2601, 1, !dbg !241 + %2604 = extractvalue { float, float, float, float } %2601, 2, !dbg !241 + %2605 = extractvalue { float, float, float, float } %2601, 3, !dbg !241 + %2606 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2589, i32 %2590, i32 %2591, i32 %2592, i32 %2599, i32 %2600) #2, !dbg !241 + %2607 = extractvalue { float, float, float, float } %2606, 0, !dbg !241 + %2608 = extractvalue { float, float, float, float } %2606, 1, !dbg !241 + %2609 = extractvalue { float, float, float, float } %2606, 2, !dbg !241 + %2610 = extractvalue { float, float, float, float } %2606, 3, !dbg !241 + %2611 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2602, float %2603, float %2604, float %2605, i32 %2593, i32 %2594, i32 %2595, i32 %2596, i32 %.v, i32 %.v97) #2, !dbg !241 + %2612 = extractvalue { float, float, float, float } %2611, 0, !dbg !241 + %2613 = extractvalue { float, float, float, float } %2611, 1, !dbg !241 + %2614 = extractvalue { float, float, float, float } %2611, 2, !dbg !241 + %2615 = extractvalue { float, float, float, float } %2611, 3, !dbg !241 + %2616 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2607, float %2608, float %2609, float %2610, i32 %2593, i32 %2594, i32 %2595, i32 %2596, i32 %.v98, i32 %.v99) #2, !dbg !241 + %2617 = extractvalue { float, float, float, float } %2616, 0, !dbg !241 + %2618 = extractvalue { float, float, float, float } %2616, 1, !dbg !241 + %2619 = extractvalue { float, float, float, float } %2616, 2, !dbg !241 + %2620 = extractvalue { float, float, float, float } %2616, 3, !dbg !241 + %2621 = and i32 %23, 2, !dbg !241 + %.not100 = icmp eq i32 %2621, 0, !dbg !241 + %2622 = select i1 %.not100, float %2612, float %2613, !dbg !241 + %2623 = select i1 %.not100, float %2613, float %2612, !dbg !241 + %2624 = select i1 %.not100, float %2614, float %2615, !dbg !241 + %2625 = select i1 %.not100, float %2615, float %2614, !dbg !241 + %2626 = select i1 %.not100, float %2617, float %2618, !dbg !241 + %2627 = select i1 %.not100, float %2618, float %2617, !dbg !241 + %2628 = select i1 %.not100, float %2619, float %2620, !dbg !241 + %2629 = select i1 %.not100, float %2620, float %2619, !dbg !241 + %2630 = and i32 %23, 1, !dbg !241 + %2631 = icmp eq i32 %2630, 0, !dbg !241 + %trunc = trunc i32 %23 to i2, !dbg !241 + %rev = tail call i2 @llvm.bitreverse.i2(i2 %trunc), !dbg !241 + %2632 = zext i2 %rev to i32, !dbg !241 + %2633 = or disjoint i32 %26, %2632, !dbg !241 + %2634 = bitcast float %2622 to i32, !dbg !241 + %2635 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2634, i32 %2633, i32 31), !dbg !241 + %2636 = bitcast float %2624 to i32, !dbg !241 + %2637 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2636, i32 %2633, i32 31), !dbg !241 + %2638 = xor i32 %2633, 2, !dbg !241 + %2639 = bitcast float %2623 to i32, !dbg !241 + %2640 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2639, i32 %2638, i32 31), !dbg !241 + %2641 = bitcast float %2625 to i32, !dbg !241 + %2642 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2641, i32 %2638, i32 31), !dbg !241 + %2643 = bitcast float %2626 to i32, !dbg !241 + %2644 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2643, i32 %2633, i32 31), !dbg !241 + %2645 = bitcast float %2628 to i32, !dbg !241 + %2646 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2645, i32 %2633, i32 31), !dbg !241 + %2647 = bitcast float %2627 to i32, !dbg !241 + %2648 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2647, i32 %2638, i32 31), !dbg !241 + %2649 = bitcast float %2629 to i32, !dbg !241 + %2650 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2649, i32 %2638, i32 31), !dbg !241 + %.v101 = select i1 %2631, i32 %2640, i32 %2635, !dbg !241 + %.v102 = select i1 %2631, i32 %2642, i32 %2637, !dbg !241 + %.v103 = select i1 %2631, i32 %2648, i32 %2644, !dbg !241 + %.v104 = select i1 %2631, i32 %2650, i32 %2646, !dbg !241 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !240 + %2651 = shl nuw nsw i32 %2630, 6, !dbg !240 + %2652 = select i1 %.not100, i32 0, i32 160, !dbg !240 + %2653 = or disjoint i32 %2652, %2651, !dbg !240 + %2654 = or disjoint i32 %2653, %26, !dbg !240 + %2655 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %2654, !dbg !240 + %2656 = load i32, ptr addrspace(3) %2655, align 4, !dbg !240 + %2657 = getelementptr inbounds nuw i8, ptr addrspace(3) %2655, i32 256, !dbg !240 + %2658 = load i32, ptr addrspace(3) %2657, align 4, !dbg !240 + %2659 = getelementptr inbounds nuw i8, ptr addrspace(3) %2655, i32 512, !dbg !240 + %2660 = load i32, ptr addrspace(3) %2659, align 4, !dbg !240 + %2661 = getelementptr inbounds nuw i8, ptr addrspace(3) %2655, i32 768, !dbg !240 + %2662 = load i32, ptr addrspace(3) %2661, align 4, !dbg !240 + %2663 = xor i32 %2654, 32, !dbg !240 + %2664 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %2663, !dbg !240 + %2665 = load i32, ptr addrspace(3) %2664, align 4, !dbg !240 + %2666 = getelementptr inbounds nuw i8, ptr addrspace(3) %2664, i32 256, !dbg !240 + %2667 = load i32, ptr addrspace(3) %2666, align 4, !dbg !240 + %2668 = getelementptr inbounds nuw i8, ptr addrspace(3) %2664, i32 512, !dbg !240 + %2669 = load i32, ptr addrspace(3) %2668, align 4, !dbg !240 + %2670 = getelementptr inbounds nuw i8, ptr addrspace(3) %2664, i32 768, !dbg !240 + %2671 = load i32, ptr addrspace(3) %2670, align 4, !dbg !240 + %2672 = select i1 %2631, i32 %2635, i32 %2640, !dbg !241 + %2673 = select i1 %2631, i32 %2637, i32 %2642, !dbg !241 + %2674 = select i1 %2631, i32 %2644, i32 %2648, !dbg !241 + %2675 = select i1 %2631, i32 %2646, i32 %2650, !dbg !241 + %2676 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2672, i32 %2673, i32 %.v101, i32 %.v102, i32 %2656, i32 %2658) #2, !dbg !242 + %2677 = extractvalue { float, float, float, float } %2676, 0, !dbg !242 + %2678 = extractvalue { float, float, float, float } %2676, 1, !dbg !242 + %2679 = extractvalue { float, float, float, float } %2676, 2, !dbg !242 + %2680 = extractvalue { float, float, float, float } %2676, 3, !dbg !242 + %2681 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2672, i32 %2673, i32 %.v101, i32 %.v102, i32 %2665, i32 %2667) #2, !dbg !242 + %2682 = extractvalue { float, float, float, float } %2681, 0, !dbg !242 + %2683 = extractvalue { float, float, float, float } %2681, 1, !dbg !242 + %2684 = extractvalue { float, float, float, float } %2681, 2, !dbg !242 + %2685 = extractvalue { float, float, float, float } %2681, 3, !dbg !242 + %2686 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2677, float %2678, float %2679, float %2680, i32 %2674, i32 %2675, i32 %.v103, i32 %.v104, i32 %2660, i32 %2662) #2, !dbg !242 + %2687 = extractvalue { float, float, float, float } %2686, 0, !dbg !242 + %2688 = extractvalue { float, float, float, float } %2686, 1, !dbg !242 + %2689 = extractvalue { float, float, float, float } %2686, 2, !dbg !242 + %2690 = extractvalue { float, float, float, float } %2686, 3, !dbg !242 + %2691 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2682, float %2683, float %2684, float %2685, i32 %2674, i32 %2675, i32 %.v103, i32 %.v104, i32 %2669, i32 %2671) #2, !dbg !242 + %2692 = extractvalue { float, float, float, float } %2691, 0, !dbg !242 + %2693 = extractvalue { float, float, float, float } %2691, 1, !dbg !242 + %2694 = extractvalue { float, float, float, float } %2691, 2, !dbg !242 + %2695 = extractvalue { float, float, float, float } %2691, 3, !dbg !242 + %2696 = fsub float 0.000000e+00, %2687, !dbg !243 + %2697 = fsub float 0.000000e+00, %2688, !dbg !243 + %2698 = fsub float 0.000000e+00, %2689, !dbg !243 + %2699 = fsub float 0.000000e+00, %2690, !dbg !243 + %2700 = fsub float 0.000000e+00, %2692, !dbg !243 + %2701 = fsub float 0.000000e+00, %2693, !dbg !243 + %2702 = fsub float 0.000000e+00, %2694, !dbg !243 + %2703 = fsub float 0.000000e+00, %2695, !dbg !243 + %2704 = select i1 %2561, float %1780, float %1781, !dbg !165 + %2705 = select i1 %2561, float %1781, float %1780, !dbg !165 + %2706 = select i1 %2561, float %1782, float %1783, !dbg !165 + %2707 = select i1 %2561, float %1783, float %1782, !dbg !165 + %2708 = select i1 %2561, float %1784, float %1785, !dbg !165 + %2709 = select i1 %2561, float %1785, float %1784, !dbg !165 + %2710 = select i1 %2561, float %1786, float %1787, !dbg !165 + %2711 = select i1 %2561, float %1787, float %1786, !dbg !165 + %2712 = bitcast float %2704 to i32, !dbg !165 + %2713 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2712, i32 %2571, i32 31), !dbg !165 + %2714 = bitcast float %2705 to i32, !dbg !165 + %2715 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2714, i32 %2574, i32 31), !dbg !165 + %2716 = bitcast float %2706 to i32, !dbg !165 + %2717 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2716, i32 %2571, i32 31), !dbg !165 + %2718 = bitcast float %2707 to i32, !dbg !165 + %2719 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2718, i32 %2574, i32 31), !dbg !165 + %2720 = bitcast float %2708 to i32, !dbg !165 + %2721 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2720, i32 %2571, i32 31), !dbg !165 + %2722 = bitcast float %2709 to i32, !dbg !165 + %2723 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2722, i32 %2574, i32 31), !dbg !165 + %2724 = bitcast float %2710 to i32, !dbg !165 + %2725 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2724, i32 %2571, i32 31), !dbg !165 + %2726 = bitcast float %2711 to i32, !dbg !165 + %2727 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2726, i32 %2574, i32 31), !dbg !165 + %.v105 = select i1 %2561, i32 %2715, i32 %2713, !dbg !165 + %.v106 = select i1 %2561, i32 %2719, i32 %2717, !dbg !165 + %.v107 = select i1 %2561, i32 %2723, i32 %2721, !dbg !165 + %.v108 = select i1 %2561, i32 %2727, i32 %2725, !dbg !165 + %2728 = bitcast float %2546 to i32, !dbg !244 + %2729 = bitcast float %.pn76.lcssa to i32, !dbg !244 + %2730 = bitcast float %.pn72.lcssa to i32, !dbg !244 + %2731 = bitcast float %2550 to i32, !dbg !244 + %2732 = bitcast float %2548 to i32, !dbg !244 + %2733 = bitcast float %.pn74.lcssa to i32, !dbg !244 + %2734 = bitcast float %.pn70.lcssa to i32, !dbg !244 + %2735 = bitcast float %2552 to i32, !dbg !244 + %2736 = select i1 %2561, i32 %2713, i32 %2715, !dbg !165 + %2737 = select i1 %2561, i32 %2717, i32 %2719, !dbg !165 + %2738 = select i1 %2561, i32 %2721, i32 %2723, !dbg !165 + %2739 = select i1 %2561, i32 %2725, i32 %2727, !dbg !165 + %2740 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2728, i32 %2729, i32 %2730, i32 %2731, i32 %2736, i32 %2737) #2, !dbg !244 + %2741 = extractvalue { float, float, float, float } %2740, 0, !dbg !244 + %2742 = extractvalue { float, float, float, float } %2740, 1, !dbg !244 + %2743 = extractvalue { float, float, float, float } %2740, 2, !dbg !244 + %2744 = extractvalue { float, float, float, float } %2740, 3, !dbg !244 + %2745 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2728, i32 %2729, i32 %2730, i32 %2731, i32 %2738, i32 %2739) #2, !dbg !244 + %2746 = extractvalue { float, float, float, float } %2745, 0, !dbg !244 + %2747 = extractvalue { float, float, float, float } %2745, 1, !dbg !244 + %2748 = extractvalue { float, float, float, float } %2745, 2, !dbg !244 + %2749 = extractvalue { float, float, float, float } %2745, 3, !dbg !244 + %2750 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2741, float %2742, float %2743, float %2744, i32 %2732, i32 %2733, i32 %2734, i32 %2735, i32 %.v105, i32 %.v106) #2, !dbg !244 + %2751 = extractvalue { float, float, float, float } %2750, 0, !dbg !244 + %2752 = extractvalue { float, float, float, float } %2750, 1, !dbg !244 + %2753 = extractvalue { float, float, float, float } %2750, 2, !dbg !244 + %2754 = extractvalue { float, float, float, float } %2750, 3, !dbg !244 + %2755 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2746, float %2747, float %2748, float %2749, i32 %2732, i32 %2733, i32 %2734, i32 %2735, i32 %.v107, i32 %.v108) #2, !dbg !244 + %2756 = extractvalue { float, float, float, float } %2755, 0, !dbg !244 + %2757 = extractvalue { float, float, float, float } %2755, 1, !dbg !244 + %2758 = extractvalue { float, float, float, float } %2755, 2, !dbg !244 + %2759 = extractvalue { float, float, float, float } %2755, 3, !dbg !244 + %2760 = select i1 %.not100, float %2751, float %2752, !dbg !244 + %2761 = select i1 %.not100, float %2752, float %2751, !dbg !244 + %2762 = select i1 %.not100, float %2753, float %2754, !dbg !244 + %2763 = select i1 %.not100, float %2754, float %2753, !dbg !244 + %2764 = select i1 %.not100, float %2756, float %2757, !dbg !244 + %2765 = select i1 %.not100, float %2757, float %2756, !dbg !244 + %2766 = select i1 %.not100, float %2758, float %2759, !dbg !244 + %2767 = select i1 %.not100, float %2759, float %2758, !dbg !244 + %2768 = bitcast float %2760 to i32, !dbg !244 + %2769 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2768, i32 %2633, i32 31), !dbg !244 + %2770 = bitcast float %2762 to i32, !dbg !244 + %2771 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2770, i32 %2633, i32 31), !dbg !244 + %2772 = bitcast float %2761 to i32, !dbg !244 + %2773 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2772, i32 %2638, i32 31), !dbg !244 + %2774 = bitcast float %2763 to i32, !dbg !244 + %2775 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2774, i32 %2638, i32 31), !dbg !244 + %2776 = bitcast float %2764 to i32, !dbg !244 + %2777 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2776, i32 %2633, i32 31), !dbg !244 + %2778 = bitcast float %2766 to i32, !dbg !244 + %2779 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2778, i32 %2633, i32 31), !dbg !244 + %2780 = bitcast float %2765 to i32, !dbg !244 + %2781 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2780, i32 %2638, i32 31), !dbg !244 + %2782 = bitcast float %2767 to i32, !dbg !244 + %2783 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2782, i32 %2638, i32 31), !dbg !244 + %.v109 = select i1 %2631, i32 %2773, i32 %2769, !dbg !244 + %.v110 = select i1 %2631, i32 %2775, i32 %2771, !dbg !244 + %.v111 = select i1 %2631, i32 %2781, i32 %2777, !dbg !244 + %.v112 = select i1 %2631, i32 %2783, i32 %2779, !dbg !244 + %2784 = icmp eq i32 %255, 0, !dbg !207 + %2785 = select i1 %2784, float %2538, float %2540, !dbg !207 + %2786 = select i1 %2784, float %2540, float %2538, !dbg !207 + %2787 = select i1 %2784, float %.pn60.lcssa, float %.pn58.lcssa, !dbg !207 + %2788 = select i1 %2784, float %.pn58.lcssa, float %.pn60.lcssa, !dbg !207 + %2789 = select i1 %2784, float %.pn56.lcssa, float %.pn54.lcssa, !dbg !207 + %2790 = select i1 %2784, float %.pn54.lcssa, float %.pn56.lcssa, !dbg !207 + %2791 = select i1 %2784, float %2542, float %2544, !dbg !207 + %2792 = select i1 %2784, float %2544, float %2542, !dbg !207 + %2793 = shl nuw nsw i32 %23, 2, !dbg !207 + %2794 = and i32 %2793, 28, !dbg !207 + %2795 = or disjoint i32 %2794, %.pre-phi560, !dbg !207 + %2796 = bitcast float %2785 to i32, !dbg !207 + %2797 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2796, i32 %2795, i32 31), !dbg !207 + %2798 = xor i32 %2795, 16, !dbg !207 + %2799 = bitcast float %2786 to i32, !dbg !207 + %2800 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2799, i32 %2798, i32 31), !dbg !207 + %2801 = bitcast float %2787 to i32, !dbg !207 + %2802 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2801, i32 %2795, i32 31), !dbg !207 + %2803 = bitcast float %2788 to i32, !dbg !207 + %2804 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2803, i32 %2798, i32 31), !dbg !207 + %2805 = bitcast float %2789 to i32, !dbg !207 + %2806 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2805, i32 %2795, i32 31), !dbg !207 + %2807 = bitcast float %2790 to i32, !dbg !207 + %2808 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2807, i32 %2798, i32 31), !dbg !207 + %2809 = bitcast float %2791 to i32, !dbg !207 + %2810 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2809, i32 %2795, i32 31), !dbg !207 + %2811 = bitcast float %2792 to i32, !dbg !207 + %2812 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2811, i32 %2798, i32 31), !dbg !207 + %.v113 = select i1 %2561, i32 %2800, i32 %2797, !dbg !207 + %.v114 = select i1 %2561, i32 %2804, i32 %2802, !dbg !207 + %.v115 = select i1 %2561, i32 %2808, i32 %2806, !dbg !207 + %.v116 = select i1 %2561, i32 %2812, i32 %2810, !dbg !207 + %2813 = select i1 %2631, i32 %2769, i32 %2773, !dbg !244 + %2814 = select i1 %2631, i32 %2771, i32 %2775, !dbg !244 + %2815 = select i1 %2631, i32 %2777, i32 %2781, !dbg !244 + %2816 = select i1 %2631, i32 %2779, i32 %2783, !dbg !244 + %2817 = select i1 %2561, i32 %2797, i32 %2800, !dbg !207 + %2818 = select i1 %2561, i32 %2802, i32 %2804, !dbg !207 + %2819 = select i1 %2561, i32 %2806, i32 %2808, !dbg !207 + %2820 = select i1 %2561, i32 %2810, i32 %2812, !dbg !207 + %2821 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2813, i32 %2814, i32 %.v109, i32 %.v110, i32 %2817, i32 %.v113) #2, !dbg !245 + %2822 = extractvalue { float, float, float, float } %2821, 0, !dbg !245 + %2823 = extractvalue { float, float, float, float } %2821, 1, !dbg !245 + %2824 = extractvalue { float, float, float, float } %2821, 2, !dbg !245 + %2825 = extractvalue { float, float, float, float } %2821, 3, !dbg !245 + %2826 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2813, i32 %2814, i32 %.v109, i32 %.v110, i32 %2819, i32 %.v115) #2, !dbg !245 + %2827 = extractvalue { float, float, float, float } %2826, 0, !dbg !245 + %2828 = extractvalue { float, float, float, float } %2826, 1, !dbg !245 + %2829 = extractvalue { float, float, float, float } %2826, 2, !dbg !245 + %2830 = extractvalue { float, float, float, float } %2826, 3, !dbg !245 + %2831 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2822, float %2823, float %2824, float %2825, i32 %2815, i32 %2816, i32 %.v111, i32 %.v112, i32 %2818, i32 %.v114) #2, !dbg !245 + %2832 = extractvalue { float, float, float, float } %2831, 0, !dbg !245 + %2833 = extractvalue { float, float, float, float } %2831, 1, !dbg !245 + %2834 = extractvalue { float, float, float, float } %2831, 2, !dbg !245 + %2835 = extractvalue { float, float, float, float } %2831, 3, !dbg !245 + %2836 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2827, float %2828, float %2829, float %2830, i32 %2815, i32 %2816, i32 %.v111, i32 %.v112, i32 %2820, i32 %.v116) #2, !dbg !245 + %2837 = extractvalue { float, float, float, float } %2836, 0, !dbg !245 + %2838 = extractvalue { float, float, float, float } %2836, 1, !dbg !245 + %2839 = extractvalue { float, float, float, float } %2836, 2, !dbg !245 + %2840 = extractvalue { float, float, float, float } %2836, 3, !dbg !245 + %2841 = fsub float 0.000000e+00, %2832, !dbg !246 + %2842 = fsub float 0.000000e+00, %2833, !dbg !246 + %2843 = fsub float 0.000000e+00, %2834, !dbg !246 + %2844 = fsub float 0.000000e+00, %2835, !dbg !246 + %2845 = fsub float 0.000000e+00, %2837, !dbg !246 + %2846 = fsub float 0.000000e+00, %2838, !dbg !246 + %2847 = fsub float 0.000000e+00, %2839, !dbg !246 + %2848 = fsub float 0.000000e+00, %2840, !dbg !246 + %2849 = select i1 %2561, float %1804, float %1805, !dbg !167 + %2850 = select i1 %2561, float %1805, float %1804, !dbg !167 + %2851 = select i1 %2561, float %1806, float %1807, !dbg !167 + %2852 = select i1 %2561, float %1807, float %1806, !dbg !167 + %2853 = select i1 %2561, float %1808, float %1809, !dbg !167 + %2854 = select i1 %2561, float %1809, float %1808, !dbg !167 + %2855 = select i1 %2561, float %1810, float %1811, !dbg !167 + %2856 = select i1 %2561, float %1811, float %1810, !dbg !167 + %2857 = bitcast float %2849 to i32, !dbg !167 + %2858 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2857, i32 %2571, i32 31), !dbg !167 + %2859 = bitcast float %2850 to i32, !dbg !167 + %2860 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2859, i32 %2574, i32 31), !dbg !167 + %2861 = bitcast float %2851 to i32, !dbg !167 + %2862 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2861, i32 %2571, i32 31), !dbg !167 + %2863 = bitcast float %2852 to i32, !dbg !167 + %2864 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2863, i32 %2574, i32 31), !dbg !167 + %2865 = bitcast float %2853 to i32, !dbg !167 + %2866 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2865, i32 %2571, i32 31), !dbg !167 + %2867 = bitcast float %2854 to i32, !dbg !167 + %2868 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2867, i32 %2574, i32 31), !dbg !167 + %2869 = bitcast float %2855 to i32, !dbg !167 + %2870 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2869, i32 %2571, i32 31), !dbg !167 + %2871 = bitcast float %2856 to i32, !dbg !167 + %2872 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2871, i32 %2574, i32 31), !dbg !167 + %.v117 = select i1 %2561, i32 %2860, i32 %2858, !dbg !167 + %.v118 = select i1 %2561, i32 %2864, i32 %2862, !dbg !167 + %.v119 = select i1 %2561, i32 %2868, i32 %2866, !dbg !167 + %.v120 = select i1 %2561, i32 %2872, i32 %2870, !dbg !167 + %2873 = bitcast float %2554 to i32, !dbg !247 + %2874 = bitcast float %.pn92.lcssa to i32, !dbg !247 + %2875 = bitcast float %.pn88.lcssa to i32, !dbg !247 + %2876 = bitcast float %2558 to i32, !dbg !247 + %2877 = bitcast float %2556 to i32, !dbg !247 + %2878 = bitcast float %.pn90.lcssa to i32, !dbg !247 + %2879 = bitcast float %.pn86.lcssa to i32, !dbg !247 + %2880 = bitcast float %2560 to i32, !dbg !247 + %2881 = select i1 %2561, i32 %2858, i32 %2860, !dbg !167 + %2882 = select i1 %2561, i32 %2862, i32 %2864, !dbg !167 + %2883 = select i1 %2561, i32 %2866, i32 %2868, !dbg !167 + %2884 = select i1 %2561, i32 %2870, i32 %2872, !dbg !167 + %2885 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2873, i32 %2874, i32 %2875, i32 %2876, i32 %2881, i32 %2882) #2, !dbg !247 + %2886 = extractvalue { float, float, float, float } %2885, 0, !dbg !247 + %2887 = extractvalue { float, float, float, float } %2885, 1, !dbg !247 + %2888 = extractvalue { float, float, float, float } %2885, 2, !dbg !247 + %2889 = extractvalue { float, float, float, float } %2885, 3, !dbg !247 + %2890 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2873, i32 %2874, i32 %2875, i32 %2876, i32 %2883, i32 %2884) #2, !dbg !247 + %2891 = extractvalue { float, float, float, float } %2890, 0, !dbg !247 + %2892 = extractvalue { float, float, float, float } %2890, 1, !dbg !247 + %2893 = extractvalue { float, float, float, float } %2890, 2, !dbg !247 + %2894 = extractvalue { float, float, float, float } %2890, 3, !dbg !247 + %2895 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2886, float %2887, float %2888, float %2889, i32 %2877, i32 %2878, i32 %2879, i32 %2880, i32 %.v117, i32 %.v118) #2, !dbg !247 + %2896 = extractvalue { float, float, float, float } %2895, 0, !dbg !247 + %2897 = extractvalue { float, float, float, float } %2895, 1, !dbg !247 + %2898 = extractvalue { float, float, float, float } %2895, 2, !dbg !247 + %2899 = extractvalue { float, float, float, float } %2895, 3, !dbg !247 + %2900 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2891, float %2892, float %2893, float %2894, i32 %2877, i32 %2878, i32 %2879, i32 %2880, i32 %.v119, i32 %.v120) #2, !dbg !247 + %2901 = extractvalue { float, float, float, float } %2900, 0, !dbg !247 + %2902 = extractvalue { float, float, float, float } %2900, 1, !dbg !247 + %2903 = extractvalue { float, float, float, float } %2900, 2, !dbg !247 + %2904 = extractvalue { float, float, float, float } %2900, 3, !dbg !247 + %2905 = select i1 %.not100, float %2896, float %2897, !dbg !247 + %2906 = select i1 %.not100, float %2897, float %2896, !dbg !247 + %2907 = select i1 %.not100, float %2898, float %2899, !dbg !247 + %2908 = select i1 %.not100, float %2899, float %2898, !dbg !247 + %2909 = select i1 %.not100, float %2901, float %2902, !dbg !247 + %2910 = select i1 %.not100, float %2902, float %2901, !dbg !247 + %2911 = select i1 %.not100, float %2903, float %2904, !dbg !247 + %2912 = select i1 %.not100, float %2904, float %2903, !dbg !247 + %2913 = bitcast float %2905 to i32, !dbg !247 + %2914 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2913, i32 %2633, i32 31), !dbg !247 + %2915 = bitcast float %2907 to i32, !dbg !247 + %2916 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2915, i32 %2633, i32 31), !dbg !247 + %2917 = bitcast float %2906 to i32, !dbg !247 + %2918 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2917, i32 %2638, i32 31), !dbg !247 + %2919 = bitcast float %2908 to i32, !dbg !247 + %2920 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2919, i32 %2638, i32 31), !dbg !247 + %2921 = bitcast float %2909 to i32, !dbg !247 + %2922 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2921, i32 %2633, i32 31), !dbg !247 + %2923 = bitcast float %2911 to i32, !dbg !247 + %2924 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2923, i32 %2633, i32 31), !dbg !247 + %2925 = bitcast float %2910 to i32, !dbg !247 + %2926 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2925, i32 %2638, i32 31), !dbg !247 + %2927 = bitcast float %2912 to i32, !dbg !247 + %2928 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2927, i32 %2638, i32 31), !dbg !247 + %.v121 = select i1 %2631, i32 %2918, i32 %2914, !dbg !247 + %.v122 = select i1 %2631, i32 %2920, i32 %2916, !dbg !247 + %.v123 = select i1 %2631, i32 %2926, i32 %2922, !dbg !247 + %.v124 = select i1 %2631, i32 %2928, i32 %2924, !dbg !247 + %2929 = select i1 %2784, float %2546, float %2548, !dbg !223 + %2930 = select i1 %2784, float %2548, float %2546, !dbg !223 + %2931 = select i1 %2784, float %.pn76.lcssa, float %.pn74.lcssa, !dbg !223 + %2932 = select i1 %2784, float %.pn74.lcssa, float %.pn76.lcssa, !dbg !223 + %2933 = select i1 %2784, float %.pn72.lcssa, float %.pn70.lcssa, !dbg !223 + %2934 = select i1 %2784, float %.pn70.lcssa, float %.pn72.lcssa, !dbg !223 + %2935 = select i1 %2784, float %2550, float %2552, !dbg !223 + %2936 = select i1 %2784, float %2552, float %2550, !dbg !223 + %2937 = bitcast float %2929 to i32, !dbg !223 + %2938 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2937, i32 %2795, i32 31), !dbg !223 + %2939 = bitcast float %2930 to i32, !dbg !223 + %2940 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2939, i32 %2798, i32 31), !dbg !223 + %2941 = bitcast float %2931 to i32, !dbg !223 + %2942 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2941, i32 %2795, i32 31), !dbg !223 + %2943 = bitcast float %2932 to i32, !dbg !223 + %2944 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2943, i32 %2798, i32 31), !dbg !223 + %2945 = bitcast float %2933 to i32, !dbg !223 + %2946 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2945, i32 %2795, i32 31), !dbg !223 + %2947 = bitcast float %2934 to i32, !dbg !223 + %2948 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2947, i32 %2798, i32 31), !dbg !223 + %2949 = bitcast float %2935 to i32, !dbg !223 + %2950 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2949, i32 %2795, i32 31), !dbg !223 + %2951 = bitcast float %2936 to i32, !dbg !223 + %2952 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %2951, i32 %2798, i32 31), !dbg !223 + %.v125 = select i1 %2561, i32 %2940, i32 %2938, !dbg !223 + %.v126 = select i1 %2561, i32 %2944, i32 %2942, !dbg !223 + %.v127 = select i1 %2561, i32 %2948, i32 %2946, !dbg !223 + %.v128 = select i1 %2561, i32 %2952, i32 %2950, !dbg !223 + %2953 = select i1 %2631, i32 %2914, i32 %2918, !dbg !247 + %2954 = select i1 %2631, i32 %2916, i32 %2920, !dbg !247 + %2955 = select i1 %2631, i32 %2922, i32 %2926, !dbg !247 + %2956 = select i1 %2631, i32 %2924, i32 %2928, !dbg !247 + %2957 = select i1 %2561, i32 %2938, i32 %2940, !dbg !223 + %2958 = select i1 %2561, i32 %2942, i32 %2944, !dbg !223 + %2959 = select i1 %2561, i32 %2946, i32 %2948, !dbg !223 + %2960 = select i1 %2561, i32 %2950, i32 %2952, !dbg !223 + %2961 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2953, i32 %2954, i32 %.v121, i32 %.v122, i32 %2957, i32 %.v125) #2, !dbg !248 + %2962 = extractvalue { float, float, float, float } %2961, 0, !dbg !248 + %2963 = extractvalue { float, float, float, float } %2961, 1, !dbg !248 + %2964 = extractvalue { float, float, float, float } %2961, 2, !dbg !248 + %2965 = extractvalue { float, float, float, float } %2961, 3, !dbg !248 + %2966 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %2953, i32 %2954, i32 %.v121, i32 %.v122, i32 %2959, i32 %.v127) #2, !dbg !248 + %2967 = extractvalue { float, float, float, float } %2966, 0, !dbg !248 + %2968 = extractvalue { float, float, float, float } %2966, 1, !dbg !248 + %2969 = extractvalue { float, float, float, float } %2966, 2, !dbg !248 + %2970 = extractvalue { float, float, float, float } %2966, 3, !dbg !248 + %2971 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2962, float %2963, float %2964, float %2965, i32 %2955, i32 %2956, i32 %.v123, i32 %.v124, i32 %2958, i32 %.v126) #2, !dbg !248 + %2972 = extractvalue { float, float, float, float } %2971, 0, !dbg !248 + %2973 = extractvalue { float, float, float, float } %2971, 1, !dbg !248 + %2974 = extractvalue { float, float, float, float } %2971, 2, !dbg !248 + %2975 = extractvalue { float, float, float, float } %2971, 3, !dbg !248 + %2976 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2967, float %2968, float %2969, float %2970, i32 %2955, i32 %2956, i32 %.v123, i32 %.v124, i32 %2960, i32 %.v128) #2, !dbg !248 + %2977 = extractvalue { float, float, float, float } %2976, 0, !dbg !248 + %2978 = extractvalue { float, float, float, float } %2976, 1, !dbg !248 + %2979 = extractvalue { float, float, float, float } %2976, 2, !dbg !248 + %2980 = extractvalue { float, float, float, float } %2976, 3, !dbg !248 + %2981 = select i1 %2631, i32 0, i32 160, !dbg !240 + %2982 = shl nuw nsw i32 %2621, 7, !dbg !240 + %2983 = or disjoint i32 %2981, %2982, !dbg !240 + %2984 = or disjoint i32 %2983, %26, !dbg !240 + %2985 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %2984, !dbg !240 + %2986 = load i32, ptr addrspace(3) %2985, align 4, !dbg !240 + %2987 = getelementptr inbounds nuw i8, ptr addrspace(3) %2985, i32 64, !dbg !240 + %2988 = load i32, ptr addrspace(3) %2987, align 4, !dbg !240 + %2989 = getelementptr inbounds nuw i8, ptr addrspace(3) %2985, i32 512, !dbg !240 + %2990 = load i32, ptr addrspace(3) %2989, align 4, !dbg !240 + %2991 = getelementptr inbounds nuw i8, ptr addrspace(3) %2985, i32 576, !dbg !240 + %2992 = load i32, ptr addrspace(3) %2991, align 4, !dbg !240 + %2993 = xor i32 %2984, 32, !dbg !240 + %2994 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %2993, !dbg !240 + %2995 = load i32, ptr addrspace(3) %2994, align 4, !dbg !240 + %2996 = getelementptr inbounds nuw i8, ptr addrspace(3) %2994, i32 64, !dbg !240 + %2997 = load i32, ptr addrspace(3) %2996, align 4, !dbg !240 + %2998 = getelementptr inbounds nuw i8, ptr addrspace(3) %2994, i32 512, !dbg !240 + %2999 = load i32, ptr addrspace(3) %2998, align 4, !dbg !240 + %3000 = getelementptr inbounds nuw i8, ptr addrspace(3) %2994, i32 576, !dbg !240 + %3001 = load i32, ptr addrspace(3) %3000, align 4, !dbg !240 + %3002 = bitcast float %1772 to i32, !dbg !249 + %3003 = bitcast float %1774 to i32, !dbg !249 + %3004 = bitcast float %1776 to i32, !dbg !249 + %3005 = bitcast float %1778 to i32, !dbg !249 + %3006 = bitcast float %1773 to i32, !dbg !249 + %3007 = bitcast float %1775 to i32, !dbg !249 + %3008 = bitcast float %1777 to i32, !dbg !249 + %3009 = bitcast float %1779 to i32, !dbg !249 + %3010 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %3002, i32 %3003, i32 %3004, i32 %3005, i32 %2986, i32 %2990) #2, !dbg !249 + %3011 = extractvalue { float, float, float, float } %3010, 0, !dbg !249 + %3012 = extractvalue { float, float, float, float } %3010, 1, !dbg !249 + %3013 = extractvalue { float, float, float, float } %3010, 2, !dbg !249 + %3014 = extractvalue { float, float, float, float } %3010, 3, !dbg !249 + %3015 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %3002, i32 %3003, i32 %3004, i32 %3005, i32 %2995, i32 %2999) #2, !dbg !249 + %3016 = extractvalue { float, float, float, float } %3015, 0, !dbg !249 + %3017 = extractvalue { float, float, float, float } %3015, 1, !dbg !249 + %3018 = extractvalue { float, float, float, float } %3015, 2, !dbg !249 + %3019 = extractvalue { float, float, float, float } %3015, 3, !dbg !249 + %3020 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3011, float %3012, float %3013, float %3014, i32 %3006, i32 %3007, i32 %3008, i32 %3009, i32 %2988, i32 %2992) #2, !dbg !249 + %3021 = extractvalue { float, float, float, float } %3020, 0, !dbg !249 + %3022 = extractvalue { float, float, float, float } %3020, 1, !dbg !249 + %3023 = extractvalue { float, float, float, float } %3020, 2, !dbg !249 + %3024 = extractvalue { float, float, float, float } %3020, 3, !dbg !249 + %3025 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3016, float %3017, float %3018, float %3019, i32 %3006, i32 %3007, i32 %3008, i32 %3009, i32 %2997, i32 %3001) #2, !dbg !249 + %3026 = extractvalue { float, float, float, float } %3025, 0, !dbg !249 + %3027 = extractvalue { float, float, float, float } %3025, 1, !dbg !249 + %3028 = extractvalue { float, float, float, float } %3025, 2, !dbg !249 + %3029 = extractvalue { float, float, float, float } %3025, 3, !dbg !249 + %3030 = select i1 %2561, float %2696, float %2697, !dbg !243 + %3031 = select i1 %2561, float %2697, float %2696, !dbg !243 + %3032 = select i1 %2561, float %2698, float %2699, !dbg !243 + %3033 = select i1 %2561, float %2699, float %2698, !dbg !243 + %3034 = select i1 %2561, float %2700, float %2701, !dbg !243 + %3035 = select i1 %2561, float %2701, float %2700, !dbg !243 + %3036 = select i1 %2561, float %2702, float %2703, !dbg !243 + %3037 = select i1 %2561, float %2703, float %2702, !dbg !243 + %3038 = bitcast float %3030 to i32, !dbg !243 + %3039 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3038, i32 %2571, i32 31), !dbg !243 + %3040 = bitcast float %3031 to i32, !dbg !243 + %3041 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3040, i32 %2574, i32 31), !dbg !243 + %3042 = bitcast float %3032 to i32, !dbg !243 + %3043 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3042, i32 %2571, i32 31), !dbg !243 + %3044 = bitcast float %3033 to i32, !dbg !243 + %3045 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3044, i32 %2574, i32 31), !dbg !243 + %3046 = bitcast float %3034 to i32, !dbg !243 + %3047 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3046, i32 %2571, i32 31), !dbg !243 + %3048 = bitcast float %3035 to i32, !dbg !243 + %3049 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3048, i32 %2574, i32 31), !dbg !243 + %3050 = bitcast float %3036 to i32, !dbg !243 + %3051 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3050, i32 %2571, i32 31), !dbg !243 + %3052 = bitcast float %3037 to i32, !dbg !243 + %3053 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3052, i32 %2574, i32 31), !dbg !243 + %.v129 = select i1 %2561, i32 %3041, i32 %3039, !dbg !243 + %.v130 = select i1 %2561, i32 %3045, i32 %3043, !dbg !243 + %.v131 = select i1 %2561, i32 %3049, i32 %3047, !dbg !243 + %.v132 = select i1 %2561, i32 %3053, i32 %3051, !dbg !243 + %3054 = bitcast float %1780 to i32, !dbg !250 + %3055 = bitcast float %1782 to i32, !dbg !250 + %3056 = bitcast float %1784 to i32, !dbg !250 + %3057 = bitcast float %1786 to i32, !dbg !250 + %3058 = bitcast float %1781 to i32, !dbg !250 + %3059 = bitcast float %1783 to i32, !dbg !250 + %3060 = bitcast float %1785 to i32, !dbg !250 + %3061 = bitcast float %1787 to i32, !dbg !250 + %3062 = select i1 %2561, i32 %3039, i32 %3041, !dbg !243 + %3063 = select i1 %2561, i32 %3043, i32 %3045, !dbg !243 + %3064 = select i1 %2561, i32 %3047, i32 %3049, !dbg !243 + %3065 = select i1 %2561, i32 %3051, i32 %3053, !dbg !243 + %3066 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3021, float %3022, float %3023, float %3024, i32 %3054, i32 %3055, i32 %3056, i32 %3057, i32 %3062, i32 %3063) #2, !dbg !250 + %3067 = extractvalue { float, float, float, float } %3066, 0, !dbg !250 + %3068 = extractvalue { float, float, float, float } %3066, 1, !dbg !250 + %3069 = extractvalue { float, float, float, float } %3066, 2, !dbg !250 + %3070 = extractvalue { float, float, float, float } %3066, 3, !dbg !250 + %3071 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3026, float %3027, float %3028, float %3029, i32 %3054, i32 %3055, i32 %3056, i32 %3057, i32 %3064, i32 %3065) #2, !dbg !250 + %3072 = extractvalue { float, float, float, float } %3071, 0, !dbg !250 + %3073 = extractvalue { float, float, float, float } %3071, 1, !dbg !250 + %3074 = extractvalue { float, float, float, float } %3071, 2, !dbg !250 + %3075 = extractvalue { float, float, float, float } %3071, 3, !dbg !250 + %3076 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3067, float %3068, float %3069, float %3070, i32 %3058, i32 %3059, i32 %3060, i32 %3061, i32 %.v129, i32 %.v130) #2, !dbg !250 + %3077 = extractvalue { float, float, float, float } %3076, 0, !dbg !250 + %3078 = extractvalue { float, float, float, float } %3076, 1, !dbg !250 + %3079 = extractvalue { float, float, float, float } %3076, 2, !dbg !250 + %3080 = extractvalue { float, float, float, float } %3076, 3, !dbg !250 + %3081 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3072, float %3073, float %3074, float %3075, i32 %3058, i32 %3059, i32 %3060, i32 %3061, i32 %.v131, i32 %.v132) #2, !dbg !250 + %3082 = extractvalue { float, float, float, float } %3081, 0, !dbg !250 + %3083 = extractvalue { float, float, float, float } %3081, 1, !dbg !250 + %3084 = extractvalue { float, float, float, float } %3081, 2, !dbg !250 + %3085 = extractvalue { float, float, float, float } %3081, 3, !dbg !250 + %3086 = select i1 %.not100, float %2546, float %2548, !dbg !223 + %3087 = select i1 %.not100, float %2548, float %2546, !dbg !223 + %3088 = select i1 %.not100, float %.pn76.lcssa, float %.pn74.lcssa, !dbg !223 + %3089 = select i1 %.not100, float %.pn74.lcssa, float %.pn76.lcssa, !dbg !223 + %3090 = select i1 %.not100, float %.pn72.lcssa, float %.pn70.lcssa, !dbg !223 + %3091 = select i1 %.not100, float %.pn70.lcssa, float %.pn72.lcssa, !dbg !223 + %3092 = select i1 %.not100, float %2550, float %2552, !dbg !223 + %3093 = select i1 %.not100, float %2552, float %2550, !dbg !223 + %3094 = bitcast float %3086 to i32, !dbg !223 + %3095 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3094, i32 %2633, i32 31), !dbg !223 + %3096 = bitcast float %3088 to i32, !dbg !223 + %3097 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3096, i32 %2633, i32 31), !dbg !223 + %3098 = bitcast float %3087 to i32, !dbg !223 + %3099 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3098, i32 %2638, i32 31), !dbg !223 + %3100 = bitcast float %3089 to i32, !dbg !223 + %3101 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3100, i32 %2638, i32 31), !dbg !223 + %3102 = bitcast float %3090 to i32, !dbg !223 + %3103 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3102, i32 %2633, i32 31), !dbg !223 + %3104 = bitcast float %3092 to i32, !dbg !223 + %3105 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3104, i32 %2633, i32 31), !dbg !223 + %3106 = bitcast float %3091 to i32, !dbg !223 + %3107 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3106, i32 %2638, i32 31), !dbg !223 + %3108 = bitcast float %3093 to i32, !dbg !223 + %3109 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3108, i32 %2638, i32 31), !dbg !223 + %.v133 = select i1 %2631, i32 %3099, i32 %3095, !dbg !223 + %.v134 = select i1 %2631, i32 %3101, i32 %3097, !dbg !223 + %.v135 = select i1 %2631, i32 %3107, i32 %3103, !dbg !223 + %.v136 = select i1 %2631, i32 %3109, i32 %3105, !dbg !223 + %3110 = select i1 %2784, float %3077, float %3078, !dbg !250 + %3111 = select i1 %2784, float %3078, float %3077, !dbg !250 + %3112 = select i1 %2784, float %3079, float %3080, !dbg !250 + %3113 = select i1 %2784, float %3080, float %3079, !dbg !250 + %3114 = select i1 %2784, float %3082, float %3083, !dbg !250 + %3115 = select i1 %2784, float %3083, float %3082, !dbg !250 + %3116 = select i1 %2784, float %3084, float %3085, !dbg !250 + %3117 = select i1 %2784, float %3085, float %3084, !dbg !250 + %3118 = bitcast float %3110 to i32, !dbg !250 + %3119 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3118, i32 %2795, i32 31), !dbg !250 + %3120 = bitcast float %3111 to i32, !dbg !250 + %3121 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3120, i32 %2798, i32 31), !dbg !250 + %3122 = bitcast float %3112 to i32, !dbg !250 + %3123 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3122, i32 %2795, i32 31), !dbg !250 + %3124 = bitcast float %3113 to i32, !dbg !250 + %3125 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3124, i32 %2798, i32 31), !dbg !250 + %3126 = bitcast float %3114 to i32, !dbg !250 + %3127 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3126, i32 %2795, i32 31), !dbg !250 + %3128 = bitcast float %3115 to i32, !dbg !250 + %3129 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3128, i32 %2798, i32 31), !dbg !250 + %3130 = bitcast float %3116 to i32, !dbg !250 + %3131 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3130, i32 %2795, i32 31), !dbg !250 + %3132 = bitcast float %3117 to i32, !dbg !250 + %3133 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3132, i32 %2798, i32 31), !dbg !250 + %.v137 = select i1 %2561, i32 %3121, i32 %3119, !dbg !250 + %.v138 = select i1 %2561, i32 %3125, i32 %3123, !dbg !250 + %.v139 = select i1 %2561, i32 %3129, i32 %3127, !dbg !250 + %.v140 = select i1 %2561, i32 %3133, i32 %3131, !dbg !250 + %3134 = select i1 %2631, i32 %3095, i32 %3099, !dbg !223 + %3135 = select i1 %2631, i32 %3097, i32 %3101, !dbg !223 + %3136 = select i1 %2631, i32 %3103, i32 %3107, !dbg !223 + %3137 = select i1 %2631, i32 %3105, i32 %3109, !dbg !223 + %3138 = select i1 %2561, i32 %3119, i32 %3121, !dbg !250 + %3139 = select i1 %2561, i32 %3123, i32 %3125, !dbg !250 + %3140 = select i1 %2561, i32 %3127, i32 %3129, !dbg !250 + %3141 = select i1 %2561, i32 %3131, i32 %3133, !dbg !250 + %3142 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %3134, i32 %3135, i32 %.v133, i32 %.v134, i32 %3138, i32 %.v137) #2, !dbg !251 + %3143 = extractvalue { float, float, float, float } %3142, 0, !dbg !251 + %3144 = extractvalue { float, float, float, float } %3142, 1, !dbg !251 + %3145 = extractvalue { float, float, float, float } %3142, 2, !dbg !251 + %3146 = extractvalue { float, float, float, float } %3142, 3, !dbg !251 + %3147 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %3134, i32 %3135, i32 %.v133, i32 %.v134, i32 %3140, i32 %.v139) #2, !dbg !251 + %3148 = extractvalue { float, float, float, float } %3147, 0, !dbg !251 + %3149 = extractvalue { float, float, float, float } %3147, 1, !dbg !251 + %3150 = extractvalue { float, float, float, float } %3147, 2, !dbg !251 + %3151 = extractvalue { float, float, float, float } %3147, 3, !dbg !251 + %3152 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3143, float %3144, float %3145, float %3146, i32 %3136, i32 %3137, i32 %.v135, i32 %.v136, i32 %3139, i32 %.v138) #2, !dbg !251 + %3153 = extractvalue { float, float, float, float } %3152, 0, !dbg !251 + %3154 = extractvalue { float, float, float, float } %3152, 1, !dbg !251 + %3155 = extractvalue { float, float, float, float } %3152, 2, !dbg !251 + %3156 = extractvalue { float, float, float, float } %3152, 3, !dbg !251 + %3157 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3148, float %3149, float %3150, float %3151, i32 %3136, i32 %3137, i32 %.v135, i32 %.v136, i32 %3141, i32 %.v140) #2, !dbg !251 + %3158 = extractvalue { float, float, float, float } %3157, 0, !dbg !251 + %3159 = extractvalue { float, float, float, float } %3157, 1, !dbg !251 + %3160 = extractvalue { float, float, float, float } %3157, 2, !dbg !251 + %3161 = extractvalue { float, float, float, float } %3157, 3, !dbg !251 + %3162 = fsub float 0.000000e+00, %3153, !dbg !252 + %3163 = fsub float 0.000000e+00, %3154, !dbg !252 + %3164 = fsub float 0.000000e+00, %3155, !dbg !252 + %3165 = fsub float 0.000000e+00, %3156, !dbg !252 + %3166 = fsub float 0.000000e+00, %3158, !dbg !252 + %3167 = fsub float 0.000000e+00, %3159, !dbg !252 + %3168 = fsub float 0.000000e+00, %3160, !dbg !252 + %3169 = fsub float 0.000000e+00, %3161, !dbg !252 + %3170 = select i1 %2561, float %2538, float %2540, !dbg !207 + %3171 = select i1 %2561, float %2540, float %2538, !dbg !207 + %3172 = select i1 %2561, float %.pn60.lcssa, float %.pn58.lcssa, !dbg !207 + %3173 = select i1 %2561, float %.pn58.lcssa, float %.pn60.lcssa, !dbg !207 + %3174 = select i1 %2561, float %.pn56.lcssa, float %.pn54.lcssa, !dbg !207 + %3175 = select i1 %2561, float %.pn54.lcssa, float %.pn56.lcssa, !dbg !207 + %3176 = select i1 %2561, float %2542, float %2544, !dbg !207 + %3177 = select i1 %2561, float %2544, float %2542, !dbg !207 + %3178 = bitcast float %3170 to i32, !dbg !207 + %3179 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3178, i32 %2571, i32 31), !dbg !207 + %3180 = bitcast float %3171 to i32, !dbg !207 + %3181 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3180, i32 %2574, i32 31), !dbg !207 + %3182 = bitcast float %3172 to i32, !dbg !207 + %3183 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3182, i32 %2571, i32 31), !dbg !207 + %3184 = bitcast float %3173 to i32, !dbg !207 + %3185 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3184, i32 %2574, i32 31), !dbg !207 + %3186 = bitcast float %3174 to i32, !dbg !207 + %3187 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3186, i32 %2571, i32 31), !dbg !207 + %3188 = bitcast float %3175 to i32, !dbg !207 + %3189 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3188, i32 %2574, i32 31), !dbg !207 + %3190 = bitcast float %3176 to i32, !dbg !207 + %3191 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3190, i32 %2571, i32 31), !dbg !207 + %3192 = bitcast float %3177 to i32, !dbg !207 + %3193 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3192, i32 %2574, i32 31), !dbg !207 + %.v141 = select i1 %2561, i32 %3181, i32 %3179, !dbg !207 + %.v142 = select i1 %2561, i32 %3185, i32 %3183, !dbg !207 + %.v143 = select i1 %2561, i32 %3189, i32 %3187, !dbg !207 + %.v144 = select i1 %2561, i32 %3193, i32 %3191, !dbg !207 + %3194 = bitcast float %1796 to i32, !dbg !253 + %3195 = bitcast float %1798 to i32, !dbg !253 + %3196 = bitcast float %1800 to i32, !dbg !253 + %3197 = bitcast float %1802 to i32, !dbg !253 + %3198 = bitcast float %1797 to i32, !dbg !253 + %3199 = bitcast float %1799 to i32, !dbg !253 + %3200 = bitcast float %1801 to i32, !dbg !253 + %3201 = bitcast float %1803 to i32, !dbg !253 + %3202 = select i1 %2561, i32 %3179, i32 %3181, !dbg !207 + %3203 = select i1 %2561, i32 %3183, i32 %3185, !dbg !207 + %3204 = select i1 %2561, i32 %3187, i32 %3189, !dbg !207 + %3205 = select i1 %2561, i32 %3191, i32 %3193, !dbg !207 + %3206 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %3194, i32 %3195, i32 %3196, i32 %3197, i32 %3202, i32 %3203) #2, !dbg !253 + %3207 = extractvalue { float, float, float, float } %3206, 0, !dbg !253 + %3208 = extractvalue { float, float, float, float } %3206, 1, !dbg !253 + %3209 = extractvalue { float, float, float, float } %3206, 2, !dbg !253 + %3210 = extractvalue { float, float, float, float } %3206, 3, !dbg !253 + %3211 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %3194, i32 %3195, i32 %3196, i32 %3197, i32 %3204, i32 %3205) #2, !dbg !253 + %3212 = extractvalue { float, float, float, float } %3211, 0, !dbg !253 + %3213 = extractvalue { float, float, float, float } %3211, 1, !dbg !253 + %3214 = extractvalue { float, float, float, float } %3211, 2, !dbg !253 + %3215 = extractvalue { float, float, float, float } %3211, 3, !dbg !253 + %3216 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3207, float %3208, float %3209, float %3210, i32 %3198, i32 %3199, i32 %3200, i32 %3201, i32 %.v141, i32 %.v142) #2, !dbg !253 + %3217 = extractvalue { float, float, float, float } %3216, 0, !dbg !253 + %3218 = extractvalue { float, float, float, float } %3216, 1, !dbg !253 + %3219 = extractvalue { float, float, float, float } %3216, 2, !dbg !253 + %3220 = extractvalue { float, float, float, float } %3216, 3, !dbg !253 + %3221 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3212, float %3213, float %3214, float %3215, i32 %3198, i32 %3199, i32 %3200, i32 %3201, i32 %.v143, i32 %.v144) #2, !dbg !253 + %3222 = extractvalue { float, float, float, float } %3221, 0, !dbg !253 + %3223 = extractvalue { float, float, float, float } %3221, 1, !dbg !253 + %3224 = extractvalue { float, float, float, float } %3221, 2, !dbg !253 + %3225 = extractvalue { float, float, float, float } %3221, 3, !dbg !253 + %3226 = select i1 %2561, float %2841, float %2842, !dbg !246 + %3227 = select i1 %2561, float %2842, float %2841, !dbg !246 + %3228 = select i1 %2561, float %2843, float %2844, !dbg !246 + %3229 = select i1 %2561, float %2844, float %2843, !dbg !246 + %3230 = select i1 %2561, float %2845, float %2846, !dbg !246 + %3231 = select i1 %2561, float %2846, float %2845, !dbg !246 + %3232 = select i1 %2561, float %2847, float %2848, !dbg !246 + %3233 = select i1 %2561, float %2848, float %2847, !dbg !246 + %3234 = bitcast float %3226 to i32, !dbg !246 + %3235 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3234, i32 %2571, i32 31), !dbg !246 + %3236 = bitcast float %3227 to i32, !dbg !246 + %3237 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3236, i32 %2574, i32 31), !dbg !246 + %3238 = bitcast float %3228 to i32, !dbg !246 + %3239 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3238, i32 %2571, i32 31), !dbg !246 + %3240 = bitcast float %3229 to i32, !dbg !246 + %3241 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3240, i32 %2574, i32 31), !dbg !246 + %3242 = bitcast float %3230 to i32, !dbg !246 + %3243 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3242, i32 %2571, i32 31), !dbg !246 + %3244 = bitcast float %3231 to i32, !dbg !246 + %3245 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3244, i32 %2574, i32 31), !dbg !246 + %3246 = bitcast float %3232 to i32, !dbg !246 + %3247 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3246, i32 %2571, i32 31), !dbg !246 + %3248 = bitcast float %3233 to i32, !dbg !246 + %3249 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3248, i32 %2574, i32 31), !dbg !246 + %.v145 = select i1 %2561, i32 %3237, i32 %3235, !dbg !246 + %.v146 = select i1 %2561, i32 %3241, i32 %3239, !dbg !246 + %.v147 = select i1 %2561, i32 %3245, i32 %3243, !dbg !246 + %.v148 = select i1 %2561, i32 %3249, i32 %3247, !dbg !246 + %3250 = bitcast float %1804 to i32, !dbg !254 + %3251 = bitcast float %1806 to i32, !dbg !254 + %3252 = bitcast float %1808 to i32, !dbg !254 + %3253 = bitcast float %1810 to i32, !dbg !254 + %3254 = bitcast float %1805 to i32, !dbg !254 + %3255 = bitcast float %1807 to i32, !dbg !254 + %3256 = bitcast float %1809 to i32, !dbg !254 + %3257 = bitcast float %1811 to i32, !dbg !254 + %3258 = select i1 %2561, i32 %3235, i32 %3237, !dbg !246 + %3259 = select i1 %2561, i32 %3239, i32 %3241, !dbg !246 + %3260 = select i1 %2561, i32 %3243, i32 %3245, !dbg !246 + %3261 = select i1 %2561, i32 %3247, i32 %3249, !dbg !246 + %3262 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3217, float %3218, float %3219, float %3220, i32 %3250, i32 %3251, i32 %3252, i32 %3253, i32 %3258, i32 %3259) #2, !dbg !254 + %3263 = extractvalue { float, float, float, float } %3262, 0, !dbg !254 + %3264 = extractvalue { float, float, float, float } %3262, 1, !dbg !254 + %3265 = extractvalue { float, float, float, float } %3262, 2, !dbg !254 + %3266 = extractvalue { float, float, float, float } %3262, 3, !dbg !254 + %3267 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3222, float %3223, float %3224, float %3225, i32 %3250, i32 %3251, i32 %3252, i32 %3253, i32 %3260, i32 %3261) #2, !dbg !254 + %3268 = extractvalue { float, float, float, float } %3267, 0, !dbg !254 + %3269 = extractvalue { float, float, float, float } %3267, 1, !dbg !254 + %3270 = extractvalue { float, float, float, float } %3267, 2, !dbg !254 + %3271 = extractvalue { float, float, float, float } %3267, 3, !dbg !254 + %3272 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3263, float %3264, float %3265, float %3266, i32 %3254, i32 %3255, i32 %3256, i32 %3257, i32 %.v145, i32 %.v146) #2, !dbg !254 + %3273 = extractvalue { float, float, float, float } %3272, 0, !dbg !254 + %3274 = extractvalue { float, float, float, float } %3272, 1, !dbg !254 + %3275 = extractvalue { float, float, float, float } %3272, 2, !dbg !254 + %3276 = extractvalue { float, float, float, float } %3272, 3, !dbg !254 + %3277 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3268, float %3269, float %3270, float %3271, i32 %3254, i32 %3255, i32 %3256, i32 %3257, i32 %.v147, i32 %.v148) #2, !dbg !254 + %3278 = extractvalue { float, float, float, float } %3277, 0, !dbg !254 + %3279 = extractvalue { float, float, float, float } %3277, 1, !dbg !254 + %3280 = extractvalue { float, float, float, float } %3277, 2, !dbg !254 + %3281 = extractvalue { float, float, float, float } %3277, 3, !dbg !254 + %3282 = select i1 %.not100, float %2554, float %2556, !dbg !239 + %3283 = select i1 %.not100, float %2556, float %2554, !dbg !239 + %3284 = select i1 %.not100, float %.pn92.lcssa, float %.pn90.lcssa, !dbg !239 + %3285 = select i1 %.not100, float %.pn90.lcssa, float %.pn92.lcssa, !dbg !239 + %3286 = select i1 %.not100, float %.pn88.lcssa, float %.pn86.lcssa, !dbg !239 + %3287 = select i1 %.not100, float %.pn86.lcssa, float %.pn88.lcssa, !dbg !239 + %3288 = select i1 %.not100, float %2558, float %2560, !dbg !239 + %3289 = select i1 %.not100, float %2560, float %2558, !dbg !239 + %3290 = bitcast float %3282 to i32, !dbg !239 + %3291 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3290, i32 %2633, i32 31), !dbg !239 + %3292 = bitcast float %3284 to i32, !dbg !239 + %3293 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3292, i32 %2633, i32 31), !dbg !239 + %3294 = bitcast float %3283 to i32, !dbg !239 + %3295 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3294, i32 %2638, i32 31), !dbg !239 + %3296 = bitcast float %3285 to i32, !dbg !239 + %3297 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3296, i32 %2638, i32 31), !dbg !239 + %3298 = bitcast float %3286 to i32, !dbg !239 + %3299 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3298, i32 %2633, i32 31), !dbg !239 + %3300 = bitcast float %3288 to i32, !dbg !239 + %3301 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3300, i32 %2633, i32 31), !dbg !239 + %3302 = bitcast float %3287 to i32, !dbg !239 + %3303 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3302, i32 %2638, i32 31), !dbg !239 + %3304 = bitcast float %3289 to i32, !dbg !239 + %3305 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3304, i32 %2638, i32 31), !dbg !239 + %.v149 = select i1 %2631, i32 %3295, i32 %3291, !dbg !239 + %.v150 = select i1 %2631, i32 %3297, i32 %3293, !dbg !239 + %.v151 = select i1 %2631, i32 %3303, i32 %3299, !dbg !239 + %.v152 = select i1 %2631, i32 %3305, i32 %3301, !dbg !239 + %3306 = select i1 %2784, float %3273, float %3274, !dbg !254 + %3307 = select i1 %2784, float %3274, float %3273, !dbg !254 + %3308 = select i1 %2784, float %3275, float %3276, !dbg !254 + %3309 = select i1 %2784, float %3276, float %3275, !dbg !254 + %3310 = select i1 %2784, float %3278, float %3279, !dbg !254 + %3311 = select i1 %2784, float %3279, float %3278, !dbg !254 + %3312 = select i1 %2784, float %3280, float %3281, !dbg !254 + %3313 = select i1 %2784, float %3281, float %3280, !dbg !254 + %3314 = bitcast float %3306 to i32, !dbg !254 + %3315 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3314, i32 %2795, i32 31), !dbg !254 + %3316 = bitcast float %3307 to i32, !dbg !254 + %3317 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3316, i32 %2798, i32 31), !dbg !254 + %3318 = bitcast float %3308 to i32, !dbg !254 + %3319 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3318, i32 %2795, i32 31), !dbg !254 + %3320 = bitcast float %3309 to i32, !dbg !254 + %3321 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3320, i32 %2798, i32 31), !dbg !254 + %3322 = bitcast float %3310 to i32, !dbg !254 + %3323 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3322, i32 %2795, i32 31), !dbg !254 + %3324 = bitcast float %3311 to i32, !dbg !254 + %3325 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3324, i32 %2798, i32 31), !dbg !254 + %3326 = bitcast float %3312 to i32, !dbg !254 + %3327 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3326, i32 %2795, i32 31), !dbg !254 + %3328 = bitcast float %3313 to i32, !dbg !254 + %3329 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3328, i32 %2798, i32 31), !dbg !254 + %.v153 = select i1 %2561, i32 %3317, i32 %3315, !dbg !254 + %.v154 = select i1 %2561, i32 %3321, i32 %3319, !dbg !254 + %.v155 = select i1 %2561, i32 %3325, i32 %3323, !dbg !254 + %.v156 = select i1 %2561, i32 %3329, i32 %3327, !dbg !254 + %3330 = select i1 %2631, i32 %3291, i32 %3295, !dbg !239 + %3331 = select i1 %2631, i32 %3293, i32 %3297, !dbg !239 + %3332 = select i1 %2631, i32 %3299, i32 %3303, !dbg !239 + %3333 = select i1 %2631, i32 %3301, i32 %3305, !dbg !239 + %3334 = select i1 %2561, i32 %3315, i32 %3317, !dbg !254 + %3335 = select i1 %2561, i32 %3319, i32 %3321, !dbg !254 + %3336 = select i1 %2561, i32 %3323, i32 %3325, !dbg !254 + %3337 = select i1 %2561, i32 %3327, i32 %3329, !dbg !254 + %3338 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %3330, i32 %3331, i32 %.v149, i32 %.v150, i32 %3334, i32 %.v153) #2, !dbg !255 + %3339 = extractvalue { float, float, float, float } %3338, 0, !dbg !255 + %3340 = extractvalue { float, float, float, float } %3338, 1, !dbg !255 + %3341 = extractvalue { float, float, float, float } %3338, 2, !dbg !255 + %3342 = extractvalue { float, float, float, float } %3338, 3, !dbg !255 + %3343 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %3330, i32 %3331, i32 %.v149, i32 %.v150, i32 %3336, i32 %.v155) #2, !dbg !255 + %3344 = extractvalue { float, float, float, float } %3343, 0, !dbg !255 + %3345 = extractvalue { float, float, float, float } %3343, 1, !dbg !255 + %3346 = extractvalue { float, float, float, float } %3343, 2, !dbg !255 + %3347 = extractvalue { float, float, float, float } %3343, 3, !dbg !255 + %3348 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3339, float %3340, float %3341, float %3342, i32 %3332, i32 %3333, i32 %.v151, i32 %.v152, i32 %3335, i32 %.v154) #2, !dbg !255 + %3349 = extractvalue { float, float, float, float } %3348, 0, !dbg !255 + %3350 = extractvalue { float, float, float, float } %3348, 1, !dbg !255 + %3351 = extractvalue { float, float, float, float } %3348, 2, !dbg !255 + %3352 = extractvalue { float, float, float, float } %3348, 3, !dbg !255 + %3353 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3344, float %3345, float %3346, float %3347, i32 %3332, i32 %3333, i32 %.v151, i32 %.v152, i32 %3337, i32 %.v156) #2, !dbg !255 + %3354 = extractvalue { float, float, float, float } %3353, 0, !dbg !255 + %3355 = extractvalue { float, float, float, float } %3353, 1, !dbg !255 + %3356 = extractvalue { float, float, float, float } %3353, 2, !dbg !255 + %3357 = extractvalue { float, float, float, float } %3353, 3, !dbg !255 + %3358 = bitcast float %1788 to i32, !dbg !256 + %3359 = bitcast float %1790 to i32, !dbg !256 + %3360 = bitcast float %1792 to i32, !dbg !256 + %3361 = bitcast float %1794 to i32, !dbg !256 + %3362 = bitcast float %1789 to i32, !dbg !256 + %3363 = bitcast float %1791 to i32, !dbg !256 + %3364 = bitcast float %1793 to i32, !dbg !256 + %3365 = bitcast float %1795 to i32, !dbg !256 + %3366 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %3358, i32 %3359, i32 %3360, i32 %3361, i32 %2986, i32 %2990) #2, !dbg !256 + %3367 = extractvalue { float, float, float, float } %3366, 0, !dbg !256 + %3368 = extractvalue { float, float, float, float } %3366, 1, !dbg !256 + %3369 = extractvalue { float, float, float, float } %3366, 2, !dbg !256 + %3370 = extractvalue { float, float, float, float } %3366, 3, !dbg !256 + %3371 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %3358, i32 %3359, i32 %3360, i32 %3361, i32 %2995, i32 %2999) #2, !dbg !256 + %3372 = extractvalue { float, float, float, float } %3371, 0, !dbg !256 + %3373 = extractvalue { float, float, float, float } %3371, 1, !dbg !256 + %3374 = extractvalue { float, float, float, float } %3371, 2, !dbg !256 + %3375 = extractvalue { float, float, float, float } %3371, 3, !dbg !256 + %3376 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3367, float %3368, float %3369, float %3370, i32 %3362, i32 %3363, i32 %3364, i32 %3365, i32 %2988, i32 %2992) #2, !dbg !256 + %3377 = extractvalue { float, float, float, float } %3376, 0, !dbg !256 + %3378 = extractvalue { float, float, float, float } %3376, 1, !dbg !256 + %3379 = extractvalue { float, float, float, float } %3376, 2, !dbg !256 + %3380 = extractvalue { float, float, float, float } %3376, 3, !dbg !256 + %3381 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3372, float %3373, float %3374, float %3375, i32 %3362, i32 %3363, i32 %3364, i32 %3365, i32 %2997, i32 %3001) #2, !dbg !256 + %3382 = extractvalue { float, float, float, float } %3381, 0, !dbg !256 + %3383 = extractvalue { float, float, float, float } %3381, 1, !dbg !256 + %3384 = extractvalue { float, float, float, float } %3381, 2, !dbg !256 + %3385 = extractvalue { float, float, float, float } %3381, 3, !dbg !256 + %3386 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3377, float %3378, float %3379, float %3380, i32 %3194, i32 %3195, i32 %3196, i32 %3197, i32 %3062, i32 %3063) #2, !dbg !257 + %3387 = extractvalue { float, float, float, float } %3386, 0, !dbg !257 + %3388 = extractvalue { float, float, float, float } %3386, 1, !dbg !257 + %3389 = extractvalue { float, float, float, float } %3386, 2, !dbg !257 + %3390 = extractvalue { float, float, float, float } %3386, 3, !dbg !257 + %3391 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3382, float %3383, float %3384, float %3385, i32 %3194, i32 %3195, i32 %3196, i32 %3197, i32 %3064, i32 %3065) #2, !dbg !257 + %3392 = extractvalue { float, float, float, float } %3391, 0, !dbg !257 + %3393 = extractvalue { float, float, float, float } %3391, 1, !dbg !257 + %3394 = extractvalue { float, float, float, float } %3391, 2, !dbg !257 + %3395 = extractvalue { float, float, float, float } %3391, 3, !dbg !257 + %3396 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3387, float %3388, float %3389, float %3390, i32 %3198, i32 %3199, i32 %3200, i32 %3201, i32 %.v129, i32 %.v130) #2, !dbg !257 + %3397 = extractvalue { float, float, float, float } %3396, 0, !dbg !257 + %3398 = extractvalue { float, float, float, float } %3396, 1, !dbg !257 + %3399 = extractvalue { float, float, float, float } %3396, 2, !dbg !257 + %3400 = extractvalue { float, float, float, float } %3396, 3, !dbg !257 + %3401 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3392, float %3393, float %3394, float %3395, i32 %3198, i32 %3199, i32 %3200, i32 %3201, i32 %.v131, i32 %.v132) #2, !dbg !257 + %3402 = extractvalue { float, float, float, float } %3401, 0, !dbg !257 + %3403 = extractvalue { float, float, float, float } %3401, 1, !dbg !257 + %3404 = extractvalue { float, float, float, float } %3401, 2, !dbg !257 + %3405 = extractvalue { float, float, float, float } %3401, 3, !dbg !257 + %3406 = select i1 %2561, float %3162, float %3163, !dbg !252 + %3407 = select i1 %2561, float %3163, float %3162, !dbg !252 + %3408 = select i1 %2561, float %3164, float %3165, !dbg !252 + %3409 = select i1 %2561, float %3165, float %3164, !dbg !252 + %3410 = select i1 %2561, float %3166, float %3167, !dbg !252 + %3411 = select i1 %2561, float %3167, float %3166, !dbg !252 + %3412 = select i1 %2561, float %3168, float %3169, !dbg !252 + %3413 = select i1 %2561, float %3169, float %3168, !dbg !252 + %3414 = bitcast float %3406 to i32, !dbg !252 + %3415 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3414, i32 %2571, i32 31), !dbg !252 + %3416 = bitcast float %3407 to i32, !dbg !252 + %3417 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3416, i32 %2574, i32 31), !dbg !252 + %3418 = bitcast float %3408 to i32, !dbg !252 + %3419 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3418, i32 %2571, i32 31), !dbg !252 + %3420 = bitcast float %3409 to i32, !dbg !252 + %3421 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3420, i32 %2574, i32 31), !dbg !252 + %3422 = bitcast float %3410 to i32, !dbg !252 + %3423 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3422, i32 %2571, i32 31), !dbg !252 + %3424 = bitcast float %3411 to i32, !dbg !252 + %3425 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3424, i32 %2574, i32 31), !dbg !252 + %3426 = bitcast float %3412 to i32, !dbg !252 + %3427 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3426, i32 %2571, i32 31), !dbg !252 + %3428 = bitcast float %3413 to i32, !dbg !252 + %3429 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3428, i32 %2574, i32 31), !dbg !252 + %.v157 = select i1 %2561, i32 %3417, i32 %3415, !dbg !252 + %.v158 = select i1 %2561, i32 %3421, i32 %3419, !dbg !252 + %.v159 = select i1 %2561, i32 %3425, i32 %3423, !dbg !252 + %.v160 = select i1 %2561, i32 %3429, i32 %3427, !dbg !252 + %3430 = select i1 %2561, i32 %3415, i32 %3417, !dbg !252 + %3431 = select i1 %2561, i32 %3419, i32 %3421, !dbg !252 + %3432 = select i1 %2561, i32 %3423, i32 %3425, !dbg !252 + %3433 = select i1 %2561, i32 %3427, i32 %3429, !dbg !252 + %3434 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3397, float %3398, float %3399, float %3400, i32 %3250, i32 %3251, i32 %3252, i32 %3253, i32 %3430, i32 %3431) #2, !dbg !258 + %3435 = extractvalue { float, float, float, float } %3434, 0, !dbg !258 + %3436 = extractvalue { float, float, float, float } %3434, 1, !dbg !258 + %3437 = extractvalue { float, float, float, float } %3434, 2, !dbg !258 + %3438 = extractvalue { float, float, float, float } %3434, 3, !dbg !258 + %3439 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3402, float %3403, float %3404, float %3405, i32 %3250, i32 %3251, i32 %3252, i32 %3253, i32 %3432, i32 %3433) #2, !dbg !258 + %3440 = extractvalue { float, float, float, float } %3439, 0, !dbg !258 + %3441 = extractvalue { float, float, float, float } %3439, 1, !dbg !258 + %3442 = extractvalue { float, float, float, float } %3439, 2, !dbg !258 + %3443 = extractvalue { float, float, float, float } %3439, 3, !dbg !258 + %3444 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3435, float %3436, float %3437, float %3438, i32 %3254, i32 %3255, i32 %3256, i32 %3257, i32 %.v157, i32 %.v158) #2, !dbg !258 + %3445 = extractvalue { float, float, float, float } %3444, 0, !dbg !258 + %3446 = extractvalue { float, float, float, float } %3444, 1, !dbg !258 + %3447 = extractvalue { float, float, float, float } %3444, 2, !dbg !258 + %3448 = extractvalue { float, float, float, float } %3444, 3, !dbg !258 + %3449 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3440, float %3441, float %3442, float %3443, i32 %3254, i32 %3255, i32 %3256, i32 %3257, i32 %.v159, i32 %.v160) #2, !dbg !258 + %3450 = extractvalue { float, float, float, float } %3449, 0, !dbg !258 + %3451 = extractvalue { float, float, float, float } %3449, 1, !dbg !258 + %3452 = extractvalue { float, float, float, float } %3449, 2, !dbg !258 + %3453 = extractvalue { float, float, float, float } %3449, 3, !dbg !258 + %3454 = select i1 %2784, float %3445, float %3446, !dbg !258 + %3455 = select i1 %2784, float %3446, float %3445, !dbg !258 + %3456 = select i1 %2784, float %3447, float %3448, !dbg !258 + %3457 = select i1 %2784, float %3448, float %3447, !dbg !258 + %3458 = select i1 %2784, float %3450, float %3451, !dbg !258 + %3459 = select i1 %2784, float %3451, float %3450, !dbg !258 + %3460 = select i1 %2784, float %3452, float %3453, !dbg !258 + %3461 = select i1 %2784, float %3453, float %3452, !dbg !258 + %3462 = bitcast float %3454 to i32, !dbg !258 + %3463 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3462, i32 %2795, i32 31), !dbg !258 + %3464 = bitcast float %3455 to i32, !dbg !258 + %3465 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3464, i32 %2798, i32 31), !dbg !258 + %3466 = bitcast float %3456 to i32, !dbg !258 + %3467 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3466, i32 %2795, i32 31), !dbg !258 + %3468 = bitcast float %3457 to i32, !dbg !258 + %3469 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3468, i32 %2798, i32 31), !dbg !258 + %3470 = bitcast float %3458 to i32, !dbg !258 + %3471 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3470, i32 %2795, i32 31), !dbg !258 + %3472 = bitcast float %3459 to i32, !dbg !258 + %3473 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3472, i32 %2798, i32 31), !dbg !258 + %3474 = bitcast float %3460 to i32, !dbg !258 + %3475 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3474, i32 %2795, i32 31), !dbg !258 + %3476 = bitcast float %3461 to i32, !dbg !258 + %3477 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %3476, i32 %2798, i32 31), !dbg !258 + %.v161 = select i1 %2561, i32 %3465, i32 %3463, !dbg !258 + %.v162 = select i1 %2561, i32 %3469, i32 %3467, !dbg !258 + %.v163 = select i1 %2561, i32 %3473, i32 %3471, !dbg !258 + %.v164 = select i1 %2561, i32 %3477, i32 %3475, !dbg !258 + %3478 = select i1 %2561, i32 %3463, i32 %3465, !dbg !258 + %3479 = select i1 %2561, i32 %3467, i32 %3469, !dbg !258 + %3480 = select i1 %2561, i32 %3471, i32 %3473, !dbg !258 + %3481 = select i1 %2561, i32 %3475, i32 %3477, !dbg !258 + %3482 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %3330, i32 %3331, i32 %.v149, i32 %.v150, i32 %3478, i32 %.v161) #2, !dbg !259 + %3483 = extractvalue { float, float, float, float } %3482, 0, !dbg !259 + %3484 = extractvalue { float, float, float, float } %3482, 1, !dbg !259 + %3485 = extractvalue { float, float, float, float } %3482, 2, !dbg !259 + %3486 = extractvalue { float, float, float, float } %3482, 3, !dbg !259 + %3487 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %3330, i32 %3331, i32 %.v149, i32 %.v150, i32 %3480, i32 %.v163) #2, !dbg !259 + %3488 = extractvalue { float, float, float, float } %3487, 0, !dbg !259 + %3489 = extractvalue { float, float, float, float } %3487, 1, !dbg !259 + %3490 = extractvalue { float, float, float, float } %3487, 2, !dbg !259 + %3491 = extractvalue { float, float, float, float } %3487, 3, !dbg !259 + %3492 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3483, float %3484, float %3485, float %3486, i32 %3332, i32 %3333, i32 %.v151, i32 %.v152, i32 %3479, i32 %.v162) #2, !dbg !259 + %3493 = extractvalue { float, float, float, float } %3492, 0, !dbg !259 + %3494 = extractvalue { float, float, float, float } %3492, 1, !dbg !259 + %3495 = extractvalue { float, float, float, float } %3492, 2, !dbg !259 + %3496 = extractvalue { float, float, float, float } %3492, 3, !dbg !259 + %3497 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3488, float %3489, float %3490, float %3491, i32 %3332, i32 %3333, i32 %.v151, i32 %.v152, i32 %3481, i32 %.v164) #2, !dbg !259 + %3498 = extractvalue { float, float, float, float } %3497, 0, !dbg !259 + %3499 = extractvalue { float, float, float, float } %3497, 1, !dbg !259 + %3500 = extractvalue { float, float, float, float } %3497, 2, !dbg !259 + %3501 = extractvalue { float, float, float, float } %3497, 3, !dbg !259 + %.idx165 = shl nuw nsw i64 %32, 12, !dbg !260 + %3502 = getelementptr i8, ptr addrspace(1) %1036, i64 %.idx165, !dbg !260 + %3503 = zext nneg i32 %1039 to i64, !dbg !261 + %3504 = getelementptr bfloat, ptr addrspace(1) %3502, i64 %3503, !dbg !261 + %.idx166 = shl nuw nsw i64 %34, 12, !dbg !262 + %3505 = getelementptr i8, ptr addrspace(1) %1036, i64 %.idx166, !dbg !262 + %3506 = getelementptr bfloat, ptr addrspace(1) %3505, i64 %3503, !dbg !263 + %3507 = or disjoint i32 %1039, 16, !dbg !264 + %3508 = zext nneg i32 %3507 to i64, !dbg !265 + %3509 = getelementptr bfloat, ptr addrspace(1) %3505, i64 %3508, !dbg !265 + %.idx167 = shl nuw nsw i64 %36, 12, !dbg !266 + %3510 = getelementptr i8, ptr addrspace(1) %1036, i64 %.idx167, !dbg !266 + %3511 = getelementptr bfloat, ptr addrspace(1) %3510, i64 %3503, !dbg !267 + %3512 = getelementptr bfloat, ptr addrspace(1) %3510, i64 %3508, !dbg !268 + %3513 = or disjoint i32 %1039, 32, !dbg !269 + %3514 = zext nneg i32 %3513 to i64, !dbg !270 + %3515 = getelementptr bfloat, ptr addrspace(1) %3510, i64 %3514, !dbg !270 + %.idx168 = shl nuw nsw i64 %38, 12, !dbg !271 + %3516 = getelementptr i8, ptr addrspace(1) %1036, i64 %.idx168, !dbg !271 + %3517 = getelementptr bfloat, ptr addrspace(1) %3516, i64 %3503, !dbg !272 + %3518 = getelementptr bfloat, ptr addrspace(1) %3516, i64 %3508, !dbg !273 + %3519 = getelementptr bfloat, ptr addrspace(1) %3516, i64 %3514, !dbg !274 + %3520 = getelementptr i8, ptr addrspace(1) %3517, i64 96, !dbg !275 + %3521 = fptrunc float %2516 to bfloat, !dbg !276 + %3522 = fptrunc float %2518 to bfloat, !dbg !276 + %3523 = fptrunc float %2520 to bfloat, !dbg !276 + %3524 = fptrunc float %2522 to bfloat, !dbg !276 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !277 + %3525 = and i32 %230, 176, !dbg !277 + %3526 = select i1 %2561, i32 0, i32 320, !dbg !277 + %3527 = shl nuw nsw i32 %1038, 6, !dbg !277 + %3528 = or i32 %3527, %2793, !dbg !277 + %3529 = and i32 %3528, 64, !dbg !277 + %3530 = or disjoint i32 %3526, %3525, !dbg !277 + %3531 = xor i32 %3530, %3529, !dbg !277 + %3532 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %3531, !dbg !277 + %3533 = insertelement <2 x bfloat> poison, bfloat %3521, i64 0, !dbg !277 + %3534 = insertelement <2 x bfloat> %3533, bfloat %3522, i64 1, !dbg !277 + %3535 = bitcast <2 x bfloat> %3534 to i32, !dbg !277 + %3536 = insertelement <2 x bfloat> poison, bfloat %3523, i64 0, !dbg !277 + %3537 = insertelement <2 x bfloat> %3536, bfloat %3524, i64 1, !dbg !277 + %3538 = bitcast <2 x bfloat> %3537 to i32, !dbg !277 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x2.b16.p3(ptr addrspace(3) %3532, i32 %3535, i32 %3538), !dbg !277 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !277 + %3539 = shl nuw nsw i32 %1038, 3, !dbg !277 + %3540 = or i32 %3539, %1037, !dbg !277 + %3541 = shl nuw nsw i32 %3540, 3, !dbg !277 + %3542 = and i32 %3541, 176, !dbg !277 + %3543 = shl nuw nsw i32 %23, 6, !dbg !277 + %3544 = and i32 %3543, 64, !dbg !277 + %3545 = and i32 %3540, 8, !dbg !277 + %3546 = icmp eq i32 %3545, 0, !dbg !277 + %3547 = select i1 %3546, i32 0, i32 320, !dbg !277 + %3548 = or disjoint i32 %3542, %3544, !dbg !277 + %3549 = xor i32 %3548, %3547, !dbg !277 + %3550 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %3549, !dbg !277 + %3551 = tail call i32 @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x1.b16.p3(ptr addrspace(3) %3550), !dbg !277 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %3551, ptr addrspace(1) %3504, i1 %33) #2, !dbg !277 + %3552 = fptrunc float %2696 to bfloat, !dbg !278 + %3553 = fptrunc float %2697 to bfloat, !dbg !278 + %3554 = fptrunc float %2698 to bfloat, !dbg !278 + %3555 = fptrunc float %2699 to bfloat, !dbg !278 + %3556 = fptrunc float %2700 to bfloat, !dbg !278 + %3557 = fptrunc float %2701 to bfloat, !dbg !278 + %3558 = fptrunc float %2702 to bfloat, !dbg !278 + %3559 = fptrunc float %2703 to bfloat, !dbg !278 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !279 + %3560 = and i32 %2793, 64, !dbg !279 + %3561 = xor i32 %3530, %3560, !dbg !279 + %3562 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %3561, !dbg !279 + %3563 = insertelement <2 x bfloat> poison, bfloat %3552, i64 0, !dbg !279 + %3564 = insertelement <2 x bfloat> %3563, bfloat %3553, i64 1, !dbg !279 + %3565 = bitcast <2 x bfloat> %3564 to i32, !dbg !279 + %3566 = insertelement <2 x bfloat> poison, bfloat %3554, i64 0, !dbg !279 + %3567 = insertelement <2 x bfloat> %3566, bfloat %3555, i64 1, !dbg !279 + %3568 = bitcast <2 x bfloat> %3567 to i32, !dbg !279 + %3569 = insertelement <2 x bfloat> poison, bfloat %3556, i64 0, !dbg !279 + %3570 = insertelement <2 x bfloat> %3569, bfloat %3557, i64 1, !dbg !279 + %3571 = bitcast <2 x bfloat> %3570 to i32, !dbg !279 + %3572 = insertelement <2 x bfloat> poison, bfloat %3558, i64 0, !dbg !279 + %3573 = insertelement <2 x bfloat> %3572, bfloat %3559, i64 1, !dbg !279 + %3574 = bitcast <2 x bfloat> %3573 to i32, !dbg !279 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %3562, i32 %3565, i32 %3568, i32 %3571, i32 %3574), !dbg !279 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !279 + %3575 = tail call i32 @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x1.b16.p3(ptr addrspace(3) %3550), !dbg !279 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %3575, ptr addrspace(1) %3506, i1 %35) #2, !dbg !279 + %3576 = fptrunc float %2538 to bfloat, !dbg !280 + %3577 = fptrunc float %2540 to bfloat, !dbg !280 + %3578 = fptrunc float %.pn60.lcssa to bfloat, !dbg !280 + %3579 = fptrunc float %.pn58.lcssa to bfloat, !dbg !280 + %3580 = fptrunc float %.pn56.lcssa to bfloat, !dbg !280 + %3581 = fptrunc float %.pn54.lcssa to bfloat, !dbg !280 + %3582 = fptrunc float %2542 to bfloat, !dbg !280 + %3583 = fptrunc float %2544 to bfloat, !dbg !280 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !281 + %3584 = insertelement <2 x bfloat> poison, bfloat %3576, i64 0, !dbg !281 + %3585 = insertelement <2 x bfloat> %3584, bfloat %3577, i64 1, !dbg !281 + %3586 = bitcast <2 x bfloat> %3585 to i32, !dbg !281 + %3587 = insertelement <2 x bfloat> poison, bfloat %3578, i64 0, !dbg !281 + %3588 = insertelement <2 x bfloat> %3587, bfloat %3579, i64 1, !dbg !281 + %3589 = bitcast <2 x bfloat> %3588 to i32, !dbg !281 + %3590 = insertelement <2 x bfloat> poison, bfloat %3580, i64 0, !dbg !281 + %3591 = insertelement <2 x bfloat> %3590, bfloat %3581, i64 1, !dbg !281 + %3592 = bitcast <2 x bfloat> %3591 to i32, !dbg !281 + %3593 = insertelement <2 x bfloat> poison, bfloat %3582, i64 0, !dbg !281 + %3594 = insertelement <2 x bfloat> %3593, bfloat %3583, i64 1, !dbg !281 + %3595 = bitcast <2 x bfloat> %3594 to i32, !dbg !281 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %3562, i32 %3586, i32 %3589, i32 %3592, i32 %3595), !dbg !281 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !281 + %3596 = tail call i32 @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x1.b16.p3(ptr addrspace(3) %3550), !dbg !281 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %3596, ptr addrspace(1) %3509, i1 %35) #2, !dbg !281 + %3597 = fptrunc float %3162 to bfloat, !dbg !282 + %3598 = fptrunc float %3163 to bfloat, !dbg !282 + %3599 = fptrunc float %3164 to bfloat, !dbg !282 + %3600 = fptrunc float %3165 to bfloat, !dbg !282 + %3601 = fptrunc float %3166 to bfloat, !dbg !282 + %3602 = fptrunc float %3167 to bfloat, !dbg !282 + %3603 = fptrunc float %3168 to bfloat, !dbg !282 + %3604 = fptrunc float %3169 to bfloat, !dbg !282 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !283 + %3605 = insertelement <2 x bfloat> poison, bfloat %3597, i64 0, !dbg !283 + %3606 = insertelement <2 x bfloat> %3605, bfloat %3598, i64 1, !dbg !283 + %3607 = bitcast <2 x bfloat> %3606 to i32, !dbg !283 + %3608 = insertelement <2 x bfloat> poison, bfloat %3599, i64 0, !dbg !283 + %3609 = insertelement <2 x bfloat> %3608, bfloat %3600, i64 1, !dbg !283 + %3610 = bitcast <2 x bfloat> %3609 to i32, !dbg !283 + %3611 = insertelement <2 x bfloat> poison, bfloat %3601, i64 0, !dbg !283 + %3612 = insertelement <2 x bfloat> %3611, bfloat %3602, i64 1, !dbg !283 + %3613 = bitcast <2 x bfloat> %3612 to i32, !dbg !283 + %3614 = insertelement <2 x bfloat> poison, bfloat %3603, i64 0, !dbg !283 + %3615 = insertelement <2 x bfloat> %3614, bfloat %3604, i64 1, !dbg !283 + %3616 = bitcast <2 x bfloat> %3615 to i32, !dbg !283 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %3562, i32 %3607, i32 %3610, i32 %3613, i32 %3616), !dbg !283 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !283 + %3617 = tail call i32 @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x1.b16.p3(ptr addrspace(3) %3550), !dbg !283 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %3617, ptr addrspace(1) %3511, i1 %37) #2, !dbg !283 + %3618 = fptrunc float %2841 to bfloat, !dbg !284 + %3619 = fptrunc float %2842 to bfloat, !dbg !284 + %3620 = fptrunc float %2843 to bfloat, !dbg !284 + %3621 = fptrunc float %2844 to bfloat, !dbg !284 + %3622 = fptrunc float %2845 to bfloat, !dbg !284 + %3623 = fptrunc float %2846 to bfloat, !dbg !284 + %3624 = fptrunc float %2847 to bfloat, !dbg !284 + %3625 = fptrunc float %2848 to bfloat, !dbg !284 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !285 + %3626 = insertelement <2 x bfloat> poison, bfloat %3618, i64 0, !dbg !285 + %3627 = insertelement <2 x bfloat> %3626, bfloat %3619, i64 1, !dbg !285 + %3628 = bitcast <2 x bfloat> %3627 to i32, !dbg !285 + %3629 = insertelement <2 x bfloat> poison, bfloat %3620, i64 0, !dbg !285 + %3630 = insertelement <2 x bfloat> %3629, bfloat %3621, i64 1, !dbg !285 + %3631 = bitcast <2 x bfloat> %3630 to i32, !dbg !285 + %3632 = insertelement <2 x bfloat> poison, bfloat %3622, i64 0, !dbg !285 + %3633 = insertelement <2 x bfloat> %3632, bfloat %3623, i64 1, !dbg !285 + %3634 = bitcast <2 x bfloat> %3633 to i32, !dbg !285 + %3635 = insertelement <2 x bfloat> poison, bfloat %3624, i64 0, !dbg !285 + %3636 = insertelement <2 x bfloat> %3635, bfloat %3625, i64 1, !dbg !285 + %3637 = bitcast <2 x bfloat> %3636 to i32, !dbg !285 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %3562, i32 %3628, i32 %3631, i32 %3634, i32 %3637), !dbg !285 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !285 + %3638 = tail call i32 @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x1.b16.p3(ptr addrspace(3) %3550), !dbg !285 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %3638, ptr addrspace(1) %3512, i1 %37) #2, !dbg !285 + %3639 = fptrunc float %2546 to bfloat, !dbg !286 + %3640 = fptrunc float %2548 to bfloat, !dbg !286 + %3641 = fptrunc float %.pn76.lcssa to bfloat, !dbg !286 + %3642 = fptrunc float %.pn74.lcssa to bfloat, !dbg !286 + %3643 = fptrunc float %.pn72.lcssa to bfloat, !dbg !286 + %3644 = fptrunc float %.pn70.lcssa to bfloat, !dbg !286 + %3645 = fptrunc float %2550 to bfloat, !dbg !286 + %3646 = fptrunc float %2552 to bfloat, !dbg !286 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !287 + %3647 = insertelement <2 x bfloat> poison, bfloat %3639, i64 0, !dbg !287 + %3648 = insertelement <2 x bfloat> %3647, bfloat %3640, i64 1, !dbg !287 + %3649 = bitcast <2 x bfloat> %3648 to i32, !dbg !287 + %3650 = insertelement <2 x bfloat> poison, bfloat %3641, i64 0, !dbg !287 + %3651 = insertelement <2 x bfloat> %3650, bfloat %3642, i64 1, !dbg !287 + %3652 = bitcast <2 x bfloat> %3651 to i32, !dbg !287 + %3653 = insertelement <2 x bfloat> poison, bfloat %3643, i64 0, !dbg !287 + %3654 = insertelement <2 x bfloat> %3653, bfloat %3644, i64 1, !dbg !287 + %3655 = bitcast <2 x bfloat> %3654 to i32, !dbg !287 + %3656 = insertelement <2 x bfloat> poison, bfloat %3645, i64 0, !dbg !287 + %3657 = insertelement <2 x bfloat> %3656, bfloat %3646, i64 1, !dbg !287 + %3658 = bitcast <2 x bfloat> %3657 to i32, !dbg !287 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %3562, i32 %3649, i32 %3652, i32 %3655, i32 %3658), !dbg !287 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !287 + %3659 = tail call i32 @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x1.b16.p3(ptr addrspace(3) %3550), !dbg !287 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %3659, ptr addrspace(1) %3515, i1 %37) #2, !dbg !287 + %3660 = insertelement <2 x float> poison, float %3493, i64 0, !dbg !288 + %3661 = insertelement <2 x float> %3660, float %3494, i64 1, !dbg !288 + %3662 = fsub <2 x float> zeroinitializer, %3661, !dbg !288 + %3663 = fptrunc <2 x float> %3662 to <2 x bfloat>, !dbg !289 + %3664 = insertelement <2 x float> poison, float %3495, i64 0, !dbg !288 + %3665 = insertelement <2 x float> %3664, float %3496, i64 1, !dbg !288 + %3666 = fsub <2 x float> zeroinitializer, %3665, !dbg !288 + %3667 = fptrunc <2 x float> %3666 to <2 x bfloat>, !dbg !289 + %3668 = insertelement <2 x float> poison, float %3498, i64 0, !dbg !288 + %3669 = insertelement <2 x float> %3668, float %3499, i64 1, !dbg !288 + %3670 = fsub <2 x float> zeroinitializer, %3669, !dbg !288 + %3671 = fptrunc <2 x float> %3670 to <2 x bfloat>, !dbg !289 + %3672 = insertelement <2 x float> poison, float %3500, i64 0, !dbg !288 + %3673 = insertelement <2 x float> %3672, float %3501, i64 1, !dbg !288 + %3674 = fsub <2 x float> zeroinitializer, %3673, !dbg !288 + %3675 = fptrunc <2 x float> %3674 to <2 x bfloat>, !dbg !289 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !290 + %3676 = bitcast <2 x bfloat> %3663 to i32, !dbg !290 + %3677 = bitcast <2 x bfloat> %3667 to i32, !dbg !290 + %3678 = bitcast <2 x bfloat> %3671 to i32, !dbg !290 + %3679 = bitcast <2 x bfloat> %3675 to i32, !dbg !290 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %3562, i32 %3676, i32 %3677, i32 %3678, i32 %3679), !dbg !290 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !290 + %3680 = tail call i32 @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x1.b16.p3(ptr addrspace(3) %3550), !dbg !290 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %3680, ptr addrspace(1) %3517, i1 %39) #2, !dbg !290 + %3681 = insertelement <2 x float> poison, float %3349, i64 0, !dbg !291 + %3682 = insertelement <2 x float> %3681, float %3350, i64 1, !dbg !291 + %3683 = fsub <2 x float> zeroinitializer, %3682, !dbg !291 + %3684 = fptrunc <2 x float> %3683 to <2 x bfloat>, !dbg !292 + %3685 = insertelement <2 x float> poison, float %3351, i64 0, !dbg !291 + %3686 = insertelement <2 x float> %3685, float %3352, i64 1, !dbg !291 + %3687 = fsub <2 x float> zeroinitializer, %3686, !dbg !291 + %3688 = fptrunc <2 x float> %3687 to <2 x bfloat>, !dbg !292 + %3689 = insertelement <2 x float> poison, float %3354, i64 0, !dbg !291 + %3690 = insertelement <2 x float> %3689, float %3355, i64 1, !dbg !291 + %3691 = fsub <2 x float> zeroinitializer, %3690, !dbg !291 + %3692 = fptrunc <2 x float> %3691 to <2 x bfloat>, !dbg !292 + %3693 = insertelement <2 x float> poison, float %3356, i64 0, !dbg !291 + %3694 = insertelement <2 x float> %3693, float %3357, i64 1, !dbg !291 + %3695 = fsub <2 x float> zeroinitializer, %3694, !dbg !291 + %3696 = fptrunc <2 x float> %3695 to <2 x bfloat>, !dbg !292 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !293 + %3697 = bitcast <2 x bfloat> %3684 to i32, !dbg !293 + %3698 = bitcast <2 x bfloat> %3688 to i32, !dbg !293 + %3699 = bitcast <2 x bfloat> %3692 to i32, !dbg !293 + %3700 = bitcast <2 x bfloat> %3696 to i32, !dbg !293 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %3562, i32 %3697, i32 %3698, i32 %3699, i32 %3700), !dbg !293 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !293 + %3701 = tail call i32 @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x1.b16.p3(ptr addrspace(3) %3550), !dbg !293 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %3701, ptr addrspace(1) %3518, i1 %39) #2, !dbg !293 + %3702 = insertelement <2 x float> poison, float %2972, i64 0, !dbg !294 + %3703 = insertelement <2 x float> %3702, float %2973, i64 1, !dbg !294 + %3704 = fsub <2 x float> zeroinitializer, %3703, !dbg !294 + %3705 = fptrunc <2 x float> %3704 to <2 x bfloat>, !dbg !295 + %3706 = insertelement <2 x float> poison, float %2974, i64 0, !dbg !294 + %3707 = insertelement <2 x float> %3706, float %2975, i64 1, !dbg !294 + %3708 = fsub <2 x float> zeroinitializer, %3707, !dbg !294 + %3709 = fptrunc <2 x float> %3708 to <2 x bfloat>, !dbg !295 + %3710 = insertelement <2 x float> poison, float %2977, i64 0, !dbg !294 + %3711 = insertelement <2 x float> %3710, float %2978, i64 1, !dbg !294 + %3712 = fsub <2 x float> zeroinitializer, %3711, !dbg !294 + %3713 = fptrunc <2 x float> %3712 to <2 x bfloat>, !dbg !295 + %3714 = insertelement <2 x float> poison, float %2979, i64 0, !dbg !294 + %3715 = insertelement <2 x float> %3714, float %2980, i64 1, !dbg !294 + %3716 = fsub <2 x float> zeroinitializer, %3715, !dbg !294 + %3717 = fptrunc <2 x float> %3716 to <2 x bfloat>, !dbg !295 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !296 + %3718 = bitcast <2 x bfloat> %3705 to i32, !dbg !296 + %3719 = bitcast <2 x bfloat> %3709 to i32, !dbg !296 + %3720 = bitcast <2 x bfloat> %3713 to i32, !dbg !296 + %3721 = bitcast <2 x bfloat> %3717 to i32, !dbg !296 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %3562, i32 %3718, i32 %3719, i32 %3720, i32 %3721), !dbg !296 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !296 + %3722 = tail call i32 @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x1.b16.p3(ptr addrspace(3) %3550), !dbg !296 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %3722, ptr addrspace(1) %3519, i1 %39) #2, !dbg !296 + %3723 = fptrunc float %2554 to bfloat, !dbg !297 + %3724 = fptrunc float %2556 to bfloat, !dbg !297 + %3725 = fptrunc float %.pn92.lcssa to bfloat, !dbg !297 + %3726 = fptrunc float %.pn90.lcssa to bfloat, !dbg !297 + %3727 = fptrunc float %.pn88.lcssa to bfloat, !dbg !297 + %3728 = fptrunc float %.pn86.lcssa to bfloat, !dbg !297 + %3729 = fptrunc float %2558 to bfloat, !dbg !297 + %3730 = fptrunc float %2560 to bfloat, !dbg !297 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !298 + %3731 = insertelement <2 x bfloat> poison, bfloat %3723, i64 0, !dbg !298 + %3732 = insertelement <2 x bfloat> %3731, bfloat %3724, i64 1, !dbg !298 + %3733 = bitcast <2 x bfloat> %3732 to i32, !dbg !298 + %3734 = insertelement <2 x bfloat> poison, bfloat %3725, i64 0, !dbg !298 + %3735 = insertelement <2 x bfloat> %3734, bfloat %3726, i64 1, !dbg !298 + %3736 = bitcast <2 x bfloat> %3735 to i32, !dbg !298 + %3737 = insertelement <2 x bfloat> poison, bfloat %3727, i64 0, !dbg !298 + %3738 = insertelement <2 x bfloat> %3737, bfloat %3728, i64 1, !dbg !298 + %3739 = bitcast <2 x bfloat> %3738 to i32, !dbg !298 + %3740 = insertelement <2 x bfloat> poison, bfloat %3729, i64 0, !dbg !298 + %3741 = insertelement <2 x bfloat> %3740, bfloat %3730, i64 1, !dbg !298 + %3742 = bitcast <2 x bfloat> %3741 to i32, !dbg !298 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %3562, i32 %3733, i32 %3736, i32 %3739, i32 %3742), !dbg !298 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !298 + %3743 = tail call i32 @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x1.b16.p3(ptr addrspace(3) %3550), !dbg !298 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %3743, ptr addrspace(1) %3520, i1 %39) #2, !dbg !298 + br label %common.ret, !dbg !299 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 65535) i32 @llvm.nvvm.read.ptx.sreg.ctaid.y() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: nounwind +declare void @llvm.nvvm.cp.async.commit.group() #2 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #3 + +; Function Attrs: nounwind +declare void @llvm.nvvm.cp.async.wait.group(i32 immarg) #2 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.idx.i32(i32, i32, i32, i32) #4 + +; Function Attrs: convergent nocallback nounwind memory(argmem: write) +declare void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x2.b16.p3(ptr addrspace(3) writeonly captures(none), i32, i32) #5 + +; Function Attrs: convergent nocallback nofree nounwind memory(argmem: read) +declare i32 @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x1.b16.p3(ptr addrspace(3) readonly captures(none)) #6 + +; Function Attrs: convergent nocallback nounwind memory(argmem: write) +declare void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) writeonly captures(none), i32, i32, i32, i32) #5 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #4 + +; Function Attrs: convergent nocallback nofree nounwind memory(argmem: read) +declare { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) readonly captures(none)) #6 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #7 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.ex2.approx.ftz.f32(float) #8 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.ex2.approx.f32(float) #8 + +; Function Attrs: nocallback nocreateundeforpoison nofree nosync nounwind speculatable willreturn memory(none) +declare i2 @llvm.bitreverse.i2(i2) #9 + +; Function Attrs: nocallback nocreateundeforpoison nofree nosync nounwind speculatable willreturn memory(none) +declare i64 @llvm.umin.i64(i64, i64) #9 + +attributes #0 = { nounwind "nvvm.reqntid"="128" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { nounwind } +attributes #3 = { convergent nocallback nounwind } +attributes #4 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #5 = { convergent nocallback nounwind memory(argmem: write) } +attributes #6 = { convergent nocallback nofree nounwind memory(argmem: read) } +attributes #7 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #8 = { mustprogress nocallback nofree nosync nounwind willreturn memory(none) } +attributes #9 = { nocallback nocreateundeforpoison nofree nosync nounwind speculatable willreturn memory(none) } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "chunk_fwd.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "chunk_gated_delta_rule_fwd_kkt_solve_kernel", linkageName: "chunk_gated_delta_rule_fwd_kkt_solve_kernel", scope: !1, file: !1, line: 40, type: !6, scopeLine: 40, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !10, !8, !8, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 70, column: 30, scope: !5) +!14 = !DILocation(line: 70, column: 36, scope: !5) +!15 = !DILocation(line: 78, column: 25, scope: !5) +!16 = !DILocation(line: 80, column: 13, scope: !5) +!17 = !DILocation(line: 80, column: 19, scope: !5) +!18 = !DILocation(line: 0, scope: !5) +!19 = !DILocation(line: 70, column: 61, scope: !5) +!20 = !DILocation(line: 70, column: 67, scope: !5) +!21 = !DILocation(line: 71, column: 34, scope: !5) +!22 = !DILocation(line: 84, column: 23, scope: !5) +!23 = !DILocation(line: 85, column: 23, scope: !5) +!24 = !DILocation(line: 86, column: 23, scope: !5) +!25 = !DILocation(line: 88, column: 16, scope: !5) +!26 = !DILocation(line: 88, column: 20, scope: !5) +!27 = !DILocation(line: 88, column: 9, scope: !5) +!28 = !DILocation(line: 91, column: 23, scope: !5) +!29 = !DILocation(line: 92, column: 21, scope: !5) +!30 = !DILocation(line: 92, column: 28, scope: !5) +!31 = !DILocation(line: 93, column: 21, scope: !5) +!32 = !DILocation(line: 93, column: 28, scope: !5) +!33 = !DILocation(line: 94, column: 21, scope: !5) +!34 = !DILocation(line: 94, column: 28, scope: !5) +!35 = !DILocation(line: 95, column: 21, scope: !5) +!36 = !DILocation(line: 95, column: 28, scope: !5) +!37 = !DILocation(line: 98, scope: !5) +!38 = !DILocation(line: 98, column: 51, scope: !5) +!39 = !DILocation(line: 98, column: 35, scope: !5) +!40 = !DILocation(line: 99, column: 51, scope: !5) +!41 = !DILocation(line: 99, column: 35, scope: !5) +!42 = !DILocation(line: 100, column: 51, scope: !5) +!43 = !DILocation(line: 100, column: 35, scope: !5) +!44 = !DILocation(line: 101, column: 51, scope: !5) +!45 = !DILocation(line: 101, column: 35, scope: !5) +!46 = !DILocation(line: 114, column: 23, scope: !5) +!47 = !DILocation(line: 102, column: 19, scope: !5) +!48 = !DILocation(line: 103, column: 19, scope: !5) +!49 = !DILocation(line: 104, column: 19, scope: !5) +!50 = !DILocation(line: 105, column: 19, scope: !5) +!51 = !DILocation(line: 109, scope: !5) +!52 = !DILocation(line: 109, column: 36, scope: !5) +!53 = !DILocation(line: 110, column: 36, scope: !5) +!54 = !DILocation(line: 111, column: 36, scope: !5) +!55 = !DILocation(line: 112, column: 36, scope: !5) +!56 = !DILocation(line: 115, column: 23, scope: !5) +!57 = !DILocation(line: 116, column: 23, scope: !5) +!58 = !DILocation(line: 117, column: 23, scope: !5) +!59 = !DILocation(line: 138, column: 38, scope: !5) +!60 = !DILocation(line: 139, column: 19, scope: !5) +!61 = !DILocation(line: 144, column: 19, scope: !5) +!62 = !DILocation(line: 139, column: 52, scope: !5) +!63 = !DILocation(line: 140, column: 23, scope: !5) +!64 = !DILocation(line: 137, column: 21, scope: !5) +!65 = !DILocation(line: 142, column: 45, scope: !5) +!66 = !DILocation(line: 144, column: 11, scope: !5) +!67 = !DILocation(line: 145, column: 56, scope: !5) +!68 = !DILocation(line: 146, column: 27, scope: !5) +!69 = !DILocation(line: 148, column: 42, scope: !5) +!70 = !DILocation(line: 148, column: 49, scope: !5) +!71 = !DILocation(line: 150, column: 49, scope: !5) +!72 = !DILocation(line: 152, column: 15, scope: !5) +!73 = !DILocation(line: 153, column: 60, scope: !5) +!74 = !DILocation(line: 154, column: 31, scope: !5) +!75 = !DILocation(line: 156, column: 46, scope: !5) +!76 = !DILocation(line: 156, column: 53, scope: !5) +!77 = !DILocation(line: 158, column: 53, scope: !5) +!78 = !DILocation(line: 159, column: 53, scope: !5) +!79 = !DILocation(line: 161, column: 19, scope: !5) +!80 = !DILocation(line: 162, column: 64, scope: !5) +!81 = !DILocation(line: 163, column: 35, scope: !5) +!82 = !DILocation(line: 165, column: 50, scope: !5) +!83 = !DILocation(line: 165, column: 57, scope: !5) +!84 = !DILocation(line: 167, column: 57, scope: !5) +!85 = !DILocation(line: 168, column: 57, scope: !5) +!86 = !DILocation(line: 169, column: 57, scope: !5) +!87 = !DILocation(line: 89, column: 9, scope: !5) +!88 = !DILocation(line: 102, column: 51, scope: !5) +!89 = !DILocation(line: 103, column: 51, scope: !5) +!90 = !DILocation(line: 104, column: 51, scope: !5) +!91 = !DILocation(line: 105, column: 51, scope: !5) +!92 = !DILocation(line: 179, column: 25, scope: !5) +!93 = !DILocation(line: 180, column: 26, scope: !5) +!94 = !DILocation(line: 183, column: 32, scope: !5) +!95 = !DILocation(line: 183, column: 49, scope: !5) +!96 = !DILocation(line: 183, column: 86, scope: !5) +!97 = !DILocation(line: 47, column: 33, scope: !98, inlinedAt: !100) +!98 = distinct !DILexicalBlockFile(scope: !5, file: !99, discriminator: 0) +!99 = !DIFile(filename: "op.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils") +!100 = !DILocation(line: 183, column: 70, scope: !101) +!101 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!102 = !DILocation(line: 183, column: 102, scope: !5) +!103 = !DILocation(line: 183, column: 17, scope: !5) +!104 = !DILocation(line: 184, column: 32, scope: !5) +!105 = !DILocation(line: 184, column: 49, scope: !5) +!106 = !DILocation(line: 184, column: 86, scope: !5) +!107 = !DILocation(line: 47, column: 33, scope: !98, inlinedAt: !108) +!108 = !DILocation(line: 184, column: 70, scope: !101) +!109 = !DILocation(line: 184, column: 102, scope: !5) +!110 = !DILocation(line: 184, column: 17, scope: !5) +!111 = !DILocation(line: 185, column: 32, scope: !5) +!112 = !DILocation(line: 185, column: 49, scope: !5) +!113 = !DILocation(line: 185, column: 86, scope: !5) +!114 = !DILocation(line: 47, column: 33, scope: !98, inlinedAt: !115) +!115 = !DILocation(line: 185, column: 70, scope: !101) +!116 = !DILocation(line: 185, column: 102, scope: !5) +!117 = !DILocation(line: 185, column: 17, scope: !5) +!118 = !DILocation(line: 186, column: 32, scope: !5) +!119 = !DILocation(line: 186, column: 49, scope: !5) +!120 = !DILocation(line: 186, column: 86, scope: !5) +!121 = !DILocation(line: 47, column: 33, scope: !98, inlinedAt: !122) +!122 = !DILocation(line: 186, column: 70, scope: !101) +!123 = !DILocation(line: 186, column: 102, scope: !5) +!124 = !DILocation(line: 186, column: 17, scope: !5) +!125 = !DILocation(line: 188, column: 43, scope: !5) +!126 = !DILocation(line: 188, column: 80, scope: !5) +!127 = !DILocation(line: 47, column: 33, scope: !98, inlinedAt: !128) +!128 = !DILocation(line: 188, column: 64, scope: !101) +!129 = !DILocation(line: 188, column: 96, scope: !5) +!130 = !DILocation(line: 188, column: 17, scope: !5) +!131 = !DILocation(line: 189, column: 43, scope: !5) +!132 = !DILocation(line: 189, column: 80, scope: !5) +!133 = !DILocation(line: 47, column: 33, scope: !98, inlinedAt: !134) +!134 = !DILocation(line: 189, column: 64, scope: !101) +!135 = !DILocation(line: 189, column: 96, scope: !5) +!136 = !DILocation(line: 189, column: 17, scope: !5) +!137 = !DILocation(line: 190, column: 43, scope: !5) +!138 = !DILocation(line: 190, column: 80, scope: !5) +!139 = !DILocation(line: 47, column: 33, scope: !98, inlinedAt: !140) +!140 = !DILocation(line: 190, column: 64, scope: !101) +!141 = !DILocation(line: 190, column: 96, scope: !5) +!142 = !DILocation(line: 190, column: 17, scope: !5) +!143 = !DILocation(line: 191, column: 43, scope: !5) +!144 = !DILocation(line: 191, column: 80, scope: !5) +!145 = !DILocation(line: 47, column: 33, scope: !98, inlinedAt: !146) +!146 = !DILocation(line: 191, column: 64, scope: !101) +!147 = !DILocation(line: 191, column: 96, scope: !5) +!148 = !DILocation(line: 191, column: 17, scope: !5) +!149 = !DILocation(line: 192, column: 43, scope: !5) +!150 = !DILocation(line: 192, column: 80, scope: !5) +!151 = !DILocation(line: 47, column: 33, scope: !98, inlinedAt: !152) +!152 = !DILocation(line: 192, column: 64, scope: !101) +!153 = !DILocation(line: 192, column: 96, scope: !5) +!154 = !DILocation(line: 192, column: 17, scope: !5) +!155 = !DILocation(line: 193, column: 43, scope: !5) +!156 = !DILocation(line: 193, column: 80, scope: !5) +!157 = !DILocation(line: 47, column: 33, scope: !98, inlinedAt: !158) +!158 = !DILocation(line: 193, column: 64, scope: !101) +!159 = !DILocation(line: 193, column: 96, scope: !5) +!160 = !DILocation(line: 193, column: 17, scope: !5) +!161 = !DILocation(line: 201, column: 20, scope: !5) +!162 = !DILocation(line: 202, column: 20, scope: !5) +!163 = !DILocation(line: 207, column: 20, scope: !5) +!164 = !DILocation(line: 203, column: 20, scope: !5) +!165 = !DILocation(line: 209, column: 20, scope: !5) +!166 = !DILocation(line: 204, column: 20, scope: !5) +!167 = !DILocation(line: 212, column: 20, scope: !5) +!168 = !DILocation(line: 208, column: 20, scope: !5) +!169 = !DILocation(line: 210, column: 20, scope: !5) +!170 = !DILocation(line: 211, column: 20, scope: !5) +!171 = !DILocation(line: 222, column: 14, scope: !5) +!172 = !DILocation(line: 223, column: 14, scope: !5) +!173 = !DILocation(line: 224, column: 14, scope: !5) +!174 = !DILocation(line: 225, column: 14, scope: !5) +!175 = !DILocation(line: 227, column: 34, scope: !5) +!176 = !DILocation(line: 227, column: 22, scope: !5) +!177 = !DILocation(line: 228, column: 61, scope: !5) +!178 = !DILocation(line: 228, column: 40, scope: !5) +!179 = !DILocation(line: 263, column: 15, scope: !180, inlinedAt: !182) +!180 = distinct !DILexicalBlockFile(scope: !5, file: !181, discriminator: 0) +!181 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!182 = !DILocation(line: 293, column: 36, scope: !180, inlinedAt: !183) +!183 = !DILocation(line: 228, column: 66, scope: !101) +!184 = !DILocation(line: 229, column: 31, scope: !5) +!185 = !DILocation(line: 229, column: 41, scope: !5) +!186 = !DILocation(line: 230, column: 48, scope: !5) +!187 = !DILocation(line: 263, column: 15, scope: !180, inlinedAt: !188) +!188 = !DILocation(line: 293, column: 36, scope: !180, inlinedAt: !189) +!189 = !DILocation(line: 230, column: 56, scope: !101) +!190 = !DILocation(line: 230, column: 24, scope: !5) +!191 = !DILocation(line: 231, column: 54, scope: !5) +!192 = !DILocation(line: 232, column: 34, scope: !5) +!193 = !DILocation(line: 232, column: 22, scope: !5) +!194 = !DILocation(line: 233, column: 61, scope: !5) +!195 = !DILocation(line: 233, column: 40, scope: !5) +!196 = !DILocation(line: 263, column: 15, scope: !180, inlinedAt: !197) +!197 = !DILocation(line: 293, column: 36, scope: !180, inlinedAt: !198) +!198 = !DILocation(line: 233, column: 66, scope: !101) +!199 = !DILocation(line: 234, column: 31, scope: !5) +!200 = !DILocation(line: 234, column: 41, scope: !5) +!201 = !DILocation(line: 235, column: 48, scope: !5) +!202 = !DILocation(line: 263, column: 15, scope: !180, inlinedAt: !203) +!203 = !DILocation(line: 293, column: 36, scope: !180, inlinedAt: !204) +!204 = !DILocation(line: 235, column: 56, scope: !101) +!205 = !DILocation(line: 235, column: 24, scope: !5) +!206 = !DILocation(line: 236, column: 54, scope: !5) +!207 = !DILocation(line: 249, column: 14, scope: !5) +!208 = !DILocation(line: 237, column: 34, scope: !5) +!209 = !DILocation(line: 237, column: 22, scope: !5) +!210 = !DILocation(line: 238, column: 61, scope: !5) +!211 = !DILocation(line: 238, column: 40, scope: !5) +!212 = !DILocation(line: 263, column: 15, scope: !180, inlinedAt: !213) +!213 = !DILocation(line: 293, column: 36, scope: !180, inlinedAt: !214) +!214 = !DILocation(line: 238, column: 66, scope: !101) +!215 = !DILocation(line: 239, column: 31, scope: !5) +!216 = !DILocation(line: 239, column: 41, scope: !5) +!217 = !DILocation(line: 240, column: 48, scope: !5) +!218 = !DILocation(line: 263, column: 15, scope: !180, inlinedAt: !219) +!219 = !DILocation(line: 293, column: 36, scope: !180, inlinedAt: !220) +!220 = !DILocation(line: 240, column: 56, scope: !101) +!221 = !DILocation(line: 240, column: 24, scope: !5) +!222 = !DILocation(line: 241, column: 54, scope: !5) +!223 = !DILocation(line: 250, column: 14, scope: !5) +!224 = !DILocation(line: 242, column: 34, scope: !5) +!225 = !DILocation(line: 242, column: 22, scope: !5) +!226 = !DILocation(line: 243, column: 61, scope: !5) +!227 = !DILocation(line: 243, column: 40, scope: !5) +!228 = !DILocation(line: 263, column: 15, scope: !180, inlinedAt: !229) +!229 = !DILocation(line: 293, column: 36, scope: !180, inlinedAt: !230) +!230 = !DILocation(line: 243, column: 66, scope: !101) +!231 = !DILocation(line: 244, column: 31, scope: !5) +!232 = !DILocation(line: 244, column: 41, scope: !5) +!233 = !DILocation(line: 245, column: 48, scope: !5) +!234 = !DILocation(line: 263, column: 15, scope: !180, inlinedAt: !235) +!235 = !DILocation(line: 293, column: 36, scope: !180, inlinedAt: !236) +!236 = !DILocation(line: 245, column: 56, scope: !101) +!237 = !DILocation(line: 245, column: 24, scope: !5) +!238 = !DILocation(line: 246, column: 54, scope: !5) +!239 = !DILocation(line: 251, column: 14, scope: !5) +!240 = !DILocation(line: 248, column: 14, scope: !5) +!241 = !DILocation(line: 258, column: 23, scope: !5) +!242 = !DILocation(line: 259, column: 8, scope: !5) +!243 = !DILocation(line: 257, column: 14, scope: !5) +!244 = !DILocation(line: 263, column: 23, scope: !5) +!245 = !DILocation(line: 264, column: 8, scope: !5) +!246 = !DILocation(line: 262, column: 14, scope: !5) +!247 = !DILocation(line: 268, column: 23, scope: !5) +!248 = !DILocation(line: 269, column: 8, scope: !5) +!249 = !DILocation(line: 275, column: 22, scope: !5) +!250 = !DILocation(line: 276, column: 22, scope: !5) +!251 = !DILocation(line: 275, column: 8, scope: !5) +!252 = !DILocation(line: 273, column: 14, scope: !5) +!253 = !DILocation(line: 281, column: 22, scope: !5) +!254 = !DILocation(line: 282, column: 22, scope: !5) +!255 = !DILocation(line: 281, column: 8, scope: !5) +!256 = !DILocation(line: 287, column: 22, scope: !5) +!257 = !DILocation(line: 288, column: 22, scope: !5) +!258 = !DILocation(line: 289, column: 22, scope: !5) +!259 = !DILocation(line: 287, column: 8, scope: !5) +!260 = !DILocation(line: 297, column: 16, scope: !5) +!261 = !DILocation(line: 297, column: 51, scope: !5) +!262 = !DILocation(line: 298, column: 16, scope: !5) +!263 = !DILocation(line: 298, column: 51, scope: !5) +!264 = !DILocation(line: 299, column: 57, scope: !5) +!265 = !DILocation(line: 299, column: 51, scope: !5) +!266 = !DILocation(line: 300, column: 16, scope: !5) +!267 = !DILocation(line: 300, column: 51, scope: !5) +!268 = !DILocation(line: 301, column: 51, scope: !5) +!269 = !DILocation(line: 302, column: 59, scope: !5) +!270 = !DILocation(line: 302, column: 51, scope: !5) +!271 = !DILocation(line: 303, column: 16, scope: !5) +!272 = !DILocation(line: 303, column: 51, scope: !5) +!273 = !DILocation(line: 304, column: 51, scope: !5) +!274 = !DILocation(line: 305, column: 51, scope: !5) +!275 = !DILocation(line: 306, column: 51, scope: !5) +!276 = !DILocation(line: 319, column: 30, scope: !5) +!277 = !DILocation(line: 319, column: 20, scope: !5) +!278 = !DILocation(line: 320, column: 30, scope: !5) +!279 = !DILocation(line: 320, column: 20, scope: !5) +!280 = !DILocation(line: 321, column: 30, scope: !5) +!281 = !DILocation(line: 321, column: 20, scope: !5) +!282 = !DILocation(line: 322, column: 30, scope: !5) +!283 = !DILocation(line: 322, column: 20, scope: !5) +!284 = !DILocation(line: 323, column: 30, scope: !5) +!285 = !DILocation(line: 323, column: 20, scope: !5) +!286 = !DILocation(line: 324, column: 30, scope: !5) +!287 = !DILocation(line: 324, column: 20, scope: !5) +!288 = !DILocation(line: 285, column: 14, scope: !5) +!289 = !DILocation(line: 325, column: 30, scope: !5) +!290 = !DILocation(line: 325, column: 20, scope: !5) +!291 = !DILocation(line: 279, column: 14, scope: !5) +!292 = !DILocation(line: 326, column: 30, scope: !5) +!293 = !DILocation(line: 326, column: 20, scope: !5) +!294 = !DILocation(line: 267, column: 14, scope: !5) +!295 = !DILocation(line: 327, column: 30, scope: !5) +!296 = !DILocation(line: 327, column: 20, scope: !5) +!297 = !DILocation(line: 328, column: 30, scope: !5) +!298 = !DILocation(line: 328, column: 20, scope: !5) +!299 = !DILocation(line: 328, column: 4, scope: !5) diff --git a/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.ptx b/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..ff857e2103bdb5159238dc291915de76c8f2a251 --- /dev/null +++ b/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.ptx @@ -0,0 +1,4016 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl chunk_gated_delta_rule_fwd_kkt_solve_kernel // -- Begin function chunk_gated_delta_rule_fwd_kkt_solve_kernel +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; + // @chunk_gated_delta_rule_fwd_kkt_solve_kernel +.visible .entry chunk_gated_delta_rule_fwd_kkt_solve_kernel( + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_fwd_kkt_solve_kernel_param_0, + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_fwd_kkt_solve_kernel_param_1, + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_fwd_kkt_solve_kernel_param_2, + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_fwd_kkt_solve_kernel_param_3, + .param .u32 chunk_gated_delta_rule_fwd_kkt_solve_kernel_param_4, + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_fwd_kkt_solve_kernel_param_5, + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_fwd_kkt_solve_kernel_param_6 +) +.reqntid 128 +{ + .reg .pred %p<92>; + .reg .b16 %rs<32>; + .reg .b32 %r<1921>; + .reg .b64 %rd<315>; + .loc 1 40 0 // chunk_fwd.py:40:0 +$L__func_begin0: + .loc 1 40 0 // chunk_fwd.py:40:0 + +// %bb.0: + ld.param.b32 %r67, [chunk_gated_delta_rule_fwd_kkt_solve_kernel_param_4]; +$L__tmp0: + .loc 1 70 30 // chunk_fwd.py:70:30 + mov.u32 %r68, %ctaid.x; + .loc 1 78 25 // chunk_fwd.py:78:25 + cvt.s64.s32 %rd2, %r67; + .loc 1 80 13 // chunk_fwd.py:80:13 + mul.wide.u32 %rd3, %r68, 64; + .loc 1 80 19 // chunk_fwd.py:80:19 + setp.gt.s64 %p2, %rd2, %rd3; + @%p2 bra $L__BB0_2; + bra.uni $L__BB0_1; +$L__BB0_2: + .loc 1 0 19 // chunk_fwd.py:0:19 + ld.param.b64 %rd57, [chunk_gated_delta_rule_fwd_kkt_solve_kernel_param_3]; + ld.param.b64 %rd56, [chunk_gated_delta_rule_fwd_kkt_solve_kernel_param_2]; + ld.param.b64 %rd55, [chunk_gated_delta_rule_fwd_kkt_solve_kernel_param_1]; + ld.param.b64 %rd54, [chunk_gated_delta_rule_fwd_kkt_solve_kernel_param_0]; + .loc 1 70 0 // chunk_fwd.py:70 + cvt.u64.u32 %rd1, %r68; + .loc 1 70 61 // chunk_fwd.py:70:61 + mov.u32 %r99, %ctaid.y; + .loc 1 70 67 // chunk_fwd.py:70:67 + cvt.u64.u32 %rd94, %r99; + .loc 1 78 25 // chunk_fwd.py:78:25 + and.b64 %rd95, %rd94, 65504; + .loc 1 71 34 // chunk_fwd.py:71:34 + and.b64 %rd96, %rd94, 31; + .loc 1 84 23 // chunk_fwd.py:84:23 + or.b64 %rd4, %rd3, 16; + .loc 1 85 23 // chunk_fwd.py:85:23 + or.b64 %rd5, %rd3, 32; + .loc 1 86 23 // chunk_fwd.py:86:23 + or.b64 %rd6, %rd3, 48; + .loc 1 88 16 // chunk_fwd.py:88:16 + mul.lo.s64 %rd97, %rd95, %rd2; + .loc 1 88 20 // chunk_fwd.py:88:20 + or.b64 %rd7, %rd97, %rd96; + .loc 1 88 9 // chunk_fwd.py:88:9 + shl.b64 %rd98, %rd7, 8; + add.s64 %rd99, %rd54, %rd98; + .loc 1 91 23 // chunk_fwd.py:91:23 + mov.u32 %r1, %tid.x; + and.b32 %r100, %r1, 120; + bfe.u32 %r101, %r1, 3, 4; + and.b32 %r2, %r1, 28; + and.b32 %r3, %r1, 7; + and.b32 %r4, %r1, 3; + and.b32 %r5, %r1, 32; + shr.u32 %r102, %r5, 2; + .loc 1 92 21 // chunk_fwd.py:92:21 + cvt.u64.u32 %rd100, %r101; + or.b64 %rd8, %rd3, %rd100; + .loc 1 92 28 // chunk_fwd.py:92:28 + setp.lt.u64 %p29, %rd8, %rd2; + .loc 1 93 21 // chunk_fwd.py:93:21 + or.b64 %rd9, %rd4, %rd100; + .loc 1 94 21 // chunk_fwd.py:94:21 + or.b64 %rd10, %rd5, %rd100; + .loc 1 95 21 // chunk_fwd.py:95:21 + or.b64 %rd11, %rd6, %rd100; + .loc 1 91 23 // chunk_fwd.py:91:23 + shl.b32 %r6, %r4, 1; + or.b32 %r7, %r6, %r102; + or.b32 %r8, %r7, 1; + or.b32 %r9, %r6, 1; + or.b32 %r103, %r6, 8; + or.b32 %r10, %r6, 9; + .loc 1 92 21 // chunk_fwd.py:92:21 + mov.b64 %rd12, {%r6, %r9}; + .loc 1 94 21 // chunk_fwd.py:94:21 + cvt.u64.u32 %rd16, %r10; + cvt.u64.u32 %rd15, %r103; + cvt.u64.u32 %rd14, %r9; + cvt.u64.u32 %rd13, %r6; + .loc 1 92 21 // chunk_fwd.py:92:21 + or.b64 %rd17, %rd3, %rd13; + or.b64 %rd18, %rd3, %rd14; + or.b64 %rd19, %rd3, %rd15; + or.b64 %rd20, %rd3, %rd16; + .loc 1 92 28 // chunk_fwd.py:92:28 + setp.lt.u64 %p13, %rd17, %rd2; + setp.lt.u64 %p14, %rd18, %rd2; + setp.lt.u64 %p15, %rd19, %rd2; + setp.lt.u64 %p16, %rd20, %rd2; + .loc 1 92 21 // chunk_fwd.py:92:21 + mov.b64 %rd21, {%r7, %r8}; + cvt.u64.u32 %rd23, %r8; + cvt.u64.u32 %rd22, %r7; + or.b64 %rd101, %rd3, %rd22; + or.b64 %rd102, %rd3, %rd23; + .loc 1 92 28 // chunk_fwd.py:92:28 + setp.lt.u64 %p12, %rd102, %rd2; + setp.lt.u64 %p11, %rd101, %rd2; + .loc 1 91 23 // chunk_fwd.py:91:23 + bfe.u32 %r11, %r1, 2, 3; + or.b32 %r12, %r11, 8; + .loc 1 92 21 // chunk_fwd.py:92:21 + mov.b64 %rd24, {%r11, %r12}; + cvt.u64.u32 %rd25, %r12; + cvt.u64.u32 %rd27, %r11; + or.b64 %rd103, %rd3, %rd27; + or.b64 %rd104, %rd3, %rd25; + .loc 1 92 28 // chunk_fwd.py:92:28 + setp.lt.u64 %p10, %rd104, %rd2; + setp.lt.u64 %p9, %rd103, %rd2; + .loc 1 93 21 // chunk_fwd.py:93:21 + or.b64 %rd26, %rd4, %rd25; + or.b64 %rd28, %rd4, %rd27; + or.b64 %rd29, %rd4, %rd13; + or.b64 %rd30, %rd4, %rd14; + or.b64 %rd31, %rd4, %rd15; + or.b64 %rd32, %rd4, %rd16; + .loc 1 93 28 // chunk_fwd.py:93:28 + setp.lt.u64 %p4, %rd26, %rd2; + setp.lt.u64 %p3, %rd28, %rd2; + setp.lt.u64 %p20, %rd32, %rd2; + setp.lt.u64 %p19, %rd31, %rd2; + setp.lt.u64 %p18, %rd30, %rd2; + setp.lt.u64 %p17, %rd29, %rd2; + .loc 1 94 21 // chunk_fwd.py:94:21 + or.b64 %rd33, %rd5, %rd25; + or.b64 %rd34, %rd5, %rd27; + or.b64 %rd35, %rd5, %rd13; + or.b64 %rd36, %rd5, %rd14; + or.b64 %rd37, %rd5, %rd15; + or.b64 %rd38, %rd5, %rd16; + .loc 1 94 28 // chunk_fwd.py:94:28 + setp.lt.u64 %p6, %rd33, %rd2; + setp.lt.u64 %p5, %rd34, %rd2; + setp.lt.u64 %p24, %rd38, %rd2; + setp.lt.u64 %p23, %rd37, %rd2; + setp.lt.u64 %p22, %rd36, %rd2; + setp.lt.u64 %p21, %rd35, %rd2; + .loc 1 95 21 // chunk_fwd.py:95:21 + or.b64 %rd39, %rd6, %rd25; + or.b64 %rd40, %rd6, %rd27; + or.b64 %rd41, %rd6, %rd13; + or.b64 %rd42, %rd6, %rd14; + or.b64 %rd43, %rd6, %rd15; + or.b64 %rd44, %rd6, %rd16; + .loc 1 95 28 // chunk_fwd.py:95:28 + setp.lt.u64 %p8, %rd39, %rd2; + setp.lt.u64 %p7, %rd40, %rd2; + setp.lt.u64 %p28, %rd44, %rd2; + setp.lt.u64 %p27, %rd43, %rd2; + setp.lt.u64 %p26, %rd42, %rd2; + setp.lt.u64 %p25, %rd41, %rd2; + .loc 1 98 0 // chunk_fwd.py:98 + shl.b64 %rd105, %rd7, 1; + add.s64 %rd106, %rd56, %rd105; + .loc 1 98 35 // chunk_fwd.py:98:35 + shl.b64 %rd107, %rd103, 6; + add.s64 %rd58, %rd106, %rd107; + shl.b64 %rd108, %rd104, 6; + add.s64 %rd59, %rd106, %rd108; + .loc 1 99 35 // chunk_fwd.py:99:35 + shl.b64 %rd109, %rd28, 6; + add.s64 %rd60, %rd106, %rd109; + shl.b64 %rd110, %rd26, 6; + add.s64 %rd61, %rd106, %rd110; + .loc 1 100 35 // chunk_fwd.py:100:35 + shl.b64 %rd111, %rd34, 6; + add.s64 %rd62, %rd106, %rd111; + shl.b64 %rd112, %rd33, 6; + add.s64 %rd63, %rd106, %rd112; + .loc 1 101 35 // chunk_fwd.py:101:35 + shl.b64 %rd113, %rd40, 6; + add.s64 %rd64, %rd106, %rd113; + shl.b64 %rd114, %rd39, 6; + add.s64 %rd65, %rd106, %rd114; + mov.b16 %rs2, 0; + .loc 1 102 19 // chunk_fwd.py:102:19 + // begin inline asm + mov.u16 %rs1, %rs2; + @%p9 ld.global.b16 { %rs1 }, [ %rd58 + 0 ]; + // end inline asm + // begin inline asm + mov.u16 %rs3, %rs2; + @%p10 ld.global.b16 { %rs3 }, [ %rd59 + 0 ]; + // end inline asm + .loc 1 103 19 // chunk_fwd.py:103:19 + // begin inline asm + mov.u16 %rs4, %rs2; + @%p3 ld.global.b16 { %rs4 }, [ %rd60 + 0 ]; + // end inline asm + // begin inline asm + mov.u16 %rs5, %rs2; + @%p4 ld.global.b16 { %rs5 }, [ %rd61 + 0 ]; + // end inline asm + .loc 1 104 19 // chunk_fwd.py:104:19 + // begin inline asm + mov.u16 %rs6, %rs2; + @%p5 ld.global.b16 { %rs6 }, [ %rd62 + 0 ]; + // end inline asm + // begin inline asm + mov.u16 %rs7, %rs2; + @%p6 ld.global.b16 { %rs7 }, [ %rd63 + 0 ]; + // end inline asm + .loc 1 105 19 // chunk_fwd.py:105:19 + // begin inline asm + mov.u16 %rs8, %rs2; + @%p7 ld.global.b16 { %rs8 }, [ %rd64 + 0 ]; + // end inline asm + // begin inline asm + mov.u16 %rs9, %rs2; + @%p8 ld.global.b16 { %rs9 }, [ %rd65 + 0 ]; + // end inline asm + .loc 1 109 0 // chunk_fwd.py:109 + shl.b64 %rd115, %rd7, 2; + add.s64 %rd116, %rd55, %rd115; + .loc 1 109 36 // chunk_fwd.py:109:36 + shl.b64 %rd117, %rd103, 7; + add.s64 %rd66, %rd116, %rd117; + shl.b64 %rd118, %rd104, 7; + add.s64 %rd67, %rd116, %rd118; + shl.b64 %rd119, %rd101, 7; + add.s64 %rd68, %rd116, %rd119; + shl.b64 %rd120, %rd102, 7; + add.s64 %rd69, %rd116, %rd120; + shl.b64 %rd121, %rd17, 7; + add.s64 %rd70, %rd116, %rd121; + shl.b64 %rd122, %rd18, 7; + add.s64 %rd71, %rd116, %rd122; + shl.b64 %rd123, %rd19, 7; + add.s64 %rd72, %rd116, %rd123; + shl.b64 %rd124, %rd20, 7; + add.s64 %rd73, %rd116, %rd124; + .loc 1 110 36 // chunk_fwd.py:110:36 + shl.b64 %rd125, %rd28, 7; + add.s64 %rd74, %rd116, %rd125; + shl.b64 %rd126, %rd26, 7; + add.s64 %rd75, %rd116, %rd126; + shl.b64 %rd127, %rd29, 7; + add.s64 %rd76, %rd116, %rd127; + shl.b64 %rd128, %rd30, 7; + add.s64 %rd77, %rd116, %rd128; + shl.b64 %rd129, %rd31, 7; + add.s64 %rd78, %rd116, %rd129; + shl.b64 %rd130, %rd32, 7; + add.s64 %rd79, %rd116, %rd130; + .loc 1 111 36 // chunk_fwd.py:111:36 + shl.b64 %rd131, %rd34, 7; + add.s64 %rd80, %rd116, %rd131; + shl.b64 %rd132, %rd33, 7; + add.s64 %rd81, %rd116, %rd132; + shl.b64 %rd133, %rd35, 7; + add.s64 %rd82, %rd116, %rd133; + shl.b64 %rd134, %rd36, 7; + add.s64 %rd83, %rd116, %rd134; + shl.b64 %rd135, %rd37, 7; + add.s64 %rd84, %rd116, %rd135; + shl.b64 %rd136, %rd38, 7; + add.s64 %rd85, %rd116, %rd136; + .loc 1 112 36 // chunk_fwd.py:112:36 + shl.b64 %rd137, %rd40, 7; + add.s64 %rd86, %rd116, %rd137; + shl.b64 %rd138, %rd39, 7; + add.s64 %rd87, %rd116, %rd138; + shl.b64 %rd139, %rd41, 7; + add.s64 %rd88, %rd116, %rd139; + shl.b64 %rd140, %rd42, 7; + add.s64 %rd89, %rd116, %rd140; + shl.b64 %rd141, %rd43, 7; + add.s64 %rd90, %rd116, %rd141; + shl.b64 %rd142, %rd44, 7; + add.s64 %rd91, %rd116, %rd142; + mov.b32 %r70, 0; + .loc 1 114 23 // chunk_fwd.py:114:23 + // begin inline asm + mov.u32 %r69, %r70; + @%p9 ld.global.b32 { %r69 }, [ %rd66 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r71, %r70; + @%p10 ld.global.b32 { %r71 }, [ %rd67 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r72, %r70; + @%p11 ld.global.b32 { %r72 }, [ %rd68 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r73, %r70; + @%p12 ld.global.b32 { %r73 }, [ %rd69 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r74, %r70; + @%p13 ld.global.b32 { %r74 }, [ %rd70 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r75, %r70; + @%p14 ld.global.b32 { %r75 }, [ %rd71 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r76, %r70; + @%p15 ld.global.b32 { %r76 }, [ %rd72 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r77, %r70; + @%p16 ld.global.b32 { %r77 }, [ %rd73 + 0 ]; + // end inline asm + .loc 1 115 23 // chunk_fwd.py:115:23 + // begin inline asm + mov.u32 %r78, %r70; + @%p3 ld.global.b32 { %r78 }, [ %rd74 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r79, %r70; + @%p4 ld.global.b32 { %r79 }, [ %rd75 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r80, %r70; + @%p17 ld.global.b32 { %r80 }, [ %rd76 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r81, %r70; + @%p18 ld.global.b32 { %r81 }, [ %rd77 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r82, %r70; + @%p19 ld.global.b32 { %r82 }, [ %rd78 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r83, %r70; + @%p20 ld.global.b32 { %r83 }, [ %rd79 + 0 ]; + // end inline asm + .loc 1 116 23 // chunk_fwd.py:116:23 + // begin inline asm + mov.u32 %r84, %r70; + @%p5 ld.global.b32 { %r84 }, [ %rd80 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r85, %r70; + @%p6 ld.global.b32 { %r85 }, [ %rd81 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r86, %r70; + @%p21 ld.global.b32 { %r86 }, [ %rd82 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r87, %r70; + @%p22 ld.global.b32 { %r87 }, [ %rd83 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r88, %r70; + @%p23 ld.global.b32 { %r88 }, [ %rd84 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r89, %r70; + @%p24 ld.global.b32 { %r89 }, [ %rd85 + 0 ]; + // end inline asm + .loc 1 117 23 // chunk_fwd.py:117:23 + // begin inline asm + mov.u32 %r90, %r70; + @%p7 ld.global.b32 { %r90 }, [ %rd86 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r91, %r70; + @%p8 ld.global.b32 { %r91 }, [ %rd87 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r92, %r70; + @%p25 ld.global.b32 { %r92 }, [ %rd88 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r93, %r70; + @%p26 ld.global.b32 { %r93 }, [ %rd89 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r94, %r70; + @%p27 ld.global.b32 { %r94 }, [ %rd90 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r95, %r70; + @%p28 ld.global.b32 { %r95 }, [ %rd91 + 0 ]; + // end inline asm + .loc 1 138 38 // chunk_fwd.py:138:38 + shl.b32 %r104, %r3, 3; + .loc 1 139 19 // chunk_fwd.py:139:19 + shl.b64 %rd143, %rd8, 13; + add.s64 %rd144, %rd99, %rd143; + .loc 1 139 52 // chunk_fwd.py:139:52 + mad.wide.u32 %rd92, %r104, 2, %rd144; + .loc 1 140 23 // chunk_fwd.py:140:23 + shl.b32 %r13, %r1, 4; + and.b32 %r105, %r13, 2032; + and.b32 %r106, %r1, 56; + shl.b32 %r107, %r106, 1; + xor.b32 %r108, %r105, %r107; + mov.b32 %r109, global_smem; + add.s32 %r96, %r109, %r108; + selp.b32 %r97, 16, 0, %p29; + // begin inline asm + cp.async.cg.shared.global [ %r96 + 0 ], [ %rd92 + 0 ], 0x10, %r97; + // end inline asm + cp.async.commit_group; + .loc 1 139 52 // chunk_fwd.py:139:52 + add.s64 %rd93, %rd92, 128; + .loc 1 140 23 // chunk_fwd.py:140:23 + bar.sync 0; + add.s32 %r98, %r96, 2048; + // begin inline asm + cp.async.cg.shared.global [ %r98 + 0 ], [ %rd93 + 0 ], 0x10, %r97; + // end inline asm + cp.async.commit_group; + shl.b32 %r110, %r3, 7; + shl.b32 %r111, %r3, 4; + shl.b32 %r112, %r1, 1; + and.b32 %r113, %r112, 48; + or.b32 %r114, %r110, %r111; + xor.b32 %r14, %r114, %r113; + xor.b32 %r15, %r14, 64; + shl.b32 %r115, %r5, 5; + or.b32 %r116, %r110, %r115; + xor.b32 %r117, %r111, %r113; + or.b32 %r16, %r116, %r117; + xor.b32 %r17, %r16, 64; + shl.b32 %r118, %r1, 7; + and.b32 %r119, %r118, 1920; + and.b32 %r18, %r1, 16; + xor.b32 %r120, %r111, %r18; + or.b32 %r19, %r120, %r119; + xor.b32 %r121, %r19, 32; + xor.b32 %r122, %r19, 64; + xor.b32 %r123, %r19, 96; + add.s32 %r223, %r96, 4096; + shl.b32 %r124, %r106, 5; + shl.b32 %r125, %r100, 1; + or.b32 %r126, %r124, %r111; + xor.b32 %r127, %r126, %r125; + add.s32 %r128, %r109, %r127; + add.s32 %r228, %r128, 6144; + shl.b32 %r129, %r3, 8; + or.b32 %r130, %r117, %r129; + add.s32 %r131, %r109, %r130; + add.s32 %r20, %r131, 6144; + add.s32 %r21, %r131, 6272; + xor.b32 %r132, %r130, 64; + add.s32 %r133, %r109, %r132; + add.s32 %r22, %r133, 6144; + add.s32 %r23, %r133, 6272; + add.s32 %r134, %r109, %r19; + add.s32 %r24, %r134, 4096; + add.s32 %r25, %r109, %r121; + add.s32 %r26, %r25, 4096; + add.s32 %r27, %r109, %r122; + add.s32 %r28, %r27, 4096; + add.s32 %r29, %r109, %r123; + add.s32 %r30, %r29, 4096; + .loc 1 137 21 // chunk_fwd.py:137:21 + shl.b32 %r31, %r1, 3; + shl.b64 %rd145, %rd1, 19; + shr.u64 %rd146, %rd94, 5; + mad.lo.s64 %rd147, %rd146, %rd2, %rd100; + shl.b64 %rd148, %rd147, 13; + add.s64 %rd149, %rd145, %rd148; + shl.b64 %rd150, %rd96, 8; + or.b64 %rd151, %rd149, %rd150; + mul.wide.u32 %rd152, %r3, 16; + or.b64 %rd153, %rd151, %rd152; + add.s64 %rd154, %rd153, %rd54; + add.s64 %rd295, %rd154, 393216; + mov.b32 %r1886, 0f00000000; + mov.b64 %rd302, {%r1886, %r1886}; + mov.b32 %r1829, 1; + mov.b32 %r1828, -1; + mov.pred %p91, -1; + mov.b64 %rd296, 0; + mov.b32 %r1887, %r1886; + mov.b32 %r1888, %r1886; + mov.b32 %r1889, %r1886; + mov.b32 %r1890, %r1886; + mov.b32 %r1891, %r1886; + mov.b32 %r1892, %r1886; + mov.b32 %r1893, %r1886; + mov.b32 %r1878, %r1886; + mov.b32 %r1879, %r1886; + mov.b32 %r1880, %r1886; + mov.b32 %r1881, %r1886; + mov.b32 %r1882, %r1886; + mov.b32 %r1883, %r1886; + mov.b32 %r1884, %r1886; + mov.b32 %r1885, %r1886; + mov.b32 %r1870, %r1886; + mov.b32 %r1871, %r1886; + mov.b32 %r1872, %r1886; + mov.b32 %r1873, %r1886; + mov.b32 %r1874, %r1886; + mov.b32 %r1875, %r1886; + mov.b32 %r1876, %r1886; + mov.b32 %r1877, %r1886; + mov.b32 %r1862, %r1886; + mov.b32 %r1863, %r1886; + mov.b32 %r1864, %r1886; + mov.b32 %r1865, %r1886; + mov.b32 %r1866, %r1886; + mov.b32 %r1867, %r1886; + mov.b32 %r1868, %r1886; + mov.b32 %r1869, %r1886; + mov.b32 %r1854, %r1886; + mov.b32 %r1855, %r1886; + mov.b32 %r1856, %r1886; + mov.b32 %r1857, %r1886; + mov.b32 %r1858, %r1886; + mov.b32 %r1859, %r1886; + mov.b32 %r1860, %r1886; + mov.b32 %r1861, %r1886; + mov.b32 %r1846, %r1886; + mov.b32 %r1847, %r1886; + mov.b32 %r1848, %r1886; + mov.b32 %r1849, %r1886; + mov.b32 %r1850, %r1886; + mov.b32 %r1851, %r1886; + mov.b32 %r1852, %r1886; + mov.b32 %r1853, %r1886; + mov.b32 %r1842, %r1886; + mov.b32 %r1843, %r1886; + mov.b32 %r1844, %r1886; + mov.b32 %r1845, %r1886; + mov.b32 %r1838, %r1886; + mov.b32 %r1839, %r1886; + mov.b32 %r1840, %r1886; + mov.b32 %r1841, %r1886; + mov.b32 %r1834, %r1886; + mov.b32 %r1835, %r1886; + mov.b32 %r1836, %r1886; + mov.b32 %r1837, %r1886; + mov.b32 %r1830, %r1886; + mov.b32 %r1831, %r1886; + mov.b32 %r1832, %r1886; + mov.b32 %r1833, %r1886; + mov.b64 %rd297, %rd302; + mov.b64 %rd298, %rd302; + mov.b64 %rd299, %rd302; + mov.b64 %rd300, %rd302; + mov.b64 %rd301, %rd302; + bra.uni $L__BB0_3; +$L__BB0_7: // in Loop: Header=BB0_3 Depth=1 + .loc 1 137 21 // chunk_fwd.py:137:21 + add.s64 %rd295, %rd295, 128; + mov.pred %p91, 0; + mov.b64 %rd296, 128; + @%p1 bra $L__BB0_3; + bra.uni $L__BB0_8; +$L__BB0_3: // =>This Inner Loop Header: Depth=1 + .loc 1 0 21 // chunk_fwd.py:0:21 + mov.pred %p1, %p91; + .loc 1 144 19 // chunk_fwd.py:144:19 + setp.le.u64 %p30, %rd2, %rd4; + .loc 1 137 21 // chunk_fwd.py:137:21 + add.s32 %r160, %r1828, 1; + setp.gt.s32 %p31, %r160, 1; + selp.b32 %r1828, 0, %r160, %p31; + .loc 1 140 23 // chunk_fwd.py:140:23 + cp.async.wait_group 1; + bar.sync 0; + shl.b32 %r161, %r1828, 11; + add.s32 %r162, %r109, %r161; + .loc 1 142 45 // chunk_fwd.py:142:45 + add.s32 %r163, %r162, %r14; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r265, %r266, %r269, %r270}, [%r163]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r267, %r268, %r271, %r272}, [%r163+1024]; + add.s32 %r164, %r162, %r15; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r273, %r274, %r277, %r278}, [%r164]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r275, %r276, %r279, %r280}, [%r164+1024]; + add.s32 %r165, %r162, %r16; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r139, %r140, %r145, %r146}, [%r165]; + add.s32 %r166, %r162, %r17; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r151, %r152, %r157, %r158}, [%r166]; + .loc 1 140 23 // chunk_fwd.py:140:23 + add.s32 %r167, %r162, %r19; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r135, %r136, %r137, %r138}, [%r167]; + add.s32 %r168, %r25, %r161; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r141, %r142, %r143, %r144}, [%r168]; + add.s32 %r169, %r27, %r161; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r147, %r148, %r149, %r150}, [%r169]; + add.s32 %r170, %r29, %r161; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r153, %r154, %r155, %r156}, [%r170]; + .loc 1 142 45 // chunk_fwd.py:142:45 + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1830, %r1831, %r1832, %r1833 }, { %r135, %r136, %r137, %r138 }, { %r139, %r140 }, { %r1830, %r1831, %r1832, %r1833 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1830, %r1831, %r1832, %r1833 }, { %r141, %r142, %r143, %r144 }, { %r145, %r146 }, { %r1830, %r1831, %r1832, %r1833 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1830, %r1831, %r1832, %r1833 }, { %r147, %r148, %r149, %r150 }, { %r151, %r152 }, { %r1830, %r1831, %r1832, %r1833 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1830, %r1831, %r1832, %r1833 }, { %r153, %r154, %r155, %r156 }, { %r157, %r158 }, { %r1830, %r1831, %r1832, %r1833 }; + // end inline asm + .loc 1 137 21 // chunk_fwd.py:137:21 + add.s32 %r171, %r1829, 1; + setp.gt.s32 %p32, %r171, 1; + selp.b32 %r1829, 0, %r171, %p32; + .loc 1 139 52 // chunk_fwd.py:139:52 + add.s64 %rd156, %rd92, %rd296; + add.s64 %rd155, %rd156, 256; + .loc 1 140 23 // chunk_fwd.py:140:23 + shl.b32 %r172, %r1829, 11; + bar.sync 0; + add.s32 %r159, %r96, %r172; + // begin inline asm + cp.async.cg.shared.global [ %r159 + 0 ], [ %rd155 + 0 ], 0x10, %r70; + // end inline asm + cp.async.commit_group; + .loc 1 144 11 // chunk_fwd.py:144:11 + @%p30 bra $L__BB0_7; +// %bb.4: // in Loop: Header=BB0_3 Depth=1 + .loc 1 0 11 // chunk_fwd.py:0:11 + setp.le.u64 %p34, %rd2, %rd5; + .loc 1 93 28 // chunk_fwd.py:93:28 + setp.lt.u64 %p33, %rd9, %rd2; + .loc 1 146 27 // chunk_fwd.py:146:27 + add.s64 %rd157, %rd295, -262144; + mov.b32 %r197, 0; + // begin inline asm + mov.u32 %r173, %r197; + mov.u32 %r174, %r197; + mov.u32 %r175, %r197; + mov.u32 %r176, %r197; + @%p33 ld.global.v4.b32 { %r173, %r174, %r175, %r176 }, [ %rd157 + 0 ]; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r223 + 0 ], { %r173, %r174, %r175, %r176 }; + // end inline asm + .loc 1 148 42 // chunk_fwd.py:148:42 + // begin inline asm + st.shared.v4.b32 [ %r228 + 0 ], { %r173, %r174, %r175, %r176 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r281, %r282, %r285, %r286}, [%r20]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r283, %r284, %r287, %r288}, [%r21]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r289, %r290, %r293, %r294}, [%r22]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r291, %r292, %r295, %r296}, [%r23]; + .loc 1 146 27 // chunk_fwd.py:146:27 + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r177, %r178, %r179, %r180}, [%r24]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r183, %r184, %r185, %r186}, [%r26]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r189, %r190, %r191, %r192}, [%r28]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r193, %r194, %r195, %r196}, [%r30]; + .loc 1 148 49 // chunk_fwd.py:148:49 + mov.b64 {%r181, %r182}, %rd301; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r181, %r182, %r1834, %r1835 }, { %r177, %r178, %r179, %r180 }, { %r281, %r282 }, { %r181, %r182, %r1834, %r1835 }; + // end inline asm + mov.b64 {%r187, %r188}, %rd302; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1836, %r1837, %r187, %r188 }, { %r177, %r178, %r179, %r180 }, { %r283, %r284 }, { %r1836, %r1837, %r187, %r188 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r181, %r182, %r1834, %r1835 }, { %r183, %r184, %r185, %r186 }, { %r285, %r286 }, { %r181, %r182, %r1834, %r1835 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1836, %r1837, %r187, %r188 }, { %r183, %r184, %r185, %r186 }, { %r287, %r288 }, { %r1836, %r1837, %r187, %r188 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r181, %r182, %r1834, %r1835 }, { %r189, %r190, %r191, %r192 }, { %r289, %r290 }, { %r181, %r182, %r1834, %r1835 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1836, %r1837, %r187, %r188 }, { %r189, %r190, %r191, %r192 }, { %r291, %r292 }, { %r1836, %r1837, %r187, %r188 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r181, %r182, %r1834, %r1835 }, { %r193, %r194, %r195, %r196 }, { %r293, %r294 }, { %r181, %r182, %r1834, %r1835 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1836, %r1837, %r187, %r188 }, { %r193, %r194, %r195, %r196 }, { %r295, %r296 }, { %r1836, %r1837, %r187, %r188 }; + // end inline asm + .loc 1 150 49 // chunk_fwd.py:150:49 + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1846, %r1847, %r1848, %r1849 }, { %r177, %r178, %r179, %r180 }, { %r265, %r266 }, { %r1846, %r1847, %r1848, %r1849 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1850, %r1851, %r1852, %r1853 }, { %r177, %r178, %r179, %r180 }, { %r267, %r268 }, { %r1850, %r1851, %r1852, %r1853 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1846, %r1847, %r1848, %r1849 }, { %r183, %r184, %r185, %r186 }, { %r269, %r270 }, { %r1846, %r1847, %r1848, %r1849 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1850, %r1851, %r1852, %r1853 }, { %r183, %r184, %r185, %r186 }, { %r271, %r272 }, { %r1850, %r1851, %r1852, %r1853 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1846, %r1847, %r1848, %r1849 }, { %r189, %r190, %r191, %r192 }, { %r273, %r274 }, { %r1846, %r1847, %r1848, %r1849 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1850, %r1851, %r1852, %r1853 }, { %r189, %r190, %r191, %r192 }, { %r275, %r276 }, { %r1850, %r1851, %r1852, %r1853 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1846, %r1847, %r1848, %r1849 }, { %r193, %r194, %r195, %r196 }, { %r277, %r278 }, { %r1846, %r1847, %r1848, %r1849 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1850, %r1851, %r1852, %r1853 }, { %r193, %r194, %r195, %r196 }, { %r279, %r280 }, { %r1850, %r1851, %r1852, %r1853 }; + // end inline asm + .loc 1 148 49 // chunk_fwd.py:148:49 + mov.b64 %rd301, {%r181, %r182}; + mov.b64 %rd302, {%r187, %r188}; + .loc 1 152 15 // chunk_fwd.py:152:15 + @%p34 bra $L__BB0_7; +// %bb.5: // in Loop: Header=BB0_3 Depth=1 + .loc 1 0 15 // chunk_fwd.py:0:15 + setp.le.u64 %p36, %rd2, %rd6; + .loc 1 94 28 // chunk_fwd.py:94:28 + setp.lt.u64 %p35, %rd10, %rd2; + .loc 1 154 31 // chunk_fwd.py:154:31 + add.s64 %rd158, %rd295, -131072; + // begin inline asm + mov.u32 %r198, %r197; + mov.u32 %r199, %r197; + mov.u32 %r200, %r197; + mov.u32 %r201, %r197; + @%p35 ld.global.v4.b32 { %r198, %r199, %r200, %r201 }, [ %rd158 + 0 ]; + // end inline asm + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r223 + 0 ], { %r198, %r199, %r200, %r201 }; + // end inline asm + .loc 1 156 46 // chunk_fwd.py:156:46 + // begin inline asm + st.shared.v4.b32 [ %r228 + 0 ], { %r198, %r199, %r200, %r201 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r297, %r298, %r301, %r302}, [%r20]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r299, %r300, %r303, %r304}, [%r21]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r305, %r306, %r309, %r310}, [%r22]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r307, %r308, %r311, %r312}, [%r23]; + .loc 1 154 31 // chunk_fwd.py:154:31 + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r202, %r203, %r204, %r205}, [%r24]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r208, %r209, %r210, %r211}, [%r26]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r214, %r215, %r216, %r217}, [%r28]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r218, %r219, %r220, %r221}, [%r30]; + .loc 1 156 53 // chunk_fwd.py:156:53 + mov.b64 {%r206, %r207}, %rd299; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r206, %r207, %r1838, %r1839 }, { %r202, %r203, %r204, %r205 }, { %r297, %r298 }, { %r206, %r207, %r1838, %r1839 }; + // end inline asm + mov.b64 {%r212, %r213}, %rd300; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1840, %r1841, %r212, %r213 }, { %r202, %r203, %r204, %r205 }, { %r299, %r300 }, { %r1840, %r1841, %r212, %r213 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r206, %r207, %r1838, %r1839 }, { %r208, %r209, %r210, %r211 }, { %r301, %r302 }, { %r206, %r207, %r1838, %r1839 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1840, %r1841, %r212, %r213 }, { %r208, %r209, %r210, %r211 }, { %r303, %r304 }, { %r1840, %r1841, %r212, %r213 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r206, %r207, %r1838, %r1839 }, { %r214, %r215, %r216, %r217 }, { %r305, %r306 }, { %r206, %r207, %r1838, %r1839 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1840, %r1841, %r212, %r213 }, { %r214, %r215, %r216, %r217 }, { %r307, %r308 }, { %r1840, %r1841, %r212, %r213 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r206, %r207, %r1838, %r1839 }, { %r218, %r219, %r220, %r221 }, { %r309, %r310 }, { %r206, %r207, %r1838, %r1839 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1840, %r1841, %r212, %r213 }, { %r218, %r219, %r220, %r221 }, { %r311, %r312 }, { %r1840, %r1841, %r212, %r213 }; + // end inline asm + .loc 1 158 53 // chunk_fwd.py:158:53 + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1854, %r1855, %r1856, %r1857 }, { %r202, %r203, %r204, %r205 }, { %r265, %r266 }, { %r1854, %r1855, %r1856, %r1857 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1858, %r1859, %r1860, %r1861 }, { %r202, %r203, %r204, %r205 }, { %r267, %r268 }, { %r1858, %r1859, %r1860, %r1861 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1854, %r1855, %r1856, %r1857 }, { %r208, %r209, %r210, %r211 }, { %r269, %r270 }, { %r1854, %r1855, %r1856, %r1857 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1858, %r1859, %r1860, %r1861 }, { %r208, %r209, %r210, %r211 }, { %r271, %r272 }, { %r1858, %r1859, %r1860, %r1861 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1854, %r1855, %r1856, %r1857 }, { %r214, %r215, %r216, %r217 }, { %r273, %r274 }, { %r1854, %r1855, %r1856, %r1857 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1858, %r1859, %r1860, %r1861 }, { %r214, %r215, %r216, %r217 }, { %r275, %r276 }, { %r1858, %r1859, %r1860, %r1861 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1854, %r1855, %r1856, %r1857 }, { %r218, %r219, %r220, %r221 }, { %r277, %r278 }, { %r1854, %r1855, %r1856, %r1857 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1858, %r1859, %r1860, %r1861 }, { %r218, %r219, %r220, %r221 }, { %r279, %r280 }, { %r1858, %r1859, %r1860, %r1861 }; + // end inline asm + .loc 1 159 53 // chunk_fwd.py:159:53 + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1862, %r1863, %r1864, %r1865 }, { %r202, %r203, %r204, %r205 }, { %r281, %r282 }, { %r1862, %r1863, %r1864, %r1865 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1866, %r1867, %r1868, %r1869 }, { %r202, %r203, %r204, %r205 }, { %r283, %r284 }, { %r1866, %r1867, %r1868, %r1869 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1862, %r1863, %r1864, %r1865 }, { %r208, %r209, %r210, %r211 }, { %r285, %r286 }, { %r1862, %r1863, %r1864, %r1865 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1866, %r1867, %r1868, %r1869 }, { %r208, %r209, %r210, %r211 }, { %r287, %r288 }, { %r1866, %r1867, %r1868, %r1869 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1862, %r1863, %r1864, %r1865 }, { %r214, %r215, %r216, %r217 }, { %r289, %r290 }, { %r1862, %r1863, %r1864, %r1865 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1866, %r1867, %r1868, %r1869 }, { %r214, %r215, %r216, %r217 }, { %r291, %r292 }, { %r1866, %r1867, %r1868, %r1869 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1862, %r1863, %r1864, %r1865 }, { %r218, %r219, %r220, %r221 }, { %r293, %r294 }, { %r1862, %r1863, %r1864, %r1865 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1866, %r1867, %r1868, %r1869 }, { %r218, %r219, %r220, %r221 }, { %r295, %r296 }, { %r1866, %r1867, %r1868, %r1869 }; + // end inline asm + .loc 1 152 15 // chunk_fwd.py:152:15 + mov.b64 %rd299, {%r206, %r207}; + mov.b64 %rd300, {%r212, %r213}; + .loc 1 161 19 // chunk_fwd.py:161:19 + @%p36 bra $L__BB0_7; +// %bb.6: // in Loop: Header=BB0_3 Depth=1 + .loc 1 95 28 // chunk_fwd.py:95:28 + setp.lt.u64 %p37, %rd11, %rd2; + mov.b32 %r222, 0; + .loc 1 163 35 // chunk_fwd.py:163:35 + // begin inline asm + mov.u32 %r224, %r222; + mov.u32 %r225, %r222; + mov.u32 %r226, %r222; + mov.u32 %r227, %r222; + @%p37 ld.global.v4.b32 { %r224, %r225, %r226, %r227 }, [ %rd295 + 0 ]; + // end inline asm + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r223 + 0 ], { %r224, %r225, %r226, %r227 }; + // end inline asm + .loc 1 165 50 // chunk_fwd.py:165:50 + // begin inline asm + st.shared.v4.b32 [ %r228 + 0 ], { %r224, %r225, %r226, %r227 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r233, %r234, %r243, %r244}, [%r20]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r235, %r236, %r247, %r248}, [%r21]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r253, %r254, %r261, %r262}, [%r22]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r255, %r256, %r263, %r264}, [%r23]; + .loc 1 163 35 // chunk_fwd.py:163:35 + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r229, %r230, %r231, %r232}, [%r24]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r239, %r240, %r241, %r242}, [%r26]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r249, %r250, %r251, %r252}, [%r28]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r257, %r258, %r259, %r260}, [%r30]; + .loc 1 165 57 // chunk_fwd.py:165:57 + mov.b64 {%r237, %r238}, %rd297; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r237, %r238, %r1842, %r1843 }, { %r229, %r230, %r231, %r232 }, { %r233, %r234 }, { %r237, %r238, %r1842, %r1843 }; + // end inline asm + mov.b64 {%r245, %r246}, %rd298; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1844, %r1845, %r245, %r246 }, { %r229, %r230, %r231, %r232 }, { %r235, %r236 }, { %r1844, %r1845, %r245, %r246 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r237, %r238, %r1842, %r1843 }, { %r239, %r240, %r241, %r242 }, { %r243, %r244 }, { %r237, %r238, %r1842, %r1843 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1844, %r1845, %r245, %r246 }, { %r239, %r240, %r241, %r242 }, { %r247, %r248 }, { %r1844, %r1845, %r245, %r246 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r237, %r238, %r1842, %r1843 }, { %r249, %r250, %r251, %r252 }, { %r253, %r254 }, { %r237, %r238, %r1842, %r1843 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1844, %r1845, %r245, %r246 }, { %r249, %r250, %r251, %r252 }, { %r255, %r256 }, { %r1844, %r1845, %r245, %r246 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r237, %r238, %r1842, %r1843 }, { %r257, %r258, %r259, %r260 }, { %r261, %r262 }, { %r237, %r238, %r1842, %r1843 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1844, %r1845, %r245, %r246 }, { %r257, %r258, %r259, %r260 }, { %r263, %r264 }, { %r1844, %r1845, %r245, %r246 }; + // end inline asm + .loc 1 167 57 // chunk_fwd.py:167:57 + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1870, %r1871, %r1872, %r1873 }, { %r229, %r230, %r231, %r232 }, { %r265, %r266 }, { %r1870, %r1871, %r1872, %r1873 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1874, %r1875, %r1876, %r1877 }, { %r229, %r230, %r231, %r232 }, { %r267, %r268 }, { %r1874, %r1875, %r1876, %r1877 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1870, %r1871, %r1872, %r1873 }, { %r239, %r240, %r241, %r242 }, { %r269, %r270 }, { %r1870, %r1871, %r1872, %r1873 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1874, %r1875, %r1876, %r1877 }, { %r239, %r240, %r241, %r242 }, { %r271, %r272 }, { %r1874, %r1875, %r1876, %r1877 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1870, %r1871, %r1872, %r1873 }, { %r249, %r250, %r251, %r252 }, { %r273, %r274 }, { %r1870, %r1871, %r1872, %r1873 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1874, %r1875, %r1876, %r1877 }, { %r249, %r250, %r251, %r252 }, { %r275, %r276 }, { %r1874, %r1875, %r1876, %r1877 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1870, %r1871, %r1872, %r1873 }, { %r257, %r258, %r259, %r260 }, { %r277, %r278 }, { %r1870, %r1871, %r1872, %r1873 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1874, %r1875, %r1876, %r1877 }, { %r257, %r258, %r259, %r260 }, { %r279, %r280 }, { %r1874, %r1875, %r1876, %r1877 }; + // end inline asm + .loc 1 168 57 // chunk_fwd.py:168:57 + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1878, %r1879, %r1880, %r1881 }, { %r229, %r230, %r231, %r232 }, { %r281, %r282 }, { %r1878, %r1879, %r1880, %r1881 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1882, %r1883, %r1884, %r1885 }, { %r229, %r230, %r231, %r232 }, { %r283, %r284 }, { %r1882, %r1883, %r1884, %r1885 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1878, %r1879, %r1880, %r1881 }, { %r239, %r240, %r241, %r242 }, { %r285, %r286 }, { %r1878, %r1879, %r1880, %r1881 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1882, %r1883, %r1884, %r1885 }, { %r239, %r240, %r241, %r242 }, { %r287, %r288 }, { %r1882, %r1883, %r1884, %r1885 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1878, %r1879, %r1880, %r1881 }, { %r249, %r250, %r251, %r252 }, { %r289, %r290 }, { %r1878, %r1879, %r1880, %r1881 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1882, %r1883, %r1884, %r1885 }, { %r249, %r250, %r251, %r252 }, { %r291, %r292 }, { %r1882, %r1883, %r1884, %r1885 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1878, %r1879, %r1880, %r1881 }, { %r257, %r258, %r259, %r260 }, { %r293, %r294 }, { %r1878, %r1879, %r1880, %r1881 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1882, %r1883, %r1884, %r1885 }, { %r257, %r258, %r259, %r260 }, { %r295, %r296 }, { %r1882, %r1883, %r1884, %r1885 }; + // end inline asm + .loc 1 169 57 // chunk_fwd.py:169:57 + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1886, %r1887, %r1888, %r1889 }, { %r229, %r230, %r231, %r232 }, { %r297, %r298 }, { %r1886, %r1887, %r1888, %r1889 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1890, %r1891, %r1892, %r1893 }, { %r229, %r230, %r231, %r232 }, { %r299, %r300 }, { %r1890, %r1891, %r1892, %r1893 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1886, %r1887, %r1888, %r1889 }, { %r239, %r240, %r241, %r242 }, { %r301, %r302 }, { %r1886, %r1887, %r1888, %r1889 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1890, %r1891, %r1892, %r1893 }, { %r239, %r240, %r241, %r242 }, { %r303, %r304 }, { %r1890, %r1891, %r1892, %r1893 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1886, %r1887, %r1888, %r1889 }, { %r249, %r250, %r251, %r252 }, { %r305, %r306 }, { %r1886, %r1887, %r1888, %r1889 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1890, %r1891, %r1892, %r1893 }, { %r249, %r250, %r251, %r252 }, { %r307, %r308 }, { %r1890, %r1891, %r1892, %r1893 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1886, %r1887, %r1888, %r1889 }, { %r257, %r258, %r259, %r260 }, { %r309, %r310 }, { %r1886, %r1887, %r1888, %r1889 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1890, %r1891, %r1892, %r1893 }, { %r257, %r258, %r259, %r260 }, { %r311, %r312 }, { %r1890, %r1891, %r1892, %r1893 }; + // end inline asm + .loc 1 152 15 // chunk_fwd.py:152:15 + mov.b64 %rd297, {%r237, %r238}; + mov.b64 %rd298, {%r245, %r246}; + bra.uni $L__BB0_7; +$L__BB0_8: // %__nv_exp2f.exit + .loc 1 102 51 // chunk_fwd.py:102:51 + mov.b32 %r313, {%rs1, %rs1}; + mov.b32 %r314, {%rs3, %rs3}; + mov.b32 {%rs10, %rs11}, %r314; + cvt.f32.bf16 %r315, %rs11; + cvt.f32.bf16 %r316, %rs10; + mov.b64 %rd160, {%r316, %r315}; + mov.b32 {%rs12, %rs13}, %r313; + cvt.f32.bf16 %r317, %rs13; + cvt.f32.bf16 %r318, %rs12; + mov.b64 %rd161, {%r318, %r317}; + .loc 1 103 51 // chunk_fwd.py:103:51 + mov.b32 %r319, {%rs4, %rs4}; + mov.b32 %r320, {%rs5, %rs5}; + mov.b32 {%rs14, %rs15}, %r320; + cvt.f32.bf16 %r321, %rs15; + cvt.f32.bf16 %r322, %rs14; + mov.b64 %rd162, {%r322, %r321}; + mov.b32 {%rs16, %rs17}, %r319; + cvt.f32.bf16 %r323, %rs17; + cvt.f32.bf16 %r324, %rs16; + mov.b64 %rd163, {%r324, %r323}; + .loc 1 104 51 // chunk_fwd.py:104:51 + mov.b32 %r325, {%rs6, %rs6}; + mov.b32 %r326, {%rs7, %rs7}; + mov.b32 {%rs18, %rs19}, %r326; + cvt.f32.bf16 %r327, %rs19; + cvt.f32.bf16 %r328, %rs18; + mov.b32 {%rs20, %rs21}, %r325; + cvt.f32.bf16 %r329, %rs21; + cvt.f32.bf16 %r330, %rs20; + .loc 1 105 51 // chunk_fwd.py:105:51 + mov.b32 %r331, {%rs8, %rs8}; + mov.b32 %r332, {%rs9, %rs9}; + mov.b32 {%rs22, %rs23}, %r332; + mov.b32 {%rs24, %rs25}, %r331; + .loc 1 137 21 // chunk_fwd.py:137:21 + cp.async.wait_group 0; + bar.sync 0; + .loc 1 179 25 // chunk_fwd.py:179:25 + mov.b64 {%r337, %r338}, %rd21; + mov.b64 {%r339, %r340}, %rd24; + setp.gt.u32 %p38, %r340, %r338; + setp.gt.u32 %p39, %r340, %r337; + setp.gt.u32 %p40, %r339, %r337; + setp.gt.u32 %p41, %r339, %r338; + setp.gt.u32 %p42, %r12, %r10; + mov.b64 {%r341, %r342}, %rd12; + setp.gt.u32 %p43, %r11, %r341; + setp.gt.u32 %p44, %r11, %r342; + .loc 1 183 86 // chunk_fwd.py:183:86 + sub.f32 %r343, %r69, %r72; + sub.f32 %r344, %r69, %r73; + sub.f32 %r345, %r71, %r72; + sub.f32 %r346, %r71, %r73; +$L__tmp1: + .loc 2 47 33 // op.py:47:33 @[ chunk_fwd.py:183:70 ] + ex2.approx.ftz.f32 %r347, %r343; + ex2.approx.ftz.f32 %r348, %r344; + ex2.approx.ftz.f32 %r349, %r345; + ex2.approx.ftz.f32 %r350, %r346; +$L__tmp2: + .loc 1 183 102 // chunk_fwd.py:183:102 + selp.f32 %r351, %r348, 0f00000000, %p9; + selp.f32 %r352, %r351, 0f00000000, %p41; + selp.f32 %r353, %r352, 0f00000000, %p12; + selp.f32 %r354, %r347, 0f00000000, %p9; + selp.f32 %r355, %r354, 0f00000000, %p40; + selp.f32 %r356, %r355, 0f00000000, %p11; + mov.b64 %rd168, {%r356, %r353}; + selp.f32 %r357, %r349, 0f00000000, %p10; + selp.f32 %r358, %r357, 0f00000000, %p39; + selp.f32 %r359, %r358, 0f00000000, %p11; + selp.f32 %r360, %r350, 0f00000000, %p10; + selp.f32 %r361, %r360, 0f00000000, %p38; + selp.f32 %r362, %r361, 0f00000000, %p12; + mov.b64 %rd169, {%r359, %r362}; + mov.b32 %r363, 0f00000000; + mov.b64 %rd170, {%r363, %r363}; + .loc 1 183 17 // chunk_fwd.py:183:17 + mov.b64 %rd171, {%r1830, %r1831}; + mov.b64 %rd172, {%r1832, %r1833}; + mul.f32x2 %rd173, %rd172, %rd169; + mul.f32x2 %rd174, %rd171, %rd168; + .loc 1 184 86 // chunk_fwd.py:184:86 + sub.f32 %r364, %r78, %r80; + sub.f32 %r365, %r78, %r81; + sub.f32 %r366, %r79, %r80; + sub.f32 %r367, %r79, %r81; + sub.f32 %r368, %r79, %r82; + sub.f32 %r369, %r79, %r83; +$L__tmp3: + .loc 2 47 33 // op.py:47:33 @[ chunk_fwd.py:184:70 ] + ex2.approx.ftz.f32 %r370, %r364; + ex2.approx.ftz.f32 %r371, %r365; + ex2.approx.ftz.f32 %r372, %r366; + ex2.approx.ftz.f32 %r373, %r367; + ex2.approx.ftz.f32 %r374, %r368; + ex2.approx.ftz.f32 %r375, %r369; +$L__tmp4: + .loc 1 184 102 // chunk_fwd.py:184:102 + selp.f32 %r376, %r371, 0f00000000, %p44; + selp.f32 %r377, %r376, 0f00000000, %p18; + selp.f32 %r378, %r377, 0f00000000, %p3; + selp.f32 %r379, %r370, 0f00000000, %p43; + selp.f32 %r380, %r379, 0f00000000, %p17; + selp.f32 %r381, %r380, 0f00000000, %p3; + mov.b64 %rd175, {%r381, %r378}; + selp.f32 %r382, %r374, 0f00000000, %p43; + selp.f32 %r383, %r382, 0f00000000, %p19; + selp.f32 %r384, %r383, 0f00000000, %p4; + selp.f32 %r385, %r375, 0f00000000, %p42; + selp.f32 %r386, %r385, 0f00000000, %p20; + selp.f32 %r387, %r386, 0f00000000, %p4; + mov.b64 %rd176, {%r384, %r387}; + selp.f32 %r388, %r372, 0f00000000, %p17; + selp.f32 %r389, %r388, 0f00000000, %p4; + selp.f32 %r390, %r373, 0f00000000, %p18; + selp.f32 %r391, %r390, 0f00000000, %p4; + .loc 1 184 17 // chunk_fwd.py:184:17 + mul.f32 %r392, %r1834, %r389; + mul.f32 %r393, %r1835, %r391; + mul.f32 %r394, %r1836, 0f00000000; + mul.f32 %r395, %r1837, 0f00000000; + mul.f32x2 %rd177, %rd302, %rd176; + mul.f32x2 %rd178, %rd301, %rd175; + .loc 1 185 86 // chunk_fwd.py:185:86 + sub.f32 %r396, %r84, %r86; + sub.f32 %r397, %r84, %r87; + sub.f32 %r398, %r85, %r86; + sub.f32 %r399, %r85, %r87; + sub.f32 %r400, %r85, %r88; + sub.f32 %r401, %r85, %r89; +$L__tmp5: + .loc 2 47 33 // op.py:47:33 @[ chunk_fwd.py:185:70 ] + ex2.approx.ftz.f32 %r402, %r396; + ex2.approx.ftz.f32 %r403, %r397; + ex2.approx.ftz.f32 %r404, %r398; + ex2.approx.ftz.f32 %r405, %r399; + ex2.approx.ftz.f32 %r406, %r400; + ex2.approx.ftz.f32 %r407, %r401; +$L__tmp6: + .loc 1 185 102 // chunk_fwd.py:185:102 + selp.f32 %r408, %r403, 0f00000000, %p44; + selp.f32 %r409, %r408, 0f00000000, %p22; + selp.f32 %r410, %r409, 0f00000000, %p5; + selp.f32 %r411, %r402, 0f00000000, %p43; + selp.f32 %r412, %r411, 0f00000000, %p21; + selp.f32 %r413, %r412, 0f00000000, %p5; + mov.b64 %rd179, {%r413, %r410}; + selp.f32 %r414, %r406, 0f00000000, %p43; + selp.f32 %r415, %r414, 0f00000000, %p23; + selp.f32 %r416, %r415, 0f00000000, %p6; + selp.f32 %r417, %r407, 0f00000000, %p42; + selp.f32 %r418, %r417, 0f00000000, %p24; + selp.f32 %r419, %r418, 0f00000000, %p6; + mov.b64 %rd180, {%r416, %r419}; + selp.f32 %r420, %r404, 0f00000000, %p21; + selp.f32 %r421, %r420, 0f00000000, %p6; + selp.f32 %r422, %r405, 0f00000000, %p22; + selp.f32 %r423, %r422, 0f00000000, %p6; + .loc 1 185 17 // chunk_fwd.py:185:17 + mul.f32 %r424, %r1838, %r421; + mul.f32 %r425, %r1839, %r423; + mul.f32 %r426, %r1840, 0f00000000; + mul.f32 %r427, %r1841, 0f00000000; + mul.f32x2 %rd181, %rd300, %rd180; + mul.f32x2 %rd182, %rd299, %rd179; + .loc 1 186 86 // chunk_fwd.py:186:86 + sub.f32 %r428, %r90, %r92; + sub.f32 %r429, %r90, %r93; + sub.f32 %r430, %r91, %r92; + sub.f32 %r431, %r91, %r93; + sub.f32 %r432, %r91, %r94; + sub.f32 %r433, %r91, %r95; +$L__tmp7: + .loc 2 47 33 // op.py:47:33 @[ chunk_fwd.py:186:70 ] + ex2.approx.ftz.f32 %r434, %r428; + ex2.approx.ftz.f32 %r435, %r429; + ex2.approx.ftz.f32 %r436, %r430; + ex2.approx.ftz.f32 %r437, %r431; + ex2.approx.ftz.f32 %r438, %r432; + ex2.approx.ftz.f32 %r439, %r433; +$L__tmp8: + .loc 1 186 102 // chunk_fwd.py:186:102 + selp.f32 %r440, %r435, 0f00000000, %p44; + selp.f32 %r441, %r440, 0f00000000, %p26; + selp.f32 %r442, %r441, 0f00000000, %p7; + selp.f32 %r443, %r434, 0f00000000, %p43; + selp.f32 %r444, %r443, 0f00000000, %p25; + selp.f32 %r445, %r444, 0f00000000, %p7; + mov.b64 %rd183, {%r445, %r442}; + selp.f32 %r446, %r438, 0f00000000, %p43; + selp.f32 %r447, %r446, 0f00000000, %p27; + selp.f32 %r448, %r447, 0f00000000, %p8; + selp.f32 %r449, %r439, 0f00000000, %p42; + selp.f32 %r450, %r449, 0f00000000, %p28; + selp.f32 %r451, %r450, 0f00000000, %p8; + mov.b64 %rd184, {%r448, %r451}; + selp.f32 %r452, %r436, 0f00000000, %p25; + selp.f32 %r453, %r452, 0f00000000, %p8; + selp.f32 %r454, %r437, 0f00000000, %p26; + selp.f32 %r455, %r454, 0f00000000, %p8; + .loc 1 186 17 // chunk_fwd.py:186:17 + mul.f32x2 %rd185, %rd298, %rd184; + mul.f32x2 %rd186, %rd297, %rd183; + .loc 1 188 80 // chunk_fwd.py:188:80 + sub.f32 %r460, %r78, %r74; + sub.f32 %r461, %r78, %r75; + sub.f32 %r462, %r79, %r74; + sub.f32 %r463, %r79, %r75; + sub.f32 %r464, %r78, %r76; + sub.f32 %r465, %r78, %r77; + sub.f32 %r466, %r79, %r76; + sub.f32 %r467, %r79, %r77; +$L__tmp9: + .loc 2 47 33 // op.py:47:33 @[ chunk_fwd.py:188:64 ] + ex2.approx.ftz.f32 %r468, %r460; + ex2.approx.ftz.f32 %r469, %r461; + ex2.approx.ftz.f32 %r470, %r462; + ex2.approx.ftz.f32 %r471, %r463; + ex2.approx.ftz.f32 %r472, %r464; + ex2.approx.ftz.f32 %r473, %r465; + ex2.approx.ftz.f32 %r474, %r466; + ex2.approx.ftz.f32 %r475, %r467; +$L__tmp10: + .loc 1 188 96 // chunk_fwd.py:188:96 + selp.f32 %r476, %r468, 0f00000000, %p13; + selp.f32 %r478, %r469, 0f00000000, %p14; + selp.f32 %r480, %r470, 0f00000000, %p13; + selp.f32 %r482, %r471, 0f00000000, %p14; + selp.f32 %r484, %r472, 0f00000000, %p15; + selp.f32 %r486, %r473, 0f00000000, %p16; + selp.f32 %r488, %r474, 0f00000000, %p15; + selp.f32 %r490, %r475, 0f00000000, %p16; + .loc 1 189 80 // chunk_fwd.py:189:80 + sub.f32 %r500, %r84, %r74; + sub.f32 %r501, %r84, %r75; + sub.f32 %r502, %r85, %r74; + sub.f32 %r503, %r85, %r75; + sub.f32 %r504, %r84, %r76; + sub.f32 %r505, %r84, %r77; + sub.f32 %r506, %r85, %r76; + sub.f32 %r507, %r85, %r77; +$L__tmp11: + .loc 2 47 33 // op.py:47:33 @[ chunk_fwd.py:189:64 ] + ex2.approx.ftz.f32 %r508, %r500; + ex2.approx.ftz.f32 %r509, %r501; + ex2.approx.ftz.f32 %r510, %r502; + ex2.approx.ftz.f32 %r511, %r503; + ex2.approx.ftz.f32 %r512, %r504; + ex2.approx.ftz.f32 %r513, %r505; + ex2.approx.ftz.f32 %r514, %r506; + ex2.approx.ftz.f32 %r515, %r507; +$L__tmp12: + .loc 1 189 96 // chunk_fwd.py:189:96 + selp.f32 %r516, %r508, 0f00000000, %p13; + selp.f32 %r518, %r509, 0f00000000, %p14; + selp.f32 %r520, %r510, 0f00000000, %p13; + selp.f32 %r522, %r511, 0f00000000, %p14; + selp.f32 %r524, %r512, 0f00000000, %p15; + selp.f32 %r526, %r513, 0f00000000, %p16; + selp.f32 %r528, %r514, 0f00000000, %p15; + selp.f32 %r530, %r515, 0f00000000, %p16; + .loc 1 190 80 // chunk_fwd.py:190:80 + sub.f32 %r540, %r84, %r80; + sub.f32 %r541, %r84, %r81; + sub.f32 %r542, %r85, %r80; + sub.f32 %r543, %r85, %r81; + sub.f32 %r544, %r84, %r82; + sub.f32 %r545, %r84, %r83; + sub.f32 %r546, %r85, %r82; + sub.f32 %r547, %r85, %r83; +$L__tmp13: + .loc 2 47 33 // op.py:47:33 @[ chunk_fwd.py:190:64 ] + ex2.approx.ftz.f32 %r548, %r540; + ex2.approx.ftz.f32 %r549, %r541; + ex2.approx.ftz.f32 %r550, %r542; + ex2.approx.ftz.f32 %r551, %r543; + ex2.approx.ftz.f32 %r552, %r544; + ex2.approx.ftz.f32 %r553, %r545; + ex2.approx.ftz.f32 %r554, %r546; + ex2.approx.ftz.f32 %r555, %r547; +$L__tmp14: + .loc 1 190 96 // chunk_fwd.py:190:96 + selp.f32 %r556, %r548, 0f00000000, %p17; + selp.f32 %r558, %r549, 0f00000000, %p18; + selp.f32 %r560, %r550, 0f00000000, %p17; + selp.f32 %r562, %r551, 0f00000000, %p18; + selp.f32 %r564, %r552, 0f00000000, %p19; + selp.f32 %r566, %r553, 0f00000000, %p20; + selp.f32 %r568, %r554, 0f00000000, %p19; + selp.f32 %r570, %r555, 0f00000000, %p20; + .loc 1 191 80 // chunk_fwd.py:191:80 + sub.f32 %r580, %r90, %r74; + sub.f32 %r581, %r90, %r75; + sub.f32 %r582, %r91, %r74; + sub.f32 %r583, %r91, %r75; + sub.f32 %r584, %r90, %r76; + sub.f32 %r585, %r90, %r77; + sub.f32 %r586, %r91, %r76; + sub.f32 %r587, %r91, %r77; +$L__tmp15: + .loc 2 47 33 // op.py:47:33 @[ chunk_fwd.py:191:64 ] + ex2.approx.ftz.f32 %r588, %r580; + ex2.approx.ftz.f32 %r589, %r581; + ex2.approx.ftz.f32 %r590, %r582; + ex2.approx.ftz.f32 %r591, %r583; + ex2.approx.ftz.f32 %r592, %r584; + ex2.approx.ftz.f32 %r593, %r585; + ex2.approx.ftz.f32 %r594, %r586; + ex2.approx.ftz.f32 %r595, %r587; +$L__tmp16: + .loc 1 191 96 // chunk_fwd.py:191:96 + selp.f32 %r596, %r588, 0f00000000, %p13; + selp.f32 %r598, %r589, 0f00000000, %p14; + selp.f32 %r600, %r590, 0f00000000, %p13; + selp.f32 %r602, %r591, 0f00000000, %p14; + selp.f32 %r604, %r592, 0f00000000, %p15; + selp.f32 %r606, %r593, 0f00000000, %p16; + selp.f32 %r608, %r594, 0f00000000, %p15; + selp.f32 %r610, %r595, 0f00000000, %p16; + .loc 1 192 80 // chunk_fwd.py:192:80 + sub.f32 %r620, %r90, %r80; + sub.f32 %r621, %r90, %r81; + sub.f32 %r622, %r91, %r80; + sub.f32 %r623, %r91, %r81; + sub.f32 %r624, %r90, %r82; + sub.f32 %r625, %r90, %r83; + sub.f32 %r626, %r91, %r82; + sub.f32 %r627, %r91, %r83; +$L__tmp17: + .loc 2 47 33 // op.py:47:33 @[ chunk_fwd.py:192:64 ] + ex2.approx.ftz.f32 %r628, %r620; + ex2.approx.ftz.f32 %r629, %r621; + ex2.approx.ftz.f32 %r630, %r622; + ex2.approx.ftz.f32 %r631, %r623; + ex2.approx.ftz.f32 %r632, %r624; + ex2.approx.ftz.f32 %r633, %r625; + ex2.approx.ftz.f32 %r634, %r626; + ex2.approx.ftz.f32 %r635, %r627; +$L__tmp18: + .loc 1 192 96 // chunk_fwd.py:192:96 + selp.f32 %r636, %r628, 0f00000000, %p17; + selp.f32 %r638, %r629, 0f00000000, %p18; + selp.f32 %r640, %r630, 0f00000000, %p17; + selp.f32 %r642, %r631, 0f00000000, %p18; + selp.f32 %r644, %r632, 0f00000000, %p19; + selp.f32 %r646, %r633, 0f00000000, %p20; + selp.f32 %r648, %r634, 0f00000000, %p19; + selp.f32 %r650, %r635, 0f00000000, %p20; + .loc 1 193 80 // chunk_fwd.py:193:80 + sub.f32 %r660, %r90, %r86; + sub.f32 %r661, %r90, %r87; + sub.f32 %r662, %r91, %r86; + sub.f32 %r663, %r91, %r87; + sub.f32 %r664, %r90, %r88; + sub.f32 %r665, %r90, %r89; + sub.f32 %r666, %r91, %r88; + sub.f32 %r667, %r91, %r89; +$L__tmp19: + .loc 2 47 33 // op.py:47:33 @[ chunk_fwd.py:193:64 ] + ex2.approx.ftz.f32 %r668, %r660; + ex2.approx.ftz.f32 %r669, %r661; + ex2.approx.ftz.f32 %r670, %r662; + ex2.approx.ftz.f32 %r671, %r663; + ex2.approx.ftz.f32 %r672, %r664; + ex2.approx.ftz.f32 %r673, %r665; + ex2.approx.ftz.f32 %r674, %r666; + ex2.approx.ftz.f32 %r675, %r667; +$L__tmp20: + .loc 1 193 96 // chunk_fwd.py:193:96 + selp.f32 %r676, %r668, 0f00000000, %p21; + selp.f32 %r678, %r669, 0f00000000, %p22; + selp.f32 %r680, %r670, 0f00000000, %p21; + selp.f32 %r682, %r671, 0f00000000, %p22; + selp.f32 %r684, %r672, 0f00000000, %p23; + selp.f32 %r686, %r673, 0f00000000, %p24; + selp.f32 %r688, %r674, 0f00000000, %p23; + selp.f32 %r690, %r675, 0f00000000, %p24; + .loc 1 222 14 // chunk_fwd.py:222:14 + mov.b64 {%r700, %r701}, %rd174; + neg.f32 %r702, %r701; + neg.f32 %r703, %r700; + mov.b64 %rd187, {%r703, %r702}; + fma.rn.f32x2 %rd304, %rd187, %rd161, %rd170; + mov.b64 {%r704, %r705}, %rd173; + neg.f32 %r706, %r705; + neg.f32 %r707, %r704; + mov.b64 %rd188, {%r707, %r706}; + fma.rn.f32x2 %rd305, %rd188, %rd160, %rd170; + .loc 1 223 14 // chunk_fwd.py:223:14 + neg.f32 %r708, %r392; + neg.f32 %r709, %r393; + neg.f32 %r710, %r394; + neg.f32 %r711, %r395; + mov.b64 {%r712, %r713}, %rd178; + neg.f32 %r714, %r713; + neg.f32 %r715, %r712; + mov.b64 %rd189, {%r715, %r714}; + mov.b64 {%r716, %r717}, %rd177; + neg.f32 %r718, %r717; + neg.f32 %r719, %r716; + mov.b64 %rd190, {%r719, %r718}; + .loc 1 224 14 // chunk_fwd.py:224:14 + mov.b64 {%r724, %r725}, %rd182; + neg.f32 %r726, %r725; + neg.f32 %r727, %r724; + mov.b64 {%r728, %r729}, %rd181; + neg.f32 %r730, %r729; + neg.f32 %r731, %r728; + .loc 1 225 14 // chunk_fwd.py:225:14 + mov.b64 {%r736, %r737}, %rd186; + mov.b64 {%r740, %r741}, %rd185; + .loc 1 227 34 // chunk_fwd.py:227:34 + sub.s64 %rd46, %rd2, %rd3; + .loc 1 227 22 // chunk_fwd.py:227:22 + setp.lt.s64 %p45, %rd46, 3; + @%p45 bra $L__BB0_11; +// %bb.9: // %.lr.ph + .loc 1 0 22 // chunk_fwd.py:0:22 + shl.b32 %r744, %r4, 4; + shr.u32 %r745, %r5, 3; + add.s32 %r746, %r109, %r744; + add.s32 %r748, %r746, %r745; + add.s32 %r750, %r748, 8; + shl.b32 %r747, %r2, 1; + add.s32 %r43, %r109, %r747; + .loc 1 227 22 // chunk_fwd.py:227:22 + min.u64 %rd47, %rd46, 16; + .loc 1 228 61 // chunk_fwd.py:228:61 + mov.b64 {%r44, %r45}, %rd304; + mov.b64 {%r46, %r47}, %rd305; + mov.b64 %rd303, 2; +$L__BB0_10: // =>This Inner Loop Header: Depth=1 + .loc 1 228 40 // chunk_fwd.py:228:40 + setp.eq.b64 %p46, %rd27, %rd303; + setp.eq.b64 %p47, %rd25, %rd303; + .loc 1 228 61 // chunk_fwd.py:228:61 + selp.f32 %r752, %r44, 0f00000000, %p46; + selp.f32 %r753, %r45, 0f00000000, %p46; + selp.f32 %r754, %r46, 0f00000000, %p47; + selp.f32 %r755, %r47, 0f00000000, %p47; +$L__tmp21: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:228:66 ] ] + add.f32 %r756, %r752, %r754; + add.f32 %r757, %r753, %r755; +$L__tmp22: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:228:66 ] + shfl.sync.bfly.b32 %r758, %r756, 16, 31, -1; +$L__tmp23: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:228:66 ] ] + add.f32 %r759, %r756, %r758; +$L__tmp24: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:228:66 ] + shfl.sync.bfly.b32 %r760, %r759, 8, 31, -1; +$L__tmp25: + .loc 1 229 31 // chunk_fwd.py:229:31 + setp.gt.u64 %p48, %rd303, %rd22; + setp.gt.u64 %p49, %rd303, %rd23; +$L__tmp26: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:228:66 ] ] + add.f32 %r761, %r759, %r760; +$L__tmp27: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:228:66 ] + shfl.sync.bfly.b32 %r762, %r761, 4, 31, -1; + shfl.sync.bfly.b32 %r763, %r757, 16, 31, -1; +$L__tmp28: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:228:66 ] ] + add.f32 %r764, %r757, %r763; +$L__tmp29: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:228:66 ] + shfl.sync.bfly.b32 %r765, %r764, 8, 31, -1; +$L__tmp30: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:228:66 ] ] + add.f32 %r766, %r764, %r765; +$L__tmp31: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:228:66 ] + shfl.sync.bfly.b32 %r767, %r766, 4, 31, -1; + mov.b64 %rd195, {%r762, %r767}; +$L__tmp32: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:228:66 ] ] + mov.b64 %rd196, {%r761, %r766}; + add.f32x2 %rd197, %rd196, %rd195; +$L__tmp33: + .loc 1 229 41 // chunk_fwd.py:229:41 + mov.b64 {%r768, %r769}, %rd197; + selp.f32 %r751, %r769, 0f00000000, %p49; + selp.f32 %r749, %r768, 0f00000000, %p48; + mov.b64 %rd198, {%r749, %r751}; + .loc 1 230 48 // chunk_fwd.py:230:48 + bar.sync 0; + // begin inline asm + st.shared.b32 [ %r748 + 0 ], %r749; + // end inline asm + // begin inline asm + st.shared.b32 [ %r750 + 0 ], %r751; + // end inline asm + bar.sync 0; + ld.shared.v2.b32 {%r770, %r771}, [%r43]; + mov.b64 {%r772, %r773}, %rd304; + mov.b64 {%r774, %r775}, %rd305; + mul.f32 %r776, %r774, %r771; + mul.f32 %r777, %r775, %r771; +$L__tmp34: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:230:56 ] ] + fma.rn.f32 %r778, %r772, %r770, %r776; + fma.rn.f32 %r779, %r773, %r770, %r777; +$L__tmp35: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:230:56 ] + shfl.sync.bfly.b32 %r780, %r778, 16, 31, -1; +$L__tmp36: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:230:56 ] ] + add.f32 %r781, %r778, %r780; +$L__tmp37: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:230:56 ] + shfl.sync.bfly.b32 %r782, %r781, 8, 31, -1; +$L__tmp38: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:230:56 ] ] + add.f32 %r783, %r781, %r782; +$L__tmp39: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:230:56 ] + shfl.sync.bfly.b32 %r784, %r783, 4, 31, -1; + shfl.sync.bfly.b32 %r785, %r779, 16, 31, -1; +$L__tmp40: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:230:56 ] ] + add.f32 %r786, %r779, %r785; +$L__tmp41: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:230:56 ] + shfl.sync.bfly.b32 %r787, %r786, 8, 31, -1; +$L__tmp42: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:230:56 ] ] + add.f32 %r788, %r786, %r787; +$L__tmp43: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:230:56 ] + shfl.sync.bfly.b32 %r789, %r788, 4, 31, -1; + mov.b64 %rd199, {%r784, %r789}; +$L__tmp44: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:230:56 ] ] + mov.b64 %rd200, {%r783, %r788}; + add.f32x2 %rd201, %rd200, %rd199; +$L__tmp45: + .loc 1 230 24 // chunk_fwd.py:230:24 + add.f32x2 %rd202, %rd198, %rd201; + .loc 1 231 54 // chunk_fwd.py:231:54 + mov.b64 {%r790, %r791}, %rd202; + selp.f32 %r792, %r791, %r775, %p47; + selp.f32 %r793, %r790, %r774, %p47; + mov.b64 %rd305, {%r793, %r792}; + selp.f32 %r794, %r791, %r773, %p46; + selp.f32 %r795, %r790, %r772, %p46; + mov.b64 %rd304, {%r795, %r794}; + .loc 1 227 22 // chunk_fwd.py:227:22 + add.s64 %rd303, %rd303, 1; + setp.ne.b64 %p50, %rd47, %rd303; + @%p50 bra $L__BB0_10; +$L__BB0_11: // %._crit_edge + .loc 1 0 0 // chunk_fwd.py:0 + cvt.f32.bf16 %r333, %rs23; + cvt.f32.bf16 %r334, %rs22; + cvt.f32.bf16 %r335, %rs25; + cvt.f32.bf16 %r336, %rs24; + mul.f32 %r456, %r1842, %r453; + mul.f32 %r457, %r1843, %r455; + mul.f32 %r458, %r1844, 0f00000000; + mul.f32 %r459, %r1845, 0f00000000; + selp.f32 %r477, %r476, 0f00000000, %p3; + selp.f32 %r479, %r478, 0f00000000, %p3; + selp.f32 %r481, %r480, 0f00000000, %p4; + selp.f32 %r483, %r482, 0f00000000, %p4; + selp.f32 %r485, %r484, 0f00000000, %p3; + selp.f32 %r487, %r486, 0f00000000, %p3; + selp.f32 %r489, %r488, 0f00000000, %p4; + selp.f32 %r491, %r490, 0f00000000, %p4; + selp.f32 %r517, %r516, 0f00000000, %p5; + selp.f32 %r519, %r518, 0f00000000, %p5; + selp.f32 %r521, %r520, 0f00000000, %p6; + selp.f32 %r523, %r522, 0f00000000, %p6; + selp.f32 %r525, %r524, 0f00000000, %p5; + selp.f32 %r527, %r526, 0f00000000, %p5; + selp.f32 %r529, %r528, 0f00000000, %p6; + selp.f32 %r531, %r530, 0f00000000, %p6; + selp.f32 %r557, %r556, 0f00000000, %p5; + selp.f32 %r559, %r558, 0f00000000, %p5; + selp.f32 %r561, %r560, 0f00000000, %p6; + selp.f32 %r563, %r562, 0f00000000, %p6; + selp.f32 %r565, %r564, 0f00000000, %p5; + selp.f32 %r567, %r566, 0f00000000, %p5; + selp.f32 %r569, %r568, 0f00000000, %p6; + selp.f32 %r571, %r570, 0f00000000, %p6; + selp.f32 %r597, %r596, 0f00000000, %p7; + selp.f32 %r599, %r598, 0f00000000, %p7; + selp.f32 %r601, %r600, 0f00000000, %p8; + selp.f32 %r603, %r602, 0f00000000, %p8; + selp.f32 %r605, %r604, 0f00000000, %p7; + selp.f32 %r607, %r606, 0f00000000, %p7; + selp.f32 %r609, %r608, 0f00000000, %p8; + selp.f32 %r611, %r610, 0f00000000, %p8; + selp.f32 %r637, %r636, 0f00000000, %p7; + selp.f32 %r639, %r638, 0f00000000, %p7; + selp.f32 %r641, %r640, 0f00000000, %p8; + selp.f32 %r643, %r642, 0f00000000, %p8; + selp.f32 %r645, %r644, 0f00000000, %p7; + selp.f32 %r647, %r646, 0f00000000, %p7; + selp.f32 %r649, %r648, 0f00000000, %p8; + selp.f32 %r651, %r650, 0f00000000, %p8; + selp.f32 %r677, %r676, 0f00000000, %p7; + selp.f32 %r679, %r678, 0f00000000, %p7; + selp.f32 %r681, %r680, 0f00000000, %p8; + selp.f32 %r683, %r682, 0f00000000, %p8; + selp.f32 %r685, %r684, 0f00000000, %p7; + selp.f32 %r687, %r686, 0f00000000, %p7; + selp.f32 %r689, %r688, 0f00000000, %p8; + selp.f32 %r691, %r690, 0f00000000, %p8; + neg.f32 %r738, %r737; + neg.f32 %r739, %r736; + neg.f32 %r742, %r741; + neg.f32 %r743, %r740; + mov.b64 %rd164, {%r328, %r327}; + mov.b64 %rd165, {%r330, %r329}; + neg.f32 %r720, %r424; + neg.f32 %r721, %r425; + neg.f32 %r722, %r426; + neg.f32 %r723, %r427; + mov.b64 %rd191, {%r727, %r726}; + mov.b64 %rd192, {%r731, %r730}; + fma.rn.f32 %r1898, %r708, %r322, 0f00000000; + fma.rn.f32 %r1899, %r709, %r322, 0f00000000; + fma.rn.f32 %r1900, %r710, %r324, 0f00000000; + fma.rn.f32 %r1901, %r711, %r324, 0f00000000; + fma.rn.f32x2 %rd307, %rd189, %rd163, %rd170; + fma.rn.f32x2 %rd308, %rd190, %rd162, %rd170; + .loc 1 232 34 // chunk_fwd.py:232:34 + sub.s64 %rd48, %rd2, %rd4; + .loc 1 232 22 // chunk_fwd.py:232:22 + setp.lt.s64 %p51, %rd48, 3; + @%p51 bra $L__BB0_15; +// %bb.12: // %.lr.ph454 + .loc 1 0 22 // chunk_fwd.py:0:22 + and.b32 %r48, %r1, 4; + and.b32 %r796, %r31, 24; + bfe.u32 %r797, %r1, 3, 2; + or.b32 %r798, %r797, %r796; + or.b32 %r49, %r798, %r48; + xor.b32 %r50, %r49, 4; + .loc 1 232 22 // chunk_fwd.py:232:22 + min.u64 %rd49, %rd48, 16; + .loc 1 233 61 // chunk_fwd.py:233:61 + mov.b64 {%r51, %r52}, %rd307; + mov.b64 {%r53, %r54}, %rd308; + mov.b64 %rd306, 2; + setp.eq.b32 %p52, %r48, 0; + mov.b32 %r1894, %r1901; + mov.b32 %r1895, %r1900; + mov.b32 %r1896, %r1899; + mov.b32 %r1897, %r1898; +$L__BB0_13: // =>This Inner Loop Header: Depth=1 + .loc 1 233 40 // chunk_fwd.py:233:40 + setp.eq.b64 %p53, %rd27, %rd306; + setp.eq.b64 %p54, %rd25, %rd306; + .loc 1 233 61 // chunk_fwd.py:233:61 + selp.f32 %r799, %r51, 0f00000000, %p53; + selp.f32 %r800, %r52, 0f00000000, %p53; + selp.f32 %r801, %r1898, 0f00000000, %p54; + selp.f32 %r802, %r1899, 0f00000000, %p54; + selp.f32 %r803, %r1900, 0f00000000, %p53; + selp.f32 %r804, %r1901, 0f00000000, %p53; + selp.f32 %r805, %r53, 0f00000000, %p54; + selp.f32 %r806, %r54, 0f00000000, %p54; +$L__tmp46: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:233:66 ] ] + add.f32 %r807, %r799, %r801; + add.f32 %r808, %r800, %r802; + add.f32 %r809, %r803, %r805; + add.f32 %r810, %r804, %r806; +$L__tmp47: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:233:66 ] + shfl.sync.bfly.b32 %r811, %r807, 16, 31, -1; +$L__tmp48: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:233:66 ] ] + add.f32 %r812, %r807, %r811; +$L__tmp49: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:233:66 ] + shfl.sync.bfly.b32 %r813, %r812, 8, 31, -1; +$L__tmp50: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:233:66 ] ] + add.f32 %r814, %r812, %r813; +$L__tmp51: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:233:66 ] + shfl.sync.bfly.b32 %r815, %r814, 4, 31, -1; + shfl.sync.bfly.b32 %r816, %r808, 16, 31, -1; +$L__tmp52: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:233:66 ] ] + add.f32 %r817, %r808, %r816; +$L__tmp53: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:233:66 ] + shfl.sync.bfly.b32 %r818, %r817, 8, 31, -1; +$L__tmp54: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:233:66 ] ] + add.f32 %r819, %r817, %r818; +$L__tmp55: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:233:66 ] + shfl.sync.bfly.b32 %r820, %r819, 4, 31, -1; + shfl.sync.bfly.b32 %r821, %r809, 16, 31, -1; +$L__tmp56: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:233:66 ] ] + add.f32 %r822, %r809, %r821; +$L__tmp57: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:233:66 ] + shfl.sync.bfly.b32 %r823, %r822, 8, 31, -1; +$L__tmp58: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:233:66 ] ] + add.f32 %r824, %r822, %r823; +$L__tmp59: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:233:66 ] + shfl.sync.bfly.b32 %r825, %r824, 4, 31, -1; + shfl.sync.bfly.b32 %r826, %r810, 16, 31, -1; +$L__tmp60: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:233:66 ] ] + add.f32 %r827, %r810, %r826; +$L__tmp61: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:233:66 ] + shfl.sync.bfly.b32 %r828, %r827, 8, 31, -1; +$L__tmp62: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:233:66 ] ] + add.f32 %r829, %r827, %r828; +$L__tmp63: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:233:66 ] + shfl.sync.bfly.b32 %r830, %r829, 4, 31, -1; + mov.b64 %rd203, {%r815, %r820}; + mov.b64 %rd204, {%r825, %r830}; +$L__tmp64: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:233:66 ] ] + mov.b64 %rd205, {%r814, %r819}; + mov.b64 %rd206, {%r824, %r829}; + add.f32x2 %rd207, %rd206, %rd204; + add.f32x2 %rd208, %rd205, %rd203; +$L__tmp65: + .loc 1 234 31 // chunk_fwd.py:234:31 + setp.gt.u64 %p55, %rd306, %rd15; + setp.gt.u64 %p56, %rd306, %rd16; + setp.gt.u64 %p57, %rd306, %rd13; + setp.gt.u64 %p58, %rd306, %rd14; + .loc 1 234 41 // chunk_fwd.py:234:41 + mov.b64 {%r831, %r832}, %rd208; + selp.f32 %r833, %r832, 0f00000000, %p58; + selp.f32 %r834, %r831, 0f00000000, %p57; + mov.b64 %rd209, {%r834, %r833}; + mov.b64 {%r835, %r836}, %rd207; + selp.f32 %r837, %r836, 0f00000000, %p56; + selp.f32 %r838, %r835, 0f00000000, %p55; + mov.b64 %rd210, {%r838, %r837}; + .loc 1 235 48 // chunk_fwd.py:235:48 + selp.f32 %r839, %r834, %r833, %p52; + selp.f32 %r840, %r833, %r834, %p52; + selp.f32 %r841, %r838, %r837, %p52; + selp.f32 %r842, %r837, %r838, %p52; + shfl.sync.idx.b32 %r843, %r839, %r49, 31, -1; + shfl.sync.idx.b32 %r844, %r841, %r49, 31, -1; + shfl.sync.idx.b32 %r845, %r840, %r50, 31, -1; + shfl.sync.idx.b32 %r846, %r842, %r50, 31, -1; + selp.b32 %r847, %r843, %r845, %p52; + selp.b32 %r848, %r844, %r846, %p52; + mov.b64 {%r849, %r850}, %rd307; + mul.f32 %r851, %r1897, %r848; + mul.f32 %r852, %r1896, %r848; + mov.b64 {%r853, %r854}, %rd308; + mul.f32 %r855, %r853, %r848; + mul.f32 %r856, %r854, %r848; +$L__tmp66: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:235:56 ] ] + fma.rn.f32 %r857, %r849, %r847, %r851; + fma.rn.f32 %r858, %r850, %r847, %r852; + fma.rn.f32 %r859, %r1895, %r847, %r855; + fma.rn.f32 %r860, %r1894, %r847, %r856; +$L__tmp67: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:235:56 ] + shfl.sync.bfly.b32 %r861, %r857, 16, 31, -1; +$L__tmp68: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:235:56 ] ] + add.f32 %r862, %r857, %r861; +$L__tmp69: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:235:56 ] + shfl.sync.bfly.b32 %r863, %r862, 8, 31, -1; +$L__tmp70: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:235:56 ] ] + add.f32 %r864, %r862, %r863; +$L__tmp71: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:235:56 ] + shfl.sync.bfly.b32 %r865, %r864, 4, 31, -1; + shfl.sync.bfly.b32 %r866, %r858, 16, 31, -1; +$L__tmp72: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:235:56 ] ] + add.f32 %r867, %r858, %r866; +$L__tmp73: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:235:56 ] + shfl.sync.bfly.b32 %r868, %r867, 8, 31, -1; +$L__tmp74: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:235:56 ] ] + add.f32 %r869, %r867, %r868; +$L__tmp75: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:235:56 ] + shfl.sync.bfly.b32 %r870, %r869, 4, 31, -1; + shfl.sync.bfly.b32 %r871, %r859, 16, 31, -1; +$L__tmp76: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:235:56 ] ] + add.f32 %r872, %r859, %r871; +$L__tmp77: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:235:56 ] + shfl.sync.bfly.b32 %r873, %r872, 8, 31, -1; +$L__tmp78: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:235:56 ] ] + add.f32 %r874, %r872, %r873; +$L__tmp79: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:235:56 ] + shfl.sync.bfly.b32 %r875, %r874, 4, 31, -1; + shfl.sync.bfly.b32 %r876, %r860, 16, 31, -1; +$L__tmp80: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:235:56 ] ] + add.f32 %r877, %r860, %r876; +$L__tmp81: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:235:56 ] + shfl.sync.bfly.b32 %r878, %r877, 8, 31, -1; +$L__tmp82: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:235:56 ] ] + add.f32 %r879, %r877, %r878; +$L__tmp83: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:235:56 ] + shfl.sync.bfly.b32 %r880, %r879, 4, 31, -1; + mov.b64 %rd211, {%r875, %r880}; + mov.b64 %rd212, {%r865, %r870}; +$L__tmp84: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:235:56 ] ] + mov.b64 %rd213, {%r874, %r879}; + mov.b64 %rd214, {%r864, %r869}; + add.f32x2 %rd215, %rd214, %rd212; + add.f32x2 %rd216, %rd213, %rd211; +$L__tmp85: + .loc 1 235 24 // chunk_fwd.py:235:24 + add.f32x2 %rd217, %rd210, %rd216; + add.f32x2 %rd218, %rd209, %rd215; + .loc 1 236 54 // chunk_fwd.py:236:54 + mov.b64 {%r881, %r882}, %rd218; + selp.f32 %r1897, %r881, %r1897, %p54; + selp.f32 %r1896, %r882, %r1896, %p54; + mov.b64 {%r883, %r884}, %rd217; + selp.f32 %r1895, %r883, %r1895, %p53; + selp.f32 %r1894, %r884, %r1894, %p53; + selp.f32 %r885, %r882, %r850, %p53; + selp.f32 %r886, %r881, %r849, %p53; + mov.b64 %rd307, {%r886, %r885}; + selp.f32 %r887, %r884, %r854, %p54; + selp.f32 %r888, %r883, %r853, %p54; + mov.b64 %rd308, {%r888, %r887}; + .loc 1 232 22 // chunk_fwd.py:232:22 + add.s64 %rd306, %rd306, 1; + setp.ne.b64 %p59, %rd49, %rd306; + @%p59 bra $L__BB0_13; +// %bb.14: // %._crit_edge455.loopexit + .loc 1 249 14 // chunk_fwd.py:249:14 + add.f32 %r1898, %r1897, 0f00000000; + add.f32 %r1899, %r1896, 0f00000000; + add.f32 %r1900, %r1895, 0f00000000; + add.f32 %r1901, %r1894, 0f00000000; +$L__BB0_15: // %._crit_edge455 + .loc 1 0 14 // chunk_fwd.py:0:14 + shl.b64 %rd159, %rd7, 7; + mov.b64 %rd166, {%r334, %r333}; + mov.b64 %rd167, {%r336, %r335}; + mul.f32 %r492, %r1846, %r477; + mul.f32 %r493, %r1847, %r479; + mul.f32 %r494, %r1848, %r481; + mul.f32 %r495, %r1849, %r483; + mul.f32 %r496, %r1850, %r485; + mul.f32 %r497, %r1851, %r487; + mul.f32 %r498, %r1852, %r489; + mul.f32 %r499, %r1853, %r491; + mul.f32 %r532, %r1854, %r517; + mul.f32 %r533, %r1855, %r519; + mul.f32 %r534, %r1856, %r521; + mul.f32 %r535, %r1857, %r523; + mul.f32 %r536, %r1858, %r525; + mul.f32 %r537, %r1859, %r527; + mul.f32 %r538, %r1860, %r529; + mul.f32 %r539, %r1861, %r531; + mul.f32 %r572, %r1862, %r557; + mul.f32 %r573, %r1863, %r559; + mul.f32 %r574, %r1864, %r561; + mul.f32 %r575, %r1865, %r563; + mul.f32 %r576, %r1866, %r565; + mul.f32 %r577, %r1867, %r567; + mul.f32 %r578, %r1868, %r569; + mul.f32 %r579, %r1869, %r571; + mul.f32 %r612, %r1870, %r597; + mul.f32 %r613, %r1871, %r599; + mul.f32 %r614, %r1872, %r601; + mul.f32 %r615, %r1873, %r603; + mul.f32 %r616, %r1874, %r605; + mul.f32 %r617, %r1875, %r607; + mul.f32 %r618, %r1876, %r609; + mul.f32 %r619, %r1877, %r611; + mul.f32 %r652, %r1878, %r637; + mul.f32 %r653, %r1879, %r639; + mul.f32 %r654, %r1880, %r641; + mul.f32 %r655, %r1881, %r643; + mul.f32 %r656, %r1882, %r645; + mul.f32 %r657, %r1883, %r647; + mul.f32 %r658, %r1884, %r649; + mul.f32 %r659, %r1885, %r651; + mul.f32 %r692, %r1886, %r677; + mul.f32 %r693, %r1887, %r679; + mul.f32 %r694, %r1888, %r681; + mul.f32 %r695, %r1889, %r683; + mul.f32 %r696, %r1890, %r685; + mul.f32 %r697, %r1891, %r687; + mul.f32 %r698, %r1892, %r689; + mul.f32 %r699, %r1893, %r691; + neg.f32 %r732, %r456; + neg.f32 %r733, %r457; + neg.f32 %r734, %r458; + neg.f32 %r735, %r459; + mov.b64 %rd193, {%r739, %r738}; + mov.b64 %rd194, {%r743, %r742}; + fma.rn.f32 %r1907, %r720, %r328, 0f00000000; + fma.rn.f32 %r1908, %r721, %r328, 0f00000000; + fma.rn.f32 %r1909, %r722, %r330, 0f00000000; + fma.rn.f32 %r1910, %r723, %r330, 0f00000000; + fma.rn.f32x2 %rd310, %rd191, %rd165, %rd170; + fma.rn.f32x2 %rd311, %rd192, %rd164, %rd170; + .loc 1 237 34 // chunk_fwd.py:237:34 + sub.s64 %rd50, %rd2, %rd5; + and.b32 %r1906, %r1, 4; + .loc 1 237 22 // chunk_fwd.py:237:22 + setp.gt.s64 %p60, %rd50, 2; + @%p60 bra $L__BB0_16; + bra.uni $L__BB0_19; +$L__BB0_16: // %.lr.ph473 + .loc 1 0 22 // chunk_fwd.py:0:22 + and.b32 %r889, %r31, 24; + bfe.u32 %r890, %r1, 3, 2; + or.b32 %r891, %r890, %r889; + or.b32 %r55, %r891, %r1906; + xor.b32 %r56, %r55, 4; + .loc 1 237 22 // chunk_fwd.py:237:22 + min.u64 %rd51, %rd50, 16; + .loc 1 238 61 // chunk_fwd.py:238:61 + mov.b64 {%r57, %r58}, %rd310; + mov.b64 {%r59, %r60}, %rd311; + mov.b64 %rd309, 2; + setp.eq.b32 %p61, %r1906, 0; + mov.b32 %r1902, %r1910; + mov.b32 %r1903, %r1909; + mov.b32 %r1904, %r1908; + mov.b32 %r1905, %r1907; +$L__BB0_17: // =>This Inner Loop Header: Depth=1 + .loc 1 238 40 // chunk_fwd.py:238:40 + setp.eq.b64 %p62, %rd27, %rd309; + setp.eq.b64 %p63, %rd25, %rd309; + .loc 1 238 61 // chunk_fwd.py:238:61 + selp.f32 %r892, %r57, 0f00000000, %p62; + selp.f32 %r893, %r58, 0f00000000, %p62; + selp.f32 %r894, %r1907, 0f00000000, %p63; + selp.f32 %r895, %r1908, 0f00000000, %p63; + selp.f32 %r896, %r1909, 0f00000000, %p62; + selp.f32 %r897, %r1910, 0f00000000, %p62; + selp.f32 %r898, %r59, 0f00000000, %p63; + selp.f32 %r899, %r60, 0f00000000, %p63; +$L__tmp86: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:238:66 ] ] + add.f32 %r900, %r892, %r894; + add.f32 %r901, %r893, %r895; + add.f32 %r902, %r896, %r898; + add.f32 %r903, %r897, %r899; +$L__tmp87: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:238:66 ] + shfl.sync.bfly.b32 %r904, %r900, 16, 31, -1; +$L__tmp88: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:238:66 ] ] + add.f32 %r905, %r900, %r904; +$L__tmp89: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:238:66 ] + shfl.sync.bfly.b32 %r906, %r905, 8, 31, -1; +$L__tmp90: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:238:66 ] ] + add.f32 %r907, %r905, %r906; +$L__tmp91: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:238:66 ] + shfl.sync.bfly.b32 %r908, %r907, 4, 31, -1; + shfl.sync.bfly.b32 %r909, %r901, 16, 31, -1; +$L__tmp92: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:238:66 ] ] + add.f32 %r910, %r901, %r909; +$L__tmp93: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:238:66 ] + shfl.sync.bfly.b32 %r911, %r910, 8, 31, -1; +$L__tmp94: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:238:66 ] ] + add.f32 %r912, %r910, %r911; +$L__tmp95: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:238:66 ] + shfl.sync.bfly.b32 %r913, %r912, 4, 31, -1; + shfl.sync.bfly.b32 %r914, %r902, 16, 31, -1; +$L__tmp96: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:238:66 ] ] + add.f32 %r915, %r902, %r914; +$L__tmp97: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:238:66 ] + shfl.sync.bfly.b32 %r916, %r915, 8, 31, -1; +$L__tmp98: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:238:66 ] ] + add.f32 %r917, %r915, %r916; +$L__tmp99: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:238:66 ] + shfl.sync.bfly.b32 %r918, %r917, 4, 31, -1; + shfl.sync.bfly.b32 %r919, %r903, 16, 31, -1; +$L__tmp100: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:238:66 ] ] + add.f32 %r920, %r903, %r919; +$L__tmp101: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:238:66 ] + shfl.sync.bfly.b32 %r921, %r920, 8, 31, -1; +$L__tmp102: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:238:66 ] ] + add.f32 %r922, %r920, %r921; +$L__tmp103: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:238:66 ] + shfl.sync.bfly.b32 %r923, %r922, 4, 31, -1; + mov.b64 %rd219, {%r908, %r913}; + mov.b64 %rd220, {%r918, %r923}; +$L__tmp104: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:238:66 ] ] + mov.b64 %rd221, {%r907, %r912}; + mov.b64 %rd222, {%r917, %r922}; + add.f32x2 %rd223, %rd222, %rd220; + add.f32x2 %rd224, %rd221, %rd219; +$L__tmp105: + .loc 1 239 31 // chunk_fwd.py:239:31 + setp.gt.u64 %p64, %rd309, %rd15; + setp.gt.u64 %p65, %rd309, %rd16; + setp.gt.u64 %p66, %rd309, %rd13; + setp.gt.u64 %p67, %rd309, %rd14; + .loc 1 239 41 // chunk_fwd.py:239:41 + mov.b64 {%r924, %r925}, %rd224; + selp.f32 %r926, %r925, 0f00000000, %p67; + selp.f32 %r927, %r924, 0f00000000, %p66; + mov.b64 %rd225, {%r927, %r926}; + mov.b64 {%r928, %r929}, %rd223; + selp.f32 %r930, %r929, 0f00000000, %p65; + selp.f32 %r931, %r928, 0f00000000, %p64; + mov.b64 %rd226, {%r931, %r930}; + .loc 1 240 48 // chunk_fwd.py:240:48 + selp.f32 %r932, %r927, %r926, %p61; + selp.f32 %r933, %r926, %r927, %p61; + selp.f32 %r934, %r931, %r930, %p61; + selp.f32 %r935, %r930, %r931, %p61; + shfl.sync.idx.b32 %r936, %r932, %r55, 31, -1; + shfl.sync.idx.b32 %r937, %r934, %r55, 31, -1; + shfl.sync.idx.b32 %r938, %r933, %r56, 31, -1; + shfl.sync.idx.b32 %r939, %r935, %r56, 31, -1; + selp.b32 %r940, %r936, %r938, %p61; + selp.b32 %r941, %r937, %r939, %p61; + mov.b64 {%r942, %r943}, %rd310; + mul.f32 %r944, %r1905, %r941; + mul.f32 %r945, %r1904, %r941; + mov.b64 {%r946, %r947}, %rd311; + mul.f32 %r948, %r946, %r941; + mul.f32 %r949, %r947, %r941; +$L__tmp106: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:240:56 ] ] + fma.rn.f32 %r950, %r942, %r940, %r944; + fma.rn.f32 %r951, %r943, %r940, %r945; + fma.rn.f32 %r952, %r1903, %r940, %r948; + fma.rn.f32 %r953, %r1902, %r940, %r949; +$L__tmp107: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:240:56 ] + shfl.sync.bfly.b32 %r954, %r950, 16, 31, -1; +$L__tmp108: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:240:56 ] ] + add.f32 %r955, %r950, %r954; +$L__tmp109: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:240:56 ] + shfl.sync.bfly.b32 %r956, %r955, 8, 31, -1; +$L__tmp110: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:240:56 ] ] + add.f32 %r957, %r955, %r956; +$L__tmp111: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:240:56 ] + shfl.sync.bfly.b32 %r958, %r957, 4, 31, -1; + shfl.sync.bfly.b32 %r959, %r951, 16, 31, -1; +$L__tmp112: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:240:56 ] ] + add.f32 %r960, %r951, %r959; +$L__tmp113: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:240:56 ] + shfl.sync.bfly.b32 %r961, %r960, 8, 31, -1; +$L__tmp114: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:240:56 ] ] + add.f32 %r962, %r960, %r961; +$L__tmp115: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:240:56 ] + shfl.sync.bfly.b32 %r963, %r962, 4, 31, -1; + shfl.sync.bfly.b32 %r964, %r952, 16, 31, -1; +$L__tmp116: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:240:56 ] ] + add.f32 %r965, %r952, %r964; +$L__tmp117: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:240:56 ] + shfl.sync.bfly.b32 %r966, %r965, 8, 31, -1; +$L__tmp118: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:240:56 ] ] + add.f32 %r967, %r965, %r966; +$L__tmp119: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:240:56 ] + shfl.sync.bfly.b32 %r968, %r967, 4, 31, -1; + shfl.sync.bfly.b32 %r969, %r953, 16, 31, -1; +$L__tmp120: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:240:56 ] ] + add.f32 %r970, %r953, %r969; +$L__tmp121: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:240:56 ] + shfl.sync.bfly.b32 %r971, %r970, 8, 31, -1; +$L__tmp122: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:240:56 ] ] + add.f32 %r972, %r970, %r971; +$L__tmp123: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:240:56 ] + shfl.sync.bfly.b32 %r973, %r972, 4, 31, -1; + mov.b64 %rd227, {%r968, %r973}; + mov.b64 %rd228, {%r958, %r963}; +$L__tmp124: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:240:56 ] ] + mov.b64 %rd229, {%r967, %r972}; + mov.b64 %rd230, {%r957, %r962}; + add.f32x2 %rd231, %rd230, %rd228; + add.f32x2 %rd232, %rd229, %rd227; +$L__tmp125: + .loc 1 240 24 // chunk_fwd.py:240:24 + add.f32x2 %rd233, %rd226, %rd232; + add.f32x2 %rd234, %rd225, %rd231; + .loc 1 241 54 // chunk_fwd.py:241:54 + mov.b64 {%r974, %r975}, %rd234; + selp.f32 %r1905, %r974, %r1905, %p63; + selp.f32 %r1904, %r975, %r1904, %p63; + mov.b64 {%r976, %r977}, %rd233; + selp.f32 %r1903, %r976, %r1903, %p62; + selp.f32 %r1902, %r977, %r1902, %p62; + selp.f32 %r978, %r975, %r943, %p62; + selp.f32 %r979, %r974, %r942, %p62; + mov.b64 %rd310, {%r979, %r978}; + selp.f32 %r980, %r977, %r947, %p63; + selp.f32 %r981, %r976, %r946, %p63; + mov.b64 %rd311, {%r981, %r980}; + .loc 1 237 22 // chunk_fwd.py:237:22 + add.s64 %rd309, %rd309, 1; + setp.ne.b64 %p68, %rd51, %rd309; + @%p68 bra $L__BB0_17; +// %bb.18: // %._crit_edge474.loopexit + .loc 1 250 14 // chunk_fwd.py:250:14 + add.f32 %r1907, %r1905, 0f00000000; + add.f32 %r1908, %r1904, 0f00000000; + add.f32 %r1909, %r1903, 0f00000000; + add.f32 %r1910, %r1902, 0f00000000; +$L__BB0_19: // %._crit_edge474 + .loc 1 0 14 // chunk_fwd.py:0:14 + add.s64 %rd45, %rd57, %rd159; + and.b32 %r32, %r1, 31; + shr.u32 %r33, %r1, 5; + shl.b32 %r34, %r3, 1; + mul.f32 %r35, %r492, %r324; + mul.f32 %r36, %r493, %r324; + mul.f32 %r37, %r494, %r322; + mul.f32 %r38, %r495, %r322; + mul.f32 %r39, %r496, %r324; + mul.f32 %r40, %r497, %r324; + mul.f32 %r41, %r498, %r322; + mul.f32 %r42, %r499, %r322; + mul.f32 %r1211, %r532, %r330; + mul.f32 %r1223, %r533, %r330; + mul.f32 %r1212, %r534, %r328; + mul.f32 %r1224, %r535, %r328; + mul.f32 %r1213, %r536, %r330; + mul.f32 %r1225, %r537, %r330; + mul.f32 %r1214, %r538, %r328; + mul.f32 %r1226, %r539, %r328; + mul.f32 %r1235, %r572, %r330; + mul.f32 %r1243, %r573, %r330; + mul.f32 %r1236, %r574, %r328; + mul.f32 %r1244, %r575, %r328; + mul.f32 %r1237, %r576, %r330; + mul.f32 %r1245, %r577, %r330; + mul.f32 %r1238, %r578, %r328; + mul.f32 %r1246, %r579, %r328; + mul.f32 %r1339, %r612, %r336; + mul.f32 %r1347, %r613, %r336; + mul.f32 %r1340, %r614, %r334; + mul.f32 %r1348, %r615, %r334; + mul.f32 %r1341, %r616, %r336; + mul.f32 %r1349, %r617, %r336; + mul.f32 %r1342, %r618, %r334; + mul.f32 %r1350, %r619, %r334; + mul.f32 %r1275, %r652, %r336; + mul.f32 %r1287, %r653, %r336; + mul.f32 %r1276, %r654, %r334; + mul.f32 %r1288, %r655, %r334; + mul.f32 %r1277, %r656, %r336; + mul.f32 %r1289, %r657, %r336; + mul.f32 %r1278, %r658, %r334; + mul.f32 %r1290, %r659, %r334; + mul.f32 %r1299, %r692, %r336; + mul.f32 %r1307, %r693, %r336; + mul.f32 %r1300, %r694, %r334; + mul.f32 %r1308, %r695, %r334; + mul.f32 %r1301, %r696, %r336; + mul.f32 %r1309, %r697, %r336; + mul.f32 %r1302, %r698, %r334; + mul.f32 %r1310, %r699, %r334; + fma.rn.f32 %r1917, %r732, %r334, 0f00000000; + fma.rn.f32 %r1918, %r733, %r334, 0f00000000; + fma.rn.f32 %r1919, %r734, %r336, 0f00000000; + fma.rn.f32 %r1920, %r735, %r336, 0f00000000; + fma.rn.f32x2 %rd313, %rd193, %rd167, %rd170; + fma.rn.f32x2 %rd314, %rd194, %rd166, %rd170; + .loc 1 242 34 // chunk_fwd.py:242:34 + sub.s64 %rd52, %rd2, %rd6; + .loc 1 242 22 // chunk_fwd.py:242:22 + setp.gt.s64 %p69, %rd52, 2; + setp.eq.b32 %p90, %r1906, 0; + and.b32 %r1916, %r31, 24; + bfe.u32 %r1915, %r1, 3, 2; + @%p69 bra $L__BB0_20; + bra.uni $L__BB0_23; +$L__BB0_20: // %.lr.ph492 + .loc 1 0 22 // chunk_fwd.py:0:22 + or.b32 %r982, %r1915, %r1916; + or.b32 %r61, %r982, %r1906; + xor.b32 %r62, %r61, 4; + .loc 1 242 22 // chunk_fwd.py:242:22 + min.u64 %rd53, %rd52, 16; + .loc 1 243 61 // chunk_fwd.py:243:61 + mov.b64 {%r63, %r64}, %rd313; + mov.b64 {%r65, %r66}, %rd314; + mov.b64 %rd312, 2; + mov.b32 %r1911, %r1920; + mov.b32 %r1912, %r1919; + mov.b32 %r1913, %r1918; + mov.b32 %r1914, %r1917; +$L__BB0_21: // =>This Inner Loop Header: Depth=1 + .loc 1 243 40 // chunk_fwd.py:243:40 + setp.eq.b64 %p70, %rd27, %rd312; + setp.eq.b64 %p71, %rd25, %rd312; + .loc 1 243 61 // chunk_fwd.py:243:61 + selp.f32 %r983, %r63, 0f00000000, %p70; + selp.f32 %r984, %r64, 0f00000000, %p70; + selp.f32 %r985, %r1917, 0f00000000, %p71; + selp.f32 %r986, %r1918, 0f00000000, %p71; + selp.f32 %r987, %r1919, 0f00000000, %p70; + selp.f32 %r988, %r1920, 0f00000000, %p70; + selp.f32 %r989, %r65, 0f00000000, %p71; + selp.f32 %r990, %r66, 0f00000000, %p71; +$L__tmp126: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:243:66 ] ] + add.f32 %r991, %r983, %r985; + add.f32 %r992, %r984, %r986; + add.f32 %r993, %r987, %r989; + add.f32 %r994, %r988, %r990; +$L__tmp127: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:243:66 ] + shfl.sync.bfly.b32 %r995, %r991, 16, 31, -1; +$L__tmp128: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:243:66 ] ] + add.f32 %r996, %r991, %r995; +$L__tmp129: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:243:66 ] + shfl.sync.bfly.b32 %r997, %r996, 8, 31, -1; +$L__tmp130: + .loc 1 244 31 // chunk_fwd.py:244:31 + setp.gt.u64 %p72, %rd312, %rd15; + setp.gt.u64 %p73, %rd312, %rd16; + setp.gt.u64 %p74, %rd312, %rd13; + setp.gt.u64 %p75, %rd312, %rd14; +$L__tmp131: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:243:66 ] ] + add.f32 %r998, %r996, %r997; +$L__tmp132: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:243:66 ] + shfl.sync.bfly.b32 %r999, %r998, 4, 31, -1; + shfl.sync.bfly.b32 %r1000, %r992, 16, 31, -1; +$L__tmp133: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:243:66 ] ] + add.f32 %r1001, %r992, %r1000; +$L__tmp134: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:243:66 ] + shfl.sync.bfly.b32 %r1002, %r1001, 8, 31, -1; +$L__tmp135: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:243:66 ] ] + add.f32 %r1003, %r1001, %r1002; +$L__tmp136: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:243:66 ] + shfl.sync.bfly.b32 %r1004, %r1003, 4, 31, -1; + shfl.sync.bfly.b32 %r1005, %r993, 16, 31, -1; +$L__tmp137: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:243:66 ] ] + add.f32 %r1006, %r993, %r1005; +$L__tmp138: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:243:66 ] + shfl.sync.bfly.b32 %r1007, %r1006, 8, 31, -1; +$L__tmp139: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:243:66 ] ] + add.f32 %r1008, %r1006, %r1007; +$L__tmp140: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:243:66 ] + shfl.sync.bfly.b32 %r1009, %r1008, 4, 31, -1; + shfl.sync.bfly.b32 %r1010, %r994, 16, 31, -1; +$L__tmp141: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:243:66 ] ] + add.f32 %r1011, %r994, %r1010; +$L__tmp142: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:243:66 ] + shfl.sync.bfly.b32 %r1012, %r1011, 8, 31, -1; +$L__tmp143: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:243:66 ] ] + add.f32 %r1013, %r1011, %r1012; +$L__tmp144: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:243:66 ] + shfl.sync.bfly.b32 %r1014, %r1013, 4, 31, -1; + mov.b64 %rd235, {%r999, %r1004}; + mov.b64 %rd236, {%r1009, %r1014}; +$L__tmp145: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:243:66 ] ] + mov.b64 %rd237, {%r998, %r1003}; + mov.b64 %rd238, {%r1008, %r1013}; + add.f32x2 %rd239, %rd238, %rd236; + add.f32x2 %rd240, %rd237, %rd235; +$L__tmp146: + .loc 1 244 41 // chunk_fwd.py:244:41 + mov.b64 {%r1015, %r1016}, %rd240; + selp.f32 %r1017, %r1016, 0f00000000, %p75; + selp.f32 %r1018, %r1015, 0f00000000, %p74; + mov.b64 %rd241, {%r1018, %r1017}; + mov.b64 {%r1019, %r1020}, %rd239; + selp.f32 %r1021, %r1020, 0f00000000, %p73; + selp.f32 %r1022, %r1019, 0f00000000, %p72; + mov.b64 %rd242, {%r1022, %r1021}; + .loc 1 245 48 // chunk_fwd.py:245:48 + selp.f32 %r1023, %r1018, %r1017, %p90; + selp.f32 %r1024, %r1017, %r1018, %p90; + selp.f32 %r1025, %r1022, %r1021, %p90; + selp.f32 %r1026, %r1021, %r1022, %p90; + shfl.sync.idx.b32 %r1027, %r1023, %r61, 31, -1; + shfl.sync.idx.b32 %r1028, %r1025, %r61, 31, -1; + shfl.sync.idx.b32 %r1029, %r1024, %r62, 31, -1; + shfl.sync.idx.b32 %r1030, %r1026, %r62, 31, -1; + selp.b32 %r1031, %r1027, %r1029, %p90; + selp.b32 %r1032, %r1028, %r1030, %p90; + mov.b64 {%r1033, %r1034}, %rd313; + mul.f32 %r1035, %r1914, %r1032; + mul.f32 %r1036, %r1913, %r1032; + mov.b64 {%r1037, %r1038}, %rd314; + mul.f32 %r1039, %r1037, %r1032; + mul.f32 %r1040, %r1038, %r1032; +$L__tmp147: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:245:56 ] ] + fma.rn.f32 %r1041, %r1033, %r1031, %r1035; + fma.rn.f32 %r1042, %r1034, %r1031, %r1036; + fma.rn.f32 %r1043, %r1912, %r1031, %r1039; + fma.rn.f32 %r1044, %r1911, %r1031, %r1040; +$L__tmp148: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:245:56 ] + shfl.sync.bfly.b32 %r1045, %r1041, 16, 31, -1; +$L__tmp149: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:245:56 ] ] + add.f32 %r1046, %r1041, %r1045; +$L__tmp150: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:245:56 ] + shfl.sync.bfly.b32 %r1047, %r1046, 8, 31, -1; +$L__tmp151: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:245:56 ] ] + add.f32 %r1048, %r1046, %r1047; +$L__tmp152: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:245:56 ] + shfl.sync.bfly.b32 %r1049, %r1048, 4, 31, -1; + shfl.sync.bfly.b32 %r1050, %r1042, 16, 31, -1; +$L__tmp153: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:245:56 ] ] + add.f32 %r1051, %r1042, %r1050; +$L__tmp154: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:245:56 ] + shfl.sync.bfly.b32 %r1052, %r1051, 8, 31, -1; +$L__tmp155: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:245:56 ] ] + add.f32 %r1053, %r1051, %r1052; +$L__tmp156: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:245:56 ] + shfl.sync.bfly.b32 %r1054, %r1053, 4, 31, -1; + shfl.sync.bfly.b32 %r1055, %r1043, 16, 31, -1; +$L__tmp157: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:245:56 ] ] + add.f32 %r1056, %r1043, %r1055; +$L__tmp158: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:245:56 ] + shfl.sync.bfly.b32 %r1057, %r1056, 8, 31, -1; +$L__tmp159: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:245:56 ] ] + add.f32 %r1058, %r1056, %r1057; +$L__tmp160: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:245:56 ] + shfl.sync.bfly.b32 %r1059, %r1058, 4, 31, -1; + shfl.sync.bfly.b32 %r1060, %r1044, 16, 31, -1; +$L__tmp161: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:245:56 ] ] + add.f32 %r1061, %r1044, %r1060; +$L__tmp162: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:245:56 ] + shfl.sync.bfly.b32 %r1062, %r1061, 8, 31, -1; +$L__tmp163: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:245:56 ] ] + add.f32 %r1063, %r1061, %r1062; +$L__tmp164: + .loc 3 293 36 // standard.py:293:36 @[ chunk_fwd.py:245:56 ] + shfl.sync.bfly.b32 %r1064, %r1063, 4, 31, -1; + mov.b64 %rd243, {%r1059, %r1064}; + mov.b64 %rd244, {%r1049, %r1054}; +$L__tmp165: + .loc 3 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ chunk_fwd.py:245:56 ] ] + mov.b64 %rd245, {%r1058, %r1063}; + mov.b64 %rd246, {%r1048, %r1053}; + add.f32x2 %rd247, %rd246, %rd244; + add.f32x2 %rd248, %rd245, %rd243; +$L__tmp166: + .loc 1 245 24 // chunk_fwd.py:245:24 + add.f32x2 %rd249, %rd242, %rd248; + add.f32x2 %rd250, %rd241, %rd247; + .loc 1 246 54 // chunk_fwd.py:246:54 + mov.b64 {%r1065, %r1066}, %rd250; + selp.f32 %r1914, %r1065, %r1914, %p71; + selp.f32 %r1913, %r1066, %r1913, %p71; + mov.b64 {%r1067, %r1068}, %rd249; + selp.f32 %r1912, %r1067, %r1912, %p70; + selp.f32 %r1911, %r1068, %r1911, %p70; + selp.f32 %r1069, %r1066, %r1034, %p70; + selp.f32 %r1070, %r1065, %r1033, %p70; + mov.b64 %rd313, {%r1070, %r1069}; + selp.f32 %r1071, %r1068, %r1038, %p71; + selp.f32 %r1072, %r1067, %r1037, %p71; + mov.b64 %rd314, {%r1072, %r1071}; + .loc 1 242 22 // chunk_fwd.py:242:22 + add.s64 %rd312, %rd312, 1; + setp.ne.b64 %p76, %rd53, %rd312; + @%p76 bra $L__BB0_21; +// %bb.22: // %._crit_edge493.loopexit + .loc 1 251 14 // chunk_fwd.py:251:14 + add.f32 %r1917, %r1914, 0f00000000; + add.f32 %r1918, %r1913, 0f00000000; + add.f32 %r1919, %r1912, 0f00000000; + add.f32 %r1920, %r1911, 0f00000000; +$L__BB0_23: // %._crit_edge493 + .loc 1 180 26 // chunk_fwd.py:180:26 + setp.eq.b32 %p80, %r12, %r8; + setp.eq.b32 %p81, %r12, %r7; + setp.eq.b32 %p82, %r11, %r8; + setp.eq.b32 %p83, %r11, %r7; + setp.eq.b32 %p84, %r12, %r10; + setp.eq.b32 %p85, %r11, %r9; + setp.eq.b32 %p86, %r11, %r6; + .loc 1 95 28 // chunk_fwd.py:95:28 + setp.lt.u64 %p79, %rd11, %rd2; + .loc 1 94 28 // chunk_fwd.py:94:28 + setp.lt.u64 %p78, %rd10, %rd2; + .loc 1 93 28 // chunk_fwd.py:93:28 + setp.lt.u64 %p77, %rd9, %rd2; + .loc 1 248 14 // chunk_fwd.py:248:14 + selp.f32 %r1389, 0f3F800000, 0f00000000, %p86; + selp.f32 %r1390, 0f3F800000, 0f00000000, %p85; + selp.f32 %r1391, 0f3F800000, 0f00000000, %p84; + selp.f32 %r1392, 0f3F800000, 0f00000000, %p83; + selp.f32 %r1393, 0f3F800000, 0f00000000, %p82; + selp.f32 %r1394, 0f3F800000, 0f00000000, %p81; + selp.f32 %r1395, 0f3F800000, 0f00000000, %p80; + mov.b64 {%r1396, %r1397}, %rd304; + add.f32 %r1074, %r1396, %r1392; + add.f32 %r1075, %r1397, %r1393; + mov.b64 {%r1398, %r1399}, %rd305; + add.f32 %r1077, %r1398, %r1394; + add.f32 %r1078, %r1399, %r1395; + bar.sync 0; + and.b32 %r1400, %r13, 320; + shl.b32 %r1401, %r4, 3; + bfe.s32 %r1402, %r1, 3, 1; + and.b32 %r1403, %r1402, 160; + xor.b32 %r1404, %r1403, %r5; + add.s32 %r1405, %r109, %r1400; + add.s32 %r1406, %r1405, %r1401; + add.s32 %r1073, %r1406, %r1404; + // begin inline asm + st.shared.v2.b32 [ %r1073 + 0 ], { %r1074, %r1075 }; + // end inline asm + add.s32 %r1076, %r1073, 512; + // begin inline asm + st.shared.v2.b32 [ %r1076 + 0 ], { %r1077, %r1078 }; + // end inline asm + .loc 1 249 14 // chunk_fwd.py:249:14 + mov.b64 {%r1407, %r1408}, %rd307; + add.f32 %r1081, %r1407, %r1389; + add.f32 %r1095, %r1408, %r1390; + mov.b64 {%r1409, %r1410}, %rd308; + add.f32 %r1082, %r1409, %r1389; + add.f32 %r1096, %r1410, %r1391; + .loc 1 250 14 // chunk_fwd.py:250:14 + mov.b64 {%r1411, %r1412}, %rd310; + add.f32 %r1125, %r1411, %r1389; + add.f32 %r1139, %r1412, %r1390; + mov.b64 {%r1413, %r1414}, %rd311; + add.f32 %r1126, %r1413, %r1389; + add.f32 %r1140, %r1414, %r1391; + .loc 1 251 14 // chunk_fwd.py:251:14 + mov.b64 {%r1415, %r1416}, %rd313; + add.f32 %r1169, %r1415, %r1389; + add.f32 %r1183, %r1416, %r1390; + mov.b64 {%r1417, %r1418}, %rd314; + add.f32 %r1170, %r1417, %r1389; + add.f32 %r1184, %r1418, %r1391; + .loc 1 207 20 // chunk_fwd.py:207:20 + selp.f32 %r1419, %r35, %r36, %p90; + selp.f32 %r1420, %r36, %r35, %p90; + selp.f32 %r1421, %r37, %r38, %p90; + selp.f32 %r1422, %r38, %r37, %p90; + selp.f32 %r1423, %r39, %r40, %p90; + selp.f32 %r1424, %r40, %r39, %p90; + selp.f32 %r1425, %r41, %r42, %p90; + selp.f32 %r1426, %r42, %r41, %p90; + or.b32 %r1427, %r1906, %r1916; + or.b32 %r1428, %r1427, %r1915; + shfl.sync.idx.b32 %r1429, %r1419, %r1428, 31, -1; + xor.b32 %r1430, %r1428, 4; + shfl.sync.idx.b32 %r1431, %r1420, %r1430, 31, -1; + shfl.sync.idx.b32 %r1432, %r1421, %r1428, 31, -1; + shfl.sync.idx.b32 %r1433, %r1422, %r1430, 31, -1; + shfl.sync.idx.b32 %r1434, %r1423, %r1428, 31, -1; + shfl.sync.idx.b32 %r1435, %r1424, %r1430, 31, -1; + shfl.sync.idx.b32 %r1436, %r1425, %r1428, 31, -1; + shfl.sync.idx.b32 %r1437, %r1426, %r1430, 31, -1; + selp.b32 %r1089, %r1431, %r1429, %p90; + selp.b32 %r1090, %r1433, %r1432, %p90; + selp.b32 %r1097, %r1435, %r1434, %p90; + selp.b32 %r1098, %r1437, %r1436, %p90; + selp.b32 %r1079, %r1429, %r1431, %p90; + selp.b32 %r1080, %r1432, %r1433, %p90; + selp.b32 %r1083, %r1434, %r1435, %p90; + selp.b32 %r1084, %r1436, %r1437, %p90; + mov.b32 %r1373, 0; + .loc 1 258 23 // chunk_fwd.py:258:23 + mov.b32 %r1085, %r1373; + mov.b32 %r1086, %r1373; + mov.b32 %r1087, %r1373; + mov.b32 %r1088, %r1373; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1085, %r1086, %r1087, %r1088 }, { %r1081, %r1898, %r1900, %r1082 }, { %r1079, %r1080 }, { %r1085, %r1086, %r1087, %r1088 }; + // end inline asm + mov.b32 %r1091, %r1373; + mov.b32 %r1092, %r1373; + mov.b32 %r1093, %r1373; + mov.b32 %r1094, %r1373; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1091, %r1092, %r1093, %r1094 }, { %r1081, %r1898, %r1900, %r1082 }, { %r1083, %r1084 }, { %r1091, %r1092, %r1093, %r1094 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1085, %r1086, %r1087, %r1088 }, { %r1095, %r1899, %r1901, %r1096 }, { %r1089, %r1090 }, { %r1085, %r1086, %r1087, %r1088 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1091, %r1092, %r1093, %r1094 }, { %r1095, %r1899, %r1901, %r1096 }, { %r1097, %r1098 }, { %r1091, %r1092, %r1093, %r1094 }; + // end inline asm + bfe.s32 %r1438, %r1, 1, 1; + and.b32 %r1439, %r1, 2; + setp.eq.b32 %p87, %r1439, 0; + selp.f32 %r1440, %r1085, %r1086, %p87; + selp.f32 %r1441, %r1086, %r1085, %p87; + selp.f32 %r1442, %r1087, %r1088, %p87; + selp.f32 %r1443, %r1088, %r1087, %p87; + selp.f32 %r1444, %r1091, %r1092, %p87; + selp.f32 %r1445, %r1092, %r1091, %p87; + selp.f32 %r1446, %r1093, %r1094, %p87; + selp.f32 %r1447, %r1094, %r1093, %p87; + and.b32 %r1448, %r1, 1; + neg.s32 %r1449, %r1448; + setp.eq.b32 %p88, %r1448, 0; + cvt.u16.u32 %rs26, %r1; + and.b16 %rs27, %rs26, 2; + shl.b16 %rs28, %rs26, 1; + shr.u16 %rs29, %rs27, 1; + or.b16 %rs30, %rs28, %rs29; + and.b16 %rs31, %rs30, 3; + cvt.u32.u16 %r1450, %rs31; + or.b32 %r1451, %r2, %r1450; + shfl.sync.idx.b32 %r1452, %r1440, %r1451, 31, -1; + shfl.sync.idx.b32 %r1453, %r1442, %r1451, 31, -1; + xor.b32 %r1454, %r1451, 2; + shfl.sync.idx.b32 %r1455, %r1441, %r1454, 31, -1; + shfl.sync.idx.b32 %r1456, %r1443, %r1454, 31, -1; + shfl.sync.idx.b32 %r1457, %r1444, %r1451, 31, -1; + shfl.sync.idx.b32 %r1458, %r1446, %r1451, 31, -1; + shfl.sync.idx.b32 %r1459, %r1445, %r1454, 31, -1; + shfl.sync.idx.b32 %r1460, %r1447, %r1454, 31, -1; + selp.b32 %r1101, %r1455, %r1452, %p88; + selp.b32 %r1102, %r1456, %r1453, %p88; + selp.b32 %r1113, %r1459, %r1457, %p88; + selp.b32 %r1114, %r1460, %r1458, %p88; + .loc 1 248 14 // chunk_fwd.py:248:14 + bar.sync 0; + shl.b32 %r1461, %r1448, 6; + and.b32 %r1462, %r1438, 160; + or.b32 %r1463, %r1462, %r1461; + or.b32 %r1464, %r1463, %r2; + add.s32 %r1465, %r109, %r1464; + ld.shared.b32 %r1103, [%r1465]; + ld.shared.b32 %r1104, [%r1465+256]; + ld.shared.b32 %r1115, [%r1465+512]; + ld.shared.b32 %r1116, [%r1465+768]; + xor.b32 %r1466, %r1464, 32; + add.s32 %r1467, %r109, %r1466; + ld.shared.b32 %r1105, [%r1467]; + ld.shared.b32 %r1106, [%r1467+256]; + ld.shared.b32 %r1121, [%r1467+512]; + ld.shared.b32 %r1122, [%r1467+768]; + .loc 1 258 23 // chunk_fwd.py:258:23 + selp.b32 %r1099, %r1452, %r1455, %p88; + selp.b32 %r1100, %r1453, %r1456, %p88; + selp.b32 %r1111, %r1457, %r1459, %p88; + selp.b32 %r1112, %r1458, %r1460, %p88; + .loc 1 259 8 // chunk_fwd.py:259:8 + mov.b32 %r1107, %r1373; + mov.b32 %r1108, %r1373; + mov.b32 %r1109, %r1373; + mov.b32 %r1110, %r1373; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1107, %r1108, %r1109, %r1110 }, { %r1099, %r1100, %r1101, %r1102 }, { %r1103, %r1104 }, { %r1107, %r1108, %r1109, %r1110 }; + // end inline asm + mov.b32 %r1117, %r1373; + mov.b32 %r1118, %r1373; + mov.b32 %r1119, %r1373; + mov.b32 %r1120, %r1373; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1117, %r1118, %r1119, %r1120 }, { %r1099, %r1100, %r1101, %r1102 }, { %r1105, %r1106 }, { %r1117, %r1118, %r1119, %r1120 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1107, %r1108, %r1109, %r1110 }, { %r1111, %r1112, %r1113, %r1114 }, { %r1115, %r1116 }, { %r1107, %r1108, %r1109, %r1110 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1117, %r1118, %r1119, %r1120 }, { %r1111, %r1112, %r1113, %r1114 }, { %r1121, %r1122 }, { %r1117, %r1118, %r1119, %r1120 }; + // end inline asm + mov.b32 %r1468, 0f00000000; + .loc 1 257 14 // chunk_fwd.py:257:14 + sub.f32 %r1469, %r1468, %r1107; + sub.f32 %r1470, %r1468, %r1108; + sub.f32 %r1471, %r1468, %r1109; + sub.f32 %r1472, %r1468, %r1110; + sub.f32 %r1473, %r1468, %r1117; + sub.f32 %r1474, %r1468, %r1118; + sub.f32 %r1475, %r1468, %r1119; + sub.f32 %r1476, %r1468, %r1120; + .loc 1 209 20 // chunk_fwd.py:209:20 + selp.f32 %r1477, %r1235, %r1243, %p90; + selp.f32 %r1478, %r1243, %r1235, %p90; + selp.f32 %r1479, %r1236, %r1244, %p90; + selp.f32 %r1480, %r1244, %r1236, %p90; + selp.f32 %r1481, %r1237, %r1245, %p90; + selp.f32 %r1482, %r1245, %r1237, %p90; + selp.f32 %r1483, %r1238, %r1246, %p90; + selp.f32 %r1484, %r1246, %r1238, %p90; + shfl.sync.idx.b32 %r1485, %r1477, %r1428, 31, -1; + shfl.sync.idx.b32 %r1486, %r1478, %r1430, 31, -1; + shfl.sync.idx.b32 %r1487, %r1479, %r1428, 31, -1; + shfl.sync.idx.b32 %r1488, %r1480, %r1430, 31, -1; + shfl.sync.idx.b32 %r1489, %r1481, %r1428, 31, -1; + shfl.sync.idx.b32 %r1490, %r1482, %r1430, 31, -1; + shfl.sync.idx.b32 %r1491, %r1483, %r1428, 31, -1; + shfl.sync.idx.b32 %r1492, %r1484, %r1430, 31, -1; + selp.b32 %r1133, %r1486, %r1485, %p90; + selp.b32 %r1134, %r1488, %r1487, %p90; + selp.b32 %r1141, %r1490, %r1489, %p90; + selp.b32 %r1142, %r1492, %r1491, %p90; + selp.b32 %r1123, %r1485, %r1486, %p90; + selp.b32 %r1124, %r1487, %r1488, %p90; + selp.b32 %r1127, %r1489, %r1490, %p90; + selp.b32 %r1128, %r1491, %r1492, %p90; + .loc 1 263 23 // chunk_fwd.py:263:23 + mov.b32 %r1129, %r1373; + mov.b32 %r1130, %r1373; + mov.b32 %r1131, %r1373; + mov.b32 %r1132, %r1373; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1129, %r1130, %r1131, %r1132 }, { %r1125, %r1907, %r1909, %r1126 }, { %r1123, %r1124 }, { %r1129, %r1130, %r1131, %r1132 }; + // end inline asm + mov.b32 %r1135, %r1373; + mov.b32 %r1136, %r1373; + mov.b32 %r1137, %r1373; + mov.b32 %r1138, %r1373; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1135, %r1136, %r1137, %r1138 }, { %r1125, %r1907, %r1909, %r1126 }, { %r1127, %r1128 }, { %r1135, %r1136, %r1137, %r1138 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1129, %r1130, %r1131, %r1132 }, { %r1139, %r1908, %r1910, %r1140 }, { %r1133, %r1134 }, { %r1129, %r1130, %r1131, %r1132 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1135, %r1136, %r1137, %r1138 }, { %r1139, %r1908, %r1910, %r1140 }, { %r1141, %r1142 }, { %r1135, %r1136, %r1137, %r1138 }; + // end inline asm + selp.f32 %r1493, %r1129, %r1130, %p87; + selp.f32 %r1494, %r1130, %r1129, %p87; + selp.f32 %r1495, %r1131, %r1132, %p87; + selp.f32 %r1496, %r1132, %r1131, %p87; + selp.f32 %r1497, %r1135, %r1136, %p87; + selp.f32 %r1498, %r1136, %r1135, %p87; + selp.f32 %r1499, %r1137, %r1138, %p87; + selp.f32 %r1500, %r1138, %r1137, %p87; + shfl.sync.idx.b32 %r1501, %r1493, %r1451, 31, -1; + shfl.sync.idx.b32 %r1502, %r1495, %r1451, 31, -1; + shfl.sync.idx.b32 %r1503, %r1494, %r1454, 31, -1; + shfl.sync.idx.b32 %r1504, %r1496, %r1454, 31, -1; + shfl.sync.idx.b32 %r1505, %r1497, %r1451, 31, -1; + shfl.sync.idx.b32 %r1506, %r1499, %r1451, 31, -1; + shfl.sync.idx.b32 %r1507, %r1498, %r1454, 31, -1; + shfl.sync.idx.b32 %r1508, %r1500, %r1454, 31, -1; + selp.b32 %r1145, %r1503, %r1501, %p88; + selp.b32 %r1146, %r1504, %r1502, %p88; + selp.b32 %r1157, %r1507, %r1505, %p88; + selp.b32 %r1158, %r1508, %r1506, %p88; + .loc 1 249 14 // chunk_fwd.py:249:14 + setp.eq.b32 %p89, %r18, 0; + selp.f32 %r1509, %r1081, %r1095, %p89; + selp.f32 %r1510, %r1095, %r1081, %p89; + selp.f32 %r1511, %r1898, %r1899, %p89; + selp.f32 %r1512, %r1899, %r1898, %p89; + selp.f32 %r1513, %r1900, %r1901, %p89; + selp.f32 %r1514, %r1901, %r1900, %p89; + selp.f32 %r1515, %r1082, %r1096, %p89; + selp.f32 %r1516, %r1096, %r1082, %p89; + shl.b32 %r1517, %r1, 2; + and.b32 %r1518, %r1517, 28; + or.b32 %r1519, %r1518, %r1915; + shfl.sync.idx.b32 %r1520, %r1509, %r1519, 31, -1; + xor.b32 %r1521, %r1519, 16; + shfl.sync.idx.b32 %r1522, %r1510, %r1521, 31, -1; + shfl.sync.idx.b32 %r1523, %r1511, %r1519, 31, -1; + shfl.sync.idx.b32 %r1524, %r1512, %r1521, 31, -1; + shfl.sync.idx.b32 %r1525, %r1513, %r1519, 31, -1; + shfl.sync.idx.b32 %r1526, %r1514, %r1521, 31, -1; + shfl.sync.idx.b32 %r1527, %r1515, %r1519, 31, -1; + shfl.sync.idx.b32 %r1528, %r1516, %r1521, 31, -1; + selp.b32 %r1148, %r1522, %r1520, %p90; + selp.b32 %r1160, %r1524, %r1523, %p90; + selp.b32 %r1150, %r1526, %r1525, %p90; + selp.b32 %r1166, %r1528, %r1527, %p90; + .loc 1 263 23 // chunk_fwd.py:263:23 + selp.b32 %r1143, %r1501, %r1503, %p88; + selp.b32 %r1144, %r1502, %r1504, %p88; + selp.b32 %r1155, %r1505, %r1507, %p88; + selp.b32 %r1156, %r1506, %r1508, %p88; + .loc 1 249 14 // chunk_fwd.py:249:14 + selp.b32 %r1147, %r1520, %r1522, %p90; + selp.b32 %r1159, %r1523, %r1524, %p90; + selp.b32 %r1149, %r1525, %r1526, %p90; + selp.b32 %r1165, %r1527, %r1528, %p90; + .loc 1 264 8 // chunk_fwd.py:264:8 + mov.b32 %r1151, %r1373; + mov.b32 %r1152, %r1373; + mov.b32 %r1153, %r1373; + mov.b32 %r1154, %r1373; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1151, %r1152, %r1153, %r1154 }, { %r1143, %r1144, %r1145, %r1146 }, { %r1147, %r1148 }, { %r1151, %r1152, %r1153, %r1154 }; + // end inline asm + mov.b32 %r1161, %r1373; + mov.b32 %r1162, %r1373; + mov.b32 %r1163, %r1373; + mov.b32 %r1164, %r1373; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1161, %r1162, %r1163, %r1164 }, { %r1143, %r1144, %r1145, %r1146 }, { %r1149, %r1150 }, { %r1161, %r1162, %r1163, %r1164 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1151, %r1152, %r1153, %r1154 }, { %r1155, %r1156, %r1157, %r1158 }, { %r1159, %r1160 }, { %r1151, %r1152, %r1153, %r1154 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1161, %r1162, %r1163, %r1164 }, { %r1155, %r1156, %r1157, %r1158 }, { %r1165, %r1166 }, { %r1161, %r1162, %r1163, %r1164 }; + // end inline asm + .loc 1 262 14 // chunk_fwd.py:262:14 + sub.f32 %r1529, %r1468, %r1151; + sub.f32 %r1530, %r1468, %r1152; + sub.f32 %r1531, %r1468, %r1153; + sub.f32 %r1532, %r1468, %r1154; + sub.f32 %r1533, %r1468, %r1161; + sub.f32 %r1534, %r1468, %r1162; + sub.f32 %r1535, %r1468, %r1163; + sub.f32 %r1536, %r1468, %r1164; + .loc 1 212 20 // chunk_fwd.py:212:20 + selp.f32 %r1537, %r1299, %r1307, %p90; + selp.f32 %r1538, %r1307, %r1299, %p90; + selp.f32 %r1539, %r1300, %r1308, %p90; + selp.f32 %r1540, %r1308, %r1300, %p90; + selp.f32 %r1541, %r1301, %r1309, %p90; + selp.f32 %r1542, %r1309, %r1301, %p90; + selp.f32 %r1543, %r1302, %r1310, %p90; + selp.f32 %r1544, %r1310, %r1302, %p90; + shfl.sync.idx.b32 %r1545, %r1537, %r1428, 31, -1; + shfl.sync.idx.b32 %r1546, %r1538, %r1430, 31, -1; + shfl.sync.idx.b32 %r1547, %r1539, %r1428, 31, -1; + shfl.sync.idx.b32 %r1548, %r1540, %r1430, 31, -1; + shfl.sync.idx.b32 %r1549, %r1541, %r1428, 31, -1; + shfl.sync.idx.b32 %r1550, %r1542, %r1430, 31, -1; + shfl.sync.idx.b32 %r1551, %r1543, %r1428, 31, -1; + shfl.sync.idx.b32 %r1552, %r1544, %r1430, 31, -1; + selp.b32 %r1177, %r1546, %r1545, %p90; + selp.b32 %r1178, %r1548, %r1547, %p90; + selp.b32 %r1185, %r1550, %r1549, %p90; + selp.b32 %r1186, %r1552, %r1551, %p90; + selp.b32 %r1167, %r1545, %r1546, %p90; + selp.b32 %r1168, %r1547, %r1548, %p90; + selp.b32 %r1171, %r1549, %r1550, %p90; + selp.b32 %r1172, %r1551, %r1552, %p90; + .loc 1 268 23 // chunk_fwd.py:268:23 + mov.b32 %r1173, %r1373; + mov.b32 %r1174, %r1373; + mov.b32 %r1175, %r1373; + mov.b32 %r1176, %r1373; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1173, %r1174, %r1175, %r1176 }, { %r1169, %r1917, %r1919, %r1170 }, { %r1167, %r1168 }, { %r1173, %r1174, %r1175, %r1176 }; + // end inline asm + mov.b32 %r1179, %r1373; + mov.b32 %r1180, %r1373; + mov.b32 %r1181, %r1373; + mov.b32 %r1182, %r1373; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1179, %r1180, %r1181, %r1182 }, { %r1169, %r1917, %r1919, %r1170 }, { %r1171, %r1172 }, { %r1179, %r1180, %r1181, %r1182 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1173, %r1174, %r1175, %r1176 }, { %r1183, %r1918, %r1920, %r1184 }, { %r1177, %r1178 }, { %r1173, %r1174, %r1175, %r1176 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1179, %r1180, %r1181, %r1182 }, { %r1183, %r1918, %r1920, %r1184 }, { %r1185, %r1186 }, { %r1179, %r1180, %r1181, %r1182 }; + // end inline asm + selp.f32 %r1553, %r1173, %r1174, %p87; + selp.f32 %r1554, %r1174, %r1173, %p87; + selp.f32 %r1555, %r1175, %r1176, %p87; + selp.f32 %r1556, %r1176, %r1175, %p87; + selp.f32 %r1557, %r1179, %r1180, %p87; + selp.f32 %r1558, %r1180, %r1179, %p87; + selp.f32 %r1559, %r1181, %r1182, %p87; + selp.f32 %r1560, %r1182, %r1181, %p87; + shfl.sync.idx.b32 %r1561, %r1553, %r1451, 31, -1; + shfl.sync.idx.b32 %r1562, %r1555, %r1451, 31, -1; + shfl.sync.idx.b32 %r1563, %r1554, %r1454, 31, -1; + shfl.sync.idx.b32 %r1564, %r1556, %r1454, 31, -1; + shfl.sync.idx.b32 %r1565, %r1557, %r1451, 31, -1; + shfl.sync.idx.b32 %r1566, %r1559, %r1451, 31, -1; + shfl.sync.idx.b32 %r1567, %r1558, %r1454, 31, -1; + shfl.sync.idx.b32 %r1568, %r1560, %r1454, 31, -1; + selp.b32 %r1189, %r1563, %r1561, %p88; + selp.b32 %r1190, %r1564, %r1562, %p88; + selp.b32 %r1201, %r1567, %r1565, %p88; + selp.b32 %r1202, %r1568, %r1566, %p88; + .loc 1 250 14 // chunk_fwd.py:250:14 + selp.f32 %r1569, %r1125, %r1139, %p89; + selp.f32 %r1570, %r1139, %r1125, %p89; + selp.f32 %r1571, %r1907, %r1908, %p89; + selp.f32 %r1572, %r1908, %r1907, %p89; + selp.f32 %r1573, %r1909, %r1910, %p89; + selp.f32 %r1574, %r1910, %r1909, %p89; + selp.f32 %r1575, %r1126, %r1140, %p89; + selp.f32 %r1576, %r1140, %r1126, %p89; + shfl.sync.idx.b32 %r1577, %r1569, %r1519, 31, -1; + shfl.sync.idx.b32 %r1578, %r1570, %r1521, 31, -1; + shfl.sync.idx.b32 %r1579, %r1571, %r1519, 31, -1; + shfl.sync.idx.b32 %r1580, %r1572, %r1521, 31, -1; + shfl.sync.idx.b32 %r1581, %r1573, %r1519, 31, -1; + shfl.sync.idx.b32 %r1582, %r1574, %r1521, 31, -1; + shfl.sync.idx.b32 %r1583, %r1575, %r1519, 31, -1; + shfl.sync.idx.b32 %r1584, %r1576, %r1521, 31, -1; + selp.b32 %r1192, %r1578, %r1577, %p90; + selp.b32 %r1204, %r1580, %r1579, %p90; + selp.b32 %r1194, %r1582, %r1581, %p90; + selp.b32 %r1210, %r1584, %r1583, %p90; + .loc 1 268 23 // chunk_fwd.py:268:23 + selp.b32 %r1187, %r1561, %r1563, %p88; + selp.b32 %r1188, %r1562, %r1564, %p88; + selp.b32 %r1199, %r1565, %r1567, %p88; + selp.b32 %r1200, %r1566, %r1568, %p88; + .loc 1 250 14 // chunk_fwd.py:250:14 + selp.b32 %r1191, %r1577, %r1578, %p90; + selp.b32 %r1203, %r1579, %r1580, %p90; + selp.b32 %r1193, %r1581, %r1582, %p90; + selp.b32 %r1209, %r1583, %r1584, %p90; + .loc 1 269 8 // chunk_fwd.py:269:8 + mov.b32 %r1195, %r1373; + mov.b32 %r1196, %r1373; + mov.b32 %r1197, %r1373; + mov.b32 %r1198, %r1373; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1195, %r1196, %r1197, %r1198 }, { %r1187, %r1188, %r1189, %r1190 }, { %r1191, %r1192 }, { %r1195, %r1196, %r1197, %r1198 }; + // end inline asm + mov.b32 %r1205, %r1373; + mov.b32 %r1206, %r1373; + mov.b32 %r1207, %r1373; + mov.b32 %r1208, %r1373; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1205, %r1206, %r1207, %r1208 }, { %r1187, %r1188, %r1189, %r1190 }, { %r1193, %r1194 }, { %r1205, %r1206, %r1207, %r1208 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1195, %r1196, %r1197, %r1198 }, { %r1199, %r1200, %r1201, %r1202 }, { %r1203, %r1204 }, { %r1195, %r1196, %r1197, %r1198 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1205, %r1206, %r1207, %r1208 }, { %r1199, %r1200, %r1201, %r1202 }, { %r1209, %r1210 }, { %r1205, %r1206, %r1207, %r1208 }; + // end inline asm + .loc 1 248 14 // chunk_fwd.py:248:14 + and.b32 %r1585, %r1449, 160; + shl.b32 %r1586, %r1439, 7; + or.b32 %r1587, %r1585, %r1586; + or.b32 %r1588, %r1587, %r2; + add.s32 %r1589, %r109, %r1588; + ld.shared.b32 %r1215, [%r1589]; + ld.shared.b32 %r1227, [%r1589+64]; + ld.shared.b32 %r1216, [%r1589+512]; + ld.shared.b32 %r1228, [%r1589+576]; + xor.b32 %r1590, %r1588, 32; + add.s32 %r1591, %r109, %r1590; + ld.shared.b32 %r1217, [%r1591]; + ld.shared.b32 %r1233, [%r1591+64]; + ld.shared.b32 %r1218, [%r1591+512]; + ld.shared.b32 %r1234, [%r1591+576]; + .loc 1 275 22 // chunk_fwd.py:275:22 + mov.b32 %r1219, %r1373; + mov.b32 %r1220, %r1373; + mov.b32 %r1221, %r1373; + mov.b32 %r1222, %r1373; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1219, %r1220, %r1221, %r1222 }, { %r1211, %r1212, %r1213, %r1214 }, { %r1215, %r1216 }, { %r1219, %r1220, %r1221, %r1222 }; + // end inline asm + mov.b32 %r1229, %r1373; + mov.b32 %r1230, %r1373; + mov.b32 %r1231, %r1373; + mov.b32 %r1232, %r1373; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1229, %r1230, %r1231, %r1232 }, { %r1211, %r1212, %r1213, %r1214 }, { %r1217, %r1218 }, { %r1229, %r1230, %r1231, %r1232 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1219, %r1220, %r1221, %r1222 }, { %r1223, %r1224, %r1225, %r1226 }, { %r1227, %r1228 }, { %r1219, %r1220, %r1221, %r1222 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1229, %r1230, %r1231, %r1232 }, { %r1223, %r1224, %r1225, %r1226 }, { %r1233, %r1234 }, { %r1229, %r1230, %r1231, %r1232 }; + // end inline asm + .loc 1 257 14 // chunk_fwd.py:257:14 + selp.f32 %r1592, %r1469, %r1470, %p90; + selp.f32 %r1593, %r1470, %r1469, %p90; + selp.f32 %r1594, %r1471, %r1472, %p90; + selp.f32 %r1595, %r1472, %r1471, %p90; + selp.f32 %r1596, %r1473, %r1474, %p90; + selp.f32 %r1597, %r1474, %r1473, %p90; + selp.f32 %r1598, %r1475, %r1476, %p90; + selp.f32 %r1599, %r1476, %r1475, %p90; + shfl.sync.idx.b32 %r1600, %r1592, %r1428, 31, -1; + shfl.sync.idx.b32 %r1601, %r1593, %r1430, 31, -1; + shfl.sync.idx.b32 %r1602, %r1594, %r1428, 31, -1; + shfl.sync.idx.b32 %r1603, %r1595, %r1430, 31, -1; + shfl.sync.idx.b32 %r1604, %r1596, %r1428, 31, -1; + shfl.sync.idx.b32 %r1605, %r1597, %r1430, 31, -1; + shfl.sync.idx.b32 %r1606, %r1598, %r1428, 31, -1; + shfl.sync.idx.b32 %r1607, %r1599, %r1430, 31, -1; + selp.b32 %r1247, %r1601, %r1600, %p90; + selp.b32 %r1248, %r1603, %r1602, %p90; + selp.b32 %r1249, %r1605, %r1604, %p90; + selp.b32 %r1250, %r1607, %r1606, %p90; + selp.b32 %r1239, %r1600, %r1601, %p90; + selp.b32 %r1240, %r1602, %r1603, %p90; + selp.b32 %r1241, %r1604, %r1605, %p90; + selp.b32 %r1242, %r1606, %r1607, %p90; + .loc 1 276 22 // chunk_fwd.py:276:22 + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1219, %r1220, %r1221, %r1222 }, { %r1235, %r1236, %r1237, %r1238 }, { %r1239, %r1240 }, { %r1219, %r1220, %r1221, %r1222 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1229, %r1230, %r1231, %r1232 }, { %r1235, %r1236, %r1237, %r1238 }, { %r1241, %r1242 }, { %r1229, %r1230, %r1231, %r1232 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1219, %r1220, %r1221, %r1222 }, { %r1243, %r1244, %r1245, %r1246 }, { %r1247, %r1248 }, { %r1219, %r1220, %r1221, %r1222 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1229, %r1230, %r1231, %r1232 }, { %r1243, %r1244, %r1245, %r1246 }, { %r1249, %r1250 }, { %r1229, %r1230, %r1231, %r1232 }; + // end inline asm + .loc 1 250 14 // chunk_fwd.py:250:14 + selp.f32 %r1608, %r1125, %r1139, %p87; + selp.f32 %r1609, %r1139, %r1125, %p87; + selp.f32 %r1610, %r1907, %r1908, %p87; + selp.f32 %r1611, %r1908, %r1907, %p87; + selp.f32 %r1612, %r1909, %r1910, %p87; + selp.f32 %r1613, %r1910, %r1909, %p87; + selp.f32 %r1614, %r1126, %r1140, %p87; + selp.f32 %r1615, %r1140, %r1126, %p87; + shfl.sync.idx.b32 %r1616, %r1608, %r1451, 31, -1; + shfl.sync.idx.b32 %r1617, %r1610, %r1451, 31, -1; + shfl.sync.idx.b32 %r1618, %r1609, %r1454, 31, -1; + shfl.sync.idx.b32 %r1619, %r1611, %r1454, 31, -1; + shfl.sync.idx.b32 %r1620, %r1612, %r1451, 31, -1; + shfl.sync.idx.b32 %r1621, %r1614, %r1451, 31, -1; + shfl.sync.idx.b32 %r1622, %r1613, %r1454, 31, -1; + shfl.sync.idx.b32 %r1623, %r1615, %r1454, 31, -1; + selp.b32 %r1253, %r1618, %r1616, %p88; + selp.b32 %r1254, %r1619, %r1617, %p88; + selp.b32 %r1265, %r1622, %r1620, %p88; + selp.b32 %r1266, %r1623, %r1621, %p88; + .loc 1 276 22 // chunk_fwd.py:276:22 + selp.f32 %r1624, %r1219, %r1220, %p89; + selp.f32 %r1625, %r1220, %r1219, %p89; + selp.f32 %r1626, %r1221, %r1222, %p89; + selp.f32 %r1627, %r1222, %r1221, %p89; + selp.f32 %r1628, %r1229, %r1230, %p89; + selp.f32 %r1629, %r1230, %r1229, %p89; + selp.f32 %r1630, %r1231, %r1232, %p89; + selp.f32 %r1631, %r1232, %r1231, %p89; + shfl.sync.idx.b32 %r1632, %r1624, %r1519, 31, -1; + shfl.sync.idx.b32 %r1633, %r1625, %r1521, 31, -1; + shfl.sync.idx.b32 %r1634, %r1626, %r1519, 31, -1; + shfl.sync.idx.b32 %r1635, %r1627, %r1521, 31, -1; + shfl.sync.idx.b32 %r1636, %r1628, %r1519, 31, -1; + shfl.sync.idx.b32 %r1637, %r1629, %r1521, 31, -1; + shfl.sync.idx.b32 %r1638, %r1630, %r1519, 31, -1; + shfl.sync.idx.b32 %r1639, %r1631, %r1521, 31, -1; + selp.b32 %r1256, %r1633, %r1632, %p90; + selp.b32 %r1268, %r1635, %r1634, %p90; + selp.b32 %r1258, %r1637, %r1636, %p90; + selp.b32 %r1274, %r1639, %r1638, %p90; + .loc 1 250 14 // chunk_fwd.py:250:14 + selp.b32 %r1251, %r1616, %r1618, %p88; + selp.b32 %r1252, %r1617, %r1619, %p88; + selp.b32 %r1263, %r1620, %r1622, %p88; + selp.b32 %r1264, %r1621, %r1623, %p88; + .loc 1 276 22 // chunk_fwd.py:276:22 + selp.b32 %r1255, %r1632, %r1633, %p90; + selp.b32 %r1267, %r1634, %r1635, %p90; + selp.b32 %r1257, %r1636, %r1637, %p90; + selp.b32 %r1273, %r1638, %r1639, %p90; + .loc 1 275 8 // chunk_fwd.py:275:8 + mov.b32 %r1259, %r1373; + mov.b32 %r1260, %r1373; + mov.b32 %r1261, %r1373; + mov.b32 %r1262, %r1373; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1259, %r1260, %r1261, %r1262 }, { %r1251, %r1252, %r1253, %r1254 }, { %r1255, %r1256 }, { %r1259, %r1260, %r1261, %r1262 }; + // end inline asm + mov.b32 %r1269, %r1373; + mov.b32 %r1270, %r1373; + mov.b32 %r1271, %r1373; + mov.b32 %r1272, %r1373; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1269, %r1270, %r1271, %r1272 }, { %r1251, %r1252, %r1253, %r1254 }, { %r1257, %r1258 }, { %r1269, %r1270, %r1271, %r1272 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1259, %r1260, %r1261, %r1262 }, { %r1263, %r1264, %r1265, %r1266 }, { %r1267, %r1268 }, { %r1259, %r1260, %r1261, %r1262 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1269, %r1270, %r1271, %r1272 }, { %r1263, %r1264, %r1265, %r1266 }, { %r1273, %r1274 }, { %r1269, %r1270, %r1271, %r1272 }; + // end inline asm + .loc 1 273 14 // chunk_fwd.py:273:14 + sub.f32 %r1640, %r1468, %r1259; + sub.f32 %r1641, %r1468, %r1260; + sub.f32 %r1642, %r1468, %r1261; + sub.f32 %r1643, %r1468, %r1262; + sub.f32 %r1644, %r1468, %r1269; + sub.f32 %r1645, %r1468, %r1270; + sub.f32 %r1646, %r1468, %r1271; + sub.f32 %r1647, %r1468, %r1272; + .loc 1 249 14 // chunk_fwd.py:249:14 + selp.f32 %r1648, %r1081, %r1095, %p90; + selp.f32 %r1649, %r1095, %r1081, %p90; + selp.f32 %r1650, %r1898, %r1899, %p90; + selp.f32 %r1651, %r1899, %r1898, %p90; + selp.f32 %r1652, %r1900, %r1901, %p90; + selp.f32 %r1653, %r1901, %r1900, %p90; + selp.f32 %r1654, %r1082, %r1096, %p90; + selp.f32 %r1655, %r1096, %r1082, %p90; + shfl.sync.idx.b32 %r1656, %r1648, %r1428, 31, -1; + shfl.sync.idx.b32 %r1657, %r1649, %r1430, 31, -1; + shfl.sync.idx.b32 %r1658, %r1650, %r1428, 31, -1; + shfl.sync.idx.b32 %r1659, %r1651, %r1430, 31, -1; + shfl.sync.idx.b32 %r1660, %r1652, %r1428, 31, -1; + shfl.sync.idx.b32 %r1661, %r1653, %r1430, 31, -1; + shfl.sync.idx.b32 %r1662, %r1654, %r1428, 31, -1; + shfl.sync.idx.b32 %r1663, %r1655, %r1430, 31, -1; + selp.b32 %r1291, %r1657, %r1656, %p90; + selp.b32 %r1292, %r1659, %r1658, %p90; + selp.b32 %r1297, %r1661, %r1660, %p90; + selp.b32 %r1298, %r1663, %r1662, %p90; + selp.b32 %r1279, %r1656, %r1657, %p90; + selp.b32 %r1280, %r1658, %r1659, %p90; + selp.b32 %r1281, %r1660, %r1661, %p90; + selp.b32 %r1282, %r1662, %r1663, %p90; + .loc 1 281 22 // chunk_fwd.py:281:22 + mov.b32 %r1283, %r1373; + mov.b32 %r1284, %r1373; + mov.b32 %r1285, %r1373; + mov.b32 %r1286, %r1373; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1283, %r1284, %r1285, %r1286 }, { %r1275, %r1276, %r1277, %r1278 }, { %r1279, %r1280 }, { %r1283, %r1284, %r1285, %r1286 }; + // end inline asm + mov.b32 %r1293, %r1373; + mov.b32 %r1294, %r1373; + mov.b32 %r1295, %r1373; + mov.b32 %r1296, %r1373; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1293, %r1294, %r1295, %r1296 }, { %r1275, %r1276, %r1277, %r1278 }, { %r1281, %r1282 }, { %r1293, %r1294, %r1295, %r1296 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1283, %r1284, %r1285, %r1286 }, { %r1287, %r1288, %r1289, %r1290 }, { %r1291, %r1292 }, { %r1283, %r1284, %r1285, %r1286 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1293, %r1294, %r1295, %r1296 }, { %r1287, %r1288, %r1289, %r1290 }, { %r1297, %r1298 }, { %r1293, %r1294, %r1295, %r1296 }; + // end inline asm + .loc 1 262 14 // chunk_fwd.py:262:14 + selp.f32 %r1664, %r1529, %r1530, %p90; + selp.f32 %r1665, %r1530, %r1529, %p90; + selp.f32 %r1666, %r1531, %r1532, %p90; + selp.f32 %r1667, %r1532, %r1531, %p90; + selp.f32 %r1668, %r1533, %r1534, %p90; + selp.f32 %r1669, %r1534, %r1533, %p90; + selp.f32 %r1670, %r1535, %r1536, %p90; + selp.f32 %r1671, %r1536, %r1535, %p90; + shfl.sync.idx.b32 %r1672, %r1664, %r1428, 31, -1; + shfl.sync.idx.b32 %r1673, %r1665, %r1430, 31, -1; + shfl.sync.idx.b32 %r1674, %r1666, %r1428, 31, -1; + shfl.sync.idx.b32 %r1675, %r1667, %r1430, 31, -1; + shfl.sync.idx.b32 %r1676, %r1668, %r1428, 31, -1; + shfl.sync.idx.b32 %r1677, %r1669, %r1430, 31, -1; + shfl.sync.idx.b32 %r1678, %r1670, %r1428, 31, -1; + shfl.sync.idx.b32 %r1679, %r1671, %r1430, 31, -1; + selp.b32 %r1311, %r1673, %r1672, %p90; + selp.b32 %r1312, %r1675, %r1674, %p90; + selp.b32 %r1313, %r1677, %r1676, %p90; + selp.b32 %r1314, %r1679, %r1678, %p90; + selp.b32 %r1303, %r1672, %r1673, %p90; + selp.b32 %r1304, %r1674, %r1675, %p90; + selp.b32 %r1305, %r1676, %r1677, %p90; + selp.b32 %r1306, %r1678, %r1679, %p90; + .loc 1 282 22 // chunk_fwd.py:282:22 + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1283, %r1284, %r1285, %r1286 }, { %r1299, %r1300, %r1301, %r1302 }, { %r1303, %r1304 }, { %r1283, %r1284, %r1285, %r1286 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1293, %r1294, %r1295, %r1296 }, { %r1299, %r1300, %r1301, %r1302 }, { %r1305, %r1306 }, { %r1293, %r1294, %r1295, %r1296 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1283, %r1284, %r1285, %r1286 }, { %r1307, %r1308, %r1309, %r1310 }, { %r1311, %r1312 }, { %r1283, %r1284, %r1285, %r1286 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1293, %r1294, %r1295, %r1296 }, { %r1307, %r1308, %r1309, %r1310 }, { %r1313, %r1314 }, { %r1293, %r1294, %r1295, %r1296 }; + // end inline asm + .loc 1 251 14 // chunk_fwd.py:251:14 + selp.f32 %r1680, %r1169, %r1183, %p87; + selp.f32 %r1681, %r1183, %r1169, %p87; + selp.f32 %r1682, %r1917, %r1918, %p87; + selp.f32 %r1683, %r1918, %r1917, %p87; + selp.f32 %r1684, %r1919, %r1920, %p87; + selp.f32 %r1685, %r1920, %r1919, %p87; + selp.f32 %r1686, %r1170, %r1184, %p87; + selp.f32 %r1687, %r1184, %r1170, %p87; + shfl.sync.idx.b32 %r1688, %r1680, %r1451, 31, -1; + shfl.sync.idx.b32 %r1689, %r1682, %r1451, 31, -1; + shfl.sync.idx.b32 %r1690, %r1681, %r1454, 31, -1; + shfl.sync.idx.b32 %r1691, %r1683, %r1454, 31, -1; + shfl.sync.idx.b32 %r1692, %r1684, %r1451, 31, -1; + shfl.sync.idx.b32 %r1693, %r1686, %r1451, 31, -1; + shfl.sync.idx.b32 %r1694, %r1685, %r1454, 31, -1; + shfl.sync.idx.b32 %r1695, %r1687, %r1454, 31, -1; + selp.b32 %r1317, %r1690, %r1688, %p88; + selp.b32 %r1318, %r1691, %r1689, %p88; + selp.b32 %r1329, %r1694, %r1692, %p88; + selp.b32 %r1330, %r1695, %r1693, %p88; + .loc 1 282 22 // chunk_fwd.py:282:22 + selp.f32 %r1696, %r1283, %r1284, %p89; + selp.f32 %r1697, %r1284, %r1283, %p89; + selp.f32 %r1698, %r1285, %r1286, %p89; + selp.f32 %r1699, %r1286, %r1285, %p89; + selp.f32 %r1700, %r1293, %r1294, %p89; + selp.f32 %r1701, %r1294, %r1293, %p89; + selp.f32 %r1702, %r1295, %r1296, %p89; + selp.f32 %r1703, %r1296, %r1295, %p89; + shfl.sync.idx.b32 %r1704, %r1696, %r1519, 31, -1; + shfl.sync.idx.b32 %r1705, %r1697, %r1521, 31, -1; + shfl.sync.idx.b32 %r1706, %r1698, %r1519, 31, -1; + shfl.sync.idx.b32 %r1707, %r1699, %r1521, 31, -1; + shfl.sync.idx.b32 %r1708, %r1700, %r1519, 31, -1; + shfl.sync.idx.b32 %r1709, %r1701, %r1521, 31, -1; + shfl.sync.idx.b32 %r1710, %r1702, %r1519, 31, -1; + shfl.sync.idx.b32 %r1711, %r1703, %r1521, 31, -1; + selp.b32 %r1320, %r1705, %r1704, %p90; + selp.b32 %r1332, %r1707, %r1706, %p90; + selp.b32 %r1322, %r1709, %r1708, %p90; + selp.b32 %r1338, %r1711, %r1710, %p90; + .loc 1 251 14 // chunk_fwd.py:251:14 + selp.b32 %r1315, %r1688, %r1690, %p88; + selp.b32 %r1316, %r1689, %r1691, %p88; + selp.b32 %r1327, %r1692, %r1694, %p88; + selp.b32 %r1328, %r1693, %r1695, %p88; + .loc 1 282 22 // chunk_fwd.py:282:22 + selp.b32 %r1319, %r1704, %r1705, %p90; + selp.b32 %r1331, %r1706, %r1707, %p90; + selp.b32 %r1321, %r1708, %r1709, %p90; + selp.b32 %r1337, %r1710, %r1711, %p90; + .loc 1 281 8 // chunk_fwd.py:281:8 + mov.b32 %r1323, %r1373; + mov.b32 %r1324, %r1373; + mov.b32 %r1325, %r1373; + mov.b32 %r1326, %r1373; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1323, %r1324, %r1325, %r1326 }, { %r1315, %r1316, %r1317, %r1318 }, { %r1319, %r1320 }, { %r1323, %r1324, %r1325, %r1326 }; + // end inline asm + mov.b32 %r1333, %r1373; + mov.b32 %r1334, %r1373; + mov.b32 %r1335, %r1373; + mov.b32 %r1336, %r1373; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1333, %r1334, %r1335, %r1336 }, { %r1315, %r1316, %r1317, %r1318 }, { %r1321, %r1322 }, { %r1333, %r1334, %r1335, %r1336 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1323, %r1324, %r1325, %r1326 }, { %r1327, %r1328, %r1329, %r1330 }, { %r1331, %r1332 }, { %r1323, %r1324, %r1325, %r1326 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1333, %r1334, %r1335, %r1336 }, { %r1327, %r1328, %r1329, %r1330 }, { %r1337, %r1338 }, { %r1333, %r1334, %r1335, %r1336 }; + // end inline asm + .loc 1 287 22 // chunk_fwd.py:287:22 + mov.b32 %r1343, %r1373; + mov.b32 %r1344, %r1373; + mov.b32 %r1345, %r1373; + mov.b32 %r1346, %r1373; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1343, %r1344, %r1345, %r1346 }, { %r1339, %r1340, %r1341, %r1342 }, { %r1215, %r1216 }, { %r1343, %r1344, %r1345, %r1346 }; + // end inline asm + mov.b32 %r1351, %r1373; + mov.b32 %r1352, %r1373; + mov.b32 %r1353, %r1373; + mov.b32 %r1354, %r1373; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1351, %r1352, %r1353, %r1354 }, { %r1339, %r1340, %r1341, %r1342 }, { %r1217, %r1218 }, { %r1351, %r1352, %r1353, %r1354 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1343, %r1344, %r1345, %r1346 }, { %r1347, %r1348, %r1349, %r1350 }, { %r1227, %r1228 }, { %r1343, %r1344, %r1345, %r1346 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1351, %r1352, %r1353, %r1354 }, { %r1347, %r1348, %r1349, %r1350 }, { %r1233, %r1234 }, { %r1351, %r1352, %r1353, %r1354 }; + // end inline asm + .loc 1 288 22 // chunk_fwd.py:288:22 + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1343, %r1344, %r1345, %r1346 }, { %r1275, %r1276, %r1277, %r1278 }, { %r1239, %r1240 }, { %r1343, %r1344, %r1345, %r1346 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1351, %r1352, %r1353, %r1354 }, { %r1275, %r1276, %r1277, %r1278 }, { %r1241, %r1242 }, { %r1351, %r1352, %r1353, %r1354 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1343, %r1344, %r1345, %r1346 }, { %r1287, %r1288, %r1289, %r1290 }, { %r1247, %r1248 }, { %r1343, %r1344, %r1345, %r1346 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1351, %r1352, %r1353, %r1354 }, { %r1287, %r1288, %r1289, %r1290 }, { %r1249, %r1250 }, { %r1351, %r1352, %r1353, %r1354 }; + // end inline asm + .loc 1 273 14 // chunk_fwd.py:273:14 + selp.f32 %r1712, %r1640, %r1641, %p90; + selp.f32 %r1713, %r1641, %r1640, %p90; + selp.f32 %r1714, %r1642, %r1643, %p90; + selp.f32 %r1715, %r1643, %r1642, %p90; + selp.f32 %r1716, %r1644, %r1645, %p90; + selp.f32 %r1717, %r1645, %r1644, %p90; + selp.f32 %r1718, %r1646, %r1647, %p90; + selp.f32 %r1719, %r1647, %r1646, %p90; + shfl.sync.idx.b32 %r1720, %r1712, %r1428, 31, -1; + shfl.sync.idx.b32 %r1721, %r1713, %r1430, 31, -1; + shfl.sync.idx.b32 %r1722, %r1714, %r1428, 31, -1; + shfl.sync.idx.b32 %r1723, %r1715, %r1430, 31, -1; + shfl.sync.idx.b32 %r1724, %r1716, %r1428, 31, -1; + shfl.sync.idx.b32 %r1725, %r1717, %r1430, 31, -1; + shfl.sync.idx.b32 %r1726, %r1718, %r1428, 31, -1; + shfl.sync.idx.b32 %r1727, %r1719, %r1430, 31, -1; + selp.b32 %r1359, %r1721, %r1720, %p90; + selp.b32 %r1360, %r1723, %r1722, %p90; + selp.b32 %r1361, %r1725, %r1724, %p90; + selp.b32 %r1362, %r1727, %r1726, %p90; + selp.b32 %r1355, %r1720, %r1721, %p90; + selp.b32 %r1356, %r1722, %r1723, %p90; + selp.b32 %r1357, %r1724, %r1725, %p90; + selp.b32 %r1358, %r1726, %r1727, %p90; + .loc 1 289 22 // chunk_fwd.py:289:22 + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1343, %r1344, %r1345, %r1346 }, { %r1299, %r1300, %r1301, %r1302 }, { %r1355, %r1356 }, { %r1343, %r1344, %r1345, %r1346 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1351, %r1352, %r1353, %r1354 }, { %r1299, %r1300, %r1301, %r1302 }, { %r1357, %r1358 }, { %r1351, %r1352, %r1353, %r1354 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1343, %r1344, %r1345, %r1346 }, { %r1307, %r1308, %r1309, %r1310 }, { %r1359, %r1360 }, { %r1343, %r1344, %r1345, %r1346 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1351, %r1352, %r1353, %r1354 }, { %r1307, %r1308, %r1309, %r1310 }, { %r1361, %r1362 }, { %r1351, %r1352, %r1353, %r1354 }; + // end inline asm + selp.f32 %r1728, %r1343, %r1344, %p89; + selp.f32 %r1729, %r1344, %r1343, %p89; + selp.f32 %r1730, %r1345, %r1346, %p89; + selp.f32 %r1731, %r1346, %r1345, %p89; + selp.f32 %r1732, %r1351, %r1352, %p89; + selp.f32 %r1733, %r1352, %r1351, %p89; + selp.f32 %r1734, %r1353, %r1354, %p89; + selp.f32 %r1735, %r1354, %r1353, %p89; + shfl.sync.idx.b32 %r1736, %r1728, %r1519, 31, -1; + shfl.sync.idx.b32 %r1737, %r1729, %r1521, 31, -1; + shfl.sync.idx.b32 %r1738, %r1730, %r1519, 31, -1; + shfl.sync.idx.b32 %r1739, %r1731, %r1521, 31, -1; + shfl.sync.idx.b32 %r1740, %r1732, %r1519, 31, -1; + shfl.sync.idx.b32 %r1741, %r1733, %r1521, 31, -1; + shfl.sync.idx.b32 %r1742, %r1734, %r1519, 31, -1; + shfl.sync.idx.b32 %r1743, %r1735, %r1521, 31, -1; + selp.b32 %r1364, %r1737, %r1736, %p90; + selp.b32 %r1372, %r1739, %r1738, %p90; + selp.b32 %r1366, %r1741, %r1740, %p90; + selp.b32 %r1378, %r1743, %r1742, %p90; + selp.b32 %r1363, %r1736, %r1737, %p90; + selp.b32 %r1371, %r1738, %r1739, %p90; + selp.b32 %r1365, %r1740, %r1741, %p90; + selp.b32 %r1377, %r1742, %r1743, %p90; + .loc 1 287 8 // chunk_fwd.py:287:8 + mov.b32 %r1367, %r1373; + mov.b32 %r1368, %r1373; + mov.b32 %r1369, %r1373; + mov.b32 %r1370, %r1373; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1367, %r1368, %r1369, %r1370 }, { %r1315, %r1316, %r1317, %r1318 }, { %r1363, %r1364 }, { %r1367, %r1368, %r1369, %r1370 }; + // end inline asm + mov.b32 %r1374, %r1373; + mov.b32 %r1375, %r1373; + mov.b32 %r1376, %r1373; + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1373, %r1374, %r1375, %r1376 }, { %r1315, %r1316, %r1317, %r1318 }, { %r1365, %r1366 }, { %r1373, %r1374, %r1375, %r1376 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1367, %r1368, %r1369, %r1370 }, { %r1327, %r1328, %r1329, %r1330 }, { %r1371, %r1372 }, { %r1367, %r1368, %r1369, %r1370 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 { %r1373, %r1374, %r1375, %r1376 }, { %r1327, %r1328, %r1329, %r1330 }, { %r1377, %r1378 }, { %r1373, %r1374, %r1375, %r1376 }; + // end inline asm + .loc 1 297 16 // chunk_fwd.py:297:16 + shl.b64 %rd261, %rd8, 12; + add.s64 %rd262, %rd45, %rd261; + .loc 1 297 51 // chunk_fwd.py:297:51 + mul.wide.u32 %rd263, %r34, 2; + add.s64 %rd251, %rd262, %rd263; + .loc 1 298 16 // chunk_fwd.py:298:16 + shl.b64 %rd264, %rd9, 12; + add.s64 %rd265, %rd45, %rd264; + .loc 1 298 51 // chunk_fwd.py:298:51 + add.s64 %rd252, %rd265, %rd263; + .loc 1 299 51 // chunk_fwd.py:299:51 + add.s64 %rd253, %rd252, 32; + .loc 1 300 16 // chunk_fwd.py:300:16 + shl.b64 %rd266, %rd10, 12; + add.s64 %rd267, %rd45, %rd266; + .loc 1 300 51 // chunk_fwd.py:300:51 + add.s64 %rd254, %rd267, %rd263; + .loc 1 301 51 // chunk_fwd.py:301:51 + add.s64 %rd255, %rd254, 32; + .loc 1 302 51 // chunk_fwd.py:302:51 + add.s64 %rd256, %rd254, 64; + .loc 1 303 16 // chunk_fwd.py:303:16 + shl.b64 %rd268, %rd11, 12; + add.s64 %rd269, %rd45, %rd268; + .loc 1 303 51 // chunk_fwd.py:303:51 + add.s64 %rd257, %rd269, %rd263; + .loc 1 304 51 // chunk_fwd.py:304:51 + add.s64 %rd258, %rd257, 32; + .loc 1 305 51 // chunk_fwd.py:305:51 + add.s64 %rd259, %rd257, 64; + .loc 1 306 51 // chunk_fwd.py:306:51 + add.s64 %rd260, %rd257, 96; + .loc 1 319 20 // chunk_fwd.py:319:20 + bar.sync 0; + and.b32 %r1744, %r13, 176; + selp.b32 %r1745, 0, 320, %p90; + shl.b32 %r1746, %r33, 6; + or.b32 %r1747, %r1746, %r1517; + and.b32 %r1748, %r1747, 64; + or.b32 %r1749, %r1745, %r1744; + xor.b32 %r1750, %r1749, %r1748; + add.s32 %r1751, %r109, %r1750; + cvt.rn.bf16x2.f32 %r1752, %r1075, %r1074; + cvt.rn.bf16x2.f32 %r1753, %r1078, %r1077; + stmatrix.sync.aligned.m8n8.x2.shared.b16 [%r1751], {%r1752, %r1753}; + bar.sync 0; + shl.b32 %r1754, %r33, 3; + or.b32 %r1755, %r1754, %r32; + shl.b32 %r1756, %r1755, 3; + and.b32 %r1757, %r1756, 176; + shl.b32 %r1758, %r1, 6; + and.b32 %r1759, %r1758, 64; + bfe.s32 %r1760, %r1755, 3, 1; + and.b32 %r1761, %r1760, 320; + or.b32 %r1762, %r1757, %r1759; + xor.b32 %r1763, %r1762, %r1761; + add.s32 %r1764, %r109, %r1763; + ldmatrix.sync.aligned.m8n8.x1.shared.b16 {%r1379}, [%r1764]; + // begin inline asm + @%p29 st.global.b32 [ %rd251 + 0 ], { %r1379 }; + // end inline asm + .loc 1 320 20 // chunk_fwd.py:320:20 + bar.sync 0; + and.b32 %r1765, %r1517, 64; + xor.b32 %r1766, %r1749, %r1765; + add.s32 %r1767, %r109, %r1766; + cvt.rn.bf16x2.f32 %r1768, %r1470, %r1469; + cvt.rn.bf16x2.f32 %r1769, %r1472, %r1471; + cvt.rn.bf16x2.f32 %r1770, %r1474, %r1473; + cvt.rn.bf16x2.f32 %r1771, %r1476, %r1475; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1767], {%r1768, %r1769, %r1770, %r1771}; + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x1.shared.b16 {%r1380}, [%r1764]; + // begin inline asm + @%p77 st.global.b32 [ %rd252 + 0 ], { %r1380 }; + // end inline asm + .loc 1 321 20 // chunk_fwd.py:321:20 + bar.sync 0; + cvt.rn.bf16x2.f32 %r1772, %r1095, %r1081; + cvt.rn.bf16x2.f32 %r1773, %r1899, %r1898; + cvt.rn.bf16x2.f32 %r1774, %r1901, %r1900; + cvt.rn.bf16x2.f32 %r1775, %r1096, %r1082; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1767], {%r1772, %r1773, %r1774, %r1775}; + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x1.shared.b16 {%r1381}, [%r1764]; + // begin inline asm + @%p77 st.global.b32 [ %rd253 + 0 ], { %r1381 }; + // end inline asm + .loc 1 322 20 // chunk_fwd.py:322:20 + bar.sync 0; + cvt.rn.bf16x2.f32 %r1776, %r1641, %r1640; + cvt.rn.bf16x2.f32 %r1777, %r1643, %r1642; + cvt.rn.bf16x2.f32 %r1778, %r1645, %r1644; + cvt.rn.bf16x2.f32 %r1779, %r1647, %r1646; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1767], {%r1776, %r1777, %r1778, %r1779}; + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x1.shared.b16 {%r1382}, [%r1764]; + // begin inline asm + @%p78 st.global.b32 [ %rd254 + 0 ], { %r1382 }; + // end inline asm + .loc 1 323 20 // chunk_fwd.py:323:20 + bar.sync 0; + cvt.rn.bf16x2.f32 %r1780, %r1530, %r1529; + cvt.rn.bf16x2.f32 %r1781, %r1532, %r1531; + cvt.rn.bf16x2.f32 %r1782, %r1534, %r1533; + cvt.rn.bf16x2.f32 %r1783, %r1536, %r1535; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1767], {%r1780, %r1781, %r1782, %r1783}; + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x1.shared.b16 {%r1383}, [%r1764]; + // begin inline asm + @%p78 st.global.b32 [ %rd255 + 0 ], { %r1383 }; + // end inline asm + .loc 1 324 20 // chunk_fwd.py:324:20 + bar.sync 0; + cvt.rn.bf16x2.f32 %r1784, %r1139, %r1125; + cvt.rn.bf16x2.f32 %r1785, %r1908, %r1907; + cvt.rn.bf16x2.f32 %r1786, %r1910, %r1909; + cvt.rn.bf16x2.f32 %r1787, %r1140, %r1126; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1767], {%r1784, %r1785, %r1786, %r1787}; + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x1.shared.b16 {%r1384}, [%r1764]; + // begin inline asm + @%p78 st.global.b32 [ %rd256 + 0 ], { %r1384 }; + // end inline asm + .loc 1 285 14 // chunk_fwd.py:285:14 + mov.b64 %rd270, {%r1367, %r1368}; + mov.b64 %rd271, {%r1468, %r1468}; + sub.f32x2 %rd272, %rd271, %rd270; + .loc 1 325 30 // chunk_fwd.py:325:30 + mov.b64 {%r1788, %r1789}, %rd272; + cvt.rn.bf16x2.f32 %r1790, %r1789, %r1788; + .loc 1 285 14 // chunk_fwd.py:285:14 + mov.b64 %rd273, {%r1369, %r1370}; + sub.f32x2 %rd274, %rd271, %rd273; + .loc 1 325 30 // chunk_fwd.py:325:30 + mov.b64 {%r1791, %r1792}, %rd274; + cvt.rn.bf16x2.f32 %r1793, %r1792, %r1791; + .loc 1 285 14 // chunk_fwd.py:285:14 + mov.b64 %rd275, {%r1373, %r1374}; + sub.f32x2 %rd276, %rd271, %rd275; + .loc 1 325 30 // chunk_fwd.py:325:30 + mov.b64 {%r1794, %r1795}, %rd276; + cvt.rn.bf16x2.f32 %r1796, %r1795, %r1794; + .loc 1 285 14 // chunk_fwd.py:285:14 + mov.b64 %rd277, {%r1375, %r1376}; + sub.f32x2 %rd278, %rd271, %rd277; + .loc 1 325 30 // chunk_fwd.py:325:30 + mov.b64 {%r1797, %r1798}, %rd278; + cvt.rn.bf16x2.f32 %r1799, %r1798, %r1797; + .loc 1 325 20 // chunk_fwd.py:325:20 + bar.sync 0; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1767], {%r1790, %r1793, %r1796, %r1799}; + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x1.shared.b16 {%r1385}, [%r1764]; + // begin inline asm + @%p79 st.global.b32 [ %rd257 + 0 ], { %r1385 }; + // end inline asm + .loc 1 279 14 // chunk_fwd.py:279:14 + mov.b64 %rd279, {%r1323, %r1324}; + sub.f32x2 %rd280, %rd271, %rd279; + .loc 1 326 30 // chunk_fwd.py:326:30 + mov.b64 {%r1800, %r1801}, %rd280; + cvt.rn.bf16x2.f32 %r1802, %r1801, %r1800; + .loc 1 279 14 // chunk_fwd.py:279:14 + mov.b64 %rd281, {%r1325, %r1326}; + sub.f32x2 %rd282, %rd271, %rd281; + .loc 1 326 30 // chunk_fwd.py:326:30 + mov.b64 {%r1803, %r1804}, %rd282; + cvt.rn.bf16x2.f32 %r1805, %r1804, %r1803; + .loc 1 279 14 // chunk_fwd.py:279:14 + mov.b64 %rd283, {%r1333, %r1334}; + sub.f32x2 %rd284, %rd271, %rd283; + .loc 1 326 30 // chunk_fwd.py:326:30 + mov.b64 {%r1806, %r1807}, %rd284; + cvt.rn.bf16x2.f32 %r1808, %r1807, %r1806; + .loc 1 279 14 // chunk_fwd.py:279:14 + mov.b64 %rd285, {%r1335, %r1336}; + sub.f32x2 %rd286, %rd271, %rd285; + .loc 1 326 30 // chunk_fwd.py:326:30 + mov.b64 {%r1809, %r1810}, %rd286; + cvt.rn.bf16x2.f32 %r1811, %r1810, %r1809; + .loc 1 326 20 // chunk_fwd.py:326:20 + bar.sync 0; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1767], {%r1802, %r1805, %r1808, %r1811}; + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x1.shared.b16 {%r1386}, [%r1764]; + // begin inline asm + @%p79 st.global.b32 [ %rd258 + 0 ], { %r1386 }; + // end inline asm + .loc 1 267 14 // chunk_fwd.py:267:14 + mov.b64 %rd287, {%r1195, %r1196}; + sub.f32x2 %rd288, %rd271, %rd287; + .loc 1 327 30 // chunk_fwd.py:327:30 + mov.b64 {%r1812, %r1813}, %rd288; + cvt.rn.bf16x2.f32 %r1814, %r1813, %r1812; + .loc 1 267 14 // chunk_fwd.py:267:14 + mov.b64 %rd289, {%r1197, %r1198}; + sub.f32x2 %rd290, %rd271, %rd289; + .loc 1 327 30 // chunk_fwd.py:327:30 + mov.b64 {%r1815, %r1816}, %rd290; + cvt.rn.bf16x2.f32 %r1817, %r1816, %r1815; + .loc 1 267 14 // chunk_fwd.py:267:14 + mov.b64 %rd291, {%r1205, %r1206}; + sub.f32x2 %rd292, %rd271, %rd291; + .loc 1 327 30 // chunk_fwd.py:327:30 + mov.b64 {%r1818, %r1819}, %rd292; + cvt.rn.bf16x2.f32 %r1820, %r1819, %r1818; + .loc 1 267 14 // chunk_fwd.py:267:14 + mov.b64 %rd293, {%r1207, %r1208}; + sub.f32x2 %rd294, %rd271, %rd293; + .loc 1 327 30 // chunk_fwd.py:327:30 + mov.b64 {%r1821, %r1822}, %rd294; + cvt.rn.bf16x2.f32 %r1823, %r1822, %r1821; + .loc 1 327 20 // chunk_fwd.py:327:20 + bar.sync 0; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1767], {%r1814, %r1817, %r1820, %r1823}; + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x1.shared.b16 {%r1387}, [%r1764]; + // begin inline asm + @%p79 st.global.b32 [ %rd259 + 0 ], { %r1387 }; + // end inline asm + .loc 1 328 20 // chunk_fwd.py:328:20 + bar.sync 0; + cvt.rn.bf16x2.f32 %r1824, %r1183, %r1169; + cvt.rn.bf16x2.f32 %r1825, %r1918, %r1917; + cvt.rn.bf16x2.f32 %r1826, %r1920, %r1919; + cvt.rn.bf16x2.f32 %r1827, %r1184, %r1170; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r1767], {%r1824, %r1825, %r1826, %r1827}; + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x1.shared.b16 {%r1388}, [%r1764]; + // begin inline asm + @%p79 st.global.b32 [ %rd260 + 0 ], { %r1388 }; + // end inline asm +$L__BB0_1: // %common.ret + .loc 1 0 0 // chunk_fwd.py:0 + ret; +$L__tmp167: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py" + .file 3 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 6 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 823 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0x330 DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 99 // DW_AT_name +.b8 104 +.b8 117 +.b8 110 +.b8 107 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 111 +.b8 112 +.b8 115 +.b8 47 +.b8 103 +.b8 97 +.b8 116 +.b8 101 +.b8 100 +.b8 95 +.b8 100 +.b8 101 +.b8 108 +.b8 116 +.b8 97 +.b8 95 +.b8 114 +.b8 117 +.b8 108 +.b8 101 +.b8 0 +.b8 2 // Abbrev [2] 0x76:0x2e DW_TAG_subprogram +.b8 99 // DW_AT_name +.b8 104 +.b8 117 +.b8 110 +.b8 107 +.b8 95 +.b8 103 +.b8 97 +.b8 116 +.b8 101 +.b8 100 +.b8 95 +.b8 100 +.b8 101 +.b8 108 +.b8 116 +.b8 97 +.b8 95 +.b8 114 +.b8 117 +.b8 108 +.b8 101 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 107 +.b8 116 +.b8 95 +.b8 115 +.b8 111 +.b8 108 +.b8 118 +.b8 101 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0xa4:0x296 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 118 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0xb9:0x18 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp2 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 183 // DW_AT_call_line +.b8 70 // DW_AT_call_column +.b8 4 // Abbrev [4] 0xd1:0x18 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp3 // DW_AT_low_pc +.b64 $L__tmp4 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 184 // DW_AT_call_line +.b8 70 // DW_AT_call_column +.b8 4 // Abbrev [4] 0xe9:0x18 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp5 // DW_AT_low_pc +.b64 $L__tmp6 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 185 // DW_AT_call_line +.b8 70 // DW_AT_call_column +.b8 4 // Abbrev [4] 0x101:0x18 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp7 // DW_AT_low_pc +.b64 $L__tmp8 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 186 // DW_AT_call_line +.b8 70 // DW_AT_call_column +.b8 4 // Abbrev [4] 0x119:0x18 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp9 // DW_AT_low_pc +.b64 $L__tmp10 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 188 // DW_AT_call_line +.b8 64 // DW_AT_call_column +.b8 4 // Abbrev [4] 0x131:0x18 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp11 // DW_AT_low_pc +.b64 $L__tmp12 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 189 // DW_AT_call_line +.b8 64 // DW_AT_call_column +.b8 4 // Abbrev [4] 0x149:0x18 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp13 // DW_AT_low_pc +.b64 $L__tmp14 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 190 // DW_AT_call_line +.b8 64 // DW_AT_call_column +.b8 4 // Abbrev [4] 0x161:0x18 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp15 // DW_AT_low_pc +.b64 $L__tmp16 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 191 // DW_AT_call_line +.b8 64 // DW_AT_call_column +.b8 4 // Abbrev [4] 0x179:0x18 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp17 // DW_AT_low_pc +.b64 $L__tmp18 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 192 // DW_AT_call_line +.b8 64 // DW_AT_call_column +.b8 4 // Abbrev [4] 0x191:0x18 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp19 // DW_AT_low_pc +.b64 $L__tmp20 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 193 // DW_AT_call_line +.b8 64 // DW_AT_call_column +.b8 5 // Abbrev [5] 0x1a9:0x32 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp21 // DW_AT_low_pc +.b64 $L__tmp33 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 228 // DW_AT_call_line +.b8 66 // DW_AT_call_column +.b8 6 // Abbrev [6] 0x1c1:0x19 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp21 // DW_AT_low_pc +.b64 $L__tmp33 // DW_AT_high_pc +.b8 3 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 5 // Abbrev [5] 0x1db:0x32 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp34 // DW_AT_low_pc +.b64 $L__tmp45 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 230 // DW_AT_call_line +.b8 56 // DW_AT_call_column +.b8 6 // Abbrev [6] 0x1f3:0x19 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp34 // DW_AT_low_pc +.b64 $L__tmp45 // DW_AT_high_pc +.b8 3 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 5 // Abbrev [5] 0x20d:0x32 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp46 // DW_AT_low_pc +.b64 $L__tmp65 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 233 // DW_AT_call_line +.b8 66 // DW_AT_call_column +.b8 6 // Abbrev [6] 0x225:0x19 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp46 // DW_AT_low_pc +.b64 $L__tmp65 // DW_AT_high_pc +.b8 3 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 5 // Abbrev [5] 0x23f:0x32 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp66 // DW_AT_low_pc +.b64 $L__tmp85 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 235 // DW_AT_call_line +.b8 56 // DW_AT_call_column +.b8 6 // Abbrev [6] 0x257:0x19 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp66 // DW_AT_low_pc +.b64 $L__tmp85 // DW_AT_high_pc +.b8 3 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 5 // Abbrev [5] 0x271:0x32 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp86 // DW_AT_low_pc +.b64 $L__tmp105 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 238 // DW_AT_call_line +.b8 66 // DW_AT_call_column +.b8 6 // Abbrev [6] 0x289:0x19 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp86 // DW_AT_low_pc +.b64 $L__tmp105 // DW_AT_high_pc +.b8 3 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 5 // Abbrev [5] 0x2a3:0x32 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp106 // DW_AT_low_pc +.b64 $L__tmp125 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 240 // DW_AT_call_line +.b8 56 // DW_AT_call_column +.b8 6 // Abbrev [6] 0x2bb:0x19 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp106 // DW_AT_low_pc +.b64 $L__tmp125 // DW_AT_high_pc +.b8 3 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 5 // Abbrev [5] 0x2d5:0x32 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp126 // DW_AT_low_pc +.b64 $L__tmp146 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 243 // DW_AT_call_line +.b8 66 // DW_AT_call_column +.b8 6 // Abbrev [6] 0x2ed:0x19 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp126 // DW_AT_low_pc +.b64 $L__tmp146 // DW_AT_high_pc +.b8 3 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 5 // Abbrev [5] 0x307:0x32 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp147 // DW_AT_low_pc +.b64 $L__tmp166 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 245 // DW_AT_call_line +.b8 56 // DW_AT_call_column +.b8 6 // Abbrev [6] 0x31f:0x19 DW_TAG_inlined_subroutine +.b32 118 // DW_AT_abstract_origin +.b64 $L__tmp147 // DW_AT_low_pc +.b64 $L__tmp166 // DW_AT_high_pc +.b8 3 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.source b/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..b6993b6e33c471a1c8466dd1e35ffc9d5b77c03f --- /dev/null +++ b/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.source @@ -0,0 +1,2482 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":40:0) +#loc550 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:0) +#loc554 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc556 = loc(unknown) +#loc559 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc563 = loc("k"(#loc)) +#loc564 = loc("g"(#loc)) +#loc565 = loc("beta"(#loc)) +#loc566 = loc("A"(#loc)) +#loc567 = loc("T"(#loc)) +#loc1059 = loc("x"(#loc550)) +#loc1060 = loc("input"(#loc554)) +#loc1061 = loc("a"(#loc559)) +#loc1062 = loc("b"(#loc559)) +module { + tt.func public @chunk_gated_delta_rule_fwd_kkt_solve_kernel(%k: !tt.ptr {tt.divisibility = 16 : i32} loc("k"(#loc)), %g: !tt.ptr {tt.divisibility = 16 : i32} loc("g"(#loc)), %beta: !tt.ptr {tt.divisibility = 16 : i32} loc("beta"(#loc)), %A: !tt.ptr {tt.divisibility = 16 : i32} loc("A"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %0 = tt.get_program_id x : i32 loc(#loc1) + %1 = arith.extsi %0 : i32 to i64 loc(#loc2) + %2 = tt.get_program_id y : i32 loc(#loc3) + %3 = arith.extsi %2 : i32 to i64 loc(#loc4) + %c32_i32 = arith.constant 32 : i32 loc(#loc5) + %c32_i64 = arith.constant 32 : i64 loc(#loc5) + %4 = arith.divsi %3, %c32_i64 : i64 loc(#loc5) + %c32_i32_0 = arith.constant 32 : i32 loc(#loc6) + %c32_i64_1 = arith.constant 32 : i64 loc(#loc6) + %5 = arith.remsi %3, %c32_i64_1 : i64 loc(#loc6) + %6 = arith.extsi %T : i32 to i64 loc(#loc7) + %7 = arith.muli %4, %6 : i64 loc(#loc7) + %8 = arith.extsi %T : i32 to i64 loc(#loc8) + %9 = arith.muli %4, %8 : i64 loc(#loc8) + %10 = arith.extsi %T : i32 to i64 loc(#loc9) + %11 = arith.addi %9, %10 : i64 loc(#loc9) + %c64_i32 = arith.constant 64 : i32 loc(#loc10) + %c64_i64 = arith.constant 64 : i64 loc(#loc10) + %12 = arith.muli %1, %c64_i64 : i64 loc(#loc10) + %13 = arith.extsi %T : i32 to i64 loc(#loc11) + %14 = arith.cmpi sge, %12, %13 : i64 loc(#loc11) + cf.cond_br %14, ^bb1, ^bb2 loc(#loc11) + ^bb1: // pred: ^bb0 + tt.return loc(#loc12) + ^bb2: // pred: ^bb0 + cf.br ^bb4 loc(#loc) + ^bb3: // no predecessors + cf.br ^bb4 loc(#loc) + ^bb4: // 2 preds: ^bb2, ^bb3 + %i_tc0 = arith.constant 64 : i32 loc(#loc568) + %i_tc0_2 = arith.constant 64 : i64 loc(#loc568) + %i_tc0_3 = arith.muli %1, %i_tc0_2 : i64 loc(#loc568) + %i_tc1 = arith.constant 64 : i32 loc(#loc569) + %i_tc1_4 = arith.constant 64 : i64 loc(#loc569) + %i_tc1_5 = arith.muli %1, %i_tc1_4 : i64 loc(#loc569) + %i_tc1_6 = arith.constant 16 : i32 loc(#loc570) + %i_tc1_7 = arith.constant 16 : i64 loc(#loc570) + %i_tc1_8 = arith.addi %i_tc1_5, %i_tc1_7 : i64 loc(#loc570) + %i_tc2 = arith.constant 64 : i32 loc(#loc571) + %i_tc2_9 = arith.constant 64 : i64 loc(#loc571) + %i_tc2_10 = arith.muli %1, %i_tc2_9 : i64 loc(#loc571) + %i_tc2_11 = arith.constant 32 : i32 loc(#loc572) + %i_tc2_12 = arith.constant 32 : i64 loc(#loc572) + %i_tc2_13 = arith.addi %i_tc2_10, %i_tc2_12 : i64 loc(#loc572) + %i_tc3 = arith.constant 64 : i32 loc(#loc573) + %i_tc3_14 = arith.constant 64 : i64 loc(#loc573) + %i_tc3_15 = arith.muli %1, %i_tc3_14 : i64 loc(#loc573) + %i_tc3_16 = arith.constant 48 : i32 loc(#loc574) + %i_tc3_17 = arith.constant 48 : i64 loc(#loc574) + %i_tc3_18 = arith.addi %i_tc3_15, %i_tc3_17 : i64 loc(#loc574) + %k_19 = arith.constant 32 : i32 loc(#loc575) + %k_20 = arith.constant 32 : i64 loc(#loc575) + %k_21 = arith.muli %7, %k_20 : i64 loc(#loc575) + %k_22 = arith.constant 1 : i32 loc(#loc576) + %k_23 = arith.constant 1 : i64 loc(#loc576) + %k_24 = arith.divsi %5, %k_23 : i64 loc(#loc576) + %k_25 = arith.addi %k_21, %k_24 : i64 loc(#loc577) + %k_26 = arith.constant 128 : i32 loc(#loc578) + %k_27 = arith.constant 128 : i64 loc(#loc578) + %k_28 = arith.muli %k_25, %k_27 : i64 loc(#loc578) + %k_29 = tt.addptr %k, %k_28 : !tt.ptr, i64 loc(#loc579) + %A_30 = arith.constant 32 : i32 loc(#loc580) + %A_31 = arith.constant 32 : i64 loc(#loc580) + %A_32 = arith.muli %7, %A_31 : i64 loc(#loc580) + %A_33 = arith.addi %A_32, %5 : i64 loc(#loc581) + %A_34 = arith.constant 64 : i32 loc(#loc582) + %A_35 = arith.constant 64 : i64 loc(#loc582) + %A_36 = arith.muli %A_33, %A_35 : i64 loc(#loc582) + %A_37 = tt.addptr %A, %A_36 : !tt.ptr, i64 loc(#loc583) + %o_i = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32> loc(#loc584) + %m_tc0 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc585) + %m_tc0_38 = tt.splat %i_tc0_3 : i64 -> tensor<16xi64> loc(#loc585) + %m_tc0_39 = arith.addi %m_tc0_38, %m_tc0 : tensor<16xi64> loc(#loc585) + %m_tc0_40 = arith.extsi %T : i32 to i64 loc(#loc586) + %m_tc0_41 = tt.splat %m_tc0_40 : i64 -> tensor<16xi64> loc(#loc586) + %m_tc0_42 = arith.cmpi slt, %m_tc0_39, %m_tc0_41 : tensor<16xi64> loc(#loc586) + %m_tc1 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc587) + %m_tc1_43 = tt.splat %i_tc1_8 : i64 -> tensor<16xi64> loc(#loc587) + %m_tc1_44 = arith.addi %m_tc1_43, %m_tc1 : tensor<16xi64> loc(#loc587) + %m_tc1_45 = arith.extsi %T : i32 to i64 loc(#loc588) + %m_tc1_46 = tt.splat %m_tc1_45 : i64 -> tensor<16xi64> loc(#loc588) + %m_tc1_47 = arith.cmpi slt, %m_tc1_44, %m_tc1_46 : tensor<16xi64> loc(#loc588) + %m_tc2 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc589) + %m_tc2_48 = tt.splat %i_tc2_13 : i64 -> tensor<16xi64> loc(#loc589) + %m_tc2_49 = arith.addi %m_tc2_48, %m_tc2 : tensor<16xi64> loc(#loc589) + %m_tc2_50 = arith.extsi %T : i32 to i64 loc(#loc590) + %m_tc2_51 = tt.splat %m_tc2_50 : i64 -> tensor<16xi64> loc(#loc590) + %m_tc2_52 = arith.cmpi slt, %m_tc2_49, %m_tc2_51 : tensor<16xi64> loc(#loc590) + %m_tc3 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc591) + %m_tc3_53 = tt.splat %i_tc3_18 : i64 -> tensor<16xi64> loc(#loc591) + %m_tc3_54 = arith.addi %m_tc3_53, %m_tc3 : tensor<16xi64> loc(#loc591) + %m_tc3_55 = arith.extsi %T : i32 to i64 loc(#loc592) + %m_tc3_56 = tt.splat %m_tc3_55 : i64 -> tensor<16xi64> loc(#loc592) + %m_tc3_57 = arith.cmpi slt, %m_tc3_54, %m_tc3_56 : tensor<16xi64> loc(#loc592) + %p_b0 = arith.constant 32 : i32 loc(#loc593) + %p_b0_58 = arith.constant 32 : i64 loc(#loc593) + %p_b0_59 = arith.muli %7, %p_b0_58 : i64 loc(#loc593) + %p_b0_60 = tt.addptr %beta, %p_b0_59 : !tt.ptr, i64 loc(#loc594) + %p_b0_61 = tt.addptr %p_b0_60, %5 : !tt.ptr, i64 loc(#loc595) + %p_b0_62 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc596) + %p_b0_63 = tt.splat %i_tc0_3 : i64 -> tensor<16xi64> loc(#loc596) + %p_b0_64 = arith.addi %p_b0_63, %p_b0_62 : tensor<16xi64> loc(#loc596) + %p_b0_65 = arith.constant 32 : i32 loc(#loc597) + %p_b0_66 = arith.constant 32 : i64 loc(#loc597) + %p_b0_67 = arith.constant dense<32> : tensor<16xi64> loc(#loc597) + %p_b0_68 = arith.muli %p_b0_64, %p_b0_67 : tensor<16xi64> loc(#loc597) + %p_b0_69 = tt.splat %p_b0_61 : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc598) + %p_b0_70 = tt.addptr %p_b0_69, %p_b0_68 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc598) + %p_b1 = arith.constant 32 : i32 loc(#loc599) + %p_b1_71 = arith.constant 32 : i64 loc(#loc599) + %p_b1_72 = arith.muli %7, %p_b1_71 : i64 loc(#loc599) + %p_b1_73 = tt.addptr %beta, %p_b1_72 : !tt.ptr, i64 loc(#loc600) + %p_b1_74 = tt.addptr %p_b1_73, %5 : !tt.ptr, i64 loc(#loc601) + %p_b1_75 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc602) + %p_b1_76 = tt.splat %i_tc1_8 : i64 -> tensor<16xi64> loc(#loc602) + %p_b1_77 = arith.addi %p_b1_76, %p_b1_75 : tensor<16xi64> loc(#loc602) + %p_b1_78 = arith.constant 32 : i32 loc(#loc603) + %p_b1_79 = arith.constant 32 : i64 loc(#loc603) + %p_b1_80 = arith.constant dense<32> : tensor<16xi64> loc(#loc603) + %p_b1_81 = arith.muli %p_b1_77, %p_b1_80 : tensor<16xi64> loc(#loc603) + %p_b1_82 = tt.splat %p_b1_74 : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc604) + %p_b1_83 = tt.addptr %p_b1_82, %p_b1_81 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc604) + %p_b2 = arith.constant 32 : i32 loc(#loc605) + %p_b2_84 = arith.constant 32 : i64 loc(#loc605) + %p_b2_85 = arith.muli %7, %p_b2_84 : i64 loc(#loc605) + %p_b2_86 = tt.addptr %beta, %p_b2_85 : !tt.ptr, i64 loc(#loc606) + %p_b2_87 = tt.addptr %p_b2_86, %5 : !tt.ptr, i64 loc(#loc607) + %p_b2_88 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc608) + %p_b2_89 = tt.splat %i_tc2_13 : i64 -> tensor<16xi64> loc(#loc608) + %p_b2_90 = arith.addi %p_b2_89, %p_b2_88 : tensor<16xi64> loc(#loc608) + %p_b2_91 = arith.constant 32 : i32 loc(#loc609) + %p_b2_92 = arith.constant 32 : i64 loc(#loc609) + %p_b2_93 = arith.constant dense<32> : tensor<16xi64> loc(#loc609) + %p_b2_94 = arith.muli %p_b2_90, %p_b2_93 : tensor<16xi64> loc(#loc609) + %p_b2_95 = tt.splat %p_b2_87 : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc610) + %p_b2_96 = tt.addptr %p_b2_95, %p_b2_94 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc610) + %p_b3 = arith.constant 32 : i32 loc(#loc611) + %p_b3_97 = arith.constant 32 : i64 loc(#loc611) + %p_b3_98 = arith.muli %7, %p_b3_97 : i64 loc(#loc611) + %p_b3_99 = tt.addptr %beta, %p_b3_98 : !tt.ptr, i64 loc(#loc612) + %p_b3_100 = tt.addptr %p_b3_99, %5 : !tt.ptr, i64 loc(#loc613) + %p_b3_101 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc614) + %p_b3_102 = tt.splat %i_tc3_18 : i64 -> tensor<16xi64> loc(#loc614) + %p_b3_103 = arith.addi %p_b3_102, %p_b3_101 : tensor<16xi64> loc(#loc614) + %p_b3_104 = arith.constant 32 : i32 loc(#loc615) + %p_b3_105 = arith.constant 32 : i64 loc(#loc615) + %p_b3_106 = arith.constant dense<32> : tensor<16xi64> loc(#loc615) + %p_b3_107 = arith.muli %p_b3_103, %p_b3_106 : tensor<16xi64> loc(#loc615) + %p_b3_108 = tt.splat %p_b3_100 : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc616) + %p_b3_109 = tt.addptr %p_b3_108, %p_b3_107 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc616) + %b_b0 = arith.constant 0.000000e+00 : f32 loc(#loc617) + %b_b0_110 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc617) + %b_b0_111 = arith.truncf %b_b0_110 : tensor<16xf32> to tensor<16xbf16> loc(#loc617) + %b_b0_112 = tt.load %p_b0_70, %m_tc0_42, %b_b0_111 : tensor<16x!tt.ptr> loc(#loc617) + %b_b0_113 = arith.extf %b_b0_112 : tensor<16xbf16> to tensor<16xf32> loc(#loc618) + %b_b1 = arith.constant 0.000000e+00 : f32 loc(#loc619) + %b_b1_114 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc619) + %b_b1_115 = arith.truncf %b_b1_114 : tensor<16xf32> to tensor<16xbf16> loc(#loc619) + %b_b1_116 = tt.load %p_b1_83, %m_tc1_47, %b_b1_115 : tensor<16x!tt.ptr> loc(#loc619) + %b_b1_117 = arith.extf %b_b1_116 : tensor<16xbf16> to tensor<16xf32> loc(#loc620) + %b_b2 = arith.constant 0.000000e+00 : f32 loc(#loc621) + %b_b2_118 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc621) + %b_b2_119 = arith.truncf %b_b2_118 : tensor<16xf32> to tensor<16xbf16> loc(#loc621) + %b_b2_120 = tt.load %p_b2_96, %m_tc2_52, %b_b2_119 : tensor<16x!tt.ptr> loc(#loc621) + %b_b2_121 = arith.extf %b_b2_120 : tensor<16xbf16> to tensor<16xf32> loc(#loc622) + %b_b3 = arith.constant 0.000000e+00 : f32 loc(#loc623) + %b_b3_122 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc623) + %b_b3_123 = arith.truncf %b_b3_122 : tensor<16xf32> to tensor<16xbf16> loc(#loc623) + %b_b3_124 = tt.load %p_b3_109, %m_tc3_57, %b_b3_123 : tensor<16x!tt.ptr> loc(#loc623) + %b_b3_125 = arith.extf %b_b3_124 : tensor<16xbf16> to tensor<16xf32> loc(#loc624) + %p_g0 = arith.constant 32 : i32 loc(#loc625) + %p_g0_126 = arith.constant 32 : i64 loc(#loc625) + %p_g0_127 = arith.muli %7, %p_g0_126 : i64 loc(#loc625) + %p_g0_128 = tt.addptr %g, %p_g0_127 : !tt.ptr, i64 loc(#loc626) + %p_g0_129 = tt.addptr %p_g0_128, %5 : !tt.ptr, i64 loc(#loc627) + %p_g0_130 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc628) + %p_g0_131 = tt.splat %i_tc0_3 : i64 -> tensor<16xi64> loc(#loc628) + %p_g0_132 = arith.addi %p_g0_131, %p_g0_130 : tensor<16xi64> loc(#loc628) + %p_g0_133 = arith.constant 32 : i32 loc(#loc629) + %p_g0_134 = arith.constant 32 : i64 loc(#loc629) + %p_g0_135 = arith.constant dense<32> : tensor<16xi64> loc(#loc629) + %p_g0_136 = arith.muli %p_g0_132, %p_g0_135 : tensor<16xi64> loc(#loc629) + %p_g0_137 = tt.splat %p_g0_129 : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc630) + %p_g0_138 = tt.addptr %p_g0_137, %p_g0_136 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc630) + %p_g1 = arith.constant 32 : i32 loc(#loc631) + %p_g1_139 = arith.constant 32 : i64 loc(#loc631) + %p_g1_140 = arith.muli %7, %p_g1_139 : i64 loc(#loc631) + %p_g1_141 = tt.addptr %g, %p_g1_140 : !tt.ptr, i64 loc(#loc632) + %p_g1_142 = tt.addptr %p_g1_141, %5 : !tt.ptr, i64 loc(#loc633) + %p_g1_143 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc634) + %p_g1_144 = tt.splat %i_tc1_8 : i64 -> tensor<16xi64> loc(#loc634) + %p_g1_145 = arith.addi %p_g1_144, %p_g1_143 : tensor<16xi64> loc(#loc634) + %p_g1_146 = arith.constant 32 : i32 loc(#loc635) + %p_g1_147 = arith.constant 32 : i64 loc(#loc635) + %p_g1_148 = arith.constant dense<32> : tensor<16xi64> loc(#loc635) + %p_g1_149 = arith.muli %p_g1_145, %p_g1_148 : tensor<16xi64> loc(#loc635) + %p_g1_150 = tt.splat %p_g1_142 : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc636) + %p_g1_151 = tt.addptr %p_g1_150, %p_g1_149 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc636) + %p_g2 = arith.constant 32 : i32 loc(#loc637) + %p_g2_152 = arith.constant 32 : i64 loc(#loc637) + %p_g2_153 = arith.muli %7, %p_g2_152 : i64 loc(#loc637) + %p_g2_154 = tt.addptr %g, %p_g2_153 : !tt.ptr, i64 loc(#loc638) + %p_g2_155 = tt.addptr %p_g2_154, %5 : !tt.ptr, i64 loc(#loc639) + %p_g2_156 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc640) + %p_g2_157 = tt.splat %i_tc2_13 : i64 -> tensor<16xi64> loc(#loc640) + %p_g2_158 = arith.addi %p_g2_157, %p_g2_156 : tensor<16xi64> loc(#loc640) + %p_g2_159 = arith.constant 32 : i32 loc(#loc641) + %p_g2_160 = arith.constant 32 : i64 loc(#loc641) + %p_g2_161 = arith.constant dense<32> : tensor<16xi64> loc(#loc641) + %p_g2_162 = arith.muli %p_g2_158, %p_g2_161 : tensor<16xi64> loc(#loc641) + %p_g2_163 = tt.splat %p_g2_155 : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc642) + %p_g2_164 = tt.addptr %p_g2_163, %p_g2_162 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc642) + %p_g3 = arith.constant 32 : i32 loc(#loc643) + %p_g3_165 = arith.constant 32 : i64 loc(#loc643) + %p_g3_166 = arith.muli %7, %p_g3_165 : i64 loc(#loc643) + %p_g3_167 = tt.addptr %g, %p_g3_166 : !tt.ptr, i64 loc(#loc644) + %p_g3_168 = tt.addptr %p_g3_167, %5 : !tt.ptr, i64 loc(#loc645) + %p_g3_169 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc646) + %p_g3_170 = tt.splat %i_tc3_18 : i64 -> tensor<16xi64> loc(#loc646) + %p_g3_171 = arith.addi %p_g3_170, %p_g3_169 : tensor<16xi64> loc(#loc646) + %p_g3_172 = arith.constant 32 : i32 loc(#loc647) + %p_g3_173 = arith.constant 32 : i64 loc(#loc647) + %p_g3_174 = arith.constant dense<32> : tensor<16xi64> loc(#loc647) + %p_g3_175 = arith.muli %p_g3_171, %p_g3_174 : tensor<16xi64> loc(#loc647) + %p_g3_176 = tt.splat %p_g3_168 : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc648) + %p_g3_177 = tt.addptr %p_g3_176, %p_g3_175 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc648) + %b_g0 = arith.constant 0.000000e+00 : f32 loc(#loc649) + %b_g0_178 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc649) + %b_g0_179 = tt.load %p_g0_138, %m_tc0_42, %b_g0_178 : tensor<16x!tt.ptr> loc(#loc649) + %b_g1 = arith.constant 0.000000e+00 : f32 loc(#loc650) + %b_g1_180 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc650) + %b_g1_181 = tt.load %p_g1_151, %m_tc1_47, %b_g1_180 : tensor<16x!tt.ptr> loc(#loc650) + %b_g2 = arith.constant 0.000000e+00 : f32 loc(#loc651) + %b_g2_182 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc651) + %b_g2_183 = tt.load %p_g2_164, %m_tc2_52, %b_g2_182 : tensor<16x!tt.ptr> loc(#loc651) + %b_g3 = arith.constant 0.000000e+00 : f32 loc(#loc652) + %b_g3_184 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc652) + %b_g3_185 = tt.load %p_g3_177, %m_tc3_57, %b_g3_184 : tensor<16x!tt.ptr> loc(#loc652) + %b_A00 = tt.call @triton.language.standard.zeros__Tc16_c16T_cfp32() : () -> tensor<16x16xf32> loc(#loc653) + %b_A11 = tt.call @triton.language.standard.zeros__Tc16_c16T_cfp32() : () -> tensor<16x16xf32> loc(#loc1063) + %b_A22 = tt.call @triton.language.standard.zeros__Tc16_c16T_cfp32() : () -> tensor<16x16xf32> loc(#loc1084) + %b_A33 = tt.call @triton.language.standard.zeros__Tc16_c16T_cfp32() : () -> tensor<16x16xf32> loc(#loc1094) + %b_A10 = tt.call @triton.language.standard.zeros__Tc16_c16T_cfp32() : () -> tensor<16x16xf32> loc(#loc1066) + %b_A20 = tt.call @triton.language.standard.zeros__Tc16_c16T_cfp32() : () -> tensor<16x16xf32> loc(#loc1086) + %b_A21 = tt.call @triton.language.standard.zeros__Tc16_c16T_cfp32() : () -> tensor<16x16xf32> loc(#loc1087) + %b_A30 = tt.call @triton.language.standard.zeros__Tc16_c16T_cfp32() : () -> tensor<16x16xf32> loc(#loc1095) + %b_A31 = tt.call @triton.language.standard.zeros__Tc16_c16T_cfp32() : () -> tensor<16x16xf32> loc(#loc1096) + %b_A32 = tt.call @triton.language.standard.zeros__Tc16_c16T_cfp32() : () -> tensor<16x16xf32> loc(#loc1097) + tt.call @triton.language.standard.cdiv__c128_c64() : () -> () loc(#loc108) + tt.call @triton.language.standard.cdiv__c128_c64() : () -> () loc(#loc108) + %c0_i32 = arith.constant 0 : i32 loc(#loc109) + %c2_i32 = arith.constant 2 : i32 loc(#loc109) + %c1_i32 = arith.constant 1 : i32 loc(#loc109) + %15 = arith.bitcast %c0_i32 : i32 to i32 loc(#loc109) + %16 = arith.bitcast %c2_i32 : i32 to i32 loc(#loc109) + %17 = arith.bitcast %c1_i32 : i32 to i32 loc(#loc109) + %18 = ub.poison : i32 loc(#loc109) + %b_A32_186:10 = scf.for %i_k = %15 to %16 step %17 iter_args(%b_A00_852 = %b_A00, %b_A11_853 = %b_A11, %b_A22_854 = %b_A22, %b_A33_855 = %b_A33, %b_A10_856 = %b_A10, %b_A20_857 = %b_A20, %b_A21_858 = %b_A21, %b_A30_859 = %b_A30, %b_A31_860 = %b_A31, %b_A32_861 = %b_A32) -> (tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>) : i32 { + %o_k = arith.constant 64 : i32 loc(#loc664) + %o_k_862 = arith.constant 64 : i32 loc(#loc664) + %o_k_863 = arith.extsi %i_k : i32 to i64 loc(#loc664) + %o_k_864 = arith.extsi %o_k_862 : i32 to i64 loc(#loc664) + %o_k_865 = arith.muli %o_k_863, %o_k_864 : i64 loc(#loc664) + %o_k_866 = arith.constant 2147483647 : i64 loc(#loc664) + %o_k_867 = arith.constant -2147483648 : i64 loc(#loc664) + %o_k_868 = arith.cmpi sle, %o_k_865, %o_k_866 : i64 loc(#loc664) + %o_k_869 = arith.cmpi sge, %o_k_865, %o_k_867 : i64 loc(#loc664) + %o_k_870 = arith.andi %o_k_868, %o_k_869 : i1 loc(#loc664) + %o_k_871 = arith.muli %i_k, %o_k_862 : i32 loc(#loc664) + %o_k_872 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc665) + %o_k_873 = tt.splat %o_k_871 : i32 -> tensor<64xi32> loc(#loc666) + %o_k_874 = arith.extsi %o_k_873 : tensor<64xi32> to tensor<64xi64> loc(#loc666) + %o_k_875 = arith.extsi %o_k_872 : tensor<64xi32> to tensor<64xi64> loc(#loc666) + %o_k_876 = arith.addi %o_k_874, %o_k_875 : tensor<64xi64> loc(#loc666) + %o_k_877 = arith.constant 2147483647 : i64 loc(#loc666) + %o_k_878 = arith.constant -2147483648 : i64 loc(#loc666) + %o_k_879 = arith.constant dense<2147483647> : tensor<64xi64> loc(#loc666) + %o_k_880 = arith.cmpi sle, %o_k_876, %o_k_879 : tensor<64xi64> loc(#loc666) + %o_k_881 = arith.constant dense<-2147483648> : tensor<64xi64> loc(#loc666) + %o_k_882 = arith.cmpi sge, %o_k_876, %o_k_881 : tensor<64xi64> loc(#loc666) + %o_k_883 = arith.andi %o_k_880, %o_k_882 : tensor<64xi1> loc(#loc666) + %o_k_884 = arith.addi %o_k_873, %o_k_872 : tensor<64xi32> loc(#loc666) + %p_k0 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc667) + %p_k0_885 = tt.splat %i_tc0_3 : i64 -> tensor<16xi64> loc(#loc667) + %p_k0_886 = arith.addi %p_k0_885, %p_k0 : tensor<16xi64> loc(#loc667) + %p_k0_887 = tt.expand_dims %p_k0_886 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc668) + %p_k0_888 = arith.constant 4096 : i32 loc(#loc669) + %p_k0_889 = arith.constant 4096 : i64 loc(#loc669) + %p_k0_890 = arith.constant dense<4096> : tensor<16x1xi64> loc(#loc669) + %p_k0_891 = arith.muli %p_k0_887, %p_k0_890 : tensor<16x1xi64> loc(#loc669) + %p_k0_892 = tt.splat %k_29 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc670) + %p_k0_893 = tt.addptr %p_k0_892, %p_k0_891 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc670) + %p_k0_894 = tt.expand_dims %o_k_884 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc671) + %p_k0_895 = tt.broadcast %p_k0_893 : tensor<16x1x!tt.ptr> -> tensor<16x64x!tt.ptr> loc(#loc672) + %p_k0_896 = tt.broadcast %p_k0_894 : tensor<1x64xi32> -> tensor<16x64xi32> loc(#loc672) + %p_k0_897 = tt.addptr %p_k0_895, %p_k0_896 : tensor<16x64x!tt.ptr>, tensor<16x64xi32> loc(#loc672) + %b_k0 = tt.expand_dims %m_tc0_42 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc673) + %b_k0_898 = tt.expand_dims %o_k_884 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc674) + %b_k0_899 = arith.constant 128 : i32 loc(#loc675) + %b_k0_900 = arith.constant dense<128> : tensor<1x64xi32> loc(#loc675) + %b_k0_901 = arith.cmpi slt, %b_k0_898, %b_k0_900 : tensor<1x64xi32> loc(#loc675) + %b_k0_902 = tt.broadcast %b_k0 : tensor<16x1xi1> -> tensor<16x64xi1> loc(#loc676) + %b_k0_903 = tt.broadcast %b_k0_901 : tensor<1x64xi1> -> tensor<16x64xi1> loc(#loc676) + %b_k0_904 = arith.andi %b_k0_902, %b_k0_903 : tensor<16x64xi1> loc(#loc676) + %b_k0_905 = arith.constant 0.000000e+00 : f32 loc(#loc677) + %b_k0_906 = arith.constant dense<0.000000e+00> : tensor<16x64xf32> loc(#loc677) + %b_k0_907 = arith.truncf %b_k0_906 : tensor<16x64xf32> to tensor<16x64xbf16> loc(#loc677) + %b_k0_908 = tt.load %p_k0_897, %b_k0_904, %b_k0_907 : tensor<16x64x!tt.ptr> loc(#loc677) + %b_A00_909 = tt.trans %b_k0_908 {order = array} : tensor<16x64xbf16> -> tensor<64x16xbf16> loc(#loc678) + %b_A00_910 = arith.constant 0.000000e+00 : f32 loc(#loc679) + %b_A00_911 = tt.dot %b_k0_908, %b_A00_909, %b_A00_852, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc679) + %61 = arith.extsi %T : i32 to i64 loc(#loc126) + %62 = arith.cmpi slt, %i_tc1_8, %61 : i64 loc(#loc126) + %63:9 = scf.if %62 -> (tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>) { + %p_k1 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc680) + %p_k1_912 = tt.splat %i_tc1_8 : i64 -> tensor<16xi64> loc(#loc680) + %p_k1_913 = arith.addi %p_k1_912, %p_k1 : tensor<16xi64> loc(#loc680) + %p_k1_914 = tt.expand_dims %p_k1_913 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc681) + %p_k1_915 = arith.constant 4096 : i32 loc(#loc682) + %p_k1_916 = arith.constant 4096 : i64 loc(#loc682) + %p_k1_917 = arith.constant dense<4096> : tensor<16x1xi64> loc(#loc682) + %p_k1_918 = arith.muli %p_k1_914, %p_k1_917 : tensor<16x1xi64> loc(#loc682) + %p_k1_919 = tt.splat %k_29 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc683) + %p_k1_920 = tt.addptr %p_k1_919, %p_k1_918 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc683) + %p_k1_921 = tt.expand_dims %o_k_884 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc684) + %p_k1_922 = tt.broadcast %p_k1_920 : tensor<16x1x!tt.ptr> -> tensor<16x64x!tt.ptr> loc(#loc685) + %p_k1_923 = tt.broadcast %p_k1_921 : tensor<1x64xi32> -> tensor<16x64xi32> loc(#loc685) + %p_k1_924 = tt.addptr %p_k1_922, %p_k1_923 : tensor<16x64x!tt.ptr>, tensor<16x64xi32> loc(#loc685) + %b_k1 = tt.expand_dims %m_tc1_47 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc686) + %b_k1_925 = tt.expand_dims %o_k_884 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc687) + %b_k1_926 = arith.constant 128 : i32 loc(#loc688) + %b_k1_927 = arith.constant dense<128> : tensor<1x64xi32> loc(#loc688) + %b_k1_928 = arith.cmpi slt, %b_k1_925, %b_k1_927 : tensor<1x64xi32> loc(#loc688) + %b_k1_929 = tt.broadcast %b_k1 : tensor<16x1xi1> -> tensor<16x64xi1> loc(#loc689) + %b_k1_930 = tt.broadcast %b_k1_928 : tensor<1x64xi1> -> tensor<16x64xi1> loc(#loc689) + %b_k1_931 = arith.andi %b_k1_929, %b_k1_930 : tensor<16x64xi1> loc(#loc689) + %b_k1_932 = arith.constant 0.000000e+00 : f32 loc(#loc690) + %b_k1_933 = arith.constant dense<0.000000e+00> : tensor<16x64xf32> loc(#loc690) + %b_k1_934 = arith.truncf %b_k1_933 : tensor<16x64xf32> to tensor<16x64xbf16> loc(#loc690) + %b_k1_935 = tt.load %p_k1_924, %b_k1_931, %b_k1_934 : tensor<16x64x!tt.ptr> loc(#loc690) + %b_A11_936 = tt.trans %b_k1_935 {order = array} : tensor<16x64xbf16> -> tensor<64x16xbf16> loc(#loc691) + %b_A11_937 = arith.constant 0.000000e+00 : f32 loc(#loc692) + %b_A11_938 = tt.dot %b_k1_935, %b_A11_936, %b_A11_853, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc1073) + %b_A10_939 = tt.trans %b_k0_908 {order = array} : tensor<16x64xbf16> -> tensor<64x16xbf16> loc(#loc693) + %b_A10_940 = arith.constant 0.000000e+00 : f32 loc(#loc694) + %b_A10_941 = tt.dot %b_k1_935, %b_A10_939, %b_A10_856, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc1074) + %64 = arith.extsi %T : i32 to i64 loc(#loc143) + %65 = arith.cmpi slt, %i_tc2_13, %64 : i64 loc(#loc143) + %b_A32_942:7 = scf.if %65 -> (tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>) { + %p_k2 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc696) + %p_k2_943 = tt.splat %i_tc2_13 : i64 -> tensor<16xi64> loc(#loc696) + %p_k2_944 = arith.addi %p_k2_943, %p_k2 : tensor<16xi64> loc(#loc696) + %p_k2_945 = tt.expand_dims %p_k2_944 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc697) + %p_k2_946 = arith.constant 4096 : i32 loc(#loc698) + %p_k2_947 = arith.constant 4096 : i64 loc(#loc698) + %p_k2_948 = arith.constant dense<4096> : tensor<16x1xi64> loc(#loc698) + %p_k2_949 = arith.muli %p_k2_945, %p_k2_948 : tensor<16x1xi64> loc(#loc698) + %p_k2_950 = tt.splat %k_29 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc699) + %p_k2_951 = tt.addptr %p_k2_950, %p_k2_949 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc699) + %p_k2_952 = tt.expand_dims %o_k_884 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc700) + %p_k2_953 = tt.broadcast %p_k2_951 : tensor<16x1x!tt.ptr> -> tensor<16x64x!tt.ptr> loc(#loc701) + %p_k2_954 = tt.broadcast %p_k2_952 : tensor<1x64xi32> -> tensor<16x64xi32> loc(#loc701) + %p_k2_955 = tt.addptr %p_k2_953, %p_k2_954 : tensor<16x64x!tt.ptr>, tensor<16x64xi32> loc(#loc701) + %b_k2 = tt.expand_dims %m_tc2_52 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc702) + %b_k2_956 = tt.expand_dims %o_k_884 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc703) + %b_k2_957 = arith.constant 128 : i32 loc(#loc704) + %b_k2_958 = arith.constant dense<128> : tensor<1x64xi32> loc(#loc704) + %b_k2_959 = arith.cmpi slt, %b_k2_956, %b_k2_958 : tensor<1x64xi32> loc(#loc704) + %b_k2_960 = tt.broadcast %b_k2 : tensor<16x1xi1> -> tensor<16x64xi1> loc(#loc705) + %b_k2_961 = tt.broadcast %b_k2_959 : tensor<1x64xi1> -> tensor<16x64xi1> loc(#loc705) + %b_k2_962 = arith.andi %b_k2_960, %b_k2_961 : tensor<16x64xi1> loc(#loc705) + %b_k2_963 = arith.constant 0.000000e+00 : f32 loc(#loc706) + %b_k2_964 = arith.constant dense<0.000000e+00> : tensor<16x64xf32> loc(#loc706) + %b_k2_965 = arith.truncf %b_k2_964 : tensor<16x64xf32> to tensor<16x64xbf16> loc(#loc706) + %b_k2_966 = tt.load %p_k2_955, %b_k2_962, %b_k2_965 : tensor<16x64x!tt.ptr> loc(#loc706) + %b_A22_967 = tt.trans %b_k2_966 {order = array} : tensor<16x64xbf16> -> tensor<64x16xbf16> loc(#loc707) + %b_A22_968 = arith.constant 0.000000e+00 : f32 loc(#loc708) + %b_A22_969 = tt.dot %b_k2_966, %b_A22_967, %b_A22_854, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc1076) + %b_A20_970 = tt.trans %b_k0_908 {order = array} : tensor<16x64xbf16> -> tensor<64x16xbf16> loc(#loc709) + %b_A20_971 = arith.constant 0.000000e+00 : f32 loc(#loc710) + %b_A20_972 = tt.dot %b_k2_966, %b_A20_970, %b_A20_857, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc1077) + %b_A21_973 = tt.trans %b_k1_935 {order = array} : tensor<16x64xbf16> -> tensor<64x16xbf16> loc(#loc711) + %b_A21_974 = arith.constant 0.000000e+00 : f32 loc(#loc712) + %b_A21_975 = tt.dot %b_k2_966, %b_A21_973, %b_A21_858, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc1078) + %66 = arith.extsi %T : i32 to i64 loc(#loc162) + %67 = arith.cmpi slt, %i_tc3_18, %66 : i64 loc(#loc162) + %b_A32_976:4 = scf.if %67 -> (tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>) { + %p_k3 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc714) + %p_k3_977 = tt.splat %i_tc3_18 : i64 -> tensor<16xi64> loc(#loc714) + %p_k3_978 = arith.addi %p_k3_977, %p_k3 : tensor<16xi64> loc(#loc714) + %p_k3_979 = tt.expand_dims %p_k3_978 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc715) + %p_k3_980 = arith.constant 4096 : i32 loc(#loc716) + %p_k3_981 = arith.constant 4096 : i64 loc(#loc716) + %p_k3_982 = arith.constant dense<4096> : tensor<16x1xi64> loc(#loc716) + %p_k3_983 = arith.muli %p_k3_979, %p_k3_982 : tensor<16x1xi64> loc(#loc716) + %p_k3_984 = tt.splat %k_29 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc717) + %p_k3_985 = tt.addptr %p_k3_984, %p_k3_983 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc717) + %p_k3_986 = tt.expand_dims %o_k_884 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc718) + %p_k3_987 = tt.broadcast %p_k3_985 : tensor<16x1x!tt.ptr> -> tensor<16x64x!tt.ptr> loc(#loc719) + %p_k3_988 = tt.broadcast %p_k3_986 : tensor<1x64xi32> -> tensor<16x64xi32> loc(#loc719) + %p_k3_989 = tt.addptr %p_k3_987, %p_k3_988 : tensor<16x64x!tt.ptr>, tensor<16x64xi32> loc(#loc719) + %b_k3 = tt.expand_dims %m_tc3_57 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc720) + %b_k3_990 = tt.expand_dims %o_k_884 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc721) + %b_k3_991 = arith.constant 128 : i32 loc(#loc722) + %b_k3_992 = arith.constant dense<128> : tensor<1x64xi32> loc(#loc722) + %b_k3_993 = arith.cmpi slt, %b_k3_990, %b_k3_992 : tensor<1x64xi32> loc(#loc722) + %b_k3_994 = tt.broadcast %b_k3 : tensor<16x1xi1> -> tensor<16x64xi1> loc(#loc723) + %b_k3_995 = tt.broadcast %b_k3_993 : tensor<1x64xi1> -> tensor<16x64xi1> loc(#loc723) + %b_k3_996 = arith.andi %b_k3_994, %b_k3_995 : tensor<16x64xi1> loc(#loc723) + %b_k3_997 = arith.constant 0.000000e+00 : f32 loc(#loc724) + %b_k3_998 = arith.constant dense<0.000000e+00> : tensor<16x64xf32> loc(#loc724) + %b_k3_999 = arith.truncf %b_k3_998 : tensor<16x64xf32> to tensor<16x64xbf16> loc(#loc724) + %b_k3_1000 = tt.load %p_k3_989, %b_k3_996, %b_k3_999 : tensor<16x64x!tt.ptr> loc(#loc724) + %b_A33_1001 = tt.trans %b_k3_1000 {order = array} : tensor<16x64xbf16> -> tensor<64x16xbf16> loc(#loc725) + %b_A33_1002 = arith.constant 0.000000e+00 : f32 loc(#loc726) + %b_A33_1003 = tt.dot %b_k3_1000, %b_A33_1001, %b_A33_855, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc1080) + %b_A30_1004 = tt.trans %b_k0_908 {order = array} : tensor<16x64xbf16> -> tensor<64x16xbf16> loc(#loc727) + %b_A30_1005 = arith.constant 0.000000e+00 : f32 loc(#loc728) + %b_A30_1006 = tt.dot %b_k3_1000, %b_A30_1004, %b_A30_859, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc1081) + %b_A31_1007 = tt.trans %b_k1_935 {order = array} : tensor<16x64xbf16> -> tensor<64x16xbf16> loc(#loc729) + %b_A31_1008 = arith.constant 0.000000e+00 : f32 loc(#loc730) + %b_A31_1009 = tt.dot %b_k3_1000, %b_A31_1007, %b_A31_860, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc1082) + %b_A32_1010 = tt.trans %b_k2_966 {order = array} : tensor<16x64xbf16> -> tensor<64x16xbf16> loc(#loc731) + %b_A32_1011 = arith.constant 0.000000e+00 : f32 loc(#loc732) + %b_A32_1012 = tt.dot %b_k3_1000, %b_A32_1010, %b_A32_861, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc1083) + scf.yield %b_A33_1003, %b_A30_1006, %b_A31_1009, %b_A32_1012 : tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32> loc(#loc1083) + } else { + scf.yield %b_A33_855, %b_A30_859, %b_A31_860, %b_A32_861 : tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32> loc(#loc163) + } loc(#loc1100) + scf.yield %b_A22_969, %b_A32_976#0, %b_A20_972, %b_A21_975, %b_A32_976#1, %b_A32_976#2, %b_A32_976#3 : tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32> loc(#loc1100) + } else { + scf.yield %b_A22_854, %b_A33_855, %b_A20_857, %b_A21_858, %b_A30_859, %b_A31_860, %b_A32_861 : tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32> loc(#loc144) + } loc(#loc1106) + scf.yield %b_A11_938, %b_A32_942#0, %b_A32_942#1, %b_A10_941, %b_A32_942#2, %b_A32_942#3, %b_A32_942#4, %b_A32_942#5, %b_A32_942#6 : tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32> loc(#loc1106) + } else { + scf.yield %b_A11_853, %b_A22_854, %b_A33_855, %b_A10_856, %b_A20_857, %b_A21_858, %b_A30_859, %b_A31_860, %b_A32_861 : tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32> loc(#loc127) + } loc(#loc127) + scf.yield %b_A00_911, %63#0, %63#1, %63#2, %63#3, %63#4, %63#5, %63#6, %63#7, %63#8 : tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32> loc(#loc183) + } loc(#loc1109) + %m_d = tt.expand_dims %o_i {axis = 1 : i32} : tensor<16xi32> -> tensor<16x1xi32> loc(#loc733) + %m_d_187 = tt.expand_dims %o_i {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc734) + %m_d_188 = tt.broadcast %m_d : tensor<16x1xi32> -> tensor<16x16xi32> loc(#loc735) + %m_d_189 = tt.broadcast %m_d_187 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc735) + %m_d_190 = arith.cmpi sgt, %m_d_188, %m_d_189 : tensor<16x16xi32> loc(#loc735) + %m_I = tt.expand_dims %o_i {axis = 1 : i32} : tensor<16xi32> -> tensor<16x1xi32> loc(#loc736) + %m_I_191 = tt.expand_dims %o_i {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc737) + %m_I_192 = tt.broadcast %m_I : tensor<16x1xi32> -> tensor<16x16xi32> loc(#loc738) + %m_I_193 = tt.broadcast %m_I_191 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc738) + %m_I_194 = arith.cmpi eq, %m_I_192, %m_I_193 : tensor<16x16xi32> loc(#loc738) + %b_A00_195 = tt.expand_dims %m_tc0_42 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc739) + %b_A00_196 = tt.broadcast %b_A00_195 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc740) + %b_A00_197 = arith.andi %m_d_190, %b_A00_196 : tensor<16x16xi1> loc(#loc740) + %b_A00_198 = tt.expand_dims %m_tc0_42 {axis = 0 : i32} : tensor<16xi1> -> tensor<1x16xi1> loc(#loc741) + %b_A00_199 = tt.broadcast %b_A00_198 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc742) + %b_A00_200 = arith.andi %b_A00_197, %b_A00_199 : tensor<16x16xi1> loc(#loc742) + %b_A00_201 = tt.expand_dims %b_g0_179 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc743) + %b_A00_202 = tt.expand_dims %b_g0_179 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc744) + %b_A00_203 = tt.broadcast %b_A00_201 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc745) + %b_A00_204 = tt.broadcast %b_A00_202 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc745) + %b_A00_205 = arith.subf %b_A00_203, %b_A00_204 : tensor<16x16xf32> loc(#loc745) + %b_A00_206 = tt.call @fla.ops.utils.op.exp2__fp32S16_16S(%b_A00_205) : (tensor<16x16xf32>) -> tensor<16x16xf32> loc(#loc746) + %b_A00_207 = arith.constant 0.000000e+00 : f32 loc(#loc747) + %b_A00_208 = arith.constant 0.000000e+00 : f32 loc(#loc747) + %b_A00_209 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc747) + %b_A00_210 = arith.select %b_A00_200, %b_A00_206, %b_A00_209 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc747) + %b_A00_211 = arith.mulf %b_A32_186#0, %b_A00_210 : tensor<16x16xf32> loc(#loc748) + %b_A11_212 = tt.expand_dims %m_tc1_47 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc749) + %b_A11_213 = tt.broadcast %b_A11_212 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc750) + %b_A11_214 = arith.andi %m_d_190, %b_A11_213 : tensor<16x16xi1> loc(#loc750) + %b_A11_215 = tt.expand_dims %m_tc1_47 {axis = 0 : i32} : tensor<16xi1> -> tensor<1x16xi1> loc(#loc751) + %b_A11_216 = tt.broadcast %b_A11_215 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc752) + %b_A11_217 = arith.andi %b_A11_214, %b_A11_216 : tensor<16x16xi1> loc(#loc752) + %b_A11_218 = tt.expand_dims %b_g1_181 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc753) + %b_A11_219 = tt.expand_dims %b_g1_181 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc754) + %b_A11_220 = tt.broadcast %b_A11_218 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc755) + %b_A11_221 = tt.broadcast %b_A11_219 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc755) + %b_A11_222 = arith.subf %b_A11_220, %b_A11_221 : tensor<16x16xf32> loc(#loc755) + %b_A11_223 = tt.call @fla.ops.utils.op.exp2__fp32S16_16S(%b_A11_222) : (tensor<16x16xf32>) -> tensor<16x16xf32> loc(#loc756) + %b_A11_224 = arith.constant 0.000000e+00 : f32 loc(#loc757) + %b_A11_225 = arith.constant 0.000000e+00 : f32 loc(#loc757) + %b_A11_226 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc757) + %b_A11_227 = arith.select %b_A11_217, %b_A11_223, %b_A11_226 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc757) + %b_A11_228 = arith.mulf %b_A32_186#1, %b_A11_227 : tensor<16x16xf32> loc(#loc758) + %b_A22_229 = tt.expand_dims %m_tc2_52 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc759) + %b_A22_230 = tt.broadcast %b_A22_229 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc760) + %b_A22_231 = arith.andi %m_d_190, %b_A22_230 : tensor<16x16xi1> loc(#loc760) + %b_A22_232 = tt.expand_dims %m_tc2_52 {axis = 0 : i32} : tensor<16xi1> -> tensor<1x16xi1> loc(#loc761) + %b_A22_233 = tt.broadcast %b_A22_232 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc762) + %b_A22_234 = arith.andi %b_A22_231, %b_A22_233 : tensor<16x16xi1> loc(#loc762) + %b_A22_235 = tt.expand_dims %b_g2_183 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc763) + %b_A22_236 = tt.expand_dims %b_g2_183 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc764) + %b_A22_237 = tt.broadcast %b_A22_235 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc765) + %b_A22_238 = tt.broadcast %b_A22_236 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc765) + %b_A22_239 = arith.subf %b_A22_237, %b_A22_238 : tensor<16x16xf32> loc(#loc765) + %b_A22_240 = tt.call @fla.ops.utils.op.exp2__fp32S16_16S(%b_A22_239) : (tensor<16x16xf32>) -> tensor<16x16xf32> loc(#loc766) + %b_A22_241 = arith.constant 0.000000e+00 : f32 loc(#loc767) + %b_A22_242 = arith.constant 0.000000e+00 : f32 loc(#loc767) + %b_A22_243 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc767) + %b_A22_244 = arith.select %b_A22_234, %b_A22_240, %b_A22_243 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc767) + %b_A22_245 = arith.mulf %b_A32_186#2, %b_A22_244 : tensor<16x16xf32> loc(#loc768) + %b_A33_246 = tt.expand_dims %m_tc3_57 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc769) + %b_A33_247 = tt.broadcast %b_A33_246 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc770) + %b_A33_248 = arith.andi %m_d_190, %b_A33_247 : tensor<16x16xi1> loc(#loc770) + %b_A33_249 = tt.expand_dims %m_tc3_57 {axis = 0 : i32} : tensor<16xi1> -> tensor<1x16xi1> loc(#loc771) + %b_A33_250 = tt.broadcast %b_A33_249 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc772) + %b_A33_251 = arith.andi %b_A33_248, %b_A33_250 : tensor<16x16xi1> loc(#loc772) + %b_A33_252 = tt.expand_dims %b_g3_185 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc773) + %b_A33_253 = tt.expand_dims %b_g3_185 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc774) + %b_A33_254 = tt.broadcast %b_A33_252 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc775) + %b_A33_255 = tt.broadcast %b_A33_253 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc775) + %b_A33_256 = arith.subf %b_A33_254, %b_A33_255 : tensor<16x16xf32> loc(#loc775) + %b_A33_257 = tt.call @fla.ops.utils.op.exp2__fp32S16_16S(%b_A33_256) : (tensor<16x16xf32>) -> tensor<16x16xf32> loc(#loc776) + %b_A33_258 = arith.constant 0.000000e+00 : f32 loc(#loc777) + %b_A33_259 = arith.constant 0.000000e+00 : f32 loc(#loc777) + %b_A33_260 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc777) + %b_A33_261 = arith.select %b_A33_251, %b_A33_257, %b_A33_260 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc777) + %b_A33_262 = arith.mulf %b_A32_186#3, %b_A33_261 : tensor<16x16xf32> loc(#loc778) + %b_A10_263 = tt.expand_dims %m_tc1_47 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc779) + %b_A10_264 = tt.expand_dims %m_tc0_42 {axis = 0 : i32} : tensor<16xi1> -> tensor<1x16xi1> loc(#loc780) + %b_A10_265 = tt.broadcast %b_A10_263 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc781) + %b_A10_266 = tt.broadcast %b_A10_264 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc781) + %b_A10_267 = arith.andi %b_A10_265, %b_A10_266 : tensor<16x16xi1> loc(#loc781) + %b_A10_268 = tt.expand_dims %b_g1_181 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc782) + %b_A10_269 = tt.expand_dims %b_g0_179 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc783) + %b_A10_270 = tt.broadcast %b_A10_268 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc784) + %b_A10_271 = tt.broadcast %b_A10_269 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc784) + %b_A10_272 = arith.subf %b_A10_270, %b_A10_271 : tensor<16x16xf32> loc(#loc784) + %b_A10_273 = tt.call @fla.ops.utils.op.exp2__fp32S16_16S(%b_A10_272) : (tensor<16x16xf32>) -> tensor<16x16xf32> loc(#loc785) + %b_A10_274 = arith.constant 0.000000e+00 : f32 loc(#loc786) + %b_A10_275 = arith.constant 0.000000e+00 : f32 loc(#loc786) + %b_A10_276 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc786) + %b_A10_277 = arith.select %b_A10_267, %b_A10_273, %b_A10_276 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc786) + %b_A10_278 = arith.mulf %b_A32_186#4, %b_A10_277 : tensor<16x16xf32> loc(#loc787) + %b_A20_279 = tt.expand_dims %m_tc2_52 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc788) + %b_A20_280 = tt.expand_dims %m_tc0_42 {axis = 0 : i32} : tensor<16xi1> -> tensor<1x16xi1> loc(#loc789) + %b_A20_281 = tt.broadcast %b_A20_279 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc790) + %b_A20_282 = tt.broadcast %b_A20_280 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc790) + %b_A20_283 = arith.andi %b_A20_281, %b_A20_282 : tensor<16x16xi1> loc(#loc790) + %b_A20_284 = tt.expand_dims %b_g2_183 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc791) + %b_A20_285 = tt.expand_dims %b_g0_179 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc792) + %b_A20_286 = tt.broadcast %b_A20_284 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc793) + %b_A20_287 = tt.broadcast %b_A20_285 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc793) + %b_A20_288 = arith.subf %b_A20_286, %b_A20_287 : tensor<16x16xf32> loc(#loc793) + %b_A20_289 = tt.call @fla.ops.utils.op.exp2__fp32S16_16S(%b_A20_288) : (tensor<16x16xf32>) -> tensor<16x16xf32> loc(#loc794) + %b_A20_290 = arith.constant 0.000000e+00 : f32 loc(#loc795) + %b_A20_291 = arith.constant 0.000000e+00 : f32 loc(#loc795) + %b_A20_292 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc795) + %b_A20_293 = arith.select %b_A20_283, %b_A20_289, %b_A20_292 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc795) + %b_A20_294 = arith.mulf %b_A32_186#5, %b_A20_293 : tensor<16x16xf32> loc(#loc796) + %b_A21_295 = tt.expand_dims %m_tc2_52 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc797) + %b_A21_296 = tt.expand_dims %m_tc1_47 {axis = 0 : i32} : tensor<16xi1> -> tensor<1x16xi1> loc(#loc798) + %b_A21_297 = tt.broadcast %b_A21_295 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc799) + %b_A21_298 = tt.broadcast %b_A21_296 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc799) + %b_A21_299 = arith.andi %b_A21_297, %b_A21_298 : tensor<16x16xi1> loc(#loc799) + %b_A21_300 = tt.expand_dims %b_g2_183 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc800) + %b_A21_301 = tt.expand_dims %b_g1_181 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc801) + %b_A21_302 = tt.broadcast %b_A21_300 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc802) + %b_A21_303 = tt.broadcast %b_A21_301 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc802) + %b_A21_304 = arith.subf %b_A21_302, %b_A21_303 : tensor<16x16xf32> loc(#loc802) + %b_A21_305 = tt.call @fla.ops.utils.op.exp2__fp32S16_16S(%b_A21_304) : (tensor<16x16xf32>) -> tensor<16x16xf32> loc(#loc803) + %b_A21_306 = arith.constant 0.000000e+00 : f32 loc(#loc804) + %b_A21_307 = arith.constant 0.000000e+00 : f32 loc(#loc804) + %b_A21_308 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc804) + %b_A21_309 = arith.select %b_A21_299, %b_A21_305, %b_A21_308 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc804) + %b_A21_310 = arith.mulf %b_A32_186#6, %b_A21_309 : tensor<16x16xf32> loc(#loc805) + %b_A30_311 = tt.expand_dims %m_tc3_57 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc806) + %b_A30_312 = tt.expand_dims %m_tc0_42 {axis = 0 : i32} : tensor<16xi1> -> tensor<1x16xi1> loc(#loc807) + %b_A30_313 = tt.broadcast %b_A30_311 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc808) + %b_A30_314 = tt.broadcast %b_A30_312 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc808) + %b_A30_315 = arith.andi %b_A30_313, %b_A30_314 : tensor<16x16xi1> loc(#loc808) + %b_A30_316 = tt.expand_dims %b_g3_185 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc809) + %b_A30_317 = tt.expand_dims %b_g0_179 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc810) + %b_A30_318 = tt.broadcast %b_A30_316 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc811) + %b_A30_319 = tt.broadcast %b_A30_317 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc811) + %b_A30_320 = arith.subf %b_A30_318, %b_A30_319 : tensor<16x16xf32> loc(#loc811) + %b_A30_321 = tt.call @fla.ops.utils.op.exp2__fp32S16_16S(%b_A30_320) : (tensor<16x16xf32>) -> tensor<16x16xf32> loc(#loc812) + %b_A30_322 = arith.constant 0.000000e+00 : f32 loc(#loc813) + %b_A30_323 = arith.constant 0.000000e+00 : f32 loc(#loc813) + %b_A30_324 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc813) + %b_A30_325 = arith.select %b_A30_315, %b_A30_321, %b_A30_324 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc813) + %b_A30_326 = arith.mulf %b_A32_186#7, %b_A30_325 : tensor<16x16xf32> loc(#loc814) + %b_A31_327 = tt.expand_dims %m_tc3_57 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc815) + %b_A31_328 = tt.expand_dims %m_tc1_47 {axis = 0 : i32} : tensor<16xi1> -> tensor<1x16xi1> loc(#loc816) + %b_A31_329 = tt.broadcast %b_A31_327 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc817) + %b_A31_330 = tt.broadcast %b_A31_328 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc817) + %b_A31_331 = arith.andi %b_A31_329, %b_A31_330 : tensor<16x16xi1> loc(#loc817) + %b_A31_332 = tt.expand_dims %b_g3_185 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc818) + %b_A31_333 = tt.expand_dims %b_g1_181 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc819) + %b_A31_334 = tt.broadcast %b_A31_332 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc820) + %b_A31_335 = tt.broadcast %b_A31_333 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc820) + %b_A31_336 = arith.subf %b_A31_334, %b_A31_335 : tensor<16x16xf32> loc(#loc820) + %b_A31_337 = tt.call @fla.ops.utils.op.exp2__fp32S16_16S(%b_A31_336) : (tensor<16x16xf32>) -> tensor<16x16xf32> loc(#loc821) + %b_A31_338 = arith.constant 0.000000e+00 : f32 loc(#loc822) + %b_A31_339 = arith.constant 0.000000e+00 : f32 loc(#loc822) + %b_A31_340 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc822) + %b_A31_341 = arith.select %b_A31_331, %b_A31_337, %b_A31_340 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc822) + %b_A31_342 = arith.mulf %b_A32_186#8, %b_A31_341 : tensor<16x16xf32> loc(#loc823) + %b_A32_343 = tt.expand_dims %m_tc3_57 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc824) + %b_A32_344 = tt.expand_dims %m_tc2_52 {axis = 0 : i32} : tensor<16xi1> -> tensor<1x16xi1> loc(#loc825) + %b_A32_345 = tt.broadcast %b_A32_343 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc826) + %b_A32_346 = tt.broadcast %b_A32_344 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc826) + %b_A32_347 = arith.andi %b_A32_345, %b_A32_346 : tensor<16x16xi1> loc(#loc826) + %b_A32_348 = tt.expand_dims %b_g3_185 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc827) + %b_A32_349 = tt.expand_dims %b_g2_183 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc828) + %b_A32_350 = tt.broadcast %b_A32_348 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc829) + %b_A32_351 = tt.broadcast %b_A32_349 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc829) + %b_A32_352 = arith.subf %b_A32_350, %b_A32_351 : tensor<16x16xf32> loc(#loc829) + %b_A32_353 = tt.call @fla.ops.utils.op.exp2__fp32S16_16S(%b_A32_352) : (tensor<16x16xf32>) -> tensor<16x16xf32> loc(#loc830) + %b_A32_354 = arith.constant 0.000000e+00 : f32 loc(#loc831) + %b_A32_355 = arith.constant 0.000000e+00 : f32 loc(#loc831) + %b_A32_356 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc831) + %b_A32_357 = arith.select %b_A32_347, %b_A32_353, %b_A32_356 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc831) + %b_A32_358 = arith.mulf %b_A32_186#9, %b_A32_357 : tensor<16x16xf32> loc(#loc832) + %b_A00_359 = tt.expand_dims %b_b0_113 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc833) + %b_A00_360 = tt.broadcast %b_A00_359 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc834) + %b_A00_361 = arith.mulf %b_A00_211, %b_A00_360 : tensor<16x16xf32> loc(#loc834) + %b_A11_362 = tt.expand_dims %b_b1_117 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc835) + %b_A11_363 = tt.broadcast %b_A11_362 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc836) + %b_A11_364 = arith.mulf %b_A11_228, %b_A11_363 : tensor<16x16xf32> loc(#loc836) + %b_A22_365 = tt.expand_dims %b_b2_121 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc837) + %b_A22_366 = tt.broadcast %b_A22_365 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc838) + %b_A22_367 = arith.mulf %b_A22_245, %b_A22_366 : tensor<16x16xf32> loc(#loc838) + %b_A33_368 = tt.expand_dims %b_b3_125 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc839) + %b_A33_369 = tt.broadcast %b_A33_368 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc840) + %b_A33_370 = arith.mulf %b_A33_262, %b_A33_369 : tensor<16x16xf32> loc(#loc840) + %b_A10_371 = tt.expand_dims %b_b1_117 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc841) + %b_A10_372 = tt.broadcast %b_A10_371 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc842) + %b_A10_373 = arith.mulf %b_A10_278, %b_A10_372 : tensor<16x16xf32> loc(#loc842) + %b_A20_374 = tt.expand_dims %b_b2_121 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc843) + %b_A20_375 = tt.broadcast %b_A20_374 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc844) + %b_A20_376 = arith.mulf %b_A20_294, %b_A20_375 : tensor<16x16xf32> loc(#loc844) + %b_A21_377 = tt.expand_dims %b_b2_121 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc845) + %b_A21_378 = tt.broadcast %b_A21_377 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc846) + %b_A21_379 = arith.mulf %b_A21_310, %b_A21_378 : tensor<16x16xf32> loc(#loc846) + %b_A30_380 = tt.expand_dims %b_b3_125 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc847) + %b_A30_381 = tt.broadcast %b_A30_380 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc848) + %b_A30_382 = arith.mulf %b_A30_326, %b_A30_381 : tensor<16x16xf32> loc(#loc848) + %b_A31_383 = tt.expand_dims %b_b3_125 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc849) + %b_A31_384 = tt.broadcast %b_A31_383 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc850) + %b_A31_385 = arith.mulf %b_A31_342, %b_A31_384 : tensor<16x16xf32> loc(#loc850) + %b_A32_386 = tt.expand_dims %b_b3_125 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc851) + %b_A32_387 = tt.broadcast %b_A32_386 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc852) + %b_A32_388 = arith.mulf %b_A32_358, %b_A32_387 : tensor<16x16xf32> loc(#loc852) + %b_Ai00 = arith.constant 0.000000e+00 : f32 loc(#loc853) + %b_Ai00_389 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc853) + %b_Ai00_390 = arith.subf %b_Ai00_389, %b_A00_361 : tensor<16x16xf32> loc(#loc853) + %b_Ai11 = arith.constant 0.000000e+00 : f32 loc(#loc854) + %b_Ai11_391 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc854) + %b_Ai11_392 = arith.subf %b_Ai11_391, %b_A11_364 : tensor<16x16xf32> loc(#loc854) + %b_Ai22 = arith.constant 0.000000e+00 : f32 loc(#loc855) + %b_Ai22_393 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc855) + %b_Ai22_394 = arith.subf %b_Ai22_393, %b_A22_367 : tensor<16x16xf32> loc(#loc855) + %b_Ai33 = arith.constant 0.000000e+00 : f32 loc(#loc856) + %b_Ai33_395 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc856) + %b_Ai33_396 = arith.subf %b_Ai33_395, %b_A33_370 : tensor<16x16xf32> loc(#loc856) + %19 = arith.extsi %T : i32 to i64 loc(#loc308) + %20 = arith.subi %19, %i_tc0_3 : i64 loc(#loc308) + %c16_i32 = arith.constant 16 : i32 loc(#loc309) + %21 = arith.extsi %c16_i32 : i32 to i64 loc(#loc309) + %22 = arith.minsi %21, %20 : i64 loc(#loc309) + %c2_i32_397 = arith.constant 2 : i32 loc(#loc310) + %c1_i32_398 = arith.constant 1 : i32 loc(#loc310) + %23 = arith.extsi %c2_i32_397 : i32 to i64 loc(#loc310) + %24 = arith.bitcast %22 : i64 to i64 loc(#loc310) + %25 = arith.extsi %c1_i32_398 : i32 to i64 loc(#loc310) + %26 = ub.poison : i64 loc(#loc310) + %b_Ai00_399 = scf.for %i = %23 to %24 step %25 iter_args(%b_Ai00_852 = %b_Ai00_390) -> (tensor<16x16xf32>) : i64 { + %b_a00 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc858) + %b_a00_853 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc858) + %b_a00_854 = arith.cmpi eq, %b_a00, %b_a00_853 : tensor<16xi64> loc(#loc858) + %b_a00_855 = tt.expand_dims %b_a00_854 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc859) + %b_a00_856 = arith.constant 0.000000e+00 : f32 loc(#loc860) + %b_a00_857 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc860) + %b_a00_858 = arith.subf %b_a00_857, %b_A00_361 : tensor<16x16xf32> loc(#loc860) + %b_a00_859 = arith.constant 0.000000e+00 : f32 loc(#loc861) + %b_a00_860 = arith.constant 0.000000e+00 : f32 loc(#loc861) + %b_a00_861 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc861) + %b_a00_862 = tt.broadcast %b_a00_855 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc861) + %b_a00_863 = arith.select %b_a00_862, %b_a00_858, %b_a00_861 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc861) + %b_a00_864 = tt.call @triton.language.standard.sum__fp32S16_16S_c0_cFalse_cNone(%b_a00_863) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc862) + %b_a00_865 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc863) + %b_a00_866 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc863) + %b_a00_867 = arith.cmpi slt, %b_a00_865, %b_a00_866 : tensor<16xi64> loc(#loc863) + %b_a00_868 = arith.constant 0.000000e+00 : f32 loc(#loc864) + %b_a00_869 = arith.constant 0.000000e+00 : f32 loc(#loc864) + %b_a00_870 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc864) + %b_a00_871 = arith.select %b_a00_867, %b_a00_864, %b_a00_870 : tensor<16xi1>, tensor<16xf32> loc(#loc864) + %b_a00_872 = tt.expand_dims %b_a00_871 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc865) + %b_a00_873 = tt.broadcast %b_a00_872 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc866) + %b_a00_874 = arith.mulf %b_a00_873, %b_Ai00_852 : tensor<16x16xf32> loc(#loc866) + %b_a00_875 = tt.call @triton.language.standard.sum__fp32S16_16S_c0_cFalse_cNone(%b_a00_874) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc867) + %b_a00_876 = arith.addf %b_a00_871, %b_a00_875 : tensor<16xf32> loc(#loc868) + %b_Ai00_877 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc869) + %b_Ai00_878 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc869) + %b_Ai00_879 = arith.cmpi eq, %b_Ai00_877, %b_Ai00_878 : tensor<16xi64> loc(#loc869) + %b_Ai00_880 = tt.expand_dims %b_Ai00_879 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc870) + %b_Ai00_881 = tt.expand_dims %b_a00_876 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc871) + %b_Ai00_882 = tt.broadcast %b_Ai00_881 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc871) + %b_Ai00_883 = tt.broadcast %b_Ai00_880 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc871) + %b_Ai00_884 = arith.select %b_Ai00_883, %b_Ai00_882, %b_Ai00_852 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc871) + scf.yield %b_Ai00_884 : tensor<16x16xf32> loc(#loc325) + } loc(#loc857) + %27 = arith.extsi %T : i32 to i64 loc(#loc326) + %28 = arith.subi %27, %i_tc1_8 : i64 loc(#loc326) + %c16_i32_400 = arith.constant 16 : i32 loc(#loc327) + %29 = arith.extsi %c16_i32_400 : i32 to i64 loc(#loc327) + %30 = arith.minsi %29, %28 : i64 loc(#loc327) + %c2_i32_401 = arith.constant 2 : i32 loc(#loc328) + %c1_i32_402 = arith.constant 1 : i32 loc(#loc328) + %31 = arith.extsi %c2_i32_401 : i32 to i64 loc(#loc328) + %32 = arith.bitcast %30 : i64 to i64 loc(#loc328) + %33 = arith.extsi %c1_i32_402 : i32 to i64 loc(#loc328) + %34 = ub.poison : i64 loc(#loc328) + %b_Ai11_403 = scf.for %i = %31 to %32 step %33 iter_args(%b_Ai11_852 = %b_Ai11_392) -> (tensor<16x16xf32>) : i64 { + %b_a11 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc873) + %b_a11_853 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc873) + %b_a11_854 = arith.cmpi eq, %b_a11, %b_a11_853 : tensor<16xi64> loc(#loc873) + %b_a11_855 = tt.expand_dims %b_a11_854 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc874) + %b_a11_856 = arith.constant 0.000000e+00 : f32 loc(#loc875) + %b_a11_857 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc875) + %b_a11_858 = arith.subf %b_a11_857, %b_A11_364 : tensor<16x16xf32> loc(#loc875) + %b_a11_859 = arith.constant 0.000000e+00 : f32 loc(#loc876) + %b_a11_860 = arith.constant 0.000000e+00 : f32 loc(#loc876) + %b_a11_861 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc876) + %b_a11_862 = tt.broadcast %b_a11_855 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc876) + %b_a11_863 = arith.select %b_a11_862, %b_a11_858, %b_a11_861 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc876) + %b_a11_864 = tt.call @triton.language.standard.sum__fp32S16_16S_c0_cFalse_cNone(%b_a11_863) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc877) + %b_a11_865 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc878) + %b_a11_866 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc878) + %b_a11_867 = arith.cmpi slt, %b_a11_865, %b_a11_866 : tensor<16xi64> loc(#loc878) + %b_a11_868 = arith.constant 0.000000e+00 : f32 loc(#loc879) + %b_a11_869 = arith.constant 0.000000e+00 : f32 loc(#loc879) + %b_a11_870 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc879) + %b_a11_871 = arith.select %b_a11_867, %b_a11_864, %b_a11_870 : tensor<16xi1>, tensor<16xf32> loc(#loc879) + %b_a11_872 = tt.expand_dims %b_a11_871 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc880) + %b_a11_873 = tt.broadcast %b_a11_872 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc881) + %b_a11_874 = arith.mulf %b_a11_873, %b_Ai11_852 : tensor<16x16xf32> loc(#loc881) + %b_a11_875 = tt.call @triton.language.standard.sum__fp32S16_16S_c0_cFalse_cNone(%b_a11_874) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc882) + %b_a11_876 = arith.addf %b_a11_871, %b_a11_875 : tensor<16xf32> loc(#loc883) + %b_Ai11_877 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc884) + %b_Ai11_878 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc884) + %b_Ai11_879 = arith.cmpi eq, %b_Ai11_877, %b_Ai11_878 : tensor<16xi64> loc(#loc884) + %b_Ai11_880 = tt.expand_dims %b_Ai11_879 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc885) + %b_Ai11_881 = tt.expand_dims %b_a11_876 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc886) + %b_Ai11_882 = tt.broadcast %b_Ai11_881 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc886) + %b_Ai11_883 = tt.broadcast %b_Ai11_880 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc886) + %b_Ai11_884 = arith.select %b_Ai11_883, %b_Ai11_882, %b_Ai11_852 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc886) + scf.yield %b_Ai11_884 : tensor<16x16xf32> loc(#loc343) + } loc(#loc872) + %35 = arith.extsi %T : i32 to i64 loc(#loc344) + %36 = arith.subi %35, %i_tc2_13 : i64 loc(#loc344) + %c16_i32_404 = arith.constant 16 : i32 loc(#loc345) + %37 = arith.extsi %c16_i32_404 : i32 to i64 loc(#loc345) + %38 = arith.minsi %37, %36 : i64 loc(#loc345) + %c2_i32_405 = arith.constant 2 : i32 loc(#loc346) + %c1_i32_406 = arith.constant 1 : i32 loc(#loc346) + %39 = arith.extsi %c2_i32_405 : i32 to i64 loc(#loc346) + %40 = arith.bitcast %38 : i64 to i64 loc(#loc346) + %41 = arith.extsi %c1_i32_406 : i32 to i64 loc(#loc346) + %42 = ub.poison : i64 loc(#loc346) + %b_Ai22_407 = scf.for %i = %39 to %40 step %41 iter_args(%b_Ai22_852 = %b_Ai22_394) -> (tensor<16x16xf32>) : i64 { + %b_a22 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc888) + %b_a22_853 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc888) + %b_a22_854 = arith.cmpi eq, %b_a22, %b_a22_853 : tensor<16xi64> loc(#loc888) + %b_a22_855 = tt.expand_dims %b_a22_854 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc889) + %b_a22_856 = arith.constant 0.000000e+00 : f32 loc(#loc890) + %b_a22_857 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc890) + %b_a22_858 = arith.subf %b_a22_857, %b_A22_367 : tensor<16x16xf32> loc(#loc890) + %b_a22_859 = arith.constant 0.000000e+00 : f32 loc(#loc891) + %b_a22_860 = arith.constant 0.000000e+00 : f32 loc(#loc891) + %b_a22_861 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc891) + %b_a22_862 = tt.broadcast %b_a22_855 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc891) + %b_a22_863 = arith.select %b_a22_862, %b_a22_858, %b_a22_861 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc891) + %b_a22_864 = tt.call @triton.language.standard.sum__fp32S16_16S_c0_cFalse_cNone(%b_a22_863) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc892) + %b_a22_865 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc893) + %b_a22_866 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc893) + %b_a22_867 = arith.cmpi slt, %b_a22_865, %b_a22_866 : tensor<16xi64> loc(#loc893) + %b_a22_868 = arith.constant 0.000000e+00 : f32 loc(#loc894) + %b_a22_869 = arith.constant 0.000000e+00 : f32 loc(#loc894) + %b_a22_870 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc894) + %b_a22_871 = arith.select %b_a22_867, %b_a22_864, %b_a22_870 : tensor<16xi1>, tensor<16xf32> loc(#loc894) + %b_a22_872 = tt.expand_dims %b_a22_871 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc895) + %b_a22_873 = tt.broadcast %b_a22_872 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc896) + %b_a22_874 = arith.mulf %b_a22_873, %b_Ai22_852 : tensor<16x16xf32> loc(#loc896) + %b_a22_875 = tt.call @triton.language.standard.sum__fp32S16_16S_c0_cFalse_cNone(%b_a22_874) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc897) + %b_a22_876 = arith.addf %b_a22_871, %b_a22_875 : tensor<16xf32> loc(#loc898) + %b_Ai22_877 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc899) + %b_Ai22_878 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc899) + %b_Ai22_879 = arith.cmpi eq, %b_Ai22_877, %b_Ai22_878 : tensor<16xi64> loc(#loc899) + %b_Ai22_880 = tt.expand_dims %b_Ai22_879 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc900) + %b_Ai22_881 = tt.expand_dims %b_a22_876 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc901) + %b_Ai22_882 = tt.broadcast %b_Ai22_881 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc901) + %b_Ai22_883 = tt.broadcast %b_Ai22_880 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc901) + %b_Ai22_884 = arith.select %b_Ai22_883, %b_Ai22_882, %b_Ai22_852 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc901) + scf.yield %b_Ai22_884 : tensor<16x16xf32> loc(#loc361) + } loc(#loc887) + %43 = arith.extsi %T : i32 to i64 loc(#loc362) + %44 = arith.subi %43, %i_tc3_18 : i64 loc(#loc362) + %c16_i32_408 = arith.constant 16 : i32 loc(#loc363) + %45 = arith.extsi %c16_i32_408 : i32 to i64 loc(#loc363) + %46 = arith.minsi %45, %44 : i64 loc(#loc363) + %c2_i32_409 = arith.constant 2 : i32 loc(#loc364) + %c1_i32_410 = arith.constant 1 : i32 loc(#loc364) + %47 = arith.extsi %c2_i32_409 : i32 to i64 loc(#loc364) + %48 = arith.bitcast %46 : i64 to i64 loc(#loc364) + %49 = arith.extsi %c1_i32_410 : i32 to i64 loc(#loc364) + %50 = ub.poison : i64 loc(#loc364) + %b_Ai33_411 = scf.for %i = %47 to %48 step %49 iter_args(%b_Ai33_852 = %b_Ai33_396) -> (tensor<16x16xf32>) : i64 { + %b_a33 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc903) + %b_a33_853 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc903) + %b_a33_854 = arith.cmpi eq, %b_a33, %b_a33_853 : tensor<16xi64> loc(#loc903) + %b_a33_855 = tt.expand_dims %b_a33_854 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc904) + %b_a33_856 = arith.constant 0.000000e+00 : f32 loc(#loc905) + %b_a33_857 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc905) + %b_a33_858 = arith.subf %b_a33_857, %b_A33_370 : tensor<16x16xf32> loc(#loc905) + %b_a33_859 = arith.constant 0.000000e+00 : f32 loc(#loc906) + %b_a33_860 = arith.constant 0.000000e+00 : f32 loc(#loc906) + %b_a33_861 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc906) + %b_a33_862 = tt.broadcast %b_a33_855 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc906) + %b_a33_863 = arith.select %b_a33_862, %b_a33_858, %b_a33_861 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc906) + %b_a33_864 = tt.call @triton.language.standard.sum__fp32S16_16S_c0_cFalse_cNone(%b_a33_863) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc907) + %b_a33_865 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc908) + %b_a33_866 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc908) + %b_a33_867 = arith.cmpi slt, %b_a33_865, %b_a33_866 : tensor<16xi64> loc(#loc908) + %b_a33_868 = arith.constant 0.000000e+00 : f32 loc(#loc909) + %b_a33_869 = arith.constant 0.000000e+00 : f32 loc(#loc909) + %b_a33_870 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc909) + %b_a33_871 = arith.select %b_a33_867, %b_a33_864, %b_a33_870 : tensor<16xi1>, tensor<16xf32> loc(#loc909) + %b_a33_872 = tt.expand_dims %b_a33_871 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc910) + %b_a33_873 = tt.broadcast %b_a33_872 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc911) + %b_a33_874 = arith.mulf %b_a33_873, %b_Ai33_852 : tensor<16x16xf32> loc(#loc911) + %b_a33_875 = tt.call @triton.language.standard.sum__fp32S16_16S_c0_cFalse_cNone(%b_a33_874) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc912) + %b_a33_876 = arith.addf %b_a33_871, %b_a33_875 : tensor<16xf32> loc(#loc913) + %b_Ai33_877 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc914) + %b_Ai33_878 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc914) + %b_Ai33_879 = arith.cmpi eq, %b_Ai33_877, %b_Ai33_878 : tensor<16xi64> loc(#loc914) + %b_Ai33_880 = tt.expand_dims %b_Ai33_879 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc915) + %b_Ai33_881 = tt.expand_dims %b_a33_876 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc916) + %b_Ai33_882 = tt.broadcast %b_Ai33_881 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc916) + %b_Ai33_883 = tt.broadcast %b_Ai33_880 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc916) + %b_Ai33_884 = arith.select %b_Ai33_883, %b_Ai33_882, %b_Ai33_852 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc916) + scf.yield %b_Ai33_884 : tensor<16x16xf32> loc(#loc379) + } loc(#loc902) + %b_Ai00_412 = arith.uitofp %m_I_194 : tensor<16x16xi1> to tensor<16x16xf32> loc(#loc917) + %b_Ai00_413 = arith.addf %b_Ai00_399, %b_Ai00_412 : tensor<16x16xf32> loc(#loc917) + %b_Ai11_414 = arith.uitofp %m_I_194 : tensor<16x16xi1> to tensor<16x16xf32> loc(#loc918) + %b_Ai11_415 = arith.addf %b_Ai11_403, %b_Ai11_414 : tensor<16x16xf32> loc(#loc918) + %b_Ai22_416 = arith.uitofp %m_I_194 : tensor<16x16xi1> to tensor<16x16xf32> loc(#loc919) + %b_Ai22_417 = arith.addf %b_Ai22_407, %b_Ai22_416 : tensor<16x16xf32> loc(#loc919) + %b_Ai33_418 = arith.uitofp %m_I_194 : tensor<16x16xi1> to tensor<16x16xf32> loc(#loc920) + %b_Ai33_419 = arith.addf %b_Ai33_411, %b_Ai33_418 : tensor<16x16xf32> loc(#loc920) + %b_Ai10 = arith.constant 0.000000e+00 : f32 loc(#loc921) + %b_Ai10_420 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc921) + %b_Ai10_421 = tt.dot %b_Ai11_415, %b_A10_373, %b_Ai10_420, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc921) + %b_Ai10_422 = arith.constant 0.000000e+00 : f32 loc(#loc922) + %b_Ai10_423 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc922) + %b_Ai10_424 = tt.dot %b_Ai10_421, %b_Ai00_413, %b_Ai10_423, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc922) + %b_Ai10_425 = arith.constant 0.000000e+00 : f32 loc(#loc923) + %b_Ai10_426 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc923) + %b_Ai10_427 = arith.subf %b_Ai10_426, %b_Ai10_424 : tensor<16x16xf32> loc(#loc923) + %b_Ai21 = arith.constant 0.000000e+00 : f32 loc(#loc924) + %b_Ai21_428 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc924) + %b_Ai21_429 = tt.dot %b_Ai22_417, %b_A21_379, %b_Ai21_428, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc924) + %b_Ai21_430 = arith.constant 0.000000e+00 : f32 loc(#loc925) + %b_Ai21_431 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc925) + %b_Ai21_432 = tt.dot %b_Ai21_429, %b_Ai11_415, %b_Ai21_431, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc925) + %b_Ai21_433 = arith.constant 0.000000e+00 : f32 loc(#loc926) + %b_Ai21_434 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc926) + %b_Ai21_435 = arith.subf %b_Ai21_434, %b_Ai21_432 : tensor<16x16xf32> loc(#loc926) + %b_Ai32 = arith.constant 0.000000e+00 : f32 loc(#loc927) + %b_Ai32_436 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc927) + %b_Ai32_437 = tt.dot %b_Ai33_419, %b_A32_388, %b_Ai32_436, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc927) + %b_Ai32_438 = arith.constant 0.000000e+00 : f32 loc(#loc928) + %b_Ai32_439 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc928) + %b_Ai32_440 = tt.dot %b_Ai32_437, %b_Ai22_417, %b_Ai32_439, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc928) + %b_Ai32_441 = arith.constant 0.000000e+00 : f32 loc(#loc929) + %b_Ai32_442 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc929) + %b_Ai32_443 = arith.subf %b_Ai32_442, %b_Ai32_440 : tensor<16x16xf32> loc(#loc929) + %b_Ai20 = arith.constant 0.000000e+00 : f32 loc(#loc930) + %b_Ai20_444 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc930) + %b_Ai20_445 = tt.dot %b_A20_376, %b_Ai00_413, %b_Ai20_444, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc930) + %b_Ai20_446 = arith.constant 0.000000e+00 : f32 loc(#loc931) + %b_Ai20_447 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc931) + %b_Ai20_448 = tt.dot %b_A21_379, %b_Ai10_427, %b_Ai20_447, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc931) + %b_Ai20_449 = arith.addf %b_Ai20_445, %b_Ai20_448 : tensor<16x16xf32> loc(#loc932) + %b_Ai20_450 = arith.constant 0.000000e+00 : f32 loc(#loc933) + %b_Ai20_451 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc933) + %b_Ai20_452 = tt.dot %b_Ai22_417, %b_Ai20_449, %b_Ai20_451, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc933) + %b_Ai20_453 = arith.constant 0.000000e+00 : f32 loc(#loc934) + %b_Ai20_454 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc934) + %b_Ai20_455 = arith.subf %b_Ai20_454, %b_Ai20_452 : tensor<16x16xf32> loc(#loc934) + %b_Ai31 = arith.constant 0.000000e+00 : f32 loc(#loc935) + %b_Ai31_456 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc935) + %b_Ai31_457 = tt.dot %b_A31_385, %b_Ai11_415, %b_Ai31_456, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc935) + %b_Ai31_458 = arith.constant 0.000000e+00 : f32 loc(#loc936) + %b_Ai31_459 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc936) + %b_Ai31_460 = tt.dot %b_A32_388, %b_Ai21_435, %b_Ai31_459, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc936) + %b_Ai31_461 = arith.addf %b_Ai31_457, %b_Ai31_460 : tensor<16x16xf32> loc(#loc937) + %b_Ai31_462 = arith.constant 0.000000e+00 : f32 loc(#loc938) + %b_Ai31_463 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc938) + %b_Ai31_464 = tt.dot %b_Ai33_419, %b_Ai31_461, %b_Ai31_463, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc938) + %b_Ai31_465 = arith.constant 0.000000e+00 : f32 loc(#loc939) + %b_Ai31_466 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc939) + %b_Ai31_467 = arith.subf %b_Ai31_466, %b_Ai31_464 : tensor<16x16xf32> loc(#loc939) + %b_Ai30 = arith.constant 0.000000e+00 : f32 loc(#loc940) + %b_Ai30_468 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc940) + %b_Ai30_469 = tt.dot %b_A30_382, %b_Ai00_413, %b_Ai30_468, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc940) + %b_Ai30_470 = arith.constant 0.000000e+00 : f32 loc(#loc941) + %b_Ai30_471 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc941) + %b_Ai30_472 = tt.dot %b_A31_385, %b_Ai10_427, %b_Ai30_471, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc941) + %b_Ai30_473 = arith.addf %b_Ai30_469, %b_Ai30_472 : tensor<16x16xf32> loc(#loc942) + %b_Ai30_474 = arith.constant 0.000000e+00 : f32 loc(#loc943) + %b_Ai30_475 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc943) + %b_Ai30_476 = tt.dot %b_A32_388, %b_Ai20_455, %b_Ai30_475, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc943) + %b_Ai30_477 = arith.addf %b_Ai30_473, %b_Ai30_476 : tensor<16x16xf32> loc(#loc944) + %b_Ai30_478 = arith.constant 0.000000e+00 : f32 loc(#loc945) + %b_Ai30_479 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc945) + %b_Ai30_480 = tt.dot %b_Ai33_419, %b_Ai30_477, %b_Ai30_479, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc945) + %b_Ai30_481 = arith.constant 0.000000e+00 : f32 loc(#loc946) + %b_Ai30_482 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc946) + %b_Ai30_483 = arith.subf %b_Ai30_482, %b_Ai30_480 : tensor<16x16xf32> loc(#loc946) + %p_A00 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc947) + %p_A00_484 = tt.splat %i_tc0_3 : i64 -> tensor<16xi64> loc(#loc947) + %p_A00_485 = arith.addi %p_A00_484, %p_A00 : tensor<16xi64> loc(#loc947) + %p_A00_486 = tt.expand_dims %p_A00_485 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc948) + %p_A00_487 = arith.constant 2048 : i32 loc(#loc949) + %p_A00_488 = arith.constant 2048 : i64 loc(#loc949) + %p_A00_489 = arith.constant dense<2048> : tensor<16x1xi64> loc(#loc949) + %p_A00_490 = arith.muli %p_A00_486, %p_A00_489 : tensor<16x1xi64> loc(#loc949) + %p_A00_491 = tt.splat %A_37 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc950) + %p_A00_492 = tt.addptr %p_A00_491, %p_A00_490 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc950) + %p_A00_493 = tt.expand_dims %o_i {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc951) + %p_A00_494 = tt.broadcast %p_A00_492 : tensor<16x1x!tt.ptr> -> tensor<16x16x!tt.ptr> loc(#loc952) + %p_A00_495 = tt.broadcast %p_A00_493 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc952) + %p_A00_496 = tt.addptr %p_A00_494, %p_A00_495 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc952) + %p_A10 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc953) + %p_A10_497 = tt.splat %i_tc1_8 : i64 -> tensor<16xi64> loc(#loc953) + %p_A10_498 = arith.addi %p_A10_497, %p_A10 : tensor<16xi64> loc(#loc953) + %p_A10_499 = tt.expand_dims %p_A10_498 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc954) + %p_A10_500 = arith.constant 2048 : i32 loc(#loc955) + %p_A10_501 = arith.constant 2048 : i64 loc(#loc955) + %p_A10_502 = arith.constant dense<2048> : tensor<16x1xi64> loc(#loc955) + %p_A10_503 = arith.muli %p_A10_499, %p_A10_502 : tensor<16x1xi64> loc(#loc955) + %p_A10_504 = tt.splat %A_37 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc956) + %p_A10_505 = tt.addptr %p_A10_504, %p_A10_503 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc956) + %p_A10_506 = tt.expand_dims %o_i {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc957) + %p_A10_507 = tt.broadcast %p_A10_505 : tensor<16x1x!tt.ptr> -> tensor<16x16x!tt.ptr> loc(#loc958) + %p_A10_508 = tt.broadcast %p_A10_506 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc958) + %p_A10_509 = tt.addptr %p_A10_507, %p_A10_508 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc958) + %p_A11 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc959) + %p_A11_510 = tt.splat %i_tc1_8 : i64 -> tensor<16xi64> loc(#loc959) + %p_A11_511 = arith.addi %p_A11_510, %p_A11 : tensor<16xi64> loc(#loc959) + %p_A11_512 = tt.expand_dims %p_A11_511 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc960) + %p_A11_513 = arith.constant 2048 : i32 loc(#loc961) + %p_A11_514 = arith.constant 2048 : i64 loc(#loc961) + %p_A11_515 = arith.constant dense<2048> : tensor<16x1xi64> loc(#loc961) + %p_A11_516 = arith.muli %p_A11_512, %p_A11_515 : tensor<16x1xi64> loc(#loc961) + %p_A11_517 = tt.splat %A_37 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc962) + %p_A11_518 = tt.addptr %p_A11_517, %p_A11_516 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc962) + %p_A11_519 = arith.constant 16 : i32 loc(#loc963) + %p_A11_520 = arith.constant 16 : i32 loc(#loc963) + %p_A11_521 = arith.constant dense<16> : tensor<16xi32> loc(#loc963) + %p_A11_522 = arith.extsi %p_A11_521 : tensor<16xi32> to tensor<16xi64> loc(#loc963) + %p_A11_523 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc963) + %p_A11_524 = arith.addi %p_A11_522, %p_A11_523 : tensor<16xi64> loc(#loc963) + %p_A11_525 = arith.constant 2147483647 : i64 loc(#loc963) + %p_A11_526 = arith.constant -2147483648 : i64 loc(#loc963) + %p_A11_527 = arith.constant dense<2147483647> : tensor<16xi64> loc(#loc963) + %p_A11_528 = arith.cmpi sle, %p_A11_524, %p_A11_527 : tensor<16xi64> loc(#loc963) + %p_A11_529 = arith.constant dense<-2147483648> : tensor<16xi64> loc(#loc963) + %p_A11_530 = arith.cmpi sge, %p_A11_524, %p_A11_529 : tensor<16xi64> loc(#loc963) + %p_A11_531 = arith.andi %p_A11_528, %p_A11_530 : tensor<16xi1> loc(#loc963) + %p_A11_532 = arith.addi %p_A11_521, %o_i : tensor<16xi32> loc(#loc963) + %p_A11_533 = tt.expand_dims %p_A11_532 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc964) + %p_A11_534 = tt.broadcast %p_A11_518 : tensor<16x1x!tt.ptr> -> tensor<16x16x!tt.ptr> loc(#loc965) + %p_A11_535 = tt.broadcast %p_A11_533 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc965) + %p_A11_536 = tt.addptr %p_A11_534, %p_A11_535 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc965) + %p_A20 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc966) + %p_A20_537 = tt.splat %i_tc2_13 : i64 -> tensor<16xi64> loc(#loc966) + %p_A20_538 = arith.addi %p_A20_537, %p_A20 : tensor<16xi64> loc(#loc966) + %p_A20_539 = tt.expand_dims %p_A20_538 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc967) + %p_A20_540 = arith.constant 2048 : i32 loc(#loc968) + %p_A20_541 = arith.constant 2048 : i64 loc(#loc968) + %p_A20_542 = arith.constant dense<2048> : tensor<16x1xi64> loc(#loc968) + %p_A20_543 = arith.muli %p_A20_539, %p_A20_542 : tensor<16x1xi64> loc(#loc968) + %p_A20_544 = tt.splat %A_37 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc969) + %p_A20_545 = tt.addptr %p_A20_544, %p_A20_543 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc969) + %p_A20_546 = tt.expand_dims %o_i {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc970) + %p_A20_547 = tt.broadcast %p_A20_545 : tensor<16x1x!tt.ptr> -> tensor<16x16x!tt.ptr> loc(#loc971) + %p_A20_548 = tt.broadcast %p_A20_546 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc971) + %p_A20_549 = tt.addptr %p_A20_547, %p_A20_548 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc971) + %p_A21 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc972) + %p_A21_550 = tt.splat %i_tc2_13 : i64 -> tensor<16xi64> loc(#loc972) + %p_A21_551 = arith.addi %p_A21_550, %p_A21 : tensor<16xi64> loc(#loc972) + %p_A21_552 = tt.expand_dims %p_A21_551 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc973) + %p_A21_553 = arith.constant 2048 : i32 loc(#loc974) + %p_A21_554 = arith.constant 2048 : i64 loc(#loc974) + %p_A21_555 = arith.constant dense<2048> : tensor<16x1xi64> loc(#loc974) + %p_A21_556 = arith.muli %p_A21_552, %p_A21_555 : tensor<16x1xi64> loc(#loc974) + %p_A21_557 = tt.splat %A_37 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc975) + %p_A21_558 = tt.addptr %p_A21_557, %p_A21_556 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc975) + %p_A21_559 = arith.constant 16 : i32 loc(#loc976) + %p_A21_560 = arith.constant 16 : i32 loc(#loc976) + %p_A21_561 = arith.constant dense<16> : tensor<16xi32> loc(#loc976) + %p_A21_562 = arith.extsi %p_A21_561 : tensor<16xi32> to tensor<16xi64> loc(#loc976) + %p_A21_563 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc976) + %p_A21_564 = arith.addi %p_A21_562, %p_A21_563 : tensor<16xi64> loc(#loc976) + %p_A21_565 = arith.constant 2147483647 : i64 loc(#loc976) + %p_A21_566 = arith.constant -2147483648 : i64 loc(#loc976) + %p_A21_567 = arith.constant dense<2147483647> : tensor<16xi64> loc(#loc976) + %p_A21_568 = arith.cmpi sle, %p_A21_564, %p_A21_567 : tensor<16xi64> loc(#loc976) + %p_A21_569 = arith.constant dense<-2147483648> : tensor<16xi64> loc(#loc976) + %p_A21_570 = arith.cmpi sge, %p_A21_564, %p_A21_569 : tensor<16xi64> loc(#loc976) + %p_A21_571 = arith.andi %p_A21_568, %p_A21_570 : tensor<16xi1> loc(#loc976) + %p_A21_572 = arith.addi %p_A21_561, %o_i : tensor<16xi32> loc(#loc976) + %p_A21_573 = tt.expand_dims %p_A21_572 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc977) + %p_A21_574 = tt.broadcast %p_A21_558 : tensor<16x1x!tt.ptr> -> tensor<16x16x!tt.ptr> loc(#loc978) + %p_A21_575 = tt.broadcast %p_A21_573 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc978) + %p_A21_576 = tt.addptr %p_A21_574, %p_A21_575 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc978) + %p_A22 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc979) + %p_A22_577 = tt.splat %i_tc2_13 : i64 -> tensor<16xi64> loc(#loc979) + %p_A22_578 = arith.addi %p_A22_577, %p_A22 : tensor<16xi64> loc(#loc979) + %p_A22_579 = tt.expand_dims %p_A22_578 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc980) + %p_A22_580 = arith.constant 2048 : i32 loc(#loc981) + %p_A22_581 = arith.constant 2048 : i64 loc(#loc981) + %p_A22_582 = arith.constant dense<2048> : tensor<16x1xi64> loc(#loc981) + %p_A22_583 = arith.muli %p_A22_579, %p_A22_582 : tensor<16x1xi64> loc(#loc981) + %p_A22_584 = tt.splat %A_37 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc982) + %p_A22_585 = tt.addptr %p_A22_584, %p_A22_583 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc982) + %p_A22_586 = arith.constant 32 : i32 loc(#loc983) + %p_A22_587 = arith.constant 32 : i32 loc(#loc983) + %p_A22_588 = arith.constant dense<32> : tensor<16xi32> loc(#loc983) + %p_A22_589 = arith.extsi %p_A22_588 : tensor<16xi32> to tensor<16xi64> loc(#loc983) + %p_A22_590 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc983) + %p_A22_591 = arith.addi %p_A22_589, %p_A22_590 : tensor<16xi64> loc(#loc983) + %p_A22_592 = arith.constant 2147483647 : i64 loc(#loc983) + %p_A22_593 = arith.constant -2147483648 : i64 loc(#loc983) + %p_A22_594 = arith.constant dense<2147483647> : tensor<16xi64> loc(#loc983) + %p_A22_595 = arith.cmpi sle, %p_A22_591, %p_A22_594 : tensor<16xi64> loc(#loc983) + %p_A22_596 = arith.constant dense<-2147483648> : tensor<16xi64> loc(#loc983) + %p_A22_597 = arith.cmpi sge, %p_A22_591, %p_A22_596 : tensor<16xi64> loc(#loc983) + %p_A22_598 = arith.andi %p_A22_595, %p_A22_597 : tensor<16xi1> loc(#loc983) + %p_A22_599 = arith.addi %p_A22_588, %o_i : tensor<16xi32> loc(#loc983) + %p_A22_600 = tt.expand_dims %p_A22_599 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc984) + %p_A22_601 = tt.broadcast %p_A22_585 : tensor<16x1x!tt.ptr> -> tensor<16x16x!tt.ptr> loc(#loc985) + %p_A22_602 = tt.broadcast %p_A22_600 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc985) + %p_A22_603 = tt.addptr %p_A22_601, %p_A22_602 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc985) + %p_A30 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc986) + %p_A30_604 = tt.splat %i_tc3_18 : i64 -> tensor<16xi64> loc(#loc986) + %p_A30_605 = arith.addi %p_A30_604, %p_A30 : tensor<16xi64> loc(#loc986) + %p_A30_606 = tt.expand_dims %p_A30_605 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc987) + %p_A30_607 = arith.constant 2048 : i32 loc(#loc988) + %p_A30_608 = arith.constant 2048 : i64 loc(#loc988) + %p_A30_609 = arith.constant dense<2048> : tensor<16x1xi64> loc(#loc988) + %p_A30_610 = arith.muli %p_A30_606, %p_A30_609 : tensor<16x1xi64> loc(#loc988) + %p_A30_611 = tt.splat %A_37 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc989) + %p_A30_612 = tt.addptr %p_A30_611, %p_A30_610 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc989) + %p_A30_613 = tt.expand_dims %o_i {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc990) + %p_A30_614 = tt.broadcast %p_A30_612 : tensor<16x1x!tt.ptr> -> tensor<16x16x!tt.ptr> loc(#loc991) + %p_A30_615 = tt.broadcast %p_A30_613 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc991) + %p_A30_616 = tt.addptr %p_A30_614, %p_A30_615 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc991) + %p_A31 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc992) + %p_A31_617 = tt.splat %i_tc3_18 : i64 -> tensor<16xi64> loc(#loc992) + %p_A31_618 = arith.addi %p_A31_617, %p_A31 : tensor<16xi64> loc(#loc992) + %p_A31_619 = tt.expand_dims %p_A31_618 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc993) + %p_A31_620 = arith.constant 2048 : i32 loc(#loc994) + %p_A31_621 = arith.constant 2048 : i64 loc(#loc994) + %p_A31_622 = arith.constant dense<2048> : tensor<16x1xi64> loc(#loc994) + %p_A31_623 = arith.muli %p_A31_619, %p_A31_622 : tensor<16x1xi64> loc(#loc994) + %p_A31_624 = tt.splat %A_37 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc995) + %p_A31_625 = tt.addptr %p_A31_624, %p_A31_623 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc995) + %p_A31_626 = arith.constant 16 : i32 loc(#loc996) + %p_A31_627 = arith.constant 16 : i32 loc(#loc996) + %p_A31_628 = arith.constant dense<16> : tensor<16xi32> loc(#loc996) + %p_A31_629 = arith.extsi %p_A31_628 : tensor<16xi32> to tensor<16xi64> loc(#loc996) + %p_A31_630 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc996) + %p_A31_631 = arith.addi %p_A31_629, %p_A31_630 : tensor<16xi64> loc(#loc996) + %p_A31_632 = arith.constant 2147483647 : i64 loc(#loc996) + %p_A31_633 = arith.constant -2147483648 : i64 loc(#loc996) + %p_A31_634 = arith.constant dense<2147483647> : tensor<16xi64> loc(#loc996) + %p_A31_635 = arith.cmpi sle, %p_A31_631, %p_A31_634 : tensor<16xi64> loc(#loc996) + %p_A31_636 = arith.constant dense<-2147483648> : tensor<16xi64> loc(#loc996) + %p_A31_637 = arith.cmpi sge, %p_A31_631, %p_A31_636 : tensor<16xi64> loc(#loc996) + %p_A31_638 = arith.andi %p_A31_635, %p_A31_637 : tensor<16xi1> loc(#loc996) + %p_A31_639 = arith.addi %p_A31_628, %o_i : tensor<16xi32> loc(#loc996) + %p_A31_640 = tt.expand_dims %p_A31_639 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc997) + %p_A31_641 = tt.broadcast %p_A31_625 : tensor<16x1x!tt.ptr> -> tensor<16x16x!tt.ptr> loc(#loc998) + %p_A31_642 = tt.broadcast %p_A31_640 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc998) + %p_A31_643 = tt.addptr %p_A31_641, %p_A31_642 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc998) + %p_A32 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc999) + %p_A32_644 = tt.splat %i_tc3_18 : i64 -> tensor<16xi64> loc(#loc999) + %p_A32_645 = arith.addi %p_A32_644, %p_A32 : tensor<16xi64> loc(#loc999) + %p_A32_646 = tt.expand_dims %p_A32_645 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc1000) + %p_A32_647 = arith.constant 2048 : i32 loc(#loc1001) + %p_A32_648 = arith.constant 2048 : i64 loc(#loc1001) + %p_A32_649 = arith.constant dense<2048> : tensor<16x1xi64> loc(#loc1001) + %p_A32_650 = arith.muli %p_A32_646, %p_A32_649 : tensor<16x1xi64> loc(#loc1001) + %p_A32_651 = tt.splat %A_37 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc1002) + %p_A32_652 = tt.addptr %p_A32_651, %p_A32_650 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc1002) + %p_A32_653 = arith.constant 32 : i32 loc(#loc1003) + %p_A32_654 = arith.constant 32 : i32 loc(#loc1003) + %p_A32_655 = arith.constant dense<32> : tensor<16xi32> loc(#loc1003) + %p_A32_656 = arith.extsi %p_A32_655 : tensor<16xi32> to tensor<16xi64> loc(#loc1003) + %p_A32_657 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc1003) + %p_A32_658 = arith.addi %p_A32_656, %p_A32_657 : tensor<16xi64> loc(#loc1003) + %p_A32_659 = arith.constant 2147483647 : i64 loc(#loc1003) + %p_A32_660 = arith.constant -2147483648 : i64 loc(#loc1003) + %p_A32_661 = arith.constant dense<2147483647> : tensor<16xi64> loc(#loc1003) + %p_A32_662 = arith.cmpi sle, %p_A32_658, %p_A32_661 : tensor<16xi64> loc(#loc1003) + %p_A32_663 = arith.constant dense<-2147483648> : tensor<16xi64> loc(#loc1003) + %p_A32_664 = arith.cmpi sge, %p_A32_658, %p_A32_663 : tensor<16xi64> loc(#loc1003) + %p_A32_665 = arith.andi %p_A32_662, %p_A32_664 : tensor<16xi1> loc(#loc1003) + %p_A32_666 = arith.addi %p_A32_655, %o_i : tensor<16xi32> loc(#loc1003) + %p_A32_667 = tt.expand_dims %p_A32_666 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc1004) + %p_A32_668 = tt.broadcast %p_A32_652 : tensor<16x1x!tt.ptr> -> tensor<16x16x!tt.ptr> loc(#loc1005) + %p_A32_669 = tt.broadcast %p_A32_667 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc1005) + %p_A32_670 = tt.addptr %p_A32_668, %p_A32_669 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc1005) + %p_A33 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc1006) + %p_A33_671 = tt.splat %i_tc3_18 : i64 -> tensor<16xi64> loc(#loc1006) + %p_A33_672 = arith.addi %p_A33_671, %p_A33 : tensor<16xi64> loc(#loc1006) + %p_A33_673 = tt.expand_dims %p_A33_672 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc1007) + %p_A33_674 = arith.constant 2048 : i32 loc(#loc1008) + %p_A33_675 = arith.constant 2048 : i64 loc(#loc1008) + %p_A33_676 = arith.constant dense<2048> : tensor<16x1xi64> loc(#loc1008) + %p_A33_677 = arith.muli %p_A33_673, %p_A33_676 : tensor<16x1xi64> loc(#loc1008) + %p_A33_678 = tt.splat %A_37 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc1009) + %p_A33_679 = tt.addptr %p_A33_678, %p_A33_677 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc1009) + %p_A33_680 = arith.constant 48 : i32 loc(#loc1010) + %p_A33_681 = arith.constant 48 : i32 loc(#loc1010) + %p_A33_682 = arith.constant dense<48> : tensor<16xi32> loc(#loc1010) + %p_A33_683 = arith.extsi %p_A33_682 : tensor<16xi32> to tensor<16xi64> loc(#loc1010) + %p_A33_684 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc1010) + %p_A33_685 = arith.addi %p_A33_683, %p_A33_684 : tensor<16xi64> loc(#loc1010) + %p_A33_686 = arith.constant 2147483647 : i64 loc(#loc1010) + %p_A33_687 = arith.constant -2147483648 : i64 loc(#loc1010) + %p_A33_688 = arith.constant dense<2147483647> : tensor<16xi64> loc(#loc1010) + %p_A33_689 = arith.cmpi sle, %p_A33_685, %p_A33_688 : tensor<16xi64> loc(#loc1010) + %p_A33_690 = arith.constant dense<-2147483648> : tensor<16xi64> loc(#loc1010) + %p_A33_691 = arith.cmpi sge, %p_A33_685, %p_A33_690 : tensor<16xi64> loc(#loc1010) + %p_A33_692 = arith.andi %p_A33_689, %p_A33_691 : tensor<16xi1> loc(#loc1010) + %p_A33_693 = arith.addi %p_A33_682, %o_i : tensor<16xi32> loc(#loc1010) + %p_A33_694 = tt.expand_dims %p_A33_693 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc1011) + %p_A33_695 = tt.broadcast %p_A33_679 : tensor<16x1x!tt.ptr> -> tensor<16x16x!tt.ptr> loc(#loc1012) + %p_A33_696 = tt.broadcast %p_A33_694 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc1012) + %p_A33_697 = tt.addptr %p_A33_695, %p_A33_696 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc1012) + %m_A0 = tt.expand_dims %m_tc0_42 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc1013) + %m_A0_698 = tt.expand_dims %o_i {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc1014) + %m_A0_699 = arith.constant 64 : i32 loc(#loc1015) + %m_A0_700 = arith.constant dense<64> : tensor<1x16xi32> loc(#loc1015) + %m_A0_701 = arith.cmpi slt, %m_A0_698, %m_A0_700 : tensor<1x16xi32> loc(#loc1015) + %m_A0_702 = tt.broadcast %m_A0 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc1016) + %m_A0_703 = tt.broadcast %m_A0_701 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc1016) + %m_A0_704 = arith.andi %m_A0_702, %m_A0_703 : tensor<16x16xi1> loc(#loc1016) + %m_A1 = tt.expand_dims %m_tc1_47 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc1017) + %m_A1_705 = tt.expand_dims %o_i {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc1018) + %m_A1_706 = arith.constant 64 : i32 loc(#loc1019) + %m_A1_707 = arith.constant dense<64> : tensor<1x16xi32> loc(#loc1019) + %m_A1_708 = arith.cmpi slt, %m_A1_705, %m_A1_707 : tensor<1x16xi32> loc(#loc1019) + %m_A1_709 = tt.broadcast %m_A1 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc1020) + %m_A1_710 = tt.broadcast %m_A1_708 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc1020) + %m_A1_711 = arith.andi %m_A1_709, %m_A1_710 : tensor<16x16xi1> loc(#loc1020) + %m_A2 = tt.expand_dims %m_tc2_52 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc1021) + %m_A2_712 = tt.expand_dims %o_i {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc1022) + %m_A2_713 = arith.constant 64 : i32 loc(#loc1023) + %m_A2_714 = arith.constant dense<64> : tensor<1x16xi32> loc(#loc1023) + %m_A2_715 = arith.cmpi slt, %m_A2_712, %m_A2_714 : tensor<1x16xi32> loc(#loc1023) + %m_A2_716 = tt.broadcast %m_A2 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc1024) + %m_A2_717 = tt.broadcast %m_A2_715 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc1024) + %m_A2_718 = arith.andi %m_A2_716, %m_A2_717 : tensor<16x16xi1> loc(#loc1024) + %m_A3 = tt.expand_dims %m_tc3_57 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc1025) + %m_A3_719 = tt.expand_dims %o_i {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc1026) + %m_A3_720 = arith.constant 64 : i32 loc(#loc1027) + %m_A3_721 = arith.constant dense<64> : tensor<1x16xi32> loc(#loc1027) + %m_A3_722 = arith.cmpi slt, %m_A3_719, %m_A3_721 : tensor<1x16xi32> loc(#loc1027) + %m_A3_723 = tt.broadcast %m_A3 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc1028) + %m_A3_724 = tt.broadcast %m_A3_722 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc1028) + %m_A3_725 = arith.andi %m_A3_723, %m_A3_724 : tensor<16x16xi1> loc(#loc1028) + %m_A11 = tt.expand_dims %m_tc1_47 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc1029) + %m_A11_726 = arith.constant 16 : i32 loc(#loc1030) + %m_A11_727 = arith.constant 16 : i32 loc(#loc1030) + %m_A11_728 = arith.constant dense<16> : tensor<16xi32> loc(#loc1030) + %m_A11_729 = arith.extsi %m_A11_728 : tensor<16xi32> to tensor<16xi64> loc(#loc1030) + %m_A11_730 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc1030) + %m_A11_731 = arith.addi %m_A11_729, %m_A11_730 : tensor<16xi64> loc(#loc1030) + %m_A11_732 = arith.constant 2147483647 : i64 loc(#loc1030) + %m_A11_733 = arith.constant -2147483648 : i64 loc(#loc1030) + %m_A11_734 = arith.constant dense<2147483647> : tensor<16xi64> loc(#loc1030) + %m_A11_735 = arith.cmpi sle, %m_A11_731, %m_A11_734 : tensor<16xi64> loc(#loc1030) + %m_A11_736 = arith.constant dense<-2147483648> : tensor<16xi64> loc(#loc1030) + %m_A11_737 = arith.cmpi sge, %m_A11_731, %m_A11_736 : tensor<16xi64> loc(#loc1030) + %m_A11_738 = arith.andi %m_A11_735, %m_A11_737 : tensor<16xi1> loc(#loc1030) + %m_A11_739 = arith.addi %m_A11_728, %o_i : tensor<16xi32> loc(#loc1030) + %m_A11_740 = tt.expand_dims %m_A11_739 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc1031) + %m_A11_741 = arith.constant 64 : i32 loc(#loc1032) + %m_A11_742 = arith.constant dense<64> : tensor<1x16xi32> loc(#loc1032) + %m_A11_743 = arith.cmpi slt, %m_A11_740, %m_A11_742 : tensor<1x16xi32> loc(#loc1032) + %m_A11_744 = tt.broadcast %m_A11 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc1033) + %m_A11_745 = tt.broadcast %m_A11_743 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc1033) + %m_A11_746 = arith.andi %m_A11_744, %m_A11_745 : tensor<16x16xi1> loc(#loc1033) + %m_A21 = tt.expand_dims %m_tc2_52 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc1034) + %m_A21_747 = arith.constant 16 : i32 loc(#loc1035) + %m_A21_748 = arith.constant 16 : i32 loc(#loc1035) + %m_A21_749 = arith.constant dense<16> : tensor<16xi32> loc(#loc1035) + %m_A21_750 = arith.extsi %m_A21_749 : tensor<16xi32> to tensor<16xi64> loc(#loc1035) + %m_A21_751 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc1035) + %m_A21_752 = arith.addi %m_A21_750, %m_A21_751 : tensor<16xi64> loc(#loc1035) + %m_A21_753 = arith.constant 2147483647 : i64 loc(#loc1035) + %m_A21_754 = arith.constant -2147483648 : i64 loc(#loc1035) + %m_A21_755 = arith.constant dense<2147483647> : tensor<16xi64> loc(#loc1035) + %m_A21_756 = arith.cmpi sle, %m_A21_752, %m_A21_755 : tensor<16xi64> loc(#loc1035) + %m_A21_757 = arith.constant dense<-2147483648> : tensor<16xi64> loc(#loc1035) + %m_A21_758 = arith.cmpi sge, %m_A21_752, %m_A21_757 : tensor<16xi64> loc(#loc1035) + %m_A21_759 = arith.andi %m_A21_756, %m_A21_758 : tensor<16xi1> loc(#loc1035) + %m_A21_760 = arith.addi %m_A21_749, %o_i : tensor<16xi32> loc(#loc1035) + %m_A21_761 = tt.expand_dims %m_A21_760 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc1036) + %m_A21_762 = arith.constant 64 : i32 loc(#loc1037) + %m_A21_763 = arith.constant dense<64> : tensor<1x16xi32> loc(#loc1037) + %m_A21_764 = arith.cmpi slt, %m_A21_761, %m_A21_763 : tensor<1x16xi32> loc(#loc1037) + %m_A21_765 = tt.broadcast %m_A21 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc1038) + %m_A21_766 = tt.broadcast %m_A21_764 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc1038) + %m_A21_767 = arith.andi %m_A21_765, %m_A21_766 : tensor<16x16xi1> loc(#loc1038) + %m_A22 = tt.expand_dims %m_tc2_52 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc1039) + %m_A22_768 = arith.constant 32 : i32 loc(#loc1040) + %m_A22_769 = arith.constant 32 : i32 loc(#loc1040) + %m_A22_770 = arith.constant dense<32> : tensor<16xi32> loc(#loc1040) + %m_A22_771 = arith.extsi %m_A22_770 : tensor<16xi32> to tensor<16xi64> loc(#loc1040) + %m_A22_772 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc1040) + %m_A22_773 = arith.addi %m_A22_771, %m_A22_772 : tensor<16xi64> loc(#loc1040) + %m_A22_774 = arith.constant 2147483647 : i64 loc(#loc1040) + %m_A22_775 = arith.constant -2147483648 : i64 loc(#loc1040) + %m_A22_776 = arith.constant dense<2147483647> : tensor<16xi64> loc(#loc1040) + %m_A22_777 = arith.cmpi sle, %m_A22_773, %m_A22_776 : tensor<16xi64> loc(#loc1040) + %m_A22_778 = arith.constant dense<-2147483648> : tensor<16xi64> loc(#loc1040) + %m_A22_779 = arith.cmpi sge, %m_A22_773, %m_A22_778 : tensor<16xi64> loc(#loc1040) + %m_A22_780 = arith.andi %m_A22_777, %m_A22_779 : tensor<16xi1> loc(#loc1040) + %m_A22_781 = arith.addi %m_A22_770, %o_i : tensor<16xi32> loc(#loc1040) + %m_A22_782 = tt.expand_dims %m_A22_781 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc1041) + %m_A22_783 = arith.constant 64 : i32 loc(#loc1042) + %m_A22_784 = arith.constant dense<64> : tensor<1x16xi32> loc(#loc1042) + %m_A22_785 = arith.cmpi slt, %m_A22_782, %m_A22_784 : tensor<1x16xi32> loc(#loc1042) + %m_A22_786 = tt.broadcast %m_A22 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc1043) + %m_A22_787 = tt.broadcast %m_A22_785 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc1043) + %m_A22_788 = arith.andi %m_A22_786, %m_A22_787 : tensor<16x16xi1> loc(#loc1043) + %m_A31 = tt.expand_dims %m_tc3_57 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc1044) + %m_A31_789 = arith.constant 16 : i32 loc(#loc1045) + %m_A31_790 = arith.constant 16 : i32 loc(#loc1045) + %m_A31_791 = arith.constant dense<16> : tensor<16xi32> loc(#loc1045) + %m_A31_792 = arith.extsi %m_A31_791 : tensor<16xi32> to tensor<16xi64> loc(#loc1045) + %m_A31_793 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc1045) + %m_A31_794 = arith.addi %m_A31_792, %m_A31_793 : tensor<16xi64> loc(#loc1045) + %m_A31_795 = arith.constant 2147483647 : i64 loc(#loc1045) + %m_A31_796 = arith.constant -2147483648 : i64 loc(#loc1045) + %m_A31_797 = arith.constant dense<2147483647> : tensor<16xi64> loc(#loc1045) + %m_A31_798 = arith.cmpi sle, %m_A31_794, %m_A31_797 : tensor<16xi64> loc(#loc1045) + %m_A31_799 = arith.constant dense<-2147483648> : tensor<16xi64> loc(#loc1045) + %m_A31_800 = arith.cmpi sge, %m_A31_794, %m_A31_799 : tensor<16xi64> loc(#loc1045) + %m_A31_801 = arith.andi %m_A31_798, %m_A31_800 : tensor<16xi1> loc(#loc1045) + %m_A31_802 = arith.addi %m_A31_791, %o_i : tensor<16xi32> loc(#loc1045) + %m_A31_803 = tt.expand_dims %m_A31_802 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc1046) + %m_A31_804 = arith.constant 64 : i32 loc(#loc1047) + %m_A31_805 = arith.constant dense<64> : tensor<1x16xi32> loc(#loc1047) + %m_A31_806 = arith.cmpi slt, %m_A31_803, %m_A31_805 : tensor<1x16xi32> loc(#loc1047) + %m_A31_807 = tt.broadcast %m_A31 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc1048) + %m_A31_808 = tt.broadcast %m_A31_806 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc1048) + %m_A31_809 = arith.andi %m_A31_807, %m_A31_808 : tensor<16x16xi1> loc(#loc1048) + %m_A32 = tt.expand_dims %m_tc3_57 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc1049) + %m_A32_810 = arith.constant 32 : i32 loc(#loc1050) + %m_A32_811 = arith.constant 32 : i32 loc(#loc1050) + %m_A32_812 = arith.constant dense<32> : tensor<16xi32> loc(#loc1050) + %m_A32_813 = arith.extsi %m_A32_812 : tensor<16xi32> to tensor<16xi64> loc(#loc1050) + %m_A32_814 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc1050) + %m_A32_815 = arith.addi %m_A32_813, %m_A32_814 : tensor<16xi64> loc(#loc1050) + %m_A32_816 = arith.constant 2147483647 : i64 loc(#loc1050) + %m_A32_817 = arith.constant -2147483648 : i64 loc(#loc1050) + %m_A32_818 = arith.constant dense<2147483647> : tensor<16xi64> loc(#loc1050) + %m_A32_819 = arith.cmpi sle, %m_A32_815, %m_A32_818 : tensor<16xi64> loc(#loc1050) + %m_A32_820 = arith.constant dense<-2147483648> : tensor<16xi64> loc(#loc1050) + %m_A32_821 = arith.cmpi sge, %m_A32_815, %m_A32_820 : tensor<16xi64> loc(#loc1050) + %m_A32_822 = arith.andi %m_A32_819, %m_A32_821 : tensor<16xi1> loc(#loc1050) + %m_A32_823 = arith.addi %m_A32_812, %o_i : tensor<16xi32> loc(#loc1050) + %m_A32_824 = tt.expand_dims %m_A32_823 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc1051) + %m_A32_825 = arith.constant 64 : i32 loc(#loc1052) + %m_A32_826 = arith.constant dense<64> : tensor<1x16xi32> loc(#loc1052) + %m_A32_827 = arith.cmpi slt, %m_A32_824, %m_A32_826 : tensor<1x16xi32> loc(#loc1052) + %m_A32_828 = tt.broadcast %m_A32 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc1053) + %m_A32_829 = tt.broadcast %m_A32_827 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc1053) + %m_A32_830 = arith.andi %m_A32_828, %m_A32_829 : tensor<16x16xi1> loc(#loc1053) + %m_A33 = tt.expand_dims %m_tc3_57 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc1054) + %m_A33_831 = arith.constant 48 : i32 loc(#loc1055) + %m_A33_832 = arith.constant 48 : i32 loc(#loc1055) + %m_A33_833 = arith.constant dense<48> : tensor<16xi32> loc(#loc1055) + %m_A33_834 = arith.extsi %m_A33_833 : tensor<16xi32> to tensor<16xi64> loc(#loc1055) + %m_A33_835 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc1055) + %m_A33_836 = arith.addi %m_A33_834, %m_A33_835 : tensor<16xi64> loc(#loc1055) + %m_A33_837 = arith.constant 2147483647 : i64 loc(#loc1055) + %m_A33_838 = arith.constant -2147483648 : i64 loc(#loc1055) + %m_A33_839 = arith.constant dense<2147483647> : tensor<16xi64> loc(#loc1055) + %m_A33_840 = arith.cmpi sle, %m_A33_836, %m_A33_839 : tensor<16xi64> loc(#loc1055) + %m_A33_841 = arith.constant dense<-2147483648> : tensor<16xi64> loc(#loc1055) + %m_A33_842 = arith.cmpi sge, %m_A33_836, %m_A33_841 : tensor<16xi64> loc(#loc1055) + %m_A33_843 = arith.andi %m_A33_840, %m_A33_842 : tensor<16xi1> loc(#loc1055) + %m_A33_844 = arith.addi %m_A33_833, %o_i : tensor<16xi32> loc(#loc1055) + %m_A33_845 = tt.expand_dims %m_A33_844 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc1056) + %m_A33_846 = arith.constant 64 : i32 loc(#loc1057) + %m_A33_847 = arith.constant dense<64> : tensor<1x16xi32> loc(#loc1057) + %m_A33_848 = arith.cmpi slt, %m_A33_845, %m_A33_847 : tensor<1x16xi32> loc(#loc1057) + %m_A33_849 = tt.broadcast %m_A33 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc1058) + %m_A33_850 = tt.broadcast %m_A33_848 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc1058) + %m_A33_851 = arith.andi %m_A33_849, %m_A33_850 : tensor<16x16xi1> loc(#loc1058) + %51 = arith.truncf %b_Ai00_413 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc522) + tt.store %p_A00_496, %51, %m_A0_704 : tensor<16x16x!tt.ptr> loc(#loc523) + %52 = arith.truncf %b_Ai10_427 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc524) + tt.store %p_A10_509, %52, %m_A1_711 : tensor<16x16x!tt.ptr> loc(#loc525) + %53 = arith.truncf %b_Ai11_415 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc526) + tt.store %p_A11_536, %53, %m_A11_746 : tensor<16x16x!tt.ptr> loc(#loc527) + %54 = arith.truncf %b_Ai20_455 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc528) + tt.store %p_A20_549, %54, %m_A2_718 : tensor<16x16x!tt.ptr> loc(#loc529) + %55 = arith.truncf %b_Ai21_435 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc530) + tt.store %p_A21_576, %55, %m_A21_767 : tensor<16x16x!tt.ptr> loc(#loc531) + %56 = arith.truncf %b_Ai22_417 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc532) + tt.store %p_A22_603, %56, %m_A22_788 : tensor<16x16x!tt.ptr> loc(#loc533) + %57 = arith.truncf %b_Ai30_483 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc534) + tt.store %p_A30_616, %57, %m_A3_725 : tensor<16x16x!tt.ptr> loc(#loc535) + %58 = arith.truncf %b_Ai31_467 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc536) + tt.store %p_A31_643, %58, %m_A31_809 : tensor<16x16x!tt.ptr> loc(#loc537) + %59 = arith.truncf %b_Ai32_443 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc538) + tt.store %p_A32_670, %59, %m_A32_830 : tensor<16x16x!tt.ptr> loc(#loc539) + %60 = arith.truncf %b_Ai33_419 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc540) + tt.store %p_A33_697, %60, %m_A33_851 : tensor<16x16x!tt.ptr> loc(#loc541) + tt.return loc(#loc542) + } loc(#loc) + tt.func private @triton.language.standard.zeros__Tc16_c16T_cfp32() -> tensor<16x16xf32> attributes {noinline = false} { + %cst = arith.constant 0.000000e+00 : f32 loc(#loc544) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc544) + tt.return %cst_0 : tensor<16x16xf32> loc(#loc545) + ^bb1: // no predecessors + %0 = ub.poison : tensor<16x16xf32> loc(#loc546) + tt.return %0 : tensor<16x16xf32> loc(#loc546) + } loc(#loc543) + tt.func private @triton.language.standard.cdiv__c128_c64() attributes {noinline = false} { + tt.return loc(#loc548) + ^bb1: // no predecessors + tt.return loc(#loc549) + } loc(#loc547) + tt.func private @fla.ops.utils.op.exp2__fp32S16_16S(%x: tensor<16x16xf32> loc("x"(#loc550))) -> tensor<16x16xf32> attributes {noinline = false} { + %0 = math.exp2 %x : tensor<16x16xf32> loc(#loc551) + tt.return %0 : tensor<16x16xf32> loc(#loc552) + ^bb1: // no predecessors + %1 = ub.poison : tensor<16x16xf32> loc(#loc553) + tt.return %1 : tensor<16x16xf32> loc(#loc553) + } loc(#loc550) + tt.func private @triton.language.standard.sum__fp32S16_16S_c0_cFalse_cNone(%input: tensor<16x16xf32> loc("input"(#loc554))) -> tensor<16xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 0 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc555) + tt.reduce.return %2 : f32 loc(#loc555) + }) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc555) + tt.return %0 : tensor<16xf32> loc(#loc557) + ^bb1: // no predecessors + %1 = ub.poison : tensor<16xf32> loc(#loc558) + tt.return %1 : tensor<16xf32> loc(#loc558) + } loc(#loc554) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc559)), %b: f32 loc("b"(#loc559))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc560) + tt.return %0 : f32 loc(#loc561) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc562) + tt.return %1 : f32 loc(#loc562) + } loc(#loc559) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":70:30) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":70:36) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":70:61) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":70:67) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":71:23) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":71:34) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":78:25) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":78:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":78:38) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":80:13) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":80:19) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":81:8) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":83:18) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":84:18) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":84:23) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":85:18) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":85:23) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":86:18) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":86:23) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":88:16) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":88:28) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":88:20) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":88:40) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":88:9) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":89:16) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":89:21) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":89:28) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":89:9) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":91:23) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":92:21) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":92:28) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":93:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":93:28) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":94:21) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":94:28) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":95:21) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":95:28) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":98:24) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":98:18) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":98:29) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":98:44) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":98:51) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":98:35) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":99:24) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":99:18) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":99:29) +#loc47 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":99:44) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":99:51) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":99:35) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":100:24) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":100:18) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":100:29) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":100:44) +#loc54 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":100:51) +#loc55 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":100:35) +#loc56 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":101:24) +#loc57 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":101:18) +#loc58 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":101:29) +#loc59 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":101:44) +#loc60 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":101:51) +#loc61 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":101:35) +#loc62 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":102:19) +#loc63 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":102:51) +#loc64 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":103:19) +#loc65 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":103:51) +#loc66 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":104:19) +#loc67 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":104:51) +#loc68 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":105:19) +#loc69 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":105:51) +#loc70 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":109:25) +#loc71 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":109:19) +#loc72 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":109:30) +#loc73 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":109:45) +#loc74 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":109:52) +#loc75 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":109:36) +#loc76 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":110:25) +#loc77 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":110:19) +#loc78 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":110:30) +#loc79 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":110:45) +#loc80 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":110:52) +#loc81 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":110:36) +#loc82 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":111:25) +#loc83 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":111:19) +#loc84 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":111:30) +#loc85 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":111:45) +#loc86 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":111:52) +#loc87 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":111:36) +#loc88 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":112:25) +#loc89 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":112:19) +#loc90 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":112:30) +#loc91 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":112:45) +#loc92 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":112:52) +#loc93 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":112:36) +#loc94 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":114:23) +#loc95 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":115:23) +#loc96 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":116:23) +#loc97 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":117:23) +#loc98 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":124:21) +#loc99 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":125:21) +#loc100 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":126:21) +#loc101 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":127:21) +#loc102 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":130:21) +#loc103 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":131:21) +#loc104 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":132:21) +#loc105 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":133:21) +#loc106 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":134:21) +#loc107 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":135:21) +#loc108 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":137:32) +#loc109 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":137:21) +#loc110 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":138:20) +#loc111 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":138:38) +#loc112 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":138:25) +#loc113 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:28) +#loc114 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:33) +#loc115 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:45) +#loc116 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:19) +#loc117 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:56) +#loc118 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:52) +#loc119 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":140:40) +#loc120 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":140:56) +#loc121 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":140:67) +#loc122 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":140:52) +#loc123 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":140:23) +#loc124 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":142:38) +#loc125 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":142:45) +#loc126 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":144:19) +#loc127 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":144:11) +#loc128 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":145:32) +#loc129 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":145:37) +#loc130 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":145:49) +#loc131 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":145:23) +#loc132 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":145:60) +#loc133 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":145:56) +#loc134 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":146:44) +#loc135 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":146:60) +#loc136 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":146:71) +#loc137 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":146:56) +#loc138 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":146:27) +#loc139 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":148:42) +#loc140 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":148:49) +#loc141 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":150:42) +#loc142 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":150:49) +#loc143 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":152:23) +#loc144 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":152:15) +#loc145 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":153:36) +#loc146 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":153:41) +#loc147 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":153:53) +#loc148 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":153:27) +#loc149 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":153:64) +#loc150 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":153:60) +#loc151 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":154:48) +#loc152 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":154:64) +#loc153 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":154:75) +#loc154 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":154:60) +#loc155 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":154:31) +#loc156 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":156:46) +#loc157 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":156:53) +#loc158 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":158:46) +#loc159 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":158:53) +#loc160 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":159:46) +#loc161 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":159:53) +#loc162 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":161:27) +#loc163 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":161:19) +#loc164 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":162:40) +#loc165 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":162:45) +#loc166 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":162:57) +#loc167 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":162:31) +#loc168 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":162:68) +#loc169 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":162:64) +#loc170 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":163:52) +#loc171 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":163:68) +#loc172 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":163:79) +#loc173 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":163:64) +#loc174 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":163:35) +#loc175 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":165:50) +#loc176 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":165:57) +#loc177 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":167:50) +#loc178 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":167:57) +#loc179 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":168:50) +#loc180 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":168:57) +#loc181 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":169:50) +#loc182 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":169:57) +#loc183 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":144:8) +#loc184 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":179:14) +#loc185 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":179:29) +#loc186 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":179:25) +#loc187 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":180:14) +#loc188 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":180:30) +#loc189 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":180:26) +#loc190 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:38) +#loc191 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:32) +#loc192 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:55) +#loc193 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:49) +#loc194 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:75) +#loc195 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:91) +#loc196 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:86) +#loc197 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:70) +#loc198 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:102) +#loc199 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:17) +#loc200 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:38) +#loc201 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:32) +#loc202 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:55) +#loc203 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:49) +#loc204 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:75) +#loc205 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:91) +#loc206 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:86) +#loc207 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:70) +#loc208 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:102) +#loc209 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:17) +#loc210 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:38) +#loc211 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:32) +#loc212 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:55) +#loc213 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:49) +#loc214 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:75) +#loc215 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:91) +#loc216 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:86) +#loc217 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:70) +#loc218 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:102) +#loc219 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:17) +#loc220 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:38) +#loc221 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:32) +#loc222 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:55) +#loc223 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:49) +#loc224 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:75) +#loc225 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:91) +#loc226 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:86) +#loc227 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:70) +#loc228 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:102) +#loc229 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:17) +#loc230 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:32) +#loc231 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:49) +#loc232 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:43) +#loc233 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:69) +#loc234 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:85) +#loc235 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:80) +#loc236 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:64) +#loc237 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:96) +#loc238 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:17) +#loc239 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:32) +#loc240 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:49) +#loc241 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:43) +#loc242 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:69) +#loc243 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:85) +#loc244 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:80) +#loc245 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:64) +#loc246 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:96) +#loc247 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:17) +#loc248 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:32) +#loc249 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:49) +#loc250 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:43) +#loc251 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:69) +#loc252 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:85) +#loc253 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:80) +#loc254 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:64) +#loc255 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:96) +#loc256 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:17) +#loc257 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:32) +#loc258 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:49) +#loc259 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:43) +#loc260 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:69) +#loc261 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:85) +#loc262 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:80) +#loc263 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:64) +#loc264 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:96) +#loc265 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:17) +#loc266 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:32) +#loc267 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:49) +#loc268 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:43) +#loc269 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:69) +#loc270 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:85) +#loc271 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:80) +#loc272 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:64) +#loc273 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:96) +#loc274 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:17) +#loc275 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:32) +#loc276 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:49) +#loc277 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:43) +#loc278 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:69) +#loc279 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:85) +#loc280 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:80) +#loc281 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:64) +#loc282 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:96) +#loc283 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:17) +#loc284 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":201:25) +#loc285 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":201:20) +#loc286 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":202:25) +#loc287 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":202:20) +#loc288 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":203:25) +#loc289 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":203:20) +#loc290 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":204:25) +#loc291 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":204:20) +#loc292 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":207:25) +#loc293 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":207:20) +#loc294 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":208:25) +#loc295 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":208:20) +#loc296 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":209:25) +#loc297 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":209:20) +#loc298 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":210:25) +#loc299 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":210:20) +#loc300 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":211:25) +#loc301 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":211:20) +#loc302 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":212:25) +#loc303 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":212:20) +#loc304 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":222:14) +#loc305 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":223:14) +#loc306 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":224:14) +#loc307 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":225:14) +#loc308 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":227:34) +#loc309 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":227:30) +#loc310 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":227:22) +#loc311 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":228:40) +#loc312 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":228:43) +#loc313 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":228:54) +#loc314 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":228:61) +#loc315 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":228:66) +#loc316 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":229:31) +#loc317 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":229:41) +#loc318 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":230:37) +#loc319 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":230:48) +#loc320 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":230:56) +#loc321 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":230:24) +#loc322 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":231:34) +#loc323 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":231:37) +#loc324 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":231:54) +#loc325 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":231:8) +#loc326 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":232:34) +#loc327 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":232:30) +#loc328 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":232:22) +#loc329 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":233:40) +#loc330 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":233:43) +#loc331 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":233:54) +#loc332 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":233:61) +#loc333 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":233:66) +#loc334 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":234:31) +#loc335 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":234:41) +#loc336 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":235:37) +#loc337 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":235:48) +#loc338 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":235:56) +#loc339 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":235:24) +#loc340 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":236:34) +#loc341 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":236:37) +#loc342 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":236:54) +#loc343 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":236:8) +#loc344 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":237:34) +#loc345 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":237:30) +#loc346 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":237:22) +#loc347 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":238:40) +#loc348 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":238:43) +#loc349 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":238:54) +#loc350 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":238:61) +#loc351 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":238:66) +#loc352 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":239:31) +#loc353 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":239:41) +#loc354 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":240:37) +#loc355 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":240:48) +#loc356 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":240:56) +#loc357 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":240:24) +#loc358 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":241:34) +#loc359 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":241:37) +#loc360 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":241:54) +#loc361 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":241:8) +#loc362 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":242:34) +#loc363 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":242:30) +#loc364 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":242:22) +#loc365 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":243:40) +#loc366 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":243:43) +#loc367 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":243:54) +#loc368 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":243:61) +#loc369 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":243:66) +#loc370 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":244:31) +#loc371 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":244:41) +#loc372 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":245:37) +#loc373 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":245:48) +#loc374 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":245:56) +#loc375 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":245:24) +#loc376 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":246:34) +#loc377 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":246:37) +#loc378 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":246:54) +#loc379 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":246:8) +#loc380 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":248:14) +#loc381 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":249:14) +#loc382 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":250:14) +#loc383 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":251:14) +#loc384 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":258:23) +#loc385 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":259:8) +#loc386 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":257:14) +#loc387 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":263:23) +#loc388 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":264:8) +#loc389 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":262:14) +#loc390 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":268:23) +#loc391 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":269:8) +#loc392 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":267:14) +#loc393 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":275:22) +#loc394 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":276:22) +#loc395 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":276:8) +#loc396 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":275:8) +#loc397 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":273:14) +#loc398 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":281:22) +#loc399 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":282:22) +#loc400 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":282:8) +#loc401 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":281:8) +#loc402 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":279:14) +#loc403 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":287:22) +#loc404 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":288:22) +#loc405 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":288:8) +#loc406 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":289:22) +#loc407 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":289:8) +#loc408 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":287:8) +#loc409 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":285:14) +#loc410 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":297:25) +#loc411 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":297:30) +#loc412 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":297:42) +#loc413 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":297:16) +#loc414 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":297:55) +#loc415 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":297:51) +#loc416 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":298:25) +#loc417 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":298:30) +#loc418 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":298:42) +#loc419 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":298:16) +#loc420 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":298:55) +#loc421 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":298:51) +#loc422 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":299:25) +#loc423 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":299:30) +#loc424 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":299:42) +#loc425 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":299:16) +#loc426 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":299:57) +#loc427 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":299:62) +#loc428 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":299:51) +#loc429 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":300:25) +#loc430 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":300:30) +#loc431 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":300:42) +#loc432 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":300:16) +#loc433 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":300:55) +#loc434 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":300:51) +#loc435 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":301:25) +#loc436 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":301:30) +#loc437 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":301:42) +#loc438 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":301:16) +#loc439 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":301:57) +#loc440 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":301:62) +#loc441 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":301:51) +#loc442 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":302:25) +#loc443 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":302:30) +#loc444 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":302:42) +#loc445 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":302:16) +#loc446 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":302:59) +#loc447 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":302:64) +#loc448 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":302:51) +#loc449 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":303:25) +#loc450 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":303:30) +#loc451 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":303:42) +#loc452 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":303:16) +#loc453 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":303:55) +#loc454 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":303:51) +#loc455 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":304:25) +#loc456 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":304:30) +#loc457 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":304:42) +#loc458 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":304:16) +#loc459 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":304:57) +#loc460 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":304:62) +#loc461 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":304:51) +#loc462 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":305:25) +#loc463 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":305:30) +#loc464 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":305:42) +#loc465 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":305:16) +#loc466 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":305:59) +#loc467 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":305:64) +#loc468 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":305:51) +#loc469 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":306:25) +#loc470 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":306:30) +#loc471 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":306:42) +#loc472 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":306:16) +#loc473 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":306:59) +#loc474 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":306:64) +#loc475 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":306:51) +#loc476 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":308:17) +#loc477 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":308:33) +#loc478 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":308:44) +#loc479 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":308:29) +#loc480 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":309:17) +#loc481 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":309:33) +#loc482 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":309:44) +#loc483 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":309:29) +#loc484 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":310:17) +#loc485 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":310:33) +#loc486 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":310:44) +#loc487 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":310:29) +#loc488 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":311:17) +#loc489 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":311:33) +#loc490 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":311:44) +#loc491 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":311:29) +#loc492 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":312:18) +#loc493 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":312:36) +#loc494 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":312:41) +#loc495 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":312:52) +#loc496 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":312:30) +#loc497 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":313:18) +#loc498 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":313:36) +#loc499 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":313:41) +#loc500 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":313:52) +#loc501 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":313:30) +#loc502 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":314:18) +#loc503 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":314:38) +#loc504 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":314:43) +#loc505 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":314:54) +#loc506 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":314:30) +#loc507 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":315:18) +#loc508 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":315:36) +#loc509 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":315:41) +#loc510 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":315:52) +#loc511 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":315:30) +#loc512 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":316:18) +#loc513 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":316:38) +#loc514 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":316:43) +#loc515 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":316:54) +#loc516 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":316:30) +#loc517 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":317:18) +#loc518 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":317:38) +#loc519 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":317:43) +#loc520 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":317:54) +#loc521 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":317:30) +#loc522 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":319:30) +#loc523 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":319:20) +#loc524 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":320:30) +#loc525 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":320:20) +#loc526 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":321:30) +#loc527 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":321:20) +#loc528 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":322:30) +#loc529 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":322:20) +#loc530 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":323:30) +#loc531 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":323:20) +#loc532 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":324:30) +#loc533 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":324:20) +#loc534 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":325:30) +#loc535 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":325:20) +#loc536 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":326:30) +#loc537 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":326:20) +#loc538 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":327:30) +#loc539 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":327:20) +#loc540 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":328:30) +#loc541 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":328:20) +#loc542 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":328:4) +#loc543 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":120:0) +#loc544 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":129:31) +#loc545 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":129:11) +#loc546 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":129:4) +#loc547 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":34:0) +#loc548 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":43:11) +#loc549 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":43:4) +#loc551 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:33) +#loc552 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:20) +#loc553 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:13) +#loc555 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc557 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc558 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc560 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc561 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc562 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc568 = loc("i_tc0"(#loc13)) +#loc569 = loc("i_tc1"(#loc14)) +#loc570 = loc("i_tc1"(#loc15)) +#loc571 = loc("i_tc2"(#loc16)) +#loc572 = loc("i_tc2"(#loc17)) +#loc573 = loc("i_tc3"(#loc18)) +#loc574 = loc("i_tc3"(#loc19)) +#loc575 = loc("k"(#loc20)) +#loc576 = loc("k"(#loc21)) +#loc577 = loc("k"(#loc22)) +#loc578 = loc("k"(#loc23)) +#loc579 = loc("k"(#loc24)) +#loc580 = loc("A"(#loc25)) +#loc581 = loc("A"(#loc26)) +#loc582 = loc("A"(#loc27)) +#loc583 = loc("A"(#loc28)) +#loc584 = loc("o_i"(#loc29)) +#loc585 = loc("m_tc0"(#loc30)) +#loc586 = loc("m_tc0"(#loc31)) +#loc587 = loc("m_tc1"(#loc32)) +#loc588 = loc("m_tc1"(#loc33)) +#loc589 = loc("m_tc2"(#loc34)) +#loc590 = loc("m_tc2"(#loc35)) +#loc591 = loc("m_tc3"(#loc36)) +#loc592 = loc("m_tc3"(#loc37)) +#loc593 = loc("p_b0"(#loc38)) +#loc594 = loc("p_b0"(#loc39)) +#loc595 = loc("p_b0"(#loc40)) +#loc596 = loc("p_b0"(#loc41)) +#loc597 = loc("p_b0"(#loc42)) +#loc598 = loc("p_b0"(#loc43)) +#loc599 = loc("p_b1"(#loc44)) +#loc600 = loc("p_b1"(#loc45)) +#loc601 = loc("p_b1"(#loc46)) +#loc602 = loc("p_b1"(#loc47)) +#loc603 = loc("p_b1"(#loc48)) +#loc604 = loc("p_b1"(#loc49)) +#loc605 = loc("p_b2"(#loc50)) +#loc606 = loc("p_b2"(#loc51)) +#loc607 = loc("p_b2"(#loc52)) +#loc608 = loc("p_b2"(#loc53)) +#loc609 = loc("p_b2"(#loc54)) +#loc610 = loc("p_b2"(#loc55)) +#loc611 = loc("p_b3"(#loc56)) +#loc612 = loc("p_b3"(#loc57)) +#loc613 = loc("p_b3"(#loc58)) +#loc614 = loc("p_b3"(#loc59)) +#loc615 = loc("p_b3"(#loc60)) +#loc616 = loc("p_b3"(#loc61)) +#loc617 = loc("b_b0"(#loc62)) +#loc618 = loc("b_b0"(#loc63)) +#loc619 = loc("b_b1"(#loc64)) +#loc620 = loc("b_b1"(#loc65)) +#loc621 = loc("b_b2"(#loc66)) +#loc622 = loc("b_b2"(#loc67)) +#loc623 = loc("b_b3"(#loc68)) +#loc624 = loc("b_b3"(#loc69)) +#loc625 = loc("p_g0"(#loc70)) +#loc626 = loc("p_g0"(#loc71)) +#loc627 = loc("p_g0"(#loc72)) +#loc628 = loc("p_g0"(#loc73)) +#loc629 = loc("p_g0"(#loc74)) +#loc630 = loc("p_g0"(#loc75)) +#loc631 = loc("p_g1"(#loc76)) +#loc632 = loc("p_g1"(#loc77)) +#loc633 = loc("p_g1"(#loc78)) +#loc634 = loc("p_g1"(#loc79)) +#loc635 = loc("p_g1"(#loc80)) +#loc636 = loc("p_g1"(#loc81)) +#loc637 = loc("p_g2"(#loc82)) +#loc638 = loc("p_g2"(#loc83)) +#loc639 = loc("p_g2"(#loc84)) +#loc640 = loc("p_g2"(#loc85)) +#loc641 = loc("p_g2"(#loc86)) +#loc642 = loc("p_g2"(#loc87)) +#loc643 = loc("p_g3"(#loc88)) +#loc644 = loc("p_g3"(#loc89)) +#loc645 = loc("p_g3"(#loc90)) +#loc646 = loc("p_g3"(#loc91)) +#loc647 = loc("p_g3"(#loc92)) +#loc648 = loc("p_g3"(#loc93)) +#loc649 = loc("b_g0"(#loc94)) +#loc650 = loc("b_g1"(#loc95)) +#loc651 = loc("b_g2"(#loc96)) +#loc652 = loc("b_g3"(#loc97)) +#loc653 = loc("b_A00"(#loc98)) +#loc654 = loc("b_A11"(#loc99)) +#loc655 = loc("b_A22"(#loc100)) +#loc656 = loc("b_A33"(#loc101)) +#loc657 = loc("b_A10"(#loc102)) +#loc658 = loc("b_A20"(#loc103)) +#loc659 = loc("b_A21"(#loc104)) +#loc660 = loc("b_A30"(#loc105)) +#loc661 = loc("b_A31"(#loc106)) +#loc662 = loc("b_A32"(#loc107)) +#loc663 = loc("b_A00"(#loc109)) +#loc664 = loc("o_k"(#loc110)) +#loc665 = loc("o_k"(#loc111)) +#loc666 = loc("o_k"(#loc112)) +#loc667 = loc("p_k0"(#loc113)) +#loc668 = loc("p_k0"(#loc114)) +#loc669 = loc("p_k0"(#loc115)) +#loc670 = loc("p_k0"(#loc116)) +#loc671 = loc("p_k0"(#loc117)) +#loc672 = loc("p_k0"(#loc118)) +#loc673 = loc("b_k0"(#loc119)) +#loc674 = loc("b_k0"(#loc120)) +#loc675 = loc("b_k0"(#loc121)) +#loc676 = loc("b_k0"(#loc122)) +#loc677 = loc("b_k0"(#loc123)) +#loc678 = loc("b_A00"(#loc124)) +#loc679 = loc("b_A00"(#loc125)) +#loc680 = loc("p_k1"(#loc128)) +#loc681 = loc("p_k1"(#loc129)) +#loc682 = loc("p_k1"(#loc130)) +#loc683 = loc("p_k1"(#loc131)) +#loc684 = loc("p_k1"(#loc132)) +#loc685 = loc("p_k1"(#loc133)) +#loc686 = loc("b_k1"(#loc134)) +#loc687 = loc("b_k1"(#loc135)) +#loc688 = loc("b_k1"(#loc136)) +#loc689 = loc("b_k1"(#loc137)) +#loc690 = loc("b_k1"(#loc138)) +#loc691 = loc("b_A11"(#loc139)) +#loc692 = loc("b_A11"(#loc140)) +#loc693 = loc("b_A10"(#loc141)) +#loc694 = loc("b_A10"(#loc142)) +#loc695 = loc("b_A22"(#loc144)) +#loc696 = loc("p_k2"(#loc145)) +#loc697 = loc("p_k2"(#loc146)) +#loc698 = loc("p_k2"(#loc147)) +#loc699 = loc("p_k2"(#loc148)) +#loc700 = loc("p_k2"(#loc149)) +#loc701 = loc("p_k2"(#loc150)) +#loc702 = loc("b_k2"(#loc151)) +#loc703 = loc("b_k2"(#loc152)) +#loc704 = loc("b_k2"(#loc153)) +#loc705 = loc("b_k2"(#loc154)) +#loc706 = loc("b_k2"(#loc155)) +#loc707 = loc("b_A22"(#loc156)) +#loc708 = loc("b_A22"(#loc157)) +#loc709 = loc("b_A20"(#loc158)) +#loc710 = loc("b_A20"(#loc159)) +#loc711 = loc("b_A21"(#loc160)) +#loc712 = loc("b_A21"(#loc161)) +#loc713 = loc("b_A33"(#loc163)) +#loc714 = loc("p_k3"(#loc164)) +#loc715 = loc("p_k3"(#loc165)) +#loc716 = loc("p_k3"(#loc166)) +#loc717 = loc("p_k3"(#loc167)) +#loc718 = loc("p_k3"(#loc168)) +#loc719 = loc("p_k3"(#loc169)) +#loc720 = loc("b_k3"(#loc170)) +#loc721 = loc("b_k3"(#loc171)) +#loc722 = loc("b_k3"(#loc172)) +#loc723 = loc("b_k3"(#loc173)) +#loc724 = loc("b_k3"(#loc174)) +#loc725 = loc("b_A33"(#loc175)) +#loc726 = loc("b_A33"(#loc176)) +#loc727 = loc("b_A30"(#loc177)) +#loc728 = loc("b_A30"(#loc178)) +#loc729 = loc("b_A31"(#loc179)) +#loc730 = loc("b_A31"(#loc180)) +#loc731 = loc("b_A32"(#loc181)) +#loc732 = loc("b_A32"(#loc182)) +#loc733 = loc("m_d"(#loc184)) +#loc734 = loc("m_d"(#loc185)) +#loc735 = loc("m_d"(#loc186)) +#loc736 = loc("m_I"(#loc187)) +#loc737 = loc("m_I"(#loc188)) +#loc738 = loc("m_I"(#loc189)) +#loc739 = loc("b_A00"(#loc190)) +#loc740 = loc("b_A00"(#loc191)) +#loc741 = loc("b_A00"(#loc192)) +#loc742 = loc("b_A00"(#loc193)) +#loc743 = loc("b_A00"(#loc194)) +#loc744 = loc("b_A00"(#loc195)) +#loc745 = loc("b_A00"(#loc196)) +#loc746 = loc("b_A00"(#loc197)) +#loc747 = loc("b_A00"(#loc198)) +#loc748 = loc("b_A00"(#loc199)) +#loc749 = loc("b_A11"(#loc200)) +#loc750 = loc("b_A11"(#loc201)) +#loc751 = loc("b_A11"(#loc202)) +#loc752 = loc("b_A11"(#loc203)) +#loc753 = loc("b_A11"(#loc204)) +#loc754 = loc("b_A11"(#loc205)) +#loc755 = loc("b_A11"(#loc206)) +#loc756 = loc("b_A11"(#loc207)) +#loc757 = loc("b_A11"(#loc208)) +#loc758 = loc("b_A11"(#loc209)) +#loc759 = loc("b_A22"(#loc210)) +#loc760 = loc("b_A22"(#loc211)) +#loc761 = loc("b_A22"(#loc212)) +#loc762 = loc("b_A22"(#loc213)) +#loc763 = loc("b_A22"(#loc214)) +#loc764 = loc("b_A22"(#loc215)) +#loc765 = loc("b_A22"(#loc216)) +#loc766 = loc("b_A22"(#loc217)) +#loc767 = loc("b_A22"(#loc218)) +#loc768 = loc("b_A22"(#loc219)) +#loc769 = loc("b_A33"(#loc220)) +#loc770 = loc("b_A33"(#loc221)) +#loc771 = loc("b_A33"(#loc222)) +#loc772 = loc("b_A33"(#loc223)) +#loc773 = loc("b_A33"(#loc224)) +#loc774 = loc("b_A33"(#loc225)) +#loc775 = loc("b_A33"(#loc226)) +#loc776 = loc("b_A33"(#loc227)) +#loc777 = loc("b_A33"(#loc228)) +#loc778 = loc("b_A33"(#loc229)) +#loc779 = loc("b_A10"(#loc230)) +#loc780 = loc("b_A10"(#loc231)) +#loc781 = loc("b_A10"(#loc232)) +#loc782 = loc("b_A10"(#loc233)) +#loc783 = loc("b_A10"(#loc234)) +#loc784 = loc("b_A10"(#loc235)) +#loc785 = loc("b_A10"(#loc236)) +#loc786 = loc("b_A10"(#loc237)) +#loc787 = loc("b_A10"(#loc238)) +#loc788 = loc("b_A20"(#loc239)) +#loc789 = loc("b_A20"(#loc240)) +#loc790 = loc("b_A20"(#loc241)) +#loc791 = loc("b_A20"(#loc242)) +#loc792 = loc("b_A20"(#loc243)) +#loc793 = loc("b_A20"(#loc244)) +#loc794 = loc("b_A20"(#loc245)) +#loc795 = loc("b_A20"(#loc246)) +#loc796 = loc("b_A20"(#loc247)) +#loc797 = loc("b_A21"(#loc248)) +#loc798 = loc("b_A21"(#loc249)) +#loc799 = loc("b_A21"(#loc250)) +#loc800 = loc("b_A21"(#loc251)) +#loc801 = loc("b_A21"(#loc252)) +#loc802 = loc("b_A21"(#loc253)) +#loc803 = loc("b_A21"(#loc254)) +#loc804 = loc("b_A21"(#loc255)) +#loc805 = loc("b_A21"(#loc256)) +#loc806 = loc("b_A30"(#loc257)) +#loc807 = loc("b_A30"(#loc258)) +#loc808 = loc("b_A30"(#loc259)) +#loc809 = loc("b_A30"(#loc260)) +#loc810 = loc("b_A30"(#loc261)) +#loc811 = loc("b_A30"(#loc262)) +#loc812 = loc("b_A30"(#loc263)) +#loc813 = loc("b_A30"(#loc264)) +#loc814 = loc("b_A30"(#loc265)) +#loc815 = loc("b_A31"(#loc266)) +#loc816 = loc("b_A31"(#loc267)) +#loc817 = loc("b_A31"(#loc268)) +#loc818 = loc("b_A31"(#loc269)) +#loc819 = loc("b_A31"(#loc270)) +#loc820 = loc("b_A31"(#loc271)) +#loc821 = loc("b_A31"(#loc272)) +#loc822 = loc("b_A31"(#loc273)) +#loc823 = loc("b_A31"(#loc274)) +#loc824 = loc("b_A32"(#loc275)) +#loc825 = loc("b_A32"(#loc276)) +#loc826 = loc("b_A32"(#loc277)) +#loc827 = loc("b_A32"(#loc278)) +#loc828 = loc("b_A32"(#loc279)) +#loc829 = loc("b_A32"(#loc280)) +#loc830 = loc("b_A32"(#loc281)) +#loc831 = loc("b_A32"(#loc282)) +#loc832 = loc("b_A32"(#loc283)) +#loc833 = loc("b_A00"(#loc284)) +#loc834 = loc("b_A00"(#loc285)) +#loc835 = loc("b_A11"(#loc286)) +#loc836 = loc("b_A11"(#loc287)) +#loc837 = loc("b_A22"(#loc288)) +#loc838 = loc("b_A22"(#loc289)) +#loc839 = loc("b_A33"(#loc290)) +#loc840 = loc("b_A33"(#loc291)) +#loc841 = loc("b_A10"(#loc292)) +#loc842 = loc("b_A10"(#loc293)) +#loc843 = loc("b_A20"(#loc294)) +#loc844 = loc("b_A20"(#loc295)) +#loc845 = loc("b_A21"(#loc296)) +#loc846 = loc("b_A21"(#loc297)) +#loc847 = loc("b_A30"(#loc298)) +#loc848 = loc("b_A30"(#loc299)) +#loc849 = loc("b_A31"(#loc300)) +#loc850 = loc("b_A31"(#loc301)) +#loc851 = loc("b_A32"(#loc302)) +#loc852 = loc("b_A32"(#loc303)) +#loc853 = loc("b_Ai00"(#loc304)) +#loc854 = loc("b_Ai11"(#loc305)) +#loc855 = loc("b_Ai22"(#loc306)) +#loc856 = loc("b_Ai33"(#loc307)) +#loc857 = loc("b_Ai00"(#loc310)) +#loc858 = loc("b_a00"(#loc311)) +#loc859 = loc("b_a00"(#loc312)) +#loc860 = loc("b_a00"(#loc313)) +#loc861 = loc("b_a00"(#loc314)) +#loc862 = loc("b_a00"(#loc315)) +#loc863 = loc("b_a00"(#loc316)) +#loc864 = loc("b_a00"(#loc317)) +#loc865 = loc("b_a00"(#loc318)) +#loc866 = loc("b_a00"(#loc319)) +#loc867 = loc("b_a00"(#loc320)) +#loc868 = loc("b_a00"(#loc321)) +#loc869 = loc("b_Ai00"(#loc322)) +#loc870 = loc("b_Ai00"(#loc323)) +#loc871 = loc("b_Ai00"(#loc324)) +#loc872 = loc("b_Ai11"(#loc328)) +#loc873 = loc("b_a11"(#loc329)) +#loc874 = loc("b_a11"(#loc330)) +#loc875 = loc("b_a11"(#loc331)) +#loc876 = loc("b_a11"(#loc332)) +#loc877 = loc("b_a11"(#loc333)) +#loc878 = loc("b_a11"(#loc334)) +#loc879 = loc("b_a11"(#loc335)) +#loc880 = loc("b_a11"(#loc336)) +#loc881 = loc("b_a11"(#loc337)) +#loc882 = loc("b_a11"(#loc338)) +#loc883 = loc("b_a11"(#loc339)) +#loc884 = loc("b_Ai11"(#loc340)) +#loc885 = loc("b_Ai11"(#loc341)) +#loc886 = loc("b_Ai11"(#loc342)) +#loc887 = loc("b_Ai22"(#loc346)) +#loc888 = loc("b_a22"(#loc347)) +#loc889 = loc("b_a22"(#loc348)) +#loc890 = loc("b_a22"(#loc349)) +#loc891 = loc("b_a22"(#loc350)) +#loc892 = loc("b_a22"(#loc351)) +#loc893 = loc("b_a22"(#loc352)) +#loc894 = loc("b_a22"(#loc353)) +#loc895 = loc("b_a22"(#loc354)) +#loc896 = loc("b_a22"(#loc355)) +#loc897 = loc("b_a22"(#loc356)) +#loc898 = loc("b_a22"(#loc357)) +#loc899 = loc("b_Ai22"(#loc358)) +#loc900 = loc("b_Ai22"(#loc359)) +#loc901 = loc("b_Ai22"(#loc360)) +#loc902 = loc("b_Ai33"(#loc364)) +#loc903 = loc("b_a33"(#loc365)) +#loc904 = loc("b_a33"(#loc366)) +#loc905 = loc("b_a33"(#loc367)) +#loc906 = loc("b_a33"(#loc368)) +#loc907 = loc("b_a33"(#loc369)) +#loc908 = loc("b_a33"(#loc370)) +#loc909 = loc("b_a33"(#loc371)) +#loc910 = loc("b_a33"(#loc372)) +#loc911 = loc("b_a33"(#loc373)) +#loc912 = loc("b_a33"(#loc374)) +#loc913 = loc("b_a33"(#loc375)) +#loc914 = loc("b_Ai33"(#loc376)) +#loc915 = loc("b_Ai33"(#loc377)) +#loc916 = loc("b_Ai33"(#loc378)) +#loc917 = loc("b_Ai00"(#loc380)) +#loc918 = loc("b_Ai11"(#loc381)) +#loc919 = loc("b_Ai22"(#loc382)) +#loc920 = loc("b_Ai33"(#loc383)) +#loc921 = loc("b_Ai10"(#loc384)) +#loc922 = loc("b_Ai10"(#loc385)) +#loc923 = loc("b_Ai10"(#loc386)) +#loc924 = loc("b_Ai21"(#loc387)) +#loc925 = loc("b_Ai21"(#loc388)) +#loc926 = loc("b_Ai21"(#loc389)) +#loc927 = loc("b_Ai32"(#loc390)) +#loc928 = loc("b_Ai32"(#loc391)) +#loc929 = loc("b_Ai32"(#loc392)) +#loc930 = loc("b_Ai20"(#loc393)) +#loc931 = loc("b_Ai20"(#loc394)) +#loc932 = loc("b_Ai20"(#loc395)) +#loc933 = loc("b_Ai20"(#loc396)) +#loc934 = loc("b_Ai20"(#loc397)) +#loc935 = loc("b_Ai31"(#loc398)) +#loc936 = loc("b_Ai31"(#loc399)) +#loc937 = loc("b_Ai31"(#loc400)) +#loc938 = loc("b_Ai31"(#loc401)) +#loc939 = loc("b_Ai31"(#loc402)) +#loc940 = loc("b_Ai30"(#loc403)) +#loc941 = loc("b_Ai30"(#loc404)) +#loc942 = loc("b_Ai30"(#loc405)) +#loc943 = loc("b_Ai30"(#loc406)) +#loc944 = loc("b_Ai30"(#loc407)) +#loc945 = loc("b_Ai30"(#loc408)) +#loc946 = loc("b_Ai30"(#loc409)) +#loc947 = loc("p_A00"(#loc410)) +#loc948 = loc("p_A00"(#loc411)) +#loc949 = loc("p_A00"(#loc412)) +#loc950 = loc("p_A00"(#loc413)) +#loc951 = loc("p_A00"(#loc414)) +#loc952 = loc("p_A00"(#loc415)) +#loc953 = loc("p_A10"(#loc416)) +#loc954 = loc("p_A10"(#loc417)) +#loc955 = loc("p_A10"(#loc418)) +#loc956 = loc("p_A10"(#loc419)) +#loc957 = loc("p_A10"(#loc420)) +#loc958 = loc("p_A10"(#loc421)) +#loc959 = loc("p_A11"(#loc422)) +#loc960 = loc("p_A11"(#loc423)) +#loc961 = loc("p_A11"(#loc424)) +#loc962 = loc("p_A11"(#loc425)) +#loc963 = loc("p_A11"(#loc426)) +#loc964 = loc("p_A11"(#loc427)) +#loc965 = loc("p_A11"(#loc428)) +#loc966 = loc("p_A20"(#loc429)) +#loc967 = loc("p_A20"(#loc430)) +#loc968 = loc("p_A20"(#loc431)) +#loc969 = loc("p_A20"(#loc432)) +#loc970 = loc("p_A20"(#loc433)) +#loc971 = loc("p_A20"(#loc434)) +#loc972 = loc("p_A21"(#loc435)) +#loc973 = loc("p_A21"(#loc436)) +#loc974 = loc("p_A21"(#loc437)) +#loc975 = loc("p_A21"(#loc438)) +#loc976 = loc("p_A21"(#loc439)) +#loc977 = loc("p_A21"(#loc440)) +#loc978 = loc("p_A21"(#loc441)) +#loc979 = loc("p_A22"(#loc442)) +#loc980 = loc("p_A22"(#loc443)) +#loc981 = loc("p_A22"(#loc444)) +#loc982 = loc("p_A22"(#loc445)) +#loc983 = loc("p_A22"(#loc446)) +#loc984 = loc("p_A22"(#loc447)) +#loc985 = loc("p_A22"(#loc448)) +#loc986 = loc("p_A30"(#loc449)) +#loc987 = loc("p_A30"(#loc450)) +#loc988 = loc("p_A30"(#loc451)) +#loc989 = loc("p_A30"(#loc452)) +#loc990 = loc("p_A30"(#loc453)) +#loc991 = loc("p_A30"(#loc454)) +#loc992 = loc("p_A31"(#loc455)) +#loc993 = loc("p_A31"(#loc456)) +#loc994 = loc("p_A31"(#loc457)) +#loc995 = loc("p_A31"(#loc458)) +#loc996 = loc("p_A31"(#loc459)) +#loc997 = loc("p_A31"(#loc460)) +#loc998 = loc("p_A31"(#loc461)) +#loc999 = loc("p_A32"(#loc462)) +#loc1000 = loc("p_A32"(#loc463)) +#loc1001 = loc("p_A32"(#loc464)) +#loc1002 = loc("p_A32"(#loc465)) +#loc1003 = loc("p_A32"(#loc466)) +#loc1004 = loc("p_A32"(#loc467)) +#loc1005 = loc("p_A32"(#loc468)) +#loc1006 = loc("p_A33"(#loc469)) +#loc1007 = loc("p_A33"(#loc470)) +#loc1008 = loc("p_A33"(#loc471)) +#loc1009 = loc("p_A33"(#loc472)) +#loc1010 = loc("p_A33"(#loc473)) +#loc1011 = loc("p_A33"(#loc474)) +#loc1012 = loc("p_A33"(#loc475)) +#loc1013 = loc("m_A0"(#loc476)) +#loc1014 = loc("m_A0"(#loc477)) +#loc1015 = loc("m_A0"(#loc478)) +#loc1016 = loc("m_A0"(#loc479)) +#loc1017 = loc("m_A1"(#loc480)) +#loc1018 = loc("m_A1"(#loc481)) +#loc1019 = loc("m_A1"(#loc482)) +#loc1020 = loc("m_A1"(#loc483)) +#loc1021 = loc("m_A2"(#loc484)) +#loc1022 = loc("m_A2"(#loc485)) +#loc1023 = loc("m_A2"(#loc486)) +#loc1024 = loc("m_A2"(#loc487)) +#loc1025 = loc("m_A3"(#loc488)) +#loc1026 = loc("m_A3"(#loc489)) +#loc1027 = loc("m_A3"(#loc490)) +#loc1028 = loc("m_A3"(#loc491)) +#loc1029 = loc("m_A11"(#loc492)) +#loc1030 = loc("m_A11"(#loc493)) +#loc1031 = loc("m_A11"(#loc494)) +#loc1032 = loc("m_A11"(#loc495)) +#loc1033 = loc("m_A11"(#loc496)) +#loc1034 = loc("m_A21"(#loc497)) +#loc1035 = loc("m_A21"(#loc498)) +#loc1036 = loc("m_A21"(#loc499)) +#loc1037 = loc("m_A21"(#loc500)) +#loc1038 = loc("m_A21"(#loc501)) +#loc1039 = loc("m_A22"(#loc502)) +#loc1040 = loc("m_A22"(#loc503)) +#loc1041 = loc("m_A22"(#loc504)) +#loc1042 = loc("m_A22"(#loc505)) +#loc1043 = loc("m_A22"(#loc506)) +#loc1044 = loc("m_A31"(#loc507)) +#loc1045 = loc("m_A31"(#loc508)) +#loc1046 = loc("m_A31"(#loc509)) +#loc1047 = loc("m_A31"(#loc510)) +#loc1048 = loc("m_A31"(#loc511)) +#loc1049 = loc("m_A32"(#loc512)) +#loc1050 = loc("m_A32"(#loc513)) +#loc1051 = loc("m_A32"(#loc514)) +#loc1052 = loc("m_A32"(#loc515)) +#loc1053 = loc("m_A32"(#loc516)) +#loc1054 = loc("m_A33"(#loc517)) +#loc1055 = loc("m_A33"(#loc518)) +#loc1056 = loc("m_A33"(#loc519)) +#loc1057 = loc("m_A33"(#loc520)) +#loc1058 = loc("m_A33"(#loc521)) +#loc1063 = loc("b_A11"(#loc654)) +#loc1064 = loc("b_A22"(#loc655)) +#loc1065 = loc("b_A33"(#loc656)) +#loc1066 = loc("b_A10"(#loc657)) +#loc1067 = loc("b_A20"(#loc658)) +#loc1068 = loc("b_A21"(#loc659)) +#loc1069 = loc("b_A30"(#loc660)) +#loc1070 = loc("b_A31"(#loc661)) +#loc1071 = loc("b_A32"(#loc662)) +#loc1072 = loc("b_A11"(#loc663)) +#loc1073 = loc("b_A11"(#loc692)) +#loc1074 = loc("b_A10"(#loc694)) +#loc1075 = loc("b_A33"(#loc695)) +#loc1076 = loc("b_A22"(#loc708)) +#loc1077 = loc("b_A20"(#loc710)) +#loc1078 = loc("b_A21"(#loc712)) +#loc1079 = loc("b_A30"(#loc713)) +#loc1080 = loc("b_A33"(#loc726)) +#loc1081 = loc("b_A30"(#loc728)) +#loc1082 = loc("b_A31"(#loc730)) +#loc1083 = loc("b_A32"(#loc732)) +#loc1084 = loc("b_A22"(#loc1064)) +#loc1085 = loc("b_A33"(#loc1065)) +#loc1086 = loc("b_A20"(#loc1067)) +#loc1087 = loc("b_A21"(#loc1068)) +#loc1088 = loc("b_A30"(#loc1069)) +#loc1089 = loc("b_A31"(#loc1070)) +#loc1090 = loc("b_A32"(#loc1071)) +#loc1091 = loc("b_A22"(#loc1072)) +#loc1092 = loc("b_A20"(#loc1075)) +#loc1093 = loc("b_A31"(#loc1079)) +#loc1094 = loc("b_A33"(#loc1085)) +#loc1095 = loc("b_A30"(#loc1088)) +#loc1096 = loc("b_A31"(#loc1089)) +#loc1097 = loc("b_A32"(#loc1090)) +#loc1098 = loc("b_A33"(#loc1091)) +#loc1099 = loc("b_A21"(#loc1092)) +#loc1100 = loc("b_A32"(#loc1093)) +#loc1101 = loc("b_A10"(#loc1098)) +#loc1102 = loc("b_A30"(#loc1099)) +#loc1103 = loc("b_A20"(#loc1101)) +#loc1104 = loc("b_A31"(#loc1102)) +#loc1105 = loc("b_A21"(#loc1103)) +#loc1106 = loc("b_A32"(#loc1104)) +#loc1107 = loc("b_A30"(#loc1105)) +#loc1108 = loc("b_A31"(#loc1107)) +#loc1109 = loc("b_A32"(#loc1108)) diff --git a/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.ttgir b/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..892ca2cdf93c07565e48b202feb584304c906126 --- /dev/null +++ b/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.ttgir @@ -0,0 +1,1442 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [4, 8], warpsPerCTA = [4, 1], order = [1, 0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1, 2], threadsPerWarp = [4, 8], warpsPerCTA = [4, 1], order = [1, 0]}> +#linear = #ttg.linear<{register = [[0, 1], [0, 8], [0, 16], [0, 32], [8, 0]], lane = [[0, 2], [0, 4], [1, 0], [2, 0], [4, 0]], warp = [[0, 0], [0, 0]], block = []}> +#linear1 = #ttg.linear<{register = [[0, 1], [0, 8], [0, 16], [0, 32]], lane = [[0, 2], [0, 4], [1, 0], [2, 0], [4, 0]], warp = [[8, 0], [0, 0]], block = []}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":40:0) +#loc1 = loc(unknown) +#loc212 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":228:66) +#loc218 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":230:56) +#loc228 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":233:66) +#loc233 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":235:56) +#loc243 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":238:66) +#loc248 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":240:56) +#loc258 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":243:66) +#loc263 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":245:56) +#mma = #ttg.nvidia_mma<{versionMajor = 2, versionMinor = 0, warpsPerCTA = [2, 2], instrShape = [16, 8]}> +#mma1 = #ttg.nvidia_mma<{versionMajor = 2, versionMinor = 0, warpsPerCTA = [4, 1], instrShape = [16, 8]}> +#shared = #ttg.swizzled_shared<{vec = 8, perPhase = 1, maxPhase = 8, order = [1, 0]}> +#shared1 = #ttg.swizzled_shared<{vec = 8, perPhase = 2, maxPhase = 2, order = [1, 0]}> +#smem = #ttg.shared_memory +#loc356 = loc("k"(#loc)) +#loc357 = loc("g"(#loc)) +#loc358 = loc("beta"(#loc)) +#loc359 = loc("A"(#loc)) +#loc360 = loc("T"(#loc)) +#loc553 = loc("b_a00"(#loc212)) +#loc558 = loc("b_a00"(#loc218)) +#loc565 = loc("b_a11"(#loc228)) +#loc570 = loc("b_a11"(#loc233)) +#loc577 = loc("b_a22"(#loc243)) +#loc582 = loc("b_a22"(#loc248)) +#loc589 = loc("b_a33"(#loc258)) +#loc594 = loc("b_a33"(#loc263)) +#loc690 = loc(callsite(#loc1 at #loc553)) +#loc692 = loc(callsite(#loc1 at #loc558)) +#loc694 = loc(callsite(#loc1 at #loc565)) +#loc696 = loc(callsite(#loc1 at #loc570)) +#loc698 = loc(callsite(#loc1 at #loc577)) +#loc700 = loc(callsite(#loc1 at #loc582)) +#loc702 = loc(callsite(#loc1 at #loc589)) +#loc704 = loc(callsite(#loc1 at #loc594)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 4 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @chunk_gated_delta_rule_fwd_kkt_solve_kernel(%k: !tt.ptr {tt.divisibility = 16 : i32} loc("k"(#loc)), %g: !tt.ptr {tt.divisibility = 16 : i32} loc("g"(#loc)), %beta: !tt.ptr {tt.divisibility = 16 : i32} loc("beta"(#loc)), %A: !tt.ptr {tt.divisibility = 16 : i32} loc("A"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<0.000000e+00> : tensor<16x16xf32, #mma> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<16x16xf32, #mma1> loc(#loc1) + %cst_1 = arith.constant dense<0.000000e+00> : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc1) + %cst_2 = arith.constant dense<0.000000e+00> : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc1) + %c0_i32 = arith.constant 0 : i32 loc(#loc1) + %c2_i32 = arith.constant 2 : i32 loc(#loc1) + %c1_i32 = arith.constant 1 : i32 loc(#loc1) + %c2_i64 = arith.constant 2 : i64 loc(#loc1) + %c1_i64 = arith.constant 1 : i64 loc(#loc1) + %cst_3 = arith.constant dense<0.000000e+00> : tensor<16x64xbf16, #blocked> loc(#loc1) + %c32_i64 = arith.constant 32 : i64 loc(#loc1) + %c64_i32 = arith.constant 64 : i32 loc(#loc1) + %c64_i64 = arith.constant 64 : i64 loc(#loc1) + %c16_i64 = arith.constant 16 : i64 loc(#loc1) + %c48_i64 = arith.constant 48 : i64 loc(#loc1) + %c128_i64 = arith.constant 128 : i64 loc(#loc1) + %cst_4 = arith.constant dense<64> : tensor<1x16xi32, #blocked1> loc(#loc1) + %cst_5 = arith.constant dense<48> : tensor<16xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc1) + %cst_6 = arith.constant dense<32> : tensor<16xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc1) + %cst_7 = arith.constant dense<16> : tensor<16xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc1) + %cst_8 = arith.constant dense<2048> : tensor<16x1xi64, #blocked1> loc(#loc1) + %cst_9 = arith.constant dense<128> : tensor<1x64xi32, #blocked> loc(#loc1) + %cst_10 = arith.constant dense<4096> : tensor<16x1xi64, #blocked> loc(#loc1) + %cst_11 = arith.constant dense<0.000000e+00> : tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc1) + %cst_12 = arith.constant dense<0.000000e+00> : tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc1) + %cst_13 = arith.constant dense<0.000000e+00> : tensor<16xbf16, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc1) + %cst_14 = arith.constant dense<0.000000e+00> : tensor<16xbf16, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc1) + %cst_15 = arith.constant dense<32> : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc1) + %cst_16 = arith.constant dense<32> : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc1) + %cst_17 = arith.constant dense<32> : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc1) + %cst_18 = arith.constant dense<32> : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc1) + %c-1_i32 = arith.constant -1 : i32 loc(#loc1) + %cst_19 = arith.constant dense<64> : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc1) + %0 = tt.get_program_id x : i32 loc(#loc2) + %1 = arith.extsi %0 : i32 to i64 loc(#loc3) + %2 = tt.get_program_id y : i32 loc(#loc4) + %3 = arith.extsi %2 : i32 to i64 loc(#loc5) + %4 = arith.divsi %3, %c32_i64 : i64 loc(#loc6) + %5 = arith.remsi %3, %c32_i64 : i64 loc(#loc7) + %6 = arith.extsi %T : i32 to i64 loc(#loc8) + %7 = arith.muli %4, %6 : i64 loc(#loc8) + %8 = arith.muli %1, %c64_i64 : i64 loc(#loc9) + %9 = arith.cmpi sge, %8, %6 : i64 loc(#loc10) + cf.cond_br %9, ^bb1, ^bb2 loc(#loc10) + ^bb1: // pred: ^bb0 + tt.return loc(#loc11) + ^bb2: // pred: ^bb0 + %i_tc1 = arith.addi %8, %c16_i64 : i64 loc(#loc361) + %i_tc2 = arith.addi %8, %c32_i64 : i64 loc(#loc362) + %i_tc3 = arith.addi %8, %c48_i64 : i64 loc(#loc363) + %k_20 = arith.muli %7, %c32_i64 : i64 loc(#loc364) + %k_21 = arith.addi %k_20, %5 : i64 loc(#loc365) + %k_22 = arith.muli %k_21, %c128_i64 : i64 loc(#loc366) + %k_23 = tt.addptr %k, %k_22 : !tt.ptr, i64 loc(#loc367) + %A_24 = arith.muli %k_21, %c64_i64 : i64 loc(#loc368) + %A_25 = tt.addptr %A, %A_24 : !tt.ptr, i64 loc(#loc369) + %o_i = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc370) + %o_i_26 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc370) + %o_i_27 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc370) + %o_i_28 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc370) + %o_i_29 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc370) + %o_i_30 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc370) + %o_i_31 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc370) + %m_tc0 = arith.extsi %o_i : tensor<16xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc371) + %m_tc0_32 = arith.extsi %o_i_26 : tensor<16xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> to tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc371) + %m_tc0_33 = arith.extsi %o_i_27 : tensor<16xi32, #ttg.slice<{dim = 1, parent = #mma}>> to tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc371) + %m_tc0_34 = arith.extsi %o_i_28 : tensor<16xi32, #ttg.slice<{dim = 1, parent = #mma1}>> to tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc371) + %m_tc0_35 = arith.extsi %o_i_30 : tensor<16xi32, #ttg.slice<{dim = 0, parent = #mma}>> to tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc371) + %m_tc0_36 = arith.extsi %o_i_31 : tensor<16xi32, #ttg.slice<{dim = 0, parent = #mma1}>> to tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc371) + %m_tc0_37 = tt.splat %8 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc371) + %m_tc0_38 = tt.splat %8 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc371) + %m_tc0_39 = tt.splat %8 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc371) + %m_tc0_40 = tt.splat %8 : i64 -> tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc371) + %m_tc0_41 = tt.splat %8 : i64 -> tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc371) + %m_tc0_42 = arith.addi %m_tc0_37, %m_tc0 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc371) + %m_tc0_43 = arith.addi %m_tc0_38, %m_tc0_32 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc371) + %m_tc0_44 = arith.addi %m_tc0_39, %m_tc0_33 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc371) + %m_tc0_45 = arith.addi %m_tc0_40, %m_tc0_35 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc371) + %m_tc0_46 = arith.addi %m_tc0_41, %m_tc0_36 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc371) + %m_tc0_47 = tt.splat %6 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc372) + %m_tc0_48 = tt.splat %6 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc372) + %m_tc0_49 = tt.splat %6 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc372) + %m_tc0_50 = tt.splat %6 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc372) + %m_tc0_51 = tt.splat %6 : i64 -> tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc372) + %m_tc0_52 = tt.splat %6 : i64 -> tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc372) + %m_tc0_53 = arith.cmpi slt, %m_tc0_42, %m_tc0_47 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc372) + %m_tc0_54 = arith.cmpi slt, %m_tc0_43, %m_tc0_48 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc372) + %m_tc0_55 = arith.cmpi slt, %m_tc0_44, %m_tc0_49 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc372) + %m_tc0_56 = arith.cmpi slt, %m_tc0_45, %m_tc0_51 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc372) + %m_tc0_57 = arith.cmpi slt, %m_tc0_46, %m_tc0_52 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc372) + %m_tc1 = tt.splat %i_tc1 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc373) + %m_tc1_58 = tt.splat %i_tc1 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc373) + %m_tc1_59 = tt.splat %i_tc1 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc373) + %m_tc1_60 = tt.splat %i_tc1 : i64 -> tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc373) + %m_tc1_61 = arith.addi %m_tc1, %m_tc0 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc373) + %m_tc1_62 = arith.addi %m_tc1_58, %m_tc0_32 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc373) + %m_tc1_63 = arith.addi %m_tc1_59, %m_tc0_34 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc373) + %m_tc1_64 = arith.addi %m_tc1_60, %m_tc0_36 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc373) + %m_tc1_65 = arith.cmpi slt, %m_tc1_61, %m_tc0_47 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc374) + %m_tc1_66 = arith.cmpi slt, %m_tc1_62, %m_tc0_48 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc374) + %m_tc1_67 = arith.cmpi slt, %m_tc1_63, %m_tc0_50 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc374) + %m_tc1_68 = arith.cmpi slt, %m_tc1_64, %m_tc0_52 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc374) + %m_tc2 = tt.splat %i_tc2 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc375) + %m_tc2_69 = tt.splat %i_tc2 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc375) + %m_tc2_70 = tt.splat %i_tc2 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc375) + %m_tc2_71 = tt.splat %i_tc2 : i64 -> tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc375) + %m_tc2_72 = arith.addi %m_tc2, %m_tc0 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc375) + %m_tc2_73 = arith.addi %m_tc2_69, %m_tc0_32 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc375) + %m_tc2_74 = arith.addi %m_tc2_70, %m_tc0_34 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc375) + %m_tc2_75 = arith.addi %m_tc2_71, %m_tc0_36 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc375) + %m_tc2_76 = arith.cmpi slt, %m_tc2_72, %m_tc0_47 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc376) + %m_tc2_77 = arith.cmpi slt, %m_tc2_73, %m_tc0_48 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc376) + %m_tc2_78 = arith.cmpi slt, %m_tc2_74, %m_tc0_50 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc376) + %m_tc2_79 = arith.cmpi slt, %m_tc2_75, %m_tc0_52 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc376) + %m_tc3 = tt.splat %i_tc3 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc377) + %m_tc3_80 = tt.splat %i_tc3 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc377) + %m_tc3_81 = tt.splat %i_tc3 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc377) + %m_tc3_82 = tt.splat %i_tc3 : i64 -> tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc377) + %m_tc3_83 = arith.addi %m_tc3, %m_tc0 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc377) + %m_tc3_84 = arith.addi %m_tc3_80, %m_tc0_32 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc377) + %m_tc3_85 = arith.addi %m_tc3_81, %m_tc0_34 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc377) + %m_tc3_86 = arith.addi %m_tc3_82, %m_tc0_36 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc377) + %m_tc3_87 = arith.cmpi slt, %m_tc3_83, %m_tc0_47 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc378) + %m_tc3_88 = arith.cmpi slt, %m_tc3_84, %m_tc0_48 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc378) + %m_tc3_89 = arith.cmpi slt, %m_tc3_85, %m_tc0_50 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc378) + %m_tc3_90 = arith.cmpi slt, %m_tc3_86, %m_tc0_52 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc378) + %p_b0 = tt.addptr %beta, %k_21 : !tt.ptr, i64 loc(#loc665) + %p_b0_91 = arith.muli %m_tc0_44, %cst_17 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc381) + %p_b0_92 = arith.muli %m_tc0_45, %cst_15 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc381) + %p_b0_93 = arith.muli %m_tc0_46, %cst_16 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc381) + %p_b0_94 = tt.splat %p_b0 : !tt.ptr -> tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc382) + %p_b0_95 = tt.splat %p_b0 : !tt.ptr -> tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc382) + %p_b0_96 = tt.addptr %p_b0_94, %p_b0_91 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma}>>, tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc382) + %p_b1 = arith.muli %m_tc1_63, %cst_18 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc383) + %p_b1_97 = arith.muli %m_tc1_64, %cst_16 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc383) + %p_b1_98 = tt.addptr %p_b0_95, %p_b1 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma1}>>, tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc384) + %p_b2 = arith.muli %m_tc2_74, %cst_18 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc385) + %p_b2_99 = arith.muli %m_tc2_75, %cst_16 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc385) + %p_b2_100 = tt.addptr %p_b0_95, %p_b2 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma1}>>, tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc386) + %p_b3 = arith.muli %m_tc3_85, %cst_18 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc387) + %p_b3_101 = arith.muli %m_tc3_86, %cst_16 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc387) + %p_b3_102 = tt.addptr %p_b0_95, %p_b3 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma1}>>, tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc388) + %b_b0 = tt.load %p_b0_96, %m_tc0_55, %cst_13 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc389) + %b_b0_103 = arith.extf %b_b0 : tensor<16xbf16, #ttg.slice<{dim = 1, parent = #mma}>> to tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc390) + %b_b1 = tt.load %p_b1_98, %m_tc1_67, %cst_14 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc391) + %b_b1_104 = arith.extf %b_b1 : tensor<16xbf16, #ttg.slice<{dim = 1, parent = #mma1}>> to tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc392) + %b_b2 = tt.load %p_b2_100, %m_tc2_78, %cst_14 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc393) + %b_b2_105 = arith.extf %b_b2 : tensor<16xbf16, #ttg.slice<{dim = 1, parent = #mma1}>> to tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc394) + %b_b3 = tt.load %p_b3_102, %m_tc3_89, %cst_14 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc395) + %b_b3_106 = arith.extf %b_b3 : tensor<16xbf16, #ttg.slice<{dim = 1, parent = #mma1}>> to tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc396) + %p_g0 = tt.addptr %g, %k_21 : !tt.ptr, i64 loc(#loc666) + %p_g0_107 = tt.splat %p_g0 : !tt.ptr -> tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc399) + %p_g0_108 = tt.splat %p_g0 : !tt.ptr -> tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc399) + %p_g0_109 = tt.splat %p_g0 : !tt.ptr -> tensor<16x!tt.ptr, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc399) + %p_g0_110 = tt.splat %p_g0 : !tt.ptr -> tensor<16x!tt.ptr, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc399) + %p_g0_111 = tt.addptr %p_g0_107, %p_b0_91 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma}>>, tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc399) + %p_g0_112 = tt.addptr %p_g0_109, %p_b0_92 : tensor<16x!tt.ptr, #ttg.slice<{dim = 0, parent = #mma}>>, tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc399) + %p_g0_113 = tt.addptr %p_g0_110, %p_b0_93 : tensor<16x!tt.ptr, #ttg.slice<{dim = 0, parent = #mma1}>>, tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc399) + %p_g1 = tt.addptr %p_g0_108, %p_b1 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma1}>>, tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc400) + %p_g1_114 = tt.addptr %p_g0_110, %p_b1_97 : tensor<16x!tt.ptr, #ttg.slice<{dim = 0, parent = #mma1}>>, tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc400) + %p_g2 = tt.addptr %p_g0_108, %p_b2 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma1}>>, tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc401) + %p_g2_115 = tt.addptr %p_g0_110, %p_b2_99 : tensor<16x!tt.ptr, #ttg.slice<{dim = 0, parent = #mma1}>>, tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc401) + %p_g3 = tt.addptr %p_g0_108, %p_b3 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma1}>>, tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc402) + %p_g3_116 = tt.addptr %p_g0_110, %p_b3_101 : tensor<16x!tt.ptr, #ttg.slice<{dim = 0, parent = #mma1}>>, tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc402) + %b_g0 = tt.load %p_g0_111, %m_tc0_55, %cst_11 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc403) + %b_g0_117 = tt.load %p_g0_112, %m_tc0_56, %cst_1 : tensor<16x!tt.ptr, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc403) + %b_g0_118 = tt.load %p_g0_113, %m_tc0_57, %cst_2 : tensor<16x!tt.ptr, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc403) + %b_g1 = tt.load %p_g1, %m_tc1_67, %cst_12 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc404) + %b_g1_119 = tt.load %p_g1_114, %m_tc1_68, %cst_2 : tensor<16x!tt.ptr, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc404) + %b_g2 = tt.load %p_g2, %m_tc2_78, %cst_12 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc405) + %b_g2_120 = tt.load %p_g2_115, %m_tc2_79, %cst_2 : tensor<16x!tt.ptr, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc405) + %b_g3 = tt.load %p_g3, %m_tc3_89, %cst_12 : tensor<16x!tt.ptr, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc406) + %b_g3_121 = tt.load %p_g3_116, %m_tc3_90, %cst_2 : tensor<16x!tt.ptr, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc406) + %o_k = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc407) + %p_k0 = tt.expand_dims %m_tc0_42 {axis = 1 : i32} : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi64, #blocked> loc(#loc408) + %p_k0_122 = arith.muli %p_k0, %cst_10 : tensor<16x1xi64, #blocked> loc(#loc409) + %p_k0_123 = tt.splat %k_23 : !tt.ptr -> tensor<16x1x!tt.ptr, #blocked> loc(#loc410) + %p_k0_124 = tt.addptr %p_k0_123, %p_k0_122 : tensor<16x1x!tt.ptr, #blocked>, tensor<16x1xi64, #blocked> loc(#loc410) + %p_k0_125 = tt.broadcast %p_k0_124 : tensor<16x1x!tt.ptr, #blocked> -> tensor<16x64x!tt.ptr, #blocked> loc(#loc411) + %b_k0 = tt.expand_dims %m_tc0_53 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi1, #blocked> loc(#loc412) + %b_k0_126 = tt.broadcast %b_k0 : tensor<16x1xi1, #blocked> -> tensor<16x64xi1, #blocked> loc(#loc413) + %10 = arith.cmpi slt, %i_tc1, %6 : i64 loc(#loc65) + %b_k0_127 = ttg.local_alloc : () -> !ttg.memdesc<2x16x64xbf16, #shared, #smem, mutable> loc(#loc414) + %p_k0_128 = tt.expand_dims %o_k {axis = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x64xi32, #blocked> loc(#loc415) + %p_k0_129 = tt.broadcast %p_k0_128 : tensor<1x64xi32, #blocked> -> tensor<16x64xi32, #blocked> loc(#loc411) + %p_k0_130 = tt.addptr %p_k0_125, %p_k0_129 : tensor<16x64x!tt.ptr, #blocked>, tensor<16x64xi32, #blocked> loc(#loc411) + %b_k0_131 = arith.cmpi slt, %p_k0_128, %cst_9 : tensor<1x64xi32, #blocked> loc(#loc416) + %b_k0_132 = tt.broadcast %b_k0_131 : tensor<1x64xi1, #blocked> -> tensor<16x64xi1, #blocked> loc(#loc413) + %b_k0_133 = arith.andi %b_k0_126, %b_k0_132 : tensor<16x64xi1, #blocked> loc(#loc413) + %b_k0_134 = ttg.memdesc_index %b_k0_127[%c0_i32] : !ttg.memdesc<2x16x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<16x64xbf16, #shared, #smem, mutable> loc(#loc414) + %b_k0_135 = ttg.async_copy_global_to_local %p_k0_130, %b_k0_134 mask %b_k0_133 other %cst_3 {contiguity = 8 : i32} : tensor<16x64x!tt.ptr, #blocked> -> <16x64xbf16, #shared, #smem, mutable> loc(#loc414) + %b_k0_136 = ttg.async_commit_group tokens %b_k0_135 loc(#loc414) + %o_k_137 = arith.addi %o_k, %cst_19 : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc417) + %p_k0_138 = tt.expand_dims %o_k_137 {axis = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x64xi32, #blocked> loc(#loc415) + %p_k0_139 = tt.broadcast %p_k0_138 : tensor<1x64xi32, #blocked> -> tensor<16x64xi32, #blocked> loc(#loc411) + %p_k0_140 = tt.addptr %p_k0_125, %p_k0_139 : tensor<16x64x!tt.ptr, #blocked>, tensor<16x64xi32, #blocked> loc(#loc411) + %b_k0_141 = arith.cmpi slt, %p_k0_138, %cst_9 : tensor<1x64xi32, #blocked> loc(#loc416) + %b_k0_142 = tt.broadcast %b_k0_141 : tensor<1x64xi1, #blocked> -> tensor<16x64xi1, #blocked> loc(#loc413) + %b_k0_143 = arith.andi %b_k0_126, %b_k0_142 : tensor<16x64xi1, #blocked> loc(#loc413) + %b_k0_144 = ttg.memdesc_index %b_k0_127[%c1_i32] : !ttg.memdesc<2x16x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<16x64xbf16, #shared, #smem, mutable> loc(#loc414) + %b_k0_145 = ttg.async_copy_global_to_local %p_k0_140, %b_k0_144 mask %b_k0_143 other %cst_3 {contiguity = 8 : i32} : tensor<16x64x!tt.ptr, #blocked> -> <16x64xbf16, #shared, #smem, mutable> loc(#loc414) + %b_k0_146 = ttg.async_commit_group tokens %b_k0_145 loc(#loc414) + %b_A32:18 = scf.for %b_A32_355 = %c0_i32 to %c2_i32 step %c1_i32 iter_args(%arg6 = %cst, %arg7 = %cst_0, %arg8 = %cst_0, %arg9 = %cst_0, %arg10 = %cst_0, %arg11 = %cst_0, %arg12 = %cst_0, %arg13 = %cst_0, %arg14 = %cst_0, %arg15 = %cst_0, %arg16 = %c1_i32, %arg17 = %c-1_i32, %b_k0_356 = %b_k0_136, %b_k0_357 = %b_k0_146, %p_k0_358 = %p_k0_129, %p_k0_359 = %p_k0_139, %b_k0_360 = %b_k0_132, %b_k0_361 = %b_k0_142) -> (tensor<16x16xf32, #mma>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, i32, i32, !ttg.async.token, !ttg.async.token, tensor<16x64xi32, #blocked>, tensor<16x64xi32, #blocked>, tensor<16x64xi1, #blocked>, tensor<16x64xi1, #blocked>) : i32 { + %b_A32_362 = arith.cmpi slt, %b_A32_355, %c0_i32 : i32 loc(#loc727) + %b_A32_363 = arith.addi %arg17, %c1_i32 : i32 loc(#loc727) + %b_A32_364 = arith.cmpi sge, %b_A32_363, %c2_i32 : i32 loc(#loc727) + %b_A32_365 = arith.select %b_A32_364, %c0_i32, %b_A32_363 : i32 loc(#loc727) + %b_k0_366 = ttg.async_wait %b_k0_356 {num = 1 : i32} loc(#loc414) + %b_k0_367 = ttg.memdesc_index %b_k0_127[%b_A32_365] : !ttg.memdesc<2x16x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<16x64xbf16, #shared, #smem, mutable> loc(#loc414) + %b_A00_368 = ttg.local_load %b_k0_367 token %b_k0_366 : !ttg.memdesc<16x64xbf16, #shared, #smem, mutable> -> tensor<16x64xbf16, #linear> loc(#loc419) + %b_A00_369 = tt.trans %b_A00_368 {order = array} : tensor<16x64xbf16, #linear> -> tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> loc(#loc420) + %b_A00_370 = ttg.local_load %b_k0_367 token %b_k0_366 : !ttg.memdesc<16x64xbf16, #shared, #smem, mutable> -> tensor<16x64xbf16, #linear1> loc(#loc419) + %b_A00_371 = tt.trans %b_A00_370 {order = array} : tensor<16x64xbf16, #linear1> -> tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> loc(#loc420) + %b_k0_372 = ttg.local_load %b_k0_367 token %b_k0_366 : !ttg.memdesc<16x64xbf16, #shared, #smem, mutable> -> tensor<16x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma, kWidth = 2}>> loc(#loc414) + %b_A00_373 = tt.dot %b_k0_372, %b_A00_371, %arg6, inputPrecision = tf32 : tensor<16x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma, kWidth = 2}>> * tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> -> tensor<16x16xf32, #mma> loc(#loc419) + %b_A32_374 = arith.addi %arg16, %c1_i32 : i32 loc(#loc727) + %b_A32_375 = arith.cmpi sge, %b_A32_374, %c2_i32 : i32 loc(#loc727) + %b_A32_376 = arith.select %b_A32_375, %c0_i32, %b_A32_374 : i32 loc(#loc727) + %b_A32_377 = arith.addi %b_A32_355, %c2_i32 : i32 loc(#loc727) + %o_k_378 = arith.muli %b_A32_377, %c64_i32 : i32 loc(#loc421) + %o_k_379 = tt.splat %o_k_378 : i32 -> tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc417) + %o_k_380 = arith.addi %o_k_379, %o_k : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc417) + %p_k0_381 = tt.expand_dims %o_k_380 {axis = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x64xi32, #blocked> loc(#loc415) + %p_k0_382 = tt.broadcast %p_k0_381 : tensor<1x64xi32, #blocked> -> tensor<16x64xi32, #blocked> loc(#loc411) + %p_k0_383 = tt.addptr %p_k0_125, %p_k0_382 : tensor<16x64x!tt.ptr, #blocked>, tensor<16x64xi32, #blocked> loc(#loc411) + %b_k0_384 = arith.cmpi slt, %p_k0_381, %cst_9 : tensor<1x64xi32, #blocked> loc(#loc416) + %b_k0_385 = tt.broadcast %b_k0_384 : tensor<1x64xi1, #blocked> -> tensor<16x64xi1, #blocked> loc(#loc413) + %b_k0_386 = arith.andi %b_k0_126, %b_k0_385 : tensor<16x64xi1, #blocked> loc(#loc413) + %b_k0_387 = ttg.memdesc_index %b_k0_127[%b_A32_376] : !ttg.memdesc<2x16x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<16x64xbf16, #shared, #smem, mutable> loc(#loc414) + %b_A32_388 = tt.splat %b_A32_362 : i1 -> tensor<16x64xi1, #blocked> loc(#loc727) + %b_A32_389 = arith.andi %b_A32_388, %b_k0_386 : tensor<16x64xi1, #blocked> loc(#loc727) + %b_k0_390 = ttg.async_copy_global_to_local %p_k0_383, %b_k0_387 mask %b_A32_389 other %cst_3 {contiguity = 8 : i32} : tensor<16x64x!tt.ptr, #blocked> -> <16x64xbf16, #shared, #smem, mutable> loc(#loc414) + %b_k0_391 = ttg.async_commit_group tokens %b_k0_390 loc(#loc414) + %39:9 = scf.if %10 -> (tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>) { + %p_k1 = tt.expand_dims %m_tc1_61 {axis = 1 : i32} : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi64, #blocked> loc(#loc422) + %p_k1_392 = arith.muli %p_k1, %cst_10 : tensor<16x1xi64, #blocked> loc(#loc423) + %p_k1_393 = tt.addptr %p_k0_123, %p_k1_392 : tensor<16x1x!tt.ptr, #blocked>, tensor<16x1xi64, #blocked> loc(#loc424) + %p_k1_394 = tt.broadcast %p_k1_393 : tensor<16x1x!tt.ptr, #blocked> -> tensor<16x64x!tt.ptr, #blocked> loc(#loc425) + %p_k1_395 = tt.addptr %p_k1_394, %p_k0_358 : tensor<16x64x!tt.ptr, #blocked>, tensor<16x64xi32, #blocked> loc(#loc425) + %b_k1 = tt.expand_dims %m_tc1_65 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi1, #blocked> loc(#loc426) + %b_k1_396 = tt.broadcast %b_k1 : tensor<16x1xi1, #blocked> -> tensor<16x64xi1, #blocked> loc(#loc427) + %b_k1_397 = arith.andi %b_k1_396, %b_k0_360 : tensor<16x64xi1, #blocked> loc(#loc427) + %b_k1_398 = tt.load %p_k1_395, %b_k1_397, %cst_3 : tensor<16x64x!tt.ptr, #blocked> loc(#loc428) + %b_k1_399 = ttg.local_alloc %b_k1_398 : (tensor<16x64xbf16, #blocked>) -> !ttg.memdesc<16x64xbf16, #shared, #smem> loc(#loc428) + %b_A11_400 = ttg.convert_layout %b_k1_398 : tensor<16x64xbf16, #blocked> -> tensor<16x64xbf16, #linear> loc(#loc429) + %b_A11_401 = tt.trans %b_A11_400 {order = array} : tensor<16x64xbf16, #linear> -> tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> loc(#loc429) + %b_k1_402 = ttg.local_load %b_k1_399 : !ttg.memdesc<16x64xbf16, #shared, #smem> -> tensor<16x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> loc(#loc428) + %b_A11_403 = tt.dot %b_k1_402, %b_A11_401, %arg7, inputPrecision = tf32 : tensor<16x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc668) + %b_A10_404 = tt.dot %b_k1_402, %b_A00_369, %arg10, inputPrecision = tf32 : tensor<16x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc669) + %40 = arith.cmpi slt, %i_tc2, %6 : i64 loc(#loc85) + %b_A32_405:7 = scf.if %40 -> (tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>) { + %p_k2 = tt.expand_dims %m_tc2_72 {axis = 1 : i32} : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi64, #blocked> loc(#loc433) + %p_k2_406 = arith.muli %p_k2, %cst_10 : tensor<16x1xi64, #blocked> loc(#loc434) + %p_k2_407 = tt.addptr %p_k0_123, %p_k2_406 : tensor<16x1x!tt.ptr, #blocked>, tensor<16x1xi64, #blocked> loc(#loc435) + %p_k2_408 = tt.broadcast %p_k2_407 : tensor<16x1x!tt.ptr, #blocked> -> tensor<16x64x!tt.ptr, #blocked> loc(#loc436) + %p_k2_409 = tt.addptr %p_k2_408, %p_k0_358 : tensor<16x64x!tt.ptr, #blocked>, tensor<16x64xi32, #blocked> loc(#loc436) + %b_k2 = tt.expand_dims %m_tc2_76 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi1, #blocked> loc(#loc437) + %b_k2_410 = tt.broadcast %b_k2 : tensor<16x1xi1, #blocked> -> tensor<16x64xi1, #blocked> loc(#loc438) + %b_k2_411 = arith.andi %b_k2_410, %b_k0_360 : tensor<16x64xi1, #blocked> loc(#loc438) + %b_k2_412 = tt.load %p_k2_409, %b_k2_411, %cst_3 : tensor<16x64x!tt.ptr, #blocked> loc(#loc439) + %b_k2_413 = ttg.local_alloc %b_k2_412 : (tensor<16x64xbf16, #blocked>) -> !ttg.memdesc<16x64xbf16, #shared, #smem> loc(#loc439) + %b_A22_414 = ttg.convert_layout %b_k2_412 : tensor<16x64xbf16, #blocked> -> tensor<16x64xbf16, #linear> loc(#loc440) + %b_A22_415 = tt.trans %b_A22_414 {order = array} : tensor<16x64xbf16, #linear> -> tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> loc(#loc440) + %b_k2_416 = ttg.local_load %b_k2_413 : !ttg.memdesc<16x64xbf16, #shared, #smem> -> tensor<16x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> loc(#loc439) + %b_A22_417 = tt.dot %b_k2_416, %b_A22_415, %arg8, inputPrecision = tf32 : tensor<16x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc671) + %b_A20_418 = tt.dot %b_k2_416, %b_A00_369, %arg11, inputPrecision = tf32 : tensor<16x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc672) + %b_A21_419 = tt.dot %b_k2_416, %b_A11_401, %arg12, inputPrecision = tf32 : tensor<16x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc673) + %41 = arith.cmpi slt, %i_tc3, %6 : i64 loc(#loc98) + %b_A32_420:4 = scf.if %41 -> (tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>) { + %p_k3 = tt.expand_dims %m_tc3_83 {axis = 1 : i32} : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi64, #blocked> loc(#loc445) + %p_k3_421 = arith.muli %p_k3, %cst_10 : tensor<16x1xi64, #blocked> loc(#loc446) + %p_k3_422 = tt.addptr %p_k0_123, %p_k3_421 : tensor<16x1x!tt.ptr, #blocked>, tensor<16x1xi64, #blocked> loc(#loc447) + %p_k3_423 = tt.broadcast %p_k3_422 : tensor<16x1x!tt.ptr, #blocked> -> tensor<16x64x!tt.ptr, #blocked> loc(#loc448) + %p_k3_424 = tt.addptr %p_k3_423, %p_k0_358 : tensor<16x64x!tt.ptr, #blocked>, tensor<16x64xi32, #blocked> loc(#loc448) + %b_k3 = tt.expand_dims %m_tc3_87 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi1, #blocked> loc(#loc449) + %b_k3_425 = tt.broadcast %b_k3 : tensor<16x1xi1, #blocked> -> tensor<16x64xi1, #blocked> loc(#loc450) + %b_k3_426 = arith.andi %b_k3_425, %b_k0_360 : tensor<16x64xi1, #blocked> loc(#loc450) + %b_k3_427 = tt.load %p_k3_424, %b_k3_426, %cst_3 : tensor<16x64x!tt.ptr, #blocked> loc(#loc451) + %b_k3_428 = ttg.local_alloc %b_k3_427 : (tensor<16x64xbf16, #blocked>) -> !ttg.memdesc<16x64xbf16, #shared, #smem> loc(#loc451) + %b_A33_429 = ttg.convert_layout %b_k3_427 : tensor<16x64xbf16, #blocked> -> tensor<16x64xbf16, #linear> loc(#loc452) + %b_A33_430 = tt.trans %b_A33_429 {order = array} : tensor<16x64xbf16, #linear> -> tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> loc(#loc452) + %b_k3_431 = ttg.local_load %b_k3_428 : !ttg.memdesc<16x64xbf16, #shared, #smem> -> tensor<16x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> loc(#loc451) + %b_A33_432 = tt.dot %b_k3_431, %b_A33_430, %arg9, inputPrecision = tf32 : tensor<16x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc675) + %b_A30_433 = tt.dot %b_k3_431, %b_A00_369, %arg13, inputPrecision = tf32 : tensor<16x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc676) + %b_A31_434 = tt.dot %b_k3_431, %b_A11_401, %arg14, inputPrecision = tf32 : tensor<16x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc677) + %b_A32_435 = tt.dot %b_k3_431, %b_A22_415, %arg15, inputPrecision = tf32 : tensor<16x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<64x16xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc678) + scf.yield %b_A33_432, %b_A30_433, %b_A31_434, %b_A32_435 : tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1> loc(#loc678) + } else { + scf.yield %arg9, %arg13, %arg14, %arg15 : tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1> loc(#loc99) + } loc(#loc718) + scf.yield %b_A22_417, %b_A32_420#0, %b_A20_418, %b_A21_419, %b_A32_420#1, %b_A32_420#2, %b_A32_420#3 : tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1> loc(#loc718) + } else { + scf.yield %arg8, %arg9, %arg11, %arg12, %arg13, %arg14, %arg15 : tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1> loc(#loc86) + } loc(#loc724) + scf.yield %b_A11_403, %b_A32_405#0, %b_A32_405#1, %b_A10_404, %b_A32_405#2, %b_A32_405#3, %b_A32_405#4, %b_A32_405#5, %b_A32_405#6 : tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1> loc(#loc724) + } else { + scf.yield %arg7, %arg8, %arg9, %arg10, %arg11, %arg12, %arg13, %arg14, %arg15 : tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1> loc(#loc74) + } loc(#loc74) + scf.yield %b_A00_373, %39#0, %39#1, %39#2, %39#3, %39#4, %39#5, %39#6, %39#7, %39#8, %b_A32_376, %b_A32_365, %b_k0_357, %b_k0_391, %p_k0_359, %p_k0_382, %b_k0_361, %b_k0_385 : tensor<16x16xf32, #mma>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, tensor<16x16xf32, #mma1>, i32, i32, !ttg.async.token, !ttg.async.token, tensor<16x64xi32, #blocked>, tensor<16x64xi32, #blocked>, tensor<16x64xi1, #blocked>, tensor<16x64xi1, #blocked> loc(#loc727) + } loc(#loc727) + %b_A32_147 = ttg.async_wait {num = 0 : i32} loc(#loc727) + ttg.local_dealloc %b_k0_127 : !ttg.memdesc<2x16x64xbf16, #shared, #smem, mutable> loc(#loc727) + %m_d = tt.expand_dims %o_i_27 {axis = 1 : i32} : tensor<16xi32, #ttg.slice<{dim = 1, parent = #mma}>> -> tensor<16x1xi32, #mma> loc(#loc457) + %m_d_148 = tt.expand_dims %o_i_28 {axis = 1 : i32} : tensor<16xi32, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xi32, #mma1> loc(#loc457) + %m_d_149 = tt.expand_dims %o_i_29 {axis = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x16xi32, #blocked1> loc(#loc458) + %m_d_150 = tt.expand_dims %o_i_30 {axis = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 0, parent = #mma}>> -> tensor<1x16xi32, #mma> loc(#loc458) + %m_d_151 = tt.expand_dims %o_i_31 {axis = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<1x16xi32, #mma1> loc(#loc458) + %m_d_152 = tt.broadcast %m_d : tensor<16x1xi32, #mma> -> tensor<16x16xi32, #mma> loc(#loc459) + %m_d_153 = tt.broadcast %m_d_148 : tensor<16x1xi32, #mma1> -> tensor<16x16xi32, #mma1> loc(#loc459) + %m_d_154 = tt.broadcast %m_d_149 : tensor<1x16xi32, #blocked1> -> tensor<16x16xi32, #blocked1> loc(#loc459) + %m_d_155 = tt.broadcast %m_d_150 : tensor<1x16xi32, #mma> -> tensor<16x16xi32, #mma> loc(#loc459) + %m_d_156 = tt.broadcast %m_d_151 : tensor<1x16xi32, #mma1> -> tensor<16x16xi32, #mma1> loc(#loc459) + %m_d_157 = arith.cmpi sgt, %m_d_152, %m_d_155 : tensor<16x16xi32, #mma> loc(#loc459) + %m_d_158 = arith.cmpi sgt, %m_d_153, %m_d_156 : tensor<16x16xi32, #mma1> loc(#loc459) + %m_I = arith.cmpi eq, %m_d_153, %m_d_156 : tensor<16x16xi32, #mma1> loc(#loc460) + %m_I_159 = arith.cmpi eq, %m_d_152, %m_d_155 : tensor<16x16xi32, #mma> loc(#loc460) + %b_A00 = tt.expand_dims %m_tc0_54 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<16x1xi1, #blocked1> loc(#loc461) + %b_A00_160 = tt.expand_dims %m_tc0_55 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #mma}>> -> tensor<16x1xi1, #mma> loc(#loc461) + %b_A00_161 = tt.broadcast %b_A00 : tensor<16x1xi1, #blocked1> -> tensor<16x16xi1, #blocked1> loc(#loc462) + %b_A00_162 = tt.broadcast %b_A00_160 : tensor<16x1xi1, #mma> -> tensor<16x16xi1, #mma> loc(#loc462) + %b_A00_163 = arith.andi %m_d_157, %b_A00_162 : tensor<16x16xi1, #mma> loc(#loc462) + %b_A00_164 = tt.expand_dims %m_tc0_56 {axis = 0 : i32} : tensor<16xi1, #ttg.slice<{dim = 0, parent = #mma}>> -> tensor<1x16xi1, #mma> loc(#loc463) + %b_A00_165 = tt.expand_dims %m_tc0_57 {axis = 0 : i32} : tensor<16xi1, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<1x16xi1, #mma1> loc(#loc463) + %b_A00_166 = tt.broadcast %b_A00_164 : tensor<1x16xi1, #mma> -> tensor<16x16xi1, #mma> loc(#loc464) + %b_A00_167 = tt.broadcast %b_A00_165 : tensor<1x16xi1, #mma1> -> tensor<16x16xi1, #mma1> loc(#loc464) + %b_A00_168 = arith.andi %b_A00_163, %b_A00_166 : tensor<16x16xi1, #mma> loc(#loc464) + %b_A00_169 = tt.expand_dims %b_g0 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma}>> -> tensor<16x1xf32, #mma> loc(#loc465) + %b_A00_170 = tt.expand_dims %b_g0_117 {axis = 0 : i32} : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma}>> -> tensor<1x16xf32, #mma> loc(#loc466) + %b_A00_171 = tt.expand_dims %b_g0_118 {axis = 0 : i32} : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<1x16xf32, #mma1> loc(#loc466) + %b_A00_172 = tt.broadcast %b_A00_169 : tensor<16x1xf32, #mma> -> tensor<16x16xf32, #mma> loc(#loc467) + %b_A00_173 = tt.broadcast %b_A00_170 : tensor<1x16xf32, #mma> -> tensor<16x16xf32, #mma> loc(#loc467) + %b_A00_174 = tt.broadcast %b_A00_171 : tensor<1x16xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc467) + %b_A00_175 = arith.subf %b_A00_172, %b_A00_173 : tensor<16x16xf32, #mma> loc(#loc467) + %b_A00_176 = math.exp2 %b_A00_175 : tensor<16x16xf32, #mma> loc(#loc679) + %b_A00_177 = arith.select %b_A00_168, %b_A00_176, %cst : tensor<16x16xi1, #mma>, tensor<16x16xf32, #mma> loc(#loc469) + %b_A00_178 = arith.mulf %b_A32#0, %b_A00_177 : tensor<16x16xf32, #mma> loc(#loc470) + %b_A11 = tt.expand_dims %m_tc1_66 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<16x1xi1, #blocked1> loc(#loc471) + %b_A11_179 = tt.expand_dims %m_tc1_67 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xi1, #mma1> loc(#loc471) + %b_A11_180 = tt.broadcast %b_A11 : tensor<16x1xi1, #blocked1> -> tensor<16x16xi1, #blocked1> loc(#loc472) + %b_A11_181 = tt.broadcast %b_A11_179 : tensor<16x1xi1, #mma1> -> tensor<16x16xi1, #mma1> loc(#loc472) + %b_A11_182 = arith.andi %m_d_158, %b_A11_181 : tensor<16x16xi1, #mma1> loc(#loc472) + %b_A11_183 = tt.expand_dims %m_tc1_68 {axis = 0 : i32} : tensor<16xi1, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<1x16xi1, #mma1> loc(#loc473) + %b_A11_184 = tt.broadcast %b_A11_183 : tensor<1x16xi1, #mma1> -> tensor<16x16xi1, #mma1> loc(#loc474) + %b_A11_185 = arith.andi %b_A11_182, %b_A11_184 : tensor<16x16xi1, #mma1> loc(#loc474) + %b_A11_186 = tt.expand_dims %b_g1 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xf32, #mma1> loc(#loc475) + %b_A11_187 = tt.expand_dims %b_g1_119 {axis = 0 : i32} : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<1x16xf32, #mma1> loc(#loc476) + %b_A11_188 = tt.broadcast %b_A11_186 : tensor<16x1xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc477) + %b_A11_189 = tt.broadcast %b_A11_187 : tensor<1x16xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc477) + %b_A11_190 = arith.subf %b_A11_188, %b_A11_189 : tensor<16x16xf32, #mma1> loc(#loc477) + %b_A11_191 = math.exp2 %b_A11_190 : tensor<16x16xf32, #mma1> loc(#loc680) + %b_A11_192 = arith.select %b_A11_185, %b_A11_191, %cst_0 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc479) + %b_A11_193 = arith.mulf %b_A32#1, %b_A11_192 : tensor<16x16xf32, #mma1> loc(#loc480) + %b_A22 = tt.expand_dims %m_tc2_77 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<16x1xi1, #blocked1> loc(#loc481) + %b_A22_194 = tt.expand_dims %m_tc2_78 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xi1, #mma1> loc(#loc481) + %b_A22_195 = tt.broadcast %b_A22 : tensor<16x1xi1, #blocked1> -> tensor<16x16xi1, #blocked1> loc(#loc482) + %b_A22_196 = tt.broadcast %b_A22_194 : tensor<16x1xi1, #mma1> -> tensor<16x16xi1, #mma1> loc(#loc482) + %b_A22_197 = arith.andi %m_d_158, %b_A22_196 : tensor<16x16xi1, #mma1> loc(#loc482) + %b_A22_198 = tt.expand_dims %m_tc2_79 {axis = 0 : i32} : tensor<16xi1, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<1x16xi1, #mma1> loc(#loc483) + %b_A22_199 = tt.broadcast %b_A22_198 : tensor<1x16xi1, #mma1> -> tensor<16x16xi1, #mma1> loc(#loc484) + %b_A22_200 = arith.andi %b_A22_197, %b_A22_199 : tensor<16x16xi1, #mma1> loc(#loc484) + %b_A22_201 = tt.expand_dims %b_g2 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xf32, #mma1> loc(#loc485) + %b_A22_202 = tt.expand_dims %b_g2_120 {axis = 0 : i32} : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<1x16xf32, #mma1> loc(#loc486) + %b_A22_203 = tt.broadcast %b_A22_201 : tensor<16x1xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc487) + %b_A22_204 = tt.broadcast %b_A22_202 : tensor<1x16xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc487) + %b_A22_205 = arith.subf %b_A22_203, %b_A22_204 : tensor<16x16xf32, #mma1> loc(#loc487) + %b_A22_206 = math.exp2 %b_A22_205 : tensor<16x16xf32, #mma1> loc(#loc681) + %b_A22_207 = arith.select %b_A22_200, %b_A22_206, %cst_0 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc489) + %b_A22_208 = arith.mulf %b_A32#2, %b_A22_207 : tensor<16x16xf32, #mma1> loc(#loc490) + %b_A33 = tt.expand_dims %m_tc3_88 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<16x1xi1, #blocked1> loc(#loc491) + %b_A33_209 = tt.expand_dims %m_tc3_89 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xi1, #mma1> loc(#loc491) + %b_A33_210 = tt.broadcast %b_A33 : tensor<16x1xi1, #blocked1> -> tensor<16x16xi1, #blocked1> loc(#loc492) + %b_A33_211 = tt.broadcast %b_A33_209 : tensor<16x1xi1, #mma1> -> tensor<16x16xi1, #mma1> loc(#loc492) + %b_A33_212 = arith.andi %m_d_158, %b_A33_211 : tensor<16x16xi1, #mma1> loc(#loc492) + %b_A33_213 = tt.expand_dims %m_tc3_90 {axis = 0 : i32} : tensor<16xi1, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<1x16xi1, #mma1> loc(#loc493) + %b_A33_214 = tt.broadcast %b_A33_213 : tensor<1x16xi1, #mma1> -> tensor<16x16xi1, #mma1> loc(#loc494) + %b_A33_215 = arith.andi %b_A33_212, %b_A33_214 : tensor<16x16xi1, #mma1> loc(#loc494) + %b_A33_216 = tt.expand_dims %b_g3 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xf32, #mma1> loc(#loc495) + %b_A33_217 = tt.expand_dims %b_g3_121 {axis = 0 : i32} : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<1x16xf32, #mma1> loc(#loc496) + %b_A33_218 = tt.broadcast %b_A33_216 : tensor<16x1xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc497) + %b_A33_219 = tt.broadcast %b_A33_217 : tensor<1x16xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc497) + %b_A33_220 = arith.subf %b_A33_218, %b_A33_219 : tensor<16x16xf32, #mma1> loc(#loc497) + %b_A33_221 = math.exp2 %b_A33_220 : tensor<16x16xf32, #mma1> loc(#loc682) + %b_A33_222 = arith.select %b_A33_215, %b_A33_221, %cst_0 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc499) + %b_A33_223 = arith.mulf %b_A32#3, %b_A33_222 : tensor<16x16xf32, #mma1> loc(#loc500) + %b_A10 = arith.andi %b_A11_181, %b_A00_167 : tensor<16x16xi1, #mma1> loc(#loc501) + %b_A10_224 = arith.subf %b_A11_188, %b_A00_174 : tensor<16x16xf32, #mma1> loc(#loc502) + %b_A10_225 = math.exp2 %b_A10_224 : tensor<16x16xf32, #mma1> loc(#loc683) + %b_A10_226 = arith.select %b_A10, %b_A10_225, %cst_0 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc504) + %b_A10_227 = arith.mulf %b_A32#4, %b_A10_226 : tensor<16x16xf32, #mma1> loc(#loc505) + %b_A20 = arith.andi %b_A22_196, %b_A00_167 : tensor<16x16xi1, #mma1> loc(#loc506) + %b_A20_228 = arith.subf %b_A22_203, %b_A00_174 : tensor<16x16xf32, #mma1> loc(#loc507) + %b_A20_229 = math.exp2 %b_A20_228 : tensor<16x16xf32, #mma1> loc(#loc684) + %b_A20_230 = arith.select %b_A20, %b_A20_229, %cst_0 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc509) + %b_A20_231 = arith.mulf %b_A32#5, %b_A20_230 : tensor<16x16xf32, #mma1> loc(#loc510) + %b_A21 = arith.andi %b_A22_196, %b_A11_184 : tensor<16x16xi1, #mma1> loc(#loc511) + %b_A21_232 = arith.subf %b_A22_203, %b_A11_189 : tensor<16x16xf32, #mma1> loc(#loc512) + %b_A21_233 = math.exp2 %b_A21_232 : tensor<16x16xf32, #mma1> loc(#loc685) + %b_A21_234 = arith.select %b_A21, %b_A21_233, %cst_0 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc514) + %b_A21_235 = arith.mulf %b_A32#6, %b_A21_234 : tensor<16x16xf32, #mma1> loc(#loc515) + %b_A30 = arith.andi %b_A33_211, %b_A00_167 : tensor<16x16xi1, #mma1> loc(#loc516) + %b_A30_236 = arith.subf %b_A33_218, %b_A00_174 : tensor<16x16xf32, #mma1> loc(#loc517) + %b_A30_237 = math.exp2 %b_A30_236 : tensor<16x16xf32, #mma1> loc(#loc686) + %b_A30_238 = arith.select %b_A30, %b_A30_237, %cst_0 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc519) + %b_A30_239 = arith.mulf %b_A32#7, %b_A30_238 : tensor<16x16xf32, #mma1> loc(#loc520) + %b_A31 = arith.andi %b_A33_211, %b_A11_184 : tensor<16x16xi1, #mma1> loc(#loc521) + %b_A31_240 = arith.subf %b_A33_218, %b_A11_189 : tensor<16x16xf32, #mma1> loc(#loc522) + %b_A31_241 = math.exp2 %b_A31_240 : tensor<16x16xf32, #mma1> loc(#loc687) + %b_A31_242 = arith.select %b_A31, %b_A31_241, %cst_0 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc524) + %b_A31_243 = arith.mulf %b_A32#8, %b_A31_242 : tensor<16x16xf32, #mma1> loc(#loc525) + %b_A32_244 = arith.andi %b_A33_211, %b_A22_199 : tensor<16x16xi1, #mma1> loc(#loc526) + %b_A32_245 = arith.subf %b_A33_218, %b_A22_204 : tensor<16x16xf32, #mma1> loc(#loc527) + %b_A32_246 = math.exp2 %b_A32_245 : tensor<16x16xf32, #mma1> loc(#loc688) + %b_A32_247 = arith.select %b_A32_244, %b_A32_246, %cst_0 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc529) + %b_A32_248 = arith.mulf %b_A32#9, %b_A32_247 : tensor<16x16xf32, #mma1> loc(#loc530) + %b_A00_249 = tt.expand_dims %b_b0_103 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma}>> -> tensor<16x1xf32, #mma> loc(#loc531) + %b_A00_250 = tt.broadcast %b_A00_249 : tensor<16x1xf32, #mma> -> tensor<16x16xf32, #mma> loc(#loc532) + %b_A00_251 = arith.mulf %b_A00_178, %b_A00_250 : tensor<16x16xf32, #mma> loc(#loc532) + %b_A11_252 = tt.expand_dims %b_b1_104 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xf32, #mma1> loc(#loc533) + %b_A11_253 = tt.broadcast %b_A11_252 : tensor<16x1xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc534) + %b_A11_254 = arith.mulf %b_A11_193, %b_A11_253 : tensor<16x16xf32, #mma1> loc(#loc534) + %b_A22_255 = tt.expand_dims %b_b2_105 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xf32, #mma1> loc(#loc535) + %b_A22_256 = tt.broadcast %b_A22_255 : tensor<16x1xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc536) + %b_A22_257 = arith.mulf %b_A22_208, %b_A22_256 : tensor<16x16xf32, #mma1> loc(#loc536) + %b_A33_258 = tt.expand_dims %b_b3_106 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xf32, #mma1> loc(#loc537) + %b_A33_259 = tt.broadcast %b_A33_258 : tensor<16x1xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc538) + %b_A33_260 = arith.mulf %b_A33_223, %b_A33_259 : tensor<16x16xf32, #mma1> loc(#loc538) + %b_A10_261 = arith.mulf %b_A10_227, %b_A11_253 : tensor<16x16xf32, #mma1> loc(#loc539) + %b_A20_262 = arith.mulf %b_A20_231, %b_A22_256 : tensor<16x16xf32, #mma1> loc(#loc540) + %b_A21_263 = arith.mulf %b_A21_235, %b_A22_256 : tensor<16x16xf32, #mma1> loc(#loc541) + %b_A30_264 = arith.mulf %b_A30_239, %b_A33_259 : tensor<16x16xf32, #mma1> loc(#loc542) + %b_A31_265 = arith.mulf %b_A31_243, %b_A33_259 : tensor<16x16xf32, #mma1> loc(#loc543) + %b_A32_266 = arith.mulf %b_A32_248, %b_A33_259 : tensor<16x16xf32, #mma1> loc(#loc544) + %b_Ai00 = arith.subf %cst, %b_A00_251 : tensor<16x16xf32, #mma> loc(#loc545) + %b_Ai11 = arith.subf %cst_0, %b_A11_254 : tensor<16x16xf32, #mma1> loc(#loc546) + %b_Ai22 = arith.subf %cst_0, %b_A22_257 : tensor<16x16xf32, #mma1> loc(#loc547) + %b_Ai33 = arith.subf %cst_0, %b_A33_260 : tensor<16x16xf32, #mma1> loc(#loc548) + %11 = arith.subi %6, %8 : i64 loc(#loc205) + %12 = arith.minsi %11, %c16_i64 : i64 loc(#loc206) + %b_Ai00_267 = scf.for %b_Ai00_355 = %c2_i64 to %12 step %c1_i64 iter_args(%b_Ai00_356 = %b_Ai00) -> (tensor<16x16xf32, #mma>) : i64 { + %b_a00 = tt.splat %b_Ai00_355 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc550) + %b_a00_357 = tt.splat %b_Ai00_355 : i64 -> tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc550) + %b_a00_358 = arith.cmpi eq, %m_tc0_33, %b_a00 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc550) + %b_a00_359 = tt.expand_dims %b_a00_358 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #mma}>> -> tensor<16x1xi1, #mma> loc(#loc551) + %b_a00_360 = tt.broadcast %b_a00_359 : tensor<16x1xi1, #mma> -> tensor<16x16xi1, #mma> loc(#loc552) + %b_a00_361 = arith.select %b_a00_360, %b_Ai00, %cst : tensor<16x16xi1, #mma>, tensor<16x16xf32, #mma> loc(#loc552) + %b_a00_362 = "tt.reduce"(%b_a00_361) <{axis = 0 : i32}> ({ + ^bb0(%b_a00_374: f32 loc(callsite(#loc1 at #loc553)), %b_a00_375: f32 loc(callsite(#loc1 at #loc553))): + %b_a00_376 = arith.addf %b_a00_374, %b_a00_375 : f32 loc(#loc708) + tt.reduce.return %b_a00_376 : f32 loc(#loc689) + }) : (tensor<16x16xf32, #mma>) -> tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc689) + %b_a00_363 = arith.cmpi slt, %m_tc0_35, %b_a00_357 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc554) + %b_a00_364 = arith.select %b_a00_363, %b_a00_362, %cst_1 : tensor<16xi1, #ttg.slice<{dim = 0, parent = #mma}>>, tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc555) + %b_a00_365 = ttg.convert_layout %b_a00_364 : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma}>> -> tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc556) + %b_a00_366 = tt.expand_dims %b_a00_365 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma}>> -> tensor<16x1xf32, #mma> loc(#loc557) + %b_a00_367 = tt.broadcast %b_a00_366 : tensor<16x1xf32, #mma> -> tensor<16x16xf32, #mma> loc(#loc556) + %b_a00_368 = arith.mulf %b_a00_367, %b_Ai00_356 : tensor<16x16xf32, #mma> loc(#loc556) + %b_a00_369 = "tt.reduce"(%b_a00_368) <{axis = 0 : i32}> ({ + ^bb0(%b_a00_374: f32 loc(callsite(#loc1 at #loc558)), %b_a00_375: f32 loc(callsite(#loc1 at #loc558))): + %b_a00_376 = arith.addf %b_a00_374, %b_a00_375 : f32 loc(#loc709) + tt.reduce.return %b_a00_376 : f32 loc(#loc691) + }) : (tensor<16x16xf32, #mma>) -> tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc691) + %b_a00_370 = arith.addf %b_a00_364, %b_a00_369 : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma}>> loc(#loc559) + %b_Ai00_371 = tt.expand_dims %b_a00_370 {axis = 0 : i32} : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma}>> -> tensor<1x16xf32, #mma> loc(#loc560) + %b_Ai00_372 = tt.broadcast %b_Ai00_371 : tensor<1x16xf32, #mma> -> tensor<16x16xf32, #mma> loc(#loc560) + %b_Ai00_373 = arith.select %b_a00_360, %b_Ai00_372, %b_Ai00_356 : tensor<16x16xi1, #mma>, tensor<16x16xf32, #mma> loc(#loc560) + scf.yield %b_Ai00_373 : tensor<16x16xf32, #mma> loc(#loc221) + } loc(#loc549) + %13 = arith.subi %6, %i_tc1 : i64 loc(#loc222) + %14 = arith.minsi %13, %c16_i64 : i64 loc(#loc223) + %b_Ai11_268 = scf.for %b_Ai11_355 = %c2_i64 to %14 step %c1_i64 iter_args(%b_Ai11_356 = %b_Ai11) -> (tensor<16x16xf32, #mma1>) : i64 { + %b_a11 = tt.splat %b_Ai11_355 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc562) + %b_a11_357 = tt.splat %b_Ai11_355 : i64 -> tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc562) + %b_a11_358 = arith.cmpi eq, %m_tc0_34, %b_a11 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc562) + %b_a11_359 = tt.expand_dims %b_a11_358 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xi1, #mma1> loc(#loc563) + %b_a11_360 = tt.broadcast %b_a11_359 : tensor<16x1xi1, #mma1> -> tensor<16x16xi1, #mma1> loc(#loc564) + %b_a11_361 = arith.select %b_a11_360, %b_Ai11, %cst_0 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc564) + %b_a11_362 = "tt.reduce"(%b_a11_361) <{axis = 0 : i32}> ({ + ^bb0(%b_a11_374: f32 loc(callsite(#loc1 at #loc565)), %b_a11_375: f32 loc(callsite(#loc1 at #loc565))): + %b_a11_376 = arith.addf %b_a11_374, %b_a11_375 : f32 loc(#loc710) + tt.reduce.return %b_a11_376 : f32 loc(#loc693) + }) : (tensor<16x16xf32, #mma1>) -> tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc693) + %b_a11_363 = arith.cmpi slt, %m_tc0_36, %b_a11_357 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc566) + %b_a11_364 = arith.select %b_a11_363, %b_a11_362, %cst_2 : tensor<16xi1, #ttg.slice<{dim = 0, parent = #mma1}>>, tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc567) + %b_a11_365 = ttg.convert_layout %b_a11_364 : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc568) + %b_a11_366 = tt.expand_dims %b_a11_365 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xf32, #mma1> loc(#loc569) + %b_a11_367 = tt.broadcast %b_a11_366 : tensor<16x1xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc568) + %b_a11_368 = arith.mulf %b_a11_367, %b_Ai11_356 : tensor<16x16xf32, #mma1> loc(#loc568) + %b_a11_369 = "tt.reduce"(%b_a11_368) <{axis = 0 : i32}> ({ + ^bb0(%b_a11_374: f32 loc(callsite(#loc1 at #loc570)), %b_a11_375: f32 loc(callsite(#loc1 at #loc570))): + %b_a11_376 = arith.addf %b_a11_374, %b_a11_375 : f32 loc(#loc711) + tt.reduce.return %b_a11_376 : f32 loc(#loc695) + }) : (tensor<16x16xf32, #mma1>) -> tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc695) + %b_a11_370 = arith.addf %b_a11_364, %b_a11_369 : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc571) + %b_Ai11_371 = tt.expand_dims %b_a11_370 {axis = 0 : i32} : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<1x16xf32, #mma1> loc(#loc572) + %b_Ai11_372 = tt.broadcast %b_Ai11_371 : tensor<1x16xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc572) + %b_Ai11_373 = arith.select %b_a11_360, %b_Ai11_372, %b_Ai11_356 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc572) + scf.yield %b_Ai11_373 : tensor<16x16xf32, #mma1> loc(#loc236) + } loc(#loc561) + %15 = arith.subi %6, %i_tc2 : i64 loc(#loc237) + %16 = arith.minsi %15, %c16_i64 : i64 loc(#loc238) + %b_Ai22_269 = scf.for %b_Ai22_355 = %c2_i64 to %16 step %c1_i64 iter_args(%b_Ai22_356 = %b_Ai22) -> (tensor<16x16xf32, #mma1>) : i64 { + %b_a22 = tt.splat %b_Ai22_355 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc574) + %b_a22_357 = tt.splat %b_Ai22_355 : i64 -> tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc574) + %b_a22_358 = arith.cmpi eq, %m_tc0_34, %b_a22 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc574) + %b_a22_359 = tt.expand_dims %b_a22_358 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xi1, #mma1> loc(#loc575) + %b_a22_360 = tt.broadcast %b_a22_359 : tensor<16x1xi1, #mma1> -> tensor<16x16xi1, #mma1> loc(#loc576) + %b_a22_361 = arith.select %b_a22_360, %b_Ai22, %cst_0 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc576) + %b_a22_362 = "tt.reduce"(%b_a22_361) <{axis = 0 : i32}> ({ + ^bb0(%b_a22_374: f32 loc(callsite(#loc1 at #loc577)), %b_a22_375: f32 loc(callsite(#loc1 at #loc577))): + %b_a22_376 = arith.addf %b_a22_374, %b_a22_375 : f32 loc(#loc712) + tt.reduce.return %b_a22_376 : f32 loc(#loc697) + }) : (tensor<16x16xf32, #mma1>) -> tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc697) + %b_a22_363 = arith.cmpi slt, %m_tc0_36, %b_a22_357 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc578) + %b_a22_364 = arith.select %b_a22_363, %b_a22_362, %cst_2 : tensor<16xi1, #ttg.slice<{dim = 0, parent = #mma1}>>, tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc579) + %b_a22_365 = ttg.convert_layout %b_a22_364 : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc580) + %b_a22_366 = tt.expand_dims %b_a22_365 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xf32, #mma1> loc(#loc581) + %b_a22_367 = tt.broadcast %b_a22_366 : tensor<16x1xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc580) + %b_a22_368 = arith.mulf %b_a22_367, %b_Ai22_356 : tensor<16x16xf32, #mma1> loc(#loc580) + %b_a22_369 = "tt.reduce"(%b_a22_368) <{axis = 0 : i32}> ({ + ^bb0(%b_a22_374: f32 loc(callsite(#loc1 at #loc582)), %b_a22_375: f32 loc(callsite(#loc1 at #loc582))): + %b_a22_376 = arith.addf %b_a22_374, %b_a22_375 : f32 loc(#loc713) + tt.reduce.return %b_a22_376 : f32 loc(#loc699) + }) : (tensor<16x16xf32, #mma1>) -> tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc699) + %b_a22_370 = arith.addf %b_a22_364, %b_a22_369 : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc583) + %b_Ai22_371 = tt.expand_dims %b_a22_370 {axis = 0 : i32} : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<1x16xf32, #mma1> loc(#loc584) + %b_Ai22_372 = tt.broadcast %b_Ai22_371 : tensor<1x16xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc584) + %b_Ai22_373 = arith.select %b_a22_360, %b_Ai22_372, %b_Ai22_356 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc584) + scf.yield %b_Ai22_373 : tensor<16x16xf32, #mma1> loc(#loc251) + } loc(#loc573) + %17 = arith.subi %6, %i_tc3 : i64 loc(#loc252) + %18 = arith.minsi %17, %c16_i64 : i64 loc(#loc253) + %b_Ai33_270 = scf.for %b_Ai33_355 = %c2_i64 to %18 step %c1_i64 iter_args(%b_Ai33_356 = %b_Ai33) -> (tensor<16x16xf32, #mma1>) : i64 { + %b_a33 = tt.splat %b_Ai33_355 : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc586) + %b_a33_357 = tt.splat %b_Ai33_355 : i64 -> tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc586) + %b_a33_358 = arith.cmpi eq, %m_tc0_34, %b_a33 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc586) + %b_a33_359 = tt.expand_dims %b_a33_358 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xi1, #mma1> loc(#loc587) + %b_a33_360 = tt.broadcast %b_a33_359 : tensor<16x1xi1, #mma1> -> tensor<16x16xi1, #mma1> loc(#loc588) + %b_a33_361 = arith.select %b_a33_360, %b_Ai33, %cst_0 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc588) + %b_a33_362 = "tt.reduce"(%b_a33_361) <{axis = 0 : i32}> ({ + ^bb0(%b_a33_374: f32 loc(callsite(#loc1 at #loc589)), %b_a33_375: f32 loc(callsite(#loc1 at #loc589))): + %b_a33_376 = arith.addf %b_a33_374, %b_a33_375 : f32 loc(#loc714) + tt.reduce.return %b_a33_376 : f32 loc(#loc701) + }) : (tensor<16x16xf32, #mma1>) -> tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc701) + %b_a33_363 = arith.cmpi slt, %m_tc0_36, %b_a33_357 : tensor<16xi64, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc590) + %b_a33_364 = arith.select %b_a33_363, %b_a33_362, %cst_2 : tensor<16xi1, #ttg.slice<{dim = 0, parent = #mma1}>>, tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc591) + %b_a33_365 = ttg.convert_layout %b_a33_364 : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> loc(#loc592) + %b_a33_366 = tt.expand_dims %b_a33_365 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #mma1}>> -> tensor<16x1xf32, #mma1> loc(#loc593) + %b_a33_367 = tt.broadcast %b_a33_366 : tensor<16x1xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc592) + %b_a33_368 = arith.mulf %b_a33_367, %b_Ai33_356 : tensor<16x16xf32, #mma1> loc(#loc592) + %b_a33_369 = "tt.reduce"(%b_a33_368) <{axis = 0 : i32}> ({ + ^bb0(%b_a33_374: f32 loc(callsite(#loc1 at #loc594)), %b_a33_375: f32 loc(callsite(#loc1 at #loc594))): + %b_a33_376 = arith.addf %b_a33_374, %b_a33_375 : f32 loc(#loc715) + tt.reduce.return %b_a33_376 : f32 loc(#loc703) + }) : (tensor<16x16xf32, #mma1>) -> tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc703) + %b_a33_370 = arith.addf %b_a33_364, %b_a33_369 : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> loc(#loc595) + %b_Ai33_371 = tt.expand_dims %b_a33_370 {axis = 0 : i32} : tensor<16xf32, #ttg.slice<{dim = 0, parent = #mma1}>> -> tensor<1x16xf32, #mma1> loc(#loc596) + %b_Ai33_372 = tt.broadcast %b_Ai33_371 : tensor<1x16xf32, #mma1> -> tensor<16x16xf32, #mma1> loc(#loc596) + %b_Ai33_373 = arith.select %b_a33_360, %b_Ai33_372, %b_Ai33_356 : tensor<16x16xi1, #mma1>, tensor<16x16xf32, #mma1> loc(#loc596) + scf.yield %b_Ai33_373 : tensor<16x16xf32, #mma1> loc(#loc266) + } loc(#loc585) + %b_Ai00_271 = arith.uitofp %m_I : tensor<16x16xi1, #mma1> to tensor<16x16xf32, #mma1> loc(#loc597) + %b_Ai00_272 = arith.uitofp %m_I_159 : tensor<16x16xi1, #mma> to tensor<16x16xf32, #mma> loc(#loc597) + %b_Ai00_273 = arith.addf %b_Ai00_267, %b_Ai00_272 : tensor<16x16xf32, #mma> loc(#loc597) + %b_Ai00_274 = ttg.local_alloc %b_Ai00_273 : (tensor<16x16xf32, #mma>) -> !ttg.memdesc<16x16xf32, #shared1, #smem> loc(#loc597) + %b_Ai11_275 = arith.addf %b_Ai11_268, %b_Ai00_271 : tensor<16x16xf32, #mma1> loc(#loc598) + %b_Ai22_276 = arith.addf %b_Ai22_269, %b_Ai00_271 : tensor<16x16xf32, #mma1> loc(#loc599) + %b_Ai33_277 = arith.addf %b_Ai33_270, %b_Ai00_271 : tensor<16x16xf32, #mma1> loc(#loc600) + %b_Ai11_278 = ttg.convert_layout %b_Ai11_275 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> loc(#loc598) + %b_A10_279 = ttg.convert_layout %b_A10_261 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> loc(#loc539) + %b_Ai10 = tt.dot %b_Ai11_278, %b_A10_279, %cst_0, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc601) + %b_Ai10_280 = ttg.convert_layout %b_Ai10 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 1}>> loc(#loc601) + %b_Ai00_281 = ttg.local_load %b_Ai00_274 : !ttg.memdesc<16x16xf32, #shared1, #smem> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 1}>> loc(#loc597) + %b_Ai10_282 = tt.dot %b_Ai10_280, %b_Ai00_281, %cst_0, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 1}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 1}>> -> tensor<16x16xf32, #mma1> loc(#loc602) + %b_Ai10_283 = arith.subf %cst_0, %b_Ai10_282 : tensor<16x16xf32, #mma1> loc(#loc603) + %b_Ai22_284 = ttg.convert_layout %b_Ai22_276 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> loc(#loc599) + %b_A21_285 = ttg.convert_layout %b_A21_263 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> loc(#loc541) + %b_Ai21 = tt.dot %b_Ai22_284, %b_A21_285, %cst_0, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc604) + %b_Ai21_286 = ttg.convert_layout %b_Ai21 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 1}>> loc(#loc604) + %b_Ai11_287 = ttg.convert_layout %b_Ai11_275 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 1}>> loc(#loc598) + %b_Ai21_288 = tt.dot %b_Ai21_286, %b_Ai11_287, %cst_0, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 1}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 1}>> -> tensor<16x16xf32, #mma1> loc(#loc605) + %b_Ai21_289 = arith.subf %cst_0, %b_Ai21_288 : tensor<16x16xf32, #mma1> loc(#loc606) + %b_Ai33_290 = ttg.convert_layout %b_Ai33_277 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> loc(#loc600) + %b_A32_291 = ttg.convert_layout %b_A32_266 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> loc(#loc544) + %b_Ai32 = tt.dot %b_Ai33_290, %b_A32_291, %cst_0, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc607) + %b_Ai32_292 = ttg.convert_layout %b_Ai32 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 1}>> loc(#loc607) + %b_Ai22_293 = ttg.convert_layout %b_Ai22_276 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 1}>> loc(#loc599) + %b_Ai32_294 = tt.dot %b_Ai32_292, %b_Ai22_293, %cst_0, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 1}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 1}>> -> tensor<16x16xf32, #mma1> loc(#loc608) + %b_Ai32_295 = arith.subf %cst_0, %b_Ai32_294 : tensor<16x16xf32, #mma1> loc(#loc609) + %b_A20_296 = ttg.convert_layout %b_A20_262 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> loc(#loc540) + %b_Ai00_297 = ttg.local_load %b_Ai00_274 : !ttg.memdesc<16x16xf32, #shared1, #smem> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> loc(#loc597) + %b_Ai20 = tt.dot %b_A20_296, %b_Ai00_297, %cst_0, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc610) + %b_A21_298 = ttg.convert_layout %b_A21_263 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> loc(#loc541) + %b_Ai10_299 = ttg.convert_layout %b_Ai10_283 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> loc(#loc603) + %b_Ai20_300 = tt.dot %b_A21_298, %b_Ai10_299, %b_Ai20, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc611) + %b_Ai22_301 = ttg.convert_layout %b_Ai22_276 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 1}>> loc(#loc599) + %b_Ai20_302 = ttg.convert_layout %b_Ai20_300 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 1}>> loc(#loc611) + %b_Ai20_303 = tt.dot %b_Ai22_301, %b_Ai20_302, %cst_0, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 1}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 1}>> -> tensor<16x16xf32, #mma1> loc(#loc612) + %b_Ai20_304 = arith.subf %cst_0, %b_Ai20_303 : tensor<16x16xf32, #mma1> loc(#loc613) + %b_A31_305 = ttg.convert_layout %b_A31_265 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> loc(#loc543) + %b_Ai11_306 = ttg.convert_layout %b_Ai11_275 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> loc(#loc598) + %b_Ai31 = tt.dot %b_A31_305, %b_Ai11_306, %cst_0, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc614) + %b_A32_307 = ttg.convert_layout %b_A32_266 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> loc(#loc544) + %b_Ai21_308 = ttg.convert_layout %b_Ai21_289 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> loc(#loc606) + %b_Ai31_309 = tt.dot %b_A32_307, %b_Ai21_308, %b_Ai31, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc615) + %b_Ai33_310 = ttg.convert_layout %b_Ai33_277 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 1}>> loc(#loc600) + %b_Ai31_311 = ttg.convert_layout %b_Ai31_309 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 1}>> loc(#loc615) + %b_Ai31_312 = tt.dot %b_Ai33_310, %b_Ai31_311, %cst_0, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 1}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 1}>> -> tensor<16x16xf32, #mma1> loc(#loc616) + %b_Ai31_313 = arith.subf %cst_0, %b_Ai31_312 : tensor<16x16xf32, #mma1> loc(#loc617) + %b_A30_314 = ttg.convert_layout %b_A30_264 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> loc(#loc542) + %b_Ai30 = tt.dot %b_A30_314, %b_Ai00_297, %cst_0, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc618) + %b_Ai30_315 = tt.dot %b_A31_305, %b_Ai10_299, %b_Ai30, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc619) + %b_Ai20_316 = ttg.convert_layout %b_Ai20_304 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> loc(#loc613) + %b_Ai30_317 = tt.dot %b_A32_307, %b_Ai20_316, %b_Ai30_315, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 2}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 2}>> -> tensor<16x16xf32, #mma1> loc(#loc620) + %b_Ai30_318 = ttg.convert_layout %b_Ai30_317 : tensor<16x16xf32, #mma1> -> tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 1}>> loc(#loc620) + %b_Ai30_319 = tt.dot %b_Ai33_310, %b_Ai30_318, %cst_0, inputPrecision = tf32 : tensor<16x16xf32, #ttg.dot_op<{opIdx = 0, parent = #mma1, kWidth = 1}>> * tensor<16x16xf32, #ttg.dot_op<{opIdx = 1, parent = #mma1, kWidth = 1}>> -> tensor<16x16xf32, #mma1> loc(#loc621) + %b_Ai30_320 = arith.subf %cst_0, %b_Ai30_319 : tensor<16x16xf32, #mma1> loc(#loc622) + %p_A00 = tt.expand_dims %m_tc0_43 {axis = 1 : i32} : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<16x1xi64, #blocked1> loc(#loc623) + %p_A00_321 = arith.muli %p_A00, %cst_8 : tensor<16x1xi64, #blocked1> loc(#loc624) + %p_A00_322 = tt.splat %A_25 : !tt.ptr -> tensor<16x1x!tt.ptr, #blocked1> loc(#loc625) + %p_A00_323 = tt.addptr %p_A00_322, %p_A00_321 : tensor<16x1x!tt.ptr, #blocked1>, tensor<16x1xi64, #blocked1> loc(#loc625) + %p_A00_324 = tt.broadcast %p_A00_323 : tensor<16x1x!tt.ptr, #blocked1> -> tensor<16x16x!tt.ptr, #blocked1> loc(#loc626) + %p_A00_325 = tt.addptr %p_A00_324, %m_d_154 : tensor<16x16x!tt.ptr, #blocked1>, tensor<16x16xi32, #blocked1> loc(#loc626) + %p_A10 = tt.expand_dims %m_tc1_62 {axis = 1 : i32} : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<16x1xi64, #blocked1> loc(#loc627) + %p_A10_326 = arith.muli %p_A10, %cst_8 : tensor<16x1xi64, #blocked1> loc(#loc628) + %p_A10_327 = tt.addptr %p_A00_322, %p_A10_326 : tensor<16x1x!tt.ptr, #blocked1>, tensor<16x1xi64, #blocked1> loc(#loc629) + %p_A10_328 = tt.broadcast %p_A10_327 : tensor<16x1x!tt.ptr, #blocked1> -> tensor<16x16x!tt.ptr, #blocked1> loc(#loc630) + %p_A10_329 = tt.addptr %p_A10_328, %m_d_154 : tensor<16x16x!tt.ptr, #blocked1>, tensor<16x16xi32, #blocked1> loc(#loc630) + %p_A11 = arith.addi %o_i_29, %cst_7 : tensor<16xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc631) + %p_A11_330 = tt.expand_dims %p_A11 {axis = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x16xi32, #blocked1> loc(#loc632) + %p_A11_331 = tt.broadcast %p_A11_330 : tensor<1x16xi32, #blocked1> -> tensor<16x16xi32, #blocked1> loc(#loc633) + %p_A11_332 = tt.addptr %p_A10_328, %p_A11_331 : tensor<16x16x!tt.ptr, #blocked1>, tensor<16x16xi32, #blocked1> loc(#loc633) + %p_A20 = tt.expand_dims %m_tc2_73 {axis = 1 : i32} : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<16x1xi64, #blocked1> loc(#loc634) + %p_A20_333 = arith.muli %p_A20, %cst_8 : tensor<16x1xi64, #blocked1> loc(#loc635) + %p_A20_334 = tt.addptr %p_A00_322, %p_A20_333 : tensor<16x1x!tt.ptr, #blocked1>, tensor<16x1xi64, #blocked1> loc(#loc636) + %p_A20_335 = tt.broadcast %p_A20_334 : tensor<16x1x!tt.ptr, #blocked1> -> tensor<16x16x!tt.ptr, #blocked1> loc(#loc637) + %p_A20_336 = tt.addptr %p_A20_335, %m_d_154 : tensor<16x16x!tt.ptr, #blocked1>, tensor<16x16xi32, #blocked1> loc(#loc637) + %p_A21 = tt.addptr %p_A20_335, %p_A11_331 : tensor<16x16x!tt.ptr, #blocked1>, tensor<16x16xi32, #blocked1> loc(#loc638) + %p_A22 = arith.addi %o_i_29, %cst_6 : tensor<16xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc639) + %p_A22_337 = tt.expand_dims %p_A22 {axis = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x16xi32, #blocked1> loc(#loc640) + %p_A22_338 = tt.broadcast %p_A22_337 : tensor<1x16xi32, #blocked1> -> tensor<16x16xi32, #blocked1> loc(#loc641) + %p_A22_339 = tt.addptr %p_A20_335, %p_A22_338 : tensor<16x16x!tt.ptr, #blocked1>, tensor<16x16xi32, #blocked1> loc(#loc641) + %p_A30 = tt.expand_dims %m_tc3_84 {axis = 1 : i32} : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<16x1xi64, #blocked1> loc(#loc642) + %p_A30_340 = arith.muli %p_A30, %cst_8 : tensor<16x1xi64, #blocked1> loc(#loc643) + %p_A30_341 = tt.addptr %p_A00_322, %p_A30_340 : tensor<16x1x!tt.ptr, #blocked1>, tensor<16x1xi64, #blocked1> loc(#loc644) + %p_A30_342 = tt.broadcast %p_A30_341 : tensor<16x1x!tt.ptr, #blocked1> -> tensor<16x16x!tt.ptr, #blocked1> loc(#loc645) + %p_A30_343 = tt.addptr %p_A30_342, %m_d_154 : tensor<16x16x!tt.ptr, #blocked1>, tensor<16x16xi32, #blocked1> loc(#loc645) + %p_A31 = tt.addptr %p_A30_342, %p_A11_331 : tensor<16x16x!tt.ptr, #blocked1>, tensor<16x16xi32, #blocked1> loc(#loc646) + %p_A32 = tt.addptr %p_A30_342, %p_A22_338 : tensor<16x16x!tt.ptr, #blocked1>, tensor<16x16xi32, #blocked1> loc(#loc647) + %p_A33 = arith.addi %o_i_29, %cst_5 : tensor<16xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc648) + %p_A33_344 = tt.expand_dims %p_A33 {axis = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x16xi32, #blocked1> loc(#loc649) + %p_A33_345 = tt.broadcast %p_A33_344 : tensor<1x16xi32, #blocked1> -> tensor<16x16xi32, #blocked1> loc(#loc650) + %p_A33_346 = tt.addptr %p_A30_342, %p_A33_345 : tensor<16x16x!tt.ptr, #blocked1>, tensor<16x16xi32, #blocked1> loc(#loc650) + %m_A0 = arith.cmpi slt, %m_d_149, %cst_4 : tensor<1x16xi32, #blocked1> loc(#loc651) + %m_A0_347 = tt.broadcast %m_A0 : tensor<1x16xi1, #blocked1> -> tensor<16x16xi1, #blocked1> loc(#loc652) + %m_A0_348 = arith.andi %b_A00_161, %m_A0_347 : tensor<16x16xi1, #blocked1> loc(#loc652) + %m_A1 = arith.andi %b_A11_180, %m_A0_347 : tensor<16x16xi1, #blocked1> loc(#loc653) + %m_A2 = arith.andi %b_A22_195, %m_A0_347 : tensor<16x16xi1, #blocked1> loc(#loc654) + %m_A3 = arith.andi %b_A33_210, %m_A0_347 : tensor<16x16xi1, #blocked1> loc(#loc655) + %m_A11 = arith.cmpi slt, %p_A11_330, %cst_4 : tensor<1x16xi32, #blocked1> loc(#loc656) + %m_A11_349 = tt.broadcast %m_A11 : tensor<1x16xi1, #blocked1> -> tensor<16x16xi1, #blocked1> loc(#loc657) + %m_A11_350 = arith.andi %b_A11_180, %m_A11_349 : tensor<16x16xi1, #blocked1> loc(#loc657) + %m_A21 = arith.andi %b_A22_195, %m_A11_349 : tensor<16x16xi1, #blocked1> loc(#loc658) + %m_A22 = arith.cmpi slt, %p_A22_337, %cst_4 : tensor<1x16xi32, #blocked1> loc(#loc659) + %m_A22_351 = tt.broadcast %m_A22 : tensor<1x16xi1, #blocked1> -> tensor<16x16xi1, #blocked1> loc(#loc660) + %m_A22_352 = arith.andi %b_A22_195, %m_A22_351 : tensor<16x16xi1, #blocked1> loc(#loc660) + %m_A31 = arith.andi %b_A33_210, %m_A11_349 : tensor<16x16xi1, #blocked1> loc(#loc661) + %m_A32 = arith.andi %b_A33_210, %m_A22_351 : tensor<16x16xi1, #blocked1> loc(#loc662) + %m_A33 = arith.cmpi slt, %p_A33_344, %cst_4 : tensor<1x16xi32, #blocked1> loc(#loc663) + %m_A33_353 = tt.broadcast %m_A33 : tensor<1x16xi1, #blocked1> -> tensor<16x16xi1, #blocked1> loc(#loc664) + %m_A33_354 = arith.andi %b_A33_210, %m_A33_353 : tensor<16x16xi1, #blocked1> loc(#loc664) + %19 = arith.truncf %b_Ai00_273 : tensor<16x16xf32, #mma> to tensor<16x16xbf16, #mma> loc(#loc335) + %20 = ttg.convert_layout %19 : tensor<16x16xbf16, #mma> -> tensor<16x16xbf16, #blocked1> loc(#loc336) + tt.store %p_A00_325, %20, %m_A0_348 : tensor<16x16x!tt.ptr, #blocked1> loc(#loc336) + %21 = arith.truncf %b_Ai10_283 : tensor<16x16xf32, #mma1> to tensor<16x16xbf16, #mma1> loc(#loc337) + %22 = ttg.convert_layout %21 : tensor<16x16xbf16, #mma1> -> tensor<16x16xbf16, #blocked1> loc(#loc338) + tt.store %p_A10_329, %22, %m_A1 : tensor<16x16x!tt.ptr, #blocked1> loc(#loc338) + %23 = arith.truncf %b_Ai11_275 : tensor<16x16xf32, #mma1> to tensor<16x16xbf16, #mma1> loc(#loc339) + %24 = ttg.convert_layout %23 : tensor<16x16xbf16, #mma1> -> tensor<16x16xbf16, #blocked1> loc(#loc340) + tt.store %p_A11_332, %24, %m_A11_350 : tensor<16x16x!tt.ptr, #blocked1> loc(#loc340) + %25 = arith.truncf %b_Ai20_304 : tensor<16x16xf32, #mma1> to tensor<16x16xbf16, #mma1> loc(#loc341) + %26 = ttg.convert_layout %25 : tensor<16x16xbf16, #mma1> -> tensor<16x16xbf16, #blocked1> loc(#loc342) + tt.store %p_A20_336, %26, %m_A2 : tensor<16x16x!tt.ptr, #blocked1> loc(#loc342) + %27 = arith.truncf %b_Ai21_289 : tensor<16x16xf32, #mma1> to tensor<16x16xbf16, #mma1> loc(#loc343) + %28 = ttg.convert_layout %27 : tensor<16x16xbf16, #mma1> -> tensor<16x16xbf16, #blocked1> loc(#loc344) + tt.store %p_A21, %28, %m_A21 : tensor<16x16x!tt.ptr, #blocked1> loc(#loc344) + %29 = arith.truncf %b_Ai22_276 : tensor<16x16xf32, #mma1> to tensor<16x16xbf16, #mma1> loc(#loc345) + %30 = ttg.convert_layout %29 : tensor<16x16xbf16, #mma1> -> tensor<16x16xbf16, #blocked1> loc(#loc346) + tt.store %p_A22_339, %30, %m_A22_352 : tensor<16x16x!tt.ptr, #blocked1> loc(#loc346) + %31 = arith.truncf %b_Ai30_320 : tensor<16x16xf32, #mma1> to tensor<16x16xbf16, #mma1> loc(#loc347) + %32 = ttg.convert_layout %31 : tensor<16x16xbf16, #mma1> -> tensor<16x16xbf16, #blocked1> loc(#loc348) + tt.store %p_A30_343, %32, %m_A3 : tensor<16x16x!tt.ptr, #blocked1> loc(#loc348) + %33 = arith.truncf %b_Ai31_313 : tensor<16x16xf32, #mma1> to tensor<16x16xbf16, #mma1> loc(#loc349) + %34 = ttg.convert_layout %33 : tensor<16x16xbf16, #mma1> -> tensor<16x16xbf16, #blocked1> loc(#loc350) + tt.store %p_A31, %34, %m_A31 : tensor<16x16x!tt.ptr, #blocked1> loc(#loc350) + %35 = arith.truncf %b_Ai32_295 : tensor<16x16xf32, #mma1> to tensor<16x16xbf16, #mma1> loc(#loc351) + %36 = ttg.convert_layout %35 : tensor<16x16xbf16, #mma1> -> tensor<16x16xbf16, #blocked1> loc(#loc352) + tt.store %p_A32, %36, %m_A32 : tensor<16x16x!tt.ptr, #blocked1> loc(#loc352) + %37 = arith.truncf %b_Ai33_277 : tensor<16x16xf32, #mma1> to tensor<16x16xbf16, #mma1> loc(#loc353) + %38 = ttg.convert_layout %37 : tensor<16x16xbf16, #mma1> -> tensor<16x16xbf16, #blocked1> loc(#loc354) + tt.store %p_A33_346, %38, %m_A33_354 : tensor<16x16x!tt.ptr, #blocked1> loc(#loc354) + tt.return loc(#loc355) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":70:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":70:36) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":70:61) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":70:67) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":71:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":71:34) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":78:25) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":80:13) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":80:19) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":81:8) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":84:23) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":85:23) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":86:23) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":88:16) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":88:20) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":88:40) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":88:9) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":89:28) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":89:9) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":91:23) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":92:21) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":92:28) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":93:21) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":93:28) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":94:21) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":94:28) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":95:21) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":95:28) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":98:29) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":98:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":98:51) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":98:35) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":99:51) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":99:35) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":100:51) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":100:35) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":101:51) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":101:35) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":102:19) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":102:51) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":103:19) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":103:51) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":104:19) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":104:51) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":105:19) +#loc47 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":105:51) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":109:30) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":109:19) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":109:36) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":110:36) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":111:36) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":112:36) +#loc54 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":114:23) +#loc55 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":115:23) +#loc56 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":116:23) +#loc57 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":117:23) +#loc58 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":138:38) +#loc59 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:33) +#loc60 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:45) +#loc61 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:19) +#loc62 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:52) +#loc63 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":140:40) +#loc64 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":140:52) +#loc65 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":144:19) +#loc66 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":140:23) +#loc67 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:56) +#loc68 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":140:67) +#loc69 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":138:25) +#loc70 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":137:21) +#loc71 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":142:45) +#loc72 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":142:38) +#loc73 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":138:20) +#loc74 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":144:11) +#loc75 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":145:37) +#loc76 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":145:49) +#loc77 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":145:23) +#loc78 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":145:56) +#loc79 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":146:44) +#loc80 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":146:56) +#loc81 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":146:27) +#loc82 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":148:42) +#loc83 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":148:49) +#loc84 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":150:49) +#loc85 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":152:23) +#loc86 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":152:15) +#loc87 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":153:41) +#loc88 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":153:53) +#loc89 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":153:27) +#loc90 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":153:60) +#loc91 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":154:48) +#loc92 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":154:60) +#loc93 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":154:31) +#loc94 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":156:46) +#loc95 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":156:53) +#loc96 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":158:53) +#loc97 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":159:53) +#loc98 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":161:27) +#loc99 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":161:19) +#loc100 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":162:45) +#loc101 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":162:57) +#loc102 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":162:31) +#loc103 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":162:64) +#loc104 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":163:52) +#loc105 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":163:64) +#loc106 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":163:35) +#loc107 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":165:50) +#loc108 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":165:57) +#loc109 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":167:57) +#loc110 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":168:57) +#loc111 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":169:57) +#loc112 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":179:14) +#loc113 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":179:29) +#loc114 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":179:25) +#loc115 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":180:26) +#loc116 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:38) +#loc117 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:32) +#loc118 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:55) +#loc119 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:49) +#loc120 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:75) +#loc121 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:91) +#loc122 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:86) +#loc123 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:33) +#loc124 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:70) +#loc125 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:102) +#loc126 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:17) +#loc127 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:38) +#loc128 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:32) +#loc129 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:55) +#loc130 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:49) +#loc131 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:75) +#loc132 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:91) +#loc133 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:86) +#loc134 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:70) +#loc135 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:102) +#loc136 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:17) +#loc137 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:38) +#loc138 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:32) +#loc139 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:55) +#loc140 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:49) +#loc141 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:75) +#loc142 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:91) +#loc143 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:86) +#loc144 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:70) +#loc145 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:102) +#loc146 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:17) +#loc147 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:38) +#loc148 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:32) +#loc149 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:55) +#loc150 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:49) +#loc151 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:75) +#loc152 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:91) +#loc153 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:86) +#loc154 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:70) +#loc155 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:102) +#loc156 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:17) +#loc157 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:43) +#loc158 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:80) +#loc159 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:64) +#loc160 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:96) +#loc161 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:17) +#loc162 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:43) +#loc163 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:80) +#loc164 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:64) +#loc165 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:96) +#loc166 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:17) +#loc167 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:43) +#loc168 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:80) +#loc169 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:64) +#loc170 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:96) +#loc171 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:17) +#loc172 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:43) +#loc173 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:80) +#loc174 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:64) +#loc175 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:96) +#loc176 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:17) +#loc177 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:43) +#loc178 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:80) +#loc179 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:64) +#loc180 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:96) +#loc181 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:17) +#loc182 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:43) +#loc183 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:80) +#loc184 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:64) +#loc185 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:96) +#loc186 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:17) +#loc187 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":201:25) +#loc188 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":201:20) +#loc189 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":202:25) +#loc190 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":202:20) +#loc191 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":203:25) +#loc192 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":203:20) +#loc193 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":204:25) +#loc194 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":204:20) +#loc195 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":207:20) +#loc196 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":208:20) +#loc197 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":209:20) +#loc198 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":210:20) +#loc199 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":211:20) +#loc200 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":212:20) +#loc201 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":222:14) +#loc202 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":223:14) +#loc203 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":224:14) +#loc204 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":225:14) +#loc205 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":227:34) +#loc206 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":227:30) +#loc207 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":227:22) +#loc208 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":228:40) +#loc209 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":228:43) +#loc210 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":228:61) +#loc211 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc213 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc214 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":229:31) +#loc215 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":229:41) +#loc216 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":230:48) +#loc217 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":230:37) +#loc219 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":230:24) +#loc220 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":231:54) +#loc221 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":231:8) +#loc222 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":232:34) +#loc223 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":232:30) +#loc224 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":232:22) +#loc225 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":233:40) +#loc226 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":233:43) +#loc227 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":233:61) +#loc229 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":234:31) +#loc230 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":234:41) +#loc231 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":235:48) +#loc232 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":235:37) +#loc234 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":235:24) +#loc235 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":236:54) +#loc236 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":236:8) +#loc237 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":237:34) +#loc238 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":237:30) +#loc239 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":237:22) +#loc240 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":238:40) +#loc241 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":238:43) +#loc242 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":238:61) +#loc244 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":239:31) +#loc245 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":239:41) +#loc246 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":240:48) +#loc247 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":240:37) +#loc249 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":240:24) +#loc250 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":241:54) +#loc251 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":241:8) +#loc252 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":242:34) +#loc253 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":242:30) +#loc254 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":242:22) +#loc255 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":243:40) +#loc256 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":243:43) +#loc257 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":243:61) +#loc259 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":244:31) +#loc260 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":244:41) +#loc261 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":245:48) +#loc262 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":245:37) +#loc264 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":245:24) +#loc265 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":246:54) +#loc266 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":246:8) +#loc267 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":248:14) +#loc268 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":249:14) +#loc269 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":250:14) +#loc270 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":251:14) +#loc271 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":258:23) +#loc272 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":259:8) +#loc273 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":257:14) +#loc274 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":263:23) +#loc275 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":264:8) +#loc276 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":262:14) +#loc277 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":268:23) +#loc278 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":269:8) +#loc279 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":267:14) +#loc280 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":275:22) +#loc281 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":276:22) +#loc282 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":275:8) +#loc283 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":273:14) +#loc284 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":281:22) +#loc285 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":282:22) +#loc286 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":281:8) +#loc287 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":279:14) +#loc288 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":287:22) +#loc289 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":288:22) +#loc290 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":289:22) +#loc291 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":287:8) +#loc292 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":285:14) +#loc293 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":297:30) +#loc294 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":297:42) +#loc295 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":297:16) +#loc296 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":297:51) +#loc297 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":298:30) +#loc298 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":298:42) +#loc299 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":298:16) +#loc300 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":298:51) +#loc301 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":299:57) +#loc302 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":299:62) +#loc303 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":299:51) +#loc304 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":300:30) +#loc305 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":300:42) +#loc306 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":300:16) +#loc307 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":300:51) +#loc308 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":301:51) +#loc309 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":302:59) +#loc310 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":302:64) +#loc311 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":302:51) +#loc312 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":303:30) +#loc313 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":303:42) +#loc314 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":303:16) +#loc315 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":303:51) +#loc316 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":304:51) +#loc317 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":305:51) +#loc318 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":306:59) +#loc319 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":306:64) +#loc320 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":306:51) +#loc321 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":308:44) +#loc322 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":308:29) +#loc323 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":309:29) +#loc324 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":310:29) +#loc325 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":311:29) +#loc326 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":312:52) +#loc327 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":312:30) +#loc328 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":313:30) +#loc329 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":314:54) +#loc330 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":314:30) +#loc331 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":315:30) +#loc332 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":316:30) +#loc333 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":317:54) +#loc334 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":317:30) +#loc335 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":319:30) +#loc336 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":319:20) +#loc337 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":320:30) +#loc338 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":320:20) +#loc339 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":321:30) +#loc340 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":321:20) +#loc341 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":322:30) +#loc342 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":322:20) +#loc343 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":323:30) +#loc344 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":323:20) +#loc345 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":324:30) +#loc346 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":324:20) +#loc347 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":325:30) +#loc348 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":325:20) +#loc349 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":326:30) +#loc350 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":326:20) +#loc351 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":327:30) +#loc352 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":327:20) +#loc353 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":328:30) +#loc354 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":328:20) +#loc355 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":328:4) +#loc361 = loc("i_tc1"(#loc12)) +#loc362 = loc("i_tc2"(#loc13)) +#loc363 = loc("i_tc3"(#loc14)) +#loc364 = loc("k"(#loc15)) +#loc365 = loc("k"(#loc16)) +#loc366 = loc("k"(#loc17)) +#loc367 = loc("k"(#loc18)) +#loc368 = loc("A"(#loc19)) +#loc369 = loc("A"(#loc20)) +#loc370 = loc("o_i"(#loc21)) +#loc371 = loc("m_tc0"(#loc22)) +#loc372 = loc("m_tc0"(#loc23)) +#loc373 = loc("m_tc1"(#loc24)) +#loc374 = loc("m_tc1"(#loc25)) +#loc375 = loc("m_tc2"(#loc26)) +#loc376 = loc("m_tc2"(#loc27)) +#loc377 = loc("m_tc3"(#loc28)) +#loc378 = loc("m_tc3"(#loc29)) +#loc379 = loc("p_b0"(#loc30)) +#loc380 = loc("p_b0"(#loc31)) +#loc381 = loc("p_b0"(#loc32)) +#loc382 = loc("p_b0"(#loc33)) +#loc383 = loc("p_b1"(#loc34)) +#loc384 = loc("p_b1"(#loc35)) +#loc385 = loc("p_b2"(#loc36)) +#loc386 = loc("p_b2"(#loc37)) +#loc387 = loc("p_b3"(#loc38)) +#loc388 = loc("p_b3"(#loc39)) +#loc389 = loc("b_b0"(#loc40)) +#loc390 = loc("b_b0"(#loc41)) +#loc391 = loc("b_b1"(#loc42)) +#loc392 = loc("b_b1"(#loc43)) +#loc393 = loc("b_b2"(#loc44)) +#loc394 = loc("b_b2"(#loc45)) +#loc395 = loc("b_b3"(#loc46)) +#loc396 = loc("b_b3"(#loc47)) +#loc397 = loc("p_g0"(#loc48)) +#loc398 = loc("p_g0"(#loc49)) +#loc399 = loc("p_g0"(#loc50)) +#loc400 = loc("p_g1"(#loc51)) +#loc401 = loc("p_g2"(#loc52)) +#loc402 = loc("p_g3"(#loc53)) +#loc403 = loc("b_g0"(#loc54)) +#loc404 = loc("b_g1"(#loc55)) +#loc405 = loc("b_g2"(#loc56)) +#loc406 = loc("b_g3"(#loc57)) +#loc407 = loc("o_k"(#loc58)) +#loc408 = loc("p_k0"(#loc59)) +#loc409 = loc("p_k0"(#loc60)) +#loc410 = loc("p_k0"(#loc61)) +#loc411 = loc("p_k0"(#loc62)) +#loc412 = loc("b_k0"(#loc63)) +#loc413 = loc("b_k0"(#loc64)) +#loc414 = loc("b_k0"(#loc66)) +#loc415 = loc("p_k0"(#loc67)) +#loc416 = loc("b_k0"(#loc68)) +#loc417 = loc("o_k"(#loc69)) +#loc418 = loc("b_A00"(#loc70)) +#loc419 = loc("b_A00"(#loc71)) +#loc420 = loc("b_A00"(#loc72)) +#loc421 = loc("o_k"(#loc73)) +#loc422 = loc("p_k1"(#loc75)) +#loc423 = loc("p_k1"(#loc76)) +#loc424 = loc("p_k1"(#loc77)) +#loc425 = loc("p_k1"(#loc78)) +#loc426 = loc("b_k1"(#loc79)) +#loc427 = loc("b_k1"(#loc80)) +#loc428 = loc("b_k1"(#loc81)) +#loc429 = loc("b_A11"(#loc82)) +#loc430 = loc("b_A11"(#loc83)) +#loc431 = loc("b_A10"(#loc84)) +#loc432 = loc("b_A22"(#loc86)) +#loc433 = loc("p_k2"(#loc87)) +#loc434 = loc("p_k2"(#loc88)) +#loc435 = loc("p_k2"(#loc89)) +#loc436 = loc("p_k2"(#loc90)) +#loc437 = loc("b_k2"(#loc91)) +#loc438 = loc("b_k2"(#loc92)) +#loc439 = loc("b_k2"(#loc93)) +#loc440 = loc("b_A22"(#loc94)) +#loc441 = loc("b_A22"(#loc95)) +#loc442 = loc("b_A20"(#loc96)) +#loc443 = loc("b_A21"(#loc97)) +#loc444 = loc("b_A33"(#loc99)) +#loc445 = loc("p_k3"(#loc100)) +#loc446 = loc("p_k3"(#loc101)) +#loc447 = loc("p_k3"(#loc102)) +#loc448 = loc("p_k3"(#loc103)) +#loc449 = loc("b_k3"(#loc104)) +#loc450 = loc("b_k3"(#loc105)) +#loc451 = loc("b_k3"(#loc106)) +#loc452 = loc("b_A33"(#loc107)) +#loc453 = loc("b_A33"(#loc108)) +#loc454 = loc("b_A30"(#loc109)) +#loc455 = loc("b_A31"(#loc110)) +#loc456 = loc("b_A32"(#loc111)) +#loc457 = loc("m_d"(#loc112)) +#loc458 = loc("m_d"(#loc113)) +#loc459 = loc("m_d"(#loc114)) +#loc460 = loc("m_I"(#loc115)) +#loc461 = loc("b_A00"(#loc116)) +#loc462 = loc("b_A00"(#loc117)) +#loc463 = loc("b_A00"(#loc118)) +#loc464 = loc("b_A00"(#loc119)) +#loc465 = loc("b_A00"(#loc120)) +#loc466 = loc("b_A00"(#loc121)) +#loc467 = loc("b_A00"(#loc122)) +#loc468 = loc("b_A00"(#loc124)) +#loc469 = loc("b_A00"(#loc125)) +#loc470 = loc("b_A00"(#loc126)) +#loc471 = loc("b_A11"(#loc127)) +#loc472 = loc("b_A11"(#loc128)) +#loc473 = loc("b_A11"(#loc129)) +#loc474 = loc("b_A11"(#loc130)) +#loc475 = loc("b_A11"(#loc131)) +#loc476 = loc("b_A11"(#loc132)) +#loc477 = loc("b_A11"(#loc133)) +#loc478 = loc("b_A11"(#loc134)) +#loc479 = loc("b_A11"(#loc135)) +#loc480 = loc("b_A11"(#loc136)) +#loc481 = loc("b_A22"(#loc137)) +#loc482 = loc("b_A22"(#loc138)) +#loc483 = loc("b_A22"(#loc139)) +#loc484 = loc("b_A22"(#loc140)) +#loc485 = loc("b_A22"(#loc141)) +#loc486 = loc("b_A22"(#loc142)) +#loc487 = loc("b_A22"(#loc143)) +#loc488 = loc("b_A22"(#loc144)) +#loc489 = loc("b_A22"(#loc145)) +#loc490 = loc("b_A22"(#loc146)) +#loc491 = loc("b_A33"(#loc147)) +#loc492 = loc("b_A33"(#loc148)) +#loc493 = loc("b_A33"(#loc149)) +#loc494 = loc("b_A33"(#loc150)) +#loc495 = loc("b_A33"(#loc151)) +#loc496 = loc("b_A33"(#loc152)) +#loc497 = loc("b_A33"(#loc153)) +#loc498 = loc("b_A33"(#loc154)) +#loc499 = loc("b_A33"(#loc155)) +#loc500 = loc("b_A33"(#loc156)) +#loc501 = loc("b_A10"(#loc157)) +#loc502 = loc("b_A10"(#loc158)) +#loc503 = loc("b_A10"(#loc159)) +#loc504 = loc("b_A10"(#loc160)) +#loc505 = loc("b_A10"(#loc161)) +#loc506 = loc("b_A20"(#loc162)) +#loc507 = loc("b_A20"(#loc163)) +#loc508 = loc("b_A20"(#loc164)) +#loc509 = loc("b_A20"(#loc165)) +#loc510 = loc("b_A20"(#loc166)) +#loc511 = loc("b_A21"(#loc167)) +#loc512 = loc("b_A21"(#loc168)) +#loc513 = loc("b_A21"(#loc169)) +#loc514 = loc("b_A21"(#loc170)) +#loc515 = loc("b_A21"(#loc171)) +#loc516 = loc("b_A30"(#loc172)) +#loc517 = loc("b_A30"(#loc173)) +#loc518 = loc("b_A30"(#loc174)) +#loc519 = loc("b_A30"(#loc175)) +#loc520 = loc("b_A30"(#loc176)) +#loc521 = loc("b_A31"(#loc177)) +#loc522 = loc("b_A31"(#loc178)) +#loc523 = loc("b_A31"(#loc179)) +#loc524 = loc("b_A31"(#loc180)) +#loc525 = loc("b_A31"(#loc181)) +#loc526 = loc("b_A32"(#loc182)) +#loc527 = loc("b_A32"(#loc183)) +#loc528 = loc("b_A32"(#loc184)) +#loc529 = loc("b_A32"(#loc185)) +#loc530 = loc("b_A32"(#loc186)) +#loc531 = loc("b_A00"(#loc187)) +#loc532 = loc("b_A00"(#loc188)) +#loc533 = loc("b_A11"(#loc189)) +#loc534 = loc("b_A11"(#loc190)) +#loc535 = loc("b_A22"(#loc191)) +#loc536 = loc("b_A22"(#loc192)) +#loc537 = loc("b_A33"(#loc193)) +#loc538 = loc("b_A33"(#loc194)) +#loc539 = loc("b_A10"(#loc195)) +#loc540 = loc("b_A20"(#loc196)) +#loc541 = loc("b_A21"(#loc197)) +#loc542 = loc("b_A30"(#loc198)) +#loc543 = loc("b_A31"(#loc199)) +#loc544 = loc("b_A32"(#loc200)) +#loc545 = loc("b_Ai00"(#loc201)) +#loc546 = loc("b_Ai11"(#loc202)) +#loc547 = loc("b_Ai22"(#loc203)) +#loc548 = loc("b_Ai33"(#loc204)) +#loc549 = loc("b_Ai00"(#loc207)) +#loc550 = loc("b_a00"(#loc208)) +#loc551 = loc("b_a00"(#loc209)) +#loc552 = loc("b_a00"(#loc210)) +#loc554 = loc("b_a00"(#loc214)) +#loc555 = loc("b_a00"(#loc215)) +#loc556 = loc("b_a00"(#loc216)) +#loc557 = loc("b_a00"(#loc217)) +#loc559 = loc("b_a00"(#loc219)) +#loc560 = loc("b_Ai00"(#loc220)) +#loc561 = loc("b_Ai11"(#loc224)) +#loc562 = loc("b_a11"(#loc225)) +#loc563 = loc("b_a11"(#loc226)) +#loc564 = loc("b_a11"(#loc227)) +#loc566 = loc("b_a11"(#loc229)) +#loc567 = loc("b_a11"(#loc230)) +#loc568 = loc("b_a11"(#loc231)) +#loc569 = loc("b_a11"(#loc232)) +#loc571 = loc("b_a11"(#loc234)) +#loc572 = loc("b_Ai11"(#loc235)) +#loc573 = loc("b_Ai22"(#loc239)) +#loc574 = loc("b_a22"(#loc240)) +#loc575 = loc("b_a22"(#loc241)) +#loc576 = loc("b_a22"(#loc242)) +#loc578 = loc("b_a22"(#loc244)) +#loc579 = loc("b_a22"(#loc245)) +#loc580 = loc("b_a22"(#loc246)) +#loc581 = loc("b_a22"(#loc247)) +#loc583 = loc("b_a22"(#loc249)) +#loc584 = loc("b_Ai22"(#loc250)) +#loc585 = loc("b_Ai33"(#loc254)) +#loc586 = loc("b_a33"(#loc255)) +#loc587 = loc("b_a33"(#loc256)) +#loc588 = loc("b_a33"(#loc257)) +#loc590 = loc("b_a33"(#loc259)) +#loc591 = loc("b_a33"(#loc260)) +#loc592 = loc("b_a33"(#loc261)) +#loc593 = loc("b_a33"(#loc262)) +#loc595 = loc("b_a33"(#loc264)) +#loc596 = loc("b_Ai33"(#loc265)) +#loc597 = loc("b_Ai00"(#loc267)) +#loc598 = loc("b_Ai11"(#loc268)) +#loc599 = loc("b_Ai22"(#loc269)) +#loc600 = loc("b_Ai33"(#loc270)) +#loc601 = loc("b_Ai10"(#loc271)) +#loc602 = loc("b_Ai10"(#loc272)) +#loc603 = loc("b_Ai10"(#loc273)) +#loc604 = loc("b_Ai21"(#loc274)) +#loc605 = loc("b_Ai21"(#loc275)) +#loc606 = loc("b_Ai21"(#loc276)) +#loc607 = loc("b_Ai32"(#loc277)) +#loc608 = loc("b_Ai32"(#loc278)) +#loc609 = loc("b_Ai32"(#loc279)) +#loc610 = loc("b_Ai20"(#loc280)) +#loc611 = loc("b_Ai20"(#loc281)) +#loc612 = loc("b_Ai20"(#loc282)) +#loc613 = loc("b_Ai20"(#loc283)) +#loc614 = loc("b_Ai31"(#loc284)) +#loc615 = loc("b_Ai31"(#loc285)) +#loc616 = loc("b_Ai31"(#loc286)) +#loc617 = loc("b_Ai31"(#loc287)) +#loc618 = loc("b_Ai30"(#loc288)) +#loc619 = loc("b_Ai30"(#loc289)) +#loc620 = loc("b_Ai30"(#loc290)) +#loc621 = loc("b_Ai30"(#loc291)) +#loc622 = loc("b_Ai30"(#loc292)) +#loc623 = loc("p_A00"(#loc293)) +#loc624 = loc("p_A00"(#loc294)) +#loc625 = loc("p_A00"(#loc295)) +#loc626 = loc("p_A00"(#loc296)) +#loc627 = loc("p_A10"(#loc297)) +#loc628 = loc("p_A10"(#loc298)) +#loc629 = loc("p_A10"(#loc299)) +#loc630 = loc("p_A10"(#loc300)) +#loc631 = loc("p_A11"(#loc301)) +#loc632 = loc("p_A11"(#loc302)) +#loc633 = loc("p_A11"(#loc303)) +#loc634 = loc("p_A20"(#loc304)) +#loc635 = loc("p_A20"(#loc305)) +#loc636 = loc("p_A20"(#loc306)) +#loc637 = loc("p_A20"(#loc307)) +#loc638 = loc("p_A21"(#loc308)) +#loc639 = loc("p_A22"(#loc309)) +#loc640 = loc("p_A22"(#loc310)) +#loc641 = loc("p_A22"(#loc311)) +#loc642 = loc("p_A30"(#loc312)) +#loc643 = loc("p_A30"(#loc313)) +#loc644 = loc("p_A30"(#loc314)) +#loc645 = loc("p_A30"(#loc315)) +#loc646 = loc("p_A31"(#loc316)) +#loc647 = loc("p_A32"(#loc317)) +#loc648 = loc("p_A33"(#loc318)) +#loc649 = loc("p_A33"(#loc319)) +#loc650 = loc("p_A33"(#loc320)) +#loc651 = loc("m_A0"(#loc321)) +#loc652 = loc("m_A0"(#loc322)) +#loc653 = loc("m_A1"(#loc323)) +#loc654 = loc("m_A2"(#loc324)) +#loc655 = loc("m_A3"(#loc325)) +#loc656 = loc("m_A11"(#loc326)) +#loc657 = loc("m_A11"(#loc327)) +#loc658 = loc("m_A21"(#loc328)) +#loc659 = loc("m_A22"(#loc329)) +#loc660 = loc("m_A22"(#loc330)) +#loc661 = loc("m_A31"(#loc331)) +#loc662 = loc("m_A32"(#loc332)) +#loc663 = loc("m_A33"(#loc333)) +#loc664 = loc("m_A33"(#loc334)) +#loc665 = loc(fused[#loc379, #loc380]) +#loc666 = loc(fused[#loc397, #loc398]) +#loc667 = loc("b_A11"(#loc418)) +#loc668 = loc("b_A11"(#loc430)) +#loc669 = loc("b_A10"(#loc431)) +#loc670 = loc("b_A33"(#loc432)) +#loc671 = loc("b_A22"(#loc441)) +#loc672 = loc("b_A20"(#loc442)) +#loc673 = loc("b_A21"(#loc443)) +#loc674 = loc("b_A30"(#loc444)) +#loc675 = loc("b_A33"(#loc453)) +#loc676 = loc("b_A30"(#loc454)) +#loc677 = loc("b_A31"(#loc455)) +#loc678 = loc("b_A32"(#loc456)) +#loc679 = loc(callsite(#loc123 at #loc468)) +#loc680 = loc(callsite(#loc123 at #loc478)) +#loc681 = loc(callsite(#loc123 at #loc488)) +#loc682 = loc(callsite(#loc123 at #loc498)) +#loc683 = loc(callsite(#loc123 at #loc503)) +#loc684 = loc(callsite(#loc123 at #loc508)) +#loc685 = loc(callsite(#loc123 at #loc513)) +#loc686 = loc(callsite(#loc123 at #loc518)) +#loc687 = loc(callsite(#loc123 at #loc523)) +#loc688 = loc(callsite(#loc123 at #loc528)) +#loc689 = loc(callsite(#loc211 at #loc553)) +#loc691 = loc(callsite(#loc211 at #loc558)) +#loc693 = loc(callsite(#loc211 at #loc565)) +#loc695 = loc(callsite(#loc211 at #loc570)) +#loc697 = loc(callsite(#loc211 at #loc577)) +#loc699 = loc(callsite(#loc211 at #loc582)) +#loc701 = loc(callsite(#loc211 at #loc589)) +#loc703 = loc(callsite(#loc211 at #loc594)) +#loc705 = loc("b_A22"(#loc667)) +#loc706 = loc("b_A20"(#loc670)) +#loc707 = loc("b_A31"(#loc674)) +#loc708 = loc(callsite(#loc213 at #loc689)) +#loc709 = loc(callsite(#loc213 at #loc691)) +#loc710 = loc(callsite(#loc213 at #loc693)) +#loc711 = loc(callsite(#loc213 at #loc695)) +#loc712 = loc(callsite(#loc213 at #loc697)) +#loc713 = loc(callsite(#loc213 at #loc699)) +#loc714 = loc(callsite(#loc213 at #loc701)) +#loc715 = loc(callsite(#loc213 at #loc703)) +#loc716 = loc("b_A33"(#loc705)) +#loc717 = loc("b_A21"(#loc706)) +#loc718 = loc("b_A32"(#loc707)) +#loc719 = loc("b_A10"(#loc716)) +#loc720 = loc("b_A30"(#loc717)) +#loc721 = loc("b_A20"(#loc719)) +#loc722 = loc("b_A31"(#loc720)) +#loc723 = loc("b_A21"(#loc721)) +#loc724 = loc("b_A32"(#loc722)) +#loc725 = loc("b_A30"(#loc723)) +#loc726 = loc("b_A31"(#loc725)) +#loc727 = loc("b_A32"(#loc726)) diff --git a/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.ttir b/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..2e1d46d0c4277b493b19c13f2d653c6df4e97f5a --- /dev/null +++ b/triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.ttir @@ -0,0 +1,1234 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":40:0) +#loc1 = loc(unknown) +#loc213 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":228:66) +#loc219 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":230:56) +#loc229 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":233:66) +#loc234 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":235:56) +#loc244 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":238:66) +#loc249 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":240:56) +#loc259 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":243:66) +#loc264 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":245:56) +#loc357 = loc("k"(#loc)) +#loc358 = loc("g"(#loc)) +#loc359 = loc("beta"(#loc)) +#loc360 = loc("A"(#loc)) +#loc361 = loc("T"(#loc)) +#loc554 = loc("b_a00"(#loc213)) +#loc559 = loc("b_a00"(#loc219)) +#loc566 = loc("b_a11"(#loc229)) +#loc571 = loc("b_a11"(#loc234)) +#loc578 = loc("b_a22"(#loc244)) +#loc583 = loc("b_a22"(#loc249)) +#loc590 = loc("b_a33"(#loc259)) +#loc595 = loc("b_a33"(#loc264)) +#loc691 = loc(callsite(#loc1 at #loc554)) +#loc693 = loc(callsite(#loc1 at #loc559)) +#loc695 = loc(callsite(#loc1 at #loc566)) +#loc697 = loc(callsite(#loc1 at #loc571)) +#loc699 = loc(callsite(#loc1 at #loc578)) +#loc701 = loc(callsite(#loc1 at #loc583)) +#loc703 = loc(callsite(#loc1 at #loc590)) +#loc705 = loc(callsite(#loc1 at #loc595)) +module { + tt.func public @chunk_gated_delta_rule_fwd_kkt_solve_kernel(%k: !tt.ptr {tt.divisibility = 16 : i32} loc("k"(#loc)), %g: !tt.ptr {tt.divisibility = 16 : i32} loc("g"(#loc)), %beta: !tt.ptr {tt.divisibility = 16 : i32} loc("beta"(#loc)), %A: !tt.ptr {tt.divisibility = 16 : i32} loc("A"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %c1_i64 = arith.constant 1 : i64 loc(#loc1) + %c2_i64 = arith.constant 2 : i64 loc(#loc1) + %cst = arith.constant dense<0.000000e+00> : tensor<16x64xbf16> loc(#loc1) + %c1_i32 = arith.constant 1 : i32 loc(#loc1) + %c2_i32 = arith.constant 2 : i32 loc(#loc1) + %c0_i32 = arith.constant 0 : i32 loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<16xbf16> loc(#loc1) + %cst_1 = arith.constant dense<64> : tensor<1x16xi32> loc(#loc1) + %cst_2 = arith.constant dense<48> : tensor<16xi32> loc(#loc1) + %cst_3 = arith.constant dense<32> : tensor<16xi32> loc(#loc1) + %cst_4 = arith.constant dense<16> : tensor<16xi32> loc(#loc1) + %cst_5 = arith.constant dense<2048> : tensor<16x1xi64> loc(#loc1) + %cst_6 = arith.constant dense<0.000000e+00> : tensor<16x16xf32> loc(#loc1) + %cst_7 = arith.constant dense<128> : tensor<1x64xi32> loc(#loc1) + %cst_8 = arith.constant dense<4096> : tensor<16x1xi64> loc(#loc1) + %cst_9 = arith.constant dense<0.000000e+00> : tensor<16xf32> loc(#loc1) + %cst_10 = arith.constant dense<32> : tensor<16xi64> loc(#loc1) + %c128_i64 = arith.constant 128 : i64 loc(#loc1) + %c48_i64 = arith.constant 48 : i64 loc(#loc1) + %c16_i64 = arith.constant 16 : i64 loc(#loc1) + %c64_i64 = arith.constant 64 : i64 loc(#loc1) + %c64_i32 = arith.constant 64 : i32 loc(#loc1) + %c32_i64 = arith.constant 32 : i64 loc(#loc1) + %0 = tt.get_program_id x : i32 loc(#loc2) + %1 = arith.extsi %0 : i32 to i64 loc(#loc3) + %2 = tt.get_program_id y : i32 loc(#loc4) + %3 = arith.extsi %2 : i32 to i64 loc(#loc5) + %4 = arith.divsi %3, %c32_i64 : i64 loc(#loc6) + %5 = arith.remsi %3, %c32_i64 : i64 loc(#loc7) + %6 = arith.extsi %T : i32 to i64 loc(#loc8) + %7 = arith.muli %4, %6 : i64 loc(#loc8) + %8 = arith.muli %1, %c64_i64 : i64 loc(#loc9) + %9 = arith.cmpi sge, %8, %6 : i64 loc(#loc10) + cf.cond_br %9, ^bb1, ^bb2 loc(#loc10) + ^bb1: // pred: ^bb0 + tt.return loc(#loc11) + ^bb2: // pred: ^bb0 + %i_tc1 = arith.addi %8, %c16_i64 : i64 loc(#loc362) + %i_tc2 = arith.addi %8, %c32_i64 : i64 loc(#loc363) + %i_tc3 = arith.addi %8, %c48_i64 : i64 loc(#loc364) + %k_11 = arith.muli %7, %c32_i64 : i64 loc(#loc365) + %k_12 = arith.addi %k_11, %5 : i64 loc(#loc366) + %k_13 = arith.muli %k_12, %c128_i64 : i64 loc(#loc367) + %k_14 = tt.addptr %k, %k_13 : !tt.ptr, i64 loc(#loc368) + %A_15 = arith.muli %k_12, %c64_i64 : i64 loc(#loc369) + %A_16 = tt.addptr %A, %A_15 : !tt.ptr, i64 loc(#loc370) + %o_i = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32> loc(#loc371) + %m_tc0 = arith.extsi %o_i : tensor<16xi32> to tensor<16xi64> loc(#loc372) + %m_tc0_17 = tt.splat %8 : i64 -> tensor<16xi64> loc(#loc372) + %m_tc0_18 = arith.addi %m_tc0_17, %m_tc0 : tensor<16xi64> loc(#loc372) + %m_tc0_19 = tt.splat %6 : i64 -> tensor<16xi64> loc(#loc373) + %m_tc0_20 = arith.cmpi slt, %m_tc0_18, %m_tc0_19 : tensor<16xi64> loc(#loc373) + %m_tc1 = tt.splat %i_tc1 : i64 -> tensor<16xi64> loc(#loc374) + %m_tc1_21 = arith.addi %m_tc1, %m_tc0 : tensor<16xi64> loc(#loc374) + %m_tc1_22 = arith.cmpi slt, %m_tc1_21, %m_tc0_19 : tensor<16xi64> loc(#loc375) + %m_tc2 = tt.splat %i_tc2 : i64 -> tensor<16xi64> loc(#loc376) + %m_tc2_23 = arith.addi %m_tc2, %m_tc0 : tensor<16xi64> loc(#loc376) + %m_tc2_24 = arith.cmpi slt, %m_tc2_23, %m_tc0_19 : tensor<16xi64> loc(#loc377) + %m_tc3 = tt.splat %i_tc3 : i64 -> tensor<16xi64> loc(#loc378) + %m_tc3_25 = arith.addi %m_tc3, %m_tc0 : tensor<16xi64> loc(#loc378) + %m_tc3_26 = arith.cmpi slt, %m_tc3_25, %m_tc0_19 : tensor<16xi64> loc(#loc379) + %p_b0 = tt.addptr %beta, %k_12 : !tt.ptr, i64 loc(#loc666) + %p_b0_27 = arith.muli %m_tc0_18, %cst_10 : tensor<16xi64> loc(#loc382) + %p_b0_28 = tt.splat %p_b0 : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc383) + %p_b0_29 = tt.addptr %p_b0_28, %p_b0_27 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc383) + %p_b1 = arith.muli %m_tc1_21, %cst_10 : tensor<16xi64> loc(#loc384) + %p_b1_30 = tt.addptr %p_b0_28, %p_b1 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc385) + %p_b2 = arith.muli %m_tc2_23, %cst_10 : tensor<16xi64> loc(#loc386) + %p_b2_31 = tt.addptr %p_b0_28, %p_b2 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc387) + %p_b3 = arith.muli %m_tc3_25, %cst_10 : tensor<16xi64> loc(#loc388) + %p_b3_32 = tt.addptr %p_b0_28, %p_b3 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc389) + %b_b0 = tt.load %p_b0_29, %m_tc0_20, %cst_0 : tensor<16x!tt.ptr> loc(#loc390) + %b_b0_33 = arith.extf %b_b0 : tensor<16xbf16> to tensor<16xf32> loc(#loc391) + %b_b1 = tt.load %p_b1_30, %m_tc1_22, %cst_0 : tensor<16x!tt.ptr> loc(#loc392) + %b_b1_34 = arith.extf %b_b1 : tensor<16xbf16> to tensor<16xf32> loc(#loc393) + %b_b2 = tt.load %p_b2_31, %m_tc2_24, %cst_0 : tensor<16x!tt.ptr> loc(#loc394) + %b_b2_35 = arith.extf %b_b2 : tensor<16xbf16> to tensor<16xf32> loc(#loc395) + %b_b3 = tt.load %p_b3_32, %m_tc3_26, %cst_0 : tensor<16x!tt.ptr> loc(#loc396) + %b_b3_36 = arith.extf %b_b3 : tensor<16xbf16> to tensor<16xf32> loc(#loc397) + %p_g0 = tt.addptr %g, %k_12 : !tt.ptr, i64 loc(#loc667) + %p_g0_37 = tt.splat %p_g0 : !tt.ptr -> tensor<16x!tt.ptr> loc(#loc400) + %p_g0_38 = tt.addptr %p_g0_37, %p_b0_27 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc400) + %p_g1 = tt.addptr %p_g0_37, %p_b1 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc401) + %p_g2 = tt.addptr %p_g0_37, %p_b2 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc402) + %p_g3 = tt.addptr %p_g0_37, %p_b3 : tensor<16x!tt.ptr>, tensor<16xi64> loc(#loc403) + %b_g0 = tt.load %p_g0_38, %m_tc0_20, %cst_9 : tensor<16x!tt.ptr> loc(#loc404) + %b_g1 = tt.load %p_g1, %m_tc1_22, %cst_9 : tensor<16x!tt.ptr> loc(#loc405) + %b_g2 = tt.load %p_g2, %m_tc2_24, %cst_9 : tensor<16x!tt.ptr> loc(#loc406) + %b_g3 = tt.load %p_g3, %m_tc3_26, %cst_9 : tensor<16x!tt.ptr> loc(#loc407) + %b_A32:10 = scf.for %i_k = %c0_i32 to %c2_i32 step %c1_i32 iter_args(%b_A00_197 = %cst_6, %b_A11_198 = %cst_6, %b_A22_199 = %cst_6, %b_A33_200 = %cst_6, %b_A10_201 = %cst_6, %b_A20_202 = %cst_6, %b_A21_203 = %cst_6, %b_A30_204 = %cst_6, %b_A31_205 = %cst_6, %b_A32_206 = %cst_6) -> (tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>) : i32 { + %o_k = arith.muli %i_k, %c64_i32 : i32 loc(#loc409) + %o_k_207 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc410) + %o_k_208 = tt.splat %o_k : i32 -> tensor<64xi32> loc(#loc411) + %o_k_209 = arith.addi %o_k_208, %o_k_207 : tensor<64xi32> loc(#loc411) + %p_k0 = tt.expand_dims %m_tc0_18 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc412) + %p_k0_210 = arith.muli %p_k0, %cst_8 : tensor<16x1xi64> loc(#loc413) + %p_k0_211 = tt.splat %k_14 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc414) + %p_k0_212 = tt.addptr %p_k0_211, %p_k0_210 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc414) + %p_k0_213 = tt.expand_dims %o_k_209 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc415) + %p_k0_214 = tt.broadcast %p_k0_212 : tensor<16x1x!tt.ptr> -> tensor<16x64x!tt.ptr> loc(#loc416) + %p_k0_215 = tt.broadcast %p_k0_213 : tensor<1x64xi32> -> tensor<16x64xi32> loc(#loc416) + %p_k0_216 = tt.addptr %p_k0_214, %p_k0_215 : tensor<16x64x!tt.ptr>, tensor<16x64xi32> loc(#loc416) + %b_k0 = tt.expand_dims %m_tc0_20 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc417) + %b_k0_217 = arith.cmpi slt, %p_k0_213, %cst_7 : tensor<1x64xi32> loc(#loc418) + %b_k0_218 = tt.broadcast %b_k0 : tensor<16x1xi1> -> tensor<16x64xi1> loc(#loc419) + %b_k0_219 = tt.broadcast %b_k0_217 : tensor<1x64xi1> -> tensor<16x64xi1> loc(#loc419) + %b_k0_220 = arith.andi %b_k0_218, %b_k0_219 : tensor<16x64xi1> loc(#loc419) + %b_k0_221 = tt.load %p_k0_216, %b_k0_220, %cst : tensor<16x64x!tt.ptr> loc(#loc420) + %b_A00_222 = tt.trans %b_k0_221 {order = array} : tensor<16x64xbf16> -> tensor<64x16xbf16> loc(#loc421) + %b_A00_223 = tt.dot %b_k0_221, %b_A00_222, %b_A00_197, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc422) + %28 = arith.cmpi slt, %i_tc1, %6 : i64 loc(#loc73) + %29:9 = scf.if %28 -> (tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>) { + %p_k1 = tt.expand_dims %m_tc1_21 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc423) + %p_k1_224 = arith.muli %p_k1, %cst_8 : tensor<16x1xi64> loc(#loc424) + %p_k1_225 = tt.addptr %p_k0_211, %p_k1_224 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc425) + %p_k1_226 = tt.broadcast %p_k1_225 : tensor<16x1x!tt.ptr> -> tensor<16x64x!tt.ptr> loc(#loc426) + %p_k1_227 = tt.addptr %p_k1_226, %p_k0_215 : tensor<16x64x!tt.ptr>, tensor<16x64xi32> loc(#loc426) + %b_k1 = tt.expand_dims %m_tc1_22 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc427) + %b_k1_228 = tt.broadcast %b_k1 : tensor<16x1xi1> -> tensor<16x64xi1> loc(#loc428) + %b_k1_229 = arith.andi %b_k1_228, %b_k0_219 : tensor<16x64xi1> loc(#loc428) + %b_k1_230 = tt.load %p_k1_227, %b_k1_229, %cst : tensor<16x64x!tt.ptr> loc(#loc429) + %b_A11_231 = tt.trans %b_k1_230 {order = array} : tensor<16x64xbf16> -> tensor<64x16xbf16> loc(#loc430) + %b_A11_232 = tt.dot %b_k1_230, %b_A11_231, %b_A11_198, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc669) + %b_A10_233 = tt.dot %b_k1_230, %b_A00_222, %b_A10_201, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc670) + %30 = arith.cmpi slt, %i_tc2, %6 : i64 loc(#loc85) + %b_A32_234:7 = scf.if %30 -> (tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>) { + %p_k2 = tt.expand_dims %m_tc2_23 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc434) + %p_k2_235 = arith.muli %p_k2, %cst_8 : tensor<16x1xi64> loc(#loc435) + %p_k2_236 = tt.addptr %p_k0_211, %p_k2_235 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc436) + %p_k2_237 = tt.broadcast %p_k2_236 : tensor<16x1x!tt.ptr> -> tensor<16x64x!tt.ptr> loc(#loc437) + %p_k2_238 = tt.addptr %p_k2_237, %p_k0_215 : tensor<16x64x!tt.ptr>, tensor<16x64xi32> loc(#loc437) + %b_k2 = tt.expand_dims %m_tc2_24 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc438) + %b_k2_239 = tt.broadcast %b_k2 : tensor<16x1xi1> -> tensor<16x64xi1> loc(#loc439) + %b_k2_240 = arith.andi %b_k2_239, %b_k0_219 : tensor<16x64xi1> loc(#loc439) + %b_k2_241 = tt.load %p_k2_238, %b_k2_240, %cst : tensor<16x64x!tt.ptr> loc(#loc440) + %b_A22_242 = tt.trans %b_k2_241 {order = array} : tensor<16x64xbf16> -> tensor<64x16xbf16> loc(#loc441) + %b_A22_243 = tt.dot %b_k2_241, %b_A22_242, %b_A22_199, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc672) + %b_A20_244 = tt.dot %b_k2_241, %b_A00_222, %b_A20_202, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc673) + %b_A21_245 = tt.dot %b_k2_241, %b_A11_231, %b_A21_203, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc674) + %31 = arith.cmpi slt, %i_tc3, %6 : i64 loc(#loc98) + %b_A32_246:4 = scf.if %31 -> (tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>) { + %p_k3 = tt.expand_dims %m_tc3_25 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc446) + %p_k3_247 = arith.muli %p_k3, %cst_8 : tensor<16x1xi64> loc(#loc447) + %p_k3_248 = tt.addptr %p_k0_211, %p_k3_247 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc448) + %p_k3_249 = tt.broadcast %p_k3_248 : tensor<16x1x!tt.ptr> -> tensor<16x64x!tt.ptr> loc(#loc449) + %p_k3_250 = tt.addptr %p_k3_249, %p_k0_215 : tensor<16x64x!tt.ptr>, tensor<16x64xi32> loc(#loc449) + %b_k3 = tt.expand_dims %m_tc3_26 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc450) + %b_k3_251 = tt.broadcast %b_k3 : tensor<16x1xi1> -> tensor<16x64xi1> loc(#loc451) + %b_k3_252 = arith.andi %b_k3_251, %b_k0_219 : tensor<16x64xi1> loc(#loc451) + %b_k3_253 = tt.load %p_k3_250, %b_k3_252, %cst : tensor<16x64x!tt.ptr> loc(#loc452) + %b_A33_254 = tt.trans %b_k3_253 {order = array} : tensor<16x64xbf16> -> tensor<64x16xbf16> loc(#loc453) + %b_A33_255 = tt.dot %b_k3_253, %b_A33_254, %b_A33_200, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc676) + %b_A30_256 = tt.dot %b_k3_253, %b_A00_222, %b_A30_204, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc677) + %b_A31_257 = tt.dot %b_k3_253, %b_A11_231, %b_A31_205, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc678) + %b_A32_258 = tt.dot %b_k3_253, %b_A22_242, %b_A32_206, inputPrecision = tf32 : tensor<16x64xbf16> * tensor<64x16xbf16> -> tensor<16x16xf32> loc(#loc679) + scf.yield %b_A33_255, %b_A30_256, %b_A31_257, %b_A32_258 : tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32> loc(#loc679) + } else { + scf.yield %b_A33_200, %b_A30_204, %b_A31_205, %b_A32_206 : tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32> loc(#loc99) + } loc(#loc719) + scf.yield %b_A22_243, %b_A32_246#0, %b_A20_244, %b_A21_245, %b_A32_246#1, %b_A32_246#2, %b_A32_246#3 : tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32> loc(#loc719) + } else { + scf.yield %b_A22_199, %b_A33_200, %b_A20_202, %b_A21_203, %b_A30_204, %b_A31_205, %b_A32_206 : tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32> loc(#loc86) + } loc(#loc725) + scf.yield %b_A11_232, %b_A32_234#0, %b_A32_234#1, %b_A10_233, %b_A32_234#2, %b_A32_234#3, %b_A32_234#4, %b_A32_234#5, %b_A32_234#6 : tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32> loc(#loc725) + } else { + scf.yield %b_A11_198, %b_A22_199, %b_A33_200, %b_A10_201, %b_A20_202, %b_A21_203, %b_A30_204, %b_A31_205, %b_A32_206 : tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32> loc(#loc74) + } loc(#loc74) + scf.yield %b_A00_223, %29#0, %29#1, %29#2, %29#3, %29#4, %29#5, %29#6, %29#7, %29#8 : tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32>, tensor<16x16xf32> loc(#loc112) + } loc(#loc728) + %m_d = tt.expand_dims %o_i {axis = 1 : i32} : tensor<16xi32> -> tensor<16x1xi32> loc(#loc458) + %m_d_39 = tt.expand_dims %o_i {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc459) + %m_d_40 = tt.broadcast %m_d : tensor<16x1xi32> -> tensor<16x16xi32> loc(#loc460) + %m_d_41 = tt.broadcast %m_d_39 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc460) + %m_d_42 = arith.cmpi sgt, %m_d_40, %m_d_41 : tensor<16x16xi32> loc(#loc460) + %m_I = arith.cmpi eq, %m_d_40, %m_d_41 : tensor<16x16xi32> loc(#loc461) + %b_A00 = tt.expand_dims %m_tc0_20 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc462) + %b_A00_43 = tt.broadcast %b_A00 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc463) + %b_A00_44 = arith.andi %m_d_42, %b_A00_43 : tensor<16x16xi1> loc(#loc463) + %b_A00_45 = tt.expand_dims %m_tc0_20 {axis = 0 : i32} : tensor<16xi1> -> tensor<1x16xi1> loc(#loc464) + %b_A00_46 = tt.broadcast %b_A00_45 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc465) + %b_A00_47 = arith.andi %b_A00_44, %b_A00_46 : tensor<16x16xi1> loc(#loc465) + %b_A00_48 = tt.expand_dims %b_g0 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc466) + %b_A00_49 = tt.expand_dims %b_g0 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc467) + %b_A00_50 = tt.broadcast %b_A00_48 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc468) + %b_A00_51 = tt.broadcast %b_A00_49 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc468) + %b_A00_52 = arith.subf %b_A00_50, %b_A00_51 : tensor<16x16xf32> loc(#loc468) + %b_A00_53 = math.exp2 %b_A00_52 : tensor<16x16xf32> loc(#loc680) + %b_A00_54 = arith.select %b_A00_47, %b_A00_53, %cst_6 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc470) + %b_A00_55 = arith.mulf %b_A32#0, %b_A00_54 : tensor<16x16xf32> loc(#loc471) + %b_A11 = tt.expand_dims %m_tc1_22 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc472) + %b_A11_56 = tt.broadcast %b_A11 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc473) + %b_A11_57 = arith.andi %m_d_42, %b_A11_56 : tensor<16x16xi1> loc(#loc473) + %b_A11_58 = tt.expand_dims %m_tc1_22 {axis = 0 : i32} : tensor<16xi1> -> tensor<1x16xi1> loc(#loc474) + %b_A11_59 = tt.broadcast %b_A11_58 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc475) + %b_A11_60 = arith.andi %b_A11_57, %b_A11_59 : tensor<16x16xi1> loc(#loc475) + %b_A11_61 = tt.expand_dims %b_g1 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc476) + %b_A11_62 = tt.expand_dims %b_g1 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc477) + %b_A11_63 = tt.broadcast %b_A11_61 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc478) + %b_A11_64 = tt.broadcast %b_A11_62 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc478) + %b_A11_65 = arith.subf %b_A11_63, %b_A11_64 : tensor<16x16xf32> loc(#loc478) + %b_A11_66 = math.exp2 %b_A11_65 : tensor<16x16xf32> loc(#loc681) + %b_A11_67 = arith.select %b_A11_60, %b_A11_66, %cst_6 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc480) + %b_A11_68 = arith.mulf %b_A32#1, %b_A11_67 : tensor<16x16xf32> loc(#loc481) + %b_A22 = tt.expand_dims %m_tc2_24 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc482) + %b_A22_69 = tt.broadcast %b_A22 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc483) + %b_A22_70 = arith.andi %m_d_42, %b_A22_69 : tensor<16x16xi1> loc(#loc483) + %b_A22_71 = tt.expand_dims %m_tc2_24 {axis = 0 : i32} : tensor<16xi1> -> tensor<1x16xi1> loc(#loc484) + %b_A22_72 = tt.broadcast %b_A22_71 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc485) + %b_A22_73 = arith.andi %b_A22_70, %b_A22_72 : tensor<16x16xi1> loc(#loc485) + %b_A22_74 = tt.expand_dims %b_g2 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc486) + %b_A22_75 = tt.expand_dims %b_g2 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc487) + %b_A22_76 = tt.broadcast %b_A22_74 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc488) + %b_A22_77 = tt.broadcast %b_A22_75 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc488) + %b_A22_78 = arith.subf %b_A22_76, %b_A22_77 : tensor<16x16xf32> loc(#loc488) + %b_A22_79 = math.exp2 %b_A22_78 : tensor<16x16xf32> loc(#loc682) + %b_A22_80 = arith.select %b_A22_73, %b_A22_79, %cst_6 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc490) + %b_A22_81 = arith.mulf %b_A32#2, %b_A22_80 : tensor<16x16xf32> loc(#loc491) + %b_A33 = tt.expand_dims %m_tc3_26 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc492) + %b_A33_82 = tt.broadcast %b_A33 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc493) + %b_A33_83 = arith.andi %m_d_42, %b_A33_82 : tensor<16x16xi1> loc(#loc493) + %b_A33_84 = tt.expand_dims %m_tc3_26 {axis = 0 : i32} : tensor<16xi1> -> tensor<1x16xi1> loc(#loc494) + %b_A33_85 = tt.broadcast %b_A33_84 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc495) + %b_A33_86 = arith.andi %b_A33_83, %b_A33_85 : tensor<16x16xi1> loc(#loc495) + %b_A33_87 = tt.expand_dims %b_g3 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc496) + %b_A33_88 = tt.expand_dims %b_g3 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc497) + %b_A33_89 = tt.broadcast %b_A33_87 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc498) + %b_A33_90 = tt.broadcast %b_A33_88 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc498) + %b_A33_91 = arith.subf %b_A33_89, %b_A33_90 : tensor<16x16xf32> loc(#loc498) + %b_A33_92 = math.exp2 %b_A33_91 : tensor<16x16xf32> loc(#loc683) + %b_A33_93 = arith.select %b_A33_86, %b_A33_92, %cst_6 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc500) + %b_A33_94 = arith.mulf %b_A32#3, %b_A33_93 : tensor<16x16xf32> loc(#loc501) + %b_A10 = arith.andi %b_A11_56, %b_A00_46 : tensor<16x16xi1> loc(#loc502) + %b_A10_95 = arith.subf %b_A11_63, %b_A00_51 : tensor<16x16xf32> loc(#loc503) + %b_A10_96 = math.exp2 %b_A10_95 : tensor<16x16xf32> loc(#loc684) + %b_A10_97 = arith.select %b_A10, %b_A10_96, %cst_6 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc505) + %b_A10_98 = arith.mulf %b_A32#4, %b_A10_97 : tensor<16x16xf32> loc(#loc506) + %b_A20 = arith.andi %b_A22_69, %b_A00_46 : tensor<16x16xi1> loc(#loc507) + %b_A20_99 = arith.subf %b_A22_76, %b_A00_51 : tensor<16x16xf32> loc(#loc508) + %b_A20_100 = math.exp2 %b_A20_99 : tensor<16x16xf32> loc(#loc685) + %b_A20_101 = arith.select %b_A20, %b_A20_100, %cst_6 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc510) + %b_A20_102 = arith.mulf %b_A32#5, %b_A20_101 : tensor<16x16xf32> loc(#loc511) + %b_A21 = arith.andi %b_A22_69, %b_A11_59 : tensor<16x16xi1> loc(#loc512) + %b_A21_103 = arith.subf %b_A22_76, %b_A11_64 : tensor<16x16xf32> loc(#loc513) + %b_A21_104 = math.exp2 %b_A21_103 : tensor<16x16xf32> loc(#loc686) + %b_A21_105 = arith.select %b_A21, %b_A21_104, %cst_6 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc515) + %b_A21_106 = arith.mulf %b_A32#6, %b_A21_105 : tensor<16x16xf32> loc(#loc516) + %b_A30 = arith.andi %b_A33_82, %b_A00_46 : tensor<16x16xi1> loc(#loc517) + %b_A30_107 = arith.subf %b_A33_89, %b_A00_51 : tensor<16x16xf32> loc(#loc518) + %b_A30_108 = math.exp2 %b_A30_107 : tensor<16x16xf32> loc(#loc687) + %b_A30_109 = arith.select %b_A30, %b_A30_108, %cst_6 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc520) + %b_A30_110 = arith.mulf %b_A32#7, %b_A30_109 : tensor<16x16xf32> loc(#loc521) + %b_A31 = arith.andi %b_A33_82, %b_A11_59 : tensor<16x16xi1> loc(#loc522) + %b_A31_111 = arith.subf %b_A33_89, %b_A11_64 : tensor<16x16xf32> loc(#loc523) + %b_A31_112 = math.exp2 %b_A31_111 : tensor<16x16xf32> loc(#loc688) + %b_A31_113 = arith.select %b_A31, %b_A31_112, %cst_6 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc525) + %b_A31_114 = arith.mulf %b_A32#8, %b_A31_113 : tensor<16x16xf32> loc(#loc526) + %b_A32_115 = arith.andi %b_A33_82, %b_A22_72 : tensor<16x16xi1> loc(#loc527) + %b_A32_116 = arith.subf %b_A33_89, %b_A22_77 : tensor<16x16xf32> loc(#loc528) + %b_A32_117 = math.exp2 %b_A32_116 : tensor<16x16xf32> loc(#loc689) + %b_A32_118 = arith.select %b_A32_115, %b_A32_117, %cst_6 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc530) + %b_A32_119 = arith.mulf %b_A32#9, %b_A32_118 : tensor<16x16xf32> loc(#loc531) + %b_A00_120 = tt.expand_dims %b_b0_33 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc532) + %b_A00_121 = tt.broadcast %b_A00_120 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc533) + %b_A00_122 = arith.mulf %b_A00_55, %b_A00_121 : tensor<16x16xf32> loc(#loc533) + %b_A11_123 = tt.expand_dims %b_b1_34 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc534) + %b_A11_124 = tt.broadcast %b_A11_123 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc535) + %b_A11_125 = arith.mulf %b_A11_68, %b_A11_124 : tensor<16x16xf32> loc(#loc535) + %b_A22_126 = tt.expand_dims %b_b2_35 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc536) + %b_A22_127 = tt.broadcast %b_A22_126 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc537) + %b_A22_128 = arith.mulf %b_A22_81, %b_A22_127 : tensor<16x16xf32> loc(#loc537) + %b_A33_129 = tt.expand_dims %b_b3_36 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc538) + %b_A33_130 = tt.broadcast %b_A33_129 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc539) + %b_A33_131 = arith.mulf %b_A33_94, %b_A33_130 : tensor<16x16xf32> loc(#loc539) + %b_A10_132 = arith.mulf %b_A10_98, %b_A11_124 : tensor<16x16xf32> loc(#loc540) + %b_A20_133 = arith.mulf %b_A20_102, %b_A22_127 : tensor<16x16xf32> loc(#loc541) + %b_A21_134 = arith.mulf %b_A21_106, %b_A22_127 : tensor<16x16xf32> loc(#loc542) + %b_A30_135 = arith.mulf %b_A30_110, %b_A33_130 : tensor<16x16xf32> loc(#loc543) + %b_A31_136 = arith.mulf %b_A31_114, %b_A33_130 : tensor<16x16xf32> loc(#loc544) + %b_A32_137 = arith.mulf %b_A32_119, %b_A33_130 : tensor<16x16xf32> loc(#loc545) + %b_Ai00 = arith.subf %cst_6, %b_A00_122 : tensor<16x16xf32> loc(#loc546) + %b_Ai11 = arith.subf %cst_6, %b_A11_125 : tensor<16x16xf32> loc(#loc547) + %b_Ai22 = arith.subf %cst_6, %b_A22_128 : tensor<16x16xf32> loc(#loc548) + %b_Ai33 = arith.subf %cst_6, %b_A33_131 : tensor<16x16xf32> loc(#loc549) + %10 = arith.subi %6, %8 : i64 loc(#loc206) + %11 = arith.minsi %10, %c16_i64 : i64 loc(#loc207) + %b_Ai00_138 = scf.for %i = %c2_i64 to %11 step %c1_i64 iter_args(%b_Ai00_197 = %b_Ai00) -> (tensor<16x16xf32>) : i64 { + %b_a00 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc551) + %b_a00_198 = arith.cmpi eq, %m_tc0, %b_a00 : tensor<16xi64> loc(#loc551) + %b_a00_199 = tt.expand_dims %b_a00_198 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc552) + %b_a00_200 = tt.broadcast %b_a00_199 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc553) + %b_a00_201 = arith.select %b_a00_200, %b_Ai00, %cst_6 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc553) + %b_a00_202 = "tt.reduce"(%b_a00_201) <{axis = 0 : i32}> ({ + ^bb0(%b_a00_213: f32 loc(callsite(#loc1 at #loc554)), %b_a00_214: f32 loc(callsite(#loc1 at #loc554))): + %b_a00_215 = arith.addf %b_a00_213, %b_a00_214 : f32 loc(#loc709) + tt.reduce.return %b_a00_215 : f32 loc(#loc690) + }) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc690) + %b_a00_203 = arith.cmpi slt, %m_tc0, %b_a00 : tensor<16xi64> loc(#loc555) + %b_a00_204 = arith.select %b_a00_203, %b_a00_202, %cst_9 : tensor<16xi1>, tensor<16xf32> loc(#loc556) + %b_a00_205 = tt.expand_dims %b_a00_204 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc557) + %b_a00_206 = tt.broadcast %b_a00_205 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc558) + %b_a00_207 = arith.mulf %b_a00_206, %b_Ai00_197 : tensor<16x16xf32> loc(#loc558) + %b_a00_208 = "tt.reduce"(%b_a00_207) <{axis = 0 : i32}> ({ + ^bb0(%b_a00_213: f32 loc(callsite(#loc1 at #loc559)), %b_a00_214: f32 loc(callsite(#loc1 at #loc559))): + %b_a00_215 = arith.addf %b_a00_213, %b_a00_214 : f32 loc(#loc710) + tt.reduce.return %b_a00_215 : f32 loc(#loc692) + }) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc692) + %b_a00_209 = arith.addf %b_a00_204, %b_a00_208 : tensor<16xf32> loc(#loc560) + %b_Ai00_210 = tt.expand_dims %b_a00_209 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc561) + %b_Ai00_211 = tt.broadcast %b_Ai00_210 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc561) + %b_Ai00_212 = arith.select %b_a00_200, %b_Ai00_211, %b_Ai00_197 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc561) + scf.yield %b_Ai00_212 : tensor<16x16xf32> loc(#loc222) + } loc(#loc550) + %12 = arith.subi %6, %i_tc1 : i64 loc(#loc223) + %13 = arith.minsi %12, %c16_i64 : i64 loc(#loc224) + %b_Ai11_139 = scf.for %i = %c2_i64 to %13 step %c1_i64 iter_args(%b_Ai11_197 = %b_Ai11) -> (tensor<16x16xf32>) : i64 { + %b_a11 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc563) + %b_a11_198 = arith.cmpi eq, %m_tc0, %b_a11 : tensor<16xi64> loc(#loc563) + %b_a11_199 = tt.expand_dims %b_a11_198 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc564) + %b_a11_200 = tt.broadcast %b_a11_199 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc565) + %b_a11_201 = arith.select %b_a11_200, %b_Ai11, %cst_6 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc565) + %b_a11_202 = "tt.reduce"(%b_a11_201) <{axis = 0 : i32}> ({ + ^bb0(%b_a11_213: f32 loc(callsite(#loc1 at #loc566)), %b_a11_214: f32 loc(callsite(#loc1 at #loc566))): + %b_a11_215 = arith.addf %b_a11_213, %b_a11_214 : f32 loc(#loc711) + tt.reduce.return %b_a11_215 : f32 loc(#loc694) + }) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc694) + %b_a11_203 = arith.cmpi slt, %m_tc0, %b_a11 : tensor<16xi64> loc(#loc567) + %b_a11_204 = arith.select %b_a11_203, %b_a11_202, %cst_9 : tensor<16xi1>, tensor<16xf32> loc(#loc568) + %b_a11_205 = tt.expand_dims %b_a11_204 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc569) + %b_a11_206 = tt.broadcast %b_a11_205 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc570) + %b_a11_207 = arith.mulf %b_a11_206, %b_Ai11_197 : tensor<16x16xf32> loc(#loc570) + %b_a11_208 = "tt.reduce"(%b_a11_207) <{axis = 0 : i32}> ({ + ^bb0(%b_a11_213: f32 loc(callsite(#loc1 at #loc571)), %b_a11_214: f32 loc(callsite(#loc1 at #loc571))): + %b_a11_215 = arith.addf %b_a11_213, %b_a11_214 : f32 loc(#loc712) + tt.reduce.return %b_a11_215 : f32 loc(#loc696) + }) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc696) + %b_a11_209 = arith.addf %b_a11_204, %b_a11_208 : tensor<16xf32> loc(#loc572) + %b_Ai11_210 = tt.expand_dims %b_a11_209 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc573) + %b_Ai11_211 = tt.broadcast %b_Ai11_210 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc573) + %b_Ai11_212 = arith.select %b_a11_200, %b_Ai11_211, %b_Ai11_197 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc573) + scf.yield %b_Ai11_212 : tensor<16x16xf32> loc(#loc237) + } loc(#loc562) + %14 = arith.subi %6, %i_tc2 : i64 loc(#loc238) + %15 = arith.minsi %14, %c16_i64 : i64 loc(#loc239) + %b_Ai22_140 = scf.for %i = %c2_i64 to %15 step %c1_i64 iter_args(%b_Ai22_197 = %b_Ai22) -> (tensor<16x16xf32>) : i64 { + %b_a22 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc575) + %b_a22_198 = arith.cmpi eq, %m_tc0, %b_a22 : tensor<16xi64> loc(#loc575) + %b_a22_199 = tt.expand_dims %b_a22_198 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc576) + %b_a22_200 = tt.broadcast %b_a22_199 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc577) + %b_a22_201 = arith.select %b_a22_200, %b_Ai22, %cst_6 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc577) + %b_a22_202 = "tt.reduce"(%b_a22_201) <{axis = 0 : i32}> ({ + ^bb0(%b_a22_213: f32 loc(callsite(#loc1 at #loc578)), %b_a22_214: f32 loc(callsite(#loc1 at #loc578))): + %b_a22_215 = arith.addf %b_a22_213, %b_a22_214 : f32 loc(#loc713) + tt.reduce.return %b_a22_215 : f32 loc(#loc698) + }) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc698) + %b_a22_203 = arith.cmpi slt, %m_tc0, %b_a22 : tensor<16xi64> loc(#loc579) + %b_a22_204 = arith.select %b_a22_203, %b_a22_202, %cst_9 : tensor<16xi1>, tensor<16xf32> loc(#loc580) + %b_a22_205 = tt.expand_dims %b_a22_204 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc581) + %b_a22_206 = tt.broadcast %b_a22_205 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc582) + %b_a22_207 = arith.mulf %b_a22_206, %b_Ai22_197 : tensor<16x16xf32> loc(#loc582) + %b_a22_208 = "tt.reduce"(%b_a22_207) <{axis = 0 : i32}> ({ + ^bb0(%b_a22_213: f32 loc(callsite(#loc1 at #loc583)), %b_a22_214: f32 loc(callsite(#loc1 at #loc583))): + %b_a22_215 = arith.addf %b_a22_213, %b_a22_214 : f32 loc(#loc714) + tt.reduce.return %b_a22_215 : f32 loc(#loc700) + }) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc700) + %b_a22_209 = arith.addf %b_a22_204, %b_a22_208 : tensor<16xf32> loc(#loc584) + %b_Ai22_210 = tt.expand_dims %b_a22_209 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc585) + %b_Ai22_211 = tt.broadcast %b_Ai22_210 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc585) + %b_Ai22_212 = arith.select %b_a22_200, %b_Ai22_211, %b_Ai22_197 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc585) + scf.yield %b_Ai22_212 : tensor<16x16xf32> loc(#loc252) + } loc(#loc574) + %16 = arith.subi %6, %i_tc3 : i64 loc(#loc253) + %17 = arith.minsi %16, %c16_i64 : i64 loc(#loc254) + %b_Ai33_141 = scf.for %i = %c2_i64 to %17 step %c1_i64 iter_args(%b_Ai33_197 = %b_Ai33) -> (tensor<16x16xf32>) : i64 { + %b_a33 = tt.splat %i : i64 -> tensor<16xi64> loc(#loc587) + %b_a33_198 = arith.cmpi eq, %m_tc0, %b_a33 : tensor<16xi64> loc(#loc587) + %b_a33_199 = tt.expand_dims %b_a33_198 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc588) + %b_a33_200 = tt.broadcast %b_a33_199 : tensor<16x1xi1> -> tensor<16x16xi1> loc(#loc589) + %b_a33_201 = arith.select %b_a33_200, %b_Ai33, %cst_6 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc589) + %b_a33_202 = "tt.reduce"(%b_a33_201) <{axis = 0 : i32}> ({ + ^bb0(%b_a33_213: f32 loc(callsite(#loc1 at #loc590)), %b_a33_214: f32 loc(callsite(#loc1 at #loc590))): + %b_a33_215 = arith.addf %b_a33_213, %b_a33_214 : f32 loc(#loc715) + tt.reduce.return %b_a33_215 : f32 loc(#loc702) + }) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc702) + %b_a33_203 = arith.cmpi slt, %m_tc0, %b_a33 : tensor<16xi64> loc(#loc591) + %b_a33_204 = arith.select %b_a33_203, %b_a33_202, %cst_9 : tensor<16xi1>, tensor<16xf32> loc(#loc592) + %b_a33_205 = tt.expand_dims %b_a33_204 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc593) + %b_a33_206 = tt.broadcast %b_a33_205 : tensor<16x1xf32> -> tensor<16x16xf32> loc(#loc594) + %b_a33_207 = arith.mulf %b_a33_206, %b_Ai33_197 : tensor<16x16xf32> loc(#loc594) + %b_a33_208 = "tt.reduce"(%b_a33_207) <{axis = 0 : i32}> ({ + ^bb0(%b_a33_213: f32 loc(callsite(#loc1 at #loc595)), %b_a33_214: f32 loc(callsite(#loc1 at #loc595))): + %b_a33_215 = arith.addf %b_a33_213, %b_a33_214 : f32 loc(#loc716) + tt.reduce.return %b_a33_215 : f32 loc(#loc704) + }) : (tensor<16x16xf32>) -> tensor<16xf32> loc(#loc704) + %b_a33_209 = arith.addf %b_a33_204, %b_a33_208 : tensor<16xf32> loc(#loc596) + %b_Ai33_210 = tt.expand_dims %b_a33_209 {axis = 0 : i32} : tensor<16xf32> -> tensor<1x16xf32> loc(#loc597) + %b_Ai33_211 = tt.broadcast %b_Ai33_210 : tensor<1x16xf32> -> tensor<16x16xf32> loc(#loc597) + %b_Ai33_212 = arith.select %b_a33_200, %b_Ai33_211, %b_Ai33_197 : tensor<16x16xi1>, tensor<16x16xf32> loc(#loc597) + scf.yield %b_Ai33_212 : tensor<16x16xf32> loc(#loc267) + } loc(#loc586) + %b_Ai00_142 = arith.uitofp %m_I : tensor<16x16xi1> to tensor<16x16xf32> loc(#loc598) + %b_Ai00_143 = arith.addf %b_Ai00_138, %b_Ai00_142 : tensor<16x16xf32> loc(#loc598) + %b_Ai11_144 = arith.addf %b_Ai11_139, %b_Ai00_142 : tensor<16x16xf32> loc(#loc599) + %b_Ai22_145 = arith.addf %b_Ai22_140, %b_Ai00_142 : tensor<16x16xf32> loc(#loc600) + %b_Ai33_146 = arith.addf %b_Ai33_141, %b_Ai00_142 : tensor<16x16xf32> loc(#loc601) + %b_Ai10 = tt.dot %b_Ai11_144, %b_A10_132, %cst_6, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc602) + %b_Ai10_147 = tt.dot %b_Ai10, %b_Ai00_143, %cst_6, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc603) + %b_Ai10_148 = arith.subf %cst_6, %b_Ai10_147 : tensor<16x16xf32> loc(#loc604) + %b_Ai21 = tt.dot %b_Ai22_145, %b_A21_134, %cst_6, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc605) + %b_Ai21_149 = tt.dot %b_Ai21, %b_Ai11_144, %cst_6, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc606) + %b_Ai21_150 = arith.subf %cst_6, %b_Ai21_149 : tensor<16x16xf32> loc(#loc607) + %b_Ai32 = tt.dot %b_Ai33_146, %b_A32_137, %cst_6, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc608) + %b_Ai32_151 = tt.dot %b_Ai32, %b_Ai22_145, %cst_6, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc609) + %b_Ai32_152 = arith.subf %cst_6, %b_Ai32_151 : tensor<16x16xf32> loc(#loc610) + %b_Ai20 = tt.dot %b_A20_133, %b_Ai00_143, %cst_6, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc611) + %b_Ai20_153 = tt.dot %b_A21_134, %b_Ai10_148, %b_Ai20, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc612) + %b_Ai20_154 = tt.dot %b_Ai22_145, %b_Ai20_153, %cst_6, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc613) + %b_Ai20_155 = arith.subf %cst_6, %b_Ai20_154 : tensor<16x16xf32> loc(#loc614) + %b_Ai31 = tt.dot %b_A31_136, %b_Ai11_144, %cst_6, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc615) + %b_Ai31_156 = tt.dot %b_A32_137, %b_Ai21_150, %b_Ai31, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc616) + %b_Ai31_157 = tt.dot %b_Ai33_146, %b_Ai31_156, %cst_6, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc617) + %b_Ai31_158 = arith.subf %cst_6, %b_Ai31_157 : tensor<16x16xf32> loc(#loc618) + %b_Ai30 = tt.dot %b_A30_135, %b_Ai00_143, %cst_6, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc619) + %b_Ai30_159 = tt.dot %b_A31_136, %b_Ai10_148, %b_Ai30, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc620) + %b_Ai30_160 = tt.dot %b_A32_137, %b_Ai20_155, %b_Ai30_159, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc621) + %b_Ai30_161 = tt.dot %b_Ai33_146, %b_Ai30_160, %cst_6, inputPrecision = tf32 : tensor<16x16xf32> * tensor<16x16xf32> -> tensor<16x16xf32> loc(#loc622) + %b_Ai30_162 = arith.subf %cst_6, %b_Ai30_161 : tensor<16x16xf32> loc(#loc623) + %p_A00 = tt.expand_dims %m_tc0_18 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc624) + %p_A00_163 = arith.muli %p_A00, %cst_5 : tensor<16x1xi64> loc(#loc625) + %p_A00_164 = tt.splat %A_16 : !tt.ptr -> tensor<16x1x!tt.ptr> loc(#loc626) + %p_A00_165 = tt.addptr %p_A00_164, %p_A00_163 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc626) + %p_A00_166 = tt.broadcast %p_A00_165 : tensor<16x1x!tt.ptr> -> tensor<16x16x!tt.ptr> loc(#loc627) + %p_A00_167 = tt.addptr %p_A00_166, %m_d_41 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc627) + %p_A10 = tt.expand_dims %m_tc1_21 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc628) + %p_A10_168 = arith.muli %p_A10, %cst_5 : tensor<16x1xi64> loc(#loc629) + %p_A10_169 = tt.addptr %p_A00_164, %p_A10_168 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc630) + %p_A10_170 = tt.broadcast %p_A10_169 : tensor<16x1x!tt.ptr> -> tensor<16x16x!tt.ptr> loc(#loc631) + %p_A10_171 = tt.addptr %p_A10_170, %m_d_41 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc631) + %p_A11 = arith.addi %o_i, %cst_4 : tensor<16xi32> loc(#loc632) + %p_A11_172 = tt.expand_dims %p_A11 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc633) + %p_A11_173 = tt.broadcast %p_A11_172 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc634) + %p_A11_174 = tt.addptr %p_A10_170, %p_A11_173 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc634) + %p_A20 = tt.expand_dims %m_tc2_23 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc635) + %p_A20_175 = arith.muli %p_A20, %cst_5 : tensor<16x1xi64> loc(#loc636) + %p_A20_176 = tt.addptr %p_A00_164, %p_A20_175 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc637) + %p_A20_177 = tt.broadcast %p_A20_176 : tensor<16x1x!tt.ptr> -> tensor<16x16x!tt.ptr> loc(#loc638) + %p_A20_178 = tt.addptr %p_A20_177, %m_d_41 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc638) + %p_A21 = tt.addptr %p_A20_177, %p_A11_173 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc639) + %p_A22 = arith.addi %o_i, %cst_3 : tensor<16xi32> loc(#loc640) + %p_A22_179 = tt.expand_dims %p_A22 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc641) + %p_A22_180 = tt.broadcast %p_A22_179 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc642) + %p_A22_181 = tt.addptr %p_A20_177, %p_A22_180 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc642) + %p_A30 = tt.expand_dims %m_tc3_25 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc643) + %p_A30_182 = arith.muli %p_A30, %cst_5 : tensor<16x1xi64> loc(#loc644) + %p_A30_183 = tt.addptr %p_A00_164, %p_A30_182 : tensor<16x1x!tt.ptr>, tensor<16x1xi64> loc(#loc645) + %p_A30_184 = tt.broadcast %p_A30_183 : tensor<16x1x!tt.ptr> -> tensor<16x16x!tt.ptr> loc(#loc646) + %p_A30_185 = tt.addptr %p_A30_184, %m_d_41 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc646) + %p_A31 = tt.addptr %p_A30_184, %p_A11_173 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc647) + %p_A32 = tt.addptr %p_A30_184, %p_A22_180 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc648) + %p_A33 = arith.addi %o_i, %cst_2 : tensor<16xi32> loc(#loc649) + %p_A33_186 = tt.expand_dims %p_A33 {axis = 0 : i32} : tensor<16xi32> -> tensor<1x16xi32> loc(#loc650) + %p_A33_187 = tt.broadcast %p_A33_186 : tensor<1x16xi32> -> tensor<16x16xi32> loc(#loc651) + %p_A33_188 = tt.addptr %p_A30_184, %p_A33_187 : tensor<16x16x!tt.ptr>, tensor<16x16xi32> loc(#loc651) + %m_A0 = arith.cmpi slt, %m_d_39, %cst_1 : tensor<1x16xi32> loc(#loc652) + %m_A0_189 = tt.broadcast %m_A0 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc653) + %m_A0_190 = arith.andi %b_A00_43, %m_A0_189 : tensor<16x16xi1> loc(#loc653) + %m_A1 = arith.andi %b_A11_56, %m_A0_189 : tensor<16x16xi1> loc(#loc654) + %m_A2 = arith.andi %b_A22_69, %m_A0_189 : tensor<16x16xi1> loc(#loc655) + %m_A3 = arith.andi %b_A33_82, %m_A0_189 : tensor<16x16xi1> loc(#loc656) + %m_A11 = arith.cmpi slt, %p_A11_172, %cst_1 : tensor<1x16xi32> loc(#loc657) + %m_A11_191 = tt.broadcast %m_A11 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc658) + %m_A11_192 = arith.andi %b_A11_56, %m_A11_191 : tensor<16x16xi1> loc(#loc658) + %m_A21 = arith.andi %b_A22_69, %m_A11_191 : tensor<16x16xi1> loc(#loc659) + %m_A22 = arith.cmpi slt, %p_A22_179, %cst_1 : tensor<1x16xi32> loc(#loc660) + %m_A22_193 = tt.broadcast %m_A22 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc661) + %m_A22_194 = arith.andi %b_A22_69, %m_A22_193 : tensor<16x16xi1> loc(#loc661) + %m_A31 = arith.andi %b_A33_82, %m_A11_191 : tensor<16x16xi1> loc(#loc662) + %m_A32 = arith.andi %b_A33_82, %m_A22_193 : tensor<16x16xi1> loc(#loc663) + %m_A33 = arith.cmpi slt, %p_A33_186, %cst_1 : tensor<1x16xi32> loc(#loc664) + %m_A33_195 = tt.broadcast %m_A33 : tensor<1x16xi1> -> tensor<16x16xi1> loc(#loc665) + %m_A33_196 = arith.andi %b_A33_82, %m_A33_195 : tensor<16x16xi1> loc(#loc665) + %18 = arith.truncf %b_Ai00_143 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc336) + tt.store %p_A00_167, %18, %m_A0_190 : tensor<16x16x!tt.ptr> loc(#loc337) + %19 = arith.truncf %b_Ai10_148 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc338) + tt.store %p_A10_171, %19, %m_A1 : tensor<16x16x!tt.ptr> loc(#loc339) + %20 = arith.truncf %b_Ai11_144 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc340) + tt.store %p_A11_174, %20, %m_A11_192 : tensor<16x16x!tt.ptr> loc(#loc341) + %21 = arith.truncf %b_Ai20_155 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc342) + tt.store %p_A20_178, %21, %m_A2 : tensor<16x16x!tt.ptr> loc(#loc343) + %22 = arith.truncf %b_Ai21_150 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc344) + tt.store %p_A21, %22, %m_A21 : tensor<16x16x!tt.ptr> loc(#loc345) + %23 = arith.truncf %b_Ai22_145 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc346) + tt.store %p_A22_181, %23, %m_A22_194 : tensor<16x16x!tt.ptr> loc(#loc347) + %24 = arith.truncf %b_Ai30_162 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc348) + tt.store %p_A30_185, %24, %m_A3 : tensor<16x16x!tt.ptr> loc(#loc349) + %25 = arith.truncf %b_Ai31_158 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc350) + tt.store %p_A31, %25, %m_A31 : tensor<16x16x!tt.ptr> loc(#loc351) + %26 = arith.truncf %b_Ai32_152 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc352) + tt.store %p_A32, %26, %m_A32 : tensor<16x16x!tt.ptr> loc(#loc353) + %27 = arith.truncf %b_Ai33_146 : tensor<16x16xf32> to tensor<16x16xbf16> loc(#loc354) + tt.store %p_A33_188, %27, %m_A33_196 : tensor<16x16x!tt.ptr> loc(#loc355) + tt.return loc(#loc356) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":70:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":70:36) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":70:61) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":70:67) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":71:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":71:34) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":78:25) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":80:13) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":80:19) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":81:8) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":84:23) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":85:23) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":86:23) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":88:16) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":88:20) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":88:40) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":88:9) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":89:28) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":89:9) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":91:23) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":92:21) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":92:28) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":93:21) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":93:28) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":94:21) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":94:28) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":95:21) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":95:28) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":98:29) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":98:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":98:51) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":98:35) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":99:51) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":99:35) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":100:51) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":100:35) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":101:51) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":101:35) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":102:19) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":102:51) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":103:19) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":103:51) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":104:19) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":104:51) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":105:19) +#loc47 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":105:51) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":109:30) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":109:19) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":109:36) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":110:36) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":111:36) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":112:36) +#loc54 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":114:23) +#loc55 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":115:23) +#loc56 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":116:23) +#loc57 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":117:23) +#loc58 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":137:21) +#loc59 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":138:20) +#loc60 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":138:38) +#loc61 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":138:25) +#loc62 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:33) +#loc63 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:45) +#loc64 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:19) +#loc65 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:56) +#loc66 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":139:52) +#loc67 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":140:40) +#loc68 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":140:67) +#loc69 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":140:52) +#loc70 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":140:23) +#loc71 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":142:38) +#loc72 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":142:45) +#loc73 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":144:19) +#loc74 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":144:11) +#loc75 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":145:37) +#loc76 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":145:49) +#loc77 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":145:23) +#loc78 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":145:56) +#loc79 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":146:44) +#loc80 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":146:56) +#loc81 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":146:27) +#loc82 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":148:42) +#loc83 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":148:49) +#loc84 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":150:49) +#loc85 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":152:23) +#loc86 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":152:15) +#loc87 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":153:41) +#loc88 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":153:53) +#loc89 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":153:27) +#loc90 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":153:60) +#loc91 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":154:48) +#loc92 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":154:60) +#loc93 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":154:31) +#loc94 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":156:46) +#loc95 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":156:53) +#loc96 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":158:53) +#loc97 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":159:53) +#loc98 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":161:27) +#loc99 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":161:19) +#loc100 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":162:45) +#loc101 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":162:57) +#loc102 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":162:31) +#loc103 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":162:64) +#loc104 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":163:52) +#loc105 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":163:64) +#loc106 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":163:35) +#loc107 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":165:50) +#loc108 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":165:57) +#loc109 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":167:57) +#loc110 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":168:57) +#loc111 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":169:57) +#loc112 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":144:8) +#loc113 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":179:14) +#loc114 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":179:29) +#loc115 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":179:25) +#loc116 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":180:26) +#loc117 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:38) +#loc118 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:32) +#loc119 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:55) +#loc120 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:49) +#loc121 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:75) +#loc122 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:91) +#loc123 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:86) +#loc124 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:33) +#loc125 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:70) +#loc126 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:102) +#loc127 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":183:17) +#loc128 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:38) +#loc129 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:32) +#loc130 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:55) +#loc131 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:49) +#loc132 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:75) +#loc133 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:91) +#loc134 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:86) +#loc135 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:70) +#loc136 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:102) +#loc137 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":184:17) +#loc138 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:38) +#loc139 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:32) +#loc140 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:55) +#loc141 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:49) +#loc142 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:75) +#loc143 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:91) +#loc144 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:86) +#loc145 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:70) +#loc146 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:102) +#loc147 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":185:17) +#loc148 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:38) +#loc149 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:32) +#loc150 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:55) +#loc151 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:49) +#loc152 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:75) +#loc153 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:91) +#loc154 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:86) +#loc155 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:70) +#loc156 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:102) +#loc157 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":186:17) +#loc158 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:43) +#loc159 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:80) +#loc160 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:64) +#loc161 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:96) +#loc162 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":188:17) +#loc163 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:43) +#loc164 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:80) +#loc165 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:64) +#loc166 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:96) +#loc167 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":189:17) +#loc168 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:43) +#loc169 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:80) +#loc170 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:64) +#loc171 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:96) +#loc172 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":190:17) +#loc173 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:43) +#loc174 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:80) +#loc175 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:64) +#loc176 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:96) +#loc177 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":191:17) +#loc178 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:43) +#loc179 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:80) +#loc180 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:64) +#loc181 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:96) +#loc182 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":192:17) +#loc183 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:43) +#loc184 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:80) +#loc185 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:64) +#loc186 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:96) +#loc187 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":193:17) +#loc188 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":201:25) +#loc189 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":201:20) +#loc190 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":202:25) +#loc191 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":202:20) +#loc192 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":203:25) +#loc193 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":203:20) +#loc194 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":204:25) +#loc195 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":204:20) +#loc196 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":207:20) +#loc197 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":208:20) +#loc198 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":209:20) +#loc199 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":210:20) +#loc200 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":211:20) +#loc201 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":212:20) +#loc202 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":222:14) +#loc203 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":223:14) +#loc204 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":224:14) +#loc205 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":225:14) +#loc206 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":227:34) +#loc207 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":227:30) +#loc208 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":227:22) +#loc209 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":228:40) +#loc210 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":228:43) +#loc211 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":228:61) +#loc212 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc214 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc215 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":229:31) +#loc216 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":229:41) +#loc217 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":230:37) +#loc218 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":230:48) +#loc220 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":230:24) +#loc221 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":231:54) +#loc222 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":231:8) +#loc223 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":232:34) +#loc224 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":232:30) +#loc225 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":232:22) +#loc226 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":233:40) +#loc227 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":233:43) +#loc228 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":233:61) +#loc230 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":234:31) +#loc231 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":234:41) +#loc232 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":235:37) +#loc233 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":235:48) +#loc235 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":235:24) +#loc236 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":236:54) +#loc237 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":236:8) +#loc238 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":237:34) +#loc239 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":237:30) +#loc240 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":237:22) +#loc241 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":238:40) +#loc242 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":238:43) +#loc243 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":238:61) +#loc245 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":239:31) +#loc246 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":239:41) +#loc247 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":240:37) +#loc248 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":240:48) +#loc250 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":240:24) +#loc251 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":241:54) +#loc252 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":241:8) +#loc253 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":242:34) +#loc254 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":242:30) +#loc255 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":242:22) +#loc256 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":243:40) +#loc257 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":243:43) +#loc258 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":243:61) +#loc260 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":244:31) +#loc261 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":244:41) +#loc262 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":245:37) +#loc263 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":245:48) +#loc265 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":245:24) +#loc266 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":246:54) +#loc267 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":246:8) +#loc268 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":248:14) +#loc269 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":249:14) +#loc270 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":250:14) +#loc271 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":251:14) +#loc272 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":258:23) +#loc273 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":259:8) +#loc274 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":257:14) +#loc275 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":263:23) +#loc276 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":264:8) +#loc277 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":262:14) +#loc278 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":268:23) +#loc279 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":269:8) +#loc280 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":267:14) +#loc281 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":275:22) +#loc282 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":276:22) +#loc283 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":275:8) +#loc284 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":273:14) +#loc285 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":281:22) +#loc286 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":282:22) +#loc287 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":281:8) +#loc288 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":279:14) +#loc289 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":287:22) +#loc290 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":288:22) +#loc291 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":289:22) +#loc292 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":287:8) +#loc293 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":285:14) +#loc294 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":297:30) +#loc295 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":297:42) +#loc296 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":297:16) +#loc297 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":297:51) +#loc298 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":298:30) +#loc299 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":298:42) +#loc300 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":298:16) +#loc301 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":298:51) +#loc302 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":299:57) +#loc303 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":299:62) +#loc304 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":299:51) +#loc305 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":300:30) +#loc306 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":300:42) +#loc307 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":300:16) +#loc308 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":300:51) +#loc309 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":301:51) +#loc310 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":302:59) +#loc311 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":302:64) +#loc312 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":302:51) +#loc313 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":303:30) +#loc314 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":303:42) +#loc315 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":303:16) +#loc316 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":303:51) +#loc317 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":304:51) +#loc318 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":305:51) +#loc319 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":306:59) +#loc320 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":306:64) +#loc321 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":306:51) +#loc322 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":308:44) +#loc323 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":308:29) +#loc324 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":309:29) +#loc325 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":310:29) +#loc326 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":311:29) +#loc327 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":312:52) +#loc328 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":312:30) +#loc329 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":313:30) +#loc330 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":314:54) +#loc331 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":314:30) +#loc332 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":315:30) +#loc333 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":316:30) +#loc334 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":317:54) +#loc335 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":317:30) +#loc336 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":319:30) +#loc337 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":319:20) +#loc338 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":320:30) +#loc339 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":320:20) +#loc340 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":321:30) +#loc341 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":321:20) +#loc342 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":322:30) +#loc343 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":322:20) +#loc344 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":323:30) +#loc345 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":323:20) +#loc346 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":324:30) +#loc347 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":324:20) +#loc348 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":325:30) +#loc349 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":325:20) +#loc350 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":326:30) +#loc351 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":326:20) +#loc352 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":327:30) +#loc353 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":327:20) +#loc354 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":328:30) +#loc355 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":328:20) +#loc356 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/gated_delta_rule/chunk_fwd.py":328:4) +#loc362 = loc("i_tc1"(#loc12)) +#loc363 = loc("i_tc2"(#loc13)) +#loc364 = loc("i_tc3"(#loc14)) +#loc365 = loc("k"(#loc15)) +#loc366 = loc("k"(#loc16)) +#loc367 = loc("k"(#loc17)) +#loc368 = loc("k"(#loc18)) +#loc369 = loc("A"(#loc19)) +#loc370 = loc("A"(#loc20)) +#loc371 = loc("o_i"(#loc21)) +#loc372 = loc("m_tc0"(#loc22)) +#loc373 = loc("m_tc0"(#loc23)) +#loc374 = loc("m_tc1"(#loc24)) +#loc375 = loc("m_tc1"(#loc25)) +#loc376 = loc("m_tc2"(#loc26)) +#loc377 = loc("m_tc2"(#loc27)) +#loc378 = loc("m_tc3"(#loc28)) +#loc379 = loc("m_tc3"(#loc29)) +#loc380 = loc("p_b0"(#loc30)) +#loc381 = loc("p_b0"(#loc31)) +#loc382 = loc("p_b0"(#loc32)) +#loc383 = loc("p_b0"(#loc33)) +#loc384 = loc("p_b1"(#loc34)) +#loc385 = loc("p_b1"(#loc35)) +#loc386 = loc("p_b2"(#loc36)) +#loc387 = loc("p_b2"(#loc37)) +#loc388 = loc("p_b3"(#loc38)) +#loc389 = loc("p_b3"(#loc39)) +#loc390 = loc("b_b0"(#loc40)) +#loc391 = loc("b_b0"(#loc41)) +#loc392 = loc("b_b1"(#loc42)) +#loc393 = loc("b_b1"(#loc43)) +#loc394 = loc("b_b2"(#loc44)) +#loc395 = loc("b_b2"(#loc45)) +#loc396 = loc("b_b3"(#loc46)) +#loc397 = loc("b_b3"(#loc47)) +#loc398 = loc("p_g0"(#loc48)) +#loc399 = loc("p_g0"(#loc49)) +#loc400 = loc("p_g0"(#loc50)) +#loc401 = loc("p_g1"(#loc51)) +#loc402 = loc("p_g2"(#loc52)) +#loc403 = loc("p_g3"(#loc53)) +#loc404 = loc("b_g0"(#loc54)) +#loc405 = loc("b_g1"(#loc55)) +#loc406 = loc("b_g2"(#loc56)) +#loc407 = loc("b_g3"(#loc57)) +#loc408 = loc("b_A00"(#loc58)) +#loc409 = loc("o_k"(#loc59)) +#loc410 = loc("o_k"(#loc60)) +#loc411 = loc("o_k"(#loc61)) +#loc412 = loc("p_k0"(#loc62)) +#loc413 = loc("p_k0"(#loc63)) +#loc414 = loc("p_k0"(#loc64)) +#loc415 = loc("p_k0"(#loc65)) +#loc416 = loc("p_k0"(#loc66)) +#loc417 = loc("b_k0"(#loc67)) +#loc418 = loc("b_k0"(#loc68)) +#loc419 = loc("b_k0"(#loc69)) +#loc420 = loc("b_k0"(#loc70)) +#loc421 = loc("b_A00"(#loc71)) +#loc422 = loc("b_A00"(#loc72)) +#loc423 = loc("p_k1"(#loc75)) +#loc424 = loc("p_k1"(#loc76)) +#loc425 = loc("p_k1"(#loc77)) +#loc426 = loc("p_k1"(#loc78)) +#loc427 = loc("b_k1"(#loc79)) +#loc428 = loc("b_k1"(#loc80)) +#loc429 = loc("b_k1"(#loc81)) +#loc430 = loc("b_A11"(#loc82)) +#loc431 = loc("b_A11"(#loc83)) +#loc432 = loc("b_A10"(#loc84)) +#loc433 = loc("b_A22"(#loc86)) +#loc434 = loc("p_k2"(#loc87)) +#loc435 = loc("p_k2"(#loc88)) +#loc436 = loc("p_k2"(#loc89)) +#loc437 = loc("p_k2"(#loc90)) +#loc438 = loc("b_k2"(#loc91)) +#loc439 = loc("b_k2"(#loc92)) +#loc440 = loc("b_k2"(#loc93)) +#loc441 = loc("b_A22"(#loc94)) +#loc442 = loc("b_A22"(#loc95)) +#loc443 = loc("b_A20"(#loc96)) +#loc444 = loc("b_A21"(#loc97)) +#loc445 = loc("b_A33"(#loc99)) +#loc446 = loc("p_k3"(#loc100)) +#loc447 = loc("p_k3"(#loc101)) +#loc448 = loc("p_k3"(#loc102)) +#loc449 = loc("p_k3"(#loc103)) +#loc450 = loc("b_k3"(#loc104)) +#loc451 = loc("b_k3"(#loc105)) +#loc452 = loc("b_k3"(#loc106)) +#loc453 = loc("b_A33"(#loc107)) +#loc454 = loc("b_A33"(#loc108)) +#loc455 = loc("b_A30"(#loc109)) +#loc456 = loc("b_A31"(#loc110)) +#loc457 = loc("b_A32"(#loc111)) +#loc458 = loc("m_d"(#loc113)) +#loc459 = loc("m_d"(#loc114)) +#loc460 = loc("m_d"(#loc115)) +#loc461 = loc("m_I"(#loc116)) +#loc462 = loc("b_A00"(#loc117)) +#loc463 = loc("b_A00"(#loc118)) +#loc464 = loc("b_A00"(#loc119)) +#loc465 = loc("b_A00"(#loc120)) +#loc466 = loc("b_A00"(#loc121)) +#loc467 = loc("b_A00"(#loc122)) +#loc468 = loc("b_A00"(#loc123)) +#loc469 = loc("b_A00"(#loc125)) +#loc470 = loc("b_A00"(#loc126)) +#loc471 = loc("b_A00"(#loc127)) +#loc472 = loc("b_A11"(#loc128)) +#loc473 = loc("b_A11"(#loc129)) +#loc474 = loc("b_A11"(#loc130)) +#loc475 = loc("b_A11"(#loc131)) +#loc476 = loc("b_A11"(#loc132)) +#loc477 = loc("b_A11"(#loc133)) +#loc478 = loc("b_A11"(#loc134)) +#loc479 = loc("b_A11"(#loc135)) +#loc480 = loc("b_A11"(#loc136)) +#loc481 = loc("b_A11"(#loc137)) +#loc482 = loc("b_A22"(#loc138)) +#loc483 = loc("b_A22"(#loc139)) +#loc484 = loc("b_A22"(#loc140)) +#loc485 = loc("b_A22"(#loc141)) +#loc486 = loc("b_A22"(#loc142)) +#loc487 = loc("b_A22"(#loc143)) +#loc488 = loc("b_A22"(#loc144)) +#loc489 = loc("b_A22"(#loc145)) +#loc490 = loc("b_A22"(#loc146)) +#loc491 = loc("b_A22"(#loc147)) +#loc492 = loc("b_A33"(#loc148)) +#loc493 = loc("b_A33"(#loc149)) +#loc494 = loc("b_A33"(#loc150)) +#loc495 = loc("b_A33"(#loc151)) +#loc496 = loc("b_A33"(#loc152)) +#loc497 = loc("b_A33"(#loc153)) +#loc498 = loc("b_A33"(#loc154)) +#loc499 = loc("b_A33"(#loc155)) +#loc500 = loc("b_A33"(#loc156)) +#loc501 = loc("b_A33"(#loc157)) +#loc502 = loc("b_A10"(#loc158)) +#loc503 = loc("b_A10"(#loc159)) +#loc504 = loc("b_A10"(#loc160)) +#loc505 = loc("b_A10"(#loc161)) +#loc506 = loc("b_A10"(#loc162)) +#loc507 = loc("b_A20"(#loc163)) +#loc508 = loc("b_A20"(#loc164)) +#loc509 = loc("b_A20"(#loc165)) +#loc510 = loc("b_A20"(#loc166)) +#loc511 = loc("b_A20"(#loc167)) +#loc512 = loc("b_A21"(#loc168)) +#loc513 = loc("b_A21"(#loc169)) +#loc514 = loc("b_A21"(#loc170)) +#loc515 = loc("b_A21"(#loc171)) +#loc516 = loc("b_A21"(#loc172)) +#loc517 = loc("b_A30"(#loc173)) +#loc518 = loc("b_A30"(#loc174)) +#loc519 = loc("b_A30"(#loc175)) +#loc520 = loc("b_A30"(#loc176)) +#loc521 = loc("b_A30"(#loc177)) +#loc522 = loc("b_A31"(#loc178)) +#loc523 = loc("b_A31"(#loc179)) +#loc524 = loc("b_A31"(#loc180)) +#loc525 = loc("b_A31"(#loc181)) +#loc526 = loc("b_A31"(#loc182)) +#loc527 = loc("b_A32"(#loc183)) +#loc528 = loc("b_A32"(#loc184)) +#loc529 = loc("b_A32"(#loc185)) +#loc530 = loc("b_A32"(#loc186)) +#loc531 = loc("b_A32"(#loc187)) +#loc532 = loc("b_A00"(#loc188)) +#loc533 = loc("b_A00"(#loc189)) +#loc534 = loc("b_A11"(#loc190)) +#loc535 = loc("b_A11"(#loc191)) +#loc536 = loc("b_A22"(#loc192)) +#loc537 = loc("b_A22"(#loc193)) +#loc538 = loc("b_A33"(#loc194)) +#loc539 = loc("b_A33"(#loc195)) +#loc540 = loc("b_A10"(#loc196)) +#loc541 = loc("b_A20"(#loc197)) +#loc542 = loc("b_A21"(#loc198)) +#loc543 = loc("b_A30"(#loc199)) +#loc544 = loc("b_A31"(#loc200)) +#loc545 = loc("b_A32"(#loc201)) +#loc546 = loc("b_Ai00"(#loc202)) +#loc547 = loc("b_Ai11"(#loc203)) +#loc548 = loc("b_Ai22"(#loc204)) +#loc549 = loc("b_Ai33"(#loc205)) +#loc550 = loc("b_Ai00"(#loc208)) +#loc551 = loc("b_a00"(#loc209)) +#loc552 = loc("b_a00"(#loc210)) +#loc553 = loc("b_a00"(#loc211)) +#loc555 = loc("b_a00"(#loc215)) +#loc556 = loc("b_a00"(#loc216)) +#loc557 = loc("b_a00"(#loc217)) +#loc558 = loc("b_a00"(#loc218)) +#loc560 = loc("b_a00"(#loc220)) +#loc561 = loc("b_Ai00"(#loc221)) +#loc562 = loc("b_Ai11"(#loc225)) +#loc563 = loc("b_a11"(#loc226)) +#loc564 = loc("b_a11"(#loc227)) +#loc565 = loc("b_a11"(#loc228)) +#loc567 = loc("b_a11"(#loc230)) +#loc568 = loc("b_a11"(#loc231)) +#loc569 = loc("b_a11"(#loc232)) +#loc570 = loc("b_a11"(#loc233)) +#loc572 = loc("b_a11"(#loc235)) +#loc573 = loc("b_Ai11"(#loc236)) +#loc574 = loc("b_Ai22"(#loc240)) +#loc575 = loc("b_a22"(#loc241)) +#loc576 = loc("b_a22"(#loc242)) +#loc577 = loc("b_a22"(#loc243)) +#loc579 = loc("b_a22"(#loc245)) +#loc580 = loc("b_a22"(#loc246)) +#loc581 = loc("b_a22"(#loc247)) +#loc582 = loc("b_a22"(#loc248)) +#loc584 = loc("b_a22"(#loc250)) +#loc585 = loc("b_Ai22"(#loc251)) +#loc586 = loc("b_Ai33"(#loc255)) +#loc587 = loc("b_a33"(#loc256)) +#loc588 = loc("b_a33"(#loc257)) +#loc589 = loc("b_a33"(#loc258)) +#loc591 = loc("b_a33"(#loc260)) +#loc592 = loc("b_a33"(#loc261)) +#loc593 = loc("b_a33"(#loc262)) +#loc594 = loc("b_a33"(#loc263)) +#loc596 = loc("b_a33"(#loc265)) +#loc597 = loc("b_Ai33"(#loc266)) +#loc598 = loc("b_Ai00"(#loc268)) +#loc599 = loc("b_Ai11"(#loc269)) +#loc600 = loc("b_Ai22"(#loc270)) +#loc601 = loc("b_Ai33"(#loc271)) +#loc602 = loc("b_Ai10"(#loc272)) +#loc603 = loc("b_Ai10"(#loc273)) +#loc604 = loc("b_Ai10"(#loc274)) +#loc605 = loc("b_Ai21"(#loc275)) +#loc606 = loc("b_Ai21"(#loc276)) +#loc607 = loc("b_Ai21"(#loc277)) +#loc608 = loc("b_Ai32"(#loc278)) +#loc609 = loc("b_Ai32"(#loc279)) +#loc610 = loc("b_Ai32"(#loc280)) +#loc611 = loc("b_Ai20"(#loc281)) +#loc612 = loc("b_Ai20"(#loc282)) +#loc613 = loc("b_Ai20"(#loc283)) +#loc614 = loc("b_Ai20"(#loc284)) +#loc615 = loc("b_Ai31"(#loc285)) +#loc616 = loc("b_Ai31"(#loc286)) +#loc617 = loc("b_Ai31"(#loc287)) +#loc618 = loc("b_Ai31"(#loc288)) +#loc619 = loc("b_Ai30"(#loc289)) +#loc620 = loc("b_Ai30"(#loc290)) +#loc621 = loc("b_Ai30"(#loc291)) +#loc622 = loc("b_Ai30"(#loc292)) +#loc623 = loc("b_Ai30"(#loc293)) +#loc624 = loc("p_A00"(#loc294)) +#loc625 = loc("p_A00"(#loc295)) +#loc626 = loc("p_A00"(#loc296)) +#loc627 = loc("p_A00"(#loc297)) +#loc628 = loc("p_A10"(#loc298)) +#loc629 = loc("p_A10"(#loc299)) +#loc630 = loc("p_A10"(#loc300)) +#loc631 = loc("p_A10"(#loc301)) +#loc632 = loc("p_A11"(#loc302)) +#loc633 = loc("p_A11"(#loc303)) +#loc634 = loc("p_A11"(#loc304)) +#loc635 = loc("p_A20"(#loc305)) +#loc636 = loc("p_A20"(#loc306)) +#loc637 = loc("p_A20"(#loc307)) +#loc638 = loc("p_A20"(#loc308)) +#loc639 = loc("p_A21"(#loc309)) +#loc640 = loc("p_A22"(#loc310)) +#loc641 = loc("p_A22"(#loc311)) +#loc642 = loc("p_A22"(#loc312)) +#loc643 = loc("p_A30"(#loc313)) +#loc644 = loc("p_A30"(#loc314)) +#loc645 = loc("p_A30"(#loc315)) +#loc646 = loc("p_A30"(#loc316)) +#loc647 = loc("p_A31"(#loc317)) +#loc648 = loc("p_A32"(#loc318)) +#loc649 = loc("p_A33"(#loc319)) +#loc650 = loc("p_A33"(#loc320)) +#loc651 = loc("p_A33"(#loc321)) +#loc652 = loc("m_A0"(#loc322)) +#loc653 = loc("m_A0"(#loc323)) +#loc654 = loc("m_A1"(#loc324)) +#loc655 = loc("m_A2"(#loc325)) +#loc656 = loc("m_A3"(#loc326)) +#loc657 = loc("m_A11"(#loc327)) +#loc658 = loc("m_A11"(#loc328)) +#loc659 = loc("m_A21"(#loc329)) +#loc660 = loc("m_A22"(#loc330)) +#loc661 = loc("m_A22"(#loc331)) +#loc662 = loc("m_A31"(#loc332)) +#loc663 = loc("m_A32"(#loc333)) +#loc664 = loc("m_A33"(#loc334)) +#loc665 = loc("m_A33"(#loc335)) +#loc666 = loc(fused[#loc380, #loc381]) +#loc667 = loc(fused[#loc398, #loc399]) +#loc668 = loc("b_A11"(#loc408)) +#loc669 = loc("b_A11"(#loc431)) +#loc670 = loc("b_A10"(#loc432)) +#loc671 = loc("b_A33"(#loc433)) +#loc672 = loc("b_A22"(#loc442)) +#loc673 = loc("b_A20"(#loc443)) +#loc674 = loc("b_A21"(#loc444)) +#loc675 = loc("b_A30"(#loc445)) +#loc676 = loc("b_A33"(#loc454)) +#loc677 = loc("b_A30"(#loc455)) +#loc678 = loc("b_A31"(#loc456)) +#loc679 = loc("b_A32"(#loc457)) +#loc680 = loc(callsite(#loc124 at #loc469)) +#loc681 = loc(callsite(#loc124 at #loc479)) +#loc682 = loc(callsite(#loc124 at #loc489)) +#loc683 = loc(callsite(#loc124 at #loc499)) +#loc684 = loc(callsite(#loc124 at #loc504)) +#loc685 = loc(callsite(#loc124 at #loc509)) +#loc686 = loc(callsite(#loc124 at #loc514)) +#loc687 = loc(callsite(#loc124 at #loc519)) +#loc688 = loc(callsite(#loc124 at #loc524)) +#loc689 = loc(callsite(#loc124 at #loc529)) +#loc690 = loc(callsite(#loc212 at #loc554)) +#loc692 = loc(callsite(#loc212 at #loc559)) +#loc694 = loc(callsite(#loc212 at #loc566)) +#loc696 = loc(callsite(#loc212 at #loc571)) +#loc698 = loc(callsite(#loc212 at #loc578)) +#loc700 = loc(callsite(#loc212 at #loc583)) +#loc702 = loc(callsite(#loc212 at #loc590)) +#loc704 = loc(callsite(#loc212 at #loc595)) +#loc706 = loc("b_A22"(#loc668)) +#loc707 = loc("b_A20"(#loc671)) +#loc708 = loc("b_A31"(#loc675)) +#loc709 = loc(callsite(#loc214 at #loc690)) +#loc710 = loc(callsite(#loc214 at #loc692)) +#loc711 = loc(callsite(#loc214 at #loc694)) +#loc712 = loc(callsite(#loc214 at #loc696)) +#loc713 = loc(callsite(#loc214 at #loc698)) +#loc714 = loc(callsite(#loc214 at #loc700)) +#loc715 = loc(callsite(#loc214 at #loc702)) +#loc716 = loc(callsite(#loc214 at #loc704)) +#loc717 = loc("b_A33"(#loc706)) +#loc718 = loc("b_A21"(#loc707)) +#loc719 = loc("b_A32"(#loc708)) +#loc720 = loc("b_A10"(#loc717)) +#loc721 = loc("b_A30"(#loc718)) +#loc722 = loc("b_A20"(#loc720)) +#loc723 = loc("b_A31"(#loc721)) +#loc724 = loc("b_A21"(#loc722)) +#loc725 = loc("b_A32"(#loc723)) +#loc726 = loc("b_A30"(#loc724)) +#loc727 = loc("b_A31"(#loc726)) +#loc728 = loc("b_A32"(#loc727)) diff --git a/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/__grp__l2norm_bwd_kernel.json b/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/__grp__l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..01b66d4038a1f646981deb6b40d514fba4a9d483 --- /dev/null +++ b/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/__grp__l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_bwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.source", "l2norm_bwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.ttir", "l2norm_bwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.ttgir", "l2norm_bwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.llir", "l2norm_bwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.ptx", "l2norm_bwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.cubin", "l2norm_bwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.cubin b/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..8c48f2d2f80629e511f2f0a059e8d7e73a59c724 --- /dev/null +++ b/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.cubin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8533299dce1bd88dc3433f984ea3cd268806d21c64a06a7a38fbc0a7677e4518 +size 631384 diff --git a/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.json b/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..7026e87422ce1eb53dc3e4db72bea7b37b579454 --- /dev/null +++ b/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"hash": "11db65f001c3c6ebdbc23952034cf24e763c295ffeabea5c29f7811ca1025f38", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 1, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 256, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_bwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.llir b/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..01ec7f43b3426709e1eefde20c10919fab06dac3 --- /dev/null +++ b/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.llir @@ -0,0 +1,9405 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_bwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, ptr addrspace(1) %3, float %4, i32 %5, ptr addrspace(1) readnone captures(none) %6, ptr addrspace(1) readnone captures(none) %7) local_unnamed_addr #0 !dbg !4 { + %9 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !12 + %10 = zext nneg i32 %9 to i64, !dbg !13 + %11 = shl nuw nsw i64 %10, 7, !dbg !14 + %12 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !15 + %13 = and i32 %12, 16, !dbg !15 + %.lobit = lshr exact i32 %13, 4, !dbg !15 + %14 = or disjoint i32 %.lobit, 2, !dbg !15 + %15 = or disjoint i32 %.lobit, 4, !dbg !15 + %16 = or disjoint i32 %.lobit, 6, !dbg !15 + %17 = shl nuw nsw i32 %12, 2, !dbg !15 + %18 = zext nneg i32 %.lobit to i64, !dbg !16 + %19 = zext nneg i32 %14 to i64, !dbg !16 + %20 = zext nneg i32 %15 to i64, !dbg !16 + %21 = zext nneg i32 %16 to i64, !dbg !16 + %22 = or disjoint i64 %11, %18, !dbg !16 + %23 = or disjoint i64 %11, %19, !dbg !16 + %24 = or disjoint i64 %11, %20, !dbg !16 + %25 = or disjoint i64 %11, %21, !dbg !16 + %26 = sext i32 %5 to i64, !dbg !17 + %27 = icmp slt i64 %22, %26, !dbg !17 + %28 = icmp slt i64 %23, %26, !dbg !17 + %29 = icmp slt i64 %24, %26, !dbg !17 + %30 = icmp slt i64 %25, %26, !dbg !17 + %31 = insertelement <32 x i32> poison, i32 %.lobit, i64 0, !dbg !15 + %32 = insertelement <16 x i32> poison, i32 %.lobit, i64 0, !dbg !15 + %33 = insertelement <8 x i32> poison, i32 %.lobit, i64 0, !dbg !15 + %34 = insertelement <4 x i32> poison, i32 %.lobit, i64 0, !dbg !15 + %35 = and i32 %17, 124, !dbg !15 + %36 = or disjoint i32 %35, 1, !dbg !15 + %37 = insertelement <2 x i32> poison, i32 %35, i64 0, !dbg !15 + %38 = insertelement <64 x i32> poison, i32 %36, i64 2, !dbg !16 + %39 = insertelement <64 x i32> %38, i32 %35, i64 3, !dbg !16 + %40 = shufflevector <32 x i32> %31, <32 x i32> poison, <64 x i32> , !dbg !16 + %41 = or disjoint <64 x i32> %40, , !dbg !16 + %42 = shufflevector <64 x i32> %39, <64 x i32> %41, <64 x i32> , !dbg !16 + %43 = shufflevector <16 x i32> %32, <16 x i32> poison, <64 x i32> , !dbg !16 + %44 = or disjoint <64 x i32> %43, , !dbg !16 + %45 = shufflevector <64 x i32> %42, <64 x i32> %44, <64 x i32> , !dbg !16 + %46 = shufflevector <8 x i32> %33, <8 x i32> poison, <64 x i32> , !dbg !16 + %47 = or disjoint <64 x i32> %46, , !dbg !16 + %48 = shufflevector <64 x i32> %45, <64 x i32> %47, <64 x i32> , !dbg !16 + %49 = shufflevector <4 x i32> %34, <4 x i32> poison, <64 x i32> , !dbg !16 + %50 = or disjoint <64 x i32> %49, , !dbg !16 + %51 = shufflevector <64 x i32> %48, <64 x i32> %50, <64 x i32> , !dbg !16 + %52 = shufflevector <2 x i32> %37, <2 x i32> poison, <64 x i32> , !dbg !16 + %53 = or disjoint <64 x i32> %52, , !dbg !16 + %54 = shufflevector <64 x i32> %53, <64 x i32> %51, <64 x i32> , !dbg !16 + %55 = zext <64 x i32> %54 to <64 x i64>, !dbg !16 + %56 = insertelement <64 x i64> poison, i64 %11, i64 0, !dbg !16 + %57 = shufflevector <64 x i64> %56, <64 x i64> poison, <64 x i32> zeroinitializer, !dbg !16 + %58 = or disjoint <64 x i64> %57, %55, !dbg !16 + %59 = insertelement <64 x i32> poison, i32 %5, i64 0, !dbg !17 + %60 = shufflevector <64 x i32> %59, <64 x i32> poison, <64 x i32> zeroinitializer, !dbg !17 + %61 = sext <64 x i32> %60 to <64 x i64>, !dbg !17 + %62 = icmp slt <64 x i64> %58, %61, !dbg !17 + %63 = and i32 %12, 15, !dbg !18 + %64 = shl nuw nsw i32 %63, 3, !dbg !18 + %65 = shl nuw nsw i64 %22, 7, !dbg !19 + %66 = shl nuw nsw i64 %23, 7, !dbg !19 + %67 = shl nuw nsw i64 %24, 7, !dbg !19 + %68 = shl nuw nsw i64 %25, 7, !dbg !19 + %69 = extractelement <64 x i64> %58, i64 63, !dbg !19 + %70 = shl nuw nsw i64 %69, 7, !dbg !19 + %71 = extractelement <64 x i64> %58, i64 62, !dbg !19 + %72 = shl nuw nsw i64 %71, 7, !dbg !19 + %73 = extractelement <64 x i64> %58, i64 61, !dbg !19 + %74 = shl nuw nsw i64 %73, 7, !dbg !19 + %75 = extractelement <64 x i64> %58, i64 60, !dbg !19 + %76 = shl nuw nsw i64 %75, 7, !dbg !19 + %77 = extractelement <64 x i64> %58, i64 59, !dbg !19 + %78 = shl nuw nsw i64 %77, 7, !dbg !19 + %79 = extractelement <64 x i64> %58, i64 58, !dbg !19 + %80 = shl nuw nsw i64 %79, 7, !dbg !19 + %81 = extractelement <64 x i64> %58, i64 57, !dbg !19 + %82 = shl nuw nsw i64 %81, 7, !dbg !19 + %83 = extractelement <64 x i64> %58, i64 56, !dbg !19 + %84 = shl nuw nsw i64 %83, 7, !dbg !19 + %85 = extractelement <64 x i64> %58, i64 55, !dbg !19 + %86 = shl nuw nsw i64 %85, 7, !dbg !19 + %87 = extractelement <64 x i64> %58, i64 54, !dbg !19 + %88 = shl nuw nsw i64 %87, 7, !dbg !19 + %89 = extractelement <64 x i64> %58, i64 53, !dbg !19 + %90 = shl nuw nsw i64 %89, 7, !dbg !19 + %91 = extractelement <64 x i64> %58, i64 52, !dbg !19 + %92 = shl nuw nsw i64 %91, 7, !dbg !19 + %93 = extractelement <64 x i64> %58, i64 51, !dbg !19 + %94 = shl nuw nsw i64 %93, 7, !dbg !19 + %95 = extractelement <64 x i64> %58, i64 50, !dbg !19 + %96 = shl nuw nsw i64 %95, 7, !dbg !19 + %97 = extractelement <64 x i64> %58, i64 49, !dbg !19 + %98 = shl nuw nsw i64 %97, 7, !dbg !19 + %99 = extractelement <64 x i64> %58, i64 48, !dbg !19 + %100 = shl nuw nsw i64 %99, 7, !dbg !19 + %101 = extractelement <64 x i64> %58, i64 47, !dbg !19 + %102 = shl nuw nsw i64 %101, 7, !dbg !19 + %103 = extractelement <64 x i64> %58, i64 46, !dbg !19 + %104 = shl nuw nsw i64 %103, 7, !dbg !19 + %105 = extractelement <64 x i64> %58, i64 45, !dbg !19 + %106 = shl nuw nsw i64 %105, 7, !dbg !19 + %107 = extractelement <64 x i64> %58, i64 44, !dbg !19 + %108 = shl nuw nsw i64 %107, 7, !dbg !19 + %109 = extractelement <64 x i64> %58, i64 43, !dbg !19 + %110 = shl nuw nsw i64 %109, 7, !dbg !19 + %111 = extractelement <64 x i64> %58, i64 42, !dbg !19 + %112 = shl nuw nsw i64 %111, 7, !dbg !19 + %113 = extractelement <64 x i64> %58, i64 41, !dbg !19 + %114 = shl nuw nsw i64 %113, 7, !dbg !19 + %115 = extractelement <64 x i64> %58, i64 40, !dbg !19 + %116 = shl nuw nsw i64 %115, 7, !dbg !19 + %117 = extractelement <64 x i64> %58, i64 39, !dbg !19 + %118 = shl nuw nsw i64 %117, 7, !dbg !19 + %119 = extractelement <64 x i64> %58, i64 38, !dbg !19 + %120 = shl nuw nsw i64 %119, 7, !dbg !19 + %121 = extractelement <64 x i64> %58, i64 37, !dbg !19 + %122 = shl nuw nsw i64 %121, 7, !dbg !19 + %123 = extractelement <64 x i64> %58, i64 36, !dbg !19 + %124 = shl nuw nsw i64 %123, 7, !dbg !19 + %125 = extractelement <64 x i64> %58, i64 35, !dbg !19 + %126 = shl nuw nsw i64 %125, 7, !dbg !19 + %127 = extractelement <64 x i64> %58, i64 34, !dbg !19 + %128 = shl nuw nsw i64 %127, 7, !dbg !19 + %129 = extractelement <64 x i64> %58, i64 33, !dbg !19 + %130 = shl nuw nsw i64 %129, 7, !dbg !19 + %131 = extractelement <64 x i64> %58, i64 32, !dbg !19 + %132 = shl nuw nsw i64 %131, 7, !dbg !19 + %133 = extractelement <64 x i64> %58, i64 31, !dbg !19 + %134 = shl nuw nsw i64 %133, 7, !dbg !19 + %135 = extractelement <64 x i64> %58, i64 30, !dbg !19 + %136 = shl nuw nsw i64 %135, 7, !dbg !19 + %137 = extractelement <64 x i64> %58, i64 29, !dbg !19 + %138 = shl nuw nsw i64 %137, 7, !dbg !19 + %139 = extractelement <64 x i64> %58, i64 28, !dbg !19 + %140 = shl nuw nsw i64 %139, 7, !dbg !19 + %141 = extractelement <64 x i64> %58, i64 27, !dbg !19 + %142 = shl nuw nsw i64 %141, 7, !dbg !19 + %143 = extractelement <64 x i64> %58, i64 26, !dbg !19 + %144 = shl nuw nsw i64 %143, 7, !dbg !19 + %145 = extractelement <64 x i64> %58, i64 25, !dbg !19 + %146 = shl nuw nsw i64 %145, 7, !dbg !19 + %147 = extractelement <64 x i64> %58, i64 24, !dbg !19 + %148 = shl nuw nsw i64 %147, 7, !dbg !19 + %149 = extractelement <64 x i64> %58, i64 23, !dbg !19 + %150 = shl nuw nsw i64 %149, 7, !dbg !19 + %151 = extractelement <64 x i64> %58, i64 22, !dbg !19 + %152 = shl nuw nsw i64 %151, 7, !dbg !19 + %153 = extractelement <64 x i64> %58, i64 21, !dbg !19 + %154 = shl nuw nsw i64 %153, 7, !dbg !19 + %155 = extractelement <64 x i64> %58, i64 20, !dbg !19 + %156 = shl nuw nsw i64 %155, 7, !dbg !19 + %157 = extractelement <64 x i64> %58, i64 19, !dbg !19 + %158 = shl nuw nsw i64 %157, 7, !dbg !19 + %159 = extractelement <64 x i64> %58, i64 18, !dbg !19 + %160 = shl nuw nsw i64 %159, 7, !dbg !19 + %161 = extractelement <64 x i64> %58, i64 17, !dbg !19 + %162 = shl nuw nsw i64 %161, 7, !dbg !19 + %163 = extractelement <64 x i64> %58, i64 16, !dbg !19 + %164 = shl nuw nsw i64 %163, 7, !dbg !19 + %165 = extractelement <64 x i64> %58, i64 15, !dbg !19 + %166 = shl nuw nsw i64 %165, 7, !dbg !19 + %167 = extractelement <64 x i64> %58, i64 14, !dbg !19 + %168 = shl nuw nsw i64 %167, 7, !dbg !19 + %169 = extractelement <64 x i64> %58, i64 13, !dbg !19 + %170 = shl nuw nsw i64 %169, 7, !dbg !19 + %171 = extractelement <64 x i64> %58, i64 12, !dbg !19 + %172 = shl nuw nsw i64 %171, 7, !dbg !19 + %173 = extractelement <64 x i64> %58, i64 11, !dbg !19 + %174 = shl nuw nsw i64 %173, 7, !dbg !19 + %175 = extractelement <64 x i64> %58, i64 10, !dbg !19 + %176 = shl nuw nsw i64 %175, 7, !dbg !19 + %177 = extractelement <64 x i64> %58, i64 9, !dbg !19 + %178 = shl nuw nsw i64 %177, 7, !dbg !19 + %179 = extractelement <64 x i64> %58, i64 8, !dbg !19 + %180 = shl nuw nsw i64 %179, 7, !dbg !19 + %181 = extractelement <64 x i64> %58, i64 7, !dbg !19 + %182 = shl nuw nsw i64 %181, 7, !dbg !19 + %183 = extractelement <64 x i64> %58, i64 6, !dbg !19 + %184 = shl nuw nsw i64 %183, 7, !dbg !19 + %185 = extractelement <64 x i64> %58, i64 5, !dbg !19 + %186 = shl nuw nsw i64 %185, 7, !dbg !19 + %187 = extractelement <64 x i64> %58, i64 4, !dbg !19 + %188 = shl nuw nsw i64 %187, 7, !dbg !19 + %189 = getelementptr bfloat, ptr addrspace(1) %0, i64 %65, !dbg !20 + %190 = getelementptr bfloat, ptr addrspace(1) %0, i64 %66, !dbg !20 + %191 = getelementptr bfloat, ptr addrspace(1) %0, i64 %67, !dbg !20 + %192 = getelementptr bfloat, ptr addrspace(1) %0, i64 %68, !dbg !20 + %193 = getelementptr bfloat, ptr addrspace(1) %0, i64 %70, !dbg !20 + %194 = getelementptr bfloat, ptr addrspace(1) %0, i64 %72, !dbg !20 + %195 = getelementptr bfloat, ptr addrspace(1) %0, i64 %74, !dbg !20 + %196 = getelementptr bfloat, ptr addrspace(1) %0, i64 %76, !dbg !20 + %197 = getelementptr bfloat, ptr addrspace(1) %0, i64 %78, !dbg !20 + %198 = getelementptr bfloat, ptr addrspace(1) %0, i64 %80, !dbg !20 + %199 = getelementptr bfloat, ptr addrspace(1) %0, i64 %82, !dbg !20 + %200 = getelementptr bfloat, ptr addrspace(1) %0, i64 %84, !dbg !20 + %201 = getelementptr bfloat, ptr addrspace(1) %0, i64 %86, !dbg !20 + %202 = getelementptr bfloat, ptr addrspace(1) %0, i64 %88, !dbg !20 + %203 = getelementptr bfloat, ptr addrspace(1) %0, i64 %90, !dbg !20 + %204 = getelementptr bfloat, ptr addrspace(1) %0, i64 %92, !dbg !20 + %205 = getelementptr bfloat, ptr addrspace(1) %0, i64 %94, !dbg !20 + %206 = getelementptr bfloat, ptr addrspace(1) %0, i64 %96, !dbg !20 + %207 = getelementptr bfloat, ptr addrspace(1) %0, i64 %98, !dbg !20 + %208 = getelementptr bfloat, ptr addrspace(1) %0, i64 %100, !dbg !20 + %209 = getelementptr bfloat, ptr addrspace(1) %0, i64 %102, !dbg !20 + %210 = getelementptr bfloat, ptr addrspace(1) %0, i64 %104, !dbg !20 + %211 = getelementptr bfloat, ptr addrspace(1) %0, i64 %106, !dbg !20 + %212 = getelementptr bfloat, ptr addrspace(1) %0, i64 %108, !dbg !20 + %213 = getelementptr bfloat, ptr addrspace(1) %0, i64 %110, !dbg !20 + %214 = getelementptr bfloat, ptr addrspace(1) %0, i64 %112, !dbg !20 + %215 = getelementptr bfloat, ptr addrspace(1) %0, i64 %114, !dbg !20 + %216 = getelementptr bfloat, ptr addrspace(1) %0, i64 %116, !dbg !20 + %217 = getelementptr bfloat, ptr addrspace(1) %0, i64 %118, !dbg !20 + %218 = getelementptr bfloat, ptr addrspace(1) %0, i64 %120, !dbg !20 + %219 = getelementptr bfloat, ptr addrspace(1) %0, i64 %122, !dbg !20 + %220 = getelementptr bfloat, ptr addrspace(1) %0, i64 %124, !dbg !20 + %221 = getelementptr bfloat, ptr addrspace(1) %0, i64 %126, !dbg !20 + %222 = getelementptr bfloat, ptr addrspace(1) %0, i64 %128, !dbg !20 + %223 = getelementptr bfloat, ptr addrspace(1) %0, i64 %130, !dbg !20 + %224 = getelementptr bfloat, ptr addrspace(1) %0, i64 %132, !dbg !20 + %225 = getelementptr bfloat, ptr addrspace(1) %0, i64 %134, !dbg !20 + %226 = getelementptr bfloat, ptr addrspace(1) %0, i64 %136, !dbg !20 + %227 = getelementptr bfloat, ptr addrspace(1) %0, i64 %138, !dbg !20 + %228 = getelementptr bfloat, ptr addrspace(1) %0, i64 %140, !dbg !20 + %229 = getelementptr bfloat, ptr addrspace(1) %0, i64 %142, !dbg !20 + %230 = getelementptr bfloat, ptr addrspace(1) %0, i64 %144, !dbg !20 + %231 = getelementptr bfloat, ptr addrspace(1) %0, i64 %146, !dbg !20 + %232 = getelementptr bfloat, ptr addrspace(1) %0, i64 %148, !dbg !20 + %233 = getelementptr bfloat, ptr addrspace(1) %0, i64 %150, !dbg !20 + %234 = getelementptr bfloat, ptr addrspace(1) %0, i64 %152, !dbg !20 + %235 = getelementptr bfloat, ptr addrspace(1) %0, i64 %154, !dbg !20 + %236 = getelementptr bfloat, ptr addrspace(1) %0, i64 %156, !dbg !20 + %237 = getelementptr bfloat, ptr addrspace(1) %0, i64 %158, !dbg !20 + %238 = getelementptr bfloat, ptr addrspace(1) %0, i64 %160, !dbg !20 + %239 = getelementptr bfloat, ptr addrspace(1) %0, i64 %162, !dbg !20 + %240 = getelementptr bfloat, ptr addrspace(1) %0, i64 %164, !dbg !20 + %241 = getelementptr bfloat, ptr addrspace(1) %0, i64 %166, !dbg !20 + %242 = getelementptr bfloat, ptr addrspace(1) %0, i64 %168, !dbg !20 + %243 = getelementptr bfloat, ptr addrspace(1) %0, i64 %170, !dbg !20 + %244 = getelementptr bfloat, ptr addrspace(1) %0, i64 %172, !dbg !20 + %245 = getelementptr bfloat, ptr addrspace(1) %0, i64 %174, !dbg !20 + %246 = getelementptr bfloat, ptr addrspace(1) %0, i64 %176, !dbg !20 + %247 = getelementptr bfloat, ptr addrspace(1) %0, i64 %178, !dbg !20 + %248 = getelementptr bfloat, ptr addrspace(1) %0, i64 %180, !dbg !20 + %249 = getelementptr bfloat, ptr addrspace(1) %0, i64 %182, !dbg !20 + %250 = getelementptr bfloat, ptr addrspace(1) %0, i64 %184, !dbg !20 + %251 = getelementptr bfloat, ptr addrspace(1) %0, i64 %186, !dbg !20 + %252 = getelementptr bfloat, ptr addrspace(1) %0, i64 %188, !dbg !20 + %253 = zext nneg i32 %64 to i64, !dbg !21 + %254 = getelementptr bfloat, ptr addrspace(1) %189, i64 %253, !dbg !21 + %255 = getelementptr bfloat, ptr addrspace(1) %190, i64 %253, !dbg !21 + %256 = getelementptr bfloat, ptr addrspace(1) %191, i64 %253, !dbg !21 + %257 = getelementptr bfloat, ptr addrspace(1) %192, i64 %253, !dbg !21 + %258 = getelementptr bfloat, ptr addrspace(1) %193, i64 %253, !dbg !21 + %259 = getelementptr bfloat, ptr addrspace(1) %194, i64 %253, !dbg !21 + %260 = getelementptr bfloat, ptr addrspace(1) %195, i64 %253, !dbg !21 + %261 = getelementptr bfloat, ptr addrspace(1) %196, i64 %253, !dbg !21 + %262 = getelementptr bfloat, ptr addrspace(1) %197, i64 %253, !dbg !21 + %263 = getelementptr bfloat, ptr addrspace(1) %198, i64 %253, !dbg !21 + %264 = getelementptr bfloat, ptr addrspace(1) %199, i64 %253, !dbg !21 + %265 = getelementptr bfloat, ptr addrspace(1) %200, i64 %253, !dbg !21 + %266 = getelementptr bfloat, ptr addrspace(1) %201, i64 %253, !dbg !21 + %267 = getelementptr bfloat, ptr addrspace(1) %202, i64 %253, !dbg !21 + %268 = getelementptr bfloat, ptr addrspace(1) %203, i64 %253, !dbg !21 + %269 = getelementptr bfloat, ptr addrspace(1) %204, i64 %253, !dbg !21 + %270 = getelementptr bfloat, ptr addrspace(1) %205, i64 %253, !dbg !21 + %271 = getelementptr bfloat, ptr addrspace(1) %206, i64 %253, !dbg !21 + %272 = getelementptr bfloat, ptr addrspace(1) %207, i64 %253, !dbg !21 + %273 = getelementptr bfloat, ptr addrspace(1) %208, i64 %253, !dbg !21 + %274 = getelementptr bfloat, ptr addrspace(1) %209, i64 %253, !dbg !21 + %275 = getelementptr bfloat, ptr addrspace(1) %210, i64 %253, !dbg !21 + %276 = getelementptr bfloat, ptr addrspace(1) %211, i64 %253, !dbg !21 + %277 = getelementptr bfloat, ptr addrspace(1) %212, i64 %253, !dbg !21 + %278 = getelementptr bfloat, ptr addrspace(1) %213, i64 %253, !dbg !21 + %279 = getelementptr bfloat, ptr addrspace(1) %214, i64 %253, !dbg !21 + %280 = getelementptr bfloat, ptr addrspace(1) %215, i64 %253, !dbg !21 + %281 = getelementptr bfloat, ptr addrspace(1) %216, i64 %253, !dbg !21 + %282 = getelementptr bfloat, ptr addrspace(1) %217, i64 %253, !dbg !21 + %283 = getelementptr bfloat, ptr addrspace(1) %218, i64 %253, !dbg !21 + %284 = getelementptr bfloat, ptr addrspace(1) %219, i64 %253, !dbg !21 + %285 = getelementptr bfloat, ptr addrspace(1) %220, i64 %253, !dbg !21 + %286 = getelementptr bfloat, ptr addrspace(1) %221, i64 %253, !dbg !21 + %287 = getelementptr bfloat, ptr addrspace(1) %222, i64 %253, !dbg !21 + %288 = getelementptr bfloat, ptr addrspace(1) %223, i64 %253, !dbg !21 + %289 = getelementptr bfloat, ptr addrspace(1) %224, i64 %253, !dbg !21 + %290 = getelementptr bfloat, ptr addrspace(1) %225, i64 %253, !dbg !21 + %291 = getelementptr bfloat, ptr addrspace(1) %226, i64 %253, !dbg !21 + %292 = getelementptr bfloat, ptr addrspace(1) %227, i64 %253, !dbg !21 + %293 = getelementptr bfloat, ptr addrspace(1) %228, i64 %253, !dbg !21 + %294 = getelementptr bfloat, ptr addrspace(1) %229, i64 %253, !dbg !21 + %295 = getelementptr bfloat, ptr addrspace(1) %230, i64 %253, !dbg !21 + %296 = getelementptr bfloat, ptr addrspace(1) %231, i64 %253, !dbg !21 + %297 = getelementptr bfloat, ptr addrspace(1) %232, i64 %253, !dbg !21 + %298 = getelementptr bfloat, ptr addrspace(1) %233, i64 %253, !dbg !21 + %299 = getelementptr bfloat, ptr addrspace(1) %234, i64 %253, !dbg !21 + %300 = getelementptr bfloat, ptr addrspace(1) %235, i64 %253, !dbg !21 + %301 = getelementptr bfloat, ptr addrspace(1) %236, i64 %253, !dbg !21 + %302 = getelementptr bfloat, ptr addrspace(1) %237, i64 %253, !dbg !21 + %303 = getelementptr bfloat, ptr addrspace(1) %238, i64 %253, !dbg !21 + %304 = getelementptr bfloat, ptr addrspace(1) %239, i64 %253, !dbg !21 + %305 = getelementptr bfloat, ptr addrspace(1) %240, i64 %253, !dbg !21 + %306 = getelementptr bfloat, ptr addrspace(1) %241, i64 %253, !dbg !21 + %307 = getelementptr bfloat, ptr addrspace(1) %242, i64 %253, !dbg !21 + %308 = getelementptr bfloat, ptr addrspace(1) %243, i64 %253, !dbg !21 + %309 = getelementptr bfloat, ptr addrspace(1) %244, i64 %253, !dbg !21 + %310 = getelementptr bfloat, ptr addrspace(1) %245, i64 %253, !dbg !21 + %311 = getelementptr bfloat, ptr addrspace(1) %246, i64 %253, !dbg !21 + %312 = getelementptr bfloat, ptr addrspace(1) %247, i64 %253, !dbg !21 + %313 = getelementptr bfloat, ptr addrspace(1) %248, i64 %253, !dbg !21 + %314 = getelementptr bfloat, ptr addrspace(1) %249, i64 %253, !dbg !21 + %315 = getelementptr bfloat, ptr addrspace(1) %250, i64 %253, !dbg !21 + %316 = getelementptr bfloat, ptr addrspace(1) %251, i64 %253, !dbg !21 + %317 = getelementptr bfloat, ptr addrspace(1) %252, i64 %253, !dbg !21 + %318 = extractelement <64 x i64> %58, i64 3, !dbg !22 + %319 = getelementptr float, ptr addrspace(1) %1, i64 %318, !dbg !22 + %320 = extractelement <64 x i64> %58, i64 2, !dbg !22 + %321 = getelementptr float, ptr addrspace(1) %1, i64 %320, !dbg !22 + %322 = extractelement <64 x i64> %58, i64 1, !dbg !22 + %323 = getelementptr float, ptr addrspace(1) %1, i64 %322, !dbg !22 + %324 = extractelement <64 x i64> %58, i64 0, !dbg !22 + %325 = getelementptr float, ptr addrspace(1) %1, i64 %324, !dbg !22 + %326 = getelementptr bfloat, ptr addrspace(1) %2, i64 %65, !dbg !23 + %327 = getelementptr bfloat, ptr addrspace(1) %2, i64 %66, !dbg !23 + %328 = getelementptr bfloat, ptr addrspace(1) %2, i64 %67, !dbg !23 + %329 = getelementptr bfloat, ptr addrspace(1) %2, i64 %68, !dbg !23 + %330 = getelementptr bfloat, ptr addrspace(1) %2, i64 %70, !dbg !23 + %331 = getelementptr bfloat, ptr addrspace(1) %2, i64 %72, !dbg !23 + %332 = getelementptr bfloat, ptr addrspace(1) %2, i64 %74, !dbg !23 + %333 = getelementptr bfloat, ptr addrspace(1) %2, i64 %76, !dbg !23 + %334 = getelementptr bfloat, ptr addrspace(1) %2, i64 %78, !dbg !23 + %335 = getelementptr bfloat, ptr addrspace(1) %2, i64 %80, !dbg !23 + %336 = getelementptr bfloat, ptr addrspace(1) %2, i64 %82, !dbg !23 + %337 = getelementptr bfloat, ptr addrspace(1) %2, i64 %84, !dbg !23 + %338 = getelementptr bfloat, ptr addrspace(1) %2, i64 %86, !dbg !23 + %339 = getelementptr bfloat, ptr addrspace(1) %2, i64 %88, !dbg !23 + %340 = getelementptr bfloat, ptr addrspace(1) %2, i64 %90, !dbg !23 + %341 = getelementptr bfloat, ptr addrspace(1) %2, i64 %92, !dbg !23 + %342 = getelementptr bfloat, ptr addrspace(1) %2, i64 %94, !dbg !23 + %343 = getelementptr bfloat, ptr addrspace(1) %2, i64 %96, !dbg !23 + %344 = getelementptr bfloat, ptr addrspace(1) %2, i64 %98, !dbg !23 + %345 = getelementptr bfloat, ptr addrspace(1) %2, i64 %100, !dbg !23 + %346 = getelementptr bfloat, ptr addrspace(1) %2, i64 %102, !dbg !23 + %347 = getelementptr bfloat, ptr addrspace(1) %2, i64 %104, !dbg !23 + %348 = getelementptr bfloat, ptr addrspace(1) %2, i64 %106, !dbg !23 + %349 = getelementptr bfloat, ptr addrspace(1) %2, i64 %108, !dbg !23 + %350 = getelementptr bfloat, ptr addrspace(1) %2, i64 %110, !dbg !23 + %351 = getelementptr bfloat, ptr addrspace(1) %2, i64 %112, !dbg !23 + %352 = getelementptr bfloat, ptr addrspace(1) %2, i64 %114, !dbg !23 + %353 = getelementptr bfloat, ptr addrspace(1) %2, i64 %116, !dbg !23 + %354 = getelementptr bfloat, ptr addrspace(1) %2, i64 %118, !dbg !23 + %355 = getelementptr bfloat, ptr addrspace(1) %2, i64 %120, !dbg !23 + %356 = getelementptr bfloat, ptr addrspace(1) %2, i64 %122, !dbg !23 + %357 = getelementptr bfloat, ptr addrspace(1) %2, i64 %124, !dbg !23 + %358 = getelementptr bfloat, ptr addrspace(1) %2, i64 %126, !dbg !23 + %359 = getelementptr bfloat, ptr addrspace(1) %2, i64 %128, !dbg !23 + %360 = getelementptr bfloat, ptr addrspace(1) %2, i64 %130, !dbg !23 + %361 = getelementptr bfloat, ptr addrspace(1) %2, i64 %132, !dbg !23 + %362 = getelementptr bfloat, ptr addrspace(1) %2, i64 %134, !dbg !23 + %363 = getelementptr bfloat, ptr addrspace(1) %2, i64 %136, !dbg !23 + %364 = getelementptr bfloat, ptr addrspace(1) %2, i64 %138, !dbg !23 + %365 = getelementptr bfloat, ptr addrspace(1) %2, i64 %140, !dbg !23 + %366 = getelementptr bfloat, ptr addrspace(1) %2, i64 %142, !dbg !23 + %367 = getelementptr bfloat, ptr addrspace(1) %2, i64 %144, !dbg !23 + %368 = getelementptr bfloat, ptr addrspace(1) %2, i64 %146, !dbg !23 + %369 = getelementptr bfloat, ptr addrspace(1) %2, i64 %148, !dbg !23 + %370 = getelementptr bfloat, ptr addrspace(1) %2, i64 %150, !dbg !23 + %371 = getelementptr bfloat, ptr addrspace(1) %2, i64 %152, !dbg !23 + %372 = getelementptr bfloat, ptr addrspace(1) %2, i64 %154, !dbg !23 + %373 = getelementptr bfloat, ptr addrspace(1) %2, i64 %156, !dbg !23 + %374 = getelementptr bfloat, ptr addrspace(1) %2, i64 %158, !dbg !23 + %375 = getelementptr bfloat, ptr addrspace(1) %2, i64 %160, !dbg !23 + %376 = getelementptr bfloat, ptr addrspace(1) %2, i64 %162, !dbg !23 + %377 = getelementptr bfloat, ptr addrspace(1) %2, i64 %164, !dbg !23 + %378 = getelementptr bfloat, ptr addrspace(1) %2, i64 %166, !dbg !23 + %379 = getelementptr bfloat, ptr addrspace(1) %2, i64 %168, !dbg !23 + %380 = getelementptr bfloat, ptr addrspace(1) %2, i64 %170, !dbg !23 + %381 = getelementptr bfloat, ptr addrspace(1) %2, i64 %172, !dbg !23 + %382 = getelementptr bfloat, ptr addrspace(1) %2, i64 %174, !dbg !23 + %383 = getelementptr bfloat, ptr addrspace(1) %2, i64 %176, !dbg !23 + %384 = getelementptr bfloat, ptr addrspace(1) %2, i64 %178, !dbg !23 + %385 = getelementptr bfloat, ptr addrspace(1) %2, i64 %180, !dbg !23 + %386 = getelementptr bfloat, ptr addrspace(1) %2, i64 %182, !dbg !23 + %387 = getelementptr bfloat, ptr addrspace(1) %2, i64 %184, !dbg !23 + %388 = getelementptr bfloat, ptr addrspace(1) %2, i64 %186, !dbg !23 + %389 = getelementptr bfloat, ptr addrspace(1) %2, i64 %188, !dbg !23 + %390 = getelementptr bfloat, ptr addrspace(1) %326, i64 %253, !dbg !24 + %391 = getelementptr bfloat, ptr addrspace(1) %327, i64 %253, !dbg !24 + %392 = getelementptr bfloat, ptr addrspace(1) %328, i64 %253, !dbg !24 + %393 = getelementptr bfloat, ptr addrspace(1) %329, i64 %253, !dbg !24 + %394 = getelementptr bfloat, ptr addrspace(1) %330, i64 %253, !dbg !24 + %395 = getelementptr bfloat, ptr addrspace(1) %331, i64 %253, !dbg !24 + %396 = getelementptr bfloat, ptr addrspace(1) %332, i64 %253, !dbg !24 + %397 = getelementptr bfloat, ptr addrspace(1) %333, i64 %253, !dbg !24 + %398 = getelementptr bfloat, ptr addrspace(1) %334, i64 %253, !dbg !24 + %399 = getelementptr bfloat, ptr addrspace(1) %335, i64 %253, !dbg !24 + %400 = getelementptr bfloat, ptr addrspace(1) %336, i64 %253, !dbg !24 + %401 = getelementptr bfloat, ptr addrspace(1) %337, i64 %253, !dbg !24 + %402 = getelementptr bfloat, ptr addrspace(1) %338, i64 %253, !dbg !24 + %403 = getelementptr bfloat, ptr addrspace(1) %339, i64 %253, !dbg !24 + %404 = getelementptr bfloat, ptr addrspace(1) %340, i64 %253, !dbg !24 + %405 = getelementptr bfloat, ptr addrspace(1) %341, i64 %253, !dbg !24 + %406 = getelementptr bfloat, ptr addrspace(1) %342, i64 %253, !dbg !24 + %407 = getelementptr bfloat, ptr addrspace(1) %343, i64 %253, !dbg !24 + %408 = getelementptr bfloat, ptr addrspace(1) %344, i64 %253, !dbg !24 + %409 = getelementptr bfloat, ptr addrspace(1) %345, i64 %253, !dbg !24 + %410 = getelementptr bfloat, ptr addrspace(1) %346, i64 %253, !dbg !24 + %411 = getelementptr bfloat, ptr addrspace(1) %347, i64 %253, !dbg !24 + %412 = getelementptr bfloat, ptr addrspace(1) %348, i64 %253, !dbg !24 + %413 = getelementptr bfloat, ptr addrspace(1) %349, i64 %253, !dbg !24 + %414 = getelementptr bfloat, ptr addrspace(1) %350, i64 %253, !dbg !24 + %415 = getelementptr bfloat, ptr addrspace(1) %351, i64 %253, !dbg !24 + %416 = getelementptr bfloat, ptr addrspace(1) %352, i64 %253, !dbg !24 + %417 = getelementptr bfloat, ptr addrspace(1) %353, i64 %253, !dbg !24 + %418 = getelementptr bfloat, ptr addrspace(1) %354, i64 %253, !dbg !24 + %419 = getelementptr bfloat, ptr addrspace(1) %355, i64 %253, !dbg !24 + %420 = getelementptr bfloat, ptr addrspace(1) %356, i64 %253, !dbg !24 + %421 = getelementptr bfloat, ptr addrspace(1) %357, i64 %253, !dbg !24 + %422 = getelementptr bfloat, ptr addrspace(1) %358, i64 %253, !dbg !24 + %423 = getelementptr bfloat, ptr addrspace(1) %359, i64 %253, !dbg !24 + %424 = getelementptr bfloat, ptr addrspace(1) %360, i64 %253, !dbg !24 + %425 = getelementptr bfloat, ptr addrspace(1) %361, i64 %253, !dbg !24 + %426 = getelementptr bfloat, ptr addrspace(1) %362, i64 %253, !dbg !24 + %427 = getelementptr bfloat, ptr addrspace(1) %363, i64 %253, !dbg !24 + %428 = getelementptr bfloat, ptr addrspace(1) %364, i64 %253, !dbg !24 + %429 = getelementptr bfloat, ptr addrspace(1) %365, i64 %253, !dbg !24 + %430 = getelementptr bfloat, ptr addrspace(1) %366, i64 %253, !dbg !24 + %431 = getelementptr bfloat, ptr addrspace(1) %367, i64 %253, !dbg !24 + %432 = getelementptr bfloat, ptr addrspace(1) %368, i64 %253, !dbg !24 + %433 = getelementptr bfloat, ptr addrspace(1) %369, i64 %253, !dbg !24 + %434 = getelementptr bfloat, ptr addrspace(1) %370, i64 %253, !dbg !24 + %435 = getelementptr bfloat, ptr addrspace(1) %371, i64 %253, !dbg !24 + %436 = getelementptr bfloat, ptr addrspace(1) %372, i64 %253, !dbg !24 + %437 = getelementptr bfloat, ptr addrspace(1) %373, i64 %253, !dbg !24 + %438 = getelementptr bfloat, ptr addrspace(1) %374, i64 %253, !dbg !24 + %439 = getelementptr bfloat, ptr addrspace(1) %375, i64 %253, !dbg !24 + %440 = getelementptr bfloat, ptr addrspace(1) %376, i64 %253, !dbg !24 + %441 = getelementptr bfloat, ptr addrspace(1) %377, i64 %253, !dbg !24 + %442 = getelementptr bfloat, ptr addrspace(1) %378, i64 %253, !dbg !24 + %443 = getelementptr bfloat, ptr addrspace(1) %379, i64 %253, !dbg !24 + %444 = getelementptr bfloat, ptr addrspace(1) %380, i64 %253, !dbg !24 + %445 = getelementptr bfloat, ptr addrspace(1) %381, i64 %253, !dbg !24 + %446 = getelementptr bfloat, ptr addrspace(1) %382, i64 %253, !dbg !24 + %447 = getelementptr bfloat, ptr addrspace(1) %383, i64 %253, !dbg !24 + %448 = getelementptr bfloat, ptr addrspace(1) %384, i64 %253, !dbg !24 + %449 = getelementptr bfloat, ptr addrspace(1) %385, i64 %253, !dbg !24 + %450 = getelementptr bfloat, ptr addrspace(1) %386, i64 %253, !dbg !24 + %451 = getelementptr bfloat, ptr addrspace(1) %387, i64 %253, !dbg !24 + %452 = getelementptr bfloat, ptr addrspace(1) %388, i64 %253, !dbg !24 + %453 = getelementptr bfloat, ptr addrspace(1) %389, i64 %253, !dbg !24 + %454 = getelementptr bfloat, ptr addrspace(1) %3, i64 %65, !dbg !25 + %455 = getelementptr bfloat, ptr addrspace(1) %3, i64 %66, !dbg !25 + %456 = getelementptr bfloat, ptr addrspace(1) %3, i64 %67, !dbg !25 + %457 = getelementptr bfloat, ptr addrspace(1) %3, i64 %68, !dbg !25 + %458 = getelementptr bfloat, ptr addrspace(1) %3, i64 %70, !dbg !25 + %459 = getelementptr bfloat, ptr addrspace(1) %3, i64 %72, !dbg !25 + %460 = getelementptr bfloat, ptr addrspace(1) %3, i64 %74, !dbg !25 + %461 = getelementptr bfloat, ptr addrspace(1) %3, i64 %76, !dbg !25 + %462 = getelementptr bfloat, ptr addrspace(1) %3, i64 %78, !dbg !25 + %463 = getelementptr bfloat, ptr addrspace(1) %3, i64 %80, !dbg !25 + %464 = getelementptr bfloat, ptr addrspace(1) %3, i64 %82, !dbg !25 + %465 = getelementptr bfloat, ptr addrspace(1) %3, i64 %84, !dbg !25 + %466 = getelementptr bfloat, ptr addrspace(1) %3, i64 %86, !dbg !25 + %467 = getelementptr bfloat, ptr addrspace(1) %3, i64 %88, !dbg !25 + %468 = getelementptr bfloat, ptr addrspace(1) %3, i64 %90, !dbg !25 + %469 = getelementptr bfloat, ptr addrspace(1) %3, i64 %92, !dbg !25 + %470 = getelementptr bfloat, ptr addrspace(1) %3, i64 %94, !dbg !25 + %471 = getelementptr bfloat, ptr addrspace(1) %3, i64 %96, !dbg !25 + %472 = getelementptr bfloat, ptr addrspace(1) %3, i64 %98, !dbg !25 + %473 = getelementptr bfloat, ptr addrspace(1) %3, i64 %100, !dbg !25 + %474 = getelementptr bfloat, ptr addrspace(1) %3, i64 %102, !dbg !25 + %475 = getelementptr bfloat, ptr addrspace(1) %3, i64 %104, !dbg !25 + %476 = getelementptr bfloat, ptr addrspace(1) %3, i64 %106, !dbg !25 + %477 = getelementptr bfloat, ptr addrspace(1) %3, i64 %108, !dbg !25 + %478 = getelementptr bfloat, ptr addrspace(1) %3, i64 %110, !dbg !25 + %479 = getelementptr bfloat, ptr addrspace(1) %3, i64 %112, !dbg !25 + %480 = getelementptr bfloat, ptr addrspace(1) %3, i64 %114, !dbg !25 + %481 = getelementptr bfloat, ptr addrspace(1) %3, i64 %116, !dbg !25 + %482 = getelementptr bfloat, ptr addrspace(1) %3, i64 %118, !dbg !25 + %483 = getelementptr bfloat, ptr addrspace(1) %3, i64 %120, !dbg !25 + %484 = getelementptr bfloat, ptr addrspace(1) %3, i64 %122, !dbg !25 + %485 = getelementptr bfloat, ptr addrspace(1) %3, i64 %124, !dbg !25 + %486 = getelementptr bfloat, ptr addrspace(1) %3, i64 %126, !dbg !25 + %487 = getelementptr bfloat, ptr addrspace(1) %3, i64 %128, !dbg !25 + %488 = getelementptr bfloat, ptr addrspace(1) %3, i64 %130, !dbg !25 + %489 = getelementptr bfloat, ptr addrspace(1) %3, i64 %132, !dbg !25 + %490 = getelementptr bfloat, ptr addrspace(1) %3, i64 %134, !dbg !25 + %491 = getelementptr bfloat, ptr addrspace(1) %3, i64 %136, !dbg !25 + %492 = getelementptr bfloat, ptr addrspace(1) %3, i64 %138, !dbg !25 + %493 = getelementptr bfloat, ptr addrspace(1) %3, i64 %140, !dbg !25 + %494 = getelementptr bfloat, ptr addrspace(1) %3, i64 %142, !dbg !25 + %495 = getelementptr bfloat, ptr addrspace(1) %3, i64 %144, !dbg !25 + %496 = getelementptr bfloat, ptr addrspace(1) %3, i64 %146, !dbg !25 + %497 = getelementptr bfloat, ptr addrspace(1) %3, i64 %148, !dbg !25 + %498 = getelementptr bfloat, ptr addrspace(1) %3, i64 %150, !dbg !25 + %499 = getelementptr bfloat, ptr addrspace(1) %3, i64 %152, !dbg !25 + %500 = getelementptr bfloat, ptr addrspace(1) %3, i64 %154, !dbg !25 + %501 = getelementptr bfloat, ptr addrspace(1) %3, i64 %156, !dbg !25 + %502 = getelementptr bfloat, ptr addrspace(1) %3, i64 %158, !dbg !25 + %503 = getelementptr bfloat, ptr addrspace(1) %3, i64 %160, !dbg !25 + %504 = getelementptr bfloat, ptr addrspace(1) %3, i64 %162, !dbg !25 + %505 = getelementptr bfloat, ptr addrspace(1) %3, i64 %164, !dbg !25 + %506 = getelementptr bfloat, ptr addrspace(1) %3, i64 %166, !dbg !25 + %507 = getelementptr bfloat, ptr addrspace(1) %3, i64 %168, !dbg !25 + %508 = getelementptr bfloat, ptr addrspace(1) %3, i64 %170, !dbg !25 + %509 = getelementptr bfloat, ptr addrspace(1) %3, i64 %172, !dbg !25 + %510 = getelementptr bfloat, ptr addrspace(1) %3, i64 %174, !dbg !25 + %511 = getelementptr bfloat, ptr addrspace(1) %3, i64 %176, !dbg !25 + %512 = getelementptr bfloat, ptr addrspace(1) %3, i64 %178, !dbg !25 + %513 = getelementptr bfloat, ptr addrspace(1) %3, i64 %180, !dbg !25 + %514 = getelementptr bfloat, ptr addrspace(1) %3, i64 %182, !dbg !25 + %515 = getelementptr bfloat, ptr addrspace(1) %3, i64 %184, !dbg !25 + %516 = getelementptr bfloat, ptr addrspace(1) %3, i64 %186, !dbg !25 + %517 = getelementptr bfloat, ptr addrspace(1) %3, i64 %188, !dbg !25 + %518 = getelementptr bfloat, ptr addrspace(1) %454, i64 %253, !dbg !26 + %519 = getelementptr bfloat, ptr addrspace(1) %455, i64 %253, !dbg !26 + %520 = getelementptr bfloat, ptr addrspace(1) %456, i64 %253, !dbg !26 + %521 = getelementptr bfloat, ptr addrspace(1) %457, i64 %253, !dbg !26 + %522 = getelementptr bfloat, ptr addrspace(1) %458, i64 %253, !dbg !26 + %523 = getelementptr bfloat, ptr addrspace(1) %459, i64 %253, !dbg !26 + %524 = getelementptr bfloat, ptr addrspace(1) %460, i64 %253, !dbg !26 + %525 = getelementptr bfloat, ptr addrspace(1) %461, i64 %253, !dbg !26 + %526 = getelementptr bfloat, ptr addrspace(1) %462, i64 %253, !dbg !26 + %527 = getelementptr bfloat, ptr addrspace(1) %463, i64 %253, !dbg !26 + %528 = getelementptr bfloat, ptr addrspace(1) %464, i64 %253, !dbg !26 + %529 = getelementptr bfloat, ptr addrspace(1) %465, i64 %253, !dbg !26 + %530 = getelementptr bfloat, ptr addrspace(1) %466, i64 %253, !dbg !26 + %531 = getelementptr bfloat, ptr addrspace(1) %467, i64 %253, !dbg !26 + %532 = getelementptr bfloat, ptr addrspace(1) %468, i64 %253, !dbg !26 + %533 = getelementptr bfloat, ptr addrspace(1) %469, i64 %253, !dbg !26 + %534 = getelementptr bfloat, ptr addrspace(1) %470, i64 %253, !dbg !26 + %535 = getelementptr bfloat, ptr addrspace(1) %471, i64 %253, !dbg !26 + %536 = getelementptr bfloat, ptr addrspace(1) %472, i64 %253, !dbg !26 + %537 = getelementptr bfloat, ptr addrspace(1) %473, i64 %253, !dbg !26 + %538 = getelementptr bfloat, ptr addrspace(1) %474, i64 %253, !dbg !26 + %539 = getelementptr bfloat, ptr addrspace(1) %475, i64 %253, !dbg !26 + %540 = getelementptr bfloat, ptr addrspace(1) %476, i64 %253, !dbg !26 + %541 = getelementptr bfloat, ptr addrspace(1) %477, i64 %253, !dbg !26 + %542 = getelementptr bfloat, ptr addrspace(1) %478, i64 %253, !dbg !26 + %543 = getelementptr bfloat, ptr addrspace(1) %479, i64 %253, !dbg !26 + %544 = getelementptr bfloat, ptr addrspace(1) %480, i64 %253, !dbg !26 + %545 = getelementptr bfloat, ptr addrspace(1) %481, i64 %253, !dbg !26 + %546 = getelementptr bfloat, ptr addrspace(1) %482, i64 %253, !dbg !26 + %547 = getelementptr bfloat, ptr addrspace(1) %483, i64 %253, !dbg !26 + %548 = getelementptr bfloat, ptr addrspace(1) %484, i64 %253, !dbg !26 + %549 = getelementptr bfloat, ptr addrspace(1) %485, i64 %253, !dbg !26 + %550 = getelementptr bfloat, ptr addrspace(1) %486, i64 %253, !dbg !26 + %551 = getelementptr bfloat, ptr addrspace(1) %487, i64 %253, !dbg !26 + %552 = getelementptr bfloat, ptr addrspace(1) %488, i64 %253, !dbg !26 + %553 = getelementptr bfloat, ptr addrspace(1) %489, i64 %253, !dbg !26 + %554 = getelementptr bfloat, ptr addrspace(1) %490, i64 %253, !dbg !26 + %555 = getelementptr bfloat, ptr addrspace(1) %491, i64 %253, !dbg !26 + %556 = getelementptr bfloat, ptr addrspace(1) %492, i64 %253, !dbg !26 + %557 = getelementptr bfloat, ptr addrspace(1) %493, i64 %253, !dbg !26 + %558 = getelementptr bfloat, ptr addrspace(1) %494, i64 %253, !dbg !26 + %559 = getelementptr bfloat, ptr addrspace(1) %495, i64 %253, !dbg !26 + %560 = getelementptr bfloat, ptr addrspace(1) %496, i64 %253, !dbg !26 + %561 = getelementptr bfloat, ptr addrspace(1) %497, i64 %253, !dbg !26 + %562 = getelementptr bfloat, ptr addrspace(1) %498, i64 %253, !dbg !26 + %563 = getelementptr bfloat, ptr addrspace(1) %499, i64 %253, !dbg !26 + %564 = getelementptr bfloat, ptr addrspace(1) %500, i64 %253, !dbg !26 + %565 = getelementptr bfloat, ptr addrspace(1) %501, i64 %253, !dbg !26 + %566 = getelementptr bfloat, ptr addrspace(1) %502, i64 %253, !dbg !26 + %567 = getelementptr bfloat, ptr addrspace(1) %503, i64 %253, !dbg !26 + %568 = getelementptr bfloat, ptr addrspace(1) %504, i64 %253, !dbg !26 + %569 = getelementptr bfloat, ptr addrspace(1) %505, i64 %253, !dbg !26 + %570 = getelementptr bfloat, ptr addrspace(1) %506, i64 %253, !dbg !26 + %571 = getelementptr bfloat, ptr addrspace(1) %507, i64 %253, !dbg !26 + %572 = getelementptr bfloat, ptr addrspace(1) %508, i64 %253, !dbg !26 + %573 = getelementptr bfloat, ptr addrspace(1) %509, i64 %253, !dbg !26 + %574 = getelementptr bfloat, ptr addrspace(1) %510, i64 %253, !dbg !26 + %575 = getelementptr bfloat, ptr addrspace(1) %511, i64 %253, !dbg !26 + %576 = getelementptr bfloat, ptr addrspace(1) %512, i64 %253, !dbg !26 + %577 = getelementptr bfloat, ptr addrspace(1) %513, i64 %253, !dbg !26 + %578 = getelementptr bfloat, ptr addrspace(1) %514, i64 %253, !dbg !26 + %579 = getelementptr bfloat, ptr addrspace(1) %515, i64 %253, !dbg !26 + %580 = getelementptr bfloat, ptr addrspace(1) %516, i64 %253, !dbg !26 + %581 = getelementptr bfloat, ptr addrspace(1) %517, i64 %253, !dbg !26 + %582 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %254, i1 %27) #5, !dbg !27 + %583 = extractvalue { i32, i32, i32, i32 } %582, 0, !dbg !27 + %584 = bitcast i32 %583 to <2 x bfloat>, !dbg !27 + %585 = extractvalue { i32, i32, i32, i32 } %582, 1, !dbg !27 + %586 = bitcast i32 %585 to <2 x bfloat>, !dbg !27 + %587 = extractvalue { i32, i32, i32, i32 } %582, 2, !dbg !27 + %588 = bitcast i32 %587 to <2 x bfloat>, !dbg !27 + %589 = extractvalue { i32, i32, i32, i32 } %582, 3, !dbg !27 + %590 = bitcast i32 %589 to <2 x bfloat>, !dbg !27 + %591 = extractelement <2 x bfloat> %584, i64 0, !dbg !27 + %592 = extractelement <2 x bfloat> %584, i64 1, !dbg !27 + %593 = extractelement <2 x bfloat> %586, i64 0, !dbg !27 + %594 = extractelement <2 x bfloat> %586, i64 1, !dbg !27 + %595 = extractelement <2 x bfloat> %588, i64 0, !dbg !27 + %596 = extractelement <2 x bfloat> %588, i64 1, !dbg !27 + %597 = extractelement <2 x bfloat> %590, i64 0, !dbg !27 + %598 = extractelement <2 x bfloat> %590, i64 1, !dbg !27 + %599 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %255, i1 %28) #5, !dbg !27 + %600 = extractvalue { i32, i32, i32, i32 } %599, 0, !dbg !27 + %601 = bitcast i32 %600 to <2 x bfloat>, !dbg !27 + %602 = extractvalue { i32, i32, i32, i32 } %599, 1, !dbg !27 + %603 = bitcast i32 %602 to <2 x bfloat>, !dbg !27 + %604 = extractvalue { i32, i32, i32, i32 } %599, 2, !dbg !27 + %605 = bitcast i32 %604 to <2 x bfloat>, !dbg !27 + %606 = extractvalue { i32, i32, i32, i32 } %599, 3, !dbg !27 + %607 = bitcast i32 %606 to <2 x bfloat>, !dbg !27 + %608 = extractelement <2 x bfloat> %601, i64 0, !dbg !27 + %609 = extractelement <2 x bfloat> %601, i64 1, !dbg !27 + %610 = extractelement <2 x bfloat> %603, i64 0, !dbg !27 + %611 = extractelement <2 x bfloat> %603, i64 1, !dbg !27 + %612 = extractelement <2 x bfloat> %605, i64 0, !dbg !27 + %613 = extractelement <2 x bfloat> %605, i64 1, !dbg !27 + %614 = extractelement <2 x bfloat> %607, i64 0, !dbg !27 + %615 = extractelement <2 x bfloat> %607, i64 1, !dbg !27 + %616 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %256, i1 %29) #5, !dbg !27 + %617 = extractvalue { i32, i32, i32, i32 } %616, 0, !dbg !27 + %618 = bitcast i32 %617 to <2 x bfloat>, !dbg !27 + %619 = extractvalue { i32, i32, i32, i32 } %616, 1, !dbg !27 + %620 = bitcast i32 %619 to <2 x bfloat>, !dbg !27 + %621 = extractvalue { i32, i32, i32, i32 } %616, 2, !dbg !27 + %622 = bitcast i32 %621 to <2 x bfloat>, !dbg !27 + %623 = extractvalue { i32, i32, i32, i32 } %616, 3, !dbg !27 + %624 = bitcast i32 %623 to <2 x bfloat>, !dbg !27 + %625 = extractelement <2 x bfloat> %618, i64 0, !dbg !27 + %626 = extractelement <2 x bfloat> %618, i64 1, !dbg !27 + %627 = extractelement <2 x bfloat> %620, i64 0, !dbg !27 + %628 = extractelement <2 x bfloat> %620, i64 1, !dbg !27 + %629 = extractelement <2 x bfloat> %622, i64 0, !dbg !27 + %630 = extractelement <2 x bfloat> %622, i64 1, !dbg !27 + %631 = extractelement <2 x bfloat> %624, i64 0, !dbg !27 + %632 = extractelement <2 x bfloat> %624, i64 1, !dbg !27 + %633 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %257, i1 %30) #5, !dbg !27 + %634 = extractvalue { i32, i32, i32, i32 } %633, 0, !dbg !27 + %635 = bitcast i32 %634 to <2 x bfloat>, !dbg !27 + %636 = extractvalue { i32, i32, i32, i32 } %633, 1, !dbg !27 + %637 = bitcast i32 %636 to <2 x bfloat>, !dbg !27 + %638 = extractvalue { i32, i32, i32, i32 } %633, 2, !dbg !27 + %639 = bitcast i32 %638 to <2 x bfloat>, !dbg !27 + %640 = extractvalue { i32, i32, i32, i32 } %633, 3, !dbg !27 + %641 = bitcast i32 %640 to <2 x bfloat>, !dbg !27 + %642 = extractelement <2 x bfloat> %635, i64 0, !dbg !27 + %643 = extractelement <2 x bfloat> %635, i64 1, !dbg !27 + %644 = extractelement <2 x bfloat> %637, i64 0, !dbg !27 + %645 = extractelement <2 x bfloat> %637, i64 1, !dbg !27 + %646 = extractelement <2 x bfloat> %639, i64 0, !dbg !27 + %647 = extractelement <2 x bfloat> %639, i64 1, !dbg !27 + %648 = extractelement <2 x bfloat> %641, i64 0, !dbg !27 + %649 = extractelement <2 x bfloat> %641, i64 1, !dbg !27 + %650 = extractelement <64 x i1> %62, i64 63, !dbg !27 + %651 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %258, i1 %650) #5, !dbg !27 + %652 = extractvalue { i32, i32, i32, i32 } %651, 0, !dbg !27 + %653 = bitcast i32 %652 to <2 x bfloat>, !dbg !27 + %654 = extractvalue { i32, i32, i32, i32 } %651, 1, !dbg !27 + %655 = bitcast i32 %654 to <2 x bfloat>, !dbg !27 + %656 = extractvalue { i32, i32, i32, i32 } %651, 2, !dbg !27 + %657 = bitcast i32 %656 to <2 x bfloat>, !dbg !27 + %658 = extractvalue { i32, i32, i32, i32 } %651, 3, !dbg !27 + %659 = bitcast i32 %658 to <2 x bfloat>, !dbg !27 + %660 = extractelement <2 x bfloat> %653, i64 0, !dbg !27 + %661 = extractelement <2 x bfloat> %653, i64 1, !dbg !27 + %662 = extractelement <2 x bfloat> %655, i64 0, !dbg !27 + %663 = extractelement <2 x bfloat> %655, i64 1, !dbg !27 + %664 = extractelement <2 x bfloat> %657, i64 0, !dbg !27 + %665 = extractelement <2 x bfloat> %657, i64 1, !dbg !27 + %666 = extractelement <2 x bfloat> %659, i64 0, !dbg !27 + %667 = extractelement <2 x bfloat> %659, i64 1, !dbg !27 + %668 = extractelement <64 x i1> %62, i64 62, !dbg !27 + %669 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %259, i1 %668) #5, !dbg !27 + %670 = extractvalue { i32, i32, i32, i32 } %669, 0, !dbg !27 + %671 = bitcast i32 %670 to <2 x bfloat>, !dbg !27 + %672 = extractvalue { i32, i32, i32, i32 } %669, 1, !dbg !27 + %673 = bitcast i32 %672 to <2 x bfloat>, !dbg !27 + %674 = extractvalue { i32, i32, i32, i32 } %669, 2, !dbg !27 + %675 = bitcast i32 %674 to <2 x bfloat>, !dbg !27 + %676 = extractvalue { i32, i32, i32, i32 } %669, 3, !dbg !27 + %677 = bitcast i32 %676 to <2 x bfloat>, !dbg !27 + %678 = extractelement <2 x bfloat> %671, i64 0, !dbg !27 + %679 = extractelement <2 x bfloat> %671, i64 1, !dbg !27 + %680 = extractelement <2 x bfloat> %673, i64 0, !dbg !27 + %681 = extractelement <2 x bfloat> %673, i64 1, !dbg !27 + %682 = extractelement <2 x bfloat> %675, i64 0, !dbg !27 + %683 = extractelement <2 x bfloat> %675, i64 1, !dbg !27 + %684 = extractelement <2 x bfloat> %677, i64 0, !dbg !27 + %685 = extractelement <2 x bfloat> %677, i64 1, !dbg !27 + %686 = extractelement <64 x i1> %62, i64 61, !dbg !27 + %687 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %260, i1 %686) #5, !dbg !27 + %688 = extractvalue { i32, i32, i32, i32 } %687, 0, !dbg !27 + %689 = bitcast i32 %688 to <2 x bfloat>, !dbg !27 + %690 = extractvalue { i32, i32, i32, i32 } %687, 1, !dbg !27 + %691 = bitcast i32 %690 to <2 x bfloat>, !dbg !27 + %692 = extractvalue { i32, i32, i32, i32 } %687, 2, !dbg !27 + %693 = bitcast i32 %692 to <2 x bfloat>, !dbg !27 + %694 = extractvalue { i32, i32, i32, i32 } %687, 3, !dbg !27 + %695 = bitcast i32 %694 to <2 x bfloat>, !dbg !27 + %696 = extractelement <2 x bfloat> %689, i64 0, !dbg !27 + %697 = extractelement <2 x bfloat> %689, i64 1, !dbg !27 + %698 = extractelement <2 x bfloat> %691, i64 0, !dbg !27 + %699 = extractelement <2 x bfloat> %691, i64 1, !dbg !27 + %700 = extractelement <2 x bfloat> %693, i64 0, !dbg !27 + %701 = extractelement <2 x bfloat> %693, i64 1, !dbg !27 + %702 = extractelement <2 x bfloat> %695, i64 0, !dbg !27 + %703 = extractelement <2 x bfloat> %695, i64 1, !dbg !27 + %704 = extractelement <64 x i1> %62, i64 60, !dbg !27 + %705 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %261, i1 %704) #5, !dbg !27 + %706 = extractvalue { i32, i32, i32, i32 } %705, 0, !dbg !27 + %707 = bitcast i32 %706 to <2 x bfloat>, !dbg !27 + %708 = extractvalue { i32, i32, i32, i32 } %705, 1, !dbg !27 + %709 = bitcast i32 %708 to <2 x bfloat>, !dbg !27 + %710 = extractvalue { i32, i32, i32, i32 } %705, 2, !dbg !27 + %711 = bitcast i32 %710 to <2 x bfloat>, !dbg !27 + %712 = extractvalue { i32, i32, i32, i32 } %705, 3, !dbg !27 + %713 = bitcast i32 %712 to <2 x bfloat>, !dbg !27 + %714 = extractelement <2 x bfloat> %707, i64 0, !dbg !27 + %715 = extractelement <2 x bfloat> %707, i64 1, !dbg !27 + %716 = extractelement <2 x bfloat> %709, i64 0, !dbg !27 + %717 = extractelement <2 x bfloat> %709, i64 1, !dbg !27 + %718 = extractelement <2 x bfloat> %711, i64 0, !dbg !27 + %719 = extractelement <2 x bfloat> %711, i64 1, !dbg !27 + %720 = extractelement <2 x bfloat> %713, i64 0, !dbg !27 + %721 = extractelement <2 x bfloat> %713, i64 1, !dbg !27 + %722 = extractelement <64 x i1> %62, i64 59, !dbg !27 + %723 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %262, i1 %722) #5, !dbg !27 + %724 = extractvalue { i32, i32, i32, i32 } %723, 0, !dbg !27 + %725 = bitcast i32 %724 to <2 x bfloat>, !dbg !27 + %726 = extractvalue { i32, i32, i32, i32 } %723, 1, !dbg !27 + %727 = bitcast i32 %726 to <2 x bfloat>, !dbg !27 + %728 = extractvalue { i32, i32, i32, i32 } %723, 2, !dbg !27 + %729 = bitcast i32 %728 to <2 x bfloat>, !dbg !27 + %730 = extractvalue { i32, i32, i32, i32 } %723, 3, !dbg !27 + %731 = bitcast i32 %730 to <2 x bfloat>, !dbg !27 + %732 = extractelement <2 x bfloat> %725, i64 0, !dbg !27 + %733 = extractelement <2 x bfloat> %725, i64 1, !dbg !27 + %734 = extractelement <2 x bfloat> %727, i64 0, !dbg !27 + %735 = extractelement <2 x bfloat> %727, i64 1, !dbg !27 + %736 = extractelement <2 x bfloat> %729, i64 0, !dbg !27 + %737 = extractelement <2 x bfloat> %729, i64 1, !dbg !27 + %738 = extractelement <2 x bfloat> %731, i64 0, !dbg !27 + %739 = extractelement <2 x bfloat> %731, i64 1, !dbg !27 + %740 = extractelement <64 x i1> %62, i64 58, !dbg !27 + %741 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %263, i1 %740) #5, !dbg !27 + %742 = extractvalue { i32, i32, i32, i32 } %741, 0, !dbg !27 + %743 = bitcast i32 %742 to <2 x bfloat>, !dbg !27 + %744 = extractvalue { i32, i32, i32, i32 } %741, 1, !dbg !27 + %745 = bitcast i32 %744 to <2 x bfloat>, !dbg !27 + %746 = extractvalue { i32, i32, i32, i32 } %741, 2, !dbg !27 + %747 = bitcast i32 %746 to <2 x bfloat>, !dbg !27 + %748 = extractvalue { i32, i32, i32, i32 } %741, 3, !dbg !27 + %749 = bitcast i32 %748 to <2 x bfloat>, !dbg !27 + %750 = extractelement <2 x bfloat> %743, i64 0, !dbg !27 + %751 = extractelement <2 x bfloat> %743, i64 1, !dbg !27 + %752 = extractelement <2 x bfloat> %745, i64 0, !dbg !27 + %753 = extractelement <2 x bfloat> %745, i64 1, !dbg !27 + %754 = extractelement <2 x bfloat> %747, i64 0, !dbg !27 + %755 = extractelement <2 x bfloat> %747, i64 1, !dbg !27 + %756 = extractelement <2 x bfloat> %749, i64 0, !dbg !27 + %757 = extractelement <2 x bfloat> %749, i64 1, !dbg !27 + %758 = extractelement <64 x i1> %62, i64 57, !dbg !27 + %759 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %264, i1 %758) #5, !dbg !27 + %760 = extractvalue { i32, i32, i32, i32 } %759, 0, !dbg !27 + %761 = bitcast i32 %760 to <2 x bfloat>, !dbg !27 + %762 = extractvalue { i32, i32, i32, i32 } %759, 1, !dbg !27 + %763 = bitcast i32 %762 to <2 x bfloat>, !dbg !27 + %764 = extractvalue { i32, i32, i32, i32 } %759, 2, !dbg !27 + %765 = bitcast i32 %764 to <2 x bfloat>, !dbg !27 + %766 = extractvalue { i32, i32, i32, i32 } %759, 3, !dbg !27 + %767 = bitcast i32 %766 to <2 x bfloat>, !dbg !27 + %768 = extractelement <2 x bfloat> %761, i64 0, !dbg !27 + %769 = extractelement <2 x bfloat> %761, i64 1, !dbg !27 + %770 = extractelement <2 x bfloat> %763, i64 0, !dbg !27 + %771 = extractelement <2 x bfloat> %763, i64 1, !dbg !27 + %772 = extractelement <2 x bfloat> %765, i64 0, !dbg !27 + %773 = extractelement <2 x bfloat> %765, i64 1, !dbg !27 + %774 = extractelement <2 x bfloat> %767, i64 0, !dbg !27 + %775 = extractelement <2 x bfloat> %767, i64 1, !dbg !27 + %776 = extractelement <64 x i1> %62, i64 56, !dbg !27 + %777 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %265, i1 %776) #5, !dbg !27 + %778 = extractvalue { i32, i32, i32, i32 } %777, 0, !dbg !27 + %779 = bitcast i32 %778 to <2 x bfloat>, !dbg !27 + %780 = extractvalue { i32, i32, i32, i32 } %777, 1, !dbg !27 + %781 = bitcast i32 %780 to <2 x bfloat>, !dbg !27 + %782 = extractvalue { i32, i32, i32, i32 } %777, 2, !dbg !27 + %783 = bitcast i32 %782 to <2 x bfloat>, !dbg !27 + %784 = extractvalue { i32, i32, i32, i32 } %777, 3, !dbg !27 + %785 = bitcast i32 %784 to <2 x bfloat>, !dbg !27 + %786 = extractelement <2 x bfloat> %779, i64 0, !dbg !27 + %787 = extractelement <2 x bfloat> %779, i64 1, !dbg !27 + %788 = extractelement <2 x bfloat> %781, i64 0, !dbg !27 + %789 = extractelement <2 x bfloat> %781, i64 1, !dbg !27 + %790 = extractelement <2 x bfloat> %783, i64 0, !dbg !27 + %791 = extractelement <2 x bfloat> %783, i64 1, !dbg !27 + %792 = extractelement <2 x bfloat> %785, i64 0, !dbg !27 + %793 = extractelement <2 x bfloat> %785, i64 1, !dbg !27 + %794 = extractelement <64 x i1> %62, i64 55, !dbg !27 + %795 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %266, i1 %794) #5, !dbg !27 + %796 = extractvalue { i32, i32, i32, i32 } %795, 0, !dbg !27 + %797 = bitcast i32 %796 to <2 x bfloat>, !dbg !27 + %798 = extractvalue { i32, i32, i32, i32 } %795, 1, !dbg !27 + %799 = bitcast i32 %798 to <2 x bfloat>, !dbg !27 + %800 = extractvalue { i32, i32, i32, i32 } %795, 2, !dbg !27 + %801 = bitcast i32 %800 to <2 x bfloat>, !dbg !27 + %802 = extractvalue { i32, i32, i32, i32 } %795, 3, !dbg !27 + %803 = bitcast i32 %802 to <2 x bfloat>, !dbg !27 + %804 = extractelement <2 x bfloat> %797, i64 0, !dbg !27 + %805 = extractelement <2 x bfloat> %797, i64 1, !dbg !27 + %806 = extractelement <2 x bfloat> %799, i64 0, !dbg !27 + %807 = extractelement <2 x bfloat> %799, i64 1, !dbg !27 + %808 = extractelement <2 x bfloat> %801, i64 0, !dbg !27 + %809 = extractelement <2 x bfloat> %801, i64 1, !dbg !27 + %810 = extractelement <2 x bfloat> %803, i64 0, !dbg !27 + %811 = extractelement <2 x bfloat> %803, i64 1, !dbg !27 + %812 = extractelement <64 x i1> %62, i64 54, !dbg !27 + %813 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %267, i1 %812) #5, !dbg !27 + %814 = extractvalue { i32, i32, i32, i32 } %813, 0, !dbg !27 + %815 = bitcast i32 %814 to <2 x bfloat>, !dbg !27 + %816 = extractvalue { i32, i32, i32, i32 } %813, 1, !dbg !27 + %817 = bitcast i32 %816 to <2 x bfloat>, !dbg !27 + %818 = extractvalue { i32, i32, i32, i32 } %813, 2, !dbg !27 + %819 = bitcast i32 %818 to <2 x bfloat>, !dbg !27 + %820 = extractvalue { i32, i32, i32, i32 } %813, 3, !dbg !27 + %821 = bitcast i32 %820 to <2 x bfloat>, !dbg !27 + %822 = extractelement <2 x bfloat> %815, i64 0, !dbg !27 + %823 = extractelement <2 x bfloat> %815, i64 1, !dbg !27 + %824 = extractelement <2 x bfloat> %817, i64 0, !dbg !27 + %825 = extractelement <2 x bfloat> %817, i64 1, !dbg !27 + %826 = extractelement <2 x bfloat> %819, i64 0, !dbg !27 + %827 = extractelement <2 x bfloat> %819, i64 1, !dbg !27 + %828 = extractelement <2 x bfloat> %821, i64 0, !dbg !27 + %829 = extractelement <2 x bfloat> %821, i64 1, !dbg !27 + %830 = extractelement <64 x i1> %62, i64 53, !dbg !27 + %831 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %268, i1 %830) #5, !dbg !27 + %832 = extractvalue { i32, i32, i32, i32 } %831, 0, !dbg !27 + %833 = bitcast i32 %832 to <2 x bfloat>, !dbg !27 + %834 = extractvalue { i32, i32, i32, i32 } %831, 1, !dbg !27 + %835 = bitcast i32 %834 to <2 x bfloat>, !dbg !27 + %836 = extractvalue { i32, i32, i32, i32 } %831, 2, !dbg !27 + %837 = bitcast i32 %836 to <2 x bfloat>, !dbg !27 + %838 = extractvalue { i32, i32, i32, i32 } %831, 3, !dbg !27 + %839 = bitcast i32 %838 to <2 x bfloat>, !dbg !27 + %840 = extractelement <2 x bfloat> %833, i64 0, !dbg !27 + %841 = extractelement <2 x bfloat> %833, i64 1, !dbg !27 + %842 = extractelement <2 x bfloat> %835, i64 0, !dbg !27 + %843 = extractelement <2 x bfloat> %835, i64 1, !dbg !27 + %844 = extractelement <2 x bfloat> %837, i64 0, !dbg !27 + %845 = extractelement <2 x bfloat> %837, i64 1, !dbg !27 + %846 = extractelement <2 x bfloat> %839, i64 0, !dbg !27 + %847 = extractelement <2 x bfloat> %839, i64 1, !dbg !27 + %848 = extractelement <64 x i1> %62, i64 52, !dbg !27 + %849 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %269, i1 %848) #5, !dbg !27 + %850 = extractvalue { i32, i32, i32, i32 } %849, 0, !dbg !27 + %851 = bitcast i32 %850 to <2 x bfloat>, !dbg !27 + %852 = extractvalue { i32, i32, i32, i32 } %849, 1, !dbg !27 + %853 = bitcast i32 %852 to <2 x bfloat>, !dbg !27 + %854 = extractvalue { i32, i32, i32, i32 } %849, 2, !dbg !27 + %855 = bitcast i32 %854 to <2 x bfloat>, !dbg !27 + %856 = extractvalue { i32, i32, i32, i32 } %849, 3, !dbg !27 + %857 = bitcast i32 %856 to <2 x bfloat>, !dbg !27 + %858 = extractelement <2 x bfloat> %851, i64 0, !dbg !27 + %859 = extractelement <2 x bfloat> %851, i64 1, !dbg !27 + %860 = extractelement <2 x bfloat> %853, i64 0, !dbg !27 + %861 = extractelement <2 x bfloat> %853, i64 1, !dbg !27 + %862 = extractelement <2 x bfloat> %855, i64 0, !dbg !27 + %863 = extractelement <2 x bfloat> %855, i64 1, !dbg !27 + %864 = extractelement <2 x bfloat> %857, i64 0, !dbg !27 + %865 = extractelement <2 x bfloat> %857, i64 1, !dbg !27 + %866 = extractelement <64 x i1> %62, i64 51, !dbg !27 + %867 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %270, i1 %866) #5, !dbg !27 + %868 = extractvalue { i32, i32, i32, i32 } %867, 0, !dbg !27 + %869 = bitcast i32 %868 to <2 x bfloat>, !dbg !27 + %870 = extractvalue { i32, i32, i32, i32 } %867, 1, !dbg !27 + %871 = bitcast i32 %870 to <2 x bfloat>, !dbg !27 + %872 = extractvalue { i32, i32, i32, i32 } %867, 2, !dbg !27 + %873 = bitcast i32 %872 to <2 x bfloat>, !dbg !27 + %874 = extractvalue { i32, i32, i32, i32 } %867, 3, !dbg !27 + %875 = bitcast i32 %874 to <2 x bfloat>, !dbg !27 + %876 = extractelement <2 x bfloat> %869, i64 0, !dbg !27 + %877 = extractelement <2 x bfloat> %869, i64 1, !dbg !27 + %878 = extractelement <2 x bfloat> %871, i64 0, !dbg !27 + %879 = extractelement <2 x bfloat> %871, i64 1, !dbg !27 + %880 = extractelement <2 x bfloat> %873, i64 0, !dbg !27 + %881 = extractelement <2 x bfloat> %873, i64 1, !dbg !27 + %882 = extractelement <2 x bfloat> %875, i64 0, !dbg !27 + %883 = extractelement <2 x bfloat> %875, i64 1, !dbg !27 + %884 = extractelement <64 x i1> %62, i64 50, !dbg !27 + %885 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %271, i1 %884) #5, !dbg !27 + %886 = extractvalue { i32, i32, i32, i32 } %885, 0, !dbg !27 + %887 = bitcast i32 %886 to <2 x bfloat>, !dbg !27 + %888 = extractvalue { i32, i32, i32, i32 } %885, 1, !dbg !27 + %889 = bitcast i32 %888 to <2 x bfloat>, !dbg !27 + %890 = extractvalue { i32, i32, i32, i32 } %885, 2, !dbg !27 + %891 = bitcast i32 %890 to <2 x bfloat>, !dbg !27 + %892 = extractvalue { i32, i32, i32, i32 } %885, 3, !dbg !27 + %893 = bitcast i32 %892 to <2 x bfloat>, !dbg !27 + %894 = extractelement <2 x bfloat> %887, i64 0, !dbg !27 + %895 = extractelement <2 x bfloat> %887, i64 1, !dbg !27 + %896 = extractelement <2 x bfloat> %889, i64 0, !dbg !27 + %897 = extractelement <2 x bfloat> %889, i64 1, !dbg !27 + %898 = extractelement <2 x bfloat> %891, i64 0, !dbg !27 + %899 = extractelement <2 x bfloat> %891, i64 1, !dbg !27 + %900 = extractelement <2 x bfloat> %893, i64 0, !dbg !27 + %901 = extractelement <2 x bfloat> %893, i64 1, !dbg !27 + %902 = extractelement <64 x i1> %62, i64 49, !dbg !27 + %903 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %272, i1 %902) #5, !dbg !27 + %904 = extractvalue { i32, i32, i32, i32 } %903, 0, !dbg !27 + %905 = bitcast i32 %904 to <2 x bfloat>, !dbg !27 + %906 = extractvalue { i32, i32, i32, i32 } %903, 1, !dbg !27 + %907 = bitcast i32 %906 to <2 x bfloat>, !dbg !27 + %908 = extractvalue { i32, i32, i32, i32 } %903, 2, !dbg !27 + %909 = bitcast i32 %908 to <2 x bfloat>, !dbg !27 + %910 = extractvalue { i32, i32, i32, i32 } %903, 3, !dbg !27 + %911 = bitcast i32 %910 to <2 x bfloat>, !dbg !27 + %912 = extractelement <2 x bfloat> %905, i64 0, !dbg !27 + %913 = extractelement <2 x bfloat> %905, i64 1, !dbg !27 + %914 = extractelement <2 x bfloat> %907, i64 0, !dbg !27 + %915 = extractelement <2 x bfloat> %907, i64 1, !dbg !27 + %916 = extractelement <2 x bfloat> %909, i64 0, !dbg !27 + %917 = extractelement <2 x bfloat> %909, i64 1, !dbg !27 + %918 = extractelement <2 x bfloat> %911, i64 0, !dbg !27 + %919 = extractelement <2 x bfloat> %911, i64 1, !dbg !27 + %920 = extractelement <64 x i1> %62, i64 48, !dbg !27 + %921 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %273, i1 %920) #5, !dbg !27 + %922 = extractvalue { i32, i32, i32, i32 } %921, 0, !dbg !27 + %923 = bitcast i32 %922 to <2 x bfloat>, !dbg !27 + %924 = extractvalue { i32, i32, i32, i32 } %921, 1, !dbg !27 + %925 = bitcast i32 %924 to <2 x bfloat>, !dbg !27 + %926 = extractvalue { i32, i32, i32, i32 } %921, 2, !dbg !27 + %927 = bitcast i32 %926 to <2 x bfloat>, !dbg !27 + %928 = extractvalue { i32, i32, i32, i32 } %921, 3, !dbg !27 + %929 = bitcast i32 %928 to <2 x bfloat>, !dbg !27 + %930 = extractelement <2 x bfloat> %923, i64 0, !dbg !27 + %931 = extractelement <2 x bfloat> %923, i64 1, !dbg !27 + %932 = extractelement <2 x bfloat> %925, i64 0, !dbg !27 + %933 = extractelement <2 x bfloat> %925, i64 1, !dbg !27 + %934 = extractelement <2 x bfloat> %927, i64 0, !dbg !27 + %935 = extractelement <2 x bfloat> %927, i64 1, !dbg !27 + %936 = extractelement <2 x bfloat> %929, i64 0, !dbg !27 + %937 = extractelement <2 x bfloat> %929, i64 1, !dbg !27 + %938 = extractelement <64 x i1> %62, i64 47, !dbg !27 + %939 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %274, i1 %938) #5, !dbg !27 + %940 = extractvalue { i32, i32, i32, i32 } %939, 0, !dbg !27 + %941 = bitcast i32 %940 to <2 x bfloat>, !dbg !27 + %942 = extractvalue { i32, i32, i32, i32 } %939, 1, !dbg !27 + %943 = bitcast i32 %942 to <2 x bfloat>, !dbg !27 + %944 = extractvalue { i32, i32, i32, i32 } %939, 2, !dbg !27 + %945 = bitcast i32 %944 to <2 x bfloat>, !dbg !27 + %946 = extractvalue { i32, i32, i32, i32 } %939, 3, !dbg !27 + %947 = bitcast i32 %946 to <2 x bfloat>, !dbg !27 + %948 = extractelement <2 x bfloat> %941, i64 0, !dbg !27 + %949 = extractelement <2 x bfloat> %941, i64 1, !dbg !27 + %950 = extractelement <2 x bfloat> %943, i64 0, !dbg !27 + %951 = extractelement <2 x bfloat> %943, i64 1, !dbg !27 + %952 = extractelement <2 x bfloat> %945, i64 0, !dbg !27 + %953 = extractelement <2 x bfloat> %945, i64 1, !dbg !27 + %954 = extractelement <2 x bfloat> %947, i64 0, !dbg !27 + %955 = extractelement <2 x bfloat> %947, i64 1, !dbg !27 + %956 = extractelement <64 x i1> %62, i64 46, !dbg !27 + %957 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %275, i1 %956) #5, !dbg !27 + %958 = extractvalue { i32, i32, i32, i32 } %957, 0, !dbg !27 + %959 = bitcast i32 %958 to <2 x bfloat>, !dbg !27 + %960 = extractvalue { i32, i32, i32, i32 } %957, 1, !dbg !27 + %961 = bitcast i32 %960 to <2 x bfloat>, !dbg !27 + %962 = extractvalue { i32, i32, i32, i32 } %957, 2, !dbg !27 + %963 = bitcast i32 %962 to <2 x bfloat>, !dbg !27 + %964 = extractvalue { i32, i32, i32, i32 } %957, 3, !dbg !27 + %965 = bitcast i32 %964 to <2 x bfloat>, !dbg !27 + %966 = extractelement <2 x bfloat> %959, i64 0, !dbg !27 + %967 = extractelement <2 x bfloat> %959, i64 1, !dbg !27 + %968 = extractelement <2 x bfloat> %961, i64 0, !dbg !27 + %969 = extractelement <2 x bfloat> %961, i64 1, !dbg !27 + %970 = extractelement <2 x bfloat> %963, i64 0, !dbg !27 + %971 = extractelement <2 x bfloat> %963, i64 1, !dbg !27 + %972 = extractelement <2 x bfloat> %965, i64 0, !dbg !27 + %973 = extractelement <2 x bfloat> %965, i64 1, !dbg !27 + %974 = extractelement <64 x i1> %62, i64 45, !dbg !27 + %975 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %276, i1 %974) #5, !dbg !27 + %976 = extractvalue { i32, i32, i32, i32 } %975, 0, !dbg !27 + %977 = bitcast i32 %976 to <2 x bfloat>, !dbg !27 + %978 = extractvalue { i32, i32, i32, i32 } %975, 1, !dbg !27 + %979 = bitcast i32 %978 to <2 x bfloat>, !dbg !27 + %980 = extractvalue { i32, i32, i32, i32 } %975, 2, !dbg !27 + %981 = bitcast i32 %980 to <2 x bfloat>, !dbg !27 + %982 = extractvalue { i32, i32, i32, i32 } %975, 3, !dbg !27 + %983 = bitcast i32 %982 to <2 x bfloat>, !dbg !27 + %984 = extractelement <2 x bfloat> %977, i64 0, !dbg !27 + %985 = extractelement <2 x bfloat> %977, i64 1, !dbg !27 + %986 = extractelement <2 x bfloat> %979, i64 0, !dbg !27 + %987 = extractelement <2 x bfloat> %979, i64 1, !dbg !27 + %988 = extractelement <2 x bfloat> %981, i64 0, !dbg !27 + %989 = extractelement <2 x bfloat> %981, i64 1, !dbg !27 + %990 = extractelement <2 x bfloat> %983, i64 0, !dbg !27 + %991 = extractelement <2 x bfloat> %983, i64 1, !dbg !27 + %992 = extractelement <64 x i1> %62, i64 44, !dbg !27 + %993 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %277, i1 %992) #5, !dbg !27 + %994 = extractvalue { i32, i32, i32, i32 } %993, 0, !dbg !27 + %995 = bitcast i32 %994 to <2 x bfloat>, !dbg !27 + %996 = extractvalue { i32, i32, i32, i32 } %993, 1, !dbg !27 + %997 = bitcast i32 %996 to <2 x bfloat>, !dbg !27 + %998 = extractvalue { i32, i32, i32, i32 } %993, 2, !dbg !27 + %999 = bitcast i32 %998 to <2 x bfloat>, !dbg !27 + %1000 = extractvalue { i32, i32, i32, i32 } %993, 3, !dbg !27 + %1001 = bitcast i32 %1000 to <2 x bfloat>, !dbg !27 + %1002 = extractelement <2 x bfloat> %995, i64 0, !dbg !27 + %1003 = extractelement <2 x bfloat> %995, i64 1, !dbg !27 + %1004 = extractelement <2 x bfloat> %997, i64 0, !dbg !27 + %1005 = extractelement <2 x bfloat> %997, i64 1, !dbg !27 + %1006 = extractelement <2 x bfloat> %999, i64 0, !dbg !27 + %1007 = extractelement <2 x bfloat> %999, i64 1, !dbg !27 + %1008 = extractelement <2 x bfloat> %1001, i64 0, !dbg !27 + %1009 = extractelement <2 x bfloat> %1001, i64 1, !dbg !27 + %1010 = extractelement <64 x i1> %62, i64 43, !dbg !27 + %1011 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %278, i1 %1010) #5, !dbg !27 + %1012 = extractvalue { i32, i32, i32, i32 } %1011, 0, !dbg !27 + %1013 = bitcast i32 %1012 to <2 x bfloat>, !dbg !27 + %1014 = extractvalue { i32, i32, i32, i32 } %1011, 1, !dbg !27 + %1015 = bitcast i32 %1014 to <2 x bfloat>, !dbg !27 + %1016 = extractvalue { i32, i32, i32, i32 } %1011, 2, !dbg !27 + %1017 = bitcast i32 %1016 to <2 x bfloat>, !dbg !27 + %1018 = extractvalue { i32, i32, i32, i32 } %1011, 3, !dbg !27 + %1019 = bitcast i32 %1018 to <2 x bfloat>, !dbg !27 + %1020 = extractelement <2 x bfloat> %1013, i64 0, !dbg !27 + %1021 = extractelement <2 x bfloat> %1013, i64 1, !dbg !27 + %1022 = extractelement <2 x bfloat> %1015, i64 0, !dbg !27 + %1023 = extractelement <2 x bfloat> %1015, i64 1, !dbg !27 + %1024 = extractelement <2 x bfloat> %1017, i64 0, !dbg !27 + %1025 = extractelement <2 x bfloat> %1017, i64 1, !dbg !27 + %1026 = extractelement <2 x bfloat> %1019, i64 0, !dbg !27 + %1027 = extractelement <2 x bfloat> %1019, i64 1, !dbg !27 + %1028 = extractelement <64 x i1> %62, i64 42, !dbg !27 + %1029 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %279, i1 %1028) #5, !dbg !27 + %1030 = extractvalue { i32, i32, i32, i32 } %1029, 0, !dbg !27 + %1031 = bitcast i32 %1030 to <2 x bfloat>, !dbg !27 + %1032 = extractvalue { i32, i32, i32, i32 } %1029, 1, !dbg !27 + %1033 = bitcast i32 %1032 to <2 x bfloat>, !dbg !27 + %1034 = extractvalue { i32, i32, i32, i32 } %1029, 2, !dbg !27 + %1035 = bitcast i32 %1034 to <2 x bfloat>, !dbg !27 + %1036 = extractvalue { i32, i32, i32, i32 } %1029, 3, !dbg !27 + %1037 = bitcast i32 %1036 to <2 x bfloat>, !dbg !27 + %1038 = extractelement <2 x bfloat> %1031, i64 0, !dbg !27 + %1039 = extractelement <2 x bfloat> %1031, i64 1, !dbg !27 + %1040 = extractelement <2 x bfloat> %1033, i64 0, !dbg !27 + %1041 = extractelement <2 x bfloat> %1033, i64 1, !dbg !27 + %1042 = extractelement <2 x bfloat> %1035, i64 0, !dbg !27 + %1043 = extractelement <2 x bfloat> %1035, i64 1, !dbg !27 + %1044 = extractelement <2 x bfloat> %1037, i64 0, !dbg !27 + %1045 = extractelement <2 x bfloat> %1037, i64 1, !dbg !27 + %1046 = extractelement <64 x i1> %62, i64 41, !dbg !27 + %1047 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %280, i1 %1046) #5, !dbg !27 + %1048 = extractvalue { i32, i32, i32, i32 } %1047, 0, !dbg !27 + %1049 = bitcast i32 %1048 to <2 x bfloat>, !dbg !27 + %1050 = extractvalue { i32, i32, i32, i32 } %1047, 1, !dbg !27 + %1051 = bitcast i32 %1050 to <2 x bfloat>, !dbg !27 + %1052 = extractvalue { i32, i32, i32, i32 } %1047, 2, !dbg !27 + %1053 = bitcast i32 %1052 to <2 x bfloat>, !dbg !27 + %1054 = extractvalue { i32, i32, i32, i32 } %1047, 3, !dbg !27 + %1055 = bitcast i32 %1054 to <2 x bfloat>, !dbg !27 + %1056 = extractelement <2 x bfloat> %1049, i64 0, !dbg !27 + %1057 = extractelement <2 x bfloat> %1049, i64 1, !dbg !27 + %1058 = extractelement <2 x bfloat> %1051, i64 0, !dbg !27 + %1059 = extractelement <2 x bfloat> %1051, i64 1, !dbg !27 + %1060 = extractelement <2 x bfloat> %1053, i64 0, !dbg !27 + %1061 = extractelement <2 x bfloat> %1053, i64 1, !dbg !27 + %1062 = extractelement <2 x bfloat> %1055, i64 0, !dbg !27 + %1063 = extractelement <2 x bfloat> %1055, i64 1, !dbg !27 + %1064 = extractelement <64 x i1> %62, i64 40, !dbg !27 + %1065 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %281, i1 %1064) #5, !dbg !27 + %1066 = extractvalue { i32, i32, i32, i32 } %1065, 0, !dbg !27 + %1067 = bitcast i32 %1066 to <2 x bfloat>, !dbg !27 + %1068 = extractvalue { i32, i32, i32, i32 } %1065, 1, !dbg !27 + %1069 = bitcast i32 %1068 to <2 x bfloat>, !dbg !27 + %1070 = extractvalue { i32, i32, i32, i32 } %1065, 2, !dbg !27 + %1071 = bitcast i32 %1070 to <2 x bfloat>, !dbg !27 + %1072 = extractvalue { i32, i32, i32, i32 } %1065, 3, !dbg !27 + %1073 = bitcast i32 %1072 to <2 x bfloat>, !dbg !27 + %1074 = extractelement <2 x bfloat> %1067, i64 0, !dbg !27 + %1075 = extractelement <2 x bfloat> %1067, i64 1, !dbg !27 + %1076 = extractelement <2 x bfloat> %1069, i64 0, !dbg !27 + %1077 = extractelement <2 x bfloat> %1069, i64 1, !dbg !27 + %1078 = extractelement <2 x bfloat> %1071, i64 0, !dbg !27 + %1079 = extractelement <2 x bfloat> %1071, i64 1, !dbg !27 + %1080 = extractelement <2 x bfloat> %1073, i64 0, !dbg !27 + %1081 = extractelement <2 x bfloat> %1073, i64 1, !dbg !27 + %1082 = extractelement <64 x i1> %62, i64 39, !dbg !27 + %1083 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %282, i1 %1082) #5, !dbg !27 + %1084 = extractvalue { i32, i32, i32, i32 } %1083, 0, !dbg !27 + %1085 = bitcast i32 %1084 to <2 x bfloat>, !dbg !27 + %1086 = extractvalue { i32, i32, i32, i32 } %1083, 1, !dbg !27 + %1087 = bitcast i32 %1086 to <2 x bfloat>, !dbg !27 + %1088 = extractvalue { i32, i32, i32, i32 } %1083, 2, !dbg !27 + %1089 = bitcast i32 %1088 to <2 x bfloat>, !dbg !27 + %1090 = extractvalue { i32, i32, i32, i32 } %1083, 3, !dbg !27 + %1091 = bitcast i32 %1090 to <2 x bfloat>, !dbg !27 + %1092 = extractelement <2 x bfloat> %1085, i64 0, !dbg !27 + %1093 = extractelement <2 x bfloat> %1085, i64 1, !dbg !27 + %1094 = extractelement <2 x bfloat> %1087, i64 0, !dbg !27 + %1095 = extractelement <2 x bfloat> %1087, i64 1, !dbg !27 + %1096 = extractelement <2 x bfloat> %1089, i64 0, !dbg !27 + %1097 = extractelement <2 x bfloat> %1089, i64 1, !dbg !27 + %1098 = extractelement <2 x bfloat> %1091, i64 0, !dbg !27 + %1099 = extractelement <2 x bfloat> %1091, i64 1, !dbg !27 + %1100 = extractelement <64 x i1> %62, i64 38, !dbg !27 + %1101 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %283, i1 %1100) #5, !dbg !27 + %1102 = extractvalue { i32, i32, i32, i32 } %1101, 0, !dbg !27 + %1103 = bitcast i32 %1102 to <2 x bfloat>, !dbg !27 + %1104 = extractvalue { i32, i32, i32, i32 } %1101, 1, !dbg !27 + %1105 = bitcast i32 %1104 to <2 x bfloat>, !dbg !27 + %1106 = extractvalue { i32, i32, i32, i32 } %1101, 2, !dbg !27 + %1107 = bitcast i32 %1106 to <2 x bfloat>, !dbg !27 + %1108 = extractvalue { i32, i32, i32, i32 } %1101, 3, !dbg !27 + %1109 = bitcast i32 %1108 to <2 x bfloat>, !dbg !27 + %1110 = extractelement <2 x bfloat> %1103, i64 0, !dbg !27 + %1111 = extractelement <2 x bfloat> %1103, i64 1, !dbg !27 + %1112 = extractelement <2 x bfloat> %1105, i64 0, !dbg !27 + %1113 = extractelement <2 x bfloat> %1105, i64 1, !dbg !27 + %1114 = extractelement <2 x bfloat> %1107, i64 0, !dbg !27 + %1115 = extractelement <2 x bfloat> %1107, i64 1, !dbg !27 + %1116 = extractelement <2 x bfloat> %1109, i64 0, !dbg !27 + %1117 = extractelement <2 x bfloat> %1109, i64 1, !dbg !27 + %1118 = extractelement <64 x i1> %62, i64 37, !dbg !27 + %1119 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %284, i1 %1118) #5, !dbg !27 + %1120 = extractvalue { i32, i32, i32, i32 } %1119, 0, !dbg !27 + %1121 = bitcast i32 %1120 to <2 x bfloat>, !dbg !27 + %1122 = extractvalue { i32, i32, i32, i32 } %1119, 1, !dbg !27 + %1123 = bitcast i32 %1122 to <2 x bfloat>, !dbg !27 + %1124 = extractvalue { i32, i32, i32, i32 } %1119, 2, !dbg !27 + %1125 = bitcast i32 %1124 to <2 x bfloat>, !dbg !27 + %1126 = extractvalue { i32, i32, i32, i32 } %1119, 3, !dbg !27 + %1127 = bitcast i32 %1126 to <2 x bfloat>, !dbg !27 + %1128 = extractelement <2 x bfloat> %1121, i64 0, !dbg !27 + %1129 = extractelement <2 x bfloat> %1121, i64 1, !dbg !27 + %1130 = extractelement <2 x bfloat> %1123, i64 0, !dbg !27 + %1131 = extractelement <2 x bfloat> %1123, i64 1, !dbg !27 + %1132 = extractelement <2 x bfloat> %1125, i64 0, !dbg !27 + %1133 = extractelement <2 x bfloat> %1125, i64 1, !dbg !27 + %1134 = extractelement <2 x bfloat> %1127, i64 0, !dbg !27 + %1135 = extractelement <2 x bfloat> %1127, i64 1, !dbg !27 + %1136 = extractelement <64 x i1> %62, i64 36, !dbg !27 + %1137 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %285, i1 %1136) #5, !dbg !27 + %1138 = extractvalue { i32, i32, i32, i32 } %1137, 0, !dbg !27 + %1139 = bitcast i32 %1138 to <2 x bfloat>, !dbg !27 + %1140 = extractvalue { i32, i32, i32, i32 } %1137, 1, !dbg !27 + %1141 = bitcast i32 %1140 to <2 x bfloat>, !dbg !27 + %1142 = extractvalue { i32, i32, i32, i32 } %1137, 2, !dbg !27 + %1143 = bitcast i32 %1142 to <2 x bfloat>, !dbg !27 + %1144 = extractvalue { i32, i32, i32, i32 } %1137, 3, !dbg !27 + %1145 = bitcast i32 %1144 to <2 x bfloat>, !dbg !27 + %1146 = extractelement <2 x bfloat> %1139, i64 0, !dbg !27 + %1147 = extractelement <2 x bfloat> %1139, i64 1, !dbg !27 + %1148 = extractelement <2 x bfloat> %1141, i64 0, !dbg !27 + %1149 = extractelement <2 x bfloat> %1141, i64 1, !dbg !27 + %1150 = extractelement <2 x bfloat> %1143, i64 0, !dbg !27 + %1151 = extractelement <2 x bfloat> %1143, i64 1, !dbg !27 + %1152 = extractelement <2 x bfloat> %1145, i64 0, !dbg !27 + %1153 = extractelement <2 x bfloat> %1145, i64 1, !dbg !27 + %1154 = extractelement <64 x i1> %62, i64 35, !dbg !27 + %1155 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %286, i1 %1154) #5, !dbg !27 + %1156 = extractvalue { i32, i32, i32, i32 } %1155, 0, !dbg !27 + %1157 = bitcast i32 %1156 to <2 x bfloat>, !dbg !27 + %1158 = extractvalue { i32, i32, i32, i32 } %1155, 1, !dbg !27 + %1159 = bitcast i32 %1158 to <2 x bfloat>, !dbg !27 + %1160 = extractvalue { i32, i32, i32, i32 } %1155, 2, !dbg !27 + %1161 = bitcast i32 %1160 to <2 x bfloat>, !dbg !27 + %1162 = extractvalue { i32, i32, i32, i32 } %1155, 3, !dbg !27 + %1163 = bitcast i32 %1162 to <2 x bfloat>, !dbg !27 + %1164 = extractelement <2 x bfloat> %1157, i64 0, !dbg !27 + %1165 = extractelement <2 x bfloat> %1157, i64 1, !dbg !27 + %1166 = extractelement <2 x bfloat> %1159, i64 0, !dbg !27 + %1167 = extractelement <2 x bfloat> %1159, i64 1, !dbg !27 + %1168 = extractelement <2 x bfloat> %1161, i64 0, !dbg !27 + %1169 = extractelement <2 x bfloat> %1161, i64 1, !dbg !27 + %1170 = extractelement <2 x bfloat> %1163, i64 0, !dbg !27 + %1171 = extractelement <2 x bfloat> %1163, i64 1, !dbg !27 + %1172 = extractelement <64 x i1> %62, i64 34, !dbg !27 + %1173 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %287, i1 %1172) #5, !dbg !27 + %1174 = extractvalue { i32, i32, i32, i32 } %1173, 0, !dbg !27 + %1175 = bitcast i32 %1174 to <2 x bfloat>, !dbg !27 + %1176 = extractvalue { i32, i32, i32, i32 } %1173, 1, !dbg !27 + %1177 = bitcast i32 %1176 to <2 x bfloat>, !dbg !27 + %1178 = extractvalue { i32, i32, i32, i32 } %1173, 2, !dbg !27 + %1179 = bitcast i32 %1178 to <2 x bfloat>, !dbg !27 + %1180 = extractvalue { i32, i32, i32, i32 } %1173, 3, !dbg !27 + %1181 = bitcast i32 %1180 to <2 x bfloat>, !dbg !27 + %1182 = extractelement <2 x bfloat> %1175, i64 0, !dbg !27 + %1183 = extractelement <2 x bfloat> %1175, i64 1, !dbg !27 + %1184 = extractelement <2 x bfloat> %1177, i64 0, !dbg !27 + %1185 = extractelement <2 x bfloat> %1177, i64 1, !dbg !27 + %1186 = extractelement <2 x bfloat> %1179, i64 0, !dbg !27 + %1187 = extractelement <2 x bfloat> %1179, i64 1, !dbg !27 + %1188 = extractelement <2 x bfloat> %1181, i64 0, !dbg !27 + %1189 = extractelement <2 x bfloat> %1181, i64 1, !dbg !27 + %1190 = extractelement <64 x i1> %62, i64 33, !dbg !27 + %1191 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %288, i1 %1190) #5, !dbg !27 + %1192 = extractvalue { i32, i32, i32, i32 } %1191, 0, !dbg !27 + %1193 = bitcast i32 %1192 to <2 x bfloat>, !dbg !27 + %1194 = extractvalue { i32, i32, i32, i32 } %1191, 1, !dbg !27 + %1195 = bitcast i32 %1194 to <2 x bfloat>, !dbg !27 + %1196 = extractvalue { i32, i32, i32, i32 } %1191, 2, !dbg !27 + %1197 = bitcast i32 %1196 to <2 x bfloat>, !dbg !27 + %1198 = extractvalue { i32, i32, i32, i32 } %1191, 3, !dbg !27 + %1199 = bitcast i32 %1198 to <2 x bfloat>, !dbg !27 + %1200 = extractelement <2 x bfloat> %1193, i64 0, !dbg !27 + %1201 = extractelement <2 x bfloat> %1193, i64 1, !dbg !27 + %1202 = extractelement <2 x bfloat> %1195, i64 0, !dbg !27 + %1203 = extractelement <2 x bfloat> %1195, i64 1, !dbg !27 + %1204 = extractelement <2 x bfloat> %1197, i64 0, !dbg !27 + %1205 = extractelement <2 x bfloat> %1197, i64 1, !dbg !27 + %1206 = extractelement <2 x bfloat> %1199, i64 0, !dbg !27 + %1207 = extractelement <2 x bfloat> %1199, i64 1, !dbg !27 + %1208 = extractelement <64 x i1> %62, i64 32, !dbg !27 + %1209 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %289, i1 %1208) #5, !dbg !27 + %1210 = extractvalue { i32, i32, i32, i32 } %1209, 0, !dbg !27 + %1211 = bitcast i32 %1210 to <2 x bfloat>, !dbg !27 + %1212 = extractvalue { i32, i32, i32, i32 } %1209, 1, !dbg !27 + %1213 = bitcast i32 %1212 to <2 x bfloat>, !dbg !27 + %1214 = extractvalue { i32, i32, i32, i32 } %1209, 2, !dbg !27 + %1215 = bitcast i32 %1214 to <2 x bfloat>, !dbg !27 + %1216 = extractvalue { i32, i32, i32, i32 } %1209, 3, !dbg !27 + %1217 = bitcast i32 %1216 to <2 x bfloat>, !dbg !27 + %1218 = extractelement <2 x bfloat> %1211, i64 0, !dbg !27 + %1219 = extractelement <2 x bfloat> %1211, i64 1, !dbg !27 + %1220 = extractelement <2 x bfloat> %1213, i64 0, !dbg !27 + %1221 = extractelement <2 x bfloat> %1213, i64 1, !dbg !27 + %1222 = extractelement <2 x bfloat> %1215, i64 0, !dbg !27 + %1223 = extractelement <2 x bfloat> %1215, i64 1, !dbg !27 + %1224 = extractelement <2 x bfloat> %1217, i64 0, !dbg !27 + %1225 = extractelement <2 x bfloat> %1217, i64 1, !dbg !27 + %1226 = extractelement <64 x i1> %62, i64 31, !dbg !27 + %1227 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %290, i1 %1226) #5, !dbg !27 + %1228 = extractvalue { i32, i32, i32, i32 } %1227, 0, !dbg !27 + %1229 = bitcast i32 %1228 to <2 x bfloat>, !dbg !27 + %1230 = extractvalue { i32, i32, i32, i32 } %1227, 1, !dbg !27 + %1231 = bitcast i32 %1230 to <2 x bfloat>, !dbg !27 + %1232 = extractvalue { i32, i32, i32, i32 } %1227, 2, !dbg !27 + %1233 = bitcast i32 %1232 to <2 x bfloat>, !dbg !27 + %1234 = extractvalue { i32, i32, i32, i32 } %1227, 3, !dbg !27 + %1235 = bitcast i32 %1234 to <2 x bfloat>, !dbg !27 + %1236 = extractelement <2 x bfloat> %1229, i64 0, !dbg !27 + %1237 = extractelement <2 x bfloat> %1229, i64 1, !dbg !27 + %1238 = extractelement <2 x bfloat> %1231, i64 0, !dbg !27 + %1239 = extractelement <2 x bfloat> %1231, i64 1, !dbg !27 + %1240 = extractelement <2 x bfloat> %1233, i64 0, !dbg !27 + %1241 = extractelement <2 x bfloat> %1233, i64 1, !dbg !27 + %1242 = extractelement <2 x bfloat> %1235, i64 0, !dbg !27 + %1243 = extractelement <2 x bfloat> %1235, i64 1, !dbg !27 + %1244 = extractelement <64 x i1> %62, i64 30, !dbg !27 + %1245 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %291, i1 %1244) #5, !dbg !27 + %1246 = extractvalue { i32, i32, i32, i32 } %1245, 0, !dbg !27 + %1247 = bitcast i32 %1246 to <2 x bfloat>, !dbg !27 + %1248 = extractvalue { i32, i32, i32, i32 } %1245, 1, !dbg !27 + %1249 = bitcast i32 %1248 to <2 x bfloat>, !dbg !27 + %1250 = extractvalue { i32, i32, i32, i32 } %1245, 2, !dbg !27 + %1251 = bitcast i32 %1250 to <2 x bfloat>, !dbg !27 + %1252 = extractvalue { i32, i32, i32, i32 } %1245, 3, !dbg !27 + %1253 = bitcast i32 %1252 to <2 x bfloat>, !dbg !27 + %1254 = extractelement <2 x bfloat> %1247, i64 0, !dbg !27 + %1255 = extractelement <2 x bfloat> %1247, i64 1, !dbg !27 + %1256 = extractelement <2 x bfloat> %1249, i64 0, !dbg !27 + %1257 = extractelement <2 x bfloat> %1249, i64 1, !dbg !27 + %1258 = extractelement <2 x bfloat> %1251, i64 0, !dbg !27 + %1259 = extractelement <2 x bfloat> %1251, i64 1, !dbg !27 + %1260 = extractelement <2 x bfloat> %1253, i64 0, !dbg !27 + %1261 = extractelement <2 x bfloat> %1253, i64 1, !dbg !27 + %1262 = extractelement <64 x i1> %62, i64 29, !dbg !27 + %1263 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %292, i1 %1262) #5, !dbg !27 + %1264 = extractvalue { i32, i32, i32, i32 } %1263, 0, !dbg !27 + %1265 = bitcast i32 %1264 to <2 x bfloat>, !dbg !27 + %1266 = extractvalue { i32, i32, i32, i32 } %1263, 1, !dbg !27 + %1267 = bitcast i32 %1266 to <2 x bfloat>, !dbg !27 + %1268 = extractvalue { i32, i32, i32, i32 } %1263, 2, !dbg !27 + %1269 = bitcast i32 %1268 to <2 x bfloat>, !dbg !27 + %1270 = extractvalue { i32, i32, i32, i32 } %1263, 3, !dbg !27 + %1271 = bitcast i32 %1270 to <2 x bfloat>, !dbg !27 + %1272 = extractelement <2 x bfloat> %1265, i64 0, !dbg !27 + %1273 = extractelement <2 x bfloat> %1265, i64 1, !dbg !27 + %1274 = extractelement <2 x bfloat> %1267, i64 0, !dbg !27 + %1275 = extractelement <2 x bfloat> %1267, i64 1, !dbg !27 + %1276 = extractelement <2 x bfloat> %1269, i64 0, !dbg !27 + %1277 = extractelement <2 x bfloat> %1269, i64 1, !dbg !27 + %1278 = extractelement <2 x bfloat> %1271, i64 0, !dbg !27 + %1279 = extractelement <2 x bfloat> %1271, i64 1, !dbg !27 + %1280 = extractelement <64 x i1> %62, i64 28, !dbg !27 + %1281 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %293, i1 %1280) #5, !dbg !27 + %1282 = extractvalue { i32, i32, i32, i32 } %1281, 0, !dbg !27 + %1283 = bitcast i32 %1282 to <2 x bfloat>, !dbg !27 + %1284 = extractvalue { i32, i32, i32, i32 } %1281, 1, !dbg !27 + %1285 = bitcast i32 %1284 to <2 x bfloat>, !dbg !27 + %1286 = extractvalue { i32, i32, i32, i32 } %1281, 2, !dbg !27 + %1287 = bitcast i32 %1286 to <2 x bfloat>, !dbg !27 + %1288 = extractvalue { i32, i32, i32, i32 } %1281, 3, !dbg !27 + %1289 = bitcast i32 %1288 to <2 x bfloat>, !dbg !27 + %1290 = extractelement <2 x bfloat> %1283, i64 0, !dbg !27 + %1291 = extractelement <2 x bfloat> %1283, i64 1, !dbg !27 + %1292 = extractelement <2 x bfloat> %1285, i64 0, !dbg !27 + %1293 = extractelement <2 x bfloat> %1285, i64 1, !dbg !27 + %1294 = extractelement <2 x bfloat> %1287, i64 0, !dbg !27 + %1295 = extractelement <2 x bfloat> %1287, i64 1, !dbg !27 + %1296 = extractelement <2 x bfloat> %1289, i64 0, !dbg !27 + %1297 = extractelement <2 x bfloat> %1289, i64 1, !dbg !27 + %1298 = extractelement <64 x i1> %62, i64 27, !dbg !27 + %1299 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %294, i1 %1298) #5, !dbg !27 + %1300 = extractvalue { i32, i32, i32, i32 } %1299, 0, !dbg !27 + %1301 = bitcast i32 %1300 to <2 x bfloat>, !dbg !27 + %1302 = extractvalue { i32, i32, i32, i32 } %1299, 1, !dbg !27 + %1303 = bitcast i32 %1302 to <2 x bfloat>, !dbg !27 + %1304 = extractvalue { i32, i32, i32, i32 } %1299, 2, !dbg !27 + %1305 = bitcast i32 %1304 to <2 x bfloat>, !dbg !27 + %1306 = extractvalue { i32, i32, i32, i32 } %1299, 3, !dbg !27 + %1307 = bitcast i32 %1306 to <2 x bfloat>, !dbg !27 + %1308 = extractelement <2 x bfloat> %1301, i64 0, !dbg !27 + %1309 = extractelement <2 x bfloat> %1301, i64 1, !dbg !27 + %1310 = extractelement <2 x bfloat> %1303, i64 0, !dbg !27 + %1311 = extractelement <2 x bfloat> %1303, i64 1, !dbg !27 + %1312 = extractelement <2 x bfloat> %1305, i64 0, !dbg !27 + %1313 = extractelement <2 x bfloat> %1305, i64 1, !dbg !27 + %1314 = extractelement <2 x bfloat> %1307, i64 0, !dbg !27 + %1315 = extractelement <2 x bfloat> %1307, i64 1, !dbg !27 + %1316 = extractelement <64 x i1> %62, i64 26, !dbg !27 + %1317 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %295, i1 %1316) #5, !dbg !27 + %1318 = extractvalue { i32, i32, i32, i32 } %1317, 0, !dbg !27 + %1319 = bitcast i32 %1318 to <2 x bfloat>, !dbg !27 + %1320 = extractvalue { i32, i32, i32, i32 } %1317, 1, !dbg !27 + %1321 = bitcast i32 %1320 to <2 x bfloat>, !dbg !27 + %1322 = extractvalue { i32, i32, i32, i32 } %1317, 2, !dbg !27 + %1323 = bitcast i32 %1322 to <2 x bfloat>, !dbg !27 + %1324 = extractvalue { i32, i32, i32, i32 } %1317, 3, !dbg !27 + %1325 = bitcast i32 %1324 to <2 x bfloat>, !dbg !27 + %1326 = extractelement <2 x bfloat> %1319, i64 0, !dbg !27 + %1327 = extractelement <2 x bfloat> %1319, i64 1, !dbg !27 + %1328 = extractelement <2 x bfloat> %1321, i64 0, !dbg !27 + %1329 = extractelement <2 x bfloat> %1321, i64 1, !dbg !27 + %1330 = extractelement <2 x bfloat> %1323, i64 0, !dbg !27 + %1331 = extractelement <2 x bfloat> %1323, i64 1, !dbg !27 + %1332 = extractelement <2 x bfloat> %1325, i64 0, !dbg !27 + %1333 = extractelement <2 x bfloat> %1325, i64 1, !dbg !27 + %1334 = extractelement <64 x i1> %62, i64 25, !dbg !27 + %1335 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %296, i1 %1334) #5, !dbg !27 + %1336 = extractvalue { i32, i32, i32, i32 } %1335, 0, !dbg !27 + %1337 = bitcast i32 %1336 to <2 x bfloat>, !dbg !27 + %1338 = extractvalue { i32, i32, i32, i32 } %1335, 1, !dbg !27 + %1339 = bitcast i32 %1338 to <2 x bfloat>, !dbg !27 + %1340 = extractvalue { i32, i32, i32, i32 } %1335, 2, !dbg !27 + %1341 = bitcast i32 %1340 to <2 x bfloat>, !dbg !27 + %1342 = extractvalue { i32, i32, i32, i32 } %1335, 3, !dbg !27 + %1343 = bitcast i32 %1342 to <2 x bfloat>, !dbg !27 + %1344 = extractelement <2 x bfloat> %1337, i64 0, !dbg !27 + %1345 = extractelement <2 x bfloat> %1337, i64 1, !dbg !27 + %1346 = extractelement <2 x bfloat> %1339, i64 0, !dbg !27 + %1347 = extractelement <2 x bfloat> %1339, i64 1, !dbg !27 + %1348 = extractelement <2 x bfloat> %1341, i64 0, !dbg !27 + %1349 = extractelement <2 x bfloat> %1341, i64 1, !dbg !27 + %1350 = extractelement <2 x bfloat> %1343, i64 0, !dbg !27 + %1351 = extractelement <2 x bfloat> %1343, i64 1, !dbg !27 + %1352 = extractelement <64 x i1> %62, i64 24, !dbg !27 + %1353 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %297, i1 %1352) #5, !dbg !27 + %1354 = extractvalue { i32, i32, i32, i32 } %1353, 0, !dbg !27 + %1355 = bitcast i32 %1354 to <2 x bfloat>, !dbg !27 + %1356 = extractvalue { i32, i32, i32, i32 } %1353, 1, !dbg !27 + %1357 = bitcast i32 %1356 to <2 x bfloat>, !dbg !27 + %1358 = extractvalue { i32, i32, i32, i32 } %1353, 2, !dbg !27 + %1359 = bitcast i32 %1358 to <2 x bfloat>, !dbg !27 + %1360 = extractvalue { i32, i32, i32, i32 } %1353, 3, !dbg !27 + %1361 = bitcast i32 %1360 to <2 x bfloat>, !dbg !27 + %1362 = extractelement <2 x bfloat> %1355, i64 0, !dbg !27 + %1363 = extractelement <2 x bfloat> %1355, i64 1, !dbg !27 + %1364 = extractelement <2 x bfloat> %1357, i64 0, !dbg !27 + %1365 = extractelement <2 x bfloat> %1357, i64 1, !dbg !27 + %1366 = extractelement <2 x bfloat> %1359, i64 0, !dbg !27 + %1367 = extractelement <2 x bfloat> %1359, i64 1, !dbg !27 + %1368 = extractelement <2 x bfloat> %1361, i64 0, !dbg !27 + %1369 = extractelement <2 x bfloat> %1361, i64 1, !dbg !27 + %1370 = extractelement <64 x i1> %62, i64 23, !dbg !27 + %1371 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %298, i1 %1370) #5, !dbg !27 + %1372 = extractvalue { i32, i32, i32, i32 } %1371, 0, !dbg !27 + %1373 = bitcast i32 %1372 to <2 x bfloat>, !dbg !27 + %1374 = extractvalue { i32, i32, i32, i32 } %1371, 1, !dbg !27 + %1375 = bitcast i32 %1374 to <2 x bfloat>, !dbg !27 + %1376 = extractvalue { i32, i32, i32, i32 } %1371, 2, !dbg !27 + %1377 = bitcast i32 %1376 to <2 x bfloat>, !dbg !27 + %1378 = extractvalue { i32, i32, i32, i32 } %1371, 3, !dbg !27 + %1379 = bitcast i32 %1378 to <2 x bfloat>, !dbg !27 + %1380 = extractelement <2 x bfloat> %1373, i64 0, !dbg !27 + %1381 = extractelement <2 x bfloat> %1373, i64 1, !dbg !27 + %1382 = extractelement <2 x bfloat> %1375, i64 0, !dbg !27 + %1383 = extractelement <2 x bfloat> %1375, i64 1, !dbg !27 + %1384 = extractelement <2 x bfloat> %1377, i64 0, !dbg !27 + %1385 = extractelement <2 x bfloat> %1377, i64 1, !dbg !27 + %1386 = extractelement <2 x bfloat> %1379, i64 0, !dbg !27 + %1387 = extractelement <2 x bfloat> %1379, i64 1, !dbg !27 + %1388 = extractelement <64 x i1> %62, i64 22, !dbg !27 + %1389 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %299, i1 %1388) #5, !dbg !27 + %1390 = extractvalue { i32, i32, i32, i32 } %1389, 0, !dbg !27 + %1391 = bitcast i32 %1390 to <2 x bfloat>, !dbg !27 + %1392 = extractvalue { i32, i32, i32, i32 } %1389, 1, !dbg !27 + %1393 = bitcast i32 %1392 to <2 x bfloat>, !dbg !27 + %1394 = extractvalue { i32, i32, i32, i32 } %1389, 2, !dbg !27 + %1395 = bitcast i32 %1394 to <2 x bfloat>, !dbg !27 + %1396 = extractvalue { i32, i32, i32, i32 } %1389, 3, !dbg !27 + %1397 = bitcast i32 %1396 to <2 x bfloat>, !dbg !27 + %1398 = extractelement <2 x bfloat> %1391, i64 0, !dbg !27 + %1399 = extractelement <2 x bfloat> %1391, i64 1, !dbg !27 + %1400 = extractelement <2 x bfloat> %1393, i64 0, !dbg !27 + %1401 = extractelement <2 x bfloat> %1393, i64 1, !dbg !27 + %1402 = extractelement <2 x bfloat> %1395, i64 0, !dbg !27 + %1403 = extractelement <2 x bfloat> %1395, i64 1, !dbg !27 + %1404 = extractelement <2 x bfloat> %1397, i64 0, !dbg !27 + %1405 = extractelement <2 x bfloat> %1397, i64 1, !dbg !27 + %1406 = extractelement <64 x i1> %62, i64 21, !dbg !27 + %1407 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %300, i1 %1406) #5, !dbg !27 + %1408 = extractvalue { i32, i32, i32, i32 } %1407, 0, !dbg !27 + %1409 = bitcast i32 %1408 to <2 x bfloat>, !dbg !27 + %1410 = extractvalue { i32, i32, i32, i32 } %1407, 1, !dbg !27 + %1411 = bitcast i32 %1410 to <2 x bfloat>, !dbg !27 + %1412 = extractvalue { i32, i32, i32, i32 } %1407, 2, !dbg !27 + %1413 = bitcast i32 %1412 to <2 x bfloat>, !dbg !27 + %1414 = extractvalue { i32, i32, i32, i32 } %1407, 3, !dbg !27 + %1415 = bitcast i32 %1414 to <2 x bfloat>, !dbg !27 + %1416 = extractelement <2 x bfloat> %1409, i64 0, !dbg !27 + %1417 = extractelement <2 x bfloat> %1409, i64 1, !dbg !27 + %1418 = extractelement <2 x bfloat> %1411, i64 0, !dbg !27 + %1419 = extractelement <2 x bfloat> %1411, i64 1, !dbg !27 + %1420 = extractelement <2 x bfloat> %1413, i64 0, !dbg !27 + %1421 = extractelement <2 x bfloat> %1413, i64 1, !dbg !27 + %1422 = extractelement <2 x bfloat> %1415, i64 0, !dbg !27 + %1423 = extractelement <2 x bfloat> %1415, i64 1, !dbg !27 + %1424 = extractelement <64 x i1> %62, i64 20, !dbg !27 + %1425 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %301, i1 %1424) #5, !dbg !27 + %1426 = extractvalue { i32, i32, i32, i32 } %1425, 0, !dbg !27 + %1427 = bitcast i32 %1426 to <2 x bfloat>, !dbg !27 + %1428 = extractvalue { i32, i32, i32, i32 } %1425, 1, !dbg !27 + %1429 = bitcast i32 %1428 to <2 x bfloat>, !dbg !27 + %1430 = extractvalue { i32, i32, i32, i32 } %1425, 2, !dbg !27 + %1431 = bitcast i32 %1430 to <2 x bfloat>, !dbg !27 + %1432 = extractvalue { i32, i32, i32, i32 } %1425, 3, !dbg !27 + %1433 = bitcast i32 %1432 to <2 x bfloat>, !dbg !27 + %1434 = extractelement <2 x bfloat> %1427, i64 0, !dbg !27 + %1435 = extractelement <2 x bfloat> %1427, i64 1, !dbg !27 + %1436 = extractelement <2 x bfloat> %1429, i64 0, !dbg !27 + %1437 = extractelement <2 x bfloat> %1429, i64 1, !dbg !27 + %1438 = extractelement <2 x bfloat> %1431, i64 0, !dbg !27 + %1439 = extractelement <2 x bfloat> %1431, i64 1, !dbg !27 + %1440 = extractelement <2 x bfloat> %1433, i64 0, !dbg !27 + %1441 = extractelement <2 x bfloat> %1433, i64 1, !dbg !27 + %1442 = extractelement <64 x i1> %62, i64 19, !dbg !27 + %1443 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %302, i1 %1442) #5, !dbg !27 + %1444 = extractvalue { i32, i32, i32, i32 } %1443, 0, !dbg !27 + %1445 = bitcast i32 %1444 to <2 x bfloat>, !dbg !27 + %1446 = extractvalue { i32, i32, i32, i32 } %1443, 1, !dbg !27 + %1447 = bitcast i32 %1446 to <2 x bfloat>, !dbg !27 + %1448 = extractvalue { i32, i32, i32, i32 } %1443, 2, !dbg !27 + %1449 = bitcast i32 %1448 to <2 x bfloat>, !dbg !27 + %1450 = extractvalue { i32, i32, i32, i32 } %1443, 3, !dbg !27 + %1451 = bitcast i32 %1450 to <2 x bfloat>, !dbg !27 + %1452 = extractelement <2 x bfloat> %1445, i64 0, !dbg !27 + %1453 = extractelement <2 x bfloat> %1445, i64 1, !dbg !27 + %1454 = extractelement <2 x bfloat> %1447, i64 0, !dbg !27 + %1455 = extractelement <2 x bfloat> %1447, i64 1, !dbg !27 + %1456 = extractelement <2 x bfloat> %1449, i64 0, !dbg !27 + %1457 = extractelement <2 x bfloat> %1449, i64 1, !dbg !27 + %1458 = extractelement <2 x bfloat> %1451, i64 0, !dbg !27 + %1459 = extractelement <2 x bfloat> %1451, i64 1, !dbg !27 + %1460 = extractelement <64 x i1> %62, i64 18, !dbg !27 + %1461 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %303, i1 %1460) #5, !dbg !27 + %1462 = extractvalue { i32, i32, i32, i32 } %1461, 0, !dbg !27 + %1463 = bitcast i32 %1462 to <2 x bfloat>, !dbg !27 + %1464 = extractvalue { i32, i32, i32, i32 } %1461, 1, !dbg !27 + %1465 = bitcast i32 %1464 to <2 x bfloat>, !dbg !27 + %1466 = extractvalue { i32, i32, i32, i32 } %1461, 2, !dbg !27 + %1467 = bitcast i32 %1466 to <2 x bfloat>, !dbg !27 + %1468 = extractvalue { i32, i32, i32, i32 } %1461, 3, !dbg !27 + %1469 = bitcast i32 %1468 to <2 x bfloat>, !dbg !27 + %1470 = extractelement <2 x bfloat> %1463, i64 0, !dbg !27 + %1471 = extractelement <2 x bfloat> %1463, i64 1, !dbg !27 + %1472 = extractelement <2 x bfloat> %1465, i64 0, !dbg !27 + %1473 = extractelement <2 x bfloat> %1465, i64 1, !dbg !27 + %1474 = extractelement <2 x bfloat> %1467, i64 0, !dbg !27 + %1475 = extractelement <2 x bfloat> %1467, i64 1, !dbg !27 + %1476 = extractelement <2 x bfloat> %1469, i64 0, !dbg !27 + %1477 = extractelement <2 x bfloat> %1469, i64 1, !dbg !27 + %1478 = extractelement <64 x i1> %62, i64 17, !dbg !27 + %1479 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %304, i1 %1478) #5, !dbg !27 + %1480 = extractvalue { i32, i32, i32, i32 } %1479, 0, !dbg !27 + %1481 = bitcast i32 %1480 to <2 x bfloat>, !dbg !27 + %1482 = extractvalue { i32, i32, i32, i32 } %1479, 1, !dbg !27 + %1483 = bitcast i32 %1482 to <2 x bfloat>, !dbg !27 + %1484 = extractvalue { i32, i32, i32, i32 } %1479, 2, !dbg !27 + %1485 = bitcast i32 %1484 to <2 x bfloat>, !dbg !27 + %1486 = extractvalue { i32, i32, i32, i32 } %1479, 3, !dbg !27 + %1487 = bitcast i32 %1486 to <2 x bfloat>, !dbg !27 + %1488 = extractelement <2 x bfloat> %1481, i64 0, !dbg !27 + %1489 = extractelement <2 x bfloat> %1481, i64 1, !dbg !27 + %1490 = extractelement <2 x bfloat> %1483, i64 0, !dbg !27 + %1491 = extractelement <2 x bfloat> %1483, i64 1, !dbg !27 + %1492 = extractelement <2 x bfloat> %1485, i64 0, !dbg !27 + %1493 = extractelement <2 x bfloat> %1485, i64 1, !dbg !27 + %1494 = extractelement <2 x bfloat> %1487, i64 0, !dbg !27 + %1495 = extractelement <2 x bfloat> %1487, i64 1, !dbg !27 + %1496 = extractelement <64 x i1> %62, i64 16, !dbg !27 + %1497 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %305, i1 %1496) #5, !dbg !27 + %1498 = extractvalue { i32, i32, i32, i32 } %1497, 0, !dbg !27 + %1499 = bitcast i32 %1498 to <2 x bfloat>, !dbg !27 + %1500 = extractvalue { i32, i32, i32, i32 } %1497, 1, !dbg !27 + %1501 = bitcast i32 %1500 to <2 x bfloat>, !dbg !27 + %1502 = extractvalue { i32, i32, i32, i32 } %1497, 2, !dbg !27 + %1503 = bitcast i32 %1502 to <2 x bfloat>, !dbg !27 + %1504 = extractvalue { i32, i32, i32, i32 } %1497, 3, !dbg !27 + %1505 = bitcast i32 %1504 to <2 x bfloat>, !dbg !27 + %1506 = extractelement <2 x bfloat> %1499, i64 0, !dbg !27 + %1507 = extractelement <2 x bfloat> %1499, i64 1, !dbg !27 + %1508 = extractelement <2 x bfloat> %1501, i64 0, !dbg !27 + %1509 = extractelement <2 x bfloat> %1501, i64 1, !dbg !27 + %1510 = extractelement <2 x bfloat> %1503, i64 0, !dbg !27 + %1511 = extractelement <2 x bfloat> %1503, i64 1, !dbg !27 + %1512 = extractelement <2 x bfloat> %1505, i64 0, !dbg !27 + %1513 = extractelement <2 x bfloat> %1505, i64 1, !dbg !27 + %1514 = extractelement <64 x i1> %62, i64 15, !dbg !27 + %1515 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %306, i1 %1514) #5, !dbg !27 + %1516 = extractvalue { i32, i32, i32, i32 } %1515, 0, !dbg !27 + %1517 = bitcast i32 %1516 to <2 x bfloat>, !dbg !27 + %1518 = extractvalue { i32, i32, i32, i32 } %1515, 1, !dbg !27 + %1519 = bitcast i32 %1518 to <2 x bfloat>, !dbg !27 + %1520 = extractvalue { i32, i32, i32, i32 } %1515, 2, !dbg !27 + %1521 = bitcast i32 %1520 to <2 x bfloat>, !dbg !27 + %1522 = extractvalue { i32, i32, i32, i32 } %1515, 3, !dbg !27 + %1523 = bitcast i32 %1522 to <2 x bfloat>, !dbg !27 + %1524 = extractelement <2 x bfloat> %1517, i64 0, !dbg !27 + %1525 = extractelement <2 x bfloat> %1517, i64 1, !dbg !27 + %1526 = extractelement <2 x bfloat> %1519, i64 0, !dbg !27 + %1527 = extractelement <2 x bfloat> %1519, i64 1, !dbg !27 + %1528 = extractelement <2 x bfloat> %1521, i64 0, !dbg !27 + %1529 = extractelement <2 x bfloat> %1521, i64 1, !dbg !27 + %1530 = extractelement <2 x bfloat> %1523, i64 0, !dbg !27 + %1531 = extractelement <2 x bfloat> %1523, i64 1, !dbg !27 + %1532 = extractelement <64 x i1> %62, i64 14, !dbg !27 + %1533 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %307, i1 %1532) #5, !dbg !27 + %1534 = extractvalue { i32, i32, i32, i32 } %1533, 0, !dbg !27 + %1535 = bitcast i32 %1534 to <2 x bfloat>, !dbg !27 + %1536 = extractvalue { i32, i32, i32, i32 } %1533, 1, !dbg !27 + %1537 = bitcast i32 %1536 to <2 x bfloat>, !dbg !27 + %1538 = extractvalue { i32, i32, i32, i32 } %1533, 2, !dbg !27 + %1539 = bitcast i32 %1538 to <2 x bfloat>, !dbg !27 + %1540 = extractvalue { i32, i32, i32, i32 } %1533, 3, !dbg !27 + %1541 = bitcast i32 %1540 to <2 x bfloat>, !dbg !27 + %1542 = extractelement <2 x bfloat> %1535, i64 0, !dbg !27 + %1543 = extractelement <2 x bfloat> %1535, i64 1, !dbg !27 + %1544 = extractelement <2 x bfloat> %1537, i64 0, !dbg !27 + %1545 = extractelement <2 x bfloat> %1537, i64 1, !dbg !27 + %1546 = extractelement <2 x bfloat> %1539, i64 0, !dbg !27 + %1547 = extractelement <2 x bfloat> %1539, i64 1, !dbg !27 + %1548 = extractelement <2 x bfloat> %1541, i64 0, !dbg !27 + %1549 = extractelement <2 x bfloat> %1541, i64 1, !dbg !27 + %1550 = extractelement <64 x i1> %62, i64 13, !dbg !27 + %1551 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %308, i1 %1550) #5, !dbg !27 + %1552 = extractvalue { i32, i32, i32, i32 } %1551, 0, !dbg !27 + %1553 = bitcast i32 %1552 to <2 x bfloat>, !dbg !27 + %1554 = extractvalue { i32, i32, i32, i32 } %1551, 1, !dbg !27 + %1555 = bitcast i32 %1554 to <2 x bfloat>, !dbg !27 + %1556 = extractvalue { i32, i32, i32, i32 } %1551, 2, !dbg !27 + %1557 = bitcast i32 %1556 to <2 x bfloat>, !dbg !27 + %1558 = extractvalue { i32, i32, i32, i32 } %1551, 3, !dbg !27 + %1559 = bitcast i32 %1558 to <2 x bfloat>, !dbg !27 + %1560 = extractelement <2 x bfloat> %1553, i64 0, !dbg !27 + %1561 = extractelement <2 x bfloat> %1553, i64 1, !dbg !27 + %1562 = extractelement <2 x bfloat> %1555, i64 0, !dbg !27 + %1563 = extractelement <2 x bfloat> %1555, i64 1, !dbg !27 + %1564 = extractelement <2 x bfloat> %1557, i64 0, !dbg !27 + %1565 = extractelement <2 x bfloat> %1557, i64 1, !dbg !27 + %1566 = extractelement <2 x bfloat> %1559, i64 0, !dbg !27 + %1567 = extractelement <2 x bfloat> %1559, i64 1, !dbg !27 + %1568 = extractelement <64 x i1> %62, i64 12, !dbg !27 + %1569 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %309, i1 %1568) #5, !dbg !27 + %1570 = extractvalue { i32, i32, i32, i32 } %1569, 0, !dbg !27 + %1571 = bitcast i32 %1570 to <2 x bfloat>, !dbg !27 + %1572 = extractvalue { i32, i32, i32, i32 } %1569, 1, !dbg !27 + %1573 = bitcast i32 %1572 to <2 x bfloat>, !dbg !27 + %1574 = extractvalue { i32, i32, i32, i32 } %1569, 2, !dbg !27 + %1575 = bitcast i32 %1574 to <2 x bfloat>, !dbg !27 + %1576 = extractvalue { i32, i32, i32, i32 } %1569, 3, !dbg !27 + %1577 = bitcast i32 %1576 to <2 x bfloat>, !dbg !27 + %1578 = extractelement <2 x bfloat> %1571, i64 0, !dbg !27 + %1579 = extractelement <2 x bfloat> %1571, i64 1, !dbg !27 + %1580 = extractelement <2 x bfloat> %1573, i64 0, !dbg !27 + %1581 = extractelement <2 x bfloat> %1573, i64 1, !dbg !27 + %1582 = extractelement <2 x bfloat> %1575, i64 0, !dbg !27 + %1583 = extractelement <2 x bfloat> %1575, i64 1, !dbg !27 + %1584 = extractelement <2 x bfloat> %1577, i64 0, !dbg !27 + %1585 = extractelement <2 x bfloat> %1577, i64 1, !dbg !27 + %1586 = extractelement <64 x i1> %62, i64 11, !dbg !27 + %1587 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %310, i1 %1586) #5, !dbg !27 + %1588 = extractvalue { i32, i32, i32, i32 } %1587, 0, !dbg !27 + %1589 = bitcast i32 %1588 to <2 x bfloat>, !dbg !27 + %1590 = extractvalue { i32, i32, i32, i32 } %1587, 1, !dbg !27 + %1591 = bitcast i32 %1590 to <2 x bfloat>, !dbg !27 + %1592 = extractvalue { i32, i32, i32, i32 } %1587, 2, !dbg !27 + %1593 = bitcast i32 %1592 to <2 x bfloat>, !dbg !27 + %1594 = extractvalue { i32, i32, i32, i32 } %1587, 3, !dbg !27 + %1595 = bitcast i32 %1594 to <2 x bfloat>, !dbg !27 + %1596 = extractelement <2 x bfloat> %1589, i64 0, !dbg !27 + %1597 = extractelement <2 x bfloat> %1589, i64 1, !dbg !27 + %1598 = extractelement <2 x bfloat> %1591, i64 0, !dbg !27 + %1599 = extractelement <2 x bfloat> %1591, i64 1, !dbg !27 + %1600 = extractelement <2 x bfloat> %1593, i64 0, !dbg !27 + %1601 = extractelement <2 x bfloat> %1593, i64 1, !dbg !27 + %1602 = extractelement <2 x bfloat> %1595, i64 0, !dbg !27 + %1603 = extractelement <2 x bfloat> %1595, i64 1, !dbg !27 + %1604 = extractelement <64 x i1> %62, i64 10, !dbg !27 + %1605 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %311, i1 %1604) #5, !dbg !27 + %1606 = extractvalue { i32, i32, i32, i32 } %1605, 0, !dbg !27 + %1607 = bitcast i32 %1606 to <2 x bfloat>, !dbg !27 + %1608 = extractvalue { i32, i32, i32, i32 } %1605, 1, !dbg !27 + %1609 = bitcast i32 %1608 to <2 x bfloat>, !dbg !27 + %1610 = extractvalue { i32, i32, i32, i32 } %1605, 2, !dbg !27 + %1611 = bitcast i32 %1610 to <2 x bfloat>, !dbg !27 + %1612 = extractvalue { i32, i32, i32, i32 } %1605, 3, !dbg !27 + %1613 = bitcast i32 %1612 to <2 x bfloat>, !dbg !27 + %1614 = extractelement <2 x bfloat> %1607, i64 0, !dbg !27 + %1615 = extractelement <2 x bfloat> %1607, i64 1, !dbg !27 + %1616 = extractelement <2 x bfloat> %1609, i64 0, !dbg !27 + %1617 = extractelement <2 x bfloat> %1609, i64 1, !dbg !27 + %1618 = extractelement <2 x bfloat> %1611, i64 0, !dbg !27 + %1619 = extractelement <2 x bfloat> %1611, i64 1, !dbg !27 + %1620 = extractelement <2 x bfloat> %1613, i64 0, !dbg !27 + %1621 = extractelement <2 x bfloat> %1613, i64 1, !dbg !27 + %1622 = extractelement <64 x i1> %62, i64 9, !dbg !27 + %1623 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %312, i1 %1622) #5, !dbg !27 + %1624 = extractvalue { i32, i32, i32, i32 } %1623, 0, !dbg !27 + %1625 = bitcast i32 %1624 to <2 x bfloat>, !dbg !27 + %1626 = extractvalue { i32, i32, i32, i32 } %1623, 1, !dbg !27 + %1627 = bitcast i32 %1626 to <2 x bfloat>, !dbg !27 + %1628 = extractvalue { i32, i32, i32, i32 } %1623, 2, !dbg !27 + %1629 = bitcast i32 %1628 to <2 x bfloat>, !dbg !27 + %1630 = extractvalue { i32, i32, i32, i32 } %1623, 3, !dbg !27 + %1631 = bitcast i32 %1630 to <2 x bfloat>, !dbg !27 + %1632 = extractelement <2 x bfloat> %1625, i64 0, !dbg !27 + %1633 = extractelement <2 x bfloat> %1625, i64 1, !dbg !27 + %1634 = extractelement <2 x bfloat> %1627, i64 0, !dbg !27 + %1635 = extractelement <2 x bfloat> %1627, i64 1, !dbg !27 + %1636 = extractelement <2 x bfloat> %1629, i64 0, !dbg !27 + %1637 = extractelement <2 x bfloat> %1629, i64 1, !dbg !27 + %1638 = extractelement <2 x bfloat> %1631, i64 0, !dbg !27 + %1639 = extractelement <2 x bfloat> %1631, i64 1, !dbg !27 + %1640 = extractelement <64 x i1> %62, i64 8, !dbg !27 + %1641 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %313, i1 %1640) #5, !dbg !27 + %1642 = extractvalue { i32, i32, i32, i32 } %1641, 0, !dbg !27 + %1643 = bitcast i32 %1642 to <2 x bfloat>, !dbg !27 + %1644 = extractvalue { i32, i32, i32, i32 } %1641, 1, !dbg !27 + %1645 = bitcast i32 %1644 to <2 x bfloat>, !dbg !27 + %1646 = extractvalue { i32, i32, i32, i32 } %1641, 2, !dbg !27 + %1647 = bitcast i32 %1646 to <2 x bfloat>, !dbg !27 + %1648 = extractvalue { i32, i32, i32, i32 } %1641, 3, !dbg !27 + %1649 = bitcast i32 %1648 to <2 x bfloat>, !dbg !27 + %1650 = extractelement <2 x bfloat> %1643, i64 0, !dbg !27 + %1651 = extractelement <2 x bfloat> %1643, i64 1, !dbg !27 + %1652 = extractelement <2 x bfloat> %1645, i64 0, !dbg !27 + %1653 = extractelement <2 x bfloat> %1645, i64 1, !dbg !27 + %1654 = extractelement <2 x bfloat> %1647, i64 0, !dbg !27 + %1655 = extractelement <2 x bfloat> %1647, i64 1, !dbg !27 + %1656 = extractelement <2 x bfloat> %1649, i64 0, !dbg !27 + %1657 = extractelement <2 x bfloat> %1649, i64 1, !dbg !27 + %1658 = extractelement <64 x i1> %62, i64 7, !dbg !27 + %1659 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %314, i1 %1658) #5, !dbg !27 + %1660 = extractvalue { i32, i32, i32, i32 } %1659, 0, !dbg !27 + %1661 = bitcast i32 %1660 to <2 x bfloat>, !dbg !27 + %1662 = extractvalue { i32, i32, i32, i32 } %1659, 1, !dbg !27 + %1663 = bitcast i32 %1662 to <2 x bfloat>, !dbg !27 + %1664 = extractvalue { i32, i32, i32, i32 } %1659, 2, !dbg !27 + %1665 = bitcast i32 %1664 to <2 x bfloat>, !dbg !27 + %1666 = extractvalue { i32, i32, i32, i32 } %1659, 3, !dbg !27 + %1667 = bitcast i32 %1666 to <2 x bfloat>, !dbg !27 + %1668 = extractelement <2 x bfloat> %1661, i64 0, !dbg !27 + %1669 = extractelement <2 x bfloat> %1661, i64 1, !dbg !27 + %1670 = extractelement <2 x bfloat> %1663, i64 0, !dbg !27 + %1671 = extractelement <2 x bfloat> %1663, i64 1, !dbg !27 + %1672 = extractelement <2 x bfloat> %1665, i64 0, !dbg !27 + %1673 = extractelement <2 x bfloat> %1665, i64 1, !dbg !27 + %1674 = extractelement <2 x bfloat> %1667, i64 0, !dbg !27 + %1675 = extractelement <2 x bfloat> %1667, i64 1, !dbg !27 + %1676 = extractelement <64 x i1> %62, i64 6, !dbg !27 + %1677 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %315, i1 %1676) #5, !dbg !27 + %1678 = extractvalue { i32, i32, i32, i32 } %1677, 0, !dbg !27 + %1679 = bitcast i32 %1678 to <2 x bfloat>, !dbg !27 + %1680 = extractvalue { i32, i32, i32, i32 } %1677, 1, !dbg !27 + %1681 = bitcast i32 %1680 to <2 x bfloat>, !dbg !27 + %1682 = extractvalue { i32, i32, i32, i32 } %1677, 2, !dbg !27 + %1683 = bitcast i32 %1682 to <2 x bfloat>, !dbg !27 + %1684 = extractvalue { i32, i32, i32, i32 } %1677, 3, !dbg !27 + %1685 = bitcast i32 %1684 to <2 x bfloat>, !dbg !27 + %1686 = extractelement <2 x bfloat> %1679, i64 0, !dbg !27 + %1687 = extractelement <2 x bfloat> %1679, i64 1, !dbg !27 + %1688 = extractelement <2 x bfloat> %1681, i64 0, !dbg !27 + %1689 = extractelement <2 x bfloat> %1681, i64 1, !dbg !27 + %1690 = extractelement <2 x bfloat> %1683, i64 0, !dbg !27 + %1691 = extractelement <2 x bfloat> %1683, i64 1, !dbg !27 + %1692 = extractelement <2 x bfloat> %1685, i64 0, !dbg !27 + %1693 = extractelement <2 x bfloat> %1685, i64 1, !dbg !27 + %1694 = extractelement <64 x i1> %62, i64 5, !dbg !27 + %1695 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %316, i1 %1694) #5, !dbg !27 + %1696 = extractvalue { i32, i32, i32, i32 } %1695, 0, !dbg !27 + %1697 = bitcast i32 %1696 to <2 x bfloat>, !dbg !27 + %1698 = extractvalue { i32, i32, i32, i32 } %1695, 1, !dbg !27 + %1699 = bitcast i32 %1698 to <2 x bfloat>, !dbg !27 + %1700 = extractvalue { i32, i32, i32, i32 } %1695, 2, !dbg !27 + %1701 = bitcast i32 %1700 to <2 x bfloat>, !dbg !27 + %1702 = extractvalue { i32, i32, i32, i32 } %1695, 3, !dbg !27 + %1703 = bitcast i32 %1702 to <2 x bfloat>, !dbg !27 + %1704 = extractelement <2 x bfloat> %1697, i64 0, !dbg !27 + %1705 = extractelement <2 x bfloat> %1697, i64 1, !dbg !27 + %1706 = extractelement <2 x bfloat> %1699, i64 0, !dbg !27 + %1707 = extractelement <2 x bfloat> %1699, i64 1, !dbg !27 + %1708 = extractelement <2 x bfloat> %1701, i64 0, !dbg !27 + %1709 = extractelement <2 x bfloat> %1701, i64 1, !dbg !27 + %1710 = extractelement <2 x bfloat> %1703, i64 0, !dbg !27 + %1711 = extractelement <2 x bfloat> %1703, i64 1, !dbg !27 + %1712 = extractelement <64 x i1> %62, i64 4, !dbg !27 + %1713 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %317, i1 %1712) #5, !dbg !27 + %1714 = extractvalue { i32, i32, i32, i32 } %1713, 0, !dbg !27 + %1715 = bitcast i32 %1714 to <2 x bfloat>, !dbg !27 + %1716 = extractvalue { i32, i32, i32, i32 } %1713, 1, !dbg !27 + %1717 = bitcast i32 %1716 to <2 x bfloat>, !dbg !27 + %1718 = extractvalue { i32, i32, i32, i32 } %1713, 2, !dbg !27 + %1719 = bitcast i32 %1718 to <2 x bfloat>, !dbg !27 + %1720 = extractvalue { i32, i32, i32, i32 } %1713, 3, !dbg !27 + %1721 = bitcast i32 %1720 to <2 x bfloat>, !dbg !27 + %1722 = extractelement <2 x bfloat> %1715, i64 0, !dbg !27 + %1723 = extractelement <2 x bfloat> %1715, i64 1, !dbg !27 + %1724 = extractelement <2 x bfloat> %1717, i64 0, !dbg !27 + %1725 = extractelement <2 x bfloat> %1717, i64 1, !dbg !27 + %1726 = extractelement <2 x bfloat> %1719, i64 0, !dbg !27 + %1727 = extractelement <2 x bfloat> %1719, i64 1, !dbg !27 + %1728 = extractelement <2 x bfloat> %1721, i64 0, !dbg !27 + %1729 = extractelement <2 x bfloat> %1721, i64 1, !dbg !27 + %1730 = fpext bfloat %591 to float, !dbg !28 + %1731 = fpext bfloat %592 to float, !dbg !28 + %1732 = fpext bfloat %593 to float, !dbg !28 + %1733 = fpext bfloat %594 to float, !dbg !28 + %1734 = fpext bfloat %595 to float, !dbg !28 + %1735 = fpext bfloat %596 to float, !dbg !28 + %1736 = fpext bfloat %597 to float, !dbg !28 + %1737 = fpext bfloat %598 to float, !dbg !28 + %1738 = fpext bfloat %608 to float, !dbg !28 + %1739 = fpext bfloat %609 to float, !dbg !28 + %1740 = fpext bfloat %610 to float, !dbg !28 + %1741 = fpext bfloat %611 to float, !dbg !28 + %1742 = fpext bfloat %612 to float, !dbg !28 + %1743 = fpext bfloat %613 to float, !dbg !28 + %1744 = fpext bfloat %614 to float, !dbg !28 + %1745 = fpext bfloat %615 to float, !dbg !28 + %1746 = fpext bfloat %625 to float, !dbg !28 + %1747 = fpext bfloat %626 to float, !dbg !28 + %1748 = fpext bfloat %627 to float, !dbg !28 + %1749 = fpext bfloat %628 to float, !dbg !28 + %1750 = fpext bfloat %629 to float, !dbg !28 + %1751 = fpext bfloat %630 to float, !dbg !28 + %1752 = fpext bfloat %631 to float, !dbg !28 + %1753 = fpext bfloat %632 to float, !dbg !28 + %1754 = fpext bfloat %642 to float, !dbg !28 + %1755 = fpext bfloat %643 to float, !dbg !28 + %1756 = fpext bfloat %644 to float, !dbg !28 + %1757 = fpext bfloat %645 to float, !dbg !28 + %1758 = fpext bfloat %646 to float, !dbg !28 + %1759 = fpext bfloat %647 to float, !dbg !28 + %1760 = fpext bfloat %648 to float, !dbg !28 + %1761 = fpext bfloat %649 to float, !dbg !28 + %1762 = fpext bfloat %660 to float, !dbg !28 + %1763 = fpext bfloat %661 to float, !dbg !28 + %1764 = fpext bfloat %662 to float, !dbg !28 + %1765 = fpext bfloat %663 to float, !dbg !28 + %1766 = fpext bfloat %664 to float, !dbg !28 + %1767 = fpext bfloat %665 to float, !dbg !28 + %1768 = fpext bfloat %666 to float, !dbg !28 + %1769 = fpext bfloat %667 to float, !dbg !28 + %1770 = fpext bfloat %678 to float, !dbg !28 + %1771 = fpext bfloat %679 to float, !dbg !28 + %1772 = fpext bfloat %680 to float, !dbg !28 + %1773 = fpext bfloat %681 to float, !dbg !28 + %1774 = fpext bfloat %682 to float, !dbg !28 + %1775 = fpext bfloat %683 to float, !dbg !28 + %1776 = fpext bfloat %684 to float, !dbg !28 + %1777 = fpext bfloat %685 to float, !dbg !28 + %1778 = fpext bfloat %696 to float, !dbg !28 + %1779 = fpext bfloat %697 to float, !dbg !28 + %1780 = fpext bfloat %698 to float, !dbg !28 + %1781 = fpext bfloat %699 to float, !dbg !28 + %1782 = fpext bfloat %700 to float, !dbg !28 + %1783 = fpext bfloat %701 to float, !dbg !28 + %1784 = fpext bfloat %702 to float, !dbg !28 + %1785 = fpext bfloat %703 to float, !dbg !28 + %1786 = fpext bfloat %714 to float, !dbg !28 + %1787 = fpext bfloat %715 to float, !dbg !28 + %1788 = fpext bfloat %716 to float, !dbg !28 + %1789 = fpext bfloat %717 to float, !dbg !28 + %1790 = fpext bfloat %718 to float, !dbg !28 + %1791 = fpext bfloat %719 to float, !dbg !28 + %1792 = fpext bfloat %720 to float, !dbg !28 + %1793 = fpext bfloat %721 to float, !dbg !28 + %1794 = fpext bfloat %732 to float, !dbg !28 + %1795 = fpext bfloat %733 to float, !dbg !28 + %1796 = fpext bfloat %734 to float, !dbg !28 + %1797 = fpext bfloat %735 to float, !dbg !28 + %1798 = fpext bfloat %736 to float, !dbg !28 + %1799 = fpext bfloat %737 to float, !dbg !28 + %1800 = fpext bfloat %738 to float, !dbg !28 + %1801 = fpext bfloat %739 to float, !dbg !28 + %1802 = fpext bfloat %750 to float, !dbg !28 + %1803 = fpext bfloat %751 to float, !dbg !28 + %1804 = fpext bfloat %752 to float, !dbg !28 + %1805 = fpext bfloat %753 to float, !dbg !28 + %1806 = fpext bfloat %754 to float, !dbg !28 + %1807 = fpext bfloat %755 to float, !dbg !28 + %1808 = fpext bfloat %756 to float, !dbg !28 + %1809 = fpext bfloat %757 to float, !dbg !28 + %1810 = fpext bfloat %768 to float, !dbg !28 + %1811 = fpext bfloat %769 to float, !dbg !28 + %1812 = fpext bfloat %770 to float, !dbg !28 + %1813 = fpext bfloat %771 to float, !dbg !28 + %1814 = fpext bfloat %772 to float, !dbg !28 + %1815 = fpext bfloat %773 to float, !dbg !28 + %1816 = fpext bfloat %774 to float, !dbg !28 + %1817 = fpext bfloat %775 to float, !dbg !28 + %1818 = fpext bfloat %786 to float, !dbg !28 + %1819 = fpext bfloat %787 to float, !dbg !28 + %1820 = fpext bfloat %788 to float, !dbg !28 + %1821 = fpext bfloat %789 to float, !dbg !28 + %1822 = fpext bfloat %790 to float, !dbg !28 + %1823 = fpext bfloat %791 to float, !dbg !28 + %1824 = fpext bfloat %792 to float, !dbg !28 + %1825 = fpext bfloat %793 to float, !dbg !28 + %1826 = fpext bfloat %804 to float, !dbg !28 + %1827 = fpext bfloat %805 to float, !dbg !28 + %1828 = fpext bfloat %806 to float, !dbg !28 + %1829 = fpext bfloat %807 to float, !dbg !28 + %1830 = fpext bfloat %808 to float, !dbg !28 + %1831 = fpext bfloat %809 to float, !dbg !28 + %1832 = fpext bfloat %810 to float, !dbg !28 + %1833 = fpext bfloat %811 to float, !dbg !28 + %1834 = fpext bfloat %822 to float, !dbg !28 + %1835 = fpext bfloat %823 to float, !dbg !28 + %1836 = fpext bfloat %824 to float, !dbg !28 + %1837 = fpext bfloat %825 to float, !dbg !28 + %1838 = fpext bfloat %826 to float, !dbg !28 + %1839 = fpext bfloat %827 to float, !dbg !28 + %1840 = fpext bfloat %828 to float, !dbg !28 + %1841 = fpext bfloat %829 to float, !dbg !28 + %1842 = fpext bfloat %840 to float, !dbg !28 + %1843 = fpext bfloat %841 to float, !dbg !28 + %1844 = fpext bfloat %842 to float, !dbg !28 + %1845 = fpext bfloat %843 to float, !dbg !28 + %1846 = fpext bfloat %844 to float, !dbg !28 + %1847 = fpext bfloat %845 to float, !dbg !28 + %1848 = fpext bfloat %846 to float, !dbg !28 + %1849 = fpext bfloat %847 to float, !dbg !28 + %1850 = fpext bfloat %858 to float, !dbg !28 + %1851 = fpext bfloat %859 to float, !dbg !28 + %1852 = fpext bfloat %860 to float, !dbg !28 + %1853 = fpext bfloat %861 to float, !dbg !28 + %1854 = fpext bfloat %862 to float, !dbg !28 + %1855 = fpext bfloat %863 to float, !dbg !28 + %1856 = fpext bfloat %864 to float, !dbg !28 + %1857 = fpext bfloat %865 to float, !dbg !28 + %1858 = fpext bfloat %876 to float, !dbg !28 + %1859 = fpext bfloat %877 to float, !dbg !28 + %1860 = fpext bfloat %878 to float, !dbg !28 + %1861 = fpext bfloat %879 to float, !dbg !28 + %1862 = fpext bfloat %880 to float, !dbg !28 + %1863 = fpext bfloat %881 to float, !dbg !28 + %1864 = fpext bfloat %882 to float, !dbg !28 + %1865 = fpext bfloat %883 to float, !dbg !28 + %1866 = fpext bfloat %894 to float, !dbg !28 + %1867 = fpext bfloat %895 to float, !dbg !28 + %1868 = fpext bfloat %896 to float, !dbg !28 + %1869 = fpext bfloat %897 to float, !dbg !28 + %1870 = fpext bfloat %898 to float, !dbg !28 + %1871 = fpext bfloat %899 to float, !dbg !28 + %1872 = fpext bfloat %900 to float, !dbg !28 + %1873 = fpext bfloat %901 to float, !dbg !28 + %1874 = fpext bfloat %912 to float, !dbg !28 + %1875 = fpext bfloat %913 to float, !dbg !28 + %1876 = fpext bfloat %914 to float, !dbg !28 + %1877 = fpext bfloat %915 to float, !dbg !28 + %1878 = fpext bfloat %916 to float, !dbg !28 + %1879 = fpext bfloat %917 to float, !dbg !28 + %1880 = fpext bfloat %918 to float, !dbg !28 + %1881 = fpext bfloat %919 to float, !dbg !28 + %1882 = fpext bfloat %930 to float, !dbg !28 + %1883 = fpext bfloat %931 to float, !dbg !28 + %1884 = fpext bfloat %932 to float, !dbg !28 + %1885 = fpext bfloat %933 to float, !dbg !28 + %1886 = fpext bfloat %934 to float, !dbg !28 + %1887 = fpext bfloat %935 to float, !dbg !28 + %1888 = fpext bfloat %936 to float, !dbg !28 + %1889 = fpext bfloat %937 to float, !dbg !28 + %1890 = fpext bfloat %948 to float, !dbg !28 + %1891 = fpext bfloat %949 to float, !dbg !28 + %1892 = fpext bfloat %950 to float, !dbg !28 + %1893 = fpext bfloat %951 to float, !dbg !28 + %1894 = fpext bfloat %952 to float, !dbg !28 + %1895 = fpext bfloat %953 to float, !dbg !28 + %1896 = fpext bfloat %954 to float, !dbg !28 + %1897 = fpext bfloat %955 to float, !dbg !28 + %1898 = fpext bfloat %966 to float, !dbg !28 + %1899 = fpext bfloat %967 to float, !dbg !28 + %1900 = fpext bfloat %968 to float, !dbg !28 + %1901 = fpext bfloat %969 to float, !dbg !28 + %1902 = fpext bfloat %970 to float, !dbg !28 + %1903 = fpext bfloat %971 to float, !dbg !28 + %1904 = fpext bfloat %972 to float, !dbg !28 + %1905 = fpext bfloat %973 to float, !dbg !28 + %1906 = fpext bfloat %984 to float, !dbg !28 + %1907 = fpext bfloat %985 to float, !dbg !28 + %1908 = fpext bfloat %986 to float, !dbg !28 + %1909 = fpext bfloat %987 to float, !dbg !28 + %1910 = fpext bfloat %988 to float, !dbg !28 + %1911 = fpext bfloat %989 to float, !dbg !28 + %1912 = fpext bfloat %990 to float, !dbg !28 + %1913 = fpext bfloat %991 to float, !dbg !28 + %1914 = fpext bfloat %1002 to float, !dbg !28 + %1915 = fpext bfloat %1003 to float, !dbg !28 + %1916 = fpext bfloat %1004 to float, !dbg !28 + %1917 = fpext bfloat %1005 to float, !dbg !28 + %1918 = fpext bfloat %1006 to float, !dbg !28 + %1919 = fpext bfloat %1007 to float, !dbg !28 + %1920 = fpext bfloat %1008 to float, !dbg !28 + %1921 = fpext bfloat %1009 to float, !dbg !28 + %1922 = fpext bfloat %1020 to float, !dbg !28 + %1923 = fpext bfloat %1021 to float, !dbg !28 + %1924 = fpext bfloat %1022 to float, !dbg !28 + %1925 = fpext bfloat %1023 to float, !dbg !28 + %1926 = fpext bfloat %1024 to float, !dbg !28 + %1927 = fpext bfloat %1025 to float, !dbg !28 + %1928 = fpext bfloat %1026 to float, !dbg !28 + %1929 = fpext bfloat %1027 to float, !dbg !28 + %1930 = fpext bfloat %1038 to float, !dbg !28 + %1931 = fpext bfloat %1039 to float, !dbg !28 + %1932 = fpext bfloat %1040 to float, !dbg !28 + %1933 = fpext bfloat %1041 to float, !dbg !28 + %1934 = fpext bfloat %1042 to float, !dbg !28 + %1935 = fpext bfloat %1043 to float, !dbg !28 + %1936 = fpext bfloat %1044 to float, !dbg !28 + %1937 = fpext bfloat %1045 to float, !dbg !28 + %1938 = fpext bfloat %1056 to float, !dbg !28 + %1939 = fpext bfloat %1057 to float, !dbg !28 + %1940 = fpext bfloat %1058 to float, !dbg !28 + %1941 = fpext bfloat %1059 to float, !dbg !28 + %1942 = fpext bfloat %1060 to float, !dbg !28 + %1943 = fpext bfloat %1061 to float, !dbg !28 + %1944 = fpext bfloat %1062 to float, !dbg !28 + %1945 = fpext bfloat %1063 to float, !dbg !28 + %1946 = fpext bfloat %1074 to float, !dbg !28 + %1947 = fpext bfloat %1075 to float, !dbg !28 + %1948 = fpext bfloat %1076 to float, !dbg !28 + %1949 = fpext bfloat %1077 to float, !dbg !28 + %1950 = fpext bfloat %1078 to float, !dbg !28 + %1951 = fpext bfloat %1079 to float, !dbg !28 + %1952 = fpext bfloat %1080 to float, !dbg !28 + %1953 = fpext bfloat %1081 to float, !dbg !28 + %1954 = fpext bfloat %1092 to float, !dbg !28 + %1955 = fpext bfloat %1093 to float, !dbg !28 + %1956 = fpext bfloat %1094 to float, !dbg !28 + %1957 = fpext bfloat %1095 to float, !dbg !28 + %1958 = fpext bfloat %1096 to float, !dbg !28 + %1959 = fpext bfloat %1097 to float, !dbg !28 + %1960 = fpext bfloat %1098 to float, !dbg !28 + %1961 = fpext bfloat %1099 to float, !dbg !28 + %1962 = fpext bfloat %1110 to float, !dbg !28 + %1963 = fpext bfloat %1111 to float, !dbg !28 + %1964 = fpext bfloat %1112 to float, !dbg !28 + %1965 = fpext bfloat %1113 to float, !dbg !28 + %1966 = fpext bfloat %1114 to float, !dbg !28 + %1967 = fpext bfloat %1115 to float, !dbg !28 + %1968 = fpext bfloat %1116 to float, !dbg !28 + %1969 = fpext bfloat %1117 to float, !dbg !28 + %1970 = fpext bfloat %1128 to float, !dbg !28 + %1971 = fpext bfloat %1129 to float, !dbg !28 + %1972 = fpext bfloat %1130 to float, !dbg !28 + %1973 = fpext bfloat %1131 to float, !dbg !28 + %1974 = fpext bfloat %1132 to float, !dbg !28 + %1975 = fpext bfloat %1133 to float, !dbg !28 + %1976 = fpext bfloat %1134 to float, !dbg !28 + %1977 = fpext bfloat %1135 to float, !dbg !28 + %1978 = fpext bfloat %1146 to float, !dbg !28 + %1979 = fpext bfloat %1147 to float, !dbg !28 + %1980 = fpext bfloat %1148 to float, !dbg !28 + %1981 = fpext bfloat %1149 to float, !dbg !28 + %1982 = fpext bfloat %1150 to float, !dbg !28 + %1983 = fpext bfloat %1151 to float, !dbg !28 + %1984 = fpext bfloat %1152 to float, !dbg !28 + %1985 = fpext bfloat %1153 to float, !dbg !28 + %1986 = fpext bfloat %1164 to float, !dbg !28 + %1987 = fpext bfloat %1165 to float, !dbg !28 + %1988 = fpext bfloat %1166 to float, !dbg !28 + %1989 = fpext bfloat %1167 to float, !dbg !28 + %1990 = fpext bfloat %1168 to float, !dbg !28 + %1991 = fpext bfloat %1169 to float, !dbg !28 + %1992 = fpext bfloat %1170 to float, !dbg !28 + %1993 = fpext bfloat %1171 to float, !dbg !28 + %1994 = fpext bfloat %1182 to float, !dbg !28 + %1995 = fpext bfloat %1183 to float, !dbg !28 + %1996 = fpext bfloat %1184 to float, !dbg !28 + %1997 = fpext bfloat %1185 to float, !dbg !28 + %1998 = fpext bfloat %1186 to float, !dbg !28 + %1999 = fpext bfloat %1187 to float, !dbg !28 + %2000 = fpext bfloat %1188 to float, !dbg !28 + %2001 = fpext bfloat %1189 to float, !dbg !28 + %2002 = fpext bfloat %1200 to float, !dbg !28 + %2003 = fpext bfloat %1201 to float, !dbg !28 + %2004 = fpext bfloat %1202 to float, !dbg !28 + %2005 = fpext bfloat %1203 to float, !dbg !28 + %2006 = fpext bfloat %1204 to float, !dbg !28 + %2007 = fpext bfloat %1205 to float, !dbg !28 + %2008 = fpext bfloat %1206 to float, !dbg !28 + %2009 = fpext bfloat %1207 to float, !dbg !28 + %2010 = fpext bfloat %1218 to float, !dbg !28 + %2011 = fpext bfloat %1219 to float, !dbg !28 + %2012 = fpext bfloat %1220 to float, !dbg !28 + %2013 = fpext bfloat %1221 to float, !dbg !28 + %2014 = fpext bfloat %1222 to float, !dbg !28 + %2015 = fpext bfloat %1223 to float, !dbg !28 + %2016 = fpext bfloat %1224 to float, !dbg !28 + %2017 = fpext bfloat %1225 to float, !dbg !28 + %2018 = fpext bfloat %1236 to float, !dbg !28 + %2019 = fpext bfloat %1237 to float, !dbg !28 + %2020 = fpext bfloat %1238 to float, !dbg !28 + %2021 = fpext bfloat %1239 to float, !dbg !28 + %2022 = fpext bfloat %1240 to float, !dbg !28 + %2023 = fpext bfloat %1241 to float, !dbg !28 + %2024 = fpext bfloat %1242 to float, !dbg !28 + %2025 = fpext bfloat %1243 to float, !dbg !28 + %2026 = fpext bfloat %1254 to float, !dbg !28 + %2027 = fpext bfloat %1255 to float, !dbg !28 + %2028 = fpext bfloat %1256 to float, !dbg !28 + %2029 = fpext bfloat %1257 to float, !dbg !28 + %2030 = fpext bfloat %1258 to float, !dbg !28 + %2031 = fpext bfloat %1259 to float, !dbg !28 + %2032 = fpext bfloat %1260 to float, !dbg !28 + %2033 = fpext bfloat %1261 to float, !dbg !28 + %2034 = fpext bfloat %1272 to float, !dbg !28 + %2035 = fpext bfloat %1273 to float, !dbg !28 + %2036 = fpext bfloat %1274 to float, !dbg !28 + %2037 = fpext bfloat %1275 to float, !dbg !28 + %2038 = fpext bfloat %1276 to float, !dbg !28 + %2039 = fpext bfloat %1277 to float, !dbg !28 + %2040 = fpext bfloat %1278 to float, !dbg !28 + %2041 = fpext bfloat %1279 to float, !dbg !28 + %2042 = fpext bfloat %1290 to float, !dbg !28 + %2043 = fpext bfloat %1291 to float, !dbg !28 + %2044 = fpext bfloat %1292 to float, !dbg !28 + %2045 = fpext bfloat %1293 to float, !dbg !28 + %2046 = fpext bfloat %1294 to float, !dbg !28 + %2047 = fpext bfloat %1295 to float, !dbg !28 + %2048 = fpext bfloat %1296 to float, !dbg !28 + %2049 = fpext bfloat %1297 to float, !dbg !28 + %2050 = fpext bfloat %1308 to float, !dbg !28 + %2051 = fpext bfloat %1309 to float, !dbg !28 + %2052 = fpext bfloat %1310 to float, !dbg !28 + %2053 = fpext bfloat %1311 to float, !dbg !28 + %2054 = fpext bfloat %1312 to float, !dbg !28 + %2055 = fpext bfloat %1313 to float, !dbg !28 + %2056 = fpext bfloat %1314 to float, !dbg !28 + %2057 = fpext bfloat %1315 to float, !dbg !28 + %2058 = fpext bfloat %1326 to float, !dbg !28 + %2059 = fpext bfloat %1327 to float, !dbg !28 + %2060 = fpext bfloat %1328 to float, !dbg !28 + %2061 = fpext bfloat %1329 to float, !dbg !28 + %2062 = fpext bfloat %1330 to float, !dbg !28 + %2063 = fpext bfloat %1331 to float, !dbg !28 + %2064 = fpext bfloat %1332 to float, !dbg !28 + %2065 = fpext bfloat %1333 to float, !dbg !28 + %2066 = fpext bfloat %1344 to float, !dbg !28 + %2067 = fpext bfloat %1345 to float, !dbg !28 + %2068 = fpext bfloat %1346 to float, !dbg !28 + %2069 = fpext bfloat %1347 to float, !dbg !28 + %2070 = fpext bfloat %1348 to float, !dbg !28 + %2071 = fpext bfloat %1349 to float, !dbg !28 + %2072 = fpext bfloat %1350 to float, !dbg !28 + %2073 = fpext bfloat %1351 to float, !dbg !28 + %2074 = fpext bfloat %1362 to float, !dbg !28 + %2075 = fpext bfloat %1363 to float, !dbg !28 + %2076 = fpext bfloat %1364 to float, !dbg !28 + %2077 = fpext bfloat %1365 to float, !dbg !28 + %2078 = fpext bfloat %1366 to float, !dbg !28 + %2079 = fpext bfloat %1367 to float, !dbg !28 + %2080 = fpext bfloat %1368 to float, !dbg !28 + %2081 = fpext bfloat %1369 to float, !dbg !28 + %2082 = fpext bfloat %1380 to float, !dbg !28 + %2083 = fpext bfloat %1381 to float, !dbg !28 + %2084 = fpext bfloat %1382 to float, !dbg !28 + %2085 = fpext bfloat %1383 to float, !dbg !28 + %2086 = fpext bfloat %1384 to float, !dbg !28 + %2087 = fpext bfloat %1385 to float, !dbg !28 + %2088 = fpext bfloat %1386 to float, !dbg !28 + %2089 = fpext bfloat %1387 to float, !dbg !28 + %2090 = fpext bfloat %1398 to float, !dbg !28 + %2091 = fpext bfloat %1399 to float, !dbg !28 + %2092 = fpext bfloat %1400 to float, !dbg !28 + %2093 = fpext bfloat %1401 to float, !dbg !28 + %2094 = fpext bfloat %1402 to float, !dbg !28 + %2095 = fpext bfloat %1403 to float, !dbg !28 + %2096 = fpext bfloat %1404 to float, !dbg !28 + %2097 = fpext bfloat %1405 to float, !dbg !28 + %2098 = fpext bfloat %1416 to float, !dbg !28 + %2099 = fpext bfloat %1417 to float, !dbg !28 + %2100 = fpext bfloat %1418 to float, !dbg !28 + %2101 = fpext bfloat %1419 to float, !dbg !28 + %2102 = fpext bfloat %1420 to float, !dbg !28 + %2103 = fpext bfloat %1421 to float, !dbg !28 + %2104 = fpext bfloat %1422 to float, !dbg !28 + %2105 = fpext bfloat %1423 to float, !dbg !28 + %2106 = fpext bfloat %1434 to float, !dbg !28 + %2107 = fpext bfloat %1435 to float, !dbg !28 + %2108 = fpext bfloat %1436 to float, !dbg !28 + %2109 = fpext bfloat %1437 to float, !dbg !28 + %2110 = fpext bfloat %1438 to float, !dbg !28 + %2111 = fpext bfloat %1439 to float, !dbg !28 + %2112 = fpext bfloat %1440 to float, !dbg !28 + %2113 = fpext bfloat %1441 to float, !dbg !28 + %2114 = fpext bfloat %1452 to float, !dbg !28 + %2115 = fpext bfloat %1453 to float, !dbg !28 + %2116 = fpext bfloat %1454 to float, !dbg !28 + %2117 = fpext bfloat %1455 to float, !dbg !28 + %2118 = fpext bfloat %1456 to float, !dbg !28 + %2119 = fpext bfloat %1457 to float, !dbg !28 + %2120 = fpext bfloat %1458 to float, !dbg !28 + %2121 = fpext bfloat %1459 to float, !dbg !28 + %2122 = fpext bfloat %1470 to float, !dbg !28 + %2123 = fpext bfloat %1471 to float, !dbg !28 + %2124 = fpext bfloat %1472 to float, !dbg !28 + %2125 = fpext bfloat %1473 to float, !dbg !28 + %2126 = fpext bfloat %1474 to float, !dbg !28 + %2127 = fpext bfloat %1475 to float, !dbg !28 + %2128 = fpext bfloat %1476 to float, !dbg !28 + %2129 = fpext bfloat %1477 to float, !dbg !28 + %2130 = fpext bfloat %1488 to float, !dbg !28 + %2131 = fpext bfloat %1489 to float, !dbg !28 + %2132 = fpext bfloat %1490 to float, !dbg !28 + %2133 = fpext bfloat %1491 to float, !dbg !28 + %2134 = fpext bfloat %1492 to float, !dbg !28 + %2135 = fpext bfloat %1493 to float, !dbg !28 + %2136 = fpext bfloat %1494 to float, !dbg !28 + %2137 = fpext bfloat %1495 to float, !dbg !28 + %2138 = fpext bfloat %1506 to float, !dbg !28 + %2139 = fpext bfloat %1507 to float, !dbg !28 + %2140 = fpext bfloat %1508 to float, !dbg !28 + %2141 = fpext bfloat %1509 to float, !dbg !28 + %2142 = fpext bfloat %1510 to float, !dbg !28 + %2143 = fpext bfloat %1511 to float, !dbg !28 + %2144 = fpext bfloat %1512 to float, !dbg !28 + %2145 = fpext bfloat %1513 to float, !dbg !28 + %2146 = fpext bfloat %1524 to float, !dbg !28 + %2147 = fpext bfloat %1525 to float, !dbg !28 + %2148 = fpext bfloat %1526 to float, !dbg !28 + %2149 = fpext bfloat %1527 to float, !dbg !28 + %2150 = fpext bfloat %1528 to float, !dbg !28 + %2151 = fpext bfloat %1529 to float, !dbg !28 + %2152 = fpext bfloat %1530 to float, !dbg !28 + %2153 = fpext bfloat %1531 to float, !dbg !28 + %2154 = fpext bfloat %1542 to float, !dbg !28 + %2155 = fpext bfloat %1543 to float, !dbg !28 + %2156 = fpext bfloat %1544 to float, !dbg !28 + %2157 = fpext bfloat %1545 to float, !dbg !28 + %2158 = fpext bfloat %1546 to float, !dbg !28 + %2159 = fpext bfloat %1547 to float, !dbg !28 + %2160 = fpext bfloat %1548 to float, !dbg !28 + %2161 = fpext bfloat %1549 to float, !dbg !28 + %2162 = fpext bfloat %1560 to float, !dbg !28 + %2163 = fpext bfloat %1561 to float, !dbg !28 + %2164 = fpext bfloat %1562 to float, !dbg !28 + %2165 = fpext bfloat %1563 to float, !dbg !28 + %2166 = fpext bfloat %1564 to float, !dbg !28 + %2167 = fpext bfloat %1565 to float, !dbg !28 + %2168 = fpext bfloat %1566 to float, !dbg !28 + %2169 = fpext bfloat %1567 to float, !dbg !28 + %2170 = fpext bfloat %1578 to float, !dbg !28 + %2171 = fpext bfloat %1579 to float, !dbg !28 + %2172 = fpext bfloat %1580 to float, !dbg !28 + %2173 = fpext bfloat %1581 to float, !dbg !28 + %2174 = fpext bfloat %1582 to float, !dbg !28 + %2175 = fpext bfloat %1583 to float, !dbg !28 + %2176 = fpext bfloat %1584 to float, !dbg !28 + %2177 = fpext bfloat %1585 to float, !dbg !28 + %2178 = fpext bfloat %1596 to float, !dbg !28 + %2179 = fpext bfloat %1597 to float, !dbg !28 + %2180 = fpext bfloat %1598 to float, !dbg !28 + %2181 = fpext bfloat %1599 to float, !dbg !28 + %2182 = fpext bfloat %1600 to float, !dbg !28 + %2183 = fpext bfloat %1601 to float, !dbg !28 + %2184 = fpext bfloat %1602 to float, !dbg !28 + %2185 = fpext bfloat %1603 to float, !dbg !28 + %2186 = fpext bfloat %1614 to float, !dbg !28 + %2187 = fpext bfloat %1615 to float, !dbg !28 + %2188 = fpext bfloat %1616 to float, !dbg !28 + %2189 = fpext bfloat %1617 to float, !dbg !28 + %2190 = fpext bfloat %1618 to float, !dbg !28 + %2191 = fpext bfloat %1619 to float, !dbg !28 + %2192 = fpext bfloat %1620 to float, !dbg !28 + %2193 = fpext bfloat %1621 to float, !dbg !28 + %2194 = fpext bfloat %1632 to float, !dbg !28 + %2195 = fpext bfloat %1633 to float, !dbg !28 + %2196 = fpext bfloat %1634 to float, !dbg !28 + %2197 = fpext bfloat %1635 to float, !dbg !28 + %2198 = fpext bfloat %1636 to float, !dbg !28 + %2199 = fpext bfloat %1637 to float, !dbg !28 + %2200 = fpext bfloat %1638 to float, !dbg !28 + %2201 = fpext bfloat %1639 to float, !dbg !28 + %2202 = fpext bfloat %1650 to float, !dbg !28 + %2203 = fpext bfloat %1651 to float, !dbg !28 + %2204 = fpext bfloat %1652 to float, !dbg !28 + %2205 = fpext bfloat %1653 to float, !dbg !28 + %2206 = fpext bfloat %1654 to float, !dbg !28 + %2207 = fpext bfloat %1655 to float, !dbg !28 + %2208 = fpext bfloat %1656 to float, !dbg !28 + %2209 = fpext bfloat %1657 to float, !dbg !28 + %2210 = fpext bfloat %1668 to float, !dbg !28 + %2211 = fpext bfloat %1669 to float, !dbg !28 + %2212 = fpext bfloat %1670 to float, !dbg !28 + %2213 = fpext bfloat %1671 to float, !dbg !28 + %2214 = fpext bfloat %1672 to float, !dbg !28 + %2215 = fpext bfloat %1673 to float, !dbg !28 + %2216 = fpext bfloat %1674 to float, !dbg !28 + %2217 = fpext bfloat %1675 to float, !dbg !28 + %2218 = fpext bfloat %1686 to float, !dbg !28 + %2219 = fpext bfloat %1687 to float, !dbg !28 + %2220 = fpext bfloat %1688 to float, !dbg !28 + %2221 = fpext bfloat %1689 to float, !dbg !28 + %2222 = fpext bfloat %1690 to float, !dbg !28 + %2223 = fpext bfloat %1691 to float, !dbg !28 + %2224 = fpext bfloat %1692 to float, !dbg !28 + %2225 = fpext bfloat %1693 to float, !dbg !28 + %2226 = fpext bfloat %1704 to float, !dbg !28 + %2227 = fpext bfloat %1705 to float, !dbg !28 + %2228 = fpext bfloat %1706 to float, !dbg !28 + %2229 = fpext bfloat %1707 to float, !dbg !28 + %2230 = fpext bfloat %1708 to float, !dbg !28 + %2231 = fpext bfloat %1709 to float, !dbg !28 + %2232 = fpext bfloat %1710 to float, !dbg !28 + %2233 = fpext bfloat %1711 to float, !dbg !28 + %2234 = fpext bfloat %1722 to float, !dbg !28 + %2235 = fpext bfloat %1723 to float, !dbg !28 + %2236 = fpext bfloat %1724 to float, !dbg !28 + %2237 = fpext bfloat %1725 to float, !dbg !28 + %2238 = fpext bfloat %1726 to float, !dbg !28 + %2239 = fpext bfloat %1727 to float, !dbg !28 + %2240 = fpext bfloat %1728 to float, !dbg !28 + %2241 = fpext bfloat %1729 to float, !dbg !28 + %2242 = extractelement <64 x i1> %62, i64 3, !dbg !29 + %2243 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %319, i1 %2242) #5, !dbg !29 + %2244 = extractelement <64 x i1> %62, i64 2, !dbg !29 + %2245 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %321, i1 %2244) #5, !dbg !29 + %2246 = extractelement <64 x i1> %62, i64 1, !dbg !29 + %2247 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %323, i1 %2246) #5, !dbg !29 + %2248 = extractelement <64 x i1> %62, i64 0, !dbg !29 + %2249 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %325, i1 %2248) #5, !dbg !29 + %2250 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %390, i1 %27) #5, !dbg !30 + %2251 = extractvalue { i32, i32, i32, i32 } %2250, 0, !dbg !30 + %2252 = bitcast i32 %2251 to <2 x bfloat>, !dbg !30 + %2253 = extractvalue { i32, i32, i32, i32 } %2250, 1, !dbg !30 + %2254 = bitcast i32 %2253 to <2 x bfloat>, !dbg !30 + %2255 = extractvalue { i32, i32, i32, i32 } %2250, 2, !dbg !30 + %2256 = bitcast i32 %2255 to <2 x bfloat>, !dbg !30 + %2257 = extractvalue { i32, i32, i32, i32 } %2250, 3, !dbg !30 + %2258 = bitcast i32 %2257 to <2 x bfloat>, !dbg !30 + %2259 = extractelement <2 x bfloat> %2252, i64 0, !dbg !30 + %2260 = extractelement <2 x bfloat> %2252, i64 1, !dbg !30 + %2261 = extractelement <2 x bfloat> %2254, i64 0, !dbg !30 + %2262 = extractelement <2 x bfloat> %2254, i64 1, !dbg !30 + %2263 = extractelement <2 x bfloat> %2256, i64 0, !dbg !30 + %2264 = extractelement <2 x bfloat> %2256, i64 1, !dbg !30 + %2265 = extractelement <2 x bfloat> %2258, i64 0, !dbg !30 + %2266 = extractelement <2 x bfloat> %2258, i64 1, !dbg !30 + %2267 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %391, i1 %28) #5, !dbg !30 + %2268 = extractvalue { i32, i32, i32, i32 } %2267, 0, !dbg !30 + %2269 = bitcast i32 %2268 to <2 x bfloat>, !dbg !30 + %2270 = extractvalue { i32, i32, i32, i32 } %2267, 1, !dbg !30 + %2271 = bitcast i32 %2270 to <2 x bfloat>, !dbg !30 + %2272 = extractvalue { i32, i32, i32, i32 } %2267, 2, !dbg !30 + %2273 = bitcast i32 %2272 to <2 x bfloat>, !dbg !30 + %2274 = extractvalue { i32, i32, i32, i32 } %2267, 3, !dbg !30 + %2275 = bitcast i32 %2274 to <2 x bfloat>, !dbg !30 + %2276 = extractelement <2 x bfloat> %2269, i64 0, !dbg !30 + %2277 = extractelement <2 x bfloat> %2269, i64 1, !dbg !30 + %2278 = extractelement <2 x bfloat> %2271, i64 0, !dbg !30 + %2279 = extractelement <2 x bfloat> %2271, i64 1, !dbg !30 + %2280 = extractelement <2 x bfloat> %2273, i64 0, !dbg !30 + %2281 = extractelement <2 x bfloat> %2273, i64 1, !dbg !30 + %2282 = extractelement <2 x bfloat> %2275, i64 0, !dbg !30 + %2283 = extractelement <2 x bfloat> %2275, i64 1, !dbg !30 + %2284 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %392, i1 %29) #5, !dbg !30 + %2285 = extractvalue { i32, i32, i32, i32 } %2284, 0, !dbg !30 + %2286 = bitcast i32 %2285 to <2 x bfloat>, !dbg !30 + %2287 = extractvalue { i32, i32, i32, i32 } %2284, 1, !dbg !30 + %2288 = bitcast i32 %2287 to <2 x bfloat>, !dbg !30 + %2289 = extractvalue { i32, i32, i32, i32 } %2284, 2, !dbg !30 + %2290 = bitcast i32 %2289 to <2 x bfloat>, !dbg !30 + %2291 = extractvalue { i32, i32, i32, i32 } %2284, 3, !dbg !30 + %2292 = bitcast i32 %2291 to <2 x bfloat>, !dbg !30 + %2293 = extractelement <2 x bfloat> %2286, i64 0, !dbg !30 + %2294 = extractelement <2 x bfloat> %2286, i64 1, !dbg !30 + %2295 = extractelement <2 x bfloat> %2288, i64 0, !dbg !30 + %2296 = extractelement <2 x bfloat> %2288, i64 1, !dbg !30 + %2297 = extractelement <2 x bfloat> %2290, i64 0, !dbg !30 + %2298 = extractelement <2 x bfloat> %2290, i64 1, !dbg !30 + %2299 = extractelement <2 x bfloat> %2292, i64 0, !dbg !30 + %2300 = extractelement <2 x bfloat> %2292, i64 1, !dbg !30 + %2301 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %393, i1 %30) #5, !dbg !30 + %2302 = extractvalue { i32, i32, i32, i32 } %2301, 0, !dbg !30 + %2303 = bitcast i32 %2302 to <2 x bfloat>, !dbg !30 + %2304 = extractvalue { i32, i32, i32, i32 } %2301, 1, !dbg !30 + %2305 = bitcast i32 %2304 to <2 x bfloat>, !dbg !30 + %2306 = extractvalue { i32, i32, i32, i32 } %2301, 2, !dbg !30 + %2307 = bitcast i32 %2306 to <2 x bfloat>, !dbg !30 + %2308 = extractvalue { i32, i32, i32, i32 } %2301, 3, !dbg !30 + %2309 = bitcast i32 %2308 to <2 x bfloat>, !dbg !30 + %2310 = extractelement <2 x bfloat> %2303, i64 0, !dbg !30 + %2311 = extractelement <2 x bfloat> %2303, i64 1, !dbg !30 + %2312 = extractelement <2 x bfloat> %2305, i64 0, !dbg !30 + %2313 = extractelement <2 x bfloat> %2305, i64 1, !dbg !30 + %2314 = extractelement <2 x bfloat> %2307, i64 0, !dbg !30 + %2315 = extractelement <2 x bfloat> %2307, i64 1, !dbg !30 + %2316 = extractelement <2 x bfloat> %2309, i64 0, !dbg !30 + %2317 = extractelement <2 x bfloat> %2309, i64 1, !dbg !30 + %2318 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %394, i1 %650) #5, !dbg !30 + %2319 = extractvalue { i32, i32, i32, i32 } %2318, 0, !dbg !30 + %2320 = bitcast i32 %2319 to <2 x bfloat>, !dbg !30 + %2321 = extractvalue { i32, i32, i32, i32 } %2318, 1, !dbg !30 + %2322 = bitcast i32 %2321 to <2 x bfloat>, !dbg !30 + %2323 = extractvalue { i32, i32, i32, i32 } %2318, 2, !dbg !30 + %2324 = bitcast i32 %2323 to <2 x bfloat>, !dbg !30 + %2325 = extractvalue { i32, i32, i32, i32 } %2318, 3, !dbg !30 + %2326 = bitcast i32 %2325 to <2 x bfloat>, !dbg !30 + %2327 = extractelement <2 x bfloat> %2320, i64 0, !dbg !30 + %2328 = extractelement <2 x bfloat> %2320, i64 1, !dbg !30 + %2329 = extractelement <2 x bfloat> %2322, i64 0, !dbg !30 + %2330 = extractelement <2 x bfloat> %2322, i64 1, !dbg !30 + %2331 = extractelement <2 x bfloat> %2324, i64 0, !dbg !30 + %2332 = extractelement <2 x bfloat> %2324, i64 1, !dbg !30 + %2333 = extractelement <2 x bfloat> %2326, i64 0, !dbg !30 + %2334 = extractelement <2 x bfloat> %2326, i64 1, !dbg !30 + %2335 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %395, i1 %668) #5, !dbg !30 + %2336 = extractvalue { i32, i32, i32, i32 } %2335, 0, !dbg !30 + %2337 = bitcast i32 %2336 to <2 x bfloat>, !dbg !30 + %2338 = extractvalue { i32, i32, i32, i32 } %2335, 1, !dbg !30 + %2339 = bitcast i32 %2338 to <2 x bfloat>, !dbg !30 + %2340 = extractvalue { i32, i32, i32, i32 } %2335, 2, !dbg !30 + %2341 = bitcast i32 %2340 to <2 x bfloat>, !dbg !30 + %2342 = extractvalue { i32, i32, i32, i32 } %2335, 3, !dbg !30 + %2343 = bitcast i32 %2342 to <2 x bfloat>, !dbg !30 + %2344 = extractelement <2 x bfloat> %2337, i64 0, !dbg !30 + %2345 = extractelement <2 x bfloat> %2337, i64 1, !dbg !30 + %2346 = extractelement <2 x bfloat> %2339, i64 0, !dbg !30 + %2347 = extractelement <2 x bfloat> %2339, i64 1, !dbg !30 + %2348 = extractelement <2 x bfloat> %2341, i64 0, !dbg !30 + %2349 = extractelement <2 x bfloat> %2341, i64 1, !dbg !30 + %2350 = extractelement <2 x bfloat> %2343, i64 0, !dbg !30 + %2351 = extractelement <2 x bfloat> %2343, i64 1, !dbg !30 + %2352 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %396, i1 %686) #5, !dbg !30 + %2353 = extractvalue { i32, i32, i32, i32 } %2352, 0, !dbg !30 + %2354 = bitcast i32 %2353 to <2 x bfloat>, !dbg !30 + %2355 = extractvalue { i32, i32, i32, i32 } %2352, 1, !dbg !30 + %2356 = bitcast i32 %2355 to <2 x bfloat>, !dbg !30 + %2357 = extractvalue { i32, i32, i32, i32 } %2352, 2, !dbg !30 + %2358 = bitcast i32 %2357 to <2 x bfloat>, !dbg !30 + %2359 = extractvalue { i32, i32, i32, i32 } %2352, 3, !dbg !30 + %2360 = bitcast i32 %2359 to <2 x bfloat>, !dbg !30 + %2361 = extractelement <2 x bfloat> %2354, i64 0, !dbg !30 + %2362 = extractelement <2 x bfloat> %2354, i64 1, !dbg !30 + %2363 = extractelement <2 x bfloat> %2356, i64 0, !dbg !30 + %2364 = extractelement <2 x bfloat> %2356, i64 1, !dbg !30 + %2365 = extractelement <2 x bfloat> %2358, i64 0, !dbg !30 + %2366 = extractelement <2 x bfloat> %2358, i64 1, !dbg !30 + %2367 = extractelement <2 x bfloat> %2360, i64 0, !dbg !30 + %2368 = extractelement <2 x bfloat> %2360, i64 1, !dbg !30 + %2369 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %397, i1 %704) #5, !dbg !30 + %2370 = extractvalue { i32, i32, i32, i32 } %2369, 0, !dbg !30 + %2371 = bitcast i32 %2370 to <2 x bfloat>, !dbg !30 + %2372 = extractvalue { i32, i32, i32, i32 } %2369, 1, !dbg !30 + %2373 = bitcast i32 %2372 to <2 x bfloat>, !dbg !30 + %2374 = extractvalue { i32, i32, i32, i32 } %2369, 2, !dbg !30 + %2375 = bitcast i32 %2374 to <2 x bfloat>, !dbg !30 + %2376 = extractvalue { i32, i32, i32, i32 } %2369, 3, !dbg !30 + %2377 = bitcast i32 %2376 to <2 x bfloat>, !dbg !30 + %2378 = extractelement <2 x bfloat> %2371, i64 0, !dbg !30 + %2379 = extractelement <2 x bfloat> %2371, i64 1, !dbg !30 + %2380 = extractelement <2 x bfloat> %2373, i64 0, !dbg !30 + %2381 = extractelement <2 x bfloat> %2373, i64 1, !dbg !30 + %2382 = extractelement <2 x bfloat> %2375, i64 0, !dbg !30 + %2383 = extractelement <2 x bfloat> %2375, i64 1, !dbg !30 + %2384 = extractelement <2 x bfloat> %2377, i64 0, !dbg !30 + %2385 = extractelement <2 x bfloat> %2377, i64 1, !dbg !30 + %2386 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %398, i1 %722) #5, !dbg !30 + %2387 = extractvalue { i32, i32, i32, i32 } %2386, 0, !dbg !30 + %2388 = bitcast i32 %2387 to <2 x bfloat>, !dbg !30 + %2389 = extractvalue { i32, i32, i32, i32 } %2386, 1, !dbg !30 + %2390 = bitcast i32 %2389 to <2 x bfloat>, !dbg !30 + %2391 = extractvalue { i32, i32, i32, i32 } %2386, 2, !dbg !30 + %2392 = bitcast i32 %2391 to <2 x bfloat>, !dbg !30 + %2393 = extractvalue { i32, i32, i32, i32 } %2386, 3, !dbg !30 + %2394 = bitcast i32 %2393 to <2 x bfloat>, !dbg !30 + %2395 = extractelement <2 x bfloat> %2388, i64 0, !dbg !30 + %2396 = extractelement <2 x bfloat> %2388, i64 1, !dbg !30 + %2397 = extractelement <2 x bfloat> %2390, i64 0, !dbg !30 + %2398 = extractelement <2 x bfloat> %2390, i64 1, !dbg !30 + %2399 = extractelement <2 x bfloat> %2392, i64 0, !dbg !30 + %2400 = extractelement <2 x bfloat> %2392, i64 1, !dbg !30 + %2401 = extractelement <2 x bfloat> %2394, i64 0, !dbg !30 + %2402 = extractelement <2 x bfloat> %2394, i64 1, !dbg !30 + %2403 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %399, i1 %740) #5, !dbg !30 + %2404 = extractvalue { i32, i32, i32, i32 } %2403, 0, !dbg !30 + %2405 = bitcast i32 %2404 to <2 x bfloat>, !dbg !30 + %2406 = extractvalue { i32, i32, i32, i32 } %2403, 1, !dbg !30 + %2407 = bitcast i32 %2406 to <2 x bfloat>, !dbg !30 + %2408 = extractvalue { i32, i32, i32, i32 } %2403, 2, !dbg !30 + %2409 = bitcast i32 %2408 to <2 x bfloat>, !dbg !30 + %2410 = extractvalue { i32, i32, i32, i32 } %2403, 3, !dbg !30 + %2411 = bitcast i32 %2410 to <2 x bfloat>, !dbg !30 + %2412 = extractelement <2 x bfloat> %2405, i64 0, !dbg !30 + %2413 = extractelement <2 x bfloat> %2405, i64 1, !dbg !30 + %2414 = extractelement <2 x bfloat> %2407, i64 0, !dbg !30 + %2415 = extractelement <2 x bfloat> %2407, i64 1, !dbg !30 + %2416 = extractelement <2 x bfloat> %2409, i64 0, !dbg !30 + %2417 = extractelement <2 x bfloat> %2409, i64 1, !dbg !30 + %2418 = extractelement <2 x bfloat> %2411, i64 0, !dbg !30 + %2419 = extractelement <2 x bfloat> %2411, i64 1, !dbg !30 + %2420 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %400, i1 %758) #5, !dbg !30 + %2421 = extractvalue { i32, i32, i32, i32 } %2420, 0, !dbg !30 + %2422 = bitcast i32 %2421 to <2 x bfloat>, !dbg !30 + %2423 = extractvalue { i32, i32, i32, i32 } %2420, 1, !dbg !30 + %2424 = bitcast i32 %2423 to <2 x bfloat>, !dbg !30 + %2425 = extractvalue { i32, i32, i32, i32 } %2420, 2, !dbg !30 + %2426 = bitcast i32 %2425 to <2 x bfloat>, !dbg !30 + %2427 = extractvalue { i32, i32, i32, i32 } %2420, 3, !dbg !30 + %2428 = bitcast i32 %2427 to <2 x bfloat>, !dbg !30 + %2429 = extractelement <2 x bfloat> %2422, i64 0, !dbg !30 + %2430 = extractelement <2 x bfloat> %2422, i64 1, !dbg !30 + %2431 = extractelement <2 x bfloat> %2424, i64 0, !dbg !30 + %2432 = extractelement <2 x bfloat> %2424, i64 1, !dbg !30 + %2433 = extractelement <2 x bfloat> %2426, i64 0, !dbg !30 + %2434 = extractelement <2 x bfloat> %2426, i64 1, !dbg !30 + %2435 = extractelement <2 x bfloat> %2428, i64 0, !dbg !30 + %2436 = extractelement <2 x bfloat> %2428, i64 1, !dbg !30 + %2437 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %401, i1 %776) #5, !dbg !30 + %2438 = extractvalue { i32, i32, i32, i32 } %2437, 0, !dbg !30 + %2439 = bitcast i32 %2438 to <2 x bfloat>, !dbg !30 + %2440 = extractvalue { i32, i32, i32, i32 } %2437, 1, !dbg !30 + %2441 = bitcast i32 %2440 to <2 x bfloat>, !dbg !30 + %2442 = extractvalue { i32, i32, i32, i32 } %2437, 2, !dbg !30 + %2443 = bitcast i32 %2442 to <2 x bfloat>, !dbg !30 + %2444 = extractvalue { i32, i32, i32, i32 } %2437, 3, !dbg !30 + %2445 = bitcast i32 %2444 to <2 x bfloat>, !dbg !30 + %2446 = extractelement <2 x bfloat> %2439, i64 0, !dbg !30 + %2447 = extractelement <2 x bfloat> %2439, i64 1, !dbg !30 + %2448 = extractelement <2 x bfloat> %2441, i64 0, !dbg !30 + %2449 = extractelement <2 x bfloat> %2441, i64 1, !dbg !30 + %2450 = extractelement <2 x bfloat> %2443, i64 0, !dbg !30 + %2451 = extractelement <2 x bfloat> %2443, i64 1, !dbg !30 + %2452 = extractelement <2 x bfloat> %2445, i64 0, !dbg !30 + %2453 = extractelement <2 x bfloat> %2445, i64 1, !dbg !30 + %2454 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %402, i1 %794) #5, !dbg !30 + %2455 = extractvalue { i32, i32, i32, i32 } %2454, 0, !dbg !30 + %2456 = bitcast i32 %2455 to <2 x bfloat>, !dbg !30 + %2457 = extractvalue { i32, i32, i32, i32 } %2454, 1, !dbg !30 + %2458 = bitcast i32 %2457 to <2 x bfloat>, !dbg !30 + %2459 = extractvalue { i32, i32, i32, i32 } %2454, 2, !dbg !30 + %2460 = bitcast i32 %2459 to <2 x bfloat>, !dbg !30 + %2461 = extractvalue { i32, i32, i32, i32 } %2454, 3, !dbg !30 + %2462 = bitcast i32 %2461 to <2 x bfloat>, !dbg !30 + %2463 = extractelement <2 x bfloat> %2456, i64 0, !dbg !30 + %2464 = extractelement <2 x bfloat> %2456, i64 1, !dbg !30 + %2465 = extractelement <2 x bfloat> %2458, i64 0, !dbg !30 + %2466 = extractelement <2 x bfloat> %2458, i64 1, !dbg !30 + %2467 = extractelement <2 x bfloat> %2460, i64 0, !dbg !30 + %2468 = extractelement <2 x bfloat> %2460, i64 1, !dbg !30 + %2469 = extractelement <2 x bfloat> %2462, i64 0, !dbg !30 + %2470 = extractelement <2 x bfloat> %2462, i64 1, !dbg !30 + %2471 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %403, i1 %812) #5, !dbg !30 + %2472 = extractvalue { i32, i32, i32, i32 } %2471, 0, !dbg !30 + %2473 = bitcast i32 %2472 to <2 x bfloat>, !dbg !30 + %2474 = extractvalue { i32, i32, i32, i32 } %2471, 1, !dbg !30 + %2475 = bitcast i32 %2474 to <2 x bfloat>, !dbg !30 + %2476 = extractvalue { i32, i32, i32, i32 } %2471, 2, !dbg !30 + %2477 = bitcast i32 %2476 to <2 x bfloat>, !dbg !30 + %2478 = extractvalue { i32, i32, i32, i32 } %2471, 3, !dbg !30 + %2479 = bitcast i32 %2478 to <2 x bfloat>, !dbg !30 + %2480 = extractelement <2 x bfloat> %2473, i64 0, !dbg !30 + %2481 = extractelement <2 x bfloat> %2473, i64 1, !dbg !30 + %2482 = extractelement <2 x bfloat> %2475, i64 0, !dbg !30 + %2483 = extractelement <2 x bfloat> %2475, i64 1, !dbg !30 + %2484 = extractelement <2 x bfloat> %2477, i64 0, !dbg !30 + %2485 = extractelement <2 x bfloat> %2477, i64 1, !dbg !30 + %2486 = extractelement <2 x bfloat> %2479, i64 0, !dbg !30 + %2487 = extractelement <2 x bfloat> %2479, i64 1, !dbg !30 + %2488 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %404, i1 %830) #5, !dbg !30 + %2489 = extractvalue { i32, i32, i32, i32 } %2488, 0, !dbg !30 + %2490 = bitcast i32 %2489 to <2 x bfloat>, !dbg !30 + %2491 = extractvalue { i32, i32, i32, i32 } %2488, 1, !dbg !30 + %2492 = bitcast i32 %2491 to <2 x bfloat>, !dbg !30 + %2493 = extractvalue { i32, i32, i32, i32 } %2488, 2, !dbg !30 + %2494 = bitcast i32 %2493 to <2 x bfloat>, !dbg !30 + %2495 = extractvalue { i32, i32, i32, i32 } %2488, 3, !dbg !30 + %2496 = bitcast i32 %2495 to <2 x bfloat>, !dbg !30 + %2497 = extractelement <2 x bfloat> %2490, i64 0, !dbg !30 + %2498 = extractelement <2 x bfloat> %2490, i64 1, !dbg !30 + %2499 = extractelement <2 x bfloat> %2492, i64 0, !dbg !30 + %2500 = extractelement <2 x bfloat> %2492, i64 1, !dbg !30 + %2501 = extractelement <2 x bfloat> %2494, i64 0, !dbg !30 + %2502 = extractelement <2 x bfloat> %2494, i64 1, !dbg !30 + %2503 = extractelement <2 x bfloat> %2496, i64 0, !dbg !30 + %2504 = extractelement <2 x bfloat> %2496, i64 1, !dbg !30 + %2505 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %405, i1 %848) #5, !dbg !30 + %2506 = extractvalue { i32, i32, i32, i32 } %2505, 0, !dbg !30 + %2507 = bitcast i32 %2506 to <2 x bfloat>, !dbg !30 + %2508 = extractvalue { i32, i32, i32, i32 } %2505, 1, !dbg !30 + %2509 = bitcast i32 %2508 to <2 x bfloat>, !dbg !30 + %2510 = extractvalue { i32, i32, i32, i32 } %2505, 2, !dbg !30 + %2511 = bitcast i32 %2510 to <2 x bfloat>, !dbg !30 + %2512 = extractvalue { i32, i32, i32, i32 } %2505, 3, !dbg !30 + %2513 = bitcast i32 %2512 to <2 x bfloat>, !dbg !30 + %2514 = extractelement <2 x bfloat> %2507, i64 0, !dbg !30 + %2515 = extractelement <2 x bfloat> %2507, i64 1, !dbg !30 + %2516 = extractelement <2 x bfloat> %2509, i64 0, !dbg !30 + %2517 = extractelement <2 x bfloat> %2509, i64 1, !dbg !30 + %2518 = extractelement <2 x bfloat> %2511, i64 0, !dbg !30 + %2519 = extractelement <2 x bfloat> %2511, i64 1, !dbg !30 + %2520 = extractelement <2 x bfloat> %2513, i64 0, !dbg !30 + %2521 = extractelement <2 x bfloat> %2513, i64 1, !dbg !30 + %2522 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %406, i1 %866) #5, !dbg !30 + %2523 = extractvalue { i32, i32, i32, i32 } %2522, 0, !dbg !30 + %2524 = bitcast i32 %2523 to <2 x bfloat>, !dbg !30 + %2525 = extractvalue { i32, i32, i32, i32 } %2522, 1, !dbg !30 + %2526 = bitcast i32 %2525 to <2 x bfloat>, !dbg !30 + %2527 = extractvalue { i32, i32, i32, i32 } %2522, 2, !dbg !30 + %2528 = bitcast i32 %2527 to <2 x bfloat>, !dbg !30 + %2529 = extractvalue { i32, i32, i32, i32 } %2522, 3, !dbg !30 + %2530 = bitcast i32 %2529 to <2 x bfloat>, !dbg !30 + %2531 = extractelement <2 x bfloat> %2524, i64 0, !dbg !30 + %2532 = extractelement <2 x bfloat> %2524, i64 1, !dbg !30 + %2533 = extractelement <2 x bfloat> %2526, i64 0, !dbg !30 + %2534 = extractelement <2 x bfloat> %2526, i64 1, !dbg !30 + %2535 = extractelement <2 x bfloat> %2528, i64 0, !dbg !30 + %2536 = extractelement <2 x bfloat> %2528, i64 1, !dbg !30 + %2537 = extractelement <2 x bfloat> %2530, i64 0, !dbg !30 + %2538 = extractelement <2 x bfloat> %2530, i64 1, !dbg !30 + %2539 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %407, i1 %884) #5, !dbg !30 + %2540 = extractvalue { i32, i32, i32, i32 } %2539, 0, !dbg !30 + %2541 = bitcast i32 %2540 to <2 x bfloat>, !dbg !30 + %2542 = extractvalue { i32, i32, i32, i32 } %2539, 1, !dbg !30 + %2543 = bitcast i32 %2542 to <2 x bfloat>, !dbg !30 + %2544 = extractvalue { i32, i32, i32, i32 } %2539, 2, !dbg !30 + %2545 = bitcast i32 %2544 to <2 x bfloat>, !dbg !30 + %2546 = extractvalue { i32, i32, i32, i32 } %2539, 3, !dbg !30 + %2547 = bitcast i32 %2546 to <2 x bfloat>, !dbg !30 + %2548 = extractelement <2 x bfloat> %2541, i64 0, !dbg !30 + %2549 = extractelement <2 x bfloat> %2541, i64 1, !dbg !30 + %2550 = extractelement <2 x bfloat> %2543, i64 0, !dbg !30 + %2551 = extractelement <2 x bfloat> %2543, i64 1, !dbg !30 + %2552 = extractelement <2 x bfloat> %2545, i64 0, !dbg !30 + %2553 = extractelement <2 x bfloat> %2545, i64 1, !dbg !30 + %2554 = extractelement <2 x bfloat> %2547, i64 0, !dbg !30 + %2555 = extractelement <2 x bfloat> %2547, i64 1, !dbg !30 + %2556 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %408, i1 %902) #5, !dbg !30 + %2557 = extractvalue { i32, i32, i32, i32 } %2556, 0, !dbg !30 + %2558 = bitcast i32 %2557 to <2 x bfloat>, !dbg !30 + %2559 = extractvalue { i32, i32, i32, i32 } %2556, 1, !dbg !30 + %2560 = bitcast i32 %2559 to <2 x bfloat>, !dbg !30 + %2561 = extractvalue { i32, i32, i32, i32 } %2556, 2, !dbg !30 + %2562 = bitcast i32 %2561 to <2 x bfloat>, !dbg !30 + %2563 = extractvalue { i32, i32, i32, i32 } %2556, 3, !dbg !30 + %2564 = bitcast i32 %2563 to <2 x bfloat>, !dbg !30 + %2565 = extractelement <2 x bfloat> %2558, i64 0, !dbg !30 + %2566 = extractelement <2 x bfloat> %2558, i64 1, !dbg !30 + %2567 = extractelement <2 x bfloat> %2560, i64 0, !dbg !30 + %2568 = extractelement <2 x bfloat> %2560, i64 1, !dbg !30 + %2569 = extractelement <2 x bfloat> %2562, i64 0, !dbg !30 + %2570 = extractelement <2 x bfloat> %2562, i64 1, !dbg !30 + %2571 = extractelement <2 x bfloat> %2564, i64 0, !dbg !30 + %2572 = extractelement <2 x bfloat> %2564, i64 1, !dbg !30 + %2573 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %409, i1 %920) #5, !dbg !30 + %2574 = extractvalue { i32, i32, i32, i32 } %2573, 0, !dbg !30 + %2575 = bitcast i32 %2574 to <2 x bfloat>, !dbg !30 + %2576 = extractvalue { i32, i32, i32, i32 } %2573, 1, !dbg !30 + %2577 = bitcast i32 %2576 to <2 x bfloat>, !dbg !30 + %2578 = extractvalue { i32, i32, i32, i32 } %2573, 2, !dbg !30 + %2579 = bitcast i32 %2578 to <2 x bfloat>, !dbg !30 + %2580 = extractvalue { i32, i32, i32, i32 } %2573, 3, !dbg !30 + %2581 = bitcast i32 %2580 to <2 x bfloat>, !dbg !30 + %2582 = extractelement <2 x bfloat> %2575, i64 0, !dbg !30 + %2583 = extractelement <2 x bfloat> %2575, i64 1, !dbg !30 + %2584 = extractelement <2 x bfloat> %2577, i64 0, !dbg !30 + %2585 = extractelement <2 x bfloat> %2577, i64 1, !dbg !30 + %2586 = extractelement <2 x bfloat> %2579, i64 0, !dbg !30 + %2587 = extractelement <2 x bfloat> %2579, i64 1, !dbg !30 + %2588 = extractelement <2 x bfloat> %2581, i64 0, !dbg !30 + %2589 = extractelement <2 x bfloat> %2581, i64 1, !dbg !30 + %2590 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %410, i1 %938) #5, !dbg !30 + %2591 = extractvalue { i32, i32, i32, i32 } %2590, 0, !dbg !30 + %2592 = bitcast i32 %2591 to <2 x bfloat>, !dbg !30 + %2593 = extractvalue { i32, i32, i32, i32 } %2590, 1, !dbg !30 + %2594 = bitcast i32 %2593 to <2 x bfloat>, !dbg !30 + %2595 = extractvalue { i32, i32, i32, i32 } %2590, 2, !dbg !30 + %2596 = bitcast i32 %2595 to <2 x bfloat>, !dbg !30 + %2597 = extractvalue { i32, i32, i32, i32 } %2590, 3, !dbg !30 + %2598 = bitcast i32 %2597 to <2 x bfloat>, !dbg !30 + %2599 = extractelement <2 x bfloat> %2592, i64 0, !dbg !30 + %2600 = extractelement <2 x bfloat> %2592, i64 1, !dbg !30 + %2601 = extractelement <2 x bfloat> %2594, i64 0, !dbg !30 + %2602 = extractelement <2 x bfloat> %2594, i64 1, !dbg !30 + %2603 = extractelement <2 x bfloat> %2596, i64 0, !dbg !30 + %2604 = extractelement <2 x bfloat> %2596, i64 1, !dbg !30 + %2605 = extractelement <2 x bfloat> %2598, i64 0, !dbg !30 + %2606 = extractelement <2 x bfloat> %2598, i64 1, !dbg !30 + %2607 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %411, i1 %956) #5, !dbg !30 + %2608 = extractvalue { i32, i32, i32, i32 } %2607, 0, !dbg !30 + %2609 = bitcast i32 %2608 to <2 x bfloat>, !dbg !30 + %2610 = extractvalue { i32, i32, i32, i32 } %2607, 1, !dbg !30 + %2611 = bitcast i32 %2610 to <2 x bfloat>, !dbg !30 + %2612 = extractvalue { i32, i32, i32, i32 } %2607, 2, !dbg !30 + %2613 = bitcast i32 %2612 to <2 x bfloat>, !dbg !30 + %2614 = extractvalue { i32, i32, i32, i32 } %2607, 3, !dbg !30 + %2615 = bitcast i32 %2614 to <2 x bfloat>, !dbg !30 + %2616 = extractelement <2 x bfloat> %2609, i64 0, !dbg !30 + %2617 = extractelement <2 x bfloat> %2609, i64 1, !dbg !30 + %2618 = extractelement <2 x bfloat> %2611, i64 0, !dbg !30 + %2619 = extractelement <2 x bfloat> %2611, i64 1, !dbg !30 + %2620 = extractelement <2 x bfloat> %2613, i64 0, !dbg !30 + %2621 = extractelement <2 x bfloat> %2613, i64 1, !dbg !30 + %2622 = extractelement <2 x bfloat> %2615, i64 0, !dbg !30 + %2623 = extractelement <2 x bfloat> %2615, i64 1, !dbg !30 + %2624 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %412, i1 %974) #5, !dbg !30 + %2625 = extractvalue { i32, i32, i32, i32 } %2624, 0, !dbg !30 + %2626 = bitcast i32 %2625 to <2 x bfloat>, !dbg !30 + %2627 = extractvalue { i32, i32, i32, i32 } %2624, 1, !dbg !30 + %2628 = bitcast i32 %2627 to <2 x bfloat>, !dbg !30 + %2629 = extractvalue { i32, i32, i32, i32 } %2624, 2, !dbg !30 + %2630 = bitcast i32 %2629 to <2 x bfloat>, !dbg !30 + %2631 = extractvalue { i32, i32, i32, i32 } %2624, 3, !dbg !30 + %2632 = bitcast i32 %2631 to <2 x bfloat>, !dbg !30 + %2633 = extractelement <2 x bfloat> %2626, i64 0, !dbg !30 + %2634 = extractelement <2 x bfloat> %2626, i64 1, !dbg !30 + %2635 = extractelement <2 x bfloat> %2628, i64 0, !dbg !30 + %2636 = extractelement <2 x bfloat> %2628, i64 1, !dbg !30 + %2637 = extractelement <2 x bfloat> %2630, i64 0, !dbg !30 + %2638 = extractelement <2 x bfloat> %2630, i64 1, !dbg !30 + %2639 = extractelement <2 x bfloat> %2632, i64 0, !dbg !30 + %2640 = extractelement <2 x bfloat> %2632, i64 1, !dbg !30 + %2641 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %413, i1 %992) #5, !dbg !30 + %2642 = extractvalue { i32, i32, i32, i32 } %2641, 0, !dbg !30 + %2643 = bitcast i32 %2642 to <2 x bfloat>, !dbg !30 + %2644 = extractvalue { i32, i32, i32, i32 } %2641, 1, !dbg !30 + %2645 = bitcast i32 %2644 to <2 x bfloat>, !dbg !30 + %2646 = extractvalue { i32, i32, i32, i32 } %2641, 2, !dbg !30 + %2647 = bitcast i32 %2646 to <2 x bfloat>, !dbg !30 + %2648 = extractvalue { i32, i32, i32, i32 } %2641, 3, !dbg !30 + %2649 = bitcast i32 %2648 to <2 x bfloat>, !dbg !30 + %2650 = extractelement <2 x bfloat> %2643, i64 0, !dbg !30 + %2651 = extractelement <2 x bfloat> %2643, i64 1, !dbg !30 + %2652 = extractelement <2 x bfloat> %2645, i64 0, !dbg !30 + %2653 = extractelement <2 x bfloat> %2645, i64 1, !dbg !30 + %2654 = extractelement <2 x bfloat> %2647, i64 0, !dbg !30 + %2655 = extractelement <2 x bfloat> %2647, i64 1, !dbg !30 + %2656 = extractelement <2 x bfloat> %2649, i64 0, !dbg !30 + %2657 = extractelement <2 x bfloat> %2649, i64 1, !dbg !30 + %2658 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %414, i1 %1010) #5, !dbg !30 + %2659 = extractvalue { i32, i32, i32, i32 } %2658, 0, !dbg !30 + %2660 = bitcast i32 %2659 to <2 x bfloat>, !dbg !30 + %2661 = extractvalue { i32, i32, i32, i32 } %2658, 1, !dbg !30 + %2662 = bitcast i32 %2661 to <2 x bfloat>, !dbg !30 + %2663 = extractvalue { i32, i32, i32, i32 } %2658, 2, !dbg !30 + %2664 = bitcast i32 %2663 to <2 x bfloat>, !dbg !30 + %2665 = extractvalue { i32, i32, i32, i32 } %2658, 3, !dbg !30 + %2666 = bitcast i32 %2665 to <2 x bfloat>, !dbg !30 + %2667 = extractelement <2 x bfloat> %2660, i64 0, !dbg !30 + %2668 = extractelement <2 x bfloat> %2660, i64 1, !dbg !30 + %2669 = extractelement <2 x bfloat> %2662, i64 0, !dbg !30 + %2670 = extractelement <2 x bfloat> %2662, i64 1, !dbg !30 + %2671 = extractelement <2 x bfloat> %2664, i64 0, !dbg !30 + %2672 = extractelement <2 x bfloat> %2664, i64 1, !dbg !30 + %2673 = extractelement <2 x bfloat> %2666, i64 0, !dbg !30 + %2674 = extractelement <2 x bfloat> %2666, i64 1, !dbg !30 + %2675 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %415, i1 %1028) #5, !dbg !30 + %2676 = extractvalue { i32, i32, i32, i32 } %2675, 0, !dbg !30 + %2677 = bitcast i32 %2676 to <2 x bfloat>, !dbg !30 + %2678 = extractvalue { i32, i32, i32, i32 } %2675, 1, !dbg !30 + %2679 = bitcast i32 %2678 to <2 x bfloat>, !dbg !30 + %2680 = extractvalue { i32, i32, i32, i32 } %2675, 2, !dbg !30 + %2681 = bitcast i32 %2680 to <2 x bfloat>, !dbg !30 + %2682 = extractvalue { i32, i32, i32, i32 } %2675, 3, !dbg !30 + %2683 = bitcast i32 %2682 to <2 x bfloat>, !dbg !30 + %2684 = extractelement <2 x bfloat> %2677, i64 0, !dbg !30 + %2685 = extractelement <2 x bfloat> %2677, i64 1, !dbg !30 + %2686 = extractelement <2 x bfloat> %2679, i64 0, !dbg !30 + %2687 = extractelement <2 x bfloat> %2679, i64 1, !dbg !30 + %2688 = extractelement <2 x bfloat> %2681, i64 0, !dbg !30 + %2689 = extractelement <2 x bfloat> %2681, i64 1, !dbg !30 + %2690 = extractelement <2 x bfloat> %2683, i64 0, !dbg !30 + %2691 = extractelement <2 x bfloat> %2683, i64 1, !dbg !30 + %2692 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %416, i1 %1046) #5, !dbg !30 + %2693 = extractvalue { i32, i32, i32, i32 } %2692, 0, !dbg !30 + %2694 = bitcast i32 %2693 to <2 x bfloat>, !dbg !30 + %2695 = extractvalue { i32, i32, i32, i32 } %2692, 1, !dbg !30 + %2696 = bitcast i32 %2695 to <2 x bfloat>, !dbg !30 + %2697 = extractvalue { i32, i32, i32, i32 } %2692, 2, !dbg !30 + %2698 = bitcast i32 %2697 to <2 x bfloat>, !dbg !30 + %2699 = extractvalue { i32, i32, i32, i32 } %2692, 3, !dbg !30 + %2700 = bitcast i32 %2699 to <2 x bfloat>, !dbg !30 + %2701 = extractelement <2 x bfloat> %2694, i64 0, !dbg !30 + %2702 = extractelement <2 x bfloat> %2694, i64 1, !dbg !30 + %2703 = extractelement <2 x bfloat> %2696, i64 0, !dbg !30 + %2704 = extractelement <2 x bfloat> %2696, i64 1, !dbg !30 + %2705 = extractelement <2 x bfloat> %2698, i64 0, !dbg !30 + %2706 = extractelement <2 x bfloat> %2698, i64 1, !dbg !30 + %2707 = extractelement <2 x bfloat> %2700, i64 0, !dbg !30 + %2708 = extractelement <2 x bfloat> %2700, i64 1, !dbg !30 + %2709 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %417, i1 %1064) #5, !dbg !30 + %2710 = extractvalue { i32, i32, i32, i32 } %2709, 0, !dbg !30 + %2711 = bitcast i32 %2710 to <2 x bfloat>, !dbg !30 + %2712 = extractvalue { i32, i32, i32, i32 } %2709, 1, !dbg !30 + %2713 = bitcast i32 %2712 to <2 x bfloat>, !dbg !30 + %2714 = extractvalue { i32, i32, i32, i32 } %2709, 2, !dbg !30 + %2715 = bitcast i32 %2714 to <2 x bfloat>, !dbg !30 + %2716 = extractvalue { i32, i32, i32, i32 } %2709, 3, !dbg !30 + %2717 = bitcast i32 %2716 to <2 x bfloat>, !dbg !30 + %2718 = extractelement <2 x bfloat> %2711, i64 0, !dbg !30 + %2719 = extractelement <2 x bfloat> %2711, i64 1, !dbg !30 + %2720 = extractelement <2 x bfloat> %2713, i64 0, !dbg !30 + %2721 = extractelement <2 x bfloat> %2713, i64 1, !dbg !30 + %2722 = extractelement <2 x bfloat> %2715, i64 0, !dbg !30 + %2723 = extractelement <2 x bfloat> %2715, i64 1, !dbg !30 + %2724 = extractelement <2 x bfloat> %2717, i64 0, !dbg !30 + %2725 = extractelement <2 x bfloat> %2717, i64 1, !dbg !30 + %2726 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %418, i1 %1082) #5, !dbg !30 + %2727 = extractvalue { i32, i32, i32, i32 } %2726, 0, !dbg !30 + %2728 = bitcast i32 %2727 to <2 x bfloat>, !dbg !30 + %2729 = extractvalue { i32, i32, i32, i32 } %2726, 1, !dbg !30 + %2730 = bitcast i32 %2729 to <2 x bfloat>, !dbg !30 + %2731 = extractvalue { i32, i32, i32, i32 } %2726, 2, !dbg !30 + %2732 = bitcast i32 %2731 to <2 x bfloat>, !dbg !30 + %2733 = extractvalue { i32, i32, i32, i32 } %2726, 3, !dbg !30 + %2734 = bitcast i32 %2733 to <2 x bfloat>, !dbg !30 + %2735 = extractelement <2 x bfloat> %2728, i64 0, !dbg !30 + %2736 = extractelement <2 x bfloat> %2728, i64 1, !dbg !30 + %2737 = extractelement <2 x bfloat> %2730, i64 0, !dbg !30 + %2738 = extractelement <2 x bfloat> %2730, i64 1, !dbg !30 + %2739 = extractelement <2 x bfloat> %2732, i64 0, !dbg !30 + %2740 = extractelement <2 x bfloat> %2732, i64 1, !dbg !30 + %2741 = extractelement <2 x bfloat> %2734, i64 0, !dbg !30 + %2742 = extractelement <2 x bfloat> %2734, i64 1, !dbg !30 + %2743 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %419, i1 %1100) #5, !dbg !30 + %2744 = extractvalue { i32, i32, i32, i32 } %2743, 0, !dbg !30 + %2745 = bitcast i32 %2744 to <2 x bfloat>, !dbg !30 + %2746 = extractvalue { i32, i32, i32, i32 } %2743, 1, !dbg !30 + %2747 = bitcast i32 %2746 to <2 x bfloat>, !dbg !30 + %2748 = extractvalue { i32, i32, i32, i32 } %2743, 2, !dbg !30 + %2749 = bitcast i32 %2748 to <2 x bfloat>, !dbg !30 + %2750 = extractvalue { i32, i32, i32, i32 } %2743, 3, !dbg !30 + %2751 = bitcast i32 %2750 to <2 x bfloat>, !dbg !30 + %2752 = extractelement <2 x bfloat> %2745, i64 0, !dbg !30 + %2753 = extractelement <2 x bfloat> %2745, i64 1, !dbg !30 + %2754 = extractelement <2 x bfloat> %2747, i64 0, !dbg !30 + %2755 = extractelement <2 x bfloat> %2747, i64 1, !dbg !30 + %2756 = extractelement <2 x bfloat> %2749, i64 0, !dbg !30 + %2757 = extractelement <2 x bfloat> %2749, i64 1, !dbg !30 + %2758 = extractelement <2 x bfloat> %2751, i64 0, !dbg !30 + %2759 = extractelement <2 x bfloat> %2751, i64 1, !dbg !30 + %2760 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %420, i1 %1118) #5, !dbg !30 + %2761 = extractvalue { i32, i32, i32, i32 } %2760, 0, !dbg !30 + %2762 = bitcast i32 %2761 to <2 x bfloat>, !dbg !30 + %2763 = extractvalue { i32, i32, i32, i32 } %2760, 1, !dbg !30 + %2764 = bitcast i32 %2763 to <2 x bfloat>, !dbg !30 + %2765 = extractvalue { i32, i32, i32, i32 } %2760, 2, !dbg !30 + %2766 = bitcast i32 %2765 to <2 x bfloat>, !dbg !30 + %2767 = extractvalue { i32, i32, i32, i32 } %2760, 3, !dbg !30 + %2768 = bitcast i32 %2767 to <2 x bfloat>, !dbg !30 + %2769 = extractelement <2 x bfloat> %2762, i64 0, !dbg !30 + %2770 = extractelement <2 x bfloat> %2762, i64 1, !dbg !30 + %2771 = extractelement <2 x bfloat> %2764, i64 0, !dbg !30 + %2772 = extractelement <2 x bfloat> %2764, i64 1, !dbg !30 + %2773 = extractelement <2 x bfloat> %2766, i64 0, !dbg !30 + %2774 = extractelement <2 x bfloat> %2766, i64 1, !dbg !30 + %2775 = extractelement <2 x bfloat> %2768, i64 0, !dbg !30 + %2776 = extractelement <2 x bfloat> %2768, i64 1, !dbg !30 + %2777 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %421, i1 %1136) #5, !dbg !30 + %2778 = extractvalue { i32, i32, i32, i32 } %2777, 0, !dbg !30 + %2779 = bitcast i32 %2778 to <2 x bfloat>, !dbg !30 + %2780 = extractvalue { i32, i32, i32, i32 } %2777, 1, !dbg !30 + %2781 = bitcast i32 %2780 to <2 x bfloat>, !dbg !30 + %2782 = extractvalue { i32, i32, i32, i32 } %2777, 2, !dbg !30 + %2783 = bitcast i32 %2782 to <2 x bfloat>, !dbg !30 + %2784 = extractvalue { i32, i32, i32, i32 } %2777, 3, !dbg !30 + %2785 = bitcast i32 %2784 to <2 x bfloat>, !dbg !30 + %2786 = extractelement <2 x bfloat> %2779, i64 0, !dbg !30 + %2787 = extractelement <2 x bfloat> %2779, i64 1, !dbg !30 + %2788 = extractelement <2 x bfloat> %2781, i64 0, !dbg !30 + %2789 = extractelement <2 x bfloat> %2781, i64 1, !dbg !30 + %2790 = extractelement <2 x bfloat> %2783, i64 0, !dbg !30 + %2791 = extractelement <2 x bfloat> %2783, i64 1, !dbg !30 + %2792 = extractelement <2 x bfloat> %2785, i64 0, !dbg !30 + %2793 = extractelement <2 x bfloat> %2785, i64 1, !dbg !30 + %2794 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %422, i1 %1154) #5, !dbg !30 + %2795 = extractvalue { i32, i32, i32, i32 } %2794, 0, !dbg !30 + %2796 = bitcast i32 %2795 to <2 x bfloat>, !dbg !30 + %2797 = extractvalue { i32, i32, i32, i32 } %2794, 1, !dbg !30 + %2798 = bitcast i32 %2797 to <2 x bfloat>, !dbg !30 + %2799 = extractvalue { i32, i32, i32, i32 } %2794, 2, !dbg !30 + %2800 = bitcast i32 %2799 to <2 x bfloat>, !dbg !30 + %2801 = extractvalue { i32, i32, i32, i32 } %2794, 3, !dbg !30 + %2802 = bitcast i32 %2801 to <2 x bfloat>, !dbg !30 + %2803 = extractelement <2 x bfloat> %2796, i64 0, !dbg !30 + %2804 = extractelement <2 x bfloat> %2796, i64 1, !dbg !30 + %2805 = extractelement <2 x bfloat> %2798, i64 0, !dbg !30 + %2806 = extractelement <2 x bfloat> %2798, i64 1, !dbg !30 + %2807 = extractelement <2 x bfloat> %2800, i64 0, !dbg !30 + %2808 = extractelement <2 x bfloat> %2800, i64 1, !dbg !30 + %2809 = extractelement <2 x bfloat> %2802, i64 0, !dbg !30 + %2810 = extractelement <2 x bfloat> %2802, i64 1, !dbg !30 + %2811 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %423, i1 %1172) #5, !dbg !30 + %2812 = extractvalue { i32, i32, i32, i32 } %2811, 0, !dbg !30 + %2813 = bitcast i32 %2812 to <2 x bfloat>, !dbg !30 + %2814 = extractvalue { i32, i32, i32, i32 } %2811, 1, !dbg !30 + %2815 = bitcast i32 %2814 to <2 x bfloat>, !dbg !30 + %2816 = extractvalue { i32, i32, i32, i32 } %2811, 2, !dbg !30 + %2817 = bitcast i32 %2816 to <2 x bfloat>, !dbg !30 + %2818 = extractvalue { i32, i32, i32, i32 } %2811, 3, !dbg !30 + %2819 = bitcast i32 %2818 to <2 x bfloat>, !dbg !30 + %2820 = extractelement <2 x bfloat> %2813, i64 0, !dbg !30 + %2821 = extractelement <2 x bfloat> %2813, i64 1, !dbg !30 + %2822 = extractelement <2 x bfloat> %2815, i64 0, !dbg !30 + %2823 = extractelement <2 x bfloat> %2815, i64 1, !dbg !30 + %2824 = extractelement <2 x bfloat> %2817, i64 0, !dbg !30 + %2825 = extractelement <2 x bfloat> %2817, i64 1, !dbg !30 + %2826 = extractelement <2 x bfloat> %2819, i64 0, !dbg !30 + %2827 = extractelement <2 x bfloat> %2819, i64 1, !dbg !30 + %2828 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %424, i1 %1190) #5, !dbg !30 + %2829 = extractvalue { i32, i32, i32, i32 } %2828, 0, !dbg !30 + %2830 = bitcast i32 %2829 to <2 x bfloat>, !dbg !30 + %2831 = extractvalue { i32, i32, i32, i32 } %2828, 1, !dbg !30 + %2832 = bitcast i32 %2831 to <2 x bfloat>, !dbg !30 + %2833 = extractvalue { i32, i32, i32, i32 } %2828, 2, !dbg !30 + %2834 = bitcast i32 %2833 to <2 x bfloat>, !dbg !30 + %2835 = extractvalue { i32, i32, i32, i32 } %2828, 3, !dbg !30 + %2836 = bitcast i32 %2835 to <2 x bfloat>, !dbg !30 + %2837 = extractelement <2 x bfloat> %2830, i64 0, !dbg !30 + %2838 = extractelement <2 x bfloat> %2830, i64 1, !dbg !30 + %2839 = extractelement <2 x bfloat> %2832, i64 0, !dbg !30 + %2840 = extractelement <2 x bfloat> %2832, i64 1, !dbg !30 + %2841 = extractelement <2 x bfloat> %2834, i64 0, !dbg !30 + %2842 = extractelement <2 x bfloat> %2834, i64 1, !dbg !30 + %2843 = extractelement <2 x bfloat> %2836, i64 0, !dbg !30 + %2844 = extractelement <2 x bfloat> %2836, i64 1, !dbg !30 + %2845 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %425, i1 %1208) #5, !dbg !30 + %2846 = extractvalue { i32, i32, i32, i32 } %2845, 0, !dbg !30 + %2847 = bitcast i32 %2846 to <2 x bfloat>, !dbg !30 + %2848 = extractvalue { i32, i32, i32, i32 } %2845, 1, !dbg !30 + %2849 = bitcast i32 %2848 to <2 x bfloat>, !dbg !30 + %2850 = extractvalue { i32, i32, i32, i32 } %2845, 2, !dbg !30 + %2851 = bitcast i32 %2850 to <2 x bfloat>, !dbg !30 + %2852 = extractvalue { i32, i32, i32, i32 } %2845, 3, !dbg !30 + %2853 = bitcast i32 %2852 to <2 x bfloat>, !dbg !30 + %2854 = extractelement <2 x bfloat> %2847, i64 0, !dbg !30 + %2855 = extractelement <2 x bfloat> %2847, i64 1, !dbg !30 + %2856 = extractelement <2 x bfloat> %2849, i64 0, !dbg !30 + %2857 = extractelement <2 x bfloat> %2849, i64 1, !dbg !30 + %2858 = extractelement <2 x bfloat> %2851, i64 0, !dbg !30 + %2859 = extractelement <2 x bfloat> %2851, i64 1, !dbg !30 + %2860 = extractelement <2 x bfloat> %2853, i64 0, !dbg !30 + %2861 = extractelement <2 x bfloat> %2853, i64 1, !dbg !30 + %2862 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %426, i1 %1226) #5, !dbg !30 + %2863 = extractvalue { i32, i32, i32, i32 } %2862, 0, !dbg !30 + %2864 = bitcast i32 %2863 to <2 x bfloat>, !dbg !30 + %2865 = extractvalue { i32, i32, i32, i32 } %2862, 1, !dbg !30 + %2866 = bitcast i32 %2865 to <2 x bfloat>, !dbg !30 + %2867 = extractvalue { i32, i32, i32, i32 } %2862, 2, !dbg !30 + %2868 = bitcast i32 %2867 to <2 x bfloat>, !dbg !30 + %2869 = extractvalue { i32, i32, i32, i32 } %2862, 3, !dbg !30 + %2870 = bitcast i32 %2869 to <2 x bfloat>, !dbg !30 + %2871 = extractelement <2 x bfloat> %2864, i64 0, !dbg !30 + %2872 = extractelement <2 x bfloat> %2864, i64 1, !dbg !30 + %2873 = extractelement <2 x bfloat> %2866, i64 0, !dbg !30 + %2874 = extractelement <2 x bfloat> %2866, i64 1, !dbg !30 + %2875 = extractelement <2 x bfloat> %2868, i64 0, !dbg !30 + %2876 = extractelement <2 x bfloat> %2868, i64 1, !dbg !30 + %2877 = extractelement <2 x bfloat> %2870, i64 0, !dbg !30 + %2878 = extractelement <2 x bfloat> %2870, i64 1, !dbg !30 + %2879 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %427, i1 %1244) #5, !dbg !30 + %2880 = extractvalue { i32, i32, i32, i32 } %2879, 0, !dbg !30 + %2881 = bitcast i32 %2880 to <2 x bfloat>, !dbg !30 + %2882 = extractvalue { i32, i32, i32, i32 } %2879, 1, !dbg !30 + %2883 = bitcast i32 %2882 to <2 x bfloat>, !dbg !30 + %2884 = extractvalue { i32, i32, i32, i32 } %2879, 2, !dbg !30 + %2885 = bitcast i32 %2884 to <2 x bfloat>, !dbg !30 + %2886 = extractvalue { i32, i32, i32, i32 } %2879, 3, !dbg !30 + %2887 = bitcast i32 %2886 to <2 x bfloat>, !dbg !30 + %2888 = extractelement <2 x bfloat> %2881, i64 0, !dbg !30 + %2889 = extractelement <2 x bfloat> %2881, i64 1, !dbg !30 + %2890 = extractelement <2 x bfloat> %2883, i64 0, !dbg !30 + %2891 = extractelement <2 x bfloat> %2883, i64 1, !dbg !30 + %2892 = extractelement <2 x bfloat> %2885, i64 0, !dbg !30 + %2893 = extractelement <2 x bfloat> %2885, i64 1, !dbg !30 + %2894 = extractelement <2 x bfloat> %2887, i64 0, !dbg !30 + %2895 = extractelement <2 x bfloat> %2887, i64 1, !dbg !30 + %2896 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %428, i1 %1262) #5, !dbg !30 + %2897 = extractvalue { i32, i32, i32, i32 } %2896, 0, !dbg !30 + %2898 = bitcast i32 %2897 to <2 x bfloat>, !dbg !30 + %2899 = extractvalue { i32, i32, i32, i32 } %2896, 1, !dbg !30 + %2900 = bitcast i32 %2899 to <2 x bfloat>, !dbg !30 + %2901 = extractvalue { i32, i32, i32, i32 } %2896, 2, !dbg !30 + %2902 = bitcast i32 %2901 to <2 x bfloat>, !dbg !30 + %2903 = extractvalue { i32, i32, i32, i32 } %2896, 3, !dbg !30 + %2904 = bitcast i32 %2903 to <2 x bfloat>, !dbg !30 + %2905 = extractelement <2 x bfloat> %2898, i64 0, !dbg !30 + %2906 = extractelement <2 x bfloat> %2898, i64 1, !dbg !30 + %2907 = extractelement <2 x bfloat> %2900, i64 0, !dbg !30 + %2908 = extractelement <2 x bfloat> %2900, i64 1, !dbg !30 + %2909 = extractelement <2 x bfloat> %2902, i64 0, !dbg !30 + %2910 = extractelement <2 x bfloat> %2902, i64 1, !dbg !30 + %2911 = extractelement <2 x bfloat> %2904, i64 0, !dbg !30 + %2912 = extractelement <2 x bfloat> %2904, i64 1, !dbg !30 + %2913 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %429, i1 %1280) #5, !dbg !30 + %2914 = extractvalue { i32, i32, i32, i32 } %2913, 0, !dbg !30 + %2915 = bitcast i32 %2914 to <2 x bfloat>, !dbg !30 + %2916 = extractvalue { i32, i32, i32, i32 } %2913, 1, !dbg !30 + %2917 = bitcast i32 %2916 to <2 x bfloat>, !dbg !30 + %2918 = extractvalue { i32, i32, i32, i32 } %2913, 2, !dbg !30 + %2919 = bitcast i32 %2918 to <2 x bfloat>, !dbg !30 + %2920 = extractvalue { i32, i32, i32, i32 } %2913, 3, !dbg !30 + %2921 = bitcast i32 %2920 to <2 x bfloat>, !dbg !30 + %2922 = extractelement <2 x bfloat> %2915, i64 0, !dbg !30 + %2923 = extractelement <2 x bfloat> %2915, i64 1, !dbg !30 + %2924 = extractelement <2 x bfloat> %2917, i64 0, !dbg !30 + %2925 = extractelement <2 x bfloat> %2917, i64 1, !dbg !30 + %2926 = extractelement <2 x bfloat> %2919, i64 0, !dbg !30 + %2927 = extractelement <2 x bfloat> %2919, i64 1, !dbg !30 + %2928 = extractelement <2 x bfloat> %2921, i64 0, !dbg !30 + %2929 = extractelement <2 x bfloat> %2921, i64 1, !dbg !30 + %2930 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %430, i1 %1298) #5, !dbg !30 + %2931 = extractvalue { i32, i32, i32, i32 } %2930, 0, !dbg !30 + %2932 = bitcast i32 %2931 to <2 x bfloat>, !dbg !30 + %2933 = extractvalue { i32, i32, i32, i32 } %2930, 1, !dbg !30 + %2934 = bitcast i32 %2933 to <2 x bfloat>, !dbg !30 + %2935 = extractvalue { i32, i32, i32, i32 } %2930, 2, !dbg !30 + %2936 = bitcast i32 %2935 to <2 x bfloat>, !dbg !30 + %2937 = extractvalue { i32, i32, i32, i32 } %2930, 3, !dbg !30 + %2938 = bitcast i32 %2937 to <2 x bfloat>, !dbg !30 + %2939 = extractelement <2 x bfloat> %2932, i64 0, !dbg !30 + %2940 = extractelement <2 x bfloat> %2932, i64 1, !dbg !30 + %2941 = extractelement <2 x bfloat> %2934, i64 0, !dbg !30 + %2942 = extractelement <2 x bfloat> %2934, i64 1, !dbg !30 + %2943 = extractelement <2 x bfloat> %2936, i64 0, !dbg !30 + %2944 = extractelement <2 x bfloat> %2936, i64 1, !dbg !30 + %2945 = extractelement <2 x bfloat> %2938, i64 0, !dbg !30 + %2946 = extractelement <2 x bfloat> %2938, i64 1, !dbg !30 + %2947 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %431, i1 %1316) #5, !dbg !30 + %2948 = extractvalue { i32, i32, i32, i32 } %2947, 0, !dbg !30 + %2949 = bitcast i32 %2948 to <2 x bfloat>, !dbg !30 + %2950 = extractvalue { i32, i32, i32, i32 } %2947, 1, !dbg !30 + %2951 = bitcast i32 %2950 to <2 x bfloat>, !dbg !30 + %2952 = extractvalue { i32, i32, i32, i32 } %2947, 2, !dbg !30 + %2953 = bitcast i32 %2952 to <2 x bfloat>, !dbg !30 + %2954 = extractvalue { i32, i32, i32, i32 } %2947, 3, !dbg !30 + %2955 = bitcast i32 %2954 to <2 x bfloat>, !dbg !30 + %2956 = extractelement <2 x bfloat> %2949, i64 0, !dbg !30 + %2957 = extractelement <2 x bfloat> %2949, i64 1, !dbg !30 + %2958 = extractelement <2 x bfloat> %2951, i64 0, !dbg !30 + %2959 = extractelement <2 x bfloat> %2951, i64 1, !dbg !30 + %2960 = extractelement <2 x bfloat> %2953, i64 0, !dbg !30 + %2961 = extractelement <2 x bfloat> %2953, i64 1, !dbg !30 + %2962 = extractelement <2 x bfloat> %2955, i64 0, !dbg !30 + %2963 = extractelement <2 x bfloat> %2955, i64 1, !dbg !30 + %2964 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %432, i1 %1334) #5, !dbg !30 + %2965 = extractvalue { i32, i32, i32, i32 } %2964, 0, !dbg !30 + %2966 = bitcast i32 %2965 to <2 x bfloat>, !dbg !30 + %2967 = extractvalue { i32, i32, i32, i32 } %2964, 1, !dbg !30 + %2968 = bitcast i32 %2967 to <2 x bfloat>, !dbg !30 + %2969 = extractvalue { i32, i32, i32, i32 } %2964, 2, !dbg !30 + %2970 = bitcast i32 %2969 to <2 x bfloat>, !dbg !30 + %2971 = extractvalue { i32, i32, i32, i32 } %2964, 3, !dbg !30 + %2972 = bitcast i32 %2971 to <2 x bfloat>, !dbg !30 + %2973 = extractelement <2 x bfloat> %2966, i64 0, !dbg !30 + %2974 = extractelement <2 x bfloat> %2966, i64 1, !dbg !30 + %2975 = extractelement <2 x bfloat> %2968, i64 0, !dbg !30 + %2976 = extractelement <2 x bfloat> %2968, i64 1, !dbg !30 + %2977 = extractelement <2 x bfloat> %2970, i64 0, !dbg !30 + %2978 = extractelement <2 x bfloat> %2970, i64 1, !dbg !30 + %2979 = extractelement <2 x bfloat> %2972, i64 0, !dbg !30 + %2980 = extractelement <2 x bfloat> %2972, i64 1, !dbg !30 + %2981 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %433, i1 %1352) #5, !dbg !30 + %2982 = extractvalue { i32, i32, i32, i32 } %2981, 0, !dbg !30 + %2983 = bitcast i32 %2982 to <2 x bfloat>, !dbg !30 + %2984 = extractvalue { i32, i32, i32, i32 } %2981, 1, !dbg !30 + %2985 = bitcast i32 %2984 to <2 x bfloat>, !dbg !30 + %2986 = extractvalue { i32, i32, i32, i32 } %2981, 2, !dbg !30 + %2987 = bitcast i32 %2986 to <2 x bfloat>, !dbg !30 + %2988 = extractvalue { i32, i32, i32, i32 } %2981, 3, !dbg !30 + %2989 = bitcast i32 %2988 to <2 x bfloat>, !dbg !30 + %2990 = extractelement <2 x bfloat> %2983, i64 0, !dbg !30 + %2991 = extractelement <2 x bfloat> %2983, i64 1, !dbg !30 + %2992 = extractelement <2 x bfloat> %2985, i64 0, !dbg !30 + %2993 = extractelement <2 x bfloat> %2985, i64 1, !dbg !30 + %2994 = extractelement <2 x bfloat> %2987, i64 0, !dbg !30 + %2995 = extractelement <2 x bfloat> %2987, i64 1, !dbg !30 + %2996 = extractelement <2 x bfloat> %2989, i64 0, !dbg !30 + %2997 = extractelement <2 x bfloat> %2989, i64 1, !dbg !30 + %2998 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %434, i1 %1370) #5, !dbg !30 + %2999 = extractvalue { i32, i32, i32, i32 } %2998, 0, !dbg !30 + %3000 = bitcast i32 %2999 to <2 x bfloat>, !dbg !30 + %3001 = extractvalue { i32, i32, i32, i32 } %2998, 1, !dbg !30 + %3002 = bitcast i32 %3001 to <2 x bfloat>, !dbg !30 + %3003 = extractvalue { i32, i32, i32, i32 } %2998, 2, !dbg !30 + %3004 = bitcast i32 %3003 to <2 x bfloat>, !dbg !30 + %3005 = extractvalue { i32, i32, i32, i32 } %2998, 3, !dbg !30 + %3006 = bitcast i32 %3005 to <2 x bfloat>, !dbg !30 + %3007 = extractelement <2 x bfloat> %3000, i64 0, !dbg !30 + %3008 = extractelement <2 x bfloat> %3000, i64 1, !dbg !30 + %3009 = extractelement <2 x bfloat> %3002, i64 0, !dbg !30 + %3010 = extractelement <2 x bfloat> %3002, i64 1, !dbg !30 + %3011 = extractelement <2 x bfloat> %3004, i64 0, !dbg !30 + %3012 = extractelement <2 x bfloat> %3004, i64 1, !dbg !30 + %3013 = extractelement <2 x bfloat> %3006, i64 0, !dbg !30 + %3014 = extractelement <2 x bfloat> %3006, i64 1, !dbg !30 + %3015 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %435, i1 %1388) #5, !dbg !30 + %3016 = extractvalue { i32, i32, i32, i32 } %3015, 0, !dbg !30 + %3017 = bitcast i32 %3016 to <2 x bfloat>, !dbg !30 + %3018 = extractvalue { i32, i32, i32, i32 } %3015, 1, !dbg !30 + %3019 = bitcast i32 %3018 to <2 x bfloat>, !dbg !30 + %3020 = extractvalue { i32, i32, i32, i32 } %3015, 2, !dbg !30 + %3021 = bitcast i32 %3020 to <2 x bfloat>, !dbg !30 + %3022 = extractvalue { i32, i32, i32, i32 } %3015, 3, !dbg !30 + %3023 = bitcast i32 %3022 to <2 x bfloat>, !dbg !30 + %3024 = extractelement <2 x bfloat> %3017, i64 0, !dbg !30 + %3025 = extractelement <2 x bfloat> %3017, i64 1, !dbg !30 + %3026 = extractelement <2 x bfloat> %3019, i64 0, !dbg !30 + %3027 = extractelement <2 x bfloat> %3019, i64 1, !dbg !30 + %3028 = extractelement <2 x bfloat> %3021, i64 0, !dbg !30 + %3029 = extractelement <2 x bfloat> %3021, i64 1, !dbg !30 + %3030 = extractelement <2 x bfloat> %3023, i64 0, !dbg !30 + %3031 = extractelement <2 x bfloat> %3023, i64 1, !dbg !30 + %3032 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %436, i1 %1406) #5, !dbg !30 + %3033 = extractvalue { i32, i32, i32, i32 } %3032, 0, !dbg !30 + %3034 = bitcast i32 %3033 to <2 x bfloat>, !dbg !30 + %3035 = extractvalue { i32, i32, i32, i32 } %3032, 1, !dbg !30 + %3036 = bitcast i32 %3035 to <2 x bfloat>, !dbg !30 + %3037 = extractvalue { i32, i32, i32, i32 } %3032, 2, !dbg !30 + %3038 = bitcast i32 %3037 to <2 x bfloat>, !dbg !30 + %3039 = extractvalue { i32, i32, i32, i32 } %3032, 3, !dbg !30 + %3040 = bitcast i32 %3039 to <2 x bfloat>, !dbg !30 + %3041 = extractelement <2 x bfloat> %3034, i64 0, !dbg !30 + %3042 = extractelement <2 x bfloat> %3034, i64 1, !dbg !30 + %3043 = extractelement <2 x bfloat> %3036, i64 0, !dbg !30 + %3044 = extractelement <2 x bfloat> %3036, i64 1, !dbg !30 + %3045 = extractelement <2 x bfloat> %3038, i64 0, !dbg !30 + %3046 = extractelement <2 x bfloat> %3038, i64 1, !dbg !30 + %3047 = extractelement <2 x bfloat> %3040, i64 0, !dbg !30 + %3048 = extractelement <2 x bfloat> %3040, i64 1, !dbg !30 + %3049 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %437, i1 %1424) #5, !dbg !30 + %3050 = extractvalue { i32, i32, i32, i32 } %3049, 0, !dbg !30 + %3051 = bitcast i32 %3050 to <2 x bfloat>, !dbg !30 + %3052 = extractvalue { i32, i32, i32, i32 } %3049, 1, !dbg !30 + %3053 = bitcast i32 %3052 to <2 x bfloat>, !dbg !30 + %3054 = extractvalue { i32, i32, i32, i32 } %3049, 2, !dbg !30 + %3055 = bitcast i32 %3054 to <2 x bfloat>, !dbg !30 + %3056 = extractvalue { i32, i32, i32, i32 } %3049, 3, !dbg !30 + %3057 = bitcast i32 %3056 to <2 x bfloat>, !dbg !30 + %3058 = extractelement <2 x bfloat> %3051, i64 0, !dbg !30 + %3059 = extractelement <2 x bfloat> %3051, i64 1, !dbg !30 + %3060 = extractelement <2 x bfloat> %3053, i64 0, !dbg !30 + %3061 = extractelement <2 x bfloat> %3053, i64 1, !dbg !30 + %3062 = extractelement <2 x bfloat> %3055, i64 0, !dbg !30 + %3063 = extractelement <2 x bfloat> %3055, i64 1, !dbg !30 + %3064 = extractelement <2 x bfloat> %3057, i64 0, !dbg !30 + %3065 = extractelement <2 x bfloat> %3057, i64 1, !dbg !30 + %3066 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %438, i1 %1442) #5, !dbg !30 + %3067 = extractvalue { i32, i32, i32, i32 } %3066, 0, !dbg !30 + %3068 = bitcast i32 %3067 to <2 x bfloat>, !dbg !30 + %3069 = extractvalue { i32, i32, i32, i32 } %3066, 1, !dbg !30 + %3070 = bitcast i32 %3069 to <2 x bfloat>, !dbg !30 + %3071 = extractvalue { i32, i32, i32, i32 } %3066, 2, !dbg !30 + %3072 = bitcast i32 %3071 to <2 x bfloat>, !dbg !30 + %3073 = extractvalue { i32, i32, i32, i32 } %3066, 3, !dbg !30 + %3074 = bitcast i32 %3073 to <2 x bfloat>, !dbg !30 + %3075 = extractelement <2 x bfloat> %3068, i64 0, !dbg !30 + %3076 = extractelement <2 x bfloat> %3068, i64 1, !dbg !30 + %3077 = extractelement <2 x bfloat> %3070, i64 0, !dbg !30 + %3078 = extractelement <2 x bfloat> %3070, i64 1, !dbg !30 + %3079 = extractelement <2 x bfloat> %3072, i64 0, !dbg !30 + %3080 = extractelement <2 x bfloat> %3072, i64 1, !dbg !30 + %3081 = extractelement <2 x bfloat> %3074, i64 0, !dbg !30 + %3082 = extractelement <2 x bfloat> %3074, i64 1, !dbg !30 + %3083 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %439, i1 %1460) #5, !dbg !30 + %3084 = extractvalue { i32, i32, i32, i32 } %3083, 0, !dbg !30 + %3085 = bitcast i32 %3084 to <2 x bfloat>, !dbg !30 + %3086 = extractvalue { i32, i32, i32, i32 } %3083, 1, !dbg !30 + %3087 = bitcast i32 %3086 to <2 x bfloat>, !dbg !30 + %3088 = extractvalue { i32, i32, i32, i32 } %3083, 2, !dbg !30 + %3089 = bitcast i32 %3088 to <2 x bfloat>, !dbg !30 + %3090 = extractvalue { i32, i32, i32, i32 } %3083, 3, !dbg !30 + %3091 = bitcast i32 %3090 to <2 x bfloat>, !dbg !30 + %3092 = extractelement <2 x bfloat> %3085, i64 0, !dbg !30 + %3093 = extractelement <2 x bfloat> %3085, i64 1, !dbg !30 + %3094 = extractelement <2 x bfloat> %3087, i64 0, !dbg !30 + %3095 = extractelement <2 x bfloat> %3087, i64 1, !dbg !30 + %3096 = extractelement <2 x bfloat> %3089, i64 0, !dbg !30 + %3097 = extractelement <2 x bfloat> %3089, i64 1, !dbg !30 + %3098 = extractelement <2 x bfloat> %3091, i64 0, !dbg !30 + %3099 = extractelement <2 x bfloat> %3091, i64 1, !dbg !30 + %3100 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %440, i1 %1478) #5, !dbg !30 + %3101 = extractvalue { i32, i32, i32, i32 } %3100, 0, !dbg !30 + %3102 = bitcast i32 %3101 to <2 x bfloat>, !dbg !30 + %3103 = extractvalue { i32, i32, i32, i32 } %3100, 1, !dbg !30 + %3104 = bitcast i32 %3103 to <2 x bfloat>, !dbg !30 + %3105 = extractvalue { i32, i32, i32, i32 } %3100, 2, !dbg !30 + %3106 = bitcast i32 %3105 to <2 x bfloat>, !dbg !30 + %3107 = extractvalue { i32, i32, i32, i32 } %3100, 3, !dbg !30 + %3108 = bitcast i32 %3107 to <2 x bfloat>, !dbg !30 + %3109 = extractelement <2 x bfloat> %3102, i64 0, !dbg !30 + %3110 = extractelement <2 x bfloat> %3102, i64 1, !dbg !30 + %3111 = extractelement <2 x bfloat> %3104, i64 0, !dbg !30 + %3112 = extractelement <2 x bfloat> %3104, i64 1, !dbg !30 + %3113 = extractelement <2 x bfloat> %3106, i64 0, !dbg !30 + %3114 = extractelement <2 x bfloat> %3106, i64 1, !dbg !30 + %3115 = extractelement <2 x bfloat> %3108, i64 0, !dbg !30 + %3116 = extractelement <2 x bfloat> %3108, i64 1, !dbg !30 + %3117 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %441, i1 %1496) #5, !dbg !30 + %3118 = extractvalue { i32, i32, i32, i32 } %3117, 0, !dbg !30 + %3119 = bitcast i32 %3118 to <2 x bfloat>, !dbg !30 + %3120 = extractvalue { i32, i32, i32, i32 } %3117, 1, !dbg !30 + %3121 = bitcast i32 %3120 to <2 x bfloat>, !dbg !30 + %3122 = extractvalue { i32, i32, i32, i32 } %3117, 2, !dbg !30 + %3123 = bitcast i32 %3122 to <2 x bfloat>, !dbg !30 + %3124 = extractvalue { i32, i32, i32, i32 } %3117, 3, !dbg !30 + %3125 = bitcast i32 %3124 to <2 x bfloat>, !dbg !30 + %3126 = extractelement <2 x bfloat> %3119, i64 0, !dbg !30 + %3127 = extractelement <2 x bfloat> %3119, i64 1, !dbg !30 + %3128 = extractelement <2 x bfloat> %3121, i64 0, !dbg !30 + %3129 = extractelement <2 x bfloat> %3121, i64 1, !dbg !30 + %3130 = extractelement <2 x bfloat> %3123, i64 0, !dbg !30 + %3131 = extractelement <2 x bfloat> %3123, i64 1, !dbg !30 + %3132 = extractelement <2 x bfloat> %3125, i64 0, !dbg !30 + %3133 = extractelement <2 x bfloat> %3125, i64 1, !dbg !30 + %3134 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %442, i1 %1514) #5, !dbg !30 + %3135 = extractvalue { i32, i32, i32, i32 } %3134, 0, !dbg !30 + %3136 = bitcast i32 %3135 to <2 x bfloat>, !dbg !30 + %3137 = extractvalue { i32, i32, i32, i32 } %3134, 1, !dbg !30 + %3138 = bitcast i32 %3137 to <2 x bfloat>, !dbg !30 + %3139 = extractvalue { i32, i32, i32, i32 } %3134, 2, !dbg !30 + %3140 = bitcast i32 %3139 to <2 x bfloat>, !dbg !30 + %3141 = extractvalue { i32, i32, i32, i32 } %3134, 3, !dbg !30 + %3142 = bitcast i32 %3141 to <2 x bfloat>, !dbg !30 + %3143 = extractelement <2 x bfloat> %3136, i64 0, !dbg !30 + %3144 = extractelement <2 x bfloat> %3136, i64 1, !dbg !30 + %3145 = extractelement <2 x bfloat> %3138, i64 0, !dbg !30 + %3146 = extractelement <2 x bfloat> %3138, i64 1, !dbg !30 + %3147 = extractelement <2 x bfloat> %3140, i64 0, !dbg !30 + %3148 = extractelement <2 x bfloat> %3140, i64 1, !dbg !30 + %3149 = extractelement <2 x bfloat> %3142, i64 0, !dbg !30 + %3150 = extractelement <2 x bfloat> %3142, i64 1, !dbg !30 + %3151 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %443, i1 %1532) #5, !dbg !30 + %3152 = extractvalue { i32, i32, i32, i32 } %3151, 0, !dbg !30 + %3153 = bitcast i32 %3152 to <2 x bfloat>, !dbg !30 + %3154 = extractvalue { i32, i32, i32, i32 } %3151, 1, !dbg !30 + %3155 = bitcast i32 %3154 to <2 x bfloat>, !dbg !30 + %3156 = extractvalue { i32, i32, i32, i32 } %3151, 2, !dbg !30 + %3157 = bitcast i32 %3156 to <2 x bfloat>, !dbg !30 + %3158 = extractvalue { i32, i32, i32, i32 } %3151, 3, !dbg !30 + %3159 = bitcast i32 %3158 to <2 x bfloat>, !dbg !30 + %3160 = extractelement <2 x bfloat> %3153, i64 0, !dbg !30 + %3161 = extractelement <2 x bfloat> %3153, i64 1, !dbg !30 + %3162 = extractelement <2 x bfloat> %3155, i64 0, !dbg !30 + %3163 = extractelement <2 x bfloat> %3155, i64 1, !dbg !30 + %3164 = extractelement <2 x bfloat> %3157, i64 0, !dbg !30 + %3165 = extractelement <2 x bfloat> %3157, i64 1, !dbg !30 + %3166 = extractelement <2 x bfloat> %3159, i64 0, !dbg !30 + %3167 = extractelement <2 x bfloat> %3159, i64 1, !dbg !30 + %3168 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %444, i1 %1550) #5, !dbg !30 + %3169 = extractvalue { i32, i32, i32, i32 } %3168, 0, !dbg !30 + %3170 = bitcast i32 %3169 to <2 x bfloat>, !dbg !30 + %3171 = extractvalue { i32, i32, i32, i32 } %3168, 1, !dbg !30 + %3172 = bitcast i32 %3171 to <2 x bfloat>, !dbg !30 + %3173 = extractvalue { i32, i32, i32, i32 } %3168, 2, !dbg !30 + %3174 = bitcast i32 %3173 to <2 x bfloat>, !dbg !30 + %3175 = extractvalue { i32, i32, i32, i32 } %3168, 3, !dbg !30 + %3176 = bitcast i32 %3175 to <2 x bfloat>, !dbg !30 + %3177 = extractelement <2 x bfloat> %3170, i64 0, !dbg !30 + %3178 = extractelement <2 x bfloat> %3170, i64 1, !dbg !30 + %3179 = extractelement <2 x bfloat> %3172, i64 0, !dbg !30 + %3180 = extractelement <2 x bfloat> %3172, i64 1, !dbg !30 + %3181 = extractelement <2 x bfloat> %3174, i64 0, !dbg !30 + %3182 = extractelement <2 x bfloat> %3174, i64 1, !dbg !30 + %3183 = extractelement <2 x bfloat> %3176, i64 0, !dbg !30 + %3184 = extractelement <2 x bfloat> %3176, i64 1, !dbg !30 + %3185 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %445, i1 %1568) #5, !dbg !30 + %3186 = extractvalue { i32, i32, i32, i32 } %3185, 0, !dbg !30 + %3187 = bitcast i32 %3186 to <2 x bfloat>, !dbg !30 + %3188 = extractvalue { i32, i32, i32, i32 } %3185, 1, !dbg !30 + %3189 = bitcast i32 %3188 to <2 x bfloat>, !dbg !30 + %3190 = extractvalue { i32, i32, i32, i32 } %3185, 2, !dbg !30 + %3191 = bitcast i32 %3190 to <2 x bfloat>, !dbg !30 + %3192 = extractvalue { i32, i32, i32, i32 } %3185, 3, !dbg !30 + %3193 = bitcast i32 %3192 to <2 x bfloat>, !dbg !30 + %3194 = extractelement <2 x bfloat> %3187, i64 0, !dbg !30 + %3195 = extractelement <2 x bfloat> %3187, i64 1, !dbg !30 + %3196 = extractelement <2 x bfloat> %3189, i64 0, !dbg !30 + %3197 = extractelement <2 x bfloat> %3189, i64 1, !dbg !30 + %3198 = extractelement <2 x bfloat> %3191, i64 0, !dbg !30 + %3199 = extractelement <2 x bfloat> %3191, i64 1, !dbg !30 + %3200 = extractelement <2 x bfloat> %3193, i64 0, !dbg !30 + %3201 = extractelement <2 x bfloat> %3193, i64 1, !dbg !30 + %3202 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %446, i1 %1586) #5, !dbg !30 + %3203 = extractvalue { i32, i32, i32, i32 } %3202, 0, !dbg !30 + %3204 = bitcast i32 %3203 to <2 x bfloat>, !dbg !30 + %3205 = extractvalue { i32, i32, i32, i32 } %3202, 1, !dbg !30 + %3206 = bitcast i32 %3205 to <2 x bfloat>, !dbg !30 + %3207 = extractvalue { i32, i32, i32, i32 } %3202, 2, !dbg !30 + %3208 = bitcast i32 %3207 to <2 x bfloat>, !dbg !30 + %3209 = extractvalue { i32, i32, i32, i32 } %3202, 3, !dbg !30 + %3210 = bitcast i32 %3209 to <2 x bfloat>, !dbg !30 + %3211 = extractelement <2 x bfloat> %3204, i64 0, !dbg !30 + %3212 = extractelement <2 x bfloat> %3204, i64 1, !dbg !30 + %3213 = extractelement <2 x bfloat> %3206, i64 0, !dbg !30 + %3214 = extractelement <2 x bfloat> %3206, i64 1, !dbg !30 + %3215 = extractelement <2 x bfloat> %3208, i64 0, !dbg !30 + %3216 = extractelement <2 x bfloat> %3208, i64 1, !dbg !30 + %3217 = extractelement <2 x bfloat> %3210, i64 0, !dbg !30 + %3218 = extractelement <2 x bfloat> %3210, i64 1, !dbg !30 + %3219 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %447, i1 %1604) #5, !dbg !30 + %3220 = extractvalue { i32, i32, i32, i32 } %3219, 0, !dbg !30 + %3221 = bitcast i32 %3220 to <2 x bfloat>, !dbg !30 + %3222 = extractvalue { i32, i32, i32, i32 } %3219, 1, !dbg !30 + %3223 = bitcast i32 %3222 to <2 x bfloat>, !dbg !30 + %3224 = extractvalue { i32, i32, i32, i32 } %3219, 2, !dbg !30 + %3225 = bitcast i32 %3224 to <2 x bfloat>, !dbg !30 + %3226 = extractvalue { i32, i32, i32, i32 } %3219, 3, !dbg !30 + %3227 = bitcast i32 %3226 to <2 x bfloat>, !dbg !30 + %3228 = extractelement <2 x bfloat> %3221, i64 0, !dbg !30 + %3229 = extractelement <2 x bfloat> %3221, i64 1, !dbg !30 + %3230 = extractelement <2 x bfloat> %3223, i64 0, !dbg !30 + %3231 = extractelement <2 x bfloat> %3223, i64 1, !dbg !30 + %3232 = extractelement <2 x bfloat> %3225, i64 0, !dbg !30 + %3233 = extractelement <2 x bfloat> %3225, i64 1, !dbg !30 + %3234 = extractelement <2 x bfloat> %3227, i64 0, !dbg !30 + %3235 = extractelement <2 x bfloat> %3227, i64 1, !dbg !30 + %3236 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %448, i1 %1622) #5, !dbg !30 + %3237 = extractvalue { i32, i32, i32, i32 } %3236, 0, !dbg !30 + %3238 = bitcast i32 %3237 to <2 x bfloat>, !dbg !30 + %3239 = extractvalue { i32, i32, i32, i32 } %3236, 1, !dbg !30 + %3240 = bitcast i32 %3239 to <2 x bfloat>, !dbg !30 + %3241 = extractvalue { i32, i32, i32, i32 } %3236, 2, !dbg !30 + %3242 = bitcast i32 %3241 to <2 x bfloat>, !dbg !30 + %3243 = extractvalue { i32, i32, i32, i32 } %3236, 3, !dbg !30 + %3244 = bitcast i32 %3243 to <2 x bfloat>, !dbg !30 + %3245 = extractelement <2 x bfloat> %3238, i64 0, !dbg !30 + %3246 = extractelement <2 x bfloat> %3238, i64 1, !dbg !30 + %3247 = extractelement <2 x bfloat> %3240, i64 0, !dbg !30 + %3248 = extractelement <2 x bfloat> %3240, i64 1, !dbg !30 + %3249 = extractelement <2 x bfloat> %3242, i64 0, !dbg !30 + %3250 = extractelement <2 x bfloat> %3242, i64 1, !dbg !30 + %3251 = extractelement <2 x bfloat> %3244, i64 0, !dbg !30 + %3252 = extractelement <2 x bfloat> %3244, i64 1, !dbg !30 + %3253 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %449, i1 %1640) #5, !dbg !30 + %3254 = extractvalue { i32, i32, i32, i32 } %3253, 0, !dbg !30 + %3255 = bitcast i32 %3254 to <2 x bfloat>, !dbg !30 + %3256 = extractvalue { i32, i32, i32, i32 } %3253, 1, !dbg !30 + %3257 = bitcast i32 %3256 to <2 x bfloat>, !dbg !30 + %3258 = extractvalue { i32, i32, i32, i32 } %3253, 2, !dbg !30 + %3259 = bitcast i32 %3258 to <2 x bfloat>, !dbg !30 + %3260 = extractvalue { i32, i32, i32, i32 } %3253, 3, !dbg !30 + %3261 = bitcast i32 %3260 to <2 x bfloat>, !dbg !30 + %3262 = extractelement <2 x bfloat> %3255, i64 0, !dbg !30 + %3263 = extractelement <2 x bfloat> %3255, i64 1, !dbg !30 + %3264 = extractelement <2 x bfloat> %3257, i64 0, !dbg !30 + %3265 = extractelement <2 x bfloat> %3257, i64 1, !dbg !30 + %3266 = extractelement <2 x bfloat> %3259, i64 0, !dbg !30 + %3267 = extractelement <2 x bfloat> %3259, i64 1, !dbg !30 + %3268 = extractelement <2 x bfloat> %3261, i64 0, !dbg !30 + %3269 = extractelement <2 x bfloat> %3261, i64 1, !dbg !30 + %3270 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %450, i1 %1658) #5, !dbg !30 + %3271 = extractvalue { i32, i32, i32, i32 } %3270, 0, !dbg !30 + %3272 = bitcast i32 %3271 to <2 x bfloat>, !dbg !30 + %3273 = extractvalue { i32, i32, i32, i32 } %3270, 1, !dbg !30 + %3274 = bitcast i32 %3273 to <2 x bfloat>, !dbg !30 + %3275 = extractvalue { i32, i32, i32, i32 } %3270, 2, !dbg !30 + %3276 = bitcast i32 %3275 to <2 x bfloat>, !dbg !30 + %3277 = extractvalue { i32, i32, i32, i32 } %3270, 3, !dbg !30 + %3278 = bitcast i32 %3277 to <2 x bfloat>, !dbg !30 + %3279 = extractelement <2 x bfloat> %3272, i64 0, !dbg !30 + %3280 = extractelement <2 x bfloat> %3272, i64 1, !dbg !30 + %3281 = extractelement <2 x bfloat> %3274, i64 0, !dbg !30 + %3282 = extractelement <2 x bfloat> %3274, i64 1, !dbg !30 + %3283 = extractelement <2 x bfloat> %3276, i64 0, !dbg !30 + %3284 = extractelement <2 x bfloat> %3276, i64 1, !dbg !30 + %3285 = extractelement <2 x bfloat> %3278, i64 0, !dbg !30 + %3286 = extractelement <2 x bfloat> %3278, i64 1, !dbg !30 + %3287 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %451, i1 %1676) #5, !dbg !30 + %3288 = extractvalue { i32, i32, i32, i32 } %3287, 0, !dbg !30 + %3289 = bitcast i32 %3288 to <2 x bfloat>, !dbg !30 + %3290 = extractvalue { i32, i32, i32, i32 } %3287, 1, !dbg !30 + %3291 = bitcast i32 %3290 to <2 x bfloat>, !dbg !30 + %3292 = extractvalue { i32, i32, i32, i32 } %3287, 2, !dbg !30 + %3293 = bitcast i32 %3292 to <2 x bfloat>, !dbg !30 + %3294 = extractvalue { i32, i32, i32, i32 } %3287, 3, !dbg !30 + %3295 = bitcast i32 %3294 to <2 x bfloat>, !dbg !30 + %3296 = extractelement <2 x bfloat> %3289, i64 0, !dbg !30 + %3297 = extractelement <2 x bfloat> %3289, i64 1, !dbg !30 + %3298 = extractelement <2 x bfloat> %3291, i64 0, !dbg !30 + %3299 = extractelement <2 x bfloat> %3291, i64 1, !dbg !30 + %3300 = extractelement <2 x bfloat> %3293, i64 0, !dbg !30 + %3301 = extractelement <2 x bfloat> %3293, i64 1, !dbg !30 + %3302 = extractelement <2 x bfloat> %3295, i64 0, !dbg !30 + %3303 = extractelement <2 x bfloat> %3295, i64 1, !dbg !30 + %3304 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %452, i1 %1694) #5, !dbg !30 + %3305 = extractvalue { i32, i32, i32, i32 } %3304, 0, !dbg !30 + %3306 = bitcast i32 %3305 to <2 x bfloat>, !dbg !30 + %3307 = extractvalue { i32, i32, i32, i32 } %3304, 1, !dbg !30 + %3308 = bitcast i32 %3307 to <2 x bfloat>, !dbg !30 + %3309 = extractvalue { i32, i32, i32, i32 } %3304, 2, !dbg !30 + %3310 = bitcast i32 %3309 to <2 x bfloat>, !dbg !30 + %3311 = extractvalue { i32, i32, i32, i32 } %3304, 3, !dbg !30 + %3312 = bitcast i32 %3311 to <2 x bfloat>, !dbg !30 + %3313 = extractelement <2 x bfloat> %3306, i64 0, !dbg !30 + %3314 = extractelement <2 x bfloat> %3306, i64 1, !dbg !30 + %3315 = extractelement <2 x bfloat> %3308, i64 0, !dbg !30 + %3316 = extractelement <2 x bfloat> %3308, i64 1, !dbg !30 + %3317 = extractelement <2 x bfloat> %3310, i64 0, !dbg !30 + %3318 = extractelement <2 x bfloat> %3310, i64 1, !dbg !30 + %3319 = extractelement <2 x bfloat> %3312, i64 0, !dbg !30 + %3320 = extractelement <2 x bfloat> %3312, i64 1, !dbg !30 + %3321 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %453, i1 %1712) #5, !dbg !30 + %3322 = extractvalue { i32, i32, i32, i32 } %3321, 0, !dbg !30 + %3323 = bitcast i32 %3322 to <2 x bfloat>, !dbg !30 + %3324 = extractvalue { i32, i32, i32, i32 } %3321, 1, !dbg !30 + %3325 = bitcast i32 %3324 to <2 x bfloat>, !dbg !30 + %3326 = extractvalue { i32, i32, i32, i32 } %3321, 2, !dbg !30 + %3327 = bitcast i32 %3326 to <2 x bfloat>, !dbg !30 + %3328 = extractvalue { i32, i32, i32, i32 } %3321, 3, !dbg !30 + %3329 = bitcast i32 %3328 to <2 x bfloat>, !dbg !30 + %3330 = extractelement <2 x bfloat> %3323, i64 0, !dbg !30 + %3331 = extractelement <2 x bfloat> %3323, i64 1, !dbg !30 + %3332 = extractelement <2 x bfloat> %3325, i64 0, !dbg !30 + %3333 = extractelement <2 x bfloat> %3325, i64 1, !dbg !30 + %3334 = extractelement <2 x bfloat> %3327, i64 0, !dbg !30 + %3335 = extractelement <2 x bfloat> %3327, i64 1, !dbg !30 + %3336 = extractelement <2 x bfloat> %3329, i64 0, !dbg !30 + %3337 = extractelement <2 x bfloat> %3329, i64 1, !dbg !30 + %3338 = fpext bfloat %2259 to float, !dbg !31 + %3339 = fpext bfloat %2260 to float, !dbg !31 + %3340 = fpext bfloat %2261 to float, !dbg !31 + %3341 = fpext bfloat %2262 to float, !dbg !31 + %3342 = fpext bfloat %2263 to float, !dbg !31 + %3343 = fpext bfloat %2264 to float, !dbg !31 + %3344 = fpext bfloat %2265 to float, !dbg !31 + %3345 = fpext bfloat %2266 to float, !dbg !31 + %3346 = fpext bfloat %2276 to float, !dbg !31 + %3347 = fpext bfloat %2277 to float, !dbg !31 + %3348 = fpext bfloat %2278 to float, !dbg !31 + %3349 = fpext bfloat %2279 to float, !dbg !31 + %3350 = fpext bfloat %2280 to float, !dbg !31 + %3351 = fpext bfloat %2281 to float, !dbg !31 + %3352 = fpext bfloat %2282 to float, !dbg !31 + %3353 = fpext bfloat %2283 to float, !dbg !31 + %3354 = fpext bfloat %2293 to float, !dbg !31 + %3355 = fpext bfloat %2294 to float, !dbg !31 + %3356 = fpext bfloat %2295 to float, !dbg !31 + %3357 = fpext bfloat %2296 to float, !dbg !31 + %3358 = fpext bfloat %2297 to float, !dbg !31 + %3359 = fpext bfloat %2298 to float, !dbg !31 + %3360 = fpext bfloat %2299 to float, !dbg !31 + %3361 = fpext bfloat %2300 to float, !dbg !31 + %3362 = fpext bfloat %2310 to float, !dbg !31 + %3363 = fpext bfloat %2311 to float, !dbg !31 + %3364 = fpext bfloat %2312 to float, !dbg !31 + %3365 = fpext bfloat %2313 to float, !dbg !31 + %3366 = fpext bfloat %2314 to float, !dbg !31 + %3367 = fpext bfloat %2315 to float, !dbg !31 + %3368 = fpext bfloat %2316 to float, !dbg !31 + %3369 = fpext bfloat %2317 to float, !dbg !31 + %3370 = fpext bfloat %2327 to float, !dbg !31 + %3371 = fpext bfloat %2328 to float, !dbg !31 + %3372 = fpext bfloat %2329 to float, !dbg !31 + %3373 = fpext bfloat %2330 to float, !dbg !31 + %3374 = fpext bfloat %2331 to float, !dbg !31 + %3375 = fpext bfloat %2332 to float, !dbg !31 + %3376 = fpext bfloat %2333 to float, !dbg !31 + %3377 = fpext bfloat %2334 to float, !dbg !31 + %3378 = fpext bfloat %2344 to float, !dbg !31 + %3379 = fpext bfloat %2345 to float, !dbg !31 + %3380 = fpext bfloat %2346 to float, !dbg !31 + %3381 = fpext bfloat %2347 to float, !dbg !31 + %3382 = fpext bfloat %2348 to float, !dbg !31 + %3383 = fpext bfloat %2349 to float, !dbg !31 + %3384 = fpext bfloat %2350 to float, !dbg !31 + %3385 = fpext bfloat %2351 to float, !dbg !31 + %3386 = fpext bfloat %2361 to float, !dbg !31 + %3387 = fpext bfloat %2362 to float, !dbg !31 + %3388 = fpext bfloat %2363 to float, !dbg !31 + %3389 = fpext bfloat %2364 to float, !dbg !31 + %3390 = fpext bfloat %2365 to float, !dbg !31 + %3391 = fpext bfloat %2366 to float, !dbg !31 + %3392 = fpext bfloat %2367 to float, !dbg !31 + %3393 = fpext bfloat %2368 to float, !dbg !31 + %3394 = fpext bfloat %2378 to float, !dbg !31 + %3395 = fpext bfloat %2379 to float, !dbg !31 + %3396 = fpext bfloat %2380 to float, !dbg !31 + %3397 = fpext bfloat %2381 to float, !dbg !31 + %3398 = fpext bfloat %2382 to float, !dbg !31 + %3399 = fpext bfloat %2383 to float, !dbg !31 + %3400 = fpext bfloat %2384 to float, !dbg !31 + %3401 = fpext bfloat %2385 to float, !dbg !31 + %3402 = fpext bfloat %2395 to float, !dbg !31 + %3403 = fpext bfloat %2396 to float, !dbg !31 + %3404 = fpext bfloat %2397 to float, !dbg !31 + %3405 = fpext bfloat %2398 to float, !dbg !31 + %3406 = fpext bfloat %2399 to float, !dbg !31 + %3407 = fpext bfloat %2400 to float, !dbg !31 + %3408 = fpext bfloat %2401 to float, !dbg !31 + %3409 = fpext bfloat %2402 to float, !dbg !31 + %3410 = fpext bfloat %2412 to float, !dbg !31 + %3411 = fpext bfloat %2413 to float, !dbg !31 + %3412 = fpext bfloat %2414 to float, !dbg !31 + %3413 = fpext bfloat %2415 to float, !dbg !31 + %3414 = fpext bfloat %2416 to float, !dbg !31 + %3415 = fpext bfloat %2417 to float, !dbg !31 + %3416 = fpext bfloat %2418 to float, !dbg !31 + %3417 = fpext bfloat %2419 to float, !dbg !31 + %3418 = fpext bfloat %2429 to float, !dbg !31 + %3419 = fpext bfloat %2430 to float, !dbg !31 + %3420 = fpext bfloat %2431 to float, !dbg !31 + %3421 = fpext bfloat %2432 to float, !dbg !31 + %3422 = fpext bfloat %2433 to float, !dbg !31 + %3423 = fpext bfloat %2434 to float, !dbg !31 + %3424 = fpext bfloat %2435 to float, !dbg !31 + %3425 = fpext bfloat %2436 to float, !dbg !31 + %3426 = fpext bfloat %2446 to float, !dbg !31 + %3427 = fpext bfloat %2447 to float, !dbg !31 + %3428 = fpext bfloat %2448 to float, !dbg !31 + %3429 = fpext bfloat %2449 to float, !dbg !31 + %3430 = fpext bfloat %2450 to float, !dbg !31 + %3431 = fpext bfloat %2451 to float, !dbg !31 + %3432 = fpext bfloat %2452 to float, !dbg !31 + %3433 = fpext bfloat %2453 to float, !dbg !31 + %3434 = fpext bfloat %2463 to float, !dbg !31 + %3435 = fpext bfloat %2464 to float, !dbg !31 + %3436 = fpext bfloat %2465 to float, !dbg !31 + %3437 = fpext bfloat %2466 to float, !dbg !31 + %3438 = fpext bfloat %2467 to float, !dbg !31 + %3439 = fpext bfloat %2468 to float, !dbg !31 + %3440 = fpext bfloat %2469 to float, !dbg !31 + %3441 = fpext bfloat %2470 to float, !dbg !31 + %3442 = fpext bfloat %2480 to float, !dbg !31 + %3443 = fpext bfloat %2481 to float, !dbg !31 + %3444 = fpext bfloat %2482 to float, !dbg !31 + %3445 = fpext bfloat %2483 to float, !dbg !31 + %3446 = fpext bfloat %2484 to float, !dbg !31 + %3447 = fpext bfloat %2485 to float, !dbg !31 + %3448 = fpext bfloat %2486 to float, !dbg !31 + %3449 = fpext bfloat %2487 to float, !dbg !31 + %3450 = fpext bfloat %2497 to float, !dbg !31 + %3451 = fpext bfloat %2498 to float, !dbg !31 + %3452 = fpext bfloat %2499 to float, !dbg !31 + %3453 = fpext bfloat %2500 to float, !dbg !31 + %3454 = fpext bfloat %2501 to float, !dbg !31 + %3455 = fpext bfloat %2502 to float, !dbg !31 + %3456 = fpext bfloat %2503 to float, !dbg !31 + %3457 = fpext bfloat %2504 to float, !dbg !31 + %3458 = fpext bfloat %2514 to float, !dbg !31 + %3459 = fpext bfloat %2515 to float, !dbg !31 + %3460 = fpext bfloat %2516 to float, !dbg !31 + %3461 = fpext bfloat %2517 to float, !dbg !31 + %3462 = fpext bfloat %2518 to float, !dbg !31 + %3463 = fpext bfloat %2519 to float, !dbg !31 + %3464 = fpext bfloat %2520 to float, !dbg !31 + %3465 = fpext bfloat %2521 to float, !dbg !31 + %3466 = fpext bfloat %2531 to float, !dbg !31 + %3467 = fpext bfloat %2532 to float, !dbg !31 + %3468 = fpext bfloat %2533 to float, !dbg !31 + %3469 = fpext bfloat %2534 to float, !dbg !31 + %3470 = fpext bfloat %2535 to float, !dbg !31 + %3471 = fpext bfloat %2536 to float, !dbg !31 + %3472 = fpext bfloat %2537 to float, !dbg !31 + %3473 = fpext bfloat %2538 to float, !dbg !31 + %3474 = fpext bfloat %2548 to float, !dbg !31 + %3475 = fpext bfloat %2549 to float, !dbg !31 + %3476 = fpext bfloat %2550 to float, !dbg !31 + %3477 = fpext bfloat %2551 to float, !dbg !31 + %3478 = fpext bfloat %2552 to float, !dbg !31 + %3479 = fpext bfloat %2553 to float, !dbg !31 + %3480 = fpext bfloat %2554 to float, !dbg !31 + %3481 = fpext bfloat %2555 to float, !dbg !31 + %3482 = fpext bfloat %2565 to float, !dbg !31 + %3483 = fpext bfloat %2566 to float, !dbg !31 + %3484 = fpext bfloat %2567 to float, !dbg !31 + %3485 = fpext bfloat %2568 to float, !dbg !31 + %3486 = fpext bfloat %2569 to float, !dbg !31 + %3487 = fpext bfloat %2570 to float, !dbg !31 + %3488 = fpext bfloat %2571 to float, !dbg !31 + %3489 = fpext bfloat %2572 to float, !dbg !31 + %3490 = fpext bfloat %2582 to float, !dbg !31 + %3491 = fpext bfloat %2583 to float, !dbg !31 + %3492 = fpext bfloat %2584 to float, !dbg !31 + %3493 = fpext bfloat %2585 to float, !dbg !31 + %3494 = fpext bfloat %2586 to float, !dbg !31 + %3495 = fpext bfloat %2587 to float, !dbg !31 + %3496 = fpext bfloat %2588 to float, !dbg !31 + %3497 = fpext bfloat %2589 to float, !dbg !31 + %3498 = fpext bfloat %2599 to float, !dbg !31 + %3499 = fpext bfloat %2600 to float, !dbg !31 + %3500 = fpext bfloat %2601 to float, !dbg !31 + %3501 = fpext bfloat %2602 to float, !dbg !31 + %3502 = fpext bfloat %2603 to float, !dbg !31 + %3503 = fpext bfloat %2604 to float, !dbg !31 + %3504 = fpext bfloat %2605 to float, !dbg !31 + %3505 = fpext bfloat %2606 to float, !dbg !31 + %3506 = fpext bfloat %2616 to float, !dbg !31 + %3507 = fpext bfloat %2617 to float, !dbg !31 + %3508 = fpext bfloat %2618 to float, !dbg !31 + %3509 = fpext bfloat %2619 to float, !dbg !31 + %3510 = fpext bfloat %2620 to float, !dbg !31 + %3511 = fpext bfloat %2621 to float, !dbg !31 + %3512 = fpext bfloat %2622 to float, !dbg !31 + %3513 = fpext bfloat %2623 to float, !dbg !31 + %3514 = fpext bfloat %2633 to float, !dbg !31 + %3515 = fpext bfloat %2634 to float, !dbg !31 + %3516 = fpext bfloat %2635 to float, !dbg !31 + %3517 = fpext bfloat %2636 to float, !dbg !31 + %3518 = fpext bfloat %2637 to float, !dbg !31 + %3519 = fpext bfloat %2638 to float, !dbg !31 + %3520 = fpext bfloat %2639 to float, !dbg !31 + %3521 = fpext bfloat %2640 to float, !dbg !31 + %3522 = fpext bfloat %2650 to float, !dbg !31 + %3523 = fpext bfloat %2651 to float, !dbg !31 + %3524 = fpext bfloat %2652 to float, !dbg !31 + %3525 = fpext bfloat %2653 to float, !dbg !31 + %3526 = fpext bfloat %2654 to float, !dbg !31 + %3527 = fpext bfloat %2655 to float, !dbg !31 + %3528 = fpext bfloat %2656 to float, !dbg !31 + %3529 = fpext bfloat %2657 to float, !dbg !31 + %3530 = fpext bfloat %2667 to float, !dbg !31 + %3531 = fpext bfloat %2668 to float, !dbg !31 + %3532 = fpext bfloat %2669 to float, !dbg !31 + %3533 = fpext bfloat %2670 to float, !dbg !31 + %3534 = fpext bfloat %2671 to float, !dbg !31 + %3535 = fpext bfloat %2672 to float, !dbg !31 + %3536 = fpext bfloat %2673 to float, !dbg !31 + %3537 = fpext bfloat %2674 to float, !dbg !31 + %3538 = fpext bfloat %2684 to float, !dbg !31 + %3539 = fpext bfloat %2685 to float, !dbg !31 + %3540 = fpext bfloat %2686 to float, !dbg !31 + %3541 = fpext bfloat %2687 to float, !dbg !31 + %3542 = fpext bfloat %2688 to float, !dbg !31 + %3543 = fpext bfloat %2689 to float, !dbg !31 + %3544 = fpext bfloat %2690 to float, !dbg !31 + %3545 = fpext bfloat %2691 to float, !dbg !31 + %3546 = fpext bfloat %2701 to float, !dbg !31 + %3547 = fpext bfloat %2702 to float, !dbg !31 + %3548 = fpext bfloat %2703 to float, !dbg !31 + %3549 = fpext bfloat %2704 to float, !dbg !31 + %3550 = fpext bfloat %2705 to float, !dbg !31 + %3551 = fpext bfloat %2706 to float, !dbg !31 + %3552 = fpext bfloat %2707 to float, !dbg !31 + %3553 = fpext bfloat %2708 to float, !dbg !31 + %3554 = fpext bfloat %2718 to float, !dbg !31 + %3555 = fpext bfloat %2719 to float, !dbg !31 + %3556 = fpext bfloat %2720 to float, !dbg !31 + %3557 = fpext bfloat %2721 to float, !dbg !31 + %3558 = fpext bfloat %2722 to float, !dbg !31 + %3559 = fpext bfloat %2723 to float, !dbg !31 + %3560 = fpext bfloat %2724 to float, !dbg !31 + %3561 = fpext bfloat %2725 to float, !dbg !31 + %3562 = fpext bfloat %2735 to float, !dbg !31 + %3563 = fpext bfloat %2736 to float, !dbg !31 + %3564 = fpext bfloat %2737 to float, !dbg !31 + %3565 = fpext bfloat %2738 to float, !dbg !31 + %3566 = fpext bfloat %2739 to float, !dbg !31 + %3567 = fpext bfloat %2740 to float, !dbg !31 + %3568 = fpext bfloat %2741 to float, !dbg !31 + %3569 = fpext bfloat %2742 to float, !dbg !31 + %3570 = fpext bfloat %2752 to float, !dbg !31 + %3571 = fpext bfloat %2753 to float, !dbg !31 + %3572 = fpext bfloat %2754 to float, !dbg !31 + %3573 = fpext bfloat %2755 to float, !dbg !31 + %3574 = fpext bfloat %2756 to float, !dbg !31 + %3575 = fpext bfloat %2757 to float, !dbg !31 + %3576 = fpext bfloat %2758 to float, !dbg !31 + %3577 = fpext bfloat %2759 to float, !dbg !31 + %3578 = fpext bfloat %2769 to float, !dbg !31 + %3579 = fpext bfloat %2770 to float, !dbg !31 + %3580 = fpext bfloat %2771 to float, !dbg !31 + %3581 = fpext bfloat %2772 to float, !dbg !31 + %3582 = fpext bfloat %2773 to float, !dbg !31 + %3583 = fpext bfloat %2774 to float, !dbg !31 + %3584 = fpext bfloat %2775 to float, !dbg !31 + %3585 = fpext bfloat %2776 to float, !dbg !31 + %3586 = fpext bfloat %2786 to float, !dbg !31 + %3587 = fpext bfloat %2787 to float, !dbg !31 + %3588 = fpext bfloat %2788 to float, !dbg !31 + %3589 = fpext bfloat %2789 to float, !dbg !31 + %3590 = fpext bfloat %2790 to float, !dbg !31 + %3591 = fpext bfloat %2791 to float, !dbg !31 + %3592 = fpext bfloat %2792 to float, !dbg !31 + %3593 = fpext bfloat %2793 to float, !dbg !31 + %3594 = fpext bfloat %2803 to float, !dbg !31 + %3595 = fpext bfloat %2804 to float, !dbg !31 + %3596 = fpext bfloat %2805 to float, !dbg !31 + %3597 = fpext bfloat %2806 to float, !dbg !31 + %3598 = fpext bfloat %2807 to float, !dbg !31 + %3599 = fpext bfloat %2808 to float, !dbg !31 + %3600 = fpext bfloat %2809 to float, !dbg !31 + %3601 = fpext bfloat %2810 to float, !dbg !31 + %3602 = fpext bfloat %2820 to float, !dbg !31 + %3603 = fpext bfloat %2821 to float, !dbg !31 + %3604 = fpext bfloat %2822 to float, !dbg !31 + %3605 = fpext bfloat %2823 to float, !dbg !31 + %3606 = fpext bfloat %2824 to float, !dbg !31 + %3607 = fpext bfloat %2825 to float, !dbg !31 + %3608 = fpext bfloat %2826 to float, !dbg !31 + %3609 = fpext bfloat %2827 to float, !dbg !31 + %3610 = fpext bfloat %2837 to float, !dbg !31 + %3611 = fpext bfloat %2838 to float, !dbg !31 + %3612 = fpext bfloat %2839 to float, !dbg !31 + %3613 = fpext bfloat %2840 to float, !dbg !31 + %3614 = fpext bfloat %2841 to float, !dbg !31 + %3615 = fpext bfloat %2842 to float, !dbg !31 + %3616 = fpext bfloat %2843 to float, !dbg !31 + %3617 = fpext bfloat %2844 to float, !dbg !31 + %3618 = fpext bfloat %2854 to float, !dbg !31 + %3619 = fpext bfloat %2855 to float, !dbg !31 + %3620 = fpext bfloat %2856 to float, !dbg !31 + %3621 = fpext bfloat %2857 to float, !dbg !31 + %3622 = fpext bfloat %2858 to float, !dbg !31 + %3623 = fpext bfloat %2859 to float, !dbg !31 + %3624 = fpext bfloat %2860 to float, !dbg !31 + %3625 = fpext bfloat %2861 to float, !dbg !31 + %3626 = fpext bfloat %2871 to float, !dbg !31 + %3627 = fpext bfloat %2872 to float, !dbg !31 + %3628 = fpext bfloat %2873 to float, !dbg !31 + %3629 = fpext bfloat %2874 to float, !dbg !31 + %3630 = fpext bfloat %2875 to float, !dbg !31 + %3631 = fpext bfloat %2876 to float, !dbg !31 + %3632 = fpext bfloat %2877 to float, !dbg !31 + %3633 = fpext bfloat %2878 to float, !dbg !31 + %3634 = fpext bfloat %2888 to float, !dbg !31 + %3635 = fpext bfloat %2889 to float, !dbg !31 + %3636 = fpext bfloat %2890 to float, !dbg !31 + %3637 = fpext bfloat %2891 to float, !dbg !31 + %3638 = fpext bfloat %2892 to float, !dbg !31 + %3639 = fpext bfloat %2893 to float, !dbg !31 + %3640 = fpext bfloat %2894 to float, !dbg !31 + %3641 = fpext bfloat %2895 to float, !dbg !31 + %3642 = fpext bfloat %2905 to float, !dbg !31 + %3643 = fpext bfloat %2906 to float, !dbg !31 + %3644 = fpext bfloat %2907 to float, !dbg !31 + %3645 = fpext bfloat %2908 to float, !dbg !31 + %3646 = fpext bfloat %2909 to float, !dbg !31 + %3647 = fpext bfloat %2910 to float, !dbg !31 + %3648 = fpext bfloat %2911 to float, !dbg !31 + %3649 = fpext bfloat %2912 to float, !dbg !31 + %3650 = fpext bfloat %2922 to float, !dbg !31 + %3651 = fpext bfloat %2923 to float, !dbg !31 + %3652 = fpext bfloat %2924 to float, !dbg !31 + %3653 = fpext bfloat %2925 to float, !dbg !31 + %3654 = fpext bfloat %2926 to float, !dbg !31 + %3655 = fpext bfloat %2927 to float, !dbg !31 + %3656 = fpext bfloat %2928 to float, !dbg !31 + %3657 = fpext bfloat %2929 to float, !dbg !31 + %3658 = fpext bfloat %2939 to float, !dbg !31 + %3659 = fpext bfloat %2940 to float, !dbg !31 + %3660 = fpext bfloat %2941 to float, !dbg !31 + %3661 = fpext bfloat %2942 to float, !dbg !31 + %3662 = fpext bfloat %2943 to float, !dbg !31 + %3663 = fpext bfloat %2944 to float, !dbg !31 + %3664 = fpext bfloat %2945 to float, !dbg !31 + %3665 = fpext bfloat %2946 to float, !dbg !31 + %3666 = fpext bfloat %2956 to float, !dbg !31 + %3667 = fpext bfloat %2957 to float, !dbg !31 + %3668 = fpext bfloat %2958 to float, !dbg !31 + %3669 = fpext bfloat %2959 to float, !dbg !31 + %3670 = fpext bfloat %2960 to float, !dbg !31 + %3671 = fpext bfloat %2961 to float, !dbg !31 + %3672 = fpext bfloat %2962 to float, !dbg !31 + %3673 = fpext bfloat %2963 to float, !dbg !31 + %3674 = fpext bfloat %2973 to float, !dbg !31 + %3675 = fpext bfloat %2974 to float, !dbg !31 + %3676 = fpext bfloat %2975 to float, !dbg !31 + %3677 = fpext bfloat %2976 to float, !dbg !31 + %3678 = fpext bfloat %2977 to float, !dbg !31 + %3679 = fpext bfloat %2978 to float, !dbg !31 + %3680 = fpext bfloat %2979 to float, !dbg !31 + %3681 = fpext bfloat %2980 to float, !dbg !31 + %3682 = fpext bfloat %2990 to float, !dbg !31 + %3683 = fpext bfloat %2991 to float, !dbg !31 + %3684 = fpext bfloat %2992 to float, !dbg !31 + %3685 = fpext bfloat %2993 to float, !dbg !31 + %3686 = fpext bfloat %2994 to float, !dbg !31 + %3687 = fpext bfloat %2995 to float, !dbg !31 + %3688 = fpext bfloat %2996 to float, !dbg !31 + %3689 = fpext bfloat %2997 to float, !dbg !31 + %3690 = fpext bfloat %3007 to float, !dbg !31 + %3691 = fpext bfloat %3008 to float, !dbg !31 + %3692 = fpext bfloat %3009 to float, !dbg !31 + %3693 = fpext bfloat %3010 to float, !dbg !31 + %3694 = fpext bfloat %3011 to float, !dbg !31 + %3695 = fpext bfloat %3012 to float, !dbg !31 + %3696 = fpext bfloat %3013 to float, !dbg !31 + %3697 = fpext bfloat %3014 to float, !dbg !31 + %3698 = fpext bfloat %3024 to float, !dbg !31 + %3699 = fpext bfloat %3025 to float, !dbg !31 + %3700 = fpext bfloat %3026 to float, !dbg !31 + %3701 = fpext bfloat %3027 to float, !dbg !31 + %3702 = fpext bfloat %3028 to float, !dbg !31 + %3703 = fpext bfloat %3029 to float, !dbg !31 + %3704 = fpext bfloat %3030 to float, !dbg !31 + %3705 = fpext bfloat %3031 to float, !dbg !31 + %3706 = fpext bfloat %3041 to float, !dbg !31 + %3707 = fpext bfloat %3042 to float, !dbg !31 + %3708 = fpext bfloat %3043 to float, !dbg !31 + %3709 = fpext bfloat %3044 to float, !dbg !31 + %3710 = fpext bfloat %3045 to float, !dbg !31 + %3711 = fpext bfloat %3046 to float, !dbg !31 + %3712 = fpext bfloat %3047 to float, !dbg !31 + %3713 = fpext bfloat %3048 to float, !dbg !31 + %3714 = fpext bfloat %3058 to float, !dbg !31 + %3715 = fpext bfloat %3059 to float, !dbg !31 + %3716 = fpext bfloat %3060 to float, !dbg !31 + %3717 = fpext bfloat %3061 to float, !dbg !31 + %3718 = fpext bfloat %3062 to float, !dbg !31 + %3719 = fpext bfloat %3063 to float, !dbg !31 + %3720 = fpext bfloat %3064 to float, !dbg !31 + %3721 = fpext bfloat %3065 to float, !dbg !31 + %3722 = fpext bfloat %3075 to float, !dbg !31 + %3723 = fpext bfloat %3076 to float, !dbg !31 + %3724 = fpext bfloat %3077 to float, !dbg !31 + %3725 = fpext bfloat %3078 to float, !dbg !31 + %3726 = fpext bfloat %3079 to float, !dbg !31 + %3727 = fpext bfloat %3080 to float, !dbg !31 + %3728 = fpext bfloat %3081 to float, !dbg !31 + %3729 = fpext bfloat %3082 to float, !dbg !31 + %3730 = fpext bfloat %3092 to float, !dbg !31 + %3731 = fpext bfloat %3093 to float, !dbg !31 + %3732 = fpext bfloat %3094 to float, !dbg !31 + %3733 = fpext bfloat %3095 to float, !dbg !31 + %3734 = fpext bfloat %3096 to float, !dbg !31 + %3735 = fpext bfloat %3097 to float, !dbg !31 + %3736 = fpext bfloat %3098 to float, !dbg !31 + %3737 = fpext bfloat %3099 to float, !dbg !31 + %3738 = fpext bfloat %3109 to float, !dbg !31 + %3739 = fpext bfloat %3110 to float, !dbg !31 + %3740 = fpext bfloat %3111 to float, !dbg !31 + %3741 = fpext bfloat %3112 to float, !dbg !31 + %3742 = fpext bfloat %3113 to float, !dbg !31 + %3743 = fpext bfloat %3114 to float, !dbg !31 + %3744 = fpext bfloat %3115 to float, !dbg !31 + %3745 = fpext bfloat %3116 to float, !dbg !31 + %3746 = fpext bfloat %3126 to float, !dbg !31 + %3747 = fpext bfloat %3127 to float, !dbg !31 + %3748 = fpext bfloat %3128 to float, !dbg !31 + %3749 = fpext bfloat %3129 to float, !dbg !31 + %3750 = fpext bfloat %3130 to float, !dbg !31 + %3751 = fpext bfloat %3131 to float, !dbg !31 + %3752 = fpext bfloat %3132 to float, !dbg !31 + %3753 = fpext bfloat %3133 to float, !dbg !31 + %3754 = fpext bfloat %3143 to float, !dbg !31 + %3755 = fpext bfloat %3144 to float, !dbg !31 + %3756 = fpext bfloat %3145 to float, !dbg !31 + %3757 = fpext bfloat %3146 to float, !dbg !31 + %3758 = fpext bfloat %3147 to float, !dbg !31 + %3759 = fpext bfloat %3148 to float, !dbg !31 + %3760 = fpext bfloat %3149 to float, !dbg !31 + %3761 = fpext bfloat %3150 to float, !dbg !31 + %3762 = fpext bfloat %3160 to float, !dbg !31 + %3763 = fpext bfloat %3161 to float, !dbg !31 + %3764 = fpext bfloat %3162 to float, !dbg !31 + %3765 = fpext bfloat %3163 to float, !dbg !31 + %3766 = fpext bfloat %3164 to float, !dbg !31 + %3767 = fpext bfloat %3165 to float, !dbg !31 + %3768 = fpext bfloat %3166 to float, !dbg !31 + %3769 = fpext bfloat %3167 to float, !dbg !31 + %3770 = fpext bfloat %3177 to float, !dbg !31 + %3771 = fpext bfloat %3178 to float, !dbg !31 + %3772 = fpext bfloat %3179 to float, !dbg !31 + %3773 = fpext bfloat %3180 to float, !dbg !31 + %3774 = fpext bfloat %3181 to float, !dbg !31 + %3775 = fpext bfloat %3182 to float, !dbg !31 + %3776 = fpext bfloat %3183 to float, !dbg !31 + %3777 = fpext bfloat %3184 to float, !dbg !31 + %3778 = fpext bfloat %3194 to float, !dbg !31 + %3779 = fpext bfloat %3195 to float, !dbg !31 + %3780 = fpext bfloat %3196 to float, !dbg !31 + %3781 = fpext bfloat %3197 to float, !dbg !31 + %3782 = fpext bfloat %3198 to float, !dbg !31 + %3783 = fpext bfloat %3199 to float, !dbg !31 + %3784 = fpext bfloat %3200 to float, !dbg !31 + %3785 = fpext bfloat %3201 to float, !dbg !31 + %3786 = fpext bfloat %3211 to float, !dbg !31 + %3787 = fpext bfloat %3212 to float, !dbg !31 + %3788 = fpext bfloat %3213 to float, !dbg !31 + %3789 = fpext bfloat %3214 to float, !dbg !31 + %3790 = fpext bfloat %3215 to float, !dbg !31 + %3791 = fpext bfloat %3216 to float, !dbg !31 + %3792 = fpext bfloat %3217 to float, !dbg !31 + %3793 = fpext bfloat %3218 to float, !dbg !31 + %3794 = fpext bfloat %3228 to float, !dbg !31 + %3795 = fpext bfloat %3229 to float, !dbg !31 + %3796 = fpext bfloat %3230 to float, !dbg !31 + %3797 = fpext bfloat %3231 to float, !dbg !31 + %3798 = fpext bfloat %3232 to float, !dbg !31 + %3799 = fpext bfloat %3233 to float, !dbg !31 + %3800 = fpext bfloat %3234 to float, !dbg !31 + %3801 = fpext bfloat %3235 to float, !dbg !31 + %3802 = fpext bfloat %3245 to float, !dbg !31 + %3803 = fpext bfloat %3246 to float, !dbg !31 + %3804 = fpext bfloat %3247 to float, !dbg !31 + %3805 = fpext bfloat %3248 to float, !dbg !31 + %3806 = fpext bfloat %3249 to float, !dbg !31 + %3807 = fpext bfloat %3250 to float, !dbg !31 + %3808 = fpext bfloat %3251 to float, !dbg !31 + %3809 = fpext bfloat %3252 to float, !dbg !31 + %3810 = fpext bfloat %3262 to float, !dbg !31 + %3811 = fpext bfloat %3263 to float, !dbg !31 + %3812 = fpext bfloat %3264 to float, !dbg !31 + %3813 = fpext bfloat %3265 to float, !dbg !31 + %3814 = fpext bfloat %3266 to float, !dbg !31 + %3815 = fpext bfloat %3267 to float, !dbg !31 + %3816 = fpext bfloat %3268 to float, !dbg !31 + %3817 = fpext bfloat %3269 to float, !dbg !31 + %3818 = fpext bfloat %3279 to float, !dbg !31 + %3819 = fpext bfloat %3280 to float, !dbg !31 + %3820 = fpext bfloat %3281 to float, !dbg !31 + %3821 = fpext bfloat %3282 to float, !dbg !31 + %3822 = fpext bfloat %3283 to float, !dbg !31 + %3823 = fpext bfloat %3284 to float, !dbg !31 + %3824 = fpext bfloat %3285 to float, !dbg !31 + %3825 = fpext bfloat %3286 to float, !dbg !31 + %3826 = fpext bfloat %3296 to float, !dbg !31 + %3827 = fpext bfloat %3297 to float, !dbg !31 + %3828 = fpext bfloat %3298 to float, !dbg !31 + %3829 = fpext bfloat %3299 to float, !dbg !31 + %3830 = fpext bfloat %3300 to float, !dbg !31 + %3831 = fpext bfloat %3301 to float, !dbg !31 + %3832 = fpext bfloat %3302 to float, !dbg !31 + %3833 = fpext bfloat %3303 to float, !dbg !31 + %3834 = fpext bfloat %3313 to float, !dbg !31 + %3835 = fpext bfloat %3314 to float, !dbg !31 + %3836 = fpext bfloat %3315 to float, !dbg !31 + %3837 = fpext bfloat %3316 to float, !dbg !31 + %3838 = fpext bfloat %3317 to float, !dbg !31 + %3839 = fpext bfloat %3318 to float, !dbg !31 + %3840 = fpext bfloat %3319 to float, !dbg !31 + %3841 = fpext bfloat %3320 to float, !dbg !31 + %3842 = fpext bfloat %3330 to float, !dbg !31 + %3843 = fpext bfloat %3331 to float, !dbg !31 + %3844 = fpext bfloat %3332 to float, !dbg !31 + %3845 = fpext bfloat %3333 to float, !dbg !31 + %3846 = fpext bfloat %3334 to float, !dbg !31 + %3847 = fpext bfloat %3335 to float, !dbg !31 + %3848 = fpext bfloat %3336 to float, !dbg !31 + %3849 = fpext bfloat %3337 to float, !dbg !31 + %3850 = shl nuw nsw i32 %63, 4, !dbg !32 + %3851 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %3850, !dbg !32 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x2.b16.p3(ptr addrspace(3) %3851, i32 %2243, i32 %2245), !dbg !32 + tail call void @llvm.nvvm.bar.warp.sync(i32 -1), !dbg !32 + %3852 = shl nuw nsw i32 %13, 3, !dbg !32 + %3853 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %3852, !dbg !32 + %3854 = load float, ptr addrspace(3) %3853, align 16, !dbg !32 + %3855 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 4, !dbg !32 + %3856 = load float, ptr addrspace(3) %3855, align 4, !dbg !32 + %3857 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 8, !dbg !32 + %3858 = load float, ptr addrspace(3) %3857, align 8, !dbg !32 + %3859 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 12, !dbg !32 + %3860 = load float, ptr addrspace(3) %3859, align 4, !dbg !32 + %3861 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 16, !dbg !32 + %3862 = load float, ptr addrspace(3) %3861, align 16, !dbg !32 + %3863 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 20, !dbg !32 + %3864 = load float, ptr addrspace(3) %3863, align 4, !dbg !32 + %3865 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 24, !dbg !32 + %3866 = load float, ptr addrspace(3) %3865, align 8, !dbg !32 + %3867 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 28, !dbg !32 + %3868 = load float, ptr addrspace(3) %3867, align 4, !dbg !32 + %3869 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 32, !dbg !32 + %3870 = load float, ptr addrspace(3) %3869, align 16, !dbg !32 + %3871 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 36, !dbg !32 + %3872 = load float, ptr addrspace(3) %3871, align 4, !dbg !32 + %3873 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 40, !dbg !32 + %3874 = load float, ptr addrspace(3) %3873, align 8, !dbg !32 + %3875 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 44, !dbg !32 + %3876 = load float, ptr addrspace(3) %3875, align 4, !dbg !32 + %3877 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 48, !dbg !32 + %3878 = load float, ptr addrspace(3) %3877, align 16, !dbg !32 + %3879 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 52, !dbg !32 + %3880 = load float, ptr addrspace(3) %3879, align 4, !dbg !32 + %3881 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 56, !dbg !32 + %3882 = load float, ptr addrspace(3) %3881, align 8, !dbg !32 + %3883 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 60, !dbg !32 + %3884 = load float, ptr addrspace(3) %3883, align 4, !dbg !32 + %3885 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 64, !dbg !32 + %3886 = load float, ptr addrspace(3) %3885, align 16, !dbg !32 + %3887 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 68, !dbg !32 + %3888 = load float, ptr addrspace(3) %3887, align 4, !dbg !32 + %3889 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 72, !dbg !32 + %3890 = load float, ptr addrspace(3) %3889, align 8, !dbg !32 + %3891 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 76, !dbg !32 + %3892 = load float, ptr addrspace(3) %3891, align 4, !dbg !32 + %3893 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 80, !dbg !32 + %3894 = load float, ptr addrspace(3) %3893, align 16, !dbg !32 + %3895 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 84, !dbg !32 + %3896 = load float, ptr addrspace(3) %3895, align 4, !dbg !32 + %3897 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 88, !dbg !32 + %3898 = load float, ptr addrspace(3) %3897, align 8, !dbg !32 + %3899 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 92, !dbg !32 + %3900 = load float, ptr addrspace(3) %3899, align 4, !dbg !32 + %3901 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 96, !dbg !32 + %3902 = load float, ptr addrspace(3) %3901, align 16, !dbg !32 + %3903 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 100, !dbg !32 + %3904 = load float, ptr addrspace(3) %3903, align 4, !dbg !32 + %3905 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 104, !dbg !32 + %3906 = load float, ptr addrspace(3) %3905, align 8, !dbg !32 + %3907 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 108, !dbg !32 + %3908 = load float, ptr addrspace(3) %3907, align 4, !dbg !32 + %3909 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 112, !dbg !32 + %3910 = load float, ptr addrspace(3) %3909, align 16, !dbg !32 + %3911 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 116, !dbg !32 + %3912 = load float, ptr addrspace(3) %3911, align 4, !dbg !32 + %3913 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 120, !dbg !32 + %3914 = load float, ptr addrspace(3) %3913, align 8, !dbg !32 + %3915 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 124, !dbg !32 + %3916 = load float, ptr addrspace(3) %3915, align 4, !dbg !32 + tail call void @llvm.nvvm.bar.warp.sync(i32 -1), !dbg !32 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x2.b16.p3(ptr addrspace(3) %3851, i32 %2247, i32 %2249), !dbg !32 + tail call void @llvm.nvvm.bar.warp.sync(i32 -1), !dbg !32 + %3917 = load float, ptr addrspace(3) %3853, align 16, !dbg !32 + %3918 = load float, ptr addrspace(3) %3855, align 4, !dbg !32 + %3919 = load float, ptr addrspace(3) %3857, align 8, !dbg !32 + %3920 = load float, ptr addrspace(3) %3859, align 4, !dbg !32 + %3921 = load float, ptr addrspace(3) %3861, align 16, !dbg !32 + %3922 = load float, ptr addrspace(3) %3863, align 4, !dbg !32 + %3923 = load float, ptr addrspace(3) %3865, align 8, !dbg !32 + %3924 = load float, ptr addrspace(3) %3867, align 4, !dbg !32 + %3925 = load float, ptr addrspace(3) %3869, align 16, !dbg !32 + %3926 = load float, ptr addrspace(3) %3871, align 4, !dbg !32 + %3927 = load float, ptr addrspace(3) %3873, align 8, !dbg !32 + %3928 = load float, ptr addrspace(3) %3875, align 4, !dbg !32 + %3929 = load float, ptr addrspace(3) %3877, align 16, !dbg !32 + %3930 = load float, ptr addrspace(3) %3879, align 4, !dbg !32 + %3931 = load float, ptr addrspace(3) %3881, align 8, !dbg !32 + %3932 = load float, ptr addrspace(3) %3883, align 4, !dbg !32 + %3933 = load float, ptr addrspace(3) %3885, align 16, !dbg !32 + %3934 = load float, ptr addrspace(3) %3887, align 4, !dbg !32 + %3935 = load float, ptr addrspace(3) %3889, align 8, !dbg !32 + %3936 = load float, ptr addrspace(3) %3891, align 4, !dbg !32 + %3937 = load float, ptr addrspace(3) %3893, align 16, !dbg !32 + %3938 = load float, ptr addrspace(3) %3895, align 4, !dbg !32 + %3939 = load float, ptr addrspace(3) %3897, align 8, !dbg !32 + %3940 = load float, ptr addrspace(3) %3899, align 4, !dbg !32 + %3941 = load float, ptr addrspace(3) %3901, align 16, !dbg !32 + %3942 = load float, ptr addrspace(3) %3903, align 4, !dbg !32 + %3943 = load float, ptr addrspace(3) %3905, align 8, !dbg !32 + %3944 = load float, ptr addrspace(3) %3907, align 4, !dbg !32 + %3945 = load float, ptr addrspace(3) %3909, align 16, !dbg !32 + %3946 = load float, ptr addrspace(3) %3911, align 4, !dbg !32 + %3947 = load float, ptr addrspace(3) %3913, align 8, !dbg !32 + %3948 = load float, ptr addrspace(3) %3915, align 4, !dbg !32 + %3949 = fmul float %3854, %3338, !dbg !33 + %3950 = fmul float %3854, %3339, !dbg !33 + %3951 = fmul float %3854, %3340, !dbg !33 + %3952 = fmul float %3854, %3341, !dbg !33 + %3953 = fmul float %3854, %3342, !dbg !33 + %3954 = fmul float %3854, %3343, !dbg !33 + %3955 = fmul float %3854, %3344, !dbg !33 + %3956 = fmul float %3854, %3345, !dbg !33 + %3957 = fmul float %3917, %3346, !dbg !33 + %3958 = fmul float %3917, %3347, !dbg !33 + %3959 = fmul float %3917, %3348, !dbg !33 + %3960 = fmul float %3917, %3349, !dbg !33 + %3961 = fmul float %3917, %3350, !dbg !33 + %3962 = fmul float %3917, %3351, !dbg !33 + %3963 = fmul float %3917, %3352, !dbg !33 + %3964 = fmul float %3917, %3353, !dbg !33 + %3965 = fmul float %3856, %3354, !dbg !33 + %3966 = fmul float %3856, %3355, !dbg !33 + %3967 = fmul float %3856, %3356, !dbg !33 + %3968 = fmul float %3856, %3357, !dbg !33 + %3969 = fmul float %3856, %3358, !dbg !33 + %3970 = fmul float %3856, %3359, !dbg !33 + %3971 = fmul float %3856, %3360, !dbg !33 + %3972 = fmul float %3856, %3361, !dbg !33 + %3973 = fmul float %3918, %3362, !dbg !33 + %3974 = fmul float %3918, %3363, !dbg !33 + %3975 = fmul float %3918, %3364, !dbg !33 + %3976 = fmul float %3918, %3365, !dbg !33 + %3977 = fmul float %3918, %3366, !dbg !33 + %3978 = fmul float %3918, %3367, !dbg !33 + %3979 = fmul float %3918, %3368, !dbg !33 + %3980 = fmul float %3918, %3369, !dbg !33 + %3981 = fmul float %3858, %3370, !dbg !33 + %3982 = fmul float %3858, %3371, !dbg !33 + %3983 = fmul float %3858, %3372, !dbg !33 + %3984 = fmul float %3858, %3373, !dbg !33 + %3985 = fmul float %3858, %3374, !dbg !33 + %3986 = fmul float %3858, %3375, !dbg !33 + %3987 = fmul float %3858, %3376, !dbg !33 + %3988 = fmul float %3858, %3377, !dbg !33 + %3989 = fmul float %3919, %3378, !dbg !33 + %3990 = fmul float %3919, %3379, !dbg !33 + %3991 = fmul float %3919, %3380, !dbg !33 + %3992 = fmul float %3919, %3381, !dbg !33 + %3993 = fmul float %3919, %3382, !dbg !33 + %3994 = fmul float %3919, %3383, !dbg !33 + %3995 = fmul float %3919, %3384, !dbg !33 + %3996 = fmul float %3919, %3385, !dbg !33 + %3997 = fmul float %3860, %3386, !dbg !33 + %3998 = fmul float %3860, %3387, !dbg !33 + %3999 = fmul float %3860, %3388, !dbg !33 + %4000 = fmul float %3860, %3389, !dbg !33 + %4001 = fmul float %3860, %3390, !dbg !33 + %4002 = fmul float %3860, %3391, !dbg !33 + %4003 = fmul float %3860, %3392, !dbg !33 + %4004 = fmul float %3860, %3393, !dbg !33 + %4005 = fmul float %3920, %3394, !dbg !33 + %4006 = fmul float %3920, %3395, !dbg !33 + %4007 = fmul float %3920, %3396, !dbg !33 + %4008 = fmul float %3920, %3397, !dbg !33 + %4009 = fmul float %3920, %3398, !dbg !33 + %4010 = fmul float %3920, %3399, !dbg !33 + %4011 = fmul float %3920, %3400, !dbg !33 + %4012 = fmul float %3920, %3401, !dbg !33 + %4013 = fmul float %3862, %3402, !dbg !33 + %4014 = fmul float %3862, %3403, !dbg !33 + %4015 = fmul float %3862, %3404, !dbg !33 + %4016 = fmul float %3862, %3405, !dbg !33 + %4017 = fmul float %3862, %3406, !dbg !33 + %4018 = fmul float %3862, %3407, !dbg !33 + %4019 = fmul float %3862, %3408, !dbg !33 + %4020 = fmul float %3862, %3409, !dbg !33 + %4021 = fmul float %3921, %3410, !dbg !33 + %4022 = fmul float %3921, %3411, !dbg !33 + %4023 = fmul float %3921, %3412, !dbg !33 + %4024 = fmul float %3921, %3413, !dbg !33 + %4025 = fmul float %3921, %3414, !dbg !33 + %4026 = fmul float %3921, %3415, !dbg !33 + %4027 = fmul float %3921, %3416, !dbg !33 + %4028 = fmul float %3921, %3417, !dbg !33 + %4029 = fmul float %3864, %3418, !dbg !33 + %4030 = fmul float %3864, %3419, !dbg !33 + %4031 = fmul float %3864, %3420, !dbg !33 + %4032 = fmul float %3864, %3421, !dbg !33 + %4033 = fmul float %3864, %3422, !dbg !33 + %4034 = fmul float %3864, %3423, !dbg !33 + %4035 = fmul float %3864, %3424, !dbg !33 + %4036 = fmul float %3864, %3425, !dbg !33 + %4037 = fmul float %3922, %3426, !dbg !33 + %4038 = fmul float %3922, %3427, !dbg !33 + %4039 = fmul float %3922, %3428, !dbg !33 + %4040 = fmul float %3922, %3429, !dbg !33 + %4041 = fmul float %3922, %3430, !dbg !33 + %4042 = fmul float %3922, %3431, !dbg !33 + %4043 = fmul float %3922, %3432, !dbg !33 + %4044 = fmul float %3922, %3433, !dbg !33 + %4045 = fmul float %3866, %3434, !dbg !33 + %4046 = fmul float %3866, %3435, !dbg !33 + %4047 = fmul float %3866, %3436, !dbg !33 + %4048 = fmul float %3866, %3437, !dbg !33 + %4049 = fmul float %3866, %3438, !dbg !33 + %4050 = fmul float %3866, %3439, !dbg !33 + %4051 = fmul float %3866, %3440, !dbg !33 + %4052 = fmul float %3866, %3441, !dbg !33 + %4053 = fmul float %3923, %3442, !dbg !33 + %4054 = fmul float %3923, %3443, !dbg !33 + %4055 = fmul float %3923, %3444, !dbg !33 + %4056 = fmul float %3923, %3445, !dbg !33 + %4057 = fmul float %3923, %3446, !dbg !33 + %4058 = fmul float %3923, %3447, !dbg !33 + %4059 = fmul float %3923, %3448, !dbg !33 + %4060 = fmul float %3923, %3449, !dbg !33 + %4061 = fmul float %3868, %3450, !dbg !33 + %4062 = fmul float %3868, %3451, !dbg !33 + %4063 = fmul float %3868, %3452, !dbg !33 + %4064 = fmul float %3868, %3453, !dbg !33 + %4065 = fmul float %3868, %3454, !dbg !33 + %4066 = fmul float %3868, %3455, !dbg !33 + %4067 = fmul float %3868, %3456, !dbg !33 + %4068 = fmul float %3868, %3457, !dbg !33 + %4069 = fmul float %3924, %3458, !dbg !33 + %4070 = fmul float %3924, %3459, !dbg !33 + %4071 = fmul float %3924, %3460, !dbg !33 + %4072 = fmul float %3924, %3461, !dbg !33 + %4073 = fmul float %3924, %3462, !dbg !33 + %4074 = fmul float %3924, %3463, !dbg !33 + %4075 = fmul float %3924, %3464, !dbg !33 + %4076 = fmul float %3924, %3465, !dbg !33 + %4077 = fmul float %3870, %3466, !dbg !33 + %4078 = fmul float %3870, %3467, !dbg !33 + %4079 = fmul float %3870, %3468, !dbg !33 + %4080 = fmul float %3870, %3469, !dbg !33 + %4081 = fmul float %3870, %3470, !dbg !33 + %4082 = fmul float %3870, %3471, !dbg !33 + %4083 = fmul float %3870, %3472, !dbg !33 + %4084 = fmul float %3870, %3473, !dbg !33 + %4085 = fmul float %3925, %3474, !dbg !33 + %4086 = fmul float %3925, %3475, !dbg !33 + %4087 = fmul float %3925, %3476, !dbg !33 + %4088 = fmul float %3925, %3477, !dbg !33 + %4089 = fmul float %3925, %3478, !dbg !33 + %4090 = fmul float %3925, %3479, !dbg !33 + %4091 = fmul float %3925, %3480, !dbg !33 + %4092 = fmul float %3925, %3481, !dbg !33 + %4093 = fmul float %3872, %3482, !dbg !33 + %4094 = fmul float %3872, %3483, !dbg !33 + %4095 = fmul float %3872, %3484, !dbg !33 + %4096 = fmul float %3872, %3485, !dbg !33 + %4097 = fmul float %3872, %3486, !dbg !33 + %4098 = fmul float %3872, %3487, !dbg !33 + %4099 = fmul float %3872, %3488, !dbg !33 + %4100 = fmul float %3872, %3489, !dbg !33 + %4101 = fmul float %3926, %3490, !dbg !33 + %4102 = fmul float %3926, %3491, !dbg !33 + %4103 = fmul float %3926, %3492, !dbg !33 + %4104 = fmul float %3926, %3493, !dbg !33 + %4105 = fmul float %3926, %3494, !dbg !33 + %4106 = fmul float %3926, %3495, !dbg !33 + %4107 = fmul float %3926, %3496, !dbg !33 + %4108 = fmul float %3926, %3497, !dbg !33 + %4109 = fmul float %3874, %3498, !dbg !33 + %4110 = fmul float %3874, %3499, !dbg !33 + %4111 = fmul float %3874, %3500, !dbg !33 + %4112 = fmul float %3874, %3501, !dbg !33 + %4113 = fmul float %3874, %3502, !dbg !33 + %4114 = fmul float %3874, %3503, !dbg !33 + %4115 = fmul float %3874, %3504, !dbg !33 + %4116 = fmul float %3874, %3505, !dbg !33 + %4117 = fmul float %3927, %3506, !dbg !33 + %4118 = fmul float %3927, %3507, !dbg !33 + %4119 = fmul float %3927, %3508, !dbg !33 + %4120 = fmul float %3927, %3509, !dbg !33 + %4121 = fmul float %3927, %3510, !dbg !33 + %4122 = fmul float %3927, %3511, !dbg !33 + %4123 = fmul float %3927, %3512, !dbg !33 + %4124 = fmul float %3927, %3513, !dbg !33 + %4125 = fmul float %3876, %3514, !dbg !33 + %4126 = fmul float %3876, %3515, !dbg !33 + %4127 = fmul float %3876, %3516, !dbg !33 + %4128 = fmul float %3876, %3517, !dbg !33 + %4129 = fmul float %3876, %3518, !dbg !33 + %4130 = fmul float %3876, %3519, !dbg !33 + %4131 = fmul float %3876, %3520, !dbg !33 + %4132 = fmul float %3876, %3521, !dbg !33 + %4133 = fmul float %3928, %3522, !dbg !33 + %4134 = fmul float %3928, %3523, !dbg !33 + %4135 = fmul float %3928, %3524, !dbg !33 + %4136 = fmul float %3928, %3525, !dbg !33 + %4137 = fmul float %3928, %3526, !dbg !33 + %4138 = fmul float %3928, %3527, !dbg !33 + %4139 = fmul float %3928, %3528, !dbg !33 + %4140 = fmul float %3928, %3529, !dbg !33 + %4141 = fmul float %3878, %3530, !dbg !33 + %4142 = fmul float %3878, %3531, !dbg !33 + %4143 = fmul float %3878, %3532, !dbg !33 + %4144 = fmul float %3878, %3533, !dbg !33 + %4145 = fmul float %3878, %3534, !dbg !33 + %4146 = fmul float %3878, %3535, !dbg !33 + %4147 = fmul float %3878, %3536, !dbg !33 + %4148 = fmul float %3878, %3537, !dbg !33 + %4149 = fmul float %3929, %3538, !dbg !33 + %4150 = fmul float %3929, %3539, !dbg !33 + %4151 = fmul float %3929, %3540, !dbg !33 + %4152 = fmul float %3929, %3541, !dbg !33 + %4153 = fmul float %3929, %3542, !dbg !33 + %4154 = fmul float %3929, %3543, !dbg !33 + %4155 = fmul float %3929, %3544, !dbg !33 + %4156 = fmul float %3929, %3545, !dbg !33 + %4157 = fmul float %3880, %3546, !dbg !33 + %4158 = fmul float %3880, %3547, !dbg !33 + %4159 = fmul float %3880, %3548, !dbg !33 + %4160 = fmul float %3880, %3549, !dbg !33 + %4161 = fmul float %3880, %3550, !dbg !33 + %4162 = fmul float %3880, %3551, !dbg !33 + %4163 = fmul float %3880, %3552, !dbg !33 + %4164 = fmul float %3880, %3553, !dbg !33 + %4165 = fmul float %3930, %3554, !dbg !33 + %4166 = fmul float %3930, %3555, !dbg !33 + %4167 = fmul float %3930, %3556, !dbg !33 + %4168 = fmul float %3930, %3557, !dbg !33 + %4169 = fmul float %3930, %3558, !dbg !33 + %4170 = fmul float %3930, %3559, !dbg !33 + %4171 = fmul float %3930, %3560, !dbg !33 + %4172 = fmul float %3930, %3561, !dbg !33 + %4173 = fmul float %3882, %3562, !dbg !33 + %4174 = fmul float %3882, %3563, !dbg !33 + %4175 = fmul float %3882, %3564, !dbg !33 + %4176 = fmul float %3882, %3565, !dbg !33 + %4177 = fmul float %3882, %3566, !dbg !33 + %4178 = fmul float %3882, %3567, !dbg !33 + %4179 = fmul float %3882, %3568, !dbg !33 + %4180 = fmul float %3882, %3569, !dbg !33 + %4181 = fmul float %3931, %3570, !dbg !33 + %4182 = fmul float %3931, %3571, !dbg !33 + %4183 = fmul float %3931, %3572, !dbg !33 + %4184 = fmul float %3931, %3573, !dbg !33 + %4185 = fmul float %3931, %3574, !dbg !33 + %4186 = fmul float %3931, %3575, !dbg !33 + %4187 = fmul float %3931, %3576, !dbg !33 + %4188 = fmul float %3931, %3577, !dbg !33 + %4189 = fmul float %3884, %3578, !dbg !33 + %4190 = fmul float %3884, %3579, !dbg !33 + %4191 = fmul float %3884, %3580, !dbg !33 + %4192 = fmul float %3884, %3581, !dbg !33 + %4193 = fmul float %3884, %3582, !dbg !33 + %4194 = fmul float %3884, %3583, !dbg !33 + %4195 = fmul float %3884, %3584, !dbg !33 + %4196 = fmul float %3884, %3585, !dbg !33 + %4197 = fmul float %3932, %3586, !dbg !33 + %4198 = fmul float %3932, %3587, !dbg !33 + %4199 = fmul float %3932, %3588, !dbg !33 + %4200 = fmul float %3932, %3589, !dbg !33 + %4201 = fmul float %3932, %3590, !dbg !33 + %4202 = fmul float %3932, %3591, !dbg !33 + %4203 = fmul float %3932, %3592, !dbg !33 + %4204 = fmul float %3932, %3593, !dbg !33 + %4205 = fmul float %3886, %3594, !dbg !33 + %4206 = fmul float %3886, %3595, !dbg !33 + %4207 = fmul float %3886, %3596, !dbg !33 + %4208 = fmul float %3886, %3597, !dbg !33 + %4209 = fmul float %3886, %3598, !dbg !33 + %4210 = fmul float %3886, %3599, !dbg !33 + %4211 = fmul float %3886, %3600, !dbg !33 + %4212 = fmul float %3886, %3601, !dbg !33 + %4213 = fmul float %3933, %3602, !dbg !33 + %4214 = fmul float %3933, %3603, !dbg !33 + %4215 = fmul float %3933, %3604, !dbg !33 + %4216 = fmul float %3933, %3605, !dbg !33 + %4217 = fmul float %3933, %3606, !dbg !33 + %4218 = fmul float %3933, %3607, !dbg !33 + %4219 = fmul float %3933, %3608, !dbg !33 + %4220 = fmul float %3933, %3609, !dbg !33 + %4221 = fmul float %3888, %3610, !dbg !33 + %4222 = fmul float %3888, %3611, !dbg !33 + %4223 = fmul float %3888, %3612, !dbg !33 + %4224 = fmul float %3888, %3613, !dbg !33 + %4225 = fmul float %3888, %3614, !dbg !33 + %4226 = fmul float %3888, %3615, !dbg !33 + %4227 = fmul float %3888, %3616, !dbg !33 + %4228 = fmul float %3888, %3617, !dbg !33 + %4229 = fmul float %3934, %3618, !dbg !33 + %4230 = fmul float %3934, %3619, !dbg !33 + %4231 = fmul float %3934, %3620, !dbg !33 + %4232 = fmul float %3934, %3621, !dbg !33 + %4233 = fmul float %3934, %3622, !dbg !33 + %4234 = fmul float %3934, %3623, !dbg !33 + %4235 = fmul float %3934, %3624, !dbg !33 + %4236 = fmul float %3934, %3625, !dbg !33 + %4237 = fmul float %3890, %3626, !dbg !33 + %4238 = fmul float %3890, %3627, !dbg !33 + %4239 = fmul float %3890, %3628, !dbg !33 + %4240 = fmul float %3890, %3629, !dbg !33 + %4241 = fmul float %3890, %3630, !dbg !33 + %4242 = fmul float %3890, %3631, !dbg !33 + %4243 = fmul float %3890, %3632, !dbg !33 + %4244 = fmul float %3890, %3633, !dbg !33 + %4245 = fmul float %3935, %3634, !dbg !33 + %4246 = fmul float %3935, %3635, !dbg !33 + %4247 = fmul float %3935, %3636, !dbg !33 + %4248 = fmul float %3935, %3637, !dbg !33 + %4249 = fmul float %3935, %3638, !dbg !33 + %4250 = fmul float %3935, %3639, !dbg !33 + %4251 = fmul float %3935, %3640, !dbg !33 + %4252 = fmul float %3935, %3641, !dbg !33 + %4253 = fmul float %3892, %3642, !dbg !33 + %4254 = fmul float %3892, %3643, !dbg !33 + %4255 = fmul float %3892, %3644, !dbg !33 + %4256 = fmul float %3892, %3645, !dbg !33 + %4257 = fmul float %3892, %3646, !dbg !33 + %4258 = fmul float %3892, %3647, !dbg !33 + %4259 = fmul float %3892, %3648, !dbg !33 + %4260 = fmul float %3892, %3649, !dbg !33 + %4261 = fmul float %3936, %3650, !dbg !33 + %4262 = fmul float %3936, %3651, !dbg !33 + %4263 = fmul float %3936, %3652, !dbg !33 + %4264 = fmul float %3936, %3653, !dbg !33 + %4265 = fmul float %3936, %3654, !dbg !33 + %4266 = fmul float %3936, %3655, !dbg !33 + %4267 = fmul float %3936, %3656, !dbg !33 + %4268 = fmul float %3936, %3657, !dbg !33 + %4269 = fmul float %3894, %3658, !dbg !33 + %4270 = fmul float %3894, %3659, !dbg !33 + %4271 = fmul float %3894, %3660, !dbg !33 + %4272 = fmul float %3894, %3661, !dbg !33 + %4273 = fmul float %3894, %3662, !dbg !33 + %4274 = fmul float %3894, %3663, !dbg !33 + %4275 = fmul float %3894, %3664, !dbg !33 + %4276 = fmul float %3894, %3665, !dbg !33 + %4277 = fmul float %3937, %3666, !dbg !33 + %4278 = fmul float %3937, %3667, !dbg !33 + %4279 = fmul float %3937, %3668, !dbg !33 + %4280 = fmul float %3937, %3669, !dbg !33 + %4281 = fmul float %3937, %3670, !dbg !33 + %4282 = fmul float %3937, %3671, !dbg !33 + %4283 = fmul float %3937, %3672, !dbg !33 + %4284 = fmul float %3937, %3673, !dbg !33 + %4285 = fmul float %3896, %3674, !dbg !33 + %4286 = fmul float %3896, %3675, !dbg !33 + %4287 = fmul float %3896, %3676, !dbg !33 + %4288 = fmul float %3896, %3677, !dbg !33 + %4289 = fmul float %3896, %3678, !dbg !33 + %4290 = fmul float %3896, %3679, !dbg !33 + %4291 = fmul float %3896, %3680, !dbg !33 + %4292 = fmul float %3896, %3681, !dbg !33 + %4293 = fmul float %3938, %3682, !dbg !33 + %4294 = fmul float %3938, %3683, !dbg !33 + %4295 = fmul float %3938, %3684, !dbg !33 + %4296 = fmul float %3938, %3685, !dbg !33 + %4297 = fmul float %3938, %3686, !dbg !33 + %4298 = fmul float %3938, %3687, !dbg !33 + %4299 = fmul float %3938, %3688, !dbg !33 + %4300 = fmul float %3938, %3689, !dbg !33 + %4301 = fmul float %3898, %3690, !dbg !33 + %4302 = fmul float %3898, %3691, !dbg !33 + %4303 = fmul float %3898, %3692, !dbg !33 + %4304 = fmul float %3898, %3693, !dbg !33 + %4305 = fmul float %3898, %3694, !dbg !33 + %4306 = fmul float %3898, %3695, !dbg !33 + %4307 = fmul float %3898, %3696, !dbg !33 + %4308 = fmul float %3898, %3697, !dbg !33 + %4309 = fmul float %3939, %3698, !dbg !33 + %4310 = fmul float %3939, %3699, !dbg !33 + %4311 = fmul float %3939, %3700, !dbg !33 + %4312 = fmul float %3939, %3701, !dbg !33 + %4313 = fmul float %3939, %3702, !dbg !33 + %4314 = fmul float %3939, %3703, !dbg !33 + %4315 = fmul float %3939, %3704, !dbg !33 + %4316 = fmul float %3939, %3705, !dbg !33 + %4317 = fmul float %3900, %3706, !dbg !33 + %4318 = fmul float %3900, %3707, !dbg !33 + %4319 = fmul float %3900, %3708, !dbg !33 + %4320 = fmul float %3900, %3709, !dbg !33 + %4321 = fmul float %3900, %3710, !dbg !33 + %4322 = fmul float %3900, %3711, !dbg !33 + %4323 = fmul float %3900, %3712, !dbg !33 + %4324 = fmul float %3900, %3713, !dbg !33 + %4325 = fmul float %3940, %3714, !dbg !33 + %4326 = fmul float %3940, %3715, !dbg !33 + %4327 = fmul float %3940, %3716, !dbg !33 + %4328 = fmul float %3940, %3717, !dbg !33 + %4329 = fmul float %3940, %3718, !dbg !33 + %4330 = fmul float %3940, %3719, !dbg !33 + %4331 = fmul float %3940, %3720, !dbg !33 + %4332 = fmul float %3940, %3721, !dbg !33 + %4333 = fmul float %3902, %3722, !dbg !33 + %4334 = fmul float %3902, %3723, !dbg !33 + %4335 = fmul float %3902, %3724, !dbg !33 + %4336 = fmul float %3902, %3725, !dbg !33 + %4337 = fmul float %3902, %3726, !dbg !33 + %4338 = fmul float %3902, %3727, !dbg !33 + %4339 = fmul float %3902, %3728, !dbg !33 + %4340 = fmul float %3902, %3729, !dbg !33 + %4341 = fmul float %3941, %3730, !dbg !33 + %4342 = fmul float %3941, %3731, !dbg !33 + %4343 = fmul float %3941, %3732, !dbg !33 + %4344 = fmul float %3941, %3733, !dbg !33 + %4345 = fmul float %3941, %3734, !dbg !33 + %4346 = fmul float %3941, %3735, !dbg !33 + %4347 = fmul float %3941, %3736, !dbg !33 + %4348 = fmul float %3941, %3737, !dbg !33 + %4349 = fmul float %3904, %3738, !dbg !33 + %4350 = fmul float %3904, %3739, !dbg !33 + %4351 = fmul float %3904, %3740, !dbg !33 + %4352 = fmul float %3904, %3741, !dbg !33 + %4353 = fmul float %3904, %3742, !dbg !33 + %4354 = fmul float %3904, %3743, !dbg !33 + %4355 = fmul float %3904, %3744, !dbg !33 + %4356 = fmul float %3904, %3745, !dbg !33 + %4357 = fmul float %3942, %3746, !dbg !33 + %4358 = fmul float %3942, %3747, !dbg !33 + %4359 = fmul float %3942, %3748, !dbg !33 + %4360 = fmul float %3942, %3749, !dbg !33 + %4361 = fmul float %3942, %3750, !dbg !33 + %4362 = fmul float %3942, %3751, !dbg !33 + %4363 = fmul float %3942, %3752, !dbg !33 + %4364 = fmul float %3942, %3753, !dbg !33 + %4365 = fmul float %3906, %3754, !dbg !33 + %4366 = fmul float %3906, %3755, !dbg !33 + %4367 = fmul float %3906, %3756, !dbg !33 + %4368 = fmul float %3906, %3757, !dbg !33 + %4369 = fmul float %3906, %3758, !dbg !33 + %4370 = fmul float %3906, %3759, !dbg !33 + %4371 = fmul float %3906, %3760, !dbg !33 + %4372 = fmul float %3906, %3761, !dbg !33 + %4373 = fmul float %3943, %3762, !dbg !33 + %4374 = fmul float %3943, %3763, !dbg !33 + %4375 = fmul float %3943, %3764, !dbg !33 + %4376 = fmul float %3943, %3765, !dbg !33 + %4377 = fmul float %3943, %3766, !dbg !33 + %4378 = fmul float %3943, %3767, !dbg !33 + %4379 = fmul float %3943, %3768, !dbg !33 + %4380 = fmul float %3943, %3769, !dbg !33 + %4381 = fmul float %3908, %3770, !dbg !33 + %4382 = fmul float %3908, %3771, !dbg !33 + %4383 = fmul float %3908, %3772, !dbg !33 + %4384 = fmul float %3908, %3773, !dbg !33 + %4385 = fmul float %3908, %3774, !dbg !33 + %4386 = fmul float %3908, %3775, !dbg !33 + %4387 = fmul float %3908, %3776, !dbg !33 + %4388 = fmul float %3908, %3777, !dbg !33 + %4389 = fmul float %3944, %3778, !dbg !33 + %4390 = fmul float %3944, %3779, !dbg !33 + %4391 = fmul float %3944, %3780, !dbg !33 + %4392 = fmul float %3944, %3781, !dbg !33 + %4393 = fmul float %3944, %3782, !dbg !33 + %4394 = fmul float %3944, %3783, !dbg !33 + %4395 = fmul float %3944, %3784, !dbg !33 + %4396 = fmul float %3944, %3785, !dbg !33 + %4397 = fmul float %3910, %3786, !dbg !33 + %4398 = fmul float %3910, %3787, !dbg !33 + %4399 = fmul float %3910, %3788, !dbg !33 + %4400 = fmul float %3910, %3789, !dbg !33 + %4401 = fmul float %3910, %3790, !dbg !33 + %4402 = fmul float %3910, %3791, !dbg !33 + %4403 = fmul float %3910, %3792, !dbg !33 + %4404 = fmul float %3910, %3793, !dbg !33 + %4405 = fmul float %3945, %3794, !dbg !33 + %4406 = fmul float %3945, %3795, !dbg !33 + %4407 = fmul float %3945, %3796, !dbg !33 + %4408 = fmul float %3945, %3797, !dbg !33 + %4409 = fmul float %3945, %3798, !dbg !33 + %4410 = fmul float %3945, %3799, !dbg !33 + %4411 = fmul float %3945, %3800, !dbg !33 + %4412 = fmul float %3945, %3801, !dbg !33 + %4413 = fmul float %3912, %3802, !dbg !33 + %4414 = fmul float %3912, %3803, !dbg !33 + %4415 = fmul float %3912, %3804, !dbg !33 + %4416 = fmul float %3912, %3805, !dbg !33 + %4417 = fmul float %3912, %3806, !dbg !33 + %4418 = fmul float %3912, %3807, !dbg !33 + %4419 = fmul float %3912, %3808, !dbg !33 + %4420 = fmul float %3912, %3809, !dbg !33 + %4421 = fmul float %3946, %3810, !dbg !33 + %4422 = fmul float %3946, %3811, !dbg !33 + %4423 = fmul float %3946, %3812, !dbg !33 + %4424 = fmul float %3946, %3813, !dbg !33 + %4425 = fmul float %3946, %3814, !dbg !33 + %4426 = fmul float %3946, %3815, !dbg !33 + %4427 = fmul float %3946, %3816, !dbg !33 + %4428 = fmul float %3946, %3817, !dbg !33 + %4429 = fmul float %3914, %3818, !dbg !33 + %4430 = fmul float %3914, %3819, !dbg !33 + %4431 = fmul float %3914, %3820, !dbg !33 + %4432 = fmul float %3914, %3821, !dbg !33 + %4433 = fmul float %3914, %3822, !dbg !33 + %4434 = fmul float %3914, %3823, !dbg !33 + %4435 = fmul float %3914, %3824, !dbg !33 + %4436 = fmul float %3914, %3825, !dbg !33 + %4437 = fmul float %3947, %3826, !dbg !33 + %4438 = fmul float %3947, %3827, !dbg !33 + %4439 = fmul float %3947, %3828, !dbg !33 + %4440 = fmul float %3947, %3829, !dbg !33 + %4441 = fmul float %3947, %3830, !dbg !33 + %4442 = fmul float %3947, %3831, !dbg !33 + %4443 = fmul float %3947, %3832, !dbg !33 + %4444 = fmul float %3947, %3833, !dbg !33 + %4445 = fmul float %3916, %3834, !dbg !33 + %4446 = fmul float %3916, %3835, !dbg !33 + %4447 = fmul float %3916, %3836, !dbg !33 + %4448 = fmul float %3916, %3837, !dbg !33 + %4449 = fmul float %3916, %3838, !dbg !33 + %4450 = fmul float %3916, %3839, !dbg !33 + %4451 = fmul float %3916, %3840, !dbg !33 + %4452 = fmul float %3916, %3841, !dbg !33 + %4453 = fmul float %3948, %3842, !dbg !33 + %4454 = fmul float %3948, %3843, !dbg !33 + %4455 = fmul float %3948, %3844, !dbg !33 + %4456 = fmul float %3948, %3845, !dbg !33 + %4457 = fmul float %3948, %3846, !dbg !33 + %4458 = fmul float %3948, %3847, !dbg !33 + %4459 = fmul float %3948, %3848, !dbg !33 + %4460 = fmul float %3948, %3849, !dbg !33 + %4461 = fmul float %1730, %3338, !dbg !34 + %4462 = fmul float %1731, %3339, !dbg !34 + %4463 = fmul float %1732, %3340, !dbg !34 + %4464 = fmul float %1733, %3341, !dbg !34 + %4465 = fmul float %1734, %3342, !dbg !34 + %4466 = fmul float %1735, %3343, !dbg !34 + %4467 = fmul float %1736, %3344, !dbg !34 + %4468 = fmul float %1737, %3345, !dbg !34 + %4469 = fmul float %1738, %3346, !dbg !34 + %4470 = fmul float %1739, %3347, !dbg !34 + %4471 = fmul float %1740, %3348, !dbg !34 + %4472 = fmul float %1741, %3349, !dbg !34 + %4473 = fmul float %1742, %3350, !dbg !34 + %4474 = fmul float %1743, %3351, !dbg !34 + %4475 = fmul float %1744, %3352, !dbg !34 + %4476 = fmul float %1745, %3353, !dbg !34 + %4477 = fmul float %1746, %3354, !dbg !34 + %4478 = fmul float %1747, %3355, !dbg !34 + %4479 = fmul float %1748, %3356, !dbg !34 + %4480 = fmul float %1749, %3357, !dbg !34 + %4481 = fmul float %1750, %3358, !dbg !34 + %4482 = fmul float %1751, %3359, !dbg !34 + %4483 = fmul float %1752, %3360, !dbg !34 + %4484 = fmul float %1753, %3361, !dbg !34 + %4485 = fmul float %1754, %3362, !dbg !34 + %4486 = fmul float %1755, %3363, !dbg !34 + %4487 = fmul float %1756, %3364, !dbg !34 + %4488 = fmul float %1757, %3365, !dbg !34 + %4489 = fmul float %1758, %3366, !dbg !34 + %4490 = fmul float %1759, %3367, !dbg !34 + %4491 = fmul float %1760, %3368, !dbg !34 + %4492 = fmul float %1761, %3369, !dbg !34 + %4493 = fmul float %1762, %3370, !dbg !34 + %4494 = fmul float %1763, %3371, !dbg !34 + %4495 = fmul float %1764, %3372, !dbg !34 + %4496 = fmul float %1765, %3373, !dbg !34 + %4497 = fmul float %1766, %3374, !dbg !34 + %4498 = fmul float %1767, %3375, !dbg !34 + %4499 = fmul float %1768, %3376, !dbg !34 + %4500 = fmul float %1769, %3377, !dbg !34 + %4501 = fmul float %1770, %3378, !dbg !34 + %4502 = fmul float %1771, %3379, !dbg !34 + %4503 = fmul float %1772, %3380, !dbg !34 + %4504 = fmul float %1773, %3381, !dbg !34 + %4505 = fmul float %1774, %3382, !dbg !34 + %4506 = fmul float %1775, %3383, !dbg !34 + %4507 = fmul float %1776, %3384, !dbg !34 + %4508 = fmul float %1777, %3385, !dbg !34 + %4509 = fmul float %1778, %3386, !dbg !34 + %4510 = fmul float %1779, %3387, !dbg !34 + %4511 = fmul float %1780, %3388, !dbg !34 + %4512 = fmul float %1781, %3389, !dbg !34 + %4513 = fmul float %1782, %3390, !dbg !34 + %4514 = fmul float %1783, %3391, !dbg !34 + %4515 = fmul float %1784, %3392, !dbg !34 + %4516 = fmul float %1785, %3393, !dbg !34 + %4517 = fmul float %1786, %3394, !dbg !34 + %4518 = fmul float %1787, %3395, !dbg !34 + %4519 = fmul float %1788, %3396, !dbg !34 + %4520 = fmul float %1789, %3397, !dbg !34 + %4521 = fmul float %1790, %3398, !dbg !34 + %4522 = fmul float %1791, %3399, !dbg !34 + %4523 = fmul float %1792, %3400, !dbg !34 + %4524 = fmul float %1793, %3401, !dbg !34 + %4525 = fmul float %1794, %3402, !dbg !34 + %4526 = fmul float %1795, %3403, !dbg !34 + %4527 = fmul float %1796, %3404, !dbg !34 + %4528 = fmul float %1797, %3405, !dbg !34 + %4529 = fmul float %1798, %3406, !dbg !34 + %4530 = fmul float %1799, %3407, !dbg !34 + %4531 = fmul float %1800, %3408, !dbg !34 + %4532 = fmul float %1801, %3409, !dbg !34 + %4533 = fmul float %1802, %3410, !dbg !34 + %4534 = fmul float %1803, %3411, !dbg !34 + %4535 = fmul float %1804, %3412, !dbg !34 + %4536 = fmul float %1805, %3413, !dbg !34 + %4537 = fmul float %1806, %3414, !dbg !34 + %4538 = fmul float %1807, %3415, !dbg !34 + %4539 = fmul float %1808, %3416, !dbg !34 + %4540 = fmul float %1809, %3417, !dbg !34 + %4541 = fmul float %1810, %3418, !dbg !34 + %4542 = fmul float %1811, %3419, !dbg !34 + %4543 = fmul float %1812, %3420, !dbg !34 + %4544 = fmul float %1813, %3421, !dbg !34 + %4545 = fmul float %1814, %3422, !dbg !34 + %4546 = fmul float %1815, %3423, !dbg !34 + %4547 = fmul float %1816, %3424, !dbg !34 + %4548 = fmul float %1817, %3425, !dbg !34 + %4549 = fmul float %1818, %3426, !dbg !34 + %4550 = fmul float %1819, %3427, !dbg !34 + %4551 = fmul float %1820, %3428, !dbg !34 + %4552 = fmul float %1821, %3429, !dbg !34 + %4553 = fmul float %1822, %3430, !dbg !34 + %4554 = fmul float %1823, %3431, !dbg !34 + %4555 = fmul float %1824, %3432, !dbg !34 + %4556 = fmul float %1825, %3433, !dbg !34 + %4557 = fmul float %1826, %3434, !dbg !34 + %4558 = fmul float %1827, %3435, !dbg !34 + %4559 = fmul float %1828, %3436, !dbg !34 + %4560 = fmul float %1829, %3437, !dbg !34 + %4561 = fmul float %1830, %3438, !dbg !34 + %4562 = fmul float %1831, %3439, !dbg !34 + %4563 = fmul float %1832, %3440, !dbg !34 + %4564 = fmul float %1833, %3441, !dbg !34 + %4565 = fmul float %1834, %3442, !dbg !34 + %4566 = fmul float %1835, %3443, !dbg !34 + %4567 = fmul float %1836, %3444, !dbg !34 + %4568 = fmul float %1837, %3445, !dbg !34 + %4569 = fmul float %1838, %3446, !dbg !34 + %4570 = fmul float %1839, %3447, !dbg !34 + %4571 = fmul float %1840, %3448, !dbg !34 + %4572 = fmul float %1841, %3449, !dbg !34 + %4573 = fmul float %1842, %3450, !dbg !34 + %4574 = fmul float %1843, %3451, !dbg !34 + %4575 = fmul float %1844, %3452, !dbg !34 + %4576 = fmul float %1845, %3453, !dbg !34 + %4577 = fmul float %1846, %3454, !dbg !34 + %4578 = fmul float %1847, %3455, !dbg !34 + %4579 = fmul float %1848, %3456, !dbg !34 + %4580 = fmul float %1849, %3457, !dbg !34 + %4581 = fmul float %1850, %3458, !dbg !34 + %4582 = fmul float %1851, %3459, !dbg !34 + %4583 = fmul float %1852, %3460, !dbg !34 + %4584 = fmul float %1853, %3461, !dbg !34 + %4585 = fmul float %1854, %3462, !dbg !34 + %4586 = fmul float %1855, %3463, !dbg !34 + %4587 = fmul float %1856, %3464, !dbg !34 + %4588 = fmul float %1857, %3465, !dbg !34 + %4589 = fmul float %1858, %3466, !dbg !34 + %4590 = fmul float %1859, %3467, !dbg !34 + %4591 = fmul float %1860, %3468, !dbg !34 + %4592 = fmul float %1861, %3469, !dbg !34 + %4593 = fmul float %1862, %3470, !dbg !34 + %4594 = fmul float %1863, %3471, !dbg !34 + %4595 = fmul float %1864, %3472, !dbg !34 + %4596 = fmul float %1865, %3473, !dbg !34 + %4597 = fmul float %1866, %3474, !dbg !34 + %4598 = fmul float %1867, %3475, !dbg !34 + %4599 = fmul float %1868, %3476, !dbg !34 + %4600 = fmul float %1869, %3477, !dbg !34 + %4601 = fmul float %1870, %3478, !dbg !34 + %4602 = fmul float %1871, %3479, !dbg !34 + %4603 = fmul float %1872, %3480, !dbg !34 + %4604 = fmul float %1873, %3481, !dbg !34 + %4605 = fmul float %1874, %3482, !dbg !34 + %4606 = fmul float %1875, %3483, !dbg !34 + %4607 = fmul float %1876, %3484, !dbg !34 + %4608 = fmul float %1877, %3485, !dbg !34 + %4609 = fmul float %1878, %3486, !dbg !34 + %4610 = fmul float %1879, %3487, !dbg !34 + %4611 = fmul float %1880, %3488, !dbg !34 + %4612 = fmul float %1881, %3489, !dbg !34 + %4613 = fmul float %1882, %3490, !dbg !34 + %4614 = fmul float %1883, %3491, !dbg !34 + %4615 = fmul float %1884, %3492, !dbg !34 + %4616 = fmul float %1885, %3493, !dbg !34 + %4617 = fmul float %1886, %3494, !dbg !34 + %4618 = fmul float %1887, %3495, !dbg !34 + %4619 = fmul float %1888, %3496, !dbg !34 + %4620 = fmul float %1889, %3497, !dbg !34 + %4621 = fmul float %1890, %3498, !dbg !34 + %4622 = fmul float %1891, %3499, !dbg !34 + %4623 = fmul float %1892, %3500, !dbg !34 + %4624 = fmul float %1893, %3501, !dbg !34 + %4625 = fmul float %1894, %3502, !dbg !34 + %4626 = fmul float %1895, %3503, !dbg !34 + %4627 = fmul float %1896, %3504, !dbg !34 + %4628 = fmul float %1897, %3505, !dbg !34 + %4629 = fmul float %1898, %3506, !dbg !34 + %4630 = fmul float %1899, %3507, !dbg !34 + %4631 = fmul float %1900, %3508, !dbg !34 + %4632 = fmul float %1901, %3509, !dbg !34 + %4633 = fmul float %1902, %3510, !dbg !34 + %4634 = fmul float %1903, %3511, !dbg !34 + %4635 = fmul float %1904, %3512, !dbg !34 + %4636 = fmul float %1905, %3513, !dbg !34 + %4637 = fmul float %1906, %3514, !dbg !34 + %4638 = fmul float %1907, %3515, !dbg !34 + %4639 = fmul float %1908, %3516, !dbg !34 + %4640 = fmul float %1909, %3517, !dbg !34 + %4641 = fmul float %1910, %3518, !dbg !34 + %4642 = fmul float %1911, %3519, !dbg !34 + %4643 = fmul float %1912, %3520, !dbg !34 + %4644 = fmul float %1913, %3521, !dbg !34 + %4645 = fmul float %1914, %3522, !dbg !34 + %4646 = fmul float %1915, %3523, !dbg !34 + %4647 = fmul float %1916, %3524, !dbg !34 + %4648 = fmul float %1917, %3525, !dbg !34 + %4649 = fmul float %1918, %3526, !dbg !34 + %4650 = fmul float %1919, %3527, !dbg !34 + %4651 = fmul float %1920, %3528, !dbg !34 + %4652 = fmul float %1921, %3529, !dbg !34 + %4653 = fmul float %1922, %3530, !dbg !34 + %4654 = fmul float %1923, %3531, !dbg !34 + %4655 = fmul float %1924, %3532, !dbg !34 + %4656 = fmul float %1925, %3533, !dbg !34 + %4657 = fmul float %1926, %3534, !dbg !34 + %4658 = fmul float %1927, %3535, !dbg !34 + %4659 = fmul float %1928, %3536, !dbg !34 + %4660 = fmul float %1929, %3537, !dbg !34 + %4661 = fmul float %1930, %3538, !dbg !34 + %4662 = fmul float %1931, %3539, !dbg !34 + %4663 = fmul float %1932, %3540, !dbg !34 + %4664 = fmul float %1933, %3541, !dbg !34 + %4665 = fmul float %1934, %3542, !dbg !34 + %4666 = fmul float %1935, %3543, !dbg !34 + %4667 = fmul float %1936, %3544, !dbg !34 + %4668 = fmul float %1937, %3545, !dbg !34 + %4669 = fmul float %1938, %3546, !dbg !34 + %4670 = fmul float %1939, %3547, !dbg !34 + %4671 = fmul float %1940, %3548, !dbg !34 + %4672 = fmul float %1941, %3549, !dbg !34 + %4673 = fmul float %1942, %3550, !dbg !34 + %4674 = fmul float %1943, %3551, !dbg !34 + %4675 = fmul float %1944, %3552, !dbg !34 + %4676 = fmul float %1945, %3553, !dbg !34 + %4677 = fmul float %1946, %3554, !dbg !34 + %4678 = fmul float %1947, %3555, !dbg !34 + %4679 = fmul float %1948, %3556, !dbg !34 + %4680 = fmul float %1949, %3557, !dbg !34 + %4681 = fmul float %1950, %3558, !dbg !34 + %4682 = fmul float %1951, %3559, !dbg !34 + %4683 = fmul float %1952, %3560, !dbg !34 + %4684 = fmul float %1953, %3561, !dbg !34 + %4685 = fmul float %1954, %3562, !dbg !34 + %4686 = fmul float %1955, %3563, !dbg !34 + %4687 = fmul float %1956, %3564, !dbg !34 + %4688 = fmul float %1957, %3565, !dbg !34 + %4689 = fmul float %1958, %3566, !dbg !34 + %4690 = fmul float %1959, %3567, !dbg !34 + %4691 = fmul float %1960, %3568, !dbg !34 + %4692 = fmul float %1961, %3569, !dbg !34 + %4693 = fmul float %1962, %3570, !dbg !34 + %4694 = fmul float %1963, %3571, !dbg !34 + %4695 = fmul float %1964, %3572, !dbg !34 + %4696 = fmul float %1965, %3573, !dbg !34 + %4697 = fmul float %1966, %3574, !dbg !34 + %4698 = fmul float %1967, %3575, !dbg !34 + %4699 = fmul float %1968, %3576, !dbg !34 + %4700 = fmul float %1969, %3577, !dbg !34 + %4701 = fmul float %1970, %3578, !dbg !34 + %4702 = fmul float %1971, %3579, !dbg !34 + %4703 = fmul float %1972, %3580, !dbg !34 + %4704 = fmul float %1973, %3581, !dbg !34 + %4705 = fmul float %1974, %3582, !dbg !34 + %4706 = fmul float %1975, %3583, !dbg !34 + %4707 = fmul float %1976, %3584, !dbg !34 + %4708 = fmul float %1977, %3585, !dbg !34 + %4709 = fmul float %1978, %3586, !dbg !34 + %4710 = fmul float %1979, %3587, !dbg !34 + %4711 = fmul float %1980, %3588, !dbg !34 + %4712 = fmul float %1981, %3589, !dbg !34 + %4713 = fmul float %1982, %3590, !dbg !34 + %4714 = fmul float %1983, %3591, !dbg !34 + %4715 = fmul float %1984, %3592, !dbg !34 + %4716 = fmul float %1985, %3593, !dbg !34 + %4717 = fmul float %1986, %3594, !dbg !34 + %4718 = fmul float %1987, %3595, !dbg !34 + %4719 = fmul float %1988, %3596, !dbg !34 + %4720 = fmul float %1989, %3597, !dbg !34 + %4721 = fmul float %1990, %3598, !dbg !34 + %4722 = fmul float %1991, %3599, !dbg !34 + %4723 = fmul float %1992, %3600, !dbg !34 + %4724 = fmul float %1993, %3601, !dbg !34 + %4725 = fmul float %1994, %3602, !dbg !34 + %4726 = fmul float %1995, %3603, !dbg !34 + %4727 = fmul float %1996, %3604, !dbg !34 + %4728 = fmul float %1997, %3605, !dbg !34 + %4729 = fmul float %1998, %3606, !dbg !34 + %4730 = fmul float %1999, %3607, !dbg !34 + %4731 = fmul float %2000, %3608, !dbg !34 + %4732 = fmul float %2001, %3609, !dbg !34 + %4733 = fmul float %2002, %3610, !dbg !34 + %4734 = fmul float %2003, %3611, !dbg !34 + %4735 = fmul float %2004, %3612, !dbg !34 + %4736 = fmul float %2005, %3613, !dbg !34 + %4737 = fmul float %2006, %3614, !dbg !34 + %4738 = fmul float %2007, %3615, !dbg !34 + %4739 = fmul float %2008, %3616, !dbg !34 + %4740 = fmul float %2009, %3617, !dbg !34 + %4741 = fmul float %2010, %3618, !dbg !34 + %4742 = fmul float %2011, %3619, !dbg !34 + %4743 = fmul float %2012, %3620, !dbg !34 + %4744 = fmul float %2013, %3621, !dbg !34 + %4745 = fmul float %2014, %3622, !dbg !34 + %4746 = fmul float %2015, %3623, !dbg !34 + %4747 = fmul float %2016, %3624, !dbg !34 + %4748 = fmul float %2017, %3625, !dbg !34 + %4749 = fmul float %2018, %3626, !dbg !34 + %4750 = fmul float %2019, %3627, !dbg !34 + %4751 = fmul float %2020, %3628, !dbg !34 + %4752 = fmul float %2021, %3629, !dbg !34 + %4753 = fmul float %2022, %3630, !dbg !34 + %4754 = fmul float %2023, %3631, !dbg !34 + %4755 = fmul float %2024, %3632, !dbg !34 + %4756 = fmul float %2025, %3633, !dbg !34 + %4757 = fmul float %2026, %3634, !dbg !34 + %4758 = fmul float %2027, %3635, !dbg !34 + %4759 = fmul float %2028, %3636, !dbg !34 + %4760 = fmul float %2029, %3637, !dbg !34 + %4761 = fmul float %2030, %3638, !dbg !34 + %4762 = fmul float %2031, %3639, !dbg !34 + %4763 = fmul float %2032, %3640, !dbg !34 + %4764 = fmul float %2033, %3641, !dbg !34 + %4765 = fmul float %2034, %3642, !dbg !34 + %4766 = fmul float %2035, %3643, !dbg !34 + %4767 = fmul float %2036, %3644, !dbg !34 + %4768 = fmul float %2037, %3645, !dbg !34 + %4769 = fmul float %2038, %3646, !dbg !34 + %4770 = fmul float %2039, %3647, !dbg !34 + %4771 = fmul float %2040, %3648, !dbg !34 + %4772 = fmul float %2041, %3649, !dbg !34 + %4773 = fmul float %2042, %3650, !dbg !34 + %4774 = fmul float %2043, %3651, !dbg !34 + %4775 = fmul float %2044, %3652, !dbg !34 + %4776 = fmul float %2045, %3653, !dbg !34 + %4777 = fmul float %2046, %3654, !dbg !34 + %4778 = fmul float %2047, %3655, !dbg !34 + %4779 = fmul float %2048, %3656, !dbg !34 + %4780 = fmul float %2049, %3657, !dbg !34 + %4781 = fmul float %2050, %3658, !dbg !34 + %4782 = fmul float %2051, %3659, !dbg !34 + %4783 = fmul float %2052, %3660, !dbg !34 + %4784 = fmul float %2053, %3661, !dbg !34 + %4785 = fmul float %2054, %3662, !dbg !34 + %4786 = fmul float %2055, %3663, !dbg !34 + %4787 = fmul float %2056, %3664, !dbg !34 + %4788 = fmul float %2057, %3665, !dbg !34 + %4789 = fmul float %2058, %3666, !dbg !34 + %4790 = fmul float %2059, %3667, !dbg !34 + %4791 = fmul float %2060, %3668, !dbg !34 + %4792 = fmul float %2061, %3669, !dbg !34 + %4793 = fmul float %2062, %3670, !dbg !34 + %4794 = fmul float %2063, %3671, !dbg !34 + %4795 = fmul float %2064, %3672, !dbg !34 + %4796 = fmul float %2065, %3673, !dbg !34 + %4797 = fmul float %2066, %3674, !dbg !34 + %4798 = fmul float %2067, %3675, !dbg !34 + %4799 = fmul float %2068, %3676, !dbg !34 + %4800 = fmul float %2069, %3677, !dbg !34 + %4801 = fmul float %2070, %3678, !dbg !34 + %4802 = fmul float %2071, %3679, !dbg !34 + %4803 = fmul float %2072, %3680, !dbg !34 + %4804 = fmul float %2073, %3681, !dbg !34 + %4805 = fmul float %2074, %3682, !dbg !34 + %4806 = fmul float %2075, %3683, !dbg !34 + %4807 = fmul float %2076, %3684, !dbg !34 + %4808 = fmul float %2077, %3685, !dbg !34 + %4809 = fmul float %2078, %3686, !dbg !34 + %4810 = fmul float %2079, %3687, !dbg !34 + %4811 = fmul float %2080, %3688, !dbg !34 + %4812 = fmul float %2081, %3689, !dbg !34 + %4813 = fmul float %2082, %3690, !dbg !34 + %4814 = fmul float %2083, %3691, !dbg !34 + %4815 = fmul float %2084, %3692, !dbg !34 + %4816 = fmul float %2085, %3693, !dbg !34 + %4817 = fmul float %2086, %3694, !dbg !34 + %4818 = fmul float %2087, %3695, !dbg !34 + %4819 = fmul float %2088, %3696, !dbg !34 + %4820 = fmul float %2089, %3697, !dbg !34 + %4821 = fmul float %2090, %3698, !dbg !34 + %4822 = fmul float %2091, %3699, !dbg !34 + %4823 = fmul float %2092, %3700, !dbg !34 + %4824 = fmul float %2093, %3701, !dbg !34 + %4825 = fmul float %2094, %3702, !dbg !34 + %4826 = fmul float %2095, %3703, !dbg !34 + %4827 = fmul float %2096, %3704, !dbg !34 + %4828 = fmul float %2097, %3705, !dbg !34 + %4829 = fmul float %2098, %3706, !dbg !34 + %4830 = fmul float %2099, %3707, !dbg !34 + %4831 = fmul float %2100, %3708, !dbg !34 + %4832 = fmul float %2101, %3709, !dbg !34 + %4833 = fmul float %2102, %3710, !dbg !34 + %4834 = fmul float %2103, %3711, !dbg !34 + %4835 = fmul float %2104, %3712, !dbg !34 + %4836 = fmul float %2105, %3713, !dbg !34 + %4837 = fmul float %2106, %3714, !dbg !34 + %4838 = fmul float %2107, %3715, !dbg !34 + %4839 = fmul float %2108, %3716, !dbg !34 + %4840 = fmul float %2109, %3717, !dbg !34 + %4841 = fmul float %2110, %3718, !dbg !34 + %4842 = fmul float %2111, %3719, !dbg !34 + %4843 = fmul float %2112, %3720, !dbg !34 + %4844 = fmul float %2113, %3721, !dbg !34 + %4845 = fmul float %2114, %3722, !dbg !34 + %4846 = fmul float %2115, %3723, !dbg !34 + %4847 = fmul float %2116, %3724, !dbg !34 + %4848 = fmul float %2117, %3725, !dbg !34 + %4849 = fmul float %2118, %3726, !dbg !34 + %4850 = fmul float %2119, %3727, !dbg !34 + %4851 = fmul float %2120, %3728, !dbg !34 + %4852 = fmul float %2121, %3729, !dbg !34 + %4853 = fmul float %2122, %3730, !dbg !34 + %4854 = fmul float %2123, %3731, !dbg !34 + %4855 = fmul float %2124, %3732, !dbg !34 + %4856 = fmul float %2125, %3733, !dbg !34 + %4857 = fmul float %2126, %3734, !dbg !34 + %4858 = fmul float %2127, %3735, !dbg !34 + %4859 = fmul float %2128, %3736, !dbg !34 + %4860 = fmul float %2129, %3737, !dbg !34 + %4861 = fmul float %2130, %3738, !dbg !34 + %4862 = fmul float %2131, %3739, !dbg !34 + %4863 = fmul float %2132, %3740, !dbg !34 + %4864 = fmul float %2133, %3741, !dbg !34 + %4865 = fmul float %2134, %3742, !dbg !34 + %4866 = fmul float %2135, %3743, !dbg !34 + %4867 = fmul float %2136, %3744, !dbg !34 + %4868 = fmul float %2137, %3745, !dbg !34 + %4869 = fmul float %2138, %3746, !dbg !34 + %4870 = fmul float %2139, %3747, !dbg !34 + %4871 = fmul float %2140, %3748, !dbg !34 + %4872 = fmul float %2141, %3749, !dbg !34 + %4873 = fmul float %2142, %3750, !dbg !34 + %4874 = fmul float %2143, %3751, !dbg !34 + %4875 = fmul float %2144, %3752, !dbg !34 + %4876 = fmul float %2145, %3753, !dbg !34 + %4877 = fmul float %2146, %3754, !dbg !34 + %4878 = fmul float %2147, %3755, !dbg !34 + %4879 = fmul float %2148, %3756, !dbg !34 + %4880 = fmul float %2149, %3757, !dbg !34 + %4881 = fmul float %2150, %3758, !dbg !34 + %4882 = fmul float %2151, %3759, !dbg !34 + %4883 = fmul float %2152, %3760, !dbg !34 + %4884 = fmul float %2153, %3761, !dbg !34 + %4885 = fmul float %2154, %3762, !dbg !34 + %4886 = fmul float %2155, %3763, !dbg !34 + %4887 = fmul float %2156, %3764, !dbg !34 + %4888 = fmul float %2157, %3765, !dbg !34 + %4889 = fmul float %2158, %3766, !dbg !34 + %4890 = fmul float %2159, %3767, !dbg !34 + %4891 = fmul float %2160, %3768, !dbg !34 + %4892 = fmul float %2161, %3769, !dbg !34 + %4893 = fmul float %2162, %3770, !dbg !34 + %4894 = fmul float %2163, %3771, !dbg !34 + %4895 = fmul float %2164, %3772, !dbg !34 + %4896 = fmul float %2165, %3773, !dbg !34 + %4897 = fmul float %2166, %3774, !dbg !34 + %4898 = fmul float %2167, %3775, !dbg !34 + %4899 = fmul float %2168, %3776, !dbg !34 + %4900 = fmul float %2169, %3777, !dbg !34 + %4901 = fmul float %2170, %3778, !dbg !34 + %4902 = fmul float %2171, %3779, !dbg !34 + %4903 = fmul float %2172, %3780, !dbg !34 + %4904 = fmul float %2173, %3781, !dbg !34 + %4905 = fmul float %2174, %3782, !dbg !34 + %4906 = fmul float %2175, %3783, !dbg !34 + %4907 = fmul float %2176, %3784, !dbg !34 + %4908 = fmul float %2177, %3785, !dbg !34 + %4909 = fmul float %2178, %3786, !dbg !34 + %4910 = fmul float %2179, %3787, !dbg !34 + %4911 = fmul float %2180, %3788, !dbg !34 + %4912 = fmul float %2181, %3789, !dbg !34 + %4913 = fmul float %2182, %3790, !dbg !34 + %4914 = fmul float %2183, %3791, !dbg !34 + %4915 = fmul float %2184, %3792, !dbg !34 + %4916 = fmul float %2185, %3793, !dbg !34 + %4917 = fmul float %2186, %3794, !dbg !34 + %4918 = fmul float %2187, %3795, !dbg !34 + %4919 = fmul float %2188, %3796, !dbg !34 + %4920 = fmul float %2189, %3797, !dbg !34 + %4921 = fmul float %2190, %3798, !dbg !34 + %4922 = fmul float %2191, %3799, !dbg !34 + %4923 = fmul float %2192, %3800, !dbg !34 + %4924 = fmul float %2193, %3801, !dbg !34 + %4925 = fmul float %2194, %3802, !dbg !34 + %4926 = fmul float %2195, %3803, !dbg !34 + %4927 = fmul float %2196, %3804, !dbg !34 + %4928 = fmul float %2197, %3805, !dbg !34 + %4929 = fmul float %2198, %3806, !dbg !34 + %4930 = fmul float %2199, %3807, !dbg !34 + %4931 = fmul float %2200, %3808, !dbg !34 + %4932 = fmul float %2201, %3809, !dbg !34 + %4933 = fmul float %2202, %3810, !dbg !34 + %4934 = fmul float %2203, %3811, !dbg !34 + %4935 = fmul float %2204, %3812, !dbg !34 + %4936 = fmul float %2205, %3813, !dbg !34 + %4937 = fmul float %2206, %3814, !dbg !34 + %4938 = fmul float %2207, %3815, !dbg !34 + %4939 = fmul float %2208, %3816, !dbg !34 + %4940 = fmul float %2209, %3817, !dbg !34 + %4941 = fmul float %2210, %3818, !dbg !34 + %4942 = fmul float %2211, %3819, !dbg !34 + %4943 = fmul float %2212, %3820, !dbg !34 + %4944 = fmul float %2213, %3821, !dbg !34 + %4945 = fmul float %2214, %3822, !dbg !34 + %4946 = fmul float %2215, %3823, !dbg !34 + %4947 = fmul float %2216, %3824, !dbg !34 + %4948 = fmul float %2217, %3825, !dbg !34 + %4949 = fmul float %2218, %3826, !dbg !34 + %4950 = fmul float %2219, %3827, !dbg !34 + %4951 = fmul float %2220, %3828, !dbg !34 + %4952 = fmul float %2221, %3829, !dbg !34 + %4953 = fmul float %2222, %3830, !dbg !34 + %4954 = fmul float %2223, %3831, !dbg !34 + %4955 = fmul float %2224, %3832, !dbg !34 + %4956 = fmul float %2225, %3833, !dbg !34 + %4957 = fmul float %2226, %3834, !dbg !34 + %4958 = fmul float %2227, %3835, !dbg !34 + %4959 = fmul float %2228, %3836, !dbg !34 + %4960 = fmul float %2229, %3837, !dbg !34 + %4961 = fmul float %2230, %3838, !dbg !34 + %4962 = fmul float %2231, %3839, !dbg !34 + %4963 = fmul float %2232, %3840, !dbg !34 + %4964 = fmul float %2233, %3841, !dbg !34 + %4965 = fmul float %2234, %3842, !dbg !34 + %4966 = fmul float %2235, %3843, !dbg !34 + %4967 = fmul float %2236, %3844, !dbg !34 + %4968 = fmul float %2237, %3845, !dbg !34 + %4969 = fmul float %2238, %3846, !dbg !34 + %4970 = fmul float %2239, %3847, !dbg !34 + %4971 = fmul float %2240, %3848, !dbg !34 + %4972 = fmul float %2241, %3849, !dbg !34 + %4973 = fadd float %4461, %4462, !dbg !35 + %4974 = fadd float %4463, %4973, !dbg !35 + %4975 = fadd float %4464, %4974, !dbg !35 + %4976 = fadd float %4465, %4975, !dbg !35 + %4977 = fadd float %4466, %4976, !dbg !35 + %4978 = fadd float %4467, %4977, !dbg !35 + %4979 = fadd float %4468, %4978, !dbg !35 + %4980 = fadd float %4469, %4470, !dbg !35 + %4981 = fadd float %4471, %4980, !dbg !35 + %4982 = fadd float %4472, %4981, !dbg !35 + %4983 = fadd float %4473, %4982, !dbg !35 + %4984 = fadd float %4474, %4983, !dbg !35 + %4985 = fadd float %4475, %4984, !dbg !35 + %4986 = fadd float %4476, %4985, !dbg !35 + %4987 = fadd float %4477, %4478, !dbg !35 + %4988 = fadd float %4479, %4987, !dbg !35 + %4989 = fadd float %4480, %4988, !dbg !35 + %4990 = fadd float %4481, %4989, !dbg !35 + %4991 = fadd float %4482, %4990, !dbg !35 + %4992 = fadd float %4483, %4991, !dbg !35 + %4993 = fadd float %4484, %4992, !dbg !35 + %4994 = fadd float %4485, %4486, !dbg !35 + %4995 = fadd float %4487, %4994, !dbg !35 + %4996 = fadd float %4488, %4995, !dbg !35 + %4997 = fadd float %4489, %4996, !dbg !35 + %4998 = fadd float %4490, %4997, !dbg !35 + %4999 = fadd float %4491, %4998, !dbg !35 + %5000 = fadd float %4492, %4999, !dbg !35 + %5001 = fadd float %4493, %4494, !dbg !35 + %5002 = fadd float %4495, %5001, !dbg !35 + %5003 = fadd float %4496, %5002, !dbg !35 + %5004 = fadd float %4497, %5003, !dbg !35 + %5005 = fadd float %4498, %5004, !dbg !35 + %5006 = fadd float %4499, %5005, !dbg !35 + %5007 = fadd float %4500, %5006, !dbg !35 + %5008 = fadd float %4501, %4502, !dbg !35 + %5009 = fadd float %4503, %5008, !dbg !35 + %5010 = fadd float %4504, %5009, !dbg !35 + %5011 = fadd float %4505, %5010, !dbg !35 + %5012 = fadd float %4506, %5011, !dbg !35 + %5013 = fadd float %4507, %5012, !dbg !35 + %5014 = fadd float %4508, %5013, !dbg !35 + %5015 = fadd float %4509, %4510, !dbg !35 + %5016 = fadd float %4511, %5015, !dbg !35 + %5017 = fadd float %4512, %5016, !dbg !35 + %5018 = fadd float %4513, %5017, !dbg !35 + %5019 = fadd float %4514, %5018, !dbg !35 + %5020 = fadd float %4515, %5019, !dbg !35 + %5021 = fadd float %4516, %5020, !dbg !35 + %5022 = fadd float %4517, %4518, !dbg !35 + %5023 = fadd float %4519, %5022, !dbg !35 + %5024 = fadd float %4520, %5023, !dbg !35 + %5025 = fadd float %4521, %5024, !dbg !35 + %5026 = fadd float %4522, %5025, !dbg !35 + %5027 = fadd float %4523, %5026, !dbg !35 + %5028 = fadd float %4524, %5027, !dbg !35 + %5029 = fadd float %4525, %4526, !dbg !35 + %5030 = fadd float %4527, %5029, !dbg !35 + %5031 = fadd float %4528, %5030, !dbg !35 + %5032 = fadd float %4529, %5031, !dbg !35 + %5033 = fadd float %4530, %5032, !dbg !35 + %5034 = fadd float %4531, %5033, !dbg !35 + %5035 = fadd float %4532, %5034, !dbg !35 + %5036 = fadd float %4533, %4534, !dbg !35 + %5037 = fadd float %4535, %5036, !dbg !35 + %5038 = fadd float %4536, %5037, !dbg !35 + %5039 = fadd float %4537, %5038, !dbg !35 + %5040 = fadd float %4538, %5039, !dbg !35 + %5041 = fadd float %4539, %5040, !dbg !35 + %5042 = fadd float %4540, %5041, !dbg !35 + %5043 = fadd float %4541, %4542, !dbg !35 + %5044 = fadd float %4543, %5043, !dbg !35 + %5045 = fadd float %4544, %5044, !dbg !35 + %5046 = fadd float %4545, %5045, !dbg !35 + %5047 = fadd float %4546, %5046, !dbg !35 + %5048 = fadd float %4547, %5047, !dbg !35 + %5049 = fadd float %4548, %5048, !dbg !35 + %5050 = fadd float %4549, %4550, !dbg !35 + %5051 = fadd float %4551, %5050, !dbg !35 + %5052 = fadd float %4552, %5051, !dbg !35 + %5053 = fadd float %4553, %5052, !dbg !35 + %5054 = fadd float %4554, %5053, !dbg !35 + %5055 = fadd float %4555, %5054, !dbg !35 + %5056 = fadd float %4556, %5055, !dbg !35 + %5057 = fadd float %4557, %4558, !dbg !35 + %5058 = fadd float %4559, %5057, !dbg !35 + %5059 = fadd float %4560, %5058, !dbg !35 + %5060 = fadd float %4561, %5059, !dbg !35 + %5061 = fadd float %4562, %5060, !dbg !35 + %5062 = fadd float %4563, %5061, !dbg !35 + %5063 = fadd float %4564, %5062, !dbg !35 + %5064 = fadd float %4565, %4566, !dbg !35 + %5065 = fadd float %4567, %5064, !dbg !35 + %5066 = fadd float %4568, %5065, !dbg !35 + %5067 = fadd float %4569, %5066, !dbg !35 + %5068 = fadd float %4570, %5067, !dbg !35 + %5069 = fadd float %4571, %5068, !dbg !35 + %5070 = fadd float %4572, %5069, !dbg !35 + %5071 = fadd float %4573, %4574, !dbg !35 + %5072 = fadd float %4575, %5071, !dbg !35 + %5073 = fadd float %4576, %5072, !dbg !35 + %5074 = fadd float %4577, %5073, !dbg !35 + %5075 = fadd float %4578, %5074, !dbg !35 + %5076 = fadd float %4579, %5075, !dbg !35 + %5077 = fadd float %4580, %5076, !dbg !35 + %5078 = fadd float %4581, %4582, !dbg !35 + %5079 = fadd float %4583, %5078, !dbg !35 + %5080 = fadd float %4584, %5079, !dbg !35 + %5081 = fadd float %4585, %5080, !dbg !35 + %5082 = fadd float %4586, %5081, !dbg !35 + %5083 = fadd float %4587, %5082, !dbg !35 + %5084 = fadd float %4588, %5083, !dbg !35 + %5085 = fadd float %4589, %4590, !dbg !35 + %5086 = fadd float %4591, %5085, !dbg !35 + %5087 = fadd float %4592, %5086, !dbg !35 + %5088 = fadd float %4593, %5087, !dbg !35 + %5089 = fadd float %4594, %5088, !dbg !35 + %5090 = fadd float %4595, %5089, !dbg !35 + %5091 = fadd float %4596, %5090, !dbg !35 + %5092 = fadd float %4597, %4598, !dbg !35 + %5093 = fadd float %4599, %5092, !dbg !35 + %5094 = fadd float %4600, %5093, !dbg !35 + %5095 = fadd float %4601, %5094, !dbg !35 + %5096 = fadd float %4602, %5095, !dbg !35 + %5097 = fadd float %4603, %5096, !dbg !35 + %5098 = fadd float %4604, %5097, !dbg !35 + %5099 = fadd float %4605, %4606, !dbg !35 + %5100 = fadd float %4607, %5099, !dbg !35 + %5101 = fadd float %4608, %5100, !dbg !35 + %5102 = fadd float %4609, %5101, !dbg !35 + %5103 = fadd float %4610, %5102, !dbg !35 + %5104 = fadd float %4611, %5103, !dbg !35 + %5105 = fadd float %4612, %5104, !dbg !35 + %5106 = fadd float %4613, %4614, !dbg !35 + %5107 = fadd float %4615, %5106, !dbg !35 + %5108 = fadd float %4616, %5107, !dbg !35 + %5109 = fadd float %4617, %5108, !dbg !35 + %5110 = fadd float %4618, %5109, !dbg !35 + %5111 = fadd float %4619, %5110, !dbg !35 + %5112 = fadd float %4620, %5111, !dbg !35 + %5113 = fadd float %4621, %4622, !dbg !35 + %5114 = fadd float %4623, %5113, !dbg !35 + %5115 = fadd float %4624, %5114, !dbg !35 + %5116 = fadd float %4625, %5115, !dbg !35 + %5117 = fadd float %4626, %5116, !dbg !35 + %5118 = fadd float %4627, %5117, !dbg !35 + %5119 = fadd float %4628, %5118, !dbg !35 + %5120 = fadd float %4629, %4630, !dbg !35 + %5121 = fadd float %4631, %5120, !dbg !35 + %5122 = fadd float %4632, %5121, !dbg !35 + %5123 = fadd float %4633, %5122, !dbg !35 + %5124 = fadd float %4634, %5123, !dbg !35 + %5125 = fadd float %4635, %5124, !dbg !35 + %5126 = fadd float %4636, %5125, !dbg !35 + %5127 = fadd float %4637, %4638, !dbg !35 + %5128 = fadd float %4639, %5127, !dbg !35 + %5129 = fadd float %4640, %5128, !dbg !35 + %5130 = fadd float %4641, %5129, !dbg !35 + %5131 = fadd float %4642, %5130, !dbg !35 + %5132 = fadd float %4643, %5131, !dbg !35 + %5133 = fadd float %4644, %5132, !dbg !35 + %5134 = fadd float %4645, %4646, !dbg !35 + %5135 = fadd float %4647, %5134, !dbg !35 + %5136 = fadd float %4648, %5135, !dbg !35 + %5137 = fadd float %4649, %5136, !dbg !35 + %5138 = fadd float %4650, %5137, !dbg !35 + %5139 = fadd float %4651, %5138, !dbg !35 + %5140 = fadd float %4652, %5139, !dbg !35 + %5141 = fadd float %4653, %4654, !dbg !35 + %5142 = fadd float %4655, %5141, !dbg !35 + %5143 = fadd float %4656, %5142, !dbg !35 + %5144 = fadd float %4657, %5143, !dbg !35 + %5145 = fadd float %4658, %5144, !dbg !35 + %5146 = fadd float %4659, %5145, !dbg !35 + %5147 = fadd float %4660, %5146, !dbg !35 + %5148 = fadd float %4661, %4662, !dbg !35 + %5149 = fadd float %4663, %5148, !dbg !35 + %5150 = fadd float %4664, %5149, !dbg !35 + %5151 = fadd float %4665, %5150, !dbg !35 + %5152 = fadd float %4666, %5151, !dbg !35 + %5153 = fadd float %4667, %5152, !dbg !35 + %5154 = fadd float %4668, %5153, !dbg !35 + %5155 = fadd float %4669, %4670, !dbg !35 + %5156 = fadd float %4671, %5155, !dbg !35 + %5157 = fadd float %4672, %5156, !dbg !35 + %5158 = fadd float %4673, %5157, !dbg !35 + %5159 = fadd float %4674, %5158, !dbg !35 + %5160 = fadd float %4675, %5159, !dbg !35 + %5161 = fadd float %4676, %5160, !dbg !35 + %5162 = fadd float %4677, %4678, !dbg !35 + %5163 = fadd float %4679, %5162, !dbg !35 + %5164 = fadd float %4680, %5163, !dbg !35 + %5165 = fadd float %4681, %5164, !dbg !35 + %5166 = fadd float %4682, %5165, !dbg !35 + %5167 = fadd float %4683, %5166, !dbg !35 + %5168 = fadd float %4684, %5167, !dbg !35 + %5169 = fadd float %4685, %4686, !dbg !35 + %5170 = fadd float %4687, %5169, !dbg !35 + %5171 = fadd float %4688, %5170, !dbg !35 + %5172 = fadd float %4689, %5171, !dbg !35 + %5173 = fadd float %4690, %5172, !dbg !35 + %5174 = fadd float %4691, %5173, !dbg !35 + %5175 = fadd float %4692, %5174, !dbg !35 + %5176 = fadd float %4693, %4694, !dbg !35 + %5177 = fadd float %4695, %5176, !dbg !35 + %5178 = fadd float %4696, %5177, !dbg !35 + %5179 = fadd float %4697, %5178, !dbg !35 + %5180 = fadd float %4698, %5179, !dbg !35 + %5181 = fadd float %4699, %5180, !dbg !35 + %5182 = fadd float %4700, %5181, !dbg !35 + %5183 = fadd float %4701, %4702, !dbg !35 + %5184 = fadd float %4703, %5183, !dbg !35 + %5185 = fadd float %4704, %5184, !dbg !35 + %5186 = fadd float %4705, %5185, !dbg !35 + %5187 = fadd float %4706, %5186, !dbg !35 + %5188 = fadd float %4707, %5187, !dbg !35 + %5189 = fadd float %4708, %5188, !dbg !35 + %5190 = fadd float %4709, %4710, !dbg !35 + %5191 = fadd float %4711, %5190, !dbg !35 + %5192 = fadd float %4712, %5191, !dbg !35 + %5193 = fadd float %4713, %5192, !dbg !35 + %5194 = fadd float %4714, %5193, !dbg !35 + %5195 = fadd float %4715, %5194, !dbg !35 + %5196 = fadd float %4716, %5195, !dbg !35 + %5197 = fadd float %4717, %4718, !dbg !35 + %5198 = fadd float %4719, %5197, !dbg !35 + %5199 = fadd float %4720, %5198, !dbg !35 + %5200 = fadd float %4721, %5199, !dbg !35 + %5201 = fadd float %4722, %5200, !dbg !35 + %5202 = fadd float %4723, %5201, !dbg !35 + %5203 = fadd float %4724, %5202, !dbg !35 + %5204 = fadd float %4725, %4726, !dbg !35 + %5205 = fadd float %4727, %5204, !dbg !35 + %5206 = fadd float %4728, %5205, !dbg !35 + %5207 = fadd float %4729, %5206, !dbg !35 + %5208 = fadd float %4730, %5207, !dbg !35 + %5209 = fadd float %4731, %5208, !dbg !35 + %5210 = fadd float %4732, %5209, !dbg !35 + %5211 = fadd float %4733, %4734, !dbg !35 + %5212 = fadd float %4735, %5211, !dbg !35 + %5213 = fadd float %4736, %5212, !dbg !35 + %5214 = fadd float %4737, %5213, !dbg !35 + %5215 = fadd float %4738, %5214, !dbg !35 + %5216 = fadd float %4739, %5215, !dbg !35 + %5217 = fadd float %4740, %5216, !dbg !35 + %5218 = fadd float %4741, %4742, !dbg !35 + %5219 = fadd float %4743, %5218, !dbg !35 + %5220 = fadd float %4744, %5219, !dbg !35 + %5221 = fadd float %4745, %5220, !dbg !35 + %5222 = fadd float %4746, %5221, !dbg !35 + %5223 = fadd float %4747, %5222, !dbg !35 + %5224 = fadd float %4748, %5223, !dbg !35 + %5225 = fadd float %4749, %4750, !dbg !35 + %5226 = fadd float %4751, %5225, !dbg !35 + %5227 = fadd float %4752, %5226, !dbg !35 + %5228 = fadd float %4753, %5227, !dbg !35 + %5229 = fadd float %4754, %5228, !dbg !35 + %5230 = fadd float %4755, %5229, !dbg !35 + %5231 = fadd float %4756, %5230, !dbg !35 + %5232 = fadd float %4757, %4758, !dbg !35 + %5233 = fadd float %4759, %5232, !dbg !35 + %5234 = fadd float %4760, %5233, !dbg !35 + %5235 = fadd float %4761, %5234, !dbg !35 + %5236 = fadd float %4762, %5235, !dbg !35 + %5237 = fadd float %4763, %5236, !dbg !35 + %5238 = fadd float %4764, %5237, !dbg !35 + %5239 = fadd float %4765, %4766, !dbg !35 + %5240 = fadd float %4767, %5239, !dbg !35 + %5241 = fadd float %4768, %5240, !dbg !35 + %5242 = fadd float %4769, %5241, !dbg !35 + %5243 = fadd float %4770, %5242, !dbg !35 + %5244 = fadd float %4771, %5243, !dbg !35 + %5245 = fadd float %4772, %5244, !dbg !35 + %5246 = fadd float %4773, %4774, !dbg !35 + %5247 = fadd float %4775, %5246, !dbg !35 + %5248 = fadd float %4776, %5247, !dbg !35 + %5249 = fadd float %4777, %5248, !dbg !35 + %5250 = fadd float %4778, %5249, !dbg !35 + %5251 = fadd float %4779, %5250, !dbg !35 + %5252 = fadd float %4780, %5251, !dbg !35 + %5253 = fadd float %4781, %4782, !dbg !35 + %5254 = fadd float %4783, %5253, !dbg !35 + %5255 = fadd float %4784, %5254, !dbg !35 + %5256 = fadd float %4785, %5255, !dbg !35 + %5257 = fadd float %4786, %5256, !dbg !35 + %5258 = fadd float %4787, %5257, !dbg !35 + %5259 = fadd float %4788, %5258, !dbg !35 + %5260 = fadd float %4789, %4790, !dbg !35 + %5261 = fadd float %4791, %5260, !dbg !35 + %5262 = fadd float %4792, %5261, !dbg !35 + %5263 = fadd float %4793, %5262, !dbg !35 + %5264 = fadd float %4794, %5263, !dbg !35 + %5265 = fadd float %4795, %5264, !dbg !35 + %5266 = fadd float %4796, %5265, !dbg !35 + %5267 = fadd float %4797, %4798, !dbg !35 + %5268 = fadd float %4799, %5267, !dbg !35 + %5269 = fadd float %4800, %5268, !dbg !35 + %5270 = fadd float %4801, %5269, !dbg !35 + %5271 = fadd float %4802, %5270, !dbg !35 + %5272 = fadd float %4803, %5271, !dbg !35 + %5273 = fadd float %4804, %5272, !dbg !35 + %5274 = fadd float %4805, %4806, !dbg !35 + %5275 = fadd float %4807, %5274, !dbg !35 + %5276 = fadd float %4808, %5275, !dbg !35 + %5277 = fadd float %4809, %5276, !dbg !35 + %5278 = fadd float %4810, %5277, !dbg !35 + %5279 = fadd float %4811, %5278, !dbg !35 + %5280 = fadd float %4812, %5279, !dbg !35 + %5281 = fadd float %4813, %4814, !dbg !35 + %5282 = fadd float %4815, %5281, !dbg !35 + %5283 = fadd float %4816, %5282, !dbg !35 + %5284 = fadd float %4817, %5283, !dbg !35 + %5285 = fadd float %4818, %5284, !dbg !35 + %5286 = fadd float %4819, %5285, !dbg !35 + %5287 = fadd float %4820, %5286, !dbg !35 + %5288 = fadd float %4821, %4822, !dbg !35 + %5289 = fadd float %4823, %5288, !dbg !35 + %5290 = fadd float %4824, %5289, !dbg !35 + %5291 = fadd float %4825, %5290, !dbg !35 + %5292 = fadd float %4826, %5291, !dbg !35 + %5293 = fadd float %4827, %5292, !dbg !35 + %5294 = fadd float %4828, %5293, !dbg !35 + %5295 = fadd float %4829, %4830, !dbg !35 + %5296 = fadd float %4831, %5295, !dbg !35 + %5297 = fadd float %4832, %5296, !dbg !35 + %5298 = fadd float %4833, %5297, !dbg !35 + %5299 = fadd float %4834, %5298, !dbg !35 + %5300 = fadd float %4835, %5299, !dbg !35 + %5301 = fadd float %4836, %5300, !dbg !35 + %5302 = fadd float %4837, %4838, !dbg !35 + %5303 = fadd float %4839, %5302, !dbg !35 + %5304 = fadd float %4840, %5303, !dbg !35 + %5305 = fadd float %4841, %5304, !dbg !35 + %5306 = fadd float %4842, %5305, !dbg !35 + %5307 = fadd float %4843, %5306, !dbg !35 + %5308 = fadd float %4844, %5307, !dbg !35 + %5309 = fadd float %4845, %4846, !dbg !35 + %5310 = fadd float %4847, %5309, !dbg !35 + %5311 = fadd float %4848, %5310, !dbg !35 + %5312 = fadd float %4849, %5311, !dbg !35 + %5313 = fadd float %4850, %5312, !dbg !35 + %5314 = fadd float %4851, %5313, !dbg !35 + %5315 = fadd float %4852, %5314, !dbg !35 + %5316 = fadd float %4853, %4854, !dbg !35 + %5317 = fadd float %4855, %5316, !dbg !35 + %5318 = fadd float %4856, %5317, !dbg !35 + %5319 = fadd float %4857, %5318, !dbg !35 + %5320 = fadd float %4858, %5319, !dbg !35 + %5321 = fadd float %4859, %5320, !dbg !35 + %5322 = fadd float %4860, %5321, !dbg !35 + %5323 = fadd float %4861, %4862, !dbg !35 + %5324 = fadd float %4863, %5323, !dbg !35 + %5325 = fadd float %4864, %5324, !dbg !35 + %5326 = fadd float %4865, %5325, !dbg !35 + %5327 = fadd float %4866, %5326, !dbg !35 + %5328 = fadd float %4867, %5327, !dbg !35 + %5329 = fadd float %4868, %5328, !dbg !35 + %5330 = fadd float %4869, %4870, !dbg !35 + %5331 = fadd float %4871, %5330, !dbg !35 + %5332 = fadd float %4872, %5331, !dbg !35 + %5333 = fadd float %4873, %5332, !dbg !35 + %5334 = fadd float %4874, %5333, !dbg !35 + %5335 = fadd float %4875, %5334, !dbg !35 + %5336 = fadd float %4876, %5335, !dbg !35 + %5337 = fadd float %4877, %4878, !dbg !35 + %5338 = fadd float %4879, %5337, !dbg !35 + %5339 = fadd float %4880, %5338, !dbg !35 + %5340 = fadd float %4881, %5339, !dbg !35 + %5341 = fadd float %4882, %5340, !dbg !35 + %5342 = fadd float %4883, %5341, !dbg !35 + %5343 = fadd float %4884, %5342, !dbg !35 + %5344 = fadd float %4885, %4886, !dbg !35 + %5345 = fadd float %4887, %5344, !dbg !35 + %5346 = fadd float %4888, %5345, !dbg !35 + %5347 = fadd float %4889, %5346, !dbg !35 + %5348 = fadd float %4890, %5347, !dbg !35 + %5349 = fadd float %4891, %5348, !dbg !35 + %5350 = fadd float %4892, %5349, !dbg !35 + %5351 = fadd float %4893, %4894, !dbg !35 + %5352 = fadd float %4895, %5351, !dbg !35 + %5353 = fadd float %4896, %5352, !dbg !35 + %5354 = fadd float %4897, %5353, !dbg !35 + %5355 = fadd float %4898, %5354, !dbg !35 + %5356 = fadd float %4899, %5355, !dbg !35 + %5357 = fadd float %4900, %5356, !dbg !35 + %5358 = fadd float %4901, %4902, !dbg !35 + %5359 = fadd float %4903, %5358, !dbg !35 + %5360 = fadd float %4904, %5359, !dbg !35 + %5361 = fadd float %4905, %5360, !dbg !35 + %5362 = fadd float %4906, %5361, !dbg !35 + %5363 = fadd float %4907, %5362, !dbg !35 + %5364 = fadd float %4908, %5363, !dbg !35 + %5365 = fadd float %4909, %4910, !dbg !35 + %5366 = fadd float %4911, %5365, !dbg !35 + %5367 = fadd float %4912, %5366, !dbg !35 + %5368 = fadd float %4913, %5367, !dbg !35 + %5369 = fadd float %4914, %5368, !dbg !35 + %5370 = fadd float %4915, %5369, !dbg !35 + %5371 = fadd float %4916, %5370, !dbg !35 + %5372 = fadd float %4917, %4918, !dbg !35 + %5373 = fadd float %4919, %5372, !dbg !35 + %5374 = fadd float %4920, %5373, !dbg !35 + %5375 = fadd float %4921, %5374, !dbg !35 + %5376 = fadd float %4922, %5375, !dbg !35 + %5377 = fadd float %4923, %5376, !dbg !35 + %5378 = fadd float %4924, %5377, !dbg !35 + %5379 = fadd float %4925, %4926, !dbg !35 + %5380 = fadd float %4927, %5379, !dbg !35 + %5381 = fadd float %4928, %5380, !dbg !35 + %5382 = fadd float %4929, %5381, !dbg !35 + %5383 = fadd float %4930, %5382, !dbg !35 + %5384 = fadd float %4931, %5383, !dbg !35 + %5385 = fadd float %4932, %5384, !dbg !35 + %5386 = fadd float %4933, %4934, !dbg !35 + %5387 = fadd float %4935, %5386, !dbg !35 + %5388 = fadd float %4936, %5387, !dbg !35 + %5389 = fadd float %4937, %5388, !dbg !35 + %5390 = fadd float %4938, %5389, !dbg !35 + %5391 = fadd float %4939, %5390, !dbg !35 + %5392 = fadd float %4940, %5391, !dbg !35 + %5393 = fadd float %4941, %4942, !dbg !35 + %5394 = fadd float %4943, %5393, !dbg !35 + %5395 = fadd float %4944, %5394, !dbg !35 + %5396 = fadd float %4945, %5395, !dbg !35 + %5397 = fadd float %4946, %5396, !dbg !35 + %5398 = fadd float %4947, %5397, !dbg !35 + %5399 = fadd float %4948, %5398, !dbg !35 + %5400 = fadd float %4949, %4950, !dbg !35 + %5401 = fadd float %4951, %5400, !dbg !35 + %5402 = fadd float %4952, %5401, !dbg !35 + %5403 = fadd float %4953, %5402, !dbg !35 + %5404 = fadd float %4954, %5403, !dbg !35 + %5405 = fadd float %4955, %5404, !dbg !35 + %5406 = fadd float %4956, %5405, !dbg !35 + %5407 = fadd float %4957, %4958, !dbg !35 + %5408 = fadd float %4959, %5407, !dbg !35 + %5409 = fadd float %4960, %5408, !dbg !35 + %5410 = fadd float %4961, %5409, !dbg !35 + %5411 = fadd float %4962, %5410, !dbg !35 + %5412 = fadd float %4963, %5411, !dbg !35 + %5413 = fadd float %4964, %5412, !dbg !35 + %5414 = fadd float %4965, %4966, !dbg !35 + %5415 = fadd float %4967, %5414, !dbg !35 + %5416 = fadd float %4968, %5415, !dbg !35 + %5417 = fadd float %4969, %5416, !dbg !35 + %5418 = fadd float %4970, %5417, !dbg !35 + %5419 = fadd float %4971, %5418, !dbg !35 + %5420 = fadd float %4972, %5419, !dbg !35 + %5421 = bitcast float %4979 to i32, !dbg !38 + %5422 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5421, i32 8, i32 31), !dbg !38 + %5423 = bitcast i32 %5422 to float, !dbg !38 + %5424 = fadd float %4979, %5423, !dbg !35 + %5425 = bitcast float %5424 to i32, !dbg !38 + %5426 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5425, i32 4, i32 31), !dbg !38 + %5427 = bitcast i32 %5426 to float, !dbg !38 + %5428 = fadd float %5424, %5427, !dbg !35 + %5429 = bitcast float %5428 to i32, !dbg !38 + %5430 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5429, i32 2, i32 31), !dbg !38 + %5431 = bitcast i32 %5430 to float, !dbg !38 + %5432 = fadd float %5428, %5431, !dbg !35 + %5433 = bitcast float %5432 to i32, !dbg !38 + %5434 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5433, i32 1, i32 31), !dbg !38 + %5435 = bitcast i32 %5434 to float, !dbg !38 + %5436 = fadd float %5432, %5435, !dbg !35 + %5437 = bitcast float %4986 to i32, !dbg !38 + %5438 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5437, i32 8, i32 31), !dbg !38 + %5439 = bitcast i32 %5438 to float, !dbg !38 + %5440 = fadd float %4986, %5439, !dbg !35 + %5441 = bitcast float %5440 to i32, !dbg !38 + %5442 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5441, i32 4, i32 31), !dbg !38 + %5443 = bitcast i32 %5442 to float, !dbg !38 + %5444 = fadd float %5440, %5443, !dbg !35 + %5445 = bitcast float %5444 to i32, !dbg !38 + %5446 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5445, i32 2, i32 31), !dbg !38 + %5447 = bitcast i32 %5446 to float, !dbg !38 + %5448 = fadd float %5444, %5447, !dbg !35 + %5449 = bitcast float %5448 to i32, !dbg !38 + %5450 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5449, i32 1, i32 31), !dbg !38 + %5451 = bitcast i32 %5450 to float, !dbg !38 + %5452 = fadd float %5448, %5451, !dbg !35 + %5453 = bitcast float %4993 to i32, !dbg !38 + %5454 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5453, i32 8, i32 31), !dbg !38 + %5455 = bitcast i32 %5454 to float, !dbg !38 + %5456 = fadd float %4993, %5455, !dbg !35 + %5457 = bitcast float %5456 to i32, !dbg !38 + %5458 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5457, i32 4, i32 31), !dbg !38 + %5459 = bitcast i32 %5458 to float, !dbg !38 + %5460 = fadd float %5456, %5459, !dbg !35 + %5461 = bitcast float %5460 to i32, !dbg !38 + %5462 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5461, i32 2, i32 31), !dbg !38 + %5463 = bitcast i32 %5462 to float, !dbg !38 + %5464 = fadd float %5460, %5463, !dbg !35 + %5465 = bitcast float %5464 to i32, !dbg !38 + %5466 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5465, i32 1, i32 31), !dbg !38 + %5467 = bitcast i32 %5466 to float, !dbg !38 + %5468 = fadd float %5464, %5467, !dbg !35 + %5469 = bitcast float %5000 to i32, !dbg !38 + %5470 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5469, i32 8, i32 31), !dbg !38 + %5471 = bitcast i32 %5470 to float, !dbg !38 + %5472 = fadd float %5000, %5471, !dbg !35 + %5473 = bitcast float %5472 to i32, !dbg !38 + %5474 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5473, i32 4, i32 31), !dbg !38 + %5475 = bitcast i32 %5474 to float, !dbg !38 + %5476 = fadd float %5472, %5475, !dbg !35 + %5477 = bitcast float %5476 to i32, !dbg !38 + %5478 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5477, i32 2, i32 31), !dbg !38 + %5479 = bitcast i32 %5478 to float, !dbg !38 + %5480 = fadd float %5476, %5479, !dbg !35 + %5481 = bitcast float %5480 to i32, !dbg !38 + %5482 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5481, i32 1, i32 31), !dbg !38 + %5483 = bitcast i32 %5482 to float, !dbg !38 + %5484 = fadd float %5480, %5483, !dbg !35 + %5485 = bitcast float %5007 to i32, !dbg !38 + %5486 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5485, i32 8, i32 31), !dbg !38 + %5487 = bitcast i32 %5486 to float, !dbg !38 + %5488 = fadd float %5007, %5487, !dbg !35 + %5489 = bitcast float %5488 to i32, !dbg !38 + %5490 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5489, i32 4, i32 31), !dbg !38 + %5491 = bitcast i32 %5490 to float, !dbg !38 + %5492 = fadd float %5488, %5491, !dbg !35 + %5493 = bitcast float %5492 to i32, !dbg !38 + %5494 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5493, i32 2, i32 31), !dbg !38 + %5495 = bitcast i32 %5494 to float, !dbg !38 + %5496 = fadd float %5492, %5495, !dbg !35 + %5497 = bitcast float %5496 to i32, !dbg !38 + %5498 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5497, i32 1, i32 31), !dbg !38 + %5499 = bitcast i32 %5498 to float, !dbg !38 + %5500 = fadd float %5496, %5499, !dbg !35 + %5501 = bitcast float %5014 to i32, !dbg !38 + %5502 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5501, i32 8, i32 31), !dbg !38 + %5503 = bitcast i32 %5502 to float, !dbg !38 + %5504 = fadd float %5014, %5503, !dbg !35 + %5505 = bitcast float %5504 to i32, !dbg !38 + %5506 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5505, i32 4, i32 31), !dbg !38 + %5507 = bitcast i32 %5506 to float, !dbg !38 + %5508 = fadd float %5504, %5507, !dbg !35 + %5509 = bitcast float %5508 to i32, !dbg !38 + %5510 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5509, i32 2, i32 31), !dbg !38 + %5511 = bitcast i32 %5510 to float, !dbg !38 + %5512 = fadd float %5508, %5511, !dbg !35 + %5513 = bitcast float %5512 to i32, !dbg !38 + %5514 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5513, i32 1, i32 31), !dbg !38 + %5515 = bitcast i32 %5514 to float, !dbg !38 + %5516 = fadd float %5512, %5515, !dbg !35 + %5517 = bitcast float %5021 to i32, !dbg !38 + %5518 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5517, i32 8, i32 31), !dbg !38 + %5519 = bitcast i32 %5518 to float, !dbg !38 + %5520 = fadd float %5021, %5519, !dbg !35 + %5521 = bitcast float %5520 to i32, !dbg !38 + %5522 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5521, i32 4, i32 31), !dbg !38 + %5523 = bitcast i32 %5522 to float, !dbg !38 + %5524 = fadd float %5520, %5523, !dbg !35 + %5525 = bitcast float %5524 to i32, !dbg !38 + %5526 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5525, i32 2, i32 31), !dbg !38 + %5527 = bitcast i32 %5526 to float, !dbg !38 + %5528 = fadd float %5524, %5527, !dbg !35 + %5529 = bitcast float %5528 to i32, !dbg !38 + %5530 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5529, i32 1, i32 31), !dbg !38 + %5531 = bitcast i32 %5530 to float, !dbg !38 + %5532 = fadd float %5528, %5531, !dbg !35 + %5533 = bitcast float %5028 to i32, !dbg !38 + %5534 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5533, i32 8, i32 31), !dbg !38 + %5535 = bitcast i32 %5534 to float, !dbg !38 + %5536 = fadd float %5028, %5535, !dbg !35 + %5537 = bitcast float %5536 to i32, !dbg !38 + %5538 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5537, i32 4, i32 31), !dbg !38 + %5539 = bitcast i32 %5538 to float, !dbg !38 + %5540 = fadd float %5536, %5539, !dbg !35 + %5541 = bitcast float %5540 to i32, !dbg !38 + %5542 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5541, i32 2, i32 31), !dbg !38 + %5543 = bitcast i32 %5542 to float, !dbg !38 + %5544 = fadd float %5540, %5543, !dbg !35 + %5545 = bitcast float %5544 to i32, !dbg !38 + %5546 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5545, i32 1, i32 31), !dbg !38 + %5547 = bitcast i32 %5546 to float, !dbg !38 + %5548 = fadd float %5544, %5547, !dbg !35 + %5549 = bitcast float %5035 to i32, !dbg !38 + %5550 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5549, i32 8, i32 31), !dbg !38 + %5551 = bitcast i32 %5550 to float, !dbg !38 + %5552 = fadd float %5035, %5551, !dbg !35 + %5553 = bitcast float %5552 to i32, !dbg !38 + %5554 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5553, i32 4, i32 31), !dbg !38 + %5555 = bitcast i32 %5554 to float, !dbg !38 + %5556 = fadd float %5552, %5555, !dbg !35 + %5557 = bitcast float %5556 to i32, !dbg !38 + %5558 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5557, i32 2, i32 31), !dbg !38 + %5559 = bitcast i32 %5558 to float, !dbg !38 + %5560 = fadd float %5556, %5559, !dbg !35 + %5561 = bitcast float %5560 to i32, !dbg !38 + %5562 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5561, i32 1, i32 31), !dbg !38 + %5563 = bitcast i32 %5562 to float, !dbg !38 + %5564 = fadd float %5560, %5563, !dbg !35 + %5565 = bitcast float %5042 to i32, !dbg !38 + %5566 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5565, i32 8, i32 31), !dbg !38 + %5567 = bitcast i32 %5566 to float, !dbg !38 + %5568 = fadd float %5042, %5567, !dbg !35 + %5569 = bitcast float %5568 to i32, !dbg !38 + %5570 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5569, i32 4, i32 31), !dbg !38 + %5571 = bitcast i32 %5570 to float, !dbg !38 + %5572 = fadd float %5568, %5571, !dbg !35 + %5573 = bitcast float %5572 to i32, !dbg !38 + %5574 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5573, i32 2, i32 31), !dbg !38 + %5575 = bitcast i32 %5574 to float, !dbg !38 + %5576 = fadd float %5572, %5575, !dbg !35 + %5577 = bitcast float %5576 to i32, !dbg !38 + %5578 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5577, i32 1, i32 31), !dbg !38 + %5579 = bitcast i32 %5578 to float, !dbg !38 + %5580 = fadd float %5576, %5579, !dbg !35 + %5581 = bitcast float %5049 to i32, !dbg !38 + %5582 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5581, i32 8, i32 31), !dbg !38 + %5583 = bitcast i32 %5582 to float, !dbg !38 + %5584 = fadd float %5049, %5583, !dbg !35 + %5585 = bitcast float %5584 to i32, !dbg !38 + %5586 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5585, i32 4, i32 31), !dbg !38 + %5587 = bitcast i32 %5586 to float, !dbg !38 + %5588 = fadd float %5584, %5587, !dbg !35 + %5589 = bitcast float %5588 to i32, !dbg !38 + %5590 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5589, i32 2, i32 31), !dbg !38 + %5591 = bitcast i32 %5590 to float, !dbg !38 + %5592 = fadd float %5588, %5591, !dbg !35 + %5593 = bitcast float %5592 to i32, !dbg !38 + %5594 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5593, i32 1, i32 31), !dbg !38 + %5595 = bitcast i32 %5594 to float, !dbg !38 + %5596 = fadd float %5592, %5595, !dbg !35 + %5597 = bitcast float %5056 to i32, !dbg !38 + %5598 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5597, i32 8, i32 31), !dbg !38 + %5599 = bitcast i32 %5598 to float, !dbg !38 + %5600 = fadd float %5056, %5599, !dbg !35 + %5601 = bitcast float %5600 to i32, !dbg !38 + %5602 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5601, i32 4, i32 31), !dbg !38 + %5603 = bitcast i32 %5602 to float, !dbg !38 + %5604 = fadd float %5600, %5603, !dbg !35 + %5605 = bitcast float %5604 to i32, !dbg !38 + %5606 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5605, i32 2, i32 31), !dbg !38 + %5607 = bitcast i32 %5606 to float, !dbg !38 + %5608 = fadd float %5604, %5607, !dbg !35 + %5609 = bitcast float %5608 to i32, !dbg !38 + %5610 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5609, i32 1, i32 31), !dbg !38 + %5611 = bitcast i32 %5610 to float, !dbg !38 + %5612 = fadd float %5608, %5611, !dbg !35 + %5613 = bitcast float %5063 to i32, !dbg !38 + %5614 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5613, i32 8, i32 31), !dbg !38 + %5615 = bitcast i32 %5614 to float, !dbg !38 + %5616 = fadd float %5063, %5615, !dbg !35 + %5617 = bitcast float %5616 to i32, !dbg !38 + %5618 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5617, i32 4, i32 31), !dbg !38 + %5619 = bitcast i32 %5618 to float, !dbg !38 + %5620 = fadd float %5616, %5619, !dbg !35 + %5621 = bitcast float %5620 to i32, !dbg !38 + %5622 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5621, i32 2, i32 31), !dbg !38 + %5623 = bitcast i32 %5622 to float, !dbg !38 + %5624 = fadd float %5620, %5623, !dbg !35 + %5625 = bitcast float %5624 to i32, !dbg !38 + %5626 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5625, i32 1, i32 31), !dbg !38 + %5627 = bitcast i32 %5626 to float, !dbg !38 + %5628 = fadd float %5624, %5627, !dbg !35 + %5629 = bitcast float %5070 to i32, !dbg !38 + %5630 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5629, i32 8, i32 31), !dbg !38 + %5631 = bitcast i32 %5630 to float, !dbg !38 + %5632 = fadd float %5070, %5631, !dbg !35 + %5633 = bitcast float %5632 to i32, !dbg !38 + %5634 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5633, i32 4, i32 31), !dbg !38 + %5635 = bitcast i32 %5634 to float, !dbg !38 + %5636 = fadd float %5632, %5635, !dbg !35 + %5637 = bitcast float %5636 to i32, !dbg !38 + %5638 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5637, i32 2, i32 31), !dbg !38 + %5639 = bitcast i32 %5638 to float, !dbg !38 + %5640 = fadd float %5636, %5639, !dbg !35 + %5641 = bitcast float %5640 to i32, !dbg !38 + %5642 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5641, i32 1, i32 31), !dbg !38 + %5643 = bitcast i32 %5642 to float, !dbg !38 + %5644 = fadd float %5640, %5643, !dbg !35 + %5645 = bitcast float %5077 to i32, !dbg !38 + %5646 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5645, i32 8, i32 31), !dbg !38 + %5647 = bitcast i32 %5646 to float, !dbg !38 + %5648 = fadd float %5077, %5647, !dbg !35 + %5649 = bitcast float %5648 to i32, !dbg !38 + %5650 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5649, i32 4, i32 31), !dbg !38 + %5651 = bitcast i32 %5650 to float, !dbg !38 + %5652 = fadd float %5648, %5651, !dbg !35 + %5653 = bitcast float %5652 to i32, !dbg !38 + %5654 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5653, i32 2, i32 31), !dbg !38 + %5655 = bitcast i32 %5654 to float, !dbg !38 + %5656 = fadd float %5652, %5655, !dbg !35 + %5657 = bitcast float %5656 to i32, !dbg !38 + %5658 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5657, i32 1, i32 31), !dbg !38 + %5659 = bitcast i32 %5658 to float, !dbg !38 + %5660 = fadd float %5656, %5659, !dbg !35 + %5661 = bitcast float %5084 to i32, !dbg !38 + %5662 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5661, i32 8, i32 31), !dbg !38 + %5663 = bitcast i32 %5662 to float, !dbg !38 + %5664 = fadd float %5084, %5663, !dbg !35 + %5665 = bitcast float %5664 to i32, !dbg !38 + %5666 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5665, i32 4, i32 31), !dbg !38 + %5667 = bitcast i32 %5666 to float, !dbg !38 + %5668 = fadd float %5664, %5667, !dbg !35 + %5669 = bitcast float %5668 to i32, !dbg !38 + %5670 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5669, i32 2, i32 31), !dbg !38 + %5671 = bitcast i32 %5670 to float, !dbg !38 + %5672 = fadd float %5668, %5671, !dbg !35 + %5673 = bitcast float %5672 to i32, !dbg !38 + %5674 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5673, i32 1, i32 31), !dbg !38 + %5675 = bitcast i32 %5674 to float, !dbg !38 + %5676 = fadd float %5672, %5675, !dbg !35 + %5677 = bitcast float %5091 to i32, !dbg !38 + %5678 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5677, i32 8, i32 31), !dbg !38 + %5679 = bitcast i32 %5678 to float, !dbg !38 + %5680 = fadd float %5091, %5679, !dbg !35 + %5681 = bitcast float %5680 to i32, !dbg !38 + %5682 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5681, i32 4, i32 31), !dbg !38 + %5683 = bitcast i32 %5682 to float, !dbg !38 + %5684 = fadd float %5680, %5683, !dbg !35 + %5685 = bitcast float %5684 to i32, !dbg !38 + %5686 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5685, i32 2, i32 31), !dbg !38 + %5687 = bitcast i32 %5686 to float, !dbg !38 + %5688 = fadd float %5684, %5687, !dbg !35 + %5689 = bitcast float %5688 to i32, !dbg !38 + %5690 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5689, i32 1, i32 31), !dbg !38 + %5691 = bitcast i32 %5690 to float, !dbg !38 + %5692 = fadd float %5688, %5691, !dbg !35 + %5693 = bitcast float %5098 to i32, !dbg !38 + %5694 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5693, i32 8, i32 31), !dbg !38 + %5695 = bitcast i32 %5694 to float, !dbg !38 + %5696 = fadd float %5098, %5695, !dbg !35 + %5697 = bitcast float %5696 to i32, !dbg !38 + %5698 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5697, i32 4, i32 31), !dbg !38 + %5699 = bitcast i32 %5698 to float, !dbg !38 + %5700 = fadd float %5696, %5699, !dbg !35 + %5701 = bitcast float %5700 to i32, !dbg !38 + %5702 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5701, i32 2, i32 31), !dbg !38 + %5703 = bitcast i32 %5702 to float, !dbg !38 + %5704 = fadd float %5700, %5703, !dbg !35 + %5705 = bitcast float %5704 to i32, !dbg !38 + %5706 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5705, i32 1, i32 31), !dbg !38 + %5707 = bitcast i32 %5706 to float, !dbg !38 + %5708 = fadd float %5704, %5707, !dbg !35 + %5709 = bitcast float %5105 to i32, !dbg !38 + %5710 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5709, i32 8, i32 31), !dbg !38 + %5711 = bitcast i32 %5710 to float, !dbg !38 + %5712 = fadd float %5105, %5711, !dbg !35 + %5713 = bitcast float %5712 to i32, !dbg !38 + %5714 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5713, i32 4, i32 31), !dbg !38 + %5715 = bitcast i32 %5714 to float, !dbg !38 + %5716 = fadd float %5712, %5715, !dbg !35 + %5717 = bitcast float %5716 to i32, !dbg !38 + %5718 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5717, i32 2, i32 31), !dbg !38 + %5719 = bitcast i32 %5718 to float, !dbg !38 + %5720 = fadd float %5716, %5719, !dbg !35 + %5721 = bitcast float %5720 to i32, !dbg !38 + %5722 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5721, i32 1, i32 31), !dbg !38 + %5723 = bitcast i32 %5722 to float, !dbg !38 + %5724 = fadd float %5720, %5723, !dbg !35 + %5725 = bitcast float %5112 to i32, !dbg !38 + %5726 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5725, i32 8, i32 31), !dbg !38 + %5727 = bitcast i32 %5726 to float, !dbg !38 + %5728 = fadd float %5112, %5727, !dbg !35 + %5729 = bitcast float %5728 to i32, !dbg !38 + %5730 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5729, i32 4, i32 31), !dbg !38 + %5731 = bitcast i32 %5730 to float, !dbg !38 + %5732 = fadd float %5728, %5731, !dbg !35 + %5733 = bitcast float %5732 to i32, !dbg !38 + %5734 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5733, i32 2, i32 31), !dbg !38 + %5735 = bitcast i32 %5734 to float, !dbg !38 + %5736 = fadd float %5732, %5735, !dbg !35 + %5737 = bitcast float %5736 to i32, !dbg !38 + %5738 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5737, i32 1, i32 31), !dbg !38 + %5739 = bitcast i32 %5738 to float, !dbg !38 + %5740 = fadd float %5736, %5739, !dbg !35 + %5741 = bitcast float %5119 to i32, !dbg !38 + %5742 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5741, i32 8, i32 31), !dbg !38 + %5743 = bitcast i32 %5742 to float, !dbg !38 + %5744 = fadd float %5119, %5743, !dbg !35 + %5745 = bitcast float %5744 to i32, !dbg !38 + %5746 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5745, i32 4, i32 31), !dbg !38 + %5747 = bitcast i32 %5746 to float, !dbg !38 + %5748 = fadd float %5744, %5747, !dbg !35 + %5749 = bitcast float %5748 to i32, !dbg !38 + %5750 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5749, i32 2, i32 31), !dbg !38 + %5751 = bitcast i32 %5750 to float, !dbg !38 + %5752 = fadd float %5748, %5751, !dbg !35 + %5753 = bitcast float %5752 to i32, !dbg !38 + %5754 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5753, i32 1, i32 31), !dbg !38 + %5755 = bitcast i32 %5754 to float, !dbg !38 + %5756 = fadd float %5752, %5755, !dbg !35 + %5757 = bitcast float %5126 to i32, !dbg !38 + %5758 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5757, i32 8, i32 31), !dbg !38 + %5759 = bitcast i32 %5758 to float, !dbg !38 + %5760 = fadd float %5126, %5759, !dbg !35 + %5761 = bitcast float %5760 to i32, !dbg !38 + %5762 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5761, i32 4, i32 31), !dbg !38 + %5763 = bitcast i32 %5762 to float, !dbg !38 + %5764 = fadd float %5760, %5763, !dbg !35 + %5765 = bitcast float %5764 to i32, !dbg !38 + %5766 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5765, i32 2, i32 31), !dbg !38 + %5767 = bitcast i32 %5766 to float, !dbg !38 + %5768 = fadd float %5764, %5767, !dbg !35 + %5769 = bitcast float %5768 to i32, !dbg !38 + %5770 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5769, i32 1, i32 31), !dbg !38 + %5771 = bitcast i32 %5770 to float, !dbg !38 + %5772 = fadd float %5768, %5771, !dbg !35 + %5773 = bitcast float %5133 to i32, !dbg !38 + %5774 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5773, i32 8, i32 31), !dbg !38 + %5775 = bitcast i32 %5774 to float, !dbg !38 + %5776 = fadd float %5133, %5775, !dbg !35 + %5777 = bitcast float %5776 to i32, !dbg !38 + %5778 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5777, i32 4, i32 31), !dbg !38 + %5779 = bitcast i32 %5778 to float, !dbg !38 + %5780 = fadd float %5776, %5779, !dbg !35 + %5781 = bitcast float %5780 to i32, !dbg !38 + %5782 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5781, i32 2, i32 31), !dbg !38 + %5783 = bitcast i32 %5782 to float, !dbg !38 + %5784 = fadd float %5780, %5783, !dbg !35 + %5785 = bitcast float %5784 to i32, !dbg !38 + %5786 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5785, i32 1, i32 31), !dbg !38 + %5787 = bitcast i32 %5786 to float, !dbg !38 + %5788 = fadd float %5784, %5787, !dbg !35 + %5789 = bitcast float %5140 to i32, !dbg !38 + %5790 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5789, i32 8, i32 31), !dbg !38 + %5791 = bitcast i32 %5790 to float, !dbg !38 + %5792 = fadd float %5140, %5791, !dbg !35 + %5793 = bitcast float %5792 to i32, !dbg !38 + %5794 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5793, i32 4, i32 31), !dbg !38 + %5795 = bitcast i32 %5794 to float, !dbg !38 + %5796 = fadd float %5792, %5795, !dbg !35 + %5797 = bitcast float %5796 to i32, !dbg !38 + %5798 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5797, i32 2, i32 31), !dbg !38 + %5799 = bitcast i32 %5798 to float, !dbg !38 + %5800 = fadd float %5796, %5799, !dbg !35 + %5801 = bitcast float %5800 to i32, !dbg !38 + %5802 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5801, i32 1, i32 31), !dbg !38 + %5803 = bitcast i32 %5802 to float, !dbg !38 + %5804 = fadd float %5800, %5803, !dbg !35 + %5805 = bitcast float %5147 to i32, !dbg !38 + %5806 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5805, i32 8, i32 31), !dbg !38 + %5807 = bitcast i32 %5806 to float, !dbg !38 + %5808 = fadd float %5147, %5807, !dbg !35 + %5809 = bitcast float %5808 to i32, !dbg !38 + %5810 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5809, i32 4, i32 31), !dbg !38 + %5811 = bitcast i32 %5810 to float, !dbg !38 + %5812 = fadd float %5808, %5811, !dbg !35 + %5813 = bitcast float %5812 to i32, !dbg !38 + %5814 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5813, i32 2, i32 31), !dbg !38 + %5815 = bitcast i32 %5814 to float, !dbg !38 + %5816 = fadd float %5812, %5815, !dbg !35 + %5817 = bitcast float %5816 to i32, !dbg !38 + %5818 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5817, i32 1, i32 31), !dbg !38 + %5819 = bitcast i32 %5818 to float, !dbg !38 + %5820 = fadd float %5816, %5819, !dbg !35 + %5821 = bitcast float %5154 to i32, !dbg !38 + %5822 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5821, i32 8, i32 31), !dbg !38 + %5823 = bitcast i32 %5822 to float, !dbg !38 + %5824 = fadd float %5154, %5823, !dbg !35 + %5825 = bitcast float %5824 to i32, !dbg !38 + %5826 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5825, i32 4, i32 31), !dbg !38 + %5827 = bitcast i32 %5826 to float, !dbg !38 + %5828 = fadd float %5824, %5827, !dbg !35 + %5829 = bitcast float %5828 to i32, !dbg !38 + %5830 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5829, i32 2, i32 31), !dbg !38 + %5831 = bitcast i32 %5830 to float, !dbg !38 + %5832 = fadd float %5828, %5831, !dbg !35 + %5833 = bitcast float %5832 to i32, !dbg !38 + %5834 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5833, i32 1, i32 31), !dbg !38 + %5835 = bitcast i32 %5834 to float, !dbg !38 + %5836 = fadd float %5832, %5835, !dbg !35 + %5837 = bitcast float %5161 to i32, !dbg !38 + %5838 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5837, i32 8, i32 31), !dbg !38 + %5839 = bitcast i32 %5838 to float, !dbg !38 + %5840 = fadd float %5161, %5839, !dbg !35 + %5841 = bitcast float %5840 to i32, !dbg !38 + %5842 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5841, i32 4, i32 31), !dbg !38 + %5843 = bitcast i32 %5842 to float, !dbg !38 + %5844 = fadd float %5840, %5843, !dbg !35 + %5845 = bitcast float %5844 to i32, !dbg !38 + %5846 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5845, i32 2, i32 31), !dbg !38 + %5847 = bitcast i32 %5846 to float, !dbg !38 + %5848 = fadd float %5844, %5847, !dbg !35 + %5849 = bitcast float %5848 to i32, !dbg !38 + %5850 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5849, i32 1, i32 31), !dbg !38 + %5851 = bitcast i32 %5850 to float, !dbg !38 + %5852 = fadd float %5848, %5851, !dbg !35 + %5853 = bitcast float %5168 to i32, !dbg !38 + %5854 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5853, i32 8, i32 31), !dbg !38 + %5855 = bitcast i32 %5854 to float, !dbg !38 + %5856 = fadd float %5168, %5855, !dbg !35 + %5857 = bitcast float %5856 to i32, !dbg !38 + %5858 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5857, i32 4, i32 31), !dbg !38 + %5859 = bitcast i32 %5858 to float, !dbg !38 + %5860 = fadd float %5856, %5859, !dbg !35 + %5861 = bitcast float %5860 to i32, !dbg !38 + %5862 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5861, i32 2, i32 31), !dbg !38 + %5863 = bitcast i32 %5862 to float, !dbg !38 + %5864 = fadd float %5860, %5863, !dbg !35 + %5865 = bitcast float %5864 to i32, !dbg !38 + %5866 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5865, i32 1, i32 31), !dbg !38 + %5867 = bitcast i32 %5866 to float, !dbg !38 + %5868 = fadd float %5864, %5867, !dbg !35 + %5869 = bitcast float %5175 to i32, !dbg !38 + %5870 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5869, i32 8, i32 31), !dbg !38 + %5871 = bitcast i32 %5870 to float, !dbg !38 + %5872 = fadd float %5175, %5871, !dbg !35 + %5873 = bitcast float %5872 to i32, !dbg !38 + %5874 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5873, i32 4, i32 31), !dbg !38 + %5875 = bitcast i32 %5874 to float, !dbg !38 + %5876 = fadd float %5872, %5875, !dbg !35 + %5877 = bitcast float %5876 to i32, !dbg !38 + %5878 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5877, i32 2, i32 31), !dbg !38 + %5879 = bitcast i32 %5878 to float, !dbg !38 + %5880 = fadd float %5876, %5879, !dbg !35 + %5881 = bitcast float %5880 to i32, !dbg !38 + %5882 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5881, i32 1, i32 31), !dbg !38 + %5883 = bitcast i32 %5882 to float, !dbg !38 + %5884 = fadd float %5880, %5883, !dbg !35 + %5885 = bitcast float %5182 to i32, !dbg !38 + %5886 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5885, i32 8, i32 31), !dbg !38 + %5887 = bitcast i32 %5886 to float, !dbg !38 + %5888 = fadd float %5182, %5887, !dbg !35 + %5889 = bitcast float %5888 to i32, !dbg !38 + %5890 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5889, i32 4, i32 31), !dbg !38 + %5891 = bitcast i32 %5890 to float, !dbg !38 + %5892 = fadd float %5888, %5891, !dbg !35 + %5893 = bitcast float %5892 to i32, !dbg !38 + %5894 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5893, i32 2, i32 31), !dbg !38 + %5895 = bitcast i32 %5894 to float, !dbg !38 + %5896 = fadd float %5892, %5895, !dbg !35 + %5897 = bitcast float %5896 to i32, !dbg !38 + %5898 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5897, i32 1, i32 31), !dbg !38 + %5899 = bitcast i32 %5898 to float, !dbg !38 + %5900 = fadd float %5896, %5899, !dbg !35 + %5901 = bitcast float %5189 to i32, !dbg !38 + %5902 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5901, i32 8, i32 31), !dbg !38 + %5903 = bitcast i32 %5902 to float, !dbg !38 + %5904 = fadd float %5189, %5903, !dbg !35 + %5905 = bitcast float %5904 to i32, !dbg !38 + %5906 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5905, i32 4, i32 31), !dbg !38 + %5907 = bitcast i32 %5906 to float, !dbg !38 + %5908 = fadd float %5904, %5907, !dbg !35 + %5909 = bitcast float %5908 to i32, !dbg !38 + %5910 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5909, i32 2, i32 31), !dbg !38 + %5911 = bitcast i32 %5910 to float, !dbg !38 + %5912 = fadd float %5908, %5911, !dbg !35 + %5913 = bitcast float %5912 to i32, !dbg !38 + %5914 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5913, i32 1, i32 31), !dbg !38 + %5915 = bitcast i32 %5914 to float, !dbg !38 + %5916 = fadd float %5912, %5915, !dbg !35 + %5917 = bitcast float %5196 to i32, !dbg !38 + %5918 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5917, i32 8, i32 31), !dbg !38 + %5919 = bitcast i32 %5918 to float, !dbg !38 + %5920 = fadd float %5196, %5919, !dbg !35 + %5921 = bitcast float %5920 to i32, !dbg !38 + %5922 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5921, i32 4, i32 31), !dbg !38 + %5923 = bitcast i32 %5922 to float, !dbg !38 + %5924 = fadd float %5920, %5923, !dbg !35 + %5925 = bitcast float %5924 to i32, !dbg !38 + %5926 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5925, i32 2, i32 31), !dbg !38 + %5927 = bitcast i32 %5926 to float, !dbg !38 + %5928 = fadd float %5924, %5927, !dbg !35 + %5929 = bitcast float %5928 to i32, !dbg !38 + %5930 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5929, i32 1, i32 31), !dbg !38 + %5931 = bitcast i32 %5930 to float, !dbg !38 + %5932 = fadd float %5928, %5931, !dbg !35 + %5933 = bitcast float %5203 to i32, !dbg !38 + %5934 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5933, i32 8, i32 31), !dbg !38 + %5935 = bitcast i32 %5934 to float, !dbg !38 + %5936 = fadd float %5203, %5935, !dbg !35 + %5937 = bitcast float %5936 to i32, !dbg !38 + %5938 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5937, i32 4, i32 31), !dbg !38 + %5939 = bitcast i32 %5938 to float, !dbg !38 + %5940 = fadd float %5936, %5939, !dbg !35 + %5941 = bitcast float %5940 to i32, !dbg !38 + %5942 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5941, i32 2, i32 31), !dbg !38 + %5943 = bitcast i32 %5942 to float, !dbg !38 + %5944 = fadd float %5940, %5943, !dbg !35 + %5945 = bitcast float %5944 to i32, !dbg !38 + %5946 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5945, i32 1, i32 31), !dbg !38 + %5947 = bitcast i32 %5946 to float, !dbg !38 + %5948 = fadd float %5944, %5947, !dbg !35 + %5949 = bitcast float %5210 to i32, !dbg !38 + %5950 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5949, i32 8, i32 31), !dbg !38 + %5951 = bitcast i32 %5950 to float, !dbg !38 + %5952 = fadd float %5210, %5951, !dbg !35 + %5953 = bitcast float %5952 to i32, !dbg !38 + %5954 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5953, i32 4, i32 31), !dbg !38 + %5955 = bitcast i32 %5954 to float, !dbg !38 + %5956 = fadd float %5952, %5955, !dbg !35 + %5957 = bitcast float %5956 to i32, !dbg !38 + %5958 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5957, i32 2, i32 31), !dbg !38 + %5959 = bitcast i32 %5958 to float, !dbg !38 + %5960 = fadd float %5956, %5959, !dbg !35 + %5961 = bitcast float %5960 to i32, !dbg !38 + %5962 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5961, i32 1, i32 31), !dbg !38 + %5963 = bitcast i32 %5962 to float, !dbg !38 + %5964 = fadd float %5960, %5963, !dbg !35 + %5965 = bitcast float %5217 to i32, !dbg !38 + %5966 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5965, i32 8, i32 31), !dbg !38 + %5967 = bitcast i32 %5966 to float, !dbg !38 + %5968 = fadd float %5217, %5967, !dbg !35 + %5969 = bitcast float %5968 to i32, !dbg !38 + %5970 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5969, i32 4, i32 31), !dbg !38 + %5971 = bitcast i32 %5970 to float, !dbg !38 + %5972 = fadd float %5968, %5971, !dbg !35 + %5973 = bitcast float %5972 to i32, !dbg !38 + %5974 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5973, i32 2, i32 31), !dbg !38 + %5975 = bitcast i32 %5974 to float, !dbg !38 + %5976 = fadd float %5972, %5975, !dbg !35 + %5977 = bitcast float %5976 to i32, !dbg !38 + %5978 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5977, i32 1, i32 31), !dbg !38 + %5979 = bitcast i32 %5978 to float, !dbg !38 + %5980 = fadd float %5976, %5979, !dbg !35 + %5981 = bitcast float %5224 to i32, !dbg !38 + %5982 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5981, i32 8, i32 31), !dbg !38 + %5983 = bitcast i32 %5982 to float, !dbg !38 + %5984 = fadd float %5224, %5983, !dbg !35 + %5985 = bitcast float %5984 to i32, !dbg !38 + %5986 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5985, i32 4, i32 31), !dbg !38 + %5987 = bitcast i32 %5986 to float, !dbg !38 + %5988 = fadd float %5984, %5987, !dbg !35 + %5989 = bitcast float %5988 to i32, !dbg !38 + %5990 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5989, i32 2, i32 31), !dbg !38 + %5991 = bitcast i32 %5990 to float, !dbg !38 + %5992 = fadd float %5988, %5991, !dbg !35 + %5993 = bitcast float %5992 to i32, !dbg !38 + %5994 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5993, i32 1, i32 31), !dbg !38 + %5995 = bitcast i32 %5994 to float, !dbg !38 + %5996 = fadd float %5992, %5995, !dbg !35 + %5997 = bitcast float %5231 to i32, !dbg !38 + %5998 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5997, i32 8, i32 31), !dbg !38 + %5999 = bitcast i32 %5998 to float, !dbg !38 + %6000 = fadd float %5231, %5999, !dbg !35 + %6001 = bitcast float %6000 to i32, !dbg !38 + %6002 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6001, i32 4, i32 31), !dbg !38 + %6003 = bitcast i32 %6002 to float, !dbg !38 + %6004 = fadd float %6000, %6003, !dbg !35 + %6005 = bitcast float %6004 to i32, !dbg !38 + %6006 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6005, i32 2, i32 31), !dbg !38 + %6007 = bitcast i32 %6006 to float, !dbg !38 + %6008 = fadd float %6004, %6007, !dbg !35 + %6009 = bitcast float %6008 to i32, !dbg !38 + %6010 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6009, i32 1, i32 31), !dbg !38 + %6011 = bitcast i32 %6010 to float, !dbg !38 + %6012 = fadd float %6008, %6011, !dbg !35 + %6013 = bitcast float %5238 to i32, !dbg !38 + %6014 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6013, i32 8, i32 31), !dbg !38 + %6015 = bitcast i32 %6014 to float, !dbg !38 + %6016 = fadd float %5238, %6015, !dbg !35 + %6017 = bitcast float %6016 to i32, !dbg !38 + %6018 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6017, i32 4, i32 31), !dbg !38 + %6019 = bitcast i32 %6018 to float, !dbg !38 + %6020 = fadd float %6016, %6019, !dbg !35 + %6021 = bitcast float %6020 to i32, !dbg !38 + %6022 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6021, i32 2, i32 31), !dbg !38 + %6023 = bitcast i32 %6022 to float, !dbg !38 + %6024 = fadd float %6020, %6023, !dbg !35 + %6025 = bitcast float %6024 to i32, !dbg !38 + %6026 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6025, i32 1, i32 31), !dbg !38 + %6027 = bitcast i32 %6026 to float, !dbg !38 + %6028 = fadd float %6024, %6027, !dbg !35 + %6029 = bitcast float %5245 to i32, !dbg !38 + %6030 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6029, i32 8, i32 31), !dbg !38 + %6031 = bitcast i32 %6030 to float, !dbg !38 + %6032 = fadd float %5245, %6031, !dbg !35 + %6033 = bitcast float %6032 to i32, !dbg !38 + %6034 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6033, i32 4, i32 31), !dbg !38 + %6035 = bitcast i32 %6034 to float, !dbg !38 + %6036 = fadd float %6032, %6035, !dbg !35 + %6037 = bitcast float %6036 to i32, !dbg !38 + %6038 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6037, i32 2, i32 31), !dbg !38 + %6039 = bitcast i32 %6038 to float, !dbg !38 + %6040 = fadd float %6036, %6039, !dbg !35 + %6041 = bitcast float %6040 to i32, !dbg !38 + %6042 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6041, i32 1, i32 31), !dbg !38 + %6043 = bitcast i32 %6042 to float, !dbg !38 + %6044 = fadd float %6040, %6043, !dbg !35 + %6045 = bitcast float %5252 to i32, !dbg !38 + %6046 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6045, i32 8, i32 31), !dbg !38 + %6047 = bitcast i32 %6046 to float, !dbg !38 + %6048 = fadd float %5252, %6047, !dbg !35 + %6049 = bitcast float %6048 to i32, !dbg !38 + %6050 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6049, i32 4, i32 31), !dbg !38 + %6051 = bitcast i32 %6050 to float, !dbg !38 + %6052 = fadd float %6048, %6051, !dbg !35 + %6053 = bitcast float %6052 to i32, !dbg !38 + %6054 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6053, i32 2, i32 31), !dbg !38 + %6055 = bitcast i32 %6054 to float, !dbg !38 + %6056 = fadd float %6052, %6055, !dbg !35 + %6057 = bitcast float %6056 to i32, !dbg !38 + %6058 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6057, i32 1, i32 31), !dbg !38 + %6059 = bitcast i32 %6058 to float, !dbg !38 + %6060 = fadd float %6056, %6059, !dbg !35 + %6061 = bitcast float %5259 to i32, !dbg !38 + %6062 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6061, i32 8, i32 31), !dbg !38 + %6063 = bitcast i32 %6062 to float, !dbg !38 + %6064 = fadd float %5259, %6063, !dbg !35 + %6065 = bitcast float %6064 to i32, !dbg !38 + %6066 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6065, i32 4, i32 31), !dbg !38 + %6067 = bitcast i32 %6066 to float, !dbg !38 + %6068 = fadd float %6064, %6067, !dbg !35 + %6069 = bitcast float %6068 to i32, !dbg !38 + %6070 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6069, i32 2, i32 31), !dbg !38 + %6071 = bitcast i32 %6070 to float, !dbg !38 + %6072 = fadd float %6068, %6071, !dbg !35 + %6073 = bitcast float %6072 to i32, !dbg !38 + %6074 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6073, i32 1, i32 31), !dbg !38 + %6075 = bitcast i32 %6074 to float, !dbg !38 + %6076 = fadd float %6072, %6075, !dbg !35 + %6077 = bitcast float %5266 to i32, !dbg !38 + %6078 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6077, i32 8, i32 31), !dbg !38 + %6079 = bitcast i32 %6078 to float, !dbg !38 + %6080 = fadd float %5266, %6079, !dbg !35 + %6081 = bitcast float %6080 to i32, !dbg !38 + %6082 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6081, i32 4, i32 31), !dbg !38 + %6083 = bitcast i32 %6082 to float, !dbg !38 + %6084 = fadd float %6080, %6083, !dbg !35 + %6085 = bitcast float %6084 to i32, !dbg !38 + %6086 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6085, i32 2, i32 31), !dbg !38 + %6087 = bitcast i32 %6086 to float, !dbg !38 + %6088 = fadd float %6084, %6087, !dbg !35 + %6089 = bitcast float %6088 to i32, !dbg !38 + %6090 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6089, i32 1, i32 31), !dbg !38 + %6091 = bitcast i32 %6090 to float, !dbg !38 + %6092 = fadd float %6088, %6091, !dbg !35 + %6093 = bitcast float %5273 to i32, !dbg !38 + %6094 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6093, i32 8, i32 31), !dbg !38 + %6095 = bitcast i32 %6094 to float, !dbg !38 + %6096 = fadd float %5273, %6095, !dbg !35 + %6097 = bitcast float %6096 to i32, !dbg !38 + %6098 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6097, i32 4, i32 31), !dbg !38 + %6099 = bitcast i32 %6098 to float, !dbg !38 + %6100 = fadd float %6096, %6099, !dbg !35 + %6101 = bitcast float %6100 to i32, !dbg !38 + %6102 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6101, i32 2, i32 31), !dbg !38 + %6103 = bitcast i32 %6102 to float, !dbg !38 + %6104 = fadd float %6100, %6103, !dbg !35 + %6105 = bitcast float %6104 to i32, !dbg !38 + %6106 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6105, i32 1, i32 31), !dbg !38 + %6107 = bitcast i32 %6106 to float, !dbg !38 + %6108 = fadd float %6104, %6107, !dbg !35 + %6109 = bitcast float %5280 to i32, !dbg !38 + %6110 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6109, i32 8, i32 31), !dbg !38 + %6111 = bitcast i32 %6110 to float, !dbg !38 + %6112 = fadd float %5280, %6111, !dbg !35 + %6113 = bitcast float %6112 to i32, !dbg !38 + %6114 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6113, i32 4, i32 31), !dbg !38 + %6115 = bitcast i32 %6114 to float, !dbg !38 + %6116 = fadd float %6112, %6115, !dbg !35 + %6117 = bitcast float %6116 to i32, !dbg !38 + %6118 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6117, i32 2, i32 31), !dbg !38 + %6119 = bitcast i32 %6118 to float, !dbg !38 + %6120 = fadd float %6116, %6119, !dbg !35 + %6121 = bitcast float %6120 to i32, !dbg !38 + %6122 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6121, i32 1, i32 31), !dbg !38 + %6123 = bitcast i32 %6122 to float, !dbg !38 + %6124 = fadd float %6120, %6123, !dbg !35 + %6125 = bitcast float %5287 to i32, !dbg !38 + %6126 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6125, i32 8, i32 31), !dbg !38 + %6127 = bitcast i32 %6126 to float, !dbg !38 + %6128 = fadd float %5287, %6127, !dbg !35 + %6129 = bitcast float %6128 to i32, !dbg !38 + %6130 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6129, i32 4, i32 31), !dbg !38 + %6131 = bitcast i32 %6130 to float, !dbg !38 + %6132 = fadd float %6128, %6131, !dbg !35 + %6133 = bitcast float %6132 to i32, !dbg !38 + %6134 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6133, i32 2, i32 31), !dbg !38 + %6135 = bitcast i32 %6134 to float, !dbg !38 + %6136 = fadd float %6132, %6135, !dbg !35 + %6137 = bitcast float %6136 to i32, !dbg !38 + %6138 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6137, i32 1, i32 31), !dbg !38 + %6139 = bitcast i32 %6138 to float, !dbg !38 + %6140 = fadd float %6136, %6139, !dbg !35 + %6141 = bitcast float %5294 to i32, !dbg !38 + %6142 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6141, i32 8, i32 31), !dbg !38 + %6143 = bitcast i32 %6142 to float, !dbg !38 + %6144 = fadd float %5294, %6143, !dbg !35 + %6145 = bitcast float %6144 to i32, !dbg !38 + %6146 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6145, i32 4, i32 31), !dbg !38 + %6147 = bitcast i32 %6146 to float, !dbg !38 + %6148 = fadd float %6144, %6147, !dbg !35 + %6149 = bitcast float %6148 to i32, !dbg !38 + %6150 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6149, i32 2, i32 31), !dbg !38 + %6151 = bitcast i32 %6150 to float, !dbg !38 + %6152 = fadd float %6148, %6151, !dbg !35 + %6153 = bitcast float %6152 to i32, !dbg !38 + %6154 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6153, i32 1, i32 31), !dbg !38 + %6155 = bitcast i32 %6154 to float, !dbg !38 + %6156 = fadd float %6152, %6155, !dbg !35 + %6157 = bitcast float %5301 to i32, !dbg !38 + %6158 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6157, i32 8, i32 31), !dbg !38 + %6159 = bitcast i32 %6158 to float, !dbg !38 + %6160 = fadd float %5301, %6159, !dbg !35 + %6161 = bitcast float %6160 to i32, !dbg !38 + %6162 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6161, i32 4, i32 31), !dbg !38 + %6163 = bitcast i32 %6162 to float, !dbg !38 + %6164 = fadd float %6160, %6163, !dbg !35 + %6165 = bitcast float %6164 to i32, !dbg !38 + %6166 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6165, i32 2, i32 31), !dbg !38 + %6167 = bitcast i32 %6166 to float, !dbg !38 + %6168 = fadd float %6164, %6167, !dbg !35 + %6169 = bitcast float %6168 to i32, !dbg !38 + %6170 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6169, i32 1, i32 31), !dbg !38 + %6171 = bitcast i32 %6170 to float, !dbg !38 + %6172 = fadd float %6168, %6171, !dbg !35 + %6173 = bitcast float %5308 to i32, !dbg !38 + %6174 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6173, i32 8, i32 31), !dbg !38 + %6175 = bitcast i32 %6174 to float, !dbg !38 + %6176 = fadd float %5308, %6175, !dbg !35 + %6177 = bitcast float %6176 to i32, !dbg !38 + %6178 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6177, i32 4, i32 31), !dbg !38 + %6179 = bitcast i32 %6178 to float, !dbg !38 + %6180 = fadd float %6176, %6179, !dbg !35 + %6181 = bitcast float %6180 to i32, !dbg !38 + %6182 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6181, i32 2, i32 31), !dbg !38 + %6183 = bitcast i32 %6182 to float, !dbg !38 + %6184 = fadd float %6180, %6183, !dbg !35 + %6185 = bitcast float %6184 to i32, !dbg !38 + %6186 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6185, i32 1, i32 31), !dbg !38 + %6187 = bitcast i32 %6186 to float, !dbg !38 + %6188 = fadd float %6184, %6187, !dbg !35 + %6189 = bitcast float %5315 to i32, !dbg !38 + %6190 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6189, i32 8, i32 31), !dbg !38 + %6191 = bitcast i32 %6190 to float, !dbg !38 + %6192 = fadd float %5315, %6191, !dbg !35 + %6193 = bitcast float %6192 to i32, !dbg !38 + %6194 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6193, i32 4, i32 31), !dbg !38 + %6195 = bitcast i32 %6194 to float, !dbg !38 + %6196 = fadd float %6192, %6195, !dbg !35 + %6197 = bitcast float %6196 to i32, !dbg !38 + %6198 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6197, i32 2, i32 31), !dbg !38 + %6199 = bitcast i32 %6198 to float, !dbg !38 + %6200 = fadd float %6196, %6199, !dbg !35 + %6201 = bitcast float %6200 to i32, !dbg !38 + %6202 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6201, i32 1, i32 31), !dbg !38 + %6203 = bitcast i32 %6202 to float, !dbg !38 + %6204 = fadd float %6200, %6203, !dbg !35 + %6205 = bitcast float %5322 to i32, !dbg !38 + %6206 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6205, i32 8, i32 31), !dbg !38 + %6207 = bitcast i32 %6206 to float, !dbg !38 + %6208 = fadd float %5322, %6207, !dbg !35 + %6209 = bitcast float %6208 to i32, !dbg !38 + %6210 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6209, i32 4, i32 31), !dbg !38 + %6211 = bitcast i32 %6210 to float, !dbg !38 + %6212 = fadd float %6208, %6211, !dbg !35 + %6213 = bitcast float %6212 to i32, !dbg !38 + %6214 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6213, i32 2, i32 31), !dbg !38 + %6215 = bitcast i32 %6214 to float, !dbg !38 + %6216 = fadd float %6212, %6215, !dbg !35 + %6217 = bitcast float %6216 to i32, !dbg !38 + %6218 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6217, i32 1, i32 31), !dbg !38 + %6219 = bitcast i32 %6218 to float, !dbg !38 + %6220 = fadd float %6216, %6219, !dbg !35 + %6221 = bitcast float %5329 to i32, !dbg !38 + %6222 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6221, i32 8, i32 31), !dbg !38 + %6223 = bitcast i32 %6222 to float, !dbg !38 + %6224 = fadd float %5329, %6223, !dbg !35 + %6225 = bitcast float %6224 to i32, !dbg !38 + %6226 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6225, i32 4, i32 31), !dbg !38 + %6227 = bitcast i32 %6226 to float, !dbg !38 + %6228 = fadd float %6224, %6227, !dbg !35 + %6229 = bitcast float %6228 to i32, !dbg !38 + %6230 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6229, i32 2, i32 31), !dbg !38 + %6231 = bitcast i32 %6230 to float, !dbg !38 + %6232 = fadd float %6228, %6231, !dbg !35 + %6233 = bitcast float %6232 to i32, !dbg !38 + %6234 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6233, i32 1, i32 31), !dbg !38 + %6235 = bitcast i32 %6234 to float, !dbg !38 + %6236 = fadd float %6232, %6235, !dbg !35 + %6237 = bitcast float %5336 to i32, !dbg !38 + %6238 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6237, i32 8, i32 31), !dbg !38 + %6239 = bitcast i32 %6238 to float, !dbg !38 + %6240 = fadd float %5336, %6239, !dbg !35 + %6241 = bitcast float %6240 to i32, !dbg !38 + %6242 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6241, i32 4, i32 31), !dbg !38 + %6243 = bitcast i32 %6242 to float, !dbg !38 + %6244 = fadd float %6240, %6243, !dbg !35 + %6245 = bitcast float %6244 to i32, !dbg !38 + %6246 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6245, i32 2, i32 31), !dbg !38 + %6247 = bitcast i32 %6246 to float, !dbg !38 + %6248 = fadd float %6244, %6247, !dbg !35 + %6249 = bitcast float %6248 to i32, !dbg !38 + %6250 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6249, i32 1, i32 31), !dbg !38 + %6251 = bitcast i32 %6250 to float, !dbg !38 + %6252 = fadd float %6248, %6251, !dbg !35 + %6253 = bitcast float %5343 to i32, !dbg !38 + %6254 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6253, i32 8, i32 31), !dbg !38 + %6255 = bitcast i32 %6254 to float, !dbg !38 + %6256 = fadd float %5343, %6255, !dbg !35 + %6257 = bitcast float %6256 to i32, !dbg !38 + %6258 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6257, i32 4, i32 31), !dbg !38 + %6259 = bitcast i32 %6258 to float, !dbg !38 + %6260 = fadd float %6256, %6259, !dbg !35 + %6261 = bitcast float %6260 to i32, !dbg !38 + %6262 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6261, i32 2, i32 31), !dbg !38 + %6263 = bitcast i32 %6262 to float, !dbg !38 + %6264 = fadd float %6260, %6263, !dbg !35 + %6265 = bitcast float %6264 to i32, !dbg !38 + %6266 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6265, i32 1, i32 31), !dbg !38 + %6267 = bitcast i32 %6266 to float, !dbg !38 + %6268 = fadd float %6264, %6267, !dbg !35 + %6269 = bitcast float %5350 to i32, !dbg !38 + %6270 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6269, i32 8, i32 31), !dbg !38 + %6271 = bitcast i32 %6270 to float, !dbg !38 + %6272 = fadd float %5350, %6271, !dbg !35 + %6273 = bitcast float %6272 to i32, !dbg !38 + %6274 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6273, i32 4, i32 31), !dbg !38 + %6275 = bitcast i32 %6274 to float, !dbg !38 + %6276 = fadd float %6272, %6275, !dbg !35 + %6277 = bitcast float %6276 to i32, !dbg !38 + %6278 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6277, i32 2, i32 31), !dbg !38 + %6279 = bitcast i32 %6278 to float, !dbg !38 + %6280 = fadd float %6276, %6279, !dbg !35 + %6281 = bitcast float %6280 to i32, !dbg !38 + %6282 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6281, i32 1, i32 31), !dbg !38 + %6283 = bitcast i32 %6282 to float, !dbg !38 + %6284 = fadd float %6280, %6283, !dbg !35 + %6285 = bitcast float %5357 to i32, !dbg !38 + %6286 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6285, i32 8, i32 31), !dbg !38 + %6287 = bitcast i32 %6286 to float, !dbg !38 + %6288 = fadd float %5357, %6287, !dbg !35 + %6289 = bitcast float %6288 to i32, !dbg !38 + %6290 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6289, i32 4, i32 31), !dbg !38 + %6291 = bitcast i32 %6290 to float, !dbg !38 + %6292 = fadd float %6288, %6291, !dbg !35 + %6293 = bitcast float %6292 to i32, !dbg !38 + %6294 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6293, i32 2, i32 31), !dbg !38 + %6295 = bitcast i32 %6294 to float, !dbg !38 + %6296 = fadd float %6292, %6295, !dbg !35 + %6297 = bitcast float %6296 to i32, !dbg !38 + %6298 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6297, i32 1, i32 31), !dbg !38 + %6299 = bitcast i32 %6298 to float, !dbg !38 + %6300 = fadd float %6296, %6299, !dbg !35 + %6301 = bitcast float %5364 to i32, !dbg !38 + %6302 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6301, i32 8, i32 31), !dbg !38 + %6303 = bitcast i32 %6302 to float, !dbg !38 + %6304 = fadd float %5364, %6303, !dbg !35 + %6305 = bitcast float %6304 to i32, !dbg !38 + %6306 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6305, i32 4, i32 31), !dbg !38 + %6307 = bitcast i32 %6306 to float, !dbg !38 + %6308 = fadd float %6304, %6307, !dbg !35 + %6309 = bitcast float %6308 to i32, !dbg !38 + %6310 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6309, i32 2, i32 31), !dbg !38 + %6311 = bitcast i32 %6310 to float, !dbg !38 + %6312 = fadd float %6308, %6311, !dbg !35 + %6313 = bitcast float %6312 to i32, !dbg !38 + %6314 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6313, i32 1, i32 31), !dbg !38 + %6315 = bitcast i32 %6314 to float, !dbg !38 + %6316 = fadd float %6312, %6315, !dbg !35 + %6317 = bitcast float %5371 to i32, !dbg !38 + %6318 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6317, i32 8, i32 31), !dbg !38 + %6319 = bitcast i32 %6318 to float, !dbg !38 + %6320 = fadd float %5371, %6319, !dbg !35 + %6321 = bitcast float %6320 to i32, !dbg !38 + %6322 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6321, i32 4, i32 31), !dbg !38 + %6323 = bitcast i32 %6322 to float, !dbg !38 + %6324 = fadd float %6320, %6323, !dbg !35 + %6325 = bitcast float %6324 to i32, !dbg !38 + %6326 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6325, i32 2, i32 31), !dbg !38 + %6327 = bitcast i32 %6326 to float, !dbg !38 + %6328 = fadd float %6324, %6327, !dbg !35 + %6329 = bitcast float %6328 to i32, !dbg !38 + %6330 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6329, i32 1, i32 31), !dbg !38 + %6331 = bitcast i32 %6330 to float, !dbg !38 + %6332 = fadd float %6328, %6331, !dbg !35 + %6333 = bitcast float %5378 to i32, !dbg !38 + %6334 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6333, i32 8, i32 31), !dbg !38 + %6335 = bitcast i32 %6334 to float, !dbg !38 + %6336 = fadd float %5378, %6335, !dbg !35 + %6337 = bitcast float %6336 to i32, !dbg !38 + %6338 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6337, i32 4, i32 31), !dbg !38 + %6339 = bitcast i32 %6338 to float, !dbg !38 + %6340 = fadd float %6336, %6339, !dbg !35 + %6341 = bitcast float %6340 to i32, !dbg !38 + %6342 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6341, i32 2, i32 31), !dbg !38 + %6343 = bitcast i32 %6342 to float, !dbg !38 + %6344 = fadd float %6340, %6343, !dbg !35 + %6345 = bitcast float %6344 to i32, !dbg !38 + %6346 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6345, i32 1, i32 31), !dbg !38 + %6347 = bitcast i32 %6346 to float, !dbg !38 + %6348 = fadd float %6344, %6347, !dbg !35 + %6349 = bitcast float %5385 to i32, !dbg !38 + %6350 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6349, i32 8, i32 31), !dbg !38 + %6351 = bitcast i32 %6350 to float, !dbg !38 + %6352 = fadd float %5385, %6351, !dbg !35 + %6353 = bitcast float %6352 to i32, !dbg !38 + %6354 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6353, i32 4, i32 31), !dbg !38 + %6355 = bitcast i32 %6354 to float, !dbg !38 + %6356 = fadd float %6352, %6355, !dbg !35 + %6357 = bitcast float %6356 to i32, !dbg !38 + %6358 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6357, i32 2, i32 31), !dbg !38 + %6359 = bitcast i32 %6358 to float, !dbg !38 + %6360 = fadd float %6356, %6359, !dbg !35 + %6361 = bitcast float %6360 to i32, !dbg !38 + %6362 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6361, i32 1, i32 31), !dbg !38 + %6363 = bitcast i32 %6362 to float, !dbg !38 + %6364 = fadd float %6360, %6363, !dbg !35 + %6365 = bitcast float %5392 to i32, !dbg !38 + %6366 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6365, i32 8, i32 31), !dbg !38 + %6367 = bitcast i32 %6366 to float, !dbg !38 + %6368 = fadd float %5392, %6367, !dbg !35 + %6369 = bitcast float %6368 to i32, !dbg !38 + %6370 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6369, i32 4, i32 31), !dbg !38 + %6371 = bitcast i32 %6370 to float, !dbg !38 + %6372 = fadd float %6368, %6371, !dbg !35 + %6373 = bitcast float %6372 to i32, !dbg !38 + %6374 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6373, i32 2, i32 31), !dbg !38 + %6375 = bitcast i32 %6374 to float, !dbg !38 + %6376 = fadd float %6372, %6375, !dbg !35 + %6377 = bitcast float %6376 to i32, !dbg !38 + %6378 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6377, i32 1, i32 31), !dbg !38 + %6379 = bitcast i32 %6378 to float, !dbg !38 + %6380 = fadd float %6376, %6379, !dbg !35 + %6381 = bitcast float %5399 to i32, !dbg !38 + %6382 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6381, i32 8, i32 31), !dbg !38 + %6383 = bitcast i32 %6382 to float, !dbg !38 + %6384 = fadd float %5399, %6383, !dbg !35 + %6385 = bitcast float %6384 to i32, !dbg !38 + %6386 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6385, i32 4, i32 31), !dbg !38 + %6387 = bitcast i32 %6386 to float, !dbg !38 + %6388 = fadd float %6384, %6387, !dbg !35 + %6389 = bitcast float %6388 to i32, !dbg !38 + %6390 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6389, i32 2, i32 31), !dbg !38 + %6391 = bitcast i32 %6390 to float, !dbg !38 + %6392 = fadd float %6388, %6391, !dbg !35 + %6393 = bitcast float %6392 to i32, !dbg !38 + %6394 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6393, i32 1, i32 31), !dbg !38 + %6395 = bitcast i32 %6394 to float, !dbg !38 + %6396 = fadd float %6392, %6395, !dbg !35 + %6397 = bitcast float %5406 to i32, !dbg !38 + %6398 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6397, i32 8, i32 31), !dbg !38 + %6399 = bitcast i32 %6398 to float, !dbg !38 + %6400 = fadd float %5406, %6399, !dbg !35 + %6401 = bitcast float %6400 to i32, !dbg !38 + %6402 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6401, i32 4, i32 31), !dbg !38 + %6403 = bitcast i32 %6402 to float, !dbg !38 + %6404 = fadd float %6400, %6403, !dbg !35 + %6405 = bitcast float %6404 to i32, !dbg !38 + %6406 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6405, i32 2, i32 31), !dbg !38 + %6407 = bitcast i32 %6406 to float, !dbg !38 + %6408 = fadd float %6404, %6407, !dbg !35 + %6409 = bitcast float %6408 to i32, !dbg !38 + %6410 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6409, i32 1, i32 31), !dbg !38 + %6411 = bitcast i32 %6410 to float, !dbg !38 + %6412 = fadd float %6408, %6411, !dbg !35 + %6413 = bitcast float %5413 to i32, !dbg !38 + %6414 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6413, i32 8, i32 31), !dbg !38 + %6415 = bitcast i32 %6414 to float, !dbg !38 + %6416 = fadd float %5413, %6415, !dbg !35 + %6417 = bitcast float %6416 to i32, !dbg !38 + %6418 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6417, i32 4, i32 31), !dbg !38 + %6419 = bitcast i32 %6418 to float, !dbg !38 + %6420 = fadd float %6416, %6419, !dbg !35 + %6421 = bitcast float %6420 to i32, !dbg !38 + %6422 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6421, i32 2, i32 31), !dbg !38 + %6423 = bitcast i32 %6422 to float, !dbg !38 + %6424 = fadd float %6420, %6423, !dbg !35 + %6425 = bitcast float %6424 to i32, !dbg !38 + %6426 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6425, i32 1, i32 31), !dbg !38 + %6427 = bitcast i32 %6426 to float, !dbg !38 + %6428 = fadd float %6424, %6427, !dbg !35 + %6429 = bitcast float %5420 to i32, !dbg !38 + %6430 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6429, i32 8, i32 31), !dbg !38 + %6431 = bitcast i32 %6430 to float, !dbg !38 + %6432 = fadd float %5420, %6431, !dbg !35 + %6433 = bitcast float %6432 to i32, !dbg !38 + %6434 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6433, i32 4, i32 31), !dbg !38 + %6435 = bitcast i32 %6434 to float, !dbg !38 + %6436 = fadd float %6432, %6435, !dbg !35 + %6437 = bitcast float %6436 to i32, !dbg !38 + %6438 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6437, i32 2, i32 31), !dbg !38 + %6439 = bitcast i32 %6438 to float, !dbg !38 + %6440 = fadd float %6436, %6439, !dbg !35 + %6441 = bitcast float %6440 to i32, !dbg !38 + %6442 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6441, i32 1, i32 31), !dbg !38 + %6443 = bitcast i32 %6442 to float, !dbg !38 + %6444 = fadd float %6440, %6443, !dbg !35 + %6445 = fmul float %5436, %1730, !dbg !41 + %6446 = fmul float %5436, %1731, !dbg !41 + %6447 = fmul float %5436, %1732, !dbg !41 + %6448 = fmul float %5436, %1733, !dbg !41 + %6449 = fmul float %5436, %1734, !dbg !41 + %6450 = fmul float %5436, %1735, !dbg !41 + %6451 = fmul float %5436, %1736, !dbg !41 + %6452 = fmul float %5436, %1737, !dbg !41 + %6453 = fmul float %5452, %1738, !dbg !41 + %6454 = fmul float %5452, %1739, !dbg !41 + %6455 = fmul float %5452, %1740, !dbg !41 + %6456 = fmul float %5452, %1741, !dbg !41 + %6457 = fmul float %5452, %1742, !dbg !41 + %6458 = fmul float %5452, %1743, !dbg !41 + %6459 = fmul float %5452, %1744, !dbg !41 + %6460 = fmul float %5452, %1745, !dbg !41 + %6461 = fmul float %5468, %1746, !dbg !41 + %6462 = fmul float %5468, %1747, !dbg !41 + %6463 = fmul float %5468, %1748, !dbg !41 + %6464 = fmul float %5468, %1749, !dbg !41 + %6465 = fmul float %5468, %1750, !dbg !41 + %6466 = fmul float %5468, %1751, !dbg !41 + %6467 = fmul float %5468, %1752, !dbg !41 + %6468 = fmul float %5468, %1753, !dbg !41 + %6469 = fmul float %5484, %1754, !dbg !41 + %6470 = fmul float %5484, %1755, !dbg !41 + %6471 = fmul float %5484, %1756, !dbg !41 + %6472 = fmul float %5484, %1757, !dbg !41 + %6473 = fmul float %5484, %1758, !dbg !41 + %6474 = fmul float %5484, %1759, !dbg !41 + %6475 = fmul float %5484, %1760, !dbg !41 + %6476 = fmul float %5484, %1761, !dbg !41 + %6477 = fmul float %5500, %1762, !dbg !41 + %6478 = fmul float %5500, %1763, !dbg !41 + %6479 = fmul float %5500, %1764, !dbg !41 + %6480 = fmul float %5500, %1765, !dbg !41 + %6481 = fmul float %5500, %1766, !dbg !41 + %6482 = fmul float %5500, %1767, !dbg !41 + %6483 = fmul float %5500, %1768, !dbg !41 + %6484 = fmul float %5500, %1769, !dbg !41 + %6485 = fmul float %5516, %1770, !dbg !41 + %6486 = fmul float %5516, %1771, !dbg !41 + %6487 = fmul float %5516, %1772, !dbg !41 + %6488 = fmul float %5516, %1773, !dbg !41 + %6489 = fmul float %5516, %1774, !dbg !41 + %6490 = fmul float %5516, %1775, !dbg !41 + %6491 = fmul float %5516, %1776, !dbg !41 + %6492 = fmul float %5516, %1777, !dbg !41 + %6493 = fmul float %5532, %1778, !dbg !41 + %6494 = fmul float %5532, %1779, !dbg !41 + %6495 = fmul float %5532, %1780, !dbg !41 + %6496 = fmul float %5532, %1781, !dbg !41 + %6497 = fmul float %5532, %1782, !dbg !41 + %6498 = fmul float %5532, %1783, !dbg !41 + %6499 = fmul float %5532, %1784, !dbg !41 + %6500 = fmul float %5532, %1785, !dbg !41 + %6501 = fmul float %5548, %1786, !dbg !41 + %6502 = fmul float %5548, %1787, !dbg !41 + %6503 = fmul float %5548, %1788, !dbg !41 + %6504 = fmul float %5548, %1789, !dbg !41 + %6505 = fmul float %5548, %1790, !dbg !41 + %6506 = fmul float %5548, %1791, !dbg !41 + %6507 = fmul float %5548, %1792, !dbg !41 + %6508 = fmul float %5548, %1793, !dbg !41 + %6509 = fmul float %5564, %1794, !dbg !41 + %6510 = fmul float %5564, %1795, !dbg !41 + %6511 = fmul float %5564, %1796, !dbg !41 + %6512 = fmul float %5564, %1797, !dbg !41 + %6513 = fmul float %5564, %1798, !dbg !41 + %6514 = fmul float %5564, %1799, !dbg !41 + %6515 = fmul float %5564, %1800, !dbg !41 + %6516 = fmul float %5564, %1801, !dbg !41 + %6517 = fmul float %5580, %1802, !dbg !41 + %6518 = fmul float %5580, %1803, !dbg !41 + %6519 = fmul float %5580, %1804, !dbg !41 + %6520 = fmul float %5580, %1805, !dbg !41 + %6521 = fmul float %5580, %1806, !dbg !41 + %6522 = fmul float %5580, %1807, !dbg !41 + %6523 = fmul float %5580, %1808, !dbg !41 + %6524 = fmul float %5580, %1809, !dbg !41 + %6525 = fmul float %5596, %1810, !dbg !41 + %6526 = fmul float %5596, %1811, !dbg !41 + %6527 = fmul float %5596, %1812, !dbg !41 + %6528 = fmul float %5596, %1813, !dbg !41 + %6529 = fmul float %5596, %1814, !dbg !41 + %6530 = fmul float %5596, %1815, !dbg !41 + %6531 = fmul float %5596, %1816, !dbg !41 + %6532 = fmul float %5596, %1817, !dbg !41 + %6533 = fmul float %5612, %1818, !dbg !41 + %6534 = fmul float %5612, %1819, !dbg !41 + %6535 = fmul float %5612, %1820, !dbg !41 + %6536 = fmul float %5612, %1821, !dbg !41 + %6537 = fmul float %5612, %1822, !dbg !41 + %6538 = fmul float %5612, %1823, !dbg !41 + %6539 = fmul float %5612, %1824, !dbg !41 + %6540 = fmul float %5612, %1825, !dbg !41 + %6541 = fmul float %5628, %1826, !dbg !41 + %6542 = fmul float %5628, %1827, !dbg !41 + %6543 = fmul float %5628, %1828, !dbg !41 + %6544 = fmul float %5628, %1829, !dbg !41 + %6545 = fmul float %5628, %1830, !dbg !41 + %6546 = fmul float %5628, %1831, !dbg !41 + %6547 = fmul float %5628, %1832, !dbg !41 + %6548 = fmul float %5628, %1833, !dbg !41 + %6549 = fmul float %5644, %1834, !dbg !41 + %6550 = fmul float %5644, %1835, !dbg !41 + %6551 = fmul float %5644, %1836, !dbg !41 + %6552 = fmul float %5644, %1837, !dbg !41 + %6553 = fmul float %5644, %1838, !dbg !41 + %6554 = fmul float %5644, %1839, !dbg !41 + %6555 = fmul float %5644, %1840, !dbg !41 + %6556 = fmul float %5644, %1841, !dbg !41 + %6557 = fmul float %5660, %1842, !dbg !41 + %6558 = fmul float %5660, %1843, !dbg !41 + %6559 = fmul float %5660, %1844, !dbg !41 + %6560 = fmul float %5660, %1845, !dbg !41 + %6561 = fmul float %5660, %1846, !dbg !41 + %6562 = fmul float %5660, %1847, !dbg !41 + %6563 = fmul float %5660, %1848, !dbg !41 + %6564 = fmul float %5660, %1849, !dbg !41 + %6565 = fmul float %5676, %1850, !dbg !41 + %6566 = fmul float %5676, %1851, !dbg !41 + %6567 = fmul float %5676, %1852, !dbg !41 + %6568 = fmul float %5676, %1853, !dbg !41 + %6569 = fmul float %5676, %1854, !dbg !41 + %6570 = fmul float %5676, %1855, !dbg !41 + %6571 = fmul float %5676, %1856, !dbg !41 + %6572 = fmul float %5676, %1857, !dbg !41 + %6573 = fmul float %5692, %1858, !dbg !41 + %6574 = fmul float %5692, %1859, !dbg !41 + %6575 = fmul float %5692, %1860, !dbg !41 + %6576 = fmul float %5692, %1861, !dbg !41 + %6577 = fmul float %5692, %1862, !dbg !41 + %6578 = fmul float %5692, %1863, !dbg !41 + %6579 = fmul float %5692, %1864, !dbg !41 + %6580 = fmul float %5692, %1865, !dbg !41 + %6581 = fmul float %5708, %1866, !dbg !41 + %6582 = fmul float %5708, %1867, !dbg !41 + %6583 = fmul float %5708, %1868, !dbg !41 + %6584 = fmul float %5708, %1869, !dbg !41 + %6585 = fmul float %5708, %1870, !dbg !41 + %6586 = fmul float %5708, %1871, !dbg !41 + %6587 = fmul float %5708, %1872, !dbg !41 + %6588 = fmul float %5708, %1873, !dbg !41 + %6589 = fmul float %5724, %1874, !dbg !41 + %6590 = fmul float %5724, %1875, !dbg !41 + %6591 = fmul float %5724, %1876, !dbg !41 + %6592 = fmul float %5724, %1877, !dbg !41 + %6593 = fmul float %5724, %1878, !dbg !41 + %6594 = fmul float %5724, %1879, !dbg !41 + %6595 = fmul float %5724, %1880, !dbg !41 + %6596 = fmul float %5724, %1881, !dbg !41 + %6597 = fmul float %5740, %1882, !dbg !41 + %6598 = fmul float %5740, %1883, !dbg !41 + %6599 = fmul float %5740, %1884, !dbg !41 + %6600 = fmul float %5740, %1885, !dbg !41 + %6601 = fmul float %5740, %1886, !dbg !41 + %6602 = fmul float %5740, %1887, !dbg !41 + %6603 = fmul float %5740, %1888, !dbg !41 + %6604 = fmul float %5740, %1889, !dbg !41 + %6605 = fmul float %5756, %1890, !dbg !41 + %6606 = fmul float %5756, %1891, !dbg !41 + %6607 = fmul float %5756, %1892, !dbg !41 + %6608 = fmul float %5756, %1893, !dbg !41 + %6609 = fmul float %5756, %1894, !dbg !41 + %6610 = fmul float %5756, %1895, !dbg !41 + %6611 = fmul float %5756, %1896, !dbg !41 + %6612 = fmul float %5756, %1897, !dbg !41 + %6613 = fmul float %5772, %1898, !dbg !41 + %6614 = fmul float %5772, %1899, !dbg !41 + %6615 = fmul float %5772, %1900, !dbg !41 + %6616 = fmul float %5772, %1901, !dbg !41 + %6617 = fmul float %5772, %1902, !dbg !41 + %6618 = fmul float %5772, %1903, !dbg !41 + %6619 = fmul float %5772, %1904, !dbg !41 + %6620 = fmul float %5772, %1905, !dbg !41 + %6621 = fmul float %5788, %1906, !dbg !41 + %6622 = fmul float %5788, %1907, !dbg !41 + %6623 = fmul float %5788, %1908, !dbg !41 + %6624 = fmul float %5788, %1909, !dbg !41 + %6625 = fmul float %5788, %1910, !dbg !41 + %6626 = fmul float %5788, %1911, !dbg !41 + %6627 = fmul float %5788, %1912, !dbg !41 + %6628 = fmul float %5788, %1913, !dbg !41 + %6629 = fmul float %5804, %1914, !dbg !41 + %6630 = fmul float %5804, %1915, !dbg !41 + %6631 = fmul float %5804, %1916, !dbg !41 + %6632 = fmul float %5804, %1917, !dbg !41 + %6633 = fmul float %5804, %1918, !dbg !41 + %6634 = fmul float %5804, %1919, !dbg !41 + %6635 = fmul float %5804, %1920, !dbg !41 + %6636 = fmul float %5804, %1921, !dbg !41 + %6637 = fmul float %5820, %1922, !dbg !41 + %6638 = fmul float %5820, %1923, !dbg !41 + %6639 = fmul float %5820, %1924, !dbg !41 + %6640 = fmul float %5820, %1925, !dbg !41 + %6641 = fmul float %5820, %1926, !dbg !41 + %6642 = fmul float %5820, %1927, !dbg !41 + %6643 = fmul float %5820, %1928, !dbg !41 + %6644 = fmul float %5820, %1929, !dbg !41 + %6645 = fmul float %5836, %1930, !dbg !41 + %6646 = fmul float %5836, %1931, !dbg !41 + %6647 = fmul float %5836, %1932, !dbg !41 + %6648 = fmul float %5836, %1933, !dbg !41 + %6649 = fmul float %5836, %1934, !dbg !41 + %6650 = fmul float %5836, %1935, !dbg !41 + %6651 = fmul float %5836, %1936, !dbg !41 + %6652 = fmul float %5836, %1937, !dbg !41 + %6653 = fmul float %5852, %1938, !dbg !41 + %6654 = fmul float %5852, %1939, !dbg !41 + %6655 = fmul float %5852, %1940, !dbg !41 + %6656 = fmul float %5852, %1941, !dbg !41 + %6657 = fmul float %5852, %1942, !dbg !41 + %6658 = fmul float %5852, %1943, !dbg !41 + %6659 = fmul float %5852, %1944, !dbg !41 + %6660 = fmul float %5852, %1945, !dbg !41 + %6661 = fmul float %5868, %1946, !dbg !41 + %6662 = fmul float %5868, %1947, !dbg !41 + %6663 = fmul float %5868, %1948, !dbg !41 + %6664 = fmul float %5868, %1949, !dbg !41 + %6665 = fmul float %5868, %1950, !dbg !41 + %6666 = fmul float %5868, %1951, !dbg !41 + %6667 = fmul float %5868, %1952, !dbg !41 + %6668 = fmul float %5868, %1953, !dbg !41 + %6669 = fmul float %5884, %1954, !dbg !41 + %6670 = fmul float %5884, %1955, !dbg !41 + %6671 = fmul float %5884, %1956, !dbg !41 + %6672 = fmul float %5884, %1957, !dbg !41 + %6673 = fmul float %5884, %1958, !dbg !41 + %6674 = fmul float %5884, %1959, !dbg !41 + %6675 = fmul float %5884, %1960, !dbg !41 + %6676 = fmul float %5884, %1961, !dbg !41 + %6677 = fmul float %5900, %1962, !dbg !41 + %6678 = fmul float %5900, %1963, !dbg !41 + %6679 = fmul float %5900, %1964, !dbg !41 + %6680 = fmul float %5900, %1965, !dbg !41 + %6681 = fmul float %5900, %1966, !dbg !41 + %6682 = fmul float %5900, %1967, !dbg !41 + %6683 = fmul float %5900, %1968, !dbg !41 + %6684 = fmul float %5900, %1969, !dbg !41 + %6685 = fmul float %5916, %1970, !dbg !41 + %6686 = fmul float %5916, %1971, !dbg !41 + %6687 = fmul float %5916, %1972, !dbg !41 + %6688 = fmul float %5916, %1973, !dbg !41 + %6689 = fmul float %5916, %1974, !dbg !41 + %6690 = fmul float %5916, %1975, !dbg !41 + %6691 = fmul float %5916, %1976, !dbg !41 + %6692 = fmul float %5916, %1977, !dbg !41 + %6693 = fmul float %5932, %1978, !dbg !41 + %6694 = fmul float %5932, %1979, !dbg !41 + %6695 = fmul float %5932, %1980, !dbg !41 + %6696 = fmul float %5932, %1981, !dbg !41 + %6697 = fmul float %5932, %1982, !dbg !41 + %6698 = fmul float %5932, %1983, !dbg !41 + %6699 = fmul float %5932, %1984, !dbg !41 + %6700 = fmul float %5932, %1985, !dbg !41 + %6701 = fmul float %5948, %1986, !dbg !41 + %6702 = fmul float %5948, %1987, !dbg !41 + %6703 = fmul float %5948, %1988, !dbg !41 + %6704 = fmul float %5948, %1989, !dbg !41 + %6705 = fmul float %5948, %1990, !dbg !41 + %6706 = fmul float %5948, %1991, !dbg !41 + %6707 = fmul float %5948, %1992, !dbg !41 + %6708 = fmul float %5948, %1993, !dbg !41 + %6709 = fmul float %5964, %1994, !dbg !41 + %6710 = fmul float %5964, %1995, !dbg !41 + %6711 = fmul float %5964, %1996, !dbg !41 + %6712 = fmul float %5964, %1997, !dbg !41 + %6713 = fmul float %5964, %1998, !dbg !41 + %6714 = fmul float %5964, %1999, !dbg !41 + %6715 = fmul float %5964, %2000, !dbg !41 + %6716 = fmul float %5964, %2001, !dbg !41 + %6717 = fmul float %5980, %2002, !dbg !41 + %6718 = fmul float %5980, %2003, !dbg !41 + %6719 = fmul float %5980, %2004, !dbg !41 + %6720 = fmul float %5980, %2005, !dbg !41 + %6721 = fmul float %5980, %2006, !dbg !41 + %6722 = fmul float %5980, %2007, !dbg !41 + %6723 = fmul float %5980, %2008, !dbg !41 + %6724 = fmul float %5980, %2009, !dbg !41 + %6725 = fmul float %5996, %2010, !dbg !41 + %6726 = fmul float %5996, %2011, !dbg !41 + %6727 = fmul float %5996, %2012, !dbg !41 + %6728 = fmul float %5996, %2013, !dbg !41 + %6729 = fmul float %5996, %2014, !dbg !41 + %6730 = fmul float %5996, %2015, !dbg !41 + %6731 = fmul float %5996, %2016, !dbg !41 + %6732 = fmul float %5996, %2017, !dbg !41 + %6733 = fmul float %6012, %2018, !dbg !41 + %6734 = fmul float %6012, %2019, !dbg !41 + %6735 = fmul float %6012, %2020, !dbg !41 + %6736 = fmul float %6012, %2021, !dbg !41 + %6737 = fmul float %6012, %2022, !dbg !41 + %6738 = fmul float %6012, %2023, !dbg !41 + %6739 = fmul float %6012, %2024, !dbg !41 + %6740 = fmul float %6012, %2025, !dbg !41 + %6741 = fmul float %6028, %2026, !dbg !41 + %6742 = fmul float %6028, %2027, !dbg !41 + %6743 = fmul float %6028, %2028, !dbg !41 + %6744 = fmul float %6028, %2029, !dbg !41 + %6745 = fmul float %6028, %2030, !dbg !41 + %6746 = fmul float %6028, %2031, !dbg !41 + %6747 = fmul float %6028, %2032, !dbg !41 + %6748 = fmul float %6028, %2033, !dbg !41 + %6749 = fmul float %6044, %2034, !dbg !41 + %6750 = fmul float %6044, %2035, !dbg !41 + %6751 = fmul float %6044, %2036, !dbg !41 + %6752 = fmul float %6044, %2037, !dbg !41 + %6753 = fmul float %6044, %2038, !dbg !41 + %6754 = fmul float %6044, %2039, !dbg !41 + %6755 = fmul float %6044, %2040, !dbg !41 + %6756 = fmul float %6044, %2041, !dbg !41 + %6757 = fmul float %6060, %2042, !dbg !41 + %6758 = fmul float %6060, %2043, !dbg !41 + %6759 = fmul float %6060, %2044, !dbg !41 + %6760 = fmul float %6060, %2045, !dbg !41 + %6761 = fmul float %6060, %2046, !dbg !41 + %6762 = fmul float %6060, %2047, !dbg !41 + %6763 = fmul float %6060, %2048, !dbg !41 + %6764 = fmul float %6060, %2049, !dbg !41 + %6765 = fmul float %6076, %2050, !dbg !41 + %6766 = fmul float %6076, %2051, !dbg !41 + %6767 = fmul float %6076, %2052, !dbg !41 + %6768 = fmul float %6076, %2053, !dbg !41 + %6769 = fmul float %6076, %2054, !dbg !41 + %6770 = fmul float %6076, %2055, !dbg !41 + %6771 = fmul float %6076, %2056, !dbg !41 + %6772 = fmul float %6076, %2057, !dbg !41 + %6773 = fmul float %6092, %2058, !dbg !41 + %6774 = fmul float %6092, %2059, !dbg !41 + %6775 = fmul float %6092, %2060, !dbg !41 + %6776 = fmul float %6092, %2061, !dbg !41 + %6777 = fmul float %6092, %2062, !dbg !41 + %6778 = fmul float %6092, %2063, !dbg !41 + %6779 = fmul float %6092, %2064, !dbg !41 + %6780 = fmul float %6092, %2065, !dbg !41 + %6781 = fmul float %6108, %2066, !dbg !41 + %6782 = fmul float %6108, %2067, !dbg !41 + %6783 = fmul float %6108, %2068, !dbg !41 + %6784 = fmul float %6108, %2069, !dbg !41 + %6785 = fmul float %6108, %2070, !dbg !41 + %6786 = fmul float %6108, %2071, !dbg !41 + %6787 = fmul float %6108, %2072, !dbg !41 + %6788 = fmul float %6108, %2073, !dbg !41 + %6789 = fmul float %6124, %2074, !dbg !41 + %6790 = fmul float %6124, %2075, !dbg !41 + %6791 = fmul float %6124, %2076, !dbg !41 + %6792 = fmul float %6124, %2077, !dbg !41 + %6793 = fmul float %6124, %2078, !dbg !41 + %6794 = fmul float %6124, %2079, !dbg !41 + %6795 = fmul float %6124, %2080, !dbg !41 + %6796 = fmul float %6124, %2081, !dbg !41 + %6797 = fmul float %6140, %2082, !dbg !41 + %6798 = fmul float %6140, %2083, !dbg !41 + %6799 = fmul float %6140, %2084, !dbg !41 + %6800 = fmul float %6140, %2085, !dbg !41 + %6801 = fmul float %6140, %2086, !dbg !41 + %6802 = fmul float %6140, %2087, !dbg !41 + %6803 = fmul float %6140, %2088, !dbg !41 + %6804 = fmul float %6140, %2089, !dbg !41 + %6805 = fmul float %6156, %2090, !dbg !41 + %6806 = fmul float %6156, %2091, !dbg !41 + %6807 = fmul float %6156, %2092, !dbg !41 + %6808 = fmul float %6156, %2093, !dbg !41 + %6809 = fmul float %6156, %2094, !dbg !41 + %6810 = fmul float %6156, %2095, !dbg !41 + %6811 = fmul float %6156, %2096, !dbg !41 + %6812 = fmul float %6156, %2097, !dbg !41 + %6813 = fmul float %6172, %2098, !dbg !41 + %6814 = fmul float %6172, %2099, !dbg !41 + %6815 = fmul float %6172, %2100, !dbg !41 + %6816 = fmul float %6172, %2101, !dbg !41 + %6817 = fmul float %6172, %2102, !dbg !41 + %6818 = fmul float %6172, %2103, !dbg !41 + %6819 = fmul float %6172, %2104, !dbg !41 + %6820 = fmul float %6172, %2105, !dbg !41 + %6821 = fmul float %6188, %2106, !dbg !41 + %6822 = fmul float %6188, %2107, !dbg !41 + %6823 = fmul float %6188, %2108, !dbg !41 + %6824 = fmul float %6188, %2109, !dbg !41 + %6825 = fmul float %6188, %2110, !dbg !41 + %6826 = fmul float %6188, %2111, !dbg !41 + %6827 = fmul float %6188, %2112, !dbg !41 + %6828 = fmul float %6188, %2113, !dbg !41 + %6829 = fmul float %6204, %2114, !dbg !41 + %6830 = fmul float %6204, %2115, !dbg !41 + %6831 = fmul float %6204, %2116, !dbg !41 + %6832 = fmul float %6204, %2117, !dbg !41 + %6833 = fmul float %6204, %2118, !dbg !41 + %6834 = fmul float %6204, %2119, !dbg !41 + %6835 = fmul float %6204, %2120, !dbg !41 + %6836 = fmul float %6204, %2121, !dbg !41 + %6837 = fmul float %6220, %2122, !dbg !41 + %6838 = fmul float %6220, %2123, !dbg !41 + %6839 = fmul float %6220, %2124, !dbg !41 + %6840 = fmul float %6220, %2125, !dbg !41 + %6841 = fmul float %6220, %2126, !dbg !41 + %6842 = fmul float %6220, %2127, !dbg !41 + %6843 = fmul float %6220, %2128, !dbg !41 + %6844 = fmul float %6220, %2129, !dbg !41 + %6845 = fmul float %6236, %2130, !dbg !41 + %6846 = fmul float %6236, %2131, !dbg !41 + %6847 = fmul float %6236, %2132, !dbg !41 + %6848 = fmul float %6236, %2133, !dbg !41 + %6849 = fmul float %6236, %2134, !dbg !41 + %6850 = fmul float %6236, %2135, !dbg !41 + %6851 = fmul float %6236, %2136, !dbg !41 + %6852 = fmul float %6236, %2137, !dbg !41 + %6853 = fmul float %6252, %2138, !dbg !41 + %6854 = fmul float %6252, %2139, !dbg !41 + %6855 = fmul float %6252, %2140, !dbg !41 + %6856 = fmul float %6252, %2141, !dbg !41 + %6857 = fmul float %6252, %2142, !dbg !41 + %6858 = fmul float %6252, %2143, !dbg !41 + %6859 = fmul float %6252, %2144, !dbg !41 + %6860 = fmul float %6252, %2145, !dbg !41 + %6861 = fmul float %6268, %2146, !dbg !41 + %6862 = fmul float %6268, %2147, !dbg !41 + %6863 = fmul float %6268, %2148, !dbg !41 + %6864 = fmul float %6268, %2149, !dbg !41 + %6865 = fmul float %6268, %2150, !dbg !41 + %6866 = fmul float %6268, %2151, !dbg !41 + %6867 = fmul float %6268, %2152, !dbg !41 + %6868 = fmul float %6268, %2153, !dbg !41 + %6869 = fmul float %6284, %2154, !dbg !41 + %6870 = fmul float %6284, %2155, !dbg !41 + %6871 = fmul float %6284, %2156, !dbg !41 + %6872 = fmul float %6284, %2157, !dbg !41 + %6873 = fmul float %6284, %2158, !dbg !41 + %6874 = fmul float %6284, %2159, !dbg !41 + %6875 = fmul float %6284, %2160, !dbg !41 + %6876 = fmul float %6284, %2161, !dbg !41 + %6877 = fmul float %6300, %2162, !dbg !41 + %6878 = fmul float %6300, %2163, !dbg !41 + %6879 = fmul float %6300, %2164, !dbg !41 + %6880 = fmul float %6300, %2165, !dbg !41 + %6881 = fmul float %6300, %2166, !dbg !41 + %6882 = fmul float %6300, %2167, !dbg !41 + %6883 = fmul float %6300, %2168, !dbg !41 + %6884 = fmul float %6300, %2169, !dbg !41 + %6885 = fmul float %6316, %2170, !dbg !41 + %6886 = fmul float %6316, %2171, !dbg !41 + %6887 = fmul float %6316, %2172, !dbg !41 + %6888 = fmul float %6316, %2173, !dbg !41 + %6889 = fmul float %6316, %2174, !dbg !41 + %6890 = fmul float %6316, %2175, !dbg !41 + %6891 = fmul float %6316, %2176, !dbg !41 + %6892 = fmul float %6316, %2177, !dbg !41 + %6893 = fmul float %6332, %2178, !dbg !41 + %6894 = fmul float %6332, %2179, !dbg !41 + %6895 = fmul float %6332, %2180, !dbg !41 + %6896 = fmul float %6332, %2181, !dbg !41 + %6897 = fmul float %6332, %2182, !dbg !41 + %6898 = fmul float %6332, %2183, !dbg !41 + %6899 = fmul float %6332, %2184, !dbg !41 + %6900 = fmul float %6332, %2185, !dbg !41 + %6901 = fmul float %6348, %2186, !dbg !41 + %6902 = fmul float %6348, %2187, !dbg !41 + %6903 = fmul float %6348, %2188, !dbg !41 + %6904 = fmul float %6348, %2189, !dbg !41 + %6905 = fmul float %6348, %2190, !dbg !41 + %6906 = fmul float %6348, %2191, !dbg !41 + %6907 = fmul float %6348, %2192, !dbg !41 + %6908 = fmul float %6348, %2193, !dbg !41 + %6909 = fmul float %6364, %2194, !dbg !41 + %6910 = fmul float %6364, %2195, !dbg !41 + %6911 = fmul float %6364, %2196, !dbg !41 + %6912 = fmul float %6364, %2197, !dbg !41 + %6913 = fmul float %6364, %2198, !dbg !41 + %6914 = fmul float %6364, %2199, !dbg !41 + %6915 = fmul float %6364, %2200, !dbg !41 + %6916 = fmul float %6364, %2201, !dbg !41 + %6917 = fmul float %6380, %2202, !dbg !41 + %6918 = fmul float %6380, %2203, !dbg !41 + %6919 = fmul float %6380, %2204, !dbg !41 + %6920 = fmul float %6380, %2205, !dbg !41 + %6921 = fmul float %6380, %2206, !dbg !41 + %6922 = fmul float %6380, %2207, !dbg !41 + %6923 = fmul float %6380, %2208, !dbg !41 + %6924 = fmul float %6380, %2209, !dbg !41 + %6925 = fmul float %6396, %2210, !dbg !41 + %6926 = fmul float %6396, %2211, !dbg !41 + %6927 = fmul float %6396, %2212, !dbg !41 + %6928 = fmul float %6396, %2213, !dbg !41 + %6929 = fmul float %6396, %2214, !dbg !41 + %6930 = fmul float %6396, %2215, !dbg !41 + %6931 = fmul float %6396, %2216, !dbg !41 + %6932 = fmul float %6396, %2217, !dbg !41 + %6933 = fmul float %6412, %2218, !dbg !41 + %6934 = fmul float %6412, %2219, !dbg !41 + %6935 = fmul float %6412, %2220, !dbg !41 + %6936 = fmul float %6412, %2221, !dbg !41 + %6937 = fmul float %6412, %2222, !dbg !41 + %6938 = fmul float %6412, %2223, !dbg !41 + %6939 = fmul float %6412, %2224, !dbg !41 + %6940 = fmul float %6412, %2225, !dbg !41 + %6941 = fmul float %6428, %2226, !dbg !41 + %6942 = fmul float %6428, %2227, !dbg !41 + %6943 = fmul float %6428, %2228, !dbg !41 + %6944 = fmul float %6428, %2229, !dbg !41 + %6945 = fmul float %6428, %2230, !dbg !41 + %6946 = fmul float %6428, %2231, !dbg !41 + %6947 = fmul float %6428, %2232, !dbg !41 + %6948 = fmul float %6428, %2233, !dbg !41 + %6949 = fmul float %6444, %2234, !dbg !41 + %6950 = fmul float %6444, %2235, !dbg !41 + %6951 = fmul float %6444, %2236, !dbg !41 + %6952 = fmul float %6444, %2237, !dbg !41 + %6953 = fmul float %6444, %2238, !dbg !41 + %6954 = fmul float %6444, %2239, !dbg !41 + %6955 = fmul float %6444, %2240, !dbg !41 + %6956 = fmul float %6444, %2241, !dbg !41 + %6957 = fmul float %3854, %6445, !dbg !42 + %6958 = fmul float %3854, %6446, !dbg !42 + %6959 = fmul float %3854, %6447, !dbg !42 + %6960 = fmul float %3854, %6448, !dbg !42 + %6961 = fmul float %3854, %6449, !dbg !42 + %6962 = fmul float %3854, %6450, !dbg !42 + %6963 = fmul float %3854, %6451, !dbg !42 + %6964 = fmul float %3854, %6452, !dbg !42 + %6965 = fmul float %3917, %6453, !dbg !42 + %6966 = fmul float %3917, %6454, !dbg !42 + %6967 = fmul float %3917, %6455, !dbg !42 + %6968 = fmul float %3917, %6456, !dbg !42 + %6969 = fmul float %3917, %6457, !dbg !42 + %6970 = fmul float %3917, %6458, !dbg !42 + %6971 = fmul float %3917, %6459, !dbg !42 + %6972 = fmul float %3917, %6460, !dbg !42 + %6973 = fmul float %3856, %6461, !dbg !42 + %6974 = fmul float %3856, %6462, !dbg !42 + %6975 = fmul float %3856, %6463, !dbg !42 + %6976 = fmul float %3856, %6464, !dbg !42 + %6977 = fmul float %3856, %6465, !dbg !42 + %6978 = fmul float %3856, %6466, !dbg !42 + %6979 = fmul float %3856, %6467, !dbg !42 + %6980 = fmul float %3856, %6468, !dbg !42 + %6981 = fmul float %3918, %6469, !dbg !42 + %6982 = fmul float %3918, %6470, !dbg !42 + %6983 = fmul float %3918, %6471, !dbg !42 + %6984 = fmul float %3918, %6472, !dbg !42 + %6985 = fmul float %3918, %6473, !dbg !42 + %6986 = fmul float %3918, %6474, !dbg !42 + %6987 = fmul float %3918, %6475, !dbg !42 + %6988 = fmul float %3918, %6476, !dbg !42 + %6989 = fmul float %3858, %6477, !dbg !42 + %6990 = fmul float %3858, %6478, !dbg !42 + %6991 = fmul float %3858, %6479, !dbg !42 + %6992 = fmul float %3858, %6480, !dbg !42 + %6993 = fmul float %3858, %6481, !dbg !42 + %6994 = fmul float %3858, %6482, !dbg !42 + %6995 = fmul float %3858, %6483, !dbg !42 + %6996 = fmul float %3858, %6484, !dbg !42 + %6997 = fmul float %3919, %6485, !dbg !42 + %6998 = fmul float %3919, %6486, !dbg !42 + %6999 = fmul float %3919, %6487, !dbg !42 + %7000 = fmul float %3919, %6488, !dbg !42 + %7001 = fmul float %3919, %6489, !dbg !42 + %7002 = fmul float %3919, %6490, !dbg !42 + %7003 = fmul float %3919, %6491, !dbg !42 + %7004 = fmul float %3919, %6492, !dbg !42 + %7005 = fmul float %3860, %6493, !dbg !42 + %7006 = fmul float %3860, %6494, !dbg !42 + %7007 = fmul float %3860, %6495, !dbg !42 + %7008 = fmul float %3860, %6496, !dbg !42 + %7009 = fmul float %3860, %6497, !dbg !42 + %7010 = fmul float %3860, %6498, !dbg !42 + %7011 = fmul float %3860, %6499, !dbg !42 + %7012 = fmul float %3860, %6500, !dbg !42 + %7013 = fmul float %3920, %6501, !dbg !42 + %7014 = fmul float %3920, %6502, !dbg !42 + %7015 = fmul float %3920, %6503, !dbg !42 + %7016 = fmul float %3920, %6504, !dbg !42 + %7017 = fmul float %3920, %6505, !dbg !42 + %7018 = fmul float %3920, %6506, !dbg !42 + %7019 = fmul float %3920, %6507, !dbg !42 + %7020 = fmul float %3920, %6508, !dbg !42 + %7021 = fmul float %3862, %6509, !dbg !42 + %7022 = fmul float %3862, %6510, !dbg !42 + %7023 = fmul float %3862, %6511, !dbg !42 + %7024 = fmul float %3862, %6512, !dbg !42 + %7025 = fmul float %3862, %6513, !dbg !42 + %7026 = fmul float %3862, %6514, !dbg !42 + %7027 = fmul float %3862, %6515, !dbg !42 + %7028 = fmul float %3862, %6516, !dbg !42 + %7029 = fmul float %3921, %6517, !dbg !42 + %7030 = fmul float %3921, %6518, !dbg !42 + %7031 = fmul float %3921, %6519, !dbg !42 + %7032 = fmul float %3921, %6520, !dbg !42 + %7033 = fmul float %3921, %6521, !dbg !42 + %7034 = fmul float %3921, %6522, !dbg !42 + %7035 = fmul float %3921, %6523, !dbg !42 + %7036 = fmul float %3921, %6524, !dbg !42 + %7037 = fmul float %3864, %6525, !dbg !42 + %7038 = fmul float %3864, %6526, !dbg !42 + %7039 = fmul float %3864, %6527, !dbg !42 + %7040 = fmul float %3864, %6528, !dbg !42 + %7041 = fmul float %3864, %6529, !dbg !42 + %7042 = fmul float %3864, %6530, !dbg !42 + %7043 = fmul float %3864, %6531, !dbg !42 + %7044 = fmul float %3864, %6532, !dbg !42 + %7045 = fmul float %3922, %6533, !dbg !42 + %7046 = fmul float %3922, %6534, !dbg !42 + %7047 = fmul float %3922, %6535, !dbg !42 + %7048 = fmul float %3922, %6536, !dbg !42 + %7049 = fmul float %3922, %6537, !dbg !42 + %7050 = fmul float %3922, %6538, !dbg !42 + %7051 = fmul float %3922, %6539, !dbg !42 + %7052 = fmul float %3922, %6540, !dbg !42 + %7053 = fmul float %3866, %6541, !dbg !42 + %7054 = fmul float %3866, %6542, !dbg !42 + %7055 = fmul float %3866, %6543, !dbg !42 + %7056 = fmul float %3866, %6544, !dbg !42 + %7057 = fmul float %3866, %6545, !dbg !42 + %7058 = fmul float %3866, %6546, !dbg !42 + %7059 = fmul float %3866, %6547, !dbg !42 + %7060 = fmul float %3866, %6548, !dbg !42 + %7061 = fmul float %3923, %6549, !dbg !42 + %7062 = fmul float %3923, %6550, !dbg !42 + %7063 = fmul float %3923, %6551, !dbg !42 + %7064 = fmul float %3923, %6552, !dbg !42 + %7065 = fmul float %3923, %6553, !dbg !42 + %7066 = fmul float %3923, %6554, !dbg !42 + %7067 = fmul float %3923, %6555, !dbg !42 + %7068 = fmul float %3923, %6556, !dbg !42 + %7069 = fmul float %3868, %6557, !dbg !42 + %7070 = fmul float %3868, %6558, !dbg !42 + %7071 = fmul float %3868, %6559, !dbg !42 + %7072 = fmul float %3868, %6560, !dbg !42 + %7073 = fmul float %3868, %6561, !dbg !42 + %7074 = fmul float %3868, %6562, !dbg !42 + %7075 = fmul float %3868, %6563, !dbg !42 + %7076 = fmul float %3868, %6564, !dbg !42 + %7077 = fmul float %3924, %6565, !dbg !42 + %7078 = fmul float %3924, %6566, !dbg !42 + %7079 = fmul float %3924, %6567, !dbg !42 + %7080 = fmul float %3924, %6568, !dbg !42 + %7081 = fmul float %3924, %6569, !dbg !42 + %7082 = fmul float %3924, %6570, !dbg !42 + %7083 = fmul float %3924, %6571, !dbg !42 + %7084 = fmul float %3924, %6572, !dbg !42 + %7085 = fmul float %3870, %6573, !dbg !42 + %7086 = fmul float %3870, %6574, !dbg !42 + %7087 = fmul float %3870, %6575, !dbg !42 + %7088 = fmul float %3870, %6576, !dbg !42 + %7089 = fmul float %3870, %6577, !dbg !42 + %7090 = fmul float %3870, %6578, !dbg !42 + %7091 = fmul float %3870, %6579, !dbg !42 + %7092 = fmul float %3870, %6580, !dbg !42 + %7093 = fmul float %3925, %6581, !dbg !42 + %7094 = fmul float %3925, %6582, !dbg !42 + %7095 = fmul float %3925, %6583, !dbg !42 + %7096 = fmul float %3925, %6584, !dbg !42 + %7097 = fmul float %3925, %6585, !dbg !42 + %7098 = fmul float %3925, %6586, !dbg !42 + %7099 = fmul float %3925, %6587, !dbg !42 + %7100 = fmul float %3925, %6588, !dbg !42 + %7101 = fmul float %3872, %6589, !dbg !42 + %7102 = fmul float %3872, %6590, !dbg !42 + %7103 = fmul float %3872, %6591, !dbg !42 + %7104 = fmul float %3872, %6592, !dbg !42 + %7105 = fmul float %3872, %6593, !dbg !42 + %7106 = fmul float %3872, %6594, !dbg !42 + %7107 = fmul float %3872, %6595, !dbg !42 + %7108 = fmul float %3872, %6596, !dbg !42 + %7109 = fmul float %3926, %6597, !dbg !42 + %7110 = fmul float %3926, %6598, !dbg !42 + %7111 = fmul float %3926, %6599, !dbg !42 + %7112 = fmul float %3926, %6600, !dbg !42 + %7113 = fmul float %3926, %6601, !dbg !42 + %7114 = fmul float %3926, %6602, !dbg !42 + %7115 = fmul float %3926, %6603, !dbg !42 + %7116 = fmul float %3926, %6604, !dbg !42 + %7117 = fmul float %3874, %6605, !dbg !42 + %7118 = fmul float %3874, %6606, !dbg !42 + %7119 = fmul float %3874, %6607, !dbg !42 + %7120 = fmul float %3874, %6608, !dbg !42 + %7121 = fmul float %3874, %6609, !dbg !42 + %7122 = fmul float %3874, %6610, !dbg !42 + %7123 = fmul float %3874, %6611, !dbg !42 + %7124 = fmul float %3874, %6612, !dbg !42 + %7125 = fmul float %3927, %6613, !dbg !42 + %7126 = fmul float %3927, %6614, !dbg !42 + %7127 = fmul float %3927, %6615, !dbg !42 + %7128 = fmul float %3927, %6616, !dbg !42 + %7129 = fmul float %3927, %6617, !dbg !42 + %7130 = fmul float %3927, %6618, !dbg !42 + %7131 = fmul float %3927, %6619, !dbg !42 + %7132 = fmul float %3927, %6620, !dbg !42 + %7133 = fmul float %3876, %6621, !dbg !42 + %7134 = fmul float %3876, %6622, !dbg !42 + %7135 = fmul float %3876, %6623, !dbg !42 + %7136 = fmul float %3876, %6624, !dbg !42 + %7137 = fmul float %3876, %6625, !dbg !42 + %7138 = fmul float %3876, %6626, !dbg !42 + %7139 = fmul float %3876, %6627, !dbg !42 + %7140 = fmul float %3876, %6628, !dbg !42 + %7141 = fmul float %3928, %6629, !dbg !42 + %7142 = fmul float %3928, %6630, !dbg !42 + %7143 = fmul float %3928, %6631, !dbg !42 + %7144 = fmul float %3928, %6632, !dbg !42 + %7145 = fmul float %3928, %6633, !dbg !42 + %7146 = fmul float %3928, %6634, !dbg !42 + %7147 = fmul float %3928, %6635, !dbg !42 + %7148 = fmul float %3928, %6636, !dbg !42 + %7149 = fmul float %3878, %6637, !dbg !42 + %7150 = fmul float %3878, %6638, !dbg !42 + %7151 = fmul float %3878, %6639, !dbg !42 + %7152 = fmul float %3878, %6640, !dbg !42 + %7153 = fmul float %3878, %6641, !dbg !42 + %7154 = fmul float %3878, %6642, !dbg !42 + %7155 = fmul float %3878, %6643, !dbg !42 + %7156 = fmul float %3878, %6644, !dbg !42 + %7157 = fmul float %3929, %6645, !dbg !42 + %7158 = fmul float %3929, %6646, !dbg !42 + %7159 = fmul float %3929, %6647, !dbg !42 + %7160 = fmul float %3929, %6648, !dbg !42 + %7161 = fmul float %3929, %6649, !dbg !42 + %7162 = fmul float %3929, %6650, !dbg !42 + %7163 = fmul float %3929, %6651, !dbg !42 + %7164 = fmul float %3929, %6652, !dbg !42 + %7165 = fmul float %3880, %6653, !dbg !42 + %7166 = fmul float %3880, %6654, !dbg !42 + %7167 = fmul float %3880, %6655, !dbg !42 + %7168 = fmul float %3880, %6656, !dbg !42 + %7169 = fmul float %3880, %6657, !dbg !42 + %7170 = fmul float %3880, %6658, !dbg !42 + %7171 = fmul float %3880, %6659, !dbg !42 + %7172 = fmul float %3880, %6660, !dbg !42 + %7173 = fmul float %3930, %6661, !dbg !42 + %7174 = fmul float %3930, %6662, !dbg !42 + %7175 = fmul float %3930, %6663, !dbg !42 + %7176 = fmul float %3930, %6664, !dbg !42 + %7177 = fmul float %3930, %6665, !dbg !42 + %7178 = fmul float %3930, %6666, !dbg !42 + %7179 = fmul float %3930, %6667, !dbg !42 + %7180 = fmul float %3930, %6668, !dbg !42 + %7181 = fmul float %3882, %6669, !dbg !42 + %7182 = fmul float %3882, %6670, !dbg !42 + %7183 = fmul float %3882, %6671, !dbg !42 + %7184 = fmul float %3882, %6672, !dbg !42 + %7185 = fmul float %3882, %6673, !dbg !42 + %7186 = fmul float %3882, %6674, !dbg !42 + %7187 = fmul float %3882, %6675, !dbg !42 + %7188 = fmul float %3882, %6676, !dbg !42 + %7189 = fmul float %3931, %6677, !dbg !42 + %7190 = fmul float %3931, %6678, !dbg !42 + %7191 = fmul float %3931, %6679, !dbg !42 + %7192 = fmul float %3931, %6680, !dbg !42 + %7193 = fmul float %3931, %6681, !dbg !42 + %7194 = fmul float %3931, %6682, !dbg !42 + %7195 = fmul float %3931, %6683, !dbg !42 + %7196 = fmul float %3931, %6684, !dbg !42 + %7197 = fmul float %3884, %6685, !dbg !42 + %7198 = fmul float %3884, %6686, !dbg !42 + %7199 = fmul float %3884, %6687, !dbg !42 + %7200 = fmul float %3884, %6688, !dbg !42 + %7201 = fmul float %3884, %6689, !dbg !42 + %7202 = fmul float %3884, %6690, !dbg !42 + %7203 = fmul float %3884, %6691, !dbg !42 + %7204 = fmul float %3884, %6692, !dbg !42 + %7205 = fmul float %3932, %6693, !dbg !42 + %7206 = fmul float %3932, %6694, !dbg !42 + %7207 = fmul float %3932, %6695, !dbg !42 + %7208 = fmul float %3932, %6696, !dbg !42 + %7209 = fmul float %3932, %6697, !dbg !42 + %7210 = fmul float %3932, %6698, !dbg !42 + %7211 = fmul float %3932, %6699, !dbg !42 + %7212 = fmul float %3932, %6700, !dbg !42 + %7213 = fmul float %3886, %6701, !dbg !42 + %7214 = fmul float %3886, %6702, !dbg !42 + %7215 = fmul float %3886, %6703, !dbg !42 + %7216 = fmul float %3886, %6704, !dbg !42 + %7217 = fmul float %3886, %6705, !dbg !42 + %7218 = fmul float %3886, %6706, !dbg !42 + %7219 = fmul float %3886, %6707, !dbg !42 + %7220 = fmul float %3886, %6708, !dbg !42 + %7221 = fmul float %3933, %6709, !dbg !42 + %7222 = fmul float %3933, %6710, !dbg !42 + %7223 = fmul float %3933, %6711, !dbg !42 + %7224 = fmul float %3933, %6712, !dbg !42 + %7225 = fmul float %3933, %6713, !dbg !42 + %7226 = fmul float %3933, %6714, !dbg !42 + %7227 = fmul float %3933, %6715, !dbg !42 + %7228 = fmul float %3933, %6716, !dbg !42 + %7229 = fmul float %3888, %6717, !dbg !42 + %7230 = fmul float %3888, %6718, !dbg !42 + %7231 = fmul float %3888, %6719, !dbg !42 + %7232 = fmul float %3888, %6720, !dbg !42 + %7233 = fmul float %3888, %6721, !dbg !42 + %7234 = fmul float %3888, %6722, !dbg !42 + %7235 = fmul float %3888, %6723, !dbg !42 + %7236 = fmul float %3888, %6724, !dbg !42 + %7237 = fmul float %3934, %6725, !dbg !42 + %7238 = fmul float %3934, %6726, !dbg !42 + %7239 = fmul float %3934, %6727, !dbg !42 + %7240 = fmul float %3934, %6728, !dbg !42 + %7241 = fmul float %3934, %6729, !dbg !42 + %7242 = fmul float %3934, %6730, !dbg !42 + %7243 = fmul float %3934, %6731, !dbg !42 + %7244 = fmul float %3934, %6732, !dbg !42 + %7245 = fmul float %3890, %6733, !dbg !42 + %7246 = fmul float %3890, %6734, !dbg !42 + %7247 = fmul float %3890, %6735, !dbg !42 + %7248 = fmul float %3890, %6736, !dbg !42 + %7249 = fmul float %3890, %6737, !dbg !42 + %7250 = fmul float %3890, %6738, !dbg !42 + %7251 = fmul float %3890, %6739, !dbg !42 + %7252 = fmul float %3890, %6740, !dbg !42 + %7253 = fmul float %3935, %6741, !dbg !42 + %7254 = fmul float %3935, %6742, !dbg !42 + %7255 = fmul float %3935, %6743, !dbg !42 + %7256 = fmul float %3935, %6744, !dbg !42 + %7257 = fmul float %3935, %6745, !dbg !42 + %7258 = fmul float %3935, %6746, !dbg !42 + %7259 = fmul float %3935, %6747, !dbg !42 + %7260 = fmul float %3935, %6748, !dbg !42 + %7261 = fmul float %3892, %6749, !dbg !42 + %7262 = fmul float %3892, %6750, !dbg !42 + %7263 = fmul float %3892, %6751, !dbg !42 + %7264 = fmul float %3892, %6752, !dbg !42 + %7265 = fmul float %3892, %6753, !dbg !42 + %7266 = fmul float %3892, %6754, !dbg !42 + %7267 = fmul float %3892, %6755, !dbg !42 + %7268 = fmul float %3892, %6756, !dbg !42 + %7269 = fmul float %3936, %6757, !dbg !42 + %7270 = fmul float %3936, %6758, !dbg !42 + %7271 = fmul float %3936, %6759, !dbg !42 + %7272 = fmul float %3936, %6760, !dbg !42 + %7273 = fmul float %3936, %6761, !dbg !42 + %7274 = fmul float %3936, %6762, !dbg !42 + %7275 = fmul float %3936, %6763, !dbg !42 + %7276 = fmul float %3936, %6764, !dbg !42 + %7277 = fmul float %3894, %6765, !dbg !42 + %7278 = fmul float %3894, %6766, !dbg !42 + %7279 = fmul float %3894, %6767, !dbg !42 + %7280 = fmul float %3894, %6768, !dbg !42 + %7281 = fmul float %3894, %6769, !dbg !42 + %7282 = fmul float %3894, %6770, !dbg !42 + %7283 = fmul float %3894, %6771, !dbg !42 + %7284 = fmul float %3894, %6772, !dbg !42 + %7285 = fmul float %3937, %6773, !dbg !42 + %7286 = fmul float %3937, %6774, !dbg !42 + %7287 = fmul float %3937, %6775, !dbg !42 + %7288 = fmul float %3937, %6776, !dbg !42 + %7289 = fmul float %3937, %6777, !dbg !42 + %7290 = fmul float %3937, %6778, !dbg !42 + %7291 = fmul float %3937, %6779, !dbg !42 + %7292 = fmul float %3937, %6780, !dbg !42 + %7293 = fmul float %3896, %6781, !dbg !42 + %7294 = fmul float %3896, %6782, !dbg !42 + %7295 = fmul float %3896, %6783, !dbg !42 + %7296 = fmul float %3896, %6784, !dbg !42 + %7297 = fmul float %3896, %6785, !dbg !42 + %7298 = fmul float %3896, %6786, !dbg !42 + %7299 = fmul float %3896, %6787, !dbg !42 + %7300 = fmul float %3896, %6788, !dbg !42 + %7301 = fmul float %3938, %6789, !dbg !42 + %7302 = fmul float %3938, %6790, !dbg !42 + %7303 = fmul float %3938, %6791, !dbg !42 + %7304 = fmul float %3938, %6792, !dbg !42 + %7305 = fmul float %3938, %6793, !dbg !42 + %7306 = fmul float %3938, %6794, !dbg !42 + %7307 = fmul float %3938, %6795, !dbg !42 + %7308 = fmul float %3938, %6796, !dbg !42 + %7309 = fmul float %3898, %6797, !dbg !42 + %7310 = fmul float %3898, %6798, !dbg !42 + %7311 = fmul float %3898, %6799, !dbg !42 + %7312 = fmul float %3898, %6800, !dbg !42 + %7313 = fmul float %3898, %6801, !dbg !42 + %7314 = fmul float %3898, %6802, !dbg !42 + %7315 = fmul float %3898, %6803, !dbg !42 + %7316 = fmul float %3898, %6804, !dbg !42 + %7317 = fmul float %3939, %6805, !dbg !42 + %7318 = fmul float %3939, %6806, !dbg !42 + %7319 = fmul float %3939, %6807, !dbg !42 + %7320 = fmul float %3939, %6808, !dbg !42 + %7321 = fmul float %3939, %6809, !dbg !42 + %7322 = fmul float %3939, %6810, !dbg !42 + %7323 = fmul float %3939, %6811, !dbg !42 + %7324 = fmul float %3939, %6812, !dbg !42 + %7325 = fmul float %3900, %6813, !dbg !42 + %7326 = fmul float %3900, %6814, !dbg !42 + %7327 = fmul float %3900, %6815, !dbg !42 + %7328 = fmul float %3900, %6816, !dbg !42 + %7329 = fmul float %3900, %6817, !dbg !42 + %7330 = fmul float %3900, %6818, !dbg !42 + %7331 = fmul float %3900, %6819, !dbg !42 + %7332 = fmul float %3900, %6820, !dbg !42 + %7333 = fmul float %3940, %6821, !dbg !42 + %7334 = fmul float %3940, %6822, !dbg !42 + %7335 = fmul float %3940, %6823, !dbg !42 + %7336 = fmul float %3940, %6824, !dbg !42 + %7337 = fmul float %3940, %6825, !dbg !42 + %7338 = fmul float %3940, %6826, !dbg !42 + %7339 = fmul float %3940, %6827, !dbg !42 + %7340 = fmul float %3940, %6828, !dbg !42 + %7341 = fmul float %3902, %6829, !dbg !42 + %7342 = fmul float %3902, %6830, !dbg !42 + %7343 = fmul float %3902, %6831, !dbg !42 + %7344 = fmul float %3902, %6832, !dbg !42 + %7345 = fmul float %3902, %6833, !dbg !42 + %7346 = fmul float %3902, %6834, !dbg !42 + %7347 = fmul float %3902, %6835, !dbg !42 + %7348 = fmul float %3902, %6836, !dbg !42 + %7349 = fmul float %3941, %6837, !dbg !42 + %7350 = fmul float %3941, %6838, !dbg !42 + %7351 = fmul float %3941, %6839, !dbg !42 + %7352 = fmul float %3941, %6840, !dbg !42 + %7353 = fmul float %3941, %6841, !dbg !42 + %7354 = fmul float %3941, %6842, !dbg !42 + %7355 = fmul float %3941, %6843, !dbg !42 + %7356 = fmul float %3941, %6844, !dbg !42 + %7357 = fmul float %3904, %6845, !dbg !42 + %7358 = fmul float %3904, %6846, !dbg !42 + %7359 = fmul float %3904, %6847, !dbg !42 + %7360 = fmul float %3904, %6848, !dbg !42 + %7361 = fmul float %3904, %6849, !dbg !42 + %7362 = fmul float %3904, %6850, !dbg !42 + %7363 = fmul float %3904, %6851, !dbg !42 + %7364 = fmul float %3904, %6852, !dbg !42 + %7365 = fmul float %3942, %6853, !dbg !42 + %7366 = fmul float %3942, %6854, !dbg !42 + %7367 = fmul float %3942, %6855, !dbg !42 + %7368 = fmul float %3942, %6856, !dbg !42 + %7369 = fmul float %3942, %6857, !dbg !42 + %7370 = fmul float %3942, %6858, !dbg !42 + %7371 = fmul float %3942, %6859, !dbg !42 + %7372 = fmul float %3942, %6860, !dbg !42 + %7373 = fmul float %3906, %6861, !dbg !42 + %7374 = fmul float %3906, %6862, !dbg !42 + %7375 = fmul float %3906, %6863, !dbg !42 + %7376 = fmul float %3906, %6864, !dbg !42 + %7377 = fmul float %3906, %6865, !dbg !42 + %7378 = fmul float %3906, %6866, !dbg !42 + %7379 = fmul float %3906, %6867, !dbg !42 + %7380 = fmul float %3906, %6868, !dbg !42 + %7381 = fmul float %3943, %6869, !dbg !42 + %7382 = fmul float %3943, %6870, !dbg !42 + %7383 = fmul float %3943, %6871, !dbg !42 + %7384 = fmul float %3943, %6872, !dbg !42 + %7385 = fmul float %3943, %6873, !dbg !42 + %7386 = fmul float %3943, %6874, !dbg !42 + %7387 = fmul float %3943, %6875, !dbg !42 + %7388 = fmul float %3943, %6876, !dbg !42 + %7389 = fmul float %3908, %6877, !dbg !42 + %7390 = fmul float %3908, %6878, !dbg !42 + %7391 = fmul float %3908, %6879, !dbg !42 + %7392 = fmul float %3908, %6880, !dbg !42 + %7393 = fmul float %3908, %6881, !dbg !42 + %7394 = fmul float %3908, %6882, !dbg !42 + %7395 = fmul float %3908, %6883, !dbg !42 + %7396 = fmul float %3908, %6884, !dbg !42 + %7397 = fmul float %3944, %6885, !dbg !42 + %7398 = fmul float %3944, %6886, !dbg !42 + %7399 = fmul float %3944, %6887, !dbg !42 + %7400 = fmul float %3944, %6888, !dbg !42 + %7401 = fmul float %3944, %6889, !dbg !42 + %7402 = fmul float %3944, %6890, !dbg !42 + %7403 = fmul float %3944, %6891, !dbg !42 + %7404 = fmul float %3944, %6892, !dbg !42 + %7405 = fmul float %3910, %6893, !dbg !42 + %7406 = fmul float %3910, %6894, !dbg !42 + %7407 = fmul float %3910, %6895, !dbg !42 + %7408 = fmul float %3910, %6896, !dbg !42 + %7409 = fmul float %3910, %6897, !dbg !42 + %7410 = fmul float %3910, %6898, !dbg !42 + %7411 = fmul float %3910, %6899, !dbg !42 + %7412 = fmul float %3910, %6900, !dbg !42 + %7413 = fmul float %3945, %6901, !dbg !42 + %7414 = fmul float %3945, %6902, !dbg !42 + %7415 = fmul float %3945, %6903, !dbg !42 + %7416 = fmul float %3945, %6904, !dbg !42 + %7417 = fmul float %3945, %6905, !dbg !42 + %7418 = fmul float %3945, %6906, !dbg !42 + %7419 = fmul float %3945, %6907, !dbg !42 + %7420 = fmul float %3945, %6908, !dbg !42 + %7421 = fmul float %3912, %6909, !dbg !42 + %7422 = fmul float %3912, %6910, !dbg !42 + %7423 = fmul float %3912, %6911, !dbg !42 + %7424 = fmul float %3912, %6912, !dbg !42 + %7425 = fmul float %3912, %6913, !dbg !42 + %7426 = fmul float %3912, %6914, !dbg !42 + %7427 = fmul float %3912, %6915, !dbg !42 + %7428 = fmul float %3912, %6916, !dbg !42 + %7429 = fmul float %3946, %6917, !dbg !42 + %7430 = fmul float %3946, %6918, !dbg !42 + %7431 = fmul float %3946, %6919, !dbg !42 + %7432 = fmul float %3946, %6920, !dbg !42 + %7433 = fmul float %3946, %6921, !dbg !42 + %7434 = fmul float %3946, %6922, !dbg !42 + %7435 = fmul float %3946, %6923, !dbg !42 + %7436 = fmul float %3946, %6924, !dbg !42 + %7437 = fmul float %3914, %6925, !dbg !42 + %7438 = fmul float %3914, %6926, !dbg !42 + %7439 = fmul float %3914, %6927, !dbg !42 + %7440 = fmul float %3914, %6928, !dbg !42 + %7441 = fmul float %3914, %6929, !dbg !42 + %7442 = fmul float %3914, %6930, !dbg !42 + %7443 = fmul float %3914, %6931, !dbg !42 + %7444 = fmul float %3914, %6932, !dbg !42 + %7445 = fmul float %3947, %6933, !dbg !42 + %7446 = fmul float %3947, %6934, !dbg !42 + %7447 = fmul float %3947, %6935, !dbg !42 + %7448 = fmul float %3947, %6936, !dbg !42 + %7449 = fmul float %3947, %6937, !dbg !42 + %7450 = fmul float %3947, %6938, !dbg !42 + %7451 = fmul float %3947, %6939, !dbg !42 + %7452 = fmul float %3947, %6940, !dbg !42 + %7453 = fmul float %3916, %6941, !dbg !42 + %7454 = fmul float %3916, %6942, !dbg !42 + %7455 = fmul float %3916, %6943, !dbg !42 + %7456 = fmul float %3916, %6944, !dbg !42 + %7457 = fmul float %3916, %6945, !dbg !42 + %7458 = fmul float %3916, %6946, !dbg !42 + %7459 = fmul float %3916, %6947, !dbg !42 + %7460 = fmul float %3916, %6948, !dbg !42 + %7461 = fmul float %3948, %6949, !dbg !42 + %7462 = fmul float %3948, %6950, !dbg !42 + %7463 = fmul float %3948, %6951, !dbg !42 + %7464 = fmul float %3948, %6952, !dbg !42 + %7465 = fmul float %3948, %6953, !dbg !42 + %7466 = fmul float %3948, %6954, !dbg !42 + %7467 = fmul float %3948, %6955, !dbg !42 + %7468 = fmul float %3948, %6956, !dbg !42 + %7469 = fsub float %3949, %6957, !dbg !43 + %7470 = fsub float %3950, %6958, !dbg !43 + %7471 = fsub float %3951, %6959, !dbg !43 + %7472 = fsub float %3952, %6960, !dbg !43 + %7473 = fsub float %3953, %6961, !dbg !43 + %7474 = fsub float %3954, %6962, !dbg !43 + %7475 = fsub float %3955, %6963, !dbg !43 + %7476 = fsub float %3956, %6964, !dbg !43 + %7477 = insertelement <2 x float> poison, float %3957, i64 0, !dbg !43 + %7478 = insertelement <2 x float> %7477, float %3958, i64 1, !dbg !43 + %7479 = insertelement <2 x float> poison, float %6965, i64 0, !dbg !43 + %7480 = insertelement <2 x float> %7479, float %6966, i64 1, !dbg !43 + %7481 = fsub <2 x float> %7478, %7480, !dbg !43 + %7482 = insertelement <2 x float> poison, float %3959, i64 0, !dbg !43 + %7483 = insertelement <2 x float> %7482, float %3960, i64 1, !dbg !43 + %7484 = insertelement <2 x float> poison, float %6967, i64 0, !dbg !43 + %7485 = insertelement <2 x float> %7484, float %6968, i64 1, !dbg !43 + %7486 = fsub <2 x float> %7483, %7485, !dbg !43 + %7487 = insertelement <2 x float> poison, float %3961, i64 0, !dbg !43 + %7488 = insertelement <2 x float> %7487, float %3962, i64 1, !dbg !43 + %7489 = insertelement <2 x float> poison, float %6969, i64 0, !dbg !43 + %7490 = insertelement <2 x float> %7489, float %6970, i64 1, !dbg !43 + %7491 = fsub <2 x float> %7488, %7490, !dbg !43 + %7492 = insertelement <2 x float> poison, float %3963, i64 0, !dbg !43 + %7493 = insertelement <2 x float> %7492, float %3964, i64 1, !dbg !43 + %7494 = insertelement <2 x float> poison, float %6971, i64 0, !dbg !43 + %7495 = insertelement <2 x float> %7494, float %6972, i64 1, !dbg !43 + %7496 = fsub <2 x float> %7493, %7495, !dbg !43 + %7497 = insertelement <2 x float> poison, float %3965, i64 0, !dbg !43 + %7498 = insertelement <2 x float> %7497, float %3966, i64 1, !dbg !43 + %7499 = insertelement <2 x float> poison, float %6973, i64 0, !dbg !43 + %7500 = insertelement <2 x float> %7499, float %6974, i64 1, !dbg !43 + %7501 = fsub <2 x float> %7498, %7500, !dbg !43 + %7502 = insertelement <2 x float> poison, float %3967, i64 0, !dbg !43 + %7503 = insertelement <2 x float> %7502, float %3968, i64 1, !dbg !43 + %7504 = insertelement <2 x float> poison, float %6975, i64 0, !dbg !43 + %7505 = insertelement <2 x float> %7504, float %6976, i64 1, !dbg !43 + %7506 = fsub <2 x float> %7503, %7505, !dbg !43 + %7507 = insertelement <2 x float> poison, float %3969, i64 0, !dbg !43 + %7508 = insertelement <2 x float> %7507, float %3970, i64 1, !dbg !43 + %7509 = insertelement <2 x float> poison, float %6977, i64 0, !dbg !43 + %7510 = insertelement <2 x float> %7509, float %6978, i64 1, !dbg !43 + %7511 = fsub <2 x float> %7508, %7510, !dbg !43 + %7512 = insertelement <2 x float> poison, float %3971, i64 0, !dbg !43 + %7513 = insertelement <2 x float> %7512, float %3972, i64 1, !dbg !43 + %7514 = insertelement <2 x float> poison, float %6979, i64 0, !dbg !43 + %7515 = insertelement <2 x float> %7514, float %6980, i64 1, !dbg !43 + %7516 = fsub <2 x float> %7513, %7515, !dbg !43 + %7517 = insertelement <2 x float> poison, float %3973, i64 0, !dbg !43 + %7518 = insertelement <2 x float> %7517, float %3974, i64 1, !dbg !43 + %7519 = insertelement <2 x float> poison, float %6981, i64 0, !dbg !43 + %7520 = insertelement <2 x float> %7519, float %6982, i64 1, !dbg !43 + %7521 = fsub <2 x float> %7518, %7520, !dbg !43 + %7522 = insertelement <2 x float> poison, float %3975, i64 0, !dbg !43 + %7523 = insertelement <2 x float> %7522, float %3976, i64 1, !dbg !43 + %7524 = insertelement <2 x float> poison, float %6983, i64 0, !dbg !43 + %7525 = insertelement <2 x float> %7524, float %6984, i64 1, !dbg !43 + %7526 = fsub <2 x float> %7523, %7525, !dbg !43 + %7527 = insertelement <2 x float> poison, float %3977, i64 0, !dbg !43 + %7528 = insertelement <2 x float> %7527, float %3978, i64 1, !dbg !43 + %7529 = insertelement <2 x float> poison, float %6985, i64 0, !dbg !43 + %7530 = insertelement <2 x float> %7529, float %6986, i64 1, !dbg !43 + %7531 = fsub <2 x float> %7528, %7530, !dbg !43 + %7532 = insertelement <2 x float> poison, float %3979, i64 0, !dbg !43 + %7533 = insertelement <2 x float> %7532, float %3980, i64 1, !dbg !43 + %7534 = insertelement <2 x float> poison, float %6987, i64 0, !dbg !43 + %7535 = insertelement <2 x float> %7534, float %6988, i64 1, !dbg !43 + %7536 = fsub <2 x float> %7533, %7535, !dbg !43 + %7537 = insertelement <2 x float> poison, float %3981, i64 0, !dbg !43 + %7538 = insertelement <2 x float> %7537, float %3982, i64 1, !dbg !43 + %7539 = insertelement <2 x float> poison, float %6989, i64 0, !dbg !43 + %7540 = insertelement <2 x float> %7539, float %6990, i64 1, !dbg !43 + %7541 = fsub <2 x float> %7538, %7540, !dbg !43 + %7542 = insertelement <2 x float> poison, float %3983, i64 0, !dbg !43 + %7543 = insertelement <2 x float> %7542, float %3984, i64 1, !dbg !43 + %7544 = insertelement <2 x float> poison, float %6991, i64 0, !dbg !43 + %7545 = insertelement <2 x float> %7544, float %6992, i64 1, !dbg !43 + %7546 = fsub <2 x float> %7543, %7545, !dbg !43 + %7547 = insertelement <2 x float> poison, float %3985, i64 0, !dbg !43 + %7548 = insertelement <2 x float> %7547, float %3986, i64 1, !dbg !43 + %7549 = insertelement <2 x float> poison, float %6993, i64 0, !dbg !43 + %7550 = insertelement <2 x float> %7549, float %6994, i64 1, !dbg !43 + %7551 = fsub <2 x float> %7548, %7550, !dbg !43 + %7552 = insertelement <2 x float> poison, float %3987, i64 0, !dbg !43 + %7553 = insertelement <2 x float> %7552, float %3988, i64 1, !dbg !43 + %7554 = insertelement <2 x float> poison, float %6995, i64 0, !dbg !43 + %7555 = insertelement <2 x float> %7554, float %6996, i64 1, !dbg !43 + %7556 = fsub <2 x float> %7553, %7555, !dbg !43 + %7557 = insertelement <2 x float> poison, float %3989, i64 0, !dbg !43 + %7558 = insertelement <2 x float> %7557, float %3990, i64 1, !dbg !43 + %7559 = insertelement <2 x float> poison, float %6997, i64 0, !dbg !43 + %7560 = insertelement <2 x float> %7559, float %6998, i64 1, !dbg !43 + %7561 = fsub <2 x float> %7558, %7560, !dbg !43 + %7562 = insertelement <2 x float> poison, float %3991, i64 0, !dbg !43 + %7563 = insertelement <2 x float> %7562, float %3992, i64 1, !dbg !43 + %7564 = insertelement <2 x float> poison, float %6999, i64 0, !dbg !43 + %7565 = insertelement <2 x float> %7564, float %7000, i64 1, !dbg !43 + %7566 = fsub <2 x float> %7563, %7565, !dbg !43 + %7567 = insertelement <2 x float> poison, float %3993, i64 0, !dbg !43 + %7568 = insertelement <2 x float> %7567, float %3994, i64 1, !dbg !43 + %7569 = insertelement <2 x float> poison, float %7001, i64 0, !dbg !43 + %7570 = insertelement <2 x float> %7569, float %7002, i64 1, !dbg !43 + %7571 = fsub <2 x float> %7568, %7570, !dbg !43 + %7572 = insertelement <2 x float> poison, float %3995, i64 0, !dbg !43 + %7573 = insertelement <2 x float> %7572, float %3996, i64 1, !dbg !43 + %7574 = insertelement <2 x float> poison, float %7003, i64 0, !dbg !43 + %7575 = insertelement <2 x float> %7574, float %7004, i64 1, !dbg !43 + %7576 = fsub <2 x float> %7573, %7575, !dbg !43 + %7577 = insertelement <2 x float> poison, float %3997, i64 0, !dbg !43 + %7578 = insertelement <2 x float> %7577, float %3998, i64 1, !dbg !43 + %7579 = insertelement <2 x float> poison, float %7005, i64 0, !dbg !43 + %7580 = insertelement <2 x float> %7579, float %7006, i64 1, !dbg !43 + %7581 = fsub <2 x float> %7578, %7580, !dbg !43 + %7582 = insertelement <2 x float> poison, float %3999, i64 0, !dbg !43 + %7583 = insertelement <2 x float> %7582, float %4000, i64 1, !dbg !43 + %7584 = insertelement <2 x float> poison, float %7007, i64 0, !dbg !43 + %7585 = insertelement <2 x float> %7584, float %7008, i64 1, !dbg !43 + %7586 = fsub <2 x float> %7583, %7585, !dbg !43 + %7587 = insertelement <2 x float> poison, float %4001, i64 0, !dbg !43 + %7588 = insertelement <2 x float> %7587, float %4002, i64 1, !dbg !43 + %7589 = insertelement <2 x float> poison, float %7009, i64 0, !dbg !43 + %7590 = insertelement <2 x float> %7589, float %7010, i64 1, !dbg !43 + %7591 = fsub <2 x float> %7588, %7590, !dbg !43 + %7592 = insertelement <2 x float> poison, float %4003, i64 0, !dbg !43 + %7593 = insertelement <2 x float> %7592, float %4004, i64 1, !dbg !43 + %7594 = insertelement <2 x float> poison, float %7011, i64 0, !dbg !43 + %7595 = insertelement <2 x float> %7594, float %7012, i64 1, !dbg !43 + %7596 = fsub <2 x float> %7593, %7595, !dbg !43 + %7597 = insertelement <2 x float> poison, float %4005, i64 0, !dbg !43 + %7598 = insertelement <2 x float> %7597, float %4006, i64 1, !dbg !43 + %7599 = insertelement <2 x float> poison, float %7013, i64 0, !dbg !43 + %7600 = insertelement <2 x float> %7599, float %7014, i64 1, !dbg !43 + %7601 = fsub <2 x float> %7598, %7600, !dbg !43 + %7602 = insertelement <2 x float> poison, float %4007, i64 0, !dbg !43 + %7603 = insertelement <2 x float> %7602, float %4008, i64 1, !dbg !43 + %7604 = insertelement <2 x float> poison, float %7015, i64 0, !dbg !43 + %7605 = insertelement <2 x float> %7604, float %7016, i64 1, !dbg !43 + %7606 = fsub <2 x float> %7603, %7605, !dbg !43 + %7607 = insertelement <2 x float> poison, float %4009, i64 0, !dbg !43 + %7608 = insertelement <2 x float> %7607, float %4010, i64 1, !dbg !43 + %7609 = insertelement <2 x float> poison, float %7017, i64 0, !dbg !43 + %7610 = insertelement <2 x float> %7609, float %7018, i64 1, !dbg !43 + %7611 = fsub <2 x float> %7608, %7610, !dbg !43 + %7612 = insertelement <2 x float> poison, float %4011, i64 0, !dbg !43 + %7613 = insertelement <2 x float> %7612, float %4012, i64 1, !dbg !43 + %7614 = insertelement <2 x float> poison, float %7019, i64 0, !dbg !43 + %7615 = insertelement <2 x float> %7614, float %7020, i64 1, !dbg !43 + %7616 = fsub <2 x float> %7613, %7615, !dbg !43 + %7617 = insertelement <2 x float> poison, float %4013, i64 0, !dbg !43 + %7618 = insertelement <2 x float> %7617, float %4014, i64 1, !dbg !43 + %7619 = insertelement <2 x float> poison, float %7021, i64 0, !dbg !43 + %7620 = insertelement <2 x float> %7619, float %7022, i64 1, !dbg !43 + %7621 = fsub <2 x float> %7618, %7620, !dbg !43 + %7622 = insertelement <2 x float> poison, float %4015, i64 0, !dbg !43 + %7623 = insertelement <2 x float> %7622, float %4016, i64 1, !dbg !43 + %7624 = insertelement <2 x float> poison, float %7023, i64 0, !dbg !43 + %7625 = insertelement <2 x float> %7624, float %7024, i64 1, !dbg !43 + %7626 = fsub <2 x float> %7623, %7625, !dbg !43 + %7627 = insertelement <2 x float> poison, float %4017, i64 0, !dbg !43 + %7628 = insertelement <2 x float> %7627, float %4018, i64 1, !dbg !43 + %7629 = insertelement <2 x float> poison, float %7025, i64 0, !dbg !43 + %7630 = insertelement <2 x float> %7629, float %7026, i64 1, !dbg !43 + %7631 = fsub <2 x float> %7628, %7630, !dbg !43 + %7632 = insertelement <2 x float> poison, float %4019, i64 0, !dbg !43 + %7633 = insertelement <2 x float> %7632, float %4020, i64 1, !dbg !43 + %7634 = insertelement <2 x float> poison, float %7027, i64 0, !dbg !43 + %7635 = insertelement <2 x float> %7634, float %7028, i64 1, !dbg !43 + %7636 = fsub <2 x float> %7633, %7635, !dbg !43 + %7637 = insertelement <2 x float> poison, float %4021, i64 0, !dbg !43 + %7638 = insertelement <2 x float> %7637, float %4022, i64 1, !dbg !43 + %7639 = insertelement <2 x float> poison, float %7029, i64 0, !dbg !43 + %7640 = insertelement <2 x float> %7639, float %7030, i64 1, !dbg !43 + %7641 = fsub <2 x float> %7638, %7640, !dbg !43 + %7642 = insertelement <2 x float> poison, float %4023, i64 0, !dbg !43 + %7643 = insertelement <2 x float> %7642, float %4024, i64 1, !dbg !43 + %7644 = insertelement <2 x float> poison, float %7031, i64 0, !dbg !43 + %7645 = insertelement <2 x float> %7644, float %7032, i64 1, !dbg !43 + %7646 = fsub <2 x float> %7643, %7645, !dbg !43 + %7647 = insertelement <2 x float> poison, float %4025, i64 0, !dbg !43 + %7648 = insertelement <2 x float> %7647, float %4026, i64 1, !dbg !43 + %7649 = insertelement <2 x float> poison, float %7033, i64 0, !dbg !43 + %7650 = insertelement <2 x float> %7649, float %7034, i64 1, !dbg !43 + %7651 = fsub <2 x float> %7648, %7650, !dbg !43 + %7652 = insertelement <2 x float> poison, float %4027, i64 0, !dbg !43 + %7653 = insertelement <2 x float> %7652, float %4028, i64 1, !dbg !43 + %7654 = insertelement <2 x float> poison, float %7035, i64 0, !dbg !43 + %7655 = insertelement <2 x float> %7654, float %7036, i64 1, !dbg !43 + %7656 = fsub <2 x float> %7653, %7655, !dbg !43 + %7657 = insertelement <2 x float> poison, float %4029, i64 0, !dbg !43 + %7658 = insertelement <2 x float> %7657, float %4030, i64 1, !dbg !43 + %7659 = insertelement <2 x float> poison, float %7037, i64 0, !dbg !43 + %7660 = insertelement <2 x float> %7659, float %7038, i64 1, !dbg !43 + %7661 = fsub <2 x float> %7658, %7660, !dbg !43 + %7662 = insertelement <2 x float> poison, float %4031, i64 0, !dbg !43 + %7663 = insertelement <2 x float> %7662, float %4032, i64 1, !dbg !43 + %7664 = insertelement <2 x float> poison, float %7039, i64 0, !dbg !43 + %7665 = insertelement <2 x float> %7664, float %7040, i64 1, !dbg !43 + %7666 = fsub <2 x float> %7663, %7665, !dbg !43 + %7667 = insertelement <2 x float> poison, float %4033, i64 0, !dbg !43 + %7668 = insertelement <2 x float> %7667, float %4034, i64 1, !dbg !43 + %7669 = insertelement <2 x float> poison, float %7041, i64 0, !dbg !43 + %7670 = insertelement <2 x float> %7669, float %7042, i64 1, !dbg !43 + %7671 = fsub <2 x float> %7668, %7670, !dbg !43 + %7672 = insertelement <2 x float> poison, float %4035, i64 0, !dbg !43 + %7673 = insertelement <2 x float> %7672, float %4036, i64 1, !dbg !43 + %7674 = insertelement <2 x float> poison, float %7043, i64 0, !dbg !43 + %7675 = insertelement <2 x float> %7674, float %7044, i64 1, !dbg !43 + %7676 = fsub <2 x float> %7673, %7675, !dbg !43 + %7677 = insertelement <2 x float> poison, float %4037, i64 0, !dbg !43 + %7678 = insertelement <2 x float> %7677, float %4038, i64 1, !dbg !43 + %7679 = insertelement <2 x float> poison, float %7045, i64 0, !dbg !43 + %7680 = insertelement <2 x float> %7679, float %7046, i64 1, !dbg !43 + %7681 = fsub <2 x float> %7678, %7680, !dbg !43 + %7682 = insertelement <2 x float> poison, float %4039, i64 0, !dbg !43 + %7683 = insertelement <2 x float> %7682, float %4040, i64 1, !dbg !43 + %7684 = insertelement <2 x float> poison, float %7047, i64 0, !dbg !43 + %7685 = insertelement <2 x float> %7684, float %7048, i64 1, !dbg !43 + %7686 = fsub <2 x float> %7683, %7685, !dbg !43 + %7687 = insertelement <2 x float> poison, float %4041, i64 0, !dbg !43 + %7688 = insertelement <2 x float> %7687, float %4042, i64 1, !dbg !43 + %7689 = insertelement <2 x float> poison, float %7049, i64 0, !dbg !43 + %7690 = insertelement <2 x float> %7689, float %7050, i64 1, !dbg !43 + %7691 = fsub <2 x float> %7688, %7690, !dbg !43 + %7692 = insertelement <2 x float> poison, float %4043, i64 0, !dbg !43 + %7693 = insertelement <2 x float> %7692, float %4044, i64 1, !dbg !43 + %7694 = insertelement <2 x float> poison, float %7051, i64 0, !dbg !43 + %7695 = insertelement <2 x float> %7694, float %7052, i64 1, !dbg !43 + %7696 = fsub <2 x float> %7693, %7695, !dbg !43 + %7697 = insertelement <2 x float> poison, float %4045, i64 0, !dbg !43 + %7698 = insertelement <2 x float> %7697, float %4046, i64 1, !dbg !43 + %7699 = insertelement <2 x float> poison, float %7053, i64 0, !dbg !43 + %7700 = insertelement <2 x float> %7699, float %7054, i64 1, !dbg !43 + %7701 = fsub <2 x float> %7698, %7700, !dbg !43 + %7702 = insertelement <2 x float> poison, float %4047, i64 0, !dbg !43 + %7703 = insertelement <2 x float> %7702, float %4048, i64 1, !dbg !43 + %7704 = insertelement <2 x float> poison, float %7055, i64 0, !dbg !43 + %7705 = insertelement <2 x float> %7704, float %7056, i64 1, !dbg !43 + %7706 = fsub <2 x float> %7703, %7705, !dbg !43 + %7707 = insertelement <2 x float> poison, float %4049, i64 0, !dbg !43 + %7708 = insertelement <2 x float> %7707, float %4050, i64 1, !dbg !43 + %7709 = insertelement <2 x float> poison, float %7057, i64 0, !dbg !43 + %7710 = insertelement <2 x float> %7709, float %7058, i64 1, !dbg !43 + %7711 = fsub <2 x float> %7708, %7710, !dbg !43 + %7712 = insertelement <2 x float> poison, float %4051, i64 0, !dbg !43 + %7713 = insertelement <2 x float> %7712, float %4052, i64 1, !dbg !43 + %7714 = insertelement <2 x float> poison, float %7059, i64 0, !dbg !43 + %7715 = insertelement <2 x float> %7714, float %7060, i64 1, !dbg !43 + %7716 = fsub <2 x float> %7713, %7715, !dbg !43 + %7717 = insertelement <2 x float> poison, float %4053, i64 0, !dbg !43 + %7718 = insertelement <2 x float> %7717, float %4054, i64 1, !dbg !43 + %7719 = insertelement <2 x float> poison, float %7061, i64 0, !dbg !43 + %7720 = insertelement <2 x float> %7719, float %7062, i64 1, !dbg !43 + %7721 = fsub <2 x float> %7718, %7720, !dbg !43 + %7722 = insertelement <2 x float> poison, float %4055, i64 0, !dbg !43 + %7723 = insertelement <2 x float> %7722, float %4056, i64 1, !dbg !43 + %7724 = insertelement <2 x float> poison, float %7063, i64 0, !dbg !43 + %7725 = insertelement <2 x float> %7724, float %7064, i64 1, !dbg !43 + %7726 = fsub <2 x float> %7723, %7725, !dbg !43 + %7727 = insertelement <2 x float> poison, float %4057, i64 0, !dbg !43 + %7728 = insertelement <2 x float> %7727, float %4058, i64 1, !dbg !43 + %7729 = insertelement <2 x float> poison, float %7065, i64 0, !dbg !43 + %7730 = insertelement <2 x float> %7729, float %7066, i64 1, !dbg !43 + %7731 = fsub <2 x float> %7728, %7730, !dbg !43 + %7732 = insertelement <2 x float> poison, float %4059, i64 0, !dbg !43 + %7733 = insertelement <2 x float> %7732, float %4060, i64 1, !dbg !43 + %7734 = insertelement <2 x float> poison, float %7067, i64 0, !dbg !43 + %7735 = insertelement <2 x float> %7734, float %7068, i64 1, !dbg !43 + %7736 = fsub <2 x float> %7733, %7735, !dbg !43 + %7737 = insertelement <2 x float> poison, float %4061, i64 0, !dbg !43 + %7738 = insertelement <2 x float> %7737, float %4062, i64 1, !dbg !43 + %7739 = insertelement <2 x float> poison, float %7069, i64 0, !dbg !43 + %7740 = insertelement <2 x float> %7739, float %7070, i64 1, !dbg !43 + %7741 = fsub <2 x float> %7738, %7740, !dbg !43 + %7742 = insertelement <2 x float> poison, float %4063, i64 0, !dbg !43 + %7743 = insertelement <2 x float> %7742, float %4064, i64 1, !dbg !43 + %7744 = insertelement <2 x float> poison, float %7071, i64 0, !dbg !43 + %7745 = insertelement <2 x float> %7744, float %7072, i64 1, !dbg !43 + %7746 = fsub <2 x float> %7743, %7745, !dbg !43 + %7747 = insertelement <2 x float> poison, float %4065, i64 0, !dbg !43 + %7748 = insertelement <2 x float> %7747, float %4066, i64 1, !dbg !43 + %7749 = insertelement <2 x float> poison, float %7073, i64 0, !dbg !43 + %7750 = insertelement <2 x float> %7749, float %7074, i64 1, !dbg !43 + %7751 = fsub <2 x float> %7748, %7750, !dbg !43 + %7752 = insertelement <2 x float> poison, float %4067, i64 0, !dbg !43 + %7753 = insertelement <2 x float> %7752, float %4068, i64 1, !dbg !43 + %7754 = insertelement <2 x float> poison, float %7075, i64 0, !dbg !43 + %7755 = insertelement <2 x float> %7754, float %7076, i64 1, !dbg !43 + %7756 = fsub <2 x float> %7753, %7755, !dbg !43 + %7757 = insertelement <2 x float> poison, float %4069, i64 0, !dbg !43 + %7758 = insertelement <2 x float> %7757, float %4070, i64 1, !dbg !43 + %7759 = insertelement <2 x float> poison, float %7077, i64 0, !dbg !43 + %7760 = insertelement <2 x float> %7759, float %7078, i64 1, !dbg !43 + %7761 = fsub <2 x float> %7758, %7760, !dbg !43 + %7762 = insertelement <2 x float> poison, float %4071, i64 0, !dbg !43 + %7763 = insertelement <2 x float> %7762, float %4072, i64 1, !dbg !43 + %7764 = insertelement <2 x float> poison, float %7079, i64 0, !dbg !43 + %7765 = insertelement <2 x float> %7764, float %7080, i64 1, !dbg !43 + %7766 = fsub <2 x float> %7763, %7765, !dbg !43 + %7767 = insertelement <2 x float> poison, float %4073, i64 0, !dbg !43 + %7768 = insertelement <2 x float> %7767, float %4074, i64 1, !dbg !43 + %7769 = insertelement <2 x float> poison, float %7081, i64 0, !dbg !43 + %7770 = insertelement <2 x float> %7769, float %7082, i64 1, !dbg !43 + %7771 = fsub <2 x float> %7768, %7770, !dbg !43 + %7772 = insertelement <2 x float> poison, float %4075, i64 0, !dbg !43 + %7773 = insertelement <2 x float> %7772, float %4076, i64 1, !dbg !43 + %7774 = insertelement <2 x float> poison, float %7083, i64 0, !dbg !43 + %7775 = insertelement <2 x float> %7774, float %7084, i64 1, !dbg !43 + %7776 = fsub <2 x float> %7773, %7775, !dbg !43 + %7777 = insertelement <2 x float> poison, float %4077, i64 0, !dbg !43 + %7778 = insertelement <2 x float> %7777, float %4078, i64 1, !dbg !43 + %7779 = insertelement <2 x float> poison, float %7085, i64 0, !dbg !43 + %7780 = insertelement <2 x float> %7779, float %7086, i64 1, !dbg !43 + %7781 = fsub <2 x float> %7778, %7780, !dbg !43 + %7782 = insertelement <2 x float> poison, float %4079, i64 0, !dbg !43 + %7783 = insertelement <2 x float> %7782, float %4080, i64 1, !dbg !43 + %7784 = insertelement <2 x float> poison, float %7087, i64 0, !dbg !43 + %7785 = insertelement <2 x float> %7784, float %7088, i64 1, !dbg !43 + %7786 = fsub <2 x float> %7783, %7785, !dbg !43 + %7787 = insertelement <2 x float> poison, float %4081, i64 0, !dbg !43 + %7788 = insertelement <2 x float> %7787, float %4082, i64 1, !dbg !43 + %7789 = insertelement <2 x float> poison, float %7089, i64 0, !dbg !43 + %7790 = insertelement <2 x float> %7789, float %7090, i64 1, !dbg !43 + %7791 = fsub <2 x float> %7788, %7790, !dbg !43 + %7792 = insertelement <2 x float> poison, float %4083, i64 0, !dbg !43 + %7793 = insertelement <2 x float> %7792, float %4084, i64 1, !dbg !43 + %7794 = insertelement <2 x float> poison, float %7091, i64 0, !dbg !43 + %7795 = insertelement <2 x float> %7794, float %7092, i64 1, !dbg !43 + %7796 = fsub <2 x float> %7793, %7795, !dbg !43 + %7797 = insertelement <2 x float> poison, float %4085, i64 0, !dbg !43 + %7798 = insertelement <2 x float> %7797, float %4086, i64 1, !dbg !43 + %7799 = insertelement <2 x float> poison, float %7093, i64 0, !dbg !43 + %7800 = insertelement <2 x float> %7799, float %7094, i64 1, !dbg !43 + %7801 = fsub <2 x float> %7798, %7800, !dbg !43 + %7802 = insertelement <2 x float> poison, float %4087, i64 0, !dbg !43 + %7803 = insertelement <2 x float> %7802, float %4088, i64 1, !dbg !43 + %7804 = insertelement <2 x float> poison, float %7095, i64 0, !dbg !43 + %7805 = insertelement <2 x float> %7804, float %7096, i64 1, !dbg !43 + %7806 = fsub <2 x float> %7803, %7805, !dbg !43 + %7807 = insertelement <2 x float> poison, float %4089, i64 0, !dbg !43 + %7808 = insertelement <2 x float> %7807, float %4090, i64 1, !dbg !43 + %7809 = insertelement <2 x float> poison, float %7097, i64 0, !dbg !43 + %7810 = insertelement <2 x float> %7809, float %7098, i64 1, !dbg !43 + %7811 = fsub <2 x float> %7808, %7810, !dbg !43 + %7812 = insertelement <2 x float> poison, float %4091, i64 0, !dbg !43 + %7813 = insertelement <2 x float> %7812, float %4092, i64 1, !dbg !43 + %7814 = insertelement <2 x float> poison, float %7099, i64 0, !dbg !43 + %7815 = insertelement <2 x float> %7814, float %7100, i64 1, !dbg !43 + %7816 = fsub <2 x float> %7813, %7815, !dbg !43 + %7817 = insertelement <2 x float> poison, float %4093, i64 0, !dbg !43 + %7818 = insertelement <2 x float> %7817, float %4094, i64 1, !dbg !43 + %7819 = insertelement <2 x float> poison, float %7101, i64 0, !dbg !43 + %7820 = insertelement <2 x float> %7819, float %7102, i64 1, !dbg !43 + %7821 = fsub <2 x float> %7818, %7820, !dbg !43 + %7822 = insertelement <2 x float> poison, float %4095, i64 0, !dbg !43 + %7823 = insertelement <2 x float> %7822, float %4096, i64 1, !dbg !43 + %7824 = insertelement <2 x float> poison, float %7103, i64 0, !dbg !43 + %7825 = insertelement <2 x float> %7824, float %7104, i64 1, !dbg !43 + %7826 = fsub <2 x float> %7823, %7825, !dbg !43 + %7827 = insertelement <2 x float> poison, float %4097, i64 0, !dbg !43 + %7828 = insertelement <2 x float> %7827, float %4098, i64 1, !dbg !43 + %7829 = insertelement <2 x float> poison, float %7105, i64 0, !dbg !43 + %7830 = insertelement <2 x float> %7829, float %7106, i64 1, !dbg !43 + %7831 = fsub <2 x float> %7828, %7830, !dbg !43 + %7832 = insertelement <2 x float> poison, float %4099, i64 0, !dbg !43 + %7833 = insertelement <2 x float> %7832, float %4100, i64 1, !dbg !43 + %7834 = insertelement <2 x float> poison, float %7107, i64 0, !dbg !43 + %7835 = insertelement <2 x float> %7834, float %7108, i64 1, !dbg !43 + %7836 = fsub <2 x float> %7833, %7835, !dbg !43 + %7837 = insertelement <2 x float> poison, float %4101, i64 0, !dbg !43 + %7838 = insertelement <2 x float> %7837, float %4102, i64 1, !dbg !43 + %7839 = insertelement <2 x float> poison, float %7109, i64 0, !dbg !43 + %7840 = insertelement <2 x float> %7839, float %7110, i64 1, !dbg !43 + %7841 = fsub <2 x float> %7838, %7840, !dbg !43 + %7842 = insertelement <2 x float> poison, float %4103, i64 0, !dbg !43 + %7843 = insertelement <2 x float> %7842, float %4104, i64 1, !dbg !43 + %7844 = insertelement <2 x float> poison, float %7111, i64 0, !dbg !43 + %7845 = insertelement <2 x float> %7844, float %7112, i64 1, !dbg !43 + %7846 = fsub <2 x float> %7843, %7845, !dbg !43 + %7847 = insertelement <2 x float> poison, float %4105, i64 0, !dbg !43 + %7848 = insertelement <2 x float> %7847, float %4106, i64 1, !dbg !43 + %7849 = insertelement <2 x float> poison, float %7113, i64 0, !dbg !43 + %7850 = insertelement <2 x float> %7849, float %7114, i64 1, !dbg !43 + %7851 = fsub <2 x float> %7848, %7850, !dbg !43 + %7852 = insertelement <2 x float> poison, float %4107, i64 0, !dbg !43 + %7853 = insertelement <2 x float> %7852, float %4108, i64 1, !dbg !43 + %7854 = insertelement <2 x float> poison, float %7115, i64 0, !dbg !43 + %7855 = insertelement <2 x float> %7854, float %7116, i64 1, !dbg !43 + %7856 = fsub <2 x float> %7853, %7855, !dbg !43 + %7857 = insertelement <2 x float> poison, float %4109, i64 0, !dbg !43 + %7858 = insertelement <2 x float> %7857, float %4110, i64 1, !dbg !43 + %7859 = insertelement <2 x float> poison, float %7117, i64 0, !dbg !43 + %7860 = insertelement <2 x float> %7859, float %7118, i64 1, !dbg !43 + %7861 = fsub <2 x float> %7858, %7860, !dbg !43 + %7862 = insertelement <2 x float> poison, float %4111, i64 0, !dbg !43 + %7863 = insertelement <2 x float> %7862, float %4112, i64 1, !dbg !43 + %7864 = insertelement <2 x float> poison, float %7119, i64 0, !dbg !43 + %7865 = insertelement <2 x float> %7864, float %7120, i64 1, !dbg !43 + %7866 = fsub <2 x float> %7863, %7865, !dbg !43 + %7867 = insertelement <2 x float> poison, float %4113, i64 0, !dbg !43 + %7868 = insertelement <2 x float> %7867, float %4114, i64 1, !dbg !43 + %7869 = insertelement <2 x float> poison, float %7121, i64 0, !dbg !43 + %7870 = insertelement <2 x float> %7869, float %7122, i64 1, !dbg !43 + %7871 = fsub <2 x float> %7868, %7870, !dbg !43 + %7872 = insertelement <2 x float> poison, float %4115, i64 0, !dbg !43 + %7873 = insertelement <2 x float> %7872, float %4116, i64 1, !dbg !43 + %7874 = insertelement <2 x float> poison, float %7123, i64 0, !dbg !43 + %7875 = insertelement <2 x float> %7874, float %7124, i64 1, !dbg !43 + %7876 = fsub <2 x float> %7873, %7875, !dbg !43 + %7877 = insertelement <2 x float> poison, float %4117, i64 0, !dbg !43 + %7878 = insertelement <2 x float> %7877, float %4118, i64 1, !dbg !43 + %7879 = insertelement <2 x float> poison, float %7125, i64 0, !dbg !43 + %7880 = insertelement <2 x float> %7879, float %7126, i64 1, !dbg !43 + %7881 = fsub <2 x float> %7878, %7880, !dbg !43 + %7882 = insertelement <2 x float> poison, float %4119, i64 0, !dbg !43 + %7883 = insertelement <2 x float> %7882, float %4120, i64 1, !dbg !43 + %7884 = insertelement <2 x float> poison, float %7127, i64 0, !dbg !43 + %7885 = insertelement <2 x float> %7884, float %7128, i64 1, !dbg !43 + %7886 = fsub <2 x float> %7883, %7885, !dbg !43 + %7887 = insertelement <2 x float> poison, float %4121, i64 0, !dbg !43 + %7888 = insertelement <2 x float> %7887, float %4122, i64 1, !dbg !43 + %7889 = insertelement <2 x float> poison, float %7129, i64 0, !dbg !43 + %7890 = insertelement <2 x float> %7889, float %7130, i64 1, !dbg !43 + %7891 = fsub <2 x float> %7888, %7890, !dbg !43 + %7892 = insertelement <2 x float> poison, float %4123, i64 0, !dbg !43 + %7893 = insertelement <2 x float> %7892, float %4124, i64 1, !dbg !43 + %7894 = insertelement <2 x float> poison, float %7131, i64 0, !dbg !43 + %7895 = insertelement <2 x float> %7894, float %7132, i64 1, !dbg !43 + %7896 = fsub <2 x float> %7893, %7895, !dbg !43 + %7897 = insertelement <2 x float> poison, float %4125, i64 0, !dbg !43 + %7898 = insertelement <2 x float> %7897, float %4126, i64 1, !dbg !43 + %7899 = insertelement <2 x float> poison, float %7133, i64 0, !dbg !43 + %7900 = insertelement <2 x float> %7899, float %7134, i64 1, !dbg !43 + %7901 = fsub <2 x float> %7898, %7900, !dbg !43 + %7902 = insertelement <2 x float> poison, float %4127, i64 0, !dbg !43 + %7903 = insertelement <2 x float> %7902, float %4128, i64 1, !dbg !43 + %7904 = insertelement <2 x float> poison, float %7135, i64 0, !dbg !43 + %7905 = insertelement <2 x float> %7904, float %7136, i64 1, !dbg !43 + %7906 = fsub <2 x float> %7903, %7905, !dbg !43 + %7907 = insertelement <2 x float> poison, float %4129, i64 0, !dbg !43 + %7908 = insertelement <2 x float> %7907, float %4130, i64 1, !dbg !43 + %7909 = insertelement <2 x float> poison, float %7137, i64 0, !dbg !43 + %7910 = insertelement <2 x float> %7909, float %7138, i64 1, !dbg !43 + %7911 = fsub <2 x float> %7908, %7910, !dbg !43 + %7912 = insertelement <2 x float> poison, float %4131, i64 0, !dbg !43 + %7913 = insertelement <2 x float> %7912, float %4132, i64 1, !dbg !43 + %7914 = insertelement <2 x float> poison, float %7139, i64 0, !dbg !43 + %7915 = insertelement <2 x float> %7914, float %7140, i64 1, !dbg !43 + %7916 = fsub <2 x float> %7913, %7915, !dbg !43 + %7917 = insertelement <2 x float> poison, float %4133, i64 0, !dbg !43 + %7918 = insertelement <2 x float> %7917, float %4134, i64 1, !dbg !43 + %7919 = insertelement <2 x float> poison, float %7141, i64 0, !dbg !43 + %7920 = insertelement <2 x float> %7919, float %7142, i64 1, !dbg !43 + %7921 = fsub <2 x float> %7918, %7920, !dbg !43 + %7922 = insertelement <2 x float> poison, float %4135, i64 0, !dbg !43 + %7923 = insertelement <2 x float> %7922, float %4136, i64 1, !dbg !43 + %7924 = insertelement <2 x float> poison, float %7143, i64 0, !dbg !43 + %7925 = insertelement <2 x float> %7924, float %7144, i64 1, !dbg !43 + %7926 = fsub <2 x float> %7923, %7925, !dbg !43 + %7927 = insertelement <2 x float> poison, float %4137, i64 0, !dbg !43 + %7928 = insertelement <2 x float> %7927, float %4138, i64 1, !dbg !43 + %7929 = insertelement <2 x float> poison, float %7145, i64 0, !dbg !43 + %7930 = insertelement <2 x float> %7929, float %7146, i64 1, !dbg !43 + %7931 = fsub <2 x float> %7928, %7930, !dbg !43 + %7932 = insertelement <2 x float> poison, float %4139, i64 0, !dbg !43 + %7933 = insertelement <2 x float> %7932, float %4140, i64 1, !dbg !43 + %7934 = insertelement <2 x float> poison, float %7147, i64 0, !dbg !43 + %7935 = insertelement <2 x float> %7934, float %7148, i64 1, !dbg !43 + %7936 = fsub <2 x float> %7933, %7935, !dbg !43 + %7937 = insertelement <2 x float> poison, float %4141, i64 0, !dbg !43 + %7938 = insertelement <2 x float> %7937, float %4142, i64 1, !dbg !43 + %7939 = insertelement <2 x float> poison, float %7149, i64 0, !dbg !43 + %7940 = insertelement <2 x float> %7939, float %7150, i64 1, !dbg !43 + %7941 = fsub <2 x float> %7938, %7940, !dbg !43 + %7942 = insertelement <2 x float> poison, float %4143, i64 0, !dbg !43 + %7943 = insertelement <2 x float> %7942, float %4144, i64 1, !dbg !43 + %7944 = insertelement <2 x float> poison, float %7151, i64 0, !dbg !43 + %7945 = insertelement <2 x float> %7944, float %7152, i64 1, !dbg !43 + %7946 = fsub <2 x float> %7943, %7945, !dbg !43 + %7947 = insertelement <2 x float> poison, float %4145, i64 0, !dbg !43 + %7948 = insertelement <2 x float> %7947, float %4146, i64 1, !dbg !43 + %7949 = insertelement <2 x float> poison, float %7153, i64 0, !dbg !43 + %7950 = insertelement <2 x float> %7949, float %7154, i64 1, !dbg !43 + %7951 = fsub <2 x float> %7948, %7950, !dbg !43 + %7952 = insertelement <2 x float> poison, float %4147, i64 0, !dbg !43 + %7953 = insertelement <2 x float> %7952, float %4148, i64 1, !dbg !43 + %7954 = insertelement <2 x float> poison, float %7155, i64 0, !dbg !43 + %7955 = insertelement <2 x float> %7954, float %7156, i64 1, !dbg !43 + %7956 = fsub <2 x float> %7953, %7955, !dbg !43 + %7957 = insertelement <2 x float> poison, float %4149, i64 0, !dbg !43 + %7958 = insertelement <2 x float> %7957, float %4150, i64 1, !dbg !43 + %7959 = insertelement <2 x float> poison, float %7157, i64 0, !dbg !43 + %7960 = insertelement <2 x float> %7959, float %7158, i64 1, !dbg !43 + %7961 = fsub <2 x float> %7958, %7960, !dbg !43 + %7962 = insertelement <2 x float> poison, float %4151, i64 0, !dbg !43 + %7963 = insertelement <2 x float> %7962, float %4152, i64 1, !dbg !43 + %7964 = insertelement <2 x float> poison, float %7159, i64 0, !dbg !43 + %7965 = insertelement <2 x float> %7964, float %7160, i64 1, !dbg !43 + %7966 = fsub <2 x float> %7963, %7965, !dbg !43 + %7967 = insertelement <2 x float> poison, float %4153, i64 0, !dbg !43 + %7968 = insertelement <2 x float> %7967, float %4154, i64 1, !dbg !43 + %7969 = insertelement <2 x float> poison, float %7161, i64 0, !dbg !43 + %7970 = insertelement <2 x float> %7969, float %7162, i64 1, !dbg !43 + %7971 = fsub <2 x float> %7968, %7970, !dbg !43 + %7972 = insertelement <2 x float> poison, float %4155, i64 0, !dbg !43 + %7973 = insertelement <2 x float> %7972, float %4156, i64 1, !dbg !43 + %7974 = insertelement <2 x float> poison, float %7163, i64 0, !dbg !43 + %7975 = insertelement <2 x float> %7974, float %7164, i64 1, !dbg !43 + %7976 = fsub <2 x float> %7973, %7975, !dbg !43 + %7977 = insertelement <2 x float> poison, float %4157, i64 0, !dbg !43 + %7978 = insertelement <2 x float> %7977, float %4158, i64 1, !dbg !43 + %7979 = insertelement <2 x float> poison, float %7165, i64 0, !dbg !43 + %7980 = insertelement <2 x float> %7979, float %7166, i64 1, !dbg !43 + %7981 = fsub <2 x float> %7978, %7980, !dbg !43 + %7982 = insertelement <2 x float> poison, float %4159, i64 0, !dbg !43 + %7983 = insertelement <2 x float> %7982, float %4160, i64 1, !dbg !43 + %7984 = insertelement <2 x float> poison, float %7167, i64 0, !dbg !43 + %7985 = insertelement <2 x float> %7984, float %7168, i64 1, !dbg !43 + %7986 = fsub <2 x float> %7983, %7985, !dbg !43 + %7987 = insertelement <2 x float> poison, float %4161, i64 0, !dbg !43 + %7988 = insertelement <2 x float> %7987, float %4162, i64 1, !dbg !43 + %7989 = insertelement <2 x float> poison, float %7169, i64 0, !dbg !43 + %7990 = insertelement <2 x float> %7989, float %7170, i64 1, !dbg !43 + %7991 = fsub <2 x float> %7988, %7990, !dbg !43 + %7992 = insertelement <2 x float> poison, float %4163, i64 0, !dbg !43 + %7993 = insertelement <2 x float> %7992, float %4164, i64 1, !dbg !43 + %7994 = insertelement <2 x float> poison, float %7171, i64 0, !dbg !43 + %7995 = insertelement <2 x float> %7994, float %7172, i64 1, !dbg !43 + %7996 = fsub <2 x float> %7993, %7995, !dbg !43 + %7997 = insertelement <2 x float> poison, float %4165, i64 0, !dbg !43 + %7998 = insertelement <2 x float> %7997, float %4166, i64 1, !dbg !43 + %7999 = insertelement <2 x float> poison, float %7173, i64 0, !dbg !43 + %8000 = insertelement <2 x float> %7999, float %7174, i64 1, !dbg !43 + %8001 = fsub <2 x float> %7998, %8000, !dbg !43 + %8002 = insertelement <2 x float> poison, float %4167, i64 0, !dbg !43 + %8003 = insertelement <2 x float> %8002, float %4168, i64 1, !dbg !43 + %8004 = insertelement <2 x float> poison, float %7175, i64 0, !dbg !43 + %8005 = insertelement <2 x float> %8004, float %7176, i64 1, !dbg !43 + %8006 = fsub <2 x float> %8003, %8005, !dbg !43 + %8007 = insertelement <2 x float> poison, float %4169, i64 0, !dbg !43 + %8008 = insertelement <2 x float> %8007, float %4170, i64 1, !dbg !43 + %8009 = insertelement <2 x float> poison, float %7177, i64 0, !dbg !43 + %8010 = insertelement <2 x float> %8009, float %7178, i64 1, !dbg !43 + %8011 = fsub <2 x float> %8008, %8010, !dbg !43 + %8012 = insertelement <2 x float> poison, float %4171, i64 0, !dbg !43 + %8013 = insertelement <2 x float> %8012, float %4172, i64 1, !dbg !43 + %8014 = insertelement <2 x float> poison, float %7179, i64 0, !dbg !43 + %8015 = insertelement <2 x float> %8014, float %7180, i64 1, !dbg !43 + %8016 = fsub <2 x float> %8013, %8015, !dbg !43 + %8017 = insertelement <2 x float> poison, float %4173, i64 0, !dbg !43 + %8018 = insertelement <2 x float> %8017, float %4174, i64 1, !dbg !43 + %8019 = insertelement <2 x float> poison, float %7181, i64 0, !dbg !43 + %8020 = insertelement <2 x float> %8019, float %7182, i64 1, !dbg !43 + %8021 = fsub <2 x float> %8018, %8020, !dbg !43 + %8022 = insertelement <2 x float> poison, float %4175, i64 0, !dbg !43 + %8023 = insertelement <2 x float> %8022, float %4176, i64 1, !dbg !43 + %8024 = insertelement <2 x float> poison, float %7183, i64 0, !dbg !43 + %8025 = insertelement <2 x float> %8024, float %7184, i64 1, !dbg !43 + %8026 = fsub <2 x float> %8023, %8025, !dbg !43 + %8027 = insertelement <2 x float> poison, float %4177, i64 0, !dbg !43 + %8028 = insertelement <2 x float> %8027, float %4178, i64 1, !dbg !43 + %8029 = insertelement <2 x float> poison, float %7185, i64 0, !dbg !43 + %8030 = insertelement <2 x float> %8029, float %7186, i64 1, !dbg !43 + %8031 = fsub <2 x float> %8028, %8030, !dbg !43 + %8032 = insertelement <2 x float> poison, float %4179, i64 0, !dbg !43 + %8033 = insertelement <2 x float> %8032, float %4180, i64 1, !dbg !43 + %8034 = insertelement <2 x float> poison, float %7187, i64 0, !dbg !43 + %8035 = insertelement <2 x float> %8034, float %7188, i64 1, !dbg !43 + %8036 = fsub <2 x float> %8033, %8035, !dbg !43 + %8037 = insertelement <2 x float> poison, float %4181, i64 0, !dbg !43 + %8038 = insertelement <2 x float> %8037, float %4182, i64 1, !dbg !43 + %8039 = insertelement <2 x float> poison, float %7189, i64 0, !dbg !43 + %8040 = insertelement <2 x float> %8039, float %7190, i64 1, !dbg !43 + %8041 = fsub <2 x float> %8038, %8040, !dbg !43 + %8042 = insertelement <2 x float> poison, float %4183, i64 0, !dbg !43 + %8043 = insertelement <2 x float> %8042, float %4184, i64 1, !dbg !43 + %8044 = insertelement <2 x float> poison, float %7191, i64 0, !dbg !43 + %8045 = insertelement <2 x float> %8044, float %7192, i64 1, !dbg !43 + %8046 = fsub <2 x float> %8043, %8045, !dbg !43 + %8047 = insertelement <2 x float> poison, float %4185, i64 0, !dbg !43 + %8048 = insertelement <2 x float> %8047, float %4186, i64 1, !dbg !43 + %8049 = insertelement <2 x float> poison, float %7193, i64 0, !dbg !43 + %8050 = insertelement <2 x float> %8049, float %7194, i64 1, !dbg !43 + %8051 = fsub <2 x float> %8048, %8050, !dbg !43 + %8052 = insertelement <2 x float> poison, float %4187, i64 0, !dbg !43 + %8053 = insertelement <2 x float> %8052, float %4188, i64 1, !dbg !43 + %8054 = insertelement <2 x float> poison, float %7195, i64 0, !dbg !43 + %8055 = insertelement <2 x float> %8054, float %7196, i64 1, !dbg !43 + %8056 = fsub <2 x float> %8053, %8055, !dbg !43 + %8057 = insertelement <2 x float> poison, float %4189, i64 0, !dbg !43 + %8058 = insertelement <2 x float> %8057, float %4190, i64 1, !dbg !43 + %8059 = insertelement <2 x float> poison, float %7197, i64 0, !dbg !43 + %8060 = insertelement <2 x float> %8059, float %7198, i64 1, !dbg !43 + %8061 = fsub <2 x float> %8058, %8060, !dbg !43 + %8062 = insertelement <2 x float> poison, float %4191, i64 0, !dbg !43 + %8063 = insertelement <2 x float> %8062, float %4192, i64 1, !dbg !43 + %8064 = insertelement <2 x float> poison, float %7199, i64 0, !dbg !43 + %8065 = insertelement <2 x float> %8064, float %7200, i64 1, !dbg !43 + %8066 = fsub <2 x float> %8063, %8065, !dbg !43 + %8067 = insertelement <2 x float> poison, float %4193, i64 0, !dbg !43 + %8068 = insertelement <2 x float> %8067, float %4194, i64 1, !dbg !43 + %8069 = insertelement <2 x float> poison, float %7201, i64 0, !dbg !43 + %8070 = insertelement <2 x float> %8069, float %7202, i64 1, !dbg !43 + %8071 = fsub <2 x float> %8068, %8070, !dbg !43 + %8072 = insertelement <2 x float> poison, float %4195, i64 0, !dbg !43 + %8073 = insertelement <2 x float> %8072, float %4196, i64 1, !dbg !43 + %8074 = insertelement <2 x float> poison, float %7203, i64 0, !dbg !43 + %8075 = insertelement <2 x float> %8074, float %7204, i64 1, !dbg !43 + %8076 = fsub <2 x float> %8073, %8075, !dbg !43 + %8077 = insertelement <2 x float> poison, float %4197, i64 0, !dbg !43 + %8078 = insertelement <2 x float> %8077, float %4198, i64 1, !dbg !43 + %8079 = insertelement <2 x float> poison, float %7205, i64 0, !dbg !43 + %8080 = insertelement <2 x float> %8079, float %7206, i64 1, !dbg !43 + %8081 = fsub <2 x float> %8078, %8080, !dbg !43 + %8082 = insertelement <2 x float> poison, float %4199, i64 0, !dbg !43 + %8083 = insertelement <2 x float> %8082, float %4200, i64 1, !dbg !43 + %8084 = insertelement <2 x float> poison, float %7207, i64 0, !dbg !43 + %8085 = insertelement <2 x float> %8084, float %7208, i64 1, !dbg !43 + %8086 = fsub <2 x float> %8083, %8085, !dbg !43 + %8087 = insertelement <2 x float> poison, float %4201, i64 0, !dbg !43 + %8088 = insertelement <2 x float> %8087, float %4202, i64 1, !dbg !43 + %8089 = insertelement <2 x float> poison, float %7209, i64 0, !dbg !43 + %8090 = insertelement <2 x float> %8089, float %7210, i64 1, !dbg !43 + %8091 = fsub <2 x float> %8088, %8090, !dbg !43 + %8092 = insertelement <2 x float> poison, float %4203, i64 0, !dbg !43 + %8093 = insertelement <2 x float> %8092, float %4204, i64 1, !dbg !43 + %8094 = insertelement <2 x float> poison, float %7211, i64 0, !dbg !43 + %8095 = insertelement <2 x float> %8094, float %7212, i64 1, !dbg !43 + %8096 = fsub <2 x float> %8093, %8095, !dbg !43 + %8097 = insertelement <2 x float> poison, float %4205, i64 0, !dbg !43 + %8098 = insertelement <2 x float> %8097, float %4206, i64 1, !dbg !43 + %8099 = insertelement <2 x float> poison, float %7213, i64 0, !dbg !43 + %8100 = insertelement <2 x float> %8099, float %7214, i64 1, !dbg !43 + %8101 = fsub <2 x float> %8098, %8100, !dbg !43 + %8102 = insertelement <2 x float> poison, float %4207, i64 0, !dbg !43 + %8103 = insertelement <2 x float> %8102, float %4208, i64 1, !dbg !43 + %8104 = insertelement <2 x float> poison, float %7215, i64 0, !dbg !43 + %8105 = insertelement <2 x float> %8104, float %7216, i64 1, !dbg !43 + %8106 = fsub <2 x float> %8103, %8105, !dbg !43 + %8107 = insertelement <2 x float> poison, float %4209, i64 0, !dbg !43 + %8108 = insertelement <2 x float> %8107, float %4210, i64 1, !dbg !43 + %8109 = insertelement <2 x float> poison, float %7217, i64 0, !dbg !43 + %8110 = insertelement <2 x float> %8109, float %7218, i64 1, !dbg !43 + %8111 = fsub <2 x float> %8108, %8110, !dbg !43 + %8112 = insertelement <2 x float> poison, float %4211, i64 0, !dbg !43 + %8113 = insertelement <2 x float> %8112, float %4212, i64 1, !dbg !43 + %8114 = insertelement <2 x float> poison, float %7219, i64 0, !dbg !43 + %8115 = insertelement <2 x float> %8114, float %7220, i64 1, !dbg !43 + %8116 = fsub <2 x float> %8113, %8115, !dbg !43 + %8117 = insertelement <2 x float> poison, float %4213, i64 0, !dbg !43 + %8118 = insertelement <2 x float> %8117, float %4214, i64 1, !dbg !43 + %8119 = insertelement <2 x float> poison, float %7221, i64 0, !dbg !43 + %8120 = insertelement <2 x float> %8119, float %7222, i64 1, !dbg !43 + %8121 = fsub <2 x float> %8118, %8120, !dbg !43 + %8122 = insertelement <2 x float> poison, float %4215, i64 0, !dbg !43 + %8123 = insertelement <2 x float> %8122, float %4216, i64 1, !dbg !43 + %8124 = insertelement <2 x float> poison, float %7223, i64 0, !dbg !43 + %8125 = insertelement <2 x float> %8124, float %7224, i64 1, !dbg !43 + %8126 = fsub <2 x float> %8123, %8125, !dbg !43 + %8127 = insertelement <2 x float> poison, float %4217, i64 0, !dbg !43 + %8128 = insertelement <2 x float> %8127, float %4218, i64 1, !dbg !43 + %8129 = insertelement <2 x float> poison, float %7225, i64 0, !dbg !43 + %8130 = insertelement <2 x float> %8129, float %7226, i64 1, !dbg !43 + %8131 = fsub <2 x float> %8128, %8130, !dbg !43 + %8132 = insertelement <2 x float> poison, float %4219, i64 0, !dbg !43 + %8133 = insertelement <2 x float> %8132, float %4220, i64 1, !dbg !43 + %8134 = insertelement <2 x float> poison, float %7227, i64 0, !dbg !43 + %8135 = insertelement <2 x float> %8134, float %7228, i64 1, !dbg !43 + %8136 = fsub <2 x float> %8133, %8135, !dbg !43 + %8137 = insertelement <2 x float> poison, float %4221, i64 0, !dbg !43 + %8138 = insertelement <2 x float> %8137, float %4222, i64 1, !dbg !43 + %8139 = insertelement <2 x float> poison, float %7229, i64 0, !dbg !43 + %8140 = insertelement <2 x float> %8139, float %7230, i64 1, !dbg !43 + %8141 = fsub <2 x float> %8138, %8140, !dbg !43 + %8142 = insertelement <2 x float> poison, float %4223, i64 0, !dbg !43 + %8143 = insertelement <2 x float> %8142, float %4224, i64 1, !dbg !43 + %8144 = insertelement <2 x float> poison, float %7231, i64 0, !dbg !43 + %8145 = insertelement <2 x float> %8144, float %7232, i64 1, !dbg !43 + %8146 = fsub <2 x float> %8143, %8145, !dbg !43 + %8147 = insertelement <2 x float> poison, float %4225, i64 0, !dbg !43 + %8148 = insertelement <2 x float> %8147, float %4226, i64 1, !dbg !43 + %8149 = insertelement <2 x float> poison, float %7233, i64 0, !dbg !43 + %8150 = insertelement <2 x float> %8149, float %7234, i64 1, !dbg !43 + %8151 = fsub <2 x float> %8148, %8150, !dbg !43 + %8152 = insertelement <2 x float> poison, float %4227, i64 0, !dbg !43 + %8153 = insertelement <2 x float> %8152, float %4228, i64 1, !dbg !43 + %8154 = insertelement <2 x float> poison, float %7235, i64 0, !dbg !43 + %8155 = insertelement <2 x float> %8154, float %7236, i64 1, !dbg !43 + %8156 = fsub <2 x float> %8153, %8155, !dbg !43 + %8157 = insertelement <2 x float> poison, float %4229, i64 0, !dbg !43 + %8158 = insertelement <2 x float> %8157, float %4230, i64 1, !dbg !43 + %8159 = insertelement <2 x float> poison, float %7237, i64 0, !dbg !43 + %8160 = insertelement <2 x float> %8159, float %7238, i64 1, !dbg !43 + %8161 = fsub <2 x float> %8158, %8160, !dbg !43 + %8162 = insertelement <2 x float> poison, float %4231, i64 0, !dbg !43 + %8163 = insertelement <2 x float> %8162, float %4232, i64 1, !dbg !43 + %8164 = insertelement <2 x float> poison, float %7239, i64 0, !dbg !43 + %8165 = insertelement <2 x float> %8164, float %7240, i64 1, !dbg !43 + %8166 = fsub <2 x float> %8163, %8165, !dbg !43 + %8167 = insertelement <2 x float> poison, float %4233, i64 0, !dbg !43 + %8168 = insertelement <2 x float> %8167, float %4234, i64 1, !dbg !43 + %8169 = insertelement <2 x float> poison, float %7241, i64 0, !dbg !43 + %8170 = insertelement <2 x float> %8169, float %7242, i64 1, !dbg !43 + %8171 = fsub <2 x float> %8168, %8170, !dbg !43 + %8172 = insertelement <2 x float> poison, float %4235, i64 0, !dbg !43 + %8173 = insertelement <2 x float> %8172, float %4236, i64 1, !dbg !43 + %8174 = insertelement <2 x float> poison, float %7243, i64 0, !dbg !43 + %8175 = insertelement <2 x float> %8174, float %7244, i64 1, !dbg !43 + %8176 = fsub <2 x float> %8173, %8175, !dbg !43 + %8177 = insertelement <2 x float> poison, float %4237, i64 0, !dbg !43 + %8178 = insertelement <2 x float> %8177, float %4238, i64 1, !dbg !43 + %8179 = insertelement <2 x float> poison, float %7245, i64 0, !dbg !43 + %8180 = insertelement <2 x float> %8179, float %7246, i64 1, !dbg !43 + %8181 = fsub <2 x float> %8178, %8180, !dbg !43 + %8182 = insertelement <2 x float> poison, float %4239, i64 0, !dbg !43 + %8183 = insertelement <2 x float> %8182, float %4240, i64 1, !dbg !43 + %8184 = insertelement <2 x float> poison, float %7247, i64 0, !dbg !43 + %8185 = insertelement <2 x float> %8184, float %7248, i64 1, !dbg !43 + %8186 = fsub <2 x float> %8183, %8185, !dbg !43 + %8187 = insertelement <2 x float> poison, float %4241, i64 0, !dbg !43 + %8188 = insertelement <2 x float> %8187, float %4242, i64 1, !dbg !43 + %8189 = insertelement <2 x float> poison, float %7249, i64 0, !dbg !43 + %8190 = insertelement <2 x float> %8189, float %7250, i64 1, !dbg !43 + %8191 = fsub <2 x float> %8188, %8190, !dbg !43 + %8192 = insertelement <2 x float> poison, float %4243, i64 0, !dbg !43 + %8193 = insertelement <2 x float> %8192, float %4244, i64 1, !dbg !43 + %8194 = insertelement <2 x float> poison, float %7251, i64 0, !dbg !43 + %8195 = insertelement <2 x float> %8194, float %7252, i64 1, !dbg !43 + %8196 = fsub <2 x float> %8193, %8195, !dbg !43 + %8197 = insertelement <2 x float> poison, float %4245, i64 0, !dbg !43 + %8198 = insertelement <2 x float> %8197, float %4246, i64 1, !dbg !43 + %8199 = insertelement <2 x float> poison, float %7253, i64 0, !dbg !43 + %8200 = insertelement <2 x float> %8199, float %7254, i64 1, !dbg !43 + %8201 = fsub <2 x float> %8198, %8200, !dbg !43 + %8202 = insertelement <2 x float> poison, float %4247, i64 0, !dbg !43 + %8203 = insertelement <2 x float> %8202, float %4248, i64 1, !dbg !43 + %8204 = insertelement <2 x float> poison, float %7255, i64 0, !dbg !43 + %8205 = insertelement <2 x float> %8204, float %7256, i64 1, !dbg !43 + %8206 = fsub <2 x float> %8203, %8205, !dbg !43 + %8207 = insertelement <2 x float> poison, float %4249, i64 0, !dbg !43 + %8208 = insertelement <2 x float> %8207, float %4250, i64 1, !dbg !43 + %8209 = insertelement <2 x float> poison, float %7257, i64 0, !dbg !43 + %8210 = insertelement <2 x float> %8209, float %7258, i64 1, !dbg !43 + %8211 = fsub <2 x float> %8208, %8210, !dbg !43 + %8212 = insertelement <2 x float> poison, float %4251, i64 0, !dbg !43 + %8213 = insertelement <2 x float> %8212, float %4252, i64 1, !dbg !43 + %8214 = insertelement <2 x float> poison, float %7259, i64 0, !dbg !43 + %8215 = insertelement <2 x float> %8214, float %7260, i64 1, !dbg !43 + %8216 = fsub <2 x float> %8213, %8215, !dbg !43 + %8217 = insertelement <2 x float> poison, float %4253, i64 0, !dbg !43 + %8218 = insertelement <2 x float> %8217, float %4254, i64 1, !dbg !43 + %8219 = insertelement <2 x float> poison, float %7261, i64 0, !dbg !43 + %8220 = insertelement <2 x float> %8219, float %7262, i64 1, !dbg !43 + %8221 = fsub <2 x float> %8218, %8220, !dbg !43 + %8222 = insertelement <2 x float> poison, float %4255, i64 0, !dbg !43 + %8223 = insertelement <2 x float> %8222, float %4256, i64 1, !dbg !43 + %8224 = insertelement <2 x float> poison, float %7263, i64 0, !dbg !43 + %8225 = insertelement <2 x float> %8224, float %7264, i64 1, !dbg !43 + %8226 = fsub <2 x float> %8223, %8225, !dbg !43 + %8227 = insertelement <2 x float> poison, float %4257, i64 0, !dbg !43 + %8228 = insertelement <2 x float> %8227, float %4258, i64 1, !dbg !43 + %8229 = insertelement <2 x float> poison, float %7265, i64 0, !dbg !43 + %8230 = insertelement <2 x float> %8229, float %7266, i64 1, !dbg !43 + %8231 = fsub <2 x float> %8228, %8230, !dbg !43 + %8232 = insertelement <2 x float> poison, float %4259, i64 0, !dbg !43 + %8233 = insertelement <2 x float> %8232, float %4260, i64 1, !dbg !43 + %8234 = insertelement <2 x float> poison, float %7267, i64 0, !dbg !43 + %8235 = insertelement <2 x float> %8234, float %7268, i64 1, !dbg !43 + %8236 = fsub <2 x float> %8233, %8235, !dbg !43 + %8237 = insertelement <2 x float> poison, float %4261, i64 0, !dbg !43 + %8238 = insertelement <2 x float> %8237, float %4262, i64 1, !dbg !43 + %8239 = insertelement <2 x float> poison, float %7269, i64 0, !dbg !43 + %8240 = insertelement <2 x float> %8239, float %7270, i64 1, !dbg !43 + %8241 = fsub <2 x float> %8238, %8240, !dbg !43 + %8242 = insertelement <2 x float> poison, float %4263, i64 0, !dbg !43 + %8243 = insertelement <2 x float> %8242, float %4264, i64 1, !dbg !43 + %8244 = insertelement <2 x float> poison, float %7271, i64 0, !dbg !43 + %8245 = insertelement <2 x float> %8244, float %7272, i64 1, !dbg !43 + %8246 = fsub <2 x float> %8243, %8245, !dbg !43 + %8247 = insertelement <2 x float> poison, float %4265, i64 0, !dbg !43 + %8248 = insertelement <2 x float> %8247, float %4266, i64 1, !dbg !43 + %8249 = insertelement <2 x float> poison, float %7273, i64 0, !dbg !43 + %8250 = insertelement <2 x float> %8249, float %7274, i64 1, !dbg !43 + %8251 = fsub <2 x float> %8248, %8250, !dbg !43 + %8252 = insertelement <2 x float> poison, float %4267, i64 0, !dbg !43 + %8253 = insertelement <2 x float> %8252, float %4268, i64 1, !dbg !43 + %8254 = insertelement <2 x float> poison, float %7275, i64 0, !dbg !43 + %8255 = insertelement <2 x float> %8254, float %7276, i64 1, !dbg !43 + %8256 = fsub <2 x float> %8253, %8255, !dbg !43 + %8257 = insertelement <2 x float> poison, float %4269, i64 0, !dbg !43 + %8258 = insertelement <2 x float> %8257, float %4270, i64 1, !dbg !43 + %8259 = insertelement <2 x float> poison, float %7277, i64 0, !dbg !43 + %8260 = insertelement <2 x float> %8259, float %7278, i64 1, !dbg !43 + %8261 = fsub <2 x float> %8258, %8260, !dbg !43 + %8262 = insertelement <2 x float> poison, float %4271, i64 0, !dbg !43 + %8263 = insertelement <2 x float> %8262, float %4272, i64 1, !dbg !43 + %8264 = insertelement <2 x float> poison, float %7279, i64 0, !dbg !43 + %8265 = insertelement <2 x float> %8264, float %7280, i64 1, !dbg !43 + %8266 = fsub <2 x float> %8263, %8265, !dbg !43 + %8267 = insertelement <2 x float> poison, float %4273, i64 0, !dbg !43 + %8268 = insertelement <2 x float> %8267, float %4274, i64 1, !dbg !43 + %8269 = insertelement <2 x float> poison, float %7281, i64 0, !dbg !43 + %8270 = insertelement <2 x float> %8269, float %7282, i64 1, !dbg !43 + %8271 = fsub <2 x float> %8268, %8270, !dbg !43 + %8272 = insertelement <2 x float> poison, float %4275, i64 0, !dbg !43 + %8273 = insertelement <2 x float> %8272, float %4276, i64 1, !dbg !43 + %8274 = insertelement <2 x float> poison, float %7283, i64 0, !dbg !43 + %8275 = insertelement <2 x float> %8274, float %7284, i64 1, !dbg !43 + %8276 = fsub <2 x float> %8273, %8275, !dbg !43 + %8277 = insertelement <2 x float> poison, float %4277, i64 0, !dbg !43 + %8278 = insertelement <2 x float> %8277, float %4278, i64 1, !dbg !43 + %8279 = insertelement <2 x float> poison, float %7285, i64 0, !dbg !43 + %8280 = insertelement <2 x float> %8279, float %7286, i64 1, !dbg !43 + %8281 = fsub <2 x float> %8278, %8280, !dbg !43 + %8282 = insertelement <2 x float> poison, float %4279, i64 0, !dbg !43 + %8283 = insertelement <2 x float> %8282, float %4280, i64 1, !dbg !43 + %8284 = insertelement <2 x float> poison, float %7287, i64 0, !dbg !43 + %8285 = insertelement <2 x float> %8284, float %7288, i64 1, !dbg !43 + %8286 = fsub <2 x float> %8283, %8285, !dbg !43 + %8287 = insertelement <2 x float> poison, float %4281, i64 0, !dbg !43 + %8288 = insertelement <2 x float> %8287, float %4282, i64 1, !dbg !43 + %8289 = insertelement <2 x float> poison, float %7289, i64 0, !dbg !43 + %8290 = insertelement <2 x float> %8289, float %7290, i64 1, !dbg !43 + %8291 = fsub <2 x float> %8288, %8290, !dbg !43 + %8292 = insertelement <2 x float> poison, float %4283, i64 0, !dbg !43 + %8293 = insertelement <2 x float> %8292, float %4284, i64 1, !dbg !43 + %8294 = insertelement <2 x float> poison, float %7291, i64 0, !dbg !43 + %8295 = insertelement <2 x float> %8294, float %7292, i64 1, !dbg !43 + %8296 = fsub <2 x float> %8293, %8295, !dbg !43 + %8297 = insertelement <2 x float> poison, float %4285, i64 0, !dbg !43 + %8298 = insertelement <2 x float> %8297, float %4286, i64 1, !dbg !43 + %8299 = insertelement <2 x float> poison, float %7293, i64 0, !dbg !43 + %8300 = insertelement <2 x float> %8299, float %7294, i64 1, !dbg !43 + %8301 = fsub <2 x float> %8298, %8300, !dbg !43 + %8302 = insertelement <2 x float> poison, float %4287, i64 0, !dbg !43 + %8303 = insertelement <2 x float> %8302, float %4288, i64 1, !dbg !43 + %8304 = insertelement <2 x float> poison, float %7295, i64 0, !dbg !43 + %8305 = insertelement <2 x float> %8304, float %7296, i64 1, !dbg !43 + %8306 = fsub <2 x float> %8303, %8305, !dbg !43 + %8307 = insertelement <2 x float> poison, float %4289, i64 0, !dbg !43 + %8308 = insertelement <2 x float> %8307, float %4290, i64 1, !dbg !43 + %8309 = insertelement <2 x float> poison, float %7297, i64 0, !dbg !43 + %8310 = insertelement <2 x float> %8309, float %7298, i64 1, !dbg !43 + %8311 = fsub <2 x float> %8308, %8310, !dbg !43 + %8312 = insertelement <2 x float> poison, float %4291, i64 0, !dbg !43 + %8313 = insertelement <2 x float> %8312, float %4292, i64 1, !dbg !43 + %8314 = insertelement <2 x float> poison, float %7299, i64 0, !dbg !43 + %8315 = insertelement <2 x float> %8314, float %7300, i64 1, !dbg !43 + %8316 = fsub <2 x float> %8313, %8315, !dbg !43 + %8317 = insertelement <2 x float> poison, float %4293, i64 0, !dbg !43 + %8318 = insertelement <2 x float> %8317, float %4294, i64 1, !dbg !43 + %8319 = insertelement <2 x float> poison, float %7301, i64 0, !dbg !43 + %8320 = insertelement <2 x float> %8319, float %7302, i64 1, !dbg !43 + %8321 = fsub <2 x float> %8318, %8320, !dbg !43 + %8322 = insertelement <2 x float> poison, float %4295, i64 0, !dbg !43 + %8323 = insertelement <2 x float> %8322, float %4296, i64 1, !dbg !43 + %8324 = insertelement <2 x float> poison, float %7303, i64 0, !dbg !43 + %8325 = insertelement <2 x float> %8324, float %7304, i64 1, !dbg !43 + %8326 = fsub <2 x float> %8323, %8325, !dbg !43 + %8327 = insertelement <2 x float> poison, float %4297, i64 0, !dbg !43 + %8328 = insertelement <2 x float> %8327, float %4298, i64 1, !dbg !43 + %8329 = insertelement <2 x float> poison, float %7305, i64 0, !dbg !43 + %8330 = insertelement <2 x float> %8329, float %7306, i64 1, !dbg !43 + %8331 = fsub <2 x float> %8328, %8330, !dbg !43 + %8332 = insertelement <2 x float> poison, float %4299, i64 0, !dbg !43 + %8333 = insertelement <2 x float> %8332, float %4300, i64 1, !dbg !43 + %8334 = insertelement <2 x float> poison, float %7307, i64 0, !dbg !43 + %8335 = insertelement <2 x float> %8334, float %7308, i64 1, !dbg !43 + %8336 = fsub <2 x float> %8333, %8335, !dbg !43 + %8337 = insertelement <2 x float> poison, float %4301, i64 0, !dbg !43 + %8338 = insertelement <2 x float> %8337, float %4302, i64 1, !dbg !43 + %8339 = insertelement <2 x float> poison, float %7309, i64 0, !dbg !43 + %8340 = insertelement <2 x float> %8339, float %7310, i64 1, !dbg !43 + %8341 = fsub <2 x float> %8338, %8340, !dbg !43 + %8342 = insertelement <2 x float> poison, float %4303, i64 0, !dbg !43 + %8343 = insertelement <2 x float> %8342, float %4304, i64 1, !dbg !43 + %8344 = insertelement <2 x float> poison, float %7311, i64 0, !dbg !43 + %8345 = insertelement <2 x float> %8344, float %7312, i64 1, !dbg !43 + %8346 = fsub <2 x float> %8343, %8345, !dbg !43 + %8347 = insertelement <2 x float> poison, float %4305, i64 0, !dbg !43 + %8348 = insertelement <2 x float> %8347, float %4306, i64 1, !dbg !43 + %8349 = insertelement <2 x float> poison, float %7313, i64 0, !dbg !43 + %8350 = insertelement <2 x float> %8349, float %7314, i64 1, !dbg !43 + %8351 = fsub <2 x float> %8348, %8350, !dbg !43 + %8352 = insertelement <2 x float> poison, float %4307, i64 0, !dbg !43 + %8353 = insertelement <2 x float> %8352, float %4308, i64 1, !dbg !43 + %8354 = insertelement <2 x float> poison, float %7315, i64 0, !dbg !43 + %8355 = insertelement <2 x float> %8354, float %7316, i64 1, !dbg !43 + %8356 = fsub <2 x float> %8353, %8355, !dbg !43 + %8357 = insertelement <2 x float> poison, float %4309, i64 0, !dbg !43 + %8358 = insertelement <2 x float> %8357, float %4310, i64 1, !dbg !43 + %8359 = insertelement <2 x float> poison, float %7317, i64 0, !dbg !43 + %8360 = insertelement <2 x float> %8359, float %7318, i64 1, !dbg !43 + %8361 = fsub <2 x float> %8358, %8360, !dbg !43 + %8362 = insertelement <2 x float> poison, float %4311, i64 0, !dbg !43 + %8363 = insertelement <2 x float> %8362, float %4312, i64 1, !dbg !43 + %8364 = insertelement <2 x float> poison, float %7319, i64 0, !dbg !43 + %8365 = insertelement <2 x float> %8364, float %7320, i64 1, !dbg !43 + %8366 = fsub <2 x float> %8363, %8365, !dbg !43 + %8367 = insertelement <2 x float> poison, float %4313, i64 0, !dbg !43 + %8368 = insertelement <2 x float> %8367, float %4314, i64 1, !dbg !43 + %8369 = insertelement <2 x float> poison, float %7321, i64 0, !dbg !43 + %8370 = insertelement <2 x float> %8369, float %7322, i64 1, !dbg !43 + %8371 = fsub <2 x float> %8368, %8370, !dbg !43 + %8372 = insertelement <2 x float> poison, float %4315, i64 0, !dbg !43 + %8373 = insertelement <2 x float> %8372, float %4316, i64 1, !dbg !43 + %8374 = insertelement <2 x float> poison, float %7323, i64 0, !dbg !43 + %8375 = insertelement <2 x float> %8374, float %7324, i64 1, !dbg !43 + %8376 = fsub <2 x float> %8373, %8375, !dbg !43 + %8377 = insertelement <2 x float> poison, float %4317, i64 0, !dbg !43 + %8378 = insertelement <2 x float> %8377, float %4318, i64 1, !dbg !43 + %8379 = insertelement <2 x float> poison, float %7325, i64 0, !dbg !43 + %8380 = insertelement <2 x float> %8379, float %7326, i64 1, !dbg !43 + %8381 = fsub <2 x float> %8378, %8380, !dbg !43 + %8382 = insertelement <2 x float> poison, float %4319, i64 0, !dbg !43 + %8383 = insertelement <2 x float> %8382, float %4320, i64 1, !dbg !43 + %8384 = insertelement <2 x float> poison, float %7327, i64 0, !dbg !43 + %8385 = insertelement <2 x float> %8384, float %7328, i64 1, !dbg !43 + %8386 = fsub <2 x float> %8383, %8385, !dbg !43 + %8387 = insertelement <2 x float> poison, float %4321, i64 0, !dbg !43 + %8388 = insertelement <2 x float> %8387, float %4322, i64 1, !dbg !43 + %8389 = insertelement <2 x float> poison, float %7329, i64 0, !dbg !43 + %8390 = insertelement <2 x float> %8389, float %7330, i64 1, !dbg !43 + %8391 = fsub <2 x float> %8388, %8390, !dbg !43 + %8392 = insertelement <2 x float> poison, float %4323, i64 0, !dbg !43 + %8393 = insertelement <2 x float> %8392, float %4324, i64 1, !dbg !43 + %8394 = insertelement <2 x float> poison, float %7331, i64 0, !dbg !43 + %8395 = insertelement <2 x float> %8394, float %7332, i64 1, !dbg !43 + %8396 = fsub <2 x float> %8393, %8395, !dbg !43 + %8397 = insertelement <2 x float> poison, float %4325, i64 0, !dbg !43 + %8398 = insertelement <2 x float> %8397, float %4326, i64 1, !dbg !43 + %8399 = insertelement <2 x float> poison, float %7333, i64 0, !dbg !43 + %8400 = insertelement <2 x float> %8399, float %7334, i64 1, !dbg !43 + %8401 = fsub <2 x float> %8398, %8400, !dbg !43 + %8402 = insertelement <2 x float> poison, float %4327, i64 0, !dbg !43 + %8403 = insertelement <2 x float> %8402, float %4328, i64 1, !dbg !43 + %8404 = insertelement <2 x float> poison, float %7335, i64 0, !dbg !43 + %8405 = insertelement <2 x float> %8404, float %7336, i64 1, !dbg !43 + %8406 = fsub <2 x float> %8403, %8405, !dbg !43 + %8407 = insertelement <2 x float> poison, float %4329, i64 0, !dbg !43 + %8408 = insertelement <2 x float> %8407, float %4330, i64 1, !dbg !43 + %8409 = insertelement <2 x float> poison, float %7337, i64 0, !dbg !43 + %8410 = insertelement <2 x float> %8409, float %7338, i64 1, !dbg !43 + %8411 = fsub <2 x float> %8408, %8410, !dbg !43 + %8412 = insertelement <2 x float> poison, float %4331, i64 0, !dbg !43 + %8413 = insertelement <2 x float> %8412, float %4332, i64 1, !dbg !43 + %8414 = insertelement <2 x float> poison, float %7339, i64 0, !dbg !43 + %8415 = insertelement <2 x float> %8414, float %7340, i64 1, !dbg !43 + %8416 = fsub <2 x float> %8413, %8415, !dbg !43 + %8417 = insertelement <2 x float> poison, float %4333, i64 0, !dbg !43 + %8418 = insertelement <2 x float> %8417, float %4334, i64 1, !dbg !43 + %8419 = insertelement <2 x float> poison, float %7341, i64 0, !dbg !43 + %8420 = insertelement <2 x float> %8419, float %7342, i64 1, !dbg !43 + %8421 = fsub <2 x float> %8418, %8420, !dbg !43 + %8422 = insertelement <2 x float> poison, float %4335, i64 0, !dbg !43 + %8423 = insertelement <2 x float> %8422, float %4336, i64 1, !dbg !43 + %8424 = insertelement <2 x float> poison, float %7343, i64 0, !dbg !43 + %8425 = insertelement <2 x float> %8424, float %7344, i64 1, !dbg !43 + %8426 = fsub <2 x float> %8423, %8425, !dbg !43 + %8427 = insertelement <2 x float> poison, float %4337, i64 0, !dbg !43 + %8428 = insertelement <2 x float> %8427, float %4338, i64 1, !dbg !43 + %8429 = insertelement <2 x float> poison, float %7345, i64 0, !dbg !43 + %8430 = insertelement <2 x float> %8429, float %7346, i64 1, !dbg !43 + %8431 = fsub <2 x float> %8428, %8430, !dbg !43 + %8432 = insertelement <2 x float> poison, float %4339, i64 0, !dbg !43 + %8433 = insertelement <2 x float> %8432, float %4340, i64 1, !dbg !43 + %8434 = insertelement <2 x float> poison, float %7347, i64 0, !dbg !43 + %8435 = insertelement <2 x float> %8434, float %7348, i64 1, !dbg !43 + %8436 = fsub <2 x float> %8433, %8435, !dbg !43 + %8437 = insertelement <2 x float> poison, float %4341, i64 0, !dbg !43 + %8438 = insertelement <2 x float> %8437, float %4342, i64 1, !dbg !43 + %8439 = insertelement <2 x float> poison, float %7349, i64 0, !dbg !43 + %8440 = insertelement <2 x float> %8439, float %7350, i64 1, !dbg !43 + %8441 = fsub <2 x float> %8438, %8440, !dbg !43 + %8442 = insertelement <2 x float> poison, float %4343, i64 0, !dbg !43 + %8443 = insertelement <2 x float> %8442, float %4344, i64 1, !dbg !43 + %8444 = insertelement <2 x float> poison, float %7351, i64 0, !dbg !43 + %8445 = insertelement <2 x float> %8444, float %7352, i64 1, !dbg !43 + %8446 = fsub <2 x float> %8443, %8445, !dbg !43 + %8447 = insertelement <2 x float> poison, float %4345, i64 0, !dbg !43 + %8448 = insertelement <2 x float> %8447, float %4346, i64 1, !dbg !43 + %8449 = insertelement <2 x float> poison, float %7353, i64 0, !dbg !43 + %8450 = insertelement <2 x float> %8449, float %7354, i64 1, !dbg !43 + %8451 = fsub <2 x float> %8448, %8450, !dbg !43 + %8452 = insertelement <2 x float> poison, float %4347, i64 0, !dbg !43 + %8453 = insertelement <2 x float> %8452, float %4348, i64 1, !dbg !43 + %8454 = insertelement <2 x float> poison, float %7355, i64 0, !dbg !43 + %8455 = insertelement <2 x float> %8454, float %7356, i64 1, !dbg !43 + %8456 = fsub <2 x float> %8453, %8455, !dbg !43 + %8457 = insertelement <2 x float> poison, float %4349, i64 0, !dbg !43 + %8458 = insertelement <2 x float> %8457, float %4350, i64 1, !dbg !43 + %8459 = insertelement <2 x float> poison, float %7357, i64 0, !dbg !43 + %8460 = insertelement <2 x float> %8459, float %7358, i64 1, !dbg !43 + %8461 = fsub <2 x float> %8458, %8460, !dbg !43 + %8462 = insertelement <2 x float> poison, float %4351, i64 0, !dbg !43 + %8463 = insertelement <2 x float> %8462, float %4352, i64 1, !dbg !43 + %8464 = insertelement <2 x float> poison, float %7359, i64 0, !dbg !43 + %8465 = insertelement <2 x float> %8464, float %7360, i64 1, !dbg !43 + %8466 = fsub <2 x float> %8463, %8465, !dbg !43 + %8467 = insertelement <2 x float> poison, float %4353, i64 0, !dbg !43 + %8468 = insertelement <2 x float> %8467, float %4354, i64 1, !dbg !43 + %8469 = insertelement <2 x float> poison, float %7361, i64 0, !dbg !43 + %8470 = insertelement <2 x float> %8469, float %7362, i64 1, !dbg !43 + %8471 = fsub <2 x float> %8468, %8470, !dbg !43 + %8472 = insertelement <2 x float> poison, float %4355, i64 0, !dbg !43 + %8473 = insertelement <2 x float> %8472, float %4356, i64 1, !dbg !43 + %8474 = insertelement <2 x float> poison, float %7363, i64 0, !dbg !43 + %8475 = insertelement <2 x float> %8474, float %7364, i64 1, !dbg !43 + %8476 = fsub <2 x float> %8473, %8475, !dbg !43 + %8477 = insertelement <2 x float> poison, float %4357, i64 0, !dbg !43 + %8478 = insertelement <2 x float> %8477, float %4358, i64 1, !dbg !43 + %8479 = insertelement <2 x float> poison, float %7365, i64 0, !dbg !43 + %8480 = insertelement <2 x float> %8479, float %7366, i64 1, !dbg !43 + %8481 = fsub <2 x float> %8478, %8480, !dbg !43 + %8482 = insertelement <2 x float> poison, float %4359, i64 0, !dbg !43 + %8483 = insertelement <2 x float> %8482, float %4360, i64 1, !dbg !43 + %8484 = insertelement <2 x float> poison, float %7367, i64 0, !dbg !43 + %8485 = insertelement <2 x float> %8484, float %7368, i64 1, !dbg !43 + %8486 = fsub <2 x float> %8483, %8485, !dbg !43 + %8487 = insertelement <2 x float> poison, float %4361, i64 0, !dbg !43 + %8488 = insertelement <2 x float> %8487, float %4362, i64 1, !dbg !43 + %8489 = insertelement <2 x float> poison, float %7369, i64 0, !dbg !43 + %8490 = insertelement <2 x float> %8489, float %7370, i64 1, !dbg !43 + %8491 = fsub <2 x float> %8488, %8490, !dbg !43 + %8492 = insertelement <2 x float> poison, float %4363, i64 0, !dbg !43 + %8493 = insertelement <2 x float> %8492, float %4364, i64 1, !dbg !43 + %8494 = insertelement <2 x float> poison, float %7371, i64 0, !dbg !43 + %8495 = insertelement <2 x float> %8494, float %7372, i64 1, !dbg !43 + %8496 = fsub <2 x float> %8493, %8495, !dbg !43 + %8497 = insertelement <2 x float> poison, float %4365, i64 0, !dbg !43 + %8498 = insertelement <2 x float> %8497, float %4366, i64 1, !dbg !43 + %8499 = insertelement <2 x float> poison, float %7373, i64 0, !dbg !43 + %8500 = insertelement <2 x float> %8499, float %7374, i64 1, !dbg !43 + %8501 = fsub <2 x float> %8498, %8500, !dbg !43 + %8502 = insertelement <2 x float> poison, float %4367, i64 0, !dbg !43 + %8503 = insertelement <2 x float> %8502, float %4368, i64 1, !dbg !43 + %8504 = insertelement <2 x float> poison, float %7375, i64 0, !dbg !43 + %8505 = insertelement <2 x float> %8504, float %7376, i64 1, !dbg !43 + %8506 = fsub <2 x float> %8503, %8505, !dbg !43 + %8507 = insertelement <2 x float> poison, float %4369, i64 0, !dbg !43 + %8508 = insertelement <2 x float> %8507, float %4370, i64 1, !dbg !43 + %8509 = insertelement <2 x float> poison, float %7377, i64 0, !dbg !43 + %8510 = insertelement <2 x float> %8509, float %7378, i64 1, !dbg !43 + %8511 = fsub <2 x float> %8508, %8510, !dbg !43 + %8512 = insertelement <2 x float> poison, float %4371, i64 0, !dbg !43 + %8513 = insertelement <2 x float> %8512, float %4372, i64 1, !dbg !43 + %8514 = insertelement <2 x float> poison, float %7379, i64 0, !dbg !43 + %8515 = insertelement <2 x float> %8514, float %7380, i64 1, !dbg !43 + %8516 = fsub <2 x float> %8513, %8515, !dbg !43 + %8517 = insertelement <2 x float> poison, float %4373, i64 0, !dbg !43 + %8518 = insertelement <2 x float> %8517, float %4374, i64 1, !dbg !43 + %8519 = insertelement <2 x float> poison, float %7381, i64 0, !dbg !43 + %8520 = insertelement <2 x float> %8519, float %7382, i64 1, !dbg !43 + %8521 = fsub <2 x float> %8518, %8520, !dbg !43 + %8522 = insertelement <2 x float> poison, float %4375, i64 0, !dbg !43 + %8523 = insertelement <2 x float> %8522, float %4376, i64 1, !dbg !43 + %8524 = insertelement <2 x float> poison, float %7383, i64 0, !dbg !43 + %8525 = insertelement <2 x float> %8524, float %7384, i64 1, !dbg !43 + %8526 = fsub <2 x float> %8523, %8525, !dbg !43 + %8527 = insertelement <2 x float> poison, float %4377, i64 0, !dbg !43 + %8528 = insertelement <2 x float> %8527, float %4378, i64 1, !dbg !43 + %8529 = insertelement <2 x float> poison, float %7385, i64 0, !dbg !43 + %8530 = insertelement <2 x float> %8529, float %7386, i64 1, !dbg !43 + %8531 = fsub <2 x float> %8528, %8530, !dbg !43 + %8532 = insertelement <2 x float> poison, float %4379, i64 0, !dbg !43 + %8533 = insertelement <2 x float> %8532, float %4380, i64 1, !dbg !43 + %8534 = insertelement <2 x float> poison, float %7387, i64 0, !dbg !43 + %8535 = insertelement <2 x float> %8534, float %7388, i64 1, !dbg !43 + %8536 = fsub <2 x float> %8533, %8535, !dbg !43 + %8537 = insertelement <2 x float> poison, float %4381, i64 0, !dbg !43 + %8538 = insertelement <2 x float> %8537, float %4382, i64 1, !dbg !43 + %8539 = insertelement <2 x float> poison, float %7389, i64 0, !dbg !43 + %8540 = insertelement <2 x float> %8539, float %7390, i64 1, !dbg !43 + %8541 = fsub <2 x float> %8538, %8540, !dbg !43 + %8542 = insertelement <2 x float> poison, float %4383, i64 0, !dbg !43 + %8543 = insertelement <2 x float> %8542, float %4384, i64 1, !dbg !43 + %8544 = insertelement <2 x float> poison, float %7391, i64 0, !dbg !43 + %8545 = insertelement <2 x float> %8544, float %7392, i64 1, !dbg !43 + %8546 = fsub <2 x float> %8543, %8545, !dbg !43 + %8547 = insertelement <2 x float> poison, float %4385, i64 0, !dbg !43 + %8548 = insertelement <2 x float> %8547, float %4386, i64 1, !dbg !43 + %8549 = insertelement <2 x float> poison, float %7393, i64 0, !dbg !43 + %8550 = insertelement <2 x float> %8549, float %7394, i64 1, !dbg !43 + %8551 = fsub <2 x float> %8548, %8550, !dbg !43 + %8552 = insertelement <2 x float> poison, float %4387, i64 0, !dbg !43 + %8553 = insertelement <2 x float> %8552, float %4388, i64 1, !dbg !43 + %8554 = insertelement <2 x float> poison, float %7395, i64 0, !dbg !43 + %8555 = insertelement <2 x float> %8554, float %7396, i64 1, !dbg !43 + %8556 = fsub <2 x float> %8553, %8555, !dbg !43 + %8557 = insertelement <2 x float> poison, float %4389, i64 0, !dbg !43 + %8558 = insertelement <2 x float> %8557, float %4390, i64 1, !dbg !43 + %8559 = insertelement <2 x float> poison, float %7397, i64 0, !dbg !43 + %8560 = insertelement <2 x float> %8559, float %7398, i64 1, !dbg !43 + %8561 = fsub <2 x float> %8558, %8560, !dbg !43 + %8562 = insertelement <2 x float> poison, float %4391, i64 0, !dbg !43 + %8563 = insertelement <2 x float> %8562, float %4392, i64 1, !dbg !43 + %8564 = insertelement <2 x float> poison, float %7399, i64 0, !dbg !43 + %8565 = insertelement <2 x float> %8564, float %7400, i64 1, !dbg !43 + %8566 = fsub <2 x float> %8563, %8565, !dbg !43 + %8567 = insertelement <2 x float> poison, float %4393, i64 0, !dbg !43 + %8568 = insertelement <2 x float> %8567, float %4394, i64 1, !dbg !43 + %8569 = insertelement <2 x float> poison, float %7401, i64 0, !dbg !43 + %8570 = insertelement <2 x float> %8569, float %7402, i64 1, !dbg !43 + %8571 = fsub <2 x float> %8568, %8570, !dbg !43 + %8572 = insertelement <2 x float> poison, float %4395, i64 0, !dbg !43 + %8573 = insertelement <2 x float> %8572, float %4396, i64 1, !dbg !43 + %8574 = insertelement <2 x float> poison, float %7403, i64 0, !dbg !43 + %8575 = insertelement <2 x float> %8574, float %7404, i64 1, !dbg !43 + %8576 = fsub <2 x float> %8573, %8575, !dbg !43 + %8577 = insertelement <2 x float> poison, float %4397, i64 0, !dbg !43 + %8578 = insertelement <2 x float> %8577, float %4398, i64 1, !dbg !43 + %8579 = insertelement <2 x float> poison, float %7405, i64 0, !dbg !43 + %8580 = insertelement <2 x float> %8579, float %7406, i64 1, !dbg !43 + %8581 = fsub <2 x float> %8578, %8580, !dbg !43 + %8582 = insertelement <2 x float> poison, float %4399, i64 0, !dbg !43 + %8583 = insertelement <2 x float> %8582, float %4400, i64 1, !dbg !43 + %8584 = insertelement <2 x float> poison, float %7407, i64 0, !dbg !43 + %8585 = insertelement <2 x float> %8584, float %7408, i64 1, !dbg !43 + %8586 = fsub <2 x float> %8583, %8585, !dbg !43 + %8587 = insertelement <2 x float> poison, float %4401, i64 0, !dbg !43 + %8588 = insertelement <2 x float> %8587, float %4402, i64 1, !dbg !43 + %8589 = insertelement <2 x float> poison, float %7409, i64 0, !dbg !43 + %8590 = insertelement <2 x float> %8589, float %7410, i64 1, !dbg !43 + %8591 = fsub <2 x float> %8588, %8590, !dbg !43 + %8592 = insertelement <2 x float> poison, float %4403, i64 0, !dbg !43 + %8593 = insertelement <2 x float> %8592, float %4404, i64 1, !dbg !43 + %8594 = insertelement <2 x float> poison, float %7411, i64 0, !dbg !43 + %8595 = insertelement <2 x float> %8594, float %7412, i64 1, !dbg !43 + %8596 = fsub <2 x float> %8593, %8595, !dbg !43 + %8597 = insertelement <2 x float> poison, float %4405, i64 0, !dbg !43 + %8598 = insertelement <2 x float> %8597, float %4406, i64 1, !dbg !43 + %8599 = insertelement <2 x float> poison, float %7413, i64 0, !dbg !43 + %8600 = insertelement <2 x float> %8599, float %7414, i64 1, !dbg !43 + %8601 = fsub <2 x float> %8598, %8600, !dbg !43 + %8602 = insertelement <2 x float> poison, float %4407, i64 0, !dbg !43 + %8603 = insertelement <2 x float> %8602, float %4408, i64 1, !dbg !43 + %8604 = insertelement <2 x float> poison, float %7415, i64 0, !dbg !43 + %8605 = insertelement <2 x float> %8604, float %7416, i64 1, !dbg !43 + %8606 = fsub <2 x float> %8603, %8605, !dbg !43 + %8607 = insertelement <2 x float> poison, float %4409, i64 0, !dbg !43 + %8608 = insertelement <2 x float> %8607, float %4410, i64 1, !dbg !43 + %8609 = insertelement <2 x float> poison, float %7417, i64 0, !dbg !43 + %8610 = insertelement <2 x float> %8609, float %7418, i64 1, !dbg !43 + %8611 = fsub <2 x float> %8608, %8610, !dbg !43 + %8612 = insertelement <2 x float> poison, float %4411, i64 0, !dbg !43 + %8613 = insertelement <2 x float> %8612, float %4412, i64 1, !dbg !43 + %8614 = insertelement <2 x float> poison, float %7419, i64 0, !dbg !43 + %8615 = insertelement <2 x float> %8614, float %7420, i64 1, !dbg !43 + %8616 = fsub <2 x float> %8613, %8615, !dbg !43 + %8617 = insertelement <2 x float> poison, float %4413, i64 0, !dbg !43 + %8618 = insertelement <2 x float> %8617, float %4414, i64 1, !dbg !43 + %8619 = insertelement <2 x float> poison, float %7421, i64 0, !dbg !43 + %8620 = insertelement <2 x float> %8619, float %7422, i64 1, !dbg !43 + %8621 = fsub <2 x float> %8618, %8620, !dbg !43 + %8622 = insertelement <2 x float> poison, float %4415, i64 0, !dbg !43 + %8623 = insertelement <2 x float> %8622, float %4416, i64 1, !dbg !43 + %8624 = insertelement <2 x float> poison, float %7423, i64 0, !dbg !43 + %8625 = insertelement <2 x float> %8624, float %7424, i64 1, !dbg !43 + %8626 = fsub <2 x float> %8623, %8625, !dbg !43 + %8627 = insertelement <2 x float> poison, float %4417, i64 0, !dbg !43 + %8628 = insertelement <2 x float> %8627, float %4418, i64 1, !dbg !43 + %8629 = insertelement <2 x float> poison, float %7425, i64 0, !dbg !43 + %8630 = insertelement <2 x float> %8629, float %7426, i64 1, !dbg !43 + %8631 = fsub <2 x float> %8628, %8630, !dbg !43 + %8632 = insertelement <2 x float> poison, float %4419, i64 0, !dbg !43 + %8633 = insertelement <2 x float> %8632, float %4420, i64 1, !dbg !43 + %8634 = insertelement <2 x float> poison, float %7427, i64 0, !dbg !43 + %8635 = insertelement <2 x float> %8634, float %7428, i64 1, !dbg !43 + %8636 = fsub <2 x float> %8633, %8635, !dbg !43 + %8637 = insertelement <2 x float> poison, float %4421, i64 0, !dbg !43 + %8638 = insertelement <2 x float> %8637, float %4422, i64 1, !dbg !43 + %8639 = insertelement <2 x float> poison, float %7429, i64 0, !dbg !43 + %8640 = insertelement <2 x float> %8639, float %7430, i64 1, !dbg !43 + %8641 = fsub <2 x float> %8638, %8640, !dbg !43 + %8642 = insertelement <2 x float> poison, float %4423, i64 0, !dbg !43 + %8643 = insertelement <2 x float> %8642, float %4424, i64 1, !dbg !43 + %8644 = insertelement <2 x float> poison, float %7431, i64 0, !dbg !43 + %8645 = insertelement <2 x float> %8644, float %7432, i64 1, !dbg !43 + %8646 = fsub <2 x float> %8643, %8645, !dbg !43 + %8647 = insertelement <2 x float> poison, float %4425, i64 0, !dbg !43 + %8648 = insertelement <2 x float> %8647, float %4426, i64 1, !dbg !43 + %8649 = insertelement <2 x float> poison, float %7433, i64 0, !dbg !43 + %8650 = insertelement <2 x float> %8649, float %7434, i64 1, !dbg !43 + %8651 = fsub <2 x float> %8648, %8650, !dbg !43 + %8652 = insertelement <2 x float> poison, float %4427, i64 0, !dbg !43 + %8653 = insertelement <2 x float> %8652, float %4428, i64 1, !dbg !43 + %8654 = insertelement <2 x float> poison, float %7435, i64 0, !dbg !43 + %8655 = insertelement <2 x float> %8654, float %7436, i64 1, !dbg !43 + %8656 = fsub <2 x float> %8653, %8655, !dbg !43 + %8657 = insertelement <2 x float> poison, float %4429, i64 0, !dbg !43 + %8658 = insertelement <2 x float> %8657, float %4430, i64 1, !dbg !43 + %8659 = insertelement <2 x float> poison, float %7437, i64 0, !dbg !43 + %8660 = insertelement <2 x float> %8659, float %7438, i64 1, !dbg !43 + %8661 = fsub <2 x float> %8658, %8660, !dbg !43 + %8662 = insertelement <2 x float> poison, float %4431, i64 0, !dbg !43 + %8663 = insertelement <2 x float> %8662, float %4432, i64 1, !dbg !43 + %8664 = insertelement <2 x float> poison, float %7439, i64 0, !dbg !43 + %8665 = insertelement <2 x float> %8664, float %7440, i64 1, !dbg !43 + %8666 = fsub <2 x float> %8663, %8665, !dbg !43 + %8667 = insertelement <2 x float> poison, float %4433, i64 0, !dbg !43 + %8668 = insertelement <2 x float> %8667, float %4434, i64 1, !dbg !43 + %8669 = insertelement <2 x float> poison, float %7441, i64 0, !dbg !43 + %8670 = insertelement <2 x float> %8669, float %7442, i64 1, !dbg !43 + %8671 = fsub <2 x float> %8668, %8670, !dbg !43 + %8672 = insertelement <2 x float> poison, float %4435, i64 0, !dbg !43 + %8673 = insertelement <2 x float> %8672, float %4436, i64 1, !dbg !43 + %8674 = insertelement <2 x float> poison, float %7443, i64 0, !dbg !43 + %8675 = insertelement <2 x float> %8674, float %7444, i64 1, !dbg !43 + %8676 = fsub <2 x float> %8673, %8675, !dbg !43 + %8677 = insertelement <2 x float> poison, float %4437, i64 0, !dbg !43 + %8678 = insertelement <2 x float> %8677, float %4438, i64 1, !dbg !43 + %8679 = insertelement <2 x float> poison, float %7445, i64 0, !dbg !43 + %8680 = insertelement <2 x float> %8679, float %7446, i64 1, !dbg !43 + %8681 = fsub <2 x float> %8678, %8680, !dbg !43 + %8682 = insertelement <2 x float> poison, float %4439, i64 0, !dbg !43 + %8683 = insertelement <2 x float> %8682, float %4440, i64 1, !dbg !43 + %8684 = insertelement <2 x float> poison, float %7447, i64 0, !dbg !43 + %8685 = insertelement <2 x float> %8684, float %7448, i64 1, !dbg !43 + %8686 = fsub <2 x float> %8683, %8685, !dbg !43 + %8687 = insertelement <2 x float> poison, float %4441, i64 0, !dbg !43 + %8688 = insertelement <2 x float> %8687, float %4442, i64 1, !dbg !43 + %8689 = insertelement <2 x float> poison, float %7449, i64 0, !dbg !43 + %8690 = insertelement <2 x float> %8689, float %7450, i64 1, !dbg !43 + %8691 = fsub <2 x float> %8688, %8690, !dbg !43 + %8692 = insertelement <2 x float> poison, float %4443, i64 0, !dbg !43 + %8693 = insertelement <2 x float> %8692, float %4444, i64 1, !dbg !43 + %8694 = insertelement <2 x float> poison, float %7451, i64 0, !dbg !43 + %8695 = insertelement <2 x float> %8694, float %7452, i64 1, !dbg !43 + %8696 = fsub <2 x float> %8693, %8695, !dbg !43 + %8697 = insertelement <2 x float> poison, float %4445, i64 0, !dbg !43 + %8698 = insertelement <2 x float> %8697, float %4446, i64 1, !dbg !43 + %8699 = insertelement <2 x float> poison, float %7453, i64 0, !dbg !43 + %8700 = insertelement <2 x float> %8699, float %7454, i64 1, !dbg !43 + %8701 = fsub <2 x float> %8698, %8700, !dbg !43 + %8702 = insertelement <2 x float> poison, float %4447, i64 0, !dbg !43 + %8703 = insertelement <2 x float> %8702, float %4448, i64 1, !dbg !43 + %8704 = insertelement <2 x float> poison, float %7455, i64 0, !dbg !43 + %8705 = insertelement <2 x float> %8704, float %7456, i64 1, !dbg !43 + %8706 = fsub <2 x float> %8703, %8705, !dbg !43 + %8707 = insertelement <2 x float> poison, float %4449, i64 0, !dbg !43 + %8708 = insertelement <2 x float> %8707, float %4450, i64 1, !dbg !43 + %8709 = insertelement <2 x float> poison, float %7457, i64 0, !dbg !43 + %8710 = insertelement <2 x float> %8709, float %7458, i64 1, !dbg !43 + %8711 = fsub <2 x float> %8708, %8710, !dbg !43 + %8712 = insertelement <2 x float> poison, float %4451, i64 0, !dbg !43 + %8713 = insertelement <2 x float> %8712, float %4452, i64 1, !dbg !43 + %8714 = insertelement <2 x float> poison, float %7459, i64 0, !dbg !43 + %8715 = insertelement <2 x float> %8714, float %7460, i64 1, !dbg !43 + %8716 = fsub <2 x float> %8713, %8715, !dbg !43 + %8717 = insertelement <2 x float> poison, float %4453, i64 0, !dbg !43 + %8718 = insertelement <2 x float> %8717, float %4454, i64 1, !dbg !43 + %8719 = insertelement <2 x float> poison, float %7461, i64 0, !dbg !43 + %8720 = insertelement <2 x float> %8719, float %7462, i64 1, !dbg !43 + %8721 = fsub <2 x float> %8718, %8720, !dbg !43 + %8722 = insertelement <2 x float> poison, float %4455, i64 0, !dbg !43 + %8723 = insertelement <2 x float> %8722, float %4456, i64 1, !dbg !43 + %8724 = insertelement <2 x float> poison, float %7463, i64 0, !dbg !43 + %8725 = insertelement <2 x float> %8724, float %7464, i64 1, !dbg !43 + %8726 = fsub <2 x float> %8723, %8725, !dbg !43 + %8727 = insertelement <2 x float> poison, float %4457, i64 0, !dbg !43 + %8728 = insertelement <2 x float> %8727, float %4458, i64 1, !dbg !43 + %8729 = insertelement <2 x float> poison, float %7465, i64 0, !dbg !43 + %8730 = insertelement <2 x float> %8729, float %7466, i64 1, !dbg !43 + %8731 = fsub <2 x float> %8728, %8730, !dbg !43 + %8732 = insertelement <2 x float> poison, float %4459, i64 0, !dbg !43 + %8733 = insertelement <2 x float> %8732, float %4460, i64 1, !dbg !43 + %8734 = insertelement <2 x float> poison, float %7467, i64 0, !dbg !43 + %8735 = insertelement <2 x float> %8734, float %7468, i64 1, !dbg !43 + %8736 = fsub <2 x float> %8733, %8735, !dbg !43 + %8737 = fptrunc float %7469 to bfloat, !dbg !44 + %8738 = fptrunc float %7470 to bfloat, !dbg !44 + %8739 = fptrunc float %7471 to bfloat, !dbg !44 + %8740 = fptrunc float %7472 to bfloat, !dbg !44 + %8741 = fptrunc float %7473 to bfloat, !dbg !44 + %8742 = fptrunc float %7474 to bfloat, !dbg !44 + %8743 = fptrunc float %7475 to bfloat, !dbg !44 + %8744 = fptrunc float %7476 to bfloat, !dbg !44 + %8745 = fptrunc <2 x float> %7481 to <2 x bfloat>, !dbg !44 + %8746 = fptrunc <2 x float> %7486 to <2 x bfloat>, !dbg !44 + %8747 = fptrunc <2 x float> %7491 to <2 x bfloat>, !dbg !44 + %8748 = fptrunc <2 x float> %7496 to <2 x bfloat>, !dbg !44 + %8749 = fptrunc <2 x float> %7501 to <2 x bfloat>, !dbg !44 + %8750 = fptrunc <2 x float> %7506 to <2 x bfloat>, !dbg !44 + %8751 = fptrunc <2 x float> %7511 to <2 x bfloat>, !dbg !44 + %8752 = fptrunc <2 x float> %7516 to <2 x bfloat>, !dbg !44 + %8753 = fptrunc <2 x float> %7521 to <2 x bfloat>, !dbg !44 + %8754 = fptrunc <2 x float> %7526 to <2 x bfloat>, !dbg !44 + %8755 = fptrunc <2 x float> %7531 to <2 x bfloat>, !dbg !44 + %8756 = fptrunc <2 x float> %7536 to <2 x bfloat>, !dbg !44 + %8757 = fptrunc <2 x float> %7541 to <2 x bfloat>, !dbg !44 + %8758 = fptrunc <2 x float> %7546 to <2 x bfloat>, !dbg !44 + %8759 = fptrunc <2 x float> %7551 to <2 x bfloat>, !dbg !44 + %8760 = fptrunc <2 x float> %7556 to <2 x bfloat>, !dbg !44 + %8761 = fptrunc <2 x float> %7561 to <2 x bfloat>, !dbg !44 + %8762 = fptrunc <2 x float> %7566 to <2 x bfloat>, !dbg !44 + %8763 = fptrunc <2 x float> %7571 to <2 x bfloat>, !dbg !44 + %8764 = fptrunc <2 x float> %7576 to <2 x bfloat>, !dbg !44 + %8765 = fptrunc <2 x float> %7581 to <2 x bfloat>, !dbg !44 + %8766 = fptrunc <2 x float> %7586 to <2 x bfloat>, !dbg !44 + %8767 = fptrunc <2 x float> %7591 to <2 x bfloat>, !dbg !44 + %8768 = fptrunc <2 x float> %7596 to <2 x bfloat>, !dbg !44 + %8769 = fptrunc <2 x float> %7601 to <2 x bfloat>, !dbg !44 + %8770 = fptrunc <2 x float> %7606 to <2 x bfloat>, !dbg !44 + %8771 = fptrunc <2 x float> %7611 to <2 x bfloat>, !dbg !44 + %8772 = fptrunc <2 x float> %7616 to <2 x bfloat>, !dbg !44 + %8773 = fptrunc <2 x float> %7621 to <2 x bfloat>, !dbg !44 + %8774 = fptrunc <2 x float> %7626 to <2 x bfloat>, !dbg !44 + %8775 = fptrunc <2 x float> %7631 to <2 x bfloat>, !dbg !44 + %8776 = fptrunc <2 x float> %7636 to <2 x bfloat>, !dbg !44 + %8777 = fptrunc <2 x float> %7641 to <2 x bfloat>, !dbg !44 + %8778 = fptrunc <2 x float> %7646 to <2 x bfloat>, !dbg !44 + %8779 = fptrunc <2 x float> %7651 to <2 x bfloat>, !dbg !44 + %8780 = fptrunc <2 x float> %7656 to <2 x bfloat>, !dbg !44 + %8781 = fptrunc <2 x float> %7661 to <2 x bfloat>, !dbg !44 + %8782 = fptrunc <2 x float> %7666 to <2 x bfloat>, !dbg !44 + %8783 = fptrunc <2 x float> %7671 to <2 x bfloat>, !dbg !44 + %8784 = fptrunc <2 x float> %7676 to <2 x bfloat>, !dbg !44 + %8785 = fptrunc <2 x float> %7681 to <2 x bfloat>, !dbg !44 + %8786 = fptrunc <2 x float> %7686 to <2 x bfloat>, !dbg !44 + %8787 = fptrunc <2 x float> %7691 to <2 x bfloat>, !dbg !44 + %8788 = fptrunc <2 x float> %7696 to <2 x bfloat>, !dbg !44 + %8789 = fptrunc <2 x float> %7701 to <2 x bfloat>, !dbg !44 + %8790 = fptrunc <2 x float> %7706 to <2 x bfloat>, !dbg !44 + %8791 = fptrunc <2 x float> %7711 to <2 x bfloat>, !dbg !44 + %8792 = fptrunc <2 x float> %7716 to <2 x bfloat>, !dbg !44 + %8793 = fptrunc <2 x float> %7721 to <2 x bfloat>, !dbg !44 + %8794 = fptrunc <2 x float> %7726 to <2 x bfloat>, !dbg !44 + %8795 = fptrunc <2 x float> %7731 to <2 x bfloat>, !dbg !44 + %8796 = fptrunc <2 x float> %7736 to <2 x bfloat>, !dbg !44 + %8797 = fptrunc <2 x float> %7741 to <2 x bfloat>, !dbg !44 + %8798 = fptrunc <2 x float> %7746 to <2 x bfloat>, !dbg !44 + %8799 = fptrunc <2 x float> %7751 to <2 x bfloat>, !dbg !44 + %8800 = fptrunc <2 x float> %7756 to <2 x bfloat>, !dbg !44 + %8801 = fptrunc <2 x float> %7761 to <2 x bfloat>, !dbg !44 + %8802 = fptrunc <2 x float> %7766 to <2 x bfloat>, !dbg !44 + %8803 = fptrunc <2 x float> %7771 to <2 x bfloat>, !dbg !44 + %8804 = fptrunc <2 x float> %7776 to <2 x bfloat>, !dbg !44 + %8805 = fptrunc <2 x float> %7781 to <2 x bfloat>, !dbg !44 + %8806 = fptrunc <2 x float> %7786 to <2 x bfloat>, !dbg !44 + %8807 = fptrunc <2 x float> %7791 to <2 x bfloat>, !dbg !44 + %8808 = fptrunc <2 x float> %7796 to <2 x bfloat>, !dbg !44 + %8809 = fptrunc <2 x float> %7801 to <2 x bfloat>, !dbg !44 + %8810 = fptrunc <2 x float> %7806 to <2 x bfloat>, !dbg !44 + %8811 = fptrunc <2 x float> %7811 to <2 x bfloat>, !dbg !44 + %8812 = fptrunc <2 x float> %7816 to <2 x bfloat>, !dbg !44 + %8813 = fptrunc <2 x float> %7821 to <2 x bfloat>, !dbg !44 + %8814 = fptrunc <2 x float> %7826 to <2 x bfloat>, !dbg !44 + %8815 = fptrunc <2 x float> %7831 to <2 x bfloat>, !dbg !44 + %8816 = fptrunc <2 x float> %7836 to <2 x bfloat>, !dbg !44 + %8817 = fptrunc <2 x float> %7841 to <2 x bfloat>, !dbg !44 + %8818 = fptrunc <2 x float> %7846 to <2 x bfloat>, !dbg !44 + %8819 = fptrunc <2 x float> %7851 to <2 x bfloat>, !dbg !44 + %8820 = fptrunc <2 x float> %7856 to <2 x bfloat>, !dbg !44 + %8821 = fptrunc <2 x float> %7861 to <2 x bfloat>, !dbg !44 + %8822 = fptrunc <2 x float> %7866 to <2 x bfloat>, !dbg !44 + %8823 = fptrunc <2 x float> %7871 to <2 x bfloat>, !dbg !44 + %8824 = fptrunc <2 x float> %7876 to <2 x bfloat>, !dbg !44 + %8825 = fptrunc <2 x float> %7881 to <2 x bfloat>, !dbg !44 + %8826 = fptrunc <2 x float> %7886 to <2 x bfloat>, !dbg !44 + %8827 = fptrunc <2 x float> %7891 to <2 x bfloat>, !dbg !44 + %8828 = fptrunc <2 x float> %7896 to <2 x bfloat>, !dbg !44 + %8829 = fptrunc <2 x float> %7901 to <2 x bfloat>, !dbg !44 + %8830 = fptrunc <2 x float> %7906 to <2 x bfloat>, !dbg !44 + %8831 = fptrunc <2 x float> %7911 to <2 x bfloat>, !dbg !44 + %8832 = fptrunc <2 x float> %7916 to <2 x bfloat>, !dbg !44 + %8833 = fptrunc <2 x float> %7921 to <2 x bfloat>, !dbg !44 + %8834 = fptrunc <2 x float> %7926 to <2 x bfloat>, !dbg !44 + %8835 = fptrunc <2 x float> %7931 to <2 x bfloat>, !dbg !44 + %8836 = fptrunc <2 x float> %7936 to <2 x bfloat>, !dbg !44 + %8837 = fptrunc <2 x float> %7941 to <2 x bfloat>, !dbg !44 + %8838 = fptrunc <2 x float> %7946 to <2 x bfloat>, !dbg !44 + %8839 = fptrunc <2 x float> %7951 to <2 x bfloat>, !dbg !44 + %8840 = fptrunc <2 x float> %7956 to <2 x bfloat>, !dbg !44 + %8841 = fptrunc <2 x float> %7961 to <2 x bfloat>, !dbg !44 + %8842 = fptrunc <2 x float> %7966 to <2 x bfloat>, !dbg !44 + %8843 = fptrunc <2 x float> %7971 to <2 x bfloat>, !dbg !44 + %8844 = fptrunc <2 x float> %7976 to <2 x bfloat>, !dbg !44 + %8845 = fptrunc <2 x float> %7981 to <2 x bfloat>, !dbg !44 + %8846 = fptrunc <2 x float> %7986 to <2 x bfloat>, !dbg !44 + %8847 = fptrunc <2 x float> %7991 to <2 x bfloat>, !dbg !44 + %8848 = fptrunc <2 x float> %7996 to <2 x bfloat>, !dbg !44 + %8849 = fptrunc <2 x float> %8001 to <2 x bfloat>, !dbg !44 + %8850 = fptrunc <2 x float> %8006 to <2 x bfloat>, !dbg !44 + %8851 = fptrunc <2 x float> %8011 to <2 x bfloat>, !dbg !44 + %8852 = fptrunc <2 x float> %8016 to <2 x bfloat>, !dbg !44 + %8853 = fptrunc <2 x float> %8021 to <2 x bfloat>, !dbg !44 + %8854 = fptrunc <2 x float> %8026 to <2 x bfloat>, !dbg !44 + %8855 = fptrunc <2 x float> %8031 to <2 x bfloat>, !dbg !44 + %8856 = fptrunc <2 x float> %8036 to <2 x bfloat>, !dbg !44 + %8857 = fptrunc <2 x float> %8041 to <2 x bfloat>, !dbg !44 + %8858 = fptrunc <2 x float> %8046 to <2 x bfloat>, !dbg !44 + %8859 = fptrunc <2 x float> %8051 to <2 x bfloat>, !dbg !44 + %8860 = fptrunc <2 x float> %8056 to <2 x bfloat>, !dbg !44 + %8861 = fptrunc <2 x float> %8061 to <2 x bfloat>, !dbg !44 + %8862 = fptrunc <2 x float> %8066 to <2 x bfloat>, !dbg !44 + %8863 = fptrunc <2 x float> %8071 to <2 x bfloat>, !dbg !44 + %8864 = fptrunc <2 x float> %8076 to <2 x bfloat>, !dbg !44 + %8865 = fptrunc <2 x float> %8081 to <2 x bfloat>, !dbg !44 + %8866 = fptrunc <2 x float> %8086 to <2 x bfloat>, !dbg !44 + %8867 = fptrunc <2 x float> %8091 to <2 x bfloat>, !dbg !44 + %8868 = fptrunc <2 x float> %8096 to <2 x bfloat>, !dbg !44 + %8869 = fptrunc <2 x float> %8101 to <2 x bfloat>, !dbg !44 + %8870 = fptrunc <2 x float> %8106 to <2 x bfloat>, !dbg !44 + %8871 = fptrunc <2 x float> %8111 to <2 x bfloat>, !dbg !44 + %8872 = fptrunc <2 x float> %8116 to <2 x bfloat>, !dbg !44 + %8873 = fptrunc <2 x float> %8121 to <2 x bfloat>, !dbg !44 + %8874 = fptrunc <2 x float> %8126 to <2 x bfloat>, !dbg !44 + %8875 = fptrunc <2 x float> %8131 to <2 x bfloat>, !dbg !44 + %8876 = fptrunc <2 x float> %8136 to <2 x bfloat>, !dbg !44 + %8877 = fptrunc <2 x float> %8141 to <2 x bfloat>, !dbg !44 + %8878 = fptrunc <2 x float> %8146 to <2 x bfloat>, !dbg !44 + %8879 = fptrunc <2 x float> %8151 to <2 x bfloat>, !dbg !44 + %8880 = fptrunc <2 x float> %8156 to <2 x bfloat>, !dbg !44 + %8881 = fptrunc <2 x float> %8161 to <2 x bfloat>, !dbg !44 + %8882 = fptrunc <2 x float> %8166 to <2 x bfloat>, !dbg !44 + %8883 = fptrunc <2 x float> %8171 to <2 x bfloat>, !dbg !44 + %8884 = fptrunc <2 x float> %8176 to <2 x bfloat>, !dbg !44 + %8885 = fptrunc <2 x float> %8181 to <2 x bfloat>, !dbg !44 + %8886 = fptrunc <2 x float> %8186 to <2 x bfloat>, !dbg !44 + %8887 = fptrunc <2 x float> %8191 to <2 x bfloat>, !dbg !44 + %8888 = fptrunc <2 x float> %8196 to <2 x bfloat>, !dbg !44 + %8889 = fptrunc <2 x float> %8201 to <2 x bfloat>, !dbg !44 + %8890 = fptrunc <2 x float> %8206 to <2 x bfloat>, !dbg !44 + %8891 = fptrunc <2 x float> %8211 to <2 x bfloat>, !dbg !44 + %8892 = fptrunc <2 x float> %8216 to <2 x bfloat>, !dbg !44 + %8893 = fptrunc <2 x float> %8221 to <2 x bfloat>, !dbg !44 + %8894 = fptrunc <2 x float> %8226 to <2 x bfloat>, !dbg !44 + %8895 = fptrunc <2 x float> %8231 to <2 x bfloat>, !dbg !44 + %8896 = fptrunc <2 x float> %8236 to <2 x bfloat>, !dbg !44 + %8897 = fptrunc <2 x float> %8241 to <2 x bfloat>, !dbg !44 + %8898 = fptrunc <2 x float> %8246 to <2 x bfloat>, !dbg !44 + %8899 = fptrunc <2 x float> %8251 to <2 x bfloat>, !dbg !44 + %8900 = fptrunc <2 x float> %8256 to <2 x bfloat>, !dbg !44 + %8901 = fptrunc <2 x float> %8261 to <2 x bfloat>, !dbg !44 + %8902 = fptrunc <2 x float> %8266 to <2 x bfloat>, !dbg !44 + %8903 = fptrunc <2 x float> %8271 to <2 x bfloat>, !dbg !44 + %8904 = fptrunc <2 x float> %8276 to <2 x bfloat>, !dbg !44 + %8905 = fptrunc <2 x float> %8281 to <2 x bfloat>, !dbg !44 + %8906 = fptrunc <2 x float> %8286 to <2 x bfloat>, !dbg !44 + %8907 = fptrunc <2 x float> %8291 to <2 x bfloat>, !dbg !44 + %8908 = fptrunc <2 x float> %8296 to <2 x bfloat>, !dbg !44 + %8909 = fptrunc <2 x float> %8301 to <2 x bfloat>, !dbg !44 + %8910 = fptrunc <2 x float> %8306 to <2 x bfloat>, !dbg !44 + %8911 = fptrunc <2 x float> %8311 to <2 x bfloat>, !dbg !44 + %8912 = fptrunc <2 x float> %8316 to <2 x bfloat>, !dbg !44 + %8913 = fptrunc <2 x float> %8321 to <2 x bfloat>, !dbg !44 + %8914 = fptrunc <2 x float> %8326 to <2 x bfloat>, !dbg !44 + %8915 = fptrunc <2 x float> %8331 to <2 x bfloat>, !dbg !44 + %8916 = fptrunc <2 x float> %8336 to <2 x bfloat>, !dbg !44 + %8917 = fptrunc <2 x float> %8341 to <2 x bfloat>, !dbg !44 + %8918 = fptrunc <2 x float> %8346 to <2 x bfloat>, !dbg !44 + %8919 = fptrunc <2 x float> %8351 to <2 x bfloat>, !dbg !44 + %8920 = fptrunc <2 x float> %8356 to <2 x bfloat>, !dbg !44 + %8921 = fptrunc <2 x float> %8361 to <2 x bfloat>, !dbg !44 + %8922 = fptrunc <2 x float> %8366 to <2 x bfloat>, !dbg !44 + %8923 = fptrunc <2 x float> %8371 to <2 x bfloat>, !dbg !44 + %8924 = fptrunc <2 x float> %8376 to <2 x bfloat>, !dbg !44 + %8925 = fptrunc <2 x float> %8381 to <2 x bfloat>, !dbg !44 + %8926 = fptrunc <2 x float> %8386 to <2 x bfloat>, !dbg !44 + %8927 = fptrunc <2 x float> %8391 to <2 x bfloat>, !dbg !44 + %8928 = fptrunc <2 x float> %8396 to <2 x bfloat>, !dbg !44 + %8929 = fptrunc <2 x float> %8401 to <2 x bfloat>, !dbg !44 + %8930 = fptrunc <2 x float> %8406 to <2 x bfloat>, !dbg !44 + %8931 = fptrunc <2 x float> %8411 to <2 x bfloat>, !dbg !44 + %8932 = fptrunc <2 x float> %8416 to <2 x bfloat>, !dbg !44 + %8933 = fptrunc <2 x float> %8421 to <2 x bfloat>, !dbg !44 + %8934 = fptrunc <2 x float> %8426 to <2 x bfloat>, !dbg !44 + %8935 = fptrunc <2 x float> %8431 to <2 x bfloat>, !dbg !44 + %8936 = fptrunc <2 x float> %8436 to <2 x bfloat>, !dbg !44 + %8937 = fptrunc <2 x float> %8441 to <2 x bfloat>, !dbg !44 + %8938 = fptrunc <2 x float> %8446 to <2 x bfloat>, !dbg !44 + %8939 = fptrunc <2 x float> %8451 to <2 x bfloat>, !dbg !44 + %8940 = fptrunc <2 x float> %8456 to <2 x bfloat>, !dbg !44 + %8941 = fptrunc <2 x float> %8461 to <2 x bfloat>, !dbg !44 + %8942 = fptrunc <2 x float> %8466 to <2 x bfloat>, !dbg !44 + %8943 = fptrunc <2 x float> %8471 to <2 x bfloat>, !dbg !44 + %8944 = fptrunc <2 x float> %8476 to <2 x bfloat>, !dbg !44 + %8945 = fptrunc <2 x float> %8481 to <2 x bfloat>, !dbg !44 + %8946 = fptrunc <2 x float> %8486 to <2 x bfloat>, !dbg !44 + %8947 = fptrunc <2 x float> %8491 to <2 x bfloat>, !dbg !44 + %8948 = fptrunc <2 x float> %8496 to <2 x bfloat>, !dbg !44 + %8949 = fptrunc <2 x float> %8501 to <2 x bfloat>, !dbg !44 + %8950 = fptrunc <2 x float> %8506 to <2 x bfloat>, !dbg !44 + %8951 = fptrunc <2 x float> %8511 to <2 x bfloat>, !dbg !44 + %8952 = fptrunc <2 x float> %8516 to <2 x bfloat>, !dbg !44 + %8953 = fptrunc <2 x float> %8521 to <2 x bfloat>, !dbg !44 + %8954 = fptrunc <2 x float> %8526 to <2 x bfloat>, !dbg !44 + %8955 = fptrunc <2 x float> %8531 to <2 x bfloat>, !dbg !44 + %8956 = fptrunc <2 x float> %8536 to <2 x bfloat>, !dbg !44 + %8957 = fptrunc <2 x float> %8541 to <2 x bfloat>, !dbg !44 + %8958 = fptrunc <2 x float> %8546 to <2 x bfloat>, !dbg !44 + %8959 = fptrunc <2 x float> %8551 to <2 x bfloat>, !dbg !44 + %8960 = fptrunc <2 x float> %8556 to <2 x bfloat>, !dbg !44 + %8961 = fptrunc <2 x float> %8561 to <2 x bfloat>, !dbg !44 + %8962 = fptrunc <2 x float> %8566 to <2 x bfloat>, !dbg !44 + %8963 = fptrunc <2 x float> %8571 to <2 x bfloat>, !dbg !44 + %8964 = fptrunc <2 x float> %8576 to <2 x bfloat>, !dbg !44 + %8965 = fptrunc <2 x float> %8581 to <2 x bfloat>, !dbg !44 + %8966 = fptrunc <2 x float> %8586 to <2 x bfloat>, !dbg !44 + %8967 = fptrunc <2 x float> %8591 to <2 x bfloat>, !dbg !44 + %8968 = fptrunc <2 x float> %8596 to <2 x bfloat>, !dbg !44 + %8969 = fptrunc <2 x float> %8601 to <2 x bfloat>, !dbg !44 + %8970 = fptrunc <2 x float> %8606 to <2 x bfloat>, !dbg !44 + %8971 = fptrunc <2 x float> %8611 to <2 x bfloat>, !dbg !44 + %8972 = fptrunc <2 x float> %8616 to <2 x bfloat>, !dbg !44 + %8973 = fptrunc <2 x float> %8621 to <2 x bfloat>, !dbg !44 + %8974 = fptrunc <2 x float> %8626 to <2 x bfloat>, !dbg !44 + %8975 = fptrunc <2 x float> %8631 to <2 x bfloat>, !dbg !44 + %8976 = fptrunc <2 x float> %8636 to <2 x bfloat>, !dbg !44 + %8977 = fptrunc <2 x float> %8641 to <2 x bfloat>, !dbg !44 + %8978 = fptrunc <2 x float> %8646 to <2 x bfloat>, !dbg !44 + %8979 = fptrunc <2 x float> %8651 to <2 x bfloat>, !dbg !44 + %8980 = fptrunc <2 x float> %8656 to <2 x bfloat>, !dbg !44 + %8981 = fptrunc <2 x float> %8661 to <2 x bfloat>, !dbg !44 + %8982 = fptrunc <2 x float> %8666 to <2 x bfloat>, !dbg !44 + %8983 = fptrunc <2 x float> %8671 to <2 x bfloat>, !dbg !44 + %8984 = fptrunc <2 x float> %8676 to <2 x bfloat>, !dbg !44 + %8985 = fptrunc <2 x float> %8681 to <2 x bfloat>, !dbg !44 + %8986 = fptrunc <2 x float> %8686 to <2 x bfloat>, !dbg !44 + %8987 = fptrunc <2 x float> %8691 to <2 x bfloat>, !dbg !44 + %8988 = fptrunc <2 x float> %8696 to <2 x bfloat>, !dbg !44 + %8989 = fptrunc <2 x float> %8701 to <2 x bfloat>, !dbg !44 + %8990 = fptrunc <2 x float> %8706 to <2 x bfloat>, !dbg !44 + %8991 = fptrunc <2 x float> %8711 to <2 x bfloat>, !dbg !44 + %8992 = fptrunc <2 x float> %8716 to <2 x bfloat>, !dbg !44 + %8993 = fptrunc <2 x float> %8721 to <2 x bfloat>, !dbg !44 + %8994 = fptrunc <2 x float> %8726 to <2 x bfloat>, !dbg !44 + %8995 = fptrunc <2 x float> %8731 to <2 x bfloat>, !dbg !44 + %8996 = fptrunc <2 x float> %8736 to <2 x bfloat>, !dbg !44 + %8997 = insertelement <2 x bfloat> poison, bfloat %8737, i64 0, !dbg !45 + %8998 = insertelement <2 x bfloat> %8997, bfloat %8738, i64 1, !dbg !45 + %8999 = bitcast <2 x bfloat> %8998 to i32, !dbg !45 + %9000 = insertelement <2 x bfloat> poison, bfloat %8739, i64 0, !dbg !45 + %9001 = insertelement <2 x bfloat> %9000, bfloat %8740, i64 1, !dbg !45 + %9002 = bitcast <2 x bfloat> %9001 to i32, !dbg !45 + %9003 = insertelement <2 x bfloat> poison, bfloat %8741, i64 0, !dbg !45 + %9004 = insertelement <2 x bfloat> %9003, bfloat %8742, i64 1, !dbg !45 + %9005 = bitcast <2 x bfloat> %9004 to i32, !dbg !45 + %9006 = insertelement <2 x bfloat> poison, bfloat %8743, i64 0, !dbg !45 + %9007 = insertelement <2 x bfloat> %9006, bfloat %8744, i64 1, !dbg !45 + %9008 = bitcast <2 x bfloat> %9007 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %8999, i32 %9002, i32 %9005, i32 %9008, ptr addrspace(1) %518, i1 %27) #5, !dbg !45 + %9009 = bitcast <2 x bfloat> %8745 to i32, !dbg !45 + %9010 = bitcast <2 x bfloat> %8746 to i32, !dbg !45 + %9011 = bitcast <2 x bfloat> %8747 to i32, !dbg !45 + %9012 = bitcast <2 x bfloat> %8748 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9009, i32 %9010, i32 %9011, i32 %9012, ptr addrspace(1) %519, i1 %28) #5, !dbg !45 + %9013 = bitcast <2 x bfloat> %8749 to i32, !dbg !45 + %9014 = bitcast <2 x bfloat> %8750 to i32, !dbg !45 + %9015 = bitcast <2 x bfloat> %8751 to i32, !dbg !45 + %9016 = bitcast <2 x bfloat> %8752 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9013, i32 %9014, i32 %9015, i32 %9016, ptr addrspace(1) %520, i1 %29) #5, !dbg !45 + %9017 = bitcast <2 x bfloat> %8753 to i32, !dbg !45 + %9018 = bitcast <2 x bfloat> %8754 to i32, !dbg !45 + %9019 = bitcast <2 x bfloat> %8755 to i32, !dbg !45 + %9020 = bitcast <2 x bfloat> %8756 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9017, i32 %9018, i32 %9019, i32 %9020, ptr addrspace(1) %521, i1 %30) #5, !dbg !45 + %9021 = bitcast <2 x bfloat> %8757 to i32, !dbg !45 + %9022 = bitcast <2 x bfloat> %8758 to i32, !dbg !45 + %9023 = bitcast <2 x bfloat> %8759 to i32, !dbg !45 + %9024 = bitcast <2 x bfloat> %8760 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9021, i32 %9022, i32 %9023, i32 %9024, ptr addrspace(1) %522, i1 %650) #5, !dbg !45 + %9025 = bitcast <2 x bfloat> %8761 to i32, !dbg !45 + %9026 = bitcast <2 x bfloat> %8762 to i32, !dbg !45 + %9027 = bitcast <2 x bfloat> %8763 to i32, !dbg !45 + %9028 = bitcast <2 x bfloat> %8764 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9025, i32 %9026, i32 %9027, i32 %9028, ptr addrspace(1) %523, i1 %668) #5, !dbg !45 + %9029 = bitcast <2 x bfloat> %8765 to i32, !dbg !45 + %9030 = bitcast <2 x bfloat> %8766 to i32, !dbg !45 + %9031 = bitcast <2 x bfloat> %8767 to i32, !dbg !45 + %9032 = bitcast <2 x bfloat> %8768 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9029, i32 %9030, i32 %9031, i32 %9032, ptr addrspace(1) %524, i1 %686) #5, !dbg !45 + %9033 = bitcast <2 x bfloat> %8769 to i32, !dbg !45 + %9034 = bitcast <2 x bfloat> %8770 to i32, !dbg !45 + %9035 = bitcast <2 x bfloat> %8771 to i32, !dbg !45 + %9036 = bitcast <2 x bfloat> %8772 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9033, i32 %9034, i32 %9035, i32 %9036, ptr addrspace(1) %525, i1 %704) #5, !dbg !45 + %9037 = bitcast <2 x bfloat> %8773 to i32, !dbg !45 + %9038 = bitcast <2 x bfloat> %8774 to i32, !dbg !45 + %9039 = bitcast <2 x bfloat> %8775 to i32, !dbg !45 + %9040 = bitcast <2 x bfloat> %8776 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9037, i32 %9038, i32 %9039, i32 %9040, ptr addrspace(1) %526, i1 %722) #5, !dbg !45 + %9041 = bitcast <2 x bfloat> %8777 to i32, !dbg !45 + %9042 = bitcast <2 x bfloat> %8778 to i32, !dbg !45 + %9043 = bitcast <2 x bfloat> %8779 to i32, !dbg !45 + %9044 = bitcast <2 x bfloat> %8780 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9041, i32 %9042, i32 %9043, i32 %9044, ptr addrspace(1) %527, i1 %740) #5, !dbg !45 + %9045 = bitcast <2 x bfloat> %8781 to i32, !dbg !45 + %9046 = bitcast <2 x bfloat> %8782 to i32, !dbg !45 + %9047 = bitcast <2 x bfloat> %8783 to i32, !dbg !45 + %9048 = bitcast <2 x bfloat> %8784 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9045, i32 %9046, i32 %9047, i32 %9048, ptr addrspace(1) %528, i1 %758) #5, !dbg !45 + %9049 = bitcast <2 x bfloat> %8785 to i32, !dbg !45 + %9050 = bitcast <2 x bfloat> %8786 to i32, !dbg !45 + %9051 = bitcast <2 x bfloat> %8787 to i32, !dbg !45 + %9052 = bitcast <2 x bfloat> %8788 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9049, i32 %9050, i32 %9051, i32 %9052, ptr addrspace(1) %529, i1 %776) #5, !dbg !45 + %9053 = bitcast <2 x bfloat> %8789 to i32, !dbg !45 + %9054 = bitcast <2 x bfloat> %8790 to i32, !dbg !45 + %9055 = bitcast <2 x bfloat> %8791 to i32, !dbg !45 + %9056 = bitcast <2 x bfloat> %8792 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9053, i32 %9054, i32 %9055, i32 %9056, ptr addrspace(1) %530, i1 %794) #5, !dbg !45 + %9057 = bitcast <2 x bfloat> %8793 to i32, !dbg !45 + %9058 = bitcast <2 x bfloat> %8794 to i32, !dbg !45 + %9059 = bitcast <2 x bfloat> %8795 to i32, !dbg !45 + %9060 = bitcast <2 x bfloat> %8796 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9057, i32 %9058, i32 %9059, i32 %9060, ptr addrspace(1) %531, i1 %812) #5, !dbg !45 + %9061 = bitcast <2 x bfloat> %8797 to i32, !dbg !45 + %9062 = bitcast <2 x bfloat> %8798 to i32, !dbg !45 + %9063 = bitcast <2 x bfloat> %8799 to i32, !dbg !45 + %9064 = bitcast <2 x bfloat> %8800 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9061, i32 %9062, i32 %9063, i32 %9064, ptr addrspace(1) %532, i1 %830) #5, !dbg !45 + %9065 = bitcast <2 x bfloat> %8801 to i32, !dbg !45 + %9066 = bitcast <2 x bfloat> %8802 to i32, !dbg !45 + %9067 = bitcast <2 x bfloat> %8803 to i32, !dbg !45 + %9068 = bitcast <2 x bfloat> %8804 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9065, i32 %9066, i32 %9067, i32 %9068, ptr addrspace(1) %533, i1 %848) #5, !dbg !45 + %9069 = bitcast <2 x bfloat> %8805 to i32, !dbg !45 + %9070 = bitcast <2 x bfloat> %8806 to i32, !dbg !45 + %9071 = bitcast <2 x bfloat> %8807 to i32, !dbg !45 + %9072 = bitcast <2 x bfloat> %8808 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9069, i32 %9070, i32 %9071, i32 %9072, ptr addrspace(1) %534, i1 %866) #5, !dbg !45 + %9073 = bitcast <2 x bfloat> %8809 to i32, !dbg !45 + %9074 = bitcast <2 x bfloat> %8810 to i32, !dbg !45 + %9075 = bitcast <2 x bfloat> %8811 to i32, !dbg !45 + %9076 = bitcast <2 x bfloat> %8812 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9073, i32 %9074, i32 %9075, i32 %9076, ptr addrspace(1) %535, i1 %884) #5, !dbg !45 + %9077 = bitcast <2 x bfloat> %8813 to i32, !dbg !45 + %9078 = bitcast <2 x bfloat> %8814 to i32, !dbg !45 + %9079 = bitcast <2 x bfloat> %8815 to i32, !dbg !45 + %9080 = bitcast <2 x bfloat> %8816 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9077, i32 %9078, i32 %9079, i32 %9080, ptr addrspace(1) %536, i1 %902) #5, !dbg !45 + %9081 = bitcast <2 x bfloat> %8817 to i32, !dbg !45 + %9082 = bitcast <2 x bfloat> %8818 to i32, !dbg !45 + %9083 = bitcast <2 x bfloat> %8819 to i32, !dbg !45 + %9084 = bitcast <2 x bfloat> %8820 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9081, i32 %9082, i32 %9083, i32 %9084, ptr addrspace(1) %537, i1 %920) #5, !dbg !45 + %9085 = bitcast <2 x bfloat> %8821 to i32, !dbg !45 + %9086 = bitcast <2 x bfloat> %8822 to i32, !dbg !45 + %9087 = bitcast <2 x bfloat> %8823 to i32, !dbg !45 + %9088 = bitcast <2 x bfloat> %8824 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9085, i32 %9086, i32 %9087, i32 %9088, ptr addrspace(1) %538, i1 %938) #5, !dbg !45 + %9089 = bitcast <2 x bfloat> %8825 to i32, !dbg !45 + %9090 = bitcast <2 x bfloat> %8826 to i32, !dbg !45 + %9091 = bitcast <2 x bfloat> %8827 to i32, !dbg !45 + %9092 = bitcast <2 x bfloat> %8828 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9089, i32 %9090, i32 %9091, i32 %9092, ptr addrspace(1) %539, i1 %956) #5, !dbg !45 + %9093 = bitcast <2 x bfloat> %8829 to i32, !dbg !45 + %9094 = bitcast <2 x bfloat> %8830 to i32, !dbg !45 + %9095 = bitcast <2 x bfloat> %8831 to i32, !dbg !45 + %9096 = bitcast <2 x bfloat> %8832 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9093, i32 %9094, i32 %9095, i32 %9096, ptr addrspace(1) %540, i1 %974) #5, !dbg !45 + %9097 = bitcast <2 x bfloat> %8833 to i32, !dbg !45 + %9098 = bitcast <2 x bfloat> %8834 to i32, !dbg !45 + %9099 = bitcast <2 x bfloat> %8835 to i32, !dbg !45 + %9100 = bitcast <2 x bfloat> %8836 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9097, i32 %9098, i32 %9099, i32 %9100, ptr addrspace(1) %541, i1 %992) #5, !dbg !45 + %9101 = bitcast <2 x bfloat> %8837 to i32, !dbg !45 + %9102 = bitcast <2 x bfloat> %8838 to i32, !dbg !45 + %9103 = bitcast <2 x bfloat> %8839 to i32, !dbg !45 + %9104 = bitcast <2 x bfloat> %8840 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9101, i32 %9102, i32 %9103, i32 %9104, ptr addrspace(1) %542, i1 %1010) #5, !dbg !45 + %9105 = bitcast <2 x bfloat> %8841 to i32, !dbg !45 + %9106 = bitcast <2 x bfloat> %8842 to i32, !dbg !45 + %9107 = bitcast <2 x bfloat> %8843 to i32, !dbg !45 + %9108 = bitcast <2 x bfloat> %8844 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9105, i32 %9106, i32 %9107, i32 %9108, ptr addrspace(1) %543, i1 %1028) #5, !dbg !45 + %9109 = bitcast <2 x bfloat> %8845 to i32, !dbg !45 + %9110 = bitcast <2 x bfloat> %8846 to i32, !dbg !45 + %9111 = bitcast <2 x bfloat> %8847 to i32, !dbg !45 + %9112 = bitcast <2 x bfloat> %8848 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9109, i32 %9110, i32 %9111, i32 %9112, ptr addrspace(1) %544, i1 %1046) #5, !dbg !45 + %9113 = bitcast <2 x bfloat> %8849 to i32, !dbg !45 + %9114 = bitcast <2 x bfloat> %8850 to i32, !dbg !45 + %9115 = bitcast <2 x bfloat> %8851 to i32, !dbg !45 + %9116 = bitcast <2 x bfloat> %8852 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9113, i32 %9114, i32 %9115, i32 %9116, ptr addrspace(1) %545, i1 %1064) #5, !dbg !45 + %9117 = bitcast <2 x bfloat> %8853 to i32, !dbg !45 + %9118 = bitcast <2 x bfloat> %8854 to i32, !dbg !45 + %9119 = bitcast <2 x bfloat> %8855 to i32, !dbg !45 + %9120 = bitcast <2 x bfloat> %8856 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9117, i32 %9118, i32 %9119, i32 %9120, ptr addrspace(1) %546, i1 %1082) #5, !dbg !45 + %9121 = bitcast <2 x bfloat> %8857 to i32, !dbg !45 + %9122 = bitcast <2 x bfloat> %8858 to i32, !dbg !45 + %9123 = bitcast <2 x bfloat> %8859 to i32, !dbg !45 + %9124 = bitcast <2 x bfloat> %8860 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9121, i32 %9122, i32 %9123, i32 %9124, ptr addrspace(1) %547, i1 %1100) #5, !dbg !45 + %9125 = bitcast <2 x bfloat> %8861 to i32, !dbg !45 + %9126 = bitcast <2 x bfloat> %8862 to i32, !dbg !45 + %9127 = bitcast <2 x bfloat> %8863 to i32, !dbg !45 + %9128 = bitcast <2 x bfloat> %8864 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9125, i32 %9126, i32 %9127, i32 %9128, ptr addrspace(1) %548, i1 %1118) #5, !dbg !45 + %9129 = bitcast <2 x bfloat> %8865 to i32, !dbg !45 + %9130 = bitcast <2 x bfloat> %8866 to i32, !dbg !45 + %9131 = bitcast <2 x bfloat> %8867 to i32, !dbg !45 + %9132 = bitcast <2 x bfloat> %8868 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9129, i32 %9130, i32 %9131, i32 %9132, ptr addrspace(1) %549, i1 %1136) #5, !dbg !45 + %9133 = bitcast <2 x bfloat> %8869 to i32, !dbg !45 + %9134 = bitcast <2 x bfloat> %8870 to i32, !dbg !45 + %9135 = bitcast <2 x bfloat> %8871 to i32, !dbg !45 + %9136 = bitcast <2 x bfloat> %8872 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9133, i32 %9134, i32 %9135, i32 %9136, ptr addrspace(1) %550, i1 %1154) #5, !dbg !45 + %9137 = bitcast <2 x bfloat> %8873 to i32, !dbg !45 + %9138 = bitcast <2 x bfloat> %8874 to i32, !dbg !45 + %9139 = bitcast <2 x bfloat> %8875 to i32, !dbg !45 + %9140 = bitcast <2 x bfloat> %8876 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9137, i32 %9138, i32 %9139, i32 %9140, ptr addrspace(1) %551, i1 %1172) #5, !dbg !45 + %9141 = bitcast <2 x bfloat> %8877 to i32, !dbg !45 + %9142 = bitcast <2 x bfloat> %8878 to i32, !dbg !45 + %9143 = bitcast <2 x bfloat> %8879 to i32, !dbg !45 + %9144 = bitcast <2 x bfloat> %8880 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9141, i32 %9142, i32 %9143, i32 %9144, ptr addrspace(1) %552, i1 %1190) #5, !dbg !45 + %9145 = bitcast <2 x bfloat> %8881 to i32, !dbg !45 + %9146 = bitcast <2 x bfloat> %8882 to i32, !dbg !45 + %9147 = bitcast <2 x bfloat> %8883 to i32, !dbg !45 + %9148 = bitcast <2 x bfloat> %8884 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9145, i32 %9146, i32 %9147, i32 %9148, ptr addrspace(1) %553, i1 %1208) #5, !dbg !45 + %9149 = bitcast <2 x bfloat> %8885 to i32, !dbg !45 + %9150 = bitcast <2 x bfloat> %8886 to i32, !dbg !45 + %9151 = bitcast <2 x bfloat> %8887 to i32, !dbg !45 + %9152 = bitcast <2 x bfloat> %8888 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9149, i32 %9150, i32 %9151, i32 %9152, ptr addrspace(1) %554, i1 %1226) #5, !dbg !45 + %9153 = bitcast <2 x bfloat> %8889 to i32, !dbg !45 + %9154 = bitcast <2 x bfloat> %8890 to i32, !dbg !45 + %9155 = bitcast <2 x bfloat> %8891 to i32, !dbg !45 + %9156 = bitcast <2 x bfloat> %8892 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9153, i32 %9154, i32 %9155, i32 %9156, ptr addrspace(1) %555, i1 %1244) #5, !dbg !45 + %9157 = bitcast <2 x bfloat> %8893 to i32, !dbg !45 + %9158 = bitcast <2 x bfloat> %8894 to i32, !dbg !45 + %9159 = bitcast <2 x bfloat> %8895 to i32, !dbg !45 + %9160 = bitcast <2 x bfloat> %8896 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9157, i32 %9158, i32 %9159, i32 %9160, ptr addrspace(1) %556, i1 %1262) #5, !dbg !45 + %9161 = bitcast <2 x bfloat> %8897 to i32, !dbg !45 + %9162 = bitcast <2 x bfloat> %8898 to i32, !dbg !45 + %9163 = bitcast <2 x bfloat> %8899 to i32, !dbg !45 + %9164 = bitcast <2 x bfloat> %8900 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9161, i32 %9162, i32 %9163, i32 %9164, ptr addrspace(1) %557, i1 %1280) #5, !dbg !45 + %9165 = bitcast <2 x bfloat> %8901 to i32, !dbg !45 + %9166 = bitcast <2 x bfloat> %8902 to i32, !dbg !45 + %9167 = bitcast <2 x bfloat> %8903 to i32, !dbg !45 + %9168 = bitcast <2 x bfloat> %8904 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9165, i32 %9166, i32 %9167, i32 %9168, ptr addrspace(1) %558, i1 %1298) #5, !dbg !45 + %9169 = bitcast <2 x bfloat> %8905 to i32, !dbg !45 + %9170 = bitcast <2 x bfloat> %8906 to i32, !dbg !45 + %9171 = bitcast <2 x bfloat> %8907 to i32, !dbg !45 + %9172 = bitcast <2 x bfloat> %8908 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9169, i32 %9170, i32 %9171, i32 %9172, ptr addrspace(1) %559, i1 %1316) #5, !dbg !45 + %9173 = bitcast <2 x bfloat> %8909 to i32, !dbg !45 + %9174 = bitcast <2 x bfloat> %8910 to i32, !dbg !45 + %9175 = bitcast <2 x bfloat> %8911 to i32, !dbg !45 + %9176 = bitcast <2 x bfloat> %8912 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9173, i32 %9174, i32 %9175, i32 %9176, ptr addrspace(1) %560, i1 %1334) #5, !dbg !45 + %9177 = bitcast <2 x bfloat> %8913 to i32, !dbg !45 + %9178 = bitcast <2 x bfloat> %8914 to i32, !dbg !45 + %9179 = bitcast <2 x bfloat> %8915 to i32, !dbg !45 + %9180 = bitcast <2 x bfloat> %8916 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9177, i32 %9178, i32 %9179, i32 %9180, ptr addrspace(1) %561, i1 %1352) #5, !dbg !45 + %9181 = bitcast <2 x bfloat> %8917 to i32, !dbg !45 + %9182 = bitcast <2 x bfloat> %8918 to i32, !dbg !45 + %9183 = bitcast <2 x bfloat> %8919 to i32, !dbg !45 + %9184 = bitcast <2 x bfloat> %8920 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9181, i32 %9182, i32 %9183, i32 %9184, ptr addrspace(1) %562, i1 %1370) #5, !dbg !45 + %9185 = bitcast <2 x bfloat> %8921 to i32, !dbg !45 + %9186 = bitcast <2 x bfloat> %8922 to i32, !dbg !45 + %9187 = bitcast <2 x bfloat> %8923 to i32, !dbg !45 + %9188 = bitcast <2 x bfloat> %8924 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9185, i32 %9186, i32 %9187, i32 %9188, ptr addrspace(1) %563, i1 %1388) #5, !dbg !45 + %9189 = bitcast <2 x bfloat> %8925 to i32, !dbg !45 + %9190 = bitcast <2 x bfloat> %8926 to i32, !dbg !45 + %9191 = bitcast <2 x bfloat> %8927 to i32, !dbg !45 + %9192 = bitcast <2 x bfloat> %8928 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9189, i32 %9190, i32 %9191, i32 %9192, ptr addrspace(1) %564, i1 %1406) #5, !dbg !45 + %9193 = bitcast <2 x bfloat> %8929 to i32, !dbg !45 + %9194 = bitcast <2 x bfloat> %8930 to i32, !dbg !45 + %9195 = bitcast <2 x bfloat> %8931 to i32, !dbg !45 + %9196 = bitcast <2 x bfloat> %8932 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9193, i32 %9194, i32 %9195, i32 %9196, ptr addrspace(1) %565, i1 %1424) #5, !dbg !45 + %9197 = bitcast <2 x bfloat> %8933 to i32, !dbg !45 + %9198 = bitcast <2 x bfloat> %8934 to i32, !dbg !45 + %9199 = bitcast <2 x bfloat> %8935 to i32, !dbg !45 + %9200 = bitcast <2 x bfloat> %8936 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9197, i32 %9198, i32 %9199, i32 %9200, ptr addrspace(1) %566, i1 %1442) #5, !dbg !45 + %9201 = bitcast <2 x bfloat> %8937 to i32, !dbg !45 + %9202 = bitcast <2 x bfloat> %8938 to i32, !dbg !45 + %9203 = bitcast <2 x bfloat> %8939 to i32, !dbg !45 + %9204 = bitcast <2 x bfloat> %8940 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9201, i32 %9202, i32 %9203, i32 %9204, ptr addrspace(1) %567, i1 %1460) #5, !dbg !45 + %9205 = bitcast <2 x bfloat> %8941 to i32, !dbg !45 + %9206 = bitcast <2 x bfloat> %8942 to i32, !dbg !45 + %9207 = bitcast <2 x bfloat> %8943 to i32, !dbg !45 + %9208 = bitcast <2 x bfloat> %8944 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9205, i32 %9206, i32 %9207, i32 %9208, ptr addrspace(1) %568, i1 %1478) #5, !dbg !45 + %9209 = bitcast <2 x bfloat> %8945 to i32, !dbg !45 + %9210 = bitcast <2 x bfloat> %8946 to i32, !dbg !45 + %9211 = bitcast <2 x bfloat> %8947 to i32, !dbg !45 + %9212 = bitcast <2 x bfloat> %8948 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9209, i32 %9210, i32 %9211, i32 %9212, ptr addrspace(1) %569, i1 %1496) #5, !dbg !45 + %9213 = bitcast <2 x bfloat> %8949 to i32, !dbg !45 + %9214 = bitcast <2 x bfloat> %8950 to i32, !dbg !45 + %9215 = bitcast <2 x bfloat> %8951 to i32, !dbg !45 + %9216 = bitcast <2 x bfloat> %8952 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9213, i32 %9214, i32 %9215, i32 %9216, ptr addrspace(1) %570, i1 %1514) #5, !dbg !45 + %9217 = bitcast <2 x bfloat> %8953 to i32, !dbg !45 + %9218 = bitcast <2 x bfloat> %8954 to i32, !dbg !45 + %9219 = bitcast <2 x bfloat> %8955 to i32, !dbg !45 + %9220 = bitcast <2 x bfloat> %8956 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9217, i32 %9218, i32 %9219, i32 %9220, ptr addrspace(1) %571, i1 %1532) #5, !dbg !45 + %9221 = bitcast <2 x bfloat> %8957 to i32, !dbg !45 + %9222 = bitcast <2 x bfloat> %8958 to i32, !dbg !45 + %9223 = bitcast <2 x bfloat> %8959 to i32, !dbg !45 + %9224 = bitcast <2 x bfloat> %8960 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9221, i32 %9222, i32 %9223, i32 %9224, ptr addrspace(1) %572, i1 %1550) #5, !dbg !45 + %9225 = bitcast <2 x bfloat> %8961 to i32, !dbg !45 + %9226 = bitcast <2 x bfloat> %8962 to i32, !dbg !45 + %9227 = bitcast <2 x bfloat> %8963 to i32, !dbg !45 + %9228 = bitcast <2 x bfloat> %8964 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9225, i32 %9226, i32 %9227, i32 %9228, ptr addrspace(1) %573, i1 %1568) #5, !dbg !45 + %9229 = bitcast <2 x bfloat> %8965 to i32, !dbg !45 + %9230 = bitcast <2 x bfloat> %8966 to i32, !dbg !45 + %9231 = bitcast <2 x bfloat> %8967 to i32, !dbg !45 + %9232 = bitcast <2 x bfloat> %8968 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9229, i32 %9230, i32 %9231, i32 %9232, ptr addrspace(1) %574, i1 %1586) #5, !dbg !45 + %9233 = bitcast <2 x bfloat> %8969 to i32, !dbg !45 + %9234 = bitcast <2 x bfloat> %8970 to i32, !dbg !45 + %9235 = bitcast <2 x bfloat> %8971 to i32, !dbg !45 + %9236 = bitcast <2 x bfloat> %8972 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9233, i32 %9234, i32 %9235, i32 %9236, ptr addrspace(1) %575, i1 %1604) #5, !dbg !45 + %9237 = bitcast <2 x bfloat> %8973 to i32, !dbg !45 + %9238 = bitcast <2 x bfloat> %8974 to i32, !dbg !45 + %9239 = bitcast <2 x bfloat> %8975 to i32, !dbg !45 + %9240 = bitcast <2 x bfloat> %8976 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9237, i32 %9238, i32 %9239, i32 %9240, ptr addrspace(1) %576, i1 %1622) #5, !dbg !45 + %9241 = bitcast <2 x bfloat> %8977 to i32, !dbg !45 + %9242 = bitcast <2 x bfloat> %8978 to i32, !dbg !45 + %9243 = bitcast <2 x bfloat> %8979 to i32, !dbg !45 + %9244 = bitcast <2 x bfloat> %8980 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9241, i32 %9242, i32 %9243, i32 %9244, ptr addrspace(1) %577, i1 %1640) #5, !dbg !45 + %9245 = bitcast <2 x bfloat> %8981 to i32, !dbg !45 + %9246 = bitcast <2 x bfloat> %8982 to i32, !dbg !45 + %9247 = bitcast <2 x bfloat> %8983 to i32, !dbg !45 + %9248 = bitcast <2 x bfloat> %8984 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9245, i32 %9246, i32 %9247, i32 %9248, ptr addrspace(1) %578, i1 %1658) #5, !dbg !45 + %9249 = bitcast <2 x bfloat> %8985 to i32, !dbg !45 + %9250 = bitcast <2 x bfloat> %8986 to i32, !dbg !45 + %9251 = bitcast <2 x bfloat> %8987 to i32, !dbg !45 + %9252 = bitcast <2 x bfloat> %8988 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9249, i32 %9250, i32 %9251, i32 %9252, ptr addrspace(1) %579, i1 %1676) #5, !dbg !45 + %9253 = bitcast <2 x bfloat> %8989 to i32, !dbg !45 + %9254 = bitcast <2 x bfloat> %8990 to i32, !dbg !45 + %9255 = bitcast <2 x bfloat> %8991 to i32, !dbg !45 + %9256 = bitcast <2 x bfloat> %8992 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9253, i32 %9254, i32 %9255, i32 %9256, ptr addrspace(1) %580, i1 %1694) #5, !dbg !45 + %9257 = bitcast <2 x bfloat> %8993 to i32, !dbg !45 + %9258 = bitcast <2 x bfloat> %8994 to i32, !dbg !45 + %9259 = bitcast <2 x bfloat> %8995 to i32, !dbg !45 + %9260 = bitcast <2 x bfloat> %8996 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9257, i32 %9258, i32 %9259, i32 %9260, ptr addrspace(1) %581, i1 %1712) #5, !dbg !45 + ret void, !dbg !46 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(argmem: write) +declare void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x2.b16.p3(ptr addrspace(3) writeonly captures(none), i32, i32) #2 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.bar.warp.sync(i32) #3 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #4 + +attributes #0 = { nounwind "nvvm.reqntid"="32" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(argmem: write) } +attributes #3 = { convergent nocallback nounwind } +attributes #4 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #5 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = distinct !DISubprogram(name: "l2norm_bwd_kernel", linkageName: "l2norm_bwd_kernel", scope: !1, file: !1, line: 93, type: !5, scopeLine: 93, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!5 = !DISubroutineType(cc: DW_CC_normal, types: !6) +!6 = !{null, !7, !9, !7, !7, !10, !11, !7, !7} +!7 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !8, size: 64, dwarfAddressSpace: 1) +!8 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!9 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !10, size: 64, dwarfAddressSpace: 1) +!10 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!11 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!12 = !DILocation(line: 94, column: 24, scope: !4) +!13 = !DILocation(line: 94, column: 30, scope: !4) +!14 = !DILocation(line: 95, column: 16, scope: !4) +!15 = !DILocation(line: 95, column: 34, scope: !4) +!16 = !DILocation(line: 95, column: 21, scope: !4) +!17 = !DILocation(line: 97, column: 16, scope: !4) +!18 = !DILocation(line: 98, column: 30, scope: !4) +!19 = !DILocation(line: 99, column: 29, scope: !4) +!20 = !DILocation(line: 99, column: 14, scope: !4) +!21 = !DILocation(line: 99, column: 33, scope: !4) +!22 = !DILocation(line: 100, column: 20, scope: !4) +!23 = !DILocation(line: 101, column: 16, scope: !4) +!24 = !DILocation(line: 101, column: 35, scope: !4) +!25 = !DILocation(line: 102, column: 16, scope: !4) +!26 = !DILocation(line: 102, column: 35, scope: !4) +!27 = !DILocation(line: 104, column: 18, scope: !4) +!28 = !DILocation(line: 104, column: 47, scope: !4) +!29 = !DILocation(line: 105, column: 21, scope: !4) +!30 = !DILocation(line: 106, column: 19, scope: !4) +!31 = !DILocation(line: 106, column: 49, scope: !4) +!32 = !DILocation(line: 107, column: 25, scope: !4) +!33 = !DILocation(line: 107, column: 18, scope: !4) +!34 = !DILocation(line: 107, column: 50, scope: !4) +!35 = !DILocation(line: 263, column: 15, scope: !36, inlinedAt: !38) +!36 = distinct !DILexicalBlockFile(scope: !4, file: !37, discriminator: 0) +!37 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!38 = !DILocation(line: 293, column: 36, scope: !36, inlinedAt: !39) +!39 = !DILocation(line: 107, column: 55, scope: !40) +!40 = distinct !DILexicalBlockFile(scope: !4, file: !1, discriminator: 0) +!41 = !DILocation(line: 107, column: 69, scope: !4) +!42 = !DILocation(line: 107, column: 75, scope: !4) +!43 = !DILocation(line: 107, column: 36, scope: !4) +!44 = !DILocation(line: 108, column: 27, scope: !4) +!45 = !DILocation(line: 108, column: 19, scope: !4) +!46 = !DILocation(line: 108, column: 4, scope: !4) diff --git a/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.ptx b/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..4869c24642edb5f12d95ab35238acc951c476a96 --- /dev/null +++ b/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.ptx @@ -0,0 +1,10132 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_bwd_kernel // -- Begin function l2norm_bwd_kernel +.extern .shared .align 16 .b8 global_smem[]; + // @l2norm_bwd_kernel +.visible .entry l2norm_bwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_2, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_3, + .param .f32 l2norm_bwd_kernel_param_4, + .param .u32 l2norm_bwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_6, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_7 +) +.reqntid 32 +{ + .reg .pred %p<115>; + .reg .b16 %rs<1184>; + .reg .b32 %r<5048>; + .reg .b64 %rd<1419>; + .loc 1 93 0 // l2norm.py:93:0 +$L__func_begin0: + .loc 1 93 0 // l2norm.py:93:0 + +// %bb.0: + ld.param.b64 %rd197, [l2norm_bwd_kernel_param_0]; + ld.param.b64 %rd198, [l2norm_bwd_kernel_param_1]; +$L__tmp0: + .loc 1 94 24 // l2norm.py:94:24 + mov.u32 %r774, %ctaid.x; + ld.param.b64 %rd199, [l2norm_bwd_kernel_param_2]; + .loc 1 95 16 // l2norm.py:95:16 + mul.wide.u32 %rd200, %r774, 128; + ld.param.b64 %rd201, [l2norm_bwd_kernel_param_3]; + .loc 1 95 34 // l2norm.py:95:34 + mov.u32 %r775, %tid.x; + and.b32 %r776, %r775, 16; + ld.param.s32 %rd202, [l2norm_bwd_kernel_param_5]; + bfe.u32 %r777, %r775, 4, 1; + or.b32 %r778, %r777, 2; + or.b32 %r779, %r777, 4; + or.b32 %r780, %r777, 6; + shl.b32 %r781, %r775, 2; + .loc 1 95 21 // l2norm.py:95:21 + cvt.u64.u32 %rd203, %r777; + cvt.u64.u32 %rd204, %r778; + cvt.u64.u32 %rd205, %r779; + cvt.u64.u32 %rd206, %r780; + or.b64 %rd207, %rd200, %rd203; + or.b64 %rd208, %rd200, %rd204; + or.b64 %rd209, %rd200, %rd205; + or.b64 %rd210, %rd200, %rd206; + .loc 1 97 16 // l2norm.py:97:16 + setp.lt.s64 %p1, %rd207, %rd202; + setp.lt.s64 %p2, %rd208, %rd202; + setp.lt.s64 %p3, %rd209, %rd202; + setp.lt.s64 %p4, %rd210, %rd202; + .loc 1 95 34 // l2norm.py:95:34 + and.b32 %r782, %r781, 124; + or.b32 %r783, %r782, 1; + .loc 1 95 21 // l2norm.py:95:21 + or.b32 %r784, %r777, 8; + or.b32 %r785, %r777, 10; + or.b32 %r786, %r777, 12; + or.b32 %r787, %r777, 14; + or.b32 %r788, %r777, 16; + or.b32 %r789, %r777, 18; + or.b32 %r790, %r777, 20; + or.b32 %r791, %r777, 22; + or.b32 %r792, %r777, 24; + or.b32 %r793, %r777, 26; + or.b32 %r794, %r777, 28; + or.b32 %r795, %r777, 30; + or.b32 %r796, %r777, 32; + or.b32 %r797, %r777, 34; + or.b32 %r798, %r777, 36; + or.b32 %r799, %r777, 38; + or.b32 %r800, %r777, 68; + or.b32 %r801, %r777, 70; + or.b32 %r802, %r777, 64; + or.b32 %r803, %r777, 66; + or.b32 %r804, %r777, 60; + or.b32 %r805, %r777, 62; + or.b32 %r806, %r777, 56; + or.b32 %r807, %r777, 58; + or.b32 %r808, %r777, 52; + or.b32 %r809, %r777, 54; + or.b32 %r810, %r777, 48; + or.b32 %r811, %r777, 50; + or.b32 %r812, %r777, 44; + or.b32 %r813, %r777, 46; + or.b32 %r814, %r777, 40; + or.b32 %r815, %r777, 42; + or.b32 %r816, %r777, 100; + or.b32 %r817, %r777, 102; + or.b32 %r818, %r777, 96; + or.b32 %r819, %r777, 98; + or.b32 %r820, %r777, 92; + or.b32 %r821, %r777, 94; + or.b32 %r822, %r777, 88; + or.b32 %r823, %r777, 90; + or.b32 %r824, %r777, 84; + or.b32 %r825, %r777, 86; + or.b32 %r826, %r777, 80; + or.b32 %r827, %r777, 82; + or.b32 %r828, %r777, 76; + or.b32 %r829, %r777, 78; + or.b32 %r830, %r777, 72; + or.b32 %r831, %r777, 74; + or.b32 %r832, %r777, 116; + or.b32 %r833, %r777, 118; + or.b32 %r834, %r777, 112; + or.b32 %r835, %r777, 114; + or.b32 %r836, %r777, 108; + or.b32 %r837, %r777, 110; + or.b32 %r838, %r777, 104; + or.b32 %r839, %r777, 106; + or.b32 %r840, %r777, 124; + or.b32 %r841, %r777, 126; + or.b32 %r842, %r777, 120; + or.b32 %r843, %r777, 122; + or.b32 %r844, %r782, 2; + or.b32 %r845, %r782, 3; + cvt.u64.u32 %rd211, %r831; + cvt.u64.u32 %rd212, %r830; + cvt.u64.u32 %rd213, %r829; + cvt.u64.u32 %rd214, %r828; + cvt.u64.u32 %rd215, %r827; + cvt.u64.u32 %rd216, %r826; + cvt.u64.u32 %rd217, %r825; + cvt.u64.u32 %rd218, %r824; + cvt.u64.u32 %rd219, %r823; + cvt.u64.u32 %rd220, %r822; + cvt.u64.u32 %rd221, %r821; + cvt.u64.u32 %rd222, %r820; + cvt.u64.u32 %rd223, %r819; + cvt.u64.u32 %rd224, %r818; + cvt.u64.u32 %rd225, %r817; + cvt.u64.u32 %rd226, %r816; + cvt.u64.u32 %rd227, %r839; + cvt.u64.u32 %rd228, %r838; + cvt.u64.u32 %rd229, %r837; + cvt.u64.u32 %rd230, %r836; + cvt.u64.u32 %rd231, %r835; + cvt.u64.u32 %rd232, %r834; + cvt.u64.u32 %rd233, %r833; + cvt.u64.u32 %rd234, %r832; + cvt.u64.u32 %rd235, %r843; + cvt.u64.u32 %rd236, %r842; + cvt.u64.u32 %rd237, %r841; + cvt.u64.u32 %rd238, %r840; + cvt.u64.u32 %rd239, %r782; + cvt.u64.u32 %rd240, %r783; + cvt.u64.u32 %rd241, %r845; + cvt.u64.u32 %rd242, %r844; + cvt.u64.u32 %rd243, %r815; + cvt.u64.u32 %rd244, %r814; + cvt.u64.u32 %rd245, %r813; + cvt.u64.u32 %rd246, %r812; + cvt.u64.u32 %rd247, %r811; + cvt.u64.u32 %rd248, %r810; + cvt.u64.u32 %rd249, %r809; + cvt.u64.u32 %rd250, %r808; + cvt.u64.u32 %rd251, %r807; + cvt.u64.u32 %rd252, %r806; + cvt.u64.u32 %rd253, %r805; + cvt.u64.u32 %rd254, %r804; + cvt.u64.u32 %rd255, %r803; + cvt.u64.u32 %rd256, %r802; + cvt.u64.u32 %rd257, %r801; + cvt.u64.u32 %rd258, %r800; + cvt.u64.u32 %rd259, %r799; + cvt.u64.u32 %rd260, %r798; + cvt.u64.u32 %rd261, %r797; + cvt.u64.u32 %rd262, %r796; + cvt.u64.u32 %rd263, %r795; + cvt.u64.u32 %rd264, %r794; + cvt.u64.u32 %rd265, %r793; + cvt.u64.u32 %rd266, %r792; + cvt.u64.u32 %rd267, %r791; + cvt.u64.u32 %rd268, %r790; + cvt.u64.u32 %rd269, %r789; + cvt.u64.u32 %rd270, %r788; + cvt.u64.u32 %rd271, %r787; + cvt.u64.u32 %rd272, %r786; + cvt.u64.u32 %rd273, %r785; + cvt.u64.u32 %rd274, %r784; + or.b64 %rd275, %rd200, %rd274; + or.b64 %rd276, %rd200, %rd273; + or.b64 %rd277, %rd200, %rd272; + or.b64 %rd278, %rd200, %rd271; + or.b64 %rd279, %rd200, %rd270; + or.b64 %rd280, %rd200, %rd269; + or.b64 %rd281, %rd200, %rd268; + or.b64 %rd282, %rd200, %rd267; + or.b64 %rd283, %rd200, %rd266; + or.b64 %rd284, %rd200, %rd265; + or.b64 %rd285, %rd200, %rd264; + or.b64 %rd286, %rd200, %rd263; + or.b64 %rd287, %rd200, %rd262; + or.b64 %rd288, %rd200, %rd261; + or.b64 %rd289, %rd200, %rd260; + or.b64 %rd290, %rd200, %rd259; + or.b64 %rd291, %rd200, %rd258; + or.b64 %rd292, %rd200, %rd257; + or.b64 %rd293, %rd200, %rd256; + or.b64 %rd294, %rd200, %rd255; + or.b64 %rd295, %rd200, %rd254; + or.b64 %rd296, %rd200, %rd253; + or.b64 %rd297, %rd200, %rd252; + or.b64 %rd298, %rd200, %rd251; + or.b64 %rd299, %rd200, %rd250; + or.b64 %rd300, %rd200, %rd249; + or.b64 %rd301, %rd200, %rd248; + or.b64 %rd302, %rd200, %rd247; + or.b64 %rd303, %rd200, %rd246; + or.b64 %rd304, %rd200, %rd245; + or.b64 %rd305, %rd200, %rd244; + or.b64 %rd306, %rd200, %rd243; + or.b64 %rd307, %rd200, %rd242; + or.b64 %rd308, %rd200, %rd241; + or.b64 %rd309, %rd200, %rd240; + or.b64 %rd310, %rd200, %rd239; + or.b64 %rd311, %rd200, %rd238; + or.b64 %rd312, %rd200, %rd237; + or.b64 %rd313, %rd200, %rd236; + or.b64 %rd314, %rd200, %rd235; + or.b64 %rd315, %rd200, %rd234; + or.b64 %rd316, %rd200, %rd233; + or.b64 %rd317, %rd200, %rd232; + or.b64 %rd318, %rd200, %rd231; + or.b64 %rd319, %rd200, %rd230; + or.b64 %rd320, %rd200, %rd229; + or.b64 %rd321, %rd200, %rd228; + or.b64 %rd322, %rd200, %rd227; + or.b64 %rd323, %rd200, %rd226; + or.b64 %rd324, %rd200, %rd225; + or.b64 %rd325, %rd200, %rd224; + or.b64 %rd326, %rd200, %rd223; + or.b64 %rd327, %rd200, %rd222; + or.b64 %rd328, %rd200, %rd221; + or.b64 %rd329, %rd200, %rd220; + or.b64 %rd330, %rd200, %rd219; + or.b64 %rd331, %rd200, %rd218; + or.b64 %rd332, %rd200, %rd217; + or.b64 %rd333, %rd200, %rd216; + or.b64 %rd334, %rd200, %rd215; + or.b64 %rd335, %rd200, %rd214; + or.b64 %rd336, %rd200, %rd213; + or.b64 %rd337, %rd200, %rd212; + or.b64 %rd338, %rd200, %rd211; + .loc 1 97 16 // l2norm.py:97:16 + setp.lt.s64 %p69, %rd338, %rd202; + selp.b16 %rs1, 1, 0, %p69; + shl.b16 %rs2, %rs1, 2; + setp.lt.s64 %p70, %rd337, %rd202; + selp.b16 %rs3, -1, 0, %p70; + shl.b16 %rs4, %rs3, 3; + or.b16 %rs5, %rs4, %rs2; + setp.lt.s64 %p71, %rd336, %rd202; + selp.b16 %rs6, 1, 0, %p71; + setp.lt.s64 %p72, %rd335, %rd202; + selp.b16 %rs7, -1, 0, %p72; + shl.b16 %rs8, %rs7, 1; + or.b16 %rs9, %rs6, %rs8; + and.b16 %rs10, %rs9, 3; + or.b16 %rs11, %rs10, %rs5; + shl.b16 %rs12, %rs11, 12; + setp.lt.s64 %p73, %rd334, %rd202; + selp.b16 %rs13, 1, 0, %p73; + shl.b16 %rs14, %rs13, 2; + setp.lt.s64 %p74, %rd333, %rd202; + selp.b16 %rs15, -1, 0, %p74; + shl.b16 %rs16, %rs15, 3; + or.b16 %rs17, %rs16, %rs14; + setp.lt.s64 %p75, %rd332, %rd202; + selp.b16 %rs18, 1, 0, %p75; + setp.lt.s64 %p76, %rd331, %rd202; + selp.b16 %rs19, -1, 0, %p76; + shl.b16 %rs20, %rs19, 1; + or.b16 %rs21, %rs18, %rs20; + and.b16 %rs22, %rs21, 3; + or.b16 %rs23, %rs22, %rs17; + and.b16 %rs24, %rs23, 15; + shl.b16 %rs25, %rs24, 8; + or.b16 %rs26, %rs12, %rs25; + setp.lt.s64 %p77, %rd330, %rd202; + selp.b16 %rs27, 1, 0, %p77; + shl.b16 %rs28, %rs27, 2; + setp.lt.s64 %p78, %rd329, %rd202; + selp.b16 %rs29, -1, 0, %p78; + shl.b16 %rs30, %rs29, 3; + or.b16 %rs31, %rs30, %rs28; + setp.lt.s64 %p79, %rd328, %rd202; + selp.b16 %rs32, 1, 0, %p79; + setp.lt.s64 %p80, %rd327, %rd202; + selp.b16 %rs33, -1, 0, %p80; + shl.b16 %rs34, %rs33, 1; + or.b16 %rs35, %rs32, %rs34; + and.b16 %rs36, %rs35, 3; + or.b16 %rs37, %rs36, %rs31; + shl.b16 %rs38, %rs37, 4; + setp.lt.s64 %p81, %rd326, %rd202; + selp.b16 %rs39, 1, 0, %p81; + shl.b16 %rs40, %rs39, 2; + setp.lt.s64 %p82, %rd325, %rd202; + selp.b16 %rs41, -1, 0, %p82; + shl.b16 %rs42, %rs41, 3; + or.b16 %rs43, %rs42, %rs40; + setp.lt.s64 %p83, %rd324, %rd202; + selp.b16 %rs44, 1, 0, %p83; + setp.lt.s64 %p84, %rd323, %rd202; + selp.b16 %rs45, -1, 0, %p84; + shl.b16 %rs46, %rs45, 1; + or.b16 %rs47, %rs44, %rs46; + and.b16 %rs48, %rs47, 3; + or.b16 %rs49, %rs48, %rs43; + and.b16 %rs50, %rs49, 15; + or.b16 %rs51, %rs50, %rs38; + and.b16 %rs52, %rs51, 255; + or.b16 %rs53, %rs52, %rs26; + cvt.u32.u16 %r846, %rs53; + shl.b32 %r847, %r846, 16; + setp.lt.s64 %p85, %rd322, %rd202; + selp.b16 %rs54, 1, 0, %p85; + shl.b16 %rs55, %rs54, 2; + setp.lt.s64 %p86, %rd321, %rd202; + selp.b16 %rs56, -1, 0, %p86; + shl.b16 %rs57, %rs56, 3; + or.b16 %rs58, %rs57, %rs55; + setp.lt.s64 %p87, %rd320, %rd202; + selp.b16 %rs59, 1, 0, %p87; + setp.lt.s64 %p88, %rd319, %rd202; + selp.b16 %rs60, -1, 0, %p88; + shl.b16 %rs61, %rs60, 1; + or.b16 %rs62, %rs59, %rs61; + and.b16 %rs63, %rs62, 3; + or.b16 %rs64, %rs63, %rs58; + shl.b16 %rs65, %rs64, 12; + setp.lt.s64 %p89, %rd318, %rd202; + selp.b16 %rs66, 1, 0, %p89; + shl.b16 %rs67, %rs66, 2; + setp.lt.s64 %p90, %rd317, %rd202; + selp.b16 %rs68, -1, 0, %p90; + shl.b16 %rs69, %rs68, 3; + or.b16 %rs70, %rs69, %rs67; + setp.lt.s64 %p91, %rd316, %rd202; + selp.b16 %rs71, 1, 0, %p91; + setp.lt.s64 %p92, %rd315, %rd202; + selp.b16 %rs72, -1, 0, %p92; + shl.b16 %rs73, %rs72, 1; + or.b16 %rs74, %rs71, %rs73; + and.b16 %rs75, %rs74, 3; + or.b16 %rs76, %rs75, %rs70; + and.b16 %rs77, %rs76, 15; + shl.b16 %rs78, %rs77, 8; + or.b16 %rs79, %rs65, %rs78; + setp.lt.s64 %p93, %rd314, %rd202; + selp.b16 %rs80, 1, 0, %p93; + shl.b16 %rs81, %rs80, 2; + setp.lt.s64 %p94, %rd313, %rd202; + selp.b16 %rs82, -1, 0, %p94; + shl.b16 %rs83, %rs82, 3; + or.b16 %rs84, %rs83, %rs81; + setp.lt.s64 %p95, %rd312, %rd202; + selp.b16 %rs85, 1, 0, %p95; + setp.lt.s64 %p96, %rd311, %rd202; + selp.b16 %rs86, -1, 0, %p96; + shl.b16 %rs87, %rs86, 1; + or.b16 %rs88, %rs85, %rs87; + and.b16 %rs89, %rs88, 3; + or.b16 %rs90, %rs89, %rs84; + shl.b16 %rs91, %rs90, 4; + setp.lt.s64 %p97, %rd310, %rd202; + selp.b16 %rs92, -1, 0, %p97; + shl.b16 %rs93, %rs92, 3; + setp.lt.s64 %p98, %rd309, %rd202; + selp.b16 %rs94, 1, 0, %p98; + shl.b16 %rs95, %rs94, 2; + or.b16 %rs96, %rs93, %rs95; + setp.lt.s64 %p38, %rd308, %rd202; + selp.b16 %rs97, 1, 0, %p38; + setp.lt.s64 %p99, %rd307, %rd202; + selp.b16 %rs98, -1, 0, %p99; + shl.b16 %rs99, %rs98, 1; + or.b16 %rs100, %rs97, %rs99; + and.b16 %rs101, %rs100, 3; + or.b16 %rs102, %rs101, %rs96; + and.b16 %rs103, %rs102, 15; + or.b16 %rs104, %rs103, %rs91; + and.b16 %rs105, %rs104, 255; + or.b16 %rs106, %rs105, %rs79; + cvt.u32.u16 %r848, %rs106; + or.b32 %r849, %r848, %r847; + cvt.u64.u32 %rd339, %r849; + setp.lt.s64 %p100, %rd306, %rd202; + selp.b16 %rs107, 1, 0, %p100; + shl.b16 %rs108, %rs107, 2; + setp.lt.s64 %p101, %rd305, %rd202; + selp.b16 %rs109, -1, 0, %p101; + shl.b16 %rs110, %rs109, 3; + or.b16 %rs111, %rs110, %rs108; + setp.lt.s64 %p102, %rd304, %rd202; + selp.b16 %rs112, 1, 0, %p102; + setp.lt.s64 %p103, %rd303, %rd202; + selp.b16 %rs113, -1, 0, %p103; + shl.b16 %rs114, %rs113, 1; + or.b16 %rs115, %rs112, %rs114; + and.b16 %rs116, %rs115, 3; + or.b16 %rs117, %rs116, %rs111; + shl.b16 %rs118, %rs117, 12; + setp.lt.s64 %p104, %rd302, %rd202; + selp.b16 %rs119, 1, 0, %p104; + shl.b16 %rs120, %rs119, 2; + setp.lt.s64 %p105, %rd301, %rd202; + selp.b16 %rs121, -1, 0, %p105; + shl.b16 %rs122, %rs121, 3; + or.b16 %rs123, %rs122, %rs120; + setp.lt.s64 %p106, %rd300, %rd202; + selp.b16 %rs124, 1, 0, %p106; + setp.lt.s64 %p107, %rd299, %rd202; + selp.b16 %rs125, -1, 0, %p107; + shl.b16 %rs126, %rs125, 1; + or.b16 %rs127, %rs124, %rs126; + and.b16 %rs128, %rs127, 3; + or.b16 %rs129, %rs128, %rs123; + and.b16 %rs130, %rs129, 15; + shl.b16 %rs131, %rs130, 8; + or.b16 %rs132, %rs118, %rs131; + setp.lt.s64 %p108, %rd298, %rd202; + selp.b16 %rs133, 1, 0, %p108; + shl.b16 %rs134, %rs133, 2; + setp.lt.s64 %p109, %rd297, %rd202; + selp.b16 %rs135, -1, 0, %p109; + shl.b16 %rs136, %rs135, 3; + or.b16 %rs137, %rs136, %rs134; + setp.lt.s64 %p110, %rd296, %rd202; + selp.b16 %rs138, 1, 0, %p110; + setp.lt.s64 %p111, %rd295, %rd202; + selp.b16 %rs139, -1, 0, %p111; + shl.b16 %rs140, %rs139, 1; + or.b16 %rs141, %rs138, %rs140; + and.b16 %rs142, %rs141, 3; + or.b16 %rs143, %rs142, %rs137; + shl.b16 %rs144, %rs143, 4; + setp.lt.s64 %p112, %rd294, %rd202; + selp.b16 %rs145, 1, 0, %p112; + shl.b16 %rs146, %rs145, 2; + setp.lt.s64 %p113, %rd293, %rd202; + selp.b16 %rs147, -1, 0, %p113; + shl.b16 %rs148, %rs147, 3; + or.b16 %rs149, %rs148, %rs146; + setp.lt.s64 %p34, %rd292, %rd202; + selp.b16 %rs150, 1, 0, %p34; + setp.lt.s64 %p114, %rd291, %rd202; + selp.b16 %rs151, -1, 0, %p114; + shl.b16 %rs152, %rs151, 1; + or.b16 %rs153, %rs150, %rs152; + and.b16 %rs154, %rs153, 3; + or.b16 %rs155, %rs154, %rs149; + and.b16 %rs156, %rs155, 15; + or.b16 %rs157, %rs156, %rs144; + and.b16 %rs158, %rs157, 255; + or.b16 %rs159, %rs158, %rs132; + cvt.u32.u16 %r850, %rs159; + shr.u32 %r851, %r850, 1; + shr.u32 %r852, %r850, 2; + shr.u32 %r853, %r850, 3; + shr.u32 %r854, %r850, 4; + shr.u32 %r855, %r850, 5; + shr.u32 %r856, %r850, 6; + shr.u32 %r857, %r850, 7; + shr.u32 %r858, %r850, 8; + shr.u32 %r859, %r850, 9; + shr.u32 %r860, %r850, 10; + shr.u32 %r861, %r850, 11; + shr.u32 %r862, %r850, 12; + shr.u32 %r863, %r850, 13; + shr.u32 %r864, %r850, 14; + shr.u32 %r865, %r850, 15; + setp.lt.s64 %p40, %rd290, %rd202; + setp.lt.s64 %p18, %rd289, %rd202; + setp.lt.s64 %p17, %rd288, %rd202; + setp.lt.s64 %p16, %rd287, %rd202; + setp.lt.s64 %p15, %rd286, %rd202; + setp.lt.s64 %p14, %rd285, %rd202; + setp.lt.s64 %p13, %rd284, %rd202; + setp.lt.s64 %p12, %rd283, %rd202; + setp.lt.s64 %p39, %rd282, %rd202; + setp.lt.s64 %p11, %rd281, %rd202; + setp.lt.s64 %p10, %rd280, %rd202; + setp.lt.s64 %p9, %rd279, %rd202; + setp.lt.s64 %p8, %rd278, %rd202; + setp.lt.s64 %p7, %rd277, %rd202; + setp.lt.s64 %p6, %rd276, %rd202; + setp.lt.s64 %p5, %rd275, %rd202; + .loc 1 98 30 // l2norm.py:98:30 + and.b32 %r866, %r775, 15; + shl.b32 %r867, %r866, 3; + .loc 1 99 14 // l2norm.py:99:14 + shl.b64 %rd340, %rd207, 8; + add.s64 %rd341, %rd197, %rd340; + shl.b64 %rd342, %rd208, 8; + add.s64 %rd343, %rd197, %rd342; + shl.b64 %rd344, %rd209, 8; + add.s64 %rd345, %rd197, %rd344; + shl.b64 %rd346, %rd210, 8; + add.s64 %rd347, %rd197, %rd346; + shl.b64 %rd348, %rd275, 8; + add.s64 %rd349, %rd197, %rd348; + shl.b64 %rd350, %rd276, 8; + add.s64 %rd351, %rd197, %rd350; + shl.b64 %rd352, %rd277, 8; + add.s64 %rd353, %rd197, %rd352; + shl.b64 %rd354, %rd278, 8; + add.s64 %rd355, %rd197, %rd354; + shl.b64 %rd356, %rd279, 8; + add.s64 %rd357, %rd197, %rd356; + shl.b64 %rd358, %rd280, 8; + add.s64 %rd359, %rd197, %rd358; + shl.b64 %rd360, %rd281, 8; + add.s64 %rd361, %rd197, %rd360; + shl.b64 %rd362, %rd282, 8; + add.s64 %rd363, %rd197, %rd362; + shl.b64 %rd364, %rd283, 8; + add.s64 %rd365, %rd197, %rd364; + shl.b64 %rd366, %rd284, 8; + add.s64 %rd367, %rd197, %rd366; + shl.b64 %rd368, %rd285, 8; + add.s64 %rd369, %rd197, %rd368; + shl.b64 %rd370, %rd286, 8; + add.s64 %rd371, %rd197, %rd370; + shl.b64 %rd372, %rd287, 8; + add.s64 %rd373, %rd197, %rd372; + shl.b64 %rd374, %rd288, 8; + add.s64 %rd375, %rd197, %rd374; + shl.b64 %rd376, %rd289, 8; + add.s64 %rd377, %rd197, %rd376; + shl.b64 %rd378, %rd290, 8; + add.s64 %rd379, %rd197, %rd378; + shl.b64 %rd380, %rd305, 8; + add.s64 %rd381, %rd197, %rd380; + shl.b64 %rd382, %rd306, 8; + add.s64 %rd383, %rd197, %rd382; + shl.b64 %rd384, %rd303, 8; + add.s64 %rd385, %rd197, %rd384; + shl.b64 %rd386, %rd304, 8; + add.s64 %rd387, %rd197, %rd386; + shl.b64 %rd388, %rd301, 8; + add.s64 %rd389, %rd197, %rd388; + shl.b64 %rd390, %rd302, 8; + add.s64 %rd391, %rd197, %rd390; + shl.b64 %rd392, %rd299, 8; + add.s64 %rd393, %rd197, %rd392; + shl.b64 %rd394, %rd300, 8; + add.s64 %rd395, %rd197, %rd394; + shl.b64 %rd396, %rd297, 8; + add.s64 %rd397, %rd197, %rd396; + shl.b64 %rd398, %rd298, 8; + add.s64 %rd399, %rd197, %rd398; + shl.b64 %rd400, %rd295, 8; + add.s64 %rd401, %rd197, %rd400; + shl.b64 %rd402, %rd296, 8; + add.s64 %rd403, %rd197, %rd402; + shl.b64 %rd404, %rd293, 8; + add.s64 %rd405, %rd197, %rd404; + shl.b64 %rd406, %rd294, 8; + add.s64 %rd407, %rd197, %rd406; + shl.b64 %rd408, %rd291, 8; + add.s64 %rd409, %rd197, %rd408; + shl.b64 %rd410, %rd292, 8; + add.s64 %rd411, %rd197, %rd410; + shl.b64 %rd412, %rd337, 8; + add.s64 %rd413, %rd197, %rd412; + shl.b64 %rd414, %rd338, 8; + add.s64 %rd415, %rd197, %rd414; + shl.b64 %rd416, %rd335, 8; + add.s64 %rd417, %rd197, %rd416; + shl.b64 %rd418, %rd336, 8; + add.s64 %rd419, %rd197, %rd418; + shl.b64 %rd420, %rd333, 8; + add.s64 %rd421, %rd197, %rd420; + shl.b64 %rd422, %rd334, 8; + add.s64 %rd423, %rd197, %rd422; + shl.b64 %rd424, %rd331, 8; + add.s64 %rd425, %rd197, %rd424; + shl.b64 %rd426, %rd332, 8; + add.s64 %rd427, %rd197, %rd426; + shl.b64 %rd428, %rd329, 8; + add.s64 %rd429, %rd197, %rd428; + shl.b64 %rd430, %rd330, 8; + add.s64 %rd431, %rd197, %rd430; + shl.b64 %rd432, %rd327, 8; + add.s64 %rd433, %rd197, %rd432; + shl.b64 %rd434, %rd328, 8; + add.s64 %rd435, %rd197, %rd434; + shl.b64 %rd436, %rd325, 8; + add.s64 %rd437, %rd197, %rd436; + shl.b64 %rd438, %rd326, 8; + add.s64 %rd439, %rd197, %rd438; + shl.b64 %rd440, %rd323, 8; + add.s64 %rd441, %rd197, %rd440; + shl.b64 %rd442, %rd324, 8; + add.s64 %rd443, %rd197, %rd442; + shl.b64 %rd444, %rd321, 8; + add.s64 %rd445, %rd197, %rd444; + shl.b64 %rd446, %rd322, 8; + add.s64 %rd447, %rd197, %rd446; + shl.b64 %rd448, %rd319, 8; + add.s64 %rd449, %rd197, %rd448; + shl.b64 %rd450, %rd320, 8; + add.s64 %rd451, %rd197, %rd450; + shl.b64 %rd452, %rd317, 8; + add.s64 %rd453, %rd197, %rd452; + shl.b64 %rd454, %rd318, 8; + add.s64 %rd455, %rd197, %rd454; + shl.b64 %rd456, %rd315, 8; + add.s64 %rd457, %rd197, %rd456; + shl.b64 %rd458, %rd316, 8; + add.s64 %rd459, %rd197, %rd458; + shl.b64 %rd460, %rd313, 8; + add.s64 %rd461, %rd197, %rd460; + shl.b64 %rd462, %rd314, 8; + add.s64 %rd463, %rd197, %rd462; + shl.b64 %rd464, %rd311, 8; + add.s64 %rd465, %rd197, %rd464; + shl.b64 %rd466, %rd312, 8; + add.s64 %rd467, %rd197, %rd466; + .loc 1 99 33 // l2norm.py:99:33 + mul.wide.u32 %rd468, %r867, 2; + add.s64 %rd1, %rd341, %rd468; + add.s64 %rd2, %rd343, %rd468; + add.s64 %rd3, %rd345, %rd468; + add.s64 %rd4, %rd347, %rd468; + add.s64 %rd5, %rd349, %rd468; + add.s64 %rd6, %rd351, %rd468; + add.s64 %rd7, %rd353, %rd468; + add.s64 %rd8, %rd355, %rd468; + add.s64 %rd9, %rd357, %rd468; + add.s64 %rd10, %rd359, %rd468; + add.s64 %rd11, %rd361, %rd468; + add.s64 %rd12, %rd363, %rd468; + add.s64 %rd13, %rd365, %rd468; + add.s64 %rd14, %rd367, %rd468; + add.s64 %rd15, %rd369, %rd468; + add.s64 %rd16, %rd371, %rd468; + add.s64 %rd17, %rd373, %rd468; + add.s64 %rd18, %rd375, %rd468; + add.s64 %rd19, %rd377, %rd468; + add.s64 %rd20, %rd379, %rd468; + add.s64 %rd21, %rd381, %rd468; + add.s64 %rd22, %rd383, %rd468; + add.s64 %rd23, %rd385, %rd468; + add.s64 %rd24, %rd387, %rd468; + add.s64 %rd25, %rd389, %rd468; + add.s64 %rd26, %rd391, %rd468; + add.s64 %rd27, %rd393, %rd468; + add.s64 %rd28, %rd395, %rd468; + add.s64 %rd29, %rd397, %rd468; + add.s64 %rd30, %rd399, %rd468; + add.s64 %rd31, %rd401, %rd468; + add.s64 %rd32, %rd403, %rd468; + add.s64 %rd33, %rd405, %rd468; + add.s64 %rd34, %rd407, %rd468; + add.s64 %rd35, %rd409, %rd468; + add.s64 %rd36, %rd411, %rd468; + add.s64 %rd37, %rd413, %rd468; + add.s64 %rd38, %rd415, %rd468; + add.s64 %rd39, %rd417, %rd468; + add.s64 %rd40, %rd419, %rd468; + add.s64 %rd41, %rd421, %rd468; + add.s64 %rd42, %rd423, %rd468; + add.s64 %rd43, %rd425, %rd468; + add.s64 %rd44, %rd427, %rd468; + add.s64 %rd45, %rd429, %rd468; + add.s64 %rd46, %rd431, %rd468; + add.s64 %rd47, %rd433, %rd468; + add.s64 %rd48, %rd435, %rd468; + add.s64 %rd49, %rd437, %rd468; + add.s64 %rd50, %rd439, %rd468; + add.s64 %rd51, %rd441, %rd468; + add.s64 %rd52, %rd443, %rd468; + add.s64 %rd53, %rd445, %rd468; + add.s64 %rd54, %rd447, %rd468; + add.s64 %rd55, %rd449, %rd468; + add.s64 %rd56, %rd451, %rd468; + add.s64 %rd57, %rd453, %rd468; + add.s64 %rd58, %rd455, %rd468; + add.s64 %rd59, %rd457, %rd468; + add.s64 %rd60, %rd459, %rd468; + add.s64 %rd61, %rd461, %rd468; + add.s64 %rd62, %rd463, %rd468; + add.s64 %rd63, %rd465, %rd468; + add.s64 %rd64, %rd467, %rd468; + .loc 1 100 20 // l2norm.py:100:20 + shl.b64 %rd469, %rd310, 2; + add.s64 %rd65, %rd198, %rd469; + shl.b64 %rd470, %rd309, 2; + add.s64 %rd66, %rd198, %rd470; + shl.b64 %rd471, %rd307, 2; + add.s64 %rd67, %rd198, %rd471; + shl.b64 %rd472, %rd308, 2; + add.s64 %rd68, %rd198, %rd472; + .loc 1 101 16 // l2norm.py:101:16 + add.s64 %rd473, %rd199, %rd340; + add.s64 %rd474, %rd199, %rd342; + add.s64 %rd475, %rd199, %rd344; + add.s64 %rd476, %rd199, %rd346; + add.s64 %rd477, %rd199, %rd348; + add.s64 %rd478, %rd199, %rd350; + add.s64 %rd479, %rd199, %rd352; + add.s64 %rd480, %rd199, %rd354; + add.s64 %rd481, %rd199, %rd356; + add.s64 %rd482, %rd199, %rd358; + add.s64 %rd483, %rd199, %rd360; + add.s64 %rd484, %rd199, %rd362; + add.s64 %rd485, %rd199, %rd364; + add.s64 %rd486, %rd199, %rd366; + add.s64 %rd487, %rd199, %rd368; + add.s64 %rd488, %rd199, %rd370; + add.s64 %rd489, %rd199, %rd372; + add.s64 %rd490, %rd199, %rd374; + add.s64 %rd491, %rd199, %rd376; + add.s64 %rd492, %rd199, %rd378; + add.s64 %rd493, %rd199, %rd380; + add.s64 %rd494, %rd199, %rd382; + add.s64 %rd495, %rd199, %rd384; + add.s64 %rd496, %rd199, %rd386; + add.s64 %rd497, %rd199, %rd388; + add.s64 %rd498, %rd199, %rd390; + add.s64 %rd499, %rd199, %rd392; + add.s64 %rd500, %rd199, %rd394; + add.s64 %rd501, %rd199, %rd396; + add.s64 %rd502, %rd199, %rd398; + add.s64 %rd503, %rd199, %rd400; + add.s64 %rd504, %rd199, %rd402; + add.s64 %rd505, %rd199, %rd404; + add.s64 %rd506, %rd199, %rd406; + add.s64 %rd507, %rd199, %rd408; + add.s64 %rd508, %rd199, %rd410; + add.s64 %rd509, %rd199, %rd412; + add.s64 %rd510, %rd199, %rd414; + add.s64 %rd511, %rd199, %rd416; + add.s64 %rd512, %rd199, %rd418; + add.s64 %rd513, %rd199, %rd420; + add.s64 %rd514, %rd199, %rd422; + add.s64 %rd515, %rd199, %rd424; + add.s64 %rd516, %rd199, %rd426; + add.s64 %rd517, %rd199, %rd428; + add.s64 %rd518, %rd199, %rd430; + add.s64 %rd519, %rd199, %rd432; + add.s64 %rd520, %rd199, %rd434; + add.s64 %rd521, %rd199, %rd436; + add.s64 %rd522, %rd199, %rd438; + add.s64 %rd523, %rd199, %rd440; + add.s64 %rd524, %rd199, %rd442; + add.s64 %rd525, %rd199, %rd444; + add.s64 %rd526, %rd199, %rd446; + add.s64 %rd527, %rd199, %rd448; + add.s64 %rd528, %rd199, %rd450; + add.s64 %rd529, %rd199, %rd452; + add.s64 %rd530, %rd199, %rd454; + add.s64 %rd531, %rd199, %rd456; + add.s64 %rd532, %rd199, %rd458; + add.s64 %rd533, %rd199, %rd460; + add.s64 %rd534, %rd199, %rd462; + add.s64 %rd535, %rd199, %rd464; + add.s64 %rd536, %rd199, %rd466; + .loc 1 101 35 // l2norm.py:101:35 + add.s64 %rd69, %rd473, %rd468; + add.s64 %rd70, %rd474, %rd468; + add.s64 %rd71, %rd475, %rd468; + add.s64 %rd72, %rd476, %rd468; + add.s64 %rd73, %rd477, %rd468; + add.s64 %rd74, %rd478, %rd468; + add.s64 %rd75, %rd479, %rd468; + add.s64 %rd76, %rd480, %rd468; + add.s64 %rd77, %rd481, %rd468; + add.s64 %rd78, %rd482, %rd468; + add.s64 %rd79, %rd483, %rd468; + add.s64 %rd80, %rd484, %rd468; + add.s64 %rd81, %rd485, %rd468; + add.s64 %rd82, %rd486, %rd468; + add.s64 %rd83, %rd487, %rd468; + add.s64 %rd84, %rd488, %rd468; + add.s64 %rd85, %rd489, %rd468; + add.s64 %rd86, %rd490, %rd468; + add.s64 %rd87, %rd491, %rd468; + add.s64 %rd88, %rd492, %rd468; + add.s64 %rd89, %rd493, %rd468; + add.s64 %rd90, %rd494, %rd468; + add.s64 %rd91, %rd495, %rd468; + add.s64 %rd92, %rd496, %rd468; + add.s64 %rd93, %rd497, %rd468; + add.s64 %rd94, %rd498, %rd468; + add.s64 %rd95, %rd499, %rd468; + add.s64 %rd96, %rd500, %rd468; + add.s64 %rd97, %rd501, %rd468; + add.s64 %rd98, %rd502, %rd468; + add.s64 %rd99, %rd503, %rd468; + add.s64 %rd100, %rd504, %rd468; + add.s64 %rd101, %rd505, %rd468; + add.s64 %rd102, %rd506, %rd468; + add.s64 %rd103, %rd507, %rd468; + add.s64 %rd104, %rd508, %rd468; + add.s64 %rd105, %rd509, %rd468; + add.s64 %rd106, %rd510, %rd468; + add.s64 %rd107, %rd511, %rd468; + add.s64 %rd108, %rd512, %rd468; + add.s64 %rd109, %rd513, %rd468; + add.s64 %rd110, %rd514, %rd468; + add.s64 %rd111, %rd515, %rd468; + add.s64 %rd112, %rd516, %rd468; + add.s64 %rd113, %rd517, %rd468; + add.s64 %rd114, %rd518, %rd468; + add.s64 %rd115, %rd519, %rd468; + add.s64 %rd116, %rd520, %rd468; + add.s64 %rd117, %rd521, %rd468; + add.s64 %rd118, %rd522, %rd468; + add.s64 %rd119, %rd523, %rd468; + add.s64 %rd120, %rd524, %rd468; + add.s64 %rd121, %rd525, %rd468; + add.s64 %rd122, %rd526, %rd468; + add.s64 %rd123, %rd527, %rd468; + add.s64 %rd124, %rd528, %rd468; + add.s64 %rd125, %rd529, %rd468; + add.s64 %rd126, %rd530, %rd468; + add.s64 %rd127, %rd531, %rd468; + add.s64 %rd128, %rd532, %rd468; + add.s64 %rd129, %rd533, %rd468; + add.s64 %rd130, %rd534, %rd468; + add.s64 %rd131, %rd535, %rd468; + add.s64 %rd132, %rd536, %rd468; + .loc 1 102 16 // l2norm.py:102:16 + add.s64 %rd537, %rd201, %rd340; + add.s64 %rd538, %rd201, %rd342; + add.s64 %rd539, %rd201, %rd344; + add.s64 %rd540, %rd201, %rd346; + add.s64 %rd541, %rd201, %rd348; + add.s64 %rd542, %rd201, %rd350; + add.s64 %rd543, %rd201, %rd352; + add.s64 %rd544, %rd201, %rd354; + add.s64 %rd545, %rd201, %rd356; + add.s64 %rd546, %rd201, %rd358; + add.s64 %rd547, %rd201, %rd360; + add.s64 %rd548, %rd201, %rd362; + add.s64 %rd549, %rd201, %rd364; + add.s64 %rd550, %rd201, %rd366; + add.s64 %rd551, %rd201, %rd368; + add.s64 %rd552, %rd201, %rd370; + add.s64 %rd553, %rd201, %rd372; + add.s64 %rd554, %rd201, %rd374; + add.s64 %rd555, %rd201, %rd376; + add.s64 %rd556, %rd201, %rd378; + add.s64 %rd557, %rd201, %rd380; + add.s64 %rd558, %rd201, %rd382; + add.s64 %rd559, %rd201, %rd384; + add.s64 %rd560, %rd201, %rd386; + add.s64 %rd561, %rd201, %rd388; + add.s64 %rd562, %rd201, %rd390; + add.s64 %rd563, %rd201, %rd392; + add.s64 %rd564, %rd201, %rd394; + add.s64 %rd565, %rd201, %rd396; + add.s64 %rd566, %rd201, %rd398; + add.s64 %rd567, %rd201, %rd400; + add.s64 %rd568, %rd201, %rd402; + add.s64 %rd569, %rd201, %rd404; + add.s64 %rd570, %rd201, %rd406; + add.s64 %rd571, %rd201, %rd408; + add.s64 %rd572, %rd201, %rd410; + add.s64 %rd573, %rd201, %rd412; + add.s64 %rd574, %rd201, %rd414; + add.s64 %rd575, %rd201, %rd416; + add.s64 %rd576, %rd201, %rd418; + add.s64 %rd577, %rd201, %rd420; + add.s64 %rd578, %rd201, %rd422; + add.s64 %rd579, %rd201, %rd424; + add.s64 %rd580, %rd201, %rd426; + add.s64 %rd581, %rd201, %rd428; + add.s64 %rd582, %rd201, %rd430; + add.s64 %rd583, %rd201, %rd432; + add.s64 %rd584, %rd201, %rd434; + add.s64 %rd585, %rd201, %rd436; + add.s64 %rd586, %rd201, %rd438; + add.s64 %rd587, %rd201, %rd440; + add.s64 %rd588, %rd201, %rd442; + add.s64 %rd589, %rd201, %rd444; + add.s64 %rd590, %rd201, %rd446; + add.s64 %rd591, %rd201, %rd448; + add.s64 %rd592, %rd201, %rd450; + add.s64 %rd593, %rd201, %rd452; + add.s64 %rd594, %rd201, %rd454; + add.s64 %rd595, %rd201, %rd456; + add.s64 %rd596, %rd201, %rd458; + add.s64 %rd597, %rd201, %rd460; + add.s64 %rd598, %rd201, %rd462; + add.s64 %rd599, %rd201, %rd464; + add.s64 %rd600, %rd201, %rd466; + .loc 1 102 35 // l2norm.py:102:35 + add.s64 %rd133, %rd537, %rd468; + add.s64 %rd134, %rd538, %rd468; + add.s64 %rd135, %rd539, %rd468; + add.s64 %rd136, %rd540, %rd468; + add.s64 %rd137, %rd541, %rd468; + add.s64 %rd138, %rd542, %rd468; + add.s64 %rd139, %rd543, %rd468; + add.s64 %rd140, %rd544, %rd468; + add.s64 %rd141, %rd545, %rd468; + add.s64 %rd142, %rd546, %rd468; + add.s64 %rd143, %rd547, %rd468; + add.s64 %rd144, %rd548, %rd468; + add.s64 %rd145, %rd549, %rd468; + add.s64 %rd146, %rd550, %rd468; + add.s64 %rd147, %rd551, %rd468; + add.s64 %rd148, %rd552, %rd468; + add.s64 %rd149, %rd553, %rd468; + add.s64 %rd150, %rd554, %rd468; + add.s64 %rd151, %rd555, %rd468; + add.s64 %rd152, %rd556, %rd468; + add.s64 %rd153, %rd557, %rd468; + add.s64 %rd154, %rd558, %rd468; + add.s64 %rd155, %rd559, %rd468; + add.s64 %rd156, %rd560, %rd468; + add.s64 %rd157, %rd561, %rd468; + add.s64 %rd158, %rd562, %rd468; + add.s64 %rd159, %rd563, %rd468; + add.s64 %rd160, %rd564, %rd468; + add.s64 %rd161, %rd565, %rd468; + add.s64 %rd162, %rd566, %rd468; + add.s64 %rd163, %rd567, %rd468; + add.s64 %rd164, %rd568, %rd468; + add.s64 %rd165, %rd569, %rd468; + add.s64 %rd166, %rd570, %rd468; + add.s64 %rd167, %rd571, %rd468; + add.s64 %rd168, %rd572, %rd468; + add.s64 %rd169, %rd573, %rd468; + add.s64 %rd170, %rd574, %rd468; + add.s64 %rd171, %rd575, %rd468; + add.s64 %rd172, %rd576, %rd468; + add.s64 %rd173, %rd577, %rd468; + add.s64 %rd174, %rd578, %rd468; + add.s64 %rd175, %rd579, %rd468; + add.s64 %rd176, %rd580, %rd468; + add.s64 %rd177, %rd581, %rd468; + add.s64 %rd178, %rd582, %rd468; + add.s64 %rd179, %rd583, %rd468; + add.s64 %rd180, %rd584, %rd468; + add.s64 %rd181, %rd585, %rd468; + add.s64 %rd182, %rd586, %rd468; + add.s64 %rd183, %rd587, %rd468; + add.s64 %rd184, %rd588, %rd468; + add.s64 %rd185, %rd589, %rd468; + add.s64 %rd186, %rd590, %rd468; + add.s64 %rd187, %rd591, %rd468; + add.s64 %rd188, %rd592, %rd468; + add.s64 %rd189, %rd593, %rd468; + add.s64 %rd190, %rd594, %rd468; + add.s64 %rd191, %rd595, %rd468; + add.s64 %rd192, %rd596, %rd468; + add.s64 %rd193, %rd597, %rd468; + add.s64 %rd194, %rd598, %rd468; + add.s64 %rd195, %rd599, %rd468; + add.s64 %rd196, %rd600, %rd468; + mov.b32 %r5, 0; + .loc 1 104 18 // l2norm.py:104:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs160, %rs161}, %r1; + mov.b32 {%rs162, %rs163}, %r2; + mov.b32 {%rs164, %rs165}, %r3; + mov.b32 {%rs166, %rs167}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs168, %rs169}, %r6; + mov.b32 {%rs170, %rs171}, %r7; + mov.b32 {%rs172, %rs173}, %r8; + mov.b32 {%rs174, %rs175}, %r9; + // begin inline asm + mov.u32 %r10, %r5; + mov.u32 %r11, %r5; + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + @%p3 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs176, %rs177}, %r10; + mov.b32 {%rs178, %rs179}, %r11; + mov.b32 {%rs180, %rs181}, %r12; + mov.b32 {%rs182, %rs183}, %r13; + // begin inline asm + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + @%p4 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ]; + // end inline asm + mov.b32 {%rs184, %rs185}, %r14; + mov.b32 {%rs186, %rs187}, %r15; + mov.b32 {%rs188, %rs189}, %r16; + mov.b32 {%rs190, %rs191}, %r17; + // begin inline asm + mov.u32 %r18, %r5; + mov.u32 %r19, %r5; + mov.u32 %r20, %r5; + mov.u32 %r21, %r5; + @%p5 ld.global.v4.b32 { %r18, %r19, %r20, %r21 }, [ %rd5 + 0 ]; + // end inline asm + mov.b32 {%rs192, %rs193}, %r18; + mov.b32 {%rs194, %rs195}, %r19; + mov.b32 {%rs196, %rs197}, %r20; + mov.b32 {%rs198, %rs199}, %r21; + // begin inline asm + mov.u32 %r22, %r5; + mov.u32 %r23, %r5; + mov.u32 %r24, %r5; + mov.u32 %r25, %r5; + @%p6 ld.global.v4.b32 { %r22, %r23, %r24, %r25 }, [ %rd6 + 0 ]; + // end inline asm + mov.b32 {%rs200, %rs201}, %r22; + mov.b32 {%rs202, %rs203}, %r23; + mov.b32 {%rs204, %rs205}, %r24; + mov.b32 {%rs206, %rs207}, %r25; + // begin inline asm + mov.u32 %r26, %r5; + mov.u32 %r27, %r5; + mov.u32 %r28, %r5; + mov.u32 %r29, %r5; + @%p7 ld.global.v4.b32 { %r26, %r27, %r28, %r29 }, [ %rd7 + 0 ]; + // end inline asm + mov.b32 {%rs208, %rs209}, %r26; + mov.b32 {%rs210, %rs211}, %r27; + mov.b32 {%rs212, %rs213}, %r28; + mov.b32 {%rs214, %rs215}, %r29; + // begin inline asm + mov.u32 %r30, %r5; + mov.u32 %r31, %r5; + mov.u32 %r32, %r5; + mov.u32 %r33, %r5; + @%p8 ld.global.v4.b32 { %r30, %r31, %r32, %r33 }, [ %rd8 + 0 ]; + // end inline asm + mov.b32 {%rs216, %rs217}, %r30; + mov.b32 {%rs218, %rs219}, %r31; + mov.b32 {%rs220, %rs221}, %r32; + mov.b32 {%rs222, %rs223}, %r33; + // begin inline asm + mov.u32 %r34, %r5; + mov.u32 %r35, %r5; + mov.u32 %r36, %r5; + mov.u32 %r37, %r5; + @%p9 ld.global.v4.b32 { %r34, %r35, %r36, %r37 }, [ %rd9 + 0 ]; + // end inline asm + mov.b32 {%rs224, %rs225}, %r34; + mov.b32 {%rs226, %rs227}, %r35; + mov.b32 {%rs228, %rs229}, %r36; + mov.b32 {%rs230, %rs231}, %r37; + // begin inline asm + mov.u32 %r38, %r5; + mov.u32 %r39, %r5; + mov.u32 %r40, %r5; + mov.u32 %r41, %r5; + @%p10 ld.global.v4.b32 { %r38, %r39, %r40, %r41 }, [ %rd10 + 0 ]; + // end inline asm + mov.b32 {%rs232, %rs233}, %r38; + mov.b32 {%rs234, %rs235}, %r39; + mov.b32 {%rs236, %rs237}, %r40; + mov.b32 {%rs238, %rs239}, %r41; + // begin inline asm + mov.u32 %r42, %r5; + mov.u32 %r43, %r5; + mov.u32 %r44, %r5; + mov.u32 %r45, %r5; + @%p11 ld.global.v4.b32 { %r42, %r43, %r44, %r45 }, [ %rd11 + 0 ]; + // end inline asm + mov.b32 {%rs240, %rs241}, %r42; + mov.b32 {%rs242, %rs243}, %r43; + mov.b32 {%rs244, %rs245}, %r44; + mov.b32 {%rs246, %rs247}, %r45; + // begin inline asm + mov.u32 %r46, %r5; + mov.u32 %r47, %r5; + mov.u32 %r48, %r5; + mov.u32 %r49, %r5; + @%p39 ld.global.v4.b32 { %r46, %r47, %r48, %r49 }, [ %rd12 + 0 ]; + // end inline asm + mov.b32 {%rs248, %rs249}, %r46; + mov.b32 {%rs250, %rs251}, %r47; + mov.b32 {%rs252, %rs253}, %r48; + mov.b32 {%rs254, %rs255}, %r49; + // begin inline asm + mov.u32 %r50, %r5; + mov.u32 %r51, %r5; + mov.u32 %r52, %r5; + mov.u32 %r53, %r5; + @%p12 ld.global.v4.b32 { %r50, %r51, %r52, %r53 }, [ %rd13 + 0 ]; + // end inline asm + mov.b32 {%rs256, %rs257}, %r50; + mov.b32 {%rs258, %rs259}, %r51; + mov.b32 {%rs260, %rs261}, %r52; + mov.b32 {%rs262, %rs263}, %r53; + // begin inline asm + mov.u32 %r54, %r5; + mov.u32 %r55, %r5; + mov.u32 %r56, %r5; + mov.u32 %r57, %r5; + @%p13 ld.global.v4.b32 { %r54, %r55, %r56, %r57 }, [ %rd14 + 0 ]; + // end inline asm + mov.b32 {%rs264, %rs265}, %r54; + mov.b32 {%rs266, %rs267}, %r55; + mov.b32 {%rs268, %rs269}, %r56; + mov.b32 {%rs270, %rs271}, %r57; + // begin inline asm + mov.u32 %r58, %r5; + mov.u32 %r59, %r5; + mov.u32 %r60, %r5; + mov.u32 %r61, %r5; + @%p14 ld.global.v4.b32 { %r58, %r59, %r60, %r61 }, [ %rd15 + 0 ]; + // end inline asm + mov.b32 {%rs272, %rs273}, %r58; + mov.b32 {%rs274, %rs275}, %r59; + mov.b32 {%rs276, %rs277}, %r60; + mov.b32 {%rs278, %rs279}, %r61; + // begin inline asm + mov.u32 %r62, %r5; + mov.u32 %r63, %r5; + mov.u32 %r64, %r5; + mov.u32 %r65, %r5; + @%p15 ld.global.v4.b32 { %r62, %r63, %r64, %r65 }, [ %rd16 + 0 ]; + // end inline asm + mov.b32 {%rs280, %rs281}, %r62; + mov.b32 {%rs282, %rs283}, %r63; + mov.b32 {%rs284, %rs285}, %r64; + mov.b32 {%rs286, %rs287}, %r65; + // begin inline asm + mov.u32 %r66, %r5; + mov.u32 %r67, %r5; + mov.u32 %r68, %r5; + mov.u32 %r69, %r5; + @%p16 ld.global.v4.b32 { %r66, %r67, %r68, %r69 }, [ %rd17 + 0 ]; + // end inline asm + mov.b32 {%rs288, %rs289}, %r66; + mov.b32 {%rs290, %rs291}, %r67; + mov.b32 {%rs292, %rs293}, %r68; + mov.b32 {%rs294, %rs295}, %r69; + // begin inline asm + mov.u32 %r70, %r5; + mov.u32 %r71, %r5; + mov.u32 %r72, %r5; + mov.u32 %r73, %r5; + @%p17 ld.global.v4.b32 { %r70, %r71, %r72, %r73 }, [ %rd18 + 0 ]; + // end inline asm + mov.b32 {%rs296, %rs297}, %r70; + mov.b32 {%rs298, %rs299}, %r71; + mov.b32 {%rs300, %rs301}, %r72; + mov.b32 {%rs302, %rs303}, %r73; + // begin inline asm + mov.u32 %r74, %r5; + mov.u32 %r75, %r5; + mov.u32 %r76, %r5; + mov.u32 %r77, %r5; + @%p18 ld.global.v4.b32 { %r74, %r75, %r76, %r77 }, [ %rd19 + 0 ]; + // end inline asm + mov.b32 {%rs304, %rs305}, %r74; + mov.b32 {%rs306, %rs307}, %r75; + mov.b32 {%rs308, %rs309}, %r76; + mov.b32 {%rs310, %rs311}, %r77; + // begin inline asm + mov.u32 %r78, %r5; + mov.u32 %r79, %r5; + mov.u32 %r80, %r5; + mov.u32 %r81, %r5; + @%p40 ld.global.v4.b32 { %r78, %r79, %r80, %r81 }, [ %rd20 + 0 ]; + // end inline asm + mov.b32 {%rs312, %rs313}, %r78; + mov.b32 {%rs314, %rs315}, %r79; + mov.b32 {%rs316, %rs317}, %r80; + mov.b32 {%rs318, %rs319}, %r81; + and.b32 %r868, %r865, 1; + setp.ne.b32 %p19, %r868, 0; + // begin inline asm + mov.u32 %r82, %r5; + mov.u32 %r83, %r5; + mov.u32 %r84, %r5; + mov.u32 %r85, %r5; + @%p19 ld.global.v4.b32 { %r82, %r83, %r84, %r85 }, [ %rd21 + 0 ]; + // end inline asm + mov.b32 {%rs320, %rs321}, %r82; + mov.b32 {%rs322, %rs323}, %r83; + mov.b32 {%rs324, %rs325}, %r84; + mov.b32 {%rs326, %rs327}, %r85; + and.b32 %r869, %r864, 1; + setp.ne.b32 %p20, %r869, 0; + // begin inline asm + mov.u32 %r86, %r5; + mov.u32 %r87, %r5; + mov.u32 %r88, %r5; + mov.u32 %r89, %r5; + @%p20 ld.global.v4.b32 { %r86, %r87, %r88, %r89 }, [ %rd22 + 0 ]; + // end inline asm + mov.b32 {%rs328, %rs329}, %r86; + mov.b32 {%rs330, %rs331}, %r87; + mov.b32 {%rs332, %rs333}, %r88; + mov.b32 {%rs334, %rs335}, %r89; + and.b32 %r870, %r863, 1; + setp.ne.b32 %p21, %r870, 0; + // begin inline asm + mov.u32 %r90, %r5; + mov.u32 %r91, %r5; + mov.u32 %r92, %r5; + mov.u32 %r93, %r5; + @%p21 ld.global.v4.b32 { %r90, %r91, %r92, %r93 }, [ %rd23 + 0 ]; + // end inline asm + mov.b32 {%rs336, %rs337}, %r90; + mov.b32 {%rs338, %rs339}, %r91; + mov.b32 {%rs340, %rs341}, %r92; + mov.b32 {%rs342, %rs343}, %r93; + and.b32 %r871, %r862, 1; + setp.ne.b32 %p22, %r871, 0; + // begin inline asm + mov.u32 %r94, %r5; + mov.u32 %r95, %r5; + mov.u32 %r96, %r5; + mov.u32 %r97, %r5; + @%p22 ld.global.v4.b32 { %r94, %r95, %r96, %r97 }, [ %rd24 + 0 ]; + // end inline asm + mov.b32 {%rs344, %rs345}, %r94; + mov.b32 {%rs346, %rs347}, %r95; + mov.b32 {%rs348, %rs349}, %r96; + mov.b32 {%rs350, %rs351}, %r97; + and.b32 %r872, %r861, 1; + setp.ne.b32 %p23, %r872, 0; + // begin inline asm + mov.u32 %r98, %r5; + mov.u32 %r99, %r5; + mov.u32 %r100, %r5; + mov.u32 %r101, %r5; + @%p23 ld.global.v4.b32 { %r98, %r99, %r100, %r101 }, [ %rd25 + 0 ]; + // end inline asm + mov.b32 {%rs352, %rs353}, %r98; + mov.b32 {%rs354, %rs355}, %r99; + mov.b32 {%rs356, %rs357}, %r100; + mov.b32 {%rs358, %rs359}, %r101; + and.b32 %r873, %r860, 1; + setp.ne.b32 %p24, %r873, 0; + // begin inline asm + mov.u32 %r102, %r5; + mov.u32 %r103, %r5; + mov.u32 %r104, %r5; + mov.u32 %r105, %r5; + @%p24 ld.global.v4.b32 { %r102, %r103, %r104, %r105 }, [ %rd26 + 0 ]; + // end inline asm + mov.b32 {%rs360, %rs361}, %r102; + mov.b32 {%rs362, %rs363}, %r103; + mov.b32 {%rs364, %rs365}, %r104; + mov.b32 {%rs366, %rs367}, %r105; + and.b32 %r874, %r859, 1; + setp.ne.b32 %p25, %r874, 0; + // begin inline asm + mov.u32 %r106, %r5; + mov.u32 %r107, %r5; + mov.u32 %r108, %r5; + mov.u32 %r109, %r5; + @%p25 ld.global.v4.b32 { %r106, %r107, %r108, %r109 }, [ %rd27 + 0 ]; + // end inline asm + mov.b32 {%rs368, %rs369}, %r106; + mov.b32 {%rs370, %rs371}, %r107; + mov.b32 {%rs372, %rs373}, %r108; + mov.b32 {%rs374, %rs375}, %r109; + and.b32 %r875, %r858, 1; + setp.ne.b32 %p26, %r875, 0; + // begin inline asm + mov.u32 %r110, %r5; + mov.u32 %r111, %r5; + mov.u32 %r112, %r5; + mov.u32 %r113, %r5; + @%p26 ld.global.v4.b32 { %r110, %r111, %r112, %r113 }, [ %rd28 + 0 ]; + // end inline asm + mov.b32 {%rs376, %rs377}, %r110; + mov.b32 {%rs378, %rs379}, %r111; + mov.b32 {%rs380, %rs381}, %r112; + mov.b32 {%rs382, %rs383}, %r113; + and.b32 %r876, %r857, 1; + setp.ne.b32 %p27, %r876, 0; + // begin inline asm + mov.u32 %r114, %r5; + mov.u32 %r115, %r5; + mov.u32 %r116, %r5; + mov.u32 %r117, %r5; + @%p27 ld.global.v4.b32 { %r114, %r115, %r116, %r117 }, [ %rd29 + 0 ]; + // end inline asm + mov.b32 {%rs384, %rs385}, %r114; + mov.b32 {%rs386, %rs387}, %r115; + mov.b32 {%rs388, %rs389}, %r116; + mov.b32 {%rs390, %rs391}, %r117; + and.b32 %r877, %r856, 1; + setp.ne.b32 %p28, %r877, 0; + // begin inline asm + mov.u32 %r118, %r5; + mov.u32 %r119, %r5; + mov.u32 %r120, %r5; + mov.u32 %r121, %r5; + @%p28 ld.global.v4.b32 { %r118, %r119, %r120, %r121 }, [ %rd30 + 0 ]; + // end inline asm + mov.b32 {%rs392, %rs393}, %r118; + mov.b32 {%rs394, %rs395}, %r119; + mov.b32 {%rs396, %rs397}, %r120; + mov.b32 {%rs398, %rs399}, %r121; + and.b32 %r878, %r855, 1; + setp.ne.b32 %p29, %r878, 0; + // begin inline asm + mov.u32 %r122, %r5; + mov.u32 %r123, %r5; + mov.u32 %r124, %r5; + mov.u32 %r125, %r5; + @%p29 ld.global.v4.b32 { %r122, %r123, %r124, %r125 }, [ %rd31 + 0 ]; + // end inline asm + mov.b32 {%rs400, %rs401}, %r122; + mov.b32 {%rs402, %rs403}, %r123; + mov.b32 {%rs404, %rs405}, %r124; + mov.b32 {%rs406, %rs407}, %r125; + and.b32 %r879, %r854, 1; + setp.ne.b32 %p30, %r879, 0; + // begin inline asm + mov.u32 %r126, %r5; + mov.u32 %r127, %r5; + mov.u32 %r128, %r5; + mov.u32 %r129, %r5; + @%p30 ld.global.v4.b32 { %r126, %r127, %r128, %r129 }, [ %rd32 + 0 ]; + // end inline asm + mov.b32 {%rs408, %rs409}, %r126; + mov.b32 {%rs410, %rs411}, %r127; + mov.b32 {%rs412, %rs413}, %r128; + mov.b32 {%rs414, %rs415}, %r129; + and.b32 %r880, %r853, 1; + setp.ne.b32 %p31, %r880, 0; + // begin inline asm + mov.u32 %r130, %r5; + mov.u32 %r131, %r5; + mov.u32 %r132, %r5; + mov.u32 %r133, %r5; + @%p31 ld.global.v4.b32 { %r130, %r131, %r132, %r133 }, [ %rd33 + 0 ]; + // end inline asm + mov.b32 {%rs416, %rs417}, %r130; + mov.b32 {%rs418, %rs419}, %r131; + mov.b32 {%rs420, %rs421}, %r132; + mov.b32 {%rs422, %rs423}, %r133; + and.b32 %r881, %r852, 1; + setp.ne.b32 %p32, %r881, 0; + // begin inline asm + mov.u32 %r134, %r5; + mov.u32 %r135, %r5; + mov.u32 %r136, %r5; + mov.u32 %r137, %r5; + @%p32 ld.global.v4.b32 { %r134, %r135, %r136, %r137 }, [ %rd34 + 0 ]; + // end inline asm + mov.b32 {%rs424, %rs425}, %r134; + mov.b32 {%rs426, %rs427}, %r135; + mov.b32 {%rs428, %rs429}, %r136; + mov.b32 {%rs430, %rs431}, %r137; + and.b32 %r882, %r851, 1; + setp.ne.b32 %p33, %r882, 0; + // begin inline asm + mov.u32 %r138, %r5; + mov.u32 %r139, %r5; + mov.u32 %r140, %r5; + mov.u32 %r141, %r5; + @%p33 ld.global.v4.b32 { %r138, %r139, %r140, %r141 }, [ %rd35 + 0 ]; + // end inline asm + mov.b32 {%rs432, %rs433}, %r138; + mov.b32 {%rs434, %rs435}, %r139; + mov.b32 {%rs436, %rs437}, %r140; + mov.b32 {%rs438, %rs439}, %r141; + // begin inline asm + mov.u32 %r142, %r5; + mov.u32 %r143, %r5; + mov.u32 %r144, %r5; + mov.u32 %r145, %r5; + @%p34 ld.global.v4.b32 { %r142, %r143, %r144, %r145 }, [ %rd36 + 0 ]; + // end inline asm + mov.b32 {%rs440, %rs441}, %r142; + mov.b32 {%rs442, %rs443}, %r143; + mov.b32 {%rs444, %rs445}, %r144; + mov.b32 {%rs446, %rs447}, %r145; + shr.u64 %rd601, %rd339, 31; + and.b64 %rd602, %rd601, 1; + setp.ne.b64 %p41, %rd602, 0; + // begin inline asm + mov.u32 %r146, %r5; + mov.u32 %r147, %r5; + mov.u32 %r148, %r5; + mov.u32 %r149, %r5; + @%p41 ld.global.v4.b32 { %r146, %r147, %r148, %r149 }, [ %rd37 + 0 ]; + // end inline asm + mov.b32 {%rs448, %rs449}, %r146; + mov.b32 {%rs450, %rs451}, %r147; + mov.b32 {%rs452, %rs453}, %r148; + mov.b32 {%rs454, %rs455}, %r149; + shr.u64 %rd603, %rd339, 30; + and.b64 %rd604, %rd603, 1; + setp.ne.b64 %p42, %rd604, 0; + // begin inline asm + mov.u32 %r150, %r5; + mov.u32 %r151, %r5; + mov.u32 %r152, %r5; + mov.u32 %r153, %r5; + @%p42 ld.global.v4.b32 { %r150, %r151, %r152, %r153 }, [ %rd38 + 0 ]; + // end inline asm + mov.b32 {%rs456, %rs457}, %r150; + mov.b32 {%rs458, %rs459}, %r151; + mov.b32 {%rs460, %rs461}, %r152; + mov.b32 {%rs462, %rs463}, %r153; + shr.u64 %rd605, %rd339, 29; + and.b64 %rd606, %rd605, 1; + setp.ne.b64 %p43, %rd606, 0; + // begin inline asm + mov.u32 %r154, %r5; + mov.u32 %r155, %r5; + mov.u32 %r156, %r5; + mov.u32 %r157, %r5; + @%p43 ld.global.v4.b32 { %r154, %r155, %r156, %r157 }, [ %rd39 + 0 ]; + // end inline asm + mov.b32 {%rs464, %rs465}, %r154; + mov.b32 {%rs466, %rs467}, %r155; + mov.b32 {%rs468, %rs469}, %r156; + mov.b32 {%rs470, %rs471}, %r157; + shr.u64 %rd607, %rd339, 28; + and.b64 %rd608, %rd607, 1; + setp.ne.b64 %p44, %rd608, 0; + // begin inline asm + mov.u32 %r158, %r5; + mov.u32 %r159, %r5; + mov.u32 %r160, %r5; + mov.u32 %r161, %r5; + @%p44 ld.global.v4.b32 { %r158, %r159, %r160, %r161 }, [ %rd40 + 0 ]; + // end inline asm + mov.b32 {%rs472, %rs473}, %r158; + mov.b32 {%rs474, %rs475}, %r159; + mov.b32 {%rs476, %rs477}, %r160; + mov.b32 {%rs478, %rs479}, %r161; + shr.u64 %rd609, %rd339, 27; + and.b64 %rd610, %rd609, 1; + setp.ne.b64 %p45, %rd610, 0; + // begin inline asm + mov.u32 %r162, %r5; + mov.u32 %r163, %r5; + mov.u32 %r164, %r5; + mov.u32 %r165, %r5; + @%p45 ld.global.v4.b32 { %r162, %r163, %r164, %r165 }, [ %rd41 + 0 ]; + // end inline asm + mov.b32 {%rs480, %rs481}, %r162; + mov.b32 {%rs482, %rs483}, %r163; + mov.b32 {%rs484, %rs485}, %r164; + mov.b32 {%rs486, %rs487}, %r165; + shr.u64 %rd611, %rd339, 26; + and.b64 %rd612, %rd611, 1; + setp.ne.b64 %p46, %rd612, 0; + // begin inline asm + mov.u32 %r166, %r5; + mov.u32 %r167, %r5; + mov.u32 %r168, %r5; + mov.u32 %r169, %r5; + @%p46 ld.global.v4.b32 { %r166, %r167, %r168, %r169 }, [ %rd42 + 0 ]; + // end inline asm + mov.b32 {%rs488, %rs489}, %r166; + mov.b32 {%rs490, %rs491}, %r167; + mov.b32 {%rs492, %rs493}, %r168; + mov.b32 {%rs494, %rs495}, %r169; + shr.u64 %rd613, %rd339, 25; + and.b64 %rd614, %rd613, 1; + setp.ne.b64 %p47, %rd614, 0; + // begin inline asm + mov.u32 %r170, %r5; + mov.u32 %r171, %r5; + mov.u32 %r172, %r5; + mov.u32 %r173, %r5; + @%p47 ld.global.v4.b32 { %r170, %r171, %r172, %r173 }, [ %rd43 + 0 ]; + // end inline asm + mov.b32 {%rs496, %rs497}, %r170; + mov.b32 {%rs498, %rs499}, %r171; + mov.b32 {%rs500, %rs501}, %r172; + mov.b32 {%rs502, %rs503}, %r173; + shr.u64 %rd615, %rd339, 24; + and.b64 %rd616, %rd615, 1; + setp.ne.b64 %p48, %rd616, 0; + // begin inline asm + mov.u32 %r174, %r5; + mov.u32 %r175, %r5; + mov.u32 %r176, %r5; + mov.u32 %r177, %r5; + @%p48 ld.global.v4.b32 { %r174, %r175, %r176, %r177 }, [ %rd44 + 0 ]; + // end inline asm + mov.b32 {%rs504, %rs505}, %r174; + mov.b32 {%rs506, %rs507}, %r175; + mov.b32 {%rs508, %rs509}, %r176; + mov.b32 {%rs510, %rs511}, %r177; + shr.u64 %rd617, %rd339, 23; + and.b64 %rd618, %rd617, 1; + setp.ne.b64 %p49, %rd618, 0; + // begin inline asm + mov.u32 %r178, %r5; + mov.u32 %r179, %r5; + mov.u32 %r180, %r5; + mov.u32 %r181, %r5; + @%p49 ld.global.v4.b32 { %r178, %r179, %r180, %r181 }, [ %rd45 + 0 ]; + // end inline asm + mov.b32 {%rs512, %rs513}, %r178; + mov.b32 {%rs514, %rs515}, %r179; + mov.b32 {%rs516, %rs517}, %r180; + mov.b32 {%rs518, %rs519}, %r181; + shr.u64 %rd619, %rd339, 22; + and.b64 %rd620, %rd619, 1; + setp.ne.b64 %p50, %rd620, 0; + // begin inline asm + mov.u32 %r182, %r5; + mov.u32 %r183, %r5; + mov.u32 %r184, %r5; + mov.u32 %r185, %r5; + @%p50 ld.global.v4.b32 { %r182, %r183, %r184, %r185 }, [ %rd46 + 0 ]; + // end inline asm + mov.b32 {%rs520, %rs521}, %r182; + mov.b32 {%rs522, %rs523}, %r183; + mov.b32 {%rs524, %rs525}, %r184; + mov.b32 {%rs526, %rs527}, %r185; + shr.u64 %rd621, %rd339, 21; + and.b64 %rd622, %rd621, 1; + setp.ne.b64 %p51, %rd622, 0; + // begin inline asm + mov.u32 %r186, %r5; + mov.u32 %r187, %r5; + mov.u32 %r188, %r5; + mov.u32 %r189, %r5; + @%p51 ld.global.v4.b32 { %r186, %r187, %r188, %r189 }, [ %rd47 + 0 ]; + // end inline asm + mov.b32 {%rs528, %rs529}, %r186; + mov.b32 {%rs530, %rs531}, %r187; + mov.b32 {%rs532, %rs533}, %r188; + mov.b32 {%rs534, %rs535}, %r189; + shr.u64 %rd623, %rd339, 20; + and.b64 %rd624, %rd623, 1; + setp.ne.b64 %p52, %rd624, 0; + // begin inline asm + mov.u32 %r190, %r5; + mov.u32 %r191, %r5; + mov.u32 %r192, %r5; + mov.u32 %r193, %r5; + @%p52 ld.global.v4.b32 { %r190, %r191, %r192, %r193 }, [ %rd48 + 0 ]; + // end inline asm + mov.b32 {%rs536, %rs537}, %r190; + mov.b32 {%rs538, %rs539}, %r191; + mov.b32 {%rs540, %rs541}, %r192; + mov.b32 {%rs542, %rs543}, %r193; + shr.u64 %rd625, %rd339, 19; + and.b64 %rd626, %rd625, 1; + setp.ne.b64 %p53, %rd626, 0; + // begin inline asm + mov.u32 %r194, %r5; + mov.u32 %r195, %r5; + mov.u32 %r196, %r5; + mov.u32 %r197, %r5; + @%p53 ld.global.v4.b32 { %r194, %r195, %r196, %r197 }, [ %rd49 + 0 ]; + // end inline asm + mov.b32 {%rs544, %rs545}, %r194; + mov.b32 {%rs546, %rs547}, %r195; + mov.b32 {%rs548, %rs549}, %r196; + mov.b32 {%rs550, %rs551}, %r197; + shr.u64 %rd627, %rd339, 18; + and.b64 %rd628, %rd627, 1; + setp.ne.b64 %p54, %rd628, 0; + // begin inline asm + mov.u32 %r198, %r5; + mov.u32 %r199, %r5; + mov.u32 %r200, %r5; + mov.u32 %r201, %r5; + @%p54 ld.global.v4.b32 { %r198, %r199, %r200, %r201 }, [ %rd50 + 0 ]; + // end inline asm + mov.b32 {%rs552, %rs553}, %r198; + mov.b32 {%rs554, %rs555}, %r199; + mov.b32 {%rs556, %rs557}, %r200; + mov.b32 {%rs558, %rs559}, %r201; + shr.u64 %rd629, %rd339, 17; + and.b64 %rd630, %rd629, 1; + setp.ne.b64 %p55, %rd630, 0; + // begin inline asm + mov.u32 %r202, %r5; + mov.u32 %r203, %r5; + mov.u32 %r204, %r5; + mov.u32 %r205, %r5; + @%p55 ld.global.v4.b32 { %r202, %r203, %r204, %r205 }, [ %rd51 + 0 ]; + // end inline asm + mov.b32 {%rs560, %rs561}, %r202; + mov.b32 {%rs562, %rs563}, %r203; + mov.b32 {%rs564, %rs565}, %r204; + mov.b32 {%rs566, %rs567}, %r205; + shr.u64 %rd631, %rd339, 16; + and.b64 %rd632, %rd631, 1; + setp.ne.b64 %p56, %rd632, 0; + // begin inline asm + mov.u32 %r206, %r5; + mov.u32 %r207, %r5; + mov.u32 %r208, %r5; + mov.u32 %r209, %r5; + @%p56 ld.global.v4.b32 { %r206, %r207, %r208, %r209 }, [ %rd52 + 0 ]; + // end inline asm + mov.b32 {%rs568, %rs569}, %r206; + mov.b32 {%rs570, %rs571}, %r207; + mov.b32 {%rs572, %rs573}, %r208; + mov.b32 {%rs574, %rs575}, %r209; + shr.u64 %rd633, %rd339, 15; + and.b64 %rd634, %rd633, 1; + setp.ne.b64 %p57, %rd634, 0; + // begin inline asm + mov.u32 %r210, %r5; + mov.u32 %r211, %r5; + mov.u32 %r212, %r5; + mov.u32 %r213, %r5; + @%p57 ld.global.v4.b32 { %r210, %r211, %r212, %r213 }, [ %rd53 + 0 ]; + // end inline asm + mov.b32 {%rs576, %rs577}, %r210; + mov.b32 {%rs578, %rs579}, %r211; + mov.b32 {%rs580, %rs581}, %r212; + mov.b32 {%rs582, %rs583}, %r213; + shr.u64 %rd635, %rd339, 14; + and.b64 %rd636, %rd635, 1; + setp.ne.b64 %p58, %rd636, 0; + // begin inline asm + mov.u32 %r214, %r5; + mov.u32 %r215, %r5; + mov.u32 %r216, %r5; + mov.u32 %r217, %r5; + @%p58 ld.global.v4.b32 { %r214, %r215, %r216, %r217 }, [ %rd54 + 0 ]; + // end inline asm + mov.b32 {%rs584, %rs585}, %r214; + mov.b32 {%rs586, %rs587}, %r215; + mov.b32 {%rs588, %rs589}, %r216; + mov.b32 {%rs590, %rs591}, %r217; + shr.u64 %rd637, %rd339, 13; + and.b64 %rd638, %rd637, 1; + setp.ne.b64 %p59, %rd638, 0; + // begin inline asm + mov.u32 %r218, %r5; + mov.u32 %r219, %r5; + mov.u32 %r220, %r5; + mov.u32 %r221, %r5; + @%p59 ld.global.v4.b32 { %r218, %r219, %r220, %r221 }, [ %rd55 + 0 ]; + // end inline asm + mov.b32 {%rs592, %rs593}, %r218; + mov.b32 {%rs594, %rs595}, %r219; + mov.b32 {%rs596, %rs597}, %r220; + mov.b32 {%rs598, %rs599}, %r221; + shr.u64 %rd639, %rd339, 12; + and.b64 %rd640, %rd639, 1; + setp.ne.b64 %p60, %rd640, 0; + // begin inline asm + mov.u32 %r222, %r5; + mov.u32 %r223, %r5; + mov.u32 %r224, %r5; + mov.u32 %r225, %r5; + @%p60 ld.global.v4.b32 { %r222, %r223, %r224, %r225 }, [ %rd56 + 0 ]; + // end inline asm + mov.b32 {%rs600, %rs601}, %r222; + mov.b32 {%rs602, %rs603}, %r223; + mov.b32 {%rs604, %rs605}, %r224; + mov.b32 {%rs606, %rs607}, %r225; + shr.u64 %rd641, %rd339, 11; + and.b64 %rd642, %rd641, 1; + setp.ne.b64 %p61, %rd642, 0; + // begin inline asm + mov.u32 %r226, %r5; + mov.u32 %r227, %r5; + mov.u32 %r228, %r5; + mov.u32 %r229, %r5; + @%p61 ld.global.v4.b32 { %r226, %r227, %r228, %r229 }, [ %rd57 + 0 ]; + // end inline asm + mov.b32 {%rs608, %rs609}, %r226; + mov.b32 {%rs610, %rs611}, %r227; + mov.b32 {%rs612, %rs613}, %r228; + mov.b32 {%rs614, %rs615}, %r229; + shr.u64 %rd643, %rd339, 10; + and.b64 %rd644, %rd643, 1; + setp.ne.b64 %p62, %rd644, 0; + // begin inline asm + mov.u32 %r230, %r5; + mov.u32 %r231, %r5; + mov.u32 %r232, %r5; + mov.u32 %r233, %r5; + @%p62 ld.global.v4.b32 { %r230, %r231, %r232, %r233 }, [ %rd58 + 0 ]; + // end inline asm + mov.b32 {%rs616, %rs617}, %r230; + mov.b32 {%rs618, %rs619}, %r231; + mov.b32 {%rs620, %rs621}, %r232; + mov.b32 {%rs622, %rs623}, %r233; + shr.u64 %rd645, %rd339, 9; + and.b64 %rd646, %rd645, 1; + setp.ne.b64 %p63, %rd646, 0; + // begin inline asm + mov.u32 %r234, %r5; + mov.u32 %r235, %r5; + mov.u32 %r236, %r5; + mov.u32 %r237, %r5; + @%p63 ld.global.v4.b32 { %r234, %r235, %r236, %r237 }, [ %rd59 + 0 ]; + // end inline asm + mov.b32 {%rs624, %rs625}, %r234; + mov.b32 {%rs626, %rs627}, %r235; + mov.b32 {%rs628, %rs629}, %r236; + mov.b32 {%rs630, %rs631}, %r237; + shr.u64 %rd647, %rd339, 8; + and.b64 %rd648, %rd647, 1; + setp.ne.b64 %p64, %rd648, 0; + // begin inline asm + mov.u32 %r238, %r5; + mov.u32 %r239, %r5; + mov.u32 %r240, %r5; + mov.u32 %r241, %r5; + @%p64 ld.global.v4.b32 { %r238, %r239, %r240, %r241 }, [ %rd60 + 0 ]; + // end inline asm + mov.b32 {%rs632, %rs633}, %r238; + mov.b32 {%rs634, %rs635}, %r239; + mov.b32 {%rs636, %rs637}, %r240; + mov.b32 {%rs638, %rs639}, %r241; + shr.u64 %rd649, %rd339, 7; + and.b64 %rd650, %rd649, 1; + setp.ne.b64 %p65, %rd650, 0; + // begin inline asm + mov.u32 %r242, %r5; + mov.u32 %r243, %r5; + mov.u32 %r244, %r5; + mov.u32 %r245, %r5; + @%p65 ld.global.v4.b32 { %r242, %r243, %r244, %r245 }, [ %rd61 + 0 ]; + // end inline asm + mov.b32 {%rs640, %rs641}, %r242; + mov.b32 {%rs642, %rs643}, %r243; + mov.b32 {%rs644, %rs645}, %r244; + mov.b32 {%rs646, %rs647}, %r245; + shr.u64 %rd651, %rd339, 6; + and.b64 %rd652, %rd651, 1; + setp.ne.b64 %p66, %rd652, 0; + // begin inline asm + mov.u32 %r246, %r5; + mov.u32 %r247, %r5; + mov.u32 %r248, %r5; + mov.u32 %r249, %r5; + @%p66 ld.global.v4.b32 { %r246, %r247, %r248, %r249 }, [ %rd62 + 0 ]; + // end inline asm + mov.b32 {%rs648, %rs649}, %r246; + mov.b32 {%rs650, %rs651}, %r247; + mov.b32 {%rs652, %rs653}, %r248; + mov.b32 {%rs654, %rs655}, %r249; + shr.u64 %rd653, %rd339, 5; + and.b64 %rd654, %rd653, 1; + setp.ne.b64 %p67, %rd654, 0; + // begin inline asm + mov.u32 %r250, %r5; + mov.u32 %r251, %r5; + mov.u32 %r252, %r5; + mov.u32 %r253, %r5; + @%p67 ld.global.v4.b32 { %r250, %r251, %r252, %r253 }, [ %rd63 + 0 ]; + // end inline asm + mov.b32 {%rs656, %rs657}, %r250; + mov.b32 {%rs658, %rs659}, %r251; + mov.b32 {%rs660, %rs661}, %r252; + mov.b32 {%rs662, %rs663}, %r253; + shr.u64 %rd655, %rd339, 4; + and.b64 %rd656, %rd655, 1; + setp.ne.b64 %p68, %rd656, 0; + // begin inline asm + mov.u32 %r254, %r5; + mov.u32 %r255, %r5; + mov.u32 %r256, %r5; + mov.u32 %r257, %r5; + @%p68 ld.global.v4.b32 { %r254, %r255, %r256, %r257 }, [ %rd64 + 0 ]; + // end inline asm + mov.b32 {%rs664, %rs665}, %r254; + mov.b32 {%rs666, %rs667}, %r255; + mov.b32 {%rs668, %rs669}, %r256; + mov.b32 {%rs670, %rs671}, %r257; + .loc 1 104 47 // l2norm.py:104:47 + cvt.f32.bf16 %r883, %rs160; + cvt.f32.bf16 %r884, %rs161; + cvt.f32.bf16 %r885, %rs162; + cvt.f32.bf16 %r886, %rs163; + cvt.f32.bf16 %r887, %rs164; + cvt.f32.bf16 %r888, %rs165; + cvt.f32.bf16 %r889, %rs166; + cvt.f32.bf16 %r890, %rs167; + cvt.f32.bf16 %r891, %rs168; + cvt.f32.bf16 %r892, %rs169; + cvt.f32.bf16 %r893, %rs170; + cvt.f32.bf16 %r894, %rs171; + cvt.f32.bf16 %r895, %rs172; + cvt.f32.bf16 %r896, %rs173; + cvt.f32.bf16 %r897, %rs174; + cvt.f32.bf16 %r898, %rs175; + cvt.f32.bf16 %r899, %rs176; + cvt.f32.bf16 %r900, %rs177; + cvt.f32.bf16 %r901, %rs178; + cvt.f32.bf16 %r902, %rs179; + cvt.f32.bf16 %r903, %rs180; + cvt.f32.bf16 %r904, %rs181; + cvt.f32.bf16 %r905, %rs182; + cvt.f32.bf16 %r906, %rs183; + cvt.f32.bf16 %r907, %rs184; + cvt.f32.bf16 %r908, %rs185; + cvt.f32.bf16 %r909, %rs186; + cvt.f32.bf16 %r910, %rs187; + cvt.f32.bf16 %r911, %rs188; + cvt.f32.bf16 %r912, %rs189; + cvt.f32.bf16 %r913, %rs190; + cvt.f32.bf16 %r914, %rs191; + cvt.f32.bf16 %r915, %rs192; + cvt.f32.bf16 %r916, %rs193; + cvt.f32.bf16 %r917, %rs194; + cvt.f32.bf16 %r918, %rs195; + cvt.f32.bf16 %r919, %rs196; + cvt.f32.bf16 %r920, %rs197; + cvt.f32.bf16 %r921, %rs198; + cvt.f32.bf16 %r922, %rs199; + cvt.f32.bf16 %r923, %rs200; + cvt.f32.bf16 %r924, %rs201; + cvt.f32.bf16 %r925, %rs202; + cvt.f32.bf16 %r926, %rs203; + cvt.f32.bf16 %r927, %rs204; + cvt.f32.bf16 %r928, %rs205; + cvt.f32.bf16 %r929, %rs206; + cvt.f32.bf16 %r930, %rs207; + cvt.f32.bf16 %r931, %rs208; + cvt.f32.bf16 %r932, %rs209; + cvt.f32.bf16 %r933, %rs210; + cvt.f32.bf16 %r934, %rs211; + cvt.f32.bf16 %r935, %rs212; + cvt.f32.bf16 %r936, %rs213; + cvt.f32.bf16 %r937, %rs214; + cvt.f32.bf16 %r938, %rs215; + cvt.f32.bf16 %r939, %rs216; + cvt.f32.bf16 %r940, %rs217; + cvt.f32.bf16 %r941, %rs218; + cvt.f32.bf16 %r942, %rs219; + cvt.f32.bf16 %r943, %rs220; + cvt.f32.bf16 %r944, %rs221; + cvt.f32.bf16 %r945, %rs222; + cvt.f32.bf16 %r946, %rs223; + cvt.f32.bf16 %r947, %rs224; + cvt.f32.bf16 %r948, %rs225; + cvt.f32.bf16 %r949, %rs226; + cvt.f32.bf16 %r950, %rs227; + cvt.f32.bf16 %r951, %rs228; + cvt.f32.bf16 %r952, %rs229; + cvt.f32.bf16 %r953, %rs230; + cvt.f32.bf16 %r954, %rs231; + cvt.f32.bf16 %r955, %rs232; + cvt.f32.bf16 %r956, %rs233; + cvt.f32.bf16 %r957, %rs234; + cvt.f32.bf16 %r958, %rs235; + cvt.f32.bf16 %r959, %rs236; + cvt.f32.bf16 %r960, %rs237; + cvt.f32.bf16 %r961, %rs238; + cvt.f32.bf16 %r962, %rs239; + cvt.f32.bf16 %r963, %rs240; + cvt.f32.bf16 %r964, %rs241; + cvt.f32.bf16 %r965, %rs242; + cvt.f32.bf16 %r966, %rs243; + cvt.f32.bf16 %r967, %rs244; + cvt.f32.bf16 %r968, %rs245; + cvt.f32.bf16 %r969, %rs246; + cvt.f32.bf16 %r970, %rs247; + cvt.f32.bf16 %r971, %rs248; + cvt.f32.bf16 %r972, %rs249; + cvt.f32.bf16 %r973, %rs250; + cvt.f32.bf16 %r974, %rs251; + cvt.f32.bf16 %r975, %rs252; + cvt.f32.bf16 %r976, %rs253; + cvt.f32.bf16 %r977, %rs254; + cvt.f32.bf16 %r978, %rs255; + cvt.f32.bf16 %r979, %rs256; + cvt.f32.bf16 %r980, %rs257; + cvt.f32.bf16 %r981, %rs258; + cvt.f32.bf16 %r982, %rs259; + cvt.f32.bf16 %r983, %rs260; + cvt.f32.bf16 %r984, %rs261; + cvt.f32.bf16 %r985, %rs262; + cvt.f32.bf16 %r986, %rs263; + cvt.f32.bf16 %r987, %rs264; + cvt.f32.bf16 %r988, %rs265; + cvt.f32.bf16 %r989, %rs266; + cvt.f32.bf16 %r990, %rs267; + cvt.f32.bf16 %r991, %rs268; + cvt.f32.bf16 %r992, %rs269; + cvt.f32.bf16 %r993, %rs270; + cvt.f32.bf16 %r994, %rs271; + cvt.f32.bf16 %r995, %rs272; + cvt.f32.bf16 %r996, %rs273; + cvt.f32.bf16 %r997, %rs274; + cvt.f32.bf16 %r998, %rs275; + cvt.f32.bf16 %r999, %rs276; + cvt.f32.bf16 %r1000, %rs277; + cvt.f32.bf16 %r1001, %rs278; + cvt.f32.bf16 %r1002, %rs279; + cvt.f32.bf16 %r1003, %rs280; + cvt.f32.bf16 %r1004, %rs281; + cvt.f32.bf16 %r1005, %rs282; + cvt.f32.bf16 %r1006, %rs283; + cvt.f32.bf16 %r1007, %rs284; + cvt.f32.bf16 %r1008, %rs285; + cvt.f32.bf16 %r1009, %rs286; + cvt.f32.bf16 %r1010, %rs287; + cvt.f32.bf16 %r1011, %rs288; + cvt.f32.bf16 %r1012, %rs289; + cvt.f32.bf16 %r1013, %rs290; + cvt.f32.bf16 %r1014, %rs291; + cvt.f32.bf16 %r1015, %rs292; + cvt.f32.bf16 %r1016, %rs293; + cvt.f32.bf16 %r1017, %rs294; + cvt.f32.bf16 %r1018, %rs295; + cvt.f32.bf16 %r1019, %rs296; + cvt.f32.bf16 %r1020, %rs297; + cvt.f32.bf16 %r1021, %rs298; + cvt.f32.bf16 %r1022, %rs299; + cvt.f32.bf16 %r1023, %rs300; + cvt.f32.bf16 %r1024, %rs301; + cvt.f32.bf16 %r1025, %rs302; + cvt.f32.bf16 %r1026, %rs303; + cvt.f32.bf16 %r1027, %rs304; + cvt.f32.bf16 %r1028, %rs305; + cvt.f32.bf16 %r1029, %rs306; + cvt.f32.bf16 %r1030, %rs307; + cvt.f32.bf16 %r1031, %rs308; + cvt.f32.bf16 %r1032, %rs309; + cvt.f32.bf16 %r1033, %rs310; + cvt.f32.bf16 %r1034, %rs311; + cvt.f32.bf16 %r1035, %rs312; + cvt.f32.bf16 %r1036, %rs313; + cvt.f32.bf16 %r1037, %rs314; + cvt.f32.bf16 %r1038, %rs315; + cvt.f32.bf16 %r1039, %rs316; + cvt.f32.bf16 %r1040, %rs317; + cvt.f32.bf16 %r1041, %rs318; + cvt.f32.bf16 %r1042, %rs319; + cvt.f32.bf16 %r1043, %rs320; + cvt.f32.bf16 %r1044, %rs321; + cvt.f32.bf16 %r1045, %rs322; + cvt.f32.bf16 %r1046, %rs323; + cvt.f32.bf16 %r1047, %rs324; + cvt.f32.bf16 %r1048, %rs325; + cvt.f32.bf16 %r1049, %rs326; + cvt.f32.bf16 %r1050, %rs327; + cvt.f32.bf16 %r1051, %rs328; + cvt.f32.bf16 %r1052, %rs329; + cvt.f32.bf16 %r1053, %rs330; + cvt.f32.bf16 %r1054, %rs331; + cvt.f32.bf16 %r1055, %rs332; + cvt.f32.bf16 %r1056, %rs333; + cvt.f32.bf16 %r1057, %rs334; + cvt.f32.bf16 %r1058, %rs335; + cvt.f32.bf16 %r1059, %rs336; + cvt.f32.bf16 %r1060, %rs337; + cvt.f32.bf16 %r1061, %rs338; + cvt.f32.bf16 %r1062, %rs339; + cvt.f32.bf16 %r1063, %rs340; + cvt.f32.bf16 %r1064, %rs341; + cvt.f32.bf16 %r1065, %rs342; + cvt.f32.bf16 %r1066, %rs343; + cvt.f32.bf16 %r1067, %rs344; + cvt.f32.bf16 %r1068, %rs345; + cvt.f32.bf16 %r1069, %rs346; + cvt.f32.bf16 %r1070, %rs347; + cvt.f32.bf16 %r1071, %rs348; + cvt.f32.bf16 %r1072, %rs349; + cvt.f32.bf16 %r1073, %rs350; + cvt.f32.bf16 %r1074, %rs351; + cvt.f32.bf16 %r1075, %rs352; + cvt.f32.bf16 %r1076, %rs353; + cvt.f32.bf16 %r1077, %rs354; + cvt.f32.bf16 %r1078, %rs355; + cvt.f32.bf16 %r1079, %rs356; + cvt.f32.bf16 %r1080, %rs357; + cvt.f32.bf16 %r1081, %rs358; + cvt.f32.bf16 %r1082, %rs359; + cvt.f32.bf16 %r1083, %rs360; + cvt.f32.bf16 %r1084, %rs361; + cvt.f32.bf16 %r1085, %rs362; + cvt.f32.bf16 %r1086, %rs363; + cvt.f32.bf16 %r1087, %rs364; + cvt.f32.bf16 %r1088, %rs365; + cvt.f32.bf16 %r1089, %rs366; + cvt.f32.bf16 %r1090, %rs367; + cvt.f32.bf16 %r1091, %rs368; + cvt.f32.bf16 %r1092, %rs369; + cvt.f32.bf16 %r1093, %rs370; + cvt.f32.bf16 %r1094, %rs371; + cvt.f32.bf16 %r1095, %rs372; + cvt.f32.bf16 %r1096, %rs373; + cvt.f32.bf16 %r1097, %rs374; + cvt.f32.bf16 %r1098, %rs375; + cvt.f32.bf16 %r1099, %rs376; + cvt.f32.bf16 %r1100, %rs377; + cvt.f32.bf16 %r1101, %rs378; + cvt.f32.bf16 %r1102, %rs379; + cvt.f32.bf16 %r1103, %rs380; + cvt.f32.bf16 %r1104, %rs381; + cvt.f32.bf16 %r1105, %rs382; + cvt.f32.bf16 %r1106, %rs383; + cvt.f32.bf16 %r1107, %rs384; + cvt.f32.bf16 %r1108, %rs385; + cvt.f32.bf16 %r1109, %rs386; + cvt.f32.bf16 %r1110, %rs387; + cvt.f32.bf16 %r1111, %rs388; + cvt.f32.bf16 %r1112, %rs389; + cvt.f32.bf16 %r1113, %rs390; + cvt.f32.bf16 %r1114, %rs391; + cvt.f32.bf16 %r1115, %rs392; + cvt.f32.bf16 %r1116, %rs393; + cvt.f32.bf16 %r1117, %rs394; + cvt.f32.bf16 %r1118, %rs395; + cvt.f32.bf16 %r1119, %rs396; + cvt.f32.bf16 %r1120, %rs397; + cvt.f32.bf16 %r1121, %rs398; + cvt.f32.bf16 %r1122, %rs399; + cvt.f32.bf16 %r1123, %rs400; + cvt.f32.bf16 %r1124, %rs401; + cvt.f32.bf16 %r1125, %rs402; + cvt.f32.bf16 %r1126, %rs403; + cvt.f32.bf16 %r1127, %rs404; + cvt.f32.bf16 %r1128, %rs405; + cvt.f32.bf16 %r1129, %rs406; + cvt.f32.bf16 %r1130, %rs407; + cvt.f32.bf16 %r1131, %rs408; + cvt.f32.bf16 %r1132, %rs409; + cvt.f32.bf16 %r1133, %rs410; + cvt.f32.bf16 %r1134, %rs411; + cvt.f32.bf16 %r1135, %rs412; + cvt.f32.bf16 %r1136, %rs413; + cvt.f32.bf16 %r1137, %rs414; + cvt.f32.bf16 %r1138, %rs415; + cvt.f32.bf16 %r1139, %rs416; + cvt.f32.bf16 %r1140, %rs417; + cvt.f32.bf16 %r1141, %rs418; + cvt.f32.bf16 %r1142, %rs419; + cvt.f32.bf16 %r1143, %rs420; + cvt.f32.bf16 %r1144, %rs421; + cvt.f32.bf16 %r1145, %rs422; + cvt.f32.bf16 %r1146, %rs423; + cvt.f32.bf16 %r1147, %rs424; + cvt.f32.bf16 %r1148, %rs425; + cvt.f32.bf16 %r1149, %rs426; + cvt.f32.bf16 %r1150, %rs427; + cvt.f32.bf16 %r1151, %rs428; + cvt.f32.bf16 %r1152, %rs429; + cvt.f32.bf16 %r1153, %rs430; + cvt.f32.bf16 %r1154, %rs431; + cvt.f32.bf16 %r1155, %rs432; + cvt.f32.bf16 %r1156, %rs433; + cvt.f32.bf16 %r1157, %rs434; + cvt.f32.bf16 %r1158, %rs435; + cvt.f32.bf16 %r1159, %rs436; + cvt.f32.bf16 %r1160, %rs437; + cvt.f32.bf16 %r1161, %rs438; + cvt.f32.bf16 %r1162, %rs439; + cvt.f32.bf16 %r1163, %rs440; + cvt.f32.bf16 %r1164, %rs441; + cvt.f32.bf16 %r1165, %rs442; + cvt.f32.bf16 %r1166, %rs443; + cvt.f32.bf16 %r1167, %rs444; + cvt.f32.bf16 %r1168, %rs445; + cvt.f32.bf16 %r1169, %rs446; + cvt.f32.bf16 %r1170, %rs447; + cvt.f32.bf16 %r1171, %rs448; + cvt.f32.bf16 %r1172, %rs449; + cvt.f32.bf16 %r1173, %rs450; + cvt.f32.bf16 %r1174, %rs451; + cvt.f32.bf16 %r1175, %rs452; + cvt.f32.bf16 %r1176, %rs453; + cvt.f32.bf16 %r1177, %rs454; + cvt.f32.bf16 %r1178, %rs455; + cvt.f32.bf16 %r1179, %rs456; + cvt.f32.bf16 %r1180, %rs457; + cvt.f32.bf16 %r1181, %rs458; + cvt.f32.bf16 %r1182, %rs459; + cvt.f32.bf16 %r1183, %rs460; + cvt.f32.bf16 %r1184, %rs461; + cvt.f32.bf16 %r1185, %rs462; + cvt.f32.bf16 %r1186, %rs463; + cvt.f32.bf16 %r1187, %rs464; + cvt.f32.bf16 %r1188, %rs465; + cvt.f32.bf16 %r1189, %rs466; + cvt.f32.bf16 %r1190, %rs467; + cvt.f32.bf16 %r1191, %rs468; + cvt.f32.bf16 %r1192, %rs469; + cvt.f32.bf16 %r1193, %rs470; + cvt.f32.bf16 %r1194, %rs471; + cvt.f32.bf16 %r1195, %rs472; + cvt.f32.bf16 %r1196, %rs473; + cvt.f32.bf16 %r1197, %rs474; + cvt.f32.bf16 %r1198, %rs475; + cvt.f32.bf16 %r1199, %rs476; + cvt.f32.bf16 %r1200, %rs477; + cvt.f32.bf16 %r1201, %rs478; + cvt.f32.bf16 %r1202, %rs479; + cvt.f32.bf16 %r1203, %rs480; + cvt.f32.bf16 %r1204, %rs481; + cvt.f32.bf16 %r1205, %rs482; + cvt.f32.bf16 %r1206, %rs483; + cvt.f32.bf16 %r1207, %rs484; + cvt.f32.bf16 %r1208, %rs485; + cvt.f32.bf16 %r1209, %rs486; + cvt.f32.bf16 %r1210, %rs487; + cvt.f32.bf16 %r1211, %rs488; + cvt.f32.bf16 %r1212, %rs489; + cvt.f32.bf16 %r1213, %rs490; + cvt.f32.bf16 %r1214, %rs491; + cvt.f32.bf16 %r1215, %rs492; + cvt.f32.bf16 %r1216, %rs493; + cvt.f32.bf16 %r1217, %rs494; + cvt.f32.bf16 %r1218, %rs495; + cvt.f32.bf16 %r1219, %rs496; + cvt.f32.bf16 %r1220, %rs497; + cvt.f32.bf16 %r1221, %rs498; + cvt.f32.bf16 %r1222, %rs499; + cvt.f32.bf16 %r1223, %rs500; + cvt.f32.bf16 %r1224, %rs501; + cvt.f32.bf16 %r1225, %rs502; + cvt.f32.bf16 %r1226, %rs503; + cvt.f32.bf16 %r1227, %rs504; + cvt.f32.bf16 %r1228, %rs505; + cvt.f32.bf16 %r1229, %rs506; + cvt.f32.bf16 %r1230, %rs507; + cvt.f32.bf16 %r1231, %rs508; + cvt.f32.bf16 %r1232, %rs509; + cvt.f32.bf16 %r1233, %rs510; + cvt.f32.bf16 %r1234, %rs511; + cvt.f32.bf16 %r1235, %rs512; + cvt.f32.bf16 %r1236, %rs513; + cvt.f32.bf16 %r1237, %rs514; + cvt.f32.bf16 %r1238, %rs515; + cvt.f32.bf16 %r1239, %rs516; + cvt.f32.bf16 %r1240, %rs517; + cvt.f32.bf16 %r1241, %rs518; + cvt.f32.bf16 %r1242, %rs519; + cvt.f32.bf16 %r1243, %rs520; + cvt.f32.bf16 %r1244, %rs521; + cvt.f32.bf16 %r1245, %rs522; + cvt.f32.bf16 %r1246, %rs523; + cvt.f32.bf16 %r1247, %rs524; + cvt.f32.bf16 %r1248, %rs525; + cvt.f32.bf16 %r1249, %rs526; + cvt.f32.bf16 %r1250, %rs527; + cvt.f32.bf16 %r1251, %rs528; + cvt.f32.bf16 %r1252, %rs529; + cvt.f32.bf16 %r1253, %rs530; + cvt.f32.bf16 %r1254, %rs531; + cvt.f32.bf16 %r1255, %rs532; + cvt.f32.bf16 %r1256, %rs533; + cvt.f32.bf16 %r1257, %rs534; + cvt.f32.bf16 %r1258, %rs535; + cvt.f32.bf16 %r1259, %rs536; + cvt.f32.bf16 %r1260, %rs537; + cvt.f32.bf16 %r1261, %rs538; + cvt.f32.bf16 %r1262, %rs539; + cvt.f32.bf16 %r1263, %rs540; + cvt.f32.bf16 %r1264, %rs541; + cvt.f32.bf16 %r1265, %rs542; + cvt.f32.bf16 %r1266, %rs543; + cvt.f32.bf16 %r1267, %rs544; + cvt.f32.bf16 %r1268, %rs545; + cvt.f32.bf16 %r1269, %rs546; + cvt.f32.bf16 %r1270, %rs547; + cvt.f32.bf16 %r1271, %rs548; + cvt.f32.bf16 %r1272, %rs549; + cvt.f32.bf16 %r1273, %rs550; + cvt.f32.bf16 %r1274, %rs551; + cvt.f32.bf16 %r1275, %rs552; + cvt.f32.bf16 %r1276, %rs553; + cvt.f32.bf16 %r1277, %rs554; + cvt.f32.bf16 %r1278, %rs555; + cvt.f32.bf16 %r1279, %rs556; + cvt.f32.bf16 %r1280, %rs557; + cvt.f32.bf16 %r1281, %rs558; + cvt.f32.bf16 %r1282, %rs559; + cvt.f32.bf16 %r1283, %rs560; + cvt.f32.bf16 %r1284, %rs561; + cvt.f32.bf16 %r1285, %rs562; + cvt.f32.bf16 %r1286, %rs563; + cvt.f32.bf16 %r1287, %rs564; + cvt.f32.bf16 %r1288, %rs565; + cvt.f32.bf16 %r1289, %rs566; + cvt.f32.bf16 %r1290, %rs567; + cvt.f32.bf16 %r1291, %rs568; + cvt.f32.bf16 %r1292, %rs569; + cvt.f32.bf16 %r1293, %rs570; + cvt.f32.bf16 %r1294, %rs571; + cvt.f32.bf16 %r1295, %rs572; + cvt.f32.bf16 %r1296, %rs573; + cvt.f32.bf16 %r1297, %rs574; + cvt.f32.bf16 %r1298, %rs575; + cvt.f32.bf16 %r1299, %rs576; + cvt.f32.bf16 %r1300, %rs577; + cvt.f32.bf16 %r1301, %rs578; + cvt.f32.bf16 %r1302, %rs579; + cvt.f32.bf16 %r1303, %rs580; + cvt.f32.bf16 %r1304, %rs581; + cvt.f32.bf16 %r1305, %rs582; + cvt.f32.bf16 %r1306, %rs583; + cvt.f32.bf16 %r1307, %rs584; + cvt.f32.bf16 %r1308, %rs585; + cvt.f32.bf16 %r1309, %rs586; + cvt.f32.bf16 %r1310, %rs587; + cvt.f32.bf16 %r1311, %rs588; + cvt.f32.bf16 %r1312, %rs589; + cvt.f32.bf16 %r1313, %rs590; + cvt.f32.bf16 %r1314, %rs591; + cvt.f32.bf16 %r1315, %rs592; + cvt.f32.bf16 %r1316, %rs593; + cvt.f32.bf16 %r1317, %rs594; + cvt.f32.bf16 %r1318, %rs595; + cvt.f32.bf16 %r1319, %rs596; + cvt.f32.bf16 %r1320, %rs597; + cvt.f32.bf16 %r1321, %rs598; + cvt.f32.bf16 %r1322, %rs599; + cvt.f32.bf16 %r1323, %rs600; + cvt.f32.bf16 %r1324, %rs601; + cvt.f32.bf16 %r1325, %rs602; + cvt.f32.bf16 %r1326, %rs603; + cvt.f32.bf16 %r1327, %rs604; + cvt.f32.bf16 %r1328, %rs605; + cvt.f32.bf16 %r1329, %rs606; + cvt.f32.bf16 %r1330, %rs607; + cvt.f32.bf16 %r1331, %rs608; + cvt.f32.bf16 %r1332, %rs609; + cvt.f32.bf16 %r1333, %rs610; + cvt.f32.bf16 %r1334, %rs611; + cvt.f32.bf16 %r1335, %rs612; + cvt.f32.bf16 %r1336, %rs613; + cvt.f32.bf16 %r1337, %rs614; + cvt.f32.bf16 %r1338, %rs615; + cvt.f32.bf16 %r1339, %rs616; + cvt.f32.bf16 %r1340, %rs617; + cvt.f32.bf16 %r1341, %rs618; + cvt.f32.bf16 %r1342, %rs619; + cvt.f32.bf16 %r1343, %rs620; + cvt.f32.bf16 %r1344, %rs621; + cvt.f32.bf16 %r1345, %rs622; + cvt.f32.bf16 %r1346, %rs623; + cvt.f32.bf16 %r1347, %rs624; + cvt.f32.bf16 %r1348, %rs625; + cvt.f32.bf16 %r1349, %rs626; + cvt.f32.bf16 %r1350, %rs627; + cvt.f32.bf16 %r1351, %rs628; + cvt.f32.bf16 %r1352, %rs629; + cvt.f32.bf16 %r1353, %rs630; + cvt.f32.bf16 %r1354, %rs631; + cvt.f32.bf16 %r1355, %rs632; + cvt.f32.bf16 %r1356, %rs633; + cvt.f32.bf16 %r1357, %rs634; + cvt.f32.bf16 %r1358, %rs635; + cvt.f32.bf16 %r1359, %rs636; + cvt.f32.bf16 %r1360, %rs637; + cvt.f32.bf16 %r1361, %rs638; + cvt.f32.bf16 %r1362, %rs639; + cvt.f32.bf16 %r1363, %rs640; + cvt.f32.bf16 %r1364, %rs641; + cvt.f32.bf16 %r1365, %rs642; + cvt.f32.bf16 %r1366, %rs643; + cvt.f32.bf16 %r1367, %rs644; + cvt.f32.bf16 %r1368, %rs645; + cvt.f32.bf16 %r1369, %rs646; + cvt.f32.bf16 %r1370, %rs647; + cvt.f32.bf16 %r1371, %rs648; + cvt.f32.bf16 %r1372, %rs649; + cvt.f32.bf16 %r1373, %rs650; + cvt.f32.bf16 %r1374, %rs651; + cvt.f32.bf16 %r1375, %rs652; + cvt.f32.bf16 %r1376, %rs653; + cvt.f32.bf16 %r1377, %rs654; + cvt.f32.bf16 %r1378, %rs655; + cvt.f32.bf16 %r1379, %rs656; + cvt.f32.bf16 %r1380, %rs657; + cvt.f32.bf16 %r1381, %rs658; + cvt.f32.bf16 %r1382, %rs659; + cvt.f32.bf16 %r1383, %rs660; + cvt.f32.bf16 %r1384, %rs661; + cvt.f32.bf16 %r1385, %rs662; + cvt.f32.bf16 %r1386, %rs663; + cvt.f32.bf16 %r1387, %rs664; + cvt.f32.bf16 %r1388, %rs665; + cvt.f32.bf16 %r1389, %rs666; + cvt.f32.bf16 %r1390, %rs667; + cvt.f32.bf16 %r1391, %rs668; + cvt.f32.bf16 %r1392, %rs669; + cvt.f32.bf16 %r1393, %rs670; + cvt.f32.bf16 %r1394, %rs671; + .loc 1 105 21 // l2norm.py:105:21 + shr.u64 %rd657, %rd339, 3; + and.b64 %rd658, %rd657, 1; + setp.ne.b64 %p35, %rd658, 0; + // begin inline asm + mov.u32 %r258, %r5; + @%p35 ld.global.b32 { %r258 }, [ %rd65 + 0 ]; + // end inline asm + shr.u64 %rd659, %rd339, 2; + and.b64 %rd660, %rd659, 1; + setp.ne.b64 %p36, %rd660, 0; + // begin inline asm + mov.u32 %r259, %r5; + @%p36 ld.global.b32 { %r259 }, [ %rd66 + 0 ]; + // end inline asm + shr.u64 %rd661, %rd339, 1; + and.b64 %rd662, %rd661, 1; + setp.ne.b64 %p37, %rd662, 0; + // begin inline asm + mov.u32 %r260, %r5; + @%p37 ld.global.b32 { %r260 }, [ %rd67 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r261, %r5; + @%p38 ld.global.b32 { %r261 }, [ %rd68 + 0 ]; + // end inline asm + .loc 1 106 19 // l2norm.py:106:19 + // begin inline asm + mov.u32 %r262, %r5; + mov.u32 %r263, %r5; + mov.u32 %r264, %r5; + mov.u32 %r265, %r5; + @%p1 ld.global.v4.b32 { %r262, %r263, %r264, %r265 }, [ %rd69 + 0 ]; + // end inline asm + mov.b32 {%rs672, %rs673}, %r262; + mov.b32 {%rs674, %rs675}, %r263; + mov.b32 {%rs676, %rs677}, %r264; + mov.b32 {%rs678, %rs679}, %r265; + // begin inline asm + mov.u32 %r266, %r5; + mov.u32 %r267, %r5; + mov.u32 %r268, %r5; + mov.u32 %r269, %r5; + @%p2 ld.global.v4.b32 { %r266, %r267, %r268, %r269 }, [ %rd70 + 0 ]; + // end inline asm + mov.b32 {%rs680, %rs681}, %r266; + mov.b32 {%rs682, %rs683}, %r267; + mov.b32 {%rs684, %rs685}, %r268; + mov.b32 {%rs686, %rs687}, %r269; + // begin inline asm + mov.u32 %r270, %r5; + mov.u32 %r271, %r5; + mov.u32 %r272, %r5; + mov.u32 %r273, %r5; + @%p3 ld.global.v4.b32 { %r270, %r271, %r272, %r273 }, [ %rd71 + 0 ]; + // end inline asm + mov.b32 {%rs688, %rs689}, %r270; + mov.b32 {%rs690, %rs691}, %r271; + mov.b32 {%rs692, %rs693}, %r272; + mov.b32 {%rs694, %rs695}, %r273; + // begin inline asm + mov.u32 %r274, %r5; + mov.u32 %r275, %r5; + mov.u32 %r276, %r5; + mov.u32 %r277, %r5; + @%p4 ld.global.v4.b32 { %r274, %r275, %r276, %r277 }, [ %rd72 + 0 ]; + // end inline asm + mov.b32 {%rs696, %rs697}, %r274; + mov.b32 {%rs698, %rs699}, %r275; + mov.b32 {%rs700, %rs701}, %r276; + mov.b32 {%rs702, %rs703}, %r277; + // begin inline asm + mov.u32 %r278, %r5; + mov.u32 %r279, %r5; + mov.u32 %r280, %r5; + mov.u32 %r281, %r5; + @%p5 ld.global.v4.b32 { %r278, %r279, %r280, %r281 }, [ %rd73 + 0 ]; + // end inline asm + mov.b32 {%rs704, %rs705}, %r278; + mov.b32 {%rs706, %rs707}, %r279; + mov.b32 {%rs708, %rs709}, %r280; + mov.b32 {%rs710, %rs711}, %r281; + // begin inline asm + mov.u32 %r282, %r5; + mov.u32 %r283, %r5; + mov.u32 %r284, %r5; + mov.u32 %r285, %r5; + @%p6 ld.global.v4.b32 { %r282, %r283, %r284, %r285 }, [ %rd74 + 0 ]; + // end inline asm + mov.b32 {%rs712, %rs713}, %r282; + mov.b32 {%rs714, %rs715}, %r283; + mov.b32 {%rs716, %rs717}, %r284; + mov.b32 {%rs718, %rs719}, %r285; + // begin inline asm + mov.u32 %r286, %r5; + mov.u32 %r287, %r5; + mov.u32 %r288, %r5; + mov.u32 %r289, %r5; + @%p7 ld.global.v4.b32 { %r286, %r287, %r288, %r289 }, [ %rd75 + 0 ]; + // end inline asm + mov.b32 {%rs720, %rs721}, %r286; + mov.b32 {%rs722, %rs723}, %r287; + mov.b32 {%rs724, %rs725}, %r288; + mov.b32 {%rs726, %rs727}, %r289; + // begin inline asm + mov.u32 %r290, %r5; + mov.u32 %r291, %r5; + mov.u32 %r292, %r5; + mov.u32 %r293, %r5; + @%p8 ld.global.v4.b32 { %r290, %r291, %r292, %r293 }, [ %rd76 + 0 ]; + // end inline asm + mov.b32 {%rs728, %rs729}, %r290; + mov.b32 {%rs730, %rs731}, %r291; + mov.b32 {%rs732, %rs733}, %r292; + mov.b32 {%rs734, %rs735}, %r293; + // begin inline asm + mov.u32 %r294, %r5; + mov.u32 %r295, %r5; + mov.u32 %r296, %r5; + mov.u32 %r297, %r5; + @%p9 ld.global.v4.b32 { %r294, %r295, %r296, %r297 }, [ %rd77 + 0 ]; + // end inline asm + mov.b32 {%rs736, %rs737}, %r294; + mov.b32 {%rs738, %rs739}, %r295; + mov.b32 {%rs740, %rs741}, %r296; + mov.b32 {%rs742, %rs743}, %r297; + // begin inline asm + mov.u32 %r298, %r5; + mov.u32 %r299, %r5; + mov.u32 %r300, %r5; + mov.u32 %r301, %r5; + @%p10 ld.global.v4.b32 { %r298, %r299, %r300, %r301 }, [ %rd78 + 0 ]; + // end inline asm + mov.b32 {%rs744, %rs745}, %r298; + mov.b32 {%rs746, %rs747}, %r299; + mov.b32 {%rs748, %rs749}, %r300; + mov.b32 {%rs750, %rs751}, %r301; + // begin inline asm + mov.u32 %r302, %r5; + mov.u32 %r303, %r5; + mov.u32 %r304, %r5; + mov.u32 %r305, %r5; + @%p11 ld.global.v4.b32 { %r302, %r303, %r304, %r305 }, [ %rd79 + 0 ]; + // end inline asm + mov.b32 {%rs752, %rs753}, %r302; + mov.b32 {%rs754, %rs755}, %r303; + mov.b32 {%rs756, %rs757}, %r304; + mov.b32 {%rs758, %rs759}, %r305; + // begin inline asm + mov.u32 %r306, %r5; + mov.u32 %r307, %r5; + mov.u32 %r308, %r5; + mov.u32 %r309, %r5; + @%p39 ld.global.v4.b32 { %r306, %r307, %r308, %r309 }, [ %rd80 + 0 ]; + // end inline asm + mov.b32 {%rs760, %rs761}, %r306; + mov.b32 {%rs762, %rs763}, %r307; + mov.b32 {%rs764, %rs765}, %r308; + mov.b32 {%rs766, %rs767}, %r309; + // begin inline asm + mov.u32 %r310, %r5; + mov.u32 %r311, %r5; + mov.u32 %r312, %r5; + mov.u32 %r313, %r5; + @%p12 ld.global.v4.b32 { %r310, %r311, %r312, %r313 }, [ %rd81 + 0 ]; + // end inline asm + mov.b32 {%rs768, %rs769}, %r310; + mov.b32 {%rs770, %rs771}, %r311; + mov.b32 {%rs772, %rs773}, %r312; + mov.b32 {%rs774, %rs775}, %r313; + // begin inline asm + mov.u32 %r314, %r5; + mov.u32 %r315, %r5; + mov.u32 %r316, %r5; + mov.u32 %r317, %r5; + @%p13 ld.global.v4.b32 { %r314, %r315, %r316, %r317 }, [ %rd82 + 0 ]; + // end inline asm + mov.b32 {%rs776, %rs777}, %r314; + mov.b32 {%rs778, %rs779}, %r315; + mov.b32 {%rs780, %rs781}, %r316; + mov.b32 {%rs782, %rs783}, %r317; + // begin inline asm + mov.u32 %r318, %r5; + mov.u32 %r319, %r5; + mov.u32 %r320, %r5; + mov.u32 %r321, %r5; + @%p14 ld.global.v4.b32 { %r318, %r319, %r320, %r321 }, [ %rd83 + 0 ]; + // end inline asm + mov.b32 {%rs784, %rs785}, %r318; + mov.b32 {%rs786, %rs787}, %r319; + mov.b32 {%rs788, %rs789}, %r320; + mov.b32 {%rs790, %rs791}, %r321; + // begin inline asm + mov.u32 %r322, %r5; + mov.u32 %r323, %r5; + mov.u32 %r324, %r5; + mov.u32 %r325, %r5; + @%p15 ld.global.v4.b32 { %r322, %r323, %r324, %r325 }, [ %rd84 + 0 ]; + // end inline asm + mov.b32 {%rs792, %rs793}, %r322; + mov.b32 {%rs794, %rs795}, %r323; + mov.b32 {%rs796, %rs797}, %r324; + mov.b32 {%rs798, %rs799}, %r325; + // begin inline asm + mov.u32 %r326, %r5; + mov.u32 %r327, %r5; + mov.u32 %r328, %r5; + mov.u32 %r329, %r5; + @%p16 ld.global.v4.b32 { %r326, %r327, %r328, %r329 }, [ %rd85 + 0 ]; + // end inline asm + mov.b32 {%rs800, %rs801}, %r326; + mov.b32 {%rs802, %rs803}, %r327; + mov.b32 {%rs804, %rs805}, %r328; + mov.b32 {%rs806, %rs807}, %r329; + // begin inline asm + mov.u32 %r330, %r5; + mov.u32 %r331, %r5; + mov.u32 %r332, %r5; + mov.u32 %r333, %r5; + @%p17 ld.global.v4.b32 { %r330, %r331, %r332, %r333 }, [ %rd86 + 0 ]; + // end inline asm + mov.b32 {%rs808, %rs809}, %r330; + mov.b32 {%rs810, %rs811}, %r331; + mov.b32 {%rs812, %rs813}, %r332; + mov.b32 {%rs814, %rs815}, %r333; + // begin inline asm + mov.u32 %r334, %r5; + mov.u32 %r335, %r5; + mov.u32 %r336, %r5; + mov.u32 %r337, %r5; + @%p18 ld.global.v4.b32 { %r334, %r335, %r336, %r337 }, [ %rd87 + 0 ]; + // end inline asm + mov.b32 {%rs816, %rs817}, %r334; + mov.b32 {%rs818, %rs819}, %r335; + mov.b32 {%rs820, %rs821}, %r336; + mov.b32 {%rs822, %rs823}, %r337; + // begin inline asm + mov.u32 %r338, %r5; + mov.u32 %r339, %r5; + mov.u32 %r340, %r5; + mov.u32 %r341, %r5; + @%p40 ld.global.v4.b32 { %r338, %r339, %r340, %r341 }, [ %rd88 + 0 ]; + // end inline asm + mov.b32 {%rs824, %rs825}, %r338; + mov.b32 {%rs826, %rs827}, %r339; + mov.b32 {%rs828, %rs829}, %r340; + mov.b32 {%rs830, %rs831}, %r341; + // begin inline asm + mov.u32 %r342, %r5; + mov.u32 %r343, %r5; + mov.u32 %r344, %r5; + mov.u32 %r345, %r5; + @%p19 ld.global.v4.b32 { %r342, %r343, %r344, %r345 }, [ %rd89 + 0 ]; + // end inline asm + mov.b32 {%rs832, %rs833}, %r342; + mov.b32 {%rs834, %rs835}, %r343; + mov.b32 {%rs836, %rs837}, %r344; + mov.b32 {%rs838, %rs839}, %r345; + // begin inline asm + mov.u32 %r346, %r5; + mov.u32 %r347, %r5; + mov.u32 %r348, %r5; + mov.u32 %r349, %r5; + @%p20 ld.global.v4.b32 { %r346, %r347, %r348, %r349 }, [ %rd90 + 0 ]; + // end inline asm + mov.b32 {%rs840, %rs841}, %r346; + mov.b32 {%rs842, %rs843}, %r347; + mov.b32 {%rs844, %rs845}, %r348; + mov.b32 {%rs846, %rs847}, %r349; + // begin inline asm + mov.u32 %r350, %r5; + mov.u32 %r351, %r5; + mov.u32 %r352, %r5; + mov.u32 %r353, %r5; + @%p21 ld.global.v4.b32 { %r350, %r351, %r352, %r353 }, [ %rd91 + 0 ]; + // end inline asm + mov.b32 {%rs848, %rs849}, %r350; + mov.b32 {%rs850, %rs851}, %r351; + mov.b32 {%rs852, %rs853}, %r352; + mov.b32 {%rs854, %rs855}, %r353; + // begin inline asm + mov.u32 %r354, %r5; + mov.u32 %r355, %r5; + mov.u32 %r356, %r5; + mov.u32 %r357, %r5; + @%p22 ld.global.v4.b32 { %r354, %r355, %r356, %r357 }, [ %rd92 + 0 ]; + // end inline asm + mov.b32 {%rs856, %rs857}, %r354; + mov.b32 {%rs858, %rs859}, %r355; + mov.b32 {%rs860, %rs861}, %r356; + mov.b32 {%rs862, %rs863}, %r357; + // begin inline asm + mov.u32 %r358, %r5; + mov.u32 %r359, %r5; + mov.u32 %r360, %r5; + mov.u32 %r361, %r5; + @%p23 ld.global.v4.b32 { %r358, %r359, %r360, %r361 }, [ %rd93 + 0 ]; + // end inline asm + mov.b32 {%rs864, %rs865}, %r358; + mov.b32 {%rs866, %rs867}, %r359; + mov.b32 {%rs868, %rs869}, %r360; + mov.b32 {%rs870, %rs871}, %r361; + // begin inline asm + mov.u32 %r362, %r5; + mov.u32 %r363, %r5; + mov.u32 %r364, %r5; + mov.u32 %r365, %r5; + @%p24 ld.global.v4.b32 { %r362, %r363, %r364, %r365 }, [ %rd94 + 0 ]; + // end inline asm + mov.b32 {%rs872, %rs873}, %r362; + mov.b32 {%rs874, %rs875}, %r363; + mov.b32 {%rs876, %rs877}, %r364; + mov.b32 {%rs878, %rs879}, %r365; + // begin inline asm + mov.u32 %r366, %r5; + mov.u32 %r367, %r5; + mov.u32 %r368, %r5; + mov.u32 %r369, %r5; + @%p25 ld.global.v4.b32 { %r366, %r367, %r368, %r369 }, [ %rd95 + 0 ]; + // end inline asm + mov.b32 {%rs880, %rs881}, %r366; + mov.b32 {%rs882, %rs883}, %r367; + mov.b32 {%rs884, %rs885}, %r368; + mov.b32 {%rs886, %rs887}, %r369; + // begin inline asm + mov.u32 %r370, %r5; + mov.u32 %r371, %r5; + mov.u32 %r372, %r5; + mov.u32 %r373, %r5; + @%p26 ld.global.v4.b32 { %r370, %r371, %r372, %r373 }, [ %rd96 + 0 ]; + // end inline asm + mov.b32 {%rs888, %rs889}, %r370; + mov.b32 {%rs890, %rs891}, %r371; + mov.b32 {%rs892, %rs893}, %r372; + mov.b32 {%rs894, %rs895}, %r373; + // begin inline asm + mov.u32 %r374, %r5; + mov.u32 %r375, %r5; + mov.u32 %r376, %r5; + mov.u32 %r377, %r5; + @%p27 ld.global.v4.b32 { %r374, %r375, %r376, %r377 }, [ %rd97 + 0 ]; + // end inline asm + mov.b32 {%rs896, %rs897}, %r374; + mov.b32 {%rs898, %rs899}, %r375; + mov.b32 {%rs900, %rs901}, %r376; + mov.b32 {%rs902, %rs903}, %r377; + // begin inline asm + mov.u32 %r378, %r5; + mov.u32 %r379, %r5; + mov.u32 %r380, %r5; + mov.u32 %r381, %r5; + @%p28 ld.global.v4.b32 { %r378, %r379, %r380, %r381 }, [ %rd98 + 0 ]; + // end inline asm + mov.b32 {%rs904, %rs905}, %r378; + mov.b32 {%rs906, %rs907}, %r379; + mov.b32 {%rs908, %rs909}, %r380; + mov.b32 {%rs910, %rs911}, %r381; + // begin inline asm + mov.u32 %r382, %r5; + mov.u32 %r383, %r5; + mov.u32 %r384, %r5; + mov.u32 %r385, %r5; + @%p29 ld.global.v4.b32 { %r382, %r383, %r384, %r385 }, [ %rd99 + 0 ]; + // end inline asm + mov.b32 {%rs912, %rs913}, %r382; + mov.b32 {%rs914, %rs915}, %r383; + mov.b32 {%rs916, %rs917}, %r384; + mov.b32 {%rs918, %rs919}, %r385; + // begin inline asm + mov.u32 %r386, %r5; + mov.u32 %r387, %r5; + mov.u32 %r388, %r5; + mov.u32 %r389, %r5; + @%p30 ld.global.v4.b32 { %r386, %r387, %r388, %r389 }, [ %rd100 + 0 ]; + // end inline asm + mov.b32 {%rs920, %rs921}, %r386; + mov.b32 {%rs922, %rs923}, %r387; + mov.b32 {%rs924, %rs925}, %r388; + mov.b32 {%rs926, %rs927}, %r389; + // begin inline asm + mov.u32 %r390, %r5; + mov.u32 %r391, %r5; + mov.u32 %r392, %r5; + mov.u32 %r393, %r5; + @%p31 ld.global.v4.b32 { %r390, %r391, %r392, %r393 }, [ %rd101 + 0 ]; + // end inline asm + mov.b32 {%rs928, %rs929}, %r390; + mov.b32 {%rs930, %rs931}, %r391; + mov.b32 {%rs932, %rs933}, %r392; + mov.b32 {%rs934, %rs935}, %r393; + // begin inline asm + mov.u32 %r394, %r5; + mov.u32 %r395, %r5; + mov.u32 %r396, %r5; + mov.u32 %r397, %r5; + @%p32 ld.global.v4.b32 { %r394, %r395, %r396, %r397 }, [ %rd102 + 0 ]; + // end inline asm + mov.b32 {%rs936, %rs937}, %r394; + mov.b32 {%rs938, %rs939}, %r395; + mov.b32 {%rs940, %rs941}, %r396; + mov.b32 {%rs942, %rs943}, %r397; + // begin inline asm + mov.u32 %r398, %r5; + mov.u32 %r399, %r5; + mov.u32 %r400, %r5; + mov.u32 %r401, %r5; + @%p33 ld.global.v4.b32 { %r398, %r399, %r400, %r401 }, [ %rd103 + 0 ]; + // end inline asm + mov.b32 {%rs944, %rs945}, %r398; + mov.b32 {%rs946, %rs947}, %r399; + mov.b32 {%rs948, %rs949}, %r400; + mov.b32 {%rs950, %rs951}, %r401; + // begin inline asm + mov.u32 %r402, %r5; + mov.u32 %r403, %r5; + mov.u32 %r404, %r5; + mov.u32 %r405, %r5; + @%p34 ld.global.v4.b32 { %r402, %r403, %r404, %r405 }, [ %rd104 + 0 ]; + // end inline asm + mov.b32 {%rs952, %rs953}, %r402; + mov.b32 {%rs954, %rs955}, %r403; + mov.b32 {%rs956, %rs957}, %r404; + mov.b32 {%rs958, %rs959}, %r405; + // begin inline asm + mov.u32 %r406, %r5; + mov.u32 %r407, %r5; + mov.u32 %r408, %r5; + mov.u32 %r409, %r5; + @%p41 ld.global.v4.b32 { %r406, %r407, %r408, %r409 }, [ %rd105 + 0 ]; + // end inline asm + mov.b32 {%rs960, %rs961}, %r406; + mov.b32 {%rs962, %rs963}, %r407; + mov.b32 {%rs964, %rs965}, %r408; + mov.b32 {%rs966, %rs967}, %r409; + // begin inline asm + mov.u32 %r410, %r5; + mov.u32 %r411, %r5; + mov.u32 %r412, %r5; + mov.u32 %r413, %r5; + @%p42 ld.global.v4.b32 { %r410, %r411, %r412, %r413 }, [ %rd106 + 0 ]; + // end inline asm + mov.b32 {%rs968, %rs969}, %r410; + mov.b32 {%rs970, %rs971}, %r411; + mov.b32 {%rs972, %rs973}, %r412; + mov.b32 {%rs974, %rs975}, %r413; + // begin inline asm + mov.u32 %r414, %r5; + mov.u32 %r415, %r5; + mov.u32 %r416, %r5; + mov.u32 %r417, %r5; + @%p43 ld.global.v4.b32 { %r414, %r415, %r416, %r417 }, [ %rd107 + 0 ]; + // end inline asm + mov.b32 {%rs976, %rs977}, %r414; + mov.b32 {%rs978, %rs979}, %r415; + mov.b32 {%rs980, %rs981}, %r416; + mov.b32 {%rs982, %rs983}, %r417; + // begin inline asm + mov.u32 %r418, %r5; + mov.u32 %r419, %r5; + mov.u32 %r420, %r5; + mov.u32 %r421, %r5; + @%p44 ld.global.v4.b32 { %r418, %r419, %r420, %r421 }, [ %rd108 + 0 ]; + // end inline asm + mov.b32 {%rs984, %rs985}, %r418; + mov.b32 {%rs986, %rs987}, %r419; + mov.b32 {%rs988, %rs989}, %r420; + mov.b32 {%rs990, %rs991}, %r421; + // begin inline asm + mov.u32 %r422, %r5; + mov.u32 %r423, %r5; + mov.u32 %r424, %r5; + mov.u32 %r425, %r5; + @%p45 ld.global.v4.b32 { %r422, %r423, %r424, %r425 }, [ %rd109 + 0 ]; + // end inline asm + mov.b32 {%rs992, %rs993}, %r422; + mov.b32 {%rs994, %rs995}, %r423; + mov.b32 {%rs996, %rs997}, %r424; + mov.b32 {%rs998, %rs999}, %r425; + // begin inline asm + mov.u32 %r426, %r5; + mov.u32 %r427, %r5; + mov.u32 %r428, %r5; + mov.u32 %r429, %r5; + @%p46 ld.global.v4.b32 { %r426, %r427, %r428, %r429 }, [ %rd110 + 0 ]; + // end inline asm + mov.b32 {%rs1000, %rs1001}, %r426; + mov.b32 {%rs1002, %rs1003}, %r427; + mov.b32 {%rs1004, %rs1005}, %r428; + mov.b32 {%rs1006, %rs1007}, %r429; + // begin inline asm + mov.u32 %r430, %r5; + mov.u32 %r431, %r5; + mov.u32 %r432, %r5; + mov.u32 %r433, %r5; + @%p47 ld.global.v4.b32 { %r430, %r431, %r432, %r433 }, [ %rd111 + 0 ]; + // end inline asm + mov.b32 {%rs1008, %rs1009}, %r430; + mov.b32 {%rs1010, %rs1011}, %r431; + mov.b32 {%rs1012, %rs1013}, %r432; + mov.b32 {%rs1014, %rs1015}, %r433; + // begin inline asm + mov.u32 %r434, %r5; + mov.u32 %r435, %r5; + mov.u32 %r436, %r5; + mov.u32 %r437, %r5; + @%p48 ld.global.v4.b32 { %r434, %r435, %r436, %r437 }, [ %rd112 + 0 ]; + // end inline asm + mov.b32 {%rs1016, %rs1017}, %r434; + mov.b32 {%rs1018, %rs1019}, %r435; + mov.b32 {%rs1020, %rs1021}, %r436; + mov.b32 {%rs1022, %rs1023}, %r437; + // begin inline asm + mov.u32 %r438, %r5; + mov.u32 %r439, %r5; + mov.u32 %r440, %r5; + mov.u32 %r441, %r5; + @%p49 ld.global.v4.b32 { %r438, %r439, %r440, %r441 }, [ %rd113 + 0 ]; + // end inline asm + mov.b32 {%rs1024, %rs1025}, %r438; + mov.b32 {%rs1026, %rs1027}, %r439; + mov.b32 {%rs1028, %rs1029}, %r440; + mov.b32 {%rs1030, %rs1031}, %r441; + // begin inline asm + mov.u32 %r442, %r5; + mov.u32 %r443, %r5; + mov.u32 %r444, %r5; + mov.u32 %r445, %r5; + @%p50 ld.global.v4.b32 { %r442, %r443, %r444, %r445 }, [ %rd114 + 0 ]; + // end inline asm + mov.b32 {%rs1032, %rs1033}, %r442; + mov.b32 {%rs1034, %rs1035}, %r443; + mov.b32 {%rs1036, %rs1037}, %r444; + mov.b32 {%rs1038, %rs1039}, %r445; + // begin inline asm + mov.u32 %r446, %r5; + mov.u32 %r447, %r5; + mov.u32 %r448, %r5; + mov.u32 %r449, %r5; + @%p51 ld.global.v4.b32 { %r446, %r447, %r448, %r449 }, [ %rd115 + 0 ]; + // end inline asm + mov.b32 {%rs1040, %rs1041}, %r446; + mov.b32 {%rs1042, %rs1043}, %r447; + mov.b32 {%rs1044, %rs1045}, %r448; + mov.b32 {%rs1046, %rs1047}, %r449; + // begin inline asm + mov.u32 %r450, %r5; + mov.u32 %r451, %r5; + mov.u32 %r452, %r5; + mov.u32 %r453, %r5; + @%p52 ld.global.v4.b32 { %r450, %r451, %r452, %r453 }, [ %rd116 + 0 ]; + // end inline asm + mov.b32 {%rs1048, %rs1049}, %r450; + mov.b32 {%rs1050, %rs1051}, %r451; + mov.b32 {%rs1052, %rs1053}, %r452; + mov.b32 {%rs1054, %rs1055}, %r453; + // begin inline asm + mov.u32 %r454, %r5; + mov.u32 %r455, %r5; + mov.u32 %r456, %r5; + mov.u32 %r457, %r5; + @%p53 ld.global.v4.b32 { %r454, %r455, %r456, %r457 }, [ %rd117 + 0 ]; + // end inline asm + mov.b32 {%rs1056, %rs1057}, %r454; + mov.b32 {%rs1058, %rs1059}, %r455; + mov.b32 {%rs1060, %rs1061}, %r456; + mov.b32 {%rs1062, %rs1063}, %r457; + // begin inline asm + mov.u32 %r458, %r5; + mov.u32 %r459, %r5; + mov.u32 %r460, %r5; + mov.u32 %r461, %r5; + @%p54 ld.global.v4.b32 { %r458, %r459, %r460, %r461 }, [ %rd118 + 0 ]; + // end inline asm + mov.b32 {%rs1064, %rs1065}, %r458; + mov.b32 {%rs1066, %rs1067}, %r459; + mov.b32 {%rs1068, %rs1069}, %r460; + mov.b32 {%rs1070, %rs1071}, %r461; + // begin inline asm + mov.u32 %r462, %r5; + mov.u32 %r463, %r5; + mov.u32 %r464, %r5; + mov.u32 %r465, %r5; + @%p55 ld.global.v4.b32 { %r462, %r463, %r464, %r465 }, [ %rd119 + 0 ]; + // end inline asm + mov.b32 {%rs1072, %rs1073}, %r462; + mov.b32 {%rs1074, %rs1075}, %r463; + mov.b32 {%rs1076, %rs1077}, %r464; + mov.b32 {%rs1078, %rs1079}, %r465; + // begin inline asm + mov.u32 %r466, %r5; + mov.u32 %r467, %r5; + mov.u32 %r468, %r5; + mov.u32 %r469, %r5; + @%p56 ld.global.v4.b32 { %r466, %r467, %r468, %r469 }, [ %rd120 + 0 ]; + // end inline asm + mov.b32 {%rs1080, %rs1081}, %r466; + mov.b32 {%rs1082, %rs1083}, %r467; + mov.b32 {%rs1084, %rs1085}, %r468; + mov.b32 {%rs1086, %rs1087}, %r469; + // begin inline asm + mov.u32 %r470, %r5; + mov.u32 %r471, %r5; + mov.u32 %r472, %r5; + mov.u32 %r473, %r5; + @%p57 ld.global.v4.b32 { %r470, %r471, %r472, %r473 }, [ %rd121 + 0 ]; + // end inline asm + mov.b32 {%rs1088, %rs1089}, %r470; + mov.b32 {%rs1090, %rs1091}, %r471; + mov.b32 {%rs1092, %rs1093}, %r472; + mov.b32 {%rs1094, %rs1095}, %r473; + // begin inline asm + mov.u32 %r474, %r5; + mov.u32 %r475, %r5; + mov.u32 %r476, %r5; + mov.u32 %r477, %r5; + @%p58 ld.global.v4.b32 { %r474, %r475, %r476, %r477 }, [ %rd122 + 0 ]; + // end inline asm + mov.b32 {%rs1096, %rs1097}, %r474; + mov.b32 {%rs1098, %rs1099}, %r475; + mov.b32 {%rs1100, %rs1101}, %r476; + mov.b32 {%rs1102, %rs1103}, %r477; + // begin inline asm + mov.u32 %r478, %r5; + mov.u32 %r479, %r5; + mov.u32 %r480, %r5; + mov.u32 %r481, %r5; + @%p59 ld.global.v4.b32 { %r478, %r479, %r480, %r481 }, [ %rd123 + 0 ]; + // end inline asm + mov.b32 {%rs1104, %rs1105}, %r478; + mov.b32 {%rs1106, %rs1107}, %r479; + mov.b32 {%rs1108, %rs1109}, %r480; + mov.b32 {%rs1110, %rs1111}, %r481; + // begin inline asm + mov.u32 %r482, %r5; + mov.u32 %r483, %r5; + mov.u32 %r484, %r5; + mov.u32 %r485, %r5; + @%p60 ld.global.v4.b32 { %r482, %r483, %r484, %r485 }, [ %rd124 + 0 ]; + // end inline asm + mov.b32 {%rs1112, %rs1113}, %r482; + mov.b32 {%rs1114, %rs1115}, %r483; + mov.b32 {%rs1116, %rs1117}, %r484; + mov.b32 {%rs1118, %rs1119}, %r485; + // begin inline asm + mov.u32 %r486, %r5; + mov.u32 %r487, %r5; + mov.u32 %r488, %r5; + mov.u32 %r489, %r5; + @%p61 ld.global.v4.b32 { %r486, %r487, %r488, %r489 }, [ %rd125 + 0 ]; + // end inline asm + mov.b32 {%rs1120, %rs1121}, %r486; + mov.b32 {%rs1122, %rs1123}, %r487; + mov.b32 {%rs1124, %rs1125}, %r488; + mov.b32 {%rs1126, %rs1127}, %r489; + // begin inline asm + mov.u32 %r490, %r5; + mov.u32 %r491, %r5; + mov.u32 %r492, %r5; + mov.u32 %r493, %r5; + @%p62 ld.global.v4.b32 { %r490, %r491, %r492, %r493 }, [ %rd126 + 0 ]; + // end inline asm + mov.b32 {%rs1128, %rs1129}, %r490; + mov.b32 {%rs1130, %rs1131}, %r491; + mov.b32 {%rs1132, %rs1133}, %r492; + mov.b32 {%rs1134, %rs1135}, %r493; + // begin inline asm + mov.u32 %r494, %r5; + mov.u32 %r495, %r5; + mov.u32 %r496, %r5; + mov.u32 %r497, %r5; + @%p63 ld.global.v4.b32 { %r494, %r495, %r496, %r497 }, [ %rd127 + 0 ]; + // end inline asm + mov.b32 {%rs1136, %rs1137}, %r494; + mov.b32 {%rs1138, %rs1139}, %r495; + mov.b32 {%rs1140, %rs1141}, %r496; + mov.b32 {%rs1142, %rs1143}, %r497; + // begin inline asm + mov.u32 %r498, %r5; + mov.u32 %r499, %r5; + mov.u32 %r500, %r5; + mov.u32 %r501, %r5; + @%p64 ld.global.v4.b32 { %r498, %r499, %r500, %r501 }, [ %rd128 + 0 ]; + // end inline asm + mov.b32 {%rs1144, %rs1145}, %r498; + mov.b32 {%rs1146, %rs1147}, %r499; + mov.b32 {%rs1148, %rs1149}, %r500; + mov.b32 {%rs1150, %rs1151}, %r501; + // begin inline asm + mov.u32 %r502, %r5; + mov.u32 %r503, %r5; + mov.u32 %r504, %r5; + mov.u32 %r505, %r5; + @%p65 ld.global.v4.b32 { %r502, %r503, %r504, %r505 }, [ %rd129 + 0 ]; + // end inline asm + mov.b32 {%rs1152, %rs1153}, %r502; + mov.b32 {%rs1154, %rs1155}, %r503; + mov.b32 {%rs1156, %rs1157}, %r504; + mov.b32 {%rs1158, %rs1159}, %r505; + // begin inline asm + mov.u32 %r506, %r5; + mov.u32 %r507, %r5; + mov.u32 %r508, %r5; + mov.u32 %r509, %r5; + @%p66 ld.global.v4.b32 { %r506, %r507, %r508, %r509 }, [ %rd130 + 0 ]; + // end inline asm + mov.b32 {%rs1160, %rs1161}, %r506; + mov.b32 {%rs1162, %rs1163}, %r507; + mov.b32 {%rs1164, %rs1165}, %r508; + mov.b32 {%rs1166, %rs1167}, %r509; + // begin inline asm + mov.u32 %r510, %r5; + mov.u32 %r511, %r5; + mov.u32 %r512, %r5; + mov.u32 %r513, %r5; + @%p67 ld.global.v4.b32 { %r510, %r511, %r512, %r513 }, [ %rd131 + 0 ]; + // end inline asm + mov.b32 {%rs1168, %rs1169}, %r510; + mov.b32 {%rs1170, %rs1171}, %r511; + mov.b32 {%rs1172, %rs1173}, %r512; + mov.b32 {%rs1174, %rs1175}, %r513; + // begin inline asm + mov.u32 %r514, %r5; + mov.u32 %r515, %r5; + mov.u32 %r516, %r5; + mov.u32 %r517, %r5; + @%p68 ld.global.v4.b32 { %r514, %r515, %r516, %r517 }, [ %rd132 + 0 ]; + // end inline asm + mov.b32 {%rs1176, %rs1177}, %r514; + mov.b32 {%rs1178, %rs1179}, %r515; + mov.b32 {%rs1180, %rs1181}, %r516; + mov.b32 {%rs1182, %rs1183}, %r517; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1395, %rs672; + cvt.f32.bf16 %r1396, %rs673; + cvt.f32.bf16 %r1397, %rs674; + cvt.f32.bf16 %r1398, %rs675; + cvt.f32.bf16 %r1399, %rs676; + cvt.f32.bf16 %r1400, %rs677; + cvt.f32.bf16 %r1401, %rs678; + cvt.f32.bf16 %r1402, %rs679; + cvt.f32.bf16 %r1403, %rs680; + cvt.f32.bf16 %r1404, %rs681; + cvt.f32.bf16 %r1405, %rs682; + cvt.f32.bf16 %r1406, %rs683; + cvt.f32.bf16 %r1407, %rs684; + cvt.f32.bf16 %r1408, %rs685; + cvt.f32.bf16 %r1409, %rs686; + cvt.f32.bf16 %r1410, %rs687; + cvt.f32.bf16 %r1411, %rs688; + cvt.f32.bf16 %r1412, %rs689; + cvt.f32.bf16 %r1413, %rs690; + cvt.f32.bf16 %r1414, %rs691; + cvt.f32.bf16 %r1415, %rs692; + cvt.f32.bf16 %r1416, %rs693; + cvt.f32.bf16 %r1417, %rs694; + cvt.f32.bf16 %r1418, %rs695; + cvt.f32.bf16 %r1419, %rs696; + cvt.f32.bf16 %r1420, %rs697; + cvt.f32.bf16 %r1421, %rs698; + cvt.f32.bf16 %r1422, %rs699; + cvt.f32.bf16 %r1423, %rs700; + cvt.f32.bf16 %r1424, %rs701; + cvt.f32.bf16 %r1425, %rs702; + cvt.f32.bf16 %r1426, %rs703; + cvt.f32.bf16 %r1427, %rs704; + cvt.f32.bf16 %r1428, %rs705; + cvt.f32.bf16 %r1429, %rs706; + cvt.f32.bf16 %r1430, %rs707; + cvt.f32.bf16 %r1431, %rs708; + cvt.f32.bf16 %r1432, %rs709; + cvt.f32.bf16 %r1433, %rs710; + cvt.f32.bf16 %r1434, %rs711; + cvt.f32.bf16 %r1435, %rs712; + cvt.f32.bf16 %r1436, %rs713; + cvt.f32.bf16 %r1437, %rs714; + cvt.f32.bf16 %r1438, %rs715; + cvt.f32.bf16 %r1439, %rs716; + cvt.f32.bf16 %r1440, %rs717; + cvt.f32.bf16 %r1441, %rs718; + cvt.f32.bf16 %r1442, %rs719; + cvt.f32.bf16 %r1443, %rs720; + cvt.f32.bf16 %r1444, %rs721; + cvt.f32.bf16 %r1445, %rs722; + cvt.f32.bf16 %r1446, %rs723; + cvt.f32.bf16 %r1447, %rs724; + cvt.f32.bf16 %r1448, %rs725; + cvt.f32.bf16 %r1449, %rs726; + cvt.f32.bf16 %r1450, %rs727; + cvt.f32.bf16 %r1451, %rs728; + cvt.f32.bf16 %r1452, %rs729; + cvt.f32.bf16 %r1453, %rs730; + cvt.f32.bf16 %r1454, %rs731; + cvt.f32.bf16 %r1455, %rs732; + cvt.f32.bf16 %r1456, %rs733; + cvt.f32.bf16 %r1457, %rs734; + cvt.f32.bf16 %r1458, %rs735; + cvt.f32.bf16 %r1459, %rs736; + cvt.f32.bf16 %r1460, %rs737; + cvt.f32.bf16 %r1461, %rs738; + cvt.f32.bf16 %r1462, %rs739; + cvt.f32.bf16 %r1463, %rs740; + cvt.f32.bf16 %r1464, %rs741; + cvt.f32.bf16 %r1465, %rs742; + cvt.f32.bf16 %r1466, %rs743; + cvt.f32.bf16 %r1467, %rs744; + cvt.f32.bf16 %r1468, %rs745; + cvt.f32.bf16 %r1469, %rs746; + cvt.f32.bf16 %r1470, %rs747; + cvt.f32.bf16 %r1471, %rs748; + cvt.f32.bf16 %r1472, %rs749; + cvt.f32.bf16 %r1473, %rs750; + cvt.f32.bf16 %r1474, %rs751; + cvt.f32.bf16 %r1475, %rs752; + cvt.f32.bf16 %r1476, %rs753; + cvt.f32.bf16 %r1477, %rs754; + cvt.f32.bf16 %r1478, %rs755; + cvt.f32.bf16 %r1479, %rs756; + cvt.f32.bf16 %r1480, %rs757; + cvt.f32.bf16 %r1481, %rs758; + cvt.f32.bf16 %r1482, %rs759; + cvt.f32.bf16 %r1483, %rs760; + cvt.f32.bf16 %r1484, %rs761; + cvt.f32.bf16 %r1485, %rs762; + cvt.f32.bf16 %r1486, %rs763; + cvt.f32.bf16 %r1487, %rs764; + cvt.f32.bf16 %r1488, %rs765; + cvt.f32.bf16 %r1489, %rs766; + cvt.f32.bf16 %r1490, %rs767; + cvt.f32.bf16 %r1491, %rs768; + cvt.f32.bf16 %r1492, %rs769; + cvt.f32.bf16 %r1493, %rs770; + cvt.f32.bf16 %r1494, %rs771; + cvt.f32.bf16 %r1495, %rs772; + cvt.f32.bf16 %r1496, %rs773; + cvt.f32.bf16 %r1497, %rs774; + cvt.f32.bf16 %r1498, %rs775; + cvt.f32.bf16 %r1499, %rs776; + cvt.f32.bf16 %r1500, %rs777; + cvt.f32.bf16 %r1501, %rs778; + cvt.f32.bf16 %r1502, %rs779; + cvt.f32.bf16 %r1503, %rs780; + cvt.f32.bf16 %r1504, %rs781; + cvt.f32.bf16 %r1505, %rs782; + cvt.f32.bf16 %r1506, %rs783; + cvt.f32.bf16 %r1507, %rs784; + cvt.f32.bf16 %r1508, %rs785; + cvt.f32.bf16 %r1509, %rs786; + cvt.f32.bf16 %r1510, %rs787; + cvt.f32.bf16 %r1511, %rs788; + cvt.f32.bf16 %r1512, %rs789; + cvt.f32.bf16 %r1513, %rs790; + cvt.f32.bf16 %r1514, %rs791; + cvt.f32.bf16 %r1515, %rs792; + cvt.f32.bf16 %r1516, %rs793; + cvt.f32.bf16 %r1517, %rs794; + cvt.f32.bf16 %r1518, %rs795; + cvt.f32.bf16 %r1519, %rs796; + cvt.f32.bf16 %r1520, %rs797; + cvt.f32.bf16 %r1521, %rs798; + cvt.f32.bf16 %r1522, %rs799; + cvt.f32.bf16 %r1523, %rs800; + cvt.f32.bf16 %r1524, %rs801; + cvt.f32.bf16 %r1525, %rs802; + cvt.f32.bf16 %r1526, %rs803; + cvt.f32.bf16 %r1527, %rs804; + cvt.f32.bf16 %r1528, %rs805; + cvt.f32.bf16 %r1529, %rs806; + cvt.f32.bf16 %r1530, %rs807; + cvt.f32.bf16 %r1531, %rs808; + cvt.f32.bf16 %r1532, %rs809; + cvt.f32.bf16 %r1533, %rs810; + cvt.f32.bf16 %r1534, %rs811; + cvt.f32.bf16 %r1535, %rs812; + cvt.f32.bf16 %r1536, %rs813; + cvt.f32.bf16 %r1537, %rs814; + cvt.f32.bf16 %r1538, %rs815; + cvt.f32.bf16 %r1539, %rs816; + cvt.f32.bf16 %r1540, %rs817; + cvt.f32.bf16 %r1541, %rs818; + cvt.f32.bf16 %r1542, %rs819; + cvt.f32.bf16 %r1543, %rs820; + cvt.f32.bf16 %r1544, %rs821; + cvt.f32.bf16 %r1545, %rs822; + cvt.f32.bf16 %r1546, %rs823; + cvt.f32.bf16 %r1547, %rs824; + cvt.f32.bf16 %r1548, %rs825; + cvt.f32.bf16 %r1549, %rs826; + cvt.f32.bf16 %r1550, %rs827; + cvt.f32.bf16 %r1551, %rs828; + cvt.f32.bf16 %r1552, %rs829; + cvt.f32.bf16 %r1553, %rs830; + cvt.f32.bf16 %r1554, %rs831; + cvt.f32.bf16 %r1555, %rs832; + cvt.f32.bf16 %r1556, %rs833; + cvt.f32.bf16 %r1557, %rs834; + cvt.f32.bf16 %r1558, %rs835; + cvt.f32.bf16 %r1559, %rs836; + cvt.f32.bf16 %r1560, %rs837; + cvt.f32.bf16 %r1561, %rs838; + cvt.f32.bf16 %r1562, %rs839; + cvt.f32.bf16 %r1563, %rs840; + cvt.f32.bf16 %r1564, %rs841; + cvt.f32.bf16 %r1565, %rs842; + cvt.f32.bf16 %r1566, %rs843; + cvt.f32.bf16 %r1567, %rs844; + cvt.f32.bf16 %r1568, %rs845; + cvt.f32.bf16 %r1569, %rs846; + cvt.f32.bf16 %r1570, %rs847; + cvt.f32.bf16 %r1571, %rs848; + cvt.f32.bf16 %r1572, %rs849; + cvt.f32.bf16 %r1573, %rs850; + cvt.f32.bf16 %r1574, %rs851; + cvt.f32.bf16 %r1575, %rs852; + cvt.f32.bf16 %r1576, %rs853; + cvt.f32.bf16 %r1577, %rs854; + cvt.f32.bf16 %r1578, %rs855; + cvt.f32.bf16 %r1579, %rs856; + cvt.f32.bf16 %r1580, %rs857; + cvt.f32.bf16 %r1581, %rs858; + cvt.f32.bf16 %r1582, %rs859; + cvt.f32.bf16 %r1583, %rs860; + cvt.f32.bf16 %r1584, %rs861; + cvt.f32.bf16 %r1585, %rs862; + cvt.f32.bf16 %r1586, %rs863; + cvt.f32.bf16 %r1587, %rs864; + cvt.f32.bf16 %r1588, %rs865; + cvt.f32.bf16 %r1589, %rs866; + cvt.f32.bf16 %r1590, %rs867; + cvt.f32.bf16 %r1591, %rs868; + cvt.f32.bf16 %r1592, %rs869; + cvt.f32.bf16 %r1593, %rs870; + cvt.f32.bf16 %r1594, %rs871; + cvt.f32.bf16 %r1595, %rs872; + cvt.f32.bf16 %r1596, %rs873; + cvt.f32.bf16 %r1597, %rs874; + cvt.f32.bf16 %r1598, %rs875; + cvt.f32.bf16 %r1599, %rs876; + cvt.f32.bf16 %r1600, %rs877; + cvt.f32.bf16 %r1601, %rs878; + cvt.f32.bf16 %r1602, %rs879; + cvt.f32.bf16 %r1603, %rs880; + cvt.f32.bf16 %r1604, %rs881; + cvt.f32.bf16 %r1605, %rs882; + cvt.f32.bf16 %r1606, %rs883; + cvt.f32.bf16 %r1607, %rs884; + cvt.f32.bf16 %r1608, %rs885; + cvt.f32.bf16 %r1609, %rs886; + cvt.f32.bf16 %r1610, %rs887; + cvt.f32.bf16 %r1611, %rs888; + cvt.f32.bf16 %r1612, %rs889; + cvt.f32.bf16 %r1613, %rs890; + cvt.f32.bf16 %r1614, %rs891; + cvt.f32.bf16 %r1615, %rs892; + cvt.f32.bf16 %r1616, %rs893; + cvt.f32.bf16 %r1617, %rs894; + cvt.f32.bf16 %r1618, %rs895; + cvt.f32.bf16 %r1619, %rs896; + cvt.f32.bf16 %r1620, %rs897; + cvt.f32.bf16 %r1621, %rs898; + cvt.f32.bf16 %r1622, %rs899; + cvt.f32.bf16 %r1623, %rs900; + cvt.f32.bf16 %r1624, %rs901; + cvt.f32.bf16 %r1625, %rs902; + cvt.f32.bf16 %r1626, %rs903; + cvt.f32.bf16 %r1627, %rs904; + cvt.f32.bf16 %r1628, %rs905; + cvt.f32.bf16 %r1629, %rs906; + cvt.f32.bf16 %r1630, %rs907; + cvt.f32.bf16 %r1631, %rs908; + cvt.f32.bf16 %r1632, %rs909; + cvt.f32.bf16 %r1633, %rs910; + cvt.f32.bf16 %r1634, %rs911; + cvt.f32.bf16 %r1635, %rs912; + cvt.f32.bf16 %r1636, %rs913; + cvt.f32.bf16 %r1637, %rs914; + cvt.f32.bf16 %r1638, %rs915; + cvt.f32.bf16 %r1639, %rs916; + cvt.f32.bf16 %r1640, %rs917; + cvt.f32.bf16 %r1641, %rs918; + cvt.f32.bf16 %r1642, %rs919; + cvt.f32.bf16 %r1643, %rs920; + cvt.f32.bf16 %r1644, %rs921; + cvt.f32.bf16 %r1645, %rs922; + cvt.f32.bf16 %r1646, %rs923; + cvt.f32.bf16 %r1647, %rs924; + cvt.f32.bf16 %r1648, %rs925; + cvt.f32.bf16 %r1649, %rs926; + cvt.f32.bf16 %r1650, %rs927; + cvt.f32.bf16 %r1651, %rs928; + cvt.f32.bf16 %r1652, %rs929; + cvt.f32.bf16 %r1653, %rs930; + cvt.f32.bf16 %r1654, %rs931; + cvt.f32.bf16 %r1655, %rs932; + cvt.f32.bf16 %r1656, %rs933; + cvt.f32.bf16 %r1657, %rs934; + cvt.f32.bf16 %r1658, %rs935; + cvt.f32.bf16 %r1659, %rs936; + cvt.f32.bf16 %r1660, %rs937; + cvt.f32.bf16 %r1661, %rs938; + cvt.f32.bf16 %r1662, %rs939; + cvt.f32.bf16 %r1663, %rs940; + cvt.f32.bf16 %r1664, %rs941; + cvt.f32.bf16 %r1665, %rs942; + cvt.f32.bf16 %r1666, %rs943; + cvt.f32.bf16 %r1667, %rs944; + cvt.f32.bf16 %r1668, %rs945; + cvt.f32.bf16 %r1669, %rs946; + cvt.f32.bf16 %r1670, %rs947; + cvt.f32.bf16 %r1671, %rs948; + cvt.f32.bf16 %r1672, %rs949; + cvt.f32.bf16 %r1673, %rs950; + cvt.f32.bf16 %r1674, %rs951; + cvt.f32.bf16 %r1675, %rs952; + cvt.f32.bf16 %r1676, %rs953; + cvt.f32.bf16 %r1677, %rs954; + cvt.f32.bf16 %r1678, %rs955; + cvt.f32.bf16 %r1679, %rs956; + cvt.f32.bf16 %r1680, %rs957; + cvt.f32.bf16 %r1681, %rs958; + cvt.f32.bf16 %r1682, %rs959; + cvt.f32.bf16 %r1683, %rs960; + cvt.f32.bf16 %r1684, %rs961; + cvt.f32.bf16 %r1685, %rs962; + cvt.f32.bf16 %r1686, %rs963; + cvt.f32.bf16 %r1687, %rs964; + cvt.f32.bf16 %r1688, %rs965; + cvt.f32.bf16 %r1689, %rs966; + cvt.f32.bf16 %r1690, %rs967; + cvt.f32.bf16 %r1691, %rs968; + cvt.f32.bf16 %r1692, %rs969; + cvt.f32.bf16 %r1693, %rs970; + cvt.f32.bf16 %r1694, %rs971; + cvt.f32.bf16 %r1695, %rs972; + cvt.f32.bf16 %r1696, %rs973; + cvt.f32.bf16 %r1697, %rs974; + cvt.f32.bf16 %r1698, %rs975; + cvt.f32.bf16 %r1699, %rs976; + cvt.f32.bf16 %r1700, %rs977; + cvt.f32.bf16 %r1701, %rs978; + cvt.f32.bf16 %r1702, %rs979; + cvt.f32.bf16 %r1703, %rs980; + cvt.f32.bf16 %r1704, %rs981; + cvt.f32.bf16 %r1705, %rs982; + cvt.f32.bf16 %r1706, %rs983; + cvt.f32.bf16 %r1707, %rs984; + cvt.f32.bf16 %r1708, %rs985; + cvt.f32.bf16 %r1709, %rs986; + cvt.f32.bf16 %r1710, %rs987; + cvt.f32.bf16 %r1711, %rs988; + cvt.f32.bf16 %r1712, %rs989; + cvt.f32.bf16 %r1713, %rs990; + cvt.f32.bf16 %r1714, %rs991; + cvt.f32.bf16 %r1715, %rs992; + cvt.f32.bf16 %r1716, %rs993; + cvt.f32.bf16 %r1717, %rs994; + cvt.f32.bf16 %r1718, %rs995; + cvt.f32.bf16 %r1719, %rs996; + cvt.f32.bf16 %r1720, %rs997; + cvt.f32.bf16 %r1721, %rs998; + cvt.f32.bf16 %r1722, %rs999; + cvt.f32.bf16 %r1723, %rs1000; + cvt.f32.bf16 %r1724, %rs1001; + cvt.f32.bf16 %r1725, %rs1002; + cvt.f32.bf16 %r1726, %rs1003; + cvt.f32.bf16 %r1727, %rs1004; + .loc 1 107 25 // l2norm.py:107:25 + mov.b32 %r1728, global_smem; + shl.b32 %r1729, %r866, 4; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1730, %rs1005; + .loc 1 107 25 // l2norm.py:107:25 + add.s32 %r1731, %r1728, %r1729; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1732, %rs1006; + .loc 1 107 25 // l2norm.py:107:25 + stmatrix.sync.aligned.m8n8.x2.shared.b16 [%r1731], {%r258, %r259}; + shl.b32 %r1733, %r776, 3; + bar.warp.sync -1; + add.s32 %r1734, %r1728, %r1733; + ld.shared.v4.b32 {%r1735, %r1736, %r1737, %r1738}, [%r1734+16]; + ld.shared.v4.b32 {%r1739, %r1740, %r1741, %r1742}, [%r1734+32]; + ld.shared.v4.b32 {%r1743, %r1744, %r1745, %r1746}, [%r1734+48]; + ld.shared.v4.b32 {%r1747, %r1748, %r1749, %r1750}, [%r1734+64]; + ld.shared.v4.b32 {%r1751, %r1752, %r1753, %r1754}, [%r1734+80]; + ld.shared.v4.b32 {%r1755, %r1756, %r1757, %r1758}, [%r1734+96]; + ld.shared.v4.b32 {%r1759, %r1760, %r1761, %r1762}, [%r1734+112]; + ld.shared.v4.b32 {%r1763, %r1764, %r1765, %r1766}, [%r1734]; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1767, %rs1007; + .loc 1 107 25 // l2norm.py:107:25 + bar.warp.sync -1; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1768, %rs1008; + .loc 1 107 25 // l2norm.py:107:25 + stmatrix.sync.aligned.m8n8.x2.shared.b16 [%r1731], {%r260, %r261}; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1769, %rs1009; + .loc 1 107 25 // l2norm.py:107:25 + bar.warp.sync -1; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1770, %rs1010; + .loc 1 107 25 // l2norm.py:107:25 + ld.shared.v4.b32 {%r1771, %r1772, %r1773, %r1774}, [%r1734]; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r1775, %r932, %r1444; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1776, %rs1011; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1777, %rs208, %rs720, %r1775; +$L__tmp2: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1778, %rs1012; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1779, %rs210, %rs722, %r1777; +$L__tmp4: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r1780, %r1774, %r1455; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1781, %rs211, %rs723, %r1779; +$L__tmp6: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1782, %rs1013; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1783, %rs212, %rs724, %r1781; +$L__tmp8: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1784, %rs1014; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1785, %rs213, %rs725, %r1783; +$L__tmp10: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r1786, %r1774, %r1453; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1787, %rs214, %rs726, %r1785; +$L__tmp12: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1788, %rs1015; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1789, %rs215, %rs727, %r1787; +$L__tmp14: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1790, %rs1016; + cvt.f32.bf16 %r1791, %rs1017; + cvt.f32.bf16 %r1792, %rs1018; + cvt.f32.bf16 %r1793, %rs1019; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r1794, %r908, %r1420; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1795, %rs1020; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1796, %rs184, %rs696, %r1794; +$L__tmp16: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1797, %rs1021; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1798, %rs186, %rs698, %r1796; +$L__tmp18: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1799, %rs1022; +$L__tmp19: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1800, %rs187, %rs699, %r1798; +$L__tmp20: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r1801, %r884, %r1396; +$L__tmp21: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1802, %rs188, %rs700, %r1800; + fma.rn.f32.bf16 %r1803, %rs160, %rs672, %r1801; + fma.rn.f32.bf16 %r1804, %rs189, %rs701, %r1802; + fma.rn.f32.bf16 %r1805, %rs162, %rs674, %r1803; + fma.rn.f32.bf16 %r1806, %rs190, %rs702, %r1804; + fma.rn.f32.bf16 %r1807, %rs163, %rs675, %r1805; + fma.rn.f32.bf16 %r1808, %rs191, %rs703, %r1806; + fma.rn.f32.bf16 %r1809, %rs164, %rs676, %r1807; +$L__tmp22: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1810, %rs1023; +$L__tmp23: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1811, %rs165, %rs677, %r1809; +$L__tmp24: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1812, %rs1024; +$L__tmp25: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1813, %rs166, %rs678, %r1811; +$L__tmp26: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1814, %rs1025; +$L__tmp27: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1815, %rs167, %rs679, %r1813; +$L__tmp28: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1816, %rs1026; + cvt.f32.bf16 %r1817, %rs1027; + cvt.f32.bf16 %r1818, %rs1028; + .loc 1 107 25 // l2norm.py:107:25 + ld.shared.v4.b32 {%r1819, %r1820, %r1821, %r1822}, [%r1734+16]; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r1823, %r892, %r1404; + .loc 1 107 25 // l2norm.py:107:25 + ld.shared.v4.b32 {%r1824, %r1825, %r1826, %r1827}, [%r1734+32]; +$L__tmp29: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1828, %rs168, %rs680, %r1823; +$L__tmp30: + .loc 1 107 25 // l2norm.py:107:25 + ld.shared.v4.b32 {%r1829, %r1830, %r1831, %r1832}, [%r1734+48]; +$L__tmp31: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1833, %rs170, %rs682, %r1828; +$L__tmp32: + .loc 1 107 25 // l2norm.py:107:25 + ld.shared.v4.b32 {%r1834, %r1835, %r1836, %r1837}, [%r1734+64]; +$L__tmp33: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1838, %rs171, %rs683, %r1833; +$L__tmp34: + .loc 1 107 25 // l2norm.py:107:25 + ld.shared.v4.b32 {%r1839, %r1840, %r1841, %r1842}, [%r1734+80]; +$L__tmp35: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1843, %rs172, %rs684, %r1838; +$L__tmp36: + .loc 1 107 25 // l2norm.py:107:25 + ld.shared.v4.b32 {%r1844, %r1845, %r1846, %r1847}, [%r1734+96]; +$L__tmp37: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1848, %rs173, %rs685, %r1843; +$L__tmp38: + .loc 1 107 25 // l2norm.py:107:25 + ld.shared.v4.b32 {%r1849, %r1850, %r1851, %r1852}, [%r1734+112]; +$L__tmp39: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1853, %rs174, %rs686, %r1848; + fma.rn.f32.bf16 %r1854, %rs175, %rs687, %r1853; +$L__tmp40: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1855, %r1815, 8, 31, -1; +$L__tmp41: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1856, %rs1029; +$L__tmp42: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1857, %r1815, %r1855; +$L__tmp43: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1858, %rs1030; +$L__tmp44: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1859, %r1857, 4, 31, -1; +$L__tmp45: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1860, %rs1031; +$L__tmp46: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1861, %r1857, %r1859; +$L__tmp47: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1862, %rs1032; +$L__tmp48: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1863, %r1861, 2, 31, -1; +$L__tmp49: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r1864, %r900, %r1412; +$L__tmp50: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1865, %r1861, %r1863; + fma.rn.f32.bf16 %r1866, %rs176, %rs688, %r1864; +$L__tmp51: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1867, %r1865, 1, 31, -1; +$L__tmp52: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1868, %rs178, %rs690, %r1866; +$L__tmp53: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1869, %r1854, 8, 31, -1; +$L__tmp54: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1870, %rs179, %rs691, %r1868; + add.f32 %r1871, %r1854, %r1869; + fma.rn.f32.bf16 %r1872, %rs180, %rs692, %r1870; +$L__tmp55: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1873, %r1871, 4, 31, -1; +$L__tmp56: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1874, %rs181, %rs693, %r1872; + add.f32 %r1875, %r1871, %r1873; + fma.rn.f32.bf16 %r1876, %rs182, %rs694, %r1874; +$L__tmp57: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1877, %r1875, 2, 31, -1; +$L__tmp58: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1878, %rs183, %rs695, %r1876; + add.f32 %r1879, %r1875, %r1877; +$L__tmp59: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1880, %rs1033; +$L__tmp60: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1881, %r1879, 1, 31, -1; +$L__tmp61: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1882, %rs1034; +$L__tmp62: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1883, %r1878, 8, 31, -1; +$L__tmp63: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r1884, %r916, %r1428; +$L__tmp64: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1885, %r1878, %r1883; + fma.rn.f32.bf16 %r1886, %rs192, %rs704, %r1884; +$L__tmp65: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1887, %r1885, 4, 31, -1; +$L__tmp66: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1888, %rs194, %rs706, %r1886; + add.f32 %r1889, %r1885, %r1887; + fma.rn.f32.bf16 %r1890, %rs195, %rs707, %r1888; +$L__tmp67: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1891, %r1889, 2, 31, -1; +$L__tmp68: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1892, %rs196, %rs708, %r1890; + add.f32 %r1893, %r1889, %r1891; + fma.rn.f32.bf16 %r1894, %rs197, %rs709, %r1892; +$L__tmp69: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1895, %r1893, 1, 31, -1; +$L__tmp70: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1896, %rs198, %rs710, %r1894; +$L__tmp71: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1897, %r1808, 8, 31, -1; +$L__tmp72: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1898, %rs199, %rs711, %r1896; + add.f32 %r1899, %r1808, %r1897; +$L__tmp73: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1900, %rs1035; +$L__tmp74: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1901, %r1899, 4, 31, -1; +$L__tmp75: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1902, %rs1036; +$L__tmp76: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1903, %r1899, %r1901; +$L__tmp77: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1904, %rs1037; +$L__tmp78: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1905, %r1903, 2, 31, -1; +$L__tmp79: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1906, %rs1038; +$L__tmp80: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1907, %r1903, %r1905; +$L__tmp81: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r1908, %r924, %r1436; +$L__tmp82: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1909, %r1907, 1, 31, -1; +$L__tmp83: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1910, %rs200, %rs712, %r1908; +$L__tmp84: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1911, %r1898, 8, 31, -1; +$L__tmp85: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1912, %rs202, %rs714, %r1910; + add.f32 %r1913, %r1898, %r1911; + fma.rn.f32.bf16 %r1914, %rs203, %rs715, %r1912; +$L__tmp86: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1915, %r1913, 4, 31, -1; +$L__tmp87: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1916, %rs204, %rs716, %r1914; + add.f32 %r1917, %r1913, %r1915; + fma.rn.f32.bf16 %r1918, %rs205, %rs717, %r1916; +$L__tmp88: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1919, %r1917, 2, 31, -1; +$L__tmp89: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1920, %rs206, %rs718, %r1918; + add.f32 %r1921, %r1917, %r1919; + fma.rn.f32.bf16 %r1922, %rs207, %rs719, %r1920; +$L__tmp90: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1923, %r1921, 1, 31, -1; +$L__tmp91: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1924, %rs1039; +$L__tmp92: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1925, %r1922, 8, 31, -1; +$L__tmp93: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1926, %rs1040; +$L__tmp94: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1927, %r1922, %r1925; +$L__tmp95: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1928, %rs1041; +$L__tmp96: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1929, %r1927, 4, 31, -1; +$L__tmp97: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1930, %rs1042; +$L__tmp98: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1931, %r1927, %r1929; +$L__tmp99: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r1932, %r940, %r1452; +$L__tmp100: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1933, %r1931, 2, 31, -1; +$L__tmp101: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1934, %rs216, %rs728, %r1932; + add.f32 %r1935, %r1931, %r1933; + fma.rn.f32.bf16 %r1936, %rs218, %rs730, %r1934; +$L__tmp102: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1937, %r1935, 1, 31, -1; +$L__tmp103: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1938, %rs219, %rs731, %r1936; +$L__tmp104: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1939, %r1789, 8, 31, -1; +$L__tmp105: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1940, %rs220, %rs732, %r1938; + add.f32 %r1941, %r1789, %r1939; + fma.rn.f32.bf16 %r1942, %rs221, %rs733, %r1940; +$L__tmp106: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1943, %r1941, 4, 31, -1; +$L__tmp107: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1944, %rs222, %rs734, %r1942; + add.f32 %r1945, %r1941, %r1943; + fma.rn.f32.bf16 %r1946, %rs223, %rs735, %r1944; +$L__tmp108: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1947, %r1945, 2, 31, -1; +$L__tmp109: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1948, %rs1043; +$L__tmp110: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1949, %r1945, %r1947; +$L__tmp111: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1950, %rs1044; +$L__tmp112: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1951, %r1949, 1, 31, -1; +$L__tmp113: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r1952, %r1774, %r1454; +$L__tmp114: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1953, %r1946, 8, 31, -1; +$L__tmp115: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd663, {%r1786, %r1952}; +$L__tmp116: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1954, %r1946, %r1953; +$L__tmp117: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1955, %rs1045; +$L__tmp118: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1956, %r1954, 4, 31, -1; +$L__tmp119: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1957, %rs1046; +$L__tmp120: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1958, %r1954, %r1956; +$L__tmp121: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r1959, %r1774, %r1451; +$L__tmp122: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1960, %r1958, 2, 31, -1; +$L__tmp123: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r1961, %r1774, %r1452; +$L__tmp124: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1962, %r1958, %r1960; +$L__tmp125: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd664, {%r1959, %r1961}; +$L__tmp126: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1963, %r1962, 1, 31, -1; +$L__tmp127: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1964, %rs1047; +$L__tmp128: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1965, %r1962, %r1963; +$L__tmp129: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1966, %rs1048; + cvt.f32.bf16 %r1967, %rs1049; + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r1968, %r1965, %r941; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1969, %rs1050; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r1970, %r1774, %r1968; + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r1971, %r1965, %r939; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1972, %rs1051; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r1973, %r1774, %r1971; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1974, %rs1052; + cvt.f32.bf16 %r1975, %rs1053; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r1976, %r1766, %r1447; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1977, %rs1054; + cvt.f32.bf16 %r1978, %rs1055; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r1979, %r1766, %r1449; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1980, %rs1056; + cvt.f32.bf16 %r1981, %rs1057; + cvt.f32.bf16 %r1982, %rs1058; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r1983, %r1766, %r1450; + mul.f32 %r1984, %r1766, %r1448; + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd665, {%r1979, %r1983}; + mov.b64 %rd666, {%r1976, %r1984}; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1985, %rs1059; +$L__tmp130: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1986, %r1949, %r1951; +$L__tmp131: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1987, %rs1060; + cvt.f32.bf16 %r1988, %rs1061; + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r1989, %r1986, %r937; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1990, %rs1062; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r1991, %r1766, %r1989; + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r1992, %r1986, %r935; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1993, %rs1063; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r1994, %r1766, %r1992; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1995, %rs1064; + cvt.f32.bf16 %r1996, %rs1065; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r1997, %r1766, %r1445; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1998, %rs1066; + cvt.f32.bf16 %r1999, %rs1067; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2000, %r1766, %r1446; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2001, %rs1068; + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd667, {%r1997, %r2000}; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2002, %r1766, %r1443; + mul.f32 %r2003, %r1766, %r1444; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2004, %rs1069; + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd668, {%r2002, %r2003}; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2005, %rs1070; + cvt.f32.bf16 %r2006, %rs1071; + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r2007, %r1986, %r933; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2008, %rs1072; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r2009, %r1766, %r2007; + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r2010, %r1986, %r931; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2011, %rs1073; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r2012, %r1766, %r2010; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2013, %rs1074; + cvt.f32.bf16 %r2014, %rs1075; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2015, %r1773, %r1439; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2016, %rs1076; + cvt.f32.bf16 %r2017, %rs1077; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2018, %r1773, %r1441; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2019, %rs1078; + cvt.f32.bf16 %r2020, %rs1079; + cvt.f32.bf16 %r2021, %rs1080; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2022, %r1773, %r1442; + mul.f32 %r2023, %r1773, %r1440; + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd669, {%r2018, %r2022}; + mov.b64 %rd670, {%r2015, %r2023}; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2024, %rs1081; +$L__tmp132: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2025, %r1935, %r1937; +$L__tmp133: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2026, %rs1082; + cvt.f32.bf16 %r2027, %rs1083; + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r2028, %r2025, %r929; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2029, %rs1084; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r2030, %r1773, %r2028; + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r2031, %r2025, %r927; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2032, %rs1085; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r2033, %r1773, %r2031; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2034, %rs1086; + cvt.f32.bf16 %r2035, %rs1087; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2036, %r1773, %r1437; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2037, %rs1088; + cvt.f32.bf16 %r2038, %rs1089; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2039, %r1773, %r1438; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2040, %rs1090; + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd671, {%r2036, %r2039}; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2041, %r1773, %r1435; + mul.f32 %r2042, %r1773, %r1436; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2043, %rs1091; + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd672, {%r2041, %r2042}; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2044, %rs1092; + cvt.f32.bf16 %r2045, %rs1093; + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r2046, %r2025, %r925; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2047, %rs1094; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r2048, %r1773, %r2046; + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r2049, %r2025, %r923; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2050, %rs1095; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r2051, %r1773, %r2049; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2052, %rs1096; + cvt.f32.bf16 %r2053, %rs1097; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2054, %r1765, %r1431; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2055, %rs1098; + cvt.f32.bf16 %r2056, %rs1099; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2057, %r1765, %r1433; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2058, %rs1100; + cvt.f32.bf16 %r2059, %rs1101; + cvt.f32.bf16 %r2060, %rs1102; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2061, %r1765, %r1434; + mul.f32 %r2062, %r1765, %r1432; + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd673, {%r2057, %r2061}; + mov.b64 %rd674, {%r2054, %r2062}; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2063, %rs1103; +$L__tmp134: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2064, %r1921, %r1923; +$L__tmp135: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2065, %rs1104; + cvt.f32.bf16 %r2066, %rs1105; + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r2067, %r2064, %r921; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2068, %rs1106; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r2069, %r1765, %r2067; + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r2070, %r2064, %r919; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2071, %rs1107; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r2072, %r1765, %r2070; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2073, %rs1108; + cvt.f32.bf16 %r2074, %rs1109; + cvt.f32.bf16 %r2075, %rs1110; + cvt.f32.bf16 %r2076, %rs1111; + cvt.f32.bf16 %r2077, %rs1112; + cvt.f32.bf16 %r2078, %rs1113; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2079, %r1276, %r1996; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2080, %rs1114; +$L__tmp136: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2081, %rs552, %rs1064, %r2079; +$L__tmp137: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2082, %r1268, %r1981; +$L__tmp138: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2083, %rs554, %rs1066, %r2081; + fma.rn.f32.bf16 %r2084, %rs544, %rs1056, %r2082; + fma.rn.f32.bf16 %r2085, %rs555, %rs1067, %r2083; + fma.rn.f32.bf16 %r2086, %rs546, %rs1058, %r2084; + fma.rn.f32.bf16 %r2087, %rs556, %rs1068, %r2085; + fma.rn.f32.bf16 %r2088, %rs547, %rs1059, %r2086; + fma.rn.f32.bf16 %r2089, %rs557, %rs1069, %r2087; + fma.rn.f32.bf16 %r2090, %rs548, %rs1060, %r2088; + fma.rn.f32.bf16 %r2091, %rs558, %rs1070, %r2089; + fma.rn.f32.bf16 %r2092, %rs549, %rs1061, %r2090; + fma.rn.f32.bf16 %r2093, %rs559, %rs1071, %r2091; + fma.rn.f32.bf16 %r2094, %rs550, %rs1062, %r2092; +$L__tmp139: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2095, %rs1115; +$L__tmp140: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2096, %rs551, %rs1063, %r2094; +$L__tmp141: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2097, %rs1116; + cvt.f32.bf16 %r2098, %rs1117; + cvt.f32.bf16 %r2099, %rs1118; + cvt.f32.bf16 %r2100, %rs1119; + cvt.f32.bf16 %r2101, %rs1120; + cvt.f32.bf16 %r2102, %rs1121; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2103, %r1260, %r1967; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2104, %rs1122; +$L__tmp142: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2105, %rs536, %rs1048, %r2103; +$L__tmp143: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2106, %r1252, %r1928; +$L__tmp144: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2107, %rs538, %rs1050, %r2105; + fma.rn.f32.bf16 %r2108, %rs528, %rs1040, %r2106; + fma.rn.f32.bf16 %r2109, %rs539, %rs1051, %r2107; + fma.rn.f32.bf16 %r2110, %rs530, %rs1042, %r2108; + fma.rn.f32.bf16 %r2111, %rs540, %rs1052, %r2109; + fma.rn.f32.bf16 %r2112, %rs531, %rs1043, %r2110; + fma.rn.f32.bf16 %r2113, %rs541, %rs1053, %r2111; + fma.rn.f32.bf16 %r2114, %rs532, %rs1044, %r2112; + fma.rn.f32.bf16 %r2115, %rs542, %rs1054, %r2113; + fma.rn.f32.bf16 %r2116, %rs533, %rs1045, %r2114; + fma.rn.f32.bf16 %r2117, %rs543, %rs1055, %r2115; + fma.rn.f32.bf16 %r2118, %rs534, %rs1046, %r2116; +$L__tmp145: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2119, %rs1123; +$L__tmp146: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2120, %rs535, %rs1047, %r2118; +$L__tmp147: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2121, %rs1124; + cvt.f32.bf16 %r2122, %rs1125; + cvt.f32.bf16 %r2123, %rs1126; + cvt.f32.bf16 %r2124, %rs1127; + cvt.f32.bf16 %r2125, %rs1128; + cvt.f32.bf16 %r2126, %rs1129; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2127, %r1244, %r1880; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2128, %rs1130; +$L__tmp148: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2129, %rs520, %rs1032, %r2127; +$L__tmp149: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2130, %r1236, %r1814; +$L__tmp150: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2131, %rs522, %rs1034, %r2129; + fma.rn.f32.bf16 %r2132, %rs512, %rs1024, %r2130; + fma.rn.f32.bf16 %r2133, %rs523, %rs1035, %r2131; + fma.rn.f32.bf16 %r2134, %rs514, %rs1026, %r2132; + fma.rn.f32.bf16 %r2135, %rs524, %rs1036, %r2133; + fma.rn.f32.bf16 %r2136, %rs515, %rs1027, %r2134; + fma.rn.f32.bf16 %r2137, %rs525, %rs1037, %r2135; + fma.rn.f32.bf16 %r2138, %rs516, %rs1028, %r2136; + fma.rn.f32.bf16 %r2139, %rs526, %rs1038, %r2137; + fma.rn.f32.bf16 %r2140, %rs517, %rs1029, %r2138; + fma.rn.f32.bf16 %r2141, %rs527, %rs1039, %r2139; + fma.rn.f32.bf16 %r2142, %rs518, %rs1030, %r2140; +$L__tmp151: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2143, %rs1131; +$L__tmp152: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2144, %rs519, %rs1031, %r2142; +$L__tmp153: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2145, %rs1132; + cvt.f32.bf16 %r2146, %rs1133; + cvt.f32.bf16 %r2147, %rs1134; + cvt.f32.bf16 %r2148, %rs1135; + cvt.f32.bf16 %r2149, %rs1136; + cvt.f32.bf16 %r2150, %rs1137; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2151, %r1228, %r1791; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2152, %rs1138; +$L__tmp154: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2153, %rs504, %rs1016, %r2151; +$L__tmp155: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2154, %r1220, %r1769; +$L__tmp156: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2155, %rs506, %rs1018, %r2153; + fma.rn.f32.bf16 %r2156, %rs496, %rs1008, %r2154; + fma.rn.f32.bf16 %r2157, %rs507, %rs1019, %r2155; + fma.rn.f32.bf16 %r2158, %rs498, %rs1010, %r2156; + fma.rn.f32.bf16 %r2159, %rs508, %rs1020, %r2157; + fma.rn.f32.bf16 %r2160, %rs499, %rs1011, %r2158; + fma.rn.f32.bf16 %r2161, %rs509, %rs1021, %r2159; + fma.rn.f32.bf16 %r2162, %rs500, %rs1012, %r2160; + fma.rn.f32.bf16 %r2163, %rs510, %rs1022, %r2161; + fma.rn.f32.bf16 %r2164, %rs501, %rs1013, %r2162; + fma.rn.f32.bf16 %r2165, %rs511, %rs1023, %r2163; + fma.rn.f32.bf16 %r2166, %rs502, %rs1014, %r2164; +$L__tmp157: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2167, %rs1139; +$L__tmp158: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2168, %rs503, %rs1015, %r2166; +$L__tmp159: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2169, %rs1140; + cvt.f32.bf16 %r2170, %rs1141; + cvt.f32.bf16 %r2171, %rs1142; + cvt.f32.bf16 %r2172, %rs1143; + cvt.f32.bf16 %r2173, %rs1144; + cvt.f32.bf16 %r2174, %rs1145; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2175, %r1212, %r1724; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2176, %rs1146; +$L__tmp160: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2177, %rs488, %rs1000, %r2175; +$L__tmp161: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2178, %r1204, %r1716; +$L__tmp162: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2179, %rs490, %rs1002, %r2177; + fma.rn.f32.bf16 %r2180, %rs480, %rs992, %r2178; + fma.rn.f32.bf16 %r2181, %rs491, %rs1003, %r2179; + fma.rn.f32.bf16 %r2182, %rs482, %rs994, %r2180; + fma.rn.f32.bf16 %r2183, %rs492, %rs1004, %r2181; + fma.rn.f32.bf16 %r2184, %rs483, %rs995, %r2182; + fma.rn.f32.bf16 %r2185, %rs493, %rs1005, %r2183; + fma.rn.f32.bf16 %r2186, %rs484, %rs996, %r2184; + fma.rn.f32.bf16 %r2187, %rs494, %rs1006, %r2185; + fma.rn.f32.bf16 %r2188, %rs485, %rs997, %r2186; + fma.rn.f32.bf16 %r2189, %rs495, %rs1007, %r2187; + fma.rn.f32.bf16 %r2190, %rs486, %rs998, %r2188; +$L__tmp163: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2191, %rs1147; +$L__tmp164: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2192, %rs487, %rs999, %r2190; +$L__tmp165: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2193, %rs1148; + cvt.f32.bf16 %r2194, %rs1149; + cvt.f32.bf16 %r2195, %rs1150; + cvt.f32.bf16 %r2196, %rs1151; + cvt.f32.bf16 %r2197, %rs1152; + cvt.f32.bf16 %r2198, %rs1153; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2199, %r1196, %r1708; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2200, %rs1154; +$L__tmp166: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2201, %rs472, %rs984, %r2199; +$L__tmp167: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2202, %r1188, %r1700; +$L__tmp168: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2203, %rs474, %rs986, %r2201; + fma.rn.f32.bf16 %r2204, %rs464, %rs976, %r2202; + fma.rn.f32.bf16 %r2205, %rs475, %rs987, %r2203; + fma.rn.f32.bf16 %r2206, %rs466, %rs978, %r2204; + fma.rn.f32.bf16 %r2207, %rs476, %rs988, %r2205; + fma.rn.f32.bf16 %r2208, %rs467, %rs979, %r2206; + fma.rn.f32.bf16 %r2209, %rs477, %rs989, %r2207; + fma.rn.f32.bf16 %r2210, %rs468, %rs980, %r2208; + fma.rn.f32.bf16 %r2211, %rs478, %rs990, %r2209; + fma.rn.f32.bf16 %r2212, %rs469, %rs981, %r2210; + fma.rn.f32.bf16 %r2213, %rs479, %rs991, %r2211; + fma.rn.f32.bf16 %r2214, %rs470, %rs982, %r2212; +$L__tmp169: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2215, %rs1155; +$L__tmp170: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2216, %rs471, %rs983, %r2214; +$L__tmp171: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2217, %rs1156; + cvt.f32.bf16 %r2218, %rs1157; + cvt.f32.bf16 %r2219, %rs1158; + cvt.f32.bf16 %r2220, %rs1159; + cvt.f32.bf16 %r2221, %rs1160; + cvt.f32.bf16 %r2222, %rs1161; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2223, %r1180, %r1692; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2224, %rs1162; +$L__tmp172: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2225, %rs456, %rs968, %r2223; +$L__tmp173: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2226, %r1172, %r1684; +$L__tmp174: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2227, %rs458, %rs970, %r2225; + fma.rn.f32.bf16 %r2228, %rs448, %rs960, %r2226; + fma.rn.f32.bf16 %r2229, %rs459, %rs971, %r2227; + fma.rn.f32.bf16 %r2230, %rs450, %rs962, %r2228; + fma.rn.f32.bf16 %r2231, %rs460, %rs972, %r2229; + fma.rn.f32.bf16 %r2232, %rs451, %rs963, %r2230; + fma.rn.f32.bf16 %r2233, %rs461, %rs973, %r2231; + fma.rn.f32.bf16 %r2234, %rs452, %rs964, %r2232; + fma.rn.f32.bf16 %r2235, %rs462, %rs974, %r2233; + fma.rn.f32.bf16 %r2236, %rs453, %rs965, %r2234; + fma.rn.f32.bf16 %r2237, %rs463, %rs975, %r2235; + fma.rn.f32.bf16 %r2238, %rs454, %rs966, %r2236; +$L__tmp175: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2239, %rs1163; +$L__tmp176: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2240, %rs455, %rs967, %r2238; +$L__tmp177: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2241, %rs1164; + cvt.f32.bf16 %r2242, %rs1165; + cvt.f32.bf16 %r2243, %rs1166; + cvt.f32.bf16 %r2244, %rs1167; + cvt.f32.bf16 %r2245, %rs1168; + cvt.f32.bf16 %r2246, %rs1169; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2247, %r1164, %r1676; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2248, %rs1170; +$L__tmp178: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2249, %rs440, %rs952, %r2247; +$L__tmp179: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2250, %r1156, %r1668; +$L__tmp180: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2251, %rs442, %rs954, %r2249; + fma.rn.f32.bf16 %r2252, %rs432, %rs944, %r2250; + fma.rn.f32.bf16 %r2253, %rs443, %rs955, %r2251; + fma.rn.f32.bf16 %r2254, %rs434, %rs946, %r2252; + fma.rn.f32.bf16 %r2255, %rs444, %rs956, %r2253; + fma.rn.f32.bf16 %r2256, %rs435, %rs947, %r2254; + fma.rn.f32.bf16 %r2257, %rs445, %rs957, %r2255; + fma.rn.f32.bf16 %r2258, %rs436, %rs948, %r2256; + fma.rn.f32.bf16 %r2259, %rs446, %rs958, %r2257; + fma.rn.f32.bf16 %r2260, %rs437, %rs949, %r2258; + fma.rn.f32.bf16 %r2261, %rs447, %rs959, %r2259; + fma.rn.f32.bf16 %r2262, %rs438, %rs950, %r2260; +$L__tmp181: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2263, %rs1171; +$L__tmp182: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2264, %rs439, %rs951, %r2262; +$L__tmp183: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2265, %rs1172; + cvt.f32.bf16 %r2266, %rs1173; + cvt.f32.bf16 %r2267, %rs1174; + cvt.f32.bf16 %r2268, %rs1175; + cvt.f32.bf16 %r2269, %rs1176; + cvt.f32.bf16 %r2270, %rs1177; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2271, %r1148, %r1660; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2272, %rs1178; +$L__tmp184: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2273, %rs424, %rs936, %r2271; +$L__tmp185: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2274, %r1140, %r1652; +$L__tmp186: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2275, %rs426, %rs938, %r2273; + fma.rn.f32.bf16 %r2276, %rs416, %rs928, %r2274; + fma.rn.f32.bf16 %r2277, %rs427, %rs939, %r2275; + fma.rn.f32.bf16 %r2278, %rs418, %rs930, %r2276; + fma.rn.f32.bf16 %r2279, %rs428, %rs940, %r2277; + fma.rn.f32.bf16 %r2280, %rs419, %rs931, %r2278; + fma.rn.f32.bf16 %r2281, %rs429, %rs941, %r2279; + fma.rn.f32.bf16 %r2282, %rs420, %rs932, %r2280; + fma.rn.f32.bf16 %r2283, %rs430, %rs942, %r2281; + fma.rn.f32.bf16 %r2284, %rs421, %rs933, %r2282; + fma.rn.f32.bf16 %r2285, %rs431, %rs943, %r2283; + fma.rn.f32.bf16 %r2286, %rs422, %rs934, %r2284; +$L__tmp187: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2287, %rs1179; +$L__tmp188: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2288, %rs423, %rs935, %r2286; +$L__tmp189: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2289, %rs1180; + cvt.f32.bf16 %r2290, %rs1181; + cvt.f32.bf16 %r2291, %rs1182; + cvt.f32.bf16 %r2292, %rs1183; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2293, %r1771, %r1403; + mul.f32 %r2294, %r1771, %r1404; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2295, %r1132, %r1644; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2296, %r1771, %r1405; +$L__tmp190: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2297, %rs408, %rs920, %r2295; +$L__tmp191: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2298, %r1771, %r1406; +$L__tmp192: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2299, %rs410, %rs922, %r2297; +$L__tmp193: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2300, %r1771, %r1407; +$L__tmp194: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2301, %rs411, %rs923, %r2299; +$L__tmp195: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2302, %r1771, %r1408; +$L__tmp196: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2303, %rs412, %rs924, %r2301; +$L__tmp197: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2304, %r1771, %r1409; +$L__tmp198: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2305, %rs413, %rs925, %r2303; +$L__tmp199: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2306, %r1771, %r1410; +$L__tmp200: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2307, %rs414, %rs926, %r2305; +$L__tmp201: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2308, %r1764, %r1411; +$L__tmp202: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2309, %rs415, %rs927, %r2307; +$L__tmp203: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2310, %r1124, %r1636; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2311, %r1764, %r1412; +$L__tmp204: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2312, %rs400, %rs912, %r2310; +$L__tmp205: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2313, %r1764, %r1413; +$L__tmp206: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2314, %rs402, %rs914, %r2312; +$L__tmp207: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2315, %r1764, %r1414; +$L__tmp208: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2316, %rs403, %rs915, %r2314; +$L__tmp209: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2317, %r1764, %r1415; +$L__tmp210: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2318, %rs404, %rs916, %r2316; +$L__tmp211: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2319, %r1764, %r1416; +$L__tmp212: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2320, %rs405, %rs917, %r2318; +$L__tmp213: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2321, %r1764, %r1417; +$L__tmp214: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2322, %rs406, %rs918, %r2320; +$L__tmp215: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2323, %r1764, %r1418; +$L__tmp216: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2324, %rs407, %rs919, %r2322; +$L__tmp217: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2325, %r1772, %r1419; + mul.f32 %r2326, %r1772, %r1420; + mul.f32 %r2327, %r1772, %r1421; + mul.f32 %r2328, %r1772, %r1422; + mul.f32 %r2329, %r1772, %r1423; + mul.f32 %r2330, %r1772, %r1424; + mul.f32 %r2331, %r1772, %r1425; + mul.f32 %r2332, %r1772, %r1426; + mul.f32 %r2333, %r1765, %r1427; + mul.f32 %r2334, %r1765, %r1428; + mul.f32 %r2335, %r1765, %r1429; + mul.f32 %r2336, %r1765, %r1430; + mul.f32 %r2337, %r1774, %r1456; + mul.f32 %r2338, %r1774, %r1457; + mul.f32 %r2339, %r1774, %r1458; + mul.f32 %r2340, %r1735, %r1459; + mul.f32 %r2341, %r1735, %r1460; + mul.f32 %r2342, %r1735, %r1461; + mul.f32 %r2343, %r1735, %r1462; + mul.f32 %r2344, %r1735, %r1463; + mul.f32 %r2345, %r1735, %r1464; + mul.f32 %r2346, %r1735, %r1465; + mul.f32 %r2347, %r1735, %r1466; + mul.f32 %r2348, %r1819, %r1467; + mul.f32 %r2349, %r1819, %r1468; + mul.f32 %r2350, %r1819, %r1469; + mul.f32 %r2351, %r1819, %r1470; + mul.f32 %r2352, %r1819, %r1471; + mul.f32 %r2353, %r1819, %r1472; + mul.f32 %r2354, %r1819, %r1473; + mul.f32 %r2355, %r1819, %r1474; + mul.f32 %r2356, %r1736, %r1475; + mul.f32 %r2357, %r1736, %r1476; + mul.f32 %r2358, %r1736, %r1477; + mul.f32 %r2359, %r1736, %r1478; + mul.f32 %r2360, %r1736, %r1479; + mul.f32 %r2361, %r1736, %r1480; + mul.f32 %r2362, %r1736, %r1481; + mul.f32 %r2363, %r1736, %r1482; + mul.f32 %r2364, %r1820, %r1483; + mul.f32 %r2365, %r1820, %r1484; + mul.f32 %r2366, %r1820, %r1485; + mul.f32 %r2367, %r1820, %r1486; + mul.f32 %r2368, %r1820, %r1487; + mul.f32 %r2369, %r1820, %r1488; + mul.f32 %r2370, %r1820, %r1489; + mul.f32 %r2371, %r1820, %r1490; + mul.f32 %r2372, %r1737, %r1491; + mul.f32 %r2373, %r1737, %r1492; + mul.f32 %r2374, %r1737, %r1493; + mul.f32 %r2375, %r1737, %r1494; + mul.f32 %r2376, %r1737, %r1495; + mul.f32 %r2377, %r1737, %r1496; + mul.f32 %r2378, %r1737, %r1497; + mul.f32 %r2379, %r1737, %r1498; + mul.f32 %r2380, %r1821, %r1499; + mul.f32 %r2381, %r1821, %r1500; + mul.f32 %r2382, %r1821, %r1501; + mul.f32 %r2383, %r1821, %r1502; + mul.f32 %r2384, %r1821, %r1503; + mul.f32 %r2385, %r1821, %r1504; + mul.f32 %r2386, %r1821, %r1505; + mul.f32 %r2387, %r1821, %r1506; + mul.f32 %r2388, %r1738, %r1507; + mul.f32 %r2389, %r1738, %r1508; + mul.f32 %r2390, %r1738, %r1509; + mul.f32 %r2391, %r1738, %r1510; + mul.f32 %r2392, %r1738, %r1511; + mul.f32 %r2393, %r1738, %r1512; + mul.f32 %r2394, %r1738, %r1513; + mul.f32 %r2395, %r1738, %r1514; + mul.f32 %r2396, %r1822, %r1515; + mul.f32 %r2397, %r1822, %r1516; + mul.f32 %r2398, %r1822, %r1517; + mul.f32 %r2399, %r1822, %r1518; + mul.f32 %r2400, %r1822, %r1519; + mul.f32 %r2401, %r1822, %r1520; + mul.f32 %r2402, %r1822, %r1521; + mul.f32 %r2403, %r1822, %r1522; + mul.f32 %r2404, %r1739, %r1523; + mul.f32 %r2405, %r1739, %r1524; + mul.f32 %r2406, %r1739, %r1525; + mul.f32 %r2407, %r1739, %r1526; + mul.f32 %r2408, %r1739, %r1527; + mul.f32 %r2409, %r1739, %r1528; + mul.f32 %r2410, %r1739, %r1529; + mul.f32 %r2411, %r1739, %r1530; + mul.f32 %r2412, %r1824, %r1531; + mul.f32 %r2413, %r1824, %r1532; + mul.f32 %r2414, %r1824, %r1533; + mul.f32 %r2415, %r1824, %r1534; + mul.f32 %r2416, %r1824, %r1535; + mul.f32 %r2417, %r1824, %r1536; + mul.f32 %r2418, %r1824, %r1537; + mul.f32 %r2419, %r1824, %r1538; + mul.f32 %r2420, %r1740, %r1539; + mul.f32 %r2421, %r1740, %r1540; + mul.f32 %r2422, %r1740, %r1541; + mul.f32 %r2423, %r1740, %r1542; + mul.f32 %r2424, %r1740, %r1543; + mul.f32 %r2425, %r1740, %r1544; + mul.f32 %r2426, %r1740, %r1545; + mul.f32 %r2427, %r1740, %r1546; + mul.f32 %r2428, %r1825, %r1547; + mul.f32 %r2429, %r1825, %r1548; + mul.f32 %r2430, %r1825, %r1549; + mul.f32 %r2431, %r1825, %r1550; + mul.f32 %r2432, %r1825, %r1551; + mul.f32 %r2433, %r1825, %r1552; + mul.f32 %r2434, %r1825, %r1553; + mul.f32 %r2435, %r1825, %r1554; + mul.f32 %r2436, %r1741, %r1555; + mul.f32 %r2437, %r1741, %r1556; + mul.f32 %r2438, %r1741, %r1557; + mul.f32 %r2439, %r1741, %r1558; + mul.f32 %r2440, %r1741, %r1559; + mul.f32 %r2441, %r1741, %r1560; + mul.f32 %r2442, %r1741, %r1561; + mul.f32 %r2443, %r1741, %r1562; + mul.f32 %r2444, %r1826, %r1563; + mul.f32 %r2445, %r1826, %r1564; + mul.f32 %r2446, %r1826, %r1565; + mul.f32 %r2447, %r1826, %r1566; + mul.f32 %r2448, %r1826, %r1567; + mul.f32 %r2449, %r1826, %r1568; + mul.f32 %r2450, %r1826, %r1569; + mul.f32 %r2451, %r1826, %r1570; + mul.f32 %r2452, %r1742, %r1571; + mul.f32 %r2453, %r1742, %r1572; + mul.f32 %r2454, %r1742, %r1573; + mul.f32 %r2455, %r1742, %r1574; + mul.f32 %r2456, %r1742, %r1575; + mul.f32 %r2457, %r1742, %r1576; + mul.f32 %r2458, %r1742, %r1577; + mul.f32 %r2459, %r1742, %r1578; + mul.f32 %r2460, %r1827, %r1579; + mul.f32 %r2461, %r1827, %r1580; + mul.f32 %r2462, %r1827, %r1581; + mul.f32 %r2463, %r1827, %r1582; + mul.f32 %r2464, %r1827, %r1583; + mul.f32 %r2465, %r1827, %r1584; + mul.f32 %r2466, %r1827, %r1585; + mul.f32 %r2467, %r1827, %r1586; + mul.f32 %r2468, %r1743, %r1587; + mul.f32 %r2469, %r1743, %r1588; + mul.f32 %r2470, %r1743, %r1589; + mul.f32 %r2471, %r1743, %r1590; + mul.f32 %r2472, %r1743, %r1591; + mul.f32 %r2473, %r1743, %r1592; + mul.f32 %r2474, %r1743, %r1593; + mul.f32 %r2475, %r1743, %r1594; + mul.f32 %r2476, %r1829, %r1595; + mul.f32 %r2477, %r1829, %r1596; + mul.f32 %r2478, %r1829, %r1597; + mul.f32 %r2479, %r1829, %r1598; + mul.f32 %r2480, %r1829, %r1599; + mul.f32 %r2481, %r1829, %r1600; + mul.f32 %r2482, %r1829, %r1601; + mul.f32 %r2483, %r1829, %r1602; + mul.f32 %r2484, %r1744, %r1603; + mul.f32 %r2485, %r1744, %r1604; + mul.f32 %r2486, %r1744, %r1605; + mul.f32 %r2487, %r1744, %r1606; + mul.f32 %r2488, %r1744, %r1607; + mul.f32 %r2489, %r1744, %r1608; + mul.f32 %r2490, %r1744, %r1609; + mul.f32 %r2491, %r1744, %r1610; + mul.f32 %r2492, %r1830, %r1611; + mul.f32 %r2493, %r1830, %r1612; + mul.f32 %r2494, %r1830, %r1613; + mul.f32 %r2495, %r1830, %r1614; + mul.f32 %r2496, %r1830, %r1615; + mul.f32 %r2497, %r1830, %r1616; + mul.f32 %r2498, %r1830, %r1617; + mul.f32 %r2499, %r1830, %r1618; + mul.f32 %r2500, %r1745, %r1619; + mul.f32 %r2501, %r1745, %r1620; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2502, %r1116, %r1628; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2503, %r1745, %r1621; +$L__tmp218: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2504, %rs392, %rs904, %r2502; +$L__tmp219: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2505, %r1745, %r1622; +$L__tmp220: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2506, %rs394, %rs906, %r2504; +$L__tmp221: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2507, %r1745, %r1623; +$L__tmp222: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2508, %rs395, %rs907, %r2506; +$L__tmp223: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2509, %r1745, %r1624; +$L__tmp224: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2510, %rs396, %rs908, %r2508; +$L__tmp225: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2511, %r1745, %r1625; +$L__tmp226: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2512, %rs397, %rs909, %r2510; +$L__tmp227: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2513, %r1745, %r1626; +$L__tmp228: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2514, %rs398, %rs910, %r2512; +$L__tmp229: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2515, %r1831, %r1627; +$L__tmp230: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2516, %rs399, %rs911, %r2514; +$L__tmp231: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2517, %r1108, %r1620; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2518, %r1831, %r1628; +$L__tmp232: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2519, %rs384, %rs896, %r2517; +$L__tmp233: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2520, %r1831, %r1629; +$L__tmp234: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2521, %rs386, %rs898, %r2519; +$L__tmp235: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2522, %r1831, %r1630; +$L__tmp236: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2523, %rs387, %rs899, %r2521; +$L__tmp237: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2524, %r1831, %r1631; +$L__tmp238: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2525, %rs388, %rs900, %r2523; +$L__tmp239: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2526, %r1831, %r1632; +$L__tmp240: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2527, %rs389, %rs901, %r2525; +$L__tmp241: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2528, %r1831, %r1633; +$L__tmp242: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2529, %rs390, %rs902, %r2527; +$L__tmp243: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2530, %r1831, %r1634; +$L__tmp244: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2531, %rs391, %rs903, %r2529; +$L__tmp245: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2532, %r1746, %r1635; + mul.f32 %r2533, %r1746, %r1636; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2534, %r1100, %r1612; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2535, %r1746, %r1637; +$L__tmp246: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2536, %rs376, %rs888, %r2534; +$L__tmp247: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2537, %r1746, %r1638; +$L__tmp248: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2538, %rs378, %rs890, %r2536; +$L__tmp249: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2539, %r1746, %r1639; +$L__tmp250: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2540, %rs379, %rs891, %r2538; +$L__tmp251: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2541, %r1746, %r1640; +$L__tmp252: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2542, %rs380, %rs892, %r2540; +$L__tmp253: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2543, %r1746, %r1641; +$L__tmp254: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2544, %rs381, %rs893, %r2542; +$L__tmp255: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2545, %r1746, %r1642; +$L__tmp256: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2546, %rs382, %rs894, %r2544; +$L__tmp257: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2547, %r1832, %r1643; +$L__tmp258: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2548, %rs383, %rs895, %r2546; +$L__tmp259: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2549, %r1092, %r1604; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2550, %r1832, %r1644; +$L__tmp260: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2551, %rs368, %rs880, %r2549; +$L__tmp261: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2552, %r1832, %r1645; +$L__tmp262: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2553, %rs370, %rs882, %r2551; +$L__tmp263: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2554, %r1832, %r1646; +$L__tmp264: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2555, %rs371, %rs883, %r2553; +$L__tmp265: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2556, %r1832, %r1647; +$L__tmp266: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2557, %rs372, %rs884, %r2555; +$L__tmp267: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2558, %r1832, %r1648; +$L__tmp268: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2559, %rs373, %rs885, %r2557; +$L__tmp269: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2560, %r1832, %r1649; +$L__tmp270: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2561, %rs374, %rs886, %r2559; +$L__tmp271: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2562, %r1832, %r1650; +$L__tmp272: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2563, %rs375, %rs887, %r2561; +$L__tmp273: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2564, %r1747, %r1651; + mul.f32 %r2565, %r1747, %r1652; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2566, %r1084, %r1596; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2567, %r1747, %r1653; +$L__tmp274: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2568, %rs360, %rs872, %r2566; +$L__tmp275: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2569, %r1747, %r1654; +$L__tmp276: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2570, %rs362, %rs874, %r2568; +$L__tmp277: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2571, %r1747, %r1655; +$L__tmp278: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2572, %rs363, %rs875, %r2570; +$L__tmp279: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2573, %r1747, %r1656; +$L__tmp280: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2574, %rs364, %rs876, %r2572; +$L__tmp281: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2575, %r1747, %r1657; +$L__tmp282: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2576, %rs365, %rs877, %r2574; +$L__tmp283: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2577, %r1747, %r1658; +$L__tmp284: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2578, %rs366, %rs878, %r2576; +$L__tmp285: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2579, %r1834, %r1659; +$L__tmp286: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2580, %rs367, %rs879, %r2578; +$L__tmp287: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2581, %r1076, %r1588; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2582, %r1834, %r1660; +$L__tmp288: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2583, %rs352, %rs864, %r2581; +$L__tmp289: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2584, %r1834, %r1661; +$L__tmp290: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2585, %rs354, %rs866, %r2583; +$L__tmp291: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2586, %r1834, %r1662; +$L__tmp292: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2587, %rs355, %rs867, %r2585; +$L__tmp293: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2588, %r1834, %r1663; +$L__tmp294: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2589, %rs356, %rs868, %r2587; +$L__tmp295: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2590, %r1834, %r1664; +$L__tmp296: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2591, %rs357, %rs869, %r2589; +$L__tmp297: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2592, %r1834, %r1665; +$L__tmp298: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2593, %rs358, %rs870, %r2591; +$L__tmp299: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2594, %r1834, %r1666; +$L__tmp300: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2595, %rs359, %rs871, %r2593; +$L__tmp301: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2596, %r1748, %r1667; + mul.f32 %r2597, %r1748, %r1668; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2598, %r1068, %r1580; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2599, %r1748, %r1669; +$L__tmp302: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2600, %rs344, %rs856, %r2598; +$L__tmp303: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2601, %r1748, %r1670; +$L__tmp304: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2602, %rs346, %rs858, %r2600; +$L__tmp305: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2603, %r1748, %r1671; +$L__tmp306: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2604, %rs347, %rs859, %r2602; +$L__tmp307: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2605, %r1748, %r1672; +$L__tmp308: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2606, %rs348, %rs860, %r2604; +$L__tmp309: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2607, %r1748, %r1673; +$L__tmp310: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2608, %rs349, %rs861, %r2606; +$L__tmp311: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2609, %r1748, %r1674; +$L__tmp312: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2610, %rs350, %rs862, %r2608; +$L__tmp313: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2611, %r1835, %r1675; +$L__tmp314: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2612, %rs351, %rs863, %r2610; +$L__tmp315: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2613, %r1060, %r1572; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2614, %r1835, %r1676; +$L__tmp316: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2615, %rs336, %rs848, %r2613; +$L__tmp317: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2616, %r1835, %r1677; +$L__tmp318: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2617, %rs338, %rs850, %r2615; +$L__tmp319: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2618, %r1835, %r1678; +$L__tmp320: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2619, %rs339, %rs851, %r2617; +$L__tmp321: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2620, %r1835, %r1679; +$L__tmp322: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2621, %rs340, %rs852, %r2619; +$L__tmp323: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2622, %r1835, %r1680; +$L__tmp324: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2623, %rs341, %rs853, %r2621; +$L__tmp325: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2624, %r1835, %r1681; +$L__tmp326: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2625, %rs342, %rs854, %r2623; +$L__tmp327: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2626, %r1835, %r1682; +$L__tmp328: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2627, %rs343, %rs855, %r2625; +$L__tmp329: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2628, %r1749, %r1683; + mul.f32 %r2629, %r1749, %r1684; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2630, %r1052, %r1564; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2631, %r1749, %r1685; +$L__tmp330: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2632, %rs328, %rs840, %r2630; +$L__tmp331: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2633, %r1749, %r1686; +$L__tmp332: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2634, %rs330, %rs842, %r2632; +$L__tmp333: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2635, %r1749, %r1687; +$L__tmp334: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2636, %rs331, %rs843, %r2634; +$L__tmp335: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2637, %r1749, %r1688; +$L__tmp336: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2638, %rs332, %rs844, %r2636; +$L__tmp337: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2639, %r1749, %r1689; +$L__tmp338: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2640, %rs333, %rs845, %r2638; +$L__tmp339: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2641, %r1749, %r1690; +$L__tmp340: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2642, %rs334, %rs846, %r2640; +$L__tmp341: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2643, %r1836, %r1691; +$L__tmp342: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2644, %rs335, %rs847, %r2642; +$L__tmp343: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2645, %r1044, %r1556; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2646, %r1836, %r1692; +$L__tmp344: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2647, %rs320, %rs832, %r2645; +$L__tmp345: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2648, %r1836, %r1693; +$L__tmp346: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2649, %rs322, %rs834, %r2647; +$L__tmp347: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2650, %r1836, %r1694; +$L__tmp348: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2651, %rs323, %rs835, %r2649; +$L__tmp349: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2652, %r1836, %r1695; +$L__tmp350: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2653, %rs324, %rs836, %r2651; +$L__tmp351: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2654, %r1836, %r1696; +$L__tmp352: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2655, %rs325, %rs837, %r2653; +$L__tmp353: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2656, %r1836, %r1697; +$L__tmp354: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2657, %rs326, %rs838, %r2655; +$L__tmp355: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2658, %r1836, %r1698; +$L__tmp356: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2659, %rs327, %rs839, %r2657; +$L__tmp357: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2660, %r1750, %r1699; + mul.f32 %r2661, %r1750, %r1700; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2662, %r1036, %r1548; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2663, %r1750, %r1701; +$L__tmp358: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2664, %rs312, %rs824, %r2662; +$L__tmp359: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2665, %r1750, %r1702; +$L__tmp360: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2666, %rs314, %rs826, %r2664; +$L__tmp361: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2667, %r1750, %r1703; +$L__tmp362: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2668, %rs315, %rs827, %r2666; +$L__tmp363: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2669, %r1750, %r1704; +$L__tmp364: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2670, %rs316, %rs828, %r2668; +$L__tmp365: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2671, %r1750, %r1705; +$L__tmp366: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2672, %rs317, %rs829, %r2670; +$L__tmp367: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2673, %r1750, %r1706; +$L__tmp368: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2674, %rs318, %rs830, %r2672; +$L__tmp369: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2675, %r1837, %r1707; +$L__tmp370: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2676, %rs319, %rs831, %r2674; +$L__tmp371: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2677, %r1028, %r1540; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2678, %r1837, %r1708; +$L__tmp372: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2679, %rs304, %rs816, %r2677; +$L__tmp373: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2680, %r1837, %r1709; +$L__tmp374: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2681, %rs306, %rs818, %r2679; +$L__tmp375: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2682, %r1837, %r1710; +$L__tmp376: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2683, %rs307, %rs819, %r2681; +$L__tmp377: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2684, %r1837, %r1711; +$L__tmp378: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2685, %rs308, %rs820, %r2683; +$L__tmp379: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2686, %r1837, %r1712; +$L__tmp380: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2687, %rs309, %rs821, %r2685; +$L__tmp381: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2688, %r1837, %r1713; +$L__tmp382: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2689, %rs310, %rs822, %r2687; +$L__tmp383: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2690, %r1837, %r1714; +$L__tmp384: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2691, %rs311, %rs823, %r2689; +$L__tmp385: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2692, %r1751, %r1715; + mul.f32 %r2693, %r1751, %r1716; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2694, %r1012, %r1524; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2695, %r1751, %r1717; +$L__tmp386: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2696, %rs288, %rs800, %r2694; +$L__tmp387: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2697, %r1751, %r1718; +$L__tmp388: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2698, %rs290, %rs802, %r2696; +$L__tmp389: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2699, %r1751, %r1719; +$L__tmp390: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2700, %rs291, %rs803, %r2698; +$L__tmp391: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2701, %r1751, %r1720; +$L__tmp392: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2702, %rs292, %rs804, %r2700; +$L__tmp393: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2703, %r1751, %r1721; +$L__tmp394: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2704, %rs293, %rs805, %r2702; +$L__tmp395: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2705, %r1751, %r1722; +$L__tmp396: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2706, %rs294, %rs806, %r2704; +$L__tmp397: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2707, %r1839, %r1723; +$L__tmp398: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2708, %rs295, %rs807, %r2706; +$L__tmp399: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2709, %r996, %r1508; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2710, %r1839, %r1724; +$L__tmp400: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2711, %rs272, %rs784, %r2709; +$L__tmp401: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2712, %r1839, %r1725; +$L__tmp402: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2713, %rs274, %rs786, %r2711; +$L__tmp403: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2714, %r1839, %r1726; +$L__tmp404: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2715, %rs275, %rs787, %r2713; +$L__tmp405: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2716, %r1839, %r1727; +$L__tmp406: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2717, %rs276, %rs788, %r2715; +$L__tmp407: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2718, %r1839, %r1730; +$L__tmp408: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2719, %rs277, %rs789, %r2717; +$L__tmp409: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2720, %r1839, %r1732; +$L__tmp410: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2721, %rs278, %rs790, %r2719; +$L__tmp411: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2722, %r1839, %r1767; +$L__tmp412: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2723, %rs279, %rs791, %r2721; +$L__tmp413: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2724, %r1752, %r1768; + mul.f32 %r2725, %r1752, %r1769; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2726, %r980, %r1492; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2727, %r1752, %r1770; +$L__tmp414: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2728, %rs256, %rs768, %r2726; +$L__tmp415: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2729, %r1752, %r1776; +$L__tmp416: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2730, %rs258, %rs770, %r2728; +$L__tmp417: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2731, %r1752, %r1778; +$L__tmp418: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2732, %rs259, %rs771, %r2730; +$L__tmp419: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2733, %r1752, %r1782; +$L__tmp420: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2734, %rs260, %rs772, %r2732; +$L__tmp421: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2735, %r1752, %r1784; +$L__tmp422: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2736, %rs261, %rs773, %r2734; +$L__tmp423: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2737, %r1752, %r1788; +$L__tmp424: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2738, %rs262, %rs774, %r2736; +$L__tmp425: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2739, %r1840, %r1790; +$L__tmp426: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2740, %rs263, %rs775, %r2738; +$L__tmp427: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2741, %r972, %r1484; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2742, %r1840, %r1791; +$L__tmp428: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2743, %rs248, %rs760, %r2741; +$L__tmp429: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2744, %r1840, %r1792; +$L__tmp430: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2745, %rs250, %rs762, %r2743; +$L__tmp431: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2746, %r1840, %r1793; +$L__tmp432: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2747, %rs251, %rs763, %r2745; +$L__tmp433: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2748, %r1840, %r1795; +$L__tmp434: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2749, %rs252, %rs764, %r2747; +$L__tmp435: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2750, %r1840, %r1797; +$L__tmp436: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2751, %rs253, %rs765, %r2749; +$L__tmp437: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2752, %r1840, %r1799; +$L__tmp438: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2753, %rs254, %rs766, %r2751; +$L__tmp439: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2754, %r1840, %r1810; +$L__tmp440: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2755, %rs255, %rs767, %r2753; +$L__tmp441: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2756, %r1753, %r1812; + mul.f32 %r2757, %r1753, %r1814; + mul.f32 %r2758, %r1753, %r1816; + mul.f32 %r2759, %r1753, %r1817; + mul.f32 %r2760, %r1753, %r1818; + mul.f32 %r2761, %r1753, %r1856; + mul.f32 %r2762, %r1753, %r1858; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2763, %r956, %r1468; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2764, %r1753, %r1860; +$L__tmp442: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2765, %rs232, %rs744, %r2763; +$L__tmp443: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2766, %r1841, %r1862; +$L__tmp444: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2767, %rs234, %rs746, %r2765; +$L__tmp445: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2768, %r948, %r1460; +$L__tmp446: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2769, %rs235, %rs747, %r2767; + fma.rn.f32.bf16 %r2770, %rs224, %rs736, %r2768; + fma.rn.f32.bf16 %r2771, %rs236, %rs748, %r2769; + fma.rn.f32.bf16 %r2772, %rs226, %rs738, %r2770; + fma.rn.f32.bf16 %r2773, %rs237, %rs749, %r2771; + fma.rn.f32.bf16 %r2774, %rs227, %rs739, %r2772; + fma.rn.f32.bf16 %r2775, %rs238, %rs750, %r2773; + fma.rn.f32.bf16 %r2776, %rs228, %rs740, %r2774; + fma.rn.f32.bf16 %r2777, %rs239, %rs751, %r2775; + fma.rn.f32.bf16 %r2778, %rs229, %rs741, %r2776; +$L__tmp447: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2779, %r1841, %r1880; +$L__tmp448: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2780, %rs230, %rs742, %r2778; +$L__tmp449: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2781, %r1841, %r1882; +$L__tmp450: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2782, %rs231, %rs743, %r2780; +$L__tmp451: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2783, %r1841, %r1900; +$L__tmp452: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2784, %r2782, 8, 31, -1; +$L__tmp453: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2785, %r1841, %r1902; +$L__tmp454: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2786, %r2782, %r2784; +$L__tmp455: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2787, %r1841, %r1904; +$L__tmp456: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2788, %r2786, 4, 31, -1; +$L__tmp457: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2789, %r1841, %r1906; +$L__tmp458: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2790, %r2786, %r2788; +$L__tmp459: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2791, %r1841, %r1924; +$L__tmp460: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2792, %r2790, 2, 31, -1; +$L__tmp461: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2793, %r1754, %r1926; +$L__tmp462: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2794, %r2790, %r2792; +$L__tmp463: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2795, %r964, %r1476; +$L__tmp464: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2796, %r2794, 1, 31, -1; +$L__tmp465: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2797, %rs240, %rs752, %r2795; +$L__tmp466: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2798, %r2777, 8, 31, -1; +$L__tmp467: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2799, %rs242, %rs754, %r2797; + add.f32 %r2800, %r2777, %r2798; + fma.rn.f32.bf16 %r2801, %rs243, %rs755, %r2799; +$L__tmp468: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2802, %r2800, 4, 31, -1; +$L__tmp469: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2803, %rs244, %rs756, %r2801; + add.f32 %r2804, %r2800, %r2802; + fma.rn.f32.bf16 %r2805, %rs245, %rs757, %r2803; +$L__tmp470: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2806, %r2804, 2, 31, -1; +$L__tmp471: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2807, %rs246, %rs758, %r2805; + add.f32 %r2808, %r2804, %r2806; + fma.rn.f32.bf16 %r2809, %rs247, %rs759, %r2807; +$L__tmp472: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2810, %r2808, 1, 31, -1; +$L__tmp473: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2811, %r1754, %r1928; +$L__tmp474: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2812, %r2809, 8, 31, -1; +$L__tmp475: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2813, %r1754, %r1930; +$L__tmp476: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2814, %r2809, %r2812; +$L__tmp477: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2815, %r1754, %r1948; +$L__tmp478: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2816, %r2814, 4, 31, -1; +$L__tmp479: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2817, %r1754, %r1950; +$L__tmp480: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2818, %r2814, %r2816; +$L__tmp481: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2819, %r1754, %r1955; +$L__tmp482: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2820, %r2818, 2, 31, -1; +$L__tmp483: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2821, %r1754, %r1957; +$L__tmp484: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2822, %r2818, %r2820; +$L__tmp485: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2823, %r1754, %r1964; +$L__tmp486: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2824, %r2822, 1, 31, -1; +$L__tmp487: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2825, %r1842, %r1966; +$L__tmp488: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2826, %r2755, 8, 31, -1; +$L__tmp489: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2827, %r988, %r1500; +$L__tmp490: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2828, %r2755, %r2826; + fma.rn.f32.bf16 %r2829, %rs264, %rs776, %r2827; +$L__tmp491: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2830, %r2828, 4, 31, -1; +$L__tmp492: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2831, %rs266, %rs778, %r2829; + add.f32 %r2832, %r2828, %r2830; + fma.rn.f32.bf16 %r2833, %rs267, %rs779, %r2831; +$L__tmp493: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2834, %r2832, 2, 31, -1; +$L__tmp494: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2835, %rs268, %rs780, %r2833; + add.f32 %r2836, %r2832, %r2834; + fma.rn.f32.bf16 %r2837, %rs269, %rs781, %r2835; +$L__tmp495: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2838, %r2836, 1, 31, -1; +$L__tmp496: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2839, %rs270, %rs782, %r2837; +$L__tmp497: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2840, %r2740, 8, 31, -1; +$L__tmp498: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2841, %rs271, %rs783, %r2839; + add.f32 %r2842, %r2740, %r2840; +$L__tmp499: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2843, %r1842, %r1967; +$L__tmp500: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2844, %r2842, 4, 31, -1; +$L__tmp501: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2845, %r1842, %r1969; +$L__tmp502: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2846, %r2842, %r2844; +$L__tmp503: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2847, %r1842, %r1972; +$L__tmp504: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2848, %r2846, 2, 31, -1; +$L__tmp505: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2849, %r1842, %r1974; +$L__tmp506: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2850, %r2846, %r2848; +$L__tmp507: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2851, %r1842, %r1975; +$L__tmp508: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2852, %r2850, 1, 31, -1; +$L__tmp509: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2853, %r1842, %r1977; +$L__tmp510: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2854, %r2841, 8, 31, -1; +$L__tmp511: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2855, %r1842, %r1978; +$L__tmp512: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2856, %r2841, %r2854; +$L__tmp513: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2857, %r1755, %r1980; +$L__tmp514: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2858, %r2856, 4, 31, -1; +$L__tmp515: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2859, %r1004, %r1516; +$L__tmp516: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2860, %r2856, %r2858; + fma.rn.f32.bf16 %r2861, %rs280, %rs792, %r2859; +$L__tmp517: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2862, %r2860, 2, 31, -1; +$L__tmp518: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2863, %rs282, %rs794, %r2861; + add.f32 %r2864, %r2860, %r2862; + fma.rn.f32.bf16 %r2865, %rs283, %rs795, %r2863; +$L__tmp519: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2866, %r2864, 1, 31, -1; +$L__tmp520: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2867, %rs284, %rs796, %r2865; +$L__tmp521: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2868, %r2723, 8, 31, -1; +$L__tmp522: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2869, %rs285, %rs797, %r2867; + add.f32 %r2870, %r2723, %r2868; + fma.rn.f32.bf16 %r2871, %rs286, %rs798, %r2869; +$L__tmp523: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2872, %r2870, 4, 31, -1; +$L__tmp524: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2873, %rs287, %rs799, %r2871; + add.f32 %r2874, %r2870, %r2872; +$L__tmp525: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2875, %r1755, %r1981; +$L__tmp526: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2876, %r2874, 2, 31, -1; +$L__tmp527: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2877, %r1755, %r1982; +$L__tmp528: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2878, %r2874, %r2876; +$L__tmp529: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2879, %r1755, %r1985; +$L__tmp530: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2880, %r2878, 1, 31, -1; +$L__tmp531: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2881, %r1755, %r1987; +$L__tmp532: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2882, %r2873, 8, 31, -1; +$L__tmp533: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2883, %r1755, %r1988; +$L__tmp534: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2884, %r2873, %r2882; +$L__tmp535: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2885, %r1755, %r1990; +$L__tmp536: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2886, %r2884, 4, 31, -1; +$L__tmp537: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2887, %r1755, %r1993; +$L__tmp538: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2888, %r2884, %r2886; +$L__tmp539: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2889, %r1844, %r1995; +$L__tmp540: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2890, %r2888, 2, 31, -1; +$L__tmp541: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2891, %r1020, %r1532; +$L__tmp542: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2892, %r2888, %r2890; + fma.rn.f32.bf16 %r2893, %rs296, %rs808, %r2891; +$L__tmp543: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2894, %r2892, 1, 31, -1; +$L__tmp544: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2895, %rs298, %rs810, %r2893; +$L__tmp545: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2896, %r2708, 8, 31, -1; +$L__tmp546: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2897, %rs299, %rs811, %r2895; + add.f32 %r2898, %r2708, %r2896; + fma.rn.f32.bf16 %r2899, %rs300, %rs812, %r2897; +$L__tmp547: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2900, %r2898, 4, 31, -1; +$L__tmp548: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2901, %rs301, %rs813, %r2899; + add.f32 %r2902, %r2898, %r2900; + fma.rn.f32.bf16 %r2903, %rs302, %rs814, %r2901; +$L__tmp549: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2904, %r2902, 2, 31, -1; +$L__tmp550: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2905, %rs303, %rs815, %r2903; + add.f32 %r2906, %r2902, %r2904; +$L__tmp551: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2907, %r1844, %r1996; +$L__tmp552: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2908, %r2906, 1, 31, -1; +$L__tmp553: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2909, %r1844, %r1998; +$L__tmp554: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2910, %r2905, 8, 31, -1; +$L__tmp555: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2911, %r1844, %r1999; +$L__tmp556: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2912, %r2905, %r2910; +$L__tmp557: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2913, %r1844, %r2001; +$L__tmp558: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2914, %r2912, 4, 31, -1; +$L__tmp559: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2915, %r1844, %r2004; +$L__tmp560: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2916, %r2912, %r2914; +$L__tmp561: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2917, %r1844, %r2005; +$L__tmp562: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2918, %r2916, 2, 31, -1; +$L__tmp563: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2919, %r1844, %r2006; +$L__tmp564: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2920, %r2916, %r2918; +$L__tmp565: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2921, %r1756, %r2008; +$L__tmp566: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2922, %r2920, 1, 31, -1; +$L__tmp567: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2923, %r1756, %r2011; +$L__tmp568: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2924, %r2691, 8, 31, -1; +$L__tmp569: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2925, %r1756, %r2013; +$L__tmp570: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2926, %r2691, %r2924; +$L__tmp571: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2927, %r1756, %r2014; +$L__tmp572: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2928, %r2926, 4, 31, -1; +$L__tmp573: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2929, %r1756, %r2016; +$L__tmp574: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2930, %r2926, %r2928; +$L__tmp575: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2931, %r1756, %r2017; +$L__tmp576: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2932, %r2930, 2, 31, -1; +$L__tmp577: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2933, %r1756, %r2019; +$L__tmp578: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2934, %r2930, %r2932; +$L__tmp579: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2935, %r1756, %r2020; +$L__tmp580: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2936, %r2934, 1, 31, -1; +$L__tmp581: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2937, %r1845, %r2021; +$L__tmp582: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2938, %r2676, 8, 31, -1; +$L__tmp583: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2939, %r1845, %r2024; +$L__tmp584: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2940, %r2676, %r2938; +$L__tmp585: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2941, %r1845, %r2026; +$L__tmp586: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2942, %r2940, 4, 31, -1; +$L__tmp587: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2943, %r1845, %r2027; +$L__tmp588: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2944, %r2940, %r2942; +$L__tmp589: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2945, %r1845, %r2029; +$L__tmp590: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2946, %r2944, 2, 31, -1; +$L__tmp591: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2947, %r1845, %r2032; +$L__tmp592: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2948, %r2944, %r2946; +$L__tmp593: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2949, %r1845, %r2034; +$L__tmp594: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2950, %r2948, 1, 31, -1; +$L__tmp595: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2951, %r1845, %r2035; +$L__tmp596: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2952, %r2659, 8, 31, -1; +$L__tmp597: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2953, %r1757, %r2037; +$L__tmp598: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2954, %r2659, %r2952; +$L__tmp599: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2955, %r1757, %r2038; +$L__tmp600: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2956, %r2954, 4, 31, -1; +$L__tmp601: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2957, %r1757, %r2040; +$L__tmp602: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2958, %r2954, %r2956; +$L__tmp603: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2959, %r1757, %r2043; +$L__tmp604: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2960, %r2958, 2, 31, -1; +$L__tmp605: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2961, %r1757, %r2044; +$L__tmp606: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2962, %r2958, %r2960; +$L__tmp607: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2963, %r1757, %r2045; +$L__tmp608: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2964, %r2962, 1, 31, -1; +$L__tmp609: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2965, %r1757, %r2047; +$L__tmp610: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2966, %r2644, 8, 31, -1; +$L__tmp611: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2967, %r1757, %r2050; +$L__tmp612: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2968, %r2644, %r2966; +$L__tmp613: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2969, %r1846, %r2052; +$L__tmp614: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2970, %r2968, 4, 31, -1; +$L__tmp615: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2971, %r1846, %r2053; +$L__tmp616: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2972, %r2968, %r2970; +$L__tmp617: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2973, %r1846, %r2055; +$L__tmp618: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2974, %r2972, 2, 31, -1; +$L__tmp619: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2975, %r1846, %r2056; +$L__tmp620: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2976, %r2972, %r2974; +$L__tmp621: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2977, %r1846, %r2058; +$L__tmp622: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2978, %r2976, 1, 31, -1; +$L__tmp623: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2979, %r1846, %r2059; +$L__tmp624: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2980, %r2627, 8, 31, -1; +$L__tmp625: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2981, %r1846, %r2060; +$L__tmp626: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2982, %r2627, %r2980; +$L__tmp627: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2983, %r1846, %r2063; +$L__tmp628: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2984, %r2982, 4, 31, -1; +$L__tmp629: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2985, %r1758, %r2065; +$L__tmp630: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2986, %r2982, %r2984; +$L__tmp631: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2987, %r1758, %r2066; +$L__tmp632: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2988, %r2986, 2, 31, -1; +$L__tmp633: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2989, %r1758, %r2068; +$L__tmp634: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2990, %r2986, %r2988; +$L__tmp635: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2991, %r1758, %r2071; +$L__tmp636: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2992, %r2990, 1, 31, -1; +$L__tmp637: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2993, %r1758, %r2073; +$L__tmp638: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2994, %r2612, 8, 31, -1; +$L__tmp639: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2995, %r1758, %r2074; +$L__tmp640: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2996, %r2612, %r2994; +$L__tmp641: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2997, %r1758, %r2075; +$L__tmp642: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2998, %r2996, 4, 31, -1; +$L__tmp643: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2999, %r1758, %r2076; +$L__tmp644: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3000, %r2996, %r2998; +$L__tmp645: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3001, %r1847, %r2077; +$L__tmp646: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3002, %r3000, 2, 31, -1; +$L__tmp647: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3003, %r1847, %r2078; +$L__tmp648: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3004, %r3000, %r3002; +$L__tmp649: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3005, %r1847, %r2080; +$L__tmp650: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3006, %r3004, 1, 31, -1; +$L__tmp651: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3007, %r1847, %r2095; +$L__tmp652: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3008, %r2595, 8, 31, -1; +$L__tmp653: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3009, %r1847, %r2097; +$L__tmp654: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3010, %r2595, %r3008; +$L__tmp655: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3011, %r1847, %r2098; +$L__tmp656: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3012, %r3010, 4, 31, -1; +$L__tmp657: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3013, %r1847, %r2099; +$L__tmp658: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3014, %r3010, %r3012; +$L__tmp659: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3015, %r1847, %r2100; +$L__tmp660: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3016, %r3014, 2, 31, -1; +$L__tmp661: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3017, %r1759, %r2101; +$L__tmp662: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3018, %r3014, %r3016; +$L__tmp663: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3019, %r1759, %r2102; +$L__tmp664: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3020, %r3018, 1, 31, -1; +$L__tmp665: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3021, %r1759, %r2104; +$L__tmp666: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3022, %r2580, 8, 31, -1; +$L__tmp667: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3023, %r1759, %r2119; +$L__tmp668: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3024, %r2580, %r3022; +$L__tmp669: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3025, %r1759, %r2121; +$L__tmp670: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3026, %r3024, 4, 31, -1; +$L__tmp671: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3027, %r1759, %r2122; +$L__tmp672: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3028, %r3024, %r3026; +$L__tmp673: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3029, %r1759, %r2123; +$L__tmp674: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3030, %r3028, 2, 31, -1; +$L__tmp675: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3031, %r1759, %r2124; +$L__tmp676: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3032, %r3028, %r3030; +$L__tmp677: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3033, %r1849, %r2125; +$L__tmp678: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3034, %r3032, 1, 31, -1; +$L__tmp679: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3035, %r1849, %r2126; +$L__tmp680: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3036, %r2563, 8, 31, -1; +$L__tmp681: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3037, %r1849, %r2128; +$L__tmp682: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3038, %r2563, %r3036; +$L__tmp683: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3039, %r1849, %r2143; +$L__tmp684: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3040, %r3038, 4, 31, -1; +$L__tmp685: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3041, %r1849, %r2145; +$L__tmp686: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3042, %r3038, %r3040; +$L__tmp687: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3043, %r1849, %r2146; +$L__tmp688: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3044, %r3042, 2, 31, -1; +$L__tmp689: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3045, %r1849, %r2147; +$L__tmp690: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3046, %r3042, %r3044; +$L__tmp691: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3047, %r1849, %r2148; +$L__tmp692: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3048, %r3046, 1, 31, -1; +$L__tmp693: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3049, %r1760, %r2149; +$L__tmp694: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3050, %r2548, 8, 31, -1; +$L__tmp695: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3051, %r1760, %r2150; +$L__tmp696: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3052, %r2548, %r3050; +$L__tmp697: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3053, %r1760, %r2152; +$L__tmp698: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3054, %r3052, 4, 31, -1; +$L__tmp699: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3055, %r1760, %r2167; +$L__tmp700: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3056, %r3052, %r3054; +$L__tmp701: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3057, %r1760, %r2169; +$L__tmp702: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3058, %r3056, 2, 31, -1; +$L__tmp703: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3059, %r1760, %r2170; +$L__tmp704: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3060, %r3056, %r3058; +$L__tmp705: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3061, %r1760, %r2171; +$L__tmp706: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3062, %r3060, 1, 31, -1; +$L__tmp707: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3063, %r1760, %r2172; +$L__tmp708: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3064, %r2531, 8, 31, -1; +$L__tmp709: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3065, %r1850, %r2173; +$L__tmp710: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3066, %r2531, %r3064; +$L__tmp711: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3067, %r1850, %r2174; +$L__tmp712: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3068, %r3066, 4, 31, -1; +$L__tmp713: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3069, %r1850, %r2176; +$L__tmp714: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3070, %r3066, %r3068; +$L__tmp715: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3071, %r1850, %r2191; +$L__tmp716: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3072, %r3070, 2, 31, -1; +$L__tmp717: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3073, %r1850, %r2193; +$L__tmp718: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3074, %r3070, %r3072; +$L__tmp719: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3075, %r1850, %r2194; +$L__tmp720: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3076, %r3074, 1, 31, -1; +$L__tmp721: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3077, %r1850, %r2195; +$L__tmp722: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3078, %r2516, 8, 31, -1; +$L__tmp723: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3079, %r1850, %r2196; +$L__tmp724: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3080, %r2516, %r3078; +$L__tmp725: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3081, %r1761, %r2197; +$L__tmp726: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3082, %r3080, 4, 31, -1; +$L__tmp727: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3083, %r1761, %r2198; +$L__tmp728: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3084, %r3080, %r3082; +$L__tmp729: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3085, %r1761, %r2200; +$L__tmp730: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3086, %r3084, 2, 31, -1; +$L__tmp731: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3087, %r1761, %r2215; +$L__tmp732: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3088, %r3084, %r3086; +$L__tmp733: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3089, %r1761, %r2217; +$L__tmp734: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3090, %r3088, 1, 31, -1; +$L__tmp735: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3091, %r1761, %r2218; +$L__tmp736: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3092, %r2324, 8, 31, -1; +$L__tmp737: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3093, %r1761, %r2219; +$L__tmp738: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3094, %r2324, %r3092; +$L__tmp739: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3095, %r1761, %r2220; +$L__tmp740: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3096, %r3094, 4, 31, -1; +$L__tmp741: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3097, %r1851, %r2221; +$L__tmp742: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3098, %r3094, %r3096; +$L__tmp743: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3099, %r1851, %r2222; +$L__tmp744: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3100, %r3098, 2, 31, -1; +$L__tmp745: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3101, %r1851, %r2224; +$L__tmp746: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3102, %r3098, %r3100; +$L__tmp747: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3103, %r1851, %r2239; +$L__tmp748: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3104, %r3102, 1, 31, -1; +$L__tmp749: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3105, %r1851, %r2241; +$L__tmp750: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3106, %r2309, 8, 31, -1; +$L__tmp751: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3107, %r1851, %r2242; +$L__tmp752: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3108, %r2309, %r3106; +$L__tmp753: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3109, %r1851, %r2243; +$L__tmp754: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3110, %r3108, 4, 31, -1; +$L__tmp755: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3111, %r1851, %r2244; +$L__tmp756: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3112, %r3108, %r3110; +$L__tmp757: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3113, %r1762, %r2245; +$L__tmp758: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3114, %r3112, 2, 31, -1; +$L__tmp759: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3115, %r1762, %r2246; +$L__tmp760: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3116, %r3112, %r3114; +$L__tmp761: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3117, %r1762, %r2248; +$L__tmp762: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3118, %r3116, 1, 31, -1; +$L__tmp763: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3119, %r1762, %r2263; +$L__tmp764: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3120, %r2288, 8, 31, -1; +$L__tmp765: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3121, %r1762, %r2265; +$L__tmp766: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3122, %r2288, %r3120; +$L__tmp767: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3123, %r1762, %r2266; +$L__tmp768: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3124, %r3122, 4, 31, -1; +$L__tmp769: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3125, %r1762, %r2267; +$L__tmp770: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3126, %r3122, %r3124; +$L__tmp771: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3127, %r1762, %r2268; +$L__tmp772: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3128, %r3126, 2, 31, -1; +$L__tmp773: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3129, %r1852, %r2269; +$L__tmp774: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3130, %r3126, %r3128; +$L__tmp775: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3131, %r1852, %r2270; +$L__tmp776: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3132, %r3130, 1, 31, -1; +$L__tmp777: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3133, %r1852, %r2272; +$L__tmp778: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3134, %r2285, 8, 31, -1; +$L__tmp779: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3135, %r1852, %r2287; +$L__tmp780: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3136, %r2285, %r3134; +$L__tmp781: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3137, %r1852, %r2289; +$L__tmp782: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3138, %r3136, 4, 31, -1; +$L__tmp783: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3139, %r1852, %r2290; +$L__tmp784: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3140, %r3136, %r3138; +$L__tmp785: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3141, %r1852, %r2291; +$L__tmp786: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3142, %r3140, 2, 31, -1; +$L__tmp787: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3143, %r1852, %r2292; +$L__tmp788: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3144, %r3140, %r3142; +$L__tmp789: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r3145, %r1284, %r2011; +$L__tmp790: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3146, %r3144, 1, 31, -1; +$L__tmp791: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r3147, %r1292, %r2024; +$L__tmp792: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3148, %r2264, 8, 31, -1; +$L__tmp793: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r3149, %r1300, %r2038; +$L__tmp794: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3150, %r2264, %r3148; +$L__tmp795: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r3151, %r1308, %r2053; +$L__tmp796: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3152, %r3150, 4, 31, -1; +$L__tmp797: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r3153, %r1316, %r2066; +$L__tmp798: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3154, %r3150, %r3152; +$L__tmp799: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r3155, %r1324, %r2078; +$L__tmp800: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3156, %r3154, 2, 31, -1; +$L__tmp801: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r3157, %r1332, %r2102; +$L__tmp802: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3158, %r3154, %r3156; +$L__tmp803: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r3159, %r1340, %r2126; +$L__tmp804: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3160, %r3158, 1, 31, -1; +$L__tmp805: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r3161, %r1348, %r2150; +$L__tmp806: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3162, %r2261, 8, 31, -1; +$L__tmp807: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r3163, %r1356, %r2174; +$L__tmp808: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3164, %r2261, %r3162; +$L__tmp809: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r3165, %r1364, %r2198; +$L__tmp810: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3166, %r3164, 4, 31, -1; +$L__tmp811: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r3167, %r1372, %r2222; +$L__tmp812: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3168, %r3164, %r3166; +$L__tmp813: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r3169, %r1380, %r2246; +$L__tmp814: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3170, %r3168, 2, 31, -1; +$L__tmp815: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r3171, %r1388, %r2270; +$L__tmp816: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3172, %r3168, %r3170; + fma.rn.f32.bf16 %r3173, %rs560, %rs1072, %r3145; +$L__tmp817: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3174, %r3172, 1, 31, -1; +$L__tmp818: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3175, %rs562, %rs1074, %r3173; +$L__tmp819: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3176, %r2240, 8, 31, -1; +$L__tmp820: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3177, %rs563, %rs1075, %r3175; + add.f32 %r3178, %r2240, %r3176; + fma.rn.f32.bf16 %r3179, %rs564, %rs1076, %r3177; +$L__tmp821: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3180, %r3178, 4, 31, -1; +$L__tmp822: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3181, %rs565, %rs1077, %r3179; + add.f32 %r3182, %r3178, %r3180; + fma.rn.f32.bf16 %r3183, %rs566, %rs1078, %r3181; +$L__tmp823: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3184, %r3182, 2, 31, -1; +$L__tmp824: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3185, %rs567, %rs1079, %r3183; + add.f32 %r3186, %r3182, %r3184; + fma.rn.f32.bf16 %r3187, %rs568, %rs1080, %r3147; +$L__tmp825: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3188, %r3186, 1, 31, -1; +$L__tmp826: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3189, %rs570, %rs1082, %r3187; +$L__tmp827: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3190, %r2237, 8, 31, -1; +$L__tmp828: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3191, %rs571, %rs1083, %r3189; + add.f32 %r3192, %r2237, %r3190; + fma.rn.f32.bf16 %r3193, %rs572, %rs1084, %r3191; +$L__tmp829: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3194, %r3192, 4, 31, -1; +$L__tmp830: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3195, %rs573, %rs1085, %r3193; + add.f32 %r3196, %r3192, %r3194; + fma.rn.f32.bf16 %r3197, %rs574, %rs1086, %r3195; +$L__tmp831: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3198, %r3196, 2, 31, -1; +$L__tmp832: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3199, %rs575, %rs1087, %r3197; + add.f32 %r3200, %r3196, %r3198; + fma.rn.f32.bf16 %r3201, %rs576, %rs1088, %r3149; +$L__tmp833: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3202, %r3200, 1, 31, -1; +$L__tmp834: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3203, %rs578, %rs1090, %r3201; +$L__tmp835: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3204, %r2216, 8, 31, -1; +$L__tmp836: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3205, %rs579, %rs1091, %r3203; + add.f32 %r3206, %r2216, %r3204; + fma.rn.f32.bf16 %r3207, %rs580, %rs1092, %r3205; +$L__tmp837: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3208, %r3206, 4, 31, -1; +$L__tmp838: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3209, %rs581, %rs1093, %r3207; + add.f32 %r3210, %r3206, %r3208; + fma.rn.f32.bf16 %r3211, %rs582, %rs1094, %r3209; +$L__tmp839: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3212, %r3210, 2, 31, -1; +$L__tmp840: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3213, %rs583, %rs1095, %r3211; + add.f32 %r3214, %r3210, %r3212; + fma.rn.f32.bf16 %r3215, %rs584, %rs1096, %r3151; +$L__tmp841: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3216, %r3214, 1, 31, -1; +$L__tmp842: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3217, %rs586, %rs1098, %r3215; +$L__tmp843: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3218, %r2213, 8, 31, -1; +$L__tmp844: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3219, %rs587, %rs1099, %r3217; + add.f32 %r3220, %r2213, %r3218; + fma.rn.f32.bf16 %r3221, %rs588, %rs1100, %r3219; +$L__tmp845: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3222, %r3220, 4, 31, -1; +$L__tmp846: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3223, %rs589, %rs1101, %r3221; + add.f32 %r3224, %r3220, %r3222; + fma.rn.f32.bf16 %r3225, %rs590, %rs1102, %r3223; +$L__tmp847: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3226, %r3224, 2, 31, -1; +$L__tmp848: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3227, %rs591, %rs1103, %r3225; + add.f32 %r3228, %r3224, %r3226; + fma.rn.f32.bf16 %r3229, %rs592, %rs1104, %r3153; +$L__tmp849: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3230, %r3228, 1, 31, -1; +$L__tmp850: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3231, %rs594, %rs1106, %r3229; +$L__tmp851: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3232, %r2192, 8, 31, -1; +$L__tmp852: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3233, %rs595, %rs1107, %r3231; + add.f32 %r3234, %r2192, %r3232; + fma.rn.f32.bf16 %r3235, %rs596, %rs1108, %r3233; +$L__tmp853: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3236, %r3234, 4, 31, -1; +$L__tmp854: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3237, %rs597, %rs1109, %r3235; + add.f32 %r3238, %r3234, %r3236; + fma.rn.f32.bf16 %r3239, %rs598, %rs1110, %r3237; +$L__tmp855: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3240, %r3238, 2, 31, -1; +$L__tmp856: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3241, %rs599, %rs1111, %r3239; + add.f32 %r3242, %r3238, %r3240; + fma.rn.f32.bf16 %r3243, %rs600, %rs1112, %r3155; +$L__tmp857: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3244, %r3242, 1, 31, -1; +$L__tmp858: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3245, %rs602, %rs1114, %r3243; +$L__tmp859: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3246, %r2189, 8, 31, -1; +$L__tmp860: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3247, %rs603, %rs1115, %r3245; + add.f32 %r3248, %r2189, %r3246; + fma.rn.f32.bf16 %r3249, %rs604, %rs1116, %r3247; +$L__tmp861: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3250, %r3248, 4, 31, -1; +$L__tmp862: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3251, %rs605, %rs1117, %r3249; + add.f32 %r3252, %r3248, %r3250; + fma.rn.f32.bf16 %r3253, %rs606, %rs1118, %r3251; +$L__tmp863: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3254, %r3252, 2, 31, -1; +$L__tmp864: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3255, %rs607, %rs1119, %r3253; + add.f32 %r3256, %r3252, %r3254; + fma.rn.f32.bf16 %r3257, %rs608, %rs1120, %r3157; +$L__tmp865: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3258, %r3256, 1, 31, -1; +$L__tmp866: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3259, %rs610, %rs1122, %r3257; +$L__tmp867: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3260, %r2168, 8, 31, -1; +$L__tmp868: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3261, %rs611, %rs1123, %r3259; + add.f32 %r3262, %r2168, %r3260; + fma.rn.f32.bf16 %r3263, %rs612, %rs1124, %r3261; +$L__tmp869: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3264, %r3262, 4, 31, -1; +$L__tmp870: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3265, %rs613, %rs1125, %r3263; + add.f32 %r3266, %r3262, %r3264; + fma.rn.f32.bf16 %r3267, %rs614, %rs1126, %r3265; +$L__tmp871: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3268, %r3266, 2, 31, -1; +$L__tmp872: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3269, %rs615, %rs1127, %r3267; + add.f32 %r3270, %r3266, %r3268; + fma.rn.f32.bf16 %r3271, %rs616, %rs1128, %r3159; +$L__tmp873: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3272, %r3270, 1, 31, -1; +$L__tmp874: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3273, %rs618, %rs1130, %r3271; +$L__tmp875: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3274, %r2165, 8, 31, -1; +$L__tmp876: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3275, %rs619, %rs1131, %r3273; + add.f32 %r3276, %r2165, %r3274; + fma.rn.f32.bf16 %r3277, %rs620, %rs1132, %r3275; +$L__tmp877: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3278, %r3276, 4, 31, -1; +$L__tmp878: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3279, %rs621, %rs1133, %r3277; + add.f32 %r3280, %r3276, %r3278; + fma.rn.f32.bf16 %r3281, %rs622, %rs1134, %r3279; +$L__tmp879: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3282, %r3280, 2, 31, -1; +$L__tmp880: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3283, %rs623, %rs1135, %r3281; + add.f32 %r3284, %r3280, %r3282; + fma.rn.f32.bf16 %r3285, %rs624, %rs1136, %r3161; +$L__tmp881: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3286, %r3284, 1, 31, -1; +$L__tmp882: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3287, %rs626, %rs1138, %r3285; +$L__tmp883: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3288, %r2144, 8, 31, -1; +$L__tmp884: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3289, %rs627, %rs1139, %r3287; + add.f32 %r3290, %r2144, %r3288; + fma.rn.f32.bf16 %r3291, %rs628, %rs1140, %r3289; +$L__tmp885: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3292, %r3290, 4, 31, -1; +$L__tmp886: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3293, %rs629, %rs1141, %r3291; + add.f32 %r3294, %r3290, %r3292; + fma.rn.f32.bf16 %r3295, %rs630, %rs1142, %r3293; +$L__tmp887: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3296, %r3294, 2, 31, -1; +$L__tmp888: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3297, %rs631, %rs1143, %r3295; + add.f32 %r3298, %r3294, %r3296; + fma.rn.f32.bf16 %r3299, %rs632, %rs1144, %r3163; +$L__tmp889: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3300, %r3298, 1, 31, -1; +$L__tmp890: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3301, %rs634, %rs1146, %r3299; +$L__tmp891: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3302, %r2141, 8, 31, -1; +$L__tmp892: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3303, %rs635, %rs1147, %r3301; + add.f32 %r3304, %r2141, %r3302; + fma.rn.f32.bf16 %r3305, %rs636, %rs1148, %r3303; +$L__tmp893: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3306, %r3304, 4, 31, -1; +$L__tmp894: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3307, %rs637, %rs1149, %r3305; + add.f32 %r3308, %r3304, %r3306; + fma.rn.f32.bf16 %r3309, %rs638, %rs1150, %r3307; +$L__tmp895: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3310, %r3308, 2, 31, -1; +$L__tmp896: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3311, %rs639, %rs1151, %r3309; + add.f32 %r3312, %r3308, %r3310; + fma.rn.f32.bf16 %r3313, %rs640, %rs1152, %r3165; +$L__tmp897: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3314, %r3312, 1, 31, -1; +$L__tmp898: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3315, %rs642, %rs1154, %r3313; +$L__tmp899: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3316, %r2120, 8, 31, -1; +$L__tmp900: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3317, %rs643, %rs1155, %r3315; + add.f32 %r3318, %r2120, %r3316; + fma.rn.f32.bf16 %r3319, %rs644, %rs1156, %r3317; +$L__tmp901: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3320, %r3318, 4, 31, -1; +$L__tmp902: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3321, %rs645, %rs1157, %r3319; + add.f32 %r3322, %r3318, %r3320; + fma.rn.f32.bf16 %r3323, %rs646, %rs1158, %r3321; +$L__tmp903: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3324, %r3322, 2, 31, -1; +$L__tmp904: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3325, %rs647, %rs1159, %r3323; + add.f32 %r3326, %r3322, %r3324; + fma.rn.f32.bf16 %r3327, %rs648, %rs1160, %r3167; +$L__tmp905: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3328, %r3326, 1, 31, -1; +$L__tmp906: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3329, %rs650, %rs1162, %r3327; +$L__tmp907: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3330, %r2117, 8, 31, -1; +$L__tmp908: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3331, %rs651, %rs1163, %r3329; + add.f32 %r3332, %r2117, %r3330; + fma.rn.f32.bf16 %r3333, %rs652, %rs1164, %r3331; +$L__tmp909: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3334, %r3332, 4, 31, -1; +$L__tmp910: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3335, %rs653, %rs1165, %r3333; + add.f32 %r3336, %r3332, %r3334; + fma.rn.f32.bf16 %r3337, %rs654, %rs1166, %r3335; +$L__tmp911: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3338, %r3336, 2, 31, -1; +$L__tmp912: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3339, %rs655, %rs1167, %r3337; + add.f32 %r3340, %r3336, %r3338; + fma.rn.f32.bf16 %r3341, %rs656, %rs1168, %r3169; +$L__tmp913: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3342, %r3340, 1, 31, -1; +$L__tmp914: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3343, %rs658, %rs1170, %r3341; +$L__tmp915: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3344, %r2096, 8, 31, -1; +$L__tmp916: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3345, %rs659, %rs1171, %r3343; + add.f32 %r3346, %r2096, %r3344; + fma.rn.f32.bf16 %r3347, %rs660, %rs1172, %r3345; +$L__tmp917: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3348, %r3346, 4, 31, -1; +$L__tmp918: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3349, %rs661, %rs1173, %r3347; + add.f32 %r3350, %r3346, %r3348; + fma.rn.f32.bf16 %r3351, %rs662, %rs1174, %r3349; +$L__tmp919: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3352, %r3350, 2, 31, -1; +$L__tmp920: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3353, %rs663, %rs1175, %r3351; + add.f32 %r3354, %r3350, %r3352; + fma.rn.f32.bf16 %r3355, %rs664, %rs1176, %r3171; +$L__tmp921: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3356, %r3354, 1, 31, -1; +$L__tmp922: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3357, %rs666, %rs1178, %r3355; +$L__tmp923: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3358, %r2093, 8, 31, -1; +$L__tmp924: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3359, %rs667, %rs1179, %r3357; + add.f32 %r3360, %r2093, %r3358; + fma.rn.f32.bf16 %r3361, %rs668, %rs1180, %r3359; +$L__tmp925: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3362, %r3360, 4, 31, -1; +$L__tmp926: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3363, %rs669, %rs1181, %r3361; + add.f32 %r3364, %r3360, %r3362; + fma.rn.f32.bf16 %r3365, %rs670, %rs1182, %r3363; +$L__tmp927: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3366, %r3364, 2, 31, -1; +$L__tmp928: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3367, %rs671, %rs1183, %r3365; + add.f32 %r3368, %r3364, %r3366; + add.f32 %r3369, %r1865, %r1867; +$L__tmp929: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3370, %r3368, 1, 31, -1; +$L__tmp930: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3371, %r1879, %r1881; +$L__tmp931: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3372, %r3185, 8, 31, -1; +$L__tmp932: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd675, {%r2335, %r2336}; +$L__tmp933: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3373, %r3185, %r3372; + add.f32 %r3374, %r1893, %r1895; +$L__tmp934: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3375, %r3373, 4, 31, -1; +$L__tmp935: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3376, %r1907, %r1909; + add.f32 %r3377, %r3373, %r3375; +$L__tmp936: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd676, {%r2333, %r2334}; +$L__tmp937: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3378, %r3377, 2, 31, -1; +$L__tmp938: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3379, %r2794, %r2796; + add.f32 %r3380, %r3377, %r3378; + add.f32 %r3381, %r2808, %r2810; +$L__tmp939: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3382, %r3380, 1, 31, -1; +$L__tmp940: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3383, %r2064, %r917; +$L__tmp941: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3384, %r3199, 8, 31, -1; +$L__tmp942: + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r3385, %r1765, %r3383; +$L__tmp943: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3386, %r3199, %r3384; + add.f32 %r3387, %r2822, %r2824; +$L__tmp944: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3388, %r3386, 4, 31, -1; +$L__tmp945: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3389, %r2064, %r915; +$L__tmp946: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3390, %r3386, %r3388; +$L__tmp947: + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r3391, %r1765, %r3389; +$L__tmp948: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3392, %r3390, 2, 31, -1; +$L__tmp949: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3393, %r2836, %r2838; + add.f32 %r3394, %r3390, %r3392; + add.f32 %r3395, %r2850, %r2852; +$L__tmp950: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3396, %r3394, 1, 31, -1; +$L__tmp951: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3397, %r2864, %r2866; +$L__tmp952: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3398, %r3213, 8, 31, -1; +$L__tmp953: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd677, {%r2331, %r2332}; +$L__tmp954: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3399, %r3213, %r3398; + add.f32 %r3400, %r2878, %r2880; +$L__tmp955: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3401, %r3399, 4, 31, -1; +$L__tmp956: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3402, %r2892, %r2894; + add.f32 %r3403, %r3399, %r3401; +$L__tmp957: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd678, {%r2329, %r2330}; +$L__tmp958: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3404, %r3403, 2, 31, -1; +$L__tmp959: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3405, %r2906, %r2908; + add.f32 %r3406, %r3403, %r3404; + add.f32 %r3407, %r2920, %r2922; +$L__tmp960: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3408, %r3406, 1, 31, -1; +$L__tmp961: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3409, %r2934, %r2936; +$L__tmp962: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3410, %r3227, 8, 31, -1; +$L__tmp963: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3411, %r3376, %r913; +$L__tmp964: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3412, %r3227, %r3410; +$L__tmp965: + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r3413, %r1772, %r3411; +$L__tmp966: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3414, %r3412, 4, 31, -1; +$L__tmp967: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3415, %r2948, %r2950; + add.f32 %r3416, %r3412, %r3414; +$L__tmp968: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3417, %r3376, %r911; +$L__tmp969: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3418, %r3416, 2, 31, -1; +$L__tmp970: + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r3419, %r1772, %r3417; +$L__tmp971: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3420, %r3416, %r3418; + add.f32 %r3421, %r2962, %r2964; +$L__tmp972: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3422, %r3420, 1, 31, -1; +$L__tmp973: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3423, %r2976, %r2978; +$L__tmp974: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3424, %r3241, 8, 31, -1; +$L__tmp975: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3425, %r2990, %r2992; + add.f32 %r3426, %r3241, %r3424; +$L__tmp976: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd679, {%r2327, %r2328}; +$L__tmp977: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3427, %r3426, 4, 31, -1; +$L__tmp978: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3428, %r3004, %r3006; + add.f32 %r3429, %r3426, %r3427; + add.f32 %r3430, %r3018, %r3020; +$L__tmp979: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3431, %r3429, 2, 31, -1; +$L__tmp980: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd680, {%r2325, %r2326}; +$L__tmp981: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3432, %r3429, %r3431; + add.f32 %r3433, %r3032, %r3034; +$L__tmp982: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3434, %r3432, 1, 31, -1; +$L__tmp983: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3435, %r3046, %r3048; +$L__tmp984: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3436, %r3255, 8, 31, -1; +$L__tmp985: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3437, %r3376, %r909; +$L__tmp986: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3438, %r3255, %r3436; +$L__tmp987: + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r3439, %r1772, %r3437; +$L__tmp988: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3440, %r3438, 4, 31, -1; +$L__tmp989: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3441, %r3060, %r3062; + add.f32 %r3442, %r3438, %r3440; +$L__tmp990: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3443, %r3376, %r907; +$L__tmp991: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3444, %r3442, 2, 31, -1; +$L__tmp992: + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r3445, %r1772, %r3443; +$L__tmp993: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3446, %r3442, %r3444; + add.f32 %r3447, %r3074, %r3076; +$L__tmp994: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3448, %r3446, 1, 31, -1; +$L__tmp995: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3449, %r3088, %r3090; +$L__tmp996: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3450, %r3269, 8, 31, -1; +$L__tmp997: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3451, %r3102, %r3104; + add.f32 %r3452, %r3269, %r3450; +$L__tmp998: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd681, {%r2321, %r2323}; +$L__tmp999: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3453, %r3452, 4, 31, -1; +$L__tmp1000: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3454, %r3116, %r3118; + add.f32 %r3455, %r3452, %r3453; + add.f32 %r3456, %r3130, %r3132; +$L__tmp1001: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3457, %r3455, 2, 31, -1; +$L__tmp1002: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd682, {%r2317, %r2319}; +$L__tmp1003: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3458, %r3455, %r3457; + add.f32 %r3459, %r3144, %r3146; +$L__tmp1004: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3460, %r3458, 1, 31, -1; +$L__tmp1005: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3461, %r3158, %r3160; +$L__tmp1006: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3462, %r3283, 8, 31, -1; +$L__tmp1007: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3463, %r3172, %r3174; + add.f32 %r3464, %r3283, %r3462; +$L__tmp1008: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3465, %r3374, %r905; +$L__tmp1009: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3466, %r3464, 4, 31, -1; +$L__tmp1010: + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r3467, %r1764, %r3465; +$L__tmp1011: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3468, %r3464, %r3466; + add.f32 %r3469, %r3186, %r3188; +$L__tmp1012: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3470, %r3468, 2, 31, -1; +$L__tmp1013: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3471, %r3200, %r3202; + add.f32 %r3472, %r3468, %r3470; +$L__tmp1014: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3473, %r3374, %r903; +$L__tmp1015: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3474, %r3472, 1, 31, -1; +$L__tmp1016: + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r3475, %r1764, %r3473; +$L__tmp1017: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3476, %r3297, 8, 31, -1; +$L__tmp1018: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3477, %r3214, %r3216; + add.f32 %r3478, %r3297, %r3476; + add.f32 %r3479, %r3228, %r3230; +$L__tmp1019: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3480, %r3478, 4, 31, -1; +$L__tmp1020: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd683, {%r2313, %r2315}; +$L__tmp1021: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3481, %r3478, %r3480; + add.f32 %r3482, %r3242, %r3244; +$L__tmp1022: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3483, %r3481, 2, 31, -1; +$L__tmp1023: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3484, %r3256, %r3258; + add.f32 %r3485, %r3481, %r3483; + add.f32 %r3486, %r3270, %r3272; +$L__tmp1024: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3487, %r3485, 1, 31, -1; +$L__tmp1025: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd684, {%r2308, %r2311}; +$L__tmp1026: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3488, %r3311, 8, 31, -1; +$L__tmp1027: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3489, %r3284, %r3286; + add.f32 %r3490, %r3311, %r3488; +$L__tmp1028: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3491, %r3374, %r901; +$L__tmp1029: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3492, %r3490, 4, 31, -1; +$L__tmp1030: + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r3493, %r1764, %r3491; +$L__tmp1031: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3494, %r3490, %r3492; + add.f32 %r3495, %r3298, %r3300; +$L__tmp1032: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3496, %r3494, 2, 31, -1; +$L__tmp1033: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3497, %r3312, %r3314; + add.f32 %r3498, %r3494, %r3496; +$L__tmp1034: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3499, %r3374, %r899; +$L__tmp1035: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3500, %r3498, 1, 31, -1; +$L__tmp1036: + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r3501, %r1764, %r3499; +$L__tmp1037: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3502, %r3325, 8, 31, -1; +$L__tmp1038: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3503, %r3326, %r3328; + add.f32 %r3504, %r3325, %r3502; + add.f32 %r3505, %r3340, %r3342; +$L__tmp1039: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3506, %r3504, 4, 31, -1; +$L__tmp1040: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd685, {%r2304, %r2306}; +$L__tmp1041: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3507, %r3504, %r3506; + add.f32 %r3508, %r3354, %r3356; +$L__tmp1042: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3509, %r3507, 2, 31, -1; +$L__tmp1043: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3510, %r3368, %r3370; + add.f32 %r3511, %r3507, %r3509; + add.f32 %r3512, %r3380, %r3382; +$L__tmp1044: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3513, %r3511, 1, 31, -1; +$L__tmp1045: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd686, {%r2300, %r2302}; +$L__tmp1046: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3514, %r3339, 8, 31, -1; +$L__tmp1047: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3515, %r3394, %r3396; + add.f32 %r3516, %r3339, %r3514; + add.f32 %r3517, %r3406, %r3408; +$L__tmp1048: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3518, %r3516, 4, 31, -1; +$L__tmp1049: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3519, %r3371, %r897; +$L__tmp1050: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3520, %r3516, %r3518; +$L__tmp1051: + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r3521, %r1771, %r3519; +$L__tmp1052: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3522, %r3520, 2, 31, -1; +$L__tmp1053: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3523, %r3420, %r3422; + add.f32 %r3524, %r3520, %r3522; + add.f32 %r3525, %r3432, %r3434; +$L__tmp1054: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3526, %r3524, 1, 31, -1; +$L__tmp1055: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3527, %r3371, %r895; +$L__tmp1056: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3528, %r3353, 8, 31, -1; +$L__tmp1057: + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r3529, %r1771, %r3527; +$L__tmp1058: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3530, %r3353, %r3528; + add.f32 %r3531, %r3446, %r3448; +$L__tmp1059: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3532, %r3530, 4, 31, -1; +$L__tmp1060: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3533, %r3458, %r3460; + add.f32 %r3534, %r3530, %r3532; +$L__tmp1061: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd687, {%r2296, %r2298}; +$L__tmp1062: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3535, %r3534, 2, 31, -1; +$L__tmp1063: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3536, %r3472, %r3474; + add.f32 %r3537, %r3534, %r3535; + add.f32 %r3538, %r3485, %r3487; +$L__tmp1064: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3539, %r3537, 1, 31, -1; +$L__tmp1065: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3540, %r3498, %r3500; +$L__tmp1066: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3541, %r3367, 8, 31, -1; +$L__tmp1067: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd688, {%r2293, %r2294}; +$L__tmp1068: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3542, %r3367, %r3541; + add.f32 %r3543, %r3511, %r3513; +$L__tmp1069: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3544, %r3542, 4, 31, -1; +$L__tmp1070: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3545, %r3371, %r893; +$L__tmp1071: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3546, %r3542, %r3544; +$L__tmp1072: + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r3547, %r1771, %r3545; +$L__tmp1073: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3548, %r3546, 2, 31, -1; +$L__tmp1074: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3549, %r3524, %r3526; + add.f32 %r3550, %r3546, %r3548; + add.f32 %r3551, %r3537, %r3539; +$L__tmp1075: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3552, %r3550, 1, 31, -1; +$L__tmp1076: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3553, %r3371, %r891; +$L__tmp1077: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3554, %r3550, %r3552; +$L__tmp1078: + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r3555, %r1771, %r3553; + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3556, %r3369, %r883; + mul.f32 %r3557, %r3369, %r884; + mul.f32 %r3558, %r3369, %r885; + mul.f32 %r3559, %r3369, %r886; + mul.f32 %r3560, %r3369, %r887; + mul.f32 %r3561, %r3369, %r888; + mul.f32 %r3562, %r3369, %r889; + mul.f32 %r3563, %r3369, %r890; + mul.f32 %r3564, %r3371, %r892; + mul.f32 %r3565, %r3371, %r894; + mul.f32 %r3566, %r3371, %r896; + mul.f32 %r3567, %r3371, %r898; + mul.f32 %r3568, %r3374, %r900; + mul.f32 %r3569, %r3374, %r902; + mul.f32 %r3570, %r3374, %r904; + mul.f32 %r3571, %r3374, %r906; + mul.f32 %r3572, %r3376, %r908; + mul.f32 %r3573, %r3376, %r910; + mul.f32 %r3574, %r3376, %r912; + mul.f32 %r3575, %r3376, %r914; + mul.f32 %r3576, %r2064, %r916; + mul.f32 %r3577, %r2064, %r918; + mul.f32 %r3578, %r2064, %r920; + mul.f32 %r3579, %r2064, %r922; + mul.f32 %r3580, %r2025, %r924; + mul.f32 %r3581, %r2025, %r926; + mul.f32 %r3582, %r2025, %r928; + mul.f32 %r3583, %r2025, %r930; + mul.f32 %r3584, %r1986, %r932; + mul.f32 %r3585, %r1986, %r934; + mul.f32 %r3586, %r1986, %r936; + mul.f32 %r3587, %r1986, %r938; + mul.f32 %r3588, %r1965, %r940; + mul.f32 %r3589, %r1965, %r942; + mul.f32 %r3590, %r1965, %r943; + mul.f32 %r3591, %r1965, %r944; + mul.f32 %r3592, %r1965, %r945; + mul.f32 %r3593, %r1965, %r946; + mul.f32 %r3594, %r3379, %r947; + mul.f32 %r3595, %r3379, %r948; + mul.f32 %r3596, %r3379, %r949; + mul.f32 %r3597, %r3379, %r950; + mul.f32 %r3598, %r3379, %r951; + mul.f32 %r3599, %r3379, %r952; + mul.f32 %r3600, %r3379, %r953; + mul.f32 %r3601, %r3379, %r954; + mul.f32 %r3602, %r3381, %r955; + mul.f32 %r3603, %r3381, %r956; + mul.f32 %r3604, %r3381, %r957; + mul.f32 %r3605, %r3381, %r958; + mul.f32 %r3606, %r3381, %r959; + mul.f32 %r3607, %r3381, %r960; + mul.f32 %r3608, %r3381, %r961; + mul.f32 %r3609, %r3381, %r962; + mul.f32 %r3610, %r3387, %r963; + mul.f32 %r3611, %r3387, %r964; + mul.f32 %r3612, %r3387, %r965; + mul.f32 %r3613, %r3387, %r966; + mul.f32 %r3614, %r3387, %r967; + mul.f32 %r3615, %r3387, %r968; + mul.f32 %r3616, %r3387, %r969; + mul.f32 %r3617, %r3387, %r970; + mul.f32 %r3618, %r3393, %r971; + mul.f32 %r3619, %r3393, %r972; + mul.f32 %r3620, %r3393, %r973; + mul.f32 %r3621, %r3393, %r974; + mul.f32 %r3622, %r3393, %r975; + mul.f32 %r3623, %r3393, %r976; + mul.f32 %r3624, %r3393, %r977; + mul.f32 %r3625, %r3393, %r978; + mul.f32 %r3626, %r3395, %r979; + mul.f32 %r3627, %r3395, %r980; + mul.f32 %r3628, %r3395, %r981; + mul.f32 %r3629, %r3395, %r982; + mul.f32 %r3630, %r3395, %r983; + mul.f32 %r3631, %r3395, %r984; + mul.f32 %r3632, %r3395, %r985; + mul.f32 %r3633, %r3395, %r986; + mul.f32 %r3634, %r3397, %r987; + mul.f32 %r3635, %r3397, %r988; + mul.f32 %r3636, %r3397, %r989; + mul.f32 %r3637, %r3397, %r990; + mul.f32 %r3638, %r3397, %r991; + mul.f32 %r3639, %r3397, %r992; + mul.f32 %r3640, %r3397, %r993; + mul.f32 %r3641, %r3397, %r994; + mul.f32 %r3642, %r3400, %r995; + mul.f32 %r3643, %r3400, %r996; + mul.f32 %r3644, %r3400, %r997; + mul.f32 %r3645, %r3400, %r998; + mul.f32 %r3646, %r3400, %r999; + mul.f32 %r3647, %r3400, %r1000; + mul.f32 %r3648, %r3400, %r1001; + mul.f32 %r3649, %r3400, %r1002; + mul.f32 %r3650, %r3402, %r1003; + mul.f32 %r3651, %r3402, %r1004; + mul.f32 %r3652, %r3402, %r1005; + mul.f32 %r3653, %r3402, %r1006; + mul.f32 %r3654, %r3402, %r1007; + mul.f32 %r3655, %r3402, %r1008; + mul.f32 %r3656, %r3402, %r1009; + mul.f32 %r3657, %r3402, %r1010; + mul.f32 %r3658, %r3405, %r1011; + mul.f32 %r3659, %r3405, %r1012; + mul.f32 %r3660, %r3405, %r1013; + mul.f32 %r3661, %r3405, %r1014; + mul.f32 %r3662, %r3405, %r1015; + mul.f32 %r3663, %r3405, %r1016; + mul.f32 %r3664, %r3405, %r1017; + mul.f32 %r3665, %r3405, %r1018; + mul.f32 %r3666, %r3407, %r1019; + mul.f32 %r3667, %r3407, %r1020; + mul.f32 %r3668, %r3407, %r1021; + mul.f32 %r3669, %r3407, %r1022; + mul.f32 %r3670, %r3407, %r1023; + mul.f32 %r3671, %r3407, %r1024; + mul.f32 %r3672, %r3407, %r1025; + mul.f32 %r3673, %r3407, %r1026; + mul.f32 %r3674, %r3409, %r1027; + mul.f32 %r3675, %r3409, %r1028; + mul.f32 %r3676, %r3409, %r1029; + mul.f32 %r3677, %r3409, %r1030; + mul.f32 %r3678, %r3409, %r1031; + mul.f32 %r3679, %r3409, %r1032; + mul.f32 %r3680, %r3409, %r1033; + mul.f32 %r3681, %r3409, %r1034; + mul.f32 %r3682, %r3415, %r1035; + mul.f32 %r3683, %r3415, %r1036; + mul.f32 %r3684, %r3415, %r1037; + mul.f32 %r3685, %r3415, %r1038; + mul.f32 %r3686, %r3415, %r1039; + mul.f32 %r3687, %r3415, %r1040; + mul.f32 %r3688, %r3415, %r1041; + mul.f32 %r3689, %r3415, %r1042; + mul.f32 %r3690, %r3421, %r1043; + mul.f32 %r3691, %r3421, %r1044; + mul.f32 %r3692, %r3421, %r1045; + mul.f32 %r3693, %r3421, %r1046; + mul.f32 %r3694, %r3421, %r1047; + mul.f32 %r3695, %r3421, %r1048; + mul.f32 %r3696, %r3421, %r1049; + mul.f32 %r3697, %r3421, %r1050; + mul.f32 %r3698, %r3423, %r1051; + mul.f32 %r3699, %r3423, %r1052; + mul.f32 %r3700, %r3423, %r1053; + mul.f32 %r3701, %r3423, %r1054; + mul.f32 %r3702, %r3423, %r1055; + mul.f32 %r3703, %r3423, %r1056; + mul.f32 %r3704, %r3423, %r1057; + mul.f32 %r3705, %r3423, %r1058; + mul.f32 %r3706, %r3425, %r1059; + mul.f32 %r3707, %r3425, %r1060; + mul.f32 %r3708, %r3425, %r1061; + mul.f32 %r3709, %r3425, %r1062; + mul.f32 %r3710, %r3425, %r1063; + mul.f32 %r3711, %r3425, %r1064; + mul.f32 %r3712, %r3425, %r1065; + mul.f32 %r3713, %r3425, %r1066; + mul.f32 %r3714, %r3428, %r1067; + mul.f32 %r3715, %r3428, %r1068; + mul.f32 %r3716, %r3428, %r1069; + mul.f32 %r3717, %r3428, %r1070; + mul.f32 %r3718, %r3428, %r1071; + mul.f32 %r3719, %r3428, %r1072; + mul.f32 %r3720, %r3428, %r1073; + mul.f32 %r3721, %r3428, %r1074; + mul.f32 %r3722, %r3430, %r1075; + mul.f32 %r3723, %r3430, %r1076; + mul.f32 %r3724, %r3430, %r1077; + mul.f32 %r3725, %r3430, %r1078; + mul.f32 %r3726, %r3430, %r1079; + mul.f32 %r3727, %r3430, %r1080; + mul.f32 %r3728, %r3430, %r1081; + mul.f32 %r3729, %r3430, %r1082; + mul.f32 %r3730, %r3433, %r1083; + mul.f32 %r3731, %r3433, %r1084; + mul.f32 %r3732, %r3433, %r1085; + mul.f32 %r3733, %r3433, %r1086; + mul.f32 %r3734, %r3433, %r1087; + mul.f32 %r3735, %r3433, %r1088; + mul.f32 %r3736, %r3433, %r1089; + mul.f32 %r3737, %r3433, %r1090; + mul.f32 %r3738, %r3435, %r1091; + mul.f32 %r3739, %r3435, %r1092; + mul.f32 %r3740, %r3435, %r1093; + mul.f32 %r3741, %r3435, %r1094; + mul.f32 %r3742, %r3435, %r1095; + mul.f32 %r3743, %r3435, %r1096; + mul.f32 %r3744, %r3435, %r1097; + mul.f32 %r3745, %r3435, %r1098; + mul.f32 %r3746, %r3441, %r1099; + mul.f32 %r3747, %r3441, %r1100; + mul.f32 %r3748, %r3441, %r1101; + mul.f32 %r3749, %r3441, %r1102; + mul.f32 %r3750, %r3441, %r1103; + mul.f32 %r3751, %r3441, %r1104; + mul.f32 %r3752, %r3441, %r1105; + mul.f32 %r3753, %r3441, %r1106; + mul.f32 %r3754, %r3447, %r1107; + mul.f32 %r3755, %r3447, %r1108; + mul.f32 %r3756, %r3447, %r1109; + mul.f32 %r3757, %r3447, %r1110; + mul.f32 %r3758, %r3447, %r1111; + mul.f32 %r3759, %r3447, %r1112; + mul.f32 %r3760, %r3447, %r1113; + mul.f32 %r3761, %r3447, %r1114; + mul.f32 %r3762, %r3449, %r1115; + mul.f32 %r3763, %r3449, %r1116; + mul.f32 %r3764, %r3449, %r1117; + mul.f32 %r3765, %r3449, %r1118; + mul.f32 %r3766, %r3449, %r1119; + mul.f32 %r3767, %r3449, %r1120; + mul.f32 %r3768, %r3449, %r1121; + mul.f32 %r3769, %r3449, %r1122; + mul.f32 %r3770, %r3451, %r1123; + mul.f32 %r3771, %r3451, %r1124; + mul.f32 %r3772, %r3451, %r1125; + mul.f32 %r3773, %r3451, %r1126; + mul.f32 %r3774, %r3451, %r1127; + mul.f32 %r3775, %r3451, %r1128; + mul.f32 %r3776, %r3451, %r1129; + mul.f32 %r3777, %r3451, %r1130; + mul.f32 %r3778, %r3454, %r1131; + mul.f32 %r3779, %r3454, %r1132; + mul.f32 %r3780, %r3454, %r1133; + mul.f32 %r3781, %r3454, %r1134; + mul.f32 %r3782, %r3454, %r1135; + mul.f32 %r3783, %r3454, %r1136; + mul.f32 %r3784, %r3454, %r1137; + mul.f32 %r3785, %r3454, %r1138; + mul.f32 %r3786, %r3456, %r1139; + mul.f32 %r3787, %r3456, %r1140; + mul.f32 %r3788, %r3456, %r1141; + mul.f32 %r3789, %r3456, %r1142; + mul.f32 %r3790, %r3456, %r1143; + mul.f32 %r3791, %r3456, %r1144; + mul.f32 %r3792, %r3456, %r1145; + mul.f32 %r3793, %r3456, %r1146; + mul.f32 %r3794, %r3459, %r1147; + mul.f32 %r3795, %r3459, %r1148; + mul.f32 %r3796, %r3459, %r1149; + mul.f32 %r3797, %r3459, %r1150; + mul.f32 %r3798, %r3459, %r1151; + mul.f32 %r3799, %r3459, %r1152; + mul.f32 %r3800, %r3459, %r1153; + mul.f32 %r3801, %r3459, %r1154; + mul.f32 %r3802, %r3461, %r1155; + mul.f32 %r3803, %r3461, %r1156; + mul.f32 %r3804, %r3461, %r1157; + mul.f32 %r3805, %r3461, %r1158; + mul.f32 %r3806, %r3461, %r1159; + mul.f32 %r3807, %r3461, %r1160; + mul.f32 %r3808, %r3461, %r1161; + mul.f32 %r3809, %r3461, %r1162; + mul.f32 %r3810, %r3463, %r1163; + mul.f32 %r3811, %r3463, %r1164; + mul.f32 %r3812, %r3463, %r1165; + mul.f32 %r3813, %r3463, %r1166; + mul.f32 %r3814, %r3463, %r1167; + mul.f32 %r3815, %r3463, %r1168; + mul.f32 %r3816, %r3463, %r1169; + mul.f32 %r3817, %r3463, %r1170; + mul.f32 %r3818, %r3469, %r1171; + mul.f32 %r3819, %r3469, %r1172; + mul.f32 %r3820, %r3469, %r1173; + mul.f32 %r3821, %r3469, %r1174; + mul.f32 %r3822, %r3469, %r1175; + mul.f32 %r3823, %r3469, %r1176; + mul.f32 %r3824, %r3469, %r1177; + mul.f32 %r3825, %r3469, %r1178; + mul.f32 %r3826, %r3471, %r1179; + mul.f32 %r3827, %r3471, %r1180; + mul.f32 %r3828, %r3471, %r1181; + mul.f32 %r3829, %r3471, %r1182; + mul.f32 %r3830, %r3471, %r1183; + mul.f32 %r3831, %r3471, %r1184; + mul.f32 %r3832, %r3471, %r1185; + mul.f32 %r3833, %r3471, %r1186; + mul.f32 %r3834, %r3477, %r1187; + mul.f32 %r3835, %r3477, %r1188; + mul.f32 %r3836, %r3477, %r1189; + mul.f32 %r3837, %r3477, %r1190; + mul.f32 %r3838, %r3477, %r1191; + mul.f32 %r3839, %r3477, %r1192; + mul.f32 %r3840, %r3477, %r1193; + mul.f32 %r3841, %r3477, %r1194; + mul.f32 %r3842, %r3479, %r1195; + mul.f32 %r3843, %r3479, %r1196; + mul.f32 %r3844, %r3479, %r1197; + mul.f32 %r3845, %r3479, %r1198; + mul.f32 %r3846, %r3479, %r1199; + mul.f32 %r3847, %r3479, %r1200; + mul.f32 %r3848, %r3479, %r1201; + mul.f32 %r3849, %r3479, %r1202; + mul.f32 %r3850, %r3482, %r1203; + mul.f32 %r3851, %r3482, %r1204; + mul.f32 %r3852, %r3482, %r1205; + mul.f32 %r3853, %r3482, %r1206; + mul.f32 %r3854, %r3482, %r1207; + mul.f32 %r3855, %r3482, %r1208; + mul.f32 %r3856, %r3482, %r1209; + mul.f32 %r3857, %r3482, %r1210; + mul.f32 %r3858, %r3484, %r1211; + mul.f32 %r3859, %r3484, %r1212; + mul.f32 %r3860, %r3484, %r1213; + mul.f32 %r3861, %r3484, %r1214; + mul.f32 %r3862, %r3484, %r1215; + mul.f32 %r3863, %r3484, %r1216; + mul.f32 %r3864, %r3484, %r1217; + mul.f32 %r3865, %r3484, %r1218; + mul.f32 %r3866, %r3486, %r1219; + mul.f32 %r3867, %r3486, %r1220; + mul.f32 %r3868, %r3486, %r1221; + mul.f32 %r3869, %r3486, %r1222; + mul.f32 %r3870, %r3486, %r1223; + mul.f32 %r3871, %r3486, %r1224; + mul.f32 %r3872, %r3486, %r1225; + mul.f32 %r3873, %r3486, %r1226; + mul.f32 %r3874, %r3489, %r1227; + mul.f32 %r3875, %r3489, %r1228; + mul.f32 %r3876, %r3489, %r1229; + mul.f32 %r3877, %r3489, %r1230; + mul.f32 %r3878, %r3489, %r1231; + mul.f32 %r3879, %r3489, %r1232; + mul.f32 %r3880, %r3489, %r1233; + mul.f32 %r3881, %r3489, %r1234; + mul.f32 %r3882, %r3495, %r1235; + mul.f32 %r3883, %r3495, %r1236; + mul.f32 %r3884, %r3495, %r1237; + mul.f32 %r3885, %r3495, %r1238; + mul.f32 %r3886, %r3495, %r1239; + mul.f32 %r3887, %r3495, %r1240; + mul.f32 %r3888, %r3495, %r1241; + mul.f32 %r3889, %r3495, %r1242; + mul.f32 %r3890, %r3497, %r1243; + mul.f32 %r3891, %r3497, %r1244; + mul.f32 %r3892, %r3497, %r1245; + mul.f32 %r3893, %r3497, %r1246; + mul.f32 %r3894, %r3497, %r1247; + mul.f32 %r3895, %r3497, %r1248; + mul.f32 %r3896, %r3497, %r1249; + mul.f32 %r3897, %r3497, %r1250; + mul.f32 %r3898, %r3503, %r1251; + mul.f32 %r3899, %r3503, %r1252; + mul.f32 %r3900, %r3503, %r1253; + mul.f32 %r3901, %r3503, %r1254; + mul.f32 %r3902, %r3503, %r1255; + mul.f32 %r3903, %r3503, %r1256; + mul.f32 %r3904, %r3503, %r1257; + mul.f32 %r3905, %r3503, %r1258; + mul.f32 %r3906, %r3505, %r1259; + mul.f32 %r3907, %r3505, %r1260; + mul.f32 %r3908, %r3505, %r1261; + mul.f32 %r3909, %r3505, %r1262; + mul.f32 %r3910, %r3505, %r1263; + mul.f32 %r3911, %r3505, %r1264; + mul.f32 %r3912, %r3505, %r1265; + mul.f32 %r3913, %r3505, %r1266; + mul.f32 %r3914, %r3508, %r1267; + mul.f32 %r3915, %r3508, %r1268; + mul.f32 %r3916, %r3508, %r1269; + mul.f32 %r3917, %r3508, %r1270; + mul.f32 %r3918, %r3508, %r1271; + mul.f32 %r3919, %r3508, %r1272; + mul.f32 %r3920, %r3508, %r1273; + mul.f32 %r3921, %r3508, %r1274; + mul.f32 %r3922, %r3510, %r1275; + mul.f32 %r3923, %r3510, %r1276; + mul.f32 %r3924, %r3510, %r1277; + mul.f32 %r3925, %r3510, %r1278; + mul.f32 %r3926, %r3510, %r1279; + mul.f32 %r3927, %r3510, %r1280; + mul.f32 %r3928, %r3510, %r1281; + mul.f32 %r3929, %r3510, %r1282; + mul.f32 %r3930, %r3512, %r1283; + mul.f32 %r3931, %r3512, %r1284; + mul.f32 %r3932, %r3512, %r1285; + mul.f32 %r3933, %r3512, %r1286; + mul.f32 %r3934, %r3512, %r1287; + mul.f32 %r3935, %r3512, %r1288; + mul.f32 %r3936, %r3512, %r1289; + mul.f32 %r3937, %r3512, %r1290; + mul.f32 %r3938, %r3515, %r1291; + mul.f32 %r3939, %r3515, %r1292; + mul.f32 %r3940, %r3515, %r1293; + mul.f32 %r3941, %r3515, %r1294; + mul.f32 %r3942, %r3515, %r1295; + mul.f32 %r3943, %r3515, %r1296; + mul.f32 %r3944, %r3515, %r1297; + mul.f32 %r3945, %r3515, %r1298; + mul.f32 %r3946, %r3517, %r1299; + mul.f32 %r3947, %r3517, %r1300; + mul.f32 %r3948, %r3517, %r1301; + mul.f32 %r3949, %r3517, %r1302; + mul.f32 %r3950, %r3517, %r1303; + mul.f32 %r3951, %r3517, %r1304; + mul.f32 %r3952, %r3517, %r1305; + mul.f32 %r3953, %r3517, %r1306; + mul.f32 %r3954, %r3523, %r1307; + mul.f32 %r3955, %r3523, %r1308; + mul.f32 %r3956, %r3523, %r1309; + mul.f32 %r3957, %r3523, %r1310; + mul.f32 %r3958, %r3523, %r1311; + mul.f32 %r3959, %r3523, %r1312; + mul.f32 %r3960, %r3523, %r1313; + mul.f32 %r3961, %r3523, %r1314; + mul.f32 %r3962, %r3525, %r1315; + mul.f32 %r3963, %r3525, %r1316; + mul.f32 %r3964, %r3525, %r1317; + mul.f32 %r3965, %r3525, %r1318; + mul.f32 %r3966, %r3525, %r1319; + mul.f32 %r3967, %r3525, %r1320; + mul.f32 %r3968, %r3525, %r1321; + mul.f32 %r3969, %r3525, %r1322; + mul.f32 %r3970, %r3531, %r1323; + mul.f32 %r3971, %r3531, %r1324; + mul.f32 %r3972, %r3531, %r1325; + mul.f32 %r3973, %r3531, %r1326; + mul.f32 %r3974, %r3531, %r1327; + mul.f32 %r3975, %r3531, %r1328; + mul.f32 %r3976, %r3531, %r1329; + mul.f32 %r3977, %r3531, %r1330; + mul.f32 %r3978, %r3533, %r1331; + mul.f32 %r3979, %r3533, %r1332; + mul.f32 %r3980, %r3533, %r1333; + mul.f32 %r3981, %r3533, %r1334; + mul.f32 %r3982, %r3533, %r1335; + mul.f32 %r3983, %r3533, %r1336; + mul.f32 %r3984, %r3533, %r1337; + mul.f32 %r3985, %r3533, %r1338; + mul.f32 %r3986, %r3536, %r1339; + mul.f32 %r3987, %r3536, %r1340; + mul.f32 %r3988, %r3536, %r1341; + mul.f32 %r3989, %r3536, %r1342; + mul.f32 %r3990, %r3536, %r1343; + mul.f32 %r3991, %r3536, %r1344; + mul.f32 %r3992, %r3536, %r1345; + mul.f32 %r3993, %r3536, %r1346; + mul.f32 %r3994, %r3538, %r1347; + mul.f32 %r3995, %r3538, %r1348; + mul.f32 %r3996, %r3538, %r1349; + mul.f32 %r3997, %r3538, %r1350; + mul.f32 %r3998, %r3538, %r1351; + mul.f32 %r3999, %r3538, %r1352; + mul.f32 %r4000, %r3538, %r1353; + mul.f32 %r4001, %r3538, %r1354; + mul.f32 %r4002, %r3540, %r1355; + mul.f32 %r4003, %r3540, %r1356; + mul.f32 %r4004, %r3540, %r1357; + mul.f32 %r4005, %r3540, %r1358; + mul.f32 %r4006, %r3540, %r1359; + mul.f32 %r4007, %r3540, %r1360; + mul.f32 %r4008, %r3540, %r1361; + mul.f32 %r4009, %r3540, %r1362; + mul.f32 %r4010, %r3543, %r1363; + mul.f32 %r4011, %r3543, %r1364; + mul.f32 %r4012, %r3543, %r1365; + mul.f32 %r4013, %r3543, %r1366; + mul.f32 %r4014, %r3543, %r1367; + mul.f32 %r4015, %r3543, %r1368; + mul.f32 %r4016, %r3543, %r1369; + mul.f32 %r4017, %r3543, %r1370; + mul.f32 %r4018, %r3549, %r1371; + mul.f32 %r4019, %r3549, %r1372; + mul.f32 %r4020, %r3549, %r1373; + mul.f32 %r4021, %r3549, %r1374; + mul.f32 %r4022, %r3549, %r1375; + mul.f32 %r4023, %r3549, %r1376; + mul.f32 %r4024, %r3549, %r1377; + mul.f32 %r4025, %r3549, %r1378; + mul.f32 %r4026, %r3551, %r1379; + mul.f32 %r4027, %r3551, %r1380; + mul.f32 %r4028, %r3551, %r1381; + mul.f32 %r4029, %r3551, %r1382; + mul.f32 %r4030, %r3551, %r1383; + mul.f32 %r4031, %r3551, %r1384; + mul.f32 %r4032, %r3551, %r1385; + mul.f32 %r4033, %r3551, %r1386; + mul.f32 %r4034, %r3554, %r1387; + mul.f32 %r4035, %r3554, %r1388; + mul.f32 %r4036, %r3554, %r1389; + mul.f32 %r4037, %r3554, %r1390; + mul.f32 %r4038, %r3554, %r1391; + mul.f32 %r4039, %r3554, %r1392; + mul.f32 %r4040, %r3554, %r1393; + mul.f32 %r4041, %r3554, %r1394; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r4042, %r1763, %r3556; + mul.f32 %r4043, %r1763, %r3557; + mul.f32 %r4044, %r1763, %r3558; + mul.f32 %r4045, %r1763, %r3559; + mul.f32 %r4046, %r1763, %r3560; + mul.f32 %r4047, %r1763, %r3561; + mul.f32 %r4048, %r1763, %r3562; + mul.f32 %r4049, %r1763, %r3563; + mul.f32 %r4050, %r1771, %r3564; + mul.f32 %r4051, %r1771, %r3565; + mul.f32 %r4052, %r1771, %r3566; + mul.f32 %r4053, %r1771, %r3567; + mul.f32 %r4054, %r1764, %r3568; + mul.f32 %r4055, %r1764, %r3569; + mul.f32 %r4056, %r1764, %r3570; + mul.f32 %r4057, %r1764, %r3571; + mul.f32 %r4058, %r1772, %r3572; + mul.f32 %r4059, %r1772, %r3573; + mul.f32 %r4060, %r1772, %r3574; + mul.f32 %r4061, %r1772, %r3575; + mul.f32 %r4062, %r1765, %r3576; + mul.f32 %r4063, %r1765, %r3577; + mul.f32 %r4064, %r1765, %r3578; + mul.f32 %r4065, %r1765, %r3579; + mul.f32 %r4066, %r1773, %r3580; + mul.f32 %r4067, %r1773, %r3581; + mul.f32 %r4068, %r1773, %r3582; + mul.f32 %r4069, %r1773, %r3583; + mul.f32 %r4070, %r1766, %r3584; + mul.f32 %r4071, %r1766, %r3585; + mul.f32 %r4072, %r1766, %r3586; + mul.f32 %r4073, %r1766, %r3587; + mul.f32 %r4074, %r1774, %r3588; + mul.f32 %r4075, %r1774, %r3589; + mul.f32 %r4076, %r1774, %r3590; + mul.f32 %r4077, %r1774, %r3591; + mul.f32 %r4078, %r1774, %r3592; + mul.f32 %r4079, %r1774, %r3593; + mul.f32 %r4080, %r1735, %r3594; + mul.f32 %r4081, %r1735, %r3595; + mul.f32 %r4082, %r1735, %r3596; + mul.f32 %r4083, %r1735, %r3597; + mul.f32 %r4084, %r1735, %r3598; + mul.f32 %r4085, %r1735, %r3599; + mul.f32 %r4086, %r1735, %r3600; + mul.f32 %r4087, %r1735, %r3601; + mul.f32 %r4088, %r1819, %r3602; + mul.f32 %r4089, %r1819, %r3603; + mul.f32 %r4090, %r1819, %r3604; + mul.f32 %r4091, %r1819, %r3605; + mul.f32 %r4092, %r1819, %r3606; + mul.f32 %r4093, %r1819, %r3607; + mul.f32 %r4094, %r1819, %r3608; + mul.f32 %r4095, %r1819, %r3609; + mul.f32 %r4096, %r1736, %r3610; + mul.f32 %r4097, %r1736, %r3611; + mul.f32 %r4098, %r1736, %r3612; + mul.f32 %r4099, %r1736, %r3613; + mul.f32 %r4100, %r1736, %r3614; + mul.f32 %r4101, %r1736, %r3615; + mul.f32 %r4102, %r1736, %r3616; + mul.f32 %r4103, %r1736, %r3617; + mul.f32 %r4104, %r1820, %r3618; + mul.f32 %r4105, %r1820, %r3619; + mul.f32 %r4106, %r1820, %r3620; + mul.f32 %r4107, %r1820, %r3621; + mul.f32 %r4108, %r1820, %r3622; + mul.f32 %r4109, %r1820, %r3623; + mul.f32 %r4110, %r1820, %r3624; + mul.f32 %r4111, %r1820, %r3625; + mul.f32 %r4112, %r1737, %r3626; + mul.f32 %r4113, %r1737, %r3627; + mul.f32 %r4114, %r1737, %r3628; + mul.f32 %r4115, %r1737, %r3629; + mul.f32 %r4116, %r1737, %r3630; + mul.f32 %r4117, %r1737, %r3631; + mul.f32 %r4118, %r1737, %r3632; + mul.f32 %r4119, %r1737, %r3633; + mul.f32 %r4120, %r1821, %r3634; + mul.f32 %r4121, %r1821, %r3635; + mul.f32 %r4122, %r1821, %r3636; + mul.f32 %r4123, %r1821, %r3637; + mul.f32 %r4124, %r1821, %r3638; + mul.f32 %r4125, %r1821, %r3639; + mul.f32 %r4126, %r1821, %r3640; + mul.f32 %r4127, %r1821, %r3641; + mul.f32 %r4128, %r1738, %r3642; + mul.f32 %r4129, %r1738, %r3643; + mul.f32 %r4130, %r1738, %r3644; + mul.f32 %r4131, %r1738, %r3645; + mul.f32 %r4132, %r1738, %r3646; + mul.f32 %r4133, %r1738, %r3647; + mul.f32 %r4134, %r1738, %r3648; + mul.f32 %r4135, %r1738, %r3649; + mul.f32 %r4136, %r1822, %r3650; + mul.f32 %r4137, %r1822, %r3651; + mul.f32 %r4138, %r1822, %r3652; + mul.f32 %r4139, %r1822, %r3653; + mul.f32 %r4140, %r1822, %r3654; + mul.f32 %r4141, %r1822, %r3655; + mul.f32 %r4142, %r1822, %r3656; + mul.f32 %r4143, %r1822, %r3657; + mul.f32 %r4144, %r1739, %r3658; + mul.f32 %r4145, %r1739, %r3659; + mul.f32 %r4146, %r1739, %r3660; + mul.f32 %r4147, %r1739, %r3661; + mul.f32 %r4148, %r1739, %r3662; + mul.f32 %r4149, %r1739, %r3663; + mul.f32 %r4150, %r1739, %r3664; + mul.f32 %r4151, %r1739, %r3665; + mul.f32 %r4152, %r1824, %r3666; + mul.f32 %r4153, %r1824, %r3667; + mul.f32 %r4154, %r1824, %r3668; + mul.f32 %r4155, %r1824, %r3669; + mul.f32 %r4156, %r1824, %r3670; + mul.f32 %r4157, %r1824, %r3671; + mul.f32 %r4158, %r1824, %r3672; + mul.f32 %r4159, %r1824, %r3673; + mul.f32 %r4160, %r1740, %r3674; + mul.f32 %r4161, %r1740, %r3675; + mul.f32 %r4162, %r1740, %r3676; + mul.f32 %r4163, %r1740, %r3677; + mul.f32 %r4164, %r1740, %r3678; + mul.f32 %r4165, %r1740, %r3679; + mul.f32 %r4166, %r1740, %r3680; + mul.f32 %r4167, %r1740, %r3681; + mul.f32 %r4168, %r1825, %r3682; + mul.f32 %r4169, %r1825, %r3683; + mul.f32 %r4170, %r1825, %r3684; + mul.f32 %r4171, %r1825, %r3685; + mul.f32 %r4172, %r1825, %r3686; + mul.f32 %r4173, %r1825, %r3687; + mul.f32 %r4174, %r1825, %r3688; + mul.f32 %r4175, %r1825, %r3689; + mul.f32 %r4176, %r1741, %r3690; + mul.f32 %r4177, %r1741, %r3691; + mul.f32 %r4178, %r1741, %r3692; + mul.f32 %r4179, %r1741, %r3693; + mul.f32 %r4180, %r1741, %r3694; + mul.f32 %r4181, %r1741, %r3695; + mul.f32 %r4182, %r1741, %r3696; + mul.f32 %r4183, %r1741, %r3697; + mul.f32 %r4184, %r1826, %r3698; + mul.f32 %r4185, %r1826, %r3699; + mul.f32 %r4186, %r1826, %r3700; + mul.f32 %r4187, %r1826, %r3701; + mul.f32 %r4188, %r1826, %r3702; + mul.f32 %r4189, %r1826, %r3703; + mul.f32 %r4190, %r1826, %r3704; + mul.f32 %r4191, %r1826, %r3705; + mul.f32 %r4192, %r1742, %r3706; + mul.f32 %r4193, %r1742, %r3707; + mul.f32 %r4194, %r1742, %r3708; + mul.f32 %r4195, %r1742, %r3709; + mul.f32 %r4196, %r1742, %r3710; + mul.f32 %r4197, %r1742, %r3711; + mul.f32 %r4198, %r1742, %r3712; + mul.f32 %r4199, %r1742, %r3713; + mul.f32 %r4200, %r1827, %r3714; + mul.f32 %r4201, %r1827, %r3715; + mul.f32 %r4202, %r1827, %r3716; + mul.f32 %r4203, %r1827, %r3717; + mul.f32 %r4204, %r1827, %r3718; + mul.f32 %r4205, %r1827, %r3719; + mul.f32 %r4206, %r1827, %r3720; + mul.f32 %r4207, %r1827, %r3721; + mul.f32 %r4208, %r1743, %r3722; + mul.f32 %r4209, %r1743, %r3723; + mul.f32 %r4210, %r1743, %r3724; + mul.f32 %r4211, %r1743, %r3725; + mul.f32 %r4212, %r1743, %r3726; + mul.f32 %r4213, %r1743, %r3727; + mul.f32 %r4214, %r1743, %r3728; + mul.f32 %r4215, %r1743, %r3729; + mul.f32 %r4216, %r1829, %r3730; + mul.f32 %r4217, %r1829, %r3731; + mul.f32 %r4218, %r1829, %r3732; + mul.f32 %r4219, %r1829, %r3733; + mul.f32 %r4220, %r1829, %r3734; + mul.f32 %r4221, %r1829, %r3735; + mul.f32 %r4222, %r1829, %r3736; + mul.f32 %r4223, %r1829, %r3737; + mul.f32 %r4224, %r1744, %r3738; + mul.f32 %r4225, %r1744, %r3739; + mul.f32 %r4226, %r1744, %r3740; + mul.f32 %r4227, %r1744, %r3741; + mul.f32 %r4228, %r1744, %r3742; + mul.f32 %r4229, %r1744, %r3743; + mul.f32 %r4230, %r1744, %r3744; + mul.f32 %r4231, %r1744, %r3745; + mul.f32 %r4232, %r1830, %r3746; + mul.f32 %r4233, %r1830, %r3747; + mul.f32 %r4234, %r1830, %r3748; + mul.f32 %r4235, %r1830, %r3749; + mul.f32 %r4236, %r1830, %r3750; + mul.f32 %r4237, %r1830, %r3751; + mul.f32 %r4238, %r1830, %r3752; + mul.f32 %r4239, %r1830, %r3753; + mul.f32 %r4240, %r1745, %r3754; + mul.f32 %r4241, %r1745, %r3755; + mul.f32 %r4242, %r1745, %r3756; + mul.f32 %r4243, %r1745, %r3757; + mul.f32 %r4244, %r1745, %r3758; + mul.f32 %r4245, %r1745, %r3759; + mul.f32 %r4246, %r1745, %r3760; + mul.f32 %r4247, %r1745, %r3761; + mul.f32 %r4248, %r1831, %r3762; + mul.f32 %r4249, %r1831, %r3763; + mul.f32 %r4250, %r1831, %r3764; + mul.f32 %r4251, %r1831, %r3765; + mul.f32 %r4252, %r1831, %r3766; + mul.f32 %r4253, %r1831, %r3767; + mul.f32 %r4254, %r1831, %r3768; + mul.f32 %r4255, %r1831, %r3769; + mul.f32 %r4256, %r1746, %r3770; + mul.f32 %r4257, %r1746, %r3771; + mul.f32 %r4258, %r1746, %r3772; + mul.f32 %r4259, %r1746, %r3773; + mul.f32 %r4260, %r1746, %r3774; + mul.f32 %r4261, %r1746, %r3775; + mul.f32 %r4262, %r1746, %r3776; + mul.f32 %r4263, %r1746, %r3777; + mul.f32 %r4264, %r1832, %r3778; + mul.f32 %r4265, %r1832, %r3779; + mul.f32 %r4266, %r1832, %r3780; + mul.f32 %r4267, %r1832, %r3781; + mul.f32 %r4268, %r1832, %r3782; + mul.f32 %r4269, %r1832, %r3783; + mul.f32 %r4270, %r1832, %r3784; + mul.f32 %r4271, %r1832, %r3785; + mul.f32 %r4272, %r1747, %r3786; + mul.f32 %r4273, %r1747, %r3787; + mul.f32 %r4274, %r1747, %r3788; + mul.f32 %r4275, %r1747, %r3789; + mul.f32 %r4276, %r1747, %r3790; + mul.f32 %r4277, %r1747, %r3791; + mul.f32 %r4278, %r1747, %r3792; + mul.f32 %r4279, %r1747, %r3793; + mul.f32 %r4280, %r1834, %r3794; + mul.f32 %r4281, %r1834, %r3795; + mul.f32 %r4282, %r1834, %r3796; + mul.f32 %r4283, %r1834, %r3797; + mul.f32 %r4284, %r1834, %r3798; + mul.f32 %r4285, %r1834, %r3799; + mul.f32 %r4286, %r1834, %r3800; + mul.f32 %r4287, %r1834, %r3801; + mul.f32 %r4288, %r1748, %r3802; + mul.f32 %r4289, %r1748, %r3803; + mul.f32 %r4290, %r1748, %r3804; + mul.f32 %r4291, %r1748, %r3805; + mul.f32 %r4292, %r1748, %r3806; + mul.f32 %r4293, %r1748, %r3807; + mul.f32 %r4294, %r1748, %r3808; + mul.f32 %r4295, %r1748, %r3809; + mul.f32 %r4296, %r1835, %r3810; + mul.f32 %r4297, %r1835, %r3811; + mul.f32 %r4298, %r1835, %r3812; + mul.f32 %r4299, %r1835, %r3813; + mul.f32 %r4300, %r1835, %r3814; + mul.f32 %r4301, %r1835, %r3815; + mul.f32 %r4302, %r1835, %r3816; + mul.f32 %r4303, %r1835, %r3817; + mul.f32 %r4304, %r1749, %r3818; + mul.f32 %r4305, %r1749, %r3819; + mul.f32 %r4306, %r1749, %r3820; + mul.f32 %r4307, %r1749, %r3821; + mul.f32 %r4308, %r1749, %r3822; + mul.f32 %r4309, %r1749, %r3823; + mul.f32 %r4310, %r1749, %r3824; + mul.f32 %r4311, %r1749, %r3825; + mul.f32 %r4312, %r1836, %r3826; + mul.f32 %r4313, %r1836, %r3827; + mul.f32 %r4314, %r1836, %r3828; + mul.f32 %r4315, %r1836, %r3829; + mul.f32 %r4316, %r1836, %r3830; + mul.f32 %r4317, %r1836, %r3831; + mul.f32 %r4318, %r1836, %r3832; + mul.f32 %r4319, %r1836, %r3833; + mul.f32 %r4320, %r1750, %r3834; + mul.f32 %r4321, %r1750, %r3835; + mul.f32 %r4322, %r1750, %r3836; + mul.f32 %r4323, %r1750, %r3837; + mul.f32 %r4324, %r1750, %r3838; + mul.f32 %r4325, %r1750, %r3839; + mul.f32 %r4326, %r1750, %r3840; + mul.f32 %r4327, %r1750, %r3841; + mul.f32 %r4328, %r1837, %r3842; + mul.f32 %r4329, %r1837, %r3843; + mul.f32 %r4330, %r1837, %r3844; + mul.f32 %r4331, %r1837, %r3845; + mul.f32 %r4332, %r1837, %r3846; + mul.f32 %r4333, %r1837, %r3847; + mul.f32 %r4334, %r1837, %r3848; + mul.f32 %r4335, %r1837, %r3849; + mul.f32 %r4336, %r1751, %r3850; + mul.f32 %r4337, %r1751, %r3851; + mul.f32 %r4338, %r1751, %r3852; + mul.f32 %r4339, %r1751, %r3853; + mul.f32 %r4340, %r1751, %r3854; + mul.f32 %r4341, %r1751, %r3855; + mul.f32 %r4342, %r1751, %r3856; + mul.f32 %r4343, %r1751, %r3857; + mul.f32 %r4344, %r1839, %r3858; + mul.f32 %r4345, %r1839, %r3859; + mul.f32 %r4346, %r1839, %r3860; + mul.f32 %r4347, %r1839, %r3861; + mul.f32 %r4348, %r1839, %r3862; + mul.f32 %r4349, %r1839, %r3863; + mul.f32 %r4350, %r1839, %r3864; + mul.f32 %r4351, %r1839, %r3865; + mul.f32 %r4352, %r1752, %r3866; + mul.f32 %r4353, %r1752, %r3867; + mul.f32 %r4354, %r1752, %r3868; + mul.f32 %r4355, %r1752, %r3869; + mul.f32 %r4356, %r1752, %r3870; + mul.f32 %r4357, %r1752, %r3871; + mul.f32 %r4358, %r1752, %r3872; + mul.f32 %r4359, %r1752, %r3873; + mul.f32 %r4360, %r1840, %r3874; + mul.f32 %r4361, %r1840, %r3875; + mul.f32 %r4362, %r1840, %r3876; + mul.f32 %r4363, %r1840, %r3877; + mul.f32 %r4364, %r1840, %r3878; + mul.f32 %r4365, %r1840, %r3879; + mul.f32 %r4366, %r1840, %r3880; + mul.f32 %r4367, %r1840, %r3881; + mul.f32 %r4368, %r1753, %r3882; + mul.f32 %r4369, %r1753, %r3883; + mul.f32 %r4370, %r1753, %r3884; + mul.f32 %r4371, %r1753, %r3885; + mul.f32 %r4372, %r1753, %r3886; + mul.f32 %r4373, %r1753, %r3887; + mul.f32 %r4374, %r1753, %r3888; + mul.f32 %r4375, %r1753, %r3889; + mul.f32 %r4376, %r1841, %r3890; + mul.f32 %r4377, %r1841, %r3891; + mul.f32 %r4378, %r1841, %r3892; + mul.f32 %r4379, %r1841, %r3893; + mul.f32 %r4380, %r1841, %r3894; + mul.f32 %r4381, %r1841, %r3895; + mul.f32 %r4382, %r1841, %r3896; + mul.f32 %r4383, %r1841, %r3897; + mul.f32 %r4384, %r1754, %r3898; + mul.f32 %r4385, %r1754, %r3899; + mul.f32 %r4386, %r1754, %r3900; + mul.f32 %r4387, %r1754, %r3901; + mul.f32 %r4388, %r1754, %r3902; + mul.f32 %r4389, %r1754, %r3903; + mul.f32 %r4390, %r1754, %r3904; + mul.f32 %r4391, %r1754, %r3905; + mul.f32 %r4392, %r1842, %r3906; + mul.f32 %r4393, %r1842, %r3907; + mul.f32 %r4394, %r1842, %r3908; + mul.f32 %r4395, %r1842, %r3909; + mul.f32 %r4396, %r1842, %r3910; + mul.f32 %r4397, %r1842, %r3911; + mul.f32 %r4398, %r1842, %r3912; + mul.f32 %r4399, %r1842, %r3913; + mul.f32 %r4400, %r1755, %r3914; + mul.f32 %r4401, %r1755, %r3915; + mul.f32 %r4402, %r1755, %r3916; + mul.f32 %r4403, %r1755, %r3917; + mul.f32 %r4404, %r1755, %r3918; + mul.f32 %r4405, %r1755, %r3919; + mul.f32 %r4406, %r1755, %r3920; + mul.f32 %r4407, %r1755, %r3921; + mul.f32 %r4408, %r1844, %r3922; + mul.f32 %r4409, %r1844, %r3923; + mul.f32 %r4410, %r1844, %r3924; + mul.f32 %r4411, %r1844, %r3925; + mul.f32 %r4412, %r1844, %r3926; + mul.f32 %r4413, %r1844, %r3927; + mul.f32 %r4414, %r1844, %r3928; + mul.f32 %r4415, %r1844, %r3929; + mul.f32 %r4416, %r1756, %r3930; + mul.f32 %r4417, %r1756, %r3931; + mul.f32 %r4418, %r1756, %r3932; + mul.f32 %r4419, %r1756, %r3933; + mul.f32 %r4420, %r1756, %r3934; + mul.f32 %r4421, %r1756, %r3935; + mul.f32 %r4422, %r1756, %r3936; + mul.f32 %r4423, %r1756, %r3937; + mul.f32 %r4424, %r1845, %r3938; + mul.f32 %r4425, %r1845, %r3939; + mul.f32 %r4426, %r1845, %r3940; + mul.f32 %r4427, %r1845, %r3941; + mul.f32 %r4428, %r1845, %r3942; + mul.f32 %r4429, %r1845, %r3943; + mul.f32 %r4430, %r1845, %r3944; + mul.f32 %r4431, %r1845, %r3945; + mul.f32 %r4432, %r1757, %r3946; + mul.f32 %r4433, %r1757, %r3947; + mul.f32 %r4434, %r1757, %r3948; + mul.f32 %r4435, %r1757, %r3949; + mul.f32 %r4436, %r1757, %r3950; + mul.f32 %r4437, %r1757, %r3951; + mul.f32 %r4438, %r1757, %r3952; + mul.f32 %r4439, %r1757, %r3953; + mul.f32 %r4440, %r1846, %r3954; + mul.f32 %r4441, %r1846, %r3955; + mul.f32 %r4442, %r1846, %r3956; + mul.f32 %r4443, %r1846, %r3957; + mul.f32 %r4444, %r1846, %r3958; + mul.f32 %r4445, %r1846, %r3959; + mul.f32 %r4446, %r1846, %r3960; + mul.f32 %r4447, %r1846, %r3961; + mul.f32 %r4448, %r1758, %r3962; + mul.f32 %r4449, %r1758, %r3963; + mul.f32 %r4450, %r1758, %r3964; + mul.f32 %r4451, %r1758, %r3965; + mul.f32 %r4452, %r1758, %r3966; + mul.f32 %r4453, %r1758, %r3967; + mul.f32 %r4454, %r1758, %r3968; + mul.f32 %r4455, %r1758, %r3969; + mul.f32 %r4456, %r1847, %r3970; + mul.f32 %r4457, %r1847, %r3971; + mul.f32 %r4458, %r1847, %r3972; + mul.f32 %r4459, %r1847, %r3973; + mul.f32 %r4460, %r1847, %r3974; + mul.f32 %r4461, %r1847, %r3975; + mul.f32 %r4462, %r1847, %r3976; + mul.f32 %r4463, %r1847, %r3977; + mul.f32 %r4464, %r1759, %r3978; + mul.f32 %r4465, %r1759, %r3979; + mul.f32 %r4466, %r1759, %r3980; + mul.f32 %r4467, %r1759, %r3981; + mul.f32 %r4468, %r1759, %r3982; + mul.f32 %r4469, %r1759, %r3983; + mul.f32 %r4470, %r1759, %r3984; + mul.f32 %r4471, %r1759, %r3985; + mul.f32 %r4472, %r1849, %r3986; + mul.f32 %r4473, %r1849, %r3987; + mul.f32 %r4474, %r1849, %r3988; + mul.f32 %r4475, %r1849, %r3989; + mul.f32 %r4476, %r1849, %r3990; + mul.f32 %r4477, %r1849, %r3991; + mul.f32 %r4478, %r1849, %r3992; + mul.f32 %r4479, %r1849, %r3993; + mul.f32 %r4480, %r1760, %r3994; + mul.f32 %r4481, %r1760, %r3995; + mul.f32 %r4482, %r1760, %r3996; + mul.f32 %r4483, %r1760, %r3997; + mul.f32 %r4484, %r1760, %r3998; + mul.f32 %r4485, %r1760, %r3999; + mul.f32 %r4486, %r1760, %r4000; + mul.f32 %r4487, %r1760, %r4001; + mul.f32 %r4488, %r1850, %r4002; + mul.f32 %r4489, %r1850, %r4003; + mul.f32 %r4490, %r1850, %r4004; + mul.f32 %r4491, %r1850, %r4005; + mul.f32 %r4492, %r1850, %r4006; + mul.f32 %r4493, %r1850, %r4007; + mul.f32 %r4494, %r1850, %r4008; + mul.f32 %r4495, %r1850, %r4009; + mul.f32 %r4496, %r1761, %r4010; + mul.f32 %r4497, %r1761, %r4011; + mul.f32 %r4498, %r1761, %r4012; + mul.f32 %r4499, %r1761, %r4013; + mul.f32 %r4500, %r1761, %r4014; + mul.f32 %r4501, %r1761, %r4015; + mul.f32 %r4502, %r1761, %r4016; + mul.f32 %r4503, %r1761, %r4017; + mul.f32 %r4504, %r1851, %r4018; + mul.f32 %r4505, %r1851, %r4019; + mul.f32 %r4506, %r1851, %r4020; + mul.f32 %r4507, %r1851, %r4021; + mul.f32 %r4508, %r1851, %r4022; + mul.f32 %r4509, %r1851, %r4023; + mul.f32 %r4510, %r1851, %r4024; + mul.f32 %r4511, %r1851, %r4025; + mul.f32 %r4512, %r1762, %r4026; + mul.f32 %r4513, %r1762, %r4027; + mul.f32 %r4514, %r1762, %r4028; + mul.f32 %r4515, %r1762, %r4029; + mul.f32 %r4516, %r1762, %r4030; + mul.f32 %r4517, %r1762, %r4031; + mul.f32 %r4518, %r1762, %r4032; + mul.f32 %r4519, %r1762, %r4033; + mul.f32 %r4520, %r1852, %r4034; + mul.f32 %r4521, %r1852, %r4035; + mul.f32 %r4522, %r1852, %r4036; + mul.f32 %r4523, %r1852, %r4037; + mul.f32 %r4524, %r1852, %r4038; + mul.f32 %r4525, %r1852, %r4039; + mul.f32 %r4526, %r1852, %r4040; + mul.f32 %r4527, %r1852, %r4041; + .loc 1 107 36 // l2norm.py:107:36 + neg.f32 %r4528, %r4042; + neg.f32 %r4529, %r4043; + fma.rn.f32 %r4530, %r1763, %r1395, %r4528; + fma.rn.f32 %r4531, %r1763, %r1396, %r4529; + neg.f32 %r4532, %r4044; + neg.f32 %r4533, %r4045; + fma.rn.f32 %r4534, %r1763, %r1397, %r4532; + fma.rn.f32 %r4535, %r1763, %r1398, %r4533; + neg.f32 %r4536, %r4046; + neg.f32 %r4537, %r4047; + fma.rn.f32 %r4538, %r1763, %r1399, %r4536; + fma.rn.f32 %r4539, %r1763, %r1400, %r4537; + neg.f32 %r4540, %r4048; + neg.f32 %r4541, %r4049; + fma.rn.f32 %r4542, %r1763, %r1401, %r4540; + fma.rn.f32 %r4543, %r1763, %r1402, %r4541; + mov.b64 %rd689, {%r3555, %r4050}; + mov.b64 %rd690, {%r3547, %r4051}; + sub.f32x2 %rd691, %rd688, %rd689; + sub.f32x2 %rd692, %rd687, %rd690; + mov.b64 %rd693, {%r3529, %r4052}; + mov.b64 %rd694, {%r3521, %r4053}; + sub.f32x2 %rd695, %rd686, %rd693; + sub.f32x2 %rd696, %rd685, %rd694; + mov.b64 %rd697, {%r3501, %r4054}; + mov.b64 %rd698, {%r3493, %r4055}; + sub.f32x2 %rd699, %rd684, %rd697; + sub.f32x2 %rd700, %rd683, %rd698; + mov.b64 %rd701, {%r3475, %r4056}; + mov.b64 %rd702, {%r3467, %r4057}; + sub.f32x2 %rd703, %rd682, %rd701; + sub.f32x2 %rd704, %rd681, %rd702; + mov.b64 %rd705, {%r3445, %r4058}; + mov.b64 %rd706, {%r3439, %r4059}; + sub.f32x2 %rd707, %rd680, %rd705; + sub.f32x2 %rd708, %rd679, %rd706; + mov.b64 %rd709, {%r3419, %r4060}; + mov.b64 %rd710, {%r3413, %r4061}; + sub.f32x2 %rd711, %rd678, %rd709; + sub.f32x2 %rd712, %rd677, %rd710; + mov.b64 %rd713, {%r3391, %r4062}; + mov.b64 %rd714, {%r3385, %r4063}; + sub.f32x2 %rd715, %rd676, %rd713; + sub.f32x2 %rd716, %rd675, %rd714; + mov.b64 %rd717, {%r2072, %r4064}; + mov.b64 %rd718, {%r2069, %r4065}; + sub.f32x2 %rd719, %rd674, %rd717; + sub.f32x2 %rd720, %rd673, %rd718; + mov.b64 %rd721, {%r2051, %r4066}; + mov.b64 %rd722, {%r2048, %r4067}; + sub.f32x2 %rd723, %rd672, %rd721; + sub.f32x2 %rd724, %rd671, %rd722; + mov.b64 %rd725, {%r2033, %r4068}; + mov.b64 %rd726, {%r2030, %r4069}; + sub.f32x2 %rd727, %rd670, %rd725; + sub.f32x2 %rd728, %rd669, %rd726; + mov.b64 %rd729, {%r2012, %r4070}; + mov.b64 %rd730, {%r2009, %r4071}; + sub.f32x2 %rd731, %rd668, %rd729; + sub.f32x2 %rd732, %rd667, %rd730; + mov.b64 %rd733, {%r1994, %r4072}; + mov.b64 %rd734, {%r1991, %r4073}; + sub.f32x2 %rd735, %rd666, %rd733; + sub.f32x2 %rd736, %rd665, %rd734; + mov.b64 %rd737, {%r1973, %r4074}; + mov.b64 %rd738, {%r1970, %r4075}; + sub.f32x2 %rd739, %rd664, %rd737; + sub.f32x2 %rd740, %rd663, %rd738; + mov.b64 %rd741, {%r1780, %r2337}; + mov.b64 %rd742, {%r4076, %r4077}; + sub.f32x2 %rd743, %rd741, %rd742; + mov.b64 %rd744, {%r2338, %r2339}; + mov.b64 %rd745, {%r4078, %r4079}; + sub.f32x2 %rd746, %rd744, %rd745; + mov.b64 %rd747, {%r2340, %r2341}; + mov.b64 %rd748, {%r4080, %r4081}; + sub.f32x2 %rd749, %rd747, %rd748; + mov.b64 %rd750, {%r2342, %r2343}; + mov.b64 %rd751, {%r4082, %r4083}; + sub.f32x2 %rd752, %rd750, %rd751; + mov.b64 %rd753, {%r2344, %r2345}; + mov.b64 %rd754, {%r4084, %r4085}; + sub.f32x2 %rd755, %rd753, %rd754; + mov.b64 %rd756, {%r2346, %r2347}; + mov.b64 %rd757, {%r4086, %r4087}; + sub.f32x2 %rd758, %rd756, %rd757; + mov.b64 %rd759, {%r2348, %r2349}; + mov.b64 %rd760, {%r4088, %r4089}; + sub.f32x2 %rd761, %rd759, %rd760; + mov.b64 %rd762, {%r2350, %r2351}; + mov.b64 %rd763, {%r4090, %r4091}; + sub.f32x2 %rd764, %rd762, %rd763; + mov.b64 %rd765, {%r2352, %r2353}; + mov.b64 %rd766, {%r4092, %r4093}; + sub.f32x2 %rd767, %rd765, %rd766; + mov.b64 %rd768, {%r2354, %r2355}; + mov.b64 %rd769, {%r4094, %r4095}; + sub.f32x2 %rd770, %rd768, %rd769; + mov.b64 %rd771, {%r2356, %r2357}; + mov.b64 %rd772, {%r4096, %r4097}; + sub.f32x2 %rd773, %rd771, %rd772; + mov.b64 %rd774, {%r2358, %r2359}; + mov.b64 %rd775, {%r4098, %r4099}; + sub.f32x2 %rd776, %rd774, %rd775; + mov.b64 %rd777, {%r2360, %r2361}; + mov.b64 %rd778, {%r4100, %r4101}; + sub.f32x2 %rd779, %rd777, %rd778; + mov.b64 %rd780, {%r2362, %r2363}; + mov.b64 %rd781, {%r4102, %r4103}; + sub.f32x2 %rd782, %rd780, %rd781; + mov.b64 %rd783, {%r2364, %r2365}; + mov.b64 %rd784, {%r4104, %r4105}; + sub.f32x2 %rd785, %rd783, %rd784; + mov.b64 %rd786, {%r2366, %r2367}; + mov.b64 %rd787, {%r4106, %r4107}; + sub.f32x2 %rd788, %rd786, %rd787; + mov.b64 %rd789, {%r2368, %r2369}; + mov.b64 %rd790, {%r4108, %r4109}; + sub.f32x2 %rd791, %rd789, %rd790; + mov.b64 %rd792, {%r2370, %r2371}; + mov.b64 %rd793, {%r4110, %r4111}; + sub.f32x2 %rd794, %rd792, %rd793; + mov.b64 %rd795, {%r2372, %r2373}; + mov.b64 %rd796, {%r4112, %r4113}; + sub.f32x2 %rd797, %rd795, %rd796; + mov.b64 %rd798, {%r2374, %r2375}; + mov.b64 %rd799, {%r4114, %r4115}; + sub.f32x2 %rd800, %rd798, %rd799; + mov.b64 %rd801, {%r2376, %r2377}; + mov.b64 %rd802, {%r4116, %r4117}; + sub.f32x2 %rd803, %rd801, %rd802; + mov.b64 %rd804, {%r2378, %r2379}; + mov.b64 %rd805, {%r4118, %r4119}; + sub.f32x2 %rd806, %rd804, %rd805; + mov.b64 %rd807, {%r2380, %r2381}; + mov.b64 %rd808, {%r4120, %r4121}; + sub.f32x2 %rd809, %rd807, %rd808; + mov.b64 %rd810, {%r2382, %r2383}; + mov.b64 %rd811, {%r4122, %r4123}; + sub.f32x2 %rd812, %rd810, %rd811; + mov.b64 %rd813, {%r2384, %r2385}; + mov.b64 %rd814, {%r4124, %r4125}; + sub.f32x2 %rd815, %rd813, %rd814; + mov.b64 %rd816, {%r2386, %r2387}; + mov.b64 %rd817, {%r4126, %r4127}; + sub.f32x2 %rd818, %rd816, %rd817; + mov.b64 %rd819, {%r2388, %r2389}; + mov.b64 %rd820, {%r4128, %r4129}; + sub.f32x2 %rd821, %rd819, %rd820; + mov.b64 %rd822, {%r2390, %r2391}; + mov.b64 %rd823, {%r4130, %r4131}; + sub.f32x2 %rd824, %rd822, %rd823; + mov.b64 %rd825, {%r2392, %r2393}; + mov.b64 %rd826, {%r4132, %r4133}; + sub.f32x2 %rd827, %rd825, %rd826; + mov.b64 %rd828, {%r2394, %r2395}; + mov.b64 %rd829, {%r4134, %r4135}; + sub.f32x2 %rd830, %rd828, %rd829; + mov.b64 %rd831, {%r2396, %r2397}; + mov.b64 %rd832, {%r4136, %r4137}; + sub.f32x2 %rd833, %rd831, %rd832; + mov.b64 %rd834, {%r2398, %r2399}; + mov.b64 %rd835, {%r4138, %r4139}; + sub.f32x2 %rd836, %rd834, %rd835; + mov.b64 %rd837, {%r2400, %r2401}; + mov.b64 %rd838, {%r4140, %r4141}; + sub.f32x2 %rd839, %rd837, %rd838; + mov.b64 %rd840, {%r2402, %r2403}; + mov.b64 %rd841, {%r4142, %r4143}; + sub.f32x2 %rd842, %rd840, %rd841; + mov.b64 %rd843, {%r2404, %r2405}; + mov.b64 %rd844, {%r4144, %r4145}; + sub.f32x2 %rd845, %rd843, %rd844; + mov.b64 %rd846, {%r2406, %r2407}; + mov.b64 %rd847, {%r4146, %r4147}; + sub.f32x2 %rd848, %rd846, %rd847; + mov.b64 %rd849, {%r2408, %r2409}; + mov.b64 %rd850, {%r4148, %r4149}; + sub.f32x2 %rd851, %rd849, %rd850; + mov.b64 %rd852, {%r2410, %r2411}; + mov.b64 %rd853, {%r4150, %r4151}; + sub.f32x2 %rd854, %rd852, %rd853; + mov.b64 %rd855, {%r2412, %r2413}; + mov.b64 %rd856, {%r4152, %r4153}; + sub.f32x2 %rd857, %rd855, %rd856; + mov.b64 %rd858, {%r2414, %r2415}; + mov.b64 %rd859, {%r4154, %r4155}; + sub.f32x2 %rd860, %rd858, %rd859; + mov.b64 %rd861, {%r2416, %r2417}; + mov.b64 %rd862, {%r4156, %r4157}; + sub.f32x2 %rd863, %rd861, %rd862; + mov.b64 %rd864, {%r2418, %r2419}; + mov.b64 %rd865, {%r4158, %r4159}; + sub.f32x2 %rd866, %rd864, %rd865; + mov.b64 %rd867, {%r2420, %r2421}; + mov.b64 %rd868, {%r4160, %r4161}; + sub.f32x2 %rd869, %rd867, %rd868; + mov.b64 %rd870, {%r2422, %r2423}; + mov.b64 %rd871, {%r4162, %r4163}; + sub.f32x2 %rd872, %rd870, %rd871; + mov.b64 %rd873, {%r2424, %r2425}; + mov.b64 %rd874, {%r4164, %r4165}; + sub.f32x2 %rd875, %rd873, %rd874; + mov.b64 %rd876, {%r2426, %r2427}; + mov.b64 %rd877, {%r4166, %r4167}; + sub.f32x2 %rd878, %rd876, %rd877; + mov.b64 %rd879, {%r2428, %r2429}; + mov.b64 %rd880, {%r4168, %r4169}; + sub.f32x2 %rd881, %rd879, %rd880; + mov.b64 %rd882, {%r2430, %r2431}; + mov.b64 %rd883, {%r4170, %r4171}; + sub.f32x2 %rd884, %rd882, %rd883; + mov.b64 %rd885, {%r2432, %r2433}; + mov.b64 %rd886, {%r4172, %r4173}; + sub.f32x2 %rd887, %rd885, %rd886; + mov.b64 %rd888, {%r2434, %r2435}; + mov.b64 %rd889, {%r4174, %r4175}; + sub.f32x2 %rd890, %rd888, %rd889; + mov.b64 %rd891, {%r2436, %r2437}; + mov.b64 %rd892, {%r4176, %r4177}; + sub.f32x2 %rd893, %rd891, %rd892; + mov.b64 %rd894, {%r2438, %r2439}; + mov.b64 %rd895, {%r4178, %r4179}; + sub.f32x2 %rd896, %rd894, %rd895; + mov.b64 %rd897, {%r2440, %r2441}; + mov.b64 %rd898, {%r4180, %r4181}; + sub.f32x2 %rd899, %rd897, %rd898; + mov.b64 %rd900, {%r2442, %r2443}; + mov.b64 %rd901, {%r4182, %r4183}; + sub.f32x2 %rd902, %rd900, %rd901; + mov.b64 %rd903, {%r2444, %r2445}; + mov.b64 %rd904, {%r4184, %r4185}; + sub.f32x2 %rd905, %rd903, %rd904; + mov.b64 %rd906, {%r2446, %r2447}; + mov.b64 %rd907, {%r4186, %r4187}; + sub.f32x2 %rd908, %rd906, %rd907; + mov.b64 %rd909, {%r2448, %r2449}; + mov.b64 %rd910, {%r4188, %r4189}; + sub.f32x2 %rd911, %rd909, %rd910; + mov.b64 %rd912, {%r2450, %r2451}; + mov.b64 %rd913, {%r4190, %r4191}; + sub.f32x2 %rd914, %rd912, %rd913; + mov.b64 %rd915, {%r2452, %r2453}; + mov.b64 %rd916, {%r4192, %r4193}; + sub.f32x2 %rd917, %rd915, %rd916; + mov.b64 %rd918, {%r2454, %r2455}; + mov.b64 %rd919, {%r4194, %r4195}; + sub.f32x2 %rd920, %rd918, %rd919; + mov.b64 %rd921, {%r2456, %r2457}; + mov.b64 %rd922, {%r4196, %r4197}; + sub.f32x2 %rd923, %rd921, %rd922; + mov.b64 %rd924, {%r2458, %r2459}; + mov.b64 %rd925, {%r4198, %r4199}; + sub.f32x2 %rd926, %rd924, %rd925; + mov.b64 %rd927, {%r2460, %r2461}; + mov.b64 %rd928, {%r4200, %r4201}; + sub.f32x2 %rd929, %rd927, %rd928; + mov.b64 %rd930, {%r2462, %r2463}; + mov.b64 %rd931, {%r4202, %r4203}; + sub.f32x2 %rd932, %rd930, %rd931; + mov.b64 %rd933, {%r2464, %r2465}; + mov.b64 %rd934, {%r4204, %r4205}; + sub.f32x2 %rd935, %rd933, %rd934; + mov.b64 %rd936, {%r2466, %r2467}; + mov.b64 %rd937, {%r4206, %r4207}; + sub.f32x2 %rd938, %rd936, %rd937; + mov.b64 %rd939, {%r2468, %r2469}; + mov.b64 %rd940, {%r4208, %r4209}; + sub.f32x2 %rd941, %rd939, %rd940; + mov.b64 %rd942, {%r2470, %r2471}; + mov.b64 %rd943, {%r4210, %r4211}; + sub.f32x2 %rd944, %rd942, %rd943; + mov.b64 %rd945, {%r2472, %r2473}; + mov.b64 %rd946, {%r4212, %r4213}; + sub.f32x2 %rd947, %rd945, %rd946; + mov.b64 %rd948, {%r2474, %r2475}; + mov.b64 %rd949, {%r4214, %r4215}; + sub.f32x2 %rd950, %rd948, %rd949; + mov.b64 %rd951, {%r2476, %r2477}; + mov.b64 %rd952, {%r4216, %r4217}; + sub.f32x2 %rd953, %rd951, %rd952; + mov.b64 %rd954, {%r2478, %r2479}; + mov.b64 %rd955, {%r4218, %r4219}; + sub.f32x2 %rd956, %rd954, %rd955; + mov.b64 %rd957, {%r2480, %r2481}; + mov.b64 %rd958, {%r4220, %r4221}; + sub.f32x2 %rd959, %rd957, %rd958; + mov.b64 %rd960, {%r2482, %r2483}; + mov.b64 %rd961, {%r4222, %r4223}; + sub.f32x2 %rd962, %rd960, %rd961; + mov.b64 %rd963, {%r2484, %r2485}; + mov.b64 %rd964, {%r4224, %r4225}; + sub.f32x2 %rd965, %rd963, %rd964; + mov.b64 %rd966, {%r2486, %r2487}; + mov.b64 %rd967, {%r4226, %r4227}; + sub.f32x2 %rd968, %rd966, %rd967; + mov.b64 %rd969, {%r2488, %r2489}; + mov.b64 %rd970, {%r4228, %r4229}; + sub.f32x2 %rd971, %rd969, %rd970; + mov.b64 %rd972, {%r2490, %r2491}; + mov.b64 %rd973, {%r4230, %r4231}; + sub.f32x2 %rd974, %rd972, %rd973; + mov.b64 %rd975, {%r2492, %r2493}; + mov.b64 %rd976, {%r4232, %r4233}; + sub.f32x2 %rd977, %rd975, %rd976; + mov.b64 %rd978, {%r2494, %r2495}; + mov.b64 %rd979, {%r4234, %r4235}; + sub.f32x2 %rd980, %rd978, %rd979; + mov.b64 %rd981, {%r2496, %r2497}; + mov.b64 %rd982, {%r4236, %r4237}; + sub.f32x2 %rd983, %rd981, %rd982; + mov.b64 %rd984, {%r2498, %r2499}; + mov.b64 %rd985, {%r4238, %r4239}; + sub.f32x2 %rd986, %rd984, %rd985; + mov.b64 %rd987, {%r2500, %r2501}; + mov.b64 %rd988, {%r4240, %r4241}; + sub.f32x2 %rd989, %rd987, %rd988; + mov.b64 %rd990, {%r2503, %r2505}; + mov.b64 %rd991, {%r4242, %r4243}; + sub.f32x2 %rd992, %rd990, %rd991; + mov.b64 %rd993, {%r2507, %r2509}; + mov.b64 %rd994, {%r4244, %r4245}; + sub.f32x2 %rd995, %rd993, %rd994; + mov.b64 %rd996, {%r2511, %r2513}; + mov.b64 %rd997, {%r4246, %r4247}; + sub.f32x2 %rd998, %rd996, %rd997; + mov.b64 %rd999, {%r2515, %r2518}; + mov.b64 %rd1000, {%r4248, %r4249}; + sub.f32x2 %rd1001, %rd999, %rd1000; + mov.b64 %rd1002, {%r2520, %r2522}; + mov.b64 %rd1003, {%r4250, %r4251}; + sub.f32x2 %rd1004, %rd1002, %rd1003; + mov.b64 %rd1005, {%r2524, %r2526}; + mov.b64 %rd1006, {%r4252, %r4253}; + sub.f32x2 %rd1007, %rd1005, %rd1006; + mov.b64 %rd1008, {%r2528, %r2530}; + mov.b64 %rd1009, {%r4254, %r4255}; + sub.f32x2 %rd1010, %rd1008, %rd1009; + mov.b64 %rd1011, {%r2532, %r2533}; + mov.b64 %rd1012, {%r4256, %r4257}; + sub.f32x2 %rd1013, %rd1011, %rd1012; + mov.b64 %rd1014, {%r2535, %r2537}; + mov.b64 %rd1015, {%r4258, %r4259}; + sub.f32x2 %rd1016, %rd1014, %rd1015; + mov.b64 %rd1017, {%r2539, %r2541}; + mov.b64 %rd1018, {%r4260, %r4261}; + sub.f32x2 %rd1019, %rd1017, %rd1018; + mov.b64 %rd1020, {%r2543, %r2545}; + mov.b64 %rd1021, {%r4262, %r4263}; + sub.f32x2 %rd1022, %rd1020, %rd1021; + mov.b64 %rd1023, {%r2547, %r2550}; + mov.b64 %rd1024, {%r4264, %r4265}; + sub.f32x2 %rd1025, %rd1023, %rd1024; + mov.b64 %rd1026, {%r2552, %r2554}; + mov.b64 %rd1027, {%r4266, %r4267}; + sub.f32x2 %rd1028, %rd1026, %rd1027; + mov.b64 %rd1029, {%r2556, %r2558}; + mov.b64 %rd1030, {%r4268, %r4269}; + sub.f32x2 %rd1031, %rd1029, %rd1030; + mov.b64 %rd1032, {%r2560, %r2562}; + mov.b64 %rd1033, {%r4270, %r4271}; + sub.f32x2 %rd1034, %rd1032, %rd1033; + mov.b64 %rd1035, {%r2564, %r2565}; + mov.b64 %rd1036, {%r4272, %r4273}; + sub.f32x2 %rd1037, %rd1035, %rd1036; + mov.b64 %rd1038, {%r2567, %r2569}; + mov.b64 %rd1039, {%r4274, %r4275}; + sub.f32x2 %rd1040, %rd1038, %rd1039; + mov.b64 %rd1041, {%r2571, %r2573}; + mov.b64 %rd1042, {%r4276, %r4277}; + sub.f32x2 %rd1043, %rd1041, %rd1042; + mov.b64 %rd1044, {%r2575, %r2577}; + mov.b64 %rd1045, {%r4278, %r4279}; + sub.f32x2 %rd1046, %rd1044, %rd1045; + mov.b64 %rd1047, {%r2579, %r2582}; + mov.b64 %rd1048, {%r4280, %r4281}; + sub.f32x2 %rd1049, %rd1047, %rd1048; + mov.b64 %rd1050, {%r2584, %r2586}; + mov.b64 %rd1051, {%r4282, %r4283}; + sub.f32x2 %rd1052, %rd1050, %rd1051; + mov.b64 %rd1053, {%r2588, %r2590}; + mov.b64 %rd1054, {%r4284, %r4285}; + sub.f32x2 %rd1055, %rd1053, %rd1054; + mov.b64 %rd1056, {%r2592, %r2594}; + mov.b64 %rd1057, {%r4286, %r4287}; + sub.f32x2 %rd1058, %rd1056, %rd1057; + mov.b64 %rd1059, {%r2596, %r2597}; + mov.b64 %rd1060, {%r4288, %r4289}; + sub.f32x2 %rd1061, %rd1059, %rd1060; + mov.b64 %rd1062, {%r2599, %r2601}; + mov.b64 %rd1063, {%r4290, %r4291}; + sub.f32x2 %rd1064, %rd1062, %rd1063; + mov.b64 %rd1065, {%r2603, %r2605}; + mov.b64 %rd1066, {%r4292, %r4293}; + sub.f32x2 %rd1067, %rd1065, %rd1066; + mov.b64 %rd1068, {%r2607, %r2609}; + mov.b64 %rd1069, {%r4294, %r4295}; + sub.f32x2 %rd1070, %rd1068, %rd1069; + mov.b64 %rd1071, {%r2611, %r2614}; + mov.b64 %rd1072, {%r4296, %r4297}; + sub.f32x2 %rd1073, %rd1071, %rd1072; + mov.b64 %rd1074, {%r2616, %r2618}; + mov.b64 %rd1075, {%r4298, %r4299}; + sub.f32x2 %rd1076, %rd1074, %rd1075; + mov.b64 %rd1077, {%r2620, %r2622}; + mov.b64 %rd1078, {%r4300, %r4301}; + sub.f32x2 %rd1079, %rd1077, %rd1078; + mov.b64 %rd1080, {%r2624, %r2626}; + mov.b64 %rd1081, {%r4302, %r4303}; + sub.f32x2 %rd1082, %rd1080, %rd1081; + mov.b64 %rd1083, {%r2628, %r2629}; + mov.b64 %rd1084, {%r4304, %r4305}; + sub.f32x2 %rd1085, %rd1083, %rd1084; + mov.b64 %rd1086, {%r2631, %r2633}; + mov.b64 %rd1087, {%r4306, %r4307}; + sub.f32x2 %rd1088, %rd1086, %rd1087; + mov.b64 %rd1089, {%r2635, %r2637}; + mov.b64 %rd1090, {%r4308, %r4309}; + sub.f32x2 %rd1091, %rd1089, %rd1090; + mov.b64 %rd1092, {%r2639, %r2641}; + mov.b64 %rd1093, {%r4310, %r4311}; + sub.f32x2 %rd1094, %rd1092, %rd1093; + mov.b64 %rd1095, {%r2643, %r2646}; + mov.b64 %rd1096, {%r4312, %r4313}; + sub.f32x2 %rd1097, %rd1095, %rd1096; + mov.b64 %rd1098, {%r2648, %r2650}; + mov.b64 %rd1099, {%r4314, %r4315}; + sub.f32x2 %rd1100, %rd1098, %rd1099; + mov.b64 %rd1101, {%r2652, %r2654}; + mov.b64 %rd1102, {%r4316, %r4317}; + sub.f32x2 %rd1103, %rd1101, %rd1102; + mov.b64 %rd1104, {%r2656, %r2658}; + mov.b64 %rd1105, {%r4318, %r4319}; + sub.f32x2 %rd1106, %rd1104, %rd1105; + mov.b64 %rd1107, {%r2660, %r2661}; + mov.b64 %rd1108, {%r4320, %r4321}; + sub.f32x2 %rd1109, %rd1107, %rd1108; + mov.b64 %rd1110, {%r2663, %r2665}; + mov.b64 %rd1111, {%r4322, %r4323}; + sub.f32x2 %rd1112, %rd1110, %rd1111; + mov.b64 %rd1113, {%r2667, %r2669}; + mov.b64 %rd1114, {%r4324, %r4325}; + sub.f32x2 %rd1115, %rd1113, %rd1114; + mov.b64 %rd1116, {%r2671, %r2673}; + mov.b64 %rd1117, {%r4326, %r4327}; + sub.f32x2 %rd1118, %rd1116, %rd1117; + mov.b64 %rd1119, {%r2675, %r2678}; + mov.b64 %rd1120, {%r4328, %r4329}; + sub.f32x2 %rd1121, %rd1119, %rd1120; + mov.b64 %rd1122, {%r2680, %r2682}; + mov.b64 %rd1123, {%r4330, %r4331}; + sub.f32x2 %rd1124, %rd1122, %rd1123; + mov.b64 %rd1125, {%r2684, %r2686}; + mov.b64 %rd1126, {%r4332, %r4333}; + sub.f32x2 %rd1127, %rd1125, %rd1126; + mov.b64 %rd1128, {%r2688, %r2690}; + mov.b64 %rd1129, {%r4334, %r4335}; + sub.f32x2 %rd1130, %rd1128, %rd1129; + mov.b64 %rd1131, {%r2692, %r2693}; + mov.b64 %rd1132, {%r4336, %r4337}; + sub.f32x2 %rd1133, %rd1131, %rd1132; + mov.b64 %rd1134, {%r2695, %r2697}; + mov.b64 %rd1135, {%r4338, %r4339}; + sub.f32x2 %rd1136, %rd1134, %rd1135; + mov.b64 %rd1137, {%r2699, %r2701}; + mov.b64 %rd1138, {%r4340, %r4341}; + sub.f32x2 %rd1139, %rd1137, %rd1138; + mov.b64 %rd1140, {%r2703, %r2705}; + mov.b64 %rd1141, {%r4342, %r4343}; + sub.f32x2 %rd1142, %rd1140, %rd1141; + mov.b64 %rd1143, {%r2707, %r2710}; + mov.b64 %rd1144, {%r4344, %r4345}; + sub.f32x2 %rd1145, %rd1143, %rd1144; + mov.b64 %rd1146, {%r2712, %r2714}; + mov.b64 %rd1147, {%r4346, %r4347}; + sub.f32x2 %rd1148, %rd1146, %rd1147; + mov.b64 %rd1149, {%r2716, %r2718}; + mov.b64 %rd1150, {%r4348, %r4349}; + sub.f32x2 %rd1151, %rd1149, %rd1150; + mov.b64 %rd1152, {%r2720, %r2722}; + mov.b64 %rd1153, {%r4350, %r4351}; + sub.f32x2 %rd1154, %rd1152, %rd1153; + mov.b64 %rd1155, {%r2724, %r2725}; + mov.b64 %rd1156, {%r4352, %r4353}; + sub.f32x2 %rd1157, %rd1155, %rd1156; + mov.b64 %rd1158, {%r2727, %r2729}; + mov.b64 %rd1159, {%r4354, %r4355}; + sub.f32x2 %rd1160, %rd1158, %rd1159; + mov.b64 %rd1161, {%r2731, %r2733}; + mov.b64 %rd1162, {%r4356, %r4357}; + sub.f32x2 %rd1163, %rd1161, %rd1162; + mov.b64 %rd1164, {%r2735, %r2737}; + mov.b64 %rd1165, {%r4358, %r4359}; + sub.f32x2 %rd1166, %rd1164, %rd1165; + mov.b64 %rd1167, {%r2739, %r2742}; + mov.b64 %rd1168, {%r4360, %r4361}; + sub.f32x2 %rd1169, %rd1167, %rd1168; + mov.b64 %rd1170, {%r2744, %r2746}; + mov.b64 %rd1171, {%r4362, %r4363}; + sub.f32x2 %rd1172, %rd1170, %rd1171; + mov.b64 %rd1173, {%r2748, %r2750}; + mov.b64 %rd1174, {%r4364, %r4365}; + sub.f32x2 %rd1175, %rd1173, %rd1174; + mov.b64 %rd1176, {%r2752, %r2754}; + mov.b64 %rd1177, {%r4366, %r4367}; + sub.f32x2 %rd1178, %rd1176, %rd1177; + mov.b64 %rd1179, {%r2756, %r2757}; + mov.b64 %rd1180, {%r4368, %r4369}; + sub.f32x2 %rd1181, %rd1179, %rd1180; + mov.b64 %rd1182, {%r2758, %r2759}; + mov.b64 %rd1183, {%r4370, %r4371}; + sub.f32x2 %rd1184, %rd1182, %rd1183; + mov.b64 %rd1185, {%r2760, %r2761}; + mov.b64 %rd1186, {%r4372, %r4373}; + sub.f32x2 %rd1187, %rd1185, %rd1186; + mov.b64 %rd1188, {%r2762, %r2764}; + mov.b64 %rd1189, {%r4374, %r4375}; + sub.f32x2 %rd1190, %rd1188, %rd1189; + mov.b64 %rd1191, {%r2766, %r2779}; + mov.b64 %rd1192, {%r4376, %r4377}; + sub.f32x2 %rd1193, %rd1191, %rd1192; + mov.b64 %rd1194, {%r2781, %r2783}; + mov.b64 %rd1195, {%r4378, %r4379}; + sub.f32x2 %rd1196, %rd1194, %rd1195; + mov.b64 %rd1197, {%r2785, %r2787}; + mov.b64 %rd1198, {%r4380, %r4381}; + sub.f32x2 %rd1199, %rd1197, %rd1198; + mov.b64 %rd1200, {%r2789, %r2791}; + mov.b64 %rd1201, {%r4382, %r4383}; + sub.f32x2 %rd1202, %rd1200, %rd1201; + mov.b64 %rd1203, {%r2793, %r2811}; + mov.b64 %rd1204, {%r4384, %r4385}; + sub.f32x2 %rd1205, %rd1203, %rd1204; + mov.b64 %rd1206, {%r2813, %r2815}; + mov.b64 %rd1207, {%r4386, %r4387}; + sub.f32x2 %rd1208, %rd1206, %rd1207; + mov.b64 %rd1209, {%r2817, %r2819}; + mov.b64 %rd1210, {%r4388, %r4389}; + sub.f32x2 %rd1211, %rd1209, %rd1210; + mov.b64 %rd1212, {%r2821, %r2823}; + mov.b64 %rd1213, {%r4390, %r4391}; + sub.f32x2 %rd1214, %rd1212, %rd1213; + mov.b64 %rd1215, {%r2825, %r2843}; + mov.b64 %rd1216, {%r4392, %r4393}; + sub.f32x2 %rd1217, %rd1215, %rd1216; + mov.b64 %rd1218, {%r2845, %r2847}; + mov.b64 %rd1219, {%r4394, %r4395}; + sub.f32x2 %rd1220, %rd1218, %rd1219; + mov.b64 %rd1221, {%r2849, %r2851}; + mov.b64 %rd1222, {%r4396, %r4397}; + sub.f32x2 %rd1223, %rd1221, %rd1222; + mov.b64 %rd1224, {%r2853, %r2855}; + mov.b64 %rd1225, {%r4398, %r4399}; + sub.f32x2 %rd1226, %rd1224, %rd1225; + mov.b64 %rd1227, {%r2857, %r2875}; + mov.b64 %rd1228, {%r4400, %r4401}; + sub.f32x2 %rd1229, %rd1227, %rd1228; + mov.b64 %rd1230, {%r2877, %r2879}; + mov.b64 %rd1231, {%r4402, %r4403}; + sub.f32x2 %rd1232, %rd1230, %rd1231; + mov.b64 %rd1233, {%r2881, %r2883}; + mov.b64 %rd1234, {%r4404, %r4405}; + sub.f32x2 %rd1235, %rd1233, %rd1234; + mov.b64 %rd1236, {%r2885, %r2887}; + mov.b64 %rd1237, {%r4406, %r4407}; + sub.f32x2 %rd1238, %rd1236, %rd1237; + mov.b64 %rd1239, {%r2889, %r2907}; + mov.b64 %rd1240, {%r4408, %r4409}; + sub.f32x2 %rd1241, %rd1239, %rd1240; + mov.b64 %rd1242, {%r2909, %r2911}; + mov.b64 %rd1243, {%r4410, %r4411}; + sub.f32x2 %rd1244, %rd1242, %rd1243; + mov.b64 %rd1245, {%r2913, %r2915}; + mov.b64 %rd1246, {%r4412, %r4413}; + sub.f32x2 %rd1247, %rd1245, %rd1246; + mov.b64 %rd1248, {%r2917, %r2919}; + mov.b64 %rd1249, {%r4414, %r4415}; + sub.f32x2 %rd1250, %rd1248, %rd1249; + mov.b64 %rd1251, {%r2921, %r2923}; + mov.b64 %rd1252, {%r4416, %r4417}; + sub.f32x2 %rd1253, %rd1251, %rd1252; + mov.b64 %rd1254, {%r2925, %r2927}; + mov.b64 %rd1255, {%r4418, %r4419}; + sub.f32x2 %rd1256, %rd1254, %rd1255; + mov.b64 %rd1257, {%r2929, %r2931}; + mov.b64 %rd1258, {%r4420, %r4421}; + sub.f32x2 %rd1259, %rd1257, %rd1258; + mov.b64 %rd1260, {%r2933, %r2935}; + mov.b64 %rd1261, {%r4422, %r4423}; + sub.f32x2 %rd1262, %rd1260, %rd1261; + mov.b64 %rd1263, {%r2937, %r2939}; + mov.b64 %rd1264, {%r4424, %r4425}; + sub.f32x2 %rd1265, %rd1263, %rd1264; + mov.b64 %rd1266, {%r2941, %r2943}; + mov.b64 %rd1267, {%r4426, %r4427}; + sub.f32x2 %rd1268, %rd1266, %rd1267; + mov.b64 %rd1269, {%r2945, %r2947}; + mov.b64 %rd1270, {%r4428, %r4429}; + sub.f32x2 %rd1271, %rd1269, %rd1270; + mov.b64 %rd1272, {%r2949, %r2951}; + mov.b64 %rd1273, {%r4430, %r4431}; + sub.f32x2 %rd1274, %rd1272, %rd1273; + mov.b64 %rd1275, {%r2953, %r2955}; + mov.b64 %rd1276, {%r4432, %r4433}; + sub.f32x2 %rd1277, %rd1275, %rd1276; + mov.b64 %rd1278, {%r2957, %r2959}; + mov.b64 %rd1279, {%r4434, %r4435}; + sub.f32x2 %rd1280, %rd1278, %rd1279; + mov.b64 %rd1281, {%r2961, %r2963}; + mov.b64 %rd1282, {%r4436, %r4437}; + sub.f32x2 %rd1283, %rd1281, %rd1282; + mov.b64 %rd1284, {%r2965, %r2967}; + mov.b64 %rd1285, {%r4438, %r4439}; + sub.f32x2 %rd1286, %rd1284, %rd1285; + mov.b64 %rd1287, {%r2969, %r2971}; + mov.b64 %rd1288, {%r4440, %r4441}; + sub.f32x2 %rd1289, %rd1287, %rd1288; + mov.b64 %rd1290, {%r2973, %r2975}; + mov.b64 %rd1291, {%r4442, %r4443}; + sub.f32x2 %rd1292, %rd1290, %rd1291; + mov.b64 %rd1293, {%r2977, %r2979}; + mov.b64 %rd1294, {%r4444, %r4445}; + sub.f32x2 %rd1295, %rd1293, %rd1294; + mov.b64 %rd1296, {%r2981, %r2983}; + mov.b64 %rd1297, {%r4446, %r4447}; + sub.f32x2 %rd1298, %rd1296, %rd1297; + mov.b64 %rd1299, {%r2985, %r2987}; + mov.b64 %rd1300, {%r4448, %r4449}; + sub.f32x2 %rd1301, %rd1299, %rd1300; + mov.b64 %rd1302, {%r2989, %r2991}; + mov.b64 %rd1303, {%r4450, %r4451}; + sub.f32x2 %rd1304, %rd1302, %rd1303; + mov.b64 %rd1305, {%r2993, %r2995}; + mov.b64 %rd1306, {%r4452, %r4453}; + sub.f32x2 %rd1307, %rd1305, %rd1306; + mov.b64 %rd1308, {%r2997, %r2999}; + mov.b64 %rd1309, {%r4454, %r4455}; + sub.f32x2 %rd1310, %rd1308, %rd1309; + mov.b64 %rd1311, {%r3001, %r3003}; + mov.b64 %rd1312, {%r4456, %r4457}; + sub.f32x2 %rd1313, %rd1311, %rd1312; + mov.b64 %rd1314, {%r3005, %r3007}; + mov.b64 %rd1315, {%r4458, %r4459}; + sub.f32x2 %rd1316, %rd1314, %rd1315; + mov.b64 %rd1317, {%r3009, %r3011}; + mov.b64 %rd1318, {%r4460, %r4461}; + sub.f32x2 %rd1319, %rd1317, %rd1318; + mov.b64 %rd1320, {%r3013, %r3015}; + mov.b64 %rd1321, {%r4462, %r4463}; + sub.f32x2 %rd1322, %rd1320, %rd1321; + mov.b64 %rd1323, {%r3017, %r3019}; + mov.b64 %rd1324, {%r4464, %r4465}; + sub.f32x2 %rd1325, %rd1323, %rd1324; + mov.b64 %rd1326, {%r3021, %r3023}; + mov.b64 %rd1327, {%r4466, %r4467}; + sub.f32x2 %rd1328, %rd1326, %rd1327; + mov.b64 %rd1329, {%r3025, %r3027}; + mov.b64 %rd1330, {%r4468, %r4469}; + sub.f32x2 %rd1331, %rd1329, %rd1330; + mov.b64 %rd1332, {%r3029, %r3031}; + mov.b64 %rd1333, {%r4470, %r4471}; + sub.f32x2 %rd1334, %rd1332, %rd1333; + mov.b64 %rd1335, {%r3033, %r3035}; + mov.b64 %rd1336, {%r4472, %r4473}; + sub.f32x2 %rd1337, %rd1335, %rd1336; + mov.b64 %rd1338, {%r3037, %r3039}; + mov.b64 %rd1339, {%r4474, %r4475}; + sub.f32x2 %rd1340, %rd1338, %rd1339; + mov.b64 %rd1341, {%r3041, %r3043}; + mov.b64 %rd1342, {%r4476, %r4477}; + sub.f32x2 %rd1343, %rd1341, %rd1342; + mov.b64 %rd1344, {%r3045, %r3047}; + mov.b64 %rd1345, {%r4478, %r4479}; + sub.f32x2 %rd1346, %rd1344, %rd1345; + mov.b64 %rd1347, {%r3049, %r3051}; + mov.b64 %rd1348, {%r4480, %r4481}; + sub.f32x2 %rd1349, %rd1347, %rd1348; + mov.b64 %rd1350, {%r3053, %r3055}; + mov.b64 %rd1351, {%r4482, %r4483}; + sub.f32x2 %rd1352, %rd1350, %rd1351; + mov.b64 %rd1353, {%r3057, %r3059}; + mov.b64 %rd1354, {%r4484, %r4485}; + sub.f32x2 %rd1355, %rd1353, %rd1354; + mov.b64 %rd1356, {%r3061, %r3063}; + mov.b64 %rd1357, {%r4486, %r4487}; + sub.f32x2 %rd1358, %rd1356, %rd1357; + mov.b64 %rd1359, {%r3065, %r3067}; + mov.b64 %rd1360, {%r4488, %r4489}; + sub.f32x2 %rd1361, %rd1359, %rd1360; + mov.b64 %rd1362, {%r3069, %r3071}; + mov.b64 %rd1363, {%r4490, %r4491}; + sub.f32x2 %rd1364, %rd1362, %rd1363; + mov.b64 %rd1365, {%r3073, %r3075}; + mov.b64 %rd1366, {%r4492, %r4493}; + sub.f32x2 %rd1367, %rd1365, %rd1366; + mov.b64 %rd1368, {%r3077, %r3079}; + mov.b64 %rd1369, {%r4494, %r4495}; + sub.f32x2 %rd1370, %rd1368, %rd1369; + mov.b64 %rd1371, {%r3081, %r3083}; + mov.b64 %rd1372, {%r4496, %r4497}; + sub.f32x2 %rd1373, %rd1371, %rd1372; + mov.b64 %rd1374, {%r3085, %r3087}; + mov.b64 %rd1375, {%r4498, %r4499}; + sub.f32x2 %rd1376, %rd1374, %rd1375; + mov.b64 %rd1377, {%r3089, %r3091}; + mov.b64 %rd1378, {%r4500, %r4501}; + sub.f32x2 %rd1379, %rd1377, %rd1378; + mov.b64 %rd1380, {%r3093, %r3095}; + mov.b64 %rd1381, {%r4502, %r4503}; + sub.f32x2 %rd1382, %rd1380, %rd1381; + mov.b64 %rd1383, {%r3097, %r3099}; + mov.b64 %rd1384, {%r4504, %r4505}; + sub.f32x2 %rd1385, %rd1383, %rd1384; + mov.b64 %rd1386, {%r3101, %r3103}; + mov.b64 %rd1387, {%r4506, %r4507}; + sub.f32x2 %rd1388, %rd1386, %rd1387; + mov.b64 %rd1389, {%r3105, %r3107}; + mov.b64 %rd1390, {%r4508, %r4509}; + sub.f32x2 %rd1391, %rd1389, %rd1390; + mov.b64 %rd1392, {%r3109, %r3111}; + mov.b64 %rd1393, {%r4510, %r4511}; + sub.f32x2 %rd1394, %rd1392, %rd1393; + mov.b64 %rd1395, {%r3113, %r3115}; + mov.b64 %rd1396, {%r4512, %r4513}; + sub.f32x2 %rd1397, %rd1395, %rd1396; + mov.b64 %rd1398, {%r3117, %r3119}; + mov.b64 %rd1399, {%r4514, %r4515}; + sub.f32x2 %rd1400, %rd1398, %rd1399; + mov.b64 %rd1401, {%r3121, %r3123}; + mov.b64 %rd1402, {%r4516, %r4517}; + sub.f32x2 %rd1403, %rd1401, %rd1402; + mov.b64 %rd1404, {%r3125, %r3127}; + mov.b64 %rd1405, {%r4518, %r4519}; + sub.f32x2 %rd1406, %rd1404, %rd1405; + mov.b64 %rd1407, {%r3129, %r3131}; + mov.b64 %rd1408, {%r4520, %r4521}; + sub.f32x2 %rd1409, %rd1407, %rd1408; + mov.b64 %rd1410, {%r3133, %r3135}; + mov.b64 %rd1411, {%r4522, %r4523}; + sub.f32x2 %rd1412, %rd1410, %rd1411; + mov.b64 %rd1413, {%r3137, %r3139}; + mov.b64 %rd1414, {%r4524, %r4525}; + sub.f32x2 %rd1415, %rd1413, %rd1414; + mov.b64 %rd1416, {%r3141, %r3143}; + mov.b64 %rd1417, {%r4526, %r4527}; + sub.f32x2 %rd1418, %rd1416, %rd1417; + .loc 1 108 27 // l2norm.py:108:27 + mov.b64 {%r4544, %r4545}, %rd691; + cvt.rn.bf16x2.f32 %r522, %r4545, %r4544; + mov.b64 {%r4546, %r4547}, %rd692; + cvt.rn.bf16x2.f32 %r523, %r4547, %r4546; + mov.b64 {%r4548, %r4549}, %rd695; + cvt.rn.bf16x2.f32 %r524, %r4549, %r4548; + mov.b64 {%r4550, %r4551}, %rd696; + cvt.rn.bf16x2.f32 %r525, %r4551, %r4550; + mov.b64 {%r4552, %r4553}, %rd699; + cvt.rn.bf16x2.f32 %r526, %r4553, %r4552; + mov.b64 {%r4554, %r4555}, %rd700; + cvt.rn.bf16x2.f32 %r527, %r4555, %r4554; + mov.b64 {%r4556, %r4557}, %rd703; + cvt.rn.bf16x2.f32 %r528, %r4557, %r4556; + mov.b64 {%r4558, %r4559}, %rd704; + cvt.rn.bf16x2.f32 %r529, %r4559, %r4558; + mov.b64 {%r4560, %r4561}, %rd707; + cvt.rn.bf16x2.f32 %r530, %r4561, %r4560; + mov.b64 {%r4562, %r4563}, %rd708; + cvt.rn.bf16x2.f32 %r531, %r4563, %r4562; + mov.b64 {%r4564, %r4565}, %rd711; + cvt.rn.bf16x2.f32 %r532, %r4565, %r4564; + mov.b64 {%r4566, %r4567}, %rd712; + cvt.rn.bf16x2.f32 %r533, %r4567, %r4566; + mov.b64 {%r4568, %r4569}, %rd715; + cvt.rn.bf16x2.f32 %r534, %r4569, %r4568; + mov.b64 {%r4570, %r4571}, %rd716; + cvt.rn.bf16x2.f32 %r535, %r4571, %r4570; + mov.b64 {%r4572, %r4573}, %rd719; + cvt.rn.bf16x2.f32 %r536, %r4573, %r4572; + mov.b64 {%r4574, %r4575}, %rd720; + cvt.rn.bf16x2.f32 %r537, %r4575, %r4574; + mov.b64 {%r4576, %r4577}, %rd723; + cvt.rn.bf16x2.f32 %r538, %r4577, %r4576; + mov.b64 {%r4578, %r4579}, %rd724; + cvt.rn.bf16x2.f32 %r539, %r4579, %r4578; + mov.b64 {%r4580, %r4581}, %rd727; + cvt.rn.bf16x2.f32 %r540, %r4581, %r4580; + mov.b64 {%r4582, %r4583}, %rd728; + cvt.rn.bf16x2.f32 %r541, %r4583, %r4582; + mov.b64 {%r4584, %r4585}, %rd731; + cvt.rn.bf16x2.f32 %r542, %r4585, %r4584; + mov.b64 {%r4586, %r4587}, %rd732; + cvt.rn.bf16x2.f32 %r543, %r4587, %r4586; + mov.b64 {%r4588, %r4589}, %rd735; + cvt.rn.bf16x2.f32 %r544, %r4589, %r4588; + mov.b64 {%r4590, %r4591}, %rd736; + cvt.rn.bf16x2.f32 %r545, %r4591, %r4590; + mov.b64 {%r4592, %r4593}, %rd739; + cvt.rn.bf16x2.f32 %r546, %r4593, %r4592; + mov.b64 {%r4594, %r4595}, %rd740; + cvt.rn.bf16x2.f32 %r547, %r4595, %r4594; + mov.b64 {%r4596, %r4597}, %rd743; + cvt.rn.bf16x2.f32 %r548, %r4597, %r4596; + mov.b64 {%r4598, %r4599}, %rd746; + cvt.rn.bf16x2.f32 %r549, %r4599, %r4598; + mov.b64 {%r4600, %r4601}, %rd749; + cvt.rn.bf16x2.f32 %r550, %r4601, %r4600; + mov.b64 {%r4602, %r4603}, %rd752; + cvt.rn.bf16x2.f32 %r551, %r4603, %r4602; + mov.b64 {%r4604, %r4605}, %rd755; + cvt.rn.bf16x2.f32 %r552, %r4605, %r4604; + mov.b64 {%r4606, %r4607}, %rd758; + cvt.rn.bf16x2.f32 %r553, %r4607, %r4606; + mov.b64 {%r4608, %r4609}, %rd761; + cvt.rn.bf16x2.f32 %r554, %r4609, %r4608; + mov.b64 {%r4610, %r4611}, %rd764; + cvt.rn.bf16x2.f32 %r555, %r4611, %r4610; + mov.b64 {%r4612, %r4613}, %rd767; + cvt.rn.bf16x2.f32 %r556, %r4613, %r4612; + mov.b64 {%r4614, %r4615}, %rd770; + cvt.rn.bf16x2.f32 %r557, %r4615, %r4614; + mov.b64 {%r4616, %r4617}, %rd773; + cvt.rn.bf16x2.f32 %r558, %r4617, %r4616; + mov.b64 {%r4618, %r4619}, %rd776; + cvt.rn.bf16x2.f32 %r559, %r4619, %r4618; + mov.b64 {%r4620, %r4621}, %rd779; + cvt.rn.bf16x2.f32 %r560, %r4621, %r4620; + mov.b64 {%r4622, %r4623}, %rd782; + cvt.rn.bf16x2.f32 %r561, %r4623, %r4622; + mov.b64 {%r4624, %r4625}, %rd785; + cvt.rn.bf16x2.f32 %r562, %r4625, %r4624; + mov.b64 {%r4626, %r4627}, %rd788; + cvt.rn.bf16x2.f32 %r563, %r4627, %r4626; + mov.b64 {%r4628, %r4629}, %rd791; + cvt.rn.bf16x2.f32 %r564, %r4629, %r4628; + mov.b64 {%r4630, %r4631}, %rd794; + cvt.rn.bf16x2.f32 %r565, %r4631, %r4630; + mov.b64 {%r4632, %r4633}, %rd797; + cvt.rn.bf16x2.f32 %r566, %r4633, %r4632; + mov.b64 {%r4634, %r4635}, %rd800; + cvt.rn.bf16x2.f32 %r567, %r4635, %r4634; + mov.b64 {%r4636, %r4637}, %rd803; + cvt.rn.bf16x2.f32 %r568, %r4637, %r4636; + mov.b64 {%r4638, %r4639}, %rd806; + cvt.rn.bf16x2.f32 %r569, %r4639, %r4638; + mov.b64 {%r4640, %r4641}, %rd809; + cvt.rn.bf16x2.f32 %r570, %r4641, %r4640; + mov.b64 {%r4642, %r4643}, %rd812; + cvt.rn.bf16x2.f32 %r571, %r4643, %r4642; + mov.b64 {%r4644, %r4645}, %rd815; + cvt.rn.bf16x2.f32 %r572, %r4645, %r4644; + mov.b64 {%r4646, %r4647}, %rd818; + cvt.rn.bf16x2.f32 %r573, %r4647, %r4646; + mov.b64 {%r4648, %r4649}, %rd821; + cvt.rn.bf16x2.f32 %r574, %r4649, %r4648; + mov.b64 {%r4650, %r4651}, %rd824; + cvt.rn.bf16x2.f32 %r575, %r4651, %r4650; + mov.b64 {%r4652, %r4653}, %rd827; + cvt.rn.bf16x2.f32 %r576, %r4653, %r4652; + mov.b64 {%r4654, %r4655}, %rd830; + cvt.rn.bf16x2.f32 %r577, %r4655, %r4654; + mov.b64 {%r4656, %r4657}, %rd833; + cvt.rn.bf16x2.f32 %r578, %r4657, %r4656; + mov.b64 {%r4658, %r4659}, %rd836; + cvt.rn.bf16x2.f32 %r579, %r4659, %r4658; + mov.b64 {%r4660, %r4661}, %rd839; + cvt.rn.bf16x2.f32 %r580, %r4661, %r4660; + mov.b64 {%r4662, %r4663}, %rd842; + cvt.rn.bf16x2.f32 %r581, %r4663, %r4662; + mov.b64 {%r4664, %r4665}, %rd845; + cvt.rn.bf16x2.f32 %r582, %r4665, %r4664; + mov.b64 {%r4666, %r4667}, %rd848; + cvt.rn.bf16x2.f32 %r583, %r4667, %r4666; + mov.b64 {%r4668, %r4669}, %rd851; + cvt.rn.bf16x2.f32 %r584, %r4669, %r4668; + mov.b64 {%r4670, %r4671}, %rd854; + cvt.rn.bf16x2.f32 %r585, %r4671, %r4670; + mov.b64 {%r4672, %r4673}, %rd857; + cvt.rn.bf16x2.f32 %r586, %r4673, %r4672; + mov.b64 {%r4674, %r4675}, %rd860; + cvt.rn.bf16x2.f32 %r587, %r4675, %r4674; + mov.b64 {%r4676, %r4677}, %rd863; + cvt.rn.bf16x2.f32 %r588, %r4677, %r4676; + mov.b64 {%r4678, %r4679}, %rd866; + cvt.rn.bf16x2.f32 %r589, %r4679, %r4678; + mov.b64 {%r4680, %r4681}, %rd869; + cvt.rn.bf16x2.f32 %r590, %r4681, %r4680; + mov.b64 {%r4682, %r4683}, %rd872; + cvt.rn.bf16x2.f32 %r591, %r4683, %r4682; + mov.b64 {%r4684, %r4685}, %rd875; + cvt.rn.bf16x2.f32 %r592, %r4685, %r4684; + mov.b64 {%r4686, %r4687}, %rd878; + cvt.rn.bf16x2.f32 %r593, %r4687, %r4686; + mov.b64 {%r4688, %r4689}, %rd881; + cvt.rn.bf16x2.f32 %r594, %r4689, %r4688; + mov.b64 {%r4690, %r4691}, %rd884; + cvt.rn.bf16x2.f32 %r595, %r4691, %r4690; + mov.b64 {%r4692, %r4693}, %rd887; + cvt.rn.bf16x2.f32 %r596, %r4693, %r4692; + mov.b64 {%r4694, %r4695}, %rd890; + cvt.rn.bf16x2.f32 %r597, %r4695, %r4694; + mov.b64 {%r4696, %r4697}, %rd893; + cvt.rn.bf16x2.f32 %r598, %r4697, %r4696; + mov.b64 {%r4698, %r4699}, %rd896; + cvt.rn.bf16x2.f32 %r599, %r4699, %r4698; + mov.b64 {%r4700, %r4701}, %rd899; + cvt.rn.bf16x2.f32 %r600, %r4701, %r4700; + mov.b64 {%r4702, %r4703}, %rd902; + cvt.rn.bf16x2.f32 %r601, %r4703, %r4702; + mov.b64 {%r4704, %r4705}, %rd905; + cvt.rn.bf16x2.f32 %r602, %r4705, %r4704; + mov.b64 {%r4706, %r4707}, %rd908; + cvt.rn.bf16x2.f32 %r603, %r4707, %r4706; + mov.b64 {%r4708, %r4709}, %rd911; + cvt.rn.bf16x2.f32 %r604, %r4709, %r4708; + mov.b64 {%r4710, %r4711}, %rd914; + cvt.rn.bf16x2.f32 %r605, %r4711, %r4710; + mov.b64 {%r4712, %r4713}, %rd917; + cvt.rn.bf16x2.f32 %r606, %r4713, %r4712; + mov.b64 {%r4714, %r4715}, %rd920; + cvt.rn.bf16x2.f32 %r607, %r4715, %r4714; + mov.b64 {%r4716, %r4717}, %rd923; + cvt.rn.bf16x2.f32 %r608, %r4717, %r4716; + mov.b64 {%r4718, %r4719}, %rd926; + cvt.rn.bf16x2.f32 %r609, %r4719, %r4718; + mov.b64 {%r4720, %r4721}, %rd929; + cvt.rn.bf16x2.f32 %r610, %r4721, %r4720; + mov.b64 {%r4722, %r4723}, %rd932; + cvt.rn.bf16x2.f32 %r611, %r4723, %r4722; + mov.b64 {%r4724, %r4725}, %rd935; + cvt.rn.bf16x2.f32 %r612, %r4725, %r4724; + mov.b64 {%r4726, %r4727}, %rd938; + cvt.rn.bf16x2.f32 %r613, %r4727, %r4726; + mov.b64 {%r4728, %r4729}, %rd941; + cvt.rn.bf16x2.f32 %r614, %r4729, %r4728; + mov.b64 {%r4730, %r4731}, %rd944; + cvt.rn.bf16x2.f32 %r615, %r4731, %r4730; + mov.b64 {%r4732, %r4733}, %rd947; + cvt.rn.bf16x2.f32 %r616, %r4733, %r4732; + mov.b64 {%r4734, %r4735}, %rd950; + cvt.rn.bf16x2.f32 %r617, %r4735, %r4734; + mov.b64 {%r4736, %r4737}, %rd953; + cvt.rn.bf16x2.f32 %r618, %r4737, %r4736; + mov.b64 {%r4738, %r4739}, %rd956; + cvt.rn.bf16x2.f32 %r619, %r4739, %r4738; + mov.b64 {%r4740, %r4741}, %rd959; + cvt.rn.bf16x2.f32 %r620, %r4741, %r4740; + mov.b64 {%r4742, %r4743}, %rd962; + cvt.rn.bf16x2.f32 %r621, %r4743, %r4742; + mov.b64 {%r4744, %r4745}, %rd965; + cvt.rn.bf16x2.f32 %r622, %r4745, %r4744; + mov.b64 {%r4746, %r4747}, %rd968; + cvt.rn.bf16x2.f32 %r623, %r4747, %r4746; + mov.b64 {%r4748, %r4749}, %rd971; + cvt.rn.bf16x2.f32 %r624, %r4749, %r4748; + mov.b64 {%r4750, %r4751}, %rd974; + cvt.rn.bf16x2.f32 %r625, %r4751, %r4750; + mov.b64 {%r4752, %r4753}, %rd977; + cvt.rn.bf16x2.f32 %r626, %r4753, %r4752; + mov.b64 {%r4754, %r4755}, %rd980; + cvt.rn.bf16x2.f32 %r627, %r4755, %r4754; + mov.b64 {%r4756, %r4757}, %rd983; + cvt.rn.bf16x2.f32 %r628, %r4757, %r4756; + mov.b64 {%r4758, %r4759}, %rd986; + cvt.rn.bf16x2.f32 %r629, %r4759, %r4758; + mov.b64 {%r4760, %r4761}, %rd989; + cvt.rn.bf16x2.f32 %r630, %r4761, %r4760; + mov.b64 {%r4762, %r4763}, %rd992; + cvt.rn.bf16x2.f32 %r631, %r4763, %r4762; + mov.b64 {%r4764, %r4765}, %rd995; + cvt.rn.bf16x2.f32 %r632, %r4765, %r4764; + mov.b64 {%r4766, %r4767}, %rd998; + cvt.rn.bf16x2.f32 %r633, %r4767, %r4766; + mov.b64 {%r4768, %r4769}, %rd1001; + cvt.rn.bf16x2.f32 %r634, %r4769, %r4768; + mov.b64 {%r4770, %r4771}, %rd1004; + cvt.rn.bf16x2.f32 %r635, %r4771, %r4770; + mov.b64 {%r4772, %r4773}, %rd1007; + cvt.rn.bf16x2.f32 %r636, %r4773, %r4772; + mov.b64 {%r4774, %r4775}, %rd1010; + cvt.rn.bf16x2.f32 %r637, %r4775, %r4774; + mov.b64 {%r4776, %r4777}, %rd1013; + cvt.rn.bf16x2.f32 %r638, %r4777, %r4776; + mov.b64 {%r4778, %r4779}, %rd1016; + cvt.rn.bf16x2.f32 %r639, %r4779, %r4778; + mov.b64 {%r4780, %r4781}, %rd1019; + cvt.rn.bf16x2.f32 %r640, %r4781, %r4780; + mov.b64 {%r4782, %r4783}, %rd1022; + cvt.rn.bf16x2.f32 %r641, %r4783, %r4782; + mov.b64 {%r4784, %r4785}, %rd1025; + cvt.rn.bf16x2.f32 %r642, %r4785, %r4784; + mov.b64 {%r4786, %r4787}, %rd1028; + cvt.rn.bf16x2.f32 %r643, %r4787, %r4786; + mov.b64 {%r4788, %r4789}, %rd1031; + cvt.rn.bf16x2.f32 %r644, %r4789, %r4788; + mov.b64 {%r4790, %r4791}, %rd1034; + cvt.rn.bf16x2.f32 %r645, %r4791, %r4790; + mov.b64 {%r4792, %r4793}, %rd1037; + cvt.rn.bf16x2.f32 %r646, %r4793, %r4792; + mov.b64 {%r4794, %r4795}, %rd1040; + cvt.rn.bf16x2.f32 %r647, %r4795, %r4794; + mov.b64 {%r4796, %r4797}, %rd1043; + cvt.rn.bf16x2.f32 %r648, %r4797, %r4796; + mov.b64 {%r4798, %r4799}, %rd1046; + cvt.rn.bf16x2.f32 %r649, %r4799, %r4798; + mov.b64 {%r4800, %r4801}, %rd1049; + cvt.rn.bf16x2.f32 %r650, %r4801, %r4800; + mov.b64 {%r4802, %r4803}, %rd1052; + cvt.rn.bf16x2.f32 %r651, %r4803, %r4802; + mov.b64 {%r4804, %r4805}, %rd1055; + cvt.rn.bf16x2.f32 %r652, %r4805, %r4804; + mov.b64 {%r4806, %r4807}, %rd1058; + cvt.rn.bf16x2.f32 %r653, %r4807, %r4806; + mov.b64 {%r4808, %r4809}, %rd1061; + cvt.rn.bf16x2.f32 %r654, %r4809, %r4808; + mov.b64 {%r4810, %r4811}, %rd1064; + cvt.rn.bf16x2.f32 %r655, %r4811, %r4810; + mov.b64 {%r4812, %r4813}, %rd1067; + cvt.rn.bf16x2.f32 %r656, %r4813, %r4812; + mov.b64 {%r4814, %r4815}, %rd1070; + cvt.rn.bf16x2.f32 %r657, %r4815, %r4814; + mov.b64 {%r4816, %r4817}, %rd1073; + cvt.rn.bf16x2.f32 %r658, %r4817, %r4816; + mov.b64 {%r4818, %r4819}, %rd1076; + cvt.rn.bf16x2.f32 %r659, %r4819, %r4818; + mov.b64 {%r4820, %r4821}, %rd1079; + cvt.rn.bf16x2.f32 %r660, %r4821, %r4820; + mov.b64 {%r4822, %r4823}, %rd1082; + cvt.rn.bf16x2.f32 %r661, %r4823, %r4822; + mov.b64 {%r4824, %r4825}, %rd1085; + cvt.rn.bf16x2.f32 %r662, %r4825, %r4824; + mov.b64 {%r4826, %r4827}, %rd1088; + cvt.rn.bf16x2.f32 %r663, %r4827, %r4826; + mov.b64 {%r4828, %r4829}, %rd1091; + cvt.rn.bf16x2.f32 %r664, %r4829, %r4828; + mov.b64 {%r4830, %r4831}, %rd1094; + cvt.rn.bf16x2.f32 %r665, %r4831, %r4830; + mov.b64 {%r4832, %r4833}, %rd1097; + cvt.rn.bf16x2.f32 %r666, %r4833, %r4832; + mov.b64 {%r4834, %r4835}, %rd1100; + cvt.rn.bf16x2.f32 %r667, %r4835, %r4834; + mov.b64 {%r4836, %r4837}, %rd1103; + cvt.rn.bf16x2.f32 %r668, %r4837, %r4836; + mov.b64 {%r4838, %r4839}, %rd1106; + cvt.rn.bf16x2.f32 %r669, %r4839, %r4838; + mov.b64 {%r4840, %r4841}, %rd1109; + cvt.rn.bf16x2.f32 %r670, %r4841, %r4840; + mov.b64 {%r4842, %r4843}, %rd1112; + cvt.rn.bf16x2.f32 %r671, %r4843, %r4842; + mov.b64 {%r4844, %r4845}, %rd1115; + cvt.rn.bf16x2.f32 %r672, %r4845, %r4844; + mov.b64 {%r4846, %r4847}, %rd1118; + cvt.rn.bf16x2.f32 %r673, %r4847, %r4846; + mov.b64 {%r4848, %r4849}, %rd1121; + cvt.rn.bf16x2.f32 %r674, %r4849, %r4848; + mov.b64 {%r4850, %r4851}, %rd1124; + cvt.rn.bf16x2.f32 %r675, %r4851, %r4850; + mov.b64 {%r4852, %r4853}, %rd1127; + cvt.rn.bf16x2.f32 %r676, %r4853, %r4852; + mov.b64 {%r4854, %r4855}, %rd1130; + cvt.rn.bf16x2.f32 %r677, %r4855, %r4854; + mov.b64 {%r4856, %r4857}, %rd1133; + cvt.rn.bf16x2.f32 %r678, %r4857, %r4856; + mov.b64 {%r4858, %r4859}, %rd1136; + cvt.rn.bf16x2.f32 %r679, %r4859, %r4858; + mov.b64 {%r4860, %r4861}, %rd1139; + cvt.rn.bf16x2.f32 %r680, %r4861, %r4860; + mov.b64 {%r4862, %r4863}, %rd1142; + cvt.rn.bf16x2.f32 %r681, %r4863, %r4862; + mov.b64 {%r4864, %r4865}, %rd1145; + cvt.rn.bf16x2.f32 %r682, %r4865, %r4864; + mov.b64 {%r4866, %r4867}, %rd1148; + cvt.rn.bf16x2.f32 %r683, %r4867, %r4866; + mov.b64 {%r4868, %r4869}, %rd1151; + cvt.rn.bf16x2.f32 %r684, %r4869, %r4868; + mov.b64 {%r4870, %r4871}, %rd1154; + cvt.rn.bf16x2.f32 %r685, %r4871, %r4870; + mov.b64 {%r4872, %r4873}, %rd1157; + cvt.rn.bf16x2.f32 %r686, %r4873, %r4872; + mov.b64 {%r4874, %r4875}, %rd1160; + cvt.rn.bf16x2.f32 %r687, %r4875, %r4874; + mov.b64 {%r4876, %r4877}, %rd1163; + cvt.rn.bf16x2.f32 %r688, %r4877, %r4876; + mov.b64 {%r4878, %r4879}, %rd1166; + cvt.rn.bf16x2.f32 %r689, %r4879, %r4878; + mov.b64 {%r4880, %r4881}, %rd1169; + cvt.rn.bf16x2.f32 %r690, %r4881, %r4880; + mov.b64 {%r4882, %r4883}, %rd1172; + cvt.rn.bf16x2.f32 %r691, %r4883, %r4882; + mov.b64 {%r4884, %r4885}, %rd1175; + cvt.rn.bf16x2.f32 %r692, %r4885, %r4884; + mov.b64 {%r4886, %r4887}, %rd1178; + cvt.rn.bf16x2.f32 %r693, %r4887, %r4886; + mov.b64 {%r4888, %r4889}, %rd1181; + cvt.rn.bf16x2.f32 %r694, %r4889, %r4888; + mov.b64 {%r4890, %r4891}, %rd1184; + cvt.rn.bf16x2.f32 %r695, %r4891, %r4890; + mov.b64 {%r4892, %r4893}, %rd1187; + cvt.rn.bf16x2.f32 %r696, %r4893, %r4892; + mov.b64 {%r4894, %r4895}, %rd1190; + cvt.rn.bf16x2.f32 %r697, %r4895, %r4894; + mov.b64 {%r4896, %r4897}, %rd1193; + cvt.rn.bf16x2.f32 %r698, %r4897, %r4896; + mov.b64 {%r4898, %r4899}, %rd1196; + cvt.rn.bf16x2.f32 %r699, %r4899, %r4898; + mov.b64 {%r4900, %r4901}, %rd1199; + cvt.rn.bf16x2.f32 %r700, %r4901, %r4900; + mov.b64 {%r4902, %r4903}, %rd1202; + cvt.rn.bf16x2.f32 %r701, %r4903, %r4902; + mov.b64 {%r4904, %r4905}, %rd1205; + cvt.rn.bf16x2.f32 %r702, %r4905, %r4904; + mov.b64 {%r4906, %r4907}, %rd1208; + cvt.rn.bf16x2.f32 %r703, %r4907, %r4906; + mov.b64 {%r4908, %r4909}, %rd1211; + cvt.rn.bf16x2.f32 %r704, %r4909, %r4908; + mov.b64 {%r4910, %r4911}, %rd1214; + cvt.rn.bf16x2.f32 %r705, %r4911, %r4910; + mov.b64 {%r4912, %r4913}, %rd1217; + cvt.rn.bf16x2.f32 %r706, %r4913, %r4912; + mov.b64 {%r4914, %r4915}, %rd1220; + cvt.rn.bf16x2.f32 %r707, %r4915, %r4914; + mov.b64 {%r4916, %r4917}, %rd1223; + cvt.rn.bf16x2.f32 %r708, %r4917, %r4916; + mov.b64 {%r4918, %r4919}, %rd1226; + cvt.rn.bf16x2.f32 %r709, %r4919, %r4918; + mov.b64 {%r4920, %r4921}, %rd1229; + cvt.rn.bf16x2.f32 %r710, %r4921, %r4920; + mov.b64 {%r4922, %r4923}, %rd1232; + cvt.rn.bf16x2.f32 %r711, %r4923, %r4922; + mov.b64 {%r4924, %r4925}, %rd1235; + cvt.rn.bf16x2.f32 %r712, %r4925, %r4924; + mov.b64 {%r4926, %r4927}, %rd1238; + cvt.rn.bf16x2.f32 %r713, %r4927, %r4926; + mov.b64 {%r4928, %r4929}, %rd1241; + cvt.rn.bf16x2.f32 %r714, %r4929, %r4928; + mov.b64 {%r4930, %r4931}, %rd1244; + cvt.rn.bf16x2.f32 %r715, %r4931, %r4930; + mov.b64 {%r4932, %r4933}, %rd1247; + cvt.rn.bf16x2.f32 %r716, %r4933, %r4932; + mov.b64 {%r4934, %r4935}, %rd1250; + cvt.rn.bf16x2.f32 %r717, %r4935, %r4934; + mov.b64 {%r4936, %r4937}, %rd1253; + cvt.rn.bf16x2.f32 %r718, %r4937, %r4936; + mov.b64 {%r4938, %r4939}, %rd1256; + cvt.rn.bf16x2.f32 %r719, %r4939, %r4938; + mov.b64 {%r4940, %r4941}, %rd1259; + cvt.rn.bf16x2.f32 %r720, %r4941, %r4940; + mov.b64 {%r4942, %r4943}, %rd1262; + cvt.rn.bf16x2.f32 %r721, %r4943, %r4942; + mov.b64 {%r4944, %r4945}, %rd1265; + cvt.rn.bf16x2.f32 %r722, %r4945, %r4944; + mov.b64 {%r4946, %r4947}, %rd1268; + cvt.rn.bf16x2.f32 %r723, %r4947, %r4946; + mov.b64 {%r4948, %r4949}, %rd1271; + cvt.rn.bf16x2.f32 %r724, %r4949, %r4948; + mov.b64 {%r4950, %r4951}, %rd1274; + cvt.rn.bf16x2.f32 %r725, %r4951, %r4950; + mov.b64 {%r4952, %r4953}, %rd1277; + cvt.rn.bf16x2.f32 %r726, %r4953, %r4952; + mov.b64 {%r4954, %r4955}, %rd1280; + cvt.rn.bf16x2.f32 %r727, %r4955, %r4954; + mov.b64 {%r4956, %r4957}, %rd1283; + cvt.rn.bf16x2.f32 %r728, %r4957, %r4956; + mov.b64 {%r4958, %r4959}, %rd1286; + cvt.rn.bf16x2.f32 %r729, %r4959, %r4958; + mov.b64 {%r4960, %r4961}, %rd1289; + cvt.rn.bf16x2.f32 %r730, %r4961, %r4960; + mov.b64 {%r4962, %r4963}, %rd1292; + cvt.rn.bf16x2.f32 %r731, %r4963, %r4962; + mov.b64 {%r4964, %r4965}, %rd1295; + cvt.rn.bf16x2.f32 %r732, %r4965, %r4964; + mov.b64 {%r4966, %r4967}, %rd1298; + cvt.rn.bf16x2.f32 %r733, %r4967, %r4966; + mov.b64 {%r4968, %r4969}, %rd1301; + cvt.rn.bf16x2.f32 %r734, %r4969, %r4968; + mov.b64 {%r4970, %r4971}, %rd1304; + cvt.rn.bf16x2.f32 %r735, %r4971, %r4970; + mov.b64 {%r4972, %r4973}, %rd1307; + cvt.rn.bf16x2.f32 %r736, %r4973, %r4972; + mov.b64 {%r4974, %r4975}, %rd1310; + cvt.rn.bf16x2.f32 %r737, %r4975, %r4974; + mov.b64 {%r4976, %r4977}, %rd1313; + cvt.rn.bf16x2.f32 %r738, %r4977, %r4976; + mov.b64 {%r4978, %r4979}, %rd1316; + cvt.rn.bf16x2.f32 %r739, %r4979, %r4978; + mov.b64 {%r4980, %r4981}, %rd1319; + cvt.rn.bf16x2.f32 %r740, %r4981, %r4980; + mov.b64 {%r4982, %r4983}, %rd1322; + cvt.rn.bf16x2.f32 %r741, %r4983, %r4982; + mov.b64 {%r4984, %r4985}, %rd1325; + cvt.rn.bf16x2.f32 %r742, %r4985, %r4984; + mov.b64 {%r4986, %r4987}, %rd1328; + cvt.rn.bf16x2.f32 %r743, %r4987, %r4986; + mov.b64 {%r4988, %r4989}, %rd1331; + cvt.rn.bf16x2.f32 %r744, %r4989, %r4988; + mov.b64 {%r4990, %r4991}, %rd1334; + cvt.rn.bf16x2.f32 %r745, %r4991, %r4990; + mov.b64 {%r4992, %r4993}, %rd1337; + cvt.rn.bf16x2.f32 %r746, %r4993, %r4992; + mov.b64 {%r4994, %r4995}, %rd1340; + cvt.rn.bf16x2.f32 %r747, %r4995, %r4994; + mov.b64 {%r4996, %r4997}, %rd1343; + cvt.rn.bf16x2.f32 %r748, %r4997, %r4996; + mov.b64 {%r4998, %r4999}, %rd1346; + cvt.rn.bf16x2.f32 %r749, %r4999, %r4998; + mov.b64 {%r5000, %r5001}, %rd1349; + cvt.rn.bf16x2.f32 %r750, %r5001, %r5000; + mov.b64 {%r5002, %r5003}, %rd1352; + cvt.rn.bf16x2.f32 %r751, %r5003, %r5002; + mov.b64 {%r5004, %r5005}, %rd1355; + cvt.rn.bf16x2.f32 %r752, %r5005, %r5004; + mov.b64 {%r5006, %r5007}, %rd1358; + cvt.rn.bf16x2.f32 %r753, %r5007, %r5006; + mov.b64 {%r5008, %r5009}, %rd1361; + cvt.rn.bf16x2.f32 %r754, %r5009, %r5008; + mov.b64 {%r5010, %r5011}, %rd1364; + cvt.rn.bf16x2.f32 %r755, %r5011, %r5010; + mov.b64 {%r5012, %r5013}, %rd1367; + cvt.rn.bf16x2.f32 %r756, %r5013, %r5012; + mov.b64 {%r5014, %r5015}, %rd1370; + cvt.rn.bf16x2.f32 %r757, %r5015, %r5014; + mov.b64 {%r5016, %r5017}, %rd1373; + cvt.rn.bf16x2.f32 %r758, %r5017, %r5016; + mov.b64 {%r5018, %r5019}, %rd1376; + cvt.rn.bf16x2.f32 %r759, %r5019, %r5018; + mov.b64 {%r5020, %r5021}, %rd1379; + cvt.rn.bf16x2.f32 %r760, %r5021, %r5020; + mov.b64 {%r5022, %r5023}, %rd1382; + cvt.rn.bf16x2.f32 %r761, %r5023, %r5022; + mov.b64 {%r5024, %r5025}, %rd1385; + cvt.rn.bf16x2.f32 %r762, %r5025, %r5024; + mov.b64 {%r5026, %r5027}, %rd1388; + cvt.rn.bf16x2.f32 %r763, %r5027, %r5026; + mov.b64 {%r5028, %r5029}, %rd1391; + cvt.rn.bf16x2.f32 %r764, %r5029, %r5028; + mov.b64 {%r5030, %r5031}, %rd1394; + cvt.rn.bf16x2.f32 %r765, %r5031, %r5030; + mov.b64 {%r5032, %r5033}, %rd1397; + cvt.rn.bf16x2.f32 %r766, %r5033, %r5032; + mov.b64 {%r5034, %r5035}, %rd1400; + cvt.rn.bf16x2.f32 %r767, %r5035, %r5034; + mov.b64 {%r5036, %r5037}, %rd1403; + cvt.rn.bf16x2.f32 %r768, %r5037, %r5036; + mov.b64 {%r5038, %r5039}, %rd1406; + cvt.rn.bf16x2.f32 %r769, %r5039, %r5038; + mov.b64 {%r5040, %r5041}, %rd1409; + cvt.rn.bf16x2.f32 %r770, %r5041, %r5040; + mov.b64 {%r5042, %r5043}, %rd1412; + cvt.rn.bf16x2.f32 %r771, %r5043, %r5042; + mov.b64 {%r5044, %r5045}, %rd1415; + cvt.rn.bf16x2.f32 %r772, %r5045, %r5044; + mov.b64 {%r5046, %r5047}, %rd1418; + cvt.rn.bf16x2.f32 %r773, %r5047, %r5046; + .loc 1 108 19 // l2norm.py:108:19 + cvt.rn.bf16x2.f32 %r518, %r4531, %r4530; + cvt.rn.bf16x2.f32 %r519, %r4535, %r4534; + cvt.rn.bf16x2.f32 %r520, %r4539, %r4538; + cvt.rn.bf16x2.f32 %r521, %r4543, %r4542; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd133 + 0 ], { %r518, %r519, %r520, %r521 }; + // end inline asm + // begin inline asm + @%p2 st.global.v4.b32 [ %rd134 + 0 ], { %r522, %r523, %r524, %r525 }; + // end inline asm + // begin inline asm + @%p3 st.global.v4.b32 [ %rd135 + 0 ], { %r526, %r527, %r528, %r529 }; + // end inline asm + // begin inline asm + @%p4 st.global.v4.b32 [ %rd136 + 0 ], { %r530, %r531, %r532, %r533 }; + // end inline asm + // begin inline asm + @%p5 st.global.v4.b32 [ %rd137 + 0 ], { %r534, %r535, %r536, %r537 }; + // end inline asm + // begin inline asm + @%p6 st.global.v4.b32 [ %rd138 + 0 ], { %r538, %r539, %r540, %r541 }; + // end inline asm + // begin inline asm + @%p7 st.global.v4.b32 [ %rd139 + 0 ], { %r542, %r543, %r544, %r545 }; + // end inline asm + // begin inline asm + @%p8 st.global.v4.b32 [ %rd140 + 0 ], { %r546, %r547, %r548, %r549 }; + // end inline asm + // begin inline asm + @%p9 st.global.v4.b32 [ %rd141 + 0 ], { %r550, %r551, %r552, %r553 }; + // end inline asm + // begin inline asm + @%p10 st.global.v4.b32 [ %rd142 + 0 ], { %r554, %r555, %r556, %r557 }; + // end inline asm + // begin inline asm + @%p11 st.global.v4.b32 [ %rd143 + 0 ], { %r558, %r559, %r560, %r561 }; + // end inline asm + // begin inline asm + @%p39 st.global.v4.b32 [ %rd144 + 0 ], { %r562, %r563, %r564, %r565 }; + // end inline asm + // begin inline asm + @%p12 st.global.v4.b32 [ %rd145 + 0 ], { %r566, %r567, %r568, %r569 }; + // end inline asm + // begin inline asm + @%p13 st.global.v4.b32 [ %rd146 + 0 ], { %r570, %r571, %r572, %r573 }; + // end inline asm + // begin inline asm + @%p14 st.global.v4.b32 [ %rd147 + 0 ], { %r574, %r575, %r576, %r577 }; + // end inline asm + // begin inline asm + @%p15 st.global.v4.b32 [ %rd148 + 0 ], { %r578, %r579, %r580, %r581 }; + // end inline asm + // begin inline asm + @%p16 st.global.v4.b32 [ %rd149 + 0 ], { %r582, %r583, %r584, %r585 }; + // end inline asm + // begin inline asm + @%p17 st.global.v4.b32 [ %rd150 + 0 ], { %r586, %r587, %r588, %r589 }; + // end inline asm + // begin inline asm + @%p18 st.global.v4.b32 [ %rd151 + 0 ], { %r590, %r591, %r592, %r593 }; + // end inline asm + // begin inline asm + @%p40 st.global.v4.b32 [ %rd152 + 0 ], { %r594, %r595, %r596, %r597 }; + // end inline asm + // begin inline asm + @%p19 st.global.v4.b32 [ %rd153 + 0 ], { %r598, %r599, %r600, %r601 }; + // end inline asm + // begin inline asm + @%p20 st.global.v4.b32 [ %rd154 + 0 ], { %r602, %r603, %r604, %r605 }; + // end inline asm + // begin inline asm + @%p21 st.global.v4.b32 [ %rd155 + 0 ], { %r606, %r607, %r608, %r609 }; + // end inline asm + // begin inline asm + @%p22 st.global.v4.b32 [ %rd156 + 0 ], { %r610, %r611, %r612, %r613 }; + // end inline asm + // begin inline asm + @%p23 st.global.v4.b32 [ %rd157 + 0 ], { %r614, %r615, %r616, %r617 }; + // end inline asm + // begin inline asm + @%p24 st.global.v4.b32 [ %rd158 + 0 ], { %r618, %r619, %r620, %r621 }; + // end inline asm + // begin inline asm + @%p25 st.global.v4.b32 [ %rd159 + 0 ], { %r622, %r623, %r624, %r625 }; + // end inline asm + // begin inline asm + @%p26 st.global.v4.b32 [ %rd160 + 0 ], { %r626, %r627, %r628, %r629 }; + // end inline asm + // begin inline asm + @%p27 st.global.v4.b32 [ %rd161 + 0 ], { %r630, %r631, %r632, %r633 }; + // end inline asm + // begin inline asm + @%p28 st.global.v4.b32 [ %rd162 + 0 ], { %r634, %r635, %r636, %r637 }; + // end inline asm + // begin inline asm + @%p29 st.global.v4.b32 [ %rd163 + 0 ], { %r638, %r639, %r640, %r641 }; + // end inline asm + // begin inline asm + @%p30 st.global.v4.b32 [ %rd164 + 0 ], { %r642, %r643, %r644, %r645 }; + // end inline asm + // begin inline asm + @%p31 st.global.v4.b32 [ %rd165 + 0 ], { %r646, %r647, %r648, %r649 }; + // end inline asm + // begin inline asm + @%p32 st.global.v4.b32 [ %rd166 + 0 ], { %r650, %r651, %r652, %r653 }; + // end inline asm + // begin inline asm + @%p33 st.global.v4.b32 [ %rd167 + 0 ], { %r654, %r655, %r656, %r657 }; + // end inline asm + // begin inline asm + @%p34 st.global.v4.b32 [ %rd168 + 0 ], { %r658, %r659, %r660, %r661 }; + // end inline asm + // begin inline asm + @%p41 st.global.v4.b32 [ %rd169 + 0 ], { %r662, %r663, %r664, %r665 }; + // end inline asm + // begin inline asm + @%p42 st.global.v4.b32 [ %rd170 + 0 ], { %r666, %r667, %r668, %r669 }; + // end inline asm + // begin inline asm + @%p43 st.global.v4.b32 [ %rd171 + 0 ], { %r670, %r671, %r672, %r673 }; + // end inline asm + // begin inline asm + @%p44 st.global.v4.b32 [ %rd172 + 0 ], { %r674, %r675, %r676, %r677 }; + // end inline asm + // begin inline asm + @%p45 st.global.v4.b32 [ %rd173 + 0 ], { %r678, %r679, %r680, %r681 }; + // end inline asm + // begin inline asm + @%p46 st.global.v4.b32 [ %rd174 + 0 ], { %r682, %r683, %r684, %r685 }; + // end inline asm + // begin inline asm + @%p47 st.global.v4.b32 [ %rd175 + 0 ], { %r686, %r687, %r688, %r689 }; + // end inline asm + // begin inline asm + @%p48 st.global.v4.b32 [ %rd176 + 0 ], { %r690, %r691, %r692, %r693 }; + // end inline asm + // begin inline asm + @%p49 st.global.v4.b32 [ %rd177 + 0 ], { %r694, %r695, %r696, %r697 }; + // end inline asm + // begin inline asm + @%p50 st.global.v4.b32 [ %rd178 + 0 ], { %r698, %r699, %r700, %r701 }; + // end inline asm + // begin inline asm + @%p51 st.global.v4.b32 [ %rd179 + 0 ], { %r702, %r703, %r704, %r705 }; + // end inline asm + // begin inline asm + @%p52 st.global.v4.b32 [ %rd180 + 0 ], { %r706, %r707, %r708, %r709 }; + // end inline asm + // begin inline asm + @%p53 st.global.v4.b32 [ %rd181 + 0 ], { %r710, %r711, %r712, %r713 }; + // end inline asm + // begin inline asm + @%p54 st.global.v4.b32 [ %rd182 + 0 ], { %r714, %r715, %r716, %r717 }; + // end inline asm + // begin inline asm + @%p55 st.global.v4.b32 [ %rd183 + 0 ], { %r718, %r719, %r720, %r721 }; + // end inline asm + // begin inline asm + @%p56 st.global.v4.b32 [ %rd184 + 0 ], { %r722, %r723, %r724, %r725 }; + // end inline asm + // begin inline asm + @%p57 st.global.v4.b32 [ %rd185 + 0 ], { %r726, %r727, %r728, %r729 }; + // end inline asm + // begin inline asm + @%p58 st.global.v4.b32 [ %rd186 + 0 ], { %r730, %r731, %r732, %r733 }; + // end inline asm + // begin inline asm + @%p59 st.global.v4.b32 [ %rd187 + 0 ], { %r734, %r735, %r736, %r737 }; + // end inline asm + // begin inline asm + @%p60 st.global.v4.b32 [ %rd188 + 0 ], { %r738, %r739, %r740, %r741 }; + // end inline asm + // begin inline asm + @%p61 st.global.v4.b32 [ %rd189 + 0 ], { %r742, %r743, %r744, %r745 }; + // end inline asm + // begin inline asm + @%p62 st.global.v4.b32 [ %rd190 + 0 ], { %r746, %r747, %r748, %r749 }; + // end inline asm + // begin inline asm + @%p63 st.global.v4.b32 [ %rd191 + 0 ], { %r750, %r751, %r752, %r753 }; + // end inline asm + // begin inline asm + @%p64 st.global.v4.b32 [ %rd192 + 0 ], { %r754, %r755, %r756, %r757 }; + // end inline asm + // begin inline asm + @%p65 st.global.v4.b32 [ %rd193 + 0 ], { %r758, %r759, %r760, %r761 }; + // end inline asm + // begin inline asm + @%p66 st.global.v4.b32 [ %rd194 + 0 ], { %r762, %r763, %r764, %r765 }; + // end inline asm + // begin inline asm + @%p67 st.global.v4.b32 [ %rd195 + 0 ], { %r766, %r767, %r768, %r769 }; + // end inline asm + // begin inline asm + @%p68 st.global.v4.b32 [ %rd196 + 0 ], { %r770, %r771, %r772, %r773 }; + // end inline asm + .loc 1 108 4 // l2norm.py:108:4 + ret; +$L__tmp1079: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 98 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp1078 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 107 // DW_AT_call_line +.b8 55 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp1078 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.source b/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..35265513bc03aa2afd3347f28178370dc92e754e --- /dev/null +++ b/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.source @@ -0,0 +1,210 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc47 = loc(unknown) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc54 = loc("y"(#loc)) +#loc55 = loc("rstd"(#loc)) +#loc56 = loc("dy"(#loc)) +#loc57 = loc("dx"(#loc)) +#loc58 = loc("eps"(#loc)) +#loc59 = loc("T"(#loc)) +#loc101 = loc("input"(#loc45)) +#loc102 = loc("a"(#loc50)) +#loc103 = loc("b"(#loc50)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc60) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc61) + %o_t = arith.constant 128 : i32 loc(#loc62) + %o_t_1 = arith.constant 128 : i64 loc(#loc62) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc62) + %o_t_3 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc63) + %o_t_4 = arith.extsi %o_t_3 : tensor<128xi32> to tensor<128xi64> loc(#loc64) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<128xi64> loc(#loc64) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<128xi64> loc(#loc64) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc65) + %m_t = arith.extsi %T : i32 to i64 loc(#loc66) + %m_t_7 = tt.splat %m_t : i64 -> tensor<128xi64> loc(#loc66) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<128xi64> loc(#loc66) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<128xi1> -> tensor<128x1xi1> loc(#loc67) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc68) + %m_x_10 = arith.constant 128 : i32 loc(#loc69) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc69) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc69) + %m_x_13 = tt.broadcast %m_x : tensor<128x1xi1> -> tensor<128x128xi1> loc(#loc70) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<128x128xi1> loc(#loc70) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<128x128xi1> loc(#loc70) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc71) + %p_y_16 = arith.constant 128 : i32 loc(#loc72) + %p_y_17 = arith.constant 128 : i64 loc(#loc72) + %p_y_18 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc72) + %p_y_19 = arith.muli %p_y, %p_y_18 : tensor<128x1xi64> loc(#loc72) + %p_y_20 = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc73) + %p_y_21 = tt.addptr %p_y_20, %p_y_19 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc73) + %p_y_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc74) + %p_y_23 = tt.broadcast %p_y_21 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc75) + %p_y_24 = tt.broadcast %p_y_22 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc75) + %p_y_25 = tt.addptr %p_y_23, %p_y_24 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc75) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr> loc(#loc76) + %p_rstd_26 = tt.addptr %p_rstd, %o_t_6 : tensor<128x!tt.ptr>, tensor<128xi64> loc(#loc76) + %p_dy = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc77) + %p_dy_27 = arith.constant 128 : i32 loc(#loc78) + %p_dy_28 = arith.constant 128 : i64 loc(#loc78) + %p_dy_29 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc78) + %p_dy_30 = arith.muli %p_dy, %p_dy_29 : tensor<128x1xi64> loc(#loc78) + %p_dy_31 = tt.splat %dy : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc79) + %p_dy_32 = tt.addptr %p_dy_31, %p_dy_30 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc79) + %p_dy_33 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc80) + %p_dy_34 = tt.broadcast %p_dy_32 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc81) + %p_dy_35 = tt.broadcast %p_dy_33 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc81) + %p_dy_36 = tt.addptr %p_dy_34, %p_dy_35 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc81) + %p_dx = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc82) + %p_dx_37 = arith.constant 128 : i32 loc(#loc83) + %p_dx_38 = arith.constant 128 : i64 loc(#loc83) + %p_dx_39 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc83) + %p_dx_40 = arith.muli %p_dx, %p_dx_39 : tensor<128x1xi64> loc(#loc83) + %p_dx_41 = tt.splat %dx : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc84) + %p_dx_42 = tt.addptr %p_dx_41, %p_dx_40 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc84) + %p_dx_43 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc85) + %p_dx_44 = tt.broadcast %p_dx_42 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc86) + %p_dx_45 = tt.broadcast %p_dx_43 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc86) + %p_dx_46 = tt.addptr %p_dx_44, %p_dx_45 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc86) + %b_y = arith.constant 0.000000e+00 : f32 loc(#loc87) + %b_y_47 = arith.constant dense<0.000000e+00> : tensor<128x128xf32> loc(#loc87) + %b_y_48 = arith.truncf %b_y_47 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc87) + %b_y_49 = tt.load %p_y_25, %m_x_15, %b_y_48 : tensor<128x128x!tt.ptr> loc(#loc87) + %b_y_50 = arith.extf %b_y_49 : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc88) + %b_rstd = arith.constant 0.000000e+00 : f32 loc(#loc89) + %b_rstd_51 = arith.constant dense<0.000000e+00> : tensor<128xf32> loc(#loc89) + %b_rstd_52 = tt.load %p_rstd_26, %m_t_8, %b_rstd_51 : tensor<128x!tt.ptr> loc(#loc89) + %b_dy = arith.constant 0.000000e+00 : f32 loc(#loc90) + %b_dy_53 = arith.constant dense<0.000000e+00> : tensor<128x128xf32> loc(#loc90) + %b_dy_54 = arith.truncf %b_dy_53 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc90) + %b_dy_55 = tt.load %p_dy_36, %m_x_15, %b_dy_54 : tensor<128x128x!tt.ptr> loc(#loc90) + %b_dy_56 = arith.extf %b_dy_55 : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc91) + %b_dx = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc92) + %b_dx_57 = tt.broadcast %b_dx : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc93) + %b_dx_58 = arith.mulf %b_dy_56, %b_dx_57 : tensor<128x128xf32> loc(#loc93) + %b_dx_59 = arith.mulf %b_dy_56, %b_y_50 : tensor<128x128xf32> loc(#loc94) + %b_dx_60 = tt.call @triton.language.standard.sum__fp32S128_128S_c1_cFalse_cNone(%b_dx_59) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc95) + %b_dx_61 = tt.expand_dims %b_dx_60 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc96) + %b_dx_62 = tt.broadcast %b_dx_61 : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc97) + %b_dx_63 = arith.mulf %b_dx_62, %b_y_50 : tensor<128x128xf32> loc(#loc97) + %b_dx_64 = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc98) + %b_dx_65 = tt.broadcast %b_dx_64 : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc99) + %b_dx_66 = arith.mulf %b_dx_63, %b_dx_65 : tensor<128x128xf32> loc(#loc99) + %b_dx_67 = arith.subf %b_dx_58, %b_dx_66 : tensor<128x128xf32> loc(#loc100) + %0 = arith.truncf %b_dx_67 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc42) + tt.store %p_dx_46, %0, %m_x_15 : tensor<128x128x!tt.ptr> loc(#loc43) + tt.return loc(#loc44) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S128_128S_c1_cFalse_cNone(%input: tensor<128x128xf32> loc("input"(#loc45))) -> tensor<128xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc46) + tt.reduce.return %2 : f32 loc(#loc46) + }) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc46) + tt.return %0 : tensor<128xf32> loc(#loc48) + ^bb1: // no predecessors + %1 = ub.poison : tensor<128xf32> loc(#loc49) + tt.return %1 : tensor<128xf32> loc(#loc49) + } loc(#loc45) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc50)), %b: f32 loc("b"(#loc50))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc51) + tt.return %0 : f32 loc(#loc52) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc53) + tt.return %1 : f32 loc(#loc53) + } loc(#loc50) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:31) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:39) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:20) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:31) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:39) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:82) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc60 = loc("i_t"(#loc1)) +#loc61 = loc("i_t"(#loc2)) +#loc62 = loc("o_t"(#loc3)) +#loc63 = loc("o_t"(#loc4)) +#loc64 = loc("o_t"(#loc5)) +#loc65 = loc("o_d"(#loc6)) +#loc66 = loc("m_t"(#loc7)) +#loc67 = loc("m_x"(#loc8)) +#loc68 = loc("m_x"(#loc9)) +#loc69 = loc("m_x"(#loc10)) +#loc70 = loc("m_x"(#loc11)) +#loc71 = loc("p_y"(#loc12)) +#loc72 = loc("p_y"(#loc13)) +#loc73 = loc("p_y"(#loc14)) +#loc74 = loc("p_y"(#loc15)) +#loc75 = loc("p_y"(#loc16)) +#loc76 = loc("p_rstd"(#loc17)) +#loc77 = loc("p_dy"(#loc18)) +#loc78 = loc("p_dy"(#loc19)) +#loc79 = loc("p_dy"(#loc20)) +#loc80 = loc("p_dy"(#loc21)) +#loc81 = loc("p_dy"(#loc22)) +#loc82 = loc("p_dx"(#loc23)) +#loc83 = loc("p_dx"(#loc24)) +#loc84 = loc("p_dx"(#loc25)) +#loc85 = loc("p_dx"(#loc26)) +#loc86 = loc("p_dx"(#loc27)) +#loc87 = loc("b_y"(#loc28)) +#loc88 = loc("b_y"(#loc29)) +#loc89 = loc("b_rstd"(#loc30)) +#loc90 = loc("b_dy"(#loc31)) +#loc91 = loc("b_dy"(#loc32)) +#loc92 = loc("b_dx"(#loc33)) +#loc93 = loc("b_dx"(#loc34)) +#loc94 = loc("b_dx"(#loc35)) +#loc95 = loc("b_dx"(#loc36)) +#loc96 = loc("b_dx"(#loc37)) +#loc97 = loc("b_dx"(#loc38)) +#loc98 = loc("b_dx"(#loc39)) +#loc99 = loc("b_dx"(#loc40)) +#loc100 = loc("b_dx"(#loc41)) diff --git a/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.ttgir b/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..30b8d0f11ffeced21ac41468aae9ea13197ceffd --- /dev/null +++ b/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.ttgir @@ -0,0 +1,154 @@ +#blocked = #ttg.blocked<{sizePerThread = [4], threadsPerWarp = [32], warpsPerCTA = [1], order = [0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [1, 1], order = [1, 0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc39 = loc("y"(#loc)) +#loc40 = loc("rstd"(#loc)) +#loc41 = loc("dy"(#loc)) +#loc42 = loc("dx"(#loc)) +#loc43 = loc("eps"(#loc)) +#loc44 = loc("T"(#loc)) +#loc72 = loc("b_dx"(#loc30)) +#loc78 = loc(callsite(#loc1 at #loc72)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 1 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %c128_i64 = arith.constant 128 : i64 loc(#loc1) + %cst = arith.constant dense<0.000000e+00> : tensor<128xf32, #blocked> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<128x128xbf16, #blocked1> loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<128x1xi64, #blocked1> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<1x128xi32, #blocked1> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc45) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc46) + %o_t = arith.muli %i_t_3, %c128_i64 : i64 loc(#loc47) + %o_t_4 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc48) + %o_t_5 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #blocked> loc(#loc48) + %o_t_6 = arith.extsi %o_t_4 : tensor<128xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> to tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc49) + %o_t_7 = arith.extsi %o_t_5 : tensor<128xi32, #blocked> to tensor<128xi64, #blocked> loc(#loc49) + %o_t_8 = tt.splat %o_t : i64 -> tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc49) + %o_t_9 = tt.splat %o_t : i64 -> tensor<128xi64, #blocked> loc(#loc49) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc49) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<128xi64, #blocked> loc(#loc49) + %m_t = arith.extsi %T : i32 to i64 loc(#loc50) + %m_t_12 = tt.splat %m_t : i64 -> tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc50) + %m_t_13 = tt.splat %m_t : i64 -> tensor<128xi64, #blocked> loc(#loc50) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc50) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<128xi64, #blocked> loc(#loc50) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<128xi1, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<128x1xi1, #blocked1> loc(#loc51) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc52) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x128xi32, #blocked1> loc(#loc52) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_2 : tensor<1x128xi32, #blocked1> loc(#loc53) + %m_x_19 = tt.broadcast %m_x : tensor<128x1xi1, #blocked1> -> tensor<128x128xi1, #blocked1> loc(#loc54) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked1> -> tensor<128x128xi1, #blocked1> loc(#loc54) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<128x128xi1, #blocked1> loc(#loc54) + %p_y = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<128x1xi64, #blocked1> loc(#loc55) + %p_y_22 = arith.muli %p_y, %cst_1 : tensor<128x1xi64, #blocked1> loc(#loc56) + %p_y_23 = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr, #blocked1> loc(#loc57) + %p_y_24 = tt.addptr %p_y_23, %p_y_22 : tensor<128x1x!tt.ptr, #blocked1>, tensor<128x1xi64, #blocked1> loc(#loc57) + %p_y_25 = tt.broadcast %p_y_24 : tensor<128x1x!tt.ptr, #blocked1> -> tensor<128x128x!tt.ptr, #blocked1> loc(#loc58) + %p_y_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked1> -> tensor<128x128xi32, #blocked1> loc(#loc58) + %p_y_27 = tt.addptr %p_y_25, %p_y_26 : tensor<128x128x!tt.ptr, #blocked1>, tensor<128x128xi32, #blocked1> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr, #blocked> loc(#loc59) + %p_rstd_28 = tt.addptr %p_rstd, %o_t_11 : tensor<128x!tt.ptr, #blocked>, tensor<128xi64, #blocked> loc(#loc59) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<128x1x!tt.ptr, #blocked1> loc(#loc60) + %p_dy_29 = tt.addptr %p_dy, %p_y_22 : tensor<128x1x!tt.ptr, #blocked1>, tensor<128x1xi64, #blocked1> loc(#loc60) + %p_dy_30 = tt.broadcast %p_dy_29 : tensor<128x1x!tt.ptr, #blocked1> -> tensor<128x128x!tt.ptr, #blocked1> loc(#loc61) + %p_dy_31 = tt.addptr %p_dy_30, %p_y_26 : tensor<128x128x!tt.ptr, #blocked1>, tensor<128x128xi32, #blocked1> loc(#loc61) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<128x1x!tt.ptr, #blocked1> loc(#loc62) + %p_dx_32 = tt.addptr %p_dx, %p_y_22 : tensor<128x1x!tt.ptr, #blocked1>, tensor<128x1xi64, #blocked1> loc(#loc62) + %p_dx_33 = tt.broadcast %p_dx_32 : tensor<128x1x!tt.ptr, #blocked1> -> tensor<128x128x!tt.ptr, #blocked1> loc(#loc63) + %p_dx_34 = tt.addptr %p_dx_33, %p_y_26 : tensor<128x128x!tt.ptr, #blocked1>, tensor<128x128xi32, #blocked1> loc(#loc63) + %b_y = tt.load %p_y_27, %m_x_21, %cst_0 : tensor<128x128x!tt.ptr, #blocked1> loc(#loc64) + %b_y_35 = arith.extf %b_y : tensor<128x128xbf16, #blocked1> to tensor<128x128xf32, #blocked1> loc(#loc65) + %b_rstd = tt.load %p_rstd_28, %m_t_15, %cst : tensor<128x!tt.ptr, #blocked> loc(#loc66) + %b_dy = tt.load %p_dy_31, %m_x_21, %cst_0 : tensor<128x128x!tt.ptr, #blocked1> loc(#loc67) + %b_dy_36 = arith.extf %b_dy : tensor<128x128xbf16, #blocked1> to tensor<128x128xf32, #blocked1> loc(#loc68) + %b_dx = ttg.convert_layout %b_rstd : tensor<128xf32, #blocked> -> tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc69) + %b_dx_37 = tt.expand_dims %b_dx {axis = 1 : i32} : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<128x1xf32, #blocked1> loc(#loc69) + %b_dx_38 = tt.broadcast %b_dx_37 : tensor<128x1xf32, #blocked1> -> tensor<128x128xf32, #blocked1> loc(#loc70) + %b_dx_39 = arith.mulf %b_dy_36, %b_dx_38 : tensor<128x128xf32, #blocked1> loc(#loc70) + %b_dx_40 = arith.mulf %b_dy_36, %b_y_35 : tensor<128x128xf32, #blocked1> loc(#loc71) + %b_dx_41 = "tt.reduce"(%b_dx_40) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_47: f32 loc(callsite(#loc1 at #loc72)), %b_dx_48: f32 loc(callsite(#loc1 at #loc72))): + %b_dx_49 = arith.addf %b_dx_47, %b_dx_48 : f32 loc(#loc79) + tt.reduce.return %b_dx_49 : f32 loc(#loc77) + }) : (tensor<128x128xf32, #blocked1>) -> tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc77) + %b_dx_42 = tt.expand_dims %b_dx_41 {axis = 1 : i32} : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<128x1xf32, #blocked1> loc(#loc73) + %b_dx_43 = tt.broadcast %b_dx_42 : tensor<128x1xf32, #blocked1> -> tensor<128x128xf32, #blocked1> loc(#loc74) + %b_dx_44 = arith.mulf %b_dx_43, %b_y_35 : tensor<128x128xf32, #blocked1> loc(#loc74) + %b_dx_45 = arith.mulf %b_dx_44, %b_dx_38 : tensor<128x128xf32, #blocked1> loc(#loc75) + %b_dx_46 = arith.subf %b_dx_39, %b_dx_45 : tensor<128x128xf32, #blocked1> loc(#loc76) + %0 = arith.truncf %b_dx_46 : tensor<128x128xf32, #blocked1> to tensor<128x128xbf16, #blocked1> loc(#loc36) + tt.store %p_dx_34, %0, %m_x_21 : tensor<128x128x!tt.ptr, #blocked1> loc(#loc37) + tt.return loc(#loc38) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc45 = loc("i_t"(#loc2)) +#loc46 = loc("i_t"(#loc3)) +#loc47 = loc("o_t"(#loc4)) +#loc48 = loc("o_t"(#loc5)) +#loc49 = loc("o_t"(#loc6)) +#loc50 = loc("m_t"(#loc7)) +#loc51 = loc("m_x"(#loc8)) +#loc52 = loc("m_x"(#loc9)) +#loc53 = loc("m_x"(#loc10)) +#loc54 = loc("m_x"(#loc11)) +#loc55 = loc("p_y"(#loc12)) +#loc56 = loc("p_y"(#loc13)) +#loc57 = loc("p_y"(#loc14)) +#loc58 = loc("p_y"(#loc15)) +#loc59 = loc("p_rstd"(#loc16)) +#loc60 = loc("p_dy"(#loc17)) +#loc61 = loc("p_dy"(#loc18)) +#loc62 = loc("p_dx"(#loc19)) +#loc63 = loc("p_dx"(#loc20)) +#loc64 = loc("b_y"(#loc21)) +#loc65 = loc("b_y"(#loc22)) +#loc66 = loc("b_rstd"(#loc23)) +#loc67 = loc("b_dy"(#loc24)) +#loc68 = loc("b_dy"(#loc25)) +#loc69 = loc("b_dx"(#loc26)) +#loc70 = loc("b_dx"(#loc27)) +#loc71 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc32)) +#loc74 = loc("b_dx"(#loc33)) +#loc75 = loc("b_dx"(#loc34)) +#loc76 = loc("b_dx"(#loc35)) +#loc77 = loc(callsite(#loc29 at #loc72)) +#loc79 = loc(callsite(#loc31 at #loc77)) diff --git a/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.ttir b/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..f242766466ced3ec80bd72ef6238b6accc66dbe7 --- /dev/null +++ b/triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.ttir @@ -0,0 +1,144 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc39 = loc("y"(#loc)) +#loc40 = loc("rstd"(#loc)) +#loc41 = loc("dy"(#loc)) +#loc42 = loc("dx"(#loc)) +#loc43 = loc("eps"(#loc)) +#loc44 = loc("T"(#loc)) +#loc72 = loc("b_dx"(#loc30)) +#loc78 = loc(callsite(#loc1 at #loc72)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<0.000000e+00> : tensor<128x128xbf16> loc(#loc1) + %b_rstd = arith.constant dense<0.000000e+00> : tensor<128xf32> loc(#loc45) + %cst_0 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc1) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc46) + %c128_i64 = arith.constant 128 : i64 loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc47) + %i_t_1 = arith.extsi %i_t : i32 to i64 loc(#loc48) + %o_t = arith.muli %i_t_1, %c128_i64 : i64 loc(#loc49) + %o_t_2 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc50) + %o_t_3 = arith.extsi %o_t_2 : tensor<128xi32> to tensor<128xi64> loc(#loc51) + %o_t_4 = tt.splat %o_t : i64 -> tensor<128xi64> loc(#loc51) + %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<128xi64> loc(#loc51) + %m_t = arith.extsi %T : i32 to i64 loc(#loc52) + %m_t_6 = tt.splat %m_t : i64 -> tensor<128xi64> loc(#loc52) + %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<128xi64> loc(#loc52) + %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<128xi1> -> tensor<128x1xi1> loc(#loc53) + %m_x_9 = tt.expand_dims %o_t_2 {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc54) + %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc46) + %m_x_11 = tt.broadcast %m_x_8 : tensor<128x1xi1> -> tensor<128x128xi1> loc(#loc55) + %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<128x128xi1> loc(#loc55) + %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<128x128xi1> loc(#loc55) + %p_y = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc56) + %p_y_14 = arith.muli %p_y, %cst_0 : tensor<128x1xi64> loc(#loc57) + %p_y_15 = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc58) + %p_y_16 = tt.addptr %p_y_15, %p_y_14 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc58) + %p_y_17 = tt.broadcast %p_y_16 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc59) + %p_y_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc59) + %p_y_19 = tt.addptr %p_y_17, %p_y_18 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc59) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr> loc(#loc60) + %p_rstd_20 = tt.addptr %p_rstd, %o_t_5 : tensor<128x!tt.ptr>, tensor<128xi64> loc(#loc60) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc61) + %p_dy_21 = tt.addptr %p_dy, %p_y_14 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc61) + %p_dy_22 = tt.broadcast %p_dy_21 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc62) + %p_dy_23 = tt.addptr %p_dy_22, %p_y_18 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc62) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc63) + %p_dx_24 = tt.addptr %p_dx, %p_y_14 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc63) + %p_dx_25 = tt.broadcast %p_dx_24 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc64) + %p_dx_26 = tt.addptr %p_dx_25, %p_y_18 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc64) + %b_y = tt.load %p_y_19, %m_x_13, %cst : tensor<128x128x!tt.ptr> loc(#loc65) + %b_y_27 = arith.extf %b_y : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc66) + %b_rstd_28 = tt.load %p_rstd_20, %m_t_7, %b_rstd : tensor<128x!tt.ptr> loc(#loc45) + %b_dy = tt.load %p_dy_23, %m_x_13, %cst : tensor<128x128x!tt.ptr> loc(#loc67) + %b_dy_29 = arith.extf %b_dy : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc68) + %b_dx = tt.expand_dims %b_rstd_28 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc69) + %b_dx_30 = tt.broadcast %b_dx : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc70) + %b_dx_31 = arith.mulf %b_dy_29, %b_dx_30 : tensor<128x128xf32> loc(#loc70) + %b_dx_32 = arith.mulf %b_dy_29, %b_y_27 : tensor<128x128xf32> loc(#loc71) + %b_dx_33 = "tt.reduce"(%b_dx_32) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_39: f32 loc(callsite(#loc1 at #loc72)), %b_dx_40: f32 loc(callsite(#loc1 at #loc72))): + %b_dx_41 = arith.addf %b_dx_39, %b_dx_40 : f32 loc(#loc79) + tt.reduce.return %b_dx_41 : f32 loc(#loc77) + }) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc77) + %b_dx_34 = tt.expand_dims %b_dx_33 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc73) + %b_dx_35 = tt.broadcast %b_dx_34 : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc74) + %b_dx_36 = arith.mulf %b_dx_35, %b_y_27 : tensor<128x128xf32> loc(#loc74) + %b_dx_37 = arith.mulf %b_dx_36, %b_dx_30 : tensor<128x128xf32> loc(#loc75) + %b_dx_38 = arith.subf %b_dx_31, %b_dx_37 : tensor<128x128xf32> loc(#loc76) + %0 = arith.truncf %b_dx_38 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc36) + tt.store %p_dx_26, %0, %m_x_13 : tensor<128x128x!tt.ptr> loc(#loc37) + tt.return loc(#loc38) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc45 = loc("b_rstd"(#loc2)) +#loc46 = loc("m_x"(#loc3)) +#loc47 = loc("i_t"(#loc4)) +#loc48 = loc("i_t"(#loc5)) +#loc49 = loc("o_t"(#loc6)) +#loc50 = loc("o_t"(#loc7)) +#loc51 = loc("o_t"(#loc8)) +#loc52 = loc("m_t"(#loc9)) +#loc53 = loc("m_x"(#loc10)) +#loc54 = loc("m_x"(#loc11)) +#loc55 = loc("m_x"(#loc12)) +#loc56 = loc("p_y"(#loc13)) +#loc57 = loc("p_y"(#loc14)) +#loc58 = loc("p_y"(#loc15)) +#loc59 = loc("p_y"(#loc16)) +#loc60 = loc("p_rstd"(#loc17)) +#loc61 = loc("p_dy"(#loc18)) +#loc62 = loc("p_dy"(#loc19)) +#loc63 = loc("p_dx"(#loc20)) +#loc64 = loc("p_dx"(#loc21)) +#loc65 = loc("b_y"(#loc22)) +#loc66 = loc("b_y"(#loc23)) +#loc67 = loc("b_dy"(#loc24)) +#loc68 = loc("b_dy"(#loc25)) +#loc69 = loc("b_dx"(#loc26)) +#loc70 = loc("b_dx"(#loc27)) +#loc71 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc32)) +#loc74 = loc("b_dx"(#loc33)) +#loc75 = loc("b_dx"(#loc34)) +#loc76 = loc("b_dx"(#loc35)) +#loc77 = loc(callsite(#loc29 at #loc72)) +#loc79 = loc(callsite(#loc31 at #loc77)) diff --git a/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/__grp__l2norm_bwd_kernel.json b/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/__grp__l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..ac00b83711d17d4c4c2ac2fc70e56998f349a12a --- /dev/null +++ b/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/__grp__l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_bwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/l2norm_bwd_kernel.source", "l2norm_bwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/l2norm_bwd_kernel.ttir", "l2norm_bwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/l2norm_bwd_kernel.ttgir", "l2norm_bwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/l2norm_bwd_kernel.llir", "l2norm_bwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/l2norm_bwd_kernel.ptx", "l2norm_bwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/l2norm_bwd_kernel.cubin", "l2norm_bwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/l2norm_bwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/l2norm_bwd_kernel.cubin b/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/l2norm_bwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..07336d0c968f05451b2d967e87305024051ba04a Binary files /dev/null and b/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/l2norm_bwd_kernel.cubin differ diff --git a/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/l2norm_bwd_kernel.json b/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..cd75737730e1ae153de2d816e407c5e8dda49c66 --- /dev/null +++ b/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"hash": "11e51f60e2bd6455888a9b2d7516accdfeaad317fc9b287f16969961108d30c9", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 2, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 0, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_bwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/l2norm_bwd_kernel.llir b/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/l2norm_bwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..67e15bc523b1d03f8b386365eee89bd04cde64d1 --- /dev/null +++ b/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/l2norm_bwd_kernel.llir @@ -0,0 +1,377 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_bwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, ptr addrspace(1) %3, float %4, i32 %5, ptr addrspace(1) readnone captures(none) %6, ptr addrspace(1) readnone captures(none) %7) local_unnamed_addr #0 !dbg !4 { + %9 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !12 + %10 = zext nneg i32 %9 to i64, !dbg !13 + %11 = shl nuw nsw i64 %10, 3, !dbg !14 + %12 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !15 + %13 = lshr i32 %12, 4, !dbg !15 + %14 = and i32 %13, 3, !dbg !15 + %15 = or disjoint i32 %14, 4, !dbg !15 + %16 = zext nneg i32 %14 to i64, !dbg !16 + %17 = zext nneg i32 %15 to i64, !dbg !16 + %18 = or disjoint i64 %11, %16, !dbg !16 + %19 = or disjoint i64 %11, %17, !dbg !16 + %20 = sext i32 %5 to i64, !dbg !17 + %21 = icmp slt i64 %18, %20, !dbg !17 + %22 = icmp slt i64 %19, %20, !dbg !17 + %23 = shl nuw nsw i32 %12, 3, !dbg !18 + %24 = and i32 %23, 120, !dbg !18 + %25 = shl nuw nsw i64 %18, 7, !dbg !19 + %26 = shl nuw nsw i64 %19, 7, !dbg !19 + %27 = getelementptr bfloat, ptr addrspace(1) %0, i64 %25, !dbg !20 + %28 = getelementptr bfloat, ptr addrspace(1) %0, i64 %26, !dbg !20 + %29 = zext nneg i32 %24 to i64, !dbg !21 + %30 = getelementptr bfloat, ptr addrspace(1) %27, i64 %29, !dbg !21 + %31 = getelementptr bfloat, ptr addrspace(1) %28, i64 %29, !dbg !21 + %32 = getelementptr float, ptr addrspace(1) %1, i64 %18, !dbg !22 + %33 = getelementptr float, ptr addrspace(1) %1, i64 %19, !dbg !22 + %34 = getelementptr bfloat, ptr addrspace(1) %2, i64 %25, !dbg !23 + %35 = getelementptr bfloat, ptr addrspace(1) %2, i64 %26, !dbg !23 + %36 = getelementptr bfloat, ptr addrspace(1) %34, i64 %29, !dbg !24 + %37 = getelementptr bfloat, ptr addrspace(1) %35, i64 %29, !dbg !24 + %38 = getelementptr bfloat, ptr addrspace(1) %3, i64 %25, !dbg !25 + %39 = getelementptr bfloat, ptr addrspace(1) %3, i64 %26, !dbg !25 + %40 = getelementptr bfloat, ptr addrspace(1) %38, i64 %29, !dbg !26 + %41 = getelementptr bfloat, ptr addrspace(1) %39, i64 %29, !dbg !26 + %42 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %30, i1 %21) #3, !dbg !27 + %43 = extractvalue { i32, i32, i32, i32 } %42, 0, !dbg !27 + %44 = bitcast i32 %43 to <2 x bfloat>, !dbg !27 + %45 = extractvalue { i32, i32, i32, i32 } %42, 1, !dbg !27 + %46 = bitcast i32 %45 to <2 x bfloat>, !dbg !27 + %47 = extractvalue { i32, i32, i32, i32 } %42, 2, !dbg !27 + %48 = bitcast i32 %47 to <2 x bfloat>, !dbg !27 + %49 = extractvalue { i32, i32, i32, i32 } %42, 3, !dbg !27 + %50 = bitcast i32 %49 to <2 x bfloat>, !dbg !27 + %51 = extractelement <2 x bfloat> %44, i64 0, !dbg !27 + %52 = extractelement <2 x bfloat> %44, i64 1, !dbg !27 + %53 = extractelement <2 x bfloat> %46, i64 0, !dbg !27 + %54 = extractelement <2 x bfloat> %46, i64 1, !dbg !27 + %55 = extractelement <2 x bfloat> %48, i64 0, !dbg !27 + %56 = extractelement <2 x bfloat> %48, i64 1, !dbg !27 + %57 = extractelement <2 x bfloat> %50, i64 0, !dbg !27 + %58 = extractelement <2 x bfloat> %50, i64 1, !dbg !27 + %59 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %31, i1 %22) #3, !dbg !27 + %60 = extractvalue { i32, i32, i32, i32 } %59, 0, !dbg !27 + %61 = bitcast i32 %60 to <2 x bfloat>, !dbg !27 + %62 = extractvalue { i32, i32, i32, i32 } %59, 1, !dbg !27 + %63 = bitcast i32 %62 to <2 x bfloat>, !dbg !27 + %64 = extractvalue { i32, i32, i32, i32 } %59, 2, !dbg !27 + %65 = bitcast i32 %64 to <2 x bfloat>, !dbg !27 + %66 = extractvalue { i32, i32, i32, i32 } %59, 3, !dbg !27 + %67 = bitcast i32 %66 to <2 x bfloat>, !dbg !27 + %68 = extractelement <2 x bfloat> %61, i64 0, !dbg !27 + %69 = extractelement <2 x bfloat> %61, i64 1, !dbg !27 + %70 = extractelement <2 x bfloat> %63, i64 0, !dbg !27 + %71 = extractelement <2 x bfloat> %63, i64 1, !dbg !27 + %72 = extractelement <2 x bfloat> %65, i64 0, !dbg !27 + %73 = extractelement <2 x bfloat> %65, i64 1, !dbg !27 + %74 = extractelement <2 x bfloat> %67, i64 0, !dbg !27 + %75 = extractelement <2 x bfloat> %67, i64 1, !dbg !27 + %76 = fpext bfloat %51 to float, !dbg !28 + %77 = fpext bfloat %52 to float, !dbg !28 + %78 = fpext bfloat %53 to float, !dbg !28 + %79 = fpext bfloat %54 to float, !dbg !28 + %80 = fpext bfloat %55 to float, !dbg !28 + %81 = fpext bfloat %56 to float, !dbg !28 + %82 = fpext bfloat %57 to float, !dbg !28 + %83 = fpext bfloat %58 to float, !dbg !28 + %84 = fpext bfloat %68 to float, !dbg !28 + %85 = fpext bfloat %69 to float, !dbg !28 + %86 = fpext bfloat %70 to float, !dbg !28 + %87 = fpext bfloat %71 to float, !dbg !28 + %88 = fpext bfloat %72 to float, !dbg !28 + %89 = fpext bfloat %73 to float, !dbg !28 + %90 = fpext bfloat %74 to float, !dbg !28 + %91 = fpext bfloat %75 to float, !dbg !28 + %92 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %32, i1 %21) #3, !dbg !29 + %93 = bitcast i32 %92 to float, !dbg !29 + %94 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %33, i1 %22) #3, !dbg !29 + %95 = bitcast i32 %94 to float, !dbg !29 + %96 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %36, i1 %21) #3, !dbg !30 + %97 = extractvalue { i32, i32, i32, i32 } %96, 0, !dbg !30 + %98 = bitcast i32 %97 to <2 x bfloat>, !dbg !30 + %99 = extractvalue { i32, i32, i32, i32 } %96, 1, !dbg !30 + %100 = bitcast i32 %99 to <2 x bfloat>, !dbg !30 + %101 = extractvalue { i32, i32, i32, i32 } %96, 2, !dbg !30 + %102 = bitcast i32 %101 to <2 x bfloat>, !dbg !30 + %103 = extractvalue { i32, i32, i32, i32 } %96, 3, !dbg !30 + %104 = bitcast i32 %103 to <2 x bfloat>, !dbg !30 + %105 = extractelement <2 x bfloat> %98, i64 0, !dbg !30 + %106 = extractelement <2 x bfloat> %98, i64 1, !dbg !30 + %107 = extractelement <2 x bfloat> %100, i64 0, !dbg !30 + %108 = extractelement <2 x bfloat> %100, i64 1, !dbg !30 + %109 = extractelement <2 x bfloat> %102, i64 0, !dbg !30 + %110 = extractelement <2 x bfloat> %102, i64 1, !dbg !30 + %111 = extractelement <2 x bfloat> %104, i64 0, !dbg !30 + %112 = extractelement <2 x bfloat> %104, i64 1, !dbg !30 + %113 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %37, i1 %22) #3, !dbg !30 + %114 = extractvalue { i32, i32, i32, i32 } %113, 0, !dbg !30 + %115 = bitcast i32 %114 to <2 x bfloat>, !dbg !30 + %116 = extractvalue { i32, i32, i32, i32 } %113, 1, !dbg !30 + %117 = bitcast i32 %116 to <2 x bfloat>, !dbg !30 + %118 = extractvalue { i32, i32, i32, i32 } %113, 2, !dbg !30 + %119 = bitcast i32 %118 to <2 x bfloat>, !dbg !30 + %120 = extractvalue { i32, i32, i32, i32 } %113, 3, !dbg !30 + %121 = bitcast i32 %120 to <2 x bfloat>, !dbg !30 + %122 = extractelement <2 x bfloat> %115, i64 0, !dbg !30 + %123 = extractelement <2 x bfloat> %115, i64 1, !dbg !30 + %124 = extractelement <2 x bfloat> %117, i64 0, !dbg !30 + %125 = extractelement <2 x bfloat> %117, i64 1, !dbg !30 + %126 = extractelement <2 x bfloat> %119, i64 0, !dbg !30 + %127 = extractelement <2 x bfloat> %119, i64 1, !dbg !30 + %128 = extractelement <2 x bfloat> %121, i64 0, !dbg !30 + %129 = extractelement <2 x bfloat> %121, i64 1, !dbg !30 + %130 = fpext bfloat %105 to float, !dbg !31 + %131 = fpext bfloat %106 to float, !dbg !31 + %132 = fpext bfloat %107 to float, !dbg !31 + %133 = fpext bfloat %108 to float, !dbg !31 + %134 = fpext bfloat %109 to float, !dbg !31 + %135 = fpext bfloat %110 to float, !dbg !31 + %136 = fpext bfloat %111 to float, !dbg !31 + %137 = fpext bfloat %112 to float, !dbg !31 + %138 = fpext bfloat %122 to float, !dbg !31 + %139 = fpext bfloat %123 to float, !dbg !31 + %140 = fpext bfloat %124 to float, !dbg !31 + %141 = fpext bfloat %125 to float, !dbg !31 + %142 = fpext bfloat %126 to float, !dbg !31 + %143 = fpext bfloat %127 to float, !dbg !31 + %144 = fpext bfloat %128 to float, !dbg !31 + %145 = fpext bfloat %129 to float, !dbg !31 + %146 = fmul float %93, %130, !dbg !32 + %147 = fmul float %93, %131, !dbg !32 + %148 = fmul float %93, %132, !dbg !32 + %149 = fmul float %93, %133, !dbg !32 + %150 = fmul float %93, %134, !dbg !32 + %151 = fmul float %93, %135, !dbg !32 + %152 = fmul float %93, %136, !dbg !32 + %153 = fmul float %93, %137, !dbg !32 + %154 = fmul float %95, %138, !dbg !32 + %155 = fmul float %95, %139, !dbg !32 + %156 = fmul float %95, %140, !dbg !32 + %157 = fmul float %95, %141, !dbg !32 + %158 = fmul float %95, %142, !dbg !32 + %159 = fmul float %95, %143, !dbg !32 + %160 = fmul float %95, %144, !dbg !32 + %161 = fmul float %95, %145, !dbg !32 + %162 = fmul float %76, %130, !dbg !33 + %163 = fmul float %77, %131, !dbg !33 + %164 = fmul float %78, %132, !dbg !33 + %165 = fmul float %79, %133, !dbg !33 + %166 = fmul float %80, %134, !dbg !33 + %167 = fmul float %81, %135, !dbg !33 + %168 = fmul float %82, %136, !dbg !33 + %169 = fmul float %83, %137, !dbg !33 + %170 = fmul float %84, %138, !dbg !33 + %171 = fmul float %85, %139, !dbg !33 + %172 = fmul float %86, %140, !dbg !33 + %173 = fmul float %87, %141, !dbg !33 + %174 = fmul float %88, %142, !dbg !33 + %175 = fmul float %89, %143, !dbg !33 + %176 = fmul float %90, %144, !dbg !33 + %177 = fmul float %91, %145, !dbg !33 + %178 = fadd float %162, %163, !dbg !34 + %179 = fadd float %164, %178, !dbg !34 + %180 = fadd float %165, %179, !dbg !34 + %181 = fadd float %166, %180, !dbg !34 + %182 = fadd float %167, %181, !dbg !34 + %183 = fadd float %168, %182, !dbg !34 + %184 = fadd float %169, %183, !dbg !34 + %185 = fadd float %170, %171, !dbg !34 + %186 = fadd float %172, %185, !dbg !34 + %187 = fadd float %173, %186, !dbg !34 + %188 = fadd float %174, %187, !dbg !34 + %189 = fadd float %175, %188, !dbg !34 + %190 = fadd float %176, %189, !dbg !34 + %191 = fadd float %177, %190, !dbg !34 + %192 = bitcast float %184 to i32, !dbg !37 + %193 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %192, i32 8, i32 31), !dbg !37 + %194 = bitcast i32 %193 to float, !dbg !37 + %195 = fadd float %184, %194, !dbg !34 + %196 = bitcast float %195 to i32, !dbg !37 + %197 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %196, i32 4, i32 31), !dbg !37 + %198 = bitcast i32 %197 to float, !dbg !37 + %199 = fadd float %195, %198, !dbg !34 + %200 = bitcast float %199 to i32, !dbg !37 + %201 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %200, i32 2, i32 31), !dbg !37 + %202 = bitcast i32 %201 to float, !dbg !37 + %203 = fadd float %199, %202, !dbg !34 + %204 = bitcast float %203 to i32, !dbg !37 + %205 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %204, i32 1, i32 31), !dbg !37 + %206 = bitcast i32 %205 to float, !dbg !37 + %207 = fadd float %203, %206, !dbg !34 + %208 = bitcast float %191 to i32, !dbg !37 + %209 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %208, i32 8, i32 31), !dbg !37 + %210 = bitcast i32 %209 to float, !dbg !37 + %211 = fadd float %191, %210, !dbg !34 + %212 = bitcast float %211 to i32, !dbg !37 + %213 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %212, i32 4, i32 31), !dbg !37 + %214 = bitcast i32 %213 to float, !dbg !37 + %215 = fadd float %211, %214, !dbg !34 + %216 = bitcast float %215 to i32, !dbg !37 + %217 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %216, i32 2, i32 31), !dbg !37 + %218 = bitcast i32 %217 to float, !dbg !37 + %219 = fadd float %215, %218, !dbg !34 + %220 = bitcast float %219 to i32, !dbg !37 + %221 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %220, i32 1, i32 31), !dbg !37 + %222 = bitcast i32 %221 to float, !dbg !37 + %223 = fadd float %219, %222, !dbg !34 + %224 = fmul float %207, %76, !dbg !40 + %225 = fmul float %207, %77, !dbg !40 + %226 = fmul float %207, %78, !dbg !40 + %227 = fmul float %207, %79, !dbg !40 + %228 = fmul float %207, %80, !dbg !40 + %229 = fmul float %207, %81, !dbg !40 + %230 = fmul float %207, %82, !dbg !40 + %231 = fmul float %207, %83, !dbg !40 + %232 = fmul float %223, %84, !dbg !40 + %233 = fmul float %223, %85, !dbg !40 + %234 = fmul float %223, %86, !dbg !40 + %235 = fmul float %223, %87, !dbg !40 + %236 = fmul float %223, %88, !dbg !40 + %237 = fmul float %223, %89, !dbg !40 + %238 = fmul float %223, %90, !dbg !40 + %239 = fmul float %223, %91, !dbg !40 + %240 = fmul float %224, %93, !dbg !41 + %241 = fmul float %225, %93, !dbg !41 + %242 = fmul float %226, %93, !dbg !41 + %243 = fmul float %227, %93, !dbg !41 + %244 = fmul float %228, %93, !dbg !41 + %245 = fmul float %229, %93, !dbg !41 + %246 = fmul float %230, %93, !dbg !41 + %247 = fmul float %231, %93, !dbg !41 + %248 = fmul float %232, %95, !dbg !41 + %249 = fmul float %233, %95, !dbg !41 + %250 = fmul float %234, %95, !dbg !41 + %251 = fmul float %235, %95, !dbg !41 + %252 = fmul float %236, %95, !dbg !41 + %253 = fmul float %237, %95, !dbg !41 + %254 = fmul float %238, %95, !dbg !41 + %255 = fmul float %239, %95, !dbg !41 + %256 = fsub float %146, %240, !dbg !42 + %257 = fsub float %147, %241, !dbg !42 + %258 = fsub float %148, %242, !dbg !42 + %259 = fsub float %149, %243, !dbg !42 + %260 = fsub float %150, %244, !dbg !42 + %261 = fsub float %151, %245, !dbg !42 + %262 = fsub float %152, %246, !dbg !42 + %263 = fsub float %153, %247, !dbg !42 + %264 = fsub float %154, %248, !dbg !42 + %265 = fsub float %155, %249, !dbg !42 + %266 = fsub float %156, %250, !dbg !42 + %267 = fsub float %157, %251, !dbg !42 + %268 = fsub float %158, %252, !dbg !42 + %269 = fsub float %159, %253, !dbg !42 + %270 = fsub float %160, %254, !dbg !42 + %271 = fsub float %161, %255, !dbg !42 + %272 = fptrunc float %256 to bfloat, !dbg !43 + %273 = fptrunc float %257 to bfloat, !dbg !43 + %274 = fptrunc float %258 to bfloat, !dbg !43 + %275 = fptrunc float %259 to bfloat, !dbg !43 + %276 = fptrunc float %260 to bfloat, !dbg !43 + %277 = fptrunc float %261 to bfloat, !dbg !43 + %278 = fptrunc float %262 to bfloat, !dbg !43 + %279 = fptrunc float %263 to bfloat, !dbg !43 + %280 = fptrunc float %264 to bfloat, !dbg !43 + %281 = fptrunc float %265 to bfloat, !dbg !43 + %282 = fptrunc float %266 to bfloat, !dbg !43 + %283 = fptrunc float %267 to bfloat, !dbg !43 + %284 = fptrunc float %268 to bfloat, !dbg !43 + %285 = fptrunc float %269 to bfloat, !dbg !43 + %286 = fptrunc float %270 to bfloat, !dbg !43 + %287 = fptrunc float %271 to bfloat, !dbg !43 + %288 = insertelement <2 x bfloat> poison, bfloat %272, i64 0, !dbg !44 + %289 = insertelement <2 x bfloat> %288, bfloat %273, i64 1, !dbg !44 + %290 = bitcast <2 x bfloat> %289 to i32, !dbg !44 + %291 = insertelement <2 x bfloat> poison, bfloat %274, i64 0, !dbg !44 + %292 = insertelement <2 x bfloat> %291, bfloat %275, i64 1, !dbg !44 + %293 = bitcast <2 x bfloat> %292 to i32, !dbg !44 + %294 = insertelement <2 x bfloat> poison, bfloat %276, i64 0, !dbg !44 + %295 = insertelement <2 x bfloat> %294, bfloat %277, i64 1, !dbg !44 + %296 = bitcast <2 x bfloat> %295 to i32, !dbg !44 + %297 = insertelement <2 x bfloat> poison, bfloat %278, i64 0, !dbg !44 + %298 = insertelement <2 x bfloat> %297, bfloat %279, i64 1, !dbg !44 + %299 = bitcast <2 x bfloat> %298 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %290, i32 %293, i32 %296, i32 %299, ptr addrspace(1) %40, i1 %21) #3, !dbg !44 + %300 = insertelement <2 x bfloat> poison, bfloat %280, i64 0, !dbg !44 + %301 = insertelement <2 x bfloat> %300, bfloat %281, i64 1, !dbg !44 + %302 = bitcast <2 x bfloat> %301 to i32, !dbg !44 + %303 = insertelement <2 x bfloat> poison, bfloat %282, i64 0, !dbg !44 + %304 = insertelement <2 x bfloat> %303, bfloat %283, i64 1, !dbg !44 + %305 = bitcast <2 x bfloat> %304 to i32, !dbg !44 + %306 = insertelement <2 x bfloat> poison, bfloat %284, i64 0, !dbg !44 + %307 = insertelement <2 x bfloat> %306, bfloat %285, i64 1, !dbg !44 + %308 = bitcast <2 x bfloat> %307 to i32, !dbg !44 + %309 = insertelement <2 x bfloat> poison, bfloat %286, i64 0, !dbg !44 + %310 = insertelement <2 x bfloat> %309, bfloat %287, i64 1, !dbg !44 + %311 = bitcast <2 x bfloat> %310 to i32, !dbg !44 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %302, i32 %305, i32 %308, i32 %311, ptr addrspace(1) %41, i1 %22) #3, !dbg !44 + ret void, !dbg !45 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +attributes #0 = { nounwind "nvvm.reqntid"="64" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = distinct !DISubprogram(name: "l2norm_bwd_kernel", linkageName: "l2norm_bwd_kernel", scope: !1, file: !1, line: 93, type: !5, scopeLine: 93, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!5 = !DISubroutineType(cc: DW_CC_normal, types: !6) +!6 = !{null, !7, !9, !7, !7, !10, !11, !7, !7} +!7 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !8, size: 64, dwarfAddressSpace: 1) +!8 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!9 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !10, size: 64, dwarfAddressSpace: 1) +!10 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!11 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!12 = !DILocation(line: 94, column: 24, scope: !4) +!13 = !DILocation(line: 94, column: 30, scope: !4) +!14 = !DILocation(line: 95, column: 16, scope: !4) +!15 = !DILocation(line: 95, column: 34, scope: !4) +!16 = !DILocation(line: 95, column: 21, scope: !4) +!17 = !DILocation(line: 97, column: 16, scope: !4) +!18 = !DILocation(line: 98, column: 30, scope: !4) +!19 = !DILocation(line: 99, column: 29, scope: !4) +!20 = !DILocation(line: 99, column: 14, scope: !4) +!21 = !DILocation(line: 99, column: 33, scope: !4) +!22 = !DILocation(line: 100, column: 20, scope: !4) +!23 = !DILocation(line: 101, column: 16, scope: !4) +!24 = !DILocation(line: 101, column: 35, scope: !4) +!25 = !DILocation(line: 102, column: 16, scope: !4) +!26 = !DILocation(line: 102, column: 35, scope: !4) +!27 = !DILocation(line: 104, column: 18, scope: !4) +!28 = !DILocation(line: 104, column: 47, scope: !4) +!29 = !DILocation(line: 105, column: 21, scope: !4) +!30 = !DILocation(line: 106, column: 19, scope: !4) +!31 = !DILocation(line: 106, column: 49, scope: !4) +!32 = !DILocation(line: 107, column: 18, scope: !4) +!33 = !DILocation(line: 107, column: 50, scope: !4) +!34 = !DILocation(line: 263, column: 15, scope: !35, inlinedAt: !37) +!35 = distinct !DILexicalBlockFile(scope: !4, file: !36, discriminator: 0) +!36 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!37 = !DILocation(line: 293, column: 36, scope: !35, inlinedAt: !38) +!38 = !DILocation(line: 107, column: 55, scope: !39) +!39 = distinct !DILexicalBlockFile(scope: !4, file: !1, discriminator: 0) +!40 = !DILocation(line: 107, column: 69, scope: !4) +!41 = !DILocation(line: 107, column: 75, scope: !4) +!42 = !DILocation(line: 107, column: 36, scope: !4) +!43 = !DILocation(line: 108, column: 27, scope: !4) +!44 = !DILocation(line: 108, column: 19, scope: !4) +!45 = !DILocation(line: 108, column: 4, scope: !4) diff --git a/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/l2norm_bwd_kernel.ptx b/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/l2norm_bwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..d1b892b9b6d105e4d6e0df9d1cb1b8445a1e72b4 --- /dev/null +++ b/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/l2norm_bwd_kernel.ptx @@ -0,0 +1,546 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_bwd_kernel // -- Begin function l2norm_bwd_kernel + // @l2norm_bwd_kernel +.visible .entry l2norm_bwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_2, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_3, + .param .f32 l2norm_bwd_kernel_param_4, + .param .u32 l2norm_bwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_6, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_7 +) +.reqntid 64 +{ + .reg .pred %p<3>; + .reg .b16 %rs<33>; + .reg .b32 %r<162>; + .reg .b64 %rd<29>; + .loc 1 93 0 // l2norm.py:93:0 +$L__func_begin0: + .loc 1 93 0 // l2norm.py:93:0 + +// %bb.0: + ld.param.b64 %rd9, [l2norm_bwd_kernel_param_0]; + ld.param.b64 %rd10, [l2norm_bwd_kernel_param_1]; +$L__tmp0: + .loc 1 94 24 // l2norm.py:94:24 + mov.u32 %r28, %ctaid.x; + ld.param.b64 %rd11, [l2norm_bwd_kernel_param_2]; + .loc 1 95 16 // l2norm.py:95:16 + mul.wide.u32 %rd12, %r28, 8; + ld.param.b64 %rd13, [l2norm_bwd_kernel_param_3]; + .loc 1 95 34 // l2norm.py:95:34 + mov.u32 %r29, %tid.x; + ld.param.s32 %rd14, [l2norm_bwd_kernel_param_5]; + bfe.u32 %r30, %r29, 4, 2; + or.b32 %r31, %r30, 4; + .loc 1 95 21 // l2norm.py:95:21 + cvt.u64.u32 %rd15, %r30; + cvt.u64.u32 %rd16, %r31; + or.b64 %rd17, %rd12, %rd15; + or.b64 %rd18, %rd12, %rd16; + .loc 1 97 16 // l2norm.py:97:16 + setp.lt.s64 %p1, %rd17, %rd14; + setp.lt.s64 %p2, %rd18, %rd14; + .loc 1 98 30 // l2norm.py:98:30 + shl.b32 %r32, %r29, 3; + and.b32 %r33, %r32, 120; + .loc 1 99 14 // l2norm.py:99:14 + shl.b64 %rd19, %rd17, 8; + add.s64 %rd20, %rd9, %rd19; + shl.b64 %rd21, %rd18, 8; + add.s64 %rd22, %rd9, %rd21; + .loc 1 99 33 // l2norm.py:99:33 + mul.wide.u32 %rd23, %r33, 2; + add.s64 %rd1, %rd20, %rd23; + add.s64 %rd2, %rd22, %rd23; + .loc 1 100 20 // l2norm.py:100:20 + shl.b64 %rd24, %rd17, 2; + add.s64 %rd3, %rd10, %rd24; + add.s64 %rd4, %rd3, 16; + .loc 1 101 16 // l2norm.py:101:16 + add.s64 %rd25, %rd11, %rd19; + add.s64 %rd26, %rd11, %rd21; + .loc 1 101 35 // l2norm.py:101:35 + add.s64 %rd5, %rd25, %rd23; + add.s64 %rd6, %rd26, %rd23; + .loc 1 102 16 // l2norm.py:102:16 + add.s64 %rd27, %rd13, %rd19; + add.s64 %rd28, %rd13, %rd21; + .loc 1 102 35 // l2norm.py:102:35 + add.s64 %rd7, %rd27, %rd23; + add.s64 %rd8, %rd28, %rd23; + mov.b32 %r5, 0; + .loc 1 104 18 // l2norm.py:104:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + mov.b32 {%rs5, %rs6}, %r3; + mov.b32 {%rs7, %rs8}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs9, %rs10}, %r6; + mov.b32 {%rs11, %rs12}, %r7; + mov.b32 {%rs13, %rs14}, %r8; + mov.b32 {%rs15, %rs16}, %r9; + .loc 1 104 47 // l2norm.py:104:47 + cvt.f32.bf16 %r34, %rs1; + cvt.f32.bf16 %r35, %rs2; + cvt.f32.bf16 %r36, %rs3; + cvt.f32.bf16 %r37, %rs4; + cvt.f32.bf16 %r38, %rs5; + cvt.f32.bf16 %r39, %rs6; + cvt.f32.bf16 %r40, %rs7; + cvt.f32.bf16 %r41, %rs8; + cvt.f32.bf16 %r42, %rs9; + cvt.f32.bf16 %r43, %rs10; + cvt.f32.bf16 %r44, %rs11; + cvt.f32.bf16 %r45, %rs12; + cvt.f32.bf16 %r46, %rs13; + cvt.f32.bf16 %r47, %rs14; + cvt.f32.bf16 %r48, %rs15; + cvt.f32.bf16 %r49, %rs16; + .loc 1 105 21 // l2norm.py:105:21 + // begin inline asm + mov.u32 %r10, %r5; + @%p1 ld.global.b32 { %r10 }, [ %rd3 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r11, %r5; + @%p2 ld.global.b32 { %r11 }, [ %rd4 + 0 ]; + // end inline asm + .loc 1 106 19 // l2norm.py:106:19 + // begin inline asm + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + @%p1 ld.global.v4.b32 { %r12, %r13, %r14, %r15 }, [ %rd5 + 0 ]; + // end inline asm + mov.b32 {%rs17, %rs18}, %r12; + mov.b32 {%rs19, %rs20}, %r13; + mov.b32 {%rs21, %rs22}, %r14; + mov.b32 {%rs23, %rs24}, %r15; + // begin inline asm + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + mov.u32 %r18, %r5; + mov.u32 %r19, %r5; + @%p2 ld.global.v4.b32 { %r16, %r17, %r18, %r19 }, [ %rd6 + 0 ]; + // end inline asm + mov.b32 {%rs25, %rs26}, %r16; + mov.b32 {%rs27, %rs28}, %r17; + mov.b32 {%rs29, %rs30}, %r18; + mov.b32 {%rs31, %rs32}, %r19; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r50, %rs17; + cvt.f32.bf16 %r51, %rs18; + cvt.f32.bf16 %r52, %rs19; + cvt.f32.bf16 %r53, %rs20; + cvt.f32.bf16 %r54, %rs21; + cvt.f32.bf16 %r55, %rs22; + cvt.f32.bf16 %r56, %rs23; + cvt.f32.bf16 %r57, %rs24; + cvt.f32.bf16 %r58, %rs25; + cvt.f32.bf16 %r59, %rs26; + cvt.f32.bf16 %r60, %rs27; + cvt.f32.bf16 %r61, %rs28; + cvt.f32.bf16 %r62, %rs29; + cvt.f32.bf16 %r63, %rs30; + cvt.f32.bf16 %r64, %rs31; + cvt.f32.bf16 %r65, %rs32; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r66, %r35, %r51; + mul.f32 %r67, %r43, %r59; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r68, %rs1, %rs17, %r66; + fma.rn.f32.bf16 %r69, %rs3, %rs19, %r68; + fma.rn.f32.bf16 %r70, %rs4, %rs20, %r69; + fma.rn.f32.bf16 %r71, %rs5, %rs21, %r70; + fma.rn.f32.bf16 %r72, %rs6, %rs22, %r71; + fma.rn.f32.bf16 %r73, %rs7, %rs23, %r72; + fma.rn.f32.bf16 %r74, %rs8, %rs24, %r73; + fma.rn.f32.bf16 %r75, %rs9, %rs25, %r67; + fma.rn.f32.bf16 %r76, %rs11, %rs27, %r75; + fma.rn.f32.bf16 %r77, %rs12, %rs28, %r76; + fma.rn.f32.bf16 %r78, %rs13, %rs29, %r77; + fma.rn.f32.bf16 %r79, %rs14, %rs30, %r78; + fma.rn.f32.bf16 %r80, %rs15, %rs31, %r79; + fma.rn.f32.bf16 %r81, %rs16, %rs32, %r80; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r82, %r74, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r83, %r74, %r82; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r84, %r83, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r85, %r83, %r84; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r86, %r85, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r87, %r85, %r86; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r88, %r87, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r89, %r87, %r88; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r90, %r81, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r91, %r81, %r90; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r92, %r91, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r93, %r91, %r92; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r94, %r93, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r95, %r93, %r94; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r96, %r95, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r97, %r95, %r96; +$L__tmp18: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r98, %r89, %r34; + mul.f32 %r99, %r89, %r35; + mul.f32 %r100, %r89, %r36; + mul.f32 %r101, %r89, %r37; + mul.f32 %r102, %r89, %r38; + mul.f32 %r103, %r89, %r39; + mul.f32 %r104, %r89, %r40; + mul.f32 %r105, %r89, %r41; + mul.f32 %r106, %r97, %r42; + mul.f32 %r107, %r97, %r43; + mul.f32 %r108, %r97, %r44; + mul.f32 %r109, %r97, %r45; + mul.f32 %r110, %r97, %r46; + mul.f32 %r111, %r97, %r47; + mul.f32 %r112, %r97, %r48; + mul.f32 %r113, %r97, %r49; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r114, %r98, %r10; + mul.f32 %r115, %r99, %r10; + mul.f32 %r116, %r100, %r10; + mul.f32 %r117, %r101, %r10; + mul.f32 %r118, %r102, %r10; + mul.f32 %r119, %r103, %r10; + mul.f32 %r120, %r104, %r10; + mul.f32 %r121, %r105, %r10; + mul.f32 %r122, %r106, %r11; + mul.f32 %r123, %r107, %r11; + mul.f32 %r124, %r108, %r11; + mul.f32 %r125, %r109, %r11; + mul.f32 %r126, %r110, %r11; + mul.f32 %r127, %r111, %r11; + mul.f32 %r128, %r112, %r11; + mul.f32 %r129, %r113, %r11; + .loc 1 107 36 // l2norm.py:107:36 + neg.f32 %r130, %r114; + fma.rn.f32 %r131, %r10, %r50, %r130; + neg.f32 %r132, %r115; + fma.rn.f32 %r133, %r10, %r51, %r132; + neg.f32 %r134, %r116; + fma.rn.f32 %r135, %r10, %r52, %r134; + neg.f32 %r136, %r117; + fma.rn.f32 %r137, %r10, %r53, %r136; + neg.f32 %r138, %r118; + fma.rn.f32 %r139, %r10, %r54, %r138; + neg.f32 %r140, %r119; + fma.rn.f32 %r141, %r10, %r55, %r140; + neg.f32 %r142, %r120; + fma.rn.f32 %r143, %r10, %r56, %r142; + neg.f32 %r144, %r121; + fma.rn.f32 %r145, %r10, %r57, %r144; + neg.f32 %r146, %r122; + fma.rn.f32 %r147, %r11, %r58, %r146; + neg.f32 %r148, %r123; + fma.rn.f32 %r149, %r11, %r59, %r148; + neg.f32 %r150, %r124; + fma.rn.f32 %r151, %r11, %r60, %r150; + neg.f32 %r152, %r125; + fma.rn.f32 %r153, %r11, %r61, %r152; + neg.f32 %r154, %r126; + fma.rn.f32 %r155, %r11, %r62, %r154; + neg.f32 %r156, %r127; + fma.rn.f32 %r157, %r11, %r63, %r156; + neg.f32 %r158, %r128; + fma.rn.f32 %r159, %r11, %r64, %r158; + neg.f32 %r160, %r129; + fma.rn.f32 %r161, %r11, %r65, %r160; + .loc 1 108 19 // l2norm.py:108:19 + cvt.rn.bf16x2.f32 %r20, %r133, %r131; + cvt.rn.bf16x2.f32 %r21, %r137, %r135; + cvt.rn.bf16x2.f32 %r22, %r141, %r139; + cvt.rn.bf16x2.f32 %r23, %r145, %r143; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd7 + 0 ], { %r20, %r21, %r22, %r23 }; + // end inline asm + cvt.rn.bf16x2.f32 %r24, %r149, %r147; + cvt.rn.bf16x2.f32 %r25, %r153, %r151; + cvt.rn.bf16x2.f32 %r26, %r157, %r155; + cvt.rn.bf16x2.f32 %r27, %r161, %r159; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd8 + 0 ], { %r24, %r25, %r26, %r27 }; + // end inline asm + .loc 1 108 4 // l2norm.py:108:4 + ret; +$L__tmp19: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 98 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp18 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 107 // DW_AT_call_line +.b8 55 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp18 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/l2norm_bwd_kernel.source b/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/l2norm_bwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..c23a4689ce5f60989f46cd34fecfe31d002a62d8 --- /dev/null +++ b/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/l2norm_bwd_kernel.source @@ -0,0 +1,210 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc47 = loc(unknown) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc54 = loc("y"(#loc)) +#loc55 = loc("rstd"(#loc)) +#loc56 = loc("dy"(#loc)) +#loc57 = loc("dx"(#loc)) +#loc58 = loc("eps"(#loc)) +#loc59 = loc("T"(#loc)) +#loc101 = loc("input"(#loc45)) +#loc102 = loc("a"(#loc50)) +#loc103 = loc("b"(#loc50)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc60) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc61) + %o_t = arith.constant 8 : i32 loc(#loc62) + %o_t_1 = arith.constant 8 : i64 loc(#loc62) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc62) + %o_t_3 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32> loc(#loc63) + %o_t_4 = arith.extsi %o_t_3 : tensor<8xi32> to tensor<8xi64> loc(#loc64) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<8xi64> loc(#loc64) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<8xi64> loc(#loc64) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc65) + %m_t = arith.extsi %T : i32 to i64 loc(#loc66) + %m_t_7 = tt.splat %m_t : i64 -> tensor<8xi64> loc(#loc66) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<8xi64> loc(#loc66) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<8xi1> -> tensor<8x1xi1> loc(#loc67) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc68) + %m_x_10 = arith.constant 128 : i32 loc(#loc69) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc69) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc69) + %m_x_13 = tt.broadcast %m_x : tensor<8x1xi1> -> tensor<8x128xi1> loc(#loc70) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<8x128xi1> loc(#loc70) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<8x128xi1> loc(#loc70) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc71) + %p_y_16 = arith.constant 128 : i32 loc(#loc72) + %p_y_17 = arith.constant 128 : i64 loc(#loc72) + %p_y_18 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc72) + %p_y_19 = arith.muli %p_y, %p_y_18 : tensor<8x1xi64> loc(#loc72) + %p_y_20 = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc73) + %p_y_21 = tt.addptr %p_y_20, %p_y_19 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc73) + %p_y_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc74) + %p_y_23 = tt.broadcast %p_y_21 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc75) + %p_y_24 = tt.broadcast %p_y_22 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc75) + %p_y_25 = tt.addptr %p_y_23, %p_y_24 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc75) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr> loc(#loc76) + %p_rstd_26 = tt.addptr %p_rstd, %o_t_6 : tensor<8x!tt.ptr>, tensor<8xi64> loc(#loc76) + %p_dy = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc77) + %p_dy_27 = arith.constant 128 : i32 loc(#loc78) + %p_dy_28 = arith.constant 128 : i64 loc(#loc78) + %p_dy_29 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc78) + %p_dy_30 = arith.muli %p_dy, %p_dy_29 : tensor<8x1xi64> loc(#loc78) + %p_dy_31 = tt.splat %dy : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc79) + %p_dy_32 = tt.addptr %p_dy_31, %p_dy_30 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc79) + %p_dy_33 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc80) + %p_dy_34 = tt.broadcast %p_dy_32 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc81) + %p_dy_35 = tt.broadcast %p_dy_33 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc81) + %p_dy_36 = tt.addptr %p_dy_34, %p_dy_35 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc81) + %p_dx = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc82) + %p_dx_37 = arith.constant 128 : i32 loc(#loc83) + %p_dx_38 = arith.constant 128 : i64 loc(#loc83) + %p_dx_39 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc83) + %p_dx_40 = arith.muli %p_dx, %p_dx_39 : tensor<8x1xi64> loc(#loc83) + %p_dx_41 = tt.splat %dx : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc84) + %p_dx_42 = tt.addptr %p_dx_41, %p_dx_40 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc84) + %p_dx_43 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc85) + %p_dx_44 = tt.broadcast %p_dx_42 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc86) + %p_dx_45 = tt.broadcast %p_dx_43 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc86) + %p_dx_46 = tt.addptr %p_dx_44, %p_dx_45 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc86) + %b_y = arith.constant 0.000000e+00 : f32 loc(#loc87) + %b_y_47 = arith.constant dense<0.000000e+00> : tensor<8x128xf32> loc(#loc87) + %b_y_48 = arith.truncf %b_y_47 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc87) + %b_y_49 = tt.load %p_y_25, %m_x_15, %b_y_48 : tensor<8x128x!tt.ptr> loc(#loc87) + %b_y_50 = arith.extf %b_y_49 : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc88) + %b_rstd = arith.constant 0.000000e+00 : f32 loc(#loc89) + %b_rstd_51 = arith.constant dense<0.000000e+00> : tensor<8xf32> loc(#loc89) + %b_rstd_52 = tt.load %p_rstd_26, %m_t_8, %b_rstd_51 : tensor<8x!tt.ptr> loc(#loc89) + %b_dy = arith.constant 0.000000e+00 : f32 loc(#loc90) + %b_dy_53 = arith.constant dense<0.000000e+00> : tensor<8x128xf32> loc(#loc90) + %b_dy_54 = arith.truncf %b_dy_53 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc90) + %b_dy_55 = tt.load %p_dy_36, %m_x_15, %b_dy_54 : tensor<8x128x!tt.ptr> loc(#loc90) + %b_dy_56 = arith.extf %b_dy_55 : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc91) + %b_dx = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc92) + %b_dx_57 = tt.broadcast %b_dx : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc93) + %b_dx_58 = arith.mulf %b_dy_56, %b_dx_57 : tensor<8x128xf32> loc(#loc93) + %b_dx_59 = arith.mulf %b_dy_56, %b_y_50 : tensor<8x128xf32> loc(#loc94) + %b_dx_60 = tt.call @triton.language.standard.sum__fp32S8_128S_c1_cFalse_cNone(%b_dx_59) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc95) + %b_dx_61 = tt.expand_dims %b_dx_60 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc96) + %b_dx_62 = tt.broadcast %b_dx_61 : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc97) + %b_dx_63 = arith.mulf %b_dx_62, %b_y_50 : tensor<8x128xf32> loc(#loc97) + %b_dx_64 = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc98) + %b_dx_65 = tt.broadcast %b_dx_64 : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc99) + %b_dx_66 = arith.mulf %b_dx_63, %b_dx_65 : tensor<8x128xf32> loc(#loc99) + %b_dx_67 = arith.subf %b_dx_58, %b_dx_66 : tensor<8x128xf32> loc(#loc100) + %0 = arith.truncf %b_dx_67 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc42) + tt.store %p_dx_46, %0, %m_x_15 : tensor<8x128x!tt.ptr> loc(#loc43) + tt.return loc(#loc44) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S8_128S_c1_cFalse_cNone(%input: tensor<8x128xf32> loc("input"(#loc45))) -> tensor<8xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc46) + tt.reduce.return %2 : f32 loc(#loc46) + }) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc46) + tt.return %0 : tensor<8xf32> loc(#loc48) + ^bb1: // no predecessors + %1 = ub.poison : tensor<8xf32> loc(#loc49) + tt.return %1 : tensor<8xf32> loc(#loc49) + } loc(#loc45) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc50)), %b: f32 loc("b"(#loc50))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc51) + tt.return %0 : f32 loc(#loc52) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc53) + tt.return %1 : f32 loc(#loc53) + } loc(#loc50) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:31) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:39) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:20) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:31) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:39) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:82) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc60 = loc("i_t"(#loc1)) +#loc61 = loc("i_t"(#loc2)) +#loc62 = loc("o_t"(#loc3)) +#loc63 = loc("o_t"(#loc4)) +#loc64 = loc("o_t"(#loc5)) +#loc65 = loc("o_d"(#loc6)) +#loc66 = loc("m_t"(#loc7)) +#loc67 = loc("m_x"(#loc8)) +#loc68 = loc("m_x"(#loc9)) +#loc69 = loc("m_x"(#loc10)) +#loc70 = loc("m_x"(#loc11)) +#loc71 = loc("p_y"(#loc12)) +#loc72 = loc("p_y"(#loc13)) +#loc73 = loc("p_y"(#loc14)) +#loc74 = loc("p_y"(#loc15)) +#loc75 = loc("p_y"(#loc16)) +#loc76 = loc("p_rstd"(#loc17)) +#loc77 = loc("p_dy"(#loc18)) +#loc78 = loc("p_dy"(#loc19)) +#loc79 = loc("p_dy"(#loc20)) +#loc80 = loc("p_dy"(#loc21)) +#loc81 = loc("p_dy"(#loc22)) +#loc82 = loc("p_dx"(#loc23)) +#loc83 = loc("p_dx"(#loc24)) +#loc84 = loc("p_dx"(#loc25)) +#loc85 = loc("p_dx"(#loc26)) +#loc86 = loc("p_dx"(#loc27)) +#loc87 = loc("b_y"(#loc28)) +#loc88 = loc("b_y"(#loc29)) +#loc89 = loc("b_rstd"(#loc30)) +#loc90 = loc("b_dy"(#loc31)) +#loc91 = loc("b_dy"(#loc32)) +#loc92 = loc("b_dx"(#loc33)) +#loc93 = loc("b_dx"(#loc34)) +#loc94 = loc("b_dx"(#loc35)) +#loc95 = loc("b_dx"(#loc36)) +#loc96 = loc("b_dx"(#loc37)) +#loc97 = loc("b_dx"(#loc38)) +#loc98 = loc("b_dx"(#loc39)) +#loc99 = loc("b_dx"(#loc40)) +#loc100 = loc("b_dx"(#loc41)) diff --git a/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/l2norm_bwd_kernel.ttgir b/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/l2norm_bwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..95c18a05dd69adbd782c62f571689185de38b58f --- /dev/null +++ b/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/l2norm_bwd_kernel.ttgir @@ -0,0 +1,146 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [2, 1], order = [1, 0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc39 = loc("y"(#loc)) +#loc40 = loc("rstd"(#loc)) +#loc41 = loc("dy"(#loc)) +#loc42 = loc("dx"(#loc)) +#loc43 = loc("eps"(#loc)) +#loc44 = loc("T"(#loc)) +#loc72 = loc("b_dx"(#loc30)) +#loc78 = loc(callsite(#loc1 at #loc72)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 2 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %c8_i64 = arith.constant 8 : i64 loc(#loc1) + %cst = arith.constant dense<0.000000e+00> : tensor<8x128xbf16, #blocked> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<8x1xi64, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc45) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc46) + %o_t = arith.muli %i_t_3, %c8_i64 : i64 loc(#loc47) + %o_t_4 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc48) + %o_t_5 = arith.extsi %o_t_4 : tensor<8xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %o_t_6 = tt.splat %o_t : i64 -> tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %o_t_7 = arith.addi %o_t_6, %o_t_5 : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49) + %m_t = arith.extsi %T : i32 to i64 loc(#loc50) + %m_t_8 = tt.splat %m_t : i64 -> tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc50) + %m_t_9 = arith.cmpi slt, %o_t_7, %m_t_8 : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc50) + %m_x = tt.expand_dims %m_t_9 {axis = 1 : i32} : tensor<8xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xi1, #blocked> loc(#loc51) + %m_x_10 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc52) + %m_x_11 = tt.expand_dims %m_x_10 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc52) + %m_x_12 = arith.cmpi slt, %m_x_11, %cst_2 : tensor<1x128xi32, #blocked> loc(#loc53) + %m_x_13 = tt.broadcast %m_x : tensor<8x1xi1, #blocked> -> tensor<8x128xi1, #blocked> loc(#loc54) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1, #blocked> -> tensor<8x128xi1, #blocked> loc(#loc54) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<8x128xi1, #blocked> loc(#loc54) + %p_y = tt.expand_dims %o_t_7 {axis = 1 : i32} : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xi64, #blocked> loc(#loc55) + %p_y_16 = arith.muli %p_y, %cst_1 : tensor<8x1xi64, #blocked> loc(#loc56) + %p_y_17 = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr, #blocked> loc(#loc57) + %p_y_18 = tt.addptr %p_y_17, %p_y_16 : tensor<8x1x!tt.ptr, #blocked>, tensor<8x1xi64, #blocked> loc(#loc57) + %p_y_19 = tt.broadcast %p_y_18 : tensor<8x1x!tt.ptr, #blocked> -> tensor<8x128x!tt.ptr, #blocked> loc(#loc58) + %p_y_20 = tt.broadcast %m_x_11 : tensor<1x128xi32, #blocked> -> tensor<8x128xi32, #blocked> loc(#loc58) + %p_y_21 = tt.addptr %p_y_19, %p_y_20 : tensor<8x128x!tt.ptr, #blocked>, tensor<8x128xi32, #blocked> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc59) + %p_rstd_22 = tt.addptr %p_rstd, %o_t_7 : tensor<8x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>>, tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc59) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<8x1x!tt.ptr, #blocked> loc(#loc60) + %p_dy_23 = tt.addptr %p_dy, %p_y_16 : tensor<8x1x!tt.ptr, #blocked>, tensor<8x1xi64, #blocked> loc(#loc60) + %p_dy_24 = tt.broadcast %p_dy_23 : tensor<8x1x!tt.ptr, #blocked> -> tensor<8x128x!tt.ptr, #blocked> loc(#loc61) + %p_dy_25 = tt.addptr %p_dy_24, %p_y_20 : tensor<8x128x!tt.ptr, #blocked>, tensor<8x128xi32, #blocked> loc(#loc61) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<8x1x!tt.ptr, #blocked> loc(#loc62) + %p_dx_26 = tt.addptr %p_dx, %p_y_16 : tensor<8x1x!tt.ptr, #blocked>, tensor<8x1xi64, #blocked> loc(#loc62) + %p_dx_27 = tt.broadcast %p_dx_26 : tensor<8x1x!tt.ptr, #blocked> -> tensor<8x128x!tt.ptr, #blocked> loc(#loc63) + %p_dx_28 = tt.addptr %p_dx_27, %p_y_20 : tensor<8x128x!tt.ptr, #blocked>, tensor<8x128xi32, #blocked> loc(#loc63) + %b_y = tt.load %p_y_21, %m_x_15, %cst : tensor<8x128x!tt.ptr, #blocked> loc(#loc64) + %b_y_29 = arith.extf %b_y : tensor<8x128xbf16, #blocked> to tensor<8x128xf32, #blocked> loc(#loc65) + %b_rstd = tt.load %p_rstd_22, %m_t_9, %cst_0 : tensor<8x!tt.ptr, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc66) + %b_dy = tt.load %p_dy_25, %m_x_15, %cst : tensor<8x128x!tt.ptr, #blocked> loc(#loc67) + %b_dy_30 = arith.extf %b_dy : tensor<8x128xbf16, #blocked> to tensor<8x128xf32, #blocked> loc(#loc68) + %b_dx = tt.expand_dims %b_rstd {axis = 1 : i32} : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xf32, #blocked> loc(#loc69) + %b_dx_31 = tt.broadcast %b_dx : tensor<8x1xf32, #blocked> -> tensor<8x128xf32, #blocked> loc(#loc70) + %b_dx_32 = arith.mulf %b_dy_30, %b_dx_31 : tensor<8x128xf32, #blocked> loc(#loc70) + %b_dx_33 = arith.mulf %b_dy_30, %b_y_29 : tensor<8x128xf32, #blocked> loc(#loc71) + %b_dx_34 = "tt.reduce"(%b_dx_33) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_40: f32 loc(callsite(#loc1 at #loc72)), %b_dx_41: f32 loc(callsite(#loc1 at #loc72))): + %b_dx_42 = arith.addf %b_dx_40, %b_dx_41 : f32 loc(#loc79) + tt.reduce.return %b_dx_42 : f32 loc(#loc77) + }) : (tensor<8x128xf32, #blocked>) -> tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc77) + %b_dx_35 = tt.expand_dims %b_dx_34 {axis = 1 : i32} : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xf32, #blocked> loc(#loc73) + %b_dx_36 = tt.broadcast %b_dx_35 : tensor<8x1xf32, #blocked> -> tensor<8x128xf32, #blocked> loc(#loc74) + %b_dx_37 = arith.mulf %b_dx_36, %b_y_29 : tensor<8x128xf32, #blocked> loc(#loc74) + %b_dx_38 = arith.mulf %b_dx_37, %b_dx_31 : tensor<8x128xf32, #blocked> loc(#loc75) + %b_dx_39 = arith.subf %b_dx_32, %b_dx_38 : tensor<8x128xf32, #blocked> loc(#loc76) + %0 = arith.truncf %b_dx_39 : tensor<8x128xf32, #blocked> to tensor<8x128xbf16, #blocked> loc(#loc36) + tt.store %p_dx_28, %0, %m_x_15 : tensor<8x128x!tt.ptr, #blocked> loc(#loc37) + tt.return loc(#loc38) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc45 = loc("i_t"(#loc2)) +#loc46 = loc("i_t"(#loc3)) +#loc47 = loc("o_t"(#loc4)) +#loc48 = loc("o_t"(#loc5)) +#loc49 = loc("o_t"(#loc6)) +#loc50 = loc("m_t"(#loc7)) +#loc51 = loc("m_x"(#loc8)) +#loc52 = loc("m_x"(#loc9)) +#loc53 = loc("m_x"(#loc10)) +#loc54 = loc("m_x"(#loc11)) +#loc55 = loc("p_y"(#loc12)) +#loc56 = loc("p_y"(#loc13)) +#loc57 = loc("p_y"(#loc14)) +#loc58 = loc("p_y"(#loc15)) +#loc59 = loc("p_rstd"(#loc16)) +#loc60 = loc("p_dy"(#loc17)) +#loc61 = loc("p_dy"(#loc18)) +#loc62 = loc("p_dx"(#loc19)) +#loc63 = loc("p_dx"(#loc20)) +#loc64 = loc("b_y"(#loc21)) +#loc65 = loc("b_y"(#loc22)) +#loc66 = loc("b_rstd"(#loc23)) +#loc67 = loc("b_dy"(#loc24)) +#loc68 = loc("b_dy"(#loc25)) +#loc69 = loc("b_dx"(#loc26)) +#loc70 = loc("b_dx"(#loc27)) +#loc71 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc32)) +#loc74 = loc("b_dx"(#loc33)) +#loc75 = loc("b_dx"(#loc34)) +#loc76 = loc("b_dx"(#loc35)) +#loc77 = loc(callsite(#loc29 at #loc72)) +#loc79 = loc(callsite(#loc31 at #loc77)) diff --git a/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/l2norm_bwd_kernel.ttir b/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/l2norm_bwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..c4761315a53ecb33821860b30f93bdae9fbfe731 --- /dev/null +++ b/triton/gpu0/CHSR6YHCXVSFLCEKTMWXKFVMZX7KVUYX7SNSQ7YWS2MWCEENGDEQ/l2norm_bwd_kernel.ttir @@ -0,0 +1,147 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc40 = loc("y"(#loc)) +#loc41 = loc("rstd"(#loc)) +#loc42 = loc("dy"(#loc)) +#loc43 = loc("dx"(#loc)) +#loc44 = loc("eps"(#loc)) +#loc45 = loc("T"(#loc)) +#loc74 = loc("b_dx"(#loc31)) +#loc80 = loc(callsite(#loc1 at #loc74)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<0.000000e+00> : tensor<8x128xbf16> loc(#loc1) + %b_rstd = arith.constant dense<0.000000e+00> : tensor<8xf32> loc(#loc46) + %cst_0 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc1) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc47) + %o_t = arith.constant 8 : i64 loc(#loc48) + %i_t = tt.get_program_id x : i32 loc(#loc49) + %i_t_1 = arith.extsi %i_t : i32 to i64 loc(#loc50) + %o_t_2 = arith.muli %i_t_1, %o_t : i64 loc(#loc48) + %o_t_3 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32> loc(#loc51) + %o_t_4 = arith.extsi %o_t_3 : tensor<8xi32> to tensor<8xi64> loc(#loc52) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<8xi64> loc(#loc52) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<8xi64> loc(#loc52) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc53) + %m_t = arith.extsi %T : i32 to i64 loc(#loc54) + %m_t_7 = tt.splat %m_t : i64 -> tensor<8xi64> loc(#loc54) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<8xi64> loc(#loc54) + %m_x_9 = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<8xi1> -> tensor<8x1xi1> loc(#loc55) + %m_x_10 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc56) + %m_x_11 = arith.cmpi slt, %m_x_10, %m_x : tensor<1x128xi32> loc(#loc47) + %m_x_12 = tt.broadcast %m_x_9 : tensor<8x1xi1> -> tensor<8x128xi1> loc(#loc57) + %m_x_13 = tt.broadcast %m_x_11 : tensor<1x128xi1> -> tensor<8x128xi1> loc(#loc57) + %m_x_14 = arith.andi %m_x_12, %m_x_13 : tensor<8x128xi1> loc(#loc57) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc58) + %p_y_15 = arith.muli %p_y, %cst_0 : tensor<8x1xi64> loc(#loc59) + %p_y_16 = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc60) + %p_y_17 = tt.addptr %p_y_16, %p_y_15 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc60) + %p_y_18 = tt.broadcast %p_y_17 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc61) + %p_y_19 = tt.broadcast %m_x_10 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc61) + %p_y_20 = tt.addptr %p_y_18, %p_y_19 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc61) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr> loc(#loc62) + %p_rstd_21 = tt.addptr %p_rstd, %o_t_6 : tensor<8x!tt.ptr>, tensor<8xi64> loc(#loc62) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc63) + %p_dy_22 = tt.addptr %p_dy, %p_y_15 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc63) + %p_dy_23 = tt.broadcast %p_dy_22 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc64) + %p_dy_24 = tt.addptr %p_dy_23, %p_y_19 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc64) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc65) + %p_dx_25 = tt.addptr %p_dx, %p_y_15 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc65) + %p_dx_26 = tt.broadcast %p_dx_25 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc66) + %p_dx_27 = tt.addptr %p_dx_26, %p_y_19 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc66) + %b_y = tt.load %p_y_20, %m_x_14, %cst : tensor<8x128x!tt.ptr> loc(#loc67) + %b_y_28 = arith.extf %b_y : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc68) + %b_rstd_29 = tt.load %p_rstd_21, %m_t_8, %b_rstd : tensor<8x!tt.ptr> loc(#loc46) + %b_dy = tt.load %p_dy_24, %m_x_14, %cst : tensor<8x128x!tt.ptr> loc(#loc69) + %b_dy_30 = arith.extf %b_dy : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc70) + %b_dx = tt.expand_dims %b_rstd_29 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc71) + %b_dx_31 = tt.broadcast %b_dx : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc72) + %b_dx_32 = arith.mulf %b_dy_30, %b_dx_31 : tensor<8x128xf32> loc(#loc72) + %b_dx_33 = arith.mulf %b_dy_30, %b_y_28 : tensor<8x128xf32> loc(#loc73) + %b_dx_34 = "tt.reduce"(%b_dx_33) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_40: f32 loc(callsite(#loc1 at #loc74)), %b_dx_41: f32 loc(callsite(#loc1 at #loc74))): + %b_dx_42 = arith.addf %b_dx_40, %b_dx_41 : f32 loc(#loc81) + tt.reduce.return %b_dx_42 : f32 loc(#loc79) + }) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc79) + %b_dx_35 = tt.expand_dims %b_dx_34 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc75) + %b_dx_36 = tt.broadcast %b_dx_35 : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc76) + %b_dx_37 = arith.mulf %b_dx_36, %b_y_28 : tensor<8x128xf32> loc(#loc76) + %b_dx_38 = arith.mulf %b_dx_37, %b_dx_31 : tensor<8x128xf32> loc(#loc77) + %b_dx_39 = arith.subf %b_dx_32, %b_dx_38 : tensor<8x128xf32> loc(#loc78) + %0 = arith.truncf %b_dx_39 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc37) + tt.store %p_dx_27, %0, %m_x_14 : tensor<8x128x!tt.ptr> loc(#loc38) + tt.return loc(#loc39) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("b_rstd"(#loc2)) +#loc47 = loc("m_x"(#loc3)) +#loc48 = loc("o_t"(#loc4)) +#loc49 = loc("i_t"(#loc5)) +#loc50 = loc("i_t"(#loc6)) +#loc51 = loc("o_t"(#loc7)) +#loc52 = loc("o_t"(#loc8)) +#loc53 = loc("o_d"(#loc9)) +#loc54 = loc("m_t"(#loc10)) +#loc55 = loc("m_x"(#loc11)) +#loc56 = loc("m_x"(#loc12)) +#loc57 = loc("m_x"(#loc13)) +#loc58 = loc("p_y"(#loc14)) +#loc59 = loc("p_y"(#loc15)) +#loc60 = loc("p_y"(#loc16)) +#loc61 = loc("p_y"(#loc17)) +#loc62 = loc("p_rstd"(#loc18)) +#loc63 = loc("p_dy"(#loc19)) +#loc64 = loc("p_dy"(#loc20)) +#loc65 = loc("p_dx"(#loc21)) +#loc66 = loc("p_dx"(#loc22)) +#loc67 = loc("b_y"(#loc23)) +#loc68 = loc("b_y"(#loc24)) +#loc69 = loc("b_dy"(#loc25)) +#loc70 = loc("b_dy"(#loc26)) +#loc71 = loc("b_dx"(#loc27)) +#loc72 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc29)) +#loc75 = loc("b_dx"(#loc33)) +#loc76 = loc("b_dx"(#loc34)) +#loc77 = loc("b_dx"(#loc35)) +#loc78 = loc("b_dx"(#loc36)) +#loc79 = loc(callsite(#loc30 at #loc74)) +#loc81 = loc(callsite(#loc32 at #loc79)) diff --git a/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/__grp__l2norm_bwd_kernel.json b/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/__grp__l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..170adadf9a3f4a3198e574afcc4d29ae5879e43a --- /dev/null +++ b/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/__grp__l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_bwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.source", "l2norm_bwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.ttir", "l2norm_bwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.ttgir", "l2norm_bwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.llir", "l2norm_bwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.ptx", "l2norm_bwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.cubin", "l2norm_bwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.cubin b/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..8c48f2d2f80629e511f2f0a059e8d7e73a59c724 --- /dev/null +++ b/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.cubin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8533299dce1bd88dc3433f984ea3cd268806d21c64a06a7a38fbc0a7677e4518 +size 631384 diff --git a/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.json b/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..148fe442606a9327c2e4fcb6931cfe92674e8cc5 --- /dev/null +++ b/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.json @@ -0,0 +1 @@ +{"hash": "11ef5dcd158e1581e8935994d596b908b37f290829e050c6302bb3085f17377c", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 1, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 256, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_bwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.llir b/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..01ec7f43b3426709e1eefde20c10919fab06dac3 --- /dev/null +++ b/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.llir @@ -0,0 +1,9405 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_bwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, ptr addrspace(1) %3, float %4, i32 %5, ptr addrspace(1) readnone captures(none) %6, ptr addrspace(1) readnone captures(none) %7) local_unnamed_addr #0 !dbg !4 { + %9 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !12 + %10 = zext nneg i32 %9 to i64, !dbg !13 + %11 = shl nuw nsw i64 %10, 7, !dbg !14 + %12 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !15 + %13 = and i32 %12, 16, !dbg !15 + %.lobit = lshr exact i32 %13, 4, !dbg !15 + %14 = or disjoint i32 %.lobit, 2, !dbg !15 + %15 = or disjoint i32 %.lobit, 4, !dbg !15 + %16 = or disjoint i32 %.lobit, 6, !dbg !15 + %17 = shl nuw nsw i32 %12, 2, !dbg !15 + %18 = zext nneg i32 %.lobit to i64, !dbg !16 + %19 = zext nneg i32 %14 to i64, !dbg !16 + %20 = zext nneg i32 %15 to i64, !dbg !16 + %21 = zext nneg i32 %16 to i64, !dbg !16 + %22 = or disjoint i64 %11, %18, !dbg !16 + %23 = or disjoint i64 %11, %19, !dbg !16 + %24 = or disjoint i64 %11, %20, !dbg !16 + %25 = or disjoint i64 %11, %21, !dbg !16 + %26 = sext i32 %5 to i64, !dbg !17 + %27 = icmp slt i64 %22, %26, !dbg !17 + %28 = icmp slt i64 %23, %26, !dbg !17 + %29 = icmp slt i64 %24, %26, !dbg !17 + %30 = icmp slt i64 %25, %26, !dbg !17 + %31 = insertelement <32 x i32> poison, i32 %.lobit, i64 0, !dbg !15 + %32 = insertelement <16 x i32> poison, i32 %.lobit, i64 0, !dbg !15 + %33 = insertelement <8 x i32> poison, i32 %.lobit, i64 0, !dbg !15 + %34 = insertelement <4 x i32> poison, i32 %.lobit, i64 0, !dbg !15 + %35 = and i32 %17, 124, !dbg !15 + %36 = or disjoint i32 %35, 1, !dbg !15 + %37 = insertelement <2 x i32> poison, i32 %35, i64 0, !dbg !15 + %38 = insertelement <64 x i32> poison, i32 %36, i64 2, !dbg !16 + %39 = insertelement <64 x i32> %38, i32 %35, i64 3, !dbg !16 + %40 = shufflevector <32 x i32> %31, <32 x i32> poison, <64 x i32> , !dbg !16 + %41 = or disjoint <64 x i32> %40, , !dbg !16 + %42 = shufflevector <64 x i32> %39, <64 x i32> %41, <64 x i32> , !dbg !16 + %43 = shufflevector <16 x i32> %32, <16 x i32> poison, <64 x i32> , !dbg !16 + %44 = or disjoint <64 x i32> %43, , !dbg !16 + %45 = shufflevector <64 x i32> %42, <64 x i32> %44, <64 x i32> , !dbg !16 + %46 = shufflevector <8 x i32> %33, <8 x i32> poison, <64 x i32> , !dbg !16 + %47 = or disjoint <64 x i32> %46, , !dbg !16 + %48 = shufflevector <64 x i32> %45, <64 x i32> %47, <64 x i32> , !dbg !16 + %49 = shufflevector <4 x i32> %34, <4 x i32> poison, <64 x i32> , !dbg !16 + %50 = or disjoint <64 x i32> %49, , !dbg !16 + %51 = shufflevector <64 x i32> %48, <64 x i32> %50, <64 x i32> , !dbg !16 + %52 = shufflevector <2 x i32> %37, <2 x i32> poison, <64 x i32> , !dbg !16 + %53 = or disjoint <64 x i32> %52, , !dbg !16 + %54 = shufflevector <64 x i32> %53, <64 x i32> %51, <64 x i32> , !dbg !16 + %55 = zext <64 x i32> %54 to <64 x i64>, !dbg !16 + %56 = insertelement <64 x i64> poison, i64 %11, i64 0, !dbg !16 + %57 = shufflevector <64 x i64> %56, <64 x i64> poison, <64 x i32> zeroinitializer, !dbg !16 + %58 = or disjoint <64 x i64> %57, %55, !dbg !16 + %59 = insertelement <64 x i32> poison, i32 %5, i64 0, !dbg !17 + %60 = shufflevector <64 x i32> %59, <64 x i32> poison, <64 x i32> zeroinitializer, !dbg !17 + %61 = sext <64 x i32> %60 to <64 x i64>, !dbg !17 + %62 = icmp slt <64 x i64> %58, %61, !dbg !17 + %63 = and i32 %12, 15, !dbg !18 + %64 = shl nuw nsw i32 %63, 3, !dbg !18 + %65 = shl nuw nsw i64 %22, 7, !dbg !19 + %66 = shl nuw nsw i64 %23, 7, !dbg !19 + %67 = shl nuw nsw i64 %24, 7, !dbg !19 + %68 = shl nuw nsw i64 %25, 7, !dbg !19 + %69 = extractelement <64 x i64> %58, i64 63, !dbg !19 + %70 = shl nuw nsw i64 %69, 7, !dbg !19 + %71 = extractelement <64 x i64> %58, i64 62, !dbg !19 + %72 = shl nuw nsw i64 %71, 7, !dbg !19 + %73 = extractelement <64 x i64> %58, i64 61, !dbg !19 + %74 = shl nuw nsw i64 %73, 7, !dbg !19 + %75 = extractelement <64 x i64> %58, i64 60, !dbg !19 + %76 = shl nuw nsw i64 %75, 7, !dbg !19 + %77 = extractelement <64 x i64> %58, i64 59, !dbg !19 + %78 = shl nuw nsw i64 %77, 7, !dbg !19 + %79 = extractelement <64 x i64> %58, i64 58, !dbg !19 + %80 = shl nuw nsw i64 %79, 7, !dbg !19 + %81 = extractelement <64 x i64> %58, i64 57, !dbg !19 + %82 = shl nuw nsw i64 %81, 7, !dbg !19 + %83 = extractelement <64 x i64> %58, i64 56, !dbg !19 + %84 = shl nuw nsw i64 %83, 7, !dbg !19 + %85 = extractelement <64 x i64> %58, i64 55, !dbg !19 + %86 = shl nuw nsw i64 %85, 7, !dbg !19 + %87 = extractelement <64 x i64> %58, i64 54, !dbg !19 + %88 = shl nuw nsw i64 %87, 7, !dbg !19 + %89 = extractelement <64 x i64> %58, i64 53, !dbg !19 + %90 = shl nuw nsw i64 %89, 7, !dbg !19 + %91 = extractelement <64 x i64> %58, i64 52, !dbg !19 + %92 = shl nuw nsw i64 %91, 7, !dbg !19 + %93 = extractelement <64 x i64> %58, i64 51, !dbg !19 + %94 = shl nuw nsw i64 %93, 7, !dbg !19 + %95 = extractelement <64 x i64> %58, i64 50, !dbg !19 + %96 = shl nuw nsw i64 %95, 7, !dbg !19 + %97 = extractelement <64 x i64> %58, i64 49, !dbg !19 + %98 = shl nuw nsw i64 %97, 7, !dbg !19 + %99 = extractelement <64 x i64> %58, i64 48, !dbg !19 + %100 = shl nuw nsw i64 %99, 7, !dbg !19 + %101 = extractelement <64 x i64> %58, i64 47, !dbg !19 + %102 = shl nuw nsw i64 %101, 7, !dbg !19 + %103 = extractelement <64 x i64> %58, i64 46, !dbg !19 + %104 = shl nuw nsw i64 %103, 7, !dbg !19 + %105 = extractelement <64 x i64> %58, i64 45, !dbg !19 + %106 = shl nuw nsw i64 %105, 7, !dbg !19 + %107 = extractelement <64 x i64> %58, i64 44, !dbg !19 + %108 = shl nuw nsw i64 %107, 7, !dbg !19 + %109 = extractelement <64 x i64> %58, i64 43, !dbg !19 + %110 = shl nuw nsw i64 %109, 7, !dbg !19 + %111 = extractelement <64 x i64> %58, i64 42, !dbg !19 + %112 = shl nuw nsw i64 %111, 7, !dbg !19 + %113 = extractelement <64 x i64> %58, i64 41, !dbg !19 + %114 = shl nuw nsw i64 %113, 7, !dbg !19 + %115 = extractelement <64 x i64> %58, i64 40, !dbg !19 + %116 = shl nuw nsw i64 %115, 7, !dbg !19 + %117 = extractelement <64 x i64> %58, i64 39, !dbg !19 + %118 = shl nuw nsw i64 %117, 7, !dbg !19 + %119 = extractelement <64 x i64> %58, i64 38, !dbg !19 + %120 = shl nuw nsw i64 %119, 7, !dbg !19 + %121 = extractelement <64 x i64> %58, i64 37, !dbg !19 + %122 = shl nuw nsw i64 %121, 7, !dbg !19 + %123 = extractelement <64 x i64> %58, i64 36, !dbg !19 + %124 = shl nuw nsw i64 %123, 7, !dbg !19 + %125 = extractelement <64 x i64> %58, i64 35, !dbg !19 + %126 = shl nuw nsw i64 %125, 7, !dbg !19 + %127 = extractelement <64 x i64> %58, i64 34, !dbg !19 + %128 = shl nuw nsw i64 %127, 7, !dbg !19 + %129 = extractelement <64 x i64> %58, i64 33, !dbg !19 + %130 = shl nuw nsw i64 %129, 7, !dbg !19 + %131 = extractelement <64 x i64> %58, i64 32, !dbg !19 + %132 = shl nuw nsw i64 %131, 7, !dbg !19 + %133 = extractelement <64 x i64> %58, i64 31, !dbg !19 + %134 = shl nuw nsw i64 %133, 7, !dbg !19 + %135 = extractelement <64 x i64> %58, i64 30, !dbg !19 + %136 = shl nuw nsw i64 %135, 7, !dbg !19 + %137 = extractelement <64 x i64> %58, i64 29, !dbg !19 + %138 = shl nuw nsw i64 %137, 7, !dbg !19 + %139 = extractelement <64 x i64> %58, i64 28, !dbg !19 + %140 = shl nuw nsw i64 %139, 7, !dbg !19 + %141 = extractelement <64 x i64> %58, i64 27, !dbg !19 + %142 = shl nuw nsw i64 %141, 7, !dbg !19 + %143 = extractelement <64 x i64> %58, i64 26, !dbg !19 + %144 = shl nuw nsw i64 %143, 7, !dbg !19 + %145 = extractelement <64 x i64> %58, i64 25, !dbg !19 + %146 = shl nuw nsw i64 %145, 7, !dbg !19 + %147 = extractelement <64 x i64> %58, i64 24, !dbg !19 + %148 = shl nuw nsw i64 %147, 7, !dbg !19 + %149 = extractelement <64 x i64> %58, i64 23, !dbg !19 + %150 = shl nuw nsw i64 %149, 7, !dbg !19 + %151 = extractelement <64 x i64> %58, i64 22, !dbg !19 + %152 = shl nuw nsw i64 %151, 7, !dbg !19 + %153 = extractelement <64 x i64> %58, i64 21, !dbg !19 + %154 = shl nuw nsw i64 %153, 7, !dbg !19 + %155 = extractelement <64 x i64> %58, i64 20, !dbg !19 + %156 = shl nuw nsw i64 %155, 7, !dbg !19 + %157 = extractelement <64 x i64> %58, i64 19, !dbg !19 + %158 = shl nuw nsw i64 %157, 7, !dbg !19 + %159 = extractelement <64 x i64> %58, i64 18, !dbg !19 + %160 = shl nuw nsw i64 %159, 7, !dbg !19 + %161 = extractelement <64 x i64> %58, i64 17, !dbg !19 + %162 = shl nuw nsw i64 %161, 7, !dbg !19 + %163 = extractelement <64 x i64> %58, i64 16, !dbg !19 + %164 = shl nuw nsw i64 %163, 7, !dbg !19 + %165 = extractelement <64 x i64> %58, i64 15, !dbg !19 + %166 = shl nuw nsw i64 %165, 7, !dbg !19 + %167 = extractelement <64 x i64> %58, i64 14, !dbg !19 + %168 = shl nuw nsw i64 %167, 7, !dbg !19 + %169 = extractelement <64 x i64> %58, i64 13, !dbg !19 + %170 = shl nuw nsw i64 %169, 7, !dbg !19 + %171 = extractelement <64 x i64> %58, i64 12, !dbg !19 + %172 = shl nuw nsw i64 %171, 7, !dbg !19 + %173 = extractelement <64 x i64> %58, i64 11, !dbg !19 + %174 = shl nuw nsw i64 %173, 7, !dbg !19 + %175 = extractelement <64 x i64> %58, i64 10, !dbg !19 + %176 = shl nuw nsw i64 %175, 7, !dbg !19 + %177 = extractelement <64 x i64> %58, i64 9, !dbg !19 + %178 = shl nuw nsw i64 %177, 7, !dbg !19 + %179 = extractelement <64 x i64> %58, i64 8, !dbg !19 + %180 = shl nuw nsw i64 %179, 7, !dbg !19 + %181 = extractelement <64 x i64> %58, i64 7, !dbg !19 + %182 = shl nuw nsw i64 %181, 7, !dbg !19 + %183 = extractelement <64 x i64> %58, i64 6, !dbg !19 + %184 = shl nuw nsw i64 %183, 7, !dbg !19 + %185 = extractelement <64 x i64> %58, i64 5, !dbg !19 + %186 = shl nuw nsw i64 %185, 7, !dbg !19 + %187 = extractelement <64 x i64> %58, i64 4, !dbg !19 + %188 = shl nuw nsw i64 %187, 7, !dbg !19 + %189 = getelementptr bfloat, ptr addrspace(1) %0, i64 %65, !dbg !20 + %190 = getelementptr bfloat, ptr addrspace(1) %0, i64 %66, !dbg !20 + %191 = getelementptr bfloat, ptr addrspace(1) %0, i64 %67, !dbg !20 + %192 = getelementptr bfloat, ptr addrspace(1) %0, i64 %68, !dbg !20 + %193 = getelementptr bfloat, ptr addrspace(1) %0, i64 %70, !dbg !20 + %194 = getelementptr bfloat, ptr addrspace(1) %0, i64 %72, !dbg !20 + %195 = getelementptr bfloat, ptr addrspace(1) %0, i64 %74, !dbg !20 + %196 = getelementptr bfloat, ptr addrspace(1) %0, i64 %76, !dbg !20 + %197 = getelementptr bfloat, ptr addrspace(1) %0, i64 %78, !dbg !20 + %198 = getelementptr bfloat, ptr addrspace(1) %0, i64 %80, !dbg !20 + %199 = getelementptr bfloat, ptr addrspace(1) %0, i64 %82, !dbg !20 + %200 = getelementptr bfloat, ptr addrspace(1) %0, i64 %84, !dbg !20 + %201 = getelementptr bfloat, ptr addrspace(1) %0, i64 %86, !dbg !20 + %202 = getelementptr bfloat, ptr addrspace(1) %0, i64 %88, !dbg !20 + %203 = getelementptr bfloat, ptr addrspace(1) %0, i64 %90, !dbg !20 + %204 = getelementptr bfloat, ptr addrspace(1) %0, i64 %92, !dbg !20 + %205 = getelementptr bfloat, ptr addrspace(1) %0, i64 %94, !dbg !20 + %206 = getelementptr bfloat, ptr addrspace(1) %0, i64 %96, !dbg !20 + %207 = getelementptr bfloat, ptr addrspace(1) %0, i64 %98, !dbg !20 + %208 = getelementptr bfloat, ptr addrspace(1) %0, i64 %100, !dbg !20 + %209 = getelementptr bfloat, ptr addrspace(1) %0, i64 %102, !dbg !20 + %210 = getelementptr bfloat, ptr addrspace(1) %0, i64 %104, !dbg !20 + %211 = getelementptr bfloat, ptr addrspace(1) %0, i64 %106, !dbg !20 + %212 = getelementptr bfloat, ptr addrspace(1) %0, i64 %108, !dbg !20 + %213 = getelementptr bfloat, ptr addrspace(1) %0, i64 %110, !dbg !20 + %214 = getelementptr bfloat, ptr addrspace(1) %0, i64 %112, !dbg !20 + %215 = getelementptr bfloat, ptr addrspace(1) %0, i64 %114, !dbg !20 + %216 = getelementptr bfloat, ptr addrspace(1) %0, i64 %116, !dbg !20 + %217 = getelementptr bfloat, ptr addrspace(1) %0, i64 %118, !dbg !20 + %218 = getelementptr bfloat, ptr addrspace(1) %0, i64 %120, !dbg !20 + %219 = getelementptr bfloat, ptr addrspace(1) %0, i64 %122, !dbg !20 + %220 = getelementptr bfloat, ptr addrspace(1) %0, i64 %124, !dbg !20 + %221 = getelementptr bfloat, ptr addrspace(1) %0, i64 %126, !dbg !20 + %222 = getelementptr bfloat, ptr addrspace(1) %0, i64 %128, !dbg !20 + %223 = getelementptr bfloat, ptr addrspace(1) %0, i64 %130, !dbg !20 + %224 = getelementptr bfloat, ptr addrspace(1) %0, i64 %132, !dbg !20 + %225 = getelementptr bfloat, ptr addrspace(1) %0, i64 %134, !dbg !20 + %226 = getelementptr bfloat, ptr addrspace(1) %0, i64 %136, !dbg !20 + %227 = getelementptr bfloat, ptr addrspace(1) %0, i64 %138, !dbg !20 + %228 = getelementptr bfloat, ptr addrspace(1) %0, i64 %140, !dbg !20 + %229 = getelementptr bfloat, ptr addrspace(1) %0, i64 %142, !dbg !20 + %230 = getelementptr bfloat, ptr addrspace(1) %0, i64 %144, !dbg !20 + %231 = getelementptr bfloat, ptr addrspace(1) %0, i64 %146, !dbg !20 + %232 = getelementptr bfloat, ptr addrspace(1) %0, i64 %148, !dbg !20 + %233 = getelementptr bfloat, ptr addrspace(1) %0, i64 %150, !dbg !20 + %234 = getelementptr bfloat, ptr addrspace(1) %0, i64 %152, !dbg !20 + %235 = getelementptr bfloat, ptr addrspace(1) %0, i64 %154, !dbg !20 + %236 = getelementptr bfloat, ptr addrspace(1) %0, i64 %156, !dbg !20 + %237 = getelementptr bfloat, ptr addrspace(1) %0, i64 %158, !dbg !20 + %238 = getelementptr bfloat, ptr addrspace(1) %0, i64 %160, !dbg !20 + %239 = getelementptr bfloat, ptr addrspace(1) %0, i64 %162, !dbg !20 + %240 = getelementptr bfloat, ptr addrspace(1) %0, i64 %164, !dbg !20 + %241 = getelementptr bfloat, ptr addrspace(1) %0, i64 %166, !dbg !20 + %242 = getelementptr bfloat, ptr addrspace(1) %0, i64 %168, !dbg !20 + %243 = getelementptr bfloat, ptr addrspace(1) %0, i64 %170, !dbg !20 + %244 = getelementptr bfloat, ptr addrspace(1) %0, i64 %172, !dbg !20 + %245 = getelementptr bfloat, ptr addrspace(1) %0, i64 %174, !dbg !20 + %246 = getelementptr bfloat, ptr addrspace(1) %0, i64 %176, !dbg !20 + %247 = getelementptr bfloat, ptr addrspace(1) %0, i64 %178, !dbg !20 + %248 = getelementptr bfloat, ptr addrspace(1) %0, i64 %180, !dbg !20 + %249 = getelementptr bfloat, ptr addrspace(1) %0, i64 %182, !dbg !20 + %250 = getelementptr bfloat, ptr addrspace(1) %0, i64 %184, !dbg !20 + %251 = getelementptr bfloat, ptr addrspace(1) %0, i64 %186, !dbg !20 + %252 = getelementptr bfloat, ptr addrspace(1) %0, i64 %188, !dbg !20 + %253 = zext nneg i32 %64 to i64, !dbg !21 + %254 = getelementptr bfloat, ptr addrspace(1) %189, i64 %253, !dbg !21 + %255 = getelementptr bfloat, ptr addrspace(1) %190, i64 %253, !dbg !21 + %256 = getelementptr bfloat, ptr addrspace(1) %191, i64 %253, !dbg !21 + %257 = getelementptr bfloat, ptr addrspace(1) %192, i64 %253, !dbg !21 + %258 = getelementptr bfloat, ptr addrspace(1) %193, i64 %253, !dbg !21 + %259 = getelementptr bfloat, ptr addrspace(1) %194, i64 %253, !dbg !21 + %260 = getelementptr bfloat, ptr addrspace(1) %195, i64 %253, !dbg !21 + %261 = getelementptr bfloat, ptr addrspace(1) %196, i64 %253, !dbg !21 + %262 = getelementptr bfloat, ptr addrspace(1) %197, i64 %253, !dbg !21 + %263 = getelementptr bfloat, ptr addrspace(1) %198, i64 %253, !dbg !21 + %264 = getelementptr bfloat, ptr addrspace(1) %199, i64 %253, !dbg !21 + %265 = getelementptr bfloat, ptr addrspace(1) %200, i64 %253, !dbg !21 + %266 = getelementptr bfloat, ptr addrspace(1) %201, i64 %253, !dbg !21 + %267 = getelementptr bfloat, ptr addrspace(1) %202, i64 %253, !dbg !21 + %268 = getelementptr bfloat, ptr addrspace(1) %203, i64 %253, !dbg !21 + %269 = getelementptr bfloat, ptr addrspace(1) %204, i64 %253, !dbg !21 + %270 = getelementptr bfloat, ptr addrspace(1) %205, i64 %253, !dbg !21 + %271 = getelementptr bfloat, ptr addrspace(1) %206, i64 %253, !dbg !21 + %272 = getelementptr bfloat, ptr addrspace(1) %207, i64 %253, !dbg !21 + %273 = getelementptr bfloat, ptr addrspace(1) %208, i64 %253, !dbg !21 + %274 = getelementptr bfloat, ptr addrspace(1) %209, i64 %253, !dbg !21 + %275 = getelementptr bfloat, ptr addrspace(1) %210, i64 %253, !dbg !21 + %276 = getelementptr bfloat, ptr addrspace(1) %211, i64 %253, !dbg !21 + %277 = getelementptr bfloat, ptr addrspace(1) %212, i64 %253, !dbg !21 + %278 = getelementptr bfloat, ptr addrspace(1) %213, i64 %253, !dbg !21 + %279 = getelementptr bfloat, ptr addrspace(1) %214, i64 %253, !dbg !21 + %280 = getelementptr bfloat, ptr addrspace(1) %215, i64 %253, !dbg !21 + %281 = getelementptr bfloat, ptr addrspace(1) %216, i64 %253, !dbg !21 + %282 = getelementptr bfloat, ptr addrspace(1) %217, i64 %253, !dbg !21 + %283 = getelementptr bfloat, ptr addrspace(1) %218, i64 %253, !dbg !21 + %284 = getelementptr bfloat, ptr addrspace(1) %219, i64 %253, !dbg !21 + %285 = getelementptr bfloat, ptr addrspace(1) %220, i64 %253, !dbg !21 + %286 = getelementptr bfloat, ptr addrspace(1) %221, i64 %253, !dbg !21 + %287 = getelementptr bfloat, ptr addrspace(1) %222, i64 %253, !dbg !21 + %288 = getelementptr bfloat, ptr addrspace(1) %223, i64 %253, !dbg !21 + %289 = getelementptr bfloat, ptr addrspace(1) %224, i64 %253, !dbg !21 + %290 = getelementptr bfloat, ptr addrspace(1) %225, i64 %253, !dbg !21 + %291 = getelementptr bfloat, ptr addrspace(1) %226, i64 %253, !dbg !21 + %292 = getelementptr bfloat, ptr addrspace(1) %227, i64 %253, !dbg !21 + %293 = getelementptr bfloat, ptr addrspace(1) %228, i64 %253, !dbg !21 + %294 = getelementptr bfloat, ptr addrspace(1) %229, i64 %253, !dbg !21 + %295 = getelementptr bfloat, ptr addrspace(1) %230, i64 %253, !dbg !21 + %296 = getelementptr bfloat, ptr addrspace(1) %231, i64 %253, !dbg !21 + %297 = getelementptr bfloat, ptr addrspace(1) %232, i64 %253, !dbg !21 + %298 = getelementptr bfloat, ptr addrspace(1) %233, i64 %253, !dbg !21 + %299 = getelementptr bfloat, ptr addrspace(1) %234, i64 %253, !dbg !21 + %300 = getelementptr bfloat, ptr addrspace(1) %235, i64 %253, !dbg !21 + %301 = getelementptr bfloat, ptr addrspace(1) %236, i64 %253, !dbg !21 + %302 = getelementptr bfloat, ptr addrspace(1) %237, i64 %253, !dbg !21 + %303 = getelementptr bfloat, ptr addrspace(1) %238, i64 %253, !dbg !21 + %304 = getelementptr bfloat, ptr addrspace(1) %239, i64 %253, !dbg !21 + %305 = getelementptr bfloat, ptr addrspace(1) %240, i64 %253, !dbg !21 + %306 = getelementptr bfloat, ptr addrspace(1) %241, i64 %253, !dbg !21 + %307 = getelementptr bfloat, ptr addrspace(1) %242, i64 %253, !dbg !21 + %308 = getelementptr bfloat, ptr addrspace(1) %243, i64 %253, !dbg !21 + %309 = getelementptr bfloat, ptr addrspace(1) %244, i64 %253, !dbg !21 + %310 = getelementptr bfloat, ptr addrspace(1) %245, i64 %253, !dbg !21 + %311 = getelementptr bfloat, ptr addrspace(1) %246, i64 %253, !dbg !21 + %312 = getelementptr bfloat, ptr addrspace(1) %247, i64 %253, !dbg !21 + %313 = getelementptr bfloat, ptr addrspace(1) %248, i64 %253, !dbg !21 + %314 = getelementptr bfloat, ptr addrspace(1) %249, i64 %253, !dbg !21 + %315 = getelementptr bfloat, ptr addrspace(1) %250, i64 %253, !dbg !21 + %316 = getelementptr bfloat, ptr addrspace(1) %251, i64 %253, !dbg !21 + %317 = getelementptr bfloat, ptr addrspace(1) %252, i64 %253, !dbg !21 + %318 = extractelement <64 x i64> %58, i64 3, !dbg !22 + %319 = getelementptr float, ptr addrspace(1) %1, i64 %318, !dbg !22 + %320 = extractelement <64 x i64> %58, i64 2, !dbg !22 + %321 = getelementptr float, ptr addrspace(1) %1, i64 %320, !dbg !22 + %322 = extractelement <64 x i64> %58, i64 1, !dbg !22 + %323 = getelementptr float, ptr addrspace(1) %1, i64 %322, !dbg !22 + %324 = extractelement <64 x i64> %58, i64 0, !dbg !22 + %325 = getelementptr float, ptr addrspace(1) %1, i64 %324, !dbg !22 + %326 = getelementptr bfloat, ptr addrspace(1) %2, i64 %65, !dbg !23 + %327 = getelementptr bfloat, ptr addrspace(1) %2, i64 %66, !dbg !23 + %328 = getelementptr bfloat, ptr addrspace(1) %2, i64 %67, !dbg !23 + %329 = getelementptr bfloat, ptr addrspace(1) %2, i64 %68, !dbg !23 + %330 = getelementptr bfloat, ptr addrspace(1) %2, i64 %70, !dbg !23 + %331 = getelementptr bfloat, ptr addrspace(1) %2, i64 %72, !dbg !23 + %332 = getelementptr bfloat, ptr addrspace(1) %2, i64 %74, !dbg !23 + %333 = getelementptr bfloat, ptr addrspace(1) %2, i64 %76, !dbg !23 + %334 = getelementptr bfloat, ptr addrspace(1) %2, i64 %78, !dbg !23 + %335 = getelementptr bfloat, ptr addrspace(1) %2, i64 %80, !dbg !23 + %336 = getelementptr bfloat, ptr addrspace(1) %2, i64 %82, !dbg !23 + %337 = getelementptr bfloat, ptr addrspace(1) %2, i64 %84, !dbg !23 + %338 = getelementptr bfloat, ptr addrspace(1) %2, i64 %86, !dbg !23 + %339 = getelementptr bfloat, ptr addrspace(1) %2, i64 %88, !dbg !23 + %340 = getelementptr bfloat, ptr addrspace(1) %2, i64 %90, !dbg !23 + %341 = getelementptr bfloat, ptr addrspace(1) %2, i64 %92, !dbg !23 + %342 = getelementptr bfloat, ptr addrspace(1) %2, i64 %94, !dbg !23 + %343 = getelementptr bfloat, ptr addrspace(1) %2, i64 %96, !dbg !23 + %344 = getelementptr bfloat, ptr addrspace(1) %2, i64 %98, !dbg !23 + %345 = getelementptr bfloat, ptr addrspace(1) %2, i64 %100, !dbg !23 + %346 = getelementptr bfloat, ptr addrspace(1) %2, i64 %102, !dbg !23 + %347 = getelementptr bfloat, ptr addrspace(1) %2, i64 %104, !dbg !23 + %348 = getelementptr bfloat, ptr addrspace(1) %2, i64 %106, !dbg !23 + %349 = getelementptr bfloat, ptr addrspace(1) %2, i64 %108, !dbg !23 + %350 = getelementptr bfloat, ptr addrspace(1) %2, i64 %110, !dbg !23 + %351 = getelementptr bfloat, ptr addrspace(1) %2, i64 %112, !dbg !23 + %352 = getelementptr bfloat, ptr addrspace(1) %2, i64 %114, !dbg !23 + %353 = getelementptr bfloat, ptr addrspace(1) %2, i64 %116, !dbg !23 + %354 = getelementptr bfloat, ptr addrspace(1) %2, i64 %118, !dbg !23 + %355 = getelementptr bfloat, ptr addrspace(1) %2, i64 %120, !dbg !23 + %356 = getelementptr bfloat, ptr addrspace(1) %2, i64 %122, !dbg !23 + %357 = getelementptr bfloat, ptr addrspace(1) %2, i64 %124, !dbg !23 + %358 = getelementptr bfloat, ptr addrspace(1) %2, i64 %126, !dbg !23 + %359 = getelementptr bfloat, ptr addrspace(1) %2, i64 %128, !dbg !23 + %360 = getelementptr bfloat, ptr addrspace(1) %2, i64 %130, !dbg !23 + %361 = getelementptr bfloat, ptr addrspace(1) %2, i64 %132, !dbg !23 + %362 = getelementptr bfloat, ptr addrspace(1) %2, i64 %134, !dbg !23 + %363 = getelementptr bfloat, ptr addrspace(1) %2, i64 %136, !dbg !23 + %364 = getelementptr bfloat, ptr addrspace(1) %2, i64 %138, !dbg !23 + %365 = getelementptr bfloat, ptr addrspace(1) %2, i64 %140, !dbg !23 + %366 = getelementptr bfloat, ptr addrspace(1) %2, i64 %142, !dbg !23 + %367 = getelementptr bfloat, ptr addrspace(1) %2, i64 %144, !dbg !23 + %368 = getelementptr bfloat, ptr addrspace(1) %2, i64 %146, !dbg !23 + %369 = getelementptr bfloat, ptr addrspace(1) %2, i64 %148, !dbg !23 + %370 = getelementptr bfloat, ptr addrspace(1) %2, i64 %150, !dbg !23 + %371 = getelementptr bfloat, ptr addrspace(1) %2, i64 %152, !dbg !23 + %372 = getelementptr bfloat, ptr addrspace(1) %2, i64 %154, !dbg !23 + %373 = getelementptr bfloat, ptr addrspace(1) %2, i64 %156, !dbg !23 + %374 = getelementptr bfloat, ptr addrspace(1) %2, i64 %158, !dbg !23 + %375 = getelementptr bfloat, ptr addrspace(1) %2, i64 %160, !dbg !23 + %376 = getelementptr bfloat, ptr addrspace(1) %2, i64 %162, !dbg !23 + %377 = getelementptr bfloat, ptr addrspace(1) %2, i64 %164, !dbg !23 + %378 = getelementptr bfloat, ptr addrspace(1) %2, i64 %166, !dbg !23 + %379 = getelementptr bfloat, ptr addrspace(1) %2, i64 %168, !dbg !23 + %380 = getelementptr bfloat, ptr addrspace(1) %2, i64 %170, !dbg !23 + %381 = getelementptr bfloat, ptr addrspace(1) %2, i64 %172, !dbg !23 + %382 = getelementptr bfloat, ptr addrspace(1) %2, i64 %174, !dbg !23 + %383 = getelementptr bfloat, ptr addrspace(1) %2, i64 %176, !dbg !23 + %384 = getelementptr bfloat, ptr addrspace(1) %2, i64 %178, !dbg !23 + %385 = getelementptr bfloat, ptr addrspace(1) %2, i64 %180, !dbg !23 + %386 = getelementptr bfloat, ptr addrspace(1) %2, i64 %182, !dbg !23 + %387 = getelementptr bfloat, ptr addrspace(1) %2, i64 %184, !dbg !23 + %388 = getelementptr bfloat, ptr addrspace(1) %2, i64 %186, !dbg !23 + %389 = getelementptr bfloat, ptr addrspace(1) %2, i64 %188, !dbg !23 + %390 = getelementptr bfloat, ptr addrspace(1) %326, i64 %253, !dbg !24 + %391 = getelementptr bfloat, ptr addrspace(1) %327, i64 %253, !dbg !24 + %392 = getelementptr bfloat, ptr addrspace(1) %328, i64 %253, !dbg !24 + %393 = getelementptr bfloat, ptr addrspace(1) %329, i64 %253, !dbg !24 + %394 = getelementptr bfloat, ptr addrspace(1) %330, i64 %253, !dbg !24 + %395 = getelementptr bfloat, ptr addrspace(1) %331, i64 %253, !dbg !24 + %396 = getelementptr bfloat, ptr addrspace(1) %332, i64 %253, !dbg !24 + %397 = getelementptr bfloat, ptr addrspace(1) %333, i64 %253, !dbg !24 + %398 = getelementptr bfloat, ptr addrspace(1) %334, i64 %253, !dbg !24 + %399 = getelementptr bfloat, ptr addrspace(1) %335, i64 %253, !dbg !24 + %400 = getelementptr bfloat, ptr addrspace(1) %336, i64 %253, !dbg !24 + %401 = getelementptr bfloat, ptr addrspace(1) %337, i64 %253, !dbg !24 + %402 = getelementptr bfloat, ptr addrspace(1) %338, i64 %253, !dbg !24 + %403 = getelementptr bfloat, ptr addrspace(1) %339, i64 %253, !dbg !24 + %404 = getelementptr bfloat, ptr addrspace(1) %340, i64 %253, !dbg !24 + %405 = getelementptr bfloat, ptr addrspace(1) %341, i64 %253, !dbg !24 + %406 = getelementptr bfloat, ptr addrspace(1) %342, i64 %253, !dbg !24 + %407 = getelementptr bfloat, ptr addrspace(1) %343, i64 %253, !dbg !24 + %408 = getelementptr bfloat, ptr addrspace(1) %344, i64 %253, !dbg !24 + %409 = getelementptr bfloat, ptr addrspace(1) %345, i64 %253, !dbg !24 + %410 = getelementptr bfloat, ptr addrspace(1) %346, i64 %253, !dbg !24 + %411 = getelementptr bfloat, ptr addrspace(1) %347, i64 %253, !dbg !24 + %412 = getelementptr bfloat, ptr addrspace(1) %348, i64 %253, !dbg !24 + %413 = getelementptr bfloat, ptr addrspace(1) %349, i64 %253, !dbg !24 + %414 = getelementptr bfloat, ptr addrspace(1) %350, i64 %253, !dbg !24 + %415 = getelementptr bfloat, ptr addrspace(1) %351, i64 %253, !dbg !24 + %416 = getelementptr bfloat, ptr addrspace(1) %352, i64 %253, !dbg !24 + %417 = getelementptr bfloat, ptr addrspace(1) %353, i64 %253, !dbg !24 + %418 = getelementptr bfloat, ptr addrspace(1) %354, i64 %253, !dbg !24 + %419 = getelementptr bfloat, ptr addrspace(1) %355, i64 %253, !dbg !24 + %420 = getelementptr bfloat, ptr addrspace(1) %356, i64 %253, !dbg !24 + %421 = getelementptr bfloat, ptr addrspace(1) %357, i64 %253, !dbg !24 + %422 = getelementptr bfloat, ptr addrspace(1) %358, i64 %253, !dbg !24 + %423 = getelementptr bfloat, ptr addrspace(1) %359, i64 %253, !dbg !24 + %424 = getelementptr bfloat, ptr addrspace(1) %360, i64 %253, !dbg !24 + %425 = getelementptr bfloat, ptr addrspace(1) %361, i64 %253, !dbg !24 + %426 = getelementptr bfloat, ptr addrspace(1) %362, i64 %253, !dbg !24 + %427 = getelementptr bfloat, ptr addrspace(1) %363, i64 %253, !dbg !24 + %428 = getelementptr bfloat, ptr addrspace(1) %364, i64 %253, !dbg !24 + %429 = getelementptr bfloat, ptr addrspace(1) %365, i64 %253, !dbg !24 + %430 = getelementptr bfloat, ptr addrspace(1) %366, i64 %253, !dbg !24 + %431 = getelementptr bfloat, ptr addrspace(1) %367, i64 %253, !dbg !24 + %432 = getelementptr bfloat, ptr addrspace(1) %368, i64 %253, !dbg !24 + %433 = getelementptr bfloat, ptr addrspace(1) %369, i64 %253, !dbg !24 + %434 = getelementptr bfloat, ptr addrspace(1) %370, i64 %253, !dbg !24 + %435 = getelementptr bfloat, ptr addrspace(1) %371, i64 %253, !dbg !24 + %436 = getelementptr bfloat, ptr addrspace(1) %372, i64 %253, !dbg !24 + %437 = getelementptr bfloat, ptr addrspace(1) %373, i64 %253, !dbg !24 + %438 = getelementptr bfloat, ptr addrspace(1) %374, i64 %253, !dbg !24 + %439 = getelementptr bfloat, ptr addrspace(1) %375, i64 %253, !dbg !24 + %440 = getelementptr bfloat, ptr addrspace(1) %376, i64 %253, !dbg !24 + %441 = getelementptr bfloat, ptr addrspace(1) %377, i64 %253, !dbg !24 + %442 = getelementptr bfloat, ptr addrspace(1) %378, i64 %253, !dbg !24 + %443 = getelementptr bfloat, ptr addrspace(1) %379, i64 %253, !dbg !24 + %444 = getelementptr bfloat, ptr addrspace(1) %380, i64 %253, !dbg !24 + %445 = getelementptr bfloat, ptr addrspace(1) %381, i64 %253, !dbg !24 + %446 = getelementptr bfloat, ptr addrspace(1) %382, i64 %253, !dbg !24 + %447 = getelementptr bfloat, ptr addrspace(1) %383, i64 %253, !dbg !24 + %448 = getelementptr bfloat, ptr addrspace(1) %384, i64 %253, !dbg !24 + %449 = getelementptr bfloat, ptr addrspace(1) %385, i64 %253, !dbg !24 + %450 = getelementptr bfloat, ptr addrspace(1) %386, i64 %253, !dbg !24 + %451 = getelementptr bfloat, ptr addrspace(1) %387, i64 %253, !dbg !24 + %452 = getelementptr bfloat, ptr addrspace(1) %388, i64 %253, !dbg !24 + %453 = getelementptr bfloat, ptr addrspace(1) %389, i64 %253, !dbg !24 + %454 = getelementptr bfloat, ptr addrspace(1) %3, i64 %65, !dbg !25 + %455 = getelementptr bfloat, ptr addrspace(1) %3, i64 %66, !dbg !25 + %456 = getelementptr bfloat, ptr addrspace(1) %3, i64 %67, !dbg !25 + %457 = getelementptr bfloat, ptr addrspace(1) %3, i64 %68, !dbg !25 + %458 = getelementptr bfloat, ptr addrspace(1) %3, i64 %70, !dbg !25 + %459 = getelementptr bfloat, ptr addrspace(1) %3, i64 %72, !dbg !25 + %460 = getelementptr bfloat, ptr addrspace(1) %3, i64 %74, !dbg !25 + %461 = getelementptr bfloat, ptr addrspace(1) %3, i64 %76, !dbg !25 + %462 = getelementptr bfloat, ptr addrspace(1) %3, i64 %78, !dbg !25 + %463 = getelementptr bfloat, ptr addrspace(1) %3, i64 %80, !dbg !25 + %464 = getelementptr bfloat, ptr addrspace(1) %3, i64 %82, !dbg !25 + %465 = getelementptr bfloat, ptr addrspace(1) %3, i64 %84, !dbg !25 + %466 = getelementptr bfloat, ptr addrspace(1) %3, i64 %86, !dbg !25 + %467 = getelementptr bfloat, ptr addrspace(1) %3, i64 %88, !dbg !25 + %468 = getelementptr bfloat, ptr addrspace(1) %3, i64 %90, !dbg !25 + %469 = getelementptr bfloat, ptr addrspace(1) %3, i64 %92, !dbg !25 + %470 = getelementptr bfloat, ptr addrspace(1) %3, i64 %94, !dbg !25 + %471 = getelementptr bfloat, ptr addrspace(1) %3, i64 %96, !dbg !25 + %472 = getelementptr bfloat, ptr addrspace(1) %3, i64 %98, !dbg !25 + %473 = getelementptr bfloat, ptr addrspace(1) %3, i64 %100, !dbg !25 + %474 = getelementptr bfloat, ptr addrspace(1) %3, i64 %102, !dbg !25 + %475 = getelementptr bfloat, ptr addrspace(1) %3, i64 %104, !dbg !25 + %476 = getelementptr bfloat, ptr addrspace(1) %3, i64 %106, !dbg !25 + %477 = getelementptr bfloat, ptr addrspace(1) %3, i64 %108, !dbg !25 + %478 = getelementptr bfloat, ptr addrspace(1) %3, i64 %110, !dbg !25 + %479 = getelementptr bfloat, ptr addrspace(1) %3, i64 %112, !dbg !25 + %480 = getelementptr bfloat, ptr addrspace(1) %3, i64 %114, !dbg !25 + %481 = getelementptr bfloat, ptr addrspace(1) %3, i64 %116, !dbg !25 + %482 = getelementptr bfloat, ptr addrspace(1) %3, i64 %118, !dbg !25 + %483 = getelementptr bfloat, ptr addrspace(1) %3, i64 %120, !dbg !25 + %484 = getelementptr bfloat, ptr addrspace(1) %3, i64 %122, !dbg !25 + %485 = getelementptr bfloat, ptr addrspace(1) %3, i64 %124, !dbg !25 + %486 = getelementptr bfloat, ptr addrspace(1) %3, i64 %126, !dbg !25 + %487 = getelementptr bfloat, ptr addrspace(1) %3, i64 %128, !dbg !25 + %488 = getelementptr bfloat, ptr addrspace(1) %3, i64 %130, !dbg !25 + %489 = getelementptr bfloat, ptr addrspace(1) %3, i64 %132, !dbg !25 + %490 = getelementptr bfloat, ptr addrspace(1) %3, i64 %134, !dbg !25 + %491 = getelementptr bfloat, ptr addrspace(1) %3, i64 %136, !dbg !25 + %492 = getelementptr bfloat, ptr addrspace(1) %3, i64 %138, !dbg !25 + %493 = getelementptr bfloat, ptr addrspace(1) %3, i64 %140, !dbg !25 + %494 = getelementptr bfloat, ptr addrspace(1) %3, i64 %142, !dbg !25 + %495 = getelementptr bfloat, ptr addrspace(1) %3, i64 %144, !dbg !25 + %496 = getelementptr bfloat, ptr addrspace(1) %3, i64 %146, !dbg !25 + %497 = getelementptr bfloat, ptr addrspace(1) %3, i64 %148, !dbg !25 + %498 = getelementptr bfloat, ptr addrspace(1) %3, i64 %150, !dbg !25 + %499 = getelementptr bfloat, ptr addrspace(1) %3, i64 %152, !dbg !25 + %500 = getelementptr bfloat, ptr addrspace(1) %3, i64 %154, !dbg !25 + %501 = getelementptr bfloat, ptr addrspace(1) %3, i64 %156, !dbg !25 + %502 = getelementptr bfloat, ptr addrspace(1) %3, i64 %158, !dbg !25 + %503 = getelementptr bfloat, ptr addrspace(1) %3, i64 %160, !dbg !25 + %504 = getelementptr bfloat, ptr addrspace(1) %3, i64 %162, !dbg !25 + %505 = getelementptr bfloat, ptr addrspace(1) %3, i64 %164, !dbg !25 + %506 = getelementptr bfloat, ptr addrspace(1) %3, i64 %166, !dbg !25 + %507 = getelementptr bfloat, ptr addrspace(1) %3, i64 %168, !dbg !25 + %508 = getelementptr bfloat, ptr addrspace(1) %3, i64 %170, !dbg !25 + %509 = getelementptr bfloat, ptr addrspace(1) %3, i64 %172, !dbg !25 + %510 = getelementptr bfloat, ptr addrspace(1) %3, i64 %174, !dbg !25 + %511 = getelementptr bfloat, ptr addrspace(1) %3, i64 %176, !dbg !25 + %512 = getelementptr bfloat, ptr addrspace(1) %3, i64 %178, !dbg !25 + %513 = getelementptr bfloat, ptr addrspace(1) %3, i64 %180, !dbg !25 + %514 = getelementptr bfloat, ptr addrspace(1) %3, i64 %182, !dbg !25 + %515 = getelementptr bfloat, ptr addrspace(1) %3, i64 %184, !dbg !25 + %516 = getelementptr bfloat, ptr addrspace(1) %3, i64 %186, !dbg !25 + %517 = getelementptr bfloat, ptr addrspace(1) %3, i64 %188, !dbg !25 + %518 = getelementptr bfloat, ptr addrspace(1) %454, i64 %253, !dbg !26 + %519 = getelementptr bfloat, ptr addrspace(1) %455, i64 %253, !dbg !26 + %520 = getelementptr bfloat, ptr addrspace(1) %456, i64 %253, !dbg !26 + %521 = getelementptr bfloat, ptr addrspace(1) %457, i64 %253, !dbg !26 + %522 = getelementptr bfloat, ptr addrspace(1) %458, i64 %253, !dbg !26 + %523 = getelementptr bfloat, ptr addrspace(1) %459, i64 %253, !dbg !26 + %524 = getelementptr bfloat, ptr addrspace(1) %460, i64 %253, !dbg !26 + %525 = getelementptr bfloat, ptr addrspace(1) %461, i64 %253, !dbg !26 + %526 = getelementptr bfloat, ptr addrspace(1) %462, i64 %253, !dbg !26 + %527 = getelementptr bfloat, ptr addrspace(1) %463, i64 %253, !dbg !26 + %528 = getelementptr bfloat, ptr addrspace(1) %464, i64 %253, !dbg !26 + %529 = getelementptr bfloat, ptr addrspace(1) %465, i64 %253, !dbg !26 + %530 = getelementptr bfloat, ptr addrspace(1) %466, i64 %253, !dbg !26 + %531 = getelementptr bfloat, ptr addrspace(1) %467, i64 %253, !dbg !26 + %532 = getelementptr bfloat, ptr addrspace(1) %468, i64 %253, !dbg !26 + %533 = getelementptr bfloat, ptr addrspace(1) %469, i64 %253, !dbg !26 + %534 = getelementptr bfloat, ptr addrspace(1) %470, i64 %253, !dbg !26 + %535 = getelementptr bfloat, ptr addrspace(1) %471, i64 %253, !dbg !26 + %536 = getelementptr bfloat, ptr addrspace(1) %472, i64 %253, !dbg !26 + %537 = getelementptr bfloat, ptr addrspace(1) %473, i64 %253, !dbg !26 + %538 = getelementptr bfloat, ptr addrspace(1) %474, i64 %253, !dbg !26 + %539 = getelementptr bfloat, ptr addrspace(1) %475, i64 %253, !dbg !26 + %540 = getelementptr bfloat, ptr addrspace(1) %476, i64 %253, !dbg !26 + %541 = getelementptr bfloat, ptr addrspace(1) %477, i64 %253, !dbg !26 + %542 = getelementptr bfloat, ptr addrspace(1) %478, i64 %253, !dbg !26 + %543 = getelementptr bfloat, ptr addrspace(1) %479, i64 %253, !dbg !26 + %544 = getelementptr bfloat, ptr addrspace(1) %480, i64 %253, !dbg !26 + %545 = getelementptr bfloat, ptr addrspace(1) %481, i64 %253, !dbg !26 + %546 = getelementptr bfloat, ptr addrspace(1) %482, i64 %253, !dbg !26 + %547 = getelementptr bfloat, ptr addrspace(1) %483, i64 %253, !dbg !26 + %548 = getelementptr bfloat, ptr addrspace(1) %484, i64 %253, !dbg !26 + %549 = getelementptr bfloat, ptr addrspace(1) %485, i64 %253, !dbg !26 + %550 = getelementptr bfloat, ptr addrspace(1) %486, i64 %253, !dbg !26 + %551 = getelementptr bfloat, ptr addrspace(1) %487, i64 %253, !dbg !26 + %552 = getelementptr bfloat, ptr addrspace(1) %488, i64 %253, !dbg !26 + %553 = getelementptr bfloat, ptr addrspace(1) %489, i64 %253, !dbg !26 + %554 = getelementptr bfloat, ptr addrspace(1) %490, i64 %253, !dbg !26 + %555 = getelementptr bfloat, ptr addrspace(1) %491, i64 %253, !dbg !26 + %556 = getelementptr bfloat, ptr addrspace(1) %492, i64 %253, !dbg !26 + %557 = getelementptr bfloat, ptr addrspace(1) %493, i64 %253, !dbg !26 + %558 = getelementptr bfloat, ptr addrspace(1) %494, i64 %253, !dbg !26 + %559 = getelementptr bfloat, ptr addrspace(1) %495, i64 %253, !dbg !26 + %560 = getelementptr bfloat, ptr addrspace(1) %496, i64 %253, !dbg !26 + %561 = getelementptr bfloat, ptr addrspace(1) %497, i64 %253, !dbg !26 + %562 = getelementptr bfloat, ptr addrspace(1) %498, i64 %253, !dbg !26 + %563 = getelementptr bfloat, ptr addrspace(1) %499, i64 %253, !dbg !26 + %564 = getelementptr bfloat, ptr addrspace(1) %500, i64 %253, !dbg !26 + %565 = getelementptr bfloat, ptr addrspace(1) %501, i64 %253, !dbg !26 + %566 = getelementptr bfloat, ptr addrspace(1) %502, i64 %253, !dbg !26 + %567 = getelementptr bfloat, ptr addrspace(1) %503, i64 %253, !dbg !26 + %568 = getelementptr bfloat, ptr addrspace(1) %504, i64 %253, !dbg !26 + %569 = getelementptr bfloat, ptr addrspace(1) %505, i64 %253, !dbg !26 + %570 = getelementptr bfloat, ptr addrspace(1) %506, i64 %253, !dbg !26 + %571 = getelementptr bfloat, ptr addrspace(1) %507, i64 %253, !dbg !26 + %572 = getelementptr bfloat, ptr addrspace(1) %508, i64 %253, !dbg !26 + %573 = getelementptr bfloat, ptr addrspace(1) %509, i64 %253, !dbg !26 + %574 = getelementptr bfloat, ptr addrspace(1) %510, i64 %253, !dbg !26 + %575 = getelementptr bfloat, ptr addrspace(1) %511, i64 %253, !dbg !26 + %576 = getelementptr bfloat, ptr addrspace(1) %512, i64 %253, !dbg !26 + %577 = getelementptr bfloat, ptr addrspace(1) %513, i64 %253, !dbg !26 + %578 = getelementptr bfloat, ptr addrspace(1) %514, i64 %253, !dbg !26 + %579 = getelementptr bfloat, ptr addrspace(1) %515, i64 %253, !dbg !26 + %580 = getelementptr bfloat, ptr addrspace(1) %516, i64 %253, !dbg !26 + %581 = getelementptr bfloat, ptr addrspace(1) %517, i64 %253, !dbg !26 + %582 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %254, i1 %27) #5, !dbg !27 + %583 = extractvalue { i32, i32, i32, i32 } %582, 0, !dbg !27 + %584 = bitcast i32 %583 to <2 x bfloat>, !dbg !27 + %585 = extractvalue { i32, i32, i32, i32 } %582, 1, !dbg !27 + %586 = bitcast i32 %585 to <2 x bfloat>, !dbg !27 + %587 = extractvalue { i32, i32, i32, i32 } %582, 2, !dbg !27 + %588 = bitcast i32 %587 to <2 x bfloat>, !dbg !27 + %589 = extractvalue { i32, i32, i32, i32 } %582, 3, !dbg !27 + %590 = bitcast i32 %589 to <2 x bfloat>, !dbg !27 + %591 = extractelement <2 x bfloat> %584, i64 0, !dbg !27 + %592 = extractelement <2 x bfloat> %584, i64 1, !dbg !27 + %593 = extractelement <2 x bfloat> %586, i64 0, !dbg !27 + %594 = extractelement <2 x bfloat> %586, i64 1, !dbg !27 + %595 = extractelement <2 x bfloat> %588, i64 0, !dbg !27 + %596 = extractelement <2 x bfloat> %588, i64 1, !dbg !27 + %597 = extractelement <2 x bfloat> %590, i64 0, !dbg !27 + %598 = extractelement <2 x bfloat> %590, i64 1, !dbg !27 + %599 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %255, i1 %28) #5, !dbg !27 + %600 = extractvalue { i32, i32, i32, i32 } %599, 0, !dbg !27 + %601 = bitcast i32 %600 to <2 x bfloat>, !dbg !27 + %602 = extractvalue { i32, i32, i32, i32 } %599, 1, !dbg !27 + %603 = bitcast i32 %602 to <2 x bfloat>, !dbg !27 + %604 = extractvalue { i32, i32, i32, i32 } %599, 2, !dbg !27 + %605 = bitcast i32 %604 to <2 x bfloat>, !dbg !27 + %606 = extractvalue { i32, i32, i32, i32 } %599, 3, !dbg !27 + %607 = bitcast i32 %606 to <2 x bfloat>, !dbg !27 + %608 = extractelement <2 x bfloat> %601, i64 0, !dbg !27 + %609 = extractelement <2 x bfloat> %601, i64 1, !dbg !27 + %610 = extractelement <2 x bfloat> %603, i64 0, !dbg !27 + %611 = extractelement <2 x bfloat> %603, i64 1, !dbg !27 + %612 = extractelement <2 x bfloat> %605, i64 0, !dbg !27 + %613 = extractelement <2 x bfloat> %605, i64 1, !dbg !27 + %614 = extractelement <2 x bfloat> %607, i64 0, !dbg !27 + %615 = extractelement <2 x bfloat> %607, i64 1, !dbg !27 + %616 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %256, i1 %29) #5, !dbg !27 + %617 = extractvalue { i32, i32, i32, i32 } %616, 0, !dbg !27 + %618 = bitcast i32 %617 to <2 x bfloat>, !dbg !27 + %619 = extractvalue { i32, i32, i32, i32 } %616, 1, !dbg !27 + %620 = bitcast i32 %619 to <2 x bfloat>, !dbg !27 + %621 = extractvalue { i32, i32, i32, i32 } %616, 2, !dbg !27 + %622 = bitcast i32 %621 to <2 x bfloat>, !dbg !27 + %623 = extractvalue { i32, i32, i32, i32 } %616, 3, !dbg !27 + %624 = bitcast i32 %623 to <2 x bfloat>, !dbg !27 + %625 = extractelement <2 x bfloat> %618, i64 0, !dbg !27 + %626 = extractelement <2 x bfloat> %618, i64 1, !dbg !27 + %627 = extractelement <2 x bfloat> %620, i64 0, !dbg !27 + %628 = extractelement <2 x bfloat> %620, i64 1, !dbg !27 + %629 = extractelement <2 x bfloat> %622, i64 0, !dbg !27 + %630 = extractelement <2 x bfloat> %622, i64 1, !dbg !27 + %631 = extractelement <2 x bfloat> %624, i64 0, !dbg !27 + %632 = extractelement <2 x bfloat> %624, i64 1, !dbg !27 + %633 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %257, i1 %30) #5, !dbg !27 + %634 = extractvalue { i32, i32, i32, i32 } %633, 0, !dbg !27 + %635 = bitcast i32 %634 to <2 x bfloat>, !dbg !27 + %636 = extractvalue { i32, i32, i32, i32 } %633, 1, !dbg !27 + %637 = bitcast i32 %636 to <2 x bfloat>, !dbg !27 + %638 = extractvalue { i32, i32, i32, i32 } %633, 2, !dbg !27 + %639 = bitcast i32 %638 to <2 x bfloat>, !dbg !27 + %640 = extractvalue { i32, i32, i32, i32 } %633, 3, !dbg !27 + %641 = bitcast i32 %640 to <2 x bfloat>, !dbg !27 + %642 = extractelement <2 x bfloat> %635, i64 0, !dbg !27 + %643 = extractelement <2 x bfloat> %635, i64 1, !dbg !27 + %644 = extractelement <2 x bfloat> %637, i64 0, !dbg !27 + %645 = extractelement <2 x bfloat> %637, i64 1, !dbg !27 + %646 = extractelement <2 x bfloat> %639, i64 0, !dbg !27 + %647 = extractelement <2 x bfloat> %639, i64 1, !dbg !27 + %648 = extractelement <2 x bfloat> %641, i64 0, !dbg !27 + %649 = extractelement <2 x bfloat> %641, i64 1, !dbg !27 + %650 = extractelement <64 x i1> %62, i64 63, !dbg !27 + %651 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %258, i1 %650) #5, !dbg !27 + %652 = extractvalue { i32, i32, i32, i32 } %651, 0, !dbg !27 + %653 = bitcast i32 %652 to <2 x bfloat>, !dbg !27 + %654 = extractvalue { i32, i32, i32, i32 } %651, 1, !dbg !27 + %655 = bitcast i32 %654 to <2 x bfloat>, !dbg !27 + %656 = extractvalue { i32, i32, i32, i32 } %651, 2, !dbg !27 + %657 = bitcast i32 %656 to <2 x bfloat>, !dbg !27 + %658 = extractvalue { i32, i32, i32, i32 } %651, 3, !dbg !27 + %659 = bitcast i32 %658 to <2 x bfloat>, !dbg !27 + %660 = extractelement <2 x bfloat> %653, i64 0, !dbg !27 + %661 = extractelement <2 x bfloat> %653, i64 1, !dbg !27 + %662 = extractelement <2 x bfloat> %655, i64 0, !dbg !27 + %663 = extractelement <2 x bfloat> %655, i64 1, !dbg !27 + %664 = extractelement <2 x bfloat> %657, i64 0, !dbg !27 + %665 = extractelement <2 x bfloat> %657, i64 1, !dbg !27 + %666 = extractelement <2 x bfloat> %659, i64 0, !dbg !27 + %667 = extractelement <2 x bfloat> %659, i64 1, !dbg !27 + %668 = extractelement <64 x i1> %62, i64 62, !dbg !27 + %669 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %259, i1 %668) #5, !dbg !27 + %670 = extractvalue { i32, i32, i32, i32 } %669, 0, !dbg !27 + %671 = bitcast i32 %670 to <2 x bfloat>, !dbg !27 + %672 = extractvalue { i32, i32, i32, i32 } %669, 1, !dbg !27 + %673 = bitcast i32 %672 to <2 x bfloat>, !dbg !27 + %674 = extractvalue { i32, i32, i32, i32 } %669, 2, !dbg !27 + %675 = bitcast i32 %674 to <2 x bfloat>, !dbg !27 + %676 = extractvalue { i32, i32, i32, i32 } %669, 3, !dbg !27 + %677 = bitcast i32 %676 to <2 x bfloat>, !dbg !27 + %678 = extractelement <2 x bfloat> %671, i64 0, !dbg !27 + %679 = extractelement <2 x bfloat> %671, i64 1, !dbg !27 + %680 = extractelement <2 x bfloat> %673, i64 0, !dbg !27 + %681 = extractelement <2 x bfloat> %673, i64 1, !dbg !27 + %682 = extractelement <2 x bfloat> %675, i64 0, !dbg !27 + %683 = extractelement <2 x bfloat> %675, i64 1, !dbg !27 + %684 = extractelement <2 x bfloat> %677, i64 0, !dbg !27 + %685 = extractelement <2 x bfloat> %677, i64 1, !dbg !27 + %686 = extractelement <64 x i1> %62, i64 61, !dbg !27 + %687 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %260, i1 %686) #5, !dbg !27 + %688 = extractvalue { i32, i32, i32, i32 } %687, 0, !dbg !27 + %689 = bitcast i32 %688 to <2 x bfloat>, !dbg !27 + %690 = extractvalue { i32, i32, i32, i32 } %687, 1, !dbg !27 + %691 = bitcast i32 %690 to <2 x bfloat>, !dbg !27 + %692 = extractvalue { i32, i32, i32, i32 } %687, 2, !dbg !27 + %693 = bitcast i32 %692 to <2 x bfloat>, !dbg !27 + %694 = extractvalue { i32, i32, i32, i32 } %687, 3, !dbg !27 + %695 = bitcast i32 %694 to <2 x bfloat>, !dbg !27 + %696 = extractelement <2 x bfloat> %689, i64 0, !dbg !27 + %697 = extractelement <2 x bfloat> %689, i64 1, !dbg !27 + %698 = extractelement <2 x bfloat> %691, i64 0, !dbg !27 + %699 = extractelement <2 x bfloat> %691, i64 1, !dbg !27 + %700 = extractelement <2 x bfloat> %693, i64 0, !dbg !27 + %701 = extractelement <2 x bfloat> %693, i64 1, !dbg !27 + %702 = extractelement <2 x bfloat> %695, i64 0, !dbg !27 + %703 = extractelement <2 x bfloat> %695, i64 1, !dbg !27 + %704 = extractelement <64 x i1> %62, i64 60, !dbg !27 + %705 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %261, i1 %704) #5, !dbg !27 + %706 = extractvalue { i32, i32, i32, i32 } %705, 0, !dbg !27 + %707 = bitcast i32 %706 to <2 x bfloat>, !dbg !27 + %708 = extractvalue { i32, i32, i32, i32 } %705, 1, !dbg !27 + %709 = bitcast i32 %708 to <2 x bfloat>, !dbg !27 + %710 = extractvalue { i32, i32, i32, i32 } %705, 2, !dbg !27 + %711 = bitcast i32 %710 to <2 x bfloat>, !dbg !27 + %712 = extractvalue { i32, i32, i32, i32 } %705, 3, !dbg !27 + %713 = bitcast i32 %712 to <2 x bfloat>, !dbg !27 + %714 = extractelement <2 x bfloat> %707, i64 0, !dbg !27 + %715 = extractelement <2 x bfloat> %707, i64 1, !dbg !27 + %716 = extractelement <2 x bfloat> %709, i64 0, !dbg !27 + %717 = extractelement <2 x bfloat> %709, i64 1, !dbg !27 + %718 = extractelement <2 x bfloat> %711, i64 0, !dbg !27 + %719 = extractelement <2 x bfloat> %711, i64 1, !dbg !27 + %720 = extractelement <2 x bfloat> %713, i64 0, !dbg !27 + %721 = extractelement <2 x bfloat> %713, i64 1, !dbg !27 + %722 = extractelement <64 x i1> %62, i64 59, !dbg !27 + %723 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %262, i1 %722) #5, !dbg !27 + %724 = extractvalue { i32, i32, i32, i32 } %723, 0, !dbg !27 + %725 = bitcast i32 %724 to <2 x bfloat>, !dbg !27 + %726 = extractvalue { i32, i32, i32, i32 } %723, 1, !dbg !27 + %727 = bitcast i32 %726 to <2 x bfloat>, !dbg !27 + %728 = extractvalue { i32, i32, i32, i32 } %723, 2, !dbg !27 + %729 = bitcast i32 %728 to <2 x bfloat>, !dbg !27 + %730 = extractvalue { i32, i32, i32, i32 } %723, 3, !dbg !27 + %731 = bitcast i32 %730 to <2 x bfloat>, !dbg !27 + %732 = extractelement <2 x bfloat> %725, i64 0, !dbg !27 + %733 = extractelement <2 x bfloat> %725, i64 1, !dbg !27 + %734 = extractelement <2 x bfloat> %727, i64 0, !dbg !27 + %735 = extractelement <2 x bfloat> %727, i64 1, !dbg !27 + %736 = extractelement <2 x bfloat> %729, i64 0, !dbg !27 + %737 = extractelement <2 x bfloat> %729, i64 1, !dbg !27 + %738 = extractelement <2 x bfloat> %731, i64 0, !dbg !27 + %739 = extractelement <2 x bfloat> %731, i64 1, !dbg !27 + %740 = extractelement <64 x i1> %62, i64 58, !dbg !27 + %741 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %263, i1 %740) #5, !dbg !27 + %742 = extractvalue { i32, i32, i32, i32 } %741, 0, !dbg !27 + %743 = bitcast i32 %742 to <2 x bfloat>, !dbg !27 + %744 = extractvalue { i32, i32, i32, i32 } %741, 1, !dbg !27 + %745 = bitcast i32 %744 to <2 x bfloat>, !dbg !27 + %746 = extractvalue { i32, i32, i32, i32 } %741, 2, !dbg !27 + %747 = bitcast i32 %746 to <2 x bfloat>, !dbg !27 + %748 = extractvalue { i32, i32, i32, i32 } %741, 3, !dbg !27 + %749 = bitcast i32 %748 to <2 x bfloat>, !dbg !27 + %750 = extractelement <2 x bfloat> %743, i64 0, !dbg !27 + %751 = extractelement <2 x bfloat> %743, i64 1, !dbg !27 + %752 = extractelement <2 x bfloat> %745, i64 0, !dbg !27 + %753 = extractelement <2 x bfloat> %745, i64 1, !dbg !27 + %754 = extractelement <2 x bfloat> %747, i64 0, !dbg !27 + %755 = extractelement <2 x bfloat> %747, i64 1, !dbg !27 + %756 = extractelement <2 x bfloat> %749, i64 0, !dbg !27 + %757 = extractelement <2 x bfloat> %749, i64 1, !dbg !27 + %758 = extractelement <64 x i1> %62, i64 57, !dbg !27 + %759 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %264, i1 %758) #5, !dbg !27 + %760 = extractvalue { i32, i32, i32, i32 } %759, 0, !dbg !27 + %761 = bitcast i32 %760 to <2 x bfloat>, !dbg !27 + %762 = extractvalue { i32, i32, i32, i32 } %759, 1, !dbg !27 + %763 = bitcast i32 %762 to <2 x bfloat>, !dbg !27 + %764 = extractvalue { i32, i32, i32, i32 } %759, 2, !dbg !27 + %765 = bitcast i32 %764 to <2 x bfloat>, !dbg !27 + %766 = extractvalue { i32, i32, i32, i32 } %759, 3, !dbg !27 + %767 = bitcast i32 %766 to <2 x bfloat>, !dbg !27 + %768 = extractelement <2 x bfloat> %761, i64 0, !dbg !27 + %769 = extractelement <2 x bfloat> %761, i64 1, !dbg !27 + %770 = extractelement <2 x bfloat> %763, i64 0, !dbg !27 + %771 = extractelement <2 x bfloat> %763, i64 1, !dbg !27 + %772 = extractelement <2 x bfloat> %765, i64 0, !dbg !27 + %773 = extractelement <2 x bfloat> %765, i64 1, !dbg !27 + %774 = extractelement <2 x bfloat> %767, i64 0, !dbg !27 + %775 = extractelement <2 x bfloat> %767, i64 1, !dbg !27 + %776 = extractelement <64 x i1> %62, i64 56, !dbg !27 + %777 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %265, i1 %776) #5, !dbg !27 + %778 = extractvalue { i32, i32, i32, i32 } %777, 0, !dbg !27 + %779 = bitcast i32 %778 to <2 x bfloat>, !dbg !27 + %780 = extractvalue { i32, i32, i32, i32 } %777, 1, !dbg !27 + %781 = bitcast i32 %780 to <2 x bfloat>, !dbg !27 + %782 = extractvalue { i32, i32, i32, i32 } %777, 2, !dbg !27 + %783 = bitcast i32 %782 to <2 x bfloat>, !dbg !27 + %784 = extractvalue { i32, i32, i32, i32 } %777, 3, !dbg !27 + %785 = bitcast i32 %784 to <2 x bfloat>, !dbg !27 + %786 = extractelement <2 x bfloat> %779, i64 0, !dbg !27 + %787 = extractelement <2 x bfloat> %779, i64 1, !dbg !27 + %788 = extractelement <2 x bfloat> %781, i64 0, !dbg !27 + %789 = extractelement <2 x bfloat> %781, i64 1, !dbg !27 + %790 = extractelement <2 x bfloat> %783, i64 0, !dbg !27 + %791 = extractelement <2 x bfloat> %783, i64 1, !dbg !27 + %792 = extractelement <2 x bfloat> %785, i64 0, !dbg !27 + %793 = extractelement <2 x bfloat> %785, i64 1, !dbg !27 + %794 = extractelement <64 x i1> %62, i64 55, !dbg !27 + %795 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %266, i1 %794) #5, !dbg !27 + %796 = extractvalue { i32, i32, i32, i32 } %795, 0, !dbg !27 + %797 = bitcast i32 %796 to <2 x bfloat>, !dbg !27 + %798 = extractvalue { i32, i32, i32, i32 } %795, 1, !dbg !27 + %799 = bitcast i32 %798 to <2 x bfloat>, !dbg !27 + %800 = extractvalue { i32, i32, i32, i32 } %795, 2, !dbg !27 + %801 = bitcast i32 %800 to <2 x bfloat>, !dbg !27 + %802 = extractvalue { i32, i32, i32, i32 } %795, 3, !dbg !27 + %803 = bitcast i32 %802 to <2 x bfloat>, !dbg !27 + %804 = extractelement <2 x bfloat> %797, i64 0, !dbg !27 + %805 = extractelement <2 x bfloat> %797, i64 1, !dbg !27 + %806 = extractelement <2 x bfloat> %799, i64 0, !dbg !27 + %807 = extractelement <2 x bfloat> %799, i64 1, !dbg !27 + %808 = extractelement <2 x bfloat> %801, i64 0, !dbg !27 + %809 = extractelement <2 x bfloat> %801, i64 1, !dbg !27 + %810 = extractelement <2 x bfloat> %803, i64 0, !dbg !27 + %811 = extractelement <2 x bfloat> %803, i64 1, !dbg !27 + %812 = extractelement <64 x i1> %62, i64 54, !dbg !27 + %813 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %267, i1 %812) #5, !dbg !27 + %814 = extractvalue { i32, i32, i32, i32 } %813, 0, !dbg !27 + %815 = bitcast i32 %814 to <2 x bfloat>, !dbg !27 + %816 = extractvalue { i32, i32, i32, i32 } %813, 1, !dbg !27 + %817 = bitcast i32 %816 to <2 x bfloat>, !dbg !27 + %818 = extractvalue { i32, i32, i32, i32 } %813, 2, !dbg !27 + %819 = bitcast i32 %818 to <2 x bfloat>, !dbg !27 + %820 = extractvalue { i32, i32, i32, i32 } %813, 3, !dbg !27 + %821 = bitcast i32 %820 to <2 x bfloat>, !dbg !27 + %822 = extractelement <2 x bfloat> %815, i64 0, !dbg !27 + %823 = extractelement <2 x bfloat> %815, i64 1, !dbg !27 + %824 = extractelement <2 x bfloat> %817, i64 0, !dbg !27 + %825 = extractelement <2 x bfloat> %817, i64 1, !dbg !27 + %826 = extractelement <2 x bfloat> %819, i64 0, !dbg !27 + %827 = extractelement <2 x bfloat> %819, i64 1, !dbg !27 + %828 = extractelement <2 x bfloat> %821, i64 0, !dbg !27 + %829 = extractelement <2 x bfloat> %821, i64 1, !dbg !27 + %830 = extractelement <64 x i1> %62, i64 53, !dbg !27 + %831 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %268, i1 %830) #5, !dbg !27 + %832 = extractvalue { i32, i32, i32, i32 } %831, 0, !dbg !27 + %833 = bitcast i32 %832 to <2 x bfloat>, !dbg !27 + %834 = extractvalue { i32, i32, i32, i32 } %831, 1, !dbg !27 + %835 = bitcast i32 %834 to <2 x bfloat>, !dbg !27 + %836 = extractvalue { i32, i32, i32, i32 } %831, 2, !dbg !27 + %837 = bitcast i32 %836 to <2 x bfloat>, !dbg !27 + %838 = extractvalue { i32, i32, i32, i32 } %831, 3, !dbg !27 + %839 = bitcast i32 %838 to <2 x bfloat>, !dbg !27 + %840 = extractelement <2 x bfloat> %833, i64 0, !dbg !27 + %841 = extractelement <2 x bfloat> %833, i64 1, !dbg !27 + %842 = extractelement <2 x bfloat> %835, i64 0, !dbg !27 + %843 = extractelement <2 x bfloat> %835, i64 1, !dbg !27 + %844 = extractelement <2 x bfloat> %837, i64 0, !dbg !27 + %845 = extractelement <2 x bfloat> %837, i64 1, !dbg !27 + %846 = extractelement <2 x bfloat> %839, i64 0, !dbg !27 + %847 = extractelement <2 x bfloat> %839, i64 1, !dbg !27 + %848 = extractelement <64 x i1> %62, i64 52, !dbg !27 + %849 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %269, i1 %848) #5, !dbg !27 + %850 = extractvalue { i32, i32, i32, i32 } %849, 0, !dbg !27 + %851 = bitcast i32 %850 to <2 x bfloat>, !dbg !27 + %852 = extractvalue { i32, i32, i32, i32 } %849, 1, !dbg !27 + %853 = bitcast i32 %852 to <2 x bfloat>, !dbg !27 + %854 = extractvalue { i32, i32, i32, i32 } %849, 2, !dbg !27 + %855 = bitcast i32 %854 to <2 x bfloat>, !dbg !27 + %856 = extractvalue { i32, i32, i32, i32 } %849, 3, !dbg !27 + %857 = bitcast i32 %856 to <2 x bfloat>, !dbg !27 + %858 = extractelement <2 x bfloat> %851, i64 0, !dbg !27 + %859 = extractelement <2 x bfloat> %851, i64 1, !dbg !27 + %860 = extractelement <2 x bfloat> %853, i64 0, !dbg !27 + %861 = extractelement <2 x bfloat> %853, i64 1, !dbg !27 + %862 = extractelement <2 x bfloat> %855, i64 0, !dbg !27 + %863 = extractelement <2 x bfloat> %855, i64 1, !dbg !27 + %864 = extractelement <2 x bfloat> %857, i64 0, !dbg !27 + %865 = extractelement <2 x bfloat> %857, i64 1, !dbg !27 + %866 = extractelement <64 x i1> %62, i64 51, !dbg !27 + %867 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %270, i1 %866) #5, !dbg !27 + %868 = extractvalue { i32, i32, i32, i32 } %867, 0, !dbg !27 + %869 = bitcast i32 %868 to <2 x bfloat>, !dbg !27 + %870 = extractvalue { i32, i32, i32, i32 } %867, 1, !dbg !27 + %871 = bitcast i32 %870 to <2 x bfloat>, !dbg !27 + %872 = extractvalue { i32, i32, i32, i32 } %867, 2, !dbg !27 + %873 = bitcast i32 %872 to <2 x bfloat>, !dbg !27 + %874 = extractvalue { i32, i32, i32, i32 } %867, 3, !dbg !27 + %875 = bitcast i32 %874 to <2 x bfloat>, !dbg !27 + %876 = extractelement <2 x bfloat> %869, i64 0, !dbg !27 + %877 = extractelement <2 x bfloat> %869, i64 1, !dbg !27 + %878 = extractelement <2 x bfloat> %871, i64 0, !dbg !27 + %879 = extractelement <2 x bfloat> %871, i64 1, !dbg !27 + %880 = extractelement <2 x bfloat> %873, i64 0, !dbg !27 + %881 = extractelement <2 x bfloat> %873, i64 1, !dbg !27 + %882 = extractelement <2 x bfloat> %875, i64 0, !dbg !27 + %883 = extractelement <2 x bfloat> %875, i64 1, !dbg !27 + %884 = extractelement <64 x i1> %62, i64 50, !dbg !27 + %885 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %271, i1 %884) #5, !dbg !27 + %886 = extractvalue { i32, i32, i32, i32 } %885, 0, !dbg !27 + %887 = bitcast i32 %886 to <2 x bfloat>, !dbg !27 + %888 = extractvalue { i32, i32, i32, i32 } %885, 1, !dbg !27 + %889 = bitcast i32 %888 to <2 x bfloat>, !dbg !27 + %890 = extractvalue { i32, i32, i32, i32 } %885, 2, !dbg !27 + %891 = bitcast i32 %890 to <2 x bfloat>, !dbg !27 + %892 = extractvalue { i32, i32, i32, i32 } %885, 3, !dbg !27 + %893 = bitcast i32 %892 to <2 x bfloat>, !dbg !27 + %894 = extractelement <2 x bfloat> %887, i64 0, !dbg !27 + %895 = extractelement <2 x bfloat> %887, i64 1, !dbg !27 + %896 = extractelement <2 x bfloat> %889, i64 0, !dbg !27 + %897 = extractelement <2 x bfloat> %889, i64 1, !dbg !27 + %898 = extractelement <2 x bfloat> %891, i64 0, !dbg !27 + %899 = extractelement <2 x bfloat> %891, i64 1, !dbg !27 + %900 = extractelement <2 x bfloat> %893, i64 0, !dbg !27 + %901 = extractelement <2 x bfloat> %893, i64 1, !dbg !27 + %902 = extractelement <64 x i1> %62, i64 49, !dbg !27 + %903 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %272, i1 %902) #5, !dbg !27 + %904 = extractvalue { i32, i32, i32, i32 } %903, 0, !dbg !27 + %905 = bitcast i32 %904 to <2 x bfloat>, !dbg !27 + %906 = extractvalue { i32, i32, i32, i32 } %903, 1, !dbg !27 + %907 = bitcast i32 %906 to <2 x bfloat>, !dbg !27 + %908 = extractvalue { i32, i32, i32, i32 } %903, 2, !dbg !27 + %909 = bitcast i32 %908 to <2 x bfloat>, !dbg !27 + %910 = extractvalue { i32, i32, i32, i32 } %903, 3, !dbg !27 + %911 = bitcast i32 %910 to <2 x bfloat>, !dbg !27 + %912 = extractelement <2 x bfloat> %905, i64 0, !dbg !27 + %913 = extractelement <2 x bfloat> %905, i64 1, !dbg !27 + %914 = extractelement <2 x bfloat> %907, i64 0, !dbg !27 + %915 = extractelement <2 x bfloat> %907, i64 1, !dbg !27 + %916 = extractelement <2 x bfloat> %909, i64 0, !dbg !27 + %917 = extractelement <2 x bfloat> %909, i64 1, !dbg !27 + %918 = extractelement <2 x bfloat> %911, i64 0, !dbg !27 + %919 = extractelement <2 x bfloat> %911, i64 1, !dbg !27 + %920 = extractelement <64 x i1> %62, i64 48, !dbg !27 + %921 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %273, i1 %920) #5, !dbg !27 + %922 = extractvalue { i32, i32, i32, i32 } %921, 0, !dbg !27 + %923 = bitcast i32 %922 to <2 x bfloat>, !dbg !27 + %924 = extractvalue { i32, i32, i32, i32 } %921, 1, !dbg !27 + %925 = bitcast i32 %924 to <2 x bfloat>, !dbg !27 + %926 = extractvalue { i32, i32, i32, i32 } %921, 2, !dbg !27 + %927 = bitcast i32 %926 to <2 x bfloat>, !dbg !27 + %928 = extractvalue { i32, i32, i32, i32 } %921, 3, !dbg !27 + %929 = bitcast i32 %928 to <2 x bfloat>, !dbg !27 + %930 = extractelement <2 x bfloat> %923, i64 0, !dbg !27 + %931 = extractelement <2 x bfloat> %923, i64 1, !dbg !27 + %932 = extractelement <2 x bfloat> %925, i64 0, !dbg !27 + %933 = extractelement <2 x bfloat> %925, i64 1, !dbg !27 + %934 = extractelement <2 x bfloat> %927, i64 0, !dbg !27 + %935 = extractelement <2 x bfloat> %927, i64 1, !dbg !27 + %936 = extractelement <2 x bfloat> %929, i64 0, !dbg !27 + %937 = extractelement <2 x bfloat> %929, i64 1, !dbg !27 + %938 = extractelement <64 x i1> %62, i64 47, !dbg !27 + %939 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %274, i1 %938) #5, !dbg !27 + %940 = extractvalue { i32, i32, i32, i32 } %939, 0, !dbg !27 + %941 = bitcast i32 %940 to <2 x bfloat>, !dbg !27 + %942 = extractvalue { i32, i32, i32, i32 } %939, 1, !dbg !27 + %943 = bitcast i32 %942 to <2 x bfloat>, !dbg !27 + %944 = extractvalue { i32, i32, i32, i32 } %939, 2, !dbg !27 + %945 = bitcast i32 %944 to <2 x bfloat>, !dbg !27 + %946 = extractvalue { i32, i32, i32, i32 } %939, 3, !dbg !27 + %947 = bitcast i32 %946 to <2 x bfloat>, !dbg !27 + %948 = extractelement <2 x bfloat> %941, i64 0, !dbg !27 + %949 = extractelement <2 x bfloat> %941, i64 1, !dbg !27 + %950 = extractelement <2 x bfloat> %943, i64 0, !dbg !27 + %951 = extractelement <2 x bfloat> %943, i64 1, !dbg !27 + %952 = extractelement <2 x bfloat> %945, i64 0, !dbg !27 + %953 = extractelement <2 x bfloat> %945, i64 1, !dbg !27 + %954 = extractelement <2 x bfloat> %947, i64 0, !dbg !27 + %955 = extractelement <2 x bfloat> %947, i64 1, !dbg !27 + %956 = extractelement <64 x i1> %62, i64 46, !dbg !27 + %957 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %275, i1 %956) #5, !dbg !27 + %958 = extractvalue { i32, i32, i32, i32 } %957, 0, !dbg !27 + %959 = bitcast i32 %958 to <2 x bfloat>, !dbg !27 + %960 = extractvalue { i32, i32, i32, i32 } %957, 1, !dbg !27 + %961 = bitcast i32 %960 to <2 x bfloat>, !dbg !27 + %962 = extractvalue { i32, i32, i32, i32 } %957, 2, !dbg !27 + %963 = bitcast i32 %962 to <2 x bfloat>, !dbg !27 + %964 = extractvalue { i32, i32, i32, i32 } %957, 3, !dbg !27 + %965 = bitcast i32 %964 to <2 x bfloat>, !dbg !27 + %966 = extractelement <2 x bfloat> %959, i64 0, !dbg !27 + %967 = extractelement <2 x bfloat> %959, i64 1, !dbg !27 + %968 = extractelement <2 x bfloat> %961, i64 0, !dbg !27 + %969 = extractelement <2 x bfloat> %961, i64 1, !dbg !27 + %970 = extractelement <2 x bfloat> %963, i64 0, !dbg !27 + %971 = extractelement <2 x bfloat> %963, i64 1, !dbg !27 + %972 = extractelement <2 x bfloat> %965, i64 0, !dbg !27 + %973 = extractelement <2 x bfloat> %965, i64 1, !dbg !27 + %974 = extractelement <64 x i1> %62, i64 45, !dbg !27 + %975 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %276, i1 %974) #5, !dbg !27 + %976 = extractvalue { i32, i32, i32, i32 } %975, 0, !dbg !27 + %977 = bitcast i32 %976 to <2 x bfloat>, !dbg !27 + %978 = extractvalue { i32, i32, i32, i32 } %975, 1, !dbg !27 + %979 = bitcast i32 %978 to <2 x bfloat>, !dbg !27 + %980 = extractvalue { i32, i32, i32, i32 } %975, 2, !dbg !27 + %981 = bitcast i32 %980 to <2 x bfloat>, !dbg !27 + %982 = extractvalue { i32, i32, i32, i32 } %975, 3, !dbg !27 + %983 = bitcast i32 %982 to <2 x bfloat>, !dbg !27 + %984 = extractelement <2 x bfloat> %977, i64 0, !dbg !27 + %985 = extractelement <2 x bfloat> %977, i64 1, !dbg !27 + %986 = extractelement <2 x bfloat> %979, i64 0, !dbg !27 + %987 = extractelement <2 x bfloat> %979, i64 1, !dbg !27 + %988 = extractelement <2 x bfloat> %981, i64 0, !dbg !27 + %989 = extractelement <2 x bfloat> %981, i64 1, !dbg !27 + %990 = extractelement <2 x bfloat> %983, i64 0, !dbg !27 + %991 = extractelement <2 x bfloat> %983, i64 1, !dbg !27 + %992 = extractelement <64 x i1> %62, i64 44, !dbg !27 + %993 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %277, i1 %992) #5, !dbg !27 + %994 = extractvalue { i32, i32, i32, i32 } %993, 0, !dbg !27 + %995 = bitcast i32 %994 to <2 x bfloat>, !dbg !27 + %996 = extractvalue { i32, i32, i32, i32 } %993, 1, !dbg !27 + %997 = bitcast i32 %996 to <2 x bfloat>, !dbg !27 + %998 = extractvalue { i32, i32, i32, i32 } %993, 2, !dbg !27 + %999 = bitcast i32 %998 to <2 x bfloat>, !dbg !27 + %1000 = extractvalue { i32, i32, i32, i32 } %993, 3, !dbg !27 + %1001 = bitcast i32 %1000 to <2 x bfloat>, !dbg !27 + %1002 = extractelement <2 x bfloat> %995, i64 0, !dbg !27 + %1003 = extractelement <2 x bfloat> %995, i64 1, !dbg !27 + %1004 = extractelement <2 x bfloat> %997, i64 0, !dbg !27 + %1005 = extractelement <2 x bfloat> %997, i64 1, !dbg !27 + %1006 = extractelement <2 x bfloat> %999, i64 0, !dbg !27 + %1007 = extractelement <2 x bfloat> %999, i64 1, !dbg !27 + %1008 = extractelement <2 x bfloat> %1001, i64 0, !dbg !27 + %1009 = extractelement <2 x bfloat> %1001, i64 1, !dbg !27 + %1010 = extractelement <64 x i1> %62, i64 43, !dbg !27 + %1011 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %278, i1 %1010) #5, !dbg !27 + %1012 = extractvalue { i32, i32, i32, i32 } %1011, 0, !dbg !27 + %1013 = bitcast i32 %1012 to <2 x bfloat>, !dbg !27 + %1014 = extractvalue { i32, i32, i32, i32 } %1011, 1, !dbg !27 + %1015 = bitcast i32 %1014 to <2 x bfloat>, !dbg !27 + %1016 = extractvalue { i32, i32, i32, i32 } %1011, 2, !dbg !27 + %1017 = bitcast i32 %1016 to <2 x bfloat>, !dbg !27 + %1018 = extractvalue { i32, i32, i32, i32 } %1011, 3, !dbg !27 + %1019 = bitcast i32 %1018 to <2 x bfloat>, !dbg !27 + %1020 = extractelement <2 x bfloat> %1013, i64 0, !dbg !27 + %1021 = extractelement <2 x bfloat> %1013, i64 1, !dbg !27 + %1022 = extractelement <2 x bfloat> %1015, i64 0, !dbg !27 + %1023 = extractelement <2 x bfloat> %1015, i64 1, !dbg !27 + %1024 = extractelement <2 x bfloat> %1017, i64 0, !dbg !27 + %1025 = extractelement <2 x bfloat> %1017, i64 1, !dbg !27 + %1026 = extractelement <2 x bfloat> %1019, i64 0, !dbg !27 + %1027 = extractelement <2 x bfloat> %1019, i64 1, !dbg !27 + %1028 = extractelement <64 x i1> %62, i64 42, !dbg !27 + %1029 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %279, i1 %1028) #5, !dbg !27 + %1030 = extractvalue { i32, i32, i32, i32 } %1029, 0, !dbg !27 + %1031 = bitcast i32 %1030 to <2 x bfloat>, !dbg !27 + %1032 = extractvalue { i32, i32, i32, i32 } %1029, 1, !dbg !27 + %1033 = bitcast i32 %1032 to <2 x bfloat>, !dbg !27 + %1034 = extractvalue { i32, i32, i32, i32 } %1029, 2, !dbg !27 + %1035 = bitcast i32 %1034 to <2 x bfloat>, !dbg !27 + %1036 = extractvalue { i32, i32, i32, i32 } %1029, 3, !dbg !27 + %1037 = bitcast i32 %1036 to <2 x bfloat>, !dbg !27 + %1038 = extractelement <2 x bfloat> %1031, i64 0, !dbg !27 + %1039 = extractelement <2 x bfloat> %1031, i64 1, !dbg !27 + %1040 = extractelement <2 x bfloat> %1033, i64 0, !dbg !27 + %1041 = extractelement <2 x bfloat> %1033, i64 1, !dbg !27 + %1042 = extractelement <2 x bfloat> %1035, i64 0, !dbg !27 + %1043 = extractelement <2 x bfloat> %1035, i64 1, !dbg !27 + %1044 = extractelement <2 x bfloat> %1037, i64 0, !dbg !27 + %1045 = extractelement <2 x bfloat> %1037, i64 1, !dbg !27 + %1046 = extractelement <64 x i1> %62, i64 41, !dbg !27 + %1047 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %280, i1 %1046) #5, !dbg !27 + %1048 = extractvalue { i32, i32, i32, i32 } %1047, 0, !dbg !27 + %1049 = bitcast i32 %1048 to <2 x bfloat>, !dbg !27 + %1050 = extractvalue { i32, i32, i32, i32 } %1047, 1, !dbg !27 + %1051 = bitcast i32 %1050 to <2 x bfloat>, !dbg !27 + %1052 = extractvalue { i32, i32, i32, i32 } %1047, 2, !dbg !27 + %1053 = bitcast i32 %1052 to <2 x bfloat>, !dbg !27 + %1054 = extractvalue { i32, i32, i32, i32 } %1047, 3, !dbg !27 + %1055 = bitcast i32 %1054 to <2 x bfloat>, !dbg !27 + %1056 = extractelement <2 x bfloat> %1049, i64 0, !dbg !27 + %1057 = extractelement <2 x bfloat> %1049, i64 1, !dbg !27 + %1058 = extractelement <2 x bfloat> %1051, i64 0, !dbg !27 + %1059 = extractelement <2 x bfloat> %1051, i64 1, !dbg !27 + %1060 = extractelement <2 x bfloat> %1053, i64 0, !dbg !27 + %1061 = extractelement <2 x bfloat> %1053, i64 1, !dbg !27 + %1062 = extractelement <2 x bfloat> %1055, i64 0, !dbg !27 + %1063 = extractelement <2 x bfloat> %1055, i64 1, !dbg !27 + %1064 = extractelement <64 x i1> %62, i64 40, !dbg !27 + %1065 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %281, i1 %1064) #5, !dbg !27 + %1066 = extractvalue { i32, i32, i32, i32 } %1065, 0, !dbg !27 + %1067 = bitcast i32 %1066 to <2 x bfloat>, !dbg !27 + %1068 = extractvalue { i32, i32, i32, i32 } %1065, 1, !dbg !27 + %1069 = bitcast i32 %1068 to <2 x bfloat>, !dbg !27 + %1070 = extractvalue { i32, i32, i32, i32 } %1065, 2, !dbg !27 + %1071 = bitcast i32 %1070 to <2 x bfloat>, !dbg !27 + %1072 = extractvalue { i32, i32, i32, i32 } %1065, 3, !dbg !27 + %1073 = bitcast i32 %1072 to <2 x bfloat>, !dbg !27 + %1074 = extractelement <2 x bfloat> %1067, i64 0, !dbg !27 + %1075 = extractelement <2 x bfloat> %1067, i64 1, !dbg !27 + %1076 = extractelement <2 x bfloat> %1069, i64 0, !dbg !27 + %1077 = extractelement <2 x bfloat> %1069, i64 1, !dbg !27 + %1078 = extractelement <2 x bfloat> %1071, i64 0, !dbg !27 + %1079 = extractelement <2 x bfloat> %1071, i64 1, !dbg !27 + %1080 = extractelement <2 x bfloat> %1073, i64 0, !dbg !27 + %1081 = extractelement <2 x bfloat> %1073, i64 1, !dbg !27 + %1082 = extractelement <64 x i1> %62, i64 39, !dbg !27 + %1083 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %282, i1 %1082) #5, !dbg !27 + %1084 = extractvalue { i32, i32, i32, i32 } %1083, 0, !dbg !27 + %1085 = bitcast i32 %1084 to <2 x bfloat>, !dbg !27 + %1086 = extractvalue { i32, i32, i32, i32 } %1083, 1, !dbg !27 + %1087 = bitcast i32 %1086 to <2 x bfloat>, !dbg !27 + %1088 = extractvalue { i32, i32, i32, i32 } %1083, 2, !dbg !27 + %1089 = bitcast i32 %1088 to <2 x bfloat>, !dbg !27 + %1090 = extractvalue { i32, i32, i32, i32 } %1083, 3, !dbg !27 + %1091 = bitcast i32 %1090 to <2 x bfloat>, !dbg !27 + %1092 = extractelement <2 x bfloat> %1085, i64 0, !dbg !27 + %1093 = extractelement <2 x bfloat> %1085, i64 1, !dbg !27 + %1094 = extractelement <2 x bfloat> %1087, i64 0, !dbg !27 + %1095 = extractelement <2 x bfloat> %1087, i64 1, !dbg !27 + %1096 = extractelement <2 x bfloat> %1089, i64 0, !dbg !27 + %1097 = extractelement <2 x bfloat> %1089, i64 1, !dbg !27 + %1098 = extractelement <2 x bfloat> %1091, i64 0, !dbg !27 + %1099 = extractelement <2 x bfloat> %1091, i64 1, !dbg !27 + %1100 = extractelement <64 x i1> %62, i64 38, !dbg !27 + %1101 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %283, i1 %1100) #5, !dbg !27 + %1102 = extractvalue { i32, i32, i32, i32 } %1101, 0, !dbg !27 + %1103 = bitcast i32 %1102 to <2 x bfloat>, !dbg !27 + %1104 = extractvalue { i32, i32, i32, i32 } %1101, 1, !dbg !27 + %1105 = bitcast i32 %1104 to <2 x bfloat>, !dbg !27 + %1106 = extractvalue { i32, i32, i32, i32 } %1101, 2, !dbg !27 + %1107 = bitcast i32 %1106 to <2 x bfloat>, !dbg !27 + %1108 = extractvalue { i32, i32, i32, i32 } %1101, 3, !dbg !27 + %1109 = bitcast i32 %1108 to <2 x bfloat>, !dbg !27 + %1110 = extractelement <2 x bfloat> %1103, i64 0, !dbg !27 + %1111 = extractelement <2 x bfloat> %1103, i64 1, !dbg !27 + %1112 = extractelement <2 x bfloat> %1105, i64 0, !dbg !27 + %1113 = extractelement <2 x bfloat> %1105, i64 1, !dbg !27 + %1114 = extractelement <2 x bfloat> %1107, i64 0, !dbg !27 + %1115 = extractelement <2 x bfloat> %1107, i64 1, !dbg !27 + %1116 = extractelement <2 x bfloat> %1109, i64 0, !dbg !27 + %1117 = extractelement <2 x bfloat> %1109, i64 1, !dbg !27 + %1118 = extractelement <64 x i1> %62, i64 37, !dbg !27 + %1119 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %284, i1 %1118) #5, !dbg !27 + %1120 = extractvalue { i32, i32, i32, i32 } %1119, 0, !dbg !27 + %1121 = bitcast i32 %1120 to <2 x bfloat>, !dbg !27 + %1122 = extractvalue { i32, i32, i32, i32 } %1119, 1, !dbg !27 + %1123 = bitcast i32 %1122 to <2 x bfloat>, !dbg !27 + %1124 = extractvalue { i32, i32, i32, i32 } %1119, 2, !dbg !27 + %1125 = bitcast i32 %1124 to <2 x bfloat>, !dbg !27 + %1126 = extractvalue { i32, i32, i32, i32 } %1119, 3, !dbg !27 + %1127 = bitcast i32 %1126 to <2 x bfloat>, !dbg !27 + %1128 = extractelement <2 x bfloat> %1121, i64 0, !dbg !27 + %1129 = extractelement <2 x bfloat> %1121, i64 1, !dbg !27 + %1130 = extractelement <2 x bfloat> %1123, i64 0, !dbg !27 + %1131 = extractelement <2 x bfloat> %1123, i64 1, !dbg !27 + %1132 = extractelement <2 x bfloat> %1125, i64 0, !dbg !27 + %1133 = extractelement <2 x bfloat> %1125, i64 1, !dbg !27 + %1134 = extractelement <2 x bfloat> %1127, i64 0, !dbg !27 + %1135 = extractelement <2 x bfloat> %1127, i64 1, !dbg !27 + %1136 = extractelement <64 x i1> %62, i64 36, !dbg !27 + %1137 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %285, i1 %1136) #5, !dbg !27 + %1138 = extractvalue { i32, i32, i32, i32 } %1137, 0, !dbg !27 + %1139 = bitcast i32 %1138 to <2 x bfloat>, !dbg !27 + %1140 = extractvalue { i32, i32, i32, i32 } %1137, 1, !dbg !27 + %1141 = bitcast i32 %1140 to <2 x bfloat>, !dbg !27 + %1142 = extractvalue { i32, i32, i32, i32 } %1137, 2, !dbg !27 + %1143 = bitcast i32 %1142 to <2 x bfloat>, !dbg !27 + %1144 = extractvalue { i32, i32, i32, i32 } %1137, 3, !dbg !27 + %1145 = bitcast i32 %1144 to <2 x bfloat>, !dbg !27 + %1146 = extractelement <2 x bfloat> %1139, i64 0, !dbg !27 + %1147 = extractelement <2 x bfloat> %1139, i64 1, !dbg !27 + %1148 = extractelement <2 x bfloat> %1141, i64 0, !dbg !27 + %1149 = extractelement <2 x bfloat> %1141, i64 1, !dbg !27 + %1150 = extractelement <2 x bfloat> %1143, i64 0, !dbg !27 + %1151 = extractelement <2 x bfloat> %1143, i64 1, !dbg !27 + %1152 = extractelement <2 x bfloat> %1145, i64 0, !dbg !27 + %1153 = extractelement <2 x bfloat> %1145, i64 1, !dbg !27 + %1154 = extractelement <64 x i1> %62, i64 35, !dbg !27 + %1155 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %286, i1 %1154) #5, !dbg !27 + %1156 = extractvalue { i32, i32, i32, i32 } %1155, 0, !dbg !27 + %1157 = bitcast i32 %1156 to <2 x bfloat>, !dbg !27 + %1158 = extractvalue { i32, i32, i32, i32 } %1155, 1, !dbg !27 + %1159 = bitcast i32 %1158 to <2 x bfloat>, !dbg !27 + %1160 = extractvalue { i32, i32, i32, i32 } %1155, 2, !dbg !27 + %1161 = bitcast i32 %1160 to <2 x bfloat>, !dbg !27 + %1162 = extractvalue { i32, i32, i32, i32 } %1155, 3, !dbg !27 + %1163 = bitcast i32 %1162 to <2 x bfloat>, !dbg !27 + %1164 = extractelement <2 x bfloat> %1157, i64 0, !dbg !27 + %1165 = extractelement <2 x bfloat> %1157, i64 1, !dbg !27 + %1166 = extractelement <2 x bfloat> %1159, i64 0, !dbg !27 + %1167 = extractelement <2 x bfloat> %1159, i64 1, !dbg !27 + %1168 = extractelement <2 x bfloat> %1161, i64 0, !dbg !27 + %1169 = extractelement <2 x bfloat> %1161, i64 1, !dbg !27 + %1170 = extractelement <2 x bfloat> %1163, i64 0, !dbg !27 + %1171 = extractelement <2 x bfloat> %1163, i64 1, !dbg !27 + %1172 = extractelement <64 x i1> %62, i64 34, !dbg !27 + %1173 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %287, i1 %1172) #5, !dbg !27 + %1174 = extractvalue { i32, i32, i32, i32 } %1173, 0, !dbg !27 + %1175 = bitcast i32 %1174 to <2 x bfloat>, !dbg !27 + %1176 = extractvalue { i32, i32, i32, i32 } %1173, 1, !dbg !27 + %1177 = bitcast i32 %1176 to <2 x bfloat>, !dbg !27 + %1178 = extractvalue { i32, i32, i32, i32 } %1173, 2, !dbg !27 + %1179 = bitcast i32 %1178 to <2 x bfloat>, !dbg !27 + %1180 = extractvalue { i32, i32, i32, i32 } %1173, 3, !dbg !27 + %1181 = bitcast i32 %1180 to <2 x bfloat>, !dbg !27 + %1182 = extractelement <2 x bfloat> %1175, i64 0, !dbg !27 + %1183 = extractelement <2 x bfloat> %1175, i64 1, !dbg !27 + %1184 = extractelement <2 x bfloat> %1177, i64 0, !dbg !27 + %1185 = extractelement <2 x bfloat> %1177, i64 1, !dbg !27 + %1186 = extractelement <2 x bfloat> %1179, i64 0, !dbg !27 + %1187 = extractelement <2 x bfloat> %1179, i64 1, !dbg !27 + %1188 = extractelement <2 x bfloat> %1181, i64 0, !dbg !27 + %1189 = extractelement <2 x bfloat> %1181, i64 1, !dbg !27 + %1190 = extractelement <64 x i1> %62, i64 33, !dbg !27 + %1191 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %288, i1 %1190) #5, !dbg !27 + %1192 = extractvalue { i32, i32, i32, i32 } %1191, 0, !dbg !27 + %1193 = bitcast i32 %1192 to <2 x bfloat>, !dbg !27 + %1194 = extractvalue { i32, i32, i32, i32 } %1191, 1, !dbg !27 + %1195 = bitcast i32 %1194 to <2 x bfloat>, !dbg !27 + %1196 = extractvalue { i32, i32, i32, i32 } %1191, 2, !dbg !27 + %1197 = bitcast i32 %1196 to <2 x bfloat>, !dbg !27 + %1198 = extractvalue { i32, i32, i32, i32 } %1191, 3, !dbg !27 + %1199 = bitcast i32 %1198 to <2 x bfloat>, !dbg !27 + %1200 = extractelement <2 x bfloat> %1193, i64 0, !dbg !27 + %1201 = extractelement <2 x bfloat> %1193, i64 1, !dbg !27 + %1202 = extractelement <2 x bfloat> %1195, i64 0, !dbg !27 + %1203 = extractelement <2 x bfloat> %1195, i64 1, !dbg !27 + %1204 = extractelement <2 x bfloat> %1197, i64 0, !dbg !27 + %1205 = extractelement <2 x bfloat> %1197, i64 1, !dbg !27 + %1206 = extractelement <2 x bfloat> %1199, i64 0, !dbg !27 + %1207 = extractelement <2 x bfloat> %1199, i64 1, !dbg !27 + %1208 = extractelement <64 x i1> %62, i64 32, !dbg !27 + %1209 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %289, i1 %1208) #5, !dbg !27 + %1210 = extractvalue { i32, i32, i32, i32 } %1209, 0, !dbg !27 + %1211 = bitcast i32 %1210 to <2 x bfloat>, !dbg !27 + %1212 = extractvalue { i32, i32, i32, i32 } %1209, 1, !dbg !27 + %1213 = bitcast i32 %1212 to <2 x bfloat>, !dbg !27 + %1214 = extractvalue { i32, i32, i32, i32 } %1209, 2, !dbg !27 + %1215 = bitcast i32 %1214 to <2 x bfloat>, !dbg !27 + %1216 = extractvalue { i32, i32, i32, i32 } %1209, 3, !dbg !27 + %1217 = bitcast i32 %1216 to <2 x bfloat>, !dbg !27 + %1218 = extractelement <2 x bfloat> %1211, i64 0, !dbg !27 + %1219 = extractelement <2 x bfloat> %1211, i64 1, !dbg !27 + %1220 = extractelement <2 x bfloat> %1213, i64 0, !dbg !27 + %1221 = extractelement <2 x bfloat> %1213, i64 1, !dbg !27 + %1222 = extractelement <2 x bfloat> %1215, i64 0, !dbg !27 + %1223 = extractelement <2 x bfloat> %1215, i64 1, !dbg !27 + %1224 = extractelement <2 x bfloat> %1217, i64 0, !dbg !27 + %1225 = extractelement <2 x bfloat> %1217, i64 1, !dbg !27 + %1226 = extractelement <64 x i1> %62, i64 31, !dbg !27 + %1227 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %290, i1 %1226) #5, !dbg !27 + %1228 = extractvalue { i32, i32, i32, i32 } %1227, 0, !dbg !27 + %1229 = bitcast i32 %1228 to <2 x bfloat>, !dbg !27 + %1230 = extractvalue { i32, i32, i32, i32 } %1227, 1, !dbg !27 + %1231 = bitcast i32 %1230 to <2 x bfloat>, !dbg !27 + %1232 = extractvalue { i32, i32, i32, i32 } %1227, 2, !dbg !27 + %1233 = bitcast i32 %1232 to <2 x bfloat>, !dbg !27 + %1234 = extractvalue { i32, i32, i32, i32 } %1227, 3, !dbg !27 + %1235 = bitcast i32 %1234 to <2 x bfloat>, !dbg !27 + %1236 = extractelement <2 x bfloat> %1229, i64 0, !dbg !27 + %1237 = extractelement <2 x bfloat> %1229, i64 1, !dbg !27 + %1238 = extractelement <2 x bfloat> %1231, i64 0, !dbg !27 + %1239 = extractelement <2 x bfloat> %1231, i64 1, !dbg !27 + %1240 = extractelement <2 x bfloat> %1233, i64 0, !dbg !27 + %1241 = extractelement <2 x bfloat> %1233, i64 1, !dbg !27 + %1242 = extractelement <2 x bfloat> %1235, i64 0, !dbg !27 + %1243 = extractelement <2 x bfloat> %1235, i64 1, !dbg !27 + %1244 = extractelement <64 x i1> %62, i64 30, !dbg !27 + %1245 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %291, i1 %1244) #5, !dbg !27 + %1246 = extractvalue { i32, i32, i32, i32 } %1245, 0, !dbg !27 + %1247 = bitcast i32 %1246 to <2 x bfloat>, !dbg !27 + %1248 = extractvalue { i32, i32, i32, i32 } %1245, 1, !dbg !27 + %1249 = bitcast i32 %1248 to <2 x bfloat>, !dbg !27 + %1250 = extractvalue { i32, i32, i32, i32 } %1245, 2, !dbg !27 + %1251 = bitcast i32 %1250 to <2 x bfloat>, !dbg !27 + %1252 = extractvalue { i32, i32, i32, i32 } %1245, 3, !dbg !27 + %1253 = bitcast i32 %1252 to <2 x bfloat>, !dbg !27 + %1254 = extractelement <2 x bfloat> %1247, i64 0, !dbg !27 + %1255 = extractelement <2 x bfloat> %1247, i64 1, !dbg !27 + %1256 = extractelement <2 x bfloat> %1249, i64 0, !dbg !27 + %1257 = extractelement <2 x bfloat> %1249, i64 1, !dbg !27 + %1258 = extractelement <2 x bfloat> %1251, i64 0, !dbg !27 + %1259 = extractelement <2 x bfloat> %1251, i64 1, !dbg !27 + %1260 = extractelement <2 x bfloat> %1253, i64 0, !dbg !27 + %1261 = extractelement <2 x bfloat> %1253, i64 1, !dbg !27 + %1262 = extractelement <64 x i1> %62, i64 29, !dbg !27 + %1263 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %292, i1 %1262) #5, !dbg !27 + %1264 = extractvalue { i32, i32, i32, i32 } %1263, 0, !dbg !27 + %1265 = bitcast i32 %1264 to <2 x bfloat>, !dbg !27 + %1266 = extractvalue { i32, i32, i32, i32 } %1263, 1, !dbg !27 + %1267 = bitcast i32 %1266 to <2 x bfloat>, !dbg !27 + %1268 = extractvalue { i32, i32, i32, i32 } %1263, 2, !dbg !27 + %1269 = bitcast i32 %1268 to <2 x bfloat>, !dbg !27 + %1270 = extractvalue { i32, i32, i32, i32 } %1263, 3, !dbg !27 + %1271 = bitcast i32 %1270 to <2 x bfloat>, !dbg !27 + %1272 = extractelement <2 x bfloat> %1265, i64 0, !dbg !27 + %1273 = extractelement <2 x bfloat> %1265, i64 1, !dbg !27 + %1274 = extractelement <2 x bfloat> %1267, i64 0, !dbg !27 + %1275 = extractelement <2 x bfloat> %1267, i64 1, !dbg !27 + %1276 = extractelement <2 x bfloat> %1269, i64 0, !dbg !27 + %1277 = extractelement <2 x bfloat> %1269, i64 1, !dbg !27 + %1278 = extractelement <2 x bfloat> %1271, i64 0, !dbg !27 + %1279 = extractelement <2 x bfloat> %1271, i64 1, !dbg !27 + %1280 = extractelement <64 x i1> %62, i64 28, !dbg !27 + %1281 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %293, i1 %1280) #5, !dbg !27 + %1282 = extractvalue { i32, i32, i32, i32 } %1281, 0, !dbg !27 + %1283 = bitcast i32 %1282 to <2 x bfloat>, !dbg !27 + %1284 = extractvalue { i32, i32, i32, i32 } %1281, 1, !dbg !27 + %1285 = bitcast i32 %1284 to <2 x bfloat>, !dbg !27 + %1286 = extractvalue { i32, i32, i32, i32 } %1281, 2, !dbg !27 + %1287 = bitcast i32 %1286 to <2 x bfloat>, !dbg !27 + %1288 = extractvalue { i32, i32, i32, i32 } %1281, 3, !dbg !27 + %1289 = bitcast i32 %1288 to <2 x bfloat>, !dbg !27 + %1290 = extractelement <2 x bfloat> %1283, i64 0, !dbg !27 + %1291 = extractelement <2 x bfloat> %1283, i64 1, !dbg !27 + %1292 = extractelement <2 x bfloat> %1285, i64 0, !dbg !27 + %1293 = extractelement <2 x bfloat> %1285, i64 1, !dbg !27 + %1294 = extractelement <2 x bfloat> %1287, i64 0, !dbg !27 + %1295 = extractelement <2 x bfloat> %1287, i64 1, !dbg !27 + %1296 = extractelement <2 x bfloat> %1289, i64 0, !dbg !27 + %1297 = extractelement <2 x bfloat> %1289, i64 1, !dbg !27 + %1298 = extractelement <64 x i1> %62, i64 27, !dbg !27 + %1299 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %294, i1 %1298) #5, !dbg !27 + %1300 = extractvalue { i32, i32, i32, i32 } %1299, 0, !dbg !27 + %1301 = bitcast i32 %1300 to <2 x bfloat>, !dbg !27 + %1302 = extractvalue { i32, i32, i32, i32 } %1299, 1, !dbg !27 + %1303 = bitcast i32 %1302 to <2 x bfloat>, !dbg !27 + %1304 = extractvalue { i32, i32, i32, i32 } %1299, 2, !dbg !27 + %1305 = bitcast i32 %1304 to <2 x bfloat>, !dbg !27 + %1306 = extractvalue { i32, i32, i32, i32 } %1299, 3, !dbg !27 + %1307 = bitcast i32 %1306 to <2 x bfloat>, !dbg !27 + %1308 = extractelement <2 x bfloat> %1301, i64 0, !dbg !27 + %1309 = extractelement <2 x bfloat> %1301, i64 1, !dbg !27 + %1310 = extractelement <2 x bfloat> %1303, i64 0, !dbg !27 + %1311 = extractelement <2 x bfloat> %1303, i64 1, !dbg !27 + %1312 = extractelement <2 x bfloat> %1305, i64 0, !dbg !27 + %1313 = extractelement <2 x bfloat> %1305, i64 1, !dbg !27 + %1314 = extractelement <2 x bfloat> %1307, i64 0, !dbg !27 + %1315 = extractelement <2 x bfloat> %1307, i64 1, !dbg !27 + %1316 = extractelement <64 x i1> %62, i64 26, !dbg !27 + %1317 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %295, i1 %1316) #5, !dbg !27 + %1318 = extractvalue { i32, i32, i32, i32 } %1317, 0, !dbg !27 + %1319 = bitcast i32 %1318 to <2 x bfloat>, !dbg !27 + %1320 = extractvalue { i32, i32, i32, i32 } %1317, 1, !dbg !27 + %1321 = bitcast i32 %1320 to <2 x bfloat>, !dbg !27 + %1322 = extractvalue { i32, i32, i32, i32 } %1317, 2, !dbg !27 + %1323 = bitcast i32 %1322 to <2 x bfloat>, !dbg !27 + %1324 = extractvalue { i32, i32, i32, i32 } %1317, 3, !dbg !27 + %1325 = bitcast i32 %1324 to <2 x bfloat>, !dbg !27 + %1326 = extractelement <2 x bfloat> %1319, i64 0, !dbg !27 + %1327 = extractelement <2 x bfloat> %1319, i64 1, !dbg !27 + %1328 = extractelement <2 x bfloat> %1321, i64 0, !dbg !27 + %1329 = extractelement <2 x bfloat> %1321, i64 1, !dbg !27 + %1330 = extractelement <2 x bfloat> %1323, i64 0, !dbg !27 + %1331 = extractelement <2 x bfloat> %1323, i64 1, !dbg !27 + %1332 = extractelement <2 x bfloat> %1325, i64 0, !dbg !27 + %1333 = extractelement <2 x bfloat> %1325, i64 1, !dbg !27 + %1334 = extractelement <64 x i1> %62, i64 25, !dbg !27 + %1335 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %296, i1 %1334) #5, !dbg !27 + %1336 = extractvalue { i32, i32, i32, i32 } %1335, 0, !dbg !27 + %1337 = bitcast i32 %1336 to <2 x bfloat>, !dbg !27 + %1338 = extractvalue { i32, i32, i32, i32 } %1335, 1, !dbg !27 + %1339 = bitcast i32 %1338 to <2 x bfloat>, !dbg !27 + %1340 = extractvalue { i32, i32, i32, i32 } %1335, 2, !dbg !27 + %1341 = bitcast i32 %1340 to <2 x bfloat>, !dbg !27 + %1342 = extractvalue { i32, i32, i32, i32 } %1335, 3, !dbg !27 + %1343 = bitcast i32 %1342 to <2 x bfloat>, !dbg !27 + %1344 = extractelement <2 x bfloat> %1337, i64 0, !dbg !27 + %1345 = extractelement <2 x bfloat> %1337, i64 1, !dbg !27 + %1346 = extractelement <2 x bfloat> %1339, i64 0, !dbg !27 + %1347 = extractelement <2 x bfloat> %1339, i64 1, !dbg !27 + %1348 = extractelement <2 x bfloat> %1341, i64 0, !dbg !27 + %1349 = extractelement <2 x bfloat> %1341, i64 1, !dbg !27 + %1350 = extractelement <2 x bfloat> %1343, i64 0, !dbg !27 + %1351 = extractelement <2 x bfloat> %1343, i64 1, !dbg !27 + %1352 = extractelement <64 x i1> %62, i64 24, !dbg !27 + %1353 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %297, i1 %1352) #5, !dbg !27 + %1354 = extractvalue { i32, i32, i32, i32 } %1353, 0, !dbg !27 + %1355 = bitcast i32 %1354 to <2 x bfloat>, !dbg !27 + %1356 = extractvalue { i32, i32, i32, i32 } %1353, 1, !dbg !27 + %1357 = bitcast i32 %1356 to <2 x bfloat>, !dbg !27 + %1358 = extractvalue { i32, i32, i32, i32 } %1353, 2, !dbg !27 + %1359 = bitcast i32 %1358 to <2 x bfloat>, !dbg !27 + %1360 = extractvalue { i32, i32, i32, i32 } %1353, 3, !dbg !27 + %1361 = bitcast i32 %1360 to <2 x bfloat>, !dbg !27 + %1362 = extractelement <2 x bfloat> %1355, i64 0, !dbg !27 + %1363 = extractelement <2 x bfloat> %1355, i64 1, !dbg !27 + %1364 = extractelement <2 x bfloat> %1357, i64 0, !dbg !27 + %1365 = extractelement <2 x bfloat> %1357, i64 1, !dbg !27 + %1366 = extractelement <2 x bfloat> %1359, i64 0, !dbg !27 + %1367 = extractelement <2 x bfloat> %1359, i64 1, !dbg !27 + %1368 = extractelement <2 x bfloat> %1361, i64 0, !dbg !27 + %1369 = extractelement <2 x bfloat> %1361, i64 1, !dbg !27 + %1370 = extractelement <64 x i1> %62, i64 23, !dbg !27 + %1371 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %298, i1 %1370) #5, !dbg !27 + %1372 = extractvalue { i32, i32, i32, i32 } %1371, 0, !dbg !27 + %1373 = bitcast i32 %1372 to <2 x bfloat>, !dbg !27 + %1374 = extractvalue { i32, i32, i32, i32 } %1371, 1, !dbg !27 + %1375 = bitcast i32 %1374 to <2 x bfloat>, !dbg !27 + %1376 = extractvalue { i32, i32, i32, i32 } %1371, 2, !dbg !27 + %1377 = bitcast i32 %1376 to <2 x bfloat>, !dbg !27 + %1378 = extractvalue { i32, i32, i32, i32 } %1371, 3, !dbg !27 + %1379 = bitcast i32 %1378 to <2 x bfloat>, !dbg !27 + %1380 = extractelement <2 x bfloat> %1373, i64 0, !dbg !27 + %1381 = extractelement <2 x bfloat> %1373, i64 1, !dbg !27 + %1382 = extractelement <2 x bfloat> %1375, i64 0, !dbg !27 + %1383 = extractelement <2 x bfloat> %1375, i64 1, !dbg !27 + %1384 = extractelement <2 x bfloat> %1377, i64 0, !dbg !27 + %1385 = extractelement <2 x bfloat> %1377, i64 1, !dbg !27 + %1386 = extractelement <2 x bfloat> %1379, i64 0, !dbg !27 + %1387 = extractelement <2 x bfloat> %1379, i64 1, !dbg !27 + %1388 = extractelement <64 x i1> %62, i64 22, !dbg !27 + %1389 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %299, i1 %1388) #5, !dbg !27 + %1390 = extractvalue { i32, i32, i32, i32 } %1389, 0, !dbg !27 + %1391 = bitcast i32 %1390 to <2 x bfloat>, !dbg !27 + %1392 = extractvalue { i32, i32, i32, i32 } %1389, 1, !dbg !27 + %1393 = bitcast i32 %1392 to <2 x bfloat>, !dbg !27 + %1394 = extractvalue { i32, i32, i32, i32 } %1389, 2, !dbg !27 + %1395 = bitcast i32 %1394 to <2 x bfloat>, !dbg !27 + %1396 = extractvalue { i32, i32, i32, i32 } %1389, 3, !dbg !27 + %1397 = bitcast i32 %1396 to <2 x bfloat>, !dbg !27 + %1398 = extractelement <2 x bfloat> %1391, i64 0, !dbg !27 + %1399 = extractelement <2 x bfloat> %1391, i64 1, !dbg !27 + %1400 = extractelement <2 x bfloat> %1393, i64 0, !dbg !27 + %1401 = extractelement <2 x bfloat> %1393, i64 1, !dbg !27 + %1402 = extractelement <2 x bfloat> %1395, i64 0, !dbg !27 + %1403 = extractelement <2 x bfloat> %1395, i64 1, !dbg !27 + %1404 = extractelement <2 x bfloat> %1397, i64 0, !dbg !27 + %1405 = extractelement <2 x bfloat> %1397, i64 1, !dbg !27 + %1406 = extractelement <64 x i1> %62, i64 21, !dbg !27 + %1407 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %300, i1 %1406) #5, !dbg !27 + %1408 = extractvalue { i32, i32, i32, i32 } %1407, 0, !dbg !27 + %1409 = bitcast i32 %1408 to <2 x bfloat>, !dbg !27 + %1410 = extractvalue { i32, i32, i32, i32 } %1407, 1, !dbg !27 + %1411 = bitcast i32 %1410 to <2 x bfloat>, !dbg !27 + %1412 = extractvalue { i32, i32, i32, i32 } %1407, 2, !dbg !27 + %1413 = bitcast i32 %1412 to <2 x bfloat>, !dbg !27 + %1414 = extractvalue { i32, i32, i32, i32 } %1407, 3, !dbg !27 + %1415 = bitcast i32 %1414 to <2 x bfloat>, !dbg !27 + %1416 = extractelement <2 x bfloat> %1409, i64 0, !dbg !27 + %1417 = extractelement <2 x bfloat> %1409, i64 1, !dbg !27 + %1418 = extractelement <2 x bfloat> %1411, i64 0, !dbg !27 + %1419 = extractelement <2 x bfloat> %1411, i64 1, !dbg !27 + %1420 = extractelement <2 x bfloat> %1413, i64 0, !dbg !27 + %1421 = extractelement <2 x bfloat> %1413, i64 1, !dbg !27 + %1422 = extractelement <2 x bfloat> %1415, i64 0, !dbg !27 + %1423 = extractelement <2 x bfloat> %1415, i64 1, !dbg !27 + %1424 = extractelement <64 x i1> %62, i64 20, !dbg !27 + %1425 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %301, i1 %1424) #5, !dbg !27 + %1426 = extractvalue { i32, i32, i32, i32 } %1425, 0, !dbg !27 + %1427 = bitcast i32 %1426 to <2 x bfloat>, !dbg !27 + %1428 = extractvalue { i32, i32, i32, i32 } %1425, 1, !dbg !27 + %1429 = bitcast i32 %1428 to <2 x bfloat>, !dbg !27 + %1430 = extractvalue { i32, i32, i32, i32 } %1425, 2, !dbg !27 + %1431 = bitcast i32 %1430 to <2 x bfloat>, !dbg !27 + %1432 = extractvalue { i32, i32, i32, i32 } %1425, 3, !dbg !27 + %1433 = bitcast i32 %1432 to <2 x bfloat>, !dbg !27 + %1434 = extractelement <2 x bfloat> %1427, i64 0, !dbg !27 + %1435 = extractelement <2 x bfloat> %1427, i64 1, !dbg !27 + %1436 = extractelement <2 x bfloat> %1429, i64 0, !dbg !27 + %1437 = extractelement <2 x bfloat> %1429, i64 1, !dbg !27 + %1438 = extractelement <2 x bfloat> %1431, i64 0, !dbg !27 + %1439 = extractelement <2 x bfloat> %1431, i64 1, !dbg !27 + %1440 = extractelement <2 x bfloat> %1433, i64 0, !dbg !27 + %1441 = extractelement <2 x bfloat> %1433, i64 1, !dbg !27 + %1442 = extractelement <64 x i1> %62, i64 19, !dbg !27 + %1443 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %302, i1 %1442) #5, !dbg !27 + %1444 = extractvalue { i32, i32, i32, i32 } %1443, 0, !dbg !27 + %1445 = bitcast i32 %1444 to <2 x bfloat>, !dbg !27 + %1446 = extractvalue { i32, i32, i32, i32 } %1443, 1, !dbg !27 + %1447 = bitcast i32 %1446 to <2 x bfloat>, !dbg !27 + %1448 = extractvalue { i32, i32, i32, i32 } %1443, 2, !dbg !27 + %1449 = bitcast i32 %1448 to <2 x bfloat>, !dbg !27 + %1450 = extractvalue { i32, i32, i32, i32 } %1443, 3, !dbg !27 + %1451 = bitcast i32 %1450 to <2 x bfloat>, !dbg !27 + %1452 = extractelement <2 x bfloat> %1445, i64 0, !dbg !27 + %1453 = extractelement <2 x bfloat> %1445, i64 1, !dbg !27 + %1454 = extractelement <2 x bfloat> %1447, i64 0, !dbg !27 + %1455 = extractelement <2 x bfloat> %1447, i64 1, !dbg !27 + %1456 = extractelement <2 x bfloat> %1449, i64 0, !dbg !27 + %1457 = extractelement <2 x bfloat> %1449, i64 1, !dbg !27 + %1458 = extractelement <2 x bfloat> %1451, i64 0, !dbg !27 + %1459 = extractelement <2 x bfloat> %1451, i64 1, !dbg !27 + %1460 = extractelement <64 x i1> %62, i64 18, !dbg !27 + %1461 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %303, i1 %1460) #5, !dbg !27 + %1462 = extractvalue { i32, i32, i32, i32 } %1461, 0, !dbg !27 + %1463 = bitcast i32 %1462 to <2 x bfloat>, !dbg !27 + %1464 = extractvalue { i32, i32, i32, i32 } %1461, 1, !dbg !27 + %1465 = bitcast i32 %1464 to <2 x bfloat>, !dbg !27 + %1466 = extractvalue { i32, i32, i32, i32 } %1461, 2, !dbg !27 + %1467 = bitcast i32 %1466 to <2 x bfloat>, !dbg !27 + %1468 = extractvalue { i32, i32, i32, i32 } %1461, 3, !dbg !27 + %1469 = bitcast i32 %1468 to <2 x bfloat>, !dbg !27 + %1470 = extractelement <2 x bfloat> %1463, i64 0, !dbg !27 + %1471 = extractelement <2 x bfloat> %1463, i64 1, !dbg !27 + %1472 = extractelement <2 x bfloat> %1465, i64 0, !dbg !27 + %1473 = extractelement <2 x bfloat> %1465, i64 1, !dbg !27 + %1474 = extractelement <2 x bfloat> %1467, i64 0, !dbg !27 + %1475 = extractelement <2 x bfloat> %1467, i64 1, !dbg !27 + %1476 = extractelement <2 x bfloat> %1469, i64 0, !dbg !27 + %1477 = extractelement <2 x bfloat> %1469, i64 1, !dbg !27 + %1478 = extractelement <64 x i1> %62, i64 17, !dbg !27 + %1479 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %304, i1 %1478) #5, !dbg !27 + %1480 = extractvalue { i32, i32, i32, i32 } %1479, 0, !dbg !27 + %1481 = bitcast i32 %1480 to <2 x bfloat>, !dbg !27 + %1482 = extractvalue { i32, i32, i32, i32 } %1479, 1, !dbg !27 + %1483 = bitcast i32 %1482 to <2 x bfloat>, !dbg !27 + %1484 = extractvalue { i32, i32, i32, i32 } %1479, 2, !dbg !27 + %1485 = bitcast i32 %1484 to <2 x bfloat>, !dbg !27 + %1486 = extractvalue { i32, i32, i32, i32 } %1479, 3, !dbg !27 + %1487 = bitcast i32 %1486 to <2 x bfloat>, !dbg !27 + %1488 = extractelement <2 x bfloat> %1481, i64 0, !dbg !27 + %1489 = extractelement <2 x bfloat> %1481, i64 1, !dbg !27 + %1490 = extractelement <2 x bfloat> %1483, i64 0, !dbg !27 + %1491 = extractelement <2 x bfloat> %1483, i64 1, !dbg !27 + %1492 = extractelement <2 x bfloat> %1485, i64 0, !dbg !27 + %1493 = extractelement <2 x bfloat> %1485, i64 1, !dbg !27 + %1494 = extractelement <2 x bfloat> %1487, i64 0, !dbg !27 + %1495 = extractelement <2 x bfloat> %1487, i64 1, !dbg !27 + %1496 = extractelement <64 x i1> %62, i64 16, !dbg !27 + %1497 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %305, i1 %1496) #5, !dbg !27 + %1498 = extractvalue { i32, i32, i32, i32 } %1497, 0, !dbg !27 + %1499 = bitcast i32 %1498 to <2 x bfloat>, !dbg !27 + %1500 = extractvalue { i32, i32, i32, i32 } %1497, 1, !dbg !27 + %1501 = bitcast i32 %1500 to <2 x bfloat>, !dbg !27 + %1502 = extractvalue { i32, i32, i32, i32 } %1497, 2, !dbg !27 + %1503 = bitcast i32 %1502 to <2 x bfloat>, !dbg !27 + %1504 = extractvalue { i32, i32, i32, i32 } %1497, 3, !dbg !27 + %1505 = bitcast i32 %1504 to <2 x bfloat>, !dbg !27 + %1506 = extractelement <2 x bfloat> %1499, i64 0, !dbg !27 + %1507 = extractelement <2 x bfloat> %1499, i64 1, !dbg !27 + %1508 = extractelement <2 x bfloat> %1501, i64 0, !dbg !27 + %1509 = extractelement <2 x bfloat> %1501, i64 1, !dbg !27 + %1510 = extractelement <2 x bfloat> %1503, i64 0, !dbg !27 + %1511 = extractelement <2 x bfloat> %1503, i64 1, !dbg !27 + %1512 = extractelement <2 x bfloat> %1505, i64 0, !dbg !27 + %1513 = extractelement <2 x bfloat> %1505, i64 1, !dbg !27 + %1514 = extractelement <64 x i1> %62, i64 15, !dbg !27 + %1515 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %306, i1 %1514) #5, !dbg !27 + %1516 = extractvalue { i32, i32, i32, i32 } %1515, 0, !dbg !27 + %1517 = bitcast i32 %1516 to <2 x bfloat>, !dbg !27 + %1518 = extractvalue { i32, i32, i32, i32 } %1515, 1, !dbg !27 + %1519 = bitcast i32 %1518 to <2 x bfloat>, !dbg !27 + %1520 = extractvalue { i32, i32, i32, i32 } %1515, 2, !dbg !27 + %1521 = bitcast i32 %1520 to <2 x bfloat>, !dbg !27 + %1522 = extractvalue { i32, i32, i32, i32 } %1515, 3, !dbg !27 + %1523 = bitcast i32 %1522 to <2 x bfloat>, !dbg !27 + %1524 = extractelement <2 x bfloat> %1517, i64 0, !dbg !27 + %1525 = extractelement <2 x bfloat> %1517, i64 1, !dbg !27 + %1526 = extractelement <2 x bfloat> %1519, i64 0, !dbg !27 + %1527 = extractelement <2 x bfloat> %1519, i64 1, !dbg !27 + %1528 = extractelement <2 x bfloat> %1521, i64 0, !dbg !27 + %1529 = extractelement <2 x bfloat> %1521, i64 1, !dbg !27 + %1530 = extractelement <2 x bfloat> %1523, i64 0, !dbg !27 + %1531 = extractelement <2 x bfloat> %1523, i64 1, !dbg !27 + %1532 = extractelement <64 x i1> %62, i64 14, !dbg !27 + %1533 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %307, i1 %1532) #5, !dbg !27 + %1534 = extractvalue { i32, i32, i32, i32 } %1533, 0, !dbg !27 + %1535 = bitcast i32 %1534 to <2 x bfloat>, !dbg !27 + %1536 = extractvalue { i32, i32, i32, i32 } %1533, 1, !dbg !27 + %1537 = bitcast i32 %1536 to <2 x bfloat>, !dbg !27 + %1538 = extractvalue { i32, i32, i32, i32 } %1533, 2, !dbg !27 + %1539 = bitcast i32 %1538 to <2 x bfloat>, !dbg !27 + %1540 = extractvalue { i32, i32, i32, i32 } %1533, 3, !dbg !27 + %1541 = bitcast i32 %1540 to <2 x bfloat>, !dbg !27 + %1542 = extractelement <2 x bfloat> %1535, i64 0, !dbg !27 + %1543 = extractelement <2 x bfloat> %1535, i64 1, !dbg !27 + %1544 = extractelement <2 x bfloat> %1537, i64 0, !dbg !27 + %1545 = extractelement <2 x bfloat> %1537, i64 1, !dbg !27 + %1546 = extractelement <2 x bfloat> %1539, i64 0, !dbg !27 + %1547 = extractelement <2 x bfloat> %1539, i64 1, !dbg !27 + %1548 = extractelement <2 x bfloat> %1541, i64 0, !dbg !27 + %1549 = extractelement <2 x bfloat> %1541, i64 1, !dbg !27 + %1550 = extractelement <64 x i1> %62, i64 13, !dbg !27 + %1551 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %308, i1 %1550) #5, !dbg !27 + %1552 = extractvalue { i32, i32, i32, i32 } %1551, 0, !dbg !27 + %1553 = bitcast i32 %1552 to <2 x bfloat>, !dbg !27 + %1554 = extractvalue { i32, i32, i32, i32 } %1551, 1, !dbg !27 + %1555 = bitcast i32 %1554 to <2 x bfloat>, !dbg !27 + %1556 = extractvalue { i32, i32, i32, i32 } %1551, 2, !dbg !27 + %1557 = bitcast i32 %1556 to <2 x bfloat>, !dbg !27 + %1558 = extractvalue { i32, i32, i32, i32 } %1551, 3, !dbg !27 + %1559 = bitcast i32 %1558 to <2 x bfloat>, !dbg !27 + %1560 = extractelement <2 x bfloat> %1553, i64 0, !dbg !27 + %1561 = extractelement <2 x bfloat> %1553, i64 1, !dbg !27 + %1562 = extractelement <2 x bfloat> %1555, i64 0, !dbg !27 + %1563 = extractelement <2 x bfloat> %1555, i64 1, !dbg !27 + %1564 = extractelement <2 x bfloat> %1557, i64 0, !dbg !27 + %1565 = extractelement <2 x bfloat> %1557, i64 1, !dbg !27 + %1566 = extractelement <2 x bfloat> %1559, i64 0, !dbg !27 + %1567 = extractelement <2 x bfloat> %1559, i64 1, !dbg !27 + %1568 = extractelement <64 x i1> %62, i64 12, !dbg !27 + %1569 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %309, i1 %1568) #5, !dbg !27 + %1570 = extractvalue { i32, i32, i32, i32 } %1569, 0, !dbg !27 + %1571 = bitcast i32 %1570 to <2 x bfloat>, !dbg !27 + %1572 = extractvalue { i32, i32, i32, i32 } %1569, 1, !dbg !27 + %1573 = bitcast i32 %1572 to <2 x bfloat>, !dbg !27 + %1574 = extractvalue { i32, i32, i32, i32 } %1569, 2, !dbg !27 + %1575 = bitcast i32 %1574 to <2 x bfloat>, !dbg !27 + %1576 = extractvalue { i32, i32, i32, i32 } %1569, 3, !dbg !27 + %1577 = bitcast i32 %1576 to <2 x bfloat>, !dbg !27 + %1578 = extractelement <2 x bfloat> %1571, i64 0, !dbg !27 + %1579 = extractelement <2 x bfloat> %1571, i64 1, !dbg !27 + %1580 = extractelement <2 x bfloat> %1573, i64 0, !dbg !27 + %1581 = extractelement <2 x bfloat> %1573, i64 1, !dbg !27 + %1582 = extractelement <2 x bfloat> %1575, i64 0, !dbg !27 + %1583 = extractelement <2 x bfloat> %1575, i64 1, !dbg !27 + %1584 = extractelement <2 x bfloat> %1577, i64 0, !dbg !27 + %1585 = extractelement <2 x bfloat> %1577, i64 1, !dbg !27 + %1586 = extractelement <64 x i1> %62, i64 11, !dbg !27 + %1587 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %310, i1 %1586) #5, !dbg !27 + %1588 = extractvalue { i32, i32, i32, i32 } %1587, 0, !dbg !27 + %1589 = bitcast i32 %1588 to <2 x bfloat>, !dbg !27 + %1590 = extractvalue { i32, i32, i32, i32 } %1587, 1, !dbg !27 + %1591 = bitcast i32 %1590 to <2 x bfloat>, !dbg !27 + %1592 = extractvalue { i32, i32, i32, i32 } %1587, 2, !dbg !27 + %1593 = bitcast i32 %1592 to <2 x bfloat>, !dbg !27 + %1594 = extractvalue { i32, i32, i32, i32 } %1587, 3, !dbg !27 + %1595 = bitcast i32 %1594 to <2 x bfloat>, !dbg !27 + %1596 = extractelement <2 x bfloat> %1589, i64 0, !dbg !27 + %1597 = extractelement <2 x bfloat> %1589, i64 1, !dbg !27 + %1598 = extractelement <2 x bfloat> %1591, i64 0, !dbg !27 + %1599 = extractelement <2 x bfloat> %1591, i64 1, !dbg !27 + %1600 = extractelement <2 x bfloat> %1593, i64 0, !dbg !27 + %1601 = extractelement <2 x bfloat> %1593, i64 1, !dbg !27 + %1602 = extractelement <2 x bfloat> %1595, i64 0, !dbg !27 + %1603 = extractelement <2 x bfloat> %1595, i64 1, !dbg !27 + %1604 = extractelement <64 x i1> %62, i64 10, !dbg !27 + %1605 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %311, i1 %1604) #5, !dbg !27 + %1606 = extractvalue { i32, i32, i32, i32 } %1605, 0, !dbg !27 + %1607 = bitcast i32 %1606 to <2 x bfloat>, !dbg !27 + %1608 = extractvalue { i32, i32, i32, i32 } %1605, 1, !dbg !27 + %1609 = bitcast i32 %1608 to <2 x bfloat>, !dbg !27 + %1610 = extractvalue { i32, i32, i32, i32 } %1605, 2, !dbg !27 + %1611 = bitcast i32 %1610 to <2 x bfloat>, !dbg !27 + %1612 = extractvalue { i32, i32, i32, i32 } %1605, 3, !dbg !27 + %1613 = bitcast i32 %1612 to <2 x bfloat>, !dbg !27 + %1614 = extractelement <2 x bfloat> %1607, i64 0, !dbg !27 + %1615 = extractelement <2 x bfloat> %1607, i64 1, !dbg !27 + %1616 = extractelement <2 x bfloat> %1609, i64 0, !dbg !27 + %1617 = extractelement <2 x bfloat> %1609, i64 1, !dbg !27 + %1618 = extractelement <2 x bfloat> %1611, i64 0, !dbg !27 + %1619 = extractelement <2 x bfloat> %1611, i64 1, !dbg !27 + %1620 = extractelement <2 x bfloat> %1613, i64 0, !dbg !27 + %1621 = extractelement <2 x bfloat> %1613, i64 1, !dbg !27 + %1622 = extractelement <64 x i1> %62, i64 9, !dbg !27 + %1623 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %312, i1 %1622) #5, !dbg !27 + %1624 = extractvalue { i32, i32, i32, i32 } %1623, 0, !dbg !27 + %1625 = bitcast i32 %1624 to <2 x bfloat>, !dbg !27 + %1626 = extractvalue { i32, i32, i32, i32 } %1623, 1, !dbg !27 + %1627 = bitcast i32 %1626 to <2 x bfloat>, !dbg !27 + %1628 = extractvalue { i32, i32, i32, i32 } %1623, 2, !dbg !27 + %1629 = bitcast i32 %1628 to <2 x bfloat>, !dbg !27 + %1630 = extractvalue { i32, i32, i32, i32 } %1623, 3, !dbg !27 + %1631 = bitcast i32 %1630 to <2 x bfloat>, !dbg !27 + %1632 = extractelement <2 x bfloat> %1625, i64 0, !dbg !27 + %1633 = extractelement <2 x bfloat> %1625, i64 1, !dbg !27 + %1634 = extractelement <2 x bfloat> %1627, i64 0, !dbg !27 + %1635 = extractelement <2 x bfloat> %1627, i64 1, !dbg !27 + %1636 = extractelement <2 x bfloat> %1629, i64 0, !dbg !27 + %1637 = extractelement <2 x bfloat> %1629, i64 1, !dbg !27 + %1638 = extractelement <2 x bfloat> %1631, i64 0, !dbg !27 + %1639 = extractelement <2 x bfloat> %1631, i64 1, !dbg !27 + %1640 = extractelement <64 x i1> %62, i64 8, !dbg !27 + %1641 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %313, i1 %1640) #5, !dbg !27 + %1642 = extractvalue { i32, i32, i32, i32 } %1641, 0, !dbg !27 + %1643 = bitcast i32 %1642 to <2 x bfloat>, !dbg !27 + %1644 = extractvalue { i32, i32, i32, i32 } %1641, 1, !dbg !27 + %1645 = bitcast i32 %1644 to <2 x bfloat>, !dbg !27 + %1646 = extractvalue { i32, i32, i32, i32 } %1641, 2, !dbg !27 + %1647 = bitcast i32 %1646 to <2 x bfloat>, !dbg !27 + %1648 = extractvalue { i32, i32, i32, i32 } %1641, 3, !dbg !27 + %1649 = bitcast i32 %1648 to <2 x bfloat>, !dbg !27 + %1650 = extractelement <2 x bfloat> %1643, i64 0, !dbg !27 + %1651 = extractelement <2 x bfloat> %1643, i64 1, !dbg !27 + %1652 = extractelement <2 x bfloat> %1645, i64 0, !dbg !27 + %1653 = extractelement <2 x bfloat> %1645, i64 1, !dbg !27 + %1654 = extractelement <2 x bfloat> %1647, i64 0, !dbg !27 + %1655 = extractelement <2 x bfloat> %1647, i64 1, !dbg !27 + %1656 = extractelement <2 x bfloat> %1649, i64 0, !dbg !27 + %1657 = extractelement <2 x bfloat> %1649, i64 1, !dbg !27 + %1658 = extractelement <64 x i1> %62, i64 7, !dbg !27 + %1659 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %314, i1 %1658) #5, !dbg !27 + %1660 = extractvalue { i32, i32, i32, i32 } %1659, 0, !dbg !27 + %1661 = bitcast i32 %1660 to <2 x bfloat>, !dbg !27 + %1662 = extractvalue { i32, i32, i32, i32 } %1659, 1, !dbg !27 + %1663 = bitcast i32 %1662 to <2 x bfloat>, !dbg !27 + %1664 = extractvalue { i32, i32, i32, i32 } %1659, 2, !dbg !27 + %1665 = bitcast i32 %1664 to <2 x bfloat>, !dbg !27 + %1666 = extractvalue { i32, i32, i32, i32 } %1659, 3, !dbg !27 + %1667 = bitcast i32 %1666 to <2 x bfloat>, !dbg !27 + %1668 = extractelement <2 x bfloat> %1661, i64 0, !dbg !27 + %1669 = extractelement <2 x bfloat> %1661, i64 1, !dbg !27 + %1670 = extractelement <2 x bfloat> %1663, i64 0, !dbg !27 + %1671 = extractelement <2 x bfloat> %1663, i64 1, !dbg !27 + %1672 = extractelement <2 x bfloat> %1665, i64 0, !dbg !27 + %1673 = extractelement <2 x bfloat> %1665, i64 1, !dbg !27 + %1674 = extractelement <2 x bfloat> %1667, i64 0, !dbg !27 + %1675 = extractelement <2 x bfloat> %1667, i64 1, !dbg !27 + %1676 = extractelement <64 x i1> %62, i64 6, !dbg !27 + %1677 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %315, i1 %1676) #5, !dbg !27 + %1678 = extractvalue { i32, i32, i32, i32 } %1677, 0, !dbg !27 + %1679 = bitcast i32 %1678 to <2 x bfloat>, !dbg !27 + %1680 = extractvalue { i32, i32, i32, i32 } %1677, 1, !dbg !27 + %1681 = bitcast i32 %1680 to <2 x bfloat>, !dbg !27 + %1682 = extractvalue { i32, i32, i32, i32 } %1677, 2, !dbg !27 + %1683 = bitcast i32 %1682 to <2 x bfloat>, !dbg !27 + %1684 = extractvalue { i32, i32, i32, i32 } %1677, 3, !dbg !27 + %1685 = bitcast i32 %1684 to <2 x bfloat>, !dbg !27 + %1686 = extractelement <2 x bfloat> %1679, i64 0, !dbg !27 + %1687 = extractelement <2 x bfloat> %1679, i64 1, !dbg !27 + %1688 = extractelement <2 x bfloat> %1681, i64 0, !dbg !27 + %1689 = extractelement <2 x bfloat> %1681, i64 1, !dbg !27 + %1690 = extractelement <2 x bfloat> %1683, i64 0, !dbg !27 + %1691 = extractelement <2 x bfloat> %1683, i64 1, !dbg !27 + %1692 = extractelement <2 x bfloat> %1685, i64 0, !dbg !27 + %1693 = extractelement <2 x bfloat> %1685, i64 1, !dbg !27 + %1694 = extractelement <64 x i1> %62, i64 5, !dbg !27 + %1695 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %316, i1 %1694) #5, !dbg !27 + %1696 = extractvalue { i32, i32, i32, i32 } %1695, 0, !dbg !27 + %1697 = bitcast i32 %1696 to <2 x bfloat>, !dbg !27 + %1698 = extractvalue { i32, i32, i32, i32 } %1695, 1, !dbg !27 + %1699 = bitcast i32 %1698 to <2 x bfloat>, !dbg !27 + %1700 = extractvalue { i32, i32, i32, i32 } %1695, 2, !dbg !27 + %1701 = bitcast i32 %1700 to <2 x bfloat>, !dbg !27 + %1702 = extractvalue { i32, i32, i32, i32 } %1695, 3, !dbg !27 + %1703 = bitcast i32 %1702 to <2 x bfloat>, !dbg !27 + %1704 = extractelement <2 x bfloat> %1697, i64 0, !dbg !27 + %1705 = extractelement <2 x bfloat> %1697, i64 1, !dbg !27 + %1706 = extractelement <2 x bfloat> %1699, i64 0, !dbg !27 + %1707 = extractelement <2 x bfloat> %1699, i64 1, !dbg !27 + %1708 = extractelement <2 x bfloat> %1701, i64 0, !dbg !27 + %1709 = extractelement <2 x bfloat> %1701, i64 1, !dbg !27 + %1710 = extractelement <2 x bfloat> %1703, i64 0, !dbg !27 + %1711 = extractelement <2 x bfloat> %1703, i64 1, !dbg !27 + %1712 = extractelement <64 x i1> %62, i64 4, !dbg !27 + %1713 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %317, i1 %1712) #5, !dbg !27 + %1714 = extractvalue { i32, i32, i32, i32 } %1713, 0, !dbg !27 + %1715 = bitcast i32 %1714 to <2 x bfloat>, !dbg !27 + %1716 = extractvalue { i32, i32, i32, i32 } %1713, 1, !dbg !27 + %1717 = bitcast i32 %1716 to <2 x bfloat>, !dbg !27 + %1718 = extractvalue { i32, i32, i32, i32 } %1713, 2, !dbg !27 + %1719 = bitcast i32 %1718 to <2 x bfloat>, !dbg !27 + %1720 = extractvalue { i32, i32, i32, i32 } %1713, 3, !dbg !27 + %1721 = bitcast i32 %1720 to <2 x bfloat>, !dbg !27 + %1722 = extractelement <2 x bfloat> %1715, i64 0, !dbg !27 + %1723 = extractelement <2 x bfloat> %1715, i64 1, !dbg !27 + %1724 = extractelement <2 x bfloat> %1717, i64 0, !dbg !27 + %1725 = extractelement <2 x bfloat> %1717, i64 1, !dbg !27 + %1726 = extractelement <2 x bfloat> %1719, i64 0, !dbg !27 + %1727 = extractelement <2 x bfloat> %1719, i64 1, !dbg !27 + %1728 = extractelement <2 x bfloat> %1721, i64 0, !dbg !27 + %1729 = extractelement <2 x bfloat> %1721, i64 1, !dbg !27 + %1730 = fpext bfloat %591 to float, !dbg !28 + %1731 = fpext bfloat %592 to float, !dbg !28 + %1732 = fpext bfloat %593 to float, !dbg !28 + %1733 = fpext bfloat %594 to float, !dbg !28 + %1734 = fpext bfloat %595 to float, !dbg !28 + %1735 = fpext bfloat %596 to float, !dbg !28 + %1736 = fpext bfloat %597 to float, !dbg !28 + %1737 = fpext bfloat %598 to float, !dbg !28 + %1738 = fpext bfloat %608 to float, !dbg !28 + %1739 = fpext bfloat %609 to float, !dbg !28 + %1740 = fpext bfloat %610 to float, !dbg !28 + %1741 = fpext bfloat %611 to float, !dbg !28 + %1742 = fpext bfloat %612 to float, !dbg !28 + %1743 = fpext bfloat %613 to float, !dbg !28 + %1744 = fpext bfloat %614 to float, !dbg !28 + %1745 = fpext bfloat %615 to float, !dbg !28 + %1746 = fpext bfloat %625 to float, !dbg !28 + %1747 = fpext bfloat %626 to float, !dbg !28 + %1748 = fpext bfloat %627 to float, !dbg !28 + %1749 = fpext bfloat %628 to float, !dbg !28 + %1750 = fpext bfloat %629 to float, !dbg !28 + %1751 = fpext bfloat %630 to float, !dbg !28 + %1752 = fpext bfloat %631 to float, !dbg !28 + %1753 = fpext bfloat %632 to float, !dbg !28 + %1754 = fpext bfloat %642 to float, !dbg !28 + %1755 = fpext bfloat %643 to float, !dbg !28 + %1756 = fpext bfloat %644 to float, !dbg !28 + %1757 = fpext bfloat %645 to float, !dbg !28 + %1758 = fpext bfloat %646 to float, !dbg !28 + %1759 = fpext bfloat %647 to float, !dbg !28 + %1760 = fpext bfloat %648 to float, !dbg !28 + %1761 = fpext bfloat %649 to float, !dbg !28 + %1762 = fpext bfloat %660 to float, !dbg !28 + %1763 = fpext bfloat %661 to float, !dbg !28 + %1764 = fpext bfloat %662 to float, !dbg !28 + %1765 = fpext bfloat %663 to float, !dbg !28 + %1766 = fpext bfloat %664 to float, !dbg !28 + %1767 = fpext bfloat %665 to float, !dbg !28 + %1768 = fpext bfloat %666 to float, !dbg !28 + %1769 = fpext bfloat %667 to float, !dbg !28 + %1770 = fpext bfloat %678 to float, !dbg !28 + %1771 = fpext bfloat %679 to float, !dbg !28 + %1772 = fpext bfloat %680 to float, !dbg !28 + %1773 = fpext bfloat %681 to float, !dbg !28 + %1774 = fpext bfloat %682 to float, !dbg !28 + %1775 = fpext bfloat %683 to float, !dbg !28 + %1776 = fpext bfloat %684 to float, !dbg !28 + %1777 = fpext bfloat %685 to float, !dbg !28 + %1778 = fpext bfloat %696 to float, !dbg !28 + %1779 = fpext bfloat %697 to float, !dbg !28 + %1780 = fpext bfloat %698 to float, !dbg !28 + %1781 = fpext bfloat %699 to float, !dbg !28 + %1782 = fpext bfloat %700 to float, !dbg !28 + %1783 = fpext bfloat %701 to float, !dbg !28 + %1784 = fpext bfloat %702 to float, !dbg !28 + %1785 = fpext bfloat %703 to float, !dbg !28 + %1786 = fpext bfloat %714 to float, !dbg !28 + %1787 = fpext bfloat %715 to float, !dbg !28 + %1788 = fpext bfloat %716 to float, !dbg !28 + %1789 = fpext bfloat %717 to float, !dbg !28 + %1790 = fpext bfloat %718 to float, !dbg !28 + %1791 = fpext bfloat %719 to float, !dbg !28 + %1792 = fpext bfloat %720 to float, !dbg !28 + %1793 = fpext bfloat %721 to float, !dbg !28 + %1794 = fpext bfloat %732 to float, !dbg !28 + %1795 = fpext bfloat %733 to float, !dbg !28 + %1796 = fpext bfloat %734 to float, !dbg !28 + %1797 = fpext bfloat %735 to float, !dbg !28 + %1798 = fpext bfloat %736 to float, !dbg !28 + %1799 = fpext bfloat %737 to float, !dbg !28 + %1800 = fpext bfloat %738 to float, !dbg !28 + %1801 = fpext bfloat %739 to float, !dbg !28 + %1802 = fpext bfloat %750 to float, !dbg !28 + %1803 = fpext bfloat %751 to float, !dbg !28 + %1804 = fpext bfloat %752 to float, !dbg !28 + %1805 = fpext bfloat %753 to float, !dbg !28 + %1806 = fpext bfloat %754 to float, !dbg !28 + %1807 = fpext bfloat %755 to float, !dbg !28 + %1808 = fpext bfloat %756 to float, !dbg !28 + %1809 = fpext bfloat %757 to float, !dbg !28 + %1810 = fpext bfloat %768 to float, !dbg !28 + %1811 = fpext bfloat %769 to float, !dbg !28 + %1812 = fpext bfloat %770 to float, !dbg !28 + %1813 = fpext bfloat %771 to float, !dbg !28 + %1814 = fpext bfloat %772 to float, !dbg !28 + %1815 = fpext bfloat %773 to float, !dbg !28 + %1816 = fpext bfloat %774 to float, !dbg !28 + %1817 = fpext bfloat %775 to float, !dbg !28 + %1818 = fpext bfloat %786 to float, !dbg !28 + %1819 = fpext bfloat %787 to float, !dbg !28 + %1820 = fpext bfloat %788 to float, !dbg !28 + %1821 = fpext bfloat %789 to float, !dbg !28 + %1822 = fpext bfloat %790 to float, !dbg !28 + %1823 = fpext bfloat %791 to float, !dbg !28 + %1824 = fpext bfloat %792 to float, !dbg !28 + %1825 = fpext bfloat %793 to float, !dbg !28 + %1826 = fpext bfloat %804 to float, !dbg !28 + %1827 = fpext bfloat %805 to float, !dbg !28 + %1828 = fpext bfloat %806 to float, !dbg !28 + %1829 = fpext bfloat %807 to float, !dbg !28 + %1830 = fpext bfloat %808 to float, !dbg !28 + %1831 = fpext bfloat %809 to float, !dbg !28 + %1832 = fpext bfloat %810 to float, !dbg !28 + %1833 = fpext bfloat %811 to float, !dbg !28 + %1834 = fpext bfloat %822 to float, !dbg !28 + %1835 = fpext bfloat %823 to float, !dbg !28 + %1836 = fpext bfloat %824 to float, !dbg !28 + %1837 = fpext bfloat %825 to float, !dbg !28 + %1838 = fpext bfloat %826 to float, !dbg !28 + %1839 = fpext bfloat %827 to float, !dbg !28 + %1840 = fpext bfloat %828 to float, !dbg !28 + %1841 = fpext bfloat %829 to float, !dbg !28 + %1842 = fpext bfloat %840 to float, !dbg !28 + %1843 = fpext bfloat %841 to float, !dbg !28 + %1844 = fpext bfloat %842 to float, !dbg !28 + %1845 = fpext bfloat %843 to float, !dbg !28 + %1846 = fpext bfloat %844 to float, !dbg !28 + %1847 = fpext bfloat %845 to float, !dbg !28 + %1848 = fpext bfloat %846 to float, !dbg !28 + %1849 = fpext bfloat %847 to float, !dbg !28 + %1850 = fpext bfloat %858 to float, !dbg !28 + %1851 = fpext bfloat %859 to float, !dbg !28 + %1852 = fpext bfloat %860 to float, !dbg !28 + %1853 = fpext bfloat %861 to float, !dbg !28 + %1854 = fpext bfloat %862 to float, !dbg !28 + %1855 = fpext bfloat %863 to float, !dbg !28 + %1856 = fpext bfloat %864 to float, !dbg !28 + %1857 = fpext bfloat %865 to float, !dbg !28 + %1858 = fpext bfloat %876 to float, !dbg !28 + %1859 = fpext bfloat %877 to float, !dbg !28 + %1860 = fpext bfloat %878 to float, !dbg !28 + %1861 = fpext bfloat %879 to float, !dbg !28 + %1862 = fpext bfloat %880 to float, !dbg !28 + %1863 = fpext bfloat %881 to float, !dbg !28 + %1864 = fpext bfloat %882 to float, !dbg !28 + %1865 = fpext bfloat %883 to float, !dbg !28 + %1866 = fpext bfloat %894 to float, !dbg !28 + %1867 = fpext bfloat %895 to float, !dbg !28 + %1868 = fpext bfloat %896 to float, !dbg !28 + %1869 = fpext bfloat %897 to float, !dbg !28 + %1870 = fpext bfloat %898 to float, !dbg !28 + %1871 = fpext bfloat %899 to float, !dbg !28 + %1872 = fpext bfloat %900 to float, !dbg !28 + %1873 = fpext bfloat %901 to float, !dbg !28 + %1874 = fpext bfloat %912 to float, !dbg !28 + %1875 = fpext bfloat %913 to float, !dbg !28 + %1876 = fpext bfloat %914 to float, !dbg !28 + %1877 = fpext bfloat %915 to float, !dbg !28 + %1878 = fpext bfloat %916 to float, !dbg !28 + %1879 = fpext bfloat %917 to float, !dbg !28 + %1880 = fpext bfloat %918 to float, !dbg !28 + %1881 = fpext bfloat %919 to float, !dbg !28 + %1882 = fpext bfloat %930 to float, !dbg !28 + %1883 = fpext bfloat %931 to float, !dbg !28 + %1884 = fpext bfloat %932 to float, !dbg !28 + %1885 = fpext bfloat %933 to float, !dbg !28 + %1886 = fpext bfloat %934 to float, !dbg !28 + %1887 = fpext bfloat %935 to float, !dbg !28 + %1888 = fpext bfloat %936 to float, !dbg !28 + %1889 = fpext bfloat %937 to float, !dbg !28 + %1890 = fpext bfloat %948 to float, !dbg !28 + %1891 = fpext bfloat %949 to float, !dbg !28 + %1892 = fpext bfloat %950 to float, !dbg !28 + %1893 = fpext bfloat %951 to float, !dbg !28 + %1894 = fpext bfloat %952 to float, !dbg !28 + %1895 = fpext bfloat %953 to float, !dbg !28 + %1896 = fpext bfloat %954 to float, !dbg !28 + %1897 = fpext bfloat %955 to float, !dbg !28 + %1898 = fpext bfloat %966 to float, !dbg !28 + %1899 = fpext bfloat %967 to float, !dbg !28 + %1900 = fpext bfloat %968 to float, !dbg !28 + %1901 = fpext bfloat %969 to float, !dbg !28 + %1902 = fpext bfloat %970 to float, !dbg !28 + %1903 = fpext bfloat %971 to float, !dbg !28 + %1904 = fpext bfloat %972 to float, !dbg !28 + %1905 = fpext bfloat %973 to float, !dbg !28 + %1906 = fpext bfloat %984 to float, !dbg !28 + %1907 = fpext bfloat %985 to float, !dbg !28 + %1908 = fpext bfloat %986 to float, !dbg !28 + %1909 = fpext bfloat %987 to float, !dbg !28 + %1910 = fpext bfloat %988 to float, !dbg !28 + %1911 = fpext bfloat %989 to float, !dbg !28 + %1912 = fpext bfloat %990 to float, !dbg !28 + %1913 = fpext bfloat %991 to float, !dbg !28 + %1914 = fpext bfloat %1002 to float, !dbg !28 + %1915 = fpext bfloat %1003 to float, !dbg !28 + %1916 = fpext bfloat %1004 to float, !dbg !28 + %1917 = fpext bfloat %1005 to float, !dbg !28 + %1918 = fpext bfloat %1006 to float, !dbg !28 + %1919 = fpext bfloat %1007 to float, !dbg !28 + %1920 = fpext bfloat %1008 to float, !dbg !28 + %1921 = fpext bfloat %1009 to float, !dbg !28 + %1922 = fpext bfloat %1020 to float, !dbg !28 + %1923 = fpext bfloat %1021 to float, !dbg !28 + %1924 = fpext bfloat %1022 to float, !dbg !28 + %1925 = fpext bfloat %1023 to float, !dbg !28 + %1926 = fpext bfloat %1024 to float, !dbg !28 + %1927 = fpext bfloat %1025 to float, !dbg !28 + %1928 = fpext bfloat %1026 to float, !dbg !28 + %1929 = fpext bfloat %1027 to float, !dbg !28 + %1930 = fpext bfloat %1038 to float, !dbg !28 + %1931 = fpext bfloat %1039 to float, !dbg !28 + %1932 = fpext bfloat %1040 to float, !dbg !28 + %1933 = fpext bfloat %1041 to float, !dbg !28 + %1934 = fpext bfloat %1042 to float, !dbg !28 + %1935 = fpext bfloat %1043 to float, !dbg !28 + %1936 = fpext bfloat %1044 to float, !dbg !28 + %1937 = fpext bfloat %1045 to float, !dbg !28 + %1938 = fpext bfloat %1056 to float, !dbg !28 + %1939 = fpext bfloat %1057 to float, !dbg !28 + %1940 = fpext bfloat %1058 to float, !dbg !28 + %1941 = fpext bfloat %1059 to float, !dbg !28 + %1942 = fpext bfloat %1060 to float, !dbg !28 + %1943 = fpext bfloat %1061 to float, !dbg !28 + %1944 = fpext bfloat %1062 to float, !dbg !28 + %1945 = fpext bfloat %1063 to float, !dbg !28 + %1946 = fpext bfloat %1074 to float, !dbg !28 + %1947 = fpext bfloat %1075 to float, !dbg !28 + %1948 = fpext bfloat %1076 to float, !dbg !28 + %1949 = fpext bfloat %1077 to float, !dbg !28 + %1950 = fpext bfloat %1078 to float, !dbg !28 + %1951 = fpext bfloat %1079 to float, !dbg !28 + %1952 = fpext bfloat %1080 to float, !dbg !28 + %1953 = fpext bfloat %1081 to float, !dbg !28 + %1954 = fpext bfloat %1092 to float, !dbg !28 + %1955 = fpext bfloat %1093 to float, !dbg !28 + %1956 = fpext bfloat %1094 to float, !dbg !28 + %1957 = fpext bfloat %1095 to float, !dbg !28 + %1958 = fpext bfloat %1096 to float, !dbg !28 + %1959 = fpext bfloat %1097 to float, !dbg !28 + %1960 = fpext bfloat %1098 to float, !dbg !28 + %1961 = fpext bfloat %1099 to float, !dbg !28 + %1962 = fpext bfloat %1110 to float, !dbg !28 + %1963 = fpext bfloat %1111 to float, !dbg !28 + %1964 = fpext bfloat %1112 to float, !dbg !28 + %1965 = fpext bfloat %1113 to float, !dbg !28 + %1966 = fpext bfloat %1114 to float, !dbg !28 + %1967 = fpext bfloat %1115 to float, !dbg !28 + %1968 = fpext bfloat %1116 to float, !dbg !28 + %1969 = fpext bfloat %1117 to float, !dbg !28 + %1970 = fpext bfloat %1128 to float, !dbg !28 + %1971 = fpext bfloat %1129 to float, !dbg !28 + %1972 = fpext bfloat %1130 to float, !dbg !28 + %1973 = fpext bfloat %1131 to float, !dbg !28 + %1974 = fpext bfloat %1132 to float, !dbg !28 + %1975 = fpext bfloat %1133 to float, !dbg !28 + %1976 = fpext bfloat %1134 to float, !dbg !28 + %1977 = fpext bfloat %1135 to float, !dbg !28 + %1978 = fpext bfloat %1146 to float, !dbg !28 + %1979 = fpext bfloat %1147 to float, !dbg !28 + %1980 = fpext bfloat %1148 to float, !dbg !28 + %1981 = fpext bfloat %1149 to float, !dbg !28 + %1982 = fpext bfloat %1150 to float, !dbg !28 + %1983 = fpext bfloat %1151 to float, !dbg !28 + %1984 = fpext bfloat %1152 to float, !dbg !28 + %1985 = fpext bfloat %1153 to float, !dbg !28 + %1986 = fpext bfloat %1164 to float, !dbg !28 + %1987 = fpext bfloat %1165 to float, !dbg !28 + %1988 = fpext bfloat %1166 to float, !dbg !28 + %1989 = fpext bfloat %1167 to float, !dbg !28 + %1990 = fpext bfloat %1168 to float, !dbg !28 + %1991 = fpext bfloat %1169 to float, !dbg !28 + %1992 = fpext bfloat %1170 to float, !dbg !28 + %1993 = fpext bfloat %1171 to float, !dbg !28 + %1994 = fpext bfloat %1182 to float, !dbg !28 + %1995 = fpext bfloat %1183 to float, !dbg !28 + %1996 = fpext bfloat %1184 to float, !dbg !28 + %1997 = fpext bfloat %1185 to float, !dbg !28 + %1998 = fpext bfloat %1186 to float, !dbg !28 + %1999 = fpext bfloat %1187 to float, !dbg !28 + %2000 = fpext bfloat %1188 to float, !dbg !28 + %2001 = fpext bfloat %1189 to float, !dbg !28 + %2002 = fpext bfloat %1200 to float, !dbg !28 + %2003 = fpext bfloat %1201 to float, !dbg !28 + %2004 = fpext bfloat %1202 to float, !dbg !28 + %2005 = fpext bfloat %1203 to float, !dbg !28 + %2006 = fpext bfloat %1204 to float, !dbg !28 + %2007 = fpext bfloat %1205 to float, !dbg !28 + %2008 = fpext bfloat %1206 to float, !dbg !28 + %2009 = fpext bfloat %1207 to float, !dbg !28 + %2010 = fpext bfloat %1218 to float, !dbg !28 + %2011 = fpext bfloat %1219 to float, !dbg !28 + %2012 = fpext bfloat %1220 to float, !dbg !28 + %2013 = fpext bfloat %1221 to float, !dbg !28 + %2014 = fpext bfloat %1222 to float, !dbg !28 + %2015 = fpext bfloat %1223 to float, !dbg !28 + %2016 = fpext bfloat %1224 to float, !dbg !28 + %2017 = fpext bfloat %1225 to float, !dbg !28 + %2018 = fpext bfloat %1236 to float, !dbg !28 + %2019 = fpext bfloat %1237 to float, !dbg !28 + %2020 = fpext bfloat %1238 to float, !dbg !28 + %2021 = fpext bfloat %1239 to float, !dbg !28 + %2022 = fpext bfloat %1240 to float, !dbg !28 + %2023 = fpext bfloat %1241 to float, !dbg !28 + %2024 = fpext bfloat %1242 to float, !dbg !28 + %2025 = fpext bfloat %1243 to float, !dbg !28 + %2026 = fpext bfloat %1254 to float, !dbg !28 + %2027 = fpext bfloat %1255 to float, !dbg !28 + %2028 = fpext bfloat %1256 to float, !dbg !28 + %2029 = fpext bfloat %1257 to float, !dbg !28 + %2030 = fpext bfloat %1258 to float, !dbg !28 + %2031 = fpext bfloat %1259 to float, !dbg !28 + %2032 = fpext bfloat %1260 to float, !dbg !28 + %2033 = fpext bfloat %1261 to float, !dbg !28 + %2034 = fpext bfloat %1272 to float, !dbg !28 + %2035 = fpext bfloat %1273 to float, !dbg !28 + %2036 = fpext bfloat %1274 to float, !dbg !28 + %2037 = fpext bfloat %1275 to float, !dbg !28 + %2038 = fpext bfloat %1276 to float, !dbg !28 + %2039 = fpext bfloat %1277 to float, !dbg !28 + %2040 = fpext bfloat %1278 to float, !dbg !28 + %2041 = fpext bfloat %1279 to float, !dbg !28 + %2042 = fpext bfloat %1290 to float, !dbg !28 + %2043 = fpext bfloat %1291 to float, !dbg !28 + %2044 = fpext bfloat %1292 to float, !dbg !28 + %2045 = fpext bfloat %1293 to float, !dbg !28 + %2046 = fpext bfloat %1294 to float, !dbg !28 + %2047 = fpext bfloat %1295 to float, !dbg !28 + %2048 = fpext bfloat %1296 to float, !dbg !28 + %2049 = fpext bfloat %1297 to float, !dbg !28 + %2050 = fpext bfloat %1308 to float, !dbg !28 + %2051 = fpext bfloat %1309 to float, !dbg !28 + %2052 = fpext bfloat %1310 to float, !dbg !28 + %2053 = fpext bfloat %1311 to float, !dbg !28 + %2054 = fpext bfloat %1312 to float, !dbg !28 + %2055 = fpext bfloat %1313 to float, !dbg !28 + %2056 = fpext bfloat %1314 to float, !dbg !28 + %2057 = fpext bfloat %1315 to float, !dbg !28 + %2058 = fpext bfloat %1326 to float, !dbg !28 + %2059 = fpext bfloat %1327 to float, !dbg !28 + %2060 = fpext bfloat %1328 to float, !dbg !28 + %2061 = fpext bfloat %1329 to float, !dbg !28 + %2062 = fpext bfloat %1330 to float, !dbg !28 + %2063 = fpext bfloat %1331 to float, !dbg !28 + %2064 = fpext bfloat %1332 to float, !dbg !28 + %2065 = fpext bfloat %1333 to float, !dbg !28 + %2066 = fpext bfloat %1344 to float, !dbg !28 + %2067 = fpext bfloat %1345 to float, !dbg !28 + %2068 = fpext bfloat %1346 to float, !dbg !28 + %2069 = fpext bfloat %1347 to float, !dbg !28 + %2070 = fpext bfloat %1348 to float, !dbg !28 + %2071 = fpext bfloat %1349 to float, !dbg !28 + %2072 = fpext bfloat %1350 to float, !dbg !28 + %2073 = fpext bfloat %1351 to float, !dbg !28 + %2074 = fpext bfloat %1362 to float, !dbg !28 + %2075 = fpext bfloat %1363 to float, !dbg !28 + %2076 = fpext bfloat %1364 to float, !dbg !28 + %2077 = fpext bfloat %1365 to float, !dbg !28 + %2078 = fpext bfloat %1366 to float, !dbg !28 + %2079 = fpext bfloat %1367 to float, !dbg !28 + %2080 = fpext bfloat %1368 to float, !dbg !28 + %2081 = fpext bfloat %1369 to float, !dbg !28 + %2082 = fpext bfloat %1380 to float, !dbg !28 + %2083 = fpext bfloat %1381 to float, !dbg !28 + %2084 = fpext bfloat %1382 to float, !dbg !28 + %2085 = fpext bfloat %1383 to float, !dbg !28 + %2086 = fpext bfloat %1384 to float, !dbg !28 + %2087 = fpext bfloat %1385 to float, !dbg !28 + %2088 = fpext bfloat %1386 to float, !dbg !28 + %2089 = fpext bfloat %1387 to float, !dbg !28 + %2090 = fpext bfloat %1398 to float, !dbg !28 + %2091 = fpext bfloat %1399 to float, !dbg !28 + %2092 = fpext bfloat %1400 to float, !dbg !28 + %2093 = fpext bfloat %1401 to float, !dbg !28 + %2094 = fpext bfloat %1402 to float, !dbg !28 + %2095 = fpext bfloat %1403 to float, !dbg !28 + %2096 = fpext bfloat %1404 to float, !dbg !28 + %2097 = fpext bfloat %1405 to float, !dbg !28 + %2098 = fpext bfloat %1416 to float, !dbg !28 + %2099 = fpext bfloat %1417 to float, !dbg !28 + %2100 = fpext bfloat %1418 to float, !dbg !28 + %2101 = fpext bfloat %1419 to float, !dbg !28 + %2102 = fpext bfloat %1420 to float, !dbg !28 + %2103 = fpext bfloat %1421 to float, !dbg !28 + %2104 = fpext bfloat %1422 to float, !dbg !28 + %2105 = fpext bfloat %1423 to float, !dbg !28 + %2106 = fpext bfloat %1434 to float, !dbg !28 + %2107 = fpext bfloat %1435 to float, !dbg !28 + %2108 = fpext bfloat %1436 to float, !dbg !28 + %2109 = fpext bfloat %1437 to float, !dbg !28 + %2110 = fpext bfloat %1438 to float, !dbg !28 + %2111 = fpext bfloat %1439 to float, !dbg !28 + %2112 = fpext bfloat %1440 to float, !dbg !28 + %2113 = fpext bfloat %1441 to float, !dbg !28 + %2114 = fpext bfloat %1452 to float, !dbg !28 + %2115 = fpext bfloat %1453 to float, !dbg !28 + %2116 = fpext bfloat %1454 to float, !dbg !28 + %2117 = fpext bfloat %1455 to float, !dbg !28 + %2118 = fpext bfloat %1456 to float, !dbg !28 + %2119 = fpext bfloat %1457 to float, !dbg !28 + %2120 = fpext bfloat %1458 to float, !dbg !28 + %2121 = fpext bfloat %1459 to float, !dbg !28 + %2122 = fpext bfloat %1470 to float, !dbg !28 + %2123 = fpext bfloat %1471 to float, !dbg !28 + %2124 = fpext bfloat %1472 to float, !dbg !28 + %2125 = fpext bfloat %1473 to float, !dbg !28 + %2126 = fpext bfloat %1474 to float, !dbg !28 + %2127 = fpext bfloat %1475 to float, !dbg !28 + %2128 = fpext bfloat %1476 to float, !dbg !28 + %2129 = fpext bfloat %1477 to float, !dbg !28 + %2130 = fpext bfloat %1488 to float, !dbg !28 + %2131 = fpext bfloat %1489 to float, !dbg !28 + %2132 = fpext bfloat %1490 to float, !dbg !28 + %2133 = fpext bfloat %1491 to float, !dbg !28 + %2134 = fpext bfloat %1492 to float, !dbg !28 + %2135 = fpext bfloat %1493 to float, !dbg !28 + %2136 = fpext bfloat %1494 to float, !dbg !28 + %2137 = fpext bfloat %1495 to float, !dbg !28 + %2138 = fpext bfloat %1506 to float, !dbg !28 + %2139 = fpext bfloat %1507 to float, !dbg !28 + %2140 = fpext bfloat %1508 to float, !dbg !28 + %2141 = fpext bfloat %1509 to float, !dbg !28 + %2142 = fpext bfloat %1510 to float, !dbg !28 + %2143 = fpext bfloat %1511 to float, !dbg !28 + %2144 = fpext bfloat %1512 to float, !dbg !28 + %2145 = fpext bfloat %1513 to float, !dbg !28 + %2146 = fpext bfloat %1524 to float, !dbg !28 + %2147 = fpext bfloat %1525 to float, !dbg !28 + %2148 = fpext bfloat %1526 to float, !dbg !28 + %2149 = fpext bfloat %1527 to float, !dbg !28 + %2150 = fpext bfloat %1528 to float, !dbg !28 + %2151 = fpext bfloat %1529 to float, !dbg !28 + %2152 = fpext bfloat %1530 to float, !dbg !28 + %2153 = fpext bfloat %1531 to float, !dbg !28 + %2154 = fpext bfloat %1542 to float, !dbg !28 + %2155 = fpext bfloat %1543 to float, !dbg !28 + %2156 = fpext bfloat %1544 to float, !dbg !28 + %2157 = fpext bfloat %1545 to float, !dbg !28 + %2158 = fpext bfloat %1546 to float, !dbg !28 + %2159 = fpext bfloat %1547 to float, !dbg !28 + %2160 = fpext bfloat %1548 to float, !dbg !28 + %2161 = fpext bfloat %1549 to float, !dbg !28 + %2162 = fpext bfloat %1560 to float, !dbg !28 + %2163 = fpext bfloat %1561 to float, !dbg !28 + %2164 = fpext bfloat %1562 to float, !dbg !28 + %2165 = fpext bfloat %1563 to float, !dbg !28 + %2166 = fpext bfloat %1564 to float, !dbg !28 + %2167 = fpext bfloat %1565 to float, !dbg !28 + %2168 = fpext bfloat %1566 to float, !dbg !28 + %2169 = fpext bfloat %1567 to float, !dbg !28 + %2170 = fpext bfloat %1578 to float, !dbg !28 + %2171 = fpext bfloat %1579 to float, !dbg !28 + %2172 = fpext bfloat %1580 to float, !dbg !28 + %2173 = fpext bfloat %1581 to float, !dbg !28 + %2174 = fpext bfloat %1582 to float, !dbg !28 + %2175 = fpext bfloat %1583 to float, !dbg !28 + %2176 = fpext bfloat %1584 to float, !dbg !28 + %2177 = fpext bfloat %1585 to float, !dbg !28 + %2178 = fpext bfloat %1596 to float, !dbg !28 + %2179 = fpext bfloat %1597 to float, !dbg !28 + %2180 = fpext bfloat %1598 to float, !dbg !28 + %2181 = fpext bfloat %1599 to float, !dbg !28 + %2182 = fpext bfloat %1600 to float, !dbg !28 + %2183 = fpext bfloat %1601 to float, !dbg !28 + %2184 = fpext bfloat %1602 to float, !dbg !28 + %2185 = fpext bfloat %1603 to float, !dbg !28 + %2186 = fpext bfloat %1614 to float, !dbg !28 + %2187 = fpext bfloat %1615 to float, !dbg !28 + %2188 = fpext bfloat %1616 to float, !dbg !28 + %2189 = fpext bfloat %1617 to float, !dbg !28 + %2190 = fpext bfloat %1618 to float, !dbg !28 + %2191 = fpext bfloat %1619 to float, !dbg !28 + %2192 = fpext bfloat %1620 to float, !dbg !28 + %2193 = fpext bfloat %1621 to float, !dbg !28 + %2194 = fpext bfloat %1632 to float, !dbg !28 + %2195 = fpext bfloat %1633 to float, !dbg !28 + %2196 = fpext bfloat %1634 to float, !dbg !28 + %2197 = fpext bfloat %1635 to float, !dbg !28 + %2198 = fpext bfloat %1636 to float, !dbg !28 + %2199 = fpext bfloat %1637 to float, !dbg !28 + %2200 = fpext bfloat %1638 to float, !dbg !28 + %2201 = fpext bfloat %1639 to float, !dbg !28 + %2202 = fpext bfloat %1650 to float, !dbg !28 + %2203 = fpext bfloat %1651 to float, !dbg !28 + %2204 = fpext bfloat %1652 to float, !dbg !28 + %2205 = fpext bfloat %1653 to float, !dbg !28 + %2206 = fpext bfloat %1654 to float, !dbg !28 + %2207 = fpext bfloat %1655 to float, !dbg !28 + %2208 = fpext bfloat %1656 to float, !dbg !28 + %2209 = fpext bfloat %1657 to float, !dbg !28 + %2210 = fpext bfloat %1668 to float, !dbg !28 + %2211 = fpext bfloat %1669 to float, !dbg !28 + %2212 = fpext bfloat %1670 to float, !dbg !28 + %2213 = fpext bfloat %1671 to float, !dbg !28 + %2214 = fpext bfloat %1672 to float, !dbg !28 + %2215 = fpext bfloat %1673 to float, !dbg !28 + %2216 = fpext bfloat %1674 to float, !dbg !28 + %2217 = fpext bfloat %1675 to float, !dbg !28 + %2218 = fpext bfloat %1686 to float, !dbg !28 + %2219 = fpext bfloat %1687 to float, !dbg !28 + %2220 = fpext bfloat %1688 to float, !dbg !28 + %2221 = fpext bfloat %1689 to float, !dbg !28 + %2222 = fpext bfloat %1690 to float, !dbg !28 + %2223 = fpext bfloat %1691 to float, !dbg !28 + %2224 = fpext bfloat %1692 to float, !dbg !28 + %2225 = fpext bfloat %1693 to float, !dbg !28 + %2226 = fpext bfloat %1704 to float, !dbg !28 + %2227 = fpext bfloat %1705 to float, !dbg !28 + %2228 = fpext bfloat %1706 to float, !dbg !28 + %2229 = fpext bfloat %1707 to float, !dbg !28 + %2230 = fpext bfloat %1708 to float, !dbg !28 + %2231 = fpext bfloat %1709 to float, !dbg !28 + %2232 = fpext bfloat %1710 to float, !dbg !28 + %2233 = fpext bfloat %1711 to float, !dbg !28 + %2234 = fpext bfloat %1722 to float, !dbg !28 + %2235 = fpext bfloat %1723 to float, !dbg !28 + %2236 = fpext bfloat %1724 to float, !dbg !28 + %2237 = fpext bfloat %1725 to float, !dbg !28 + %2238 = fpext bfloat %1726 to float, !dbg !28 + %2239 = fpext bfloat %1727 to float, !dbg !28 + %2240 = fpext bfloat %1728 to float, !dbg !28 + %2241 = fpext bfloat %1729 to float, !dbg !28 + %2242 = extractelement <64 x i1> %62, i64 3, !dbg !29 + %2243 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %319, i1 %2242) #5, !dbg !29 + %2244 = extractelement <64 x i1> %62, i64 2, !dbg !29 + %2245 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %321, i1 %2244) #5, !dbg !29 + %2246 = extractelement <64 x i1> %62, i64 1, !dbg !29 + %2247 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %323, i1 %2246) #5, !dbg !29 + %2248 = extractelement <64 x i1> %62, i64 0, !dbg !29 + %2249 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %325, i1 %2248) #5, !dbg !29 + %2250 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %390, i1 %27) #5, !dbg !30 + %2251 = extractvalue { i32, i32, i32, i32 } %2250, 0, !dbg !30 + %2252 = bitcast i32 %2251 to <2 x bfloat>, !dbg !30 + %2253 = extractvalue { i32, i32, i32, i32 } %2250, 1, !dbg !30 + %2254 = bitcast i32 %2253 to <2 x bfloat>, !dbg !30 + %2255 = extractvalue { i32, i32, i32, i32 } %2250, 2, !dbg !30 + %2256 = bitcast i32 %2255 to <2 x bfloat>, !dbg !30 + %2257 = extractvalue { i32, i32, i32, i32 } %2250, 3, !dbg !30 + %2258 = bitcast i32 %2257 to <2 x bfloat>, !dbg !30 + %2259 = extractelement <2 x bfloat> %2252, i64 0, !dbg !30 + %2260 = extractelement <2 x bfloat> %2252, i64 1, !dbg !30 + %2261 = extractelement <2 x bfloat> %2254, i64 0, !dbg !30 + %2262 = extractelement <2 x bfloat> %2254, i64 1, !dbg !30 + %2263 = extractelement <2 x bfloat> %2256, i64 0, !dbg !30 + %2264 = extractelement <2 x bfloat> %2256, i64 1, !dbg !30 + %2265 = extractelement <2 x bfloat> %2258, i64 0, !dbg !30 + %2266 = extractelement <2 x bfloat> %2258, i64 1, !dbg !30 + %2267 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %391, i1 %28) #5, !dbg !30 + %2268 = extractvalue { i32, i32, i32, i32 } %2267, 0, !dbg !30 + %2269 = bitcast i32 %2268 to <2 x bfloat>, !dbg !30 + %2270 = extractvalue { i32, i32, i32, i32 } %2267, 1, !dbg !30 + %2271 = bitcast i32 %2270 to <2 x bfloat>, !dbg !30 + %2272 = extractvalue { i32, i32, i32, i32 } %2267, 2, !dbg !30 + %2273 = bitcast i32 %2272 to <2 x bfloat>, !dbg !30 + %2274 = extractvalue { i32, i32, i32, i32 } %2267, 3, !dbg !30 + %2275 = bitcast i32 %2274 to <2 x bfloat>, !dbg !30 + %2276 = extractelement <2 x bfloat> %2269, i64 0, !dbg !30 + %2277 = extractelement <2 x bfloat> %2269, i64 1, !dbg !30 + %2278 = extractelement <2 x bfloat> %2271, i64 0, !dbg !30 + %2279 = extractelement <2 x bfloat> %2271, i64 1, !dbg !30 + %2280 = extractelement <2 x bfloat> %2273, i64 0, !dbg !30 + %2281 = extractelement <2 x bfloat> %2273, i64 1, !dbg !30 + %2282 = extractelement <2 x bfloat> %2275, i64 0, !dbg !30 + %2283 = extractelement <2 x bfloat> %2275, i64 1, !dbg !30 + %2284 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %392, i1 %29) #5, !dbg !30 + %2285 = extractvalue { i32, i32, i32, i32 } %2284, 0, !dbg !30 + %2286 = bitcast i32 %2285 to <2 x bfloat>, !dbg !30 + %2287 = extractvalue { i32, i32, i32, i32 } %2284, 1, !dbg !30 + %2288 = bitcast i32 %2287 to <2 x bfloat>, !dbg !30 + %2289 = extractvalue { i32, i32, i32, i32 } %2284, 2, !dbg !30 + %2290 = bitcast i32 %2289 to <2 x bfloat>, !dbg !30 + %2291 = extractvalue { i32, i32, i32, i32 } %2284, 3, !dbg !30 + %2292 = bitcast i32 %2291 to <2 x bfloat>, !dbg !30 + %2293 = extractelement <2 x bfloat> %2286, i64 0, !dbg !30 + %2294 = extractelement <2 x bfloat> %2286, i64 1, !dbg !30 + %2295 = extractelement <2 x bfloat> %2288, i64 0, !dbg !30 + %2296 = extractelement <2 x bfloat> %2288, i64 1, !dbg !30 + %2297 = extractelement <2 x bfloat> %2290, i64 0, !dbg !30 + %2298 = extractelement <2 x bfloat> %2290, i64 1, !dbg !30 + %2299 = extractelement <2 x bfloat> %2292, i64 0, !dbg !30 + %2300 = extractelement <2 x bfloat> %2292, i64 1, !dbg !30 + %2301 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %393, i1 %30) #5, !dbg !30 + %2302 = extractvalue { i32, i32, i32, i32 } %2301, 0, !dbg !30 + %2303 = bitcast i32 %2302 to <2 x bfloat>, !dbg !30 + %2304 = extractvalue { i32, i32, i32, i32 } %2301, 1, !dbg !30 + %2305 = bitcast i32 %2304 to <2 x bfloat>, !dbg !30 + %2306 = extractvalue { i32, i32, i32, i32 } %2301, 2, !dbg !30 + %2307 = bitcast i32 %2306 to <2 x bfloat>, !dbg !30 + %2308 = extractvalue { i32, i32, i32, i32 } %2301, 3, !dbg !30 + %2309 = bitcast i32 %2308 to <2 x bfloat>, !dbg !30 + %2310 = extractelement <2 x bfloat> %2303, i64 0, !dbg !30 + %2311 = extractelement <2 x bfloat> %2303, i64 1, !dbg !30 + %2312 = extractelement <2 x bfloat> %2305, i64 0, !dbg !30 + %2313 = extractelement <2 x bfloat> %2305, i64 1, !dbg !30 + %2314 = extractelement <2 x bfloat> %2307, i64 0, !dbg !30 + %2315 = extractelement <2 x bfloat> %2307, i64 1, !dbg !30 + %2316 = extractelement <2 x bfloat> %2309, i64 0, !dbg !30 + %2317 = extractelement <2 x bfloat> %2309, i64 1, !dbg !30 + %2318 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %394, i1 %650) #5, !dbg !30 + %2319 = extractvalue { i32, i32, i32, i32 } %2318, 0, !dbg !30 + %2320 = bitcast i32 %2319 to <2 x bfloat>, !dbg !30 + %2321 = extractvalue { i32, i32, i32, i32 } %2318, 1, !dbg !30 + %2322 = bitcast i32 %2321 to <2 x bfloat>, !dbg !30 + %2323 = extractvalue { i32, i32, i32, i32 } %2318, 2, !dbg !30 + %2324 = bitcast i32 %2323 to <2 x bfloat>, !dbg !30 + %2325 = extractvalue { i32, i32, i32, i32 } %2318, 3, !dbg !30 + %2326 = bitcast i32 %2325 to <2 x bfloat>, !dbg !30 + %2327 = extractelement <2 x bfloat> %2320, i64 0, !dbg !30 + %2328 = extractelement <2 x bfloat> %2320, i64 1, !dbg !30 + %2329 = extractelement <2 x bfloat> %2322, i64 0, !dbg !30 + %2330 = extractelement <2 x bfloat> %2322, i64 1, !dbg !30 + %2331 = extractelement <2 x bfloat> %2324, i64 0, !dbg !30 + %2332 = extractelement <2 x bfloat> %2324, i64 1, !dbg !30 + %2333 = extractelement <2 x bfloat> %2326, i64 0, !dbg !30 + %2334 = extractelement <2 x bfloat> %2326, i64 1, !dbg !30 + %2335 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %395, i1 %668) #5, !dbg !30 + %2336 = extractvalue { i32, i32, i32, i32 } %2335, 0, !dbg !30 + %2337 = bitcast i32 %2336 to <2 x bfloat>, !dbg !30 + %2338 = extractvalue { i32, i32, i32, i32 } %2335, 1, !dbg !30 + %2339 = bitcast i32 %2338 to <2 x bfloat>, !dbg !30 + %2340 = extractvalue { i32, i32, i32, i32 } %2335, 2, !dbg !30 + %2341 = bitcast i32 %2340 to <2 x bfloat>, !dbg !30 + %2342 = extractvalue { i32, i32, i32, i32 } %2335, 3, !dbg !30 + %2343 = bitcast i32 %2342 to <2 x bfloat>, !dbg !30 + %2344 = extractelement <2 x bfloat> %2337, i64 0, !dbg !30 + %2345 = extractelement <2 x bfloat> %2337, i64 1, !dbg !30 + %2346 = extractelement <2 x bfloat> %2339, i64 0, !dbg !30 + %2347 = extractelement <2 x bfloat> %2339, i64 1, !dbg !30 + %2348 = extractelement <2 x bfloat> %2341, i64 0, !dbg !30 + %2349 = extractelement <2 x bfloat> %2341, i64 1, !dbg !30 + %2350 = extractelement <2 x bfloat> %2343, i64 0, !dbg !30 + %2351 = extractelement <2 x bfloat> %2343, i64 1, !dbg !30 + %2352 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %396, i1 %686) #5, !dbg !30 + %2353 = extractvalue { i32, i32, i32, i32 } %2352, 0, !dbg !30 + %2354 = bitcast i32 %2353 to <2 x bfloat>, !dbg !30 + %2355 = extractvalue { i32, i32, i32, i32 } %2352, 1, !dbg !30 + %2356 = bitcast i32 %2355 to <2 x bfloat>, !dbg !30 + %2357 = extractvalue { i32, i32, i32, i32 } %2352, 2, !dbg !30 + %2358 = bitcast i32 %2357 to <2 x bfloat>, !dbg !30 + %2359 = extractvalue { i32, i32, i32, i32 } %2352, 3, !dbg !30 + %2360 = bitcast i32 %2359 to <2 x bfloat>, !dbg !30 + %2361 = extractelement <2 x bfloat> %2354, i64 0, !dbg !30 + %2362 = extractelement <2 x bfloat> %2354, i64 1, !dbg !30 + %2363 = extractelement <2 x bfloat> %2356, i64 0, !dbg !30 + %2364 = extractelement <2 x bfloat> %2356, i64 1, !dbg !30 + %2365 = extractelement <2 x bfloat> %2358, i64 0, !dbg !30 + %2366 = extractelement <2 x bfloat> %2358, i64 1, !dbg !30 + %2367 = extractelement <2 x bfloat> %2360, i64 0, !dbg !30 + %2368 = extractelement <2 x bfloat> %2360, i64 1, !dbg !30 + %2369 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %397, i1 %704) #5, !dbg !30 + %2370 = extractvalue { i32, i32, i32, i32 } %2369, 0, !dbg !30 + %2371 = bitcast i32 %2370 to <2 x bfloat>, !dbg !30 + %2372 = extractvalue { i32, i32, i32, i32 } %2369, 1, !dbg !30 + %2373 = bitcast i32 %2372 to <2 x bfloat>, !dbg !30 + %2374 = extractvalue { i32, i32, i32, i32 } %2369, 2, !dbg !30 + %2375 = bitcast i32 %2374 to <2 x bfloat>, !dbg !30 + %2376 = extractvalue { i32, i32, i32, i32 } %2369, 3, !dbg !30 + %2377 = bitcast i32 %2376 to <2 x bfloat>, !dbg !30 + %2378 = extractelement <2 x bfloat> %2371, i64 0, !dbg !30 + %2379 = extractelement <2 x bfloat> %2371, i64 1, !dbg !30 + %2380 = extractelement <2 x bfloat> %2373, i64 0, !dbg !30 + %2381 = extractelement <2 x bfloat> %2373, i64 1, !dbg !30 + %2382 = extractelement <2 x bfloat> %2375, i64 0, !dbg !30 + %2383 = extractelement <2 x bfloat> %2375, i64 1, !dbg !30 + %2384 = extractelement <2 x bfloat> %2377, i64 0, !dbg !30 + %2385 = extractelement <2 x bfloat> %2377, i64 1, !dbg !30 + %2386 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %398, i1 %722) #5, !dbg !30 + %2387 = extractvalue { i32, i32, i32, i32 } %2386, 0, !dbg !30 + %2388 = bitcast i32 %2387 to <2 x bfloat>, !dbg !30 + %2389 = extractvalue { i32, i32, i32, i32 } %2386, 1, !dbg !30 + %2390 = bitcast i32 %2389 to <2 x bfloat>, !dbg !30 + %2391 = extractvalue { i32, i32, i32, i32 } %2386, 2, !dbg !30 + %2392 = bitcast i32 %2391 to <2 x bfloat>, !dbg !30 + %2393 = extractvalue { i32, i32, i32, i32 } %2386, 3, !dbg !30 + %2394 = bitcast i32 %2393 to <2 x bfloat>, !dbg !30 + %2395 = extractelement <2 x bfloat> %2388, i64 0, !dbg !30 + %2396 = extractelement <2 x bfloat> %2388, i64 1, !dbg !30 + %2397 = extractelement <2 x bfloat> %2390, i64 0, !dbg !30 + %2398 = extractelement <2 x bfloat> %2390, i64 1, !dbg !30 + %2399 = extractelement <2 x bfloat> %2392, i64 0, !dbg !30 + %2400 = extractelement <2 x bfloat> %2392, i64 1, !dbg !30 + %2401 = extractelement <2 x bfloat> %2394, i64 0, !dbg !30 + %2402 = extractelement <2 x bfloat> %2394, i64 1, !dbg !30 + %2403 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %399, i1 %740) #5, !dbg !30 + %2404 = extractvalue { i32, i32, i32, i32 } %2403, 0, !dbg !30 + %2405 = bitcast i32 %2404 to <2 x bfloat>, !dbg !30 + %2406 = extractvalue { i32, i32, i32, i32 } %2403, 1, !dbg !30 + %2407 = bitcast i32 %2406 to <2 x bfloat>, !dbg !30 + %2408 = extractvalue { i32, i32, i32, i32 } %2403, 2, !dbg !30 + %2409 = bitcast i32 %2408 to <2 x bfloat>, !dbg !30 + %2410 = extractvalue { i32, i32, i32, i32 } %2403, 3, !dbg !30 + %2411 = bitcast i32 %2410 to <2 x bfloat>, !dbg !30 + %2412 = extractelement <2 x bfloat> %2405, i64 0, !dbg !30 + %2413 = extractelement <2 x bfloat> %2405, i64 1, !dbg !30 + %2414 = extractelement <2 x bfloat> %2407, i64 0, !dbg !30 + %2415 = extractelement <2 x bfloat> %2407, i64 1, !dbg !30 + %2416 = extractelement <2 x bfloat> %2409, i64 0, !dbg !30 + %2417 = extractelement <2 x bfloat> %2409, i64 1, !dbg !30 + %2418 = extractelement <2 x bfloat> %2411, i64 0, !dbg !30 + %2419 = extractelement <2 x bfloat> %2411, i64 1, !dbg !30 + %2420 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %400, i1 %758) #5, !dbg !30 + %2421 = extractvalue { i32, i32, i32, i32 } %2420, 0, !dbg !30 + %2422 = bitcast i32 %2421 to <2 x bfloat>, !dbg !30 + %2423 = extractvalue { i32, i32, i32, i32 } %2420, 1, !dbg !30 + %2424 = bitcast i32 %2423 to <2 x bfloat>, !dbg !30 + %2425 = extractvalue { i32, i32, i32, i32 } %2420, 2, !dbg !30 + %2426 = bitcast i32 %2425 to <2 x bfloat>, !dbg !30 + %2427 = extractvalue { i32, i32, i32, i32 } %2420, 3, !dbg !30 + %2428 = bitcast i32 %2427 to <2 x bfloat>, !dbg !30 + %2429 = extractelement <2 x bfloat> %2422, i64 0, !dbg !30 + %2430 = extractelement <2 x bfloat> %2422, i64 1, !dbg !30 + %2431 = extractelement <2 x bfloat> %2424, i64 0, !dbg !30 + %2432 = extractelement <2 x bfloat> %2424, i64 1, !dbg !30 + %2433 = extractelement <2 x bfloat> %2426, i64 0, !dbg !30 + %2434 = extractelement <2 x bfloat> %2426, i64 1, !dbg !30 + %2435 = extractelement <2 x bfloat> %2428, i64 0, !dbg !30 + %2436 = extractelement <2 x bfloat> %2428, i64 1, !dbg !30 + %2437 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %401, i1 %776) #5, !dbg !30 + %2438 = extractvalue { i32, i32, i32, i32 } %2437, 0, !dbg !30 + %2439 = bitcast i32 %2438 to <2 x bfloat>, !dbg !30 + %2440 = extractvalue { i32, i32, i32, i32 } %2437, 1, !dbg !30 + %2441 = bitcast i32 %2440 to <2 x bfloat>, !dbg !30 + %2442 = extractvalue { i32, i32, i32, i32 } %2437, 2, !dbg !30 + %2443 = bitcast i32 %2442 to <2 x bfloat>, !dbg !30 + %2444 = extractvalue { i32, i32, i32, i32 } %2437, 3, !dbg !30 + %2445 = bitcast i32 %2444 to <2 x bfloat>, !dbg !30 + %2446 = extractelement <2 x bfloat> %2439, i64 0, !dbg !30 + %2447 = extractelement <2 x bfloat> %2439, i64 1, !dbg !30 + %2448 = extractelement <2 x bfloat> %2441, i64 0, !dbg !30 + %2449 = extractelement <2 x bfloat> %2441, i64 1, !dbg !30 + %2450 = extractelement <2 x bfloat> %2443, i64 0, !dbg !30 + %2451 = extractelement <2 x bfloat> %2443, i64 1, !dbg !30 + %2452 = extractelement <2 x bfloat> %2445, i64 0, !dbg !30 + %2453 = extractelement <2 x bfloat> %2445, i64 1, !dbg !30 + %2454 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %402, i1 %794) #5, !dbg !30 + %2455 = extractvalue { i32, i32, i32, i32 } %2454, 0, !dbg !30 + %2456 = bitcast i32 %2455 to <2 x bfloat>, !dbg !30 + %2457 = extractvalue { i32, i32, i32, i32 } %2454, 1, !dbg !30 + %2458 = bitcast i32 %2457 to <2 x bfloat>, !dbg !30 + %2459 = extractvalue { i32, i32, i32, i32 } %2454, 2, !dbg !30 + %2460 = bitcast i32 %2459 to <2 x bfloat>, !dbg !30 + %2461 = extractvalue { i32, i32, i32, i32 } %2454, 3, !dbg !30 + %2462 = bitcast i32 %2461 to <2 x bfloat>, !dbg !30 + %2463 = extractelement <2 x bfloat> %2456, i64 0, !dbg !30 + %2464 = extractelement <2 x bfloat> %2456, i64 1, !dbg !30 + %2465 = extractelement <2 x bfloat> %2458, i64 0, !dbg !30 + %2466 = extractelement <2 x bfloat> %2458, i64 1, !dbg !30 + %2467 = extractelement <2 x bfloat> %2460, i64 0, !dbg !30 + %2468 = extractelement <2 x bfloat> %2460, i64 1, !dbg !30 + %2469 = extractelement <2 x bfloat> %2462, i64 0, !dbg !30 + %2470 = extractelement <2 x bfloat> %2462, i64 1, !dbg !30 + %2471 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %403, i1 %812) #5, !dbg !30 + %2472 = extractvalue { i32, i32, i32, i32 } %2471, 0, !dbg !30 + %2473 = bitcast i32 %2472 to <2 x bfloat>, !dbg !30 + %2474 = extractvalue { i32, i32, i32, i32 } %2471, 1, !dbg !30 + %2475 = bitcast i32 %2474 to <2 x bfloat>, !dbg !30 + %2476 = extractvalue { i32, i32, i32, i32 } %2471, 2, !dbg !30 + %2477 = bitcast i32 %2476 to <2 x bfloat>, !dbg !30 + %2478 = extractvalue { i32, i32, i32, i32 } %2471, 3, !dbg !30 + %2479 = bitcast i32 %2478 to <2 x bfloat>, !dbg !30 + %2480 = extractelement <2 x bfloat> %2473, i64 0, !dbg !30 + %2481 = extractelement <2 x bfloat> %2473, i64 1, !dbg !30 + %2482 = extractelement <2 x bfloat> %2475, i64 0, !dbg !30 + %2483 = extractelement <2 x bfloat> %2475, i64 1, !dbg !30 + %2484 = extractelement <2 x bfloat> %2477, i64 0, !dbg !30 + %2485 = extractelement <2 x bfloat> %2477, i64 1, !dbg !30 + %2486 = extractelement <2 x bfloat> %2479, i64 0, !dbg !30 + %2487 = extractelement <2 x bfloat> %2479, i64 1, !dbg !30 + %2488 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %404, i1 %830) #5, !dbg !30 + %2489 = extractvalue { i32, i32, i32, i32 } %2488, 0, !dbg !30 + %2490 = bitcast i32 %2489 to <2 x bfloat>, !dbg !30 + %2491 = extractvalue { i32, i32, i32, i32 } %2488, 1, !dbg !30 + %2492 = bitcast i32 %2491 to <2 x bfloat>, !dbg !30 + %2493 = extractvalue { i32, i32, i32, i32 } %2488, 2, !dbg !30 + %2494 = bitcast i32 %2493 to <2 x bfloat>, !dbg !30 + %2495 = extractvalue { i32, i32, i32, i32 } %2488, 3, !dbg !30 + %2496 = bitcast i32 %2495 to <2 x bfloat>, !dbg !30 + %2497 = extractelement <2 x bfloat> %2490, i64 0, !dbg !30 + %2498 = extractelement <2 x bfloat> %2490, i64 1, !dbg !30 + %2499 = extractelement <2 x bfloat> %2492, i64 0, !dbg !30 + %2500 = extractelement <2 x bfloat> %2492, i64 1, !dbg !30 + %2501 = extractelement <2 x bfloat> %2494, i64 0, !dbg !30 + %2502 = extractelement <2 x bfloat> %2494, i64 1, !dbg !30 + %2503 = extractelement <2 x bfloat> %2496, i64 0, !dbg !30 + %2504 = extractelement <2 x bfloat> %2496, i64 1, !dbg !30 + %2505 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %405, i1 %848) #5, !dbg !30 + %2506 = extractvalue { i32, i32, i32, i32 } %2505, 0, !dbg !30 + %2507 = bitcast i32 %2506 to <2 x bfloat>, !dbg !30 + %2508 = extractvalue { i32, i32, i32, i32 } %2505, 1, !dbg !30 + %2509 = bitcast i32 %2508 to <2 x bfloat>, !dbg !30 + %2510 = extractvalue { i32, i32, i32, i32 } %2505, 2, !dbg !30 + %2511 = bitcast i32 %2510 to <2 x bfloat>, !dbg !30 + %2512 = extractvalue { i32, i32, i32, i32 } %2505, 3, !dbg !30 + %2513 = bitcast i32 %2512 to <2 x bfloat>, !dbg !30 + %2514 = extractelement <2 x bfloat> %2507, i64 0, !dbg !30 + %2515 = extractelement <2 x bfloat> %2507, i64 1, !dbg !30 + %2516 = extractelement <2 x bfloat> %2509, i64 0, !dbg !30 + %2517 = extractelement <2 x bfloat> %2509, i64 1, !dbg !30 + %2518 = extractelement <2 x bfloat> %2511, i64 0, !dbg !30 + %2519 = extractelement <2 x bfloat> %2511, i64 1, !dbg !30 + %2520 = extractelement <2 x bfloat> %2513, i64 0, !dbg !30 + %2521 = extractelement <2 x bfloat> %2513, i64 1, !dbg !30 + %2522 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %406, i1 %866) #5, !dbg !30 + %2523 = extractvalue { i32, i32, i32, i32 } %2522, 0, !dbg !30 + %2524 = bitcast i32 %2523 to <2 x bfloat>, !dbg !30 + %2525 = extractvalue { i32, i32, i32, i32 } %2522, 1, !dbg !30 + %2526 = bitcast i32 %2525 to <2 x bfloat>, !dbg !30 + %2527 = extractvalue { i32, i32, i32, i32 } %2522, 2, !dbg !30 + %2528 = bitcast i32 %2527 to <2 x bfloat>, !dbg !30 + %2529 = extractvalue { i32, i32, i32, i32 } %2522, 3, !dbg !30 + %2530 = bitcast i32 %2529 to <2 x bfloat>, !dbg !30 + %2531 = extractelement <2 x bfloat> %2524, i64 0, !dbg !30 + %2532 = extractelement <2 x bfloat> %2524, i64 1, !dbg !30 + %2533 = extractelement <2 x bfloat> %2526, i64 0, !dbg !30 + %2534 = extractelement <2 x bfloat> %2526, i64 1, !dbg !30 + %2535 = extractelement <2 x bfloat> %2528, i64 0, !dbg !30 + %2536 = extractelement <2 x bfloat> %2528, i64 1, !dbg !30 + %2537 = extractelement <2 x bfloat> %2530, i64 0, !dbg !30 + %2538 = extractelement <2 x bfloat> %2530, i64 1, !dbg !30 + %2539 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %407, i1 %884) #5, !dbg !30 + %2540 = extractvalue { i32, i32, i32, i32 } %2539, 0, !dbg !30 + %2541 = bitcast i32 %2540 to <2 x bfloat>, !dbg !30 + %2542 = extractvalue { i32, i32, i32, i32 } %2539, 1, !dbg !30 + %2543 = bitcast i32 %2542 to <2 x bfloat>, !dbg !30 + %2544 = extractvalue { i32, i32, i32, i32 } %2539, 2, !dbg !30 + %2545 = bitcast i32 %2544 to <2 x bfloat>, !dbg !30 + %2546 = extractvalue { i32, i32, i32, i32 } %2539, 3, !dbg !30 + %2547 = bitcast i32 %2546 to <2 x bfloat>, !dbg !30 + %2548 = extractelement <2 x bfloat> %2541, i64 0, !dbg !30 + %2549 = extractelement <2 x bfloat> %2541, i64 1, !dbg !30 + %2550 = extractelement <2 x bfloat> %2543, i64 0, !dbg !30 + %2551 = extractelement <2 x bfloat> %2543, i64 1, !dbg !30 + %2552 = extractelement <2 x bfloat> %2545, i64 0, !dbg !30 + %2553 = extractelement <2 x bfloat> %2545, i64 1, !dbg !30 + %2554 = extractelement <2 x bfloat> %2547, i64 0, !dbg !30 + %2555 = extractelement <2 x bfloat> %2547, i64 1, !dbg !30 + %2556 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %408, i1 %902) #5, !dbg !30 + %2557 = extractvalue { i32, i32, i32, i32 } %2556, 0, !dbg !30 + %2558 = bitcast i32 %2557 to <2 x bfloat>, !dbg !30 + %2559 = extractvalue { i32, i32, i32, i32 } %2556, 1, !dbg !30 + %2560 = bitcast i32 %2559 to <2 x bfloat>, !dbg !30 + %2561 = extractvalue { i32, i32, i32, i32 } %2556, 2, !dbg !30 + %2562 = bitcast i32 %2561 to <2 x bfloat>, !dbg !30 + %2563 = extractvalue { i32, i32, i32, i32 } %2556, 3, !dbg !30 + %2564 = bitcast i32 %2563 to <2 x bfloat>, !dbg !30 + %2565 = extractelement <2 x bfloat> %2558, i64 0, !dbg !30 + %2566 = extractelement <2 x bfloat> %2558, i64 1, !dbg !30 + %2567 = extractelement <2 x bfloat> %2560, i64 0, !dbg !30 + %2568 = extractelement <2 x bfloat> %2560, i64 1, !dbg !30 + %2569 = extractelement <2 x bfloat> %2562, i64 0, !dbg !30 + %2570 = extractelement <2 x bfloat> %2562, i64 1, !dbg !30 + %2571 = extractelement <2 x bfloat> %2564, i64 0, !dbg !30 + %2572 = extractelement <2 x bfloat> %2564, i64 1, !dbg !30 + %2573 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %409, i1 %920) #5, !dbg !30 + %2574 = extractvalue { i32, i32, i32, i32 } %2573, 0, !dbg !30 + %2575 = bitcast i32 %2574 to <2 x bfloat>, !dbg !30 + %2576 = extractvalue { i32, i32, i32, i32 } %2573, 1, !dbg !30 + %2577 = bitcast i32 %2576 to <2 x bfloat>, !dbg !30 + %2578 = extractvalue { i32, i32, i32, i32 } %2573, 2, !dbg !30 + %2579 = bitcast i32 %2578 to <2 x bfloat>, !dbg !30 + %2580 = extractvalue { i32, i32, i32, i32 } %2573, 3, !dbg !30 + %2581 = bitcast i32 %2580 to <2 x bfloat>, !dbg !30 + %2582 = extractelement <2 x bfloat> %2575, i64 0, !dbg !30 + %2583 = extractelement <2 x bfloat> %2575, i64 1, !dbg !30 + %2584 = extractelement <2 x bfloat> %2577, i64 0, !dbg !30 + %2585 = extractelement <2 x bfloat> %2577, i64 1, !dbg !30 + %2586 = extractelement <2 x bfloat> %2579, i64 0, !dbg !30 + %2587 = extractelement <2 x bfloat> %2579, i64 1, !dbg !30 + %2588 = extractelement <2 x bfloat> %2581, i64 0, !dbg !30 + %2589 = extractelement <2 x bfloat> %2581, i64 1, !dbg !30 + %2590 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %410, i1 %938) #5, !dbg !30 + %2591 = extractvalue { i32, i32, i32, i32 } %2590, 0, !dbg !30 + %2592 = bitcast i32 %2591 to <2 x bfloat>, !dbg !30 + %2593 = extractvalue { i32, i32, i32, i32 } %2590, 1, !dbg !30 + %2594 = bitcast i32 %2593 to <2 x bfloat>, !dbg !30 + %2595 = extractvalue { i32, i32, i32, i32 } %2590, 2, !dbg !30 + %2596 = bitcast i32 %2595 to <2 x bfloat>, !dbg !30 + %2597 = extractvalue { i32, i32, i32, i32 } %2590, 3, !dbg !30 + %2598 = bitcast i32 %2597 to <2 x bfloat>, !dbg !30 + %2599 = extractelement <2 x bfloat> %2592, i64 0, !dbg !30 + %2600 = extractelement <2 x bfloat> %2592, i64 1, !dbg !30 + %2601 = extractelement <2 x bfloat> %2594, i64 0, !dbg !30 + %2602 = extractelement <2 x bfloat> %2594, i64 1, !dbg !30 + %2603 = extractelement <2 x bfloat> %2596, i64 0, !dbg !30 + %2604 = extractelement <2 x bfloat> %2596, i64 1, !dbg !30 + %2605 = extractelement <2 x bfloat> %2598, i64 0, !dbg !30 + %2606 = extractelement <2 x bfloat> %2598, i64 1, !dbg !30 + %2607 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %411, i1 %956) #5, !dbg !30 + %2608 = extractvalue { i32, i32, i32, i32 } %2607, 0, !dbg !30 + %2609 = bitcast i32 %2608 to <2 x bfloat>, !dbg !30 + %2610 = extractvalue { i32, i32, i32, i32 } %2607, 1, !dbg !30 + %2611 = bitcast i32 %2610 to <2 x bfloat>, !dbg !30 + %2612 = extractvalue { i32, i32, i32, i32 } %2607, 2, !dbg !30 + %2613 = bitcast i32 %2612 to <2 x bfloat>, !dbg !30 + %2614 = extractvalue { i32, i32, i32, i32 } %2607, 3, !dbg !30 + %2615 = bitcast i32 %2614 to <2 x bfloat>, !dbg !30 + %2616 = extractelement <2 x bfloat> %2609, i64 0, !dbg !30 + %2617 = extractelement <2 x bfloat> %2609, i64 1, !dbg !30 + %2618 = extractelement <2 x bfloat> %2611, i64 0, !dbg !30 + %2619 = extractelement <2 x bfloat> %2611, i64 1, !dbg !30 + %2620 = extractelement <2 x bfloat> %2613, i64 0, !dbg !30 + %2621 = extractelement <2 x bfloat> %2613, i64 1, !dbg !30 + %2622 = extractelement <2 x bfloat> %2615, i64 0, !dbg !30 + %2623 = extractelement <2 x bfloat> %2615, i64 1, !dbg !30 + %2624 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %412, i1 %974) #5, !dbg !30 + %2625 = extractvalue { i32, i32, i32, i32 } %2624, 0, !dbg !30 + %2626 = bitcast i32 %2625 to <2 x bfloat>, !dbg !30 + %2627 = extractvalue { i32, i32, i32, i32 } %2624, 1, !dbg !30 + %2628 = bitcast i32 %2627 to <2 x bfloat>, !dbg !30 + %2629 = extractvalue { i32, i32, i32, i32 } %2624, 2, !dbg !30 + %2630 = bitcast i32 %2629 to <2 x bfloat>, !dbg !30 + %2631 = extractvalue { i32, i32, i32, i32 } %2624, 3, !dbg !30 + %2632 = bitcast i32 %2631 to <2 x bfloat>, !dbg !30 + %2633 = extractelement <2 x bfloat> %2626, i64 0, !dbg !30 + %2634 = extractelement <2 x bfloat> %2626, i64 1, !dbg !30 + %2635 = extractelement <2 x bfloat> %2628, i64 0, !dbg !30 + %2636 = extractelement <2 x bfloat> %2628, i64 1, !dbg !30 + %2637 = extractelement <2 x bfloat> %2630, i64 0, !dbg !30 + %2638 = extractelement <2 x bfloat> %2630, i64 1, !dbg !30 + %2639 = extractelement <2 x bfloat> %2632, i64 0, !dbg !30 + %2640 = extractelement <2 x bfloat> %2632, i64 1, !dbg !30 + %2641 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %413, i1 %992) #5, !dbg !30 + %2642 = extractvalue { i32, i32, i32, i32 } %2641, 0, !dbg !30 + %2643 = bitcast i32 %2642 to <2 x bfloat>, !dbg !30 + %2644 = extractvalue { i32, i32, i32, i32 } %2641, 1, !dbg !30 + %2645 = bitcast i32 %2644 to <2 x bfloat>, !dbg !30 + %2646 = extractvalue { i32, i32, i32, i32 } %2641, 2, !dbg !30 + %2647 = bitcast i32 %2646 to <2 x bfloat>, !dbg !30 + %2648 = extractvalue { i32, i32, i32, i32 } %2641, 3, !dbg !30 + %2649 = bitcast i32 %2648 to <2 x bfloat>, !dbg !30 + %2650 = extractelement <2 x bfloat> %2643, i64 0, !dbg !30 + %2651 = extractelement <2 x bfloat> %2643, i64 1, !dbg !30 + %2652 = extractelement <2 x bfloat> %2645, i64 0, !dbg !30 + %2653 = extractelement <2 x bfloat> %2645, i64 1, !dbg !30 + %2654 = extractelement <2 x bfloat> %2647, i64 0, !dbg !30 + %2655 = extractelement <2 x bfloat> %2647, i64 1, !dbg !30 + %2656 = extractelement <2 x bfloat> %2649, i64 0, !dbg !30 + %2657 = extractelement <2 x bfloat> %2649, i64 1, !dbg !30 + %2658 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %414, i1 %1010) #5, !dbg !30 + %2659 = extractvalue { i32, i32, i32, i32 } %2658, 0, !dbg !30 + %2660 = bitcast i32 %2659 to <2 x bfloat>, !dbg !30 + %2661 = extractvalue { i32, i32, i32, i32 } %2658, 1, !dbg !30 + %2662 = bitcast i32 %2661 to <2 x bfloat>, !dbg !30 + %2663 = extractvalue { i32, i32, i32, i32 } %2658, 2, !dbg !30 + %2664 = bitcast i32 %2663 to <2 x bfloat>, !dbg !30 + %2665 = extractvalue { i32, i32, i32, i32 } %2658, 3, !dbg !30 + %2666 = bitcast i32 %2665 to <2 x bfloat>, !dbg !30 + %2667 = extractelement <2 x bfloat> %2660, i64 0, !dbg !30 + %2668 = extractelement <2 x bfloat> %2660, i64 1, !dbg !30 + %2669 = extractelement <2 x bfloat> %2662, i64 0, !dbg !30 + %2670 = extractelement <2 x bfloat> %2662, i64 1, !dbg !30 + %2671 = extractelement <2 x bfloat> %2664, i64 0, !dbg !30 + %2672 = extractelement <2 x bfloat> %2664, i64 1, !dbg !30 + %2673 = extractelement <2 x bfloat> %2666, i64 0, !dbg !30 + %2674 = extractelement <2 x bfloat> %2666, i64 1, !dbg !30 + %2675 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %415, i1 %1028) #5, !dbg !30 + %2676 = extractvalue { i32, i32, i32, i32 } %2675, 0, !dbg !30 + %2677 = bitcast i32 %2676 to <2 x bfloat>, !dbg !30 + %2678 = extractvalue { i32, i32, i32, i32 } %2675, 1, !dbg !30 + %2679 = bitcast i32 %2678 to <2 x bfloat>, !dbg !30 + %2680 = extractvalue { i32, i32, i32, i32 } %2675, 2, !dbg !30 + %2681 = bitcast i32 %2680 to <2 x bfloat>, !dbg !30 + %2682 = extractvalue { i32, i32, i32, i32 } %2675, 3, !dbg !30 + %2683 = bitcast i32 %2682 to <2 x bfloat>, !dbg !30 + %2684 = extractelement <2 x bfloat> %2677, i64 0, !dbg !30 + %2685 = extractelement <2 x bfloat> %2677, i64 1, !dbg !30 + %2686 = extractelement <2 x bfloat> %2679, i64 0, !dbg !30 + %2687 = extractelement <2 x bfloat> %2679, i64 1, !dbg !30 + %2688 = extractelement <2 x bfloat> %2681, i64 0, !dbg !30 + %2689 = extractelement <2 x bfloat> %2681, i64 1, !dbg !30 + %2690 = extractelement <2 x bfloat> %2683, i64 0, !dbg !30 + %2691 = extractelement <2 x bfloat> %2683, i64 1, !dbg !30 + %2692 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %416, i1 %1046) #5, !dbg !30 + %2693 = extractvalue { i32, i32, i32, i32 } %2692, 0, !dbg !30 + %2694 = bitcast i32 %2693 to <2 x bfloat>, !dbg !30 + %2695 = extractvalue { i32, i32, i32, i32 } %2692, 1, !dbg !30 + %2696 = bitcast i32 %2695 to <2 x bfloat>, !dbg !30 + %2697 = extractvalue { i32, i32, i32, i32 } %2692, 2, !dbg !30 + %2698 = bitcast i32 %2697 to <2 x bfloat>, !dbg !30 + %2699 = extractvalue { i32, i32, i32, i32 } %2692, 3, !dbg !30 + %2700 = bitcast i32 %2699 to <2 x bfloat>, !dbg !30 + %2701 = extractelement <2 x bfloat> %2694, i64 0, !dbg !30 + %2702 = extractelement <2 x bfloat> %2694, i64 1, !dbg !30 + %2703 = extractelement <2 x bfloat> %2696, i64 0, !dbg !30 + %2704 = extractelement <2 x bfloat> %2696, i64 1, !dbg !30 + %2705 = extractelement <2 x bfloat> %2698, i64 0, !dbg !30 + %2706 = extractelement <2 x bfloat> %2698, i64 1, !dbg !30 + %2707 = extractelement <2 x bfloat> %2700, i64 0, !dbg !30 + %2708 = extractelement <2 x bfloat> %2700, i64 1, !dbg !30 + %2709 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %417, i1 %1064) #5, !dbg !30 + %2710 = extractvalue { i32, i32, i32, i32 } %2709, 0, !dbg !30 + %2711 = bitcast i32 %2710 to <2 x bfloat>, !dbg !30 + %2712 = extractvalue { i32, i32, i32, i32 } %2709, 1, !dbg !30 + %2713 = bitcast i32 %2712 to <2 x bfloat>, !dbg !30 + %2714 = extractvalue { i32, i32, i32, i32 } %2709, 2, !dbg !30 + %2715 = bitcast i32 %2714 to <2 x bfloat>, !dbg !30 + %2716 = extractvalue { i32, i32, i32, i32 } %2709, 3, !dbg !30 + %2717 = bitcast i32 %2716 to <2 x bfloat>, !dbg !30 + %2718 = extractelement <2 x bfloat> %2711, i64 0, !dbg !30 + %2719 = extractelement <2 x bfloat> %2711, i64 1, !dbg !30 + %2720 = extractelement <2 x bfloat> %2713, i64 0, !dbg !30 + %2721 = extractelement <2 x bfloat> %2713, i64 1, !dbg !30 + %2722 = extractelement <2 x bfloat> %2715, i64 0, !dbg !30 + %2723 = extractelement <2 x bfloat> %2715, i64 1, !dbg !30 + %2724 = extractelement <2 x bfloat> %2717, i64 0, !dbg !30 + %2725 = extractelement <2 x bfloat> %2717, i64 1, !dbg !30 + %2726 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %418, i1 %1082) #5, !dbg !30 + %2727 = extractvalue { i32, i32, i32, i32 } %2726, 0, !dbg !30 + %2728 = bitcast i32 %2727 to <2 x bfloat>, !dbg !30 + %2729 = extractvalue { i32, i32, i32, i32 } %2726, 1, !dbg !30 + %2730 = bitcast i32 %2729 to <2 x bfloat>, !dbg !30 + %2731 = extractvalue { i32, i32, i32, i32 } %2726, 2, !dbg !30 + %2732 = bitcast i32 %2731 to <2 x bfloat>, !dbg !30 + %2733 = extractvalue { i32, i32, i32, i32 } %2726, 3, !dbg !30 + %2734 = bitcast i32 %2733 to <2 x bfloat>, !dbg !30 + %2735 = extractelement <2 x bfloat> %2728, i64 0, !dbg !30 + %2736 = extractelement <2 x bfloat> %2728, i64 1, !dbg !30 + %2737 = extractelement <2 x bfloat> %2730, i64 0, !dbg !30 + %2738 = extractelement <2 x bfloat> %2730, i64 1, !dbg !30 + %2739 = extractelement <2 x bfloat> %2732, i64 0, !dbg !30 + %2740 = extractelement <2 x bfloat> %2732, i64 1, !dbg !30 + %2741 = extractelement <2 x bfloat> %2734, i64 0, !dbg !30 + %2742 = extractelement <2 x bfloat> %2734, i64 1, !dbg !30 + %2743 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %419, i1 %1100) #5, !dbg !30 + %2744 = extractvalue { i32, i32, i32, i32 } %2743, 0, !dbg !30 + %2745 = bitcast i32 %2744 to <2 x bfloat>, !dbg !30 + %2746 = extractvalue { i32, i32, i32, i32 } %2743, 1, !dbg !30 + %2747 = bitcast i32 %2746 to <2 x bfloat>, !dbg !30 + %2748 = extractvalue { i32, i32, i32, i32 } %2743, 2, !dbg !30 + %2749 = bitcast i32 %2748 to <2 x bfloat>, !dbg !30 + %2750 = extractvalue { i32, i32, i32, i32 } %2743, 3, !dbg !30 + %2751 = bitcast i32 %2750 to <2 x bfloat>, !dbg !30 + %2752 = extractelement <2 x bfloat> %2745, i64 0, !dbg !30 + %2753 = extractelement <2 x bfloat> %2745, i64 1, !dbg !30 + %2754 = extractelement <2 x bfloat> %2747, i64 0, !dbg !30 + %2755 = extractelement <2 x bfloat> %2747, i64 1, !dbg !30 + %2756 = extractelement <2 x bfloat> %2749, i64 0, !dbg !30 + %2757 = extractelement <2 x bfloat> %2749, i64 1, !dbg !30 + %2758 = extractelement <2 x bfloat> %2751, i64 0, !dbg !30 + %2759 = extractelement <2 x bfloat> %2751, i64 1, !dbg !30 + %2760 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %420, i1 %1118) #5, !dbg !30 + %2761 = extractvalue { i32, i32, i32, i32 } %2760, 0, !dbg !30 + %2762 = bitcast i32 %2761 to <2 x bfloat>, !dbg !30 + %2763 = extractvalue { i32, i32, i32, i32 } %2760, 1, !dbg !30 + %2764 = bitcast i32 %2763 to <2 x bfloat>, !dbg !30 + %2765 = extractvalue { i32, i32, i32, i32 } %2760, 2, !dbg !30 + %2766 = bitcast i32 %2765 to <2 x bfloat>, !dbg !30 + %2767 = extractvalue { i32, i32, i32, i32 } %2760, 3, !dbg !30 + %2768 = bitcast i32 %2767 to <2 x bfloat>, !dbg !30 + %2769 = extractelement <2 x bfloat> %2762, i64 0, !dbg !30 + %2770 = extractelement <2 x bfloat> %2762, i64 1, !dbg !30 + %2771 = extractelement <2 x bfloat> %2764, i64 0, !dbg !30 + %2772 = extractelement <2 x bfloat> %2764, i64 1, !dbg !30 + %2773 = extractelement <2 x bfloat> %2766, i64 0, !dbg !30 + %2774 = extractelement <2 x bfloat> %2766, i64 1, !dbg !30 + %2775 = extractelement <2 x bfloat> %2768, i64 0, !dbg !30 + %2776 = extractelement <2 x bfloat> %2768, i64 1, !dbg !30 + %2777 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %421, i1 %1136) #5, !dbg !30 + %2778 = extractvalue { i32, i32, i32, i32 } %2777, 0, !dbg !30 + %2779 = bitcast i32 %2778 to <2 x bfloat>, !dbg !30 + %2780 = extractvalue { i32, i32, i32, i32 } %2777, 1, !dbg !30 + %2781 = bitcast i32 %2780 to <2 x bfloat>, !dbg !30 + %2782 = extractvalue { i32, i32, i32, i32 } %2777, 2, !dbg !30 + %2783 = bitcast i32 %2782 to <2 x bfloat>, !dbg !30 + %2784 = extractvalue { i32, i32, i32, i32 } %2777, 3, !dbg !30 + %2785 = bitcast i32 %2784 to <2 x bfloat>, !dbg !30 + %2786 = extractelement <2 x bfloat> %2779, i64 0, !dbg !30 + %2787 = extractelement <2 x bfloat> %2779, i64 1, !dbg !30 + %2788 = extractelement <2 x bfloat> %2781, i64 0, !dbg !30 + %2789 = extractelement <2 x bfloat> %2781, i64 1, !dbg !30 + %2790 = extractelement <2 x bfloat> %2783, i64 0, !dbg !30 + %2791 = extractelement <2 x bfloat> %2783, i64 1, !dbg !30 + %2792 = extractelement <2 x bfloat> %2785, i64 0, !dbg !30 + %2793 = extractelement <2 x bfloat> %2785, i64 1, !dbg !30 + %2794 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %422, i1 %1154) #5, !dbg !30 + %2795 = extractvalue { i32, i32, i32, i32 } %2794, 0, !dbg !30 + %2796 = bitcast i32 %2795 to <2 x bfloat>, !dbg !30 + %2797 = extractvalue { i32, i32, i32, i32 } %2794, 1, !dbg !30 + %2798 = bitcast i32 %2797 to <2 x bfloat>, !dbg !30 + %2799 = extractvalue { i32, i32, i32, i32 } %2794, 2, !dbg !30 + %2800 = bitcast i32 %2799 to <2 x bfloat>, !dbg !30 + %2801 = extractvalue { i32, i32, i32, i32 } %2794, 3, !dbg !30 + %2802 = bitcast i32 %2801 to <2 x bfloat>, !dbg !30 + %2803 = extractelement <2 x bfloat> %2796, i64 0, !dbg !30 + %2804 = extractelement <2 x bfloat> %2796, i64 1, !dbg !30 + %2805 = extractelement <2 x bfloat> %2798, i64 0, !dbg !30 + %2806 = extractelement <2 x bfloat> %2798, i64 1, !dbg !30 + %2807 = extractelement <2 x bfloat> %2800, i64 0, !dbg !30 + %2808 = extractelement <2 x bfloat> %2800, i64 1, !dbg !30 + %2809 = extractelement <2 x bfloat> %2802, i64 0, !dbg !30 + %2810 = extractelement <2 x bfloat> %2802, i64 1, !dbg !30 + %2811 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %423, i1 %1172) #5, !dbg !30 + %2812 = extractvalue { i32, i32, i32, i32 } %2811, 0, !dbg !30 + %2813 = bitcast i32 %2812 to <2 x bfloat>, !dbg !30 + %2814 = extractvalue { i32, i32, i32, i32 } %2811, 1, !dbg !30 + %2815 = bitcast i32 %2814 to <2 x bfloat>, !dbg !30 + %2816 = extractvalue { i32, i32, i32, i32 } %2811, 2, !dbg !30 + %2817 = bitcast i32 %2816 to <2 x bfloat>, !dbg !30 + %2818 = extractvalue { i32, i32, i32, i32 } %2811, 3, !dbg !30 + %2819 = bitcast i32 %2818 to <2 x bfloat>, !dbg !30 + %2820 = extractelement <2 x bfloat> %2813, i64 0, !dbg !30 + %2821 = extractelement <2 x bfloat> %2813, i64 1, !dbg !30 + %2822 = extractelement <2 x bfloat> %2815, i64 0, !dbg !30 + %2823 = extractelement <2 x bfloat> %2815, i64 1, !dbg !30 + %2824 = extractelement <2 x bfloat> %2817, i64 0, !dbg !30 + %2825 = extractelement <2 x bfloat> %2817, i64 1, !dbg !30 + %2826 = extractelement <2 x bfloat> %2819, i64 0, !dbg !30 + %2827 = extractelement <2 x bfloat> %2819, i64 1, !dbg !30 + %2828 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %424, i1 %1190) #5, !dbg !30 + %2829 = extractvalue { i32, i32, i32, i32 } %2828, 0, !dbg !30 + %2830 = bitcast i32 %2829 to <2 x bfloat>, !dbg !30 + %2831 = extractvalue { i32, i32, i32, i32 } %2828, 1, !dbg !30 + %2832 = bitcast i32 %2831 to <2 x bfloat>, !dbg !30 + %2833 = extractvalue { i32, i32, i32, i32 } %2828, 2, !dbg !30 + %2834 = bitcast i32 %2833 to <2 x bfloat>, !dbg !30 + %2835 = extractvalue { i32, i32, i32, i32 } %2828, 3, !dbg !30 + %2836 = bitcast i32 %2835 to <2 x bfloat>, !dbg !30 + %2837 = extractelement <2 x bfloat> %2830, i64 0, !dbg !30 + %2838 = extractelement <2 x bfloat> %2830, i64 1, !dbg !30 + %2839 = extractelement <2 x bfloat> %2832, i64 0, !dbg !30 + %2840 = extractelement <2 x bfloat> %2832, i64 1, !dbg !30 + %2841 = extractelement <2 x bfloat> %2834, i64 0, !dbg !30 + %2842 = extractelement <2 x bfloat> %2834, i64 1, !dbg !30 + %2843 = extractelement <2 x bfloat> %2836, i64 0, !dbg !30 + %2844 = extractelement <2 x bfloat> %2836, i64 1, !dbg !30 + %2845 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %425, i1 %1208) #5, !dbg !30 + %2846 = extractvalue { i32, i32, i32, i32 } %2845, 0, !dbg !30 + %2847 = bitcast i32 %2846 to <2 x bfloat>, !dbg !30 + %2848 = extractvalue { i32, i32, i32, i32 } %2845, 1, !dbg !30 + %2849 = bitcast i32 %2848 to <2 x bfloat>, !dbg !30 + %2850 = extractvalue { i32, i32, i32, i32 } %2845, 2, !dbg !30 + %2851 = bitcast i32 %2850 to <2 x bfloat>, !dbg !30 + %2852 = extractvalue { i32, i32, i32, i32 } %2845, 3, !dbg !30 + %2853 = bitcast i32 %2852 to <2 x bfloat>, !dbg !30 + %2854 = extractelement <2 x bfloat> %2847, i64 0, !dbg !30 + %2855 = extractelement <2 x bfloat> %2847, i64 1, !dbg !30 + %2856 = extractelement <2 x bfloat> %2849, i64 0, !dbg !30 + %2857 = extractelement <2 x bfloat> %2849, i64 1, !dbg !30 + %2858 = extractelement <2 x bfloat> %2851, i64 0, !dbg !30 + %2859 = extractelement <2 x bfloat> %2851, i64 1, !dbg !30 + %2860 = extractelement <2 x bfloat> %2853, i64 0, !dbg !30 + %2861 = extractelement <2 x bfloat> %2853, i64 1, !dbg !30 + %2862 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %426, i1 %1226) #5, !dbg !30 + %2863 = extractvalue { i32, i32, i32, i32 } %2862, 0, !dbg !30 + %2864 = bitcast i32 %2863 to <2 x bfloat>, !dbg !30 + %2865 = extractvalue { i32, i32, i32, i32 } %2862, 1, !dbg !30 + %2866 = bitcast i32 %2865 to <2 x bfloat>, !dbg !30 + %2867 = extractvalue { i32, i32, i32, i32 } %2862, 2, !dbg !30 + %2868 = bitcast i32 %2867 to <2 x bfloat>, !dbg !30 + %2869 = extractvalue { i32, i32, i32, i32 } %2862, 3, !dbg !30 + %2870 = bitcast i32 %2869 to <2 x bfloat>, !dbg !30 + %2871 = extractelement <2 x bfloat> %2864, i64 0, !dbg !30 + %2872 = extractelement <2 x bfloat> %2864, i64 1, !dbg !30 + %2873 = extractelement <2 x bfloat> %2866, i64 0, !dbg !30 + %2874 = extractelement <2 x bfloat> %2866, i64 1, !dbg !30 + %2875 = extractelement <2 x bfloat> %2868, i64 0, !dbg !30 + %2876 = extractelement <2 x bfloat> %2868, i64 1, !dbg !30 + %2877 = extractelement <2 x bfloat> %2870, i64 0, !dbg !30 + %2878 = extractelement <2 x bfloat> %2870, i64 1, !dbg !30 + %2879 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %427, i1 %1244) #5, !dbg !30 + %2880 = extractvalue { i32, i32, i32, i32 } %2879, 0, !dbg !30 + %2881 = bitcast i32 %2880 to <2 x bfloat>, !dbg !30 + %2882 = extractvalue { i32, i32, i32, i32 } %2879, 1, !dbg !30 + %2883 = bitcast i32 %2882 to <2 x bfloat>, !dbg !30 + %2884 = extractvalue { i32, i32, i32, i32 } %2879, 2, !dbg !30 + %2885 = bitcast i32 %2884 to <2 x bfloat>, !dbg !30 + %2886 = extractvalue { i32, i32, i32, i32 } %2879, 3, !dbg !30 + %2887 = bitcast i32 %2886 to <2 x bfloat>, !dbg !30 + %2888 = extractelement <2 x bfloat> %2881, i64 0, !dbg !30 + %2889 = extractelement <2 x bfloat> %2881, i64 1, !dbg !30 + %2890 = extractelement <2 x bfloat> %2883, i64 0, !dbg !30 + %2891 = extractelement <2 x bfloat> %2883, i64 1, !dbg !30 + %2892 = extractelement <2 x bfloat> %2885, i64 0, !dbg !30 + %2893 = extractelement <2 x bfloat> %2885, i64 1, !dbg !30 + %2894 = extractelement <2 x bfloat> %2887, i64 0, !dbg !30 + %2895 = extractelement <2 x bfloat> %2887, i64 1, !dbg !30 + %2896 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %428, i1 %1262) #5, !dbg !30 + %2897 = extractvalue { i32, i32, i32, i32 } %2896, 0, !dbg !30 + %2898 = bitcast i32 %2897 to <2 x bfloat>, !dbg !30 + %2899 = extractvalue { i32, i32, i32, i32 } %2896, 1, !dbg !30 + %2900 = bitcast i32 %2899 to <2 x bfloat>, !dbg !30 + %2901 = extractvalue { i32, i32, i32, i32 } %2896, 2, !dbg !30 + %2902 = bitcast i32 %2901 to <2 x bfloat>, !dbg !30 + %2903 = extractvalue { i32, i32, i32, i32 } %2896, 3, !dbg !30 + %2904 = bitcast i32 %2903 to <2 x bfloat>, !dbg !30 + %2905 = extractelement <2 x bfloat> %2898, i64 0, !dbg !30 + %2906 = extractelement <2 x bfloat> %2898, i64 1, !dbg !30 + %2907 = extractelement <2 x bfloat> %2900, i64 0, !dbg !30 + %2908 = extractelement <2 x bfloat> %2900, i64 1, !dbg !30 + %2909 = extractelement <2 x bfloat> %2902, i64 0, !dbg !30 + %2910 = extractelement <2 x bfloat> %2902, i64 1, !dbg !30 + %2911 = extractelement <2 x bfloat> %2904, i64 0, !dbg !30 + %2912 = extractelement <2 x bfloat> %2904, i64 1, !dbg !30 + %2913 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %429, i1 %1280) #5, !dbg !30 + %2914 = extractvalue { i32, i32, i32, i32 } %2913, 0, !dbg !30 + %2915 = bitcast i32 %2914 to <2 x bfloat>, !dbg !30 + %2916 = extractvalue { i32, i32, i32, i32 } %2913, 1, !dbg !30 + %2917 = bitcast i32 %2916 to <2 x bfloat>, !dbg !30 + %2918 = extractvalue { i32, i32, i32, i32 } %2913, 2, !dbg !30 + %2919 = bitcast i32 %2918 to <2 x bfloat>, !dbg !30 + %2920 = extractvalue { i32, i32, i32, i32 } %2913, 3, !dbg !30 + %2921 = bitcast i32 %2920 to <2 x bfloat>, !dbg !30 + %2922 = extractelement <2 x bfloat> %2915, i64 0, !dbg !30 + %2923 = extractelement <2 x bfloat> %2915, i64 1, !dbg !30 + %2924 = extractelement <2 x bfloat> %2917, i64 0, !dbg !30 + %2925 = extractelement <2 x bfloat> %2917, i64 1, !dbg !30 + %2926 = extractelement <2 x bfloat> %2919, i64 0, !dbg !30 + %2927 = extractelement <2 x bfloat> %2919, i64 1, !dbg !30 + %2928 = extractelement <2 x bfloat> %2921, i64 0, !dbg !30 + %2929 = extractelement <2 x bfloat> %2921, i64 1, !dbg !30 + %2930 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %430, i1 %1298) #5, !dbg !30 + %2931 = extractvalue { i32, i32, i32, i32 } %2930, 0, !dbg !30 + %2932 = bitcast i32 %2931 to <2 x bfloat>, !dbg !30 + %2933 = extractvalue { i32, i32, i32, i32 } %2930, 1, !dbg !30 + %2934 = bitcast i32 %2933 to <2 x bfloat>, !dbg !30 + %2935 = extractvalue { i32, i32, i32, i32 } %2930, 2, !dbg !30 + %2936 = bitcast i32 %2935 to <2 x bfloat>, !dbg !30 + %2937 = extractvalue { i32, i32, i32, i32 } %2930, 3, !dbg !30 + %2938 = bitcast i32 %2937 to <2 x bfloat>, !dbg !30 + %2939 = extractelement <2 x bfloat> %2932, i64 0, !dbg !30 + %2940 = extractelement <2 x bfloat> %2932, i64 1, !dbg !30 + %2941 = extractelement <2 x bfloat> %2934, i64 0, !dbg !30 + %2942 = extractelement <2 x bfloat> %2934, i64 1, !dbg !30 + %2943 = extractelement <2 x bfloat> %2936, i64 0, !dbg !30 + %2944 = extractelement <2 x bfloat> %2936, i64 1, !dbg !30 + %2945 = extractelement <2 x bfloat> %2938, i64 0, !dbg !30 + %2946 = extractelement <2 x bfloat> %2938, i64 1, !dbg !30 + %2947 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %431, i1 %1316) #5, !dbg !30 + %2948 = extractvalue { i32, i32, i32, i32 } %2947, 0, !dbg !30 + %2949 = bitcast i32 %2948 to <2 x bfloat>, !dbg !30 + %2950 = extractvalue { i32, i32, i32, i32 } %2947, 1, !dbg !30 + %2951 = bitcast i32 %2950 to <2 x bfloat>, !dbg !30 + %2952 = extractvalue { i32, i32, i32, i32 } %2947, 2, !dbg !30 + %2953 = bitcast i32 %2952 to <2 x bfloat>, !dbg !30 + %2954 = extractvalue { i32, i32, i32, i32 } %2947, 3, !dbg !30 + %2955 = bitcast i32 %2954 to <2 x bfloat>, !dbg !30 + %2956 = extractelement <2 x bfloat> %2949, i64 0, !dbg !30 + %2957 = extractelement <2 x bfloat> %2949, i64 1, !dbg !30 + %2958 = extractelement <2 x bfloat> %2951, i64 0, !dbg !30 + %2959 = extractelement <2 x bfloat> %2951, i64 1, !dbg !30 + %2960 = extractelement <2 x bfloat> %2953, i64 0, !dbg !30 + %2961 = extractelement <2 x bfloat> %2953, i64 1, !dbg !30 + %2962 = extractelement <2 x bfloat> %2955, i64 0, !dbg !30 + %2963 = extractelement <2 x bfloat> %2955, i64 1, !dbg !30 + %2964 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %432, i1 %1334) #5, !dbg !30 + %2965 = extractvalue { i32, i32, i32, i32 } %2964, 0, !dbg !30 + %2966 = bitcast i32 %2965 to <2 x bfloat>, !dbg !30 + %2967 = extractvalue { i32, i32, i32, i32 } %2964, 1, !dbg !30 + %2968 = bitcast i32 %2967 to <2 x bfloat>, !dbg !30 + %2969 = extractvalue { i32, i32, i32, i32 } %2964, 2, !dbg !30 + %2970 = bitcast i32 %2969 to <2 x bfloat>, !dbg !30 + %2971 = extractvalue { i32, i32, i32, i32 } %2964, 3, !dbg !30 + %2972 = bitcast i32 %2971 to <2 x bfloat>, !dbg !30 + %2973 = extractelement <2 x bfloat> %2966, i64 0, !dbg !30 + %2974 = extractelement <2 x bfloat> %2966, i64 1, !dbg !30 + %2975 = extractelement <2 x bfloat> %2968, i64 0, !dbg !30 + %2976 = extractelement <2 x bfloat> %2968, i64 1, !dbg !30 + %2977 = extractelement <2 x bfloat> %2970, i64 0, !dbg !30 + %2978 = extractelement <2 x bfloat> %2970, i64 1, !dbg !30 + %2979 = extractelement <2 x bfloat> %2972, i64 0, !dbg !30 + %2980 = extractelement <2 x bfloat> %2972, i64 1, !dbg !30 + %2981 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %433, i1 %1352) #5, !dbg !30 + %2982 = extractvalue { i32, i32, i32, i32 } %2981, 0, !dbg !30 + %2983 = bitcast i32 %2982 to <2 x bfloat>, !dbg !30 + %2984 = extractvalue { i32, i32, i32, i32 } %2981, 1, !dbg !30 + %2985 = bitcast i32 %2984 to <2 x bfloat>, !dbg !30 + %2986 = extractvalue { i32, i32, i32, i32 } %2981, 2, !dbg !30 + %2987 = bitcast i32 %2986 to <2 x bfloat>, !dbg !30 + %2988 = extractvalue { i32, i32, i32, i32 } %2981, 3, !dbg !30 + %2989 = bitcast i32 %2988 to <2 x bfloat>, !dbg !30 + %2990 = extractelement <2 x bfloat> %2983, i64 0, !dbg !30 + %2991 = extractelement <2 x bfloat> %2983, i64 1, !dbg !30 + %2992 = extractelement <2 x bfloat> %2985, i64 0, !dbg !30 + %2993 = extractelement <2 x bfloat> %2985, i64 1, !dbg !30 + %2994 = extractelement <2 x bfloat> %2987, i64 0, !dbg !30 + %2995 = extractelement <2 x bfloat> %2987, i64 1, !dbg !30 + %2996 = extractelement <2 x bfloat> %2989, i64 0, !dbg !30 + %2997 = extractelement <2 x bfloat> %2989, i64 1, !dbg !30 + %2998 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %434, i1 %1370) #5, !dbg !30 + %2999 = extractvalue { i32, i32, i32, i32 } %2998, 0, !dbg !30 + %3000 = bitcast i32 %2999 to <2 x bfloat>, !dbg !30 + %3001 = extractvalue { i32, i32, i32, i32 } %2998, 1, !dbg !30 + %3002 = bitcast i32 %3001 to <2 x bfloat>, !dbg !30 + %3003 = extractvalue { i32, i32, i32, i32 } %2998, 2, !dbg !30 + %3004 = bitcast i32 %3003 to <2 x bfloat>, !dbg !30 + %3005 = extractvalue { i32, i32, i32, i32 } %2998, 3, !dbg !30 + %3006 = bitcast i32 %3005 to <2 x bfloat>, !dbg !30 + %3007 = extractelement <2 x bfloat> %3000, i64 0, !dbg !30 + %3008 = extractelement <2 x bfloat> %3000, i64 1, !dbg !30 + %3009 = extractelement <2 x bfloat> %3002, i64 0, !dbg !30 + %3010 = extractelement <2 x bfloat> %3002, i64 1, !dbg !30 + %3011 = extractelement <2 x bfloat> %3004, i64 0, !dbg !30 + %3012 = extractelement <2 x bfloat> %3004, i64 1, !dbg !30 + %3013 = extractelement <2 x bfloat> %3006, i64 0, !dbg !30 + %3014 = extractelement <2 x bfloat> %3006, i64 1, !dbg !30 + %3015 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %435, i1 %1388) #5, !dbg !30 + %3016 = extractvalue { i32, i32, i32, i32 } %3015, 0, !dbg !30 + %3017 = bitcast i32 %3016 to <2 x bfloat>, !dbg !30 + %3018 = extractvalue { i32, i32, i32, i32 } %3015, 1, !dbg !30 + %3019 = bitcast i32 %3018 to <2 x bfloat>, !dbg !30 + %3020 = extractvalue { i32, i32, i32, i32 } %3015, 2, !dbg !30 + %3021 = bitcast i32 %3020 to <2 x bfloat>, !dbg !30 + %3022 = extractvalue { i32, i32, i32, i32 } %3015, 3, !dbg !30 + %3023 = bitcast i32 %3022 to <2 x bfloat>, !dbg !30 + %3024 = extractelement <2 x bfloat> %3017, i64 0, !dbg !30 + %3025 = extractelement <2 x bfloat> %3017, i64 1, !dbg !30 + %3026 = extractelement <2 x bfloat> %3019, i64 0, !dbg !30 + %3027 = extractelement <2 x bfloat> %3019, i64 1, !dbg !30 + %3028 = extractelement <2 x bfloat> %3021, i64 0, !dbg !30 + %3029 = extractelement <2 x bfloat> %3021, i64 1, !dbg !30 + %3030 = extractelement <2 x bfloat> %3023, i64 0, !dbg !30 + %3031 = extractelement <2 x bfloat> %3023, i64 1, !dbg !30 + %3032 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %436, i1 %1406) #5, !dbg !30 + %3033 = extractvalue { i32, i32, i32, i32 } %3032, 0, !dbg !30 + %3034 = bitcast i32 %3033 to <2 x bfloat>, !dbg !30 + %3035 = extractvalue { i32, i32, i32, i32 } %3032, 1, !dbg !30 + %3036 = bitcast i32 %3035 to <2 x bfloat>, !dbg !30 + %3037 = extractvalue { i32, i32, i32, i32 } %3032, 2, !dbg !30 + %3038 = bitcast i32 %3037 to <2 x bfloat>, !dbg !30 + %3039 = extractvalue { i32, i32, i32, i32 } %3032, 3, !dbg !30 + %3040 = bitcast i32 %3039 to <2 x bfloat>, !dbg !30 + %3041 = extractelement <2 x bfloat> %3034, i64 0, !dbg !30 + %3042 = extractelement <2 x bfloat> %3034, i64 1, !dbg !30 + %3043 = extractelement <2 x bfloat> %3036, i64 0, !dbg !30 + %3044 = extractelement <2 x bfloat> %3036, i64 1, !dbg !30 + %3045 = extractelement <2 x bfloat> %3038, i64 0, !dbg !30 + %3046 = extractelement <2 x bfloat> %3038, i64 1, !dbg !30 + %3047 = extractelement <2 x bfloat> %3040, i64 0, !dbg !30 + %3048 = extractelement <2 x bfloat> %3040, i64 1, !dbg !30 + %3049 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %437, i1 %1424) #5, !dbg !30 + %3050 = extractvalue { i32, i32, i32, i32 } %3049, 0, !dbg !30 + %3051 = bitcast i32 %3050 to <2 x bfloat>, !dbg !30 + %3052 = extractvalue { i32, i32, i32, i32 } %3049, 1, !dbg !30 + %3053 = bitcast i32 %3052 to <2 x bfloat>, !dbg !30 + %3054 = extractvalue { i32, i32, i32, i32 } %3049, 2, !dbg !30 + %3055 = bitcast i32 %3054 to <2 x bfloat>, !dbg !30 + %3056 = extractvalue { i32, i32, i32, i32 } %3049, 3, !dbg !30 + %3057 = bitcast i32 %3056 to <2 x bfloat>, !dbg !30 + %3058 = extractelement <2 x bfloat> %3051, i64 0, !dbg !30 + %3059 = extractelement <2 x bfloat> %3051, i64 1, !dbg !30 + %3060 = extractelement <2 x bfloat> %3053, i64 0, !dbg !30 + %3061 = extractelement <2 x bfloat> %3053, i64 1, !dbg !30 + %3062 = extractelement <2 x bfloat> %3055, i64 0, !dbg !30 + %3063 = extractelement <2 x bfloat> %3055, i64 1, !dbg !30 + %3064 = extractelement <2 x bfloat> %3057, i64 0, !dbg !30 + %3065 = extractelement <2 x bfloat> %3057, i64 1, !dbg !30 + %3066 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %438, i1 %1442) #5, !dbg !30 + %3067 = extractvalue { i32, i32, i32, i32 } %3066, 0, !dbg !30 + %3068 = bitcast i32 %3067 to <2 x bfloat>, !dbg !30 + %3069 = extractvalue { i32, i32, i32, i32 } %3066, 1, !dbg !30 + %3070 = bitcast i32 %3069 to <2 x bfloat>, !dbg !30 + %3071 = extractvalue { i32, i32, i32, i32 } %3066, 2, !dbg !30 + %3072 = bitcast i32 %3071 to <2 x bfloat>, !dbg !30 + %3073 = extractvalue { i32, i32, i32, i32 } %3066, 3, !dbg !30 + %3074 = bitcast i32 %3073 to <2 x bfloat>, !dbg !30 + %3075 = extractelement <2 x bfloat> %3068, i64 0, !dbg !30 + %3076 = extractelement <2 x bfloat> %3068, i64 1, !dbg !30 + %3077 = extractelement <2 x bfloat> %3070, i64 0, !dbg !30 + %3078 = extractelement <2 x bfloat> %3070, i64 1, !dbg !30 + %3079 = extractelement <2 x bfloat> %3072, i64 0, !dbg !30 + %3080 = extractelement <2 x bfloat> %3072, i64 1, !dbg !30 + %3081 = extractelement <2 x bfloat> %3074, i64 0, !dbg !30 + %3082 = extractelement <2 x bfloat> %3074, i64 1, !dbg !30 + %3083 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %439, i1 %1460) #5, !dbg !30 + %3084 = extractvalue { i32, i32, i32, i32 } %3083, 0, !dbg !30 + %3085 = bitcast i32 %3084 to <2 x bfloat>, !dbg !30 + %3086 = extractvalue { i32, i32, i32, i32 } %3083, 1, !dbg !30 + %3087 = bitcast i32 %3086 to <2 x bfloat>, !dbg !30 + %3088 = extractvalue { i32, i32, i32, i32 } %3083, 2, !dbg !30 + %3089 = bitcast i32 %3088 to <2 x bfloat>, !dbg !30 + %3090 = extractvalue { i32, i32, i32, i32 } %3083, 3, !dbg !30 + %3091 = bitcast i32 %3090 to <2 x bfloat>, !dbg !30 + %3092 = extractelement <2 x bfloat> %3085, i64 0, !dbg !30 + %3093 = extractelement <2 x bfloat> %3085, i64 1, !dbg !30 + %3094 = extractelement <2 x bfloat> %3087, i64 0, !dbg !30 + %3095 = extractelement <2 x bfloat> %3087, i64 1, !dbg !30 + %3096 = extractelement <2 x bfloat> %3089, i64 0, !dbg !30 + %3097 = extractelement <2 x bfloat> %3089, i64 1, !dbg !30 + %3098 = extractelement <2 x bfloat> %3091, i64 0, !dbg !30 + %3099 = extractelement <2 x bfloat> %3091, i64 1, !dbg !30 + %3100 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %440, i1 %1478) #5, !dbg !30 + %3101 = extractvalue { i32, i32, i32, i32 } %3100, 0, !dbg !30 + %3102 = bitcast i32 %3101 to <2 x bfloat>, !dbg !30 + %3103 = extractvalue { i32, i32, i32, i32 } %3100, 1, !dbg !30 + %3104 = bitcast i32 %3103 to <2 x bfloat>, !dbg !30 + %3105 = extractvalue { i32, i32, i32, i32 } %3100, 2, !dbg !30 + %3106 = bitcast i32 %3105 to <2 x bfloat>, !dbg !30 + %3107 = extractvalue { i32, i32, i32, i32 } %3100, 3, !dbg !30 + %3108 = bitcast i32 %3107 to <2 x bfloat>, !dbg !30 + %3109 = extractelement <2 x bfloat> %3102, i64 0, !dbg !30 + %3110 = extractelement <2 x bfloat> %3102, i64 1, !dbg !30 + %3111 = extractelement <2 x bfloat> %3104, i64 0, !dbg !30 + %3112 = extractelement <2 x bfloat> %3104, i64 1, !dbg !30 + %3113 = extractelement <2 x bfloat> %3106, i64 0, !dbg !30 + %3114 = extractelement <2 x bfloat> %3106, i64 1, !dbg !30 + %3115 = extractelement <2 x bfloat> %3108, i64 0, !dbg !30 + %3116 = extractelement <2 x bfloat> %3108, i64 1, !dbg !30 + %3117 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %441, i1 %1496) #5, !dbg !30 + %3118 = extractvalue { i32, i32, i32, i32 } %3117, 0, !dbg !30 + %3119 = bitcast i32 %3118 to <2 x bfloat>, !dbg !30 + %3120 = extractvalue { i32, i32, i32, i32 } %3117, 1, !dbg !30 + %3121 = bitcast i32 %3120 to <2 x bfloat>, !dbg !30 + %3122 = extractvalue { i32, i32, i32, i32 } %3117, 2, !dbg !30 + %3123 = bitcast i32 %3122 to <2 x bfloat>, !dbg !30 + %3124 = extractvalue { i32, i32, i32, i32 } %3117, 3, !dbg !30 + %3125 = bitcast i32 %3124 to <2 x bfloat>, !dbg !30 + %3126 = extractelement <2 x bfloat> %3119, i64 0, !dbg !30 + %3127 = extractelement <2 x bfloat> %3119, i64 1, !dbg !30 + %3128 = extractelement <2 x bfloat> %3121, i64 0, !dbg !30 + %3129 = extractelement <2 x bfloat> %3121, i64 1, !dbg !30 + %3130 = extractelement <2 x bfloat> %3123, i64 0, !dbg !30 + %3131 = extractelement <2 x bfloat> %3123, i64 1, !dbg !30 + %3132 = extractelement <2 x bfloat> %3125, i64 0, !dbg !30 + %3133 = extractelement <2 x bfloat> %3125, i64 1, !dbg !30 + %3134 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %442, i1 %1514) #5, !dbg !30 + %3135 = extractvalue { i32, i32, i32, i32 } %3134, 0, !dbg !30 + %3136 = bitcast i32 %3135 to <2 x bfloat>, !dbg !30 + %3137 = extractvalue { i32, i32, i32, i32 } %3134, 1, !dbg !30 + %3138 = bitcast i32 %3137 to <2 x bfloat>, !dbg !30 + %3139 = extractvalue { i32, i32, i32, i32 } %3134, 2, !dbg !30 + %3140 = bitcast i32 %3139 to <2 x bfloat>, !dbg !30 + %3141 = extractvalue { i32, i32, i32, i32 } %3134, 3, !dbg !30 + %3142 = bitcast i32 %3141 to <2 x bfloat>, !dbg !30 + %3143 = extractelement <2 x bfloat> %3136, i64 0, !dbg !30 + %3144 = extractelement <2 x bfloat> %3136, i64 1, !dbg !30 + %3145 = extractelement <2 x bfloat> %3138, i64 0, !dbg !30 + %3146 = extractelement <2 x bfloat> %3138, i64 1, !dbg !30 + %3147 = extractelement <2 x bfloat> %3140, i64 0, !dbg !30 + %3148 = extractelement <2 x bfloat> %3140, i64 1, !dbg !30 + %3149 = extractelement <2 x bfloat> %3142, i64 0, !dbg !30 + %3150 = extractelement <2 x bfloat> %3142, i64 1, !dbg !30 + %3151 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %443, i1 %1532) #5, !dbg !30 + %3152 = extractvalue { i32, i32, i32, i32 } %3151, 0, !dbg !30 + %3153 = bitcast i32 %3152 to <2 x bfloat>, !dbg !30 + %3154 = extractvalue { i32, i32, i32, i32 } %3151, 1, !dbg !30 + %3155 = bitcast i32 %3154 to <2 x bfloat>, !dbg !30 + %3156 = extractvalue { i32, i32, i32, i32 } %3151, 2, !dbg !30 + %3157 = bitcast i32 %3156 to <2 x bfloat>, !dbg !30 + %3158 = extractvalue { i32, i32, i32, i32 } %3151, 3, !dbg !30 + %3159 = bitcast i32 %3158 to <2 x bfloat>, !dbg !30 + %3160 = extractelement <2 x bfloat> %3153, i64 0, !dbg !30 + %3161 = extractelement <2 x bfloat> %3153, i64 1, !dbg !30 + %3162 = extractelement <2 x bfloat> %3155, i64 0, !dbg !30 + %3163 = extractelement <2 x bfloat> %3155, i64 1, !dbg !30 + %3164 = extractelement <2 x bfloat> %3157, i64 0, !dbg !30 + %3165 = extractelement <2 x bfloat> %3157, i64 1, !dbg !30 + %3166 = extractelement <2 x bfloat> %3159, i64 0, !dbg !30 + %3167 = extractelement <2 x bfloat> %3159, i64 1, !dbg !30 + %3168 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %444, i1 %1550) #5, !dbg !30 + %3169 = extractvalue { i32, i32, i32, i32 } %3168, 0, !dbg !30 + %3170 = bitcast i32 %3169 to <2 x bfloat>, !dbg !30 + %3171 = extractvalue { i32, i32, i32, i32 } %3168, 1, !dbg !30 + %3172 = bitcast i32 %3171 to <2 x bfloat>, !dbg !30 + %3173 = extractvalue { i32, i32, i32, i32 } %3168, 2, !dbg !30 + %3174 = bitcast i32 %3173 to <2 x bfloat>, !dbg !30 + %3175 = extractvalue { i32, i32, i32, i32 } %3168, 3, !dbg !30 + %3176 = bitcast i32 %3175 to <2 x bfloat>, !dbg !30 + %3177 = extractelement <2 x bfloat> %3170, i64 0, !dbg !30 + %3178 = extractelement <2 x bfloat> %3170, i64 1, !dbg !30 + %3179 = extractelement <2 x bfloat> %3172, i64 0, !dbg !30 + %3180 = extractelement <2 x bfloat> %3172, i64 1, !dbg !30 + %3181 = extractelement <2 x bfloat> %3174, i64 0, !dbg !30 + %3182 = extractelement <2 x bfloat> %3174, i64 1, !dbg !30 + %3183 = extractelement <2 x bfloat> %3176, i64 0, !dbg !30 + %3184 = extractelement <2 x bfloat> %3176, i64 1, !dbg !30 + %3185 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %445, i1 %1568) #5, !dbg !30 + %3186 = extractvalue { i32, i32, i32, i32 } %3185, 0, !dbg !30 + %3187 = bitcast i32 %3186 to <2 x bfloat>, !dbg !30 + %3188 = extractvalue { i32, i32, i32, i32 } %3185, 1, !dbg !30 + %3189 = bitcast i32 %3188 to <2 x bfloat>, !dbg !30 + %3190 = extractvalue { i32, i32, i32, i32 } %3185, 2, !dbg !30 + %3191 = bitcast i32 %3190 to <2 x bfloat>, !dbg !30 + %3192 = extractvalue { i32, i32, i32, i32 } %3185, 3, !dbg !30 + %3193 = bitcast i32 %3192 to <2 x bfloat>, !dbg !30 + %3194 = extractelement <2 x bfloat> %3187, i64 0, !dbg !30 + %3195 = extractelement <2 x bfloat> %3187, i64 1, !dbg !30 + %3196 = extractelement <2 x bfloat> %3189, i64 0, !dbg !30 + %3197 = extractelement <2 x bfloat> %3189, i64 1, !dbg !30 + %3198 = extractelement <2 x bfloat> %3191, i64 0, !dbg !30 + %3199 = extractelement <2 x bfloat> %3191, i64 1, !dbg !30 + %3200 = extractelement <2 x bfloat> %3193, i64 0, !dbg !30 + %3201 = extractelement <2 x bfloat> %3193, i64 1, !dbg !30 + %3202 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %446, i1 %1586) #5, !dbg !30 + %3203 = extractvalue { i32, i32, i32, i32 } %3202, 0, !dbg !30 + %3204 = bitcast i32 %3203 to <2 x bfloat>, !dbg !30 + %3205 = extractvalue { i32, i32, i32, i32 } %3202, 1, !dbg !30 + %3206 = bitcast i32 %3205 to <2 x bfloat>, !dbg !30 + %3207 = extractvalue { i32, i32, i32, i32 } %3202, 2, !dbg !30 + %3208 = bitcast i32 %3207 to <2 x bfloat>, !dbg !30 + %3209 = extractvalue { i32, i32, i32, i32 } %3202, 3, !dbg !30 + %3210 = bitcast i32 %3209 to <2 x bfloat>, !dbg !30 + %3211 = extractelement <2 x bfloat> %3204, i64 0, !dbg !30 + %3212 = extractelement <2 x bfloat> %3204, i64 1, !dbg !30 + %3213 = extractelement <2 x bfloat> %3206, i64 0, !dbg !30 + %3214 = extractelement <2 x bfloat> %3206, i64 1, !dbg !30 + %3215 = extractelement <2 x bfloat> %3208, i64 0, !dbg !30 + %3216 = extractelement <2 x bfloat> %3208, i64 1, !dbg !30 + %3217 = extractelement <2 x bfloat> %3210, i64 0, !dbg !30 + %3218 = extractelement <2 x bfloat> %3210, i64 1, !dbg !30 + %3219 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %447, i1 %1604) #5, !dbg !30 + %3220 = extractvalue { i32, i32, i32, i32 } %3219, 0, !dbg !30 + %3221 = bitcast i32 %3220 to <2 x bfloat>, !dbg !30 + %3222 = extractvalue { i32, i32, i32, i32 } %3219, 1, !dbg !30 + %3223 = bitcast i32 %3222 to <2 x bfloat>, !dbg !30 + %3224 = extractvalue { i32, i32, i32, i32 } %3219, 2, !dbg !30 + %3225 = bitcast i32 %3224 to <2 x bfloat>, !dbg !30 + %3226 = extractvalue { i32, i32, i32, i32 } %3219, 3, !dbg !30 + %3227 = bitcast i32 %3226 to <2 x bfloat>, !dbg !30 + %3228 = extractelement <2 x bfloat> %3221, i64 0, !dbg !30 + %3229 = extractelement <2 x bfloat> %3221, i64 1, !dbg !30 + %3230 = extractelement <2 x bfloat> %3223, i64 0, !dbg !30 + %3231 = extractelement <2 x bfloat> %3223, i64 1, !dbg !30 + %3232 = extractelement <2 x bfloat> %3225, i64 0, !dbg !30 + %3233 = extractelement <2 x bfloat> %3225, i64 1, !dbg !30 + %3234 = extractelement <2 x bfloat> %3227, i64 0, !dbg !30 + %3235 = extractelement <2 x bfloat> %3227, i64 1, !dbg !30 + %3236 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %448, i1 %1622) #5, !dbg !30 + %3237 = extractvalue { i32, i32, i32, i32 } %3236, 0, !dbg !30 + %3238 = bitcast i32 %3237 to <2 x bfloat>, !dbg !30 + %3239 = extractvalue { i32, i32, i32, i32 } %3236, 1, !dbg !30 + %3240 = bitcast i32 %3239 to <2 x bfloat>, !dbg !30 + %3241 = extractvalue { i32, i32, i32, i32 } %3236, 2, !dbg !30 + %3242 = bitcast i32 %3241 to <2 x bfloat>, !dbg !30 + %3243 = extractvalue { i32, i32, i32, i32 } %3236, 3, !dbg !30 + %3244 = bitcast i32 %3243 to <2 x bfloat>, !dbg !30 + %3245 = extractelement <2 x bfloat> %3238, i64 0, !dbg !30 + %3246 = extractelement <2 x bfloat> %3238, i64 1, !dbg !30 + %3247 = extractelement <2 x bfloat> %3240, i64 0, !dbg !30 + %3248 = extractelement <2 x bfloat> %3240, i64 1, !dbg !30 + %3249 = extractelement <2 x bfloat> %3242, i64 0, !dbg !30 + %3250 = extractelement <2 x bfloat> %3242, i64 1, !dbg !30 + %3251 = extractelement <2 x bfloat> %3244, i64 0, !dbg !30 + %3252 = extractelement <2 x bfloat> %3244, i64 1, !dbg !30 + %3253 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %449, i1 %1640) #5, !dbg !30 + %3254 = extractvalue { i32, i32, i32, i32 } %3253, 0, !dbg !30 + %3255 = bitcast i32 %3254 to <2 x bfloat>, !dbg !30 + %3256 = extractvalue { i32, i32, i32, i32 } %3253, 1, !dbg !30 + %3257 = bitcast i32 %3256 to <2 x bfloat>, !dbg !30 + %3258 = extractvalue { i32, i32, i32, i32 } %3253, 2, !dbg !30 + %3259 = bitcast i32 %3258 to <2 x bfloat>, !dbg !30 + %3260 = extractvalue { i32, i32, i32, i32 } %3253, 3, !dbg !30 + %3261 = bitcast i32 %3260 to <2 x bfloat>, !dbg !30 + %3262 = extractelement <2 x bfloat> %3255, i64 0, !dbg !30 + %3263 = extractelement <2 x bfloat> %3255, i64 1, !dbg !30 + %3264 = extractelement <2 x bfloat> %3257, i64 0, !dbg !30 + %3265 = extractelement <2 x bfloat> %3257, i64 1, !dbg !30 + %3266 = extractelement <2 x bfloat> %3259, i64 0, !dbg !30 + %3267 = extractelement <2 x bfloat> %3259, i64 1, !dbg !30 + %3268 = extractelement <2 x bfloat> %3261, i64 0, !dbg !30 + %3269 = extractelement <2 x bfloat> %3261, i64 1, !dbg !30 + %3270 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %450, i1 %1658) #5, !dbg !30 + %3271 = extractvalue { i32, i32, i32, i32 } %3270, 0, !dbg !30 + %3272 = bitcast i32 %3271 to <2 x bfloat>, !dbg !30 + %3273 = extractvalue { i32, i32, i32, i32 } %3270, 1, !dbg !30 + %3274 = bitcast i32 %3273 to <2 x bfloat>, !dbg !30 + %3275 = extractvalue { i32, i32, i32, i32 } %3270, 2, !dbg !30 + %3276 = bitcast i32 %3275 to <2 x bfloat>, !dbg !30 + %3277 = extractvalue { i32, i32, i32, i32 } %3270, 3, !dbg !30 + %3278 = bitcast i32 %3277 to <2 x bfloat>, !dbg !30 + %3279 = extractelement <2 x bfloat> %3272, i64 0, !dbg !30 + %3280 = extractelement <2 x bfloat> %3272, i64 1, !dbg !30 + %3281 = extractelement <2 x bfloat> %3274, i64 0, !dbg !30 + %3282 = extractelement <2 x bfloat> %3274, i64 1, !dbg !30 + %3283 = extractelement <2 x bfloat> %3276, i64 0, !dbg !30 + %3284 = extractelement <2 x bfloat> %3276, i64 1, !dbg !30 + %3285 = extractelement <2 x bfloat> %3278, i64 0, !dbg !30 + %3286 = extractelement <2 x bfloat> %3278, i64 1, !dbg !30 + %3287 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %451, i1 %1676) #5, !dbg !30 + %3288 = extractvalue { i32, i32, i32, i32 } %3287, 0, !dbg !30 + %3289 = bitcast i32 %3288 to <2 x bfloat>, !dbg !30 + %3290 = extractvalue { i32, i32, i32, i32 } %3287, 1, !dbg !30 + %3291 = bitcast i32 %3290 to <2 x bfloat>, !dbg !30 + %3292 = extractvalue { i32, i32, i32, i32 } %3287, 2, !dbg !30 + %3293 = bitcast i32 %3292 to <2 x bfloat>, !dbg !30 + %3294 = extractvalue { i32, i32, i32, i32 } %3287, 3, !dbg !30 + %3295 = bitcast i32 %3294 to <2 x bfloat>, !dbg !30 + %3296 = extractelement <2 x bfloat> %3289, i64 0, !dbg !30 + %3297 = extractelement <2 x bfloat> %3289, i64 1, !dbg !30 + %3298 = extractelement <2 x bfloat> %3291, i64 0, !dbg !30 + %3299 = extractelement <2 x bfloat> %3291, i64 1, !dbg !30 + %3300 = extractelement <2 x bfloat> %3293, i64 0, !dbg !30 + %3301 = extractelement <2 x bfloat> %3293, i64 1, !dbg !30 + %3302 = extractelement <2 x bfloat> %3295, i64 0, !dbg !30 + %3303 = extractelement <2 x bfloat> %3295, i64 1, !dbg !30 + %3304 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %452, i1 %1694) #5, !dbg !30 + %3305 = extractvalue { i32, i32, i32, i32 } %3304, 0, !dbg !30 + %3306 = bitcast i32 %3305 to <2 x bfloat>, !dbg !30 + %3307 = extractvalue { i32, i32, i32, i32 } %3304, 1, !dbg !30 + %3308 = bitcast i32 %3307 to <2 x bfloat>, !dbg !30 + %3309 = extractvalue { i32, i32, i32, i32 } %3304, 2, !dbg !30 + %3310 = bitcast i32 %3309 to <2 x bfloat>, !dbg !30 + %3311 = extractvalue { i32, i32, i32, i32 } %3304, 3, !dbg !30 + %3312 = bitcast i32 %3311 to <2 x bfloat>, !dbg !30 + %3313 = extractelement <2 x bfloat> %3306, i64 0, !dbg !30 + %3314 = extractelement <2 x bfloat> %3306, i64 1, !dbg !30 + %3315 = extractelement <2 x bfloat> %3308, i64 0, !dbg !30 + %3316 = extractelement <2 x bfloat> %3308, i64 1, !dbg !30 + %3317 = extractelement <2 x bfloat> %3310, i64 0, !dbg !30 + %3318 = extractelement <2 x bfloat> %3310, i64 1, !dbg !30 + %3319 = extractelement <2 x bfloat> %3312, i64 0, !dbg !30 + %3320 = extractelement <2 x bfloat> %3312, i64 1, !dbg !30 + %3321 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %453, i1 %1712) #5, !dbg !30 + %3322 = extractvalue { i32, i32, i32, i32 } %3321, 0, !dbg !30 + %3323 = bitcast i32 %3322 to <2 x bfloat>, !dbg !30 + %3324 = extractvalue { i32, i32, i32, i32 } %3321, 1, !dbg !30 + %3325 = bitcast i32 %3324 to <2 x bfloat>, !dbg !30 + %3326 = extractvalue { i32, i32, i32, i32 } %3321, 2, !dbg !30 + %3327 = bitcast i32 %3326 to <2 x bfloat>, !dbg !30 + %3328 = extractvalue { i32, i32, i32, i32 } %3321, 3, !dbg !30 + %3329 = bitcast i32 %3328 to <2 x bfloat>, !dbg !30 + %3330 = extractelement <2 x bfloat> %3323, i64 0, !dbg !30 + %3331 = extractelement <2 x bfloat> %3323, i64 1, !dbg !30 + %3332 = extractelement <2 x bfloat> %3325, i64 0, !dbg !30 + %3333 = extractelement <2 x bfloat> %3325, i64 1, !dbg !30 + %3334 = extractelement <2 x bfloat> %3327, i64 0, !dbg !30 + %3335 = extractelement <2 x bfloat> %3327, i64 1, !dbg !30 + %3336 = extractelement <2 x bfloat> %3329, i64 0, !dbg !30 + %3337 = extractelement <2 x bfloat> %3329, i64 1, !dbg !30 + %3338 = fpext bfloat %2259 to float, !dbg !31 + %3339 = fpext bfloat %2260 to float, !dbg !31 + %3340 = fpext bfloat %2261 to float, !dbg !31 + %3341 = fpext bfloat %2262 to float, !dbg !31 + %3342 = fpext bfloat %2263 to float, !dbg !31 + %3343 = fpext bfloat %2264 to float, !dbg !31 + %3344 = fpext bfloat %2265 to float, !dbg !31 + %3345 = fpext bfloat %2266 to float, !dbg !31 + %3346 = fpext bfloat %2276 to float, !dbg !31 + %3347 = fpext bfloat %2277 to float, !dbg !31 + %3348 = fpext bfloat %2278 to float, !dbg !31 + %3349 = fpext bfloat %2279 to float, !dbg !31 + %3350 = fpext bfloat %2280 to float, !dbg !31 + %3351 = fpext bfloat %2281 to float, !dbg !31 + %3352 = fpext bfloat %2282 to float, !dbg !31 + %3353 = fpext bfloat %2283 to float, !dbg !31 + %3354 = fpext bfloat %2293 to float, !dbg !31 + %3355 = fpext bfloat %2294 to float, !dbg !31 + %3356 = fpext bfloat %2295 to float, !dbg !31 + %3357 = fpext bfloat %2296 to float, !dbg !31 + %3358 = fpext bfloat %2297 to float, !dbg !31 + %3359 = fpext bfloat %2298 to float, !dbg !31 + %3360 = fpext bfloat %2299 to float, !dbg !31 + %3361 = fpext bfloat %2300 to float, !dbg !31 + %3362 = fpext bfloat %2310 to float, !dbg !31 + %3363 = fpext bfloat %2311 to float, !dbg !31 + %3364 = fpext bfloat %2312 to float, !dbg !31 + %3365 = fpext bfloat %2313 to float, !dbg !31 + %3366 = fpext bfloat %2314 to float, !dbg !31 + %3367 = fpext bfloat %2315 to float, !dbg !31 + %3368 = fpext bfloat %2316 to float, !dbg !31 + %3369 = fpext bfloat %2317 to float, !dbg !31 + %3370 = fpext bfloat %2327 to float, !dbg !31 + %3371 = fpext bfloat %2328 to float, !dbg !31 + %3372 = fpext bfloat %2329 to float, !dbg !31 + %3373 = fpext bfloat %2330 to float, !dbg !31 + %3374 = fpext bfloat %2331 to float, !dbg !31 + %3375 = fpext bfloat %2332 to float, !dbg !31 + %3376 = fpext bfloat %2333 to float, !dbg !31 + %3377 = fpext bfloat %2334 to float, !dbg !31 + %3378 = fpext bfloat %2344 to float, !dbg !31 + %3379 = fpext bfloat %2345 to float, !dbg !31 + %3380 = fpext bfloat %2346 to float, !dbg !31 + %3381 = fpext bfloat %2347 to float, !dbg !31 + %3382 = fpext bfloat %2348 to float, !dbg !31 + %3383 = fpext bfloat %2349 to float, !dbg !31 + %3384 = fpext bfloat %2350 to float, !dbg !31 + %3385 = fpext bfloat %2351 to float, !dbg !31 + %3386 = fpext bfloat %2361 to float, !dbg !31 + %3387 = fpext bfloat %2362 to float, !dbg !31 + %3388 = fpext bfloat %2363 to float, !dbg !31 + %3389 = fpext bfloat %2364 to float, !dbg !31 + %3390 = fpext bfloat %2365 to float, !dbg !31 + %3391 = fpext bfloat %2366 to float, !dbg !31 + %3392 = fpext bfloat %2367 to float, !dbg !31 + %3393 = fpext bfloat %2368 to float, !dbg !31 + %3394 = fpext bfloat %2378 to float, !dbg !31 + %3395 = fpext bfloat %2379 to float, !dbg !31 + %3396 = fpext bfloat %2380 to float, !dbg !31 + %3397 = fpext bfloat %2381 to float, !dbg !31 + %3398 = fpext bfloat %2382 to float, !dbg !31 + %3399 = fpext bfloat %2383 to float, !dbg !31 + %3400 = fpext bfloat %2384 to float, !dbg !31 + %3401 = fpext bfloat %2385 to float, !dbg !31 + %3402 = fpext bfloat %2395 to float, !dbg !31 + %3403 = fpext bfloat %2396 to float, !dbg !31 + %3404 = fpext bfloat %2397 to float, !dbg !31 + %3405 = fpext bfloat %2398 to float, !dbg !31 + %3406 = fpext bfloat %2399 to float, !dbg !31 + %3407 = fpext bfloat %2400 to float, !dbg !31 + %3408 = fpext bfloat %2401 to float, !dbg !31 + %3409 = fpext bfloat %2402 to float, !dbg !31 + %3410 = fpext bfloat %2412 to float, !dbg !31 + %3411 = fpext bfloat %2413 to float, !dbg !31 + %3412 = fpext bfloat %2414 to float, !dbg !31 + %3413 = fpext bfloat %2415 to float, !dbg !31 + %3414 = fpext bfloat %2416 to float, !dbg !31 + %3415 = fpext bfloat %2417 to float, !dbg !31 + %3416 = fpext bfloat %2418 to float, !dbg !31 + %3417 = fpext bfloat %2419 to float, !dbg !31 + %3418 = fpext bfloat %2429 to float, !dbg !31 + %3419 = fpext bfloat %2430 to float, !dbg !31 + %3420 = fpext bfloat %2431 to float, !dbg !31 + %3421 = fpext bfloat %2432 to float, !dbg !31 + %3422 = fpext bfloat %2433 to float, !dbg !31 + %3423 = fpext bfloat %2434 to float, !dbg !31 + %3424 = fpext bfloat %2435 to float, !dbg !31 + %3425 = fpext bfloat %2436 to float, !dbg !31 + %3426 = fpext bfloat %2446 to float, !dbg !31 + %3427 = fpext bfloat %2447 to float, !dbg !31 + %3428 = fpext bfloat %2448 to float, !dbg !31 + %3429 = fpext bfloat %2449 to float, !dbg !31 + %3430 = fpext bfloat %2450 to float, !dbg !31 + %3431 = fpext bfloat %2451 to float, !dbg !31 + %3432 = fpext bfloat %2452 to float, !dbg !31 + %3433 = fpext bfloat %2453 to float, !dbg !31 + %3434 = fpext bfloat %2463 to float, !dbg !31 + %3435 = fpext bfloat %2464 to float, !dbg !31 + %3436 = fpext bfloat %2465 to float, !dbg !31 + %3437 = fpext bfloat %2466 to float, !dbg !31 + %3438 = fpext bfloat %2467 to float, !dbg !31 + %3439 = fpext bfloat %2468 to float, !dbg !31 + %3440 = fpext bfloat %2469 to float, !dbg !31 + %3441 = fpext bfloat %2470 to float, !dbg !31 + %3442 = fpext bfloat %2480 to float, !dbg !31 + %3443 = fpext bfloat %2481 to float, !dbg !31 + %3444 = fpext bfloat %2482 to float, !dbg !31 + %3445 = fpext bfloat %2483 to float, !dbg !31 + %3446 = fpext bfloat %2484 to float, !dbg !31 + %3447 = fpext bfloat %2485 to float, !dbg !31 + %3448 = fpext bfloat %2486 to float, !dbg !31 + %3449 = fpext bfloat %2487 to float, !dbg !31 + %3450 = fpext bfloat %2497 to float, !dbg !31 + %3451 = fpext bfloat %2498 to float, !dbg !31 + %3452 = fpext bfloat %2499 to float, !dbg !31 + %3453 = fpext bfloat %2500 to float, !dbg !31 + %3454 = fpext bfloat %2501 to float, !dbg !31 + %3455 = fpext bfloat %2502 to float, !dbg !31 + %3456 = fpext bfloat %2503 to float, !dbg !31 + %3457 = fpext bfloat %2504 to float, !dbg !31 + %3458 = fpext bfloat %2514 to float, !dbg !31 + %3459 = fpext bfloat %2515 to float, !dbg !31 + %3460 = fpext bfloat %2516 to float, !dbg !31 + %3461 = fpext bfloat %2517 to float, !dbg !31 + %3462 = fpext bfloat %2518 to float, !dbg !31 + %3463 = fpext bfloat %2519 to float, !dbg !31 + %3464 = fpext bfloat %2520 to float, !dbg !31 + %3465 = fpext bfloat %2521 to float, !dbg !31 + %3466 = fpext bfloat %2531 to float, !dbg !31 + %3467 = fpext bfloat %2532 to float, !dbg !31 + %3468 = fpext bfloat %2533 to float, !dbg !31 + %3469 = fpext bfloat %2534 to float, !dbg !31 + %3470 = fpext bfloat %2535 to float, !dbg !31 + %3471 = fpext bfloat %2536 to float, !dbg !31 + %3472 = fpext bfloat %2537 to float, !dbg !31 + %3473 = fpext bfloat %2538 to float, !dbg !31 + %3474 = fpext bfloat %2548 to float, !dbg !31 + %3475 = fpext bfloat %2549 to float, !dbg !31 + %3476 = fpext bfloat %2550 to float, !dbg !31 + %3477 = fpext bfloat %2551 to float, !dbg !31 + %3478 = fpext bfloat %2552 to float, !dbg !31 + %3479 = fpext bfloat %2553 to float, !dbg !31 + %3480 = fpext bfloat %2554 to float, !dbg !31 + %3481 = fpext bfloat %2555 to float, !dbg !31 + %3482 = fpext bfloat %2565 to float, !dbg !31 + %3483 = fpext bfloat %2566 to float, !dbg !31 + %3484 = fpext bfloat %2567 to float, !dbg !31 + %3485 = fpext bfloat %2568 to float, !dbg !31 + %3486 = fpext bfloat %2569 to float, !dbg !31 + %3487 = fpext bfloat %2570 to float, !dbg !31 + %3488 = fpext bfloat %2571 to float, !dbg !31 + %3489 = fpext bfloat %2572 to float, !dbg !31 + %3490 = fpext bfloat %2582 to float, !dbg !31 + %3491 = fpext bfloat %2583 to float, !dbg !31 + %3492 = fpext bfloat %2584 to float, !dbg !31 + %3493 = fpext bfloat %2585 to float, !dbg !31 + %3494 = fpext bfloat %2586 to float, !dbg !31 + %3495 = fpext bfloat %2587 to float, !dbg !31 + %3496 = fpext bfloat %2588 to float, !dbg !31 + %3497 = fpext bfloat %2589 to float, !dbg !31 + %3498 = fpext bfloat %2599 to float, !dbg !31 + %3499 = fpext bfloat %2600 to float, !dbg !31 + %3500 = fpext bfloat %2601 to float, !dbg !31 + %3501 = fpext bfloat %2602 to float, !dbg !31 + %3502 = fpext bfloat %2603 to float, !dbg !31 + %3503 = fpext bfloat %2604 to float, !dbg !31 + %3504 = fpext bfloat %2605 to float, !dbg !31 + %3505 = fpext bfloat %2606 to float, !dbg !31 + %3506 = fpext bfloat %2616 to float, !dbg !31 + %3507 = fpext bfloat %2617 to float, !dbg !31 + %3508 = fpext bfloat %2618 to float, !dbg !31 + %3509 = fpext bfloat %2619 to float, !dbg !31 + %3510 = fpext bfloat %2620 to float, !dbg !31 + %3511 = fpext bfloat %2621 to float, !dbg !31 + %3512 = fpext bfloat %2622 to float, !dbg !31 + %3513 = fpext bfloat %2623 to float, !dbg !31 + %3514 = fpext bfloat %2633 to float, !dbg !31 + %3515 = fpext bfloat %2634 to float, !dbg !31 + %3516 = fpext bfloat %2635 to float, !dbg !31 + %3517 = fpext bfloat %2636 to float, !dbg !31 + %3518 = fpext bfloat %2637 to float, !dbg !31 + %3519 = fpext bfloat %2638 to float, !dbg !31 + %3520 = fpext bfloat %2639 to float, !dbg !31 + %3521 = fpext bfloat %2640 to float, !dbg !31 + %3522 = fpext bfloat %2650 to float, !dbg !31 + %3523 = fpext bfloat %2651 to float, !dbg !31 + %3524 = fpext bfloat %2652 to float, !dbg !31 + %3525 = fpext bfloat %2653 to float, !dbg !31 + %3526 = fpext bfloat %2654 to float, !dbg !31 + %3527 = fpext bfloat %2655 to float, !dbg !31 + %3528 = fpext bfloat %2656 to float, !dbg !31 + %3529 = fpext bfloat %2657 to float, !dbg !31 + %3530 = fpext bfloat %2667 to float, !dbg !31 + %3531 = fpext bfloat %2668 to float, !dbg !31 + %3532 = fpext bfloat %2669 to float, !dbg !31 + %3533 = fpext bfloat %2670 to float, !dbg !31 + %3534 = fpext bfloat %2671 to float, !dbg !31 + %3535 = fpext bfloat %2672 to float, !dbg !31 + %3536 = fpext bfloat %2673 to float, !dbg !31 + %3537 = fpext bfloat %2674 to float, !dbg !31 + %3538 = fpext bfloat %2684 to float, !dbg !31 + %3539 = fpext bfloat %2685 to float, !dbg !31 + %3540 = fpext bfloat %2686 to float, !dbg !31 + %3541 = fpext bfloat %2687 to float, !dbg !31 + %3542 = fpext bfloat %2688 to float, !dbg !31 + %3543 = fpext bfloat %2689 to float, !dbg !31 + %3544 = fpext bfloat %2690 to float, !dbg !31 + %3545 = fpext bfloat %2691 to float, !dbg !31 + %3546 = fpext bfloat %2701 to float, !dbg !31 + %3547 = fpext bfloat %2702 to float, !dbg !31 + %3548 = fpext bfloat %2703 to float, !dbg !31 + %3549 = fpext bfloat %2704 to float, !dbg !31 + %3550 = fpext bfloat %2705 to float, !dbg !31 + %3551 = fpext bfloat %2706 to float, !dbg !31 + %3552 = fpext bfloat %2707 to float, !dbg !31 + %3553 = fpext bfloat %2708 to float, !dbg !31 + %3554 = fpext bfloat %2718 to float, !dbg !31 + %3555 = fpext bfloat %2719 to float, !dbg !31 + %3556 = fpext bfloat %2720 to float, !dbg !31 + %3557 = fpext bfloat %2721 to float, !dbg !31 + %3558 = fpext bfloat %2722 to float, !dbg !31 + %3559 = fpext bfloat %2723 to float, !dbg !31 + %3560 = fpext bfloat %2724 to float, !dbg !31 + %3561 = fpext bfloat %2725 to float, !dbg !31 + %3562 = fpext bfloat %2735 to float, !dbg !31 + %3563 = fpext bfloat %2736 to float, !dbg !31 + %3564 = fpext bfloat %2737 to float, !dbg !31 + %3565 = fpext bfloat %2738 to float, !dbg !31 + %3566 = fpext bfloat %2739 to float, !dbg !31 + %3567 = fpext bfloat %2740 to float, !dbg !31 + %3568 = fpext bfloat %2741 to float, !dbg !31 + %3569 = fpext bfloat %2742 to float, !dbg !31 + %3570 = fpext bfloat %2752 to float, !dbg !31 + %3571 = fpext bfloat %2753 to float, !dbg !31 + %3572 = fpext bfloat %2754 to float, !dbg !31 + %3573 = fpext bfloat %2755 to float, !dbg !31 + %3574 = fpext bfloat %2756 to float, !dbg !31 + %3575 = fpext bfloat %2757 to float, !dbg !31 + %3576 = fpext bfloat %2758 to float, !dbg !31 + %3577 = fpext bfloat %2759 to float, !dbg !31 + %3578 = fpext bfloat %2769 to float, !dbg !31 + %3579 = fpext bfloat %2770 to float, !dbg !31 + %3580 = fpext bfloat %2771 to float, !dbg !31 + %3581 = fpext bfloat %2772 to float, !dbg !31 + %3582 = fpext bfloat %2773 to float, !dbg !31 + %3583 = fpext bfloat %2774 to float, !dbg !31 + %3584 = fpext bfloat %2775 to float, !dbg !31 + %3585 = fpext bfloat %2776 to float, !dbg !31 + %3586 = fpext bfloat %2786 to float, !dbg !31 + %3587 = fpext bfloat %2787 to float, !dbg !31 + %3588 = fpext bfloat %2788 to float, !dbg !31 + %3589 = fpext bfloat %2789 to float, !dbg !31 + %3590 = fpext bfloat %2790 to float, !dbg !31 + %3591 = fpext bfloat %2791 to float, !dbg !31 + %3592 = fpext bfloat %2792 to float, !dbg !31 + %3593 = fpext bfloat %2793 to float, !dbg !31 + %3594 = fpext bfloat %2803 to float, !dbg !31 + %3595 = fpext bfloat %2804 to float, !dbg !31 + %3596 = fpext bfloat %2805 to float, !dbg !31 + %3597 = fpext bfloat %2806 to float, !dbg !31 + %3598 = fpext bfloat %2807 to float, !dbg !31 + %3599 = fpext bfloat %2808 to float, !dbg !31 + %3600 = fpext bfloat %2809 to float, !dbg !31 + %3601 = fpext bfloat %2810 to float, !dbg !31 + %3602 = fpext bfloat %2820 to float, !dbg !31 + %3603 = fpext bfloat %2821 to float, !dbg !31 + %3604 = fpext bfloat %2822 to float, !dbg !31 + %3605 = fpext bfloat %2823 to float, !dbg !31 + %3606 = fpext bfloat %2824 to float, !dbg !31 + %3607 = fpext bfloat %2825 to float, !dbg !31 + %3608 = fpext bfloat %2826 to float, !dbg !31 + %3609 = fpext bfloat %2827 to float, !dbg !31 + %3610 = fpext bfloat %2837 to float, !dbg !31 + %3611 = fpext bfloat %2838 to float, !dbg !31 + %3612 = fpext bfloat %2839 to float, !dbg !31 + %3613 = fpext bfloat %2840 to float, !dbg !31 + %3614 = fpext bfloat %2841 to float, !dbg !31 + %3615 = fpext bfloat %2842 to float, !dbg !31 + %3616 = fpext bfloat %2843 to float, !dbg !31 + %3617 = fpext bfloat %2844 to float, !dbg !31 + %3618 = fpext bfloat %2854 to float, !dbg !31 + %3619 = fpext bfloat %2855 to float, !dbg !31 + %3620 = fpext bfloat %2856 to float, !dbg !31 + %3621 = fpext bfloat %2857 to float, !dbg !31 + %3622 = fpext bfloat %2858 to float, !dbg !31 + %3623 = fpext bfloat %2859 to float, !dbg !31 + %3624 = fpext bfloat %2860 to float, !dbg !31 + %3625 = fpext bfloat %2861 to float, !dbg !31 + %3626 = fpext bfloat %2871 to float, !dbg !31 + %3627 = fpext bfloat %2872 to float, !dbg !31 + %3628 = fpext bfloat %2873 to float, !dbg !31 + %3629 = fpext bfloat %2874 to float, !dbg !31 + %3630 = fpext bfloat %2875 to float, !dbg !31 + %3631 = fpext bfloat %2876 to float, !dbg !31 + %3632 = fpext bfloat %2877 to float, !dbg !31 + %3633 = fpext bfloat %2878 to float, !dbg !31 + %3634 = fpext bfloat %2888 to float, !dbg !31 + %3635 = fpext bfloat %2889 to float, !dbg !31 + %3636 = fpext bfloat %2890 to float, !dbg !31 + %3637 = fpext bfloat %2891 to float, !dbg !31 + %3638 = fpext bfloat %2892 to float, !dbg !31 + %3639 = fpext bfloat %2893 to float, !dbg !31 + %3640 = fpext bfloat %2894 to float, !dbg !31 + %3641 = fpext bfloat %2895 to float, !dbg !31 + %3642 = fpext bfloat %2905 to float, !dbg !31 + %3643 = fpext bfloat %2906 to float, !dbg !31 + %3644 = fpext bfloat %2907 to float, !dbg !31 + %3645 = fpext bfloat %2908 to float, !dbg !31 + %3646 = fpext bfloat %2909 to float, !dbg !31 + %3647 = fpext bfloat %2910 to float, !dbg !31 + %3648 = fpext bfloat %2911 to float, !dbg !31 + %3649 = fpext bfloat %2912 to float, !dbg !31 + %3650 = fpext bfloat %2922 to float, !dbg !31 + %3651 = fpext bfloat %2923 to float, !dbg !31 + %3652 = fpext bfloat %2924 to float, !dbg !31 + %3653 = fpext bfloat %2925 to float, !dbg !31 + %3654 = fpext bfloat %2926 to float, !dbg !31 + %3655 = fpext bfloat %2927 to float, !dbg !31 + %3656 = fpext bfloat %2928 to float, !dbg !31 + %3657 = fpext bfloat %2929 to float, !dbg !31 + %3658 = fpext bfloat %2939 to float, !dbg !31 + %3659 = fpext bfloat %2940 to float, !dbg !31 + %3660 = fpext bfloat %2941 to float, !dbg !31 + %3661 = fpext bfloat %2942 to float, !dbg !31 + %3662 = fpext bfloat %2943 to float, !dbg !31 + %3663 = fpext bfloat %2944 to float, !dbg !31 + %3664 = fpext bfloat %2945 to float, !dbg !31 + %3665 = fpext bfloat %2946 to float, !dbg !31 + %3666 = fpext bfloat %2956 to float, !dbg !31 + %3667 = fpext bfloat %2957 to float, !dbg !31 + %3668 = fpext bfloat %2958 to float, !dbg !31 + %3669 = fpext bfloat %2959 to float, !dbg !31 + %3670 = fpext bfloat %2960 to float, !dbg !31 + %3671 = fpext bfloat %2961 to float, !dbg !31 + %3672 = fpext bfloat %2962 to float, !dbg !31 + %3673 = fpext bfloat %2963 to float, !dbg !31 + %3674 = fpext bfloat %2973 to float, !dbg !31 + %3675 = fpext bfloat %2974 to float, !dbg !31 + %3676 = fpext bfloat %2975 to float, !dbg !31 + %3677 = fpext bfloat %2976 to float, !dbg !31 + %3678 = fpext bfloat %2977 to float, !dbg !31 + %3679 = fpext bfloat %2978 to float, !dbg !31 + %3680 = fpext bfloat %2979 to float, !dbg !31 + %3681 = fpext bfloat %2980 to float, !dbg !31 + %3682 = fpext bfloat %2990 to float, !dbg !31 + %3683 = fpext bfloat %2991 to float, !dbg !31 + %3684 = fpext bfloat %2992 to float, !dbg !31 + %3685 = fpext bfloat %2993 to float, !dbg !31 + %3686 = fpext bfloat %2994 to float, !dbg !31 + %3687 = fpext bfloat %2995 to float, !dbg !31 + %3688 = fpext bfloat %2996 to float, !dbg !31 + %3689 = fpext bfloat %2997 to float, !dbg !31 + %3690 = fpext bfloat %3007 to float, !dbg !31 + %3691 = fpext bfloat %3008 to float, !dbg !31 + %3692 = fpext bfloat %3009 to float, !dbg !31 + %3693 = fpext bfloat %3010 to float, !dbg !31 + %3694 = fpext bfloat %3011 to float, !dbg !31 + %3695 = fpext bfloat %3012 to float, !dbg !31 + %3696 = fpext bfloat %3013 to float, !dbg !31 + %3697 = fpext bfloat %3014 to float, !dbg !31 + %3698 = fpext bfloat %3024 to float, !dbg !31 + %3699 = fpext bfloat %3025 to float, !dbg !31 + %3700 = fpext bfloat %3026 to float, !dbg !31 + %3701 = fpext bfloat %3027 to float, !dbg !31 + %3702 = fpext bfloat %3028 to float, !dbg !31 + %3703 = fpext bfloat %3029 to float, !dbg !31 + %3704 = fpext bfloat %3030 to float, !dbg !31 + %3705 = fpext bfloat %3031 to float, !dbg !31 + %3706 = fpext bfloat %3041 to float, !dbg !31 + %3707 = fpext bfloat %3042 to float, !dbg !31 + %3708 = fpext bfloat %3043 to float, !dbg !31 + %3709 = fpext bfloat %3044 to float, !dbg !31 + %3710 = fpext bfloat %3045 to float, !dbg !31 + %3711 = fpext bfloat %3046 to float, !dbg !31 + %3712 = fpext bfloat %3047 to float, !dbg !31 + %3713 = fpext bfloat %3048 to float, !dbg !31 + %3714 = fpext bfloat %3058 to float, !dbg !31 + %3715 = fpext bfloat %3059 to float, !dbg !31 + %3716 = fpext bfloat %3060 to float, !dbg !31 + %3717 = fpext bfloat %3061 to float, !dbg !31 + %3718 = fpext bfloat %3062 to float, !dbg !31 + %3719 = fpext bfloat %3063 to float, !dbg !31 + %3720 = fpext bfloat %3064 to float, !dbg !31 + %3721 = fpext bfloat %3065 to float, !dbg !31 + %3722 = fpext bfloat %3075 to float, !dbg !31 + %3723 = fpext bfloat %3076 to float, !dbg !31 + %3724 = fpext bfloat %3077 to float, !dbg !31 + %3725 = fpext bfloat %3078 to float, !dbg !31 + %3726 = fpext bfloat %3079 to float, !dbg !31 + %3727 = fpext bfloat %3080 to float, !dbg !31 + %3728 = fpext bfloat %3081 to float, !dbg !31 + %3729 = fpext bfloat %3082 to float, !dbg !31 + %3730 = fpext bfloat %3092 to float, !dbg !31 + %3731 = fpext bfloat %3093 to float, !dbg !31 + %3732 = fpext bfloat %3094 to float, !dbg !31 + %3733 = fpext bfloat %3095 to float, !dbg !31 + %3734 = fpext bfloat %3096 to float, !dbg !31 + %3735 = fpext bfloat %3097 to float, !dbg !31 + %3736 = fpext bfloat %3098 to float, !dbg !31 + %3737 = fpext bfloat %3099 to float, !dbg !31 + %3738 = fpext bfloat %3109 to float, !dbg !31 + %3739 = fpext bfloat %3110 to float, !dbg !31 + %3740 = fpext bfloat %3111 to float, !dbg !31 + %3741 = fpext bfloat %3112 to float, !dbg !31 + %3742 = fpext bfloat %3113 to float, !dbg !31 + %3743 = fpext bfloat %3114 to float, !dbg !31 + %3744 = fpext bfloat %3115 to float, !dbg !31 + %3745 = fpext bfloat %3116 to float, !dbg !31 + %3746 = fpext bfloat %3126 to float, !dbg !31 + %3747 = fpext bfloat %3127 to float, !dbg !31 + %3748 = fpext bfloat %3128 to float, !dbg !31 + %3749 = fpext bfloat %3129 to float, !dbg !31 + %3750 = fpext bfloat %3130 to float, !dbg !31 + %3751 = fpext bfloat %3131 to float, !dbg !31 + %3752 = fpext bfloat %3132 to float, !dbg !31 + %3753 = fpext bfloat %3133 to float, !dbg !31 + %3754 = fpext bfloat %3143 to float, !dbg !31 + %3755 = fpext bfloat %3144 to float, !dbg !31 + %3756 = fpext bfloat %3145 to float, !dbg !31 + %3757 = fpext bfloat %3146 to float, !dbg !31 + %3758 = fpext bfloat %3147 to float, !dbg !31 + %3759 = fpext bfloat %3148 to float, !dbg !31 + %3760 = fpext bfloat %3149 to float, !dbg !31 + %3761 = fpext bfloat %3150 to float, !dbg !31 + %3762 = fpext bfloat %3160 to float, !dbg !31 + %3763 = fpext bfloat %3161 to float, !dbg !31 + %3764 = fpext bfloat %3162 to float, !dbg !31 + %3765 = fpext bfloat %3163 to float, !dbg !31 + %3766 = fpext bfloat %3164 to float, !dbg !31 + %3767 = fpext bfloat %3165 to float, !dbg !31 + %3768 = fpext bfloat %3166 to float, !dbg !31 + %3769 = fpext bfloat %3167 to float, !dbg !31 + %3770 = fpext bfloat %3177 to float, !dbg !31 + %3771 = fpext bfloat %3178 to float, !dbg !31 + %3772 = fpext bfloat %3179 to float, !dbg !31 + %3773 = fpext bfloat %3180 to float, !dbg !31 + %3774 = fpext bfloat %3181 to float, !dbg !31 + %3775 = fpext bfloat %3182 to float, !dbg !31 + %3776 = fpext bfloat %3183 to float, !dbg !31 + %3777 = fpext bfloat %3184 to float, !dbg !31 + %3778 = fpext bfloat %3194 to float, !dbg !31 + %3779 = fpext bfloat %3195 to float, !dbg !31 + %3780 = fpext bfloat %3196 to float, !dbg !31 + %3781 = fpext bfloat %3197 to float, !dbg !31 + %3782 = fpext bfloat %3198 to float, !dbg !31 + %3783 = fpext bfloat %3199 to float, !dbg !31 + %3784 = fpext bfloat %3200 to float, !dbg !31 + %3785 = fpext bfloat %3201 to float, !dbg !31 + %3786 = fpext bfloat %3211 to float, !dbg !31 + %3787 = fpext bfloat %3212 to float, !dbg !31 + %3788 = fpext bfloat %3213 to float, !dbg !31 + %3789 = fpext bfloat %3214 to float, !dbg !31 + %3790 = fpext bfloat %3215 to float, !dbg !31 + %3791 = fpext bfloat %3216 to float, !dbg !31 + %3792 = fpext bfloat %3217 to float, !dbg !31 + %3793 = fpext bfloat %3218 to float, !dbg !31 + %3794 = fpext bfloat %3228 to float, !dbg !31 + %3795 = fpext bfloat %3229 to float, !dbg !31 + %3796 = fpext bfloat %3230 to float, !dbg !31 + %3797 = fpext bfloat %3231 to float, !dbg !31 + %3798 = fpext bfloat %3232 to float, !dbg !31 + %3799 = fpext bfloat %3233 to float, !dbg !31 + %3800 = fpext bfloat %3234 to float, !dbg !31 + %3801 = fpext bfloat %3235 to float, !dbg !31 + %3802 = fpext bfloat %3245 to float, !dbg !31 + %3803 = fpext bfloat %3246 to float, !dbg !31 + %3804 = fpext bfloat %3247 to float, !dbg !31 + %3805 = fpext bfloat %3248 to float, !dbg !31 + %3806 = fpext bfloat %3249 to float, !dbg !31 + %3807 = fpext bfloat %3250 to float, !dbg !31 + %3808 = fpext bfloat %3251 to float, !dbg !31 + %3809 = fpext bfloat %3252 to float, !dbg !31 + %3810 = fpext bfloat %3262 to float, !dbg !31 + %3811 = fpext bfloat %3263 to float, !dbg !31 + %3812 = fpext bfloat %3264 to float, !dbg !31 + %3813 = fpext bfloat %3265 to float, !dbg !31 + %3814 = fpext bfloat %3266 to float, !dbg !31 + %3815 = fpext bfloat %3267 to float, !dbg !31 + %3816 = fpext bfloat %3268 to float, !dbg !31 + %3817 = fpext bfloat %3269 to float, !dbg !31 + %3818 = fpext bfloat %3279 to float, !dbg !31 + %3819 = fpext bfloat %3280 to float, !dbg !31 + %3820 = fpext bfloat %3281 to float, !dbg !31 + %3821 = fpext bfloat %3282 to float, !dbg !31 + %3822 = fpext bfloat %3283 to float, !dbg !31 + %3823 = fpext bfloat %3284 to float, !dbg !31 + %3824 = fpext bfloat %3285 to float, !dbg !31 + %3825 = fpext bfloat %3286 to float, !dbg !31 + %3826 = fpext bfloat %3296 to float, !dbg !31 + %3827 = fpext bfloat %3297 to float, !dbg !31 + %3828 = fpext bfloat %3298 to float, !dbg !31 + %3829 = fpext bfloat %3299 to float, !dbg !31 + %3830 = fpext bfloat %3300 to float, !dbg !31 + %3831 = fpext bfloat %3301 to float, !dbg !31 + %3832 = fpext bfloat %3302 to float, !dbg !31 + %3833 = fpext bfloat %3303 to float, !dbg !31 + %3834 = fpext bfloat %3313 to float, !dbg !31 + %3835 = fpext bfloat %3314 to float, !dbg !31 + %3836 = fpext bfloat %3315 to float, !dbg !31 + %3837 = fpext bfloat %3316 to float, !dbg !31 + %3838 = fpext bfloat %3317 to float, !dbg !31 + %3839 = fpext bfloat %3318 to float, !dbg !31 + %3840 = fpext bfloat %3319 to float, !dbg !31 + %3841 = fpext bfloat %3320 to float, !dbg !31 + %3842 = fpext bfloat %3330 to float, !dbg !31 + %3843 = fpext bfloat %3331 to float, !dbg !31 + %3844 = fpext bfloat %3332 to float, !dbg !31 + %3845 = fpext bfloat %3333 to float, !dbg !31 + %3846 = fpext bfloat %3334 to float, !dbg !31 + %3847 = fpext bfloat %3335 to float, !dbg !31 + %3848 = fpext bfloat %3336 to float, !dbg !31 + %3849 = fpext bfloat %3337 to float, !dbg !31 + %3850 = shl nuw nsw i32 %63, 4, !dbg !32 + %3851 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %3850, !dbg !32 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x2.b16.p3(ptr addrspace(3) %3851, i32 %2243, i32 %2245), !dbg !32 + tail call void @llvm.nvvm.bar.warp.sync(i32 -1), !dbg !32 + %3852 = shl nuw nsw i32 %13, 3, !dbg !32 + %3853 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %3852, !dbg !32 + %3854 = load float, ptr addrspace(3) %3853, align 16, !dbg !32 + %3855 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 4, !dbg !32 + %3856 = load float, ptr addrspace(3) %3855, align 4, !dbg !32 + %3857 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 8, !dbg !32 + %3858 = load float, ptr addrspace(3) %3857, align 8, !dbg !32 + %3859 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 12, !dbg !32 + %3860 = load float, ptr addrspace(3) %3859, align 4, !dbg !32 + %3861 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 16, !dbg !32 + %3862 = load float, ptr addrspace(3) %3861, align 16, !dbg !32 + %3863 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 20, !dbg !32 + %3864 = load float, ptr addrspace(3) %3863, align 4, !dbg !32 + %3865 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 24, !dbg !32 + %3866 = load float, ptr addrspace(3) %3865, align 8, !dbg !32 + %3867 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 28, !dbg !32 + %3868 = load float, ptr addrspace(3) %3867, align 4, !dbg !32 + %3869 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 32, !dbg !32 + %3870 = load float, ptr addrspace(3) %3869, align 16, !dbg !32 + %3871 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 36, !dbg !32 + %3872 = load float, ptr addrspace(3) %3871, align 4, !dbg !32 + %3873 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 40, !dbg !32 + %3874 = load float, ptr addrspace(3) %3873, align 8, !dbg !32 + %3875 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 44, !dbg !32 + %3876 = load float, ptr addrspace(3) %3875, align 4, !dbg !32 + %3877 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 48, !dbg !32 + %3878 = load float, ptr addrspace(3) %3877, align 16, !dbg !32 + %3879 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 52, !dbg !32 + %3880 = load float, ptr addrspace(3) %3879, align 4, !dbg !32 + %3881 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 56, !dbg !32 + %3882 = load float, ptr addrspace(3) %3881, align 8, !dbg !32 + %3883 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 60, !dbg !32 + %3884 = load float, ptr addrspace(3) %3883, align 4, !dbg !32 + %3885 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 64, !dbg !32 + %3886 = load float, ptr addrspace(3) %3885, align 16, !dbg !32 + %3887 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 68, !dbg !32 + %3888 = load float, ptr addrspace(3) %3887, align 4, !dbg !32 + %3889 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 72, !dbg !32 + %3890 = load float, ptr addrspace(3) %3889, align 8, !dbg !32 + %3891 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 76, !dbg !32 + %3892 = load float, ptr addrspace(3) %3891, align 4, !dbg !32 + %3893 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 80, !dbg !32 + %3894 = load float, ptr addrspace(3) %3893, align 16, !dbg !32 + %3895 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 84, !dbg !32 + %3896 = load float, ptr addrspace(3) %3895, align 4, !dbg !32 + %3897 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 88, !dbg !32 + %3898 = load float, ptr addrspace(3) %3897, align 8, !dbg !32 + %3899 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 92, !dbg !32 + %3900 = load float, ptr addrspace(3) %3899, align 4, !dbg !32 + %3901 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 96, !dbg !32 + %3902 = load float, ptr addrspace(3) %3901, align 16, !dbg !32 + %3903 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 100, !dbg !32 + %3904 = load float, ptr addrspace(3) %3903, align 4, !dbg !32 + %3905 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 104, !dbg !32 + %3906 = load float, ptr addrspace(3) %3905, align 8, !dbg !32 + %3907 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 108, !dbg !32 + %3908 = load float, ptr addrspace(3) %3907, align 4, !dbg !32 + %3909 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 112, !dbg !32 + %3910 = load float, ptr addrspace(3) %3909, align 16, !dbg !32 + %3911 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 116, !dbg !32 + %3912 = load float, ptr addrspace(3) %3911, align 4, !dbg !32 + %3913 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 120, !dbg !32 + %3914 = load float, ptr addrspace(3) %3913, align 8, !dbg !32 + %3915 = getelementptr inbounds nuw i8, ptr addrspace(3) %3853, i32 124, !dbg !32 + %3916 = load float, ptr addrspace(3) %3915, align 4, !dbg !32 + tail call void @llvm.nvvm.bar.warp.sync(i32 -1), !dbg !32 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x2.b16.p3(ptr addrspace(3) %3851, i32 %2247, i32 %2249), !dbg !32 + tail call void @llvm.nvvm.bar.warp.sync(i32 -1), !dbg !32 + %3917 = load float, ptr addrspace(3) %3853, align 16, !dbg !32 + %3918 = load float, ptr addrspace(3) %3855, align 4, !dbg !32 + %3919 = load float, ptr addrspace(3) %3857, align 8, !dbg !32 + %3920 = load float, ptr addrspace(3) %3859, align 4, !dbg !32 + %3921 = load float, ptr addrspace(3) %3861, align 16, !dbg !32 + %3922 = load float, ptr addrspace(3) %3863, align 4, !dbg !32 + %3923 = load float, ptr addrspace(3) %3865, align 8, !dbg !32 + %3924 = load float, ptr addrspace(3) %3867, align 4, !dbg !32 + %3925 = load float, ptr addrspace(3) %3869, align 16, !dbg !32 + %3926 = load float, ptr addrspace(3) %3871, align 4, !dbg !32 + %3927 = load float, ptr addrspace(3) %3873, align 8, !dbg !32 + %3928 = load float, ptr addrspace(3) %3875, align 4, !dbg !32 + %3929 = load float, ptr addrspace(3) %3877, align 16, !dbg !32 + %3930 = load float, ptr addrspace(3) %3879, align 4, !dbg !32 + %3931 = load float, ptr addrspace(3) %3881, align 8, !dbg !32 + %3932 = load float, ptr addrspace(3) %3883, align 4, !dbg !32 + %3933 = load float, ptr addrspace(3) %3885, align 16, !dbg !32 + %3934 = load float, ptr addrspace(3) %3887, align 4, !dbg !32 + %3935 = load float, ptr addrspace(3) %3889, align 8, !dbg !32 + %3936 = load float, ptr addrspace(3) %3891, align 4, !dbg !32 + %3937 = load float, ptr addrspace(3) %3893, align 16, !dbg !32 + %3938 = load float, ptr addrspace(3) %3895, align 4, !dbg !32 + %3939 = load float, ptr addrspace(3) %3897, align 8, !dbg !32 + %3940 = load float, ptr addrspace(3) %3899, align 4, !dbg !32 + %3941 = load float, ptr addrspace(3) %3901, align 16, !dbg !32 + %3942 = load float, ptr addrspace(3) %3903, align 4, !dbg !32 + %3943 = load float, ptr addrspace(3) %3905, align 8, !dbg !32 + %3944 = load float, ptr addrspace(3) %3907, align 4, !dbg !32 + %3945 = load float, ptr addrspace(3) %3909, align 16, !dbg !32 + %3946 = load float, ptr addrspace(3) %3911, align 4, !dbg !32 + %3947 = load float, ptr addrspace(3) %3913, align 8, !dbg !32 + %3948 = load float, ptr addrspace(3) %3915, align 4, !dbg !32 + %3949 = fmul float %3854, %3338, !dbg !33 + %3950 = fmul float %3854, %3339, !dbg !33 + %3951 = fmul float %3854, %3340, !dbg !33 + %3952 = fmul float %3854, %3341, !dbg !33 + %3953 = fmul float %3854, %3342, !dbg !33 + %3954 = fmul float %3854, %3343, !dbg !33 + %3955 = fmul float %3854, %3344, !dbg !33 + %3956 = fmul float %3854, %3345, !dbg !33 + %3957 = fmul float %3917, %3346, !dbg !33 + %3958 = fmul float %3917, %3347, !dbg !33 + %3959 = fmul float %3917, %3348, !dbg !33 + %3960 = fmul float %3917, %3349, !dbg !33 + %3961 = fmul float %3917, %3350, !dbg !33 + %3962 = fmul float %3917, %3351, !dbg !33 + %3963 = fmul float %3917, %3352, !dbg !33 + %3964 = fmul float %3917, %3353, !dbg !33 + %3965 = fmul float %3856, %3354, !dbg !33 + %3966 = fmul float %3856, %3355, !dbg !33 + %3967 = fmul float %3856, %3356, !dbg !33 + %3968 = fmul float %3856, %3357, !dbg !33 + %3969 = fmul float %3856, %3358, !dbg !33 + %3970 = fmul float %3856, %3359, !dbg !33 + %3971 = fmul float %3856, %3360, !dbg !33 + %3972 = fmul float %3856, %3361, !dbg !33 + %3973 = fmul float %3918, %3362, !dbg !33 + %3974 = fmul float %3918, %3363, !dbg !33 + %3975 = fmul float %3918, %3364, !dbg !33 + %3976 = fmul float %3918, %3365, !dbg !33 + %3977 = fmul float %3918, %3366, !dbg !33 + %3978 = fmul float %3918, %3367, !dbg !33 + %3979 = fmul float %3918, %3368, !dbg !33 + %3980 = fmul float %3918, %3369, !dbg !33 + %3981 = fmul float %3858, %3370, !dbg !33 + %3982 = fmul float %3858, %3371, !dbg !33 + %3983 = fmul float %3858, %3372, !dbg !33 + %3984 = fmul float %3858, %3373, !dbg !33 + %3985 = fmul float %3858, %3374, !dbg !33 + %3986 = fmul float %3858, %3375, !dbg !33 + %3987 = fmul float %3858, %3376, !dbg !33 + %3988 = fmul float %3858, %3377, !dbg !33 + %3989 = fmul float %3919, %3378, !dbg !33 + %3990 = fmul float %3919, %3379, !dbg !33 + %3991 = fmul float %3919, %3380, !dbg !33 + %3992 = fmul float %3919, %3381, !dbg !33 + %3993 = fmul float %3919, %3382, !dbg !33 + %3994 = fmul float %3919, %3383, !dbg !33 + %3995 = fmul float %3919, %3384, !dbg !33 + %3996 = fmul float %3919, %3385, !dbg !33 + %3997 = fmul float %3860, %3386, !dbg !33 + %3998 = fmul float %3860, %3387, !dbg !33 + %3999 = fmul float %3860, %3388, !dbg !33 + %4000 = fmul float %3860, %3389, !dbg !33 + %4001 = fmul float %3860, %3390, !dbg !33 + %4002 = fmul float %3860, %3391, !dbg !33 + %4003 = fmul float %3860, %3392, !dbg !33 + %4004 = fmul float %3860, %3393, !dbg !33 + %4005 = fmul float %3920, %3394, !dbg !33 + %4006 = fmul float %3920, %3395, !dbg !33 + %4007 = fmul float %3920, %3396, !dbg !33 + %4008 = fmul float %3920, %3397, !dbg !33 + %4009 = fmul float %3920, %3398, !dbg !33 + %4010 = fmul float %3920, %3399, !dbg !33 + %4011 = fmul float %3920, %3400, !dbg !33 + %4012 = fmul float %3920, %3401, !dbg !33 + %4013 = fmul float %3862, %3402, !dbg !33 + %4014 = fmul float %3862, %3403, !dbg !33 + %4015 = fmul float %3862, %3404, !dbg !33 + %4016 = fmul float %3862, %3405, !dbg !33 + %4017 = fmul float %3862, %3406, !dbg !33 + %4018 = fmul float %3862, %3407, !dbg !33 + %4019 = fmul float %3862, %3408, !dbg !33 + %4020 = fmul float %3862, %3409, !dbg !33 + %4021 = fmul float %3921, %3410, !dbg !33 + %4022 = fmul float %3921, %3411, !dbg !33 + %4023 = fmul float %3921, %3412, !dbg !33 + %4024 = fmul float %3921, %3413, !dbg !33 + %4025 = fmul float %3921, %3414, !dbg !33 + %4026 = fmul float %3921, %3415, !dbg !33 + %4027 = fmul float %3921, %3416, !dbg !33 + %4028 = fmul float %3921, %3417, !dbg !33 + %4029 = fmul float %3864, %3418, !dbg !33 + %4030 = fmul float %3864, %3419, !dbg !33 + %4031 = fmul float %3864, %3420, !dbg !33 + %4032 = fmul float %3864, %3421, !dbg !33 + %4033 = fmul float %3864, %3422, !dbg !33 + %4034 = fmul float %3864, %3423, !dbg !33 + %4035 = fmul float %3864, %3424, !dbg !33 + %4036 = fmul float %3864, %3425, !dbg !33 + %4037 = fmul float %3922, %3426, !dbg !33 + %4038 = fmul float %3922, %3427, !dbg !33 + %4039 = fmul float %3922, %3428, !dbg !33 + %4040 = fmul float %3922, %3429, !dbg !33 + %4041 = fmul float %3922, %3430, !dbg !33 + %4042 = fmul float %3922, %3431, !dbg !33 + %4043 = fmul float %3922, %3432, !dbg !33 + %4044 = fmul float %3922, %3433, !dbg !33 + %4045 = fmul float %3866, %3434, !dbg !33 + %4046 = fmul float %3866, %3435, !dbg !33 + %4047 = fmul float %3866, %3436, !dbg !33 + %4048 = fmul float %3866, %3437, !dbg !33 + %4049 = fmul float %3866, %3438, !dbg !33 + %4050 = fmul float %3866, %3439, !dbg !33 + %4051 = fmul float %3866, %3440, !dbg !33 + %4052 = fmul float %3866, %3441, !dbg !33 + %4053 = fmul float %3923, %3442, !dbg !33 + %4054 = fmul float %3923, %3443, !dbg !33 + %4055 = fmul float %3923, %3444, !dbg !33 + %4056 = fmul float %3923, %3445, !dbg !33 + %4057 = fmul float %3923, %3446, !dbg !33 + %4058 = fmul float %3923, %3447, !dbg !33 + %4059 = fmul float %3923, %3448, !dbg !33 + %4060 = fmul float %3923, %3449, !dbg !33 + %4061 = fmul float %3868, %3450, !dbg !33 + %4062 = fmul float %3868, %3451, !dbg !33 + %4063 = fmul float %3868, %3452, !dbg !33 + %4064 = fmul float %3868, %3453, !dbg !33 + %4065 = fmul float %3868, %3454, !dbg !33 + %4066 = fmul float %3868, %3455, !dbg !33 + %4067 = fmul float %3868, %3456, !dbg !33 + %4068 = fmul float %3868, %3457, !dbg !33 + %4069 = fmul float %3924, %3458, !dbg !33 + %4070 = fmul float %3924, %3459, !dbg !33 + %4071 = fmul float %3924, %3460, !dbg !33 + %4072 = fmul float %3924, %3461, !dbg !33 + %4073 = fmul float %3924, %3462, !dbg !33 + %4074 = fmul float %3924, %3463, !dbg !33 + %4075 = fmul float %3924, %3464, !dbg !33 + %4076 = fmul float %3924, %3465, !dbg !33 + %4077 = fmul float %3870, %3466, !dbg !33 + %4078 = fmul float %3870, %3467, !dbg !33 + %4079 = fmul float %3870, %3468, !dbg !33 + %4080 = fmul float %3870, %3469, !dbg !33 + %4081 = fmul float %3870, %3470, !dbg !33 + %4082 = fmul float %3870, %3471, !dbg !33 + %4083 = fmul float %3870, %3472, !dbg !33 + %4084 = fmul float %3870, %3473, !dbg !33 + %4085 = fmul float %3925, %3474, !dbg !33 + %4086 = fmul float %3925, %3475, !dbg !33 + %4087 = fmul float %3925, %3476, !dbg !33 + %4088 = fmul float %3925, %3477, !dbg !33 + %4089 = fmul float %3925, %3478, !dbg !33 + %4090 = fmul float %3925, %3479, !dbg !33 + %4091 = fmul float %3925, %3480, !dbg !33 + %4092 = fmul float %3925, %3481, !dbg !33 + %4093 = fmul float %3872, %3482, !dbg !33 + %4094 = fmul float %3872, %3483, !dbg !33 + %4095 = fmul float %3872, %3484, !dbg !33 + %4096 = fmul float %3872, %3485, !dbg !33 + %4097 = fmul float %3872, %3486, !dbg !33 + %4098 = fmul float %3872, %3487, !dbg !33 + %4099 = fmul float %3872, %3488, !dbg !33 + %4100 = fmul float %3872, %3489, !dbg !33 + %4101 = fmul float %3926, %3490, !dbg !33 + %4102 = fmul float %3926, %3491, !dbg !33 + %4103 = fmul float %3926, %3492, !dbg !33 + %4104 = fmul float %3926, %3493, !dbg !33 + %4105 = fmul float %3926, %3494, !dbg !33 + %4106 = fmul float %3926, %3495, !dbg !33 + %4107 = fmul float %3926, %3496, !dbg !33 + %4108 = fmul float %3926, %3497, !dbg !33 + %4109 = fmul float %3874, %3498, !dbg !33 + %4110 = fmul float %3874, %3499, !dbg !33 + %4111 = fmul float %3874, %3500, !dbg !33 + %4112 = fmul float %3874, %3501, !dbg !33 + %4113 = fmul float %3874, %3502, !dbg !33 + %4114 = fmul float %3874, %3503, !dbg !33 + %4115 = fmul float %3874, %3504, !dbg !33 + %4116 = fmul float %3874, %3505, !dbg !33 + %4117 = fmul float %3927, %3506, !dbg !33 + %4118 = fmul float %3927, %3507, !dbg !33 + %4119 = fmul float %3927, %3508, !dbg !33 + %4120 = fmul float %3927, %3509, !dbg !33 + %4121 = fmul float %3927, %3510, !dbg !33 + %4122 = fmul float %3927, %3511, !dbg !33 + %4123 = fmul float %3927, %3512, !dbg !33 + %4124 = fmul float %3927, %3513, !dbg !33 + %4125 = fmul float %3876, %3514, !dbg !33 + %4126 = fmul float %3876, %3515, !dbg !33 + %4127 = fmul float %3876, %3516, !dbg !33 + %4128 = fmul float %3876, %3517, !dbg !33 + %4129 = fmul float %3876, %3518, !dbg !33 + %4130 = fmul float %3876, %3519, !dbg !33 + %4131 = fmul float %3876, %3520, !dbg !33 + %4132 = fmul float %3876, %3521, !dbg !33 + %4133 = fmul float %3928, %3522, !dbg !33 + %4134 = fmul float %3928, %3523, !dbg !33 + %4135 = fmul float %3928, %3524, !dbg !33 + %4136 = fmul float %3928, %3525, !dbg !33 + %4137 = fmul float %3928, %3526, !dbg !33 + %4138 = fmul float %3928, %3527, !dbg !33 + %4139 = fmul float %3928, %3528, !dbg !33 + %4140 = fmul float %3928, %3529, !dbg !33 + %4141 = fmul float %3878, %3530, !dbg !33 + %4142 = fmul float %3878, %3531, !dbg !33 + %4143 = fmul float %3878, %3532, !dbg !33 + %4144 = fmul float %3878, %3533, !dbg !33 + %4145 = fmul float %3878, %3534, !dbg !33 + %4146 = fmul float %3878, %3535, !dbg !33 + %4147 = fmul float %3878, %3536, !dbg !33 + %4148 = fmul float %3878, %3537, !dbg !33 + %4149 = fmul float %3929, %3538, !dbg !33 + %4150 = fmul float %3929, %3539, !dbg !33 + %4151 = fmul float %3929, %3540, !dbg !33 + %4152 = fmul float %3929, %3541, !dbg !33 + %4153 = fmul float %3929, %3542, !dbg !33 + %4154 = fmul float %3929, %3543, !dbg !33 + %4155 = fmul float %3929, %3544, !dbg !33 + %4156 = fmul float %3929, %3545, !dbg !33 + %4157 = fmul float %3880, %3546, !dbg !33 + %4158 = fmul float %3880, %3547, !dbg !33 + %4159 = fmul float %3880, %3548, !dbg !33 + %4160 = fmul float %3880, %3549, !dbg !33 + %4161 = fmul float %3880, %3550, !dbg !33 + %4162 = fmul float %3880, %3551, !dbg !33 + %4163 = fmul float %3880, %3552, !dbg !33 + %4164 = fmul float %3880, %3553, !dbg !33 + %4165 = fmul float %3930, %3554, !dbg !33 + %4166 = fmul float %3930, %3555, !dbg !33 + %4167 = fmul float %3930, %3556, !dbg !33 + %4168 = fmul float %3930, %3557, !dbg !33 + %4169 = fmul float %3930, %3558, !dbg !33 + %4170 = fmul float %3930, %3559, !dbg !33 + %4171 = fmul float %3930, %3560, !dbg !33 + %4172 = fmul float %3930, %3561, !dbg !33 + %4173 = fmul float %3882, %3562, !dbg !33 + %4174 = fmul float %3882, %3563, !dbg !33 + %4175 = fmul float %3882, %3564, !dbg !33 + %4176 = fmul float %3882, %3565, !dbg !33 + %4177 = fmul float %3882, %3566, !dbg !33 + %4178 = fmul float %3882, %3567, !dbg !33 + %4179 = fmul float %3882, %3568, !dbg !33 + %4180 = fmul float %3882, %3569, !dbg !33 + %4181 = fmul float %3931, %3570, !dbg !33 + %4182 = fmul float %3931, %3571, !dbg !33 + %4183 = fmul float %3931, %3572, !dbg !33 + %4184 = fmul float %3931, %3573, !dbg !33 + %4185 = fmul float %3931, %3574, !dbg !33 + %4186 = fmul float %3931, %3575, !dbg !33 + %4187 = fmul float %3931, %3576, !dbg !33 + %4188 = fmul float %3931, %3577, !dbg !33 + %4189 = fmul float %3884, %3578, !dbg !33 + %4190 = fmul float %3884, %3579, !dbg !33 + %4191 = fmul float %3884, %3580, !dbg !33 + %4192 = fmul float %3884, %3581, !dbg !33 + %4193 = fmul float %3884, %3582, !dbg !33 + %4194 = fmul float %3884, %3583, !dbg !33 + %4195 = fmul float %3884, %3584, !dbg !33 + %4196 = fmul float %3884, %3585, !dbg !33 + %4197 = fmul float %3932, %3586, !dbg !33 + %4198 = fmul float %3932, %3587, !dbg !33 + %4199 = fmul float %3932, %3588, !dbg !33 + %4200 = fmul float %3932, %3589, !dbg !33 + %4201 = fmul float %3932, %3590, !dbg !33 + %4202 = fmul float %3932, %3591, !dbg !33 + %4203 = fmul float %3932, %3592, !dbg !33 + %4204 = fmul float %3932, %3593, !dbg !33 + %4205 = fmul float %3886, %3594, !dbg !33 + %4206 = fmul float %3886, %3595, !dbg !33 + %4207 = fmul float %3886, %3596, !dbg !33 + %4208 = fmul float %3886, %3597, !dbg !33 + %4209 = fmul float %3886, %3598, !dbg !33 + %4210 = fmul float %3886, %3599, !dbg !33 + %4211 = fmul float %3886, %3600, !dbg !33 + %4212 = fmul float %3886, %3601, !dbg !33 + %4213 = fmul float %3933, %3602, !dbg !33 + %4214 = fmul float %3933, %3603, !dbg !33 + %4215 = fmul float %3933, %3604, !dbg !33 + %4216 = fmul float %3933, %3605, !dbg !33 + %4217 = fmul float %3933, %3606, !dbg !33 + %4218 = fmul float %3933, %3607, !dbg !33 + %4219 = fmul float %3933, %3608, !dbg !33 + %4220 = fmul float %3933, %3609, !dbg !33 + %4221 = fmul float %3888, %3610, !dbg !33 + %4222 = fmul float %3888, %3611, !dbg !33 + %4223 = fmul float %3888, %3612, !dbg !33 + %4224 = fmul float %3888, %3613, !dbg !33 + %4225 = fmul float %3888, %3614, !dbg !33 + %4226 = fmul float %3888, %3615, !dbg !33 + %4227 = fmul float %3888, %3616, !dbg !33 + %4228 = fmul float %3888, %3617, !dbg !33 + %4229 = fmul float %3934, %3618, !dbg !33 + %4230 = fmul float %3934, %3619, !dbg !33 + %4231 = fmul float %3934, %3620, !dbg !33 + %4232 = fmul float %3934, %3621, !dbg !33 + %4233 = fmul float %3934, %3622, !dbg !33 + %4234 = fmul float %3934, %3623, !dbg !33 + %4235 = fmul float %3934, %3624, !dbg !33 + %4236 = fmul float %3934, %3625, !dbg !33 + %4237 = fmul float %3890, %3626, !dbg !33 + %4238 = fmul float %3890, %3627, !dbg !33 + %4239 = fmul float %3890, %3628, !dbg !33 + %4240 = fmul float %3890, %3629, !dbg !33 + %4241 = fmul float %3890, %3630, !dbg !33 + %4242 = fmul float %3890, %3631, !dbg !33 + %4243 = fmul float %3890, %3632, !dbg !33 + %4244 = fmul float %3890, %3633, !dbg !33 + %4245 = fmul float %3935, %3634, !dbg !33 + %4246 = fmul float %3935, %3635, !dbg !33 + %4247 = fmul float %3935, %3636, !dbg !33 + %4248 = fmul float %3935, %3637, !dbg !33 + %4249 = fmul float %3935, %3638, !dbg !33 + %4250 = fmul float %3935, %3639, !dbg !33 + %4251 = fmul float %3935, %3640, !dbg !33 + %4252 = fmul float %3935, %3641, !dbg !33 + %4253 = fmul float %3892, %3642, !dbg !33 + %4254 = fmul float %3892, %3643, !dbg !33 + %4255 = fmul float %3892, %3644, !dbg !33 + %4256 = fmul float %3892, %3645, !dbg !33 + %4257 = fmul float %3892, %3646, !dbg !33 + %4258 = fmul float %3892, %3647, !dbg !33 + %4259 = fmul float %3892, %3648, !dbg !33 + %4260 = fmul float %3892, %3649, !dbg !33 + %4261 = fmul float %3936, %3650, !dbg !33 + %4262 = fmul float %3936, %3651, !dbg !33 + %4263 = fmul float %3936, %3652, !dbg !33 + %4264 = fmul float %3936, %3653, !dbg !33 + %4265 = fmul float %3936, %3654, !dbg !33 + %4266 = fmul float %3936, %3655, !dbg !33 + %4267 = fmul float %3936, %3656, !dbg !33 + %4268 = fmul float %3936, %3657, !dbg !33 + %4269 = fmul float %3894, %3658, !dbg !33 + %4270 = fmul float %3894, %3659, !dbg !33 + %4271 = fmul float %3894, %3660, !dbg !33 + %4272 = fmul float %3894, %3661, !dbg !33 + %4273 = fmul float %3894, %3662, !dbg !33 + %4274 = fmul float %3894, %3663, !dbg !33 + %4275 = fmul float %3894, %3664, !dbg !33 + %4276 = fmul float %3894, %3665, !dbg !33 + %4277 = fmul float %3937, %3666, !dbg !33 + %4278 = fmul float %3937, %3667, !dbg !33 + %4279 = fmul float %3937, %3668, !dbg !33 + %4280 = fmul float %3937, %3669, !dbg !33 + %4281 = fmul float %3937, %3670, !dbg !33 + %4282 = fmul float %3937, %3671, !dbg !33 + %4283 = fmul float %3937, %3672, !dbg !33 + %4284 = fmul float %3937, %3673, !dbg !33 + %4285 = fmul float %3896, %3674, !dbg !33 + %4286 = fmul float %3896, %3675, !dbg !33 + %4287 = fmul float %3896, %3676, !dbg !33 + %4288 = fmul float %3896, %3677, !dbg !33 + %4289 = fmul float %3896, %3678, !dbg !33 + %4290 = fmul float %3896, %3679, !dbg !33 + %4291 = fmul float %3896, %3680, !dbg !33 + %4292 = fmul float %3896, %3681, !dbg !33 + %4293 = fmul float %3938, %3682, !dbg !33 + %4294 = fmul float %3938, %3683, !dbg !33 + %4295 = fmul float %3938, %3684, !dbg !33 + %4296 = fmul float %3938, %3685, !dbg !33 + %4297 = fmul float %3938, %3686, !dbg !33 + %4298 = fmul float %3938, %3687, !dbg !33 + %4299 = fmul float %3938, %3688, !dbg !33 + %4300 = fmul float %3938, %3689, !dbg !33 + %4301 = fmul float %3898, %3690, !dbg !33 + %4302 = fmul float %3898, %3691, !dbg !33 + %4303 = fmul float %3898, %3692, !dbg !33 + %4304 = fmul float %3898, %3693, !dbg !33 + %4305 = fmul float %3898, %3694, !dbg !33 + %4306 = fmul float %3898, %3695, !dbg !33 + %4307 = fmul float %3898, %3696, !dbg !33 + %4308 = fmul float %3898, %3697, !dbg !33 + %4309 = fmul float %3939, %3698, !dbg !33 + %4310 = fmul float %3939, %3699, !dbg !33 + %4311 = fmul float %3939, %3700, !dbg !33 + %4312 = fmul float %3939, %3701, !dbg !33 + %4313 = fmul float %3939, %3702, !dbg !33 + %4314 = fmul float %3939, %3703, !dbg !33 + %4315 = fmul float %3939, %3704, !dbg !33 + %4316 = fmul float %3939, %3705, !dbg !33 + %4317 = fmul float %3900, %3706, !dbg !33 + %4318 = fmul float %3900, %3707, !dbg !33 + %4319 = fmul float %3900, %3708, !dbg !33 + %4320 = fmul float %3900, %3709, !dbg !33 + %4321 = fmul float %3900, %3710, !dbg !33 + %4322 = fmul float %3900, %3711, !dbg !33 + %4323 = fmul float %3900, %3712, !dbg !33 + %4324 = fmul float %3900, %3713, !dbg !33 + %4325 = fmul float %3940, %3714, !dbg !33 + %4326 = fmul float %3940, %3715, !dbg !33 + %4327 = fmul float %3940, %3716, !dbg !33 + %4328 = fmul float %3940, %3717, !dbg !33 + %4329 = fmul float %3940, %3718, !dbg !33 + %4330 = fmul float %3940, %3719, !dbg !33 + %4331 = fmul float %3940, %3720, !dbg !33 + %4332 = fmul float %3940, %3721, !dbg !33 + %4333 = fmul float %3902, %3722, !dbg !33 + %4334 = fmul float %3902, %3723, !dbg !33 + %4335 = fmul float %3902, %3724, !dbg !33 + %4336 = fmul float %3902, %3725, !dbg !33 + %4337 = fmul float %3902, %3726, !dbg !33 + %4338 = fmul float %3902, %3727, !dbg !33 + %4339 = fmul float %3902, %3728, !dbg !33 + %4340 = fmul float %3902, %3729, !dbg !33 + %4341 = fmul float %3941, %3730, !dbg !33 + %4342 = fmul float %3941, %3731, !dbg !33 + %4343 = fmul float %3941, %3732, !dbg !33 + %4344 = fmul float %3941, %3733, !dbg !33 + %4345 = fmul float %3941, %3734, !dbg !33 + %4346 = fmul float %3941, %3735, !dbg !33 + %4347 = fmul float %3941, %3736, !dbg !33 + %4348 = fmul float %3941, %3737, !dbg !33 + %4349 = fmul float %3904, %3738, !dbg !33 + %4350 = fmul float %3904, %3739, !dbg !33 + %4351 = fmul float %3904, %3740, !dbg !33 + %4352 = fmul float %3904, %3741, !dbg !33 + %4353 = fmul float %3904, %3742, !dbg !33 + %4354 = fmul float %3904, %3743, !dbg !33 + %4355 = fmul float %3904, %3744, !dbg !33 + %4356 = fmul float %3904, %3745, !dbg !33 + %4357 = fmul float %3942, %3746, !dbg !33 + %4358 = fmul float %3942, %3747, !dbg !33 + %4359 = fmul float %3942, %3748, !dbg !33 + %4360 = fmul float %3942, %3749, !dbg !33 + %4361 = fmul float %3942, %3750, !dbg !33 + %4362 = fmul float %3942, %3751, !dbg !33 + %4363 = fmul float %3942, %3752, !dbg !33 + %4364 = fmul float %3942, %3753, !dbg !33 + %4365 = fmul float %3906, %3754, !dbg !33 + %4366 = fmul float %3906, %3755, !dbg !33 + %4367 = fmul float %3906, %3756, !dbg !33 + %4368 = fmul float %3906, %3757, !dbg !33 + %4369 = fmul float %3906, %3758, !dbg !33 + %4370 = fmul float %3906, %3759, !dbg !33 + %4371 = fmul float %3906, %3760, !dbg !33 + %4372 = fmul float %3906, %3761, !dbg !33 + %4373 = fmul float %3943, %3762, !dbg !33 + %4374 = fmul float %3943, %3763, !dbg !33 + %4375 = fmul float %3943, %3764, !dbg !33 + %4376 = fmul float %3943, %3765, !dbg !33 + %4377 = fmul float %3943, %3766, !dbg !33 + %4378 = fmul float %3943, %3767, !dbg !33 + %4379 = fmul float %3943, %3768, !dbg !33 + %4380 = fmul float %3943, %3769, !dbg !33 + %4381 = fmul float %3908, %3770, !dbg !33 + %4382 = fmul float %3908, %3771, !dbg !33 + %4383 = fmul float %3908, %3772, !dbg !33 + %4384 = fmul float %3908, %3773, !dbg !33 + %4385 = fmul float %3908, %3774, !dbg !33 + %4386 = fmul float %3908, %3775, !dbg !33 + %4387 = fmul float %3908, %3776, !dbg !33 + %4388 = fmul float %3908, %3777, !dbg !33 + %4389 = fmul float %3944, %3778, !dbg !33 + %4390 = fmul float %3944, %3779, !dbg !33 + %4391 = fmul float %3944, %3780, !dbg !33 + %4392 = fmul float %3944, %3781, !dbg !33 + %4393 = fmul float %3944, %3782, !dbg !33 + %4394 = fmul float %3944, %3783, !dbg !33 + %4395 = fmul float %3944, %3784, !dbg !33 + %4396 = fmul float %3944, %3785, !dbg !33 + %4397 = fmul float %3910, %3786, !dbg !33 + %4398 = fmul float %3910, %3787, !dbg !33 + %4399 = fmul float %3910, %3788, !dbg !33 + %4400 = fmul float %3910, %3789, !dbg !33 + %4401 = fmul float %3910, %3790, !dbg !33 + %4402 = fmul float %3910, %3791, !dbg !33 + %4403 = fmul float %3910, %3792, !dbg !33 + %4404 = fmul float %3910, %3793, !dbg !33 + %4405 = fmul float %3945, %3794, !dbg !33 + %4406 = fmul float %3945, %3795, !dbg !33 + %4407 = fmul float %3945, %3796, !dbg !33 + %4408 = fmul float %3945, %3797, !dbg !33 + %4409 = fmul float %3945, %3798, !dbg !33 + %4410 = fmul float %3945, %3799, !dbg !33 + %4411 = fmul float %3945, %3800, !dbg !33 + %4412 = fmul float %3945, %3801, !dbg !33 + %4413 = fmul float %3912, %3802, !dbg !33 + %4414 = fmul float %3912, %3803, !dbg !33 + %4415 = fmul float %3912, %3804, !dbg !33 + %4416 = fmul float %3912, %3805, !dbg !33 + %4417 = fmul float %3912, %3806, !dbg !33 + %4418 = fmul float %3912, %3807, !dbg !33 + %4419 = fmul float %3912, %3808, !dbg !33 + %4420 = fmul float %3912, %3809, !dbg !33 + %4421 = fmul float %3946, %3810, !dbg !33 + %4422 = fmul float %3946, %3811, !dbg !33 + %4423 = fmul float %3946, %3812, !dbg !33 + %4424 = fmul float %3946, %3813, !dbg !33 + %4425 = fmul float %3946, %3814, !dbg !33 + %4426 = fmul float %3946, %3815, !dbg !33 + %4427 = fmul float %3946, %3816, !dbg !33 + %4428 = fmul float %3946, %3817, !dbg !33 + %4429 = fmul float %3914, %3818, !dbg !33 + %4430 = fmul float %3914, %3819, !dbg !33 + %4431 = fmul float %3914, %3820, !dbg !33 + %4432 = fmul float %3914, %3821, !dbg !33 + %4433 = fmul float %3914, %3822, !dbg !33 + %4434 = fmul float %3914, %3823, !dbg !33 + %4435 = fmul float %3914, %3824, !dbg !33 + %4436 = fmul float %3914, %3825, !dbg !33 + %4437 = fmul float %3947, %3826, !dbg !33 + %4438 = fmul float %3947, %3827, !dbg !33 + %4439 = fmul float %3947, %3828, !dbg !33 + %4440 = fmul float %3947, %3829, !dbg !33 + %4441 = fmul float %3947, %3830, !dbg !33 + %4442 = fmul float %3947, %3831, !dbg !33 + %4443 = fmul float %3947, %3832, !dbg !33 + %4444 = fmul float %3947, %3833, !dbg !33 + %4445 = fmul float %3916, %3834, !dbg !33 + %4446 = fmul float %3916, %3835, !dbg !33 + %4447 = fmul float %3916, %3836, !dbg !33 + %4448 = fmul float %3916, %3837, !dbg !33 + %4449 = fmul float %3916, %3838, !dbg !33 + %4450 = fmul float %3916, %3839, !dbg !33 + %4451 = fmul float %3916, %3840, !dbg !33 + %4452 = fmul float %3916, %3841, !dbg !33 + %4453 = fmul float %3948, %3842, !dbg !33 + %4454 = fmul float %3948, %3843, !dbg !33 + %4455 = fmul float %3948, %3844, !dbg !33 + %4456 = fmul float %3948, %3845, !dbg !33 + %4457 = fmul float %3948, %3846, !dbg !33 + %4458 = fmul float %3948, %3847, !dbg !33 + %4459 = fmul float %3948, %3848, !dbg !33 + %4460 = fmul float %3948, %3849, !dbg !33 + %4461 = fmul float %1730, %3338, !dbg !34 + %4462 = fmul float %1731, %3339, !dbg !34 + %4463 = fmul float %1732, %3340, !dbg !34 + %4464 = fmul float %1733, %3341, !dbg !34 + %4465 = fmul float %1734, %3342, !dbg !34 + %4466 = fmul float %1735, %3343, !dbg !34 + %4467 = fmul float %1736, %3344, !dbg !34 + %4468 = fmul float %1737, %3345, !dbg !34 + %4469 = fmul float %1738, %3346, !dbg !34 + %4470 = fmul float %1739, %3347, !dbg !34 + %4471 = fmul float %1740, %3348, !dbg !34 + %4472 = fmul float %1741, %3349, !dbg !34 + %4473 = fmul float %1742, %3350, !dbg !34 + %4474 = fmul float %1743, %3351, !dbg !34 + %4475 = fmul float %1744, %3352, !dbg !34 + %4476 = fmul float %1745, %3353, !dbg !34 + %4477 = fmul float %1746, %3354, !dbg !34 + %4478 = fmul float %1747, %3355, !dbg !34 + %4479 = fmul float %1748, %3356, !dbg !34 + %4480 = fmul float %1749, %3357, !dbg !34 + %4481 = fmul float %1750, %3358, !dbg !34 + %4482 = fmul float %1751, %3359, !dbg !34 + %4483 = fmul float %1752, %3360, !dbg !34 + %4484 = fmul float %1753, %3361, !dbg !34 + %4485 = fmul float %1754, %3362, !dbg !34 + %4486 = fmul float %1755, %3363, !dbg !34 + %4487 = fmul float %1756, %3364, !dbg !34 + %4488 = fmul float %1757, %3365, !dbg !34 + %4489 = fmul float %1758, %3366, !dbg !34 + %4490 = fmul float %1759, %3367, !dbg !34 + %4491 = fmul float %1760, %3368, !dbg !34 + %4492 = fmul float %1761, %3369, !dbg !34 + %4493 = fmul float %1762, %3370, !dbg !34 + %4494 = fmul float %1763, %3371, !dbg !34 + %4495 = fmul float %1764, %3372, !dbg !34 + %4496 = fmul float %1765, %3373, !dbg !34 + %4497 = fmul float %1766, %3374, !dbg !34 + %4498 = fmul float %1767, %3375, !dbg !34 + %4499 = fmul float %1768, %3376, !dbg !34 + %4500 = fmul float %1769, %3377, !dbg !34 + %4501 = fmul float %1770, %3378, !dbg !34 + %4502 = fmul float %1771, %3379, !dbg !34 + %4503 = fmul float %1772, %3380, !dbg !34 + %4504 = fmul float %1773, %3381, !dbg !34 + %4505 = fmul float %1774, %3382, !dbg !34 + %4506 = fmul float %1775, %3383, !dbg !34 + %4507 = fmul float %1776, %3384, !dbg !34 + %4508 = fmul float %1777, %3385, !dbg !34 + %4509 = fmul float %1778, %3386, !dbg !34 + %4510 = fmul float %1779, %3387, !dbg !34 + %4511 = fmul float %1780, %3388, !dbg !34 + %4512 = fmul float %1781, %3389, !dbg !34 + %4513 = fmul float %1782, %3390, !dbg !34 + %4514 = fmul float %1783, %3391, !dbg !34 + %4515 = fmul float %1784, %3392, !dbg !34 + %4516 = fmul float %1785, %3393, !dbg !34 + %4517 = fmul float %1786, %3394, !dbg !34 + %4518 = fmul float %1787, %3395, !dbg !34 + %4519 = fmul float %1788, %3396, !dbg !34 + %4520 = fmul float %1789, %3397, !dbg !34 + %4521 = fmul float %1790, %3398, !dbg !34 + %4522 = fmul float %1791, %3399, !dbg !34 + %4523 = fmul float %1792, %3400, !dbg !34 + %4524 = fmul float %1793, %3401, !dbg !34 + %4525 = fmul float %1794, %3402, !dbg !34 + %4526 = fmul float %1795, %3403, !dbg !34 + %4527 = fmul float %1796, %3404, !dbg !34 + %4528 = fmul float %1797, %3405, !dbg !34 + %4529 = fmul float %1798, %3406, !dbg !34 + %4530 = fmul float %1799, %3407, !dbg !34 + %4531 = fmul float %1800, %3408, !dbg !34 + %4532 = fmul float %1801, %3409, !dbg !34 + %4533 = fmul float %1802, %3410, !dbg !34 + %4534 = fmul float %1803, %3411, !dbg !34 + %4535 = fmul float %1804, %3412, !dbg !34 + %4536 = fmul float %1805, %3413, !dbg !34 + %4537 = fmul float %1806, %3414, !dbg !34 + %4538 = fmul float %1807, %3415, !dbg !34 + %4539 = fmul float %1808, %3416, !dbg !34 + %4540 = fmul float %1809, %3417, !dbg !34 + %4541 = fmul float %1810, %3418, !dbg !34 + %4542 = fmul float %1811, %3419, !dbg !34 + %4543 = fmul float %1812, %3420, !dbg !34 + %4544 = fmul float %1813, %3421, !dbg !34 + %4545 = fmul float %1814, %3422, !dbg !34 + %4546 = fmul float %1815, %3423, !dbg !34 + %4547 = fmul float %1816, %3424, !dbg !34 + %4548 = fmul float %1817, %3425, !dbg !34 + %4549 = fmul float %1818, %3426, !dbg !34 + %4550 = fmul float %1819, %3427, !dbg !34 + %4551 = fmul float %1820, %3428, !dbg !34 + %4552 = fmul float %1821, %3429, !dbg !34 + %4553 = fmul float %1822, %3430, !dbg !34 + %4554 = fmul float %1823, %3431, !dbg !34 + %4555 = fmul float %1824, %3432, !dbg !34 + %4556 = fmul float %1825, %3433, !dbg !34 + %4557 = fmul float %1826, %3434, !dbg !34 + %4558 = fmul float %1827, %3435, !dbg !34 + %4559 = fmul float %1828, %3436, !dbg !34 + %4560 = fmul float %1829, %3437, !dbg !34 + %4561 = fmul float %1830, %3438, !dbg !34 + %4562 = fmul float %1831, %3439, !dbg !34 + %4563 = fmul float %1832, %3440, !dbg !34 + %4564 = fmul float %1833, %3441, !dbg !34 + %4565 = fmul float %1834, %3442, !dbg !34 + %4566 = fmul float %1835, %3443, !dbg !34 + %4567 = fmul float %1836, %3444, !dbg !34 + %4568 = fmul float %1837, %3445, !dbg !34 + %4569 = fmul float %1838, %3446, !dbg !34 + %4570 = fmul float %1839, %3447, !dbg !34 + %4571 = fmul float %1840, %3448, !dbg !34 + %4572 = fmul float %1841, %3449, !dbg !34 + %4573 = fmul float %1842, %3450, !dbg !34 + %4574 = fmul float %1843, %3451, !dbg !34 + %4575 = fmul float %1844, %3452, !dbg !34 + %4576 = fmul float %1845, %3453, !dbg !34 + %4577 = fmul float %1846, %3454, !dbg !34 + %4578 = fmul float %1847, %3455, !dbg !34 + %4579 = fmul float %1848, %3456, !dbg !34 + %4580 = fmul float %1849, %3457, !dbg !34 + %4581 = fmul float %1850, %3458, !dbg !34 + %4582 = fmul float %1851, %3459, !dbg !34 + %4583 = fmul float %1852, %3460, !dbg !34 + %4584 = fmul float %1853, %3461, !dbg !34 + %4585 = fmul float %1854, %3462, !dbg !34 + %4586 = fmul float %1855, %3463, !dbg !34 + %4587 = fmul float %1856, %3464, !dbg !34 + %4588 = fmul float %1857, %3465, !dbg !34 + %4589 = fmul float %1858, %3466, !dbg !34 + %4590 = fmul float %1859, %3467, !dbg !34 + %4591 = fmul float %1860, %3468, !dbg !34 + %4592 = fmul float %1861, %3469, !dbg !34 + %4593 = fmul float %1862, %3470, !dbg !34 + %4594 = fmul float %1863, %3471, !dbg !34 + %4595 = fmul float %1864, %3472, !dbg !34 + %4596 = fmul float %1865, %3473, !dbg !34 + %4597 = fmul float %1866, %3474, !dbg !34 + %4598 = fmul float %1867, %3475, !dbg !34 + %4599 = fmul float %1868, %3476, !dbg !34 + %4600 = fmul float %1869, %3477, !dbg !34 + %4601 = fmul float %1870, %3478, !dbg !34 + %4602 = fmul float %1871, %3479, !dbg !34 + %4603 = fmul float %1872, %3480, !dbg !34 + %4604 = fmul float %1873, %3481, !dbg !34 + %4605 = fmul float %1874, %3482, !dbg !34 + %4606 = fmul float %1875, %3483, !dbg !34 + %4607 = fmul float %1876, %3484, !dbg !34 + %4608 = fmul float %1877, %3485, !dbg !34 + %4609 = fmul float %1878, %3486, !dbg !34 + %4610 = fmul float %1879, %3487, !dbg !34 + %4611 = fmul float %1880, %3488, !dbg !34 + %4612 = fmul float %1881, %3489, !dbg !34 + %4613 = fmul float %1882, %3490, !dbg !34 + %4614 = fmul float %1883, %3491, !dbg !34 + %4615 = fmul float %1884, %3492, !dbg !34 + %4616 = fmul float %1885, %3493, !dbg !34 + %4617 = fmul float %1886, %3494, !dbg !34 + %4618 = fmul float %1887, %3495, !dbg !34 + %4619 = fmul float %1888, %3496, !dbg !34 + %4620 = fmul float %1889, %3497, !dbg !34 + %4621 = fmul float %1890, %3498, !dbg !34 + %4622 = fmul float %1891, %3499, !dbg !34 + %4623 = fmul float %1892, %3500, !dbg !34 + %4624 = fmul float %1893, %3501, !dbg !34 + %4625 = fmul float %1894, %3502, !dbg !34 + %4626 = fmul float %1895, %3503, !dbg !34 + %4627 = fmul float %1896, %3504, !dbg !34 + %4628 = fmul float %1897, %3505, !dbg !34 + %4629 = fmul float %1898, %3506, !dbg !34 + %4630 = fmul float %1899, %3507, !dbg !34 + %4631 = fmul float %1900, %3508, !dbg !34 + %4632 = fmul float %1901, %3509, !dbg !34 + %4633 = fmul float %1902, %3510, !dbg !34 + %4634 = fmul float %1903, %3511, !dbg !34 + %4635 = fmul float %1904, %3512, !dbg !34 + %4636 = fmul float %1905, %3513, !dbg !34 + %4637 = fmul float %1906, %3514, !dbg !34 + %4638 = fmul float %1907, %3515, !dbg !34 + %4639 = fmul float %1908, %3516, !dbg !34 + %4640 = fmul float %1909, %3517, !dbg !34 + %4641 = fmul float %1910, %3518, !dbg !34 + %4642 = fmul float %1911, %3519, !dbg !34 + %4643 = fmul float %1912, %3520, !dbg !34 + %4644 = fmul float %1913, %3521, !dbg !34 + %4645 = fmul float %1914, %3522, !dbg !34 + %4646 = fmul float %1915, %3523, !dbg !34 + %4647 = fmul float %1916, %3524, !dbg !34 + %4648 = fmul float %1917, %3525, !dbg !34 + %4649 = fmul float %1918, %3526, !dbg !34 + %4650 = fmul float %1919, %3527, !dbg !34 + %4651 = fmul float %1920, %3528, !dbg !34 + %4652 = fmul float %1921, %3529, !dbg !34 + %4653 = fmul float %1922, %3530, !dbg !34 + %4654 = fmul float %1923, %3531, !dbg !34 + %4655 = fmul float %1924, %3532, !dbg !34 + %4656 = fmul float %1925, %3533, !dbg !34 + %4657 = fmul float %1926, %3534, !dbg !34 + %4658 = fmul float %1927, %3535, !dbg !34 + %4659 = fmul float %1928, %3536, !dbg !34 + %4660 = fmul float %1929, %3537, !dbg !34 + %4661 = fmul float %1930, %3538, !dbg !34 + %4662 = fmul float %1931, %3539, !dbg !34 + %4663 = fmul float %1932, %3540, !dbg !34 + %4664 = fmul float %1933, %3541, !dbg !34 + %4665 = fmul float %1934, %3542, !dbg !34 + %4666 = fmul float %1935, %3543, !dbg !34 + %4667 = fmul float %1936, %3544, !dbg !34 + %4668 = fmul float %1937, %3545, !dbg !34 + %4669 = fmul float %1938, %3546, !dbg !34 + %4670 = fmul float %1939, %3547, !dbg !34 + %4671 = fmul float %1940, %3548, !dbg !34 + %4672 = fmul float %1941, %3549, !dbg !34 + %4673 = fmul float %1942, %3550, !dbg !34 + %4674 = fmul float %1943, %3551, !dbg !34 + %4675 = fmul float %1944, %3552, !dbg !34 + %4676 = fmul float %1945, %3553, !dbg !34 + %4677 = fmul float %1946, %3554, !dbg !34 + %4678 = fmul float %1947, %3555, !dbg !34 + %4679 = fmul float %1948, %3556, !dbg !34 + %4680 = fmul float %1949, %3557, !dbg !34 + %4681 = fmul float %1950, %3558, !dbg !34 + %4682 = fmul float %1951, %3559, !dbg !34 + %4683 = fmul float %1952, %3560, !dbg !34 + %4684 = fmul float %1953, %3561, !dbg !34 + %4685 = fmul float %1954, %3562, !dbg !34 + %4686 = fmul float %1955, %3563, !dbg !34 + %4687 = fmul float %1956, %3564, !dbg !34 + %4688 = fmul float %1957, %3565, !dbg !34 + %4689 = fmul float %1958, %3566, !dbg !34 + %4690 = fmul float %1959, %3567, !dbg !34 + %4691 = fmul float %1960, %3568, !dbg !34 + %4692 = fmul float %1961, %3569, !dbg !34 + %4693 = fmul float %1962, %3570, !dbg !34 + %4694 = fmul float %1963, %3571, !dbg !34 + %4695 = fmul float %1964, %3572, !dbg !34 + %4696 = fmul float %1965, %3573, !dbg !34 + %4697 = fmul float %1966, %3574, !dbg !34 + %4698 = fmul float %1967, %3575, !dbg !34 + %4699 = fmul float %1968, %3576, !dbg !34 + %4700 = fmul float %1969, %3577, !dbg !34 + %4701 = fmul float %1970, %3578, !dbg !34 + %4702 = fmul float %1971, %3579, !dbg !34 + %4703 = fmul float %1972, %3580, !dbg !34 + %4704 = fmul float %1973, %3581, !dbg !34 + %4705 = fmul float %1974, %3582, !dbg !34 + %4706 = fmul float %1975, %3583, !dbg !34 + %4707 = fmul float %1976, %3584, !dbg !34 + %4708 = fmul float %1977, %3585, !dbg !34 + %4709 = fmul float %1978, %3586, !dbg !34 + %4710 = fmul float %1979, %3587, !dbg !34 + %4711 = fmul float %1980, %3588, !dbg !34 + %4712 = fmul float %1981, %3589, !dbg !34 + %4713 = fmul float %1982, %3590, !dbg !34 + %4714 = fmul float %1983, %3591, !dbg !34 + %4715 = fmul float %1984, %3592, !dbg !34 + %4716 = fmul float %1985, %3593, !dbg !34 + %4717 = fmul float %1986, %3594, !dbg !34 + %4718 = fmul float %1987, %3595, !dbg !34 + %4719 = fmul float %1988, %3596, !dbg !34 + %4720 = fmul float %1989, %3597, !dbg !34 + %4721 = fmul float %1990, %3598, !dbg !34 + %4722 = fmul float %1991, %3599, !dbg !34 + %4723 = fmul float %1992, %3600, !dbg !34 + %4724 = fmul float %1993, %3601, !dbg !34 + %4725 = fmul float %1994, %3602, !dbg !34 + %4726 = fmul float %1995, %3603, !dbg !34 + %4727 = fmul float %1996, %3604, !dbg !34 + %4728 = fmul float %1997, %3605, !dbg !34 + %4729 = fmul float %1998, %3606, !dbg !34 + %4730 = fmul float %1999, %3607, !dbg !34 + %4731 = fmul float %2000, %3608, !dbg !34 + %4732 = fmul float %2001, %3609, !dbg !34 + %4733 = fmul float %2002, %3610, !dbg !34 + %4734 = fmul float %2003, %3611, !dbg !34 + %4735 = fmul float %2004, %3612, !dbg !34 + %4736 = fmul float %2005, %3613, !dbg !34 + %4737 = fmul float %2006, %3614, !dbg !34 + %4738 = fmul float %2007, %3615, !dbg !34 + %4739 = fmul float %2008, %3616, !dbg !34 + %4740 = fmul float %2009, %3617, !dbg !34 + %4741 = fmul float %2010, %3618, !dbg !34 + %4742 = fmul float %2011, %3619, !dbg !34 + %4743 = fmul float %2012, %3620, !dbg !34 + %4744 = fmul float %2013, %3621, !dbg !34 + %4745 = fmul float %2014, %3622, !dbg !34 + %4746 = fmul float %2015, %3623, !dbg !34 + %4747 = fmul float %2016, %3624, !dbg !34 + %4748 = fmul float %2017, %3625, !dbg !34 + %4749 = fmul float %2018, %3626, !dbg !34 + %4750 = fmul float %2019, %3627, !dbg !34 + %4751 = fmul float %2020, %3628, !dbg !34 + %4752 = fmul float %2021, %3629, !dbg !34 + %4753 = fmul float %2022, %3630, !dbg !34 + %4754 = fmul float %2023, %3631, !dbg !34 + %4755 = fmul float %2024, %3632, !dbg !34 + %4756 = fmul float %2025, %3633, !dbg !34 + %4757 = fmul float %2026, %3634, !dbg !34 + %4758 = fmul float %2027, %3635, !dbg !34 + %4759 = fmul float %2028, %3636, !dbg !34 + %4760 = fmul float %2029, %3637, !dbg !34 + %4761 = fmul float %2030, %3638, !dbg !34 + %4762 = fmul float %2031, %3639, !dbg !34 + %4763 = fmul float %2032, %3640, !dbg !34 + %4764 = fmul float %2033, %3641, !dbg !34 + %4765 = fmul float %2034, %3642, !dbg !34 + %4766 = fmul float %2035, %3643, !dbg !34 + %4767 = fmul float %2036, %3644, !dbg !34 + %4768 = fmul float %2037, %3645, !dbg !34 + %4769 = fmul float %2038, %3646, !dbg !34 + %4770 = fmul float %2039, %3647, !dbg !34 + %4771 = fmul float %2040, %3648, !dbg !34 + %4772 = fmul float %2041, %3649, !dbg !34 + %4773 = fmul float %2042, %3650, !dbg !34 + %4774 = fmul float %2043, %3651, !dbg !34 + %4775 = fmul float %2044, %3652, !dbg !34 + %4776 = fmul float %2045, %3653, !dbg !34 + %4777 = fmul float %2046, %3654, !dbg !34 + %4778 = fmul float %2047, %3655, !dbg !34 + %4779 = fmul float %2048, %3656, !dbg !34 + %4780 = fmul float %2049, %3657, !dbg !34 + %4781 = fmul float %2050, %3658, !dbg !34 + %4782 = fmul float %2051, %3659, !dbg !34 + %4783 = fmul float %2052, %3660, !dbg !34 + %4784 = fmul float %2053, %3661, !dbg !34 + %4785 = fmul float %2054, %3662, !dbg !34 + %4786 = fmul float %2055, %3663, !dbg !34 + %4787 = fmul float %2056, %3664, !dbg !34 + %4788 = fmul float %2057, %3665, !dbg !34 + %4789 = fmul float %2058, %3666, !dbg !34 + %4790 = fmul float %2059, %3667, !dbg !34 + %4791 = fmul float %2060, %3668, !dbg !34 + %4792 = fmul float %2061, %3669, !dbg !34 + %4793 = fmul float %2062, %3670, !dbg !34 + %4794 = fmul float %2063, %3671, !dbg !34 + %4795 = fmul float %2064, %3672, !dbg !34 + %4796 = fmul float %2065, %3673, !dbg !34 + %4797 = fmul float %2066, %3674, !dbg !34 + %4798 = fmul float %2067, %3675, !dbg !34 + %4799 = fmul float %2068, %3676, !dbg !34 + %4800 = fmul float %2069, %3677, !dbg !34 + %4801 = fmul float %2070, %3678, !dbg !34 + %4802 = fmul float %2071, %3679, !dbg !34 + %4803 = fmul float %2072, %3680, !dbg !34 + %4804 = fmul float %2073, %3681, !dbg !34 + %4805 = fmul float %2074, %3682, !dbg !34 + %4806 = fmul float %2075, %3683, !dbg !34 + %4807 = fmul float %2076, %3684, !dbg !34 + %4808 = fmul float %2077, %3685, !dbg !34 + %4809 = fmul float %2078, %3686, !dbg !34 + %4810 = fmul float %2079, %3687, !dbg !34 + %4811 = fmul float %2080, %3688, !dbg !34 + %4812 = fmul float %2081, %3689, !dbg !34 + %4813 = fmul float %2082, %3690, !dbg !34 + %4814 = fmul float %2083, %3691, !dbg !34 + %4815 = fmul float %2084, %3692, !dbg !34 + %4816 = fmul float %2085, %3693, !dbg !34 + %4817 = fmul float %2086, %3694, !dbg !34 + %4818 = fmul float %2087, %3695, !dbg !34 + %4819 = fmul float %2088, %3696, !dbg !34 + %4820 = fmul float %2089, %3697, !dbg !34 + %4821 = fmul float %2090, %3698, !dbg !34 + %4822 = fmul float %2091, %3699, !dbg !34 + %4823 = fmul float %2092, %3700, !dbg !34 + %4824 = fmul float %2093, %3701, !dbg !34 + %4825 = fmul float %2094, %3702, !dbg !34 + %4826 = fmul float %2095, %3703, !dbg !34 + %4827 = fmul float %2096, %3704, !dbg !34 + %4828 = fmul float %2097, %3705, !dbg !34 + %4829 = fmul float %2098, %3706, !dbg !34 + %4830 = fmul float %2099, %3707, !dbg !34 + %4831 = fmul float %2100, %3708, !dbg !34 + %4832 = fmul float %2101, %3709, !dbg !34 + %4833 = fmul float %2102, %3710, !dbg !34 + %4834 = fmul float %2103, %3711, !dbg !34 + %4835 = fmul float %2104, %3712, !dbg !34 + %4836 = fmul float %2105, %3713, !dbg !34 + %4837 = fmul float %2106, %3714, !dbg !34 + %4838 = fmul float %2107, %3715, !dbg !34 + %4839 = fmul float %2108, %3716, !dbg !34 + %4840 = fmul float %2109, %3717, !dbg !34 + %4841 = fmul float %2110, %3718, !dbg !34 + %4842 = fmul float %2111, %3719, !dbg !34 + %4843 = fmul float %2112, %3720, !dbg !34 + %4844 = fmul float %2113, %3721, !dbg !34 + %4845 = fmul float %2114, %3722, !dbg !34 + %4846 = fmul float %2115, %3723, !dbg !34 + %4847 = fmul float %2116, %3724, !dbg !34 + %4848 = fmul float %2117, %3725, !dbg !34 + %4849 = fmul float %2118, %3726, !dbg !34 + %4850 = fmul float %2119, %3727, !dbg !34 + %4851 = fmul float %2120, %3728, !dbg !34 + %4852 = fmul float %2121, %3729, !dbg !34 + %4853 = fmul float %2122, %3730, !dbg !34 + %4854 = fmul float %2123, %3731, !dbg !34 + %4855 = fmul float %2124, %3732, !dbg !34 + %4856 = fmul float %2125, %3733, !dbg !34 + %4857 = fmul float %2126, %3734, !dbg !34 + %4858 = fmul float %2127, %3735, !dbg !34 + %4859 = fmul float %2128, %3736, !dbg !34 + %4860 = fmul float %2129, %3737, !dbg !34 + %4861 = fmul float %2130, %3738, !dbg !34 + %4862 = fmul float %2131, %3739, !dbg !34 + %4863 = fmul float %2132, %3740, !dbg !34 + %4864 = fmul float %2133, %3741, !dbg !34 + %4865 = fmul float %2134, %3742, !dbg !34 + %4866 = fmul float %2135, %3743, !dbg !34 + %4867 = fmul float %2136, %3744, !dbg !34 + %4868 = fmul float %2137, %3745, !dbg !34 + %4869 = fmul float %2138, %3746, !dbg !34 + %4870 = fmul float %2139, %3747, !dbg !34 + %4871 = fmul float %2140, %3748, !dbg !34 + %4872 = fmul float %2141, %3749, !dbg !34 + %4873 = fmul float %2142, %3750, !dbg !34 + %4874 = fmul float %2143, %3751, !dbg !34 + %4875 = fmul float %2144, %3752, !dbg !34 + %4876 = fmul float %2145, %3753, !dbg !34 + %4877 = fmul float %2146, %3754, !dbg !34 + %4878 = fmul float %2147, %3755, !dbg !34 + %4879 = fmul float %2148, %3756, !dbg !34 + %4880 = fmul float %2149, %3757, !dbg !34 + %4881 = fmul float %2150, %3758, !dbg !34 + %4882 = fmul float %2151, %3759, !dbg !34 + %4883 = fmul float %2152, %3760, !dbg !34 + %4884 = fmul float %2153, %3761, !dbg !34 + %4885 = fmul float %2154, %3762, !dbg !34 + %4886 = fmul float %2155, %3763, !dbg !34 + %4887 = fmul float %2156, %3764, !dbg !34 + %4888 = fmul float %2157, %3765, !dbg !34 + %4889 = fmul float %2158, %3766, !dbg !34 + %4890 = fmul float %2159, %3767, !dbg !34 + %4891 = fmul float %2160, %3768, !dbg !34 + %4892 = fmul float %2161, %3769, !dbg !34 + %4893 = fmul float %2162, %3770, !dbg !34 + %4894 = fmul float %2163, %3771, !dbg !34 + %4895 = fmul float %2164, %3772, !dbg !34 + %4896 = fmul float %2165, %3773, !dbg !34 + %4897 = fmul float %2166, %3774, !dbg !34 + %4898 = fmul float %2167, %3775, !dbg !34 + %4899 = fmul float %2168, %3776, !dbg !34 + %4900 = fmul float %2169, %3777, !dbg !34 + %4901 = fmul float %2170, %3778, !dbg !34 + %4902 = fmul float %2171, %3779, !dbg !34 + %4903 = fmul float %2172, %3780, !dbg !34 + %4904 = fmul float %2173, %3781, !dbg !34 + %4905 = fmul float %2174, %3782, !dbg !34 + %4906 = fmul float %2175, %3783, !dbg !34 + %4907 = fmul float %2176, %3784, !dbg !34 + %4908 = fmul float %2177, %3785, !dbg !34 + %4909 = fmul float %2178, %3786, !dbg !34 + %4910 = fmul float %2179, %3787, !dbg !34 + %4911 = fmul float %2180, %3788, !dbg !34 + %4912 = fmul float %2181, %3789, !dbg !34 + %4913 = fmul float %2182, %3790, !dbg !34 + %4914 = fmul float %2183, %3791, !dbg !34 + %4915 = fmul float %2184, %3792, !dbg !34 + %4916 = fmul float %2185, %3793, !dbg !34 + %4917 = fmul float %2186, %3794, !dbg !34 + %4918 = fmul float %2187, %3795, !dbg !34 + %4919 = fmul float %2188, %3796, !dbg !34 + %4920 = fmul float %2189, %3797, !dbg !34 + %4921 = fmul float %2190, %3798, !dbg !34 + %4922 = fmul float %2191, %3799, !dbg !34 + %4923 = fmul float %2192, %3800, !dbg !34 + %4924 = fmul float %2193, %3801, !dbg !34 + %4925 = fmul float %2194, %3802, !dbg !34 + %4926 = fmul float %2195, %3803, !dbg !34 + %4927 = fmul float %2196, %3804, !dbg !34 + %4928 = fmul float %2197, %3805, !dbg !34 + %4929 = fmul float %2198, %3806, !dbg !34 + %4930 = fmul float %2199, %3807, !dbg !34 + %4931 = fmul float %2200, %3808, !dbg !34 + %4932 = fmul float %2201, %3809, !dbg !34 + %4933 = fmul float %2202, %3810, !dbg !34 + %4934 = fmul float %2203, %3811, !dbg !34 + %4935 = fmul float %2204, %3812, !dbg !34 + %4936 = fmul float %2205, %3813, !dbg !34 + %4937 = fmul float %2206, %3814, !dbg !34 + %4938 = fmul float %2207, %3815, !dbg !34 + %4939 = fmul float %2208, %3816, !dbg !34 + %4940 = fmul float %2209, %3817, !dbg !34 + %4941 = fmul float %2210, %3818, !dbg !34 + %4942 = fmul float %2211, %3819, !dbg !34 + %4943 = fmul float %2212, %3820, !dbg !34 + %4944 = fmul float %2213, %3821, !dbg !34 + %4945 = fmul float %2214, %3822, !dbg !34 + %4946 = fmul float %2215, %3823, !dbg !34 + %4947 = fmul float %2216, %3824, !dbg !34 + %4948 = fmul float %2217, %3825, !dbg !34 + %4949 = fmul float %2218, %3826, !dbg !34 + %4950 = fmul float %2219, %3827, !dbg !34 + %4951 = fmul float %2220, %3828, !dbg !34 + %4952 = fmul float %2221, %3829, !dbg !34 + %4953 = fmul float %2222, %3830, !dbg !34 + %4954 = fmul float %2223, %3831, !dbg !34 + %4955 = fmul float %2224, %3832, !dbg !34 + %4956 = fmul float %2225, %3833, !dbg !34 + %4957 = fmul float %2226, %3834, !dbg !34 + %4958 = fmul float %2227, %3835, !dbg !34 + %4959 = fmul float %2228, %3836, !dbg !34 + %4960 = fmul float %2229, %3837, !dbg !34 + %4961 = fmul float %2230, %3838, !dbg !34 + %4962 = fmul float %2231, %3839, !dbg !34 + %4963 = fmul float %2232, %3840, !dbg !34 + %4964 = fmul float %2233, %3841, !dbg !34 + %4965 = fmul float %2234, %3842, !dbg !34 + %4966 = fmul float %2235, %3843, !dbg !34 + %4967 = fmul float %2236, %3844, !dbg !34 + %4968 = fmul float %2237, %3845, !dbg !34 + %4969 = fmul float %2238, %3846, !dbg !34 + %4970 = fmul float %2239, %3847, !dbg !34 + %4971 = fmul float %2240, %3848, !dbg !34 + %4972 = fmul float %2241, %3849, !dbg !34 + %4973 = fadd float %4461, %4462, !dbg !35 + %4974 = fadd float %4463, %4973, !dbg !35 + %4975 = fadd float %4464, %4974, !dbg !35 + %4976 = fadd float %4465, %4975, !dbg !35 + %4977 = fadd float %4466, %4976, !dbg !35 + %4978 = fadd float %4467, %4977, !dbg !35 + %4979 = fadd float %4468, %4978, !dbg !35 + %4980 = fadd float %4469, %4470, !dbg !35 + %4981 = fadd float %4471, %4980, !dbg !35 + %4982 = fadd float %4472, %4981, !dbg !35 + %4983 = fadd float %4473, %4982, !dbg !35 + %4984 = fadd float %4474, %4983, !dbg !35 + %4985 = fadd float %4475, %4984, !dbg !35 + %4986 = fadd float %4476, %4985, !dbg !35 + %4987 = fadd float %4477, %4478, !dbg !35 + %4988 = fadd float %4479, %4987, !dbg !35 + %4989 = fadd float %4480, %4988, !dbg !35 + %4990 = fadd float %4481, %4989, !dbg !35 + %4991 = fadd float %4482, %4990, !dbg !35 + %4992 = fadd float %4483, %4991, !dbg !35 + %4993 = fadd float %4484, %4992, !dbg !35 + %4994 = fadd float %4485, %4486, !dbg !35 + %4995 = fadd float %4487, %4994, !dbg !35 + %4996 = fadd float %4488, %4995, !dbg !35 + %4997 = fadd float %4489, %4996, !dbg !35 + %4998 = fadd float %4490, %4997, !dbg !35 + %4999 = fadd float %4491, %4998, !dbg !35 + %5000 = fadd float %4492, %4999, !dbg !35 + %5001 = fadd float %4493, %4494, !dbg !35 + %5002 = fadd float %4495, %5001, !dbg !35 + %5003 = fadd float %4496, %5002, !dbg !35 + %5004 = fadd float %4497, %5003, !dbg !35 + %5005 = fadd float %4498, %5004, !dbg !35 + %5006 = fadd float %4499, %5005, !dbg !35 + %5007 = fadd float %4500, %5006, !dbg !35 + %5008 = fadd float %4501, %4502, !dbg !35 + %5009 = fadd float %4503, %5008, !dbg !35 + %5010 = fadd float %4504, %5009, !dbg !35 + %5011 = fadd float %4505, %5010, !dbg !35 + %5012 = fadd float %4506, %5011, !dbg !35 + %5013 = fadd float %4507, %5012, !dbg !35 + %5014 = fadd float %4508, %5013, !dbg !35 + %5015 = fadd float %4509, %4510, !dbg !35 + %5016 = fadd float %4511, %5015, !dbg !35 + %5017 = fadd float %4512, %5016, !dbg !35 + %5018 = fadd float %4513, %5017, !dbg !35 + %5019 = fadd float %4514, %5018, !dbg !35 + %5020 = fadd float %4515, %5019, !dbg !35 + %5021 = fadd float %4516, %5020, !dbg !35 + %5022 = fadd float %4517, %4518, !dbg !35 + %5023 = fadd float %4519, %5022, !dbg !35 + %5024 = fadd float %4520, %5023, !dbg !35 + %5025 = fadd float %4521, %5024, !dbg !35 + %5026 = fadd float %4522, %5025, !dbg !35 + %5027 = fadd float %4523, %5026, !dbg !35 + %5028 = fadd float %4524, %5027, !dbg !35 + %5029 = fadd float %4525, %4526, !dbg !35 + %5030 = fadd float %4527, %5029, !dbg !35 + %5031 = fadd float %4528, %5030, !dbg !35 + %5032 = fadd float %4529, %5031, !dbg !35 + %5033 = fadd float %4530, %5032, !dbg !35 + %5034 = fadd float %4531, %5033, !dbg !35 + %5035 = fadd float %4532, %5034, !dbg !35 + %5036 = fadd float %4533, %4534, !dbg !35 + %5037 = fadd float %4535, %5036, !dbg !35 + %5038 = fadd float %4536, %5037, !dbg !35 + %5039 = fadd float %4537, %5038, !dbg !35 + %5040 = fadd float %4538, %5039, !dbg !35 + %5041 = fadd float %4539, %5040, !dbg !35 + %5042 = fadd float %4540, %5041, !dbg !35 + %5043 = fadd float %4541, %4542, !dbg !35 + %5044 = fadd float %4543, %5043, !dbg !35 + %5045 = fadd float %4544, %5044, !dbg !35 + %5046 = fadd float %4545, %5045, !dbg !35 + %5047 = fadd float %4546, %5046, !dbg !35 + %5048 = fadd float %4547, %5047, !dbg !35 + %5049 = fadd float %4548, %5048, !dbg !35 + %5050 = fadd float %4549, %4550, !dbg !35 + %5051 = fadd float %4551, %5050, !dbg !35 + %5052 = fadd float %4552, %5051, !dbg !35 + %5053 = fadd float %4553, %5052, !dbg !35 + %5054 = fadd float %4554, %5053, !dbg !35 + %5055 = fadd float %4555, %5054, !dbg !35 + %5056 = fadd float %4556, %5055, !dbg !35 + %5057 = fadd float %4557, %4558, !dbg !35 + %5058 = fadd float %4559, %5057, !dbg !35 + %5059 = fadd float %4560, %5058, !dbg !35 + %5060 = fadd float %4561, %5059, !dbg !35 + %5061 = fadd float %4562, %5060, !dbg !35 + %5062 = fadd float %4563, %5061, !dbg !35 + %5063 = fadd float %4564, %5062, !dbg !35 + %5064 = fadd float %4565, %4566, !dbg !35 + %5065 = fadd float %4567, %5064, !dbg !35 + %5066 = fadd float %4568, %5065, !dbg !35 + %5067 = fadd float %4569, %5066, !dbg !35 + %5068 = fadd float %4570, %5067, !dbg !35 + %5069 = fadd float %4571, %5068, !dbg !35 + %5070 = fadd float %4572, %5069, !dbg !35 + %5071 = fadd float %4573, %4574, !dbg !35 + %5072 = fadd float %4575, %5071, !dbg !35 + %5073 = fadd float %4576, %5072, !dbg !35 + %5074 = fadd float %4577, %5073, !dbg !35 + %5075 = fadd float %4578, %5074, !dbg !35 + %5076 = fadd float %4579, %5075, !dbg !35 + %5077 = fadd float %4580, %5076, !dbg !35 + %5078 = fadd float %4581, %4582, !dbg !35 + %5079 = fadd float %4583, %5078, !dbg !35 + %5080 = fadd float %4584, %5079, !dbg !35 + %5081 = fadd float %4585, %5080, !dbg !35 + %5082 = fadd float %4586, %5081, !dbg !35 + %5083 = fadd float %4587, %5082, !dbg !35 + %5084 = fadd float %4588, %5083, !dbg !35 + %5085 = fadd float %4589, %4590, !dbg !35 + %5086 = fadd float %4591, %5085, !dbg !35 + %5087 = fadd float %4592, %5086, !dbg !35 + %5088 = fadd float %4593, %5087, !dbg !35 + %5089 = fadd float %4594, %5088, !dbg !35 + %5090 = fadd float %4595, %5089, !dbg !35 + %5091 = fadd float %4596, %5090, !dbg !35 + %5092 = fadd float %4597, %4598, !dbg !35 + %5093 = fadd float %4599, %5092, !dbg !35 + %5094 = fadd float %4600, %5093, !dbg !35 + %5095 = fadd float %4601, %5094, !dbg !35 + %5096 = fadd float %4602, %5095, !dbg !35 + %5097 = fadd float %4603, %5096, !dbg !35 + %5098 = fadd float %4604, %5097, !dbg !35 + %5099 = fadd float %4605, %4606, !dbg !35 + %5100 = fadd float %4607, %5099, !dbg !35 + %5101 = fadd float %4608, %5100, !dbg !35 + %5102 = fadd float %4609, %5101, !dbg !35 + %5103 = fadd float %4610, %5102, !dbg !35 + %5104 = fadd float %4611, %5103, !dbg !35 + %5105 = fadd float %4612, %5104, !dbg !35 + %5106 = fadd float %4613, %4614, !dbg !35 + %5107 = fadd float %4615, %5106, !dbg !35 + %5108 = fadd float %4616, %5107, !dbg !35 + %5109 = fadd float %4617, %5108, !dbg !35 + %5110 = fadd float %4618, %5109, !dbg !35 + %5111 = fadd float %4619, %5110, !dbg !35 + %5112 = fadd float %4620, %5111, !dbg !35 + %5113 = fadd float %4621, %4622, !dbg !35 + %5114 = fadd float %4623, %5113, !dbg !35 + %5115 = fadd float %4624, %5114, !dbg !35 + %5116 = fadd float %4625, %5115, !dbg !35 + %5117 = fadd float %4626, %5116, !dbg !35 + %5118 = fadd float %4627, %5117, !dbg !35 + %5119 = fadd float %4628, %5118, !dbg !35 + %5120 = fadd float %4629, %4630, !dbg !35 + %5121 = fadd float %4631, %5120, !dbg !35 + %5122 = fadd float %4632, %5121, !dbg !35 + %5123 = fadd float %4633, %5122, !dbg !35 + %5124 = fadd float %4634, %5123, !dbg !35 + %5125 = fadd float %4635, %5124, !dbg !35 + %5126 = fadd float %4636, %5125, !dbg !35 + %5127 = fadd float %4637, %4638, !dbg !35 + %5128 = fadd float %4639, %5127, !dbg !35 + %5129 = fadd float %4640, %5128, !dbg !35 + %5130 = fadd float %4641, %5129, !dbg !35 + %5131 = fadd float %4642, %5130, !dbg !35 + %5132 = fadd float %4643, %5131, !dbg !35 + %5133 = fadd float %4644, %5132, !dbg !35 + %5134 = fadd float %4645, %4646, !dbg !35 + %5135 = fadd float %4647, %5134, !dbg !35 + %5136 = fadd float %4648, %5135, !dbg !35 + %5137 = fadd float %4649, %5136, !dbg !35 + %5138 = fadd float %4650, %5137, !dbg !35 + %5139 = fadd float %4651, %5138, !dbg !35 + %5140 = fadd float %4652, %5139, !dbg !35 + %5141 = fadd float %4653, %4654, !dbg !35 + %5142 = fadd float %4655, %5141, !dbg !35 + %5143 = fadd float %4656, %5142, !dbg !35 + %5144 = fadd float %4657, %5143, !dbg !35 + %5145 = fadd float %4658, %5144, !dbg !35 + %5146 = fadd float %4659, %5145, !dbg !35 + %5147 = fadd float %4660, %5146, !dbg !35 + %5148 = fadd float %4661, %4662, !dbg !35 + %5149 = fadd float %4663, %5148, !dbg !35 + %5150 = fadd float %4664, %5149, !dbg !35 + %5151 = fadd float %4665, %5150, !dbg !35 + %5152 = fadd float %4666, %5151, !dbg !35 + %5153 = fadd float %4667, %5152, !dbg !35 + %5154 = fadd float %4668, %5153, !dbg !35 + %5155 = fadd float %4669, %4670, !dbg !35 + %5156 = fadd float %4671, %5155, !dbg !35 + %5157 = fadd float %4672, %5156, !dbg !35 + %5158 = fadd float %4673, %5157, !dbg !35 + %5159 = fadd float %4674, %5158, !dbg !35 + %5160 = fadd float %4675, %5159, !dbg !35 + %5161 = fadd float %4676, %5160, !dbg !35 + %5162 = fadd float %4677, %4678, !dbg !35 + %5163 = fadd float %4679, %5162, !dbg !35 + %5164 = fadd float %4680, %5163, !dbg !35 + %5165 = fadd float %4681, %5164, !dbg !35 + %5166 = fadd float %4682, %5165, !dbg !35 + %5167 = fadd float %4683, %5166, !dbg !35 + %5168 = fadd float %4684, %5167, !dbg !35 + %5169 = fadd float %4685, %4686, !dbg !35 + %5170 = fadd float %4687, %5169, !dbg !35 + %5171 = fadd float %4688, %5170, !dbg !35 + %5172 = fadd float %4689, %5171, !dbg !35 + %5173 = fadd float %4690, %5172, !dbg !35 + %5174 = fadd float %4691, %5173, !dbg !35 + %5175 = fadd float %4692, %5174, !dbg !35 + %5176 = fadd float %4693, %4694, !dbg !35 + %5177 = fadd float %4695, %5176, !dbg !35 + %5178 = fadd float %4696, %5177, !dbg !35 + %5179 = fadd float %4697, %5178, !dbg !35 + %5180 = fadd float %4698, %5179, !dbg !35 + %5181 = fadd float %4699, %5180, !dbg !35 + %5182 = fadd float %4700, %5181, !dbg !35 + %5183 = fadd float %4701, %4702, !dbg !35 + %5184 = fadd float %4703, %5183, !dbg !35 + %5185 = fadd float %4704, %5184, !dbg !35 + %5186 = fadd float %4705, %5185, !dbg !35 + %5187 = fadd float %4706, %5186, !dbg !35 + %5188 = fadd float %4707, %5187, !dbg !35 + %5189 = fadd float %4708, %5188, !dbg !35 + %5190 = fadd float %4709, %4710, !dbg !35 + %5191 = fadd float %4711, %5190, !dbg !35 + %5192 = fadd float %4712, %5191, !dbg !35 + %5193 = fadd float %4713, %5192, !dbg !35 + %5194 = fadd float %4714, %5193, !dbg !35 + %5195 = fadd float %4715, %5194, !dbg !35 + %5196 = fadd float %4716, %5195, !dbg !35 + %5197 = fadd float %4717, %4718, !dbg !35 + %5198 = fadd float %4719, %5197, !dbg !35 + %5199 = fadd float %4720, %5198, !dbg !35 + %5200 = fadd float %4721, %5199, !dbg !35 + %5201 = fadd float %4722, %5200, !dbg !35 + %5202 = fadd float %4723, %5201, !dbg !35 + %5203 = fadd float %4724, %5202, !dbg !35 + %5204 = fadd float %4725, %4726, !dbg !35 + %5205 = fadd float %4727, %5204, !dbg !35 + %5206 = fadd float %4728, %5205, !dbg !35 + %5207 = fadd float %4729, %5206, !dbg !35 + %5208 = fadd float %4730, %5207, !dbg !35 + %5209 = fadd float %4731, %5208, !dbg !35 + %5210 = fadd float %4732, %5209, !dbg !35 + %5211 = fadd float %4733, %4734, !dbg !35 + %5212 = fadd float %4735, %5211, !dbg !35 + %5213 = fadd float %4736, %5212, !dbg !35 + %5214 = fadd float %4737, %5213, !dbg !35 + %5215 = fadd float %4738, %5214, !dbg !35 + %5216 = fadd float %4739, %5215, !dbg !35 + %5217 = fadd float %4740, %5216, !dbg !35 + %5218 = fadd float %4741, %4742, !dbg !35 + %5219 = fadd float %4743, %5218, !dbg !35 + %5220 = fadd float %4744, %5219, !dbg !35 + %5221 = fadd float %4745, %5220, !dbg !35 + %5222 = fadd float %4746, %5221, !dbg !35 + %5223 = fadd float %4747, %5222, !dbg !35 + %5224 = fadd float %4748, %5223, !dbg !35 + %5225 = fadd float %4749, %4750, !dbg !35 + %5226 = fadd float %4751, %5225, !dbg !35 + %5227 = fadd float %4752, %5226, !dbg !35 + %5228 = fadd float %4753, %5227, !dbg !35 + %5229 = fadd float %4754, %5228, !dbg !35 + %5230 = fadd float %4755, %5229, !dbg !35 + %5231 = fadd float %4756, %5230, !dbg !35 + %5232 = fadd float %4757, %4758, !dbg !35 + %5233 = fadd float %4759, %5232, !dbg !35 + %5234 = fadd float %4760, %5233, !dbg !35 + %5235 = fadd float %4761, %5234, !dbg !35 + %5236 = fadd float %4762, %5235, !dbg !35 + %5237 = fadd float %4763, %5236, !dbg !35 + %5238 = fadd float %4764, %5237, !dbg !35 + %5239 = fadd float %4765, %4766, !dbg !35 + %5240 = fadd float %4767, %5239, !dbg !35 + %5241 = fadd float %4768, %5240, !dbg !35 + %5242 = fadd float %4769, %5241, !dbg !35 + %5243 = fadd float %4770, %5242, !dbg !35 + %5244 = fadd float %4771, %5243, !dbg !35 + %5245 = fadd float %4772, %5244, !dbg !35 + %5246 = fadd float %4773, %4774, !dbg !35 + %5247 = fadd float %4775, %5246, !dbg !35 + %5248 = fadd float %4776, %5247, !dbg !35 + %5249 = fadd float %4777, %5248, !dbg !35 + %5250 = fadd float %4778, %5249, !dbg !35 + %5251 = fadd float %4779, %5250, !dbg !35 + %5252 = fadd float %4780, %5251, !dbg !35 + %5253 = fadd float %4781, %4782, !dbg !35 + %5254 = fadd float %4783, %5253, !dbg !35 + %5255 = fadd float %4784, %5254, !dbg !35 + %5256 = fadd float %4785, %5255, !dbg !35 + %5257 = fadd float %4786, %5256, !dbg !35 + %5258 = fadd float %4787, %5257, !dbg !35 + %5259 = fadd float %4788, %5258, !dbg !35 + %5260 = fadd float %4789, %4790, !dbg !35 + %5261 = fadd float %4791, %5260, !dbg !35 + %5262 = fadd float %4792, %5261, !dbg !35 + %5263 = fadd float %4793, %5262, !dbg !35 + %5264 = fadd float %4794, %5263, !dbg !35 + %5265 = fadd float %4795, %5264, !dbg !35 + %5266 = fadd float %4796, %5265, !dbg !35 + %5267 = fadd float %4797, %4798, !dbg !35 + %5268 = fadd float %4799, %5267, !dbg !35 + %5269 = fadd float %4800, %5268, !dbg !35 + %5270 = fadd float %4801, %5269, !dbg !35 + %5271 = fadd float %4802, %5270, !dbg !35 + %5272 = fadd float %4803, %5271, !dbg !35 + %5273 = fadd float %4804, %5272, !dbg !35 + %5274 = fadd float %4805, %4806, !dbg !35 + %5275 = fadd float %4807, %5274, !dbg !35 + %5276 = fadd float %4808, %5275, !dbg !35 + %5277 = fadd float %4809, %5276, !dbg !35 + %5278 = fadd float %4810, %5277, !dbg !35 + %5279 = fadd float %4811, %5278, !dbg !35 + %5280 = fadd float %4812, %5279, !dbg !35 + %5281 = fadd float %4813, %4814, !dbg !35 + %5282 = fadd float %4815, %5281, !dbg !35 + %5283 = fadd float %4816, %5282, !dbg !35 + %5284 = fadd float %4817, %5283, !dbg !35 + %5285 = fadd float %4818, %5284, !dbg !35 + %5286 = fadd float %4819, %5285, !dbg !35 + %5287 = fadd float %4820, %5286, !dbg !35 + %5288 = fadd float %4821, %4822, !dbg !35 + %5289 = fadd float %4823, %5288, !dbg !35 + %5290 = fadd float %4824, %5289, !dbg !35 + %5291 = fadd float %4825, %5290, !dbg !35 + %5292 = fadd float %4826, %5291, !dbg !35 + %5293 = fadd float %4827, %5292, !dbg !35 + %5294 = fadd float %4828, %5293, !dbg !35 + %5295 = fadd float %4829, %4830, !dbg !35 + %5296 = fadd float %4831, %5295, !dbg !35 + %5297 = fadd float %4832, %5296, !dbg !35 + %5298 = fadd float %4833, %5297, !dbg !35 + %5299 = fadd float %4834, %5298, !dbg !35 + %5300 = fadd float %4835, %5299, !dbg !35 + %5301 = fadd float %4836, %5300, !dbg !35 + %5302 = fadd float %4837, %4838, !dbg !35 + %5303 = fadd float %4839, %5302, !dbg !35 + %5304 = fadd float %4840, %5303, !dbg !35 + %5305 = fadd float %4841, %5304, !dbg !35 + %5306 = fadd float %4842, %5305, !dbg !35 + %5307 = fadd float %4843, %5306, !dbg !35 + %5308 = fadd float %4844, %5307, !dbg !35 + %5309 = fadd float %4845, %4846, !dbg !35 + %5310 = fadd float %4847, %5309, !dbg !35 + %5311 = fadd float %4848, %5310, !dbg !35 + %5312 = fadd float %4849, %5311, !dbg !35 + %5313 = fadd float %4850, %5312, !dbg !35 + %5314 = fadd float %4851, %5313, !dbg !35 + %5315 = fadd float %4852, %5314, !dbg !35 + %5316 = fadd float %4853, %4854, !dbg !35 + %5317 = fadd float %4855, %5316, !dbg !35 + %5318 = fadd float %4856, %5317, !dbg !35 + %5319 = fadd float %4857, %5318, !dbg !35 + %5320 = fadd float %4858, %5319, !dbg !35 + %5321 = fadd float %4859, %5320, !dbg !35 + %5322 = fadd float %4860, %5321, !dbg !35 + %5323 = fadd float %4861, %4862, !dbg !35 + %5324 = fadd float %4863, %5323, !dbg !35 + %5325 = fadd float %4864, %5324, !dbg !35 + %5326 = fadd float %4865, %5325, !dbg !35 + %5327 = fadd float %4866, %5326, !dbg !35 + %5328 = fadd float %4867, %5327, !dbg !35 + %5329 = fadd float %4868, %5328, !dbg !35 + %5330 = fadd float %4869, %4870, !dbg !35 + %5331 = fadd float %4871, %5330, !dbg !35 + %5332 = fadd float %4872, %5331, !dbg !35 + %5333 = fadd float %4873, %5332, !dbg !35 + %5334 = fadd float %4874, %5333, !dbg !35 + %5335 = fadd float %4875, %5334, !dbg !35 + %5336 = fadd float %4876, %5335, !dbg !35 + %5337 = fadd float %4877, %4878, !dbg !35 + %5338 = fadd float %4879, %5337, !dbg !35 + %5339 = fadd float %4880, %5338, !dbg !35 + %5340 = fadd float %4881, %5339, !dbg !35 + %5341 = fadd float %4882, %5340, !dbg !35 + %5342 = fadd float %4883, %5341, !dbg !35 + %5343 = fadd float %4884, %5342, !dbg !35 + %5344 = fadd float %4885, %4886, !dbg !35 + %5345 = fadd float %4887, %5344, !dbg !35 + %5346 = fadd float %4888, %5345, !dbg !35 + %5347 = fadd float %4889, %5346, !dbg !35 + %5348 = fadd float %4890, %5347, !dbg !35 + %5349 = fadd float %4891, %5348, !dbg !35 + %5350 = fadd float %4892, %5349, !dbg !35 + %5351 = fadd float %4893, %4894, !dbg !35 + %5352 = fadd float %4895, %5351, !dbg !35 + %5353 = fadd float %4896, %5352, !dbg !35 + %5354 = fadd float %4897, %5353, !dbg !35 + %5355 = fadd float %4898, %5354, !dbg !35 + %5356 = fadd float %4899, %5355, !dbg !35 + %5357 = fadd float %4900, %5356, !dbg !35 + %5358 = fadd float %4901, %4902, !dbg !35 + %5359 = fadd float %4903, %5358, !dbg !35 + %5360 = fadd float %4904, %5359, !dbg !35 + %5361 = fadd float %4905, %5360, !dbg !35 + %5362 = fadd float %4906, %5361, !dbg !35 + %5363 = fadd float %4907, %5362, !dbg !35 + %5364 = fadd float %4908, %5363, !dbg !35 + %5365 = fadd float %4909, %4910, !dbg !35 + %5366 = fadd float %4911, %5365, !dbg !35 + %5367 = fadd float %4912, %5366, !dbg !35 + %5368 = fadd float %4913, %5367, !dbg !35 + %5369 = fadd float %4914, %5368, !dbg !35 + %5370 = fadd float %4915, %5369, !dbg !35 + %5371 = fadd float %4916, %5370, !dbg !35 + %5372 = fadd float %4917, %4918, !dbg !35 + %5373 = fadd float %4919, %5372, !dbg !35 + %5374 = fadd float %4920, %5373, !dbg !35 + %5375 = fadd float %4921, %5374, !dbg !35 + %5376 = fadd float %4922, %5375, !dbg !35 + %5377 = fadd float %4923, %5376, !dbg !35 + %5378 = fadd float %4924, %5377, !dbg !35 + %5379 = fadd float %4925, %4926, !dbg !35 + %5380 = fadd float %4927, %5379, !dbg !35 + %5381 = fadd float %4928, %5380, !dbg !35 + %5382 = fadd float %4929, %5381, !dbg !35 + %5383 = fadd float %4930, %5382, !dbg !35 + %5384 = fadd float %4931, %5383, !dbg !35 + %5385 = fadd float %4932, %5384, !dbg !35 + %5386 = fadd float %4933, %4934, !dbg !35 + %5387 = fadd float %4935, %5386, !dbg !35 + %5388 = fadd float %4936, %5387, !dbg !35 + %5389 = fadd float %4937, %5388, !dbg !35 + %5390 = fadd float %4938, %5389, !dbg !35 + %5391 = fadd float %4939, %5390, !dbg !35 + %5392 = fadd float %4940, %5391, !dbg !35 + %5393 = fadd float %4941, %4942, !dbg !35 + %5394 = fadd float %4943, %5393, !dbg !35 + %5395 = fadd float %4944, %5394, !dbg !35 + %5396 = fadd float %4945, %5395, !dbg !35 + %5397 = fadd float %4946, %5396, !dbg !35 + %5398 = fadd float %4947, %5397, !dbg !35 + %5399 = fadd float %4948, %5398, !dbg !35 + %5400 = fadd float %4949, %4950, !dbg !35 + %5401 = fadd float %4951, %5400, !dbg !35 + %5402 = fadd float %4952, %5401, !dbg !35 + %5403 = fadd float %4953, %5402, !dbg !35 + %5404 = fadd float %4954, %5403, !dbg !35 + %5405 = fadd float %4955, %5404, !dbg !35 + %5406 = fadd float %4956, %5405, !dbg !35 + %5407 = fadd float %4957, %4958, !dbg !35 + %5408 = fadd float %4959, %5407, !dbg !35 + %5409 = fadd float %4960, %5408, !dbg !35 + %5410 = fadd float %4961, %5409, !dbg !35 + %5411 = fadd float %4962, %5410, !dbg !35 + %5412 = fadd float %4963, %5411, !dbg !35 + %5413 = fadd float %4964, %5412, !dbg !35 + %5414 = fadd float %4965, %4966, !dbg !35 + %5415 = fadd float %4967, %5414, !dbg !35 + %5416 = fadd float %4968, %5415, !dbg !35 + %5417 = fadd float %4969, %5416, !dbg !35 + %5418 = fadd float %4970, %5417, !dbg !35 + %5419 = fadd float %4971, %5418, !dbg !35 + %5420 = fadd float %4972, %5419, !dbg !35 + %5421 = bitcast float %4979 to i32, !dbg !38 + %5422 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5421, i32 8, i32 31), !dbg !38 + %5423 = bitcast i32 %5422 to float, !dbg !38 + %5424 = fadd float %4979, %5423, !dbg !35 + %5425 = bitcast float %5424 to i32, !dbg !38 + %5426 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5425, i32 4, i32 31), !dbg !38 + %5427 = bitcast i32 %5426 to float, !dbg !38 + %5428 = fadd float %5424, %5427, !dbg !35 + %5429 = bitcast float %5428 to i32, !dbg !38 + %5430 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5429, i32 2, i32 31), !dbg !38 + %5431 = bitcast i32 %5430 to float, !dbg !38 + %5432 = fadd float %5428, %5431, !dbg !35 + %5433 = bitcast float %5432 to i32, !dbg !38 + %5434 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5433, i32 1, i32 31), !dbg !38 + %5435 = bitcast i32 %5434 to float, !dbg !38 + %5436 = fadd float %5432, %5435, !dbg !35 + %5437 = bitcast float %4986 to i32, !dbg !38 + %5438 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5437, i32 8, i32 31), !dbg !38 + %5439 = bitcast i32 %5438 to float, !dbg !38 + %5440 = fadd float %4986, %5439, !dbg !35 + %5441 = bitcast float %5440 to i32, !dbg !38 + %5442 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5441, i32 4, i32 31), !dbg !38 + %5443 = bitcast i32 %5442 to float, !dbg !38 + %5444 = fadd float %5440, %5443, !dbg !35 + %5445 = bitcast float %5444 to i32, !dbg !38 + %5446 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5445, i32 2, i32 31), !dbg !38 + %5447 = bitcast i32 %5446 to float, !dbg !38 + %5448 = fadd float %5444, %5447, !dbg !35 + %5449 = bitcast float %5448 to i32, !dbg !38 + %5450 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5449, i32 1, i32 31), !dbg !38 + %5451 = bitcast i32 %5450 to float, !dbg !38 + %5452 = fadd float %5448, %5451, !dbg !35 + %5453 = bitcast float %4993 to i32, !dbg !38 + %5454 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5453, i32 8, i32 31), !dbg !38 + %5455 = bitcast i32 %5454 to float, !dbg !38 + %5456 = fadd float %4993, %5455, !dbg !35 + %5457 = bitcast float %5456 to i32, !dbg !38 + %5458 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5457, i32 4, i32 31), !dbg !38 + %5459 = bitcast i32 %5458 to float, !dbg !38 + %5460 = fadd float %5456, %5459, !dbg !35 + %5461 = bitcast float %5460 to i32, !dbg !38 + %5462 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5461, i32 2, i32 31), !dbg !38 + %5463 = bitcast i32 %5462 to float, !dbg !38 + %5464 = fadd float %5460, %5463, !dbg !35 + %5465 = bitcast float %5464 to i32, !dbg !38 + %5466 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5465, i32 1, i32 31), !dbg !38 + %5467 = bitcast i32 %5466 to float, !dbg !38 + %5468 = fadd float %5464, %5467, !dbg !35 + %5469 = bitcast float %5000 to i32, !dbg !38 + %5470 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5469, i32 8, i32 31), !dbg !38 + %5471 = bitcast i32 %5470 to float, !dbg !38 + %5472 = fadd float %5000, %5471, !dbg !35 + %5473 = bitcast float %5472 to i32, !dbg !38 + %5474 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5473, i32 4, i32 31), !dbg !38 + %5475 = bitcast i32 %5474 to float, !dbg !38 + %5476 = fadd float %5472, %5475, !dbg !35 + %5477 = bitcast float %5476 to i32, !dbg !38 + %5478 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5477, i32 2, i32 31), !dbg !38 + %5479 = bitcast i32 %5478 to float, !dbg !38 + %5480 = fadd float %5476, %5479, !dbg !35 + %5481 = bitcast float %5480 to i32, !dbg !38 + %5482 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5481, i32 1, i32 31), !dbg !38 + %5483 = bitcast i32 %5482 to float, !dbg !38 + %5484 = fadd float %5480, %5483, !dbg !35 + %5485 = bitcast float %5007 to i32, !dbg !38 + %5486 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5485, i32 8, i32 31), !dbg !38 + %5487 = bitcast i32 %5486 to float, !dbg !38 + %5488 = fadd float %5007, %5487, !dbg !35 + %5489 = bitcast float %5488 to i32, !dbg !38 + %5490 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5489, i32 4, i32 31), !dbg !38 + %5491 = bitcast i32 %5490 to float, !dbg !38 + %5492 = fadd float %5488, %5491, !dbg !35 + %5493 = bitcast float %5492 to i32, !dbg !38 + %5494 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5493, i32 2, i32 31), !dbg !38 + %5495 = bitcast i32 %5494 to float, !dbg !38 + %5496 = fadd float %5492, %5495, !dbg !35 + %5497 = bitcast float %5496 to i32, !dbg !38 + %5498 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5497, i32 1, i32 31), !dbg !38 + %5499 = bitcast i32 %5498 to float, !dbg !38 + %5500 = fadd float %5496, %5499, !dbg !35 + %5501 = bitcast float %5014 to i32, !dbg !38 + %5502 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5501, i32 8, i32 31), !dbg !38 + %5503 = bitcast i32 %5502 to float, !dbg !38 + %5504 = fadd float %5014, %5503, !dbg !35 + %5505 = bitcast float %5504 to i32, !dbg !38 + %5506 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5505, i32 4, i32 31), !dbg !38 + %5507 = bitcast i32 %5506 to float, !dbg !38 + %5508 = fadd float %5504, %5507, !dbg !35 + %5509 = bitcast float %5508 to i32, !dbg !38 + %5510 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5509, i32 2, i32 31), !dbg !38 + %5511 = bitcast i32 %5510 to float, !dbg !38 + %5512 = fadd float %5508, %5511, !dbg !35 + %5513 = bitcast float %5512 to i32, !dbg !38 + %5514 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5513, i32 1, i32 31), !dbg !38 + %5515 = bitcast i32 %5514 to float, !dbg !38 + %5516 = fadd float %5512, %5515, !dbg !35 + %5517 = bitcast float %5021 to i32, !dbg !38 + %5518 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5517, i32 8, i32 31), !dbg !38 + %5519 = bitcast i32 %5518 to float, !dbg !38 + %5520 = fadd float %5021, %5519, !dbg !35 + %5521 = bitcast float %5520 to i32, !dbg !38 + %5522 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5521, i32 4, i32 31), !dbg !38 + %5523 = bitcast i32 %5522 to float, !dbg !38 + %5524 = fadd float %5520, %5523, !dbg !35 + %5525 = bitcast float %5524 to i32, !dbg !38 + %5526 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5525, i32 2, i32 31), !dbg !38 + %5527 = bitcast i32 %5526 to float, !dbg !38 + %5528 = fadd float %5524, %5527, !dbg !35 + %5529 = bitcast float %5528 to i32, !dbg !38 + %5530 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5529, i32 1, i32 31), !dbg !38 + %5531 = bitcast i32 %5530 to float, !dbg !38 + %5532 = fadd float %5528, %5531, !dbg !35 + %5533 = bitcast float %5028 to i32, !dbg !38 + %5534 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5533, i32 8, i32 31), !dbg !38 + %5535 = bitcast i32 %5534 to float, !dbg !38 + %5536 = fadd float %5028, %5535, !dbg !35 + %5537 = bitcast float %5536 to i32, !dbg !38 + %5538 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5537, i32 4, i32 31), !dbg !38 + %5539 = bitcast i32 %5538 to float, !dbg !38 + %5540 = fadd float %5536, %5539, !dbg !35 + %5541 = bitcast float %5540 to i32, !dbg !38 + %5542 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5541, i32 2, i32 31), !dbg !38 + %5543 = bitcast i32 %5542 to float, !dbg !38 + %5544 = fadd float %5540, %5543, !dbg !35 + %5545 = bitcast float %5544 to i32, !dbg !38 + %5546 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5545, i32 1, i32 31), !dbg !38 + %5547 = bitcast i32 %5546 to float, !dbg !38 + %5548 = fadd float %5544, %5547, !dbg !35 + %5549 = bitcast float %5035 to i32, !dbg !38 + %5550 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5549, i32 8, i32 31), !dbg !38 + %5551 = bitcast i32 %5550 to float, !dbg !38 + %5552 = fadd float %5035, %5551, !dbg !35 + %5553 = bitcast float %5552 to i32, !dbg !38 + %5554 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5553, i32 4, i32 31), !dbg !38 + %5555 = bitcast i32 %5554 to float, !dbg !38 + %5556 = fadd float %5552, %5555, !dbg !35 + %5557 = bitcast float %5556 to i32, !dbg !38 + %5558 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5557, i32 2, i32 31), !dbg !38 + %5559 = bitcast i32 %5558 to float, !dbg !38 + %5560 = fadd float %5556, %5559, !dbg !35 + %5561 = bitcast float %5560 to i32, !dbg !38 + %5562 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5561, i32 1, i32 31), !dbg !38 + %5563 = bitcast i32 %5562 to float, !dbg !38 + %5564 = fadd float %5560, %5563, !dbg !35 + %5565 = bitcast float %5042 to i32, !dbg !38 + %5566 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5565, i32 8, i32 31), !dbg !38 + %5567 = bitcast i32 %5566 to float, !dbg !38 + %5568 = fadd float %5042, %5567, !dbg !35 + %5569 = bitcast float %5568 to i32, !dbg !38 + %5570 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5569, i32 4, i32 31), !dbg !38 + %5571 = bitcast i32 %5570 to float, !dbg !38 + %5572 = fadd float %5568, %5571, !dbg !35 + %5573 = bitcast float %5572 to i32, !dbg !38 + %5574 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5573, i32 2, i32 31), !dbg !38 + %5575 = bitcast i32 %5574 to float, !dbg !38 + %5576 = fadd float %5572, %5575, !dbg !35 + %5577 = bitcast float %5576 to i32, !dbg !38 + %5578 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5577, i32 1, i32 31), !dbg !38 + %5579 = bitcast i32 %5578 to float, !dbg !38 + %5580 = fadd float %5576, %5579, !dbg !35 + %5581 = bitcast float %5049 to i32, !dbg !38 + %5582 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5581, i32 8, i32 31), !dbg !38 + %5583 = bitcast i32 %5582 to float, !dbg !38 + %5584 = fadd float %5049, %5583, !dbg !35 + %5585 = bitcast float %5584 to i32, !dbg !38 + %5586 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5585, i32 4, i32 31), !dbg !38 + %5587 = bitcast i32 %5586 to float, !dbg !38 + %5588 = fadd float %5584, %5587, !dbg !35 + %5589 = bitcast float %5588 to i32, !dbg !38 + %5590 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5589, i32 2, i32 31), !dbg !38 + %5591 = bitcast i32 %5590 to float, !dbg !38 + %5592 = fadd float %5588, %5591, !dbg !35 + %5593 = bitcast float %5592 to i32, !dbg !38 + %5594 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5593, i32 1, i32 31), !dbg !38 + %5595 = bitcast i32 %5594 to float, !dbg !38 + %5596 = fadd float %5592, %5595, !dbg !35 + %5597 = bitcast float %5056 to i32, !dbg !38 + %5598 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5597, i32 8, i32 31), !dbg !38 + %5599 = bitcast i32 %5598 to float, !dbg !38 + %5600 = fadd float %5056, %5599, !dbg !35 + %5601 = bitcast float %5600 to i32, !dbg !38 + %5602 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5601, i32 4, i32 31), !dbg !38 + %5603 = bitcast i32 %5602 to float, !dbg !38 + %5604 = fadd float %5600, %5603, !dbg !35 + %5605 = bitcast float %5604 to i32, !dbg !38 + %5606 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5605, i32 2, i32 31), !dbg !38 + %5607 = bitcast i32 %5606 to float, !dbg !38 + %5608 = fadd float %5604, %5607, !dbg !35 + %5609 = bitcast float %5608 to i32, !dbg !38 + %5610 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5609, i32 1, i32 31), !dbg !38 + %5611 = bitcast i32 %5610 to float, !dbg !38 + %5612 = fadd float %5608, %5611, !dbg !35 + %5613 = bitcast float %5063 to i32, !dbg !38 + %5614 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5613, i32 8, i32 31), !dbg !38 + %5615 = bitcast i32 %5614 to float, !dbg !38 + %5616 = fadd float %5063, %5615, !dbg !35 + %5617 = bitcast float %5616 to i32, !dbg !38 + %5618 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5617, i32 4, i32 31), !dbg !38 + %5619 = bitcast i32 %5618 to float, !dbg !38 + %5620 = fadd float %5616, %5619, !dbg !35 + %5621 = bitcast float %5620 to i32, !dbg !38 + %5622 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5621, i32 2, i32 31), !dbg !38 + %5623 = bitcast i32 %5622 to float, !dbg !38 + %5624 = fadd float %5620, %5623, !dbg !35 + %5625 = bitcast float %5624 to i32, !dbg !38 + %5626 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5625, i32 1, i32 31), !dbg !38 + %5627 = bitcast i32 %5626 to float, !dbg !38 + %5628 = fadd float %5624, %5627, !dbg !35 + %5629 = bitcast float %5070 to i32, !dbg !38 + %5630 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5629, i32 8, i32 31), !dbg !38 + %5631 = bitcast i32 %5630 to float, !dbg !38 + %5632 = fadd float %5070, %5631, !dbg !35 + %5633 = bitcast float %5632 to i32, !dbg !38 + %5634 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5633, i32 4, i32 31), !dbg !38 + %5635 = bitcast i32 %5634 to float, !dbg !38 + %5636 = fadd float %5632, %5635, !dbg !35 + %5637 = bitcast float %5636 to i32, !dbg !38 + %5638 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5637, i32 2, i32 31), !dbg !38 + %5639 = bitcast i32 %5638 to float, !dbg !38 + %5640 = fadd float %5636, %5639, !dbg !35 + %5641 = bitcast float %5640 to i32, !dbg !38 + %5642 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5641, i32 1, i32 31), !dbg !38 + %5643 = bitcast i32 %5642 to float, !dbg !38 + %5644 = fadd float %5640, %5643, !dbg !35 + %5645 = bitcast float %5077 to i32, !dbg !38 + %5646 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5645, i32 8, i32 31), !dbg !38 + %5647 = bitcast i32 %5646 to float, !dbg !38 + %5648 = fadd float %5077, %5647, !dbg !35 + %5649 = bitcast float %5648 to i32, !dbg !38 + %5650 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5649, i32 4, i32 31), !dbg !38 + %5651 = bitcast i32 %5650 to float, !dbg !38 + %5652 = fadd float %5648, %5651, !dbg !35 + %5653 = bitcast float %5652 to i32, !dbg !38 + %5654 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5653, i32 2, i32 31), !dbg !38 + %5655 = bitcast i32 %5654 to float, !dbg !38 + %5656 = fadd float %5652, %5655, !dbg !35 + %5657 = bitcast float %5656 to i32, !dbg !38 + %5658 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5657, i32 1, i32 31), !dbg !38 + %5659 = bitcast i32 %5658 to float, !dbg !38 + %5660 = fadd float %5656, %5659, !dbg !35 + %5661 = bitcast float %5084 to i32, !dbg !38 + %5662 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5661, i32 8, i32 31), !dbg !38 + %5663 = bitcast i32 %5662 to float, !dbg !38 + %5664 = fadd float %5084, %5663, !dbg !35 + %5665 = bitcast float %5664 to i32, !dbg !38 + %5666 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5665, i32 4, i32 31), !dbg !38 + %5667 = bitcast i32 %5666 to float, !dbg !38 + %5668 = fadd float %5664, %5667, !dbg !35 + %5669 = bitcast float %5668 to i32, !dbg !38 + %5670 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5669, i32 2, i32 31), !dbg !38 + %5671 = bitcast i32 %5670 to float, !dbg !38 + %5672 = fadd float %5668, %5671, !dbg !35 + %5673 = bitcast float %5672 to i32, !dbg !38 + %5674 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5673, i32 1, i32 31), !dbg !38 + %5675 = bitcast i32 %5674 to float, !dbg !38 + %5676 = fadd float %5672, %5675, !dbg !35 + %5677 = bitcast float %5091 to i32, !dbg !38 + %5678 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5677, i32 8, i32 31), !dbg !38 + %5679 = bitcast i32 %5678 to float, !dbg !38 + %5680 = fadd float %5091, %5679, !dbg !35 + %5681 = bitcast float %5680 to i32, !dbg !38 + %5682 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5681, i32 4, i32 31), !dbg !38 + %5683 = bitcast i32 %5682 to float, !dbg !38 + %5684 = fadd float %5680, %5683, !dbg !35 + %5685 = bitcast float %5684 to i32, !dbg !38 + %5686 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5685, i32 2, i32 31), !dbg !38 + %5687 = bitcast i32 %5686 to float, !dbg !38 + %5688 = fadd float %5684, %5687, !dbg !35 + %5689 = bitcast float %5688 to i32, !dbg !38 + %5690 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5689, i32 1, i32 31), !dbg !38 + %5691 = bitcast i32 %5690 to float, !dbg !38 + %5692 = fadd float %5688, %5691, !dbg !35 + %5693 = bitcast float %5098 to i32, !dbg !38 + %5694 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5693, i32 8, i32 31), !dbg !38 + %5695 = bitcast i32 %5694 to float, !dbg !38 + %5696 = fadd float %5098, %5695, !dbg !35 + %5697 = bitcast float %5696 to i32, !dbg !38 + %5698 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5697, i32 4, i32 31), !dbg !38 + %5699 = bitcast i32 %5698 to float, !dbg !38 + %5700 = fadd float %5696, %5699, !dbg !35 + %5701 = bitcast float %5700 to i32, !dbg !38 + %5702 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5701, i32 2, i32 31), !dbg !38 + %5703 = bitcast i32 %5702 to float, !dbg !38 + %5704 = fadd float %5700, %5703, !dbg !35 + %5705 = bitcast float %5704 to i32, !dbg !38 + %5706 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5705, i32 1, i32 31), !dbg !38 + %5707 = bitcast i32 %5706 to float, !dbg !38 + %5708 = fadd float %5704, %5707, !dbg !35 + %5709 = bitcast float %5105 to i32, !dbg !38 + %5710 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5709, i32 8, i32 31), !dbg !38 + %5711 = bitcast i32 %5710 to float, !dbg !38 + %5712 = fadd float %5105, %5711, !dbg !35 + %5713 = bitcast float %5712 to i32, !dbg !38 + %5714 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5713, i32 4, i32 31), !dbg !38 + %5715 = bitcast i32 %5714 to float, !dbg !38 + %5716 = fadd float %5712, %5715, !dbg !35 + %5717 = bitcast float %5716 to i32, !dbg !38 + %5718 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5717, i32 2, i32 31), !dbg !38 + %5719 = bitcast i32 %5718 to float, !dbg !38 + %5720 = fadd float %5716, %5719, !dbg !35 + %5721 = bitcast float %5720 to i32, !dbg !38 + %5722 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5721, i32 1, i32 31), !dbg !38 + %5723 = bitcast i32 %5722 to float, !dbg !38 + %5724 = fadd float %5720, %5723, !dbg !35 + %5725 = bitcast float %5112 to i32, !dbg !38 + %5726 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5725, i32 8, i32 31), !dbg !38 + %5727 = bitcast i32 %5726 to float, !dbg !38 + %5728 = fadd float %5112, %5727, !dbg !35 + %5729 = bitcast float %5728 to i32, !dbg !38 + %5730 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5729, i32 4, i32 31), !dbg !38 + %5731 = bitcast i32 %5730 to float, !dbg !38 + %5732 = fadd float %5728, %5731, !dbg !35 + %5733 = bitcast float %5732 to i32, !dbg !38 + %5734 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5733, i32 2, i32 31), !dbg !38 + %5735 = bitcast i32 %5734 to float, !dbg !38 + %5736 = fadd float %5732, %5735, !dbg !35 + %5737 = bitcast float %5736 to i32, !dbg !38 + %5738 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5737, i32 1, i32 31), !dbg !38 + %5739 = bitcast i32 %5738 to float, !dbg !38 + %5740 = fadd float %5736, %5739, !dbg !35 + %5741 = bitcast float %5119 to i32, !dbg !38 + %5742 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5741, i32 8, i32 31), !dbg !38 + %5743 = bitcast i32 %5742 to float, !dbg !38 + %5744 = fadd float %5119, %5743, !dbg !35 + %5745 = bitcast float %5744 to i32, !dbg !38 + %5746 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5745, i32 4, i32 31), !dbg !38 + %5747 = bitcast i32 %5746 to float, !dbg !38 + %5748 = fadd float %5744, %5747, !dbg !35 + %5749 = bitcast float %5748 to i32, !dbg !38 + %5750 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5749, i32 2, i32 31), !dbg !38 + %5751 = bitcast i32 %5750 to float, !dbg !38 + %5752 = fadd float %5748, %5751, !dbg !35 + %5753 = bitcast float %5752 to i32, !dbg !38 + %5754 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5753, i32 1, i32 31), !dbg !38 + %5755 = bitcast i32 %5754 to float, !dbg !38 + %5756 = fadd float %5752, %5755, !dbg !35 + %5757 = bitcast float %5126 to i32, !dbg !38 + %5758 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5757, i32 8, i32 31), !dbg !38 + %5759 = bitcast i32 %5758 to float, !dbg !38 + %5760 = fadd float %5126, %5759, !dbg !35 + %5761 = bitcast float %5760 to i32, !dbg !38 + %5762 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5761, i32 4, i32 31), !dbg !38 + %5763 = bitcast i32 %5762 to float, !dbg !38 + %5764 = fadd float %5760, %5763, !dbg !35 + %5765 = bitcast float %5764 to i32, !dbg !38 + %5766 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5765, i32 2, i32 31), !dbg !38 + %5767 = bitcast i32 %5766 to float, !dbg !38 + %5768 = fadd float %5764, %5767, !dbg !35 + %5769 = bitcast float %5768 to i32, !dbg !38 + %5770 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5769, i32 1, i32 31), !dbg !38 + %5771 = bitcast i32 %5770 to float, !dbg !38 + %5772 = fadd float %5768, %5771, !dbg !35 + %5773 = bitcast float %5133 to i32, !dbg !38 + %5774 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5773, i32 8, i32 31), !dbg !38 + %5775 = bitcast i32 %5774 to float, !dbg !38 + %5776 = fadd float %5133, %5775, !dbg !35 + %5777 = bitcast float %5776 to i32, !dbg !38 + %5778 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5777, i32 4, i32 31), !dbg !38 + %5779 = bitcast i32 %5778 to float, !dbg !38 + %5780 = fadd float %5776, %5779, !dbg !35 + %5781 = bitcast float %5780 to i32, !dbg !38 + %5782 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5781, i32 2, i32 31), !dbg !38 + %5783 = bitcast i32 %5782 to float, !dbg !38 + %5784 = fadd float %5780, %5783, !dbg !35 + %5785 = bitcast float %5784 to i32, !dbg !38 + %5786 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5785, i32 1, i32 31), !dbg !38 + %5787 = bitcast i32 %5786 to float, !dbg !38 + %5788 = fadd float %5784, %5787, !dbg !35 + %5789 = bitcast float %5140 to i32, !dbg !38 + %5790 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5789, i32 8, i32 31), !dbg !38 + %5791 = bitcast i32 %5790 to float, !dbg !38 + %5792 = fadd float %5140, %5791, !dbg !35 + %5793 = bitcast float %5792 to i32, !dbg !38 + %5794 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5793, i32 4, i32 31), !dbg !38 + %5795 = bitcast i32 %5794 to float, !dbg !38 + %5796 = fadd float %5792, %5795, !dbg !35 + %5797 = bitcast float %5796 to i32, !dbg !38 + %5798 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5797, i32 2, i32 31), !dbg !38 + %5799 = bitcast i32 %5798 to float, !dbg !38 + %5800 = fadd float %5796, %5799, !dbg !35 + %5801 = bitcast float %5800 to i32, !dbg !38 + %5802 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5801, i32 1, i32 31), !dbg !38 + %5803 = bitcast i32 %5802 to float, !dbg !38 + %5804 = fadd float %5800, %5803, !dbg !35 + %5805 = bitcast float %5147 to i32, !dbg !38 + %5806 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5805, i32 8, i32 31), !dbg !38 + %5807 = bitcast i32 %5806 to float, !dbg !38 + %5808 = fadd float %5147, %5807, !dbg !35 + %5809 = bitcast float %5808 to i32, !dbg !38 + %5810 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5809, i32 4, i32 31), !dbg !38 + %5811 = bitcast i32 %5810 to float, !dbg !38 + %5812 = fadd float %5808, %5811, !dbg !35 + %5813 = bitcast float %5812 to i32, !dbg !38 + %5814 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5813, i32 2, i32 31), !dbg !38 + %5815 = bitcast i32 %5814 to float, !dbg !38 + %5816 = fadd float %5812, %5815, !dbg !35 + %5817 = bitcast float %5816 to i32, !dbg !38 + %5818 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5817, i32 1, i32 31), !dbg !38 + %5819 = bitcast i32 %5818 to float, !dbg !38 + %5820 = fadd float %5816, %5819, !dbg !35 + %5821 = bitcast float %5154 to i32, !dbg !38 + %5822 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5821, i32 8, i32 31), !dbg !38 + %5823 = bitcast i32 %5822 to float, !dbg !38 + %5824 = fadd float %5154, %5823, !dbg !35 + %5825 = bitcast float %5824 to i32, !dbg !38 + %5826 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5825, i32 4, i32 31), !dbg !38 + %5827 = bitcast i32 %5826 to float, !dbg !38 + %5828 = fadd float %5824, %5827, !dbg !35 + %5829 = bitcast float %5828 to i32, !dbg !38 + %5830 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5829, i32 2, i32 31), !dbg !38 + %5831 = bitcast i32 %5830 to float, !dbg !38 + %5832 = fadd float %5828, %5831, !dbg !35 + %5833 = bitcast float %5832 to i32, !dbg !38 + %5834 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5833, i32 1, i32 31), !dbg !38 + %5835 = bitcast i32 %5834 to float, !dbg !38 + %5836 = fadd float %5832, %5835, !dbg !35 + %5837 = bitcast float %5161 to i32, !dbg !38 + %5838 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5837, i32 8, i32 31), !dbg !38 + %5839 = bitcast i32 %5838 to float, !dbg !38 + %5840 = fadd float %5161, %5839, !dbg !35 + %5841 = bitcast float %5840 to i32, !dbg !38 + %5842 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5841, i32 4, i32 31), !dbg !38 + %5843 = bitcast i32 %5842 to float, !dbg !38 + %5844 = fadd float %5840, %5843, !dbg !35 + %5845 = bitcast float %5844 to i32, !dbg !38 + %5846 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5845, i32 2, i32 31), !dbg !38 + %5847 = bitcast i32 %5846 to float, !dbg !38 + %5848 = fadd float %5844, %5847, !dbg !35 + %5849 = bitcast float %5848 to i32, !dbg !38 + %5850 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5849, i32 1, i32 31), !dbg !38 + %5851 = bitcast i32 %5850 to float, !dbg !38 + %5852 = fadd float %5848, %5851, !dbg !35 + %5853 = bitcast float %5168 to i32, !dbg !38 + %5854 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5853, i32 8, i32 31), !dbg !38 + %5855 = bitcast i32 %5854 to float, !dbg !38 + %5856 = fadd float %5168, %5855, !dbg !35 + %5857 = bitcast float %5856 to i32, !dbg !38 + %5858 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5857, i32 4, i32 31), !dbg !38 + %5859 = bitcast i32 %5858 to float, !dbg !38 + %5860 = fadd float %5856, %5859, !dbg !35 + %5861 = bitcast float %5860 to i32, !dbg !38 + %5862 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5861, i32 2, i32 31), !dbg !38 + %5863 = bitcast i32 %5862 to float, !dbg !38 + %5864 = fadd float %5860, %5863, !dbg !35 + %5865 = bitcast float %5864 to i32, !dbg !38 + %5866 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5865, i32 1, i32 31), !dbg !38 + %5867 = bitcast i32 %5866 to float, !dbg !38 + %5868 = fadd float %5864, %5867, !dbg !35 + %5869 = bitcast float %5175 to i32, !dbg !38 + %5870 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5869, i32 8, i32 31), !dbg !38 + %5871 = bitcast i32 %5870 to float, !dbg !38 + %5872 = fadd float %5175, %5871, !dbg !35 + %5873 = bitcast float %5872 to i32, !dbg !38 + %5874 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5873, i32 4, i32 31), !dbg !38 + %5875 = bitcast i32 %5874 to float, !dbg !38 + %5876 = fadd float %5872, %5875, !dbg !35 + %5877 = bitcast float %5876 to i32, !dbg !38 + %5878 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5877, i32 2, i32 31), !dbg !38 + %5879 = bitcast i32 %5878 to float, !dbg !38 + %5880 = fadd float %5876, %5879, !dbg !35 + %5881 = bitcast float %5880 to i32, !dbg !38 + %5882 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5881, i32 1, i32 31), !dbg !38 + %5883 = bitcast i32 %5882 to float, !dbg !38 + %5884 = fadd float %5880, %5883, !dbg !35 + %5885 = bitcast float %5182 to i32, !dbg !38 + %5886 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5885, i32 8, i32 31), !dbg !38 + %5887 = bitcast i32 %5886 to float, !dbg !38 + %5888 = fadd float %5182, %5887, !dbg !35 + %5889 = bitcast float %5888 to i32, !dbg !38 + %5890 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5889, i32 4, i32 31), !dbg !38 + %5891 = bitcast i32 %5890 to float, !dbg !38 + %5892 = fadd float %5888, %5891, !dbg !35 + %5893 = bitcast float %5892 to i32, !dbg !38 + %5894 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5893, i32 2, i32 31), !dbg !38 + %5895 = bitcast i32 %5894 to float, !dbg !38 + %5896 = fadd float %5892, %5895, !dbg !35 + %5897 = bitcast float %5896 to i32, !dbg !38 + %5898 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5897, i32 1, i32 31), !dbg !38 + %5899 = bitcast i32 %5898 to float, !dbg !38 + %5900 = fadd float %5896, %5899, !dbg !35 + %5901 = bitcast float %5189 to i32, !dbg !38 + %5902 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5901, i32 8, i32 31), !dbg !38 + %5903 = bitcast i32 %5902 to float, !dbg !38 + %5904 = fadd float %5189, %5903, !dbg !35 + %5905 = bitcast float %5904 to i32, !dbg !38 + %5906 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5905, i32 4, i32 31), !dbg !38 + %5907 = bitcast i32 %5906 to float, !dbg !38 + %5908 = fadd float %5904, %5907, !dbg !35 + %5909 = bitcast float %5908 to i32, !dbg !38 + %5910 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5909, i32 2, i32 31), !dbg !38 + %5911 = bitcast i32 %5910 to float, !dbg !38 + %5912 = fadd float %5908, %5911, !dbg !35 + %5913 = bitcast float %5912 to i32, !dbg !38 + %5914 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5913, i32 1, i32 31), !dbg !38 + %5915 = bitcast i32 %5914 to float, !dbg !38 + %5916 = fadd float %5912, %5915, !dbg !35 + %5917 = bitcast float %5196 to i32, !dbg !38 + %5918 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5917, i32 8, i32 31), !dbg !38 + %5919 = bitcast i32 %5918 to float, !dbg !38 + %5920 = fadd float %5196, %5919, !dbg !35 + %5921 = bitcast float %5920 to i32, !dbg !38 + %5922 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5921, i32 4, i32 31), !dbg !38 + %5923 = bitcast i32 %5922 to float, !dbg !38 + %5924 = fadd float %5920, %5923, !dbg !35 + %5925 = bitcast float %5924 to i32, !dbg !38 + %5926 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5925, i32 2, i32 31), !dbg !38 + %5927 = bitcast i32 %5926 to float, !dbg !38 + %5928 = fadd float %5924, %5927, !dbg !35 + %5929 = bitcast float %5928 to i32, !dbg !38 + %5930 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5929, i32 1, i32 31), !dbg !38 + %5931 = bitcast i32 %5930 to float, !dbg !38 + %5932 = fadd float %5928, %5931, !dbg !35 + %5933 = bitcast float %5203 to i32, !dbg !38 + %5934 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5933, i32 8, i32 31), !dbg !38 + %5935 = bitcast i32 %5934 to float, !dbg !38 + %5936 = fadd float %5203, %5935, !dbg !35 + %5937 = bitcast float %5936 to i32, !dbg !38 + %5938 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5937, i32 4, i32 31), !dbg !38 + %5939 = bitcast i32 %5938 to float, !dbg !38 + %5940 = fadd float %5936, %5939, !dbg !35 + %5941 = bitcast float %5940 to i32, !dbg !38 + %5942 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5941, i32 2, i32 31), !dbg !38 + %5943 = bitcast i32 %5942 to float, !dbg !38 + %5944 = fadd float %5940, %5943, !dbg !35 + %5945 = bitcast float %5944 to i32, !dbg !38 + %5946 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5945, i32 1, i32 31), !dbg !38 + %5947 = bitcast i32 %5946 to float, !dbg !38 + %5948 = fadd float %5944, %5947, !dbg !35 + %5949 = bitcast float %5210 to i32, !dbg !38 + %5950 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5949, i32 8, i32 31), !dbg !38 + %5951 = bitcast i32 %5950 to float, !dbg !38 + %5952 = fadd float %5210, %5951, !dbg !35 + %5953 = bitcast float %5952 to i32, !dbg !38 + %5954 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5953, i32 4, i32 31), !dbg !38 + %5955 = bitcast i32 %5954 to float, !dbg !38 + %5956 = fadd float %5952, %5955, !dbg !35 + %5957 = bitcast float %5956 to i32, !dbg !38 + %5958 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5957, i32 2, i32 31), !dbg !38 + %5959 = bitcast i32 %5958 to float, !dbg !38 + %5960 = fadd float %5956, %5959, !dbg !35 + %5961 = bitcast float %5960 to i32, !dbg !38 + %5962 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5961, i32 1, i32 31), !dbg !38 + %5963 = bitcast i32 %5962 to float, !dbg !38 + %5964 = fadd float %5960, %5963, !dbg !35 + %5965 = bitcast float %5217 to i32, !dbg !38 + %5966 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5965, i32 8, i32 31), !dbg !38 + %5967 = bitcast i32 %5966 to float, !dbg !38 + %5968 = fadd float %5217, %5967, !dbg !35 + %5969 = bitcast float %5968 to i32, !dbg !38 + %5970 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5969, i32 4, i32 31), !dbg !38 + %5971 = bitcast i32 %5970 to float, !dbg !38 + %5972 = fadd float %5968, %5971, !dbg !35 + %5973 = bitcast float %5972 to i32, !dbg !38 + %5974 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5973, i32 2, i32 31), !dbg !38 + %5975 = bitcast i32 %5974 to float, !dbg !38 + %5976 = fadd float %5972, %5975, !dbg !35 + %5977 = bitcast float %5976 to i32, !dbg !38 + %5978 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5977, i32 1, i32 31), !dbg !38 + %5979 = bitcast i32 %5978 to float, !dbg !38 + %5980 = fadd float %5976, %5979, !dbg !35 + %5981 = bitcast float %5224 to i32, !dbg !38 + %5982 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5981, i32 8, i32 31), !dbg !38 + %5983 = bitcast i32 %5982 to float, !dbg !38 + %5984 = fadd float %5224, %5983, !dbg !35 + %5985 = bitcast float %5984 to i32, !dbg !38 + %5986 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5985, i32 4, i32 31), !dbg !38 + %5987 = bitcast i32 %5986 to float, !dbg !38 + %5988 = fadd float %5984, %5987, !dbg !35 + %5989 = bitcast float %5988 to i32, !dbg !38 + %5990 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5989, i32 2, i32 31), !dbg !38 + %5991 = bitcast i32 %5990 to float, !dbg !38 + %5992 = fadd float %5988, %5991, !dbg !35 + %5993 = bitcast float %5992 to i32, !dbg !38 + %5994 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5993, i32 1, i32 31), !dbg !38 + %5995 = bitcast i32 %5994 to float, !dbg !38 + %5996 = fadd float %5992, %5995, !dbg !35 + %5997 = bitcast float %5231 to i32, !dbg !38 + %5998 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %5997, i32 8, i32 31), !dbg !38 + %5999 = bitcast i32 %5998 to float, !dbg !38 + %6000 = fadd float %5231, %5999, !dbg !35 + %6001 = bitcast float %6000 to i32, !dbg !38 + %6002 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6001, i32 4, i32 31), !dbg !38 + %6003 = bitcast i32 %6002 to float, !dbg !38 + %6004 = fadd float %6000, %6003, !dbg !35 + %6005 = bitcast float %6004 to i32, !dbg !38 + %6006 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6005, i32 2, i32 31), !dbg !38 + %6007 = bitcast i32 %6006 to float, !dbg !38 + %6008 = fadd float %6004, %6007, !dbg !35 + %6009 = bitcast float %6008 to i32, !dbg !38 + %6010 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6009, i32 1, i32 31), !dbg !38 + %6011 = bitcast i32 %6010 to float, !dbg !38 + %6012 = fadd float %6008, %6011, !dbg !35 + %6013 = bitcast float %5238 to i32, !dbg !38 + %6014 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6013, i32 8, i32 31), !dbg !38 + %6015 = bitcast i32 %6014 to float, !dbg !38 + %6016 = fadd float %5238, %6015, !dbg !35 + %6017 = bitcast float %6016 to i32, !dbg !38 + %6018 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6017, i32 4, i32 31), !dbg !38 + %6019 = bitcast i32 %6018 to float, !dbg !38 + %6020 = fadd float %6016, %6019, !dbg !35 + %6021 = bitcast float %6020 to i32, !dbg !38 + %6022 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6021, i32 2, i32 31), !dbg !38 + %6023 = bitcast i32 %6022 to float, !dbg !38 + %6024 = fadd float %6020, %6023, !dbg !35 + %6025 = bitcast float %6024 to i32, !dbg !38 + %6026 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6025, i32 1, i32 31), !dbg !38 + %6027 = bitcast i32 %6026 to float, !dbg !38 + %6028 = fadd float %6024, %6027, !dbg !35 + %6029 = bitcast float %5245 to i32, !dbg !38 + %6030 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6029, i32 8, i32 31), !dbg !38 + %6031 = bitcast i32 %6030 to float, !dbg !38 + %6032 = fadd float %5245, %6031, !dbg !35 + %6033 = bitcast float %6032 to i32, !dbg !38 + %6034 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6033, i32 4, i32 31), !dbg !38 + %6035 = bitcast i32 %6034 to float, !dbg !38 + %6036 = fadd float %6032, %6035, !dbg !35 + %6037 = bitcast float %6036 to i32, !dbg !38 + %6038 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6037, i32 2, i32 31), !dbg !38 + %6039 = bitcast i32 %6038 to float, !dbg !38 + %6040 = fadd float %6036, %6039, !dbg !35 + %6041 = bitcast float %6040 to i32, !dbg !38 + %6042 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6041, i32 1, i32 31), !dbg !38 + %6043 = bitcast i32 %6042 to float, !dbg !38 + %6044 = fadd float %6040, %6043, !dbg !35 + %6045 = bitcast float %5252 to i32, !dbg !38 + %6046 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6045, i32 8, i32 31), !dbg !38 + %6047 = bitcast i32 %6046 to float, !dbg !38 + %6048 = fadd float %5252, %6047, !dbg !35 + %6049 = bitcast float %6048 to i32, !dbg !38 + %6050 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6049, i32 4, i32 31), !dbg !38 + %6051 = bitcast i32 %6050 to float, !dbg !38 + %6052 = fadd float %6048, %6051, !dbg !35 + %6053 = bitcast float %6052 to i32, !dbg !38 + %6054 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6053, i32 2, i32 31), !dbg !38 + %6055 = bitcast i32 %6054 to float, !dbg !38 + %6056 = fadd float %6052, %6055, !dbg !35 + %6057 = bitcast float %6056 to i32, !dbg !38 + %6058 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6057, i32 1, i32 31), !dbg !38 + %6059 = bitcast i32 %6058 to float, !dbg !38 + %6060 = fadd float %6056, %6059, !dbg !35 + %6061 = bitcast float %5259 to i32, !dbg !38 + %6062 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6061, i32 8, i32 31), !dbg !38 + %6063 = bitcast i32 %6062 to float, !dbg !38 + %6064 = fadd float %5259, %6063, !dbg !35 + %6065 = bitcast float %6064 to i32, !dbg !38 + %6066 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6065, i32 4, i32 31), !dbg !38 + %6067 = bitcast i32 %6066 to float, !dbg !38 + %6068 = fadd float %6064, %6067, !dbg !35 + %6069 = bitcast float %6068 to i32, !dbg !38 + %6070 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6069, i32 2, i32 31), !dbg !38 + %6071 = bitcast i32 %6070 to float, !dbg !38 + %6072 = fadd float %6068, %6071, !dbg !35 + %6073 = bitcast float %6072 to i32, !dbg !38 + %6074 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6073, i32 1, i32 31), !dbg !38 + %6075 = bitcast i32 %6074 to float, !dbg !38 + %6076 = fadd float %6072, %6075, !dbg !35 + %6077 = bitcast float %5266 to i32, !dbg !38 + %6078 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6077, i32 8, i32 31), !dbg !38 + %6079 = bitcast i32 %6078 to float, !dbg !38 + %6080 = fadd float %5266, %6079, !dbg !35 + %6081 = bitcast float %6080 to i32, !dbg !38 + %6082 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6081, i32 4, i32 31), !dbg !38 + %6083 = bitcast i32 %6082 to float, !dbg !38 + %6084 = fadd float %6080, %6083, !dbg !35 + %6085 = bitcast float %6084 to i32, !dbg !38 + %6086 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6085, i32 2, i32 31), !dbg !38 + %6087 = bitcast i32 %6086 to float, !dbg !38 + %6088 = fadd float %6084, %6087, !dbg !35 + %6089 = bitcast float %6088 to i32, !dbg !38 + %6090 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6089, i32 1, i32 31), !dbg !38 + %6091 = bitcast i32 %6090 to float, !dbg !38 + %6092 = fadd float %6088, %6091, !dbg !35 + %6093 = bitcast float %5273 to i32, !dbg !38 + %6094 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6093, i32 8, i32 31), !dbg !38 + %6095 = bitcast i32 %6094 to float, !dbg !38 + %6096 = fadd float %5273, %6095, !dbg !35 + %6097 = bitcast float %6096 to i32, !dbg !38 + %6098 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6097, i32 4, i32 31), !dbg !38 + %6099 = bitcast i32 %6098 to float, !dbg !38 + %6100 = fadd float %6096, %6099, !dbg !35 + %6101 = bitcast float %6100 to i32, !dbg !38 + %6102 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6101, i32 2, i32 31), !dbg !38 + %6103 = bitcast i32 %6102 to float, !dbg !38 + %6104 = fadd float %6100, %6103, !dbg !35 + %6105 = bitcast float %6104 to i32, !dbg !38 + %6106 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6105, i32 1, i32 31), !dbg !38 + %6107 = bitcast i32 %6106 to float, !dbg !38 + %6108 = fadd float %6104, %6107, !dbg !35 + %6109 = bitcast float %5280 to i32, !dbg !38 + %6110 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6109, i32 8, i32 31), !dbg !38 + %6111 = bitcast i32 %6110 to float, !dbg !38 + %6112 = fadd float %5280, %6111, !dbg !35 + %6113 = bitcast float %6112 to i32, !dbg !38 + %6114 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6113, i32 4, i32 31), !dbg !38 + %6115 = bitcast i32 %6114 to float, !dbg !38 + %6116 = fadd float %6112, %6115, !dbg !35 + %6117 = bitcast float %6116 to i32, !dbg !38 + %6118 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6117, i32 2, i32 31), !dbg !38 + %6119 = bitcast i32 %6118 to float, !dbg !38 + %6120 = fadd float %6116, %6119, !dbg !35 + %6121 = bitcast float %6120 to i32, !dbg !38 + %6122 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6121, i32 1, i32 31), !dbg !38 + %6123 = bitcast i32 %6122 to float, !dbg !38 + %6124 = fadd float %6120, %6123, !dbg !35 + %6125 = bitcast float %5287 to i32, !dbg !38 + %6126 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6125, i32 8, i32 31), !dbg !38 + %6127 = bitcast i32 %6126 to float, !dbg !38 + %6128 = fadd float %5287, %6127, !dbg !35 + %6129 = bitcast float %6128 to i32, !dbg !38 + %6130 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6129, i32 4, i32 31), !dbg !38 + %6131 = bitcast i32 %6130 to float, !dbg !38 + %6132 = fadd float %6128, %6131, !dbg !35 + %6133 = bitcast float %6132 to i32, !dbg !38 + %6134 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6133, i32 2, i32 31), !dbg !38 + %6135 = bitcast i32 %6134 to float, !dbg !38 + %6136 = fadd float %6132, %6135, !dbg !35 + %6137 = bitcast float %6136 to i32, !dbg !38 + %6138 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6137, i32 1, i32 31), !dbg !38 + %6139 = bitcast i32 %6138 to float, !dbg !38 + %6140 = fadd float %6136, %6139, !dbg !35 + %6141 = bitcast float %5294 to i32, !dbg !38 + %6142 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6141, i32 8, i32 31), !dbg !38 + %6143 = bitcast i32 %6142 to float, !dbg !38 + %6144 = fadd float %5294, %6143, !dbg !35 + %6145 = bitcast float %6144 to i32, !dbg !38 + %6146 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6145, i32 4, i32 31), !dbg !38 + %6147 = bitcast i32 %6146 to float, !dbg !38 + %6148 = fadd float %6144, %6147, !dbg !35 + %6149 = bitcast float %6148 to i32, !dbg !38 + %6150 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6149, i32 2, i32 31), !dbg !38 + %6151 = bitcast i32 %6150 to float, !dbg !38 + %6152 = fadd float %6148, %6151, !dbg !35 + %6153 = bitcast float %6152 to i32, !dbg !38 + %6154 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6153, i32 1, i32 31), !dbg !38 + %6155 = bitcast i32 %6154 to float, !dbg !38 + %6156 = fadd float %6152, %6155, !dbg !35 + %6157 = bitcast float %5301 to i32, !dbg !38 + %6158 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6157, i32 8, i32 31), !dbg !38 + %6159 = bitcast i32 %6158 to float, !dbg !38 + %6160 = fadd float %5301, %6159, !dbg !35 + %6161 = bitcast float %6160 to i32, !dbg !38 + %6162 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6161, i32 4, i32 31), !dbg !38 + %6163 = bitcast i32 %6162 to float, !dbg !38 + %6164 = fadd float %6160, %6163, !dbg !35 + %6165 = bitcast float %6164 to i32, !dbg !38 + %6166 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6165, i32 2, i32 31), !dbg !38 + %6167 = bitcast i32 %6166 to float, !dbg !38 + %6168 = fadd float %6164, %6167, !dbg !35 + %6169 = bitcast float %6168 to i32, !dbg !38 + %6170 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6169, i32 1, i32 31), !dbg !38 + %6171 = bitcast i32 %6170 to float, !dbg !38 + %6172 = fadd float %6168, %6171, !dbg !35 + %6173 = bitcast float %5308 to i32, !dbg !38 + %6174 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6173, i32 8, i32 31), !dbg !38 + %6175 = bitcast i32 %6174 to float, !dbg !38 + %6176 = fadd float %5308, %6175, !dbg !35 + %6177 = bitcast float %6176 to i32, !dbg !38 + %6178 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6177, i32 4, i32 31), !dbg !38 + %6179 = bitcast i32 %6178 to float, !dbg !38 + %6180 = fadd float %6176, %6179, !dbg !35 + %6181 = bitcast float %6180 to i32, !dbg !38 + %6182 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6181, i32 2, i32 31), !dbg !38 + %6183 = bitcast i32 %6182 to float, !dbg !38 + %6184 = fadd float %6180, %6183, !dbg !35 + %6185 = bitcast float %6184 to i32, !dbg !38 + %6186 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6185, i32 1, i32 31), !dbg !38 + %6187 = bitcast i32 %6186 to float, !dbg !38 + %6188 = fadd float %6184, %6187, !dbg !35 + %6189 = bitcast float %5315 to i32, !dbg !38 + %6190 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6189, i32 8, i32 31), !dbg !38 + %6191 = bitcast i32 %6190 to float, !dbg !38 + %6192 = fadd float %5315, %6191, !dbg !35 + %6193 = bitcast float %6192 to i32, !dbg !38 + %6194 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6193, i32 4, i32 31), !dbg !38 + %6195 = bitcast i32 %6194 to float, !dbg !38 + %6196 = fadd float %6192, %6195, !dbg !35 + %6197 = bitcast float %6196 to i32, !dbg !38 + %6198 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6197, i32 2, i32 31), !dbg !38 + %6199 = bitcast i32 %6198 to float, !dbg !38 + %6200 = fadd float %6196, %6199, !dbg !35 + %6201 = bitcast float %6200 to i32, !dbg !38 + %6202 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6201, i32 1, i32 31), !dbg !38 + %6203 = bitcast i32 %6202 to float, !dbg !38 + %6204 = fadd float %6200, %6203, !dbg !35 + %6205 = bitcast float %5322 to i32, !dbg !38 + %6206 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6205, i32 8, i32 31), !dbg !38 + %6207 = bitcast i32 %6206 to float, !dbg !38 + %6208 = fadd float %5322, %6207, !dbg !35 + %6209 = bitcast float %6208 to i32, !dbg !38 + %6210 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6209, i32 4, i32 31), !dbg !38 + %6211 = bitcast i32 %6210 to float, !dbg !38 + %6212 = fadd float %6208, %6211, !dbg !35 + %6213 = bitcast float %6212 to i32, !dbg !38 + %6214 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6213, i32 2, i32 31), !dbg !38 + %6215 = bitcast i32 %6214 to float, !dbg !38 + %6216 = fadd float %6212, %6215, !dbg !35 + %6217 = bitcast float %6216 to i32, !dbg !38 + %6218 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6217, i32 1, i32 31), !dbg !38 + %6219 = bitcast i32 %6218 to float, !dbg !38 + %6220 = fadd float %6216, %6219, !dbg !35 + %6221 = bitcast float %5329 to i32, !dbg !38 + %6222 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6221, i32 8, i32 31), !dbg !38 + %6223 = bitcast i32 %6222 to float, !dbg !38 + %6224 = fadd float %5329, %6223, !dbg !35 + %6225 = bitcast float %6224 to i32, !dbg !38 + %6226 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6225, i32 4, i32 31), !dbg !38 + %6227 = bitcast i32 %6226 to float, !dbg !38 + %6228 = fadd float %6224, %6227, !dbg !35 + %6229 = bitcast float %6228 to i32, !dbg !38 + %6230 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6229, i32 2, i32 31), !dbg !38 + %6231 = bitcast i32 %6230 to float, !dbg !38 + %6232 = fadd float %6228, %6231, !dbg !35 + %6233 = bitcast float %6232 to i32, !dbg !38 + %6234 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6233, i32 1, i32 31), !dbg !38 + %6235 = bitcast i32 %6234 to float, !dbg !38 + %6236 = fadd float %6232, %6235, !dbg !35 + %6237 = bitcast float %5336 to i32, !dbg !38 + %6238 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6237, i32 8, i32 31), !dbg !38 + %6239 = bitcast i32 %6238 to float, !dbg !38 + %6240 = fadd float %5336, %6239, !dbg !35 + %6241 = bitcast float %6240 to i32, !dbg !38 + %6242 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6241, i32 4, i32 31), !dbg !38 + %6243 = bitcast i32 %6242 to float, !dbg !38 + %6244 = fadd float %6240, %6243, !dbg !35 + %6245 = bitcast float %6244 to i32, !dbg !38 + %6246 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6245, i32 2, i32 31), !dbg !38 + %6247 = bitcast i32 %6246 to float, !dbg !38 + %6248 = fadd float %6244, %6247, !dbg !35 + %6249 = bitcast float %6248 to i32, !dbg !38 + %6250 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6249, i32 1, i32 31), !dbg !38 + %6251 = bitcast i32 %6250 to float, !dbg !38 + %6252 = fadd float %6248, %6251, !dbg !35 + %6253 = bitcast float %5343 to i32, !dbg !38 + %6254 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6253, i32 8, i32 31), !dbg !38 + %6255 = bitcast i32 %6254 to float, !dbg !38 + %6256 = fadd float %5343, %6255, !dbg !35 + %6257 = bitcast float %6256 to i32, !dbg !38 + %6258 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6257, i32 4, i32 31), !dbg !38 + %6259 = bitcast i32 %6258 to float, !dbg !38 + %6260 = fadd float %6256, %6259, !dbg !35 + %6261 = bitcast float %6260 to i32, !dbg !38 + %6262 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6261, i32 2, i32 31), !dbg !38 + %6263 = bitcast i32 %6262 to float, !dbg !38 + %6264 = fadd float %6260, %6263, !dbg !35 + %6265 = bitcast float %6264 to i32, !dbg !38 + %6266 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6265, i32 1, i32 31), !dbg !38 + %6267 = bitcast i32 %6266 to float, !dbg !38 + %6268 = fadd float %6264, %6267, !dbg !35 + %6269 = bitcast float %5350 to i32, !dbg !38 + %6270 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6269, i32 8, i32 31), !dbg !38 + %6271 = bitcast i32 %6270 to float, !dbg !38 + %6272 = fadd float %5350, %6271, !dbg !35 + %6273 = bitcast float %6272 to i32, !dbg !38 + %6274 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6273, i32 4, i32 31), !dbg !38 + %6275 = bitcast i32 %6274 to float, !dbg !38 + %6276 = fadd float %6272, %6275, !dbg !35 + %6277 = bitcast float %6276 to i32, !dbg !38 + %6278 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6277, i32 2, i32 31), !dbg !38 + %6279 = bitcast i32 %6278 to float, !dbg !38 + %6280 = fadd float %6276, %6279, !dbg !35 + %6281 = bitcast float %6280 to i32, !dbg !38 + %6282 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6281, i32 1, i32 31), !dbg !38 + %6283 = bitcast i32 %6282 to float, !dbg !38 + %6284 = fadd float %6280, %6283, !dbg !35 + %6285 = bitcast float %5357 to i32, !dbg !38 + %6286 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6285, i32 8, i32 31), !dbg !38 + %6287 = bitcast i32 %6286 to float, !dbg !38 + %6288 = fadd float %5357, %6287, !dbg !35 + %6289 = bitcast float %6288 to i32, !dbg !38 + %6290 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6289, i32 4, i32 31), !dbg !38 + %6291 = bitcast i32 %6290 to float, !dbg !38 + %6292 = fadd float %6288, %6291, !dbg !35 + %6293 = bitcast float %6292 to i32, !dbg !38 + %6294 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6293, i32 2, i32 31), !dbg !38 + %6295 = bitcast i32 %6294 to float, !dbg !38 + %6296 = fadd float %6292, %6295, !dbg !35 + %6297 = bitcast float %6296 to i32, !dbg !38 + %6298 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6297, i32 1, i32 31), !dbg !38 + %6299 = bitcast i32 %6298 to float, !dbg !38 + %6300 = fadd float %6296, %6299, !dbg !35 + %6301 = bitcast float %5364 to i32, !dbg !38 + %6302 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6301, i32 8, i32 31), !dbg !38 + %6303 = bitcast i32 %6302 to float, !dbg !38 + %6304 = fadd float %5364, %6303, !dbg !35 + %6305 = bitcast float %6304 to i32, !dbg !38 + %6306 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6305, i32 4, i32 31), !dbg !38 + %6307 = bitcast i32 %6306 to float, !dbg !38 + %6308 = fadd float %6304, %6307, !dbg !35 + %6309 = bitcast float %6308 to i32, !dbg !38 + %6310 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6309, i32 2, i32 31), !dbg !38 + %6311 = bitcast i32 %6310 to float, !dbg !38 + %6312 = fadd float %6308, %6311, !dbg !35 + %6313 = bitcast float %6312 to i32, !dbg !38 + %6314 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6313, i32 1, i32 31), !dbg !38 + %6315 = bitcast i32 %6314 to float, !dbg !38 + %6316 = fadd float %6312, %6315, !dbg !35 + %6317 = bitcast float %5371 to i32, !dbg !38 + %6318 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6317, i32 8, i32 31), !dbg !38 + %6319 = bitcast i32 %6318 to float, !dbg !38 + %6320 = fadd float %5371, %6319, !dbg !35 + %6321 = bitcast float %6320 to i32, !dbg !38 + %6322 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6321, i32 4, i32 31), !dbg !38 + %6323 = bitcast i32 %6322 to float, !dbg !38 + %6324 = fadd float %6320, %6323, !dbg !35 + %6325 = bitcast float %6324 to i32, !dbg !38 + %6326 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6325, i32 2, i32 31), !dbg !38 + %6327 = bitcast i32 %6326 to float, !dbg !38 + %6328 = fadd float %6324, %6327, !dbg !35 + %6329 = bitcast float %6328 to i32, !dbg !38 + %6330 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6329, i32 1, i32 31), !dbg !38 + %6331 = bitcast i32 %6330 to float, !dbg !38 + %6332 = fadd float %6328, %6331, !dbg !35 + %6333 = bitcast float %5378 to i32, !dbg !38 + %6334 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6333, i32 8, i32 31), !dbg !38 + %6335 = bitcast i32 %6334 to float, !dbg !38 + %6336 = fadd float %5378, %6335, !dbg !35 + %6337 = bitcast float %6336 to i32, !dbg !38 + %6338 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6337, i32 4, i32 31), !dbg !38 + %6339 = bitcast i32 %6338 to float, !dbg !38 + %6340 = fadd float %6336, %6339, !dbg !35 + %6341 = bitcast float %6340 to i32, !dbg !38 + %6342 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6341, i32 2, i32 31), !dbg !38 + %6343 = bitcast i32 %6342 to float, !dbg !38 + %6344 = fadd float %6340, %6343, !dbg !35 + %6345 = bitcast float %6344 to i32, !dbg !38 + %6346 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6345, i32 1, i32 31), !dbg !38 + %6347 = bitcast i32 %6346 to float, !dbg !38 + %6348 = fadd float %6344, %6347, !dbg !35 + %6349 = bitcast float %5385 to i32, !dbg !38 + %6350 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6349, i32 8, i32 31), !dbg !38 + %6351 = bitcast i32 %6350 to float, !dbg !38 + %6352 = fadd float %5385, %6351, !dbg !35 + %6353 = bitcast float %6352 to i32, !dbg !38 + %6354 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6353, i32 4, i32 31), !dbg !38 + %6355 = bitcast i32 %6354 to float, !dbg !38 + %6356 = fadd float %6352, %6355, !dbg !35 + %6357 = bitcast float %6356 to i32, !dbg !38 + %6358 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6357, i32 2, i32 31), !dbg !38 + %6359 = bitcast i32 %6358 to float, !dbg !38 + %6360 = fadd float %6356, %6359, !dbg !35 + %6361 = bitcast float %6360 to i32, !dbg !38 + %6362 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6361, i32 1, i32 31), !dbg !38 + %6363 = bitcast i32 %6362 to float, !dbg !38 + %6364 = fadd float %6360, %6363, !dbg !35 + %6365 = bitcast float %5392 to i32, !dbg !38 + %6366 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6365, i32 8, i32 31), !dbg !38 + %6367 = bitcast i32 %6366 to float, !dbg !38 + %6368 = fadd float %5392, %6367, !dbg !35 + %6369 = bitcast float %6368 to i32, !dbg !38 + %6370 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6369, i32 4, i32 31), !dbg !38 + %6371 = bitcast i32 %6370 to float, !dbg !38 + %6372 = fadd float %6368, %6371, !dbg !35 + %6373 = bitcast float %6372 to i32, !dbg !38 + %6374 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6373, i32 2, i32 31), !dbg !38 + %6375 = bitcast i32 %6374 to float, !dbg !38 + %6376 = fadd float %6372, %6375, !dbg !35 + %6377 = bitcast float %6376 to i32, !dbg !38 + %6378 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6377, i32 1, i32 31), !dbg !38 + %6379 = bitcast i32 %6378 to float, !dbg !38 + %6380 = fadd float %6376, %6379, !dbg !35 + %6381 = bitcast float %5399 to i32, !dbg !38 + %6382 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6381, i32 8, i32 31), !dbg !38 + %6383 = bitcast i32 %6382 to float, !dbg !38 + %6384 = fadd float %5399, %6383, !dbg !35 + %6385 = bitcast float %6384 to i32, !dbg !38 + %6386 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6385, i32 4, i32 31), !dbg !38 + %6387 = bitcast i32 %6386 to float, !dbg !38 + %6388 = fadd float %6384, %6387, !dbg !35 + %6389 = bitcast float %6388 to i32, !dbg !38 + %6390 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6389, i32 2, i32 31), !dbg !38 + %6391 = bitcast i32 %6390 to float, !dbg !38 + %6392 = fadd float %6388, %6391, !dbg !35 + %6393 = bitcast float %6392 to i32, !dbg !38 + %6394 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6393, i32 1, i32 31), !dbg !38 + %6395 = bitcast i32 %6394 to float, !dbg !38 + %6396 = fadd float %6392, %6395, !dbg !35 + %6397 = bitcast float %5406 to i32, !dbg !38 + %6398 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6397, i32 8, i32 31), !dbg !38 + %6399 = bitcast i32 %6398 to float, !dbg !38 + %6400 = fadd float %5406, %6399, !dbg !35 + %6401 = bitcast float %6400 to i32, !dbg !38 + %6402 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6401, i32 4, i32 31), !dbg !38 + %6403 = bitcast i32 %6402 to float, !dbg !38 + %6404 = fadd float %6400, %6403, !dbg !35 + %6405 = bitcast float %6404 to i32, !dbg !38 + %6406 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6405, i32 2, i32 31), !dbg !38 + %6407 = bitcast i32 %6406 to float, !dbg !38 + %6408 = fadd float %6404, %6407, !dbg !35 + %6409 = bitcast float %6408 to i32, !dbg !38 + %6410 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6409, i32 1, i32 31), !dbg !38 + %6411 = bitcast i32 %6410 to float, !dbg !38 + %6412 = fadd float %6408, %6411, !dbg !35 + %6413 = bitcast float %5413 to i32, !dbg !38 + %6414 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6413, i32 8, i32 31), !dbg !38 + %6415 = bitcast i32 %6414 to float, !dbg !38 + %6416 = fadd float %5413, %6415, !dbg !35 + %6417 = bitcast float %6416 to i32, !dbg !38 + %6418 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6417, i32 4, i32 31), !dbg !38 + %6419 = bitcast i32 %6418 to float, !dbg !38 + %6420 = fadd float %6416, %6419, !dbg !35 + %6421 = bitcast float %6420 to i32, !dbg !38 + %6422 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6421, i32 2, i32 31), !dbg !38 + %6423 = bitcast i32 %6422 to float, !dbg !38 + %6424 = fadd float %6420, %6423, !dbg !35 + %6425 = bitcast float %6424 to i32, !dbg !38 + %6426 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6425, i32 1, i32 31), !dbg !38 + %6427 = bitcast i32 %6426 to float, !dbg !38 + %6428 = fadd float %6424, %6427, !dbg !35 + %6429 = bitcast float %5420 to i32, !dbg !38 + %6430 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6429, i32 8, i32 31), !dbg !38 + %6431 = bitcast i32 %6430 to float, !dbg !38 + %6432 = fadd float %5420, %6431, !dbg !35 + %6433 = bitcast float %6432 to i32, !dbg !38 + %6434 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6433, i32 4, i32 31), !dbg !38 + %6435 = bitcast i32 %6434 to float, !dbg !38 + %6436 = fadd float %6432, %6435, !dbg !35 + %6437 = bitcast float %6436 to i32, !dbg !38 + %6438 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6437, i32 2, i32 31), !dbg !38 + %6439 = bitcast i32 %6438 to float, !dbg !38 + %6440 = fadd float %6436, %6439, !dbg !35 + %6441 = bitcast float %6440 to i32, !dbg !38 + %6442 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %6441, i32 1, i32 31), !dbg !38 + %6443 = bitcast i32 %6442 to float, !dbg !38 + %6444 = fadd float %6440, %6443, !dbg !35 + %6445 = fmul float %5436, %1730, !dbg !41 + %6446 = fmul float %5436, %1731, !dbg !41 + %6447 = fmul float %5436, %1732, !dbg !41 + %6448 = fmul float %5436, %1733, !dbg !41 + %6449 = fmul float %5436, %1734, !dbg !41 + %6450 = fmul float %5436, %1735, !dbg !41 + %6451 = fmul float %5436, %1736, !dbg !41 + %6452 = fmul float %5436, %1737, !dbg !41 + %6453 = fmul float %5452, %1738, !dbg !41 + %6454 = fmul float %5452, %1739, !dbg !41 + %6455 = fmul float %5452, %1740, !dbg !41 + %6456 = fmul float %5452, %1741, !dbg !41 + %6457 = fmul float %5452, %1742, !dbg !41 + %6458 = fmul float %5452, %1743, !dbg !41 + %6459 = fmul float %5452, %1744, !dbg !41 + %6460 = fmul float %5452, %1745, !dbg !41 + %6461 = fmul float %5468, %1746, !dbg !41 + %6462 = fmul float %5468, %1747, !dbg !41 + %6463 = fmul float %5468, %1748, !dbg !41 + %6464 = fmul float %5468, %1749, !dbg !41 + %6465 = fmul float %5468, %1750, !dbg !41 + %6466 = fmul float %5468, %1751, !dbg !41 + %6467 = fmul float %5468, %1752, !dbg !41 + %6468 = fmul float %5468, %1753, !dbg !41 + %6469 = fmul float %5484, %1754, !dbg !41 + %6470 = fmul float %5484, %1755, !dbg !41 + %6471 = fmul float %5484, %1756, !dbg !41 + %6472 = fmul float %5484, %1757, !dbg !41 + %6473 = fmul float %5484, %1758, !dbg !41 + %6474 = fmul float %5484, %1759, !dbg !41 + %6475 = fmul float %5484, %1760, !dbg !41 + %6476 = fmul float %5484, %1761, !dbg !41 + %6477 = fmul float %5500, %1762, !dbg !41 + %6478 = fmul float %5500, %1763, !dbg !41 + %6479 = fmul float %5500, %1764, !dbg !41 + %6480 = fmul float %5500, %1765, !dbg !41 + %6481 = fmul float %5500, %1766, !dbg !41 + %6482 = fmul float %5500, %1767, !dbg !41 + %6483 = fmul float %5500, %1768, !dbg !41 + %6484 = fmul float %5500, %1769, !dbg !41 + %6485 = fmul float %5516, %1770, !dbg !41 + %6486 = fmul float %5516, %1771, !dbg !41 + %6487 = fmul float %5516, %1772, !dbg !41 + %6488 = fmul float %5516, %1773, !dbg !41 + %6489 = fmul float %5516, %1774, !dbg !41 + %6490 = fmul float %5516, %1775, !dbg !41 + %6491 = fmul float %5516, %1776, !dbg !41 + %6492 = fmul float %5516, %1777, !dbg !41 + %6493 = fmul float %5532, %1778, !dbg !41 + %6494 = fmul float %5532, %1779, !dbg !41 + %6495 = fmul float %5532, %1780, !dbg !41 + %6496 = fmul float %5532, %1781, !dbg !41 + %6497 = fmul float %5532, %1782, !dbg !41 + %6498 = fmul float %5532, %1783, !dbg !41 + %6499 = fmul float %5532, %1784, !dbg !41 + %6500 = fmul float %5532, %1785, !dbg !41 + %6501 = fmul float %5548, %1786, !dbg !41 + %6502 = fmul float %5548, %1787, !dbg !41 + %6503 = fmul float %5548, %1788, !dbg !41 + %6504 = fmul float %5548, %1789, !dbg !41 + %6505 = fmul float %5548, %1790, !dbg !41 + %6506 = fmul float %5548, %1791, !dbg !41 + %6507 = fmul float %5548, %1792, !dbg !41 + %6508 = fmul float %5548, %1793, !dbg !41 + %6509 = fmul float %5564, %1794, !dbg !41 + %6510 = fmul float %5564, %1795, !dbg !41 + %6511 = fmul float %5564, %1796, !dbg !41 + %6512 = fmul float %5564, %1797, !dbg !41 + %6513 = fmul float %5564, %1798, !dbg !41 + %6514 = fmul float %5564, %1799, !dbg !41 + %6515 = fmul float %5564, %1800, !dbg !41 + %6516 = fmul float %5564, %1801, !dbg !41 + %6517 = fmul float %5580, %1802, !dbg !41 + %6518 = fmul float %5580, %1803, !dbg !41 + %6519 = fmul float %5580, %1804, !dbg !41 + %6520 = fmul float %5580, %1805, !dbg !41 + %6521 = fmul float %5580, %1806, !dbg !41 + %6522 = fmul float %5580, %1807, !dbg !41 + %6523 = fmul float %5580, %1808, !dbg !41 + %6524 = fmul float %5580, %1809, !dbg !41 + %6525 = fmul float %5596, %1810, !dbg !41 + %6526 = fmul float %5596, %1811, !dbg !41 + %6527 = fmul float %5596, %1812, !dbg !41 + %6528 = fmul float %5596, %1813, !dbg !41 + %6529 = fmul float %5596, %1814, !dbg !41 + %6530 = fmul float %5596, %1815, !dbg !41 + %6531 = fmul float %5596, %1816, !dbg !41 + %6532 = fmul float %5596, %1817, !dbg !41 + %6533 = fmul float %5612, %1818, !dbg !41 + %6534 = fmul float %5612, %1819, !dbg !41 + %6535 = fmul float %5612, %1820, !dbg !41 + %6536 = fmul float %5612, %1821, !dbg !41 + %6537 = fmul float %5612, %1822, !dbg !41 + %6538 = fmul float %5612, %1823, !dbg !41 + %6539 = fmul float %5612, %1824, !dbg !41 + %6540 = fmul float %5612, %1825, !dbg !41 + %6541 = fmul float %5628, %1826, !dbg !41 + %6542 = fmul float %5628, %1827, !dbg !41 + %6543 = fmul float %5628, %1828, !dbg !41 + %6544 = fmul float %5628, %1829, !dbg !41 + %6545 = fmul float %5628, %1830, !dbg !41 + %6546 = fmul float %5628, %1831, !dbg !41 + %6547 = fmul float %5628, %1832, !dbg !41 + %6548 = fmul float %5628, %1833, !dbg !41 + %6549 = fmul float %5644, %1834, !dbg !41 + %6550 = fmul float %5644, %1835, !dbg !41 + %6551 = fmul float %5644, %1836, !dbg !41 + %6552 = fmul float %5644, %1837, !dbg !41 + %6553 = fmul float %5644, %1838, !dbg !41 + %6554 = fmul float %5644, %1839, !dbg !41 + %6555 = fmul float %5644, %1840, !dbg !41 + %6556 = fmul float %5644, %1841, !dbg !41 + %6557 = fmul float %5660, %1842, !dbg !41 + %6558 = fmul float %5660, %1843, !dbg !41 + %6559 = fmul float %5660, %1844, !dbg !41 + %6560 = fmul float %5660, %1845, !dbg !41 + %6561 = fmul float %5660, %1846, !dbg !41 + %6562 = fmul float %5660, %1847, !dbg !41 + %6563 = fmul float %5660, %1848, !dbg !41 + %6564 = fmul float %5660, %1849, !dbg !41 + %6565 = fmul float %5676, %1850, !dbg !41 + %6566 = fmul float %5676, %1851, !dbg !41 + %6567 = fmul float %5676, %1852, !dbg !41 + %6568 = fmul float %5676, %1853, !dbg !41 + %6569 = fmul float %5676, %1854, !dbg !41 + %6570 = fmul float %5676, %1855, !dbg !41 + %6571 = fmul float %5676, %1856, !dbg !41 + %6572 = fmul float %5676, %1857, !dbg !41 + %6573 = fmul float %5692, %1858, !dbg !41 + %6574 = fmul float %5692, %1859, !dbg !41 + %6575 = fmul float %5692, %1860, !dbg !41 + %6576 = fmul float %5692, %1861, !dbg !41 + %6577 = fmul float %5692, %1862, !dbg !41 + %6578 = fmul float %5692, %1863, !dbg !41 + %6579 = fmul float %5692, %1864, !dbg !41 + %6580 = fmul float %5692, %1865, !dbg !41 + %6581 = fmul float %5708, %1866, !dbg !41 + %6582 = fmul float %5708, %1867, !dbg !41 + %6583 = fmul float %5708, %1868, !dbg !41 + %6584 = fmul float %5708, %1869, !dbg !41 + %6585 = fmul float %5708, %1870, !dbg !41 + %6586 = fmul float %5708, %1871, !dbg !41 + %6587 = fmul float %5708, %1872, !dbg !41 + %6588 = fmul float %5708, %1873, !dbg !41 + %6589 = fmul float %5724, %1874, !dbg !41 + %6590 = fmul float %5724, %1875, !dbg !41 + %6591 = fmul float %5724, %1876, !dbg !41 + %6592 = fmul float %5724, %1877, !dbg !41 + %6593 = fmul float %5724, %1878, !dbg !41 + %6594 = fmul float %5724, %1879, !dbg !41 + %6595 = fmul float %5724, %1880, !dbg !41 + %6596 = fmul float %5724, %1881, !dbg !41 + %6597 = fmul float %5740, %1882, !dbg !41 + %6598 = fmul float %5740, %1883, !dbg !41 + %6599 = fmul float %5740, %1884, !dbg !41 + %6600 = fmul float %5740, %1885, !dbg !41 + %6601 = fmul float %5740, %1886, !dbg !41 + %6602 = fmul float %5740, %1887, !dbg !41 + %6603 = fmul float %5740, %1888, !dbg !41 + %6604 = fmul float %5740, %1889, !dbg !41 + %6605 = fmul float %5756, %1890, !dbg !41 + %6606 = fmul float %5756, %1891, !dbg !41 + %6607 = fmul float %5756, %1892, !dbg !41 + %6608 = fmul float %5756, %1893, !dbg !41 + %6609 = fmul float %5756, %1894, !dbg !41 + %6610 = fmul float %5756, %1895, !dbg !41 + %6611 = fmul float %5756, %1896, !dbg !41 + %6612 = fmul float %5756, %1897, !dbg !41 + %6613 = fmul float %5772, %1898, !dbg !41 + %6614 = fmul float %5772, %1899, !dbg !41 + %6615 = fmul float %5772, %1900, !dbg !41 + %6616 = fmul float %5772, %1901, !dbg !41 + %6617 = fmul float %5772, %1902, !dbg !41 + %6618 = fmul float %5772, %1903, !dbg !41 + %6619 = fmul float %5772, %1904, !dbg !41 + %6620 = fmul float %5772, %1905, !dbg !41 + %6621 = fmul float %5788, %1906, !dbg !41 + %6622 = fmul float %5788, %1907, !dbg !41 + %6623 = fmul float %5788, %1908, !dbg !41 + %6624 = fmul float %5788, %1909, !dbg !41 + %6625 = fmul float %5788, %1910, !dbg !41 + %6626 = fmul float %5788, %1911, !dbg !41 + %6627 = fmul float %5788, %1912, !dbg !41 + %6628 = fmul float %5788, %1913, !dbg !41 + %6629 = fmul float %5804, %1914, !dbg !41 + %6630 = fmul float %5804, %1915, !dbg !41 + %6631 = fmul float %5804, %1916, !dbg !41 + %6632 = fmul float %5804, %1917, !dbg !41 + %6633 = fmul float %5804, %1918, !dbg !41 + %6634 = fmul float %5804, %1919, !dbg !41 + %6635 = fmul float %5804, %1920, !dbg !41 + %6636 = fmul float %5804, %1921, !dbg !41 + %6637 = fmul float %5820, %1922, !dbg !41 + %6638 = fmul float %5820, %1923, !dbg !41 + %6639 = fmul float %5820, %1924, !dbg !41 + %6640 = fmul float %5820, %1925, !dbg !41 + %6641 = fmul float %5820, %1926, !dbg !41 + %6642 = fmul float %5820, %1927, !dbg !41 + %6643 = fmul float %5820, %1928, !dbg !41 + %6644 = fmul float %5820, %1929, !dbg !41 + %6645 = fmul float %5836, %1930, !dbg !41 + %6646 = fmul float %5836, %1931, !dbg !41 + %6647 = fmul float %5836, %1932, !dbg !41 + %6648 = fmul float %5836, %1933, !dbg !41 + %6649 = fmul float %5836, %1934, !dbg !41 + %6650 = fmul float %5836, %1935, !dbg !41 + %6651 = fmul float %5836, %1936, !dbg !41 + %6652 = fmul float %5836, %1937, !dbg !41 + %6653 = fmul float %5852, %1938, !dbg !41 + %6654 = fmul float %5852, %1939, !dbg !41 + %6655 = fmul float %5852, %1940, !dbg !41 + %6656 = fmul float %5852, %1941, !dbg !41 + %6657 = fmul float %5852, %1942, !dbg !41 + %6658 = fmul float %5852, %1943, !dbg !41 + %6659 = fmul float %5852, %1944, !dbg !41 + %6660 = fmul float %5852, %1945, !dbg !41 + %6661 = fmul float %5868, %1946, !dbg !41 + %6662 = fmul float %5868, %1947, !dbg !41 + %6663 = fmul float %5868, %1948, !dbg !41 + %6664 = fmul float %5868, %1949, !dbg !41 + %6665 = fmul float %5868, %1950, !dbg !41 + %6666 = fmul float %5868, %1951, !dbg !41 + %6667 = fmul float %5868, %1952, !dbg !41 + %6668 = fmul float %5868, %1953, !dbg !41 + %6669 = fmul float %5884, %1954, !dbg !41 + %6670 = fmul float %5884, %1955, !dbg !41 + %6671 = fmul float %5884, %1956, !dbg !41 + %6672 = fmul float %5884, %1957, !dbg !41 + %6673 = fmul float %5884, %1958, !dbg !41 + %6674 = fmul float %5884, %1959, !dbg !41 + %6675 = fmul float %5884, %1960, !dbg !41 + %6676 = fmul float %5884, %1961, !dbg !41 + %6677 = fmul float %5900, %1962, !dbg !41 + %6678 = fmul float %5900, %1963, !dbg !41 + %6679 = fmul float %5900, %1964, !dbg !41 + %6680 = fmul float %5900, %1965, !dbg !41 + %6681 = fmul float %5900, %1966, !dbg !41 + %6682 = fmul float %5900, %1967, !dbg !41 + %6683 = fmul float %5900, %1968, !dbg !41 + %6684 = fmul float %5900, %1969, !dbg !41 + %6685 = fmul float %5916, %1970, !dbg !41 + %6686 = fmul float %5916, %1971, !dbg !41 + %6687 = fmul float %5916, %1972, !dbg !41 + %6688 = fmul float %5916, %1973, !dbg !41 + %6689 = fmul float %5916, %1974, !dbg !41 + %6690 = fmul float %5916, %1975, !dbg !41 + %6691 = fmul float %5916, %1976, !dbg !41 + %6692 = fmul float %5916, %1977, !dbg !41 + %6693 = fmul float %5932, %1978, !dbg !41 + %6694 = fmul float %5932, %1979, !dbg !41 + %6695 = fmul float %5932, %1980, !dbg !41 + %6696 = fmul float %5932, %1981, !dbg !41 + %6697 = fmul float %5932, %1982, !dbg !41 + %6698 = fmul float %5932, %1983, !dbg !41 + %6699 = fmul float %5932, %1984, !dbg !41 + %6700 = fmul float %5932, %1985, !dbg !41 + %6701 = fmul float %5948, %1986, !dbg !41 + %6702 = fmul float %5948, %1987, !dbg !41 + %6703 = fmul float %5948, %1988, !dbg !41 + %6704 = fmul float %5948, %1989, !dbg !41 + %6705 = fmul float %5948, %1990, !dbg !41 + %6706 = fmul float %5948, %1991, !dbg !41 + %6707 = fmul float %5948, %1992, !dbg !41 + %6708 = fmul float %5948, %1993, !dbg !41 + %6709 = fmul float %5964, %1994, !dbg !41 + %6710 = fmul float %5964, %1995, !dbg !41 + %6711 = fmul float %5964, %1996, !dbg !41 + %6712 = fmul float %5964, %1997, !dbg !41 + %6713 = fmul float %5964, %1998, !dbg !41 + %6714 = fmul float %5964, %1999, !dbg !41 + %6715 = fmul float %5964, %2000, !dbg !41 + %6716 = fmul float %5964, %2001, !dbg !41 + %6717 = fmul float %5980, %2002, !dbg !41 + %6718 = fmul float %5980, %2003, !dbg !41 + %6719 = fmul float %5980, %2004, !dbg !41 + %6720 = fmul float %5980, %2005, !dbg !41 + %6721 = fmul float %5980, %2006, !dbg !41 + %6722 = fmul float %5980, %2007, !dbg !41 + %6723 = fmul float %5980, %2008, !dbg !41 + %6724 = fmul float %5980, %2009, !dbg !41 + %6725 = fmul float %5996, %2010, !dbg !41 + %6726 = fmul float %5996, %2011, !dbg !41 + %6727 = fmul float %5996, %2012, !dbg !41 + %6728 = fmul float %5996, %2013, !dbg !41 + %6729 = fmul float %5996, %2014, !dbg !41 + %6730 = fmul float %5996, %2015, !dbg !41 + %6731 = fmul float %5996, %2016, !dbg !41 + %6732 = fmul float %5996, %2017, !dbg !41 + %6733 = fmul float %6012, %2018, !dbg !41 + %6734 = fmul float %6012, %2019, !dbg !41 + %6735 = fmul float %6012, %2020, !dbg !41 + %6736 = fmul float %6012, %2021, !dbg !41 + %6737 = fmul float %6012, %2022, !dbg !41 + %6738 = fmul float %6012, %2023, !dbg !41 + %6739 = fmul float %6012, %2024, !dbg !41 + %6740 = fmul float %6012, %2025, !dbg !41 + %6741 = fmul float %6028, %2026, !dbg !41 + %6742 = fmul float %6028, %2027, !dbg !41 + %6743 = fmul float %6028, %2028, !dbg !41 + %6744 = fmul float %6028, %2029, !dbg !41 + %6745 = fmul float %6028, %2030, !dbg !41 + %6746 = fmul float %6028, %2031, !dbg !41 + %6747 = fmul float %6028, %2032, !dbg !41 + %6748 = fmul float %6028, %2033, !dbg !41 + %6749 = fmul float %6044, %2034, !dbg !41 + %6750 = fmul float %6044, %2035, !dbg !41 + %6751 = fmul float %6044, %2036, !dbg !41 + %6752 = fmul float %6044, %2037, !dbg !41 + %6753 = fmul float %6044, %2038, !dbg !41 + %6754 = fmul float %6044, %2039, !dbg !41 + %6755 = fmul float %6044, %2040, !dbg !41 + %6756 = fmul float %6044, %2041, !dbg !41 + %6757 = fmul float %6060, %2042, !dbg !41 + %6758 = fmul float %6060, %2043, !dbg !41 + %6759 = fmul float %6060, %2044, !dbg !41 + %6760 = fmul float %6060, %2045, !dbg !41 + %6761 = fmul float %6060, %2046, !dbg !41 + %6762 = fmul float %6060, %2047, !dbg !41 + %6763 = fmul float %6060, %2048, !dbg !41 + %6764 = fmul float %6060, %2049, !dbg !41 + %6765 = fmul float %6076, %2050, !dbg !41 + %6766 = fmul float %6076, %2051, !dbg !41 + %6767 = fmul float %6076, %2052, !dbg !41 + %6768 = fmul float %6076, %2053, !dbg !41 + %6769 = fmul float %6076, %2054, !dbg !41 + %6770 = fmul float %6076, %2055, !dbg !41 + %6771 = fmul float %6076, %2056, !dbg !41 + %6772 = fmul float %6076, %2057, !dbg !41 + %6773 = fmul float %6092, %2058, !dbg !41 + %6774 = fmul float %6092, %2059, !dbg !41 + %6775 = fmul float %6092, %2060, !dbg !41 + %6776 = fmul float %6092, %2061, !dbg !41 + %6777 = fmul float %6092, %2062, !dbg !41 + %6778 = fmul float %6092, %2063, !dbg !41 + %6779 = fmul float %6092, %2064, !dbg !41 + %6780 = fmul float %6092, %2065, !dbg !41 + %6781 = fmul float %6108, %2066, !dbg !41 + %6782 = fmul float %6108, %2067, !dbg !41 + %6783 = fmul float %6108, %2068, !dbg !41 + %6784 = fmul float %6108, %2069, !dbg !41 + %6785 = fmul float %6108, %2070, !dbg !41 + %6786 = fmul float %6108, %2071, !dbg !41 + %6787 = fmul float %6108, %2072, !dbg !41 + %6788 = fmul float %6108, %2073, !dbg !41 + %6789 = fmul float %6124, %2074, !dbg !41 + %6790 = fmul float %6124, %2075, !dbg !41 + %6791 = fmul float %6124, %2076, !dbg !41 + %6792 = fmul float %6124, %2077, !dbg !41 + %6793 = fmul float %6124, %2078, !dbg !41 + %6794 = fmul float %6124, %2079, !dbg !41 + %6795 = fmul float %6124, %2080, !dbg !41 + %6796 = fmul float %6124, %2081, !dbg !41 + %6797 = fmul float %6140, %2082, !dbg !41 + %6798 = fmul float %6140, %2083, !dbg !41 + %6799 = fmul float %6140, %2084, !dbg !41 + %6800 = fmul float %6140, %2085, !dbg !41 + %6801 = fmul float %6140, %2086, !dbg !41 + %6802 = fmul float %6140, %2087, !dbg !41 + %6803 = fmul float %6140, %2088, !dbg !41 + %6804 = fmul float %6140, %2089, !dbg !41 + %6805 = fmul float %6156, %2090, !dbg !41 + %6806 = fmul float %6156, %2091, !dbg !41 + %6807 = fmul float %6156, %2092, !dbg !41 + %6808 = fmul float %6156, %2093, !dbg !41 + %6809 = fmul float %6156, %2094, !dbg !41 + %6810 = fmul float %6156, %2095, !dbg !41 + %6811 = fmul float %6156, %2096, !dbg !41 + %6812 = fmul float %6156, %2097, !dbg !41 + %6813 = fmul float %6172, %2098, !dbg !41 + %6814 = fmul float %6172, %2099, !dbg !41 + %6815 = fmul float %6172, %2100, !dbg !41 + %6816 = fmul float %6172, %2101, !dbg !41 + %6817 = fmul float %6172, %2102, !dbg !41 + %6818 = fmul float %6172, %2103, !dbg !41 + %6819 = fmul float %6172, %2104, !dbg !41 + %6820 = fmul float %6172, %2105, !dbg !41 + %6821 = fmul float %6188, %2106, !dbg !41 + %6822 = fmul float %6188, %2107, !dbg !41 + %6823 = fmul float %6188, %2108, !dbg !41 + %6824 = fmul float %6188, %2109, !dbg !41 + %6825 = fmul float %6188, %2110, !dbg !41 + %6826 = fmul float %6188, %2111, !dbg !41 + %6827 = fmul float %6188, %2112, !dbg !41 + %6828 = fmul float %6188, %2113, !dbg !41 + %6829 = fmul float %6204, %2114, !dbg !41 + %6830 = fmul float %6204, %2115, !dbg !41 + %6831 = fmul float %6204, %2116, !dbg !41 + %6832 = fmul float %6204, %2117, !dbg !41 + %6833 = fmul float %6204, %2118, !dbg !41 + %6834 = fmul float %6204, %2119, !dbg !41 + %6835 = fmul float %6204, %2120, !dbg !41 + %6836 = fmul float %6204, %2121, !dbg !41 + %6837 = fmul float %6220, %2122, !dbg !41 + %6838 = fmul float %6220, %2123, !dbg !41 + %6839 = fmul float %6220, %2124, !dbg !41 + %6840 = fmul float %6220, %2125, !dbg !41 + %6841 = fmul float %6220, %2126, !dbg !41 + %6842 = fmul float %6220, %2127, !dbg !41 + %6843 = fmul float %6220, %2128, !dbg !41 + %6844 = fmul float %6220, %2129, !dbg !41 + %6845 = fmul float %6236, %2130, !dbg !41 + %6846 = fmul float %6236, %2131, !dbg !41 + %6847 = fmul float %6236, %2132, !dbg !41 + %6848 = fmul float %6236, %2133, !dbg !41 + %6849 = fmul float %6236, %2134, !dbg !41 + %6850 = fmul float %6236, %2135, !dbg !41 + %6851 = fmul float %6236, %2136, !dbg !41 + %6852 = fmul float %6236, %2137, !dbg !41 + %6853 = fmul float %6252, %2138, !dbg !41 + %6854 = fmul float %6252, %2139, !dbg !41 + %6855 = fmul float %6252, %2140, !dbg !41 + %6856 = fmul float %6252, %2141, !dbg !41 + %6857 = fmul float %6252, %2142, !dbg !41 + %6858 = fmul float %6252, %2143, !dbg !41 + %6859 = fmul float %6252, %2144, !dbg !41 + %6860 = fmul float %6252, %2145, !dbg !41 + %6861 = fmul float %6268, %2146, !dbg !41 + %6862 = fmul float %6268, %2147, !dbg !41 + %6863 = fmul float %6268, %2148, !dbg !41 + %6864 = fmul float %6268, %2149, !dbg !41 + %6865 = fmul float %6268, %2150, !dbg !41 + %6866 = fmul float %6268, %2151, !dbg !41 + %6867 = fmul float %6268, %2152, !dbg !41 + %6868 = fmul float %6268, %2153, !dbg !41 + %6869 = fmul float %6284, %2154, !dbg !41 + %6870 = fmul float %6284, %2155, !dbg !41 + %6871 = fmul float %6284, %2156, !dbg !41 + %6872 = fmul float %6284, %2157, !dbg !41 + %6873 = fmul float %6284, %2158, !dbg !41 + %6874 = fmul float %6284, %2159, !dbg !41 + %6875 = fmul float %6284, %2160, !dbg !41 + %6876 = fmul float %6284, %2161, !dbg !41 + %6877 = fmul float %6300, %2162, !dbg !41 + %6878 = fmul float %6300, %2163, !dbg !41 + %6879 = fmul float %6300, %2164, !dbg !41 + %6880 = fmul float %6300, %2165, !dbg !41 + %6881 = fmul float %6300, %2166, !dbg !41 + %6882 = fmul float %6300, %2167, !dbg !41 + %6883 = fmul float %6300, %2168, !dbg !41 + %6884 = fmul float %6300, %2169, !dbg !41 + %6885 = fmul float %6316, %2170, !dbg !41 + %6886 = fmul float %6316, %2171, !dbg !41 + %6887 = fmul float %6316, %2172, !dbg !41 + %6888 = fmul float %6316, %2173, !dbg !41 + %6889 = fmul float %6316, %2174, !dbg !41 + %6890 = fmul float %6316, %2175, !dbg !41 + %6891 = fmul float %6316, %2176, !dbg !41 + %6892 = fmul float %6316, %2177, !dbg !41 + %6893 = fmul float %6332, %2178, !dbg !41 + %6894 = fmul float %6332, %2179, !dbg !41 + %6895 = fmul float %6332, %2180, !dbg !41 + %6896 = fmul float %6332, %2181, !dbg !41 + %6897 = fmul float %6332, %2182, !dbg !41 + %6898 = fmul float %6332, %2183, !dbg !41 + %6899 = fmul float %6332, %2184, !dbg !41 + %6900 = fmul float %6332, %2185, !dbg !41 + %6901 = fmul float %6348, %2186, !dbg !41 + %6902 = fmul float %6348, %2187, !dbg !41 + %6903 = fmul float %6348, %2188, !dbg !41 + %6904 = fmul float %6348, %2189, !dbg !41 + %6905 = fmul float %6348, %2190, !dbg !41 + %6906 = fmul float %6348, %2191, !dbg !41 + %6907 = fmul float %6348, %2192, !dbg !41 + %6908 = fmul float %6348, %2193, !dbg !41 + %6909 = fmul float %6364, %2194, !dbg !41 + %6910 = fmul float %6364, %2195, !dbg !41 + %6911 = fmul float %6364, %2196, !dbg !41 + %6912 = fmul float %6364, %2197, !dbg !41 + %6913 = fmul float %6364, %2198, !dbg !41 + %6914 = fmul float %6364, %2199, !dbg !41 + %6915 = fmul float %6364, %2200, !dbg !41 + %6916 = fmul float %6364, %2201, !dbg !41 + %6917 = fmul float %6380, %2202, !dbg !41 + %6918 = fmul float %6380, %2203, !dbg !41 + %6919 = fmul float %6380, %2204, !dbg !41 + %6920 = fmul float %6380, %2205, !dbg !41 + %6921 = fmul float %6380, %2206, !dbg !41 + %6922 = fmul float %6380, %2207, !dbg !41 + %6923 = fmul float %6380, %2208, !dbg !41 + %6924 = fmul float %6380, %2209, !dbg !41 + %6925 = fmul float %6396, %2210, !dbg !41 + %6926 = fmul float %6396, %2211, !dbg !41 + %6927 = fmul float %6396, %2212, !dbg !41 + %6928 = fmul float %6396, %2213, !dbg !41 + %6929 = fmul float %6396, %2214, !dbg !41 + %6930 = fmul float %6396, %2215, !dbg !41 + %6931 = fmul float %6396, %2216, !dbg !41 + %6932 = fmul float %6396, %2217, !dbg !41 + %6933 = fmul float %6412, %2218, !dbg !41 + %6934 = fmul float %6412, %2219, !dbg !41 + %6935 = fmul float %6412, %2220, !dbg !41 + %6936 = fmul float %6412, %2221, !dbg !41 + %6937 = fmul float %6412, %2222, !dbg !41 + %6938 = fmul float %6412, %2223, !dbg !41 + %6939 = fmul float %6412, %2224, !dbg !41 + %6940 = fmul float %6412, %2225, !dbg !41 + %6941 = fmul float %6428, %2226, !dbg !41 + %6942 = fmul float %6428, %2227, !dbg !41 + %6943 = fmul float %6428, %2228, !dbg !41 + %6944 = fmul float %6428, %2229, !dbg !41 + %6945 = fmul float %6428, %2230, !dbg !41 + %6946 = fmul float %6428, %2231, !dbg !41 + %6947 = fmul float %6428, %2232, !dbg !41 + %6948 = fmul float %6428, %2233, !dbg !41 + %6949 = fmul float %6444, %2234, !dbg !41 + %6950 = fmul float %6444, %2235, !dbg !41 + %6951 = fmul float %6444, %2236, !dbg !41 + %6952 = fmul float %6444, %2237, !dbg !41 + %6953 = fmul float %6444, %2238, !dbg !41 + %6954 = fmul float %6444, %2239, !dbg !41 + %6955 = fmul float %6444, %2240, !dbg !41 + %6956 = fmul float %6444, %2241, !dbg !41 + %6957 = fmul float %3854, %6445, !dbg !42 + %6958 = fmul float %3854, %6446, !dbg !42 + %6959 = fmul float %3854, %6447, !dbg !42 + %6960 = fmul float %3854, %6448, !dbg !42 + %6961 = fmul float %3854, %6449, !dbg !42 + %6962 = fmul float %3854, %6450, !dbg !42 + %6963 = fmul float %3854, %6451, !dbg !42 + %6964 = fmul float %3854, %6452, !dbg !42 + %6965 = fmul float %3917, %6453, !dbg !42 + %6966 = fmul float %3917, %6454, !dbg !42 + %6967 = fmul float %3917, %6455, !dbg !42 + %6968 = fmul float %3917, %6456, !dbg !42 + %6969 = fmul float %3917, %6457, !dbg !42 + %6970 = fmul float %3917, %6458, !dbg !42 + %6971 = fmul float %3917, %6459, !dbg !42 + %6972 = fmul float %3917, %6460, !dbg !42 + %6973 = fmul float %3856, %6461, !dbg !42 + %6974 = fmul float %3856, %6462, !dbg !42 + %6975 = fmul float %3856, %6463, !dbg !42 + %6976 = fmul float %3856, %6464, !dbg !42 + %6977 = fmul float %3856, %6465, !dbg !42 + %6978 = fmul float %3856, %6466, !dbg !42 + %6979 = fmul float %3856, %6467, !dbg !42 + %6980 = fmul float %3856, %6468, !dbg !42 + %6981 = fmul float %3918, %6469, !dbg !42 + %6982 = fmul float %3918, %6470, !dbg !42 + %6983 = fmul float %3918, %6471, !dbg !42 + %6984 = fmul float %3918, %6472, !dbg !42 + %6985 = fmul float %3918, %6473, !dbg !42 + %6986 = fmul float %3918, %6474, !dbg !42 + %6987 = fmul float %3918, %6475, !dbg !42 + %6988 = fmul float %3918, %6476, !dbg !42 + %6989 = fmul float %3858, %6477, !dbg !42 + %6990 = fmul float %3858, %6478, !dbg !42 + %6991 = fmul float %3858, %6479, !dbg !42 + %6992 = fmul float %3858, %6480, !dbg !42 + %6993 = fmul float %3858, %6481, !dbg !42 + %6994 = fmul float %3858, %6482, !dbg !42 + %6995 = fmul float %3858, %6483, !dbg !42 + %6996 = fmul float %3858, %6484, !dbg !42 + %6997 = fmul float %3919, %6485, !dbg !42 + %6998 = fmul float %3919, %6486, !dbg !42 + %6999 = fmul float %3919, %6487, !dbg !42 + %7000 = fmul float %3919, %6488, !dbg !42 + %7001 = fmul float %3919, %6489, !dbg !42 + %7002 = fmul float %3919, %6490, !dbg !42 + %7003 = fmul float %3919, %6491, !dbg !42 + %7004 = fmul float %3919, %6492, !dbg !42 + %7005 = fmul float %3860, %6493, !dbg !42 + %7006 = fmul float %3860, %6494, !dbg !42 + %7007 = fmul float %3860, %6495, !dbg !42 + %7008 = fmul float %3860, %6496, !dbg !42 + %7009 = fmul float %3860, %6497, !dbg !42 + %7010 = fmul float %3860, %6498, !dbg !42 + %7011 = fmul float %3860, %6499, !dbg !42 + %7012 = fmul float %3860, %6500, !dbg !42 + %7013 = fmul float %3920, %6501, !dbg !42 + %7014 = fmul float %3920, %6502, !dbg !42 + %7015 = fmul float %3920, %6503, !dbg !42 + %7016 = fmul float %3920, %6504, !dbg !42 + %7017 = fmul float %3920, %6505, !dbg !42 + %7018 = fmul float %3920, %6506, !dbg !42 + %7019 = fmul float %3920, %6507, !dbg !42 + %7020 = fmul float %3920, %6508, !dbg !42 + %7021 = fmul float %3862, %6509, !dbg !42 + %7022 = fmul float %3862, %6510, !dbg !42 + %7023 = fmul float %3862, %6511, !dbg !42 + %7024 = fmul float %3862, %6512, !dbg !42 + %7025 = fmul float %3862, %6513, !dbg !42 + %7026 = fmul float %3862, %6514, !dbg !42 + %7027 = fmul float %3862, %6515, !dbg !42 + %7028 = fmul float %3862, %6516, !dbg !42 + %7029 = fmul float %3921, %6517, !dbg !42 + %7030 = fmul float %3921, %6518, !dbg !42 + %7031 = fmul float %3921, %6519, !dbg !42 + %7032 = fmul float %3921, %6520, !dbg !42 + %7033 = fmul float %3921, %6521, !dbg !42 + %7034 = fmul float %3921, %6522, !dbg !42 + %7035 = fmul float %3921, %6523, !dbg !42 + %7036 = fmul float %3921, %6524, !dbg !42 + %7037 = fmul float %3864, %6525, !dbg !42 + %7038 = fmul float %3864, %6526, !dbg !42 + %7039 = fmul float %3864, %6527, !dbg !42 + %7040 = fmul float %3864, %6528, !dbg !42 + %7041 = fmul float %3864, %6529, !dbg !42 + %7042 = fmul float %3864, %6530, !dbg !42 + %7043 = fmul float %3864, %6531, !dbg !42 + %7044 = fmul float %3864, %6532, !dbg !42 + %7045 = fmul float %3922, %6533, !dbg !42 + %7046 = fmul float %3922, %6534, !dbg !42 + %7047 = fmul float %3922, %6535, !dbg !42 + %7048 = fmul float %3922, %6536, !dbg !42 + %7049 = fmul float %3922, %6537, !dbg !42 + %7050 = fmul float %3922, %6538, !dbg !42 + %7051 = fmul float %3922, %6539, !dbg !42 + %7052 = fmul float %3922, %6540, !dbg !42 + %7053 = fmul float %3866, %6541, !dbg !42 + %7054 = fmul float %3866, %6542, !dbg !42 + %7055 = fmul float %3866, %6543, !dbg !42 + %7056 = fmul float %3866, %6544, !dbg !42 + %7057 = fmul float %3866, %6545, !dbg !42 + %7058 = fmul float %3866, %6546, !dbg !42 + %7059 = fmul float %3866, %6547, !dbg !42 + %7060 = fmul float %3866, %6548, !dbg !42 + %7061 = fmul float %3923, %6549, !dbg !42 + %7062 = fmul float %3923, %6550, !dbg !42 + %7063 = fmul float %3923, %6551, !dbg !42 + %7064 = fmul float %3923, %6552, !dbg !42 + %7065 = fmul float %3923, %6553, !dbg !42 + %7066 = fmul float %3923, %6554, !dbg !42 + %7067 = fmul float %3923, %6555, !dbg !42 + %7068 = fmul float %3923, %6556, !dbg !42 + %7069 = fmul float %3868, %6557, !dbg !42 + %7070 = fmul float %3868, %6558, !dbg !42 + %7071 = fmul float %3868, %6559, !dbg !42 + %7072 = fmul float %3868, %6560, !dbg !42 + %7073 = fmul float %3868, %6561, !dbg !42 + %7074 = fmul float %3868, %6562, !dbg !42 + %7075 = fmul float %3868, %6563, !dbg !42 + %7076 = fmul float %3868, %6564, !dbg !42 + %7077 = fmul float %3924, %6565, !dbg !42 + %7078 = fmul float %3924, %6566, !dbg !42 + %7079 = fmul float %3924, %6567, !dbg !42 + %7080 = fmul float %3924, %6568, !dbg !42 + %7081 = fmul float %3924, %6569, !dbg !42 + %7082 = fmul float %3924, %6570, !dbg !42 + %7083 = fmul float %3924, %6571, !dbg !42 + %7084 = fmul float %3924, %6572, !dbg !42 + %7085 = fmul float %3870, %6573, !dbg !42 + %7086 = fmul float %3870, %6574, !dbg !42 + %7087 = fmul float %3870, %6575, !dbg !42 + %7088 = fmul float %3870, %6576, !dbg !42 + %7089 = fmul float %3870, %6577, !dbg !42 + %7090 = fmul float %3870, %6578, !dbg !42 + %7091 = fmul float %3870, %6579, !dbg !42 + %7092 = fmul float %3870, %6580, !dbg !42 + %7093 = fmul float %3925, %6581, !dbg !42 + %7094 = fmul float %3925, %6582, !dbg !42 + %7095 = fmul float %3925, %6583, !dbg !42 + %7096 = fmul float %3925, %6584, !dbg !42 + %7097 = fmul float %3925, %6585, !dbg !42 + %7098 = fmul float %3925, %6586, !dbg !42 + %7099 = fmul float %3925, %6587, !dbg !42 + %7100 = fmul float %3925, %6588, !dbg !42 + %7101 = fmul float %3872, %6589, !dbg !42 + %7102 = fmul float %3872, %6590, !dbg !42 + %7103 = fmul float %3872, %6591, !dbg !42 + %7104 = fmul float %3872, %6592, !dbg !42 + %7105 = fmul float %3872, %6593, !dbg !42 + %7106 = fmul float %3872, %6594, !dbg !42 + %7107 = fmul float %3872, %6595, !dbg !42 + %7108 = fmul float %3872, %6596, !dbg !42 + %7109 = fmul float %3926, %6597, !dbg !42 + %7110 = fmul float %3926, %6598, !dbg !42 + %7111 = fmul float %3926, %6599, !dbg !42 + %7112 = fmul float %3926, %6600, !dbg !42 + %7113 = fmul float %3926, %6601, !dbg !42 + %7114 = fmul float %3926, %6602, !dbg !42 + %7115 = fmul float %3926, %6603, !dbg !42 + %7116 = fmul float %3926, %6604, !dbg !42 + %7117 = fmul float %3874, %6605, !dbg !42 + %7118 = fmul float %3874, %6606, !dbg !42 + %7119 = fmul float %3874, %6607, !dbg !42 + %7120 = fmul float %3874, %6608, !dbg !42 + %7121 = fmul float %3874, %6609, !dbg !42 + %7122 = fmul float %3874, %6610, !dbg !42 + %7123 = fmul float %3874, %6611, !dbg !42 + %7124 = fmul float %3874, %6612, !dbg !42 + %7125 = fmul float %3927, %6613, !dbg !42 + %7126 = fmul float %3927, %6614, !dbg !42 + %7127 = fmul float %3927, %6615, !dbg !42 + %7128 = fmul float %3927, %6616, !dbg !42 + %7129 = fmul float %3927, %6617, !dbg !42 + %7130 = fmul float %3927, %6618, !dbg !42 + %7131 = fmul float %3927, %6619, !dbg !42 + %7132 = fmul float %3927, %6620, !dbg !42 + %7133 = fmul float %3876, %6621, !dbg !42 + %7134 = fmul float %3876, %6622, !dbg !42 + %7135 = fmul float %3876, %6623, !dbg !42 + %7136 = fmul float %3876, %6624, !dbg !42 + %7137 = fmul float %3876, %6625, !dbg !42 + %7138 = fmul float %3876, %6626, !dbg !42 + %7139 = fmul float %3876, %6627, !dbg !42 + %7140 = fmul float %3876, %6628, !dbg !42 + %7141 = fmul float %3928, %6629, !dbg !42 + %7142 = fmul float %3928, %6630, !dbg !42 + %7143 = fmul float %3928, %6631, !dbg !42 + %7144 = fmul float %3928, %6632, !dbg !42 + %7145 = fmul float %3928, %6633, !dbg !42 + %7146 = fmul float %3928, %6634, !dbg !42 + %7147 = fmul float %3928, %6635, !dbg !42 + %7148 = fmul float %3928, %6636, !dbg !42 + %7149 = fmul float %3878, %6637, !dbg !42 + %7150 = fmul float %3878, %6638, !dbg !42 + %7151 = fmul float %3878, %6639, !dbg !42 + %7152 = fmul float %3878, %6640, !dbg !42 + %7153 = fmul float %3878, %6641, !dbg !42 + %7154 = fmul float %3878, %6642, !dbg !42 + %7155 = fmul float %3878, %6643, !dbg !42 + %7156 = fmul float %3878, %6644, !dbg !42 + %7157 = fmul float %3929, %6645, !dbg !42 + %7158 = fmul float %3929, %6646, !dbg !42 + %7159 = fmul float %3929, %6647, !dbg !42 + %7160 = fmul float %3929, %6648, !dbg !42 + %7161 = fmul float %3929, %6649, !dbg !42 + %7162 = fmul float %3929, %6650, !dbg !42 + %7163 = fmul float %3929, %6651, !dbg !42 + %7164 = fmul float %3929, %6652, !dbg !42 + %7165 = fmul float %3880, %6653, !dbg !42 + %7166 = fmul float %3880, %6654, !dbg !42 + %7167 = fmul float %3880, %6655, !dbg !42 + %7168 = fmul float %3880, %6656, !dbg !42 + %7169 = fmul float %3880, %6657, !dbg !42 + %7170 = fmul float %3880, %6658, !dbg !42 + %7171 = fmul float %3880, %6659, !dbg !42 + %7172 = fmul float %3880, %6660, !dbg !42 + %7173 = fmul float %3930, %6661, !dbg !42 + %7174 = fmul float %3930, %6662, !dbg !42 + %7175 = fmul float %3930, %6663, !dbg !42 + %7176 = fmul float %3930, %6664, !dbg !42 + %7177 = fmul float %3930, %6665, !dbg !42 + %7178 = fmul float %3930, %6666, !dbg !42 + %7179 = fmul float %3930, %6667, !dbg !42 + %7180 = fmul float %3930, %6668, !dbg !42 + %7181 = fmul float %3882, %6669, !dbg !42 + %7182 = fmul float %3882, %6670, !dbg !42 + %7183 = fmul float %3882, %6671, !dbg !42 + %7184 = fmul float %3882, %6672, !dbg !42 + %7185 = fmul float %3882, %6673, !dbg !42 + %7186 = fmul float %3882, %6674, !dbg !42 + %7187 = fmul float %3882, %6675, !dbg !42 + %7188 = fmul float %3882, %6676, !dbg !42 + %7189 = fmul float %3931, %6677, !dbg !42 + %7190 = fmul float %3931, %6678, !dbg !42 + %7191 = fmul float %3931, %6679, !dbg !42 + %7192 = fmul float %3931, %6680, !dbg !42 + %7193 = fmul float %3931, %6681, !dbg !42 + %7194 = fmul float %3931, %6682, !dbg !42 + %7195 = fmul float %3931, %6683, !dbg !42 + %7196 = fmul float %3931, %6684, !dbg !42 + %7197 = fmul float %3884, %6685, !dbg !42 + %7198 = fmul float %3884, %6686, !dbg !42 + %7199 = fmul float %3884, %6687, !dbg !42 + %7200 = fmul float %3884, %6688, !dbg !42 + %7201 = fmul float %3884, %6689, !dbg !42 + %7202 = fmul float %3884, %6690, !dbg !42 + %7203 = fmul float %3884, %6691, !dbg !42 + %7204 = fmul float %3884, %6692, !dbg !42 + %7205 = fmul float %3932, %6693, !dbg !42 + %7206 = fmul float %3932, %6694, !dbg !42 + %7207 = fmul float %3932, %6695, !dbg !42 + %7208 = fmul float %3932, %6696, !dbg !42 + %7209 = fmul float %3932, %6697, !dbg !42 + %7210 = fmul float %3932, %6698, !dbg !42 + %7211 = fmul float %3932, %6699, !dbg !42 + %7212 = fmul float %3932, %6700, !dbg !42 + %7213 = fmul float %3886, %6701, !dbg !42 + %7214 = fmul float %3886, %6702, !dbg !42 + %7215 = fmul float %3886, %6703, !dbg !42 + %7216 = fmul float %3886, %6704, !dbg !42 + %7217 = fmul float %3886, %6705, !dbg !42 + %7218 = fmul float %3886, %6706, !dbg !42 + %7219 = fmul float %3886, %6707, !dbg !42 + %7220 = fmul float %3886, %6708, !dbg !42 + %7221 = fmul float %3933, %6709, !dbg !42 + %7222 = fmul float %3933, %6710, !dbg !42 + %7223 = fmul float %3933, %6711, !dbg !42 + %7224 = fmul float %3933, %6712, !dbg !42 + %7225 = fmul float %3933, %6713, !dbg !42 + %7226 = fmul float %3933, %6714, !dbg !42 + %7227 = fmul float %3933, %6715, !dbg !42 + %7228 = fmul float %3933, %6716, !dbg !42 + %7229 = fmul float %3888, %6717, !dbg !42 + %7230 = fmul float %3888, %6718, !dbg !42 + %7231 = fmul float %3888, %6719, !dbg !42 + %7232 = fmul float %3888, %6720, !dbg !42 + %7233 = fmul float %3888, %6721, !dbg !42 + %7234 = fmul float %3888, %6722, !dbg !42 + %7235 = fmul float %3888, %6723, !dbg !42 + %7236 = fmul float %3888, %6724, !dbg !42 + %7237 = fmul float %3934, %6725, !dbg !42 + %7238 = fmul float %3934, %6726, !dbg !42 + %7239 = fmul float %3934, %6727, !dbg !42 + %7240 = fmul float %3934, %6728, !dbg !42 + %7241 = fmul float %3934, %6729, !dbg !42 + %7242 = fmul float %3934, %6730, !dbg !42 + %7243 = fmul float %3934, %6731, !dbg !42 + %7244 = fmul float %3934, %6732, !dbg !42 + %7245 = fmul float %3890, %6733, !dbg !42 + %7246 = fmul float %3890, %6734, !dbg !42 + %7247 = fmul float %3890, %6735, !dbg !42 + %7248 = fmul float %3890, %6736, !dbg !42 + %7249 = fmul float %3890, %6737, !dbg !42 + %7250 = fmul float %3890, %6738, !dbg !42 + %7251 = fmul float %3890, %6739, !dbg !42 + %7252 = fmul float %3890, %6740, !dbg !42 + %7253 = fmul float %3935, %6741, !dbg !42 + %7254 = fmul float %3935, %6742, !dbg !42 + %7255 = fmul float %3935, %6743, !dbg !42 + %7256 = fmul float %3935, %6744, !dbg !42 + %7257 = fmul float %3935, %6745, !dbg !42 + %7258 = fmul float %3935, %6746, !dbg !42 + %7259 = fmul float %3935, %6747, !dbg !42 + %7260 = fmul float %3935, %6748, !dbg !42 + %7261 = fmul float %3892, %6749, !dbg !42 + %7262 = fmul float %3892, %6750, !dbg !42 + %7263 = fmul float %3892, %6751, !dbg !42 + %7264 = fmul float %3892, %6752, !dbg !42 + %7265 = fmul float %3892, %6753, !dbg !42 + %7266 = fmul float %3892, %6754, !dbg !42 + %7267 = fmul float %3892, %6755, !dbg !42 + %7268 = fmul float %3892, %6756, !dbg !42 + %7269 = fmul float %3936, %6757, !dbg !42 + %7270 = fmul float %3936, %6758, !dbg !42 + %7271 = fmul float %3936, %6759, !dbg !42 + %7272 = fmul float %3936, %6760, !dbg !42 + %7273 = fmul float %3936, %6761, !dbg !42 + %7274 = fmul float %3936, %6762, !dbg !42 + %7275 = fmul float %3936, %6763, !dbg !42 + %7276 = fmul float %3936, %6764, !dbg !42 + %7277 = fmul float %3894, %6765, !dbg !42 + %7278 = fmul float %3894, %6766, !dbg !42 + %7279 = fmul float %3894, %6767, !dbg !42 + %7280 = fmul float %3894, %6768, !dbg !42 + %7281 = fmul float %3894, %6769, !dbg !42 + %7282 = fmul float %3894, %6770, !dbg !42 + %7283 = fmul float %3894, %6771, !dbg !42 + %7284 = fmul float %3894, %6772, !dbg !42 + %7285 = fmul float %3937, %6773, !dbg !42 + %7286 = fmul float %3937, %6774, !dbg !42 + %7287 = fmul float %3937, %6775, !dbg !42 + %7288 = fmul float %3937, %6776, !dbg !42 + %7289 = fmul float %3937, %6777, !dbg !42 + %7290 = fmul float %3937, %6778, !dbg !42 + %7291 = fmul float %3937, %6779, !dbg !42 + %7292 = fmul float %3937, %6780, !dbg !42 + %7293 = fmul float %3896, %6781, !dbg !42 + %7294 = fmul float %3896, %6782, !dbg !42 + %7295 = fmul float %3896, %6783, !dbg !42 + %7296 = fmul float %3896, %6784, !dbg !42 + %7297 = fmul float %3896, %6785, !dbg !42 + %7298 = fmul float %3896, %6786, !dbg !42 + %7299 = fmul float %3896, %6787, !dbg !42 + %7300 = fmul float %3896, %6788, !dbg !42 + %7301 = fmul float %3938, %6789, !dbg !42 + %7302 = fmul float %3938, %6790, !dbg !42 + %7303 = fmul float %3938, %6791, !dbg !42 + %7304 = fmul float %3938, %6792, !dbg !42 + %7305 = fmul float %3938, %6793, !dbg !42 + %7306 = fmul float %3938, %6794, !dbg !42 + %7307 = fmul float %3938, %6795, !dbg !42 + %7308 = fmul float %3938, %6796, !dbg !42 + %7309 = fmul float %3898, %6797, !dbg !42 + %7310 = fmul float %3898, %6798, !dbg !42 + %7311 = fmul float %3898, %6799, !dbg !42 + %7312 = fmul float %3898, %6800, !dbg !42 + %7313 = fmul float %3898, %6801, !dbg !42 + %7314 = fmul float %3898, %6802, !dbg !42 + %7315 = fmul float %3898, %6803, !dbg !42 + %7316 = fmul float %3898, %6804, !dbg !42 + %7317 = fmul float %3939, %6805, !dbg !42 + %7318 = fmul float %3939, %6806, !dbg !42 + %7319 = fmul float %3939, %6807, !dbg !42 + %7320 = fmul float %3939, %6808, !dbg !42 + %7321 = fmul float %3939, %6809, !dbg !42 + %7322 = fmul float %3939, %6810, !dbg !42 + %7323 = fmul float %3939, %6811, !dbg !42 + %7324 = fmul float %3939, %6812, !dbg !42 + %7325 = fmul float %3900, %6813, !dbg !42 + %7326 = fmul float %3900, %6814, !dbg !42 + %7327 = fmul float %3900, %6815, !dbg !42 + %7328 = fmul float %3900, %6816, !dbg !42 + %7329 = fmul float %3900, %6817, !dbg !42 + %7330 = fmul float %3900, %6818, !dbg !42 + %7331 = fmul float %3900, %6819, !dbg !42 + %7332 = fmul float %3900, %6820, !dbg !42 + %7333 = fmul float %3940, %6821, !dbg !42 + %7334 = fmul float %3940, %6822, !dbg !42 + %7335 = fmul float %3940, %6823, !dbg !42 + %7336 = fmul float %3940, %6824, !dbg !42 + %7337 = fmul float %3940, %6825, !dbg !42 + %7338 = fmul float %3940, %6826, !dbg !42 + %7339 = fmul float %3940, %6827, !dbg !42 + %7340 = fmul float %3940, %6828, !dbg !42 + %7341 = fmul float %3902, %6829, !dbg !42 + %7342 = fmul float %3902, %6830, !dbg !42 + %7343 = fmul float %3902, %6831, !dbg !42 + %7344 = fmul float %3902, %6832, !dbg !42 + %7345 = fmul float %3902, %6833, !dbg !42 + %7346 = fmul float %3902, %6834, !dbg !42 + %7347 = fmul float %3902, %6835, !dbg !42 + %7348 = fmul float %3902, %6836, !dbg !42 + %7349 = fmul float %3941, %6837, !dbg !42 + %7350 = fmul float %3941, %6838, !dbg !42 + %7351 = fmul float %3941, %6839, !dbg !42 + %7352 = fmul float %3941, %6840, !dbg !42 + %7353 = fmul float %3941, %6841, !dbg !42 + %7354 = fmul float %3941, %6842, !dbg !42 + %7355 = fmul float %3941, %6843, !dbg !42 + %7356 = fmul float %3941, %6844, !dbg !42 + %7357 = fmul float %3904, %6845, !dbg !42 + %7358 = fmul float %3904, %6846, !dbg !42 + %7359 = fmul float %3904, %6847, !dbg !42 + %7360 = fmul float %3904, %6848, !dbg !42 + %7361 = fmul float %3904, %6849, !dbg !42 + %7362 = fmul float %3904, %6850, !dbg !42 + %7363 = fmul float %3904, %6851, !dbg !42 + %7364 = fmul float %3904, %6852, !dbg !42 + %7365 = fmul float %3942, %6853, !dbg !42 + %7366 = fmul float %3942, %6854, !dbg !42 + %7367 = fmul float %3942, %6855, !dbg !42 + %7368 = fmul float %3942, %6856, !dbg !42 + %7369 = fmul float %3942, %6857, !dbg !42 + %7370 = fmul float %3942, %6858, !dbg !42 + %7371 = fmul float %3942, %6859, !dbg !42 + %7372 = fmul float %3942, %6860, !dbg !42 + %7373 = fmul float %3906, %6861, !dbg !42 + %7374 = fmul float %3906, %6862, !dbg !42 + %7375 = fmul float %3906, %6863, !dbg !42 + %7376 = fmul float %3906, %6864, !dbg !42 + %7377 = fmul float %3906, %6865, !dbg !42 + %7378 = fmul float %3906, %6866, !dbg !42 + %7379 = fmul float %3906, %6867, !dbg !42 + %7380 = fmul float %3906, %6868, !dbg !42 + %7381 = fmul float %3943, %6869, !dbg !42 + %7382 = fmul float %3943, %6870, !dbg !42 + %7383 = fmul float %3943, %6871, !dbg !42 + %7384 = fmul float %3943, %6872, !dbg !42 + %7385 = fmul float %3943, %6873, !dbg !42 + %7386 = fmul float %3943, %6874, !dbg !42 + %7387 = fmul float %3943, %6875, !dbg !42 + %7388 = fmul float %3943, %6876, !dbg !42 + %7389 = fmul float %3908, %6877, !dbg !42 + %7390 = fmul float %3908, %6878, !dbg !42 + %7391 = fmul float %3908, %6879, !dbg !42 + %7392 = fmul float %3908, %6880, !dbg !42 + %7393 = fmul float %3908, %6881, !dbg !42 + %7394 = fmul float %3908, %6882, !dbg !42 + %7395 = fmul float %3908, %6883, !dbg !42 + %7396 = fmul float %3908, %6884, !dbg !42 + %7397 = fmul float %3944, %6885, !dbg !42 + %7398 = fmul float %3944, %6886, !dbg !42 + %7399 = fmul float %3944, %6887, !dbg !42 + %7400 = fmul float %3944, %6888, !dbg !42 + %7401 = fmul float %3944, %6889, !dbg !42 + %7402 = fmul float %3944, %6890, !dbg !42 + %7403 = fmul float %3944, %6891, !dbg !42 + %7404 = fmul float %3944, %6892, !dbg !42 + %7405 = fmul float %3910, %6893, !dbg !42 + %7406 = fmul float %3910, %6894, !dbg !42 + %7407 = fmul float %3910, %6895, !dbg !42 + %7408 = fmul float %3910, %6896, !dbg !42 + %7409 = fmul float %3910, %6897, !dbg !42 + %7410 = fmul float %3910, %6898, !dbg !42 + %7411 = fmul float %3910, %6899, !dbg !42 + %7412 = fmul float %3910, %6900, !dbg !42 + %7413 = fmul float %3945, %6901, !dbg !42 + %7414 = fmul float %3945, %6902, !dbg !42 + %7415 = fmul float %3945, %6903, !dbg !42 + %7416 = fmul float %3945, %6904, !dbg !42 + %7417 = fmul float %3945, %6905, !dbg !42 + %7418 = fmul float %3945, %6906, !dbg !42 + %7419 = fmul float %3945, %6907, !dbg !42 + %7420 = fmul float %3945, %6908, !dbg !42 + %7421 = fmul float %3912, %6909, !dbg !42 + %7422 = fmul float %3912, %6910, !dbg !42 + %7423 = fmul float %3912, %6911, !dbg !42 + %7424 = fmul float %3912, %6912, !dbg !42 + %7425 = fmul float %3912, %6913, !dbg !42 + %7426 = fmul float %3912, %6914, !dbg !42 + %7427 = fmul float %3912, %6915, !dbg !42 + %7428 = fmul float %3912, %6916, !dbg !42 + %7429 = fmul float %3946, %6917, !dbg !42 + %7430 = fmul float %3946, %6918, !dbg !42 + %7431 = fmul float %3946, %6919, !dbg !42 + %7432 = fmul float %3946, %6920, !dbg !42 + %7433 = fmul float %3946, %6921, !dbg !42 + %7434 = fmul float %3946, %6922, !dbg !42 + %7435 = fmul float %3946, %6923, !dbg !42 + %7436 = fmul float %3946, %6924, !dbg !42 + %7437 = fmul float %3914, %6925, !dbg !42 + %7438 = fmul float %3914, %6926, !dbg !42 + %7439 = fmul float %3914, %6927, !dbg !42 + %7440 = fmul float %3914, %6928, !dbg !42 + %7441 = fmul float %3914, %6929, !dbg !42 + %7442 = fmul float %3914, %6930, !dbg !42 + %7443 = fmul float %3914, %6931, !dbg !42 + %7444 = fmul float %3914, %6932, !dbg !42 + %7445 = fmul float %3947, %6933, !dbg !42 + %7446 = fmul float %3947, %6934, !dbg !42 + %7447 = fmul float %3947, %6935, !dbg !42 + %7448 = fmul float %3947, %6936, !dbg !42 + %7449 = fmul float %3947, %6937, !dbg !42 + %7450 = fmul float %3947, %6938, !dbg !42 + %7451 = fmul float %3947, %6939, !dbg !42 + %7452 = fmul float %3947, %6940, !dbg !42 + %7453 = fmul float %3916, %6941, !dbg !42 + %7454 = fmul float %3916, %6942, !dbg !42 + %7455 = fmul float %3916, %6943, !dbg !42 + %7456 = fmul float %3916, %6944, !dbg !42 + %7457 = fmul float %3916, %6945, !dbg !42 + %7458 = fmul float %3916, %6946, !dbg !42 + %7459 = fmul float %3916, %6947, !dbg !42 + %7460 = fmul float %3916, %6948, !dbg !42 + %7461 = fmul float %3948, %6949, !dbg !42 + %7462 = fmul float %3948, %6950, !dbg !42 + %7463 = fmul float %3948, %6951, !dbg !42 + %7464 = fmul float %3948, %6952, !dbg !42 + %7465 = fmul float %3948, %6953, !dbg !42 + %7466 = fmul float %3948, %6954, !dbg !42 + %7467 = fmul float %3948, %6955, !dbg !42 + %7468 = fmul float %3948, %6956, !dbg !42 + %7469 = fsub float %3949, %6957, !dbg !43 + %7470 = fsub float %3950, %6958, !dbg !43 + %7471 = fsub float %3951, %6959, !dbg !43 + %7472 = fsub float %3952, %6960, !dbg !43 + %7473 = fsub float %3953, %6961, !dbg !43 + %7474 = fsub float %3954, %6962, !dbg !43 + %7475 = fsub float %3955, %6963, !dbg !43 + %7476 = fsub float %3956, %6964, !dbg !43 + %7477 = insertelement <2 x float> poison, float %3957, i64 0, !dbg !43 + %7478 = insertelement <2 x float> %7477, float %3958, i64 1, !dbg !43 + %7479 = insertelement <2 x float> poison, float %6965, i64 0, !dbg !43 + %7480 = insertelement <2 x float> %7479, float %6966, i64 1, !dbg !43 + %7481 = fsub <2 x float> %7478, %7480, !dbg !43 + %7482 = insertelement <2 x float> poison, float %3959, i64 0, !dbg !43 + %7483 = insertelement <2 x float> %7482, float %3960, i64 1, !dbg !43 + %7484 = insertelement <2 x float> poison, float %6967, i64 0, !dbg !43 + %7485 = insertelement <2 x float> %7484, float %6968, i64 1, !dbg !43 + %7486 = fsub <2 x float> %7483, %7485, !dbg !43 + %7487 = insertelement <2 x float> poison, float %3961, i64 0, !dbg !43 + %7488 = insertelement <2 x float> %7487, float %3962, i64 1, !dbg !43 + %7489 = insertelement <2 x float> poison, float %6969, i64 0, !dbg !43 + %7490 = insertelement <2 x float> %7489, float %6970, i64 1, !dbg !43 + %7491 = fsub <2 x float> %7488, %7490, !dbg !43 + %7492 = insertelement <2 x float> poison, float %3963, i64 0, !dbg !43 + %7493 = insertelement <2 x float> %7492, float %3964, i64 1, !dbg !43 + %7494 = insertelement <2 x float> poison, float %6971, i64 0, !dbg !43 + %7495 = insertelement <2 x float> %7494, float %6972, i64 1, !dbg !43 + %7496 = fsub <2 x float> %7493, %7495, !dbg !43 + %7497 = insertelement <2 x float> poison, float %3965, i64 0, !dbg !43 + %7498 = insertelement <2 x float> %7497, float %3966, i64 1, !dbg !43 + %7499 = insertelement <2 x float> poison, float %6973, i64 0, !dbg !43 + %7500 = insertelement <2 x float> %7499, float %6974, i64 1, !dbg !43 + %7501 = fsub <2 x float> %7498, %7500, !dbg !43 + %7502 = insertelement <2 x float> poison, float %3967, i64 0, !dbg !43 + %7503 = insertelement <2 x float> %7502, float %3968, i64 1, !dbg !43 + %7504 = insertelement <2 x float> poison, float %6975, i64 0, !dbg !43 + %7505 = insertelement <2 x float> %7504, float %6976, i64 1, !dbg !43 + %7506 = fsub <2 x float> %7503, %7505, !dbg !43 + %7507 = insertelement <2 x float> poison, float %3969, i64 0, !dbg !43 + %7508 = insertelement <2 x float> %7507, float %3970, i64 1, !dbg !43 + %7509 = insertelement <2 x float> poison, float %6977, i64 0, !dbg !43 + %7510 = insertelement <2 x float> %7509, float %6978, i64 1, !dbg !43 + %7511 = fsub <2 x float> %7508, %7510, !dbg !43 + %7512 = insertelement <2 x float> poison, float %3971, i64 0, !dbg !43 + %7513 = insertelement <2 x float> %7512, float %3972, i64 1, !dbg !43 + %7514 = insertelement <2 x float> poison, float %6979, i64 0, !dbg !43 + %7515 = insertelement <2 x float> %7514, float %6980, i64 1, !dbg !43 + %7516 = fsub <2 x float> %7513, %7515, !dbg !43 + %7517 = insertelement <2 x float> poison, float %3973, i64 0, !dbg !43 + %7518 = insertelement <2 x float> %7517, float %3974, i64 1, !dbg !43 + %7519 = insertelement <2 x float> poison, float %6981, i64 0, !dbg !43 + %7520 = insertelement <2 x float> %7519, float %6982, i64 1, !dbg !43 + %7521 = fsub <2 x float> %7518, %7520, !dbg !43 + %7522 = insertelement <2 x float> poison, float %3975, i64 0, !dbg !43 + %7523 = insertelement <2 x float> %7522, float %3976, i64 1, !dbg !43 + %7524 = insertelement <2 x float> poison, float %6983, i64 0, !dbg !43 + %7525 = insertelement <2 x float> %7524, float %6984, i64 1, !dbg !43 + %7526 = fsub <2 x float> %7523, %7525, !dbg !43 + %7527 = insertelement <2 x float> poison, float %3977, i64 0, !dbg !43 + %7528 = insertelement <2 x float> %7527, float %3978, i64 1, !dbg !43 + %7529 = insertelement <2 x float> poison, float %6985, i64 0, !dbg !43 + %7530 = insertelement <2 x float> %7529, float %6986, i64 1, !dbg !43 + %7531 = fsub <2 x float> %7528, %7530, !dbg !43 + %7532 = insertelement <2 x float> poison, float %3979, i64 0, !dbg !43 + %7533 = insertelement <2 x float> %7532, float %3980, i64 1, !dbg !43 + %7534 = insertelement <2 x float> poison, float %6987, i64 0, !dbg !43 + %7535 = insertelement <2 x float> %7534, float %6988, i64 1, !dbg !43 + %7536 = fsub <2 x float> %7533, %7535, !dbg !43 + %7537 = insertelement <2 x float> poison, float %3981, i64 0, !dbg !43 + %7538 = insertelement <2 x float> %7537, float %3982, i64 1, !dbg !43 + %7539 = insertelement <2 x float> poison, float %6989, i64 0, !dbg !43 + %7540 = insertelement <2 x float> %7539, float %6990, i64 1, !dbg !43 + %7541 = fsub <2 x float> %7538, %7540, !dbg !43 + %7542 = insertelement <2 x float> poison, float %3983, i64 0, !dbg !43 + %7543 = insertelement <2 x float> %7542, float %3984, i64 1, !dbg !43 + %7544 = insertelement <2 x float> poison, float %6991, i64 0, !dbg !43 + %7545 = insertelement <2 x float> %7544, float %6992, i64 1, !dbg !43 + %7546 = fsub <2 x float> %7543, %7545, !dbg !43 + %7547 = insertelement <2 x float> poison, float %3985, i64 0, !dbg !43 + %7548 = insertelement <2 x float> %7547, float %3986, i64 1, !dbg !43 + %7549 = insertelement <2 x float> poison, float %6993, i64 0, !dbg !43 + %7550 = insertelement <2 x float> %7549, float %6994, i64 1, !dbg !43 + %7551 = fsub <2 x float> %7548, %7550, !dbg !43 + %7552 = insertelement <2 x float> poison, float %3987, i64 0, !dbg !43 + %7553 = insertelement <2 x float> %7552, float %3988, i64 1, !dbg !43 + %7554 = insertelement <2 x float> poison, float %6995, i64 0, !dbg !43 + %7555 = insertelement <2 x float> %7554, float %6996, i64 1, !dbg !43 + %7556 = fsub <2 x float> %7553, %7555, !dbg !43 + %7557 = insertelement <2 x float> poison, float %3989, i64 0, !dbg !43 + %7558 = insertelement <2 x float> %7557, float %3990, i64 1, !dbg !43 + %7559 = insertelement <2 x float> poison, float %6997, i64 0, !dbg !43 + %7560 = insertelement <2 x float> %7559, float %6998, i64 1, !dbg !43 + %7561 = fsub <2 x float> %7558, %7560, !dbg !43 + %7562 = insertelement <2 x float> poison, float %3991, i64 0, !dbg !43 + %7563 = insertelement <2 x float> %7562, float %3992, i64 1, !dbg !43 + %7564 = insertelement <2 x float> poison, float %6999, i64 0, !dbg !43 + %7565 = insertelement <2 x float> %7564, float %7000, i64 1, !dbg !43 + %7566 = fsub <2 x float> %7563, %7565, !dbg !43 + %7567 = insertelement <2 x float> poison, float %3993, i64 0, !dbg !43 + %7568 = insertelement <2 x float> %7567, float %3994, i64 1, !dbg !43 + %7569 = insertelement <2 x float> poison, float %7001, i64 0, !dbg !43 + %7570 = insertelement <2 x float> %7569, float %7002, i64 1, !dbg !43 + %7571 = fsub <2 x float> %7568, %7570, !dbg !43 + %7572 = insertelement <2 x float> poison, float %3995, i64 0, !dbg !43 + %7573 = insertelement <2 x float> %7572, float %3996, i64 1, !dbg !43 + %7574 = insertelement <2 x float> poison, float %7003, i64 0, !dbg !43 + %7575 = insertelement <2 x float> %7574, float %7004, i64 1, !dbg !43 + %7576 = fsub <2 x float> %7573, %7575, !dbg !43 + %7577 = insertelement <2 x float> poison, float %3997, i64 0, !dbg !43 + %7578 = insertelement <2 x float> %7577, float %3998, i64 1, !dbg !43 + %7579 = insertelement <2 x float> poison, float %7005, i64 0, !dbg !43 + %7580 = insertelement <2 x float> %7579, float %7006, i64 1, !dbg !43 + %7581 = fsub <2 x float> %7578, %7580, !dbg !43 + %7582 = insertelement <2 x float> poison, float %3999, i64 0, !dbg !43 + %7583 = insertelement <2 x float> %7582, float %4000, i64 1, !dbg !43 + %7584 = insertelement <2 x float> poison, float %7007, i64 0, !dbg !43 + %7585 = insertelement <2 x float> %7584, float %7008, i64 1, !dbg !43 + %7586 = fsub <2 x float> %7583, %7585, !dbg !43 + %7587 = insertelement <2 x float> poison, float %4001, i64 0, !dbg !43 + %7588 = insertelement <2 x float> %7587, float %4002, i64 1, !dbg !43 + %7589 = insertelement <2 x float> poison, float %7009, i64 0, !dbg !43 + %7590 = insertelement <2 x float> %7589, float %7010, i64 1, !dbg !43 + %7591 = fsub <2 x float> %7588, %7590, !dbg !43 + %7592 = insertelement <2 x float> poison, float %4003, i64 0, !dbg !43 + %7593 = insertelement <2 x float> %7592, float %4004, i64 1, !dbg !43 + %7594 = insertelement <2 x float> poison, float %7011, i64 0, !dbg !43 + %7595 = insertelement <2 x float> %7594, float %7012, i64 1, !dbg !43 + %7596 = fsub <2 x float> %7593, %7595, !dbg !43 + %7597 = insertelement <2 x float> poison, float %4005, i64 0, !dbg !43 + %7598 = insertelement <2 x float> %7597, float %4006, i64 1, !dbg !43 + %7599 = insertelement <2 x float> poison, float %7013, i64 0, !dbg !43 + %7600 = insertelement <2 x float> %7599, float %7014, i64 1, !dbg !43 + %7601 = fsub <2 x float> %7598, %7600, !dbg !43 + %7602 = insertelement <2 x float> poison, float %4007, i64 0, !dbg !43 + %7603 = insertelement <2 x float> %7602, float %4008, i64 1, !dbg !43 + %7604 = insertelement <2 x float> poison, float %7015, i64 0, !dbg !43 + %7605 = insertelement <2 x float> %7604, float %7016, i64 1, !dbg !43 + %7606 = fsub <2 x float> %7603, %7605, !dbg !43 + %7607 = insertelement <2 x float> poison, float %4009, i64 0, !dbg !43 + %7608 = insertelement <2 x float> %7607, float %4010, i64 1, !dbg !43 + %7609 = insertelement <2 x float> poison, float %7017, i64 0, !dbg !43 + %7610 = insertelement <2 x float> %7609, float %7018, i64 1, !dbg !43 + %7611 = fsub <2 x float> %7608, %7610, !dbg !43 + %7612 = insertelement <2 x float> poison, float %4011, i64 0, !dbg !43 + %7613 = insertelement <2 x float> %7612, float %4012, i64 1, !dbg !43 + %7614 = insertelement <2 x float> poison, float %7019, i64 0, !dbg !43 + %7615 = insertelement <2 x float> %7614, float %7020, i64 1, !dbg !43 + %7616 = fsub <2 x float> %7613, %7615, !dbg !43 + %7617 = insertelement <2 x float> poison, float %4013, i64 0, !dbg !43 + %7618 = insertelement <2 x float> %7617, float %4014, i64 1, !dbg !43 + %7619 = insertelement <2 x float> poison, float %7021, i64 0, !dbg !43 + %7620 = insertelement <2 x float> %7619, float %7022, i64 1, !dbg !43 + %7621 = fsub <2 x float> %7618, %7620, !dbg !43 + %7622 = insertelement <2 x float> poison, float %4015, i64 0, !dbg !43 + %7623 = insertelement <2 x float> %7622, float %4016, i64 1, !dbg !43 + %7624 = insertelement <2 x float> poison, float %7023, i64 0, !dbg !43 + %7625 = insertelement <2 x float> %7624, float %7024, i64 1, !dbg !43 + %7626 = fsub <2 x float> %7623, %7625, !dbg !43 + %7627 = insertelement <2 x float> poison, float %4017, i64 0, !dbg !43 + %7628 = insertelement <2 x float> %7627, float %4018, i64 1, !dbg !43 + %7629 = insertelement <2 x float> poison, float %7025, i64 0, !dbg !43 + %7630 = insertelement <2 x float> %7629, float %7026, i64 1, !dbg !43 + %7631 = fsub <2 x float> %7628, %7630, !dbg !43 + %7632 = insertelement <2 x float> poison, float %4019, i64 0, !dbg !43 + %7633 = insertelement <2 x float> %7632, float %4020, i64 1, !dbg !43 + %7634 = insertelement <2 x float> poison, float %7027, i64 0, !dbg !43 + %7635 = insertelement <2 x float> %7634, float %7028, i64 1, !dbg !43 + %7636 = fsub <2 x float> %7633, %7635, !dbg !43 + %7637 = insertelement <2 x float> poison, float %4021, i64 0, !dbg !43 + %7638 = insertelement <2 x float> %7637, float %4022, i64 1, !dbg !43 + %7639 = insertelement <2 x float> poison, float %7029, i64 0, !dbg !43 + %7640 = insertelement <2 x float> %7639, float %7030, i64 1, !dbg !43 + %7641 = fsub <2 x float> %7638, %7640, !dbg !43 + %7642 = insertelement <2 x float> poison, float %4023, i64 0, !dbg !43 + %7643 = insertelement <2 x float> %7642, float %4024, i64 1, !dbg !43 + %7644 = insertelement <2 x float> poison, float %7031, i64 0, !dbg !43 + %7645 = insertelement <2 x float> %7644, float %7032, i64 1, !dbg !43 + %7646 = fsub <2 x float> %7643, %7645, !dbg !43 + %7647 = insertelement <2 x float> poison, float %4025, i64 0, !dbg !43 + %7648 = insertelement <2 x float> %7647, float %4026, i64 1, !dbg !43 + %7649 = insertelement <2 x float> poison, float %7033, i64 0, !dbg !43 + %7650 = insertelement <2 x float> %7649, float %7034, i64 1, !dbg !43 + %7651 = fsub <2 x float> %7648, %7650, !dbg !43 + %7652 = insertelement <2 x float> poison, float %4027, i64 0, !dbg !43 + %7653 = insertelement <2 x float> %7652, float %4028, i64 1, !dbg !43 + %7654 = insertelement <2 x float> poison, float %7035, i64 0, !dbg !43 + %7655 = insertelement <2 x float> %7654, float %7036, i64 1, !dbg !43 + %7656 = fsub <2 x float> %7653, %7655, !dbg !43 + %7657 = insertelement <2 x float> poison, float %4029, i64 0, !dbg !43 + %7658 = insertelement <2 x float> %7657, float %4030, i64 1, !dbg !43 + %7659 = insertelement <2 x float> poison, float %7037, i64 0, !dbg !43 + %7660 = insertelement <2 x float> %7659, float %7038, i64 1, !dbg !43 + %7661 = fsub <2 x float> %7658, %7660, !dbg !43 + %7662 = insertelement <2 x float> poison, float %4031, i64 0, !dbg !43 + %7663 = insertelement <2 x float> %7662, float %4032, i64 1, !dbg !43 + %7664 = insertelement <2 x float> poison, float %7039, i64 0, !dbg !43 + %7665 = insertelement <2 x float> %7664, float %7040, i64 1, !dbg !43 + %7666 = fsub <2 x float> %7663, %7665, !dbg !43 + %7667 = insertelement <2 x float> poison, float %4033, i64 0, !dbg !43 + %7668 = insertelement <2 x float> %7667, float %4034, i64 1, !dbg !43 + %7669 = insertelement <2 x float> poison, float %7041, i64 0, !dbg !43 + %7670 = insertelement <2 x float> %7669, float %7042, i64 1, !dbg !43 + %7671 = fsub <2 x float> %7668, %7670, !dbg !43 + %7672 = insertelement <2 x float> poison, float %4035, i64 0, !dbg !43 + %7673 = insertelement <2 x float> %7672, float %4036, i64 1, !dbg !43 + %7674 = insertelement <2 x float> poison, float %7043, i64 0, !dbg !43 + %7675 = insertelement <2 x float> %7674, float %7044, i64 1, !dbg !43 + %7676 = fsub <2 x float> %7673, %7675, !dbg !43 + %7677 = insertelement <2 x float> poison, float %4037, i64 0, !dbg !43 + %7678 = insertelement <2 x float> %7677, float %4038, i64 1, !dbg !43 + %7679 = insertelement <2 x float> poison, float %7045, i64 0, !dbg !43 + %7680 = insertelement <2 x float> %7679, float %7046, i64 1, !dbg !43 + %7681 = fsub <2 x float> %7678, %7680, !dbg !43 + %7682 = insertelement <2 x float> poison, float %4039, i64 0, !dbg !43 + %7683 = insertelement <2 x float> %7682, float %4040, i64 1, !dbg !43 + %7684 = insertelement <2 x float> poison, float %7047, i64 0, !dbg !43 + %7685 = insertelement <2 x float> %7684, float %7048, i64 1, !dbg !43 + %7686 = fsub <2 x float> %7683, %7685, !dbg !43 + %7687 = insertelement <2 x float> poison, float %4041, i64 0, !dbg !43 + %7688 = insertelement <2 x float> %7687, float %4042, i64 1, !dbg !43 + %7689 = insertelement <2 x float> poison, float %7049, i64 0, !dbg !43 + %7690 = insertelement <2 x float> %7689, float %7050, i64 1, !dbg !43 + %7691 = fsub <2 x float> %7688, %7690, !dbg !43 + %7692 = insertelement <2 x float> poison, float %4043, i64 0, !dbg !43 + %7693 = insertelement <2 x float> %7692, float %4044, i64 1, !dbg !43 + %7694 = insertelement <2 x float> poison, float %7051, i64 0, !dbg !43 + %7695 = insertelement <2 x float> %7694, float %7052, i64 1, !dbg !43 + %7696 = fsub <2 x float> %7693, %7695, !dbg !43 + %7697 = insertelement <2 x float> poison, float %4045, i64 0, !dbg !43 + %7698 = insertelement <2 x float> %7697, float %4046, i64 1, !dbg !43 + %7699 = insertelement <2 x float> poison, float %7053, i64 0, !dbg !43 + %7700 = insertelement <2 x float> %7699, float %7054, i64 1, !dbg !43 + %7701 = fsub <2 x float> %7698, %7700, !dbg !43 + %7702 = insertelement <2 x float> poison, float %4047, i64 0, !dbg !43 + %7703 = insertelement <2 x float> %7702, float %4048, i64 1, !dbg !43 + %7704 = insertelement <2 x float> poison, float %7055, i64 0, !dbg !43 + %7705 = insertelement <2 x float> %7704, float %7056, i64 1, !dbg !43 + %7706 = fsub <2 x float> %7703, %7705, !dbg !43 + %7707 = insertelement <2 x float> poison, float %4049, i64 0, !dbg !43 + %7708 = insertelement <2 x float> %7707, float %4050, i64 1, !dbg !43 + %7709 = insertelement <2 x float> poison, float %7057, i64 0, !dbg !43 + %7710 = insertelement <2 x float> %7709, float %7058, i64 1, !dbg !43 + %7711 = fsub <2 x float> %7708, %7710, !dbg !43 + %7712 = insertelement <2 x float> poison, float %4051, i64 0, !dbg !43 + %7713 = insertelement <2 x float> %7712, float %4052, i64 1, !dbg !43 + %7714 = insertelement <2 x float> poison, float %7059, i64 0, !dbg !43 + %7715 = insertelement <2 x float> %7714, float %7060, i64 1, !dbg !43 + %7716 = fsub <2 x float> %7713, %7715, !dbg !43 + %7717 = insertelement <2 x float> poison, float %4053, i64 0, !dbg !43 + %7718 = insertelement <2 x float> %7717, float %4054, i64 1, !dbg !43 + %7719 = insertelement <2 x float> poison, float %7061, i64 0, !dbg !43 + %7720 = insertelement <2 x float> %7719, float %7062, i64 1, !dbg !43 + %7721 = fsub <2 x float> %7718, %7720, !dbg !43 + %7722 = insertelement <2 x float> poison, float %4055, i64 0, !dbg !43 + %7723 = insertelement <2 x float> %7722, float %4056, i64 1, !dbg !43 + %7724 = insertelement <2 x float> poison, float %7063, i64 0, !dbg !43 + %7725 = insertelement <2 x float> %7724, float %7064, i64 1, !dbg !43 + %7726 = fsub <2 x float> %7723, %7725, !dbg !43 + %7727 = insertelement <2 x float> poison, float %4057, i64 0, !dbg !43 + %7728 = insertelement <2 x float> %7727, float %4058, i64 1, !dbg !43 + %7729 = insertelement <2 x float> poison, float %7065, i64 0, !dbg !43 + %7730 = insertelement <2 x float> %7729, float %7066, i64 1, !dbg !43 + %7731 = fsub <2 x float> %7728, %7730, !dbg !43 + %7732 = insertelement <2 x float> poison, float %4059, i64 0, !dbg !43 + %7733 = insertelement <2 x float> %7732, float %4060, i64 1, !dbg !43 + %7734 = insertelement <2 x float> poison, float %7067, i64 0, !dbg !43 + %7735 = insertelement <2 x float> %7734, float %7068, i64 1, !dbg !43 + %7736 = fsub <2 x float> %7733, %7735, !dbg !43 + %7737 = insertelement <2 x float> poison, float %4061, i64 0, !dbg !43 + %7738 = insertelement <2 x float> %7737, float %4062, i64 1, !dbg !43 + %7739 = insertelement <2 x float> poison, float %7069, i64 0, !dbg !43 + %7740 = insertelement <2 x float> %7739, float %7070, i64 1, !dbg !43 + %7741 = fsub <2 x float> %7738, %7740, !dbg !43 + %7742 = insertelement <2 x float> poison, float %4063, i64 0, !dbg !43 + %7743 = insertelement <2 x float> %7742, float %4064, i64 1, !dbg !43 + %7744 = insertelement <2 x float> poison, float %7071, i64 0, !dbg !43 + %7745 = insertelement <2 x float> %7744, float %7072, i64 1, !dbg !43 + %7746 = fsub <2 x float> %7743, %7745, !dbg !43 + %7747 = insertelement <2 x float> poison, float %4065, i64 0, !dbg !43 + %7748 = insertelement <2 x float> %7747, float %4066, i64 1, !dbg !43 + %7749 = insertelement <2 x float> poison, float %7073, i64 0, !dbg !43 + %7750 = insertelement <2 x float> %7749, float %7074, i64 1, !dbg !43 + %7751 = fsub <2 x float> %7748, %7750, !dbg !43 + %7752 = insertelement <2 x float> poison, float %4067, i64 0, !dbg !43 + %7753 = insertelement <2 x float> %7752, float %4068, i64 1, !dbg !43 + %7754 = insertelement <2 x float> poison, float %7075, i64 0, !dbg !43 + %7755 = insertelement <2 x float> %7754, float %7076, i64 1, !dbg !43 + %7756 = fsub <2 x float> %7753, %7755, !dbg !43 + %7757 = insertelement <2 x float> poison, float %4069, i64 0, !dbg !43 + %7758 = insertelement <2 x float> %7757, float %4070, i64 1, !dbg !43 + %7759 = insertelement <2 x float> poison, float %7077, i64 0, !dbg !43 + %7760 = insertelement <2 x float> %7759, float %7078, i64 1, !dbg !43 + %7761 = fsub <2 x float> %7758, %7760, !dbg !43 + %7762 = insertelement <2 x float> poison, float %4071, i64 0, !dbg !43 + %7763 = insertelement <2 x float> %7762, float %4072, i64 1, !dbg !43 + %7764 = insertelement <2 x float> poison, float %7079, i64 0, !dbg !43 + %7765 = insertelement <2 x float> %7764, float %7080, i64 1, !dbg !43 + %7766 = fsub <2 x float> %7763, %7765, !dbg !43 + %7767 = insertelement <2 x float> poison, float %4073, i64 0, !dbg !43 + %7768 = insertelement <2 x float> %7767, float %4074, i64 1, !dbg !43 + %7769 = insertelement <2 x float> poison, float %7081, i64 0, !dbg !43 + %7770 = insertelement <2 x float> %7769, float %7082, i64 1, !dbg !43 + %7771 = fsub <2 x float> %7768, %7770, !dbg !43 + %7772 = insertelement <2 x float> poison, float %4075, i64 0, !dbg !43 + %7773 = insertelement <2 x float> %7772, float %4076, i64 1, !dbg !43 + %7774 = insertelement <2 x float> poison, float %7083, i64 0, !dbg !43 + %7775 = insertelement <2 x float> %7774, float %7084, i64 1, !dbg !43 + %7776 = fsub <2 x float> %7773, %7775, !dbg !43 + %7777 = insertelement <2 x float> poison, float %4077, i64 0, !dbg !43 + %7778 = insertelement <2 x float> %7777, float %4078, i64 1, !dbg !43 + %7779 = insertelement <2 x float> poison, float %7085, i64 0, !dbg !43 + %7780 = insertelement <2 x float> %7779, float %7086, i64 1, !dbg !43 + %7781 = fsub <2 x float> %7778, %7780, !dbg !43 + %7782 = insertelement <2 x float> poison, float %4079, i64 0, !dbg !43 + %7783 = insertelement <2 x float> %7782, float %4080, i64 1, !dbg !43 + %7784 = insertelement <2 x float> poison, float %7087, i64 0, !dbg !43 + %7785 = insertelement <2 x float> %7784, float %7088, i64 1, !dbg !43 + %7786 = fsub <2 x float> %7783, %7785, !dbg !43 + %7787 = insertelement <2 x float> poison, float %4081, i64 0, !dbg !43 + %7788 = insertelement <2 x float> %7787, float %4082, i64 1, !dbg !43 + %7789 = insertelement <2 x float> poison, float %7089, i64 0, !dbg !43 + %7790 = insertelement <2 x float> %7789, float %7090, i64 1, !dbg !43 + %7791 = fsub <2 x float> %7788, %7790, !dbg !43 + %7792 = insertelement <2 x float> poison, float %4083, i64 0, !dbg !43 + %7793 = insertelement <2 x float> %7792, float %4084, i64 1, !dbg !43 + %7794 = insertelement <2 x float> poison, float %7091, i64 0, !dbg !43 + %7795 = insertelement <2 x float> %7794, float %7092, i64 1, !dbg !43 + %7796 = fsub <2 x float> %7793, %7795, !dbg !43 + %7797 = insertelement <2 x float> poison, float %4085, i64 0, !dbg !43 + %7798 = insertelement <2 x float> %7797, float %4086, i64 1, !dbg !43 + %7799 = insertelement <2 x float> poison, float %7093, i64 0, !dbg !43 + %7800 = insertelement <2 x float> %7799, float %7094, i64 1, !dbg !43 + %7801 = fsub <2 x float> %7798, %7800, !dbg !43 + %7802 = insertelement <2 x float> poison, float %4087, i64 0, !dbg !43 + %7803 = insertelement <2 x float> %7802, float %4088, i64 1, !dbg !43 + %7804 = insertelement <2 x float> poison, float %7095, i64 0, !dbg !43 + %7805 = insertelement <2 x float> %7804, float %7096, i64 1, !dbg !43 + %7806 = fsub <2 x float> %7803, %7805, !dbg !43 + %7807 = insertelement <2 x float> poison, float %4089, i64 0, !dbg !43 + %7808 = insertelement <2 x float> %7807, float %4090, i64 1, !dbg !43 + %7809 = insertelement <2 x float> poison, float %7097, i64 0, !dbg !43 + %7810 = insertelement <2 x float> %7809, float %7098, i64 1, !dbg !43 + %7811 = fsub <2 x float> %7808, %7810, !dbg !43 + %7812 = insertelement <2 x float> poison, float %4091, i64 0, !dbg !43 + %7813 = insertelement <2 x float> %7812, float %4092, i64 1, !dbg !43 + %7814 = insertelement <2 x float> poison, float %7099, i64 0, !dbg !43 + %7815 = insertelement <2 x float> %7814, float %7100, i64 1, !dbg !43 + %7816 = fsub <2 x float> %7813, %7815, !dbg !43 + %7817 = insertelement <2 x float> poison, float %4093, i64 0, !dbg !43 + %7818 = insertelement <2 x float> %7817, float %4094, i64 1, !dbg !43 + %7819 = insertelement <2 x float> poison, float %7101, i64 0, !dbg !43 + %7820 = insertelement <2 x float> %7819, float %7102, i64 1, !dbg !43 + %7821 = fsub <2 x float> %7818, %7820, !dbg !43 + %7822 = insertelement <2 x float> poison, float %4095, i64 0, !dbg !43 + %7823 = insertelement <2 x float> %7822, float %4096, i64 1, !dbg !43 + %7824 = insertelement <2 x float> poison, float %7103, i64 0, !dbg !43 + %7825 = insertelement <2 x float> %7824, float %7104, i64 1, !dbg !43 + %7826 = fsub <2 x float> %7823, %7825, !dbg !43 + %7827 = insertelement <2 x float> poison, float %4097, i64 0, !dbg !43 + %7828 = insertelement <2 x float> %7827, float %4098, i64 1, !dbg !43 + %7829 = insertelement <2 x float> poison, float %7105, i64 0, !dbg !43 + %7830 = insertelement <2 x float> %7829, float %7106, i64 1, !dbg !43 + %7831 = fsub <2 x float> %7828, %7830, !dbg !43 + %7832 = insertelement <2 x float> poison, float %4099, i64 0, !dbg !43 + %7833 = insertelement <2 x float> %7832, float %4100, i64 1, !dbg !43 + %7834 = insertelement <2 x float> poison, float %7107, i64 0, !dbg !43 + %7835 = insertelement <2 x float> %7834, float %7108, i64 1, !dbg !43 + %7836 = fsub <2 x float> %7833, %7835, !dbg !43 + %7837 = insertelement <2 x float> poison, float %4101, i64 0, !dbg !43 + %7838 = insertelement <2 x float> %7837, float %4102, i64 1, !dbg !43 + %7839 = insertelement <2 x float> poison, float %7109, i64 0, !dbg !43 + %7840 = insertelement <2 x float> %7839, float %7110, i64 1, !dbg !43 + %7841 = fsub <2 x float> %7838, %7840, !dbg !43 + %7842 = insertelement <2 x float> poison, float %4103, i64 0, !dbg !43 + %7843 = insertelement <2 x float> %7842, float %4104, i64 1, !dbg !43 + %7844 = insertelement <2 x float> poison, float %7111, i64 0, !dbg !43 + %7845 = insertelement <2 x float> %7844, float %7112, i64 1, !dbg !43 + %7846 = fsub <2 x float> %7843, %7845, !dbg !43 + %7847 = insertelement <2 x float> poison, float %4105, i64 0, !dbg !43 + %7848 = insertelement <2 x float> %7847, float %4106, i64 1, !dbg !43 + %7849 = insertelement <2 x float> poison, float %7113, i64 0, !dbg !43 + %7850 = insertelement <2 x float> %7849, float %7114, i64 1, !dbg !43 + %7851 = fsub <2 x float> %7848, %7850, !dbg !43 + %7852 = insertelement <2 x float> poison, float %4107, i64 0, !dbg !43 + %7853 = insertelement <2 x float> %7852, float %4108, i64 1, !dbg !43 + %7854 = insertelement <2 x float> poison, float %7115, i64 0, !dbg !43 + %7855 = insertelement <2 x float> %7854, float %7116, i64 1, !dbg !43 + %7856 = fsub <2 x float> %7853, %7855, !dbg !43 + %7857 = insertelement <2 x float> poison, float %4109, i64 0, !dbg !43 + %7858 = insertelement <2 x float> %7857, float %4110, i64 1, !dbg !43 + %7859 = insertelement <2 x float> poison, float %7117, i64 0, !dbg !43 + %7860 = insertelement <2 x float> %7859, float %7118, i64 1, !dbg !43 + %7861 = fsub <2 x float> %7858, %7860, !dbg !43 + %7862 = insertelement <2 x float> poison, float %4111, i64 0, !dbg !43 + %7863 = insertelement <2 x float> %7862, float %4112, i64 1, !dbg !43 + %7864 = insertelement <2 x float> poison, float %7119, i64 0, !dbg !43 + %7865 = insertelement <2 x float> %7864, float %7120, i64 1, !dbg !43 + %7866 = fsub <2 x float> %7863, %7865, !dbg !43 + %7867 = insertelement <2 x float> poison, float %4113, i64 0, !dbg !43 + %7868 = insertelement <2 x float> %7867, float %4114, i64 1, !dbg !43 + %7869 = insertelement <2 x float> poison, float %7121, i64 0, !dbg !43 + %7870 = insertelement <2 x float> %7869, float %7122, i64 1, !dbg !43 + %7871 = fsub <2 x float> %7868, %7870, !dbg !43 + %7872 = insertelement <2 x float> poison, float %4115, i64 0, !dbg !43 + %7873 = insertelement <2 x float> %7872, float %4116, i64 1, !dbg !43 + %7874 = insertelement <2 x float> poison, float %7123, i64 0, !dbg !43 + %7875 = insertelement <2 x float> %7874, float %7124, i64 1, !dbg !43 + %7876 = fsub <2 x float> %7873, %7875, !dbg !43 + %7877 = insertelement <2 x float> poison, float %4117, i64 0, !dbg !43 + %7878 = insertelement <2 x float> %7877, float %4118, i64 1, !dbg !43 + %7879 = insertelement <2 x float> poison, float %7125, i64 0, !dbg !43 + %7880 = insertelement <2 x float> %7879, float %7126, i64 1, !dbg !43 + %7881 = fsub <2 x float> %7878, %7880, !dbg !43 + %7882 = insertelement <2 x float> poison, float %4119, i64 0, !dbg !43 + %7883 = insertelement <2 x float> %7882, float %4120, i64 1, !dbg !43 + %7884 = insertelement <2 x float> poison, float %7127, i64 0, !dbg !43 + %7885 = insertelement <2 x float> %7884, float %7128, i64 1, !dbg !43 + %7886 = fsub <2 x float> %7883, %7885, !dbg !43 + %7887 = insertelement <2 x float> poison, float %4121, i64 0, !dbg !43 + %7888 = insertelement <2 x float> %7887, float %4122, i64 1, !dbg !43 + %7889 = insertelement <2 x float> poison, float %7129, i64 0, !dbg !43 + %7890 = insertelement <2 x float> %7889, float %7130, i64 1, !dbg !43 + %7891 = fsub <2 x float> %7888, %7890, !dbg !43 + %7892 = insertelement <2 x float> poison, float %4123, i64 0, !dbg !43 + %7893 = insertelement <2 x float> %7892, float %4124, i64 1, !dbg !43 + %7894 = insertelement <2 x float> poison, float %7131, i64 0, !dbg !43 + %7895 = insertelement <2 x float> %7894, float %7132, i64 1, !dbg !43 + %7896 = fsub <2 x float> %7893, %7895, !dbg !43 + %7897 = insertelement <2 x float> poison, float %4125, i64 0, !dbg !43 + %7898 = insertelement <2 x float> %7897, float %4126, i64 1, !dbg !43 + %7899 = insertelement <2 x float> poison, float %7133, i64 0, !dbg !43 + %7900 = insertelement <2 x float> %7899, float %7134, i64 1, !dbg !43 + %7901 = fsub <2 x float> %7898, %7900, !dbg !43 + %7902 = insertelement <2 x float> poison, float %4127, i64 0, !dbg !43 + %7903 = insertelement <2 x float> %7902, float %4128, i64 1, !dbg !43 + %7904 = insertelement <2 x float> poison, float %7135, i64 0, !dbg !43 + %7905 = insertelement <2 x float> %7904, float %7136, i64 1, !dbg !43 + %7906 = fsub <2 x float> %7903, %7905, !dbg !43 + %7907 = insertelement <2 x float> poison, float %4129, i64 0, !dbg !43 + %7908 = insertelement <2 x float> %7907, float %4130, i64 1, !dbg !43 + %7909 = insertelement <2 x float> poison, float %7137, i64 0, !dbg !43 + %7910 = insertelement <2 x float> %7909, float %7138, i64 1, !dbg !43 + %7911 = fsub <2 x float> %7908, %7910, !dbg !43 + %7912 = insertelement <2 x float> poison, float %4131, i64 0, !dbg !43 + %7913 = insertelement <2 x float> %7912, float %4132, i64 1, !dbg !43 + %7914 = insertelement <2 x float> poison, float %7139, i64 0, !dbg !43 + %7915 = insertelement <2 x float> %7914, float %7140, i64 1, !dbg !43 + %7916 = fsub <2 x float> %7913, %7915, !dbg !43 + %7917 = insertelement <2 x float> poison, float %4133, i64 0, !dbg !43 + %7918 = insertelement <2 x float> %7917, float %4134, i64 1, !dbg !43 + %7919 = insertelement <2 x float> poison, float %7141, i64 0, !dbg !43 + %7920 = insertelement <2 x float> %7919, float %7142, i64 1, !dbg !43 + %7921 = fsub <2 x float> %7918, %7920, !dbg !43 + %7922 = insertelement <2 x float> poison, float %4135, i64 0, !dbg !43 + %7923 = insertelement <2 x float> %7922, float %4136, i64 1, !dbg !43 + %7924 = insertelement <2 x float> poison, float %7143, i64 0, !dbg !43 + %7925 = insertelement <2 x float> %7924, float %7144, i64 1, !dbg !43 + %7926 = fsub <2 x float> %7923, %7925, !dbg !43 + %7927 = insertelement <2 x float> poison, float %4137, i64 0, !dbg !43 + %7928 = insertelement <2 x float> %7927, float %4138, i64 1, !dbg !43 + %7929 = insertelement <2 x float> poison, float %7145, i64 0, !dbg !43 + %7930 = insertelement <2 x float> %7929, float %7146, i64 1, !dbg !43 + %7931 = fsub <2 x float> %7928, %7930, !dbg !43 + %7932 = insertelement <2 x float> poison, float %4139, i64 0, !dbg !43 + %7933 = insertelement <2 x float> %7932, float %4140, i64 1, !dbg !43 + %7934 = insertelement <2 x float> poison, float %7147, i64 0, !dbg !43 + %7935 = insertelement <2 x float> %7934, float %7148, i64 1, !dbg !43 + %7936 = fsub <2 x float> %7933, %7935, !dbg !43 + %7937 = insertelement <2 x float> poison, float %4141, i64 0, !dbg !43 + %7938 = insertelement <2 x float> %7937, float %4142, i64 1, !dbg !43 + %7939 = insertelement <2 x float> poison, float %7149, i64 0, !dbg !43 + %7940 = insertelement <2 x float> %7939, float %7150, i64 1, !dbg !43 + %7941 = fsub <2 x float> %7938, %7940, !dbg !43 + %7942 = insertelement <2 x float> poison, float %4143, i64 0, !dbg !43 + %7943 = insertelement <2 x float> %7942, float %4144, i64 1, !dbg !43 + %7944 = insertelement <2 x float> poison, float %7151, i64 0, !dbg !43 + %7945 = insertelement <2 x float> %7944, float %7152, i64 1, !dbg !43 + %7946 = fsub <2 x float> %7943, %7945, !dbg !43 + %7947 = insertelement <2 x float> poison, float %4145, i64 0, !dbg !43 + %7948 = insertelement <2 x float> %7947, float %4146, i64 1, !dbg !43 + %7949 = insertelement <2 x float> poison, float %7153, i64 0, !dbg !43 + %7950 = insertelement <2 x float> %7949, float %7154, i64 1, !dbg !43 + %7951 = fsub <2 x float> %7948, %7950, !dbg !43 + %7952 = insertelement <2 x float> poison, float %4147, i64 0, !dbg !43 + %7953 = insertelement <2 x float> %7952, float %4148, i64 1, !dbg !43 + %7954 = insertelement <2 x float> poison, float %7155, i64 0, !dbg !43 + %7955 = insertelement <2 x float> %7954, float %7156, i64 1, !dbg !43 + %7956 = fsub <2 x float> %7953, %7955, !dbg !43 + %7957 = insertelement <2 x float> poison, float %4149, i64 0, !dbg !43 + %7958 = insertelement <2 x float> %7957, float %4150, i64 1, !dbg !43 + %7959 = insertelement <2 x float> poison, float %7157, i64 0, !dbg !43 + %7960 = insertelement <2 x float> %7959, float %7158, i64 1, !dbg !43 + %7961 = fsub <2 x float> %7958, %7960, !dbg !43 + %7962 = insertelement <2 x float> poison, float %4151, i64 0, !dbg !43 + %7963 = insertelement <2 x float> %7962, float %4152, i64 1, !dbg !43 + %7964 = insertelement <2 x float> poison, float %7159, i64 0, !dbg !43 + %7965 = insertelement <2 x float> %7964, float %7160, i64 1, !dbg !43 + %7966 = fsub <2 x float> %7963, %7965, !dbg !43 + %7967 = insertelement <2 x float> poison, float %4153, i64 0, !dbg !43 + %7968 = insertelement <2 x float> %7967, float %4154, i64 1, !dbg !43 + %7969 = insertelement <2 x float> poison, float %7161, i64 0, !dbg !43 + %7970 = insertelement <2 x float> %7969, float %7162, i64 1, !dbg !43 + %7971 = fsub <2 x float> %7968, %7970, !dbg !43 + %7972 = insertelement <2 x float> poison, float %4155, i64 0, !dbg !43 + %7973 = insertelement <2 x float> %7972, float %4156, i64 1, !dbg !43 + %7974 = insertelement <2 x float> poison, float %7163, i64 0, !dbg !43 + %7975 = insertelement <2 x float> %7974, float %7164, i64 1, !dbg !43 + %7976 = fsub <2 x float> %7973, %7975, !dbg !43 + %7977 = insertelement <2 x float> poison, float %4157, i64 0, !dbg !43 + %7978 = insertelement <2 x float> %7977, float %4158, i64 1, !dbg !43 + %7979 = insertelement <2 x float> poison, float %7165, i64 0, !dbg !43 + %7980 = insertelement <2 x float> %7979, float %7166, i64 1, !dbg !43 + %7981 = fsub <2 x float> %7978, %7980, !dbg !43 + %7982 = insertelement <2 x float> poison, float %4159, i64 0, !dbg !43 + %7983 = insertelement <2 x float> %7982, float %4160, i64 1, !dbg !43 + %7984 = insertelement <2 x float> poison, float %7167, i64 0, !dbg !43 + %7985 = insertelement <2 x float> %7984, float %7168, i64 1, !dbg !43 + %7986 = fsub <2 x float> %7983, %7985, !dbg !43 + %7987 = insertelement <2 x float> poison, float %4161, i64 0, !dbg !43 + %7988 = insertelement <2 x float> %7987, float %4162, i64 1, !dbg !43 + %7989 = insertelement <2 x float> poison, float %7169, i64 0, !dbg !43 + %7990 = insertelement <2 x float> %7989, float %7170, i64 1, !dbg !43 + %7991 = fsub <2 x float> %7988, %7990, !dbg !43 + %7992 = insertelement <2 x float> poison, float %4163, i64 0, !dbg !43 + %7993 = insertelement <2 x float> %7992, float %4164, i64 1, !dbg !43 + %7994 = insertelement <2 x float> poison, float %7171, i64 0, !dbg !43 + %7995 = insertelement <2 x float> %7994, float %7172, i64 1, !dbg !43 + %7996 = fsub <2 x float> %7993, %7995, !dbg !43 + %7997 = insertelement <2 x float> poison, float %4165, i64 0, !dbg !43 + %7998 = insertelement <2 x float> %7997, float %4166, i64 1, !dbg !43 + %7999 = insertelement <2 x float> poison, float %7173, i64 0, !dbg !43 + %8000 = insertelement <2 x float> %7999, float %7174, i64 1, !dbg !43 + %8001 = fsub <2 x float> %7998, %8000, !dbg !43 + %8002 = insertelement <2 x float> poison, float %4167, i64 0, !dbg !43 + %8003 = insertelement <2 x float> %8002, float %4168, i64 1, !dbg !43 + %8004 = insertelement <2 x float> poison, float %7175, i64 0, !dbg !43 + %8005 = insertelement <2 x float> %8004, float %7176, i64 1, !dbg !43 + %8006 = fsub <2 x float> %8003, %8005, !dbg !43 + %8007 = insertelement <2 x float> poison, float %4169, i64 0, !dbg !43 + %8008 = insertelement <2 x float> %8007, float %4170, i64 1, !dbg !43 + %8009 = insertelement <2 x float> poison, float %7177, i64 0, !dbg !43 + %8010 = insertelement <2 x float> %8009, float %7178, i64 1, !dbg !43 + %8011 = fsub <2 x float> %8008, %8010, !dbg !43 + %8012 = insertelement <2 x float> poison, float %4171, i64 0, !dbg !43 + %8013 = insertelement <2 x float> %8012, float %4172, i64 1, !dbg !43 + %8014 = insertelement <2 x float> poison, float %7179, i64 0, !dbg !43 + %8015 = insertelement <2 x float> %8014, float %7180, i64 1, !dbg !43 + %8016 = fsub <2 x float> %8013, %8015, !dbg !43 + %8017 = insertelement <2 x float> poison, float %4173, i64 0, !dbg !43 + %8018 = insertelement <2 x float> %8017, float %4174, i64 1, !dbg !43 + %8019 = insertelement <2 x float> poison, float %7181, i64 0, !dbg !43 + %8020 = insertelement <2 x float> %8019, float %7182, i64 1, !dbg !43 + %8021 = fsub <2 x float> %8018, %8020, !dbg !43 + %8022 = insertelement <2 x float> poison, float %4175, i64 0, !dbg !43 + %8023 = insertelement <2 x float> %8022, float %4176, i64 1, !dbg !43 + %8024 = insertelement <2 x float> poison, float %7183, i64 0, !dbg !43 + %8025 = insertelement <2 x float> %8024, float %7184, i64 1, !dbg !43 + %8026 = fsub <2 x float> %8023, %8025, !dbg !43 + %8027 = insertelement <2 x float> poison, float %4177, i64 0, !dbg !43 + %8028 = insertelement <2 x float> %8027, float %4178, i64 1, !dbg !43 + %8029 = insertelement <2 x float> poison, float %7185, i64 0, !dbg !43 + %8030 = insertelement <2 x float> %8029, float %7186, i64 1, !dbg !43 + %8031 = fsub <2 x float> %8028, %8030, !dbg !43 + %8032 = insertelement <2 x float> poison, float %4179, i64 0, !dbg !43 + %8033 = insertelement <2 x float> %8032, float %4180, i64 1, !dbg !43 + %8034 = insertelement <2 x float> poison, float %7187, i64 0, !dbg !43 + %8035 = insertelement <2 x float> %8034, float %7188, i64 1, !dbg !43 + %8036 = fsub <2 x float> %8033, %8035, !dbg !43 + %8037 = insertelement <2 x float> poison, float %4181, i64 0, !dbg !43 + %8038 = insertelement <2 x float> %8037, float %4182, i64 1, !dbg !43 + %8039 = insertelement <2 x float> poison, float %7189, i64 0, !dbg !43 + %8040 = insertelement <2 x float> %8039, float %7190, i64 1, !dbg !43 + %8041 = fsub <2 x float> %8038, %8040, !dbg !43 + %8042 = insertelement <2 x float> poison, float %4183, i64 0, !dbg !43 + %8043 = insertelement <2 x float> %8042, float %4184, i64 1, !dbg !43 + %8044 = insertelement <2 x float> poison, float %7191, i64 0, !dbg !43 + %8045 = insertelement <2 x float> %8044, float %7192, i64 1, !dbg !43 + %8046 = fsub <2 x float> %8043, %8045, !dbg !43 + %8047 = insertelement <2 x float> poison, float %4185, i64 0, !dbg !43 + %8048 = insertelement <2 x float> %8047, float %4186, i64 1, !dbg !43 + %8049 = insertelement <2 x float> poison, float %7193, i64 0, !dbg !43 + %8050 = insertelement <2 x float> %8049, float %7194, i64 1, !dbg !43 + %8051 = fsub <2 x float> %8048, %8050, !dbg !43 + %8052 = insertelement <2 x float> poison, float %4187, i64 0, !dbg !43 + %8053 = insertelement <2 x float> %8052, float %4188, i64 1, !dbg !43 + %8054 = insertelement <2 x float> poison, float %7195, i64 0, !dbg !43 + %8055 = insertelement <2 x float> %8054, float %7196, i64 1, !dbg !43 + %8056 = fsub <2 x float> %8053, %8055, !dbg !43 + %8057 = insertelement <2 x float> poison, float %4189, i64 0, !dbg !43 + %8058 = insertelement <2 x float> %8057, float %4190, i64 1, !dbg !43 + %8059 = insertelement <2 x float> poison, float %7197, i64 0, !dbg !43 + %8060 = insertelement <2 x float> %8059, float %7198, i64 1, !dbg !43 + %8061 = fsub <2 x float> %8058, %8060, !dbg !43 + %8062 = insertelement <2 x float> poison, float %4191, i64 0, !dbg !43 + %8063 = insertelement <2 x float> %8062, float %4192, i64 1, !dbg !43 + %8064 = insertelement <2 x float> poison, float %7199, i64 0, !dbg !43 + %8065 = insertelement <2 x float> %8064, float %7200, i64 1, !dbg !43 + %8066 = fsub <2 x float> %8063, %8065, !dbg !43 + %8067 = insertelement <2 x float> poison, float %4193, i64 0, !dbg !43 + %8068 = insertelement <2 x float> %8067, float %4194, i64 1, !dbg !43 + %8069 = insertelement <2 x float> poison, float %7201, i64 0, !dbg !43 + %8070 = insertelement <2 x float> %8069, float %7202, i64 1, !dbg !43 + %8071 = fsub <2 x float> %8068, %8070, !dbg !43 + %8072 = insertelement <2 x float> poison, float %4195, i64 0, !dbg !43 + %8073 = insertelement <2 x float> %8072, float %4196, i64 1, !dbg !43 + %8074 = insertelement <2 x float> poison, float %7203, i64 0, !dbg !43 + %8075 = insertelement <2 x float> %8074, float %7204, i64 1, !dbg !43 + %8076 = fsub <2 x float> %8073, %8075, !dbg !43 + %8077 = insertelement <2 x float> poison, float %4197, i64 0, !dbg !43 + %8078 = insertelement <2 x float> %8077, float %4198, i64 1, !dbg !43 + %8079 = insertelement <2 x float> poison, float %7205, i64 0, !dbg !43 + %8080 = insertelement <2 x float> %8079, float %7206, i64 1, !dbg !43 + %8081 = fsub <2 x float> %8078, %8080, !dbg !43 + %8082 = insertelement <2 x float> poison, float %4199, i64 0, !dbg !43 + %8083 = insertelement <2 x float> %8082, float %4200, i64 1, !dbg !43 + %8084 = insertelement <2 x float> poison, float %7207, i64 0, !dbg !43 + %8085 = insertelement <2 x float> %8084, float %7208, i64 1, !dbg !43 + %8086 = fsub <2 x float> %8083, %8085, !dbg !43 + %8087 = insertelement <2 x float> poison, float %4201, i64 0, !dbg !43 + %8088 = insertelement <2 x float> %8087, float %4202, i64 1, !dbg !43 + %8089 = insertelement <2 x float> poison, float %7209, i64 0, !dbg !43 + %8090 = insertelement <2 x float> %8089, float %7210, i64 1, !dbg !43 + %8091 = fsub <2 x float> %8088, %8090, !dbg !43 + %8092 = insertelement <2 x float> poison, float %4203, i64 0, !dbg !43 + %8093 = insertelement <2 x float> %8092, float %4204, i64 1, !dbg !43 + %8094 = insertelement <2 x float> poison, float %7211, i64 0, !dbg !43 + %8095 = insertelement <2 x float> %8094, float %7212, i64 1, !dbg !43 + %8096 = fsub <2 x float> %8093, %8095, !dbg !43 + %8097 = insertelement <2 x float> poison, float %4205, i64 0, !dbg !43 + %8098 = insertelement <2 x float> %8097, float %4206, i64 1, !dbg !43 + %8099 = insertelement <2 x float> poison, float %7213, i64 0, !dbg !43 + %8100 = insertelement <2 x float> %8099, float %7214, i64 1, !dbg !43 + %8101 = fsub <2 x float> %8098, %8100, !dbg !43 + %8102 = insertelement <2 x float> poison, float %4207, i64 0, !dbg !43 + %8103 = insertelement <2 x float> %8102, float %4208, i64 1, !dbg !43 + %8104 = insertelement <2 x float> poison, float %7215, i64 0, !dbg !43 + %8105 = insertelement <2 x float> %8104, float %7216, i64 1, !dbg !43 + %8106 = fsub <2 x float> %8103, %8105, !dbg !43 + %8107 = insertelement <2 x float> poison, float %4209, i64 0, !dbg !43 + %8108 = insertelement <2 x float> %8107, float %4210, i64 1, !dbg !43 + %8109 = insertelement <2 x float> poison, float %7217, i64 0, !dbg !43 + %8110 = insertelement <2 x float> %8109, float %7218, i64 1, !dbg !43 + %8111 = fsub <2 x float> %8108, %8110, !dbg !43 + %8112 = insertelement <2 x float> poison, float %4211, i64 0, !dbg !43 + %8113 = insertelement <2 x float> %8112, float %4212, i64 1, !dbg !43 + %8114 = insertelement <2 x float> poison, float %7219, i64 0, !dbg !43 + %8115 = insertelement <2 x float> %8114, float %7220, i64 1, !dbg !43 + %8116 = fsub <2 x float> %8113, %8115, !dbg !43 + %8117 = insertelement <2 x float> poison, float %4213, i64 0, !dbg !43 + %8118 = insertelement <2 x float> %8117, float %4214, i64 1, !dbg !43 + %8119 = insertelement <2 x float> poison, float %7221, i64 0, !dbg !43 + %8120 = insertelement <2 x float> %8119, float %7222, i64 1, !dbg !43 + %8121 = fsub <2 x float> %8118, %8120, !dbg !43 + %8122 = insertelement <2 x float> poison, float %4215, i64 0, !dbg !43 + %8123 = insertelement <2 x float> %8122, float %4216, i64 1, !dbg !43 + %8124 = insertelement <2 x float> poison, float %7223, i64 0, !dbg !43 + %8125 = insertelement <2 x float> %8124, float %7224, i64 1, !dbg !43 + %8126 = fsub <2 x float> %8123, %8125, !dbg !43 + %8127 = insertelement <2 x float> poison, float %4217, i64 0, !dbg !43 + %8128 = insertelement <2 x float> %8127, float %4218, i64 1, !dbg !43 + %8129 = insertelement <2 x float> poison, float %7225, i64 0, !dbg !43 + %8130 = insertelement <2 x float> %8129, float %7226, i64 1, !dbg !43 + %8131 = fsub <2 x float> %8128, %8130, !dbg !43 + %8132 = insertelement <2 x float> poison, float %4219, i64 0, !dbg !43 + %8133 = insertelement <2 x float> %8132, float %4220, i64 1, !dbg !43 + %8134 = insertelement <2 x float> poison, float %7227, i64 0, !dbg !43 + %8135 = insertelement <2 x float> %8134, float %7228, i64 1, !dbg !43 + %8136 = fsub <2 x float> %8133, %8135, !dbg !43 + %8137 = insertelement <2 x float> poison, float %4221, i64 0, !dbg !43 + %8138 = insertelement <2 x float> %8137, float %4222, i64 1, !dbg !43 + %8139 = insertelement <2 x float> poison, float %7229, i64 0, !dbg !43 + %8140 = insertelement <2 x float> %8139, float %7230, i64 1, !dbg !43 + %8141 = fsub <2 x float> %8138, %8140, !dbg !43 + %8142 = insertelement <2 x float> poison, float %4223, i64 0, !dbg !43 + %8143 = insertelement <2 x float> %8142, float %4224, i64 1, !dbg !43 + %8144 = insertelement <2 x float> poison, float %7231, i64 0, !dbg !43 + %8145 = insertelement <2 x float> %8144, float %7232, i64 1, !dbg !43 + %8146 = fsub <2 x float> %8143, %8145, !dbg !43 + %8147 = insertelement <2 x float> poison, float %4225, i64 0, !dbg !43 + %8148 = insertelement <2 x float> %8147, float %4226, i64 1, !dbg !43 + %8149 = insertelement <2 x float> poison, float %7233, i64 0, !dbg !43 + %8150 = insertelement <2 x float> %8149, float %7234, i64 1, !dbg !43 + %8151 = fsub <2 x float> %8148, %8150, !dbg !43 + %8152 = insertelement <2 x float> poison, float %4227, i64 0, !dbg !43 + %8153 = insertelement <2 x float> %8152, float %4228, i64 1, !dbg !43 + %8154 = insertelement <2 x float> poison, float %7235, i64 0, !dbg !43 + %8155 = insertelement <2 x float> %8154, float %7236, i64 1, !dbg !43 + %8156 = fsub <2 x float> %8153, %8155, !dbg !43 + %8157 = insertelement <2 x float> poison, float %4229, i64 0, !dbg !43 + %8158 = insertelement <2 x float> %8157, float %4230, i64 1, !dbg !43 + %8159 = insertelement <2 x float> poison, float %7237, i64 0, !dbg !43 + %8160 = insertelement <2 x float> %8159, float %7238, i64 1, !dbg !43 + %8161 = fsub <2 x float> %8158, %8160, !dbg !43 + %8162 = insertelement <2 x float> poison, float %4231, i64 0, !dbg !43 + %8163 = insertelement <2 x float> %8162, float %4232, i64 1, !dbg !43 + %8164 = insertelement <2 x float> poison, float %7239, i64 0, !dbg !43 + %8165 = insertelement <2 x float> %8164, float %7240, i64 1, !dbg !43 + %8166 = fsub <2 x float> %8163, %8165, !dbg !43 + %8167 = insertelement <2 x float> poison, float %4233, i64 0, !dbg !43 + %8168 = insertelement <2 x float> %8167, float %4234, i64 1, !dbg !43 + %8169 = insertelement <2 x float> poison, float %7241, i64 0, !dbg !43 + %8170 = insertelement <2 x float> %8169, float %7242, i64 1, !dbg !43 + %8171 = fsub <2 x float> %8168, %8170, !dbg !43 + %8172 = insertelement <2 x float> poison, float %4235, i64 0, !dbg !43 + %8173 = insertelement <2 x float> %8172, float %4236, i64 1, !dbg !43 + %8174 = insertelement <2 x float> poison, float %7243, i64 0, !dbg !43 + %8175 = insertelement <2 x float> %8174, float %7244, i64 1, !dbg !43 + %8176 = fsub <2 x float> %8173, %8175, !dbg !43 + %8177 = insertelement <2 x float> poison, float %4237, i64 0, !dbg !43 + %8178 = insertelement <2 x float> %8177, float %4238, i64 1, !dbg !43 + %8179 = insertelement <2 x float> poison, float %7245, i64 0, !dbg !43 + %8180 = insertelement <2 x float> %8179, float %7246, i64 1, !dbg !43 + %8181 = fsub <2 x float> %8178, %8180, !dbg !43 + %8182 = insertelement <2 x float> poison, float %4239, i64 0, !dbg !43 + %8183 = insertelement <2 x float> %8182, float %4240, i64 1, !dbg !43 + %8184 = insertelement <2 x float> poison, float %7247, i64 0, !dbg !43 + %8185 = insertelement <2 x float> %8184, float %7248, i64 1, !dbg !43 + %8186 = fsub <2 x float> %8183, %8185, !dbg !43 + %8187 = insertelement <2 x float> poison, float %4241, i64 0, !dbg !43 + %8188 = insertelement <2 x float> %8187, float %4242, i64 1, !dbg !43 + %8189 = insertelement <2 x float> poison, float %7249, i64 0, !dbg !43 + %8190 = insertelement <2 x float> %8189, float %7250, i64 1, !dbg !43 + %8191 = fsub <2 x float> %8188, %8190, !dbg !43 + %8192 = insertelement <2 x float> poison, float %4243, i64 0, !dbg !43 + %8193 = insertelement <2 x float> %8192, float %4244, i64 1, !dbg !43 + %8194 = insertelement <2 x float> poison, float %7251, i64 0, !dbg !43 + %8195 = insertelement <2 x float> %8194, float %7252, i64 1, !dbg !43 + %8196 = fsub <2 x float> %8193, %8195, !dbg !43 + %8197 = insertelement <2 x float> poison, float %4245, i64 0, !dbg !43 + %8198 = insertelement <2 x float> %8197, float %4246, i64 1, !dbg !43 + %8199 = insertelement <2 x float> poison, float %7253, i64 0, !dbg !43 + %8200 = insertelement <2 x float> %8199, float %7254, i64 1, !dbg !43 + %8201 = fsub <2 x float> %8198, %8200, !dbg !43 + %8202 = insertelement <2 x float> poison, float %4247, i64 0, !dbg !43 + %8203 = insertelement <2 x float> %8202, float %4248, i64 1, !dbg !43 + %8204 = insertelement <2 x float> poison, float %7255, i64 0, !dbg !43 + %8205 = insertelement <2 x float> %8204, float %7256, i64 1, !dbg !43 + %8206 = fsub <2 x float> %8203, %8205, !dbg !43 + %8207 = insertelement <2 x float> poison, float %4249, i64 0, !dbg !43 + %8208 = insertelement <2 x float> %8207, float %4250, i64 1, !dbg !43 + %8209 = insertelement <2 x float> poison, float %7257, i64 0, !dbg !43 + %8210 = insertelement <2 x float> %8209, float %7258, i64 1, !dbg !43 + %8211 = fsub <2 x float> %8208, %8210, !dbg !43 + %8212 = insertelement <2 x float> poison, float %4251, i64 0, !dbg !43 + %8213 = insertelement <2 x float> %8212, float %4252, i64 1, !dbg !43 + %8214 = insertelement <2 x float> poison, float %7259, i64 0, !dbg !43 + %8215 = insertelement <2 x float> %8214, float %7260, i64 1, !dbg !43 + %8216 = fsub <2 x float> %8213, %8215, !dbg !43 + %8217 = insertelement <2 x float> poison, float %4253, i64 0, !dbg !43 + %8218 = insertelement <2 x float> %8217, float %4254, i64 1, !dbg !43 + %8219 = insertelement <2 x float> poison, float %7261, i64 0, !dbg !43 + %8220 = insertelement <2 x float> %8219, float %7262, i64 1, !dbg !43 + %8221 = fsub <2 x float> %8218, %8220, !dbg !43 + %8222 = insertelement <2 x float> poison, float %4255, i64 0, !dbg !43 + %8223 = insertelement <2 x float> %8222, float %4256, i64 1, !dbg !43 + %8224 = insertelement <2 x float> poison, float %7263, i64 0, !dbg !43 + %8225 = insertelement <2 x float> %8224, float %7264, i64 1, !dbg !43 + %8226 = fsub <2 x float> %8223, %8225, !dbg !43 + %8227 = insertelement <2 x float> poison, float %4257, i64 0, !dbg !43 + %8228 = insertelement <2 x float> %8227, float %4258, i64 1, !dbg !43 + %8229 = insertelement <2 x float> poison, float %7265, i64 0, !dbg !43 + %8230 = insertelement <2 x float> %8229, float %7266, i64 1, !dbg !43 + %8231 = fsub <2 x float> %8228, %8230, !dbg !43 + %8232 = insertelement <2 x float> poison, float %4259, i64 0, !dbg !43 + %8233 = insertelement <2 x float> %8232, float %4260, i64 1, !dbg !43 + %8234 = insertelement <2 x float> poison, float %7267, i64 0, !dbg !43 + %8235 = insertelement <2 x float> %8234, float %7268, i64 1, !dbg !43 + %8236 = fsub <2 x float> %8233, %8235, !dbg !43 + %8237 = insertelement <2 x float> poison, float %4261, i64 0, !dbg !43 + %8238 = insertelement <2 x float> %8237, float %4262, i64 1, !dbg !43 + %8239 = insertelement <2 x float> poison, float %7269, i64 0, !dbg !43 + %8240 = insertelement <2 x float> %8239, float %7270, i64 1, !dbg !43 + %8241 = fsub <2 x float> %8238, %8240, !dbg !43 + %8242 = insertelement <2 x float> poison, float %4263, i64 0, !dbg !43 + %8243 = insertelement <2 x float> %8242, float %4264, i64 1, !dbg !43 + %8244 = insertelement <2 x float> poison, float %7271, i64 0, !dbg !43 + %8245 = insertelement <2 x float> %8244, float %7272, i64 1, !dbg !43 + %8246 = fsub <2 x float> %8243, %8245, !dbg !43 + %8247 = insertelement <2 x float> poison, float %4265, i64 0, !dbg !43 + %8248 = insertelement <2 x float> %8247, float %4266, i64 1, !dbg !43 + %8249 = insertelement <2 x float> poison, float %7273, i64 0, !dbg !43 + %8250 = insertelement <2 x float> %8249, float %7274, i64 1, !dbg !43 + %8251 = fsub <2 x float> %8248, %8250, !dbg !43 + %8252 = insertelement <2 x float> poison, float %4267, i64 0, !dbg !43 + %8253 = insertelement <2 x float> %8252, float %4268, i64 1, !dbg !43 + %8254 = insertelement <2 x float> poison, float %7275, i64 0, !dbg !43 + %8255 = insertelement <2 x float> %8254, float %7276, i64 1, !dbg !43 + %8256 = fsub <2 x float> %8253, %8255, !dbg !43 + %8257 = insertelement <2 x float> poison, float %4269, i64 0, !dbg !43 + %8258 = insertelement <2 x float> %8257, float %4270, i64 1, !dbg !43 + %8259 = insertelement <2 x float> poison, float %7277, i64 0, !dbg !43 + %8260 = insertelement <2 x float> %8259, float %7278, i64 1, !dbg !43 + %8261 = fsub <2 x float> %8258, %8260, !dbg !43 + %8262 = insertelement <2 x float> poison, float %4271, i64 0, !dbg !43 + %8263 = insertelement <2 x float> %8262, float %4272, i64 1, !dbg !43 + %8264 = insertelement <2 x float> poison, float %7279, i64 0, !dbg !43 + %8265 = insertelement <2 x float> %8264, float %7280, i64 1, !dbg !43 + %8266 = fsub <2 x float> %8263, %8265, !dbg !43 + %8267 = insertelement <2 x float> poison, float %4273, i64 0, !dbg !43 + %8268 = insertelement <2 x float> %8267, float %4274, i64 1, !dbg !43 + %8269 = insertelement <2 x float> poison, float %7281, i64 0, !dbg !43 + %8270 = insertelement <2 x float> %8269, float %7282, i64 1, !dbg !43 + %8271 = fsub <2 x float> %8268, %8270, !dbg !43 + %8272 = insertelement <2 x float> poison, float %4275, i64 0, !dbg !43 + %8273 = insertelement <2 x float> %8272, float %4276, i64 1, !dbg !43 + %8274 = insertelement <2 x float> poison, float %7283, i64 0, !dbg !43 + %8275 = insertelement <2 x float> %8274, float %7284, i64 1, !dbg !43 + %8276 = fsub <2 x float> %8273, %8275, !dbg !43 + %8277 = insertelement <2 x float> poison, float %4277, i64 0, !dbg !43 + %8278 = insertelement <2 x float> %8277, float %4278, i64 1, !dbg !43 + %8279 = insertelement <2 x float> poison, float %7285, i64 0, !dbg !43 + %8280 = insertelement <2 x float> %8279, float %7286, i64 1, !dbg !43 + %8281 = fsub <2 x float> %8278, %8280, !dbg !43 + %8282 = insertelement <2 x float> poison, float %4279, i64 0, !dbg !43 + %8283 = insertelement <2 x float> %8282, float %4280, i64 1, !dbg !43 + %8284 = insertelement <2 x float> poison, float %7287, i64 0, !dbg !43 + %8285 = insertelement <2 x float> %8284, float %7288, i64 1, !dbg !43 + %8286 = fsub <2 x float> %8283, %8285, !dbg !43 + %8287 = insertelement <2 x float> poison, float %4281, i64 0, !dbg !43 + %8288 = insertelement <2 x float> %8287, float %4282, i64 1, !dbg !43 + %8289 = insertelement <2 x float> poison, float %7289, i64 0, !dbg !43 + %8290 = insertelement <2 x float> %8289, float %7290, i64 1, !dbg !43 + %8291 = fsub <2 x float> %8288, %8290, !dbg !43 + %8292 = insertelement <2 x float> poison, float %4283, i64 0, !dbg !43 + %8293 = insertelement <2 x float> %8292, float %4284, i64 1, !dbg !43 + %8294 = insertelement <2 x float> poison, float %7291, i64 0, !dbg !43 + %8295 = insertelement <2 x float> %8294, float %7292, i64 1, !dbg !43 + %8296 = fsub <2 x float> %8293, %8295, !dbg !43 + %8297 = insertelement <2 x float> poison, float %4285, i64 0, !dbg !43 + %8298 = insertelement <2 x float> %8297, float %4286, i64 1, !dbg !43 + %8299 = insertelement <2 x float> poison, float %7293, i64 0, !dbg !43 + %8300 = insertelement <2 x float> %8299, float %7294, i64 1, !dbg !43 + %8301 = fsub <2 x float> %8298, %8300, !dbg !43 + %8302 = insertelement <2 x float> poison, float %4287, i64 0, !dbg !43 + %8303 = insertelement <2 x float> %8302, float %4288, i64 1, !dbg !43 + %8304 = insertelement <2 x float> poison, float %7295, i64 0, !dbg !43 + %8305 = insertelement <2 x float> %8304, float %7296, i64 1, !dbg !43 + %8306 = fsub <2 x float> %8303, %8305, !dbg !43 + %8307 = insertelement <2 x float> poison, float %4289, i64 0, !dbg !43 + %8308 = insertelement <2 x float> %8307, float %4290, i64 1, !dbg !43 + %8309 = insertelement <2 x float> poison, float %7297, i64 0, !dbg !43 + %8310 = insertelement <2 x float> %8309, float %7298, i64 1, !dbg !43 + %8311 = fsub <2 x float> %8308, %8310, !dbg !43 + %8312 = insertelement <2 x float> poison, float %4291, i64 0, !dbg !43 + %8313 = insertelement <2 x float> %8312, float %4292, i64 1, !dbg !43 + %8314 = insertelement <2 x float> poison, float %7299, i64 0, !dbg !43 + %8315 = insertelement <2 x float> %8314, float %7300, i64 1, !dbg !43 + %8316 = fsub <2 x float> %8313, %8315, !dbg !43 + %8317 = insertelement <2 x float> poison, float %4293, i64 0, !dbg !43 + %8318 = insertelement <2 x float> %8317, float %4294, i64 1, !dbg !43 + %8319 = insertelement <2 x float> poison, float %7301, i64 0, !dbg !43 + %8320 = insertelement <2 x float> %8319, float %7302, i64 1, !dbg !43 + %8321 = fsub <2 x float> %8318, %8320, !dbg !43 + %8322 = insertelement <2 x float> poison, float %4295, i64 0, !dbg !43 + %8323 = insertelement <2 x float> %8322, float %4296, i64 1, !dbg !43 + %8324 = insertelement <2 x float> poison, float %7303, i64 0, !dbg !43 + %8325 = insertelement <2 x float> %8324, float %7304, i64 1, !dbg !43 + %8326 = fsub <2 x float> %8323, %8325, !dbg !43 + %8327 = insertelement <2 x float> poison, float %4297, i64 0, !dbg !43 + %8328 = insertelement <2 x float> %8327, float %4298, i64 1, !dbg !43 + %8329 = insertelement <2 x float> poison, float %7305, i64 0, !dbg !43 + %8330 = insertelement <2 x float> %8329, float %7306, i64 1, !dbg !43 + %8331 = fsub <2 x float> %8328, %8330, !dbg !43 + %8332 = insertelement <2 x float> poison, float %4299, i64 0, !dbg !43 + %8333 = insertelement <2 x float> %8332, float %4300, i64 1, !dbg !43 + %8334 = insertelement <2 x float> poison, float %7307, i64 0, !dbg !43 + %8335 = insertelement <2 x float> %8334, float %7308, i64 1, !dbg !43 + %8336 = fsub <2 x float> %8333, %8335, !dbg !43 + %8337 = insertelement <2 x float> poison, float %4301, i64 0, !dbg !43 + %8338 = insertelement <2 x float> %8337, float %4302, i64 1, !dbg !43 + %8339 = insertelement <2 x float> poison, float %7309, i64 0, !dbg !43 + %8340 = insertelement <2 x float> %8339, float %7310, i64 1, !dbg !43 + %8341 = fsub <2 x float> %8338, %8340, !dbg !43 + %8342 = insertelement <2 x float> poison, float %4303, i64 0, !dbg !43 + %8343 = insertelement <2 x float> %8342, float %4304, i64 1, !dbg !43 + %8344 = insertelement <2 x float> poison, float %7311, i64 0, !dbg !43 + %8345 = insertelement <2 x float> %8344, float %7312, i64 1, !dbg !43 + %8346 = fsub <2 x float> %8343, %8345, !dbg !43 + %8347 = insertelement <2 x float> poison, float %4305, i64 0, !dbg !43 + %8348 = insertelement <2 x float> %8347, float %4306, i64 1, !dbg !43 + %8349 = insertelement <2 x float> poison, float %7313, i64 0, !dbg !43 + %8350 = insertelement <2 x float> %8349, float %7314, i64 1, !dbg !43 + %8351 = fsub <2 x float> %8348, %8350, !dbg !43 + %8352 = insertelement <2 x float> poison, float %4307, i64 0, !dbg !43 + %8353 = insertelement <2 x float> %8352, float %4308, i64 1, !dbg !43 + %8354 = insertelement <2 x float> poison, float %7315, i64 0, !dbg !43 + %8355 = insertelement <2 x float> %8354, float %7316, i64 1, !dbg !43 + %8356 = fsub <2 x float> %8353, %8355, !dbg !43 + %8357 = insertelement <2 x float> poison, float %4309, i64 0, !dbg !43 + %8358 = insertelement <2 x float> %8357, float %4310, i64 1, !dbg !43 + %8359 = insertelement <2 x float> poison, float %7317, i64 0, !dbg !43 + %8360 = insertelement <2 x float> %8359, float %7318, i64 1, !dbg !43 + %8361 = fsub <2 x float> %8358, %8360, !dbg !43 + %8362 = insertelement <2 x float> poison, float %4311, i64 0, !dbg !43 + %8363 = insertelement <2 x float> %8362, float %4312, i64 1, !dbg !43 + %8364 = insertelement <2 x float> poison, float %7319, i64 0, !dbg !43 + %8365 = insertelement <2 x float> %8364, float %7320, i64 1, !dbg !43 + %8366 = fsub <2 x float> %8363, %8365, !dbg !43 + %8367 = insertelement <2 x float> poison, float %4313, i64 0, !dbg !43 + %8368 = insertelement <2 x float> %8367, float %4314, i64 1, !dbg !43 + %8369 = insertelement <2 x float> poison, float %7321, i64 0, !dbg !43 + %8370 = insertelement <2 x float> %8369, float %7322, i64 1, !dbg !43 + %8371 = fsub <2 x float> %8368, %8370, !dbg !43 + %8372 = insertelement <2 x float> poison, float %4315, i64 0, !dbg !43 + %8373 = insertelement <2 x float> %8372, float %4316, i64 1, !dbg !43 + %8374 = insertelement <2 x float> poison, float %7323, i64 0, !dbg !43 + %8375 = insertelement <2 x float> %8374, float %7324, i64 1, !dbg !43 + %8376 = fsub <2 x float> %8373, %8375, !dbg !43 + %8377 = insertelement <2 x float> poison, float %4317, i64 0, !dbg !43 + %8378 = insertelement <2 x float> %8377, float %4318, i64 1, !dbg !43 + %8379 = insertelement <2 x float> poison, float %7325, i64 0, !dbg !43 + %8380 = insertelement <2 x float> %8379, float %7326, i64 1, !dbg !43 + %8381 = fsub <2 x float> %8378, %8380, !dbg !43 + %8382 = insertelement <2 x float> poison, float %4319, i64 0, !dbg !43 + %8383 = insertelement <2 x float> %8382, float %4320, i64 1, !dbg !43 + %8384 = insertelement <2 x float> poison, float %7327, i64 0, !dbg !43 + %8385 = insertelement <2 x float> %8384, float %7328, i64 1, !dbg !43 + %8386 = fsub <2 x float> %8383, %8385, !dbg !43 + %8387 = insertelement <2 x float> poison, float %4321, i64 0, !dbg !43 + %8388 = insertelement <2 x float> %8387, float %4322, i64 1, !dbg !43 + %8389 = insertelement <2 x float> poison, float %7329, i64 0, !dbg !43 + %8390 = insertelement <2 x float> %8389, float %7330, i64 1, !dbg !43 + %8391 = fsub <2 x float> %8388, %8390, !dbg !43 + %8392 = insertelement <2 x float> poison, float %4323, i64 0, !dbg !43 + %8393 = insertelement <2 x float> %8392, float %4324, i64 1, !dbg !43 + %8394 = insertelement <2 x float> poison, float %7331, i64 0, !dbg !43 + %8395 = insertelement <2 x float> %8394, float %7332, i64 1, !dbg !43 + %8396 = fsub <2 x float> %8393, %8395, !dbg !43 + %8397 = insertelement <2 x float> poison, float %4325, i64 0, !dbg !43 + %8398 = insertelement <2 x float> %8397, float %4326, i64 1, !dbg !43 + %8399 = insertelement <2 x float> poison, float %7333, i64 0, !dbg !43 + %8400 = insertelement <2 x float> %8399, float %7334, i64 1, !dbg !43 + %8401 = fsub <2 x float> %8398, %8400, !dbg !43 + %8402 = insertelement <2 x float> poison, float %4327, i64 0, !dbg !43 + %8403 = insertelement <2 x float> %8402, float %4328, i64 1, !dbg !43 + %8404 = insertelement <2 x float> poison, float %7335, i64 0, !dbg !43 + %8405 = insertelement <2 x float> %8404, float %7336, i64 1, !dbg !43 + %8406 = fsub <2 x float> %8403, %8405, !dbg !43 + %8407 = insertelement <2 x float> poison, float %4329, i64 0, !dbg !43 + %8408 = insertelement <2 x float> %8407, float %4330, i64 1, !dbg !43 + %8409 = insertelement <2 x float> poison, float %7337, i64 0, !dbg !43 + %8410 = insertelement <2 x float> %8409, float %7338, i64 1, !dbg !43 + %8411 = fsub <2 x float> %8408, %8410, !dbg !43 + %8412 = insertelement <2 x float> poison, float %4331, i64 0, !dbg !43 + %8413 = insertelement <2 x float> %8412, float %4332, i64 1, !dbg !43 + %8414 = insertelement <2 x float> poison, float %7339, i64 0, !dbg !43 + %8415 = insertelement <2 x float> %8414, float %7340, i64 1, !dbg !43 + %8416 = fsub <2 x float> %8413, %8415, !dbg !43 + %8417 = insertelement <2 x float> poison, float %4333, i64 0, !dbg !43 + %8418 = insertelement <2 x float> %8417, float %4334, i64 1, !dbg !43 + %8419 = insertelement <2 x float> poison, float %7341, i64 0, !dbg !43 + %8420 = insertelement <2 x float> %8419, float %7342, i64 1, !dbg !43 + %8421 = fsub <2 x float> %8418, %8420, !dbg !43 + %8422 = insertelement <2 x float> poison, float %4335, i64 0, !dbg !43 + %8423 = insertelement <2 x float> %8422, float %4336, i64 1, !dbg !43 + %8424 = insertelement <2 x float> poison, float %7343, i64 0, !dbg !43 + %8425 = insertelement <2 x float> %8424, float %7344, i64 1, !dbg !43 + %8426 = fsub <2 x float> %8423, %8425, !dbg !43 + %8427 = insertelement <2 x float> poison, float %4337, i64 0, !dbg !43 + %8428 = insertelement <2 x float> %8427, float %4338, i64 1, !dbg !43 + %8429 = insertelement <2 x float> poison, float %7345, i64 0, !dbg !43 + %8430 = insertelement <2 x float> %8429, float %7346, i64 1, !dbg !43 + %8431 = fsub <2 x float> %8428, %8430, !dbg !43 + %8432 = insertelement <2 x float> poison, float %4339, i64 0, !dbg !43 + %8433 = insertelement <2 x float> %8432, float %4340, i64 1, !dbg !43 + %8434 = insertelement <2 x float> poison, float %7347, i64 0, !dbg !43 + %8435 = insertelement <2 x float> %8434, float %7348, i64 1, !dbg !43 + %8436 = fsub <2 x float> %8433, %8435, !dbg !43 + %8437 = insertelement <2 x float> poison, float %4341, i64 0, !dbg !43 + %8438 = insertelement <2 x float> %8437, float %4342, i64 1, !dbg !43 + %8439 = insertelement <2 x float> poison, float %7349, i64 0, !dbg !43 + %8440 = insertelement <2 x float> %8439, float %7350, i64 1, !dbg !43 + %8441 = fsub <2 x float> %8438, %8440, !dbg !43 + %8442 = insertelement <2 x float> poison, float %4343, i64 0, !dbg !43 + %8443 = insertelement <2 x float> %8442, float %4344, i64 1, !dbg !43 + %8444 = insertelement <2 x float> poison, float %7351, i64 0, !dbg !43 + %8445 = insertelement <2 x float> %8444, float %7352, i64 1, !dbg !43 + %8446 = fsub <2 x float> %8443, %8445, !dbg !43 + %8447 = insertelement <2 x float> poison, float %4345, i64 0, !dbg !43 + %8448 = insertelement <2 x float> %8447, float %4346, i64 1, !dbg !43 + %8449 = insertelement <2 x float> poison, float %7353, i64 0, !dbg !43 + %8450 = insertelement <2 x float> %8449, float %7354, i64 1, !dbg !43 + %8451 = fsub <2 x float> %8448, %8450, !dbg !43 + %8452 = insertelement <2 x float> poison, float %4347, i64 0, !dbg !43 + %8453 = insertelement <2 x float> %8452, float %4348, i64 1, !dbg !43 + %8454 = insertelement <2 x float> poison, float %7355, i64 0, !dbg !43 + %8455 = insertelement <2 x float> %8454, float %7356, i64 1, !dbg !43 + %8456 = fsub <2 x float> %8453, %8455, !dbg !43 + %8457 = insertelement <2 x float> poison, float %4349, i64 0, !dbg !43 + %8458 = insertelement <2 x float> %8457, float %4350, i64 1, !dbg !43 + %8459 = insertelement <2 x float> poison, float %7357, i64 0, !dbg !43 + %8460 = insertelement <2 x float> %8459, float %7358, i64 1, !dbg !43 + %8461 = fsub <2 x float> %8458, %8460, !dbg !43 + %8462 = insertelement <2 x float> poison, float %4351, i64 0, !dbg !43 + %8463 = insertelement <2 x float> %8462, float %4352, i64 1, !dbg !43 + %8464 = insertelement <2 x float> poison, float %7359, i64 0, !dbg !43 + %8465 = insertelement <2 x float> %8464, float %7360, i64 1, !dbg !43 + %8466 = fsub <2 x float> %8463, %8465, !dbg !43 + %8467 = insertelement <2 x float> poison, float %4353, i64 0, !dbg !43 + %8468 = insertelement <2 x float> %8467, float %4354, i64 1, !dbg !43 + %8469 = insertelement <2 x float> poison, float %7361, i64 0, !dbg !43 + %8470 = insertelement <2 x float> %8469, float %7362, i64 1, !dbg !43 + %8471 = fsub <2 x float> %8468, %8470, !dbg !43 + %8472 = insertelement <2 x float> poison, float %4355, i64 0, !dbg !43 + %8473 = insertelement <2 x float> %8472, float %4356, i64 1, !dbg !43 + %8474 = insertelement <2 x float> poison, float %7363, i64 0, !dbg !43 + %8475 = insertelement <2 x float> %8474, float %7364, i64 1, !dbg !43 + %8476 = fsub <2 x float> %8473, %8475, !dbg !43 + %8477 = insertelement <2 x float> poison, float %4357, i64 0, !dbg !43 + %8478 = insertelement <2 x float> %8477, float %4358, i64 1, !dbg !43 + %8479 = insertelement <2 x float> poison, float %7365, i64 0, !dbg !43 + %8480 = insertelement <2 x float> %8479, float %7366, i64 1, !dbg !43 + %8481 = fsub <2 x float> %8478, %8480, !dbg !43 + %8482 = insertelement <2 x float> poison, float %4359, i64 0, !dbg !43 + %8483 = insertelement <2 x float> %8482, float %4360, i64 1, !dbg !43 + %8484 = insertelement <2 x float> poison, float %7367, i64 0, !dbg !43 + %8485 = insertelement <2 x float> %8484, float %7368, i64 1, !dbg !43 + %8486 = fsub <2 x float> %8483, %8485, !dbg !43 + %8487 = insertelement <2 x float> poison, float %4361, i64 0, !dbg !43 + %8488 = insertelement <2 x float> %8487, float %4362, i64 1, !dbg !43 + %8489 = insertelement <2 x float> poison, float %7369, i64 0, !dbg !43 + %8490 = insertelement <2 x float> %8489, float %7370, i64 1, !dbg !43 + %8491 = fsub <2 x float> %8488, %8490, !dbg !43 + %8492 = insertelement <2 x float> poison, float %4363, i64 0, !dbg !43 + %8493 = insertelement <2 x float> %8492, float %4364, i64 1, !dbg !43 + %8494 = insertelement <2 x float> poison, float %7371, i64 0, !dbg !43 + %8495 = insertelement <2 x float> %8494, float %7372, i64 1, !dbg !43 + %8496 = fsub <2 x float> %8493, %8495, !dbg !43 + %8497 = insertelement <2 x float> poison, float %4365, i64 0, !dbg !43 + %8498 = insertelement <2 x float> %8497, float %4366, i64 1, !dbg !43 + %8499 = insertelement <2 x float> poison, float %7373, i64 0, !dbg !43 + %8500 = insertelement <2 x float> %8499, float %7374, i64 1, !dbg !43 + %8501 = fsub <2 x float> %8498, %8500, !dbg !43 + %8502 = insertelement <2 x float> poison, float %4367, i64 0, !dbg !43 + %8503 = insertelement <2 x float> %8502, float %4368, i64 1, !dbg !43 + %8504 = insertelement <2 x float> poison, float %7375, i64 0, !dbg !43 + %8505 = insertelement <2 x float> %8504, float %7376, i64 1, !dbg !43 + %8506 = fsub <2 x float> %8503, %8505, !dbg !43 + %8507 = insertelement <2 x float> poison, float %4369, i64 0, !dbg !43 + %8508 = insertelement <2 x float> %8507, float %4370, i64 1, !dbg !43 + %8509 = insertelement <2 x float> poison, float %7377, i64 0, !dbg !43 + %8510 = insertelement <2 x float> %8509, float %7378, i64 1, !dbg !43 + %8511 = fsub <2 x float> %8508, %8510, !dbg !43 + %8512 = insertelement <2 x float> poison, float %4371, i64 0, !dbg !43 + %8513 = insertelement <2 x float> %8512, float %4372, i64 1, !dbg !43 + %8514 = insertelement <2 x float> poison, float %7379, i64 0, !dbg !43 + %8515 = insertelement <2 x float> %8514, float %7380, i64 1, !dbg !43 + %8516 = fsub <2 x float> %8513, %8515, !dbg !43 + %8517 = insertelement <2 x float> poison, float %4373, i64 0, !dbg !43 + %8518 = insertelement <2 x float> %8517, float %4374, i64 1, !dbg !43 + %8519 = insertelement <2 x float> poison, float %7381, i64 0, !dbg !43 + %8520 = insertelement <2 x float> %8519, float %7382, i64 1, !dbg !43 + %8521 = fsub <2 x float> %8518, %8520, !dbg !43 + %8522 = insertelement <2 x float> poison, float %4375, i64 0, !dbg !43 + %8523 = insertelement <2 x float> %8522, float %4376, i64 1, !dbg !43 + %8524 = insertelement <2 x float> poison, float %7383, i64 0, !dbg !43 + %8525 = insertelement <2 x float> %8524, float %7384, i64 1, !dbg !43 + %8526 = fsub <2 x float> %8523, %8525, !dbg !43 + %8527 = insertelement <2 x float> poison, float %4377, i64 0, !dbg !43 + %8528 = insertelement <2 x float> %8527, float %4378, i64 1, !dbg !43 + %8529 = insertelement <2 x float> poison, float %7385, i64 0, !dbg !43 + %8530 = insertelement <2 x float> %8529, float %7386, i64 1, !dbg !43 + %8531 = fsub <2 x float> %8528, %8530, !dbg !43 + %8532 = insertelement <2 x float> poison, float %4379, i64 0, !dbg !43 + %8533 = insertelement <2 x float> %8532, float %4380, i64 1, !dbg !43 + %8534 = insertelement <2 x float> poison, float %7387, i64 0, !dbg !43 + %8535 = insertelement <2 x float> %8534, float %7388, i64 1, !dbg !43 + %8536 = fsub <2 x float> %8533, %8535, !dbg !43 + %8537 = insertelement <2 x float> poison, float %4381, i64 0, !dbg !43 + %8538 = insertelement <2 x float> %8537, float %4382, i64 1, !dbg !43 + %8539 = insertelement <2 x float> poison, float %7389, i64 0, !dbg !43 + %8540 = insertelement <2 x float> %8539, float %7390, i64 1, !dbg !43 + %8541 = fsub <2 x float> %8538, %8540, !dbg !43 + %8542 = insertelement <2 x float> poison, float %4383, i64 0, !dbg !43 + %8543 = insertelement <2 x float> %8542, float %4384, i64 1, !dbg !43 + %8544 = insertelement <2 x float> poison, float %7391, i64 0, !dbg !43 + %8545 = insertelement <2 x float> %8544, float %7392, i64 1, !dbg !43 + %8546 = fsub <2 x float> %8543, %8545, !dbg !43 + %8547 = insertelement <2 x float> poison, float %4385, i64 0, !dbg !43 + %8548 = insertelement <2 x float> %8547, float %4386, i64 1, !dbg !43 + %8549 = insertelement <2 x float> poison, float %7393, i64 0, !dbg !43 + %8550 = insertelement <2 x float> %8549, float %7394, i64 1, !dbg !43 + %8551 = fsub <2 x float> %8548, %8550, !dbg !43 + %8552 = insertelement <2 x float> poison, float %4387, i64 0, !dbg !43 + %8553 = insertelement <2 x float> %8552, float %4388, i64 1, !dbg !43 + %8554 = insertelement <2 x float> poison, float %7395, i64 0, !dbg !43 + %8555 = insertelement <2 x float> %8554, float %7396, i64 1, !dbg !43 + %8556 = fsub <2 x float> %8553, %8555, !dbg !43 + %8557 = insertelement <2 x float> poison, float %4389, i64 0, !dbg !43 + %8558 = insertelement <2 x float> %8557, float %4390, i64 1, !dbg !43 + %8559 = insertelement <2 x float> poison, float %7397, i64 0, !dbg !43 + %8560 = insertelement <2 x float> %8559, float %7398, i64 1, !dbg !43 + %8561 = fsub <2 x float> %8558, %8560, !dbg !43 + %8562 = insertelement <2 x float> poison, float %4391, i64 0, !dbg !43 + %8563 = insertelement <2 x float> %8562, float %4392, i64 1, !dbg !43 + %8564 = insertelement <2 x float> poison, float %7399, i64 0, !dbg !43 + %8565 = insertelement <2 x float> %8564, float %7400, i64 1, !dbg !43 + %8566 = fsub <2 x float> %8563, %8565, !dbg !43 + %8567 = insertelement <2 x float> poison, float %4393, i64 0, !dbg !43 + %8568 = insertelement <2 x float> %8567, float %4394, i64 1, !dbg !43 + %8569 = insertelement <2 x float> poison, float %7401, i64 0, !dbg !43 + %8570 = insertelement <2 x float> %8569, float %7402, i64 1, !dbg !43 + %8571 = fsub <2 x float> %8568, %8570, !dbg !43 + %8572 = insertelement <2 x float> poison, float %4395, i64 0, !dbg !43 + %8573 = insertelement <2 x float> %8572, float %4396, i64 1, !dbg !43 + %8574 = insertelement <2 x float> poison, float %7403, i64 0, !dbg !43 + %8575 = insertelement <2 x float> %8574, float %7404, i64 1, !dbg !43 + %8576 = fsub <2 x float> %8573, %8575, !dbg !43 + %8577 = insertelement <2 x float> poison, float %4397, i64 0, !dbg !43 + %8578 = insertelement <2 x float> %8577, float %4398, i64 1, !dbg !43 + %8579 = insertelement <2 x float> poison, float %7405, i64 0, !dbg !43 + %8580 = insertelement <2 x float> %8579, float %7406, i64 1, !dbg !43 + %8581 = fsub <2 x float> %8578, %8580, !dbg !43 + %8582 = insertelement <2 x float> poison, float %4399, i64 0, !dbg !43 + %8583 = insertelement <2 x float> %8582, float %4400, i64 1, !dbg !43 + %8584 = insertelement <2 x float> poison, float %7407, i64 0, !dbg !43 + %8585 = insertelement <2 x float> %8584, float %7408, i64 1, !dbg !43 + %8586 = fsub <2 x float> %8583, %8585, !dbg !43 + %8587 = insertelement <2 x float> poison, float %4401, i64 0, !dbg !43 + %8588 = insertelement <2 x float> %8587, float %4402, i64 1, !dbg !43 + %8589 = insertelement <2 x float> poison, float %7409, i64 0, !dbg !43 + %8590 = insertelement <2 x float> %8589, float %7410, i64 1, !dbg !43 + %8591 = fsub <2 x float> %8588, %8590, !dbg !43 + %8592 = insertelement <2 x float> poison, float %4403, i64 0, !dbg !43 + %8593 = insertelement <2 x float> %8592, float %4404, i64 1, !dbg !43 + %8594 = insertelement <2 x float> poison, float %7411, i64 0, !dbg !43 + %8595 = insertelement <2 x float> %8594, float %7412, i64 1, !dbg !43 + %8596 = fsub <2 x float> %8593, %8595, !dbg !43 + %8597 = insertelement <2 x float> poison, float %4405, i64 0, !dbg !43 + %8598 = insertelement <2 x float> %8597, float %4406, i64 1, !dbg !43 + %8599 = insertelement <2 x float> poison, float %7413, i64 0, !dbg !43 + %8600 = insertelement <2 x float> %8599, float %7414, i64 1, !dbg !43 + %8601 = fsub <2 x float> %8598, %8600, !dbg !43 + %8602 = insertelement <2 x float> poison, float %4407, i64 0, !dbg !43 + %8603 = insertelement <2 x float> %8602, float %4408, i64 1, !dbg !43 + %8604 = insertelement <2 x float> poison, float %7415, i64 0, !dbg !43 + %8605 = insertelement <2 x float> %8604, float %7416, i64 1, !dbg !43 + %8606 = fsub <2 x float> %8603, %8605, !dbg !43 + %8607 = insertelement <2 x float> poison, float %4409, i64 0, !dbg !43 + %8608 = insertelement <2 x float> %8607, float %4410, i64 1, !dbg !43 + %8609 = insertelement <2 x float> poison, float %7417, i64 0, !dbg !43 + %8610 = insertelement <2 x float> %8609, float %7418, i64 1, !dbg !43 + %8611 = fsub <2 x float> %8608, %8610, !dbg !43 + %8612 = insertelement <2 x float> poison, float %4411, i64 0, !dbg !43 + %8613 = insertelement <2 x float> %8612, float %4412, i64 1, !dbg !43 + %8614 = insertelement <2 x float> poison, float %7419, i64 0, !dbg !43 + %8615 = insertelement <2 x float> %8614, float %7420, i64 1, !dbg !43 + %8616 = fsub <2 x float> %8613, %8615, !dbg !43 + %8617 = insertelement <2 x float> poison, float %4413, i64 0, !dbg !43 + %8618 = insertelement <2 x float> %8617, float %4414, i64 1, !dbg !43 + %8619 = insertelement <2 x float> poison, float %7421, i64 0, !dbg !43 + %8620 = insertelement <2 x float> %8619, float %7422, i64 1, !dbg !43 + %8621 = fsub <2 x float> %8618, %8620, !dbg !43 + %8622 = insertelement <2 x float> poison, float %4415, i64 0, !dbg !43 + %8623 = insertelement <2 x float> %8622, float %4416, i64 1, !dbg !43 + %8624 = insertelement <2 x float> poison, float %7423, i64 0, !dbg !43 + %8625 = insertelement <2 x float> %8624, float %7424, i64 1, !dbg !43 + %8626 = fsub <2 x float> %8623, %8625, !dbg !43 + %8627 = insertelement <2 x float> poison, float %4417, i64 0, !dbg !43 + %8628 = insertelement <2 x float> %8627, float %4418, i64 1, !dbg !43 + %8629 = insertelement <2 x float> poison, float %7425, i64 0, !dbg !43 + %8630 = insertelement <2 x float> %8629, float %7426, i64 1, !dbg !43 + %8631 = fsub <2 x float> %8628, %8630, !dbg !43 + %8632 = insertelement <2 x float> poison, float %4419, i64 0, !dbg !43 + %8633 = insertelement <2 x float> %8632, float %4420, i64 1, !dbg !43 + %8634 = insertelement <2 x float> poison, float %7427, i64 0, !dbg !43 + %8635 = insertelement <2 x float> %8634, float %7428, i64 1, !dbg !43 + %8636 = fsub <2 x float> %8633, %8635, !dbg !43 + %8637 = insertelement <2 x float> poison, float %4421, i64 0, !dbg !43 + %8638 = insertelement <2 x float> %8637, float %4422, i64 1, !dbg !43 + %8639 = insertelement <2 x float> poison, float %7429, i64 0, !dbg !43 + %8640 = insertelement <2 x float> %8639, float %7430, i64 1, !dbg !43 + %8641 = fsub <2 x float> %8638, %8640, !dbg !43 + %8642 = insertelement <2 x float> poison, float %4423, i64 0, !dbg !43 + %8643 = insertelement <2 x float> %8642, float %4424, i64 1, !dbg !43 + %8644 = insertelement <2 x float> poison, float %7431, i64 0, !dbg !43 + %8645 = insertelement <2 x float> %8644, float %7432, i64 1, !dbg !43 + %8646 = fsub <2 x float> %8643, %8645, !dbg !43 + %8647 = insertelement <2 x float> poison, float %4425, i64 0, !dbg !43 + %8648 = insertelement <2 x float> %8647, float %4426, i64 1, !dbg !43 + %8649 = insertelement <2 x float> poison, float %7433, i64 0, !dbg !43 + %8650 = insertelement <2 x float> %8649, float %7434, i64 1, !dbg !43 + %8651 = fsub <2 x float> %8648, %8650, !dbg !43 + %8652 = insertelement <2 x float> poison, float %4427, i64 0, !dbg !43 + %8653 = insertelement <2 x float> %8652, float %4428, i64 1, !dbg !43 + %8654 = insertelement <2 x float> poison, float %7435, i64 0, !dbg !43 + %8655 = insertelement <2 x float> %8654, float %7436, i64 1, !dbg !43 + %8656 = fsub <2 x float> %8653, %8655, !dbg !43 + %8657 = insertelement <2 x float> poison, float %4429, i64 0, !dbg !43 + %8658 = insertelement <2 x float> %8657, float %4430, i64 1, !dbg !43 + %8659 = insertelement <2 x float> poison, float %7437, i64 0, !dbg !43 + %8660 = insertelement <2 x float> %8659, float %7438, i64 1, !dbg !43 + %8661 = fsub <2 x float> %8658, %8660, !dbg !43 + %8662 = insertelement <2 x float> poison, float %4431, i64 0, !dbg !43 + %8663 = insertelement <2 x float> %8662, float %4432, i64 1, !dbg !43 + %8664 = insertelement <2 x float> poison, float %7439, i64 0, !dbg !43 + %8665 = insertelement <2 x float> %8664, float %7440, i64 1, !dbg !43 + %8666 = fsub <2 x float> %8663, %8665, !dbg !43 + %8667 = insertelement <2 x float> poison, float %4433, i64 0, !dbg !43 + %8668 = insertelement <2 x float> %8667, float %4434, i64 1, !dbg !43 + %8669 = insertelement <2 x float> poison, float %7441, i64 0, !dbg !43 + %8670 = insertelement <2 x float> %8669, float %7442, i64 1, !dbg !43 + %8671 = fsub <2 x float> %8668, %8670, !dbg !43 + %8672 = insertelement <2 x float> poison, float %4435, i64 0, !dbg !43 + %8673 = insertelement <2 x float> %8672, float %4436, i64 1, !dbg !43 + %8674 = insertelement <2 x float> poison, float %7443, i64 0, !dbg !43 + %8675 = insertelement <2 x float> %8674, float %7444, i64 1, !dbg !43 + %8676 = fsub <2 x float> %8673, %8675, !dbg !43 + %8677 = insertelement <2 x float> poison, float %4437, i64 0, !dbg !43 + %8678 = insertelement <2 x float> %8677, float %4438, i64 1, !dbg !43 + %8679 = insertelement <2 x float> poison, float %7445, i64 0, !dbg !43 + %8680 = insertelement <2 x float> %8679, float %7446, i64 1, !dbg !43 + %8681 = fsub <2 x float> %8678, %8680, !dbg !43 + %8682 = insertelement <2 x float> poison, float %4439, i64 0, !dbg !43 + %8683 = insertelement <2 x float> %8682, float %4440, i64 1, !dbg !43 + %8684 = insertelement <2 x float> poison, float %7447, i64 0, !dbg !43 + %8685 = insertelement <2 x float> %8684, float %7448, i64 1, !dbg !43 + %8686 = fsub <2 x float> %8683, %8685, !dbg !43 + %8687 = insertelement <2 x float> poison, float %4441, i64 0, !dbg !43 + %8688 = insertelement <2 x float> %8687, float %4442, i64 1, !dbg !43 + %8689 = insertelement <2 x float> poison, float %7449, i64 0, !dbg !43 + %8690 = insertelement <2 x float> %8689, float %7450, i64 1, !dbg !43 + %8691 = fsub <2 x float> %8688, %8690, !dbg !43 + %8692 = insertelement <2 x float> poison, float %4443, i64 0, !dbg !43 + %8693 = insertelement <2 x float> %8692, float %4444, i64 1, !dbg !43 + %8694 = insertelement <2 x float> poison, float %7451, i64 0, !dbg !43 + %8695 = insertelement <2 x float> %8694, float %7452, i64 1, !dbg !43 + %8696 = fsub <2 x float> %8693, %8695, !dbg !43 + %8697 = insertelement <2 x float> poison, float %4445, i64 0, !dbg !43 + %8698 = insertelement <2 x float> %8697, float %4446, i64 1, !dbg !43 + %8699 = insertelement <2 x float> poison, float %7453, i64 0, !dbg !43 + %8700 = insertelement <2 x float> %8699, float %7454, i64 1, !dbg !43 + %8701 = fsub <2 x float> %8698, %8700, !dbg !43 + %8702 = insertelement <2 x float> poison, float %4447, i64 0, !dbg !43 + %8703 = insertelement <2 x float> %8702, float %4448, i64 1, !dbg !43 + %8704 = insertelement <2 x float> poison, float %7455, i64 0, !dbg !43 + %8705 = insertelement <2 x float> %8704, float %7456, i64 1, !dbg !43 + %8706 = fsub <2 x float> %8703, %8705, !dbg !43 + %8707 = insertelement <2 x float> poison, float %4449, i64 0, !dbg !43 + %8708 = insertelement <2 x float> %8707, float %4450, i64 1, !dbg !43 + %8709 = insertelement <2 x float> poison, float %7457, i64 0, !dbg !43 + %8710 = insertelement <2 x float> %8709, float %7458, i64 1, !dbg !43 + %8711 = fsub <2 x float> %8708, %8710, !dbg !43 + %8712 = insertelement <2 x float> poison, float %4451, i64 0, !dbg !43 + %8713 = insertelement <2 x float> %8712, float %4452, i64 1, !dbg !43 + %8714 = insertelement <2 x float> poison, float %7459, i64 0, !dbg !43 + %8715 = insertelement <2 x float> %8714, float %7460, i64 1, !dbg !43 + %8716 = fsub <2 x float> %8713, %8715, !dbg !43 + %8717 = insertelement <2 x float> poison, float %4453, i64 0, !dbg !43 + %8718 = insertelement <2 x float> %8717, float %4454, i64 1, !dbg !43 + %8719 = insertelement <2 x float> poison, float %7461, i64 0, !dbg !43 + %8720 = insertelement <2 x float> %8719, float %7462, i64 1, !dbg !43 + %8721 = fsub <2 x float> %8718, %8720, !dbg !43 + %8722 = insertelement <2 x float> poison, float %4455, i64 0, !dbg !43 + %8723 = insertelement <2 x float> %8722, float %4456, i64 1, !dbg !43 + %8724 = insertelement <2 x float> poison, float %7463, i64 0, !dbg !43 + %8725 = insertelement <2 x float> %8724, float %7464, i64 1, !dbg !43 + %8726 = fsub <2 x float> %8723, %8725, !dbg !43 + %8727 = insertelement <2 x float> poison, float %4457, i64 0, !dbg !43 + %8728 = insertelement <2 x float> %8727, float %4458, i64 1, !dbg !43 + %8729 = insertelement <2 x float> poison, float %7465, i64 0, !dbg !43 + %8730 = insertelement <2 x float> %8729, float %7466, i64 1, !dbg !43 + %8731 = fsub <2 x float> %8728, %8730, !dbg !43 + %8732 = insertelement <2 x float> poison, float %4459, i64 0, !dbg !43 + %8733 = insertelement <2 x float> %8732, float %4460, i64 1, !dbg !43 + %8734 = insertelement <2 x float> poison, float %7467, i64 0, !dbg !43 + %8735 = insertelement <2 x float> %8734, float %7468, i64 1, !dbg !43 + %8736 = fsub <2 x float> %8733, %8735, !dbg !43 + %8737 = fptrunc float %7469 to bfloat, !dbg !44 + %8738 = fptrunc float %7470 to bfloat, !dbg !44 + %8739 = fptrunc float %7471 to bfloat, !dbg !44 + %8740 = fptrunc float %7472 to bfloat, !dbg !44 + %8741 = fptrunc float %7473 to bfloat, !dbg !44 + %8742 = fptrunc float %7474 to bfloat, !dbg !44 + %8743 = fptrunc float %7475 to bfloat, !dbg !44 + %8744 = fptrunc float %7476 to bfloat, !dbg !44 + %8745 = fptrunc <2 x float> %7481 to <2 x bfloat>, !dbg !44 + %8746 = fptrunc <2 x float> %7486 to <2 x bfloat>, !dbg !44 + %8747 = fptrunc <2 x float> %7491 to <2 x bfloat>, !dbg !44 + %8748 = fptrunc <2 x float> %7496 to <2 x bfloat>, !dbg !44 + %8749 = fptrunc <2 x float> %7501 to <2 x bfloat>, !dbg !44 + %8750 = fptrunc <2 x float> %7506 to <2 x bfloat>, !dbg !44 + %8751 = fptrunc <2 x float> %7511 to <2 x bfloat>, !dbg !44 + %8752 = fptrunc <2 x float> %7516 to <2 x bfloat>, !dbg !44 + %8753 = fptrunc <2 x float> %7521 to <2 x bfloat>, !dbg !44 + %8754 = fptrunc <2 x float> %7526 to <2 x bfloat>, !dbg !44 + %8755 = fptrunc <2 x float> %7531 to <2 x bfloat>, !dbg !44 + %8756 = fptrunc <2 x float> %7536 to <2 x bfloat>, !dbg !44 + %8757 = fptrunc <2 x float> %7541 to <2 x bfloat>, !dbg !44 + %8758 = fptrunc <2 x float> %7546 to <2 x bfloat>, !dbg !44 + %8759 = fptrunc <2 x float> %7551 to <2 x bfloat>, !dbg !44 + %8760 = fptrunc <2 x float> %7556 to <2 x bfloat>, !dbg !44 + %8761 = fptrunc <2 x float> %7561 to <2 x bfloat>, !dbg !44 + %8762 = fptrunc <2 x float> %7566 to <2 x bfloat>, !dbg !44 + %8763 = fptrunc <2 x float> %7571 to <2 x bfloat>, !dbg !44 + %8764 = fptrunc <2 x float> %7576 to <2 x bfloat>, !dbg !44 + %8765 = fptrunc <2 x float> %7581 to <2 x bfloat>, !dbg !44 + %8766 = fptrunc <2 x float> %7586 to <2 x bfloat>, !dbg !44 + %8767 = fptrunc <2 x float> %7591 to <2 x bfloat>, !dbg !44 + %8768 = fptrunc <2 x float> %7596 to <2 x bfloat>, !dbg !44 + %8769 = fptrunc <2 x float> %7601 to <2 x bfloat>, !dbg !44 + %8770 = fptrunc <2 x float> %7606 to <2 x bfloat>, !dbg !44 + %8771 = fptrunc <2 x float> %7611 to <2 x bfloat>, !dbg !44 + %8772 = fptrunc <2 x float> %7616 to <2 x bfloat>, !dbg !44 + %8773 = fptrunc <2 x float> %7621 to <2 x bfloat>, !dbg !44 + %8774 = fptrunc <2 x float> %7626 to <2 x bfloat>, !dbg !44 + %8775 = fptrunc <2 x float> %7631 to <2 x bfloat>, !dbg !44 + %8776 = fptrunc <2 x float> %7636 to <2 x bfloat>, !dbg !44 + %8777 = fptrunc <2 x float> %7641 to <2 x bfloat>, !dbg !44 + %8778 = fptrunc <2 x float> %7646 to <2 x bfloat>, !dbg !44 + %8779 = fptrunc <2 x float> %7651 to <2 x bfloat>, !dbg !44 + %8780 = fptrunc <2 x float> %7656 to <2 x bfloat>, !dbg !44 + %8781 = fptrunc <2 x float> %7661 to <2 x bfloat>, !dbg !44 + %8782 = fptrunc <2 x float> %7666 to <2 x bfloat>, !dbg !44 + %8783 = fptrunc <2 x float> %7671 to <2 x bfloat>, !dbg !44 + %8784 = fptrunc <2 x float> %7676 to <2 x bfloat>, !dbg !44 + %8785 = fptrunc <2 x float> %7681 to <2 x bfloat>, !dbg !44 + %8786 = fptrunc <2 x float> %7686 to <2 x bfloat>, !dbg !44 + %8787 = fptrunc <2 x float> %7691 to <2 x bfloat>, !dbg !44 + %8788 = fptrunc <2 x float> %7696 to <2 x bfloat>, !dbg !44 + %8789 = fptrunc <2 x float> %7701 to <2 x bfloat>, !dbg !44 + %8790 = fptrunc <2 x float> %7706 to <2 x bfloat>, !dbg !44 + %8791 = fptrunc <2 x float> %7711 to <2 x bfloat>, !dbg !44 + %8792 = fptrunc <2 x float> %7716 to <2 x bfloat>, !dbg !44 + %8793 = fptrunc <2 x float> %7721 to <2 x bfloat>, !dbg !44 + %8794 = fptrunc <2 x float> %7726 to <2 x bfloat>, !dbg !44 + %8795 = fptrunc <2 x float> %7731 to <2 x bfloat>, !dbg !44 + %8796 = fptrunc <2 x float> %7736 to <2 x bfloat>, !dbg !44 + %8797 = fptrunc <2 x float> %7741 to <2 x bfloat>, !dbg !44 + %8798 = fptrunc <2 x float> %7746 to <2 x bfloat>, !dbg !44 + %8799 = fptrunc <2 x float> %7751 to <2 x bfloat>, !dbg !44 + %8800 = fptrunc <2 x float> %7756 to <2 x bfloat>, !dbg !44 + %8801 = fptrunc <2 x float> %7761 to <2 x bfloat>, !dbg !44 + %8802 = fptrunc <2 x float> %7766 to <2 x bfloat>, !dbg !44 + %8803 = fptrunc <2 x float> %7771 to <2 x bfloat>, !dbg !44 + %8804 = fptrunc <2 x float> %7776 to <2 x bfloat>, !dbg !44 + %8805 = fptrunc <2 x float> %7781 to <2 x bfloat>, !dbg !44 + %8806 = fptrunc <2 x float> %7786 to <2 x bfloat>, !dbg !44 + %8807 = fptrunc <2 x float> %7791 to <2 x bfloat>, !dbg !44 + %8808 = fptrunc <2 x float> %7796 to <2 x bfloat>, !dbg !44 + %8809 = fptrunc <2 x float> %7801 to <2 x bfloat>, !dbg !44 + %8810 = fptrunc <2 x float> %7806 to <2 x bfloat>, !dbg !44 + %8811 = fptrunc <2 x float> %7811 to <2 x bfloat>, !dbg !44 + %8812 = fptrunc <2 x float> %7816 to <2 x bfloat>, !dbg !44 + %8813 = fptrunc <2 x float> %7821 to <2 x bfloat>, !dbg !44 + %8814 = fptrunc <2 x float> %7826 to <2 x bfloat>, !dbg !44 + %8815 = fptrunc <2 x float> %7831 to <2 x bfloat>, !dbg !44 + %8816 = fptrunc <2 x float> %7836 to <2 x bfloat>, !dbg !44 + %8817 = fptrunc <2 x float> %7841 to <2 x bfloat>, !dbg !44 + %8818 = fptrunc <2 x float> %7846 to <2 x bfloat>, !dbg !44 + %8819 = fptrunc <2 x float> %7851 to <2 x bfloat>, !dbg !44 + %8820 = fptrunc <2 x float> %7856 to <2 x bfloat>, !dbg !44 + %8821 = fptrunc <2 x float> %7861 to <2 x bfloat>, !dbg !44 + %8822 = fptrunc <2 x float> %7866 to <2 x bfloat>, !dbg !44 + %8823 = fptrunc <2 x float> %7871 to <2 x bfloat>, !dbg !44 + %8824 = fptrunc <2 x float> %7876 to <2 x bfloat>, !dbg !44 + %8825 = fptrunc <2 x float> %7881 to <2 x bfloat>, !dbg !44 + %8826 = fptrunc <2 x float> %7886 to <2 x bfloat>, !dbg !44 + %8827 = fptrunc <2 x float> %7891 to <2 x bfloat>, !dbg !44 + %8828 = fptrunc <2 x float> %7896 to <2 x bfloat>, !dbg !44 + %8829 = fptrunc <2 x float> %7901 to <2 x bfloat>, !dbg !44 + %8830 = fptrunc <2 x float> %7906 to <2 x bfloat>, !dbg !44 + %8831 = fptrunc <2 x float> %7911 to <2 x bfloat>, !dbg !44 + %8832 = fptrunc <2 x float> %7916 to <2 x bfloat>, !dbg !44 + %8833 = fptrunc <2 x float> %7921 to <2 x bfloat>, !dbg !44 + %8834 = fptrunc <2 x float> %7926 to <2 x bfloat>, !dbg !44 + %8835 = fptrunc <2 x float> %7931 to <2 x bfloat>, !dbg !44 + %8836 = fptrunc <2 x float> %7936 to <2 x bfloat>, !dbg !44 + %8837 = fptrunc <2 x float> %7941 to <2 x bfloat>, !dbg !44 + %8838 = fptrunc <2 x float> %7946 to <2 x bfloat>, !dbg !44 + %8839 = fptrunc <2 x float> %7951 to <2 x bfloat>, !dbg !44 + %8840 = fptrunc <2 x float> %7956 to <2 x bfloat>, !dbg !44 + %8841 = fptrunc <2 x float> %7961 to <2 x bfloat>, !dbg !44 + %8842 = fptrunc <2 x float> %7966 to <2 x bfloat>, !dbg !44 + %8843 = fptrunc <2 x float> %7971 to <2 x bfloat>, !dbg !44 + %8844 = fptrunc <2 x float> %7976 to <2 x bfloat>, !dbg !44 + %8845 = fptrunc <2 x float> %7981 to <2 x bfloat>, !dbg !44 + %8846 = fptrunc <2 x float> %7986 to <2 x bfloat>, !dbg !44 + %8847 = fptrunc <2 x float> %7991 to <2 x bfloat>, !dbg !44 + %8848 = fptrunc <2 x float> %7996 to <2 x bfloat>, !dbg !44 + %8849 = fptrunc <2 x float> %8001 to <2 x bfloat>, !dbg !44 + %8850 = fptrunc <2 x float> %8006 to <2 x bfloat>, !dbg !44 + %8851 = fptrunc <2 x float> %8011 to <2 x bfloat>, !dbg !44 + %8852 = fptrunc <2 x float> %8016 to <2 x bfloat>, !dbg !44 + %8853 = fptrunc <2 x float> %8021 to <2 x bfloat>, !dbg !44 + %8854 = fptrunc <2 x float> %8026 to <2 x bfloat>, !dbg !44 + %8855 = fptrunc <2 x float> %8031 to <2 x bfloat>, !dbg !44 + %8856 = fptrunc <2 x float> %8036 to <2 x bfloat>, !dbg !44 + %8857 = fptrunc <2 x float> %8041 to <2 x bfloat>, !dbg !44 + %8858 = fptrunc <2 x float> %8046 to <2 x bfloat>, !dbg !44 + %8859 = fptrunc <2 x float> %8051 to <2 x bfloat>, !dbg !44 + %8860 = fptrunc <2 x float> %8056 to <2 x bfloat>, !dbg !44 + %8861 = fptrunc <2 x float> %8061 to <2 x bfloat>, !dbg !44 + %8862 = fptrunc <2 x float> %8066 to <2 x bfloat>, !dbg !44 + %8863 = fptrunc <2 x float> %8071 to <2 x bfloat>, !dbg !44 + %8864 = fptrunc <2 x float> %8076 to <2 x bfloat>, !dbg !44 + %8865 = fptrunc <2 x float> %8081 to <2 x bfloat>, !dbg !44 + %8866 = fptrunc <2 x float> %8086 to <2 x bfloat>, !dbg !44 + %8867 = fptrunc <2 x float> %8091 to <2 x bfloat>, !dbg !44 + %8868 = fptrunc <2 x float> %8096 to <2 x bfloat>, !dbg !44 + %8869 = fptrunc <2 x float> %8101 to <2 x bfloat>, !dbg !44 + %8870 = fptrunc <2 x float> %8106 to <2 x bfloat>, !dbg !44 + %8871 = fptrunc <2 x float> %8111 to <2 x bfloat>, !dbg !44 + %8872 = fptrunc <2 x float> %8116 to <2 x bfloat>, !dbg !44 + %8873 = fptrunc <2 x float> %8121 to <2 x bfloat>, !dbg !44 + %8874 = fptrunc <2 x float> %8126 to <2 x bfloat>, !dbg !44 + %8875 = fptrunc <2 x float> %8131 to <2 x bfloat>, !dbg !44 + %8876 = fptrunc <2 x float> %8136 to <2 x bfloat>, !dbg !44 + %8877 = fptrunc <2 x float> %8141 to <2 x bfloat>, !dbg !44 + %8878 = fptrunc <2 x float> %8146 to <2 x bfloat>, !dbg !44 + %8879 = fptrunc <2 x float> %8151 to <2 x bfloat>, !dbg !44 + %8880 = fptrunc <2 x float> %8156 to <2 x bfloat>, !dbg !44 + %8881 = fptrunc <2 x float> %8161 to <2 x bfloat>, !dbg !44 + %8882 = fptrunc <2 x float> %8166 to <2 x bfloat>, !dbg !44 + %8883 = fptrunc <2 x float> %8171 to <2 x bfloat>, !dbg !44 + %8884 = fptrunc <2 x float> %8176 to <2 x bfloat>, !dbg !44 + %8885 = fptrunc <2 x float> %8181 to <2 x bfloat>, !dbg !44 + %8886 = fptrunc <2 x float> %8186 to <2 x bfloat>, !dbg !44 + %8887 = fptrunc <2 x float> %8191 to <2 x bfloat>, !dbg !44 + %8888 = fptrunc <2 x float> %8196 to <2 x bfloat>, !dbg !44 + %8889 = fptrunc <2 x float> %8201 to <2 x bfloat>, !dbg !44 + %8890 = fptrunc <2 x float> %8206 to <2 x bfloat>, !dbg !44 + %8891 = fptrunc <2 x float> %8211 to <2 x bfloat>, !dbg !44 + %8892 = fptrunc <2 x float> %8216 to <2 x bfloat>, !dbg !44 + %8893 = fptrunc <2 x float> %8221 to <2 x bfloat>, !dbg !44 + %8894 = fptrunc <2 x float> %8226 to <2 x bfloat>, !dbg !44 + %8895 = fptrunc <2 x float> %8231 to <2 x bfloat>, !dbg !44 + %8896 = fptrunc <2 x float> %8236 to <2 x bfloat>, !dbg !44 + %8897 = fptrunc <2 x float> %8241 to <2 x bfloat>, !dbg !44 + %8898 = fptrunc <2 x float> %8246 to <2 x bfloat>, !dbg !44 + %8899 = fptrunc <2 x float> %8251 to <2 x bfloat>, !dbg !44 + %8900 = fptrunc <2 x float> %8256 to <2 x bfloat>, !dbg !44 + %8901 = fptrunc <2 x float> %8261 to <2 x bfloat>, !dbg !44 + %8902 = fptrunc <2 x float> %8266 to <2 x bfloat>, !dbg !44 + %8903 = fptrunc <2 x float> %8271 to <2 x bfloat>, !dbg !44 + %8904 = fptrunc <2 x float> %8276 to <2 x bfloat>, !dbg !44 + %8905 = fptrunc <2 x float> %8281 to <2 x bfloat>, !dbg !44 + %8906 = fptrunc <2 x float> %8286 to <2 x bfloat>, !dbg !44 + %8907 = fptrunc <2 x float> %8291 to <2 x bfloat>, !dbg !44 + %8908 = fptrunc <2 x float> %8296 to <2 x bfloat>, !dbg !44 + %8909 = fptrunc <2 x float> %8301 to <2 x bfloat>, !dbg !44 + %8910 = fptrunc <2 x float> %8306 to <2 x bfloat>, !dbg !44 + %8911 = fptrunc <2 x float> %8311 to <2 x bfloat>, !dbg !44 + %8912 = fptrunc <2 x float> %8316 to <2 x bfloat>, !dbg !44 + %8913 = fptrunc <2 x float> %8321 to <2 x bfloat>, !dbg !44 + %8914 = fptrunc <2 x float> %8326 to <2 x bfloat>, !dbg !44 + %8915 = fptrunc <2 x float> %8331 to <2 x bfloat>, !dbg !44 + %8916 = fptrunc <2 x float> %8336 to <2 x bfloat>, !dbg !44 + %8917 = fptrunc <2 x float> %8341 to <2 x bfloat>, !dbg !44 + %8918 = fptrunc <2 x float> %8346 to <2 x bfloat>, !dbg !44 + %8919 = fptrunc <2 x float> %8351 to <2 x bfloat>, !dbg !44 + %8920 = fptrunc <2 x float> %8356 to <2 x bfloat>, !dbg !44 + %8921 = fptrunc <2 x float> %8361 to <2 x bfloat>, !dbg !44 + %8922 = fptrunc <2 x float> %8366 to <2 x bfloat>, !dbg !44 + %8923 = fptrunc <2 x float> %8371 to <2 x bfloat>, !dbg !44 + %8924 = fptrunc <2 x float> %8376 to <2 x bfloat>, !dbg !44 + %8925 = fptrunc <2 x float> %8381 to <2 x bfloat>, !dbg !44 + %8926 = fptrunc <2 x float> %8386 to <2 x bfloat>, !dbg !44 + %8927 = fptrunc <2 x float> %8391 to <2 x bfloat>, !dbg !44 + %8928 = fptrunc <2 x float> %8396 to <2 x bfloat>, !dbg !44 + %8929 = fptrunc <2 x float> %8401 to <2 x bfloat>, !dbg !44 + %8930 = fptrunc <2 x float> %8406 to <2 x bfloat>, !dbg !44 + %8931 = fptrunc <2 x float> %8411 to <2 x bfloat>, !dbg !44 + %8932 = fptrunc <2 x float> %8416 to <2 x bfloat>, !dbg !44 + %8933 = fptrunc <2 x float> %8421 to <2 x bfloat>, !dbg !44 + %8934 = fptrunc <2 x float> %8426 to <2 x bfloat>, !dbg !44 + %8935 = fptrunc <2 x float> %8431 to <2 x bfloat>, !dbg !44 + %8936 = fptrunc <2 x float> %8436 to <2 x bfloat>, !dbg !44 + %8937 = fptrunc <2 x float> %8441 to <2 x bfloat>, !dbg !44 + %8938 = fptrunc <2 x float> %8446 to <2 x bfloat>, !dbg !44 + %8939 = fptrunc <2 x float> %8451 to <2 x bfloat>, !dbg !44 + %8940 = fptrunc <2 x float> %8456 to <2 x bfloat>, !dbg !44 + %8941 = fptrunc <2 x float> %8461 to <2 x bfloat>, !dbg !44 + %8942 = fptrunc <2 x float> %8466 to <2 x bfloat>, !dbg !44 + %8943 = fptrunc <2 x float> %8471 to <2 x bfloat>, !dbg !44 + %8944 = fptrunc <2 x float> %8476 to <2 x bfloat>, !dbg !44 + %8945 = fptrunc <2 x float> %8481 to <2 x bfloat>, !dbg !44 + %8946 = fptrunc <2 x float> %8486 to <2 x bfloat>, !dbg !44 + %8947 = fptrunc <2 x float> %8491 to <2 x bfloat>, !dbg !44 + %8948 = fptrunc <2 x float> %8496 to <2 x bfloat>, !dbg !44 + %8949 = fptrunc <2 x float> %8501 to <2 x bfloat>, !dbg !44 + %8950 = fptrunc <2 x float> %8506 to <2 x bfloat>, !dbg !44 + %8951 = fptrunc <2 x float> %8511 to <2 x bfloat>, !dbg !44 + %8952 = fptrunc <2 x float> %8516 to <2 x bfloat>, !dbg !44 + %8953 = fptrunc <2 x float> %8521 to <2 x bfloat>, !dbg !44 + %8954 = fptrunc <2 x float> %8526 to <2 x bfloat>, !dbg !44 + %8955 = fptrunc <2 x float> %8531 to <2 x bfloat>, !dbg !44 + %8956 = fptrunc <2 x float> %8536 to <2 x bfloat>, !dbg !44 + %8957 = fptrunc <2 x float> %8541 to <2 x bfloat>, !dbg !44 + %8958 = fptrunc <2 x float> %8546 to <2 x bfloat>, !dbg !44 + %8959 = fptrunc <2 x float> %8551 to <2 x bfloat>, !dbg !44 + %8960 = fptrunc <2 x float> %8556 to <2 x bfloat>, !dbg !44 + %8961 = fptrunc <2 x float> %8561 to <2 x bfloat>, !dbg !44 + %8962 = fptrunc <2 x float> %8566 to <2 x bfloat>, !dbg !44 + %8963 = fptrunc <2 x float> %8571 to <2 x bfloat>, !dbg !44 + %8964 = fptrunc <2 x float> %8576 to <2 x bfloat>, !dbg !44 + %8965 = fptrunc <2 x float> %8581 to <2 x bfloat>, !dbg !44 + %8966 = fptrunc <2 x float> %8586 to <2 x bfloat>, !dbg !44 + %8967 = fptrunc <2 x float> %8591 to <2 x bfloat>, !dbg !44 + %8968 = fptrunc <2 x float> %8596 to <2 x bfloat>, !dbg !44 + %8969 = fptrunc <2 x float> %8601 to <2 x bfloat>, !dbg !44 + %8970 = fptrunc <2 x float> %8606 to <2 x bfloat>, !dbg !44 + %8971 = fptrunc <2 x float> %8611 to <2 x bfloat>, !dbg !44 + %8972 = fptrunc <2 x float> %8616 to <2 x bfloat>, !dbg !44 + %8973 = fptrunc <2 x float> %8621 to <2 x bfloat>, !dbg !44 + %8974 = fptrunc <2 x float> %8626 to <2 x bfloat>, !dbg !44 + %8975 = fptrunc <2 x float> %8631 to <2 x bfloat>, !dbg !44 + %8976 = fptrunc <2 x float> %8636 to <2 x bfloat>, !dbg !44 + %8977 = fptrunc <2 x float> %8641 to <2 x bfloat>, !dbg !44 + %8978 = fptrunc <2 x float> %8646 to <2 x bfloat>, !dbg !44 + %8979 = fptrunc <2 x float> %8651 to <2 x bfloat>, !dbg !44 + %8980 = fptrunc <2 x float> %8656 to <2 x bfloat>, !dbg !44 + %8981 = fptrunc <2 x float> %8661 to <2 x bfloat>, !dbg !44 + %8982 = fptrunc <2 x float> %8666 to <2 x bfloat>, !dbg !44 + %8983 = fptrunc <2 x float> %8671 to <2 x bfloat>, !dbg !44 + %8984 = fptrunc <2 x float> %8676 to <2 x bfloat>, !dbg !44 + %8985 = fptrunc <2 x float> %8681 to <2 x bfloat>, !dbg !44 + %8986 = fptrunc <2 x float> %8686 to <2 x bfloat>, !dbg !44 + %8987 = fptrunc <2 x float> %8691 to <2 x bfloat>, !dbg !44 + %8988 = fptrunc <2 x float> %8696 to <2 x bfloat>, !dbg !44 + %8989 = fptrunc <2 x float> %8701 to <2 x bfloat>, !dbg !44 + %8990 = fptrunc <2 x float> %8706 to <2 x bfloat>, !dbg !44 + %8991 = fptrunc <2 x float> %8711 to <2 x bfloat>, !dbg !44 + %8992 = fptrunc <2 x float> %8716 to <2 x bfloat>, !dbg !44 + %8993 = fptrunc <2 x float> %8721 to <2 x bfloat>, !dbg !44 + %8994 = fptrunc <2 x float> %8726 to <2 x bfloat>, !dbg !44 + %8995 = fptrunc <2 x float> %8731 to <2 x bfloat>, !dbg !44 + %8996 = fptrunc <2 x float> %8736 to <2 x bfloat>, !dbg !44 + %8997 = insertelement <2 x bfloat> poison, bfloat %8737, i64 0, !dbg !45 + %8998 = insertelement <2 x bfloat> %8997, bfloat %8738, i64 1, !dbg !45 + %8999 = bitcast <2 x bfloat> %8998 to i32, !dbg !45 + %9000 = insertelement <2 x bfloat> poison, bfloat %8739, i64 0, !dbg !45 + %9001 = insertelement <2 x bfloat> %9000, bfloat %8740, i64 1, !dbg !45 + %9002 = bitcast <2 x bfloat> %9001 to i32, !dbg !45 + %9003 = insertelement <2 x bfloat> poison, bfloat %8741, i64 0, !dbg !45 + %9004 = insertelement <2 x bfloat> %9003, bfloat %8742, i64 1, !dbg !45 + %9005 = bitcast <2 x bfloat> %9004 to i32, !dbg !45 + %9006 = insertelement <2 x bfloat> poison, bfloat %8743, i64 0, !dbg !45 + %9007 = insertelement <2 x bfloat> %9006, bfloat %8744, i64 1, !dbg !45 + %9008 = bitcast <2 x bfloat> %9007 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %8999, i32 %9002, i32 %9005, i32 %9008, ptr addrspace(1) %518, i1 %27) #5, !dbg !45 + %9009 = bitcast <2 x bfloat> %8745 to i32, !dbg !45 + %9010 = bitcast <2 x bfloat> %8746 to i32, !dbg !45 + %9011 = bitcast <2 x bfloat> %8747 to i32, !dbg !45 + %9012 = bitcast <2 x bfloat> %8748 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9009, i32 %9010, i32 %9011, i32 %9012, ptr addrspace(1) %519, i1 %28) #5, !dbg !45 + %9013 = bitcast <2 x bfloat> %8749 to i32, !dbg !45 + %9014 = bitcast <2 x bfloat> %8750 to i32, !dbg !45 + %9015 = bitcast <2 x bfloat> %8751 to i32, !dbg !45 + %9016 = bitcast <2 x bfloat> %8752 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9013, i32 %9014, i32 %9015, i32 %9016, ptr addrspace(1) %520, i1 %29) #5, !dbg !45 + %9017 = bitcast <2 x bfloat> %8753 to i32, !dbg !45 + %9018 = bitcast <2 x bfloat> %8754 to i32, !dbg !45 + %9019 = bitcast <2 x bfloat> %8755 to i32, !dbg !45 + %9020 = bitcast <2 x bfloat> %8756 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9017, i32 %9018, i32 %9019, i32 %9020, ptr addrspace(1) %521, i1 %30) #5, !dbg !45 + %9021 = bitcast <2 x bfloat> %8757 to i32, !dbg !45 + %9022 = bitcast <2 x bfloat> %8758 to i32, !dbg !45 + %9023 = bitcast <2 x bfloat> %8759 to i32, !dbg !45 + %9024 = bitcast <2 x bfloat> %8760 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9021, i32 %9022, i32 %9023, i32 %9024, ptr addrspace(1) %522, i1 %650) #5, !dbg !45 + %9025 = bitcast <2 x bfloat> %8761 to i32, !dbg !45 + %9026 = bitcast <2 x bfloat> %8762 to i32, !dbg !45 + %9027 = bitcast <2 x bfloat> %8763 to i32, !dbg !45 + %9028 = bitcast <2 x bfloat> %8764 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9025, i32 %9026, i32 %9027, i32 %9028, ptr addrspace(1) %523, i1 %668) #5, !dbg !45 + %9029 = bitcast <2 x bfloat> %8765 to i32, !dbg !45 + %9030 = bitcast <2 x bfloat> %8766 to i32, !dbg !45 + %9031 = bitcast <2 x bfloat> %8767 to i32, !dbg !45 + %9032 = bitcast <2 x bfloat> %8768 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9029, i32 %9030, i32 %9031, i32 %9032, ptr addrspace(1) %524, i1 %686) #5, !dbg !45 + %9033 = bitcast <2 x bfloat> %8769 to i32, !dbg !45 + %9034 = bitcast <2 x bfloat> %8770 to i32, !dbg !45 + %9035 = bitcast <2 x bfloat> %8771 to i32, !dbg !45 + %9036 = bitcast <2 x bfloat> %8772 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9033, i32 %9034, i32 %9035, i32 %9036, ptr addrspace(1) %525, i1 %704) #5, !dbg !45 + %9037 = bitcast <2 x bfloat> %8773 to i32, !dbg !45 + %9038 = bitcast <2 x bfloat> %8774 to i32, !dbg !45 + %9039 = bitcast <2 x bfloat> %8775 to i32, !dbg !45 + %9040 = bitcast <2 x bfloat> %8776 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9037, i32 %9038, i32 %9039, i32 %9040, ptr addrspace(1) %526, i1 %722) #5, !dbg !45 + %9041 = bitcast <2 x bfloat> %8777 to i32, !dbg !45 + %9042 = bitcast <2 x bfloat> %8778 to i32, !dbg !45 + %9043 = bitcast <2 x bfloat> %8779 to i32, !dbg !45 + %9044 = bitcast <2 x bfloat> %8780 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9041, i32 %9042, i32 %9043, i32 %9044, ptr addrspace(1) %527, i1 %740) #5, !dbg !45 + %9045 = bitcast <2 x bfloat> %8781 to i32, !dbg !45 + %9046 = bitcast <2 x bfloat> %8782 to i32, !dbg !45 + %9047 = bitcast <2 x bfloat> %8783 to i32, !dbg !45 + %9048 = bitcast <2 x bfloat> %8784 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9045, i32 %9046, i32 %9047, i32 %9048, ptr addrspace(1) %528, i1 %758) #5, !dbg !45 + %9049 = bitcast <2 x bfloat> %8785 to i32, !dbg !45 + %9050 = bitcast <2 x bfloat> %8786 to i32, !dbg !45 + %9051 = bitcast <2 x bfloat> %8787 to i32, !dbg !45 + %9052 = bitcast <2 x bfloat> %8788 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9049, i32 %9050, i32 %9051, i32 %9052, ptr addrspace(1) %529, i1 %776) #5, !dbg !45 + %9053 = bitcast <2 x bfloat> %8789 to i32, !dbg !45 + %9054 = bitcast <2 x bfloat> %8790 to i32, !dbg !45 + %9055 = bitcast <2 x bfloat> %8791 to i32, !dbg !45 + %9056 = bitcast <2 x bfloat> %8792 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9053, i32 %9054, i32 %9055, i32 %9056, ptr addrspace(1) %530, i1 %794) #5, !dbg !45 + %9057 = bitcast <2 x bfloat> %8793 to i32, !dbg !45 + %9058 = bitcast <2 x bfloat> %8794 to i32, !dbg !45 + %9059 = bitcast <2 x bfloat> %8795 to i32, !dbg !45 + %9060 = bitcast <2 x bfloat> %8796 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9057, i32 %9058, i32 %9059, i32 %9060, ptr addrspace(1) %531, i1 %812) #5, !dbg !45 + %9061 = bitcast <2 x bfloat> %8797 to i32, !dbg !45 + %9062 = bitcast <2 x bfloat> %8798 to i32, !dbg !45 + %9063 = bitcast <2 x bfloat> %8799 to i32, !dbg !45 + %9064 = bitcast <2 x bfloat> %8800 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9061, i32 %9062, i32 %9063, i32 %9064, ptr addrspace(1) %532, i1 %830) #5, !dbg !45 + %9065 = bitcast <2 x bfloat> %8801 to i32, !dbg !45 + %9066 = bitcast <2 x bfloat> %8802 to i32, !dbg !45 + %9067 = bitcast <2 x bfloat> %8803 to i32, !dbg !45 + %9068 = bitcast <2 x bfloat> %8804 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9065, i32 %9066, i32 %9067, i32 %9068, ptr addrspace(1) %533, i1 %848) #5, !dbg !45 + %9069 = bitcast <2 x bfloat> %8805 to i32, !dbg !45 + %9070 = bitcast <2 x bfloat> %8806 to i32, !dbg !45 + %9071 = bitcast <2 x bfloat> %8807 to i32, !dbg !45 + %9072 = bitcast <2 x bfloat> %8808 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9069, i32 %9070, i32 %9071, i32 %9072, ptr addrspace(1) %534, i1 %866) #5, !dbg !45 + %9073 = bitcast <2 x bfloat> %8809 to i32, !dbg !45 + %9074 = bitcast <2 x bfloat> %8810 to i32, !dbg !45 + %9075 = bitcast <2 x bfloat> %8811 to i32, !dbg !45 + %9076 = bitcast <2 x bfloat> %8812 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9073, i32 %9074, i32 %9075, i32 %9076, ptr addrspace(1) %535, i1 %884) #5, !dbg !45 + %9077 = bitcast <2 x bfloat> %8813 to i32, !dbg !45 + %9078 = bitcast <2 x bfloat> %8814 to i32, !dbg !45 + %9079 = bitcast <2 x bfloat> %8815 to i32, !dbg !45 + %9080 = bitcast <2 x bfloat> %8816 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9077, i32 %9078, i32 %9079, i32 %9080, ptr addrspace(1) %536, i1 %902) #5, !dbg !45 + %9081 = bitcast <2 x bfloat> %8817 to i32, !dbg !45 + %9082 = bitcast <2 x bfloat> %8818 to i32, !dbg !45 + %9083 = bitcast <2 x bfloat> %8819 to i32, !dbg !45 + %9084 = bitcast <2 x bfloat> %8820 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9081, i32 %9082, i32 %9083, i32 %9084, ptr addrspace(1) %537, i1 %920) #5, !dbg !45 + %9085 = bitcast <2 x bfloat> %8821 to i32, !dbg !45 + %9086 = bitcast <2 x bfloat> %8822 to i32, !dbg !45 + %9087 = bitcast <2 x bfloat> %8823 to i32, !dbg !45 + %9088 = bitcast <2 x bfloat> %8824 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9085, i32 %9086, i32 %9087, i32 %9088, ptr addrspace(1) %538, i1 %938) #5, !dbg !45 + %9089 = bitcast <2 x bfloat> %8825 to i32, !dbg !45 + %9090 = bitcast <2 x bfloat> %8826 to i32, !dbg !45 + %9091 = bitcast <2 x bfloat> %8827 to i32, !dbg !45 + %9092 = bitcast <2 x bfloat> %8828 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9089, i32 %9090, i32 %9091, i32 %9092, ptr addrspace(1) %539, i1 %956) #5, !dbg !45 + %9093 = bitcast <2 x bfloat> %8829 to i32, !dbg !45 + %9094 = bitcast <2 x bfloat> %8830 to i32, !dbg !45 + %9095 = bitcast <2 x bfloat> %8831 to i32, !dbg !45 + %9096 = bitcast <2 x bfloat> %8832 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9093, i32 %9094, i32 %9095, i32 %9096, ptr addrspace(1) %540, i1 %974) #5, !dbg !45 + %9097 = bitcast <2 x bfloat> %8833 to i32, !dbg !45 + %9098 = bitcast <2 x bfloat> %8834 to i32, !dbg !45 + %9099 = bitcast <2 x bfloat> %8835 to i32, !dbg !45 + %9100 = bitcast <2 x bfloat> %8836 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9097, i32 %9098, i32 %9099, i32 %9100, ptr addrspace(1) %541, i1 %992) #5, !dbg !45 + %9101 = bitcast <2 x bfloat> %8837 to i32, !dbg !45 + %9102 = bitcast <2 x bfloat> %8838 to i32, !dbg !45 + %9103 = bitcast <2 x bfloat> %8839 to i32, !dbg !45 + %9104 = bitcast <2 x bfloat> %8840 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9101, i32 %9102, i32 %9103, i32 %9104, ptr addrspace(1) %542, i1 %1010) #5, !dbg !45 + %9105 = bitcast <2 x bfloat> %8841 to i32, !dbg !45 + %9106 = bitcast <2 x bfloat> %8842 to i32, !dbg !45 + %9107 = bitcast <2 x bfloat> %8843 to i32, !dbg !45 + %9108 = bitcast <2 x bfloat> %8844 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9105, i32 %9106, i32 %9107, i32 %9108, ptr addrspace(1) %543, i1 %1028) #5, !dbg !45 + %9109 = bitcast <2 x bfloat> %8845 to i32, !dbg !45 + %9110 = bitcast <2 x bfloat> %8846 to i32, !dbg !45 + %9111 = bitcast <2 x bfloat> %8847 to i32, !dbg !45 + %9112 = bitcast <2 x bfloat> %8848 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9109, i32 %9110, i32 %9111, i32 %9112, ptr addrspace(1) %544, i1 %1046) #5, !dbg !45 + %9113 = bitcast <2 x bfloat> %8849 to i32, !dbg !45 + %9114 = bitcast <2 x bfloat> %8850 to i32, !dbg !45 + %9115 = bitcast <2 x bfloat> %8851 to i32, !dbg !45 + %9116 = bitcast <2 x bfloat> %8852 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9113, i32 %9114, i32 %9115, i32 %9116, ptr addrspace(1) %545, i1 %1064) #5, !dbg !45 + %9117 = bitcast <2 x bfloat> %8853 to i32, !dbg !45 + %9118 = bitcast <2 x bfloat> %8854 to i32, !dbg !45 + %9119 = bitcast <2 x bfloat> %8855 to i32, !dbg !45 + %9120 = bitcast <2 x bfloat> %8856 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9117, i32 %9118, i32 %9119, i32 %9120, ptr addrspace(1) %546, i1 %1082) #5, !dbg !45 + %9121 = bitcast <2 x bfloat> %8857 to i32, !dbg !45 + %9122 = bitcast <2 x bfloat> %8858 to i32, !dbg !45 + %9123 = bitcast <2 x bfloat> %8859 to i32, !dbg !45 + %9124 = bitcast <2 x bfloat> %8860 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9121, i32 %9122, i32 %9123, i32 %9124, ptr addrspace(1) %547, i1 %1100) #5, !dbg !45 + %9125 = bitcast <2 x bfloat> %8861 to i32, !dbg !45 + %9126 = bitcast <2 x bfloat> %8862 to i32, !dbg !45 + %9127 = bitcast <2 x bfloat> %8863 to i32, !dbg !45 + %9128 = bitcast <2 x bfloat> %8864 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9125, i32 %9126, i32 %9127, i32 %9128, ptr addrspace(1) %548, i1 %1118) #5, !dbg !45 + %9129 = bitcast <2 x bfloat> %8865 to i32, !dbg !45 + %9130 = bitcast <2 x bfloat> %8866 to i32, !dbg !45 + %9131 = bitcast <2 x bfloat> %8867 to i32, !dbg !45 + %9132 = bitcast <2 x bfloat> %8868 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9129, i32 %9130, i32 %9131, i32 %9132, ptr addrspace(1) %549, i1 %1136) #5, !dbg !45 + %9133 = bitcast <2 x bfloat> %8869 to i32, !dbg !45 + %9134 = bitcast <2 x bfloat> %8870 to i32, !dbg !45 + %9135 = bitcast <2 x bfloat> %8871 to i32, !dbg !45 + %9136 = bitcast <2 x bfloat> %8872 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9133, i32 %9134, i32 %9135, i32 %9136, ptr addrspace(1) %550, i1 %1154) #5, !dbg !45 + %9137 = bitcast <2 x bfloat> %8873 to i32, !dbg !45 + %9138 = bitcast <2 x bfloat> %8874 to i32, !dbg !45 + %9139 = bitcast <2 x bfloat> %8875 to i32, !dbg !45 + %9140 = bitcast <2 x bfloat> %8876 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9137, i32 %9138, i32 %9139, i32 %9140, ptr addrspace(1) %551, i1 %1172) #5, !dbg !45 + %9141 = bitcast <2 x bfloat> %8877 to i32, !dbg !45 + %9142 = bitcast <2 x bfloat> %8878 to i32, !dbg !45 + %9143 = bitcast <2 x bfloat> %8879 to i32, !dbg !45 + %9144 = bitcast <2 x bfloat> %8880 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9141, i32 %9142, i32 %9143, i32 %9144, ptr addrspace(1) %552, i1 %1190) #5, !dbg !45 + %9145 = bitcast <2 x bfloat> %8881 to i32, !dbg !45 + %9146 = bitcast <2 x bfloat> %8882 to i32, !dbg !45 + %9147 = bitcast <2 x bfloat> %8883 to i32, !dbg !45 + %9148 = bitcast <2 x bfloat> %8884 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9145, i32 %9146, i32 %9147, i32 %9148, ptr addrspace(1) %553, i1 %1208) #5, !dbg !45 + %9149 = bitcast <2 x bfloat> %8885 to i32, !dbg !45 + %9150 = bitcast <2 x bfloat> %8886 to i32, !dbg !45 + %9151 = bitcast <2 x bfloat> %8887 to i32, !dbg !45 + %9152 = bitcast <2 x bfloat> %8888 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9149, i32 %9150, i32 %9151, i32 %9152, ptr addrspace(1) %554, i1 %1226) #5, !dbg !45 + %9153 = bitcast <2 x bfloat> %8889 to i32, !dbg !45 + %9154 = bitcast <2 x bfloat> %8890 to i32, !dbg !45 + %9155 = bitcast <2 x bfloat> %8891 to i32, !dbg !45 + %9156 = bitcast <2 x bfloat> %8892 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9153, i32 %9154, i32 %9155, i32 %9156, ptr addrspace(1) %555, i1 %1244) #5, !dbg !45 + %9157 = bitcast <2 x bfloat> %8893 to i32, !dbg !45 + %9158 = bitcast <2 x bfloat> %8894 to i32, !dbg !45 + %9159 = bitcast <2 x bfloat> %8895 to i32, !dbg !45 + %9160 = bitcast <2 x bfloat> %8896 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9157, i32 %9158, i32 %9159, i32 %9160, ptr addrspace(1) %556, i1 %1262) #5, !dbg !45 + %9161 = bitcast <2 x bfloat> %8897 to i32, !dbg !45 + %9162 = bitcast <2 x bfloat> %8898 to i32, !dbg !45 + %9163 = bitcast <2 x bfloat> %8899 to i32, !dbg !45 + %9164 = bitcast <2 x bfloat> %8900 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9161, i32 %9162, i32 %9163, i32 %9164, ptr addrspace(1) %557, i1 %1280) #5, !dbg !45 + %9165 = bitcast <2 x bfloat> %8901 to i32, !dbg !45 + %9166 = bitcast <2 x bfloat> %8902 to i32, !dbg !45 + %9167 = bitcast <2 x bfloat> %8903 to i32, !dbg !45 + %9168 = bitcast <2 x bfloat> %8904 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9165, i32 %9166, i32 %9167, i32 %9168, ptr addrspace(1) %558, i1 %1298) #5, !dbg !45 + %9169 = bitcast <2 x bfloat> %8905 to i32, !dbg !45 + %9170 = bitcast <2 x bfloat> %8906 to i32, !dbg !45 + %9171 = bitcast <2 x bfloat> %8907 to i32, !dbg !45 + %9172 = bitcast <2 x bfloat> %8908 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9169, i32 %9170, i32 %9171, i32 %9172, ptr addrspace(1) %559, i1 %1316) #5, !dbg !45 + %9173 = bitcast <2 x bfloat> %8909 to i32, !dbg !45 + %9174 = bitcast <2 x bfloat> %8910 to i32, !dbg !45 + %9175 = bitcast <2 x bfloat> %8911 to i32, !dbg !45 + %9176 = bitcast <2 x bfloat> %8912 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9173, i32 %9174, i32 %9175, i32 %9176, ptr addrspace(1) %560, i1 %1334) #5, !dbg !45 + %9177 = bitcast <2 x bfloat> %8913 to i32, !dbg !45 + %9178 = bitcast <2 x bfloat> %8914 to i32, !dbg !45 + %9179 = bitcast <2 x bfloat> %8915 to i32, !dbg !45 + %9180 = bitcast <2 x bfloat> %8916 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9177, i32 %9178, i32 %9179, i32 %9180, ptr addrspace(1) %561, i1 %1352) #5, !dbg !45 + %9181 = bitcast <2 x bfloat> %8917 to i32, !dbg !45 + %9182 = bitcast <2 x bfloat> %8918 to i32, !dbg !45 + %9183 = bitcast <2 x bfloat> %8919 to i32, !dbg !45 + %9184 = bitcast <2 x bfloat> %8920 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9181, i32 %9182, i32 %9183, i32 %9184, ptr addrspace(1) %562, i1 %1370) #5, !dbg !45 + %9185 = bitcast <2 x bfloat> %8921 to i32, !dbg !45 + %9186 = bitcast <2 x bfloat> %8922 to i32, !dbg !45 + %9187 = bitcast <2 x bfloat> %8923 to i32, !dbg !45 + %9188 = bitcast <2 x bfloat> %8924 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9185, i32 %9186, i32 %9187, i32 %9188, ptr addrspace(1) %563, i1 %1388) #5, !dbg !45 + %9189 = bitcast <2 x bfloat> %8925 to i32, !dbg !45 + %9190 = bitcast <2 x bfloat> %8926 to i32, !dbg !45 + %9191 = bitcast <2 x bfloat> %8927 to i32, !dbg !45 + %9192 = bitcast <2 x bfloat> %8928 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9189, i32 %9190, i32 %9191, i32 %9192, ptr addrspace(1) %564, i1 %1406) #5, !dbg !45 + %9193 = bitcast <2 x bfloat> %8929 to i32, !dbg !45 + %9194 = bitcast <2 x bfloat> %8930 to i32, !dbg !45 + %9195 = bitcast <2 x bfloat> %8931 to i32, !dbg !45 + %9196 = bitcast <2 x bfloat> %8932 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9193, i32 %9194, i32 %9195, i32 %9196, ptr addrspace(1) %565, i1 %1424) #5, !dbg !45 + %9197 = bitcast <2 x bfloat> %8933 to i32, !dbg !45 + %9198 = bitcast <2 x bfloat> %8934 to i32, !dbg !45 + %9199 = bitcast <2 x bfloat> %8935 to i32, !dbg !45 + %9200 = bitcast <2 x bfloat> %8936 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9197, i32 %9198, i32 %9199, i32 %9200, ptr addrspace(1) %566, i1 %1442) #5, !dbg !45 + %9201 = bitcast <2 x bfloat> %8937 to i32, !dbg !45 + %9202 = bitcast <2 x bfloat> %8938 to i32, !dbg !45 + %9203 = bitcast <2 x bfloat> %8939 to i32, !dbg !45 + %9204 = bitcast <2 x bfloat> %8940 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9201, i32 %9202, i32 %9203, i32 %9204, ptr addrspace(1) %567, i1 %1460) #5, !dbg !45 + %9205 = bitcast <2 x bfloat> %8941 to i32, !dbg !45 + %9206 = bitcast <2 x bfloat> %8942 to i32, !dbg !45 + %9207 = bitcast <2 x bfloat> %8943 to i32, !dbg !45 + %9208 = bitcast <2 x bfloat> %8944 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9205, i32 %9206, i32 %9207, i32 %9208, ptr addrspace(1) %568, i1 %1478) #5, !dbg !45 + %9209 = bitcast <2 x bfloat> %8945 to i32, !dbg !45 + %9210 = bitcast <2 x bfloat> %8946 to i32, !dbg !45 + %9211 = bitcast <2 x bfloat> %8947 to i32, !dbg !45 + %9212 = bitcast <2 x bfloat> %8948 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9209, i32 %9210, i32 %9211, i32 %9212, ptr addrspace(1) %569, i1 %1496) #5, !dbg !45 + %9213 = bitcast <2 x bfloat> %8949 to i32, !dbg !45 + %9214 = bitcast <2 x bfloat> %8950 to i32, !dbg !45 + %9215 = bitcast <2 x bfloat> %8951 to i32, !dbg !45 + %9216 = bitcast <2 x bfloat> %8952 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9213, i32 %9214, i32 %9215, i32 %9216, ptr addrspace(1) %570, i1 %1514) #5, !dbg !45 + %9217 = bitcast <2 x bfloat> %8953 to i32, !dbg !45 + %9218 = bitcast <2 x bfloat> %8954 to i32, !dbg !45 + %9219 = bitcast <2 x bfloat> %8955 to i32, !dbg !45 + %9220 = bitcast <2 x bfloat> %8956 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9217, i32 %9218, i32 %9219, i32 %9220, ptr addrspace(1) %571, i1 %1532) #5, !dbg !45 + %9221 = bitcast <2 x bfloat> %8957 to i32, !dbg !45 + %9222 = bitcast <2 x bfloat> %8958 to i32, !dbg !45 + %9223 = bitcast <2 x bfloat> %8959 to i32, !dbg !45 + %9224 = bitcast <2 x bfloat> %8960 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9221, i32 %9222, i32 %9223, i32 %9224, ptr addrspace(1) %572, i1 %1550) #5, !dbg !45 + %9225 = bitcast <2 x bfloat> %8961 to i32, !dbg !45 + %9226 = bitcast <2 x bfloat> %8962 to i32, !dbg !45 + %9227 = bitcast <2 x bfloat> %8963 to i32, !dbg !45 + %9228 = bitcast <2 x bfloat> %8964 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9225, i32 %9226, i32 %9227, i32 %9228, ptr addrspace(1) %573, i1 %1568) #5, !dbg !45 + %9229 = bitcast <2 x bfloat> %8965 to i32, !dbg !45 + %9230 = bitcast <2 x bfloat> %8966 to i32, !dbg !45 + %9231 = bitcast <2 x bfloat> %8967 to i32, !dbg !45 + %9232 = bitcast <2 x bfloat> %8968 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9229, i32 %9230, i32 %9231, i32 %9232, ptr addrspace(1) %574, i1 %1586) #5, !dbg !45 + %9233 = bitcast <2 x bfloat> %8969 to i32, !dbg !45 + %9234 = bitcast <2 x bfloat> %8970 to i32, !dbg !45 + %9235 = bitcast <2 x bfloat> %8971 to i32, !dbg !45 + %9236 = bitcast <2 x bfloat> %8972 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9233, i32 %9234, i32 %9235, i32 %9236, ptr addrspace(1) %575, i1 %1604) #5, !dbg !45 + %9237 = bitcast <2 x bfloat> %8973 to i32, !dbg !45 + %9238 = bitcast <2 x bfloat> %8974 to i32, !dbg !45 + %9239 = bitcast <2 x bfloat> %8975 to i32, !dbg !45 + %9240 = bitcast <2 x bfloat> %8976 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9237, i32 %9238, i32 %9239, i32 %9240, ptr addrspace(1) %576, i1 %1622) #5, !dbg !45 + %9241 = bitcast <2 x bfloat> %8977 to i32, !dbg !45 + %9242 = bitcast <2 x bfloat> %8978 to i32, !dbg !45 + %9243 = bitcast <2 x bfloat> %8979 to i32, !dbg !45 + %9244 = bitcast <2 x bfloat> %8980 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9241, i32 %9242, i32 %9243, i32 %9244, ptr addrspace(1) %577, i1 %1640) #5, !dbg !45 + %9245 = bitcast <2 x bfloat> %8981 to i32, !dbg !45 + %9246 = bitcast <2 x bfloat> %8982 to i32, !dbg !45 + %9247 = bitcast <2 x bfloat> %8983 to i32, !dbg !45 + %9248 = bitcast <2 x bfloat> %8984 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9245, i32 %9246, i32 %9247, i32 %9248, ptr addrspace(1) %578, i1 %1658) #5, !dbg !45 + %9249 = bitcast <2 x bfloat> %8985 to i32, !dbg !45 + %9250 = bitcast <2 x bfloat> %8986 to i32, !dbg !45 + %9251 = bitcast <2 x bfloat> %8987 to i32, !dbg !45 + %9252 = bitcast <2 x bfloat> %8988 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9249, i32 %9250, i32 %9251, i32 %9252, ptr addrspace(1) %579, i1 %1676) #5, !dbg !45 + %9253 = bitcast <2 x bfloat> %8989 to i32, !dbg !45 + %9254 = bitcast <2 x bfloat> %8990 to i32, !dbg !45 + %9255 = bitcast <2 x bfloat> %8991 to i32, !dbg !45 + %9256 = bitcast <2 x bfloat> %8992 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9253, i32 %9254, i32 %9255, i32 %9256, ptr addrspace(1) %580, i1 %1694) #5, !dbg !45 + %9257 = bitcast <2 x bfloat> %8993 to i32, !dbg !45 + %9258 = bitcast <2 x bfloat> %8994 to i32, !dbg !45 + %9259 = bitcast <2 x bfloat> %8995 to i32, !dbg !45 + %9260 = bitcast <2 x bfloat> %8996 to i32, !dbg !45 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %9257, i32 %9258, i32 %9259, i32 %9260, ptr addrspace(1) %581, i1 %1712) #5, !dbg !45 + ret void, !dbg !46 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(argmem: write) +declare void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x2.b16.p3(ptr addrspace(3) writeonly captures(none), i32, i32) #2 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.bar.warp.sync(i32) #3 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #4 + +attributes #0 = { nounwind "nvvm.reqntid"="32" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(argmem: write) } +attributes #3 = { convergent nocallback nounwind } +attributes #4 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #5 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = distinct !DISubprogram(name: "l2norm_bwd_kernel", linkageName: "l2norm_bwd_kernel", scope: !1, file: !1, line: 93, type: !5, scopeLine: 93, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!5 = !DISubroutineType(cc: DW_CC_normal, types: !6) +!6 = !{null, !7, !9, !7, !7, !10, !11, !7, !7} +!7 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !8, size: 64, dwarfAddressSpace: 1) +!8 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!9 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !10, size: 64, dwarfAddressSpace: 1) +!10 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!11 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!12 = !DILocation(line: 94, column: 24, scope: !4) +!13 = !DILocation(line: 94, column: 30, scope: !4) +!14 = !DILocation(line: 95, column: 16, scope: !4) +!15 = !DILocation(line: 95, column: 34, scope: !4) +!16 = !DILocation(line: 95, column: 21, scope: !4) +!17 = !DILocation(line: 97, column: 16, scope: !4) +!18 = !DILocation(line: 98, column: 30, scope: !4) +!19 = !DILocation(line: 99, column: 29, scope: !4) +!20 = !DILocation(line: 99, column: 14, scope: !4) +!21 = !DILocation(line: 99, column: 33, scope: !4) +!22 = !DILocation(line: 100, column: 20, scope: !4) +!23 = !DILocation(line: 101, column: 16, scope: !4) +!24 = !DILocation(line: 101, column: 35, scope: !4) +!25 = !DILocation(line: 102, column: 16, scope: !4) +!26 = !DILocation(line: 102, column: 35, scope: !4) +!27 = !DILocation(line: 104, column: 18, scope: !4) +!28 = !DILocation(line: 104, column: 47, scope: !4) +!29 = !DILocation(line: 105, column: 21, scope: !4) +!30 = !DILocation(line: 106, column: 19, scope: !4) +!31 = !DILocation(line: 106, column: 49, scope: !4) +!32 = !DILocation(line: 107, column: 25, scope: !4) +!33 = !DILocation(line: 107, column: 18, scope: !4) +!34 = !DILocation(line: 107, column: 50, scope: !4) +!35 = !DILocation(line: 263, column: 15, scope: !36, inlinedAt: !38) +!36 = distinct !DILexicalBlockFile(scope: !4, file: !37, discriminator: 0) +!37 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!38 = !DILocation(line: 293, column: 36, scope: !36, inlinedAt: !39) +!39 = !DILocation(line: 107, column: 55, scope: !40) +!40 = distinct !DILexicalBlockFile(scope: !4, file: !1, discriminator: 0) +!41 = !DILocation(line: 107, column: 69, scope: !4) +!42 = !DILocation(line: 107, column: 75, scope: !4) +!43 = !DILocation(line: 107, column: 36, scope: !4) +!44 = !DILocation(line: 108, column: 27, scope: !4) +!45 = !DILocation(line: 108, column: 19, scope: !4) +!46 = !DILocation(line: 108, column: 4, scope: !4) diff --git a/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.ptx b/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..4869c24642edb5f12d95ab35238acc951c476a96 --- /dev/null +++ b/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.ptx @@ -0,0 +1,10132 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_bwd_kernel // -- Begin function l2norm_bwd_kernel +.extern .shared .align 16 .b8 global_smem[]; + // @l2norm_bwd_kernel +.visible .entry l2norm_bwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_2, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_3, + .param .f32 l2norm_bwd_kernel_param_4, + .param .u32 l2norm_bwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_6, + .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_7 +) +.reqntid 32 +{ + .reg .pred %p<115>; + .reg .b16 %rs<1184>; + .reg .b32 %r<5048>; + .reg .b64 %rd<1419>; + .loc 1 93 0 // l2norm.py:93:0 +$L__func_begin0: + .loc 1 93 0 // l2norm.py:93:0 + +// %bb.0: + ld.param.b64 %rd197, [l2norm_bwd_kernel_param_0]; + ld.param.b64 %rd198, [l2norm_bwd_kernel_param_1]; +$L__tmp0: + .loc 1 94 24 // l2norm.py:94:24 + mov.u32 %r774, %ctaid.x; + ld.param.b64 %rd199, [l2norm_bwd_kernel_param_2]; + .loc 1 95 16 // l2norm.py:95:16 + mul.wide.u32 %rd200, %r774, 128; + ld.param.b64 %rd201, [l2norm_bwd_kernel_param_3]; + .loc 1 95 34 // l2norm.py:95:34 + mov.u32 %r775, %tid.x; + and.b32 %r776, %r775, 16; + ld.param.s32 %rd202, [l2norm_bwd_kernel_param_5]; + bfe.u32 %r777, %r775, 4, 1; + or.b32 %r778, %r777, 2; + or.b32 %r779, %r777, 4; + or.b32 %r780, %r777, 6; + shl.b32 %r781, %r775, 2; + .loc 1 95 21 // l2norm.py:95:21 + cvt.u64.u32 %rd203, %r777; + cvt.u64.u32 %rd204, %r778; + cvt.u64.u32 %rd205, %r779; + cvt.u64.u32 %rd206, %r780; + or.b64 %rd207, %rd200, %rd203; + or.b64 %rd208, %rd200, %rd204; + or.b64 %rd209, %rd200, %rd205; + or.b64 %rd210, %rd200, %rd206; + .loc 1 97 16 // l2norm.py:97:16 + setp.lt.s64 %p1, %rd207, %rd202; + setp.lt.s64 %p2, %rd208, %rd202; + setp.lt.s64 %p3, %rd209, %rd202; + setp.lt.s64 %p4, %rd210, %rd202; + .loc 1 95 34 // l2norm.py:95:34 + and.b32 %r782, %r781, 124; + or.b32 %r783, %r782, 1; + .loc 1 95 21 // l2norm.py:95:21 + or.b32 %r784, %r777, 8; + or.b32 %r785, %r777, 10; + or.b32 %r786, %r777, 12; + or.b32 %r787, %r777, 14; + or.b32 %r788, %r777, 16; + or.b32 %r789, %r777, 18; + or.b32 %r790, %r777, 20; + or.b32 %r791, %r777, 22; + or.b32 %r792, %r777, 24; + or.b32 %r793, %r777, 26; + or.b32 %r794, %r777, 28; + or.b32 %r795, %r777, 30; + or.b32 %r796, %r777, 32; + or.b32 %r797, %r777, 34; + or.b32 %r798, %r777, 36; + or.b32 %r799, %r777, 38; + or.b32 %r800, %r777, 68; + or.b32 %r801, %r777, 70; + or.b32 %r802, %r777, 64; + or.b32 %r803, %r777, 66; + or.b32 %r804, %r777, 60; + or.b32 %r805, %r777, 62; + or.b32 %r806, %r777, 56; + or.b32 %r807, %r777, 58; + or.b32 %r808, %r777, 52; + or.b32 %r809, %r777, 54; + or.b32 %r810, %r777, 48; + or.b32 %r811, %r777, 50; + or.b32 %r812, %r777, 44; + or.b32 %r813, %r777, 46; + or.b32 %r814, %r777, 40; + or.b32 %r815, %r777, 42; + or.b32 %r816, %r777, 100; + or.b32 %r817, %r777, 102; + or.b32 %r818, %r777, 96; + or.b32 %r819, %r777, 98; + or.b32 %r820, %r777, 92; + or.b32 %r821, %r777, 94; + or.b32 %r822, %r777, 88; + or.b32 %r823, %r777, 90; + or.b32 %r824, %r777, 84; + or.b32 %r825, %r777, 86; + or.b32 %r826, %r777, 80; + or.b32 %r827, %r777, 82; + or.b32 %r828, %r777, 76; + or.b32 %r829, %r777, 78; + or.b32 %r830, %r777, 72; + or.b32 %r831, %r777, 74; + or.b32 %r832, %r777, 116; + or.b32 %r833, %r777, 118; + or.b32 %r834, %r777, 112; + or.b32 %r835, %r777, 114; + or.b32 %r836, %r777, 108; + or.b32 %r837, %r777, 110; + or.b32 %r838, %r777, 104; + or.b32 %r839, %r777, 106; + or.b32 %r840, %r777, 124; + or.b32 %r841, %r777, 126; + or.b32 %r842, %r777, 120; + or.b32 %r843, %r777, 122; + or.b32 %r844, %r782, 2; + or.b32 %r845, %r782, 3; + cvt.u64.u32 %rd211, %r831; + cvt.u64.u32 %rd212, %r830; + cvt.u64.u32 %rd213, %r829; + cvt.u64.u32 %rd214, %r828; + cvt.u64.u32 %rd215, %r827; + cvt.u64.u32 %rd216, %r826; + cvt.u64.u32 %rd217, %r825; + cvt.u64.u32 %rd218, %r824; + cvt.u64.u32 %rd219, %r823; + cvt.u64.u32 %rd220, %r822; + cvt.u64.u32 %rd221, %r821; + cvt.u64.u32 %rd222, %r820; + cvt.u64.u32 %rd223, %r819; + cvt.u64.u32 %rd224, %r818; + cvt.u64.u32 %rd225, %r817; + cvt.u64.u32 %rd226, %r816; + cvt.u64.u32 %rd227, %r839; + cvt.u64.u32 %rd228, %r838; + cvt.u64.u32 %rd229, %r837; + cvt.u64.u32 %rd230, %r836; + cvt.u64.u32 %rd231, %r835; + cvt.u64.u32 %rd232, %r834; + cvt.u64.u32 %rd233, %r833; + cvt.u64.u32 %rd234, %r832; + cvt.u64.u32 %rd235, %r843; + cvt.u64.u32 %rd236, %r842; + cvt.u64.u32 %rd237, %r841; + cvt.u64.u32 %rd238, %r840; + cvt.u64.u32 %rd239, %r782; + cvt.u64.u32 %rd240, %r783; + cvt.u64.u32 %rd241, %r845; + cvt.u64.u32 %rd242, %r844; + cvt.u64.u32 %rd243, %r815; + cvt.u64.u32 %rd244, %r814; + cvt.u64.u32 %rd245, %r813; + cvt.u64.u32 %rd246, %r812; + cvt.u64.u32 %rd247, %r811; + cvt.u64.u32 %rd248, %r810; + cvt.u64.u32 %rd249, %r809; + cvt.u64.u32 %rd250, %r808; + cvt.u64.u32 %rd251, %r807; + cvt.u64.u32 %rd252, %r806; + cvt.u64.u32 %rd253, %r805; + cvt.u64.u32 %rd254, %r804; + cvt.u64.u32 %rd255, %r803; + cvt.u64.u32 %rd256, %r802; + cvt.u64.u32 %rd257, %r801; + cvt.u64.u32 %rd258, %r800; + cvt.u64.u32 %rd259, %r799; + cvt.u64.u32 %rd260, %r798; + cvt.u64.u32 %rd261, %r797; + cvt.u64.u32 %rd262, %r796; + cvt.u64.u32 %rd263, %r795; + cvt.u64.u32 %rd264, %r794; + cvt.u64.u32 %rd265, %r793; + cvt.u64.u32 %rd266, %r792; + cvt.u64.u32 %rd267, %r791; + cvt.u64.u32 %rd268, %r790; + cvt.u64.u32 %rd269, %r789; + cvt.u64.u32 %rd270, %r788; + cvt.u64.u32 %rd271, %r787; + cvt.u64.u32 %rd272, %r786; + cvt.u64.u32 %rd273, %r785; + cvt.u64.u32 %rd274, %r784; + or.b64 %rd275, %rd200, %rd274; + or.b64 %rd276, %rd200, %rd273; + or.b64 %rd277, %rd200, %rd272; + or.b64 %rd278, %rd200, %rd271; + or.b64 %rd279, %rd200, %rd270; + or.b64 %rd280, %rd200, %rd269; + or.b64 %rd281, %rd200, %rd268; + or.b64 %rd282, %rd200, %rd267; + or.b64 %rd283, %rd200, %rd266; + or.b64 %rd284, %rd200, %rd265; + or.b64 %rd285, %rd200, %rd264; + or.b64 %rd286, %rd200, %rd263; + or.b64 %rd287, %rd200, %rd262; + or.b64 %rd288, %rd200, %rd261; + or.b64 %rd289, %rd200, %rd260; + or.b64 %rd290, %rd200, %rd259; + or.b64 %rd291, %rd200, %rd258; + or.b64 %rd292, %rd200, %rd257; + or.b64 %rd293, %rd200, %rd256; + or.b64 %rd294, %rd200, %rd255; + or.b64 %rd295, %rd200, %rd254; + or.b64 %rd296, %rd200, %rd253; + or.b64 %rd297, %rd200, %rd252; + or.b64 %rd298, %rd200, %rd251; + or.b64 %rd299, %rd200, %rd250; + or.b64 %rd300, %rd200, %rd249; + or.b64 %rd301, %rd200, %rd248; + or.b64 %rd302, %rd200, %rd247; + or.b64 %rd303, %rd200, %rd246; + or.b64 %rd304, %rd200, %rd245; + or.b64 %rd305, %rd200, %rd244; + or.b64 %rd306, %rd200, %rd243; + or.b64 %rd307, %rd200, %rd242; + or.b64 %rd308, %rd200, %rd241; + or.b64 %rd309, %rd200, %rd240; + or.b64 %rd310, %rd200, %rd239; + or.b64 %rd311, %rd200, %rd238; + or.b64 %rd312, %rd200, %rd237; + or.b64 %rd313, %rd200, %rd236; + or.b64 %rd314, %rd200, %rd235; + or.b64 %rd315, %rd200, %rd234; + or.b64 %rd316, %rd200, %rd233; + or.b64 %rd317, %rd200, %rd232; + or.b64 %rd318, %rd200, %rd231; + or.b64 %rd319, %rd200, %rd230; + or.b64 %rd320, %rd200, %rd229; + or.b64 %rd321, %rd200, %rd228; + or.b64 %rd322, %rd200, %rd227; + or.b64 %rd323, %rd200, %rd226; + or.b64 %rd324, %rd200, %rd225; + or.b64 %rd325, %rd200, %rd224; + or.b64 %rd326, %rd200, %rd223; + or.b64 %rd327, %rd200, %rd222; + or.b64 %rd328, %rd200, %rd221; + or.b64 %rd329, %rd200, %rd220; + or.b64 %rd330, %rd200, %rd219; + or.b64 %rd331, %rd200, %rd218; + or.b64 %rd332, %rd200, %rd217; + or.b64 %rd333, %rd200, %rd216; + or.b64 %rd334, %rd200, %rd215; + or.b64 %rd335, %rd200, %rd214; + or.b64 %rd336, %rd200, %rd213; + or.b64 %rd337, %rd200, %rd212; + or.b64 %rd338, %rd200, %rd211; + .loc 1 97 16 // l2norm.py:97:16 + setp.lt.s64 %p69, %rd338, %rd202; + selp.b16 %rs1, 1, 0, %p69; + shl.b16 %rs2, %rs1, 2; + setp.lt.s64 %p70, %rd337, %rd202; + selp.b16 %rs3, -1, 0, %p70; + shl.b16 %rs4, %rs3, 3; + or.b16 %rs5, %rs4, %rs2; + setp.lt.s64 %p71, %rd336, %rd202; + selp.b16 %rs6, 1, 0, %p71; + setp.lt.s64 %p72, %rd335, %rd202; + selp.b16 %rs7, -1, 0, %p72; + shl.b16 %rs8, %rs7, 1; + or.b16 %rs9, %rs6, %rs8; + and.b16 %rs10, %rs9, 3; + or.b16 %rs11, %rs10, %rs5; + shl.b16 %rs12, %rs11, 12; + setp.lt.s64 %p73, %rd334, %rd202; + selp.b16 %rs13, 1, 0, %p73; + shl.b16 %rs14, %rs13, 2; + setp.lt.s64 %p74, %rd333, %rd202; + selp.b16 %rs15, -1, 0, %p74; + shl.b16 %rs16, %rs15, 3; + or.b16 %rs17, %rs16, %rs14; + setp.lt.s64 %p75, %rd332, %rd202; + selp.b16 %rs18, 1, 0, %p75; + setp.lt.s64 %p76, %rd331, %rd202; + selp.b16 %rs19, -1, 0, %p76; + shl.b16 %rs20, %rs19, 1; + or.b16 %rs21, %rs18, %rs20; + and.b16 %rs22, %rs21, 3; + or.b16 %rs23, %rs22, %rs17; + and.b16 %rs24, %rs23, 15; + shl.b16 %rs25, %rs24, 8; + or.b16 %rs26, %rs12, %rs25; + setp.lt.s64 %p77, %rd330, %rd202; + selp.b16 %rs27, 1, 0, %p77; + shl.b16 %rs28, %rs27, 2; + setp.lt.s64 %p78, %rd329, %rd202; + selp.b16 %rs29, -1, 0, %p78; + shl.b16 %rs30, %rs29, 3; + or.b16 %rs31, %rs30, %rs28; + setp.lt.s64 %p79, %rd328, %rd202; + selp.b16 %rs32, 1, 0, %p79; + setp.lt.s64 %p80, %rd327, %rd202; + selp.b16 %rs33, -1, 0, %p80; + shl.b16 %rs34, %rs33, 1; + or.b16 %rs35, %rs32, %rs34; + and.b16 %rs36, %rs35, 3; + or.b16 %rs37, %rs36, %rs31; + shl.b16 %rs38, %rs37, 4; + setp.lt.s64 %p81, %rd326, %rd202; + selp.b16 %rs39, 1, 0, %p81; + shl.b16 %rs40, %rs39, 2; + setp.lt.s64 %p82, %rd325, %rd202; + selp.b16 %rs41, -1, 0, %p82; + shl.b16 %rs42, %rs41, 3; + or.b16 %rs43, %rs42, %rs40; + setp.lt.s64 %p83, %rd324, %rd202; + selp.b16 %rs44, 1, 0, %p83; + setp.lt.s64 %p84, %rd323, %rd202; + selp.b16 %rs45, -1, 0, %p84; + shl.b16 %rs46, %rs45, 1; + or.b16 %rs47, %rs44, %rs46; + and.b16 %rs48, %rs47, 3; + or.b16 %rs49, %rs48, %rs43; + and.b16 %rs50, %rs49, 15; + or.b16 %rs51, %rs50, %rs38; + and.b16 %rs52, %rs51, 255; + or.b16 %rs53, %rs52, %rs26; + cvt.u32.u16 %r846, %rs53; + shl.b32 %r847, %r846, 16; + setp.lt.s64 %p85, %rd322, %rd202; + selp.b16 %rs54, 1, 0, %p85; + shl.b16 %rs55, %rs54, 2; + setp.lt.s64 %p86, %rd321, %rd202; + selp.b16 %rs56, -1, 0, %p86; + shl.b16 %rs57, %rs56, 3; + or.b16 %rs58, %rs57, %rs55; + setp.lt.s64 %p87, %rd320, %rd202; + selp.b16 %rs59, 1, 0, %p87; + setp.lt.s64 %p88, %rd319, %rd202; + selp.b16 %rs60, -1, 0, %p88; + shl.b16 %rs61, %rs60, 1; + or.b16 %rs62, %rs59, %rs61; + and.b16 %rs63, %rs62, 3; + or.b16 %rs64, %rs63, %rs58; + shl.b16 %rs65, %rs64, 12; + setp.lt.s64 %p89, %rd318, %rd202; + selp.b16 %rs66, 1, 0, %p89; + shl.b16 %rs67, %rs66, 2; + setp.lt.s64 %p90, %rd317, %rd202; + selp.b16 %rs68, -1, 0, %p90; + shl.b16 %rs69, %rs68, 3; + or.b16 %rs70, %rs69, %rs67; + setp.lt.s64 %p91, %rd316, %rd202; + selp.b16 %rs71, 1, 0, %p91; + setp.lt.s64 %p92, %rd315, %rd202; + selp.b16 %rs72, -1, 0, %p92; + shl.b16 %rs73, %rs72, 1; + or.b16 %rs74, %rs71, %rs73; + and.b16 %rs75, %rs74, 3; + or.b16 %rs76, %rs75, %rs70; + and.b16 %rs77, %rs76, 15; + shl.b16 %rs78, %rs77, 8; + or.b16 %rs79, %rs65, %rs78; + setp.lt.s64 %p93, %rd314, %rd202; + selp.b16 %rs80, 1, 0, %p93; + shl.b16 %rs81, %rs80, 2; + setp.lt.s64 %p94, %rd313, %rd202; + selp.b16 %rs82, -1, 0, %p94; + shl.b16 %rs83, %rs82, 3; + or.b16 %rs84, %rs83, %rs81; + setp.lt.s64 %p95, %rd312, %rd202; + selp.b16 %rs85, 1, 0, %p95; + setp.lt.s64 %p96, %rd311, %rd202; + selp.b16 %rs86, -1, 0, %p96; + shl.b16 %rs87, %rs86, 1; + or.b16 %rs88, %rs85, %rs87; + and.b16 %rs89, %rs88, 3; + or.b16 %rs90, %rs89, %rs84; + shl.b16 %rs91, %rs90, 4; + setp.lt.s64 %p97, %rd310, %rd202; + selp.b16 %rs92, -1, 0, %p97; + shl.b16 %rs93, %rs92, 3; + setp.lt.s64 %p98, %rd309, %rd202; + selp.b16 %rs94, 1, 0, %p98; + shl.b16 %rs95, %rs94, 2; + or.b16 %rs96, %rs93, %rs95; + setp.lt.s64 %p38, %rd308, %rd202; + selp.b16 %rs97, 1, 0, %p38; + setp.lt.s64 %p99, %rd307, %rd202; + selp.b16 %rs98, -1, 0, %p99; + shl.b16 %rs99, %rs98, 1; + or.b16 %rs100, %rs97, %rs99; + and.b16 %rs101, %rs100, 3; + or.b16 %rs102, %rs101, %rs96; + and.b16 %rs103, %rs102, 15; + or.b16 %rs104, %rs103, %rs91; + and.b16 %rs105, %rs104, 255; + or.b16 %rs106, %rs105, %rs79; + cvt.u32.u16 %r848, %rs106; + or.b32 %r849, %r848, %r847; + cvt.u64.u32 %rd339, %r849; + setp.lt.s64 %p100, %rd306, %rd202; + selp.b16 %rs107, 1, 0, %p100; + shl.b16 %rs108, %rs107, 2; + setp.lt.s64 %p101, %rd305, %rd202; + selp.b16 %rs109, -1, 0, %p101; + shl.b16 %rs110, %rs109, 3; + or.b16 %rs111, %rs110, %rs108; + setp.lt.s64 %p102, %rd304, %rd202; + selp.b16 %rs112, 1, 0, %p102; + setp.lt.s64 %p103, %rd303, %rd202; + selp.b16 %rs113, -1, 0, %p103; + shl.b16 %rs114, %rs113, 1; + or.b16 %rs115, %rs112, %rs114; + and.b16 %rs116, %rs115, 3; + or.b16 %rs117, %rs116, %rs111; + shl.b16 %rs118, %rs117, 12; + setp.lt.s64 %p104, %rd302, %rd202; + selp.b16 %rs119, 1, 0, %p104; + shl.b16 %rs120, %rs119, 2; + setp.lt.s64 %p105, %rd301, %rd202; + selp.b16 %rs121, -1, 0, %p105; + shl.b16 %rs122, %rs121, 3; + or.b16 %rs123, %rs122, %rs120; + setp.lt.s64 %p106, %rd300, %rd202; + selp.b16 %rs124, 1, 0, %p106; + setp.lt.s64 %p107, %rd299, %rd202; + selp.b16 %rs125, -1, 0, %p107; + shl.b16 %rs126, %rs125, 1; + or.b16 %rs127, %rs124, %rs126; + and.b16 %rs128, %rs127, 3; + or.b16 %rs129, %rs128, %rs123; + and.b16 %rs130, %rs129, 15; + shl.b16 %rs131, %rs130, 8; + or.b16 %rs132, %rs118, %rs131; + setp.lt.s64 %p108, %rd298, %rd202; + selp.b16 %rs133, 1, 0, %p108; + shl.b16 %rs134, %rs133, 2; + setp.lt.s64 %p109, %rd297, %rd202; + selp.b16 %rs135, -1, 0, %p109; + shl.b16 %rs136, %rs135, 3; + or.b16 %rs137, %rs136, %rs134; + setp.lt.s64 %p110, %rd296, %rd202; + selp.b16 %rs138, 1, 0, %p110; + setp.lt.s64 %p111, %rd295, %rd202; + selp.b16 %rs139, -1, 0, %p111; + shl.b16 %rs140, %rs139, 1; + or.b16 %rs141, %rs138, %rs140; + and.b16 %rs142, %rs141, 3; + or.b16 %rs143, %rs142, %rs137; + shl.b16 %rs144, %rs143, 4; + setp.lt.s64 %p112, %rd294, %rd202; + selp.b16 %rs145, 1, 0, %p112; + shl.b16 %rs146, %rs145, 2; + setp.lt.s64 %p113, %rd293, %rd202; + selp.b16 %rs147, -1, 0, %p113; + shl.b16 %rs148, %rs147, 3; + or.b16 %rs149, %rs148, %rs146; + setp.lt.s64 %p34, %rd292, %rd202; + selp.b16 %rs150, 1, 0, %p34; + setp.lt.s64 %p114, %rd291, %rd202; + selp.b16 %rs151, -1, 0, %p114; + shl.b16 %rs152, %rs151, 1; + or.b16 %rs153, %rs150, %rs152; + and.b16 %rs154, %rs153, 3; + or.b16 %rs155, %rs154, %rs149; + and.b16 %rs156, %rs155, 15; + or.b16 %rs157, %rs156, %rs144; + and.b16 %rs158, %rs157, 255; + or.b16 %rs159, %rs158, %rs132; + cvt.u32.u16 %r850, %rs159; + shr.u32 %r851, %r850, 1; + shr.u32 %r852, %r850, 2; + shr.u32 %r853, %r850, 3; + shr.u32 %r854, %r850, 4; + shr.u32 %r855, %r850, 5; + shr.u32 %r856, %r850, 6; + shr.u32 %r857, %r850, 7; + shr.u32 %r858, %r850, 8; + shr.u32 %r859, %r850, 9; + shr.u32 %r860, %r850, 10; + shr.u32 %r861, %r850, 11; + shr.u32 %r862, %r850, 12; + shr.u32 %r863, %r850, 13; + shr.u32 %r864, %r850, 14; + shr.u32 %r865, %r850, 15; + setp.lt.s64 %p40, %rd290, %rd202; + setp.lt.s64 %p18, %rd289, %rd202; + setp.lt.s64 %p17, %rd288, %rd202; + setp.lt.s64 %p16, %rd287, %rd202; + setp.lt.s64 %p15, %rd286, %rd202; + setp.lt.s64 %p14, %rd285, %rd202; + setp.lt.s64 %p13, %rd284, %rd202; + setp.lt.s64 %p12, %rd283, %rd202; + setp.lt.s64 %p39, %rd282, %rd202; + setp.lt.s64 %p11, %rd281, %rd202; + setp.lt.s64 %p10, %rd280, %rd202; + setp.lt.s64 %p9, %rd279, %rd202; + setp.lt.s64 %p8, %rd278, %rd202; + setp.lt.s64 %p7, %rd277, %rd202; + setp.lt.s64 %p6, %rd276, %rd202; + setp.lt.s64 %p5, %rd275, %rd202; + .loc 1 98 30 // l2norm.py:98:30 + and.b32 %r866, %r775, 15; + shl.b32 %r867, %r866, 3; + .loc 1 99 14 // l2norm.py:99:14 + shl.b64 %rd340, %rd207, 8; + add.s64 %rd341, %rd197, %rd340; + shl.b64 %rd342, %rd208, 8; + add.s64 %rd343, %rd197, %rd342; + shl.b64 %rd344, %rd209, 8; + add.s64 %rd345, %rd197, %rd344; + shl.b64 %rd346, %rd210, 8; + add.s64 %rd347, %rd197, %rd346; + shl.b64 %rd348, %rd275, 8; + add.s64 %rd349, %rd197, %rd348; + shl.b64 %rd350, %rd276, 8; + add.s64 %rd351, %rd197, %rd350; + shl.b64 %rd352, %rd277, 8; + add.s64 %rd353, %rd197, %rd352; + shl.b64 %rd354, %rd278, 8; + add.s64 %rd355, %rd197, %rd354; + shl.b64 %rd356, %rd279, 8; + add.s64 %rd357, %rd197, %rd356; + shl.b64 %rd358, %rd280, 8; + add.s64 %rd359, %rd197, %rd358; + shl.b64 %rd360, %rd281, 8; + add.s64 %rd361, %rd197, %rd360; + shl.b64 %rd362, %rd282, 8; + add.s64 %rd363, %rd197, %rd362; + shl.b64 %rd364, %rd283, 8; + add.s64 %rd365, %rd197, %rd364; + shl.b64 %rd366, %rd284, 8; + add.s64 %rd367, %rd197, %rd366; + shl.b64 %rd368, %rd285, 8; + add.s64 %rd369, %rd197, %rd368; + shl.b64 %rd370, %rd286, 8; + add.s64 %rd371, %rd197, %rd370; + shl.b64 %rd372, %rd287, 8; + add.s64 %rd373, %rd197, %rd372; + shl.b64 %rd374, %rd288, 8; + add.s64 %rd375, %rd197, %rd374; + shl.b64 %rd376, %rd289, 8; + add.s64 %rd377, %rd197, %rd376; + shl.b64 %rd378, %rd290, 8; + add.s64 %rd379, %rd197, %rd378; + shl.b64 %rd380, %rd305, 8; + add.s64 %rd381, %rd197, %rd380; + shl.b64 %rd382, %rd306, 8; + add.s64 %rd383, %rd197, %rd382; + shl.b64 %rd384, %rd303, 8; + add.s64 %rd385, %rd197, %rd384; + shl.b64 %rd386, %rd304, 8; + add.s64 %rd387, %rd197, %rd386; + shl.b64 %rd388, %rd301, 8; + add.s64 %rd389, %rd197, %rd388; + shl.b64 %rd390, %rd302, 8; + add.s64 %rd391, %rd197, %rd390; + shl.b64 %rd392, %rd299, 8; + add.s64 %rd393, %rd197, %rd392; + shl.b64 %rd394, %rd300, 8; + add.s64 %rd395, %rd197, %rd394; + shl.b64 %rd396, %rd297, 8; + add.s64 %rd397, %rd197, %rd396; + shl.b64 %rd398, %rd298, 8; + add.s64 %rd399, %rd197, %rd398; + shl.b64 %rd400, %rd295, 8; + add.s64 %rd401, %rd197, %rd400; + shl.b64 %rd402, %rd296, 8; + add.s64 %rd403, %rd197, %rd402; + shl.b64 %rd404, %rd293, 8; + add.s64 %rd405, %rd197, %rd404; + shl.b64 %rd406, %rd294, 8; + add.s64 %rd407, %rd197, %rd406; + shl.b64 %rd408, %rd291, 8; + add.s64 %rd409, %rd197, %rd408; + shl.b64 %rd410, %rd292, 8; + add.s64 %rd411, %rd197, %rd410; + shl.b64 %rd412, %rd337, 8; + add.s64 %rd413, %rd197, %rd412; + shl.b64 %rd414, %rd338, 8; + add.s64 %rd415, %rd197, %rd414; + shl.b64 %rd416, %rd335, 8; + add.s64 %rd417, %rd197, %rd416; + shl.b64 %rd418, %rd336, 8; + add.s64 %rd419, %rd197, %rd418; + shl.b64 %rd420, %rd333, 8; + add.s64 %rd421, %rd197, %rd420; + shl.b64 %rd422, %rd334, 8; + add.s64 %rd423, %rd197, %rd422; + shl.b64 %rd424, %rd331, 8; + add.s64 %rd425, %rd197, %rd424; + shl.b64 %rd426, %rd332, 8; + add.s64 %rd427, %rd197, %rd426; + shl.b64 %rd428, %rd329, 8; + add.s64 %rd429, %rd197, %rd428; + shl.b64 %rd430, %rd330, 8; + add.s64 %rd431, %rd197, %rd430; + shl.b64 %rd432, %rd327, 8; + add.s64 %rd433, %rd197, %rd432; + shl.b64 %rd434, %rd328, 8; + add.s64 %rd435, %rd197, %rd434; + shl.b64 %rd436, %rd325, 8; + add.s64 %rd437, %rd197, %rd436; + shl.b64 %rd438, %rd326, 8; + add.s64 %rd439, %rd197, %rd438; + shl.b64 %rd440, %rd323, 8; + add.s64 %rd441, %rd197, %rd440; + shl.b64 %rd442, %rd324, 8; + add.s64 %rd443, %rd197, %rd442; + shl.b64 %rd444, %rd321, 8; + add.s64 %rd445, %rd197, %rd444; + shl.b64 %rd446, %rd322, 8; + add.s64 %rd447, %rd197, %rd446; + shl.b64 %rd448, %rd319, 8; + add.s64 %rd449, %rd197, %rd448; + shl.b64 %rd450, %rd320, 8; + add.s64 %rd451, %rd197, %rd450; + shl.b64 %rd452, %rd317, 8; + add.s64 %rd453, %rd197, %rd452; + shl.b64 %rd454, %rd318, 8; + add.s64 %rd455, %rd197, %rd454; + shl.b64 %rd456, %rd315, 8; + add.s64 %rd457, %rd197, %rd456; + shl.b64 %rd458, %rd316, 8; + add.s64 %rd459, %rd197, %rd458; + shl.b64 %rd460, %rd313, 8; + add.s64 %rd461, %rd197, %rd460; + shl.b64 %rd462, %rd314, 8; + add.s64 %rd463, %rd197, %rd462; + shl.b64 %rd464, %rd311, 8; + add.s64 %rd465, %rd197, %rd464; + shl.b64 %rd466, %rd312, 8; + add.s64 %rd467, %rd197, %rd466; + .loc 1 99 33 // l2norm.py:99:33 + mul.wide.u32 %rd468, %r867, 2; + add.s64 %rd1, %rd341, %rd468; + add.s64 %rd2, %rd343, %rd468; + add.s64 %rd3, %rd345, %rd468; + add.s64 %rd4, %rd347, %rd468; + add.s64 %rd5, %rd349, %rd468; + add.s64 %rd6, %rd351, %rd468; + add.s64 %rd7, %rd353, %rd468; + add.s64 %rd8, %rd355, %rd468; + add.s64 %rd9, %rd357, %rd468; + add.s64 %rd10, %rd359, %rd468; + add.s64 %rd11, %rd361, %rd468; + add.s64 %rd12, %rd363, %rd468; + add.s64 %rd13, %rd365, %rd468; + add.s64 %rd14, %rd367, %rd468; + add.s64 %rd15, %rd369, %rd468; + add.s64 %rd16, %rd371, %rd468; + add.s64 %rd17, %rd373, %rd468; + add.s64 %rd18, %rd375, %rd468; + add.s64 %rd19, %rd377, %rd468; + add.s64 %rd20, %rd379, %rd468; + add.s64 %rd21, %rd381, %rd468; + add.s64 %rd22, %rd383, %rd468; + add.s64 %rd23, %rd385, %rd468; + add.s64 %rd24, %rd387, %rd468; + add.s64 %rd25, %rd389, %rd468; + add.s64 %rd26, %rd391, %rd468; + add.s64 %rd27, %rd393, %rd468; + add.s64 %rd28, %rd395, %rd468; + add.s64 %rd29, %rd397, %rd468; + add.s64 %rd30, %rd399, %rd468; + add.s64 %rd31, %rd401, %rd468; + add.s64 %rd32, %rd403, %rd468; + add.s64 %rd33, %rd405, %rd468; + add.s64 %rd34, %rd407, %rd468; + add.s64 %rd35, %rd409, %rd468; + add.s64 %rd36, %rd411, %rd468; + add.s64 %rd37, %rd413, %rd468; + add.s64 %rd38, %rd415, %rd468; + add.s64 %rd39, %rd417, %rd468; + add.s64 %rd40, %rd419, %rd468; + add.s64 %rd41, %rd421, %rd468; + add.s64 %rd42, %rd423, %rd468; + add.s64 %rd43, %rd425, %rd468; + add.s64 %rd44, %rd427, %rd468; + add.s64 %rd45, %rd429, %rd468; + add.s64 %rd46, %rd431, %rd468; + add.s64 %rd47, %rd433, %rd468; + add.s64 %rd48, %rd435, %rd468; + add.s64 %rd49, %rd437, %rd468; + add.s64 %rd50, %rd439, %rd468; + add.s64 %rd51, %rd441, %rd468; + add.s64 %rd52, %rd443, %rd468; + add.s64 %rd53, %rd445, %rd468; + add.s64 %rd54, %rd447, %rd468; + add.s64 %rd55, %rd449, %rd468; + add.s64 %rd56, %rd451, %rd468; + add.s64 %rd57, %rd453, %rd468; + add.s64 %rd58, %rd455, %rd468; + add.s64 %rd59, %rd457, %rd468; + add.s64 %rd60, %rd459, %rd468; + add.s64 %rd61, %rd461, %rd468; + add.s64 %rd62, %rd463, %rd468; + add.s64 %rd63, %rd465, %rd468; + add.s64 %rd64, %rd467, %rd468; + .loc 1 100 20 // l2norm.py:100:20 + shl.b64 %rd469, %rd310, 2; + add.s64 %rd65, %rd198, %rd469; + shl.b64 %rd470, %rd309, 2; + add.s64 %rd66, %rd198, %rd470; + shl.b64 %rd471, %rd307, 2; + add.s64 %rd67, %rd198, %rd471; + shl.b64 %rd472, %rd308, 2; + add.s64 %rd68, %rd198, %rd472; + .loc 1 101 16 // l2norm.py:101:16 + add.s64 %rd473, %rd199, %rd340; + add.s64 %rd474, %rd199, %rd342; + add.s64 %rd475, %rd199, %rd344; + add.s64 %rd476, %rd199, %rd346; + add.s64 %rd477, %rd199, %rd348; + add.s64 %rd478, %rd199, %rd350; + add.s64 %rd479, %rd199, %rd352; + add.s64 %rd480, %rd199, %rd354; + add.s64 %rd481, %rd199, %rd356; + add.s64 %rd482, %rd199, %rd358; + add.s64 %rd483, %rd199, %rd360; + add.s64 %rd484, %rd199, %rd362; + add.s64 %rd485, %rd199, %rd364; + add.s64 %rd486, %rd199, %rd366; + add.s64 %rd487, %rd199, %rd368; + add.s64 %rd488, %rd199, %rd370; + add.s64 %rd489, %rd199, %rd372; + add.s64 %rd490, %rd199, %rd374; + add.s64 %rd491, %rd199, %rd376; + add.s64 %rd492, %rd199, %rd378; + add.s64 %rd493, %rd199, %rd380; + add.s64 %rd494, %rd199, %rd382; + add.s64 %rd495, %rd199, %rd384; + add.s64 %rd496, %rd199, %rd386; + add.s64 %rd497, %rd199, %rd388; + add.s64 %rd498, %rd199, %rd390; + add.s64 %rd499, %rd199, %rd392; + add.s64 %rd500, %rd199, %rd394; + add.s64 %rd501, %rd199, %rd396; + add.s64 %rd502, %rd199, %rd398; + add.s64 %rd503, %rd199, %rd400; + add.s64 %rd504, %rd199, %rd402; + add.s64 %rd505, %rd199, %rd404; + add.s64 %rd506, %rd199, %rd406; + add.s64 %rd507, %rd199, %rd408; + add.s64 %rd508, %rd199, %rd410; + add.s64 %rd509, %rd199, %rd412; + add.s64 %rd510, %rd199, %rd414; + add.s64 %rd511, %rd199, %rd416; + add.s64 %rd512, %rd199, %rd418; + add.s64 %rd513, %rd199, %rd420; + add.s64 %rd514, %rd199, %rd422; + add.s64 %rd515, %rd199, %rd424; + add.s64 %rd516, %rd199, %rd426; + add.s64 %rd517, %rd199, %rd428; + add.s64 %rd518, %rd199, %rd430; + add.s64 %rd519, %rd199, %rd432; + add.s64 %rd520, %rd199, %rd434; + add.s64 %rd521, %rd199, %rd436; + add.s64 %rd522, %rd199, %rd438; + add.s64 %rd523, %rd199, %rd440; + add.s64 %rd524, %rd199, %rd442; + add.s64 %rd525, %rd199, %rd444; + add.s64 %rd526, %rd199, %rd446; + add.s64 %rd527, %rd199, %rd448; + add.s64 %rd528, %rd199, %rd450; + add.s64 %rd529, %rd199, %rd452; + add.s64 %rd530, %rd199, %rd454; + add.s64 %rd531, %rd199, %rd456; + add.s64 %rd532, %rd199, %rd458; + add.s64 %rd533, %rd199, %rd460; + add.s64 %rd534, %rd199, %rd462; + add.s64 %rd535, %rd199, %rd464; + add.s64 %rd536, %rd199, %rd466; + .loc 1 101 35 // l2norm.py:101:35 + add.s64 %rd69, %rd473, %rd468; + add.s64 %rd70, %rd474, %rd468; + add.s64 %rd71, %rd475, %rd468; + add.s64 %rd72, %rd476, %rd468; + add.s64 %rd73, %rd477, %rd468; + add.s64 %rd74, %rd478, %rd468; + add.s64 %rd75, %rd479, %rd468; + add.s64 %rd76, %rd480, %rd468; + add.s64 %rd77, %rd481, %rd468; + add.s64 %rd78, %rd482, %rd468; + add.s64 %rd79, %rd483, %rd468; + add.s64 %rd80, %rd484, %rd468; + add.s64 %rd81, %rd485, %rd468; + add.s64 %rd82, %rd486, %rd468; + add.s64 %rd83, %rd487, %rd468; + add.s64 %rd84, %rd488, %rd468; + add.s64 %rd85, %rd489, %rd468; + add.s64 %rd86, %rd490, %rd468; + add.s64 %rd87, %rd491, %rd468; + add.s64 %rd88, %rd492, %rd468; + add.s64 %rd89, %rd493, %rd468; + add.s64 %rd90, %rd494, %rd468; + add.s64 %rd91, %rd495, %rd468; + add.s64 %rd92, %rd496, %rd468; + add.s64 %rd93, %rd497, %rd468; + add.s64 %rd94, %rd498, %rd468; + add.s64 %rd95, %rd499, %rd468; + add.s64 %rd96, %rd500, %rd468; + add.s64 %rd97, %rd501, %rd468; + add.s64 %rd98, %rd502, %rd468; + add.s64 %rd99, %rd503, %rd468; + add.s64 %rd100, %rd504, %rd468; + add.s64 %rd101, %rd505, %rd468; + add.s64 %rd102, %rd506, %rd468; + add.s64 %rd103, %rd507, %rd468; + add.s64 %rd104, %rd508, %rd468; + add.s64 %rd105, %rd509, %rd468; + add.s64 %rd106, %rd510, %rd468; + add.s64 %rd107, %rd511, %rd468; + add.s64 %rd108, %rd512, %rd468; + add.s64 %rd109, %rd513, %rd468; + add.s64 %rd110, %rd514, %rd468; + add.s64 %rd111, %rd515, %rd468; + add.s64 %rd112, %rd516, %rd468; + add.s64 %rd113, %rd517, %rd468; + add.s64 %rd114, %rd518, %rd468; + add.s64 %rd115, %rd519, %rd468; + add.s64 %rd116, %rd520, %rd468; + add.s64 %rd117, %rd521, %rd468; + add.s64 %rd118, %rd522, %rd468; + add.s64 %rd119, %rd523, %rd468; + add.s64 %rd120, %rd524, %rd468; + add.s64 %rd121, %rd525, %rd468; + add.s64 %rd122, %rd526, %rd468; + add.s64 %rd123, %rd527, %rd468; + add.s64 %rd124, %rd528, %rd468; + add.s64 %rd125, %rd529, %rd468; + add.s64 %rd126, %rd530, %rd468; + add.s64 %rd127, %rd531, %rd468; + add.s64 %rd128, %rd532, %rd468; + add.s64 %rd129, %rd533, %rd468; + add.s64 %rd130, %rd534, %rd468; + add.s64 %rd131, %rd535, %rd468; + add.s64 %rd132, %rd536, %rd468; + .loc 1 102 16 // l2norm.py:102:16 + add.s64 %rd537, %rd201, %rd340; + add.s64 %rd538, %rd201, %rd342; + add.s64 %rd539, %rd201, %rd344; + add.s64 %rd540, %rd201, %rd346; + add.s64 %rd541, %rd201, %rd348; + add.s64 %rd542, %rd201, %rd350; + add.s64 %rd543, %rd201, %rd352; + add.s64 %rd544, %rd201, %rd354; + add.s64 %rd545, %rd201, %rd356; + add.s64 %rd546, %rd201, %rd358; + add.s64 %rd547, %rd201, %rd360; + add.s64 %rd548, %rd201, %rd362; + add.s64 %rd549, %rd201, %rd364; + add.s64 %rd550, %rd201, %rd366; + add.s64 %rd551, %rd201, %rd368; + add.s64 %rd552, %rd201, %rd370; + add.s64 %rd553, %rd201, %rd372; + add.s64 %rd554, %rd201, %rd374; + add.s64 %rd555, %rd201, %rd376; + add.s64 %rd556, %rd201, %rd378; + add.s64 %rd557, %rd201, %rd380; + add.s64 %rd558, %rd201, %rd382; + add.s64 %rd559, %rd201, %rd384; + add.s64 %rd560, %rd201, %rd386; + add.s64 %rd561, %rd201, %rd388; + add.s64 %rd562, %rd201, %rd390; + add.s64 %rd563, %rd201, %rd392; + add.s64 %rd564, %rd201, %rd394; + add.s64 %rd565, %rd201, %rd396; + add.s64 %rd566, %rd201, %rd398; + add.s64 %rd567, %rd201, %rd400; + add.s64 %rd568, %rd201, %rd402; + add.s64 %rd569, %rd201, %rd404; + add.s64 %rd570, %rd201, %rd406; + add.s64 %rd571, %rd201, %rd408; + add.s64 %rd572, %rd201, %rd410; + add.s64 %rd573, %rd201, %rd412; + add.s64 %rd574, %rd201, %rd414; + add.s64 %rd575, %rd201, %rd416; + add.s64 %rd576, %rd201, %rd418; + add.s64 %rd577, %rd201, %rd420; + add.s64 %rd578, %rd201, %rd422; + add.s64 %rd579, %rd201, %rd424; + add.s64 %rd580, %rd201, %rd426; + add.s64 %rd581, %rd201, %rd428; + add.s64 %rd582, %rd201, %rd430; + add.s64 %rd583, %rd201, %rd432; + add.s64 %rd584, %rd201, %rd434; + add.s64 %rd585, %rd201, %rd436; + add.s64 %rd586, %rd201, %rd438; + add.s64 %rd587, %rd201, %rd440; + add.s64 %rd588, %rd201, %rd442; + add.s64 %rd589, %rd201, %rd444; + add.s64 %rd590, %rd201, %rd446; + add.s64 %rd591, %rd201, %rd448; + add.s64 %rd592, %rd201, %rd450; + add.s64 %rd593, %rd201, %rd452; + add.s64 %rd594, %rd201, %rd454; + add.s64 %rd595, %rd201, %rd456; + add.s64 %rd596, %rd201, %rd458; + add.s64 %rd597, %rd201, %rd460; + add.s64 %rd598, %rd201, %rd462; + add.s64 %rd599, %rd201, %rd464; + add.s64 %rd600, %rd201, %rd466; + .loc 1 102 35 // l2norm.py:102:35 + add.s64 %rd133, %rd537, %rd468; + add.s64 %rd134, %rd538, %rd468; + add.s64 %rd135, %rd539, %rd468; + add.s64 %rd136, %rd540, %rd468; + add.s64 %rd137, %rd541, %rd468; + add.s64 %rd138, %rd542, %rd468; + add.s64 %rd139, %rd543, %rd468; + add.s64 %rd140, %rd544, %rd468; + add.s64 %rd141, %rd545, %rd468; + add.s64 %rd142, %rd546, %rd468; + add.s64 %rd143, %rd547, %rd468; + add.s64 %rd144, %rd548, %rd468; + add.s64 %rd145, %rd549, %rd468; + add.s64 %rd146, %rd550, %rd468; + add.s64 %rd147, %rd551, %rd468; + add.s64 %rd148, %rd552, %rd468; + add.s64 %rd149, %rd553, %rd468; + add.s64 %rd150, %rd554, %rd468; + add.s64 %rd151, %rd555, %rd468; + add.s64 %rd152, %rd556, %rd468; + add.s64 %rd153, %rd557, %rd468; + add.s64 %rd154, %rd558, %rd468; + add.s64 %rd155, %rd559, %rd468; + add.s64 %rd156, %rd560, %rd468; + add.s64 %rd157, %rd561, %rd468; + add.s64 %rd158, %rd562, %rd468; + add.s64 %rd159, %rd563, %rd468; + add.s64 %rd160, %rd564, %rd468; + add.s64 %rd161, %rd565, %rd468; + add.s64 %rd162, %rd566, %rd468; + add.s64 %rd163, %rd567, %rd468; + add.s64 %rd164, %rd568, %rd468; + add.s64 %rd165, %rd569, %rd468; + add.s64 %rd166, %rd570, %rd468; + add.s64 %rd167, %rd571, %rd468; + add.s64 %rd168, %rd572, %rd468; + add.s64 %rd169, %rd573, %rd468; + add.s64 %rd170, %rd574, %rd468; + add.s64 %rd171, %rd575, %rd468; + add.s64 %rd172, %rd576, %rd468; + add.s64 %rd173, %rd577, %rd468; + add.s64 %rd174, %rd578, %rd468; + add.s64 %rd175, %rd579, %rd468; + add.s64 %rd176, %rd580, %rd468; + add.s64 %rd177, %rd581, %rd468; + add.s64 %rd178, %rd582, %rd468; + add.s64 %rd179, %rd583, %rd468; + add.s64 %rd180, %rd584, %rd468; + add.s64 %rd181, %rd585, %rd468; + add.s64 %rd182, %rd586, %rd468; + add.s64 %rd183, %rd587, %rd468; + add.s64 %rd184, %rd588, %rd468; + add.s64 %rd185, %rd589, %rd468; + add.s64 %rd186, %rd590, %rd468; + add.s64 %rd187, %rd591, %rd468; + add.s64 %rd188, %rd592, %rd468; + add.s64 %rd189, %rd593, %rd468; + add.s64 %rd190, %rd594, %rd468; + add.s64 %rd191, %rd595, %rd468; + add.s64 %rd192, %rd596, %rd468; + add.s64 %rd193, %rd597, %rd468; + add.s64 %rd194, %rd598, %rd468; + add.s64 %rd195, %rd599, %rd468; + add.s64 %rd196, %rd600, %rd468; + mov.b32 %r5, 0; + .loc 1 104 18 // l2norm.py:104:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs160, %rs161}, %r1; + mov.b32 {%rs162, %rs163}, %r2; + mov.b32 {%rs164, %rs165}, %r3; + mov.b32 {%rs166, %rs167}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs168, %rs169}, %r6; + mov.b32 {%rs170, %rs171}, %r7; + mov.b32 {%rs172, %rs173}, %r8; + mov.b32 {%rs174, %rs175}, %r9; + // begin inline asm + mov.u32 %r10, %r5; + mov.u32 %r11, %r5; + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + @%p3 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs176, %rs177}, %r10; + mov.b32 {%rs178, %rs179}, %r11; + mov.b32 {%rs180, %rs181}, %r12; + mov.b32 {%rs182, %rs183}, %r13; + // begin inline asm + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + @%p4 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ]; + // end inline asm + mov.b32 {%rs184, %rs185}, %r14; + mov.b32 {%rs186, %rs187}, %r15; + mov.b32 {%rs188, %rs189}, %r16; + mov.b32 {%rs190, %rs191}, %r17; + // begin inline asm + mov.u32 %r18, %r5; + mov.u32 %r19, %r5; + mov.u32 %r20, %r5; + mov.u32 %r21, %r5; + @%p5 ld.global.v4.b32 { %r18, %r19, %r20, %r21 }, [ %rd5 + 0 ]; + // end inline asm + mov.b32 {%rs192, %rs193}, %r18; + mov.b32 {%rs194, %rs195}, %r19; + mov.b32 {%rs196, %rs197}, %r20; + mov.b32 {%rs198, %rs199}, %r21; + // begin inline asm + mov.u32 %r22, %r5; + mov.u32 %r23, %r5; + mov.u32 %r24, %r5; + mov.u32 %r25, %r5; + @%p6 ld.global.v4.b32 { %r22, %r23, %r24, %r25 }, [ %rd6 + 0 ]; + // end inline asm + mov.b32 {%rs200, %rs201}, %r22; + mov.b32 {%rs202, %rs203}, %r23; + mov.b32 {%rs204, %rs205}, %r24; + mov.b32 {%rs206, %rs207}, %r25; + // begin inline asm + mov.u32 %r26, %r5; + mov.u32 %r27, %r5; + mov.u32 %r28, %r5; + mov.u32 %r29, %r5; + @%p7 ld.global.v4.b32 { %r26, %r27, %r28, %r29 }, [ %rd7 + 0 ]; + // end inline asm + mov.b32 {%rs208, %rs209}, %r26; + mov.b32 {%rs210, %rs211}, %r27; + mov.b32 {%rs212, %rs213}, %r28; + mov.b32 {%rs214, %rs215}, %r29; + // begin inline asm + mov.u32 %r30, %r5; + mov.u32 %r31, %r5; + mov.u32 %r32, %r5; + mov.u32 %r33, %r5; + @%p8 ld.global.v4.b32 { %r30, %r31, %r32, %r33 }, [ %rd8 + 0 ]; + // end inline asm + mov.b32 {%rs216, %rs217}, %r30; + mov.b32 {%rs218, %rs219}, %r31; + mov.b32 {%rs220, %rs221}, %r32; + mov.b32 {%rs222, %rs223}, %r33; + // begin inline asm + mov.u32 %r34, %r5; + mov.u32 %r35, %r5; + mov.u32 %r36, %r5; + mov.u32 %r37, %r5; + @%p9 ld.global.v4.b32 { %r34, %r35, %r36, %r37 }, [ %rd9 + 0 ]; + // end inline asm + mov.b32 {%rs224, %rs225}, %r34; + mov.b32 {%rs226, %rs227}, %r35; + mov.b32 {%rs228, %rs229}, %r36; + mov.b32 {%rs230, %rs231}, %r37; + // begin inline asm + mov.u32 %r38, %r5; + mov.u32 %r39, %r5; + mov.u32 %r40, %r5; + mov.u32 %r41, %r5; + @%p10 ld.global.v4.b32 { %r38, %r39, %r40, %r41 }, [ %rd10 + 0 ]; + // end inline asm + mov.b32 {%rs232, %rs233}, %r38; + mov.b32 {%rs234, %rs235}, %r39; + mov.b32 {%rs236, %rs237}, %r40; + mov.b32 {%rs238, %rs239}, %r41; + // begin inline asm + mov.u32 %r42, %r5; + mov.u32 %r43, %r5; + mov.u32 %r44, %r5; + mov.u32 %r45, %r5; + @%p11 ld.global.v4.b32 { %r42, %r43, %r44, %r45 }, [ %rd11 + 0 ]; + // end inline asm + mov.b32 {%rs240, %rs241}, %r42; + mov.b32 {%rs242, %rs243}, %r43; + mov.b32 {%rs244, %rs245}, %r44; + mov.b32 {%rs246, %rs247}, %r45; + // begin inline asm + mov.u32 %r46, %r5; + mov.u32 %r47, %r5; + mov.u32 %r48, %r5; + mov.u32 %r49, %r5; + @%p39 ld.global.v4.b32 { %r46, %r47, %r48, %r49 }, [ %rd12 + 0 ]; + // end inline asm + mov.b32 {%rs248, %rs249}, %r46; + mov.b32 {%rs250, %rs251}, %r47; + mov.b32 {%rs252, %rs253}, %r48; + mov.b32 {%rs254, %rs255}, %r49; + // begin inline asm + mov.u32 %r50, %r5; + mov.u32 %r51, %r5; + mov.u32 %r52, %r5; + mov.u32 %r53, %r5; + @%p12 ld.global.v4.b32 { %r50, %r51, %r52, %r53 }, [ %rd13 + 0 ]; + // end inline asm + mov.b32 {%rs256, %rs257}, %r50; + mov.b32 {%rs258, %rs259}, %r51; + mov.b32 {%rs260, %rs261}, %r52; + mov.b32 {%rs262, %rs263}, %r53; + // begin inline asm + mov.u32 %r54, %r5; + mov.u32 %r55, %r5; + mov.u32 %r56, %r5; + mov.u32 %r57, %r5; + @%p13 ld.global.v4.b32 { %r54, %r55, %r56, %r57 }, [ %rd14 + 0 ]; + // end inline asm + mov.b32 {%rs264, %rs265}, %r54; + mov.b32 {%rs266, %rs267}, %r55; + mov.b32 {%rs268, %rs269}, %r56; + mov.b32 {%rs270, %rs271}, %r57; + // begin inline asm + mov.u32 %r58, %r5; + mov.u32 %r59, %r5; + mov.u32 %r60, %r5; + mov.u32 %r61, %r5; + @%p14 ld.global.v4.b32 { %r58, %r59, %r60, %r61 }, [ %rd15 + 0 ]; + // end inline asm + mov.b32 {%rs272, %rs273}, %r58; + mov.b32 {%rs274, %rs275}, %r59; + mov.b32 {%rs276, %rs277}, %r60; + mov.b32 {%rs278, %rs279}, %r61; + // begin inline asm + mov.u32 %r62, %r5; + mov.u32 %r63, %r5; + mov.u32 %r64, %r5; + mov.u32 %r65, %r5; + @%p15 ld.global.v4.b32 { %r62, %r63, %r64, %r65 }, [ %rd16 + 0 ]; + // end inline asm + mov.b32 {%rs280, %rs281}, %r62; + mov.b32 {%rs282, %rs283}, %r63; + mov.b32 {%rs284, %rs285}, %r64; + mov.b32 {%rs286, %rs287}, %r65; + // begin inline asm + mov.u32 %r66, %r5; + mov.u32 %r67, %r5; + mov.u32 %r68, %r5; + mov.u32 %r69, %r5; + @%p16 ld.global.v4.b32 { %r66, %r67, %r68, %r69 }, [ %rd17 + 0 ]; + // end inline asm + mov.b32 {%rs288, %rs289}, %r66; + mov.b32 {%rs290, %rs291}, %r67; + mov.b32 {%rs292, %rs293}, %r68; + mov.b32 {%rs294, %rs295}, %r69; + // begin inline asm + mov.u32 %r70, %r5; + mov.u32 %r71, %r5; + mov.u32 %r72, %r5; + mov.u32 %r73, %r5; + @%p17 ld.global.v4.b32 { %r70, %r71, %r72, %r73 }, [ %rd18 + 0 ]; + // end inline asm + mov.b32 {%rs296, %rs297}, %r70; + mov.b32 {%rs298, %rs299}, %r71; + mov.b32 {%rs300, %rs301}, %r72; + mov.b32 {%rs302, %rs303}, %r73; + // begin inline asm + mov.u32 %r74, %r5; + mov.u32 %r75, %r5; + mov.u32 %r76, %r5; + mov.u32 %r77, %r5; + @%p18 ld.global.v4.b32 { %r74, %r75, %r76, %r77 }, [ %rd19 + 0 ]; + // end inline asm + mov.b32 {%rs304, %rs305}, %r74; + mov.b32 {%rs306, %rs307}, %r75; + mov.b32 {%rs308, %rs309}, %r76; + mov.b32 {%rs310, %rs311}, %r77; + // begin inline asm + mov.u32 %r78, %r5; + mov.u32 %r79, %r5; + mov.u32 %r80, %r5; + mov.u32 %r81, %r5; + @%p40 ld.global.v4.b32 { %r78, %r79, %r80, %r81 }, [ %rd20 + 0 ]; + // end inline asm + mov.b32 {%rs312, %rs313}, %r78; + mov.b32 {%rs314, %rs315}, %r79; + mov.b32 {%rs316, %rs317}, %r80; + mov.b32 {%rs318, %rs319}, %r81; + and.b32 %r868, %r865, 1; + setp.ne.b32 %p19, %r868, 0; + // begin inline asm + mov.u32 %r82, %r5; + mov.u32 %r83, %r5; + mov.u32 %r84, %r5; + mov.u32 %r85, %r5; + @%p19 ld.global.v4.b32 { %r82, %r83, %r84, %r85 }, [ %rd21 + 0 ]; + // end inline asm + mov.b32 {%rs320, %rs321}, %r82; + mov.b32 {%rs322, %rs323}, %r83; + mov.b32 {%rs324, %rs325}, %r84; + mov.b32 {%rs326, %rs327}, %r85; + and.b32 %r869, %r864, 1; + setp.ne.b32 %p20, %r869, 0; + // begin inline asm + mov.u32 %r86, %r5; + mov.u32 %r87, %r5; + mov.u32 %r88, %r5; + mov.u32 %r89, %r5; + @%p20 ld.global.v4.b32 { %r86, %r87, %r88, %r89 }, [ %rd22 + 0 ]; + // end inline asm + mov.b32 {%rs328, %rs329}, %r86; + mov.b32 {%rs330, %rs331}, %r87; + mov.b32 {%rs332, %rs333}, %r88; + mov.b32 {%rs334, %rs335}, %r89; + and.b32 %r870, %r863, 1; + setp.ne.b32 %p21, %r870, 0; + // begin inline asm + mov.u32 %r90, %r5; + mov.u32 %r91, %r5; + mov.u32 %r92, %r5; + mov.u32 %r93, %r5; + @%p21 ld.global.v4.b32 { %r90, %r91, %r92, %r93 }, [ %rd23 + 0 ]; + // end inline asm + mov.b32 {%rs336, %rs337}, %r90; + mov.b32 {%rs338, %rs339}, %r91; + mov.b32 {%rs340, %rs341}, %r92; + mov.b32 {%rs342, %rs343}, %r93; + and.b32 %r871, %r862, 1; + setp.ne.b32 %p22, %r871, 0; + // begin inline asm + mov.u32 %r94, %r5; + mov.u32 %r95, %r5; + mov.u32 %r96, %r5; + mov.u32 %r97, %r5; + @%p22 ld.global.v4.b32 { %r94, %r95, %r96, %r97 }, [ %rd24 + 0 ]; + // end inline asm + mov.b32 {%rs344, %rs345}, %r94; + mov.b32 {%rs346, %rs347}, %r95; + mov.b32 {%rs348, %rs349}, %r96; + mov.b32 {%rs350, %rs351}, %r97; + and.b32 %r872, %r861, 1; + setp.ne.b32 %p23, %r872, 0; + // begin inline asm + mov.u32 %r98, %r5; + mov.u32 %r99, %r5; + mov.u32 %r100, %r5; + mov.u32 %r101, %r5; + @%p23 ld.global.v4.b32 { %r98, %r99, %r100, %r101 }, [ %rd25 + 0 ]; + // end inline asm + mov.b32 {%rs352, %rs353}, %r98; + mov.b32 {%rs354, %rs355}, %r99; + mov.b32 {%rs356, %rs357}, %r100; + mov.b32 {%rs358, %rs359}, %r101; + and.b32 %r873, %r860, 1; + setp.ne.b32 %p24, %r873, 0; + // begin inline asm + mov.u32 %r102, %r5; + mov.u32 %r103, %r5; + mov.u32 %r104, %r5; + mov.u32 %r105, %r5; + @%p24 ld.global.v4.b32 { %r102, %r103, %r104, %r105 }, [ %rd26 + 0 ]; + // end inline asm + mov.b32 {%rs360, %rs361}, %r102; + mov.b32 {%rs362, %rs363}, %r103; + mov.b32 {%rs364, %rs365}, %r104; + mov.b32 {%rs366, %rs367}, %r105; + and.b32 %r874, %r859, 1; + setp.ne.b32 %p25, %r874, 0; + // begin inline asm + mov.u32 %r106, %r5; + mov.u32 %r107, %r5; + mov.u32 %r108, %r5; + mov.u32 %r109, %r5; + @%p25 ld.global.v4.b32 { %r106, %r107, %r108, %r109 }, [ %rd27 + 0 ]; + // end inline asm + mov.b32 {%rs368, %rs369}, %r106; + mov.b32 {%rs370, %rs371}, %r107; + mov.b32 {%rs372, %rs373}, %r108; + mov.b32 {%rs374, %rs375}, %r109; + and.b32 %r875, %r858, 1; + setp.ne.b32 %p26, %r875, 0; + // begin inline asm + mov.u32 %r110, %r5; + mov.u32 %r111, %r5; + mov.u32 %r112, %r5; + mov.u32 %r113, %r5; + @%p26 ld.global.v4.b32 { %r110, %r111, %r112, %r113 }, [ %rd28 + 0 ]; + // end inline asm + mov.b32 {%rs376, %rs377}, %r110; + mov.b32 {%rs378, %rs379}, %r111; + mov.b32 {%rs380, %rs381}, %r112; + mov.b32 {%rs382, %rs383}, %r113; + and.b32 %r876, %r857, 1; + setp.ne.b32 %p27, %r876, 0; + // begin inline asm + mov.u32 %r114, %r5; + mov.u32 %r115, %r5; + mov.u32 %r116, %r5; + mov.u32 %r117, %r5; + @%p27 ld.global.v4.b32 { %r114, %r115, %r116, %r117 }, [ %rd29 + 0 ]; + // end inline asm + mov.b32 {%rs384, %rs385}, %r114; + mov.b32 {%rs386, %rs387}, %r115; + mov.b32 {%rs388, %rs389}, %r116; + mov.b32 {%rs390, %rs391}, %r117; + and.b32 %r877, %r856, 1; + setp.ne.b32 %p28, %r877, 0; + // begin inline asm + mov.u32 %r118, %r5; + mov.u32 %r119, %r5; + mov.u32 %r120, %r5; + mov.u32 %r121, %r5; + @%p28 ld.global.v4.b32 { %r118, %r119, %r120, %r121 }, [ %rd30 + 0 ]; + // end inline asm + mov.b32 {%rs392, %rs393}, %r118; + mov.b32 {%rs394, %rs395}, %r119; + mov.b32 {%rs396, %rs397}, %r120; + mov.b32 {%rs398, %rs399}, %r121; + and.b32 %r878, %r855, 1; + setp.ne.b32 %p29, %r878, 0; + // begin inline asm + mov.u32 %r122, %r5; + mov.u32 %r123, %r5; + mov.u32 %r124, %r5; + mov.u32 %r125, %r5; + @%p29 ld.global.v4.b32 { %r122, %r123, %r124, %r125 }, [ %rd31 + 0 ]; + // end inline asm + mov.b32 {%rs400, %rs401}, %r122; + mov.b32 {%rs402, %rs403}, %r123; + mov.b32 {%rs404, %rs405}, %r124; + mov.b32 {%rs406, %rs407}, %r125; + and.b32 %r879, %r854, 1; + setp.ne.b32 %p30, %r879, 0; + // begin inline asm + mov.u32 %r126, %r5; + mov.u32 %r127, %r5; + mov.u32 %r128, %r5; + mov.u32 %r129, %r5; + @%p30 ld.global.v4.b32 { %r126, %r127, %r128, %r129 }, [ %rd32 + 0 ]; + // end inline asm + mov.b32 {%rs408, %rs409}, %r126; + mov.b32 {%rs410, %rs411}, %r127; + mov.b32 {%rs412, %rs413}, %r128; + mov.b32 {%rs414, %rs415}, %r129; + and.b32 %r880, %r853, 1; + setp.ne.b32 %p31, %r880, 0; + // begin inline asm + mov.u32 %r130, %r5; + mov.u32 %r131, %r5; + mov.u32 %r132, %r5; + mov.u32 %r133, %r5; + @%p31 ld.global.v4.b32 { %r130, %r131, %r132, %r133 }, [ %rd33 + 0 ]; + // end inline asm + mov.b32 {%rs416, %rs417}, %r130; + mov.b32 {%rs418, %rs419}, %r131; + mov.b32 {%rs420, %rs421}, %r132; + mov.b32 {%rs422, %rs423}, %r133; + and.b32 %r881, %r852, 1; + setp.ne.b32 %p32, %r881, 0; + // begin inline asm + mov.u32 %r134, %r5; + mov.u32 %r135, %r5; + mov.u32 %r136, %r5; + mov.u32 %r137, %r5; + @%p32 ld.global.v4.b32 { %r134, %r135, %r136, %r137 }, [ %rd34 + 0 ]; + // end inline asm + mov.b32 {%rs424, %rs425}, %r134; + mov.b32 {%rs426, %rs427}, %r135; + mov.b32 {%rs428, %rs429}, %r136; + mov.b32 {%rs430, %rs431}, %r137; + and.b32 %r882, %r851, 1; + setp.ne.b32 %p33, %r882, 0; + // begin inline asm + mov.u32 %r138, %r5; + mov.u32 %r139, %r5; + mov.u32 %r140, %r5; + mov.u32 %r141, %r5; + @%p33 ld.global.v4.b32 { %r138, %r139, %r140, %r141 }, [ %rd35 + 0 ]; + // end inline asm + mov.b32 {%rs432, %rs433}, %r138; + mov.b32 {%rs434, %rs435}, %r139; + mov.b32 {%rs436, %rs437}, %r140; + mov.b32 {%rs438, %rs439}, %r141; + // begin inline asm + mov.u32 %r142, %r5; + mov.u32 %r143, %r5; + mov.u32 %r144, %r5; + mov.u32 %r145, %r5; + @%p34 ld.global.v4.b32 { %r142, %r143, %r144, %r145 }, [ %rd36 + 0 ]; + // end inline asm + mov.b32 {%rs440, %rs441}, %r142; + mov.b32 {%rs442, %rs443}, %r143; + mov.b32 {%rs444, %rs445}, %r144; + mov.b32 {%rs446, %rs447}, %r145; + shr.u64 %rd601, %rd339, 31; + and.b64 %rd602, %rd601, 1; + setp.ne.b64 %p41, %rd602, 0; + // begin inline asm + mov.u32 %r146, %r5; + mov.u32 %r147, %r5; + mov.u32 %r148, %r5; + mov.u32 %r149, %r5; + @%p41 ld.global.v4.b32 { %r146, %r147, %r148, %r149 }, [ %rd37 + 0 ]; + // end inline asm + mov.b32 {%rs448, %rs449}, %r146; + mov.b32 {%rs450, %rs451}, %r147; + mov.b32 {%rs452, %rs453}, %r148; + mov.b32 {%rs454, %rs455}, %r149; + shr.u64 %rd603, %rd339, 30; + and.b64 %rd604, %rd603, 1; + setp.ne.b64 %p42, %rd604, 0; + // begin inline asm + mov.u32 %r150, %r5; + mov.u32 %r151, %r5; + mov.u32 %r152, %r5; + mov.u32 %r153, %r5; + @%p42 ld.global.v4.b32 { %r150, %r151, %r152, %r153 }, [ %rd38 + 0 ]; + // end inline asm + mov.b32 {%rs456, %rs457}, %r150; + mov.b32 {%rs458, %rs459}, %r151; + mov.b32 {%rs460, %rs461}, %r152; + mov.b32 {%rs462, %rs463}, %r153; + shr.u64 %rd605, %rd339, 29; + and.b64 %rd606, %rd605, 1; + setp.ne.b64 %p43, %rd606, 0; + // begin inline asm + mov.u32 %r154, %r5; + mov.u32 %r155, %r5; + mov.u32 %r156, %r5; + mov.u32 %r157, %r5; + @%p43 ld.global.v4.b32 { %r154, %r155, %r156, %r157 }, [ %rd39 + 0 ]; + // end inline asm + mov.b32 {%rs464, %rs465}, %r154; + mov.b32 {%rs466, %rs467}, %r155; + mov.b32 {%rs468, %rs469}, %r156; + mov.b32 {%rs470, %rs471}, %r157; + shr.u64 %rd607, %rd339, 28; + and.b64 %rd608, %rd607, 1; + setp.ne.b64 %p44, %rd608, 0; + // begin inline asm + mov.u32 %r158, %r5; + mov.u32 %r159, %r5; + mov.u32 %r160, %r5; + mov.u32 %r161, %r5; + @%p44 ld.global.v4.b32 { %r158, %r159, %r160, %r161 }, [ %rd40 + 0 ]; + // end inline asm + mov.b32 {%rs472, %rs473}, %r158; + mov.b32 {%rs474, %rs475}, %r159; + mov.b32 {%rs476, %rs477}, %r160; + mov.b32 {%rs478, %rs479}, %r161; + shr.u64 %rd609, %rd339, 27; + and.b64 %rd610, %rd609, 1; + setp.ne.b64 %p45, %rd610, 0; + // begin inline asm + mov.u32 %r162, %r5; + mov.u32 %r163, %r5; + mov.u32 %r164, %r5; + mov.u32 %r165, %r5; + @%p45 ld.global.v4.b32 { %r162, %r163, %r164, %r165 }, [ %rd41 + 0 ]; + // end inline asm + mov.b32 {%rs480, %rs481}, %r162; + mov.b32 {%rs482, %rs483}, %r163; + mov.b32 {%rs484, %rs485}, %r164; + mov.b32 {%rs486, %rs487}, %r165; + shr.u64 %rd611, %rd339, 26; + and.b64 %rd612, %rd611, 1; + setp.ne.b64 %p46, %rd612, 0; + // begin inline asm + mov.u32 %r166, %r5; + mov.u32 %r167, %r5; + mov.u32 %r168, %r5; + mov.u32 %r169, %r5; + @%p46 ld.global.v4.b32 { %r166, %r167, %r168, %r169 }, [ %rd42 + 0 ]; + // end inline asm + mov.b32 {%rs488, %rs489}, %r166; + mov.b32 {%rs490, %rs491}, %r167; + mov.b32 {%rs492, %rs493}, %r168; + mov.b32 {%rs494, %rs495}, %r169; + shr.u64 %rd613, %rd339, 25; + and.b64 %rd614, %rd613, 1; + setp.ne.b64 %p47, %rd614, 0; + // begin inline asm + mov.u32 %r170, %r5; + mov.u32 %r171, %r5; + mov.u32 %r172, %r5; + mov.u32 %r173, %r5; + @%p47 ld.global.v4.b32 { %r170, %r171, %r172, %r173 }, [ %rd43 + 0 ]; + // end inline asm + mov.b32 {%rs496, %rs497}, %r170; + mov.b32 {%rs498, %rs499}, %r171; + mov.b32 {%rs500, %rs501}, %r172; + mov.b32 {%rs502, %rs503}, %r173; + shr.u64 %rd615, %rd339, 24; + and.b64 %rd616, %rd615, 1; + setp.ne.b64 %p48, %rd616, 0; + // begin inline asm + mov.u32 %r174, %r5; + mov.u32 %r175, %r5; + mov.u32 %r176, %r5; + mov.u32 %r177, %r5; + @%p48 ld.global.v4.b32 { %r174, %r175, %r176, %r177 }, [ %rd44 + 0 ]; + // end inline asm + mov.b32 {%rs504, %rs505}, %r174; + mov.b32 {%rs506, %rs507}, %r175; + mov.b32 {%rs508, %rs509}, %r176; + mov.b32 {%rs510, %rs511}, %r177; + shr.u64 %rd617, %rd339, 23; + and.b64 %rd618, %rd617, 1; + setp.ne.b64 %p49, %rd618, 0; + // begin inline asm + mov.u32 %r178, %r5; + mov.u32 %r179, %r5; + mov.u32 %r180, %r5; + mov.u32 %r181, %r5; + @%p49 ld.global.v4.b32 { %r178, %r179, %r180, %r181 }, [ %rd45 + 0 ]; + // end inline asm + mov.b32 {%rs512, %rs513}, %r178; + mov.b32 {%rs514, %rs515}, %r179; + mov.b32 {%rs516, %rs517}, %r180; + mov.b32 {%rs518, %rs519}, %r181; + shr.u64 %rd619, %rd339, 22; + and.b64 %rd620, %rd619, 1; + setp.ne.b64 %p50, %rd620, 0; + // begin inline asm + mov.u32 %r182, %r5; + mov.u32 %r183, %r5; + mov.u32 %r184, %r5; + mov.u32 %r185, %r5; + @%p50 ld.global.v4.b32 { %r182, %r183, %r184, %r185 }, [ %rd46 + 0 ]; + // end inline asm + mov.b32 {%rs520, %rs521}, %r182; + mov.b32 {%rs522, %rs523}, %r183; + mov.b32 {%rs524, %rs525}, %r184; + mov.b32 {%rs526, %rs527}, %r185; + shr.u64 %rd621, %rd339, 21; + and.b64 %rd622, %rd621, 1; + setp.ne.b64 %p51, %rd622, 0; + // begin inline asm + mov.u32 %r186, %r5; + mov.u32 %r187, %r5; + mov.u32 %r188, %r5; + mov.u32 %r189, %r5; + @%p51 ld.global.v4.b32 { %r186, %r187, %r188, %r189 }, [ %rd47 + 0 ]; + // end inline asm + mov.b32 {%rs528, %rs529}, %r186; + mov.b32 {%rs530, %rs531}, %r187; + mov.b32 {%rs532, %rs533}, %r188; + mov.b32 {%rs534, %rs535}, %r189; + shr.u64 %rd623, %rd339, 20; + and.b64 %rd624, %rd623, 1; + setp.ne.b64 %p52, %rd624, 0; + // begin inline asm + mov.u32 %r190, %r5; + mov.u32 %r191, %r5; + mov.u32 %r192, %r5; + mov.u32 %r193, %r5; + @%p52 ld.global.v4.b32 { %r190, %r191, %r192, %r193 }, [ %rd48 + 0 ]; + // end inline asm + mov.b32 {%rs536, %rs537}, %r190; + mov.b32 {%rs538, %rs539}, %r191; + mov.b32 {%rs540, %rs541}, %r192; + mov.b32 {%rs542, %rs543}, %r193; + shr.u64 %rd625, %rd339, 19; + and.b64 %rd626, %rd625, 1; + setp.ne.b64 %p53, %rd626, 0; + // begin inline asm + mov.u32 %r194, %r5; + mov.u32 %r195, %r5; + mov.u32 %r196, %r5; + mov.u32 %r197, %r5; + @%p53 ld.global.v4.b32 { %r194, %r195, %r196, %r197 }, [ %rd49 + 0 ]; + // end inline asm + mov.b32 {%rs544, %rs545}, %r194; + mov.b32 {%rs546, %rs547}, %r195; + mov.b32 {%rs548, %rs549}, %r196; + mov.b32 {%rs550, %rs551}, %r197; + shr.u64 %rd627, %rd339, 18; + and.b64 %rd628, %rd627, 1; + setp.ne.b64 %p54, %rd628, 0; + // begin inline asm + mov.u32 %r198, %r5; + mov.u32 %r199, %r5; + mov.u32 %r200, %r5; + mov.u32 %r201, %r5; + @%p54 ld.global.v4.b32 { %r198, %r199, %r200, %r201 }, [ %rd50 + 0 ]; + // end inline asm + mov.b32 {%rs552, %rs553}, %r198; + mov.b32 {%rs554, %rs555}, %r199; + mov.b32 {%rs556, %rs557}, %r200; + mov.b32 {%rs558, %rs559}, %r201; + shr.u64 %rd629, %rd339, 17; + and.b64 %rd630, %rd629, 1; + setp.ne.b64 %p55, %rd630, 0; + // begin inline asm + mov.u32 %r202, %r5; + mov.u32 %r203, %r5; + mov.u32 %r204, %r5; + mov.u32 %r205, %r5; + @%p55 ld.global.v4.b32 { %r202, %r203, %r204, %r205 }, [ %rd51 + 0 ]; + // end inline asm + mov.b32 {%rs560, %rs561}, %r202; + mov.b32 {%rs562, %rs563}, %r203; + mov.b32 {%rs564, %rs565}, %r204; + mov.b32 {%rs566, %rs567}, %r205; + shr.u64 %rd631, %rd339, 16; + and.b64 %rd632, %rd631, 1; + setp.ne.b64 %p56, %rd632, 0; + // begin inline asm + mov.u32 %r206, %r5; + mov.u32 %r207, %r5; + mov.u32 %r208, %r5; + mov.u32 %r209, %r5; + @%p56 ld.global.v4.b32 { %r206, %r207, %r208, %r209 }, [ %rd52 + 0 ]; + // end inline asm + mov.b32 {%rs568, %rs569}, %r206; + mov.b32 {%rs570, %rs571}, %r207; + mov.b32 {%rs572, %rs573}, %r208; + mov.b32 {%rs574, %rs575}, %r209; + shr.u64 %rd633, %rd339, 15; + and.b64 %rd634, %rd633, 1; + setp.ne.b64 %p57, %rd634, 0; + // begin inline asm + mov.u32 %r210, %r5; + mov.u32 %r211, %r5; + mov.u32 %r212, %r5; + mov.u32 %r213, %r5; + @%p57 ld.global.v4.b32 { %r210, %r211, %r212, %r213 }, [ %rd53 + 0 ]; + // end inline asm + mov.b32 {%rs576, %rs577}, %r210; + mov.b32 {%rs578, %rs579}, %r211; + mov.b32 {%rs580, %rs581}, %r212; + mov.b32 {%rs582, %rs583}, %r213; + shr.u64 %rd635, %rd339, 14; + and.b64 %rd636, %rd635, 1; + setp.ne.b64 %p58, %rd636, 0; + // begin inline asm + mov.u32 %r214, %r5; + mov.u32 %r215, %r5; + mov.u32 %r216, %r5; + mov.u32 %r217, %r5; + @%p58 ld.global.v4.b32 { %r214, %r215, %r216, %r217 }, [ %rd54 + 0 ]; + // end inline asm + mov.b32 {%rs584, %rs585}, %r214; + mov.b32 {%rs586, %rs587}, %r215; + mov.b32 {%rs588, %rs589}, %r216; + mov.b32 {%rs590, %rs591}, %r217; + shr.u64 %rd637, %rd339, 13; + and.b64 %rd638, %rd637, 1; + setp.ne.b64 %p59, %rd638, 0; + // begin inline asm + mov.u32 %r218, %r5; + mov.u32 %r219, %r5; + mov.u32 %r220, %r5; + mov.u32 %r221, %r5; + @%p59 ld.global.v4.b32 { %r218, %r219, %r220, %r221 }, [ %rd55 + 0 ]; + // end inline asm + mov.b32 {%rs592, %rs593}, %r218; + mov.b32 {%rs594, %rs595}, %r219; + mov.b32 {%rs596, %rs597}, %r220; + mov.b32 {%rs598, %rs599}, %r221; + shr.u64 %rd639, %rd339, 12; + and.b64 %rd640, %rd639, 1; + setp.ne.b64 %p60, %rd640, 0; + // begin inline asm + mov.u32 %r222, %r5; + mov.u32 %r223, %r5; + mov.u32 %r224, %r5; + mov.u32 %r225, %r5; + @%p60 ld.global.v4.b32 { %r222, %r223, %r224, %r225 }, [ %rd56 + 0 ]; + // end inline asm + mov.b32 {%rs600, %rs601}, %r222; + mov.b32 {%rs602, %rs603}, %r223; + mov.b32 {%rs604, %rs605}, %r224; + mov.b32 {%rs606, %rs607}, %r225; + shr.u64 %rd641, %rd339, 11; + and.b64 %rd642, %rd641, 1; + setp.ne.b64 %p61, %rd642, 0; + // begin inline asm + mov.u32 %r226, %r5; + mov.u32 %r227, %r5; + mov.u32 %r228, %r5; + mov.u32 %r229, %r5; + @%p61 ld.global.v4.b32 { %r226, %r227, %r228, %r229 }, [ %rd57 + 0 ]; + // end inline asm + mov.b32 {%rs608, %rs609}, %r226; + mov.b32 {%rs610, %rs611}, %r227; + mov.b32 {%rs612, %rs613}, %r228; + mov.b32 {%rs614, %rs615}, %r229; + shr.u64 %rd643, %rd339, 10; + and.b64 %rd644, %rd643, 1; + setp.ne.b64 %p62, %rd644, 0; + // begin inline asm + mov.u32 %r230, %r5; + mov.u32 %r231, %r5; + mov.u32 %r232, %r5; + mov.u32 %r233, %r5; + @%p62 ld.global.v4.b32 { %r230, %r231, %r232, %r233 }, [ %rd58 + 0 ]; + // end inline asm + mov.b32 {%rs616, %rs617}, %r230; + mov.b32 {%rs618, %rs619}, %r231; + mov.b32 {%rs620, %rs621}, %r232; + mov.b32 {%rs622, %rs623}, %r233; + shr.u64 %rd645, %rd339, 9; + and.b64 %rd646, %rd645, 1; + setp.ne.b64 %p63, %rd646, 0; + // begin inline asm + mov.u32 %r234, %r5; + mov.u32 %r235, %r5; + mov.u32 %r236, %r5; + mov.u32 %r237, %r5; + @%p63 ld.global.v4.b32 { %r234, %r235, %r236, %r237 }, [ %rd59 + 0 ]; + // end inline asm + mov.b32 {%rs624, %rs625}, %r234; + mov.b32 {%rs626, %rs627}, %r235; + mov.b32 {%rs628, %rs629}, %r236; + mov.b32 {%rs630, %rs631}, %r237; + shr.u64 %rd647, %rd339, 8; + and.b64 %rd648, %rd647, 1; + setp.ne.b64 %p64, %rd648, 0; + // begin inline asm + mov.u32 %r238, %r5; + mov.u32 %r239, %r5; + mov.u32 %r240, %r5; + mov.u32 %r241, %r5; + @%p64 ld.global.v4.b32 { %r238, %r239, %r240, %r241 }, [ %rd60 + 0 ]; + // end inline asm + mov.b32 {%rs632, %rs633}, %r238; + mov.b32 {%rs634, %rs635}, %r239; + mov.b32 {%rs636, %rs637}, %r240; + mov.b32 {%rs638, %rs639}, %r241; + shr.u64 %rd649, %rd339, 7; + and.b64 %rd650, %rd649, 1; + setp.ne.b64 %p65, %rd650, 0; + // begin inline asm + mov.u32 %r242, %r5; + mov.u32 %r243, %r5; + mov.u32 %r244, %r5; + mov.u32 %r245, %r5; + @%p65 ld.global.v4.b32 { %r242, %r243, %r244, %r245 }, [ %rd61 + 0 ]; + // end inline asm + mov.b32 {%rs640, %rs641}, %r242; + mov.b32 {%rs642, %rs643}, %r243; + mov.b32 {%rs644, %rs645}, %r244; + mov.b32 {%rs646, %rs647}, %r245; + shr.u64 %rd651, %rd339, 6; + and.b64 %rd652, %rd651, 1; + setp.ne.b64 %p66, %rd652, 0; + // begin inline asm + mov.u32 %r246, %r5; + mov.u32 %r247, %r5; + mov.u32 %r248, %r5; + mov.u32 %r249, %r5; + @%p66 ld.global.v4.b32 { %r246, %r247, %r248, %r249 }, [ %rd62 + 0 ]; + // end inline asm + mov.b32 {%rs648, %rs649}, %r246; + mov.b32 {%rs650, %rs651}, %r247; + mov.b32 {%rs652, %rs653}, %r248; + mov.b32 {%rs654, %rs655}, %r249; + shr.u64 %rd653, %rd339, 5; + and.b64 %rd654, %rd653, 1; + setp.ne.b64 %p67, %rd654, 0; + // begin inline asm + mov.u32 %r250, %r5; + mov.u32 %r251, %r5; + mov.u32 %r252, %r5; + mov.u32 %r253, %r5; + @%p67 ld.global.v4.b32 { %r250, %r251, %r252, %r253 }, [ %rd63 + 0 ]; + // end inline asm + mov.b32 {%rs656, %rs657}, %r250; + mov.b32 {%rs658, %rs659}, %r251; + mov.b32 {%rs660, %rs661}, %r252; + mov.b32 {%rs662, %rs663}, %r253; + shr.u64 %rd655, %rd339, 4; + and.b64 %rd656, %rd655, 1; + setp.ne.b64 %p68, %rd656, 0; + // begin inline asm + mov.u32 %r254, %r5; + mov.u32 %r255, %r5; + mov.u32 %r256, %r5; + mov.u32 %r257, %r5; + @%p68 ld.global.v4.b32 { %r254, %r255, %r256, %r257 }, [ %rd64 + 0 ]; + // end inline asm + mov.b32 {%rs664, %rs665}, %r254; + mov.b32 {%rs666, %rs667}, %r255; + mov.b32 {%rs668, %rs669}, %r256; + mov.b32 {%rs670, %rs671}, %r257; + .loc 1 104 47 // l2norm.py:104:47 + cvt.f32.bf16 %r883, %rs160; + cvt.f32.bf16 %r884, %rs161; + cvt.f32.bf16 %r885, %rs162; + cvt.f32.bf16 %r886, %rs163; + cvt.f32.bf16 %r887, %rs164; + cvt.f32.bf16 %r888, %rs165; + cvt.f32.bf16 %r889, %rs166; + cvt.f32.bf16 %r890, %rs167; + cvt.f32.bf16 %r891, %rs168; + cvt.f32.bf16 %r892, %rs169; + cvt.f32.bf16 %r893, %rs170; + cvt.f32.bf16 %r894, %rs171; + cvt.f32.bf16 %r895, %rs172; + cvt.f32.bf16 %r896, %rs173; + cvt.f32.bf16 %r897, %rs174; + cvt.f32.bf16 %r898, %rs175; + cvt.f32.bf16 %r899, %rs176; + cvt.f32.bf16 %r900, %rs177; + cvt.f32.bf16 %r901, %rs178; + cvt.f32.bf16 %r902, %rs179; + cvt.f32.bf16 %r903, %rs180; + cvt.f32.bf16 %r904, %rs181; + cvt.f32.bf16 %r905, %rs182; + cvt.f32.bf16 %r906, %rs183; + cvt.f32.bf16 %r907, %rs184; + cvt.f32.bf16 %r908, %rs185; + cvt.f32.bf16 %r909, %rs186; + cvt.f32.bf16 %r910, %rs187; + cvt.f32.bf16 %r911, %rs188; + cvt.f32.bf16 %r912, %rs189; + cvt.f32.bf16 %r913, %rs190; + cvt.f32.bf16 %r914, %rs191; + cvt.f32.bf16 %r915, %rs192; + cvt.f32.bf16 %r916, %rs193; + cvt.f32.bf16 %r917, %rs194; + cvt.f32.bf16 %r918, %rs195; + cvt.f32.bf16 %r919, %rs196; + cvt.f32.bf16 %r920, %rs197; + cvt.f32.bf16 %r921, %rs198; + cvt.f32.bf16 %r922, %rs199; + cvt.f32.bf16 %r923, %rs200; + cvt.f32.bf16 %r924, %rs201; + cvt.f32.bf16 %r925, %rs202; + cvt.f32.bf16 %r926, %rs203; + cvt.f32.bf16 %r927, %rs204; + cvt.f32.bf16 %r928, %rs205; + cvt.f32.bf16 %r929, %rs206; + cvt.f32.bf16 %r930, %rs207; + cvt.f32.bf16 %r931, %rs208; + cvt.f32.bf16 %r932, %rs209; + cvt.f32.bf16 %r933, %rs210; + cvt.f32.bf16 %r934, %rs211; + cvt.f32.bf16 %r935, %rs212; + cvt.f32.bf16 %r936, %rs213; + cvt.f32.bf16 %r937, %rs214; + cvt.f32.bf16 %r938, %rs215; + cvt.f32.bf16 %r939, %rs216; + cvt.f32.bf16 %r940, %rs217; + cvt.f32.bf16 %r941, %rs218; + cvt.f32.bf16 %r942, %rs219; + cvt.f32.bf16 %r943, %rs220; + cvt.f32.bf16 %r944, %rs221; + cvt.f32.bf16 %r945, %rs222; + cvt.f32.bf16 %r946, %rs223; + cvt.f32.bf16 %r947, %rs224; + cvt.f32.bf16 %r948, %rs225; + cvt.f32.bf16 %r949, %rs226; + cvt.f32.bf16 %r950, %rs227; + cvt.f32.bf16 %r951, %rs228; + cvt.f32.bf16 %r952, %rs229; + cvt.f32.bf16 %r953, %rs230; + cvt.f32.bf16 %r954, %rs231; + cvt.f32.bf16 %r955, %rs232; + cvt.f32.bf16 %r956, %rs233; + cvt.f32.bf16 %r957, %rs234; + cvt.f32.bf16 %r958, %rs235; + cvt.f32.bf16 %r959, %rs236; + cvt.f32.bf16 %r960, %rs237; + cvt.f32.bf16 %r961, %rs238; + cvt.f32.bf16 %r962, %rs239; + cvt.f32.bf16 %r963, %rs240; + cvt.f32.bf16 %r964, %rs241; + cvt.f32.bf16 %r965, %rs242; + cvt.f32.bf16 %r966, %rs243; + cvt.f32.bf16 %r967, %rs244; + cvt.f32.bf16 %r968, %rs245; + cvt.f32.bf16 %r969, %rs246; + cvt.f32.bf16 %r970, %rs247; + cvt.f32.bf16 %r971, %rs248; + cvt.f32.bf16 %r972, %rs249; + cvt.f32.bf16 %r973, %rs250; + cvt.f32.bf16 %r974, %rs251; + cvt.f32.bf16 %r975, %rs252; + cvt.f32.bf16 %r976, %rs253; + cvt.f32.bf16 %r977, %rs254; + cvt.f32.bf16 %r978, %rs255; + cvt.f32.bf16 %r979, %rs256; + cvt.f32.bf16 %r980, %rs257; + cvt.f32.bf16 %r981, %rs258; + cvt.f32.bf16 %r982, %rs259; + cvt.f32.bf16 %r983, %rs260; + cvt.f32.bf16 %r984, %rs261; + cvt.f32.bf16 %r985, %rs262; + cvt.f32.bf16 %r986, %rs263; + cvt.f32.bf16 %r987, %rs264; + cvt.f32.bf16 %r988, %rs265; + cvt.f32.bf16 %r989, %rs266; + cvt.f32.bf16 %r990, %rs267; + cvt.f32.bf16 %r991, %rs268; + cvt.f32.bf16 %r992, %rs269; + cvt.f32.bf16 %r993, %rs270; + cvt.f32.bf16 %r994, %rs271; + cvt.f32.bf16 %r995, %rs272; + cvt.f32.bf16 %r996, %rs273; + cvt.f32.bf16 %r997, %rs274; + cvt.f32.bf16 %r998, %rs275; + cvt.f32.bf16 %r999, %rs276; + cvt.f32.bf16 %r1000, %rs277; + cvt.f32.bf16 %r1001, %rs278; + cvt.f32.bf16 %r1002, %rs279; + cvt.f32.bf16 %r1003, %rs280; + cvt.f32.bf16 %r1004, %rs281; + cvt.f32.bf16 %r1005, %rs282; + cvt.f32.bf16 %r1006, %rs283; + cvt.f32.bf16 %r1007, %rs284; + cvt.f32.bf16 %r1008, %rs285; + cvt.f32.bf16 %r1009, %rs286; + cvt.f32.bf16 %r1010, %rs287; + cvt.f32.bf16 %r1011, %rs288; + cvt.f32.bf16 %r1012, %rs289; + cvt.f32.bf16 %r1013, %rs290; + cvt.f32.bf16 %r1014, %rs291; + cvt.f32.bf16 %r1015, %rs292; + cvt.f32.bf16 %r1016, %rs293; + cvt.f32.bf16 %r1017, %rs294; + cvt.f32.bf16 %r1018, %rs295; + cvt.f32.bf16 %r1019, %rs296; + cvt.f32.bf16 %r1020, %rs297; + cvt.f32.bf16 %r1021, %rs298; + cvt.f32.bf16 %r1022, %rs299; + cvt.f32.bf16 %r1023, %rs300; + cvt.f32.bf16 %r1024, %rs301; + cvt.f32.bf16 %r1025, %rs302; + cvt.f32.bf16 %r1026, %rs303; + cvt.f32.bf16 %r1027, %rs304; + cvt.f32.bf16 %r1028, %rs305; + cvt.f32.bf16 %r1029, %rs306; + cvt.f32.bf16 %r1030, %rs307; + cvt.f32.bf16 %r1031, %rs308; + cvt.f32.bf16 %r1032, %rs309; + cvt.f32.bf16 %r1033, %rs310; + cvt.f32.bf16 %r1034, %rs311; + cvt.f32.bf16 %r1035, %rs312; + cvt.f32.bf16 %r1036, %rs313; + cvt.f32.bf16 %r1037, %rs314; + cvt.f32.bf16 %r1038, %rs315; + cvt.f32.bf16 %r1039, %rs316; + cvt.f32.bf16 %r1040, %rs317; + cvt.f32.bf16 %r1041, %rs318; + cvt.f32.bf16 %r1042, %rs319; + cvt.f32.bf16 %r1043, %rs320; + cvt.f32.bf16 %r1044, %rs321; + cvt.f32.bf16 %r1045, %rs322; + cvt.f32.bf16 %r1046, %rs323; + cvt.f32.bf16 %r1047, %rs324; + cvt.f32.bf16 %r1048, %rs325; + cvt.f32.bf16 %r1049, %rs326; + cvt.f32.bf16 %r1050, %rs327; + cvt.f32.bf16 %r1051, %rs328; + cvt.f32.bf16 %r1052, %rs329; + cvt.f32.bf16 %r1053, %rs330; + cvt.f32.bf16 %r1054, %rs331; + cvt.f32.bf16 %r1055, %rs332; + cvt.f32.bf16 %r1056, %rs333; + cvt.f32.bf16 %r1057, %rs334; + cvt.f32.bf16 %r1058, %rs335; + cvt.f32.bf16 %r1059, %rs336; + cvt.f32.bf16 %r1060, %rs337; + cvt.f32.bf16 %r1061, %rs338; + cvt.f32.bf16 %r1062, %rs339; + cvt.f32.bf16 %r1063, %rs340; + cvt.f32.bf16 %r1064, %rs341; + cvt.f32.bf16 %r1065, %rs342; + cvt.f32.bf16 %r1066, %rs343; + cvt.f32.bf16 %r1067, %rs344; + cvt.f32.bf16 %r1068, %rs345; + cvt.f32.bf16 %r1069, %rs346; + cvt.f32.bf16 %r1070, %rs347; + cvt.f32.bf16 %r1071, %rs348; + cvt.f32.bf16 %r1072, %rs349; + cvt.f32.bf16 %r1073, %rs350; + cvt.f32.bf16 %r1074, %rs351; + cvt.f32.bf16 %r1075, %rs352; + cvt.f32.bf16 %r1076, %rs353; + cvt.f32.bf16 %r1077, %rs354; + cvt.f32.bf16 %r1078, %rs355; + cvt.f32.bf16 %r1079, %rs356; + cvt.f32.bf16 %r1080, %rs357; + cvt.f32.bf16 %r1081, %rs358; + cvt.f32.bf16 %r1082, %rs359; + cvt.f32.bf16 %r1083, %rs360; + cvt.f32.bf16 %r1084, %rs361; + cvt.f32.bf16 %r1085, %rs362; + cvt.f32.bf16 %r1086, %rs363; + cvt.f32.bf16 %r1087, %rs364; + cvt.f32.bf16 %r1088, %rs365; + cvt.f32.bf16 %r1089, %rs366; + cvt.f32.bf16 %r1090, %rs367; + cvt.f32.bf16 %r1091, %rs368; + cvt.f32.bf16 %r1092, %rs369; + cvt.f32.bf16 %r1093, %rs370; + cvt.f32.bf16 %r1094, %rs371; + cvt.f32.bf16 %r1095, %rs372; + cvt.f32.bf16 %r1096, %rs373; + cvt.f32.bf16 %r1097, %rs374; + cvt.f32.bf16 %r1098, %rs375; + cvt.f32.bf16 %r1099, %rs376; + cvt.f32.bf16 %r1100, %rs377; + cvt.f32.bf16 %r1101, %rs378; + cvt.f32.bf16 %r1102, %rs379; + cvt.f32.bf16 %r1103, %rs380; + cvt.f32.bf16 %r1104, %rs381; + cvt.f32.bf16 %r1105, %rs382; + cvt.f32.bf16 %r1106, %rs383; + cvt.f32.bf16 %r1107, %rs384; + cvt.f32.bf16 %r1108, %rs385; + cvt.f32.bf16 %r1109, %rs386; + cvt.f32.bf16 %r1110, %rs387; + cvt.f32.bf16 %r1111, %rs388; + cvt.f32.bf16 %r1112, %rs389; + cvt.f32.bf16 %r1113, %rs390; + cvt.f32.bf16 %r1114, %rs391; + cvt.f32.bf16 %r1115, %rs392; + cvt.f32.bf16 %r1116, %rs393; + cvt.f32.bf16 %r1117, %rs394; + cvt.f32.bf16 %r1118, %rs395; + cvt.f32.bf16 %r1119, %rs396; + cvt.f32.bf16 %r1120, %rs397; + cvt.f32.bf16 %r1121, %rs398; + cvt.f32.bf16 %r1122, %rs399; + cvt.f32.bf16 %r1123, %rs400; + cvt.f32.bf16 %r1124, %rs401; + cvt.f32.bf16 %r1125, %rs402; + cvt.f32.bf16 %r1126, %rs403; + cvt.f32.bf16 %r1127, %rs404; + cvt.f32.bf16 %r1128, %rs405; + cvt.f32.bf16 %r1129, %rs406; + cvt.f32.bf16 %r1130, %rs407; + cvt.f32.bf16 %r1131, %rs408; + cvt.f32.bf16 %r1132, %rs409; + cvt.f32.bf16 %r1133, %rs410; + cvt.f32.bf16 %r1134, %rs411; + cvt.f32.bf16 %r1135, %rs412; + cvt.f32.bf16 %r1136, %rs413; + cvt.f32.bf16 %r1137, %rs414; + cvt.f32.bf16 %r1138, %rs415; + cvt.f32.bf16 %r1139, %rs416; + cvt.f32.bf16 %r1140, %rs417; + cvt.f32.bf16 %r1141, %rs418; + cvt.f32.bf16 %r1142, %rs419; + cvt.f32.bf16 %r1143, %rs420; + cvt.f32.bf16 %r1144, %rs421; + cvt.f32.bf16 %r1145, %rs422; + cvt.f32.bf16 %r1146, %rs423; + cvt.f32.bf16 %r1147, %rs424; + cvt.f32.bf16 %r1148, %rs425; + cvt.f32.bf16 %r1149, %rs426; + cvt.f32.bf16 %r1150, %rs427; + cvt.f32.bf16 %r1151, %rs428; + cvt.f32.bf16 %r1152, %rs429; + cvt.f32.bf16 %r1153, %rs430; + cvt.f32.bf16 %r1154, %rs431; + cvt.f32.bf16 %r1155, %rs432; + cvt.f32.bf16 %r1156, %rs433; + cvt.f32.bf16 %r1157, %rs434; + cvt.f32.bf16 %r1158, %rs435; + cvt.f32.bf16 %r1159, %rs436; + cvt.f32.bf16 %r1160, %rs437; + cvt.f32.bf16 %r1161, %rs438; + cvt.f32.bf16 %r1162, %rs439; + cvt.f32.bf16 %r1163, %rs440; + cvt.f32.bf16 %r1164, %rs441; + cvt.f32.bf16 %r1165, %rs442; + cvt.f32.bf16 %r1166, %rs443; + cvt.f32.bf16 %r1167, %rs444; + cvt.f32.bf16 %r1168, %rs445; + cvt.f32.bf16 %r1169, %rs446; + cvt.f32.bf16 %r1170, %rs447; + cvt.f32.bf16 %r1171, %rs448; + cvt.f32.bf16 %r1172, %rs449; + cvt.f32.bf16 %r1173, %rs450; + cvt.f32.bf16 %r1174, %rs451; + cvt.f32.bf16 %r1175, %rs452; + cvt.f32.bf16 %r1176, %rs453; + cvt.f32.bf16 %r1177, %rs454; + cvt.f32.bf16 %r1178, %rs455; + cvt.f32.bf16 %r1179, %rs456; + cvt.f32.bf16 %r1180, %rs457; + cvt.f32.bf16 %r1181, %rs458; + cvt.f32.bf16 %r1182, %rs459; + cvt.f32.bf16 %r1183, %rs460; + cvt.f32.bf16 %r1184, %rs461; + cvt.f32.bf16 %r1185, %rs462; + cvt.f32.bf16 %r1186, %rs463; + cvt.f32.bf16 %r1187, %rs464; + cvt.f32.bf16 %r1188, %rs465; + cvt.f32.bf16 %r1189, %rs466; + cvt.f32.bf16 %r1190, %rs467; + cvt.f32.bf16 %r1191, %rs468; + cvt.f32.bf16 %r1192, %rs469; + cvt.f32.bf16 %r1193, %rs470; + cvt.f32.bf16 %r1194, %rs471; + cvt.f32.bf16 %r1195, %rs472; + cvt.f32.bf16 %r1196, %rs473; + cvt.f32.bf16 %r1197, %rs474; + cvt.f32.bf16 %r1198, %rs475; + cvt.f32.bf16 %r1199, %rs476; + cvt.f32.bf16 %r1200, %rs477; + cvt.f32.bf16 %r1201, %rs478; + cvt.f32.bf16 %r1202, %rs479; + cvt.f32.bf16 %r1203, %rs480; + cvt.f32.bf16 %r1204, %rs481; + cvt.f32.bf16 %r1205, %rs482; + cvt.f32.bf16 %r1206, %rs483; + cvt.f32.bf16 %r1207, %rs484; + cvt.f32.bf16 %r1208, %rs485; + cvt.f32.bf16 %r1209, %rs486; + cvt.f32.bf16 %r1210, %rs487; + cvt.f32.bf16 %r1211, %rs488; + cvt.f32.bf16 %r1212, %rs489; + cvt.f32.bf16 %r1213, %rs490; + cvt.f32.bf16 %r1214, %rs491; + cvt.f32.bf16 %r1215, %rs492; + cvt.f32.bf16 %r1216, %rs493; + cvt.f32.bf16 %r1217, %rs494; + cvt.f32.bf16 %r1218, %rs495; + cvt.f32.bf16 %r1219, %rs496; + cvt.f32.bf16 %r1220, %rs497; + cvt.f32.bf16 %r1221, %rs498; + cvt.f32.bf16 %r1222, %rs499; + cvt.f32.bf16 %r1223, %rs500; + cvt.f32.bf16 %r1224, %rs501; + cvt.f32.bf16 %r1225, %rs502; + cvt.f32.bf16 %r1226, %rs503; + cvt.f32.bf16 %r1227, %rs504; + cvt.f32.bf16 %r1228, %rs505; + cvt.f32.bf16 %r1229, %rs506; + cvt.f32.bf16 %r1230, %rs507; + cvt.f32.bf16 %r1231, %rs508; + cvt.f32.bf16 %r1232, %rs509; + cvt.f32.bf16 %r1233, %rs510; + cvt.f32.bf16 %r1234, %rs511; + cvt.f32.bf16 %r1235, %rs512; + cvt.f32.bf16 %r1236, %rs513; + cvt.f32.bf16 %r1237, %rs514; + cvt.f32.bf16 %r1238, %rs515; + cvt.f32.bf16 %r1239, %rs516; + cvt.f32.bf16 %r1240, %rs517; + cvt.f32.bf16 %r1241, %rs518; + cvt.f32.bf16 %r1242, %rs519; + cvt.f32.bf16 %r1243, %rs520; + cvt.f32.bf16 %r1244, %rs521; + cvt.f32.bf16 %r1245, %rs522; + cvt.f32.bf16 %r1246, %rs523; + cvt.f32.bf16 %r1247, %rs524; + cvt.f32.bf16 %r1248, %rs525; + cvt.f32.bf16 %r1249, %rs526; + cvt.f32.bf16 %r1250, %rs527; + cvt.f32.bf16 %r1251, %rs528; + cvt.f32.bf16 %r1252, %rs529; + cvt.f32.bf16 %r1253, %rs530; + cvt.f32.bf16 %r1254, %rs531; + cvt.f32.bf16 %r1255, %rs532; + cvt.f32.bf16 %r1256, %rs533; + cvt.f32.bf16 %r1257, %rs534; + cvt.f32.bf16 %r1258, %rs535; + cvt.f32.bf16 %r1259, %rs536; + cvt.f32.bf16 %r1260, %rs537; + cvt.f32.bf16 %r1261, %rs538; + cvt.f32.bf16 %r1262, %rs539; + cvt.f32.bf16 %r1263, %rs540; + cvt.f32.bf16 %r1264, %rs541; + cvt.f32.bf16 %r1265, %rs542; + cvt.f32.bf16 %r1266, %rs543; + cvt.f32.bf16 %r1267, %rs544; + cvt.f32.bf16 %r1268, %rs545; + cvt.f32.bf16 %r1269, %rs546; + cvt.f32.bf16 %r1270, %rs547; + cvt.f32.bf16 %r1271, %rs548; + cvt.f32.bf16 %r1272, %rs549; + cvt.f32.bf16 %r1273, %rs550; + cvt.f32.bf16 %r1274, %rs551; + cvt.f32.bf16 %r1275, %rs552; + cvt.f32.bf16 %r1276, %rs553; + cvt.f32.bf16 %r1277, %rs554; + cvt.f32.bf16 %r1278, %rs555; + cvt.f32.bf16 %r1279, %rs556; + cvt.f32.bf16 %r1280, %rs557; + cvt.f32.bf16 %r1281, %rs558; + cvt.f32.bf16 %r1282, %rs559; + cvt.f32.bf16 %r1283, %rs560; + cvt.f32.bf16 %r1284, %rs561; + cvt.f32.bf16 %r1285, %rs562; + cvt.f32.bf16 %r1286, %rs563; + cvt.f32.bf16 %r1287, %rs564; + cvt.f32.bf16 %r1288, %rs565; + cvt.f32.bf16 %r1289, %rs566; + cvt.f32.bf16 %r1290, %rs567; + cvt.f32.bf16 %r1291, %rs568; + cvt.f32.bf16 %r1292, %rs569; + cvt.f32.bf16 %r1293, %rs570; + cvt.f32.bf16 %r1294, %rs571; + cvt.f32.bf16 %r1295, %rs572; + cvt.f32.bf16 %r1296, %rs573; + cvt.f32.bf16 %r1297, %rs574; + cvt.f32.bf16 %r1298, %rs575; + cvt.f32.bf16 %r1299, %rs576; + cvt.f32.bf16 %r1300, %rs577; + cvt.f32.bf16 %r1301, %rs578; + cvt.f32.bf16 %r1302, %rs579; + cvt.f32.bf16 %r1303, %rs580; + cvt.f32.bf16 %r1304, %rs581; + cvt.f32.bf16 %r1305, %rs582; + cvt.f32.bf16 %r1306, %rs583; + cvt.f32.bf16 %r1307, %rs584; + cvt.f32.bf16 %r1308, %rs585; + cvt.f32.bf16 %r1309, %rs586; + cvt.f32.bf16 %r1310, %rs587; + cvt.f32.bf16 %r1311, %rs588; + cvt.f32.bf16 %r1312, %rs589; + cvt.f32.bf16 %r1313, %rs590; + cvt.f32.bf16 %r1314, %rs591; + cvt.f32.bf16 %r1315, %rs592; + cvt.f32.bf16 %r1316, %rs593; + cvt.f32.bf16 %r1317, %rs594; + cvt.f32.bf16 %r1318, %rs595; + cvt.f32.bf16 %r1319, %rs596; + cvt.f32.bf16 %r1320, %rs597; + cvt.f32.bf16 %r1321, %rs598; + cvt.f32.bf16 %r1322, %rs599; + cvt.f32.bf16 %r1323, %rs600; + cvt.f32.bf16 %r1324, %rs601; + cvt.f32.bf16 %r1325, %rs602; + cvt.f32.bf16 %r1326, %rs603; + cvt.f32.bf16 %r1327, %rs604; + cvt.f32.bf16 %r1328, %rs605; + cvt.f32.bf16 %r1329, %rs606; + cvt.f32.bf16 %r1330, %rs607; + cvt.f32.bf16 %r1331, %rs608; + cvt.f32.bf16 %r1332, %rs609; + cvt.f32.bf16 %r1333, %rs610; + cvt.f32.bf16 %r1334, %rs611; + cvt.f32.bf16 %r1335, %rs612; + cvt.f32.bf16 %r1336, %rs613; + cvt.f32.bf16 %r1337, %rs614; + cvt.f32.bf16 %r1338, %rs615; + cvt.f32.bf16 %r1339, %rs616; + cvt.f32.bf16 %r1340, %rs617; + cvt.f32.bf16 %r1341, %rs618; + cvt.f32.bf16 %r1342, %rs619; + cvt.f32.bf16 %r1343, %rs620; + cvt.f32.bf16 %r1344, %rs621; + cvt.f32.bf16 %r1345, %rs622; + cvt.f32.bf16 %r1346, %rs623; + cvt.f32.bf16 %r1347, %rs624; + cvt.f32.bf16 %r1348, %rs625; + cvt.f32.bf16 %r1349, %rs626; + cvt.f32.bf16 %r1350, %rs627; + cvt.f32.bf16 %r1351, %rs628; + cvt.f32.bf16 %r1352, %rs629; + cvt.f32.bf16 %r1353, %rs630; + cvt.f32.bf16 %r1354, %rs631; + cvt.f32.bf16 %r1355, %rs632; + cvt.f32.bf16 %r1356, %rs633; + cvt.f32.bf16 %r1357, %rs634; + cvt.f32.bf16 %r1358, %rs635; + cvt.f32.bf16 %r1359, %rs636; + cvt.f32.bf16 %r1360, %rs637; + cvt.f32.bf16 %r1361, %rs638; + cvt.f32.bf16 %r1362, %rs639; + cvt.f32.bf16 %r1363, %rs640; + cvt.f32.bf16 %r1364, %rs641; + cvt.f32.bf16 %r1365, %rs642; + cvt.f32.bf16 %r1366, %rs643; + cvt.f32.bf16 %r1367, %rs644; + cvt.f32.bf16 %r1368, %rs645; + cvt.f32.bf16 %r1369, %rs646; + cvt.f32.bf16 %r1370, %rs647; + cvt.f32.bf16 %r1371, %rs648; + cvt.f32.bf16 %r1372, %rs649; + cvt.f32.bf16 %r1373, %rs650; + cvt.f32.bf16 %r1374, %rs651; + cvt.f32.bf16 %r1375, %rs652; + cvt.f32.bf16 %r1376, %rs653; + cvt.f32.bf16 %r1377, %rs654; + cvt.f32.bf16 %r1378, %rs655; + cvt.f32.bf16 %r1379, %rs656; + cvt.f32.bf16 %r1380, %rs657; + cvt.f32.bf16 %r1381, %rs658; + cvt.f32.bf16 %r1382, %rs659; + cvt.f32.bf16 %r1383, %rs660; + cvt.f32.bf16 %r1384, %rs661; + cvt.f32.bf16 %r1385, %rs662; + cvt.f32.bf16 %r1386, %rs663; + cvt.f32.bf16 %r1387, %rs664; + cvt.f32.bf16 %r1388, %rs665; + cvt.f32.bf16 %r1389, %rs666; + cvt.f32.bf16 %r1390, %rs667; + cvt.f32.bf16 %r1391, %rs668; + cvt.f32.bf16 %r1392, %rs669; + cvt.f32.bf16 %r1393, %rs670; + cvt.f32.bf16 %r1394, %rs671; + .loc 1 105 21 // l2norm.py:105:21 + shr.u64 %rd657, %rd339, 3; + and.b64 %rd658, %rd657, 1; + setp.ne.b64 %p35, %rd658, 0; + // begin inline asm + mov.u32 %r258, %r5; + @%p35 ld.global.b32 { %r258 }, [ %rd65 + 0 ]; + // end inline asm + shr.u64 %rd659, %rd339, 2; + and.b64 %rd660, %rd659, 1; + setp.ne.b64 %p36, %rd660, 0; + // begin inline asm + mov.u32 %r259, %r5; + @%p36 ld.global.b32 { %r259 }, [ %rd66 + 0 ]; + // end inline asm + shr.u64 %rd661, %rd339, 1; + and.b64 %rd662, %rd661, 1; + setp.ne.b64 %p37, %rd662, 0; + // begin inline asm + mov.u32 %r260, %r5; + @%p37 ld.global.b32 { %r260 }, [ %rd67 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r261, %r5; + @%p38 ld.global.b32 { %r261 }, [ %rd68 + 0 ]; + // end inline asm + .loc 1 106 19 // l2norm.py:106:19 + // begin inline asm + mov.u32 %r262, %r5; + mov.u32 %r263, %r5; + mov.u32 %r264, %r5; + mov.u32 %r265, %r5; + @%p1 ld.global.v4.b32 { %r262, %r263, %r264, %r265 }, [ %rd69 + 0 ]; + // end inline asm + mov.b32 {%rs672, %rs673}, %r262; + mov.b32 {%rs674, %rs675}, %r263; + mov.b32 {%rs676, %rs677}, %r264; + mov.b32 {%rs678, %rs679}, %r265; + // begin inline asm + mov.u32 %r266, %r5; + mov.u32 %r267, %r5; + mov.u32 %r268, %r5; + mov.u32 %r269, %r5; + @%p2 ld.global.v4.b32 { %r266, %r267, %r268, %r269 }, [ %rd70 + 0 ]; + // end inline asm + mov.b32 {%rs680, %rs681}, %r266; + mov.b32 {%rs682, %rs683}, %r267; + mov.b32 {%rs684, %rs685}, %r268; + mov.b32 {%rs686, %rs687}, %r269; + // begin inline asm + mov.u32 %r270, %r5; + mov.u32 %r271, %r5; + mov.u32 %r272, %r5; + mov.u32 %r273, %r5; + @%p3 ld.global.v4.b32 { %r270, %r271, %r272, %r273 }, [ %rd71 + 0 ]; + // end inline asm + mov.b32 {%rs688, %rs689}, %r270; + mov.b32 {%rs690, %rs691}, %r271; + mov.b32 {%rs692, %rs693}, %r272; + mov.b32 {%rs694, %rs695}, %r273; + // begin inline asm + mov.u32 %r274, %r5; + mov.u32 %r275, %r5; + mov.u32 %r276, %r5; + mov.u32 %r277, %r5; + @%p4 ld.global.v4.b32 { %r274, %r275, %r276, %r277 }, [ %rd72 + 0 ]; + // end inline asm + mov.b32 {%rs696, %rs697}, %r274; + mov.b32 {%rs698, %rs699}, %r275; + mov.b32 {%rs700, %rs701}, %r276; + mov.b32 {%rs702, %rs703}, %r277; + // begin inline asm + mov.u32 %r278, %r5; + mov.u32 %r279, %r5; + mov.u32 %r280, %r5; + mov.u32 %r281, %r5; + @%p5 ld.global.v4.b32 { %r278, %r279, %r280, %r281 }, [ %rd73 + 0 ]; + // end inline asm + mov.b32 {%rs704, %rs705}, %r278; + mov.b32 {%rs706, %rs707}, %r279; + mov.b32 {%rs708, %rs709}, %r280; + mov.b32 {%rs710, %rs711}, %r281; + // begin inline asm + mov.u32 %r282, %r5; + mov.u32 %r283, %r5; + mov.u32 %r284, %r5; + mov.u32 %r285, %r5; + @%p6 ld.global.v4.b32 { %r282, %r283, %r284, %r285 }, [ %rd74 + 0 ]; + // end inline asm + mov.b32 {%rs712, %rs713}, %r282; + mov.b32 {%rs714, %rs715}, %r283; + mov.b32 {%rs716, %rs717}, %r284; + mov.b32 {%rs718, %rs719}, %r285; + // begin inline asm + mov.u32 %r286, %r5; + mov.u32 %r287, %r5; + mov.u32 %r288, %r5; + mov.u32 %r289, %r5; + @%p7 ld.global.v4.b32 { %r286, %r287, %r288, %r289 }, [ %rd75 + 0 ]; + // end inline asm + mov.b32 {%rs720, %rs721}, %r286; + mov.b32 {%rs722, %rs723}, %r287; + mov.b32 {%rs724, %rs725}, %r288; + mov.b32 {%rs726, %rs727}, %r289; + // begin inline asm + mov.u32 %r290, %r5; + mov.u32 %r291, %r5; + mov.u32 %r292, %r5; + mov.u32 %r293, %r5; + @%p8 ld.global.v4.b32 { %r290, %r291, %r292, %r293 }, [ %rd76 + 0 ]; + // end inline asm + mov.b32 {%rs728, %rs729}, %r290; + mov.b32 {%rs730, %rs731}, %r291; + mov.b32 {%rs732, %rs733}, %r292; + mov.b32 {%rs734, %rs735}, %r293; + // begin inline asm + mov.u32 %r294, %r5; + mov.u32 %r295, %r5; + mov.u32 %r296, %r5; + mov.u32 %r297, %r5; + @%p9 ld.global.v4.b32 { %r294, %r295, %r296, %r297 }, [ %rd77 + 0 ]; + // end inline asm + mov.b32 {%rs736, %rs737}, %r294; + mov.b32 {%rs738, %rs739}, %r295; + mov.b32 {%rs740, %rs741}, %r296; + mov.b32 {%rs742, %rs743}, %r297; + // begin inline asm + mov.u32 %r298, %r5; + mov.u32 %r299, %r5; + mov.u32 %r300, %r5; + mov.u32 %r301, %r5; + @%p10 ld.global.v4.b32 { %r298, %r299, %r300, %r301 }, [ %rd78 + 0 ]; + // end inline asm + mov.b32 {%rs744, %rs745}, %r298; + mov.b32 {%rs746, %rs747}, %r299; + mov.b32 {%rs748, %rs749}, %r300; + mov.b32 {%rs750, %rs751}, %r301; + // begin inline asm + mov.u32 %r302, %r5; + mov.u32 %r303, %r5; + mov.u32 %r304, %r5; + mov.u32 %r305, %r5; + @%p11 ld.global.v4.b32 { %r302, %r303, %r304, %r305 }, [ %rd79 + 0 ]; + // end inline asm + mov.b32 {%rs752, %rs753}, %r302; + mov.b32 {%rs754, %rs755}, %r303; + mov.b32 {%rs756, %rs757}, %r304; + mov.b32 {%rs758, %rs759}, %r305; + // begin inline asm + mov.u32 %r306, %r5; + mov.u32 %r307, %r5; + mov.u32 %r308, %r5; + mov.u32 %r309, %r5; + @%p39 ld.global.v4.b32 { %r306, %r307, %r308, %r309 }, [ %rd80 + 0 ]; + // end inline asm + mov.b32 {%rs760, %rs761}, %r306; + mov.b32 {%rs762, %rs763}, %r307; + mov.b32 {%rs764, %rs765}, %r308; + mov.b32 {%rs766, %rs767}, %r309; + // begin inline asm + mov.u32 %r310, %r5; + mov.u32 %r311, %r5; + mov.u32 %r312, %r5; + mov.u32 %r313, %r5; + @%p12 ld.global.v4.b32 { %r310, %r311, %r312, %r313 }, [ %rd81 + 0 ]; + // end inline asm + mov.b32 {%rs768, %rs769}, %r310; + mov.b32 {%rs770, %rs771}, %r311; + mov.b32 {%rs772, %rs773}, %r312; + mov.b32 {%rs774, %rs775}, %r313; + // begin inline asm + mov.u32 %r314, %r5; + mov.u32 %r315, %r5; + mov.u32 %r316, %r5; + mov.u32 %r317, %r5; + @%p13 ld.global.v4.b32 { %r314, %r315, %r316, %r317 }, [ %rd82 + 0 ]; + // end inline asm + mov.b32 {%rs776, %rs777}, %r314; + mov.b32 {%rs778, %rs779}, %r315; + mov.b32 {%rs780, %rs781}, %r316; + mov.b32 {%rs782, %rs783}, %r317; + // begin inline asm + mov.u32 %r318, %r5; + mov.u32 %r319, %r5; + mov.u32 %r320, %r5; + mov.u32 %r321, %r5; + @%p14 ld.global.v4.b32 { %r318, %r319, %r320, %r321 }, [ %rd83 + 0 ]; + // end inline asm + mov.b32 {%rs784, %rs785}, %r318; + mov.b32 {%rs786, %rs787}, %r319; + mov.b32 {%rs788, %rs789}, %r320; + mov.b32 {%rs790, %rs791}, %r321; + // begin inline asm + mov.u32 %r322, %r5; + mov.u32 %r323, %r5; + mov.u32 %r324, %r5; + mov.u32 %r325, %r5; + @%p15 ld.global.v4.b32 { %r322, %r323, %r324, %r325 }, [ %rd84 + 0 ]; + // end inline asm + mov.b32 {%rs792, %rs793}, %r322; + mov.b32 {%rs794, %rs795}, %r323; + mov.b32 {%rs796, %rs797}, %r324; + mov.b32 {%rs798, %rs799}, %r325; + // begin inline asm + mov.u32 %r326, %r5; + mov.u32 %r327, %r5; + mov.u32 %r328, %r5; + mov.u32 %r329, %r5; + @%p16 ld.global.v4.b32 { %r326, %r327, %r328, %r329 }, [ %rd85 + 0 ]; + // end inline asm + mov.b32 {%rs800, %rs801}, %r326; + mov.b32 {%rs802, %rs803}, %r327; + mov.b32 {%rs804, %rs805}, %r328; + mov.b32 {%rs806, %rs807}, %r329; + // begin inline asm + mov.u32 %r330, %r5; + mov.u32 %r331, %r5; + mov.u32 %r332, %r5; + mov.u32 %r333, %r5; + @%p17 ld.global.v4.b32 { %r330, %r331, %r332, %r333 }, [ %rd86 + 0 ]; + // end inline asm + mov.b32 {%rs808, %rs809}, %r330; + mov.b32 {%rs810, %rs811}, %r331; + mov.b32 {%rs812, %rs813}, %r332; + mov.b32 {%rs814, %rs815}, %r333; + // begin inline asm + mov.u32 %r334, %r5; + mov.u32 %r335, %r5; + mov.u32 %r336, %r5; + mov.u32 %r337, %r5; + @%p18 ld.global.v4.b32 { %r334, %r335, %r336, %r337 }, [ %rd87 + 0 ]; + // end inline asm + mov.b32 {%rs816, %rs817}, %r334; + mov.b32 {%rs818, %rs819}, %r335; + mov.b32 {%rs820, %rs821}, %r336; + mov.b32 {%rs822, %rs823}, %r337; + // begin inline asm + mov.u32 %r338, %r5; + mov.u32 %r339, %r5; + mov.u32 %r340, %r5; + mov.u32 %r341, %r5; + @%p40 ld.global.v4.b32 { %r338, %r339, %r340, %r341 }, [ %rd88 + 0 ]; + // end inline asm + mov.b32 {%rs824, %rs825}, %r338; + mov.b32 {%rs826, %rs827}, %r339; + mov.b32 {%rs828, %rs829}, %r340; + mov.b32 {%rs830, %rs831}, %r341; + // begin inline asm + mov.u32 %r342, %r5; + mov.u32 %r343, %r5; + mov.u32 %r344, %r5; + mov.u32 %r345, %r5; + @%p19 ld.global.v4.b32 { %r342, %r343, %r344, %r345 }, [ %rd89 + 0 ]; + // end inline asm + mov.b32 {%rs832, %rs833}, %r342; + mov.b32 {%rs834, %rs835}, %r343; + mov.b32 {%rs836, %rs837}, %r344; + mov.b32 {%rs838, %rs839}, %r345; + // begin inline asm + mov.u32 %r346, %r5; + mov.u32 %r347, %r5; + mov.u32 %r348, %r5; + mov.u32 %r349, %r5; + @%p20 ld.global.v4.b32 { %r346, %r347, %r348, %r349 }, [ %rd90 + 0 ]; + // end inline asm + mov.b32 {%rs840, %rs841}, %r346; + mov.b32 {%rs842, %rs843}, %r347; + mov.b32 {%rs844, %rs845}, %r348; + mov.b32 {%rs846, %rs847}, %r349; + // begin inline asm + mov.u32 %r350, %r5; + mov.u32 %r351, %r5; + mov.u32 %r352, %r5; + mov.u32 %r353, %r5; + @%p21 ld.global.v4.b32 { %r350, %r351, %r352, %r353 }, [ %rd91 + 0 ]; + // end inline asm + mov.b32 {%rs848, %rs849}, %r350; + mov.b32 {%rs850, %rs851}, %r351; + mov.b32 {%rs852, %rs853}, %r352; + mov.b32 {%rs854, %rs855}, %r353; + // begin inline asm + mov.u32 %r354, %r5; + mov.u32 %r355, %r5; + mov.u32 %r356, %r5; + mov.u32 %r357, %r5; + @%p22 ld.global.v4.b32 { %r354, %r355, %r356, %r357 }, [ %rd92 + 0 ]; + // end inline asm + mov.b32 {%rs856, %rs857}, %r354; + mov.b32 {%rs858, %rs859}, %r355; + mov.b32 {%rs860, %rs861}, %r356; + mov.b32 {%rs862, %rs863}, %r357; + // begin inline asm + mov.u32 %r358, %r5; + mov.u32 %r359, %r5; + mov.u32 %r360, %r5; + mov.u32 %r361, %r5; + @%p23 ld.global.v4.b32 { %r358, %r359, %r360, %r361 }, [ %rd93 + 0 ]; + // end inline asm + mov.b32 {%rs864, %rs865}, %r358; + mov.b32 {%rs866, %rs867}, %r359; + mov.b32 {%rs868, %rs869}, %r360; + mov.b32 {%rs870, %rs871}, %r361; + // begin inline asm + mov.u32 %r362, %r5; + mov.u32 %r363, %r5; + mov.u32 %r364, %r5; + mov.u32 %r365, %r5; + @%p24 ld.global.v4.b32 { %r362, %r363, %r364, %r365 }, [ %rd94 + 0 ]; + // end inline asm + mov.b32 {%rs872, %rs873}, %r362; + mov.b32 {%rs874, %rs875}, %r363; + mov.b32 {%rs876, %rs877}, %r364; + mov.b32 {%rs878, %rs879}, %r365; + // begin inline asm + mov.u32 %r366, %r5; + mov.u32 %r367, %r5; + mov.u32 %r368, %r5; + mov.u32 %r369, %r5; + @%p25 ld.global.v4.b32 { %r366, %r367, %r368, %r369 }, [ %rd95 + 0 ]; + // end inline asm + mov.b32 {%rs880, %rs881}, %r366; + mov.b32 {%rs882, %rs883}, %r367; + mov.b32 {%rs884, %rs885}, %r368; + mov.b32 {%rs886, %rs887}, %r369; + // begin inline asm + mov.u32 %r370, %r5; + mov.u32 %r371, %r5; + mov.u32 %r372, %r5; + mov.u32 %r373, %r5; + @%p26 ld.global.v4.b32 { %r370, %r371, %r372, %r373 }, [ %rd96 + 0 ]; + // end inline asm + mov.b32 {%rs888, %rs889}, %r370; + mov.b32 {%rs890, %rs891}, %r371; + mov.b32 {%rs892, %rs893}, %r372; + mov.b32 {%rs894, %rs895}, %r373; + // begin inline asm + mov.u32 %r374, %r5; + mov.u32 %r375, %r5; + mov.u32 %r376, %r5; + mov.u32 %r377, %r5; + @%p27 ld.global.v4.b32 { %r374, %r375, %r376, %r377 }, [ %rd97 + 0 ]; + // end inline asm + mov.b32 {%rs896, %rs897}, %r374; + mov.b32 {%rs898, %rs899}, %r375; + mov.b32 {%rs900, %rs901}, %r376; + mov.b32 {%rs902, %rs903}, %r377; + // begin inline asm + mov.u32 %r378, %r5; + mov.u32 %r379, %r5; + mov.u32 %r380, %r5; + mov.u32 %r381, %r5; + @%p28 ld.global.v4.b32 { %r378, %r379, %r380, %r381 }, [ %rd98 + 0 ]; + // end inline asm + mov.b32 {%rs904, %rs905}, %r378; + mov.b32 {%rs906, %rs907}, %r379; + mov.b32 {%rs908, %rs909}, %r380; + mov.b32 {%rs910, %rs911}, %r381; + // begin inline asm + mov.u32 %r382, %r5; + mov.u32 %r383, %r5; + mov.u32 %r384, %r5; + mov.u32 %r385, %r5; + @%p29 ld.global.v4.b32 { %r382, %r383, %r384, %r385 }, [ %rd99 + 0 ]; + // end inline asm + mov.b32 {%rs912, %rs913}, %r382; + mov.b32 {%rs914, %rs915}, %r383; + mov.b32 {%rs916, %rs917}, %r384; + mov.b32 {%rs918, %rs919}, %r385; + // begin inline asm + mov.u32 %r386, %r5; + mov.u32 %r387, %r5; + mov.u32 %r388, %r5; + mov.u32 %r389, %r5; + @%p30 ld.global.v4.b32 { %r386, %r387, %r388, %r389 }, [ %rd100 + 0 ]; + // end inline asm + mov.b32 {%rs920, %rs921}, %r386; + mov.b32 {%rs922, %rs923}, %r387; + mov.b32 {%rs924, %rs925}, %r388; + mov.b32 {%rs926, %rs927}, %r389; + // begin inline asm + mov.u32 %r390, %r5; + mov.u32 %r391, %r5; + mov.u32 %r392, %r5; + mov.u32 %r393, %r5; + @%p31 ld.global.v4.b32 { %r390, %r391, %r392, %r393 }, [ %rd101 + 0 ]; + // end inline asm + mov.b32 {%rs928, %rs929}, %r390; + mov.b32 {%rs930, %rs931}, %r391; + mov.b32 {%rs932, %rs933}, %r392; + mov.b32 {%rs934, %rs935}, %r393; + // begin inline asm + mov.u32 %r394, %r5; + mov.u32 %r395, %r5; + mov.u32 %r396, %r5; + mov.u32 %r397, %r5; + @%p32 ld.global.v4.b32 { %r394, %r395, %r396, %r397 }, [ %rd102 + 0 ]; + // end inline asm + mov.b32 {%rs936, %rs937}, %r394; + mov.b32 {%rs938, %rs939}, %r395; + mov.b32 {%rs940, %rs941}, %r396; + mov.b32 {%rs942, %rs943}, %r397; + // begin inline asm + mov.u32 %r398, %r5; + mov.u32 %r399, %r5; + mov.u32 %r400, %r5; + mov.u32 %r401, %r5; + @%p33 ld.global.v4.b32 { %r398, %r399, %r400, %r401 }, [ %rd103 + 0 ]; + // end inline asm + mov.b32 {%rs944, %rs945}, %r398; + mov.b32 {%rs946, %rs947}, %r399; + mov.b32 {%rs948, %rs949}, %r400; + mov.b32 {%rs950, %rs951}, %r401; + // begin inline asm + mov.u32 %r402, %r5; + mov.u32 %r403, %r5; + mov.u32 %r404, %r5; + mov.u32 %r405, %r5; + @%p34 ld.global.v4.b32 { %r402, %r403, %r404, %r405 }, [ %rd104 + 0 ]; + // end inline asm + mov.b32 {%rs952, %rs953}, %r402; + mov.b32 {%rs954, %rs955}, %r403; + mov.b32 {%rs956, %rs957}, %r404; + mov.b32 {%rs958, %rs959}, %r405; + // begin inline asm + mov.u32 %r406, %r5; + mov.u32 %r407, %r5; + mov.u32 %r408, %r5; + mov.u32 %r409, %r5; + @%p41 ld.global.v4.b32 { %r406, %r407, %r408, %r409 }, [ %rd105 + 0 ]; + // end inline asm + mov.b32 {%rs960, %rs961}, %r406; + mov.b32 {%rs962, %rs963}, %r407; + mov.b32 {%rs964, %rs965}, %r408; + mov.b32 {%rs966, %rs967}, %r409; + // begin inline asm + mov.u32 %r410, %r5; + mov.u32 %r411, %r5; + mov.u32 %r412, %r5; + mov.u32 %r413, %r5; + @%p42 ld.global.v4.b32 { %r410, %r411, %r412, %r413 }, [ %rd106 + 0 ]; + // end inline asm + mov.b32 {%rs968, %rs969}, %r410; + mov.b32 {%rs970, %rs971}, %r411; + mov.b32 {%rs972, %rs973}, %r412; + mov.b32 {%rs974, %rs975}, %r413; + // begin inline asm + mov.u32 %r414, %r5; + mov.u32 %r415, %r5; + mov.u32 %r416, %r5; + mov.u32 %r417, %r5; + @%p43 ld.global.v4.b32 { %r414, %r415, %r416, %r417 }, [ %rd107 + 0 ]; + // end inline asm + mov.b32 {%rs976, %rs977}, %r414; + mov.b32 {%rs978, %rs979}, %r415; + mov.b32 {%rs980, %rs981}, %r416; + mov.b32 {%rs982, %rs983}, %r417; + // begin inline asm + mov.u32 %r418, %r5; + mov.u32 %r419, %r5; + mov.u32 %r420, %r5; + mov.u32 %r421, %r5; + @%p44 ld.global.v4.b32 { %r418, %r419, %r420, %r421 }, [ %rd108 + 0 ]; + // end inline asm + mov.b32 {%rs984, %rs985}, %r418; + mov.b32 {%rs986, %rs987}, %r419; + mov.b32 {%rs988, %rs989}, %r420; + mov.b32 {%rs990, %rs991}, %r421; + // begin inline asm + mov.u32 %r422, %r5; + mov.u32 %r423, %r5; + mov.u32 %r424, %r5; + mov.u32 %r425, %r5; + @%p45 ld.global.v4.b32 { %r422, %r423, %r424, %r425 }, [ %rd109 + 0 ]; + // end inline asm + mov.b32 {%rs992, %rs993}, %r422; + mov.b32 {%rs994, %rs995}, %r423; + mov.b32 {%rs996, %rs997}, %r424; + mov.b32 {%rs998, %rs999}, %r425; + // begin inline asm + mov.u32 %r426, %r5; + mov.u32 %r427, %r5; + mov.u32 %r428, %r5; + mov.u32 %r429, %r5; + @%p46 ld.global.v4.b32 { %r426, %r427, %r428, %r429 }, [ %rd110 + 0 ]; + // end inline asm + mov.b32 {%rs1000, %rs1001}, %r426; + mov.b32 {%rs1002, %rs1003}, %r427; + mov.b32 {%rs1004, %rs1005}, %r428; + mov.b32 {%rs1006, %rs1007}, %r429; + // begin inline asm + mov.u32 %r430, %r5; + mov.u32 %r431, %r5; + mov.u32 %r432, %r5; + mov.u32 %r433, %r5; + @%p47 ld.global.v4.b32 { %r430, %r431, %r432, %r433 }, [ %rd111 + 0 ]; + // end inline asm + mov.b32 {%rs1008, %rs1009}, %r430; + mov.b32 {%rs1010, %rs1011}, %r431; + mov.b32 {%rs1012, %rs1013}, %r432; + mov.b32 {%rs1014, %rs1015}, %r433; + // begin inline asm + mov.u32 %r434, %r5; + mov.u32 %r435, %r5; + mov.u32 %r436, %r5; + mov.u32 %r437, %r5; + @%p48 ld.global.v4.b32 { %r434, %r435, %r436, %r437 }, [ %rd112 + 0 ]; + // end inline asm + mov.b32 {%rs1016, %rs1017}, %r434; + mov.b32 {%rs1018, %rs1019}, %r435; + mov.b32 {%rs1020, %rs1021}, %r436; + mov.b32 {%rs1022, %rs1023}, %r437; + // begin inline asm + mov.u32 %r438, %r5; + mov.u32 %r439, %r5; + mov.u32 %r440, %r5; + mov.u32 %r441, %r5; + @%p49 ld.global.v4.b32 { %r438, %r439, %r440, %r441 }, [ %rd113 + 0 ]; + // end inline asm + mov.b32 {%rs1024, %rs1025}, %r438; + mov.b32 {%rs1026, %rs1027}, %r439; + mov.b32 {%rs1028, %rs1029}, %r440; + mov.b32 {%rs1030, %rs1031}, %r441; + // begin inline asm + mov.u32 %r442, %r5; + mov.u32 %r443, %r5; + mov.u32 %r444, %r5; + mov.u32 %r445, %r5; + @%p50 ld.global.v4.b32 { %r442, %r443, %r444, %r445 }, [ %rd114 + 0 ]; + // end inline asm + mov.b32 {%rs1032, %rs1033}, %r442; + mov.b32 {%rs1034, %rs1035}, %r443; + mov.b32 {%rs1036, %rs1037}, %r444; + mov.b32 {%rs1038, %rs1039}, %r445; + // begin inline asm + mov.u32 %r446, %r5; + mov.u32 %r447, %r5; + mov.u32 %r448, %r5; + mov.u32 %r449, %r5; + @%p51 ld.global.v4.b32 { %r446, %r447, %r448, %r449 }, [ %rd115 + 0 ]; + // end inline asm + mov.b32 {%rs1040, %rs1041}, %r446; + mov.b32 {%rs1042, %rs1043}, %r447; + mov.b32 {%rs1044, %rs1045}, %r448; + mov.b32 {%rs1046, %rs1047}, %r449; + // begin inline asm + mov.u32 %r450, %r5; + mov.u32 %r451, %r5; + mov.u32 %r452, %r5; + mov.u32 %r453, %r5; + @%p52 ld.global.v4.b32 { %r450, %r451, %r452, %r453 }, [ %rd116 + 0 ]; + // end inline asm + mov.b32 {%rs1048, %rs1049}, %r450; + mov.b32 {%rs1050, %rs1051}, %r451; + mov.b32 {%rs1052, %rs1053}, %r452; + mov.b32 {%rs1054, %rs1055}, %r453; + // begin inline asm + mov.u32 %r454, %r5; + mov.u32 %r455, %r5; + mov.u32 %r456, %r5; + mov.u32 %r457, %r5; + @%p53 ld.global.v4.b32 { %r454, %r455, %r456, %r457 }, [ %rd117 + 0 ]; + // end inline asm + mov.b32 {%rs1056, %rs1057}, %r454; + mov.b32 {%rs1058, %rs1059}, %r455; + mov.b32 {%rs1060, %rs1061}, %r456; + mov.b32 {%rs1062, %rs1063}, %r457; + // begin inline asm + mov.u32 %r458, %r5; + mov.u32 %r459, %r5; + mov.u32 %r460, %r5; + mov.u32 %r461, %r5; + @%p54 ld.global.v4.b32 { %r458, %r459, %r460, %r461 }, [ %rd118 + 0 ]; + // end inline asm + mov.b32 {%rs1064, %rs1065}, %r458; + mov.b32 {%rs1066, %rs1067}, %r459; + mov.b32 {%rs1068, %rs1069}, %r460; + mov.b32 {%rs1070, %rs1071}, %r461; + // begin inline asm + mov.u32 %r462, %r5; + mov.u32 %r463, %r5; + mov.u32 %r464, %r5; + mov.u32 %r465, %r5; + @%p55 ld.global.v4.b32 { %r462, %r463, %r464, %r465 }, [ %rd119 + 0 ]; + // end inline asm + mov.b32 {%rs1072, %rs1073}, %r462; + mov.b32 {%rs1074, %rs1075}, %r463; + mov.b32 {%rs1076, %rs1077}, %r464; + mov.b32 {%rs1078, %rs1079}, %r465; + // begin inline asm + mov.u32 %r466, %r5; + mov.u32 %r467, %r5; + mov.u32 %r468, %r5; + mov.u32 %r469, %r5; + @%p56 ld.global.v4.b32 { %r466, %r467, %r468, %r469 }, [ %rd120 + 0 ]; + // end inline asm + mov.b32 {%rs1080, %rs1081}, %r466; + mov.b32 {%rs1082, %rs1083}, %r467; + mov.b32 {%rs1084, %rs1085}, %r468; + mov.b32 {%rs1086, %rs1087}, %r469; + // begin inline asm + mov.u32 %r470, %r5; + mov.u32 %r471, %r5; + mov.u32 %r472, %r5; + mov.u32 %r473, %r5; + @%p57 ld.global.v4.b32 { %r470, %r471, %r472, %r473 }, [ %rd121 + 0 ]; + // end inline asm + mov.b32 {%rs1088, %rs1089}, %r470; + mov.b32 {%rs1090, %rs1091}, %r471; + mov.b32 {%rs1092, %rs1093}, %r472; + mov.b32 {%rs1094, %rs1095}, %r473; + // begin inline asm + mov.u32 %r474, %r5; + mov.u32 %r475, %r5; + mov.u32 %r476, %r5; + mov.u32 %r477, %r5; + @%p58 ld.global.v4.b32 { %r474, %r475, %r476, %r477 }, [ %rd122 + 0 ]; + // end inline asm + mov.b32 {%rs1096, %rs1097}, %r474; + mov.b32 {%rs1098, %rs1099}, %r475; + mov.b32 {%rs1100, %rs1101}, %r476; + mov.b32 {%rs1102, %rs1103}, %r477; + // begin inline asm + mov.u32 %r478, %r5; + mov.u32 %r479, %r5; + mov.u32 %r480, %r5; + mov.u32 %r481, %r5; + @%p59 ld.global.v4.b32 { %r478, %r479, %r480, %r481 }, [ %rd123 + 0 ]; + // end inline asm + mov.b32 {%rs1104, %rs1105}, %r478; + mov.b32 {%rs1106, %rs1107}, %r479; + mov.b32 {%rs1108, %rs1109}, %r480; + mov.b32 {%rs1110, %rs1111}, %r481; + // begin inline asm + mov.u32 %r482, %r5; + mov.u32 %r483, %r5; + mov.u32 %r484, %r5; + mov.u32 %r485, %r5; + @%p60 ld.global.v4.b32 { %r482, %r483, %r484, %r485 }, [ %rd124 + 0 ]; + // end inline asm + mov.b32 {%rs1112, %rs1113}, %r482; + mov.b32 {%rs1114, %rs1115}, %r483; + mov.b32 {%rs1116, %rs1117}, %r484; + mov.b32 {%rs1118, %rs1119}, %r485; + // begin inline asm + mov.u32 %r486, %r5; + mov.u32 %r487, %r5; + mov.u32 %r488, %r5; + mov.u32 %r489, %r5; + @%p61 ld.global.v4.b32 { %r486, %r487, %r488, %r489 }, [ %rd125 + 0 ]; + // end inline asm + mov.b32 {%rs1120, %rs1121}, %r486; + mov.b32 {%rs1122, %rs1123}, %r487; + mov.b32 {%rs1124, %rs1125}, %r488; + mov.b32 {%rs1126, %rs1127}, %r489; + // begin inline asm + mov.u32 %r490, %r5; + mov.u32 %r491, %r5; + mov.u32 %r492, %r5; + mov.u32 %r493, %r5; + @%p62 ld.global.v4.b32 { %r490, %r491, %r492, %r493 }, [ %rd126 + 0 ]; + // end inline asm + mov.b32 {%rs1128, %rs1129}, %r490; + mov.b32 {%rs1130, %rs1131}, %r491; + mov.b32 {%rs1132, %rs1133}, %r492; + mov.b32 {%rs1134, %rs1135}, %r493; + // begin inline asm + mov.u32 %r494, %r5; + mov.u32 %r495, %r5; + mov.u32 %r496, %r5; + mov.u32 %r497, %r5; + @%p63 ld.global.v4.b32 { %r494, %r495, %r496, %r497 }, [ %rd127 + 0 ]; + // end inline asm + mov.b32 {%rs1136, %rs1137}, %r494; + mov.b32 {%rs1138, %rs1139}, %r495; + mov.b32 {%rs1140, %rs1141}, %r496; + mov.b32 {%rs1142, %rs1143}, %r497; + // begin inline asm + mov.u32 %r498, %r5; + mov.u32 %r499, %r5; + mov.u32 %r500, %r5; + mov.u32 %r501, %r5; + @%p64 ld.global.v4.b32 { %r498, %r499, %r500, %r501 }, [ %rd128 + 0 ]; + // end inline asm + mov.b32 {%rs1144, %rs1145}, %r498; + mov.b32 {%rs1146, %rs1147}, %r499; + mov.b32 {%rs1148, %rs1149}, %r500; + mov.b32 {%rs1150, %rs1151}, %r501; + // begin inline asm + mov.u32 %r502, %r5; + mov.u32 %r503, %r5; + mov.u32 %r504, %r5; + mov.u32 %r505, %r5; + @%p65 ld.global.v4.b32 { %r502, %r503, %r504, %r505 }, [ %rd129 + 0 ]; + // end inline asm + mov.b32 {%rs1152, %rs1153}, %r502; + mov.b32 {%rs1154, %rs1155}, %r503; + mov.b32 {%rs1156, %rs1157}, %r504; + mov.b32 {%rs1158, %rs1159}, %r505; + // begin inline asm + mov.u32 %r506, %r5; + mov.u32 %r507, %r5; + mov.u32 %r508, %r5; + mov.u32 %r509, %r5; + @%p66 ld.global.v4.b32 { %r506, %r507, %r508, %r509 }, [ %rd130 + 0 ]; + // end inline asm + mov.b32 {%rs1160, %rs1161}, %r506; + mov.b32 {%rs1162, %rs1163}, %r507; + mov.b32 {%rs1164, %rs1165}, %r508; + mov.b32 {%rs1166, %rs1167}, %r509; + // begin inline asm + mov.u32 %r510, %r5; + mov.u32 %r511, %r5; + mov.u32 %r512, %r5; + mov.u32 %r513, %r5; + @%p67 ld.global.v4.b32 { %r510, %r511, %r512, %r513 }, [ %rd131 + 0 ]; + // end inline asm + mov.b32 {%rs1168, %rs1169}, %r510; + mov.b32 {%rs1170, %rs1171}, %r511; + mov.b32 {%rs1172, %rs1173}, %r512; + mov.b32 {%rs1174, %rs1175}, %r513; + // begin inline asm + mov.u32 %r514, %r5; + mov.u32 %r515, %r5; + mov.u32 %r516, %r5; + mov.u32 %r517, %r5; + @%p68 ld.global.v4.b32 { %r514, %r515, %r516, %r517 }, [ %rd132 + 0 ]; + // end inline asm + mov.b32 {%rs1176, %rs1177}, %r514; + mov.b32 {%rs1178, %rs1179}, %r515; + mov.b32 {%rs1180, %rs1181}, %r516; + mov.b32 {%rs1182, %rs1183}, %r517; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1395, %rs672; + cvt.f32.bf16 %r1396, %rs673; + cvt.f32.bf16 %r1397, %rs674; + cvt.f32.bf16 %r1398, %rs675; + cvt.f32.bf16 %r1399, %rs676; + cvt.f32.bf16 %r1400, %rs677; + cvt.f32.bf16 %r1401, %rs678; + cvt.f32.bf16 %r1402, %rs679; + cvt.f32.bf16 %r1403, %rs680; + cvt.f32.bf16 %r1404, %rs681; + cvt.f32.bf16 %r1405, %rs682; + cvt.f32.bf16 %r1406, %rs683; + cvt.f32.bf16 %r1407, %rs684; + cvt.f32.bf16 %r1408, %rs685; + cvt.f32.bf16 %r1409, %rs686; + cvt.f32.bf16 %r1410, %rs687; + cvt.f32.bf16 %r1411, %rs688; + cvt.f32.bf16 %r1412, %rs689; + cvt.f32.bf16 %r1413, %rs690; + cvt.f32.bf16 %r1414, %rs691; + cvt.f32.bf16 %r1415, %rs692; + cvt.f32.bf16 %r1416, %rs693; + cvt.f32.bf16 %r1417, %rs694; + cvt.f32.bf16 %r1418, %rs695; + cvt.f32.bf16 %r1419, %rs696; + cvt.f32.bf16 %r1420, %rs697; + cvt.f32.bf16 %r1421, %rs698; + cvt.f32.bf16 %r1422, %rs699; + cvt.f32.bf16 %r1423, %rs700; + cvt.f32.bf16 %r1424, %rs701; + cvt.f32.bf16 %r1425, %rs702; + cvt.f32.bf16 %r1426, %rs703; + cvt.f32.bf16 %r1427, %rs704; + cvt.f32.bf16 %r1428, %rs705; + cvt.f32.bf16 %r1429, %rs706; + cvt.f32.bf16 %r1430, %rs707; + cvt.f32.bf16 %r1431, %rs708; + cvt.f32.bf16 %r1432, %rs709; + cvt.f32.bf16 %r1433, %rs710; + cvt.f32.bf16 %r1434, %rs711; + cvt.f32.bf16 %r1435, %rs712; + cvt.f32.bf16 %r1436, %rs713; + cvt.f32.bf16 %r1437, %rs714; + cvt.f32.bf16 %r1438, %rs715; + cvt.f32.bf16 %r1439, %rs716; + cvt.f32.bf16 %r1440, %rs717; + cvt.f32.bf16 %r1441, %rs718; + cvt.f32.bf16 %r1442, %rs719; + cvt.f32.bf16 %r1443, %rs720; + cvt.f32.bf16 %r1444, %rs721; + cvt.f32.bf16 %r1445, %rs722; + cvt.f32.bf16 %r1446, %rs723; + cvt.f32.bf16 %r1447, %rs724; + cvt.f32.bf16 %r1448, %rs725; + cvt.f32.bf16 %r1449, %rs726; + cvt.f32.bf16 %r1450, %rs727; + cvt.f32.bf16 %r1451, %rs728; + cvt.f32.bf16 %r1452, %rs729; + cvt.f32.bf16 %r1453, %rs730; + cvt.f32.bf16 %r1454, %rs731; + cvt.f32.bf16 %r1455, %rs732; + cvt.f32.bf16 %r1456, %rs733; + cvt.f32.bf16 %r1457, %rs734; + cvt.f32.bf16 %r1458, %rs735; + cvt.f32.bf16 %r1459, %rs736; + cvt.f32.bf16 %r1460, %rs737; + cvt.f32.bf16 %r1461, %rs738; + cvt.f32.bf16 %r1462, %rs739; + cvt.f32.bf16 %r1463, %rs740; + cvt.f32.bf16 %r1464, %rs741; + cvt.f32.bf16 %r1465, %rs742; + cvt.f32.bf16 %r1466, %rs743; + cvt.f32.bf16 %r1467, %rs744; + cvt.f32.bf16 %r1468, %rs745; + cvt.f32.bf16 %r1469, %rs746; + cvt.f32.bf16 %r1470, %rs747; + cvt.f32.bf16 %r1471, %rs748; + cvt.f32.bf16 %r1472, %rs749; + cvt.f32.bf16 %r1473, %rs750; + cvt.f32.bf16 %r1474, %rs751; + cvt.f32.bf16 %r1475, %rs752; + cvt.f32.bf16 %r1476, %rs753; + cvt.f32.bf16 %r1477, %rs754; + cvt.f32.bf16 %r1478, %rs755; + cvt.f32.bf16 %r1479, %rs756; + cvt.f32.bf16 %r1480, %rs757; + cvt.f32.bf16 %r1481, %rs758; + cvt.f32.bf16 %r1482, %rs759; + cvt.f32.bf16 %r1483, %rs760; + cvt.f32.bf16 %r1484, %rs761; + cvt.f32.bf16 %r1485, %rs762; + cvt.f32.bf16 %r1486, %rs763; + cvt.f32.bf16 %r1487, %rs764; + cvt.f32.bf16 %r1488, %rs765; + cvt.f32.bf16 %r1489, %rs766; + cvt.f32.bf16 %r1490, %rs767; + cvt.f32.bf16 %r1491, %rs768; + cvt.f32.bf16 %r1492, %rs769; + cvt.f32.bf16 %r1493, %rs770; + cvt.f32.bf16 %r1494, %rs771; + cvt.f32.bf16 %r1495, %rs772; + cvt.f32.bf16 %r1496, %rs773; + cvt.f32.bf16 %r1497, %rs774; + cvt.f32.bf16 %r1498, %rs775; + cvt.f32.bf16 %r1499, %rs776; + cvt.f32.bf16 %r1500, %rs777; + cvt.f32.bf16 %r1501, %rs778; + cvt.f32.bf16 %r1502, %rs779; + cvt.f32.bf16 %r1503, %rs780; + cvt.f32.bf16 %r1504, %rs781; + cvt.f32.bf16 %r1505, %rs782; + cvt.f32.bf16 %r1506, %rs783; + cvt.f32.bf16 %r1507, %rs784; + cvt.f32.bf16 %r1508, %rs785; + cvt.f32.bf16 %r1509, %rs786; + cvt.f32.bf16 %r1510, %rs787; + cvt.f32.bf16 %r1511, %rs788; + cvt.f32.bf16 %r1512, %rs789; + cvt.f32.bf16 %r1513, %rs790; + cvt.f32.bf16 %r1514, %rs791; + cvt.f32.bf16 %r1515, %rs792; + cvt.f32.bf16 %r1516, %rs793; + cvt.f32.bf16 %r1517, %rs794; + cvt.f32.bf16 %r1518, %rs795; + cvt.f32.bf16 %r1519, %rs796; + cvt.f32.bf16 %r1520, %rs797; + cvt.f32.bf16 %r1521, %rs798; + cvt.f32.bf16 %r1522, %rs799; + cvt.f32.bf16 %r1523, %rs800; + cvt.f32.bf16 %r1524, %rs801; + cvt.f32.bf16 %r1525, %rs802; + cvt.f32.bf16 %r1526, %rs803; + cvt.f32.bf16 %r1527, %rs804; + cvt.f32.bf16 %r1528, %rs805; + cvt.f32.bf16 %r1529, %rs806; + cvt.f32.bf16 %r1530, %rs807; + cvt.f32.bf16 %r1531, %rs808; + cvt.f32.bf16 %r1532, %rs809; + cvt.f32.bf16 %r1533, %rs810; + cvt.f32.bf16 %r1534, %rs811; + cvt.f32.bf16 %r1535, %rs812; + cvt.f32.bf16 %r1536, %rs813; + cvt.f32.bf16 %r1537, %rs814; + cvt.f32.bf16 %r1538, %rs815; + cvt.f32.bf16 %r1539, %rs816; + cvt.f32.bf16 %r1540, %rs817; + cvt.f32.bf16 %r1541, %rs818; + cvt.f32.bf16 %r1542, %rs819; + cvt.f32.bf16 %r1543, %rs820; + cvt.f32.bf16 %r1544, %rs821; + cvt.f32.bf16 %r1545, %rs822; + cvt.f32.bf16 %r1546, %rs823; + cvt.f32.bf16 %r1547, %rs824; + cvt.f32.bf16 %r1548, %rs825; + cvt.f32.bf16 %r1549, %rs826; + cvt.f32.bf16 %r1550, %rs827; + cvt.f32.bf16 %r1551, %rs828; + cvt.f32.bf16 %r1552, %rs829; + cvt.f32.bf16 %r1553, %rs830; + cvt.f32.bf16 %r1554, %rs831; + cvt.f32.bf16 %r1555, %rs832; + cvt.f32.bf16 %r1556, %rs833; + cvt.f32.bf16 %r1557, %rs834; + cvt.f32.bf16 %r1558, %rs835; + cvt.f32.bf16 %r1559, %rs836; + cvt.f32.bf16 %r1560, %rs837; + cvt.f32.bf16 %r1561, %rs838; + cvt.f32.bf16 %r1562, %rs839; + cvt.f32.bf16 %r1563, %rs840; + cvt.f32.bf16 %r1564, %rs841; + cvt.f32.bf16 %r1565, %rs842; + cvt.f32.bf16 %r1566, %rs843; + cvt.f32.bf16 %r1567, %rs844; + cvt.f32.bf16 %r1568, %rs845; + cvt.f32.bf16 %r1569, %rs846; + cvt.f32.bf16 %r1570, %rs847; + cvt.f32.bf16 %r1571, %rs848; + cvt.f32.bf16 %r1572, %rs849; + cvt.f32.bf16 %r1573, %rs850; + cvt.f32.bf16 %r1574, %rs851; + cvt.f32.bf16 %r1575, %rs852; + cvt.f32.bf16 %r1576, %rs853; + cvt.f32.bf16 %r1577, %rs854; + cvt.f32.bf16 %r1578, %rs855; + cvt.f32.bf16 %r1579, %rs856; + cvt.f32.bf16 %r1580, %rs857; + cvt.f32.bf16 %r1581, %rs858; + cvt.f32.bf16 %r1582, %rs859; + cvt.f32.bf16 %r1583, %rs860; + cvt.f32.bf16 %r1584, %rs861; + cvt.f32.bf16 %r1585, %rs862; + cvt.f32.bf16 %r1586, %rs863; + cvt.f32.bf16 %r1587, %rs864; + cvt.f32.bf16 %r1588, %rs865; + cvt.f32.bf16 %r1589, %rs866; + cvt.f32.bf16 %r1590, %rs867; + cvt.f32.bf16 %r1591, %rs868; + cvt.f32.bf16 %r1592, %rs869; + cvt.f32.bf16 %r1593, %rs870; + cvt.f32.bf16 %r1594, %rs871; + cvt.f32.bf16 %r1595, %rs872; + cvt.f32.bf16 %r1596, %rs873; + cvt.f32.bf16 %r1597, %rs874; + cvt.f32.bf16 %r1598, %rs875; + cvt.f32.bf16 %r1599, %rs876; + cvt.f32.bf16 %r1600, %rs877; + cvt.f32.bf16 %r1601, %rs878; + cvt.f32.bf16 %r1602, %rs879; + cvt.f32.bf16 %r1603, %rs880; + cvt.f32.bf16 %r1604, %rs881; + cvt.f32.bf16 %r1605, %rs882; + cvt.f32.bf16 %r1606, %rs883; + cvt.f32.bf16 %r1607, %rs884; + cvt.f32.bf16 %r1608, %rs885; + cvt.f32.bf16 %r1609, %rs886; + cvt.f32.bf16 %r1610, %rs887; + cvt.f32.bf16 %r1611, %rs888; + cvt.f32.bf16 %r1612, %rs889; + cvt.f32.bf16 %r1613, %rs890; + cvt.f32.bf16 %r1614, %rs891; + cvt.f32.bf16 %r1615, %rs892; + cvt.f32.bf16 %r1616, %rs893; + cvt.f32.bf16 %r1617, %rs894; + cvt.f32.bf16 %r1618, %rs895; + cvt.f32.bf16 %r1619, %rs896; + cvt.f32.bf16 %r1620, %rs897; + cvt.f32.bf16 %r1621, %rs898; + cvt.f32.bf16 %r1622, %rs899; + cvt.f32.bf16 %r1623, %rs900; + cvt.f32.bf16 %r1624, %rs901; + cvt.f32.bf16 %r1625, %rs902; + cvt.f32.bf16 %r1626, %rs903; + cvt.f32.bf16 %r1627, %rs904; + cvt.f32.bf16 %r1628, %rs905; + cvt.f32.bf16 %r1629, %rs906; + cvt.f32.bf16 %r1630, %rs907; + cvt.f32.bf16 %r1631, %rs908; + cvt.f32.bf16 %r1632, %rs909; + cvt.f32.bf16 %r1633, %rs910; + cvt.f32.bf16 %r1634, %rs911; + cvt.f32.bf16 %r1635, %rs912; + cvt.f32.bf16 %r1636, %rs913; + cvt.f32.bf16 %r1637, %rs914; + cvt.f32.bf16 %r1638, %rs915; + cvt.f32.bf16 %r1639, %rs916; + cvt.f32.bf16 %r1640, %rs917; + cvt.f32.bf16 %r1641, %rs918; + cvt.f32.bf16 %r1642, %rs919; + cvt.f32.bf16 %r1643, %rs920; + cvt.f32.bf16 %r1644, %rs921; + cvt.f32.bf16 %r1645, %rs922; + cvt.f32.bf16 %r1646, %rs923; + cvt.f32.bf16 %r1647, %rs924; + cvt.f32.bf16 %r1648, %rs925; + cvt.f32.bf16 %r1649, %rs926; + cvt.f32.bf16 %r1650, %rs927; + cvt.f32.bf16 %r1651, %rs928; + cvt.f32.bf16 %r1652, %rs929; + cvt.f32.bf16 %r1653, %rs930; + cvt.f32.bf16 %r1654, %rs931; + cvt.f32.bf16 %r1655, %rs932; + cvt.f32.bf16 %r1656, %rs933; + cvt.f32.bf16 %r1657, %rs934; + cvt.f32.bf16 %r1658, %rs935; + cvt.f32.bf16 %r1659, %rs936; + cvt.f32.bf16 %r1660, %rs937; + cvt.f32.bf16 %r1661, %rs938; + cvt.f32.bf16 %r1662, %rs939; + cvt.f32.bf16 %r1663, %rs940; + cvt.f32.bf16 %r1664, %rs941; + cvt.f32.bf16 %r1665, %rs942; + cvt.f32.bf16 %r1666, %rs943; + cvt.f32.bf16 %r1667, %rs944; + cvt.f32.bf16 %r1668, %rs945; + cvt.f32.bf16 %r1669, %rs946; + cvt.f32.bf16 %r1670, %rs947; + cvt.f32.bf16 %r1671, %rs948; + cvt.f32.bf16 %r1672, %rs949; + cvt.f32.bf16 %r1673, %rs950; + cvt.f32.bf16 %r1674, %rs951; + cvt.f32.bf16 %r1675, %rs952; + cvt.f32.bf16 %r1676, %rs953; + cvt.f32.bf16 %r1677, %rs954; + cvt.f32.bf16 %r1678, %rs955; + cvt.f32.bf16 %r1679, %rs956; + cvt.f32.bf16 %r1680, %rs957; + cvt.f32.bf16 %r1681, %rs958; + cvt.f32.bf16 %r1682, %rs959; + cvt.f32.bf16 %r1683, %rs960; + cvt.f32.bf16 %r1684, %rs961; + cvt.f32.bf16 %r1685, %rs962; + cvt.f32.bf16 %r1686, %rs963; + cvt.f32.bf16 %r1687, %rs964; + cvt.f32.bf16 %r1688, %rs965; + cvt.f32.bf16 %r1689, %rs966; + cvt.f32.bf16 %r1690, %rs967; + cvt.f32.bf16 %r1691, %rs968; + cvt.f32.bf16 %r1692, %rs969; + cvt.f32.bf16 %r1693, %rs970; + cvt.f32.bf16 %r1694, %rs971; + cvt.f32.bf16 %r1695, %rs972; + cvt.f32.bf16 %r1696, %rs973; + cvt.f32.bf16 %r1697, %rs974; + cvt.f32.bf16 %r1698, %rs975; + cvt.f32.bf16 %r1699, %rs976; + cvt.f32.bf16 %r1700, %rs977; + cvt.f32.bf16 %r1701, %rs978; + cvt.f32.bf16 %r1702, %rs979; + cvt.f32.bf16 %r1703, %rs980; + cvt.f32.bf16 %r1704, %rs981; + cvt.f32.bf16 %r1705, %rs982; + cvt.f32.bf16 %r1706, %rs983; + cvt.f32.bf16 %r1707, %rs984; + cvt.f32.bf16 %r1708, %rs985; + cvt.f32.bf16 %r1709, %rs986; + cvt.f32.bf16 %r1710, %rs987; + cvt.f32.bf16 %r1711, %rs988; + cvt.f32.bf16 %r1712, %rs989; + cvt.f32.bf16 %r1713, %rs990; + cvt.f32.bf16 %r1714, %rs991; + cvt.f32.bf16 %r1715, %rs992; + cvt.f32.bf16 %r1716, %rs993; + cvt.f32.bf16 %r1717, %rs994; + cvt.f32.bf16 %r1718, %rs995; + cvt.f32.bf16 %r1719, %rs996; + cvt.f32.bf16 %r1720, %rs997; + cvt.f32.bf16 %r1721, %rs998; + cvt.f32.bf16 %r1722, %rs999; + cvt.f32.bf16 %r1723, %rs1000; + cvt.f32.bf16 %r1724, %rs1001; + cvt.f32.bf16 %r1725, %rs1002; + cvt.f32.bf16 %r1726, %rs1003; + cvt.f32.bf16 %r1727, %rs1004; + .loc 1 107 25 // l2norm.py:107:25 + mov.b32 %r1728, global_smem; + shl.b32 %r1729, %r866, 4; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1730, %rs1005; + .loc 1 107 25 // l2norm.py:107:25 + add.s32 %r1731, %r1728, %r1729; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1732, %rs1006; + .loc 1 107 25 // l2norm.py:107:25 + stmatrix.sync.aligned.m8n8.x2.shared.b16 [%r1731], {%r258, %r259}; + shl.b32 %r1733, %r776, 3; + bar.warp.sync -1; + add.s32 %r1734, %r1728, %r1733; + ld.shared.v4.b32 {%r1735, %r1736, %r1737, %r1738}, [%r1734+16]; + ld.shared.v4.b32 {%r1739, %r1740, %r1741, %r1742}, [%r1734+32]; + ld.shared.v4.b32 {%r1743, %r1744, %r1745, %r1746}, [%r1734+48]; + ld.shared.v4.b32 {%r1747, %r1748, %r1749, %r1750}, [%r1734+64]; + ld.shared.v4.b32 {%r1751, %r1752, %r1753, %r1754}, [%r1734+80]; + ld.shared.v4.b32 {%r1755, %r1756, %r1757, %r1758}, [%r1734+96]; + ld.shared.v4.b32 {%r1759, %r1760, %r1761, %r1762}, [%r1734+112]; + ld.shared.v4.b32 {%r1763, %r1764, %r1765, %r1766}, [%r1734]; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1767, %rs1007; + .loc 1 107 25 // l2norm.py:107:25 + bar.warp.sync -1; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1768, %rs1008; + .loc 1 107 25 // l2norm.py:107:25 + stmatrix.sync.aligned.m8n8.x2.shared.b16 [%r1731], {%r260, %r261}; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1769, %rs1009; + .loc 1 107 25 // l2norm.py:107:25 + bar.warp.sync -1; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1770, %rs1010; + .loc 1 107 25 // l2norm.py:107:25 + ld.shared.v4.b32 {%r1771, %r1772, %r1773, %r1774}, [%r1734]; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r1775, %r932, %r1444; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1776, %rs1011; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1777, %rs208, %rs720, %r1775; +$L__tmp2: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1778, %rs1012; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1779, %rs210, %rs722, %r1777; +$L__tmp4: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r1780, %r1774, %r1455; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1781, %rs211, %rs723, %r1779; +$L__tmp6: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1782, %rs1013; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1783, %rs212, %rs724, %r1781; +$L__tmp8: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1784, %rs1014; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1785, %rs213, %rs725, %r1783; +$L__tmp10: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r1786, %r1774, %r1453; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1787, %rs214, %rs726, %r1785; +$L__tmp12: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1788, %rs1015; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1789, %rs215, %rs727, %r1787; +$L__tmp14: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1790, %rs1016; + cvt.f32.bf16 %r1791, %rs1017; + cvt.f32.bf16 %r1792, %rs1018; + cvt.f32.bf16 %r1793, %rs1019; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r1794, %r908, %r1420; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1795, %rs1020; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1796, %rs184, %rs696, %r1794; +$L__tmp16: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1797, %rs1021; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1798, %rs186, %rs698, %r1796; +$L__tmp18: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1799, %rs1022; +$L__tmp19: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1800, %rs187, %rs699, %r1798; +$L__tmp20: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r1801, %r884, %r1396; +$L__tmp21: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1802, %rs188, %rs700, %r1800; + fma.rn.f32.bf16 %r1803, %rs160, %rs672, %r1801; + fma.rn.f32.bf16 %r1804, %rs189, %rs701, %r1802; + fma.rn.f32.bf16 %r1805, %rs162, %rs674, %r1803; + fma.rn.f32.bf16 %r1806, %rs190, %rs702, %r1804; + fma.rn.f32.bf16 %r1807, %rs163, %rs675, %r1805; + fma.rn.f32.bf16 %r1808, %rs191, %rs703, %r1806; + fma.rn.f32.bf16 %r1809, %rs164, %rs676, %r1807; +$L__tmp22: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1810, %rs1023; +$L__tmp23: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1811, %rs165, %rs677, %r1809; +$L__tmp24: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1812, %rs1024; +$L__tmp25: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1813, %rs166, %rs678, %r1811; +$L__tmp26: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1814, %rs1025; +$L__tmp27: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1815, %rs167, %rs679, %r1813; +$L__tmp28: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1816, %rs1026; + cvt.f32.bf16 %r1817, %rs1027; + cvt.f32.bf16 %r1818, %rs1028; + .loc 1 107 25 // l2norm.py:107:25 + ld.shared.v4.b32 {%r1819, %r1820, %r1821, %r1822}, [%r1734+16]; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r1823, %r892, %r1404; + .loc 1 107 25 // l2norm.py:107:25 + ld.shared.v4.b32 {%r1824, %r1825, %r1826, %r1827}, [%r1734+32]; +$L__tmp29: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1828, %rs168, %rs680, %r1823; +$L__tmp30: + .loc 1 107 25 // l2norm.py:107:25 + ld.shared.v4.b32 {%r1829, %r1830, %r1831, %r1832}, [%r1734+48]; +$L__tmp31: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1833, %rs170, %rs682, %r1828; +$L__tmp32: + .loc 1 107 25 // l2norm.py:107:25 + ld.shared.v4.b32 {%r1834, %r1835, %r1836, %r1837}, [%r1734+64]; +$L__tmp33: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1838, %rs171, %rs683, %r1833; +$L__tmp34: + .loc 1 107 25 // l2norm.py:107:25 + ld.shared.v4.b32 {%r1839, %r1840, %r1841, %r1842}, [%r1734+80]; +$L__tmp35: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1843, %rs172, %rs684, %r1838; +$L__tmp36: + .loc 1 107 25 // l2norm.py:107:25 + ld.shared.v4.b32 {%r1844, %r1845, %r1846, %r1847}, [%r1734+96]; +$L__tmp37: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1848, %rs173, %rs685, %r1843; +$L__tmp38: + .loc 1 107 25 // l2norm.py:107:25 + ld.shared.v4.b32 {%r1849, %r1850, %r1851, %r1852}, [%r1734+112]; +$L__tmp39: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1853, %rs174, %rs686, %r1848; + fma.rn.f32.bf16 %r1854, %rs175, %rs687, %r1853; +$L__tmp40: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1855, %r1815, 8, 31, -1; +$L__tmp41: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1856, %rs1029; +$L__tmp42: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1857, %r1815, %r1855; +$L__tmp43: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1858, %rs1030; +$L__tmp44: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1859, %r1857, 4, 31, -1; +$L__tmp45: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1860, %rs1031; +$L__tmp46: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1861, %r1857, %r1859; +$L__tmp47: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1862, %rs1032; +$L__tmp48: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1863, %r1861, 2, 31, -1; +$L__tmp49: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r1864, %r900, %r1412; +$L__tmp50: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1865, %r1861, %r1863; + fma.rn.f32.bf16 %r1866, %rs176, %rs688, %r1864; +$L__tmp51: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1867, %r1865, 1, 31, -1; +$L__tmp52: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1868, %rs178, %rs690, %r1866; +$L__tmp53: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1869, %r1854, 8, 31, -1; +$L__tmp54: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1870, %rs179, %rs691, %r1868; + add.f32 %r1871, %r1854, %r1869; + fma.rn.f32.bf16 %r1872, %rs180, %rs692, %r1870; +$L__tmp55: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1873, %r1871, 4, 31, -1; +$L__tmp56: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1874, %rs181, %rs693, %r1872; + add.f32 %r1875, %r1871, %r1873; + fma.rn.f32.bf16 %r1876, %rs182, %rs694, %r1874; +$L__tmp57: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1877, %r1875, 2, 31, -1; +$L__tmp58: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1878, %rs183, %rs695, %r1876; + add.f32 %r1879, %r1875, %r1877; +$L__tmp59: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1880, %rs1033; +$L__tmp60: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1881, %r1879, 1, 31, -1; +$L__tmp61: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1882, %rs1034; +$L__tmp62: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1883, %r1878, 8, 31, -1; +$L__tmp63: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r1884, %r916, %r1428; +$L__tmp64: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1885, %r1878, %r1883; + fma.rn.f32.bf16 %r1886, %rs192, %rs704, %r1884; +$L__tmp65: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1887, %r1885, 4, 31, -1; +$L__tmp66: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1888, %rs194, %rs706, %r1886; + add.f32 %r1889, %r1885, %r1887; + fma.rn.f32.bf16 %r1890, %rs195, %rs707, %r1888; +$L__tmp67: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1891, %r1889, 2, 31, -1; +$L__tmp68: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1892, %rs196, %rs708, %r1890; + add.f32 %r1893, %r1889, %r1891; + fma.rn.f32.bf16 %r1894, %rs197, %rs709, %r1892; +$L__tmp69: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1895, %r1893, 1, 31, -1; +$L__tmp70: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1896, %rs198, %rs710, %r1894; +$L__tmp71: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1897, %r1808, 8, 31, -1; +$L__tmp72: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1898, %rs199, %rs711, %r1896; + add.f32 %r1899, %r1808, %r1897; +$L__tmp73: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1900, %rs1035; +$L__tmp74: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1901, %r1899, 4, 31, -1; +$L__tmp75: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1902, %rs1036; +$L__tmp76: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1903, %r1899, %r1901; +$L__tmp77: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1904, %rs1037; +$L__tmp78: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1905, %r1903, 2, 31, -1; +$L__tmp79: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1906, %rs1038; +$L__tmp80: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1907, %r1903, %r1905; +$L__tmp81: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r1908, %r924, %r1436; +$L__tmp82: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1909, %r1907, 1, 31, -1; +$L__tmp83: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1910, %rs200, %rs712, %r1908; +$L__tmp84: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1911, %r1898, 8, 31, -1; +$L__tmp85: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1912, %rs202, %rs714, %r1910; + add.f32 %r1913, %r1898, %r1911; + fma.rn.f32.bf16 %r1914, %rs203, %rs715, %r1912; +$L__tmp86: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1915, %r1913, 4, 31, -1; +$L__tmp87: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1916, %rs204, %rs716, %r1914; + add.f32 %r1917, %r1913, %r1915; + fma.rn.f32.bf16 %r1918, %rs205, %rs717, %r1916; +$L__tmp88: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1919, %r1917, 2, 31, -1; +$L__tmp89: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1920, %rs206, %rs718, %r1918; + add.f32 %r1921, %r1917, %r1919; + fma.rn.f32.bf16 %r1922, %rs207, %rs719, %r1920; +$L__tmp90: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1923, %r1921, 1, 31, -1; +$L__tmp91: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1924, %rs1039; +$L__tmp92: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1925, %r1922, 8, 31, -1; +$L__tmp93: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1926, %rs1040; +$L__tmp94: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1927, %r1922, %r1925; +$L__tmp95: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1928, %rs1041; +$L__tmp96: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1929, %r1927, 4, 31, -1; +$L__tmp97: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1930, %rs1042; +$L__tmp98: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1931, %r1927, %r1929; +$L__tmp99: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r1932, %r940, %r1452; +$L__tmp100: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1933, %r1931, 2, 31, -1; +$L__tmp101: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1934, %rs216, %rs728, %r1932; + add.f32 %r1935, %r1931, %r1933; + fma.rn.f32.bf16 %r1936, %rs218, %rs730, %r1934; +$L__tmp102: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1937, %r1935, 1, 31, -1; +$L__tmp103: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1938, %rs219, %rs731, %r1936; +$L__tmp104: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1939, %r1789, 8, 31, -1; +$L__tmp105: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1940, %rs220, %rs732, %r1938; + add.f32 %r1941, %r1789, %r1939; + fma.rn.f32.bf16 %r1942, %rs221, %rs733, %r1940; +$L__tmp106: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1943, %r1941, 4, 31, -1; +$L__tmp107: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r1944, %rs222, %rs734, %r1942; + add.f32 %r1945, %r1941, %r1943; + fma.rn.f32.bf16 %r1946, %rs223, %rs735, %r1944; +$L__tmp108: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1947, %r1945, 2, 31, -1; +$L__tmp109: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1948, %rs1043; +$L__tmp110: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1949, %r1945, %r1947; +$L__tmp111: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1950, %rs1044; +$L__tmp112: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1951, %r1949, 1, 31, -1; +$L__tmp113: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r1952, %r1774, %r1454; +$L__tmp114: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1953, %r1946, 8, 31, -1; +$L__tmp115: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd663, {%r1786, %r1952}; +$L__tmp116: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1954, %r1946, %r1953; +$L__tmp117: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1955, %rs1045; +$L__tmp118: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1956, %r1954, 4, 31, -1; +$L__tmp119: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1957, %rs1046; +$L__tmp120: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1958, %r1954, %r1956; +$L__tmp121: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r1959, %r1774, %r1451; +$L__tmp122: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1960, %r1958, 2, 31, -1; +$L__tmp123: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r1961, %r1774, %r1452; +$L__tmp124: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1962, %r1958, %r1960; +$L__tmp125: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd664, {%r1959, %r1961}; +$L__tmp126: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r1963, %r1962, 1, 31, -1; +$L__tmp127: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1964, %rs1047; +$L__tmp128: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1965, %r1962, %r1963; +$L__tmp129: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1966, %rs1048; + cvt.f32.bf16 %r1967, %rs1049; + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r1968, %r1965, %r941; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1969, %rs1050; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r1970, %r1774, %r1968; + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r1971, %r1965, %r939; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1972, %rs1051; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r1973, %r1774, %r1971; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1974, %rs1052; + cvt.f32.bf16 %r1975, %rs1053; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r1976, %r1766, %r1447; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1977, %rs1054; + cvt.f32.bf16 %r1978, %rs1055; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r1979, %r1766, %r1449; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1980, %rs1056; + cvt.f32.bf16 %r1981, %rs1057; + cvt.f32.bf16 %r1982, %rs1058; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r1983, %r1766, %r1450; + mul.f32 %r1984, %r1766, %r1448; + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd665, {%r1979, %r1983}; + mov.b64 %rd666, {%r1976, %r1984}; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1985, %rs1059; +$L__tmp130: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r1986, %r1949, %r1951; +$L__tmp131: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1987, %rs1060; + cvt.f32.bf16 %r1988, %rs1061; + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r1989, %r1986, %r937; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1990, %rs1062; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r1991, %r1766, %r1989; + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r1992, %r1986, %r935; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1993, %rs1063; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r1994, %r1766, %r1992; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1995, %rs1064; + cvt.f32.bf16 %r1996, %rs1065; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r1997, %r1766, %r1445; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r1998, %rs1066; + cvt.f32.bf16 %r1999, %rs1067; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2000, %r1766, %r1446; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2001, %rs1068; + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd667, {%r1997, %r2000}; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2002, %r1766, %r1443; + mul.f32 %r2003, %r1766, %r1444; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2004, %rs1069; + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd668, {%r2002, %r2003}; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2005, %rs1070; + cvt.f32.bf16 %r2006, %rs1071; + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r2007, %r1986, %r933; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2008, %rs1072; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r2009, %r1766, %r2007; + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r2010, %r1986, %r931; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2011, %rs1073; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r2012, %r1766, %r2010; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2013, %rs1074; + cvt.f32.bf16 %r2014, %rs1075; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2015, %r1773, %r1439; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2016, %rs1076; + cvt.f32.bf16 %r2017, %rs1077; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2018, %r1773, %r1441; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2019, %rs1078; + cvt.f32.bf16 %r2020, %rs1079; + cvt.f32.bf16 %r2021, %rs1080; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2022, %r1773, %r1442; + mul.f32 %r2023, %r1773, %r1440; + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd669, {%r2018, %r2022}; + mov.b64 %rd670, {%r2015, %r2023}; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2024, %rs1081; +$L__tmp132: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2025, %r1935, %r1937; +$L__tmp133: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2026, %rs1082; + cvt.f32.bf16 %r2027, %rs1083; + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r2028, %r2025, %r929; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2029, %rs1084; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r2030, %r1773, %r2028; + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r2031, %r2025, %r927; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2032, %rs1085; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r2033, %r1773, %r2031; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2034, %rs1086; + cvt.f32.bf16 %r2035, %rs1087; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2036, %r1773, %r1437; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2037, %rs1088; + cvt.f32.bf16 %r2038, %rs1089; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2039, %r1773, %r1438; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2040, %rs1090; + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd671, {%r2036, %r2039}; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2041, %r1773, %r1435; + mul.f32 %r2042, %r1773, %r1436; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2043, %rs1091; + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd672, {%r2041, %r2042}; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2044, %rs1092; + cvt.f32.bf16 %r2045, %rs1093; + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r2046, %r2025, %r925; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2047, %rs1094; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r2048, %r1773, %r2046; + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r2049, %r2025, %r923; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2050, %rs1095; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r2051, %r1773, %r2049; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2052, %rs1096; + cvt.f32.bf16 %r2053, %rs1097; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2054, %r1765, %r1431; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2055, %rs1098; + cvt.f32.bf16 %r2056, %rs1099; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2057, %r1765, %r1433; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2058, %rs1100; + cvt.f32.bf16 %r2059, %rs1101; + cvt.f32.bf16 %r2060, %rs1102; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2061, %r1765, %r1434; + mul.f32 %r2062, %r1765, %r1432; + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd673, {%r2057, %r2061}; + mov.b64 %rd674, {%r2054, %r2062}; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2063, %rs1103; +$L__tmp134: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2064, %r1921, %r1923; +$L__tmp135: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2065, %rs1104; + cvt.f32.bf16 %r2066, %rs1105; + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r2067, %r2064, %r921; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2068, %rs1106; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r2069, %r1765, %r2067; + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r2070, %r2064, %r919; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2071, %rs1107; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r2072, %r1765, %r2070; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2073, %rs1108; + cvt.f32.bf16 %r2074, %rs1109; + cvt.f32.bf16 %r2075, %rs1110; + cvt.f32.bf16 %r2076, %rs1111; + cvt.f32.bf16 %r2077, %rs1112; + cvt.f32.bf16 %r2078, %rs1113; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2079, %r1276, %r1996; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2080, %rs1114; +$L__tmp136: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2081, %rs552, %rs1064, %r2079; +$L__tmp137: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2082, %r1268, %r1981; +$L__tmp138: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2083, %rs554, %rs1066, %r2081; + fma.rn.f32.bf16 %r2084, %rs544, %rs1056, %r2082; + fma.rn.f32.bf16 %r2085, %rs555, %rs1067, %r2083; + fma.rn.f32.bf16 %r2086, %rs546, %rs1058, %r2084; + fma.rn.f32.bf16 %r2087, %rs556, %rs1068, %r2085; + fma.rn.f32.bf16 %r2088, %rs547, %rs1059, %r2086; + fma.rn.f32.bf16 %r2089, %rs557, %rs1069, %r2087; + fma.rn.f32.bf16 %r2090, %rs548, %rs1060, %r2088; + fma.rn.f32.bf16 %r2091, %rs558, %rs1070, %r2089; + fma.rn.f32.bf16 %r2092, %rs549, %rs1061, %r2090; + fma.rn.f32.bf16 %r2093, %rs559, %rs1071, %r2091; + fma.rn.f32.bf16 %r2094, %rs550, %rs1062, %r2092; +$L__tmp139: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2095, %rs1115; +$L__tmp140: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2096, %rs551, %rs1063, %r2094; +$L__tmp141: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2097, %rs1116; + cvt.f32.bf16 %r2098, %rs1117; + cvt.f32.bf16 %r2099, %rs1118; + cvt.f32.bf16 %r2100, %rs1119; + cvt.f32.bf16 %r2101, %rs1120; + cvt.f32.bf16 %r2102, %rs1121; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2103, %r1260, %r1967; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2104, %rs1122; +$L__tmp142: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2105, %rs536, %rs1048, %r2103; +$L__tmp143: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2106, %r1252, %r1928; +$L__tmp144: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2107, %rs538, %rs1050, %r2105; + fma.rn.f32.bf16 %r2108, %rs528, %rs1040, %r2106; + fma.rn.f32.bf16 %r2109, %rs539, %rs1051, %r2107; + fma.rn.f32.bf16 %r2110, %rs530, %rs1042, %r2108; + fma.rn.f32.bf16 %r2111, %rs540, %rs1052, %r2109; + fma.rn.f32.bf16 %r2112, %rs531, %rs1043, %r2110; + fma.rn.f32.bf16 %r2113, %rs541, %rs1053, %r2111; + fma.rn.f32.bf16 %r2114, %rs532, %rs1044, %r2112; + fma.rn.f32.bf16 %r2115, %rs542, %rs1054, %r2113; + fma.rn.f32.bf16 %r2116, %rs533, %rs1045, %r2114; + fma.rn.f32.bf16 %r2117, %rs543, %rs1055, %r2115; + fma.rn.f32.bf16 %r2118, %rs534, %rs1046, %r2116; +$L__tmp145: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2119, %rs1123; +$L__tmp146: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2120, %rs535, %rs1047, %r2118; +$L__tmp147: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2121, %rs1124; + cvt.f32.bf16 %r2122, %rs1125; + cvt.f32.bf16 %r2123, %rs1126; + cvt.f32.bf16 %r2124, %rs1127; + cvt.f32.bf16 %r2125, %rs1128; + cvt.f32.bf16 %r2126, %rs1129; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2127, %r1244, %r1880; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2128, %rs1130; +$L__tmp148: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2129, %rs520, %rs1032, %r2127; +$L__tmp149: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2130, %r1236, %r1814; +$L__tmp150: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2131, %rs522, %rs1034, %r2129; + fma.rn.f32.bf16 %r2132, %rs512, %rs1024, %r2130; + fma.rn.f32.bf16 %r2133, %rs523, %rs1035, %r2131; + fma.rn.f32.bf16 %r2134, %rs514, %rs1026, %r2132; + fma.rn.f32.bf16 %r2135, %rs524, %rs1036, %r2133; + fma.rn.f32.bf16 %r2136, %rs515, %rs1027, %r2134; + fma.rn.f32.bf16 %r2137, %rs525, %rs1037, %r2135; + fma.rn.f32.bf16 %r2138, %rs516, %rs1028, %r2136; + fma.rn.f32.bf16 %r2139, %rs526, %rs1038, %r2137; + fma.rn.f32.bf16 %r2140, %rs517, %rs1029, %r2138; + fma.rn.f32.bf16 %r2141, %rs527, %rs1039, %r2139; + fma.rn.f32.bf16 %r2142, %rs518, %rs1030, %r2140; +$L__tmp151: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2143, %rs1131; +$L__tmp152: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2144, %rs519, %rs1031, %r2142; +$L__tmp153: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2145, %rs1132; + cvt.f32.bf16 %r2146, %rs1133; + cvt.f32.bf16 %r2147, %rs1134; + cvt.f32.bf16 %r2148, %rs1135; + cvt.f32.bf16 %r2149, %rs1136; + cvt.f32.bf16 %r2150, %rs1137; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2151, %r1228, %r1791; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2152, %rs1138; +$L__tmp154: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2153, %rs504, %rs1016, %r2151; +$L__tmp155: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2154, %r1220, %r1769; +$L__tmp156: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2155, %rs506, %rs1018, %r2153; + fma.rn.f32.bf16 %r2156, %rs496, %rs1008, %r2154; + fma.rn.f32.bf16 %r2157, %rs507, %rs1019, %r2155; + fma.rn.f32.bf16 %r2158, %rs498, %rs1010, %r2156; + fma.rn.f32.bf16 %r2159, %rs508, %rs1020, %r2157; + fma.rn.f32.bf16 %r2160, %rs499, %rs1011, %r2158; + fma.rn.f32.bf16 %r2161, %rs509, %rs1021, %r2159; + fma.rn.f32.bf16 %r2162, %rs500, %rs1012, %r2160; + fma.rn.f32.bf16 %r2163, %rs510, %rs1022, %r2161; + fma.rn.f32.bf16 %r2164, %rs501, %rs1013, %r2162; + fma.rn.f32.bf16 %r2165, %rs511, %rs1023, %r2163; + fma.rn.f32.bf16 %r2166, %rs502, %rs1014, %r2164; +$L__tmp157: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2167, %rs1139; +$L__tmp158: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2168, %rs503, %rs1015, %r2166; +$L__tmp159: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2169, %rs1140; + cvt.f32.bf16 %r2170, %rs1141; + cvt.f32.bf16 %r2171, %rs1142; + cvt.f32.bf16 %r2172, %rs1143; + cvt.f32.bf16 %r2173, %rs1144; + cvt.f32.bf16 %r2174, %rs1145; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2175, %r1212, %r1724; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2176, %rs1146; +$L__tmp160: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2177, %rs488, %rs1000, %r2175; +$L__tmp161: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2178, %r1204, %r1716; +$L__tmp162: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2179, %rs490, %rs1002, %r2177; + fma.rn.f32.bf16 %r2180, %rs480, %rs992, %r2178; + fma.rn.f32.bf16 %r2181, %rs491, %rs1003, %r2179; + fma.rn.f32.bf16 %r2182, %rs482, %rs994, %r2180; + fma.rn.f32.bf16 %r2183, %rs492, %rs1004, %r2181; + fma.rn.f32.bf16 %r2184, %rs483, %rs995, %r2182; + fma.rn.f32.bf16 %r2185, %rs493, %rs1005, %r2183; + fma.rn.f32.bf16 %r2186, %rs484, %rs996, %r2184; + fma.rn.f32.bf16 %r2187, %rs494, %rs1006, %r2185; + fma.rn.f32.bf16 %r2188, %rs485, %rs997, %r2186; + fma.rn.f32.bf16 %r2189, %rs495, %rs1007, %r2187; + fma.rn.f32.bf16 %r2190, %rs486, %rs998, %r2188; +$L__tmp163: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2191, %rs1147; +$L__tmp164: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2192, %rs487, %rs999, %r2190; +$L__tmp165: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2193, %rs1148; + cvt.f32.bf16 %r2194, %rs1149; + cvt.f32.bf16 %r2195, %rs1150; + cvt.f32.bf16 %r2196, %rs1151; + cvt.f32.bf16 %r2197, %rs1152; + cvt.f32.bf16 %r2198, %rs1153; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2199, %r1196, %r1708; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2200, %rs1154; +$L__tmp166: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2201, %rs472, %rs984, %r2199; +$L__tmp167: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2202, %r1188, %r1700; +$L__tmp168: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2203, %rs474, %rs986, %r2201; + fma.rn.f32.bf16 %r2204, %rs464, %rs976, %r2202; + fma.rn.f32.bf16 %r2205, %rs475, %rs987, %r2203; + fma.rn.f32.bf16 %r2206, %rs466, %rs978, %r2204; + fma.rn.f32.bf16 %r2207, %rs476, %rs988, %r2205; + fma.rn.f32.bf16 %r2208, %rs467, %rs979, %r2206; + fma.rn.f32.bf16 %r2209, %rs477, %rs989, %r2207; + fma.rn.f32.bf16 %r2210, %rs468, %rs980, %r2208; + fma.rn.f32.bf16 %r2211, %rs478, %rs990, %r2209; + fma.rn.f32.bf16 %r2212, %rs469, %rs981, %r2210; + fma.rn.f32.bf16 %r2213, %rs479, %rs991, %r2211; + fma.rn.f32.bf16 %r2214, %rs470, %rs982, %r2212; +$L__tmp169: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2215, %rs1155; +$L__tmp170: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2216, %rs471, %rs983, %r2214; +$L__tmp171: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2217, %rs1156; + cvt.f32.bf16 %r2218, %rs1157; + cvt.f32.bf16 %r2219, %rs1158; + cvt.f32.bf16 %r2220, %rs1159; + cvt.f32.bf16 %r2221, %rs1160; + cvt.f32.bf16 %r2222, %rs1161; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2223, %r1180, %r1692; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2224, %rs1162; +$L__tmp172: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2225, %rs456, %rs968, %r2223; +$L__tmp173: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2226, %r1172, %r1684; +$L__tmp174: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2227, %rs458, %rs970, %r2225; + fma.rn.f32.bf16 %r2228, %rs448, %rs960, %r2226; + fma.rn.f32.bf16 %r2229, %rs459, %rs971, %r2227; + fma.rn.f32.bf16 %r2230, %rs450, %rs962, %r2228; + fma.rn.f32.bf16 %r2231, %rs460, %rs972, %r2229; + fma.rn.f32.bf16 %r2232, %rs451, %rs963, %r2230; + fma.rn.f32.bf16 %r2233, %rs461, %rs973, %r2231; + fma.rn.f32.bf16 %r2234, %rs452, %rs964, %r2232; + fma.rn.f32.bf16 %r2235, %rs462, %rs974, %r2233; + fma.rn.f32.bf16 %r2236, %rs453, %rs965, %r2234; + fma.rn.f32.bf16 %r2237, %rs463, %rs975, %r2235; + fma.rn.f32.bf16 %r2238, %rs454, %rs966, %r2236; +$L__tmp175: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2239, %rs1163; +$L__tmp176: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2240, %rs455, %rs967, %r2238; +$L__tmp177: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2241, %rs1164; + cvt.f32.bf16 %r2242, %rs1165; + cvt.f32.bf16 %r2243, %rs1166; + cvt.f32.bf16 %r2244, %rs1167; + cvt.f32.bf16 %r2245, %rs1168; + cvt.f32.bf16 %r2246, %rs1169; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2247, %r1164, %r1676; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2248, %rs1170; +$L__tmp178: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2249, %rs440, %rs952, %r2247; +$L__tmp179: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2250, %r1156, %r1668; +$L__tmp180: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2251, %rs442, %rs954, %r2249; + fma.rn.f32.bf16 %r2252, %rs432, %rs944, %r2250; + fma.rn.f32.bf16 %r2253, %rs443, %rs955, %r2251; + fma.rn.f32.bf16 %r2254, %rs434, %rs946, %r2252; + fma.rn.f32.bf16 %r2255, %rs444, %rs956, %r2253; + fma.rn.f32.bf16 %r2256, %rs435, %rs947, %r2254; + fma.rn.f32.bf16 %r2257, %rs445, %rs957, %r2255; + fma.rn.f32.bf16 %r2258, %rs436, %rs948, %r2256; + fma.rn.f32.bf16 %r2259, %rs446, %rs958, %r2257; + fma.rn.f32.bf16 %r2260, %rs437, %rs949, %r2258; + fma.rn.f32.bf16 %r2261, %rs447, %rs959, %r2259; + fma.rn.f32.bf16 %r2262, %rs438, %rs950, %r2260; +$L__tmp181: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2263, %rs1171; +$L__tmp182: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2264, %rs439, %rs951, %r2262; +$L__tmp183: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2265, %rs1172; + cvt.f32.bf16 %r2266, %rs1173; + cvt.f32.bf16 %r2267, %rs1174; + cvt.f32.bf16 %r2268, %rs1175; + cvt.f32.bf16 %r2269, %rs1176; + cvt.f32.bf16 %r2270, %rs1177; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2271, %r1148, %r1660; + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2272, %rs1178; +$L__tmp184: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2273, %rs424, %rs936, %r2271; +$L__tmp185: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2274, %r1140, %r1652; +$L__tmp186: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2275, %rs426, %rs938, %r2273; + fma.rn.f32.bf16 %r2276, %rs416, %rs928, %r2274; + fma.rn.f32.bf16 %r2277, %rs427, %rs939, %r2275; + fma.rn.f32.bf16 %r2278, %rs418, %rs930, %r2276; + fma.rn.f32.bf16 %r2279, %rs428, %rs940, %r2277; + fma.rn.f32.bf16 %r2280, %rs419, %rs931, %r2278; + fma.rn.f32.bf16 %r2281, %rs429, %rs941, %r2279; + fma.rn.f32.bf16 %r2282, %rs420, %rs932, %r2280; + fma.rn.f32.bf16 %r2283, %rs430, %rs942, %r2281; + fma.rn.f32.bf16 %r2284, %rs421, %rs933, %r2282; + fma.rn.f32.bf16 %r2285, %rs431, %rs943, %r2283; + fma.rn.f32.bf16 %r2286, %rs422, %rs934, %r2284; +$L__tmp187: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2287, %rs1179; +$L__tmp188: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2288, %rs423, %rs935, %r2286; +$L__tmp189: + .loc 1 106 49 // l2norm.py:106:49 + cvt.f32.bf16 %r2289, %rs1180; + cvt.f32.bf16 %r2290, %rs1181; + cvt.f32.bf16 %r2291, %rs1182; + cvt.f32.bf16 %r2292, %rs1183; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2293, %r1771, %r1403; + mul.f32 %r2294, %r1771, %r1404; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2295, %r1132, %r1644; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2296, %r1771, %r1405; +$L__tmp190: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2297, %rs408, %rs920, %r2295; +$L__tmp191: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2298, %r1771, %r1406; +$L__tmp192: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2299, %rs410, %rs922, %r2297; +$L__tmp193: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2300, %r1771, %r1407; +$L__tmp194: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2301, %rs411, %rs923, %r2299; +$L__tmp195: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2302, %r1771, %r1408; +$L__tmp196: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2303, %rs412, %rs924, %r2301; +$L__tmp197: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2304, %r1771, %r1409; +$L__tmp198: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2305, %rs413, %rs925, %r2303; +$L__tmp199: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2306, %r1771, %r1410; +$L__tmp200: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2307, %rs414, %rs926, %r2305; +$L__tmp201: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2308, %r1764, %r1411; +$L__tmp202: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2309, %rs415, %rs927, %r2307; +$L__tmp203: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2310, %r1124, %r1636; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2311, %r1764, %r1412; +$L__tmp204: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2312, %rs400, %rs912, %r2310; +$L__tmp205: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2313, %r1764, %r1413; +$L__tmp206: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2314, %rs402, %rs914, %r2312; +$L__tmp207: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2315, %r1764, %r1414; +$L__tmp208: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2316, %rs403, %rs915, %r2314; +$L__tmp209: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2317, %r1764, %r1415; +$L__tmp210: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2318, %rs404, %rs916, %r2316; +$L__tmp211: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2319, %r1764, %r1416; +$L__tmp212: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2320, %rs405, %rs917, %r2318; +$L__tmp213: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2321, %r1764, %r1417; +$L__tmp214: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2322, %rs406, %rs918, %r2320; +$L__tmp215: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2323, %r1764, %r1418; +$L__tmp216: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2324, %rs407, %rs919, %r2322; +$L__tmp217: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2325, %r1772, %r1419; + mul.f32 %r2326, %r1772, %r1420; + mul.f32 %r2327, %r1772, %r1421; + mul.f32 %r2328, %r1772, %r1422; + mul.f32 %r2329, %r1772, %r1423; + mul.f32 %r2330, %r1772, %r1424; + mul.f32 %r2331, %r1772, %r1425; + mul.f32 %r2332, %r1772, %r1426; + mul.f32 %r2333, %r1765, %r1427; + mul.f32 %r2334, %r1765, %r1428; + mul.f32 %r2335, %r1765, %r1429; + mul.f32 %r2336, %r1765, %r1430; + mul.f32 %r2337, %r1774, %r1456; + mul.f32 %r2338, %r1774, %r1457; + mul.f32 %r2339, %r1774, %r1458; + mul.f32 %r2340, %r1735, %r1459; + mul.f32 %r2341, %r1735, %r1460; + mul.f32 %r2342, %r1735, %r1461; + mul.f32 %r2343, %r1735, %r1462; + mul.f32 %r2344, %r1735, %r1463; + mul.f32 %r2345, %r1735, %r1464; + mul.f32 %r2346, %r1735, %r1465; + mul.f32 %r2347, %r1735, %r1466; + mul.f32 %r2348, %r1819, %r1467; + mul.f32 %r2349, %r1819, %r1468; + mul.f32 %r2350, %r1819, %r1469; + mul.f32 %r2351, %r1819, %r1470; + mul.f32 %r2352, %r1819, %r1471; + mul.f32 %r2353, %r1819, %r1472; + mul.f32 %r2354, %r1819, %r1473; + mul.f32 %r2355, %r1819, %r1474; + mul.f32 %r2356, %r1736, %r1475; + mul.f32 %r2357, %r1736, %r1476; + mul.f32 %r2358, %r1736, %r1477; + mul.f32 %r2359, %r1736, %r1478; + mul.f32 %r2360, %r1736, %r1479; + mul.f32 %r2361, %r1736, %r1480; + mul.f32 %r2362, %r1736, %r1481; + mul.f32 %r2363, %r1736, %r1482; + mul.f32 %r2364, %r1820, %r1483; + mul.f32 %r2365, %r1820, %r1484; + mul.f32 %r2366, %r1820, %r1485; + mul.f32 %r2367, %r1820, %r1486; + mul.f32 %r2368, %r1820, %r1487; + mul.f32 %r2369, %r1820, %r1488; + mul.f32 %r2370, %r1820, %r1489; + mul.f32 %r2371, %r1820, %r1490; + mul.f32 %r2372, %r1737, %r1491; + mul.f32 %r2373, %r1737, %r1492; + mul.f32 %r2374, %r1737, %r1493; + mul.f32 %r2375, %r1737, %r1494; + mul.f32 %r2376, %r1737, %r1495; + mul.f32 %r2377, %r1737, %r1496; + mul.f32 %r2378, %r1737, %r1497; + mul.f32 %r2379, %r1737, %r1498; + mul.f32 %r2380, %r1821, %r1499; + mul.f32 %r2381, %r1821, %r1500; + mul.f32 %r2382, %r1821, %r1501; + mul.f32 %r2383, %r1821, %r1502; + mul.f32 %r2384, %r1821, %r1503; + mul.f32 %r2385, %r1821, %r1504; + mul.f32 %r2386, %r1821, %r1505; + mul.f32 %r2387, %r1821, %r1506; + mul.f32 %r2388, %r1738, %r1507; + mul.f32 %r2389, %r1738, %r1508; + mul.f32 %r2390, %r1738, %r1509; + mul.f32 %r2391, %r1738, %r1510; + mul.f32 %r2392, %r1738, %r1511; + mul.f32 %r2393, %r1738, %r1512; + mul.f32 %r2394, %r1738, %r1513; + mul.f32 %r2395, %r1738, %r1514; + mul.f32 %r2396, %r1822, %r1515; + mul.f32 %r2397, %r1822, %r1516; + mul.f32 %r2398, %r1822, %r1517; + mul.f32 %r2399, %r1822, %r1518; + mul.f32 %r2400, %r1822, %r1519; + mul.f32 %r2401, %r1822, %r1520; + mul.f32 %r2402, %r1822, %r1521; + mul.f32 %r2403, %r1822, %r1522; + mul.f32 %r2404, %r1739, %r1523; + mul.f32 %r2405, %r1739, %r1524; + mul.f32 %r2406, %r1739, %r1525; + mul.f32 %r2407, %r1739, %r1526; + mul.f32 %r2408, %r1739, %r1527; + mul.f32 %r2409, %r1739, %r1528; + mul.f32 %r2410, %r1739, %r1529; + mul.f32 %r2411, %r1739, %r1530; + mul.f32 %r2412, %r1824, %r1531; + mul.f32 %r2413, %r1824, %r1532; + mul.f32 %r2414, %r1824, %r1533; + mul.f32 %r2415, %r1824, %r1534; + mul.f32 %r2416, %r1824, %r1535; + mul.f32 %r2417, %r1824, %r1536; + mul.f32 %r2418, %r1824, %r1537; + mul.f32 %r2419, %r1824, %r1538; + mul.f32 %r2420, %r1740, %r1539; + mul.f32 %r2421, %r1740, %r1540; + mul.f32 %r2422, %r1740, %r1541; + mul.f32 %r2423, %r1740, %r1542; + mul.f32 %r2424, %r1740, %r1543; + mul.f32 %r2425, %r1740, %r1544; + mul.f32 %r2426, %r1740, %r1545; + mul.f32 %r2427, %r1740, %r1546; + mul.f32 %r2428, %r1825, %r1547; + mul.f32 %r2429, %r1825, %r1548; + mul.f32 %r2430, %r1825, %r1549; + mul.f32 %r2431, %r1825, %r1550; + mul.f32 %r2432, %r1825, %r1551; + mul.f32 %r2433, %r1825, %r1552; + mul.f32 %r2434, %r1825, %r1553; + mul.f32 %r2435, %r1825, %r1554; + mul.f32 %r2436, %r1741, %r1555; + mul.f32 %r2437, %r1741, %r1556; + mul.f32 %r2438, %r1741, %r1557; + mul.f32 %r2439, %r1741, %r1558; + mul.f32 %r2440, %r1741, %r1559; + mul.f32 %r2441, %r1741, %r1560; + mul.f32 %r2442, %r1741, %r1561; + mul.f32 %r2443, %r1741, %r1562; + mul.f32 %r2444, %r1826, %r1563; + mul.f32 %r2445, %r1826, %r1564; + mul.f32 %r2446, %r1826, %r1565; + mul.f32 %r2447, %r1826, %r1566; + mul.f32 %r2448, %r1826, %r1567; + mul.f32 %r2449, %r1826, %r1568; + mul.f32 %r2450, %r1826, %r1569; + mul.f32 %r2451, %r1826, %r1570; + mul.f32 %r2452, %r1742, %r1571; + mul.f32 %r2453, %r1742, %r1572; + mul.f32 %r2454, %r1742, %r1573; + mul.f32 %r2455, %r1742, %r1574; + mul.f32 %r2456, %r1742, %r1575; + mul.f32 %r2457, %r1742, %r1576; + mul.f32 %r2458, %r1742, %r1577; + mul.f32 %r2459, %r1742, %r1578; + mul.f32 %r2460, %r1827, %r1579; + mul.f32 %r2461, %r1827, %r1580; + mul.f32 %r2462, %r1827, %r1581; + mul.f32 %r2463, %r1827, %r1582; + mul.f32 %r2464, %r1827, %r1583; + mul.f32 %r2465, %r1827, %r1584; + mul.f32 %r2466, %r1827, %r1585; + mul.f32 %r2467, %r1827, %r1586; + mul.f32 %r2468, %r1743, %r1587; + mul.f32 %r2469, %r1743, %r1588; + mul.f32 %r2470, %r1743, %r1589; + mul.f32 %r2471, %r1743, %r1590; + mul.f32 %r2472, %r1743, %r1591; + mul.f32 %r2473, %r1743, %r1592; + mul.f32 %r2474, %r1743, %r1593; + mul.f32 %r2475, %r1743, %r1594; + mul.f32 %r2476, %r1829, %r1595; + mul.f32 %r2477, %r1829, %r1596; + mul.f32 %r2478, %r1829, %r1597; + mul.f32 %r2479, %r1829, %r1598; + mul.f32 %r2480, %r1829, %r1599; + mul.f32 %r2481, %r1829, %r1600; + mul.f32 %r2482, %r1829, %r1601; + mul.f32 %r2483, %r1829, %r1602; + mul.f32 %r2484, %r1744, %r1603; + mul.f32 %r2485, %r1744, %r1604; + mul.f32 %r2486, %r1744, %r1605; + mul.f32 %r2487, %r1744, %r1606; + mul.f32 %r2488, %r1744, %r1607; + mul.f32 %r2489, %r1744, %r1608; + mul.f32 %r2490, %r1744, %r1609; + mul.f32 %r2491, %r1744, %r1610; + mul.f32 %r2492, %r1830, %r1611; + mul.f32 %r2493, %r1830, %r1612; + mul.f32 %r2494, %r1830, %r1613; + mul.f32 %r2495, %r1830, %r1614; + mul.f32 %r2496, %r1830, %r1615; + mul.f32 %r2497, %r1830, %r1616; + mul.f32 %r2498, %r1830, %r1617; + mul.f32 %r2499, %r1830, %r1618; + mul.f32 %r2500, %r1745, %r1619; + mul.f32 %r2501, %r1745, %r1620; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2502, %r1116, %r1628; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2503, %r1745, %r1621; +$L__tmp218: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2504, %rs392, %rs904, %r2502; +$L__tmp219: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2505, %r1745, %r1622; +$L__tmp220: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2506, %rs394, %rs906, %r2504; +$L__tmp221: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2507, %r1745, %r1623; +$L__tmp222: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2508, %rs395, %rs907, %r2506; +$L__tmp223: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2509, %r1745, %r1624; +$L__tmp224: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2510, %rs396, %rs908, %r2508; +$L__tmp225: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2511, %r1745, %r1625; +$L__tmp226: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2512, %rs397, %rs909, %r2510; +$L__tmp227: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2513, %r1745, %r1626; +$L__tmp228: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2514, %rs398, %rs910, %r2512; +$L__tmp229: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2515, %r1831, %r1627; +$L__tmp230: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2516, %rs399, %rs911, %r2514; +$L__tmp231: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2517, %r1108, %r1620; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2518, %r1831, %r1628; +$L__tmp232: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2519, %rs384, %rs896, %r2517; +$L__tmp233: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2520, %r1831, %r1629; +$L__tmp234: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2521, %rs386, %rs898, %r2519; +$L__tmp235: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2522, %r1831, %r1630; +$L__tmp236: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2523, %rs387, %rs899, %r2521; +$L__tmp237: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2524, %r1831, %r1631; +$L__tmp238: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2525, %rs388, %rs900, %r2523; +$L__tmp239: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2526, %r1831, %r1632; +$L__tmp240: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2527, %rs389, %rs901, %r2525; +$L__tmp241: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2528, %r1831, %r1633; +$L__tmp242: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2529, %rs390, %rs902, %r2527; +$L__tmp243: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2530, %r1831, %r1634; +$L__tmp244: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2531, %rs391, %rs903, %r2529; +$L__tmp245: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2532, %r1746, %r1635; + mul.f32 %r2533, %r1746, %r1636; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2534, %r1100, %r1612; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2535, %r1746, %r1637; +$L__tmp246: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2536, %rs376, %rs888, %r2534; +$L__tmp247: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2537, %r1746, %r1638; +$L__tmp248: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2538, %rs378, %rs890, %r2536; +$L__tmp249: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2539, %r1746, %r1639; +$L__tmp250: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2540, %rs379, %rs891, %r2538; +$L__tmp251: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2541, %r1746, %r1640; +$L__tmp252: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2542, %rs380, %rs892, %r2540; +$L__tmp253: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2543, %r1746, %r1641; +$L__tmp254: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2544, %rs381, %rs893, %r2542; +$L__tmp255: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2545, %r1746, %r1642; +$L__tmp256: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2546, %rs382, %rs894, %r2544; +$L__tmp257: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2547, %r1832, %r1643; +$L__tmp258: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2548, %rs383, %rs895, %r2546; +$L__tmp259: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2549, %r1092, %r1604; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2550, %r1832, %r1644; +$L__tmp260: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2551, %rs368, %rs880, %r2549; +$L__tmp261: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2552, %r1832, %r1645; +$L__tmp262: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2553, %rs370, %rs882, %r2551; +$L__tmp263: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2554, %r1832, %r1646; +$L__tmp264: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2555, %rs371, %rs883, %r2553; +$L__tmp265: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2556, %r1832, %r1647; +$L__tmp266: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2557, %rs372, %rs884, %r2555; +$L__tmp267: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2558, %r1832, %r1648; +$L__tmp268: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2559, %rs373, %rs885, %r2557; +$L__tmp269: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2560, %r1832, %r1649; +$L__tmp270: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2561, %rs374, %rs886, %r2559; +$L__tmp271: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2562, %r1832, %r1650; +$L__tmp272: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2563, %rs375, %rs887, %r2561; +$L__tmp273: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2564, %r1747, %r1651; + mul.f32 %r2565, %r1747, %r1652; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2566, %r1084, %r1596; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2567, %r1747, %r1653; +$L__tmp274: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2568, %rs360, %rs872, %r2566; +$L__tmp275: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2569, %r1747, %r1654; +$L__tmp276: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2570, %rs362, %rs874, %r2568; +$L__tmp277: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2571, %r1747, %r1655; +$L__tmp278: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2572, %rs363, %rs875, %r2570; +$L__tmp279: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2573, %r1747, %r1656; +$L__tmp280: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2574, %rs364, %rs876, %r2572; +$L__tmp281: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2575, %r1747, %r1657; +$L__tmp282: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2576, %rs365, %rs877, %r2574; +$L__tmp283: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2577, %r1747, %r1658; +$L__tmp284: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2578, %rs366, %rs878, %r2576; +$L__tmp285: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2579, %r1834, %r1659; +$L__tmp286: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2580, %rs367, %rs879, %r2578; +$L__tmp287: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2581, %r1076, %r1588; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2582, %r1834, %r1660; +$L__tmp288: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2583, %rs352, %rs864, %r2581; +$L__tmp289: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2584, %r1834, %r1661; +$L__tmp290: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2585, %rs354, %rs866, %r2583; +$L__tmp291: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2586, %r1834, %r1662; +$L__tmp292: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2587, %rs355, %rs867, %r2585; +$L__tmp293: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2588, %r1834, %r1663; +$L__tmp294: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2589, %rs356, %rs868, %r2587; +$L__tmp295: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2590, %r1834, %r1664; +$L__tmp296: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2591, %rs357, %rs869, %r2589; +$L__tmp297: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2592, %r1834, %r1665; +$L__tmp298: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2593, %rs358, %rs870, %r2591; +$L__tmp299: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2594, %r1834, %r1666; +$L__tmp300: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2595, %rs359, %rs871, %r2593; +$L__tmp301: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2596, %r1748, %r1667; + mul.f32 %r2597, %r1748, %r1668; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2598, %r1068, %r1580; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2599, %r1748, %r1669; +$L__tmp302: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2600, %rs344, %rs856, %r2598; +$L__tmp303: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2601, %r1748, %r1670; +$L__tmp304: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2602, %rs346, %rs858, %r2600; +$L__tmp305: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2603, %r1748, %r1671; +$L__tmp306: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2604, %rs347, %rs859, %r2602; +$L__tmp307: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2605, %r1748, %r1672; +$L__tmp308: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2606, %rs348, %rs860, %r2604; +$L__tmp309: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2607, %r1748, %r1673; +$L__tmp310: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2608, %rs349, %rs861, %r2606; +$L__tmp311: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2609, %r1748, %r1674; +$L__tmp312: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2610, %rs350, %rs862, %r2608; +$L__tmp313: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2611, %r1835, %r1675; +$L__tmp314: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2612, %rs351, %rs863, %r2610; +$L__tmp315: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2613, %r1060, %r1572; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2614, %r1835, %r1676; +$L__tmp316: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2615, %rs336, %rs848, %r2613; +$L__tmp317: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2616, %r1835, %r1677; +$L__tmp318: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2617, %rs338, %rs850, %r2615; +$L__tmp319: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2618, %r1835, %r1678; +$L__tmp320: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2619, %rs339, %rs851, %r2617; +$L__tmp321: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2620, %r1835, %r1679; +$L__tmp322: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2621, %rs340, %rs852, %r2619; +$L__tmp323: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2622, %r1835, %r1680; +$L__tmp324: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2623, %rs341, %rs853, %r2621; +$L__tmp325: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2624, %r1835, %r1681; +$L__tmp326: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2625, %rs342, %rs854, %r2623; +$L__tmp327: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2626, %r1835, %r1682; +$L__tmp328: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2627, %rs343, %rs855, %r2625; +$L__tmp329: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2628, %r1749, %r1683; + mul.f32 %r2629, %r1749, %r1684; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2630, %r1052, %r1564; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2631, %r1749, %r1685; +$L__tmp330: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2632, %rs328, %rs840, %r2630; +$L__tmp331: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2633, %r1749, %r1686; +$L__tmp332: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2634, %rs330, %rs842, %r2632; +$L__tmp333: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2635, %r1749, %r1687; +$L__tmp334: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2636, %rs331, %rs843, %r2634; +$L__tmp335: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2637, %r1749, %r1688; +$L__tmp336: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2638, %rs332, %rs844, %r2636; +$L__tmp337: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2639, %r1749, %r1689; +$L__tmp338: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2640, %rs333, %rs845, %r2638; +$L__tmp339: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2641, %r1749, %r1690; +$L__tmp340: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2642, %rs334, %rs846, %r2640; +$L__tmp341: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2643, %r1836, %r1691; +$L__tmp342: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2644, %rs335, %rs847, %r2642; +$L__tmp343: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2645, %r1044, %r1556; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2646, %r1836, %r1692; +$L__tmp344: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2647, %rs320, %rs832, %r2645; +$L__tmp345: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2648, %r1836, %r1693; +$L__tmp346: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2649, %rs322, %rs834, %r2647; +$L__tmp347: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2650, %r1836, %r1694; +$L__tmp348: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2651, %rs323, %rs835, %r2649; +$L__tmp349: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2652, %r1836, %r1695; +$L__tmp350: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2653, %rs324, %rs836, %r2651; +$L__tmp351: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2654, %r1836, %r1696; +$L__tmp352: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2655, %rs325, %rs837, %r2653; +$L__tmp353: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2656, %r1836, %r1697; +$L__tmp354: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2657, %rs326, %rs838, %r2655; +$L__tmp355: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2658, %r1836, %r1698; +$L__tmp356: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2659, %rs327, %rs839, %r2657; +$L__tmp357: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2660, %r1750, %r1699; + mul.f32 %r2661, %r1750, %r1700; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2662, %r1036, %r1548; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2663, %r1750, %r1701; +$L__tmp358: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2664, %rs312, %rs824, %r2662; +$L__tmp359: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2665, %r1750, %r1702; +$L__tmp360: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2666, %rs314, %rs826, %r2664; +$L__tmp361: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2667, %r1750, %r1703; +$L__tmp362: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2668, %rs315, %rs827, %r2666; +$L__tmp363: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2669, %r1750, %r1704; +$L__tmp364: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2670, %rs316, %rs828, %r2668; +$L__tmp365: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2671, %r1750, %r1705; +$L__tmp366: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2672, %rs317, %rs829, %r2670; +$L__tmp367: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2673, %r1750, %r1706; +$L__tmp368: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2674, %rs318, %rs830, %r2672; +$L__tmp369: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2675, %r1837, %r1707; +$L__tmp370: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2676, %rs319, %rs831, %r2674; +$L__tmp371: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2677, %r1028, %r1540; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2678, %r1837, %r1708; +$L__tmp372: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2679, %rs304, %rs816, %r2677; +$L__tmp373: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2680, %r1837, %r1709; +$L__tmp374: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2681, %rs306, %rs818, %r2679; +$L__tmp375: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2682, %r1837, %r1710; +$L__tmp376: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2683, %rs307, %rs819, %r2681; +$L__tmp377: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2684, %r1837, %r1711; +$L__tmp378: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2685, %rs308, %rs820, %r2683; +$L__tmp379: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2686, %r1837, %r1712; +$L__tmp380: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2687, %rs309, %rs821, %r2685; +$L__tmp381: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2688, %r1837, %r1713; +$L__tmp382: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2689, %rs310, %rs822, %r2687; +$L__tmp383: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2690, %r1837, %r1714; +$L__tmp384: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2691, %rs311, %rs823, %r2689; +$L__tmp385: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2692, %r1751, %r1715; + mul.f32 %r2693, %r1751, %r1716; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2694, %r1012, %r1524; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2695, %r1751, %r1717; +$L__tmp386: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2696, %rs288, %rs800, %r2694; +$L__tmp387: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2697, %r1751, %r1718; +$L__tmp388: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2698, %rs290, %rs802, %r2696; +$L__tmp389: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2699, %r1751, %r1719; +$L__tmp390: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2700, %rs291, %rs803, %r2698; +$L__tmp391: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2701, %r1751, %r1720; +$L__tmp392: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2702, %rs292, %rs804, %r2700; +$L__tmp393: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2703, %r1751, %r1721; +$L__tmp394: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2704, %rs293, %rs805, %r2702; +$L__tmp395: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2705, %r1751, %r1722; +$L__tmp396: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2706, %rs294, %rs806, %r2704; +$L__tmp397: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2707, %r1839, %r1723; +$L__tmp398: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2708, %rs295, %rs807, %r2706; +$L__tmp399: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2709, %r996, %r1508; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2710, %r1839, %r1724; +$L__tmp400: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2711, %rs272, %rs784, %r2709; +$L__tmp401: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2712, %r1839, %r1725; +$L__tmp402: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2713, %rs274, %rs786, %r2711; +$L__tmp403: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2714, %r1839, %r1726; +$L__tmp404: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2715, %rs275, %rs787, %r2713; +$L__tmp405: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2716, %r1839, %r1727; +$L__tmp406: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2717, %rs276, %rs788, %r2715; +$L__tmp407: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2718, %r1839, %r1730; +$L__tmp408: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2719, %rs277, %rs789, %r2717; +$L__tmp409: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2720, %r1839, %r1732; +$L__tmp410: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2721, %rs278, %rs790, %r2719; +$L__tmp411: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2722, %r1839, %r1767; +$L__tmp412: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2723, %rs279, %rs791, %r2721; +$L__tmp413: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2724, %r1752, %r1768; + mul.f32 %r2725, %r1752, %r1769; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2726, %r980, %r1492; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2727, %r1752, %r1770; +$L__tmp414: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2728, %rs256, %rs768, %r2726; +$L__tmp415: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2729, %r1752, %r1776; +$L__tmp416: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2730, %rs258, %rs770, %r2728; +$L__tmp417: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2731, %r1752, %r1778; +$L__tmp418: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2732, %rs259, %rs771, %r2730; +$L__tmp419: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2733, %r1752, %r1782; +$L__tmp420: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2734, %rs260, %rs772, %r2732; +$L__tmp421: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2735, %r1752, %r1784; +$L__tmp422: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2736, %rs261, %rs773, %r2734; +$L__tmp423: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2737, %r1752, %r1788; +$L__tmp424: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2738, %rs262, %rs774, %r2736; +$L__tmp425: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2739, %r1840, %r1790; +$L__tmp426: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2740, %rs263, %rs775, %r2738; +$L__tmp427: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2741, %r972, %r1484; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2742, %r1840, %r1791; +$L__tmp428: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2743, %rs248, %rs760, %r2741; +$L__tmp429: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2744, %r1840, %r1792; +$L__tmp430: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2745, %rs250, %rs762, %r2743; +$L__tmp431: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2746, %r1840, %r1793; +$L__tmp432: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2747, %rs251, %rs763, %r2745; +$L__tmp433: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2748, %r1840, %r1795; +$L__tmp434: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2749, %rs252, %rs764, %r2747; +$L__tmp435: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2750, %r1840, %r1797; +$L__tmp436: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2751, %rs253, %rs765, %r2749; +$L__tmp437: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2752, %r1840, %r1799; +$L__tmp438: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2753, %rs254, %rs766, %r2751; +$L__tmp439: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2754, %r1840, %r1810; +$L__tmp440: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2755, %rs255, %rs767, %r2753; +$L__tmp441: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2756, %r1753, %r1812; + mul.f32 %r2757, %r1753, %r1814; + mul.f32 %r2758, %r1753, %r1816; + mul.f32 %r2759, %r1753, %r1817; + mul.f32 %r2760, %r1753, %r1818; + mul.f32 %r2761, %r1753, %r1856; + mul.f32 %r2762, %r1753, %r1858; + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2763, %r956, %r1468; + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2764, %r1753, %r1860; +$L__tmp442: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2765, %rs232, %rs744, %r2763; +$L__tmp443: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2766, %r1841, %r1862; +$L__tmp444: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2767, %rs234, %rs746, %r2765; +$L__tmp445: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2768, %r948, %r1460; +$L__tmp446: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2769, %rs235, %rs747, %r2767; + fma.rn.f32.bf16 %r2770, %rs224, %rs736, %r2768; + fma.rn.f32.bf16 %r2771, %rs236, %rs748, %r2769; + fma.rn.f32.bf16 %r2772, %rs226, %rs738, %r2770; + fma.rn.f32.bf16 %r2773, %rs237, %rs749, %r2771; + fma.rn.f32.bf16 %r2774, %rs227, %rs739, %r2772; + fma.rn.f32.bf16 %r2775, %rs238, %rs750, %r2773; + fma.rn.f32.bf16 %r2776, %rs228, %rs740, %r2774; + fma.rn.f32.bf16 %r2777, %rs239, %rs751, %r2775; + fma.rn.f32.bf16 %r2778, %rs229, %rs741, %r2776; +$L__tmp447: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2779, %r1841, %r1880; +$L__tmp448: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2780, %rs230, %rs742, %r2778; +$L__tmp449: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2781, %r1841, %r1882; +$L__tmp450: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2782, %rs231, %rs743, %r2780; +$L__tmp451: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2783, %r1841, %r1900; +$L__tmp452: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2784, %r2782, 8, 31, -1; +$L__tmp453: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2785, %r1841, %r1902; +$L__tmp454: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2786, %r2782, %r2784; +$L__tmp455: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2787, %r1841, %r1904; +$L__tmp456: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2788, %r2786, 4, 31, -1; +$L__tmp457: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2789, %r1841, %r1906; +$L__tmp458: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2790, %r2786, %r2788; +$L__tmp459: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2791, %r1841, %r1924; +$L__tmp460: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2792, %r2790, 2, 31, -1; +$L__tmp461: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2793, %r1754, %r1926; +$L__tmp462: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2794, %r2790, %r2792; +$L__tmp463: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2795, %r964, %r1476; +$L__tmp464: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2796, %r2794, 1, 31, -1; +$L__tmp465: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2797, %rs240, %rs752, %r2795; +$L__tmp466: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2798, %r2777, 8, 31, -1; +$L__tmp467: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2799, %rs242, %rs754, %r2797; + add.f32 %r2800, %r2777, %r2798; + fma.rn.f32.bf16 %r2801, %rs243, %rs755, %r2799; +$L__tmp468: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2802, %r2800, 4, 31, -1; +$L__tmp469: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2803, %rs244, %rs756, %r2801; + add.f32 %r2804, %r2800, %r2802; + fma.rn.f32.bf16 %r2805, %rs245, %rs757, %r2803; +$L__tmp470: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2806, %r2804, 2, 31, -1; +$L__tmp471: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2807, %rs246, %rs758, %r2805; + add.f32 %r2808, %r2804, %r2806; + fma.rn.f32.bf16 %r2809, %rs247, %rs759, %r2807; +$L__tmp472: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2810, %r2808, 1, 31, -1; +$L__tmp473: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2811, %r1754, %r1928; +$L__tmp474: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2812, %r2809, 8, 31, -1; +$L__tmp475: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2813, %r1754, %r1930; +$L__tmp476: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2814, %r2809, %r2812; +$L__tmp477: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2815, %r1754, %r1948; +$L__tmp478: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2816, %r2814, 4, 31, -1; +$L__tmp479: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2817, %r1754, %r1950; +$L__tmp480: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2818, %r2814, %r2816; +$L__tmp481: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2819, %r1754, %r1955; +$L__tmp482: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2820, %r2818, 2, 31, -1; +$L__tmp483: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2821, %r1754, %r1957; +$L__tmp484: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2822, %r2818, %r2820; +$L__tmp485: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2823, %r1754, %r1964; +$L__tmp486: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2824, %r2822, 1, 31, -1; +$L__tmp487: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2825, %r1842, %r1966; +$L__tmp488: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2826, %r2755, 8, 31, -1; +$L__tmp489: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2827, %r988, %r1500; +$L__tmp490: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2828, %r2755, %r2826; + fma.rn.f32.bf16 %r2829, %rs264, %rs776, %r2827; +$L__tmp491: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2830, %r2828, 4, 31, -1; +$L__tmp492: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2831, %rs266, %rs778, %r2829; + add.f32 %r2832, %r2828, %r2830; + fma.rn.f32.bf16 %r2833, %rs267, %rs779, %r2831; +$L__tmp493: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2834, %r2832, 2, 31, -1; +$L__tmp494: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2835, %rs268, %rs780, %r2833; + add.f32 %r2836, %r2832, %r2834; + fma.rn.f32.bf16 %r2837, %rs269, %rs781, %r2835; +$L__tmp495: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2838, %r2836, 1, 31, -1; +$L__tmp496: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2839, %rs270, %rs782, %r2837; +$L__tmp497: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2840, %r2740, 8, 31, -1; +$L__tmp498: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2841, %rs271, %rs783, %r2839; + add.f32 %r2842, %r2740, %r2840; +$L__tmp499: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2843, %r1842, %r1967; +$L__tmp500: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2844, %r2842, 4, 31, -1; +$L__tmp501: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2845, %r1842, %r1969; +$L__tmp502: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2846, %r2842, %r2844; +$L__tmp503: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2847, %r1842, %r1972; +$L__tmp504: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2848, %r2846, 2, 31, -1; +$L__tmp505: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2849, %r1842, %r1974; +$L__tmp506: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2850, %r2846, %r2848; +$L__tmp507: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2851, %r1842, %r1975; +$L__tmp508: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2852, %r2850, 1, 31, -1; +$L__tmp509: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2853, %r1842, %r1977; +$L__tmp510: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2854, %r2841, 8, 31, -1; +$L__tmp511: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2855, %r1842, %r1978; +$L__tmp512: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2856, %r2841, %r2854; +$L__tmp513: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2857, %r1755, %r1980; +$L__tmp514: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2858, %r2856, 4, 31, -1; +$L__tmp515: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2859, %r1004, %r1516; +$L__tmp516: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2860, %r2856, %r2858; + fma.rn.f32.bf16 %r2861, %rs280, %rs792, %r2859; +$L__tmp517: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2862, %r2860, 2, 31, -1; +$L__tmp518: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2863, %rs282, %rs794, %r2861; + add.f32 %r2864, %r2860, %r2862; + fma.rn.f32.bf16 %r2865, %rs283, %rs795, %r2863; +$L__tmp519: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2866, %r2864, 1, 31, -1; +$L__tmp520: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2867, %rs284, %rs796, %r2865; +$L__tmp521: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2868, %r2723, 8, 31, -1; +$L__tmp522: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2869, %rs285, %rs797, %r2867; + add.f32 %r2870, %r2723, %r2868; + fma.rn.f32.bf16 %r2871, %rs286, %rs798, %r2869; +$L__tmp523: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2872, %r2870, 4, 31, -1; +$L__tmp524: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2873, %rs287, %rs799, %r2871; + add.f32 %r2874, %r2870, %r2872; +$L__tmp525: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2875, %r1755, %r1981; +$L__tmp526: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2876, %r2874, 2, 31, -1; +$L__tmp527: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2877, %r1755, %r1982; +$L__tmp528: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2878, %r2874, %r2876; +$L__tmp529: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2879, %r1755, %r1985; +$L__tmp530: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2880, %r2878, 1, 31, -1; +$L__tmp531: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2881, %r1755, %r1987; +$L__tmp532: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2882, %r2873, 8, 31, -1; +$L__tmp533: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2883, %r1755, %r1988; +$L__tmp534: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2884, %r2873, %r2882; +$L__tmp535: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2885, %r1755, %r1990; +$L__tmp536: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2886, %r2884, 4, 31, -1; +$L__tmp537: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2887, %r1755, %r1993; +$L__tmp538: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2888, %r2884, %r2886; +$L__tmp539: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2889, %r1844, %r1995; +$L__tmp540: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2890, %r2888, 2, 31, -1; +$L__tmp541: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r2891, %r1020, %r1532; +$L__tmp542: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2892, %r2888, %r2890; + fma.rn.f32.bf16 %r2893, %rs296, %rs808, %r2891; +$L__tmp543: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2894, %r2892, 1, 31, -1; +$L__tmp544: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2895, %rs298, %rs810, %r2893; +$L__tmp545: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2896, %r2708, 8, 31, -1; +$L__tmp546: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2897, %rs299, %rs811, %r2895; + add.f32 %r2898, %r2708, %r2896; + fma.rn.f32.bf16 %r2899, %rs300, %rs812, %r2897; +$L__tmp547: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2900, %r2898, 4, 31, -1; +$L__tmp548: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2901, %rs301, %rs813, %r2899; + add.f32 %r2902, %r2898, %r2900; + fma.rn.f32.bf16 %r2903, %rs302, %rs814, %r2901; +$L__tmp549: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2904, %r2902, 2, 31, -1; +$L__tmp550: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r2905, %rs303, %rs815, %r2903; + add.f32 %r2906, %r2902, %r2904; +$L__tmp551: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2907, %r1844, %r1996; +$L__tmp552: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2908, %r2906, 1, 31, -1; +$L__tmp553: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2909, %r1844, %r1998; +$L__tmp554: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2910, %r2905, 8, 31, -1; +$L__tmp555: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2911, %r1844, %r1999; +$L__tmp556: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2912, %r2905, %r2910; +$L__tmp557: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2913, %r1844, %r2001; +$L__tmp558: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2914, %r2912, 4, 31, -1; +$L__tmp559: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2915, %r1844, %r2004; +$L__tmp560: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2916, %r2912, %r2914; +$L__tmp561: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2917, %r1844, %r2005; +$L__tmp562: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2918, %r2916, 2, 31, -1; +$L__tmp563: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2919, %r1844, %r2006; +$L__tmp564: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2920, %r2916, %r2918; +$L__tmp565: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2921, %r1756, %r2008; +$L__tmp566: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2922, %r2920, 1, 31, -1; +$L__tmp567: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2923, %r1756, %r2011; +$L__tmp568: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2924, %r2691, 8, 31, -1; +$L__tmp569: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2925, %r1756, %r2013; +$L__tmp570: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2926, %r2691, %r2924; +$L__tmp571: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2927, %r1756, %r2014; +$L__tmp572: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2928, %r2926, 4, 31, -1; +$L__tmp573: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2929, %r1756, %r2016; +$L__tmp574: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2930, %r2926, %r2928; +$L__tmp575: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2931, %r1756, %r2017; +$L__tmp576: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2932, %r2930, 2, 31, -1; +$L__tmp577: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2933, %r1756, %r2019; +$L__tmp578: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2934, %r2930, %r2932; +$L__tmp579: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2935, %r1756, %r2020; +$L__tmp580: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2936, %r2934, 1, 31, -1; +$L__tmp581: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2937, %r1845, %r2021; +$L__tmp582: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2938, %r2676, 8, 31, -1; +$L__tmp583: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2939, %r1845, %r2024; +$L__tmp584: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2940, %r2676, %r2938; +$L__tmp585: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2941, %r1845, %r2026; +$L__tmp586: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2942, %r2940, 4, 31, -1; +$L__tmp587: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2943, %r1845, %r2027; +$L__tmp588: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2944, %r2940, %r2942; +$L__tmp589: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2945, %r1845, %r2029; +$L__tmp590: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2946, %r2944, 2, 31, -1; +$L__tmp591: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2947, %r1845, %r2032; +$L__tmp592: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2948, %r2944, %r2946; +$L__tmp593: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2949, %r1845, %r2034; +$L__tmp594: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2950, %r2948, 1, 31, -1; +$L__tmp595: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2951, %r1845, %r2035; +$L__tmp596: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2952, %r2659, 8, 31, -1; +$L__tmp597: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2953, %r1757, %r2037; +$L__tmp598: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2954, %r2659, %r2952; +$L__tmp599: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2955, %r1757, %r2038; +$L__tmp600: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2956, %r2954, 4, 31, -1; +$L__tmp601: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2957, %r1757, %r2040; +$L__tmp602: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2958, %r2954, %r2956; +$L__tmp603: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2959, %r1757, %r2043; +$L__tmp604: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2960, %r2958, 2, 31, -1; +$L__tmp605: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2961, %r1757, %r2044; +$L__tmp606: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2962, %r2958, %r2960; +$L__tmp607: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2963, %r1757, %r2045; +$L__tmp608: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2964, %r2962, 1, 31, -1; +$L__tmp609: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2965, %r1757, %r2047; +$L__tmp610: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2966, %r2644, 8, 31, -1; +$L__tmp611: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2967, %r1757, %r2050; +$L__tmp612: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2968, %r2644, %r2966; +$L__tmp613: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2969, %r1846, %r2052; +$L__tmp614: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2970, %r2968, 4, 31, -1; +$L__tmp615: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2971, %r1846, %r2053; +$L__tmp616: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2972, %r2968, %r2970; +$L__tmp617: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2973, %r1846, %r2055; +$L__tmp618: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2974, %r2972, 2, 31, -1; +$L__tmp619: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2975, %r1846, %r2056; +$L__tmp620: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2976, %r2972, %r2974; +$L__tmp621: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2977, %r1846, %r2058; +$L__tmp622: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2978, %r2976, 1, 31, -1; +$L__tmp623: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2979, %r1846, %r2059; +$L__tmp624: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2980, %r2627, 8, 31, -1; +$L__tmp625: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2981, %r1846, %r2060; +$L__tmp626: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2982, %r2627, %r2980; +$L__tmp627: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2983, %r1846, %r2063; +$L__tmp628: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2984, %r2982, 4, 31, -1; +$L__tmp629: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2985, %r1758, %r2065; +$L__tmp630: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2986, %r2982, %r2984; +$L__tmp631: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2987, %r1758, %r2066; +$L__tmp632: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2988, %r2986, 2, 31, -1; +$L__tmp633: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2989, %r1758, %r2068; +$L__tmp634: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2990, %r2986, %r2988; +$L__tmp635: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2991, %r1758, %r2071; +$L__tmp636: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2992, %r2990, 1, 31, -1; +$L__tmp637: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2993, %r1758, %r2073; +$L__tmp638: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2994, %r2612, 8, 31, -1; +$L__tmp639: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2995, %r1758, %r2074; +$L__tmp640: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r2996, %r2612, %r2994; +$L__tmp641: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2997, %r1758, %r2075; +$L__tmp642: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r2998, %r2996, 4, 31, -1; +$L__tmp643: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r2999, %r1758, %r2076; +$L__tmp644: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3000, %r2996, %r2998; +$L__tmp645: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3001, %r1847, %r2077; +$L__tmp646: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3002, %r3000, 2, 31, -1; +$L__tmp647: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3003, %r1847, %r2078; +$L__tmp648: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3004, %r3000, %r3002; +$L__tmp649: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3005, %r1847, %r2080; +$L__tmp650: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3006, %r3004, 1, 31, -1; +$L__tmp651: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3007, %r1847, %r2095; +$L__tmp652: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3008, %r2595, 8, 31, -1; +$L__tmp653: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3009, %r1847, %r2097; +$L__tmp654: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3010, %r2595, %r3008; +$L__tmp655: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3011, %r1847, %r2098; +$L__tmp656: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3012, %r3010, 4, 31, -1; +$L__tmp657: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3013, %r1847, %r2099; +$L__tmp658: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3014, %r3010, %r3012; +$L__tmp659: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3015, %r1847, %r2100; +$L__tmp660: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3016, %r3014, 2, 31, -1; +$L__tmp661: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3017, %r1759, %r2101; +$L__tmp662: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3018, %r3014, %r3016; +$L__tmp663: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3019, %r1759, %r2102; +$L__tmp664: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3020, %r3018, 1, 31, -1; +$L__tmp665: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3021, %r1759, %r2104; +$L__tmp666: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3022, %r2580, 8, 31, -1; +$L__tmp667: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3023, %r1759, %r2119; +$L__tmp668: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3024, %r2580, %r3022; +$L__tmp669: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3025, %r1759, %r2121; +$L__tmp670: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3026, %r3024, 4, 31, -1; +$L__tmp671: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3027, %r1759, %r2122; +$L__tmp672: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3028, %r3024, %r3026; +$L__tmp673: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3029, %r1759, %r2123; +$L__tmp674: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3030, %r3028, 2, 31, -1; +$L__tmp675: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3031, %r1759, %r2124; +$L__tmp676: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3032, %r3028, %r3030; +$L__tmp677: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3033, %r1849, %r2125; +$L__tmp678: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3034, %r3032, 1, 31, -1; +$L__tmp679: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3035, %r1849, %r2126; +$L__tmp680: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3036, %r2563, 8, 31, -1; +$L__tmp681: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3037, %r1849, %r2128; +$L__tmp682: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3038, %r2563, %r3036; +$L__tmp683: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3039, %r1849, %r2143; +$L__tmp684: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3040, %r3038, 4, 31, -1; +$L__tmp685: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3041, %r1849, %r2145; +$L__tmp686: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3042, %r3038, %r3040; +$L__tmp687: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3043, %r1849, %r2146; +$L__tmp688: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3044, %r3042, 2, 31, -1; +$L__tmp689: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3045, %r1849, %r2147; +$L__tmp690: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3046, %r3042, %r3044; +$L__tmp691: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3047, %r1849, %r2148; +$L__tmp692: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3048, %r3046, 1, 31, -1; +$L__tmp693: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3049, %r1760, %r2149; +$L__tmp694: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3050, %r2548, 8, 31, -1; +$L__tmp695: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3051, %r1760, %r2150; +$L__tmp696: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3052, %r2548, %r3050; +$L__tmp697: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3053, %r1760, %r2152; +$L__tmp698: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3054, %r3052, 4, 31, -1; +$L__tmp699: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3055, %r1760, %r2167; +$L__tmp700: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3056, %r3052, %r3054; +$L__tmp701: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3057, %r1760, %r2169; +$L__tmp702: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3058, %r3056, 2, 31, -1; +$L__tmp703: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3059, %r1760, %r2170; +$L__tmp704: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3060, %r3056, %r3058; +$L__tmp705: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3061, %r1760, %r2171; +$L__tmp706: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3062, %r3060, 1, 31, -1; +$L__tmp707: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3063, %r1760, %r2172; +$L__tmp708: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3064, %r2531, 8, 31, -1; +$L__tmp709: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3065, %r1850, %r2173; +$L__tmp710: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3066, %r2531, %r3064; +$L__tmp711: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3067, %r1850, %r2174; +$L__tmp712: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3068, %r3066, 4, 31, -1; +$L__tmp713: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3069, %r1850, %r2176; +$L__tmp714: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3070, %r3066, %r3068; +$L__tmp715: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3071, %r1850, %r2191; +$L__tmp716: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3072, %r3070, 2, 31, -1; +$L__tmp717: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3073, %r1850, %r2193; +$L__tmp718: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3074, %r3070, %r3072; +$L__tmp719: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3075, %r1850, %r2194; +$L__tmp720: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3076, %r3074, 1, 31, -1; +$L__tmp721: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3077, %r1850, %r2195; +$L__tmp722: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3078, %r2516, 8, 31, -1; +$L__tmp723: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3079, %r1850, %r2196; +$L__tmp724: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3080, %r2516, %r3078; +$L__tmp725: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3081, %r1761, %r2197; +$L__tmp726: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3082, %r3080, 4, 31, -1; +$L__tmp727: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3083, %r1761, %r2198; +$L__tmp728: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3084, %r3080, %r3082; +$L__tmp729: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3085, %r1761, %r2200; +$L__tmp730: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3086, %r3084, 2, 31, -1; +$L__tmp731: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3087, %r1761, %r2215; +$L__tmp732: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3088, %r3084, %r3086; +$L__tmp733: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3089, %r1761, %r2217; +$L__tmp734: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3090, %r3088, 1, 31, -1; +$L__tmp735: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3091, %r1761, %r2218; +$L__tmp736: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3092, %r2324, 8, 31, -1; +$L__tmp737: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3093, %r1761, %r2219; +$L__tmp738: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3094, %r2324, %r3092; +$L__tmp739: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3095, %r1761, %r2220; +$L__tmp740: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3096, %r3094, 4, 31, -1; +$L__tmp741: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3097, %r1851, %r2221; +$L__tmp742: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3098, %r3094, %r3096; +$L__tmp743: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3099, %r1851, %r2222; +$L__tmp744: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3100, %r3098, 2, 31, -1; +$L__tmp745: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3101, %r1851, %r2224; +$L__tmp746: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3102, %r3098, %r3100; +$L__tmp747: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3103, %r1851, %r2239; +$L__tmp748: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3104, %r3102, 1, 31, -1; +$L__tmp749: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3105, %r1851, %r2241; +$L__tmp750: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3106, %r2309, 8, 31, -1; +$L__tmp751: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3107, %r1851, %r2242; +$L__tmp752: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3108, %r2309, %r3106; +$L__tmp753: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3109, %r1851, %r2243; +$L__tmp754: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3110, %r3108, 4, 31, -1; +$L__tmp755: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3111, %r1851, %r2244; +$L__tmp756: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3112, %r3108, %r3110; +$L__tmp757: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3113, %r1762, %r2245; +$L__tmp758: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3114, %r3112, 2, 31, -1; +$L__tmp759: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3115, %r1762, %r2246; +$L__tmp760: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3116, %r3112, %r3114; +$L__tmp761: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3117, %r1762, %r2248; +$L__tmp762: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3118, %r3116, 1, 31, -1; +$L__tmp763: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3119, %r1762, %r2263; +$L__tmp764: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3120, %r2288, 8, 31, -1; +$L__tmp765: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3121, %r1762, %r2265; +$L__tmp766: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3122, %r2288, %r3120; +$L__tmp767: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3123, %r1762, %r2266; +$L__tmp768: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3124, %r3122, 4, 31, -1; +$L__tmp769: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3125, %r1762, %r2267; +$L__tmp770: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3126, %r3122, %r3124; +$L__tmp771: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3127, %r1762, %r2268; +$L__tmp772: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3128, %r3126, 2, 31, -1; +$L__tmp773: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3129, %r1852, %r2269; +$L__tmp774: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3130, %r3126, %r3128; +$L__tmp775: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3131, %r1852, %r2270; +$L__tmp776: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3132, %r3130, 1, 31, -1; +$L__tmp777: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3133, %r1852, %r2272; +$L__tmp778: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3134, %r2285, 8, 31, -1; +$L__tmp779: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3135, %r1852, %r2287; +$L__tmp780: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3136, %r2285, %r3134; +$L__tmp781: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3137, %r1852, %r2289; +$L__tmp782: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3138, %r3136, 4, 31, -1; +$L__tmp783: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3139, %r1852, %r2290; +$L__tmp784: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3140, %r3136, %r3138; +$L__tmp785: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3141, %r1852, %r2291; +$L__tmp786: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3142, %r3140, 2, 31, -1; +$L__tmp787: + .loc 1 107 18 // l2norm.py:107:18 + mul.f32 %r3143, %r1852, %r2292; +$L__tmp788: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3144, %r3140, %r3142; +$L__tmp789: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r3145, %r1284, %r2011; +$L__tmp790: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3146, %r3144, 1, 31, -1; +$L__tmp791: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r3147, %r1292, %r2024; +$L__tmp792: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3148, %r2264, 8, 31, -1; +$L__tmp793: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r3149, %r1300, %r2038; +$L__tmp794: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3150, %r2264, %r3148; +$L__tmp795: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r3151, %r1308, %r2053; +$L__tmp796: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3152, %r3150, 4, 31, -1; +$L__tmp797: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r3153, %r1316, %r2066; +$L__tmp798: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3154, %r3150, %r3152; +$L__tmp799: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r3155, %r1324, %r2078; +$L__tmp800: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3156, %r3154, 2, 31, -1; +$L__tmp801: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r3157, %r1332, %r2102; +$L__tmp802: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3158, %r3154, %r3156; +$L__tmp803: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r3159, %r1340, %r2126; +$L__tmp804: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3160, %r3158, 1, 31, -1; +$L__tmp805: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r3161, %r1348, %r2150; +$L__tmp806: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3162, %r2261, 8, 31, -1; +$L__tmp807: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r3163, %r1356, %r2174; +$L__tmp808: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3164, %r2261, %r3162; +$L__tmp809: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r3165, %r1364, %r2198; +$L__tmp810: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3166, %r3164, 4, 31, -1; +$L__tmp811: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r3167, %r1372, %r2222; +$L__tmp812: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3168, %r3164, %r3166; +$L__tmp813: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r3169, %r1380, %r2246; +$L__tmp814: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3170, %r3168, 2, 31, -1; +$L__tmp815: + .loc 1 107 50 // l2norm.py:107:50 + mul.f32 %r3171, %r1388, %r2270; +$L__tmp816: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3172, %r3168, %r3170; + fma.rn.f32.bf16 %r3173, %rs560, %rs1072, %r3145; +$L__tmp817: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3174, %r3172, 1, 31, -1; +$L__tmp818: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3175, %rs562, %rs1074, %r3173; +$L__tmp819: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3176, %r2240, 8, 31, -1; +$L__tmp820: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3177, %rs563, %rs1075, %r3175; + add.f32 %r3178, %r2240, %r3176; + fma.rn.f32.bf16 %r3179, %rs564, %rs1076, %r3177; +$L__tmp821: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3180, %r3178, 4, 31, -1; +$L__tmp822: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3181, %rs565, %rs1077, %r3179; + add.f32 %r3182, %r3178, %r3180; + fma.rn.f32.bf16 %r3183, %rs566, %rs1078, %r3181; +$L__tmp823: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3184, %r3182, 2, 31, -1; +$L__tmp824: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3185, %rs567, %rs1079, %r3183; + add.f32 %r3186, %r3182, %r3184; + fma.rn.f32.bf16 %r3187, %rs568, %rs1080, %r3147; +$L__tmp825: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3188, %r3186, 1, 31, -1; +$L__tmp826: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3189, %rs570, %rs1082, %r3187; +$L__tmp827: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3190, %r2237, 8, 31, -1; +$L__tmp828: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3191, %rs571, %rs1083, %r3189; + add.f32 %r3192, %r2237, %r3190; + fma.rn.f32.bf16 %r3193, %rs572, %rs1084, %r3191; +$L__tmp829: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3194, %r3192, 4, 31, -1; +$L__tmp830: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3195, %rs573, %rs1085, %r3193; + add.f32 %r3196, %r3192, %r3194; + fma.rn.f32.bf16 %r3197, %rs574, %rs1086, %r3195; +$L__tmp831: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3198, %r3196, 2, 31, -1; +$L__tmp832: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3199, %rs575, %rs1087, %r3197; + add.f32 %r3200, %r3196, %r3198; + fma.rn.f32.bf16 %r3201, %rs576, %rs1088, %r3149; +$L__tmp833: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3202, %r3200, 1, 31, -1; +$L__tmp834: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3203, %rs578, %rs1090, %r3201; +$L__tmp835: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3204, %r2216, 8, 31, -1; +$L__tmp836: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3205, %rs579, %rs1091, %r3203; + add.f32 %r3206, %r2216, %r3204; + fma.rn.f32.bf16 %r3207, %rs580, %rs1092, %r3205; +$L__tmp837: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3208, %r3206, 4, 31, -1; +$L__tmp838: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3209, %rs581, %rs1093, %r3207; + add.f32 %r3210, %r3206, %r3208; + fma.rn.f32.bf16 %r3211, %rs582, %rs1094, %r3209; +$L__tmp839: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3212, %r3210, 2, 31, -1; +$L__tmp840: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3213, %rs583, %rs1095, %r3211; + add.f32 %r3214, %r3210, %r3212; + fma.rn.f32.bf16 %r3215, %rs584, %rs1096, %r3151; +$L__tmp841: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3216, %r3214, 1, 31, -1; +$L__tmp842: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3217, %rs586, %rs1098, %r3215; +$L__tmp843: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3218, %r2213, 8, 31, -1; +$L__tmp844: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3219, %rs587, %rs1099, %r3217; + add.f32 %r3220, %r2213, %r3218; + fma.rn.f32.bf16 %r3221, %rs588, %rs1100, %r3219; +$L__tmp845: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3222, %r3220, 4, 31, -1; +$L__tmp846: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3223, %rs589, %rs1101, %r3221; + add.f32 %r3224, %r3220, %r3222; + fma.rn.f32.bf16 %r3225, %rs590, %rs1102, %r3223; +$L__tmp847: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3226, %r3224, 2, 31, -1; +$L__tmp848: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3227, %rs591, %rs1103, %r3225; + add.f32 %r3228, %r3224, %r3226; + fma.rn.f32.bf16 %r3229, %rs592, %rs1104, %r3153; +$L__tmp849: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3230, %r3228, 1, 31, -1; +$L__tmp850: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3231, %rs594, %rs1106, %r3229; +$L__tmp851: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3232, %r2192, 8, 31, -1; +$L__tmp852: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3233, %rs595, %rs1107, %r3231; + add.f32 %r3234, %r2192, %r3232; + fma.rn.f32.bf16 %r3235, %rs596, %rs1108, %r3233; +$L__tmp853: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3236, %r3234, 4, 31, -1; +$L__tmp854: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3237, %rs597, %rs1109, %r3235; + add.f32 %r3238, %r3234, %r3236; + fma.rn.f32.bf16 %r3239, %rs598, %rs1110, %r3237; +$L__tmp855: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3240, %r3238, 2, 31, -1; +$L__tmp856: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3241, %rs599, %rs1111, %r3239; + add.f32 %r3242, %r3238, %r3240; + fma.rn.f32.bf16 %r3243, %rs600, %rs1112, %r3155; +$L__tmp857: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3244, %r3242, 1, 31, -1; +$L__tmp858: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3245, %rs602, %rs1114, %r3243; +$L__tmp859: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3246, %r2189, 8, 31, -1; +$L__tmp860: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3247, %rs603, %rs1115, %r3245; + add.f32 %r3248, %r2189, %r3246; + fma.rn.f32.bf16 %r3249, %rs604, %rs1116, %r3247; +$L__tmp861: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3250, %r3248, 4, 31, -1; +$L__tmp862: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3251, %rs605, %rs1117, %r3249; + add.f32 %r3252, %r3248, %r3250; + fma.rn.f32.bf16 %r3253, %rs606, %rs1118, %r3251; +$L__tmp863: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3254, %r3252, 2, 31, -1; +$L__tmp864: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3255, %rs607, %rs1119, %r3253; + add.f32 %r3256, %r3252, %r3254; + fma.rn.f32.bf16 %r3257, %rs608, %rs1120, %r3157; +$L__tmp865: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3258, %r3256, 1, 31, -1; +$L__tmp866: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3259, %rs610, %rs1122, %r3257; +$L__tmp867: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3260, %r2168, 8, 31, -1; +$L__tmp868: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3261, %rs611, %rs1123, %r3259; + add.f32 %r3262, %r2168, %r3260; + fma.rn.f32.bf16 %r3263, %rs612, %rs1124, %r3261; +$L__tmp869: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3264, %r3262, 4, 31, -1; +$L__tmp870: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3265, %rs613, %rs1125, %r3263; + add.f32 %r3266, %r3262, %r3264; + fma.rn.f32.bf16 %r3267, %rs614, %rs1126, %r3265; +$L__tmp871: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3268, %r3266, 2, 31, -1; +$L__tmp872: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3269, %rs615, %rs1127, %r3267; + add.f32 %r3270, %r3266, %r3268; + fma.rn.f32.bf16 %r3271, %rs616, %rs1128, %r3159; +$L__tmp873: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3272, %r3270, 1, 31, -1; +$L__tmp874: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3273, %rs618, %rs1130, %r3271; +$L__tmp875: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3274, %r2165, 8, 31, -1; +$L__tmp876: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3275, %rs619, %rs1131, %r3273; + add.f32 %r3276, %r2165, %r3274; + fma.rn.f32.bf16 %r3277, %rs620, %rs1132, %r3275; +$L__tmp877: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3278, %r3276, 4, 31, -1; +$L__tmp878: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3279, %rs621, %rs1133, %r3277; + add.f32 %r3280, %r3276, %r3278; + fma.rn.f32.bf16 %r3281, %rs622, %rs1134, %r3279; +$L__tmp879: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3282, %r3280, 2, 31, -1; +$L__tmp880: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3283, %rs623, %rs1135, %r3281; + add.f32 %r3284, %r3280, %r3282; + fma.rn.f32.bf16 %r3285, %rs624, %rs1136, %r3161; +$L__tmp881: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3286, %r3284, 1, 31, -1; +$L__tmp882: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3287, %rs626, %rs1138, %r3285; +$L__tmp883: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3288, %r2144, 8, 31, -1; +$L__tmp884: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3289, %rs627, %rs1139, %r3287; + add.f32 %r3290, %r2144, %r3288; + fma.rn.f32.bf16 %r3291, %rs628, %rs1140, %r3289; +$L__tmp885: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3292, %r3290, 4, 31, -1; +$L__tmp886: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3293, %rs629, %rs1141, %r3291; + add.f32 %r3294, %r3290, %r3292; + fma.rn.f32.bf16 %r3295, %rs630, %rs1142, %r3293; +$L__tmp887: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3296, %r3294, 2, 31, -1; +$L__tmp888: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3297, %rs631, %rs1143, %r3295; + add.f32 %r3298, %r3294, %r3296; + fma.rn.f32.bf16 %r3299, %rs632, %rs1144, %r3163; +$L__tmp889: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3300, %r3298, 1, 31, -1; +$L__tmp890: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3301, %rs634, %rs1146, %r3299; +$L__tmp891: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3302, %r2141, 8, 31, -1; +$L__tmp892: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3303, %rs635, %rs1147, %r3301; + add.f32 %r3304, %r2141, %r3302; + fma.rn.f32.bf16 %r3305, %rs636, %rs1148, %r3303; +$L__tmp893: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3306, %r3304, 4, 31, -1; +$L__tmp894: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3307, %rs637, %rs1149, %r3305; + add.f32 %r3308, %r3304, %r3306; + fma.rn.f32.bf16 %r3309, %rs638, %rs1150, %r3307; +$L__tmp895: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3310, %r3308, 2, 31, -1; +$L__tmp896: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3311, %rs639, %rs1151, %r3309; + add.f32 %r3312, %r3308, %r3310; + fma.rn.f32.bf16 %r3313, %rs640, %rs1152, %r3165; +$L__tmp897: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3314, %r3312, 1, 31, -1; +$L__tmp898: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3315, %rs642, %rs1154, %r3313; +$L__tmp899: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3316, %r2120, 8, 31, -1; +$L__tmp900: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3317, %rs643, %rs1155, %r3315; + add.f32 %r3318, %r2120, %r3316; + fma.rn.f32.bf16 %r3319, %rs644, %rs1156, %r3317; +$L__tmp901: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3320, %r3318, 4, 31, -1; +$L__tmp902: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3321, %rs645, %rs1157, %r3319; + add.f32 %r3322, %r3318, %r3320; + fma.rn.f32.bf16 %r3323, %rs646, %rs1158, %r3321; +$L__tmp903: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3324, %r3322, 2, 31, -1; +$L__tmp904: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3325, %rs647, %rs1159, %r3323; + add.f32 %r3326, %r3322, %r3324; + fma.rn.f32.bf16 %r3327, %rs648, %rs1160, %r3167; +$L__tmp905: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3328, %r3326, 1, 31, -1; +$L__tmp906: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3329, %rs650, %rs1162, %r3327; +$L__tmp907: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3330, %r2117, 8, 31, -1; +$L__tmp908: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3331, %rs651, %rs1163, %r3329; + add.f32 %r3332, %r2117, %r3330; + fma.rn.f32.bf16 %r3333, %rs652, %rs1164, %r3331; +$L__tmp909: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3334, %r3332, 4, 31, -1; +$L__tmp910: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3335, %rs653, %rs1165, %r3333; + add.f32 %r3336, %r3332, %r3334; + fma.rn.f32.bf16 %r3337, %rs654, %rs1166, %r3335; +$L__tmp911: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3338, %r3336, 2, 31, -1; +$L__tmp912: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3339, %rs655, %rs1167, %r3337; + add.f32 %r3340, %r3336, %r3338; + fma.rn.f32.bf16 %r3341, %rs656, %rs1168, %r3169; +$L__tmp913: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3342, %r3340, 1, 31, -1; +$L__tmp914: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3343, %rs658, %rs1170, %r3341; +$L__tmp915: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3344, %r2096, 8, 31, -1; +$L__tmp916: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3345, %rs659, %rs1171, %r3343; + add.f32 %r3346, %r2096, %r3344; + fma.rn.f32.bf16 %r3347, %rs660, %rs1172, %r3345; +$L__tmp917: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3348, %r3346, 4, 31, -1; +$L__tmp918: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3349, %rs661, %rs1173, %r3347; + add.f32 %r3350, %r3346, %r3348; + fma.rn.f32.bf16 %r3351, %rs662, %rs1174, %r3349; +$L__tmp919: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3352, %r3350, 2, 31, -1; +$L__tmp920: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3353, %rs663, %rs1175, %r3351; + add.f32 %r3354, %r3350, %r3352; + fma.rn.f32.bf16 %r3355, %rs664, %rs1176, %r3171; +$L__tmp921: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3356, %r3354, 1, 31, -1; +$L__tmp922: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3357, %rs666, %rs1178, %r3355; +$L__tmp923: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3358, %r2093, 8, 31, -1; +$L__tmp924: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3359, %rs667, %rs1179, %r3357; + add.f32 %r3360, %r2093, %r3358; + fma.rn.f32.bf16 %r3361, %rs668, %rs1180, %r3359; +$L__tmp925: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3362, %r3360, 4, 31, -1; +$L__tmp926: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3363, %rs669, %rs1181, %r3361; + add.f32 %r3364, %r3360, %r3362; + fma.rn.f32.bf16 %r3365, %rs670, %rs1182, %r3363; +$L__tmp927: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3366, %r3364, 2, 31, -1; +$L__tmp928: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + fma.rn.f32.bf16 %r3367, %rs671, %rs1183, %r3365; + add.f32 %r3368, %r3364, %r3366; + add.f32 %r3369, %r1865, %r1867; +$L__tmp929: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3370, %r3368, 1, 31, -1; +$L__tmp930: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3371, %r1879, %r1881; +$L__tmp931: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3372, %r3185, 8, 31, -1; +$L__tmp932: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd675, {%r2335, %r2336}; +$L__tmp933: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3373, %r3185, %r3372; + add.f32 %r3374, %r1893, %r1895; +$L__tmp934: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3375, %r3373, 4, 31, -1; +$L__tmp935: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3376, %r1907, %r1909; + add.f32 %r3377, %r3373, %r3375; +$L__tmp936: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd676, {%r2333, %r2334}; +$L__tmp937: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3378, %r3377, 2, 31, -1; +$L__tmp938: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3379, %r2794, %r2796; + add.f32 %r3380, %r3377, %r3378; + add.f32 %r3381, %r2808, %r2810; +$L__tmp939: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3382, %r3380, 1, 31, -1; +$L__tmp940: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3383, %r2064, %r917; +$L__tmp941: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3384, %r3199, 8, 31, -1; +$L__tmp942: + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r3385, %r1765, %r3383; +$L__tmp943: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3386, %r3199, %r3384; + add.f32 %r3387, %r2822, %r2824; +$L__tmp944: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3388, %r3386, 4, 31, -1; +$L__tmp945: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3389, %r2064, %r915; +$L__tmp946: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3390, %r3386, %r3388; +$L__tmp947: + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r3391, %r1765, %r3389; +$L__tmp948: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3392, %r3390, 2, 31, -1; +$L__tmp949: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3393, %r2836, %r2838; + add.f32 %r3394, %r3390, %r3392; + add.f32 %r3395, %r2850, %r2852; +$L__tmp950: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3396, %r3394, 1, 31, -1; +$L__tmp951: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3397, %r2864, %r2866; +$L__tmp952: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3398, %r3213, 8, 31, -1; +$L__tmp953: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd677, {%r2331, %r2332}; +$L__tmp954: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3399, %r3213, %r3398; + add.f32 %r3400, %r2878, %r2880; +$L__tmp955: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3401, %r3399, 4, 31, -1; +$L__tmp956: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3402, %r2892, %r2894; + add.f32 %r3403, %r3399, %r3401; +$L__tmp957: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd678, {%r2329, %r2330}; +$L__tmp958: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3404, %r3403, 2, 31, -1; +$L__tmp959: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3405, %r2906, %r2908; + add.f32 %r3406, %r3403, %r3404; + add.f32 %r3407, %r2920, %r2922; +$L__tmp960: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3408, %r3406, 1, 31, -1; +$L__tmp961: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3409, %r2934, %r2936; +$L__tmp962: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3410, %r3227, 8, 31, -1; +$L__tmp963: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3411, %r3376, %r913; +$L__tmp964: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3412, %r3227, %r3410; +$L__tmp965: + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r3413, %r1772, %r3411; +$L__tmp966: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3414, %r3412, 4, 31, -1; +$L__tmp967: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3415, %r2948, %r2950; + add.f32 %r3416, %r3412, %r3414; +$L__tmp968: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3417, %r3376, %r911; +$L__tmp969: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3418, %r3416, 2, 31, -1; +$L__tmp970: + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r3419, %r1772, %r3417; +$L__tmp971: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3420, %r3416, %r3418; + add.f32 %r3421, %r2962, %r2964; +$L__tmp972: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3422, %r3420, 1, 31, -1; +$L__tmp973: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3423, %r2976, %r2978; +$L__tmp974: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3424, %r3241, 8, 31, -1; +$L__tmp975: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3425, %r2990, %r2992; + add.f32 %r3426, %r3241, %r3424; +$L__tmp976: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd679, {%r2327, %r2328}; +$L__tmp977: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3427, %r3426, 4, 31, -1; +$L__tmp978: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3428, %r3004, %r3006; + add.f32 %r3429, %r3426, %r3427; + add.f32 %r3430, %r3018, %r3020; +$L__tmp979: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3431, %r3429, 2, 31, -1; +$L__tmp980: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd680, {%r2325, %r2326}; +$L__tmp981: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3432, %r3429, %r3431; + add.f32 %r3433, %r3032, %r3034; +$L__tmp982: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3434, %r3432, 1, 31, -1; +$L__tmp983: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3435, %r3046, %r3048; +$L__tmp984: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3436, %r3255, 8, 31, -1; +$L__tmp985: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3437, %r3376, %r909; +$L__tmp986: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3438, %r3255, %r3436; +$L__tmp987: + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r3439, %r1772, %r3437; +$L__tmp988: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3440, %r3438, 4, 31, -1; +$L__tmp989: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3441, %r3060, %r3062; + add.f32 %r3442, %r3438, %r3440; +$L__tmp990: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3443, %r3376, %r907; +$L__tmp991: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3444, %r3442, 2, 31, -1; +$L__tmp992: + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r3445, %r1772, %r3443; +$L__tmp993: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3446, %r3442, %r3444; + add.f32 %r3447, %r3074, %r3076; +$L__tmp994: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3448, %r3446, 1, 31, -1; +$L__tmp995: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3449, %r3088, %r3090; +$L__tmp996: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3450, %r3269, 8, 31, -1; +$L__tmp997: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3451, %r3102, %r3104; + add.f32 %r3452, %r3269, %r3450; +$L__tmp998: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd681, {%r2321, %r2323}; +$L__tmp999: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3453, %r3452, 4, 31, -1; +$L__tmp1000: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3454, %r3116, %r3118; + add.f32 %r3455, %r3452, %r3453; + add.f32 %r3456, %r3130, %r3132; +$L__tmp1001: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3457, %r3455, 2, 31, -1; +$L__tmp1002: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd682, {%r2317, %r2319}; +$L__tmp1003: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3458, %r3455, %r3457; + add.f32 %r3459, %r3144, %r3146; +$L__tmp1004: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3460, %r3458, 1, 31, -1; +$L__tmp1005: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3461, %r3158, %r3160; +$L__tmp1006: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3462, %r3283, 8, 31, -1; +$L__tmp1007: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3463, %r3172, %r3174; + add.f32 %r3464, %r3283, %r3462; +$L__tmp1008: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3465, %r3374, %r905; +$L__tmp1009: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3466, %r3464, 4, 31, -1; +$L__tmp1010: + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r3467, %r1764, %r3465; +$L__tmp1011: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3468, %r3464, %r3466; + add.f32 %r3469, %r3186, %r3188; +$L__tmp1012: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3470, %r3468, 2, 31, -1; +$L__tmp1013: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3471, %r3200, %r3202; + add.f32 %r3472, %r3468, %r3470; +$L__tmp1014: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3473, %r3374, %r903; +$L__tmp1015: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3474, %r3472, 1, 31, -1; +$L__tmp1016: + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r3475, %r1764, %r3473; +$L__tmp1017: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3476, %r3297, 8, 31, -1; +$L__tmp1018: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3477, %r3214, %r3216; + add.f32 %r3478, %r3297, %r3476; + add.f32 %r3479, %r3228, %r3230; +$L__tmp1019: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3480, %r3478, 4, 31, -1; +$L__tmp1020: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd683, {%r2313, %r2315}; +$L__tmp1021: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3481, %r3478, %r3480; + add.f32 %r3482, %r3242, %r3244; +$L__tmp1022: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3483, %r3481, 2, 31, -1; +$L__tmp1023: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3484, %r3256, %r3258; + add.f32 %r3485, %r3481, %r3483; + add.f32 %r3486, %r3270, %r3272; +$L__tmp1024: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3487, %r3485, 1, 31, -1; +$L__tmp1025: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd684, {%r2308, %r2311}; +$L__tmp1026: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3488, %r3311, 8, 31, -1; +$L__tmp1027: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3489, %r3284, %r3286; + add.f32 %r3490, %r3311, %r3488; +$L__tmp1028: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3491, %r3374, %r901; +$L__tmp1029: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3492, %r3490, 4, 31, -1; +$L__tmp1030: + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r3493, %r1764, %r3491; +$L__tmp1031: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3494, %r3490, %r3492; + add.f32 %r3495, %r3298, %r3300; +$L__tmp1032: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3496, %r3494, 2, 31, -1; +$L__tmp1033: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3497, %r3312, %r3314; + add.f32 %r3498, %r3494, %r3496; +$L__tmp1034: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3499, %r3374, %r899; +$L__tmp1035: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3500, %r3498, 1, 31, -1; +$L__tmp1036: + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r3501, %r1764, %r3499; +$L__tmp1037: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3502, %r3325, 8, 31, -1; +$L__tmp1038: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3503, %r3326, %r3328; + add.f32 %r3504, %r3325, %r3502; + add.f32 %r3505, %r3340, %r3342; +$L__tmp1039: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3506, %r3504, 4, 31, -1; +$L__tmp1040: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd685, {%r2304, %r2306}; +$L__tmp1041: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3507, %r3504, %r3506; + add.f32 %r3508, %r3354, %r3356; +$L__tmp1042: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3509, %r3507, 2, 31, -1; +$L__tmp1043: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3510, %r3368, %r3370; + add.f32 %r3511, %r3507, %r3509; + add.f32 %r3512, %r3380, %r3382; +$L__tmp1044: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3513, %r3511, 1, 31, -1; +$L__tmp1045: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd686, {%r2300, %r2302}; +$L__tmp1046: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3514, %r3339, 8, 31, -1; +$L__tmp1047: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3515, %r3394, %r3396; + add.f32 %r3516, %r3339, %r3514; + add.f32 %r3517, %r3406, %r3408; +$L__tmp1048: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3518, %r3516, 4, 31, -1; +$L__tmp1049: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3519, %r3371, %r897; +$L__tmp1050: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3520, %r3516, %r3518; +$L__tmp1051: + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r3521, %r1771, %r3519; +$L__tmp1052: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3522, %r3520, 2, 31, -1; +$L__tmp1053: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3523, %r3420, %r3422; + add.f32 %r3524, %r3520, %r3522; + add.f32 %r3525, %r3432, %r3434; +$L__tmp1054: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3526, %r3524, 1, 31, -1; +$L__tmp1055: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3527, %r3371, %r895; +$L__tmp1056: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3528, %r3353, 8, 31, -1; +$L__tmp1057: + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r3529, %r1771, %r3527; +$L__tmp1058: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3530, %r3353, %r3528; + add.f32 %r3531, %r3446, %r3448; +$L__tmp1059: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3532, %r3530, 4, 31, -1; +$L__tmp1060: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3533, %r3458, %r3460; + add.f32 %r3534, %r3530, %r3532; +$L__tmp1061: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd687, {%r2296, %r2298}; +$L__tmp1062: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3535, %r3534, 2, 31, -1; +$L__tmp1063: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3536, %r3472, %r3474; + add.f32 %r3537, %r3534, %r3535; + add.f32 %r3538, %r3485, %r3487; +$L__tmp1064: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3539, %r3537, 1, 31, -1; +$L__tmp1065: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3540, %r3498, %r3500; +$L__tmp1066: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3541, %r3367, 8, 31, -1; +$L__tmp1067: + .loc 1 107 36 // l2norm.py:107:36 + mov.b64 %rd688, {%r2293, %r2294}; +$L__tmp1068: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3542, %r3367, %r3541; + add.f32 %r3543, %r3511, %r3513; +$L__tmp1069: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3544, %r3542, 4, 31, -1; +$L__tmp1070: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3545, %r3371, %r893; +$L__tmp1071: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3546, %r3542, %r3544; +$L__tmp1072: + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r3547, %r1771, %r3545; +$L__tmp1073: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3548, %r3546, 2, 31, -1; +$L__tmp1074: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3549, %r3524, %r3526; + add.f32 %r3550, %r3546, %r3548; + add.f32 %r3551, %r3537, %r3539; +$L__tmp1075: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ] + shfl.sync.bfly.b32 %r3552, %r3550, 1, 31, -1; +$L__tmp1076: + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3553, %r3371, %r891; +$L__tmp1077: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ] + add.f32 %r3554, %r3550, %r3552; +$L__tmp1078: + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r3555, %r1771, %r3553; + .loc 1 107 69 // l2norm.py:107:69 + mul.f32 %r3556, %r3369, %r883; + mul.f32 %r3557, %r3369, %r884; + mul.f32 %r3558, %r3369, %r885; + mul.f32 %r3559, %r3369, %r886; + mul.f32 %r3560, %r3369, %r887; + mul.f32 %r3561, %r3369, %r888; + mul.f32 %r3562, %r3369, %r889; + mul.f32 %r3563, %r3369, %r890; + mul.f32 %r3564, %r3371, %r892; + mul.f32 %r3565, %r3371, %r894; + mul.f32 %r3566, %r3371, %r896; + mul.f32 %r3567, %r3371, %r898; + mul.f32 %r3568, %r3374, %r900; + mul.f32 %r3569, %r3374, %r902; + mul.f32 %r3570, %r3374, %r904; + mul.f32 %r3571, %r3374, %r906; + mul.f32 %r3572, %r3376, %r908; + mul.f32 %r3573, %r3376, %r910; + mul.f32 %r3574, %r3376, %r912; + mul.f32 %r3575, %r3376, %r914; + mul.f32 %r3576, %r2064, %r916; + mul.f32 %r3577, %r2064, %r918; + mul.f32 %r3578, %r2064, %r920; + mul.f32 %r3579, %r2064, %r922; + mul.f32 %r3580, %r2025, %r924; + mul.f32 %r3581, %r2025, %r926; + mul.f32 %r3582, %r2025, %r928; + mul.f32 %r3583, %r2025, %r930; + mul.f32 %r3584, %r1986, %r932; + mul.f32 %r3585, %r1986, %r934; + mul.f32 %r3586, %r1986, %r936; + mul.f32 %r3587, %r1986, %r938; + mul.f32 %r3588, %r1965, %r940; + mul.f32 %r3589, %r1965, %r942; + mul.f32 %r3590, %r1965, %r943; + mul.f32 %r3591, %r1965, %r944; + mul.f32 %r3592, %r1965, %r945; + mul.f32 %r3593, %r1965, %r946; + mul.f32 %r3594, %r3379, %r947; + mul.f32 %r3595, %r3379, %r948; + mul.f32 %r3596, %r3379, %r949; + mul.f32 %r3597, %r3379, %r950; + mul.f32 %r3598, %r3379, %r951; + mul.f32 %r3599, %r3379, %r952; + mul.f32 %r3600, %r3379, %r953; + mul.f32 %r3601, %r3379, %r954; + mul.f32 %r3602, %r3381, %r955; + mul.f32 %r3603, %r3381, %r956; + mul.f32 %r3604, %r3381, %r957; + mul.f32 %r3605, %r3381, %r958; + mul.f32 %r3606, %r3381, %r959; + mul.f32 %r3607, %r3381, %r960; + mul.f32 %r3608, %r3381, %r961; + mul.f32 %r3609, %r3381, %r962; + mul.f32 %r3610, %r3387, %r963; + mul.f32 %r3611, %r3387, %r964; + mul.f32 %r3612, %r3387, %r965; + mul.f32 %r3613, %r3387, %r966; + mul.f32 %r3614, %r3387, %r967; + mul.f32 %r3615, %r3387, %r968; + mul.f32 %r3616, %r3387, %r969; + mul.f32 %r3617, %r3387, %r970; + mul.f32 %r3618, %r3393, %r971; + mul.f32 %r3619, %r3393, %r972; + mul.f32 %r3620, %r3393, %r973; + mul.f32 %r3621, %r3393, %r974; + mul.f32 %r3622, %r3393, %r975; + mul.f32 %r3623, %r3393, %r976; + mul.f32 %r3624, %r3393, %r977; + mul.f32 %r3625, %r3393, %r978; + mul.f32 %r3626, %r3395, %r979; + mul.f32 %r3627, %r3395, %r980; + mul.f32 %r3628, %r3395, %r981; + mul.f32 %r3629, %r3395, %r982; + mul.f32 %r3630, %r3395, %r983; + mul.f32 %r3631, %r3395, %r984; + mul.f32 %r3632, %r3395, %r985; + mul.f32 %r3633, %r3395, %r986; + mul.f32 %r3634, %r3397, %r987; + mul.f32 %r3635, %r3397, %r988; + mul.f32 %r3636, %r3397, %r989; + mul.f32 %r3637, %r3397, %r990; + mul.f32 %r3638, %r3397, %r991; + mul.f32 %r3639, %r3397, %r992; + mul.f32 %r3640, %r3397, %r993; + mul.f32 %r3641, %r3397, %r994; + mul.f32 %r3642, %r3400, %r995; + mul.f32 %r3643, %r3400, %r996; + mul.f32 %r3644, %r3400, %r997; + mul.f32 %r3645, %r3400, %r998; + mul.f32 %r3646, %r3400, %r999; + mul.f32 %r3647, %r3400, %r1000; + mul.f32 %r3648, %r3400, %r1001; + mul.f32 %r3649, %r3400, %r1002; + mul.f32 %r3650, %r3402, %r1003; + mul.f32 %r3651, %r3402, %r1004; + mul.f32 %r3652, %r3402, %r1005; + mul.f32 %r3653, %r3402, %r1006; + mul.f32 %r3654, %r3402, %r1007; + mul.f32 %r3655, %r3402, %r1008; + mul.f32 %r3656, %r3402, %r1009; + mul.f32 %r3657, %r3402, %r1010; + mul.f32 %r3658, %r3405, %r1011; + mul.f32 %r3659, %r3405, %r1012; + mul.f32 %r3660, %r3405, %r1013; + mul.f32 %r3661, %r3405, %r1014; + mul.f32 %r3662, %r3405, %r1015; + mul.f32 %r3663, %r3405, %r1016; + mul.f32 %r3664, %r3405, %r1017; + mul.f32 %r3665, %r3405, %r1018; + mul.f32 %r3666, %r3407, %r1019; + mul.f32 %r3667, %r3407, %r1020; + mul.f32 %r3668, %r3407, %r1021; + mul.f32 %r3669, %r3407, %r1022; + mul.f32 %r3670, %r3407, %r1023; + mul.f32 %r3671, %r3407, %r1024; + mul.f32 %r3672, %r3407, %r1025; + mul.f32 %r3673, %r3407, %r1026; + mul.f32 %r3674, %r3409, %r1027; + mul.f32 %r3675, %r3409, %r1028; + mul.f32 %r3676, %r3409, %r1029; + mul.f32 %r3677, %r3409, %r1030; + mul.f32 %r3678, %r3409, %r1031; + mul.f32 %r3679, %r3409, %r1032; + mul.f32 %r3680, %r3409, %r1033; + mul.f32 %r3681, %r3409, %r1034; + mul.f32 %r3682, %r3415, %r1035; + mul.f32 %r3683, %r3415, %r1036; + mul.f32 %r3684, %r3415, %r1037; + mul.f32 %r3685, %r3415, %r1038; + mul.f32 %r3686, %r3415, %r1039; + mul.f32 %r3687, %r3415, %r1040; + mul.f32 %r3688, %r3415, %r1041; + mul.f32 %r3689, %r3415, %r1042; + mul.f32 %r3690, %r3421, %r1043; + mul.f32 %r3691, %r3421, %r1044; + mul.f32 %r3692, %r3421, %r1045; + mul.f32 %r3693, %r3421, %r1046; + mul.f32 %r3694, %r3421, %r1047; + mul.f32 %r3695, %r3421, %r1048; + mul.f32 %r3696, %r3421, %r1049; + mul.f32 %r3697, %r3421, %r1050; + mul.f32 %r3698, %r3423, %r1051; + mul.f32 %r3699, %r3423, %r1052; + mul.f32 %r3700, %r3423, %r1053; + mul.f32 %r3701, %r3423, %r1054; + mul.f32 %r3702, %r3423, %r1055; + mul.f32 %r3703, %r3423, %r1056; + mul.f32 %r3704, %r3423, %r1057; + mul.f32 %r3705, %r3423, %r1058; + mul.f32 %r3706, %r3425, %r1059; + mul.f32 %r3707, %r3425, %r1060; + mul.f32 %r3708, %r3425, %r1061; + mul.f32 %r3709, %r3425, %r1062; + mul.f32 %r3710, %r3425, %r1063; + mul.f32 %r3711, %r3425, %r1064; + mul.f32 %r3712, %r3425, %r1065; + mul.f32 %r3713, %r3425, %r1066; + mul.f32 %r3714, %r3428, %r1067; + mul.f32 %r3715, %r3428, %r1068; + mul.f32 %r3716, %r3428, %r1069; + mul.f32 %r3717, %r3428, %r1070; + mul.f32 %r3718, %r3428, %r1071; + mul.f32 %r3719, %r3428, %r1072; + mul.f32 %r3720, %r3428, %r1073; + mul.f32 %r3721, %r3428, %r1074; + mul.f32 %r3722, %r3430, %r1075; + mul.f32 %r3723, %r3430, %r1076; + mul.f32 %r3724, %r3430, %r1077; + mul.f32 %r3725, %r3430, %r1078; + mul.f32 %r3726, %r3430, %r1079; + mul.f32 %r3727, %r3430, %r1080; + mul.f32 %r3728, %r3430, %r1081; + mul.f32 %r3729, %r3430, %r1082; + mul.f32 %r3730, %r3433, %r1083; + mul.f32 %r3731, %r3433, %r1084; + mul.f32 %r3732, %r3433, %r1085; + mul.f32 %r3733, %r3433, %r1086; + mul.f32 %r3734, %r3433, %r1087; + mul.f32 %r3735, %r3433, %r1088; + mul.f32 %r3736, %r3433, %r1089; + mul.f32 %r3737, %r3433, %r1090; + mul.f32 %r3738, %r3435, %r1091; + mul.f32 %r3739, %r3435, %r1092; + mul.f32 %r3740, %r3435, %r1093; + mul.f32 %r3741, %r3435, %r1094; + mul.f32 %r3742, %r3435, %r1095; + mul.f32 %r3743, %r3435, %r1096; + mul.f32 %r3744, %r3435, %r1097; + mul.f32 %r3745, %r3435, %r1098; + mul.f32 %r3746, %r3441, %r1099; + mul.f32 %r3747, %r3441, %r1100; + mul.f32 %r3748, %r3441, %r1101; + mul.f32 %r3749, %r3441, %r1102; + mul.f32 %r3750, %r3441, %r1103; + mul.f32 %r3751, %r3441, %r1104; + mul.f32 %r3752, %r3441, %r1105; + mul.f32 %r3753, %r3441, %r1106; + mul.f32 %r3754, %r3447, %r1107; + mul.f32 %r3755, %r3447, %r1108; + mul.f32 %r3756, %r3447, %r1109; + mul.f32 %r3757, %r3447, %r1110; + mul.f32 %r3758, %r3447, %r1111; + mul.f32 %r3759, %r3447, %r1112; + mul.f32 %r3760, %r3447, %r1113; + mul.f32 %r3761, %r3447, %r1114; + mul.f32 %r3762, %r3449, %r1115; + mul.f32 %r3763, %r3449, %r1116; + mul.f32 %r3764, %r3449, %r1117; + mul.f32 %r3765, %r3449, %r1118; + mul.f32 %r3766, %r3449, %r1119; + mul.f32 %r3767, %r3449, %r1120; + mul.f32 %r3768, %r3449, %r1121; + mul.f32 %r3769, %r3449, %r1122; + mul.f32 %r3770, %r3451, %r1123; + mul.f32 %r3771, %r3451, %r1124; + mul.f32 %r3772, %r3451, %r1125; + mul.f32 %r3773, %r3451, %r1126; + mul.f32 %r3774, %r3451, %r1127; + mul.f32 %r3775, %r3451, %r1128; + mul.f32 %r3776, %r3451, %r1129; + mul.f32 %r3777, %r3451, %r1130; + mul.f32 %r3778, %r3454, %r1131; + mul.f32 %r3779, %r3454, %r1132; + mul.f32 %r3780, %r3454, %r1133; + mul.f32 %r3781, %r3454, %r1134; + mul.f32 %r3782, %r3454, %r1135; + mul.f32 %r3783, %r3454, %r1136; + mul.f32 %r3784, %r3454, %r1137; + mul.f32 %r3785, %r3454, %r1138; + mul.f32 %r3786, %r3456, %r1139; + mul.f32 %r3787, %r3456, %r1140; + mul.f32 %r3788, %r3456, %r1141; + mul.f32 %r3789, %r3456, %r1142; + mul.f32 %r3790, %r3456, %r1143; + mul.f32 %r3791, %r3456, %r1144; + mul.f32 %r3792, %r3456, %r1145; + mul.f32 %r3793, %r3456, %r1146; + mul.f32 %r3794, %r3459, %r1147; + mul.f32 %r3795, %r3459, %r1148; + mul.f32 %r3796, %r3459, %r1149; + mul.f32 %r3797, %r3459, %r1150; + mul.f32 %r3798, %r3459, %r1151; + mul.f32 %r3799, %r3459, %r1152; + mul.f32 %r3800, %r3459, %r1153; + mul.f32 %r3801, %r3459, %r1154; + mul.f32 %r3802, %r3461, %r1155; + mul.f32 %r3803, %r3461, %r1156; + mul.f32 %r3804, %r3461, %r1157; + mul.f32 %r3805, %r3461, %r1158; + mul.f32 %r3806, %r3461, %r1159; + mul.f32 %r3807, %r3461, %r1160; + mul.f32 %r3808, %r3461, %r1161; + mul.f32 %r3809, %r3461, %r1162; + mul.f32 %r3810, %r3463, %r1163; + mul.f32 %r3811, %r3463, %r1164; + mul.f32 %r3812, %r3463, %r1165; + mul.f32 %r3813, %r3463, %r1166; + mul.f32 %r3814, %r3463, %r1167; + mul.f32 %r3815, %r3463, %r1168; + mul.f32 %r3816, %r3463, %r1169; + mul.f32 %r3817, %r3463, %r1170; + mul.f32 %r3818, %r3469, %r1171; + mul.f32 %r3819, %r3469, %r1172; + mul.f32 %r3820, %r3469, %r1173; + mul.f32 %r3821, %r3469, %r1174; + mul.f32 %r3822, %r3469, %r1175; + mul.f32 %r3823, %r3469, %r1176; + mul.f32 %r3824, %r3469, %r1177; + mul.f32 %r3825, %r3469, %r1178; + mul.f32 %r3826, %r3471, %r1179; + mul.f32 %r3827, %r3471, %r1180; + mul.f32 %r3828, %r3471, %r1181; + mul.f32 %r3829, %r3471, %r1182; + mul.f32 %r3830, %r3471, %r1183; + mul.f32 %r3831, %r3471, %r1184; + mul.f32 %r3832, %r3471, %r1185; + mul.f32 %r3833, %r3471, %r1186; + mul.f32 %r3834, %r3477, %r1187; + mul.f32 %r3835, %r3477, %r1188; + mul.f32 %r3836, %r3477, %r1189; + mul.f32 %r3837, %r3477, %r1190; + mul.f32 %r3838, %r3477, %r1191; + mul.f32 %r3839, %r3477, %r1192; + mul.f32 %r3840, %r3477, %r1193; + mul.f32 %r3841, %r3477, %r1194; + mul.f32 %r3842, %r3479, %r1195; + mul.f32 %r3843, %r3479, %r1196; + mul.f32 %r3844, %r3479, %r1197; + mul.f32 %r3845, %r3479, %r1198; + mul.f32 %r3846, %r3479, %r1199; + mul.f32 %r3847, %r3479, %r1200; + mul.f32 %r3848, %r3479, %r1201; + mul.f32 %r3849, %r3479, %r1202; + mul.f32 %r3850, %r3482, %r1203; + mul.f32 %r3851, %r3482, %r1204; + mul.f32 %r3852, %r3482, %r1205; + mul.f32 %r3853, %r3482, %r1206; + mul.f32 %r3854, %r3482, %r1207; + mul.f32 %r3855, %r3482, %r1208; + mul.f32 %r3856, %r3482, %r1209; + mul.f32 %r3857, %r3482, %r1210; + mul.f32 %r3858, %r3484, %r1211; + mul.f32 %r3859, %r3484, %r1212; + mul.f32 %r3860, %r3484, %r1213; + mul.f32 %r3861, %r3484, %r1214; + mul.f32 %r3862, %r3484, %r1215; + mul.f32 %r3863, %r3484, %r1216; + mul.f32 %r3864, %r3484, %r1217; + mul.f32 %r3865, %r3484, %r1218; + mul.f32 %r3866, %r3486, %r1219; + mul.f32 %r3867, %r3486, %r1220; + mul.f32 %r3868, %r3486, %r1221; + mul.f32 %r3869, %r3486, %r1222; + mul.f32 %r3870, %r3486, %r1223; + mul.f32 %r3871, %r3486, %r1224; + mul.f32 %r3872, %r3486, %r1225; + mul.f32 %r3873, %r3486, %r1226; + mul.f32 %r3874, %r3489, %r1227; + mul.f32 %r3875, %r3489, %r1228; + mul.f32 %r3876, %r3489, %r1229; + mul.f32 %r3877, %r3489, %r1230; + mul.f32 %r3878, %r3489, %r1231; + mul.f32 %r3879, %r3489, %r1232; + mul.f32 %r3880, %r3489, %r1233; + mul.f32 %r3881, %r3489, %r1234; + mul.f32 %r3882, %r3495, %r1235; + mul.f32 %r3883, %r3495, %r1236; + mul.f32 %r3884, %r3495, %r1237; + mul.f32 %r3885, %r3495, %r1238; + mul.f32 %r3886, %r3495, %r1239; + mul.f32 %r3887, %r3495, %r1240; + mul.f32 %r3888, %r3495, %r1241; + mul.f32 %r3889, %r3495, %r1242; + mul.f32 %r3890, %r3497, %r1243; + mul.f32 %r3891, %r3497, %r1244; + mul.f32 %r3892, %r3497, %r1245; + mul.f32 %r3893, %r3497, %r1246; + mul.f32 %r3894, %r3497, %r1247; + mul.f32 %r3895, %r3497, %r1248; + mul.f32 %r3896, %r3497, %r1249; + mul.f32 %r3897, %r3497, %r1250; + mul.f32 %r3898, %r3503, %r1251; + mul.f32 %r3899, %r3503, %r1252; + mul.f32 %r3900, %r3503, %r1253; + mul.f32 %r3901, %r3503, %r1254; + mul.f32 %r3902, %r3503, %r1255; + mul.f32 %r3903, %r3503, %r1256; + mul.f32 %r3904, %r3503, %r1257; + mul.f32 %r3905, %r3503, %r1258; + mul.f32 %r3906, %r3505, %r1259; + mul.f32 %r3907, %r3505, %r1260; + mul.f32 %r3908, %r3505, %r1261; + mul.f32 %r3909, %r3505, %r1262; + mul.f32 %r3910, %r3505, %r1263; + mul.f32 %r3911, %r3505, %r1264; + mul.f32 %r3912, %r3505, %r1265; + mul.f32 %r3913, %r3505, %r1266; + mul.f32 %r3914, %r3508, %r1267; + mul.f32 %r3915, %r3508, %r1268; + mul.f32 %r3916, %r3508, %r1269; + mul.f32 %r3917, %r3508, %r1270; + mul.f32 %r3918, %r3508, %r1271; + mul.f32 %r3919, %r3508, %r1272; + mul.f32 %r3920, %r3508, %r1273; + mul.f32 %r3921, %r3508, %r1274; + mul.f32 %r3922, %r3510, %r1275; + mul.f32 %r3923, %r3510, %r1276; + mul.f32 %r3924, %r3510, %r1277; + mul.f32 %r3925, %r3510, %r1278; + mul.f32 %r3926, %r3510, %r1279; + mul.f32 %r3927, %r3510, %r1280; + mul.f32 %r3928, %r3510, %r1281; + mul.f32 %r3929, %r3510, %r1282; + mul.f32 %r3930, %r3512, %r1283; + mul.f32 %r3931, %r3512, %r1284; + mul.f32 %r3932, %r3512, %r1285; + mul.f32 %r3933, %r3512, %r1286; + mul.f32 %r3934, %r3512, %r1287; + mul.f32 %r3935, %r3512, %r1288; + mul.f32 %r3936, %r3512, %r1289; + mul.f32 %r3937, %r3512, %r1290; + mul.f32 %r3938, %r3515, %r1291; + mul.f32 %r3939, %r3515, %r1292; + mul.f32 %r3940, %r3515, %r1293; + mul.f32 %r3941, %r3515, %r1294; + mul.f32 %r3942, %r3515, %r1295; + mul.f32 %r3943, %r3515, %r1296; + mul.f32 %r3944, %r3515, %r1297; + mul.f32 %r3945, %r3515, %r1298; + mul.f32 %r3946, %r3517, %r1299; + mul.f32 %r3947, %r3517, %r1300; + mul.f32 %r3948, %r3517, %r1301; + mul.f32 %r3949, %r3517, %r1302; + mul.f32 %r3950, %r3517, %r1303; + mul.f32 %r3951, %r3517, %r1304; + mul.f32 %r3952, %r3517, %r1305; + mul.f32 %r3953, %r3517, %r1306; + mul.f32 %r3954, %r3523, %r1307; + mul.f32 %r3955, %r3523, %r1308; + mul.f32 %r3956, %r3523, %r1309; + mul.f32 %r3957, %r3523, %r1310; + mul.f32 %r3958, %r3523, %r1311; + mul.f32 %r3959, %r3523, %r1312; + mul.f32 %r3960, %r3523, %r1313; + mul.f32 %r3961, %r3523, %r1314; + mul.f32 %r3962, %r3525, %r1315; + mul.f32 %r3963, %r3525, %r1316; + mul.f32 %r3964, %r3525, %r1317; + mul.f32 %r3965, %r3525, %r1318; + mul.f32 %r3966, %r3525, %r1319; + mul.f32 %r3967, %r3525, %r1320; + mul.f32 %r3968, %r3525, %r1321; + mul.f32 %r3969, %r3525, %r1322; + mul.f32 %r3970, %r3531, %r1323; + mul.f32 %r3971, %r3531, %r1324; + mul.f32 %r3972, %r3531, %r1325; + mul.f32 %r3973, %r3531, %r1326; + mul.f32 %r3974, %r3531, %r1327; + mul.f32 %r3975, %r3531, %r1328; + mul.f32 %r3976, %r3531, %r1329; + mul.f32 %r3977, %r3531, %r1330; + mul.f32 %r3978, %r3533, %r1331; + mul.f32 %r3979, %r3533, %r1332; + mul.f32 %r3980, %r3533, %r1333; + mul.f32 %r3981, %r3533, %r1334; + mul.f32 %r3982, %r3533, %r1335; + mul.f32 %r3983, %r3533, %r1336; + mul.f32 %r3984, %r3533, %r1337; + mul.f32 %r3985, %r3533, %r1338; + mul.f32 %r3986, %r3536, %r1339; + mul.f32 %r3987, %r3536, %r1340; + mul.f32 %r3988, %r3536, %r1341; + mul.f32 %r3989, %r3536, %r1342; + mul.f32 %r3990, %r3536, %r1343; + mul.f32 %r3991, %r3536, %r1344; + mul.f32 %r3992, %r3536, %r1345; + mul.f32 %r3993, %r3536, %r1346; + mul.f32 %r3994, %r3538, %r1347; + mul.f32 %r3995, %r3538, %r1348; + mul.f32 %r3996, %r3538, %r1349; + mul.f32 %r3997, %r3538, %r1350; + mul.f32 %r3998, %r3538, %r1351; + mul.f32 %r3999, %r3538, %r1352; + mul.f32 %r4000, %r3538, %r1353; + mul.f32 %r4001, %r3538, %r1354; + mul.f32 %r4002, %r3540, %r1355; + mul.f32 %r4003, %r3540, %r1356; + mul.f32 %r4004, %r3540, %r1357; + mul.f32 %r4005, %r3540, %r1358; + mul.f32 %r4006, %r3540, %r1359; + mul.f32 %r4007, %r3540, %r1360; + mul.f32 %r4008, %r3540, %r1361; + mul.f32 %r4009, %r3540, %r1362; + mul.f32 %r4010, %r3543, %r1363; + mul.f32 %r4011, %r3543, %r1364; + mul.f32 %r4012, %r3543, %r1365; + mul.f32 %r4013, %r3543, %r1366; + mul.f32 %r4014, %r3543, %r1367; + mul.f32 %r4015, %r3543, %r1368; + mul.f32 %r4016, %r3543, %r1369; + mul.f32 %r4017, %r3543, %r1370; + mul.f32 %r4018, %r3549, %r1371; + mul.f32 %r4019, %r3549, %r1372; + mul.f32 %r4020, %r3549, %r1373; + mul.f32 %r4021, %r3549, %r1374; + mul.f32 %r4022, %r3549, %r1375; + mul.f32 %r4023, %r3549, %r1376; + mul.f32 %r4024, %r3549, %r1377; + mul.f32 %r4025, %r3549, %r1378; + mul.f32 %r4026, %r3551, %r1379; + mul.f32 %r4027, %r3551, %r1380; + mul.f32 %r4028, %r3551, %r1381; + mul.f32 %r4029, %r3551, %r1382; + mul.f32 %r4030, %r3551, %r1383; + mul.f32 %r4031, %r3551, %r1384; + mul.f32 %r4032, %r3551, %r1385; + mul.f32 %r4033, %r3551, %r1386; + mul.f32 %r4034, %r3554, %r1387; + mul.f32 %r4035, %r3554, %r1388; + mul.f32 %r4036, %r3554, %r1389; + mul.f32 %r4037, %r3554, %r1390; + mul.f32 %r4038, %r3554, %r1391; + mul.f32 %r4039, %r3554, %r1392; + mul.f32 %r4040, %r3554, %r1393; + mul.f32 %r4041, %r3554, %r1394; + .loc 1 107 75 // l2norm.py:107:75 + mul.f32 %r4042, %r1763, %r3556; + mul.f32 %r4043, %r1763, %r3557; + mul.f32 %r4044, %r1763, %r3558; + mul.f32 %r4045, %r1763, %r3559; + mul.f32 %r4046, %r1763, %r3560; + mul.f32 %r4047, %r1763, %r3561; + mul.f32 %r4048, %r1763, %r3562; + mul.f32 %r4049, %r1763, %r3563; + mul.f32 %r4050, %r1771, %r3564; + mul.f32 %r4051, %r1771, %r3565; + mul.f32 %r4052, %r1771, %r3566; + mul.f32 %r4053, %r1771, %r3567; + mul.f32 %r4054, %r1764, %r3568; + mul.f32 %r4055, %r1764, %r3569; + mul.f32 %r4056, %r1764, %r3570; + mul.f32 %r4057, %r1764, %r3571; + mul.f32 %r4058, %r1772, %r3572; + mul.f32 %r4059, %r1772, %r3573; + mul.f32 %r4060, %r1772, %r3574; + mul.f32 %r4061, %r1772, %r3575; + mul.f32 %r4062, %r1765, %r3576; + mul.f32 %r4063, %r1765, %r3577; + mul.f32 %r4064, %r1765, %r3578; + mul.f32 %r4065, %r1765, %r3579; + mul.f32 %r4066, %r1773, %r3580; + mul.f32 %r4067, %r1773, %r3581; + mul.f32 %r4068, %r1773, %r3582; + mul.f32 %r4069, %r1773, %r3583; + mul.f32 %r4070, %r1766, %r3584; + mul.f32 %r4071, %r1766, %r3585; + mul.f32 %r4072, %r1766, %r3586; + mul.f32 %r4073, %r1766, %r3587; + mul.f32 %r4074, %r1774, %r3588; + mul.f32 %r4075, %r1774, %r3589; + mul.f32 %r4076, %r1774, %r3590; + mul.f32 %r4077, %r1774, %r3591; + mul.f32 %r4078, %r1774, %r3592; + mul.f32 %r4079, %r1774, %r3593; + mul.f32 %r4080, %r1735, %r3594; + mul.f32 %r4081, %r1735, %r3595; + mul.f32 %r4082, %r1735, %r3596; + mul.f32 %r4083, %r1735, %r3597; + mul.f32 %r4084, %r1735, %r3598; + mul.f32 %r4085, %r1735, %r3599; + mul.f32 %r4086, %r1735, %r3600; + mul.f32 %r4087, %r1735, %r3601; + mul.f32 %r4088, %r1819, %r3602; + mul.f32 %r4089, %r1819, %r3603; + mul.f32 %r4090, %r1819, %r3604; + mul.f32 %r4091, %r1819, %r3605; + mul.f32 %r4092, %r1819, %r3606; + mul.f32 %r4093, %r1819, %r3607; + mul.f32 %r4094, %r1819, %r3608; + mul.f32 %r4095, %r1819, %r3609; + mul.f32 %r4096, %r1736, %r3610; + mul.f32 %r4097, %r1736, %r3611; + mul.f32 %r4098, %r1736, %r3612; + mul.f32 %r4099, %r1736, %r3613; + mul.f32 %r4100, %r1736, %r3614; + mul.f32 %r4101, %r1736, %r3615; + mul.f32 %r4102, %r1736, %r3616; + mul.f32 %r4103, %r1736, %r3617; + mul.f32 %r4104, %r1820, %r3618; + mul.f32 %r4105, %r1820, %r3619; + mul.f32 %r4106, %r1820, %r3620; + mul.f32 %r4107, %r1820, %r3621; + mul.f32 %r4108, %r1820, %r3622; + mul.f32 %r4109, %r1820, %r3623; + mul.f32 %r4110, %r1820, %r3624; + mul.f32 %r4111, %r1820, %r3625; + mul.f32 %r4112, %r1737, %r3626; + mul.f32 %r4113, %r1737, %r3627; + mul.f32 %r4114, %r1737, %r3628; + mul.f32 %r4115, %r1737, %r3629; + mul.f32 %r4116, %r1737, %r3630; + mul.f32 %r4117, %r1737, %r3631; + mul.f32 %r4118, %r1737, %r3632; + mul.f32 %r4119, %r1737, %r3633; + mul.f32 %r4120, %r1821, %r3634; + mul.f32 %r4121, %r1821, %r3635; + mul.f32 %r4122, %r1821, %r3636; + mul.f32 %r4123, %r1821, %r3637; + mul.f32 %r4124, %r1821, %r3638; + mul.f32 %r4125, %r1821, %r3639; + mul.f32 %r4126, %r1821, %r3640; + mul.f32 %r4127, %r1821, %r3641; + mul.f32 %r4128, %r1738, %r3642; + mul.f32 %r4129, %r1738, %r3643; + mul.f32 %r4130, %r1738, %r3644; + mul.f32 %r4131, %r1738, %r3645; + mul.f32 %r4132, %r1738, %r3646; + mul.f32 %r4133, %r1738, %r3647; + mul.f32 %r4134, %r1738, %r3648; + mul.f32 %r4135, %r1738, %r3649; + mul.f32 %r4136, %r1822, %r3650; + mul.f32 %r4137, %r1822, %r3651; + mul.f32 %r4138, %r1822, %r3652; + mul.f32 %r4139, %r1822, %r3653; + mul.f32 %r4140, %r1822, %r3654; + mul.f32 %r4141, %r1822, %r3655; + mul.f32 %r4142, %r1822, %r3656; + mul.f32 %r4143, %r1822, %r3657; + mul.f32 %r4144, %r1739, %r3658; + mul.f32 %r4145, %r1739, %r3659; + mul.f32 %r4146, %r1739, %r3660; + mul.f32 %r4147, %r1739, %r3661; + mul.f32 %r4148, %r1739, %r3662; + mul.f32 %r4149, %r1739, %r3663; + mul.f32 %r4150, %r1739, %r3664; + mul.f32 %r4151, %r1739, %r3665; + mul.f32 %r4152, %r1824, %r3666; + mul.f32 %r4153, %r1824, %r3667; + mul.f32 %r4154, %r1824, %r3668; + mul.f32 %r4155, %r1824, %r3669; + mul.f32 %r4156, %r1824, %r3670; + mul.f32 %r4157, %r1824, %r3671; + mul.f32 %r4158, %r1824, %r3672; + mul.f32 %r4159, %r1824, %r3673; + mul.f32 %r4160, %r1740, %r3674; + mul.f32 %r4161, %r1740, %r3675; + mul.f32 %r4162, %r1740, %r3676; + mul.f32 %r4163, %r1740, %r3677; + mul.f32 %r4164, %r1740, %r3678; + mul.f32 %r4165, %r1740, %r3679; + mul.f32 %r4166, %r1740, %r3680; + mul.f32 %r4167, %r1740, %r3681; + mul.f32 %r4168, %r1825, %r3682; + mul.f32 %r4169, %r1825, %r3683; + mul.f32 %r4170, %r1825, %r3684; + mul.f32 %r4171, %r1825, %r3685; + mul.f32 %r4172, %r1825, %r3686; + mul.f32 %r4173, %r1825, %r3687; + mul.f32 %r4174, %r1825, %r3688; + mul.f32 %r4175, %r1825, %r3689; + mul.f32 %r4176, %r1741, %r3690; + mul.f32 %r4177, %r1741, %r3691; + mul.f32 %r4178, %r1741, %r3692; + mul.f32 %r4179, %r1741, %r3693; + mul.f32 %r4180, %r1741, %r3694; + mul.f32 %r4181, %r1741, %r3695; + mul.f32 %r4182, %r1741, %r3696; + mul.f32 %r4183, %r1741, %r3697; + mul.f32 %r4184, %r1826, %r3698; + mul.f32 %r4185, %r1826, %r3699; + mul.f32 %r4186, %r1826, %r3700; + mul.f32 %r4187, %r1826, %r3701; + mul.f32 %r4188, %r1826, %r3702; + mul.f32 %r4189, %r1826, %r3703; + mul.f32 %r4190, %r1826, %r3704; + mul.f32 %r4191, %r1826, %r3705; + mul.f32 %r4192, %r1742, %r3706; + mul.f32 %r4193, %r1742, %r3707; + mul.f32 %r4194, %r1742, %r3708; + mul.f32 %r4195, %r1742, %r3709; + mul.f32 %r4196, %r1742, %r3710; + mul.f32 %r4197, %r1742, %r3711; + mul.f32 %r4198, %r1742, %r3712; + mul.f32 %r4199, %r1742, %r3713; + mul.f32 %r4200, %r1827, %r3714; + mul.f32 %r4201, %r1827, %r3715; + mul.f32 %r4202, %r1827, %r3716; + mul.f32 %r4203, %r1827, %r3717; + mul.f32 %r4204, %r1827, %r3718; + mul.f32 %r4205, %r1827, %r3719; + mul.f32 %r4206, %r1827, %r3720; + mul.f32 %r4207, %r1827, %r3721; + mul.f32 %r4208, %r1743, %r3722; + mul.f32 %r4209, %r1743, %r3723; + mul.f32 %r4210, %r1743, %r3724; + mul.f32 %r4211, %r1743, %r3725; + mul.f32 %r4212, %r1743, %r3726; + mul.f32 %r4213, %r1743, %r3727; + mul.f32 %r4214, %r1743, %r3728; + mul.f32 %r4215, %r1743, %r3729; + mul.f32 %r4216, %r1829, %r3730; + mul.f32 %r4217, %r1829, %r3731; + mul.f32 %r4218, %r1829, %r3732; + mul.f32 %r4219, %r1829, %r3733; + mul.f32 %r4220, %r1829, %r3734; + mul.f32 %r4221, %r1829, %r3735; + mul.f32 %r4222, %r1829, %r3736; + mul.f32 %r4223, %r1829, %r3737; + mul.f32 %r4224, %r1744, %r3738; + mul.f32 %r4225, %r1744, %r3739; + mul.f32 %r4226, %r1744, %r3740; + mul.f32 %r4227, %r1744, %r3741; + mul.f32 %r4228, %r1744, %r3742; + mul.f32 %r4229, %r1744, %r3743; + mul.f32 %r4230, %r1744, %r3744; + mul.f32 %r4231, %r1744, %r3745; + mul.f32 %r4232, %r1830, %r3746; + mul.f32 %r4233, %r1830, %r3747; + mul.f32 %r4234, %r1830, %r3748; + mul.f32 %r4235, %r1830, %r3749; + mul.f32 %r4236, %r1830, %r3750; + mul.f32 %r4237, %r1830, %r3751; + mul.f32 %r4238, %r1830, %r3752; + mul.f32 %r4239, %r1830, %r3753; + mul.f32 %r4240, %r1745, %r3754; + mul.f32 %r4241, %r1745, %r3755; + mul.f32 %r4242, %r1745, %r3756; + mul.f32 %r4243, %r1745, %r3757; + mul.f32 %r4244, %r1745, %r3758; + mul.f32 %r4245, %r1745, %r3759; + mul.f32 %r4246, %r1745, %r3760; + mul.f32 %r4247, %r1745, %r3761; + mul.f32 %r4248, %r1831, %r3762; + mul.f32 %r4249, %r1831, %r3763; + mul.f32 %r4250, %r1831, %r3764; + mul.f32 %r4251, %r1831, %r3765; + mul.f32 %r4252, %r1831, %r3766; + mul.f32 %r4253, %r1831, %r3767; + mul.f32 %r4254, %r1831, %r3768; + mul.f32 %r4255, %r1831, %r3769; + mul.f32 %r4256, %r1746, %r3770; + mul.f32 %r4257, %r1746, %r3771; + mul.f32 %r4258, %r1746, %r3772; + mul.f32 %r4259, %r1746, %r3773; + mul.f32 %r4260, %r1746, %r3774; + mul.f32 %r4261, %r1746, %r3775; + mul.f32 %r4262, %r1746, %r3776; + mul.f32 %r4263, %r1746, %r3777; + mul.f32 %r4264, %r1832, %r3778; + mul.f32 %r4265, %r1832, %r3779; + mul.f32 %r4266, %r1832, %r3780; + mul.f32 %r4267, %r1832, %r3781; + mul.f32 %r4268, %r1832, %r3782; + mul.f32 %r4269, %r1832, %r3783; + mul.f32 %r4270, %r1832, %r3784; + mul.f32 %r4271, %r1832, %r3785; + mul.f32 %r4272, %r1747, %r3786; + mul.f32 %r4273, %r1747, %r3787; + mul.f32 %r4274, %r1747, %r3788; + mul.f32 %r4275, %r1747, %r3789; + mul.f32 %r4276, %r1747, %r3790; + mul.f32 %r4277, %r1747, %r3791; + mul.f32 %r4278, %r1747, %r3792; + mul.f32 %r4279, %r1747, %r3793; + mul.f32 %r4280, %r1834, %r3794; + mul.f32 %r4281, %r1834, %r3795; + mul.f32 %r4282, %r1834, %r3796; + mul.f32 %r4283, %r1834, %r3797; + mul.f32 %r4284, %r1834, %r3798; + mul.f32 %r4285, %r1834, %r3799; + mul.f32 %r4286, %r1834, %r3800; + mul.f32 %r4287, %r1834, %r3801; + mul.f32 %r4288, %r1748, %r3802; + mul.f32 %r4289, %r1748, %r3803; + mul.f32 %r4290, %r1748, %r3804; + mul.f32 %r4291, %r1748, %r3805; + mul.f32 %r4292, %r1748, %r3806; + mul.f32 %r4293, %r1748, %r3807; + mul.f32 %r4294, %r1748, %r3808; + mul.f32 %r4295, %r1748, %r3809; + mul.f32 %r4296, %r1835, %r3810; + mul.f32 %r4297, %r1835, %r3811; + mul.f32 %r4298, %r1835, %r3812; + mul.f32 %r4299, %r1835, %r3813; + mul.f32 %r4300, %r1835, %r3814; + mul.f32 %r4301, %r1835, %r3815; + mul.f32 %r4302, %r1835, %r3816; + mul.f32 %r4303, %r1835, %r3817; + mul.f32 %r4304, %r1749, %r3818; + mul.f32 %r4305, %r1749, %r3819; + mul.f32 %r4306, %r1749, %r3820; + mul.f32 %r4307, %r1749, %r3821; + mul.f32 %r4308, %r1749, %r3822; + mul.f32 %r4309, %r1749, %r3823; + mul.f32 %r4310, %r1749, %r3824; + mul.f32 %r4311, %r1749, %r3825; + mul.f32 %r4312, %r1836, %r3826; + mul.f32 %r4313, %r1836, %r3827; + mul.f32 %r4314, %r1836, %r3828; + mul.f32 %r4315, %r1836, %r3829; + mul.f32 %r4316, %r1836, %r3830; + mul.f32 %r4317, %r1836, %r3831; + mul.f32 %r4318, %r1836, %r3832; + mul.f32 %r4319, %r1836, %r3833; + mul.f32 %r4320, %r1750, %r3834; + mul.f32 %r4321, %r1750, %r3835; + mul.f32 %r4322, %r1750, %r3836; + mul.f32 %r4323, %r1750, %r3837; + mul.f32 %r4324, %r1750, %r3838; + mul.f32 %r4325, %r1750, %r3839; + mul.f32 %r4326, %r1750, %r3840; + mul.f32 %r4327, %r1750, %r3841; + mul.f32 %r4328, %r1837, %r3842; + mul.f32 %r4329, %r1837, %r3843; + mul.f32 %r4330, %r1837, %r3844; + mul.f32 %r4331, %r1837, %r3845; + mul.f32 %r4332, %r1837, %r3846; + mul.f32 %r4333, %r1837, %r3847; + mul.f32 %r4334, %r1837, %r3848; + mul.f32 %r4335, %r1837, %r3849; + mul.f32 %r4336, %r1751, %r3850; + mul.f32 %r4337, %r1751, %r3851; + mul.f32 %r4338, %r1751, %r3852; + mul.f32 %r4339, %r1751, %r3853; + mul.f32 %r4340, %r1751, %r3854; + mul.f32 %r4341, %r1751, %r3855; + mul.f32 %r4342, %r1751, %r3856; + mul.f32 %r4343, %r1751, %r3857; + mul.f32 %r4344, %r1839, %r3858; + mul.f32 %r4345, %r1839, %r3859; + mul.f32 %r4346, %r1839, %r3860; + mul.f32 %r4347, %r1839, %r3861; + mul.f32 %r4348, %r1839, %r3862; + mul.f32 %r4349, %r1839, %r3863; + mul.f32 %r4350, %r1839, %r3864; + mul.f32 %r4351, %r1839, %r3865; + mul.f32 %r4352, %r1752, %r3866; + mul.f32 %r4353, %r1752, %r3867; + mul.f32 %r4354, %r1752, %r3868; + mul.f32 %r4355, %r1752, %r3869; + mul.f32 %r4356, %r1752, %r3870; + mul.f32 %r4357, %r1752, %r3871; + mul.f32 %r4358, %r1752, %r3872; + mul.f32 %r4359, %r1752, %r3873; + mul.f32 %r4360, %r1840, %r3874; + mul.f32 %r4361, %r1840, %r3875; + mul.f32 %r4362, %r1840, %r3876; + mul.f32 %r4363, %r1840, %r3877; + mul.f32 %r4364, %r1840, %r3878; + mul.f32 %r4365, %r1840, %r3879; + mul.f32 %r4366, %r1840, %r3880; + mul.f32 %r4367, %r1840, %r3881; + mul.f32 %r4368, %r1753, %r3882; + mul.f32 %r4369, %r1753, %r3883; + mul.f32 %r4370, %r1753, %r3884; + mul.f32 %r4371, %r1753, %r3885; + mul.f32 %r4372, %r1753, %r3886; + mul.f32 %r4373, %r1753, %r3887; + mul.f32 %r4374, %r1753, %r3888; + mul.f32 %r4375, %r1753, %r3889; + mul.f32 %r4376, %r1841, %r3890; + mul.f32 %r4377, %r1841, %r3891; + mul.f32 %r4378, %r1841, %r3892; + mul.f32 %r4379, %r1841, %r3893; + mul.f32 %r4380, %r1841, %r3894; + mul.f32 %r4381, %r1841, %r3895; + mul.f32 %r4382, %r1841, %r3896; + mul.f32 %r4383, %r1841, %r3897; + mul.f32 %r4384, %r1754, %r3898; + mul.f32 %r4385, %r1754, %r3899; + mul.f32 %r4386, %r1754, %r3900; + mul.f32 %r4387, %r1754, %r3901; + mul.f32 %r4388, %r1754, %r3902; + mul.f32 %r4389, %r1754, %r3903; + mul.f32 %r4390, %r1754, %r3904; + mul.f32 %r4391, %r1754, %r3905; + mul.f32 %r4392, %r1842, %r3906; + mul.f32 %r4393, %r1842, %r3907; + mul.f32 %r4394, %r1842, %r3908; + mul.f32 %r4395, %r1842, %r3909; + mul.f32 %r4396, %r1842, %r3910; + mul.f32 %r4397, %r1842, %r3911; + mul.f32 %r4398, %r1842, %r3912; + mul.f32 %r4399, %r1842, %r3913; + mul.f32 %r4400, %r1755, %r3914; + mul.f32 %r4401, %r1755, %r3915; + mul.f32 %r4402, %r1755, %r3916; + mul.f32 %r4403, %r1755, %r3917; + mul.f32 %r4404, %r1755, %r3918; + mul.f32 %r4405, %r1755, %r3919; + mul.f32 %r4406, %r1755, %r3920; + mul.f32 %r4407, %r1755, %r3921; + mul.f32 %r4408, %r1844, %r3922; + mul.f32 %r4409, %r1844, %r3923; + mul.f32 %r4410, %r1844, %r3924; + mul.f32 %r4411, %r1844, %r3925; + mul.f32 %r4412, %r1844, %r3926; + mul.f32 %r4413, %r1844, %r3927; + mul.f32 %r4414, %r1844, %r3928; + mul.f32 %r4415, %r1844, %r3929; + mul.f32 %r4416, %r1756, %r3930; + mul.f32 %r4417, %r1756, %r3931; + mul.f32 %r4418, %r1756, %r3932; + mul.f32 %r4419, %r1756, %r3933; + mul.f32 %r4420, %r1756, %r3934; + mul.f32 %r4421, %r1756, %r3935; + mul.f32 %r4422, %r1756, %r3936; + mul.f32 %r4423, %r1756, %r3937; + mul.f32 %r4424, %r1845, %r3938; + mul.f32 %r4425, %r1845, %r3939; + mul.f32 %r4426, %r1845, %r3940; + mul.f32 %r4427, %r1845, %r3941; + mul.f32 %r4428, %r1845, %r3942; + mul.f32 %r4429, %r1845, %r3943; + mul.f32 %r4430, %r1845, %r3944; + mul.f32 %r4431, %r1845, %r3945; + mul.f32 %r4432, %r1757, %r3946; + mul.f32 %r4433, %r1757, %r3947; + mul.f32 %r4434, %r1757, %r3948; + mul.f32 %r4435, %r1757, %r3949; + mul.f32 %r4436, %r1757, %r3950; + mul.f32 %r4437, %r1757, %r3951; + mul.f32 %r4438, %r1757, %r3952; + mul.f32 %r4439, %r1757, %r3953; + mul.f32 %r4440, %r1846, %r3954; + mul.f32 %r4441, %r1846, %r3955; + mul.f32 %r4442, %r1846, %r3956; + mul.f32 %r4443, %r1846, %r3957; + mul.f32 %r4444, %r1846, %r3958; + mul.f32 %r4445, %r1846, %r3959; + mul.f32 %r4446, %r1846, %r3960; + mul.f32 %r4447, %r1846, %r3961; + mul.f32 %r4448, %r1758, %r3962; + mul.f32 %r4449, %r1758, %r3963; + mul.f32 %r4450, %r1758, %r3964; + mul.f32 %r4451, %r1758, %r3965; + mul.f32 %r4452, %r1758, %r3966; + mul.f32 %r4453, %r1758, %r3967; + mul.f32 %r4454, %r1758, %r3968; + mul.f32 %r4455, %r1758, %r3969; + mul.f32 %r4456, %r1847, %r3970; + mul.f32 %r4457, %r1847, %r3971; + mul.f32 %r4458, %r1847, %r3972; + mul.f32 %r4459, %r1847, %r3973; + mul.f32 %r4460, %r1847, %r3974; + mul.f32 %r4461, %r1847, %r3975; + mul.f32 %r4462, %r1847, %r3976; + mul.f32 %r4463, %r1847, %r3977; + mul.f32 %r4464, %r1759, %r3978; + mul.f32 %r4465, %r1759, %r3979; + mul.f32 %r4466, %r1759, %r3980; + mul.f32 %r4467, %r1759, %r3981; + mul.f32 %r4468, %r1759, %r3982; + mul.f32 %r4469, %r1759, %r3983; + mul.f32 %r4470, %r1759, %r3984; + mul.f32 %r4471, %r1759, %r3985; + mul.f32 %r4472, %r1849, %r3986; + mul.f32 %r4473, %r1849, %r3987; + mul.f32 %r4474, %r1849, %r3988; + mul.f32 %r4475, %r1849, %r3989; + mul.f32 %r4476, %r1849, %r3990; + mul.f32 %r4477, %r1849, %r3991; + mul.f32 %r4478, %r1849, %r3992; + mul.f32 %r4479, %r1849, %r3993; + mul.f32 %r4480, %r1760, %r3994; + mul.f32 %r4481, %r1760, %r3995; + mul.f32 %r4482, %r1760, %r3996; + mul.f32 %r4483, %r1760, %r3997; + mul.f32 %r4484, %r1760, %r3998; + mul.f32 %r4485, %r1760, %r3999; + mul.f32 %r4486, %r1760, %r4000; + mul.f32 %r4487, %r1760, %r4001; + mul.f32 %r4488, %r1850, %r4002; + mul.f32 %r4489, %r1850, %r4003; + mul.f32 %r4490, %r1850, %r4004; + mul.f32 %r4491, %r1850, %r4005; + mul.f32 %r4492, %r1850, %r4006; + mul.f32 %r4493, %r1850, %r4007; + mul.f32 %r4494, %r1850, %r4008; + mul.f32 %r4495, %r1850, %r4009; + mul.f32 %r4496, %r1761, %r4010; + mul.f32 %r4497, %r1761, %r4011; + mul.f32 %r4498, %r1761, %r4012; + mul.f32 %r4499, %r1761, %r4013; + mul.f32 %r4500, %r1761, %r4014; + mul.f32 %r4501, %r1761, %r4015; + mul.f32 %r4502, %r1761, %r4016; + mul.f32 %r4503, %r1761, %r4017; + mul.f32 %r4504, %r1851, %r4018; + mul.f32 %r4505, %r1851, %r4019; + mul.f32 %r4506, %r1851, %r4020; + mul.f32 %r4507, %r1851, %r4021; + mul.f32 %r4508, %r1851, %r4022; + mul.f32 %r4509, %r1851, %r4023; + mul.f32 %r4510, %r1851, %r4024; + mul.f32 %r4511, %r1851, %r4025; + mul.f32 %r4512, %r1762, %r4026; + mul.f32 %r4513, %r1762, %r4027; + mul.f32 %r4514, %r1762, %r4028; + mul.f32 %r4515, %r1762, %r4029; + mul.f32 %r4516, %r1762, %r4030; + mul.f32 %r4517, %r1762, %r4031; + mul.f32 %r4518, %r1762, %r4032; + mul.f32 %r4519, %r1762, %r4033; + mul.f32 %r4520, %r1852, %r4034; + mul.f32 %r4521, %r1852, %r4035; + mul.f32 %r4522, %r1852, %r4036; + mul.f32 %r4523, %r1852, %r4037; + mul.f32 %r4524, %r1852, %r4038; + mul.f32 %r4525, %r1852, %r4039; + mul.f32 %r4526, %r1852, %r4040; + mul.f32 %r4527, %r1852, %r4041; + .loc 1 107 36 // l2norm.py:107:36 + neg.f32 %r4528, %r4042; + neg.f32 %r4529, %r4043; + fma.rn.f32 %r4530, %r1763, %r1395, %r4528; + fma.rn.f32 %r4531, %r1763, %r1396, %r4529; + neg.f32 %r4532, %r4044; + neg.f32 %r4533, %r4045; + fma.rn.f32 %r4534, %r1763, %r1397, %r4532; + fma.rn.f32 %r4535, %r1763, %r1398, %r4533; + neg.f32 %r4536, %r4046; + neg.f32 %r4537, %r4047; + fma.rn.f32 %r4538, %r1763, %r1399, %r4536; + fma.rn.f32 %r4539, %r1763, %r1400, %r4537; + neg.f32 %r4540, %r4048; + neg.f32 %r4541, %r4049; + fma.rn.f32 %r4542, %r1763, %r1401, %r4540; + fma.rn.f32 %r4543, %r1763, %r1402, %r4541; + mov.b64 %rd689, {%r3555, %r4050}; + mov.b64 %rd690, {%r3547, %r4051}; + sub.f32x2 %rd691, %rd688, %rd689; + sub.f32x2 %rd692, %rd687, %rd690; + mov.b64 %rd693, {%r3529, %r4052}; + mov.b64 %rd694, {%r3521, %r4053}; + sub.f32x2 %rd695, %rd686, %rd693; + sub.f32x2 %rd696, %rd685, %rd694; + mov.b64 %rd697, {%r3501, %r4054}; + mov.b64 %rd698, {%r3493, %r4055}; + sub.f32x2 %rd699, %rd684, %rd697; + sub.f32x2 %rd700, %rd683, %rd698; + mov.b64 %rd701, {%r3475, %r4056}; + mov.b64 %rd702, {%r3467, %r4057}; + sub.f32x2 %rd703, %rd682, %rd701; + sub.f32x2 %rd704, %rd681, %rd702; + mov.b64 %rd705, {%r3445, %r4058}; + mov.b64 %rd706, {%r3439, %r4059}; + sub.f32x2 %rd707, %rd680, %rd705; + sub.f32x2 %rd708, %rd679, %rd706; + mov.b64 %rd709, {%r3419, %r4060}; + mov.b64 %rd710, {%r3413, %r4061}; + sub.f32x2 %rd711, %rd678, %rd709; + sub.f32x2 %rd712, %rd677, %rd710; + mov.b64 %rd713, {%r3391, %r4062}; + mov.b64 %rd714, {%r3385, %r4063}; + sub.f32x2 %rd715, %rd676, %rd713; + sub.f32x2 %rd716, %rd675, %rd714; + mov.b64 %rd717, {%r2072, %r4064}; + mov.b64 %rd718, {%r2069, %r4065}; + sub.f32x2 %rd719, %rd674, %rd717; + sub.f32x2 %rd720, %rd673, %rd718; + mov.b64 %rd721, {%r2051, %r4066}; + mov.b64 %rd722, {%r2048, %r4067}; + sub.f32x2 %rd723, %rd672, %rd721; + sub.f32x2 %rd724, %rd671, %rd722; + mov.b64 %rd725, {%r2033, %r4068}; + mov.b64 %rd726, {%r2030, %r4069}; + sub.f32x2 %rd727, %rd670, %rd725; + sub.f32x2 %rd728, %rd669, %rd726; + mov.b64 %rd729, {%r2012, %r4070}; + mov.b64 %rd730, {%r2009, %r4071}; + sub.f32x2 %rd731, %rd668, %rd729; + sub.f32x2 %rd732, %rd667, %rd730; + mov.b64 %rd733, {%r1994, %r4072}; + mov.b64 %rd734, {%r1991, %r4073}; + sub.f32x2 %rd735, %rd666, %rd733; + sub.f32x2 %rd736, %rd665, %rd734; + mov.b64 %rd737, {%r1973, %r4074}; + mov.b64 %rd738, {%r1970, %r4075}; + sub.f32x2 %rd739, %rd664, %rd737; + sub.f32x2 %rd740, %rd663, %rd738; + mov.b64 %rd741, {%r1780, %r2337}; + mov.b64 %rd742, {%r4076, %r4077}; + sub.f32x2 %rd743, %rd741, %rd742; + mov.b64 %rd744, {%r2338, %r2339}; + mov.b64 %rd745, {%r4078, %r4079}; + sub.f32x2 %rd746, %rd744, %rd745; + mov.b64 %rd747, {%r2340, %r2341}; + mov.b64 %rd748, {%r4080, %r4081}; + sub.f32x2 %rd749, %rd747, %rd748; + mov.b64 %rd750, {%r2342, %r2343}; + mov.b64 %rd751, {%r4082, %r4083}; + sub.f32x2 %rd752, %rd750, %rd751; + mov.b64 %rd753, {%r2344, %r2345}; + mov.b64 %rd754, {%r4084, %r4085}; + sub.f32x2 %rd755, %rd753, %rd754; + mov.b64 %rd756, {%r2346, %r2347}; + mov.b64 %rd757, {%r4086, %r4087}; + sub.f32x2 %rd758, %rd756, %rd757; + mov.b64 %rd759, {%r2348, %r2349}; + mov.b64 %rd760, {%r4088, %r4089}; + sub.f32x2 %rd761, %rd759, %rd760; + mov.b64 %rd762, {%r2350, %r2351}; + mov.b64 %rd763, {%r4090, %r4091}; + sub.f32x2 %rd764, %rd762, %rd763; + mov.b64 %rd765, {%r2352, %r2353}; + mov.b64 %rd766, {%r4092, %r4093}; + sub.f32x2 %rd767, %rd765, %rd766; + mov.b64 %rd768, {%r2354, %r2355}; + mov.b64 %rd769, {%r4094, %r4095}; + sub.f32x2 %rd770, %rd768, %rd769; + mov.b64 %rd771, {%r2356, %r2357}; + mov.b64 %rd772, {%r4096, %r4097}; + sub.f32x2 %rd773, %rd771, %rd772; + mov.b64 %rd774, {%r2358, %r2359}; + mov.b64 %rd775, {%r4098, %r4099}; + sub.f32x2 %rd776, %rd774, %rd775; + mov.b64 %rd777, {%r2360, %r2361}; + mov.b64 %rd778, {%r4100, %r4101}; + sub.f32x2 %rd779, %rd777, %rd778; + mov.b64 %rd780, {%r2362, %r2363}; + mov.b64 %rd781, {%r4102, %r4103}; + sub.f32x2 %rd782, %rd780, %rd781; + mov.b64 %rd783, {%r2364, %r2365}; + mov.b64 %rd784, {%r4104, %r4105}; + sub.f32x2 %rd785, %rd783, %rd784; + mov.b64 %rd786, {%r2366, %r2367}; + mov.b64 %rd787, {%r4106, %r4107}; + sub.f32x2 %rd788, %rd786, %rd787; + mov.b64 %rd789, {%r2368, %r2369}; + mov.b64 %rd790, {%r4108, %r4109}; + sub.f32x2 %rd791, %rd789, %rd790; + mov.b64 %rd792, {%r2370, %r2371}; + mov.b64 %rd793, {%r4110, %r4111}; + sub.f32x2 %rd794, %rd792, %rd793; + mov.b64 %rd795, {%r2372, %r2373}; + mov.b64 %rd796, {%r4112, %r4113}; + sub.f32x2 %rd797, %rd795, %rd796; + mov.b64 %rd798, {%r2374, %r2375}; + mov.b64 %rd799, {%r4114, %r4115}; + sub.f32x2 %rd800, %rd798, %rd799; + mov.b64 %rd801, {%r2376, %r2377}; + mov.b64 %rd802, {%r4116, %r4117}; + sub.f32x2 %rd803, %rd801, %rd802; + mov.b64 %rd804, {%r2378, %r2379}; + mov.b64 %rd805, {%r4118, %r4119}; + sub.f32x2 %rd806, %rd804, %rd805; + mov.b64 %rd807, {%r2380, %r2381}; + mov.b64 %rd808, {%r4120, %r4121}; + sub.f32x2 %rd809, %rd807, %rd808; + mov.b64 %rd810, {%r2382, %r2383}; + mov.b64 %rd811, {%r4122, %r4123}; + sub.f32x2 %rd812, %rd810, %rd811; + mov.b64 %rd813, {%r2384, %r2385}; + mov.b64 %rd814, {%r4124, %r4125}; + sub.f32x2 %rd815, %rd813, %rd814; + mov.b64 %rd816, {%r2386, %r2387}; + mov.b64 %rd817, {%r4126, %r4127}; + sub.f32x2 %rd818, %rd816, %rd817; + mov.b64 %rd819, {%r2388, %r2389}; + mov.b64 %rd820, {%r4128, %r4129}; + sub.f32x2 %rd821, %rd819, %rd820; + mov.b64 %rd822, {%r2390, %r2391}; + mov.b64 %rd823, {%r4130, %r4131}; + sub.f32x2 %rd824, %rd822, %rd823; + mov.b64 %rd825, {%r2392, %r2393}; + mov.b64 %rd826, {%r4132, %r4133}; + sub.f32x2 %rd827, %rd825, %rd826; + mov.b64 %rd828, {%r2394, %r2395}; + mov.b64 %rd829, {%r4134, %r4135}; + sub.f32x2 %rd830, %rd828, %rd829; + mov.b64 %rd831, {%r2396, %r2397}; + mov.b64 %rd832, {%r4136, %r4137}; + sub.f32x2 %rd833, %rd831, %rd832; + mov.b64 %rd834, {%r2398, %r2399}; + mov.b64 %rd835, {%r4138, %r4139}; + sub.f32x2 %rd836, %rd834, %rd835; + mov.b64 %rd837, {%r2400, %r2401}; + mov.b64 %rd838, {%r4140, %r4141}; + sub.f32x2 %rd839, %rd837, %rd838; + mov.b64 %rd840, {%r2402, %r2403}; + mov.b64 %rd841, {%r4142, %r4143}; + sub.f32x2 %rd842, %rd840, %rd841; + mov.b64 %rd843, {%r2404, %r2405}; + mov.b64 %rd844, {%r4144, %r4145}; + sub.f32x2 %rd845, %rd843, %rd844; + mov.b64 %rd846, {%r2406, %r2407}; + mov.b64 %rd847, {%r4146, %r4147}; + sub.f32x2 %rd848, %rd846, %rd847; + mov.b64 %rd849, {%r2408, %r2409}; + mov.b64 %rd850, {%r4148, %r4149}; + sub.f32x2 %rd851, %rd849, %rd850; + mov.b64 %rd852, {%r2410, %r2411}; + mov.b64 %rd853, {%r4150, %r4151}; + sub.f32x2 %rd854, %rd852, %rd853; + mov.b64 %rd855, {%r2412, %r2413}; + mov.b64 %rd856, {%r4152, %r4153}; + sub.f32x2 %rd857, %rd855, %rd856; + mov.b64 %rd858, {%r2414, %r2415}; + mov.b64 %rd859, {%r4154, %r4155}; + sub.f32x2 %rd860, %rd858, %rd859; + mov.b64 %rd861, {%r2416, %r2417}; + mov.b64 %rd862, {%r4156, %r4157}; + sub.f32x2 %rd863, %rd861, %rd862; + mov.b64 %rd864, {%r2418, %r2419}; + mov.b64 %rd865, {%r4158, %r4159}; + sub.f32x2 %rd866, %rd864, %rd865; + mov.b64 %rd867, {%r2420, %r2421}; + mov.b64 %rd868, {%r4160, %r4161}; + sub.f32x2 %rd869, %rd867, %rd868; + mov.b64 %rd870, {%r2422, %r2423}; + mov.b64 %rd871, {%r4162, %r4163}; + sub.f32x2 %rd872, %rd870, %rd871; + mov.b64 %rd873, {%r2424, %r2425}; + mov.b64 %rd874, {%r4164, %r4165}; + sub.f32x2 %rd875, %rd873, %rd874; + mov.b64 %rd876, {%r2426, %r2427}; + mov.b64 %rd877, {%r4166, %r4167}; + sub.f32x2 %rd878, %rd876, %rd877; + mov.b64 %rd879, {%r2428, %r2429}; + mov.b64 %rd880, {%r4168, %r4169}; + sub.f32x2 %rd881, %rd879, %rd880; + mov.b64 %rd882, {%r2430, %r2431}; + mov.b64 %rd883, {%r4170, %r4171}; + sub.f32x2 %rd884, %rd882, %rd883; + mov.b64 %rd885, {%r2432, %r2433}; + mov.b64 %rd886, {%r4172, %r4173}; + sub.f32x2 %rd887, %rd885, %rd886; + mov.b64 %rd888, {%r2434, %r2435}; + mov.b64 %rd889, {%r4174, %r4175}; + sub.f32x2 %rd890, %rd888, %rd889; + mov.b64 %rd891, {%r2436, %r2437}; + mov.b64 %rd892, {%r4176, %r4177}; + sub.f32x2 %rd893, %rd891, %rd892; + mov.b64 %rd894, {%r2438, %r2439}; + mov.b64 %rd895, {%r4178, %r4179}; + sub.f32x2 %rd896, %rd894, %rd895; + mov.b64 %rd897, {%r2440, %r2441}; + mov.b64 %rd898, {%r4180, %r4181}; + sub.f32x2 %rd899, %rd897, %rd898; + mov.b64 %rd900, {%r2442, %r2443}; + mov.b64 %rd901, {%r4182, %r4183}; + sub.f32x2 %rd902, %rd900, %rd901; + mov.b64 %rd903, {%r2444, %r2445}; + mov.b64 %rd904, {%r4184, %r4185}; + sub.f32x2 %rd905, %rd903, %rd904; + mov.b64 %rd906, {%r2446, %r2447}; + mov.b64 %rd907, {%r4186, %r4187}; + sub.f32x2 %rd908, %rd906, %rd907; + mov.b64 %rd909, {%r2448, %r2449}; + mov.b64 %rd910, {%r4188, %r4189}; + sub.f32x2 %rd911, %rd909, %rd910; + mov.b64 %rd912, {%r2450, %r2451}; + mov.b64 %rd913, {%r4190, %r4191}; + sub.f32x2 %rd914, %rd912, %rd913; + mov.b64 %rd915, {%r2452, %r2453}; + mov.b64 %rd916, {%r4192, %r4193}; + sub.f32x2 %rd917, %rd915, %rd916; + mov.b64 %rd918, {%r2454, %r2455}; + mov.b64 %rd919, {%r4194, %r4195}; + sub.f32x2 %rd920, %rd918, %rd919; + mov.b64 %rd921, {%r2456, %r2457}; + mov.b64 %rd922, {%r4196, %r4197}; + sub.f32x2 %rd923, %rd921, %rd922; + mov.b64 %rd924, {%r2458, %r2459}; + mov.b64 %rd925, {%r4198, %r4199}; + sub.f32x2 %rd926, %rd924, %rd925; + mov.b64 %rd927, {%r2460, %r2461}; + mov.b64 %rd928, {%r4200, %r4201}; + sub.f32x2 %rd929, %rd927, %rd928; + mov.b64 %rd930, {%r2462, %r2463}; + mov.b64 %rd931, {%r4202, %r4203}; + sub.f32x2 %rd932, %rd930, %rd931; + mov.b64 %rd933, {%r2464, %r2465}; + mov.b64 %rd934, {%r4204, %r4205}; + sub.f32x2 %rd935, %rd933, %rd934; + mov.b64 %rd936, {%r2466, %r2467}; + mov.b64 %rd937, {%r4206, %r4207}; + sub.f32x2 %rd938, %rd936, %rd937; + mov.b64 %rd939, {%r2468, %r2469}; + mov.b64 %rd940, {%r4208, %r4209}; + sub.f32x2 %rd941, %rd939, %rd940; + mov.b64 %rd942, {%r2470, %r2471}; + mov.b64 %rd943, {%r4210, %r4211}; + sub.f32x2 %rd944, %rd942, %rd943; + mov.b64 %rd945, {%r2472, %r2473}; + mov.b64 %rd946, {%r4212, %r4213}; + sub.f32x2 %rd947, %rd945, %rd946; + mov.b64 %rd948, {%r2474, %r2475}; + mov.b64 %rd949, {%r4214, %r4215}; + sub.f32x2 %rd950, %rd948, %rd949; + mov.b64 %rd951, {%r2476, %r2477}; + mov.b64 %rd952, {%r4216, %r4217}; + sub.f32x2 %rd953, %rd951, %rd952; + mov.b64 %rd954, {%r2478, %r2479}; + mov.b64 %rd955, {%r4218, %r4219}; + sub.f32x2 %rd956, %rd954, %rd955; + mov.b64 %rd957, {%r2480, %r2481}; + mov.b64 %rd958, {%r4220, %r4221}; + sub.f32x2 %rd959, %rd957, %rd958; + mov.b64 %rd960, {%r2482, %r2483}; + mov.b64 %rd961, {%r4222, %r4223}; + sub.f32x2 %rd962, %rd960, %rd961; + mov.b64 %rd963, {%r2484, %r2485}; + mov.b64 %rd964, {%r4224, %r4225}; + sub.f32x2 %rd965, %rd963, %rd964; + mov.b64 %rd966, {%r2486, %r2487}; + mov.b64 %rd967, {%r4226, %r4227}; + sub.f32x2 %rd968, %rd966, %rd967; + mov.b64 %rd969, {%r2488, %r2489}; + mov.b64 %rd970, {%r4228, %r4229}; + sub.f32x2 %rd971, %rd969, %rd970; + mov.b64 %rd972, {%r2490, %r2491}; + mov.b64 %rd973, {%r4230, %r4231}; + sub.f32x2 %rd974, %rd972, %rd973; + mov.b64 %rd975, {%r2492, %r2493}; + mov.b64 %rd976, {%r4232, %r4233}; + sub.f32x2 %rd977, %rd975, %rd976; + mov.b64 %rd978, {%r2494, %r2495}; + mov.b64 %rd979, {%r4234, %r4235}; + sub.f32x2 %rd980, %rd978, %rd979; + mov.b64 %rd981, {%r2496, %r2497}; + mov.b64 %rd982, {%r4236, %r4237}; + sub.f32x2 %rd983, %rd981, %rd982; + mov.b64 %rd984, {%r2498, %r2499}; + mov.b64 %rd985, {%r4238, %r4239}; + sub.f32x2 %rd986, %rd984, %rd985; + mov.b64 %rd987, {%r2500, %r2501}; + mov.b64 %rd988, {%r4240, %r4241}; + sub.f32x2 %rd989, %rd987, %rd988; + mov.b64 %rd990, {%r2503, %r2505}; + mov.b64 %rd991, {%r4242, %r4243}; + sub.f32x2 %rd992, %rd990, %rd991; + mov.b64 %rd993, {%r2507, %r2509}; + mov.b64 %rd994, {%r4244, %r4245}; + sub.f32x2 %rd995, %rd993, %rd994; + mov.b64 %rd996, {%r2511, %r2513}; + mov.b64 %rd997, {%r4246, %r4247}; + sub.f32x2 %rd998, %rd996, %rd997; + mov.b64 %rd999, {%r2515, %r2518}; + mov.b64 %rd1000, {%r4248, %r4249}; + sub.f32x2 %rd1001, %rd999, %rd1000; + mov.b64 %rd1002, {%r2520, %r2522}; + mov.b64 %rd1003, {%r4250, %r4251}; + sub.f32x2 %rd1004, %rd1002, %rd1003; + mov.b64 %rd1005, {%r2524, %r2526}; + mov.b64 %rd1006, {%r4252, %r4253}; + sub.f32x2 %rd1007, %rd1005, %rd1006; + mov.b64 %rd1008, {%r2528, %r2530}; + mov.b64 %rd1009, {%r4254, %r4255}; + sub.f32x2 %rd1010, %rd1008, %rd1009; + mov.b64 %rd1011, {%r2532, %r2533}; + mov.b64 %rd1012, {%r4256, %r4257}; + sub.f32x2 %rd1013, %rd1011, %rd1012; + mov.b64 %rd1014, {%r2535, %r2537}; + mov.b64 %rd1015, {%r4258, %r4259}; + sub.f32x2 %rd1016, %rd1014, %rd1015; + mov.b64 %rd1017, {%r2539, %r2541}; + mov.b64 %rd1018, {%r4260, %r4261}; + sub.f32x2 %rd1019, %rd1017, %rd1018; + mov.b64 %rd1020, {%r2543, %r2545}; + mov.b64 %rd1021, {%r4262, %r4263}; + sub.f32x2 %rd1022, %rd1020, %rd1021; + mov.b64 %rd1023, {%r2547, %r2550}; + mov.b64 %rd1024, {%r4264, %r4265}; + sub.f32x2 %rd1025, %rd1023, %rd1024; + mov.b64 %rd1026, {%r2552, %r2554}; + mov.b64 %rd1027, {%r4266, %r4267}; + sub.f32x2 %rd1028, %rd1026, %rd1027; + mov.b64 %rd1029, {%r2556, %r2558}; + mov.b64 %rd1030, {%r4268, %r4269}; + sub.f32x2 %rd1031, %rd1029, %rd1030; + mov.b64 %rd1032, {%r2560, %r2562}; + mov.b64 %rd1033, {%r4270, %r4271}; + sub.f32x2 %rd1034, %rd1032, %rd1033; + mov.b64 %rd1035, {%r2564, %r2565}; + mov.b64 %rd1036, {%r4272, %r4273}; + sub.f32x2 %rd1037, %rd1035, %rd1036; + mov.b64 %rd1038, {%r2567, %r2569}; + mov.b64 %rd1039, {%r4274, %r4275}; + sub.f32x2 %rd1040, %rd1038, %rd1039; + mov.b64 %rd1041, {%r2571, %r2573}; + mov.b64 %rd1042, {%r4276, %r4277}; + sub.f32x2 %rd1043, %rd1041, %rd1042; + mov.b64 %rd1044, {%r2575, %r2577}; + mov.b64 %rd1045, {%r4278, %r4279}; + sub.f32x2 %rd1046, %rd1044, %rd1045; + mov.b64 %rd1047, {%r2579, %r2582}; + mov.b64 %rd1048, {%r4280, %r4281}; + sub.f32x2 %rd1049, %rd1047, %rd1048; + mov.b64 %rd1050, {%r2584, %r2586}; + mov.b64 %rd1051, {%r4282, %r4283}; + sub.f32x2 %rd1052, %rd1050, %rd1051; + mov.b64 %rd1053, {%r2588, %r2590}; + mov.b64 %rd1054, {%r4284, %r4285}; + sub.f32x2 %rd1055, %rd1053, %rd1054; + mov.b64 %rd1056, {%r2592, %r2594}; + mov.b64 %rd1057, {%r4286, %r4287}; + sub.f32x2 %rd1058, %rd1056, %rd1057; + mov.b64 %rd1059, {%r2596, %r2597}; + mov.b64 %rd1060, {%r4288, %r4289}; + sub.f32x2 %rd1061, %rd1059, %rd1060; + mov.b64 %rd1062, {%r2599, %r2601}; + mov.b64 %rd1063, {%r4290, %r4291}; + sub.f32x2 %rd1064, %rd1062, %rd1063; + mov.b64 %rd1065, {%r2603, %r2605}; + mov.b64 %rd1066, {%r4292, %r4293}; + sub.f32x2 %rd1067, %rd1065, %rd1066; + mov.b64 %rd1068, {%r2607, %r2609}; + mov.b64 %rd1069, {%r4294, %r4295}; + sub.f32x2 %rd1070, %rd1068, %rd1069; + mov.b64 %rd1071, {%r2611, %r2614}; + mov.b64 %rd1072, {%r4296, %r4297}; + sub.f32x2 %rd1073, %rd1071, %rd1072; + mov.b64 %rd1074, {%r2616, %r2618}; + mov.b64 %rd1075, {%r4298, %r4299}; + sub.f32x2 %rd1076, %rd1074, %rd1075; + mov.b64 %rd1077, {%r2620, %r2622}; + mov.b64 %rd1078, {%r4300, %r4301}; + sub.f32x2 %rd1079, %rd1077, %rd1078; + mov.b64 %rd1080, {%r2624, %r2626}; + mov.b64 %rd1081, {%r4302, %r4303}; + sub.f32x2 %rd1082, %rd1080, %rd1081; + mov.b64 %rd1083, {%r2628, %r2629}; + mov.b64 %rd1084, {%r4304, %r4305}; + sub.f32x2 %rd1085, %rd1083, %rd1084; + mov.b64 %rd1086, {%r2631, %r2633}; + mov.b64 %rd1087, {%r4306, %r4307}; + sub.f32x2 %rd1088, %rd1086, %rd1087; + mov.b64 %rd1089, {%r2635, %r2637}; + mov.b64 %rd1090, {%r4308, %r4309}; + sub.f32x2 %rd1091, %rd1089, %rd1090; + mov.b64 %rd1092, {%r2639, %r2641}; + mov.b64 %rd1093, {%r4310, %r4311}; + sub.f32x2 %rd1094, %rd1092, %rd1093; + mov.b64 %rd1095, {%r2643, %r2646}; + mov.b64 %rd1096, {%r4312, %r4313}; + sub.f32x2 %rd1097, %rd1095, %rd1096; + mov.b64 %rd1098, {%r2648, %r2650}; + mov.b64 %rd1099, {%r4314, %r4315}; + sub.f32x2 %rd1100, %rd1098, %rd1099; + mov.b64 %rd1101, {%r2652, %r2654}; + mov.b64 %rd1102, {%r4316, %r4317}; + sub.f32x2 %rd1103, %rd1101, %rd1102; + mov.b64 %rd1104, {%r2656, %r2658}; + mov.b64 %rd1105, {%r4318, %r4319}; + sub.f32x2 %rd1106, %rd1104, %rd1105; + mov.b64 %rd1107, {%r2660, %r2661}; + mov.b64 %rd1108, {%r4320, %r4321}; + sub.f32x2 %rd1109, %rd1107, %rd1108; + mov.b64 %rd1110, {%r2663, %r2665}; + mov.b64 %rd1111, {%r4322, %r4323}; + sub.f32x2 %rd1112, %rd1110, %rd1111; + mov.b64 %rd1113, {%r2667, %r2669}; + mov.b64 %rd1114, {%r4324, %r4325}; + sub.f32x2 %rd1115, %rd1113, %rd1114; + mov.b64 %rd1116, {%r2671, %r2673}; + mov.b64 %rd1117, {%r4326, %r4327}; + sub.f32x2 %rd1118, %rd1116, %rd1117; + mov.b64 %rd1119, {%r2675, %r2678}; + mov.b64 %rd1120, {%r4328, %r4329}; + sub.f32x2 %rd1121, %rd1119, %rd1120; + mov.b64 %rd1122, {%r2680, %r2682}; + mov.b64 %rd1123, {%r4330, %r4331}; + sub.f32x2 %rd1124, %rd1122, %rd1123; + mov.b64 %rd1125, {%r2684, %r2686}; + mov.b64 %rd1126, {%r4332, %r4333}; + sub.f32x2 %rd1127, %rd1125, %rd1126; + mov.b64 %rd1128, {%r2688, %r2690}; + mov.b64 %rd1129, {%r4334, %r4335}; + sub.f32x2 %rd1130, %rd1128, %rd1129; + mov.b64 %rd1131, {%r2692, %r2693}; + mov.b64 %rd1132, {%r4336, %r4337}; + sub.f32x2 %rd1133, %rd1131, %rd1132; + mov.b64 %rd1134, {%r2695, %r2697}; + mov.b64 %rd1135, {%r4338, %r4339}; + sub.f32x2 %rd1136, %rd1134, %rd1135; + mov.b64 %rd1137, {%r2699, %r2701}; + mov.b64 %rd1138, {%r4340, %r4341}; + sub.f32x2 %rd1139, %rd1137, %rd1138; + mov.b64 %rd1140, {%r2703, %r2705}; + mov.b64 %rd1141, {%r4342, %r4343}; + sub.f32x2 %rd1142, %rd1140, %rd1141; + mov.b64 %rd1143, {%r2707, %r2710}; + mov.b64 %rd1144, {%r4344, %r4345}; + sub.f32x2 %rd1145, %rd1143, %rd1144; + mov.b64 %rd1146, {%r2712, %r2714}; + mov.b64 %rd1147, {%r4346, %r4347}; + sub.f32x2 %rd1148, %rd1146, %rd1147; + mov.b64 %rd1149, {%r2716, %r2718}; + mov.b64 %rd1150, {%r4348, %r4349}; + sub.f32x2 %rd1151, %rd1149, %rd1150; + mov.b64 %rd1152, {%r2720, %r2722}; + mov.b64 %rd1153, {%r4350, %r4351}; + sub.f32x2 %rd1154, %rd1152, %rd1153; + mov.b64 %rd1155, {%r2724, %r2725}; + mov.b64 %rd1156, {%r4352, %r4353}; + sub.f32x2 %rd1157, %rd1155, %rd1156; + mov.b64 %rd1158, {%r2727, %r2729}; + mov.b64 %rd1159, {%r4354, %r4355}; + sub.f32x2 %rd1160, %rd1158, %rd1159; + mov.b64 %rd1161, {%r2731, %r2733}; + mov.b64 %rd1162, {%r4356, %r4357}; + sub.f32x2 %rd1163, %rd1161, %rd1162; + mov.b64 %rd1164, {%r2735, %r2737}; + mov.b64 %rd1165, {%r4358, %r4359}; + sub.f32x2 %rd1166, %rd1164, %rd1165; + mov.b64 %rd1167, {%r2739, %r2742}; + mov.b64 %rd1168, {%r4360, %r4361}; + sub.f32x2 %rd1169, %rd1167, %rd1168; + mov.b64 %rd1170, {%r2744, %r2746}; + mov.b64 %rd1171, {%r4362, %r4363}; + sub.f32x2 %rd1172, %rd1170, %rd1171; + mov.b64 %rd1173, {%r2748, %r2750}; + mov.b64 %rd1174, {%r4364, %r4365}; + sub.f32x2 %rd1175, %rd1173, %rd1174; + mov.b64 %rd1176, {%r2752, %r2754}; + mov.b64 %rd1177, {%r4366, %r4367}; + sub.f32x2 %rd1178, %rd1176, %rd1177; + mov.b64 %rd1179, {%r2756, %r2757}; + mov.b64 %rd1180, {%r4368, %r4369}; + sub.f32x2 %rd1181, %rd1179, %rd1180; + mov.b64 %rd1182, {%r2758, %r2759}; + mov.b64 %rd1183, {%r4370, %r4371}; + sub.f32x2 %rd1184, %rd1182, %rd1183; + mov.b64 %rd1185, {%r2760, %r2761}; + mov.b64 %rd1186, {%r4372, %r4373}; + sub.f32x2 %rd1187, %rd1185, %rd1186; + mov.b64 %rd1188, {%r2762, %r2764}; + mov.b64 %rd1189, {%r4374, %r4375}; + sub.f32x2 %rd1190, %rd1188, %rd1189; + mov.b64 %rd1191, {%r2766, %r2779}; + mov.b64 %rd1192, {%r4376, %r4377}; + sub.f32x2 %rd1193, %rd1191, %rd1192; + mov.b64 %rd1194, {%r2781, %r2783}; + mov.b64 %rd1195, {%r4378, %r4379}; + sub.f32x2 %rd1196, %rd1194, %rd1195; + mov.b64 %rd1197, {%r2785, %r2787}; + mov.b64 %rd1198, {%r4380, %r4381}; + sub.f32x2 %rd1199, %rd1197, %rd1198; + mov.b64 %rd1200, {%r2789, %r2791}; + mov.b64 %rd1201, {%r4382, %r4383}; + sub.f32x2 %rd1202, %rd1200, %rd1201; + mov.b64 %rd1203, {%r2793, %r2811}; + mov.b64 %rd1204, {%r4384, %r4385}; + sub.f32x2 %rd1205, %rd1203, %rd1204; + mov.b64 %rd1206, {%r2813, %r2815}; + mov.b64 %rd1207, {%r4386, %r4387}; + sub.f32x2 %rd1208, %rd1206, %rd1207; + mov.b64 %rd1209, {%r2817, %r2819}; + mov.b64 %rd1210, {%r4388, %r4389}; + sub.f32x2 %rd1211, %rd1209, %rd1210; + mov.b64 %rd1212, {%r2821, %r2823}; + mov.b64 %rd1213, {%r4390, %r4391}; + sub.f32x2 %rd1214, %rd1212, %rd1213; + mov.b64 %rd1215, {%r2825, %r2843}; + mov.b64 %rd1216, {%r4392, %r4393}; + sub.f32x2 %rd1217, %rd1215, %rd1216; + mov.b64 %rd1218, {%r2845, %r2847}; + mov.b64 %rd1219, {%r4394, %r4395}; + sub.f32x2 %rd1220, %rd1218, %rd1219; + mov.b64 %rd1221, {%r2849, %r2851}; + mov.b64 %rd1222, {%r4396, %r4397}; + sub.f32x2 %rd1223, %rd1221, %rd1222; + mov.b64 %rd1224, {%r2853, %r2855}; + mov.b64 %rd1225, {%r4398, %r4399}; + sub.f32x2 %rd1226, %rd1224, %rd1225; + mov.b64 %rd1227, {%r2857, %r2875}; + mov.b64 %rd1228, {%r4400, %r4401}; + sub.f32x2 %rd1229, %rd1227, %rd1228; + mov.b64 %rd1230, {%r2877, %r2879}; + mov.b64 %rd1231, {%r4402, %r4403}; + sub.f32x2 %rd1232, %rd1230, %rd1231; + mov.b64 %rd1233, {%r2881, %r2883}; + mov.b64 %rd1234, {%r4404, %r4405}; + sub.f32x2 %rd1235, %rd1233, %rd1234; + mov.b64 %rd1236, {%r2885, %r2887}; + mov.b64 %rd1237, {%r4406, %r4407}; + sub.f32x2 %rd1238, %rd1236, %rd1237; + mov.b64 %rd1239, {%r2889, %r2907}; + mov.b64 %rd1240, {%r4408, %r4409}; + sub.f32x2 %rd1241, %rd1239, %rd1240; + mov.b64 %rd1242, {%r2909, %r2911}; + mov.b64 %rd1243, {%r4410, %r4411}; + sub.f32x2 %rd1244, %rd1242, %rd1243; + mov.b64 %rd1245, {%r2913, %r2915}; + mov.b64 %rd1246, {%r4412, %r4413}; + sub.f32x2 %rd1247, %rd1245, %rd1246; + mov.b64 %rd1248, {%r2917, %r2919}; + mov.b64 %rd1249, {%r4414, %r4415}; + sub.f32x2 %rd1250, %rd1248, %rd1249; + mov.b64 %rd1251, {%r2921, %r2923}; + mov.b64 %rd1252, {%r4416, %r4417}; + sub.f32x2 %rd1253, %rd1251, %rd1252; + mov.b64 %rd1254, {%r2925, %r2927}; + mov.b64 %rd1255, {%r4418, %r4419}; + sub.f32x2 %rd1256, %rd1254, %rd1255; + mov.b64 %rd1257, {%r2929, %r2931}; + mov.b64 %rd1258, {%r4420, %r4421}; + sub.f32x2 %rd1259, %rd1257, %rd1258; + mov.b64 %rd1260, {%r2933, %r2935}; + mov.b64 %rd1261, {%r4422, %r4423}; + sub.f32x2 %rd1262, %rd1260, %rd1261; + mov.b64 %rd1263, {%r2937, %r2939}; + mov.b64 %rd1264, {%r4424, %r4425}; + sub.f32x2 %rd1265, %rd1263, %rd1264; + mov.b64 %rd1266, {%r2941, %r2943}; + mov.b64 %rd1267, {%r4426, %r4427}; + sub.f32x2 %rd1268, %rd1266, %rd1267; + mov.b64 %rd1269, {%r2945, %r2947}; + mov.b64 %rd1270, {%r4428, %r4429}; + sub.f32x2 %rd1271, %rd1269, %rd1270; + mov.b64 %rd1272, {%r2949, %r2951}; + mov.b64 %rd1273, {%r4430, %r4431}; + sub.f32x2 %rd1274, %rd1272, %rd1273; + mov.b64 %rd1275, {%r2953, %r2955}; + mov.b64 %rd1276, {%r4432, %r4433}; + sub.f32x2 %rd1277, %rd1275, %rd1276; + mov.b64 %rd1278, {%r2957, %r2959}; + mov.b64 %rd1279, {%r4434, %r4435}; + sub.f32x2 %rd1280, %rd1278, %rd1279; + mov.b64 %rd1281, {%r2961, %r2963}; + mov.b64 %rd1282, {%r4436, %r4437}; + sub.f32x2 %rd1283, %rd1281, %rd1282; + mov.b64 %rd1284, {%r2965, %r2967}; + mov.b64 %rd1285, {%r4438, %r4439}; + sub.f32x2 %rd1286, %rd1284, %rd1285; + mov.b64 %rd1287, {%r2969, %r2971}; + mov.b64 %rd1288, {%r4440, %r4441}; + sub.f32x2 %rd1289, %rd1287, %rd1288; + mov.b64 %rd1290, {%r2973, %r2975}; + mov.b64 %rd1291, {%r4442, %r4443}; + sub.f32x2 %rd1292, %rd1290, %rd1291; + mov.b64 %rd1293, {%r2977, %r2979}; + mov.b64 %rd1294, {%r4444, %r4445}; + sub.f32x2 %rd1295, %rd1293, %rd1294; + mov.b64 %rd1296, {%r2981, %r2983}; + mov.b64 %rd1297, {%r4446, %r4447}; + sub.f32x2 %rd1298, %rd1296, %rd1297; + mov.b64 %rd1299, {%r2985, %r2987}; + mov.b64 %rd1300, {%r4448, %r4449}; + sub.f32x2 %rd1301, %rd1299, %rd1300; + mov.b64 %rd1302, {%r2989, %r2991}; + mov.b64 %rd1303, {%r4450, %r4451}; + sub.f32x2 %rd1304, %rd1302, %rd1303; + mov.b64 %rd1305, {%r2993, %r2995}; + mov.b64 %rd1306, {%r4452, %r4453}; + sub.f32x2 %rd1307, %rd1305, %rd1306; + mov.b64 %rd1308, {%r2997, %r2999}; + mov.b64 %rd1309, {%r4454, %r4455}; + sub.f32x2 %rd1310, %rd1308, %rd1309; + mov.b64 %rd1311, {%r3001, %r3003}; + mov.b64 %rd1312, {%r4456, %r4457}; + sub.f32x2 %rd1313, %rd1311, %rd1312; + mov.b64 %rd1314, {%r3005, %r3007}; + mov.b64 %rd1315, {%r4458, %r4459}; + sub.f32x2 %rd1316, %rd1314, %rd1315; + mov.b64 %rd1317, {%r3009, %r3011}; + mov.b64 %rd1318, {%r4460, %r4461}; + sub.f32x2 %rd1319, %rd1317, %rd1318; + mov.b64 %rd1320, {%r3013, %r3015}; + mov.b64 %rd1321, {%r4462, %r4463}; + sub.f32x2 %rd1322, %rd1320, %rd1321; + mov.b64 %rd1323, {%r3017, %r3019}; + mov.b64 %rd1324, {%r4464, %r4465}; + sub.f32x2 %rd1325, %rd1323, %rd1324; + mov.b64 %rd1326, {%r3021, %r3023}; + mov.b64 %rd1327, {%r4466, %r4467}; + sub.f32x2 %rd1328, %rd1326, %rd1327; + mov.b64 %rd1329, {%r3025, %r3027}; + mov.b64 %rd1330, {%r4468, %r4469}; + sub.f32x2 %rd1331, %rd1329, %rd1330; + mov.b64 %rd1332, {%r3029, %r3031}; + mov.b64 %rd1333, {%r4470, %r4471}; + sub.f32x2 %rd1334, %rd1332, %rd1333; + mov.b64 %rd1335, {%r3033, %r3035}; + mov.b64 %rd1336, {%r4472, %r4473}; + sub.f32x2 %rd1337, %rd1335, %rd1336; + mov.b64 %rd1338, {%r3037, %r3039}; + mov.b64 %rd1339, {%r4474, %r4475}; + sub.f32x2 %rd1340, %rd1338, %rd1339; + mov.b64 %rd1341, {%r3041, %r3043}; + mov.b64 %rd1342, {%r4476, %r4477}; + sub.f32x2 %rd1343, %rd1341, %rd1342; + mov.b64 %rd1344, {%r3045, %r3047}; + mov.b64 %rd1345, {%r4478, %r4479}; + sub.f32x2 %rd1346, %rd1344, %rd1345; + mov.b64 %rd1347, {%r3049, %r3051}; + mov.b64 %rd1348, {%r4480, %r4481}; + sub.f32x2 %rd1349, %rd1347, %rd1348; + mov.b64 %rd1350, {%r3053, %r3055}; + mov.b64 %rd1351, {%r4482, %r4483}; + sub.f32x2 %rd1352, %rd1350, %rd1351; + mov.b64 %rd1353, {%r3057, %r3059}; + mov.b64 %rd1354, {%r4484, %r4485}; + sub.f32x2 %rd1355, %rd1353, %rd1354; + mov.b64 %rd1356, {%r3061, %r3063}; + mov.b64 %rd1357, {%r4486, %r4487}; + sub.f32x2 %rd1358, %rd1356, %rd1357; + mov.b64 %rd1359, {%r3065, %r3067}; + mov.b64 %rd1360, {%r4488, %r4489}; + sub.f32x2 %rd1361, %rd1359, %rd1360; + mov.b64 %rd1362, {%r3069, %r3071}; + mov.b64 %rd1363, {%r4490, %r4491}; + sub.f32x2 %rd1364, %rd1362, %rd1363; + mov.b64 %rd1365, {%r3073, %r3075}; + mov.b64 %rd1366, {%r4492, %r4493}; + sub.f32x2 %rd1367, %rd1365, %rd1366; + mov.b64 %rd1368, {%r3077, %r3079}; + mov.b64 %rd1369, {%r4494, %r4495}; + sub.f32x2 %rd1370, %rd1368, %rd1369; + mov.b64 %rd1371, {%r3081, %r3083}; + mov.b64 %rd1372, {%r4496, %r4497}; + sub.f32x2 %rd1373, %rd1371, %rd1372; + mov.b64 %rd1374, {%r3085, %r3087}; + mov.b64 %rd1375, {%r4498, %r4499}; + sub.f32x2 %rd1376, %rd1374, %rd1375; + mov.b64 %rd1377, {%r3089, %r3091}; + mov.b64 %rd1378, {%r4500, %r4501}; + sub.f32x2 %rd1379, %rd1377, %rd1378; + mov.b64 %rd1380, {%r3093, %r3095}; + mov.b64 %rd1381, {%r4502, %r4503}; + sub.f32x2 %rd1382, %rd1380, %rd1381; + mov.b64 %rd1383, {%r3097, %r3099}; + mov.b64 %rd1384, {%r4504, %r4505}; + sub.f32x2 %rd1385, %rd1383, %rd1384; + mov.b64 %rd1386, {%r3101, %r3103}; + mov.b64 %rd1387, {%r4506, %r4507}; + sub.f32x2 %rd1388, %rd1386, %rd1387; + mov.b64 %rd1389, {%r3105, %r3107}; + mov.b64 %rd1390, {%r4508, %r4509}; + sub.f32x2 %rd1391, %rd1389, %rd1390; + mov.b64 %rd1392, {%r3109, %r3111}; + mov.b64 %rd1393, {%r4510, %r4511}; + sub.f32x2 %rd1394, %rd1392, %rd1393; + mov.b64 %rd1395, {%r3113, %r3115}; + mov.b64 %rd1396, {%r4512, %r4513}; + sub.f32x2 %rd1397, %rd1395, %rd1396; + mov.b64 %rd1398, {%r3117, %r3119}; + mov.b64 %rd1399, {%r4514, %r4515}; + sub.f32x2 %rd1400, %rd1398, %rd1399; + mov.b64 %rd1401, {%r3121, %r3123}; + mov.b64 %rd1402, {%r4516, %r4517}; + sub.f32x2 %rd1403, %rd1401, %rd1402; + mov.b64 %rd1404, {%r3125, %r3127}; + mov.b64 %rd1405, {%r4518, %r4519}; + sub.f32x2 %rd1406, %rd1404, %rd1405; + mov.b64 %rd1407, {%r3129, %r3131}; + mov.b64 %rd1408, {%r4520, %r4521}; + sub.f32x2 %rd1409, %rd1407, %rd1408; + mov.b64 %rd1410, {%r3133, %r3135}; + mov.b64 %rd1411, {%r4522, %r4523}; + sub.f32x2 %rd1412, %rd1410, %rd1411; + mov.b64 %rd1413, {%r3137, %r3139}; + mov.b64 %rd1414, {%r4524, %r4525}; + sub.f32x2 %rd1415, %rd1413, %rd1414; + mov.b64 %rd1416, {%r3141, %r3143}; + mov.b64 %rd1417, {%r4526, %r4527}; + sub.f32x2 %rd1418, %rd1416, %rd1417; + .loc 1 108 27 // l2norm.py:108:27 + mov.b64 {%r4544, %r4545}, %rd691; + cvt.rn.bf16x2.f32 %r522, %r4545, %r4544; + mov.b64 {%r4546, %r4547}, %rd692; + cvt.rn.bf16x2.f32 %r523, %r4547, %r4546; + mov.b64 {%r4548, %r4549}, %rd695; + cvt.rn.bf16x2.f32 %r524, %r4549, %r4548; + mov.b64 {%r4550, %r4551}, %rd696; + cvt.rn.bf16x2.f32 %r525, %r4551, %r4550; + mov.b64 {%r4552, %r4553}, %rd699; + cvt.rn.bf16x2.f32 %r526, %r4553, %r4552; + mov.b64 {%r4554, %r4555}, %rd700; + cvt.rn.bf16x2.f32 %r527, %r4555, %r4554; + mov.b64 {%r4556, %r4557}, %rd703; + cvt.rn.bf16x2.f32 %r528, %r4557, %r4556; + mov.b64 {%r4558, %r4559}, %rd704; + cvt.rn.bf16x2.f32 %r529, %r4559, %r4558; + mov.b64 {%r4560, %r4561}, %rd707; + cvt.rn.bf16x2.f32 %r530, %r4561, %r4560; + mov.b64 {%r4562, %r4563}, %rd708; + cvt.rn.bf16x2.f32 %r531, %r4563, %r4562; + mov.b64 {%r4564, %r4565}, %rd711; + cvt.rn.bf16x2.f32 %r532, %r4565, %r4564; + mov.b64 {%r4566, %r4567}, %rd712; + cvt.rn.bf16x2.f32 %r533, %r4567, %r4566; + mov.b64 {%r4568, %r4569}, %rd715; + cvt.rn.bf16x2.f32 %r534, %r4569, %r4568; + mov.b64 {%r4570, %r4571}, %rd716; + cvt.rn.bf16x2.f32 %r535, %r4571, %r4570; + mov.b64 {%r4572, %r4573}, %rd719; + cvt.rn.bf16x2.f32 %r536, %r4573, %r4572; + mov.b64 {%r4574, %r4575}, %rd720; + cvt.rn.bf16x2.f32 %r537, %r4575, %r4574; + mov.b64 {%r4576, %r4577}, %rd723; + cvt.rn.bf16x2.f32 %r538, %r4577, %r4576; + mov.b64 {%r4578, %r4579}, %rd724; + cvt.rn.bf16x2.f32 %r539, %r4579, %r4578; + mov.b64 {%r4580, %r4581}, %rd727; + cvt.rn.bf16x2.f32 %r540, %r4581, %r4580; + mov.b64 {%r4582, %r4583}, %rd728; + cvt.rn.bf16x2.f32 %r541, %r4583, %r4582; + mov.b64 {%r4584, %r4585}, %rd731; + cvt.rn.bf16x2.f32 %r542, %r4585, %r4584; + mov.b64 {%r4586, %r4587}, %rd732; + cvt.rn.bf16x2.f32 %r543, %r4587, %r4586; + mov.b64 {%r4588, %r4589}, %rd735; + cvt.rn.bf16x2.f32 %r544, %r4589, %r4588; + mov.b64 {%r4590, %r4591}, %rd736; + cvt.rn.bf16x2.f32 %r545, %r4591, %r4590; + mov.b64 {%r4592, %r4593}, %rd739; + cvt.rn.bf16x2.f32 %r546, %r4593, %r4592; + mov.b64 {%r4594, %r4595}, %rd740; + cvt.rn.bf16x2.f32 %r547, %r4595, %r4594; + mov.b64 {%r4596, %r4597}, %rd743; + cvt.rn.bf16x2.f32 %r548, %r4597, %r4596; + mov.b64 {%r4598, %r4599}, %rd746; + cvt.rn.bf16x2.f32 %r549, %r4599, %r4598; + mov.b64 {%r4600, %r4601}, %rd749; + cvt.rn.bf16x2.f32 %r550, %r4601, %r4600; + mov.b64 {%r4602, %r4603}, %rd752; + cvt.rn.bf16x2.f32 %r551, %r4603, %r4602; + mov.b64 {%r4604, %r4605}, %rd755; + cvt.rn.bf16x2.f32 %r552, %r4605, %r4604; + mov.b64 {%r4606, %r4607}, %rd758; + cvt.rn.bf16x2.f32 %r553, %r4607, %r4606; + mov.b64 {%r4608, %r4609}, %rd761; + cvt.rn.bf16x2.f32 %r554, %r4609, %r4608; + mov.b64 {%r4610, %r4611}, %rd764; + cvt.rn.bf16x2.f32 %r555, %r4611, %r4610; + mov.b64 {%r4612, %r4613}, %rd767; + cvt.rn.bf16x2.f32 %r556, %r4613, %r4612; + mov.b64 {%r4614, %r4615}, %rd770; + cvt.rn.bf16x2.f32 %r557, %r4615, %r4614; + mov.b64 {%r4616, %r4617}, %rd773; + cvt.rn.bf16x2.f32 %r558, %r4617, %r4616; + mov.b64 {%r4618, %r4619}, %rd776; + cvt.rn.bf16x2.f32 %r559, %r4619, %r4618; + mov.b64 {%r4620, %r4621}, %rd779; + cvt.rn.bf16x2.f32 %r560, %r4621, %r4620; + mov.b64 {%r4622, %r4623}, %rd782; + cvt.rn.bf16x2.f32 %r561, %r4623, %r4622; + mov.b64 {%r4624, %r4625}, %rd785; + cvt.rn.bf16x2.f32 %r562, %r4625, %r4624; + mov.b64 {%r4626, %r4627}, %rd788; + cvt.rn.bf16x2.f32 %r563, %r4627, %r4626; + mov.b64 {%r4628, %r4629}, %rd791; + cvt.rn.bf16x2.f32 %r564, %r4629, %r4628; + mov.b64 {%r4630, %r4631}, %rd794; + cvt.rn.bf16x2.f32 %r565, %r4631, %r4630; + mov.b64 {%r4632, %r4633}, %rd797; + cvt.rn.bf16x2.f32 %r566, %r4633, %r4632; + mov.b64 {%r4634, %r4635}, %rd800; + cvt.rn.bf16x2.f32 %r567, %r4635, %r4634; + mov.b64 {%r4636, %r4637}, %rd803; + cvt.rn.bf16x2.f32 %r568, %r4637, %r4636; + mov.b64 {%r4638, %r4639}, %rd806; + cvt.rn.bf16x2.f32 %r569, %r4639, %r4638; + mov.b64 {%r4640, %r4641}, %rd809; + cvt.rn.bf16x2.f32 %r570, %r4641, %r4640; + mov.b64 {%r4642, %r4643}, %rd812; + cvt.rn.bf16x2.f32 %r571, %r4643, %r4642; + mov.b64 {%r4644, %r4645}, %rd815; + cvt.rn.bf16x2.f32 %r572, %r4645, %r4644; + mov.b64 {%r4646, %r4647}, %rd818; + cvt.rn.bf16x2.f32 %r573, %r4647, %r4646; + mov.b64 {%r4648, %r4649}, %rd821; + cvt.rn.bf16x2.f32 %r574, %r4649, %r4648; + mov.b64 {%r4650, %r4651}, %rd824; + cvt.rn.bf16x2.f32 %r575, %r4651, %r4650; + mov.b64 {%r4652, %r4653}, %rd827; + cvt.rn.bf16x2.f32 %r576, %r4653, %r4652; + mov.b64 {%r4654, %r4655}, %rd830; + cvt.rn.bf16x2.f32 %r577, %r4655, %r4654; + mov.b64 {%r4656, %r4657}, %rd833; + cvt.rn.bf16x2.f32 %r578, %r4657, %r4656; + mov.b64 {%r4658, %r4659}, %rd836; + cvt.rn.bf16x2.f32 %r579, %r4659, %r4658; + mov.b64 {%r4660, %r4661}, %rd839; + cvt.rn.bf16x2.f32 %r580, %r4661, %r4660; + mov.b64 {%r4662, %r4663}, %rd842; + cvt.rn.bf16x2.f32 %r581, %r4663, %r4662; + mov.b64 {%r4664, %r4665}, %rd845; + cvt.rn.bf16x2.f32 %r582, %r4665, %r4664; + mov.b64 {%r4666, %r4667}, %rd848; + cvt.rn.bf16x2.f32 %r583, %r4667, %r4666; + mov.b64 {%r4668, %r4669}, %rd851; + cvt.rn.bf16x2.f32 %r584, %r4669, %r4668; + mov.b64 {%r4670, %r4671}, %rd854; + cvt.rn.bf16x2.f32 %r585, %r4671, %r4670; + mov.b64 {%r4672, %r4673}, %rd857; + cvt.rn.bf16x2.f32 %r586, %r4673, %r4672; + mov.b64 {%r4674, %r4675}, %rd860; + cvt.rn.bf16x2.f32 %r587, %r4675, %r4674; + mov.b64 {%r4676, %r4677}, %rd863; + cvt.rn.bf16x2.f32 %r588, %r4677, %r4676; + mov.b64 {%r4678, %r4679}, %rd866; + cvt.rn.bf16x2.f32 %r589, %r4679, %r4678; + mov.b64 {%r4680, %r4681}, %rd869; + cvt.rn.bf16x2.f32 %r590, %r4681, %r4680; + mov.b64 {%r4682, %r4683}, %rd872; + cvt.rn.bf16x2.f32 %r591, %r4683, %r4682; + mov.b64 {%r4684, %r4685}, %rd875; + cvt.rn.bf16x2.f32 %r592, %r4685, %r4684; + mov.b64 {%r4686, %r4687}, %rd878; + cvt.rn.bf16x2.f32 %r593, %r4687, %r4686; + mov.b64 {%r4688, %r4689}, %rd881; + cvt.rn.bf16x2.f32 %r594, %r4689, %r4688; + mov.b64 {%r4690, %r4691}, %rd884; + cvt.rn.bf16x2.f32 %r595, %r4691, %r4690; + mov.b64 {%r4692, %r4693}, %rd887; + cvt.rn.bf16x2.f32 %r596, %r4693, %r4692; + mov.b64 {%r4694, %r4695}, %rd890; + cvt.rn.bf16x2.f32 %r597, %r4695, %r4694; + mov.b64 {%r4696, %r4697}, %rd893; + cvt.rn.bf16x2.f32 %r598, %r4697, %r4696; + mov.b64 {%r4698, %r4699}, %rd896; + cvt.rn.bf16x2.f32 %r599, %r4699, %r4698; + mov.b64 {%r4700, %r4701}, %rd899; + cvt.rn.bf16x2.f32 %r600, %r4701, %r4700; + mov.b64 {%r4702, %r4703}, %rd902; + cvt.rn.bf16x2.f32 %r601, %r4703, %r4702; + mov.b64 {%r4704, %r4705}, %rd905; + cvt.rn.bf16x2.f32 %r602, %r4705, %r4704; + mov.b64 {%r4706, %r4707}, %rd908; + cvt.rn.bf16x2.f32 %r603, %r4707, %r4706; + mov.b64 {%r4708, %r4709}, %rd911; + cvt.rn.bf16x2.f32 %r604, %r4709, %r4708; + mov.b64 {%r4710, %r4711}, %rd914; + cvt.rn.bf16x2.f32 %r605, %r4711, %r4710; + mov.b64 {%r4712, %r4713}, %rd917; + cvt.rn.bf16x2.f32 %r606, %r4713, %r4712; + mov.b64 {%r4714, %r4715}, %rd920; + cvt.rn.bf16x2.f32 %r607, %r4715, %r4714; + mov.b64 {%r4716, %r4717}, %rd923; + cvt.rn.bf16x2.f32 %r608, %r4717, %r4716; + mov.b64 {%r4718, %r4719}, %rd926; + cvt.rn.bf16x2.f32 %r609, %r4719, %r4718; + mov.b64 {%r4720, %r4721}, %rd929; + cvt.rn.bf16x2.f32 %r610, %r4721, %r4720; + mov.b64 {%r4722, %r4723}, %rd932; + cvt.rn.bf16x2.f32 %r611, %r4723, %r4722; + mov.b64 {%r4724, %r4725}, %rd935; + cvt.rn.bf16x2.f32 %r612, %r4725, %r4724; + mov.b64 {%r4726, %r4727}, %rd938; + cvt.rn.bf16x2.f32 %r613, %r4727, %r4726; + mov.b64 {%r4728, %r4729}, %rd941; + cvt.rn.bf16x2.f32 %r614, %r4729, %r4728; + mov.b64 {%r4730, %r4731}, %rd944; + cvt.rn.bf16x2.f32 %r615, %r4731, %r4730; + mov.b64 {%r4732, %r4733}, %rd947; + cvt.rn.bf16x2.f32 %r616, %r4733, %r4732; + mov.b64 {%r4734, %r4735}, %rd950; + cvt.rn.bf16x2.f32 %r617, %r4735, %r4734; + mov.b64 {%r4736, %r4737}, %rd953; + cvt.rn.bf16x2.f32 %r618, %r4737, %r4736; + mov.b64 {%r4738, %r4739}, %rd956; + cvt.rn.bf16x2.f32 %r619, %r4739, %r4738; + mov.b64 {%r4740, %r4741}, %rd959; + cvt.rn.bf16x2.f32 %r620, %r4741, %r4740; + mov.b64 {%r4742, %r4743}, %rd962; + cvt.rn.bf16x2.f32 %r621, %r4743, %r4742; + mov.b64 {%r4744, %r4745}, %rd965; + cvt.rn.bf16x2.f32 %r622, %r4745, %r4744; + mov.b64 {%r4746, %r4747}, %rd968; + cvt.rn.bf16x2.f32 %r623, %r4747, %r4746; + mov.b64 {%r4748, %r4749}, %rd971; + cvt.rn.bf16x2.f32 %r624, %r4749, %r4748; + mov.b64 {%r4750, %r4751}, %rd974; + cvt.rn.bf16x2.f32 %r625, %r4751, %r4750; + mov.b64 {%r4752, %r4753}, %rd977; + cvt.rn.bf16x2.f32 %r626, %r4753, %r4752; + mov.b64 {%r4754, %r4755}, %rd980; + cvt.rn.bf16x2.f32 %r627, %r4755, %r4754; + mov.b64 {%r4756, %r4757}, %rd983; + cvt.rn.bf16x2.f32 %r628, %r4757, %r4756; + mov.b64 {%r4758, %r4759}, %rd986; + cvt.rn.bf16x2.f32 %r629, %r4759, %r4758; + mov.b64 {%r4760, %r4761}, %rd989; + cvt.rn.bf16x2.f32 %r630, %r4761, %r4760; + mov.b64 {%r4762, %r4763}, %rd992; + cvt.rn.bf16x2.f32 %r631, %r4763, %r4762; + mov.b64 {%r4764, %r4765}, %rd995; + cvt.rn.bf16x2.f32 %r632, %r4765, %r4764; + mov.b64 {%r4766, %r4767}, %rd998; + cvt.rn.bf16x2.f32 %r633, %r4767, %r4766; + mov.b64 {%r4768, %r4769}, %rd1001; + cvt.rn.bf16x2.f32 %r634, %r4769, %r4768; + mov.b64 {%r4770, %r4771}, %rd1004; + cvt.rn.bf16x2.f32 %r635, %r4771, %r4770; + mov.b64 {%r4772, %r4773}, %rd1007; + cvt.rn.bf16x2.f32 %r636, %r4773, %r4772; + mov.b64 {%r4774, %r4775}, %rd1010; + cvt.rn.bf16x2.f32 %r637, %r4775, %r4774; + mov.b64 {%r4776, %r4777}, %rd1013; + cvt.rn.bf16x2.f32 %r638, %r4777, %r4776; + mov.b64 {%r4778, %r4779}, %rd1016; + cvt.rn.bf16x2.f32 %r639, %r4779, %r4778; + mov.b64 {%r4780, %r4781}, %rd1019; + cvt.rn.bf16x2.f32 %r640, %r4781, %r4780; + mov.b64 {%r4782, %r4783}, %rd1022; + cvt.rn.bf16x2.f32 %r641, %r4783, %r4782; + mov.b64 {%r4784, %r4785}, %rd1025; + cvt.rn.bf16x2.f32 %r642, %r4785, %r4784; + mov.b64 {%r4786, %r4787}, %rd1028; + cvt.rn.bf16x2.f32 %r643, %r4787, %r4786; + mov.b64 {%r4788, %r4789}, %rd1031; + cvt.rn.bf16x2.f32 %r644, %r4789, %r4788; + mov.b64 {%r4790, %r4791}, %rd1034; + cvt.rn.bf16x2.f32 %r645, %r4791, %r4790; + mov.b64 {%r4792, %r4793}, %rd1037; + cvt.rn.bf16x2.f32 %r646, %r4793, %r4792; + mov.b64 {%r4794, %r4795}, %rd1040; + cvt.rn.bf16x2.f32 %r647, %r4795, %r4794; + mov.b64 {%r4796, %r4797}, %rd1043; + cvt.rn.bf16x2.f32 %r648, %r4797, %r4796; + mov.b64 {%r4798, %r4799}, %rd1046; + cvt.rn.bf16x2.f32 %r649, %r4799, %r4798; + mov.b64 {%r4800, %r4801}, %rd1049; + cvt.rn.bf16x2.f32 %r650, %r4801, %r4800; + mov.b64 {%r4802, %r4803}, %rd1052; + cvt.rn.bf16x2.f32 %r651, %r4803, %r4802; + mov.b64 {%r4804, %r4805}, %rd1055; + cvt.rn.bf16x2.f32 %r652, %r4805, %r4804; + mov.b64 {%r4806, %r4807}, %rd1058; + cvt.rn.bf16x2.f32 %r653, %r4807, %r4806; + mov.b64 {%r4808, %r4809}, %rd1061; + cvt.rn.bf16x2.f32 %r654, %r4809, %r4808; + mov.b64 {%r4810, %r4811}, %rd1064; + cvt.rn.bf16x2.f32 %r655, %r4811, %r4810; + mov.b64 {%r4812, %r4813}, %rd1067; + cvt.rn.bf16x2.f32 %r656, %r4813, %r4812; + mov.b64 {%r4814, %r4815}, %rd1070; + cvt.rn.bf16x2.f32 %r657, %r4815, %r4814; + mov.b64 {%r4816, %r4817}, %rd1073; + cvt.rn.bf16x2.f32 %r658, %r4817, %r4816; + mov.b64 {%r4818, %r4819}, %rd1076; + cvt.rn.bf16x2.f32 %r659, %r4819, %r4818; + mov.b64 {%r4820, %r4821}, %rd1079; + cvt.rn.bf16x2.f32 %r660, %r4821, %r4820; + mov.b64 {%r4822, %r4823}, %rd1082; + cvt.rn.bf16x2.f32 %r661, %r4823, %r4822; + mov.b64 {%r4824, %r4825}, %rd1085; + cvt.rn.bf16x2.f32 %r662, %r4825, %r4824; + mov.b64 {%r4826, %r4827}, %rd1088; + cvt.rn.bf16x2.f32 %r663, %r4827, %r4826; + mov.b64 {%r4828, %r4829}, %rd1091; + cvt.rn.bf16x2.f32 %r664, %r4829, %r4828; + mov.b64 {%r4830, %r4831}, %rd1094; + cvt.rn.bf16x2.f32 %r665, %r4831, %r4830; + mov.b64 {%r4832, %r4833}, %rd1097; + cvt.rn.bf16x2.f32 %r666, %r4833, %r4832; + mov.b64 {%r4834, %r4835}, %rd1100; + cvt.rn.bf16x2.f32 %r667, %r4835, %r4834; + mov.b64 {%r4836, %r4837}, %rd1103; + cvt.rn.bf16x2.f32 %r668, %r4837, %r4836; + mov.b64 {%r4838, %r4839}, %rd1106; + cvt.rn.bf16x2.f32 %r669, %r4839, %r4838; + mov.b64 {%r4840, %r4841}, %rd1109; + cvt.rn.bf16x2.f32 %r670, %r4841, %r4840; + mov.b64 {%r4842, %r4843}, %rd1112; + cvt.rn.bf16x2.f32 %r671, %r4843, %r4842; + mov.b64 {%r4844, %r4845}, %rd1115; + cvt.rn.bf16x2.f32 %r672, %r4845, %r4844; + mov.b64 {%r4846, %r4847}, %rd1118; + cvt.rn.bf16x2.f32 %r673, %r4847, %r4846; + mov.b64 {%r4848, %r4849}, %rd1121; + cvt.rn.bf16x2.f32 %r674, %r4849, %r4848; + mov.b64 {%r4850, %r4851}, %rd1124; + cvt.rn.bf16x2.f32 %r675, %r4851, %r4850; + mov.b64 {%r4852, %r4853}, %rd1127; + cvt.rn.bf16x2.f32 %r676, %r4853, %r4852; + mov.b64 {%r4854, %r4855}, %rd1130; + cvt.rn.bf16x2.f32 %r677, %r4855, %r4854; + mov.b64 {%r4856, %r4857}, %rd1133; + cvt.rn.bf16x2.f32 %r678, %r4857, %r4856; + mov.b64 {%r4858, %r4859}, %rd1136; + cvt.rn.bf16x2.f32 %r679, %r4859, %r4858; + mov.b64 {%r4860, %r4861}, %rd1139; + cvt.rn.bf16x2.f32 %r680, %r4861, %r4860; + mov.b64 {%r4862, %r4863}, %rd1142; + cvt.rn.bf16x2.f32 %r681, %r4863, %r4862; + mov.b64 {%r4864, %r4865}, %rd1145; + cvt.rn.bf16x2.f32 %r682, %r4865, %r4864; + mov.b64 {%r4866, %r4867}, %rd1148; + cvt.rn.bf16x2.f32 %r683, %r4867, %r4866; + mov.b64 {%r4868, %r4869}, %rd1151; + cvt.rn.bf16x2.f32 %r684, %r4869, %r4868; + mov.b64 {%r4870, %r4871}, %rd1154; + cvt.rn.bf16x2.f32 %r685, %r4871, %r4870; + mov.b64 {%r4872, %r4873}, %rd1157; + cvt.rn.bf16x2.f32 %r686, %r4873, %r4872; + mov.b64 {%r4874, %r4875}, %rd1160; + cvt.rn.bf16x2.f32 %r687, %r4875, %r4874; + mov.b64 {%r4876, %r4877}, %rd1163; + cvt.rn.bf16x2.f32 %r688, %r4877, %r4876; + mov.b64 {%r4878, %r4879}, %rd1166; + cvt.rn.bf16x2.f32 %r689, %r4879, %r4878; + mov.b64 {%r4880, %r4881}, %rd1169; + cvt.rn.bf16x2.f32 %r690, %r4881, %r4880; + mov.b64 {%r4882, %r4883}, %rd1172; + cvt.rn.bf16x2.f32 %r691, %r4883, %r4882; + mov.b64 {%r4884, %r4885}, %rd1175; + cvt.rn.bf16x2.f32 %r692, %r4885, %r4884; + mov.b64 {%r4886, %r4887}, %rd1178; + cvt.rn.bf16x2.f32 %r693, %r4887, %r4886; + mov.b64 {%r4888, %r4889}, %rd1181; + cvt.rn.bf16x2.f32 %r694, %r4889, %r4888; + mov.b64 {%r4890, %r4891}, %rd1184; + cvt.rn.bf16x2.f32 %r695, %r4891, %r4890; + mov.b64 {%r4892, %r4893}, %rd1187; + cvt.rn.bf16x2.f32 %r696, %r4893, %r4892; + mov.b64 {%r4894, %r4895}, %rd1190; + cvt.rn.bf16x2.f32 %r697, %r4895, %r4894; + mov.b64 {%r4896, %r4897}, %rd1193; + cvt.rn.bf16x2.f32 %r698, %r4897, %r4896; + mov.b64 {%r4898, %r4899}, %rd1196; + cvt.rn.bf16x2.f32 %r699, %r4899, %r4898; + mov.b64 {%r4900, %r4901}, %rd1199; + cvt.rn.bf16x2.f32 %r700, %r4901, %r4900; + mov.b64 {%r4902, %r4903}, %rd1202; + cvt.rn.bf16x2.f32 %r701, %r4903, %r4902; + mov.b64 {%r4904, %r4905}, %rd1205; + cvt.rn.bf16x2.f32 %r702, %r4905, %r4904; + mov.b64 {%r4906, %r4907}, %rd1208; + cvt.rn.bf16x2.f32 %r703, %r4907, %r4906; + mov.b64 {%r4908, %r4909}, %rd1211; + cvt.rn.bf16x2.f32 %r704, %r4909, %r4908; + mov.b64 {%r4910, %r4911}, %rd1214; + cvt.rn.bf16x2.f32 %r705, %r4911, %r4910; + mov.b64 {%r4912, %r4913}, %rd1217; + cvt.rn.bf16x2.f32 %r706, %r4913, %r4912; + mov.b64 {%r4914, %r4915}, %rd1220; + cvt.rn.bf16x2.f32 %r707, %r4915, %r4914; + mov.b64 {%r4916, %r4917}, %rd1223; + cvt.rn.bf16x2.f32 %r708, %r4917, %r4916; + mov.b64 {%r4918, %r4919}, %rd1226; + cvt.rn.bf16x2.f32 %r709, %r4919, %r4918; + mov.b64 {%r4920, %r4921}, %rd1229; + cvt.rn.bf16x2.f32 %r710, %r4921, %r4920; + mov.b64 {%r4922, %r4923}, %rd1232; + cvt.rn.bf16x2.f32 %r711, %r4923, %r4922; + mov.b64 {%r4924, %r4925}, %rd1235; + cvt.rn.bf16x2.f32 %r712, %r4925, %r4924; + mov.b64 {%r4926, %r4927}, %rd1238; + cvt.rn.bf16x2.f32 %r713, %r4927, %r4926; + mov.b64 {%r4928, %r4929}, %rd1241; + cvt.rn.bf16x2.f32 %r714, %r4929, %r4928; + mov.b64 {%r4930, %r4931}, %rd1244; + cvt.rn.bf16x2.f32 %r715, %r4931, %r4930; + mov.b64 {%r4932, %r4933}, %rd1247; + cvt.rn.bf16x2.f32 %r716, %r4933, %r4932; + mov.b64 {%r4934, %r4935}, %rd1250; + cvt.rn.bf16x2.f32 %r717, %r4935, %r4934; + mov.b64 {%r4936, %r4937}, %rd1253; + cvt.rn.bf16x2.f32 %r718, %r4937, %r4936; + mov.b64 {%r4938, %r4939}, %rd1256; + cvt.rn.bf16x2.f32 %r719, %r4939, %r4938; + mov.b64 {%r4940, %r4941}, %rd1259; + cvt.rn.bf16x2.f32 %r720, %r4941, %r4940; + mov.b64 {%r4942, %r4943}, %rd1262; + cvt.rn.bf16x2.f32 %r721, %r4943, %r4942; + mov.b64 {%r4944, %r4945}, %rd1265; + cvt.rn.bf16x2.f32 %r722, %r4945, %r4944; + mov.b64 {%r4946, %r4947}, %rd1268; + cvt.rn.bf16x2.f32 %r723, %r4947, %r4946; + mov.b64 {%r4948, %r4949}, %rd1271; + cvt.rn.bf16x2.f32 %r724, %r4949, %r4948; + mov.b64 {%r4950, %r4951}, %rd1274; + cvt.rn.bf16x2.f32 %r725, %r4951, %r4950; + mov.b64 {%r4952, %r4953}, %rd1277; + cvt.rn.bf16x2.f32 %r726, %r4953, %r4952; + mov.b64 {%r4954, %r4955}, %rd1280; + cvt.rn.bf16x2.f32 %r727, %r4955, %r4954; + mov.b64 {%r4956, %r4957}, %rd1283; + cvt.rn.bf16x2.f32 %r728, %r4957, %r4956; + mov.b64 {%r4958, %r4959}, %rd1286; + cvt.rn.bf16x2.f32 %r729, %r4959, %r4958; + mov.b64 {%r4960, %r4961}, %rd1289; + cvt.rn.bf16x2.f32 %r730, %r4961, %r4960; + mov.b64 {%r4962, %r4963}, %rd1292; + cvt.rn.bf16x2.f32 %r731, %r4963, %r4962; + mov.b64 {%r4964, %r4965}, %rd1295; + cvt.rn.bf16x2.f32 %r732, %r4965, %r4964; + mov.b64 {%r4966, %r4967}, %rd1298; + cvt.rn.bf16x2.f32 %r733, %r4967, %r4966; + mov.b64 {%r4968, %r4969}, %rd1301; + cvt.rn.bf16x2.f32 %r734, %r4969, %r4968; + mov.b64 {%r4970, %r4971}, %rd1304; + cvt.rn.bf16x2.f32 %r735, %r4971, %r4970; + mov.b64 {%r4972, %r4973}, %rd1307; + cvt.rn.bf16x2.f32 %r736, %r4973, %r4972; + mov.b64 {%r4974, %r4975}, %rd1310; + cvt.rn.bf16x2.f32 %r737, %r4975, %r4974; + mov.b64 {%r4976, %r4977}, %rd1313; + cvt.rn.bf16x2.f32 %r738, %r4977, %r4976; + mov.b64 {%r4978, %r4979}, %rd1316; + cvt.rn.bf16x2.f32 %r739, %r4979, %r4978; + mov.b64 {%r4980, %r4981}, %rd1319; + cvt.rn.bf16x2.f32 %r740, %r4981, %r4980; + mov.b64 {%r4982, %r4983}, %rd1322; + cvt.rn.bf16x2.f32 %r741, %r4983, %r4982; + mov.b64 {%r4984, %r4985}, %rd1325; + cvt.rn.bf16x2.f32 %r742, %r4985, %r4984; + mov.b64 {%r4986, %r4987}, %rd1328; + cvt.rn.bf16x2.f32 %r743, %r4987, %r4986; + mov.b64 {%r4988, %r4989}, %rd1331; + cvt.rn.bf16x2.f32 %r744, %r4989, %r4988; + mov.b64 {%r4990, %r4991}, %rd1334; + cvt.rn.bf16x2.f32 %r745, %r4991, %r4990; + mov.b64 {%r4992, %r4993}, %rd1337; + cvt.rn.bf16x2.f32 %r746, %r4993, %r4992; + mov.b64 {%r4994, %r4995}, %rd1340; + cvt.rn.bf16x2.f32 %r747, %r4995, %r4994; + mov.b64 {%r4996, %r4997}, %rd1343; + cvt.rn.bf16x2.f32 %r748, %r4997, %r4996; + mov.b64 {%r4998, %r4999}, %rd1346; + cvt.rn.bf16x2.f32 %r749, %r4999, %r4998; + mov.b64 {%r5000, %r5001}, %rd1349; + cvt.rn.bf16x2.f32 %r750, %r5001, %r5000; + mov.b64 {%r5002, %r5003}, %rd1352; + cvt.rn.bf16x2.f32 %r751, %r5003, %r5002; + mov.b64 {%r5004, %r5005}, %rd1355; + cvt.rn.bf16x2.f32 %r752, %r5005, %r5004; + mov.b64 {%r5006, %r5007}, %rd1358; + cvt.rn.bf16x2.f32 %r753, %r5007, %r5006; + mov.b64 {%r5008, %r5009}, %rd1361; + cvt.rn.bf16x2.f32 %r754, %r5009, %r5008; + mov.b64 {%r5010, %r5011}, %rd1364; + cvt.rn.bf16x2.f32 %r755, %r5011, %r5010; + mov.b64 {%r5012, %r5013}, %rd1367; + cvt.rn.bf16x2.f32 %r756, %r5013, %r5012; + mov.b64 {%r5014, %r5015}, %rd1370; + cvt.rn.bf16x2.f32 %r757, %r5015, %r5014; + mov.b64 {%r5016, %r5017}, %rd1373; + cvt.rn.bf16x2.f32 %r758, %r5017, %r5016; + mov.b64 {%r5018, %r5019}, %rd1376; + cvt.rn.bf16x2.f32 %r759, %r5019, %r5018; + mov.b64 {%r5020, %r5021}, %rd1379; + cvt.rn.bf16x2.f32 %r760, %r5021, %r5020; + mov.b64 {%r5022, %r5023}, %rd1382; + cvt.rn.bf16x2.f32 %r761, %r5023, %r5022; + mov.b64 {%r5024, %r5025}, %rd1385; + cvt.rn.bf16x2.f32 %r762, %r5025, %r5024; + mov.b64 {%r5026, %r5027}, %rd1388; + cvt.rn.bf16x2.f32 %r763, %r5027, %r5026; + mov.b64 {%r5028, %r5029}, %rd1391; + cvt.rn.bf16x2.f32 %r764, %r5029, %r5028; + mov.b64 {%r5030, %r5031}, %rd1394; + cvt.rn.bf16x2.f32 %r765, %r5031, %r5030; + mov.b64 {%r5032, %r5033}, %rd1397; + cvt.rn.bf16x2.f32 %r766, %r5033, %r5032; + mov.b64 {%r5034, %r5035}, %rd1400; + cvt.rn.bf16x2.f32 %r767, %r5035, %r5034; + mov.b64 {%r5036, %r5037}, %rd1403; + cvt.rn.bf16x2.f32 %r768, %r5037, %r5036; + mov.b64 {%r5038, %r5039}, %rd1406; + cvt.rn.bf16x2.f32 %r769, %r5039, %r5038; + mov.b64 {%r5040, %r5041}, %rd1409; + cvt.rn.bf16x2.f32 %r770, %r5041, %r5040; + mov.b64 {%r5042, %r5043}, %rd1412; + cvt.rn.bf16x2.f32 %r771, %r5043, %r5042; + mov.b64 {%r5044, %r5045}, %rd1415; + cvt.rn.bf16x2.f32 %r772, %r5045, %r5044; + mov.b64 {%r5046, %r5047}, %rd1418; + cvt.rn.bf16x2.f32 %r773, %r5047, %r5046; + .loc 1 108 19 // l2norm.py:108:19 + cvt.rn.bf16x2.f32 %r518, %r4531, %r4530; + cvt.rn.bf16x2.f32 %r519, %r4535, %r4534; + cvt.rn.bf16x2.f32 %r520, %r4539, %r4538; + cvt.rn.bf16x2.f32 %r521, %r4543, %r4542; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd133 + 0 ], { %r518, %r519, %r520, %r521 }; + // end inline asm + // begin inline asm + @%p2 st.global.v4.b32 [ %rd134 + 0 ], { %r522, %r523, %r524, %r525 }; + // end inline asm + // begin inline asm + @%p3 st.global.v4.b32 [ %rd135 + 0 ], { %r526, %r527, %r528, %r529 }; + // end inline asm + // begin inline asm + @%p4 st.global.v4.b32 [ %rd136 + 0 ], { %r530, %r531, %r532, %r533 }; + // end inline asm + // begin inline asm + @%p5 st.global.v4.b32 [ %rd137 + 0 ], { %r534, %r535, %r536, %r537 }; + // end inline asm + // begin inline asm + @%p6 st.global.v4.b32 [ %rd138 + 0 ], { %r538, %r539, %r540, %r541 }; + // end inline asm + // begin inline asm + @%p7 st.global.v4.b32 [ %rd139 + 0 ], { %r542, %r543, %r544, %r545 }; + // end inline asm + // begin inline asm + @%p8 st.global.v4.b32 [ %rd140 + 0 ], { %r546, %r547, %r548, %r549 }; + // end inline asm + // begin inline asm + @%p9 st.global.v4.b32 [ %rd141 + 0 ], { %r550, %r551, %r552, %r553 }; + // end inline asm + // begin inline asm + @%p10 st.global.v4.b32 [ %rd142 + 0 ], { %r554, %r555, %r556, %r557 }; + // end inline asm + // begin inline asm + @%p11 st.global.v4.b32 [ %rd143 + 0 ], { %r558, %r559, %r560, %r561 }; + // end inline asm + // begin inline asm + @%p39 st.global.v4.b32 [ %rd144 + 0 ], { %r562, %r563, %r564, %r565 }; + // end inline asm + // begin inline asm + @%p12 st.global.v4.b32 [ %rd145 + 0 ], { %r566, %r567, %r568, %r569 }; + // end inline asm + // begin inline asm + @%p13 st.global.v4.b32 [ %rd146 + 0 ], { %r570, %r571, %r572, %r573 }; + // end inline asm + // begin inline asm + @%p14 st.global.v4.b32 [ %rd147 + 0 ], { %r574, %r575, %r576, %r577 }; + // end inline asm + // begin inline asm + @%p15 st.global.v4.b32 [ %rd148 + 0 ], { %r578, %r579, %r580, %r581 }; + // end inline asm + // begin inline asm + @%p16 st.global.v4.b32 [ %rd149 + 0 ], { %r582, %r583, %r584, %r585 }; + // end inline asm + // begin inline asm + @%p17 st.global.v4.b32 [ %rd150 + 0 ], { %r586, %r587, %r588, %r589 }; + // end inline asm + // begin inline asm + @%p18 st.global.v4.b32 [ %rd151 + 0 ], { %r590, %r591, %r592, %r593 }; + // end inline asm + // begin inline asm + @%p40 st.global.v4.b32 [ %rd152 + 0 ], { %r594, %r595, %r596, %r597 }; + // end inline asm + // begin inline asm + @%p19 st.global.v4.b32 [ %rd153 + 0 ], { %r598, %r599, %r600, %r601 }; + // end inline asm + // begin inline asm + @%p20 st.global.v4.b32 [ %rd154 + 0 ], { %r602, %r603, %r604, %r605 }; + // end inline asm + // begin inline asm + @%p21 st.global.v4.b32 [ %rd155 + 0 ], { %r606, %r607, %r608, %r609 }; + // end inline asm + // begin inline asm + @%p22 st.global.v4.b32 [ %rd156 + 0 ], { %r610, %r611, %r612, %r613 }; + // end inline asm + // begin inline asm + @%p23 st.global.v4.b32 [ %rd157 + 0 ], { %r614, %r615, %r616, %r617 }; + // end inline asm + // begin inline asm + @%p24 st.global.v4.b32 [ %rd158 + 0 ], { %r618, %r619, %r620, %r621 }; + // end inline asm + // begin inline asm + @%p25 st.global.v4.b32 [ %rd159 + 0 ], { %r622, %r623, %r624, %r625 }; + // end inline asm + // begin inline asm + @%p26 st.global.v4.b32 [ %rd160 + 0 ], { %r626, %r627, %r628, %r629 }; + // end inline asm + // begin inline asm + @%p27 st.global.v4.b32 [ %rd161 + 0 ], { %r630, %r631, %r632, %r633 }; + // end inline asm + // begin inline asm + @%p28 st.global.v4.b32 [ %rd162 + 0 ], { %r634, %r635, %r636, %r637 }; + // end inline asm + // begin inline asm + @%p29 st.global.v4.b32 [ %rd163 + 0 ], { %r638, %r639, %r640, %r641 }; + // end inline asm + // begin inline asm + @%p30 st.global.v4.b32 [ %rd164 + 0 ], { %r642, %r643, %r644, %r645 }; + // end inline asm + // begin inline asm + @%p31 st.global.v4.b32 [ %rd165 + 0 ], { %r646, %r647, %r648, %r649 }; + // end inline asm + // begin inline asm + @%p32 st.global.v4.b32 [ %rd166 + 0 ], { %r650, %r651, %r652, %r653 }; + // end inline asm + // begin inline asm + @%p33 st.global.v4.b32 [ %rd167 + 0 ], { %r654, %r655, %r656, %r657 }; + // end inline asm + // begin inline asm + @%p34 st.global.v4.b32 [ %rd168 + 0 ], { %r658, %r659, %r660, %r661 }; + // end inline asm + // begin inline asm + @%p41 st.global.v4.b32 [ %rd169 + 0 ], { %r662, %r663, %r664, %r665 }; + // end inline asm + // begin inline asm + @%p42 st.global.v4.b32 [ %rd170 + 0 ], { %r666, %r667, %r668, %r669 }; + // end inline asm + // begin inline asm + @%p43 st.global.v4.b32 [ %rd171 + 0 ], { %r670, %r671, %r672, %r673 }; + // end inline asm + // begin inline asm + @%p44 st.global.v4.b32 [ %rd172 + 0 ], { %r674, %r675, %r676, %r677 }; + // end inline asm + // begin inline asm + @%p45 st.global.v4.b32 [ %rd173 + 0 ], { %r678, %r679, %r680, %r681 }; + // end inline asm + // begin inline asm + @%p46 st.global.v4.b32 [ %rd174 + 0 ], { %r682, %r683, %r684, %r685 }; + // end inline asm + // begin inline asm + @%p47 st.global.v4.b32 [ %rd175 + 0 ], { %r686, %r687, %r688, %r689 }; + // end inline asm + // begin inline asm + @%p48 st.global.v4.b32 [ %rd176 + 0 ], { %r690, %r691, %r692, %r693 }; + // end inline asm + // begin inline asm + @%p49 st.global.v4.b32 [ %rd177 + 0 ], { %r694, %r695, %r696, %r697 }; + // end inline asm + // begin inline asm + @%p50 st.global.v4.b32 [ %rd178 + 0 ], { %r698, %r699, %r700, %r701 }; + // end inline asm + // begin inline asm + @%p51 st.global.v4.b32 [ %rd179 + 0 ], { %r702, %r703, %r704, %r705 }; + // end inline asm + // begin inline asm + @%p52 st.global.v4.b32 [ %rd180 + 0 ], { %r706, %r707, %r708, %r709 }; + // end inline asm + // begin inline asm + @%p53 st.global.v4.b32 [ %rd181 + 0 ], { %r710, %r711, %r712, %r713 }; + // end inline asm + // begin inline asm + @%p54 st.global.v4.b32 [ %rd182 + 0 ], { %r714, %r715, %r716, %r717 }; + // end inline asm + // begin inline asm + @%p55 st.global.v4.b32 [ %rd183 + 0 ], { %r718, %r719, %r720, %r721 }; + // end inline asm + // begin inline asm + @%p56 st.global.v4.b32 [ %rd184 + 0 ], { %r722, %r723, %r724, %r725 }; + // end inline asm + // begin inline asm + @%p57 st.global.v4.b32 [ %rd185 + 0 ], { %r726, %r727, %r728, %r729 }; + // end inline asm + // begin inline asm + @%p58 st.global.v4.b32 [ %rd186 + 0 ], { %r730, %r731, %r732, %r733 }; + // end inline asm + // begin inline asm + @%p59 st.global.v4.b32 [ %rd187 + 0 ], { %r734, %r735, %r736, %r737 }; + // end inline asm + // begin inline asm + @%p60 st.global.v4.b32 [ %rd188 + 0 ], { %r738, %r739, %r740, %r741 }; + // end inline asm + // begin inline asm + @%p61 st.global.v4.b32 [ %rd189 + 0 ], { %r742, %r743, %r744, %r745 }; + // end inline asm + // begin inline asm + @%p62 st.global.v4.b32 [ %rd190 + 0 ], { %r746, %r747, %r748, %r749 }; + // end inline asm + // begin inline asm + @%p63 st.global.v4.b32 [ %rd191 + 0 ], { %r750, %r751, %r752, %r753 }; + // end inline asm + // begin inline asm + @%p64 st.global.v4.b32 [ %rd192 + 0 ], { %r754, %r755, %r756, %r757 }; + // end inline asm + // begin inline asm + @%p65 st.global.v4.b32 [ %rd193 + 0 ], { %r758, %r759, %r760, %r761 }; + // end inline asm + // begin inline asm + @%p66 st.global.v4.b32 [ %rd194 + 0 ], { %r762, %r763, %r764, %r765 }; + // end inline asm + // begin inline asm + @%p67 st.global.v4.b32 [ %rd195 + 0 ], { %r766, %r767, %r768, %r769 }; + // end inline asm + // begin inline asm + @%p68 st.global.v4.b32 [ %rd196 + 0 ], { %r770, %r771, %r772, %r773 }; + // end inline asm + .loc 1 108 4 // l2norm.py:108:4 + ret; +$L__tmp1079: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 98 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp1078 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 107 // DW_AT_call_line +.b8 55 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp1078 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.source b/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..35265513bc03aa2afd3347f28178370dc92e754e --- /dev/null +++ b/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.source @@ -0,0 +1,210 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc47 = loc(unknown) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc54 = loc("y"(#loc)) +#loc55 = loc("rstd"(#loc)) +#loc56 = loc("dy"(#loc)) +#loc57 = loc("dx"(#loc)) +#loc58 = loc("eps"(#loc)) +#loc59 = loc("T"(#loc)) +#loc101 = loc("input"(#loc45)) +#loc102 = loc("a"(#loc50)) +#loc103 = loc("b"(#loc50)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc60) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc61) + %o_t = arith.constant 128 : i32 loc(#loc62) + %o_t_1 = arith.constant 128 : i64 loc(#loc62) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc62) + %o_t_3 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc63) + %o_t_4 = arith.extsi %o_t_3 : tensor<128xi32> to tensor<128xi64> loc(#loc64) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<128xi64> loc(#loc64) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<128xi64> loc(#loc64) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc65) + %m_t = arith.extsi %T : i32 to i64 loc(#loc66) + %m_t_7 = tt.splat %m_t : i64 -> tensor<128xi64> loc(#loc66) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<128xi64> loc(#loc66) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<128xi1> -> tensor<128x1xi1> loc(#loc67) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc68) + %m_x_10 = arith.constant 128 : i32 loc(#loc69) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc69) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc69) + %m_x_13 = tt.broadcast %m_x : tensor<128x1xi1> -> tensor<128x128xi1> loc(#loc70) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<128x128xi1> loc(#loc70) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<128x128xi1> loc(#loc70) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc71) + %p_y_16 = arith.constant 128 : i32 loc(#loc72) + %p_y_17 = arith.constant 128 : i64 loc(#loc72) + %p_y_18 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc72) + %p_y_19 = arith.muli %p_y, %p_y_18 : tensor<128x1xi64> loc(#loc72) + %p_y_20 = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc73) + %p_y_21 = tt.addptr %p_y_20, %p_y_19 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc73) + %p_y_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc74) + %p_y_23 = tt.broadcast %p_y_21 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc75) + %p_y_24 = tt.broadcast %p_y_22 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc75) + %p_y_25 = tt.addptr %p_y_23, %p_y_24 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc75) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr> loc(#loc76) + %p_rstd_26 = tt.addptr %p_rstd, %o_t_6 : tensor<128x!tt.ptr>, tensor<128xi64> loc(#loc76) + %p_dy = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc77) + %p_dy_27 = arith.constant 128 : i32 loc(#loc78) + %p_dy_28 = arith.constant 128 : i64 loc(#loc78) + %p_dy_29 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc78) + %p_dy_30 = arith.muli %p_dy, %p_dy_29 : tensor<128x1xi64> loc(#loc78) + %p_dy_31 = tt.splat %dy : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc79) + %p_dy_32 = tt.addptr %p_dy_31, %p_dy_30 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc79) + %p_dy_33 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc80) + %p_dy_34 = tt.broadcast %p_dy_32 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc81) + %p_dy_35 = tt.broadcast %p_dy_33 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc81) + %p_dy_36 = tt.addptr %p_dy_34, %p_dy_35 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc81) + %p_dx = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc82) + %p_dx_37 = arith.constant 128 : i32 loc(#loc83) + %p_dx_38 = arith.constant 128 : i64 loc(#loc83) + %p_dx_39 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc83) + %p_dx_40 = arith.muli %p_dx, %p_dx_39 : tensor<128x1xi64> loc(#loc83) + %p_dx_41 = tt.splat %dx : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc84) + %p_dx_42 = tt.addptr %p_dx_41, %p_dx_40 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc84) + %p_dx_43 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc85) + %p_dx_44 = tt.broadcast %p_dx_42 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc86) + %p_dx_45 = tt.broadcast %p_dx_43 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc86) + %p_dx_46 = tt.addptr %p_dx_44, %p_dx_45 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc86) + %b_y = arith.constant 0.000000e+00 : f32 loc(#loc87) + %b_y_47 = arith.constant dense<0.000000e+00> : tensor<128x128xf32> loc(#loc87) + %b_y_48 = arith.truncf %b_y_47 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc87) + %b_y_49 = tt.load %p_y_25, %m_x_15, %b_y_48 : tensor<128x128x!tt.ptr> loc(#loc87) + %b_y_50 = arith.extf %b_y_49 : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc88) + %b_rstd = arith.constant 0.000000e+00 : f32 loc(#loc89) + %b_rstd_51 = arith.constant dense<0.000000e+00> : tensor<128xf32> loc(#loc89) + %b_rstd_52 = tt.load %p_rstd_26, %m_t_8, %b_rstd_51 : tensor<128x!tt.ptr> loc(#loc89) + %b_dy = arith.constant 0.000000e+00 : f32 loc(#loc90) + %b_dy_53 = arith.constant dense<0.000000e+00> : tensor<128x128xf32> loc(#loc90) + %b_dy_54 = arith.truncf %b_dy_53 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc90) + %b_dy_55 = tt.load %p_dy_36, %m_x_15, %b_dy_54 : tensor<128x128x!tt.ptr> loc(#loc90) + %b_dy_56 = arith.extf %b_dy_55 : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc91) + %b_dx = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc92) + %b_dx_57 = tt.broadcast %b_dx : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc93) + %b_dx_58 = arith.mulf %b_dy_56, %b_dx_57 : tensor<128x128xf32> loc(#loc93) + %b_dx_59 = arith.mulf %b_dy_56, %b_y_50 : tensor<128x128xf32> loc(#loc94) + %b_dx_60 = tt.call @triton.language.standard.sum__fp32S128_128S_c1_cFalse_cNone(%b_dx_59) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc95) + %b_dx_61 = tt.expand_dims %b_dx_60 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc96) + %b_dx_62 = tt.broadcast %b_dx_61 : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc97) + %b_dx_63 = arith.mulf %b_dx_62, %b_y_50 : tensor<128x128xf32> loc(#loc97) + %b_dx_64 = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc98) + %b_dx_65 = tt.broadcast %b_dx_64 : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc99) + %b_dx_66 = arith.mulf %b_dx_63, %b_dx_65 : tensor<128x128xf32> loc(#loc99) + %b_dx_67 = arith.subf %b_dx_58, %b_dx_66 : tensor<128x128xf32> loc(#loc100) + %0 = arith.truncf %b_dx_67 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc42) + tt.store %p_dx_46, %0, %m_x_15 : tensor<128x128x!tt.ptr> loc(#loc43) + tt.return loc(#loc44) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S128_128S_c1_cFalse_cNone(%input: tensor<128x128xf32> loc("input"(#loc45))) -> tensor<128xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc46) + tt.reduce.return %2 : f32 loc(#loc46) + }) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc46) + tt.return %0 : tensor<128xf32> loc(#loc48) + ^bb1: // no predecessors + %1 = ub.poison : tensor<128xf32> loc(#loc49) + tt.return %1 : tensor<128xf32> loc(#loc49) + } loc(#loc45) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc50)), %b: f32 loc("b"(#loc50))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc51) + tt.return %0 : f32 loc(#loc52) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc53) + tt.return %1 : f32 loc(#loc53) + } loc(#loc50) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:31) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:39) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:20) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:31) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:39) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:82) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc60 = loc("i_t"(#loc1)) +#loc61 = loc("i_t"(#loc2)) +#loc62 = loc("o_t"(#loc3)) +#loc63 = loc("o_t"(#loc4)) +#loc64 = loc("o_t"(#loc5)) +#loc65 = loc("o_d"(#loc6)) +#loc66 = loc("m_t"(#loc7)) +#loc67 = loc("m_x"(#loc8)) +#loc68 = loc("m_x"(#loc9)) +#loc69 = loc("m_x"(#loc10)) +#loc70 = loc("m_x"(#loc11)) +#loc71 = loc("p_y"(#loc12)) +#loc72 = loc("p_y"(#loc13)) +#loc73 = loc("p_y"(#loc14)) +#loc74 = loc("p_y"(#loc15)) +#loc75 = loc("p_y"(#loc16)) +#loc76 = loc("p_rstd"(#loc17)) +#loc77 = loc("p_dy"(#loc18)) +#loc78 = loc("p_dy"(#loc19)) +#loc79 = loc("p_dy"(#loc20)) +#loc80 = loc("p_dy"(#loc21)) +#loc81 = loc("p_dy"(#loc22)) +#loc82 = loc("p_dx"(#loc23)) +#loc83 = loc("p_dx"(#loc24)) +#loc84 = loc("p_dx"(#loc25)) +#loc85 = loc("p_dx"(#loc26)) +#loc86 = loc("p_dx"(#loc27)) +#loc87 = loc("b_y"(#loc28)) +#loc88 = loc("b_y"(#loc29)) +#loc89 = loc("b_rstd"(#loc30)) +#loc90 = loc("b_dy"(#loc31)) +#loc91 = loc("b_dy"(#loc32)) +#loc92 = loc("b_dx"(#loc33)) +#loc93 = loc("b_dx"(#loc34)) +#loc94 = loc("b_dx"(#loc35)) +#loc95 = loc("b_dx"(#loc36)) +#loc96 = loc("b_dx"(#loc37)) +#loc97 = loc("b_dx"(#loc38)) +#loc98 = loc("b_dx"(#loc39)) +#loc99 = loc("b_dx"(#loc40)) +#loc100 = loc("b_dx"(#loc41)) diff --git a/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.ttgir b/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..30b8d0f11ffeced21ac41468aae9ea13197ceffd --- /dev/null +++ b/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.ttgir @@ -0,0 +1,154 @@ +#blocked = #ttg.blocked<{sizePerThread = [4], threadsPerWarp = [32], warpsPerCTA = [1], order = [0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [1, 1], order = [1, 0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc39 = loc("y"(#loc)) +#loc40 = loc("rstd"(#loc)) +#loc41 = loc("dy"(#loc)) +#loc42 = loc("dx"(#loc)) +#loc43 = loc("eps"(#loc)) +#loc44 = loc("T"(#loc)) +#loc72 = loc("b_dx"(#loc30)) +#loc78 = loc(callsite(#loc1 at #loc72)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 1 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %c128_i64 = arith.constant 128 : i64 loc(#loc1) + %cst = arith.constant dense<0.000000e+00> : tensor<128xf32, #blocked> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<128x128xbf16, #blocked1> loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<128x1xi64, #blocked1> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<1x128xi32, #blocked1> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc45) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc46) + %o_t = arith.muli %i_t_3, %c128_i64 : i64 loc(#loc47) + %o_t_4 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc48) + %o_t_5 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #blocked> loc(#loc48) + %o_t_6 = arith.extsi %o_t_4 : tensor<128xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> to tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc49) + %o_t_7 = arith.extsi %o_t_5 : tensor<128xi32, #blocked> to tensor<128xi64, #blocked> loc(#loc49) + %o_t_8 = tt.splat %o_t : i64 -> tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc49) + %o_t_9 = tt.splat %o_t : i64 -> tensor<128xi64, #blocked> loc(#loc49) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc49) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<128xi64, #blocked> loc(#loc49) + %m_t = arith.extsi %T : i32 to i64 loc(#loc50) + %m_t_12 = tt.splat %m_t : i64 -> tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc50) + %m_t_13 = tt.splat %m_t : i64 -> tensor<128xi64, #blocked> loc(#loc50) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc50) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<128xi64, #blocked> loc(#loc50) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<128xi1, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<128x1xi1, #blocked1> loc(#loc51) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc52) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x128xi32, #blocked1> loc(#loc52) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_2 : tensor<1x128xi32, #blocked1> loc(#loc53) + %m_x_19 = tt.broadcast %m_x : tensor<128x1xi1, #blocked1> -> tensor<128x128xi1, #blocked1> loc(#loc54) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked1> -> tensor<128x128xi1, #blocked1> loc(#loc54) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<128x128xi1, #blocked1> loc(#loc54) + %p_y = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<128x1xi64, #blocked1> loc(#loc55) + %p_y_22 = arith.muli %p_y, %cst_1 : tensor<128x1xi64, #blocked1> loc(#loc56) + %p_y_23 = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr, #blocked1> loc(#loc57) + %p_y_24 = tt.addptr %p_y_23, %p_y_22 : tensor<128x1x!tt.ptr, #blocked1>, tensor<128x1xi64, #blocked1> loc(#loc57) + %p_y_25 = tt.broadcast %p_y_24 : tensor<128x1x!tt.ptr, #blocked1> -> tensor<128x128x!tt.ptr, #blocked1> loc(#loc58) + %p_y_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked1> -> tensor<128x128xi32, #blocked1> loc(#loc58) + %p_y_27 = tt.addptr %p_y_25, %p_y_26 : tensor<128x128x!tt.ptr, #blocked1>, tensor<128x128xi32, #blocked1> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr, #blocked> loc(#loc59) + %p_rstd_28 = tt.addptr %p_rstd, %o_t_11 : tensor<128x!tt.ptr, #blocked>, tensor<128xi64, #blocked> loc(#loc59) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<128x1x!tt.ptr, #blocked1> loc(#loc60) + %p_dy_29 = tt.addptr %p_dy, %p_y_22 : tensor<128x1x!tt.ptr, #blocked1>, tensor<128x1xi64, #blocked1> loc(#loc60) + %p_dy_30 = tt.broadcast %p_dy_29 : tensor<128x1x!tt.ptr, #blocked1> -> tensor<128x128x!tt.ptr, #blocked1> loc(#loc61) + %p_dy_31 = tt.addptr %p_dy_30, %p_y_26 : tensor<128x128x!tt.ptr, #blocked1>, tensor<128x128xi32, #blocked1> loc(#loc61) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<128x1x!tt.ptr, #blocked1> loc(#loc62) + %p_dx_32 = tt.addptr %p_dx, %p_y_22 : tensor<128x1x!tt.ptr, #blocked1>, tensor<128x1xi64, #blocked1> loc(#loc62) + %p_dx_33 = tt.broadcast %p_dx_32 : tensor<128x1x!tt.ptr, #blocked1> -> tensor<128x128x!tt.ptr, #blocked1> loc(#loc63) + %p_dx_34 = tt.addptr %p_dx_33, %p_y_26 : tensor<128x128x!tt.ptr, #blocked1>, tensor<128x128xi32, #blocked1> loc(#loc63) + %b_y = tt.load %p_y_27, %m_x_21, %cst_0 : tensor<128x128x!tt.ptr, #blocked1> loc(#loc64) + %b_y_35 = arith.extf %b_y : tensor<128x128xbf16, #blocked1> to tensor<128x128xf32, #blocked1> loc(#loc65) + %b_rstd = tt.load %p_rstd_28, %m_t_15, %cst : tensor<128x!tt.ptr, #blocked> loc(#loc66) + %b_dy = tt.load %p_dy_31, %m_x_21, %cst_0 : tensor<128x128x!tt.ptr, #blocked1> loc(#loc67) + %b_dy_36 = arith.extf %b_dy : tensor<128x128xbf16, #blocked1> to tensor<128x128xf32, #blocked1> loc(#loc68) + %b_dx = ttg.convert_layout %b_rstd : tensor<128xf32, #blocked> -> tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc69) + %b_dx_37 = tt.expand_dims %b_dx {axis = 1 : i32} : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<128x1xf32, #blocked1> loc(#loc69) + %b_dx_38 = tt.broadcast %b_dx_37 : tensor<128x1xf32, #blocked1> -> tensor<128x128xf32, #blocked1> loc(#loc70) + %b_dx_39 = arith.mulf %b_dy_36, %b_dx_38 : tensor<128x128xf32, #blocked1> loc(#loc70) + %b_dx_40 = arith.mulf %b_dy_36, %b_y_35 : tensor<128x128xf32, #blocked1> loc(#loc71) + %b_dx_41 = "tt.reduce"(%b_dx_40) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_47: f32 loc(callsite(#loc1 at #loc72)), %b_dx_48: f32 loc(callsite(#loc1 at #loc72))): + %b_dx_49 = arith.addf %b_dx_47, %b_dx_48 : f32 loc(#loc79) + tt.reduce.return %b_dx_49 : f32 loc(#loc77) + }) : (tensor<128x128xf32, #blocked1>) -> tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc77) + %b_dx_42 = tt.expand_dims %b_dx_41 {axis = 1 : i32} : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<128x1xf32, #blocked1> loc(#loc73) + %b_dx_43 = tt.broadcast %b_dx_42 : tensor<128x1xf32, #blocked1> -> tensor<128x128xf32, #blocked1> loc(#loc74) + %b_dx_44 = arith.mulf %b_dx_43, %b_y_35 : tensor<128x128xf32, #blocked1> loc(#loc74) + %b_dx_45 = arith.mulf %b_dx_44, %b_dx_38 : tensor<128x128xf32, #blocked1> loc(#loc75) + %b_dx_46 = arith.subf %b_dx_39, %b_dx_45 : tensor<128x128xf32, #blocked1> loc(#loc76) + %0 = arith.truncf %b_dx_46 : tensor<128x128xf32, #blocked1> to tensor<128x128xbf16, #blocked1> loc(#loc36) + tt.store %p_dx_34, %0, %m_x_21 : tensor<128x128x!tt.ptr, #blocked1> loc(#loc37) + tt.return loc(#loc38) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc45 = loc("i_t"(#loc2)) +#loc46 = loc("i_t"(#loc3)) +#loc47 = loc("o_t"(#loc4)) +#loc48 = loc("o_t"(#loc5)) +#loc49 = loc("o_t"(#loc6)) +#loc50 = loc("m_t"(#loc7)) +#loc51 = loc("m_x"(#loc8)) +#loc52 = loc("m_x"(#loc9)) +#loc53 = loc("m_x"(#loc10)) +#loc54 = loc("m_x"(#loc11)) +#loc55 = loc("p_y"(#loc12)) +#loc56 = loc("p_y"(#loc13)) +#loc57 = loc("p_y"(#loc14)) +#loc58 = loc("p_y"(#loc15)) +#loc59 = loc("p_rstd"(#loc16)) +#loc60 = loc("p_dy"(#loc17)) +#loc61 = loc("p_dy"(#loc18)) +#loc62 = loc("p_dx"(#loc19)) +#loc63 = loc("p_dx"(#loc20)) +#loc64 = loc("b_y"(#loc21)) +#loc65 = loc("b_y"(#loc22)) +#loc66 = loc("b_rstd"(#loc23)) +#loc67 = loc("b_dy"(#loc24)) +#loc68 = loc("b_dy"(#loc25)) +#loc69 = loc("b_dx"(#loc26)) +#loc70 = loc("b_dx"(#loc27)) +#loc71 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc32)) +#loc74 = loc("b_dx"(#loc33)) +#loc75 = loc("b_dx"(#loc34)) +#loc76 = loc("b_dx"(#loc35)) +#loc77 = loc(callsite(#loc29 at #loc72)) +#loc79 = loc(callsite(#loc31 at #loc77)) diff --git a/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.ttir b/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..f242766466ced3ec80bd72ef6238b6accc66dbe7 --- /dev/null +++ b/triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.ttir @@ -0,0 +1,144 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0) +#loc1 = loc(unknown) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55) +#loc39 = loc("y"(#loc)) +#loc40 = loc("rstd"(#loc)) +#loc41 = loc("dy"(#loc)) +#loc42 = loc("dx"(#loc)) +#loc43 = loc("eps"(#loc)) +#loc44 = loc("T"(#loc)) +#loc72 = loc("b_dx"(#loc30)) +#loc78 = loc(callsite(#loc1 at #loc72)) +module { + tt.func public @l2norm_bwd_kernel(%y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<0.000000e+00> : tensor<128x128xbf16> loc(#loc1) + %b_rstd = arith.constant dense<0.000000e+00> : tensor<128xf32> loc(#loc45) + %cst_0 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc1) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc46) + %c128_i64 = arith.constant 128 : i64 loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc47) + %i_t_1 = arith.extsi %i_t : i32 to i64 loc(#loc48) + %o_t = arith.muli %i_t_1, %c128_i64 : i64 loc(#loc49) + %o_t_2 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc50) + %o_t_3 = arith.extsi %o_t_2 : tensor<128xi32> to tensor<128xi64> loc(#loc51) + %o_t_4 = tt.splat %o_t : i64 -> tensor<128xi64> loc(#loc51) + %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<128xi64> loc(#loc51) + %m_t = arith.extsi %T : i32 to i64 loc(#loc52) + %m_t_6 = tt.splat %m_t : i64 -> tensor<128xi64> loc(#loc52) + %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<128xi64> loc(#loc52) + %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<128xi1> -> tensor<128x1xi1> loc(#loc53) + %m_x_9 = tt.expand_dims %o_t_2 {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc54) + %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc46) + %m_x_11 = tt.broadcast %m_x_8 : tensor<128x1xi1> -> tensor<128x128xi1> loc(#loc55) + %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<128x128xi1> loc(#loc55) + %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<128x128xi1> loc(#loc55) + %p_y = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc56) + %p_y_14 = arith.muli %p_y, %cst_0 : tensor<128x1xi64> loc(#loc57) + %p_y_15 = tt.splat %y : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc58) + %p_y_16 = tt.addptr %p_y_15, %p_y_14 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc58) + %p_y_17 = tt.broadcast %p_y_16 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc59) + %p_y_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc59) + %p_y_19 = tt.addptr %p_y_17, %p_y_18 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc59) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<128x!tt.ptr> loc(#loc60) + %p_rstd_20 = tt.addptr %p_rstd, %o_t_5 : tensor<128x!tt.ptr>, tensor<128xi64> loc(#loc60) + %p_dy = tt.splat %dy : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc61) + %p_dy_21 = tt.addptr %p_dy, %p_y_14 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc61) + %p_dy_22 = tt.broadcast %p_dy_21 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc62) + %p_dy_23 = tt.addptr %p_dy_22, %p_y_18 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc62) + %p_dx = tt.splat %dx : !tt.ptr -> tensor<128x1x!tt.ptr> loc(#loc63) + %p_dx_24 = tt.addptr %p_dx, %p_y_14 : tensor<128x1x!tt.ptr>, tensor<128x1xi64> loc(#loc63) + %p_dx_25 = tt.broadcast %p_dx_24 : tensor<128x1x!tt.ptr> -> tensor<128x128x!tt.ptr> loc(#loc64) + %p_dx_26 = tt.addptr %p_dx_25, %p_y_18 : tensor<128x128x!tt.ptr>, tensor<128x128xi32> loc(#loc64) + %b_y = tt.load %p_y_19, %m_x_13, %cst : tensor<128x128x!tt.ptr> loc(#loc65) + %b_y_27 = arith.extf %b_y : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc66) + %b_rstd_28 = tt.load %p_rstd_20, %m_t_7, %b_rstd : tensor<128x!tt.ptr> loc(#loc45) + %b_dy = tt.load %p_dy_23, %m_x_13, %cst : tensor<128x128x!tt.ptr> loc(#loc67) + %b_dy_29 = arith.extf %b_dy : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc68) + %b_dx = tt.expand_dims %b_rstd_28 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc69) + %b_dx_30 = tt.broadcast %b_dx : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc70) + %b_dx_31 = arith.mulf %b_dy_29, %b_dx_30 : tensor<128x128xf32> loc(#loc70) + %b_dx_32 = arith.mulf %b_dy_29, %b_y_27 : tensor<128x128xf32> loc(#loc71) + %b_dx_33 = "tt.reduce"(%b_dx_32) <{axis = 1 : i32}> ({ + ^bb0(%b_dx_39: f32 loc(callsite(#loc1 at #loc72)), %b_dx_40: f32 loc(callsite(#loc1 at #loc72))): + %b_dx_41 = arith.addf %b_dx_39, %b_dx_40 : f32 loc(#loc79) + tt.reduce.return %b_dx_41 : f32 loc(#loc77) + }) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc77) + %b_dx_34 = tt.expand_dims %b_dx_33 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc73) + %b_dx_35 = tt.broadcast %b_dx_34 : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc74) + %b_dx_36 = arith.mulf %b_dx_35, %b_y_27 : tensor<128x128xf32> loc(#loc74) + %b_dx_37 = arith.mulf %b_dx_36, %b_dx_30 : tensor<128x128xf32> loc(#loc75) + %b_dx_38 = arith.subf %b_dx_31, %b_dx_37 : tensor<128x128xf32> loc(#loc76) + %0 = arith.truncf %b_dx_38 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc36) + tt.store %p_dx_26, %0, %m_x_13 : tensor<128x128x!tt.ptr> loc(#loc37) + tt.return loc(#loc38) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4) +#loc45 = loc("b_rstd"(#loc2)) +#loc46 = loc("m_x"(#loc3)) +#loc47 = loc("i_t"(#loc4)) +#loc48 = loc("i_t"(#loc5)) +#loc49 = loc("o_t"(#loc6)) +#loc50 = loc("o_t"(#loc7)) +#loc51 = loc("o_t"(#loc8)) +#loc52 = loc("m_t"(#loc9)) +#loc53 = loc("m_x"(#loc10)) +#loc54 = loc("m_x"(#loc11)) +#loc55 = loc("m_x"(#loc12)) +#loc56 = loc("p_y"(#loc13)) +#loc57 = loc("p_y"(#loc14)) +#loc58 = loc("p_y"(#loc15)) +#loc59 = loc("p_y"(#loc16)) +#loc60 = loc("p_rstd"(#loc17)) +#loc61 = loc("p_dy"(#loc18)) +#loc62 = loc("p_dy"(#loc19)) +#loc63 = loc("p_dx"(#loc20)) +#loc64 = loc("p_dx"(#loc21)) +#loc65 = loc("b_y"(#loc22)) +#loc66 = loc("b_y"(#loc23)) +#loc67 = loc("b_dy"(#loc24)) +#loc68 = loc("b_dy"(#loc25)) +#loc69 = loc("b_dx"(#loc26)) +#loc70 = loc("b_dx"(#loc27)) +#loc71 = loc("b_dx"(#loc28)) +#loc73 = loc("b_dx"(#loc32)) +#loc74 = loc("b_dx"(#loc33)) +#loc75 = loc("b_dx"(#loc34)) +#loc76 = loc("b_dx"(#loc35)) +#loc77 = loc(callsite(#loc29 at #loc72)) +#loc79 = loc(callsite(#loc31 at #loc77)) diff --git a/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/__grp__l2norm_fwd_kernel.json b/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/__grp__l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..9e7646e9a07f4a69f39500655e1f1dcadeb5b03f --- /dev/null +++ b/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/__grp__l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_fwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/l2norm_fwd_kernel.source", "l2norm_fwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/l2norm_fwd_kernel.ttir", "l2norm_fwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/l2norm_fwd_kernel.ttgir", "l2norm_fwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/l2norm_fwd_kernel.llir", "l2norm_fwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/l2norm_fwd_kernel.ptx", "l2norm_fwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/l2norm_fwd_kernel.cubin", "l2norm_fwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/l2norm_fwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/l2norm_fwd_kernel.cubin b/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/l2norm_fwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..1f89a2895b6326b97021c87a8058e85107b2f2e9 Binary files /dev/null and b/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/l2norm_fwd_kernel.cubin differ diff --git a/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/l2norm_fwd_kernel.json b/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..dec9aa0077d32b4800da64900894bdde0693e614 --- /dev/null +++ b/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"hash": "12a80e39702372ebe26ec0d84ad1288b94169a3a9b9302158d527b5ba7fab4f5", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 4, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 128, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_fwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/l2norm_fwd_kernel.llir b/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/l2norm_fwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..44be3da80c161d57bf55b2625361f13bf1780f63 --- /dev/null +++ b/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/l2norm_fwd_kernel.llir @@ -0,0 +1,633 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 +@.str.2 = private unnamed_addr constant [17 x i8] c"__CUDA_PREC_SQRT\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_fwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, float %3, i32 %4, ptr addrspace(1) readnone captures(none) %5, ptr addrspace(1) readnone captures(none) %6) local_unnamed_addr #0 !dbg !5 { + %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %9 = zext nneg i32 %8 to i64, !dbg !14 + %10 = shl nuw nsw i64 %9, 5, !dbg !15 + %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !16 + %12 = and i32 %11, 112, !dbg !16 + %13 = lshr exact i32 %12, 4, !dbg !16 + %14 = or disjoint i32 %13, 8, !dbg !16 + %15 = or disjoint i32 %13, 16, !dbg !16 + %16 = or disjoint i32 %13, 24, !dbg !16 + %17 = zext nneg i32 %13 to i64, !dbg !17 + %18 = zext nneg i32 %14 to i64, !dbg !17 + %19 = zext nneg i32 %15 to i64, !dbg !17 + %20 = zext nneg i32 %16 to i64, !dbg !17 + %21 = or disjoint i64 %10, %17, !dbg !17 + %22 = or disjoint i64 %10, %18, !dbg !17 + %23 = or disjoint i64 %10, %19, !dbg !17 + %24 = or disjoint i64 %10, %20, !dbg !17 + %25 = sext i32 %4 to i64, !dbg !18 + %26 = icmp slt i64 %21, %25, !dbg !18 + %27 = icmp slt i64 %22, %25, !dbg !18 + %28 = icmp slt i64 %23, %25, !dbg !18 + %29 = icmp slt i64 %24, %25, !dbg !18 + %30 = shl nuw nsw i32 %11, 3, !dbg !19 + %31 = and i32 %30, 120, !dbg !19 + %32 = shl nuw nsw i64 %21, 7, !dbg !20 + %33 = shl nuw nsw i64 %22, 7, !dbg !20 + %34 = shl nuw nsw i64 %23, 7, !dbg !20 + %35 = shl nuw nsw i64 %24, 7, !dbg !20 + %36 = getelementptr bfloat, ptr addrspace(1) %0, i64 %32, !dbg !21 + %37 = getelementptr bfloat, ptr addrspace(1) %0, i64 %33, !dbg !21 + %38 = getelementptr bfloat, ptr addrspace(1) %0, i64 %34, !dbg !21 + %39 = getelementptr bfloat, ptr addrspace(1) %0, i64 %35, !dbg !21 + %40 = zext nneg i32 %31 to i64, !dbg !22 + %41 = getelementptr bfloat, ptr addrspace(1) %36, i64 %40, !dbg !22 + %42 = getelementptr bfloat, ptr addrspace(1) %37, i64 %40, !dbg !22 + %43 = getelementptr bfloat, ptr addrspace(1) %38, i64 %40, !dbg !22 + %44 = getelementptr bfloat, ptr addrspace(1) %39, i64 %40, !dbg !22 + %45 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %41, i1 %26) #6, !dbg !23 + %46 = extractvalue { i32, i32, i32, i32 } %45, 0, !dbg !23 + %47 = bitcast i32 %46 to <2 x bfloat>, !dbg !23 + %48 = extractvalue { i32, i32, i32, i32 } %45, 1, !dbg !23 + %49 = bitcast i32 %48 to <2 x bfloat>, !dbg !23 + %50 = extractvalue { i32, i32, i32, i32 } %45, 2, !dbg !23 + %51 = bitcast i32 %50 to <2 x bfloat>, !dbg !23 + %52 = extractvalue { i32, i32, i32, i32 } %45, 3, !dbg !23 + %53 = bitcast i32 %52 to <2 x bfloat>, !dbg !23 + %54 = extractelement <2 x bfloat> %47, i64 0, !dbg !23 + %55 = extractelement <2 x bfloat> %47, i64 1, !dbg !23 + %56 = extractelement <2 x bfloat> %49, i64 0, !dbg !23 + %57 = extractelement <2 x bfloat> %49, i64 1, !dbg !23 + %58 = extractelement <2 x bfloat> %51, i64 0, !dbg !23 + %59 = extractelement <2 x bfloat> %51, i64 1, !dbg !23 + %60 = extractelement <2 x bfloat> %53, i64 0, !dbg !23 + %61 = extractelement <2 x bfloat> %53, i64 1, !dbg !23 + %62 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %42, i1 %27) #6, !dbg !23 + %63 = extractvalue { i32, i32, i32, i32 } %62, 0, !dbg !23 + %64 = bitcast i32 %63 to <2 x bfloat>, !dbg !23 + %65 = extractvalue { i32, i32, i32, i32 } %62, 1, !dbg !23 + %66 = bitcast i32 %65 to <2 x bfloat>, !dbg !23 + %67 = extractvalue { i32, i32, i32, i32 } %62, 2, !dbg !23 + %68 = bitcast i32 %67 to <2 x bfloat>, !dbg !23 + %69 = extractvalue { i32, i32, i32, i32 } %62, 3, !dbg !23 + %70 = bitcast i32 %69 to <2 x bfloat>, !dbg !23 + %71 = extractelement <2 x bfloat> %64, i64 0, !dbg !23 + %72 = extractelement <2 x bfloat> %64, i64 1, !dbg !23 + %73 = extractelement <2 x bfloat> %66, i64 0, !dbg !23 + %74 = extractelement <2 x bfloat> %66, i64 1, !dbg !23 + %75 = extractelement <2 x bfloat> %68, i64 0, !dbg !23 + %76 = extractelement <2 x bfloat> %68, i64 1, !dbg !23 + %77 = extractelement <2 x bfloat> %70, i64 0, !dbg !23 + %78 = extractelement <2 x bfloat> %70, i64 1, !dbg !23 + %79 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %43, i1 %28) #6, !dbg !23 + %80 = extractvalue { i32, i32, i32, i32 } %79, 0, !dbg !23 + %81 = bitcast i32 %80 to <2 x bfloat>, !dbg !23 + %82 = extractvalue { i32, i32, i32, i32 } %79, 1, !dbg !23 + %83 = bitcast i32 %82 to <2 x bfloat>, !dbg !23 + %84 = extractvalue { i32, i32, i32, i32 } %79, 2, !dbg !23 + %85 = bitcast i32 %84 to <2 x bfloat>, !dbg !23 + %86 = extractvalue { i32, i32, i32, i32 } %79, 3, !dbg !23 + %87 = bitcast i32 %86 to <2 x bfloat>, !dbg !23 + %88 = extractelement <2 x bfloat> %81, i64 0, !dbg !23 + %89 = extractelement <2 x bfloat> %81, i64 1, !dbg !23 + %90 = extractelement <2 x bfloat> %83, i64 0, !dbg !23 + %91 = extractelement <2 x bfloat> %83, i64 1, !dbg !23 + %92 = extractelement <2 x bfloat> %85, i64 0, !dbg !23 + %93 = extractelement <2 x bfloat> %85, i64 1, !dbg !23 + %94 = extractelement <2 x bfloat> %87, i64 0, !dbg !23 + %95 = extractelement <2 x bfloat> %87, i64 1, !dbg !23 + %96 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %44, i1 %29) #6, !dbg !23 + %97 = extractvalue { i32, i32, i32, i32 } %96, 0, !dbg !23 + %98 = bitcast i32 %97 to <2 x bfloat>, !dbg !23 + %99 = extractvalue { i32, i32, i32, i32 } %96, 1, !dbg !23 + %100 = bitcast i32 %99 to <2 x bfloat>, !dbg !23 + %101 = extractvalue { i32, i32, i32, i32 } %96, 2, !dbg !23 + %102 = bitcast i32 %101 to <2 x bfloat>, !dbg !23 + %103 = extractvalue { i32, i32, i32, i32 } %96, 3, !dbg !23 + %104 = bitcast i32 %103 to <2 x bfloat>, !dbg !23 + %105 = extractelement <2 x bfloat> %98, i64 0, !dbg !23 + %106 = extractelement <2 x bfloat> %98, i64 1, !dbg !23 + %107 = extractelement <2 x bfloat> %100, i64 0, !dbg !23 + %108 = extractelement <2 x bfloat> %100, i64 1, !dbg !23 + %109 = extractelement <2 x bfloat> %102, i64 0, !dbg !23 + %110 = extractelement <2 x bfloat> %102, i64 1, !dbg !23 + %111 = extractelement <2 x bfloat> %104, i64 0, !dbg !23 + %112 = extractelement <2 x bfloat> %104, i64 1, !dbg !23 + %113 = fpext bfloat %54 to float, !dbg !24 + %114 = fpext bfloat %55 to float, !dbg !24 + %115 = fpext bfloat %56 to float, !dbg !24 + %116 = fpext bfloat %57 to float, !dbg !24 + %117 = fpext bfloat %58 to float, !dbg !24 + %118 = fpext bfloat %59 to float, !dbg !24 + %119 = fpext bfloat %60 to float, !dbg !24 + %120 = fpext bfloat %61 to float, !dbg !24 + %121 = fpext bfloat %71 to float, !dbg !24 + %122 = fpext bfloat %72 to float, !dbg !24 + %123 = fpext bfloat %73 to float, !dbg !24 + %124 = fpext bfloat %74 to float, !dbg !24 + %125 = fpext bfloat %75 to float, !dbg !24 + %126 = fpext bfloat %76 to float, !dbg !24 + %127 = fpext bfloat %77 to float, !dbg !24 + %128 = fpext bfloat %78 to float, !dbg !24 + %129 = fpext bfloat %88 to float, !dbg !24 + %130 = fpext bfloat %89 to float, !dbg !24 + %131 = fpext bfloat %90 to float, !dbg !24 + %132 = fpext bfloat %91 to float, !dbg !24 + %133 = fpext bfloat %92 to float, !dbg !24 + %134 = fpext bfloat %93 to float, !dbg !24 + %135 = fpext bfloat %94 to float, !dbg !24 + %136 = fpext bfloat %95 to float, !dbg !24 + %137 = fpext bfloat %105 to float, !dbg !24 + %138 = fpext bfloat %106 to float, !dbg !24 + %139 = fpext bfloat %107 to float, !dbg !24 + %140 = fpext bfloat %108 to float, !dbg !24 + %141 = fpext bfloat %109 to float, !dbg !24 + %142 = fpext bfloat %110 to float, !dbg !24 + %143 = fpext bfloat %111 to float, !dbg !24 + %144 = fpext bfloat %112 to float, !dbg !24 + %145 = fmul float %113, %113, !dbg !25 + %146 = fmul float %114, %114, !dbg !25 + %147 = fmul float %115, %115, !dbg !25 + %148 = fmul float %116, %116, !dbg !25 + %149 = fmul float %117, %117, !dbg !25 + %150 = fmul float %118, %118, !dbg !25 + %151 = fmul float %119, %119, !dbg !25 + %152 = fmul float %120, %120, !dbg !25 + %153 = fmul float %121, %121, !dbg !25 + %154 = fmul float %122, %122, !dbg !25 + %155 = fmul float %123, %123, !dbg !25 + %156 = fmul float %124, %124, !dbg !25 + %157 = fmul float %125, %125, !dbg !25 + %158 = fmul float %126, %126, !dbg !25 + %159 = fmul float %127, %127, !dbg !25 + %160 = fmul float %128, %128, !dbg !25 + %161 = fmul float %129, %129, !dbg !25 + %162 = fmul float %130, %130, !dbg !25 + %163 = fmul float %131, %131, !dbg !25 + %164 = fmul float %132, %132, !dbg !25 + %165 = fmul float %133, %133, !dbg !25 + %166 = fmul float %134, %134, !dbg !25 + %167 = fmul float %135, %135, !dbg !25 + %168 = fmul float %136, %136, !dbg !25 + %169 = fmul float %137, %137, !dbg !25 + %170 = fmul float %138, %138, !dbg !25 + %171 = fmul float %139, %139, !dbg !25 + %172 = fmul float %140, %140, !dbg !25 + %173 = fmul float %141, %141, !dbg !25 + %174 = fmul float %142, %142, !dbg !25 + %175 = fmul float %143, %143, !dbg !25 + %176 = fmul float %144, %144, !dbg !25 + %177 = fadd float %145, %146, !dbg !26 + %178 = fadd float %147, %177, !dbg !26 + %179 = fadd float %148, %178, !dbg !26 + %180 = fadd float %149, %179, !dbg !26 + %181 = fadd float %150, %180, !dbg !26 + %182 = fadd float %151, %181, !dbg !26 + %183 = fadd float %152, %182, !dbg !26 + %184 = fadd float %153, %154, !dbg !26 + %185 = fadd float %155, %184, !dbg !26 + %186 = fadd float %156, %185, !dbg !26 + %187 = fadd float %157, %186, !dbg !26 + %188 = fadd float %158, %187, !dbg !26 + %189 = fadd float %159, %188, !dbg !26 + %190 = fadd float %160, %189, !dbg !26 + %191 = fadd float %161, %162, !dbg !26 + %192 = fadd float %163, %191, !dbg !26 + %193 = fadd float %164, %192, !dbg !26 + %194 = fadd float %165, %193, !dbg !26 + %195 = fadd float %166, %194, !dbg !26 + %196 = fadd float %167, %195, !dbg !26 + %197 = fadd float %168, %196, !dbg !26 + %198 = fadd float %169, %170, !dbg !26 + %199 = fadd float %171, %198, !dbg !26 + %200 = fadd float %172, %199, !dbg !26 + %201 = fadd float %173, %200, !dbg !26 + %202 = fadd float %174, %201, !dbg !26 + %203 = fadd float %175, %202, !dbg !26 + %204 = fadd float %176, %203, !dbg !26 + %205 = bitcast float %183 to i32, !dbg !29 + %206 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %205, i32 8, i32 31), !dbg !29 + %207 = bitcast i32 %206 to float, !dbg !29 + %208 = fadd float %183, %207, !dbg !26 + %209 = bitcast float %208 to i32, !dbg !29 + %210 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %209, i32 4, i32 31), !dbg !29 + %211 = bitcast i32 %210 to float, !dbg !29 + %212 = fadd float %208, %211, !dbg !26 + %213 = bitcast float %212 to i32, !dbg !29 + %214 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %213, i32 2, i32 31), !dbg !29 + %215 = bitcast i32 %214 to float, !dbg !29 + %216 = fadd float %212, %215, !dbg !26 + %217 = bitcast float %216 to i32, !dbg !29 + %218 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %217, i32 1, i32 31), !dbg !29 + %219 = bitcast i32 %218 to float, !dbg !29 + %220 = fadd float %216, %219, !dbg !26 + %221 = bitcast float %190 to i32, !dbg !29 + %222 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %221, i32 8, i32 31), !dbg !29 + %223 = bitcast i32 %222 to float, !dbg !29 + %224 = fadd float %190, %223, !dbg !26 + %225 = bitcast float %224 to i32, !dbg !29 + %226 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %225, i32 4, i32 31), !dbg !29 + %227 = bitcast i32 %226 to float, !dbg !29 + %228 = fadd float %224, %227, !dbg !26 + %229 = bitcast float %228 to i32, !dbg !29 + %230 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %229, i32 2, i32 31), !dbg !29 + %231 = bitcast i32 %230 to float, !dbg !29 + %232 = fadd float %228, %231, !dbg !26 + %233 = bitcast float %232 to i32, !dbg !29 + %234 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %233, i32 1, i32 31), !dbg !29 + %235 = bitcast i32 %234 to float, !dbg !29 + %236 = fadd float %232, %235, !dbg !26 + %237 = bitcast float %197 to i32, !dbg !29 + %238 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %237, i32 8, i32 31), !dbg !29 + %239 = bitcast i32 %238 to float, !dbg !29 + %240 = fadd float %197, %239, !dbg !26 + %241 = bitcast float %240 to i32, !dbg !29 + %242 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %241, i32 4, i32 31), !dbg !29 + %243 = bitcast i32 %242 to float, !dbg !29 + %244 = fadd float %240, %243, !dbg !26 + %245 = bitcast float %244 to i32, !dbg !29 + %246 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %245, i32 2, i32 31), !dbg !29 + %247 = bitcast i32 %246 to float, !dbg !29 + %248 = fadd float %244, %247, !dbg !26 + %249 = bitcast float %248 to i32, !dbg !29 + %250 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %249, i32 1, i32 31), !dbg !29 + %251 = bitcast i32 %250 to float, !dbg !29 + %252 = fadd float %248, %251, !dbg !26 + %253 = bitcast float %204 to i32, !dbg !29 + %254 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %253, i32 8, i32 31), !dbg !29 + %255 = bitcast i32 %254 to float, !dbg !29 + %256 = fadd float %204, %255, !dbg !26 + %257 = bitcast float %256 to i32, !dbg !29 + %258 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %257, i32 4, i32 31), !dbg !29 + %259 = bitcast i32 %258 to float, !dbg !29 + %260 = fadd float %256, %259, !dbg !26 + %261 = bitcast float %260 to i32, !dbg !29 + %262 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %261, i32 2, i32 31), !dbg !29 + %263 = bitcast i32 %262 to float, !dbg !29 + %264 = fadd float %260, %263, !dbg !26 + %265 = bitcast float %264 to i32, !dbg !29 + %266 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %265, i32 1, i32 31), !dbg !29 + %267 = bitcast i32 %266 to float, !dbg !29 + %268 = fadd float %264, %267, !dbg !26 + %269 = fadd float %3, %220, !dbg !32 + %270 = fadd float %3, %236, !dbg !32 + %271 = fadd float %3, %252, !dbg !32 + %272 = fadd float %3, %268, !dbg !32 + %273 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i = icmp eq i32 %273, 0, !dbg !33 + %274 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i = icmp eq i32 %274, 0, !dbg !33 + br i1 %.not.i, label %280, label %275, !dbg !33 + +275: ; preds = %7 + br i1 %.not1.i, label %278, label %276, !dbg !33 + +276: ; preds = %275 + %277 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %269) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +278: ; preds = %275 + %279 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %269) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +280: ; preds = %7 + br i1 %.not1.i, label %283, label %281, !dbg !33 + +281: ; preds = %280 + %282 = tail call float @llvm.nvvm.sqrt.rn.f(float %269) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +283: ; preds = %280 + %284 = tail call float @llvm.nvvm.sqrt.approx.f(float %269) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +__nv_sqrtf.exit: ; preds = %276, %278, %281, %283 + %.0.i = phi float [ %277, %276 ], [ %279, %278 ], [ %282, %281 ], [ %284, %283 ], !dbg !33 + %285 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i1 = icmp eq i32 %285, 0, !dbg !33 + %286 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i4 = icmp eq i32 %286, 0, !dbg !33 + br i1 %.not.i1, label %292, label %287, !dbg !33 + +287: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %290, label %288, !dbg !33 + +288: ; preds = %287 + %289 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %270) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +290: ; preds = %287 + %291 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %270) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +292: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %295, label %293, !dbg !33 + +293: ; preds = %292 + %294 = tail call float @llvm.nvvm.sqrt.rn.f(float %270) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +295: ; preds = %292 + %296 = tail call float @llvm.nvvm.sqrt.approx.f(float %270) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +__nv_sqrtf.exit5: ; preds = %288, %290, %293, %295 + %.0.i3 = phi float [ %289, %288 ], [ %291, %290 ], [ %294, %293 ], [ %296, %295 ], !dbg !33 + %297 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i6 = icmp eq i32 %297, 0, !dbg !33 + %298 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i9 = icmp eq i32 %298, 0, !dbg !33 + br i1 %.not.i6, label %304, label %299, !dbg !33 + +299: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %302, label %300, !dbg !33 + +300: ; preds = %299 + %301 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %271) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +302: ; preds = %299 + %303 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %271) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +304: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %307, label %305, !dbg !33 + +305: ; preds = %304 + %306 = tail call float @llvm.nvvm.sqrt.rn.f(float %271) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +307: ; preds = %304 + %308 = tail call float @llvm.nvvm.sqrt.approx.f(float %271) #6, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +__nv_sqrtf.exit10: ; preds = %300, %302, %305, %307 + %.0.i8 = phi float [ %301, %300 ], [ %303, %302 ], [ %306, %305 ], [ %308, %307 ], !dbg !33 + %309 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i11 = icmp eq i32 %309, 0, !dbg !33 + %310 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i14 = icmp eq i32 %310, 0, !dbg !33 + br i1 %.not.i11, label %316, label %311, !dbg !33 + +311: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %314, label %312, !dbg !33 + +312: ; preds = %311 + %313 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %272) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +314: ; preds = %311 + %315 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %272) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +316: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %319, label %317, !dbg !33 + +317: ; preds = %316 + %318 = tail call float @llvm.nvvm.sqrt.rn.f(float %272) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +319: ; preds = %316 + %320 = tail call float @llvm.nvvm.sqrt.approx.f(float %272) #6, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +__nv_sqrtf.exit15: ; preds = %312, %314, %317, %319 + %.0.i13 = phi float [ %313, %312 ], [ %315, %314 ], [ %318, %317 ], [ %320, %319 ], !dbg !33 + %321 = and i32 %11, 31, !dbg !16 + %322 = zext nneg i32 %321 to i64, !dbg !17 + %323 = or disjoint i64 %10, %322, !dbg !17 + %324 = getelementptr float, ptr addrspace(1) %2, i64 %323, !dbg !34 + %325 = getelementptr bfloat, ptr addrspace(1) %1, i64 %35, !dbg !35 + %326 = getelementptr bfloat, ptr addrspace(1) %325, i64 %40, !dbg !36 + %327 = getelementptr bfloat, ptr addrspace(1) %1, i64 %34, !dbg !35 + %328 = getelementptr bfloat, ptr addrspace(1) %327, i64 %40, !dbg !36 + %329 = getelementptr bfloat, ptr addrspace(1) %1, i64 %33, !dbg !35 + %330 = getelementptr bfloat, ptr addrspace(1) %329, i64 %40, !dbg !36 + %331 = getelementptr bfloat, ptr addrspace(1) %1, i64 %32, !dbg !35 + %332 = getelementptr bfloat, ptr addrspace(1) %331, i64 %40, !dbg !36 + %333 = icmp slt i64 %323, %25, !dbg !18 + %334 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i), !dbg !37 + %335 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i3), !dbg !37 + %336 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i8), !dbg !37 + %337 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i13), !dbg !37 + %338 = fmul float %334, %113, !dbg !38 + %339 = fmul float %334, %114, !dbg !38 + %340 = fmul float %334, %115, !dbg !38 + %341 = fmul float %334, %116, !dbg !38 + %342 = fmul float %334, %117, !dbg !38 + %343 = fmul float %334, %118, !dbg !38 + %344 = fmul float %334, %119, !dbg !38 + %345 = fmul float %334, %120, !dbg !38 + %346 = fmul float %335, %121, !dbg !38 + %347 = fmul float %335, %122, !dbg !38 + %348 = fmul float %335, %123, !dbg !38 + %349 = fmul float %335, %124, !dbg !38 + %350 = fmul float %335, %125, !dbg !38 + %351 = fmul float %335, %126, !dbg !38 + %352 = fmul float %335, %127, !dbg !38 + %353 = fmul float %335, %128, !dbg !38 + %354 = fmul float %336, %129, !dbg !38 + %355 = fmul float %336, %130, !dbg !38 + %356 = fmul float %336, %131, !dbg !38 + %357 = fmul float %336, %132, !dbg !38 + %358 = fmul float %336, %133, !dbg !38 + %359 = fmul float %336, %134, !dbg !38 + %360 = fmul float %336, %135, !dbg !38 + %361 = fmul float %336, %136, !dbg !38 + %362 = fmul float %337, %137, !dbg !38 + %363 = fmul float %337, %138, !dbg !38 + %364 = fmul float %337, %139, !dbg !38 + %365 = fmul float %337, %140, !dbg !38 + %366 = fmul float %337, %141, !dbg !38 + %367 = fmul float %337, %142, !dbg !38 + %368 = fmul float %337, %143, !dbg !38 + %369 = fmul float %337, %144, !dbg !38 + %370 = fptrunc float %338 to bfloat, !dbg !39 + %371 = fptrunc float %339 to bfloat, !dbg !39 + %372 = fptrunc float %340 to bfloat, !dbg !39 + %373 = fptrunc float %341 to bfloat, !dbg !39 + %374 = fptrunc float %342 to bfloat, !dbg !39 + %375 = fptrunc float %343 to bfloat, !dbg !39 + %376 = fptrunc float %344 to bfloat, !dbg !39 + %377 = fptrunc float %345 to bfloat, !dbg !39 + %378 = fptrunc float %346 to bfloat, !dbg !39 + %379 = fptrunc float %347 to bfloat, !dbg !39 + %380 = fptrunc float %348 to bfloat, !dbg !39 + %381 = fptrunc float %349 to bfloat, !dbg !39 + %382 = fptrunc float %350 to bfloat, !dbg !39 + %383 = fptrunc float %351 to bfloat, !dbg !39 + %384 = fptrunc float %352 to bfloat, !dbg !39 + %385 = fptrunc float %353 to bfloat, !dbg !39 + %386 = fptrunc float %354 to bfloat, !dbg !39 + %387 = fptrunc float %355 to bfloat, !dbg !39 + %388 = fptrunc float %356 to bfloat, !dbg !39 + %389 = fptrunc float %357 to bfloat, !dbg !39 + %390 = fptrunc float %358 to bfloat, !dbg !39 + %391 = fptrunc float %359 to bfloat, !dbg !39 + %392 = fptrunc float %360 to bfloat, !dbg !39 + %393 = fptrunc float %361 to bfloat, !dbg !39 + %394 = fptrunc float %362 to bfloat, !dbg !39 + %395 = fptrunc float %363 to bfloat, !dbg !39 + %396 = fptrunc float %364 to bfloat, !dbg !39 + %397 = fptrunc float %365 to bfloat, !dbg !39 + %398 = fptrunc float %366 to bfloat, !dbg !39 + %399 = fptrunc float %367 to bfloat, !dbg !39 + %400 = fptrunc float %368 to bfloat, !dbg !39 + %401 = fptrunc float %369 to bfloat, !dbg !39 + %402 = insertelement <2 x bfloat> poison, bfloat %370, i64 0, !dbg !40 + %403 = insertelement <2 x bfloat> %402, bfloat %371, i64 1, !dbg !40 + %404 = bitcast <2 x bfloat> %403 to i32, !dbg !40 + %405 = insertelement <2 x bfloat> poison, bfloat %372, i64 0, !dbg !40 + %406 = insertelement <2 x bfloat> %405, bfloat %373, i64 1, !dbg !40 + %407 = bitcast <2 x bfloat> %406 to i32, !dbg !40 + %408 = insertelement <2 x bfloat> poison, bfloat %374, i64 0, !dbg !40 + %409 = insertelement <2 x bfloat> %408, bfloat %375, i64 1, !dbg !40 + %410 = bitcast <2 x bfloat> %409 to i32, !dbg !40 + %411 = insertelement <2 x bfloat> poison, bfloat %376, i64 0, !dbg !40 + %412 = insertelement <2 x bfloat> %411, bfloat %377, i64 1, !dbg !40 + %413 = bitcast <2 x bfloat> %412 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %404, i32 %407, i32 %410, i32 %413, ptr addrspace(1) %332, i1 %26) #6, !dbg !40 + %414 = insertelement <2 x bfloat> poison, bfloat %378, i64 0, !dbg !40 + %415 = insertelement <2 x bfloat> %414, bfloat %379, i64 1, !dbg !40 + %416 = bitcast <2 x bfloat> %415 to i32, !dbg !40 + %417 = insertelement <2 x bfloat> poison, bfloat %380, i64 0, !dbg !40 + %418 = insertelement <2 x bfloat> %417, bfloat %381, i64 1, !dbg !40 + %419 = bitcast <2 x bfloat> %418 to i32, !dbg !40 + %420 = insertelement <2 x bfloat> poison, bfloat %382, i64 0, !dbg !40 + %421 = insertelement <2 x bfloat> %420, bfloat %383, i64 1, !dbg !40 + %422 = bitcast <2 x bfloat> %421 to i32, !dbg !40 + %423 = insertelement <2 x bfloat> poison, bfloat %384, i64 0, !dbg !40 + %424 = insertelement <2 x bfloat> %423, bfloat %385, i64 1, !dbg !40 + %425 = bitcast <2 x bfloat> %424 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %416, i32 %419, i32 %422, i32 %425, ptr addrspace(1) %330, i1 %27) #6, !dbg !40 + %426 = insertelement <2 x bfloat> poison, bfloat %386, i64 0, !dbg !40 + %427 = insertelement <2 x bfloat> %426, bfloat %387, i64 1, !dbg !40 + %428 = bitcast <2 x bfloat> %427 to i32, !dbg !40 + %429 = insertelement <2 x bfloat> poison, bfloat %388, i64 0, !dbg !40 + %430 = insertelement <2 x bfloat> %429, bfloat %389, i64 1, !dbg !40 + %431 = bitcast <2 x bfloat> %430 to i32, !dbg !40 + %432 = insertelement <2 x bfloat> poison, bfloat %390, i64 0, !dbg !40 + %433 = insertelement <2 x bfloat> %432, bfloat %391, i64 1, !dbg !40 + %434 = bitcast <2 x bfloat> %433 to i32, !dbg !40 + %435 = insertelement <2 x bfloat> poison, bfloat %392, i64 0, !dbg !40 + %436 = insertelement <2 x bfloat> %435, bfloat %393, i64 1, !dbg !40 + %437 = bitcast <2 x bfloat> %436 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %428, i32 %431, i32 %434, i32 %437, ptr addrspace(1) %328, i1 %28) #6, !dbg !40 + %438 = insertelement <2 x bfloat> poison, bfloat %394, i64 0, !dbg !40 + %439 = insertelement <2 x bfloat> %438, bfloat %395, i64 1, !dbg !40 + %440 = bitcast <2 x bfloat> %439 to i32, !dbg !40 + %441 = insertelement <2 x bfloat> poison, bfloat %396, i64 0, !dbg !40 + %442 = insertelement <2 x bfloat> %441, bfloat %397, i64 1, !dbg !40 + %443 = bitcast <2 x bfloat> %442 to i32, !dbg !40 + %444 = insertelement <2 x bfloat> poison, bfloat %398, i64 0, !dbg !40 + %445 = insertelement <2 x bfloat> %444, bfloat %399, i64 1, !dbg !40 + %446 = bitcast <2 x bfloat> %445 to i32, !dbg !40 + %447 = insertelement <2 x bfloat> poison, bfloat %400, i64 0, !dbg !40 + %448 = insertelement <2 x bfloat> %447, bfloat %401, i64 1, !dbg !40 + %449 = bitcast <2 x bfloat> %448 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %440, i32 %443, i32 %446, i32 %449, ptr addrspace(1) %326, i1 %29) #6, !dbg !40 + %450 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %12, !dbg !41 + %451 = bitcast float %334 to i32, !dbg !41 + %452 = bitcast float %335 to i32, !dbg !41 + %453 = bitcast float %336 to i32, !dbg !41 + %454 = bitcast float %337 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %450, i32 %451, i32 %452, i32 %453, i32 %454) #6, !dbg !41 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !41 + %455 = shl nuw nsw i32 %11, 4, !dbg !41 + %456 = and i32 %455, 112, !dbg !41 + %457 = lshr i32 %11, 1, !dbg !41 + %458 = and i32 %457, 12, !dbg !41 + %459 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %456, !dbg !41 + %460 = getelementptr inbounds nuw i8, ptr addrspace(3) %459, i32 %458, !dbg !41 + %461 = load i32, ptr addrspace(3) %460, align 4, !dbg !41 + %462 = and i32 %11, 96, !dbg !41 + %463 = icmp eq i32 %462, 0, !dbg !41 + %464 = and i1 %463, %333, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %461, ptr addrspace(1) %324, i1 %464) #6, !dbg !41 + ret void, !dbg !42 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.div.full(float, float) #3 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #4 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #5 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.f(float) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="128" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) } +attributes #4 = { convergent nocallback nounwind } +attributes #5 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #6 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "l2norm_fwd_kernel", linkageName: "l2norm_fwd_kernel", scope: !1, file: !1, line: 69, type: !6, scopeLine: 69, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !10, !11, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 70, column: 24, scope: !5) +!14 = !DILocation(line: 70, column: 30, scope: !5) +!15 = !DILocation(line: 71, column: 16, scope: !5) +!16 = !DILocation(line: 71, column: 34, scope: !5) +!17 = !DILocation(line: 71, column: 21, scope: !5) +!18 = !DILocation(line: 73, column: 16, scope: !5) +!19 = !DILocation(line: 74, column: 30, scope: !5) +!20 = !DILocation(line: 75, column: 29, scope: !5) +!21 = !DILocation(line: 75, column: 14, scope: !5) +!22 = !DILocation(line: 75, column: 33, scope: !5) +!23 = !DILocation(line: 79, column: 18, scope: !5) +!24 = !DILocation(line: 79, column: 47, scope: !5) +!25 = !DILocation(line: 80, column: 38, scope: !5) +!26 = !DILocation(line: 263, column: 15, scope: !27, inlinedAt: !29) +!27 = distinct !DILexicalBlockFile(scope: !5, file: !28, discriminator: 0) +!28 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!29 = !DILocation(line: 293, column: 36, scope: !27, inlinedAt: !30) +!30 = !DILocation(line: 80, column: 43, scope: !31) +!31 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!32 = !DILocation(line: 80, column: 48, scope: !5) +!33 = !DILocation(line: 80, column: 25, scope: !5) +!34 = !DILocation(line: 77, column: 20, scope: !5) +!35 = !DILocation(line: 76, column: 14, scope: !5) +!36 = !DILocation(line: 76, column: 33, scope: !5) +!37 = !DILocation(line: 80, column: 17, scope: !5) +!38 = !DILocation(line: 81, column: 16, scope: !5) +!39 = !DILocation(line: 83, column: 25, scope: !5) +!40 = !DILocation(line: 83, column: 18, scope: !5) +!41 = !DILocation(line: 84, column: 21, scope: !5) +!42 = !DILocation(line: 84, column: 4, scope: !5) diff --git a/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/l2norm_fwd_kernel.ptx b/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/l2norm_fwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..1698399b7f34e882d9efc122449b4c78b312ebcc --- /dev/null +++ b/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/l2norm_fwd_kernel.ptx @@ -0,0 +1,642 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_fwd_kernel // -- Begin function l2norm_fwd_kernel +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; +.global .align 1 .b8 _$_str_$_2[17] = {95, 95, 67, 85, 68, 65, 95, 80, 82, 69, 67, 95, 83, 81, 82, 84}; + // @l2norm_fwd_kernel +.visible .entry l2norm_fwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_2, + .param .f32 l2norm_fwd_kernel_param_3, + .param .u32 l2norm_fwd_kernel_param_4, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_6 +) +.reqntid 128 +{ + .reg .pred %p<8>; + .reg .b16 %rs<33>; + .reg .b32 %r<196>; + .reg .b64 %rd<39>; + .loc 1 69 0 // l2norm.py:69:0 +$L__func_begin0: + .loc 1 69 0 // l2norm.py:69:0 + +// %bb.0: // %__nv_sqrtf.exit + ld.param.b64 %rd10, [l2norm_fwd_kernel_param_0]; + ld.param.b64 %rd11, [l2norm_fwd_kernel_param_1]; +$L__tmp0: + .loc 1 70 24 // l2norm.py:70:24 + mov.u32 %r40, %ctaid.x; + ld.param.b64 %rd12, [l2norm_fwd_kernel_param_2]; + .loc 1 71 16 // l2norm.py:71:16 + mul.wide.u32 %rd13, %r40, 32; + ld.param.b32 %r41, [l2norm_fwd_kernel_param_3]; + ld.param.s32 %rd14, [l2norm_fwd_kernel_param_4]; + .loc 1 71 34 // l2norm.py:71:34 + mov.u32 %r42, %tid.x; + and.b32 %r43, %r42, 112; + bfe.u32 %r44, %r42, 4, 3; + or.b32 %r45, %r44, 8; + or.b32 %r46, %r44, 16; + or.b32 %r47, %r44, 24; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd15, %r44; + cvt.u64.u32 %rd16, %r45; + cvt.u64.u32 %rd17, %r46; + cvt.u64.u32 %rd18, %r47; + or.b64 %rd19, %rd13, %rd15; + or.b64 %rd20, %rd13, %rd16; + or.b64 %rd21, %rd13, %rd17; + or.b64 %rd22, %rd13, %rd18; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p1, %rd19, %rd14; + setp.lt.s64 %p2, %rd20, %rd14; + setp.lt.s64 %p3, %rd21, %rd14; + setp.lt.s64 %p4, %rd22, %rd14; + .loc 1 74 30 // l2norm.py:74:30 + shl.b32 %r48, %r42, 3; + and.b32 %r49, %r48, 120; + .loc 1 75 14 // l2norm.py:75:14 + shl.b64 %rd23, %rd19, 8; + add.s64 %rd24, %rd10, %rd23; + shl.b64 %rd25, %rd20, 8; + add.s64 %rd26, %rd10, %rd25; + shl.b64 %rd27, %rd21, 8; + add.s64 %rd28, %rd10, %rd27; + shl.b64 %rd29, %rd22, 8; + add.s64 %rd30, %rd10, %rd29; + .loc 1 75 33 // l2norm.py:75:33 + mul.wide.u32 %rd31, %r49, 2; + add.s64 %rd1, %rd24, %rd31; + add.s64 %rd2, %rd26, %rd31; + add.s64 %rd3, %rd28, %rd31; + add.s64 %rd4, %rd30, %rd31; + mov.b32 %r5, 0; + .loc 1 79 18 // l2norm.py:79:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + mov.b32 {%rs5, %rs6}, %r3; + mov.b32 {%rs7, %rs8}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs9, %rs10}, %r6; + mov.b32 {%rs11, %rs12}, %r7; + mov.b32 {%rs13, %rs14}, %r8; + mov.b32 {%rs15, %rs16}, %r9; + // begin inline asm + mov.u32 %r10, %r5; + mov.u32 %r11, %r5; + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + @%p3 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs17, %rs18}, %r10; + mov.b32 {%rs19, %rs20}, %r11; + mov.b32 {%rs21, %rs22}, %r12; + mov.b32 {%rs23, %rs24}, %r13; + // begin inline asm + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + @%p4 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ]; + // end inline asm + mov.b32 {%rs25, %rs26}, %r14; + mov.b32 {%rs27, %rs28}, %r15; + mov.b32 {%rs29, %rs30}, %r16; + mov.b32 {%rs31, %rs32}, %r17; + .loc 1 79 47 // l2norm.py:79:47 + cvt.f32.bf16 %r50, %rs1; + cvt.f32.bf16 %r51, %rs2; + cvt.f32.bf16 %r52, %rs3; + cvt.f32.bf16 %r53, %rs4; + cvt.f32.bf16 %r54, %rs5; + cvt.f32.bf16 %r55, %rs6; + cvt.f32.bf16 %r56, %rs7; + cvt.f32.bf16 %r57, %rs8; + cvt.f32.bf16 %r58, %rs9; + cvt.f32.bf16 %r59, %rs10; + cvt.f32.bf16 %r60, %rs11; + cvt.f32.bf16 %r61, %rs12; + cvt.f32.bf16 %r62, %rs13; + cvt.f32.bf16 %r63, %rs14; + cvt.f32.bf16 %r64, %rs15; + cvt.f32.bf16 %r65, %rs16; + cvt.f32.bf16 %r66, %rs17; + cvt.f32.bf16 %r67, %rs18; + cvt.f32.bf16 %r68, %rs19; + cvt.f32.bf16 %r69, %rs20; + cvt.f32.bf16 %r70, %rs21; + cvt.f32.bf16 %r71, %rs22; + cvt.f32.bf16 %r72, %rs23; + cvt.f32.bf16 %r73, %rs24; + cvt.f32.bf16 %r74, %rs25; + cvt.f32.bf16 %r75, %rs26; + cvt.f32.bf16 %r76, %rs27; + cvt.f32.bf16 %r77, %rs28; + cvt.f32.bf16 %r78, %rs29; + cvt.f32.bf16 %r79, %rs30; + cvt.f32.bf16 %r80, %rs31; + cvt.f32.bf16 %r81, %rs32; + .loc 1 80 38 // l2norm.py:80:38 + mul.f32 %r82, %r51, %r51; + mul.f32 %r83, %r59, %r59; + mul.f32 %r84, %r67, %r67; + mul.f32 %r85, %r75, %r75; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + fma.rn.f32.bf16 %r86, %rs1, %rs1, %r82; + fma.rn.f32.bf16 %r87, %rs3, %rs3, %r86; + fma.rn.f32.bf16 %r88, %rs4, %rs4, %r87; + fma.rn.f32.bf16 %r89, %rs5, %rs5, %r88; + fma.rn.f32.bf16 %r90, %rs6, %rs6, %r89; + fma.rn.f32.bf16 %r91, %rs7, %rs7, %r90; + fma.rn.f32.bf16 %r92, %rs8, %rs8, %r91; + fma.rn.f32.bf16 %r93, %rs9, %rs9, %r83; + fma.rn.f32.bf16 %r94, %rs11, %rs11, %r93; + fma.rn.f32.bf16 %r95, %rs12, %rs12, %r94; + fma.rn.f32.bf16 %r96, %rs13, %rs13, %r95; + fma.rn.f32.bf16 %r97, %rs14, %rs14, %r96; + fma.rn.f32.bf16 %r98, %rs15, %rs15, %r97; + fma.rn.f32.bf16 %r99, %rs16, %rs16, %r98; + fma.rn.f32.bf16 %r100, %rs17, %rs17, %r84; + fma.rn.f32.bf16 %r101, %rs19, %rs19, %r100; + fma.rn.f32.bf16 %r102, %rs20, %rs20, %r101; + fma.rn.f32.bf16 %r103, %rs21, %rs21, %r102; + fma.rn.f32.bf16 %r104, %rs22, %rs22, %r103; + fma.rn.f32.bf16 %r105, %rs23, %rs23, %r104; + fma.rn.f32.bf16 %r106, %rs24, %rs24, %r105; + fma.rn.f32.bf16 %r107, %rs25, %rs25, %r85; + fma.rn.f32.bf16 %r108, %rs27, %rs27, %r107; + fma.rn.f32.bf16 %r109, %rs28, %rs28, %r108; + fma.rn.f32.bf16 %r110, %rs29, %rs29, %r109; + fma.rn.f32.bf16 %r111, %rs30, %rs30, %r110; + fma.rn.f32.bf16 %r112, %rs31, %rs31, %r111; + fma.rn.f32.bf16 %r113, %rs32, %rs32, %r112; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r114, %r92, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r115, %r92, %r114; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r116, %r115, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r117, %r115, %r116; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r118, %r117, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r119, %r117, %r118; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r120, %r119, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r121, %r119, %r120; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r122, %r99, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r123, %r99, %r122; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r124, %r123, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r125, %r123, %r124; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r126, %r125, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r127, %r125, %r126; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r128, %r127, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r129, %r127, %r128; +$L__tmp18: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r130, %r106, 8, 31, -1; +$L__tmp19: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r131, %r106, %r130; +$L__tmp20: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r132, %r131, 4, 31, -1; +$L__tmp21: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r133, %r131, %r132; +$L__tmp22: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r134, %r133, 2, 31, -1; +$L__tmp23: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r135, %r133, %r134; +$L__tmp24: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r136, %r135, 1, 31, -1; +$L__tmp25: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r137, %r135, %r136; +$L__tmp26: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r138, %r113, 8, 31, -1; +$L__tmp27: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r139, %r113, %r138; +$L__tmp28: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r140, %r139, 4, 31, -1; +$L__tmp29: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r141, %r139, %r140; +$L__tmp30: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r142, %r141, 2, 31, -1; +$L__tmp31: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r143, %r141, %r142; +$L__tmp32: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r144, %r143, 1, 31, -1; +$L__tmp33: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r145, %r143, %r144; +$L__tmp34: + .loc 1 80 48 // l2norm.py:80:48 + add.f32 %r146, %r41, %r121; + add.f32 %r147, %r41, %r129; + add.f32 %r148, %r41, %r137; + add.f32 %r149, %r41, %r145; + .loc 1 80 25 // l2norm.py:80:25 + sqrt.approx.ftz.f32 %r150, %r146; + sqrt.approx.ftz.f32 %r151, %r147; + sqrt.approx.ftz.f32 %r152, %r148; + sqrt.approx.ftz.f32 %r153, %r149; + .loc 1 71 34 // l2norm.py:71:34 + and.b32 %r154, %r42, 31; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd32, %r154; + or.b64 %rd33, %rd13, %rd32; + .loc 1 77 20 // l2norm.py:77:20 + shl.b64 %rd34, %rd33, 2; + add.s64 %rd9, %rd12, %rd34; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd35, %rd11, %rd29; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd8, %rd35, %rd31; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd36, %rd11, %rd27; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd7, %rd36, %rd31; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd37, %rd11, %rd25; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd6, %rd37, %rd31; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd38, %rd11, %rd23; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd5, %rd38, %rd31; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p6, %rd33, %rd14; + mov.b32 %r155, 0f3F800000; + .loc 1 80 17 // l2norm.py:80:17 + div.full.f32 %r35, %r155, %r150; + div.full.f32 %r36, %r155, %r151; + div.full.f32 %r37, %r155, %r152; + div.full.f32 %r38, %r155, %r153; + .loc 1 81 16 // l2norm.py:81:16 + mul.f32 %r156, %r35, %r50; + mul.f32 %r157, %r35, %r51; + mul.f32 %r158, %r35, %r52; + mul.f32 %r159, %r35, %r53; + mul.f32 %r160, %r35, %r54; + mul.f32 %r161, %r35, %r55; + mul.f32 %r162, %r35, %r56; + mul.f32 %r163, %r35, %r57; + mul.f32 %r164, %r36, %r58; + mul.f32 %r165, %r36, %r59; + mul.f32 %r166, %r36, %r60; + mul.f32 %r167, %r36, %r61; + mul.f32 %r168, %r36, %r62; + mul.f32 %r169, %r36, %r63; + mul.f32 %r170, %r36, %r64; + mul.f32 %r171, %r36, %r65; + mul.f32 %r172, %r37, %r66; + mul.f32 %r173, %r37, %r67; + mul.f32 %r174, %r37, %r68; + mul.f32 %r175, %r37, %r69; + mul.f32 %r176, %r37, %r70; + mul.f32 %r177, %r37, %r71; + mul.f32 %r178, %r37, %r72; + mul.f32 %r179, %r37, %r73; + mul.f32 %r180, %r38, %r74; + mul.f32 %r181, %r38, %r75; + mul.f32 %r182, %r38, %r76; + mul.f32 %r183, %r38, %r77; + mul.f32 %r184, %r38, %r78; + mul.f32 %r185, %r38, %r79; + mul.f32 %r186, %r38, %r80; + mul.f32 %r187, %r38, %r81; + .loc 1 83 18 // l2norm.py:83:18 + cvt.rn.bf16x2.f32 %r18, %r157, %r156; + cvt.rn.bf16x2.f32 %r19, %r159, %r158; + cvt.rn.bf16x2.f32 %r20, %r161, %r160; + cvt.rn.bf16x2.f32 %r21, %r163, %r162; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd5 + 0 ], { %r18, %r19, %r20, %r21 }; + // end inline asm + cvt.rn.bf16x2.f32 %r22, %r165, %r164; + cvt.rn.bf16x2.f32 %r23, %r167, %r166; + cvt.rn.bf16x2.f32 %r24, %r169, %r168; + cvt.rn.bf16x2.f32 %r25, %r171, %r170; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd6 + 0 ], { %r22, %r23, %r24, %r25 }; + // end inline asm + cvt.rn.bf16x2.f32 %r26, %r173, %r172; + cvt.rn.bf16x2.f32 %r27, %r175, %r174; + cvt.rn.bf16x2.f32 %r28, %r177, %r176; + cvt.rn.bf16x2.f32 %r29, %r179, %r178; + // begin inline asm + @%p3 st.global.v4.b32 [ %rd7 + 0 ], { %r26, %r27, %r28, %r29 }; + // end inline asm + cvt.rn.bf16x2.f32 %r30, %r181, %r180; + cvt.rn.bf16x2.f32 %r31, %r183, %r182; + cvt.rn.bf16x2.f32 %r32, %r185, %r184; + cvt.rn.bf16x2.f32 %r33, %r187, %r186; + // begin inline asm + @%p4 st.global.v4.b32 [ %rd8 + 0 ], { %r30, %r31, %r32, %r33 }; + // end inline asm + .loc 1 84 21 // l2norm.py:84:21 + mov.b32 %r188, global_smem; + add.s32 %r34, %r188, %r43; + // begin inline asm + st.shared.v4.b32 [ %r34 + 0 ], { %r35, %r36, %r37, %r38 }; + // end inline asm + bar.sync 0; + shl.b32 %r189, %r42, 4; + and.b32 %r190, %r189, 112; + shr.u32 %r191, %r42, 1; + and.b32 %r192, %r191, 12; + add.s32 %r193, %r188, %r190; + add.s32 %r194, %r193, %r192; + ld.shared.b32 %r39, [%r194]; + and.b32 %r195, %r42, 96; + setp.eq.b32 %p7, %r195, 0; + and.pred %p5, %p7, %p6; + // begin inline asm + @%p5 st.global.b32 [ %rd9 + 0 ], { %r39 }; + // end inline asm + .loc 1 84 4 // l2norm.py:84:4 + ret; +$L__tmp35: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp34 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 80 // DW_AT_call_line +.b8 43 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp34 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/l2norm_fwd_kernel.source b/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/l2norm_fwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..d9fa83aac0600b1e5d35d95ed8aaad5c5636e577 --- /dev/null +++ b/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/l2norm_fwd_kernel.source @@ -0,0 +1,172 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc38 = loc(unknown) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc45 = loc("x"(#loc)) +#loc46 = loc("y"(#loc)) +#loc47 = loc("rstd"(#loc)) +#loc48 = loc("eps"(#loc)) +#loc49 = loc("T"(#loc)) +#loc81 = loc("input"(#loc36)) +#loc82 = loc("a"(#loc41)) +#loc83 = loc("b"(#loc41)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc50) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc51) + %o_t = arith.constant 32 : i32 loc(#loc52) + %o_t_1 = arith.constant 32 : i64 loc(#loc52) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc52) + %o_t_3 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32> loc(#loc53) + %o_t_4 = arith.extsi %o_t_3 : tensor<32xi32> to tensor<32xi64> loc(#loc54) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<32xi64> loc(#loc54) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<32xi64> loc(#loc54) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc55) + %m_t = arith.extsi %T : i32 to i64 loc(#loc56) + %m_t_7 = tt.splat %m_t : i64 -> tensor<32xi64> loc(#loc56) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<32xi64> loc(#loc56) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<32xi1> -> tensor<32x1xi1> loc(#loc57) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc58) + %m_x_10 = arith.constant 128 : i32 loc(#loc59) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc59) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc59) + %m_x_13 = tt.broadcast %m_x : tensor<32x1xi1> -> tensor<32x128xi1> loc(#loc60) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<32x128xi1> loc(#loc60) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<32x128xi1> loc(#loc60) + %p_x = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<32xi64> -> tensor<32x1xi64> loc(#loc61) + %p_x_16 = arith.constant 128 : i32 loc(#loc62) + %p_x_17 = arith.constant 128 : i64 loc(#loc62) + %p_x_18 = arith.constant dense<128> : tensor<32x1xi64> loc(#loc62) + %p_x_19 = arith.muli %p_x, %p_x_18 : tensor<32x1xi64> loc(#loc62) + %p_x_20 = tt.splat %x : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc63) + %p_x_21 = tt.addptr %p_x_20, %p_x_19 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc63) + %p_x_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc64) + %p_x_23 = tt.broadcast %p_x_21 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc65) + %p_x_24 = tt.broadcast %p_x_22 : tensor<1x128xi32> -> tensor<32x128xi32> loc(#loc65) + %p_x_25 = tt.addptr %p_x_23, %p_x_24 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc65) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<32xi64> -> tensor<32x1xi64> loc(#loc66) + %p_y_26 = arith.constant 128 : i32 loc(#loc67) + %p_y_27 = arith.constant 128 : i64 loc(#loc67) + %p_y_28 = arith.constant dense<128> : tensor<32x1xi64> loc(#loc67) + %p_y_29 = arith.muli %p_y, %p_y_28 : tensor<32x1xi64> loc(#loc67) + %p_y_30 = tt.splat %y : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc68) + %p_y_31 = tt.addptr %p_y_30, %p_y_29 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc68) + %p_y_32 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc69) + %p_y_33 = tt.broadcast %p_y_31 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc70) + %p_y_34 = tt.broadcast %p_y_32 : tensor<1x128xi32> -> tensor<32x128xi32> loc(#loc70) + %p_y_35 = tt.addptr %p_y_33, %p_y_34 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc70) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<32x!tt.ptr> loc(#loc71) + %p_rstd_36 = tt.addptr %p_rstd, %o_t_6 : tensor<32x!tt.ptr>, tensor<32xi64> loc(#loc71) + %b_x = arith.constant 0.000000e+00 : f32 loc(#loc72) + %b_x_37 = arith.constant dense<0.000000e+00> : tensor<32x128xf32> loc(#loc72) + %b_x_38 = arith.truncf %b_x_37 : tensor<32x128xf32> to tensor<32x128xbf16> loc(#loc72) + %b_x_39 = tt.load %p_x_25, %m_x_15, %b_x_38 : tensor<32x128x!tt.ptr> loc(#loc72) + %b_x_40 = arith.extf %b_x_39 : tensor<32x128xbf16> to tensor<32x128xf32> loc(#loc73) + %b_rstd = arith.mulf %b_x_40, %b_x_40 : tensor<32x128xf32> loc(#loc74) + %b_rstd_41 = tt.call @triton.language.standard.sum__fp32S32_128S_c1_cFalse_cNone(%b_rstd) : (tensor<32x128xf32>) -> tensor<32xf32> loc(#loc75) + %b_rstd_42 = tt.splat %eps : f32 -> tensor<32xf32> loc(#loc76) + %b_rstd_43 = arith.addf %b_rstd_41, %b_rstd_42 : tensor<32xf32> loc(#loc76) + %b_rstd_44 = math.sqrt %b_rstd_43 : tensor<32xf32> loc(#loc77) + %b_rstd_45 = arith.constant 1 : i32 loc(#loc78) + %b_rstd_46 = arith.constant 1.000000e+00 : f32 loc(#loc78) + %b_rstd_47 = arith.constant dense<1.000000e+00> : tensor<32xf32> loc(#loc78) + %b_rstd_48 = arith.divf %b_rstd_47, %b_rstd_44 : tensor<32xf32> loc(#loc78) + %b_y = tt.expand_dims %b_rstd_48 {axis = 1 : i32} : tensor<32xf32> -> tensor<32x1xf32> loc(#loc79) + %b_y_49 = tt.broadcast %b_y : tensor<32x1xf32> -> tensor<32x128xf32> loc(#loc80) + %b_y_50 = arith.mulf %b_x_40, %b_y_49 : tensor<32x128xf32> loc(#loc80) + %0 = arith.truncf %b_y_50 : tensor<32x128xf32> to tensor<32x128xbf16> loc(#loc32) + tt.store %p_y_35, %0, %m_x_15 : tensor<32x128x!tt.ptr> loc(#loc33) + tt.store %p_rstd_36, %b_rstd_48, %m_t_8 : tensor<32x!tt.ptr> loc(#loc34) + tt.return loc(#loc35) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S32_128S_c1_cFalse_cNone(%input: tensor<32x128xf32> loc("input"(#loc36))) -> tensor<32xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc37) + tt.reduce.return %2 : f32 loc(#loc37) + }) : (tensor<32x128xf32>) -> tensor<32xf32> loc(#loc37) + tt.return %0 : tensor<32xf32> loc(#loc39) + ^bb1: // no predecessors + %1 = ub.poison : tensor<32xf32> loc(#loc40) + tt.return %1 : tensor<32xf32> loc(#loc40) + } loc(#loc36) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc41)), %b: f32 loc("b"(#loc41))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc42) + tt.return %0 : f32 loc(#loc43) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc44) + tt.return %1 : f32 loc(#loc44) + } loc(#loc41) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:18) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:29) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:37) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc50 = loc("i_t"(#loc1)) +#loc51 = loc("i_t"(#loc2)) +#loc52 = loc("o_t"(#loc3)) +#loc53 = loc("o_t"(#loc4)) +#loc54 = loc("o_t"(#loc5)) +#loc55 = loc("o_d"(#loc6)) +#loc56 = loc("m_t"(#loc7)) +#loc57 = loc("m_x"(#loc8)) +#loc58 = loc("m_x"(#loc9)) +#loc59 = loc("m_x"(#loc10)) +#loc60 = loc("m_x"(#loc11)) +#loc61 = loc("p_x"(#loc12)) +#loc62 = loc("p_x"(#loc13)) +#loc63 = loc("p_x"(#loc14)) +#loc64 = loc("p_x"(#loc15)) +#loc65 = loc("p_x"(#loc16)) +#loc66 = loc("p_y"(#loc17)) +#loc67 = loc("p_y"(#loc18)) +#loc68 = loc("p_y"(#loc19)) +#loc69 = loc("p_y"(#loc20)) +#loc70 = loc("p_y"(#loc21)) +#loc71 = loc("p_rstd"(#loc22)) +#loc72 = loc("b_x"(#loc23)) +#loc73 = loc("b_x"(#loc24)) +#loc74 = loc("b_rstd"(#loc25)) +#loc75 = loc("b_rstd"(#loc26)) +#loc76 = loc("b_rstd"(#loc27)) +#loc77 = loc("b_rstd"(#loc28)) +#loc78 = loc("b_rstd"(#loc29)) +#loc79 = loc("b_y"(#loc30)) +#loc80 = loc("b_y"(#loc31)) diff --git a/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/l2norm_fwd_kernel.ttgir b/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/l2norm_fwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..dfcba2ba2295c1b1e3528c3ef5161ce0aff782f8 --- /dev/null +++ b/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/l2norm_fwd_kernel.ttgir @@ -0,0 +1,135 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [4, 1], order = [1, 0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [4], order = [0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc1 = loc(unknown) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc34 = loc("x"(#loc)) +#loc35 = loc("y"(#loc)) +#loc36 = loc("rstd"(#loc)) +#loc37 = loc("eps"(#loc)) +#loc38 = loc("T"(#loc)) +#loc59 = loc("b_rstd"(#loc23)) +#loc66 = loc(callsite(#loc1 at #loc59)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 4 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<1.000000e+00> : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<32x128xbf16, #blocked> loc(#loc1) + %c32_i64 = arith.constant 32 : i64 loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<32x1xi64, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc39) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc40) + %o_t = arith.muli %i_t_3, %c32_i64 : i64 loc(#loc41) + %o_t_4 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc42) + %o_t_5 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32, #blocked1> loc(#loc42) + %o_t_6 = arith.extsi %o_t_4 : tensor<32xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_7 = arith.extsi %o_t_5 : tensor<32xi32, #blocked1> to tensor<32xi64, #blocked1> loc(#loc43) + %o_t_8 = tt.splat %o_t : i64 -> tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_9 = tt.splat %o_t : i64 -> tensor<32xi64, #blocked1> loc(#loc43) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<32xi64, #blocked1> loc(#loc43) + %m_t = arith.extsi %T : i32 to i64 loc(#loc44) + %m_t_12 = tt.splat %m_t : i64 -> tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_13 = tt.splat %m_t : i64 -> tensor<32xi64, #blocked1> loc(#loc44) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<32xi64, #blocked1> loc(#loc44) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<32xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32x1xi1, #blocked> loc(#loc45) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc46) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc46) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_1 : tensor<1x128xi32, #blocked> loc(#loc47) + %m_x_19 = tt.broadcast %m_x : tensor<32x1xi1, #blocked> -> tensor<32x128xi1, #blocked> loc(#loc48) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<32x128xi1, #blocked> loc(#loc48) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<32x128xi1, #blocked> loc(#loc48) + %p_x = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32x1xi64, #blocked> loc(#loc49) + %p_x_22 = arith.muli %p_x, %cst_2 : tensor<32x1xi64, #blocked> loc(#loc50) + %p_x_23 = tt.splat %x : !tt.ptr -> tensor<32x1x!tt.ptr, #blocked> loc(#loc51) + %p_x_24 = tt.addptr %p_x_23, %p_x_22 : tensor<32x1x!tt.ptr, #blocked>, tensor<32x1xi64, #blocked> loc(#loc51) + %p_x_25 = tt.broadcast %p_x_24 : tensor<32x1x!tt.ptr, #blocked> -> tensor<32x128x!tt.ptr, #blocked> loc(#loc52) + %p_x_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<32x128xi32, #blocked> loc(#loc52) + %p_x_27 = tt.addptr %p_x_25, %p_x_26 : tensor<32x128x!tt.ptr, #blocked>, tensor<32x128xi32, #blocked> loc(#loc52) + %p_y = tt.splat %y : !tt.ptr -> tensor<32x1x!tt.ptr, #blocked> loc(#loc53) + %p_y_28 = tt.addptr %p_y, %p_x_22 : tensor<32x1x!tt.ptr, #blocked>, tensor<32x1xi64, #blocked> loc(#loc53) + %p_y_29 = tt.broadcast %p_y_28 : tensor<32x1x!tt.ptr, #blocked> -> tensor<32x128x!tt.ptr, #blocked> loc(#loc54) + %p_y_30 = tt.addptr %p_y_29, %p_x_26 : tensor<32x128x!tt.ptr, #blocked>, tensor<32x128xi32, #blocked> loc(#loc54) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<32x!tt.ptr, #blocked1> loc(#loc55) + %p_rstd_31 = tt.addptr %p_rstd, %o_t_11 : tensor<32x!tt.ptr, #blocked1>, tensor<32xi64, #blocked1> loc(#loc55) + %b_x = tt.load %p_x_27, %m_x_21, %cst_0 : tensor<32x128x!tt.ptr, #blocked> loc(#loc56) + %b_x_32 = arith.extf %b_x : tensor<32x128xbf16, #blocked> to tensor<32x128xf32, #blocked> loc(#loc57) + %b_rstd = arith.mulf %b_x_32, %b_x_32 : tensor<32x128xf32, #blocked> loc(#loc58) + %b_rstd_33 = "tt.reduce"(%b_rstd) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_40: f32 loc(callsite(#loc1 at #loc59)), %b_rstd_41: f32 loc(callsite(#loc1 at #loc59))): + %b_rstd_42 = arith.addf %b_rstd_40, %b_rstd_41 : f32 loc(#loc67) + tt.reduce.return %b_rstd_42 : f32 loc(#loc65) + }) : (tensor<32x128xf32, #blocked>) -> tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc65) + %b_rstd_34 = tt.splat %eps : f32 -> tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_36 = math.sqrt %b_rstd_35 : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc61) + %b_rstd_37 = arith.divf %cst, %b_rstd_36 : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc62) + %b_y = tt.expand_dims %b_rstd_37 {axis = 1 : i32} : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32x1xf32, #blocked> loc(#loc63) + %b_y_38 = tt.broadcast %b_y : tensor<32x1xf32, #blocked> -> tensor<32x128xf32, #blocked> loc(#loc64) + %b_y_39 = arith.mulf %b_x_32, %b_y_38 : tensor<32x128xf32, #blocked> loc(#loc64) + %0 = arith.truncf %b_y_39 : tensor<32x128xf32, #blocked> to tensor<32x128xbf16, #blocked> loc(#loc30) + tt.store %p_y_30, %0, %m_x_21 : tensor<32x128x!tt.ptr, #blocked> loc(#loc31) + %1 = ttg.convert_layout %b_rstd_37 : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32xf32, #blocked1> loc(#loc32) + tt.store %p_rstd_31, %1, %m_t_15 : tensor<32x!tt.ptr, #blocked1> loc(#loc32) + tt.return loc(#loc33) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc39 = loc("i_t"(#loc2)) +#loc40 = loc("i_t"(#loc3)) +#loc41 = loc("o_t"(#loc4)) +#loc42 = loc("o_t"(#loc5)) +#loc43 = loc("o_t"(#loc6)) +#loc44 = loc("m_t"(#loc7)) +#loc45 = loc("m_x"(#loc8)) +#loc46 = loc("m_x"(#loc9)) +#loc47 = loc("m_x"(#loc10)) +#loc48 = loc("m_x"(#loc11)) +#loc49 = loc("p_x"(#loc12)) +#loc50 = loc("p_x"(#loc13)) +#loc51 = loc("p_x"(#loc14)) +#loc52 = loc("p_x"(#loc15)) +#loc53 = loc("p_y"(#loc16)) +#loc54 = loc("p_y"(#loc17)) +#loc55 = loc("p_rstd"(#loc18)) +#loc56 = loc("b_x"(#loc19)) +#loc57 = loc("b_x"(#loc20)) +#loc58 = loc("b_rstd"(#loc21)) +#loc60 = loc("b_rstd"(#loc25)) +#loc61 = loc("b_rstd"(#loc26)) +#loc62 = loc("b_rstd"(#loc27)) +#loc63 = loc("b_y"(#loc28)) +#loc64 = loc("b_y"(#loc29)) +#loc65 = loc(callsite(#loc22 at #loc59)) +#loc67 = loc(callsite(#loc24 at #loc65)) diff --git a/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/l2norm_fwd_kernel.ttir b/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/l2norm_fwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..3688281bceaf7f8abdf8568104090be5e0170141 --- /dev/null +++ b/triton/gpu0/CKUA4OLQENZOXYTOYDMEVUJIROKBNGR2TOJQEFMNKJ5VXJ72WT2Q/l2norm_fwd_kernel.ttir @@ -0,0 +1,128 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc3 = loc(unknown) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc35 = loc("x"(#loc)) +#loc36 = loc("y"(#loc)) +#loc37 = loc("rstd"(#loc)) +#loc38 = loc("eps"(#loc)) +#loc39 = loc("T"(#loc)) +#loc62 = loc("b_rstd"(#loc25)) +#loc68 = loc(callsite(#loc3 at #loc62)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %b_x = arith.constant dense<0.000000e+00> : tensor<32x128xbf16> loc(#loc40) + %b_rstd = arith.constant dense<1.000000e+00> : tensor<32xf32> loc(#loc41) + %cst = arith.constant dense<128> : tensor<32x1xi64> loc(#loc3) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc42) + %o_t = arith.constant 32 : i64 loc(#loc43) + %i_t = tt.get_program_id x : i32 loc(#loc44) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc45) + %o_t_1 = arith.muli %i_t_0, %o_t : i64 loc(#loc43) + %o_t_2 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32> loc(#loc46) + %o_t_3 = arith.extsi %o_t_2 : tensor<32xi32> to tensor<32xi64> loc(#loc47) + %o_t_4 = tt.splat %o_t_1 : i64 -> tensor<32xi64> loc(#loc47) + %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<32xi64> loc(#loc47) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc48) + %m_t = arith.extsi %T : i32 to i64 loc(#loc49) + %m_t_6 = tt.splat %m_t : i64 -> tensor<32xi64> loc(#loc49) + %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<32xi64> loc(#loc49) + %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<32xi1> -> tensor<32x1xi1> loc(#loc50) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc51) + %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc42) + %m_x_11 = tt.broadcast %m_x_8 : tensor<32x1xi1> -> tensor<32x128xi1> loc(#loc52) + %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<32x128xi1> loc(#loc52) + %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<32x128xi1> loc(#loc52) + %p_x = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<32xi64> -> tensor<32x1xi64> loc(#loc53) + %p_x_14 = arith.muli %p_x, %cst : tensor<32x1xi64> loc(#loc54) + %p_x_15 = tt.splat %x : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc55) + %p_x_16 = tt.addptr %p_x_15, %p_x_14 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc55) + %p_x_17 = tt.broadcast %p_x_16 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc56) + %p_x_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<32x128xi32> loc(#loc56) + %p_x_19 = tt.addptr %p_x_17, %p_x_18 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc56) + %p_y = tt.splat %y : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc57) + %p_y_20 = tt.addptr %p_y, %p_x_14 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc57) + %p_y_21 = tt.broadcast %p_y_20 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc58) + %p_y_22 = tt.addptr %p_y_21, %p_x_18 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<32x!tt.ptr> loc(#loc59) + %p_rstd_23 = tt.addptr %p_rstd, %o_t_5 : tensor<32x!tt.ptr>, tensor<32xi64> loc(#loc59) + %b_x_24 = tt.load %p_x_19, %m_x_13, %b_x : tensor<32x128x!tt.ptr> loc(#loc40) + %b_x_25 = arith.extf %b_x_24 : tensor<32x128xbf16> to tensor<32x128xf32> loc(#loc60) + %b_rstd_26 = arith.mulf %b_x_25, %b_x_25 : tensor<32x128xf32> loc(#loc61) + %b_rstd_27 = "tt.reduce"(%b_rstd_26) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_34: f32 loc(callsite(#loc3 at #loc62)), %b_rstd_35: f32 loc(callsite(#loc3 at #loc62))): + %b_rstd_36 = arith.addf %b_rstd_34, %b_rstd_35 : f32 loc(#loc69) + tt.reduce.return %b_rstd_36 : f32 loc(#loc67) + }) : (tensor<32x128xf32>) -> tensor<32xf32> loc(#loc67) + %b_rstd_28 = tt.splat %eps : f32 -> tensor<32xf32> loc(#loc63) + %b_rstd_29 = arith.addf %b_rstd_27, %b_rstd_28 : tensor<32xf32> loc(#loc63) + %b_rstd_30 = math.sqrt %b_rstd_29 : tensor<32xf32> loc(#loc64) + %b_rstd_31 = arith.divf %b_rstd, %b_rstd_30 : tensor<32xf32> loc(#loc41) + %b_y = tt.expand_dims %b_rstd_31 {axis = 1 : i32} : tensor<32xf32> -> tensor<32x1xf32> loc(#loc65) + %b_y_32 = tt.broadcast %b_y : tensor<32x1xf32> -> tensor<32x128xf32> loc(#loc66) + %b_y_33 = arith.mulf %b_x_25, %b_y_32 : tensor<32x128xf32> loc(#loc66) + %0 = arith.truncf %b_y_33 : tensor<32x128xf32> to tensor<32x128xbf16> loc(#loc31) + tt.store %p_y_22, %0, %m_x_13 : tensor<32x128x!tt.ptr> loc(#loc32) + tt.store %p_rstd_23, %b_rstd_31, %m_t_7 : tensor<32x!tt.ptr> loc(#loc33) + tt.return loc(#loc34) + } loc(#loc) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc40 = loc("b_x"(#loc1)) +#loc41 = loc("b_rstd"(#loc2)) +#loc42 = loc("m_x"(#loc4)) +#loc43 = loc("o_t"(#loc5)) +#loc44 = loc("i_t"(#loc6)) +#loc45 = loc("i_t"(#loc7)) +#loc46 = loc("o_t"(#loc8)) +#loc47 = loc("o_t"(#loc9)) +#loc48 = loc("o_d"(#loc10)) +#loc49 = loc("m_t"(#loc11)) +#loc50 = loc("m_x"(#loc12)) +#loc51 = loc("m_x"(#loc13)) +#loc52 = loc("m_x"(#loc14)) +#loc53 = loc("p_x"(#loc15)) +#loc54 = loc("p_x"(#loc16)) +#loc55 = loc("p_x"(#loc17)) +#loc56 = loc("p_x"(#loc18)) +#loc57 = loc("p_y"(#loc19)) +#loc58 = loc("p_y"(#loc20)) +#loc59 = loc("p_rstd"(#loc21)) +#loc60 = loc("b_x"(#loc22)) +#loc61 = loc("b_rstd"(#loc23)) +#loc63 = loc("b_rstd"(#loc27)) +#loc64 = loc("b_rstd"(#loc28)) +#loc65 = loc("b_y"(#loc29)) +#loc66 = loc("b_y"(#loc30)) +#loc67 = loc(callsite(#loc24 at #loc62)) +#loc69 = loc(callsite(#loc26 at #loc67)) diff --git a/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/__grp__l2norm_fwd_kernel.json b/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/__grp__l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..e96c9f2e910ed048c1eb396f9c82f3f3a833cc06 --- /dev/null +++ b/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/__grp__l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_fwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/l2norm_fwd_kernel.source", "l2norm_fwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/l2norm_fwd_kernel.ttir", "l2norm_fwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/l2norm_fwd_kernel.ttgir", "l2norm_fwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/l2norm_fwd_kernel.llir", "l2norm_fwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/l2norm_fwd_kernel.ptx", "l2norm_fwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/l2norm_fwd_kernel.cubin", "l2norm_fwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/l2norm_fwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/l2norm_fwd_kernel.cubin b/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/l2norm_fwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..d21943cbb588cffc6c90d5a9a5b8738c44506d0d Binary files /dev/null and b/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/l2norm_fwd_kernel.cubin differ diff --git a/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/l2norm_fwd_kernel.json b/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..2f1f77866f2d5f7b0f06f43f3cbc1cd6b091fada --- /dev/null +++ b/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"hash": "12dc83f552b10e8bb2752df2346137df2f35d3258d6f03399ab594aa0479e0a5", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 4, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 32, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_fwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/l2norm_fwd_kernel.llir b/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/l2norm_fwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..639e9082bdf936e023ff352182cf149819527738 --- /dev/null +++ b/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/l2norm_fwd_kernel.llir @@ -0,0 +1,249 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 +@.str.2 = private unnamed_addr constant [17 x i8] c"__CUDA_PREC_SQRT\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_fwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, float %3, i32 %4, ptr addrspace(1) readnone captures(none) %5, ptr addrspace(1) readnone captures(none) %6) local_unnamed_addr #0 !dbg !5 { + %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %9 = zext nneg i32 %8 to i64, !dbg !14 + %10 = shl nuw nsw i64 %9, 3, !dbg !15 + %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !16 + %12 = and i32 %11, 112, !dbg !16 + %13 = lshr exact i32 %12, 4, !dbg !16 + %14 = zext nneg i32 %13 to i64, !dbg !17 + %15 = or disjoint i64 %10, %14, !dbg !17 + %16 = sext i32 %4 to i64, !dbg !18 + %17 = icmp slt i64 %15, %16, !dbg !18 + %18 = shl nuw nsw i32 %11, 3, !dbg !19 + %19 = and i32 %18, 120, !dbg !19 + %20 = shl nuw nsw i64 %15, 7, !dbg !20 + %21 = getelementptr bfloat, ptr addrspace(1) %0, i64 %20, !dbg !21 + %22 = zext nneg i32 %19 to i64, !dbg !22 + %23 = getelementptr bfloat, ptr addrspace(1) %21, i64 %22, !dbg !22 + %24 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %23, i1 %17) #6, !dbg !23 + %25 = extractvalue { i32, i32, i32, i32 } %24, 0, !dbg !23 + %26 = bitcast i32 %25 to <2 x bfloat>, !dbg !23 + %27 = extractvalue { i32, i32, i32, i32 } %24, 1, !dbg !23 + %28 = bitcast i32 %27 to <2 x bfloat>, !dbg !23 + %29 = extractvalue { i32, i32, i32, i32 } %24, 2, !dbg !23 + %30 = bitcast i32 %29 to <2 x bfloat>, !dbg !23 + %31 = extractvalue { i32, i32, i32, i32 } %24, 3, !dbg !23 + %32 = bitcast i32 %31 to <2 x bfloat>, !dbg !23 + %33 = extractelement <2 x bfloat> %26, i64 0, !dbg !23 + %34 = extractelement <2 x bfloat> %26, i64 1, !dbg !23 + %35 = extractelement <2 x bfloat> %28, i64 0, !dbg !23 + %36 = extractelement <2 x bfloat> %28, i64 1, !dbg !23 + %37 = extractelement <2 x bfloat> %30, i64 0, !dbg !23 + %38 = extractelement <2 x bfloat> %30, i64 1, !dbg !23 + %39 = extractelement <2 x bfloat> %32, i64 0, !dbg !23 + %40 = extractelement <2 x bfloat> %32, i64 1, !dbg !23 + %41 = fpext bfloat %33 to float, !dbg !24 + %42 = fpext bfloat %34 to float, !dbg !24 + %43 = fpext bfloat %35 to float, !dbg !24 + %44 = fpext bfloat %36 to float, !dbg !24 + %45 = fpext bfloat %37 to float, !dbg !24 + %46 = fpext bfloat %38 to float, !dbg !24 + %47 = fpext bfloat %39 to float, !dbg !24 + %48 = fpext bfloat %40 to float, !dbg !24 + %49 = fmul float %41, %41, !dbg !25 + %50 = fmul float %42, %42, !dbg !25 + %51 = fmul float %43, %43, !dbg !25 + %52 = fmul float %44, %44, !dbg !25 + %53 = fmul float %45, %45, !dbg !25 + %54 = fmul float %46, %46, !dbg !25 + %55 = fmul float %47, %47, !dbg !25 + %56 = fmul float %48, %48, !dbg !25 + %57 = fadd float %49, %50, !dbg !26 + %58 = fadd float %51, %57, !dbg !26 + %59 = fadd float %52, %58, !dbg !26 + %60 = fadd float %53, %59, !dbg !26 + %61 = fadd float %54, %60, !dbg !26 + %62 = fadd float %55, %61, !dbg !26 + %63 = fadd float %56, %62, !dbg !26 + %64 = bitcast float %63 to i32, !dbg !29 + %65 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %64, i32 8, i32 31), !dbg !29 + %66 = bitcast i32 %65 to float, !dbg !29 + %67 = fadd float %63, %66, !dbg !26 + %68 = bitcast float %67 to i32, !dbg !29 + %69 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %68, i32 4, i32 31), !dbg !29 + %70 = bitcast i32 %69 to float, !dbg !29 + %71 = fadd float %67, %70, !dbg !26 + %72 = bitcast float %71 to i32, !dbg !29 + %73 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %72, i32 2, i32 31), !dbg !29 + %74 = bitcast i32 %73 to float, !dbg !29 + %75 = fadd float %71, %74, !dbg !26 + %76 = bitcast float %75 to i32, !dbg !29 + %77 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %76, i32 1, i32 31), !dbg !29 + %78 = bitcast i32 %77 to float, !dbg !29 + %79 = fadd float %75, %78, !dbg !26 + %80 = fadd float %3, %79, !dbg !32 + %81 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i = icmp eq i32 %81, 0, !dbg !33 + %82 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i = icmp eq i32 %82, 0, !dbg !33 + br i1 %.not.i, label %88, label %83, !dbg !33 + +83: ; preds = %7 + br i1 %.not1.i, label %86, label %84, !dbg !33 + +84: ; preds = %83 + %85 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %80) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +86: ; preds = %83 + %87 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %80) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +88: ; preds = %7 + br i1 %.not1.i, label %91, label %89, !dbg !33 + +89: ; preds = %88 + %90 = tail call float @llvm.nvvm.sqrt.rn.f(float %80) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +91: ; preds = %88 + %92 = tail call float @llvm.nvvm.sqrt.approx.f(float %80) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +__nv_sqrtf.exit: ; preds = %84, %86, %89, %91 + %.0.i = phi float [ %85, %84 ], [ %87, %86 ], [ %90, %89 ], [ %92, %91 ], !dbg !33 + %93 = and i32 %11, 7, !dbg !16 + %94 = zext nneg i32 %93 to i64, !dbg !17 + %95 = or disjoint i64 %10, %94, !dbg !17 + %96 = getelementptr float, ptr addrspace(1) %2, i64 %95, !dbg !34 + %97 = getelementptr bfloat, ptr addrspace(1) %1, i64 %20, !dbg !35 + %98 = getelementptr bfloat, ptr addrspace(1) %97, i64 %22, !dbg !36 + %99 = icmp slt i64 %95, %16, !dbg !18 + %100 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i), !dbg !37 + %101 = fmul float %100, %41, !dbg !38 + %102 = fmul float %100, %42, !dbg !38 + %103 = fmul float %100, %43, !dbg !38 + %104 = fmul float %100, %44, !dbg !38 + %105 = fmul float %100, %45, !dbg !38 + %106 = fmul float %100, %46, !dbg !38 + %107 = fmul float %100, %47, !dbg !38 + %108 = fmul float %100, %48, !dbg !38 + %109 = fptrunc float %101 to bfloat, !dbg !39 + %110 = fptrunc float %102 to bfloat, !dbg !39 + %111 = fptrunc float %103 to bfloat, !dbg !39 + %112 = fptrunc float %104 to bfloat, !dbg !39 + %113 = fptrunc float %105 to bfloat, !dbg !39 + %114 = fptrunc float %106 to bfloat, !dbg !39 + %115 = fptrunc float %107 to bfloat, !dbg !39 + %116 = fptrunc float %108 to bfloat, !dbg !39 + %117 = insertelement <2 x bfloat> poison, bfloat %109, i64 0, !dbg !40 + %118 = insertelement <2 x bfloat> %117, bfloat %110, i64 1, !dbg !40 + %119 = bitcast <2 x bfloat> %118 to i32, !dbg !40 + %120 = insertelement <2 x bfloat> poison, bfloat %111, i64 0, !dbg !40 + %121 = insertelement <2 x bfloat> %120, bfloat %112, i64 1, !dbg !40 + %122 = bitcast <2 x bfloat> %121 to i32, !dbg !40 + %123 = insertelement <2 x bfloat> poison, bfloat %113, i64 0, !dbg !40 + %124 = insertelement <2 x bfloat> %123, bfloat %114, i64 1, !dbg !40 + %125 = bitcast <2 x bfloat> %124 to i32, !dbg !40 + %126 = insertelement <2 x bfloat> poison, bfloat %115, i64 0, !dbg !40 + %127 = insertelement <2 x bfloat> %126, bfloat %116, i64 1, !dbg !40 + %128 = bitcast <2 x bfloat> %127 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %119, i32 %122, i32 %125, i32 %128, ptr addrspace(1) %98, i1 %17) #6, !dbg !40 + %129 = lshr exact i32 %12, 2, !dbg !41 + %130 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %129, !dbg !41 + %131 = bitcast float %100 to <1 x i32>, !dbg !41 + tail call void asm sideeffect "st.shared.b32 [ $0 + 0 ], $1;", "r,r"(ptr addrspace(3) %130, <1 x i32> %131) #6, !dbg !41 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !41 + %132 = shl nuw nsw i32 %93, 2, !dbg !41 + %133 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %132, !dbg !41 + %134 = load i32, ptr addrspace(3) %133, align 4, !dbg !41 + %135 = and i32 %11, 120, !dbg !41 + %136 = icmp eq i32 %135, 0, !dbg !41 + %137 = and i1 %136, %99, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %134, ptr addrspace(1) %96, i1 %137) #6, !dbg !41 + ret void, !dbg !42 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.div.full(float, float) #3 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #4 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #5 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.f(float) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="128" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) } +attributes #4 = { convergent nocallback nounwind } +attributes #5 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #6 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "l2norm_fwd_kernel", linkageName: "l2norm_fwd_kernel", scope: !1, file: !1, line: 69, type: !6, scopeLine: 69, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !10, !11, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 70, column: 24, scope: !5) +!14 = !DILocation(line: 70, column: 30, scope: !5) +!15 = !DILocation(line: 71, column: 16, scope: !5) +!16 = !DILocation(line: 71, column: 34, scope: !5) +!17 = !DILocation(line: 71, column: 21, scope: !5) +!18 = !DILocation(line: 73, column: 16, scope: !5) +!19 = !DILocation(line: 74, column: 30, scope: !5) +!20 = !DILocation(line: 75, column: 29, scope: !5) +!21 = !DILocation(line: 75, column: 14, scope: !5) +!22 = !DILocation(line: 75, column: 33, scope: !5) +!23 = !DILocation(line: 79, column: 18, scope: !5) +!24 = !DILocation(line: 79, column: 47, scope: !5) +!25 = !DILocation(line: 80, column: 38, scope: !5) +!26 = !DILocation(line: 263, column: 15, scope: !27, inlinedAt: !29) +!27 = distinct !DILexicalBlockFile(scope: !5, file: !28, discriminator: 0) +!28 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!29 = !DILocation(line: 293, column: 36, scope: !27, inlinedAt: !30) +!30 = !DILocation(line: 80, column: 43, scope: !31) +!31 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!32 = !DILocation(line: 80, column: 48, scope: !5) +!33 = !DILocation(line: 80, column: 25, scope: !5) +!34 = !DILocation(line: 77, column: 20, scope: !5) +!35 = !DILocation(line: 76, column: 14, scope: !5) +!36 = !DILocation(line: 76, column: 33, scope: !5) +!37 = !DILocation(line: 80, column: 17, scope: !5) +!38 = !DILocation(line: 81, column: 16, scope: !5) +!39 = !DILocation(line: 83, column: 25, scope: !5) +!40 = !DILocation(line: 83, column: 18, scope: !5) +!41 = !DILocation(line: 84, column: 21, scope: !5) +!42 = !DILocation(line: 84, column: 4, scope: !5) diff --git a/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/l2norm_fwd_kernel.ptx b/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/l2norm_fwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..4aa67ae72e3c97c8320f79c6dfc50a50840500f4 --- /dev/null +++ b/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/l2norm_fwd_kernel.ptx @@ -0,0 +1,399 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_fwd_kernel // -- Begin function l2norm_fwd_kernel +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; +.global .align 1 .b8 _$_str_$_2[17] = {95, 95, 67, 85, 68, 65, 95, 80, 82, 69, 67, 95, 83, 81, 82, 84}; + // @l2norm_fwd_kernel +.visible .entry l2norm_fwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_2, + .param .f32 l2norm_fwd_kernel_param_3, + .param .u32 l2norm_fwd_kernel_param_4, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_6 +) +.reqntid 128 +{ + .reg .pred %p<5>; + .reg .b16 %rs<9>; + .reg .b32 %r<61>; + .reg .b64 %rd<18>; + .loc 1 69 0 // l2norm.py:69:0 +$L__func_begin0: + .loc 1 69 0 // l2norm.py:69:0 + +// %bb.0: // %__nv_sqrtf.exit + ld.param.b64 %rd4, [l2norm_fwd_kernel_param_0]; + ld.param.b64 %rd5, [l2norm_fwd_kernel_param_1]; +$L__tmp0: + .loc 1 70 24 // l2norm.py:70:24 + mov.u32 %r13, %ctaid.x; + ld.param.b64 %rd6, [l2norm_fwd_kernel_param_2]; + .loc 1 71 16 // l2norm.py:71:16 + mul.wide.u32 %rd7, %r13, 8; + ld.param.b32 %r14, [l2norm_fwd_kernel_param_3]; + ld.param.s32 %rd8, [l2norm_fwd_kernel_param_4]; + .loc 1 71 34 // l2norm.py:71:34 + mov.u32 %r15, %tid.x; + and.b32 %r16, %r15, 112; + bfe.u32 %r17, %r15, 4, 3; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd9, %r17; + or.b64 %rd10, %rd7, %rd9; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p1, %rd10, %rd8; + .loc 1 74 30 // l2norm.py:74:30 + shl.b32 %r18, %r15, 3; + and.b32 %r19, %r18, 120; + .loc 1 75 14 // l2norm.py:75:14 + shl.b64 %rd11, %rd10, 8; + add.s64 %rd12, %rd4, %rd11; + .loc 1 75 33 // l2norm.py:75:33 + mul.wide.u32 %rd13, %r19, 2; + add.s64 %rd1, %rd12, %rd13; + mov.b32 %r5, 0; + .loc 1 79 18 // l2norm.py:79:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + mov.b32 {%rs5, %rs6}, %r3; + mov.b32 {%rs7, %rs8}, %r4; + .loc 1 79 47 // l2norm.py:79:47 + cvt.f32.bf16 %r20, %rs1; + cvt.f32.bf16 %r21, %rs2; + cvt.f32.bf16 %r22, %rs3; + cvt.f32.bf16 %r23, %rs4; + cvt.f32.bf16 %r24, %rs5; + cvt.f32.bf16 %r25, %rs6; + cvt.f32.bf16 %r26, %rs7; + cvt.f32.bf16 %r27, %rs8; + .loc 1 80 38 // l2norm.py:80:38 + mul.f32 %r28, %r21, %r21; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + fma.rn.f32.bf16 %r29, %rs1, %rs1, %r28; + fma.rn.f32.bf16 %r30, %rs3, %rs3, %r29; + fma.rn.f32.bf16 %r31, %rs4, %rs4, %r30; + fma.rn.f32.bf16 %r32, %rs5, %rs5, %r31; + fma.rn.f32.bf16 %r33, %rs6, %rs6, %r32; + fma.rn.f32.bf16 %r34, %rs7, %rs7, %r33; + fma.rn.f32.bf16 %r35, %rs8, %rs8, %r34; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r36, %r35, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r37, %r35, %r36; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r38, %r37, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r39, %r37, %r38; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r40, %r39, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r41, %r39, %r40; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r42, %r41, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r43, %r41, %r42; +$L__tmp10: + .loc 1 80 48 // l2norm.py:80:48 + add.f32 %r44, %r14, %r43; + .loc 1 80 25 // l2norm.py:80:25 + sqrt.approx.ftz.f32 %r45, %r44; + .loc 1 71 34 // l2norm.py:71:34 + and.b32 %r46, %r15, 7; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd14, %r46; + or.b64 %rd15, %rd7, %rd14; + .loc 1 77 20 // l2norm.py:77:20 + shl.b64 %rd16, %rd15, 2; + add.s64 %rd3, %rd6, %rd16; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd17, %rd5, %rd11; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd2, %rd17, %rd13; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p3, %rd15, %rd8; + mov.b32 %r47, 0f3F800000; + .loc 1 80 17 // l2norm.py:80:17 + div.full.f32 %r11, %r47, %r45; + .loc 1 81 16 // l2norm.py:81:16 + mul.f32 %r48, %r11, %r20; + mul.f32 %r49, %r11, %r21; + mul.f32 %r50, %r11, %r22; + mul.f32 %r51, %r11, %r23; + mul.f32 %r52, %r11, %r24; + mul.f32 %r53, %r11, %r25; + mul.f32 %r54, %r11, %r26; + mul.f32 %r55, %r11, %r27; + .loc 1 83 18 // l2norm.py:83:18 + cvt.rn.bf16x2.f32 %r6, %r49, %r48; + cvt.rn.bf16x2.f32 %r7, %r51, %r50; + cvt.rn.bf16x2.f32 %r8, %r53, %r52; + cvt.rn.bf16x2.f32 %r9, %r55, %r54; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd2 + 0 ], { %r6, %r7, %r8, %r9 }; + // end inline asm + .loc 1 84 21 // l2norm.py:84:21 + shr.u32 %r56, %r16, 2; + mov.b32 %r57, global_smem; + add.s32 %r10, %r57, %r56; + // begin inline asm + st.shared.b32 [ %r10 + 0 ], %r11; + // end inline asm + bar.sync 0; + shl.b32 %r58, %r46, 2; + add.s32 %r59, %r57, %r58; + ld.shared.b32 %r12, [%r59]; + and.b32 %r60, %r15, 120; + setp.eq.b32 %p4, %r60, 0; + and.pred %p2, %p4, %p3; + // begin inline asm + @%p2 st.global.b32 [ %rd3 + 0 ], { %r12 }; + // end inline asm + .loc 1 84 4 // l2norm.py:84:4 + ret; +$L__tmp11: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp10 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 80 // DW_AT_call_line +.b8 43 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp10 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/l2norm_fwd_kernel.source b/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/l2norm_fwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..c7c5a735c81381ccb0efd0d2a8f8d86240068e11 --- /dev/null +++ b/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/l2norm_fwd_kernel.source @@ -0,0 +1,172 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc38 = loc(unknown) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc45 = loc("x"(#loc)) +#loc46 = loc("y"(#loc)) +#loc47 = loc("rstd"(#loc)) +#loc48 = loc("eps"(#loc)) +#loc49 = loc("T"(#loc)) +#loc81 = loc("input"(#loc36)) +#loc82 = loc("a"(#loc41)) +#loc83 = loc("b"(#loc41)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc50) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc51) + %o_t = arith.constant 8 : i32 loc(#loc52) + %o_t_1 = arith.constant 8 : i64 loc(#loc52) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc52) + %o_t_3 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32> loc(#loc53) + %o_t_4 = arith.extsi %o_t_3 : tensor<8xi32> to tensor<8xi64> loc(#loc54) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<8xi64> loc(#loc54) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<8xi64> loc(#loc54) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc55) + %m_t = arith.extsi %T : i32 to i64 loc(#loc56) + %m_t_7 = tt.splat %m_t : i64 -> tensor<8xi64> loc(#loc56) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<8xi64> loc(#loc56) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<8xi1> -> tensor<8x1xi1> loc(#loc57) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc58) + %m_x_10 = arith.constant 128 : i32 loc(#loc59) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc59) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc59) + %m_x_13 = tt.broadcast %m_x : tensor<8x1xi1> -> tensor<8x128xi1> loc(#loc60) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<8x128xi1> loc(#loc60) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<8x128xi1> loc(#loc60) + %p_x = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc61) + %p_x_16 = arith.constant 128 : i32 loc(#loc62) + %p_x_17 = arith.constant 128 : i64 loc(#loc62) + %p_x_18 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc62) + %p_x_19 = arith.muli %p_x, %p_x_18 : tensor<8x1xi64> loc(#loc62) + %p_x_20 = tt.splat %x : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc63) + %p_x_21 = tt.addptr %p_x_20, %p_x_19 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc63) + %p_x_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc64) + %p_x_23 = tt.broadcast %p_x_21 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc65) + %p_x_24 = tt.broadcast %p_x_22 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc65) + %p_x_25 = tt.addptr %p_x_23, %p_x_24 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc65) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc66) + %p_y_26 = arith.constant 128 : i32 loc(#loc67) + %p_y_27 = arith.constant 128 : i64 loc(#loc67) + %p_y_28 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc67) + %p_y_29 = arith.muli %p_y, %p_y_28 : tensor<8x1xi64> loc(#loc67) + %p_y_30 = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc68) + %p_y_31 = tt.addptr %p_y_30, %p_y_29 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc68) + %p_y_32 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc69) + %p_y_33 = tt.broadcast %p_y_31 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc70) + %p_y_34 = tt.broadcast %p_y_32 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc70) + %p_y_35 = tt.addptr %p_y_33, %p_y_34 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc70) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr> loc(#loc71) + %p_rstd_36 = tt.addptr %p_rstd, %o_t_6 : tensor<8x!tt.ptr>, tensor<8xi64> loc(#loc71) + %b_x = arith.constant 0.000000e+00 : f32 loc(#loc72) + %b_x_37 = arith.constant dense<0.000000e+00> : tensor<8x128xf32> loc(#loc72) + %b_x_38 = arith.truncf %b_x_37 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc72) + %b_x_39 = tt.load %p_x_25, %m_x_15, %b_x_38 : tensor<8x128x!tt.ptr> loc(#loc72) + %b_x_40 = arith.extf %b_x_39 : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc73) + %b_rstd = arith.mulf %b_x_40, %b_x_40 : tensor<8x128xf32> loc(#loc74) + %b_rstd_41 = tt.call @triton.language.standard.sum__fp32S8_128S_c1_cFalse_cNone(%b_rstd) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc75) + %b_rstd_42 = tt.splat %eps : f32 -> tensor<8xf32> loc(#loc76) + %b_rstd_43 = arith.addf %b_rstd_41, %b_rstd_42 : tensor<8xf32> loc(#loc76) + %b_rstd_44 = math.sqrt %b_rstd_43 : tensor<8xf32> loc(#loc77) + %b_rstd_45 = arith.constant 1 : i32 loc(#loc78) + %b_rstd_46 = arith.constant 1.000000e+00 : f32 loc(#loc78) + %b_rstd_47 = arith.constant dense<1.000000e+00> : tensor<8xf32> loc(#loc78) + %b_rstd_48 = arith.divf %b_rstd_47, %b_rstd_44 : tensor<8xf32> loc(#loc78) + %b_y = tt.expand_dims %b_rstd_48 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc79) + %b_y_49 = tt.broadcast %b_y : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc80) + %b_y_50 = arith.mulf %b_x_40, %b_y_49 : tensor<8x128xf32> loc(#loc80) + %0 = arith.truncf %b_y_50 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc32) + tt.store %p_y_35, %0, %m_x_15 : tensor<8x128x!tt.ptr> loc(#loc33) + tt.store %p_rstd_36, %b_rstd_48, %m_t_8 : tensor<8x!tt.ptr> loc(#loc34) + tt.return loc(#loc35) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S8_128S_c1_cFalse_cNone(%input: tensor<8x128xf32> loc("input"(#loc36))) -> tensor<8xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc37) + tt.reduce.return %2 : f32 loc(#loc37) + }) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc37) + tt.return %0 : tensor<8xf32> loc(#loc39) + ^bb1: // no predecessors + %1 = ub.poison : tensor<8xf32> loc(#loc40) + tt.return %1 : tensor<8xf32> loc(#loc40) + } loc(#loc36) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc41)), %b: f32 loc("b"(#loc41))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc42) + tt.return %0 : f32 loc(#loc43) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc44) + tt.return %1 : f32 loc(#loc44) + } loc(#loc41) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:18) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:29) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:37) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc50 = loc("i_t"(#loc1)) +#loc51 = loc("i_t"(#loc2)) +#loc52 = loc("o_t"(#loc3)) +#loc53 = loc("o_t"(#loc4)) +#loc54 = loc("o_t"(#loc5)) +#loc55 = loc("o_d"(#loc6)) +#loc56 = loc("m_t"(#loc7)) +#loc57 = loc("m_x"(#loc8)) +#loc58 = loc("m_x"(#loc9)) +#loc59 = loc("m_x"(#loc10)) +#loc60 = loc("m_x"(#loc11)) +#loc61 = loc("p_x"(#loc12)) +#loc62 = loc("p_x"(#loc13)) +#loc63 = loc("p_x"(#loc14)) +#loc64 = loc("p_x"(#loc15)) +#loc65 = loc("p_x"(#loc16)) +#loc66 = loc("p_y"(#loc17)) +#loc67 = loc("p_y"(#loc18)) +#loc68 = loc("p_y"(#loc19)) +#loc69 = loc("p_y"(#loc20)) +#loc70 = loc("p_y"(#loc21)) +#loc71 = loc("p_rstd"(#loc22)) +#loc72 = loc("b_x"(#loc23)) +#loc73 = loc("b_x"(#loc24)) +#loc74 = loc("b_rstd"(#loc25)) +#loc75 = loc("b_rstd"(#loc26)) +#loc76 = loc("b_rstd"(#loc27)) +#loc77 = loc("b_rstd"(#loc28)) +#loc78 = loc("b_rstd"(#loc29)) +#loc79 = loc("b_y"(#loc30)) +#loc80 = loc("b_y"(#loc31)) diff --git a/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/l2norm_fwd_kernel.ttgir b/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/l2norm_fwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..72b0f913da168c2c233286b0953ca7e41e1d4463 --- /dev/null +++ b/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/l2norm_fwd_kernel.ttgir @@ -0,0 +1,135 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [4, 1], order = [1, 0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [4], order = [0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc1 = loc(unknown) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc34 = loc("x"(#loc)) +#loc35 = loc("y"(#loc)) +#loc36 = loc("rstd"(#loc)) +#loc37 = loc("eps"(#loc)) +#loc38 = loc("T"(#loc)) +#loc59 = loc("b_rstd"(#loc23)) +#loc66 = loc(callsite(#loc1 at #loc59)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 4 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<1.000000e+00> : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<8x128xbf16, #blocked> loc(#loc1) + %c8_i64 = arith.constant 8 : i64 loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<8x1xi64, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc39) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc40) + %o_t = arith.muli %i_t_3, %c8_i64 : i64 loc(#loc41) + %o_t_4 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc42) + %o_t_5 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32, #blocked1> loc(#loc42) + %o_t_6 = arith.extsi %o_t_4 : tensor<8xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_7 = arith.extsi %o_t_5 : tensor<8xi32, #blocked1> to tensor<8xi64, #blocked1> loc(#loc43) + %o_t_8 = tt.splat %o_t : i64 -> tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_9 = tt.splat %o_t : i64 -> tensor<8xi64, #blocked1> loc(#loc43) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<8xi64, #blocked1> loc(#loc43) + %m_t = arith.extsi %T : i32 to i64 loc(#loc44) + %m_t_12 = tt.splat %m_t : i64 -> tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_13 = tt.splat %m_t : i64 -> tensor<8xi64, #blocked1> loc(#loc44) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<8xi64, #blocked1> loc(#loc44) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<8xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xi1, #blocked> loc(#loc45) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc46) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc46) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_1 : tensor<1x128xi32, #blocked> loc(#loc47) + %m_x_19 = tt.broadcast %m_x : tensor<8x1xi1, #blocked> -> tensor<8x128xi1, #blocked> loc(#loc48) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<8x128xi1, #blocked> loc(#loc48) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<8x128xi1, #blocked> loc(#loc48) + %p_x = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xi64, #blocked> loc(#loc49) + %p_x_22 = arith.muli %p_x, %cst_2 : tensor<8x1xi64, #blocked> loc(#loc50) + %p_x_23 = tt.splat %x : !tt.ptr -> tensor<8x1x!tt.ptr, #blocked> loc(#loc51) + %p_x_24 = tt.addptr %p_x_23, %p_x_22 : tensor<8x1x!tt.ptr, #blocked>, tensor<8x1xi64, #blocked> loc(#loc51) + %p_x_25 = tt.broadcast %p_x_24 : tensor<8x1x!tt.ptr, #blocked> -> tensor<8x128x!tt.ptr, #blocked> loc(#loc52) + %p_x_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<8x128xi32, #blocked> loc(#loc52) + %p_x_27 = tt.addptr %p_x_25, %p_x_26 : tensor<8x128x!tt.ptr, #blocked>, tensor<8x128xi32, #blocked> loc(#loc52) + %p_y = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr, #blocked> loc(#loc53) + %p_y_28 = tt.addptr %p_y, %p_x_22 : tensor<8x1x!tt.ptr, #blocked>, tensor<8x1xi64, #blocked> loc(#loc53) + %p_y_29 = tt.broadcast %p_y_28 : tensor<8x1x!tt.ptr, #blocked> -> tensor<8x128x!tt.ptr, #blocked> loc(#loc54) + %p_y_30 = tt.addptr %p_y_29, %p_x_26 : tensor<8x128x!tt.ptr, #blocked>, tensor<8x128xi32, #blocked> loc(#loc54) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr, #blocked1> loc(#loc55) + %p_rstd_31 = tt.addptr %p_rstd, %o_t_11 : tensor<8x!tt.ptr, #blocked1>, tensor<8xi64, #blocked1> loc(#loc55) + %b_x = tt.load %p_x_27, %m_x_21, %cst_0 : tensor<8x128x!tt.ptr, #blocked> loc(#loc56) + %b_x_32 = arith.extf %b_x : tensor<8x128xbf16, #blocked> to tensor<8x128xf32, #blocked> loc(#loc57) + %b_rstd = arith.mulf %b_x_32, %b_x_32 : tensor<8x128xf32, #blocked> loc(#loc58) + %b_rstd_33 = "tt.reduce"(%b_rstd) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_40: f32 loc(callsite(#loc1 at #loc59)), %b_rstd_41: f32 loc(callsite(#loc1 at #loc59))): + %b_rstd_42 = arith.addf %b_rstd_40, %b_rstd_41 : f32 loc(#loc67) + tt.reduce.return %b_rstd_42 : f32 loc(#loc65) + }) : (tensor<8x128xf32, #blocked>) -> tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc65) + %b_rstd_34 = tt.splat %eps : f32 -> tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_36 = math.sqrt %b_rstd_35 : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc61) + %b_rstd_37 = arith.divf %cst, %b_rstd_36 : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc62) + %b_y = tt.expand_dims %b_rstd_37 {axis = 1 : i32} : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xf32, #blocked> loc(#loc63) + %b_y_38 = tt.broadcast %b_y : tensor<8x1xf32, #blocked> -> tensor<8x128xf32, #blocked> loc(#loc64) + %b_y_39 = arith.mulf %b_x_32, %b_y_38 : tensor<8x128xf32, #blocked> loc(#loc64) + %0 = arith.truncf %b_y_39 : tensor<8x128xf32, #blocked> to tensor<8x128xbf16, #blocked> loc(#loc30) + tt.store %p_y_30, %0, %m_x_21 : tensor<8x128x!tt.ptr, #blocked> loc(#loc31) + %1 = ttg.convert_layout %b_rstd_37 : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8xf32, #blocked1> loc(#loc32) + tt.store %p_rstd_31, %1, %m_t_15 : tensor<8x!tt.ptr, #blocked1> loc(#loc32) + tt.return loc(#loc33) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc39 = loc("i_t"(#loc2)) +#loc40 = loc("i_t"(#loc3)) +#loc41 = loc("o_t"(#loc4)) +#loc42 = loc("o_t"(#loc5)) +#loc43 = loc("o_t"(#loc6)) +#loc44 = loc("m_t"(#loc7)) +#loc45 = loc("m_x"(#loc8)) +#loc46 = loc("m_x"(#loc9)) +#loc47 = loc("m_x"(#loc10)) +#loc48 = loc("m_x"(#loc11)) +#loc49 = loc("p_x"(#loc12)) +#loc50 = loc("p_x"(#loc13)) +#loc51 = loc("p_x"(#loc14)) +#loc52 = loc("p_x"(#loc15)) +#loc53 = loc("p_y"(#loc16)) +#loc54 = loc("p_y"(#loc17)) +#loc55 = loc("p_rstd"(#loc18)) +#loc56 = loc("b_x"(#loc19)) +#loc57 = loc("b_x"(#loc20)) +#loc58 = loc("b_rstd"(#loc21)) +#loc60 = loc("b_rstd"(#loc25)) +#loc61 = loc("b_rstd"(#loc26)) +#loc62 = loc("b_rstd"(#loc27)) +#loc63 = loc("b_y"(#loc28)) +#loc64 = loc("b_y"(#loc29)) +#loc65 = loc(callsite(#loc22 at #loc59)) +#loc67 = loc(callsite(#loc24 at #loc65)) diff --git a/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/l2norm_fwd_kernel.ttir b/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/l2norm_fwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..8bbf17da02489a0d195ed0bd020180f9e4ac2496 --- /dev/null +++ b/triton/gpu0/CLOIH5KSWEHIXMTVFXZDIYJX34XTLUZFRVXQGOM2WWKKUBDZ4CSQ/l2norm_fwd_kernel.ttir @@ -0,0 +1,128 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc3 = loc(unknown) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc35 = loc("x"(#loc)) +#loc36 = loc("y"(#loc)) +#loc37 = loc("rstd"(#loc)) +#loc38 = loc("eps"(#loc)) +#loc39 = loc("T"(#loc)) +#loc62 = loc("b_rstd"(#loc25)) +#loc68 = loc(callsite(#loc3 at #loc62)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %b_x = arith.constant dense<0.000000e+00> : tensor<8x128xbf16> loc(#loc40) + %b_rstd = arith.constant dense<1.000000e+00> : tensor<8xf32> loc(#loc41) + %cst = arith.constant dense<128> : tensor<8x1xi64> loc(#loc3) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc42) + %o_t = arith.constant 8 : i64 loc(#loc43) + %i_t = tt.get_program_id x : i32 loc(#loc44) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc45) + %o_t_1 = arith.muli %i_t_0, %o_t : i64 loc(#loc43) + %o_t_2 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32> loc(#loc46) + %o_t_3 = arith.extsi %o_t_2 : tensor<8xi32> to tensor<8xi64> loc(#loc47) + %o_t_4 = tt.splat %o_t_1 : i64 -> tensor<8xi64> loc(#loc47) + %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<8xi64> loc(#loc47) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc48) + %m_t = arith.extsi %T : i32 to i64 loc(#loc49) + %m_t_6 = tt.splat %m_t : i64 -> tensor<8xi64> loc(#loc49) + %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<8xi64> loc(#loc49) + %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<8xi1> -> tensor<8x1xi1> loc(#loc50) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc51) + %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc42) + %m_x_11 = tt.broadcast %m_x_8 : tensor<8x1xi1> -> tensor<8x128xi1> loc(#loc52) + %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<8x128xi1> loc(#loc52) + %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<8x128xi1> loc(#loc52) + %p_x = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc53) + %p_x_14 = arith.muli %p_x, %cst : tensor<8x1xi64> loc(#loc54) + %p_x_15 = tt.splat %x : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc55) + %p_x_16 = tt.addptr %p_x_15, %p_x_14 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc55) + %p_x_17 = tt.broadcast %p_x_16 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc56) + %p_x_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc56) + %p_x_19 = tt.addptr %p_x_17, %p_x_18 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc56) + %p_y = tt.splat %y : !tt.ptr -> tensor<8x1x!tt.ptr> loc(#loc57) + %p_y_20 = tt.addptr %p_y, %p_x_14 : tensor<8x1x!tt.ptr>, tensor<8x1xi64> loc(#loc57) + %p_y_21 = tt.broadcast %p_y_20 : tensor<8x1x!tt.ptr> -> tensor<8x128x!tt.ptr> loc(#loc58) + %p_y_22 = tt.addptr %p_y_21, %p_x_18 : tensor<8x128x!tt.ptr>, tensor<8x128xi32> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<8x!tt.ptr> loc(#loc59) + %p_rstd_23 = tt.addptr %p_rstd, %o_t_5 : tensor<8x!tt.ptr>, tensor<8xi64> loc(#loc59) + %b_x_24 = tt.load %p_x_19, %m_x_13, %b_x : tensor<8x128x!tt.ptr> loc(#loc40) + %b_x_25 = arith.extf %b_x_24 : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc60) + %b_rstd_26 = arith.mulf %b_x_25, %b_x_25 : tensor<8x128xf32> loc(#loc61) + %b_rstd_27 = "tt.reduce"(%b_rstd_26) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_34: f32 loc(callsite(#loc3 at #loc62)), %b_rstd_35: f32 loc(callsite(#loc3 at #loc62))): + %b_rstd_36 = arith.addf %b_rstd_34, %b_rstd_35 : f32 loc(#loc69) + tt.reduce.return %b_rstd_36 : f32 loc(#loc67) + }) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc67) + %b_rstd_28 = tt.splat %eps : f32 -> tensor<8xf32> loc(#loc63) + %b_rstd_29 = arith.addf %b_rstd_27, %b_rstd_28 : tensor<8xf32> loc(#loc63) + %b_rstd_30 = math.sqrt %b_rstd_29 : tensor<8xf32> loc(#loc64) + %b_rstd_31 = arith.divf %b_rstd, %b_rstd_30 : tensor<8xf32> loc(#loc41) + %b_y = tt.expand_dims %b_rstd_31 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc65) + %b_y_32 = tt.broadcast %b_y : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc66) + %b_y_33 = arith.mulf %b_x_25, %b_y_32 : tensor<8x128xf32> loc(#loc66) + %0 = arith.truncf %b_y_33 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc31) + tt.store %p_y_22, %0, %m_x_13 : tensor<8x128x!tt.ptr> loc(#loc32) + tt.store %p_rstd_23, %b_rstd_31, %m_t_7 : tensor<8x!tt.ptr> loc(#loc33) + tt.return loc(#loc34) + } loc(#loc) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc40 = loc("b_x"(#loc1)) +#loc41 = loc("b_rstd"(#loc2)) +#loc42 = loc("m_x"(#loc4)) +#loc43 = loc("o_t"(#loc5)) +#loc44 = loc("i_t"(#loc6)) +#loc45 = loc("i_t"(#loc7)) +#loc46 = loc("o_t"(#loc8)) +#loc47 = loc("o_t"(#loc9)) +#loc48 = loc("o_d"(#loc10)) +#loc49 = loc("m_t"(#loc11)) +#loc50 = loc("m_x"(#loc12)) +#loc51 = loc("m_x"(#loc13)) +#loc52 = loc("m_x"(#loc14)) +#loc53 = loc("p_x"(#loc15)) +#loc54 = loc("p_x"(#loc16)) +#loc55 = loc("p_x"(#loc17)) +#loc56 = loc("p_x"(#loc18)) +#loc57 = loc("p_y"(#loc19)) +#loc58 = loc("p_y"(#loc20)) +#loc59 = loc("p_rstd"(#loc21)) +#loc60 = loc("b_x"(#loc22)) +#loc61 = loc("b_rstd"(#loc23)) +#loc63 = loc("b_rstd"(#loc27)) +#loc64 = loc("b_rstd"(#loc28)) +#loc65 = loc("b_y"(#loc29)) +#loc66 = loc("b_y"(#loc30)) +#loc67 = loc(callsite(#loc24 at #loc62)) +#loc69 = loc(callsite(#loc26 at #loc67)) diff --git a/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/__grp__l2norm_fwd_kernel.json b/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/__grp__l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..a58bea0f438c3c1653af5e60d4653143c659c488 --- /dev/null +++ b/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/__grp__l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_fwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.source", "l2norm_fwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.ttir", "l2norm_fwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.ttgir", "l2norm_fwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.llir", "l2norm_fwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.ptx", "l2norm_fwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.cubin", "l2norm_fwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.cubin b/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..e617d532bfd0aa8e273513ca942044878259763a --- /dev/null +++ b/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.cubin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:90dbfe30b76c691acb0425888b284030758d8f8ba0f1e207b84590a8b84ef881 +size 160296 diff --git a/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.json b/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..de4b762a1f46b3fff52db6af8706ff116bbadc76 --- /dev/null +++ b/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"hash": "1307baa9c0b88d3a83649374d98349b8dc8699f85d1a4ddd7fb791fcc761f740", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 1, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 256, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_fwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.llir b/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..e58d03c0890ee8ee7e1bc22729785de0e4699ee4 --- /dev/null +++ b/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.llir @@ -0,0 +1,4174 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 +@.str.2 = private unnamed_addr constant [17 x i8] c"__CUDA_PREC_SQRT\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_fwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, float %3, i32 %4, ptr addrspace(1) readnone captures(none) %5, ptr addrspace(1) readnone captures(none) %6) local_unnamed_addr #0 !dbg !5 { + %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %9 = zext nneg i32 %8 to i64, !dbg !14 + %10 = shl nuw nsw i64 %9, 6, !dbg !15 + %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !16 + %12 = and i32 %11, 16, !dbg !16 + %.lobit = lshr exact i32 %12, 4, !dbg !16 + %13 = or disjoint i32 %.lobit, 2, !dbg !16 + %14 = insertelement <2 x i32> poison, i32 %.lobit, i64 0, !dbg !16 + %15 = insertelement <4 x i32> poison, i32 %.lobit, i64 0, !dbg !16 + %16 = insertelement <8 x i32> poison, i32 %.lobit, i64 0, !dbg !16 + %17 = insertelement <16 x i32> poison, i32 %.lobit, i64 0, !dbg !16 + %18 = insertelement <32 x i32> poison, i32 %13, i64 30, !dbg !17 + %19 = insertelement <32 x i32> %18, i32 %.lobit, i64 31, !dbg !17 + %20 = shufflevector <16 x i32> %17, <16 x i32> poison, <32 x i32> , !dbg !17 + %21 = or disjoint <32 x i32> %20, , !dbg !17 + %22 = shufflevector <32 x i32> %21, <32 x i32> %19, <32 x i32> , !dbg !17 + %23 = shufflevector <8 x i32> %16, <8 x i32> poison, <32 x i32> , !dbg !17 + %24 = or disjoint <32 x i32> %23, , !dbg !17 + %25 = shufflevector <32 x i32> %22, <32 x i32> %24, <32 x i32> , !dbg !17 + %26 = shufflevector <4 x i32> %15, <4 x i32> poison, <32 x i32> , !dbg !17 + %27 = or disjoint <32 x i32> %26, , !dbg !17 + %28 = shufflevector <32 x i32> %25, <32 x i32> %27, <32 x i32> , !dbg !17 + %29 = shufflevector <2 x i32> %14, <2 x i32> poison, <32 x i32> , !dbg !17 + %30 = or disjoint <32 x i32> %29, , !dbg !17 + %31 = shufflevector <32 x i32> %28, <32 x i32> %30, <32 x i32> , !dbg !17 + %32 = zext <32 x i32> %31 to <32 x i64>, !dbg !17 + %33 = insertelement <32 x i64> poison, i64 %10, i64 0, !dbg !17 + %34 = shufflevector <32 x i64> %33, <32 x i64> poison, <32 x i32> zeroinitializer, !dbg !17 + %35 = or disjoint <32 x i64> %34, %32, !dbg !17 + %36 = insertelement <32 x i32> poison, i32 %4, i64 0, !dbg !18 + %37 = shufflevector <32 x i32> %36, <32 x i32> poison, <32 x i32> zeroinitializer, !dbg !18 + %38 = sext <32 x i32> %37 to <32 x i64>, !dbg !18 + %39 = icmp slt <32 x i64> %35, %38, !dbg !18 + %40 = and i32 %11, 15, !dbg !19 + %41 = shl nuw nsw i32 %40, 3, !dbg !19 + %42 = extractelement <32 x i64> %35, i64 31, !dbg !20 + %43 = shl nuw nsw i64 %42, 7, !dbg !20 + %44 = extractelement <32 x i64> %35, i64 30, !dbg !20 + %45 = shl nuw nsw i64 %44, 7, !dbg !20 + %46 = extractelement <32 x i64> %35, i64 29, !dbg !20 + %47 = shl nuw nsw i64 %46, 7, !dbg !20 + %48 = extractelement <32 x i64> %35, i64 28, !dbg !20 + %49 = shl nuw nsw i64 %48, 7, !dbg !20 + %50 = extractelement <32 x i64> %35, i64 27, !dbg !20 + %51 = shl nuw nsw i64 %50, 7, !dbg !20 + %52 = extractelement <32 x i64> %35, i64 26, !dbg !20 + %53 = shl nuw nsw i64 %52, 7, !dbg !20 + %54 = extractelement <32 x i64> %35, i64 25, !dbg !20 + %55 = shl nuw nsw i64 %54, 7, !dbg !20 + %56 = extractelement <32 x i64> %35, i64 24, !dbg !20 + %57 = shl nuw nsw i64 %56, 7, !dbg !20 + %58 = extractelement <32 x i64> %35, i64 23, !dbg !20 + %59 = shl nuw nsw i64 %58, 7, !dbg !20 + %60 = extractelement <32 x i64> %35, i64 22, !dbg !20 + %61 = shl nuw nsw i64 %60, 7, !dbg !20 + %62 = extractelement <32 x i64> %35, i64 21, !dbg !20 + %63 = shl nuw nsw i64 %62, 7, !dbg !20 + %64 = extractelement <32 x i64> %35, i64 20, !dbg !20 + %65 = shl nuw nsw i64 %64, 7, !dbg !20 + %66 = extractelement <32 x i64> %35, i64 19, !dbg !20 + %67 = shl nuw nsw i64 %66, 7, !dbg !20 + %68 = extractelement <32 x i64> %35, i64 18, !dbg !20 + %69 = shl nuw nsw i64 %68, 7, !dbg !20 + %70 = extractelement <32 x i64> %35, i64 17, !dbg !20 + %71 = shl nuw nsw i64 %70, 7, !dbg !20 + %72 = extractelement <32 x i64> %35, i64 16, !dbg !20 + %73 = shl nuw nsw i64 %72, 7, !dbg !20 + %74 = extractelement <32 x i64> %35, i64 15, !dbg !20 + %75 = shl nuw nsw i64 %74, 7, !dbg !20 + %76 = extractelement <32 x i64> %35, i64 14, !dbg !20 + %77 = shl nuw nsw i64 %76, 7, !dbg !20 + %78 = extractelement <32 x i64> %35, i64 13, !dbg !20 + %79 = shl nuw nsw i64 %78, 7, !dbg !20 + %80 = extractelement <32 x i64> %35, i64 12, !dbg !20 + %81 = shl nuw nsw i64 %80, 7, !dbg !20 + %82 = extractelement <32 x i64> %35, i64 11, !dbg !20 + %83 = shl nuw nsw i64 %82, 7, !dbg !20 + %84 = extractelement <32 x i64> %35, i64 10, !dbg !20 + %85 = shl nuw nsw i64 %84, 7, !dbg !20 + %86 = extractelement <32 x i64> %35, i64 9, !dbg !20 + %87 = shl nuw nsw i64 %86, 7, !dbg !20 + %88 = extractelement <32 x i64> %35, i64 8, !dbg !20 + %89 = shl nuw nsw i64 %88, 7, !dbg !20 + %90 = extractelement <32 x i64> %35, i64 7, !dbg !20 + %91 = shl nuw nsw i64 %90, 7, !dbg !20 + %92 = extractelement <32 x i64> %35, i64 6, !dbg !20 + %93 = shl nuw nsw i64 %92, 7, !dbg !20 + %94 = extractelement <32 x i64> %35, i64 5, !dbg !20 + %95 = shl nuw nsw i64 %94, 7, !dbg !20 + %96 = extractelement <32 x i64> %35, i64 4, !dbg !20 + %97 = shl nuw nsw i64 %96, 7, !dbg !20 + %98 = extractelement <32 x i64> %35, i64 3, !dbg !20 + %99 = shl nuw nsw i64 %98, 7, !dbg !20 + %100 = extractelement <32 x i64> %35, i64 2, !dbg !20 + %101 = shl nuw nsw i64 %100, 7, !dbg !20 + %102 = extractelement <32 x i64> %35, i64 1, !dbg !20 + %103 = shl nuw nsw i64 %102, 7, !dbg !20 + %104 = extractelement <32 x i64> %35, i64 0, !dbg !20 + %105 = shl nuw nsw i64 %104, 7, !dbg !20 + %106 = getelementptr bfloat, ptr addrspace(1) %0, i64 %43, !dbg !21 + %107 = getelementptr bfloat, ptr addrspace(1) %0, i64 %45, !dbg !21 + %108 = getelementptr bfloat, ptr addrspace(1) %0, i64 %47, !dbg !21 + %109 = getelementptr bfloat, ptr addrspace(1) %0, i64 %49, !dbg !21 + %110 = getelementptr bfloat, ptr addrspace(1) %0, i64 %51, !dbg !21 + %111 = getelementptr bfloat, ptr addrspace(1) %0, i64 %53, !dbg !21 + %112 = getelementptr bfloat, ptr addrspace(1) %0, i64 %55, !dbg !21 + %113 = getelementptr bfloat, ptr addrspace(1) %0, i64 %57, !dbg !21 + %114 = getelementptr bfloat, ptr addrspace(1) %0, i64 %59, !dbg !21 + %115 = getelementptr bfloat, ptr addrspace(1) %0, i64 %61, !dbg !21 + %116 = getelementptr bfloat, ptr addrspace(1) %0, i64 %63, !dbg !21 + %117 = getelementptr bfloat, ptr addrspace(1) %0, i64 %65, !dbg !21 + %118 = getelementptr bfloat, ptr addrspace(1) %0, i64 %67, !dbg !21 + %119 = getelementptr bfloat, ptr addrspace(1) %0, i64 %69, !dbg !21 + %120 = getelementptr bfloat, ptr addrspace(1) %0, i64 %71, !dbg !21 + %121 = getelementptr bfloat, ptr addrspace(1) %0, i64 %73, !dbg !21 + %122 = getelementptr bfloat, ptr addrspace(1) %0, i64 %75, !dbg !21 + %123 = getelementptr bfloat, ptr addrspace(1) %0, i64 %77, !dbg !21 + %124 = getelementptr bfloat, ptr addrspace(1) %0, i64 %79, !dbg !21 + %125 = getelementptr bfloat, ptr addrspace(1) %0, i64 %81, !dbg !21 + %126 = getelementptr bfloat, ptr addrspace(1) %0, i64 %83, !dbg !21 + %127 = getelementptr bfloat, ptr addrspace(1) %0, i64 %85, !dbg !21 + %128 = getelementptr bfloat, ptr addrspace(1) %0, i64 %87, !dbg !21 + %129 = getelementptr bfloat, ptr addrspace(1) %0, i64 %89, !dbg !21 + %130 = getelementptr bfloat, ptr addrspace(1) %0, i64 %91, !dbg !21 + %131 = getelementptr bfloat, ptr addrspace(1) %0, i64 %93, !dbg !21 + %132 = getelementptr bfloat, ptr addrspace(1) %0, i64 %95, !dbg !21 + %133 = getelementptr bfloat, ptr addrspace(1) %0, i64 %97, !dbg !21 + %134 = getelementptr bfloat, ptr addrspace(1) %0, i64 %99, !dbg !21 + %135 = getelementptr bfloat, ptr addrspace(1) %0, i64 %101, !dbg !21 + %136 = getelementptr bfloat, ptr addrspace(1) %0, i64 %103, !dbg !21 + %137 = getelementptr bfloat, ptr addrspace(1) %0, i64 %105, !dbg !21 + %138 = zext nneg i32 %41 to i64, !dbg !22 + %139 = getelementptr bfloat, ptr addrspace(1) %106, i64 %138, !dbg !22 + %140 = getelementptr bfloat, ptr addrspace(1) %107, i64 %138, !dbg !22 + %141 = getelementptr bfloat, ptr addrspace(1) %108, i64 %138, !dbg !22 + %142 = getelementptr bfloat, ptr addrspace(1) %109, i64 %138, !dbg !22 + %143 = getelementptr bfloat, ptr addrspace(1) %110, i64 %138, !dbg !22 + %144 = getelementptr bfloat, ptr addrspace(1) %111, i64 %138, !dbg !22 + %145 = getelementptr bfloat, ptr addrspace(1) %112, i64 %138, !dbg !22 + %146 = getelementptr bfloat, ptr addrspace(1) %113, i64 %138, !dbg !22 + %147 = getelementptr bfloat, ptr addrspace(1) %114, i64 %138, !dbg !22 + %148 = getelementptr bfloat, ptr addrspace(1) %115, i64 %138, !dbg !22 + %149 = getelementptr bfloat, ptr addrspace(1) %116, i64 %138, !dbg !22 + %150 = getelementptr bfloat, ptr addrspace(1) %117, i64 %138, !dbg !22 + %151 = getelementptr bfloat, ptr addrspace(1) %118, i64 %138, !dbg !22 + %152 = getelementptr bfloat, ptr addrspace(1) %119, i64 %138, !dbg !22 + %153 = getelementptr bfloat, ptr addrspace(1) %120, i64 %138, !dbg !22 + %154 = getelementptr bfloat, ptr addrspace(1) %121, i64 %138, !dbg !22 + %155 = getelementptr bfloat, ptr addrspace(1) %122, i64 %138, !dbg !22 + %156 = getelementptr bfloat, ptr addrspace(1) %123, i64 %138, !dbg !22 + %157 = getelementptr bfloat, ptr addrspace(1) %124, i64 %138, !dbg !22 + %158 = getelementptr bfloat, ptr addrspace(1) %125, i64 %138, !dbg !22 + %159 = getelementptr bfloat, ptr addrspace(1) %126, i64 %138, !dbg !22 + %160 = getelementptr bfloat, ptr addrspace(1) %127, i64 %138, !dbg !22 + %161 = getelementptr bfloat, ptr addrspace(1) %128, i64 %138, !dbg !22 + %162 = getelementptr bfloat, ptr addrspace(1) %129, i64 %138, !dbg !22 + %163 = getelementptr bfloat, ptr addrspace(1) %130, i64 %138, !dbg !22 + %164 = getelementptr bfloat, ptr addrspace(1) %131, i64 %138, !dbg !22 + %165 = getelementptr bfloat, ptr addrspace(1) %132, i64 %138, !dbg !22 + %166 = getelementptr bfloat, ptr addrspace(1) %133, i64 %138, !dbg !22 + %167 = getelementptr bfloat, ptr addrspace(1) %134, i64 %138, !dbg !22 + %168 = getelementptr bfloat, ptr addrspace(1) %135, i64 %138, !dbg !22 + %169 = getelementptr bfloat, ptr addrspace(1) %136, i64 %138, !dbg !22 + %170 = getelementptr bfloat, ptr addrspace(1) %137, i64 %138, !dbg !22 + %171 = extractelement <32 x i1> %39, i64 31, !dbg !23 + %172 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %139, i1 %171) #7, !dbg !23 + %173 = extractvalue { i32, i32, i32, i32 } %172, 0, !dbg !23 + %174 = bitcast i32 %173 to <2 x bfloat>, !dbg !23 + %175 = extractvalue { i32, i32, i32, i32 } %172, 1, !dbg !23 + %176 = bitcast i32 %175 to <2 x bfloat>, !dbg !23 + %177 = extractvalue { i32, i32, i32, i32 } %172, 2, !dbg !23 + %178 = bitcast i32 %177 to <2 x bfloat>, !dbg !23 + %179 = extractvalue { i32, i32, i32, i32 } %172, 3, !dbg !23 + %180 = bitcast i32 %179 to <2 x bfloat>, !dbg !23 + %181 = extractelement <2 x bfloat> %174, i64 0, !dbg !23 + %182 = extractelement <2 x bfloat> %174, i64 1, !dbg !23 + %183 = extractelement <2 x bfloat> %176, i64 0, !dbg !23 + %184 = extractelement <2 x bfloat> %176, i64 1, !dbg !23 + %185 = extractelement <2 x bfloat> %178, i64 0, !dbg !23 + %186 = extractelement <2 x bfloat> %178, i64 1, !dbg !23 + %187 = extractelement <2 x bfloat> %180, i64 0, !dbg !23 + %188 = extractelement <2 x bfloat> %180, i64 1, !dbg !23 + %189 = extractelement <32 x i1> %39, i64 30, !dbg !23 + %190 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %140, i1 %189) #7, !dbg !23 + %191 = extractvalue { i32, i32, i32, i32 } %190, 0, !dbg !23 + %192 = bitcast i32 %191 to <2 x bfloat>, !dbg !23 + %193 = extractvalue { i32, i32, i32, i32 } %190, 1, !dbg !23 + %194 = bitcast i32 %193 to <2 x bfloat>, !dbg !23 + %195 = extractvalue { i32, i32, i32, i32 } %190, 2, !dbg !23 + %196 = bitcast i32 %195 to <2 x bfloat>, !dbg !23 + %197 = extractvalue { i32, i32, i32, i32 } %190, 3, !dbg !23 + %198 = bitcast i32 %197 to <2 x bfloat>, !dbg !23 + %199 = extractelement <2 x bfloat> %192, i64 0, !dbg !23 + %200 = extractelement <2 x bfloat> %192, i64 1, !dbg !23 + %201 = extractelement <2 x bfloat> %194, i64 0, !dbg !23 + %202 = extractelement <2 x bfloat> %194, i64 1, !dbg !23 + %203 = extractelement <2 x bfloat> %196, i64 0, !dbg !23 + %204 = extractelement <2 x bfloat> %196, i64 1, !dbg !23 + %205 = extractelement <2 x bfloat> %198, i64 0, !dbg !23 + %206 = extractelement <2 x bfloat> %198, i64 1, !dbg !23 + %207 = extractelement <32 x i1> %39, i64 29, !dbg !23 + %208 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %141, i1 %207) #7, !dbg !23 + %209 = extractvalue { i32, i32, i32, i32 } %208, 0, !dbg !23 + %210 = bitcast i32 %209 to <2 x bfloat>, !dbg !23 + %211 = extractvalue { i32, i32, i32, i32 } %208, 1, !dbg !23 + %212 = bitcast i32 %211 to <2 x bfloat>, !dbg !23 + %213 = extractvalue { i32, i32, i32, i32 } %208, 2, !dbg !23 + %214 = bitcast i32 %213 to <2 x bfloat>, !dbg !23 + %215 = extractvalue { i32, i32, i32, i32 } %208, 3, !dbg !23 + %216 = bitcast i32 %215 to <2 x bfloat>, !dbg !23 + %217 = extractelement <2 x bfloat> %210, i64 0, !dbg !23 + %218 = extractelement <2 x bfloat> %210, i64 1, !dbg !23 + %219 = extractelement <2 x bfloat> %212, i64 0, !dbg !23 + %220 = extractelement <2 x bfloat> %212, i64 1, !dbg !23 + %221 = extractelement <2 x bfloat> %214, i64 0, !dbg !23 + %222 = extractelement <2 x bfloat> %214, i64 1, !dbg !23 + %223 = extractelement <2 x bfloat> %216, i64 0, !dbg !23 + %224 = extractelement <2 x bfloat> %216, i64 1, !dbg !23 + %225 = extractelement <32 x i1> %39, i64 28, !dbg !23 + %226 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %142, i1 %225) #7, !dbg !23 + %227 = extractvalue { i32, i32, i32, i32 } %226, 0, !dbg !23 + %228 = bitcast i32 %227 to <2 x bfloat>, !dbg !23 + %229 = extractvalue { i32, i32, i32, i32 } %226, 1, !dbg !23 + %230 = bitcast i32 %229 to <2 x bfloat>, !dbg !23 + %231 = extractvalue { i32, i32, i32, i32 } %226, 2, !dbg !23 + %232 = bitcast i32 %231 to <2 x bfloat>, !dbg !23 + %233 = extractvalue { i32, i32, i32, i32 } %226, 3, !dbg !23 + %234 = bitcast i32 %233 to <2 x bfloat>, !dbg !23 + %235 = extractelement <2 x bfloat> %228, i64 0, !dbg !23 + %236 = extractelement <2 x bfloat> %228, i64 1, !dbg !23 + %237 = extractelement <2 x bfloat> %230, i64 0, !dbg !23 + %238 = extractelement <2 x bfloat> %230, i64 1, !dbg !23 + %239 = extractelement <2 x bfloat> %232, i64 0, !dbg !23 + %240 = extractelement <2 x bfloat> %232, i64 1, !dbg !23 + %241 = extractelement <2 x bfloat> %234, i64 0, !dbg !23 + %242 = extractelement <2 x bfloat> %234, i64 1, !dbg !23 + %243 = extractelement <32 x i1> %39, i64 27, !dbg !23 + %244 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %143, i1 %243) #7, !dbg !23 + %245 = extractvalue { i32, i32, i32, i32 } %244, 0, !dbg !23 + %246 = bitcast i32 %245 to <2 x bfloat>, !dbg !23 + %247 = extractvalue { i32, i32, i32, i32 } %244, 1, !dbg !23 + %248 = bitcast i32 %247 to <2 x bfloat>, !dbg !23 + %249 = extractvalue { i32, i32, i32, i32 } %244, 2, !dbg !23 + %250 = bitcast i32 %249 to <2 x bfloat>, !dbg !23 + %251 = extractvalue { i32, i32, i32, i32 } %244, 3, !dbg !23 + %252 = bitcast i32 %251 to <2 x bfloat>, !dbg !23 + %253 = extractelement <2 x bfloat> %246, i64 0, !dbg !23 + %254 = extractelement <2 x bfloat> %246, i64 1, !dbg !23 + %255 = extractelement <2 x bfloat> %248, i64 0, !dbg !23 + %256 = extractelement <2 x bfloat> %248, i64 1, !dbg !23 + %257 = extractelement <2 x bfloat> %250, i64 0, !dbg !23 + %258 = extractelement <2 x bfloat> %250, i64 1, !dbg !23 + %259 = extractelement <2 x bfloat> %252, i64 0, !dbg !23 + %260 = extractelement <2 x bfloat> %252, i64 1, !dbg !23 + %261 = extractelement <32 x i1> %39, i64 26, !dbg !23 + %262 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %144, i1 %261) #7, !dbg !23 + %263 = extractvalue { i32, i32, i32, i32 } %262, 0, !dbg !23 + %264 = bitcast i32 %263 to <2 x bfloat>, !dbg !23 + %265 = extractvalue { i32, i32, i32, i32 } %262, 1, !dbg !23 + %266 = bitcast i32 %265 to <2 x bfloat>, !dbg !23 + %267 = extractvalue { i32, i32, i32, i32 } %262, 2, !dbg !23 + %268 = bitcast i32 %267 to <2 x bfloat>, !dbg !23 + %269 = extractvalue { i32, i32, i32, i32 } %262, 3, !dbg !23 + %270 = bitcast i32 %269 to <2 x bfloat>, !dbg !23 + %271 = extractelement <2 x bfloat> %264, i64 0, !dbg !23 + %272 = extractelement <2 x bfloat> %264, i64 1, !dbg !23 + %273 = extractelement <2 x bfloat> %266, i64 0, !dbg !23 + %274 = extractelement <2 x bfloat> %266, i64 1, !dbg !23 + %275 = extractelement <2 x bfloat> %268, i64 0, !dbg !23 + %276 = extractelement <2 x bfloat> %268, i64 1, !dbg !23 + %277 = extractelement <2 x bfloat> %270, i64 0, !dbg !23 + %278 = extractelement <2 x bfloat> %270, i64 1, !dbg !23 + %279 = extractelement <32 x i1> %39, i64 25, !dbg !23 + %280 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %145, i1 %279) #7, !dbg !23 + %281 = extractvalue { i32, i32, i32, i32 } %280, 0, !dbg !23 + %282 = bitcast i32 %281 to <2 x bfloat>, !dbg !23 + %283 = extractvalue { i32, i32, i32, i32 } %280, 1, !dbg !23 + %284 = bitcast i32 %283 to <2 x bfloat>, !dbg !23 + %285 = extractvalue { i32, i32, i32, i32 } %280, 2, !dbg !23 + %286 = bitcast i32 %285 to <2 x bfloat>, !dbg !23 + %287 = extractvalue { i32, i32, i32, i32 } %280, 3, !dbg !23 + %288 = bitcast i32 %287 to <2 x bfloat>, !dbg !23 + %289 = extractelement <2 x bfloat> %282, i64 0, !dbg !23 + %290 = extractelement <2 x bfloat> %282, i64 1, !dbg !23 + %291 = extractelement <2 x bfloat> %284, i64 0, !dbg !23 + %292 = extractelement <2 x bfloat> %284, i64 1, !dbg !23 + %293 = extractelement <2 x bfloat> %286, i64 0, !dbg !23 + %294 = extractelement <2 x bfloat> %286, i64 1, !dbg !23 + %295 = extractelement <2 x bfloat> %288, i64 0, !dbg !23 + %296 = extractelement <2 x bfloat> %288, i64 1, !dbg !23 + %297 = extractelement <32 x i1> %39, i64 24, !dbg !23 + %298 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %146, i1 %297) #7, !dbg !23 + %299 = extractvalue { i32, i32, i32, i32 } %298, 0, !dbg !23 + %300 = bitcast i32 %299 to <2 x bfloat>, !dbg !23 + %301 = extractvalue { i32, i32, i32, i32 } %298, 1, !dbg !23 + %302 = bitcast i32 %301 to <2 x bfloat>, !dbg !23 + %303 = extractvalue { i32, i32, i32, i32 } %298, 2, !dbg !23 + %304 = bitcast i32 %303 to <2 x bfloat>, !dbg !23 + %305 = extractvalue { i32, i32, i32, i32 } %298, 3, !dbg !23 + %306 = bitcast i32 %305 to <2 x bfloat>, !dbg !23 + %307 = extractelement <2 x bfloat> %300, i64 0, !dbg !23 + %308 = extractelement <2 x bfloat> %300, i64 1, !dbg !23 + %309 = extractelement <2 x bfloat> %302, i64 0, !dbg !23 + %310 = extractelement <2 x bfloat> %302, i64 1, !dbg !23 + %311 = extractelement <2 x bfloat> %304, i64 0, !dbg !23 + %312 = extractelement <2 x bfloat> %304, i64 1, !dbg !23 + %313 = extractelement <2 x bfloat> %306, i64 0, !dbg !23 + %314 = extractelement <2 x bfloat> %306, i64 1, !dbg !23 + %315 = extractelement <32 x i1> %39, i64 23, !dbg !23 + %316 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %147, i1 %315) #7, !dbg !23 + %317 = extractvalue { i32, i32, i32, i32 } %316, 0, !dbg !23 + %318 = bitcast i32 %317 to <2 x bfloat>, !dbg !23 + %319 = extractvalue { i32, i32, i32, i32 } %316, 1, !dbg !23 + %320 = bitcast i32 %319 to <2 x bfloat>, !dbg !23 + %321 = extractvalue { i32, i32, i32, i32 } %316, 2, !dbg !23 + %322 = bitcast i32 %321 to <2 x bfloat>, !dbg !23 + %323 = extractvalue { i32, i32, i32, i32 } %316, 3, !dbg !23 + %324 = bitcast i32 %323 to <2 x bfloat>, !dbg !23 + %325 = extractelement <2 x bfloat> %318, i64 0, !dbg !23 + %326 = extractelement <2 x bfloat> %318, i64 1, !dbg !23 + %327 = extractelement <2 x bfloat> %320, i64 0, !dbg !23 + %328 = extractelement <2 x bfloat> %320, i64 1, !dbg !23 + %329 = extractelement <2 x bfloat> %322, i64 0, !dbg !23 + %330 = extractelement <2 x bfloat> %322, i64 1, !dbg !23 + %331 = extractelement <2 x bfloat> %324, i64 0, !dbg !23 + %332 = extractelement <2 x bfloat> %324, i64 1, !dbg !23 + %333 = extractelement <32 x i1> %39, i64 22, !dbg !23 + %334 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %148, i1 %333) #7, !dbg !23 + %335 = extractvalue { i32, i32, i32, i32 } %334, 0, !dbg !23 + %336 = bitcast i32 %335 to <2 x bfloat>, !dbg !23 + %337 = extractvalue { i32, i32, i32, i32 } %334, 1, !dbg !23 + %338 = bitcast i32 %337 to <2 x bfloat>, !dbg !23 + %339 = extractvalue { i32, i32, i32, i32 } %334, 2, !dbg !23 + %340 = bitcast i32 %339 to <2 x bfloat>, !dbg !23 + %341 = extractvalue { i32, i32, i32, i32 } %334, 3, !dbg !23 + %342 = bitcast i32 %341 to <2 x bfloat>, !dbg !23 + %343 = extractelement <2 x bfloat> %336, i64 0, !dbg !23 + %344 = extractelement <2 x bfloat> %336, i64 1, !dbg !23 + %345 = extractelement <2 x bfloat> %338, i64 0, !dbg !23 + %346 = extractelement <2 x bfloat> %338, i64 1, !dbg !23 + %347 = extractelement <2 x bfloat> %340, i64 0, !dbg !23 + %348 = extractelement <2 x bfloat> %340, i64 1, !dbg !23 + %349 = extractelement <2 x bfloat> %342, i64 0, !dbg !23 + %350 = extractelement <2 x bfloat> %342, i64 1, !dbg !23 + %351 = extractelement <32 x i1> %39, i64 21, !dbg !23 + %352 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %149, i1 %351) #7, !dbg !23 + %353 = extractvalue { i32, i32, i32, i32 } %352, 0, !dbg !23 + %354 = bitcast i32 %353 to <2 x bfloat>, !dbg !23 + %355 = extractvalue { i32, i32, i32, i32 } %352, 1, !dbg !23 + %356 = bitcast i32 %355 to <2 x bfloat>, !dbg !23 + %357 = extractvalue { i32, i32, i32, i32 } %352, 2, !dbg !23 + %358 = bitcast i32 %357 to <2 x bfloat>, !dbg !23 + %359 = extractvalue { i32, i32, i32, i32 } %352, 3, !dbg !23 + %360 = bitcast i32 %359 to <2 x bfloat>, !dbg !23 + %361 = extractelement <2 x bfloat> %354, i64 0, !dbg !23 + %362 = extractelement <2 x bfloat> %354, i64 1, !dbg !23 + %363 = extractelement <2 x bfloat> %356, i64 0, !dbg !23 + %364 = extractelement <2 x bfloat> %356, i64 1, !dbg !23 + %365 = extractelement <2 x bfloat> %358, i64 0, !dbg !23 + %366 = extractelement <2 x bfloat> %358, i64 1, !dbg !23 + %367 = extractelement <2 x bfloat> %360, i64 0, !dbg !23 + %368 = extractelement <2 x bfloat> %360, i64 1, !dbg !23 + %369 = extractelement <32 x i1> %39, i64 20, !dbg !23 + %370 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %150, i1 %369) #7, !dbg !23 + %371 = extractvalue { i32, i32, i32, i32 } %370, 0, !dbg !23 + %372 = bitcast i32 %371 to <2 x bfloat>, !dbg !23 + %373 = extractvalue { i32, i32, i32, i32 } %370, 1, !dbg !23 + %374 = bitcast i32 %373 to <2 x bfloat>, !dbg !23 + %375 = extractvalue { i32, i32, i32, i32 } %370, 2, !dbg !23 + %376 = bitcast i32 %375 to <2 x bfloat>, !dbg !23 + %377 = extractvalue { i32, i32, i32, i32 } %370, 3, !dbg !23 + %378 = bitcast i32 %377 to <2 x bfloat>, !dbg !23 + %379 = extractelement <2 x bfloat> %372, i64 0, !dbg !23 + %380 = extractelement <2 x bfloat> %372, i64 1, !dbg !23 + %381 = extractelement <2 x bfloat> %374, i64 0, !dbg !23 + %382 = extractelement <2 x bfloat> %374, i64 1, !dbg !23 + %383 = extractelement <2 x bfloat> %376, i64 0, !dbg !23 + %384 = extractelement <2 x bfloat> %376, i64 1, !dbg !23 + %385 = extractelement <2 x bfloat> %378, i64 0, !dbg !23 + %386 = extractelement <2 x bfloat> %378, i64 1, !dbg !23 + %387 = extractelement <32 x i1> %39, i64 19, !dbg !23 + %388 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %151, i1 %387) #7, !dbg !23 + %389 = extractvalue { i32, i32, i32, i32 } %388, 0, !dbg !23 + %390 = bitcast i32 %389 to <2 x bfloat>, !dbg !23 + %391 = extractvalue { i32, i32, i32, i32 } %388, 1, !dbg !23 + %392 = bitcast i32 %391 to <2 x bfloat>, !dbg !23 + %393 = extractvalue { i32, i32, i32, i32 } %388, 2, !dbg !23 + %394 = bitcast i32 %393 to <2 x bfloat>, !dbg !23 + %395 = extractvalue { i32, i32, i32, i32 } %388, 3, !dbg !23 + %396 = bitcast i32 %395 to <2 x bfloat>, !dbg !23 + %397 = extractelement <2 x bfloat> %390, i64 0, !dbg !23 + %398 = extractelement <2 x bfloat> %390, i64 1, !dbg !23 + %399 = extractelement <2 x bfloat> %392, i64 0, !dbg !23 + %400 = extractelement <2 x bfloat> %392, i64 1, !dbg !23 + %401 = extractelement <2 x bfloat> %394, i64 0, !dbg !23 + %402 = extractelement <2 x bfloat> %394, i64 1, !dbg !23 + %403 = extractelement <2 x bfloat> %396, i64 0, !dbg !23 + %404 = extractelement <2 x bfloat> %396, i64 1, !dbg !23 + %405 = extractelement <32 x i1> %39, i64 18, !dbg !23 + %406 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %152, i1 %405) #7, !dbg !23 + %407 = extractvalue { i32, i32, i32, i32 } %406, 0, !dbg !23 + %408 = bitcast i32 %407 to <2 x bfloat>, !dbg !23 + %409 = extractvalue { i32, i32, i32, i32 } %406, 1, !dbg !23 + %410 = bitcast i32 %409 to <2 x bfloat>, !dbg !23 + %411 = extractvalue { i32, i32, i32, i32 } %406, 2, !dbg !23 + %412 = bitcast i32 %411 to <2 x bfloat>, !dbg !23 + %413 = extractvalue { i32, i32, i32, i32 } %406, 3, !dbg !23 + %414 = bitcast i32 %413 to <2 x bfloat>, !dbg !23 + %415 = extractelement <2 x bfloat> %408, i64 0, !dbg !23 + %416 = extractelement <2 x bfloat> %408, i64 1, !dbg !23 + %417 = extractelement <2 x bfloat> %410, i64 0, !dbg !23 + %418 = extractelement <2 x bfloat> %410, i64 1, !dbg !23 + %419 = extractelement <2 x bfloat> %412, i64 0, !dbg !23 + %420 = extractelement <2 x bfloat> %412, i64 1, !dbg !23 + %421 = extractelement <2 x bfloat> %414, i64 0, !dbg !23 + %422 = extractelement <2 x bfloat> %414, i64 1, !dbg !23 + %423 = extractelement <32 x i1> %39, i64 17, !dbg !23 + %424 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %153, i1 %423) #7, !dbg !23 + %425 = extractvalue { i32, i32, i32, i32 } %424, 0, !dbg !23 + %426 = bitcast i32 %425 to <2 x bfloat>, !dbg !23 + %427 = extractvalue { i32, i32, i32, i32 } %424, 1, !dbg !23 + %428 = bitcast i32 %427 to <2 x bfloat>, !dbg !23 + %429 = extractvalue { i32, i32, i32, i32 } %424, 2, !dbg !23 + %430 = bitcast i32 %429 to <2 x bfloat>, !dbg !23 + %431 = extractvalue { i32, i32, i32, i32 } %424, 3, !dbg !23 + %432 = bitcast i32 %431 to <2 x bfloat>, !dbg !23 + %433 = extractelement <2 x bfloat> %426, i64 0, !dbg !23 + %434 = extractelement <2 x bfloat> %426, i64 1, !dbg !23 + %435 = extractelement <2 x bfloat> %428, i64 0, !dbg !23 + %436 = extractelement <2 x bfloat> %428, i64 1, !dbg !23 + %437 = extractelement <2 x bfloat> %430, i64 0, !dbg !23 + %438 = extractelement <2 x bfloat> %430, i64 1, !dbg !23 + %439 = extractelement <2 x bfloat> %432, i64 0, !dbg !23 + %440 = extractelement <2 x bfloat> %432, i64 1, !dbg !23 + %441 = extractelement <32 x i1> %39, i64 16, !dbg !23 + %442 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %154, i1 %441) #7, !dbg !23 + %443 = extractvalue { i32, i32, i32, i32 } %442, 0, !dbg !23 + %444 = bitcast i32 %443 to <2 x bfloat>, !dbg !23 + %445 = extractvalue { i32, i32, i32, i32 } %442, 1, !dbg !23 + %446 = bitcast i32 %445 to <2 x bfloat>, !dbg !23 + %447 = extractvalue { i32, i32, i32, i32 } %442, 2, !dbg !23 + %448 = bitcast i32 %447 to <2 x bfloat>, !dbg !23 + %449 = extractvalue { i32, i32, i32, i32 } %442, 3, !dbg !23 + %450 = bitcast i32 %449 to <2 x bfloat>, !dbg !23 + %451 = extractelement <2 x bfloat> %444, i64 0, !dbg !23 + %452 = extractelement <2 x bfloat> %444, i64 1, !dbg !23 + %453 = extractelement <2 x bfloat> %446, i64 0, !dbg !23 + %454 = extractelement <2 x bfloat> %446, i64 1, !dbg !23 + %455 = extractelement <2 x bfloat> %448, i64 0, !dbg !23 + %456 = extractelement <2 x bfloat> %448, i64 1, !dbg !23 + %457 = extractelement <2 x bfloat> %450, i64 0, !dbg !23 + %458 = extractelement <2 x bfloat> %450, i64 1, !dbg !23 + %459 = extractelement <32 x i1> %39, i64 15, !dbg !23 + %460 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %155, i1 %459) #7, !dbg !23 + %461 = extractvalue { i32, i32, i32, i32 } %460, 0, !dbg !23 + %462 = bitcast i32 %461 to <2 x bfloat>, !dbg !23 + %463 = extractvalue { i32, i32, i32, i32 } %460, 1, !dbg !23 + %464 = bitcast i32 %463 to <2 x bfloat>, !dbg !23 + %465 = extractvalue { i32, i32, i32, i32 } %460, 2, !dbg !23 + %466 = bitcast i32 %465 to <2 x bfloat>, !dbg !23 + %467 = extractvalue { i32, i32, i32, i32 } %460, 3, !dbg !23 + %468 = bitcast i32 %467 to <2 x bfloat>, !dbg !23 + %469 = extractelement <2 x bfloat> %462, i64 0, !dbg !23 + %470 = extractelement <2 x bfloat> %462, i64 1, !dbg !23 + %471 = extractelement <2 x bfloat> %464, i64 0, !dbg !23 + %472 = extractelement <2 x bfloat> %464, i64 1, !dbg !23 + %473 = extractelement <2 x bfloat> %466, i64 0, !dbg !23 + %474 = extractelement <2 x bfloat> %466, i64 1, !dbg !23 + %475 = extractelement <2 x bfloat> %468, i64 0, !dbg !23 + %476 = extractelement <2 x bfloat> %468, i64 1, !dbg !23 + %477 = extractelement <32 x i1> %39, i64 14, !dbg !23 + %478 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %156, i1 %477) #7, !dbg !23 + %479 = extractvalue { i32, i32, i32, i32 } %478, 0, !dbg !23 + %480 = bitcast i32 %479 to <2 x bfloat>, !dbg !23 + %481 = extractvalue { i32, i32, i32, i32 } %478, 1, !dbg !23 + %482 = bitcast i32 %481 to <2 x bfloat>, !dbg !23 + %483 = extractvalue { i32, i32, i32, i32 } %478, 2, !dbg !23 + %484 = bitcast i32 %483 to <2 x bfloat>, !dbg !23 + %485 = extractvalue { i32, i32, i32, i32 } %478, 3, !dbg !23 + %486 = bitcast i32 %485 to <2 x bfloat>, !dbg !23 + %487 = extractelement <2 x bfloat> %480, i64 0, !dbg !23 + %488 = extractelement <2 x bfloat> %480, i64 1, !dbg !23 + %489 = extractelement <2 x bfloat> %482, i64 0, !dbg !23 + %490 = extractelement <2 x bfloat> %482, i64 1, !dbg !23 + %491 = extractelement <2 x bfloat> %484, i64 0, !dbg !23 + %492 = extractelement <2 x bfloat> %484, i64 1, !dbg !23 + %493 = extractelement <2 x bfloat> %486, i64 0, !dbg !23 + %494 = extractelement <2 x bfloat> %486, i64 1, !dbg !23 + %495 = extractelement <32 x i1> %39, i64 13, !dbg !23 + %496 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %157, i1 %495) #7, !dbg !23 + %497 = extractvalue { i32, i32, i32, i32 } %496, 0, !dbg !23 + %498 = bitcast i32 %497 to <2 x bfloat>, !dbg !23 + %499 = extractvalue { i32, i32, i32, i32 } %496, 1, !dbg !23 + %500 = bitcast i32 %499 to <2 x bfloat>, !dbg !23 + %501 = extractvalue { i32, i32, i32, i32 } %496, 2, !dbg !23 + %502 = bitcast i32 %501 to <2 x bfloat>, !dbg !23 + %503 = extractvalue { i32, i32, i32, i32 } %496, 3, !dbg !23 + %504 = bitcast i32 %503 to <2 x bfloat>, !dbg !23 + %505 = extractelement <2 x bfloat> %498, i64 0, !dbg !23 + %506 = extractelement <2 x bfloat> %498, i64 1, !dbg !23 + %507 = extractelement <2 x bfloat> %500, i64 0, !dbg !23 + %508 = extractelement <2 x bfloat> %500, i64 1, !dbg !23 + %509 = extractelement <2 x bfloat> %502, i64 0, !dbg !23 + %510 = extractelement <2 x bfloat> %502, i64 1, !dbg !23 + %511 = extractelement <2 x bfloat> %504, i64 0, !dbg !23 + %512 = extractelement <2 x bfloat> %504, i64 1, !dbg !23 + %513 = extractelement <32 x i1> %39, i64 12, !dbg !23 + %514 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %158, i1 %513) #7, !dbg !23 + %515 = extractvalue { i32, i32, i32, i32 } %514, 0, !dbg !23 + %516 = bitcast i32 %515 to <2 x bfloat>, !dbg !23 + %517 = extractvalue { i32, i32, i32, i32 } %514, 1, !dbg !23 + %518 = bitcast i32 %517 to <2 x bfloat>, !dbg !23 + %519 = extractvalue { i32, i32, i32, i32 } %514, 2, !dbg !23 + %520 = bitcast i32 %519 to <2 x bfloat>, !dbg !23 + %521 = extractvalue { i32, i32, i32, i32 } %514, 3, !dbg !23 + %522 = bitcast i32 %521 to <2 x bfloat>, !dbg !23 + %523 = extractelement <2 x bfloat> %516, i64 0, !dbg !23 + %524 = extractelement <2 x bfloat> %516, i64 1, !dbg !23 + %525 = extractelement <2 x bfloat> %518, i64 0, !dbg !23 + %526 = extractelement <2 x bfloat> %518, i64 1, !dbg !23 + %527 = extractelement <2 x bfloat> %520, i64 0, !dbg !23 + %528 = extractelement <2 x bfloat> %520, i64 1, !dbg !23 + %529 = extractelement <2 x bfloat> %522, i64 0, !dbg !23 + %530 = extractelement <2 x bfloat> %522, i64 1, !dbg !23 + %531 = extractelement <32 x i1> %39, i64 11, !dbg !23 + %532 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %159, i1 %531) #7, !dbg !23 + %533 = extractvalue { i32, i32, i32, i32 } %532, 0, !dbg !23 + %534 = bitcast i32 %533 to <2 x bfloat>, !dbg !23 + %535 = extractvalue { i32, i32, i32, i32 } %532, 1, !dbg !23 + %536 = bitcast i32 %535 to <2 x bfloat>, !dbg !23 + %537 = extractvalue { i32, i32, i32, i32 } %532, 2, !dbg !23 + %538 = bitcast i32 %537 to <2 x bfloat>, !dbg !23 + %539 = extractvalue { i32, i32, i32, i32 } %532, 3, !dbg !23 + %540 = bitcast i32 %539 to <2 x bfloat>, !dbg !23 + %541 = extractelement <2 x bfloat> %534, i64 0, !dbg !23 + %542 = extractelement <2 x bfloat> %534, i64 1, !dbg !23 + %543 = extractelement <2 x bfloat> %536, i64 0, !dbg !23 + %544 = extractelement <2 x bfloat> %536, i64 1, !dbg !23 + %545 = extractelement <2 x bfloat> %538, i64 0, !dbg !23 + %546 = extractelement <2 x bfloat> %538, i64 1, !dbg !23 + %547 = extractelement <2 x bfloat> %540, i64 0, !dbg !23 + %548 = extractelement <2 x bfloat> %540, i64 1, !dbg !23 + %549 = extractelement <32 x i1> %39, i64 10, !dbg !23 + %550 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %160, i1 %549) #7, !dbg !23 + %551 = extractvalue { i32, i32, i32, i32 } %550, 0, !dbg !23 + %552 = bitcast i32 %551 to <2 x bfloat>, !dbg !23 + %553 = extractvalue { i32, i32, i32, i32 } %550, 1, !dbg !23 + %554 = bitcast i32 %553 to <2 x bfloat>, !dbg !23 + %555 = extractvalue { i32, i32, i32, i32 } %550, 2, !dbg !23 + %556 = bitcast i32 %555 to <2 x bfloat>, !dbg !23 + %557 = extractvalue { i32, i32, i32, i32 } %550, 3, !dbg !23 + %558 = bitcast i32 %557 to <2 x bfloat>, !dbg !23 + %559 = extractelement <2 x bfloat> %552, i64 0, !dbg !23 + %560 = extractelement <2 x bfloat> %552, i64 1, !dbg !23 + %561 = extractelement <2 x bfloat> %554, i64 0, !dbg !23 + %562 = extractelement <2 x bfloat> %554, i64 1, !dbg !23 + %563 = extractelement <2 x bfloat> %556, i64 0, !dbg !23 + %564 = extractelement <2 x bfloat> %556, i64 1, !dbg !23 + %565 = extractelement <2 x bfloat> %558, i64 0, !dbg !23 + %566 = extractelement <2 x bfloat> %558, i64 1, !dbg !23 + %567 = extractelement <32 x i1> %39, i64 9, !dbg !23 + %568 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %161, i1 %567) #7, !dbg !23 + %569 = extractvalue { i32, i32, i32, i32 } %568, 0, !dbg !23 + %570 = bitcast i32 %569 to <2 x bfloat>, !dbg !23 + %571 = extractvalue { i32, i32, i32, i32 } %568, 1, !dbg !23 + %572 = bitcast i32 %571 to <2 x bfloat>, !dbg !23 + %573 = extractvalue { i32, i32, i32, i32 } %568, 2, !dbg !23 + %574 = bitcast i32 %573 to <2 x bfloat>, !dbg !23 + %575 = extractvalue { i32, i32, i32, i32 } %568, 3, !dbg !23 + %576 = bitcast i32 %575 to <2 x bfloat>, !dbg !23 + %577 = extractelement <2 x bfloat> %570, i64 0, !dbg !23 + %578 = extractelement <2 x bfloat> %570, i64 1, !dbg !23 + %579 = extractelement <2 x bfloat> %572, i64 0, !dbg !23 + %580 = extractelement <2 x bfloat> %572, i64 1, !dbg !23 + %581 = extractelement <2 x bfloat> %574, i64 0, !dbg !23 + %582 = extractelement <2 x bfloat> %574, i64 1, !dbg !23 + %583 = extractelement <2 x bfloat> %576, i64 0, !dbg !23 + %584 = extractelement <2 x bfloat> %576, i64 1, !dbg !23 + %585 = extractelement <32 x i1> %39, i64 8, !dbg !23 + %586 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %162, i1 %585) #7, !dbg !23 + %587 = extractvalue { i32, i32, i32, i32 } %586, 0, !dbg !23 + %588 = bitcast i32 %587 to <2 x bfloat>, !dbg !23 + %589 = extractvalue { i32, i32, i32, i32 } %586, 1, !dbg !23 + %590 = bitcast i32 %589 to <2 x bfloat>, !dbg !23 + %591 = extractvalue { i32, i32, i32, i32 } %586, 2, !dbg !23 + %592 = bitcast i32 %591 to <2 x bfloat>, !dbg !23 + %593 = extractvalue { i32, i32, i32, i32 } %586, 3, !dbg !23 + %594 = bitcast i32 %593 to <2 x bfloat>, !dbg !23 + %595 = extractelement <2 x bfloat> %588, i64 0, !dbg !23 + %596 = extractelement <2 x bfloat> %588, i64 1, !dbg !23 + %597 = extractelement <2 x bfloat> %590, i64 0, !dbg !23 + %598 = extractelement <2 x bfloat> %590, i64 1, !dbg !23 + %599 = extractelement <2 x bfloat> %592, i64 0, !dbg !23 + %600 = extractelement <2 x bfloat> %592, i64 1, !dbg !23 + %601 = extractelement <2 x bfloat> %594, i64 0, !dbg !23 + %602 = extractelement <2 x bfloat> %594, i64 1, !dbg !23 + %603 = extractelement <32 x i1> %39, i64 7, !dbg !23 + %604 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %163, i1 %603) #7, !dbg !23 + %605 = extractvalue { i32, i32, i32, i32 } %604, 0, !dbg !23 + %606 = bitcast i32 %605 to <2 x bfloat>, !dbg !23 + %607 = extractvalue { i32, i32, i32, i32 } %604, 1, !dbg !23 + %608 = bitcast i32 %607 to <2 x bfloat>, !dbg !23 + %609 = extractvalue { i32, i32, i32, i32 } %604, 2, !dbg !23 + %610 = bitcast i32 %609 to <2 x bfloat>, !dbg !23 + %611 = extractvalue { i32, i32, i32, i32 } %604, 3, !dbg !23 + %612 = bitcast i32 %611 to <2 x bfloat>, !dbg !23 + %613 = extractelement <2 x bfloat> %606, i64 0, !dbg !23 + %614 = extractelement <2 x bfloat> %606, i64 1, !dbg !23 + %615 = extractelement <2 x bfloat> %608, i64 0, !dbg !23 + %616 = extractelement <2 x bfloat> %608, i64 1, !dbg !23 + %617 = extractelement <2 x bfloat> %610, i64 0, !dbg !23 + %618 = extractelement <2 x bfloat> %610, i64 1, !dbg !23 + %619 = extractelement <2 x bfloat> %612, i64 0, !dbg !23 + %620 = extractelement <2 x bfloat> %612, i64 1, !dbg !23 + %621 = extractelement <32 x i1> %39, i64 6, !dbg !23 + %622 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %164, i1 %621) #7, !dbg !23 + %623 = extractvalue { i32, i32, i32, i32 } %622, 0, !dbg !23 + %624 = bitcast i32 %623 to <2 x bfloat>, !dbg !23 + %625 = extractvalue { i32, i32, i32, i32 } %622, 1, !dbg !23 + %626 = bitcast i32 %625 to <2 x bfloat>, !dbg !23 + %627 = extractvalue { i32, i32, i32, i32 } %622, 2, !dbg !23 + %628 = bitcast i32 %627 to <2 x bfloat>, !dbg !23 + %629 = extractvalue { i32, i32, i32, i32 } %622, 3, !dbg !23 + %630 = bitcast i32 %629 to <2 x bfloat>, !dbg !23 + %631 = extractelement <2 x bfloat> %624, i64 0, !dbg !23 + %632 = extractelement <2 x bfloat> %624, i64 1, !dbg !23 + %633 = extractelement <2 x bfloat> %626, i64 0, !dbg !23 + %634 = extractelement <2 x bfloat> %626, i64 1, !dbg !23 + %635 = extractelement <2 x bfloat> %628, i64 0, !dbg !23 + %636 = extractelement <2 x bfloat> %628, i64 1, !dbg !23 + %637 = extractelement <2 x bfloat> %630, i64 0, !dbg !23 + %638 = extractelement <2 x bfloat> %630, i64 1, !dbg !23 + %639 = extractelement <32 x i1> %39, i64 5, !dbg !23 + %640 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %165, i1 %639) #7, !dbg !23 + %641 = extractvalue { i32, i32, i32, i32 } %640, 0, !dbg !23 + %642 = bitcast i32 %641 to <2 x bfloat>, !dbg !23 + %643 = extractvalue { i32, i32, i32, i32 } %640, 1, !dbg !23 + %644 = bitcast i32 %643 to <2 x bfloat>, !dbg !23 + %645 = extractvalue { i32, i32, i32, i32 } %640, 2, !dbg !23 + %646 = bitcast i32 %645 to <2 x bfloat>, !dbg !23 + %647 = extractvalue { i32, i32, i32, i32 } %640, 3, !dbg !23 + %648 = bitcast i32 %647 to <2 x bfloat>, !dbg !23 + %649 = extractelement <2 x bfloat> %642, i64 0, !dbg !23 + %650 = extractelement <2 x bfloat> %642, i64 1, !dbg !23 + %651 = extractelement <2 x bfloat> %644, i64 0, !dbg !23 + %652 = extractelement <2 x bfloat> %644, i64 1, !dbg !23 + %653 = extractelement <2 x bfloat> %646, i64 0, !dbg !23 + %654 = extractelement <2 x bfloat> %646, i64 1, !dbg !23 + %655 = extractelement <2 x bfloat> %648, i64 0, !dbg !23 + %656 = extractelement <2 x bfloat> %648, i64 1, !dbg !23 + %657 = extractelement <32 x i1> %39, i64 4, !dbg !23 + %658 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %166, i1 %657) #7, !dbg !23 + %659 = extractvalue { i32, i32, i32, i32 } %658, 0, !dbg !23 + %660 = bitcast i32 %659 to <2 x bfloat>, !dbg !23 + %661 = extractvalue { i32, i32, i32, i32 } %658, 1, !dbg !23 + %662 = bitcast i32 %661 to <2 x bfloat>, !dbg !23 + %663 = extractvalue { i32, i32, i32, i32 } %658, 2, !dbg !23 + %664 = bitcast i32 %663 to <2 x bfloat>, !dbg !23 + %665 = extractvalue { i32, i32, i32, i32 } %658, 3, !dbg !23 + %666 = bitcast i32 %665 to <2 x bfloat>, !dbg !23 + %667 = extractelement <2 x bfloat> %660, i64 0, !dbg !23 + %668 = extractelement <2 x bfloat> %660, i64 1, !dbg !23 + %669 = extractelement <2 x bfloat> %662, i64 0, !dbg !23 + %670 = extractelement <2 x bfloat> %662, i64 1, !dbg !23 + %671 = extractelement <2 x bfloat> %664, i64 0, !dbg !23 + %672 = extractelement <2 x bfloat> %664, i64 1, !dbg !23 + %673 = extractelement <2 x bfloat> %666, i64 0, !dbg !23 + %674 = extractelement <2 x bfloat> %666, i64 1, !dbg !23 + %675 = extractelement <32 x i1> %39, i64 3, !dbg !23 + %676 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %167, i1 %675) #7, !dbg !23 + %677 = extractvalue { i32, i32, i32, i32 } %676, 0, !dbg !23 + %678 = bitcast i32 %677 to <2 x bfloat>, !dbg !23 + %679 = extractvalue { i32, i32, i32, i32 } %676, 1, !dbg !23 + %680 = bitcast i32 %679 to <2 x bfloat>, !dbg !23 + %681 = extractvalue { i32, i32, i32, i32 } %676, 2, !dbg !23 + %682 = bitcast i32 %681 to <2 x bfloat>, !dbg !23 + %683 = extractvalue { i32, i32, i32, i32 } %676, 3, !dbg !23 + %684 = bitcast i32 %683 to <2 x bfloat>, !dbg !23 + %685 = extractelement <2 x bfloat> %678, i64 0, !dbg !23 + %686 = extractelement <2 x bfloat> %678, i64 1, !dbg !23 + %687 = extractelement <2 x bfloat> %680, i64 0, !dbg !23 + %688 = extractelement <2 x bfloat> %680, i64 1, !dbg !23 + %689 = extractelement <2 x bfloat> %682, i64 0, !dbg !23 + %690 = extractelement <2 x bfloat> %682, i64 1, !dbg !23 + %691 = extractelement <2 x bfloat> %684, i64 0, !dbg !23 + %692 = extractelement <2 x bfloat> %684, i64 1, !dbg !23 + %693 = extractelement <32 x i1> %39, i64 2, !dbg !23 + %694 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %168, i1 %693) #7, !dbg !23 + %695 = extractvalue { i32, i32, i32, i32 } %694, 0, !dbg !23 + %696 = bitcast i32 %695 to <2 x bfloat>, !dbg !23 + %697 = extractvalue { i32, i32, i32, i32 } %694, 1, !dbg !23 + %698 = bitcast i32 %697 to <2 x bfloat>, !dbg !23 + %699 = extractvalue { i32, i32, i32, i32 } %694, 2, !dbg !23 + %700 = bitcast i32 %699 to <2 x bfloat>, !dbg !23 + %701 = extractvalue { i32, i32, i32, i32 } %694, 3, !dbg !23 + %702 = bitcast i32 %701 to <2 x bfloat>, !dbg !23 + %703 = extractelement <2 x bfloat> %696, i64 0, !dbg !23 + %704 = extractelement <2 x bfloat> %696, i64 1, !dbg !23 + %705 = extractelement <2 x bfloat> %698, i64 0, !dbg !23 + %706 = extractelement <2 x bfloat> %698, i64 1, !dbg !23 + %707 = extractelement <2 x bfloat> %700, i64 0, !dbg !23 + %708 = extractelement <2 x bfloat> %700, i64 1, !dbg !23 + %709 = extractelement <2 x bfloat> %702, i64 0, !dbg !23 + %710 = extractelement <2 x bfloat> %702, i64 1, !dbg !23 + %711 = extractelement <32 x i1> %39, i64 1, !dbg !23 + %712 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %169, i1 %711) #7, !dbg !23 + %713 = extractvalue { i32, i32, i32, i32 } %712, 0, !dbg !23 + %714 = bitcast i32 %713 to <2 x bfloat>, !dbg !23 + %715 = extractvalue { i32, i32, i32, i32 } %712, 1, !dbg !23 + %716 = bitcast i32 %715 to <2 x bfloat>, !dbg !23 + %717 = extractvalue { i32, i32, i32, i32 } %712, 2, !dbg !23 + %718 = bitcast i32 %717 to <2 x bfloat>, !dbg !23 + %719 = extractvalue { i32, i32, i32, i32 } %712, 3, !dbg !23 + %720 = bitcast i32 %719 to <2 x bfloat>, !dbg !23 + %721 = extractelement <2 x bfloat> %714, i64 0, !dbg !23 + %722 = extractelement <2 x bfloat> %714, i64 1, !dbg !23 + %723 = extractelement <2 x bfloat> %716, i64 0, !dbg !23 + %724 = extractelement <2 x bfloat> %716, i64 1, !dbg !23 + %725 = extractelement <2 x bfloat> %718, i64 0, !dbg !23 + %726 = extractelement <2 x bfloat> %718, i64 1, !dbg !23 + %727 = extractelement <2 x bfloat> %720, i64 0, !dbg !23 + %728 = extractelement <2 x bfloat> %720, i64 1, !dbg !23 + %729 = extractelement <32 x i1> %39, i64 0, !dbg !23 + %730 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %170, i1 %729) #7, !dbg !23 + %731 = extractvalue { i32, i32, i32, i32 } %730, 0, !dbg !23 + %732 = bitcast i32 %731 to <2 x bfloat>, !dbg !23 + %733 = extractvalue { i32, i32, i32, i32 } %730, 1, !dbg !23 + %734 = bitcast i32 %733 to <2 x bfloat>, !dbg !23 + %735 = extractvalue { i32, i32, i32, i32 } %730, 2, !dbg !23 + %736 = bitcast i32 %735 to <2 x bfloat>, !dbg !23 + %737 = extractvalue { i32, i32, i32, i32 } %730, 3, !dbg !23 + %738 = bitcast i32 %737 to <2 x bfloat>, !dbg !23 + %739 = extractelement <2 x bfloat> %732, i64 0, !dbg !23 + %740 = extractelement <2 x bfloat> %732, i64 1, !dbg !23 + %741 = extractelement <2 x bfloat> %734, i64 0, !dbg !23 + %742 = extractelement <2 x bfloat> %734, i64 1, !dbg !23 + %743 = extractelement <2 x bfloat> %736, i64 0, !dbg !23 + %744 = extractelement <2 x bfloat> %736, i64 1, !dbg !23 + %745 = extractelement <2 x bfloat> %738, i64 0, !dbg !23 + %746 = extractelement <2 x bfloat> %738, i64 1, !dbg !23 + %747 = fpext bfloat %181 to float, !dbg !24 + %748 = fpext bfloat %182 to float, !dbg !24 + %749 = fpext bfloat %183 to float, !dbg !24 + %750 = fpext bfloat %184 to float, !dbg !24 + %751 = fpext bfloat %185 to float, !dbg !24 + %752 = fpext bfloat %186 to float, !dbg !24 + %753 = fpext bfloat %187 to float, !dbg !24 + %754 = fpext bfloat %188 to float, !dbg !24 + %755 = fpext bfloat %199 to float, !dbg !24 + %756 = fpext bfloat %200 to float, !dbg !24 + %757 = fpext bfloat %201 to float, !dbg !24 + %758 = fpext bfloat %202 to float, !dbg !24 + %759 = fpext bfloat %203 to float, !dbg !24 + %760 = fpext bfloat %204 to float, !dbg !24 + %761 = fpext bfloat %205 to float, !dbg !24 + %762 = fpext bfloat %206 to float, !dbg !24 + %763 = fpext bfloat %217 to float, !dbg !24 + %764 = fpext bfloat %218 to float, !dbg !24 + %765 = fpext bfloat %219 to float, !dbg !24 + %766 = fpext bfloat %220 to float, !dbg !24 + %767 = fpext bfloat %221 to float, !dbg !24 + %768 = fpext bfloat %222 to float, !dbg !24 + %769 = fpext bfloat %223 to float, !dbg !24 + %770 = fpext bfloat %224 to float, !dbg !24 + %771 = fpext bfloat %235 to float, !dbg !24 + %772 = fpext bfloat %236 to float, !dbg !24 + %773 = fpext bfloat %237 to float, !dbg !24 + %774 = fpext bfloat %238 to float, !dbg !24 + %775 = fpext bfloat %239 to float, !dbg !24 + %776 = fpext bfloat %240 to float, !dbg !24 + %777 = fpext bfloat %241 to float, !dbg !24 + %778 = fpext bfloat %242 to float, !dbg !24 + %779 = fpext bfloat %253 to float, !dbg !24 + %780 = fpext bfloat %254 to float, !dbg !24 + %781 = fpext bfloat %255 to float, !dbg !24 + %782 = fpext bfloat %256 to float, !dbg !24 + %783 = fpext bfloat %257 to float, !dbg !24 + %784 = fpext bfloat %258 to float, !dbg !24 + %785 = fpext bfloat %259 to float, !dbg !24 + %786 = fpext bfloat %260 to float, !dbg !24 + %787 = fpext bfloat %271 to float, !dbg !24 + %788 = fpext bfloat %272 to float, !dbg !24 + %789 = fpext bfloat %273 to float, !dbg !24 + %790 = fpext bfloat %274 to float, !dbg !24 + %791 = fpext bfloat %275 to float, !dbg !24 + %792 = fpext bfloat %276 to float, !dbg !24 + %793 = fpext bfloat %277 to float, !dbg !24 + %794 = fpext bfloat %278 to float, !dbg !24 + %795 = fpext bfloat %289 to float, !dbg !24 + %796 = fpext bfloat %290 to float, !dbg !24 + %797 = fpext bfloat %291 to float, !dbg !24 + %798 = fpext bfloat %292 to float, !dbg !24 + %799 = fpext bfloat %293 to float, !dbg !24 + %800 = fpext bfloat %294 to float, !dbg !24 + %801 = fpext bfloat %295 to float, !dbg !24 + %802 = fpext bfloat %296 to float, !dbg !24 + %803 = fpext bfloat %307 to float, !dbg !24 + %804 = fpext bfloat %308 to float, !dbg !24 + %805 = fpext bfloat %309 to float, !dbg !24 + %806 = fpext bfloat %310 to float, !dbg !24 + %807 = fpext bfloat %311 to float, !dbg !24 + %808 = fpext bfloat %312 to float, !dbg !24 + %809 = fpext bfloat %313 to float, !dbg !24 + %810 = fpext bfloat %314 to float, !dbg !24 + %811 = fpext bfloat %325 to float, !dbg !24 + %812 = fpext bfloat %326 to float, !dbg !24 + %813 = fpext bfloat %327 to float, !dbg !24 + %814 = fpext bfloat %328 to float, !dbg !24 + %815 = fpext bfloat %329 to float, !dbg !24 + %816 = fpext bfloat %330 to float, !dbg !24 + %817 = fpext bfloat %331 to float, !dbg !24 + %818 = fpext bfloat %332 to float, !dbg !24 + %819 = fpext bfloat %343 to float, !dbg !24 + %820 = fpext bfloat %344 to float, !dbg !24 + %821 = fpext bfloat %345 to float, !dbg !24 + %822 = fpext bfloat %346 to float, !dbg !24 + %823 = fpext bfloat %347 to float, !dbg !24 + %824 = fpext bfloat %348 to float, !dbg !24 + %825 = fpext bfloat %349 to float, !dbg !24 + %826 = fpext bfloat %350 to float, !dbg !24 + %827 = fpext bfloat %361 to float, !dbg !24 + %828 = fpext bfloat %362 to float, !dbg !24 + %829 = fpext bfloat %363 to float, !dbg !24 + %830 = fpext bfloat %364 to float, !dbg !24 + %831 = fpext bfloat %365 to float, !dbg !24 + %832 = fpext bfloat %366 to float, !dbg !24 + %833 = fpext bfloat %367 to float, !dbg !24 + %834 = fpext bfloat %368 to float, !dbg !24 + %835 = fpext bfloat %379 to float, !dbg !24 + %836 = fpext bfloat %380 to float, !dbg !24 + %837 = fpext bfloat %381 to float, !dbg !24 + %838 = fpext bfloat %382 to float, !dbg !24 + %839 = fpext bfloat %383 to float, !dbg !24 + %840 = fpext bfloat %384 to float, !dbg !24 + %841 = fpext bfloat %385 to float, !dbg !24 + %842 = fpext bfloat %386 to float, !dbg !24 + %843 = fpext bfloat %397 to float, !dbg !24 + %844 = fpext bfloat %398 to float, !dbg !24 + %845 = fpext bfloat %399 to float, !dbg !24 + %846 = fpext bfloat %400 to float, !dbg !24 + %847 = fpext bfloat %401 to float, !dbg !24 + %848 = fpext bfloat %402 to float, !dbg !24 + %849 = fpext bfloat %403 to float, !dbg !24 + %850 = fpext bfloat %404 to float, !dbg !24 + %851 = fpext bfloat %415 to float, !dbg !24 + %852 = fpext bfloat %416 to float, !dbg !24 + %853 = fpext bfloat %417 to float, !dbg !24 + %854 = fpext bfloat %418 to float, !dbg !24 + %855 = fpext bfloat %419 to float, !dbg !24 + %856 = fpext bfloat %420 to float, !dbg !24 + %857 = fpext bfloat %421 to float, !dbg !24 + %858 = fpext bfloat %422 to float, !dbg !24 + %859 = fpext bfloat %433 to float, !dbg !24 + %860 = fpext bfloat %434 to float, !dbg !24 + %861 = fpext bfloat %435 to float, !dbg !24 + %862 = fpext bfloat %436 to float, !dbg !24 + %863 = fpext bfloat %437 to float, !dbg !24 + %864 = fpext bfloat %438 to float, !dbg !24 + %865 = fpext bfloat %439 to float, !dbg !24 + %866 = fpext bfloat %440 to float, !dbg !24 + %867 = fpext bfloat %451 to float, !dbg !24 + %868 = fpext bfloat %452 to float, !dbg !24 + %869 = fpext bfloat %453 to float, !dbg !24 + %870 = fpext bfloat %454 to float, !dbg !24 + %871 = fpext bfloat %455 to float, !dbg !24 + %872 = fpext bfloat %456 to float, !dbg !24 + %873 = fpext bfloat %457 to float, !dbg !24 + %874 = fpext bfloat %458 to float, !dbg !24 + %875 = fpext bfloat %469 to float, !dbg !24 + %876 = fpext bfloat %470 to float, !dbg !24 + %877 = fpext bfloat %471 to float, !dbg !24 + %878 = fpext bfloat %472 to float, !dbg !24 + %879 = fpext bfloat %473 to float, !dbg !24 + %880 = fpext bfloat %474 to float, !dbg !24 + %881 = fpext bfloat %475 to float, !dbg !24 + %882 = fpext bfloat %476 to float, !dbg !24 + %883 = fpext bfloat %487 to float, !dbg !24 + %884 = fpext bfloat %488 to float, !dbg !24 + %885 = fpext bfloat %489 to float, !dbg !24 + %886 = fpext bfloat %490 to float, !dbg !24 + %887 = fpext bfloat %491 to float, !dbg !24 + %888 = fpext bfloat %492 to float, !dbg !24 + %889 = fpext bfloat %493 to float, !dbg !24 + %890 = fpext bfloat %494 to float, !dbg !24 + %891 = fpext bfloat %505 to float, !dbg !24 + %892 = fpext bfloat %506 to float, !dbg !24 + %893 = fpext bfloat %507 to float, !dbg !24 + %894 = fpext bfloat %508 to float, !dbg !24 + %895 = fpext bfloat %509 to float, !dbg !24 + %896 = fpext bfloat %510 to float, !dbg !24 + %897 = fpext bfloat %511 to float, !dbg !24 + %898 = fpext bfloat %512 to float, !dbg !24 + %899 = fpext bfloat %523 to float, !dbg !24 + %900 = fpext bfloat %524 to float, !dbg !24 + %901 = fpext bfloat %525 to float, !dbg !24 + %902 = fpext bfloat %526 to float, !dbg !24 + %903 = fpext bfloat %527 to float, !dbg !24 + %904 = fpext bfloat %528 to float, !dbg !24 + %905 = fpext bfloat %529 to float, !dbg !24 + %906 = fpext bfloat %530 to float, !dbg !24 + %907 = fpext bfloat %541 to float, !dbg !24 + %908 = fpext bfloat %542 to float, !dbg !24 + %909 = fpext bfloat %543 to float, !dbg !24 + %910 = fpext bfloat %544 to float, !dbg !24 + %911 = fpext bfloat %545 to float, !dbg !24 + %912 = fpext bfloat %546 to float, !dbg !24 + %913 = fpext bfloat %547 to float, !dbg !24 + %914 = fpext bfloat %548 to float, !dbg !24 + %915 = fpext bfloat %559 to float, !dbg !24 + %916 = fpext bfloat %560 to float, !dbg !24 + %917 = fpext bfloat %561 to float, !dbg !24 + %918 = fpext bfloat %562 to float, !dbg !24 + %919 = fpext bfloat %563 to float, !dbg !24 + %920 = fpext bfloat %564 to float, !dbg !24 + %921 = fpext bfloat %565 to float, !dbg !24 + %922 = fpext bfloat %566 to float, !dbg !24 + %923 = fpext bfloat %577 to float, !dbg !24 + %924 = fpext bfloat %578 to float, !dbg !24 + %925 = fpext bfloat %579 to float, !dbg !24 + %926 = fpext bfloat %580 to float, !dbg !24 + %927 = fpext bfloat %581 to float, !dbg !24 + %928 = fpext bfloat %582 to float, !dbg !24 + %929 = fpext bfloat %583 to float, !dbg !24 + %930 = fpext bfloat %584 to float, !dbg !24 + %931 = fpext bfloat %595 to float, !dbg !24 + %932 = fpext bfloat %596 to float, !dbg !24 + %933 = fpext bfloat %597 to float, !dbg !24 + %934 = fpext bfloat %598 to float, !dbg !24 + %935 = fpext bfloat %599 to float, !dbg !24 + %936 = fpext bfloat %600 to float, !dbg !24 + %937 = fpext bfloat %601 to float, !dbg !24 + %938 = fpext bfloat %602 to float, !dbg !24 + %939 = fpext bfloat %613 to float, !dbg !24 + %940 = fpext bfloat %614 to float, !dbg !24 + %941 = fpext bfloat %615 to float, !dbg !24 + %942 = fpext bfloat %616 to float, !dbg !24 + %943 = fpext bfloat %617 to float, !dbg !24 + %944 = fpext bfloat %618 to float, !dbg !24 + %945 = fpext bfloat %619 to float, !dbg !24 + %946 = fpext bfloat %620 to float, !dbg !24 + %947 = fpext bfloat %631 to float, !dbg !24 + %948 = fpext bfloat %632 to float, !dbg !24 + %949 = fpext bfloat %633 to float, !dbg !24 + %950 = fpext bfloat %634 to float, !dbg !24 + %951 = fpext bfloat %635 to float, !dbg !24 + %952 = fpext bfloat %636 to float, !dbg !24 + %953 = fpext bfloat %637 to float, !dbg !24 + %954 = fpext bfloat %638 to float, !dbg !24 + %955 = fpext bfloat %649 to float, !dbg !24 + %956 = fpext bfloat %650 to float, !dbg !24 + %957 = fpext bfloat %651 to float, !dbg !24 + %958 = fpext bfloat %652 to float, !dbg !24 + %959 = fpext bfloat %653 to float, !dbg !24 + %960 = fpext bfloat %654 to float, !dbg !24 + %961 = fpext bfloat %655 to float, !dbg !24 + %962 = fpext bfloat %656 to float, !dbg !24 + %963 = fpext bfloat %667 to float, !dbg !24 + %964 = fpext bfloat %668 to float, !dbg !24 + %965 = fpext bfloat %669 to float, !dbg !24 + %966 = fpext bfloat %670 to float, !dbg !24 + %967 = fpext bfloat %671 to float, !dbg !24 + %968 = fpext bfloat %672 to float, !dbg !24 + %969 = fpext bfloat %673 to float, !dbg !24 + %970 = fpext bfloat %674 to float, !dbg !24 + %971 = fpext bfloat %685 to float, !dbg !24 + %972 = fpext bfloat %686 to float, !dbg !24 + %973 = fpext bfloat %687 to float, !dbg !24 + %974 = fpext bfloat %688 to float, !dbg !24 + %975 = fpext bfloat %689 to float, !dbg !24 + %976 = fpext bfloat %690 to float, !dbg !24 + %977 = fpext bfloat %691 to float, !dbg !24 + %978 = fpext bfloat %692 to float, !dbg !24 + %979 = fpext bfloat %703 to float, !dbg !24 + %980 = fpext bfloat %704 to float, !dbg !24 + %981 = fpext bfloat %705 to float, !dbg !24 + %982 = fpext bfloat %706 to float, !dbg !24 + %983 = fpext bfloat %707 to float, !dbg !24 + %984 = fpext bfloat %708 to float, !dbg !24 + %985 = fpext bfloat %709 to float, !dbg !24 + %986 = fpext bfloat %710 to float, !dbg !24 + %987 = fpext bfloat %721 to float, !dbg !24 + %988 = fpext bfloat %722 to float, !dbg !24 + %989 = fpext bfloat %723 to float, !dbg !24 + %990 = fpext bfloat %724 to float, !dbg !24 + %991 = fpext bfloat %725 to float, !dbg !24 + %992 = fpext bfloat %726 to float, !dbg !24 + %993 = fpext bfloat %727 to float, !dbg !24 + %994 = fpext bfloat %728 to float, !dbg !24 + %995 = fpext bfloat %739 to float, !dbg !24 + %996 = fpext bfloat %740 to float, !dbg !24 + %997 = fpext bfloat %741 to float, !dbg !24 + %998 = fpext bfloat %742 to float, !dbg !24 + %999 = fpext bfloat %743 to float, !dbg !24 + %1000 = fpext bfloat %744 to float, !dbg !24 + %1001 = fpext bfloat %745 to float, !dbg !24 + %1002 = fpext bfloat %746 to float, !dbg !24 + %1003 = fmul float %747, %747, !dbg !25 + %1004 = fmul float %748, %748, !dbg !25 + %1005 = fmul float %749, %749, !dbg !25 + %1006 = fmul float %750, %750, !dbg !25 + %1007 = fmul float %751, %751, !dbg !25 + %1008 = fmul float %752, %752, !dbg !25 + %1009 = fmul float %753, %753, !dbg !25 + %1010 = fmul float %754, %754, !dbg !25 + %1011 = fmul float %755, %755, !dbg !25 + %1012 = fmul float %756, %756, !dbg !25 + %1013 = fmul float %757, %757, !dbg !25 + %1014 = fmul float %758, %758, !dbg !25 + %1015 = fmul float %759, %759, !dbg !25 + %1016 = fmul float %760, %760, !dbg !25 + %1017 = fmul float %761, %761, !dbg !25 + %1018 = fmul float %762, %762, !dbg !25 + %1019 = fmul float %763, %763, !dbg !25 + %1020 = fmul float %764, %764, !dbg !25 + %1021 = fmul float %765, %765, !dbg !25 + %1022 = fmul float %766, %766, !dbg !25 + %1023 = fmul float %767, %767, !dbg !25 + %1024 = fmul float %768, %768, !dbg !25 + %1025 = fmul float %769, %769, !dbg !25 + %1026 = fmul float %770, %770, !dbg !25 + %1027 = fmul float %771, %771, !dbg !25 + %1028 = fmul float %772, %772, !dbg !25 + %1029 = fmul float %773, %773, !dbg !25 + %1030 = fmul float %774, %774, !dbg !25 + %1031 = fmul float %775, %775, !dbg !25 + %1032 = fmul float %776, %776, !dbg !25 + %1033 = fmul float %777, %777, !dbg !25 + %1034 = fmul float %778, %778, !dbg !25 + %1035 = fmul float %779, %779, !dbg !25 + %1036 = fmul float %780, %780, !dbg !25 + %1037 = fmul float %781, %781, !dbg !25 + %1038 = fmul float %782, %782, !dbg !25 + %1039 = fmul float %783, %783, !dbg !25 + %1040 = fmul float %784, %784, !dbg !25 + %1041 = fmul float %785, %785, !dbg !25 + %1042 = fmul float %786, %786, !dbg !25 + %1043 = fmul float %787, %787, !dbg !25 + %1044 = fmul float %788, %788, !dbg !25 + %1045 = fmul float %789, %789, !dbg !25 + %1046 = fmul float %790, %790, !dbg !25 + %1047 = fmul float %791, %791, !dbg !25 + %1048 = fmul float %792, %792, !dbg !25 + %1049 = fmul float %793, %793, !dbg !25 + %1050 = fmul float %794, %794, !dbg !25 + %1051 = fmul float %795, %795, !dbg !25 + %1052 = fmul float %796, %796, !dbg !25 + %1053 = fmul float %797, %797, !dbg !25 + %1054 = fmul float %798, %798, !dbg !25 + %1055 = fmul float %799, %799, !dbg !25 + %1056 = fmul float %800, %800, !dbg !25 + %1057 = fmul float %801, %801, !dbg !25 + %1058 = fmul float %802, %802, !dbg !25 + %1059 = fmul float %803, %803, !dbg !25 + %1060 = fmul float %804, %804, !dbg !25 + %1061 = fmul float %805, %805, !dbg !25 + %1062 = fmul float %806, %806, !dbg !25 + %1063 = fmul float %807, %807, !dbg !25 + %1064 = fmul float %808, %808, !dbg !25 + %1065 = fmul float %809, %809, !dbg !25 + %1066 = fmul float %810, %810, !dbg !25 + %1067 = fmul float %811, %811, !dbg !25 + %1068 = fmul float %812, %812, !dbg !25 + %1069 = fmul float %813, %813, !dbg !25 + %1070 = fmul float %814, %814, !dbg !25 + %1071 = fmul float %815, %815, !dbg !25 + %1072 = fmul float %816, %816, !dbg !25 + %1073 = fmul float %817, %817, !dbg !25 + %1074 = fmul float %818, %818, !dbg !25 + %1075 = fmul float %819, %819, !dbg !25 + %1076 = fmul float %820, %820, !dbg !25 + %1077 = fmul float %821, %821, !dbg !25 + %1078 = fmul float %822, %822, !dbg !25 + %1079 = fmul float %823, %823, !dbg !25 + %1080 = fmul float %824, %824, !dbg !25 + %1081 = fmul float %825, %825, !dbg !25 + %1082 = fmul float %826, %826, !dbg !25 + %1083 = fmul float %827, %827, !dbg !25 + %1084 = fmul float %828, %828, !dbg !25 + %1085 = fmul float %829, %829, !dbg !25 + %1086 = fmul float %830, %830, !dbg !25 + %1087 = fmul float %831, %831, !dbg !25 + %1088 = fmul float %832, %832, !dbg !25 + %1089 = fmul float %833, %833, !dbg !25 + %1090 = fmul float %834, %834, !dbg !25 + %1091 = fmul float %835, %835, !dbg !25 + %1092 = fmul float %836, %836, !dbg !25 + %1093 = fmul float %837, %837, !dbg !25 + %1094 = fmul float %838, %838, !dbg !25 + %1095 = fmul float %839, %839, !dbg !25 + %1096 = fmul float %840, %840, !dbg !25 + %1097 = fmul float %841, %841, !dbg !25 + %1098 = fmul float %842, %842, !dbg !25 + %1099 = fmul float %843, %843, !dbg !25 + %1100 = fmul float %844, %844, !dbg !25 + %1101 = fmul float %845, %845, !dbg !25 + %1102 = fmul float %846, %846, !dbg !25 + %1103 = fmul float %847, %847, !dbg !25 + %1104 = fmul float %848, %848, !dbg !25 + %1105 = fmul float %849, %849, !dbg !25 + %1106 = fmul float %850, %850, !dbg !25 + %1107 = fmul float %851, %851, !dbg !25 + %1108 = fmul float %852, %852, !dbg !25 + %1109 = fmul float %853, %853, !dbg !25 + %1110 = fmul float %854, %854, !dbg !25 + %1111 = fmul float %855, %855, !dbg !25 + %1112 = fmul float %856, %856, !dbg !25 + %1113 = fmul float %857, %857, !dbg !25 + %1114 = fmul float %858, %858, !dbg !25 + %1115 = fmul float %859, %859, !dbg !25 + %1116 = fmul float %860, %860, !dbg !25 + %1117 = fmul float %861, %861, !dbg !25 + %1118 = fmul float %862, %862, !dbg !25 + %1119 = fmul float %863, %863, !dbg !25 + %1120 = fmul float %864, %864, !dbg !25 + %1121 = fmul float %865, %865, !dbg !25 + %1122 = fmul float %866, %866, !dbg !25 + %1123 = fmul float %867, %867, !dbg !25 + %1124 = fmul float %868, %868, !dbg !25 + %1125 = fmul float %869, %869, !dbg !25 + %1126 = fmul float %870, %870, !dbg !25 + %1127 = fmul float %871, %871, !dbg !25 + %1128 = fmul float %872, %872, !dbg !25 + %1129 = fmul float %873, %873, !dbg !25 + %1130 = fmul float %874, %874, !dbg !25 + %1131 = fmul float %875, %875, !dbg !25 + %1132 = fmul float %876, %876, !dbg !25 + %1133 = fmul float %877, %877, !dbg !25 + %1134 = fmul float %878, %878, !dbg !25 + %1135 = fmul float %879, %879, !dbg !25 + %1136 = fmul float %880, %880, !dbg !25 + %1137 = fmul float %881, %881, !dbg !25 + %1138 = fmul float %882, %882, !dbg !25 + %1139 = fmul float %883, %883, !dbg !25 + %1140 = fmul float %884, %884, !dbg !25 + %1141 = fmul float %885, %885, !dbg !25 + %1142 = fmul float %886, %886, !dbg !25 + %1143 = fmul float %887, %887, !dbg !25 + %1144 = fmul float %888, %888, !dbg !25 + %1145 = fmul float %889, %889, !dbg !25 + %1146 = fmul float %890, %890, !dbg !25 + %1147 = fmul float %891, %891, !dbg !25 + %1148 = fmul float %892, %892, !dbg !25 + %1149 = fmul float %893, %893, !dbg !25 + %1150 = fmul float %894, %894, !dbg !25 + %1151 = fmul float %895, %895, !dbg !25 + %1152 = fmul float %896, %896, !dbg !25 + %1153 = fmul float %897, %897, !dbg !25 + %1154 = fmul float %898, %898, !dbg !25 + %1155 = fmul float %899, %899, !dbg !25 + %1156 = fmul float %900, %900, !dbg !25 + %1157 = fmul float %901, %901, !dbg !25 + %1158 = fmul float %902, %902, !dbg !25 + %1159 = fmul float %903, %903, !dbg !25 + %1160 = fmul float %904, %904, !dbg !25 + %1161 = fmul float %905, %905, !dbg !25 + %1162 = fmul float %906, %906, !dbg !25 + %1163 = fmul float %907, %907, !dbg !25 + %1164 = fmul float %908, %908, !dbg !25 + %1165 = fmul float %909, %909, !dbg !25 + %1166 = fmul float %910, %910, !dbg !25 + %1167 = fmul float %911, %911, !dbg !25 + %1168 = fmul float %912, %912, !dbg !25 + %1169 = fmul float %913, %913, !dbg !25 + %1170 = fmul float %914, %914, !dbg !25 + %1171 = fmul float %915, %915, !dbg !25 + %1172 = fmul float %916, %916, !dbg !25 + %1173 = fmul float %917, %917, !dbg !25 + %1174 = fmul float %918, %918, !dbg !25 + %1175 = fmul float %919, %919, !dbg !25 + %1176 = fmul float %920, %920, !dbg !25 + %1177 = fmul float %921, %921, !dbg !25 + %1178 = fmul float %922, %922, !dbg !25 + %1179 = fmul float %923, %923, !dbg !25 + %1180 = fmul float %924, %924, !dbg !25 + %1181 = fmul float %925, %925, !dbg !25 + %1182 = fmul float %926, %926, !dbg !25 + %1183 = fmul float %927, %927, !dbg !25 + %1184 = fmul float %928, %928, !dbg !25 + %1185 = fmul float %929, %929, !dbg !25 + %1186 = fmul float %930, %930, !dbg !25 + %1187 = fmul float %931, %931, !dbg !25 + %1188 = fmul float %932, %932, !dbg !25 + %1189 = fmul float %933, %933, !dbg !25 + %1190 = fmul float %934, %934, !dbg !25 + %1191 = fmul float %935, %935, !dbg !25 + %1192 = fmul float %936, %936, !dbg !25 + %1193 = fmul float %937, %937, !dbg !25 + %1194 = fmul float %938, %938, !dbg !25 + %1195 = fmul float %939, %939, !dbg !25 + %1196 = fmul float %940, %940, !dbg !25 + %1197 = fmul float %941, %941, !dbg !25 + %1198 = fmul float %942, %942, !dbg !25 + %1199 = fmul float %943, %943, !dbg !25 + %1200 = fmul float %944, %944, !dbg !25 + %1201 = fmul float %945, %945, !dbg !25 + %1202 = fmul float %946, %946, !dbg !25 + %1203 = fmul float %947, %947, !dbg !25 + %1204 = fmul float %948, %948, !dbg !25 + %1205 = fmul float %949, %949, !dbg !25 + %1206 = fmul float %950, %950, !dbg !25 + %1207 = fmul float %951, %951, !dbg !25 + %1208 = fmul float %952, %952, !dbg !25 + %1209 = fmul float %953, %953, !dbg !25 + %1210 = fmul float %954, %954, !dbg !25 + %1211 = fmul float %955, %955, !dbg !25 + %1212 = fmul float %956, %956, !dbg !25 + %1213 = fmul float %957, %957, !dbg !25 + %1214 = fmul float %958, %958, !dbg !25 + %1215 = fmul float %959, %959, !dbg !25 + %1216 = fmul float %960, %960, !dbg !25 + %1217 = fmul float %961, %961, !dbg !25 + %1218 = fmul float %962, %962, !dbg !25 + %1219 = fmul float %963, %963, !dbg !25 + %1220 = fmul float %964, %964, !dbg !25 + %1221 = fmul float %965, %965, !dbg !25 + %1222 = fmul float %966, %966, !dbg !25 + %1223 = fmul float %967, %967, !dbg !25 + %1224 = fmul float %968, %968, !dbg !25 + %1225 = fmul float %969, %969, !dbg !25 + %1226 = fmul float %970, %970, !dbg !25 + %1227 = fmul float %971, %971, !dbg !25 + %1228 = fmul float %972, %972, !dbg !25 + %1229 = fmul float %973, %973, !dbg !25 + %1230 = fmul float %974, %974, !dbg !25 + %1231 = fmul float %975, %975, !dbg !25 + %1232 = fmul float %976, %976, !dbg !25 + %1233 = fmul float %977, %977, !dbg !25 + %1234 = fmul float %978, %978, !dbg !25 + %1235 = fmul float %979, %979, !dbg !25 + %1236 = fmul float %980, %980, !dbg !25 + %1237 = fmul float %981, %981, !dbg !25 + %1238 = fmul float %982, %982, !dbg !25 + %1239 = fmul float %983, %983, !dbg !25 + %1240 = fmul float %984, %984, !dbg !25 + %1241 = fmul float %985, %985, !dbg !25 + %1242 = fmul float %986, %986, !dbg !25 + %1243 = fmul float %987, %987, !dbg !25 + %1244 = fmul float %988, %988, !dbg !25 + %1245 = fmul float %989, %989, !dbg !25 + %1246 = fmul float %990, %990, !dbg !25 + %1247 = fmul float %991, %991, !dbg !25 + %1248 = fmul float %992, %992, !dbg !25 + %1249 = fmul float %993, %993, !dbg !25 + %1250 = fmul float %994, %994, !dbg !25 + %1251 = fmul float %995, %995, !dbg !25 + %1252 = fmul float %996, %996, !dbg !25 + %1253 = fmul float %997, %997, !dbg !25 + %1254 = fmul float %998, %998, !dbg !25 + %1255 = fmul float %999, %999, !dbg !25 + %1256 = fmul float %1000, %1000, !dbg !25 + %1257 = fmul float %1001, %1001, !dbg !25 + %1258 = fmul float %1002, %1002, !dbg !25 + %1259 = fadd float %1003, %1004, !dbg !26 + %1260 = fadd float %1005, %1259, !dbg !26 + %1261 = fadd float %1006, %1260, !dbg !26 + %1262 = fadd float %1007, %1261, !dbg !26 + %1263 = fadd float %1008, %1262, !dbg !26 + %1264 = fadd float %1009, %1263, !dbg !26 + %1265 = fadd float %1010, %1264, !dbg !26 + %1266 = fadd float %1011, %1012, !dbg !26 + %1267 = fadd float %1013, %1266, !dbg !26 + %1268 = fadd float %1014, %1267, !dbg !26 + %1269 = fadd float %1015, %1268, !dbg !26 + %1270 = fadd float %1016, %1269, !dbg !26 + %1271 = fadd float %1017, %1270, !dbg !26 + %1272 = fadd float %1018, %1271, !dbg !26 + %1273 = fadd float %1019, %1020, !dbg !26 + %1274 = fadd float %1021, %1273, !dbg !26 + %1275 = fadd float %1022, %1274, !dbg !26 + %1276 = fadd float %1023, %1275, !dbg !26 + %1277 = fadd float %1024, %1276, !dbg !26 + %1278 = fadd float %1025, %1277, !dbg !26 + %1279 = fadd float %1026, %1278, !dbg !26 + %1280 = fadd float %1027, %1028, !dbg !26 + %1281 = fadd float %1029, %1280, !dbg !26 + %1282 = fadd float %1030, %1281, !dbg !26 + %1283 = fadd float %1031, %1282, !dbg !26 + %1284 = fadd float %1032, %1283, !dbg !26 + %1285 = fadd float %1033, %1284, !dbg !26 + %1286 = fadd float %1034, %1285, !dbg !26 + %1287 = fadd float %1035, %1036, !dbg !26 + %1288 = fadd float %1037, %1287, !dbg !26 + %1289 = fadd float %1038, %1288, !dbg !26 + %1290 = fadd float %1039, %1289, !dbg !26 + %1291 = fadd float %1040, %1290, !dbg !26 + %1292 = fadd float %1041, %1291, !dbg !26 + %1293 = fadd float %1042, %1292, !dbg !26 + %1294 = fadd float %1043, %1044, !dbg !26 + %1295 = fadd float %1045, %1294, !dbg !26 + %1296 = fadd float %1046, %1295, !dbg !26 + %1297 = fadd float %1047, %1296, !dbg !26 + %1298 = fadd float %1048, %1297, !dbg !26 + %1299 = fadd float %1049, %1298, !dbg !26 + %1300 = fadd float %1050, %1299, !dbg !26 + %1301 = fadd float %1051, %1052, !dbg !26 + %1302 = fadd float %1053, %1301, !dbg !26 + %1303 = fadd float %1054, %1302, !dbg !26 + %1304 = fadd float %1055, %1303, !dbg !26 + %1305 = fadd float %1056, %1304, !dbg !26 + %1306 = fadd float %1057, %1305, !dbg !26 + %1307 = fadd float %1058, %1306, !dbg !26 + %1308 = fadd float %1059, %1060, !dbg !26 + %1309 = fadd float %1061, %1308, !dbg !26 + %1310 = fadd float %1062, %1309, !dbg !26 + %1311 = fadd float %1063, %1310, !dbg !26 + %1312 = fadd float %1064, %1311, !dbg !26 + %1313 = fadd float %1065, %1312, !dbg !26 + %1314 = fadd float %1066, %1313, !dbg !26 + %1315 = fadd float %1067, %1068, !dbg !26 + %1316 = fadd float %1069, %1315, !dbg !26 + %1317 = fadd float %1070, %1316, !dbg !26 + %1318 = fadd float %1071, %1317, !dbg !26 + %1319 = fadd float %1072, %1318, !dbg !26 + %1320 = fadd float %1073, %1319, !dbg !26 + %1321 = fadd float %1074, %1320, !dbg !26 + %1322 = fadd float %1075, %1076, !dbg !26 + %1323 = fadd float %1077, %1322, !dbg !26 + %1324 = fadd float %1078, %1323, !dbg !26 + %1325 = fadd float %1079, %1324, !dbg !26 + %1326 = fadd float %1080, %1325, !dbg !26 + %1327 = fadd float %1081, %1326, !dbg !26 + %1328 = fadd float %1082, %1327, !dbg !26 + %1329 = fadd float %1083, %1084, !dbg !26 + %1330 = fadd float %1085, %1329, !dbg !26 + %1331 = fadd float %1086, %1330, !dbg !26 + %1332 = fadd float %1087, %1331, !dbg !26 + %1333 = fadd float %1088, %1332, !dbg !26 + %1334 = fadd float %1089, %1333, !dbg !26 + %1335 = fadd float %1090, %1334, !dbg !26 + %1336 = fadd float %1091, %1092, !dbg !26 + %1337 = fadd float %1093, %1336, !dbg !26 + %1338 = fadd float %1094, %1337, !dbg !26 + %1339 = fadd float %1095, %1338, !dbg !26 + %1340 = fadd float %1096, %1339, !dbg !26 + %1341 = fadd float %1097, %1340, !dbg !26 + %1342 = fadd float %1098, %1341, !dbg !26 + %1343 = fadd float %1099, %1100, !dbg !26 + %1344 = fadd float %1101, %1343, !dbg !26 + %1345 = fadd float %1102, %1344, !dbg !26 + %1346 = fadd float %1103, %1345, !dbg !26 + %1347 = fadd float %1104, %1346, !dbg !26 + %1348 = fadd float %1105, %1347, !dbg !26 + %1349 = fadd float %1106, %1348, !dbg !26 + %1350 = fadd float %1107, %1108, !dbg !26 + %1351 = fadd float %1109, %1350, !dbg !26 + %1352 = fadd float %1110, %1351, !dbg !26 + %1353 = fadd float %1111, %1352, !dbg !26 + %1354 = fadd float %1112, %1353, !dbg !26 + %1355 = fadd float %1113, %1354, !dbg !26 + %1356 = fadd float %1114, %1355, !dbg !26 + %1357 = fadd float %1115, %1116, !dbg !26 + %1358 = fadd float %1117, %1357, !dbg !26 + %1359 = fadd float %1118, %1358, !dbg !26 + %1360 = fadd float %1119, %1359, !dbg !26 + %1361 = fadd float %1120, %1360, !dbg !26 + %1362 = fadd float %1121, %1361, !dbg !26 + %1363 = fadd float %1122, %1362, !dbg !26 + %1364 = fadd float %1123, %1124, !dbg !26 + %1365 = fadd float %1125, %1364, !dbg !26 + %1366 = fadd float %1126, %1365, !dbg !26 + %1367 = fadd float %1127, %1366, !dbg !26 + %1368 = fadd float %1128, %1367, !dbg !26 + %1369 = fadd float %1129, %1368, !dbg !26 + %1370 = fadd float %1130, %1369, !dbg !26 + %1371 = fadd float %1131, %1132, !dbg !26 + %1372 = fadd float %1133, %1371, !dbg !26 + %1373 = fadd float %1134, %1372, !dbg !26 + %1374 = fadd float %1135, %1373, !dbg !26 + %1375 = fadd float %1136, %1374, !dbg !26 + %1376 = fadd float %1137, %1375, !dbg !26 + %1377 = fadd float %1138, %1376, !dbg !26 + %1378 = fadd float %1139, %1140, !dbg !26 + %1379 = fadd float %1141, %1378, !dbg !26 + %1380 = fadd float %1142, %1379, !dbg !26 + %1381 = fadd float %1143, %1380, !dbg !26 + %1382 = fadd float %1144, %1381, !dbg !26 + %1383 = fadd float %1145, %1382, !dbg !26 + %1384 = fadd float %1146, %1383, !dbg !26 + %1385 = fadd float %1147, %1148, !dbg !26 + %1386 = fadd float %1149, %1385, !dbg !26 + %1387 = fadd float %1150, %1386, !dbg !26 + %1388 = fadd float %1151, %1387, !dbg !26 + %1389 = fadd float %1152, %1388, !dbg !26 + %1390 = fadd float %1153, %1389, !dbg !26 + %1391 = fadd float %1154, %1390, !dbg !26 + %1392 = fadd float %1155, %1156, !dbg !26 + %1393 = fadd float %1157, %1392, !dbg !26 + %1394 = fadd float %1158, %1393, !dbg !26 + %1395 = fadd float %1159, %1394, !dbg !26 + %1396 = fadd float %1160, %1395, !dbg !26 + %1397 = fadd float %1161, %1396, !dbg !26 + %1398 = fadd float %1162, %1397, !dbg !26 + %1399 = fadd float %1163, %1164, !dbg !26 + %1400 = fadd float %1165, %1399, !dbg !26 + %1401 = fadd float %1166, %1400, !dbg !26 + %1402 = fadd float %1167, %1401, !dbg !26 + %1403 = fadd float %1168, %1402, !dbg !26 + %1404 = fadd float %1169, %1403, !dbg !26 + %1405 = fadd float %1170, %1404, !dbg !26 + %1406 = fadd float %1171, %1172, !dbg !26 + %1407 = fadd float %1173, %1406, !dbg !26 + %1408 = fadd float %1174, %1407, !dbg !26 + %1409 = fadd float %1175, %1408, !dbg !26 + %1410 = fadd float %1176, %1409, !dbg !26 + %1411 = fadd float %1177, %1410, !dbg !26 + %1412 = fadd float %1178, %1411, !dbg !26 + %1413 = fadd float %1179, %1180, !dbg !26 + %1414 = fadd float %1181, %1413, !dbg !26 + %1415 = fadd float %1182, %1414, !dbg !26 + %1416 = fadd float %1183, %1415, !dbg !26 + %1417 = fadd float %1184, %1416, !dbg !26 + %1418 = fadd float %1185, %1417, !dbg !26 + %1419 = fadd float %1186, %1418, !dbg !26 + %1420 = fadd float %1187, %1188, !dbg !26 + %1421 = fadd float %1189, %1420, !dbg !26 + %1422 = fadd float %1190, %1421, !dbg !26 + %1423 = fadd float %1191, %1422, !dbg !26 + %1424 = fadd float %1192, %1423, !dbg !26 + %1425 = fadd float %1193, %1424, !dbg !26 + %1426 = fadd float %1194, %1425, !dbg !26 + %1427 = fadd float %1195, %1196, !dbg !26 + %1428 = fadd float %1197, %1427, !dbg !26 + %1429 = fadd float %1198, %1428, !dbg !26 + %1430 = fadd float %1199, %1429, !dbg !26 + %1431 = fadd float %1200, %1430, !dbg !26 + %1432 = fadd float %1201, %1431, !dbg !26 + %1433 = fadd float %1202, %1432, !dbg !26 + %1434 = fadd float %1203, %1204, !dbg !26 + %1435 = fadd float %1205, %1434, !dbg !26 + %1436 = fadd float %1206, %1435, !dbg !26 + %1437 = fadd float %1207, %1436, !dbg !26 + %1438 = fadd float %1208, %1437, !dbg !26 + %1439 = fadd float %1209, %1438, !dbg !26 + %1440 = fadd float %1210, %1439, !dbg !26 + %1441 = fadd float %1211, %1212, !dbg !26 + %1442 = fadd float %1213, %1441, !dbg !26 + %1443 = fadd float %1214, %1442, !dbg !26 + %1444 = fadd float %1215, %1443, !dbg !26 + %1445 = fadd float %1216, %1444, !dbg !26 + %1446 = fadd float %1217, %1445, !dbg !26 + %1447 = fadd float %1218, %1446, !dbg !26 + %1448 = fadd float %1219, %1220, !dbg !26 + %1449 = fadd float %1221, %1448, !dbg !26 + %1450 = fadd float %1222, %1449, !dbg !26 + %1451 = fadd float %1223, %1450, !dbg !26 + %1452 = fadd float %1224, %1451, !dbg !26 + %1453 = fadd float %1225, %1452, !dbg !26 + %1454 = fadd float %1226, %1453, !dbg !26 + %1455 = fadd float %1227, %1228, !dbg !26 + %1456 = fadd float %1229, %1455, !dbg !26 + %1457 = fadd float %1230, %1456, !dbg !26 + %1458 = fadd float %1231, %1457, !dbg !26 + %1459 = fadd float %1232, %1458, !dbg !26 + %1460 = fadd float %1233, %1459, !dbg !26 + %1461 = fadd float %1234, %1460, !dbg !26 + %1462 = fadd float %1235, %1236, !dbg !26 + %1463 = fadd float %1237, %1462, !dbg !26 + %1464 = fadd float %1238, %1463, !dbg !26 + %1465 = fadd float %1239, %1464, !dbg !26 + %1466 = fadd float %1240, %1465, !dbg !26 + %1467 = fadd float %1241, %1466, !dbg !26 + %1468 = fadd float %1242, %1467, !dbg !26 + %1469 = fadd float %1243, %1244, !dbg !26 + %1470 = fadd float %1245, %1469, !dbg !26 + %1471 = fadd float %1246, %1470, !dbg !26 + %1472 = fadd float %1247, %1471, !dbg !26 + %1473 = fadd float %1248, %1472, !dbg !26 + %1474 = fadd float %1249, %1473, !dbg !26 + %1475 = fadd float %1250, %1474, !dbg !26 + %1476 = fadd float %1251, %1252, !dbg !26 + %1477 = fadd float %1253, %1476, !dbg !26 + %1478 = fadd float %1254, %1477, !dbg !26 + %1479 = fadd float %1255, %1478, !dbg !26 + %1480 = fadd float %1256, %1479, !dbg !26 + %1481 = fadd float %1257, %1480, !dbg !26 + %1482 = fadd float %1258, %1481, !dbg !26 + %1483 = bitcast float %1265 to i32, !dbg !29 + %1484 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1483, i32 8, i32 31), !dbg !29 + %1485 = bitcast i32 %1484 to float, !dbg !29 + %1486 = fadd float %1265, %1485, !dbg !26 + %1487 = bitcast float %1486 to i32, !dbg !29 + %1488 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1487, i32 4, i32 31), !dbg !29 + %1489 = bitcast i32 %1488 to float, !dbg !29 + %1490 = fadd float %1486, %1489, !dbg !26 + %1491 = bitcast float %1490 to i32, !dbg !29 + %1492 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1491, i32 2, i32 31), !dbg !29 + %1493 = bitcast i32 %1492 to float, !dbg !29 + %1494 = fadd float %1490, %1493, !dbg !26 + %1495 = bitcast float %1494 to i32, !dbg !29 + %1496 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1495, i32 1, i32 31), !dbg !29 + %1497 = bitcast i32 %1496 to float, !dbg !29 + %1498 = fadd float %1494, %1497, !dbg !26 + %1499 = bitcast float %1272 to i32, !dbg !29 + %1500 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1499, i32 8, i32 31), !dbg !29 + %1501 = bitcast i32 %1500 to float, !dbg !29 + %1502 = fadd float %1272, %1501, !dbg !26 + %1503 = bitcast float %1502 to i32, !dbg !29 + %1504 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1503, i32 4, i32 31), !dbg !29 + %1505 = bitcast i32 %1504 to float, !dbg !29 + %1506 = fadd float %1502, %1505, !dbg !26 + %1507 = bitcast float %1506 to i32, !dbg !29 + %1508 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1507, i32 2, i32 31), !dbg !29 + %1509 = bitcast i32 %1508 to float, !dbg !29 + %1510 = fadd float %1506, %1509, !dbg !26 + %1511 = bitcast float %1510 to i32, !dbg !29 + %1512 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1511, i32 1, i32 31), !dbg !29 + %1513 = bitcast i32 %1512 to float, !dbg !29 + %1514 = fadd float %1510, %1513, !dbg !26 + %1515 = bitcast float %1279 to i32, !dbg !29 + %1516 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1515, i32 8, i32 31), !dbg !29 + %1517 = bitcast i32 %1516 to float, !dbg !29 + %1518 = fadd float %1279, %1517, !dbg !26 + %1519 = bitcast float %1518 to i32, !dbg !29 + %1520 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1519, i32 4, i32 31), !dbg !29 + %1521 = bitcast i32 %1520 to float, !dbg !29 + %1522 = fadd float %1518, %1521, !dbg !26 + %1523 = bitcast float %1522 to i32, !dbg !29 + %1524 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1523, i32 2, i32 31), !dbg !29 + %1525 = bitcast i32 %1524 to float, !dbg !29 + %1526 = fadd float %1522, %1525, !dbg !26 + %1527 = bitcast float %1526 to i32, !dbg !29 + %1528 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1527, i32 1, i32 31), !dbg !29 + %1529 = bitcast i32 %1528 to float, !dbg !29 + %1530 = fadd float %1526, %1529, !dbg !26 + %1531 = bitcast float %1286 to i32, !dbg !29 + %1532 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1531, i32 8, i32 31), !dbg !29 + %1533 = bitcast i32 %1532 to float, !dbg !29 + %1534 = fadd float %1286, %1533, !dbg !26 + %1535 = bitcast float %1534 to i32, !dbg !29 + %1536 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1535, i32 4, i32 31), !dbg !29 + %1537 = bitcast i32 %1536 to float, !dbg !29 + %1538 = fadd float %1534, %1537, !dbg !26 + %1539 = bitcast float %1538 to i32, !dbg !29 + %1540 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1539, i32 2, i32 31), !dbg !29 + %1541 = bitcast i32 %1540 to float, !dbg !29 + %1542 = fadd float %1538, %1541, !dbg !26 + %1543 = bitcast float %1542 to i32, !dbg !29 + %1544 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1543, i32 1, i32 31), !dbg !29 + %1545 = bitcast i32 %1544 to float, !dbg !29 + %1546 = fadd float %1542, %1545, !dbg !26 + %1547 = bitcast float %1293 to i32, !dbg !29 + %1548 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1547, i32 8, i32 31), !dbg !29 + %1549 = bitcast i32 %1548 to float, !dbg !29 + %1550 = fadd float %1293, %1549, !dbg !26 + %1551 = bitcast float %1550 to i32, !dbg !29 + %1552 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1551, i32 4, i32 31), !dbg !29 + %1553 = bitcast i32 %1552 to float, !dbg !29 + %1554 = fadd float %1550, %1553, !dbg !26 + %1555 = bitcast float %1554 to i32, !dbg !29 + %1556 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1555, i32 2, i32 31), !dbg !29 + %1557 = bitcast i32 %1556 to float, !dbg !29 + %1558 = fadd float %1554, %1557, !dbg !26 + %1559 = bitcast float %1558 to i32, !dbg !29 + %1560 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1559, i32 1, i32 31), !dbg !29 + %1561 = bitcast i32 %1560 to float, !dbg !29 + %1562 = fadd float %1558, %1561, !dbg !26 + %1563 = bitcast float %1300 to i32, !dbg !29 + %1564 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1563, i32 8, i32 31), !dbg !29 + %1565 = bitcast i32 %1564 to float, !dbg !29 + %1566 = fadd float %1300, %1565, !dbg !26 + %1567 = bitcast float %1566 to i32, !dbg !29 + %1568 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1567, i32 4, i32 31), !dbg !29 + %1569 = bitcast i32 %1568 to float, !dbg !29 + %1570 = fadd float %1566, %1569, !dbg !26 + %1571 = bitcast float %1570 to i32, !dbg !29 + %1572 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1571, i32 2, i32 31), !dbg !29 + %1573 = bitcast i32 %1572 to float, !dbg !29 + %1574 = fadd float %1570, %1573, !dbg !26 + %1575 = bitcast float %1574 to i32, !dbg !29 + %1576 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1575, i32 1, i32 31), !dbg !29 + %1577 = bitcast i32 %1576 to float, !dbg !29 + %1578 = fadd float %1574, %1577, !dbg !26 + %1579 = bitcast float %1307 to i32, !dbg !29 + %1580 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1579, i32 8, i32 31), !dbg !29 + %1581 = bitcast i32 %1580 to float, !dbg !29 + %1582 = fadd float %1307, %1581, !dbg !26 + %1583 = bitcast float %1582 to i32, !dbg !29 + %1584 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1583, i32 4, i32 31), !dbg !29 + %1585 = bitcast i32 %1584 to float, !dbg !29 + %1586 = fadd float %1582, %1585, !dbg !26 + %1587 = bitcast float %1586 to i32, !dbg !29 + %1588 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1587, i32 2, i32 31), !dbg !29 + %1589 = bitcast i32 %1588 to float, !dbg !29 + %1590 = fadd float %1586, %1589, !dbg !26 + %1591 = bitcast float %1590 to i32, !dbg !29 + %1592 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1591, i32 1, i32 31), !dbg !29 + %1593 = bitcast i32 %1592 to float, !dbg !29 + %1594 = fadd float %1590, %1593, !dbg !26 + %1595 = bitcast float %1314 to i32, !dbg !29 + %1596 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1595, i32 8, i32 31), !dbg !29 + %1597 = bitcast i32 %1596 to float, !dbg !29 + %1598 = fadd float %1314, %1597, !dbg !26 + %1599 = bitcast float %1598 to i32, !dbg !29 + %1600 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1599, i32 4, i32 31), !dbg !29 + %1601 = bitcast i32 %1600 to float, !dbg !29 + %1602 = fadd float %1598, %1601, !dbg !26 + %1603 = bitcast float %1602 to i32, !dbg !29 + %1604 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1603, i32 2, i32 31), !dbg !29 + %1605 = bitcast i32 %1604 to float, !dbg !29 + %1606 = fadd float %1602, %1605, !dbg !26 + %1607 = bitcast float %1606 to i32, !dbg !29 + %1608 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1607, i32 1, i32 31), !dbg !29 + %1609 = bitcast i32 %1608 to float, !dbg !29 + %1610 = fadd float %1606, %1609, !dbg !26 + %1611 = bitcast float %1321 to i32, !dbg !29 + %1612 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1611, i32 8, i32 31), !dbg !29 + %1613 = bitcast i32 %1612 to float, !dbg !29 + %1614 = fadd float %1321, %1613, !dbg !26 + %1615 = bitcast float %1614 to i32, !dbg !29 + %1616 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1615, i32 4, i32 31), !dbg !29 + %1617 = bitcast i32 %1616 to float, !dbg !29 + %1618 = fadd float %1614, %1617, !dbg !26 + %1619 = bitcast float %1618 to i32, !dbg !29 + %1620 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1619, i32 2, i32 31), !dbg !29 + %1621 = bitcast i32 %1620 to float, !dbg !29 + %1622 = fadd float %1618, %1621, !dbg !26 + %1623 = bitcast float %1622 to i32, !dbg !29 + %1624 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1623, i32 1, i32 31), !dbg !29 + %1625 = bitcast i32 %1624 to float, !dbg !29 + %1626 = fadd float %1622, %1625, !dbg !26 + %1627 = bitcast float %1328 to i32, !dbg !29 + %1628 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1627, i32 8, i32 31), !dbg !29 + %1629 = bitcast i32 %1628 to float, !dbg !29 + %1630 = fadd float %1328, %1629, !dbg !26 + %1631 = bitcast float %1630 to i32, !dbg !29 + %1632 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1631, i32 4, i32 31), !dbg !29 + %1633 = bitcast i32 %1632 to float, !dbg !29 + %1634 = fadd float %1630, %1633, !dbg !26 + %1635 = bitcast float %1634 to i32, !dbg !29 + %1636 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1635, i32 2, i32 31), !dbg !29 + %1637 = bitcast i32 %1636 to float, !dbg !29 + %1638 = fadd float %1634, %1637, !dbg !26 + %1639 = bitcast float %1638 to i32, !dbg !29 + %1640 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1639, i32 1, i32 31), !dbg !29 + %1641 = bitcast i32 %1640 to float, !dbg !29 + %1642 = fadd float %1638, %1641, !dbg !26 + %1643 = bitcast float %1335 to i32, !dbg !29 + %1644 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1643, i32 8, i32 31), !dbg !29 + %1645 = bitcast i32 %1644 to float, !dbg !29 + %1646 = fadd float %1335, %1645, !dbg !26 + %1647 = bitcast float %1646 to i32, !dbg !29 + %1648 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1647, i32 4, i32 31), !dbg !29 + %1649 = bitcast i32 %1648 to float, !dbg !29 + %1650 = fadd float %1646, %1649, !dbg !26 + %1651 = bitcast float %1650 to i32, !dbg !29 + %1652 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1651, i32 2, i32 31), !dbg !29 + %1653 = bitcast i32 %1652 to float, !dbg !29 + %1654 = fadd float %1650, %1653, !dbg !26 + %1655 = bitcast float %1654 to i32, !dbg !29 + %1656 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1655, i32 1, i32 31), !dbg !29 + %1657 = bitcast i32 %1656 to float, !dbg !29 + %1658 = fadd float %1654, %1657, !dbg !26 + %1659 = bitcast float %1342 to i32, !dbg !29 + %1660 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1659, i32 8, i32 31), !dbg !29 + %1661 = bitcast i32 %1660 to float, !dbg !29 + %1662 = fadd float %1342, %1661, !dbg !26 + %1663 = bitcast float %1662 to i32, !dbg !29 + %1664 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1663, i32 4, i32 31), !dbg !29 + %1665 = bitcast i32 %1664 to float, !dbg !29 + %1666 = fadd float %1662, %1665, !dbg !26 + %1667 = bitcast float %1666 to i32, !dbg !29 + %1668 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1667, i32 2, i32 31), !dbg !29 + %1669 = bitcast i32 %1668 to float, !dbg !29 + %1670 = fadd float %1666, %1669, !dbg !26 + %1671 = bitcast float %1670 to i32, !dbg !29 + %1672 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1671, i32 1, i32 31), !dbg !29 + %1673 = bitcast i32 %1672 to float, !dbg !29 + %1674 = fadd float %1670, %1673, !dbg !26 + %1675 = bitcast float %1349 to i32, !dbg !29 + %1676 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1675, i32 8, i32 31), !dbg !29 + %1677 = bitcast i32 %1676 to float, !dbg !29 + %1678 = fadd float %1349, %1677, !dbg !26 + %1679 = bitcast float %1678 to i32, !dbg !29 + %1680 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1679, i32 4, i32 31), !dbg !29 + %1681 = bitcast i32 %1680 to float, !dbg !29 + %1682 = fadd float %1678, %1681, !dbg !26 + %1683 = bitcast float %1682 to i32, !dbg !29 + %1684 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1683, i32 2, i32 31), !dbg !29 + %1685 = bitcast i32 %1684 to float, !dbg !29 + %1686 = fadd float %1682, %1685, !dbg !26 + %1687 = bitcast float %1686 to i32, !dbg !29 + %1688 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1687, i32 1, i32 31), !dbg !29 + %1689 = bitcast i32 %1688 to float, !dbg !29 + %1690 = fadd float %1686, %1689, !dbg !26 + %1691 = bitcast float %1356 to i32, !dbg !29 + %1692 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1691, i32 8, i32 31), !dbg !29 + %1693 = bitcast i32 %1692 to float, !dbg !29 + %1694 = fadd float %1356, %1693, !dbg !26 + %1695 = bitcast float %1694 to i32, !dbg !29 + %1696 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1695, i32 4, i32 31), !dbg !29 + %1697 = bitcast i32 %1696 to float, !dbg !29 + %1698 = fadd float %1694, %1697, !dbg !26 + %1699 = bitcast float %1698 to i32, !dbg !29 + %1700 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1699, i32 2, i32 31), !dbg !29 + %1701 = bitcast i32 %1700 to float, !dbg !29 + %1702 = fadd float %1698, %1701, !dbg !26 + %1703 = bitcast float %1702 to i32, !dbg !29 + %1704 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1703, i32 1, i32 31), !dbg !29 + %1705 = bitcast i32 %1704 to float, !dbg !29 + %1706 = fadd float %1702, %1705, !dbg !26 + %1707 = bitcast float %1363 to i32, !dbg !29 + %1708 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1707, i32 8, i32 31), !dbg !29 + %1709 = bitcast i32 %1708 to float, !dbg !29 + %1710 = fadd float %1363, %1709, !dbg !26 + %1711 = bitcast float %1710 to i32, !dbg !29 + %1712 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1711, i32 4, i32 31), !dbg !29 + %1713 = bitcast i32 %1712 to float, !dbg !29 + %1714 = fadd float %1710, %1713, !dbg !26 + %1715 = bitcast float %1714 to i32, !dbg !29 + %1716 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1715, i32 2, i32 31), !dbg !29 + %1717 = bitcast i32 %1716 to float, !dbg !29 + %1718 = fadd float %1714, %1717, !dbg !26 + %1719 = bitcast float %1718 to i32, !dbg !29 + %1720 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1719, i32 1, i32 31), !dbg !29 + %1721 = bitcast i32 %1720 to float, !dbg !29 + %1722 = fadd float %1718, %1721, !dbg !26 + %1723 = bitcast float %1370 to i32, !dbg !29 + %1724 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1723, i32 8, i32 31), !dbg !29 + %1725 = bitcast i32 %1724 to float, !dbg !29 + %1726 = fadd float %1370, %1725, !dbg !26 + %1727 = bitcast float %1726 to i32, !dbg !29 + %1728 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1727, i32 4, i32 31), !dbg !29 + %1729 = bitcast i32 %1728 to float, !dbg !29 + %1730 = fadd float %1726, %1729, !dbg !26 + %1731 = bitcast float %1730 to i32, !dbg !29 + %1732 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1731, i32 2, i32 31), !dbg !29 + %1733 = bitcast i32 %1732 to float, !dbg !29 + %1734 = fadd float %1730, %1733, !dbg !26 + %1735 = bitcast float %1734 to i32, !dbg !29 + %1736 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1735, i32 1, i32 31), !dbg !29 + %1737 = bitcast i32 %1736 to float, !dbg !29 + %1738 = fadd float %1734, %1737, !dbg !26 + %1739 = bitcast float %1377 to i32, !dbg !29 + %1740 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1739, i32 8, i32 31), !dbg !29 + %1741 = bitcast i32 %1740 to float, !dbg !29 + %1742 = fadd float %1377, %1741, !dbg !26 + %1743 = bitcast float %1742 to i32, !dbg !29 + %1744 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1743, i32 4, i32 31), !dbg !29 + %1745 = bitcast i32 %1744 to float, !dbg !29 + %1746 = fadd float %1742, %1745, !dbg !26 + %1747 = bitcast float %1746 to i32, !dbg !29 + %1748 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1747, i32 2, i32 31), !dbg !29 + %1749 = bitcast i32 %1748 to float, !dbg !29 + %1750 = fadd float %1746, %1749, !dbg !26 + %1751 = bitcast float %1750 to i32, !dbg !29 + %1752 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1751, i32 1, i32 31), !dbg !29 + %1753 = bitcast i32 %1752 to float, !dbg !29 + %1754 = fadd float %1750, %1753, !dbg !26 + %1755 = bitcast float %1384 to i32, !dbg !29 + %1756 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1755, i32 8, i32 31), !dbg !29 + %1757 = bitcast i32 %1756 to float, !dbg !29 + %1758 = fadd float %1384, %1757, !dbg !26 + %1759 = bitcast float %1758 to i32, !dbg !29 + %1760 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1759, i32 4, i32 31), !dbg !29 + %1761 = bitcast i32 %1760 to float, !dbg !29 + %1762 = fadd float %1758, %1761, !dbg !26 + %1763 = bitcast float %1762 to i32, !dbg !29 + %1764 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1763, i32 2, i32 31), !dbg !29 + %1765 = bitcast i32 %1764 to float, !dbg !29 + %1766 = fadd float %1762, %1765, !dbg !26 + %1767 = bitcast float %1766 to i32, !dbg !29 + %1768 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1767, i32 1, i32 31), !dbg !29 + %1769 = bitcast i32 %1768 to float, !dbg !29 + %1770 = fadd float %1766, %1769, !dbg !26 + %1771 = bitcast float %1391 to i32, !dbg !29 + %1772 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1771, i32 8, i32 31), !dbg !29 + %1773 = bitcast i32 %1772 to float, !dbg !29 + %1774 = fadd float %1391, %1773, !dbg !26 + %1775 = bitcast float %1774 to i32, !dbg !29 + %1776 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1775, i32 4, i32 31), !dbg !29 + %1777 = bitcast i32 %1776 to float, !dbg !29 + %1778 = fadd float %1774, %1777, !dbg !26 + %1779 = bitcast float %1778 to i32, !dbg !29 + %1780 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1779, i32 2, i32 31), !dbg !29 + %1781 = bitcast i32 %1780 to float, !dbg !29 + %1782 = fadd float %1778, %1781, !dbg !26 + %1783 = bitcast float %1782 to i32, !dbg !29 + %1784 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1783, i32 1, i32 31), !dbg !29 + %1785 = bitcast i32 %1784 to float, !dbg !29 + %1786 = fadd float %1782, %1785, !dbg !26 + %1787 = bitcast float %1398 to i32, !dbg !29 + %1788 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1787, i32 8, i32 31), !dbg !29 + %1789 = bitcast i32 %1788 to float, !dbg !29 + %1790 = fadd float %1398, %1789, !dbg !26 + %1791 = bitcast float %1790 to i32, !dbg !29 + %1792 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1791, i32 4, i32 31), !dbg !29 + %1793 = bitcast i32 %1792 to float, !dbg !29 + %1794 = fadd float %1790, %1793, !dbg !26 + %1795 = bitcast float %1794 to i32, !dbg !29 + %1796 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1795, i32 2, i32 31), !dbg !29 + %1797 = bitcast i32 %1796 to float, !dbg !29 + %1798 = fadd float %1794, %1797, !dbg !26 + %1799 = bitcast float %1798 to i32, !dbg !29 + %1800 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1799, i32 1, i32 31), !dbg !29 + %1801 = bitcast i32 %1800 to float, !dbg !29 + %1802 = fadd float %1798, %1801, !dbg !26 + %1803 = bitcast float %1405 to i32, !dbg !29 + %1804 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1803, i32 8, i32 31), !dbg !29 + %1805 = bitcast i32 %1804 to float, !dbg !29 + %1806 = fadd float %1405, %1805, !dbg !26 + %1807 = bitcast float %1806 to i32, !dbg !29 + %1808 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1807, i32 4, i32 31), !dbg !29 + %1809 = bitcast i32 %1808 to float, !dbg !29 + %1810 = fadd float %1806, %1809, !dbg !26 + %1811 = bitcast float %1810 to i32, !dbg !29 + %1812 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1811, i32 2, i32 31), !dbg !29 + %1813 = bitcast i32 %1812 to float, !dbg !29 + %1814 = fadd float %1810, %1813, !dbg !26 + %1815 = bitcast float %1814 to i32, !dbg !29 + %1816 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1815, i32 1, i32 31), !dbg !29 + %1817 = bitcast i32 %1816 to float, !dbg !29 + %1818 = fadd float %1814, %1817, !dbg !26 + %1819 = bitcast float %1412 to i32, !dbg !29 + %1820 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1819, i32 8, i32 31), !dbg !29 + %1821 = bitcast i32 %1820 to float, !dbg !29 + %1822 = fadd float %1412, %1821, !dbg !26 + %1823 = bitcast float %1822 to i32, !dbg !29 + %1824 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1823, i32 4, i32 31), !dbg !29 + %1825 = bitcast i32 %1824 to float, !dbg !29 + %1826 = fadd float %1822, %1825, !dbg !26 + %1827 = bitcast float %1826 to i32, !dbg !29 + %1828 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1827, i32 2, i32 31), !dbg !29 + %1829 = bitcast i32 %1828 to float, !dbg !29 + %1830 = fadd float %1826, %1829, !dbg !26 + %1831 = bitcast float %1830 to i32, !dbg !29 + %1832 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1831, i32 1, i32 31), !dbg !29 + %1833 = bitcast i32 %1832 to float, !dbg !29 + %1834 = fadd float %1830, %1833, !dbg !26 + %1835 = bitcast float %1419 to i32, !dbg !29 + %1836 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1835, i32 8, i32 31), !dbg !29 + %1837 = bitcast i32 %1836 to float, !dbg !29 + %1838 = fadd float %1419, %1837, !dbg !26 + %1839 = bitcast float %1838 to i32, !dbg !29 + %1840 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1839, i32 4, i32 31), !dbg !29 + %1841 = bitcast i32 %1840 to float, !dbg !29 + %1842 = fadd float %1838, %1841, !dbg !26 + %1843 = bitcast float %1842 to i32, !dbg !29 + %1844 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1843, i32 2, i32 31), !dbg !29 + %1845 = bitcast i32 %1844 to float, !dbg !29 + %1846 = fadd float %1842, %1845, !dbg !26 + %1847 = bitcast float %1846 to i32, !dbg !29 + %1848 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1847, i32 1, i32 31), !dbg !29 + %1849 = bitcast i32 %1848 to float, !dbg !29 + %1850 = fadd float %1846, %1849, !dbg !26 + %1851 = bitcast float %1426 to i32, !dbg !29 + %1852 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1851, i32 8, i32 31), !dbg !29 + %1853 = bitcast i32 %1852 to float, !dbg !29 + %1854 = fadd float %1426, %1853, !dbg !26 + %1855 = bitcast float %1854 to i32, !dbg !29 + %1856 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1855, i32 4, i32 31), !dbg !29 + %1857 = bitcast i32 %1856 to float, !dbg !29 + %1858 = fadd float %1854, %1857, !dbg !26 + %1859 = bitcast float %1858 to i32, !dbg !29 + %1860 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1859, i32 2, i32 31), !dbg !29 + %1861 = bitcast i32 %1860 to float, !dbg !29 + %1862 = fadd float %1858, %1861, !dbg !26 + %1863 = bitcast float %1862 to i32, !dbg !29 + %1864 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1863, i32 1, i32 31), !dbg !29 + %1865 = bitcast i32 %1864 to float, !dbg !29 + %1866 = fadd float %1862, %1865, !dbg !26 + %1867 = bitcast float %1433 to i32, !dbg !29 + %1868 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1867, i32 8, i32 31), !dbg !29 + %1869 = bitcast i32 %1868 to float, !dbg !29 + %1870 = fadd float %1433, %1869, !dbg !26 + %1871 = bitcast float %1870 to i32, !dbg !29 + %1872 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1871, i32 4, i32 31), !dbg !29 + %1873 = bitcast i32 %1872 to float, !dbg !29 + %1874 = fadd float %1870, %1873, !dbg !26 + %1875 = bitcast float %1874 to i32, !dbg !29 + %1876 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1875, i32 2, i32 31), !dbg !29 + %1877 = bitcast i32 %1876 to float, !dbg !29 + %1878 = fadd float %1874, %1877, !dbg !26 + %1879 = bitcast float %1878 to i32, !dbg !29 + %1880 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1879, i32 1, i32 31), !dbg !29 + %1881 = bitcast i32 %1880 to float, !dbg !29 + %1882 = fadd float %1878, %1881, !dbg !26 + %1883 = bitcast float %1440 to i32, !dbg !29 + %1884 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1883, i32 8, i32 31), !dbg !29 + %1885 = bitcast i32 %1884 to float, !dbg !29 + %1886 = fadd float %1440, %1885, !dbg !26 + %1887 = bitcast float %1886 to i32, !dbg !29 + %1888 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1887, i32 4, i32 31), !dbg !29 + %1889 = bitcast i32 %1888 to float, !dbg !29 + %1890 = fadd float %1886, %1889, !dbg !26 + %1891 = bitcast float %1890 to i32, !dbg !29 + %1892 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1891, i32 2, i32 31), !dbg !29 + %1893 = bitcast i32 %1892 to float, !dbg !29 + %1894 = fadd float %1890, %1893, !dbg !26 + %1895 = bitcast float %1894 to i32, !dbg !29 + %1896 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1895, i32 1, i32 31), !dbg !29 + %1897 = bitcast i32 %1896 to float, !dbg !29 + %1898 = fadd float %1894, %1897, !dbg !26 + %1899 = bitcast float %1447 to i32, !dbg !29 + %1900 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1899, i32 8, i32 31), !dbg !29 + %1901 = bitcast i32 %1900 to float, !dbg !29 + %1902 = fadd float %1447, %1901, !dbg !26 + %1903 = bitcast float %1902 to i32, !dbg !29 + %1904 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1903, i32 4, i32 31), !dbg !29 + %1905 = bitcast i32 %1904 to float, !dbg !29 + %1906 = fadd float %1902, %1905, !dbg !26 + %1907 = bitcast float %1906 to i32, !dbg !29 + %1908 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1907, i32 2, i32 31), !dbg !29 + %1909 = bitcast i32 %1908 to float, !dbg !29 + %1910 = fadd float %1906, %1909, !dbg !26 + %1911 = bitcast float %1910 to i32, !dbg !29 + %1912 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1911, i32 1, i32 31), !dbg !29 + %1913 = bitcast i32 %1912 to float, !dbg !29 + %1914 = fadd float %1910, %1913, !dbg !26 + %1915 = bitcast float %1454 to i32, !dbg !29 + %1916 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1915, i32 8, i32 31), !dbg !29 + %1917 = bitcast i32 %1916 to float, !dbg !29 + %1918 = fadd float %1454, %1917, !dbg !26 + %1919 = bitcast float %1918 to i32, !dbg !29 + %1920 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1919, i32 4, i32 31), !dbg !29 + %1921 = bitcast i32 %1920 to float, !dbg !29 + %1922 = fadd float %1918, %1921, !dbg !26 + %1923 = bitcast float %1922 to i32, !dbg !29 + %1924 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1923, i32 2, i32 31), !dbg !29 + %1925 = bitcast i32 %1924 to float, !dbg !29 + %1926 = fadd float %1922, %1925, !dbg !26 + %1927 = bitcast float %1926 to i32, !dbg !29 + %1928 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1927, i32 1, i32 31), !dbg !29 + %1929 = bitcast i32 %1928 to float, !dbg !29 + %1930 = fadd float %1926, %1929, !dbg !26 + %1931 = bitcast float %1461 to i32, !dbg !29 + %1932 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1931, i32 8, i32 31), !dbg !29 + %1933 = bitcast i32 %1932 to float, !dbg !29 + %1934 = fadd float %1461, %1933, !dbg !26 + %1935 = bitcast float %1934 to i32, !dbg !29 + %1936 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1935, i32 4, i32 31), !dbg !29 + %1937 = bitcast i32 %1936 to float, !dbg !29 + %1938 = fadd float %1934, %1937, !dbg !26 + %1939 = bitcast float %1938 to i32, !dbg !29 + %1940 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1939, i32 2, i32 31), !dbg !29 + %1941 = bitcast i32 %1940 to float, !dbg !29 + %1942 = fadd float %1938, %1941, !dbg !26 + %1943 = bitcast float %1942 to i32, !dbg !29 + %1944 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1943, i32 1, i32 31), !dbg !29 + %1945 = bitcast i32 %1944 to float, !dbg !29 + %1946 = fadd float %1942, %1945, !dbg !26 + %1947 = bitcast float %1468 to i32, !dbg !29 + %1948 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1947, i32 8, i32 31), !dbg !29 + %1949 = bitcast i32 %1948 to float, !dbg !29 + %1950 = fadd float %1468, %1949, !dbg !26 + %1951 = bitcast float %1950 to i32, !dbg !29 + %1952 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1951, i32 4, i32 31), !dbg !29 + %1953 = bitcast i32 %1952 to float, !dbg !29 + %1954 = fadd float %1950, %1953, !dbg !26 + %1955 = bitcast float %1954 to i32, !dbg !29 + %1956 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1955, i32 2, i32 31), !dbg !29 + %1957 = bitcast i32 %1956 to float, !dbg !29 + %1958 = fadd float %1954, %1957, !dbg !26 + %1959 = bitcast float %1958 to i32, !dbg !29 + %1960 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1959, i32 1, i32 31), !dbg !29 + %1961 = bitcast i32 %1960 to float, !dbg !29 + %1962 = fadd float %1958, %1961, !dbg !26 + %1963 = bitcast float %1475 to i32, !dbg !29 + %1964 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1963, i32 8, i32 31), !dbg !29 + %1965 = bitcast i32 %1964 to float, !dbg !29 + %1966 = fadd float %1475, %1965, !dbg !26 + %1967 = bitcast float %1966 to i32, !dbg !29 + %1968 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1967, i32 4, i32 31), !dbg !29 + %1969 = bitcast i32 %1968 to float, !dbg !29 + %1970 = fadd float %1966, %1969, !dbg !26 + %1971 = bitcast float %1970 to i32, !dbg !29 + %1972 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1971, i32 2, i32 31), !dbg !29 + %1973 = bitcast i32 %1972 to float, !dbg !29 + %1974 = fadd float %1970, %1973, !dbg !26 + %1975 = bitcast float %1974 to i32, !dbg !29 + %1976 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1975, i32 1, i32 31), !dbg !29 + %1977 = bitcast i32 %1976 to float, !dbg !29 + %1978 = fadd float %1974, %1977, !dbg !26 + %1979 = bitcast float %1482 to i32, !dbg !29 + %1980 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1979, i32 8, i32 31), !dbg !29 + %1981 = bitcast i32 %1980 to float, !dbg !29 + %1982 = fadd float %1482, %1981, !dbg !26 + %1983 = bitcast float %1982 to i32, !dbg !29 + %1984 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1983, i32 4, i32 31), !dbg !29 + %1985 = bitcast i32 %1984 to float, !dbg !29 + %1986 = fadd float %1982, %1985, !dbg !26 + %1987 = bitcast float %1986 to i32, !dbg !29 + %1988 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1987, i32 2, i32 31), !dbg !29 + %1989 = bitcast i32 %1988 to float, !dbg !29 + %1990 = fadd float %1986, %1989, !dbg !26 + %1991 = bitcast float %1990 to i32, !dbg !29 + %1992 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %1991, i32 1, i32 31), !dbg !29 + %1993 = bitcast i32 %1992 to float, !dbg !29 + %1994 = fadd float %1990, %1993, !dbg !26 + %1995 = fadd float %3, %1498, !dbg !32 + %1996 = fadd float %3, %1514, !dbg !32 + %1997 = fadd float %3, %1530, !dbg !32 + %1998 = fadd float %3, %1546, !dbg !32 + %1999 = fadd float %3, %1562, !dbg !32 + %2000 = fadd float %3, %1578, !dbg !32 + %2001 = fadd float %3, %1594, !dbg !32 + %2002 = fadd float %3, %1610, !dbg !32 + %2003 = fadd float %3, %1626, !dbg !32 + %2004 = fadd float %3, %1642, !dbg !32 + %2005 = fadd float %3, %1658, !dbg !32 + %2006 = fadd float %3, %1674, !dbg !32 + %2007 = fadd float %3, %1690, !dbg !32 + %2008 = fadd float %3, %1706, !dbg !32 + %2009 = fadd float %3, %1722, !dbg !32 + %2010 = fadd float %3, %1738, !dbg !32 + %2011 = fadd float %3, %1754, !dbg !32 + %2012 = fadd float %3, %1770, !dbg !32 + %2013 = fadd float %3, %1786, !dbg !32 + %2014 = fadd float %3, %1802, !dbg !32 + %2015 = fadd float %3, %1818, !dbg !32 + %2016 = fadd float %3, %1834, !dbg !32 + %2017 = fadd float %3, %1850, !dbg !32 + %2018 = fadd float %3, %1866, !dbg !32 + %2019 = fadd float %3, %1882, !dbg !32 + %2020 = fadd float %3, %1898, !dbg !32 + %2021 = fadd float %3, %1914, !dbg !32 + %2022 = fadd float %3, %1930, !dbg !32 + %2023 = fadd float %3, %1946, !dbg !32 + %2024 = fadd float %3, %1962, !dbg !32 + %2025 = fadd float %3, %1978, !dbg !32 + %2026 = fadd float %3, %1994, !dbg !32 + %2027 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i = icmp eq i32 %2027, 0, !dbg !33 + %2028 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i = icmp eq i32 %2028, 0, !dbg !33 + br i1 %.not.i, label %2034, label %2029, !dbg !33 + +2029: ; preds = %7 + br i1 %.not1.i, label %2032, label %2030, !dbg !33 + +2030: ; preds = %2029 + %2031 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1995) #7, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +2032: ; preds = %2029 + %2033 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1995) #7, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +2034: ; preds = %7 + br i1 %.not1.i, label %2037, label %2035, !dbg !33 + +2035: ; preds = %2034 + %2036 = tail call float @llvm.nvvm.sqrt.rn.f(float %1995) #7, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +2037: ; preds = %2034 + %2038 = tail call float @llvm.nvvm.sqrt.approx.f(float %1995) #7, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +__nv_sqrtf.exit: ; preds = %2030, %2032, %2035, %2037 + %.0.i = phi float [ %2031, %2030 ], [ %2033, %2032 ], [ %2036, %2035 ], [ %2038, %2037 ], !dbg !33 + %2039 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i1 = icmp eq i32 %2039, 0, !dbg !33 + %2040 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i4 = icmp eq i32 %2040, 0, !dbg !33 + br i1 %.not.i1, label %2046, label %2041, !dbg !33 + +2041: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %2044, label %2042, !dbg !33 + +2042: ; preds = %2041 + %2043 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1996) #7, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +2044: ; preds = %2041 + %2045 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1996) #7, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +2046: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %2049, label %2047, !dbg !33 + +2047: ; preds = %2046 + %2048 = tail call float @llvm.nvvm.sqrt.rn.f(float %1996) #7, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +2049: ; preds = %2046 + %2050 = tail call float @llvm.nvvm.sqrt.approx.f(float %1996) #7, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +__nv_sqrtf.exit5: ; preds = %2042, %2044, %2047, %2049 + %.0.i3 = phi float [ %2043, %2042 ], [ %2045, %2044 ], [ %2048, %2047 ], [ %2050, %2049 ], !dbg !33 + %2051 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i6 = icmp eq i32 %2051, 0, !dbg !33 + %2052 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i9 = icmp eq i32 %2052, 0, !dbg !33 + br i1 %.not.i6, label %2058, label %2053, !dbg !33 + +2053: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %2056, label %2054, !dbg !33 + +2054: ; preds = %2053 + %2055 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1997) #7, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +2056: ; preds = %2053 + %2057 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1997) #7, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +2058: ; preds = %__nv_sqrtf.exit5 + br i1 %.not1.i9, label %2061, label %2059, !dbg !33 + +2059: ; preds = %2058 + %2060 = tail call float @llvm.nvvm.sqrt.rn.f(float %1997) #7, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +2061: ; preds = %2058 + %2062 = tail call float @llvm.nvvm.sqrt.approx.f(float %1997) #7, !dbg !33 + br label %__nv_sqrtf.exit10, !dbg !33 + +__nv_sqrtf.exit10: ; preds = %2054, %2056, %2059, %2061 + %.0.i8 = phi float [ %2055, %2054 ], [ %2057, %2056 ], [ %2060, %2059 ], [ %2062, %2061 ], !dbg !33 + %2063 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i11 = icmp eq i32 %2063, 0, !dbg !33 + %2064 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i14 = icmp eq i32 %2064, 0, !dbg !33 + br i1 %.not.i11, label %2070, label %2065, !dbg !33 + +2065: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %2068, label %2066, !dbg !33 + +2066: ; preds = %2065 + %2067 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1998) #7, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +2068: ; preds = %2065 + %2069 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1998) #7, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +2070: ; preds = %__nv_sqrtf.exit10 + br i1 %.not1.i14, label %2073, label %2071, !dbg !33 + +2071: ; preds = %2070 + %2072 = tail call float @llvm.nvvm.sqrt.rn.f(float %1998) #7, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +2073: ; preds = %2070 + %2074 = tail call float @llvm.nvvm.sqrt.approx.f(float %1998) #7, !dbg !33 + br label %__nv_sqrtf.exit15, !dbg !33 + +__nv_sqrtf.exit15: ; preds = %2066, %2068, %2071, %2073 + %.0.i13 = phi float [ %2067, %2066 ], [ %2069, %2068 ], [ %2072, %2071 ], [ %2074, %2073 ], !dbg !33 + %2075 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i16 = icmp eq i32 %2075, 0, !dbg !33 + %2076 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i19 = icmp eq i32 %2076, 0, !dbg !33 + br i1 %.not.i16, label %2082, label %2077, !dbg !33 + +2077: ; preds = %__nv_sqrtf.exit15 + br i1 %.not1.i19, label %2080, label %2078, !dbg !33 + +2078: ; preds = %2077 + %2079 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %1999) #7, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +2080: ; preds = %2077 + %2081 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %1999) #7, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +2082: ; preds = %__nv_sqrtf.exit15 + br i1 %.not1.i19, label %2085, label %2083, !dbg !33 + +2083: ; preds = %2082 + %2084 = tail call float @llvm.nvvm.sqrt.rn.f(float %1999) #7, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +2085: ; preds = %2082 + %2086 = tail call float @llvm.nvvm.sqrt.approx.f(float %1999) #7, !dbg !33 + br label %__nv_sqrtf.exit20, !dbg !33 + +__nv_sqrtf.exit20: ; preds = %2078, %2080, %2083, %2085 + %.0.i18 = phi float [ %2079, %2078 ], [ %2081, %2080 ], [ %2084, %2083 ], [ %2086, %2085 ], !dbg !33 + %2087 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i21 = icmp eq i32 %2087, 0, !dbg !33 + %2088 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i24 = icmp eq i32 %2088, 0, !dbg !33 + br i1 %.not.i21, label %2094, label %2089, !dbg !33 + +2089: ; preds = %__nv_sqrtf.exit20 + br i1 %.not1.i24, label %2092, label %2090, !dbg !33 + +2090: ; preds = %2089 + %2091 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2000) #7, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +2092: ; preds = %2089 + %2093 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2000) #7, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +2094: ; preds = %__nv_sqrtf.exit20 + br i1 %.not1.i24, label %2097, label %2095, !dbg !33 + +2095: ; preds = %2094 + %2096 = tail call float @llvm.nvvm.sqrt.rn.f(float %2000) #7, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +2097: ; preds = %2094 + %2098 = tail call float @llvm.nvvm.sqrt.approx.f(float %2000) #7, !dbg !33 + br label %__nv_sqrtf.exit25, !dbg !33 + +__nv_sqrtf.exit25: ; preds = %2090, %2092, %2095, %2097 + %.0.i23 = phi float [ %2091, %2090 ], [ %2093, %2092 ], [ %2096, %2095 ], [ %2098, %2097 ], !dbg !33 + %2099 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i26 = icmp eq i32 %2099, 0, !dbg !33 + %2100 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i29 = icmp eq i32 %2100, 0, !dbg !33 + br i1 %.not.i26, label %2106, label %2101, !dbg !33 + +2101: ; preds = %__nv_sqrtf.exit25 + br i1 %.not1.i29, label %2104, label %2102, !dbg !33 + +2102: ; preds = %2101 + %2103 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2001) #7, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +2104: ; preds = %2101 + %2105 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2001) #7, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +2106: ; preds = %__nv_sqrtf.exit25 + br i1 %.not1.i29, label %2109, label %2107, !dbg !33 + +2107: ; preds = %2106 + %2108 = tail call float @llvm.nvvm.sqrt.rn.f(float %2001) #7, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +2109: ; preds = %2106 + %2110 = tail call float @llvm.nvvm.sqrt.approx.f(float %2001) #7, !dbg !33 + br label %__nv_sqrtf.exit30, !dbg !33 + +__nv_sqrtf.exit30: ; preds = %2102, %2104, %2107, %2109 + %.0.i28 = phi float [ %2103, %2102 ], [ %2105, %2104 ], [ %2108, %2107 ], [ %2110, %2109 ], !dbg !33 + %2111 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i31 = icmp eq i32 %2111, 0, !dbg !33 + %2112 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i34 = icmp eq i32 %2112, 0, !dbg !33 + br i1 %.not.i31, label %2118, label %2113, !dbg !33 + +2113: ; preds = %__nv_sqrtf.exit30 + br i1 %.not1.i34, label %2116, label %2114, !dbg !33 + +2114: ; preds = %2113 + %2115 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2002) #7, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +2116: ; preds = %2113 + %2117 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2002) #7, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +2118: ; preds = %__nv_sqrtf.exit30 + br i1 %.not1.i34, label %2121, label %2119, !dbg !33 + +2119: ; preds = %2118 + %2120 = tail call float @llvm.nvvm.sqrt.rn.f(float %2002) #7, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +2121: ; preds = %2118 + %2122 = tail call float @llvm.nvvm.sqrt.approx.f(float %2002) #7, !dbg !33 + br label %__nv_sqrtf.exit35, !dbg !33 + +__nv_sqrtf.exit35: ; preds = %2114, %2116, %2119, %2121 + %.0.i33 = phi float [ %2115, %2114 ], [ %2117, %2116 ], [ %2120, %2119 ], [ %2122, %2121 ], !dbg !33 + %2123 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i36 = icmp eq i32 %2123, 0, !dbg !33 + %2124 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i39 = icmp eq i32 %2124, 0, !dbg !33 + br i1 %.not.i36, label %2130, label %2125, !dbg !33 + +2125: ; preds = %__nv_sqrtf.exit35 + br i1 %.not1.i39, label %2128, label %2126, !dbg !33 + +2126: ; preds = %2125 + %2127 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2003) #7, !dbg !33 + br label %__nv_sqrtf.exit40, !dbg !33 + +2128: ; preds = %2125 + %2129 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2003) #7, !dbg !33 + br label %__nv_sqrtf.exit40, !dbg !33 + +2130: ; preds = %__nv_sqrtf.exit35 + br i1 %.not1.i39, label %2133, label %2131, !dbg !33 + +2131: ; preds = %2130 + %2132 = tail call float @llvm.nvvm.sqrt.rn.f(float %2003) #7, !dbg !33 + br label %__nv_sqrtf.exit40, !dbg !33 + +2133: ; preds = %2130 + %2134 = tail call float @llvm.nvvm.sqrt.approx.f(float %2003) #7, !dbg !33 + br label %__nv_sqrtf.exit40, !dbg !33 + +__nv_sqrtf.exit40: ; preds = %2126, %2128, %2131, %2133 + %.0.i38 = phi float [ %2127, %2126 ], [ %2129, %2128 ], [ %2132, %2131 ], [ %2134, %2133 ], !dbg !33 + %2135 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i41 = icmp eq i32 %2135, 0, !dbg !33 + %2136 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i44 = icmp eq i32 %2136, 0, !dbg !33 + br i1 %.not.i41, label %2142, label %2137, !dbg !33 + +2137: ; preds = %__nv_sqrtf.exit40 + br i1 %.not1.i44, label %2140, label %2138, !dbg !33 + +2138: ; preds = %2137 + %2139 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2004) #7, !dbg !33 + br label %__nv_sqrtf.exit45, !dbg !33 + +2140: ; preds = %2137 + %2141 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2004) #7, !dbg !33 + br label %__nv_sqrtf.exit45, !dbg !33 + +2142: ; preds = %__nv_sqrtf.exit40 + br i1 %.not1.i44, label %2145, label %2143, !dbg !33 + +2143: ; preds = %2142 + %2144 = tail call float @llvm.nvvm.sqrt.rn.f(float %2004) #7, !dbg !33 + br label %__nv_sqrtf.exit45, !dbg !33 + +2145: ; preds = %2142 + %2146 = tail call float @llvm.nvvm.sqrt.approx.f(float %2004) #7, !dbg !33 + br label %__nv_sqrtf.exit45, !dbg !33 + +__nv_sqrtf.exit45: ; preds = %2138, %2140, %2143, %2145 + %.0.i43 = phi float [ %2139, %2138 ], [ %2141, %2140 ], [ %2144, %2143 ], [ %2146, %2145 ], !dbg !33 + %2147 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i46 = icmp eq i32 %2147, 0, !dbg !33 + %2148 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i49 = icmp eq i32 %2148, 0, !dbg !33 + br i1 %.not.i46, label %2154, label %2149, !dbg !33 + +2149: ; preds = %__nv_sqrtf.exit45 + br i1 %.not1.i49, label %2152, label %2150, !dbg !33 + +2150: ; preds = %2149 + %2151 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2005) #7, !dbg !33 + br label %__nv_sqrtf.exit50, !dbg !33 + +2152: ; preds = %2149 + %2153 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2005) #7, !dbg !33 + br label %__nv_sqrtf.exit50, !dbg !33 + +2154: ; preds = %__nv_sqrtf.exit45 + br i1 %.not1.i49, label %2157, label %2155, !dbg !33 + +2155: ; preds = %2154 + %2156 = tail call float @llvm.nvvm.sqrt.rn.f(float %2005) #7, !dbg !33 + br label %__nv_sqrtf.exit50, !dbg !33 + +2157: ; preds = %2154 + %2158 = tail call float @llvm.nvvm.sqrt.approx.f(float %2005) #7, !dbg !33 + br label %__nv_sqrtf.exit50, !dbg !33 + +__nv_sqrtf.exit50: ; preds = %2150, %2152, %2155, %2157 + %.0.i48 = phi float [ %2151, %2150 ], [ %2153, %2152 ], [ %2156, %2155 ], [ %2158, %2157 ], !dbg !33 + %2159 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i51 = icmp eq i32 %2159, 0, !dbg !33 + %2160 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i54 = icmp eq i32 %2160, 0, !dbg !33 + br i1 %.not.i51, label %2166, label %2161, !dbg !33 + +2161: ; preds = %__nv_sqrtf.exit50 + br i1 %.not1.i54, label %2164, label %2162, !dbg !33 + +2162: ; preds = %2161 + %2163 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2006) #7, !dbg !33 + br label %__nv_sqrtf.exit55, !dbg !33 + +2164: ; preds = %2161 + %2165 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2006) #7, !dbg !33 + br label %__nv_sqrtf.exit55, !dbg !33 + +2166: ; preds = %__nv_sqrtf.exit50 + br i1 %.not1.i54, label %2169, label %2167, !dbg !33 + +2167: ; preds = %2166 + %2168 = tail call float @llvm.nvvm.sqrt.rn.f(float %2006) #7, !dbg !33 + br label %__nv_sqrtf.exit55, !dbg !33 + +2169: ; preds = %2166 + %2170 = tail call float @llvm.nvvm.sqrt.approx.f(float %2006) #7, !dbg !33 + br label %__nv_sqrtf.exit55, !dbg !33 + +__nv_sqrtf.exit55: ; preds = %2162, %2164, %2167, %2169 + %.0.i53 = phi float [ %2163, %2162 ], [ %2165, %2164 ], [ %2168, %2167 ], [ %2170, %2169 ], !dbg !33 + %2171 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i56 = icmp eq i32 %2171, 0, !dbg !33 + %2172 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i59 = icmp eq i32 %2172, 0, !dbg !33 + br i1 %.not.i56, label %2178, label %2173, !dbg !33 + +2173: ; preds = %__nv_sqrtf.exit55 + br i1 %.not1.i59, label %2176, label %2174, !dbg !33 + +2174: ; preds = %2173 + %2175 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2007) #7, !dbg !33 + br label %__nv_sqrtf.exit60, !dbg !33 + +2176: ; preds = %2173 + %2177 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2007) #7, !dbg !33 + br label %__nv_sqrtf.exit60, !dbg !33 + +2178: ; preds = %__nv_sqrtf.exit55 + br i1 %.not1.i59, label %2181, label %2179, !dbg !33 + +2179: ; preds = %2178 + %2180 = tail call float @llvm.nvvm.sqrt.rn.f(float %2007) #7, !dbg !33 + br label %__nv_sqrtf.exit60, !dbg !33 + +2181: ; preds = %2178 + %2182 = tail call float @llvm.nvvm.sqrt.approx.f(float %2007) #7, !dbg !33 + br label %__nv_sqrtf.exit60, !dbg !33 + +__nv_sqrtf.exit60: ; preds = %2174, %2176, %2179, %2181 + %.0.i58 = phi float [ %2175, %2174 ], [ %2177, %2176 ], [ %2180, %2179 ], [ %2182, %2181 ], !dbg !33 + %2183 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i61 = icmp eq i32 %2183, 0, !dbg !33 + %2184 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i64 = icmp eq i32 %2184, 0, !dbg !33 + br i1 %.not.i61, label %2190, label %2185, !dbg !33 + +2185: ; preds = %__nv_sqrtf.exit60 + br i1 %.not1.i64, label %2188, label %2186, !dbg !33 + +2186: ; preds = %2185 + %2187 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2008) #7, !dbg !33 + br label %__nv_sqrtf.exit65, !dbg !33 + +2188: ; preds = %2185 + %2189 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2008) #7, !dbg !33 + br label %__nv_sqrtf.exit65, !dbg !33 + +2190: ; preds = %__nv_sqrtf.exit60 + br i1 %.not1.i64, label %2193, label %2191, !dbg !33 + +2191: ; preds = %2190 + %2192 = tail call float @llvm.nvvm.sqrt.rn.f(float %2008) #7, !dbg !33 + br label %__nv_sqrtf.exit65, !dbg !33 + +2193: ; preds = %2190 + %2194 = tail call float @llvm.nvvm.sqrt.approx.f(float %2008) #7, !dbg !33 + br label %__nv_sqrtf.exit65, !dbg !33 + +__nv_sqrtf.exit65: ; preds = %2186, %2188, %2191, %2193 + %.0.i63 = phi float [ %2187, %2186 ], [ %2189, %2188 ], [ %2192, %2191 ], [ %2194, %2193 ], !dbg !33 + %2195 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i66 = icmp eq i32 %2195, 0, !dbg !33 + %2196 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i69 = icmp eq i32 %2196, 0, !dbg !33 + br i1 %.not.i66, label %2202, label %2197, !dbg !33 + +2197: ; preds = %__nv_sqrtf.exit65 + br i1 %.not1.i69, label %2200, label %2198, !dbg !33 + +2198: ; preds = %2197 + %2199 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2009) #7, !dbg !33 + br label %__nv_sqrtf.exit70, !dbg !33 + +2200: ; preds = %2197 + %2201 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2009) #7, !dbg !33 + br label %__nv_sqrtf.exit70, !dbg !33 + +2202: ; preds = %__nv_sqrtf.exit65 + br i1 %.not1.i69, label %2205, label %2203, !dbg !33 + +2203: ; preds = %2202 + %2204 = tail call float @llvm.nvvm.sqrt.rn.f(float %2009) #7, !dbg !33 + br label %__nv_sqrtf.exit70, !dbg !33 + +2205: ; preds = %2202 + %2206 = tail call float @llvm.nvvm.sqrt.approx.f(float %2009) #7, !dbg !33 + br label %__nv_sqrtf.exit70, !dbg !33 + +__nv_sqrtf.exit70: ; preds = %2198, %2200, %2203, %2205 + %.0.i68 = phi float [ %2199, %2198 ], [ %2201, %2200 ], [ %2204, %2203 ], [ %2206, %2205 ], !dbg !33 + %2207 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i71 = icmp eq i32 %2207, 0, !dbg !33 + %2208 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i74 = icmp eq i32 %2208, 0, !dbg !33 + br i1 %.not.i71, label %2214, label %2209, !dbg !33 + +2209: ; preds = %__nv_sqrtf.exit70 + br i1 %.not1.i74, label %2212, label %2210, !dbg !33 + +2210: ; preds = %2209 + %2211 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2010) #7, !dbg !33 + br label %__nv_sqrtf.exit75, !dbg !33 + +2212: ; preds = %2209 + %2213 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2010) #7, !dbg !33 + br label %__nv_sqrtf.exit75, !dbg !33 + +2214: ; preds = %__nv_sqrtf.exit70 + br i1 %.not1.i74, label %2217, label %2215, !dbg !33 + +2215: ; preds = %2214 + %2216 = tail call float @llvm.nvvm.sqrt.rn.f(float %2010) #7, !dbg !33 + br label %__nv_sqrtf.exit75, !dbg !33 + +2217: ; preds = %2214 + %2218 = tail call float @llvm.nvvm.sqrt.approx.f(float %2010) #7, !dbg !33 + br label %__nv_sqrtf.exit75, !dbg !33 + +__nv_sqrtf.exit75: ; preds = %2210, %2212, %2215, %2217 + %.0.i73 = phi float [ %2211, %2210 ], [ %2213, %2212 ], [ %2216, %2215 ], [ %2218, %2217 ], !dbg !33 + %2219 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i76 = icmp eq i32 %2219, 0, !dbg !33 + %2220 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i79 = icmp eq i32 %2220, 0, !dbg !33 + br i1 %.not.i76, label %2226, label %2221, !dbg !33 + +2221: ; preds = %__nv_sqrtf.exit75 + br i1 %.not1.i79, label %2224, label %2222, !dbg !33 + +2222: ; preds = %2221 + %2223 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2011) #7, !dbg !33 + br label %__nv_sqrtf.exit80, !dbg !33 + +2224: ; preds = %2221 + %2225 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2011) #7, !dbg !33 + br label %__nv_sqrtf.exit80, !dbg !33 + +2226: ; preds = %__nv_sqrtf.exit75 + br i1 %.not1.i79, label %2229, label %2227, !dbg !33 + +2227: ; preds = %2226 + %2228 = tail call float @llvm.nvvm.sqrt.rn.f(float %2011) #7, !dbg !33 + br label %__nv_sqrtf.exit80, !dbg !33 + +2229: ; preds = %2226 + %2230 = tail call float @llvm.nvvm.sqrt.approx.f(float %2011) #7, !dbg !33 + br label %__nv_sqrtf.exit80, !dbg !33 + +__nv_sqrtf.exit80: ; preds = %2222, %2224, %2227, %2229 + %.0.i78 = phi float [ %2223, %2222 ], [ %2225, %2224 ], [ %2228, %2227 ], [ %2230, %2229 ], !dbg !33 + %2231 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i81 = icmp eq i32 %2231, 0, !dbg !33 + %2232 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i84 = icmp eq i32 %2232, 0, !dbg !33 + br i1 %.not.i81, label %2238, label %2233, !dbg !33 + +2233: ; preds = %__nv_sqrtf.exit80 + br i1 %.not1.i84, label %2236, label %2234, !dbg !33 + +2234: ; preds = %2233 + %2235 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2012) #7, !dbg !33 + br label %__nv_sqrtf.exit85, !dbg !33 + +2236: ; preds = %2233 + %2237 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2012) #7, !dbg !33 + br label %__nv_sqrtf.exit85, !dbg !33 + +2238: ; preds = %__nv_sqrtf.exit80 + br i1 %.not1.i84, label %2241, label %2239, !dbg !33 + +2239: ; preds = %2238 + %2240 = tail call float @llvm.nvvm.sqrt.rn.f(float %2012) #7, !dbg !33 + br label %__nv_sqrtf.exit85, !dbg !33 + +2241: ; preds = %2238 + %2242 = tail call float @llvm.nvvm.sqrt.approx.f(float %2012) #7, !dbg !33 + br label %__nv_sqrtf.exit85, !dbg !33 + +__nv_sqrtf.exit85: ; preds = %2234, %2236, %2239, %2241 + %.0.i83 = phi float [ %2235, %2234 ], [ %2237, %2236 ], [ %2240, %2239 ], [ %2242, %2241 ], !dbg !33 + %2243 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i86 = icmp eq i32 %2243, 0, !dbg !33 + %2244 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i89 = icmp eq i32 %2244, 0, !dbg !33 + br i1 %.not.i86, label %2250, label %2245, !dbg !33 + +2245: ; preds = %__nv_sqrtf.exit85 + br i1 %.not1.i89, label %2248, label %2246, !dbg !33 + +2246: ; preds = %2245 + %2247 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2013) #7, !dbg !33 + br label %__nv_sqrtf.exit90, !dbg !33 + +2248: ; preds = %2245 + %2249 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2013) #7, !dbg !33 + br label %__nv_sqrtf.exit90, !dbg !33 + +2250: ; preds = %__nv_sqrtf.exit85 + br i1 %.not1.i89, label %2253, label %2251, !dbg !33 + +2251: ; preds = %2250 + %2252 = tail call float @llvm.nvvm.sqrt.rn.f(float %2013) #7, !dbg !33 + br label %__nv_sqrtf.exit90, !dbg !33 + +2253: ; preds = %2250 + %2254 = tail call float @llvm.nvvm.sqrt.approx.f(float %2013) #7, !dbg !33 + br label %__nv_sqrtf.exit90, !dbg !33 + +__nv_sqrtf.exit90: ; preds = %2246, %2248, %2251, %2253 + %.0.i88 = phi float [ %2247, %2246 ], [ %2249, %2248 ], [ %2252, %2251 ], [ %2254, %2253 ], !dbg !33 + %2255 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i91 = icmp eq i32 %2255, 0, !dbg !33 + %2256 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i94 = icmp eq i32 %2256, 0, !dbg !33 + br i1 %.not.i91, label %2262, label %2257, !dbg !33 + +2257: ; preds = %__nv_sqrtf.exit90 + br i1 %.not1.i94, label %2260, label %2258, !dbg !33 + +2258: ; preds = %2257 + %2259 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2014) #7, !dbg !33 + br label %__nv_sqrtf.exit95, !dbg !33 + +2260: ; preds = %2257 + %2261 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2014) #7, !dbg !33 + br label %__nv_sqrtf.exit95, !dbg !33 + +2262: ; preds = %__nv_sqrtf.exit90 + br i1 %.not1.i94, label %2265, label %2263, !dbg !33 + +2263: ; preds = %2262 + %2264 = tail call float @llvm.nvvm.sqrt.rn.f(float %2014) #7, !dbg !33 + br label %__nv_sqrtf.exit95, !dbg !33 + +2265: ; preds = %2262 + %2266 = tail call float @llvm.nvvm.sqrt.approx.f(float %2014) #7, !dbg !33 + br label %__nv_sqrtf.exit95, !dbg !33 + +__nv_sqrtf.exit95: ; preds = %2258, %2260, %2263, %2265 + %.0.i93 = phi float [ %2259, %2258 ], [ %2261, %2260 ], [ %2264, %2263 ], [ %2266, %2265 ], !dbg !33 + %2267 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i96 = icmp eq i32 %2267, 0, !dbg !33 + %2268 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i99 = icmp eq i32 %2268, 0, !dbg !33 + br i1 %.not.i96, label %2274, label %2269, !dbg !33 + +2269: ; preds = %__nv_sqrtf.exit95 + br i1 %.not1.i99, label %2272, label %2270, !dbg !33 + +2270: ; preds = %2269 + %2271 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2015) #7, !dbg !33 + br label %__nv_sqrtf.exit100, !dbg !33 + +2272: ; preds = %2269 + %2273 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2015) #7, !dbg !33 + br label %__nv_sqrtf.exit100, !dbg !33 + +2274: ; preds = %__nv_sqrtf.exit95 + br i1 %.not1.i99, label %2277, label %2275, !dbg !33 + +2275: ; preds = %2274 + %2276 = tail call float @llvm.nvvm.sqrt.rn.f(float %2015) #7, !dbg !33 + br label %__nv_sqrtf.exit100, !dbg !33 + +2277: ; preds = %2274 + %2278 = tail call float @llvm.nvvm.sqrt.approx.f(float %2015) #7, !dbg !33 + br label %__nv_sqrtf.exit100, !dbg !33 + +__nv_sqrtf.exit100: ; preds = %2270, %2272, %2275, %2277 + %.0.i98 = phi float [ %2271, %2270 ], [ %2273, %2272 ], [ %2276, %2275 ], [ %2278, %2277 ], !dbg !33 + %2279 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i101 = icmp eq i32 %2279, 0, !dbg !33 + %2280 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i104 = icmp eq i32 %2280, 0, !dbg !33 + br i1 %.not.i101, label %2286, label %2281, !dbg !33 + +2281: ; preds = %__nv_sqrtf.exit100 + br i1 %.not1.i104, label %2284, label %2282, !dbg !33 + +2282: ; preds = %2281 + %2283 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2016) #7, !dbg !33 + br label %__nv_sqrtf.exit105, !dbg !33 + +2284: ; preds = %2281 + %2285 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2016) #7, !dbg !33 + br label %__nv_sqrtf.exit105, !dbg !33 + +2286: ; preds = %__nv_sqrtf.exit100 + br i1 %.not1.i104, label %2289, label %2287, !dbg !33 + +2287: ; preds = %2286 + %2288 = tail call float @llvm.nvvm.sqrt.rn.f(float %2016) #7, !dbg !33 + br label %__nv_sqrtf.exit105, !dbg !33 + +2289: ; preds = %2286 + %2290 = tail call float @llvm.nvvm.sqrt.approx.f(float %2016) #7, !dbg !33 + br label %__nv_sqrtf.exit105, !dbg !33 + +__nv_sqrtf.exit105: ; preds = %2282, %2284, %2287, %2289 + %.0.i103 = phi float [ %2283, %2282 ], [ %2285, %2284 ], [ %2288, %2287 ], [ %2290, %2289 ], !dbg !33 + %2291 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i106 = icmp eq i32 %2291, 0, !dbg !33 + %2292 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i109 = icmp eq i32 %2292, 0, !dbg !33 + br i1 %.not.i106, label %2298, label %2293, !dbg !33 + +2293: ; preds = %__nv_sqrtf.exit105 + br i1 %.not1.i109, label %2296, label %2294, !dbg !33 + +2294: ; preds = %2293 + %2295 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2017) #7, !dbg !33 + br label %__nv_sqrtf.exit110, !dbg !33 + +2296: ; preds = %2293 + %2297 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2017) #7, !dbg !33 + br label %__nv_sqrtf.exit110, !dbg !33 + +2298: ; preds = %__nv_sqrtf.exit105 + br i1 %.not1.i109, label %2301, label %2299, !dbg !33 + +2299: ; preds = %2298 + %2300 = tail call float @llvm.nvvm.sqrt.rn.f(float %2017) #7, !dbg !33 + br label %__nv_sqrtf.exit110, !dbg !33 + +2301: ; preds = %2298 + %2302 = tail call float @llvm.nvvm.sqrt.approx.f(float %2017) #7, !dbg !33 + br label %__nv_sqrtf.exit110, !dbg !33 + +__nv_sqrtf.exit110: ; preds = %2294, %2296, %2299, %2301 + %.0.i108 = phi float [ %2295, %2294 ], [ %2297, %2296 ], [ %2300, %2299 ], [ %2302, %2301 ], !dbg !33 + %2303 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i111 = icmp eq i32 %2303, 0, !dbg !33 + %2304 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i114 = icmp eq i32 %2304, 0, !dbg !33 + br i1 %.not.i111, label %2310, label %2305, !dbg !33 + +2305: ; preds = %__nv_sqrtf.exit110 + br i1 %.not1.i114, label %2308, label %2306, !dbg !33 + +2306: ; preds = %2305 + %2307 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2018) #7, !dbg !33 + br label %__nv_sqrtf.exit115, !dbg !33 + +2308: ; preds = %2305 + %2309 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2018) #7, !dbg !33 + br label %__nv_sqrtf.exit115, !dbg !33 + +2310: ; preds = %__nv_sqrtf.exit110 + br i1 %.not1.i114, label %2313, label %2311, !dbg !33 + +2311: ; preds = %2310 + %2312 = tail call float @llvm.nvvm.sqrt.rn.f(float %2018) #7, !dbg !33 + br label %__nv_sqrtf.exit115, !dbg !33 + +2313: ; preds = %2310 + %2314 = tail call float @llvm.nvvm.sqrt.approx.f(float %2018) #7, !dbg !33 + br label %__nv_sqrtf.exit115, !dbg !33 + +__nv_sqrtf.exit115: ; preds = %2306, %2308, %2311, %2313 + %.0.i113 = phi float [ %2307, %2306 ], [ %2309, %2308 ], [ %2312, %2311 ], [ %2314, %2313 ], !dbg !33 + %2315 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i116 = icmp eq i32 %2315, 0, !dbg !33 + %2316 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i119 = icmp eq i32 %2316, 0, !dbg !33 + br i1 %.not.i116, label %2322, label %2317, !dbg !33 + +2317: ; preds = %__nv_sqrtf.exit115 + br i1 %.not1.i119, label %2320, label %2318, !dbg !33 + +2318: ; preds = %2317 + %2319 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2019) #7, !dbg !33 + br label %__nv_sqrtf.exit120, !dbg !33 + +2320: ; preds = %2317 + %2321 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2019) #7, !dbg !33 + br label %__nv_sqrtf.exit120, !dbg !33 + +2322: ; preds = %__nv_sqrtf.exit115 + br i1 %.not1.i119, label %2325, label %2323, !dbg !33 + +2323: ; preds = %2322 + %2324 = tail call float @llvm.nvvm.sqrt.rn.f(float %2019) #7, !dbg !33 + br label %__nv_sqrtf.exit120, !dbg !33 + +2325: ; preds = %2322 + %2326 = tail call float @llvm.nvvm.sqrt.approx.f(float %2019) #7, !dbg !33 + br label %__nv_sqrtf.exit120, !dbg !33 + +__nv_sqrtf.exit120: ; preds = %2318, %2320, %2323, %2325 + %.0.i118 = phi float [ %2319, %2318 ], [ %2321, %2320 ], [ %2324, %2323 ], [ %2326, %2325 ], !dbg !33 + %2327 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i121 = icmp eq i32 %2327, 0, !dbg !33 + %2328 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i124 = icmp eq i32 %2328, 0, !dbg !33 + br i1 %.not.i121, label %2334, label %2329, !dbg !33 + +2329: ; preds = %__nv_sqrtf.exit120 + br i1 %.not1.i124, label %2332, label %2330, !dbg !33 + +2330: ; preds = %2329 + %2331 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2020) #7, !dbg !33 + br label %__nv_sqrtf.exit125, !dbg !33 + +2332: ; preds = %2329 + %2333 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2020) #7, !dbg !33 + br label %__nv_sqrtf.exit125, !dbg !33 + +2334: ; preds = %__nv_sqrtf.exit120 + br i1 %.not1.i124, label %2337, label %2335, !dbg !33 + +2335: ; preds = %2334 + %2336 = tail call float @llvm.nvvm.sqrt.rn.f(float %2020) #7, !dbg !33 + br label %__nv_sqrtf.exit125, !dbg !33 + +2337: ; preds = %2334 + %2338 = tail call float @llvm.nvvm.sqrt.approx.f(float %2020) #7, !dbg !33 + br label %__nv_sqrtf.exit125, !dbg !33 + +__nv_sqrtf.exit125: ; preds = %2330, %2332, %2335, %2337 + %.0.i123 = phi float [ %2331, %2330 ], [ %2333, %2332 ], [ %2336, %2335 ], [ %2338, %2337 ], !dbg !33 + %2339 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i126 = icmp eq i32 %2339, 0, !dbg !33 + %2340 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i129 = icmp eq i32 %2340, 0, !dbg !33 + br i1 %.not.i126, label %2346, label %2341, !dbg !33 + +2341: ; preds = %__nv_sqrtf.exit125 + br i1 %.not1.i129, label %2344, label %2342, !dbg !33 + +2342: ; preds = %2341 + %2343 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2021) #7, !dbg !33 + br label %__nv_sqrtf.exit130, !dbg !33 + +2344: ; preds = %2341 + %2345 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2021) #7, !dbg !33 + br label %__nv_sqrtf.exit130, !dbg !33 + +2346: ; preds = %__nv_sqrtf.exit125 + br i1 %.not1.i129, label %2349, label %2347, !dbg !33 + +2347: ; preds = %2346 + %2348 = tail call float @llvm.nvvm.sqrt.rn.f(float %2021) #7, !dbg !33 + br label %__nv_sqrtf.exit130, !dbg !33 + +2349: ; preds = %2346 + %2350 = tail call float @llvm.nvvm.sqrt.approx.f(float %2021) #7, !dbg !33 + br label %__nv_sqrtf.exit130, !dbg !33 + +__nv_sqrtf.exit130: ; preds = %2342, %2344, %2347, %2349 + %.0.i128 = phi float [ %2343, %2342 ], [ %2345, %2344 ], [ %2348, %2347 ], [ %2350, %2349 ], !dbg !33 + %2351 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i131 = icmp eq i32 %2351, 0, !dbg !33 + %2352 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i134 = icmp eq i32 %2352, 0, !dbg !33 + br i1 %.not.i131, label %2358, label %2353, !dbg !33 + +2353: ; preds = %__nv_sqrtf.exit130 + br i1 %.not1.i134, label %2356, label %2354, !dbg !33 + +2354: ; preds = %2353 + %2355 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2022) #7, !dbg !33 + br label %__nv_sqrtf.exit135, !dbg !33 + +2356: ; preds = %2353 + %2357 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2022) #7, !dbg !33 + br label %__nv_sqrtf.exit135, !dbg !33 + +2358: ; preds = %__nv_sqrtf.exit130 + br i1 %.not1.i134, label %2361, label %2359, !dbg !33 + +2359: ; preds = %2358 + %2360 = tail call float @llvm.nvvm.sqrt.rn.f(float %2022) #7, !dbg !33 + br label %__nv_sqrtf.exit135, !dbg !33 + +2361: ; preds = %2358 + %2362 = tail call float @llvm.nvvm.sqrt.approx.f(float %2022) #7, !dbg !33 + br label %__nv_sqrtf.exit135, !dbg !33 + +__nv_sqrtf.exit135: ; preds = %2354, %2356, %2359, %2361 + %.0.i133 = phi float [ %2355, %2354 ], [ %2357, %2356 ], [ %2360, %2359 ], [ %2362, %2361 ], !dbg !33 + %2363 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i136 = icmp eq i32 %2363, 0, !dbg !33 + %2364 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i139 = icmp eq i32 %2364, 0, !dbg !33 + br i1 %.not.i136, label %2370, label %2365, !dbg !33 + +2365: ; preds = %__nv_sqrtf.exit135 + br i1 %.not1.i139, label %2368, label %2366, !dbg !33 + +2366: ; preds = %2365 + %2367 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2023) #7, !dbg !33 + br label %__nv_sqrtf.exit140, !dbg !33 + +2368: ; preds = %2365 + %2369 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2023) #7, !dbg !33 + br label %__nv_sqrtf.exit140, !dbg !33 + +2370: ; preds = %__nv_sqrtf.exit135 + br i1 %.not1.i139, label %2373, label %2371, !dbg !33 + +2371: ; preds = %2370 + %2372 = tail call float @llvm.nvvm.sqrt.rn.f(float %2023) #7, !dbg !33 + br label %__nv_sqrtf.exit140, !dbg !33 + +2373: ; preds = %2370 + %2374 = tail call float @llvm.nvvm.sqrt.approx.f(float %2023) #7, !dbg !33 + br label %__nv_sqrtf.exit140, !dbg !33 + +__nv_sqrtf.exit140: ; preds = %2366, %2368, %2371, %2373 + %.0.i138 = phi float [ %2367, %2366 ], [ %2369, %2368 ], [ %2372, %2371 ], [ %2374, %2373 ], !dbg !33 + %2375 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i141 = icmp eq i32 %2375, 0, !dbg !33 + %2376 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i144 = icmp eq i32 %2376, 0, !dbg !33 + br i1 %.not.i141, label %2382, label %2377, !dbg !33 + +2377: ; preds = %__nv_sqrtf.exit140 + br i1 %.not1.i144, label %2380, label %2378, !dbg !33 + +2378: ; preds = %2377 + %2379 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2024) #7, !dbg !33 + br label %__nv_sqrtf.exit145, !dbg !33 + +2380: ; preds = %2377 + %2381 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2024) #7, !dbg !33 + br label %__nv_sqrtf.exit145, !dbg !33 + +2382: ; preds = %__nv_sqrtf.exit140 + br i1 %.not1.i144, label %2385, label %2383, !dbg !33 + +2383: ; preds = %2382 + %2384 = tail call float @llvm.nvvm.sqrt.rn.f(float %2024) #7, !dbg !33 + br label %__nv_sqrtf.exit145, !dbg !33 + +2385: ; preds = %2382 + %2386 = tail call float @llvm.nvvm.sqrt.approx.f(float %2024) #7, !dbg !33 + br label %__nv_sqrtf.exit145, !dbg !33 + +__nv_sqrtf.exit145: ; preds = %2378, %2380, %2383, %2385 + %.0.i143 = phi float [ %2379, %2378 ], [ %2381, %2380 ], [ %2384, %2383 ], [ %2386, %2385 ], !dbg !33 + %2387 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i146 = icmp eq i32 %2387, 0, !dbg !33 + %2388 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i149 = icmp eq i32 %2388, 0, !dbg !33 + br i1 %.not.i146, label %2394, label %2389, !dbg !33 + +2389: ; preds = %__nv_sqrtf.exit145 + br i1 %.not1.i149, label %2392, label %2390, !dbg !33 + +2390: ; preds = %2389 + %2391 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2025) #7, !dbg !33 + br label %__nv_sqrtf.exit150, !dbg !33 + +2392: ; preds = %2389 + %2393 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2025) #7, !dbg !33 + br label %__nv_sqrtf.exit150, !dbg !33 + +2394: ; preds = %__nv_sqrtf.exit145 + br i1 %.not1.i149, label %2397, label %2395, !dbg !33 + +2395: ; preds = %2394 + %2396 = tail call float @llvm.nvvm.sqrt.rn.f(float %2025) #7, !dbg !33 + br label %__nv_sqrtf.exit150, !dbg !33 + +2397: ; preds = %2394 + %2398 = tail call float @llvm.nvvm.sqrt.approx.f(float %2025) #7, !dbg !33 + br label %__nv_sqrtf.exit150, !dbg !33 + +__nv_sqrtf.exit150: ; preds = %2390, %2392, %2395, %2397 + %.0.i148 = phi float [ %2391, %2390 ], [ %2393, %2392 ], [ %2396, %2395 ], [ %2398, %2397 ], !dbg !33 + %2399 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #7, !dbg !33 + %.not.i151 = icmp eq i32 %2399, 0, !dbg !33 + %2400 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #7, !dbg !33 + %.not1.i154 = icmp eq i32 %2400, 0, !dbg !33 + br i1 %.not.i151, label %2406, label %2401, !dbg !33 + +2401: ; preds = %__nv_sqrtf.exit150 + br i1 %.not1.i154, label %2404, label %2402, !dbg !33 + +2402: ; preds = %2401 + %2403 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %2026) #7, !dbg !33 + br label %__nv_sqrtf.exit155, !dbg !33 + +2404: ; preds = %2401 + %2405 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %2026) #7, !dbg !33 + br label %__nv_sqrtf.exit155, !dbg !33 + +2406: ; preds = %__nv_sqrtf.exit150 + br i1 %.not1.i154, label %2409, label %2407, !dbg !33 + +2407: ; preds = %2406 + %2408 = tail call float @llvm.nvvm.sqrt.rn.f(float %2026) #7, !dbg !33 + br label %__nv_sqrtf.exit155, !dbg !33 + +2409: ; preds = %2406 + %2410 = tail call float @llvm.nvvm.sqrt.approx.f(float %2026) #7, !dbg !33 + br label %__nv_sqrtf.exit155, !dbg !33 + +__nv_sqrtf.exit155: ; preds = %2402, %2404, %2407, %2409 + %.0.i153 = phi float [ %2403, %2402 ], [ %2405, %2404 ], [ %2408, %2407 ], [ %2410, %2409 ], !dbg !33 + %2411 = sext i32 %4 to i64, !dbg !18 + %2412 = shl nuw nsw i32 %11, 1, !dbg !16 + %2413 = and i32 %2412, 62, !dbg !16 + %2414 = or disjoint i32 %2413, 1, !dbg !16 + %2415 = zext nneg i32 %2414 to i64, !dbg !17 + %2416 = or disjoint i64 %10, %2415, !dbg !17 + %2417 = getelementptr float, ptr addrspace(1) %2, i64 %2416, !dbg !34 + %2418 = zext nneg i32 %2413 to i64, !dbg !17 + %2419 = or disjoint i64 %10, %2418, !dbg !17 + %2420 = getelementptr float, ptr addrspace(1) %2, i64 %2419, !dbg !34 + %2421 = getelementptr bfloat, ptr addrspace(1) %1, i64 %105, !dbg !35 + %2422 = getelementptr bfloat, ptr addrspace(1) %2421, i64 %138, !dbg !36 + %2423 = getelementptr bfloat, ptr addrspace(1) %1, i64 %103, !dbg !35 + %2424 = getelementptr bfloat, ptr addrspace(1) %2423, i64 %138, !dbg !36 + %2425 = getelementptr bfloat, ptr addrspace(1) %1, i64 %101, !dbg !35 + %2426 = getelementptr bfloat, ptr addrspace(1) %2425, i64 %138, !dbg !36 + %2427 = getelementptr bfloat, ptr addrspace(1) %1, i64 %99, !dbg !35 + %2428 = getelementptr bfloat, ptr addrspace(1) %2427, i64 %138, !dbg !36 + %2429 = getelementptr bfloat, ptr addrspace(1) %1, i64 %97, !dbg !35 + %2430 = getelementptr bfloat, ptr addrspace(1) %2429, i64 %138, !dbg !36 + %2431 = getelementptr bfloat, ptr addrspace(1) %1, i64 %95, !dbg !35 + %2432 = getelementptr bfloat, ptr addrspace(1) %2431, i64 %138, !dbg !36 + %2433 = getelementptr bfloat, ptr addrspace(1) %1, i64 %93, !dbg !35 + %2434 = getelementptr bfloat, ptr addrspace(1) %2433, i64 %138, !dbg !36 + %2435 = getelementptr bfloat, ptr addrspace(1) %1, i64 %91, !dbg !35 + %2436 = getelementptr bfloat, ptr addrspace(1) %2435, i64 %138, !dbg !36 + %2437 = getelementptr bfloat, ptr addrspace(1) %1, i64 %89, !dbg !35 + %2438 = getelementptr bfloat, ptr addrspace(1) %2437, i64 %138, !dbg !36 + %2439 = getelementptr bfloat, ptr addrspace(1) %1, i64 %87, !dbg !35 + %2440 = getelementptr bfloat, ptr addrspace(1) %2439, i64 %138, !dbg !36 + %2441 = getelementptr bfloat, ptr addrspace(1) %1, i64 %85, !dbg !35 + %2442 = getelementptr bfloat, ptr addrspace(1) %2441, i64 %138, !dbg !36 + %2443 = getelementptr bfloat, ptr addrspace(1) %1, i64 %83, !dbg !35 + %2444 = getelementptr bfloat, ptr addrspace(1) %2443, i64 %138, !dbg !36 + %2445 = getelementptr bfloat, ptr addrspace(1) %1, i64 %81, !dbg !35 + %2446 = getelementptr bfloat, ptr addrspace(1) %2445, i64 %138, !dbg !36 + %2447 = getelementptr bfloat, ptr addrspace(1) %1, i64 %79, !dbg !35 + %2448 = getelementptr bfloat, ptr addrspace(1) %2447, i64 %138, !dbg !36 + %2449 = getelementptr bfloat, ptr addrspace(1) %1, i64 %77, !dbg !35 + %2450 = getelementptr bfloat, ptr addrspace(1) %2449, i64 %138, !dbg !36 + %2451 = getelementptr bfloat, ptr addrspace(1) %1, i64 %75, !dbg !35 + %2452 = getelementptr bfloat, ptr addrspace(1) %2451, i64 %138, !dbg !36 + %2453 = getelementptr bfloat, ptr addrspace(1) %1, i64 %73, !dbg !35 + %2454 = getelementptr bfloat, ptr addrspace(1) %2453, i64 %138, !dbg !36 + %2455 = getelementptr bfloat, ptr addrspace(1) %1, i64 %71, !dbg !35 + %2456 = getelementptr bfloat, ptr addrspace(1) %2455, i64 %138, !dbg !36 + %2457 = getelementptr bfloat, ptr addrspace(1) %1, i64 %69, !dbg !35 + %2458 = getelementptr bfloat, ptr addrspace(1) %2457, i64 %138, !dbg !36 + %2459 = getelementptr bfloat, ptr addrspace(1) %1, i64 %67, !dbg !35 + %2460 = getelementptr bfloat, ptr addrspace(1) %2459, i64 %138, !dbg !36 + %2461 = getelementptr bfloat, ptr addrspace(1) %1, i64 %65, !dbg !35 + %2462 = getelementptr bfloat, ptr addrspace(1) %2461, i64 %138, !dbg !36 + %2463 = getelementptr bfloat, ptr addrspace(1) %1, i64 %63, !dbg !35 + %2464 = getelementptr bfloat, ptr addrspace(1) %2463, i64 %138, !dbg !36 + %2465 = getelementptr bfloat, ptr addrspace(1) %1, i64 %61, !dbg !35 + %2466 = getelementptr bfloat, ptr addrspace(1) %2465, i64 %138, !dbg !36 + %2467 = getelementptr bfloat, ptr addrspace(1) %1, i64 %59, !dbg !35 + %2468 = getelementptr bfloat, ptr addrspace(1) %2467, i64 %138, !dbg !36 + %2469 = getelementptr bfloat, ptr addrspace(1) %1, i64 %57, !dbg !35 + %2470 = getelementptr bfloat, ptr addrspace(1) %2469, i64 %138, !dbg !36 + %2471 = getelementptr bfloat, ptr addrspace(1) %1, i64 %55, !dbg !35 + %2472 = getelementptr bfloat, ptr addrspace(1) %2471, i64 %138, !dbg !36 + %2473 = getelementptr bfloat, ptr addrspace(1) %1, i64 %53, !dbg !35 + %2474 = getelementptr bfloat, ptr addrspace(1) %2473, i64 %138, !dbg !36 + %2475 = getelementptr bfloat, ptr addrspace(1) %1, i64 %51, !dbg !35 + %2476 = getelementptr bfloat, ptr addrspace(1) %2475, i64 %138, !dbg !36 + %2477 = getelementptr bfloat, ptr addrspace(1) %1, i64 %49, !dbg !35 + %2478 = getelementptr bfloat, ptr addrspace(1) %2477, i64 %138, !dbg !36 + %2479 = getelementptr bfloat, ptr addrspace(1) %1, i64 %47, !dbg !35 + %2480 = getelementptr bfloat, ptr addrspace(1) %2479, i64 %138, !dbg !36 + %2481 = getelementptr bfloat, ptr addrspace(1) %1, i64 %45, !dbg !35 + %2482 = getelementptr bfloat, ptr addrspace(1) %2481, i64 %138, !dbg !36 + %2483 = getelementptr bfloat, ptr addrspace(1) %1, i64 %43, !dbg !35 + %2484 = getelementptr bfloat, ptr addrspace(1) %2483, i64 %138, !dbg !36 + %2485 = icmp slt i64 %2416, %2411, !dbg !18 + %2486 = icmp slt i64 %2419, %2411, !dbg !18 + %2487 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i), !dbg !37 + %2488 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i3), !dbg !37 + %2489 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i8), !dbg !37 + %2490 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i13), !dbg !37 + %2491 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i18), !dbg !37 + %2492 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i23), !dbg !37 + %2493 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i28), !dbg !37 + %2494 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i33), !dbg !37 + %2495 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i38), !dbg !37 + %2496 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i43), !dbg !37 + %2497 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i48), !dbg !37 + %2498 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i53), !dbg !37 + %2499 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i58), !dbg !37 + %2500 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i63), !dbg !37 + %2501 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i68), !dbg !37 + %2502 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i73), !dbg !37 + %2503 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i78), !dbg !37 + %2504 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i83), !dbg !37 + %2505 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i88), !dbg !37 + %2506 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i93), !dbg !37 + %2507 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i98), !dbg !37 + %2508 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i103), !dbg !37 + %2509 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i108), !dbg !37 + %2510 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i113), !dbg !37 + %2511 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i118), !dbg !37 + %2512 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i123), !dbg !37 + %2513 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i128), !dbg !37 + %2514 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i133), !dbg !37 + %2515 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i138), !dbg !37 + %2516 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i143), !dbg !37 + %2517 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i148), !dbg !37 + %2518 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i153), !dbg !37 + %2519 = fmul float %2487, %747, !dbg !38 + %2520 = fmul float %2487, %748, !dbg !38 + %2521 = fmul float %2487, %749, !dbg !38 + %2522 = fmul float %2487, %750, !dbg !38 + %2523 = fmul float %2487, %751, !dbg !38 + %2524 = fmul float %2487, %752, !dbg !38 + %2525 = fmul float %2487, %753, !dbg !38 + %2526 = fmul float %2487, %754, !dbg !38 + %2527 = fmul float %2488, %755, !dbg !38 + %2528 = fmul float %2488, %756, !dbg !38 + %2529 = fmul float %2488, %757, !dbg !38 + %2530 = fmul float %2488, %758, !dbg !38 + %2531 = fmul float %2488, %759, !dbg !38 + %2532 = fmul float %2488, %760, !dbg !38 + %2533 = fmul float %2488, %761, !dbg !38 + %2534 = fmul float %2488, %762, !dbg !38 + %2535 = fmul float %2489, %763, !dbg !38 + %2536 = fmul float %2489, %764, !dbg !38 + %2537 = fmul float %2489, %765, !dbg !38 + %2538 = fmul float %2489, %766, !dbg !38 + %2539 = fmul float %2489, %767, !dbg !38 + %2540 = fmul float %2489, %768, !dbg !38 + %2541 = fmul float %2489, %769, !dbg !38 + %2542 = fmul float %2489, %770, !dbg !38 + %2543 = fmul float %2490, %771, !dbg !38 + %2544 = fmul float %2490, %772, !dbg !38 + %2545 = fmul float %2490, %773, !dbg !38 + %2546 = fmul float %2490, %774, !dbg !38 + %2547 = fmul float %2490, %775, !dbg !38 + %2548 = fmul float %2490, %776, !dbg !38 + %2549 = fmul float %2490, %777, !dbg !38 + %2550 = fmul float %2490, %778, !dbg !38 + %2551 = fmul float %2491, %779, !dbg !38 + %2552 = fmul float %2491, %780, !dbg !38 + %2553 = fmul float %2491, %781, !dbg !38 + %2554 = fmul float %2491, %782, !dbg !38 + %2555 = fmul float %2491, %783, !dbg !38 + %2556 = fmul float %2491, %784, !dbg !38 + %2557 = fmul float %2491, %785, !dbg !38 + %2558 = fmul float %2491, %786, !dbg !38 + %2559 = fmul float %2492, %787, !dbg !38 + %2560 = fmul float %2492, %788, !dbg !38 + %2561 = fmul float %2492, %789, !dbg !38 + %2562 = fmul float %2492, %790, !dbg !38 + %2563 = fmul float %2492, %791, !dbg !38 + %2564 = fmul float %2492, %792, !dbg !38 + %2565 = fmul float %2492, %793, !dbg !38 + %2566 = fmul float %2492, %794, !dbg !38 + %2567 = fmul float %2493, %795, !dbg !38 + %2568 = fmul float %2493, %796, !dbg !38 + %2569 = fmul float %2493, %797, !dbg !38 + %2570 = fmul float %2493, %798, !dbg !38 + %2571 = fmul float %2493, %799, !dbg !38 + %2572 = fmul float %2493, %800, !dbg !38 + %2573 = fmul float %2493, %801, !dbg !38 + %2574 = fmul float %2493, %802, !dbg !38 + %2575 = fmul float %2494, %803, !dbg !38 + %2576 = fmul float %2494, %804, !dbg !38 + %2577 = fmul float %2494, %805, !dbg !38 + %2578 = fmul float %2494, %806, !dbg !38 + %2579 = fmul float %2494, %807, !dbg !38 + %2580 = fmul float %2494, %808, !dbg !38 + %2581 = fmul float %2494, %809, !dbg !38 + %2582 = fmul float %2494, %810, !dbg !38 + %2583 = fmul float %2495, %811, !dbg !38 + %2584 = fmul float %2495, %812, !dbg !38 + %2585 = fmul float %2495, %813, !dbg !38 + %2586 = fmul float %2495, %814, !dbg !38 + %2587 = fmul float %2495, %815, !dbg !38 + %2588 = fmul float %2495, %816, !dbg !38 + %2589 = fmul float %2495, %817, !dbg !38 + %2590 = fmul float %2495, %818, !dbg !38 + %2591 = fmul float %2496, %819, !dbg !38 + %2592 = fmul float %2496, %820, !dbg !38 + %2593 = fmul float %2496, %821, !dbg !38 + %2594 = fmul float %2496, %822, !dbg !38 + %2595 = fmul float %2496, %823, !dbg !38 + %2596 = fmul float %2496, %824, !dbg !38 + %2597 = fmul float %2496, %825, !dbg !38 + %2598 = fmul float %2496, %826, !dbg !38 + %2599 = fmul float %2497, %827, !dbg !38 + %2600 = fmul float %2497, %828, !dbg !38 + %2601 = fmul float %2497, %829, !dbg !38 + %2602 = fmul float %2497, %830, !dbg !38 + %2603 = fmul float %2497, %831, !dbg !38 + %2604 = fmul float %2497, %832, !dbg !38 + %2605 = fmul float %2497, %833, !dbg !38 + %2606 = fmul float %2497, %834, !dbg !38 + %2607 = fmul float %2498, %835, !dbg !38 + %2608 = fmul float %2498, %836, !dbg !38 + %2609 = fmul float %2498, %837, !dbg !38 + %2610 = fmul float %2498, %838, !dbg !38 + %2611 = fmul float %2498, %839, !dbg !38 + %2612 = fmul float %2498, %840, !dbg !38 + %2613 = fmul float %2498, %841, !dbg !38 + %2614 = fmul float %2498, %842, !dbg !38 + %2615 = fmul float %2499, %843, !dbg !38 + %2616 = fmul float %2499, %844, !dbg !38 + %2617 = fmul float %2499, %845, !dbg !38 + %2618 = fmul float %2499, %846, !dbg !38 + %2619 = fmul float %2499, %847, !dbg !38 + %2620 = fmul float %2499, %848, !dbg !38 + %2621 = fmul float %2499, %849, !dbg !38 + %2622 = fmul float %2499, %850, !dbg !38 + %2623 = fmul float %2500, %851, !dbg !38 + %2624 = fmul float %2500, %852, !dbg !38 + %2625 = fmul float %2500, %853, !dbg !38 + %2626 = fmul float %2500, %854, !dbg !38 + %2627 = fmul float %2500, %855, !dbg !38 + %2628 = fmul float %2500, %856, !dbg !38 + %2629 = fmul float %2500, %857, !dbg !38 + %2630 = fmul float %2500, %858, !dbg !38 + %2631 = fmul float %2501, %859, !dbg !38 + %2632 = fmul float %2501, %860, !dbg !38 + %2633 = fmul float %2501, %861, !dbg !38 + %2634 = fmul float %2501, %862, !dbg !38 + %2635 = fmul float %2501, %863, !dbg !38 + %2636 = fmul float %2501, %864, !dbg !38 + %2637 = fmul float %2501, %865, !dbg !38 + %2638 = fmul float %2501, %866, !dbg !38 + %2639 = fmul float %2502, %867, !dbg !38 + %2640 = fmul float %2502, %868, !dbg !38 + %2641 = fmul float %2502, %869, !dbg !38 + %2642 = fmul float %2502, %870, !dbg !38 + %2643 = fmul float %2502, %871, !dbg !38 + %2644 = fmul float %2502, %872, !dbg !38 + %2645 = fmul float %2502, %873, !dbg !38 + %2646 = fmul float %2502, %874, !dbg !38 + %2647 = fmul float %2503, %875, !dbg !38 + %2648 = fmul float %2503, %876, !dbg !38 + %2649 = fmul float %2503, %877, !dbg !38 + %2650 = fmul float %2503, %878, !dbg !38 + %2651 = fmul float %2503, %879, !dbg !38 + %2652 = fmul float %2503, %880, !dbg !38 + %2653 = fmul float %2503, %881, !dbg !38 + %2654 = fmul float %2503, %882, !dbg !38 + %2655 = fmul float %2504, %883, !dbg !38 + %2656 = fmul float %2504, %884, !dbg !38 + %2657 = fmul float %2504, %885, !dbg !38 + %2658 = fmul float %2504, %886, !dbg !38 + %2659 = fmul float %2504, %887, !dbg !38 + %2660 = fmul float %2504, %888, !dbg !38 + %2661 = fmul float %2504, %889, !dbg !38 + %2662 = fmul float %2504, %890, !dbg !38 + %2663 = fmul float %2505, %891, !dbg !38 + %2664 = fmul float %2505, %892, !dbg !38 + %2665 = fmul float %2505, %893, !dbg !38 + %2666 = fmul float %2505, %894, !dbg !38 + %2667 = fmul float %2505, %895, !dbg !38 + %2668 = fmul float %2505, %896, !dbg !38 + %2669 = fmul float %2505, %897, !dbg !38 + %2670 = fmul float %2505, %898, !dbg !38 + %2671 = fmul float %2506, %899, !dbg !38 + %2672 = fmul float %2506, %900, !dbg !38 + %2673 = fmul float %2506, %901, !dbg !38 + %2674 = fmul float %2506, %902, !dbg !38 + %2675 = fmul float %2506, %903, !dbg !38 + %2676 = fmul float %2506, %904, !dbg !38 + %2677 = fmul float %2506, %905, !dbg !38 + %2678 = fmul float %2506, %906, !dbg !38 + %2679 = fmul float %2507, %907, !dbg !38 + %2680 = fmul float %2507, %908, !dbg !38 + %2681 = fmul float %2507, %909, !dbg !38 + %2682 = fmul float %2507, %910, !dbg !38 + %2683 = fmul float %2507, %911, !dbg !38 + %2684 = fmul float %2507, %912, !dbg !38 + %2685 = fmul float %2507, %913, !dbg !38 + %2686 = fmul float %2507, %914, !dbg !38 + %2687 = fmul float %2508, %915, !dbg !38 + %2688 = fmul float %2508, %916, !dbg !38 + %2689 = fmul float %2508, %917, !dbg !38 + %2690 = fmul float %2508, %918, !dbg !38 + %2691 = fmul float %2508, %919, !dbg !38 + %2692 = fmul float %2508, %920, !dbg !38 + %2693 = fmul float %2508, %921, !dbg !38 + %2694 = fmul float %2508, %922, !dbg !38 + %2695 = fmul float %2509, %923, !dbg !38 + %2696 = fmul float %2509, %924, !dbg !38 + %2697 = fmul float %2509, %925, !dbg !38 + %2698 = fmul float %2509, %926, !dbg !38 + %2699 = fmul float %2509, %927, !dbg !38 + %2700 = fmul float %2509, %928, !dbg !38 + %2701 = fmul float %2509, %929, !dbg !38 + %2702 = fmul float %2509, %930, !dbg !38 + %2703 = fmul float %2510, %931, !dbg !38 + %2704 = fmul float %2510, %932, !dbg !38 + %2705 = fmul float %2510, %933, !dbg !38 + %2706 = fmul float %2510, %934, !dbg !38 + %2707 = fmul float %2510, %935, !dbg !38 + %2708 = fmul float %2510, %936, !dbg !38 + %2709 = fmul float %2510, %937, !dbg !38 + %2710 = fmul float %2510, %938, !dbg !38 + %2711 = fmul float %2511, %939, !dbg !38 + %2712 = fmul float %2511, %940, !dbg !38 + %2713 = fmul float %2511, %941, !dbg !38 + %2714 = fmul float %2511, %942, !dbg !38 + %2715 = fmul float %2511, %943, !dbg !38 + %2716 = fmul float %2511, %944, !dbg !38 + %2717 = fmul float %2511, %945, !dbg !38 + %2718 = fmul float %2511, %946, !dbg !38 + %2719 = fmul float %2512, %947, !dbg !38 + %2720 = fmul float %2512, %948, !dbg !38 + %2721 = fmul float %2512, %949, !dbg !38 + %2722 = fmul float %2512, %950, !dbg !38 + %2723 = fmul float %2512, %951, !dbg !38 + %2724 = fmul float %2512, %952, !dbg !38 + %2725 = fmul float %2512, %953, !dbg !38 + %2726 = fmul float %2512, %954, !dbg !38 + %2727 = fmul float %2513, %955, !dbg !38 + %2728 = fmul float %2513, %956, !dbg !38 + %2729 = fmul float %2513, %957, !dbg !38 + %2730 = fmul float %2513, %958, !dbg !38 + %2731 = fmul float %2513, %959, !dbg !38 + %2732 = fmul float %2513, %960, !dbg !38 + %2733 = fmul float %2513, %961, !dbg !38 + %2734 = fmul float %2513, %962, !dbg !38 + %2735 = fmul float %2514, %963, !dbg !38 + %2736 = fmul float %2514, %964, !dbg !38 + %2737 = fmul float %2514, %965, !dbg !38 + %2738 = fmul float %2514, %966, !dbg !38 + %2739 = fmul float %2514, %967, !dbg !38 + %2740 = fmul float %2514, %968, !dbg !38 + %2741 = fmul float %2514, %969, !dbg !38 + %2742 = fmul float %2514, %970, !dbg !38 + %2743 = fmul float %2515, %971, !dbg !38 + %2744 = fmul float %2515, %972, !dbg !38 + %2745 = fmul float %2515, %973, !dbg !38 + %2746 = fmul float %2515, %974, !dbg !38 + %2747 = fmul float %2515, %975, !dbg !38 + %2748 = fmul float %2515, %976, !dbg !38 + %2749 = fmul float %2515, %977, !dbg !38 + %2750 = fmul float %2515, %978, !dbg !38 + %2751 = fmul float %2516, %979, !dbg !38 + %2752 = fmul float %2516, %980, !dbg !38 + %2753 = fmul float %2516, %981, !dbg !38 + %2754 = fmul float %2516, %982, !dbg !38 + %2755 = fmul float %2516, %983, !dbg !38 + %2756 = fmul float %2516, %984, !dbg !38 + %2757 = fmul float %2516, %985, !dbg !38 + %2758 = fmul float %2516, %986, !dbg !38 + %2759 = fmul float %2517, %987, !dbg !38 + %2760 = fmul float %2517, %988, !dbg !38 + %2761 = fmul float %2517, %989, !dbg !38 + %2762 = fmul float %2517, %990, !dbg !38 + %2763 = fmul float %2517, %991, !dbg !38 + %2764 = fmul float %2517, %992, !dbg !38 + %2765 = fmul float %2517, %993, !dbg !38 + %2766 = fmul float %2517, %994, !dbg !38 + %2767 = fmul float %2518, %995, !dbg !38 + %2768 = fmul float %2518, %996, !dbg !38 + %2769 = fmul float %2518, %997, !dbg !38 + %2770 = fmul float %2518, %998, !dbg !38 + %2771 = fmul float %2518, %999, !dbg !38 + %2772 = fmul float %2518, %1000, !dbg !38 + %2773 = fmul float %2518, %1001, !dbg !38 + %2774 = fmul float %2518, %1002, !dbg !38 + %2775 = fptrunc float %2519 to bfloat, !dbg !39 + %2776 = fptrunc float %2520 to bfloat, !dbg !39 + %2777 = fptrunc float %2521 to bfloat, !dbg !39 + %2778 = fptrunc float %2522 to bfloat, !dbg !39 + %2779 = fptrunc float %2523 to bfloat, !dbg !39 + %2780 = fptrunc float %2524 to bfloat, !dbg !39 + %2781 = fptrunc float %2525 to bfloat, !dbg !39 + %2782 = fptrunc float %2526 to bfloat, !dbg !39 + %2783 = fptrunc float %2527 to bfloat, !dbg !39 + %2784 = fptrunc float %2528 to bfloat, !dbg !39 + %2785 = fptrunc float %2529 to bfloat, !dbg !39 + %2786 = fptrunc float %2530 to bfloat, !dbg !39 + %2787 = fptrunc float %2531 to bfloat, !dbg !39 + %2788 = fptrunc float %2532 to bfloat, !dbg !39 + %2789 = fptrunc float %2533 to bfloat, !dbg !39 + %2790 = fptrunc float %2534 to bfloat, !dbg !39 + %2791 = fptrunc float %2535 to bfloat, !dbg !39 + %2792 = fptrunc float %2536 to bfloat, !dbg !39 + %2793 = fptrunc float %2537 to bfloat, !dbg !39 + %2794 = fptrunc float %2538 to bfloat, !dbg !39 + %2795 = fptrunc float %2539 to bfloat, !dbg !39 + %2796 = fptrunc float %2540 to bfloat, !dbg !39 + %2797 = fptrunc float %2541 to bfloat, !dbg !39 + %2798 = fptrunc float %2542 to bfloat, !dbg !39 + %2799 = fptrunc float %2543 to bfloat, !dbg !39 + %2800 = fptrunc float %2544 to bfloat, !dbg !39 + %2801 = fptrunc float %2545 to bfloat, !dbg !39 + %2802 = fptrunc float %2546 to bfloat, !dbg !39 + %2803 = fptrunc float %2547 to bfloat, !dbg !39 + %2804 = fptrunc float %2548 to bfloat, !dbg !39 + %2805 = fptrunc float %2549 to bfloat, !dbg !39 + %2806 = fptrunc float %2550 to bfloat, !dbg !39 + %2807 = fptrunc float %2551 to bfloat, !dbg !39 + %2808 = fptrunc float %2552 to bfloat, !dbg !39 + %2809 = fptrunc float %2553 to bfloat, !dbg !39 + %2810 = fptrunc float %2554 to bfloat, !dbg !39 + %2811 = fptrunc float %2555 to bfloat, !dbg !39 + %2812 = fptrunc float %2556 to bfloat, !dbg !39 + %2813 = fptrunc float %2557 to bfloat, !dbg !39 + %2814 = fptrunc float %2558 to bfloat, !dbg !39 + %2815 = fptrunc float %2559 to bfloat, !dbg !39 + %2816 = fptrunc float %2560 to bfloat, !dbg !39 + %2817 = fptrunc float %2561 to bfloat, !dbg !39 + %2818 = fptrunc float %2562 to bfloat, !dbg !39 + %2819 = fptrunc float %2563 to bfloat, !dbg !39 + %2820 = fptrunc float %2564 to bfloat, !dbg !39 + %2821 = fptrunc float %2565 to bfloat, !dbg !39 + %2822 = fptrunc float %2566 to bfloat, !dbg !39 + %2823 = fptrunc float %2567 to bfloat, !dbg !39 + %2824 = fptrunc float %2568 to bfloat, !dbg !39 + %2825 = fptrunc float %2569 to bfloat, !dbg !39 + %2826 = fptrunc float %2570 to bfloat, !dbg !39 + %2827 = fptrunc float %2571 to bfloat, !dbg !39 + %2828 = fptrunc float %2572 to bfloat, !dbg !39 + %2829 = fptrunc float %2573 to bfloat, !dbg !39 + %2830 = fptrunc float %2574 to bfloat, !dbg !39 + %2831 = fptrunc float %2575 to bfloat, !dbg !39 + %2832 = fptrunc float %2576 to bfloat, !dbg !39 + %2833 = fptrunc float %2577 to bfloat, !dbg !39 + %2834 = fptrunc float %2578 to bfloat, !dbg !39 + %2835 = fptrunc float %2579 to bfloat, !dbg !39 + %2836 = fptrunc float %2580 to bfloat, !dbg !39 + %2837 = fptrunc float %2581 to bfloat, !dbg !39 + %2838 = fptrunc float %2582 to bfloat, !dbg !39 + %2839 = fptrunc float %2583 to bfloat, !dbg !39 + %2840 = fptrunc float %2584 to bfloat, !dbg !39 + %2841 = fptrunc float %2585 to bfloat, !dbg !39 + %2842 = fptrunc float %2586 to bfloat, !dbg !39 + %2843 = fptrunc float %2587 to bfloat, !dbg !39 + %2844 = fptrunc float %2588 to bfloat, !dbg !39 + %2845 = fptrunc float %2589 to bfloat, !dbg !39 + %2846 = fptrunc float %2590 to bfloat, !dbg !39 + %2847 = fptrunc float %2591 to bfloat, !dbg !39 + %2848 = fptrunc float %2592 to bfloat, !dbg !39 + %2849 = fptrunc float %2593 to bfloat, !dbg !39 + %2850 = fptrunc float %2594 to bfloat, !dbg !39 + %2851 = fptrunc float %2595 to bfloat, !dbg !39 + %2852 = fptrunc float %2596 to bfloat, !dbg !39 + %2853 = fptrunc float %2597 to bfloat, !dbg !39 + %2854 = fptrunc float %2598 to bfloat, !dbg !39 + %2855 = fptrunc float %2599 to bfloat, !dbg !39 + %2856 = fptrunc float %2600 to bfloat, !dbg !39 + %2857 = fptrunc float %2601 to bfloat, !dbg !39 + %2858 = fptrunc float %2602 to bfloat, !dbg !39 + %2859 = fptrunc float %2603 to bfloat, !dbg !39 + %2860 = fptrunc float %2604 to bfloat, !dbg !39 + %2861 = fptrunc float %2605 to bfloat, !dbg !39 + %2862 = fptrunc float %2606 to bfloat, !dbg !39 + %2863 = fptrunc float %2607 to bfloat, !dbg !39 + %2864 = fptrunc float %2608 to bfloat, !dbg !39 + %2865 = fptrunc float %2609 to bfloat, !dbg !39 + %2866 = fptrunc float %2610 to bfloat, !dbg !39 + %2867 = fptrunc float %2611 to bfloat, !dbg !39 + %2868 = fptrunc float %2612 to bfloat, !dbg !39 + %2869 = fptrunc float %2613 to bfloat, !dbg !39 + %2870 = fptrunc float %2614 to bfloat, !dbg !39 + %2871 = fptrunc float %2615 to bfloat, !dbg !39 + %2872 = fptrunc float %2616 to bfloat, !dbg !39 + %2873 = fptrunc float %2617 to bfloat, !dbg !39 + %2874 = fptrunc float %2618 to bfloat, !dbg !39 + %2875 = fptrunc float %2619 to bfloat, !dbg !39 + %2876 = fptrunc float %2620 to bfloat, !dbg !39 + %2877 = fptrunc float %2621 to bfloat, !dbg !39 + %2878 = fptrunc float %2622 to bfloat, !dbg !39 + %2879 = fptrunc float %2623 to bfloat, !dbg !39 + %2880 = fptrunc float %2624 to bfloat, !dbg !39 + %2881 = fptrunc float %2625 to bfloat, !dbg !39 + %2882 = fptrunc float %2626 to bfloat, !dbg !39 + %2883 = fptrunc float %2627 to bfloat, !dbg !39 + %2884 = fptrunc float %2628 to bfloat, !dbg !39 + %2885 = fptrunc float %2629 to bfloat, !dbg !39 + %2886 = fptrunc float %2630 to bfloat, !dbg !39 + %2887 = fptrunc float %2631 to bfloat, !dbg !39 + %2888 = fptrunc float %2632 to bfloat, !dbg !39 + %2889 = fptrunc float %2633 to bfloat, !dbg !39 + %2890 = fptrunc float %2634 to bfloat, !dbg !39 + %2891 = fptrunc float %2635 to bfloat, !dbg !39 + %2892 = fptrunc float %2636 to bfloat, !dbg !39 + %2893 = fptrunc float %2637 to bfloat, !dbg !39 + %2894 = fptrunc float %2638 to bfloat, !dbg !39 + %2895 = fptrunc float %2639 to bfloat, !dbg !39 + %2896 = fptrunc float %2640 to bfloat, !dbg !39 + %2897 = fptrunc float %2641 to bfloat, !dbg !39 + %2898 = fptrunc float %2642 to bfloat, !dbg !39 + %2899 = fptrunc float %2643 to bfloat, !dbg !39 + %2900 = fptrunc float %2644 to bfloat, !dbg !39 + %2901 = fptrunc float %2645 to bfloat, !dbg !39 + %2902 = fptrunc float %2646 to bfloat, !dbg !39 + %2903 = fptrunc float %2647 to bfloat, !dbg !39 + %2904 = fptrunc float %2648 to bfloat, !dbg !39 + %2905 = fptrunc float %2649 to bfloat, !dbg !39 + %2906 = fptrunc float %2650 to bfloat, !dbg !39 + %2907 = fptrunc float %2651 to bfloat, !dbg !39 + %2908 = fptrunc float %2652 to bfloat, !dbg !39 + %2909 = fptrunc float %2653 to bfloat, !dbg !39 + %2910 = fptrunc float %2654 to bfloat, !dbg !39 + %2911 = fptrunc float %2655 to bfloat, !dbg !39 + %2912 = fptrunc float %2656 to bfloat, !dbg !39 + %2913 = fptrunc float %2657 to bfloat, !dbg !39 + %2914 = fptrunc float %2658 to bfloat, !dbg !39 + %2915 = fptrunc float %2659 to bfloat, !dbg !39 + %2916 = fptrunc float %2660 to bfloat, !dbg !39 + %2917 = fptrunc float %2661 to bfloat, !dbg !39 + %2918 = fptrunc float %2662 to bfloat, !dbg !39 + %2919 = fptrunc float %2663 to bfloat, !dbg !39 + %2920 = fptrunc float %2664 to bfloat, !dbg !39 + %2921 = fptrunc float %2665 to bfloat, !dbg !39 + %2922 = fptrunc float %2666 to bfloat, !dbg !39 + %2923 = fptrunc float %2667 to bfloat, !dbg !39 + %2924 = fptrunc float %2668 to bfloat, !dbg !39 + %2925 = fptrunc float %2669 to bfloat, !dbg !39 + %2926 = fptrunc float %2670 to bfloat, !dbg !39 + %2927 = fptrunc float %2671 to bfloat, !dbg !39 + %2928 = fptrunc float %2672 to bfloat, !dbg !39 + %2929 = fptrunc float %2673 to bfloat, !dbg !39 + %2930 = fptrunc float %2674 to bfloat, !dbg !39 + %2931 = fptrunc float %2675 to bfloat, !dbg !39 + %2932 = fptrunc float %2676 to bfloat, !dbg !39 + %2933 = fptrunc float %2677 to bfloat, !dbg !39 + %2934 = fptrunc float %2678 to bfloat, !dbg !39 + %2935 = fptrunc float %2679 to bfloat, !dbg !39 + %2936 = fptrunc float %2680 to bfloat, !dbg !39 + %2937 = fptrunc float %2681 to bfloat, !dbg !39 + %2938 = fptrunc float %2682 to bfloat, !dbg !39 + %2939 = fptrunc float %2683 to bfloat, !dbg !39 + %2940 = fptrunc float %2684 to bfloat, !dbg !39 + %2941 = fptrunc float %2685 to bfloat, !dbg !39 + %2942 = fptrunc float %2686 to bfloat, !dbg !39 + %2943 = fptrunc float %2687 to bfloat, !dbg !39 + %2944 = fptrunc float %2688 to bfloat, !dbg !39 + %2945 = fptrunc float %2689 to bfloat, !dbg !39 + %2946 = fptrunc float %2690 to bfloat, !dbg !39 + %2947 = fptrunc float %2691 to bfloat, !dbg !39 + %2948 = fptrunc float %2692 to bfloat, !dbg !39 + %2949 = fptrunc float %2693 to bfloat, !dbg !39 + %2950 = fptrunc float %2694 to bfloat, !dbg !39 + %2951 = fptrunc float %2695 to bfloat, !dbg !39 + %2952 = fptrunc float %2696 to bfloat, !dbg !39 + %2953 = fptrunc float %2697 to bfloat, !dbg !39 + %2954 = fptrunc float %2698 to bfloat, !dbg !39 + %2955 = fptrunc float %2699 to bfloat, !dbg !39 + %2956 = fptrunc float %2700 to bfloat, !dbg !39 + %2957 = fptrunc float %2701 to bfloat, !dbg !39 + %2958 = fptrunc float %2702 to bfloat, !dbg !39 + %2959 = fptrunc float %2703 to bfloat, !dbg !39 + %2960 = fptrunc float %2704 to bfloat, !dbg !39 + %2961 = fptrunc float %2705 to bfloat, !dbg !39 + %2962 = fptrunc float %2706 to bfloat, !dbg !39 + %2963 = fptrunc float %2707 to bfloat, !dbg !39 + %2964 = fptrunc float %2708 to bfloat, !dbg !39 + %2965 = fptrunc float %2709 to bfloat, !dbg !39 + %2966 = fptrunc float %2710 to bfloat, !dbg !39 + %2967 = fptrunc float %2711 to bfloat, !dbg !39 + %2968 = fptrunc float %2712 to bfloat, !dbg !39 + %2969 = fptrunc float %2713 to bfloat, !dbg !39 + %2970 = fptrunc float %2714 to bfloat, !dbg !39 + %2971 = fptrunc float %2715 to bfloat, !dbg !39 + %2972 = fptrunc float %2716 to bfloat, !dbg !39 + %2973 = fptrunc float %2717 to bfloat, !dbg !39 + %2974 = fptrunc float %2718 to bfloat, !dbg !39 + %2975 = fptrunc float %2719 to bfloat, !dbg !39 + %2976 = fptrunc float %2720 to bfloat, !dbg !39 + %2977 = fptrunc float %2721 to bfloat, !dbg !39 + %2978 = fptrunc float %2722 to bfloat, !dbg !39 + %2979 = fptrunc float %2723 to bfloat, !dbg !39 + %2980 = fptrunc float %2724 to bfloat, !dbg !39 + %2981 = fptrunc float %2725 to bfloat, !dbg !39 + %2982 = fptrunc float %2726 to bfloat, !dbg !39 + %2983 = fptrunc float %2727 to bfloat, !dbg !39 + %2984 = fptrunc float %2728 to bfloat, !dbg !39 + %2985 = fptrunc float %2729 to bfloat, !dbg !39 + %2986 = fptrunc float %2730 to bfloat, !dbg !39 + %2987 = fptrunc float %2731 to bfloat, !dbg !39 + %2988 = fptrunc float %2732 to bfloat, !dbg !39 + %2989 = fptrunc float %2733 to bfloat, !dbg !39 + %2990 = fptrunc float %2734 to bfloat, !dbg !39 + %2991 = fptrunc float %2735 to bfloat, !dbg !39 + %2992 = fptrunc float %2736 to bfloat, !dbg !39 + %2993 = fptrunc float %2737 to bfloat, !dbg !39 + %2994 = fptrunc float %2738 to bfloat, !dbg !39 + %2995 = fptrunc float %2739 to bfloat, !dbg !39 + %2996 = fptrunc float %2740 to bfloat, !dbg !39 + %2997 = fptrunc float %2741 to bfloat, !dbg !39 + %2998 = fptrunc float %2742 to bfloat, !dbg !39 + %2999 = fptrunc float %2743 to bfloat, !dbg !39 + %3000 = fptrunc float %2744 to bfloat, !dbg !39 + %3001 = fptrunc float %2745 to bfloat, !dbg !39 + %3002 = fptrunc float %2746 to bfloat, !dbg !39 + %3003 = fptrunc float %2747 to bfloat, !dbg !39 + %3004 = fptrunc float %2748 to bfloat, !dbg !39 + %3005 = fptrunc float %2749 to bfloat, !dbg !39 + %3006 = fptrunc float %2750 to bfloat, !dbg !39 + %3007 = fptrunc float %2751 to bfloat, !dbg !39 + %3008 = fptrunc float %2752 to bfloat, !dbg !39 + %3009 = fptrunc float %2753 to bfloat, !dbg !39 + %3010 = fptrunc float %2754 to bfloat, !dbg !39 + %3011 = fptrunc float %2755 to bfloat, !dbg !39 + %3012 = fptrunc float %2756 to bfloat, !dbg !39 + %3013 = fptrunc float %2757 to bfloat, !dbg !39 + %3014 = fptrunc float %2758 to bfloat, !dbg !39 + %3015 = fptrunc float %2759 to bfloat, !dbg !39 + %3016 = fptrunc float %2760 to bfloat, !dbg !39 + %3017 = fptrunc float %2761 to bfloat, !dbg !39 + %3018 = fptrunc float %2762 to bfloat, !dbg !39 + %3019 = fptrunc float %2763 to bfloat, !dbg !39 + %3020 = fptrunc float %2764 to bfloat, !dbg !39 + %3021 = fptrunc float %2765 to bfloat, !dbg !39 + %3022 = fptrunc float %2766 to bfloat, !dbg !39 + %3023 = fptrunc float %2767 to bfloat, !dbg !39 + %3024 = fptrunc float %2768 to bfloat, !dbg !39 + %3025 = fptrunc float %2769 to bfloat, !dbg !39 + %3026 = fptrunc float %2770 to bfloat, !dbg !39 + %3027 = fptrunc float %2771 to bfloat, !dbg !39 + %3028 = fptrunc float %2772 to bfloat, !dbg !39 + %3029 = fptrunc float %2773 to bfloat, !dbg !39 + %3030 = fptrunc float %2774 to bfloat, !dbg !39 + %3031 = insertelement <2 x bfloat> poison, bfloat %2775, i64 0, !dbg !40 + %3032 = insertelement <2 x bfloat> %3031, bfloat %2776, i64 1, !dbg !40 + %3033 = bitcast <2 x bfloat> %3032 to i32, !dbg !40 + %3034 = insertelement <2 x bfloat> poison, bfloat %2777, i64 0, !dbg !40 + %3035 = insertelement <2 x bfloat> %3034, bfloat %2778, i64 1, !dbg !40 + %3036 = bitcast <2 x bfloat> %3035 to i32, !dbg !40 + %3037 = insertelement <2 x bfloat> poison, bfloat %2779, i64 0, !dbg !40 + %3038 = insertelement <2 x bfloat> %3037, bfloat %2780, i64 1, !dbg !40 + %3039 = bitcast <2 x bfloat> %3038 to i32, !dbg !40 + %3040 = insertelement <2 x bfloat> poison, bfloat %2781, i64 0, !dbg !40 + %3041 = insertelement <2 x bfloat> %3040, bfloat %2782, i64 1, !dbg !40 + %3042 = bitcast <2 x bfloat> %3041 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3033, i32 %3036, i32 %3039, i32 %3042, ptr addrspace(1) %2484, i1 %171) #7, !dbg !40 + %3043 = insertelement <2 x bfloat> poison, bfloat %2783, i64 0, !dbg !40 + %3044 = insertelement <2 x bfloat> %3043, bfloat %2784, i64 1, !dbg !40 + %3045 = bitcast <2 x bfloat> %3044 to i32, !dbg !40 + %3046 = insertelement <2 x bfloat> poison, bfloat %2785, i64 0, !dbg !40 + %3047 = insertelement <2 x bfloat> %3046, bfloat %2786, i64 1, !dbg !40 + %3048 = bitcast <2 x bfloat> %3047 to i32, !dbg !40 + %3049 = insertelement <2 x bfloat> poison, bfloat %2787, i64 0, !dbg !40 + %3050 = insertelement <2 x bfloat> %3049, bfloat %2788, i64 1, !dbg !40 + %3051 = bitcast <2 x bfloat> %3050 to i32, !dbg !40 + %3052 = insertelement <2 x bfloat> poison, bfloat %2789, i64 0, !dbg !40 + %3053 = insertelement <2 x bfloat> %3052, bfloat %2790, i64 1, !dbg !40 + %3054 = bitcast <2 x bfloat> %3053 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3045, i32 %3048, i32 %3051, i32 %3054, ptr addrspace(1) %2482, i1 %189) #7, !dbg !40 + %3055 = insertelement <2 x bfloat> poison, bfloat %2791, i64 0, !dbg !40 + %3056 = insertelement <2 x bfloat> %3055, bfloat %2792, i64 1, !dbg !40 + %3057 = bitcast <2 x bfloat> %3056 to i32, !dbg !40 + %3058 = insertelement <2 x bfloat> poison, bfloat %2793, i64 0, !dbg !40 + %3059 = insertelement <2 x bfloat> %3058, bfloat %2794, i64 1, !dbg !40 + %3060 = bitcast <2 x bfloat> %3059 to i32, !dbg !40 + %3061 = insertelement <2 x bfloat> poison, bfloat %2795, i64 0, !dbg !40 + %3062 = insertelement <2 x bfloat> %3061, bfloat %2796, i64 1, !dbg !40 + %3063 = bitcast <2 x bfloat> %3062 to i32, !dbg !40 + %3064 = insertelement <2 x bfloat> poison, bfloat %2797, i64 0, !dbg !40 + %3065 = insertelement <2 x bfloat> %3064, bfloat %2798, i64 1, !dbg !40 + %3066 = bitcast <2 x bfloat> %3065 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3057, i32 %3060, i32 %3063, i32 %3066, ptr addrspace(1) %2480, i1 %207) #7, !dbg !40 + %3067 = insertelement <2 x bfloat> poison, bfloat %2799, i64 0, !dbg !40 + %3068 = insertelement <2 x bfloat> %3067, bfloat %2800, i64 1, !dbg !40 + %3069 = bitcast <2 x bfloat> %3068 to i32, !dbg !40 + %3070 = insertelement <2 x bfloat> poison, bfloat %2801, i64 0, !dbg !40 + %3071 = insertelement <2 x bfloat> %3070, bfloat %2802, i64 1, !dbg !40 + %3072 = bitcast <2 x bfloat> %3071 to i32, !dbg !40 + %3073 = insertelement <2 x bfloat> poison, bfloat %2803, i64 0, !dbg !40 + %3074 = insertelement <2 x bfloat> %3073, bfloat %2804, i64 1, !dbg !40 + %3075 = bitcast <2 x bfloat> %3074 to i32, !dbg !40 + %3076 = insertelement <2 x bfloat> poison, bfloat %2805, i64 0, !dbg !40 + %3077 = insertelement <2 x bfloat> %3076, bfloat %2806, i64 1, !dbg !40 + %3078 = bitcast <2 x bfloat> %3077 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3069, i32 %3072, i32 %3075, i32 %3078, ptr addrspace(1) %2478, i1 %225) #7, !dbg !40 + %3079 = insertelement <2 x bfloat> poison, bfloat %2807, i64 0, !dbg !40 + %3080 = insertelement <2 x bfloat> %3079, bfloat %2808, i64 1, !dbg !40 + %3081 = bitcast <2 x bfloat> %3080 to i32, !dbg !40 + %3082 = insertelement <2 x bfloat> poison, bfloat %2809, i64 0, !dbg !40 + %3083 = insertelement <2 x bfloat> %3082, bfloat %2810, i64 1, !dbg !40 + %3084 = bitcast <2 x bfloat> %3083 to i32, !dbg !40 + %3085 = insertelement <2 x bfloat> poison, bfloat %2811, i64 0, !dbg !40 + %3086 = insertelement <2 x bfloat> %3085, bfloat %2812, i64 1, !dbg !40 + %3087 = bitcast <2 x bfloat> %3086 to i32, !dbg !40 + %3088 = insertelement <2 x bfloat> poison, bfloat %2813, i64 0, !dbg !40 + %3089 = insertelement <2 x bfloat> %3088, bfloat %2814, i64 1, !dbg !40 + %3090 = bitcast <2 x bfloat> %3089 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3081, i32 %3084, i32 %3087, i32 %3090, ptr addrspace(1) %2476, i1 %243) #7, !dbg !40 + %3091 = insertelement <2 x bfloat> poison, bfloat %2815, i64 0, !dbg !40 + %3092 = insertelement <2 x bfloat> %3091, bfloat %2816, i64 1, !dbg !40 + %3093 = bitcast <2 x bfloat> %3092 to i32, !dbg !40 + %3094 = insertelement <2 x bfloat> poison, bfloat %2817, i64 0, !dbg !40 + %3095 = insertelement <2 x bfloat> %3094, bfloat %2818, i64 1, !dbg !40 + %3096 = bitcast <2 x bfloat> %3095 to i32, !dbg !40 + %3097 = insertelement <2 x bfloat> poison, bfloat %2819, i64 0, !dbg !40 + %3098 = insertelement <2 x bfloat> %3097, bfloat %2820, i64 1, !dbg !40 + %3099 = bitcast <2 x bfloat> %3098 to i32, !dbg !40 + %3100 = insertelement <2 x bfloat> poison, bfloat %2821, i64 0, !dbg !40 + %3101 = insertelement <2 x bfloat> %3100, bfloat %2822, i64 1, !dbg !40 + %3102 = bitcast <2 x bfloat> %3101 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3093, i32 %3096, i32 %3099, i32 %3102, ptr addrspace(1) %2474, i1 %261) #7, !dbg !40 + %3103 = insertelement <2 x bfloat> poison, bfloat %2823, i64 0, !dbg !40 + %3104 = insertelement <2 x bfloat> %3103, bfloat %2824, i64 1, !dbg !40 + %3105 = bitcast <2 x bfloat> %3104 to i32, !dbg !40 + %3106 = insertelement <2 x bfloat> poison, bfloat %2825, i64 0, !dbg !40 + %3107 = insertelement <2 x bfloat> %3106, bfloat %2826, i64 1, !dbg !40 + %3108 = bitcast <2 x bfloat> %3107 to i32, !dbg !40 + %3109 = insertelement <2 x bfloat> poison, bfloat %2827, i64 0, !dbg !40 + %3110 = insertelement <2 x bfloat> %3109, bfloat %2828, i64 1, !dbg !40 + %3111 = bitcast <2 x bfloat> %3110 to i32, !dbg !40 + %3112 = insertelement <2 x bfloat> poison, bfloat %2829, i64 0, !dbg !40 + %3113 = insertelement <2 x bfloat> %3112, bfloat %2830, i64 1, !dbg !40 + %3114 = bitcast <2 x bfloat> %3113 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3105, i32 %3108, i32 %3111, i32 %3114, ptr addrspace(1) %2472, i1 %279) #7, !dbg !40 + %3115 = insertelement <2 x bfloat> poison, bfloat %2831, i64 0, !dbg !40 + %3116 = insertelement <2 x bfloat> %3115, bfloat %2832, i64 1, !dbg !40 + %3117 = bitcast <2 x bfloat> %3116 to i32, !dbg !40 + %3118 = insertelement <2 x bfloat> poison, bfloat %2833, i64 0, !dbg !40 + %3119 = insertelement <2 x bfloat> %3118, bfloat %2834, i64 1, !dbg !40 + %3120 = bitcast <2 x bfloat> %3119 to i32, !dbg !40 + %3121 = insertelement <2 x bfloat> poison, bfloat %2835, i64 0, !dbg !40 + %3122 = insertelement <2 x bfloat> %3121, bfloat %2836, i64 1, !dbg !40 + %3123 = bitcast <2 x bfloat> %3122 to i32, !dbg !40 + %3124 = insertelement <2 x bfloat> poison, bfloat %2837, i64 0, !dbg !40 + %3125 = insertelement <2 x bfloat> %3124, bfloat %2838, i64 1, !dbg !40 + %3126 = bitcast <2 x bfloat> %3125 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3117, i32 %3120, i32 %3123, i32 %3126, ptr addrspace(1) %2470, i1 %297) #7, !dbg !40 + %3127 = insertelement <2 x bfloat> poison, bfloat %2839, i64 0, !dbg !40 + %3128 = insertelement <2 x bfloat> %3127, bfloat %2840, i64 1, !dbg !40 + %3129 = bitcast <2 x bfloat> %3128 to i32, !dbg !40 + %3130 = insertelement <2 x bfloat> poison, bfloat %2841, i64 0, !dbg !40 + %3131 = insertelement <2 x bfloat> %3130, bfloat %2842, i64 1, !dbg !40 + %3132 = bitcast <2 x bfloat> %3131 to i32, !dbg !40 + %3133 = insertelement <2 x bfloat> poison, bfloat %2843, i64 0, !dbg !40 + %3134 = insertelement <2 x bfloat> %3133, bfloat %2844, i64 1, !dbg !40 + %3135 = bitcast <2 x bfloat> %3134 to i32, !dbg !40 + %3136 = insertelement <2 x bfloat> poison, bfloat %2845, i64 0, !dbg !40 + %3137 = insertelement <2 x bfloat> %3136, bfloat %2846, i64 1, !dbg !40 + %3138 = bitcast <2 x bfloat> %3137 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3129, i32 %3132, i32 %3135, i32 %3138, ptr addrspace(1) %2468, i1 %315) #7, !dbg !40 + %3139 = insertelement <2 x bfloat> poison, bfloat %2847, i64 0, !dbg !40 + %3140 = insertelement <2 x bfloat> %3139, bfloat %2848, i64 1, !dbg !40 + %3141 = bitcast <2 x bfloat> %3140 to i32, !dbg !40 + %3142 = insertelement <2 x bfloat> poison, bfloat %2849, i64 0, !dbg !40 + %3143 = insertelement <2 x bfloat> %3142, bfloat %2850, i64 1, !dbg !40 + %3144 = bitcast <2 x bfloat> %3143 to i32, !dbg !40 + %3145 = insertelement <2 x bfloat> poison, bfloat %2851, i64 0, !dbg !40 + %3146 = insertelement <2 x bfloat> %3145, bfloat %2852, i64 1, !dbg !40 + %3147 = bitcast <2 x bfloat> %3146 to i32, !dbg !40 + %3148 = insertelement <2 x bfloat> poison, bfloat %2853, i64 0, !dbg !40 + %3149 = insertelement <2 x bfloat> %3148, bfloat %2854, i64 1, !dbg !40 + %3150 = bitcast <2 x bfloat> %3149 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3141, i32 %3144, i32 %3147, i32 %3150, ptr addrspace(1) %2466, i1 %333) #7, !dbg !40 + %3151 = insertelement <2 x bfloat> poison, bfloat %2855, i64 0, !dbg !40 + %3152 = insertelement <2 x bfloat> %3151, bfloat %2856, i64 1, !dbg !40 + %3153 = bitcast <2 x bfloat> %3152 to i32, !dbg !40 + %3154 = insertelement <2 x bfloat> poison, bfloat %2857, i64 0, !dbg !40 + %3155 = insertelement <2 x bfloat> %3154, bfloat %2858, i64 1, !dbg !40 + %3156 = bitcast <2 x bfloat> %3155 to i32, !dbg !40 + %3157 = insertelement <2 x bfloat> poison, bfloat %2859, i64 0, !dbg !40 + %3158 = insertelement <2 x bfloat> %3157, bfloat %2860, i64 1, !dbg !40 + %3159 = bitcast <2 x bfloat> %3158 to i32, !dbg !40 + %3160 = insertelement <2 x bfloat> poison, bfloat %2861, i64 0, !dbg !40 + %3161 = insertelement <2 x bfloat> %3160, bfloat %2862, i64 1, !dbg !40 + %3162 = bitcast <2 x bfloat> %3161 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3153, i32 %3156, i32 %3159, i32 %3162, ptr addrspace(1) %2464, i1 %351) #7, !dbg !40 + %3163 = insertelement <2 x bfloat> poison, bfloat %2863, i64 0, !dbg !40 + %3164 = insertelement <2 x bfloat> %3163, bfloat %2864, i64 1, !dbg !40 + %3165 = bitcast <2 x bfloat> %3164 to i32, !dbg !40 + %3166 = insertelement <2 x bfloat> poison, bfloat %2865, i64 0, !dbg !40 + %3167 = insertelement <2 x bfloat> %3166, bfloat %2866, i64 1, !dbg !40 + %3168 = bitcast <2 x bfloat> %3167 to i32, !dbg !40 + %3169 = insertelement <2 x bfloat> poison, bfloat %2867, i64 0, !dbg !40 + %3170 = insertelement <2 x bfloat> %3169, bfloat %2868, i64 1, !dbg !40 + %3171 = bitcast <2 x bfloat> %3170 to i32, !dbg !40 + %3172 = insertelement <2 x bfloat> poison, bfloat %2869, i64 0, !dbg !40 + %3173 = insertelement <2 x bfloat> %3172, bfloat %2870, i64 1, !dbg !40 + %3174 = bitcast <2 x bfloat> %3173 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3165, i32 %3168, i32 %3171, i32 %3174, ptr addrspace(1) %2462, i1 %369) #7, !dbg !40 + %3175 = insertelement <2 x bfloat> poison, bfloat %2871, i64 0, !dbg !40 + %3176 = insertelement <2 x bfloat> %3175, bfloat %2872, i64 1, !dbg !40 + %3177 = bitcast <2 x bfloat> %3176 to i32, !dbg !40 + %3178 = insertelement <2 x bfloat> poison, bfloat %2873, i64 0, !dbg !40 + %3179 = insertelement <2 x bfloat> %3178, bfloat %2874, i64 1, !dbg !40 + %3180 = bitcast <2 x bfloat> %3179 to i32, !dbg !40 + %3181 = insertelement <2 x bfloat> poison, bfloat %2875, i64 0, !dbg !40 + %3182 = insertelement <2 x bfloat> %3181, bfloat %2876, i64 1, !dbg !40 + %3183 = bitcast <2 x bfloat> %3182 to i32, !dbg !40 + %3184 = insertelement <2 x bfloat> poison, bfloat %2877, i64 0, !dbg !40 + %3185 = insertelement <2 x bfloat> %3184, bfloat %2878, i64 1, !dbg !40 + %3186 = bitcast <2 x bfloat> %3185 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3177, i32 %3180, i32 %3183, i32 %3186, ptr addrspace(1) %2460, i1 %387) #7, !dbg !40 + %3187 = insertelement <2 x bfloat> poison, bfloat %2879, i64 0, !dbg !40 + %3188 = insertelement <2 x bfloat> %3187, bfloat %2880, i64 1, !dbg !40 + %3189 = bitcast <2 x bfloat> %3188 to i32, !dbg !40 + %3190 = insertelement <2 x bfloat> poison, bfloat %2881, i64 0, !dbg !40 + %3191 = insertelement <2 x bfloat> %3190, bfloat %2882, i64 1, !dbg !40 + %3192 = bitcast <2 x bfloat> %3191 to i32, !dbg !40 + %3193 = insertelement <2 x bfloat> poison, bfloat %2883, i64 0, !dbg !40 + %3194 = insertelement <2 x bfloat> %3193, bfloat %2884, i64 1, !dbg !40 + %3195 = bitcast <2 x bfloat> %3194 to i32, !dbg !40 + %3196 = insertelement <2 x bfloat> poison, bfloat %2885, i64 0, !dbg !40 + %3197 = insertelement <2 x bfloat> %3196, bfloat %2886, i64 1, !dbg !40 + %3198 = bitcast <2 x bfloat> %3197 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3189, i32 %3192, i32 %3195, i32 %3198, ptr addrspace(1) %2458, i1 %405) #7, !dbg !40 + %3199 = insertelement <2 x bfloat> poison, bfloat %2887, i64 0, !dbg !40 + %3200 = insertelement <2 x bfloat> %3199, bfloat %2888, i64 1, !dbg !40 + %3201 = bitcast <2 x bfloat> %3200 to i32, !dbg !40 + %3202 = insertelement <2 x bfloat> poison, bfloat %2889, i64 0, !dbg !40 + %3203 = insertelement <2 x bfloat> %3202, bfloat %2890, i64 1, !dbg !40 + %3204 = bitcast <2 x bfloat> %3203 to i32, !dbg !40 + %3205 = insertelement <2 x bfloat> poison, bfloat %2891, i64 0, !dbg !40 + %3206 = insertelement <2 x bfloat> %3205, bfloat %2892, i64 1, !dbg !40 + %3207 = bitcast <2 x bfloat> %3206 to i32, !dbg !40 + %3208 = insertelement <2 x bfloat> poison, bfloat %2893, i64 0, !dbg !40 + %3209 = insertelement <2 x bfloat> %3208, bfloat %2894, i64 1, !dbg !40 + %3210 = bitcast <2 x bfloat> %3209 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3201, i32 %3204, i32 %3207, i32 %3210, ptr addrspace(1) %2456, i1 %423) #7, !dbg !40 + %3211 = insertelement <2 x bfloat> poison, bfloat %2895, i64 0, !dbg !40 + %3212 = insertelement <2 x bfloat> %3211, bfloat %2896, i64 1, !dbg !40 + %3213 = bitcast <2 x bfloat> %3212 to i32, !dbg !40 + %3214 = insertelement <2 x bfloat> poison, bfloat %2897, i64 0, !dbg !40 + %3215 = insertelement <2 x bfloat> %3214, bfloat %2898, i64 1, !dbg !40 + %3216 = bitcast <2 x bfloat> %3215 to i32, !dbg !40 + %3217 = insertelement <2 x bfloat> poison, bfloat %2899, i64 0, !dbg !40 + %3218 = insertelement <2 x bfloat> %3217, bfloat %2900, i64 1, !dbg !40 + %3219 = bitcast <2 x bfloat> %3218 to i32, !dbg !40 + %3220 = insertelement <2 x bfloat> poison, bfloat %2901, i64 0, !dbg !40 + %3221 = insertelement <2 x bfloat> %3220, bfloat %2902, i64 1, !dbg !40 + %3222 = bitcast <2 x bfloat> %3221 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3213, i32 %3216, i32 %3219, i32 %3222, ptr addrspace(1) %2454, i1 %441) #7, !dbg !40 + %3223 = insertelement <2 x bfloat> poison, bfloat %2903, i64 0, !dbg !40 + %3224 = insertelement <2 x bfloat> %3223, bfloat %2904, i64 1, !dbg !40 + %3225 = bitcast <2 x bfloat> %3224 to i32, !dbg !40 + %3226 = insertelement <2 x bfloat> poison, bfloat %2905, i64 0, !dbg !40 + %3227 = insertelement <2 x bfloat> %3226, bfloat %2906, i64 1, !dbg !40 + %3228 = bitcast <2 x bfloat> %3227 to i32, !dbg !40 + %3229 = insertelement <2 x bfloat> poison, bfloat %2907, i64 0, !dbg !40 + %3230 = insertelement <2 x bfloat> %3229, bfloat %2908, i64 1, !dbg !40 + %3231 = bitcast <2 x bfloat> %3230 to i32, !dbg !40 + %3232 = insertelement <2 x bfloat> poison, bfloat %2909, i64 0, !dbg !40 + %3233 = insertelement <2 x bfloat> %3232, bfloat %2910, i64 1, !dbg !40 + %3234 = bitcast <2 x bfloat> %3233 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3225, i32 %3228, i32 %3231, i32 %3234, ptr addrspace(1) %2452, i1 %459) #7, !dbg !40 + %3235 = insertelement <2 x bfloat> poison, bfloat %2911, i64 0, !dbg !40 + %3236 = insertelement <2 x bfloat> %3235, bfloat %2912, i64 1, !dbg !40 + %3237 = bitcast <2 x bfloat> %3236 to i32, !dbg !40 + %3238 = insertelement <2 x bfloat> poison, bfloat %2913, i64 0, !dbg !40 + %3239 = insertelement <2 x bfloat> %3238, bfloat %2914, i64 1, !dbg !40 + %3240 = bitcast <2 x bfloat> %3239 to i32, !dbg !40 + %3241 = insertelement <2 x bfloat> poison, bfloat %2915, i64 0, !dbg !40 + %3242 = insertelement <2 x bfloat> %3241, bfloat %2916, i64 1, !dbg !40 + %3243 = bitcast <2 x bfloat> %3242 to i32, !dbg !40 + %3244 = insertelement <2 x bfloat> poison, bfloat %2917, i64 0, !dbg !40 + %3245 = insertelement <2 x bfloat> %3244, bfloat %2918, i64 1, !dbg !40 + %3246 = bitcast <2 x bfloat> %3245 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3237, i32 %3240, i32 %3243, i32 %3246, ptr addrspace(1) %2450, i1 %477) #7, !dbg !40 + %3247 = insertelement <2 x bfloat> poison, bfloat %2919, i64 0, !dbg !40 + %3248 = insertelement <2 x bfloat> %3247, bfloat %2920, i64 1, !dbg !40 + %3249 = bitcast <2 x bfloat> %3248 to i32, !dbg !40 + %3250 = insertelement <2 x bfloat> poison, bfloat %2921, i64 0, !dbg !40 + %3251 = insertelement <2 x bfloat> %3250, bfloat %2922, i64 1, !dbg !40 + %3252 = bitcast <2 x bfloat> %3251 to i32, !dbg !40 + %3253 = insertelement <2 x bfloat> poison, bfloat %2923, i64 0, !dbg !40 + %3254 = insertelement <2 x bfloat> %3253, bfloat %2924, i64 1, !dbg !40 + %3255 = bitcast <2 x bfloat> %3254 to i32, !dbg !40 + %3256 = insertelement <2 x bfloat> poison, bfloat %2925, i64 0, !dbg !40 + %3257 = insertelement <2 x bfloat> %3256, bfloat %2926, i64 1, !dbg !40 + %3258 = bitcast <2 x bfloat> %3257 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3249, i32 %3252, i32 %3255, i32 %3258, ptr addrspace(1) %2448, i1 %495) #7, !dbg !40 + %3259 = insertelement <2 x bfloat> poison, bfloat %2927, i64 0, !dbg !40 + %3260 = insertelement <2 x bfloat> %3259, bfloat %2928, i64 1, !dbg !40 + %3261 = bitcast <2 x bfloat> %3260 to i32, !dbg !40 + %3262 = insertelement <2 x bfloat> poison, bfloat %2929, i64 0, !dbg !40 + %3263 = insertelement <2 x bfloat> %3262, bfloat %2930, i64 1, !dbg !40 + %3264 = bitcast <2 x bfloat> %3263 to i32, !dbg !40 + %3265 = insertelement <2 x bfloat> poison, bfloat %2931, i64 0, !dbg !40 + %3266 = insertelement <2 x bfloat> %3265, bfloat %2932, i64 1, !dbg !40 + %3267 = bitcast <2 x bfloat> %3266 to i32, !dbg !40 + %3268 = insertelement <2 x bfloat> poison, bfloat %2933, i64 0, !dbg !40 + %3269 = insertelement <2 x bfloat> %3268, bfloat %2934, i64 1, !dbg !40 + %3270 = bitcast <2 x bfloat> %3269 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3261, i32 %3264, i32 %3267, i32 %3270, ptr addrspace(1) %2446, i1 %513) #7, !dbg !40 + %3271 = insertelement <2 x bfloat> poison, bfloat %2935, i64 0, !dbg !40 + %3272 = insertelement <2 x bfloat> %3271, bfloat %2936, i64 1, !dbg !40 + %3273 = bitcast <2 x bfloat> %3272 to i32, !dbg !40 + %3274 = insertelement <2 x bfloat> poison, bfloat %2937, i64 0, !dbg !40 + %3275 = insertelement <2 x bfloat> %3274, bfloat %2938, i64 1, !dbg !40 + %3276 = bitcast <2 x bfloat> %3275 to i32, !dbg !40 + %3277 = insertelement <2 x bfloat> poison, bfloat %2939, i64 0, !dbg !40 + %3278 = insertelement <2 x bfloat> %3277, bfloat %2940, i64 1, !dbg !40 + %3279 = bitcast <2 x bfloat> %3278 to i32, !dbg !40 + %3280 = insertelement <2 x bfloat> poison, bfloat %2941, i64 0, !dbg !40 + %3281 = insertelement <2 x bfloat> %3280, bfloat %2942, i64 1, !dbg !40 + %3282 = bitcast <2 x bfloat> %3281 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3273, i32 %3276, i32 %3279, i32 %3282, ptr addrspace(1) %2444, i1 %531) #7, !dbg !40 + %3283 = insertelement <2 x bfloat> poison, bfloat %2943, i64 0, !dbg !40 + %3284 = insertelement <2 x bfloat> %3283, bfloat %2944, i64 1, !dbg !40 + %3285 = bitcast <2 x bfloat> %3284 to i32, !dbg !40 + %3286 = insertelement <2 x bfloat> poison, bfloat %2945, i64 0, !dbg !40 + %3287 = insertelement <2 x bfloat> %3286, bfloat %2946, i64 1, !dbg !40 + %3288 = bitcast <2 x bfloat> %3287 to i32, !dbg !40 + %3289 = insertelement <2 x bfloat> poison, bfloat %2947, i64 0, !dbg !40 + %3290 = insertelement <2 x bfloat> %3289, bfloat %2948, i64 1, !dbg !40 + %3291 = bitcast <2 x bfloat> %3290 to i32, !dbg !40 + %3292 = insertelement <2 x bfloat> poison, bfloat %2949, i64 0, !dbg !40 + %3293 = insertelement <2 x bfloat> %3292, bfloat %2950, i64 1, !dbg !40 + %3294 = bitcast <2 x bfloat> %3293 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3285, i32 %3288, i32 %3291, i32 %3294, ptr addrspace(1) %2442, i1 %549) #7, !dbg !40 + %3295 = insertelement <2 x bfloat> poison, bfloat %2951, i64 0, !dbg !40 + %3296 = insertelement <2 x bfloat> %3295, bfloat %2952, i64 1, !dbg !40 + %3297 = bitcast <2 x bfloat> %3296 to i32, !dbg !40 + %3298 = insertelement <2 x bfloat> poison, bfloat %2953, i64 0, !dbg !40 + %3299 = insertelement <2 x bfloat> %3298, bfloat %2954, i64 1, !dbg !40 + %3300 = bitcast <2 x bfloat> %3299 to i32, !dbg !40 + %3301 = insertelement <2 x bfloat> poison, bfloat %2955, i64 0, !dbg !40 + %3302 = insertelement <2 x bfloat> %3301, bfloat %2956, i64 1, !dbg !40 + %3303 = bitcast <2 x bfloat> %3302 to i32, !dbg !40 + %3304 = insertelement <2 x bfloat> poison, bfloat %2957, i64 0, !dbg !40 + %3305 = insertelement <2 x bfloat> %3304, bfloat %2958, i64 1, !dbg !40 + %3306 = bitcast <2 x bfloat> %3305 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3297, i32 %3300, i32 %3303, i32 %3306, ptr addrspace(1) %2440, i1 %567) #7, !dbg !40 + %3307 = insertelement <2 x bfloat> poison, bfloat %2959, i64 0, !dbg !40 + %3308 = insertelement <2 x bfloat> %3307, bfloat %2960, i64 1, !dbg !40 + %3309 = bitcast <2 x bfloat> %3308 to i32, !dbg !40 + %3310 = insertelement <2 x bfloat> poison, bfloat %2961, i64 0, !dbg !40 + %3311 = insertelement <2 x bfloat> %3310, bfloat %2962, i64 1, !dbg !40 + %3312 = bitcast <2 x bfloat> %3311 to i32, !dbg !40 + %3313 = insertelement <2 x bfloat> poison, bfloat %2963, i64 0, !dbg !40 + %3314 = insertelement <2 x bfloat> %3313, bfloat %2964, i64 1, !dbg !40 + %3315 = bitcast <2 x bfloat> %3314 to i32, !dbg !40 + %3316 = insertelement <2 x bfloat> poison, bfloat %2965, i64 0, !dbg !40 + %3317 = insertelement <2 x bfloat> %3316, bfloat %2966, i64 1, !dbg !40 + %3318 = bitcast <2 x bfloat> %3317 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3309, i32 %3312, i32 %3315, i32 %3318, ptr addrspace(1) %2438, i1 %585) #7, !dbg !40 + %3319 = insertelement <2 x bfloat> poison, bfloat %2967, i64 0, !dbg !40 + %3320 = insertelement <2 x bfloat> %3319, bfloat %2968, i64 1, !dbg !40 + %3321 = bitcast <2 x bfloat> %3320 to i32, !dbg !40 + %3322 = insertelement <2 x bfloat> poison, bfloat %2969, i64 0, !dbg !40 + %3323 = insertelement <2 x bfloat> %3322, bfloat %2970, i64 1, !dbg !40 + %3324 = bitcast <2 x bfloat> %3323 to i32, !dbg !40 + %3325 = insertelement <2 x bfloat> poison, bfloat %2971, i64 0, !dbg !40 + %3326 = insertelement <2 x bfloat> %3325, bfloat %2972, i64 1, !dbg !40 + %3327 = bitcast <2 x bfloat> %3326 to i32, !dbg !40 + %3328 = insertelement <2 x bfloat> poison, bfloat %2973, i64 0, !dbg !40 + %3329 = insertelement <2 x bfloat> %3328, bfloat %2974, i64 1, !dbg !40 + %3330 = bitcast <2 x bfloat> %3329 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3321, i32 %3324, i32 %3327, i32 %3330, ptr addrspace(1) %2436, i1 %603) #7, !dbg !40 + %3331 = insertelement <2 x bfloat> poison, bfloat %2975, i64 0, !dbg !40 + %3332 = insertelement <2 x bfloat> %3331, bfloat %2976, i64 1, !dbg !40 + %3333 = bitcast <2 x bfloat> %3332 to i32, !dbg !40 + %3334 = insertelement <2 x bfloat> poison, bfloat %2977, i64 0, !dbg !40 + %3335 = insertelement <2 x bfloat> %3334, bfloat %2978, i64 1, !dbg !40 + %3336 = bitcast <2 x bfloat> %3335 to i32, !dbg !40 + %3337 = insertelement <2 x bfloat> poison, bfloat %2979, i64 0, !dbg !40 + %3338 = insertelement <2 x bfloat> %3337, bfloat %2980, i64 1, !dbg !40 + %3339 = bitcast <2 x bfloat> %3338 to i32, !dbg !40 + %3340 = insertelement <2 x bfloat> poison, bfloat %2981, i64 0, !dbg !40 + %3341 = insertelement <2 x bfloat> %3340, bfloat %2982, i64 1, !dbg !40 + %3342 = bitcast <2 x bfloat> %3341 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3333, i32 %3336, i32 %3339, i32 %3342, ptr addrspace(1) %2434, i1 %621) #7, !dbg !40 + %3343 = insertelement <2 x bfloat> poison, bfloat %2983, i64 0, !dbg !40 + %3344 = insertelement <2 x bfloat> %3343, bfloat %2984, i64 1, !dbg !40 + %3345 = bitcast <2 x bfloat> %3344 to i32, !dbg !40 + %3346 = insertelement <2 x bfloat> poison, bfloat %2985, i64 0, !dbg !40 + %3347 = insertelement <2 x bfloat> %3346, bfloat %2986, i64 1, !dbg !40 + %3348 = bitcast <2 x bfloat> %3347 to i32, !dbg !40 + %3349 = insertelement <2 x bfloat> poison, bfloat %2987, i64 0, !dbg !40 + %3350 = insertelement <2 x bfloat> %3349, bfloat %2988, i64 1, !dbg !40 + %3351 = bitcast <2 x bfloat> %3350 to i32, !dbg !40 + %3352 = insertelement <2 x bfloat> poison, bfloat %2989, i64 0, !dbg !40 + %3353 = insertelement <2 x bfloat> %3352, bfloat %2990, i64 1, !dbg !40 + %3354 = bitcast <2 x bfloat> %3353 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3345, i32 %3348, i32 %3351, i32 %3354, ptr addrspace(1) %2432, i1 %639) #7, !dbg !40 + %3355 = insertelement <2 x bfloat> poison, bfloat %2991, i64 0, !dbg !40 + %3356 = insertelement <2 x bfloat> %3355, bfloat %2992, i64 1, !dbg !40 + %3357 = bitcast <2 x bfloat> %3356 to i32, !dbg !40 + %3358 = insertelement <2 x bfloat> poison, bfloat %2993, i64 0, !dbg !40 + %3359 = insertelement <2 x bfloat> %3358, bfloat %2994, i64 1, !dbg !40 + %3360 = bitcast <2 x bfloat> %3359 to i32, !dbg !40 + %3361 = insertelement <2 x bfloat> poison, bfloat %2995, i64 0, !dbg !40 + %3362 = insertelement <2 x bfloat> %3361, bfloat %2996, i64 1, !dbg !40 + %3363 = bitcast <2 x bfloat> %3362 to i32, !dbg !40 + %3364 = insertelement <2 x bfloat> poison, bfloat %2997, i64 0, !dbg !40 + %3365 = insertelement <2 x bfloat> %3364, bfloat %2998, i64 1, !dbg !40 + %3366 = bitcast <2 x bfloat> %3365 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3357, i32 %3360, i32 %3363, i32 %3366, ptr addrspace(1) %2430, i1 %657) #7, !dbg !40 + %3367 = insertelement <2 x bfloat> poison, bfloat %2999, i64 0, !dbg !40 + %3368 = insertelement <2 x bfloat> %3367, bfloat %3000, i64 1, !dbg !40 + %3369 = bitcast <2 x bfloat> %3368 to i32, !dbg !40 + %3370 = insertelement <2 x bfloat> poison, bfloat %3001, i64 0, !dbg !40 + %3371 = insertelement <2 x bfloat> %3370, bfloat %3002, i64 1, !dbg !40 + %3372 = bitcast <2 x bfloat> %3371 to i32, !dbg !40 + %3373 = insertelement <2 x bfloat> poison, bfloat %3003, i64 0, !dbg !40 + %3374 = insertelement <2 x bfloat> %3373, bfloat %3004, i64 1, !dbg !40 + %3375 = bitcast <2 x bfloat> %3374 to i32, !dbg !40 + %3376 = insertelement <2 x bfloat> poison, bfloat %3005, i64 0, !dbg !40 + %3377 = insertelement <2 x bfloat> %3376, bfloat %3006, i64 1, !dbg !40 + %3378 = bitcast <2 x bfloat> %3377 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3369, i32 %3372, i32 %3375, i32 %3378, ptr addrspace(1) %2428, i1 %675) #7, !dbg !40 + %3379 = insertelement <2 x bfloat> poison, bfloat %3007, i64 0, !dbg !40 + %3380 = insertelement <2 x bfloat> %3379, bfloat %3008, i64 1, !dbg !40 + %3381 = bitcast <2 x bfloat> %3380 to i32, !dbg !40 + %3382 = insertelement <2 x bfloat> poison, bfloat %3009, i64 0, !dbg !40 + %3383 = insertelement <2 x bfloat> %3382, bfloat %3010, i64 1, !dbg !40 + %3384 = bitcast <2 x bfloat> %3383 to i32, !dbg !40 + %3385 = insertelement <2 x bfloat> poison, bfloat %3011, i64 0, !dbg !40 + %3386 = insertelement <2 x bfloat> %3385, bfloat %3012, i64 1, !dbg !40 + %3387 = bitcast <2 x bfloat> %3386 to i32, !dbg !40 + %3388 = insertelement <2 x bfloat> poison, bfloat %3013, i64 0, !dbg !40 + %3389 = insertelement <2 x bfloat> %3388, bfloat %3014, i64 1, !dbg !40 + %3390 = bitcast <2 x bfloat> %3389 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3381, i32 %3384, i32 %3387, i32 %3390, ptr addrspace(1) %2426, i1 %693) #7, !dbg !40 + %3391 = insertelement <2 x bfloat> poison, bfloat %3015, i64 0, !dbg !40 + %3392 = insertelement <2 x bfloat> %3391, bfloat %3016, i64 1, !dbg !40 + %3393 = bitcast <2 x bfloat> %3392 to i32, !dbg !40 + %3394 = insertelement <2 x bfloat> poison, bfloat %3017, i64 0, !dbg !40 + %3395 = insertelement <2 x bfloat> %3394, bfloat %3018, i64 1, !dbg !40 + %3396 = bitcast <2 x bfloat> %3395 to i32, !dbg !40 + %3397 = insertelement <2 x bfloat> poison, bfloat %3019, i64 0, !dbg !40 + %3398 = insertelement <2 x bfloat> %3397, bfloat %3020, i64 1, !dbg !40 + %3399 = bitcast <2 x bfloat> %3398 to i32, !dbg !40 + %3400 = insertelement <2 x bfloat> poison, bfloat %3021, i64 0, !dbg !40 + %3401 = insertelement <2 x bfloat> %3400, bfloat %3022, i64 1, !dbg !40 + %3402 = bitcast <2 x bfloat> %3401 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3393, i32 %3396, i32 %3399, i32 %3402, ptr addrspace(1) %2424, i1 %711) #7, !dbg !40 + %3403 = insertelement <2 x bfloat> poison, bfloat %3023, i64 0, !dbg !40 + %3404 = insertelement <2 x bfloat> %3403, bfloat %3024, i64 1, !dbg !40 + %3405 = bitcast <2 x bfloat> %3404 to i32, !dbg !40 + %3406 = insertelement <2 x bfloat> poison, bfloat %3025, i64 0, !dbg !40 + %3407 = insertelement <2 x bfloat> %3406, bfloat %3026, i64 1, !dbg !40 + %3408 = bitcast <2 x bfloat> %3407 to i32, !dbg !40 + %3409 = insertelement <2 x bfloat> poison, bfloat %3027, i64 0, !dbg !40 + %3410 = insertelement <2 x bfloat> %3409, bfloat %3028, i64 1, !dbg !40 + %3411 = bitcast <2 x bfloat> %3410 to i32, !dbg !40 + %3412 = insertelement <2 x bfloat> poison, bfloat %3029, i64 0, !dbg !40 + %3413 = insertelement <2 x bfloat> %3412, bfloat %3030, i64 1, !dbg !40 + %3414 = bitcast <2 x bfloat> %3413 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %3405, i32 %3408, i32 %3411, i32 %3414, ptr addrspace(1) %2422, i1 %729) #7, !dbg !40 + %3415 = shl nuw nsw i32 %12, 3, !dbg !41 + %3416 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %3415, !dbg !41 + %3417 = bitcast float %2487 to i32, !dbg !41 + %3418 = bitcast float %2488 to i32, !dbg !41 + %3419 = bitcast float %2489 to i32, !dbg !41 + %3420 = bitcast float %2490 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %3416, i32 %3417, i32 %3418, i32 %3419, i32 %3420) #7, !dbg !41 + %3421 = getelementptr inbounds nuw i8, ptr addrspace(3) %3416, i32 16, !dbg !41 + %3422 = bitcast float %2491 to i32, !dbg !41 + %3423 = bitcast float %2492 to i32, !dbg !41 + %3424 = bitcast float %2493 to i32, !dbg !41 + %3425 = bitcast float %2494 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %3421, i32 %3422, i32 %3423, i32 %3424, i32 %3425) #7, !dbg !41 + %3426 = getelementptr inbounds nuw i8, ptr addrspace(3) %3416, i32 32, !dbg !41 + %3427 = bitcast float %2495 to i32, !dbg !41 + %3428 = bitcast float %2496 to i32, !dbg !41 + %3429 = bitcast float %2497 to i32, !dbg !41 + %3430 = bitcast float %2498 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %3426, i32 %3427, i32 %3428, i32 %3429, i32 %3430) #7, !dbg !41 + %3431 = getelementptr inbounds nuw i8, ptr addrspace(3) %3416, i32 48, !dbg !41 + %3432 = bitcast float %2499 to i32, !dbg !41 + %3433 = bitcast float %2500 to i32, !dbg !41 + %3434 = bitcast float %2501 to i32, !dbg !41 + %3435 = bitcast float %2502 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %3431, i32 %3432, i32 %3433, i32 %3434, i32 %3435) #7, !dbg !41 + %3436 = getelementptr inbounds nuw i8, ptr addrspace(3) %3416, i32 64, !dbg !41 + %3437 = bitcast float %2503 to i32, !dbg !41 + %3438 = bitcast float %2504 to i32, !dbg !41 + %3439 = bitcast float %2505 to i32, !dbg !41 + %3440 = bitcast float %2506 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %3436, i32 %3437, i32 %3438, i32 %3439, i32 %3440) #7, !dbg !41 + %3441 = getelementptr inbounds nuw i8, ptr addrspace(3) %3416, i32 80, !dbg !41 + %3442 = bitcast float %2507 to i32, !dbg !41 + %3443 = bitcast float %2508 to i32, !dbg !41 + %3444 = bitcast float %2509 to i32, !dbg !41 + %3445 = bitcast float %2510 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %3441, i32 %3442, i32 %3443, i32 %3444, i32 %3445) #7, !dbg !41 + %3446 = getelementptr inbounds nuw i8, ptr addrspace(3) %3416, i32 96, !dbg !41 + %3447 = bitcast float %2511 to i32, !dbg !41 + %3448 = bitcast float %2512 to i32, !dbg !41 + %3449 = bitcast float %2513 to i32, !dbg !41 + %3450 = bitcast float %2514 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %3446, i32 %3447, i32 %3448, i32 %3449, i32 %3450) #7, !dbg !41 + %3451 = getelementptr inbounds nuw i8, ptr addrspace(3) %3416, i32 112, !dbg !41 + %3452 = bitcast float %2515 to i32, !dbg !41 + %3453 = bitcast float %2516 to i32, !dbg !41 + %3454 = bitcast float %2517 to i32, !dbg !41 + %3455 = bitcast float %2518 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %3451, i32 %3452, i32 %3453, i32 %3454, i32 %3455) #7, !dbg !41 + tail call void @llvm.nvvm.bar.warp.sync(i32 -1), !dbg !41 + %3456 = shl nuw nsw i32 %40, 4, !dbg !41 + %3457 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %3456, !dbg !41 + %3458 = tail call { i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.b16.p3(ptr addrspace(3) %3457), !dbg !41 + %3459 = extractvalue { i32, i32 } %3458, 0, !dbg !41 + %3460 = extractvalue { i32, i32 } %3458, 1, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %3459, ptr addrspace(1) %2420, i1 %2486) #7, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %3460, ptr addrspace(1) %2417, i1 %2485) #7, !dbg !41 + ret void, !dbg !42 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.div.full(float, float) #3 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.bar.warp.sync(i32) #4 + +; Function Attrs: convergent nocallback nofree nounwind memory(argmem: read) +declare { i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.b16.p3(ptr addrspace(3) readonly captures(none)) #5 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #6 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.f(float) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="32" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) } +attributes #4 = { convergent nocallback nounwind } +attributes #5 = { convergent nocallback nofree nounwind memory(argmem: read) } +attributes #6 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #7 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "l2norm_fwd_kernel", linkageName: "l2norm_fwd_kernel", scope: !1, file: !1, line: 69, type: !6, scopeLine: 69, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !10, !11, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 70, column: 24, scope: !5) +!14 = !DILocation(line: 70, column: 30, scope: !5) +!15 = !DILocation(line: 71, column: 16, scope: !5) +!16 = !DILocation(line: 71, column: 34, scope: !5) +!17 = !DILocation(line: 71, column: 21, scope: !5) +!18 = !DILocation(line: 73, column: 16, scope: !5) +!19 = !DILocation(line: 74, column: 30, scope: !5) +!20 = !DILocation(line: 75, column: 29, scope: !5) +!21 = !DILocation(line: 75, column: 14, scope: !5) +!22 = !DILocation(line: 75, column: 33, scope: !5) +!23 = !DILocation(line: 79, column: 18, scope: !5) +!24 = !DILocation(line: 79, column: 47, scope: !5) +!25 = !DILocation(line: 80, column: 38, scope: !5) +!26 = !DILocation(line: 263, column: 15, scope: !27, inlinedAt: !29) +!27 = distinct !DILexicalBlockFile(scope: !5, file: !28, discriminator: 0) +!28 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!29 = !DILocation(line: 293, column: 36, scope: !27, inlinedAt: !30) +!30 = !DILocation(line: 80, column: 43, scope: !31) +!31 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!32 = !DILocation(line: 80, column: 48, scope: !5) +!33 = !DILocation(line: 80, column: 25, scope: !5) +!34 = !DILocation(line: 77, column: 20, scope: !5) +!35 = !DILocation(line: 76, column: 14, scope: !5) +!36 = !DILocation(line: 76, column: 33, scope: !5) +!37 = !DILocation(line: 80, column: 17, scope: !5) +!38 = !DILocation(line: 81, column: 16, scope: !5) +!39 = !DILocation(line: 83, column: 25, scope: !5) +!40 = !DILocation(line: 83, column: 18, scope: !5) +!41 = !DILocation(line: 84, column: 21, scope: !5) +!42 = !DILocation(line: 84, column: 4, scope: !5) diff --git a/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.ptx b/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..0bc04fbb0005fa7b3bcfc7cbd415407a1aba40a9 --- /dev/null +++ b/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.ptx @@ -0,0 +1,3012 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_fwd_kernel // -- Begin function l2norm_fwd_kernel +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; +.global .align 1 .b8 _$_str_$_2[17] = {95, 95, 67, 85, 68, 65, 95, 80, 82, 69, 67, 95, 83, 81, 82, 84}; + // @l2norm_fwd_kernel +.visible .entry l2norm_fwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_2, + .param .f32 l2norm_fwd_kernel_param_3, + .param .u32 l2norm_fwd_kernel_param_4, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_6 +) +.reqntid 32 +{ + .reg .pred %p<50>; + .reg .b16 %rs<310>; + .reg .b32 %r<1465>; + .reg .b64 %rd<238>; + .loc 1 69 0 // l2norm.py:69:0 +$L__func_begin0: + .loc 1 69 0 // l2norm.py:69:0 + +// %bb.0: // %__nv_sqrtf.exit + ld.param.b64 %rd67, [l2norm_fwd_kernel_param_0]; + ld.param.b64 %rd68, [l2norm_fwd_kernel_param_1]; +$L__tmp0: + .loc 1 70 24 // l2norm.py:70:24 + mov.u32 %r300, %ctaid.x; + ld.param.b64 %rd69, [l2norm_fwd_kernel_param_2]; + .loc 1 71 16 // l2norm.py:71:16 + mul.wide.u32 %rd70, %r300, 64; + ld.param.b32 %r301, [l2norm_fwd_kernel_param_3]; + ld.param.s32 %rd71, [l2norm_fwd_kernel_param_4]; + .loc 1 71 34 // l2norm.py:71:34 + mov.u32 %r302, %tid.x; + and.b32 %r303, %r302, 16; + bfe.u32 %r304, %r302, 4, 1; + or.b32 %r305, %r304, 2; + .loc 1 71 21 // l2norm.py:71:21 + or.b32 %r306, %r304, 60; + or.b32 %r307, %r304, 62; + or.b32 %r308, %r304, 56; + or.b32 %r309, %r304, 58; + or.b32 %r310, %r304, 52; + or.b32 %r311, %r304, 54; + or.b32 %r312, %r304, 48; + or.b32 %r313, %r304, 50; + or.b32 %r314, %r304, 44; + or.b32 %r315, %r304, 46; + or.b32 %r316, %r304, 40; + or.b32 %r317, %r304, 42; + or.b32 %r318, %r304, 36; + or.b32 %r319, %r304, 38; + or.b32 %r320, %r304, 32; + or.b32 %r321, %r304, 34; + or.b32 %r322, %r304, 16; + or.b32 %r323, %r304, 18; + or.b32 %r324, %r304, 20; + or.b32 %r325, %r304, 22; + or.b32 %r326, %r304, 24; + or.b32 %r327, %r304, 26; + or.b32 %r328, %r304, 28; + or.b32 %r329, %r304, 30; + or.b32 %r330, %r304, 8; + or.b32 %r331, %r304, 10; + or.b32 %r332, %r304, 12; + or.b32 %r333, %r304, 14; + or.b32 %r334, %r304, 4; + or.b32 %r335, %r304, 6; + cvt.u64.u32 %rd72, %r321; + cvt.u64.u32 %rd73, %r320; + cvt.u64.u32 %rd74, %r319; + cvt.u64.u32 %rd75, %r318; + cvt.u64.u32 %rd76, %r317; + cvt.u64.u32 %rd77, %r316; + cvt.u64.u32 %rd78, %r315; + cvt.u64.u32 %rd79, %r314; + cvt.u64.u32 %rd80, %r313; + cvt.u64.u32 %rd81, %r312; + cvt.u64.u32 %rd82, %r311; + cvt.u64.u32 %rd83, %r310; + cvt.u64.u32 %rd84, %r309; + cvt.u64.u32 %rd85, %r308; + cvt.u64.u32 %rd86, %r307; + cvt.u64.u32 %rd87, %r306; + cvt.u64.u32 %rd88, %r329; + cvt.u64.u32 %rd89, %r328; + cvt.u64.u32 %rd90, %r327; + cvt.u64.u32 %rd91, %r326; + cvt.u64.u32 %rd92, %r325; + cvt.u64.u32 %rd93, %r324; + cvt.u64.u32 %rd94, %r323; + cvt.u64.u32 %rd95, %r322; + cvt.u64.u32 %rd96, %r333; + cvt.u64.u32 %rd97, %r332; + cvt.u64.u32 %rd98, %r331; + cvt.u64.u32 %rd99, %r330; + cvt.u64.u32 %rd100, %r335; + cvt.u64.u32 %rd101, %r334; + cvt.u64.u32 %rd102, %r305; + cvt.u64.u32 %rd103, %r304; + or.b64 %rd104, %rd70, %rd103; + or.b64 %rd105, %rd70, %rd102; + or.b64 %rd106, %rd70, %rd101; + or.b64 %rd107, %rd70, %rd100; + or.b64 %rd108, %rd70, %rd99; + or.b64 %rd109, %rd70, %rd98; + or.b64 %rd110, %rd70, %rd97; + or.b64 %rd111, %rd70, %rd96; + or.b64 %rd112, %rd70, %rd95; + or.b64 %rd113, %rd70, %rd94; + or.b64 %rd114, %rd70, %rd93; + or.b64 %rd115, %rd70, %rd92; + or.b64 %rd116, %rd70, %rd91; + or.b64 %rd117, %rd70, %rd90; + or.b64 %rd118, %rd70, %rd89; + or.b64 %rd119, %rd70, %rd88; + or.b64 %rd120, %rd70, %rd87; + or.b64 %rd121, %rd70, %rd86; + or.b64 %rd122, %rd70, %rd85; + or.b64 %rd123, %rd70, %rd84; + or.b64 %rd124, %rd70, %rd83; + or.b64 %rd125, %rd70, %rd82; + or.b64 %rd126, %rd70, %rd81; + or.b64 %rd127, %rd70, %rd80; + or.b64 %rd128, %rd70, %rd79; + or.b64 %rd129, %rd70, %rd78; + or.b64 %rd130, %rd70, %rd77; + or.b64 %rd131, %rd70, %rd76; + or.b64 %rd132, %rd70, %rd75; + or.b64 %rd133, %rd70, %rd74; + or.b64 %rd134, %rd70, %rd73; + or.b64 %rd135, %rd70, %rd72; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p35, %rd135, %rd71; + selp.b16 %rs1, 1, 0, %p35; + shl.b16 %rs2, %rs1, 2; + setp.lt.s64 %p36, %rd134, %rd71; + selp.b16 %rs3, -1, 0, %p36; + shl.b16 %rs4, %rs3, 3; + or.b16 %rs5, %rs4, %rs2; + setp.lt.s64 %p37, %rd133, %rd71; + selp.b16 %rs6, 1, 0, %p37; + setp.lt.s64 %p38, %rd132, %rd71; + selp.b16 %rs7, -1, 0, %p38; + shl.b16 %rs8, %rs7, 1; + or.b16 %rs9, %rs6, %rs8; + and.b16 %rs10, %rs9, 3; + or.b16 %rs11, %rs10, %rs5; + shl.b16 %rs12, %rs11, 12; + setp.lt.s64 %p39, %rd131, %rd71; + selp.b16 %rs13, 1, 0, %p39; + shl.b16 %rs14, %rs13, 2; + setp.lt.s64 %p40, %rd130, %rd71; + selp.b16 %rs15, -1, 0, %p40; + shl.b16 %rs16, %rs15, 3; + or.b16 %rs17, %rs16, %rs14; + setp.lt.s64 %p41, %rd129, %rd71; + selp.b16 %rs18, 1, 0, %p41; + setp.lt.s64 %p42, %rd128, %rd71; + selp.b16 %rs19, -1, 0, %p42; + shl.b16 %rs20, %rs19, 1; + or.b16 %rs21, %rs18, %rs20; + and.b16 %rs22, %rs21, 3; + or.b16 %rs23, %rs22, %rs17; + and.b16 %rs24, %rs23, 15; + shl.b16 %rs25, %rs24, 8; + or.b16 %rs26, %rs12, %rs25; + setp.lt.s64 %p43, %rd127, %rd71; + selp.b16 %rs27, 1, 0, %p43; + shl.b16 %rs28, %rs27, 2; + setp.lt.s64 %p44, %rd126, %rd71; + selp.b16 %rs29, -1, 0, %p44; + shl.b16 %rs30, %rs29, 3; + or.b16 %rs31, %rs30, %rs28; + setp.lt.s64 %p45, %rd125, %rd71; + selp.b16 %rs32, 1, 0, %p45; + setp.lt.s64 %p46, %rd124, %rd71; + selp.b16 %rs33, -1, 0, %p46; + shl.b16 %rs34, %rs33, 1; + or.b16 %rs35, %rs32, %rs34; + and.b16 %rs36, %rs35, 3; + or.b16 %rs37, %rs36, %rs31; + shl.b16 %rs38, %rs37, 4; + setp.lt.s64 %p47, %rd123, %rd71; + selp.b16 %rs39, 1, 0, %p47; + shl.b16 %rs40, %rs39, 2; + setp.lt.s64 %p48, %rd122, %rd71; + selp.b16 %rs41, -1, 0, %p48; + shl.b16 %rs42, %rs41, 3; + or.b16 %rs43, %rs42, %rs40; + setp.lt.s64 %p32, %rd121, %rd71; + selp.b16 %rs44, 1, 0, %p32; + setp.lt.s64 %p49, %rd120, %rd71; + selp.b16 %rs45, -1, 0, %p49; + shl.b16 %rs46, %rs45, 1; + or.b16 %rs47, %rs44, %rs46; + and.b16 %rs48, %rs47, 3; + or.b16 %rs49, %rs48, %rs43; + and.b16 %rs50, %rs49, 15; + or.b16 %rs51, %rs50, %rs38; + and.b16 %rs52, %rs51, 255; + or.b16 %rs53, %rs52, %rs26; + cvt.u32.u16 %r336, %rs53; + setp.lt.s64 %p16, %rd119, %rd71; + setp.lt.s64 %p15, %rd118, %rd71; + setp.lt.s64 %p14, %rd117, %rd71; + setp.lt.s64 %p13, %rd116, %rd71; + setp.lt.s64 %p12, %rd115, %rd71; + setp.lt.s64 %p11, %rd114, %rd71; + setp.lt.s64 %p10, %rd113, %rd71; + setp.lt.s64 %p9, %rd112, %rd71; + setp.lt.s64 %p1, %rd111, %rd71; + setp.lt.s64 %p8, %rd110, %rd71; + setp.lt.s64 %p7, %rd109, %rd71; + setp.lt.s64 %p6, %rd108, %rd71; + setp.lt.s64 %p5, %rd107, %rd71; + setp.lt.s64 %p4, %rd106, %rd71; + setp.lt.s64 %p3, %rd105, %rd71; + setp.lt.s64 %p2, %rd104, %rd71; + .loc 1 74 30 // l2norm.py:74:30 + and.b32 %r337, %r302, 15; + shl.b32 %r338, %r337, 3; + .loc 1 75 14 // l2norm.py:75:14 + shl.b64 %rd136, %rd104, 8; + add.s64 %rd137, %rd67, %rd136; + shl.b64 %rd138, %rd105, 8; + add.s64 %rd139, %rd67, %rd138; + shl.b64 %rd140, %rd106, 8; + add.s64 %rd141, %rd67, %rd140; + shl.b64 %rd142, %rd107, 8; + add.s64 %rd143, %rd67, %rd142; + shl.b64 %rd144, %rd108, 8; + add.s64 %rd145, %rd67, %rd144; + shl.b64 %rd146, %rd109, 8; + add.s64 %rd147, %rd67, %rd146; + shl.b64 %rd148, %rd110, 8; + add.s64 %rd149, %rd67, %rd148; + shl.b64 %rd150, %rd111, 8; + add.s64 %rd151, %rd67, %rd150; + shl.b64 %rd152, %rd112, 8; + add.s64 %rd153, %rd67, %rd152; + shl.b64 %rd154, %rd113, 8; + add.s64 %rd155, %rd67, %rd154; + shl.b64 %rd156, %rd114, 8; + add.s64 %rd157, %rd67, %rd156; + shl.b64 %rd158, %rd115, 8; + add.s64 %rd159, %rd67, %rd158; + shl.b64 %rd160, %rd116, 8; + add.s64 %rd161, %rd67, %rd160; + shl.b64 %rd162, %rd117, 8; + add.s64 %rd163, %rd67, %rd162; + shl.b64 %rd164, %rd118, 8; + add.s64 %rd165, %rd67, %rd164; + shl.b64 %rd166, %rd119, 8; + add.s64 %rd167, %rd67, %rd166; + shl.b64 %rd168, %rd134, 8; + add.s64 %rd169, %rd67, %rd168; + shl.b64 %rd170, %rd135, 8; + add.s64 %rd171, %rd67, %rd170; + shl.b64 %rd172, %rd132, 8; + add.s64 %rd173, %rd67, %rd172; + shl.b64 %rd174, %rd133, 8; + add.s64 %rd175, %rd67, %rd174; + shl.b64 %rd176, %rd130, 8; + add.s64 %rd177, %rd67, %rd176; + shl.b64 %rd178, %rd131, 8; + add.s64 %rd179, %rd67, %rd178; + shl.b64 %rd180, %rd128, 8; + add.s64 %rd181, %rd67, %rd180; + shl.b64 %rd182, %rd129, 8; + add.s64 %rd183, %rd67, %rd182; + shl.b64 %rd184, %rd126, 8; + add.s64 %rd185, %rd67, %rd184; + shl.b64 %rd186, %rd127, 8; + add.s64 %rd187, %rd67, %rd186; + shl.b64 %rd188, %rd124, 8; + add.s64 %rd189, %rd67, %rd188; + shl.b64 %rd190, %rd125, 8; + add.s64 %rd191, %rd67, %rd190; + shl.b64 %rd192, %rd122, 8; + add.s64 %rd193, %rd67, %rd192; + shl.b64 %rd194, %rd123, 8; + add.s64 %rd195, %rd67, %rd194; + shl.b64 %rd196, %rd120, 8; + add.s64 %rd197, %rd67, %rd196; + shl.b64 %rd198, %rd121, 8; + add.s64 %rd199, %rd67, %rd198; + .loc 1 75 33 // l2norm.py:75:33 + mul.wide.u32 %rd200, %r338, 2; + add.s64 %rd1, %rd137, %rd200; + add.s64 %rd2, %rd139, %rd200; + add.s64 %rd3, %rd141, %rd200; + add.s64 %rd4, %rd143, %rd200; + add.s64 %rd5, %rd145, %rd200; + add.s64 %rd6, %rd147, %rd200; + add.s64 %rd7, %rd149, %rd200; + add.s64 %rd8, %rd151, %rd200; + add.s64 %rd9, %rd153, %rd200; + add.s64 %rd10, %rd155, %rd200; + add.s64 %rd11, %rd157, %rd200; + add.s64 %rd12, %rd159, %rd200; + add.s64 %rd13, %rd161, %rd200; + add.s64 %rd14, %rd163, %rd200; + add.s64 %rd15, %rd165, %rd200; + add.s64 %rd16, %rd167, %rd200; + add.s64 %rd17, %rd169, %rd200; + add.s64 %rd18, %rd171, %rd200; + add.s64 %rd19, %rd173, %rd200; + add.s64 %rd20, %rd175, %rd200; + add.s64 %rd21, %rd177, %rd200; + add.s64 %rd22, %rd179, %rd200; + add.s64 %rd23, %rd181, %rd200; + add.s64 %rd24, %rd183, %rd200; + add.s64 %rd25, %rd185, %rd200; + add.s64 %rd26, %rd187, %rd200; + add.s64 %rd27, %rd189, %rd200; + add.s64 %rd28, %rd191, %rd200; + add.s64 %rd29, %rd193, %rd200; + add.s64 %rd30, %rd195, %rd200; + add.s64 %rd31, %rd197, %rd200; + add.s64 %rd32, %rd199, %rd200; + mov.b32 %r5, 0; + .loc 1 79 18 // l2norm.py:79:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p2 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs54, %rs55}, %r1; + mov.b32 {%rs56, %rs57}, %r2; + mov.b32 {%rs58, %rs59}, %r3; + mov.b32 {%rs60, %rs61}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p3 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs62, %rs63}, %r6; + mov.b32 {%rs64, %rs65}, %r7; + mov.b32 {%rs66, %rs67}, %r8; + mov.b32 {%rs68, %rs69}, %r9; + // begin inline asm + mov.u32 %r10, %r5; + mov.u32 %r11, %r5; + mov.u32 %r12, %r5; + mov.u32 %r13, %r5; + @%p4 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ]; + // end inline asm + mov.b32 {%rs70, %rs71}, %r10; + mov.b32 {%rs72, %rs73}, %r11; + mov.b32 {%rs74, %rs75}, %r12; + mov.b32 {%rs76, %rs77}, %r13; + // begin inline asm + mov.u32 %r14, %r5; + mov.u32 %r15, %r5; + mov.u32 %r16, %r5; + mov.u32 %r17, %r5; + @%p5 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ]; + // end inline asm + mov.b32 {%rs78, %rs79}, %r14; + mov.b32 {%rs80, %rs81}, %r15; + mov.b32 {%rs82, %rs83}, %r16; + mov.b32 {%rs84, %rs85}, %r17; + // begin inline asm + mov.u32 %r18, %r5; + mov.u32 %r19, %r5; + mov.u32 %r20, %r5; + mov.u32 %r21, %r5; + @%p6 ld.global.v4.b32 { %r18, %r19, %r20, %r21 }, [ %rd5 + 0 ]; + // end inline asm + mov.b32 {%rs86, %rs87}, %r18; + mov.b32 {%rs88, %rs89}, %r19; + mov.b32 {%rs90, %rs91}, %r20; + mov.b32 {%rs92, %rs93}, %r21; + // begin inline asm + mov.u32 %r22, %r5; + mov.u32 %r23, %r5; + mov.u32 %r24, %r5; + mov.u32 %r25, %r5; + @%p7 ld.global.v4.b32 { %r22, %r23, %r24, %r25 }, [ %rd6 + 0 ]; + // end inline asm + mov.b32 {%rs94, %rs95}, %r22; + mov.b32 {%rs96, %rs97}, %r23; + mov.b32 {%rs98, %rs99}, %r24; + mov.b32 {%rs100, %rs101}, %r25; + // begin inline asm + mov.u32 %r26, %r5; + mov.u32 %r27, %r5; + mov.u32 %r28, %r5; + mov.u32 %r29, %r5; + @%p8 ld.global.v4.b32 { %r26, %r27, %r28, %r29 }, [ %rd7 + 0 ]; + // end inline asm + mov.b32 {%rs102, %rs103}, %r26; + mov.b32 {%rs104, %rs105}, %r27; + mov.b32 {%rs106, %rs107}, %r28; + mov.b32 {%rs108, %rs109}, %r29; + // begin inline asm + mov.u32 %r30, %r5; + mov.u32 %r31, %r5; + mov.u32 %r32, %r5; + mov.u32 %r33, %r5; + @%p1 ld.global.v4.b32 { %r30, %r31, %r32, %r33 }, [ %rd8 + 0 ]; + // end inline asm + mov.b32 {%rs110, %rs111}, %r30; + mov.b32 {%rs112, %rs113}, %r31; + mov.b32 {%rs114, %rs115}, %r32; + mov.b32 {%rs116, %rs117}, %r33; + // begin inline asm + mov.u32 %r34, %r5; + mov.u32 %r35, %r5; + mov.u32 %r36, %r5; + mov.u32 %r37, %r5; + @%p9 ld.global.v4.b32 { %r34, %r35, %r36, %r37 }, [ %rd9 + 0 ]; + // end inline asm + mov.b32 {%rs118, %rs119}, %r34; + mov.b32 {%rs120, %rs121}, %r35; + mov.b32 {%rs122, %rs123}, %r36; + mov.b32 {%rs124, %rs125}, %r37; + // begin inline asm + mov.u32 %r38, %r5; + mov.u32 %r39, %r5; + mov.u32 %r40, %r5; + mov.u32 %r41, %r5; + @%p10 ld.global.v4.b32 { %r38, %r39, %r40, %r41 }, [ %rd10 + 0 ]; + // end inline asm + mov.b32 {%rs126, %rs127}, %r38; + mov.b32 {%rs128, %rs129}, %r39; + mov.b32 {%rs130, %rs131}, %r40; + mov.b32 {%rs132, %rs133}, %r41; + // begin inline asm + mov.u32 %r42, %r5; + mov.u32 %r43, %r5; + mov.u32 %r44, %r5; + mov.u32 %r45, %r5; + @%p11 ld.global.v4.b32 { %r42, %r43, %r44, %r45 }, [ %rd11 + 0 ]; + // end inline asm + mov.b32 {%rs134, %rs135}, %r42; + mov.b32 {%rs136, %rs137}, %r43; + mov.b32 {%rs138, %rs139}, %r44; + mov.b32 {%rs140, %rs141}, %r45; + // begin inline asm + mov.u32 %r46, %r5; + mov.u32 %r47, %r5; + mov.u32 %r48, %r5; + mov.u32 %r49, %r5; + @%p12 ld.global.v4.b32 { %r46, %r47, %r48, %r49 }, [ %rd12 + 0 ]; + // end inline asm + mov.b32 {%rs142, %rs143}, %r46; + mov.b32 {%rs144, %rs145}, %r47; + mov.b32 {%rs146, %rs147}, %r48; + mov.b32 {%rs148, %rs149}, %r49; + // begin inline asm + mov.u32 %r50, %r5; + mov.u32 %r51, %r5; + mov.u32 %r52, %r5; + mov.u32 %r53, %r5; + @%p13 ld.global.v4.b32 { %r50, %r51, %r52, %r53 }, [ %rd13 + 0 ]; + // end inline asm + mov.b32 {%rs150, %rs151}, %r50; + mov.b32 {%rs152, %rs153}, %r51; + mov.b32 {%rs154, %rs155}, %r52; + mov.b32 {%rs156, %rs157}, %r53; + // begin inline asm + mov.u32 %r54, %r5; + mov.u32 %r55, %r5; + mov.u32 %r56, %r5; + mov.u32 %r57, %r5; + @%p14 ld.global.v4.b32 { %r54, %r55, %r56, %r57 }, [ %rd14 + 0 ]; + // end inline asm + mov.b32 {%rs158, %rs159}, %r54; + mov.b32 {%rs160, %rs161}, %r55; + mov.b32 {%rs162, %rs163}, %r56; + mov.b32 {%rs164, %rs165}, %r57; + // begin inline asm + mov.u32 %r58, %r5; + mov.u32 %r59, %r5; + mov.u32 %r60, %r5; + mov.u32 %r61, %r5; + @%p15 ld.global.v4.b32 { %r58, %r59, %r60, %r61 }, [ %rd15 + 0 ]; + // end inline asm + mov.b32 {%rs166, %rs167}, %r58; + mov.b32 {%rs168, %rs169}, %r59; + mov.b32 {%rs170, %rs171}, %r60; + mov.b32 {%rs172, %rs173}, %r61; + // begin inline asm + mov.u32 %r62, %r5; + mov.u32 %r63, %r5; + mov.u32 %r64, %r5; + mov.u32 %r65, %r5; + @%p16 ld.global.v4.b32 { %r62, %r63, %r64, %r65 }, [ %rd16 + 0 ]; + // end inline asm + mov.b32 {%rs174, %rs175}, %r62; + mov.b32 {%rs176, %rs177}, %r63; + mov.b32 {%rs178, %rs179}, %r64; + mov.b32 {%rs180, %rs181}, %r65; + shr.u32 %r339, %r336, 15; + and.b32 %r340, %r339, 1; + setp.ne.b32 %p17, %r340, 0; + // begin inline asm + mov.u32 %r66, %r5; + mov.u32 %r67, %r5; + mov.u32 %r68, %r5; + mov.u32 %r69, %r5; + @%p17 ld.global.v4.b32 { %r66, %r67, %r68, %r69 }, [ %rd17 + 0 ]; + // end inline asm + mov.b32 {%rs182, %rs183}, %r66; + mov.b32 {%rs184, %rs185}, %r67; + mov.b32 {%rs186, %rs187}, %r68; + mov.b32 {%rs188, %rs189}, %r69; + shr.u32 %r341, %r336, 14; + and.b32 %r342, %r341, 1; + setp.ne.b32 %p18, %r342, 0; + // begin inline asm + mov.u32 %r70, %r5; + mov.u32 %r71, %r5; + mov.u32 %r72, %r5; + mov.u32 %r73, %r5; + @%p18 ld.global.v4.b32 { %r70, %r71, %r72, %r73 }, [ %rd18 + 0 ]; + // end inline asm + mov.b32 {%rs190, %rs191}, %r70; + mov.b32 {%rs192, %rs193}, %r71; + mov.b32 {%rs194, %rs195}, %r72; + mov.b32 {%rs196, %rs197}, %r73; + shr.u32 %r343, %r336, 13; + and.b32 %r344, %r343, 1; + setp.ne.b32 %p19, %r344, 0; + // begin inline asm + mov.u32 %r74, %r5; + mov.u32 %r75, %r5; + mov.u32 %r76, %r5; + mov.u32 %r77, %r5; + @%p19 ld.global.v4.b32 { %r74, %r75, %r76, %r77 }, [ %rd19 + 0 ]; + // end inline asm + mov.b32 {%rs198, %rs199}, %r74; + mov.b32 {%rs200, %rs201}, %r75; + mov.b32 {%rs202, %rs203}, %r76; + mov.b32 {%rs204, %rs205}, %r77; + shr.u32 %r345, %r336, 12; + and.b32 %r346, %r345, 1; + setp.ne.b32 %p20, %r346, 0; + // begin inline asm + mov.u32 %r78, %r5; + mov.u32 %r79, %r5; + mov.u32 %r80, %r5; + mov.u32 %r81, %r5; + @%p20 ld.global.v4.b32 { %r78, %r79, %r80, %r81 }, [ %rd20 + 0 ]; + // end inline asm + mov.b32 {%rs206, %rs207}, %r78; + mov.b32 {%rs208, %rs209}, %r79; + mov.b32 {%rs210, %rs211}, %r80; + mov.b32 {%rs212, %rs213}, %r81; + shr.u32 %r347, %r336, 11; + and.b32 %r348, %r347, 1; + setp.ne.b32 %p21, %r348, 0; + // begin inline asm + mov.u32 %r82, %r5; + mov.u32 %r83, %r5; + mov.u32 %r84, %r5; + mov.u32 %r85, %r5; + @%p21 ld.global.v4.b32 { %r82, %r83, %r84, %r85 }, [ %rd21 + 0 ]; + // end inline asm + mov.b32 {%rs214, %rs215}, %r82; + mov.b32 {%rs216, %rs217}, %r83; + mov.b32 {%rs218, %rs219}, %r84; + mov.b32 {%rs220, %rs221}, %r85; + shr.u32 %r349, %r336, 10; + and.b32 %r350, %r349, 1; + setp.ne.b32 %p22, %r350, 0; + // begin inline asm + mov.u32 %r86, %r5; + mov.u32 %r87, %r5; + mov.u32 %r88, %r5; + mov.u32 %r89, %r5; + @%p22 ld.global.v4.b32 { %r86, %r87, %r88, %r89 }, [ %rd22 + 0 ]; + // end inline asm + mov.b32 {%rs222, %rs223}, %r86; + mov.b32 {%rs224, %rs225}, %r87; + mov.b32 {%rs226, %rs227}, %r88; + mov.b32 {%rs228, %rs229}, %r89; + shr.u32 %r351, %r336, 9; + and.b32 %r352, %r351, 1; + setp.ne.b32 %p23, %r352, 0; + // begin inline asm + mov.u32 %r90, %r5; + mov.u32 %r91, %r5; + mov.u32 %r92, %r5; + mov.u32 %r93, %r5; + @%p23 ld.global.v4.b32 { %r90, %r91, %r92, %r93 }, [ %rd23 + 0 ]; + // end inline asm + mov.b32 {%rs230, %rs231}, %r90; + mov.b32 {%rs232, %rs233}, %r91; + mov.b32 {%rs234, %rs235}, %r92; + mov.b32 {%rs236, %rs237}, %r93; + shr.u32 %r353, %r336, 8; + and.b32 %r354, %r353, 1; + setp.ne.b32 %p24, %r354, 0; + // begin inline asm + mov.u32 %r94, %r5; + mov.u32 %r95, %r5; + mov.u32 %r96, %r5; + mov.u32 %r97, %r5; + @%p24 ld.global.v4.b32 { %r94, %r95, %r96, %r97 }, [ %rd24 + 0 ]; + // end inline asm + mov.b32 {%rs238, %rs239}, %r94; + mov.b32 {%rs240, %rs241}, %r95; + mov.b32 {%rs242, %rs243}, %r96; + mov.b32 {%rs244, %rs245}, %r97; + shr.u32 %r355, %r336, 7; + and.b32 %r356, %r355, 1; + setp.ne.b32 %p25, %r356, 0; + // begin inline asm + mov.u32 %r98, %r5; + mov.u32 %r99, %r5; + mov.u32 %r100, %r5; + mov.u32 %r101, %r5; + @%p25 ld.global.v4.b32 { %r98, %r99, %r100, %r101 }, [ %rd25 + 0 ]; + // end inline asm + mov.b32 {%rs246, %rs247}, %r98; + mov.b32 {%rs248, %rs249}, %r99; + mov.b32 {%rs250, %rs251}, %r100; + mov.b32 {%rs252, %rs253}, %r101; + shr.u32 %r357, %r336, 6; + and.b32 %r358, %r357, 1; + setp.ne.b32 %p26, %r358, 0; + // begin inline asm + mov.u32 %r102, %r5; + mov.u32 %r103, %r5; + mov.u32 %r104, %r5; + mov.u32 %r105, %r5; + @%p26 ld.global.v4.b32 { %r102, %r103, %r104, %r105 }, [ %rd26 + 0 ]; + // end inline asm + mov.b32 {%rs254, %rs255}, %r102; + mov.b32 {%rs256, %rs257}, %r103; + mov.b32 {%rs258, %rs259}, %r104; + mov.b32 {%rs260, %rs261}, %r105; + shr.u32 %r359, %r336, 5; + and.b32 %r360, %r359, 1; + setp.ne.b32 %p27, %r360, 0; + // begin inline asm + mov.u32 %r106, %r5; + mov.u32 %r107, %r5; + mov.u32 %r108, %r5; + mov.u32 %r109, %r5; + @%p27 ld.global.v4.b32 { %r106, %r107, %r108, %r109 }, [ %rd27 + 0 ]; + // end inline asm + mov.b32 {%rs262, %rs263}, %r106; + mov.b32 {%rs264, %rs265}, %r107; + mov.b32 {%rs266, %rs267}, %r108; + mov.b32 {%rs268, %rs269}, %r109; + shr.u32 %r361, %r336, 4; + and.b32 %r362, %r361, 1; + setp.ne.b32 %p28, %r362, 0; + // begin inline asm + mov.u32 %r110, %r5; + mov.u32 %r111, %r5; + mov.u32 %r112, %r5; + mov.u32 %r113, %r5; + @%p28 ld.global.v4.b32 { %r110, %r111, %r112, %r113 }, [ %rd28 + 0 ]; + // end inline asm + mov.b32 {%rs270, %rs271}, %r110; + mov.b32 {%rs272, %rs273}, %r111; + mov.b32 {%rs274, %rs275}, %r112; + mov.b32 {%rs276, %rs277}, %r113; + shr.u32 %r363, %r336, 3; + and.b32 %r364, %r363, 1; + setp.ne.b32 %p29, %r364, 0; + // begin inline asm + mov.u32 %r114, %r5; + mov.u32 %r115, %r5; + mov.u32 %r116, %r5; + mov.u32 %r117, %r5; + @%p29 ld.global.v4.b32 { %r114, %r115, %r116, %r117 }, [ %rd29 + 0 ]; + // end inline asm + mov.b32 {%rs278, %rs279}, %r114; + mov.b32 {%rs280, %rs281}, %r115; + mov.b32 {%rs282, %rs283}, %r116; + mov.b32 {%rs284, %rs285}, %r117; + shr.u32 %r365, %r336, 2; + and.b32 %r366, %r365, 1; + setp.ne.b32 %p30, %r366, 0; + // begin inline asm + mov.u32 %r118, %r5; + mov.u32 %r119, %r5; + mov.u32 %r120, %r5; + mov.u32 %r121, %r5; + @%p30 ld.global.v4.b32 { %r118, %r119, %r120, %r121 }, [ %rd30 + 0 ]; + // end inline asm + mov.b32 {%rs286, %rs287}, %r118; + mov.b32 {%rs288, %rs289}, %r119; + mov.b32 {%rs290, %rs291}, %r120; + mov.b32 {%rs292, %rs293}, %r121; + shr.u32 %r367, %r336, 1; + and.b32 %r368, %r367, 1; + setp.ne.b32 %p31, %r368, 0; + // begin inline asm + mov.u32 %r122, %r5; + mov.u32 %r123, %r5; + mov.u32 %r124, %r5; + mov.u32 %r125, %r5; + @%p31 ld.global.v4.b32 { %r122, %r123, %r124, %r125 }, [ %rd31 + 0 ]; + // end inline asm + mov.b32 {%rs294, %rs295}, %r122; + mov.b32 {%rs296, %rs297}, %r123; + mov.b32 {%rs298, %rs299}, %r124; + mov.b32 {%rs300, %rs301}, %r125; + // begin inline asm + mov.u32 %r126, %r5; + mov.u32 %r127, %r5; + mov.u32 %r128, %r5; + mov.u32 %r129, %r5; + @%p32 ld.global.v4.b32 { %r126, %r127, %r128, %r129 }, [ %rd32 + 0 ]; + // end inline asm + mov.b32 {%rs302, %rs303}, %r126; + mov.b32 {%rs304, %rs305}, %r127; + mov.b32 {%rs306, %rs307}, %r128; + mov.b32 {%rs308, %rs309}, %r129; + .loc 1 79 47 // l2norm.py:79:47 + cvt.f32.bf16 %r369, %rs54; + cvt.f32.bf16 %r370, %rs55; + cvt.f32.bf16 %r371, %rs56; + cvt.f32.bf16 %r372, %rs57; + cvt.f32.bf16 %r373, %rs58; + cvt.f32.bf16 %r374, %rs59; + cvt.f32.bf16 %r375, %rs60; + cvt.f32.bf16 %r376, %rs61; + cvt.f32.bf16 %r377, %rs62; + cvt.f32.bf16 %r378, %rs63; + cvt.f32.bf16 %r379, %rs64; + cvt.f32.bf16 %r380, %rs65; + cvt.f32.bf16 %r381, %rs66; + cvt.f32.bf16 %r382, %rs67; + cvt.f32.bf16 %r383, %rs68; + cvt.f32.bf16 %r384, %rs69; + cvt.f32.bf16 %r385, %rs70; + cvt.f32.bf16 %r386, %rs71; + cvt.f32.bf16 %r387, %rs72; + cvt.f32.bf16 %r388, %rs73; + cvt.f32.bf16 %r389, %rs74; + cvt.f32.bf16 %r390, %rs75; + cvt.f32.bf16 %r391, %rs76; + cvt.f32.bf16 %r392, %rs77; + cvt.f32.bf16 %r393, %rs78; + cvt.f32.bf16 %r394, %rs79; + cvt.f32.bf16 %r395, %rs80; + cvt.f32.bf16 %r396, %rs81; + cvt.f32.bf16 %r397, %rs82; + cvt.f32.bf16 %r398, %rs83; + cvt.f32.bf16 %r399, %rs84; + cvt.f32.bf16 %r400, %rs85; + cvt.f32.bf16 %r401, %rs86; + cvt.f32.bf16 %r402, %rs87; + cvt.f32.bf16 %r403, %rs88; + cvt.f32.bf16 %r404, %rs89; + cvt.f32.bf16 %r405, %rs90; + cvt.f32.bf16 %r406, %rs91; + cvt.f32.bf16 %r407, %rs92; + cvt.f32.bf16 %r408, %rs93; + cvt.f32.bf16 %r409, %rs94; + cvt.f32.bf16 %r410, %rs95; + cvt.f32.bf16 %r411, %rs96; + cvt.f32.bf16 %r412, %rs97; + cvt.f32.bf16 %r413, %rs98; + cvt.f32.bf16 %r414, %rs99; + cvt.f32.bf16 %r415, %rs100; + cvt.f32.bf16 %r416, %rs101; + cvt.f32.bf16 %r417, %rs102; + cvt.f32.bf16 %r418, %rs103; + cvt.f32.bf16 %r419, %rs104; + cvt.f32.bf16 %r420, %rs105; + cvt.f32.bf16 %r421, %rs106; + cvt.f32.bf16 %r422, %rs107; + cvt.f32.bf16 %r423, %rs108; + cvt.f32.bf16 %r424, %rs109; + cvt.f32.bf16 %r425, %rs110; + cvt.f32.bf16 %r426, %rs111; + cvt.f32.bf16 %r427, %rs112; + cvt.f32.bf16 %r428, %rs113; + cvt.f32.bf16 %r429, %rs114; + cvt.f32.bf16 %r430, %rs115; + cvt.f32.bf16 %r431, %rs116; + cvt.f32.bf16 %r432, %rs117; + cvt.f32.bf16 %r433, %rs118; + cvt.f32.bf16 %r434, %rs119; + cvt.f32.bf16 %r435, %rs120; + cvt.f32.bf16 %r436, %rs121; + cvt.f32.bf16 %r437, %rs122; + cvt.f32.bf16 %r438, %rs123; + cvt.f32.bf16 %r439, %rs124; + cvt.f32.bf16 %r440, %rs125; + cvt.f32.bf16 %r441, %rs126; + cvt.f32.bf16 %r442, %rs127; + cvt.f32.bf16 %r443, %rs128; + cvt.f32.bf16 %r444, %rs129; + cvt.f32.bf16 %r445, %rs130; + cvt.f32.bf16 %r446, %rs131; + cvt.f32.bf16 %r447, %rs132; + cvt.f32.bf16 %r448, %rs133; + cvt.f32.bf16 %r449, %rs134; + cvt.f32.bf16 %r450, %rs135; + cvt.f32.bf16 %r451, %rs136; + cvt.f32.bf16 %r452, %rs137; + cvt.f32.bf16 %r453, %rs138; + cvt.f32.bf16 %r454, %rs139; + cvt.f32.bf16 %r455, %rs140; + cvt.f32.bf16 %r456, %rs141; + cvt.f32.bf16 %r457, %rs142; + cvt.f32.bf16 %r458, %rs143; + cvt.f32.bf16 %r459, %rs144; + cvt.f32.bf16 %r460, %rs145; + cvt.f32.bf16 %r461, %rs146; + cvt.f32.bf16 %r462, %rs147; + cvt.f32.bf16 %r463, %rs148; + cvt.f32.bf16 %r464, %rs149; + cvt.f32.bf16 %r465, %rs150; + cvt.f32.bf16 %r466, %rs151; + cvt.f32.bf16 %r467, %rs152; + cvt.f32.bf16 %r468, %rs153; + cvt.f32.bf16 %r469, %rs154; + cvt.f32.bf16 %r470, %rs155; + cvt.f32.bf16 %r471, %rs156; + cvt.f32.bf16 %r472, %rs157; + cvt.f32.bf16 %r473, %rs158; + cvt.f32.bf16 %r474, %rs159; + cvt.f32.bf16 %r475, %rs160; + cvt.f32.bf16 %r476, %rs161; + cvt.f32.bf16 %r477, %rs162; + cvt.f32.bf16 %r478, %rs163; + cvt.f32.bf16 %r479, %rs164; + cvt.f32.bf16 %r480, %rs165; + cvt.f32.bf16 %r481, %rs166; + cvt.f32.bf16 %r482, %rs167; + cvt.f32.bf16 %r483, %rs168; + cvt.f32.bf16 %r484, %rs169; + cvt.f32.bf16 %r485, %rs170; + cvt.f32.bf16 %r486, %rs171; + cvt.f32.bf16 %r487, %rs172; + cvt.f32.bf16 %r488, %rs173; + cvt.f32.bf16 %r489, %rs174; + cvt.f32.bf16 %r490, %rs175; + cvt.f32.bf16 %r491, %rs176; + cvt.f32.bf16 %r492, %rs177; + cvt.f32.bf16 %r493, %rs178; + cvt.f32.bf16 %r494, %rs179; + cvt.f32.bf16 %r495, %rs180; + cvt.f32.bf16 %r496, %rs181; + cvt.f32.bf16 %r497, %rs182; + cvt.f32.bf16 %r498, %rs183; + cvt.f32.bf16 %r499, %rs184; + cvt.f32.bf16 %r500, %rs185; + cvt.f32.bf16 %r501, %rs186; + cvt.f32.bf16 %r502, %rs187; + cvt.f32.bf16 %r503, %rs188; + cvt.f32.bf16 %r504, %rs189; + cvt.f32.bf16 %r505, %rs190; + cvt.f32.bf16 %r506, %rs191; + cvt.f32.bf16 %r507, %rs192; + cvt.f32.bf16 %r508, %rs193; + cvt.f32.bf16 %r509, %rs194; + cvt.f32.bf16 %r510, %rs195; + cvt.f32.bf16 %r511, %rs196; + cvt.f32.bf16 %r512, %rs197; + cvt.f32.bf16 %r513, %rs198; + cvt.f32.bf16 %r514, %rs199; + cvt.f32.bf16 %r515, %rs200; + cvt.f32.bf16 %r516, %rs201; + cvt.f32.bf16 %r517, %rs202; + cvt.f32.bf16 %r518, %rs203; + cvt.f32.bf16 %r519, %rs204; + cvt.f32.bf16 %r520, %rs205; + cvt.f32.bf16 %r521, %rs206; + cvt.f32.bf16 %r522, %rs207; + cvt.f32.bf16 %r523, %rs208; + cvt.f32.bf16 %r524, %rs209; + cvt.f32.bf16 %r525, %rs210; + cvt.f32.bf16 %r526, %rs211; + cvt.f32.bf16 %r527, %rs212; + cvt.f32.bf16 %r528, %rs213; + cvt.f32.bf16 %r529, %rs214; + cvt.f32.bf16 %r530, %rs215; + cvt.f32.bf16 %r531, %rs216; + cvt.f32.bf16 %r532, %rs217; + cvt.f32.bf16 %r533, %rs218; + cvt.f32.bf16 %r534, %rs219; + cvt.f32.bf16 %r535, %rs220; + cvt.f32.bf16 %r536, %rs221; + cvt.f32.bf16 %r537, %rs222; + cvt.f32.bf16 %r538, %rs223; + cvt.f32.bf16 %r539, %rs224; + cvt.f32.bf16 %r540, %rs225; + cvt.f32.bf16 %r541, %rs226; + cvt.f32.bf16 %r542, %rs227; + cvt.f32.bf16 %r543, %rs228; + cvt.f32.bf16 %r544, %rs229; + cvt.f32.bf16 %r545, %rs230; + cvt.f32.bf16 %r546, %rs231; + cvt.f32.bf16 %r547, %rs232; + cvt.f32.bf16 %r548, %rs233; + cvt.f32.bf16 %r549, %rs234; + cvt.f32.bf16 %r550, %rs235; + cvt.f32.bf16 %r551, %rs236; + cvt.f32.bf16 %r552, %rs237; + cvt.f32.bf16 %r553, %rs238; + cvt.f32.bf16 %r554, %rs239; + cvt.f32.bf16 %r555, %rs240; + cvt.f32.bf16 %r556, %rs241; + cvt.f32.bf16 %r557, %rs242; + cvt.f32.bf16 %r558, %rs243; + cvt.f32.bf16 %r559, %rs244; + cvt.f32.bf16 %r560, %rs245; + cvt.f32.bf16 %r561, %rs246; + cvt.f32.bf16 %r562, %rs247; + cvt.f32.bf16 %r563, %rs248; + cvt.f32.bf16 %r564, %rs249; + cvt.f32.bf16 %r565, %rs250; + cvt.f32.bf16 %r566, %rs251; + cvt.f32.bf16 %r567, %rs252; + cvt.f32.bf16 %r568, %rs253; + cvt.f32.bf16 %r569, %rs254; + cvt.f32.bf16 %r570, %rs255; + cvt.f32.bf16 %r571, %rs256; + cvt.f32.bf16 %r572, %rs257; + cvt.f32.bf16 %r573, %rs258; + cvt.f32.bf16 %r574, %rs259; + cvt.f32.bf16 %r575, %rs260; + cvt.f32.bf16 %r576, %rs261; + cvt.f32.bf16 %r577, %rs262; + cvt.f32.bf16 %r578, %rs263; + cvt.f32.bf16 %r579, %rs264; + cvt.f32.bf16 %r580, %rs265; + cvt.f32.bf16 %r581, %rs266; + cvt.f32.bf16 %r582, %rs267; + cvt.f32.bf16 %r583, %rs268; + cvt.f32.bf16 %r584, %rs269; + cvt.f32.bf16 %r585, %rs270; + cvt.f32.bf16 %r586, %rs271; + cvt.f32.bf16 %r587, %rs272; + cvt.f32.bf16 %r588, %rs273; + cvt.f32.bf16 %r589, %rs274; + cvt.f32.bf16 %r590, %rs275; + cvt.f32.bf16 %r591, %rs276; + cvt.f32.bf16 %r592, %rs277; + cvt.f32.bf16 %r593, %rs278; + cvt.f32.bf16 %r594, %rs279; + cvt.f32.bf16 %r595, %rs280; + cvt.f32.bf16 %r596, %rs281; + cvt.f32.bf16 %r597, %rs282; + cvt.f32.bf16 %r598, %rs283; + cvt.f32.bf16 %r599, %rs284; + cvt.f32.bf16 %r600, %rs285; + cvt.f32.bf16 %r601, %rs286; + cvt.f32.bf16 %r602, %rs287; + cvt.f32.bf16 %r603, %rs288; + cvt.f32.bf16 %r604, %rs289; + cvt.f32.bf16 %r605, %rs290; + cvt.f32.bf16 %r606, %rs291; + cvt.f32.bf16 %r607, %rs292; + cvt.f32.bf16 %r608, %rs293; + cvt.f32.bf16 %r609, %rs294; + cvt.f32.bf16 %r610, %rs295; + cvt.f32.bf16 %r611, %rs296; + cvt.f32.bf16 %r612, %rs297; + cvt.f32.bf16 %r613, %rs298; + cvt.f32.bf16 %r614, %rs299; + cvt.f32.bf16 %r615, %rs300; + cvt.f32.bf16 %r616, %rs301; + cvt.f32.bf16 %r617, %rs302; + cvt.f32.bf16 %r618, %rs303; + cvt.f32.bf16 %r619, %rs304; + cvt.f32.bf16 %r620, %rs305; + cvt.f32.bf16 %r621, %rs306; + cvt.f32.bf16 %r622, %rs307; + cvt.f32.bf16 %r623, %rs308; + cvt.f32.bf16 %r624, %rs309; + .loc 1 80 38 // l2norm.py:80:38 + mul.f32 %r625, %r370, %r370; + mul.f32 %r626, %r378, %r378; + mul.f32 %r627, %r386, %r386; + mul.f32 %r628, %r394, %r394; + mul.f32 %r629, %r402, %r402; + mul.f32 %r630, %r410, %r410; + mul.f32 %r631, %r418, %r418; + mul.f32 %r632, %r426, %r426; + mul.f32 %r633, %r434, %r434; + mul.f32 %r634, %r442, %r442; + mul.f32 %r635, %r450, %r450; + mul.f32 %r636, %r458, %r458; + mul.f32 %r637, %r466, %r466; + mul.f32 %r638, %r474, %r474; + mul.f32 %r639, %r482, %r482; + mul.f32 %r640, %r490, %r490; + mul.f32 %r641, %r498, %r498; + mul.f32 %r642, %r506, %r506; + mul.f32 %r643, %r514, %r514; + mul.f32 %r644, %r522, %r522; + mul.f32 %r645, %r530, %r530; + mul.f32 %r646, %r538, %r538; + mul.f32 %r647, %r546, %r546; + mul.f32 %r648, %r554, %r554; + mul.f32 %r649, %r562, %r562; + mul.f32 %r650, %r570, %r570; + mul.f32 %r651, %r578, %r578; + mul.f32 %r652, %r586, %r586; + mul.f32 %r653, %r594, %r594; + mul.f32 %r654, %r602, %r602; + mul.f32 %r655, %r610, %r610; + mul.f32 %r656, %r618, %r618; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + fma.rn.f32.bf16 %r657, %rs54, %rs54, %r625; + fma.rn.f32.bf16 %r658, %rs56, %rs56, %r657; + fma.rn.f32.bf16 %r659, %rs57, %rs57, %r658; + fma.rn.f32.bf16 %r660, %rs58, %rs58, %r659; + fma.rn.f32.bf16 %r661, %rs59, %rs59, %r660; + fma.rn.f32.bf16 %r662, %rs60, %rs60, %r661; + fma.rn.f32.bf16 %r663, %rs61, %rs61, %r662; + fma.rn.f32.bf16 %r664, %rs62, %rs62, %r626; + fma.rn.f32.bf16 %r665, %rs64, %rs64, %r664; + fma.rn.f32.bf16 %r666, %rs65, %rs65, %r665; + fma.rn.f32.bf16 %r667, %rs66, %rs66, %r666; + fma.rn.f32.bf16 %r668, %rs67, %rs67, %r667; + fma.rn.f32.bf16 %r669, %rs68, %rs68, %r668; + fma.rn.f32.bf16 %r670, %rs69, %rs69, %r669; + fma.rn.f32.bf16 %r671, %rs70, %rs70, %r627; + fma.rn.f32.bf16 %r672, %rs72, %rs72, %r671; + fma.rn.f32.bf16 %r673, %rs73, %rs73, %r672; + fma.rn.f32.bf16 %r674, %rs74, %rs74, %r673; + fma.rn.f32.bf16 %r675, %rs75, %rs75, %r674; + fma.rn.f32.bf16 %r676, %rs76, %rs76, %r675; + fma.rn.f32.bf16 %r677, %rs77, %rs77, %r676; + fma.rn.f32.bf16 %r678, %rs78, %rs78, %r628; + fma.rn.f32.bf16 %r679, %rs80, %rs80, %r678; + fma.rn.f32.bf16 %r680, %rs81, %rs81, %r679; + fma.rn.f32.bf16 %r681, %rs82, %rs82, %r680; + fma.rn.f32.bf16 %r682, %rs83, %rs83, %r681; + fma.rn.f32.bf16 %r683, %rs84, %rs84, %r682; + fma.rn.f32.bf16 %r684, %rs85, %rs85, %r683; + fma.rn.f32.bf16 %r685, %rs86, %rs86, %r629; + fma.rn.f32.bf16 %r686, %rs88, %rs88, %r685; + fma.rn.f32.bf16 %r687, %rs89, %rs89, %r686; + fma.rn.f32.bf16 %r688, %rs90, %rs90, %r687; + fma.rn.f32.bf16 %r689, %rs91, %rs91, %r688; + fma.rn.f32.bf16 %r690, %rs92, %rs92, %r689; + fma.rn.f32.bf16 %r691, %rs93, %rs93, %r690; + fma.rn.f32.bf16 %r692, %rs94, %rs94, %r630; + fma.rn.f32.bf16 %r693, %rs96, %rs96, %r692; + fma.rn.f32.bf16 %r694, %rs97, %rs97, %r693; + fma.rn.f32.bf16 %r695, %rs98, %rs98, %r694; + fma.rn.f32.bf16 %r696, %rs99, %rs99, %r695; + fma.rn.f32.bf16 %r697, %rs100, %rs100, %r696; + fma.rn.f32.bf16 %r698, %rs101, %rs101, %r697; + fma.rn.f32.bf16 %r699, %rs102, %rs102, %r631; + fma.rn.f32.bf16 %r700, %rs104, %rs104, %r699; + fma.rn.f32.bf16 %r701, %rs105, %rs105, %r700; + fma.rn.f32.bf16 %r702, %rs106, %rs106, %r701; + fma.rn.f32.bf16 %r703, %rs107, %rs107, %r702; + fma.rn.f32.bf16 %r704, %rs108, %rs108, %r703; + fma.rn.f32.bf16 %r705, %rs109, %rs109, %r704; + fma.rn.f32.bf16 %r706, %rs110, %rs110, %r632; + fma.rn.f32.bf16 %r707, %rs112, %rs112, %r706; + fma.rn.f32.bf16 %r708, %rs113, %rs113, %r707; + fma.rn.f32.bf16 %r709, %rs114, %rs114, %r708; + fma.rn.f32.bf16 %r710, %rs115, %rs115, %r709; + fma.rn.f32.bf16 %r711, %rs116, %rs116, %r710; + fma.rn.f32.bf16 %r712, %rs117, %rs117, %r711; + fma.rn.f32.bf16 %r713, %rs118, %rs118, %r633; + fma.rn.f32.bf16 %r714, %rs120, %rs120, %r713; + fma.rn.f32.bf16 %r715, %rs121, %rs121, %r714; + fma.rn.f32.bf16 %r716, %rs122, %rs122, %r715; + fma.rn.f32.bf16 %r717, %rs123, %rs123, %r716; + fma.rn.f32.bf16 %r718, %rs124, %rs124, %r717; + fma.rn.f32.bf16 %r719, %rs125, %rs125, %r718; + fma.rn.f32.bf16 %r720, %rs126, %rs126, %r634; + fma.rn.f32.bf16 %r721, %rs128, %rs128, %r720; + fma.rn.f32.bf16 %r722, %rs129, %rs129, %r721; + fma.rn.f32.bf16 %r723, %rs130, %rs130, %r722; + fma.rn.f32.bf16 %r724, %rs131, %rs131, %r723; + fma.rn.f32.bf16 %r725, %rs132, %rs132, %r724; + fma.rn.f32.bf16 %r726, %rs133, %rs133, %r725; + fma.rn.f32.bf16 %r727, %rs134, %rs134, %r635; + fma.rn.f32.bf16 %r728, %rs136, %rs136, %r727; + fma.rn.f32.bf16 %r729, %rs137, %rs137, %r728; + fma.rn.f32.bf16 %r730, %rs138, %rs138, %r729; + fma.rn.f32.bf16 %r731, %rs139, %rs139, %r730; + fma.rn.f32.bf16 %r732, %rs140, %rs140, %r731; + fma.rn.f32.bf16 %r733, %rs141, %rs141, %r732; + fma.rn.f32.bf16 %r734, %rs142, %rs142, %r636; + fma.rn.f32.bf16 %r735, %rs144, %rs144, %r734; + fma.rn.f32.bf16 %r736, %rs145, %rs145, %r735; + fma.rn.f32.bf16 %r737, %rs146, %rs146, %r736; + fma.rn.f32.bf16 %r738, %rs147, %rs147, %r737; + fma.rn.f32.bf16 %r739, %rs148, %rs148, %r738; + fma.rn.f32.bf16 %r740, %rs149, %rs149, %r739; + fma.rn.f32.bf16 %r741, %rs150, %rs150, %r637; + fma.rn.f32.bf16 %r742, %rs152, %rs152, %r741; + fma.rn.f32.bf16 %r743, %rs153, %rs153, %r742; + fma.rn.f32.bf16 %r744, %rs154, %rs154, %r743; + fma.rn.f32.bf16 %r745, %rs155, %rs155, %r744; + fma.rn.f32.bf16 %r746, %rs156, %rs156, %r745; + fma.rn.f32.bf16 %r747, %rs157, %rs157, %r746; + fma.rn.f32.bf16 %r748, %rs158, %rs158, %r638; + fma.rn.f32.bf16 %r749, %rs160, %rs160, %r748; + fma.rn.f32.bf16 %r750, %rs161, %rs161, %r749; + fma.rn.f32.bf16 %r751, %rs162, %rs162, %r750; + fma.rn.f32.bf16 %r752, %rs163, %rs163, %r751; + fma.rn.f32.bf16 %r753, %rs164, %rs164, %r752; + fma.rn.f32.bf16 %r754, %rs165, %rs165, %r753; + fma.rn.f32.bf16 %r755, %rs166, %rs166, %r639; + fma.rn.f32.bf16 %r756, %rs168, %rs168, %r755; + fma.rn.f32.bf16 %r757, %rs169, %rs169, %r756; + fma.rn.f32.bf16 %r758, %rs170, %rs170, %r757; + fma.rn.f32.bf16 %r759, %rs171, %rs171, %r758; + fma.rn.f32.bf16 %r760, %rs172, %rs172, %r759; + fma.rn.f32.bf16 %r761, %rs173, %rs173, %r760; + fma.rn.f32.bf16 %r762, %rs174, %rs174, %r640; + fma.rn.f32.bf16 %r763, %rs176, %rs176, %r762; + fma.rn.f32.bf16 %r764, %rs177, %rs177, %r763; + fma.rn.f32.bf16 %r765, %rs178, %rs178, %r764; + fma.rn.f32.bf16 %r766, %rs179, %rs179, %r765; + fma.rn.f32.bf16 %r767, %rs180, %rs180, %r766; + fma.rn.f32.bf16 %r768, %rs181, %rs181, %r767; + fma.rn.f32.bf16 %r769, %rs182, %rs182, %r641; + fma.rn.f32.bf16 %r770, %rs184, %rs184, %r769; + fma.rn.f32.bf16 %r771, %rs185, %rs185, %r770; + fma.rn.f32.bf16 %r772, %rs186, %rs186, %r771; + fma.rn.f32.bf16 %r773, %rs187, %rs187, %r772; + fma.rn.f32.bf16 %r774, %rs188, %rs188, %r773; + fma.rn.f32.bf16 %r775, %rs189, %rs189, %r774; + fma.rn.f32.bf16 %r776, %rs190, %rs190, %r642; + fma.rn.f32.bf16 %r777, %rs192, %rs192, %r776; + fma.rn.f32.bf16 %r778, %rs193, %rs193, %r777; + fma.rn.f32.bf16 %r779, %rs194, %rs194, %r778; + fma.rn.f32.bf16 %r780, %rs195, %rs195, %r779; + fma.rn.f32.bf16 %r781, %rs196, %rs196, %r780; + fma.rn.f32.bf16 %r782, %rs197, %rs197, %r781; + fma.rn.f32.bf16 %r783, %rs198, %rs198, %r643; + fma.rn.f32.bf16 %r784, %rs200, %rs200, %r783; + fma.rn.f32.bf16 %r785, %rs201, %rs201, %r784; + fma.rn.f32.bf16 %r786, %rs202, %rs202, %r785; + fma.rn.f32.bf16 %r787, %rs203, %rs203, %r786; + fma.rn.f32.bf16 %r788, %rs204, %rs204, %r787; + fma.rn.f32.bf16 %r789, %rs205, %rs205, %r788; + fma.rn.f32.bf16 %r790, %rs206, %rs206, %r644; + fma.rn.f32.bf16 %r791, %rs208, %rs208, %r790; + fma.rn.f32.bf16 %r792, %rs209, %rs209, %r791; + fma.rn.f32.bf16 %r793, %rs210, %rs210, %r792; + fma.rn.f32.bf16 %r794, %rs211, %rs211, %r793; + fma.rn.f32.bf16 %r795, %rs212, %rs212, %r794; + fma.rn.f32.bf16 %r796, %rs213, %rs213, %r795; + fma.rn.f32.bf16 %r797, %rs214, %rs214, %r645; + fma.rn.f32.bf16 %r798, %rs216, %rs216, %r797; + fma.rn.f32.bf16 %r799, %rs217, %rs217, %r798; + fma.rn.f32.bf16 %r800, %rs218, %rs218, %r799; + fma.rn.f32.bf16 %r801, %rs219, %rs219, %r800; + fma.rn.f32.bf16 %r802, %rs220, %rs220, %r801; + fma.rn.f32.bf16 %r803, %rs221, %rs221, %r802; + fma.rn.f32.bf16 %r804, %rs222, %rs222, %r646; + fma.rn.f32.bf16 %r805, %rs224, %rs224, %r804; + fma.rn.f32.bf16 %r806, %rs225, %rs225, %r805; + fma.rn.f32.bf16 %r807, %rs226, %rs226, %r806; + fma.rn.f32.bf16 %r808, %rs227, %rs227, %r807; + fma.rn.f32.bf16 %r809, %rs228, %rs228, %r808; + fma.rn.f32.bf16 %r810, %rs229, %rs229, %r809; + fma.rn.f32.bf16 %r811, %rs230, %rs230, %r647; + fma.rn.f32.bf16 %r812, %rs232, %rs232, %r811; + fma.rn.f32.bf16 %r813, %rs233, %rs233, %r812; + fma.rn.f32.bf16 %r814, %rs234, %rs234, %r813; + fma.rn.f32.bf16 %r815, %rs235, %rs235, %r814; + fma.rn.f32.bf16 %r816, %rs236, %rs236, %r815; + fma.rn.f32.bf16 %r817, %rs237, %rs237, %r816; + fma.rn.f32.bf16 %r818, %rs238, %rs238, %r648; + fma.rn.f32.bf16 %r819, %rs240, %rs240, %r818; + fma.rn.f32.bf16 %r820, %rs241, %rs241, %r819; + fma.rn.f32.bf16 %r821, %rs242, %rs242, %r820; + fma.rn.f32.bf16 %r822, %rs243, %rs243, %r821; + fma.rn.f32.bf16 %r823, %rs244, %rs244, %r822; + fma.rn.f32.bf16 %r824, %rs245, %rs245, %r823; + fma.rn.f32.bf16 %r825, %rs246, %rs246, %r649; + fma.rn.f32.bf16 %r826, %rs248, %rs248, %r825; + fma.rn.f32.bf16 %r827, %rs249, %rs249, %r826; + fma.rn.f32.bf16 %r828, %rs250, %rs250, %r827; + fma.rn.f32.bf16 %r829, %rs251, %rs251, %r828; + fma.rn.f32.bf16 %r830, %rs252, %rs252, %r829; + fma.rn.f32.bf16 %r831, %rs253, %rs253, %r830; + fma.rn.f32.bf16 %r832, %rs254, %rs254, %r650; + fma.rn.f32.bf16 %r833, %rs256, %rs256, %r832; + fma.rn.f32.bf16 %r834, %rs257, %rs257, %r833; + fma.rn.f32.bf16 %r835, %rs258, %rs258, %r834; + fma.rn.f32.bf16 %r836, %rs259, %rs259, %r835; + fma.rn.f32.bf16 %r837, %rs260, %rs260, %r836; + fma.rn.f32.bf16 %r838, %rs261, %rs261, %r837; + fma.rn.f32.bf16 %r839, %rs262, %rs262, %r651; + fma.rn.f32.bf16 %r840, %rs264, %rs264, %r839; + fma.rn.f32.bf16 %r841, %rs265, %rs265, %r840; + fma.rn.f32.bf16 %r842, %rs266, %rs266, %r841; + fma.rn.f32.bf16 %r843, %rs267, %rs267, %r842; + fma.rn.f32.bf16 %r844, %rs268, %rs268, %r843; + fma.rn.f32.bf16 %r845, %rs269, %rs269, %r844; + fma.rn.f32.bf16 %r846, %rs270, %rs270, %r652; + fma.rn.f32.bf16 %r847, %rs272, %rs272, %r846; + fma.rn.f32.bf16 %r848, %rs273, %rs273, %r847; + fma.rn.f32.bf16 %r849, %rs274, %rs274, %r848; + fma.rn.f32.bf16 %r850, %rs275, %rs275, %r849; + fma.rn.f32.bf16 %r851, %rs276, %rs276, %r850; + fma.rn.f32.bf16 %r852, %rs277, %rs277, %r851; + fma.rn.f32.bf16 %r853, %rs278, %rs278, %r653; + fma.rn.f32.bf16 %r854, %rs280, %rs280, %r853; + fma.rn.f32.bf16 %r855, %rs281, %rs281, %r854; + fma.rn.f32.bf16 %r856, %rs282, %rs282, %r855; + fma.rn.f32.bf16 %r857, %rs283, %rs283, %r856; + fma.rn.f32.bf16 %r858, %rs284, %rs284, %r857; + fma.rn.f32.bf16 %r859, %rs285, %rs285, %r858; + fma.rn.f32.bf16 %r860, %rs286, %rs286, %r654; + fma.rn.f32.bf16 %r861, %rs288, %rs288, %r860; + fma.rn.f32.bf16 %r862, %rs289, %rs289, %r861; + fma.rn.f32.bf16 %r863, %rs290, %rs290, %r862; + fma.rn.f32.bf16 %r864, %rs291, %rs291, %r863; + fma.rn.f32.bf16 %r865, %rs292, %rs292, %r864; + fma.rn.f32.bf16 %r866, %rs293, %rs293, %r865; + fma.rn.f32.bf16 %r867, %rs294, %rs294, %r655; + fma.rn.f32.bf16 %r868, %rs296, %rs296, %r867; + fma.rn.f32.bf16 %r869, %rs297, %rs297, %r868; + fma.rn.f32.bf16 %r870, %rs298, %rs298, %r869; + fma.rn.f32.bf16 %r871, %rs299, %rs299, %r870; + fma.rn.f32.bf16 %r872, %rs300, %rs300, %r871; + fma.rn.f32.bf16 %r873, %rs301, %rs301, %r872; + fma.rn.f32.bf16 %r874, %rs302, %rs302, %r656; + fma.rn.f32.bf16 %r875, %rs304, %rs304, %r874; + fma.rn.f32.bf16 %r876, %rs305, %rs305, %r875; + fma.rn.f32.bf16 %r877, %rs306, %rs306, %r876; + fma.rn.f32.bf16 %r878, %rs307, %rs307, %r877; + fma.rn.f32.bf16 %r879, %rs308, %rs308, %r878; + fma.rn.f32.bf16 %r880, %rs309, %rs309, %r879; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r881, %r663, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r882, %r663, %r881; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r883, %r882, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r884, %r882, %r883; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r885, %r884, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r886, %r884, %r885; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r887, %r886, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r888, %r886, %r887; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r889, %r670, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r890, %r670, %r889; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r891, %r890, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r892, %r890, %r891; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r893, %r892, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r894, %r892, %r893; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r895, %r894, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r896, %r894, %r895; +$L__tmp18: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r897, %r677, 8, 31, -1; +$L__tmp19: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r898, %r677, %r897; +$L__tmp20: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r899, %r898, 4, 31, -1; +$L__tmp21: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r900, %r898, %r899; +$L__tmp22: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r901, %r900, 2, 31, -1; +$L__tmp23: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r902, %r900, %r901; +$L__tmp24: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r903, %r902, 1, 31, -1; +$L__tmp25: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r904, %r902, %r903; +$L__tmp26: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r905, %r684, 8, 31, -1; +$L__tmp27: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r906, %r684, %r905; +$L__tmp28: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r907, %r906, 4, 31, -1; +$L__tmp29: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r908, %r906, %r907; +$L__tmp30: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r909, %r908, 2, 31, -1; +$L__tmp31: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r910, %r908, %r909; +$L__tmp32: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r911, %r910, 1, 31, -1; +$L__tmp33: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r912, %r910, %r911; +$L__tmp34: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r913, %r691, 8, 31, -1; +$L__tmp35: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r914, %r691, %r913; +$L__tmp36: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r915, %r914, 4, 31, -1; +$L__tmp37: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r916, %r914, %r915; +$L__tmp38: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r917, %r916, 2, 31, -1; +$L__tmp39: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r918, %r916, %r917; +$L__tmp40: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r919, %r918, 1, 31, -1; +$L__tmp41: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r920, %r918, %r919; +$L__tmp42: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r921, %r698, 8, 31, -1; +$L__tmp43: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r922, %r698, %r921; +$L__tmp44: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r923, %r922, 4, 31, -1; +$L__tmp45: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r924, %r922, %r923; +$L__tmp46: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r925, %r924, 2, 31, -1; +$L__tmp47: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r926, %r924, %r925; +$L__tmp48: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r927, %r926, 1, 31, -1; +$L__tmp49: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r928, %r926, %r927; +$L__tmp50: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r929, %r705, 8, 31, -1; +$L__tmp51: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r930, %r705, %r929; +$L__tmp52: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r931, %r930, 4, 31, -1; +$L__tmp53: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r932, %r930, %r931; +$L__tmp54: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r933, %r932, 2, 31, -1; +$L__tmp55: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r934, %r932, %r933; +$L__tmp56: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r935, %r934, 1, 31, -1; +$L__tmp57: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r936, %r934, %r935; +$L__tmp58: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r937, %r712, 8, 31, -1; +$L__tmp59: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r938, %r712, %r937; +$L__tmp60: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r939, %r938, 4, 31, -1; +$L__tmp61: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r940, %r938, %r939; +$L__tmp62: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r941, %r940, 2, 31, -1; +$L__tmp63: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r942, %r940, %r941; +$L__tmp64: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r943, %r942, 1, 31, -1; +$L__tmp65: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r944, %r942, %r943; +$L__tmp66: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r945, %r719, 8, 31, -1; +$L__tmp67: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r946, %r719, %r945; +$L__tmp68: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r947, %r946, 4, 31, -1; +$L__tmp69: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r948, %r946, %r947; +$L__tmp70: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r949, %r948, 2, 31, -1; +$L__tmp71: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r950, %r948, %r949; +$L__tmp72: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r951, %r950, 1, 31, -1; +$L__tmp73: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r952, %r950, %r951; +$L__tmp74: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r953, %r726, 8, 31, -1; +$L__tmp75: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r954, %r726, %r953; +$L__tmp76: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r955, %r954, 4, 31, -1; +$L__tmp77: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r956, %r954, %r955; +$L__tmp78: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r957, %r956, 2, 31, -1; +$L__tmp79: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r958, %r956, %r957; +$L__tmp80: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r959, %r958, 1, 31, -1; +$L__tmp81: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r960, %r958, %r959; +$L__tmp82: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r961, %r733, 8, 31, -1; +$L__tmp83: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r962, %r733, %r961; +$L__tmp84: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r963, %r962, 4, 31, -1; +$L__tmp85: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r964, %r962, %r963; +$L__tmp86: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r965, %r964, 2, 31, -1; +$L__tmp87: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r966, %r964, %r965; +$L__tmp88: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r967, %r966, 1, 31, -1; +$L__tmp89: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r968, %r966, %r967; +$L__tmp90: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r969, %r740, 8, 31, -1; +$L__tmp91: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r970, %r740, %r969; +$L__tmp92: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r971, %r970, 4, 31, -1; +$L__tmp93: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r972, %r970, %r971; +$L__tmp94: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r973, %r972, 2, 31, -1; +$L__tmp95: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r974, %r972, %r973; +$L__tmp96: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r975, %r974, 1, 31, -1; +$L__tmp97: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r976, %r974, %r975; +$L__tmp98: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r977, %r747, 8, 31, -1; +$L__tmp99: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r978, %r747, %r977; +$L__tmp100: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r979, %r978, 4, 31, -1; +$L__tmp101: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r980, %r978, %r979; +$L__tmp102: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r981, %r980, 2, 31, -1; +$L__tmp103: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r982, %r980, %r981; +$L__tmp104: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r983, %r982, 1, 31, -1; +$L__tmp105: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r984, %r982, %r983; +$L__tmp106: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r985, %r754, 8, 31, -1; +$L__tmp107: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r986, %r754, %r985; +$L__tmp108: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r987, %r986, 4, 31, -1; +$L__tmp109: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r988, %r986, %r987; +$L__tmp110: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r989, %r988, 2, 31, -1; +$L__tmp111: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r990, %r988, %r989; +$L__tmp112: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r991, %r990, 1, 31, -1; +$L__tmp113: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r992, %r990, %r991; +$L__tmp114: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r993, %r761, 8, 31, -1; +$L__tmp115: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r994, %r761, %r993; +$L__tmp116: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r995, %r994, 4, 31, -1; +$L__tmp117: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r996, %r994, %r995; +$L__tmp118: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r997, %r996, 2, 31, -1; +$L__tmp119: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r998, %r996, %r997; +$L__tmp120: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r999, %r998, 1, 31, -1; +$L__tmp121: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1000, %r998, %r999; +$L__tmp122: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1001, %r768, 8, 31, -1; +$L__tmp123: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1002, %r768, %r1001; +$L__tmp124: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1003, %r1002, 4, 31, -1; +$L__tmp125: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1004, %r1002, %r1003; +$L__tmp126: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1005, %r1004, 2, 31, -1; +$L__tmp127: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1006, %r1004, %r1005; +$L__tmp128: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1007, %r1006, 1, 31, -1; +$L__tmp129: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1008, %r1006, %r1007; +$L__tmp130: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1009, %r775, 8, 31, -1; +$L__tmp131: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1010, %r775, %r1009; +$L__tmp132: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1011, %r1010, 4, 31, -1; +$L__tmp133: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1012, %r1010, %r1011; +$L__tmp134: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1013, %r1012, 2, 31, -1; +$L__tmp135: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1014, %r1012, %r1013; +$L__tmp136: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1015, %r1014, 1, 31, -1; +$L__tmp137: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1016, %r1014, %r1015; +$L__tmp138: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1017, %r782, 8, 31, -1; +$L__tmp139: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1018, %r782, %r1017; +$L__tmp140: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1019, %r1018, 4, 31, -1; +$L__tmp141: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1020, %r1018, %r1019; +$L__tmp142: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1021, %r1020, 2, 31, -1; +$L__tmp143: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1022, %r1020, %r1021; +$L__tmp144: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1023, %r1022, 1, 31, -1; +$L__tmp145: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1024, %r1022, %r1023; +$L__tmp146: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1025, %r789, 8, 31, -1; +$L__tmp147: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1026, %r789, %r1025; +$L__tmp148: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1027, %r1026, 4, 31, -1; +$L__tmp149: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1028, %r1026, %r1027; +$L__tmp150: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1029, %r1028, 2, 31, -1; +$L__tmp151: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1030, %r1028, %r1029; +$L__tmp152: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1031, %r1030, 1, 31, -1; +$L__tmp153: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1032, %r1030, %r1031; +$L__tmp154: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1033, %r796, 8, 31, -1; +$L__tmp155: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1034, %r796, %r1033; +$L__tmp156: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1035, %r1034, 4, 31, -1; +$L__tmp157: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1036, %r1034, %r1035; +$L__tmp158: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1037, %r1036, 2, 31, -1; +$L__tmp159: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1038, %r1036, %r1037; +$L__tmp160: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1039, %r1038, 1, 31, -1; +$L__tmp161: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1040, %r1038, %r1039; +$L__tmp162: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1041, %r803, 8, 31, -1; +$L__tmp163: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1042, %r803, %r1041; +$L__tmp164: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1043, %r1042, 4, 31, -1; +$L__tmp165: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1044, %r1042, %r1043; +$L__tmp166: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1045, %r1044, 2, 31, -1; +$L__tmp167: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1046, %r1044, %r1045; +$L__tmp168: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1047, %r1046, 1, 31, -1; +$L__tmp169: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1048, %r1046, %r1047; +$L__tmp170: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1049, %r810, 8, 31, -1; +$L__tmp171: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1050, %r810, %r1049; +$L__tmp172: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1051, %r1050, 4, 31, -1; +$L__tmp173: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1052, %r1050, %r1051; +$L__tmp174: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1053, %r1052, 2, 31, -1; +$L__tmp175: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1054, %r1052, %r1053; +$L__tmp176: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1055, %r1054, 1, 31, -1; +$L__tmp177: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1056, %r1054, %r1055; +$L__tmp178: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1057, %r817, 8, 31, -1; +$L__tmp179: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1058, %r817, %r1057; +$L__tmp180: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1059, %r1058, 4, 31, -1; +$L__tmp181: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1060, %r1058, %r1059; +$L__tmp182: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1061, %r1060, 2, 31, -1; +$L__tmp183: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1062, %r1060, %r1061; +$L__tmp184: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1063, %r1062, 1, 31, -1; +$L__tmp185: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1064, %r1062, %r1063; +$L__tmp186: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1065, %r824, 8, 31, -1; +$L__tmp187: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1066, %r824, %r1065; +$L__tmp188: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1067, %r1066, 4, 31, -1; +$L__tmp189: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1068, %r1066, %r1067; +$L__tmp190: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1069, %r1068, 2, 31, -1; +$L__tmp191: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1070, %r1068, %r1069; +$L__tmp192: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1071, %r1070, 1, 31, -1; +$L__tmp193: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1072, %r1070, %r1071; +$L__tmp194: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1073, %r831, 8, 31, -1; +$L__tmp195: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1074, %r831, %r1073; +$L__tmp196: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1075, %r1074, 4, 31, -1; +$L__tmp197: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1076, %r1074, %r1075; +$L__tmp198: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1077, %r1076, 2, 31, -1; +$L__tmp199: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1078, %r1076, %r1077; +$L__tmp200: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1079, %r1078, 1, 31, -1; +$L__tmp201: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1080, %r1078, %r1079; +$L__tmp202: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1081, %r838, 8, 31, -1; +$L__tmp203: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1082, %r838, %r1081; +$L__tmp204: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1083, %r1082, 4, 31, -1; +$L__tmp205: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1084, %r1082, %r1083; +$L__tmp206: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1085, %r1084, 2, 31, -1; +$L__tmp207: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1086, %r1084, %r1085; +$L__tmp208: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1087, %r1086, 1, 31, -1; +$L__tmp209: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1088, %r1086, %r1087; +$L__tmp210: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1089, %r845, 8, 31, -1; +$L__tmp211: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1090, %r845, %r1089; +$L__tmp212: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1091, %r1090, 4, 31, -1; +$L__tmp213: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1092, %r1090, %r1091; +$L__tmp214: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1093, %r1092, 2, 31, -1; +$L__tmp215: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1094, %r1092, %r1093; +$L__tmp216: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1095, %r1094, 1, 31, -1; +$L__tmp217: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1096, %r1094, %r1095; +$L__tmp218: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1097, %r852, 8, 31, -1; +$L__tmp219: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1098, %r852, %r1097; +$L__tmp220: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1099, %r1098, 4, 31, -1; +$L__tmp221: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1100, %r1098, %r1099; +$L__tmp222: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1101, %r1100, 2, 31, -1; +$L__tmp223: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1102, %r1100, %r1101; +$L__tmp224: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1103, %r1102, 1, 31, -1; +$L__tmp225: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1104, %r1102, %r1103; +$L__tmp226: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1105, %r859, 8, 31, -1; +$L__tmp227: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1106, %r859, %r1105; +$L__tmp228: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1107, %r1106, 4, 31, -1; +$L__tmp229: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1108, %r1106, %r1107; +$L__tmp230: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1109, %r1108, 2, 31, -1; +$L__tmp231: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1110, %r1108, %r1109; +$L__tmp232: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1111, %r1110, 1, 31, -1; +$L__tmp233: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1112, %r1110, %r1111; +$L__tmp234: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1113, %r866, 8, 31, -1; +$L__tmp235: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1114, %r866, %r1113; +$L__tmp236: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1115, %r1114, 4, 31, -1; +$L__tmp237: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1116, %r1114, %r1115; +$L__tmp238: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1117, %r1116, 2, 31, -1; +$L__tmp239: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1118, %r1116, %r1117; +$L__tmp240: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1119, %r1118, 1, 31, -1; +$L__tmp241: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1120, %r1118, %r1119; +$L__tmp242: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1121, %r873, 8, 31, -1; +$L__tmp243: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1122, %r873, %r1121; +$L__tmp244: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1123, %r1122, 4, 31, -1; +$L__tmp245: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1124, %r1122, %r1123; +$L__tmp246: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1125, %r1124, 2, 31, -1; +$L__tmp247: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1126, %r1124, %r1125; +$L__tmp248: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1127, %r1126, 1, 31, -1; +$L__tmp249: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1128, %r1126, %r1127; +$L__tmp250: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1129, %r880, 8, 31, -1; +$L__tmp251: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1130, %r880, %r1129; +$L__tmp252: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1131, %r1130, 4, 31, -1; +$L__tmp253: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1132, %r1130, %r1131; +$L__tmp254: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1133, %r1132, 2, 31, -1; +$L__tmp255: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1134, %r1132, %r1133; +$L__tmp256: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r1135, %r1134, 1, 31, -1; +$L__tmp257: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r1136, %r1134, %r1135; +$L__tmp258: + .loc 1 80 48 // l2norm.py:80:48 + add.f32 %r1137, %r301, %r888; + add.f32 %r1138, %r301, %r896; + add.f32 %r1139, %r301, %r904; + add.f32 %r1140, %r301, %r912; + add.f32 %r1141, %r301, %r920; + add.f32 %r1142, %r301, %r928; + add.f32 %r1143, %r301, %r936; + add.f32 %r1144, %r301, %r944; + add.f32 %r1145, %r301, %r952; + add.f32 %r1146, %r301, %r960; + add.f32 %r1147, %r301, %r968; + add.f32 %r1148, %r301, %r976; + add.f32 %r1149, %r301, %r984; + add.f32 %r1150, %r301, %r992; + add.f32 %r1151, %r301, %r1000; + add.f32 %r1152, %r301, %r1008; + add.f32 %r1153, %r301, %r1016; + add.f32 %r1154, %r301, %r1024; + add.f32 %r1155, %r301, %r1032; + add.f32 %r1156, %r301, %r1040; + add.f32 %r1157, %r301, %r1048; + add.f32 %r1158, %r301, %r1056; + add.f32 %r1159, %r301, %r1064; + add.f32 %r1160, %r301, %r1072; + add.f32 %r1161, %r301, %r1080; + add.f32 %r1162, %r301, %r1088; + add.f32 %r1163, %r301, %r1096; + add.f32 %r1164, %r301, %r1104; + add.f32 %r1165, %r301, %r1112; + add.f32 %r1166, %r301, %r1120; + add.f32 %r1167, %r301, %r1128; + add.f32 %r1168, %r301, %r1136; + .loc 1 80 25 // l2norm.py:80:25 + sqrt.approx.ftz.f32 %r1169, %r1137; + sqrt.approx.ftz.f32 %r1170, %r1138; + sqrt.approx.ftz.f32 %r1171, %r1139; + sqrt.approx.ftz.f32 %r1172, %r1140; + sqrt.approx.ftz.f32 %r1173, %r1141; + sqrt.approx.ftz.f32 %r1174, %r1142; + sqrt.approx.ftz.f32 %r1175, %r1143; + sqrt.approx.ftz.f32 %r1176, %r1144; + sqrt.approx.ftz.f32 %r1177, %r1145; + sqrt.approx.ftz.f32 %r1178, %r1146; + sqrt.approx.ftz.f32 %r1179, %r1147; + sqrt.approx.ftz.f32 %r1180, %r1148; + sqrt.approx.ftz.f32 %r1181, %r1149; + sqrt.approx.ftz.f32 %r1182, %r1150; + sqrt.approx.ftz.f32 %r1183, %r1151; + sqrt.approx.ftz.f32 %r1184, %r1152; + sqrt.approx.ftz.f32 %r1185, %r1153; + sqrt.approx.ftz.f32 %r1186, %r1154; + sqrt.approx.ftz.f32 %r1187, %r1155; + sqrt.approx.ftz.f32 %r1188, %r1156; + sqrt.approx.ftz.f32 %r1189, %r1157; + sqrt.approx.ftz.f32 %r1190, %r1158; + sqrt.approx.ftz.f32 %r1191, %r1159; + sqrt.approx.ftz.f32 %r1192, %r1160; + sqrt.approx.ftz.f32 %r1193, %r1161; + sqrt.approx.ftz.f32 %r1194, %r1162; + sqrt.approx.ftz.f32 %r1195, %r1163; + sqrt.approx.ftz.f32 %r1196, %r1164; + sqrt.approx.ftz.f32 %r1197, %r1165; + sqrt.approx.ftz.f32 %r1198, %r1166; + sqrt.approx.ftz.f32 %r1199, %r1167; + sqrt.approx.ftz.f32 %r1200, %r1168; + .loc 1 71 34 // l2norm.py:71:34 + shl.b32 %r1201, %r302, 1; + and.b32 %r1202, %r1201, 62; + or.b32 %r1203, %r1202, 1; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd201, %r1203; + or.b64 %rd202, %rd70, %rd201; + cvt.u64.u32 %rd203, %r1202; + or.b64 %rd204, %rd70, %rd203; + .loc 1 77 20 // l2norm.py:77:20 + shl.b64 %rd205, %rd204, 2; + add.s64 %rd65, %rd69, %rd205; + add.s64 %rd66, %rd65, 4; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd206, %rd68, %rd198; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd64, %rd206, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd207, %rd68, %rd196; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd63, %rd207, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd208, %rd68, %rd194; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd62, %rd208, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd209, %rd68, %rd192; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd61, %rd209, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd210, %rd68, %rd190; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd60, %rd210, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd211, %rd68, %rd188; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd59, %rd211, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd212, %rd68, %rd186; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd58, %rd212, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd213, %rd68, %rd184; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd57, %rd213, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd214, %rd68, %rd182; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd56, %rd214, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd215, %rd68, %rd180; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd55, %rd215, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd216, %rd68, %rd178; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd54, %rd216, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd217, %rd68, %rd176; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd53, %rd217, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd218, %rd68, %rd174; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd52, %rd218, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd219, %rd68, %rd172; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd51, %rd219, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd220, %rd68, %rd170; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd50, %rd220, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd221, %rd68, %rd168; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd49, %rd221, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd222, %rd68, %rd166; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd48, %rd222, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd223, %rd68, %rd164; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd47, %rd223, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd224, %rd68, %rd162; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd46, %rd224, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd225, %rd68, %rd160; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd45, %rd225, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd226, %rd68, %rd158; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd44, %rd226, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd227, %rd68, %rd156; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd43, %rd227, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd228, %rd68, %rd154; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd42, %rd228, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd229, %rd68, %rd152; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd41, %rd229, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd230, %rd68, %rd150; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd40, %rd230, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd231, %rd68, %rd148; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd39, %rd231, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd232, %rd68, %rd146; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd38, %rd232, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd233, %rd68, %rd144; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd37, %rd233, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd234, %rd68, %rd142; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd36, %rd234, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd235, %rd68, %rd140; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd35, %rd235, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd236, %rd68, %rd138; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd34, %rd236, %rd200; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd237, %rd68, %rd136; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd33, %rd237, %rd200; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p34, %rd202, %rd71; + setp.lt.s64 %p33, %rd204, %rd71; + mov.b32 %r1204, 0f3F800000; + .loc 1 80 17 // l2norm.py:80:17 + div.full.f32 %r259, %r1204, %r1169; + div.full.f32 %r260, %r1204, %r1170; + div.full.f32 %r261, %r1204, %r1171; + div.full.f32 %r262, %r1204, %r1172; + div.full.f32 %r264, %r1204, %r1173; + div.full.f32 %r265, %r1204, %r1174; + div.full.f32 %r266, %r1204, %r1175; + div.full.f32 %r267, %r1204, %r1176; + div.full.f32 %r269, %r1204, %r1177; + div.full.f32 %r270, %r1204, %r1178; + div.full.f32 %r271, %r1204, %r1179; + div.full.f32 %r272, %r1204, %r1180; + div.full.f32 %r274, %r1204, %r1181; + div.full.f32 %r275, %r1204, %r1182; + div.full.f32 %r276, %r1204, %r1183; + div.full.f32 %r277, %r1204, %r1184; + div.full.f32 %r279, %r1204, %r1185; + div.full.f32 %r280, %r1204, %r1186; + div.full.f32 %r281, %r1204, %r1187; + div.full.f32 %r282, %r1204, %r1188; + div.full.f32 %r284, %r1204, %r1189; + div.full.f32 %r285, %r1204, %r1190; + div.full.f32 %r286, %r1204, %r1191; + div.full.f32 %r287, %r1204, %r1192; + div.full.f32 %r289, %r1204, %r1193; + div.full.f32 %r290, %r1204, %r1194; + div.full.f32 %r291, %r1204, %r1195; + div.full.f32 %r292, %r1204, %r1196; + div.full.f32 %r294, %r1204, %r1197; + div.full.f32 %r295, %r1204, %r1198; + div.full.f32 %r296, %r1204, %r1199; + div.full.f32 %r297, %r1204, %r1200; + .loc 1 81 16 // l2norm.py:81:16 + mul.f32 %r1205, %r259, %r369; + mul.f32 %r1206, %r259, %r370; + mul.f32 %r1207, %r259, %r371; + mul.f32 %r1208, %r259, %r372; + mul.f32 %r1209, %r259, %r373; + mul.f32 %r1210, %r259, %r374; + mul.f32 %r1211, %r259, %r375; + mul.f32 %r1212, %r259, %r376; + mul.f32 %r1213, %r260, %r377; + mul.f32 %r1214, %r260, %r378; + mul.f32 %r1215, %r260, %r379; + mul.f32 %r1216, %r260, %r380; + mul.f32 %r1217, %r260, %r381; + mul.f32 %r1218, %r260, %r382; + mul.f32 %r1219, %r260, %r383; + mul.f32 %r1220, %r260, %r384; + mul.f32 %r1221, %r261, %r385; + mul.f32 %r1222, %r261, %r386; + mul.f32 %r1223, %r261, %r387; + mul.f32 %r1224, %r261, %r388; + mul.f32 %r1225, %r261, %r389; + mul.f32 %r1226, %r261, %r390; + mul.f32 %r1227, %r261, %r391; + mul.f32 %r1228, %r261, %r392; + mul.f32 %r1229, %r262, %r393; + mul.f32 %r1230, %r262, %r394; + mul.f32 %r1231, %r262, %r395; + mul.f32 %r1232, %r262, %r396; + mul.f32 %r1233, %r262, %r397; + mul.f32 %r1234, %r262, %r398; + mul.f32 %r1235, %r262, %r399; + mul.f32 %r1236, %r262, %r400; + mul.f32 %r1237, %r264, %r401; + mul.f32 %r1238, %r264, %r402; + mul.f32 %r1239, %r264, %r403; + mul.f32 %r1240, %r264, %r404; + mul.f32 %r1241, %r264, %r405; + mul.f32 %r1242, %r264, %r406; + mul.f32 %r1243, %r264, %r407; + mul.f32 %r1244, %r264, %r408; + mul.f32 %r1245, %r265, %r409; + mul.f32 %r1246, %r265, %r410; + mul.f32 %r1247, %r265, %r411; + mul.f32 %r1248, %r265, %r412; + mul.f32 %r1249, %r265, %r413; + mul.f32 %r1250, %r265, %r414; + mul.f32 %r1251, %r265, %r415; + mul.f32 %r1252, %r265, %r416; + mul.f32 %r1253, %r266, %r417; + mul.f32 %r1254, %r266, %r418; + mul.f32 %r1255, %r266, %r419; + mul.f32 %r1256, %r266, %r420; + mul.f32 %r1257, %r266, %r421; + mul.f32 %r1258, %r266, %r422; + mul.f32 %r1259, %r266, %r423; + mul.f32 %r1260, %r266, %r424; + mul.f32 %r1261, %r267, %r425; + mul.f32 %r1262, %r267, %r426; + mul.f32 %r1263, %r267, %r427; + mul.f32 %r1264, %r267, %r428; + mul.f32 %r1265, %r267, %r429; + mul.f32 %r1266, %r267, %r430; + mul.f32 %r1267, %r267, %r431; + mul.f32 %r1268, %r267, %r432; + mul.f32 %r1269, %r269, %r433; + mul.f32 %r1270, %r269, %r434; + mul.f32 %r1271, %r269, %r435; + mul.f32 %r1272, %r269, %r436; + mul.f32 %r1273, %r269, %r437; + mul.f32 %r1274, %r269, %r438; + mul.f32 %r1275, %r269, %r439; + mul.f32 %r1276, %r269, %r440; + mul.f32 %r1277, %r270, %r441; + mul.f32 %r1278, %r270, %r442; + mul.f32 %r1279, %r270, %r443; + mul.f32 %r1280, %r270, %r444; + mul.f32 %r1281, %r270, %r445; + mul.f32 %r1282, %r270, %r446; + mul.f32 %r1283, %r270, %r447; + mul.f32 %r1284, %r270, %r448; + mul.f32 %r1285, %r271, %r449; + mul.f32 %r1286, %r271, %r450; + mul.f32 %r1287, %r271, %r451; + mul.f32 %r1288, %r271, %r452; + mul.f32 %r1289, %r271, %r453; + mul.f32 %r1290, %r271, %r454; + mul.f32 %r1291, %r271, %r455; + mul.f32 %r1292, %r271, %r456; + mul.f32 %r1293, %r272, %r457; + mul.f32 %r1294, %r272, %r458; + mul.f32 %r1295, %r272, %r459; + mul.f32 %r1296, %r272, %r460; + mul.f32 %r1297, %r272, %r461; + mul.f32 %r1298, %r272, %r462; + mul.f32 %r1299, %r272, %r463; + mul.f32 %r1300, %r272, %r464; + mul.f32 %r1301, %r274, %r465; + mul.f32 %r1302, %r274, %r466; + mul.f32 %r1303, %r274, %r467; + mul.f32 %r1304, %r274, %r468; + mul.f32 %r1305, %r274, %r469; + mul.f32 %r1306, %r274, %r470; + mul.f32 %r1307, %r274, %r471; + mul.f32 %r1308, %r274, %r472; + mul.f32 %r1309, %r275, %r473; + mul.f32 %r1310, %r275, %r474; + mul.f32 %r1311, %r275, %r475; + mul.f32 %r1312, %r275, %r476; + mul.f32 %r1313, %r275, %r477; + mul.f32 %r1314, %r275, %r478; + mul.f32 %r1315, %r275, %r479; + mul.f32 %r1316, %r275, %r480; + mul.f32 %r1317, %r276, %r481; + mul.f32 %r1318, %r276, %r482; + mul.f32 %r1319, %r276, %r483; + mul.f32 %r1320, %r276, %r484; + mul.f32 %r1321, %r276, %r485; + mul.f32 %r1322, %r276, %r486; + mul.f32 %r1323, %r276, %r487; + mul.f32 %r1324, %r276, %r488; + mul.f32 %r1325, %r277, %r489; + mul.f32 %r1326, %r277, %r490; + mul.f32 %r1327, %r277, %r491; + mul.f32 %r1328, %r277, %r492; + mul.f32 %r1329, %r277, %r493; + mul.f32 %r1330, %r277, %r494; + mul.f32 %r1331, %r277, %r495; + mul.f32 %r1332, %r277, %r496; + mul.f32 %r1333, %r279, %r497; + mul.f32 %r1334, %r279, %r498; + mul.f32 %r1335, %r279, %r499; + mul.f32 %r1336, %r279, %r500; + mul.f32 %r1337, %r279, %r501; + mul.f32 %r1338, %r279, %r502; + mul.f32 %r1339, %r279, %r503; + mul.f32 %r1340, %r279, %r504; + mul.f32 %r1341, %r280, %r505; + mul.f32 %r1342, %r280, %r506; + mul.f32 %r1343, %r280, %r507; + mul.f32 %r1344, %r280, %r508; + mul.f32 %r1345, %r280, %r509; + mul.f32 %r1346, %r280, %r510; + mul.f32 %r1347, %r280, %r511; + mul.f32 %r1348, %r280, %r512; + mul.f32 %r1349, %r281, %r513; + mul.f32 %r1350, %r281, %r514; + mul.f32 %r1351, %r281, %r515; + mul.f32 %r1352, %r281, %r516; + mul.f32 %r1353, %r281, %r517; + mul.f32 %r1354, %r281, %r518; + mul.f32 %r1355, %r281, %r519; + mul.f32 %r1356, %r281, %r520; + mul.f32 %r1357, %r282, %r521; + mul.f32 %r1358, %r282, %r522; + mul.f32 %r1359, %r282, %r523; + mul.f32 %r1360, %r282, %r524; + mul.f32 %r1361, %r282, %r525; + mul.f32 %r1362, %r282, %r526; + mul.f32 %r1363, %r282, %r527; + mul.f32 %r1364, %r282, %r528; + mul.f32 %r1365, %r284, %r529; + mul.f32 %r1366, %r284, %r530; + mul.f32 %r1367, %r284, %r531; + mul.f32 %r1368, %r284, %r532; + mul.f32 %r1369, %r284, %r533; + mul.f32 %r1370, %r284, %r534; + mul.f32 %r1371, %r284, %r535; + mul.f32 %r1372, %r284, %r536; + mul.f32 %r1373, %r285, %r537; + mul.f32 %r1374, %r285, %r538; + mul.f32 %r1375, %r285, %r539; + mul.f32 %r1376, %r285, %r540; + mul.f32 %r1377, %r285, %r541; + mul.f32 %r1378, %r285, %r542; + mul.f32 %r1379, %r285, %r543; + mul.f32 %r1380, %r285, %r544; + mul.f32 %r1381, %r286, %r545; + mul.f32 %r1382, %r286, %r546; + mul.f32 %r1383, %r286, %r547; + mul.f32 %r1384, %r286, %r548; + mul.f32 %r1385, %r286, %r549; + mul.f32 %r1386, %r286, %r550; + mul.f32 %r1387, %r286, %r551; + mul.f32 %r1388, %r286, %r552; + mul.f32 %r1389, %r287, %r553; + mul.f32 %r1390, %r287, %r554; + mul.f32 %r1391, %r287, %r555; + mul.f32 %r1392, %r287, %r556; + mul.f32 %r1393, %r287, %r557; + mul.f32 %r1394, %r287, %r558; + mul.f32 %r1395, %r287, %r559; + mul.f32 %r1396, %r287, %r560; + mul.f32 %r1397, %r289, %r561; + mul.f32 %r1398, %r289, %r562; + mul.f32 %r1399, %r289, %r563; + mul.f32 %r1400, %r289, %r564; + mul.f32 %r1401, %r289, %r565; + mul.f32 %r1402, %r289, %r566; + mul.f32 %r1403, %r289, %r567; + mul.f32 %r1404, %r289, %r568; + mul.f32 %r1405, %r290, %r569; + mul.f32 %r1406, %r290, %r570; + mul.f32 %r1407, %r290, %r571; + mul.f32 %r1408, %r290, %r572; + mul.f32 %r1409, %r290, %r573; + mul.f32 %r1410, %r290, %r574; + mul.f32 %r1411, %r290, %r575; + mul.f32 %r1412, %r290, %r576; + mul.f32 %r1413, %r291, %r577; + mul.f32 %r1414, %r291, %r578; + mul.f32 %r1415, %r291, %r579; + mul.f32 %r1416, %r291, %r580; + mul.f32 %r1417, %r291, %r581; + mul.f32 %r1418, %r291, %r582; + mul.f32 %r1419, %r291, %r583; + mul.f32 %r1420, %r291, %r584; + mul.f32 %r1421, %r292, %r585; + mul.f32 %r1422, %r292, %r586; + mul.f32 %r1423, %r292, %r587; + mul.f32 %r1424, %r292, %r588; + mul.f32 %r1425, %r292, %r589; + mul.f32 %r1426, %r292, %r590; + mul.f32 %r1427, %r292, %r591; + mul.f32 %r1428, %r292, %r592; + mul.f32 %r1429, %r294, %r593; + mul.f32 %r1430, %r294, %r594; + mul.f32 %r1431, %r294, %r595; + mul.f32 %r1432, %r294, %r596; + mul.f32 %r1433, %r294, %r597; + mul.f32 %r1434, %r294, %r598; + mul.f32 %r1435, %r294, %r599; + mul.f32 %r1436, %r294, %r600; + mul.f32 %r1437, %r295, %r601; + mul.f32 %r1438, %r295, %r602; + mul.f32 %r1439, %r295, %r603; + mul.f32 %r1440, %r295, %r604; + mul.f32 %r1441, %r295, %r605; + mul.f32 %r1442, %r295, %r606; + mul.f32 %r1443, %r295, %r607; + mul.f32 %r1444, %r295, %r608; + mul.f32 %r1445, %r296, %r609; + mul.f32 %r1446, %r296, %r610; + mul.f32 %r1447, %r296, %r611; + mul.f32 %r1448, %r296, %r612; + mul.f32 %r1449, %r296, %r613; + mul.f32 %r1450, %r296, %r614; + mul.f32 %r1451, %r296, %r615; + mul.f32 %r1452, %r296, %r616; + mul.f32 %r1453, %r297, %r617; + mul.f32 %r1454, %r297, %r618; + mul.f32 %r1455, %r297, %r619; + mul.f32 %r1456, %r297, %r620; + mul.f32 %r1457, %r297, %r621; + mul.f32 %r1458, %r297, %r622; + mul.f32 %r1459, %r297, %r623; + mul.f32 %r1460, %r297, %r624; + .loc 1 83 18 // l2norm.py:83:18 + cvt.rn.bf16x2.f32 %r130, %r1206, %r1205; + cvt.rn.bf16x2.f32 %r131, %r1208, %r1207; + cvt.rn.bf16x2.f32 %r132, %r1210, %r1209; + cvt.rn.bf16x2.f32 %r133, %r1212, %r1211; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd33 + 0 ], { %r130, %r131, %r132, %r133 }; + // end inline asm + cvt.rn.bf16x2.f32 %r134, %r1214, %r1213; + cvt.rn.bf16x2.f32 %r135, %r1216, %r1215; + cvt.rn.bf16x2.f32 %r136, %r1218, %r1217; + cvt.rn.bf16x2.f32 %r137, %r1220, %r1219; + // begin inline asm + @%p3 st.global.v4.b32 [ %rd34 + 0 ], { %r134, %r135, %r136, %r137 }; + // end inline asm + cvt.rn.bf16x2.f32 %r138, %r1222, %r1221; + cvt.rn.bf16x2.f32 %r139, %r1224, %r1223; + cvt.rn.bf16x2.f32 %r140, %r1226, %r1225; + cvt.rn.bf16x2.f32 %r141, %r1228, %r1227; + // begin inline asm + @%p4 st.global.v4.b32 [ %rd35 + 0 ], { %r138, %r139, %r140, %r141 }; + // end inline asm + cvt.rn.bf16x2.f32 %r142, %r1230, %r1229; + cvt.rn.bf16x2.f32 %r143, %r1232, %r1231; + cvt.rn.bf16x2.f32 %r144, %r1234, %r1233; + cvt.rn.bf16x2.f32 %r145, %r1236, %r1235; + // begin inline asm + @%p5 st.global.v4.b32 [ %rd36 + 0 ], { %r142, %r143, %r144, %r145 }; + // end inline asm + cvt.rn.bf16x2.f32 %r146, %r1238, %r1237; + cvt.rn.bf16x2.f32 %r147, %r1240, %r1239; + cvt.rn.bf16x2.f32 %r148, %r1242, %r1241; + cvt.rn.bf16x2.f32 %r149, %r1244, %r1243; + // begin inline asm + @%p6 st.global.v4.b32 [ %rd37 + 0 ], { %r146, %r147, %r148, %r149 }; + // end inline asm + cvt.rn.bf16x2.f32 %r150, %r1246, %r1245; + cvt.rn.bf16x2.f32 %r151, %r1248, %r1247; + cvt.rn.bf16x2.f32 %r152, %r1250, %r1249; + cvt.rn.bf16x2.f32 %r153, %r1252, %r1251; + // begin inline asm + @%p7 st.global.v4.b32 [ %rd38 + 0 ], { %r150, %r151, %r152, %r153 }; + // end inline asm + cvt.rn.bf16x2.f32 %r154, %r1254, %r1253; + cvt.rn.bf16x2.f32 %r155, %r1256, %r1255; + cvt.rn.bf16x2.f32 %r156, %r1258, %r1257; + cvt.rn.bf16x2.f32 %r157, %r1260, %r1259; + // begin inline asm + @%p8 st.global.v4.b32 [ %rd39 + 0 ], { %r154, %r155, %r156, %r157 }; + // end inline asm + cvt.rn.bf16x2.f32 %r158, %r1262, %r1261; + cvt.rn.bf16x2.f32 %r159, %r1264, %r1263; + cvt.rn.bf16x2.f32 %r160, %r1266, %r1265; + cvt.rn.bf16x2.f32 %r161, %r1268, %r1267; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd40 + 0 ], { %r158, %r159, %r160, %r161 }; + // end inline asm + cvt.rn.bf16x2.f32 %r162, %r1270, %r1269; + cvt.rn.bf16x2.f32 %r163, %r1272, %r1271; + cvt.rn.bf16x2.f32 %r164, %r1274, %r1273; + cvt.rn.bf16x2.f32 %r165, %r1276, %r1275; + // begin inline asm + @%p9 st.global.v4.b32 [ %rd41 + 0 ], { %r162, %r163, %r164, %r165 }; + // end inline asm + cvt.rn.bf16x2.f32 %r166, %r1278, %r1277; + cvt.rn.bf16x2.f32 %r167, %r1280, %r1279; + cvt.rn.bf16x2.f32 %r168, %r1282, %r1281; + cvt.rn.bf16x2.f32 %r169, %r1284, %r1283; + // begin inline asm + @%p10 st.global.v4.b32 [ %rd42 + 0 ], { %r166, %r167, %r168, %r169 }; + // end inline asm + cvt.rn.bf16x2.f32 %r170, %r1286, %r1285; + cvt.rn.bf16x2.f32 %r171, %r1288, %r1287; + cvt.rn.bf16x2.f32 %r172, %r1290, %r1289; + cvt.rn.bf16x2.f32 %r173, %r1292, %r1291; + // begin inline asm + @%p11 st.global.v4.b32 [ %rd43 + 0 ], { %r170, %r171, %r172, %r173 }; + // end inline asm + cvt.rn.bf16x2.f32 %r174, %r1294, %r1293; + cvt.rn.bf16x2.f32 %r175, %r1296, %r1295; + cvt.rn.bf16x2.f32 %r176, %r1298, %r1297; + cvt.rn.bf16x2.f32 %r177, %r1300, %r1299; + // begin inline asm + @%p12 st.global.v4.b32 [ %rd44 + 0 ], { %r174, %r175, %r176, %r177 }; + // end inline asm + cvt.rn.bf16x2.f32 %r178, %r1302, %r1301; + cvt.rn.bf16x2.f32 %r179, %r1304, %r1303; + cvt.rn.bf16x2.f32 %r180, %r1306, %r1305; + cvt.rn.bf16x2.f32 %r181, %r1308, %r1307; + // begin inline asm + @%p13 st.global.v4.b32 [ %rd45 + 0 ], { %r178, %r179, %r180, %r181 }; + // end inline asm + cvt.rn.bf16x2.f32 %r182, %r1310, %r1309; + cvt.rn.bf16x2.f32 %r183, %r1312, %r1311; + cvt.rn.bf16x2.f32 %r184, %r1314, %r1313; + cvt.rn.bf16x2.f32 %r185, %r1316, %r1315; + // begin inline asm + @%p14 st.global.v4.b32 [ %rd46 + 0 ], { %r182, %r183, %r184, %r185 }; + // end inline asm + cvt.rn.bf16x2.f32 %r186, %r1318, %r1317; + cvt.rn.bf16x2.f32 %r187, %r1320, %r1319; + cvt.rn.bf16x2.f32 %r188, %r1322, %r1321; + cvt.rn.bf16x2.f32 %r189, %r1324, %r1323; + // begin inline asm + @%p15 st.global.v4.b32 [ %rd47 + 0 ], { %r186, %r187, %r188, %r189 }; + // end inline asm + cvt.rn.bf16x2.f32 %r190, %r1326, %r1325; + cvt.rn.bf16x2.f32 %r191, %r1328, %r1327; + cvt.rn.bf16x2.f32 %r192, %r1330, %r1329; + cvt.rn.bf16x2.f32 %r193, %r1332, %r1331; + // begin inline asm + @%p16 st.global.v4.b32 [ %rd48 + 0 ], { %r190, %r191, %r192, %r193 }; + // end inline asm + cvt.rn.bf16x2.f32 %r194, %r1334, %r1333; + cvt.rn.bf16x2.f32 %r195, %r1336, %r1335; + cvt.rn.bf16x2.f32 %r196, %r1338, %r1337; + cvt.rn.bf16x2.f32 %r197, %r1340, %r1339; + // begin inline asm + @%p17 st.global.v4.b32 [ %rd49 + 0 ], { %r194, %r195, %r196, %r197 }; + // end inline asm + cvt.rn.bf16x2.f32 %r198, %r1342, %r1341; + cvt.rn.bf16x2.f32 %r199, %r1344, %r1343; + cvt.rn.bf16x2.f32 %r200, %r1346, %r1345; + cvt.rn.bf16x2.f32 %r201, %r1348, %r1347; + // begin inline asm + @%p18 st.global.v4.b32 [ %rd50 + 0 ], { %r198, %r199, %r200, %r201 }; + // end inline asm + cvt.rn.bf16x2.f32 %r202, %r1350, %r1349; + cvt.rn.bf16x2.f32 %r203, %r1352, %r1351; + cvt.rn.bf16x2.f32 %r204, %r1354, %r1353; + cvt.rn.bf16x2.f32 %r205, %r1356, %r1355; + // begin inline asm + @%p19 st.global.v4.b32 [ %rd51 + 0 ], { %r202, %r203, %r204, %r205 }; + // end inline asm + cvt.rn.bf16x2.f32 %r206, %r1358, %r1357; + cvt.rn.bf16x2.f32 %r207, %r1360, %r1359; + cvt.rn.bf16x2.f32 %r208, %r1362, %r1361; + cvt.rn.bf16x2.f32 %r209, %r1364, %r1363; + // begin inline asm + @%p20 st.global.v4.b32 [ %rd52 + 0 ], { %r206, %r207, %r208, %r209 }; + // end inline asm + cvt.rn.bf16x2.f32 %r210, %r1366, %r1365; + cvt.rn.bf16x2.f32 %r211, %r1368, %r1367; + cvt.rn.bf16x2.f32 %r212, %r1370, %r1369; + cvt.rn.bf16x2.f32 %r213, %r1372, %r1371; + // begin inline asm + @%p21 st.global.v4.b32 [ %rd53 + 0 ], { %r210, %r211, %r212, %r213 }; + // end inline asm + cvt.rn.bf16x2.f32 %r214, %r1374, %r1373; + cvt.rn.bf16x2.f32 %r215, %r1376, %r1375; + cvt.rn.bf16x2.f32 %r216, %r1378, %r1377; + cvt.rn.bf16x2.f32 %r217, %r1380, %r1379; + // begin inline asm + @%p22 st.global.v4.b32 [ %rd54 + 0 ], { %r214, %r215, %r216, %r217 }; + // end inline asm + cvt.rn.bf16x2.f32 %r218, %r1382, %r1381; + cvt.rn.bf16x2.f32 %r219, %r1384, %r1383; + cvt.rn.bf16x2.f32 %r220, %r1386, %r1385; + cvt.rn.bf16x2.f32 %r221, %r1388, %r1387; + // begin inline asm + @%p23 st.global.v4.b32 [ %rd55 + 0 ], { %r218, %r219, %r220, %r221 }; + // end inline asm + cvt.rn.bf16x2.f32 %r222, %r1390, %r1389; + cvt.rn.bf16x2.f32 %r223, %r1392, %r1391; + cvt.rn.bf16x2.f32 %r224, %r1394, %r1393; + cvt.rn.bf16x2.f32 %r225, %r1396, %r1395; + // begin inline asm + @%p24 st.global.v4.b32 [ %rd56 + 0 ], { %r222, %r223, %r224, %r225 }; + // end inline asm + cvt.rn.bf16x2.f32 %r226, %r1398, %r1397; + cvt.rn.bf16x2.f32 %r227, %r1400, %r1399; + cvt.rn.bf16x2.f32 %r228, %r1402, %r1401; + cvt.rn.bf16x2.f32 %r229, %r1404, %r1403; + // begin inline asm + @%p25 st.global.v4.b32 [ %rd57 + 0 ], { %r226, %r227, %r228, %r229 }; + // end inline asm + cvt.rn.bf16x2.f32 %r230, %r1406, %r1405; + cvt.rn.bf16x2.f32 %r231, %r1408, %r1407; + cvt.rn.bf16x2.f32 %r232, %r1410, %r1409; + cvt.rn.bf16x2.f32 %r233, %r1412, %r1411; + // begin inline asm + @%p26 st.global.v4.b32 [ %rd58 + 0 ], { %r230, %r231, %r232, %r233 }; + // end inline asm + cvt.rn.bf16x2.f32 %r234, %r1414, %r1413; + cvt.rn.bf16x2.f32 %r235, %r1416, %r1415; + cvt.rn.bf16x2.f32 %r236, %r1418, %r1417; + cvt.rn.bf16x2.f32 %r237, %r1420, %r1419; + // begin inline asm + @%p27 st.global.v4.b32 [ %rd59 + 0 ], { %r234, %r235, %r236, %r237 }; + // end inline asm + cvt.rn.bf16x2.f32 %r238, %r1422, %r1421; + cvt.rn.bf16x2.f32 %r239, %r1424, %r1423; + cvt.rn.bf16x2.f32 %r240, %r1426, %r1425; + cvt.rn.bf16x2.f32 %r241, %r1428, %r1427; + // begin inline asm + @%p28 st.global.v4.b32 [ %rd60 + 0 ], { %r238, %r239, %r240, %r241 }; + // end inline asm + cvt.rn.bf16x2.f32 %r242, %r1430, %r1429; + cvt.rn.bf16x2.f32 %r243, %r1432, %r1431; + cvt.rn.bf16x2.f32 %r244, %r1434, %r1433; + cvt.rn.bf16x2.f32 %r245, %r1436, %r1435; + // begin inline asm + @%p29 st.global.v4.b32 [ %rd61 + 0 ], { %r242, %r243, %r244, %r245 }; + // end inline asm + cvt.rn.bf16x2.f32 %r246, %r1438, %r1437; + cvt.rn.bf16x2.f32 %r247, %r1440, %r1439; + cvt.rn.bf16x2.f32 %r248, %r1442, %r1441; + cvt.rn.bf16x2.f32 %r249, %r1444, %r1443; + // begin inline asm + @%p30 st.global.v4.b32 [ %rd62 + 0 ], { %r246, %r247, %r248, %r249 }; + // end inline asm + cvt.rn.bf16x2.f32 %r250, %r1446, %r1445; + cvt.rn.bf16x2.f32 %r251, %r1448, %r1447; + cvt.rn.bf16x2.f32 %r252, %r1450, %r1449; + cvt.rn.bf16x2.f32 %r253, %r1452, %r1451; + // begin inline asm + @%p31 st.global.v4.b32 [ %rd63 + 0 ], { %r250, %r251, %r252, %r253 }; + // end inline asm + cvt.rn.bf16x2.f32 %r254, %r1454, %r1453; + cvt.rn.bf16x2.f32 %r255, %r1456, %r1455; + cvt.rn.bf16x2.f32 %r256, %r1458, %r1457; + cvt.rn.bf16x2.f32 %r257, %r1460, %r1459; + // begin inline asm + @%p32 st.global.v4.b32 [ %rd64 + 0 ], { %r254, %r255, %r256, %r257 }; + // end inline asm + .loc 1 84 21 // l2norm.py:84:21 + shl.b32 %r1461, %r303, 3; + mov.b32 %r1462, global_smem; + add.s32 %r258, %r1462, %r1461; + // begin inline asm + st.shared.v4.b32 [ %r258 + 0 ], { %r259, %r260, %r261, %r262 }; + // end inline asm + add.s32 %r263, %r258, 16; + // begin inline asm + st.shared.v4.b32 [ %r263 + 0 ], { %r264, %r265, %r266, %r267 }; + // end inline asm + add.s32 %r268, %r258, 32; + // begin inline asm + st.shared.v4.b32 [ %r268 + 0 ], { %r269, %r270, %r271, %r272 }; + // end inline asm + add.s32 %r273, %r258, 48; + // begin inline asm + st.shared.v4.b32 [ %r273 + 0 ], { %r274, %r275, %r276, %r277 }; + // end inline asm + add.s32 %r278, %r258, 64; + // begin inline asm + st.shared.v4.b32 [ %r278 + 0 ], { %r279, %r280, %r281, %r282 }; + // end inline asm + add.s32 %r283, %r258, 80; + // begin inline asm + st.shared.v4.b32 [ %r283 + 0 ], { %r284, %r285, %r286, %r287 }; + // end inline asm + add.s32 %r288, %r258, 96; + // begin inline asm + st.shared.v4.b32 [ %r288 + 0 ], { %r289, %r290, %r291, %r292 }; + // end inline asm + add.s32 %r293, %r258, 112; + // begin inline asm + st.shared.v4.b32 [ %r293 + 0 ], { %r294, %r295, %r296, %r297 }; + // end inline asm + bar.warp.sync -1; + shl.b32 %r1463, %r337, 4; + add.s32 %r1464, %r1462, %r1463; + ldmatrix.sync.aligned.m8n8.x2.shared.b16 {%r298, %r299}, [%r1464]; + // begin inline asm + @%p33 st.global.b32 [ %rd65 + 0 ], { %r298 }; + // end inline asm + // begin inline asm + @%p34 st.global.b32 [ %rd66 + 0 ], { %r299 }; + // end inline asm + .loc 1 84 4 // l2norm.py:84:4 + ret; +$L__tmp259: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp258 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 80 // DW_AT_call_line +.b8 43 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp258 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.source b/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..b5000c470f5e0683aea9e8045b2d0a8c059ea3d1 --- /dev/null +++ b/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.source @@ -0,0 +1,172 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc38 = loc(unknown) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc45 = loc("x"(#loc)) +#loc46 = loc("y"(#loc)) +#loc47 = loc("rstd"(#loc)) +#loc48 = loc("eps"(#loc)) +#loc49 = loc("T"(#loc)) +#loc81 = loc("input"(#loc36)) +#loc82 = loc("a"(#loc41)) +#loc83 = loc("b"(#loc41)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc50) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc51) + %o_t = arith.constant 64 : i32 loc(#loc52) + %o_t_1 = arith.constant 64 : i64 loc(#loc52) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc52) + %o_t_3 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc53) + %o_t_4 = arith.extsi %o_t_3 : tensor<64xi32> to tensor<64xi64> loc(#loc54) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<64xi64> loc(#loc54) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<64xi64> loc(#loc54) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc55) + %m_t = arith.extsi %T : i32 to i64 loc(#loc56) + %m_t_7 = tt.splat %m_t : i64 -> tensor<64xi64> loc(#loc56) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<64xi64> loc(#loc56) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc57) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc58) + %m_x_10 = arith.constant 128 : i32 loc(#loc59) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc59) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc59) + %m_x_13 = tt.broadcast %m_x : tensor<64x1xi1> -> tensor<64x128xi1> loc(#loc60) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<64x128xi1> loc(#loc60) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<64x128xi1> loc(#loc60) + %p_x = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc61) + %p_x_16 = arith.constant 128 : i32 loc(#loc62) + %p_x_17 = arith.constant 128 : i64 loc(#loc62) + %p_x_18 = arith.constant dense<128> : tensor<64x1xi64> loc(#loc62) + %p_x_19 = arith.muli %p_x, %p_x_18 : tensor<64x1xi64> loc(#loc62) + %p_x_20 = tt.splat %x : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc63) + %p_x_21 = tt.addptr %p_x_20, %p_x_19 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc63) + %p_x_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc64) + %p_x_23 = tt.broadcast %p_x_21 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc65) + %p_x_24 = tt.broadcast %p_x_22 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc65) + %p_x_25 = tt.addptr %p_x_23, %p_x_24 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc65) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc66) + %p_y_26 = arith.constant 128 : i32 loc(#loc67) + %p_y_27 = arith.constant 128 : i64 loc(#loc67) + %p_y_28 = arith.constant dense<128> : tensor<64x1xi64> loc(#loc67) + %p_y_29 = arith.muli %p_y, %p_y_28 : tensor<64x1xi64> loc(#loc67) + %p_y_30 = tt.splat %y : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc68) + %p_y_31 = tt.addptr %p_y_30, %p_y_29 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc68) + %p_y_32 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc69) + %p_y_33 = tt.broadcast %p_y_31 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc70) + %p_y_34 = tt.broadcast %p_y_32 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc70) + %p_y_35 = tt.addptr %p_y_33, %p_y_34 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc70) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<64x!tt.ptr> loc(#loc71) + %p_rstd_36 = tt.addptr %p_rstd, %o_t_6 : tensor<64x!tt.ptr>, tensor<64xi64> loc(#loc71) + %b_x = arith.constant 0.000000e+00 : f32 loc(#loc72) + %b_x_37 = arith.constant dense<0.000000e+00> : tensor<64x128xf32> loc(#loc72) + %b_x_38 = arith.truncf %b_x_37 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc72) + %b_x_39 = tt.load %p_x_25, %m_x_15, %b_x_38 : tensor<64x128x!tt.ptr> loc(#loc72) + %b_x_40 = arith.extf %b_x_39 : tensor<64x128xbf16> to tensor<64x128xf32> loc(#loc73) + %b_rstd = arith.mulf %b_x_40, %b_x_40 : tensor<64x128xf32> loc(#loc74) + %b_rstd_41 = tt.call @triton.language.standard.sum__fp32S64_128S_c1_cFalse_cNone(%b_rstd) : (tensor<64x128xf32>) -> tensor<64xf32> loc(#loc75) + %b_rstd_42 = tt.splat %eps : f32 -> tensor<64xf32> loc(#loc76) + %b_rstd_43 = arith.addf %b_rstd_41, %b_rstd_42 : tensor<64xf32> loc(#loc76) + %b_rstd_44 = math.sqrt %b_rstd_43 : tensor<64xf32> loc(#loc77) + %b_rstd_45 = arith.constant 1 : i32 loc(#loc78) + %b_rstd_46 = arith.constant 1.000000e+00 : f32 loc(#loc78) + %b_rstd_47 = arith.constant dense<1.000000e+00> : tensor<64xf32> loc(#loc78) + %b_rstd_48 = arith.divf %b_rstd_47, %b_rstd_44 : tensor<64xf32> loc(#loc78) + %b_y = tt.expand_dims %b_rstd_48 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc79) + %b_y_49 = tt.broadcast %b_y : tensor<64x1xf32> -> tensor<64x128xf32> loc(#loc80) + %b_y_50 = arith.mulf %b_x_40, %b_y_49 : tensor<64x128xf32> loc(#loc80) + %0 = arith.truncf %b_y_50 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc32) + tt.store %p_y_35, %0, %m_x_15 : tensor<64x128x!tt.ptr> loc(#loc33) + tt.store %p_rstd_36, %b_rstd_48, %m_t_8 : tensor<64x!tt.ptr> loc(#loc34) + tt.return loc(#loc35) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S64_128S_c1_cFalse_cNone(%input: tensor<64x128xf32> loc("input"(#loc36))) -> tensor<64xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc37) + tt.reduce.return %2 : f32 loc(#loc37) + }) : (tensor<64x128xf32>) -> tensor<64xf32> loc(#loc37) + tt.return %0 : tensor<64xf32> loc(#loc39) + ^bb1: // no predecessors + %1 = ub.poison : tensor<64xf32> loc(#loc40) + tt.return %1 : tensor<64xf32> loc(#loc40) + } loc(#loc36) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc41)), %b: f32 loc("b"(#loc41))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc42) + tt.return %0 : f32 loc(#loc43) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc44) + tt.return %1 : f32 loc(#loc44) + } loc(#loc41) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:18) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:29) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:37) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc50 = loc("i_t"(#loc1)) +#loc51 = loc("i_t"(#loc2)) +#loc52 = loc("o_t"(#loc3)) +#loc53 = loc("o_t"(#loc4)) +#loc54 = loc("o_t"(#loc5)) +#loc55 = loc("o_d"(#loc6)) +#loc56 = loc("m_t"(#loc7)) +#loc57 = loc("m_x"(#loc8)) +#loc58 = loc("m_x"(#loc9)) +#loc59 = loc("m_x"(#loc10)) +#loc60 = loc("m_x"(#loc11)) +#loc61 = loc("p_x"(#loc12)) +#loc62 = loc("p_x"(#loc13)) +#loc63 = loc("p_x"(#loc14)) +#loc64 = loc("p_x"(#loc15)) +#loc65 = loc("p_x"(#loc16)) +#loc66 = loc("p_y"(#loc17)) +#loc67 = loc("p_y"(#loc18)) +#loc68 = loc("p_y"(#loc19)) +#loc69 = loc("p_y"(#loc20)) +#loc70 = loc("p_y"(#loc21)) +#loc71 = loc("p_rstd"(#loc22)) +#loc72 = loc("b_x"(#loc23)) +#loc73 = loc("b_x"(#loc24)) +#loc74 = loc("b_rstd"(#loc25)) +#loc75 = loc("b_rstd"(#loc26)) +#loc76 = loc("b_rstd"(#loc27)) +#loc77 = loc("b_rstd"(#loc28)) +#loc78 = loc("b_rstd"(#loc29)) +#loc79 = loc("b_y"(#loc30)) +#loc80 = loc("b_y"(#loc31)) diff --git a/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.ttgir b/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..f27aa354afea36a6da1b03d28871bee46654b0dd --- /dev/null +++ b/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.ttgir @@ -0,0 +1,135 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [1, 1], order = [1, 0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [2], threadsPerWarp = [32], warpsPerCTA = [1], order = [0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc1 = loc(unknown) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc34 = loc("x"(#loc)) +#loc35 = loc("y"(#loc)) +#loc36 = loc("rstd"(#loc)) +#loc37 = loc("eps"(#loc)) +#loc38 = loc("T"(#loc)) +#loc59 = loc("b_rstd"(#loc23)) +#loc66 = loc(callsite(#loc1 at #loc59)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 1 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<1.000000e+00> : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<64x128xbf16, #blocked> loc(#loc1) + %c64_i64 = arith.constant 64 : i64 loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<64x1xi64, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc39) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc40) + %o_t = arith.muli %i_t_3, %c64_i64 : i64 loc(#loc41) + %o_t_4 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc42) + %o_t_5 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #blocked1> loc(#loc42) + %o_t_6 = arith.extsi %o_t_4 : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_7 = arith.extsi %o_t_5 : tensor<64xi32, #blocked1> to tensor<64xi64, #blocked1> loc(#loc43) + %o_t_8 = tt.splat %o_t : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_9 = tt.splat %o_t : i64 -> tensor<64xi64, #blocked1> loc(#loc43) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<64xi64, #blocked1> loc(#loc43) + %m_t = arith.extsi %T : i32 to i64 loc(#loc44) + %m_t_12 = tt.splat %m_t : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_13 = tt.splat %m_t : i64 -> tensor<64xi64, #blocked1> loc(#loc44) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<64xi64, #blocked1> loc(#loc44) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<64xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64x1xi1, #blocked> loc(#loc45) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc46) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc46) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_1 : tensor<1x128xi32, #blocked> loc(#loc47) + %m_x_19 = tt.broadcast %m_x : tensor<64x1xi1, #blocked> -> tensor<64x128xi1, #blocked> loc(#loc48) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<64x128xi1, #blocked> loc(#loc48) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<64x128xi1, #blocked> loc(#loc48) + %p_x = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64x1xi64, #blocked> loc(#loc49) + %p_x_22 = arith.muli %p_x, %cst_2 : tensor<64x1xi64, #blocked> loc(#loc50) + %p_x_23 = tt.splat %x : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked> loc(#loc51) + %p_x_24 = tt.addptr %p_x_23, %p_x_22 : tensor<64x1x!tt.ptr, #blocked>, tensor<64x1xi64, #blocked> loc(#loc51) + %p_x_25 = tt.broadcast %p_x_24 : tensor<64x1x!tt.ptr, #blocked> -> tensor<64x128x!tt.ptr, #blocked> loc(#loc52) + %p_x_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<64x128xi32, #blocked> loc(#loc52) + %p_x_27 = tt.addptr %p_x_25, %p_x_26 : tensor<64x128x!tt.ptr, #blocked>, tensor<64x128xi32, #blocked> loc(#loc52) + %p_y = tt.splat %y : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked> loc(#loc53) + %p_y_28 = tt.addptr %p_y, %p_x_22 : tensor<64x1x!tt.ptr, #blocked>, tensor<64x1xi64, #blocked> loc(#loc53) + %p_y_29 = tt.broadcast %p_y_28 : tensor<64x1x!tt.ptr, #blocked> -> tensor<64x128x!tt.ptr, #blocked> loc(#loc54) + %p_y_30 = tt.addptr %p_y_29, %p_x_26 : tensor<64x128x!tt.ptr, #blocked>, tensor<64x128xi32, #blocked> loc(#loc54) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<64x!tt.ptr, #blocked1> loc(#loc55) + %p_rstd_31 = tt.addptr %p_rstd, %o_t_11 : tensor<64x!tt.ptr, #blocked1>, tensor<64xi64, #blocked1> loc(#loc55) + %b_x = tt.load %p_x_27, %m_x_21, %cst_0 : tensor<64x128x!tt.ptr, #blocked> loc(#loc56) + %b_x_32 = arith.extf %b_x : tensor<64x128xbf16, #blocked> to tensor<64x128xf32, #blocked> loc(#loc57) + %b_rstd = arith.mulf %b_x_32, %b_x_32 : tensor<64x128xf32, #blocked> loc(#loc58) + %b_rstd_33 = "tt.reduce"(%b_rstd) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_40: f32 loc(callsite(#loc1 at #loc59)), %b_rstd_41: f32 loc(callsite(#loc1 at #loc59))): + %b_rstd_42 = arith.addf %b_rstd_40, %b_rstd_41 : f32 loc(#loc67) + tt.reduce.return %b_rstd_42 : f32 loc(#loc65) + }) : (tensor<64x128xf32, #blocked>) -> tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc65) + %b_rstd_34 = tt.splat %eps : f32 -> tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_36 = math.sqrt %b_rstd_35 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc61) + %b_rstd_37 = arith.divf %cst, %b_rstd_36 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc62) + %b_y = tt.expand_dims %b_rstd_37 {axis = 1 : i32} : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64x1xf32, #blocked> loc(#loc63) + %b_y_38 = tt.broadcast %b_y : tensor<64x1xf32, #blocked> -> tensor<64x128xf32, #blocked> loc(#loc64) + %b_y_39 = arith.mulf %b_x_32, %b_y_38 : tensor<64x128xf32, #blocked> loc(#loc64) + %0 = arith.truncf %b_y_39 : tensor<64x128xf32, #blocked> to tensor<64x128xbf16, #blocked> loc(#loc30) + tt.store %p_y_30, %0, %m_x_21 : tensor<64x128x!tt.ptr, #blocked> loc(#loc31) + %1 = ttg.convert_layout %b_rstd_37 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<64xf32, #blocked1> loc(#loc32) + tt.store %p_rstd_31, %1, %m_t_15 : tensor<64x!tt.ptr, #blocked1> loc(#loc32) + tt.return loc(#loc33) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc39 = loc("i_t"(#loc2)) +#loc40 = loc("i_t"(#loc3)) +#loc41 = loc("o_t"(#loc4)) +#loc42 = loc("o_t"(#loc5)) +#loc43 = loc("o_t"(#loc6)) +#loc44 = loc("m_t"(#loc7)) +#loc45 = loc("m_x"(#loc8)) +#loc46 = loc("m_x"(#loc9)) +#loc47 = loc("m_x"(#loc10)) +#loc48 = loc("m_x"(#loc11)) +#loc49 = loc("p_x"(#loc12)) +#loc50 = loc("p_x"(#loc13)) +#loc51 = loc("p_x"(#loc14)) +#loc52 = loc("p_x"(#loc15)) +#loc53 = loc("p_y"(#loc16)) +#loc54 = loc("p_y"(#loc17)) +#loc55 = loc("p_rstd"(#loc18)) +#loc56 = loc("b_x"(#loc19)) +#loc57 = loc("b_x"(#loc20)) +#loc58 = loc("b_rstd"(#loc21)) +#loc60 = loc("b_rstd"(#loc25)) +#loc61 = loc("b_rstd"(#loc26)) +#loc62 = loc("b_rstd"(#loc27)) +#loc63 = loc("b_y"(#loc28)) +#loc64 = loc("b_y"(#loc29)) +#loc65 = loc(callsite(#loc22 at #loc59)) +#loc67 = loc(callsite(#loc24 at #loc65)) diff --git a/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.ttir b/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..85ebb8b128ceed91f07ab82f5c0fe10a876f6335 --- /dev/null +++ b/triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.ttir @@ -0,0 +1,128 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc3 = loc(unknown) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc35 = loc("x"(#loc)) +#loc36 = loc("y"(#loc)) +#loc37 = loc("rstd"(#loc)) +#loc38 = loc("eps"(#loc)) +#loc39 = loc("T"(#loc)) +#loc62 = loc("b_rstd"(#loc25)) +#loc68 = loc(callsite(#loc3 at #loc62)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %b_x = arith.constant dense<0.000000e+00> : tensor<64x128xbf16> loc(#loc40) + %b_rstd = arith.constant dense<1.000000e+00> : tensor<64xf32> loc(#loc41) + %cst = arith.constant dense<128> : tensor<64x1xi64> loc(#loc3) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc42) + %o_t = arith.constant 64 : i64 loc(#loc43) + %i_t = tt.get_program_id x : i32 loc(#loc44) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc45) + %o_t_1 = arith.muli %i_t_0, %o_t : i64 loc(#loc43) + %o_t_2 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc46) + %o_t_3 = arith.extsi %o_t_2 : tensor<64xi32> to tensor<64xi64> loc(#loc47) + %o_t_4 = tt.splat %o_t_1 : i64 -> tensor<64xi64> loc(#loc47) + %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<64xi64> loc(#loc47) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc48) + %m_t = arith.extsi %T : i32 to i64 loc(#loc49) + %m_t_6 = tt.splat %m_t : i64 -> tensor<64xi64> loc(#loc49) + %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<64xi64> loc(#loc49) + %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc50) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc51) + %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc42) + %m_x_11 = tt.broadcast %m_x_8 : tensor<64x1xi1> -> tensor<64x128xi1> loc(#loc52) + %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<64x128xi1> loc(#loc52) + %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<64x128xi1> loc(#loc52) + %p_x = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc53) + %p_x_14 = arith.muli %p_x, %cst : tensor<64x1xi64> loc(#loc54) + %p_x_15 = tt.splat %x : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc55) + %p_x_16 = tt.addptr %p_x_15, %p_x_14 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc55) + %p_x_17 = tt.broadcast %p_x_16 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc56) + %p_x_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<64x128xi32> loc(#loc56) + %p_x_19 = tt.addptr %p_x_17, %p_x_18 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc56) + %p_y = tt.splat %y : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc57) + %p_y_20 = tt.addptr %p_y, %p_x_14 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc57) + %p_y_21 = tt.broadcast %p_y_20 : tensor<64x1x!tt.ptr> -> tensor<64x128x!tt.ptr> loc(#loc58) + %p_y_22 = tt.addptr %p_y_21, %p_x_18 : tensor<64x128x!tt.ptr>, tensor<64x128xi32> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<64x!tt.ptr> loc(#loc59) + %p_rstd_23 = tt.addptr %p_rstd, %o_t_5 : tensor<64x!tt.ptr>, tensor<64xi64> loc(#loc59) + %b_x_24 = tt.load %p_x_19, %m_x_13, %b_x : tensor<64x128x!tt.ptr> loc(#loc40) + %b_x_25 = arith.extf %b_x_24 : tensor<64x128xbf16> to tensor<64x128xf32> loc(#loc60) + %b_rstd_26 = arith.mulf %b_x_25, %b_x_25 : tensor<64x128xf32> loc(#loc61) + %b_rstd_27 = "tt.reduce"(%b_rstd_26) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_34: f32 loc(callsite(#loc3 at #loc62)), %b_rstd_35: f32 loc(callsite(#loc3 at #loc62))): + %b_rstd_36 = arith.addf %b_rstd_34, %b_rstd_35 : f32 loc(#loc69) + tt.reduce.return %b_rstd_36 : f32 loc(#loc67) + }) : (tensor<64x128xf32>) -> tensor<64xf32> loc(#loc67) + %b_rstd_28 = tt.splat %eps : f32 -> tensor<64xf32> loc(#loc63) + %b_rstd_29 = arith.addf %b_rstd_27, %b_rstd_28 : tensor<64xf32> loc(#loc63) + %b_rstd_30 = math.sqrt %b_rstd_29 : tensor<64xf32> loc(#loc64) + %b_rstd_31 = arith.divf %b_rstd, %b_rstd_30 : tensor<64xf32> loc(#loc41) + %b_y = tt.expand_dims %b_rstd_31 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc65) + %b_y_32 = tt.broadcast %b_y : tensor<64x1xf32> -> tensor<64x128xf32> loc(#loc66) + %b_y_33 = arith.mulf %b_x_25, %b_y_32 : tensor<64x128xf32> loc(#loc66) + %0 = arith.truncf %b_y_33 : tensor<64x128xf32> to tensor<64x128xbf16> loc(#loc31) + tt.store %p_y_22, %0, %m_x_13 : tensor<64x128x!tt.ptr> loc(#loc32) + tt.store %p_rstd_23, %b_rstd_31, %m_t_7 : tensor<64x!tt.ptr> loc(#loc33) + tt.return loc(#loc34) + } loc(#loc) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc40 = loc("b_x"(#loc1)) +#loc41 = loc("b_rstd"(#loc2)) +#loc42 = loc("m_x"(#loc4)) +#loc43 = loc("o_t"(#loc5)) +#loc44 = loc("i_t"(#loc6)) +#loc45 = loc("i_t"(#loc7)) +#loc46 = loc("o_t"(#loc8)) +#loc47 = loc("o_t"(#loc9)) +#loc48 = loc("o_d"(#loc10)) +#loc49 = loc("m_t"(#loc11)) +#loc50 = loc("m_x"(#loc12)) +#loc51 = loc("m_x"(#loc13)) +#loc52 = loc("m_x"(#loc14)) +#loc53 = loc("p_x"(#loc15)) +#loc54 = loc("p_x"(#loc16)) +#loc55 = loc("p_x"(#loc17)) +#loc56 = loc("p_x"(#loc18)) +#loc57 = loc("p_y"(#loc19)) +#loc58 = loc("p_y"(#loc20)) +#loc59 = loc("p_rstd"(#loc21)) +#loc60 = loc("b_x"(#loc22)) +#loc61 = loc("b_rstd"(#loc23)) +#loc63 = loc("b_rstd"(#loc27)) +#loc64 = loc("b_rstd"(#loc28)) +#loc65 = loc("b_y"(#loc29)) +#loc66 = loc("b_y"(#loc30)) +#loc67 = loc(callsite(#loc24 at #loc62)) +#loc69 = loc(callsite(#loc26 at #loc67)) diff --git a/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/__grp__chunk_gated_delta_rule_fwd_kernel_h_blockdim64.json b/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/__grp__chunk_gated_delta_rule_fwd_kernel_h_blockdim64.json new file mode 100644 index 0000000000000000000000000000000000000000..29c5288979d91fe4cbe84e469bfe3101775897cd --- /dev/null +++ b/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/__grp__chunk_gated_delta_rule_fwd_kernel_h_blockdim64.json @@ -0,0 +1 @@ +{"child_paths": {"chunk_gated_delta_rule_fwd_kernel_h_blockdim64.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.source", "chunk_gated_delta_rule_fwd_kernel_h_blockdim64.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.ttir", "chunk_gated_delta_rule_fwd_kernel_h_blockdim64.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.ttgir", "chunk_gated_delta_rule_fwd_kernel_h_blockdim64.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.llir", "chunk_gated_delta_rule_fwd_kernel_h_blockdim64.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.ptx", "chunk_gated_delta_rule_fwd_kernel_h_blockdim64.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.cubin", "chunk_gated_delta_rule_fwd_kernel_h_blockdim64.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.json"}} \ No newline at end of file diff --git a/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.cubin b/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.cubin new file mode 100644 index 0000000000000000000000000000000000000000..125b016ce6f6a6fd0da18b80c136319891ed398d --- /dev/null +++ b/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.cubin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d8b4468f684acc92e8b3770dd79f6af8dbab397a384b45b414266af09cbc0854 +size 227952 diff --git a/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.json b/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.json new file mode 100644 index 0000000000000000000000000000000000000000..9f9ec6112c319d1e157d24520045fa339158670a --- /dev/null +++ b/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.json @@ -0,0 +1 @@ +{"hash": "130e9177f56f18459ed09dea93dc2c4916b2752a621a972694113556926fa37f", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 2, "num_ctas": 1, "num_stages": 4, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 142092, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "chunk_gated_delta_rule_fwd_kernel_h_blockdim64"} \ No newline at end of file diff --git a/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.llir b/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.llir new file mode 100644 index 0000000000000000000000000000000000000000..ce69f485ebbd079041740aab53de439b4b65667f --- /dev/null +++ b/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.llir @@ -0,0 +1,4480 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @chunk_gated_delta_rule_fwd_kernel_h_blockdim64(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, ptr addrspace(1) %3, ptr addrspace(1) %4, ptr addrspace(1) %5, i32 %6, ptr addrspace(1) readnone captures(none) %7, ptr addrspace(1) readnone captures(none) %8) local_unnamed_addr #0 !dbg !5 { + %10 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %11 = zext nneg i32 %10 to i64, !dbg !14 + %12 = lshr i64 %11, 1, !dbg !15 + %13 = lshr i64 %11, 6, !dbg !16 + %14 = and i64 %12, 31, !dbg !17 + %15 = sext i32 %6 to i64, !dbg !18 + %16 = add i32 %6, 63, !dbg !19 + %17 = shl nuw nsw i64 %13, 5, !dbg !18 + %18 = mul nsw i64 %17, %15, !dbg !24 + %19 = or disjoint i64 %18, %14, !dbg !25 + %20 = shl i64 %19, 7, !dbg !26 + %21 = getelementptr bfloat, ptr addrspace(1) %1, i64 %20, !dbg !27 + %22 = getelementptr bfloat, ptr addrspace(1) %0, i64 %20, !dbg !28 + %23 = getelementptr bfloat, ptr addrspace(1) %2, i64 %20, !dbg !29 + %24 = getelementptr bfloat, ptr addrspace(1) %3, i64 %20, !dbg !30 + %25 = shl nuw nsw i64 %11, 6, !dbg !31 + %26 = and i64 %25, 64, !dbg !31 + %27 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !32 + %28 = and i32 %27, 7, !dbg !32 + %29 = shl nuw nsw i32 %28, 3, !dbg !32 + %30 = and i32 %27, 56, !dbg !32 + %31 = lshr exact i32 %30, 3, !dbg !32 + %32 = or disjoint i32 %31, 8, !dbg !32 + %33 = or disjoint i32 %31, 16, !dbg !32 + %34 = or disjoint i32 %31, 24, !dbg !32 + %35 = or disjoint i32 %31, 32, !dbg !32 + %36 = or disjoint i32 %31, 40, !dbg !32 + %37 = or disjoint i32 %31, 48, !dbg !32 + %38 = or disjoint i32 %31, 56, !dbg !32 + %39 = and i32 %27, 28, !dbg !32 + %40 = lshr exact i32 %39, 2, !dbg !32 + %41 = and i32 %27, 32, !dbg !32 + %42 = lshr exact i32 %41, 1, !dbg !32 + %43 = or disjoint i32 %42, %40, !dbg !32 + %44 = or disjoint i32 %43, 8, !dbg !32 + %45 = or disjoint i32 %43, 32, !dbg !32 + %46 = or disjoint i32 %43, 40, !dbg !32 + %47 = and i32 %27, 63, !dbg !32 + %48 = zext nneg i32 %29 to i64, !dbg !33 + %49 = zext nneg i32 %31 to i64, !dbg !33 + %50 = zext nneg i32 %32 to i64, !dbg !33 + %51 = zext nneg i32 %33 to i64, !dbg !33 + %52 = zext nneg i32 %34 to i64, !dbg !33 + %53 = zext nneg i32 %35 to i64, !dbg !33 + %54 = zext nneg i32 %36 to i64, !dbg !33 + %55 = zext nneg i32 %37 to i64, !dbg !33 + %56 = zext nneg i32 %38 to i64, !dbg !33 + %57 = insertelement <4 x i32> poison, i32 %46, i64 0, !dbg !33 + %58 = insertelement <4 x i32> %57, i32 %45, i64 1, !dbg !33 + %59 = insertelement <4 x i32> %58, i32 %44, i64 2, !dbg !33 + %60 = insertelement <4 x i32> %59, i32 %43, i64 3, !dbg !33 + %61 = zext <4 x i32> %60 to <4 x i64>, !dbg !33 + %62 = zext nneg i32 %47 to i64, !dbg !33 + %63 = or disjoint i64 %26, %48, !dbg !33 + %64 = or disjoint i32 %29, 64, !dbg !34 + %65 = getelementptr float, ptr addrspace(1) %4, i64 %19, !dbg !35 + %66 = getelementptr bfloat, ptr addrspace(1) %22, i64 %48, !dbg !36 + %67 = zext nneg i32 %64 to i64, !dbg !37 + %68 = getelementptr bfloat, ptr addrspace(1) %22, i64 %67, !dbg !37 + %69 = icmp sgt i32 %16, 63, !dbg !38 + %70 = icmp slt i32 %31, %6, !dbg !39 + %71 = icmp slt i32 %32, %6, !dbg !39 + %72 = icmp slt i32 %33, %6, !dbg !39 + %73 = icmp slt i32 %34, %6, !dbg !39 + %74 = icmp slt i32 %35, %6, !dbg !39 + %75 = icmp slt i32 %36, %6, !dbg !39 + %76 = icmp slt i32 %37, %6, !dbg !39 + %77 = icmp slt i32 %38, %6, !dbg !39 + %78 = icmp slt i32 %47, %6, !dbg !39 + %79 = shl nuw nsw i64 %49, 12, !dbg !40 + %80 = shl nuw nsw i64 %50, 12, !dbg !40 + %81 = shl nuw nsw i64 %51, 12, !dbg !40 + %82 = shl nuw nsw i64 %52, 12, !dbg !40 + %83 = shl nuw nsw i64 %53, 12, !dbg !40 + %84 = shl nuw nsw i64 %54, 12, !dbg !40 + %85 = shl nuw nsw i64 %55, 12, !dbg !40 + %86 = shl nuw nsw i64 %56, 12, !dbg !40 + %87 = getelementptr bfloat, ptr addrspace(1) %23, i64 %79, !dbg !41 + %88 = getelementptr bfloat, ptr addrspace(1) %23, i64 %80, !dbg !41 + %89 = getelementptr bfloat, ptr addrspace(1) %23, i64 %81, !dbg !41 + %90 = getelementptr bfloat, ptr addrspace(1) %23, i64 %82, !dbg !41 + %91 = getelementptr bfloat, ptr addrspace(1) %23, i64 %83, !dbg !41 + %92 = getelementptr bfloat, ptr addrspace(1) %23, i64 %84, !dbg !41 + %93 = getelementptr bfloat, ptr addrspace(1) %23, i64 %85, !dbg !41 + %94 = getelementptr bfloat, ptr addrspace(1) %23, i64 %86, !dbg !41 + %95 = getelementptr bfloat, ptr addrspace(1) %87, i64 %48, !dbg !42 + %96 = getelementptr bfloat, ptr addrspace(1) %88, i64 %48, !dbg !42 + %97 = getelementptr bfloat, ptr addrspace(1) %89, i64 %48, !dbg !42 + %98 = getelementptr bfloat, ptr addrspace(1) %90, i64 %48, !dbg !42 + %99 = getelementptr bfloat, ptr addrspace(1) %91, i64 %48, !dbg !42 + %100 = getelementptr bfloat, ptr addrspace(1) %92, i64 %48, !dbg !42 + %101 = getelementptr bfloat, ptr addrspace(1) %93, i64 %48, !dbg !42 + %102 = getelementptr bfloat, ptr addrspace(1) %94, i64 %48, !dbg !42 + %103 = and i1 %70, %69, !dbg !38 + %104 = and i1 %71, %69, !dbg !38 + %105 = and i1 %72, %69, !dbg !38 + %106 = and i1 %73, %69, !dbg !38 + %107 = and i1 %74, %69, !dbg !38 + %108 = and i1 %75, %69, !dbg !38 + %109 = and i1 %76, %69, !dbg !38 + %110 = and i1 %77, %69, !dbg !38 + %111 = shl nuw nsw i32 %47, 4, !dbg !43 + %112 = shl nuw nsw i32 %30, 1, !dbg !43 + %113 = xor i32 %111, %112, !dbg !43 + %114 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %113, !dbg !43 + %115 = select i1 %103, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %114, ptr addrspace(1) %95, i32 %115) #2, !dbg !43 + %116 = or disjoint i32 %113, 1024, !dbg !43 + %117 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %116, !dbg !43 + %118 = select i1 %104, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %117, ptr addrspace(1) %96, i32 %118) #2, !dbg !43 + %119 = or disjoint i32 %113, 2048, !dbg !43 + %120 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %119, !dbg !43 + %121 = select i1 %105, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %120, ptr addrspace(1) %97, i32 %121) #2, !dbg !43 + %122 = or disjoint i32 %113, 3072, !dbg !43 + %123 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %122, !dbg !43 + %124 = select i1 %106, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %123, ptr addrspace(1) %98, i32 %124) #2, !dbg !43 + %125 = or disjoint i32 %113, 4096, !dbg !43 + %126 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %125, !dbg !43 + %127 = select i1 %107, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %126, ptr addrspace(1) %99, i32 %127) #2, !dbg !43 + %128 = or disjoint i32 %113, 5120, !dbg !43 + %129 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %128, !dbg !43 + %130 = select i1 %108, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %129, ptr addrspace(1) %100, i32 %130) #2, !dbg !43 + %131 = or disjoint i32 %113, 6144, !dbg !43 + %132 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %131, !dbg !43 + %133 = select i1 %109, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %132, ptr addrspace(1) %101, i32 %133) #2, !dbg !43 + %134 = or disjoint i32 %113, 7168, !dbg !43 + %135 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %134, !dbg !43 + %136 = select i1 %110, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %135, ptr addrspace(1) %102, i32 %136) #2, !dbg !43 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !43 + %137 = getelementptr bfloat, ptr addrspace(1) %87, i64 %67, !dbg !44 + %138 = getelementptr bfloat, ptr addrspace(1) %88, i64 %67, !dbg !44 + %139 = getelementptr bfloat, ptr addrspace(1) %89, i64 %67, !dbg !44 + %140 = getelementptr bfloat, ptr addrspace(1) %90, i64 %67, !dbg !44 + %141 = getelementptr bfloat, ptr addrspace(1) %91, i64 %67, !dbg !44 + %142 = getelementptr bfloat, ptr addrspace(1) %92, i64 %67, !dbg !44 + %143 = getelementptr bfloat, ptr addrspace(1) %93, i64 %67, !dbg !44 + %144 = getelementptr bfloat, ptr addrspace(1) %94, i64 %67, !dbg !44 + %145 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 24576), i32 %113, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %145, ptr addrspace(1) %137, i32 %115) #2, !dbg !45 + %146 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 24576), i32 %116, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %146, ptr addrspace(1) %138, i32 %118) #2, !dbg !45 + %147 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 24576), i32 %119, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %147, ptr addrspace(1) %139, i32 %121) #2, !dbg !45 + %148 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 24576), i32 %122, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %148, ptr addrspace(1) %140, i32 %124) #2, !dbg !45 + %149 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 24576), i32 %125, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %149, ptr addrspace(1) %141, i32 %127) #2, !dbg !45 + %150 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 24576), i32 %128, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %150, ptr addrspace(1) %142, i32 %130) #2, !dbg !45 + %151 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 24576), i32 %131, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %151, ptr addrspace(1) %143, i32 %133) #2, !dbg !45 + %152 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 24576), i32 %134, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %152, ptr addrspace(1) %144, i32 %136) #2, !dbg !45 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !45 + %153 = getelementptr bfloat, ptr addrspace(1) %21, i64 %79, !dbg !46 + %154 = getelementptr bfloat, ptr addrspace(1) %21, i64 %80, !dbg !46 + %155 = getelementptr bfloat, ptr addrspace(1) %21, i64 %81, !dbg !46 + %156 = getelementptr bfloat, ptr addrspace(1) %21, i64 %82, !dbg !46 + %157 = getelementptr bfloat, ptr addrspace(1) %21, i64 %83, !dbg !46 + %158 = getelementptr bfloat, ptr addrspace(1) %21, i64 %84, !dbg !46 + %159 = getelementptr bfloat, ptr addrspace(1) %21, i64 %85, !dbg !46 + %160 = getelementptr bfloat, ptr addrspace(1) %21, i64 %86, !dbg !46 + %161 = getelementptr bfloat, ptr addrspace(1) %153, i64 %63, !dbg !47 + %162 = getelementptr bfloat, ptr addrspace(1) %154, i64 %63, !dbg !47 + %163 = getelementptr bfloat, ptr addrspace(1) %155, i64 %63, !dbg !47 + %164 = getelementptr bfloat, ptr addrspace(1) %156, i64 %63, !dbg !47 + %165 = getelementptr bfloat, ptr addrspace(1) %157, i64 %63, !dbg !47 + %166 = getelementptr bfloat, ptr addrspace(1) %158, i64 %63, !dbg !47 + %167 = getelementptr bfloat, ptr addrspace(1) %159, i64 %63, !dbg !47 + %168 = getelementptr bfloat, ptr addrspace(1) %160, i64 %63, !dbg !47 + %169 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 49152), i32 %113, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %169, ptr addrspace(1) %161, i32 %115) #2, !dbg !48 + %170 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 49152), i32 %116, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %170, ptr addrspace(1) %162, i32 %118) #2, !dbg !48 + %171 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 49152), i32 %119, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %171, ptr addrspace(1) %163, i32 %121) #2, !dbg !48 + %172 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 49152), i32 %122, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %172, ptr addrspace(1) %164, i32 %124) #2, !dbg !48 + %173 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 49152), i32 %125, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %173, ptr addrspace(1) %165, i32 %127) #2, !dbg !48 + %174 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 49152), i32 %128, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %174, ptr addrspace(1) %166, i32 %130) #2, !dbg !48 + %175 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 49152), i32 %131, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %175, ptr addrspace(1) %167, i32 %133) #2, !dbg !48 + %176 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 49152), i32 %134, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %176, ptr addrspace(1) %168, i32 %136) #2, !dbg !48 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !48 + %177 = tail call i32 @llvm.smin.i32(i32 %6, i32 64), !dbg !49 + %178 = shl i32 %177, 5, !dbg !50 + %179 = add i32 %178, -32, !dbg !50 + %180 = sext i32 %179 to i64, !dbg !51 + %181 = getelementptr float, ptr addrspace(1) %4, i64 %18, !dbg !52 + %182 = getelementptr float, ptr addrspace(1) %181, i64 %180, !dbg !52 + %183 = getelementptr float, ptr addrspace(1) %182, i64 %14, !dbg !52 + %184 = icmp eq i32 %47, 0, !dbg !53 + %185 = select i1 %69, i32 4, i32 0, !dbg !53 + tail call void asm sideeffect "@$3 cp.async.ca.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x4, $2;", "r,l,r,b"(ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 142080), ptr addrspace(1) %183, i32 %185, i1 %184) #2, !dbg !53 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !53 + %.idx = shl nuw nsw i64 %62, 7, !dbg !54 + %186 = getelementptr i8, ptr addrspace(1) %65, i64 %.idx, !dbg !54 + %187 = and i1 %78, %69, !dbg !38 + %188 = shl nuw nsw i32 %47, 2, !dbg !55 + %189 = getelementptr i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 141312), i32 %188, !dbg !55 + %190 = select i1 %187, i32 4, i32 0, !dbg !55 + tail call void asm sideeffect "cp.async.ca.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x4, $2;", "r,l,r"(ptr addrspace(3) %189, ptr addrspace(1) %186, i32 %190) #2, !dbg !55 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !55 + %191 = getelementptr bfloat, ptr addrspace(1) %66, i64 %79, !dbg !56 + %192 = getelementptr bfloat, ptr addrspace(1) %66, i64 %80, !dbg !56 + %193 = getelementptr bfloat, ptr addrspace(1) %66, i64 %81, !dbg !56 + %194 = getelementptr bfloat, ptr addrspace(1) %66, i64 %82, !dbg !56 + %195 = getelementptr bfloat, ptr addrspace(1) %66, i64 %83, !dbg !56 + %196 = getelementptr bfloat, ptr addrspace(1) %66, i64 %84, !dbg !56 + %197 = getelementptr bfloat, ptr addrspace(1) %66, i64 %85, !dbg !56 + %198 = getelementptr bfloat, ptr addrspace(1) %66, i64 %86, !dbg !56 + %199 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73728), i32 %113, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %199, ptr addrspace(1) %191, i32 %115) #2, !dbg !57 + %200 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73728), i32 %116, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %200, ptr addrspace(1) %192, i32 %118) #2, !dbg !57 + %201 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73728), i32 %119, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %201, ptr addrspace(1) %193, i32 %121) #2, !dbg !57 + %202 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73728), i32 %122, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %202, ptr addrspace(1) %194, i32 %124) #2, !dbg !57 + %203 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73728), i32 %125, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %203, ptr addrspace(1) %195, i32 %127) #2, !dbg !57 + %204 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73728), i32 %128, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %204, ptr addrspace(1) %196, i32 %130) #2, !dbg !57 + %205 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73728), i32 %131, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %205, ptr addrspace(1) %197, i32 %133) #2, !dbg !57 + %206 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73728), i32 %134, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %206, ptr addrspace(1) %198, i32 %136) #2, !dbg !57 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !57 + %207 = getelementptr bfloat, ptr addrspace(1) %68, i64 %79, !dbg !58 + %208 = getelementptr bfloat, ptr addrspace(1) %68, i64 %80, !dbg !58 + %209 = getelementptr bfloat, ptr addrspace(1) %68, i64 %81, !dbg !58 + %210 = getelementptr bfloat, ptr addrspace(1) %68, i64 %82, !dbg !58 + %211 = getelementptr bfloat, ptr addrspace(1) %68, i64 %83, !dbg !58 + %212 = getelementptr bfloat, ptr addrspace(1) %68, i64 %84, !dbg !58 + %213 = getelementptr bfloat, ptr addrspace(1) %68, i64 %85, !dbg !58 + %214 = getelementptr bfloat, ptr addrspace(1) %68, i64 %86, !dbg !58 + %215 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 98304), i32 %113, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %215, ptr addrspace(1) %207, i32 %115) #2, !dbg !59 + %216 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 98304), i32 %116, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %216, ptr addrspace(1) %208, i32 %118) #2, !dbg !59 + %217 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 98304), i32 %119, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %217, ptr addrspace(1) %209, i32 %121) #2, !dbg !59 + %218 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 98304), i32 %122, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %218, ptr addrspace(1) %210, i32 %124) #2, !dbg !59 + %219 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 98304), i32 %125, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %219, ptr addrspace(1) %211, i32 %127) #2, !dbg !59 + %220 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 98304), i32 %128, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %220, ptr addrspace(1) %212, i32 %130) #2, !dbg !59 + %221 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 98304), i32 %131, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %221, ptr addrspace(1) %213, i32 %133) #2, !dbg !59 + %222 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 98304), i32 %134, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %222, ptr addrspace(1) %214, i32 %136) #2, !dbg !59 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !59 + %223 = icmp sgt i32 %16, 127, !dbg !38 + %224 = or disjoint i64 %49, 64, !dbg !60 + %225 = or disjoint i64 %50, 64, !dbg !60 + %226 = or disjoint i64 %51, 64, !dbg !60 + %227 = or disjoint i64 %52, 64, !dbg !60 + %228 = or disjoint i64 %53, 64, !dbg !60 + %229 = or disjoint i64 %54, 64, !dbg !60 + %230 = or disjoint i64 %55, 64, !dbg !60 + %231 = or disjoint i64 %56, 64, !dbg !60 + %232 = or disjoint i64 %62, 64, !dbg !60 + %233 = icmp slt i64 %224, %15, !dbg !39 + %234 = icmp slt i64 %225, %15, !dbg !39 + %235 = icmp slt i64 %226, %15, !dbg !39 + %236 = icmp slt i64 %227, %15, !dbg !39 + %237 = icmp slt i64 %228, %15, !dbg !39 + %238 = icmp slt i64 %229, %15, !dbg !39 + %239 = icmp slt i64 %230, %15, !dbg !39 + %240 = icmp slt i64 %231, %15, !dbg !39 + %241 = icmp slt i64 %232, %15, !dbg !39 + %242 = shl nuw nsw i64 %224, 12, !dbg !40 + %243 = shl nuw nsw i64 %225, 12, !dbg !40 + %244 = shl nuw nsw i64 %226, 12, !dbg !40 + %245 = shl nuw nsw i64 %227, 12, !dbg !40 + %246 = shl nuw nsw i64 %228, 12, !dbg !40 + %247 = shl nuw nsw i64 %229, 12, !dbg !40 + %248 = shl nuw nsw i64 %230, 12, !dbg !40 + %249 = shl nuw nsw i64 %231, 12, !dbg !40 + %250 = getelementptr bfloat, ptr addrspace(1) %23, i64 %242, !dbg !41 + %251 = getelementptr bfloat, ptr addrspace(1) %23, i64 %243, !dbg !41 + %252 = getelementptr bfloat, ptr addrspace(1) %23, i64 %244, !dbg !41 + %253 = getelementptr bfloat, ptr addrspace(1) %23, i64 %245, !dbg !41 + %254 = getelementptr bfloat, ptr addrspace(1) %23, i64 %246, !dbg !41 + %255 = getelementptr bfloat, ptr addrspace(1) %23, i64 %247, !dbg !41 + %256 = getelementptr bfloat, ptr addrspace(1) %23, i64 %248, !dbg !41 + %257 = getelementptr bfloat, ptr addrspace(1) %23, i64 %249, !dbg !41 + %258 = getelementptr bfloat, ptr addrspace(1) %250, i64 %48, !dbg !42 + %259 = getelementptr bfloat, ptr addrspace(1) %251, i64 %48, !dbg !42 + %260 = getelementptr bfloat, ptr addrspace(1) %252, i64 %48, !dbg !42 + %261 = getelementptr bfloat, ptr addrspace(1) %253, i64 %48, !dbg !42 + %262 = getelementptr bfloat, ptr addrspace(1) %254, i64 %48, !dbg !42 + %263 = getelementptr bfloat, ptr addrspace(1) %255, i64 %48, !dbg !42 + %264 = getelementptr bfloat, ptr addrspace(1) %256, i64 %48, !dbg !42 + %265 = getelementptr bfloat, ptr addrspace(1) %257, i64 %48, !dbg !42 + %266 = and i1 %223, %233, !dbg !38 + %267 = and i1 %223, %234, !dbg !38 + %268 = and i1 %223, %235, !dbg !38 + %269 = and i1 %223, %236, !dbg !38 + %270 = and i1 %223, %237, !dbg !38 + %271 = and i1 %223, %238, !dbg !38 + %272 = and i1 %223, %239, !dbg !38 + %273 = and i1 %223, %240, !dbg !38 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !43 + %274 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 8192), i32 %113, !dbg !43 + %275 = select i1 %266, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %274, ptr addrspace(1) %258, i32 %275) #2, !dbg !43 + %276 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 8192), i32 %116, !dbg !43 + %277 = select i1 %267, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %276, ptr addrspace(1) %259, i32 %277) #2, !dbg !43 + %278 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 8192), i32 %119, !dbg !43 + %279 = select i1 %268, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %278, ptr addrspace(1) %260, i32 %279) #2, !dbg !43 + %280 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 8192), i32 %122, !dbg !43 + %281 = select i1 %269, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %280, ptr addrspace(1) %261, i32 %281) #2, !dbg !43 + %282 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 8192), i32 %125, !dbg !43 + %283 = select i1 %270, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %282, ptr addrspace(1) %262, i32 %283) #2, !dbg !43 + %284 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 8192), i32 %128, !dbg !43 + %285 = select i1 %271, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %284, ptr addrspace(1) %263, i32 %285) #2, !dbg !43 + %286 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 8192), i32 %131, !dbg !43 + %287 = select i1 %272, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %286, ptr addrspace(1) %264, i32 %287) #2, !dbg !43 + %288 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 8192), i32 %134, !dbg !43 + %289 = select i1 %273, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %288, ptr addrspace(1) %265, i32 %289) #2, !dbg !43 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !43 + %290 = getelementptr bfloat, ptr addrspace(1) %250, i64 %67, !dbg !44 + %291 = getelementptr bfloat, ptr addrspace(1) %251, i64 %67, !dbg !44 + %292 = getelementptr bfloat, ptr addrspace(1) %252, i64 %67, !dbg !44 + %293 = getelementptr bfloat, ptr addrspace(1) %253, i64 %67, !dbg !44 + %294 = getelementptr bfloat, ptr addrspace(1) %254, i64 %67, !dbg !44 + %295 = getelementptr bfloat, ptr addrspace(1) %255, i64 %67, !dbg !44 + %296 = getelementptr bfloat, ptr addrspace(1) %256, i64 %67, !dbg !44 + %297 = getelementptr bfloat, ptr addrspace(1) %257, i64 %67, !dbg !44 + %298 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 32768), i32 %113, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %298, ptr addrspace(1) %290, i32 %275) #2, !dbg !45 + %299 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 32768), i32 %116, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %299, ptr addrspace(1) %291, i32 %277) #2, !dbg !45 + %300 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 32768), i32 %119, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %300, ptr addrspace(1) %292, i32 %279) #2, !dbg !45 + %301 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 32768), i32 %122, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %301, ptr addrspace(1) %293, i32 %281) #2, !dbg !45 + %302 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 32768), i32 %125, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %302, ptr addrspace(1) %294, i32 %283) #2, !dbg !45 + %303 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 32768), i32 %128, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %303, ptr addrspace(1) %295, i32 %285) #2, !dbg !45 + %304 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 32768), i32 %131, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %304, ptr addrspace(1) %296, i32 %287) #2, !dbg !45 + %305 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 32768), i32 %134, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %305, ptr addrspace(1) %297, i32 %289) #2, !dbg !45 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !45 + %306 = getelementptr bfloat, ptr addrspace(1) %21, i64 %242, !dbg !46 + %307 = getelementptr bfloat, ptr addrspace(1) %21, i64 %243, !dbg !46 + %308 = getelementptr bfloat, ptr addrspace(1) %21, i64 %244, !dbg !46 + %309 = getelementptr bfloat, ptr addrspace(1) %21, i64 %245, !dbg !46 + %310 = getelementptr bfloat, ptr addrspace(1) %21, i64 %246, !dbg !46 + %311 = getelementptr bfloat, ptr addrspace(1) %21, i64 %247, !dbg !46 + %312 = getelementptr bfloat, ptr addrspace(1) %21, i64 %248, !dbg !46 + %313 = getelementptr bfloat, ptr addrspace(1) %21, i64 %249, !dbg !46 + %314 = getelementptr bfloat, ptr addrspace(1) %306, i64 %63, !dbg !47 + %315 = getelementptr bfloat, ptr addrspace(1) %307, i64 %63, !dbg !47 + %316 = getelementptr bfloat, ptr addrspace(1) %308, i64 %63, !dbg !47 + %317 = getelementptr bfloat, ptr addrspace(1) %309, i64 %63, !dbg !47 + %318 = getelementptr bfloat, ptr addrspace(1) %310, i64 %63, !dbg !47 + %319 = getelementptr bfloat, ptr addrspace(1) %311, i64 %63, !dbg !47 + %320 = getelementptr bfloat, ptr addrspace(1) %312, i64 %63, !dbg !47 + %321 = getelementptr bfloat, ptr addrspace(1) %313, i64 %63, !dbg !47 + %322 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 57344), i32 %113, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %322, ptr addrspace(1) %314, i32 %275) #2, !dbg !48 + %323 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 57344), i32 %116, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %323, ptr addrspace(1) %315, i32 %277) #2, !dbg !48 + %324 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 57344), i32 %119, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %324, ptr addrspace(1) %316, i32 %279) #2, !dbg !48 + %325 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 57344), i32 %122, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %325, ptr addrspace(1) %317, i32 %281) #2, !dbg !48 + %326 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 57344), i32 %125, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %326, ptr addrspace(1) %318, i32 %283) #2, !dbg !48 + %327 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 57344), i32 %128, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %327, ptr addrspace(1) %319, i32 %285) #2, !dbg !48 + %328 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 57344), i32 %131, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %328, ptr addrspace(1) %320, i32 %287) #2, !dbg !48 + %329 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 57344), i32 %134, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %329, ptr addrspace(1) %321, i32 %289) #2, !dbg !48 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !48 + %330 = tail call i32 @llvm.smin.i32(i32 %6, i32 128), !dbg !49 + %331 = shl i32 %330, 5, !dbg !50 + %332 = add i32 %331, -32, !dbg !50 + %333 = sext i32 %332 to i64, !dbg !51 + %334 = getelementptr float, ptr addrspace(1) %181, i64 %333, !dbg !52 + %335 = getelementptr float, ptr addrspace(1) %334, i64 %14, !dbg !52 + %336 = select i1 %223, i32 4, i32 0, !dbg !53 + tail call void asm sideeffect "@$3 cp.async.ca.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x4, $2;", "r,l,r,b"(ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 142084), ptr addrspace(1) %335, i32 %336, i1 %184) #2, !dbg !53 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !53 + %.idx2 = shl nuw nsw i64 %232, 7, !dbg !54 + %337 = getelementptr i8, ptr addrspace(1) %65, i64 %.idx2, !dbg !54 + %338 = and i1 %223, %241, !dbg !38 + %339 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 141568), i32 %188, !dbg !55 + %340 = select i1 %338, i32 4, i32 0, !dbg !55 + tail call void asm sideeffect "cp.async.ca.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x4, $2;", "r,l,r"(ptr addrspace(3) %339, ptr addrspace(1) %337, i32 %340) #2, !dbg !55 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !55 + %341 = getelementptr bfloat, ptr addrspace(1) %66, i64 %242, !dbg !56 + %342 = getelementptr bfloat, ptr addrspace(1) %66, i64 %243, !dbg !56 + %343 = getelementptr bfloat, ptr addrspace(1) %66, i64 %244, !dbg !56 + %344 = getelementptr bfloat, ptr addrspace(1) %66, i64 %245, !dbg !56 + %345 = getelementptr bfloat, ptr addrspace(1) %66, i64 %246, !dbg !56 + %346 = getelementptr bfloat, ptr addrspace(1) %66, i64 %247, !dbg !56 + %347 = getelementptr bfloat, ptr addrspace(1) %66, i64 %248, !dbg !56 + %348 = getelementptr bfloat, ptr addrspace(1) %66, i64 %249, !dbg !56 + %349 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 81920), i32 %113, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %349, ptr addrspace(1) %341, i32 %275) #2, !dbg !57 + %350 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 81920), i32 %116, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %350, ptr addrspace(1) %342, i32 %277) #2, !dbg !57 + %351 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 81920), i32 %119, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %351, ptr addrspace(1) %343, i32 %279) #2, !dbg !57 + %352 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 81920), i32 %122, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %352, ptr addrspace(1) %344, i32 %281) #2, !dbg !57 + %353 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 81920), i32 %125, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %353, ptr addrspace(1) %345, i32 %283) #2, !dbg !57 + %354 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 81920), i32 %128, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %354, ptr addrspace(1) %346, i32 %285) #2, !dbg !57 + %355 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 81920), i32 %131, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %355, ptr addrspace(1) %347, i32 %287) #2, !dbg !57 + %356 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 81920), i32 %134, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %356, ptr addrspace(1) %348, i32 %289) #2, !dbg !57 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !57 + %357 = getelementptr bfloat, ptr addrspace(1) %68, i64 %242, !dbg !58 + %358 = getelementptr bfloat, ptr addrspace(1) %68, i64 %243, !dbg !58 + %359 = getelementptr bfloat, ptr addrspace(1) %68, i64 %244, !dbg !58 + %360 = getelementptr bfloat, ptr addrspace(1) %68, i64 %245, !dbg !58 + %361 = getelementptr bfloat, ptr addrspace(1) %68, i64 %246, !dbg !58 + %362 = getelementptr bfloat, ptr addrspace(1) %68, i64 %247, !dbg !58 + %363 = getelementptr bfloat, ptr addrspace(1) %68, i64 %248, !dbg !58 + %364 = getelementptr bfloat, ptr addrspace(1) %68, i64 %249, !dbg !58 + %365 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 106496), i32 %113, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %365, ptr addrspace(1) %357, i32 %275) #2, !dbg !59 + %366 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 106496), i32 %116, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %366, ptr addrspace(1) %358, i32 %277) #2, !dbg !59 + %367 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 106496), i32 %119, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %367, ptr addrspace(1) %359, i32 %279) #2, !dbg !59 + %368 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 106496), i32 %122, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %368, ptr addrspace(1) %360, i32 %281) #2, !dbg !59 + %369 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 106496), i32 %125, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %369, ptr addrspace(1) %361, i32 %283) #2, !dbg !59 + %370 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 106496), i32 %128, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %370, ptr addrspace(1) %362, i32 %285) #2, !dbg !59 + %371 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 106496), i32 %131, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %371, ptr addrspace(1) %363, i32 %287) #2, !dbg !59 + %372 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 106496), i32 %134, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %372, ptr addrspace(1) %364, i32 %289) #2, !dbg !59 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !59 + %373 = icmp sgt i32 %16, 191, !dbg !38 + %374 = or disjoint i64 %49, 128, !dbg !60 + %375 = or disjoint i64 %50, 128, !dbg !60 + %376 = or disjoint i64 %51, 128, !dbg !60 + %377 = or disjoint i64 %52, 128, !dbg !60 + %378 = or disjoint i64 %53, 128, !dbg !60 + %379 = or disjoint i64 %54, 128, !dbg !60 + %380 = or disjoint i64 %55, 128, !dbg !60 + %381 = or disjoint i64 %56, 128, !dbg !60 + %382 = or disjoint i64 %62, 128, !dbg !60 + %383 = icmp slt i64 %374, %15, !dbg !39 + %384 = icmp slt i64 %375, %15, !dbg !39 + %385 = icmp slt i64 %376, %15, !dbg !39 + %386 = icmp slt i64 %377, %15, !dbg !39 + %387 = icmp slt i64 %378, %15, !dbg !39 + %388 = icmp slt i64 %379, %15, !dbg !39 + %389 = icmp slt i64 %380, %15, !dbg !39 + %390 = icmp slt i64 %381, %15, !dbg !39 + %391 = icmp slt i64 %382, %15, !dbg !39 + %392 = shl nuw nsw i64 %374, 12, !dbg !40 + %393 = shl nuw nsw i64 %375, 12, !dbg !40 + %394 = shl nuw nsw i64 %376, 12, !dbg !40 + %395 = shl nuw nsw i64 %377, 12, !dbg !40 + %396 = shl nuw nsw i64 %378, 12, !dbg !40 + %397 = shl nuw nsw i64 %379, 12, !dbg !40 + %398 = shl nuw nsw i64 %380, 12, !dbg !40 + %399 = shl nuw nsw i64 %381, 12, !dbg !40 + %400 = getelementptr bfloat, ptr addrspace(1) %23, i64 %392, !dbg !41 + %401 = getelementptr bfloat, ptr addrspace(1) %23, i64 %393, !dbg !41 + %402 = getelementptr bfloat, ptr addrspace(1) %23, i64 %394, !dbg !41 + %403 = getelementptr bfloat, ptr addrspace(1) %23, i64 %395, !dbg !41 + %404 = getelementptr bfloat, ptr addrspace(1) %23, i64 %396, !dbg !41 + %405 = getelementptr bfloat, ptr addrspace(1) %23, i64 %397, !dbg !41 + %406 = getelementptr bfloat, ptr addrspace(1) %23, i64 %398, !dbg !41 + %407 = getelementptr bfloat, ptr addrspace(1) %23, i64 %399, !dbg !41 + %408 = getelementptr bfloat, ptr addrspace(1) %400, i64 %48, !dbg !42 + %409 = getelementptr bfloat, ptr addrspace(1) %401, i64 %48, !dbg !42 + %410 = getelementptr bfloat, ptr addrspace(1) %402, i64 %48, !dbg !42 + %411 = getelementptr bfloat, ptr addrspace(1) %403, i64 %48, !dbg !42 + %412 = getelementptr bfloat, ptr addrspace(1) %404, i64 %48, !dbg !42 + %413 = getelementptr bfloat, ptr addrspace(1) %405, i64 %48, !dbg !42 + %414 = getelementptr bfloat, ptr addrspace(1) %406, i64 %48, !dbg !42 + %415 = getelementptr bfloat, ptr addrspace(1) %407, i64 %48, !dbg !42 + %416 = and i1 %373, %383, !dbg !38 + %417 = and i1 %373, %384, !dbg !38 + %418 = and i1 %373, %385, !dbg !38 + %419 = and i1 %373, %386, !dbg !38 + %420 = and i1 %373, %387, !dbg !38 + %421 = and i1 %373, %388, !dbg !38 + %422 = and i1 %373, %389, !dbg !38 + %423 = and i1 %373, %390, !dbg !38 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !43 + %424 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %113, !dbg !43 + %425 = select i1 %416, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %424, ptr addrspace(1) %408, i32 %425) #2, !dbg !43 + %426 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %116, !dbg !43 + %427 = select i1 %417, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %426, ptr addrspace(1) %409, i32 %427) #2, !dbg !43 + %428 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %119, !dbg !43 + %429 = select i1 %418, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %428, ptr addrspace(1) %410, i32 %429) #2, !dbg !43 + %430 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %122, !dbg !43 + %431 = select i1 %419, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %430, ptr addrspace(1) %411, i32 %431) #2, !dbg !43 + %432 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %125, !dbg !43 + %433 = select i1 %420, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %432, ptr addrspace(1) %412, i32 %433) #2, !dbg !43 + %434 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %128, !dbg !43 + %435 = select i1 %421, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %434, ptr addrspace(1) %413, i32 %435) #2, !dbg !43 + %436 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %131, !dbg !43 + %437 = select i1 %422, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %436, ptr addrspace(1) %414, i32 %437) #2, !dbg !43 + %438 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %134, !dbg !43 + %439 = select i1 %423, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %438, ptr addrspace(1) %415, i32 %439) #2, !dbg !43 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !43 + %440 = getelementptr bfloat, ptr addrspace(1) %400, i64 %67, !dbg !44 + %441 = getelementptr bfloat, ptr addrspace(1) %401, i64 %67, !dbg !44 + %442 = getelementptr bfloat, ptr addrspace(1) %402, i64 %67, !dbg !44 + %443 = getelementptr bfloat, ptr addrspace(1) %403, i64 %67, !dbg !44 + %444 = getelementptr bfloat, ptr addrspace(1) %404, i64 %67, !dbg !44 + %445 = getelementptr bfloat, ptr addrspace(1) %405, i64 %67, !dbg !44 + %446 = getelementptr bfloat, ptr addrspace(1) %406, i64 %67, !dbg !44 + %447 = getelementptr bfloat, ptr addrspace(1) %407, i64 %67, !dbg !44 + %448 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 40960), i32 %113, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %448, ptr addrspace(1) %440, i32 %425) #2, !dbg !45 + %449 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 40960), i32 %116, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %449, ptr addrspace(1) %441, i32 %427) #2, !dbg !45 + %450 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 40960), i32 %119, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %450, ptr addrspace(1) %442, i32 %429) #2, !dbg !45 + %451 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 40960), i32 %122, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %451, ptr addrspace(1) %443, i32 %431) #2, !dbg !45 + %452 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 40960), i32 %125, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %452, ptr addrspace(1) %444, i32 %433) #2, !dbg !45 + %453 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 40960), i32 %128, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %453, ptr addrspace(1) %445, i32 %435) #2, !dbg !45 + %454 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 40960), i32 %131, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %454, ptr addrspace(1) %446, i32 %437) #2, !dbg !45 + %455 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 40960), i32 %134, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %455, ptr addrspace(1) %447, i32 %439) #2, !dbg !45 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !45 + %456 = getelementptr bfloat, ptr addrspace(1) %21, i64 %392, !dbg !46 + %457 = getelementptr bfloat, ptr addrspace(1) %21, i64 %393, !dbg !46 + %458 = getelementptr bfloat, ptr addrspace(1) %21, i64 %394, !dbg !46 + %459 = getelementptr bfloat, ptr addrspace(1) %21, i64 %395, !dbg !46 + %460 = getelementptr bfloat, ptr addrspace(1) %21, i64 %396, !dbg !46 + %461 = getelementptr bfloat, ptr addrspace(1) %21, i64 %397, !dbg !46 + %462 = getelementptr bfloat, ptr addrspace(1) %21, i64 %398, !dbg !46 + %463 = getelementptr bfloat, ptr addrspace(1) %21, i64 %399, !dbg !46 + %464 = getelementptr bfloat, ptr addrspace(1) %456, i64 %63, !dbg !47 + %465 = getelementptr bfloat, ptr addrspace(1) %457, i64 %63, !dbg !47 + %466 = getelementptr bfloat, ptr addrspace(1) %458, i64 %63, !dbg !47 + %467 = getelementptr bfloat, ptr addrspace(1) %459, i64 %63, !dbg !47 + %468 = getelementptr bfloat, ptr addrspace(1) %460, i64 %63, !dbg !47 + %469 = getelementptr bfloat, ptr addrspace(1) %461, i64 %63, !dbg !47 + %470 = getelementptr bfloat, ptr addrspace(1) %462, i64 %63, !dbg !47 + %471 = getelementptr bfloat, ptr addrspace(1) %463, i64 %63, !dbg !47 + %472 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 65536), i32 %113, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %472, ptr addrspace(1) %464, i32 %425) #2, !dbg !48 + %473 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 65536), i32 %116, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %473, ptr addrspace(1) %465, i32 %427) #2, !dbg !48 + %474 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 65536), i32 %119, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %474, ptr addrspace(1) %466, i32 %429) #2, !dbg !48 + %475 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 65536), i32 %122, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %475, ptr addrspace(1) %467, i32 %431) #2, !dbg !48 + %476 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 65536), i32 %125, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %476, ptr addrspace(1) %468, i32 %433) #2, !dbg !48 + %477 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 65536), i32 %128, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %477, ptr addrspace(1) %469, i32 %435) #2, !dbg !48 + %478 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 65536), i32 %131, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %478, ptr addrspace(1) %470, i32 %437) #2, !dbg !48 + %479 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 65536), i32 %134, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %479, ptr addrspace(1) %471, i32 %439) #2, !dbg !48 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !48 + %480 = tail call i32 @llvm.smin.i32(i32 %6, i32 192), !dbg !49 + %481 = shl i32 %480, 5, !dbg !50 + %482 = add i32 %481, -32, !dbg !50 + %483 = sext i32 %482 to i64, !dbg !51 + %484 = getelementptr float, ptr addrspace(1) %181, i64 %483, !dbg !52 + %485 = getelementptr float, ptr addrspace(1) %484, i64 %14, !dbg !52 + %486 = select i1 %373, i32 4, i32 0, !dbg !53 + tail call void asm sideeffect "@$3 cp.async.ca.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x4, $2;", "r,l,r,b"(ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 142088), ptr addrspace(1) %485, i32 %486, i1 %184) #2, !dbg !53 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !53 + %.idx3 = shl nuw nsw i64 %382, 7, !dbg !54 + %487 = getelementptr i8, ptr addrspace(1) %65, i64 %.idx3, !dbg !54 + %488 = and i1 %373, %391, !dbg !38 + %489 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 141824), i32 %188, !dbg !55 + %490 = select i1 %488, i32 4, i32 0, !dbg !55 + tail call void asm sideeffect "cp.async.ca.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x4, $2;", "r,l,r"(ptr addrspace(3) %489, ptr addrspace(1) %487, i32 %490) #2, !dbg !55 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !55 + %491 = getelementptr bfloat, ptr addrspace(1) %66, i64 %392, !dbg !56 + %492 = getelementptr bfloat, ptr addrspace(1) %66, i64 %393, !dbg !56 + %493 = getelementptr bfloat, ptr addrspace(1) %66, i64 %394, !dbg !56 + %494 = getelementptr bfloat, ptr addrspace(1) %66, i64 %395, !dbg !56 + %495 = getelementptr bfloat, ptr addrspace(1) %66, i64 %396, !dbg !56 + %496 = getelementptr bfloat, ptr addrspace(1) %66, i64 %397, !dbg !56 + %497 = getelementptr bfloat, ptr addrspace(1) %66, i64 %398, !dbg !56 + %498 = getelementptr bfloat, ptr addrspace(1) %66, i64 %399, !dbg !56 + %499 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 90112), i32 %113, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %499, ptr addrspace(1) %491, i32 %425) #2, !dbg !57 + %500 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 90112), i32 %116, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %500, ptr addrspace(1) %492, i32 %427) #2, !dbg !57 + %501 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 90112), i32 %119, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %501, ptr addrspace(1) %493, i32 %429) #2, !dbg !57 + %502 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 90112), i32 %122, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %502, ptr addrspace(1) %494, i32 %431) #2, !dbg !57 + %503 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 90112), i32 %125, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %503, ptr addrspace(1) %495, i32 %433) #2, !dbg !57 + %504 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 90112), i32 %128, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %504, ptr addrspace(1) %496, i32 %435) #2, !dbg !57 + %505 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 90112), i32 %131, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %505, ptr addrspace(1) %497, i32 %437) #2, !dbg !57 + %506 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 90112), i32 %134, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %506, ptr addrspace(1) %498, i32 %439) #2, !dbg !57 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !57 + %507 = getelementptr bfloat, ptr addrspace(1) %68, i64 %392, !dbg !58 + %508 = getelementptr bfloat, ptr addrspace(1) %68, i64 %393, !dbg !58 + %509 = getelementptr bfloat, ptr addrspace(1) %68, i64 %394, !dbg !58 + %510 = getelementptr bfloat, ptr addrspace(1) %68, i64 %395, !dbg !58 + %511 = getelementptr bfloat, ptr addrspace(1) %68, i64 %396, !dbg !58 + %512 = getelementptr bfloat, ptr addrspace(1) %68, i64 %397, !dbg !58 + %513 = getelementptr bfloat, ptr addrspace(1) %68, i64 %398, !dbg !58 + %514 = getelementptr bfloat, ptr addrspace(1) %68, i64 %399, !dbg !58 + %515 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 114688), i32 %113, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %515, ptr addrspace(1) %507, i32 %425) #2, !dbg !59 + %516 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 114688), i32 %116, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %516, ptr addrspace(1) %508, i32 %427) #2, !dbg !59 + %517 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 114688), i32 %119, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %517, ptr addrspace(1) %509, i32 %429) #2, !dbg !59 + %518 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 114688), i32 %122, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %518, ptr addrspace(1) %510, i32 %431) #2, !dbg !59 + %519 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 114688), i32 %125, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %519, ptr addrspace(1) %511, i32 %433) #2, !dbg !59 + %520 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 114688), i32 %128, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %520, ptr addrspace(1) %512, i32 %435) #2, !dbg !59 + %521 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 114688), i32 %131, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %521, ptr addrspace(1) %513, i32 %437) #2, !dbg !59 + %522 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 114688), i32 %134, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %522, ptr addrspace(1) %514, i32 %439) #2, !dbg !59 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !59 + %invariant.gep = getelementptr bfloat, ptr addrspace(1) %24, i64 %63, !dbg !38 + %invariant.gep527 = getelementptr bfloat, ptr addrspace(1) %21, i64 %63, !dbg !38 + %invariant.gep543 = getelementptr float, ptr addrspace(1) %181, i64 %14, !dbg !38 + br i1 %69, label %.lr.ph, label %._crit_edge, !dbg !38 + +.lr.ph: ; preds = %9 + %523 = lshr i32 %16, 6, !dbg !61 + %524 = zext nneg i32 %523 to i64, !dbg !62 + %factor.op.mul = mul nuw nsw i64 %13, %524, !dbg !38 + %525 = or disjoint <4 x i64> %61, splat (i64 128), !dbg !60 + %526 = or disjoint <4 x i64> %61, splat (i64 64), !dbg !60 + %527 = insertelement <4 x i32> poison, i32 %6, i64 0, !dbg !39 + %528 = shufflevector <4 x i32> %527, <4 x i32> poison, <4 x i32> zeroinitializer, !dbg !39 + %529 = icmp ult <4 x i32> %60, %528, !dbg !39 + %530 = shl nuw nsw i32 %38, 7, !dbg !63 + %531 = shl nuw nsw i32 %37, 7, !dbg !63 + %532 = shl nuw nsw i32 %36, 7, !dbg !63 + %533 = shl nuw nsw i32 %35, 7, !dbg !63 + %534 = shl nuw nsw i32 %34, 7, !dbg !63 + %535 = shl nuw nsw i32 %33, 7, !dbg !63 + %536 = shl nuw nsw i32 %32, 7, !dbg !63 + %537 = shl nuw nsw i32 %30, 4, !dbg !63 + %538 = add nsw i32 %523, -3 + %.reass = shl i64 %factor.op.mul, 20 + %539 = getelementptr i8, ptr addrspace(1) %5, i64 %.reass + %.idx496 = shl nuw nsw i64 %14, 15 + %540 = getelementptr i8, ptr addrspace(1) %539, i64 %.idx496 + %541 = zext nneg i32 %537 to i64 + %542 = zext nneg i32 %536 to i64 + %543 = zext nneg i32 %535 to i64 + %544 = zext nneg i32 %534 to i64 + %545 = zext nneg i32 %533 to i64 + %546 = zext nneg i32 %532 to i64 + %547 = zext nneg i32 %531 to i64 + %548 = zext nneg i32 %530 to i64 + %549 = shl nuw nsw i32 %27, 7 + %550 = and i32 %549, 1920 + %551 = shl nuw nsw i32 %28, 4 + %552 = and i32 %27, 16 + %553 = shl nuw nsw i32 %41, 6 + %554 = or disjoint i32 %550, %551 + %555 = xor i32 %554, %552 + %556 = or disjoint i32 %555, %553 + %557 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %556 + %558 = or disjoint i32 %556, 4096 + %559 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %558 + %560 = xor i32 %556, 32 + %561 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %560 + %562 = or disjoint i32 %560, 4096 + %563 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %562 + %564 = xor i32 %556, 64 + %565 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %564 + %566 = or disjoint i32 %564, 4096 + %567 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %566 + %568 = xor i32 %556, 96 + %569 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %568 + %570 = or disjoint i32 %568, 4096 + %571 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %570 + %572 = shl nuw nsw i32 %27, 3 + %573 = and i32 %572, 384 + %574 = shl nuw nsw i32 %27, 5 + %575 = and i32 %574, 32 + %576 = and i32 %27, 2 + %577 = icmp eq i32 %576, 0 + %578 = select i1 %577, i32 0, i32 1088 + %579 = and i32 %27, 4 + %580 = icmp eq i32 %579, 0 + %581 = shl nuw nsw i32 %579, 2 + %582 = and i32 %27, 8 + %583 = icmp eq i32 %582, 0 + %584 = select i1 %583, i32 0, i32 544 + %585 = or disjoint i32 %578, %581 + %586 = or disjoint i32 %573, %575 + %587 = xor i32 %586, %584 + %588 = or disjoint i32 %585, %587 + %589 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %588 + %590 = xor i32 %588, 64 + %591 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %590 + %592 = shl nuw nsw i32 %27, 6 + %593 = and i32 %592, 1088 + %594 = shl nuw nsw i32 %30, 2 + %595 = shl nuw nsw i32 %576, 3 + %596 = select i1 %580, i32 0, i32 544 + %597 = xor i32 %593, %594 + %598 = xor i32 %597, %596 + %599 = or disjoint i32 %598, %595 + %600 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %599 + %601 = or disjoint i32 %599, 256 + %602 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %601 + %603 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %556 + %604 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %558 + %605 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %560 + %606 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %562 + %607 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %564 + %608 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %566 + %609 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %568 + %610 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %570 + %611 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 139264), i32 %588 + %612 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 139264), i32 %590 + %613 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 139264), i32 %599 + %614 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 139264), i32 %601 + %615 = and i32 %549, 3968 + %616 = or disjoint i32 %551, %615 + %617 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %616 + %618 = or disjoint i32 %616, 4096 + %619 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %618 + %620 = xor i32 %616, 16 + %621 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %620 + %622 = or disjoint i32 %620, 4096 + %623 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %622 + %624 = xor i32 %616, 32 + %625 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %624 + %626 = or disjoint i32 %624, 4096 + %627 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %626 + %628 = xor i32 %616, 48 + %629 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %628 + %630 = or disjoint i32 %628, 4096 + %631 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %630 + %632 = xor i32 %616, 64 + %633 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %632 + %634 = or disjoint i32 %632, 4096 + %635 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %634 + %636 = xor i32 %616, 80 + %637 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %636 + %638 = or disjoint i32 %636, 4096 + %639 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %638 + %640 = xor i32 %616, 96 + %641 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %640 + %642 = or disjoint i32 %640, 4096 + %643 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %642 + %644 = xor i32 %616, 112 + %645 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %644 + %646 = or disjoint i32 %644, 4096 + %647 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %646 + %648 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %616 + %649 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %618 + %650 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %620 + %651 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %622 + %652 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %624 + %653 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %626 + %654 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %628 + %655 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %630 + %656 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %632 + %657 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %634 + %658 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %636 + %659 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %638 + %660 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %640 + %661 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %642 + %662 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %644 + %663 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 131072), i32 %646 + %664 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %588 + %665 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %590 + %666 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %599 + %667 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 122880), i32 %601 + %668 = shl nuw nsw i32 %41, 1 + %invariant.gep607 = getelementptr i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 141312), i32 %668, !dbg !38 + %invariant.gep609 = getelementptr i8, ptr addrspace(3) %invariant.gep607, i32 %39, !dbg !38 + %669 = shl nuw nsw i32 %582, 1 + %670 = shl nuw nsw i32 %552, 6 + %671 = mul nuw nsw i32 %28, 144 + %672 = or disjoint i32 %669, %41 + %673 = xor i32 %671, %672 + %674 = or disjoint i32 %673, %670 + %675 = or disjoint i32 %674, 2048 + %676 = or disjoint i32 %674, 4096 + %677 = or disjoint i32 %674, 6144 + %678 = xor i32 %674, 64 + %679 = or disjoint i32 %678, 2048 + %680 = or disjoint i32 %678, 4096 + %681 = or disjoint i32 %678, 6144 + %682 = sext i32 %538 to i64, !dbg !38 + %683 = zext nneg <4 x i32> %528 to <4 x i64> + br label %684, !dbg !38 + +684: ; preds = %.lr.ph, %__nv_exp2f.exit512 + %indvars.iv = phi i64 [ 0, %.lr.ph ], [ %indvars.iv.next, %__nv_exp2f.exit512 ] + %.pn243594 = phi i1 [ %390, %.lr.ph ], [ %3772, %__nv_exp2f.exit512 ] + %.pn259593 = phi i1 [ %389, %.lr.ph ], [ %3771, %__nv_exp2f.exit512 ] + %.pn275592 = phi i1 [ %388, %.lr.ph ], [ %3770, %__nv_exp2f.exit512 ] + %.pn291591 = phi i1 [ %387, %.lr.ph ], [ %3769, %__nv_exp2f.exit512 ] + %.pn307590 = phi i1 [ %386, %.lr.ph ], [ %3768, %__nv_exp2f.exit512 ] + %.pn323589 = phi i1 [ %385, %.lr.ph ], [ %3767, %__nv_exp2f.exit512 ] + %.pn339588 = phi i1 [ %384, %.lr.ph ], [ %3766, %__nv_exp2f.exit512 ] + %.pn355587 = phi i1 [ %383, %.lr.ph ], [ %3765, %__nv_exp2f.exit512 ] + %.pn243.pn586 = phi i1 [ %240, %.lr.ph ], [ %.pn243594, %__nv_exp2f.exit512 ] + %.pn259.pn585 = phi i1 [ %239, %.lr.ph ], [ %.pn259593, %__nv_exp2f.exit512 ] + %.pn275.pn584 = phi i1 [ %238, %.lr.ph ], [ %.pn275592, %__nv_exp2f.exit512 ] + %.pn291.pn583 = phi i1 [ %237, %.lr.ph ], [ %.pn291591, %__nv_exp2f.exit512 ] + %.pn307.pn582 = phi i1 [ %236, %.lr.ph ], [ %.pn307590, %__nv_exp2f.exit512 ] + %.pn323.pn581 = phi i1 [ %235, %.lr.ph ], [ %.pn323589, %__nv_exp2f.exit512 ] + %.pn339.pn580 = phi i1 [ %234, %.lr.ph ], [ %.pn339588, %__nv_exp2f.exit512 ] + %.pn355.pn579 = phi i1 [ %233, %.lr.ph ], [ %.pn355587, %__nv_exp2f.exit512 ] + %.pn243.pn.pn578 = phi i1 [ %77, %.lr.ph ], [ %.pn243.pn586, %__nv_exp2f.exit512 ] + %.pn259.pn.pn577 = phi i1 [ %76, %.lr.ph ], [ %.pn259.pn585, %__nv_exp2f.exit512 ] + %.pn275.pn.pn576 = phi i1 [ %75, %.lr.ph ], [ %.pn275.pn584, %__nv_exp2f.exit512 ] + %.pn291.pn.pn575 = phi i1 [ %74, %.lr.ph ], [ %.pn291.pn583, %__nv_exp2f.exit512 ] + %.pn307.pn.pn574 = phi i1 [ %73, %.lr.ph ], [ %.pn307.pn582, %__nv_exp2f.exit512 ] + %.pn323.pn.pn573 = phi i1 [ %72, %.lr.ph ], [ %.pn323.pn581, %__nv_exp2f.exit512 ] + %.pn339.pn.pn572 = phi i1 [ %71, %.lr.ph ], [ %.pn339.pn580, %__nv_exp2f.exit512 ] + %.pn355.pn.pn571 = phi i1 [ %70, %.lr.ph ], [ %.pn355.pn579, %__nv_exp2f.exit512 ] + %.pn4570 = phi i64 [ %399, %.lr.ph ], [ %3781, %__nv_exp2f.exit512 ] + %.pn20569 = phi i64 [ %398, %.lr.ph ], [ %3780, %__nv_exp2f.exit512 ] + %.pn36568 = phi i64 [ %397, %.lr.ph ], [ %3779, %__nv_exp2f.exit512 ] + %.pn52567 = phi i64 [ %396, %.lr.ph ], [ %3778, %__nv_exp2f.exit512 ] + %.pn68566 = phi i64 [ %395, %.lr.ph ], [ %3777, %__nv_exp2f.exit512 ] + %.pn84565 = phi i64 [ %394, %.lr.ph ], [ %3776, %__nv_exp2f.exit512 ] + %.pn100564 = phi i64 [ %393, %.lr.ph ], [ %3775, %__nv_exp2f.exit512 ] + %.pn116563 = phi i64 [ %392, %.lr.ph ], [ %3774, %__nv_exp2f.exit512 ] + %.pn4.pn562 = phi i64 [ %249, %.lr.ph ], [ %.pn4570, %__nv_exp2f.exit512 ] + %.pn20.pn561 = phi i64 [ %248, %.lr.ph ], [ %.pn20569, %__nv_exp2f.exit512 ] + %.pn36.pn560 = phi i64 [ %247, %.lr.ph ], [ %.pn36568, %__nv_exp2f.exit512 ] + %.pn52.pn559 = phi i64 [ %246, %.lr.ph ], [ %.pn52567, %__nv_exp2f.exit512 ] + %.pn68.pn558 = phi i64 [ %245, %.lr.ph ], [ %.pn68566, %__nv_exp2f.exit512 ] + %.pn84.pn557 = phi i64 [ %244, %.lr.ph ], [ %.pn84565, %__nv_exp2f.exit512 ] + %.pn100.pn556 = phi i64 [ %243, %.lr.ph ], [ %.pn100564, %__nv_exp2f.exit512 ] + %.pn116.pn555 = phi i64 [ %242, %.lr.ph ], [ %.pn116563, %__nv_exp2f.exit512 ] + %.pn4.pn.pn554 = phi i64 [ %86, %.lr.ph ], [ %.pn4.pn562, %__nv_exp2f.exit512 ] + %.pn20.pn.pn553 = phi i64 [ %85, %.lr.ph ], [ %.pn20.pn561, %__nv_exp2f.exit512 ] + %.pn36.pn.pn552 = phi i64 [ %84, %.lr.ph ], [ %.pn36.pn560, %__nv_exp2f.exit512 ] + %.pn52.pn.pn551 = phi i64 [ %83, %.lr.ph ], [ %.pn52.pn559, %__nv_exp2f.exit512 ] + %.pn68.pn.pn550 = phi i64 [ %82, %.lr.ph ], [ %.pn68.pn558, %__nv_exp2f.exit512 ] + %.pn84.pn.pn549 = phi i64 [ %81, %.lr.ph ], [ %.pn84.pn557, %__nv_exp2f.exit512 ] + %.pn100.pn.pn548 = phi i64 [ %80, %.lr.ph ], [ %.pn100.pn556, %__nv_exp2f.exit512 ] + %.pn116.pn.pn547 = phi i64 [ %79, %.lr.ph ], [ %.pn116.pn555, %__nv_exp2f.exit512 ] + %685 = phi i32 [ -1, %.lr.ph ], [ %822, %__nv_exp2f.exit512 ] + %686 = phi i32 [ 2, %.lr.ph ], [ %3750, %__nv_exp2f.exit512 ] + %687 = phi float [ 0.000000e+00, %.lr.ph ], [ %3669, %__nv_exp2f.exit512 ] + %688 = phi float [ 0.000000e+00, %.lr.ph ], [ %3670, %__nv_exp2f.exit512 ] + %689 = phi float [ 0.000000e+00, %.lr.ph ], [ %3671, %__nv_exp2f.exit512 ] + %690 = phi float [ 0.000000e+00, %.lr.ph ], [ %3672, %__nv_exp2f.exit512 ] + %691 = phi float [ 0.000000e+00, %.lr.ph ], [ %3674, %__nv_exp2f.exit512 ] + %692 = phi float [ 0.000000e+00, %.lr.ph ], [ %3675, %__nv_exp2f.exit512 ] + %693 = phi float [ 0.000000e+00, %.lr.ph ], [ %3676, %__nv_exp2f.exit512 ] + %694 = phi float [ 0.000000e+00, %.lr.ph ], [ %3677, %__nv_exp2f.exit512 ] + %695 = phi float [ 0.000000e+00, %.lr.ph ], [ %3679, %__nv_exp2f.exit512 ] + %696 = phi float [ 0.000000e+00, %.lr.ph ], [ %3680, %__nv_exp2f.exit512 ] + %697 = phi float [ 0.000000e+00, %.lr.ph ], [ %3681, %__nv_exp2f.exit512 ] + %698 = phi float [ 0.000000e+00, %.lr.ph ], [ %3682, %__nv_exp2f.exit512 ] + %699 = phi float [ 0.000000e+00, %.lr.ph ], [ %3684, %__nv_exp2f.exit512 ] + %700 = phi float [ 0.000000e+00, %.lr.ph ], [ %3685, %__nv_exp2f.exit512 ] + %701 = phi float [ 0.000000e+00, %.lr.ph ], [ %3686, %__nv_exp2f.exit512 ] + %702 = phi float [ 0.000000e+00, %.lr.ph ], [ %3687, %__nv_exp2f.exit512 ] + %703 = phi float [ 0.000000e+00, %.lr.ph ], [ %3689, %__nv_exp2f.exit512 ] + %704 = phi float [ 0.000000e+00, %.lr.ph ], [ %3690, %__nv_exp2f.exit512 ] + %705 = phi float [ 0.000000e+00, %.lr.ph ], [ %3691, %__nv_exp2f.exit512 ] + %706 = phi float [ 0.000000e+00, %.lr.ph ], [ %3692, %__nv_exp2f.exit512 ] + %707 = phi float [ 0.000000e+00, %.lr.ph ], [ %3694, %__nv_exp2f.exit512 ] + %708 = phi float [ 0.000000e+00, %.lr.ph ], [ %3695, %__nv_exp2f.exit512 ] + %709 = phi float [ 0.000000e+00, %.lr.ph ], [ %3696, %__nv_exp2f.exit512 ] + %710 = phi float [ 0.000000e+00, %.lr.ph ], [ %3697, %__nv_exp2f.exit512 ] + %711 = phi float [ 0.000000e+00, %.lr.ph ], [ %3699, %__nv_exp2f.exit512 ] + %712 = phi float [ 0.000000e+00, %.lr.ph ], [ %3700, %__nv_exp2f.exit512 ] + %713 = phi float [ 0.000000e+00, %.lr.ph ], [ %3701, %__nv_exp2f.exit512 ] + %714 = phi float [ 0.000000e+00, %.lr.ph ], [ %3702, %__nv_exp2f.exit512 ] + %715 = phi float [ 0.000000e+00, %.lr.ph ], [ %3704, %__nv_exp2f.exit512 ] + %716 = phi float [ 0.000000e+00, %.lr.ph ], [ %3705, %__nv_exp2f.exit512 ] + %717 = phi float [ 0.000000e+00, %.lr.ph ], [ %3706, %__nv_exp2f.exit512 ] + %718 = phi float [ 0.000000e+00, %.lr.ph ], [ %3707, %__nv_exp2f.exit512 ] + %719 = phi float [ 0.000000e+00, %.lr.ph ], [ %3709, %__nv_exp2f.exit512 ] + %720 = phi float [ 0.000000e+00, %.lr.ph ], [ %3710, %__nv_exp2f.exit512 ] + %721 = phi float [ 0.000000e+00, %.lr.ph ], [ %3711, %__nv_exp2f.exit512 ] + %722 = phi float [ 0.000000e+00, %.lr.ph ], [ %3712, %__nv_exp2f.exit512 ] + %723 = phi float [ 0.000000e+00, %.lr.ph ], [ %3714, %__nv_exp2f.exit512 ] + %724 = phi float [ 0.000000e+00, %.lr.ph ], [ %3715, %__nv_exp2f.exit512 ] + %725 = phi float [ 0.000000e+00, %.lr.ph ], [ %3716, %__nv_exp2f.exit512 ] + %726 = phi float [ 0.000000e+00, %.lr.ph ], [ %3717, %__nv_exp2f.exit512 ] + %727 = phi float [ 0.000000e+00, %.lr.ph ], [ %3719, %__nv_exp2f.exit512 ] + %728 = phi float [ 0.000000e+00, %.lr.ph ], [ %3720, %__nv_exp2f.exit512 ] + %729 = phi float [ 0.000000e+00, %.lr.ph ], [ %3721, %__nv_exp2f.exit512 ] + %730 = phi float [ 0.000000e+00, %.lr.ph ], [ %3722, %__nv_exp2f.exit512 ] + %731 = phi float [ 0.000000e+00, %.lr.ph ], [ %3724, %__nv_exp2f.exit512 ] + %732 = phi float [ 0.000000e+00, %.lr.ph ], [ %3725, %__nv_exp2f.exit512 ] + %733 = phi float [ 0.000000e+00, %.lr.ph ], [ %3726, %__nv_exp2f.exit512 ] + %734 = phi float [ 0.000000e+00, %.lr.ph ], [ %3727, %__nv_exp2f.exit512 ] + %735 = phi float [ 0.000000e+00, %.lr.ph ], [ %3729, %__nv_exp2f.exit512 ] + %736 = phi float [ 0.000000e+00, %.lr.ph ], [ %3730, %__nv_exp2f.exit512 ] + %737 = phi float [ 0.000000e+00, %.lr.ph ], [ %3731, %__nv_exp2f.exit512 ] + %738 = phi float [ 0.000000e+00, %.lr.ph ], [ %3732, %__nv_exp2f.exit512 ] + %739 = phi float [ 0.000000e+00, %.lr.ph ], [ %3734, %__nv_exp2f.exit512 ] + %740 = phi float [ 0.000000e+00, %.lr.ph ], [ %3735, %__nv_exp2f.exit512 ] + %741 = phi float [ 0.000000e+00, %.lr.ph ], [ %3736, %__nv_exp2f.exit512 ] + %742 = phi float [ 0.000000e+00, %.lr.ph ], [ %3737, %__nv_exp2f.exit512 ] + %743 = phi float [ 0.000000e+00, %.lr.ph ], [ %3739, %__nv_exp2f.exit512 ] + %744 = phi float [ 0.000000e+00, %.lr.ph ], [ %3740, %__nv_exp2f.exit512 ] + %745 = phi float [ 0.000000e+00, %.lr.ph ], [ %3741, %__nv_exp2f.exit512 ] + %746 = phi float [ 0.000000e+00, %.lr.ph ], [ %3742, %__nv_exp2f.exit512 ] + %747 = phi float [ 0.000000e+00, %.lr.ph ], [ %3744, %__nv_exp2f.exit512 ] + %748 = phi float [ 0.000000e+00, %.lr.ph ], [ %3745, %__nv_exp2f.exit512 ] + %749 = phi float [ 0.000000e+00, %.lr.ph ], [ %3746, %__nv_exp2f.exit512 ] + %750 = phi float [ 0.000000e+00, %.lr.ph ], [ %3747, %__nv_exp2f.exit512 ] + %751 = phi float [ 0.000000e+00, %.lr.ph ], [ %3300, %__nv_exp2f.exit512 ] + %752 = phi float [ 0.000000e+00, %.lr.ph ], [ %3301, %__nv_exp2f.exit512 ] + %753 = phi float [ 0.000000e+00, %.lr.ph ], [ %3302, %__nv_exp2f.exit512 ] + %754 = phi float [ 0.000000e+00, %.lr.ph ], [ %3303, %__nv_exp2f.exit512 ] + %755 = phi float [ 0.000000e+00, %.lr.ph ], [ %3305, %__nv_exp2f.exit512 ] + %756 = phi float [ 0.000000e+00, %.lr.ph ], [ %3306, %__nv_exp2f.exit512 ] + %757 = phi float [ 0.000000e+00, %.lr.ph ], [ %3307, %__nv_exp2f.exit512 ] + %758 = phi float [ 0.000000e+00, %.lr.ph ], [ %3308, %__nv_exp2f.exit512 ] + %759 = phi float [ 0.000000e+00, %.lr.ph ], [ %3310, %__nv_exp2f.exit512 ] + %760 = phi float [ 0.000000e+00, %.lr.ph ], [ %3311, %__nv_exp2f.exit512 ] + %761 = phi float [ 0.000000e+00, %.lr.ph ], [ %3312, %__nv_exp2f.exit512 ] + %762 = phi float [ 0.000000e+00, %.lr.ph ], [ %3313, %__nv_exp2f.exit512 ] + %763 = phi float [ 0.000000e+00, %.lr.ph ], [ %3315, %__nv_exp2f.exit512 ] + %764 = phi float [ 0.000000e+00, %.lr.ph ], [ %3316, %__nv_exp2f.exit512 ] + %765 = phi float [ 0.000000e+00, %.lr.ph ], [ %3317, %__nv_exp2f.exit512 ] + %766 = phi float [ 0.000000e+00, %.lr.ph ], [ %3318, %__nv_exp2f.exit512 ] + %767 = phi float [ 0.000000e+00, %.lr.ph ], [ %3320, %__nv_exp2f.exit512 ] + %768 = phi float [ 0.000000e+00, %.lr.ph ], [ %3321, %__nv_exp2f.exit512 ] + %769 = phi float [ 0.000000e+00, %.lr.ph ], [ %3322, %__nv_exp2f.exit512 ] + %770 = phi float [ 0.000000e+00, %.lr.ph ], [ %3323, %__nv_exp2f.exit512 ] + %771 = phi float [ 0.000000e+00, %.lr.ph ], [ %3325, %__nv_exp2f.exit512 ] + %772 = phi float [ 0.000000e+00, %.lr.ph ], [ %3326, %__nv_exp2f.exit512 ] + %773 = phi float [ 0.000000e+00, %.lr.ph ], [ %3327, %__nv_exp2f.exit512 ] + %774 = phi float [ 0.000000e+00, %.lr.ph ], [ %3328, %__nv_exp2f.exit512 ] + %775 = phi float [ 0.000000e+00, %.lr.ph ], [ %3330, %__nv_exp2f.exit512 ] + %776 = phi float [ 0.000000e+00, %.lr.ph ], [ %3331, %__nv_exp2f.exit512 ] + %777 = phi float [ 0.000000e+00, %.lr.ph ], [ %3332, %__nv_exp2f.exit512 ] + %778 = phi float [ 0.000000e+00, %.lr.ph ], [ %3333, %__nv_exp2f.exit512 ] + %779 = phi float [ 0.000000e+00, %.lr.ph ], [ %3335, %__nv_exp2f.exit512 ] + %780 = phi float [ 0.000000e+00, %.lr.ph ], [ %3336, %__nv_exp2f.exit512 ] + %781 = phi float [ 0.000000e+00, %.lr.ph ], [ %3337, %__nv_exp2f.exit512 ] + %782 = phi float [ 0.000000e+00, %.lr.ph ], [ %3338, %__nv_exp2f.exit512 ] + %783 = phi float [ 0.000000e+00, %.lr.ph ], [ %3340, %__nv_exp2f.exit512 ] + %784 = phi float [ 0.000000e+00, %.lr.ph ], [ %3341, %__nv_exp2f.exit512 ] + %785 = phi float [ 0.000000e+00, %.lr.ph ], [ %3342, %__nv_exp2f.exit512 ] + %786 = phi float [ 0.000000e+00, %.lr.ph ], [ %3343, %__nv_exp2f.exit512 ] + %787 = phi float [ 0.000000e+00, %.lr.ph ], [ %3345, %__nv_exp2f.exit512 ] + %788 = phi float [ 0.000000e+00, %.lr.ph ], [ %3346, %__nv_exp2f.exit512 ] + %789 = phi float [ 0.000000e+00, %.lr.ph ], [ %3347, %__nv_exp2f.exit512 ] + %790 = phi float [ 0.000000e+00, %.lr.ph ], [ %3348, %__nv_exp2f.exit512 ] + %791 = phi float [ 0.000000e+00, %.lr.ph ], [ %3350, %__nv_exp2f.exit512 ] + %792 = phi float [ 0.000000e+00, %.lr.ph ], [ %3351, %__nv_exp2f.exit512 ] + %793 = phi float [ 0.000000e+00, %.lr.ph ], [ %3352, %__nv_exp2f.exit512 ] + %794 = phi float [ 0.000000e+00, %.lr.ph ], [ %3353, %__nv_exp2f.exit512 ] + %795 = phi float [ 0.000000e+00, %.lr.ph ], [ %3355, %__nv_exp2f.exit512 ] + %796 = phi float [ 0.000000e+00, %.lr.ph ], [ %3356, %__nv_exp2f.exit512 ] + %797 = phi float [ 0.000000e+00, %.lr.ph ], [ %3357, %__nv_exp2f.exit512 ] + %798 = phi float [ 0.000000e+00, %.lr.ph ], [ %3358, %__nv_exp2f.exit512 ] + %799 = phi float [ 0.000000e+00, %.lr.ph ], [ %3360, %__nv_exp2f.exit512 ] + %800 = phi float [ 0.000000e+00, %.lr.ph ], [ %3361, %__nv_exp2f.exit512 ] + %801 = phi float [ 0.000000e+00, %.lr.ph ], [ %3362, %__nv_exp2f.exit512 ] + %802 = phi float [ 0.000000e+00, %.lr.ph ], [ %3363, %__nv_exp2f.exit512 ] + %803 = phi float [ 0.000000e+00, %.lr.ph ], [ %3365, %__nv_exp2f.exit512 ] + %804 = phi float [ 0.000000e+00, %.lr.ph ], [ %3366, %__nv_exp2f.exit512 ] + %805 = phi float [ 0.000000e+00, %.lr.ph ], [ %3367, %__nv_exp2f.exit512 ] + %806 = phi float [ 0.000000e+00, %.lr.ph ], [ %3368, %__nv_exp2f.exit512 ] + %807 = phi float [ 0.000000e+00, %.lr.ph ], [ %3370, %__nv_exp2f.exit512 ] + %808 = phi float [ 0.000000e+00, %.lr.ph ], [ %3371, %__nv_exp2f.exit512 ] + %809 = phi float [ 0.000000e+00, %.lr.ph ], [ %3372, %__nv_exp2f.exit512 ] + %810 = phi float [ 0.000000e+00, %.lr.ph ], [ %3373, %__nv_exp2f.exit512 ] + %811 = phi float [ 0.000000e+00, %.lr.ph ], [ %3375, %__nv_exp2f.exit512 ] + %812 = phi float [ 0.000000e+00, %.lr.ph ], [ %3376, %__nv_exp2f.exit512 ] + %813 = phi float [ 0.000000e+00, %.lr.ph ], [ %3377, %__nv_exp2f.exit512 ] + %814 = phi float [ 0.000000e+00, %.lr.ph ], [ %3378, %__nv_exp2f.exit512 ] + %815 = phi <4 x i1> [ %529, %.lr.ph ], [ %818, %__nv_exp2f.exit512 ] + %816 = phi <4 x i64> [ %526, %.lr.ph ], [ %817, %__nv_exp2f.exit512 ] + %817 = phi <4 x i64> [ %525, %.lr.ph ], [ %3763, %__nv_exp2f.exit512 ] + %818 = icmp slt <4 x i64> %816, %683, !dbg !39 + %819 = icmp slt i64 %indvars.iv, %682, !dbg !38 + %820 = add i32 %685, 1, !dbg !38 + %821 = icmp sgt i32 %820, 2, !dbg !38 + %822 = select i1 %821, i32 0, i32 %820, !dbg !38 + %.idx497 = shl nuw nsw i64 %indvars.iv, 20, !dbg !64 + %823 = getelementptr i8, ptr addrspace(1) %540, i64 %.idx497, !dbg !64 + %824 = getelementptr bfloat, ptr addrspace(1) %823, i64 %541, !dbg !65 + %825 = getelementptr bfloat, ptr addrspace(1) %823, i64 %542, !dbg !65 + %826 = getelementptr bfloat, ptr addrspace(1) %823, i64 %543, !dbg !65 + %827 = getelementptr bfloat, ptr addrspace(1) %823, i64 %544, !dbg !65 + %828 = getelementptr bfloat, ptr addrspace(1) %823, i64 %545, !dbg !65 + %829 = getelementptr bfloat, ptr addrspace(1) %823, i64 %546, !dbg !65 + %830 = getelementptr bfloat, ptr addrspace(1) %823, i64 %547, !dbg !65 + %831 = getelementptr bfloat, ptr addrspace(1) %823, i64 %548, !dbg !65 + %832 = getelementptr bfloat, ptr addrspace(1) %824, i64 %63, !dbg !66 + %833 = getelementptr bfloat, ptr addrspace(1) %825, i64 %63, !dbg !66 + %834 = getelementptr bfloat, ptr addrspace(1) %826, i64 %63, !dbg !66 + %835 = getelementptr bfloat, ptr addrspace(1) %827, i64 %63, !dbg !66 + %836 = getelementptr bfloat, ptr addrspace(1) %828, i64 %63, !dbg !66 + %837 = getelementptr bfloat, ptr addrspace(1) %829, i64 %63, !dbg !66 + %838 = getelementptr bfloat, ptr addrspace(1) %830, i64 %63, !dbg !66 + %839 = getelementptr bfloat, ptr addrspace(1) %831, i64 %63, !dbg !66 + %840 = fptrunc float %751 to bfloat, !dbg !67 + %841 = fptrunc float %752 to bfloat, !dbg !67 + %842 = fptrunc float %753 to bfloat, !dbg !67 + %843 = fptrunc float %754 to bfloat, !dbg !67 + %844 = fptrunc float %755 to bfloat, !dbg !67 + %845 = fptrunc float %756 to bfloat, !dbg !67 + %846 = fptrunc float %757 to bfloat, !dbg !67 + %847 = fptrunc float %758 to bfloat, !dbg !67 + %848 = fptrunc float %759 to bfloat, !dbg !67 + %849 = fptrunc float %760 to bfloat, !dbg !67 + %850 = fptrunc float %761 to bfloat, !dbg !67 + %851 = fptrunc float %762 to bfloat, !dbg !67 + %852 = fptrunc float %763 to bfloat, !dbg !67 + %853 = fptrunc float %764 to bfloat, !dbg !67 + %854 = fptrunc float %765 to bfloat, !dbg !67 + %855 = fptrunc float %766 to bfloat, !dbg !67 + %856 = fptrunc float %767 to bfloat, !dbg !67 + %857 = fptrunc float %768 to bfloat, !dbg !67 + %858 = fptrunc float %769 to bfloat, !dbg !67 + %859 = fptrunc float %770 to bfloat, !dbg !67 + %860 = fptrunc float %771 to bfloat, !dbg !67 + %861 = fptrunc float %772 to bfloat, !dbg !67 + %862 = fptrunc float %773 to bfloat, !dbg !67 + %863 = fptrunc float %774 to bfloat, !dbg !67 + %864 = fptrunc float %775 to bfloat, !dbg !67 + %865 = fptrunc float %776 to bfloat, !dbg !67 + %866 = fptrunc float %777 to bfloat, !dbg !67 + %867 = fptrunc float %778 to bfloat, !dbg !67 + %868 = fptrunc float %779 to bfloat, !dbg !67 + %869 = fptrunc float %780 to bfloat, !dbg !67 + %870 = fptrunc float %781 to bfloat, !dbg !67 + %871 = fptrunc float %782 to bfloat, !dbg !67 + %872 = fptrunc float %783 to bfloat, !dbg !67 + %873 = fptrunc float %784 to bfloat, !dbg !67 + %874 = fptrunc float %785 to bfloat, !dbg !67 + %875 = fptrunc float %786 to bfloat, !dbg !67 + %876 = fptrunc float %787 to bfloat, !dbg !67 + %877 = fptrunc float %788 to bfloat, !dbg !67 + %878 = fptrunc float %789 to bfloat, !dbg !67 + %879 = fptrunc float %790 to bfloat, !dbg !67 + %880 = fptrunc float %791 to bfloat, !dbg !67 + %881 = fptrunc float %792 to bfloat, !dbg !67 + %882 = fptrunc float %793 to bfloat, !dbg !67 + %883 = fptrunc float %794 to bfloat, !dbg !67 + %884 = fptrunc float %795 to bfloat, !dbg !67 + %885 = fptrunc float %796 to bfloat, !dbg !67 + %886 = fptrunc float %797 to bfloat, !dbg !67 + %887 = fptrunc float %798 to bfloat, !dbg !67 + %888 = fptrunc float %799 to bfloat, !dbg !67 + %889 = fptrunc float %800 to bfloat, !dbg !67 + %890 = fptrunc float %801 to bfloat, !dbg !67 + %891 = fptrunc float %802 to bfloat, !dbg !67 + %892 = fptrunc float %803 to bfloat, !dbg !67 + %893 = fptrunc float %804 to bfloat, !dbg !67 + %894 = fptrunc float %805 to bfloat, !dbg !67 + %895 = fptrunc float %806 to bfloat, !dbg !67 + %896 = fptrunc float %807 to bfloat, !dbg !67 + %897 = fptrunc float %808 to bfloat, !dbg !67 + %898 = fptrunc float %809 to bfloat, !dbg !67 + %899 = fptrunc float %810 to bfloat, !dbg !67 + %900 = fptrunc float %811 to bfloat, !dbg !67 + %901 = fptrunc float %812 to bfloat, !dbg !67 + %902 = fptrunc float %813 to bfloat, !dbg !67 + %903 = fptrunc float %814 to bfloat, !dbg !67 + %904 = insertelement <2 x bfloat> poison, bfloat %840, i64 0, !dbg !67 + %905 = insertelement <2 x bfloat> %904, bfloat %841, i64 1, !dbg !67 + %906 = bitcast <2 x bfloat> %905 to i32, !dbg !67 + %907 = insertelement <2 x bfloat> poison, bfloat %842, i64 0, !dbg !67 + %908 = insertelement <2 x bfloat> %907, bfloat %843, i64 1, !dbg !67 + %909 = bitcast <2 x bfloat> %908 to i32, !dbg !67 + %910 = insertelement <2 x bfloat> poison, bfloat %844, i64 0, !dbg !67 + %911 = insertelement <2 x bfloat> %910, bfloat %845, i64 1, !dbg !67 + %912 = bitcast <2 x bfloat> %911 to i32, !dbg !67 + %913 = insertelement <2 x bfloat> poison, bfloat %846, i64 0, !dbg !67 + %914 = insertelement <2 x bfloat> %913, bfloat %847, i64 1, !dbg !67 + %915 = bitcast <2 x bfloat> %914 to i32, !dbg !67 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %557, i32 %906, i32 %909, i32 %912, i32 %915), !dbg !67 + %916 = insertelement <2 x bfloat> poison, bfloat %872, i64 0, !dbg !67 + %917 = insertelement <2 x bfloat> %916, bfloat %873, i64 1, !dbg !67 + %918 = bitcast <2 x bfloat> %917 to i32, !dbg !67 + %919 = insertelement <2 x bfloat> poison, bfloat %874, i64 0, !dbg !67 + %920 = insertelement <2 x bfloat> %919, bfloat %875, i64 1, !dbg !67 + %921 = bitcast <2 x bfloat> %920 to i32, !dbg !67 + %922 = insertelement <2 x bfloat> poison, bfloat %876, i64 0, !dbg !67 + %923 = insertelement <2 x bfloat> %922, bfloat %877, i64 1, !dbg !67 + %924 = bitcast <2 x bfloat> %923 to i32, !dbg !67 + %925 = insertelement <2 x bfloat> poison, bfloat %878, i64 0, !dbg !67 + %926 = insertelement <2 x bfloat> %925, bfloat %879, i64 1, !dbg !67 + %927 = bitcast <2 x bfloat> %926 to i32, !dbg !67 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %559, i32 %918, i32 %921, i32 %924, i32 %927), !dbg !67 + %928 = insertelement <2 x bfloat> poison, bfloat %848, i64 0, !dbg !67 + %929 = insertelement <2 x bfloat> %928, bfloat %849, i64 1, !dbg !67 + %930 = bitcast <2 x bfloat> %929 to i32, !dbg !67 + %931 = insertelement <2 x bfloat> poison, bfloat %850, i64 0, !dbg !67 + %932 = insertelement <2 x bfloat> %931, bfloat %851, i64 1, !dbg !67 + %933 = bitcast <2 x bfloat> %932 to i32, !dbg !67 + %934 = insertelement <2 x bfloat> poison, bfloat %852, i64 0, !dbg !67 + %935 = insertelement <2 x bfloat> %934, bfloat %853, i64 1, !dbg !67 + %936 = bitcast <2 x bfloat> %935 to i32, !dbg !67 + %937 = insertelement <2 x bfloat> poison, bfloat %854, i64 0, !dbg !67 + %938 = insertelement <2 x bfloat> %937, bfloat %855, i64 1, !dbg !67 + %939 = bitcast <2 x bfloat> %938 to i32, !dbg !67 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %561, i32 %930, i32 %933, i32 %936, i32 %939), !dbg !67 + %940 = insertelement <2 x bfloat> poison, bfloat %880, i64 0, !dbg !67 + %941 = insertelement <2 x bfloat> %940, bfloat %881, i64 1, !dbg !67 + %942 = bitcast <2 x bfloat> %941 to i32, !dbg !67 + %943 = insertelement <2 x bfloat> poison, bfloat %882, i64 0, !dbg !67 + %944 = insertelement <2 x bfloat> %943, bfloat %883, i64 1, !dbg !67 + %945 = bitcast <2 x bfloat> %944 to i32, !dbg !67 + %946 = insertelement <2 x bfloat> poison, bfloat %884, i64 0, !dbg !67 + %947 = insertelement <2 x bfloat> %946, bfloat %885, i64 1, !dbg !67 + %948 = bitcast <2 x bfloat> %947 to i32, !dbg !67 + %949 = insertelement <2 x bfloat> poison, bfloat %886, i64 0, !dbg !67 + %950 = insertelement <2 x bfloat> %949, bfloat %887, i64 1, !dbg !67 + %951 = bitcast <2 x bfloat> %950 to i32, !dbg !67 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %563, i32 %942, i32 %945, i32 %948, i32 %951), !dbg !67 + %952 = insertelement <2 x bfloat> poison, bfloat %856, i64 0, !dbg !67 + %953 = insertelement <2 x bfloat> %952, bfloat %857, i64 1, !dbg !67 + %954 = bitcast <2 x bfloat> %953 to i32, !dbg !67 + %955 = insertelement <2 x bfloat> poison, bfloat %858, i64 0, !dbg !67 + %956 = insertelement <2 x bfloat> %955, bfloat %859, i64 1, !dbg !67 + %957 = bitcast <2 x bfloat> %956 to i32, !dbg !67 + %958 = insertelement <2 x bfloat> poison, bfloat %860, i64 0, !dbg !67 + %959 = insertelement <2 x bfloat> %958, bfloat %861, i64 1, !dbg !67 + %960 = bitcast <2 x bfloat> %959 to i32, !dbg !67 + %961 = insertelement <2 x bfloat> poison, bfloat %862, i64 0, !dbg !67 + %962 = insertelement <2 x bfloat> %961, bfloat %863, i64 1, !dbg !67 + %963 = bitcast <2 x bfloat> %962 to i32, !dbg !67 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %565, i32 %954, i32 %957, i32 %960, i32 %963), !dbg !67 + %964 = insertelement <2 x bfloat> poison, bfloat %888, i64 0, !dbg !67 + %965 = insertelement <2 x bfloat> %964, bfloat %889, i64 1, !dbg !67 + %966 = bitcast <2 x bfloat> %965 to i32, !dbg !67 + %967 = insertelement <2 x bfloat> poison, bfloat %890, i64 0, !dbg !67 + %968 = insertelement <2 x bfloat> %967, bfloat %891, i64 1, !dbg !67 + %969 = bitcast <2 x bfloat> %968 to i32, !dbg !67 + %970 = insertelement <2 x bfloat> poison, bfloat %892, i64 0, !dbg !67 + %971 = insertelement <2 x bfloat> %970, bfloat %893, i64 1, !dbg !67 + %972 = bitcast <2 x bfloat> %971 to i32, !dbg !67 + %973 = insertelement <2 x bfloat> poison, bfloat %894, i64 0, !dbg !67 + %974 = insertelement <2 x bfloat> %973, bfloat %895, i64 1, !dbg !67 + %975 = bitcast <2 x bfloat> %974 to i32, !dbg !67 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %567, i32 %966, i32 %969, i32 %972, i32 %975), !dbg !67 + %976 = insertelement <2 x bfloat> poison, bfloat %864, i64 0, !dbg !67 + %977 = insertelement <2 x bfloat> %976, bfloat %865, i64 1, !dbg !67 + %978 = bitcast <2 x bfloat> %977 to i32, !dbg !67 + %979 = insertelement <2 x bfloat> poison, bfloat %866, i64 0, !dbg !67 + %980 = insertelement <2 x bfloat> %979, bfloat %867, i64 1, !dbg !67 + %981 = bitcast <2 x bfloat> %980 to i32, !dbg !67 + %982 = insertelement <2 x bfloat> poison, bfloat %868, i64 0, !dbg !67 + %983 = insertelement <2 x bfloat> %982, bfloat %869, i64 1, !dbg !67 + %984 = bitcast <2 x bfloat> %983 to i32, !dbg !67 + %985 = insertelement <2 x bfloat> poison, bfloat %870, i64 0, !dbg !67 + %986 = insertelement <2 x bfloat> %985, bfloat %871, i64 1, !dbg !67 + %987 = bitcast <2 x bfloat> %986 to i32, !dbg !67 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %569, i32 %978, i32 %981, i32 %984, i32 %987), !dbg !67 + %988 = insertelement <2 x bfloat> poison, bfloat %896, i64 0, !dbg !67 + %989 = insertelement <2 x bfloat> %988, bfloat %897, i64 1, !dbg !67 + %990 = bitcast <2 x bfloat> %989 to i32, !dbg !67 + %991 = insertelement <2 x bfloat> poison, bfloat %898, i64 0, !dbg !67 + %992 = insertelement <2 x bfloat> %991, bfloat %899, i64 1, !dbg !67 + %993 = bitcast <2 x bfloat> %992 to i32, !dbg !67 + %994 = insertelement <2 x bfloat> poison, bfloat %900, i64 0, !dbg !67 + %995 = insertelement <2 x bfloat> %994, bfloat %901, i64 1, !dbg !67 + %996 = bitcast <2 x bfloat> %995 to i32, !dbg !67 + %997 = insertelement <2 x bfloat> poison, bfloat %902, i64 0, !dbg !67 + %998 = insertelement <2 x bfloat> %997, bfloat %903, i64 1, !dbg !67 + %999 = bitcast <2 x bfloat> %998 to i32, !dbg !67 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %571, i32 %990, i32 %993, i32 %996, i32 %999), !dbg !67 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %589, i32 %906, i32 %912, i32 %930, i32 %936) #2, !dbg !68 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %591, i32 %954, i32 %960, i32 %978, i32 %984) #2, !dbg !68 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !68 + %1000 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %600), !dbg !68 + %1001 = extractvalue { i32, i32, i32, i32 } %1000, 0, !dbg !68 + %1002 = extractvalue { i32, i32, i32, i32 } %1000, 1, !dbg !68 + %1003 = extractvalue { i32, i32, i32, i32 } %1000, 2, !dbg !68 + %1004 = extractvalue { i32, i32, i32, i32 } %1000, 3, !dbg !68 + %1005 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %602), !dbg !68 + %1006 = extractvalue { i32, i32, i32, i32 } %1005, 0, !dbg !68 + %1007 = extractvalue { i32, i32, i32, i32 } %1005, 1, !dbg !68 + %1008 = extractvalue { i32, i32, i32, i32 } %1005, 2, !dbg !68 + %1009 = extractvalue { i32, i32, i32, i32 } %1005, 3, !dbg !68 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !68 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %589, i32 %909, i32 %915, i32 %933, i32 %939) #2, !dbg !68 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %591, i32 %957, i32 %963, i32 %981, i32 %987) #2, !dbg !68 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !68 + %1010 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %600), !dbg !68 + %1011 = extractvalue { i32, i32, i32, i32 } %1010, 0, !dbg !68 + %1012 = extractvalue { i32, i32, i32, i32 } %1010, 1, !dbg !68 + %1013 = extractvalue { i32, i32, i32, i32 } %1010, 2, !dbg !68 + %1014 = extractvalue { i32, i32, i32, i32 } %1010, 3, !dbg !68 + %1015 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %602), !dbg !68 + %1016 = extractvalue { i32, i32, i32, i32 } %1015, 0, !dbg !68 + %1017 = extractvalue { i32, i32, i32, i32 } %1015, 1, !dbg !68 + %1018 = extractvalue { i32, i32, i32, i32 } %1015, 2, !dbg !68 + %1019 = extractvalue { i32, i32, i32, i32 } %1015, 3, !dbg !68 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !68 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %589, i32 %918, i32 %924, i32 %942, i32 %948) #2, !dbg !68 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %591, i32 %966, i32 %972, i32 %990, i32 %996) #2, !dbg !68 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !68 + %1020 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %600), !dbg !68 + %1021 = extractvalue { i32, i32, i32, i32 } %1020, 0, !dbg !68 + %1022 = extractvalue { i32, i32, i32, i32 } %1020, 1, !dbg !68 + %1023 = extractvalue { i32, i32, i32, i32 } %1020, 2, !dbg !68 + %1024 = extractvalue { i32, i32, i32, i32 } %1020, 3, !dbg !68 + %1025 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %602), !dbg !68 + %1026 = extractvalue { i32, i32, i32, i32 } %1025, 0, !dbg !68 + %1027 = extractvalue { i32, i32, i32, i32 } %1025, 1, !dbg !68 + %1028 = extractvalue { i32, i32, i32, i32 } %1025, 2, !dbg !68 + %1029 = extractvalue { i32, i32, i32, i32 } %1025, 3, !dbg !68 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !68 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %589, i32 %921, i32 %927, i32 %945, i32 %951) #2, !dbg !68 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %591, i32 %969, i32 %975, i32 %993, i32 %999) #2, !dbg !68 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !68 + %1030 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %600), !dbg !68 + %1031 = extractvalue { i32, i32, i32, i32 } %1030, 0, !dbg !68 + %1032 = extractvalue { i32, i32, i32, i32 } %1030, 1, !dbg !68 + %1033 = extractvalue { i32, i32, i32, i32 } %1030, 2, !dbg !68 + %1034 = extractvalue { i32, i32, i32, i32 } %1030, 3, !dbg !68 + %1035 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %602), !dbg !68 + %1036 = extractvalue { i32, i32, i32, i32 } %1035, 0, !dbg !68 + %1037 = extractvalue { i32, i32, i32, i32 } %1035, 1, !dbg !68 + %1038 = extractvalue { i32, i32, i32, i32 } %1035, 2, !dbg !68 + %1039 = extractvalue { i32, i32, i32, i32 } %1035, 3, !dbg !68 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1001, i32 %1002, i32 %1003, i32 %1004, ptr addrspace(1) %832, i1 true) #2, !dbg !68 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1011, i32 %1012, i32 %1013, i32 %1014, ptr addrspace(1) %833, i1 true) #2, !dbg !68 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1006, i32 %1007, i32 %1008, i32 %1009, ptr addrspace(1) %834, i1 true) #2, !dbg !68 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1016, i32 %1017, i32 %1018, i32 %1019, ptr addrspace(1) %835, i1 true) #2, !dbg !68 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1021, i32 %1022, i32 %1023, i32 %1024, ptr addrspace(1) %836, i1 true) #2, !dbg !68 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1031, i32 %1032, i32 %1033, i32 %1034, ptr addrspace(1) %837, i1 true) #2, !dbg !68 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1026, i32 %1027, i32 %1028, i32 %1029, ptr addrspace(1) %838, i1 true) #2, !dbg !68 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1036, i32 %1037, i32 %1038, i32 %1039, ptr addrspace(1) %839, i1 true) #2, !dbg !68 + %1040 = getelementptr i8, ptr addrspace(1) %824, i64 16384, !dbg !69 + %1041 = getelementptr i8, ptr addrspace(1) %824, i64 18432, !dbg !69 + %1042 = getelementptr i8, ptr addrspace(1) %824, i64 20480, !dbg !69 + %1043 = getelementptr i8, ptr addrspace(1) %824, i64 22528, !dbg !69 + %1044 = getelementptr i8, ptr addrspace(1) %824, i64 24576, !dbg !69 + %1045 = getelementptr i8, ptr addrspace(1) %824, i64 26624, !dbg !69 + %1046 = getelementptr i8, ptr addrspace(1) %824, i64 28672, !dbg !69 + %1047 = getelementptr i8, ptr addrspace(1) %824, i64 30720, !dbg !69 + %1048 = getelementptr bfloat, ptr addrspace(1) %1040, i64 %63, !dbg !70 + %1049 = getelementptr bfloat, ptr addrspace(1) %1041, i64 %63, !dbg !70 + %1050 = getelementptr bfloat, ptr addrspace(1) %1042, i64 %63, !dbg !70 + %1051 = getelementptr bfloat, ptr addrspace(1) %1043, i64 %63, !dbg !70 + %1052 = getelementptr bfloat, ptr addrspace(1) %1044, i64 %63, !dbg !70 + %1053 = getelementptr bfloat, ptr addrspace(1) %1045, i64 %63, !dbg !70 + %1054 = getelementptr bfloat, ptr addrspace(1) %1046, i64 %63, !dbg !70 + %1055 = getelementptr bfloat, ptr addrspace(1) %1047, i64 %63, !dbg !70 + %1056 = fptrunc float %687 to bfloat, !dbg !71 + %1057 = fptrunc float %688 to bfloat, !dbg !71 + %1058 = fptrunc float %689 to bfloat, !dbg !71 + %1059 = fptrunc float %690 to bfloat, !dbg !71 + %1060 = fptrunc float %691 to bfloat, !dbg !71 + %1061 = fptrunc float %692 to bfloat, !dbg !71 + %1062 = fptrunc float %693 to bfloat, !dbg !71 + %1063 = fptrunc float %694 to bfloat, !dbg !71 + %1064 = fptrunc float %695 to bfloat, !dbg !71 + %1065 = fptrunc float %696 to bfloat, !dbg !71 + %1066 = fptrunc float %697 to bfloat, !dbg !71 + %1067 = fptrunc float %698 to bfloat, !dbg !71 + %1068 = fptrunc float %699 to bfloat, !dbg !71 + %1069 = fptrunc float %700 to bfloat, !dbg !71 + %1070 = fptrunc float %701 to bfloat, !dbg !71 + %1071 = fptrunc float %702 to bfloat, !dbg !71 + %1072 = fptrunc float %703 to bfloat, !dbg !71 + %1073 = fptrunc float %704 to bfloat, !dbg !71 + %1074 = fptrunc float %705 to bfloat, !dbg !71 + %1075 = fptrunc float %706 to bfloat, !dbg !71 + %1076 = fptrunc float %707 to bfloat, !dbg !71 + %1077 = fptrunc float %708 to bfloat, !dbg !71 + %1078 = fptrunc float %709 to bfloat, !dbg !71 + %1079 = fptrunc float %710 to bfloat, !dbg !71 + %1080 = fptrunc float %711 to bfloat, !dbg !71 + %1081 = fptrunc float %712 to bfloat, !dbg !71 + %1082 = fptrunc float %713 to bfloat, !dbg !71 + %1083 = fptrunc float %714 to bfloat, !dbg !71 + %1084 = fptrunc float %715 to bfloat, !dbg !71 + %1085 = fptrunc float %716 to bfloat, !dbg !71 + %1086 = fptrunc float %717 to bfloat, !dbg !71 + %1087 = fptrunc float %718 to bfloat, !dbg !71 + %1088 = fptrunc float %719 to bfloat, !dbg !71 + %1089 = fptrunc float %720 to bfloat, !dbg !71 + %1090 = fptrunc float %721 to bfloat, !dbg !71 + %1091 = fptrunc float %722 to bfloat, !dbg !71 + %1092 = fptrunc float %723 to bfloat, !dbg !71 + %1093 = fptrunc float %724 to bfloat, !dbg !71 + %1094 = fptrunc float %725 to bfloat, !dbg !71 + %1095 = fptrunc float %726 to bfloat, !dbg !71 + %1096 = fptrunc float %727 to bfloat, !dbg !71 + %1097 = fptrunc float %728 to bfloat, !dbg !71 + %1098 = fptrunc float %729 to bfloat, !dbg !71 + %1099 = fptrunc float %730 to bfloat, !dbg !71 + %1100 = fptrunc float %731 to bfloat, !dbg !71 + %1101 = fptrunc float %732 to bfloat, !dbg !71 + %1102 = fptrunc float %733 to bfloat, !dbg !71 + %1103 = fptrunc float %734 to bfloat, !dbg !71 + %1104 = fptrunc float %735 to bfloat, !dbg !71 + %1105 = fptrunc float %736 to bfloat, !dbg !71 + %1106 = fptrunc float %737 to bfloat, !dbg !71 + %1107 = fptrunc float %738 to bfloat, !dbg !71 + %1108 = fptrunc float %739 to bfloat, !dbg !71 + %1109 = fptrunc float %740 to bfloat, !dbg !71 + %1110 = fptrunc float %741 to bfloat, !dbg !71 + %1111 = fptrunc float %742 to bfloat, !dbg !71 + %1112 = fptrunc float %743 to bfloat, !dbg !71 + %1113 = fptrunc float %744 to bfloat, !dbg !71 + %1114 = fptrunc float %745 to bfloat, !dbg !71 + %1115 = fptrunc float %746 to bfloat, !dbg !71 + %1116 = fptrunc float %747 to bfloat, !dbg !71 + %1117 = fptrunc float %748 to bfloat, !dbg !71 + %1118 = fptrunc float %749 to bfloat, !dbg !71 + %1119 = fptrunc float %750 to bfloat, !dbg !71 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !71 + %1120 = insertelement <2 x bfloat> poison, bfloat %1056, i64 0, !dbg !71 + %1121 = insertelement <2 x bfloat> %1120, bfloat %1057, i64 1, !dbg !71 + %1122 = bitcast <2 x bfloat> %1121 to i32, !dbg !71 + %1123 = insertelement <2 x bfloat> poison, bfloat %1058, i64 0, !dbg !71 + %1124 = insertelement <2 x bfloat> %1123, bfloat %1059, i64 1, !dbg !71 + %1125 = bitcast <2 x bfloat> %1124 to i32, !dbg !71 + %1126 = insertelement <2 x bfloat> poison, bfloat %1060, i64 0, !dbg !71 + %1127 = insertelement <2 x bfloat> %1126, bfloat %1061, i64 1, !dbg !71 + %1128 = bitcast <2 x bfloat> %1127 to i32, !dbg !71 + %1129 = insertelement <2 x bfloat> poison, bfloat %1062, i64 0, !dbg !71 + %1130 = insertelement <2 x bfloat> %1129, bfloat %1063, i64 1, !dbg !71 + %1131 = bitcast <2 x bfloat> %1130 to i32, !dbg !71 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %603, i32 %1122, i32 %1125, i32 %1128, i32 %1131), !dbg !71 + %1132 = insertelement <2 x bfloat> poison, bfloat %1088, i64 0, !dbg !71 + %1133 = insertelement <2 x bfloat> %1132, bfloat %1089, i64 1, !dbg !71 + %1134 = bitcast <2 x bfloat> %1133 to i32, !dbg !71 + %1135 = insertelement <2 x bfloat> poison, bfloat %1090, i64 0, !dbg !71 + %1136 = insertelement <2 x bfloat> %1135, bfloat %1091, i64 1, !dbg !71 + %1137 = bitcast <2 x bfloat> %1136 to i32, !dbg !71 + %1138 = insertelement <2 x bfloat> poison, bfloat %1092, i64 0, !dbg !71 + %1139 = insertelement <2 x bfloat> %1138, bfloat %1093, i64 1, !dbg !71 + %1140 = bitcast <2 x bfloat> %1139 to i32, !dbg !71 + %1141 = insertelement <2 x bfloat> poison, bfloat %1094, i64 0, !dbg !71 + %1142 = insertelement <2 x bfloat> %1141, bfloat %1095, i64 1, !dbg !71 + %1143 = bitcast <2 x bfloat> %1142 to i32, !dbg !71 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %604, i32 %1134, i32 %1137, i32 %1140, i32 %1143), !dbg !71 + %1144 = insertelement <2 x bfloat> poison, bfloat %1064, i64 0, !dbg !71 + %1145 = insertelement <2 x bfloat> %1144, bfloat %1065, i64 1, !dbg !71 + %1146 = bitcast <2 x bfloat> %1145 to i32, !dbg !71 + %1147 = insertelement <2 x bfloat> poison, bfloat %1066, i64 0, !dbg !71 + %1148 = insertelement <2 x bfloat> %1147, bfloat %1067, i64 1, !dbg !71 + %1149 = bitcast <2 x bfloat> %1148 to i32, !dbg !71 + %1150 = insertelement <2 x bfloat> poison, bfloat %1068, i64 0, !dbg !71 + %1151 = insertelement <2 x bfloat> %1150, bfloat %1069, i64 1, !dbg !71 + %1152 = bitcast <2 x bfloat> %1151 to i32, !dbg !71 + %1153 = insertelement <2 x bfloat> poison, bfloat %1070, i64 0, !dbg !71 + %1154 = insertelement <2 x bfloat> %1153, bfloat %1071, i64 1, !dbg !71 + %1155 = bitcast <2 x bfloat> %1154 to i32, !dbg !71 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %605, i32 %1146, i32 %1149, i32 %1152, i32 %1155), !dbg !71 + %1156 = insertelement <2 x bfloat> poison, bfloat %1096, i64 0, !dbg !71 + %1157 = insertelement <2 x bfloat> %1156, bfloat %1097, i64 1, !dbg !71 + %1158 = bitcast <2 x bfloat> %1157 to i32, !dbg !71 + %1159 = insertelement <2 x bfloat> poison, bfloat %1098, i64 0, !dbg !71 + %1160 = insertelement <2 x bfloat> %1159, bfloat %1099, i64 1, !dbg !71 + %1161 = bitcast <2 x bfloat> %1160 to i32, !dbg !71 + %1162 = insertelement <2 x bfloat> poison, bfloat %1100, i64 0, !dbg !71 + %1163 = insertelement <2 x bfloat> %1162, bfloat %1101, i64 1, !dbg !71 + %1164 = bitcast <2 x bfloat> %1163 to i32, !dbg !71 + %1165 = insertelement <2 x bfloat> poison, bfloat %1102, i64 0, !dbg !71 + %1166 = insertelement <2 x bfloat> %1165, bfloat %1103, i64 1, !dbg !71 + %1167 = bitcast <2 x bfloat> %1166 to i32, !dbg !71 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %606, i32 %1158, i32 %1161, i32 %1164, i32 %1167), !dbg !71 + %1168 = insertelement <2 x bfloat> poison, bfloat %1072, i64 0, !dbg !71 + %1169 = insertelement <2 x bfloat> %1168, bfloat %1073, i64 1, !dbg !71 + %1170 = bitcast <2 x bfloat> %1169 to i32, !dbg !71 + %1171 = insertelement <2 x bfloat> poison, bfloat %1074, i64 0, !dbg !71 + %1172 = insertelement <2 x bfloat> %1171, bfloat %1075, i64 1, !dbg !71 + %1173 = bitcast <2 x bfloat> %1172 to i32, !dbg !71 + %1174 = insertelement <2 x bfloat> poison, bfloat %1076, i64 0, !dbg !71 + %1175 = insertelement <2 x bfloat> %1174, bfloat %1077, i64 1, !dbg !71 + %1176 = bitcast <2 x bfloat> %1175 to i32, !dbg !71 + %1177 = insertelement <2 x bfloat> poison, bfloat %1078, i64 0, !dbg !71 + %1178 = insertelement <2 x bfloat> %1177, bfloat %1079, i64 1, !dbg !71 + %1179 = bitcast <2 x bfloat> %1178 to i32, !dbg !71 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %607, i32 %1170, i32 %1173, i32 %1176, i32 %1179), !dbg !71 + %1180 = insertelement <2 x bfloat> poison, bfloat %1104, i64 0, !dbg !71 + %1181 = insertelement <2 x bfloat> %1180, bfloat %1105, i64 1, !dbg !71 + %1182 = bitcast <2 x bfloat> %1181 to i32, !dbg !71 + %1183 = insertelement <2 x bfloat> poison, bfloat %1106, i64 0, !dbg !71 + %1184 = insertelement <2 x bfloat> %1183, bfloat %1107, i64 1, !dbg !71 + %1185 = bitcast <2 x bfloat> %1184 to i32, !dbg !71 + %1186 = insertelement <2 x bfloat> poison, bfloat %1108, i64 0, !dbg !71 + %1187 = insertelement <2 x bfloat> %1186, bfloat %1109, i64 1, !dbg !71 + %1188 = bitcast <2 x bfloat> %1187 to i32, !dbg !71 + %1189 = insertelement <2 x bfloat> poison, bfloat %1110, i64 0, !dbg !71 + %1190 = insertelement <2 x bfloat> %1189, bfloat %1111, i64 1, !dbg !71 + %1191 = bitcast <2 x bfloat> %1190 to i32, !dbg !71 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %608, i32 %1182, i32 %1185, i32 %1188, i32 %1191), !dbg !71 + %1192 = insertelement <2 x bfloat> poison, bfloat %1080, i64 0, !dbg !71 + %1193 = insertelement <2 x bfloat> %1192, bfloat %1081, i64 1, !dbg !71 + %1194 = bitcast <2 x bfloat> %1193 to i32, !dbg !71 + %1195 = insertelement <2 x bfloat> poison, bfloat %1082, i64 0, !dbg !71 + %1196 = insertelement <2 x bfloat> %1195, bfloat %1083, i64 1, !dbg !71 + %1197 = bitcast <2 x bfloat> %1196 to i32, !dbg !71 + %1198 = insertelement <2 x bfloat> poison, bfloat %1084, i64 0, !dbg !71 + %1199 = insertelement <2 x bfloat> %1198, bfloat %1085, i64 1, !dbg !71 + %1200 = bitcast <2 x bfloat> %1199 to i32, !dbg !71 + %1201 = insertelement <2 x bfloat> poison, bfloat %1086, i64 0, !dbg !71 + %1202 = insertelement <2 x bfloat> %1201, bfloat %1087, i64 1, !dbg !71 + %1203 = bitcast <2 x bfloat> %1202 to i32, !dbg !71 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %609, i32 %1194, i32 %1197, i32 %1200, i32 %1203), !dbg !71 + %1204 = insertelement <2 x bfloat> poison, bfloat %1112, i64 0, !dbg !71 + %1205 = insertelement <2 x bfloat> %1204, bfloat %1113, i64 1, !dbg !71 + %1206 = bitcast <2 x bfloat> %1205 to i32, !dbg !71 + %1207 = insertelement <2 x bfloat> poison, bfloat %1114, i64 0, !dbg !71 + %1208 = insertelement <2 x bfloat> %1207, bfloat %1115, i64 1, !dbg !71 + %1209 = bitcast <2 x bfloat> %1208 to i32, !dbg !71 + %1210 = insertelement <2 x bfloat> poison, bfloat %1116, i64 0, !dbg !71 + %1211 = insertelement <2 x bfloat> %1210, bfloat %1117, i64 1, !dbg !71 + %1212 = bitcast <2 x bfloat> %1211 to i32, !dbg !71 + %1213 = insertelement <2 x bfloat> poison, bfloat %1118, i64 0, !dbg !71 + %1214 = insertelement <2 x bfloat> %1213, bfloat %1119, i64 1, !dbg !71 + %1215 = bitcast <2 x bfloat> %1214 to i32, !dbg !71 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %610, i32 %1206, i32 %1209, i32 %1212, i32 %1215), !dbg !71 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %611, i32 %1122, i32 %1128, i32 %1146, i32 %1152) #2, !dbg !72 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %612, i32 %1170, i32 %1176, i32 %1194, i32 %1200) #2, !dbg !72 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !72 + %1216 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %613), !dbg !72 + %1217 = extractvalue { i32, i32, i32, i32 } %1216, 0, !dbg !72 + %1218 = extractvalue { i32, i32, i32, i32 } %1216, 1, !dbg !72 + %1219 = extractvalue { i32, i32, i32, i32 } %1216, 2, !dbg !72 + %1220 = extractvalue { i32, i32, i32, i32 } %1216, 3, !dbg !72 + %1221 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %614), !dbg !72 + %1222 = extractvalue { i32, i32, i32, i32 } %1221, 0, !dbg !72 + %1223 = extractvalue { i32, i32, i32, i32 } %1221, 1, !dbg !72 + %1224 = extractvalue { i32, i32, i32, i32 } %1221, 2, !dbg !72 + %1225 = extractvalue { i32, i32, i32, i32 } %1221, 3, !dbg !72 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !72 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %611, i32 %1125, i32 %1131, i32 %1149, i32 %1155) #2, !dbg !72 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %612, i32 %1173, i32 %1179, i32 %1197, i32 %1203) #2, !dbg !72 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !72 + %1226 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %613), !dbg !72 + %1227 = extractvalue { i32, i32, i32, i32 } %1226, 0, !dbg !72 + %1228 = extractvalue { i32, i32, i32, i32 } %1226, 1, !dbg !72 + %1229 = extractvalue { i32, i32, i32, i32 } %1226, 2, !dbg !72 + %1230 = extractvalue { i32, i32, i32, i32 } %1226, 3, !dbg !72 + %1231 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %614), !dbg !72 + %1232 = extractvalue { i32, i32, i32, i32 } %1231, 0, !dbg !72 + %1233 = extractvalue { i32, i32, i32, i32 } %1231, 1, !dbg !72 + %1234 = extractvalue { i32, i32, i32, i32 } %1231, 2, !dbg !72 + %1235 = extractvalue { i32, i32, i32, i32 } %1231, 3, !dbg !72 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !72 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %611, i32 %1134, i32 %1140, i32 %1158, i32 %1164) #2, !dbg !72 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %612, i32 %1182, i32 %1188, i32 %1206, i32 %1212) #2, !dbg !72 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !72 + %1236 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %613), !dbg !72 + %1237 = extractvalue { i32, i32, i32, i32 } %1236, 0, !dbg !72 + %1238 = extractvalue { i32, i32, i32, i32 } %1236, 1, !dbg !72 + %1239 = extractvalue { i32, i32, i32, i32 } %1236, 2, !dbg !72 + %1240 = extractvalue { i32, i32, i32, i32 } %1236, 3, !dbg !72 + %1241 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %614), !dbg !72 + %1242 = extractvalue { i32, i32, i32, i32 } %1241, 0, !dbg !72 + %1243 = extractvalue { i32, i32, i32, i32 } %1241, 1, !dbg !72 + %1244 = extractvalue { i32, i32, i32, i32 } %1241, 2, !dbg !72 + %1245 = extractvalue { i32, i32, i32, i32 } %1241, 3, !dbg !72 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !72 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %611, i32 %1137, i32 %1143, i32 %1161, i32 %1167) #2, !dbg !72 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %612, i32 %1185, i32 %1191, i32 %1209, i32 %1215) #2, !dbg !72 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !72 + %1246 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %613), !dbg !72 + %1247 = extractvalue { i32, i32, i32, i32 } %1246, 0, !dbg !72 + %1248 = extractvalue { i32, i32, i32, i32 } %1246, 1, !dbg !72 + %1249 = extractvalue { i32, i32, i32, i32 } %1246, 2, !dbg !72 + %1250 = extractvalue { i32, i32, i32, i32 } %1246, 3, !dbg !72 + %1251 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %614), !dbg !72 + %1252 = extractvalue { i32, i32, i32, i32 } %1251, 0, !dbg !72 + %1253 = extractvalue { i32, i32, i32, i32 } %1251, 1, !dbg !72 + %1254 = extractvalue { i32, i32, i32, i32 } %1251, 2, !dbg !72 + %1255 = extractvalue { i32, i32, i32, i32 } %1251, 3, !dbg !72 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1217, i32 %1218, i32 %1219, i32 %1220, ptr addrspace(1) %1048, i1 true) #2, !dbg !72 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1227, i32 %1228, i32 %1229, i32 %1230, ptr addrspace(1) %1049, i1 true) #2, !dbg !72 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1222, i32 %1223, i32 %1224, i32 %1225, ptr addrspace(1) %1050, i1 true) #2, !dbg !72 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1232, i32 %1233, i32 %1234, i32 %1235, ptr addrspace(1) %1051, i1 true) #2, !dbg !72 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1237, i32 %1238, i32 %1239, i32 %1240, ptr addrspace(1) %1052, i1 true) #2, !dbg !72 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1247, i32 %1248, i32 %1249, i32 %1250, ptr addrspace(1) %1053, i1 true) #2, !dbg !72 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1242, i32 %1243, i32 %1244, i32 %1245, ptr addrspace(1) %1054, i1 true) #2, !dbg !72 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1252, i32 %1253, i32 %1254, i32 %1255, ptr addrspace(1) %1055, i1 true) #2, !dbg !72 + tail call void @llvm.nvvm.cp.async.wait.group(i32 14), !dbg !43 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !43 + %1256 = shl i32 %822, 12, !dbg !43 + %1257 = getelementptr bfloat, ptr addrspace(3) @global_smem, i32 %1256, !dbg !43 + %1258 = getelementptr inbounds nuw i8, ptr addrspace(3) %1257, i32 %556, !dbg !43 + %1259 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %1258), !dbg !43 + %1260 = extractvalue { i32, i32, i32, i32 } %1259, 0, !dbg !43 + %1261 = extractvalue { i32, i32, i32, i32 } %1259, 1, !dbg !43 + %1262 = extractvalue { i32, i32, i32, i32 } %1259, 2, !dbg !43 + %1263 = extractvalue { i32, i32, i32, i32 } %1259, 3, !dbg !43 + %1264 = getelementptr inbounds nuw i8, ptr addrspace(3) %1257, i32 %558, !dbg !43 + %1265 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %1264), !dbg !43 + %1266 = extractvalue { i32, i32, i32, i32 } %1265, 0, !dbg !43 + %1267 = extractvalue { i32, i32, i32, i32 } %1265, 1, !dbg !43 + %1268 = extractvalue { i32, i32, i32, i32 } %1265, 2, !dbg !43 + %1269 = extractvalue { i32, i32, i32, i32 } %1265, 3, !dbg !43 + %1270 = getelementptr inbounds nuw i8, ptr addrspace(3) %1257, i32 %560, !dbg !43 + %1271 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %1270), !dbg !43 + %1272 = extractvalue { i32, i32, i32, i32 } %1271, 0, !dbg !43 + %1273 = extractvalue { i32, i32, i32, i32 } %1271, 1, !dbg !43 + %1274 = extractvalue { i32, i32, i32, i32 } %1271, 2, !dbg !43 + %1275 = extractvalue { i32, i32, i32, i32 } %1271, 3, !dbg !43 + %1276 = getelementptr inbounds nuw i8, ptr addrspace(3) %1257, i32 %562, !dbg !43 + %1277 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %1276), !dbg !43 + %1278 = extractvalue { i32, i32, i32, i32 } %1277, 0, !dbg !43 + %1279 = extractvalue { i32, i32, i32, i32 } %1277, 1, !dbg !43 + %1280 = extractvalue { i32, i32, i32, i32 } %1277, 2, !dbg !43 + %1281 = extractvalue { i32, i32, i32, i32 } %1277, 3, !dbg !43 + %1282 = getelementptr inbounds nuw i8, ptr addrspace(3) %1257, i32 %564, !dbg !43 + %1283 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %1282), !dbg !43 + %1284 = extractvalue { i32, i32, i32, i32 } %1283, 0, !dbg !43 + %1285 = extractvalue { i32, i32, i32, i32 } %1283, 1, !dbg !43 + %1286 = extractvalue { i32, i32, i32, i32 } %1283, 2, !dbg !43 + %1287 = extractvalue { i32, i32, i32, i32 } %1283, 3, !dbg !43 + %1288 = getelementptr inbounds nuw i8, ptr addrspace(3) %1257, i32 %566, !dbg !43 + %1289 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %1288), !dbg !43 + %1290 = extractvalue { i32, i32, i32, i32 } %1289, 0, !dbg !43 + %1291 = extractvalue { i32, i32, i32, i32 } %1289, 1, !dbg !43 + %1292 = extractvalue { i32, i32, i32, i32 } %1289, 2, !dbg !43 + %1293 = extractvalue { i32, i32, i32, i32 } %1289, 3, !dbg !43 + %1294 = getelementptr inbounds nuw i8, ptr addrspace(3) %1257, i32 %568, !dbg !43 + %1295 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %1294), !dbg !43 + %1296 = extractvalue { i32, i32, i32, i32 } %1295, 0, !dbg !43 + %1297 = extractvalue { i32, i32, i32, i32 } %1295, 1, !dbg !43 + %1298 = extractvalue { i32, i32, i32, i32 } %1295, 2, !dbg !43 + %1299 = extractvalue { i32, i32, i32, i32 } %1295, 3, !dbg !43 + %1300 = getelementptr inbounds nuw i8, ptr addrspace(3) %1257, i32 %570, !dbg !43 + %1301 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %1300), !dbg !43 + %1302 = extractvalue { i32, i32, i32, i32 } %1301, 0, !dbg !43 + %1303 = extractvalue { i32, i32, i32, i32 } %1301, 1, !dbg !43 + %1304 = extractvalue { i32, i32, i32, i32 } %1301, 2, !dbg !43 + %1305 = extractvalue { i32, i32, i32, i32 } %1301, 3, !dbg !43 + %1306 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %617), !dbg !67 + %1307 = extractvalue { i32, i32, i32, i32 } %1306, 0, !dbg !67 + %1308 = extractvalue { i32, i32, i32, i32 } %1306, 1, !dbg !67 + %1309 = extractvalue { i32, i32, i32, i32 } %1306, 2, !dbg !67 + %1310 = extractvalue { i32, i32, i32, i32 } %1306, 3, !dbg !67 + %1311 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %619), !dbg !67 + %1312 = extractvalue { i32, i32, i32, i32 } %1311, 0, !dbg !67 + %1313 = extractvalue { i32, i32, i32, i32 } %1311, 1, !dbg !67 + %1314 = extractvalue { i32, i32, i32, i32 } %1311, 2, !dbg !67 + %1315 = extractvalue { i32, i32, i32, i32 } %1311, 3, !dbg !67 + %1316 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %621), !dbg !67 + %1317 = extractvalue { i32, i32, i32, i32 } %1316, 0, !dbg !67 + %1318 = extractvalue { i32, i32, i32, i32 } %1316, 1, !dbg !67 + %1319 = extractvalue { i32, i32, i32, i32 } %1316, 2, !dbg !67 + %1320 = extractvalue { i32, i32, i32, i32 } %1316, 3, !dbg !67 + %1321 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %623), !dbg !67 + %1322 = extractvalue { i32, i32, i32, i32 } %1321, 0, !dbg !67 + %1323 = extractvalue { i32, i32, i32, i32 } %1321, 1, !dbg !67 + %1324 = extractvalue { i32, i32, i32, i32 } %1321, 2, !dbg !67 + %1325 = extractvalue { i32, i32, i32, i32 } %1321, 3, !dbg !67 + %1326 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %625), !dbg !67 + %1327 = extractvalue { i32, i32, i32, i32 } %1326, 0, !dbg !67 + %1328 = extractvalue { i32, i32, i32, i32 } %1326, 1, !dbg !67 + %1329 = extractvalue { i32, i32, i32, i32 } %1326, 2, !dbg !67 + %1330 = extractvalue { i32, i32, i32, i32 } %1326, 3, !dbg !67 + %1331 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %627), !dbg !67 + %1332 = extractvalue { i32, i32, i32, i32 } %1331, 0, !dbg !67 + %1333 = extractvalue { i32, i32, i32, i32 } %1331, 1, !dbg !67 + %1334 = extractvalue { i32, i32, i32, i32 } %1331, 2, !dbg !67 + %1335 = extractvalue { i32, i32, i32, i32 } %1331, 3, !dbg !67 + %1336 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %629), !dbg !67 + %1337 = extractvalue { i32, i32, i32, i32 } %1336, 0, !dbg !67 + %1338 = extractvalue { i32, i32, i32, i32 } %1336, 1, !dbg !67 + %1339 = extractvalue { i32, i32, i32, i32 } %1336, 2, !dbg !67 + %1340 = extractvalue { i32, i32, i32, i32 } %1336, 3, !dbg !67 + %1341 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %631), !dbg !67 + %1342 = extractvalue { i32, i32, i32, i32 } %1341, 0, !dbg !67 + %1343 = extractvalue { i32, i32, i32, i32 } %1341, 1, !dbg !67 + %1344 = extractvalue { i32, i32, i32, i32 } %1341, 2, !dbg !67 + %1345 = extractvalue { i32, i32, i32, i32 } %1341, 3, !dbg !67 + %1346 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %633), !dbg !67 + %1347 = extractvalue { i32, i32, i32, i32 } %1346, 0, !dbg !67 + %1348 = extractvalue { i32, i32, i32, i32 } %1346, 1, !dbg !67 + %1349 = extractvalue { i32, i32, i32, i32 } %1346, 2, !dbg !67 + %1350 = extractvalue { i32, i32, i32, i32 } %1346, 3, !dbg !67 + %1351 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %635), !dbg !67 + %1352 = extractvalue { i32, i32, i32, i32 } %1351, 0, !dbg !67 + %1353 = extractvalue { i32, i32, i32, i32 } %1351, 1, !dbg !67 + %1354 = extractvalue { i32, i32, i32, i32 } %1351, 2, !dbg !67 + %1355 = extractvalue { i32, i32, i32, i32 } %1351, 3, !dbg !67 + %1356 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %637), !dbg !67 + %1357 = extractvalue { i32, i32, i32, i32 } %1356, 0, !dbg !67 + %1358 = extractvalue { i32, i32, i32, i32 } %1356, 1, !dbg !67 + %1359 = extractvalue { i32, i32, i32, i32 } %1356, 2, !dbg !67 + %1360 = extractvalue { i32, i32, i32, i32 } %1356, 3, !dbg !67 + %1361 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %639), !dbg !67 + %1362 = extractvalue { i32, i32, i32, i32 } %1361, 0, !dbg !67 + %1363 = extractvalue { i32, i32, i32, i32 } %1361, 1, !dbg !67 + %1364 = extractvalue { i32, i32, i32, i32 } %1361, 2, !dbg !67 + %1365 = extractvalue { i32, i32, i32, i32 } %1361, 3, !dbg !67 + %1366 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %641), !dbg !67 + %1367 = extractvalue { i32, i32, i32, i32 } %1366, 0, !dbg !67 + %1368 = extractvalue { i32, i32, i32, i32 } %1366, 1, !dbg !67 + %1369 = extractvalue { i32, i32, i32, i32 } %1366, 2, !dbg !67 + %1370 = extractvalue { i32, i32, i32, i32 } %1366, 3, !dbg !67 + %1371 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %643), !dbg !67 + %1372 = extractvalue { i32, i32, i32, i32 } %1371, 0, !dbg !67 + %1373 = extractvalue { i32, i32, i32, i32 } %1371, 1, !dbg !67 + %1374 = extractvalue { i32, i32, i32, i32 } %1371, 2, !dbg !67 + %1375 = extractvalue { i32, i32, i32, i32 } %1371, 3, !dbg !67 + %1376 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %645), !dbg !67 + %1377 = extractvalue { i32, i32, i32, i32 } %1376, 0, !dbg !67 + %1378 = extractvalue { i32, i32, i32, i32 } %1376, 1, !dbg !67 + %1379 = extractvalue { i32, i32, i32, i32 } %1376, 2, !dbg !67 + %1380 = extractvalue { i32, i32, i32, i32 } %1376, 3, !dbg !67 + %1381 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %647), !dbg !67 + %1382 = extractvalue { i32, i32, i32, i32 } %1381, 0, !dbg !67 + %1383 = extractvalue { i32, i32, i32, i32 } %1381, 1, !dbg !67 + %1384 = extractvalue { i32, i32, i32, i32 } %1381, 2, !dbg !67 + %1385 = extractvalue { i32, i32, i32, i32 } %1381, 3, !dbg !67 + %1386 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1260, i32 %1261, i32 %1262, i32 %1263, i32 %1307, i32 %1308) #2, !dbg !73 + %1387 = extractvalue { float, float, float, float } %1386, 0, !dbg !73 + %1388 = extractvalue { float, float, float, float } %1386, 1, !dbg !73 + %1389 = extractvalue { float, float, float, float } %1386, 2, !dbg !73 + %1390 = extractvalue { float, float, float, float } %1386, 3, !dbg !73 + %1391 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1260, i32 %1261, i32 %1262, i32 %1263, i32 %1317, i32 %1318) #2, !dbg !73 + %1392 = extractvalue { float, float, float, float } %1391, 0, !dbg !73 + %1393 = extractvalue { float, float, float, float } %1391, 1, !dbg !73 + %1394 = extractvalue { float, float, float, float } %1391, 2, !dbg !73 + %1395 = extractvalue { float, float, float, float } %1391, 3, !dbg !73 + %1396 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1260, i32 %1261, i32 %1262, i32 %1263, i32 %1327, i32 %1328) #2, !dbg !73 + %1397 = extractvalue { float, float, float, float } %1396, 0, !dbg !73 + %1398 = extractvalue { float, float, float, float } %1396, 1, !dbg !73 + %1399 = extractvalue { float, float, float, float } %1396, 2, !dbg !73 + %1400 = extractvalue { float, float, float, float } %1396, 3, !dbg !73 + %1401 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1260, i32 %1261, i32 %1262, i32 %1263, i32 %1337, i32 %1338) #2, !dbg !73 + %1402 = extractvalue { float, float, float, float } %1401, 0, !dbg !73 + %1403 = extractvalue { float, float, float, float } %1401, 1, !dbg !73 + %1404 = extractvalue { float, float, float, float } %1401, 2, !dbg !73 + %1405 = extractvalue { float, float, float, float } %1401, 3, !dbg !73 + %1406 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1260, i32 %1261, i32 %1262, i32 %1263, i32 %1347, i32 %1348) #2, !dbg !73 + %1407 = extractvalue { float, float, float, float } %1406, 0, !dbg !73 + %1408 = extractvalue { float, float, float, float } %1406, 1, !dbg !73 + %1409 = extractvalue { float, float, float, float } %1406, 2, !dbg !73 + %1410 = extractvalue { float, float, float, float } %1406, 3, !dbg !73 + %1411 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1260, i32 %1261, i32 %1262, i32 %1263, i32 %1357, i32 %1358) #2, !dbg !73 + %1412 = extractvalue { float, float, float, float } %1411, 0, !dbg !73 + %1413 = extractvalue { float, float, float, float } %1411, 1, !dbg !73 + %1414 = extractvalue { float, float, float, float } %1411, 2, !dbg !73 + %1415 = extractvalue { float, float, float, float } %1411, 3, !dbg !73 + %1416 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1260, i32 %1261, i32 %1262, i32 %1263, i32 %1367, i32 %1368) #2, !dbg !73 + %1417 = extractvalue { float, float, float, float } %1416, 0, !dbg !73 + %1418 = extractvalue { float, float, float, float } %1416, 1, !dbg !73 + %1419 = extractvalue { float, float, float, float } %1416, 2, !dbg !73 + %1420 = extractvalue { float, float, float, float } %1416, 3, !dbg !73 + %1421 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1260, i32 %1261, i32 %1262, i32 %1263, i32 %1377, i32 %1378) #2, !dbg !73 + %1422 = extractvalue { float, float, float, float } %1421, 0, !dbg !73 + %1423 = extractvalue { float, float, float, float } %1421, 1, !dbg !73 + %1424 = extractvalue { float, float, float, float } %1421, 2, !dbg !73 + %1425 = extractvalue { float, float, float, float } %1421, 3, !dbg !73 + %1426 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1266, i32 %1267, i32 %1268, i32 %1269, i32 %1307, i32 %1308) #2, !dbg !73 + %1427 = extractvalue { float, float, float, float } %1426, 0, !dbg !73 + %1428 = extractvalue { float, float, float, float } %1426, 1, !dbg !73 + %1429 = extractvalue { float, float, float, float } %1426, 2, !dbg !73 + %1430 = extractvalue { float, float, float, float } %1426, 3, !dbg !73 + %1431 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1266, i32 %1267, i32 %1268, i32 %1269, i32 %1317, i32 %1318) #2, !dbg !73 + %1432 = extractvalue { float, float, float, float } %1431, 0, !dbg !73 + %1433 = extractvalue { float, float, float, float } %1431, 1, !dbg !73 + %1434 = extractvalue { float, float, float, float } %1431, 2, !dbg !73 + %1435 = extractvalue { float, float, float, float } %1431, 3, !dbg !73 + %1436 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1266, i32 %1267, i32 %1268, i32 %1269, i32 %1327, i32 %1328) #2, !dbg !73 + %1437 = extractvalue { float, float, float, float } %1436, 0, !dbg !73 + %1438 = extractvalue { float, float, float, float } %1436, 1, !dbg !73 + %1439 = extractvalue { float, float, float, float } %1436, 2, !dbg !73 + %1440 = extractvalue { float, float, float, float } %1436, 3, !dbg !73 + %1441 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1266, i32 %1267, i32 %1268, i32 %1269, i32 %1337, i32 %1338) #2, !dbg !73 + %1442 = extractvalue { float, float, float, float } %1441, 0, !dbg !73 + %1443 = extractvalue { float, float, float, float } %1441, 1, !dbg !73 + %1444 = extractvalue { float, float, float, float } %1441, 2, !dbg !73 + %1445 = extractvalue { float, float, float, float } %1441, 3, !dbg !73 + %1446 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1266, i32 %1267, i32 %1268, i32 %1269, i32 %1347, i32 %1348) #2, !dbg !73 + %1447 = extractvalue { float, float, float, float } %1446, 0, !dbg !73 + %1448 = extractvalue { float, float, float, float } %1446, 1, !dbg !73 + %1449 = extractvalue { float, float, float, float } %1446, 2, !dbg !73 + %1450 = extractvalue { float, float, float, float } %1446, 3, !dbg !73 + %1451 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1266, i32 %1267, i32 %1268, i32 %1269, i32 %1357, i32 %1358) #2, !dbg !73 + %1452 = extractvalue { float, float, float, float } %1451, 0, !dbg !73 + %1453 = extractvalue { float, float, float, float } %1451, 1, !dbg !73 + %1454 = extractvalue { float, float, float, float } %1451, 2, !dbg !73 + %1455 = extractvalue { float, float, float, float } %1451, 3, !dbg !73 + %1456 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1266, i32 %1267, i32 %1268, i32 %1269, i32 %1367, i32 %1368) #2, !dbg !73 + %1457 = extractvalue { float, float, float, float } %1456, 0, !dbg !73 + %1458 = extractvalue { float, float, float, float } %1456, 1, !dbg !73 + %1459 = extractvalue { float, float, float, float } %1456, 2, !dbg !73 + %1460 = extractvalue { float, float, float, float } %1456, 3, !dbg !73 + %1461 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, i32 %1266, i32 %1267, i32 %1268, i32 %1269, i32 %1377, i32 %1378) #2, !dbg !73 + %1462 = extractvalue { float, float, float, float } %1461, 0, !dbg !73 + %1463 = extractvalue { float, float, float, float } %1461, 1, !dbg !73 + %1464 = extractvalue { float, float, float, float } %1461, 2, !dbg !73 + %1465 = extractvalue { float, float, float, float } %1461, 3, !dbg !73 + %1466 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1387, float %1388, float %1389, float %1390, i32 %1272, i32 %1273, i32 %1274, i32 %1275, i32 %1309, i32 %1310) #2, !dbg !73 + %1467 = extractvalue { float, float, float, float } %1466, 0, !dbg !73 + %1468 = extractvalue { float, float, float, float } %1466, 1, !dbg !73 + %1469 = extractvalue { float, float, float, float } %1466, 2, !dbg !73 + %1470 = extractvalue { float, float, float, float } %1466, 3, !dbg !73 + %1471 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1392, float %1393, float %1394, float %1395, i32 %1272, i32 %1273, i32 %1274, i32 %1275, i32 %1319, i32 %1320) #2, !dbg !73 + %1472 = extractvalue { float, float, float, float } %1471, 0, !dbg !73 + %1473 = extractvalue { float, float, float, float } %1471, 1, !dbg !73 + %1474 = extractvalue { float, float, float, float } %1471, 2, !dbg !73 + %1475 = extractvalue { float, float, float, float } %1471, 3, !dbg !73 + %1476 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1397, float %1398, float %1399, float %1400, i32 %1272, i32 %1273, i32 %1274, i32 %1275, i32 %1329, i32 %1330) #2, !dbg !73 + %1477 = extractvalue { float, float, float, float } %1476, 0, !dbg !73 + %1478 = extractvalue { float, float, float, float } %1476, 1, !dbg !73 + %1479 = extractvalue { float, float, float, float } %1476, 2, !dbg !73 + %1480 = extractvalue { float, float, float, float } %1476, 3, !dbg !73 + %1481 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1402, float %1403, float %1404, float %1405, i32 %1272, i32 %1273, i32 %1274, i32 %1275, i32 %1339, i32 %1340) #2, !dbg !73 + %1482 = extractvalue { float, float, float, float } %1481, 0, !dbg !73 + %1483 = extractvalue { float, float, float, float } %1481, 1, !dbg !73 + %1484 = extractvalue { float, float, float, float } %1481, 2, !dbg !73 + %1485 = extractvalue { float, float, float, float } %1481, 3, !dbg !73 + %1486 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1407, float %1408, float %1409, float %1410, i32 %1272, i32 %1273, i32 %1274, i32 %1275, i32 %1349, i32 %1350) #2, !dbg !73 + %1487 = extractvalue { float, float, float, float } %1486, 0, !dbg !73 + %1488 = extractvalue { float, float, float, float } %1486, 1, !dbg !73 + %1489 = extractvalue { float, float, float, float } %1486, 2, !dbg !73 + %1490 = extractvalue { float, float, float, float } %1486, 3, !dbg !73 + %1491 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1412, float %1413, float %1414, float %1415, i32 %1272, i32 %1273, i32 %1274, i32 %1275, i32 %1359, i32 %1360) #2, !dbg !73 + %1492 = extractvalue { float, float, float, float } %1491, 0, !dbg !73 + %1493 = extractvalue { float, float, float, float } %1491, 1, !dbg !73 + %1494 = extractvalue { float, float, float, float } %1491, 2, !dbg !73 + %1495 = extractvalue { float, float, float, float } %1491, 3, !dbg !73 + %1496 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1417, float %1418, float %1419, float %1420, i32 %1272, i32 %1273, i32 %1274, i32 %1275, i32 %1369, i32 %1370) #2, !dbg !73 + %1497 = extractvalue { float, float, float, float } %1496, 0, !dbg !73 + %1498 = extractvalue { float, float, float, float } %1496, 1, !dbg !73 + %1499 = extractvalue { float, float, float, float } %1496, 2, !dbg !73 + %1500 = extractvalue { float, float, float, float } %1496, 3, !dbg !73 + %1501 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1422, float %1423, float %1424, float %1425, i32 %1272, i32 %1273, i32 %1274, i32 %1275, i32 %1379, i32 %1380) #2, !dbg !73 + %1502 = extractvalue { float, float, float, float } %1501, 0, !dbg !73 + %1503 = extractvalue { float, float, float, float } %1501, 1, !dbg !73 + %1504 = extractvalue { float, float, float, float } %1501, 2, !dbg !73 + %1505 = extractvalue { float, float, float, float } %1501, 3, !dbg !73 + %1506 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1427, float %1428, float %1429, float %1430, i32 %1278, i32 %1279, i32 %1280, i32 %1281, i32 %1309, i32 %1310) #2, !dbg !73 + %1507 = extractvalue { float, float, float, float } %1506, 0, !dbg !73 + %1508 = extractvalue { float, float, float, float } %1506, 1, !dbg !73 + %1509 = extractvalue { float, float, float, float } %1506, 2, !dbg !73 + %1510 = extractvalue { float, float, float, float } %1506, 3, !dbg !73 + %1511 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1432, float %1433, float %1434, float %1435, i32 %1278, i32 %1279, i32 %1280, i32 %1281, i32 %1319, i32 %1320) #2, !dbg !73 + %1512 = extractvalue { float, float, float, float } %1511, 0, !dbg !73 + %1513 = extractvalue { float, float, float, float } %1511, 1, !dbg !73 + %1514 = extractvalue { float, float, float, float } %1511, 2, !dbg !73 + %1515 = extractvalue { float, float, float, float } %1511, 3, !dbg !73 + %1516 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1437, float %1438, float %1439, float %1440, i32 %1278, i32 %1279, i32 %1280, i32 %1281, i32 %1329, i32 %1330) #2, !dbg !73 + %1517 = extractvalue { float, float, float, float } %1516, 0, !dbg !73 + %1518 = extractvalue { float, float, float, float } %1516, 1, !dbg !73 + %1519 = extractvalue { float, float, float, float } %1516, 2, !dbg !73 + %1520 = extractvalue { float, float, float, float } %1516, 3, !dbg !73 + %1521 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1442, float %1443, float %1444, float %1445, i32 %1278, i32 %1279, i32 %1280, i32 %1281, i32 %1339, i32 %1340) #2, !dbg !73 + %1522 = extractvalue { float, float, float, float } %1521, 0, !dbg !73 + %1523 = extractvalue { float, float, float, float } %1521, 1, !dbg !73 + %1524 = extractvalue { float, float, float, float } %1521, 2, !dbg !73 + %1525 = extractvalue { float, float, float, float } %1521, 3, !dbg !73 + %1526 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1447, float %1448, float %1449, float %1450, i32 %1278, i32 %1279, i32 %1280, i32 %1281, i32 %1349, i32 %1350) #2, !dbg !73 + %1527 = extractvalue { float, float, float, float } %1526, 0, !dbg !73 + %1528 = extractvalue { float, float, float, float } %1526, 1, !dbg !73 + %1529 = extractvalue { float, float, float, float } %1526, 2, !dbg !73 + %1530 = extractvalue { float, float, float, float } %1526, 3, !dbg !73 + %1531 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1452, float %1453, float %1454, float %1455, i32 %1278, i32 %1279, i32 %1280, i32 %1281, i32 %1359, i32 %1360) #2, !dbg !73 + %1532 = extractvalue { float, float, float, float } %1531, 0, !dbg !73 + %1533 = extractvalue { float, float, float, float } %1531, 1, !dbg !73 + %1534 = extractvalue { float, float, float, float } %1531, 2, !dbg !73 + %1535 = extractvalue { float, float, float, float } %1531, 3, !dbg !73 + %1536 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1457, float %1458, float %1459, float %1460, i32 %1278, i32 %1279, i32 %1280, i32 %1281, i32 %1369, i32 %1370) #2, !dbg !73 + %1537 = extractvalue { float, float, float, float } %1536, 0, !dbg !73 + %1538 = extractvalue { float, float, float, float } %1536, 1, !dbg !73 + %1539 = extractvalue { float, float, float, float } %1536, 2, !dbg !73 + %1540 = extractvalue { float, float, float, float } %1536, 3, !dbg !73 + %1541 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1462, float %1463, float %1464, float %1465, i32 %1278, i32 %1279, i32 %1280, i32 %1281, i32 %1379, i32 %1380) #2, !dbg !73 + %1542 = extractvalue { float, float, float, float } %1541, 0, !dbg !73 + %1543 = extractvalue { float, float, float, float } %1541, 1, !dbg !73 + %1544 = extractvalue { float, float, float, float } %1541, 2, !dbg !73 + %1545 = extractvalue { float, float, float, float } %1541, 3, !dbg !73 + %1546 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1467, float %1468, float %1469, float %1470, i32 %1284, i32 %1285, i32 %1286, i32 %1287, i32 %1312, i32 %1313) #2, !dbg !73 + %1547 = extractvalue { float, float, float, float } %1546, 0, !dbg !73 + %1548 = extractvalue { float, float, float, float } %1546, 1, !dbg !73 + %1549 = extractvalue { float, float, float, float } %1546, 2, !dbg !73 + %1550 = extractvalue { float, float, float, float } %1546, 3, !dbg !73 + %1551 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1472, float %1473, float %1474, float %1475, i32 %1284, i32 %1285, i32 %1286, i32 %1287, i32 %1322, i32 %1323) #2, !dbg !73 + %1552 = extractvalue { float, float, float, float } %1551, 0, !dbg !73 + %1553 = extractvalue { float, float, float, float } %1551, 1, !dbg !73 + %1554 = extractvalue { float, float, float, float } %1551, 2, !dbg !73 + %1555 = extractvalue { float, float, float, float } %1551, 3, !dbg !73 + %1556 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1477, float %1478, float %1479, float %1480, i32 %1284, i32 %1285, i32 %1286, i32 %1287, i32 %1332, i32 %1333) #2, !dbg !73 + %1557 = extractvalue { float, float, float, float } %1556, 0, !dbg !73 + %1558 = extractvalue { float, float, float, float } %1556, 1, !dbg !73 + %1559 = extractvalue { float, float, float, float } %1556, 2, !dbg !73 + %1560 = extractvalue { float, float, float, float } %1556, 3, !dbg !73 + %1561 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1482, float %1483, float %1484, float %1485, i32 %1284, i32 %1285, i32 %1286, i32 %1287, i32 %1342, i32 %1343) #2, !dbg !73 + %1562 = extractvalue { float, float, float, float } %1561, 0, !dbg !73 + %1563 = extractvalue { float, float, float, float } %1561, 1, !dbg !73 + %1564 = extractvalue { float, float, float, float } %1561, 2, !dbg !73 + %1565 = extractvalue { float, float, float, float } %1561, 3, !dbg !73 + %1566 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1487, float %1488, float %1489, float %1490, i32 %1284, i32 %1285, i32 %1286, i32 %1287, i32 %1352, i32 %1353) #2, !dbg !73 + %1567 = extractvalue { float, float, float, float } %1566, 0, !dbg !73 + %1568 = extractvalue { float, float, float, float } %1566, 1, !dbg !73 + %1569 = extractvalue { float, float, float, float } %1566, 2, !dbg !73 + %1570 = extractvalue { float, float, float, float } %1566, 3, !dbg !73 + %1571 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1492, float %1493, float %1494, float %1495, i32 %1284, i32 %1285, i32 %1286, i32 %1287, i32 %1362, i32 %1363) #2, !dbg !73 + %1572 = extractvalue { float, float, float, float } %1571, 0, !dbg !73 + %1573 = extractvalue { float, float, float, float } %1571, 1, !dbg !73 + %1574 = extractvalue { float, float, float, float } %1571, 2, !dbg !73 + %1575 = extractvalue { float, float, float, float } %1571, 3, !dbg !73 + %1576 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1497, float %1498, float %1499, float %1500, i32 %1284, i32 %1285, i32 %1286, i32 %1287, i32 %1372, i32 %1373) #2, !dbg !73 + %1577 = extractvalue { float, float, float, float } %1576, 0, !dbg !73 + %1578 = extractvalue { float, float, float, float } %1576, 1, !dbg !73 + %1579 = extractvalue { float, float, float, float } %1576, 2, !dbg !73 + %1580 = extractvalue { float, float, float, float } %1576, 3, !dbg !73 + %1581 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1502, float %1503, float %1504, float %1505, i32 %1284, i32 %1285, i32 %1286, i32 %1287, i32 %1382, i32 %1383) #2, !dbg !73 + %1582 = extractvalue { float, float, float, float } %1581, 0, !dbg !73 + %1583 = extractvalue { float, float, float, float } %1581, 1, !dbg !73 + %1584 = extractvalue { float, float, float, float } %1581, 2, !dbg !73 + %1585 = extractvalue { float, float, float, float } %1581, 3, !dbg !73 + %1586 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1507, float %1508, float %1509, float %1510, i32 %1290, i32 %1291, i32 %1292, i32 %1293, i32 %1312, i32 %1313) #2, !dbg !73 + %1587 = extractvalue { float, float, float, float } %1586, 0, !dbg !73 + %1588 = extractvalue { float, float, float, float } %1586, 1, !dbg !73 + %1589 = extractvalue { float, float, float, float } %1586, 2, !dbg !73 + %1590 = extractvalue { float, float, float, float } %1586, 3, !dbg !73 + %1591 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1512, float %1513, float %1514, float %1515, i32 %1290, i32 %1291, i32 %1292, i32 %1293, i32 %1322, i32 %1323) #2, !dbg !73 + %1592 = extractvalue { float, float, float, float } %1591, 0, !dbg !73 + %1593 = extractvalue { float, float, float, float } %1591, 1, !dbg !73 + %1594 = extractvalue { float, float, float, float } %1591, 2, !dbg !73 + %1595 = extractvalue { float, float, float, float } %1591, 3, !dbg !73 + %1596 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1517, float %1518, float %1519, float %1520, i32 %1290, i32 %1291, i32 %1292, i32 %1293, i32 %1332, i32 %1333) #2, !dbg !73 + %1597 = extractvalue { float, float, float, float } %1596, 0, !dbg !73 + %1598 = extractvalue { float, float, float, float } %1596, 1, !dbg !73 + %1599 = extractvalue { float, float, float, float } %1596, 2, !dbg !73 + %1600 = extractvalue { float, float, float, float } %1596, 3, !dbg !73 + %1601 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1522, float %1523, float %1524, float %1525, i32 %1290, i32 %1291, i32 %1292, i32 %1293, i32 %1342, i32 %1343) #2, !dbg !73 + %1602 = extractvalue { float, float, float, float } %1601, 0, !dbg !73 + %1603 = extractvalue { float, float, float, float } %1601, 1, !dbg !73 + %1604 = extractvalue { float, float, float, float } %1601, 2, !dbg !73 + %1605 = extractvalue { float, float, float, float } %1601, 3, !dbg !73 + %1606 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1527, float %1528, float %1529, float %1530, i32 %1290, i32 %1291, i32 %1292, i32 %1293, i32 %1352, i32 %1353) #2, !dbg !73 + %1607 = extractvalue { float, float, float, float } %1606, 0, !dbg !73 + %1608 = extractvalue { float, float, float, float } %1606, 1, !dbg !73 + %1609 = extractvalue { float, float, float, float } %1606, 2, !dbg !73 + %1610 = extractvalue { float, float, float, float } %1606, 3, !dbg !73 + %1611 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1532, float %1533, float %1534, float %1535, i32 %1290, i32 %1291, i32 %1292, i32 %1293, i32 %1362, i32 %1363) #2, !dbg !73 + %1612 = extractvalue { float, float, float, float } %1611, 0, !dbg !73 + %1613 = extractvalue { float, float, float, float } %1611, 1, !dbg !73 + %1614 = extractvalue { float, float, float, float } %1611, 2, !dbg !73 + %1615 = extractvalue { float, float, float, float } %1611, 3, !dbg !73 + %1616 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1537, float %1538, float %1539, float %1540, i32 %1290, i32 %1291, i32 %1292, i32 %1293, i32 %1372, i32 %1373) #2, !dbg !73 + %1617 = extractvalue { float, float, float, float } %1616, 0, !dbg !73 + %1618 = extractvalue { float, float, float, float } %1616, 1, !dbg !73 + %1619 = extractvalue { float, float, float, float } %1616, 2, !dbg !73 + %1620 = extractvalue { float, float, float, float } %1616, 3, !dbg !73 + %1621 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1542, float %1543, float %1544, float %1545, i32 %1290, i32 %1291, i32 %1292, i32 %1293, i32 %1382, i32 %1383) #2, !dbg !73 + %1622 = extractvalue { float, float, float, float } %1621, 0, !dbg !73 + %1623 = extractvalue { float, float, float, float } %1621, 1, !dbg !73 + %1624 = extractvalue { float, float, float, float } %1621, 2, !dbg !73 + %1625 = extractvalue { float, float, float, float } %1621, 3, !dbg !73 + %1626 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1547, float %1548, float %1549, float %1550, i32 %1296, i32 %1297, i32 %1298, i32 %1299, i32 %1314, i32 %1315) #2, !dbg !73 + %1627 = extractvalue { float, float, float, float } %1626, 0, !dbg !73 + %1628 = extractvalue { float, float, float, float } %1626, 1, !dbg !73 + %1629 = extractvalue { float, float, float, float } %1626, 2, !dbg !73 + %1630 = extractvalue { float, float, float, float } %1626, 3, !dbg !73 + %1631 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1552, float %1553, float %1554, float %1555, i32 %1296, i32 %1297, i32 %1298, i32 %1299, i32 %1324, i32 %1325) #2, !dbg !73 + %1632 = extractvalue { float, float, float, float } %1631, 0, !dbg !73 + %1633 = extractvalue { float, float, float, float } %1631, 1, !dbg !73 + %1634 = extractvalue { float, float, float, float } %1631, 2, !dbg !73 + %1635 = extractvalue { float, float, float, float } %1631, 3, !dbg !73 + %1636 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1557, float %1558, float %1559, float %1560, i32 %1296, i32 %1297, i32 %1298, i32 %1299, i32 %1334, i32 %1335) #2, !dbg !73 + %1637 = extractvalue { float, float, float, float } %1636, 0, !dbg !73 + %1638 = extractvalue { float, float, float, float } %1636, 1, !dbg !73 + %1639 = extractvalue { float, float, float, float } %1636, 2, !dbg !73 + %1640 = extractvalue { float, float, float, float } %1636, 3, !dbg !73 + %1641 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1562, float %1563, float %1564, float %1565, i32 %1296, i32 %1297, i32 %1298, i32 %1299, i32 %1344, i32 %1345) #2, !dbg !73 + %1642 = extractvalue { float, float, float, float } %1641, 0, !dbg !73 + %1643 = extractvalue { float, float, float, float } %1641, 1, !dbg !73 + %1644 = extractvalue { float, float, float, float } %1641, 2, !dbg !73 + %1645 = extractvalue { float, float, float, float } %1641, 3, !dbg !73 + %1646 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1567, float %1568, float %1569, float %1570, i32 %1296, i32 %1297, i32 %1298, i32 %1299, i32 %1354, i32 %1355) #2, !dbg !73 + %1647 = extractvalue { float, float, float, float } %1646, 0, !dbg !73 + %1648 = extractvalue { float, float, float, float } %1646, 1, !dbg !73 + %1649 = extractvalue { float, float, float, float } %1646, 2, !dbg !73 + %1650 = extractvalue { float, float, float, float } %1646, 3, !dbg !73 + %1651 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1572, float %1573, float %1574, float %1575, i32 %1296, i32 %1297, i32 %1298, i32 %1299, i32 %1364, i32 %1365) #2, !dbg !73 + %1652 = extractvalue { float, float, float, float } %1651, 0, !dbg !73 + %1653 = extractvalue { float, float, float, float } %1651, 1, !dbg !73 + %1654 = extractvalue { float, float, float, float } %1651, 2, !dbg !73 + %1655 = extractvalue { float, float, float, float } %1651, 3, !dbg !73 + %1656 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1577, float %1578, float %1579, float %1580, i32 %1296, i32 %1297, i32 %1298, i32 %1299, i32 %1374, i32 %1375) #2, !dbg !73 + %1657 = extractvalue { float, float, float, float } %1656, 0, !dbg !73 + %1658 = extractvalue { float, float, float, float } %1656, 1, !dbg !73 + %1659 = extractvalue { float, float, float, float } %1656, 2, !dbg !73 + %1660 = extractvalue { float, float, float, float } %1656, 3, !dbg !73 + %1661 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1582, float %1583, float %1584, float %1585, i32 %1296, i32 %1297, i32 %1298, i32 %1299, i32 %1384, i32 %1385) #2, !dbg !73 + %1662 = extractvalue { float, float, float, float } %1661, 0, !dbg !73 + %1663 = extractvalue { float, float, float, float } %1661, 1, !dbg !73 + %1664 = extractvalue { float, float, float, float } %1661, 2, !dbg !73 + %1665 = extractvalue { float, float, float, float } %1661, 3, !dbg !73 + %1666 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1587, float %1588, float %1589, float %1590, i32 %1302, i32 %1303, i32 %1304, i32 %1305, i32 %1314, i32 %1315) #2, !dbg !73 + %1667 = extractvalue { float, float, float, float } %1666, 0, !dbg !73 + %1668 = extractvalue { float, float, float, float } %1666, 1, !dbg !73 + %1669 = extractvalue { float, float, float, float } %1666, 2, !dbg !73 + %1670 = extractvalue { float, float, float, float } %1666, 3, !dbg !73 + %1671 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1592, float %1593, float %1594, float %1595, i32 %1302, i32 %1303, i32 %1304, i32 %1305, i32 %1324, i32 %1325) #2, !dbg !73 + %1672 = extractvalue { float, float, float, float } %1671, 0, !dbg !73 + %1673 = extractvalue { float, float, float, float } %1671, 1, !dbg !73 + %1674 = extractvalue { float, float, float, float } %1671, 2, !dbg !73 + %1675 = extractvalue { float, float, float, float } %1671, 3, !dbg !73 + %1676 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1597, float %1598, float %1599, float %1600, i32 %1302, i32 %1303, i32 %1304, i32 %1305, i32 %1334, i32 %1335) #2, !dbg !73 + %1677 = extractvalue { float, float, float, float } %1676, 0, !dbg !73 + %1678 = extractvalue { float, float, float, float } %1676, 1, !dbg !73 + %1679 = extractvalue { float, float, float, float } %1676, 2, !dbg !73 + %1680 = extractvalue { float, float, float, float } %1676, 3, !dbg !73 + %1681 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1602, float %1603, float %1604, float %1605, i32 %1302, i32 %1303, i32 %1304, i32 %1305, i32 %1344, i32 %1345) #2, !dbg !73 + %1682 = extractvalue { float, float, float, float } %1681, 0, !dbg !73 + %1683 = extractvalue { float, float, float, float } %1681, 1, !dbg !73 + %1684 = extractvalue { float, float, float, float } %1681, 2, !dbg !73 + %1685 = extractvalue { float, float, float, float } %1681, 3, !dbg !73 + %1686 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1607, float %1608, float %1609, float %1610, i32 %1302, i32 %1303, i32 %1304, i32 %1305, i32 %1354, i32 %1355) #2, !dbg !73 + %1687 = extractvalue { float, float, float, float } %1686, 0, !dbg !73 + %1688 = extractvalue { float, float, float, float } %1686, 1, !dbg !73 + %1689 = extractvalue { float, float, float, float } %1686, 2, !dbg !73 + %1690 = extractvalue { float, float, float, float } %1686, 3, !dbg !73 + %1691 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1612, float %1613, float %1614, float %1615, i32 %1302, i32 %1303, i32 %1304, i32 %1305, i32 %1364, i32 %1365) #2, !dbg !73 + %1692 = extractvalue { float, float, float, float } %1691, 0, !dbg !73 + %1693 = extractvalue { float, float, float, float } %1691, 1, !dbg !73 + %1694 = extractvalue { float, float, float, float } %1691, 2, !dbg !73 + %1695 = extractvalue { float, float, float, float } %1691, 3, !dbg !73 + %1696 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1617, float %1618, float %1619, float %1620, i32 %1302, i32 %1303, i32 %1304, i32 %1305, i32 %1374, i32 %1375) #2, !dbg !73 + %1697 = extractvalue { float, float, float, float } %1696, 0, !dbg !73 + %1698 = extractvalue { float, float, float, float } %1696, 1, !dbg !73 + %1699 = extractvalue { float, float, float, float } %1696, 2, !dbg !73 + %1700 = extractvalue { float, float, float, float } %1696, 3, !dbg !73 + %1701 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1622, float %1623, float %1624, float %1625, i32 %1302, i32 %1303, i32 %1304, i32 %1305, i32 %1384, i32 %1385) #2, !dbg !73 + %1702 = extractvalue { float, float, float, float } %1701, 0, !dbg !73 + %1703 = extractvalue { float, float, float, float } %1701, 1, !dbg !73 + %1704 = extractvalue { float, float, float, float } %1701, 2, !dbg !73 + %1705 = extractvalue { float, float, float, float } %1701, 3, !dbg !73 + %1706 = getelementptr bfloat, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 24576), i32 %1256, !dbg !45 + %1707 = getelementptr inbounds nuw i8, ptr addrspace(3) %1706, i32 %556, !dbg !45 + %1708 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %1707), !dbg !45 + %1709 = extractvalue { i32, i32, i32, i32 } %1708, 0, !dbg !45 + %1710 = extractvalue { i32, i32, i32, i32 } %1708, 1, !dbg !45 + %1711 = extractvalue { i32, i32, i32, i32 } %1708, 2, !dbg !45 + %1712 = extractvalue { i32, i32, i32, i32 } %1708, 3, !dbg !45 + %1713 = getelementptr inbounds nuw i8, ptr addrspace(3) %1706, i32 %558, !dbg !45 + %1714 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %1713), !dbg !45 + %1715 = extractvalue { i32, i32, i32, i32 } %1714, 0, !dbg !45 + %1716 = extractvalue { i32, i32, i32, i32 } %1714, 1, !dbg !45 + %1717 = extractvalue { i32, i32, i32, i32 } %1714, 2, !dbg !45 + %1718 = extractvalue { i32, i32, i32, i32 } %1714, 3, !dbg !45 + %1719 = getelementptr inbounds nuw i8, ptr addrspace(3) %1706, i32 %560, !dbg !45 + %1720 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %1719), !dbg !45 + %1721 = extractvalue { i32, i32, i32, i32 } %1720, 0, !dbg !45 + %1722 = extractvalue { i32, i32, i32, i32 } %1720, 1, !dbg !45 + %1723 = extractvalue { i32, i32, i32, i32 } %1720, 2, !dbg !45 + %1724 = extractvalue { i32, i32, i32, i32 } %1720, 3, !dbg !45 + %1725 = getelementptr inbounds nuw i8, ptr addrspace(3) %1706, i32 %562, !dbg !45 + %1726 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %1725), !dbg !45 + %1727 = extractvalue { i32, i32, i32, i32 } %1726, 0, !dbg !45 + %1728 = extractvalue { i32, i32, i32, i32 } %1726, 1, !dbg !45 + %1729 = extractvalue { i32, i32, i32, i32 } %1726, 2, !dbg !45 + %1730 = extractvalue { i32, i32, i32, i32 } %1726, 3, !dbg !45 + %1731 = getelementptr inbounds nuw i8, ptr addrspace(3) %1706, i32 %564, !dbg !45 + %1732 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %1731), !dbg !45 + %1733 = extractvalue { i32, i32, i32, i32 } %1732, 0, !dbg !45 + %1734 = extractvalue { i32, i32, i32, i32 } %1732, 1, !dbg !45 + %1735 = extractvalue { i32, i32, i32, i32 } %1732, 2, !dbg !45 + %1736 = extractvalue { i32, i32, i32, i32 } %1732, 3, !dbg !45 + %1737 = getelementptr inbounds nuw i8, ptr addrspace(3) %1706, i32 %566, !dbg !45 + %1738 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %1737), !dbg !45 + %1739 = extractvalue { i32, i32, i32, i32 } %1738, 0, !dbg !45 + %1740 = extractvalue { i32, i32, i32, i32 } %1738, 1, !dbg !45 + %1741 = extractvalue { i32, i32, i32, i32 } %1738, 2, !dbg !45 + %1742 = extractvalue { i32, i32, i32, i32 } %1738, 3, !dbg !45 + %1743 = getelementptr inbounds nuw i8, ptr addrspace(3) %1706, i32 %568, !dbg !45 + %1744 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %1743), !dbg !45 + %1745 = extractvalue { i32, i32, i32, i32 } %1744, 0, !dbg !45 + %1746 = extractvalue { i32, i32, i32, i32 } %1744, 1, !dbg !45 + %1747 = extractvalue { i32, i32, i32, i32 } %1744, 2, !dbg !45 + %1748 = extractvalue { i32, i32, i32, i32 } %1744, 3, !dbg !45 + %1749 = getelementptr inbounds nuw i8, ptr addrspace(3) %1706, i32 %570, !dbg !45 + %1750 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %1749), !dbg !45 + %1751 = extractvalue { i32, i32, i32, i32 } %1750, 0, !dbg !45 + %1752 = extractvalue { i32, i32, i32, i32 } %1750, 1, !dbg !45 + %1753 = extractvalue { i32, i32, i32, i32 } %1750, 2, !dbg !45 + %1754 = extractvalue { i32, i32, i32, i32 } %1750, 3, !dbg !45 + %1755 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %648), !dbg !71 + %1756 = extractvalue { i32, i32, i32, i32 } %1755, 0, !dbg !71 + %1757 = extractvalue { i32, i32, i32, i32 } %1755, 1, !dbg !71 + %1758 = extractvalue { i32, i32, i32, i32 } %1755, 2, !dbg !71 + %1759 = extractvalue { i32, i32, i32, i32 } %1755, 3, !dbg !71 + %1760 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %649), !dbg !71 + %1761 = extractvalue { i32, i32, i32, i32 } %1760, 0, !dbg !71 + %1762 = extractvalue { i32, i32, i32, i32 } %1760, 1, !dbg !71 + %1763 = extractvalue { i32, i32, i32, i32 } %1760, 2, !dbg !71 + %1764 = extractvalue { i32, i32, i32, i32 } %1760, 3, !dbg !71 + %1765 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %650), !dbg !71 + %1766 = extractvalue { i32, i32, i32, i32 } %1765, 0, !dbg !71 + %1767 = extractvalue { i32, i32, i32, i32 } %1765, 1, !dbg !71 + %1768 = extractvalue { i32, i32, i32, i32 } %1765, 2, !dbg !71 + %1769 = extractvalue { i32, i32, i32, i32 } %1765, 3, !dbg !71 + %1770 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %651), !dbg !71 + %1771 = extractvalue { i32, i32, i32, i32 } %1770, 0, !dbg !71 + %1772 = extractvalue { i32, i32, i32, i32 } %1770, 1, !dbg !71 + %1773 = extractvalue { i32, i32, i32, i32 } %1770, 2, !dbg !71 + %1774 = extractvalue { i32, i32, i32, i32 } %1770, 3, !dbg !71 + %1775 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %652), !dbg !71 + %1776 = extractvalue { i32, i32, i32, i32 } %1775, 0, !dbg !71 + %1777 = extractvalue { i32, i32, i32, i32 } %1775, 1, !dbg !71 + %1778 = extractvalue { i32, i32, i32, i32 } %1775, 2, !dbg !71 + %1779 = extractvalue { i32, i32, i32, i32 } %1775, 3, !dbg !71 + %1780 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %653), !dbg !71 + %1781 = extractvalue { i32, i32, i32, i32 } %1780, 0, !dbg !71 + %1782 = extractvalue { i32, i32, i32, i32 } %1780, 1, !dbg !71 + %1783 = extractvalue { i32, i32, i32, i32 } %1780, 2, !dbg !71 + %1784 = extractvalue { i32, i32, i32, i32 } %1780, 3, !dbg !71 + %1785 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %654), !dbg !71 + %1786 = extractvalue { i32, i32, i32, i32 } %1785, 0, !dbg !71 + %1787 = extractvalue { i32, i32, i32, i32 } %1785, 1, !dbg !71 + %1788 = extractvalue { i32, i32, i32, i32 } %1785, 2, !dbg !71 + %1789 = extractvalue { i32, i32, i32, i32 } %1785, 3, !dbg !71 + %1790 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %655), !dbg !71 + %1791 = extractvalue { i32, i32, i32, i32 } %1790, 0, !dbg !71 + %1792 = extractvalue { i32, i32, i32, i32 } %1790, 1, !dbg !71 + %1793 = extractvalue { i32, i32, i32, i32 } %1790, 2, !dbg !71 + %1794 = extractvalue { i32, i32, i32, i32 } %1790, 3, !dbg !71 + %1795 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %656), !dbg !71 + %1796 = extractvalue { i32, i32, i32, i32 } %1795, 0, !dbg !71 + %1797 = extractvalue { i32, i32, i32, i32 } %1795, 1, !dbg !71 + %1798 = extractvalue { i32, i32, i32, i32 } %1795, 2, !dbg !71 + %1799 = extractvalue { i32, i32, i32, i32 } %1795, 3, !dbg !71 + %1800 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %657), !dbg !71 + %1801 = extractvalue { i32, i32, i32, i32 } %1800, 0, !dbg !71 + %1802 = extractvalue { i32, i32, i32, i32 } %1800, 1, !dbg !71 + %1803 = extractvalue { i32, i32, i32, i32 } %1800, 2, !dbg !71 + %1804 = extractvalue { i32, i32, i32, i32 } %1800, 3, !dbg !71 + %1805 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %658), !dbg !71 + %1806 = extractvalue { i32, i32, i32, i32 } %1805, 0, !dbg !71 + %1807 = extractvalue { i32, i32, i32, i32 } %1805, 1, !dbg !71 + %1808 = extractvalue { i32, i32, i32, i32 } %1805, 2, !dbg !71 + %1809 = extractvalue { i32, i32, i32, i32 } %1805, 3, !dbg !71 + %1810 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %659), !dbg !71 + %1811 = extractvalue { i32, i32, i32, i32 } %1810, 0, !dbg !71 + %1812 = extractvalue { i32, i32, i32, i32 } %1810, 1, !dbg !71 + %1813 = extractvalue { i32, i32, i32, i32 } %1810, 2, !dbg !71 + %1814 = extractvalue { i32, i32, i32, i32 } %1810, 3, !dbg !71 + %1815 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %660), !dbg !71 + %1816 = extractvalue { i32, i32, i32, i32 } %1815, 0, !dbg !71 + %1817 = extractvalue { i32, i32, i32, i32 } %1815, 1, !dbg !71 + %1818 = extractvalue { i32, i32, i32, i32 } %1815, 2, !dbg !71 + %1819 = extractvalue { i32, i32, i32, i32 } %1815, 3, !dbg !71 + %1820 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %661), !dbg !71 + %1821 = extractvalue { i32, i32, i32, i32 } %1820, 0, !dbg !71 + %1822 = extractvalue { i32, i32, i32, i32 } %1820, 1, !dbg !71 + %1823 = extractvalue { i32, i32, i32, i32 } %1820, 2, !dbg !71 + %1824 = extractvalue { i32, i32, i32, i32 } %1820, 3, !dbg !71 + %1825 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %662), !dbg !71 + %1826 = extractvalue { i32, i32, i32, i32 } %1825, 0, !dbg !71 + %1827 = extractvalue { i32, i32, i32, i32 } %1825, 1, !dbg !71 + %1828 = extractvalue { i32, i32, i32, i32 } %1825, 2, !dbg !71 + %1829 = extractvalue { i32, i32, i32, i32 } %1825, 3, !dbg !71 + %1830 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %663), !dbg !71 + %1831 = extractvalue { i32, i32, i32, i32 } %1830, 0, !dbg !71 + %1832 = extractvalue { i32, i32, i32, i32 } %1830, 1, !dbg !71 + %1833 = extractvalue { i32, i32, i32, i32 } %1830, 2, !dbg !71 + %1834 = extractvalue { i32, i32, i32, i32 } %1830, 3, !dbg !71 + %1835 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1627, float %1628, float %1629, float %1630, i32 %1709, i32 %1710, i32 %1711, i32 %1712, i32 %1756, i32 %1757) #2, !dbg !74 + %1836 = extractvalue { float, float, float, float } %1835, 0, !dbg !74 + %1837 = extractvalue { float, float, float, float } %1835, 1, !dbg !74 + %1838 = extractvalue { float, float, float, float } %1835, 2, !dbg !74 + %1839 = extractvalue { float, float, float, float } %1835, 3, !dbg !74 + %1840 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1632, float %1633, float %1634, float %1635, i32 %1709, i32 %1710, i32 %1711, i32 %1712, i32 %1766, i32 %1767) #2, !dbg !74 + %1841 = extractvalue { float, float, float, float } %1840, 0, !dbg !74 + %1842 = extractvalue { float, float, float, float } %1840, 1, !dbg !74 + %1843 = extractvalue { float, float, float, float } %1840, 2, !dbg !74 + %1844 = extractvalue { float, float, float, float } %1840, 3, !dbg !74 + %1845 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1637, float %1638, float %1639, float %1640, i32 %1709, i32 %1710, i32 %1711, i32 %1712, i32 %1776, i32 %1777) #2, !dbg !74 + %1846 = extractvalue { float, float, float, float } %1845, 0, !dbg !74 + %1847 = extractvalue { float, float, float, float } %1845, 1, !dbg !74 + %1848 = extractvalue { float, float, float, float } %1845, 2, !dbg !74 + %1849 = extractvalue { float, float, float, float } %1845, 3, !dbg !74 + %1850 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1642, float %1643, float %1644, float %1645, i32 %1709, i32 %1710, i32 %1711, i32 %1712, i32 %1786, i32 %1787) #2, !dbg !74 + %1851 = extractvalue { float, float, float, float } %1850, 0, !dbg !74 + %1852 = extractvalue { float, float, float, float } %1850, 1, !dbg !74 + %1853 = extractvalue { float, float, float, float } %1850, 2, !dbg !74 + %1854 = extractvalue { float, float, float, float } %1850, 3, !dbg !74 + %1855 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1647, float %1648, float %1649, float %1650, i32 %1709, i32 %1710, i32 %1711, i32 %1712, i32 %1796, i32 %1797) #2, !dbg !74 + %1856 = extractvalue { float, float, float, float } %1855, 0, !dbg !74 + %1857 = extractvalue { float, float, float, float } %1855, 1, !dbg !74 + %1858 = extractvalue { float, float, float, float } %1855, 2, !dbg !74 + %1859 = extractvalue { float, float, float, float } %1855, 3, !dbg !74 + %1860 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1652, float %1653, float %1654, float %1655, i32 %1709, i32 %1710, i32 %1711, i32 %1712, i32 %1806, i32 %1807) #2, !dbg !74 + %1861 = extractvalue { float, float, float, float } %1860, 0, !dbg !74 + %1862 = extractvalue { float, float, float, float } %1860, 1, !dbg !74 + %1863 = extractvalue { float, float, float, float } %1860, 2, !dbg !74 + %1864 = extractvalue { float, float, float, float } %1860, 3, !dbg !74 + %1865 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1657, float %1658, float %1659, float %1660, i32 %1709, i32 %1710, i32 %1711, i32 %1712, i32 %1816, i32 %1817) #2, !dbg !74 + %1866 = extractvalue { float, float, float, float } %1865, 0, !dbg !74 + %1867 = extractvalue { float, float, float, float } %1865, 1, !dbg !74 + %1868 = extractvalue { float, float, float, float } %1865, 2, !dbg !74 + %1869 = extractvalue { float, float, float, float } %1865, 3, !dbg !74 + %1870 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1662, float %1663, float %1664, float %1665, i32 %1709, i32 %1710, i32 %1711, i32 %1712, i32 %1826, i32 %1827) #2, !dbg !74 + %1871 = extractvalue { float, float, float, float } %1870, 0, !dbg !74 + %1872 = extractvalue { float, float, float, float } %1870, 1, !dbg !74 + %1873 = extractvalue { float, float, float, float } %1870, 2, !dbg !74 + %1874 = extractvalue { float, float, float, float } %1870, 3, !dbg !74 + %1875 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1667, float %1668, float %1669, float %1670, i32 %1715, i32 %1716, i32 %1717, i32 %1718, i32 %1756, i32 %1757) #2, !dbg !74 + %1876 = extractvalue { float, float, float, float } %1875, 0, !dbg !74 + %1877 = extractvalue { float, float, float, float } %1875, 1, !dbg !74 + %1878 = extractvalue { float, float, float, float } %1875, 2, !dbg !74 + %1879 = extractvalue { float, float, float, float } %1875, 3, !dbg !74 + %1880 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1672, float %1673, float %1674, float %1675, i32 %1715, i32 %1716, i32 %1717, i32 %1718, i32 %1766, i32 %1767) #2, !dbg !74 + %1881 = extractvalue { float, float, float, float } %1880, 0, !dbg !74 + %1882 = extractvalue { float, float, float, float } %1880, 1, !dbg !74 + %1883 = extractvalue { float, float, float, float } %1880, 2, !dbg !74 + %1884 = extractvalue { float, float, float, float } %1880, 3, !dbg !74 + %1885 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1677, float %1678, float %1679, float %1680, i32 %1715, i32 %1716, i32 %1717, i32 %1718, i32 %1776, i32 %1777) #2, !dbg !74 + %1886 = extractvalue { float, float, float, float } %1885, 0, !dbg !74 + %1887 = extractvalue { float, float, float, float } %1885, 1, !dbg !74 + %1888 = extractvalue { float, float, float, float } %1885, 2, !dbg !74 + %1889 = extractvalue { float, float, float, float } %1885, 3, !dbg !74 + %1890 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1682, float %1683, float %1684, float %1685, i32 %1715, i32 %1716, i32 %1717, i32 %1718, i32 %1786, i32 %1787) #2, !dbg !74 + %1891 = extractvalue { float, float, float, float } %1890, 0, !dbg !74 + %1892 = extractvalue { float, float, float, float } %1890, 1, !dbg !74 + %1893 = extractvalue { float, float, float, float } %1890, 2, !dbg !74 + %1894 = extractvalue { float, float, float, float } %1890, 3, !dbg !74 + %1895 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1687, float %1688, float %1689, float %1690, i32 %1715, i32 %1716, i32 %1717, i32 %1718, i32 %1796, i32 %1797) #2, !dbg !74 + %1896 = extractvalue { float, float, float, float } %1895, 0, !dbg !74 + %1897 = extractvalue { float, float, float, float } %1895, 1, !dbg !74 + %1898 = extractvalue { float, float, float, float } %1895, 2, !dbg !74 + %1899 = extractvalue { float, float, float, float } %1895, 3, !dbg !74 + %1900 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1692, float %1693, float %1694, float %1695, i32 %1715, i32 %1716, i32 %1717, i32 %1718, i32 %1806, i32 %1807) #2, !dbg !74 + %1901 = extractvalue { float, float, float, float } %1900, 0, !dbg !74 + %1902 = extractvalue { float, float, float, float } %1900, 1, !dbg !74 + %1903 = extractvalue { float, float, float, float } %1900, 2, !dbg !74 + %1904 = extractvalue { float, float, float, float } %1900, 3, !dbg !74 + %1905 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1697, float %1698, float %1699, float %1700, i32 %1715, i32 %1716, i32 %1717, i32 %1718, i32 %1816, i32 %1817) #2, !dbg !74 + %1906 = extractvalue { float, float, float, float } %1905, 0, !dbg !74 + %1907 = extractvalue { float, float, float, float } %1905, 1, !dbg !74 + %1908 = extractvalue { float, float, float, float } %1905, 2, !dbg !74 + %1909 = extractvalue { float, float, float, float } %1905, 3, !dbg !74 + %1910 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1702, float %1703, float %1704, float %1705, i32 %1715, i32 %1716, i32 %1717, i32 %1718, i32 %1826, i32 %1827) #2, !dbg !74 + %1911 = extractvalue { float, float, float, float } %1910, 0, !dbg !74 + %1912 = extractvalue { float, float, float, float } %1910, 1, !dbg !74 + %1913 = extractvalue { float, float, float, float } %1910, 2, !dbg !74 + %1914 = extractvalue { float, float, float, float } %1910, 3, !dbg !74 + %1915 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1836, float %1837, float %1838, float %1839, i32 %1721, i32 %1722, i32 %1723, i32 %1724, i32 %1758, i32 %1759) #2, !dbg !74 + %1916 = extractvalue { float, float, float, float } %1915, 0, !dbg !74 + %1917 = extractvalue { float, float, float, float } %1915, 1, !dbg !74 + %1918 = extractvalue { float, float, float, float } %1915, 2, !dbg !74 + %1919 = extractvalue { float, float, float, float } %1915, 3, !dbg !74 + %1920 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1841, float %1842, float %1843, float %1844, i32 %1721, i32 %1722, i32 %1723, i32 %1724, i32 %1768, i32 %1769) #2, !dbg !74 + %1921 = extractvalue { float, float, float, float } %1920, 0, !dbg !74 + %1922 = extractvalue { float, float, float, float } %1920, 1, !dbg !74 + %1923 = extractvalue { float, float, float, float } %1920, 2, !dbg !74 + %1924 = extractvalue { float, float, float, float } %1920, 3, !dbg !74 + %1925 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1846, float %1847, float %1848, float %1849, i32 %1721, i32 %1722, i32 %1723, i32 %1724, i32 %1778, i32 %1779) #2, !dbg !74 + %1926 = extractvalue { float, float, float, float } %1925, 0, !dbg !74 + %1927 = extractvalue { float, float, float, float } %1925, 1, !dbg !74 + %1928 = extractvalue { float, float, float, float } %1925, 2, !dbg !74 + %1929 = extractvalue { float, float, float, float } %1925, 3, !dbg !74 + %1930 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1851, float %1852, float %1853, float %1854, i32 %1721, i32 %1722, i32 %1723, i32 %1724, i32 %1788, i32 %1789) #2, !dbg !74 + %1931 = extractvalue { float, float, float, float } %1930, 0, !dbg !74 + %1932 = extractvalue { float, float, float, float } %1930, 1, !dbg !74 + %1933 = extractvalue { float, float, float, float } %1930, 2, !dbg !74 + %1934 = extractvalue { float, float, float, float } %1930, 3, !dbg !74 + %1935 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1856, float %1857, float %1858, float %1859, i32 %1721, i32 %1722, i32 %1723, i32 %1724, i32 %1798, i32 %1799) #2, !dbg !74 + %1936 = extractvalue { float, float, float, float } %1935, 0, !dbg !74 + %1937 = extractvalue { float, float, float, float } %1935, 1, !dbg !74 + %1938 = extractvalue { float, float, float, float } %1935, 2, !dbg !74 + %1939 = extractvalue { float, float, float, float } %1935, 3, !dbg !74 + %1940 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1861, float %1862, float %1863, float %1864, i32 %1721, i32 %1722, i32 %1723, i32 %1724, i32 %1808, i32 %1809) #2, !dbg !74 + %1941 = extractvalue { float, float, float, float } %1940, 0, !dbg !74 + %1942 = extractvalue { float, float, float, float } %1940, 1, !dbg !74 + %1943 = extractvalue { float, float, float, float } %1940, 2, !dbg !74 + %1944 = extractvalue { float, float, float, float } %1940, 3, !dbg !74 + %1945 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1866, float %1867, float %1868, float %1869, i32 %1721, i32 %1722, i32 %1723, i32 %1724, i32 %1818, i32 %1819) #2, !dbg !74 + %1946 = extractvalue { float, float, float, float } %1945, 0, !dbg !74 + %1947 = extractvalue { float, float, float, float } %1945, 1, !dbg !74 + %1948 = extractvalue { float, float, float, float } %1945, 2, !dbg !74 + %1949 = extractvalue { float, float, float, float } %1945, 3, !dbg !74 + %1950 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1871, float %1872, float %1873, float %1874, i32 %1721, i32 %1722, i32 %1723, i32 %1724, i32 %1828, i32 %1829) #2, !dbg !74 + %1951 = extractvalue { float, float, float, float } %1950, 0, !dbg !74 + %1952 = extractvalue { float, float, float, float } %1950, 1, !dbg !74 + %1953 = extractvalue { float, float, float, float } %1950, 2, !dbg !74 + %1954 = extractvalue { float, float, float, float } %1950, 3, !dbg !74 + %1955 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1876, float %1877, float %1878, float %1879, i32 %1727, i32 %1728, i32 %1729, i32 %1730, i32 %1758, i32 %1759) #2, !dbg !74 + %1956 = extractvalue { float, float, float, float } %1955, 0, !dbg !74 + %1957 = extractvalue { float, float, float, float } %1955, 1, !dbg !74 + %1958 = extractvalue { float, float, float, float } %1955, 2, !dbg !74 + %1959 = extractvalue { float, float, float, float } %1955, 3, !dbg !74 + %1960 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1881, float %1882, float %1883, float %1884, i32 %1727, i32 %1728, i32 %1729, i32 %1730, i32 %1768, i32 %1769) #2, !dbg !74 + %1961 = extractvalue { float, float, float, float } %1960, 0, !dbg !74 + %1962 = extractvalue { float, float, float, float } %1960, 1, !dbg !74 + %1963 = extractvalue { float, float, float, float } %1960, 2, !dbg !74 + %1964 = extractvalue { float, float, float, float } %1960, 3, !dbg !74 + %1965 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1886, float %1887, float %1888, float %1889, i32 %1727, i32 %1728, i32 %1729, i32 %1730, i32 %1778, i32 %1779) #2, !dbg !74 + %1966 = extractvalue { float, float, float, float } %1965, 0, !dbg !74 + %1967 = extractvalue { float, float, float, float } %1965, 1, !dbg !74 + %1968 = extractvalue { float, float, float, float } %1965, 2, !dbg !74 + %1969 = extractvalue { float, float, float, float } %1965, 3, !dbg !74 + %1970 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1891, float %1892, float %1893, float %1894, i32 %1727, i32 %1728, i32 %1729, i32 %1730, i32 %1788, i32 %1789) #2, !dbg !74 + %1971 = extractvalue { float, float, float, float } %1970, 0, !dbg !74 + %1972 = extractvalue { float, float, float, float } %1970, 1, !dbg !74 + %1973 = extractvalue { float, float, float, float } %1970, 2, !dbg !74 + %1974 = extractvalue { float, float, float, float } %1970, 3, !dbg !74 + %1975 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1896, float %1897, float %1898, float %1899, i32 %1727, i32 %1728, i32 %1729, i32 %1730, i32 %1798, i32 %1799) #2, !dbg !74 + %1976 = extractvalue { float, float, float, float } %1975, 0, !dbg !74 + %1977 = extractvalue { float, float, float, float } %1975, 1, !dbg !74 + %1978 = extractvalue { float, float, float, float } %1975, 2, !dbg !74 + %1979 = extractvalue { float, float, float, float } %1975, 3, !dbg !74 + %1980 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1901, float %1902, float %1903, float %1904, i32 %1727, i32 %1728, i32 %1729, i32 %1730, i32 %1808, i32 %1809) #2, !dbg !74 + %1981 = extractvalue { float, float, float, float } %1980, 0, !dbg !74 + %1982 = extractvalue { float, float, float, float } %1980, 1, !dbg !74 + %1983 = extractvalue { float, float, float, float } %1980, 2, !dbg !74 + %1984 = extractvalue { float, float, float, float } %1980, 3, !dbg !74 + %1985 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1906, float %1907, float %1908, float %1909, i32 %1727, i32 %1728, i32 %1729, i32 %1730, i32 %1818, i32 %1819) #2, !dbg !74 + %1986 = extractvalue { float, float, float, float } %1985, 0, !dbg !74 + %1987 = extractvalue { float, float, float, float } %1985, 1, !dbg !74 + %1988 = extractvalue { float, float, float, float } %1985, 2, !dbg !74 + %1989 = extractvalue { float, float, float, float } %1985, 3, !dbg !74 + %1990 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1911, float %1912, float %1913, float %1914, i32 %1727, i32 %1728, i32 %1729, i32 %1730, i32 %1828, i32 %1829) #2, !dbg !74 + %1991 = extractvalue { float, float, float, float } %1990, 0, !dbg !74 + %1992 = extractvalue { float, float, float, float } %1990, 1, !dbg !74 + %1993 = extractvalue { float, float, float, float } %1990, 2, !dbg !74 + %1994 = extractvalue { float, float, float, float } %1990, 3, !dbg !74 + %1995 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1916, float %1917, float %1918, float %1919, i32 %1733, i32 %1734, i32 %1735, i32 %1736, i32 %1761, i32 %1762) #2, !dbg !74 + %1996 = extractvalue { float, float, float, float } %1995, 0, !dbg !74 + %1997 = extractvalue { float, float, float, float } %1995, 1, !dbg !74 + %1998 = extractvalue { float, float, float, float } %1995, 2, !dbg !74 + %1999 = extractvalue { float, float, float, float } %1995, 3, !dbg !74 + %2000 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1921, float %1922, float %1923, float %1924, i32 %1733, i32 %1734, i32 %1735, i32 %1736, i32 %1771, i32 %1772) #2, !dbg !74 + %2001 = extractvalue { float, float, float, float } %2000, 0, !dbg !74 + %2002 = extractvalue { float, float, float, float } %2000, 1, !dbg !74 + %2003 = extractvalue { float, float, float, float } %2000, 2, !dbg !74 + %2004 = extractvalue { float, float, float, float } %2000, 3, !dbg !74 + %2005 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1926, float %1927, float %1928, float %1929, i32 %1733, i32 %1734, i32 %1735, i32 %1736, i32 %1781, i32 %1782) #2, !dbg !74 + %2006 = extractvalue { float, float, float, float } %2005, 0, !dbg !74 + %2007 = extractvalue { float, float, float, float } %2005, 1, !dbg !74 + %2008 = extractvalue { float, float, float, float } %2005, 2, !dbg !74 + %2009 = extractvalue { float, float, float, float } %2005, 3, !dbg !74 + %2010 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1931, float %1932, float %1933, float %1934, i32 %1733, i32 %1734, i32 %1735, i32 %1736, i32 %1791, i32 %1792) #2, !dbg !74 + %2011 = extractvalue { float, float, float, float } %2010, 0, !dbg !74 + %2012 = extractvalue { float, float, float, float } %2010, 1, !dbg !74 + %2013 = extractvalue { float, float, float, float } %2010, 2, !dbg !74 + %2014 = extractvalue { float, float, float, float } %2010, 3, !dbg !74 + %2015 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1936, float %1937, float %1938, float %1939, i32 %1733, i32 %1734, i32 %1735, i32 %1736, i32 %1801, i32 %1802) #2, !dbg !74 + %2016 = extractvalue { float, float, float, float } %2015, 0, !dbg !74 + %2017 = extractvalue { float, float, float, float } %2015, 1, !dbg !74 + %2018 = extractvalue { float, float, float, float } %2015, 2, !dbg !74 + %2019 = extractvalue { float, float, float, float } %2015, 3, !dbg !74 + %2020 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1941, float %1942, float %1943, float %1944, i32 %1733, i32 %1734, i32 %1735, i32 %1736, i32 %1811, i32 %1812) #2, !dbg !74 + %2021 = extractvalue { float, float, float, float } %2020, 0, !dbg !74 + %2022 = extractvalue { float, float, float, float } %2020, 1, !dbg !74 + %2023 = extractvalue { float, float, float, float } %2020, 2, !dbg !74 + %2024 = extractvalue { float, float, float, float } %2020, 3, !dbg !74 + %2025 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1946, float %1947, float %1948, float %1949, i32 %1733, i32 %1734, i32 %1735, i32 %1736, i32 %1821, i32 %1822) #2, !dbg !74 + %2026 = extractvalue { float, float, float, float } %2025, 0, !dbg !74 + %2027 = extractvalue { float, float, float, float } %2025, 1, !dbg !74 + %2028 = extractvalue { float, float, float, float } %2025, 2, !dbg !74 + %2029 = extractvalue { float, float, float, float } %2025, 3, !dbg !74 + %2030 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1951, float %1952, float %1953, float %1954, i32 %1733, i32 %1734, i32 %1735, i32 %1736, i32 %1831, i32 %1832) #2, !dbg !74 + %2031 = extractvalue { float, float, float, float } %2030, 0, !dbg !74 + %2032 = extractvalue { float, float, float, float } %2030, 1, !dbg !74 + %2033 = extractvalue { float, float, float, float } %2030, 2, !dbg !74 + %2034 = extractvalue { float, float, float, float } %2030, 3, !dbg !74 + %2035 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1956, float %1957, float %1958, float %1959, i32 %1739, i32 %1740, i32 %1741, i32 %1742, i32 %1761, i32 %1762) #2, !dbg !74 + %2036 = extractvalue { float, float, float, float } %2035, 0, !dbg !74 + %2037 = extractvalue { float, float, float, float } %2035, 1, !dbg !74 + %2038 = extractvalue { float, float, float, float } %2035, 2, !dbg !74 + %2039 = extractvalue { float, float, float, float } %2035, 3, !dbg !74 + %2040 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1961, float %1962, float %1963, float %1964, i32 %1739, i32 %1740, i32 %1741, i32 %1742, i32 %1771, i32 %1772) #2, !dbg !74 + %2041 = extractvalue { float, float, float, float } %2040, 0, !dbg !74 + %2042 = extractvalue { float, float, float, float } %2040, 1, !dbg !74 + %2043 = extractvalue { float, float, float, float } %2040, 2, !dbg !74 + %2044 = extractvalue { float, float, float, float } %2040, 3, !dbg !74 + %2045 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1966, float %1967, float %1968, float %1969, i32 %1739, i32 %1740, i32 %1741, i32 %1742, i32 %1781, i32 %1782) #2, !dbg !74 + %2046 = extractvalue { float, float, float, float } %2045, 0, !dbg !74 + %2047 = extractvalue { float, float, float, float } %2045, 1, !dbg !74 + %2048 = extractvalue { float, float, float, float } %2045, 2, !dbg !74 + %2049 = extractvalue { float, float, float, float } %2045, 3, !dbg !74 + %2050 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1971, float %1972, float %1973, float %1974, i32 %1739, i32 %1740, i32 %1741, i32 %1742, i32 %1791, i32 %1792) #2, !dbg !74 + %2051 = extractvalue { float, float, float, float } %2050, 0, !dbg !74 + %2052 = extractvalue { float, float, float, float } %2050, 1, !dbg !74 + %2053 = extractvalue { float, float, float, float } %2050, 2, !dbg !74 + %2054 = extractvalue { float, float, float, float } %2050, 3, !dbg !74 + %2055 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1976, float %1977, float %1978, float %1979, i32 %1739, i32 %1740, i32 %1741, i32 %1742, i32 %1801, i32 %1802) #2, !dbg !74 + %2056 = extractvalue { float, float, float, float } %2055, 0, !dbg !74 + %2057 = extractvalue { float, float, float, float } %2055, 1, !dbg !74 + %2058 = extractvalue { float, float, float, float } %2055, 2, !dbg !74 + %2059 = extractvalue { float, float, float, float } %2055, 3, !dbg !74 + %2060 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1981, float %1982, float %1983, float %1984, i32 %1739, i32 %1740, i32 %1741, i32 %1742, i32 %1811, i32 %1812) #2, !dbg !74 + %2061 = extractvalue { float, float, float, float } %2060, 0, !dbg !74 + %2062 = extractvalue { float, float, float, float } %2060, 1, !dbg !74 + %2063 = extractvalue { float, float, float, float } %2060, 2, !dbg !74 + %2064 = extractvalue { float, float, float, float } %2060, 3, !dbg !74 + %2065 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1986, float %1987, float %1988, float %1989, i32 %1739, i32 %1740, i32 %1741, i32 %1742, i32 %1821, i32 %1822) #2, !dbg !74 + %2066 = extractvalue { float, float, float, float } %2065, 0, !dbg !74 + %2067 = extractvalue { float, float, float, float } %2065, 1, !dbg !74 + %2068 = extractvalue { float, float, float, float } %2065, 2, !dbg !74 + %2069 = extractvalue { float, float, float, float } %2065, 3, !dbg !74 + %2070 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1991, float %1992, float %1993, float %1994, i32 %1739, i32 %1740, i32 %1741, i32 %1742, i32 %1831, i32 %1832) #2, !dbg !74 + %2071 = extractvalue { float, float, float, float } %2070, 0, !dbg !74 + %2072 = extractvalue { float, float, float, float } %2070, 1, !dbg !74 + %2073 = extractvalue { float, float, float, float } %2070, 2, !dbg !74 + %2074 = extractvalue { float, float, float, float } %2070, 3, !dbg !74 + %2075 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %1996, float %1997, float %1998, float %1999, i32 %1745, i32 %1746, i32 %1747, i32 %1748, i32 %1763, i32 %1764) #2, !dbg !74 + %2076 = extractvalue { float, float, float, float } %2075, 0, !dbg !74 + %2077 = extractvalue { float, float, float, float } %2075, 1, !dbg !74 + %2078 = extractvalue { float, float, float, float } %2075, 2, !dbg !74 + %2079 = extractvalue { float, float, float, float } %2075, 3, !dbg !74 + %2080 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2001, float %2002, float %2003, float %2004, i32 %1745, i32 %1746, i32 %1747, i32 %1748, i32 %1773, i32 %1774) #2, !dbg !74 + %2081 = extractvalue { float, float, float, float } %2080, 0, !dbg !74 + %2082 = extractvalue { float, float, float, float } %2080, 1, !dbg !74 + %2083 = extractvalue { float, float, float, float } %2080, 2, !dbg !74 + %2084 = extractvalue { float, float, float, float } %2080, 3, !dbg !74 + %2085 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2006, float %2007, float %2008, float %2009, i32 %1745, i32 %1746, i32 %1747, i32 %1748, i32 %1783, i32 %1784) #2, !dbg !74 + %2086 = extractvalue { float, float, float, float } %2085, 0, !dbg !74 + %2087 = extractvalue { float, float, float, float } %2085, 1, !dbg !74 + %2088 = extractvalue { float, float, float, float } %2085, 2, !dbg !74 + %2089 = extractvalue { float, float, float, float } %2085, 3, !dbg !74 + %2090 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2011, float %2012, float %2013, float %2014, i32 %1745, i32 %1746, i32 %1747, i32 %1748, i32 %1793, i32 %1794) #2, !dbg !74 + %2091 = extractvalue { float, float, float, float } %2090, 0, !dbg !74 + %2092 = extractvalue { float, float, float, float } %2090, 1, !dbg !74 + %2093 = extractvalue { float, float, float, float } %2090, 2, !dbg !74 + %2094 = extractvalue { float, float, float, float } %2090, 3, !dbg !74 + %2095 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2016, float %2017, float %2018, float %2019, i32 %1745, i32 %1746, i32 %1747, i32 %1748, i32 %1803, i32 %1804) #2, !dbg !74 + %2096 = extractvalue { float, float, float, float } %2095, 0, !dbg !74 + %2097 = extractvalue { float, float, float, float } %2095, 1, !dbg !74 + %2098 = extractvalue { float, float, float, float } %2095, 2, !dbg !74 + %2099 = extractvalue { float, float, float, float } %2095, 3, !dbg !74 + %2100 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2021, float %2022, float %2023, float %2024, i32 %1745, i32 %1746, i32 %1747, i32 %1748, i32 %1813, i32 %1814) #2, !dbg !74 + %2101 = extractvalue { float, float, float, float } %2100, 0, !dbg !74 + %2102 = extractvalue { float, float, float, float } %2100, 1, !dbg !74 + %2103 = extractvalue { float, float, float, float } %2100, 2, !dbg !74 + %2104 = extractvalue { float, float, float, float } %2100, 3, !dbg !74 + %2105 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2026, float %2027, float %2028, float %2029, i32 %1745, i32 %1746, i32 %1747, i32 %1748, i32 %1823, i32 %1824) #2, !dbg !74 + %2106 = extractvalue { float, float, float, float } %2105, 0, !dbg !74 + %2107 = extractvalue { float, float, float, float } %2105, 1, !dbg !74 + %2108 = extractvalue { float, float, float, float } %2105, 2, !dbg !74 + %2109 = extractvalue { float, float, float, float } %2105, 3, !dbg !74 + %2110 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2031, float %2032, float %2033, float %2034, i32 %1745, i32 %1746, i32 %1747, i32 %1748, i32 %1833, i32 %1834) #2, !dbg !74 + %2111 = extractvalue { float, float, float, float } %2110, 0, !dbg !74 + %2112 = extractvalue { float, float, float, float } %2110, 1, !dbg !74 + %2113 = extractvalue { float, float, float, float } %2110, 2, !dbg !74 + %2114 = extractvalue { float, float, float, float } %2110, 3, !dbg !74 + %2115 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2036, float %2037, float %2038, float %2039, i32 %1751, i32 %1752, i32 %1753, i32 %1754, i32 %1763, i32 %1764) #2, !dbg !74 + %2116 = extractvalue { float, float, float, float } %2115, 0, !dbg !74 + %2117 = extractvalue { float, float, float, float } %2115, 1, !dbg !74 + %2118 = extractvalue { float, float, float, float } %2115, 2, !dbg !74 + %2119 = extractvalue { float, float, float, float } %2115, 3, !dbg !74 + %2120 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2041, float %2042, float %2043, float %2044, i32 %1751, i32 %1752, i32 %1753, i32 %1754, i32 %1773, i32 %1774) #2, !dbg !74 + %2121 = extractvalue { float, float, float, float } %2120, 0, !dbg !74 + %2122 = extractvalue { float, float, float, float } %2120, 1, !dbg !74 + %2123 = extractvalue { float, float, float, float } %2120, 2, !dbg !74 + %2124 = extractvalue { float, float, float, float } %2120, 3, !dbg !74 + %2125 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2046, float %2047, float %2048, float %2049, i32 %1751, i32 %1752, i32 %1753, i32 %1754, i32 %1783, i32 %1784) #2, !dbg !74 + %2126 = extractvalue { float, float, float, float } %2125, 0, !dbg !74 + %2127 = extractvalue { float, float, float, float } %2125, 1, !dbg !74 + %2128 = extractvalue { float, float, float, float } %2125, 2, !dbg !74 + %2129 = extractvalue { float, float, float, float } %2125, 3, !dbg !74 + %2130 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2051, float %2052, float %2053, float %2054, i32 %1751, i32 %1752, i32 %1753, i32 %1754, i32 %1793, i32 %1794) #2, !dbg !74 + %2131 = extractvalue { float, float, float, float } %2130, 0, !dbg !74 + %2132 = extractvalue { float, float, float, float } %2130, 1, !dbg !74 + %2133 = extractvalue { float, float, float, float } %2130, 2, !dbg !74 + %2134 = extractvalue { float, float, float, float } %2130, 3, !dbg !74 + %2135 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2056, float %2057, float %2058, float %2059, i32 %1751, i32 %1752, i32 %1753, i32 %1754, i32 %1803, i32 %1804) #2, !dbg !74 + %2136 = extractvalue { float, float, float, float } %2135, 0, !dbg !74 + %2137 = extractvalue { float, float, float, float } %2135, 1, !dbg !74 + %2138 = extractvalue { float, float, float, float } %2135, 2, !dbg !74 + %2139 = extractvalue { float, float, float, float } %2135, 3, !dbg !74 + %2140 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2061, float %2062, float %2063, float %2064, i32 %1751, i32 %1752, i32 %1753, i32 %1754, i32 %1813, i32 %1814) #2, !dbg !74 + %2141 = extractvalue { float, float, float, float } %2140, 0, !dbg !74 + %2142 = extractvalue { float, float, float, float } %2140, 1, !dbg !74 + %2143 = extractvalue { float, float, float, float } %2140, 2, !dbg !74 + %2144 = extractvalue { float, float, float, float } %2140, 3, !dbg !74 + %2145 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2066, float %2067, float %2068, float %2069, i32 %1751, i32 %1752, i32 %1753, i32 %1754, i32 %1823, i32 %1824) #2, !dbg !74 + %2146 = extractvalue { float, float, float, float } %2145, 0, !dbg !74 + %2147 = extractvalue { float, float, float, float } %2145, 1, !dbg !74 + %2148 = extractvalue { float, float, float, float } %2145, 2, !dbg !74 + %2149 = extractvalue { float, float, float, float } %2145, 3, !dbg !74 + %2150 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2071, float %2072, float %2073, float %2074, i32 %1751, i32 %1752, i32 %1753, i32 %1754, i32 %1833, i32 %1834) #2, !dbg !74 + %2151 = extractvalue { float, float, float, float } %2150, 0, !dbg !74 + %2152 = extractvalue { float, float, float, float } %2150, 1, !dbg !74 + %2153 = extractvalue { float, float, float, float } %2150, 2, !dbg !74 + %2154 = extractvalue { float, float, float, float } %2150, 3, !dbg !74 + %2155 = getelementptr bfloat, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 49152), i32 %1256, !dbg !48 + %2156 = getelementptr inbounds nuw i8, ptr addrspace(3) %2155, i32 %556, !dbg !75 + %2157 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %2156), !dbg !75 + %2158 = extractvalue { i32, i32, i32, i32 } %2157, 0, !dbg !75 + %2159 = bitcast i32 %2158 to <2 x bfloat>, !dbg !75 + %2160 = extractvalue { i32, i32, i32, i32 } %2157, 1, !dbg !75 + %2161 = bitcast i32 %2160 to <2 x bfloat>, !dbg !75 + %2162 = extractvalue { i32, i32, i32, i32 } %2157, 2, !dbg !75 + %2163 = bitcast i32 %2162 to <2 x bfloat>, !dbg !75 + %2164 = extractvalue { i32, i32, i32, i32 } %2157, 3, !dbg !75 + %2165 = bitcast i32 %2164 to <2 x bfloat>, !dbg !75 + %2166 = getelementptr inbounds nuw i8, ptr addrspace(3) %2155, i32 %558, !dbg !75 + %2167 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %2166), !dbg !75 + %2168 = extractvalue { i32, i32, i32, i32 } %2167, 0, !dbg !75 + %2169 = bitcast i32 %2168 to <2 x bfloat>, !dbg !75 + %2170 = extractvalue { i32, i32, i32, i32 } %2167, 1, !dbg !75 + %2171 = bitcast i32 %2170 to <2 x bfloat>, !dbg !75 + %2172 = extractvalue { i32, i32, i32, i32 } %2167, 2, !dbg !75 + %2173 = bitcast i32 %2172 to <2 x bfloat>, !dbg !75 + %2174 = extractvalue { i32, i32, i32, i32 } %2167, 3, !dbg !75 + %2175 = bitcast i32 %2174 to <2 x bfloat>, !dbg !75 + %2176 = getelementptr inbounds nuw i8, ptr addrspace(3) %2155, i32 %560, !dbg !75 + %2177 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %2176), !dbg !75 + %2178 = extractvalue { i32, i32, i32, i32 } %2177, 0, !dbg !75 + %2179 = bitcast i32 %2178 to <2 x bfloat>, !dbg !75 + %2180 = extractvalue { i32, i32, i32, i32 } %2177, 1, !dbg !75 + %2181 = bitcast i32 %2180 to <2 x bfloat>, !dbg !75 + %2182 = extractvalue { i32, i32, i32, i32 } %2177, 2, !dbg !75 + %2183 = bitcast i32 %2182 to <2 x bfloat>, !dbg !75 + %2184 = extractvalue { i32, i32, i32, i32 } %2177, 3, !dbg !75 + %2185 = bitcast i32 %2184 to <2 x bfloat>, !dbg !75 + %2186 = getelementptr inbounds nuw i8, ptr addrspace(3) %2155, i32 %562, !dbg !75 + %2187 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %2186), !dbg !75 + %2188 = extractvalue { i32, i32, i32, i32 } %2187, 0, !dbg !75 + %2189 = bitcast i32 %2188 to <2 x bfloat>, !dbg !75 + %2190 = extractvalue { i32, i32, i32, i32 } %2187, 1, !dbg !75 + %2191 = bitcast i32 %2190 to <2 x bfloat>, !dbg !75 + %2192 = extractvalue { i32, i32, i32, i32 } %2187, 2, !dbg !75 + %2193 = bitcast i32 %2192 to <2 x bfloat>, !dbg !75 + %2194 = extractvalue { i32, i32, i32, i32 } %2187, 3, !dbg !75 + %2195 = bitcast i32 %2194 to <2 x bfloat>, !dbg !75 + %2196 = getelementptr inbounds nuw i8, ptr addrspace(3) %2155, i32 %564, !dbg !75 + %2197 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %2196), !dbg !75 + %2198 = extractvalue { i32, i32, i32, i32 } %2197, 0, !dbg !75 + %2199 = bitcast i32 %2198 to <2 x bfloat>, !dbg !75 + %2200 = extractvalue { i32, i32, i32, i32 } %2197, 1, !dbg !75 + %2201 = bitcast i32 %2200 to <2 x bfloat>, !dbg !75 + %2202 = extractvalue { i32, i32, i32, i32 } %2197, 2, !dbg !75 + %2203 = bitcast i32 %2202 to <2 x bfloat>, !dbg !75 + %2204 = extractvalue { i32, i32, i32, i32 } %2197, 3, !dbg !75 + %2205 = bitcast i32 %2204 to <2 x bfloat>, !dbg !75 + %2206 = getelementptr inbounds nuw i8, ptr addrspace(3) %2155, i32 %566, !dbg !75 + %2207 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %2206), !dbg !75 + %2208 = extractvalue { i32, i32, i32, i32 } %2207, 0, !dbg !75 + %2209 = bitcast i32 %2208 to <2 x bfloat>, !dbg !75 + %2210 = extractvalue { i32, i32, i32, i32 } %2207, 1, !dbg !75 + %2211 = bitcast i32 %2210 to <2 x bfloat>, !dbg !75 + %2212 = extractvalue { i32, i32, i32, i32 } %2207, 2, !dbg !75 + %2213 = bitcast i32 %2212 to <2 x bfloat>, !dbg !75 + %2214 = extractvalue { i32, i32, i32, i32 } %2207, 3, !dbg !75 + %2215 = bitcast i32 %2214 to <2 x bfloat>, !dbg !75 + %2216 = getelementptr inbounds nuw i8, ptr addrspace(3) %2155, i32 %568, !dbg !75 + %2217 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %2216), !dbg !75 + %2218 = extractvalue { i32, i32, i32, i32 } %2217, 0, !dbg !75 + %2219 = bitcast i32 %2218 to <2 x bfloat>, !dbg !75 + %2220 = extractvalue { i32, i32, i32, i32 } %2217, 1, !dbg !75 + %2221 = bitcast i32 %2220 to <2 x bfloat>, !dbg !75 + %2222 = extractvalue { i32, i32, i32, i32 } %2217, 2, !dbg !75 + %2223 = bitcast i32 %2222 to <2 x bfloat>, !dbg !75 + %2224 = extractvalue { i32, i32, i32, i32 } %2217, 3, !dbg !75 + %2225 = bitcast i32 %2224 to <2 x bfloat>, !dbg !75 + %2226 = getelementptr inbounds nuw i8, ptr addrspace(3) %2155, i32 %570, !dbg !75 + %2227 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %2226), !dbg !75 + %2228 = extractvalue { i32, i32, i32, i32 } %2227, 0, !dbg !75 + %2229 = bitcast i32 %2228 to <2 x bfloat>, !dbg !75 + %2230 = extractvalue { i32, i32, i32, i32 } %2227, 1, !dbg !75 + %2231 = bitcast i32 %2230 to <2 x bfloat>, !dbg !75 + %2232 = extractvalue { i32, i32, i32, i32 } %2227, 2, !dbg !75 + %2233 = bitcast i32 %2232 to <2 x bfloat>, !dbg !75 + %2234 = extractvalue { i32, i32, i32, i32 } %2227, 3, !dbg !75 + %2235 = bitcast i32 %2234 to <2 x bfloat>, !dbg !75 + %gep = getelementptr bfloat, ptr addrspace(1) %invariant.gep, i64 %.pn116.pn.pn547, !dbg !76 + %gep514 = getelementptr bfloat, ptr addrspace(1) %invariant.gep, i64 %.pn100.pn.pn548, !dbg !76 + %gep516 = getelementptr bfloat, ptr addrspace(1) %invariant.gep, i64 %.pn84.pn.pn549, !dbg !76 + %gep518 = getelementptr bfloat, ptr addrspace(1) %invariant.gep, i64 %.pn68.pn.pn550, !dbg !76 + %gep520 = getelementptr bfloat, ptr addrspace(1) %invariant.gep, i64 %.pn52.pn.pn551, !dbg !76 + %gep522 = getelementptr bfloat, ptr addrspace(1) %invariant.gep, i64 %.pn36.pn.pn552, !dbg !76 + %gep524 = getelementptr bfloat, ptr addrspace(1) %invariant.gep, i64 %.pn20.pn.pn553, !dbg !76 + %gep526 = getelementptr bfloat, ptr addrspace(1) %invariant.gep, i64 %.pn4.pn.pn554, !dbg !76 + %2236 = fpext <2 x bfloat> %2159 to <2 x float>, !dbg !75 + %2237 = insertelement <2 x float> poison, float %2076, i64 0, !dbg !75 + %2238 = insertelement <2 x float> %2237, float %2077, i64 1, !dbg !75 + %2239 = fsub <2 x float> %2236, %2238, !dbg !75 + %2240 = fptrunc <2 x float> %2239 to <2 x bfloat>, !dbg !77 + %2241 = fpext <2 x bfloat> %2161 to <2 x float>, !dbg !75 + %2242 = insertelement <2 x float> poison, float %2078, i64 0, !dbg !75 + %2243 = insertelement <2 x float> %2242, float %2079, i64 1, !dbg !75 + %2244 = fsub <2 x float> %2241, %2243, !dbg !75 + %2245 = fptrunc <2 x float> %2244 to <2 x bfloat>, !dbg !77 + %2246 = fpext <2 x bfloat> %2163 to <2 x float>, !dbg !75 + %2247 = insertelement <2 x float> poison, float %2081, i64 0, !dbg !75 + %2248 = insertelement <2 x float> %2247, float %2082, i64 1, !dbg !75 + %2249 = fsub <2 x float> %2246, %2248, !dbg !75 + %2250 = fptrunc <2 x float> %2249 to <2 x bfloat>, !dbg !77 + %2251 = fpext <2 x bfloat> %2165 to <2 x float>, !dbg !75 + %2252 = insertelement <2 x float> poison, float %2083, i64 0, !dbg !75 + %2253 = insertelement <2 x float> %2252, float %2084, i64 1, !dbg !75 + %2254 = fsub <2 x float> %2251, %2253, !dbg !75 + %2255 = fptrunc <2 x float> %2254 to <2 x bfloat>, !dbg !77 + %2256 = fpext <2 x bfloat> %2179 to <2 x float>, !dbg !75 + %2257 = insertelement <2 x float> poison, float %2086, i64 0, !dbg !75 + %2258 = insertelement <2 x float> %2257, float %2087, i64 1, !dbg !75 + %2259 = fsub <2 x float> %2256, %2258, !dbg !75 + %2260 = fptrunc <2 x float> %2259 to <2 x bfloat>, !dbg !77 + %2261 = fpext <2 x bfloat> %2181 to <2 x float>, !dbg !75 + %2262 = insertelement <2 x float> poison, float %2088, i64 0, !dbg !75 + %2263 = insertelement <2 x float> %2262, float %2089, i64 1, !dbg !75 + %2264 = fsub <2 x float> %2261, %2263, !dbg !75 + %2265 = fptrunc <2 x float> %2264 to <2 x bfloat>, !dbg !77 + %2266 = fpext <2 x bfloat> %2183 to <2 x float>, !dbg !75 + %2267 = insertelement <2 x float> poison, float %2091, i64 0, !dbg !75 + %2268 = insertelement <2 x float> %2267, float %2092, i64 1, !dbg !75 + %2269 = fsub <2 x float> %2266, %2268, !dbg !75 + %2270 = fptrunc <2 x float> %2269 to <2 x bfloat>, !dbg !77 + %2271 = fpext <2 x bfloat> %2185 to <2 x float>, !dbg !75 + %2272 = insertelement <2 x float> poison, float %2093, i64 0, !dbg !75 + %2273 = insertelement <2 x float> %2272, float %2094, i64 1, !dbg !75 + %2274 = fsub <2 x float> %2271, %2273, !dbg !75 + %2275 = fptrunc <2 x float> %2274 to <2 x bfloat>, !dbg !77 + %2276 = fpext <2 x bfloat> %2199 to <2 x float>, !dbg !75 + %2277 = insertelement <2 x float> poison, float %2096, i64 0, !dbg !75 + %2278 = insertelement <2 x float> %2277, float %2097, i64 1, !dbg !75 + %2279 = fsub <2 x float> %2276, %2278, !dbg !75 + %2280 = fptrunc <2 x float> %2279 to <2 x bfloat>, !dbg !77 + %2281 = fpext <2 x bfloat> %2201 to <2 x float>, !dbg !75 + %2282 = insertelement <2 x float> poison, float %2098, i64 0, !dbg !75 + %2283 = insertelement <2 x float> %2282, float %2099, i64 1, !dbg !75 + %2284 = fsub <2 x float> %2281, %2283, !dbg !75 + %2285 = fptrunc <2 x float> %2284 to <2 x bfloat>, !dbg !77 + %2286 = fpext <2 x bfloat> %2203 to <2 x float>, !dbg !75 + %2287 = insertelement <2 x float> poison, float %2101, i64 0, !dbg !75 + %2288 = insertelement <2 x float> %2287, float %2102, i64 1, !dbg !75 + %2289 = fsub <2 x float> %2286, %2288, !dbg !75 + %2290 = fptrunc <2 x float> %2289 to <2 x bfloat>, !dbg !77 + %2291 = fpext <2 x bfloat> %2205 to <2 x float>, !dbg !75 + %2292 = insertelement <2 x float> poison, float %2103, i64 0, !dbg !75 + %2293 = insertelement <2 x float> %2292, float %2104, i64 1, !dbg !75 + %2294 = fsub <2 x float> %2291, %2293, !dbg !75 + %2295 = fptrunc <2 x float> %2294 to <2 x bfloat>, !dbg !77 + %2296 = fpext <2 x bfloat> %2219 to <2 x float>, !dbg !75 + %2297 = insertelement <2 x float> poison, float %2106, i64 0, !dbg !75 + %2298 = insertelement <2 x float> %2297, float %2107, i64 1, !dbg !75 + %2299 = fsub <2 x float> %2296, %2298, !dbg !75 + %2300 = fptrunc <2 x float> %2299 to <2 x bfloat>, !dbg !77 + %2301 = fpext <2 x bfloat> %2221 to <2 x float>, !dbg !75 + %2302 = insertelement <2 x float> poison, float %2108, i64 0, !dbg !75 + %2303 = insertelement <2 x float> %2302, float %2109, i64 1, !dbg !75 + %2304 = fsub <2 x float> %2301, %2303, !dbg !75 + %2305 = fptrunc <2 x float> %2304 to <2 x bfloat>, !dbg !77 + %2306 = fpext <2 x bfloat> %2223 to <2 x float>, !dbg !75 + %2307 = insertelement <2 x float> poison, float %2111, i64 0, !dbg !75 + %2308 = insertelement <2 x float> %2307, float %2112, i64 1, !dbg !75 + %2309 = fsub <2 x float> %2306, %2308, !dbg !75 + %2310 = fptrunc <2 x float> %2309 to <2 x bfloat>, !dbg !77 + %2311 = fpext <2 x bfloat> %2225 to <2 x float>, !dbg !75 + %2312 = insertelement <2 x float> poison, float %2113, i64 0, !dbg !75 + %2313 = insertelement <2 x float> %2312, float %2114, i64 1, !dbg !75 + %2314 = fsub <2 x float> %2311, %2313, !dbg !75 + %2315 = fptrunc <2 x float> %2314 to <2 x bfloat>, !dbg !77 + %2316 = fpext <2 x bfloat> %2169 to <2 x float>, !dbg !75 + %2317 = insertelement <2 x float> poison, float %2116, i64 0, !dbg !75 + %2318 = insertelement <2 x float> %2317, float %2117, i64 1, !dbg !75 + %2319 = fsub <2 x float> %2316, %2318, !dbg !75 + %2320 = fptrunc <2 x float> %2319 to <2 x bfloat>, !dbg !77 + %2321 = fpext <2 x bfloat> %2171 to <2 x float>, !dbg !75 + %2322 = insertelement <2 x float> poison, float %2118, i64 0, !dbg !75 + %2323 = insertelement <2 x float> %2322, float %2119, i64 1, !dbg !75 + %2324 = fsub <2 x float> %2321, %2323, !dbg !75 + %2325 = fptrunc <2 x float> %2324 to <2 x bfloat>, !dbg !77 + %2326 = fpext <2 x bfloat> %2173 to <2 x float>, !dbg !75 + %2327 = insertelement <2 x float> poison, float %2121, i64 0, !dbg !75 + %2328 = insertelement <2 x float> %2327, float %2122, i64 1, !dbg !75 + %2329 = fsub <2 x float> %2326, %2328, !dbg !75 + %2330 = fptrunc <2 x float> %2329 to <2 x bfloat>, !dbg !77 + %2331 = fpext <2 x bfloat> %2175 to <2 x float>, !dbg !75 + %2332 = insertelement <2 x float> poison, float %2123, i64 0, !dbg !75 + %2333 = insertelement <2 x float> %2332, float %2124, i64 1, !dbg !75 + %2334 = fsub <2 x float> %2331, %2333, !dbg !75 + %2335 = fptrunc <2 x float> %2334 to <2 x bfloat>, !dbg !77 + %2336 = fpext <2 x bfloat> %2189 to <2 x float>, !dbg !75 + %2337 = insertelement <2 x float> poison, float %2126, i64 0, !dbg !75 + %2338 = insertelement <2 x float> %2337, float %2127, i64 1, !dbg !75 + %2339 = fsub <2 x float> %2336, %2338, !dbg !75 + %2340 = fptrunc <2 x float> %2339 to <2 x bfloat>, !dbg !77 + %2341 = fpext <2 x bfloat> %2191 to <2 x float>, !dbg !75 + %2342 = insertelement <2 x float> poison, float %2128, i64 0, !dbg !75 + %2343 = insertelement <2 x float> %2342, float %2129, i64 1, !dbg !75 + %2344 = fsub <2 x float> %2341, %2343, !dbg !75 + %2345 = fptrunc <2 x float> %2344 to <2 x bfloat>, !dbg !77 + %2346 = fpext <2 x bfloat> %2193 to <2 x float>, !dbg !75 + %2347 = insertelement <2 x float> poison, float %2131, i64 0, !dbg !75 + %2348 = insertelement <2 x float> %2347, float %2132, i64 1, !dbg !75 + %2349 = fsub <2 x float> %2346, %2348, !dbg !75 + %2350 = fptrunc <2 x float> %2349 to <2 x bfloat>, !dbg !77 + %2351 = fpext <2 x bfloat> %2195 to <2 x float>, !dbg !75 + %2352 = insertelement <2 x float> poison, float %2133, i64 0, !dbg !75 + %2353 = insertelement <2 x float> %2352, float %2134, i64 1, !dbg !75 + %2354 = fsub <2 x float> %2351, %2353, !dbg !75 + %2355 = fptrunc <2 x float> %2354 to <2 x bfloat>, !dbg !77 + %2356 = fpext <2 x bfloat> %2209 to <2 x float>, !dbg !75 + %2357 = insertelement <2 x float> poison, float %2136, i64 0, !dbg !75 + %2358 = insertelement <2 x float> %2357, float %2137, i64 1, !dbg !75 + %2359 = fsub <2 x float> %2356, %2358, !dbg !75 + %2360 = fptrunc <2 x float> %2359 to <2 x bfloat>, !dbg !77 + %2361 = fpext <2 x bfloat> %2211 to <2 x float>, !dbg !75 + %2362 = insertelement <2 x float> poison, float %2138, i64 0, !dbg !75 + %2363 = insertelement <2 x float> %2362, float %2139, i64 1, !dbg !75 + %2364 = fsub <2 x float> %2361, %2363, !dbg !75 + %2365 = fptrunc <2 x float> %2364 to <2 x bfloat>, !dbg !77 + %2366 = fpext <2 x bfloat> %2213 to <2 x float>, !dbg !75 + %2367 = insertelement <2 x float> poison, float %2141, i64 0, !dbg !75 + %2368 = insertelement <2 x float> %2367, float %2142, i64 1, !dbg !75 + %2369 = fsub <2 x float> %2366, %2368, !dbg !75 + %2370 = fptrunc <2 x float> %2369 to <2 x bfloat>, !dbg !77 + %2371 = fpext <2 x bfloat> %2215 to <2 x float>, !dbg !75 + %2372 = insertelement <2 x float> poison, float %2143, i64 0, !dbg !75 + %2373 = insertelement <2 x float> %2372, float %2144, i64 1, !dbg !75 + %2374 = fsub <2 x float> %2371, %2373, !dbg !75 + %2375 = fptrunc <2 x float> %2374 to <2 x bfloat>, !dbg !77 + %2376 = fpext <2 x bfloat> %2229 to <2 x float>, !dbg !75 + %2377 = insertelement <2 x float> poison, float %2146, i64 0, !dbg !75 + %2378 = insertelement <2 x float> %2377, float %2147, i64 1, !dbg !75 + %2379 = fsub <2 x float> %2376, %2378, !dbg !75 + %2380 = fptrunc <2 x float> %2379 to <2 x bfloat>, !dbg !77 + %2381 = fpext <2 x bfloat> %2231 to <2 x float>, !dbg !75 + %2382 = insertelement <2 x float> poison, float %2148, i64 0, !dbg !75 + %2383 = insertelement <2 x float> %2382, float %2149, i64 1, !dbg !75 + %2384 = fsub <2 x float> %2381, %2383, !dbg !75 + %2385 = fptrunc <2 x float> %2384 to <2 x bfloat>, !dbg !77 + %2386 = fpext <2 x bfloat> %2233 to <2 x float>, !dbg !75 + %2387 = insertelement <2 x float> poison, float %2151, i64 0, !dbg !75 + %2388 = insertelement <2 x float> %2387, float %2152, i64 1, !dbg !75 + %2389 = fsub <2 x float> %2386, %2388, !dbg !75 + %2390 = fptrunc <2 x float> %2389 to <2 x bfloat>, !dbg !77 + %2391 = fpext <2 x bfloat> %2235 to <2 x float>, !dbg !75 + %2392 = insertelement <2 x float> poison, float %2153, i64 0, !dbg !75 + %2393 = insertelement <2 x float> %2392, float %2154, i64 1, !dbg !75 + %2394 = fsub <2 x float> %2391, %2393, !dbg !75 + %2395 = fptrunc <2 x float> %2394 to <2 x bfloat>, !dbg !77 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !77 + %2396 = bitcast <2 x bfloat> %2240 to i32, !dbg !77 + %2397 = bitcast <2 x bfloat> %2250 to i32, !dbg !77 + %2398 = bitcast <2 x bfloat> %2260 to i32, !dbg !77 + %2399 = bitcast <2 x bfloat> %2270 to i32, !dbg !77 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %664, i32 %2396, i32 %2397, i32 %2398, i32 %2399) #2, !dbg !77 + %2400 = bitcast <2 x bfloat> %2280 to i32, !dbg !77 + %2401 = bitcast <2 x bfloat> %2290 to i32, !dbg !77 + %2402 = bitcast <2 x bfloat> %2300 to i32, !dbg !77 + %2403 = bitcast <2 x bfloat> %2310 to i32, !dbg !77 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %665, i32 %2400, i32 %2401, i32 %2402, i32 %2403) #2, !dbg !77 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !77 + %2404 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %666), !dbg !77 + %2405 = extractvalue { i32, i32, i32, i32 } %2404, 0, !dbg !77 + %2406 = extractvalue { i32, i32, i32, i32 } %2404, 1, !dbg !77 + %2407 = extractvalue { i32, i32, i32, i32 } %2404, 2, !dbg !77 + %2408 = extractvalue { i32, i32, i32, i32 } %2404, 3, !dbg !77 + %2409 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %667), !dbg !77 + %2410 = extractvalue { i32, i32, i32, i32 } %2409, 0, !dbg !77 + %2411 = extractvalue { i32, i32, i32, i32 } %2409, 1, !dbg !77 + %2412 = extractvalue { i32, i32, i32, i32 } %2409, 2, !dbg !77 + %2413 = extractvalue { i32, i32, i32, i32 } %2409, 3, !dbg !77 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !77 + %2414 = bitcast <2 x bfloat> %2245 to i32, !dbg !77 + %2415 = bitcast <2 x bfloat> %2255 to i32, !dbg !77 + %2416 = bitcast <2 x bfloat> %2265 to i32, !dbg !77 + %2417 = bitcast <2 x bfloat> %2275 to i32, !dbg !77 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %664, i32 %2414, i32 %2415, i32 %2416, i32 %2417) #2, !dbg !77 + %2418 = bitcast <2 x bfloat> %2285 to i32, !dbg !77 + %2419 = bitcast <2 x bfloat> %2295 to i32, !dbg !77 + %2420 = bitcast <2 x bfloat> %2305 to i32, !dbg !77 + %2421 = bitcast <2 x bfloat> %2315 to i32, !dbg !77 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %665, i32 %2418, i32 %2419, i32 %2420, i32 %2421) #2, !dbg !77 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !77 + %2422 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %666), !dbg !77 + %2423 = extractvalue { i32, i32, i32, i32 } %2422, 0, !dbg !77 + %2424 = extractvalue { i32, i32, i32, i32 } %2422, 1, !dbg !77 + %2425 = extractvalue { i32, i32, i32, i32 } %2422, 2, !dbg !77 + %2426 = extractvalue { i32, i32, i32, i32 } %2422, 3, !dbg !77 + %2427 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %667), !dbg !77 + %2428 = extractvalue { i32, i32, i32, i32 } %2427, 0, !dbg !77 + %2429 = extractvalue { i32, i32, i32, i32 } %2427, 1, !dbg !77 + %2430 = extractvalue { i32, i32, i32, i32 } %2427, 2, !dbg !77 + %2431 = extractvalue { i32, i32, i32, i32 } %2427, 3, !dbg !77 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !77 + %2432 = bitcast <2 x bfloat> %2320 to i32, !dbg !77 + %2433 = bitcast <2 x bfloat> %2330 to i32, !dbg !77 + %2434 = bitcast <2 x bfloat> %2340 to i32, !dbg !77 + %2435 = bitcast <2 x bfloat> %2350 to i32, !dbg !77 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %664, i32 %2432, i32 %2433, i32 %2434, i32 %2435) #2, !dbg !77 + %2436 = bitcast <2 x bfloat> %2360 to i32, !dbg !77 + %2437 = bitcast <2 x bfloat> %2370 to i32, !dbg !77 + %2438 = bitcast <2 x bfloat> %2380 to i32, !dbg !77 + %2439 = bitcast <2 x bfloat> %2390 to i32, !dbg !77 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %665, i32 %2436, i32 %2437, i32 %2438, i32 %2439) #2, !dbg !77 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !77 + %2440 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %666), !dbg !77 + %2441 = extractvalue { i32, i32, i32, i32 } %2440, 0, !dbg !77 + %2442 = extractvalue { i32, i32, i32, i32 } %2440, 1, !dbg !77 + %2443 = extractvalue { i32, i32, i32, i32 } %2440, 2, !dbg !77 + %2444 = extractvalue { i32, i32, i32, i32 } %2440, 3, !dbg !77 + %2445 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %667), !dbg !77 + %2446 = extractvalue { i32, i32, i32, i32 } %2445, 0, !dbg !77 + %2447 = extractvalue { i32, i32, i32, i32 } %2445, 1, !dbg !77 + %2448 = extractvalue { i32, i32, i32, i32 } %2445, 2, !dbg !77 + %2449 = extractvalue { i32, i32, i32, i32 } %2445, 3, !dbg !77 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !77 + %2450 = bitcast <2 x bfloat> %2325 to i32, !dbg !77 + %2451 = bitcast <2 x bfloat> %2335 to i32, !dbg !77 + %2452 = bitcast <2 x bfloat> %2345 to i32, !dbg !77 + %2453 = bitcast <2 x bfloat> %2355 to i32, !dbg !77 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %664, i32 %2450, i32 %2451, i32 %2452, i32 %2453) #2, !dbg !77 + %2454 = bitcast <2 x bfloat> %2365 to i32, !dbg !77 + %2455 = bitcast <2 x bfloat> %2375 to i32, !dbg !77 + %2456 = bitcast <2 x bfloat> %2385 to i32, !dbg !77 + %2457 = bitcast <2 x bfloat> %2395 to i32, !dbg !77 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %665, i32 %2454, i32 %2455, i32 %2456, i32 %2457) #2, !dbg !77 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !77 + %2458 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %666), !dbg !77 + %2459 = extractvalue { i32, i32, i32, i32 } %2458, 0, !dbg !77 + %2460 = extractvalue { i32, i32, i32, i32 } %2458, 1, !dbg !77 + %2461 = extractvalue { i32, i32, i32, i32 } %2458, 2, !dbg !77 + %2462 = extractvalue { i32, i32, i32, i32 } %2458, 3, !dbg !77 + %2463 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %667), !dbg !77 + %2464 = extractvalue { i32, i32, i32, i32 } %2463, 0, !dbg !77 + %2465 = extractvalue { i32, i32, i32, i32 } %2463, 1, !dbg !77 + %2466 = extractvalue { i32, i32, i32, i32 } %2463, 2, !dbg !77 + %2467 = extractvalue { i32, i32, i32, i32 } %2463, 3, !dbg !77 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %2405, i32 %2406, i32 %2407, i32 %2408, ptr addrspace(1) %gep, i1 %.pn355.pn.pn571) #2, !dbg !78 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %2423, i32 %2424, i32 %2425, i32 %2426, ptr addrspace(1) %gep514, i1 %.pn339.pn.pn572) #2, !dbg !78 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %2410, i32 %2411, i32 %2412, i32 %2413, ptr addrspace(1) %gep516, i1 %.pn323.pn.pn573) #2, !dbg !78 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %2428, i32 %2429, i32 %2430, i32 %2431, ptr addrspace(1) %gep518, i1 %.pn307.pn.pn574) #2, !dbg !78 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %2441, i32 %2442, i32 %2443, i32 %2444, ptr addrspace(1) %gep520, i1 %.pn291.pn.pn575) #2, !dbg !78 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %2459, i32 %2460, i32 %2461, i32 %2462, ptr addrspace(1) %gep522, i1 %.pn275.pn.pn576) #2, !dbg !78 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %2446, i32 %2447, i32 %2448, i32 %2449, ptr addrspace(1) %gep524, i1 %.pn259.pn.pn577) #2, !dbg !78 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %2464, i32 %2465, i32 %2466, i32 %2467, ptr addrspace(1) %gep526, i1 %.pn243.pn.pn578) #2, !dbg !78 + %2468 = getelementptr float, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 142080), i32 %822, !dbg !53 + %2469 = load float, ptr addrspace(3) %2468, align 4, !dbg !53 + %.idx498 = shl i32 %822, 8, !dbg !55 + %gep610 = getelementptr i8, ptr addrspace(3) %invariant.gep609, i32 %.idx498, !dbg !55 + %2470 = load float, ptr addrspace(3) %gep610, align 4, !dbg !55 + %2471 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep610, i32 32, !dbg !55 + %2472 = load float, ptr addrspace(3) %2471, align 4, !dbg !55 + %2473 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep610, i32 128, !dbg !55 + %2474 = load float, ptr addrspace(3) %2473, align 4, !dbg !55 + %2475 = getelementptr inbounds nuw i8, ptr addrspace(3) %gep610, i32 160, !dbg !55 + %2476 = load float, ptr addrspace(3) %2475, align 4, !dbg !55 + %2477 = fsub float %2469, %2470, !dbg !79 + %2478 = fsub float %2469, %2472, !dbg !79 + %2479 = fsub float %2469, %2474, !dbg !79 + %2480 = fsub float %2469, %2476, !dbg !79 + %2481 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !80 + %.not.i = icmp eq i32 %2481, 0, !dbg !80 + br i1 %.not.i, label %2484, label %2482, !dbg !80 + +2482: ; preds = %684 + %2483 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %2477), !dbg !80 + br label %__nv_exp2f.exit, !dbg !80 + +2484: ; preds = %684 + %2485 = tail call float @llvm.nvvm.ex2.approx.f32(float %2477), !dbg !80 + br label %__nv_exp2f.exit, !dbg !80 + +__nv_exp2f.exit: ; preds = %2482, %2484 + %.0.i = phi float [ %2483, %2482 ], [ %2485, %2484 ], !dbg !80 + %2486 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !80 + %.not.i501 = icmp eq i32 %2486, 0, !dbg !80 + br i1 %.not.i501, label %2489, label %2487, !dbg !80 + +2487: ; preds = %__nv_exp2f.exit + %2488 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %2478), !dbg !80 + br label %__nv_exp2f.exit503, !dbg !80 + +2489: ; preds = %__nv_exp2f.exit + %2490 = tail call float @llvm.nvvm.ex2.approx.f32(float %2478), !dbg !80 + br label %__nv_exp2f.exit503, !dbg !80 + +__nv_exp2f.exit503: ; preds = %2487, %2489 + %.0.i502 = phi float [ %2488, %2487 ], [ %2490, %2489 ], !dbg !80 + %2491 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !80 + %.not.i504 = icmp eq i32 %2491, 0, !dbg !80 + br i1 %.not.i504, label %2494, label %2492, !dbg !80 + +2492: ; preds = %__nv_exp2f.exit503 + %2493 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %2479), !dbg !80 + br label %__nv_exp2f.exit506, !dbg !80 + +2494: ; preds = %__nv_exp2f.exit503 + %2495 = tail call float @llvm.nvvm.ex2.approx.f32(float %2479), !dbg !80 + br label %__nv_exp2f.exit506, !dbg !80 + +__nv_exp2f.exit506: ; preds = %2492, %2494 + %.0.i505 = phi float [ %2493, %2492 ], [ %2495, %2494 ], !dbg !80 + %2496 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !80 + %.not.i507 = icmp eq i32 %2496, 0, !dbg !80 + br i1 %.not.i507, label %2499, label %2497, !dbg !80 + +2497: ; preds = %__nv_exp2f.exit506 + %2498 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %2480), !dbg !80 + br label %__nv_exp2f.exit509, !dbg !80 + +2499: ; preds = %__nv_exp2f.exit506 + %2500 = tail call float @llvm.nvvm.ex2.approx.f32(float %2480), !dbg !80 + br label %__nv_exp2f.exit509, !dbg !80 + +__nv_exp2f.exit509: ; preds = %2497, %2499 + %.0.i508 = phi float [ %2498, %2497 ], [ %2500, %2499 ], !dbg !80 + %2501 = extractelement <4 x i1> %815, i64 3, !dbg !84 + %2502 = select i1 %2501, float %.0.i, float 0.000000e+00, !dbg !84 + %2503 = extractelement <4 x i1> %815, i64 2, !dbg !84 + %2504 = select i1 %2503, float %.0.i502, float 0.000000e+00, !dbg !84 + %2505 = extractelement <4 x i1> %815, i64 1, !dbg !84 + %2506 = select i1 %2505, float %.0.i505, float 0.000000e+00, !dbg !84 + %2507 = extractelement <4 x i1> %815, i64 0, !dbg !84 + %2508 = select i1 %2507, float %.0.i508, float 0.000000e+00, !dbg !84 + %2509 = extractelement <2 x float> %2239, i64 0, !dbg !85 + %2510 = fmul float %2509, %2502, !dbg !85 + %2511 = extractelement <2 x float> %2239, i64 1, !dbg !85 + %2512 = fmul float %2511, %2502, !dbg !85 + %2513 = extractelement <2 x float> %2244, i64 0, !dbg !85 + %2514 = fmul float %2513, %2504, !dbg !85 + %2515 = extractelement <2 x float> %2244, i64 1, !dbg !85 + %2516 = fmul float %2515, %2504, !dbg !85 + %2517 = extractelement <2 x float> %2249, i64 0, !dbg !85 + %2518 = fmul float %2517, %2502, !dbg !85 + %2519 = extractelement <2 x float> %2249, i64 1, !dbg !85 + %2520 = fmul float %2519, %2502, !dbg !85 + %2521 = extractelement <2 x float> %2254, i64 0, !dbg !85 + %2522 = fmul float %2521, %2504, !dbg !85 + %2523 = extractelement <2 x float> %2254, i64 1, !dbg !85 + %2524 = fmul float %2523, %2504, !dbg !85 + %2525 = extractelement <2 x float> %2259, i64 0, !dbg !85 + %2526 = fmul float %2525, %2502, !dbg !85 + %2527 = extractelement <2 x float> %2259, i64 1, !dbg !85 + %2528 = fmul float %2527, %2502, !dbg !85 + %2529 = extractelement <2 x float> %2264, i64 0, !dbg !85 + %2530 = fmul float %2529, %2504, !dbg !85 + %2531 = extractelement <2 x float> %2264, i64 1, !dbg !85 + %2532 = fmul float %2531, %2504, !dbg !85 + %2533 = extractelement <2 x float> %2269, i64 0, !dbg !85 + %2534 = fmul float %2533, %2502, !dbg !85 + %2535 = extractelement <2 x float> %2269, i64 1, !dbg !85 + %2536 = fmul float %2535, %2502, !dbg !85 + %2537 = extractelement <2 x float> %2274, i64 0, !dbg !85 + %2538 = fmul float %2537, %2504, !dbg !85 + %2539 = extractelement <2 x float> %2274, i64 1, !dbg !85 + %2540 = fmul float %2539, %2504, !dbg !85 + %2541 = extractelement <2 x float> %2279, i64 0, !dbg !85 + %2542 = fmul float %2541, %2502, !dbg !85 + %2543 = extractelement <2 x float> %2279, i64 1, !dbg !85 + %2544 = fmul float %2543, %2502, !dbg !85 + %2545 = extractelement <2 x float> %2284, i64 0, !dbg !85 + %2546 = fmul float %2545, %2504, !dbg !85 + %2547 = extractelement <2 x float> %2284, i64 1, !dbg !85 + %2548 = fmul float %2547, %2504, !dbg !85 + %2549 = extractelement <2 x float> %2289, i64 0, !dbg !85 + %2550 = fmul float %2549, %2502, !dbg !85 + %2551 = extractelement <2 x float> %2289, i64 1, !dbg !85 + %2552 = fmul float %2551, %2502, !dbg !85 + %2553 = extractelement <2 x float> %2294, i64 0, !dbg !85 + %2554 = fmul float %2553, %2504, !dbg !85 + %2555 = extractelement <2 x float> %2294, i64 1, !dbg !85 + %2556 = fmul float %2555, %2504, !dbg !85 + %2557 = extractelement <2 x float> %2299, i64 0, !dbg !85 + %2558 = fmul float %2557, %2502, !dbg !85 + %2559 = extractelement <2 x float> %2299, i64 1, !dbg !85 + %2560 = fmul float %2559, %2502, !dbg !85 + %2561 = extractelement <2 x float> %2304, i64 0, !dbg !85 + %2562 = fmul float %2561, %2504, !dbg !85 + %2563 = extractelement <2 x float> %2304, i64 1, !dbg !85 + %2564 = fmul float %2563, %2504, !dbg !85 + %2565 = extractelement <2 x float> %2309, i64 0, !dbg !85 + %2566 = fmul float %2565, %2502, !dbg !85 + %2567 = extractelement <2 x float> %2309, i64 1, !dbg !85 + %2568 = fmul float %2567, %2502, !dbg !85 + %2569 = extractelement <2 x float> %2314, i64 0, !dbg !85 + %2570 = fmul float %2569, %2504, !dbg !85 + %2571 = extractelement <2 x float> %2314, i64 1, !dbg !85 + %2572 = fmul float %2571, %2504, !dbg !85 + %2573 = extractelement <2 x float> %2319, i64 0, !dbg !85 + %2574 = fmul float %2573, %2506, !dbg !85 + %2575 = extractelement <2 x float> %2319, i64 1, !dbg !85 + %2576 = fmul float %2575, %2506, !dbg !85 + %2577 = extractelement <2 x float> %2324, i64 0, !dbg !85 + %2578 = fmul float %2577, %2508, !dbg !85 + %2579 = extractelement <2 x float> %2324, i64 1, !dbg !85 + %2580 = fmul float %2579, %2508, !dbg !85 + %2581 = extractelement <2 x float> %2329, i64 0, !dbg !85 + %2582 = fmul float %2581, %2506, !dbg !85 + %2583 = extractelement <2 x float> %2329, i64 1, !dbg !85 + %2584 = fmul float %2583, %2506, !dbg !85 + %2585 = extractelement <2 x float> %2334, i64 0, !dbg !85 + %2586 = fmul float %2585, %2508, !dbg !85 + %2587 = extractelement <2 x float> %2334, i64 1, !dbg !85 + %2588 = fmul float %2587, %2508, !dbg !85 + %2589 = extractelement <2 x float> %2339, i64 0, !dbg !85 + %2590 = fmul float %2589, %2506, !dbg !85 + %2591 = extractelement <2 x float> %2339, i64 1, !dbg !85 + %2592 = fmul float %2591, %2506, !dbg !85 + %2593 = extractelement <2 x float> %2344, i64 0, !dbg !85 + %2594 = fmul float %2593, %2508, !dbg !85 + %2595 = extractelement <2 x float> %2344, i64 1, !dbg !85 + %2596 = fmul float %2595, %2508, !dbg !85 + %2597 = extractelement <2 x float> %2349, i64 0, !dbg !85 + %2598 = fmul float %2597, %2506, !dbg !85 + %2599 = extractelement <2 x float> %2349, i64 1, !dbg !85 + %2600 = fmul float %2599, %2506, !dbg !85 + %2601 = extractelement <2 x float> %2354, i64 0, !dbg !85 + %2602 = fmul float %2601, %2508, !dbg !85 + %2603 = extractelement <2 x float> %2354, i64 1, !dbg !85 + %2604 = fmul float %2603, %2508, !dbg !85 + %2605 = extractelement <2 x float> %2359, i64 0, !dbg !85 + %2606 = fmul float %2605, %2506, !dbg !85 + %2607 = extractelement <2 x float> %2359, i64 1, !dbg !85 + %2608 = fmul float %2607, %2506, !dbg !85 + %2609 = extractelement <2 x float> %2364, i64 0, !dbg !85 + %2610 = fmul float %2609, %2508, !dbg !85 + %2611 = extractelement <2 x float> %2364, i64 1, !dbg !85 + %2612 = fmul float %2611, %2508, !dbg !85 + %2613 = extractelement <2 x float> %2369, i64 0, !dbg !85 + %2614 = fmul float %2613, %2506, !dbg !85 + %2615 = extractelement <2 x float> %2369, i64 1, !dbg !85 + %2616 = fmul float %2615, %2506, !dbg !85 + %2617 = extractelement <2 x float> %2374, i64 0, !dbg !85 + %2618 = fmul float %2617, %2508, !dbg !85 + %2619 = extractelement <2 x float> %2374, i64 1, !dbg !85 + %2620 = fmul float %2619, %2508, !dbg !85 + %2621 = extractelement <2 x float> %2379, i64 0, !dbg !85 + %2622 = fmul float %2621, %2506, !dbg !85 + %2623 = extractelement <2 x float> %2379, i64 1, !dbg !85 + %2624 = fmul float %2623, %2506, !dbg !85 + %2625 = extractelement <2 x float> %2384, i64 0, !dbg !85 + %2626 = fmul float %2625, %2508, !dbg !85 + %2627 = extractelement <2 x float> %2384, i64 1, !dbg !85 + %2628 = fmul float %2627, %2508, !dbg !85 + %2629 = extractelement <2 x float> %2389, i64 0, !dbg !85 + %2630 = fmul float %2629, %2506, !dbg !85 + %2631 = extractelement <2 x float> %2389, i64 1, !dbg !85 + %2632 = fmul float %2631, %2506, !dbg !85 + %2633 = extractelement <2 x float> %2394, i64 0, !dbg !85 + %2634 = fmul float %2633, %2508, !dbg !85 + %2635 = extractelement <2 x float> %2394, i64 1, !dbg !85 + %2636 = fmul float %2635, %2508, !dbg !85 + %2637 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #2, !dbg !86 + %.not.i510 = icmp eq i32 %2637, 0, !dbg !86 + br i1 %.not.i510, label %2640, label %2638, !dbg !86 + +2638: ; preds = %__nv_exp2f.exit509 + %2639 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %2469), !dbg !86 + br label %__nv_exp2f.exit512, !dbg !86 + +2640: ; preds = %__nv_exp2f.exit509 + %2641 = tail call float @llvm.nvvm.ex2.approx.f32(float %2469), !dbg !86 + br label %__nv_exp2f.exit512, !dbg !86 + +__nv_exp2f.exit512: ; preds = %2638, %2640 + %.0.i511 = phi float [ %2639, %2638 ], [ %2641, %2640 ], !dbg !86 + %2642 = fmul float %751, %.0.i511, !dbg !88 + %2643 = fmul float %752, %.0.i511, !dbg !88 + %2644 = fmul float %753, %.0.i511, !dbg !88 + %2645 = fmul float %754, %.0.i511, !dbg !88 + %2646 = fmul float %755, %.0.i511, !dbg !88 + %2647 = fmul float %756, %.0.i511, !dbg !88 + %2648 = fmul float %757, %.0.i511, !dbg !88 + %2649 = fmul float %758, %.0.i511, !dbg !88 + %2650 = fmul float %759, %.0.i511, !dbg !88 + %2651 = fmul float %760, %.0.i511, !dbg !88 + %2652 = fmul float %761, %.0.i511, !dbg !88 + %2653 = fmul float %762, %.0.i511, !dbg !88 + %2654 = fmul float %763, %.0.i511, !dbg !88 + %2655 = fmul float %764, %.0.i511, !dbg !88 + %2656 = fmul float %765, %.0.i511, !dbg !88 + %2657 = fmul float %766, %.0.i511, !dbg !88 + %2658 = fmul float %767, %.0.i511, !dbg !88 + %2659 = fmul float %768, %.0.i511, !dbg !88 + %2660 = fmul float %769, %.0.i511, !dbg !88 + %2661 = fmul float %770, %.0.i511, !dbg !88 + %2662 = fmul float %771, %.0.i511, !dbg !88 + %2663 = fmul float %772, %.0.i511, !dbg !88 + %2664 = fmul float %773, %.0.i511, !dbg !88 + %2665 = fmul float %774, %.0.i511, !dbg !88 + %2666 = fmul float %775, %.0.i511, !dbg !88 + %2667 = fmul float %776, %.0.i511, !dbg !88 + %2668 = fmul float %777, %.0.i511, !dbg !88 + %2669 = fmul float %778, %.0.i511, !dbg !88 + %2670 = fmul float %779, %.0.i511, !dbg !88 + %2671 = fmul float %780, %.0.i511, !dbg !88 + %2672 = fmul float %781, %.0.i511, !dbg !88 + %2673 = fmul float %782, %.0.i511, !dbg !88 + %2674 = fmul float %783, %.0.i511, !dbg !88 + %2675 = fmul float %784, %.0.i511, !dbg !88 + %2676 = fmul float %785, %.0.i511, !dbg !88 + %2677 = fmul float %786, %.0.i511, !dbg !88 + %2678 = fmul float %787, %.0.i511, !dbg !88 + %2679 = fmul float %788, %.0.i511, !dbg !88 + %2680 = fmul float %789, %.0.i511, !dbg !88 + %2681 = fmul float %790, %.0.i511, !dbg !88 + %2682 = fmul float %791, %.0.i511, !dbg !88 + %2683 = fmul float %792, %.0.i511, !dbg !88 + %2684 = fmul float %793, %.0.i511, !dbg !88 + %2685 = fmul float %794, %.0.i511, !dbg !88 + %2686 = fmul float %795, %.0.i511, !dbg !88 + %2687 = fmul float %796, %.0.i511, !dbg !88 + %2688 = fmul float %797, %.0.i511, !dbg !88 + %2689 = fmul float %798, %.0.i511, !dbg !88 + %2690 = fmul float %799, %.0.i511, !dbg !88 + %2691 = fmul float %800, %.0.i511, !dbg !88 + %2692 = fmul float %801, %.0.i511, !dbg !88 + %2693 = fmul float %802, %.0.i511, !dbg !88 + %2694 = fmul float %803, %.0.i511, !dbg !88 + %2695 = fmul float %804, %.0.i511, !dbg !88 + %2696 = fmul float %805, %.0.i511, !dbg !88 + %2697 = fmul float %806, %.0.i511, !dbg !88 + %2698 = fmul float %807, %.0.i511, !dbg !88 + %2699 = fmul float %808, %.0.i511, !dbg !88 + %2700 = fmul float %809, %.0.i511, !dbg !88 + %2701 = fmul float %810, %.0.i511, !dbg !88 + %2702 = fmul float %811, %.0.i511, !dbg !88 + %2703 = fmul float %812, %.0.i511, !dbg !88 + %2704 = fmul float %813, %.0.i511, !dbg !88 + %2705 = fmul float %814, %.0.i511, !dbg !88 + %2706 = fmul float %687, %.0.i511, !dbg !89 + %2707 = fmul float %688, %.0.i511, !dbg !89 + %2708 = fmul float %689, %.0.i511, !dbg !89 + %2709 = fmul float %690, %.0.i511, !dbg !89 + %2710 = fmul float %691, %.0.i511, !dbg !89 + %2711 = fmul float %692, %.0.i511, !dbg !89 + %2712 = fmul float %693, %.0.i511, !dbg !89 + %2713 = fmul float %694, %.0.i511, !dbg !89 + %2714 = fmul float %695, %.0.i511, !dbg !89 + %2715 = fmul float %696, %.0.i511, !dbg !89 + %2716 = fmul float %697, %.0.i511, !dbg !89 + %2717 = fmul float %698, %.0.i511, !dbg !89 + %2718 = fmul float %699, %.0.i511, !dbg !89 + %2719 = fmul float %700, %.0.i511, !dbg !89 + %2720 = fmul float %701, %.0.i511, !dbg !89 + %2721 = fmul float %702, %.0.i511, !dbg !89 + %2722 = fmul float %703, %.0.i511, !dbg !89 + %2723 = fmul float %704, %.0.i511, !dbg !89 + %2724 = fmul float %705, %.0.i511, !dbg !89 + %2725 = fmul float %706, %.0.i511, !dbg !89 + %2726 = fmul float %707, %.0.i511, !dbg !89 + %2727 = fmul float %708, %.0.i511, !dbg !89 + %2728 = fmul float %709, %.0.i511, !dbg !89 + %2729 = fmul float %710, %.0.i511, !dbg !89 + %2730 = fmul float %711, %.0.i511, !dbg !89 + %2731 = fmul float %712, %.0.i511, !dbg !89 + %2732 = fmul float %713, %.0.i511, !dbg !89 + %2733 = fmul float %714, %.0.i511, !dbg !89 + %2734 = fmul float %715, %.0.i511, !dbg !89 + %2735 = fmul float %716, %.0.i511, !dbg !89 + %2736 = fmul float %717, %.0.i511, !dbg !89 + %2737 = fmul float %718, %.0.i511, !dbg !89 + %2738 = fmul float %719, %.0.i511, !dbg !89 + %2739 = fmul float %720, %.0.i511, !dbg !89 + %2740 = fmul float %721, %.0.i511, !dbg !89 + %2741 = fmul float %722, %.0.i511, !dbg !89 + %2742 = fmul float %723, %.0.i511, !dbg !89 + %2743 = fmul float %724, %.0.i511, !dbg !89 + %2744 = fmul float %725, %.0.i511, !dbg !89 + %2745 = fmul float %726, %.0.i511, !dbg !89 + %2746 = fmul float %727, %.0.i511, !dbg !89 + %2747 = fmul float %728, %.0.i511, !dbg !89 + %2748 = fmul float %729, %.0.i511, !dbg !89 + %2749 = fmul float %730, %.0.i511, !dbg !89 + %2750 = fmul float %731, %.0.i511, !dbg !89 + %2751 = fmul float %732, %.0.i511, !dbg !89 + %2752 = fmul float %733, %.0.i511, !dbg !89 + %2753 = fmul float %734, %.0.i511, !dbg !89 + %2754 = fmul float %735, %.0.i511, !dbg !89 + %2755 = fmul float %736, %.0.i511, !dbg !89 + %2756 = fmul float %737, %.0.i511, !dbg !89 + %2757 = fmul float %738, %.0.i511, !dbg !89 + %2758 = fmul float %739, %.0.i511, !dbg !89 + %2759 = fmul float %740, %.0.i511, !dbg !89 + %2760 = fmul float %741, %.0.i511, !dbg !89 + %2761 = fmul float %742, %.0.i511, !dbg !89 + %2762 = fmul float %743, %.0.i511, !dbg !89 + %2763 = fmul float %744, %.0.i511, !dbg !89 + %2764 = fmul float %745, %.0.i511, !dbg !89 + %2765 = fmul float %746, %.0.i511, !dbg !89 + %2766 = fmul float %747, %.0.i511, !dbg !89 + %2767 = fmul float %748, %.0.i511, !dbg !89 + %2768 = fmul float %749, %.0.i511, !dbg !89 + %2769 = fmul float %750, %.0.i511, !dbg !89 + %2770 = fptrunc float %2510 to bfloat, !dbg !90 + %2771 = fptrunc float %2512 to bfloat, !dbg !90 + %2772 = fptrunc float %2514 to bfloat, !dbg !90 + %2773 = fptrunc float %2516 to bfloat, !dbg !90 + %2774 = fptrunc float %2518 to bfloat, !dbg !90 + %2775 = fptrunc float %2520 to bfloat, !dbg !90 + %2776 = fptrunc float %2522 to bfloat, !dbg !90 + %2777 = fptrunc float %2524 to bfloat, !dbg !90 + %2778 = fptrunc float %2526 to bfloat, !dbg !90 + %2779 = fptrunc float %2528 to bfloat, !dbg !90 + %2780 = fptrunc float %2530 to bfloat, !dbg !90 + %2781 = fptrunc float %2532 to bfloat, !dbg !90 + %2782 = fptrunc float %2534 to bfloat, !dbg !90 + %2783 = fptrunc float %2536 to bfloat, !dbg !90 + %2784 = fptrunc float %2538 to bfloat, !dbg !90 + %2785 = fptrunc float %2540 to bfloat, !dbg !90 + %2786 = fptrunc float %2542 to bfloat, !dbg !90 + %2787 = fptrunc float %2544 to bfloat, !dbg !90 + %2788 = fptrunc float %2546 to bfloat, !dbg !90 + %2789 = fptrunc float %2548 to bfloat, !dbg !90 + %2790 = fptrunc float %2550 to bfloat, !dbg !90 + %2791 = fptrunc float %2552 to bfloat, !dbg !90 + %2792 = fptrunc float %2554 to bfloat, !dbg !90 + %2793 = fptrunc float %2556 to bfloat, !dbg !90 + %2794 = fptrunc float %2558 to bfloat, !dbg !90 + %2795 = fptrunc float %2560 to bfloat, !dbg !90 + %2796 = fptrunc float %2562 to bfloat, !dbg !90 + %2797 = fptrunc float %2564 to bfloat, !dbg !90 + %2798 = fptrunc float %2566 to bfloat, !dbg !90 + %2799 = fptrunc float %2568 to bfloat, !dbg !90 + %2800 = fptrunc float %2570 to bfloat, !dbg !90 + %2801 = fptrunc float %2572 to bfloat, !dbg !90 + %2802 = fptrunc float %2574 to bfloat, !dbg !90 + %2803 = fptrunc float %2576 to bfloat, !dbg !90 + %2804 = fptrunc float %2578 to bfloat, !dbg !90 + %2805 = fptrunc float %2580 to bfloat, !dbg !90 + %2806 = fptrunc float %2582 to bfloat, !dbg !90 + %2807 = fptrunc float %2584 to bfloat, !dbg !90 + %2808 = fptrunc float %2586 to bfloat, !dbg !90 + %2809 = fptrunc float %2588 to bfloat, !dbg !90 + %2810 = fptrunc float %2590 to bfloat, !dbg !90 + %2811 = fptrunc float %2592 to bfloat, !dbg !90 + %2812 = fptrunc float %2594 to bfloat, !dbg !90 + %2813 = fptrunc float %2596 to bfloat, !dbg !90 + %2814 = fptrunc float %2598 to bfloat, !dbg !90 + %2815 = fptrunc float %2600 to bfloat, !dbg !90 + %2816 = fptrunc float %2602 to bfloat, !dbg !90 + %2817 = fptrunc float %2604 to bfloat, !dbg !90 + %2818 = fptrunc float %2606 to bfloat, !dbg !90 + %2819 = fptrunc float %2608 to bfloat, !dbg !90 + %2820 = fptrunc float %2610 to bfloat, !dbg !90 + %2821 = fptrunc float %2612 to bfloat, !dbg !90 + %2822 = fptrunc float %2614 to bfloat, !dbg !90 + %2823 = fptrunc float %2616 to bfloat, !dbg !90 + %2824 = fptrunc float %2618 to bfloat, !dbg !90 + %2825 = fptrunc float %2620 to bfloat, !dbg !90 + %2826 = fptrunc float %2622 to bfloat, !dbg !90 + %2827 = fptrunc float %2624 to bfloat, !dbg !90 + %2828 = fptrunc float %2626 to bfloat, !dbg !90 + %2829 = fptrunc float %2628 to bfloat, !dbg !90 + %2830 = fptrunc float %2630 to bfloat, !dbg !90 + %2831 = fptrunc float %2632 to bfloat, !dbg !90 + %2832 = fptrunc float %2634 to bfloat, !dbg !90 + %2833 = fptrunc float %2636 to bfloat, !dbg !90 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !90 + %2834 = insertelement <2 x bfloat> poison, bfloat %2770, i64 0, !dbg !90 + %2835 = insertelement <2 x bfloat> %2834, bfloat %2771, i64 1, !dbg !90 + %2836 = bitcast <2 x bfloat> %2835 to i32, !dbg !90 + %2837 = insertelement <2 x bfloat> poison, bfloat %2772, i64 0, !dbg !90 + %2838 = insertelement <2 x bfloat> %2837, bfloat %2773, i64 1, !dbg !90 + %2839 = bitcast <2 x bfloat> %2838 to i32, !dbg !90 + %2840 = insertelement <2 x bfloat> poison, bfloat %2774, i64 0, !dbg !90 + %2841 = insertelement <2 x bfloat> %2840, bfloat %2775, i64 1, !dbg !90 + %2842 = bitcast <2 x bfloat> %2841 to i32, !dbg !90 + %2843 = insertelement <2 x bfloat> poison, bfloat %2776, i64 0, !dbg !90 + %2844 = insertelement <2 x bfloat> %2843, bfloat %2777, i64 1, !dbg !90 + %2845 = bitcast <2 x bfloat> %2844 to i32, !dbg !90 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %557, i32 %2836, i32 %2839, i32 %2842, i32 %2845), !dbg !90 + %2846 = insertelement <2 x bfloat> poison, bfloat %2802, i64 0, !dbg !90 + %2847 = insertelement <2 x bfloat> %2846, bfloat %2803, i64 1, !dbg !90 + %2848 = bitcast <2 x bfloat> %2847 to i32, !dbg !90 + %2849 = insertelement <2 x bfloat> poison, bfloat %2804, i64 0, !dbg !90 + %2850 = insertelement <2 x bfloat> %2849, bfloat %2805, i64 1, !dbg !90 + %2851 = bitcast <2 x bfloat> %2850 to i32, !dbg !90 + %2852 = insertelement <2 x bfloat> poison, bfloat %2806, i64 0, !dbg !90 + %2853 = insertelement <2 x bfloat> %2852, bfloat %2807, i64 1, !dbg !90 + %2854 = bitcast <2 x bfloat> %2853 to i32, !dbg !90 + %2855 = insertelement <2 x bfloat> poison, bfloat %2808, i64 0, !dbg !90 + %2856 = insertelement <2 x bfloat> %2855, bfloat %2809, i64 1, !dbg !90 + %2857 = bitcast <2 x bfloat> %2856 to i32, !dbg !90 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %559, i32 %2848, i32 %2851, i32 %2854, i32 %2857), !dbg !90 + %2858 = insertelement <2 x bfloat> poison, bfloat %2778, i64 0, !dbg !90 + %2859 = insertelement <2 x bfloat> %2858, bfloat %2779, i64 1, !dbg !90 + %2860 = bitcast <2 x bfloat> %2859 to i32, !dbg !90 + %2861 = insertelement <2 x bfloat> poison, bfloat %2780, i64 0, !dbg !90 + %2862 = insertelement <2 x bfloat> %2861, bfloat %2781, i64 1, !dbg !90 + %2863 = bitcast <2 x bfloat> %2862 to i32, !dbg !90 + %2864 = insertelement <2 x bfloat> poison, bfloat %2782, i64 0, !dbg !90 + %2865 = insertelement <2 x bfloat> %2864, bfloat %2783, i64 1, !dbg !90 + %2866 = bitcast <2 x bfloat> %2865 to i32, !dbg !90 + %2867 = insertelement <2 x bfloat> poison, bfloat %2784, i64 0, !dbg !90 + %2868 = insertelement <2 x bfloat> %2867, bfloat %2785, i64 1, !dbg !90 + %2869 = bitcast <2 x bfloat> %2868 to i32, !dbg !90 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %561, i32 %2860, i32 %2863, i32 %2866, i32 %2869), !dbg !90 + %2870 = insertelement <2 x bfloat> poison, bfloat %2810, i64 0, !dbg !90 + %2871 = insertelement <2 x bfloat> %2870, bfloat %2811, i64 1, !dbg !90 + %2872 = bitcast <2 x bfloat> %2871 to i32, !dbg !90 + %2873 = insertelement <2 x bfloat> poison, bfloat %2812, i64 0, !dbg !90 + %2874 = insertelement <2 x bfloat> %2873, bfloat %2813, i64 1, !dbg !90 + %2875 = bitcast <2 x bfloat> %2874 to i32, !dbg !90 + %2876 = insertelement <2 x bfloat> poison, bfloat %2814, i64 0, !dbg !90 + %2877 = insertelement <2 x bfloat> %2876, bfloat %2815, i64 1, !dbg !90 + %2878 = bitcast <2 x bfloat> %2877 to i32, !dbg !90 + %2879 = insertelement <2 x bfloat> poison, bfloat %2816, i64 0, !dbg !90 + %2880 = insertelement <2 x bfloat> %2879, bfloat %2817, i64 1, !dbg !90 + %2881 = bitcast <2 x bfloat> %2880 to i32, !dbg !90 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %563, i32 %2872, i32 %2875, i32 %2878, i32 %2881), !dbg !90 + %2882 = insertelement <2 x bfloat> poison, bfloat %2786, i64 0, !dbg !90 + %2883 = insertelement <2 x bfloat> %2882, bfloat %2787, i64 1, !dbg !90 + %2884 = bitcast <2 x bfloat> %2883 to i32, !dbg !90 + %2885 = insertelement <2 x bfloat> poison, bfloat %2788, i64 0, !dbg !90 + %2886 = insertelement <2 x bfloat> %2885, bfloat %2789, i64 1, !dbg !90 + %2887 = bitcast <2 x bfloat> %2886 to i32, !dbg !90 + %2888 = insertelement <2 x bfloat> poison, bfloat %2790, i64 0, !dbg !90 + %2889 = insertelement <2 x bfloat> %2888, bfloat %2791, i64 1, !dbg !90 + %2890 = bitcast <2 x bfloat> %2889 to i32, !dbg !90 + %2891 = insertelement <2 x bfloat> poison, bfloat %2792, i64 0, !dbg !90 + %2892 = insertelement <2 x bfloat> %2891, bfloat %2793, i64 1, !dbg !90 + %2893 = bitcast <2 x bfloat> %2892 to i32, !dbg !90 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %565, i32 %2884, i32 %2887, i32 %2890, i32 %2893), !dbg !90 + %2894 = insertelement <2 x bfloat> poison, bfloat %2818, i64 0, !dbg !90 + %2895 = insertelement <2 x bfloat> %2894, bfloat %2819, i64 1, !dbg !90 + %2896 = bitcast <2 x bfloat> %2895 to i32, !dbg !90 + %2897 = insertelement <2 x bfloat> poison, bfloat %2820, i64 0, !dbg !90 + %2898 = insertelement <2 x bfloat> %2897, bfloat %2821, i64 1, !dbg !90 + %2899 = bitcast <2 x bfloat> %2898 to i32, !dbg !90 + %2900 = insertelement <2 x bfloat> poison, bfloat %2822, i64 0, !dbg !90 + %2901 = insertelement <2 x bfloat> %2900, bfloat %2823, i64 1, !dbg !90 + %2902 = bitcast <2 x bfloat> %2901 to i32, !dbg !90 + %2903 = insertelement <2 x bfloat> poison, bfloat %2824, i64 0, !dbg !90 + %2904 = insertelement <2 x bfloat> %2903, bfloat %2825, i64 1, !dbg !90 + %2905 = bitcast <2 x bfloat> %2904 to i32, !dbg !90 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %567, i32 %2896, i32 %2899, i32 %2902, i32 %2905), !dbg !90 + %2906 = insertelement <2 x bfloat> poison, bfloat %2794, i64 0, !dbg !90 + %2907 = insertelement <2 x bfloat> %2906, bfloat %2795, i64 1, !dbg !90 + %2908 = bitcast <2 x bfloat> %2907 to i32, !dbg !90 + %2909 = insertelement <2 x bfloat> poison, bfloat %2796, i64 0, !dbg !90 + %2910 = insertelement <2 x bfloat> %2909, bfloat %2797, i64 1, !dbg !90 + %2911 = bitcast <2 x bfloat> %2910 to i32, !dbg !90 + %2912 = insertelement <2 x bfloat> poison, bfloat %2798, i64 0, !dbg !90 + %2913 = insertelement <2 x bfloat> %2912, bfloat %2799, i64 1, !dbg !90 + %2914 = bitcast <2 x bfloat> %2913 to i32, !dbg !90 + %2915 = insertelement <2 x bfloat> poison, bfloat %2800, i64 0, !dbg !90 + %2916 = insertelement <2 x bfloat> %2915, bfloat %2801, i64 1, !dbg !90 + %2917 = bitcast <2 x bfloat> %2916 to i32, !dbg !90 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %569, i32 %2908, i32 %2911, i32 %2914, i32 %2917), !dbg !90 + %2918 = insertelement <2 x bfloat> poison, bfloat %2826, i64 0, !dbg !90 + %2919 = insertelement <2 x bfloat> %2918, bfloat %2827, i64 1, !dbg !90 + %2920 = bitcast <2 x bfloat> %2919 to i32, !dbg !90 + %2921 = insertelement <2 x bfloat> poison, bfloat %2828, i64 0, !dbg !90 + %2922 = insertelement <2 x bfloat> %2921, bfloat %2829, i64 1, !dbg !90 + %2923 = bitcast <2 x bfloat> %2922 to i32, !dbg !90 + %2924 = insertelement <2 x bfloat> poison, bfloat %2830, i64 0, !dbg !90 + %2925 = insertelement <2 x bfloat> %2924, bfloat %2831, i64 1, !dbg !90 + %2926 = bitcast <2 x bfloat> %2925 to i32, !dbg !90 + %2927 = insertelement <2 x bfloat> poison, bfloat %2832, i64 0, !dbg !90 + %2928 = insertelement <2 x bfloat> %2927, bfloat %2833, i64 1, !dbg !90 + %2929 = bitcast <2 x bfloat> %2928 to i32, !dbg !90 + tail call void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) nonnull %571, i32 %2920, i32 %2923, i32 %2926, i32 %2929), !dbg !90 + %2930 = getelementptr bfloat, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73728), i32 %1256, !dbg !57 + %2931 = getelementptr inbounds nuw i8, ptr addrspace(3) %2930, i32 %674, !dbg !57 + %2932 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %2931), !dbg !57 + %2933 = extractvalue { i32, i32, i32, i32 } %2932, 0, !dbg !57 + %2934 = extractvalue { i32, i32, i32, i32 } %2932, 1, !dbg !57 + %2935 = extractvalue { i32, i32, i32, i32 } %2932, 2, !dbg !57 + %2936 = extractvalue { i32, i32, i32, i32 } %2932, 3, !dbg !57 + %2937 = getelementptr inbounds nuw i8, ptr addrspace(3) %2930, i32 %675, !dbg !57 + %2938 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %2937), !dbg !57 + %2939 = extractvalue { i32, i32, i32, i32 } %2938, 0, !dbg !57 + %2940 = extractvalue { i32, i32, i32, i32 } %2938, 1, !dbg !57 + %2941 = extractvalue { i32, i32, i32, i32 } %2938, 2, !dbg !57 + %2942 = extractvalue { i32, i32, i32, i32 } %2938, 3, !dbg !57 + %2943 = getelementptr inbounds nuw i8, ptr addrspace(3) %2930, i32 %676, !dbg !57 + %2944 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %2943), !dbg !57 + %2945 = extractvalue { i32, i32, i32, i32 } %2944, 0, !dbg !57 + %2946 = extractvalue { i32, i32, i32, i32 } %2944, 1, !dbg !57 + %2947 = extractvalue { i32, i32, i32, i32 } %2944, 2, !dbg !57 + %2948 = extractvalue { i32, i32, i32, i32 } %2944, 3, !dbg !57 + %2949 = getelementptr inbounds nuw i8, ptr addrspace(3) %2930, i32 %677, !dbg !57 + %2950 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %2949), !dbg !57 + %2951 = extractvalue { i32, i32, i32, i32 } %2950, 0, !dbg !57 + %2952 = extractvalue { i32, i32, i32, i32 } %2950, 1, !dbg !57 + %2953 = extractvalue { i32, i32, i32, i32 } %2950, 2, !dbg !57 + %2954 = extractvalue { i32, i32, i32, i32 } %2950, 3, !dbg !57 + %2955 = getelementptr inbounds nuw i8, ptr addrspace(3) %2930, i32 %678, !dbg !57 + %2956 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %2955), !dbg !57 + %2957 = extractvalue { i32, i32, i32, i32 } %2956, 0, !dbg !57 + %2958 = extractvalue { i32, i32, i32, i32 } %2956, 1, !dbg !57 + %2959 = extractvalue { i32, i32, i32, i32 } %2956, 2, !dbg !57 + %2960 = extractvalue { i32, i32, i32, i32 } %2956, 3, !dbg !57 + %2961 = getelementptr inbounds nuw i8, ptr addrspace(3) %2930, i32 %679, !dbg !57 + %2962 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %2961), !dbg !57 + %2963 = extractvalue { i32, i32, i32, i32 } %2962, 0, !dbg !57 + %2964 = extractvalue { i32, i32, i32, i32 } %2962, 1, !dbg !57 + %2965 = extractvalue { i32, i32, i32, i32 } %2962, 2, !dbg !57 + %2966 = extractvalue { i32, i32, i32, i32 } %2962, 3, !dbg !57 + %2967 = getelementptr inbounds nuw i8, ptr addrspace(3) %2930, i32 %680, !dbg !57 + %2968 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %2967), !dbg !57 + %2969 = extractvalue { i32, i32, i32, i32 } %2968, 0, !dbg !57 + %2970 = extractvalue { i32, i32, i32, i32 } %2968, 1, !dbg !57 + %2971 = extractvalue { i32, i32, i32, i32 } %2968, 2, !dbg !57 + %2972 = extractvalue { i32, i32, i32, i32 } %2968, 3, !dbg !57 + %2973 = getelementptr inbounds nuw i8, ptr addrspace(3) %2930, i32 %681, !dbg !57 + %2974 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %2973), !dbg !57 + %2975 = extractvalue { i32, i32, i32, i32 } %2974, 0, !dbg !57 + %2976 = extractvalue { i32, i32, i32, i32 } %2974, 1, !dbg !57 + %2977 = extractvalue { i32, i32, i32, i32 } %2974, 2, !dbg !57 + %2978 = extractvalue { i32, i32, i32, i32 } %2974, 3, !dbg !57 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !90 + %2979 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %617), !dbg !90 + %2980 = extractvalue { i32, i32, i32, i32 } %2979, 0, !dbg !90 + %2981 = extractvalue { i32, i32, i32, i32 } %2979, 1, !dbg !90 + %2982 = extractvalue { i32, i32, i32, i32 } %2979, 2, !dbg !90 + %2983 = extractvalue { i32, i32, i32, i32 } %2979, 3, !dbg !90 + %2984 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %619), !dbg !90 + %2985 = extractvalue { i32, i32, i32, i32 } %2984, 0, !dbg !90 + %2986 = extractvalue { i32, i32, i32, i32 } %2984, 1, !dbg !90 + %2987 = extractvalue { i32, i32, i32, i32 } %2984, 2, !dbg !90 + %2988 = extractvalue { i32, i32, i32, i32 } %2984, 3, !dbg !90 + %2989 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %621), !dbg !90 + %2990 = extractvalue { i32, i32, i32, i32 } %2989, 0, !dbg !90 + %2991 = extractvalue { i32, i32, i32, i32 } %2989, 1, !dbg !90 + %2992 = extractvalue { i32, i32, i32, i32 } %2989, 2, !dbg !90 + %2993 = extractvalue { i32, i32, i32, i32 } %2989, 3, !dbg !90 + %2994 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %623), !dbg !90 + %2995 = extractvalue { i32, i32, i32, i32 } %2994, 0, !dbg !90 + %2996 = extractvalue { i32, i32, i32, i32 } %2994, 1, !dbg !90 + %2997 = extractvalue { i32, i32, i32, i32 } %2994, 2, !dbg !90 + %2998 = extractvalue { i32, i32, i32, i32 } %2994, 3, !dbg !90 + %2999 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %625), !dbg !90 + %3000 = extractvalue { i32, i32, i32, i32 } %2999, 0, !dbg !90 + %3001 = extractvalue { i32, i32, i32, i32 } %2999, 1, !dbg !90 + %3002 = extractvalue { i32, i32, i32, i32 } %2999, 2, !dbg !90 + %3003 = extractvalue { i32, i32, i32, i32 } %2999, 3, !dbg !90 + %3004 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %627), !dbg !90 + %3005 = extractvalue { i32, i32, i32, i32 } %3004, 0, !dbg !90 + %3006 = extractvalue { i32, i32, i32, i32 } %3004, 1, !dbg !90 + %3007 = extractvalue { i32, i32, i32, i32 } %3004, 2, !dbg !90 + %3008 = extractvalue { i32, i32, i32, i32 } %3004, 3, !dbg !90 + %3009 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %629), !dbg !90 + %3010 = extractvalue { i32, i32, i32, i32 } %3009, 0, !dbg !90 + %3011 = extractvalue { i32, i32, i32, i32 } %3009, 1, !dbg !90 + %3012 = extractvalue { i32, i32, i32, i32 } %3009, 2, !dbg !90 + %3013 = extractvalue { i32, i32, i32, i32 } %3009, 3, !dbg !90 + %3014 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %631), !dbg !90 + %3015 = extractvalue { i32, i32, i32, i32 } %3014, 0, !dbg !90 + %3016 = extractvalue { i32, i32, i32, i32 } %3014, 1, !dbg !90 + %3017 = extractvalue { i32, i32, i32, i32 } %3014, 2, !dbg !90 + %3018 = extractvalue { i32, i32, i32, i32 } %3014, 3, !dbg !90 + %3019 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %633), !dbg !90 + %3020 = extractvalue { i32, i32, i32, i32 } %3019, 0, !dbg !90 + %3021 = extractvalue { i32, i32, i32, i32 } %3019, 1, !dbg !90 + %3022 = extractvalue { i32, i32, i32, i32 } %3019, 2, !dbg !90 + %3023 = extractvalue { i32, i32, i32, i32 } %3019, 3, !dbg !90 + %3024 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %635), !dbg !90 + %3025 = extractvalue { i32, i32, i32, i32 } %3024, 0, !dbg !90 + %3026 = extractvalue { i32, i32, i32, i32 } %3024, 1, !dbg !90 + %3027 = extractvalue { i32, i32, i32, i32 } %3024, 2, !dbg !90 + %3028 = extractvalue { i32, i32, i32, i32 } %3024, 3, !dbg !90 + %3029 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %637), !dbg !90 + %3030 = extractvalue { i32, i32, i32, i32 } %3029, 0, !dbg !90 + %3031 = extractvalue { i32, i32, i32, i32 } %3029, 1, !dbg !90 + %3032 = extractvalue { i32, i32, i32, i32 } %3029, 2, !dbg !90 + %3033 = extractvalue { i32, i32, i32, i32 } %3029, 3, !dbg !90 + %3034 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %639), !dbg !90 + %3035 = extractvalue { i32, i32, i32, i32 } %3034, 0, !dbg !90 + %3036 = extractvalue { i32, i32, i32, i32 } %3034, 1, !dbg !90 + %3037 = extractvalue { i32, i32, i32, i32 } %3034, 2, !dbg !90 + %3038 = extractvalue { i32, i32, i32, i32 } %3034, 3, !dbg !90 + %3039 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %641), !dbg !90 + %3040 = extractvalue { i32, i32, i32, i32 } %3039, 0, !dbg !90 + %3041 = extractvalue { i32, i32, i32, i32 } %3039, 1, !dbg !90 + %3042 = extractvalue { i32, i32, i32, i32 } %3039, 2, !dbg !90 + %3043 = extractvalue { i32, i32, i32, i32 } %3039, 3, !dbg !90 + %3044 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %643), !dbg !90 + %3045 = extractvalue { i32, i32, i32, i32 } %3044, 0, !dbg !90 + %3046 = extractvalue { i32, i32, i32, i32 } %3044, 1, !dbg !90 + %3047 = extractvalue { i32, i32, i32, i32 } %3044, 2, !dbg !90 + %3048 = extractvalue { i32, i32, i32, i32 } %3044, 3, !dbg !90 + %3049 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %645), !dbg !90 + %3050 = extractvalue { i32, i32, i32, i32 } %3049, 0, !dbg !90 + %3051 = extractvalue { i32, i32, i32, i32 } %3049, 1, !dbg !90 + %3052 = extractvalue { i32, i32, i32, i32 } %3049, 2, !dbg !90 + %3053 = extractvalue { i32, i32, i32, i32 } %3049, 3, !dbg !90 + %3054 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %647), !dbg !90 + %3055 = extractvalue { i32, i32, i32, i32 } %3054, 0, !dbg !90 + %3056 = extractvalue { i32, i32, i32, i32 } %3054, 1, !dbg !90 + %3057 = extractvalue { i32, i32, i32, i32 } %3054, 2, !dbg !90 + %3058 = extractvalue { i32, i32, i32, i32 } %3054, 3, !dbg !90 + %3059 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2642, float %2643, float %2644, float %2645, i32 %2933, i32 %2934, i32 %2935, i32 %2936, i32 %2980, i32 %2981) #2, !dbg !91 + %3060 = extractvalue { float, float, float, float } %3059, 0, !dbg !91 + %3061 = extractvalue { float, float, float, float } %3059, 1, !dbg !91 + %3062 = extractvalue { float, float, float, float } %3059, 2, !dbg !91 + %3063 = extractvalue { float, float, float, float } %3059, 3, !dbg !91 + %3064 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2646, float %2647, float %2648, float %2649, i32 %2933, i32 %2934, i32 %2935, i32 %2936, i32 %2990, i32 %2991) #2, !dbg !91 + %3065 = extractvalue { float, float, float, float } %3064, 0, !dbg !91 + %3066 = extractvalue { float, float, float, float } %3064, 1, !dbg !91 + %3067 = extractvalue { float, float, float, float } %3064, 2, !dbg !91 + %3068 = extractvalue { float, float, float, float } %3064, 3, !dbg !91 + %3069 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2650, float %2651, float %2652, float %2653, i32 %2933, i32 %2934, i32 %2935, i32 %2936, i32 %3000, i32 %3001) #2, !dbg !91 + %3070 = extractvalue { float, float, float, float } %3069, 0, !dbg !91 + %3071 = extractvalue { float, float, float, float } %3069, 1, !dbg !91 + %3072 = extractvalue { float, float, float, float } %3069, 2, !dbg !91 + %3073 = extractvalue { float, float, float, float } %3069, 3, !dbg !91 + %3074 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2654, float %2655, float %2656, float %2657, i32 %2933, i32 %2934, i32 %2935, i32 %2936, i32 %3010, i32 %3011) #2, !dbg !91 + %3075 = extractvalue { float, float, float, float } %3074, 0, !dbg !91 + %3076 = extractvalue { float, float, float, float } %3074, 1, !dbg !91 + %3077 = extractvalue { float, float, float, float } %3074, 2, !dbg !91 + %3078 = extractvalue { float, float, float, float } %3074, 3, !dbg !91 + %3079 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2658, float %2659, float %2660, float %2661, i32 %2933, i32 %2934, i32 %2935, i32 %2936, i32 %3020, i32 %3021) #2, !dbg !91 + %3080 = extractvalue { float, float, float, float } %3079, 0, !dbg !91 + %3081 = extractvalue { float, float, float, float } %3079, 1, !dbg !91 + %3082 = extractvalue { float, float, float, float } %3079, 2, !dbg !91 + %3083 = extractvalue { float, float, float, float } %3079, 3, !dbg !91 + %3084 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2662, float %2663, float %2664, float %2665, i32 %2933, i32 %2934, i32 %2935, i32 %2936, i32 %3030, i32 %3031) #2, !dbg !91 + %3085 = extractvalue { float, float, float, float } %3084, 0, !dbg !91 + %3086 = extractvalue { float, float, float, float } %3084, 1, !dbg !91 + %3087 = extractvalue { float, float, float, float } %3084, 2, !dbg !91 + %3088 = extractvalue { float, float, float, float } %3084, 3, !dbg !91 + %3089 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2666, float %2667, float %2668, float %2669, i32 %2933, i32 %2934, i32 %2935, i32 %2936, i32 %3040, i32 %3041) #2, !dbg !91 + %3090 = extractvalue { float, float, float, float } %3089, 0, !dbg !91 + %3091 = extractvalue { float, float, float, float } %3089, 1, !dbg !91 + %3092 = extractvalue { float, float, float, float } %3089, 2, !dbg !91 + %3093 = extractvalue { float, float, float, float } %3089, 3, !dbg !91 + %3094 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2670, float %2671, float %2672, float %2673, i32 %2933, i32 %2934, i32 %2935, i32 %2936, i32 %3050, i32 %3051) #2, !dbg !91 + %3095 = extractvalue { float, float, float, float } %3094, 0, !dbg !91 + %3096 = extractvalue { float, float, float, float } %3094, 1, !dbg !91 + %3097 = extractvalue { float, float, float, float } %3094, 2, !dbg !91 + %3098 = extractvalue { float, float, float, float } %3094, 3, !dbg !91 + %3099 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2674, float %2675, float %2676, float %2677, i32 %2957, i32 %2958, i32 %2959, i32 %2960, i32 %2980, i32 %2981) #2, !dbg !91 + %3100 = extractvalue { float, float, float, float } %3099, 0, !dbg !91 + %3101 = extractvalue { float, float, float, float } %3099, 1, !dbg !91 + %3102 = extractvalue { float, float, float, float } %3099, 2, !dbg !91 + %3103 = extractvalue { float, float, float, float } %3099, 3, !dbg !91 + %3104 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2678, float %2679, float %2680, float %2681, i32 %2957, i32 %2958, i32 %2959, i32 %2960, i32 %2990, i32 %2991) #2, !dbg !91 + %3105 = extractvalue { float, float, float, float } %3104, 0, !dbg !91 + %3106 = extractvalue { float, float, float, float } %3104, 1, !dbg !91 + %3107 = extractvalue { float, float, float, float } %3104, 2, !dbg !91 + %3108 = extractvalue { float, float, float, float } %3104, 3, !dbg !91 + %3109 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2682, float %2683, float %2684, float %2685, i32 %2957, i32 %2958, i32 %2959, i32 %2960, i32 %3000, i32 %3001) #2, !dbg !91 + %3110 = extractvalue { float, float, float, float } %3109, 0, !dbg !91 + %3111 = extractvalue { float, float, float, float } %3109, 1, !dbg !91 + %3112 = extractvalue { float, float, float, float } %3109, 2, !dbg !91 + %3113 = extractvalue { float, float, float, float } %3109, 3, !dbg !91 + %3114 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2686, float %2687, float %2688, float %2689, i32 %2957, i32 %2958, i32 %2959, i32 %2960, i32 %3010, i32 %3011) #2, !dbg !91 + %3115 = extractvalue { float, float, float, float } %3114, 0, !dbg !91 + %3116 = extractvalue { float, float, float, float } %3114, 1, !dbg !91 + %3117 = extractvalue { float, float, float, float } %3114, 2, !dbg !91 + %3118 = extractvalue { float, float, float, float } %3114, 3, !dbg !91 + %3119 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2690, float %2691, float %2692, float %2693, i32 %2957, i32 %2958, i32 %2959, i32 %2960, i32 %3020, i32 %3021) #2, !dbg !91 + %3120 = extractvalue { float, float, float, float } %3119, 0, !dbg !91 + %3121 = extractvalue { float, float, float, float } %3119, 1, !dbg !91 + %3122 = extractvalue { float, float, float, float } %3119, 2, !dbg !91 + %3123 = extractvalue { float, float, float, float } %3119, 3, !dbg !91 + %3124 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2694, float %2695, float %2696, float %2697, i32 %2957, i32 %2958, i32 %2959, i32 %2960, i32 %3030, i32 %3031) #2, !dbg !91 + %3125 = extractvalue { float, float, float, float } %3124, 0, !dbg !91 + %3126 = extractvalue { float, float, float, float } %3124, 1, !dbg !91 + %3127 = extractvalue { float, float, float, float } %3124, 2, !dbg !91 + %3128 = extractvalue { float, float, float, float } %3124, 3, !dbg !91 + %3129 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2698, float %2699, float %2700, float %2701, i32 %2957, i32 %2958, i32 %2959, i32 %2960, i32 %3040, i32 %3041) #2, !dbg !91 + %3130 = extractvalue { float, float, float, float } %3129, 0, !dbg !91 + %3131 = extractvalue { float, float, float, float } %3129, 1, !dbg !91 + %3132 = extractvalue { float, float, float, float } %3129, 2, !dbg !91 + %3133 = extractvalue { float, float, float, float } %3129, 3, !dbg !91 + %3134 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2702, float %2703, float %2704, float %2705, i32 %2957, i32 %2958, i32 %2959, i32 %2960, i32 %3050, i32 %3051) #2, !dbg !91 + %3135 = extractvalue { float, float, float, float } %3134, 0, !dbg !91 + %3136 = extractvalue { float, float, float, float } %3134, 1, !dbg !91 + %3137 = extractvalue { float, float, float, float } %3134, 2, !dbg !91 + %3138 = extractvalue { float, float, float, float } %3134, 3, !dbg !91 + %3139 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3060, float %3061, float %3062, float %3063, i32 %2939, i32 %2940, i32 %2941, i32 %2942, i32 %2982, i32 %2983) #2, !dbg !91 + %3140 = extractvalue { float, float, float, float } %3139, 0, !dbg !91 + %3141 = extractvalue { float, float, float, float } %3139, 1, !dbg !91 + %3142 = extractvalue { float, float, float, float } %3139, 2, !dbg !91 + %3143 = extractvalue { float, float, float, float } %3139, 3, !dbg !91 + %3144 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3065, float %3066, float %3067, float %3068, i32 %2939, i32 %2940, i32 %2941, i32 %2942, i32 %2992, i32 %2993) #2, !dbg !91 + %3145 = extractvalue { float, float, float, float } %3144, 0, !dbg !91 + %3146 = extractvalue { float, float, float, float } %3144, 1, !dbg !91 + %3147 = extractvalue { float, float, float, float } %3144, 2, !dbg !91 + %3148 = extractvalue { float, float, float, float } %3144, 3, !dbg !91 + %3149 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3070, float %3071, float %3072, float %3073, i32 %2939, i32 %2940, i32 %2941, i32 %2942, i32 %3002, i32 %3003) #2, !dbg !91 + %3150 = extractvalue { float, float, float, float } %3149, 0, !dbg !91 + %3151 = extractvalue { float, float, float, float } %3149, 1, !dbg !91 + %3152 = extractvalue { float, float, float, float } %3149, 2, !dbg !91 + %3153 = extractvalue { float, float, float, float } %3149, 3, !dbg !91 + %3154 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3075, float %3076, float %3077, float %3078, i32 %2939, i32 %2940, i32 %2941, i32 %2942, i32 %3012, i32 %3013) #2, !dbg !91 + %3155 = extractvalue { float, float, float, float } %3154, 0, !dbg !91 + %3156 = extractvalue { float, float, float, float } %3154, 1, !dbg !91 + %3157 = extractvalue { float, float, float, float } %3154, 2, !dbg !91 + %3158 = extractvalue { float, float, float, float } %3154, 3, !dbg !91 + %3159 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3080, float %3081, float %3082, float %3083, i32 %2939, i32 %2940, i32 %2941, i32 %2942, i32 %3022, i32 %3023) #2, !dbg !91 + %3160 = extractvalue { float, float, float, float } %3159, 0, !dbg !91 + %3161 = extractvalue { float, float, float, float } %3159, 1, !dbg !91 + %3162 = extractvalue { float, float, float, float } %3159, 2, !dbg !91 + %3163 = extractvalue { float, float, float, float } %3159, 3, !dbg !91 + %3164 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3085, float %3086, float %3087, float %3088, i32 %2939, i32 %2940, i32 %2941, i32 %2942, i32 %3032, i32 %3033) #2, !dbg !91 + %3165 = extractvalue { float, float, float, float } %3164, 0, !dbg !91 + %3166 = extractvalue { float, float, float, float } %3164, 1, !dbg !91 + %3167 = extractvalue { float, float, float, float } %3164, 2, !dbg !91 + %3168 = extractvalue { float, float, float, float } %3164, 3, !dbg !91 + %3169 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3090, float %3091, float %3092, float %3093, i32 %2939, i32 %2940, i32 %2941, i32 %2942, i32 %3042, i32 %3043) #2, !dbg !91 + %3170 = extractvalue { float, float, float, float } %3169, 0, !dbg !91 + %3171 = extractvalue { float, float, float, float } %3169, 1, !dbg !91 + %3172 = extractvalue { float, float, float, float } %3169, 2, !dbg !91 + %3173 = extractvalue { float, float, float, float } %3169, 3, !dbg !91 + %3174 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3095, float %3096, float %3097, float %3098, i32 %2939, i32 %2940, i32 %2941, i32 %2942, i32 %3052, i32 %3053) #2, !dbg !91 + %3175 = extractvalue { float, float, float, float } %3174, 0, !dbg !91 + %3176 = extractvalue { float, float, float, float } %3174, 1, !dbg !91 + %3177 = extractvalue { float, float, float, float } %3174, 2, !dbg !91 + %3178 = extractvalue { float, float, float, float } %3174, 3, !dbg !91 + %3179 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3100, float %3101, float %3102, float %3103, i32 %2963, i32 %2964, i32 %2965, i32 %2966, i32 %2982, i32 %2983) #2, !dbg !91 + %3180 = extractvalue { float, float, float, float } %3179, 0, !dbg !91 + %3181 = extractvalue { float, float, float, float } %3179, 1, !dbg !91 + %3182 = extractvalue { float, float, float, float } %3179, 2, !dbg !91 + %3183 = extractvalue { float, float, float, float } %3179, 3, !dbg !91 + %3184 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3105, float %3106, float %3107, float %3108, i32 %2963, i32 %2964, i32 %2965, i32 %2966, i32 %2992, i32 %2993) #2, !dbg !91 + %3185 = extractvalue { float, float, float, float } %3184, 0, !dbg !91 + %3186 = extractvalue { float, float, float, float } %3184, 1, !dbg !91 + %3187 = extractvalue { float, float, float, float } %3184, 2, !dbg !91 + %3188 = extractvalue { float, float, float, float } %3184, 3, !dbg !91 + %3189 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3110, float %3111, float %3112, float %3113, i32 %2963, i32 %2964, i32 %2965, i32 %2966, i32 %3002, i32 %3003) #2, !dbg !91 + %3190 = extractvalue { float, float, float, float } %3189, 0, !dbg !91 + %3191 = extractvalue { float, float, float, float } %3189, 1, !dbg !91 + %3192 = extractvalue { float, float, float, float } %3189, 2, !dbg !91 + %3193 = extractvalue { float, float, float, float } %3189, 3, !dbg !91 + %3194 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3115, float %3116, float %3117, float %3118, i32 %2963, i32 %2964, i32 %2965, i32 %2966, i32 %3012, i32 %3013) #2, !dbg !91 + %3195 = extractvalue { float, float, float, float } %3194, 0, !dbg !91 + %3196 = extractvalue { float, float, float, float } %3194, 1, !dbg !91 + %3197 = extractvalue { float, float, float, float } %3194, 2, !dbg !91 + %3198 = extractvalue { float, float, float, float } %3194, 3, !dbg !91 + %3199 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3120, float %3121, float %3122, float %3123, i32 %2963, i32 %2964, i32 %2965, i32 %2966, i32 %3022, i32 %3023) #2, !dbg !91 + %3200 = extractvalue { float, float, float, float } %3199, 0, !dbg !91 + %3201 = extractvalue { float, float, float, float } %3199, 1, !dbg !91 + %3202 = extractvalue { float, float, float, float } %3199, 2, !dbg !91 + %3203 = extractvalue { float, float, float, float } %3199, 3, !dbg !91 + %3204 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3125, float %3126, float %3127, float %3128, i32 %2963, i32 %2964, i32 %2965, i32 %2966, i32 %3032, i32 %3033) #2, !dbg !91 + %3205 = extractvalue { float, float, float, float } %3204, 0, !dbg !91 + %3206 = extractvalue { float, float, float, float } %3204, 1, !dbg !91 + %3207 = extractvalue { float, float, float, float } %3204, 2, !dbg !91 + %3208 = extractvalue { float, float, float, float } %3204, 3, !dbg !91 + %3209 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3130, float %3131, float %3132, float %3133, i32 %2963, i32 %2964, i32 %2965, i32 %2966, i32 %3042, i32 %3043) #2, !dbg !91 + %3210 = extractvalue { float, float, float, float } %3209, 0, !dbg !91 + %3211 = extractvalue { float, float, float, float } %3209, 1, !dbg !91 + %3212 = extractvalue { float, float, float, float } %3209, 2, !dbg !91 + %3213 = extractvalue { float, float, float, float } %3209, 3, !dbg !91 + %3214 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3135, float %3136, float %3137, float %3138, i32 %2963, i32 %2964, i32 %2965, i32 %2966, i32 %3052, i32 %3053) #2, !dbg !91 + %3215 = extractvalue { float, float, float, float } %3214, 0, !dbg !91 + %3216 = extractvalue { float, float, float, float } %3214, 1, !dbg !91 + %3217 = extractvalue { float, float, float, float } %3214, 2, !dbg !91 + %3218 = extractvalue { float, float, float, float } %3214, 3, !dbg !91 + %3219 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3140, float %3141, float %3142, float %3143, i32 %2945, i32 %2946, i32 %2947, i32 %2948, i32 %2985, i32 %2986) #2, !dbg !91 + %3220 = extractvalue { float, float, float, float } %3219, 0, !dbg !91 + %3221 = extractvalue { float, float, float, float } %3219, 1, !dbg !91 + %3222 = extractvalue { float, float, float, float } %3219, 2, !dbg !91 + %3223 = extractvalue { float, float, float, float } %3219, 3, !dbg !91 + %3224 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3145, float %3146, float %3147, float %3148, i32 %2945, i32 %2946, i32 %2947, i32 %2948, i32 %2995, i32 %2996) #2, !dbg !91 + %3225 = extractvalue { float, float, float, float } %3224, 0, !dbg !91 + %3226 = extractvalue { float, float, float, float } %3224, 1, !dbg !91 + %3227 = extractvalue { float, float, float, float } %3224, 2, !dbg !91 + %3228 = extractvalue { float, float, float, float } %3224, 3, !dbg !91 + %3229 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3150, float %3151, float %3152, float %3153, i32 %2945, i32 %2946, i32 %2947, i32 %2948, i32 %3005, i32 %3006) #2, !dbg !91 + %3230 = extractvalue { float, float, float, float } %3229, 0, !dbg !91 + %3231 = extractvalue { float, float, float, float } %3229, 1, !dbg !91 + %3232 = extractvalue { float, float, float, float } %3229, 2, !dbg !91 + %3233 = extractvalue { float, float, float, float } %3229, 3, !dbg !91 + %3234 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3155, float %3156, float %3157, float %3158, i32 %2945, i32 %2946, i32 %2947, i32 %2948, i32 %3015, i32 %3016) #2, !dbg !91 + %3235 = extractvalue { float, float, float, float } %3234, 0, !dbg !91 + %3236 = extractvalue { float, float, float, float } %3234, 1, !dbg !91 + %3237 = extractvalue { float, float, float, float } %3234, 2, !dbg !91 + %3238 = extractvalue { float, float, float, float } %3234, 3, !dbg !91 + %3239 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3160, float %3161, float %3162, float %3163, i32 %2945, i32 %2946, i32 %2947, i32 %2948, i32 %3025, i32 %3026) #2, !dbg !91 + %3240 = extractvalue { float, float, float, float } %3239, 0, !dbg !91 + %3241 = extractvalue { float, float, float, float } %3239, 1, !dbg !91 + %3242 = extractvalue { float, float, float, float } %3239, 2, !dbg !91 + %3243 = extractvalue { float, float, float, float } %3239, 3, !dbg !91 + %3244 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3165, float %3166, float %3167, float %3168, i32 %2945, i32 %2946, i32 %2947, i32 %2948, i32 %3035, i32 %3036) #2, !dbg !91 + %3245 = extractvalue { float, float, float, float } %3244, 0, !dbg !91 + %3246 = extractvalue { float, float, float, float } %3244, 1, !dbg !91 + %3247 = extractvalue { float, float, float, float } %3244, 2, !dbg !91 + %3248 = extractvalue { float, float, float, float } %3244, 3, !dbg !91 + %3249 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3170, float %3171, float %3172, float %3173, i32 %2945, i32 %2946, i32 %2947, i32 %2948, i32 %3045, i32 %3046) #2, !dbg !91 + %3250 = extractvalue { float, float, float, float } %3249, 0, !dbg !91 + %3251 = extractvalue { float, float, float, float } %3249, 1, !dbg !91 + %3252 = extractvalue { float, float, float, float } %3249, 2, !dbg !91 + %3253 = extractvalue { float, float, float, float } %3249, 3, !dbg !91 + %3254 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3175, float %3176, float %3177, float %3178, i32 %2945, i32 %2946, i32 %2947, i32 %2948, i32 %3055, i32 %3056) #2, !dbg !91 + %3255 = extractvalue { float, float, float, float } %3254, 0, !dbg !91 + %3256 = extractvalue { float, float, float, float } %3254, 1, !dbg !91 + %3257 = extractvalue { float, float, float, float } %3254, 2, !dbg !91 + %3258 = extractvalue { float, float, float, float } %3254, 3, !dbg !91 + %3259 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3180, float %3181, float %3182, float %3183, i32 %2969, i32 %2970, i32 %2971, i32 %2972, i32 %2985, i32 %2986) #2, !dbg !91 + %3260 = extractvalue { float, float, float, float } %3259, 0, !dbg !91 + %3261 = extractvalue { float, float, float, float } %3259, 1, !dbg !91 + %3262 = extractvalue { float, float, float, float } %3259, 2, !dbg !91 + %3263 = extractvalue { float, float, float, float } %3259, 3, !dbg !91 + %3264 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3185, float %3186, float %3187, float %3188, i32 %2969, i32 %2970, i32 %2971, i32 %2972, i32 %2995, i32 %2996) #2, !dbg !91 + %3265 = extractvalue { float, float, float, float } %3264, 0, !dbg !91 + %3266 = extractvalue { float, float, float, float } %3264, 1, !dbg !91 + %3267 = extractvalue { float, float, float, float } %3264, 2, !dbg !91 + %3268 = extractvalue { float, float, float, float } %3264, 3, !dbg !91 + %3269 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3190, float %3191, float %3192, float %3193, i32 %2969, i32 %2970, i32 %2971, i32 %2972, i32 %3005, i32 %3006) #2, !dbg !91 + %3270 = extractvalue { float, float, float, float } %3269, 0, !dbg !91 + %3271 = extractvalue { float, float, float, float } %3269, 1, !dbg !91 + %3272 = extractvalue { float, float, float, float } %3269, 2, !dbg !91 + %3273 = extractvalue { float, float, float, float } %3269, 3, !dbg !91 + %3274 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3195, float %3196, float %3197, float %3198, i32 %2969, i32 %2970, i32 %2971, i32 %2972, i32 %3015, i32 %3016) #2, !dbg !91 + %3275 = extractvalue { float, float, float, float } %3274, 0, !dbg !91 + %3276 = extractvalue { float, float, float, float } %3274, 1, !dbg !91 + %3277 = extractvalue { float, float, float, float } %3274, 2, !dbg !91 + %3278 = extractvalue { float, float, float, float } %3274, 3, !dbg !91 + %3279 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3200, float %3201, float %3202, float %3203, i32 %2969, i32 %2970, i32 %2971, i32 %2972, i32 %3025, i32 %3026) #2, !dbg !91 + %3280 = extractvalue { float, float, float, float } %3279, 0, !dbg !91 + %3281 = extractvalue { float, float, float, float } %3279, 1, !dbg !91 + %3282 = extractvalue { float, float, float, float } %3279, 2, !dbg !91 + %3283 = extractvalue { float, float, float, float } %3279, 3, !dbg !91 + %3284 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3205, float %3206, float %3207, float %3208, i32 %2969, i32 %2970, i32 %2971, i32 %2972, i32 %3035, i32 %3036) #2, !dbg !91 + %3285 = extractvalue { float, float, float, float } %3284, 0, !dbg !91 + %3286 = extractvalue { float, float, float, float } %3284, 1, !dbg !91 + %3287 = extractvalue { float, float, float, float } %3284, 2, !dbg !91 + %3288 = extractvalue { float, float, float, float } %3284, 3, !dbg !91 + %3289 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3210, float %3211, float %3212, float %3213, i32 %2969, i32 %2970, i32 %2971, i32 %2972, i32 %3045, i32 %3046) #2, !dbg !91 + %3290 = extractvalue { float, float, float, float } %3289, 0, !dbg !91 + %3291 = extractvalue { float, float, float, float } %3289, 1, !dbg !91 + %3292 = extractvalue { float, float, float, float } %3289, 2, !dbg !91 + %3293 = extractvalue { float, float, float, float } %3289, 3, !dbg !91 + %3294 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3215, float %3216, float %3217, float %3218, i32 %2969, i32 %2970, i32 %2971, i32 %2972, i32 %3055, i32 %3056) #2, !dbg !91 + %3295 = extractvalue { float, float, float, float } %3294, 0, !dbg !91 + %3296 = extractvalue { float, float, float, float } %3294, 1, !dbg !91 + %3297 = extractvalue { float, float, float, float } %3294, 2, !dbg !91 + %3298 = extractvalue { float, float, float, float } %3294, 3, !dbg !91 + %3299 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3220, float %3221, float %3222, float %3223, i32 %2951, i32 %2952, i32 %2953, i32 %2954, i32 %2987, i32 %2988) #2, !dbg !91 + %3300 = extractvalue { float, float, float, float } %3299, 0, !dbg !91 + %3301 = extractvalue { float, float, float, float } %3299, 1, !dbg !91 + %3302 = extractvalue { float, float, float, float } %3299, 2, !dbg !91 + %3303 = extractvalue { float, float, float, float } %3299, 3, !dbg !91 + %3304 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3225, float %3226, float %3227, float %3228, i32 %2951, i32 %2952, i32 %2953, i32 %2954, i32 %2997, i32 %2998) #2, !dbg !91 + %3305 = extractvalue { float, float, float, float } %3304, 0, !dbg !91 + %3306 = extractvalue { float, float, float, float } %3304, 1, !dbg !91 + %3307 = extractvalue { float, float, float, float } %3304, 2, !dbg !91 + %3308 = extractvalue { float, float, float, float } %3304, 3, !dbg !91 + %3309 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3230, float %3231, float %3232, float %3233, i32 %2951, i32 %2952, i32 %2953, i32 %2954, i32 %3007, i32 %3008) #2, !dbg !91 + %3310 = extractvalue { float, float, float, float } %3309, 0, !dbg !91 + %3311 = extractvalue { float, float, float, float } %3309, 1, !dbg !91 + %3312 = extractvalue { float, float, float, float } %3309, 2, !dbg !91 + %3313 = extractvalue { float, float, float, float } %3309, 3, !dbg !91 + %3314 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3235, float %3236, float %3237, float %3238, i32 %2951, i32 %2952, i32 %2953, i32 %2954, i32 %3017, i32 %3018) #2, !dbg !91 + %3315 = extractvalue { float, float, float, float } %3314, 0, !dbg !91 + %3316 = extractvalue { float, float, float, float } %3314, 1, !dbg !91 + %3317 = extractvalue { float, float, float, float } %3314, 2, !dbg !91 + %3318 = extractvalue { float, float, float, float } %3314, 3, !dbg !91 + %3319 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3240, float %3241, float %3242, float %3243, i32 %2951, i32 %2952, i32 %2953, i32 %2954, i32 %3027, i32 %3028) #2, !dbg !91 + %3320 = extractvalue { float, float, float, float } %3319, 0, !dbg !91 + %3321 = extractvalue { float, float, float, float } %3319, 1, !dbg !91 + %3322 = extractvalue { float, float, float, float } %3319, 2, !dbg !91 + %3323 = extractvalue { float, float, float, float } %3319, 3, !dbg !91 + %3324 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3245, float %3246, float %3247, float %3248, i32 %2951, i32 %2952, i32 %2953, i32 %2954, i32 %3037, i32 %3038) #2, !dbg !91 + %3325 = extractvalue { float, float, float, float } %3324, 0, !dbg !91 + %3326 = extractvalue { float, float, float, float } %3324, 1, !dbg !91 + %3327 = extractvalue { float, float, float, float } %3324, 2, !dbg !91 + %3328 = extractvalue { float, float, float, float } %3324, 3, !dbg !91 + %3329 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3250, float %3251, float %3252, float %3253, i32 %2951, i32 %2952, i32 %2953, i32 %2954, i32 %3047, i32 %3048) #2, !dbg !91 + %3330 = extractvalue { float, float, float, float } %3329, 0, !dbg !91 + %3331 = extractvalue { float, float, float, float } %3329, 1, !dbg !91 + %3332 = extractvalue { float, float, float, float } %3329, 2, !dbg !91 + %3333 = extractvalue { float, float, float, float } %3329, 3, !dbg !91 + %3334 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3255, float %3256, float %3257, float %3258, i32 %2951, i32 %2952, i32 %2953, i32 %2954, i32 %3057, i32 %3058) #2, !dbg !91 + %3335 = extractvalue { float, float, float, float } %3334, 0, !dbg !91 + %3336 = extractvalue { float, float, float, float } %3334, 1, !dbg !91 + %3337 = extractvalue { float, float, float, float } %3334, 2, !dbg !91 + %3338 = extractvalue { float, float, float, float } %3334, 3, !dbg !91 + %3339 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3260, float %3261, float %3262, float %3263, i32 %2975, i32 %2976, i32 %2977, i32 %2978, i32 %2987, i32 %2988) #2, !dbg !91 + %3340 = extractvalue { float, float, float, float } %3339, 0, !dbg !91 + %3341 = extractvalue { float, float, float, float } %3339, 1, !dbg !91 + %3342 = extractvalue { float, float, float, float } %3339, 2, !dbg !91 + %3343 = extractvalue { float, float, float, float } %3339, 3, !dbg !91 + %3344 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3265, float %3266, float %3267, float %3268, i32 %2975, i32 %2976, i32 %2977, i32 %2978, i32 %2997, i32 %2998) #2, !dbg !91 + %3345 = extractvalue { float, float, float, float } %3344, 0, !dbg !91 + %3346 = extractvalue { float, float, float, float } %3344, 1, !dbg !91 + %3347 = extractvalue { float, float, float, float } %3344, 2, !dbg !91 + %3348 = extractvalue { float, float, float, float } %3344, 3, !dbg !91 + %3349 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3270, float %3271, float %3272, float %3273, i32 %2975, i32 %2976, i32 %2977, i32 %2978, i32 %3007, i32 %3008) #2, !dbg !91 + %3350 = extractvalue { float, float, float, float } %3349, 0, !dbg !91 + %3351 = extractvalue { float, float, float, float } %3349, 1, !dbg !91 + %3352 = extractvalue { float, float, float, float } %3349, 2, !dbg !91 + %3353 = extractvalue { float, float, float, float } %3349, 3, !dbg !91 + %3354 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3275, float %3276, float %3277, float %3278, i32 %2975, i32 %2976, i32 %2977, i32 %2978, i32 %3017, i32 %3018) #2, !dbg !91 + %3355 = extractvalue { float, float, float, float } %3354, 0, !dbg !91 + %3356 = extractvalue { float, float, float, float } %3354, 1, !dbg !91 + %3357 = extractvalue { float, float, float, float } %3354, 2, !dbg !91 + %3358 = extractvalue { float, float, float, float } %3354, 3, !dbg !91 + %3359 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3280, float %3281, float %3282, float %3283, i32 %2975, i32 %2976, i32 %2977, i32 %2978, i32 %3027, i32 %3028) #2, !dbg !91 + %3360 = extractvalue { float, float, float, float } %3359, 0, !dbg !91 + %3361 = extractvalue { float, float, float, float } %3359, 1, !dbg !91 + %3362 = extractvalue { float, float, float, float } %3359, 2, !dbg !91 + %3363 = extractvalue { float, float, float, float } %3359, 3, !dbg !91 + %3364 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3285, float %3286, float %3287, float %3288, i32 %2975, i32 %2976, i32 %2977, i32 %2978, i32 %3037, i32 %3038) #2, !dbg !91 + %3365 = extractvalue { float, float, float, float } %3364, 0, !dbg !91 + %3366 = extractvalue { float, float, float, float } %3364, 1, !dbg !91 + %3367 = extractvalue { float, float, float, float } %3364, 2, !dbg !91 + %3368 = extractvalue { float, float, float, float } %3364, 3, !dbg !91 + %3369 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3290, float %3291, float %3292, float %3293, i32 %2975, i32 %2976, i32 %2977, i32 %2978, i32 %3047, i32 %3048) #2, !dbg !91 + %3370 = extractvalue { float, float, float, float } %3369, 0, !dbg !91 + %3371 = extractvalue { float, float, float, float } %3369, 1, !dbg !91 + %3372 = extractvalue { float, float, float, float } %3369, 2, !dbg !91 + %3373 = extractvalue { float, float, float, float } %3369, 3, !dbg !91 + %3374 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3295, float %3296, float %3297, float %3298, i32 %2975, i32 %2976, i32 %2977, i32 %2978, i32 %3057, i32 %3058) #2, !dbg !91 + %3375 = extractvalue { float, float, float, float } %3374, 0, !dbg !91 + %3376 = extractvalue { float, float, float, float } %3374, 1, !dbg !91 + %3377 = extractvalue { float, float, float, float } %3374, 2, !dbg !91 + %3378 = extractvalue { float, float, float, float } %3374, 3, !dbg !91 + %3379 = getelementptr bfloat, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 98304), i32 %1256, !dbg !59 + %3380 = getelementptr inbounds nuw i8, ptr addrspace(3) %3379, i32 %674, !dbg !59 + %3381 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %3380), !dbg !59 + %3382 = extractvalue { i32, i32, i32, i32 } %3381, 0, !dbg !59 + %3383 = extractvalue { i32, i32, i32, i32 } %3381, 1, !dbg !59 + %3384 = extractvalue { i32, i32, i32, i32 } %3381, 2, !dbg !59 + %3385 = extractvalue { i32, i32, i32, i32 } %3381, 3, !dbg !59 + %3386 = getelementptr inbounds nuw i8, ptr addrspace(3) %3379, i32 %675, !dbg !59 + %3387 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %3386), !dbg !59 + %3388 = extractvalue { i32, i32, i32, i32 } %3387, 0, !dbg !59 + %3389 = extractvalue { i32, i32, i32, i32 } %3387, 1, !dbg !59 + %3390 = extractvalue { i32, i32, i32, i32 } %3387, 2, !dbg !59 + %3391 = extractvalue { i32, i32, i32, i32 } %3387, 3, !dbg !59 + %3392 = getelementptr inbounds nuw i8, ptr addrspace(3) %3379, i32 %676, !dbg !59 + %3393 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %3392), !dbg !59 + %3394 = extractvalue { i32, i32, i32, i32 } %3393, 0, !dbg !59 + %3395 = extractvalue { i32, i32, i32, i32 } %3393, 1, !dbg !59 + %3396 = extractvalue { i32, i32, i32, i32 } %3393, 2, !dbg !59 + %3397 = extractvalue { i32, i32, i32, i32 } %3393, 3, !dbg !59 + %3398 = getelementptr inbounds nuw i8, ptr addrspace(3) %3379, i32 %677, !dbg !59 + %3399 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %3398), !dbg !59 + %3400 = extractvalue { i32, i32, i32, i32 } %3399, 0, !dbg !59 + %3401 = extractvalue { i32, i32, i32, i32 } %3399, 1, !dbg !59 + %3402 = extractvalue { i32, i32, i32, i32 } %3399, 2, !dbg !59 + %3403 = extractvalue { i32, i32, i32, i32 } %3399, 3, !dbg !59 + %3404 = getelementptr inbounds nuw i8, ptr addrspace(3) %3379, i32 %678, !dbg !59 + %3405 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) %3404), !dbg !59 + %3406 = extractvalue { i32, i32, i32, i32 } %3405, 0, !dbg !59 + %3407 = extractvalue { i32, i32, i32, i32 } %3405, 1, !dbg !59 + %3408 = extractvalue { i32, i32, i32, i32 } %3405, 2, !dbg !59 + %3409 = extractvalue { i32, i32, i32, i32 } %3405, 3, !dbg !59 + %3410 = getelementptr inbounds nuw i8, ptr addrspace(3) %3379, i32 %679, !dbg !59 + %3411 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %3410), !dbg !59 + %3412 = extractvalue { i32, i32, i32, i32 } %3411, 0, !dbg !59 + %3413 = extractvalue { i32, i32, i32, i32 } %3411, 1, !dbg !59 + %3414 = extractvalue { i32, i32, i32, i32 } %3411, 2, !dbg !59 + %3415 = extractvalue { i32, i32, i32, i32 } %3411, 3, !dbg !59 + %3416 = getelementptr inbounds nuw i8, ptr addrspace(3) %3379, i32 %680, !dbg !59 + %3417 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %3416), !dbg !59 + %3418 = extractvalue { i32, i32, i32, i32 } %3417, 0, !dbg !59 + %3419 = extractvalue { i32, i32, i32, i32 } %3417, 1, !dbg !59 + %3420 = extractvalue { i32, i32, i32, i32 } %3417, 2, !dbg !59 + %3421 = extractvalue { i32, i32, i32, i32 } %3417, 3, !dbg !59 + %3422 = getelementptr inbounds nuw i8, ptr addrspace(3) %3379, i32 %681, !dbg !59 + %3423 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) nonnull %3422), !dbg !59 + %3424 = extractvalue { i32, i32, i32, i32 } %3423, 0, !dbg !59 + %3425 = extractvalue { i32, i32, i32, i32 } %3423, 1, !dbg !59 + %3426 = extractvalue { i32, i32, i32, i32 } %3423, 2, !dbg !59 + %3427 = extractvalue { i32, i32, i32, i32 } %3423, 3, !dbg !59 + %3428 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2706, float %2707, float %2708, float %2709, i32 %3382, i32 %3383, i32 %3384, i32 %3385, i32 %2980, i32 %2981) #2, !dbg !92 + %3429 = extractvalue { float, float, float, float } %3428, 0, !dbg !92 + %3430 = extractvalue { float, float, float, float } %3428, 1, !dbg !92 + %3431 = extractvalue { float, float, float, float } %3428, 2, !dbg !92 + %3432 = extractvalue { float, float, float, float } %3428, 3, !dbg !92 + %3433 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2710, float %2711, float %2712, float %2713, i32 %3382, i32 %3383, i32 %3384, i32 %3385, i32 %2990, i32 %2991) #2, !dbg !92 + %3434 = extractvalue { float, float, float, float } %3433, 0, !dbg !92 + %3435 = extractvalue { float, float, float, float } %3433, 1, !dbg !92 + %3436 = extractvalue { float, float, float, float } %3433, 2, !dbg !92 + %3437 = extractvalue { float, float, float, float } %3433, 3, !dbg !92 + %3438 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2714, float %2715, float %2716, float %2717, i32 %3382, i32 %3383, i32 %3384, i32 %3385, i32 %3000, i32 %3001) #2, !dbg !92 + %3439 = extractvalue { float, float, float, float } %3438, 0, !dbg !92 + %3440 = extractvalue { float, float, float, float } %3438, 1, !dbg !92 + %3441 = extractvalue { float, float, float, float } %3438, 2, !dbg !92 + %3442 = extractvalue { float, float, float, float } %3438, 3, !dbg !92 + %3443 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2718, float %2719, float %2720, float %2721, i32 %3382, i32 %3383, i32 %3384, i32 %3385, i32 %3010, i32 %3011) #2, !dbg !92 + %3444 = extractvalue { float, float, float, float } %3443, 0, !dbg !92 + %3445 = extractvalue { float, float, float, float } %3443, 1, !dbg !92 + %3446 = extractvalue { float, float, float, float } %3443, 2, !dbg !92 + %3447 = extractvalue { float, float, float, float } %3443, 3, !dbg !92 + %3448 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2722, float %2723, float %2724, float %2725, i32 %3382, i32 %3383, i32 %3384, i32 %3385, i32 %3020, i32 %3021) #2, !dbg !92 + %3449 = extractvalue { float, float, float, float } %3448, 0, !dbg !92 + %3450 = extractvalue { float, float, float, float } %3448, 1, !dbg !92 + %3451 = extractvalue { float, float, float, float } %3448, 2, !dbg !92 + %3452 = extractvalue { float, float, float, float } %3448, 3, !dbg !92 + %3453 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2726, float %2727, float %2728, float %2729, i32 %3382, i32 %3383, i32 %3384, i32 %3385, i32 %3030, i32 %3031) #2, !dbg !92 + %3454 = extractvalue { float, float, float, float } %3453, 0, !dbg !92 + %3455 = extractvalue { float, float, float, float } %3453, 1, !dbg !92 + %3456 = extractvalue { float, float, float, float } %3453, 2, !dbg !92 + %3457 = extractvalue { float, float, float, float } %3453, 3, !dbg !92 + %3458 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2730, float %2731, float %2732, float %2733, i32 %3382, i32 %3383, i32 %3384, i32 %3385, i32 %3040, i32 %3041) #2, !dbg !92 + %3459 = extractvalue { float, float, float, float } %3458, 0, !dbg !92 + %3460 = extractvalue { float, float, float, float } %3458, 1, !dbg !92 + %3461 = extractvalue { float, float, float, float } %3458, 2, !dbg !92 + %3462 = extractvalue { float, float, float, float } %3458, 3, !dbg !92 + %3463 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2734, float %2735, float %2736, float %2737, i32 %3382, i32 %3383, i32 %3384, i32 %3385, i32 %3050, i32 %3051) #2, !dbg !92 + %3464 = extractvalue { float, float, float, float } %3463, 0, !dbg !92 + %3465 = extractvalue { float, float, float, float } %3463, 1, !dbg !92 + %3466 = extractvalue { float, float, float, float } %3463, 2, !dbg !92 + %3467 = extractvalue { float, float, float, float } %3463, 3, !dbg !92 + %3468 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2738, float %2739, float %2740, float %2741, i32 %3406, i32 %3407, i32 %3408, i32 %3409, i32 %2980, i32 %2981) #2, !dbg !92 + %3469 = extractvalue { float, float, float, float } %3468, 0, !dbg !92 + %3470 = extractvalue { float, float, float, float } %3468, 1, !dbg !92 + %3471 = extractvalue { float, float, float, float } %3468, 2, !dbg !92 + %3472 = extractvalue { float, float, float, float } %3468, 3, !dbg !92 + %3473 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2742, float %2743, float %2744, float %2745, i32 %3406, i32 %3407, i32 %3408, i32 %3409, i32 %2990, i32 %2991) #2, !dbg !92 + %3474 = extractvalue { float, float, float, float } %3473, 0, !dbg !92 + %3475 = extractvalue { float, float, float, float } %3473, 1, !dbg !92 + %3476 = extractvalue { float, float, float, float } %3473, 2, !dbg !92 + %3477 = extractvalue { float, float, float, float } %3473, 3, !dbg !92 + %3478 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2746, float %2747, float %2748, float %2749, i32 %3406, i32 %3407, i32 %3408, i32 %3409, i32 %3000, i32 %3001) #2, !dbg !92 + %3479 = extractvalue { float, float, float, float } %3478, 0, !dbg !92 + %3480 = extractvalue { float, float, float, float } %3478, 1, !dbg !92 + %3481 = extractvalue { float, float, float, float } %3478, 2, !dbg !92 + %3482 = extractvalue { float, float, float, float } %3478, 3, !dbg !92 + %3483 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2750, float %2751, float %2752, float %2753, i32 %3406, i32 %3407, i32 %3408, i32 %3409, i32 %3010, i32 %3011) #2, !dbg !92 + %3484 = extractvalue { float, float, float, float } %3483, 0, !dbg !92 + %3485 = extractvalue { float, float, float, float } %3483, 1, !dbg !92 + %3486 = extractvalue { float, float, float, float } %3483, 2, !dbg !92 + %3487 = extractvalue { float, float, float, float } %3483, 3, !dbg !92 + %3488 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2754, float %2755, float %2756, float %2757, i32 %3406, i32 %3407, i32 %3408, i32 %3409, i32 %3020, i32 %3021) #2, !dbg !92 + %3489 = extractvalue { float, float, float, float } %3488, 0, !dbg !92 + %3490 = extractvalue { float, float, float, float } %3488, 1, !dbg !92 + %3491 = extractvalue { float, float, float, float } %3488, 2, !dbg !92 + %3492 = extractvalue { float, float, float, float } %3488, 3, !dbg !92 + %3493 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2758, float %2759, float %2760, float %2761, i32 %3406, i32 %3407, i32 %3408, i32 %3409, i32 %3030, i32 %3031) #2, !dbg !92 + %3494 = extractvalue { float, float, float, float } %3493, 0, !dbg !92 + %3495 = extractvalue { float, float, float, float } %3493, 1, !dbg !92 + %3496 = extractvalue { float, float, float, float } %3493, 2, !dbg !92 + %3497 = extractvalue { float, float, float, float } %3493, 3, !dbg !92 + %3498 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2762, float %2763, float %2764, float %2765, i32 %3406, i32 %3407, i32 %3408, i32 %3409, i32 %3040, i32 %3041) #2, !dbg !92 + %3499 = extractvalue { float, float, float, float } %3498, 0, !dbg !92 + %3500 = extractvalue { float, float, float, float } %3498, 1, !dbg !92 + %3501 = extractvalue { float, float, float, float } %3498, 2, !dbg !92 + %3502 = extractvalue { float, float, float, float } %3498, 3, !dbg !92 + %3503 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %2766, float %2767, float %2768, float %2769, i32 %3406, i32 %3407, i32 %3408, i32 %3409, i32 %3050, i32 %3051) #2, !dbg !92 + %3504 = extractvalue { float, float, float, float } %3503, 0, !dbg !92 + %3505 = extractvalue { float, float, float, float } %3503, 1, !dbg !92 + %3506 = extractvalue { float, float, float, float } %3503, 2, !dbg !92 + %3507 = extractvalue { float, float, float, float } %3503, 3, !dbg !92 + %3508 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3429, float %3430, float %3431, float %3432, i32 %3388, i32 %3389, i32 %3390, i32 %3391, i32 %2982, i32 %2983) #2, !dbg !92 + %3509 = extractvalue { float, float, float, float } %3508, 0, !dbg !92 + %3510 = extractvalue { float, float, float, float } %3508, 1, !dbg !92 + %3511 = extractvalue { float, float, float, float } %3508, 2, !dbg !92 + %3512 = extractvalue { float, float, float, float } %3508, 3, !dbg !92 + %3513 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3434, float %3435, float %3436, float %3437, i32 %3388, i32 %3389, i32 %3390, i32 %3391, i32 %2992, i32 %2993) #2, !dbg !92 + %3514 = extractvalue { float, float, float, float } %3513, 0, !dbg !92 + %3515 = extractvalue { float, float, float, float } %3513, 1, !dbg !92 + %3516 = extractvalue { float, float, float, float } %3513, 2, !dbg !92 + %3517 = extractvalue { float, float, float, float } %3513, 3, !dbg !92 + %3518 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3439, float %3440, float %3441, float %3442, i32 %3388, i32 %3389, i32 %3390, i32 %3391, i32 %3002, i32 %3003) #2, !dbg !92 + %3519 = extractvalue { float, float, float, float } %3518, 0, !dbg !92 + %3520 = extractvalue { float, float, float, float } %3518, 1, !dbg !92 + %3521 = extractvalue { float, float, float, float } %3518, 2, !dbg !92 + %3522 = extractvalue { float, float, float, float } %3518, 3, !dbg !92 + %3523 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3444, float %3445, float %3446, float %3447, i32 %3388, i32 %3389, i32 %3390, i32 %3391, i32 %3012, i32 %3013) #2, !dbg !92 + %3524 = extractvalue { float, float, float, float } %3523, 0, !dbg !92 + %3525 = extractvalue { float, float, float, float } %3523, 1, !dbg !92 + %3526 = extractvalue { float, float, float, float } %3523, 2, !dbg !92 + %3527 = extractvalue { float, float, float, float } %3523, 3, !dbg !92 + %3528 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3449, float %3450, float %3451, float %3452, i32 %3388, i32 %3389, i32 %3390, i32 %3391, i32 %3022, i32 %3023) #2, !dbg !92 + %3529 = extractvalue { float, float, float, float } %3528, 0, !dbg !92 + %3530 = extractvalue { float, float, float, float } %3528, 1, !dbg !92 + %3531 = extractvalue { float, float, float, float } %3528, 2, !dbg !92 + %3532 = extractvalue { float, float, float, float } %3528, 3, !dbg !92 + %3533 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3454, float %3455, float %3456, float %3457, i32 %3388, i32 %3389, i32 %3390, i32 %3391, i32 %3032, i32 %3033) #2, !dbg !92 + %3534 = extractvalue { float, float, float, float } %3533, 0, !dbg !92 + %3535 = extractvalue { float, float, float, float } %3533, 1, !dbg !92 + %3536 = extractvalue { float, float, float, float } %3533, 2, !dbg !92 + %3537 = extractvalue { float, float, float, float } %3533, 3, !dbg !92 + %3538 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3459, float %3460, float %3461, float %3462, i32 %3388, i32 %3389, i32 %3390, i32 %3391, i32 %3042, i32 %3043) #2, !dbg !92 + %3539 = extractvalue { float, float, float, float } %3538, 0, !dbg !92 + %3540 = extractvalue { float, float, float, float } %3538, 1, !dbg !92 + %3541 = extractvalue { float, float, float, float } %3538, 2, !dbg !92 + %3542 = extractvalue { float, float, float, float } %3538, 3, !dbg !92 + %3543 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3464, float %3465, float %3466, float %3467, i32 %3388, i32 %3389, i32 %3390, i32 %3391, i32 %3052, i32 %3053) #2, !dbg !92 + %3544 = extractvalue { float, float, float, float } %3543, 0, !dbg !92 + %3545 = extractvalue { float, float, float, float } %3543, 1, !dbg !92 + %3546 = extractvalue { float, float, float, float } %3543, 2, !dbg !92 + %3547 = extractvalue { float, float, float, float } %3543, 3, !dbg !92 + %3548 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3469, float %3470, float %3471, float %3472, i32 %3412, i32 %3413, i32 %3414, i32 %3415, i32 %2982, i32 %2983) #2, !dbg !92 + %3549 = extractvalue { float, float, float, float } %3548, 0, !dbg !92 + %3550 = extractvalue { float, float, float, float } %3548, 1, !dbg !92 + %3551 = extractvalue { float, float, float, float } %3548, 2, !dbg !92 + %3552 = extractvalue { float, float, float, float } %3548, 3, !dbg !92 + %3553 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3474, float %3475, float %3476, float %3477, i32 %3412, i32 %3413, i32 %3414, i32 %3415, i32 %2992, i32 %2993) #2, !dbg !92 + %3554 = extractvalue { float, float, float, float } %3553, 0, !dbg !92 + %3555 = extractvalue { float, float, float, float } %3553, 1, !dbg !92 + %3556 = extractvalue { float, float, float, float } %3553, 2, !dbg !92 + %3557 = extractvalue { float, float, float, float } %3553, 3, !dbg !92 + %3558 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3479, float %3480, float %3481, float %3482, i32 %3412, i32 %3413, i32 %3414, i32 %3415, i32 %3002, i32 %3003) #2, !dbg !92 + %3559 = extractvalue { float, float, float, float } %3558, 0, !dbg !92 + %3560 = extractvalue { float, float, float, float } %3558, 1, !dbg !92 + %3561 = extractvalue { float, float, float, float } %3558, 2, !dbg !92 + %3562 = extractvalue { float, float, float, float } %3558, 3, !dbg !92 + %3563 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3484, float %3485, float %3486, float %3487, i32 %3412, i32 %3413, i32 %3414, i32 %3415, i32 %3012, i32 %3013) #2, !dbg !92 + %3564 = extractvalue { float, float, float, float } %3563, 0, !dbg !92 + %3565 = extractvalue { float, float, float, float } %3563, 1, !dbg !92 + %3566 = extractvalue { float, float, float, float } %3563, 2, !dbg !92 + %3567 = extractvalue { float, float, float, float } %3563, 3, !dbg !92 + %3568 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3489, float %3490, float %3491, float %3492, i32 %3412, i32 %3413, i32 %3414, i32 %3415, i32 %3022, i32 %3023) #2, !dbg !92 + %3569 = extractvalue { float, float, float, float } %3568, 0, !dbg !92 + %3570 = extractvalue { float, float, float, float } %3568, 1, !dbg !92 + %3571 = extractvalue { float, float, float, float } %3568, 2, !dbg !92 + %3572 = extractvalue { float, float, float, float } %3568, 3, !dbg !92 + %3573 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3494, float %3495, float %3496, float %3497, i32 %3412, i32 %3413, i32 %3414, i32 %3415, i32 %3032, i32 %3033) #2, !dbg !92 + %3574 = extractvalue { float, float, float, float } %3573, 0, !dbg !92 + %3575 = extractvalue { float, float, float, float } %3573, 1, !dbg !92 + %3576 = extractvalue { float, float, float, float } %3573, 2, !dbg !92 + %3577 = extractvalue { float, float, float, float } %3573, 3, !dbg !92 + %3578 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3499, float %3500, float %3501, float %3502, i32 %3412, i32 %3413, i32 %3414, i32 %3415, i32 %3042, i32 %3043) #2, !dbg !92 + %3579 = extractvalue { float, float, float, float } %3578, 0, !dbg !92 + %3580 = extractvalue { float, float, float, float } %3578, 1, !dbg !92 + %3581 = extractvalue { float, float, float, float } %3578, 2, !dbg !92 + %3582 = extractvalue { float, float, float, float } %3578, 3, !dbg !92 + %3583 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3504, float %3505, float %3506, float %3507, i32 %3412, i32 %3413, i32 %3414, i32 %3415, i32 %3052, i32 %3053) #2, !dbg !92 + %3584 = extractvalue { float, float, float, float } %3583, 0, !dbg !92 + %3585 = extractvalue { float, float, float, float } %3583, 1, !dbg !92 + %3586 = extractvalue { float, float, float, float } %3583, 2, !dbg !92 + %3587 = extractvalue { float, float, float, float } %3583, 3, !dbg !92 + %3588 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3509, float %3510, float %3511, float %3512, i32 %3394, i32 %3395, i32 %3396, i32 %3397, i32 %2985, i32 %2986) #2, !dbg !92 + %3589 = extractvalue { float, float, float, float } %3588, 0, !dbg !92 + %3590 = extractvalue { float, float, float, float } %3588, 1, !dbg !92 + %3591 = extractvalue { float, float, float, float } %3588, 2, !dbg !92 + %3592 = extractvalue { float, float, float, float } %3588, 3, !dbg !92 + %3593 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3514, float %3515, float %3516, float %3517, i32 %3394, i32 %3395, i32 %3396, i32 %3397, i32 %2995, i32 %2996) #2, !dbg !92 + %3594 = extractvalue { float, float, float, float } %3593, 0, !dbg !92 + %3595 = extractvalue { float, float, float, float } %3593, 1, !dbg !92 + %3596 = extractvalue { float, float, float, float } %3593, 2, !dbg !92 + %3597 = extractvalue { float, float, float, float } %3593, 3, !dbg !92 + %3598 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3519, float %3520, float %3521, float %3522, i32 %3394, i32 %3395, i32 %3396, i32 %3397, i32 %3005, i32 %3006) #2, !dbg !92 + %3599 = extractvalue { float, float, float, float } %3598, 0, !dbg !92 + %3600 = extractvalue { float, float, float, float } %3598, 1, !dbg !92 + %3601 = extractvalue { float, float, float, float } %3598, 2, !dbg !92 + %3602 = extractvalue { float, float, float, float } %3598, 3, !dbg !92 + %3603 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3524, float %3525, float %3526, float %3527, i32 %3394, i32 %3395, i32 %3396, i32 %3397, i32 %3015, i32 %3016) #2, !dbg !92 + %3604 = extractvalue { float, float, float, float } %3603, 0, !dbg !92 + %3605 = extractvalue { float, float, float, float } %3603, 1, !dbg !92 + %3606 = extractvalue { float, float, float, float } %3603, 2, !dbg !92 + %3607 = extractvalue { float, float, float, float } %3603, 3, !dbg !92 + %3608 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3529, float %3530, float %3531, float %3532, i32 %3394, i32 %3395, i32 %3396, i32 %3397, i32 %3025, i32 %3026) #2, !dbg !92 + %3609 = extractvalue { float, float, float, float } %3608, 0, !dbg !92 + %3610 = extractvalue { float, float, float, float } %3608, 1, !dbg !92 + %3611 = extractvalue { float, float, float, float } %3608, 2, !dbg !92 + %3612 = extractvalue { float, float, float, float } %3608, 3, !dbg !92 + %3613 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3534, float %3535, float %3536, float %3537, i32 %3394, i32 %3395, i32 %3396, i32 %3397, i32 %3035, i32 %3036) #2, !dbg !92 + %3614 = extractvalue { float, float, float, float } %3613, 0, !dbg !92 + %3615 = extractvalue { float, float, float, float } %3613, 1, !dbg !92 + %3616 = extractvalue { float, float, float, float } %3613, 2, !dbg !92 + %3617 = extractvalue { float, float, float, float } %3613, 3, !dbg !92 + %3618 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3539, float %3540, float %3541, float %3542, i32 %3394, i32 %3395, i32 %3396, i32 %3397, i32 %3045, i32 %3046) #2, !dbg !92 + %3619 = extractvalue { float, float, float, float } %3618, 0, !dbg !92 + %3620 = extractvalue { float, float, float, float } %3618, 1, !dbg !92 + %3621 = extractvalue { float, float, float, float } %3618, 2, !dbg !92 + %3622 = extractvalue { float, float, float, float } %3618, 3, !dbg !92 + %3623 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3544, float %3545, float %3546, float %3547, i32 %3394, i32 %3395, i32 %3396, i32 %3397, i32 %3055, i32 %3056) #2, !dbg !92 + %3624 = extractvalue { float, float, float, float } %3623, 0, !dbg !92 + %3625 = extractvalue { float, float, float, float } %3623, 1, !dbg !92 + %3626 = extractvalue { float, float, float, float } %3623, 2, !dbg !92 + %3627 = extractvalue { float, float, float, float } %3623, 3, !dbg !92 + %3628 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3549, float %3550, float %3551, float %3552, i32 %3418, i32 %3419, i32 %3420, i32 %3421, i32 %2985, i32 %2986) #2, !dbg !92 + %3629 = extractvalue { float, float, float, float } %3628, 0, !dbg !92 + %3630 = extractvalue { float, float, float, float } %3628, 1, !dbg !92 + %3631 = extractvalue { float, float, float, float } %3628, 2, !dbg !92 + %3632 = extractvalue { float, float, float, float } %3628, 3, !dbg !92 + %3633 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3554, float %3555, float %3556, float %3557, i32 %3418, i32 %3419, i32 %3420, i32 %3421, i32 %2995, i32 %2996) #2, !dbg !92 + %3634 = extractvalue { float, float, float, float } %3633, 0, !dbg !92 + %3635 = extractvalue { float, float, float, float } %3633, 1, !dbg !92 + %3636 = extractvalue { float, float, float, float } %3633, 2, !dbg !92 + %3637 = extractvalue { float, float, float, float } %3633, 3, !dbg !92 + %3638 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3559, float %3560, float %3561, float %3562, i32 %3418, i32 %3419, i32 %3420, i32 %3421, i32 %3005, i32 %3006) #2, !dbg !92 + %3639 = extractvalue { float, float, float, float } %3638, 0, !dbg !92 + %3640 = extractvalue { float, float, float, float } %3638, 1, !dbg !92 + %3641 = extractvalue { float, float, float, float } %3638, 2, !dbg !92 + %3642 = extractvalue { float, float, float, float } %3638, 3, !dbg !92 + %3643 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3564, float %3565, float %3566, float %3567, i32 %3418, i32 %3419, i32 %3420, i32 %3421, i32 %3015, i32 %3016) #2, !dbg !92 + %3644 = extractvalue { float, float, float, float } %3643, 0, !dbg !92 + %3645 = extractvalue { float, float, float, float } %3643, 1, !dbg !92 + %3646 = extractvalue { float, float, float, float } %3643, 2, !dbg !92 + %3647 = extractvalue { float, float, float, float } %3643, 3, !dbg !92 + %3648 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3569, float %3570, float %3571, float %3572, i32 %3418, i32 %3419, i32 %3420, i32 %3421, i32 %3025, i32 %3026) #2, !dbg !92 + %3649 = extractvalue { float, float, float, float } %3648, 0, !dbg !92 + %3650 = extractvalue { float, float, float, float } %3648, 1, !dbg !92 + %3651 = extractvalue { float, float, float, float } %3648, 2, !dbg !92 + %3652 = extractvalue { float, float, float, float } %3648, 3, !dbg !92 + %3653 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3574, float %3575, float %3576, float %3577, i32 %3418, i32 %3419, i32 %3420, i32 %3421, i32 %3035, i32 %3036) #2, !dbg !92 + %3654 = extractvalue { float, float, float, float } %3653, 0, !dbg !92 + %3655 = extractvalue { float, float, float, float } %3653, 1, !dbg !92 + %3656 = extractvalue { float, float, float, float } %3653, 2, !dbg !92 + %3657 = extractvalue { float, float, float, float } %3653, 3, !dbg !92 + %3658 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3579, float %3580, float %3581, float %3582, i32 %3418, i32 %3419, i32 %3420, i32 %3421, i32 %3045, i32 %3046) #2, !dbg !92 + %3659 = extractvalue { float, float, float, float } %3658, 0, !dbg !92 + %3660 = extractvalue { float, float, float, float } %3658, 1, !dbg !92 + %3661 = extractvalue { float, float, float, float } %3658, 2, !dbg !92 + %3662 = extractvalue { float, float, float, float } %3658, 3, !dbg !92 + %3663 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3584, float %3585, float %3586, float %3587, i32 %3418, i32 %3419, i32 %3420, i32 %3421, i32 %3055, i32 %3056) #2, !dbg !92 + %3664 = extractvalue { float, float, float, float } %3663, 0, !dbg !92 + %3665 = extractvalue { float, float, float, float } %3663, 1, !dbg !92 + %3666 = extractvalue { float, float, float, float } %3663, 2, !dbg !92 + %3667 = extractvalue { float, float, float, float } %3663, 3, !dbg !92 + %3668 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3589, float %3590, float %3591, float %3592, i32 %3400, i32 %3401, i32 %3402, i32 %3403, i32 %2987, i32 %2988) #2, !dbg !92 + %3669 = extractvalue { float, float, float, float } %3668, 0, !dbg !92 + %3670 = extractvalue { float, float, float, float } %3668, 1, !dbg !92 + %3671 = extractvalue { float, float, float, float } %3668, 2, !dbg !92 + %3672 = extractvalue { float, float, float, float } %3668, 3, !dbg !92 + %3673 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3594, float %3595, float %3596, float %3597, i32 %3400, i32 %3401, i32 %3402, i32 %3403, i32 %2997, i32 %2998) #2, !dbg !92 + %3674 = extractvalue { float, float, float, float } %3673, 0, !dbg !92 + %3675 = extractvalue { float, float, float, float } %3673, 1, !dbg !92 + %3676 = extractvalue { float, float, float, float } %3673, 2, !dbg !92 + %3677 = extractvalue { float, float, float, float } %3673, 3, !dbg !92 + %3678 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3599, float %3600, float %3601, float %3602, i32 %3400, i32 %3401, i32 %3402, i32 %3403, i32 %3007, i32 %3008) #2, !dbg !92 + %3679 = extractvalue { float, float, float, float } %3678, 0, !dbg !92 + %3680 = extractvalue { float, float, float, float } %3678, 1, !dbg !92 + %3681 = extractvalue { float, float, float, float } %3678, 2, !dbg !92 + %3682 = extractvalue { float, float, float, float } %3678, 3, !dbg !92 + %3683 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3604, float %3605, float %3606, float %3607, i32 %3400, i32 %3401, i32 %3402, i32 %3403, i32 %3017, i32 %3018) #2, !dbg !92 + %3684 = extractvalue { float, float, float, float } %3683, 0, !dbg !92 + %3685 = extractvalue { float, float, float, float } %3683, 1, !dbg !92 + %3686 = extractvalue { float, float, float, float } %3683, 2, !dbg !92 + %3687 = extractvalue { float, float, float, float } %3683, 3, !dbg !92 + %3688 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3609, float %3610, float %3611, float %3612, i32 %3400, i32 %3401, i32 %3402, i32 %3403, i32 %3027, i32 %3028) #2, !dbg !92 + %3689 = extractvalue { float, float, float, float } %3688, 0, !dbg !92 + %3690 = extractvalue { float, float, float, float } %3688, 1, !dbg !92 + %3691 = extractvalue { float, float, float, float } %3688, 2, !dbg !92 + %3692 = extractvalue { float, float, float, float } %3688, 3, !dbg !92 + %3693 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3614, float %3615, float %3616, float %3617, i32 %3400, i32 %3401, i32 %3402, i32 %3403, i32 %3037, i32 %3038) #2, !dbg !92 + %3694 = extractvalue { float, float, float, float } %3693, 0, !dbg !92 + %3695 = extractvalue { float, float, float, float } %3693, 1, !dbg !92 + %3696 = extractvalue { float, float, float, float } %3693, 2, !dbg !92 + %3697 = extractvalue { float, float, float, float } %3693, 3, !dbg !92 + %3698 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3619, float %3620, float %3621, float %3622, i32 %3400, i32 %3401, i32 %3402, i32 %3403, i32 %3047, i32 %3048) #2, !dbg !92 + %3699 = extractvalue { float, float, float, float } %3698, 0, !dbg !92 + %3700 = extractvalue { float, float, float, float } %3698, 1, !dbg !92 + %3701 = extractvalue { float, float, float, float } %3698, 2, !dbg !92 + %3702 = extractvalue { float, float, float, float } %3698, 3, !dbg !92 + %3703 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3624, float %3625, float %3626, float %3627, i32 %3400, i32 %3401, i32 %3402, i32 %3403, i32 %3057, i32 %3058) #2, !dbg !92 + %3704 = extractvalue { float, float, float, float } %3703, 0, !dbg !92 + %3705 = extractvalue { float, float, float, float } %3703, 1, !dbg !92 + %3706 = extractvalue { float, float, float, float } %3703, 2, !dbg !92 + %3707 = extractvalue { float, float, float, float } %3703, 3, !dbg !92 + %3708 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3629, float %3630, float %3631, float %3632, i32 %3424, i32 %3425, i32 %3426, i32 %3427, i32 %2987, i32 %2988) #2, !dbg !92 + %3709 = extractvalue { float, float, float, float } %3708, 0, !dbg !92 + %3710 = extractvalue { float, float, float, float } %3708, 1, !dbg !92 + %3711 = extractvalue { float, float, float, float } %3708, 2, !dbg !92 + %3712 = extractvalue { float, float, float, float } %3708, 3, !dbg !92 + %3713 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3634, float %3635, float %3636, float %3637, i32 %3424, i32 %3425, i32 %3426, i32 %3427, i32 %2997, i32 %2998) #2, !dbg !92 + %3714 = extractvalue { float, float, float, float } %3713, 0, !dbg !92 + %3715 = extractvalue { float, float, float, float } %3713, 1, !dbg !92 + %3716 = extractvalue { float, float, float, float } %3713, 2, !dbg !92 + %3717 = extractvalue { float, float, float, float } %3713, 3, !dbg !92 + %3718 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3639, float %3640, float %3641, float %3642, i32 %3424, i32 %3425, i32 %3426, i32 %3427, i32 %3007, i32 %3008) #2, !dbg !92 + %3719 = extractvalue { float, float, float, float } %3718, 0, !dbg !92 + %3720 = extractvalue { float, float, float, float } %3718, 1, !dbg !92 + %3721 = extractvalue { float, float, float, float } %3718, 2, !dbg !92 + %3722 = extractvalue { float, float, float, float } %3718, 3, !dbg !92 + %3723 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3644, float %3645, float %3646, float %3647, i32 %3424, i32 %3425, i32 %3426, i32 %3427, i32 %3017, i32 %3018) #2, !dbg !92 + %3724 = extractvalue { float, float, float, float } %3723, 0, !dbg !92 + %3725 = extractvalue { float, float, float, float } %3723, 1, !dbg !92 + %3726 = extractvalue { float, float, float, float } %3723, 2, !dbg !92 + %3727 = extractvalue { float, float, float, float } %3723, 3, !dbg !92 + %3728 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3649, float %3650, float %3651, float %3652, i32 %3424, i32 %3425, i32 %3426, i32 %3427, i32 %3027, i32 %3028) #2, !dbg !92 + %3729 = extractvalue { float, float, float, float } %3728, 0, !dbg !92 + %3730 = extractvalue { float, float, float, float } %3728, 1, !dbg !92 + %3731 = extractvalue { float, float, float, float } %3728, 2, !dbg !92 + %3732 = extractvalue { float, float, float, float } %3728, 3, !dbg !92 + %3733 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3654, float %3655, float %3656, float %3657, i32 %3424, i32 %3425, i32 %3426, i32 %3427, i32 %3037, i32 %3038) #2, !dbg !92 + %3734 = extractvalue { float, float, float, float } %3733, 0, !dbg !92 + %3735 = extractvalue { float, float, float, float } %3733, 1, !dbg !92 + %3736 = extractvalue { float, float, float, float } %3733, 2, !dbg !92 + %3737 = extractvalue { float, float, float, float } %3733, 3, !dbg !92 + %3738 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3659, float %3660, float %3661, float %3662, i32 %3424, i32 %3425, i32 %3426, i32 %3427, i32 %3047, i32 %3048) #2, !dbg !92 + %3739 = extractvalue { float, float, float, float } %3738, 0, !dbg !92 + %3740 = extractvalue { float, float, float, float } %3738, 1, !dbg !92 + %3741 = extractvalue { float, float, float, float } %3738, 2, !dbg !92 + %3742 = extractvalue { float, float, float, float } %3738, 3, !dbg !92 + %3743 = tail call { float, float, float, float } asm sideeffect "mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { $0, $1, $2, $3 }, { $8, $9, $10, $11 }, { $12, $13 }, { $4, $5, $6, $7 };", "=f,=f,=f,=f,0,1,2,3,r,r,r,r,r,r"(float %3664, float %3665, float %3666, float %3667, i32 %3424, i32 %3425, i32 %3426, i32 %3427, i32 %3057, i32 %3058) #2, !dbg !92 + %3744 = extractvalue { float, float, float, float } %3743, 0, !dbg !92 + %3745 = extractvalue { float, float, float, float } %3743, 1, !dbg !92 + %3746 = extractvalue { float, float, float, float } %3743, 2, !dbg !92 + %3747 = extractvalue { float, float, float, float } %3743, 3, !dbg !92 + %3748 = add i32 %686, 1, !dbg !38 + %3749 = icmp sgt i32 %3748, 2, !dbg !38 + %3750 = select i1 %3749, i32 0, i32 %3748, !dbg !38 + %3751 = shl i64 %indvars.iv, 6, !dbg !93 + %3752 = add i64 %3751, 192, !dbg !93 + %3753 = or disjoint i64 %3752, %49, !dbg !60 + %3754 = or disjoint i64 %3752, %50, !dbg !60 + %3755 = or disjoint i64 %3752, %51, !dbg !60 + %3756 = or disjoint i64 %3752, %52, !dbg !60 + %3757 = or disjoint i64 %3752, %53, !dbg !60 + %3758 = or disjoint i64 %3752, %54, !dbg !60 + %3759 = or disjoint i64 %3752, %55, !dbg !60 + %3760 = or disjoint i64 %3752, %56, !dbg !60 + %3761 = insertelement <4 x i64> poison, i64 %3752, i64 0, !dbg !60 + %3762 = shufflevector <4 x i64> %3761, <4 x i64> poison, <4 x i32> zeroinitializer, !dbg !60 + %3763 = or disjoint <4 x i64> %3762, %61, !dbg !60 + %3764 = or disjoint i64 %3752, %62, !dbg !60 + %3765 = icmp slt i64 %3753, %15, !dbg !39 + %3766 = icmp slt i64 %3754, %15, !dbg !39 + %3767 = icmp slt i64 %3755, %15, !dbg !39 + %3768 = icmp slt i64 %3756, %15, !dbg !39 + %3769 = icmp slt i64 %3757, %15, !dbg !39 + %3770 = icmp slt i64 %3758, %15, !dbg !39 + %3771 = icmp slt i64 %3759, %15, !dbg !39 + %3772 = icmp slt i64 %3760, %15, !dbg !39 + %3773 = icmp slt i64 %3764, %15, !dbg !39 + %3774 = shl nuw nsw i64 %3753, 12, !dbg !40 + %3775 = shl nuw nsw i64 %3754, 12, !dbg !40 + %3776 = shl nuw nsw i64 %3755, 12, !dbg !40 + %3777 = shl nuw nsw i64 %3756, 12, !dbg !40 + %3778 = shl nuw nsw i64 %3757, 12, !dbg !40 + %3779 = shl nuw nsw i64 %3758, 12, !dbg !40 + %3780 = shl nuw nsw i64 %3759, 12, !dbg !40 + %3781 = shl nuw nsw i64 %3760, 12, !dbg !40 + %3782 = getelementptr bfloat, ptr addrspace(1) %23, i64 %3774, !dbg !41 + %3783 = getelementptr bfloat, ptr addrspace(1) %23, i64 %3775, !dbg !41 + %3784 = getelementptr bfloat, ptr addrspace(1) %23, i64 %3776, !dbg !41 + %3785 = getelementptr bfloat, ptr addrspace(1) %23, i64 %3777, !dbg !41 + %3786 = getelementptr bfloat, ptr addrspace(1) %23, i64 %3778, !dbg !41 + %3787 = getelementptr bfloat, ptr addrspace(1) %23, i64 %3779, !dbg !41 + %3788 = getelementptr bfloat, ptr addrspace(1) %23, i64 %3780, !dbg !41 + %3789 = getelementptr bfloat, ptr addrspace(1) %23, i64 %3781, !dbg !41 + %3790 = getelementptr bfloat, ptr addrspace(1) %3782, i64 %48, !dbg !42 + %3791 = getelementptr bfloat, ptr addrspace(1) %3783, i64 %48, !dbg !42 + %3792 = getelementptr bfloat, ptr addrspace(1) %3784, i64 %48, !dbg !42 + %3793 = getelementptr bfloat, ptr addrspace(1) %3785, i64 %48, !dbg !42 + %3794 = getelementptr bfloat, ptr addrspace(1) %3786, i64 %48, !dbg !42 + %3795 = getelementptr bfloat, ptr addrspace(1) %3787, i64 %48, !dbg !42 + %3796 = getelementptr bfloat, ptr addrspace(1) %3788, i64 %48, !dbg !42 + %3797 = getelementptr bfloat, ptr addrspace(1) %3789, i64 %48, !dbg !42 + %3798 = shl i32 %3750, 12, !dbg !43 + %3799 = getelementptr bfloat, ptr addrspace(3) @global_smem, i32 %3798, !dbg !43 + %3800 = and i1 %819, %3765, !dbg !38 + %3801 = and i1 %819, %3766, !dbg !38 + %3802 = and i1 %819, %3767, !dbg !38 + %3803 = and i1 %819, %3768, !dbg !38 + %3804 = and i1 %819, %3769, !dbg !38 + %3805 = and i1 %819, %3770, !dbg !38 + %3806 = and i1 %819, %3771, !dbg !38 + %3807 = and i1 %819, %3772, !dbg !38 + %3808 = getelementptr inbounds nuw i8, ptr addrspace(3) %3799, i32 %113, !dbg !43 + %3809 = select i1 %3800, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %3808, ptr addrspace(1) %3790, i32 %3809) #2, !dbg !43 + %3810 = getelementptr inbounds nuw i8, ptr addrspace(3) %3799, i32 %116, !dbg !43 + %3811 = select i1 %3801, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3810, ptr addrspace(1) %3791, i32 %3811) #2, !dbg !43 + %3812 = getelementptr inbounds nuw i8, ptr addrspace(3) %3799, i32 %119, !dbg !43 + %3813 = select i1 %3802, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3812, ptr addrspace(1) %3792, i32 %3813) #2, !dbg !43 + %3814 = getelementptr inbounds nuw i8, ptr addrspace(3) %3799, i32 %122, !dbg !43 + %3815 = select i1 %3803, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3814, ptr addrspace(1) %3793, i32 %3815) #2, !dbg !43 + %3816 = getelementptr inbounds nuw i8, ptr addrspace(3) %3799, i32 %125, !dbg !43 + %3817 = select i1 %3804, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3816, ptr addrspace(1) %3794, i32 %3817) #2, !dbg !43 + %3818 = getelementptr inbounds nuw i8, ptr addrspace(3) %3799, i32 %128, !dbg !43 + %3819 = select i1 %3805, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3818, ptr addrspace(1) %3795, i32 %3819) #2, !dbg !43 + %3820 = getelementptr inbounds nuw i8, ptr addrspace(3) %3799, i32 %131, !dbg !43 + %3821 = select i1 %3806, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3820, ptr addrspace(1) %3796, i32 %3821) #2, !dbg !43 + %3822 = getelementptr inbounds nuw i8, ptr addrspace(3) %3799, i32 %134, !dbg !43 + %3823 = select i1 %3807, i32 16, i32 0, !dbg !43 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3822, ptr addrspace(1) %3797, i32 %3823) #2, !dbg !43 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !43 + %3824 = getelementptr bfloat, ptr addrspace(1) %3782, i64 %67, !dbg !44 + %3825 = getelementptr bfloat, ptr addrspace(1) %3783, i64 %67, !dbg !44 + %3826 = getelementptr bfloat, ptr addrspace(1) %3784, i64 %67, !dbg !44 + %3827 = getelementptr bfloat, ptr addrspace(1) %3785, i64 %67, !dbg !44 + %3828 = getelementptr bfloat, ptr addrspace(1) %3786, i64 %67, !dbg !44 + %3829 = getelementptr bfloat, ptr addrspace(1) %3787, i64 %67, !dbg !44 + %3830 = getelementptr bfloat, ptr addrspace(1) %3788, i64 %67, !dbg !44 + %3831 = getelementptr bfloat, ptr addrspace(1) %3789, i64 %67, !dbg !44 + %3832 = getelementptr bfloat, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 24576), i32 %3798, !dbg !45 + %3833 = getelementptr inbounds nuw i8, ptr addrspace(3) %3832, i32 %113, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %3833, ptr addrspace(1) %3824, i32 %3809) #2, !dbg !45 + %3834 = getelementptr inbounds nuw i8, ptr addrspace(3) %3832, i32 %116, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3834, ptr addrspace(1) %3825, i32 %3811) #2, !dbg !45 + %3835 = getelementptr inbounds nuw i8, ptr addrspace(3) %3832, i32 %119, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3835, ptr addrspace(1) %3826, i32 %3813) #2, !dbg !45 + %3836 = getelementptr inbounds nuw i8, ptr addrspace(3) %3832, i32 %122, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3836, ptr addrspace(1) %3827, i32 %3815) #2, !dbg !45 + %3837 = getelementptr inbounds nuw i8, ptr addrspace(3) %3832, i32 %125, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3837, ptr addrspace(1) %3828, i32 %3817) #2, !dbg !45 + %3838 = getelementptr inbounds nuw i8, ptr addrspace(3) %3832, i32 %128, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3838, ptr addrspace(1) %3829, i32 %3819) #2, !dbg !45 + %3839 = getelementptr inbounds nuw i8, ptr addrspace(3) %3832, i32 %131, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3839, ptr addrspace(1) %3830, i32 %3821) #2, !dbg !45 + %3840 = getelementptr inbounds nuw i8, ptr addrspace(3) %3832, i32 %134, !dbg !45 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3840, ptr addrspace(1) %3831, i32 %3823) #2, !dbg !45 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !45 + %gep528 = getelementptr bfloat, ptr addrspace(1) %invariant.gep527, i64 %3774, !dbg !47 + %gep530 = getelementptr bfloat, ptr addrspace(1) %invariant.gep527, i64 %3775, !dbg !47 + %gep532 = getelementptr bfloat, ptr addrspace(1) %invariant.gep527, i64 %3776, !dbg !47 + %gep534 = getelementptr bfloat, ptr addrspace(1) %invariant.gep527, i64 %3777, !dbg !47 + %gep536 = getelementptr bfloat, ptr addrspace(1) %invariant.gep527, i64 %3778, !dbg !47 + %gep538 = getelementptr bfloat, ptr addrspace(1) %invariant.gep527, i64 %3779, !dbg !47 + %gep540 = getelementptr bfloat, ptr addrspace(1) %invariant.gep527, i64 %3780, !dbg !47 + %gep542 = getelementptr bfloat, ptr addrspace(1) %invariant.gep527, i64 %3781, !dbg !47 + %3841 = getelementptr bfloat, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 49152), i32 %3798, !dbg !48 + %3842 = getelementptr inbounds nuw i8, ptr addrspace(3) %3841, i32 %113, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %3842, ptr addrspace(1) %gep528, i32 %3809) #2, !dbg !48 + %3843 = getelementptr inbounds nuw i8, ptr addrspace(3) %3841, i32 %116, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3843, ptr addrspace(1) %gep530, i32 %3811) #2, !dbg !48 + %3844 = getelementptr inbounds nuw i8, ptr addrspace(3) %3841, i32 %119, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3844, ptr addrspace(1) %gep532, i32 %3813) #2, !dbg !48 + %3845 = getelementptr inbounds nuw i8, ptr addrspace(3) %3841, i32 %122, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3845, ptr addrspace(1) %gep534, i32 %3815) #2, !dbg !48 + %3846 = getelementptr inbounds nuw i8, ptr addrspace(3) %3841, i32 %125, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3846, ptr addrspace(1) %gep536, i32 %3817) #2, !dbg !48 + %3847 = getelementptr inbounds nuw i8, ptr addrspace(3) %3841, i32 %128, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3847, ptr addrspace(1) %gep538, i32 %3819) #2, !dbg !48 + %3848 = getelementptr inbounds nuw i8, ptr addrspace(3) %3841, i32 %131, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3848, ptr addrspace(1) %gep540, i32 %3821) #2, !dbg !48 + %3849 = getelementptr inbounds nuw i8, ptr addrspace(3) %3841, i32 %134, !dbg !48 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3849, ptr addrspace(1) %gep542, i32 %3823) #2, !dbg !48 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !48 + %indvars.iv.tr = trunc i64 %indvars.iv to i32, !dbg !49 + %3850 = shl i32 %indvars.iv.tr, 6, !dbg !49 + %3851 = add i32 %3850, 256, !dbg !49 + %3852 = tail call i32 @llvm.smin.i32(i32 %3851, i32 %6), !dbg !49 + %3853 = shl i32 %3852, 5, !dbg !50 + %3854 = add i32 %3853, -32, !dbg !50 + %3855 = sext i32 %3854 to i64, !dbg !51 + %gep544 = getelementptr float, ptr addrspace(1) %invariant.gep543, i64 %3855, !dbg !52 + %3856 = getelementptr float, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 142080), i32 %3750, !dbg !53 + %3857 = select i1 %819, i32 4, i32 0, !dbg !53 + tail call void asm sideeffect "@$3 cp.async.ca.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x4, $2;", "r,l,r,b"(ptr addrspace(3) %3856, ptr addrspace(1) %gep544, i32 %3857, i1 %184) #2, !dbg !53 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !53 + %.idx499 = shl nuw nsw i64 %3764, 7, !dbg !54 + %3858 = getelementptr i8, ptr addrspace(1) %65, i64 %.idx499, !dbg !54 + %.idx500 = shl i32 %3750, 8, !dbg !55 + %3859 = and i1 %819, %3773, !dbg !38 + %gep546 = getelementptr i8, ptr addrspace(3) %189, i32 %.idx500, !dbg !55 + %3860 = select i1 %3859, i32 4, i32 0, !dbg !55 + tail call void asm sideeffect "cp.async.ca.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x4, $2;", "r,l,r"(ptr addrspace(3) %gep546, ptr addrspace(1) %3858, i32 %3860) #2, !dbg !55 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !55 + %3861 = getelementptr bfloat, ptr addrspace(1) %66, i64 %3774, !dbg !56 + %3862 = getelementptr bfloat, ptr addrspace(1) %66, i64 %3775, !dbg !56 + %3863 = getelementptr bfloat, ptr addrspace(1) %66, i64 %3776, !dbg !56 + %3864 = getelementptr bfloat, ptr addrspace(1) %66, i64 %3777, !dbg !56 + %3865 = getelementptr bfloat, ptr addrspace(1) %66, i64 %3778, !dbg !56 + %3866 = getelementptr bfloat, ptr addrspace(1) %66, i64 %3779, !dbg !56 + %3867 = getelementptr bfloat, ptr addrspace(1) %66, i64 %3780, !dbg !56 + %3868 = getelementptr bfloat, ptr addrspace(1) %66, i64 %3781, !dbg !56 + %3869 = getelementptr bfloat, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73728), i32 %3798, !dbg !57 + %3870 = getelementptr inbounds nuw i8, ptr addrspace(3) %3869, i32 %113, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %3870, ptr addrspace(1) %3861, i32 %3809) #2, !dbg !57 + %3871 = getelementptr inbounds nuw i8, ptr addrspace(3) %3869, i32 %116, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3871, ptr addrspace(1) %3862, i32 %3811) #2, !dbg !57 + %3872 = getelementptr inbounds nuw i8, ptr addrspace(3) %3869, i32 %119, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3872, ptr addrspace(1) %3863, i32 %3813) #2, !dbg !57 + %3873 = getelementptr inbounds nuw i8, ptr addrspace(3) %3869, i32 %122, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3873, ptr addrspace(1) %3864, i32 %3815) #2, !dbg !57 + %3874 = getelementptr inbounds nuw i8, ptr addrspace(3) %3869, i32 %125, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3874, ptr addrspace(1) %3865, i32 %3817) #2, !dbg !57 + %3875 = getelementptr inbounds nuw i8, ptr addrspace(3) %3869, i32 %128, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3875, ptr addrspace(1) %3866, i32 %3819) #2, !dbg !57 + %3876 = getelementptr inbounds nuw i8, ptr addrspace(3) %3869, i32 %131, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3876, ptr addrspace(1) %3867, i32 %3821) #2, !dbg !57 + %3877 = getelementptr inbounds nuw i8, ptr addrspace(3) %3869, i32 %134, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3877, ptr addrspace(1) %3868, i32 %3823) #2, !dbg !57 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !57 + %3878 = getelementptr bfloat, ptr addrspace(1) %68, i64 %3774, !dbg !58 + %3879 = getelementptr bfloat, ptr addrspace(1) %68, i64 %3775, !dbg !58 + %3880 = getelementptr bfloat, ptr addrspace(1) %68, i64 %3776, !dbg !58 + %3881 = getelementptr bfloat, ptr addrspace(1) %68, i64 %3777, !dbg !58 + %3882 = getelementptr bfloat, ptr addrspace(1) %68, i64 %3778, !dbg !58 + %3883 = getelementptr bfloat, ptr addrspace(1) %68, i64 %3779, !dbg !58 + %3884 = getelementptr bfloat, ptr addrspace(1) %68, i64 %3780, !dbg !58 + %3885 = getelementptr bfloat, ptr addrspace(1) %68, i64 %3781, !dbg !58 + %3886 = getelementptr bfloat, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 98304), i32 %3798, !dbg !59 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !59 + %3887 = getelementptr inbounds nuw i8, ptr addrspace(3) %3886, i32 %113, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %3887, ptr addrspace(1) %3878, i32 %3809) #2, !dbg !59 + %3888 = getelementptr inbounds nuw i8, ptr addrspace(3) %3886, i32 %116, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3888, ptr addrspace(1) %3879, i32 %3811) #2, !dbg !59 + %3889 = getelementptr inbounds nuw i8, ptr addrspace(3) %3886, i32 %119, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3889, ptr addrspace(1) %3880, i32 %3813) #2, !dbg !59 + %3890 = getelementptr inbounds nuw i8, ptr addrspace(3) %3886, i32 %122, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3890, ptr addrspace(1) %3881, i32 %3815) #2, !dbg !59 + %3891 = getelementptr inbounds nuw i8, ptr addrspace(3) %3886, i32 %125, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3891, ptr addrspace(1) %3882, i32 %3817) #2, !dbg !59 + %3892 = getelementptr inbounds nuw i8, ptr addrspace(3) %3886, i32 %128, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3892, ptr addrspace(1) %3883, i32 %3819) #2, !dbg !59 + %3893 = getelementptr inbounds nuw i8, ptr addrspace(3) %3886, i32 %131, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3893, ptr addrspace(1) %3884, i32 %3821) #2, !dbg !59 + %3894 = getelementptr inbounds nuw i8, ptr addrspace(3) %3886, i32 %134, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %3894, ptr addrspace(1) %3885, i32 %3823) #2, !dbg !59 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !59 + %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1, !dbg !38 + %exitcond.not = icmp eq i64 %indvars.iv.next, %524, !dbg !38 + br i1 %exitcond.not, label %._crit_edge, label %684, !dbg !38 + +._crit_edge: ; preds = %__nv_exp2f.exit512, %9 + tail call void @llvm.nvvm.cp.async.wait.group(i32 0), !dbg !38 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !38 + ret void, !dbg !94 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: nounwind +declare void @llvm.nvvm.cp.async.commit.group() #2 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind speculatable willreturn memory(none) +declare i32 @llvm.smin.i32(i32, i32) #3 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #4 + +; Function Attrs: nounwind +declare void @llvm.nvvm.cp.async.wait.group(i32 immarg) #2 + +; Function Attrs: convergent nocallback nounwind memory(argmem: write) +declare void @llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) writeonly captures(none), i32, i32, i32, i32) #5 + +; Function Attrs: convergent nocallback nofree nounwind memory(argmem: read) +declare { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) readonly captures(none)) #6 + +; Function Attrs: convergent nocallback nofree nounwind memory(argmem: read) +declare { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3) readonly captures(none)) #6 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #7 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.ex2.approx.ftz.f32(float) #8 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.ex2.approx.f32(float) #8 + +attributes #0 = { nounwind "nvvm.reqntid"="64" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { nounwind } +attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind speculatable willreturn memory(none) } +attributes #4 = { convergent nocallback nounwind } +attributes #5 = { convergent nocallback nounwind memory(argmem: write) } +attributes #6 = { convergent nocallback nofree nounwind memory(argmem: read) } +attributes #7 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #8 = { mustprogress nocallback nofree nosync nounwind willreturn memory(none) } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "chunk_delta_h.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "chunk_gated_delta_rule_fwd_kernel_h_blockdim64", linkageName: "chunk_gated_delta_rule_fwd_kernel_h_blockdim64", scope: !1, file: !1, line: 59, type: !6, scopeLine: 59, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !8, !8, !10, !8, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 86, column: 24, scope: !5) +!14 = !DILocation(line: 86, column: 30, scope: !5) +!15 = !DILocation(line: 88, column: 34, scope: !5) +!16 = !DILocation(line: 89, column: 23, scope: !5) +!17 = !DILocation(line: 89, column: 34, scope: !5) +!18 = !DILocation(line: 96, column: 25, scope: !5) +!19 = !DILocation(line: 43, column: 17, scope: !20, inlinedAt: !22) +!20 = distinct !DILexicalBlockFile(scope: !5, file: !21, discriminator: 0) +!21 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!22 = !DILocation(line: 97, column: 24, scope: !23) +!23 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!24 = !DILocation(line: 119, column: 16, scope: !5) +!25 = !DILocation(line: 119, column: 21, scope: !5) +!26 = !DILocation(line: 119, column: 41, scope: !5) +!27 = !DILocation(line: 119, column: 9, scope: !5) +!28 = !DILocation(line: 120, column: 9, scope: !5) +!29 = !DILocation(line: 121, column: 9, scope: !5) +!30 = !DILocation(line: 123, column: 17, scope: !5) +!31 = !DILocation(line: 131, column: 16, scope: !5) +!32 = !DILocation(line: 131, column: 34, scope: !5) +!33 = !DILocation(line: 131, column: 21, scope: !5) +!34 = !DILocation(line: 135, column: 16, scope: !5) +!35 = !DILocation(line: 248, column: 22, scope: !5) +!36 = !DILocation(line: 290, column: 18, scope: !5) +!37 = !DILocation(line: 297, column: 22, scope: !5) +!38 = !DILocation(line: 175, column: 21, scope: !5) +!39 = !DILocation(line: 178, column: 20, scope: !5) +!40 = !DILocation(line: 211, column: 34, scope: !5) +!41 = !DILocation(line: 211, column: 18, scope: !5) +!42 = !DILocation(line: 211, column: 42, scope: !5) +!43 = !DILocation(line: 212, column: 22, scope: !5) +!44 = !DILocation(line: 218, column: 46, scope: !5) +!45 = !DILocation(line: 219, column: 26, scope: !5) +!46 = !DILocation(line: 238, column: 18, scope: !5) +!47 = !DILocation(line: 238, column: 42, scope: !5) +!48 = !DILocation(line: 239, column: 22, scope: !5) +!49 = !DILocation(line: 245, column: 39, scope: !5) +!50 = !DILocation(line: 247, column: 58, scope: !5) +!51 = !DILocation(line: 247, column: 47, scope: !5) +!52 = !DILocation(line: 247, column: 35, scope: !5) +!53 = !DILocation(line: 247, column: 31, scope: !5) +!54 = !DILocation(line: 248, column: 54, scope: !5) +!55 = !DILocation(line: 249, column: 26, scope: !5) +!56 = !DILocation(line: 290, column: 34, scope: !5) +!57 = !DILocation(line: 291, column: 22, scope: !5) +!58 = !DILocation(line: 297, column: 38, scope: !5) +!59 = !DILocation(line: 298, column: 26, scope: !5) +!60 = !DILocation(line: 177, column: 31, scope: !5) +!61 = !DILocation(line: 43, column: 30, scope: !20, inlinedAt: !22) +!62 = !DILocation(line: 98, column: 20, scope: !5) +!63 = !DILocation(line: 183, column: 60, scope: !5) +!64 = !DILocation(line: 0, scope: !5) +!65 = !DILocation(line: 183, column: 44, scope: !5) +!66 = !DILocation(line: 183, column: 64, scope: !5) +!67 = !DILocation(line: 185, column: 31, scope: !5) +!68 = !DILocation(line: 185, column: 23, scope: !5) +!69 = !DILocation(line: 191, column: 48, scope: !5) +!70 = !DILocation(line: 191, column: 68, scope: !5) +!71 = !DILocation(line: 193, column: 35, scope: !5) +!72 = !DILocation(line: 193, column: 27, scope: !5) +!73 = !DILocation(line: 216, column: 30, scope: !5) +!74 = !DILocation(line: 223, column: 54, scope: !5) +!75 = !DILocation(line: 239, column: 74, scope: !5) +!76 = !DILocation(line: 242, column: 50, scope: !5) +!77 = !DILocation(line: 243, column: 33, scope: !5) +!78 = !DILocation(line: 243, column: 26, scope: !5) +!79 = !DILocation(line: 250, column: 54, scope: !5) +!80 = !DILocation(line: 47, column: 33, scope: !81, inlinedAt: !83) +!81 = distinct !DILexicalBlockFile(scope: !5, file: !82, discriminator: 0) +!82 = !DIFile(filename: "op.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils") +!83 = !DILocation(line: 250, column: 43, scope: !23) +!84 = !DILocation(line: 250, column: 60, scope: !5) +!85 = !DILocation(line: 250, column: 24, scope: !5) +!86 = !DILocation(line: 47, column: 33, scope: !81, inlinedAt: !87) +!87 = !DILocation(line: 251, column: 28, scope: !23) +!88 = !DILocation(line: 252, column: 20, scope: !5) +!89 = !DILocation(line: 254, column: 24, scope: !5) +!90 = !DILocation(line: 288, column: 21, scope: !5) +!91 = !DILocation(line: 295, column: 36, scope: !5) +!92 = !DILocation(line: 302, column: 40, scope: !5) +!93 = !DILocation(line: 177, column: 26, scope: !5) +!94 = !DILocation(line: 318, column: 4, scope: !5) diff --git a/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.ptx b/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.ptx new file mode 100644 index 0000000000000000000000000000000000000000..2798633bce152f52da0b6c17dedde4a0afbed7bf --- /dev/null +++ b/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.ptx @@ -0,0 +1,3794 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl chunk_gated_delta_rule_fwd_kernel_h_blockdim64 // -- Begin function chunk_gated_delta_rule_fwd_kernel_h_blockdim64 +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; + // @chunk_gated_delta_rule_fwd_kernel_h_blockdim64 +.visible .entry chunk_gated_delta_rule_fwd_kernel_h_blockdim64( + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_fwd_kernel_h_blockdim64_param_0, + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_fwd_kernel_h_blockdim64_param_1, + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_fwd_kernel_h_blockdim64_param_2, + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_fwd_kernel_h_blockdim64_param_3, + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_fwd_kernel_h_blockdim64_param_4, + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_fwd_kernel_h_blockdim64_param_5, + .param .u32 chunk_gated_delta_rule_fwd_kernel_h_blockdim64_param_6, + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_fwd_kernel_h_blockdim64_param_7, + .param .u64 .ptr .global .align 1 chunk_gated_delta_rule_fwd_kernel_h_blockdim64_param_8 +) +.reqntid 64 +{ + .reg .pred %p<56>; + .reg .b16 %rs<65>; + .reg .b32 %r<1420>; + .reg .b64 %rd<483>; + .loc 1 59 0 // chunk_delta_h.py:59:0 +$L__func_begin0: + .loc 1 59 0 // chunk_delta_h.py:59:0 + +// %bb.0: + ld.param.b32 %r71, [chunk_gated_delta_rule_fwd_kernel_h_blockdim64_param_6]; + ld.param.b64 %rd27, [chunk_gated_delta_rule_fwd_kernel_h_blockdim64_param_4]; + ld.param.b64 %rd25, [chunk_gated_delta_rule_fwd_kernel_h_blockdim64_param_2]; + ld.param.b64 %rd24, [chunk_gated_delta_rule_fwd_kernel_h_blockdim64_param_1]; + ld.param.b64 %rd23, [chunk_gated_delta_rule_fwd_kernel_h_blockdim64_param_0]; +$L__tmp0: + .loc 1 86 24 // chunk_delta_h.py:86:24 + mov.u32 %r1, %ctaid.x; + .loc 1 86 30 // chunk_delta_h.py:86:30 + cvt.u64.u32 %rd155, %r1; + .loc 1 89 23 // chunk_delta_h.py:89:23 + shr.u64 %rd1, %rd155, 6; + .loc 1 89 34 // chunk_delta_h.py:89:34 + bfe.u64 %rd2, %rd155, 1, 5; + .loc 1 96 25 // chunk_delta_h.py:96:25 + cvt.s64.s32 %rd3, %r71; +$L__tmp1: + .loc 2 43 17 // standard.py:43:17 @[ chunk_delta_h.py:97:24 ] + add.s32 %r2, %r71, 63; +$L__tmp2: + .loc 1 119 16 // chunk_delta_h.py:119:16 + mul.lo.s64 %rd156, %rd1, %rd3; + shl.b64 %rd157, %rd156, 5; + .loc 1 119 21 // chunk_delta_h.py:119:21 + or.b64 %rd158, %rd157, %rd2; + .loc 1 119 9 // chunk_delta_h.py:119:9 + shl.b64 %rd159, %rd158, 8; + add.s64 %rd160, %rd24, %rd159; + .loc 1 120 9 // chunk_delta_h.py:120:9 + add.s64 %rd161, %rd23, %rd159; + .loc 1 121 9 // chunk_delta_h.py:121:9 + add.s64 %rd162, %rd25, %rd159; + .loc 1 131 16 // chunk_delta_h.py:131:16 + mul.wide.u32 %rd163, %r1, 64; + and.b64 %rd164, %rd163, 64; + .loc 1 131 34 // chunk_delta_h.py:131:34 + mov.u32 %r3, %tid.x; + and.b32 %r4, %r3, 7; + shl.b32 %r228, %r4, 3; + and.b32 %r5, %r3, 56; + bfe.u32 %r229, %r3, 3, 3; + or.b32 %r230, %r229, 8; + or.b32 %r231, %r229, 16; + or.b32 %r232, %r229, 24; + or.b32 %r233, %r229, 32; + or.b32 %r234, %r229, 40; + or.b32 %r235, %r229, 48; + or.b32 %r236, %r229, 56; + and.b32 %r243, %r3, 63; + .loc 1 131 21 // chunk_delta_h.py:131:21 + cvt.u64.u32 %rd165, %r228; + cvt.u64.u32 %rd4, %r229; + cvt.u64.u32 %rd166, %r230; + cvt.u64.u32 %rd167, %r231; + cvt.u64.u32 %rd168, %r232; + cvt.u64.u32 %rd169, %r233; + cvt.u64.u32 %rd170, %r234; + cvt.u64.u32 %rd171, %r235; + cvt.u64.u32 %rd172, %r236; + cvt.u64.u32 %rd11, %r243; + or.b64 %rd173, %rd164, %rd165; + .loc 1 248 22 // chunk_delta_h.py:248:22 + shl.b64 %rd174, %rd158, 2; + add.s64 %rd175, %rd27, %rd174; + .loc 1 290 18 // chunk_delta_h.py:290:18 + mul.wide.u32 %rd176, %r228, 2; + add.s64 %rd177, %rd161, %rd176; + .loc 1 175 21 // chunk_delta_h.py:175:21 + setp.lt.s32 %p14, %r2, 64; + setp.gt.s32 %p15, %r2, 63; + .loc 1 178 20 // chunk_delta_h.py:178:20 + setp.lt.s32 %p51, %r229, %r71; + setp.lt.s32 %p50, %r230, %r71; + setp.lt.s32 %p49, %r231, %r71; + setp.lt.s32 %p48, %r232, %r71; + setp.lt.s32 %p47, %r233, %r71; + setp.lt.s32 %p46, %r234, %r71; + setp.lt.s32 %p45, %r235, %r71; + setp.lt.s32 %p44, %r236, %r71; + setp.lt.s32 %p16, %r243, %r71; + .loc 1 211 18 // chunk_delta_h.py:211:18 + mul.wide.u32 %rd178, %r229, 8192; + add.s64 %rd179, %rd162, %rd178; + mul.wide.u32 %rd180, %r230, 8192; + add.s64 %rd181, %rd162, %rd180; + mul.wide.u32 %rd182, %r231, 8192; + add.s64 %rd183, %rd162, %rd182; + mul.wide.u32 %rd184, %r232, 8192; + add.s64 %rd185, %rd162, %rd184; + mul.wide.u32 %rd186, %r233, 8192; + add.s64 %rd187, %rd162, %rd186; + mul.wide.u32 %rd188, %r234, 8192; + add.s64 %rd189, %rd162, %rd188; + mul.wide.u32 %rd190, %r235, 8192; + add.s64 %rd191, %rd162, %rd190; + mul.wide.u32 %rd192, %r236, 8192; + add.s64 %rd193, %rd162, %rd192; + .loc 1 211 42 // chunk_delta_h.py:211:42 + add.s64 %rd29, %rd179, %rd176; + add.s64 %rd30, %rd181, %rd176; + add.s64 %rd31, %rd183, %rd176; + add.s64 %rd32, %rd185, %rd176; + add.s64 %rd33, %rd187, %rd176; + add.s64 %rd34, %rd189, %rd176; + add.s64 %rd35, %rd191, %rd176; + add.s64 %rd36, %rd193, %rd176; + .loc 1 212 22 // chunk_delta_h.py:212:22 + shl.b32 %r244, %r243, 4; + shl.b32 %r245, %r5, 1; + xor.b32 %r8, %r244, %r245; + mov.b32 %r246, global_smem; + add.s32 %r72, %r246, %r8; + selp.b32 %r247, 16, 0, %p15; + selp.b32 %r81, %r247, 0, %p51; + // begin inline asm + cp.async.cg.shared.global [ %r72 + 0 ], [ %rd29 + 0 ], 0x10, %r81; + // end inline asm + add.s32 %r73, %r72, 1024; + selp.b32 %r83, %r247, 0, %p50; + // begin inline asm + cp.async.cg.shared.global [ %r73 + 0 ], [ %rd30 + 0 ], 0x10, %r83; + // end inline asm + add.s32 %r74, %r72, 2048; + selp.b32 %r85, %r247, 0, %p49; + // begin inline asm + cp.async.cg.shared.global [ %r74 + 0 ], [ %rd31 + 0 ], 0x10, %r85; + // end inline asm + add.s32 %r75, %r72, 3072; + selp.b32 %r87, %r247, 0, %p48; + // begin inline asm + cp.async.cg.shared.global [ %r75 + 0 ], [ %rd32 + 0 ], 0x10, %r87; + // end inline asm + add.s32 %r76, %r72, 4096; + selp.b32 %r89, %r247, 0, %p47; + // begin inline asm + cp.async.cg.shared.global [ %r76 + 0 ], [ %rd33 + 0 ], 0x10, %r89; + // end inline asm + add.s32 %r77, %r72, 5120; + selp.b32 %r91, %r247, 0, %p46; + // begin inline asm + cp.async.cg.shared.global [ %r77 + 0 ], [ %rd34 + 0 ], 0x10, %r91; + // end inline asm + add.s32 %r78, %r72, 6144; + selp.b32 %r93, %r247, 0, %p45; + // begin inline asm + cp.async.cg.shared.global [ %r78 + 0 ], [ %rd35 + 0 ], 0x10, %r93; + // end inline asm + add.s32 %r79, %r72, 7168; + selp.b32 %r95, %r247, 0, %p44; + // begin inline asm + cp.async.cg.shared.global [ %r79 + 0 ], [ %rd36 + 0 ], 0x10, %r95; + // end inline asm + cp.async.commit_group; + .loc 1 218 46 // chunk_delta_h.py:218:46 + add.s64 %rd37, %rd29, 128; + add.s64 %rd38, %rd30, 128; + add.s64 %rd39, %rd31, 128; + add.s64 %rd40, %rd32, 128; + add.s64 %rd41, %rd33, 128; + add.s64 %rd42, %rd34, 128; + add.s64 %rd43, %rd35, 128; + add.s64 %rd44, %rd36, 128; + .loc 1 219 26 // chunk_delta_h.py:219:26 + add.s32 %r80, %r72, 24576; + // begin inline asm + cp.async.cg.shared.global [ %r80 + 0 ], [ %rd37 + 0 ], 0x10, %r81; + // end inline asm + add.s32 %r82, %r72, 25600; + // begin inline asm + cp.async.cg.shared.global [ %r82 + 0 ], [ %rd38 + 0 ], 0x10, %r83; + // end inline asm + add.s32 %r84, %r72, 26624; + // begin inline asm + cp.async.cg.shared.global [ %r84 + 0 ], [ %rd39 + 0 ], 0x10, %r85; + // end inline asm + add.s32 %r86, %r72, 27648; + // begin inline asm + cp.async.cg.shared.global [ %r86 + 0 ], [ %rd40 + 0 ], 0x10, %r87; + // end inline asm + add.s32 %r88, %r72, 28672; + // begin inline asm + cp.async.cg.shared.global [ %r88 + 0 ], [ %rd41 + 0 ], 0x10, %r89; + // end inline asm + add.s32 %r90, %r72, 29696; + // begin inline asm + cp.async.cg.shared.global [ %r90 + 0 ], [ %rd42 + 0 ], 0x10, %r91; + // end inline asm + add.s32 %r92, %r72, 30720; + // begin inline asm + cp.async.cg.shared.global [ %r92 + 0 ], [ %rd43 + 0 ], 0x10, %r93; + // end inline asm + add.s32 %r94, %r72, 31744; + // begin inline asm + cp.async.cg.shared.global [ %r94 + 0 ], [ %rd44 + 0 ], 0x10, %r95; + // end inline asm + cp.async.commit_group; + .loc 1 238 18 // chunk_delta_h.py:238:18 + add.s64 %rd194, %rd160, %rd178; + add.s64 %rd195, %rd160, %rd180; + add.s64 %rd196, %rd160, %rd182; + add.s64 %rd197, %rd160, %rd184; + add.s64 %rd198, %rd160, %rd186; + add.s64 %rd199, %rd160, %rd188; + add.s64 %rd200, %rd160, %rd190; + add.s64 %rd201, %rd160, %rd192; + .loc 1 238 42 // chunk_delta_h.py:238:42 + shl.b64 %rd202, %rd173, 1; + add.s64 %rd45, %rd194, %rd202; + add.s64 %rd46, %rd195, %rd202; + add.s64 %rd47, %rd196, %rd202; + add.s64 %rd48, %rd197, %rd202; + add.s64 %rd49, %rd198, %rd202; + add.s64 %rd50, %rd199, %rd202; + add.s64 %rd51, %rd200, %rd202; + add.s64 %rd52, %rd201, %rd202; + .loc 1 239 22 // chunk_delta_h.py:239:22 + add.s32 %r96, %r72, 49152; + // begin inline asm + cp.async.cg.shared.global [ %r96 + 0 ], [ %rd45 + 0 ], 0x10, %r81; + // end inline asm + add.s32 %r97, %r72, 50176; + // begin inline asm + cp.async.cg.shared.global [ %r97 + 0 ], [ %rd46 + 0 ], 0x10, %r83; + // end inline asm + add.s32 %r98, %r72, 51200; + // begin inline asm + cp.async.cg.shared.global [ %r98 + 0 ], [ %rd47 + 0 ], 0x10, %r85; + // end inline asm + add.s32 %r99, %r72, 52224; + // begin inline asm + cp.async.cg.shared.global [ %r99 + 0 ], [ %rd48 + 0 ], 0x10, %r87; + // end inline asm + add.s32 %r100, %r72, 53248; + // begin inline asm + cp.async.cg.shared.global [ %r100 + 0 ], [ %rd49 + 0 ], 0x10, %r89; + // end inline asm + add.s32 %r101, %r72, 54272; + // begin inline asm + cp.async.cg.shared.global [ %r101 + 0 ], [ %rd50 + 0 ], 0x10, %r91; + // end inline asm + add.s32 %r102, %r72, 55296; + // begin inline asm + cp.async.cg.shared.global [ %r102 + 0 ], [ %rd51 + 0 ], 0x10, %r93; + // end inline asm + add.s32 %r103, %r72, 56320; + // begin inline asm + cp.async.cg.shared.global [ %r103 + 0 ], [ %rd52 + 0 ], 0x10, %r95; + // end inline asm + cp.async.commit_group; + .loc 1 245 39 // chunk_delta_h.py:245:39 + min.s32 %r248, %r71, 64; + .loc 1 247 58 // chunk_delta_h.py:247:58 + shl.b32 %r249, %r248, 5; + add.s32 %r250, %r249, -32; + .loc 1 247 35 // chunk_delta_h.py:247:35 + shl.b64 %rd203, %rd156, 7; + add.s64 %rd204, %rd27, %rd203; + shl.b64 %rd205, %rd2, 2; + .loc 1 175 21 // chunk_delta_h.py:175:21 + add.s64 %rd13, %rd204, %rd205; + .loc 1 247 35 // chunk_delta_h.py:247:35 + mad.wide.s32 %rd53, %r250, 4, %rd13; + .loc 1 247 31 // chunk_delta_h.py:247:31 + setp.eq.b32 %p13, %r243, 0; + selp.b32 %r105, 4, 0, %p15; + add.s32 %r104, %r246, 142080; + // begin inline asm + @%p13 cp.async.ca.shared.global [ %r104 + 0 ], [ %rd53 + 0 ], 0x4, %r105; + // end inline asm + cp.async.commit_group; + .loc 1 248 54 // chunk_delta_h.py:248:54 + mul.wide.u32 %rd12, %r243, 128; + add.s64 %rd54, %rd175, %rd12; + .loc 1 249 26 // chunk_delta_h.py:249:26 + shl.b32 %r251, %r243, 2; + add.s32 %r9, %r246, %r251; + add.s32 %r106, %r9, 141312; + selp.b32 %r107, %r105, 0, %p16; + // begin inline asm + cp.async.ca.shared.global [ %r106 + 0 ], [ %rd54 + 0 ], 0x4, %r107; + // end inline asm + cp.async.commit_group; + .loc 1 290 34 // chunk_delta_h.py:290:34 + add.s64 %rd55, %rd177, %rd178; + add.s64 %rd56, %rd177, %rd180; + add.s64 %rd57, %rd177, %rd182; + add.s64 %rd58, %rd177, %rd184; + add.s64 %rd59, %rd177, %rd186; + add.s64 %rd60, %rd177, %rd188; + add.s64 %rd61, %rd177, %rd190; + add.s64 %rd62, %rd177, %rd192; + .loc 1 291 22 // chunk_delta_h.py:291:22 + add.s32 %r108, %r72, 73728; + // begin inline asm + cp.async.cg.shared.global [ %r108 + 0 ], [ %rd55 + 0 ], 0x10, %r81; + // end inline asm + add.s32 %r109, %r72, 74752; + // begin inline asm + cp.async.cg.shared.global [ %r109 + 0 ], [ %rd56 + 0 ], 0x10, %r83; + // end inline asm + add.s32 %r110, %r72, 75776; + // begin inline asm + cp.async.cg.shared.global [ %r110 + 0 ], [ %rd57 + 0 ], 0x10, %r85; + // end inline asm + add.s32 %r111, %r72, 76800; + // begin inline asm + cp.async.cg.shared.global [ %r111 + 0 ], [ %rd58 + 0 ], 0x10, %r87; + // end inline asm + add.s32 %r112, %r72, 77824; + // begin inline asm + cp.async.cg.shared.global [ %r112 + 0 ], [ %rd59 + 0 ], 0x10, %r89; + // end inline asm + add.s32 %r113, %r72, 78848; + // begin inline asm + cp.async.cg.shared.global [ %r113 + 0 ], [ %rd60 + 0 ], 0x10, %r91; + // end inline asm + add.s32 %r114, %r72, 79872; + // begin inline asm + cp.async.cg.shared.global [ %r114 + 0 ], [ %rd61 + 0 ], 0x10, %r93; + // end inline asm + add.s32 %r115, %r72, 80896; + // begin inline asm + cp.async.cg.shared.global [ %r115 + 0 ], [ %rd62 + 0 ], 0x10, %r95; + // end inline asm + cp.async.commit_group; + .loc 1 297 38 // chunk_delta_h.py:297:38 + add.s64 %rd63, %rd55, 128; + add.s64 %rd64, %rd56, 128; + add.s64 %rd65, %rd57, 128; + add.s64 %rd66, %rd58, 128; + add.s64 %rd67, %rd59, 128; + add.s64 %rd68, %rd60, 128; + add.s64 %rd69, %rd61, 128; + add.s64 %rd70, %rd62, 128; + .loc 1 298 26 // chunk_delta_h.py:298:26 + add.s32 %r116, %r72, 98304; + // begin inline asm + cp.async.cg.shared.global [ %r116 + 0 ], [ %rd63 + 0 ], 0x10, %r81; + // end inline asm + add.s32 %r117, %r72, 99328; + // begin inline asm + cp.async.cg.shared.global [ %r117 + 0 ], [ %rd64 + 0 ], 0x10, %r83; + // end inline asm + add.s32 %r118, %r72, 100352; + // begin inline asm + cp.async.cg.shared.global [ %r118 + 0 ], [ %rd65 + 0 ], 0x10, %r85; + // end inline asm + add.s32 %r119, %r72, 101376; + // begin inline asm + cp.async.cg.shared.global [ %r119 + 0 ], [ %rd66 + 0 ], 0x10, %r87; + // end inline asm + add.s32 %r120, %r72, 102400; + // begin inline asm + cp.async.cg.shared.global [ %r120 + 0 ], [ %rd67 + 0 ], 0x10, %r89; + // end inline asm + add.s32 %r121, %r72, 103424; + // begin inline asm + cp.async.cg.shared.global [ %r121 + 0 ], [ %rd68 + 0 ], 0x10, %r91; + // end inline asm + add.s32 %r122, %r72, 104448; + // begin inline asm + cp.async.cg.shared.global [ %r122 + 0 ], [ %rd69 + 0 ], 0x10, %r93; + // end inline asm + add.s32 %r123, %r72, 105472; + // begin inline asm + cp.async.cg.shared.global [ %r123 + 0 ], [ %rd70 + 0 ], 0x10, %r95; + // end inline asm + cp.async.commit_group; + .loc 1 175 21 // chunk_delta_h.py:175:21 + setp.gt.s32 %p17, %r2, 127; + .loc 1 177 31 // chunk_delta_h.py:177:31 + or.b64 %rd206, %rd4, 64; + or.b64 %rd207, %rd166, 64; + or.b64 %rd208, %rd167, 64; + or.b64 %rd209, %rd168, 64; + or.b64 %rd210, %rd169, 64; + or.b64 %rd211, %rd170, 64; + or.b64 %rd212, %rd171, 64; + or.b64 %rd213, %rd172, 64; + or.b64 %rd214, %rd11, 64; + .loc 1 178 20 // chunk_delta_h.py:178:20 + setp.lt.s64 %p43, %rd206, %rd3; + setp.lt.s64 %p42, %rd207, %rd3; + setp.lt.s64 %p41, %rd208, %rd3; + setp.lt.s64 %p40, %rd209, %rd3; + setp.lt.s64 %p39, %rd210, %rd3; + setp.lt.s64 %p38, %rd211, %rd3; + setp.lt.s64 %p37, %rd212, %rd3; + setp.lt.s64 %p36, %rd213, %rd3; + setp.lt.s64 %p18, %rd214, %rd3; + .loc 1 211 18 // chunk_delta_h.py:211:18 + shl.b64 %rd215, %rd206, 13; + add.s64 %rd216, %rd162, %rd215; + shl.b64 %rd217, %rd207, 13; + add.s64 %rd218, %rd162, %rd217; + shl.b64 %rd219, %rd208, 13; + add.s64 %rd220, %rd162, %rd219; + shl.b64 %rd221, %rd209, 13; + add.s64 %rd222, %rd162, %rd221; + shl.b64 %rd223, %rd210, 13; + add.s64 %rd224, %rd162, %rd223; + shl.b64 %rd225, %rd211, 13; + add.s64 %rd226, %rd162, %rd225; + shl.b64 %rd227, %rd212, 13; + add.s64 %rd228, %rd162, %rd227; + shl.b64 %rd229, %rd213, 13; + add.s64 %rd230, %rd162, %rd229; + .loc 1 211 42 // chunk_delta_h.py:211:42 + add.s64 %rd71, %rd216, %rd176; + add.s64 %rd72, %rd218, %rd176; + add.s64 %rd73, %rd220, %rd176; + add.s64 %rd74, %rd222, %rd176; + add.s64 %rd75, %rd224, %rd176; + add.s64 %rd76, %rd226, %rd176; + add.s64 %rd77, %rd228, %rd176; + add.s64 %rd78, %rd230, %rd176; + .loc 1 212 22 // chunk_delta_h.py:212:22 + bar.sync 0; + add.s32 %r124, %r72, 8192; + selp.b32 %r252, 16, 0, %p43; + selp.b32 %r133, %r252, 0, %p17; + // begin inline asm + cp.async.cg.shared.global [ %r124 + 0 ], [ %rd71 + 0 ], 0x10, %r133; + // end inline asm + add.s32 %r125, %r72, 9216; + selp.b32 %r253, 16, 0, %p42; + selp.b32 %r135, %r253, 0, %p17; + // begin inline asm + cp.async.cg.shared.global [ %r125 + 0 ], [ %rd72 + 0 ], 0x10, %r135; + // end inline asm + add.s32 %r126, %r72, 10240; + selp.b32 %r254, 16, 0, %p41; + selp.b32 %r137, %r254, 0, %p17; + // begin inline asm + cp.async.cg.shared.global [ %r126 + 0 ], [ %rd73 + 0 ], 0x10, %r137; + // end inline asm + add.s32 %r127, %r72, 11264; + selp.b32 %r255, 16, 0, %p40; + selp.b32 %r139, %r255, 0, %p17; + // begin inline asm + cp.async.cg.shared.global [ %r127 + 0 ], [ %rd74 + 0 ], 0x10, %r139; + // end inline asm + add.s32 %r128, %r72, 12288; + selp.b32 %r256, 16, 0, %p39; + selp.b32 %r141, %r256, 0, %p17; + // begin inline asm + cp.async.cg.shared.global [ %r128 + 0 ], [ %rd75 + 0 ], 0x10, %r141; + // end inline asm + add.s32 %r129, %r72, 13312; + selp.b32 %r257, 16, 0, %p38; + selp.b32 %r143, %r257, 0, %p17; + // begin inline asm + cp.async.cg.shared.global [ %r129 + 0 ], [ %rd76 + 0 ], 0x10, %r143; + // end inline asm + add.s32 %r130, %r72, 14336; + selp.b32 %r258, 16, 0, %p37; + selp.b32 %r145, %r258, 0, %p17; + // begin inline asm + cp.async.cg.shared.global [ %r130 + 0 ], [ %rd77 + 0 ], 0x10, %r145; + // end inline asm + add.s32 %r131, %r72, 15360; + selp.b32 %r259, 16, 0, %p36; + selp.b32 %r147, %r259, 0, %p17; + // begin inline asm + cp.async.cg.shared.global [ %r131 + 0 ], [ %rd78 + 0 ], 0x10, %r147; + // end inline asm + cp.async.commit_group; + .loc 1 218 46 // chunk_delta_h.py:218:46 + add.s64 %rd79, %rd71, 128; + add.s64 %rd80, %rd72, 128; + add.s64 %rd81, %rd73, 128; + add.s64 %rd82, %rd74, 128; + add.s64 %rd83, %rd75, 128; + add.s64 %rd84, %rd76, 128; + add.s64 %rd85, %rd77, 128; + add.s64 %rd86, %rd78, 128; + .loc 1 219 26 // chunk_delta_h.py:219:26 + add.s32 %r132, %r72, 32768; + // begin inline asm + cp.async.cg.shared.global [ %r132 + 0 ], [ %rd79 + 0 ], 0x10, %r133; + // end inline asm + add.s32 %r134, %r72, 33792; + // begin inline asm + cp.async.cg.shared.global [ %r134 + 0 ], [ %rd80 + 0 ], 0x10, %r135; + // end inline asm + add.s32 %r136, %r72, 34816; + // begin inline asm + cp.async.cg.shared.global [ %r136 + 0 ], [ %rd81 + 0 ], 0x10, %r137; + // end inline asm + add.s32 %r138, %r72, 35840; + // begin inline asm + cp.async.cg.shared.global [ %r138 + 0 ], [ %rd82 + 0 ], 0x10, %r139; + // end inline asm + add.s32 %r140, %r72, 36864; + // begin inline asm + cp.async.cg.shared.global [ %r140 + 0 ], [ %rd83 + 0 ], 0x10, %r141; + // end inline asm + add.s32 %r142, %r72, 37888; + // begin inline asm + cp.async.cg.shared.global [ %r142 + 0 ], [ %rd84 + 0 ], 0x10, %r143; + // end inline asm + add.s32 %r144, %r72, 38912; + // begin inline asm + cp.async.cg.shared.global [ %r144 + 0 ], [ %rd85 + 0 ], 0x10, %r145; + // end inline asm + add.s32 %r146, %r72, 39936; + // begin inline asm + cp.async.cg.shared.global [ %r146 + 0 ], [ %rd86 + 0 ], 0x10, %r147; + // end inline asm + cp.async.commit_group; + .loc 1 238 18 // chunk_delta_h.py:238:18 + add.s64 %rd231, %rd160, %rd215; + add.s64 %rd232, %rd160, %rd217; + add.s64 %rd233, %rd160, %rd219; + add.s64 %rd234, %rd160, %rd221; + add.s64 %rd235, %rd160, %rd223; + add.s64 %rd236, %rd160, %rd225; + add.s64 %rd237, %rd160, %rd227; + add.s64 %rd238, %rd160, %rd229; + .loc 1 238 42 // chunk_delta_h.py:238:42 + add.s64 %rd87, %rd231, %rd202; + add.s64 %rd88, %rd232, %rd202; + add.s64 %rd89, %rd233, %rd202; + add.s64 %rd90, %rd234, %rd202; + add.s64 %rd91, %rd235, %rd202; + add.s64 %rd92, %rd236, %rd202; + add.s64 %rd93, %rd237, %rd202; + add.s64 %rd94, %rd238, %rd202; + .loc 1 239 22 // chunk_delta_h.py:239:22 + add.s32 %r148, %r72, 57344; + // begin inline asm + cp.async.cg.shared.global [ %r148 + 0 ], [ %rd87 + 0 ], 0x10, %r133; + // end inline asm + add.s32 %r149, %r72, 58368; + // begin inline asm + cp.async.cg.shared.global [ %r149 + 0 ], [ %rd88 + 0 ], 0x10, %r135; + // end inline asm + add.s32 %r150, %r72, 59392; + // begin inline asm + cp.async.cg.shared.global [ %r150 + 0 ], [ %rd89 + 0 ], 0x10, %r137; + // end inline asm + add.s32 %r151, %r72, 60416; + // begin inline asm + cp.async.cg.shared.global [ %r151 + 0 ], [ %rd90 + 0 ], 0x10, %r139; + // end inline asm + add.s32 %r152, %r72, 61440; + // begin inline asm + cp.async.cg.shared.global [ %r152 + 0 ], [ %rd91 + 0 ], 0x10, %r141; + // end inline asm + add.s32 %r153, %r72, 62464; + // begin inline asm + cp.async.cg.shared.global [ %r153 + 0 ], [ %rd92 + 0 ], 0x10, %r143; + // end inline asm + add.s32 %r154, %r72, 63488; + // begin inline asm + cp.async.cg.shared.global [ %r154 + 0 ], [ %rd93 + 0 ], 0x10, %r145; + // end inline asm + add.s32 %r155, %r72, 64512; + // begin inline asm + cp.async.cg.shared.global [ %r155 + 0 ], [ %rd94 + 0 ], 0x10, %r147; + // end inline asm + cp.async.commit_group; + .loc 1 245 39 // chunk_delta_h.py:245:39 + min.s32 %r260, %r71, 128; + .loc 1 247 58 // chunk_delta_h.py:247:58 + shl.b32 %r261, %r260, 5; + add.s32 %r262, %r261, -32; + .loc 1 247 35 // chunk_delta_h.py:247:35 + mad.wide.s32 %rd95, %r262, 4, %rd13; + .loc 1 247 31 // chunk_delta_h.py:247:31 + selp.b32 %r157, 4, 0, %p17; + add.s32 %r156, %r246, 142084; + // begin inline asm + @%p13 cp.async.ca.shared.global [ %r156 + 0 ], [ %rd95 + 0 ], 0x4, %r157; + // end inline asm + cp.async.commit_group; + .loc 1 248 54 // chunk_delta_h.py:248:54 + shl.b64 %rd239, %rd214, 7; + add.s64 %rd96, %rd175, %rd239; + .loc 1 249 26 // chunk_delta_h.py:249:26 + add.s32 %r158, %r9, 141568; + selp.b32 %r263, 4, 0, %p18; + selp.b32 %r159, %r263, 0, %p17; + // begin inline asm + cp.async.ca.shared.global [ %r158 + 0 ], [ %rd96 + 0 ], 0x4, %r159; + // end inline asm + cp.async.commit_group; + .loc 1 290 34 // chunk_delta_h.py:290:34 + add.s64 %rd97, %rd177, %rd215; + add.s64 %rd98, %rd177, %rd217; + add.s64 %rd99, %rd177, %rd219; + add.s64 %rd100, %rd177, %rd221; + add.s64 %rd101, %rd177, %rd223; + add.s64 %rd102, %rd177, %rd225; + add.s64 %rd103, %rd177, %rd227; + add.s64 %rd104, %rd177, %rd229; + .loc 1 291 22 // chunk_delta_h.py:291:22 + add.s32 %r160, %r72, 81920; + // begin inline asm + cp.async.cg.shared.global [ %r160 + 0 ], [ %rd97 + 0 ], 0x10, %r133; + // end inline asm + add.s32 %r161, %r72, 82944; + // begin inline asm + cp.async.cg.shared.global [ %r161 + 0 ], [ %rd98 + 0 ], 0x10, %r135; + // end inline asm + add.s32 %r162, %r72, 83968; + // begin inline asm + cp.async.cg.shared.global [ %r162 + 0 ], [ %rd99 + 0 ], 0x10, %r137; + // end inline asm + add.s32 %r163, %r72, 84992; + // begin inline asm + cp.async.cg.shared.global [ %r163 + 0 ], [ %rd100 + 0 ], 0x10, %r139; + // end inline asm + add.s32 %r164, %r72, 86016; + // begin inline asm + cp.async.cg.shared.global [ %r164 + 0 ], [ %rd101 + 0 ], 0x10, %r141; + // end inline asm + add.s32 %r165, %r72, 87040; + // begin inline asm + cp.async.cg.shared.global [ %r165 + 0 ], [ %rd102 + 0 ], 0x10, %r143; + // end inline asm + add.s32 %r166, %r72, 88064; + // begin inline asm + cp.async.cg.shared.global [ %r166 + 0 ], [ %rd103 + 0 ], 0x10, %r145; + // end inline asm + add.s32 %r167, %r72, 89088; + // begin inline asm + cp.async.cg.shared.global [ %r167 + 0 ], [ %rd104 + 0 ], 0x10, %r147; + // end inline asm + cp.async.commit_group; + .loc 1 297 38 // chunk_delta_h.py:297:38 + add.s64 %rd105, %rd97, 128; + add.s64 %rd106, %rd98, 128; + add.s64 %rd107, %rd99, 128; + add.s64 %rd108, %rd100, 128; + add.s64 %rd109, %rd101, 128; + add.s64 %rd110, %rd102, 128; + add.s64 %rd111, %rd103, 128; + add.s64 %rd112, %rd104, 128; + .loc 1 298 26 // chunk_delta_h.py:298:26 + add.s32 %r168, %r72, 106496; + // begin inline asm + cp.async.cg.shared.global [ %r168 + 0 ], [ %rd105 + 0 ], 0x10, %r133; + // end inline asm + add.s32 %r169, %r72, 107520; + // begin inline asm + cp.async.cg.shared.global [ %r169 + 0 ], [ %rd106 + 0 ], 0x10, %r135; + // end inline asm + add.s32 %r170, %r72, 108544; + // begin inline asm + cp.async.cg.shared.global [ %r170 + 0 ], [ %rd107 + 0 ], 0x10, %r137; + // end inline asm + add.s32 %r171, %r72, 109568; + // begin inline asm + cp.async.cg.shared.global [ %r171 + 0 ], [ %rd108 + 0 ], 0x10, %r139; + // end inline asm + add.s32 %r172, %r72, 110592; + // begin inline asm + cp.async.cg.shared.global [ %r172 + 0 ], [ %rd109 + 0 ], 0x10, %r141; + // end inline asm + add.s32 %r173, %r72, 111616; + // begin inline asm + cp.async.cg.shared.global [ %r173 + 0 ], [ %rd110 + 0 ], 0x10, %r143; + // end inline asm + add.s32 %r174, %r72, 112640; + // begin inline asm + cp.async.cg.shared.global [ %r174 + 0 ], [ %rd111 + 0 ], 0x10, %r145; + // end inline asm + add.s32 %r175, %r72, 113664; + // begin inline asm + cp.async.cg.shared.global [ %r175 + 0 ], [ %rd112 + 0 ], 0x10, %r147; + // end inline asm + cp.async.commit_group; + .loc 1 175 21 // chunk_delta_h.py:175:21 + setp.gt.s32 %p19, %r2, 191; + .loc 1 177 31 // chunk_delta_h.py:177:31 + or.b64 %rd240, %rd4, 128; + or.b64 %rd241, %rd166, 128; + or.b64 %rd242, %rd167, 128; + or.b64 %rd243, %rd168, 128; + or.b64 %rd244, %rd169, 128; + or.b64 %rd245, %rd170, 128; + or.b64 %rd246, %rd171, 128; + or.b64 %rd247, %rd172, 128; + or.b64 %rd248, %rd11, 128; + .loc 1 178 20 // chunk_delta_h.py:178:20 + setp.lt.s64 %p35, %rd240, %rd3; + setp.lt.s64 %p34, %rd241, %rd3; + setp.lt.s64 %p33, %rd242, %rd3; + setp.lt.s64 %p32, %rd243, %rd3; + setp.lt.s64 %p31, %rd244, %rd3; + setp.lt.s64 %p30, %rd245, %rd3; + setp.lt.s64 %p29, %rd246, %rd3; + setp.lt.s64 %p28, %rd247, %rd3; + setp.lt.s64 %p20, %rd248, %rd3; + .loc 1 211 18 // chunk_delta_h.py:211:18 + shl.b64 %rd249, %rd240, 13; + add.s64 %rd250, %rd162, %rd249; + shl.b64 %rd251, %rd241, 13; + add.s64 %rd252, %rd162, %rd251; + shl.b64 %rd253, %rd242, 13; + add.s64 %rd254, %rd162, %rd253; + shl.b64 %rd255, %rd243, 13; + add.s64 %rd256, %rd162, %rd255; + shl.b64 %rd257, %rd244, 13; + add.s64 %rd258, %rd162, %rd257; + shl.b64 %rd259, %rd245, 13; + add.s64 %rd260, %rd162, %rd259; + shl.b64 %rd261, %rd246, 13; + add.s64 %rd262, %rd162, %rd261; + shl.b64 %rd263, %rd247, 13; + add.s64 %rd264, %rd162, %rd263; + .loc 1 211 42 // chunk_delta_h.py:211:42 + add.s64 %rd113, %rd250, %rd176; + add.s64 %rd114, %rd252, %rd176; + add.s64 %rd115, %rd254, %rd176; + add.s64 %rd116, %rd256, %rd176; + add.s64 %rd117, %rd258, %rd176; + add.s64 %rd118, %rd260, %rd176; + add.s64 %rd119, %rd262, %rd176; + add.s64 %rd120, %rd264, %rd176; + .loc 1 212 22 // chunk_delta_h.py:212:22 + bar.sync 0; + add.s32 %r176, %r72, 16384; + selp.b32 %r264, 16, 0, %p35; + selp.b32 %r185, %r264, 0, %p19; + // begin inline asm + cp.async.cg.shared.global [ %r176 + 0 ], [ %rd113 + 0 ], 0x10, %r185; + // end inline asm + add.s32 %r177, %r72, 17408; + selp.b32 %r265, 16, 0, %p34; + selp.b32 %r187, %r265, 0, %p19; + // begin inline asm + cp.async.cg.shared.global [ %r177 + 0 ], [ %rd114 + 0 ], 0x10, %r187; + // end inline asm + add.s32 %r178, %r72, 18432; + selp.b32 %r266, 16, 0, %p33; + selp.b32 %r189, %r266, 0, %p19; + // begin inline asm + cp.async.cg.shared.global [ %r178 + 0 ], [ %rd115 + 0 ], 0x10, %r189; + // end inline asm + add.s32 %r179, %r72, 19456; + selp.b32 %r267, 16, 0, %p32; + selp.b32 %r191, %r267, 0, %p19; + // begin inline asm + cp.async.cg.shared.global [ %r179 + 0 ], [ %rd116 + 0 ], 0x10, %r191; + // end inline asm + add.s32 %r180, %r72, 20480; + selp.b32 %r268, 16, 0, %p31; + selp.b32 %r193, %r268, 0, %p19; + // begin inline asm + cp.async.cg.shared.global [ %r180 + 0 ], [ %rd117 + 0 ], 0x10, %r193; + // end inline asm + add.s32 %r181, %r72, 21504; + selp.b32 %r269, 16, 0, %p30; + selp.b32 %r195, %r269, 0, %p19; + // begin inline asm + cp.async.cg.shared.global [ %r181 + 0 ], [ %rd118 + 0 ], 0x10, %r195; + // end inline asm + add.s32 %r182, %r72, 22528; + selp.b32 %r270, 16, 0, %p29; + selp.b32 %r197, %r270, 0, %p19; + // begin inline asm + cp.async.cg.shared.global [ %r182 + 0 ], [ %rd119 + 0 ], 0x10, %r197; + // end inline asm + add.s32 %r183, %r72, 23552; + selp.b32 %r271, 16, 0, %p28; + selp.b32 %r199, %r271, 0, %p19; + // begin inline asm + cp.async.cg.shared.global [ %r183 + 0 ], [ %rd120 + 0 ], 0x10, %r199; + // end inline asm + cp.async.commit_group; + .loc 1 218 46 // chunk_delta_h.py:218:46 + add.s64 %rd121, %rd113, 128; + add.s64 %rd122, %rd114, 128; + add.s64 %rd123, %rd115, 128; + add.s64 %rd124, %rd116, 128; + add.s64 %rd125, %rd117, 128; + add.s64 %rd126, %rd118, 128; + add.s64 %rd127, %rd119, 128; + add.s64 %rd128, %rd120, 128; + .loc 1 219 26 // chunk_delta_h.py:219:26 + add.s32 %r184, %r72, 40960; + // begin inline asm + cp.async.cg.shared.global [ %r184 + 0 ], [ %rd121 + 0 ], 0x10, %r185; + // end inline asm + add.s32 %r186, %r72, 41984; + // begin inline asm + cp.async.cg.shared.global [ %r186 + 0 ], [ %rd122 + 0 ], 0x10, %r187; + // end inline asm + add.s32 %r188, %r72, 43008; + // begin inline asm + cp.async.cg.shared.global [ %r188 + 0 ], [ %rd123 + 0 ], 0x10, %r189; + // end inline asm + add.s32 %r190, %r72, 44032; + // begin inline asm + cp.async.cg.shared.global [ %r190 + 0 ], [ %rd124 + 0 ], 0x10, %r191; + // end inline asm + add.s32 %r192, %r72, 45056; + // begin inline asm + cp.async.cg.shared.global [ %r192 + 0 ], [ %rd125 + 0 ], 0x10, %r193; + // end inline asm + add.s32 %r194, %r72, 46080; + // begin inline asm + cp.async.cg.shared.global [ %r194 + 0 ], [ %rd126 + 0 ], 0x10, %r195; + // end inline asm + add.s32 %r196, %r72, 47104; + // begin inline asm + cp.async.cg.shared.global [ %r196 + 0 ], [ %rd127 + 0 ], 0x10, %r197; + // end inline asm + add.s32 %r198, %r72, 48128; + // begin inline asm + cp.async.cg.shared.global [ %r198 + 0 ], [ %rd128 + 0 ], 0x10, %r199; + // end inline asm + cp.async.commit_group; + .loc 1 238 18 // chunk_delta_h.py:238:18 + add.s64 %rd265, %rd160, %rd249; + add.s64 %rd266, %rd160, %rd251; + add.s64 %rd267, %rd160, %rd253; + add.s64 %rd268, %rd160, %rd255; + add.s64 %rd269, %rd160, %rd257; + add.s64 %rd270, %rd160, %rd259; + add.s64 %rd271, %rd160, %rd261; + add.s64 %rd272, %rd160, %rd263; + .loc 1 238 42 // chunk_delta_h.py:238:42 + add.s64 %rd129, %rd265, %rd202; + add.s64 %rd130, %rd266, %rd202; + add.s64 %rd131, %rd267, %rd202; + add.s64 %rd132, %rd268, %rd202; + add.s64 %rd133, %rd269, %rd202; + add.s64 %rd134, %rd270, %rd202; + add.s64 %rd135, %rd271, %rd202; + add.s64 %rd136, %rd272, %rd202; + .loc 1 239 22 // chunk_delta_h.py:239:22 + add.s32 %r200, %r72, 65536; + // begin inline asm + cp.async.cg.shared.global [ %r200 + 0 ], [ %rd129 + 0 ], 0x10, %r185; + // end inline asm + add.s32 %r201, %r72, 66560; + // begin inline asm + cp.async.cg.shared.global [ %r201 + 0 ], [ %rd130 + 0 ], 0x10, %r187; + // end inline asm + add.s32 %r202, %r72, 67584; + // begin inline asm + cp.async.cg.shared.global [ %r202 + 0 ], [ %rd131 + 0 ], 0x10, %r189; + // end inline asm + add.s32 %r203, %r72, 68608; + // begin inline asm + cp.async.cg.shared.global [ %r203 + 0 ], [ %rd132 + 0 ], 0x10, %r191; + // end inline asm + add.s32 %r204, %r72, 69632; + // begin inline asm + cp.async.cg.shared.global [ %r204 + 0 ], [ %rd133 + 0 ], 0x10, %r193; + // end inline asm + add.s32 %r205, %r72, 70656; + // begin inline asm + cp.async.cg.shared.global [ %r205 + 0 ], [ %rd134 + 0 ], 0x10, %r195; + // end inline asm + add.s32 %r206, %r72, 71680; + // begin inline asm + cp.async.cg.shared.global [ %r206 + 0 ], [ %rd135 + 0 ], 0x10, %r197; + // end inline asm + add.s32 %r207, %r72, 72704; + // begin inline asm + cp.async.cg.shared.global [ %r207 + 0 ], [ %rd136 + 0 ], 0x10, %r199; + // end inline asm + cp.async.commit_group; + .loc 1 245 39 // chunk_delta_h.py:245:39 + min.s32 %r272, %r71, 192; + .loc 1 247 58 // chunk_delta_h.py:247:58 + shl.b32 %r273, %r272, 5; + add.s32 %r274, %r273, -32; + .loc 1 247 35 // chunk_delta_h.py:247:35 + mad.wide.s32 %rd137, %r274, 4, %rd13; + .loc 1 247 31 // chunk_delta_h.py:247:31 + selp.b32 %r209, 4, 0, %p19; + add.s32 %r208, %r246, 142088; + // begin inline asm + @%p13 cp.async.ca.shared.global [ %r208 + 0 ], [ %rd137 + 0 ], 0x4, %r209; + // end inline asm + cp.async.commit_group; + .loc 1 248 54 // chunk_delta_h.py:248:54 + shl.b64 %rd273, %rd248, 7; + add.s64 %rd138, %rd175, %rd273; + .loc 1 249 26 // chunk_delta_h.py:249:26 + add.s32 %r210, %r9, 141824; + selp.b32 %r275, 4, 0, %p20; + selp.b32 %r211, %r275, 0, %p19; + // begin inline asm + cp.async.ca.shared.global [ %r210 + 0 ], [ %rd138 + 0 ], 0x4, %r211; + // end inline asm + cp.async.commit_group; + .loc 1 290 34 // chunk_delta_h.py:290:34 + add.s64 %rd139, %rd177, %rd249; + add.s64 %rd140, %rd177, %rd251; + add.s64 %rd141, %rd177, %rd253; + add.s64 %rd142, %rd177, %rd255; + add.s64 %rd143, %rd177, %rd257; + add.s64 %rd144, %rd177, %rd259; + add.s64 %rd145, %rd177, %rd261; + add.s64 %rd146, %rd177, %rd263; + .loc 1 291 22 // chunk_delta_h.py:291:22 + add.s32 %r212, %r72, 90112; + // begin inline asm + cp.async.cg.shared.global [ %r212 + 0 ], [ %rd139 + 0 ], 0x10, %r185; + // end inline asm + add.s32 %r213, %r72, 91136; + // begin inline asm + cp.async.cg.shared.global [ %r213 + 0 ], [ %rd140 + 0 ], 0x10, %r187; + // end inline asm + add.s32 %r214, %r72, 92160; + // begin inline asm + cp.async.cg.shared.global [ %r214 + 0 ], [ %rd141 + 0 ], 0x10, %r189; + // end inline asm + add.s32 %r215, %r72, 93184; + // begin inline asm + cp.async.cg.shared.global [ %r215 + 0 ], [ %rd142 + 0 ], 0x10, %r191; + // end inline asm + add.s32 %r216, %r72, 94208; + // begin inline asm + cp.async.cg.shared.global [ %r216 + 0 ], [ %rd143 + 0 ], 0x10, %r193; + // end inline asm + add.s32 %r217, %r72, 95232; + // begin inline asm + cp.async.cg.shared.global [ %r217 + 0 ], [ %rd144 + 0 ], 0x10, %r195; + // end inline asm + add.s32 %r218, %r72, 96256; + // begin inline asm + cp.async.cg.shared.global [ %r218 + 0 ], [ %rd145 + 0 ], 0x10, %r197; + // end inline asm + add.s32 %r219, %r72, 97280; + // begin inline asm + cp.async.cg.shared.global [ %r219 + 0 ], [ %rd146 + 0 ], 0x10, %r199; + // end inline asm + cp.async.commit_group; + .loc 1 297 38 // chunk_delta_h.py:297:38 + add.s64 %rd147, %rd139, 128; + add.s64 %rd148, %rd140, 128; + add.s64 %rd149, %rd141, 128; + add.s64 %rd150, %rd142, 128; + add.s64 %rd151, %rd143, 128; + add.s64 %rd152, %rd144, 128; + add.s64 %rd153, %rd145, 128; + add.s64 %rd154, %rd146, 128; + .loc 1 298 26 // chunk_delta_h.py:298:26 + add.s32 %r220, %r72, 114688; + // begin inline asm + cp.async.cg.shared.global [ %r220 + 0 ], [ %rd147 + 0 ], 0x10, %r185; + // end inline asm + add.s32 %r221, %r72, 115712; + // begin inline asm + cp.async.cg.shared.global [ %r221 + 0 ], [ %rd148 + 0 ], 0x10, %r187; + // end inline asm + add.s32 %r222, %r72, 116736; + // begin inline asm + cp.async.cg.shared.global [ %r222 + 0 ], [ %rd149 + 0 ], 0x10, %r189; + // end inline asm + add.s32 %r223, %r72, 117760; + // begin inline asm + cp.async.cg.shared.global [ %r223 + 0 ], [ %rd150 + 0 ], 0x10, %r191; + // end inline asm + add.s32 %r224, %r72, 118784; + // begin inline asm + cp.async.cg.shared.global [ %r224 + 0 ], [ %rd151 + 0 ], 0x10, %r193; + // end inline asm + add.s32 %r225, %r72, 119808; + // begin inline asm + cp.async.cg.shared.global [ %r225 + 0 ], [ %rd152 + 0 ], 0x10, %r195; + // end inline asm + add.s32 %r226, %r72, 120832; + // begin inline asm + cp.async.cg.shared.global [ %r226 + 0 ], [ %rd153 + 0 ], 0x10, %r197; + // end inline asm + add.s32 %r227, %r72, 121856; + // begin inline asm + cp.async.cg.shared.global [ %r227 + 0 ], [ %rd154 + 0 ], 0x10, %r199; + // end inline asm + cp.async.commit_group; + .loc 1 175 21 // chunk_delta_h.py:175:21 + @%p14 bra $L__BB0_3; +// %bb.1: // %.lr.ph + .loc 1 0 21 // chunk_delta_h.py:0:21 + ld.param.b64 %rd28, [chunk_gated_delta_rule_fwd_kernel_h_blockdim64_param_5]; + ld.param.b64 %rd26, [chunk_gated_delta_rule_fwd_kernel_h_blockdim64_param_3]; + and.b32 %r6, %r3, 28; + bfe.u32 %r237, %r3, 2, 3; + and.b32 %r7, %r3, 32; + shr.u32 %r238, %r7, 1; + or.b32 %r239, %r238, %r237; + or.b32 %r240, %r239, 8; + or.b32 %r241, %r239, 32; + or.b32 %r242, %r239, 40; + mov.b64 %rd6, {%r240, %r239}; + mov.b64 %rd5, {%r242, %r241}; + cvt.u64.u32 %rd10, %r239; + cvt.u64.u32 %rd9, %r240; + cvt.u64.u32 %rd8, %r241; + cvt.u64.u32 %rd7, %r242; +$L__tmp3: + .loc 2 43 30 // standard.py:43:30 @[ chunk_delta_h.py:97:24 ] + shr.u32 %r276, %r2, 6; +$L__tmp4: + .loc 1 98 20 // chunk_delta_h.py:98:20 + cvt.u64.u32 %rd274, %r276; + .loc 1 175 21 // chunk_delta_h.py:175:21 + mul.lo.s64 %rd275, %rd1, %rd274; + .loc 1 177 31 // chunk_delta_h.py:177:31 + or.b64 %rd479, %rd7, 128; + or.b64 %rd480, %rd8, 128; + or.b64 %rd481, %rd9, 128; + or.b64 %rd482, %rd10, 128; + or.b64 %rd475, %rd7, 64; + or.b64 %rd476, %rd8, 64; + or.b64 %rd477, %rd9, 64; + or.b64 %rd478, %rd10, 64; + .loc 1 178 20 // chunk_delta_h.py:178:20 + mov.b64 {%r277, %r278}, %rd5; + setp.lt.u32 %p53, %r278, %r71; + setp.lt.u32 %p52, %r277, %r71; + mov.b64 {%r279, %r280}, %rd6; + setp.lt.u32 %p55, %r280, %r71; + setp.lt.u32 %p54, %r279, %r71; + add.s32 %r281, %r276, -3; + shl.b64 %rd276, %rd275, 20; + shl.b64 %rd277, %rd2, 15; + shl.b32 %r282, %r3, 7; + and.b32 %r283, %r282, 1920; + shl.b32 %r284, %r4, 4; + and.b32 %r285, %r3, 16; + shl.b32 %r286, %r7, 6; + or.b32 %r287, %r283, %r284; + xor.b32 %r288, %r287, %r285; + or.b32 %r10, %r288, %r286; + add.s32 %r289, %r246, %r10; + add.s32 %r11, %r289, 122880; + add.s32 %r12, %r289, 126976; + xor.b32 %r13, %r10, 32; + add.s32 %r290, %r246, %r13; + add.s32 %r14, %r290, 122880; + add.s32 %r15, %r290, 126976; + xor.b32 %r16, %r10, 64; + add.s32 %r291, %r246, %r16; + add.s32 %r17, %r291, 122880; + add.s32 %r18, %r291, 126976; + xor.b32 %r19, %r10, 96; + add.s32 %r292, %r246, %r19; + add.s32 %r20, %r292, 122880; + add.s32 %r21, %r292, 126976; + shl.b32 %r293, %r3, 3; + and.b32 %r294, %r293, 384; + shl.b32 %r295, %r3, 5; + and.b32 %r296, %r295, 32; + and.b32 %r297, %r3, 2; + bfe.s32 %r298, %r3, 1, 1; + and.b32 %r299, %r298, 1088; + bfe.s32 %r300, %r3, 2, 1; + and.b32 %r301, %r3, 4; + shl.b32 %r302, %r301, 2; + and.b32 %r303, %r3, 8; + bfe.s32 %r304, %r3, 3, 1; + and.b32 %r305, %r304, 544; + or.b32 %r306, %r299, %r302; + or.b32 %r307, %r294, %r296; + xor.b32 %r308, %r307, %r305; + or.b32 %r309, %r306, %r308; + add.s32 %r310, %r246, %r309; + add.s32 %r347, %r310, 131072; + xor.b32 %r311, %r309, 64; + add.s32 %r312, %r246, %r311; + add.s32 %r352, %r312, 131072; + shl.b32 %r313, %r3, 6; + and.b32 %r314, %r313, 1088; + shl.b32 %r315, %r5, 2; + shl.b32 %r316, %r297, 3; + and.b32 %r317, %r300, 544; + xor.b32 %r318, %r314, %r315; + xor.b32 %r319, %r318, %r317; + or.b32 %r320, %r319, %r316; + add.s32 %r321, %r246, %r320; + add.s32 %r22, %r321, 131072; + add.s32 %r23, %r321, 131328; + add.s32 %r24, %r289, 131072; + add.s32 %r25, %r289, 135168; + add.s32 %r26, %r290, 131072; + add.s32 %r27, %r290, 135168; + add.s32 %r28, %r291, 131072; + add.s32 %r29, %r291, 135168; + add.s32 %r30, %r292, 131072; + add.s32 %r31, %r292, 135168; + add.s32 %r413, %r310, 139264; + add.s32 %r418, %r312, 139264; + add.s32 %r32, %r321, 139264; + add.s32 %r33, %r321, 139520; + and.b32 %r322, %r282, 3968; + or.b32 %r323, %r284, %r322; + add.s32 %r324, %r246, %r323; + add.s32 %r34, %r324, 122880; + add.s32 %r35, %r324, 126976; + xor.b32 %r325, %r323, 16; + add.s32 %r326, %r246, %r325; + add.s32 %r36, %r326, 122880; + add.s32 %r37, %r326, 126976; + xor.b32 %r327, %r323, 32; + add.s32 %r328, %r246, %r327; + add.s32 %r38, %r328, 122880; + add.s32 %r39, %r328, 126976; + xor.b32 %r329, %r323, 48; + add.s32 %r330, %r246, %r329; + add.s32 %r40, %r330, 122880; + add.s32 %r41, %r330, 126976; + xor.b32 %r331, %r323, 64; + add.s32 %r332, %r246, %r331; + add.s32 %r42, %r332, 122880; + add.s32 %r43, %r332, 126976; + xor.b32 %r333, %r323, 80; + add.s32 %r334, %r246, %r333; + add.s32 %r44, %r334, 122880; + add.s32 %r45, %r334, 126976; + xor.b32 %r335, %r323, 96; + add.s32 %r336, %r246, %r335; + add.s32 %r46, %r336, 122880; + add.s32 %r47, %r336, 126976; + xor.b32 %r337, %r323, 112; + add.s32 %r338, %r246, %r337; + add.s32 %r48, %r338, 122880; + add.s32 %r49, %r338, 126976; + add.s32 %r50, %r324, 131072; + add.s32 %r51, %r324, 135168; + add.s32 %r52, %r326, 131072; + add.s32 %r53, %r326, 135168; + add.s32 %r54, %r328, 131072; + add.s32 %r55, %r328, 135168; + add.s32 %r56, %r330, 131072; + add.s32 %r57, %r330, 135168; + add.s32 %r58, %r332, 131072; + add.s32 %r59, %r332, 135168; + add.s32 %r60, %r334, 131072; + add.s32 %r61, %r334, 135168; + add.s32 %r62, %r336, 131072; + add.s32 %r63, %r336, 135168; + add.s32 %r64, %r338, 131072; + add.s32 %r65, %r338, 135168; + add.s32 %r735, %r310, 122880; + add.s32 %r740, %r312, 122880; + add.s32 %r66, %r321, 122880; + add.s32 %r67, %r321, 123136; + shl.b32 %r339, %r7, 1; + .loc 1 175 21 // chunk_delta_h.py:175:21 + add.s32 %r340, %r246, %r339; + add.s32 %r68, %r340, %r6; + shl.b32 %r341, %r303, 1; + shl.b32 %r342, %r285, 6; + mul.lo.s32 %r343, %r4, 144; + or.b32 %r344, %r341, %r7; + xor.b32 %r345, %r343, %r344; + or.b32 %r69, %r345, %r342; + xor.b32 %r70, %r69, 64; + cvt.s64.s32 %rd14, %r281; + cvt.u64.u32 %rd15, %r71; + mul.wide.u32 %rd16, %r276, 64; + add.s64 %rd278, %rd203, %rd12; + add.s64 %rd279, %rd278, %rd205; + add.s64 %rd280, %rd279, %rd27; + add.s64 %rd472, %rd280, 24576; + or.b64 %rd17, %rd11, 192; + mul.wide.u32 %rd18, %r4, 16; + shl.b64 %rd281, %rd156, 13; + shl.b64 %rd282, %rd4, 13; + add.s64 %rd283, %rd281, %rd282; + shl.b64 %rd284, %rd2, 8; + or.b64 %rd285, %rd283, %rd284; + add.s64 %rd471, %rd23, %rd285; + and.b32 %r346, %r1, 1; + mul.wide.u32 %rd286, %r346, 128; + or.b64 %rd287, %rd285, %rd286; + add.s64 %rd470, %rd24, %rd287; + add.s64 %rd469, %rd25, %rd285; + add.s64 %rd468, %rd26, %rd287; + or.b64 %rd288, %rd276, %rd277; + shl.b64 %rd289, %rd4, 8; + or.b64 %rd290, %rd288, %rd289; + or.b64 %rd291, %rd290, %rd286; + add.s64 %rd467, %rd28, %rd291; + mov.b32 %r969, 0f00000000; + mov.b32 %r1419, 2; + mov.b32 %r1418, -1; + mov.b64 %rd473, 0; + mov.b64 %rd474, %rd473; + mov.b32 %r970, %r969; + mov.b32 %r971, %r969; + mov.b32 %r972, %r969; + mov.b32 %r977, %r969; + mov.b32 %r978, %r969; + mov.b32 %r979, %r969; + mov.b32 %r980, %r969; + mov.b32 %r981, %r969; + mov.b32 %r982, %r969; + mov.b32 %r983, %r969; + mov.b32 %r984, %r969; + mov.b32 %r985, %r969; + mov.b32 %r986, %r969; + mov.b32 %r987, %r969; + mov.b32 %r988, %r969; + mov.b32 %r989, %r969; + mov.b32 %r990, %r969; + mov.b32 %r991, %r969; + mov.b32 %r992, %r969; + mov.b32 %r993, %r969; + mov.b32 %r994, %r969; + mov.b32 %r995, %r969; + mov.b32 %r996, %r969; + mov.b32 %r997, %r969; + mov.b32 %r998, %r969; + mov.b32 %r999, %r969; + mov.b32 %r1000, %r969; + mov.b32 %r1001, %r969; + mov.b32 %r1002, %r969; + mov.b32 %r1003, %r969; + mov.b32 %r1004, %r969; + mov.b32 %r1005, %r969; + mov.b32 %r1006, %r969; + mov.b32 %r1007, %r969; + mov.b32 %r1008, %r969; + mov.b32 %r1013, %r969; + mov.b32 %r1014, %r969; + mov.b32 %r1015, %r969; + mov.b32 %r1016, %r969; + mov.b32 %r1017, %r969; + mov.b32 %r1018, %r969; + mov.b32 %r1019, %r969; + mov.b32 %r1020, %r969; + mov.b32 %r1021, %r969; + mov.b32 %r1022, %r969; + mov.b32 %r1023, %r969; + mov.b32 %r1024, %r969; + mov.b32 %r1025, %r969; + mov.b32 %r1026, %r969; + mov.b32 %r1027, %r969; + mov.b32 %r1028, %r969; + mov.b32 %r1029, %r969; + mov.b32 %r1030, %r969; + mov.b32 %r1031, %r969; + mov.b32 %r1032, %r969; + mov.b32 %r1033, %r969; + mov.b32 %r1034, %r969; + mov.b32 %r1035, %r969; + mov.b32 %r1036, %r969; + mov.b32 %r1037, %r969; + mov.b32 %r1038, %r969; + mov.b32 %r1039, %r969; + mov.b32 %r1040, %r969; + mov.b32 %r825, %r969; + mov.b32 %r826, %r969; + mov.b32 %r827, %r969; + mov.b32 %r828, %r969; + mov.b32 %r835, %r969; + mov.b32 %r836, %r969; + mov.b32 %r837, %r969; + mov.b32 %r838, %r969; + mov.b32 %r841, %r969; + mov.b32 %r842, %r969; + mov.b32 %r843, %r969; + mov.b32 %r844, %r969; + mov.b32 %r847, %r969; + mov.b32 %r848, %r969; + mov.b32 %r849, %r969; + mov.b32 %r850, %r969; + mov.b32 %r853, %r969; + mov.b32 %r854, %r969; + mov.b32 %r855, %r969; + mov.b32 %r856, %r969; + mov.b32 %r859, %r969; + mov.b32 %r860, %r969; + mov.b32 %r861, %r969; + mov.b32 %r862, %r969; + mov.b32 %r865, %r969; + mov.b32 %r866, %r969; + mov.b32 %r867, %r969; + mov.b32 %r868, %r969; + mov.b32 %r871, %r969; + mov.b32 %r872, %r969; + mov.b32 %r873, %r969; + mov.b32 %r874, %r969; + mov.b32 %r877, %r969; + mov.b32 %r878, %r969; + mov.b32 %r879, %r969; + mov.b32 %r880, %r969; + mov.b32 %r885, %r969; + mov.b32 %r886, %r969; + mov.b32 %r887, %r969; + mov.b32 %r888, %r969; + mov.b32 %r889, %r969; + mov.b32 %r890, %r969; + mov.b32 %r891, %r969; + mov.b32 %r892, %r969; + mov.b32 %r893, %r969; + mov.b32 %r894, %r969; + mov.b32 %r895, %r969; + mov.b32 %r896, %r969; + mov.b32 %r897, %r969; + mov.b32 %r898, %r969; + mov.b32 %r899, %r969; + mov.b32 %r900, %r969; + mov.b32 %r901, %r969; + mov.b32 %r902, %r969; + mov.b32 %r903, %r969; + mov.b32 %r904, %r969; + mov.b32 %r905, %r969; + mov.b32 %r906, %r969; + mov.b32 %r907, %r969; + mov.b32 %r908, %r969; + mov.b32 %r909, %r969; + mov.b32 %r910, %r969; + mov.b32 %r911, %r969; + mov.b32 %r912, %r969; +$L__BB0_2: // %__nv_exp2f.exit + // =>This Inner Loop Header: Depth=1 + .loc 1 0 21 // chunk_delta_h.py:0:21 + mov.pred %p8, %p51; + mov.pred %p7, %p50; + mov.pred %p6, %p49; + mov.pred %p5, %p48; + mov.pred %p4, %p47; + mov.pred %p3, %p46; + mov.pred %p2, %p45; + mov.pred %p1, %p44; + mov.pred %p51, %p43; + mov.pred %p50, %p42; + mov.pred %p49, %p41; + mov.pred %p48, %p40; + mov.pred %p47, %p39; + mov.pred %p46, %p38; + mov.pred %p45, %p37; + mov.pred %p44, %p36; + mov.pred %p43, %p35; + mov.pred %p42, %p34; + mov.pred %p41, %p33; + mov.pred %p40, %p32; + mov.pred %p39, %p31; + mov.pred %p38, %p30; + mov.pred %p37, %p29; + mov.pred %p36, %p28; + cvt.u32.u64 %r1109, %rd11; + .loc 1 247 31 // chunk_delta_h.py:247:31 + setp.eq.b32 %p22, %r1109, 0; + .loc 1 178 20 // chunk_delta_h.py:178:20 + setp.lt.s64 %p9, %rd475, %rd15; + setp.lt.s64 %p10, %rd476, %rd15; + setp.lt.s64 %p11, %rd477, %rd15; + setp.lt.s64 %p12, %rd478, %rd15; + .loc 1 175 21 // chunk_delta_h.py:175:21 + setp.lt.s64 %p23, %rd474, %rd14; + add.s32 %r1110, %r1418, 1; + setp.gt.s32 %p24, %r1110, 2; + selp.b32 %r1418, 0, %r1110, %p24; + .loc 1 183 64 // chunk_delta_h.py:183:64 + add.s64 %rd292, %rd467, %rd18; + add.s64 %rd293, %rd292, 2048; + add.s64 %rd294, %rd292, 4096; + add.s64 %rd295, %rd292, 6144; + add.s64 %rd296, %rd292, 8192; + add.s64 %rd297, %rd292, 10240; + add.s64 %rd298, %rd292, 12288; + .loc 1 185 31 // chunk_delta_h.py:185:31 + add.s64 %rd299, %rd292, 14336; + cvt.rn.bf16x2.f32 %r348, %r826, %r825; + cvt.rn.bf16x2.f32 %r357, %r828, %r827; + cvt.rn.bf16x2.f32 %r349, %r836, %r835; + cvt.rn.bf16x2.f32 %r358, %r838, %r837; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r11], {%r348, %r357, %r349, %r358}; + cvt.rn.bf16x2.f32 %r365, %r878, %r877; + cvt.rn.bf16x2.f32 %r373, %r880, %r879; + cvt.rn.bf16x2.f32 %r366, %r886, %r885; + cvt.rn.bf16x2.f32 %r374, %r888, %r887; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r12], {%r365, %r373, %r366, %r374}; + cvt.rn.bf16x2.f32 %r350, %r842, %r841; + cvt.rn.bf16x2.f32 %r359, %r844, %r843; + cvt.rn.bf16x2.f32 %r351, %r848, %r847; + cvt.rn.bf16x2.f32 %r360, %r850, %r849; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r14], {%r350, %r359, %r351, %r360}; + cvt.rn.bf16x2.f32 %r367, %r890, %r889; + cvt.rn.bf16x2.f32 %r375, %r892, %r891; + cvt.rn.bf16x2.f32 %r368, %r894, %r893; + cvt.rn.bf16x2.f32 %r376, %r896, %r895; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r15], {%r367, %r375, %r368, %r376}; + cvt.rn.bf16x2.f32 %r353, %r854, %r853; + cvt.rn.bf16x2.f32 %r361, %r856, %r855; + cvt.rn.bf16x2.f32 %r354, %r860, %r859; + cvt.rn.bf16x2.f32 %r362, %r862, %r861; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r17], {%r353, %r361, %r354, %r362}; + cvt.rn.bf16x2.f32 %r369, %r898, %r897; + cvt.rn.bf16x2.f32 %r377, %r900, %r899; + cvt.rn.bf16x2.f32 %r370, %r902, %r901; + cvt.rn.bf16x2.f32 %r378, %r904, %r903; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r18], {%r369, %r377, %r370, %r378}; + cvt.rn.bf16x2.f32 %r355, %r866, %r865; + cvt.rn.bf16x2.f32 %r363, %r868, %r867; + cvt.rn.bf16x2.f32 %r356, %r872, %r871; + cvt.rn.bf16x2.f32 %r364, %r874, %r873; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r20], {%r355, %r363, %r356, %r364}; + cvt.rn.bf16x2.f32 %r371, %r906, %r905; + cvt.rn.bf16x2.f32 %r379, %r908, %r907; + cvt.rn.bf16x2.f32 %r372, %r910, %r909; + cvt.rn.bf16x2.f32 %r380, %r912, %r911; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r21], {%r371, %r379, %r372, %r380}; + .loc 1 185 23 // chunk_delta_h.py:185:23 + // begin inline asm + st.shared.v4.b32 [ %r347 + 0 ], { %r348, %r349, %r350, %r351 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r352 + 0 ], { %r353, %r354, %r355, %r356 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r381, %r382, %r383, %r384}, [%r22]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r389, %r390, %r391, %r392}, [%r23]; + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r347 + 0 ], { %r357, %r358, %r359, %r360 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r352 + 0 ], { %r361, %r362, %r363, %r364 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r385, %r386, %r387, %r388}, [%r22]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r393, %r394, %r395, %r396}, [%r23]; + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r347 + 0 ], { %r365, %r366, %r367, %r368 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r352 + 0 ], { %r369, %r370, %r371, %r372 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r397, %r398, %r399, %r400}, [%r22]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r405, %r406, %r407, %r408}, [%r23]; + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r347 + 0 ], { %r373, %r374, %r375, %r376 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r352 + 0 ], { %r377, %r378, %r379, %r380 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r401, %r402, %r403, %r404}, [%r22]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r409, %r410, %r411, %r412}, [%r23]; + mov.pred %p21, -1; + // begin inline asm + @%p21 st.global.v4.b32 [ %rd292 + 0 ], { %r381, %r382, %r383, %r384 }; + // end inline asm + // begin inline asm + @%p21 st.global.v4.b32 [ %rd293 + 0 ], { %r385, %r386, %r387, %r388 }; + // end inline asm + // begin inline asm + @%p21 st.global.v4.b32 [ %rd294 + 0 ], { %r389, %r390, %r391, %r392 }; + // end inline asm + // begin inline asm + @%p21 st.global.v4.b32 [ %rd295 + 0 ], { %r393, %r394, %r395, %r396 }; + // end inline asm + // begin inline asm + @%p21 st.global.v4.b32 [ %rd296 + 0 ], { %r397, %r398, %r399, %r400 }; + // end inline asm + // begin inline asm + @%p21 st.global.v4.b32 [ %rd297 + 0 ], { %r401, %r402, %r403, %r404 }; + // end inline asm + // begin inline asm + @%p21 st.global.v4.b32 [ %rd298 + 0 ], { %r405, %r406, %r407, %r408 }; + // end inline asm + // begin inline asm + @%p21 st.global.v4.b32 [ %rd299 + 0 ], { %r409, %r410, %r411, %r412 }; + // end inline asm + .loc 1 191 68 // chunk_delta_h.py:191:68 + add.s64 %rd300, %rd292, 16384; + add.s64 %rd301, %rd292, 18432; + add.s64 %rd302, %rd292, 20480; + add.s64 %rd303, %rd292, 22528; + add.s64 %rd304, %rd292, 24576; + add.s64 %rd305, %rd292, 26624; + add.s64 %rd306, %rd292, 28672; + .loc 1 193 35 // chunk_delta_h.py:193:35 + add.s64 %rd307, %rd292, 30720; + bar.sync 0; + cvt.rn.bf16x2.f32 %r414, %r970, %r969; + cvt.rn.bf16x2.f32 %r423, %r972, %r971; + cvt.rn.bf16x2.f32 %r415, %r978, %r977; + cvt.rn.bf16x2.f32 %r424, %r980, %r979; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r24], {%r414, %r423, %r415, %r424}; + cvt.rn.bf16x2.f32 %r431, %r1006, %r1005; + cvt.rn.bf16x2.f32 %r439, %r1008, %r1007; + cvt.rn.bf16x2.f32 %r432, %r1014, %r1013; + cvt.rn.bf16x2.f32 %r440, %r1016, %r1015; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r25], {%r431, %r439, %r432, %r440}; + cvt.rn.bf16x2.f32 %r416, %r982, %r981; + cvt.rn.bf16x2.f32 %r425, %r984, %r983; + cvt.rn.bf16x2.f32 %r417, %r986, %r985; + cvt.rn.bf16x2.f32 %r426, %r988, %r987; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r26], {%r416, %r425, %r417, %r426}; + cvt.rn.bf16x2.f32 %r433, %r1018, %r1017; + cvt.rn.bf16x2.f32 %r441, %r1020, %r1019; + cvt.rn.bf16x2.f32 %r434, %r1022, %r1021; + cvt.rn.bf16x2.f32 %r442, %r1024, %r1023; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r27], {%r433, %r441, %r434, %r442}; + cvt.rn.bf16x2.f32 %r419, %r990, %r989; + cvt.rn.bf16x2.f32 %r427, %r992, %r991; + cvt.rn.bf16x2.f32 %r420, %r994, %r993; + cvt.rn.bf16x2.f32 %r428, %r996, %r995; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r28], {%r419, %r427, %r420, %r428}; + cvt.rn.bf16x2.f32 %r435, %r1026, %r1025; + cvt.rn.bf16x2.f32 %r443, %r1028, %r1027; + cvt.rn.bf16x2.f32 %r436, %r1030, %r1029; + cvt.rn.bf16x2.f32 %r444, %r1032, %r1031; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r29], {%r435, %r443, %r436, %r444}; + cvt.rn.bf16x2.f32 %r421, %r998, %r997; + cvt.rn.bf16x2.f32 %r429, %r1000, %r999; + cvt.rn.bf16x2.f32 %r422, %r1002, %r1001; + cvt.rn.bf16x2.f32 %r430, %r1004, %r1003; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r30], {%r421, %r429, %r422, %r430}; + cvt.rn.bf16x2.f32 %r437, %r1034, %r1033; + cvt.rn.bf16x2.f32 %r445, %r1036, %r1035; + cvt.rn.bf16x2.f32 %r438, %r1038, %r1037; + cvt.rn.bf16x2.f32 %r446, %r1040, %r1039; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r31], {%r437, %r445, %r438, %r446}; + .loc 1 193 27 // chunk_delta_h.py:193:27 + // begin inline asm + st.shared.v4.b32 [ %r413 + 0 ], { %r414, %r415, %r416, %r417 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r418 + 0 ], { %r419, %r420, %r421, %r422 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r447, %r448, %r449, %r450}, [%r32]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r455, %r456, %r457, %r458}, [%r33]; + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r413 + 0 ], { %r423, %r424, %r425, %r426 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r418 + 0 ], { %r427, %r428, %r429, %r430 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r451, %r452, %r453, %r454}, [%r32]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r459, %r460, %r461, %r462}, [%r33]; + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r413 + 0 ], { %r431, %r432, %r433, %r434 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r418 + 0 ], { %r435, %r436, %r437, %r438 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r463, %r464, %r465, %r466}, [%r32]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r471, %r472, %r473, %r474}, [%r33]; + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r413 + 0 ], { %r439, %r440, %r441, %r442 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r418 + 0 ], { %r443, %r444, %r445, %r446 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r467, %r468, %r469, %r470}, [%r32]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r475, %r476, %r477, %r478}, [%r33]; + // begin inline asm + @%p21 st.global.v4.b32 [ %rd300 + 0 ], { %r447, %r448, %r449, %r450 }; + // end inline asm + // begin inline asm + @%p21 st.global.v4.b32 [ %rd301 + 0 ], { %r451, %r452, %r453, %r454 }; + // end inline asm + // begin inline asm + @%p21 st.global.v4.b32 [ %rd302 + 0 ], { %r455, %r456, %r457, %r458 }; + // end inline asm + // begin inline asm + @%p21 st.global.v4.b32 [ %rd303 + 0 ], { %r459, %r460, %r461, %r462 }; + // end inline asm + // begin inline asm + @%p21 st.global.v4.b32 [ %rd304 + 0 ], { %r463, %r464, %r465, %r466 }; + // end inline asm + // begin inline asm + @%p21 st.global.v4.b32 [ %rd305 + 0 ], { %r467, %r468, %r469, %r470 }; + // end inline asm + // begin inline asm + @%p21 st.global.v4.b32 [ %rd306 + 0 ], { %r471, %r472, %r473, %r474 }; + // end inline asm + // begin inline asm + @%p21 st.global.v4.b32 [ %rd307 + 0 ], { %r475, %r476, %r477, %r478 }; + // end inline asm + .loc 1 212 22 // chunk_delta_h.py:212:22 + cp.async.wait_group 14; + bar.sync 0; + shl.b32 %r1111, %r1418, 13; + add.s32 %r1112, %r246, %r1111; + add.s32 %r1113, %r1112, %r10; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r479, %r480, %r481, %r482}, [%r1113]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r499, %r500, %r501, %r502}, [%r1113+4096]; + add.s32 %r1114, %r1112, %r13; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r507, %r508, %r509, %r510}, [%r1114]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r559, %r560, %r561, %r562}, [%r1114+4096]; + add.s32 %r1115, %r1112, %r16; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r591, %r592, %r593, %r594}, [%r1115]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r611, %r612, %r613, %r614}, [%r1115+4096]; + add.s32 %r1116, %r1112, %r19; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r615, %r616, %r617, %r618}, [%r1116]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r635, %r636, %r637, %r638}, [%r1116+4096]; + .loc 1 185 31 // chunk_delta_h.py:185:31 + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r483, %r484, %r511, %r512}, [%r34]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r595, %r596, %r619, %r620}, [%r35]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r485, %r486, %r517, %r518}, [%r36]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r597, %r598, %r621, %r622}, [%r37]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r487, %r488, %r523, %r524}, [%r38]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r599, %r600, %r623, %r624}, [%r39]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r489, %r490, %r529, %r530}, [%r40]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r601, %r602, %r625, %r626}, [%r41]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r491, %r492, %r535, %r536}, [%r42]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r603, %r604, %r627, %r628}, [%r43]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r493, %r494, %r541, %r542}, [%r44]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r605, %r606, %r629, %r630}, [%r45]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r495, %r496, %r547, %r548}, [%r46]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r607, %r608, %r631, %r632}, [%r47]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r497, %r498, %r553, %r554}, [%r48]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r609, %r610, %r633, %r634}, [%r49]; + mov.b32 %r590, 0; + .loc 1 216 30 // chunk_delta_h.py:216:30 + mov.b32 %r503, %r590; + mov.b32 %r504, %r590; + mov.b32 %r505, %r590; + mov.b32 %r506, %r590; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r503, %r504, %r505, %r506 }, { %r479, %r480, %r481, %r482 }, { %r483, %r484 }, { %r503, %r504, %r505, %r506 }; + // end inline asm + mov.b32 %r513, %r590; + mov.b32 %r514, %r590; + mov.b32 %r515, %r590; + mov.b32 %r516, %r590; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r513, %r514, %r515, %r516 }, { %r479, %r480, %r481, %r482 }, { %r485, %r486 }, { %r513, %r514, %r515, %r516 }; + // end inline asm + mov.b32 %r519, %r590; + mov.b32 %r520, %r590; + mov.b32 %r521, %r590; + mov.b32 %r522, %r590; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r519, %r520, %r521, %r522 }, { %r479, %r480, %r481, %r482 }, { %r487, %r488 }, { %r519, %r520, %r521, %r522 }; + // end inline asm + mov.b32 %r525, %r590; + mov.b32 %r526, %r590; + mov.b32 %r527, %r590; + mov.b32 %r528, %r590; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r525, %r526, %r527, %r528 }, { %r479, %r480, %r481, %r482 }, { %r489, %r490 }, { %r525, %r526, %r527, %r528 }; + // end inline asm + mov.b32 %r531, %r590; + mov.b32 %r532, %r590; + mov.b32 %r533, %r590; + mov.b32 %r534, %r590; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r531, %r532, %r533, %r534 }, { %r479, %r480, %r481, %r482 }, { %r491, %r492 }, { %r531, %r532, %r533, %r534 }; + // end inline asm + mov.b32 %r537, %r590; + mov.b32 %r538, %r590; + mov.b32 %r539, %r590; + mov.b32 %r540, %r590; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r537, %r538, %r539, %r540 }, { %r479, %r480, %r481, %r482 }, { %r493, %r494 }, { %r537, %r538, %r539, %r540 }; + // end inline asm + mov.b32 %r543, %r590; + mov.b32 %r544, %r590; + mov.b32 %r545, %r590; + mov.b32 %r546, %r590; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r543, %r544, %r545, %r546 }, { %r479, %r480, %r481, %r482 }, { %r495, %r496 }, { %r543, %r544, %r545, %r546 }; + // end inline asm + mov.b32 %r549, %r590; + mov.b32 %r550, %r590; + mov.b32 %r551, %r590; + mov.b32 %r552, %r590; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r549, %r550, %r551, %r552 }, { %r479, %r480, %r481, %r482 }, { %r497, %r498 }, { %r549, %r550, %r551, %r552 }; + // end inline asm + mov.b32 %r555, %r590; + mov.b32 %r556, %r590; + mov.b32 %r557, %r590; + mov.b32 %r558, %r590; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r555, %r556, %r557, %r558 }, { %r499, %r500, %r501, %r502 }, { %r483, %r484 }, { %r555, %r556, %r557, %r558 }; + // end inline asm + mov.b32 %r563, %r590; + mov.b32 %r564, %r590; + mov.b32 %r565, %r590; + mov.b32 %r566, %r590; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r563, %r564, %r565, %r566 }, { %r499, %r500, %r501, %r502 }, { %r485, %r486 }, { %r563, %r564, %r565, %r566 }; + // end inline asm + mov.b32 %r567, %r590; + mov.b32 %r568, %r590; + mov.b32 %r569, %r590; + mov.b32 %r570, %r590; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r567, %r568, %r569, %r570 }, { %r499, %r500, %r501, %r502 }, { %r487, %r488 }, { %r567, %r568, %r569, %r570 }; + // end inline asm + mov.b32 %r571, %r590; + mov.b32 %r572, %r590; + mov.b32 %r573, %r590; + mov.b32 %r574, %r590; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r571, %r572, %r573, %r574 }, { %r499, %r500, %r501, %r502 }, { %r489, %r490 }, { %r571, %r572, %r573, %r574 }; + // end inline asm + mov.b32 %r575, %r590; + mov.b32 %r576, %r590; + mov.b32 %r577, %r590; + mov.b32 %r578, %r590; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r575, %r576, %r577, %r578 }, { %r499, %r500, %r501, %r502 }, { %r491, %r492 }, { %r575, %r576, %r577, %r578 }; + // end inline asm + mov.b32 %r579, %r590; + mov.b32 %r580, %r590; + mov.b32 %r581, %r590; + mov.b32 %r582, %r590; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r579, %r580, %r581, %r582 }, { %r499, %r500, %r501, %r502 }, { %r493, %r494 }, { %r579, %r580, %r581, %r582 }; + // end inline asm + mov.b32 %r583, %r590; + mov.b32 %r584, %r590; + mov.b32 %r585, %r590; + mov.b32 %r586, %r590; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r583, %r584, %r585, %r586 }, { %r499, %r500, %r501, %r502 }, { %r495, %r496 }, { %r583, %r584, %r585, %r586 }; + // end inline asm + mov.b32 %r587, %r590; + mov.b32 %r588, %r590; + mov.b32 %r589, %r590; + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r587, %r588, %r589, %r590 }, { %r499, %r500, %r501, %r502 }, { %r497, %r498 }, { %r587, %r588, %r589, %r590 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r503, %r504, %r505, %r506 }, { %r507, %r508, %r509, %r510 }, { %r511, %r512 }, { %r503, %r504, %r505, %r506 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r513, %r514, %r515, %r516 }, { %r507, %r508, %r509, %r510 }, { %r517, %r518 }, { %r513, %r514, %r515, %r516 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r519, %r520, %r521, %r522 }, { %r507, %r508, %r509, %r510 }, { %r523, %r524 }, { %r519, %r520, %r521, %r522 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r525, %r526, %r527, %r528 }, { %r507, %r508, %r509, %r510 }, { %r529, %r530 }, { %r525, %r526, %r527, %r528 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r531, %r532, %r533, %r534 }, { %r507, %r508, %r509, %r510 }, { %r535, %r536 }, { %r531, %r532, %r533, %r534 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r537, %r538, %r539, %r540 }, { %r507, %r508, %r509, %r510 }, { %r541, %r542 }, { %r537, %r538, %r539, %r540 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r543, %r544, %r545, %r546 }, { %r507, %r508, %r509, %r510 }, { %r547, %r548 }, { %r543, %r544, %r545, %r546 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r549, %r550, %r551, %r552 }, { %r507, %r508, %r509, %r510 }, { %r553, %r554 }, { %r549, %r550, %r551, %r552 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r555, %r556, %r557, %r558 }, { %r559, %r560, %r561, %r562 }, { %r511, %r512 }, { %r555, %r556, %r557, %r558 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r563, %r564, %r565, %r566 }, { %r559, %r560, %r561, %r562 }, { %r517, %r518 }, { %r563, %r564, %r565, %r566 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r567, %r568, %r569, %r570 }, { %r559, %r560, %r561, %r562 }, { %r523, %r524 }, { %r567, %r568, %r569, %r570 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r571, %r572, %r573, %r574 }, { %r559, %r560, %r561, %r562 }, { %r529, %r530 }, { %r571, %r572, %r573, %r574 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r575, %r576, %r577, %r578 }, { %r559, %r560, %r561, %r562 }, { %r535, %r536 }, { %r575, %r576, %r577, %r578 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r579, %r580, %r581, %r582 }, { %r559, %r560, %r561, %r562 }, { %r541, %r542 }, { %r579, %r580, %r581, %r582 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r583, %r584, %r585, %r586 }, { %r559, %r560, %r561, %r562 }, { %r547, %r548 }, { %r583, %r584, %r585, %r586 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r587, %r588, %r589, %r590 }, { %r559, %r560, %r561, %r562 }, { %r553, %r554 }, { %r587, %r588, %r589, %r590 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r503, %r504, %r505, %r506 }, { %r591, %r592, %r593, %r594 }, { %r595, %r596 }, { %r503, %r504, %r505, %r506 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r513, %r514, %r515, %r516 }, { %r591, %r592, %r593, %r594 }, { %r597, %r598 }, { %r513, %r514, %r515, %r516 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r519, %r520, %r521, %r522 }, { %r591, %r592, %r593, %r594 }, { %r599, %r600 }, { %r519, %r520, %r521, %r522 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r525, %r526, %r527, %r528 }, { %r591, %r592, %r593, %r594 }, { %r601, %r602 }, { %r525, %r526, %r527, %r528 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r531, %r532, %r533, %r534 }, { %r591, %r592, %r593, %r594 }, { %r603, %r604 }, { %r531, %r532, %r533, %r534 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r537, %r538, %r539, %r540 }, { %r591, %r592, %r593, %r594 }, { %r605, %r606 }, { %r537, %r538, %r539, %r540 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r543, %r544, %r545, %r546 }, { %r591, %r592, %r593, %r594 }, { %r607, %r608 }, { %r543, %r544, %r545, %r546 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r549, %r550, %r551, %r552 }, { %r591, %r592, %r593, %r594 }, { %r609, %r610 }, { %r549, %r550, %r551, %r552 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r555, %r556, %r557, %r558 }, { %r611, %r612, %r613, %r614 }, { %r595, %r596 }, { %r555, %r556, %r557, %r558 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r563, %r564, %r565, %r566 }, { %r611, %r612, %r613, %r614 }, { %r597, %r598 }, { %r563, %r564, %r565, %r566 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r567, %r568, %r569, %r570 }, { %r611, %r612, %r613, %r614 }, { %r599, %r600 }, { %r567, %r568, %r569, %r570 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r571, %r572, %r573, %r574 }, { %r611, %r612, %r613, %r614 }, { %r601, %r602 }, { %r571, %r572, %r573, %r574 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r575, %r576, %r577, %r578 }, { %r611, %r612, %r613, %r614 }, { %r603, %r604 }, { %r575, %r576, %r577, %r578 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r579, %r580, %r581, %r582 }, { %r611, %r612, %r613, %r614 }, { %r605, %r606 }, { %r579, %r580, %r581, %r582 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r583, %r584, %r585, %r586 }, { %r611, %r612, %r613, %r614 }, { %r607, %r608 }, { %r583, %r584, %r585, %r586 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r587, %r588, %r589, %r590 }, { %r611, %r612, %r613, %r614 }, { %r609, %r610 }, { %r587, %r588, %r589, %r590 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r503, %r504, %r505, %r506 }, { %r615, %r616, %r617, %r618 }, { %r619, %r620 }, { %r503, %r504, %r505, %r506 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r513, %r514, %r515, %r516 }, { %r615, %r616, %r617, %r618 }, { %r621, %r622 }, { %r513, %r514, %r515, %r516 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r519, %r520, %r521, %r522 }, { %r615, %r616, %r617, %r618 }, { %r623, %r624 }, { %r519, %r520, %r521, %r522 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r525, %r526, %r527, %r528 }, { %r615, %r616, %r617, %r618 }, { %r625, %r626 }, { %r525, %r526, %r527, %r528 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r531, %r532, %r533, %r534 }, { %r615, %r616, %r617, %r618 }, { %r627, %r628 }, { %r531, %r532, %r533, %r534 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r537, %r538, %r539, %r540 }, { %r615, %r616, %r617, %r618 }, { %r629, %r630 }, { %r537, %r538, %r539, %r540 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r543, %r544, %r545, %r546 }, { %r615, %r616, %r617, %r618 }, { %r631, %r632 }, { %r543, %r544, %r545, %r546 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r549, %r550, %r551, %r552 }, { %r615, %r616, %r617, %r618 }, { %r633, %r634 }, { %r549, %r550, %r551, %r552 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r555, %r556, %r557, %r558 }, { %r635, %r636, %r637, %r638 }, { %r619, %r620 }, { %r555, %r556, %r557, %r558 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r563, %r564, %r565, %r566 }, { %r635, %r636, %r637, %r638 }, { %r621, %r622 }, { %r563, %r564, %r565, %r566 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r567, %r568, %r569, %r570 }, { %r635, %r636, %r637, %r638 }, { %r623, %r624 }, { %r567, %r568, %r569, %r570 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r571, %r572, %r573, %r574 }, { %r635, %r636, %r637, %r638 }, { %r625, %r626 }, { %r571, %r572, %r573, %r574 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r575, %r576, %r577, %r578 }, { %r635, %r636, %r637, %r638 }, { %r627, %r628 }, { %r575, %r576, %r577, %r578 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r579, %r580, %r581, %r582 }, { %r635, %r636, %r637, %r638 }, { %r629, %r630 }, { %r579, %r580, %r581, %r582 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r583, %r584, %r585, %r586 }, { %r635, %r636, %r637, %r638 }, { %r631, %r632 }, { %r583, %r584, %r585, %r586 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r587, %r588, %r589, %r590 }, { %r635, %r636, %r637, %r638 }, { %r633, %r634 }, { %r587, %r588, %r589, %r590 }; + // end inline asm + .loc 1 219 26 // chunk_delta_h.py:219:26 + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r639, %r640, %r641, %r642}, [%r1113+24576]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r659, %r660, %r661, %r662}, [%r1113+28672]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r663, %r664, %r665, %r666}, [%r1114+24576]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r683, %r684, %r685, %r686}, [%r1114+28672]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r687, %r688, %r689, %r690}, [%r1115+24576]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r707, %r708, %r709, %r710}, [%r1115+28672]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r711, %r712, %r713, %r714}, [%r1116+24576]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r731, %r732, %r733, %r734}, [%r1116+28672]; + .loc 1 193 35 // chunk_delta_h.py:193:35 + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r643, %r644, %r667, %r668}, [%r50]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r691, %r692, %r715, %r716}, [%r51]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r645, %r646, %r669, %r670}, [%r52]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r693, %r694, %r717, %r718}, [%r53]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r647, %r648, %r671, %r672}, [%r54]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r695, %r696, %r719, %r720}, [%r55]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r649, %r650, %r673, %r674}, [%r56]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r697, %r698, %r721, %r722}, [%r57]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r651, %r652, %r675, %r676}, [%r58]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r699, %r700, %r723, %r724}, [%r59]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r653, %r654, %r677, %r678}, [%r60]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r701, %r702, %r725, %r726}, [%r61]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r655, %r656, %r679, %r680}, [%r62]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r703, %r704, %r727, %r728}, [%r63]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r657, %r658, %r681, %r682}, [%r64]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r705, %r706, %r729, %r730}, [%r65]; + .loc 1 223 54 // chunk_delta_h.py:223:54 + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r503, %r504, %r505, %r506 }, { %r639, %r640, %r641, %r642 }, { %r643, %r644 }, { %r503, %r504, %r505, %r506 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r513, %r514, %r515, %r516 }, { %r639, %r640, %r641, %r642 }, { %r645, %r646 }, { %r513, %r514, %r515, %r516 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r519, %r520, %r521, %r522 }, { %r639, %r640, %r641, %r642 }, { %r647, %r648 }, { %r519, %r520, %r521, %r522 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r525, %r526, %r527, %r528 }, { %r639, %r640, %r641, %r642 }, { %r649, %r650 }, { %r525, %r526, %r527, %r528 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r531, %r532, %r533, %r534 }, { %r639, %r640, %r641, %r642 }, { %r651, %r652 }, { %r531, %r532, %r533, %r534 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r537, %r538, %r539, %r540 }, { %r639, %r640, %r641, %r642 }, { %r653, %r654 }, { %r537, %r538, %r539, %r540 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r543, %r544, %r545, %r546 }, { %r639, %r640, %r641, %r642 }, { %r655, %r656 }, { %r543, %r544, %r545, %r546 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r549, %r550, %r551, %r552 }, { %r639, %r640, %r641, %r642 }, { %r657, %r658 }, { %r549, %r550, %r551, %r552 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r555, %r556, %r557, %r558 }, { %r659, %r660, %r661, %r662 }, { %r643, %r644 }, { %r555, %r556, %r557, %r558 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r563, %r564, %r565, %r566 }, { %r659, %r660, %r661, %r662 }, { %r645, %r646 }, { %r563, %r564, %r565, %r566 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r567, %r568, %r569, %r570 }, { %r659, %r660, %r661, %r662 }, { %r647, %r648 }, { %r567, %r568, %r569, %r570 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r571, %r572, %r573, %r574 }, { %r659, %r660, %r661, %r662 }, { %r649, %r650 }, { %r571, %r572, %r573, %r574 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r575, %r576, %r577, %r578 }, { %r659, %r660, %r661, %r662 }, { %r651, %r652 }, { %r575, %r576, %r577, %r578 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r579, %r580, %r581, %r582 }, { %r659, %r660, %r661, %r662 }, { %r653, %r654 }, { %r579, %r580, %r581, %r582 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r583, %r584, %r585, %r586 }, { %r659, %r660, %r661, %r662 }, { %r655, %r656 }, { %r583, %r584, %r585, %r586 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r587, %r588, %r589, %r590 }, { %r659, %r660, %r661, %r662 }, { %r657, %r658 }, { %r587, %r588, %r589, %r590 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r503, %r504, %r505, %r506 }, { %r663, %r664, %r665, %r666 }, { %r667, %r668 }, { %r503, %r504, %r505, %r506 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r513, %r514, %r515, %r516 }, { %r663, %r664, %r665, %r666 }, { %r669, %r670 }, { %r513, %r514, %r515, %r516 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r519, %r520, %r521, %r522 }, { %r663, %r664, %r665, %r666 }, { %r671, %r672 }, { %r519, %r520, %r521, %r522 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r525, %r526, %r527, %r528 }, { %r663, %r664, %r665, %r666 }, { %r673, %r674 }, { %r525, %r526, %r527, %r528 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r531, %r532, %r533, %r534 }, { %r663, %r664, %r665, %r666 }, { %r675, %r676 }, { %r531, %r532, %r533, %r534 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r537, %r538, %r539, %r540 }, { %r663, %r664, %r665, %r666 }, { %r677, %r678 }, { %r537, %r538, %r539, %r540 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r543, %r544, %r545, %r546 }, { %r663, %r664, %r665, %r666 }, { %r679, %r680 }, { %r543, %r544, %r545, %r546 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r549, %r550, %r551, %r552 }, { %r663, %r664, %r665, %r666 }, { %r681, %r682 }, { %r549, %r550, %r551, %r552 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r555, %r556, %r557, %r558 }, { %r683, %r684, %r685, %r686 }, { %r667, %r668 }, { %r555, %r556, %r557, %r558 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r563, %r564, %r565, %r566 }, { %r683, %r684, %r685, %r686 }, { %r669, %r670 }, { %r563, %r564, %r565, %r566 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r567, %r568, %r569, %r570 }, { %r683, %r684, %r685, %r686 }, { %r671, %r672 }, { %r567, %r568, %r569, %r570 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r571, %r572, %r573, %r574 }, { %r683, %r684, %r685, %r686 }, { %r673, %r674 }, { %r571, %r572, %r573, %r574 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r575, %r576, %r577, %r578 }, { %r683, %r684, %r685, %r686 }, { %r675, %r676 }, { %r575, %r576, %r577, %r578 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r579, %r580, %r581, %r582 }, { %r683, %r684, %r685, %r686 }, { %r677, %r678 }, { %r579, %r580, %r581, %r582 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r583, %r584, %r585, %r586 }, { %r683, %r684, %r685, %r686 }, { %r679, %r680 }, { %r583, %r584, %r585, %r586 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r587, %r588, %r589, %r590 }, { %r683, %r684, %r685, %r686 }, { %r681, %r682 }, { %r587, %r588, %r589, %r590 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r503, %r504, %r505, %r506 }, { %r687, %r688, %r689, %r690 }, { %r691, %r692 }, { %r503, %r504, %r505, %r506 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r513, %r514, %r515, %r516 }, { %r687, %r688, %r689, %r690 }, { %r693, %r694 }, { %r513, %r514, %r515, %r516 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r519, %r520, %r521, %r522 }, { %r687, %r688, %r689, %r690 }, { %r695, %r696 }, { %r519, %r520, %r521, %r522 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r525, %r526, %r527, %r528 }, { %r687, %r688, %r689, %r690 }, { %r697, %r698 }, { %r525, %r526, %r527, %r528 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r531, %r532, %r533, %r534 }, { %r687, %r688, %r689, %r690 }, { %r699, %r700 }, { %r531, %r532, %r533, %r534 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r537, %r538, %r539, %r540 }, { %r687, %r688, %r689, %r690 }, { %r701, %r702 }, { %r537, %r538, %r539, %r540 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r543, %r544, %r545, %r546 }, { %r687, %r688, %r689, %r690 }, { %r703, %r704 }, { %r543, %r544, %r545, %r546 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r549, %r550, %r551, %r552 }, { %r687, %r688, %r689, %r690 }, { %r705, %r706 }, { %r549, %r550, %r551, %r552 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r555, %r556, %r557, %r558 }, { %r707, %r708, %r709, %r710 }, { %r691, %r692 }, { %r555, %r556, %r557, %r558 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r563, %r564, %r565, %r566 }, { %r707, %r708, %r709, %r710 }, { %r693, %r694 }, { %r563, %r564, %r565, %r566 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r567, %r568, %r569, %r570 }, { %r707, %r708, %r709, %r710 }, { %r695, %r696 }, { %r567, %r568, %r569, %r570 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r571, %r572, %r573, %r574 }, { %r707, %r708, %r709, %r710 }, { %r697, %r698 }, { %r571, %r572, %r573, %r574 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r575, %r576, %r577, %r578 }, { %r707, %r708, %r709, %r710 }, { %r699, %r700 }, { %r575, %r576, %r577, %r578 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r579, %r580, %r581, %r582 }, { %r707, %r708, %r709, %r710 }, { %r701, %r702 }, { %r579, %r580, %r581, %r582 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r583, %r584, %r585, %r586 }, { %r707, %r708, %r709, %r710 }, { %r703, %r704 }, { %r583, %r584, %r585, %r586 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r587, %r588, %r589, %r590 }, { %r707, %r708, %r709, %r710 }, { %r705, %r706 }, { %r587, %r588, %r589, %r590 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r503, %r504, %r505, %r506 }, { %r711, %r712, %r713, %r714 }, { %r715, %r716 }, { %r503, %r504, %r505, %r506 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r513, %r514, %r515, %r516 }, { %r711, %r712, %r713, %r714 }, { %r717, %r718 }, { %r513, %r514, %r515, %r516 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r519, %r520, %r521, %r522 }, { %r711, %r712, %r713, %r714 }, { %r719, %r720 }, { %r519, %r520, %r521, %r522 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r525, %r526, %r527, %r528 }, { %r711, %r712, %r713, %r714 }, { %r721, %r722 }, { %r525, %r526, %r527, %r528 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r531, %r532, %r533, %r534 }, { %r711, %r712, %r713, %r714 }, { %r723, %r724 }, { %r531, %r532, %r533, %r534 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r537, %r538, %r539, %r540 }, { %r711, %r712, %r713, %r714 }, { %r725, %r726 }, { %r537, %r538, %r539, %r540 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r543, %r544, %r545, %r546 }, { %r711, %r712, %r713, %r714 }, { %r727, %r728 }, { %r543, %r544, %r545, %r546 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r549, %r550, %r551, %r552 }, { %r711, %r712, %r713, %r714 }, { %r729, %r730 }, { %r549, %r550, %r551, %r552 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r555, %r556, %r557, %r558 }, { %r731, %r732, %r733, %r734 }, { %r715, %r716 }, { %r555, %r556, %r557, %r558 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r563, %r564, %r565, %r566 }, { %r731, %r732, %r733, %r734 }, { %r717, %r718 }, { %r563, %r564, %r565, %r566 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r567, %r568, %r569, %r570 }, { %r731, %r732, %r733, %r734 }, { %r719, %r720 }, { %r567, %r568, %r569, %r570 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r571, %r572, %r573, %r574 }, { %r731, %r732, %r733, %r734 }, { %r721, %r722 }, { %r571, %r572, %r573, %r574 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r575, %r576, %r577, %r578 }, { %r731, %r732, %r733, %r734 }, { %r723, %r724 }, { %r575, %r576, %r577, %r578 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r579, %r580, %r581, %r582 }, { %r731, %r732, %r733, %r734 }, { %r725, %r726 }, { %r579, %r580, %r581, %r582 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r583, %r584, %r585, %r586 }, { %r731, %r732, %r733, %r734 }, { %r727, %r728 }, { %r583, %r584, %r585, %r586 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r587, %r588, %r589, %r590 }, { %r731, %r732, %r733, %r734 }, { %r729, %r730 }, { %r587, %r588, %r589, %r590 }; + // end inline asm + .loc 1 239 74 // chunk_delta_h.py:239:74 + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r1117, %r1118, %r1119, %r1120}, [%r1113+49152]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r1121, %r1122, %r1123, %r1124}, [%r1113+53248]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r1125, %r1126, %r1127, %r1128}, [%r1114+49152]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r1129, %r1130, %r1131, %r1132}, [%r1114+53248]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r1133, %r1134, %r1135, %r1136}, [%r1115+49152]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r1137, %r1138, %r1139, %r1140}, [%r1115+53248]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r1141, %r1142, %r1143, %r1144}, [%r1116+49152]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r1145, %r1146, %r1147, %r1148}, [%r1116+53248]; + .loc 1 242 50 // chunk_delta_h.py:242:50 + add.s64 %rd308, %rd468, %rd18; + add.s64 %rd309, %rd308, 65536; + add.s64 %rd310, %rd308, 131072; + add.s64 %rd311, %rd308, 196608; + add.s64 %rd312, %rd308, 262144; + add.s64 %rd313, %rd308, 327680; + add.s64 %rd314, %rd308, 393216; + .loc 1 239 74 // chunk_delta_h.py:239:74 + add.s64 %rd315, %rd308, 458752; + mov.b32 {%rs1, %rs2}, %r1117; + cvt.f32.bf16 %r1149, %rs2; + cvt.f32.bf16 %r1150, %rs1; + mov.b64 %rd357, {%r1150, %r1149}; + mov.b64 %rd358, {%r503, %r504}; + sub.f32x2 %rd359, %rd357, %rd358; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1151, %r1152}, %rd359; + cvt.rn.bf16x2.f32 %r736, %r1152, %r1151; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs3, %rs4}, %r1118; + cvt.f32.bf16 %r1153, %rs4; + cvt.f32.bf16 %r1154, %rs3; + mov.b64 %rd360, {%r1154, %r1153}; + mov.b64 %rd361, {%r505, %r506}; + sub.f32x2 %rd362, %rd360, %rd361; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1155, %r1156}, %rd362; + cvt.rn.bf16x2.f32 %r745, %r1156, %r1155; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs5, %rs6}, %r1119; + cvt.f32.bf16 %r1157, %rs6; + cvt.f32.bf16 %r1158, %rs5; + mov.b64 %rd363, {%r1158, %r1157}; + mov.b64 %rd364, {%r513, %r514}; + sub.f32x2 %rd365, %rd363, %rd364; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1159, %r1160}, %rd365; + cvt.rn.bf16x2.f32 %r737, %r1160, %r1159; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs7, %rs8}, %r1120; + cvt.f32.bf16 %r1161, %rs8; + cvt.f32.bf16 %r1162, %rs7; + mov.b64 %rd366, {%r1162, %r1161}; + mov.b64 %rd367, {%r515, %r516}; + sub.f32x2 %rd368, %rd366, %rd367; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1163, %r1164}, %rd368; + cvt.rn.bf16x2.f32 %r746, %r1164, %r1163; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs9, %rs10}, %r1125; + cvt.f32.bf16 %r1165, %rs10; + cvt.f32.bf16 %r1166, %rs9; + mov.b64 %rd369, {%r1166, %r1165}; + mov.b64 %rd370, {%r519, %r520}; + sub.f32x2 %rd371, %rd369, %rd370; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1167, %r1168}, %rd371; + cvt.rn.bf16x2.f32 %r738, %r1168, %r1167; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs11, %rs12}, %r1126; + cvt.f32.bf16 %r1169, %rs12; + cvt.f32.bf16 %r1170, %rs11; + mov.b64 %rd372, {%r1170, %r1169}; + mov.b64 %rd373, {%r521, %r522}; + sub.f32x2 %rd374, %rd372, %rd373; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1171, %r1172}, %rd374; + cvt.rn.bf16x2.f32 %r747, %r1172, %r1171; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs13, %rs14}, %r1127; + cvt.f32.bf16 %r1173, %rs14; + cvt.f32.bf16 %r1174, %rs13; + mov.b64 %rd375, {%r1174, %r1173}; + mov.b64 %rd376, {%r525, %r526}; + sub.f32x2 %rd377, %rd375, %rd376; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1175, %r1176}, %rd377; + cvt.rn.bf16x2.f32 %r739, %r1176, %r1175; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs15, %rs16}, %r1128; + cvt.f32.bf16 %r1177, %rs16; + cvt.f32.bf16 %r1178, %rs15; + mov.b64 %rd378, {%r1178, %r1177}; + mov.b64 %rd379, {%r527, %r528}; + sub.f32x2 %rd380, %rd378, %rd379; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1179, %r1180}, %rd380; + cvt.rn.bf16x2.f32 %r748, %r1180, %r1179; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs17, %rs18}, %r1133; + cvt.f32.bf16 %r1181, %rs18; + cvt.f32.bf16 %r1182, %rs17; + mov.b64 %rd381, {%r1182, %r1181}; + mov.b64 %rd382, {%r531, %r532}; + sub.f32x2 %rd383, %rd381, %rd382; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1183, %r1184}, %rd383; + cvt.rn.bf16x2.f32 %r741, %r1184, %r1183; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs19, %rs20}, %r1134; + cvt.f32.bf16 %r1185, %rs20; + cvt.f32.bf16 %r1186, %rs19; + mov.b64 %rd384, {%r1186, %r1185}; + mov.b64 %rd385, {%r533, %r534}; + sub.f32x2 %rd386, %rd384, %rd385; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1187, %r1188}, %rd386; + cvt.rn.bf16x2.f32 %r749, %r1188, %r1187; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs21, %rs22}, %r1135; + cvt.f32.bf16 %r1189, %rs22; + cvt.f32.bf16 %r1190, %rs21; + mov.b64 %rd387, {%r1190, %r1189}; + mov.b64 %rd388, {%r537, %r538}; + sub.f32x2 %rd389, %rd387, %rd388; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1191, %r1192}, %rd389; + cvt.rn.bf16x2.f32 %r742, %r1192, %r1191; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs23, %rs24}, %r1136; + cvt.f32.bf16 %r1193, %rs24; + cvt.f32.bf16 %r1194, %rs23; + mov.b64 %rd390, {%r1194, %r1193}; + mov.b64 %rd391, {%r539, %r540}; + sub.f32x2 %rd392, %rd390, %rd391; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1195, %r1196}, %rd392; + cvt.rn.bf16x2.f32 %r750, %r1196, %r1195; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs25, %rs26}, %r1141; + cvt.f32.bf16 %r1197, %rs26; + cvt.f32.bf16 %r1198, %rs25; + mov.b64 %rd393, {%r1198, %r1197}; + mov.b64 %rd394, {%r543, %r544}; + sub.f32x2 %rd395, %rd393, %rd394; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1199, %r1200}, %rd395; + cvt.rn.bf16x2.f32 %r743, %r1200, %r1199; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs27, %rs28}, %r1142; + cvt.f32.bf16 %r1201, %rs28; + cvt.f32.bf16 %r1202, %rs27; + mov.b64 %rd396, {%r1202, %r1201}; + mov.b64 %rd397, {%r545, %r546}; + sub.f32x2 %rd398, %rd396, %rd397; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1203, %r1204}, %rd398; + cvt.rn.bf16x2.f32 %r751, %r1204, %r1203; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs29, %rs30}, %r1143; + cvt.f32.bf16 %r1205, %rs30; + cvt.f32.bf16 %r1206, %rs29; + mov.b64 %rd399, {%r1206, %r1205}; + mov.b64 %rd400, {%r549, %r550}; + sub.f32x2 %rd401, %rd399, %rd400; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1207, %r1208}, %rd401; + cvt.rn.bf16x2.f32 %r744, %r1208, %r1207; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs31, %rs32}, %r1144; + cvt.f32.bf16 %r1209, %rs32; + cvt.f32.bf16 %r1210, %rs31; + mov.b64 %rd402, {%r1210, %r1209}; + mov.b64 %rd403, {%r551, %r552}; + sub.f32x2 %rd404, %rd402, %rd403; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1211, %r1212}, %rd404; + cvt.rn.bf16x2.f32 %r752, %r1212, %r1211; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs33, %rs34}, %r1121; + cvt.f32.bf16 %r1213, %rs34; + cvt.f32.bf16 %r1214, %rs33; + mov.b64 %rd405, {%r1214, %r1213}; + mov.b64 %rd406, {%r555, %r556}; + sub.f32x2 %rd407, %rd405, %rd406; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1215, %r1216}, %rd407; + cvt.rn.bf16x2.f32 %r753, %r1216, %r1215; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs35, %rs36}, %r1122; + cvt.f32.bf16 %r1217, %rs36; + cvt.f32.bf16 %r1218, %rs35; + mov.b64 %rd408, {%r1218, %r1217}; + mov.b64 %rd409, {%r557, %r558}; + sub.f32x2 %rd410, %rd408, %rd409; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1219, %r1220}, %rd410; + cvt.rn.bf16x2.f32 %r761, %r1220, %r1219; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs37, %rs38}, %r1123; + cvt.f32.bf16 %r1221, %rs38; + cvt.f32.bf16 %r1222, %rs37; + mov.b64 %rd411, {%r1222, %r1221}; + mov.b64 %rd412, {%r563, %r564}; + sub.f32x2 %rd413, %rd411, %rd412; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1223, %r1224}, %rd413; + cvt.rn.bf16x2.f32 %r754, %r1224, %r1223; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs39, %rs40}, %r1124; + cvt.f32.bf16 %r1225, %rs40; + cvt.f32.bf16 %r1226, %rs39; + mov.b64 %rd414, {%r1226, %r1225}; + mov.b64 %rd415, {%r565, %r566}; + sub.f32x2 %rd416, %rd414, %rd415; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1227, %r1228}, %rd416; + cvt.rn.bf16x2.f32 %r762, %r1228, %r1227; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs41, %rs42}, %r1129; + cvt.f32.bf16 %r1229, %rs42; + cvt.f32.bf16 %r1230, %rs41; + mov.b64 %rd417, {%r1230, %r1229}; + mov.b64 %rd418, {%r567, %r568}; + sub.f32x2 %rd419, %rd417, %rd418; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1231, %r1232}, %rd419; + cvt.rn.bf16x2.f32 %r755, %r1232, %r1231; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs43, %rs44}, %r1130; + cvt.f32.bf16 %r1233, %rs44; + cvt.f32.bf16 %r1234, %rs43; + mov.b64 %rd420, {%r1234, %r1233}; + mov.b64 %rd421, {%r569, %r570}; + sub.f32x2 %rd422, %rd420, %rd421; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1235, %r1236}, %rd422; + cvt.rn.bf16x2.f32 %r763, %r1236, %r1235; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs45, %rs46}, %r1131; + cvt.f32.bf16 %r1237, %rs46; + cvt.f32.bf16 %r1238, %rs45; + mov.b64 %rd423, {%r1238, %r1237}; + mov.b64 %rd424, {%r571, %r572}; + sub.f32x2 %rd425, %rd423, %rd424; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1239, %r1240}, %rd425; + cvt.rn.bf16x2.f32 %r756, %r1240, %r1239; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs47, %rs48}, %r1132; + cvt.f32.bf16 %r1241, %rs48; + cvt.f32.bf16 %r1242, %rs47; + mov.b64 %rd426, {%r1242, %r1241}; + mov.b64 %rd427, {%r573, %r574}; + sub.f32x2 %rd428, %rd426, %rd427; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1243, %r1244}, %rd428; + cvt.rn.bf16x2.f32 %r764, %r1244, %r1243; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs49, %rs50}, %r1137; + cvt.f32.bf16 %r1245, %rs50; + cvt.f32.bf16 %r1246, %rs49; + mov.b64 %rd429, {%r1246, %r1245}; + mov.b64 %rd430, {%r575, %r576}; + sub.f32x2 %rd431, %rd429, %rd430; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1247, %r1248}, %rd431; + cvt.rn.bf16x2.f32 %r757, %r1248, %r1247; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs51, %rs52}, %r1138; + cvt.f32.bf16 %r1249, %rs52; + cvt.f32.bf16 %r1250, %rs51; + mov.b64 %rd432, {%r1250, %r1249}; + mov.b64 %rd433, {%r577, %r578}; + sub.f32x2 %rd434, %rd432, %rd433; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1251, %r1252}, %rd434; + cvt.rn.bf16x2.f32 %r765, %r1252, %r1251; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs53, %rs54}, %r1139; + cvt.f32.bf16 %r1253, %rs54; + cvt.f32.bf16 %r1254, %rs53; + mov.b64 %rd435, {%r1254, %r1253}; + mov.b64 %rd436, {%r579, %r580}; + sub.f32x2 %rd437, %rd435, %rd436; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1255, %r1256}, %rd437; + cvt.rn.bf16x2.f32 %r758, %r1256, %r1255; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs55, %rs56}, %r1140; + cvt.f32.bf16 %r1257, %rs56; + cvt.f32.bf16 %r1258, %rs55; + mov.b64 %rd438, {%r1258, %r1257}; + mov.b64 %rd439, {%r581, %r582}; + sub.f32x2 %rd440, %rd438, %rd439; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1259, %r1260}, %rd440; + cvt.rn.bf16x2.f32 %r766, %r1260, %r1259; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs57, %rs58}, %r1145; + cvt.f32.bf16 %r1261, %rs58; + cvt.f32.bf16 %r1262, %rs57; + mov.b64 %rd441, {%r1262, %r1261}; + mov.b64 %rd442, {%r583, %r584}; + sub.f32x2 %rd443, %rd441, %rd442; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1263, %r1264}, %rd443; + cvt.rn.bf16x2.f32 %r759, %r1264, %r1263; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs59, %rs60}, %r1146; + cvt.f32.bf16 %r1265, %rs60; + cvt.f32.bf16 %r1266, %rs59; + mov.b64 %rd444, {%r1266, %r1265}; + mov.b64 %rd445, {%r585, %r586}; + sub.f32x2 %rd446, %rd444, %rd445; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1267, %r1268}, %rd446; + cvt.rn.bf16x2.f32 %r767, %r1268, %r1267; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs61, %rs62}, %r1147; + cvt.f32.bf16 %r1269, %rs62; + cvt.f32.bf16 %r1270, %rs61; + mov.b64 %rd447, {%r1270, %r1269}; + mov.b64 %rd448, {%r587, %r588}; + sub.f32x2 %rd449, %rd447, %rd448; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1271, %r1272}, %rd449; + cvt.rn.bf16x2.f32 %r760, %r1272, %r1271; + .loc 1 239 74 // chunk_delta_h.py:239:74 + mov.b32 {%rs63, %rs64}, %r1148; + cvt.f32.bf16 %r1273, %rs64; + cvt.f32.bf16 %r1274, %rs63; + mov.b64 %rd450, {%r1274, %r1273}; + mov.b64 %rd451, {%r589, %r590}; + sub.f32x2 %rd452, %rd450, %rd451; + .loc 1 243 33 // chunk_delta_h.py:243:33 + mov.b64 {%r1275, %r1276}, %rd452; + cvt.rn.bf16x2.f32 %r768, %r1276, %r1275; + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r735 + 0 ], { %r736, %r737, %r738, %r739 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r740 + 0 ], { %r741, %r742, %r743, %r744 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r769, %r770, %r771, %r772}, [%r66]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r777, %r778, %r779, %r780}, [%r67]; + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r735 + 0 ], { %r745, %r746, %r747, %r748 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r740 + 0 ], { %r749, %r750, %r751, %r752 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r773, %r774, %r775, %r776}, [%r66]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r781, %r782, %r783, %r784}, [%r67]; + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r735 + 0 ], { %r753, %r754, %r755, %r756 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r740 + 0 ], { %r757, %r758, %r759, %r760 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r785, %r786, %r787, %r788}, [%r66]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r793, %r794, %r795, %r796}, [%r67]; + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r735 + 0 ], { %r761, %r762, %r763, %r764 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r740 + 0 ], { %r765, %r766, %r767, %r768 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r789, %r790, %r791, %r792}, [%r66]; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r797, %r798, %r799, %r800}, [%r67]; + .loc 1 243 26 // chunk_delta_h.py:243:26 + // begin inline asm + @%p8 st.global.v4.b32 [ %rd308 + 0 ], { %r769, %r770, %r771, %r772 }; + // end inline asm + // begin inline asm + @%p7 st.global.v4.b32 [ %rd309 + 0 ], { %r773, %r774, %r775, %r776 }; + // end inline asm + // begin inline asm + @%p6 st.global.v4.b32 [ %rd310 + 0 ], { %r777, %r778, %r779, %r780 }; + // end inline asm + // begin inline asm + @%p5 st.global.v4.b32 [ %rd311 + 0 ], { %r781, %r782, %r783, %r784 }; + // end inline asm + // begin inline asm + @%p4 st.global.v4.b32 [ %rd312 + 0 ], { %r785, %r786, %r787, %r788 }; + // end inline asm + // begin inline asm + @%p3 st.global.v4.b32 [ %rd313 + 0 ], { %r789, %r790, %r791, %r792 }; + // end inline asm + // begin inline asm + @%p2 st.global.v4.b32 [ %rd314 + 0 ], { %r793, %r794, %r795, %r796 }; + // end inline asm + // begin inline asm + @%p1 st.global.v4.b32 [ %rd315 + 0 ], { %r797, %r798, %r799, %r800 }; + // end inline asm + .loc 1 247 31 // chunk_delta_h.py:247:31 + shl.b32 %r1277, %r1418, 2; + add.s32 %r1278, %r246, %r1277; + ld.shared.b32 %r1279, [%r1278+142080]; + .loc 1 249 26 // chunk_delta_h.py:249:26 + shl.b32 %r1280, %r1418, 8; + add.s32 %r1281, %r68, %r1280; + ld.shared.b32 %r1282, [%r1281+141312]; + ld.shared.b32 %r1283, [%r1281+141344]; + ld.shared.b32 %r1284, [%r1281+141440]; + ld.shared.b32 %r1285, [%r1281+141472]; + .loc 1 250 54 // chunk_delta_h.py:250:54 + sub.f32 %r1286, %r1279, %r1282; + sub.f32 %r1287, %r1279, %r1283; + sub.f32 %r1288, %r1279, %r1284; + sub.f32 %r1289, %r1279, %r1285; +$L__tmp5: + .loc 3 47 33 // op.py:47:33 @[ chunk_delta_h.py:250:43 ] + ex2.approx.ftz.f32 %r1290, %r1286; + ex2.approx.ftz.f32 %r1291, %r1287; + ex2.approx.ftz.f32 %r1292, %r1288; + ex2.approx.ftz.f32 %r1293, %r1289; +$L__tmp6: + .loc 1 250 60 // chunk_delta_h.py:250:60 + selp.f32 %r1294, %r1290, 0f00000000, %p55; + selp.f32 %r1295, %r1291, 0f00000000, %p54; + selp.f32 %r1296, %r1292, 0f00000000, %p53; + selp.f32 %r1297, %r1293, 0f00000000, %p52; + .loc 1 250 24 // chunk_delta_h.py:250:24 + mul.f32 %r1298, %r1151, %r1294; + mul.f32 %r1299, %r1152, %r1294; + mul.f32 %r1300, %r1155, %r1295; + mul.f32 %r1301, %r1156, %r1295; + mul.f32 %r1302, %r1159, %r1294; + mul.f32 %r1303, %r1160, %r1294; + mul.f32 %r1304, %r1163, %r1295; + mul.f32 %r1305, %r1164, %r1295; + mul.f32 %r1306, %r1167, %r1294; + mul.f32 %r1307, %r1168, %r1294; + mul.f32 %r1308, %r1171, %r1295; + mul.f32 %r1309, %r1172, %r1295; + mul.f32 %r1310, %r1175, %r1294; + mul.f32 %r1311, %r1176, %r1294; + mul.f32 %r1312, %r1179, %r1295; + mul.f32 %r1313, %r1180, %r1295; + mul.f32 %r1314, %r1183, %r1294; + mul.f32 %r1315, %r1184, %r1294; + mul.f32 %r1316, %r1187, %r1295; + mul.f32 %r1317, %r1188, %r1295; + mul.f32 %r1318, %r1191, %r1294; + mul.f32 %r1319, %r1192, %r1294; + mul.f32 %r1320, %r1195, %r1295; + mul.f32 %r1321, %r1196, %r1295; + mul.f32 %r1322, %r1199, %r1294; + mul.f32 %r1323, %r1200, %r1294; + mul.f32 %r1324, %r1203, %r1295; + mul.f32 %r1325, %r1204, %r1295; + mul.f32 %r1326, %r1207, %r1294; + mul.f32 %r1327, %r1208, %r1294; + mul.f32 %r1328, %r1211, %r1295; + mul.f32 %r1329, %r1212, %r1295; + mul.f32 %r1330, %r1215, %r1296; + mul.f32 %r1331, %r1216, %r1296; + mul.f32 %r1332, %r1219, %r1297; + mul.f32 %r1333, %r1220, %r1297; + mul.f32 %r1334, %r1223, %r1296; + mul.f32 %r1335, %r1224, %r1296; + mul.f32 %r1336, %r1227, %r1297; + mul.f32 %r1337, %r1228, %r1297; + mul.f32 %r1338, %r1231, %r1296; + mul.f32 %r1339, %r1232, %r1296; + mul.f32 %r1340, %r1235, %r1297; + mul.f32 %r1341, %r1236, %r1297; + mul.f32 %r1342, %r1239, %r1296; + mul.f32 %r1343, %r1240, %r1296; + mul.f32 %r1344, %r1243, %r1297; + mul.f32 %r1345, %r1244, %r1297; + mul.f32 %r1346, %r1247, %r1296; + mul.f32 %r1347, %r1248, %r1296; + mul.f32 %r1348, %r1251, %r1297; + mul.f32 %r1349, %r1252, %r1297; + mul.f32 %r1350, %r1255, %r1296; + mul.f32 %r1351, %r1256, %r1296; + mul.f32 %r1352, %r1259, %r1297; + mul.f32 %r1353, %r1260, %r1297; + mul.f32 %r1354, %r1263, %r1296; + mul.f32 %r1355, %r1264, %r1296; + mul.f32 %r1356, %r1267, %r1297; + mul.f32 %r1357, %r1268, %r1297; + mul.f32 %r1358, %r1271, %r1296; + mul.f32 %r1359, %r1272, %r1296; + mul.f32 %r1360, %r1275, %r1297; + mul.f32 %r1361, %r1276, %r1297; +$L__tmp7: + .loc 3 47 33 // op.py:47:33 @[ chunk_delta_h.py:251:28 ] + ex2.approx.ftz.f32 %r1362, %r1279; +$L__tmp8: + .loc 1 252 20 // chunk_delta_h.py:252:20 + mul.f32 %r825, %r825, %r1362; + mul.f32 %r826, %r826, %r1362; + mul.f32 %r827, %r827, %r1362; + mul.f32 %r828, %r828, %r1362; + mul.f32 %r835, %r835, %r1362; + mul.f32 %r836, %r836, %r1362; + mul.f32 %r837, %r837, %r1362; + mul.f32 %r838, %r838, %r1362; + mul.f32 %r841, %r841, %r1362; + mul.f32 %r842, %r842, %r1362; + mul.f32 %r843, %r843, %r1362; + mul.f32 %r844, %r844, %r1362; + mul.f32 %r847, %r847, %r1362; + mul.f32 %r848, %r848, %r1362; + mul.f32 %r849, %r849, %r1362; + mul.f32 %r850, %r850, %r1362; + mul.f32 %r853, %r853, %r1362; + mul.f32 %r854, %r854, %r1362; + mul.f32 %r855, %r855, %r1362; + mul.f32 %r856, %r856, %r1362; + mul.f32 %r859, %r859, %r1362; + mul.f32 %r860, %r860, %r1362; + mul.f32 %r861, %r861, %r1362; + mul.f32 %r862, %r862, %r1362; + mul.f32 %r865, %r865, %r1362; + mul.f32 %r866, %r866, %r1362; + mul.f32 %r867, %r867, %r1362; + mul.f32 %r868, %r868, %r1362; + mul.f32 %r871, %r871, %r1362; + mul.f32 %r872, %r872, %r1362; + mul.f32 %r873, %r873, %r1362; + mul.f32 %r874, %r874, %r1362; + mul.f32 %r877, %r877, %r1362; + mul.f32 %r878, %r878, %r1362; + mul.f32 %r879, %r879, %r1362; + mul.f32 %r880, %r880, %r1362; + mul.f32 %r885, %r885, %r1362; + mul.f32 %r886, %r886, %r1362; + mul.f32 %r887, %r887, %r1362; + mul.f32 %r888, %r888, %r1362; + mul.f32 %r889, %r889, %r1362; + mul.f32 %r890, %r890, %r1362; + mul.f32 %r891, %r891, %r1362; + mul.f32 %r892, %r892, %r1362; + mul.f32 %r893, %r893, %r1362; + mul.f32 %r894, %r894, %r1362; + mul.f32 %r895, %r895, %r1362; + mul.f32 %r896, %r896, %r1362; + mul.f32 %r897, %r897, %r1362; + mul.f32 %r898, %r898, %r1362; + mul.f32 %r899, %r899, %r1362; + mul.f32 %r900, %r900, %r1362; + mul.f32 %r901, %r901, %r1362; + mul.f32 %r902, %r902, %r1362; + mul.f32 %r903, %r903, %r1362; + mul.f32 %r904, %r904, %r1362; + mul.f32 %r905, %r905, %r1362; + mul.f32 %r906, %r906, %r1362; + mul.f32 %r907, %r907, %r1362; + mul.f32 %r908, %r908, %r1362; + mul.f32 %r909, %r909, %r1362; + mul.f32 %r910, %r910, %r1362; + mul.f32 %r911, %r911, %r1362; + mul.f32 %r912, %r912, %r1362; + .loc 1 254 24 // chunk_delta_h.py:254:24 + mul.f32 %r969, %r969, %r1362; + mul.f32 %r970, %r970, %r1362; + mul.f32 %r971, %r971, %r1362; + mul.f32 %r972, %r972, %r1362; + mul.f32 %r977, %r977, %r1362; + mul.f32 %r978, %r978, %r1362; + mul.f32 %r979, %r979, %r1362; + mul.f32 %r980, %r980, %r1362; + mul.f32 %r981, %r981, %r1362; + mul.f32 %r982, %r982, %r1362; + mul.f32 %r983, %r983, %r1362; + mul.f32 %r984, %r984, %r1362; + mul.f32 %r985, %r985, %r1362; + mul.f32 %r986, %r986, %r1362; + mul.f32 %r987, %r987, %r1362; + mul.f32 %r988, %r988, %r1362; + mul.f32 %r989, %r989, %r1362; + mul.f32 %r990, %r990, %r1362; + mul.f32 %r991, %r991, %r1362; + mul.f32 %r992, %r992, %r1362; + mul.f32 %r993, %r993, %r1362; + mul.f32 %r994, %r994, %r1362; + mul.f32 %r995, %r995, %r1362; + mul.f32 %r996, %r996, %r1362; + mul.f32 %r997, %r997, %r1362; + mul.f32 %r998, %r998, %r1362; + mul.f32 %r999, %r999, %r1362; + mul.f32 %r1000, %r1000, %r1362; + mul.f32 %r1001, %r1001, %r1362; + mul.f32 %r1002, %r1002, %r1362; + mul.f32 %r1003, %r1003, %r1362; + mul.f32 %r1004, %r1004, %r1362; + mul.f32 %r1005, %r1005, %r1362; + mul.f32 %r1006, %r1006, %r1362; + mul.f32 %r1007, %r1007, %r1362; + mul.f32 %r1008, %r1008, %r1362; + mul.f32 %r1013, %r1013, %r1362; + mul.f32 %r1014, %r1014, %r1362; + mul.f32 %r1015, %r1015, %r1362; + mul.f32 %r1016, %r1016, %r1362; + mul.f32 %r1017, %r1017, %r1362; + mul.f32 %r1018, %r1018, %r1362; + mul.f32 %r1019, %r1019, %r1362; + mul.f32 %r1020, %r1020, %r1362; + mul.f32 %r1021, %r1021, %r1362; + mul.f32 %r1022, %r1022, %r1362; + mul.f32 %r1023, %r1023, %r1362; + mul.f32 %r1024, %r1024, %r1362; + mul.f32 %r1025, %r1025, %r1362; + mul.f32 %r1026, %r1026, %r1362; + mul.f32 %r1027, %r1027, %r1362; + mul.f32 %r1028, %r1028, %r1362; + mul.f32 %r1029, %r1029, %r1362; + mul.f32 %r1030, %r1030, %r1362; + mul.f32 %r1031, %r1031, %r1362; + mul.f32 %r1032, %r1032, %r1362; + mul.f32 %r1033, %r1033, %r1362; + mul.f32 %r1034, %r1034, %r1362; + mul.f32 %r1035, %r1035, %r1362; + mul.f32 %r1036, %r1036, %r1362; + mul.f32 %r1037, %r1037, %r1362; + mul.f32 %r1038, %r1038, %r1362; + mul.f32 %r1039, %r1039, %r1362; + mul.f32 %r1040, %r1040, %r1362; + .loc 1 288 21 // chunk_delta_h.py:288:21 + bar.sync 0; + cvt.rn.bf16x2.f32 %r1363, %r1299, %r1298; + cvt.rn.bf16x2.f32 %r1364, %r1301, %r1300; + cvt.rn.bf16x2.f32 %r1365, %r1303, %r1302; + cvt.rn.bf16x2.f32 %r1366, %r1305, %r1304; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r11], {%r1363, %r1364, %r1365, %r1366}; + cvt.rn.bf16x2.f32 %r1367, %r1331, %r1330; + cvt.rn.bf16x2.f32 %r1368, %r1333, %r1332; + cvt.rn.bf16x2.f32 %r1369, %r1335, %r1334; + cvt.rn.bf16x2.f32 %r1370, %r1337, %r1336; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r12], {%r1367, %r1368, %r1369, %r1370}; + cvt.rn.bf16x2.f32 %r1371, %r1307, %r1306; + cvt.rn.bf16x2.f32 %r1372, %r1309, %r1308; + cvt.rn.bf16x2.f32 %r1373, %r1311, %r1310; + cvt.rn.bf16x2.f32 %r1374, %r1313, %r1312; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r14], {%r1371, %r1372, %r1373, %r1374}; + cvt.rn.bf16x2.f32 %r1375, %r1339, %r1338; + cvt.rn.bf16x2.f32 %r1376, %r1341, %r1340; + cvt.rn.bf16x2.f32 %r1377, %r1343, %r1342; + cvt.rn.bf16x2.f32 %r1378, %r1345, %r1344; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r15], {%r1375, %r1376, %r1377, %r1378}; + cvt.rn.bf16x2.f32 %r1379, %r1315, %r1314; + cvt.rn.bf16x2.f32 %r1380, %r1317, %r1316; + cvt.rn.bf16x2.f32 %r1381, %r1319, %r1318; + cvt.rn.bf16x2.f32 %r1382, %r1321, %r1320; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r17], {%r1379, %r1380, %r1381, %r1382}; + cvt.rn.bf16x2.f32 %r1383, %r1347, %r1346; + cvt.rn.bf16x2.f32 %r1384, %r1349, %r1348; + cvt.rn.bf16x2.f32 %r1385, %r1351, %r1350; + cvt.rn.bf16x2.f32 %r1386, %r1353, %r1352; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r18], {%r1383, %r1384, %r1385, %r1386}; + cvt.rn.bf16x2.f32 %r1387, %r1323, %r1322; + cvt.rn.bf16x2.f32 %r1388, %r1325, %r1324; + cvt.rn.bf16x2.f32 %r1389, %r1327, %r1326; + cvt.rn.bf16x2.f32 %r1390, %r1329, %r1328; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r20], {%r1387, %r1388, %r1389, %r1390}; + cvt.rn.bf16x2.f32 %r1391, %r1355, %r1354; + cvt.rn.bf16x2.f32 %r1392, %r1357, %r1356; + cvt.rn.bf16x2.f32 %r1393, %r1359, %r1358; + cvt.rn.bf16x2.f32 %r1394, %r1361, %r1360; + stmatrix.sync.aligned.m8n8.x4.shared.b16 [%r21], {%r1391, %r1392, %r1393, %r1394}; + .loc 1 291 22 // chunk_delta_h.py:291:22 + add.s32 %r1395, %r1112, %r69; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r801, %r802, %r803, %r804}, [%r1395+73728]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r829, %r830, %r831, %r832}, [%r1395+75776]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r913, %r914, %r915, %r916}, [%r1395+77824]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r937, %r938, %r939, %r940}, [%r1395+79872]; + add.s32 %r1396, %r1112, %r70; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r821, %r822, %r823, %r824}, [%r1396+73728]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r881, %r882, %r883, %r884}, [%r1396+75776]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r933, %r934, %r935, %r936}, [%r1396+77824]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r957, %r958, %r959, %r960}, [%r1396+79872]; + .loc 1 288 21 // chunk_delta_h.py:288:21 + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r805, %r806, %r833, %r834}, [%r34]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r917, %r918, %r941, %r942}, [%r35]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r807, %r808, %r839, %r840}, [%r36]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r919, %r920, %r943, %r944}, [%r37]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r809, %r810, %r845, %r846}, [%r38]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r921, %r922, %r945, %r946}, [%r39]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r811, %r812, %r851, %r852}, [%r40]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r923, %r924, %r947, %r948}, [%r41]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r813, %r814, %r857, %r858}, [%r42]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r925, %r926, %r949, %r950}, [%r43]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r815, %r816, %r863, %r864}, [%r44]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r927, %r928, %r951, %r952}, [%r45]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r817, %r818, %r869, %r870}, [%r46]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r929, %r930, %r953, %r954}, [%r47]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r819, %r820, %r875, %r876}, [%r48]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r931, %r932, %r955, %r956}, [%r49]; + .loc 1 295 36 // chunk_delta_h.py:295:36 + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r825, %r826, %r827, %r828 }, { %r801, %r802, %r803, %r804 }, { %r805, %r806 }, { %r825, %r826, %r827, %r828 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r835, %r836, %r837, %r838 }, { %r801, %r802, %r803, %r804 }, { %r807, %r808 }, { %r835, %r836, %r837, %r838 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r841, %r842, %r843, %r844 }, { %r801, %r802, %r803, %r804 }, { %r809, %r810 }, { %r841, %r842, %r843, %r844 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r847, %r848, %r849, %r850 }, { %r801, %r802, %r803, %r804 }, { %r811, %r812 }, { %r847, %r848, %r849, %r850 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r853, %r854, %r855, %r856 }, { %r801, %r802, %r803, %r804 }, { %r813, %r814 }, { %r853, %r854, %r855, %r856 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r859, %r860, %r861, %r862 }, { %r801, %r802, %r803, %r804 }, { %r815, %r816 }, { %r859, %r860, %r861, %r862 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r865, %r866, %r867, %r868 }, { %r801, %r802, %r803, %r804 }, { %r817, %r818 }, { %r865, %r866, %r867, %r868 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r871, %r872, %r873, %r874 }, { %r801, %r802, %r803, %r804 }, { %r819, %r820 }, { %r871, %r872, %r873, %r874 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r877, %r878, %r879, %r880 }, { %r821, %r822, %r823, %r824 }, { %r805, %r806 }, { %r877, %r878, %r879, %r880 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r885, %r886, %r887, %r888 }, { %r821, %r822, %r823, %r824 }, { %r807, %r808 }, { %r885, %r886, %r887, %r888 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r889, %r890, %r891, %r892 }, { %r821, %r822, %r823, %r824 }, { %r809, %r810 }, { %r889, %r890, %r891, %r892 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r893, %r894, %r895, %r896 }, { %r821, %r822, %r823, %r824 }, { %r811, %r812 }, { %r893, %r894, %r895, %r896 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r897, %r898, %r899, %r900 }, { %r821, %r822, %r823, %r824 }, { %r813, %r814 }, { %r897, %r898, %r899, %r900 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r901, %r902, %r903, %r904 }, { %r821, %r822, %r823, %r824 }, { %r815, %r816 }, { %r901, %r902, %r903, %r904 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r905, %r906, %r907, %r908 }, { %r821, %r822, %r823, %r824 }, { %r817, %r818 }, { %r905, %r906, %r907, %r908 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r909, %r910, %r911, %r912 }, { %r821, %r822, %r823, %r824 }, { %r819, %r820 }, { %r909, %r910, %r911, %r912 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r825, %r826, %r827, %r828 }, { %r829, %r830, %r831, %r832 }, { %r833, %r834 }, { %r825, %r826, %r827, %r828 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r835, %r836, %r837, %r838 }, { %r829, %r830, %r831, %r832 }, { %r839, %r840 }, { %r835, %r836, %r837, %r838 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r841, %r842, %r843, %r844 }, { %r829, %r830, %r831, %r832 }, { %r845, %r846 }, { %r841, %r842, %r843, %r844 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r847, %r848, %r849, %r850 }, { %r829, %r830, %r831, %r832 }, { %r851, %r852 }, { %r847, %r848, %r849, %r850 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r853, %r854, %r855, %r856 }, { %r829, %r830, %r831, %r832 }, { %r857, %r858 }, { %r853, %r854, %r855, %r856 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r859, %r860, %r861, %r862 }, { %r829, %r830, %r831, %r832 }, { %r863, %r864 }, { %r859, %r860, %r861, %r862 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r865, %r866, %r867, %r868 }, { %r829, %r830, %r831, %r832 }, { %r869, %r870 }, { %r865, %r866, %r867, %r868 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r871, %r872, %r873, %r874 }, { %r829, %r830, %r831, %r832 }, { %r875, %r876 }, { %r871, %r872, %r873, %r874 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r877, %r878, %r879, %r880 }, { %r881, %r882, %r883, %r884 }, { %r833, %r834 }, { %r877, %r878, %r879, %r880 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r885, %r886, %r887, %r888 }, { %r881, %r882, %r883, %r884 }, { %r839, %r840 }, { %r885, %r886, %r887, %r888 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r889, %r890, %r891, %r892 }, { %r881, %r882, %r883, %r884 }, { %r845, %r846 }, { %r889, %r890, %r891, %r892 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r893, %r894, %r895, %r896 }, { %r881, %r882, %r883, %r884 }, { %r851, %r852 }, { %r893, %r894, %r895, %r896 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r897, %r898, %r899, %r900 }, { %r881, %r882, %r883, %r884 }, { %r857, %r858 }, { %r897, %r898, %r899, %r900 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r901, %r902, %r903, %r904 }, { %r881, %r882, %r883, %r884 }, { %r863, %r864 }, { %r901, %r902, %r903, %r904 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r905, %r906, %r907, %r908 }, { %r881, %r882, %r883, %r884 }, { %r869, %r870 }, { %r905, %r906, %r907, %r908 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r909, %r910, %r911, %r912 }, { %r881, %r882, %r883, %r884 }, { %r875, %r876 }, { %r909, %r910, %r911, %r912 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r825, %r826, %r827, %r828 }, { %r913, %r914, %r915, %r916 }, { %r917, %r918 }, { %r825, %r826, %r827, %r828 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r835, %r836, %r837, %r838 }, { %r913, %r914, %r915, %r916 }, { %r919, %r920 }, { %r835, %r836, %r837, %r838 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r841, %r842, %r843, %r844 }, { %r913, %r914, %r915, %r916 }, { %r921, %r922 }, { %r841, %r842, %r843, %r844 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r847, %r848, %r849, %r850 }, { %r913, %r914, %r915, %r916 }, { %r923, %r924 }, { %r847, %r848, %r849, %r850 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r853, %r854, %r855, %r856 }, { %r913, %r914, %r915, %r916 }, { %r925, %r926 }, { %r853, %r854, %r855, %r856 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r859, %r860, %r861, %r862 }, { %r913, %r914, %r915, %r916 }, { %r927, %r928 }, { %r859, %r860, %r861, %r862 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r865, %r866, %r867, %r868 }, { %r913, %r914, %r915, %r916 }, { %r929, %r930 }, { %r865, %r866, %r867, %r868 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r871, %r872, %r873, %r874 }, { %r913, %r914, %r915, %r916 }, { %r931, %r932 }, { %r871, %r872, %r873, %r874 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r877, %r878, %r879, %r880 }, { %r933, %r934, %r935, %r936 }, { %r917, %r918 }, { %r877, %r878, %r879, %r880 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r885, %r886, %r887, %r888 }, { %r933, %r934, %r935, %r936 }, { %r919, %r920 }, { %r885, %r886, %r887, %r888 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r889, %r890, %r891, %r892 }, { %r933, %r934, %r935, %r936 }, { %r921, %r922 }, { %r889, %r890, %r891, %r892 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r893, %r894, %r895, %r896 }, { %r933, %r934, %r935, %r936 }, { %r923, %r924 }, { %r893, %r894, %r895, %r896 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r897, %r898, %r899, %r900 }, { %r933, %r934, %r935, %r936 }, { %r925, %r926 }, { %r897, %r898, %r899, %r900 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r901, %r902, %r903, %r904 }, { %r933, %r934, %r935, %r936 }, { %r927, %r928 }, { %r901, %r902, %r903, %r904 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r905, %r906, %r907, %r908 }, { %r933, %r934, %r935, %r936 }, { %r929, %r930 }, { %r905, %r906, %r907, %r908 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r909, %r910, %r911, %r912 }, { %r933, %r934, %r935, %r936 }, { %r931, %r932 }, { %r909, %r910, %r911, %r912 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r825, %r826, %r827, %r828 }, { %r937, %r938, %r939, %r940 }, { %r941, %r942 }, { %r825, %r826, %r827, %r828 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r835, %r836, %r837, %r838 }, { %r937, %r938, %r939, %r940 }, { %r943, %r944 }, { %r835, %r836, %r837, %r838 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r841, %r842, %r843, %r844 }, { %r937, %r938, %r939, %r940 }, { %r945, %r946 }, { %r841, %r842, %r843, %r844 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r847, %r848, %r849, %r850 }, { %r937, %r938, %r939, %r940 }, { %r947, %r948 }, { %r847, %r848, %r849, %r850 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r853, %r854, %r855, %r856 }, { %r937, %r938, %r939, %r940 }, { %r949, %r950 }, { %r853, %r854, %r855, %r856 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r859, %r860, %r861, %r862 }, { %r937, %r938, %r939, %r940 }, { %r951, %r952 }, { %r859, %r860, %r861, %r862 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r865, %r866, %r867, %r868 }, { %r937, %r938, %r939, %r940 }, { %r953, %r954 }, { %r865, %r866, %r867, %r868 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r871, %r872, %r873, %r874 }, { %r937, %r938, %r939, %r940 }, { %r955, %r956 }, { %r871, %r872, %r873, %r874 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r877, %r878, %r879, %r880 }, { %r957, %r958, %r959, %r960 }, { %r941, %r942 }, { %r877, %r878, %r879, %r880 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r885, %r886, %r887, %r888 }, { %r957, %r958, %r959, %r960 }, { %r943, %r944 }, { %r885, %r886, %r887, %r888 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r889, %r890, %r891, %r892 }, { %r957, %r958, %r959, %r960 }, { %r945, %r946 }, { %r889, %r890, %r891, %r892 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r893, %r894, %r895, %r896 }, { %r957, %r958, %r959, %r960 }, { %r947, %r948 }, { %r893, %r894, %r895, %r896 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r897, %r898, %r899, %r900 }, { %r957, %r958, %r959, %r960 }, { %r949, %r950 }, { %r897, %r898, %r899, %r900 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r901, %r902, %r903, %r904 }, { %r957, %r958, %r959, %r960 }, { %r951, %r952 }, { %r901, %r902, %r903, %r904 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r905, %r906, %r907, %r908 }, { %r957, %r958, %r959, %r960 }, { %r953, %r954 }, { %r905, %r906, %r907, %r908 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r909, %r910, %r911, %r912 }, { %r957, %r958, %r959, %r960 }, { %r955, %r956 }, { %r909, %r910, %r911, %r912 }; + // end inline asm + .loc 1 298 26 // chunk_delta_h.py:298:26 + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r961, %r962, %r963, %r964}, [%r1395+98304]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r973, %r974, %r975, %r976}, [%r1395+100352]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r1041, %r1042, %r1043, %r1044}, [%r1395+102400]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r1049, %r1050, %r1051, %r1052}, [%r1395+104448]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r965, %r966, %r967, %r968}, [%r1396+98304]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r1009, %r1010, %r1011, %r1012}, [%r1396+100352]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r1045, %r1046, %r1047, %r1048}, [%r1396+102400]; + ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 {%r1053, %r1054, %r1055, %r1056}, [%r1396+104448]; + .loc 1 302 40 // chunk_delta_h.py:302:40 + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r969, %r970, %r971, %r972 }, { %r961, %r962, %r963, %r964 }, { %r805, %r806 }, { %r969, %r970, %r971, %r972 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r977, %r978, %r979, %r980 }, { %r961, %r962, %r963, %r964 }, { %r807, %r808 }, { %r977, %r978, %r979, %r980 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r981, %r982, %r983, %r984 }, { %r961, %r962, %r963, %r964 }, { %r809, %r810 }, { %r981, %r982, %r983, %r984 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r985, %r986, %r987, %r988 }, { %r961, %r962, %r963, %r964 }, { %r811, %r812 }, { %r985, %r986, %r987, %r988 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r989, %r990, %r991, %r992 }, { %r961, %r962, %r963, %r964 }, { %r813, %r814 }, { %r989, %r990, %r991, %r992 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r993, %r994, %r995, %r996 }, { %r961, %r962, %r963, %r964 }, { %r815, %r816 }, { %r993, %r994, %r995, %r996 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r997, %r998, %r999, %r1000 }, { %r961, %r962, %r963, %r964 }, { %r817, %r818 }, { %r997, %r998, %r999, %r1000 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1001, %r1002, %r1003, %r1004 }, { %r961, %r962, %r963, %r964 }, { %r819, %r820 }, { %r1001, %r1002, %r1003, %r1004 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1005, %r1006, %r1007, %r1008 }, { %r965, %r966, %r967, %r968 }, { %r805, %r806 }, { %r1005, %r1006, %r1007, %r1008 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1013, %r1014, %r1015, %r1016 }, { %r965, %r966, %r967, %r968 }, { %r807, %r808 }, { %r1013, %r1014, %r1015, %r1016 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1017, %r1018, %r1019, %r1020 }, { %r965, %r966, %r967, %r968 }, { %r809, %r810 }, { %r1017, %r1018, %r1019, %r1020 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1021, %r1022, %r1023, %r1024 }, { %r965, %r966, %r967, %r968 }, { %r811, %r812 }, { %r1021, %r1022, %r1023, %r1024 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1025, %r1026, %r1027, %r1028 }, { %r965, %r966, %r967, %r968 }, { %r813, %r814 }, { %r1025, %r1026, %r1027, %r1028 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1029, %r1030, %r1031, %r1032 }, { %r965, %r966, %r967, %r968 }, { %r815, %r816 }, { %r1029, %r1030, %r1031, %r1032 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1033, %r1034, %r1035, %r1036 }, { %r965, %r966, %r967, %r968 }, { %r817, %r818 }, { %r1033, %r1034, %r1035, %r1036 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1037, %r1038, %r1039, %r1040 }, { %r965, %r966, %r967, %r968 }, { %r819, %r820 }, { %r1037, %r1038, %r1039, %r1040 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r969, %r970, %r971, %r972 }, { %r973, %r974, %r975, %r976 }, { %r833, %r834 }, { %r969, %r970, %r971, %r972 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r977, %r978, %r979, %r980 }, { %r973, %r974, %r975, %r976 }, { %r839, %r840 }, { %r977, %r978, %r979, %r980 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r981, %r982, %r983, %r984 }, { %r973, %r974, %r975, %r976 }, { %r845, %r846 }, { %r981, %r982, %r983, %r984 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r985, %r986, %r987, %r988 }, { %r973, %r974, %r975, %r976 }, { %r851, %r852 }, { %r985, %r986, %r987, %r988 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r989, %r990, %r991, %r992 }, { %r973, %r974, %r975, %r976 }, { %r857, %r858 }, { %r989, %r990, %r991, %r992 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r993, %r994, %r995, %r996 }, { %r973, %r974, %r975, %r976 }, { %r863, %r864 }, { %r993, %r994, %r995, %r996 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r997, %r998, %r999, %r1000 }, { %r973, %r974, %r975, %r976 }, { %r869, %r870 }, { %r997, %r998, %r999, %r1000 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1001, %r1002, %r1003, %r1004 }, { %r973, %r974, %r975, %r976 }, { %r875, %r876 }, { %r1001, %r1002, %r1003, %r1004 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1005, %r1006, %r1007, %r1008 }, { %r1009, %r1010, %r1011, %r1012 }, { %r833, %r834 }, { %r1005, %r1006, %r1007, %r1008 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1013, %r1014, %r1015, %r1016 }, { %r1009, %r1010, %r1011, %r1012 }, { %r839, %r840 }, { %r1013, %r1014, %r1015, %r1016 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1017, %r1018, %r1019, %r1020 }, { %r1009, %r1010, %r1011, %r1012 }, { %r845, %r846 }, { %r1017, %r1018, %r1019, %r1020 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1021, %r1022, %r1023, %r1024 }, { %r1009, %r1010, %r1011, %r1012 }, { %r851, %r852 }, { %r1021, %r1022, %r1023, %r1024 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1025, %r1026, %r1027, %r1028 }, { %r1009, %r1010, %r1011, %r1012 }, { %r857, %r858 }, { %r1025, %r1026, %r1027, %r1028 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1029, %r1030, %r1031, %r1032 }, { %r1009, %r1010, %r1011, %r1012 }, { %r863, %r864 }, { %r1029, %r1030, %r1031, %r1032 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1033, %r1034, %r1035, %r1036 }, { %r1009, %r1010, %r1011, %r1012 }, { %r869, %r870 }, { %r1033, %r1034, %r1035, %r1036 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1037, %r1038, %r1039, %r1040 }, { %r1009, %r1010, %r1011, %r1012 }, { %r875, %r876 }, { %r1037, %r1038, %r1039, %r1040 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r969, %r970, %r971, %r972 }, { %r1041, %r1042, %r1043, %r1044 }, { %r917, %r918 }, { %r969, %r970, %r971, %r972 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r977, %r978, %r979, %r980 }, { %r1041, %r1042, %r1043, %r1044 }, { %r919, %r920 }, { %r977, %r978, %r979, %r980 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r981, %r982, %r983, %r984 }, { %r1041, %r1042, %r1043, %r1044 }, { %r921, %r922 }, { %r981, %r982, %r983, %r984 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r985, %r986, %r987, %r988 }, { %r1041, %r1042, %r1043, %r1044 }, { %r923, %r924 }, { %r985, %r986, %r987, %r988 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r989, %r990, %r991, %r992 }, { %r1041, %r1042, %r1043, %r1044 }, { %r925, %r926 }, { %r989, %r990, %r991, %r992 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r993, %r994, %r995, %r996 }, { %r1041, %r1042, %r1043, %r1044 }, { %r927, %r928 }, { %r993, %r994, %r995, %r996 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r997, %r998, %r999, %r1000 }, { %r1041, %r1042, %r1043, %r1044 }, { %r929, %r930 }, { %r997, %r998, %r999, %r1000 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1001, %r1002, %r1003, %r1004 }, { %r1041, %r1042, %r1043, %r1044 }, { %r931, %r932 }, { %r1001, %r1002, %r1003, %r1004 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1005, %r1006, %r1007, %r1008 }, { %r1045, %r1046, %r1047, %r1048 }, { %r917, %r918 }, { %r1005, %r1006, %r1007, %r1008 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1013, %r1014, %r1015, %r1016 }, { %r1045, %r1046, %r1047, %r1048 }, { %r919, %r920 }, { %r1013, %r1014, %r1015, %r1016 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1017, %r1018, %r1019, %r1020 }, { %r1045, %r1046, %r1047, %r1048 }, { %r921, %r922 }, { %r1017, %r1018, %r1019, %r1020 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1021, %r1022, %r1023, %r1024 }, { %r1045, %r1046, %r1047, %r1048 }, { %r923, %r924 }, { %r1021, %r1022, %r1023, %r1024 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1025, %r1026, %r1027, %r1028 }, { %r1045, %r1046, %r1047, %r1048 }, { %r925, %r926 }, { %r1025, %r1026, %r1027, %r1028 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1029, %r1030, %r1031, %r1032 }, { %r1045, %r1046, %r1047, %r1048 }, { %r927, %r928 }, { %r1029, %r1030, %r1031, %r1032 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1033, %r1034, %r1035, %r1036 }, { %r1045, %r1046, %r1047, %r1048 }, { %r929, %r930 }, { %r1033, %r1034, %r1035, %r1036 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1037, %r1038, %r1039, %r1040 }, { %r1045, %r1046, %r1047, %r1048 }, { %r931, %r932 }, { %r1037, %r1038, %r1039, %r1040 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r969, %r970, %r971, %r972 }, { %r1049, %r1050, %r1051, %r1052 }, { %r941, %r942 }, { %r969, %r970, %r971, %r972 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r977, %r978, %r979, %r980 }, { %r1049, %r1050, %r1051, %r1052 }, { %r943, %r944 }, { %r977, %r978, %r979, %r980 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r981, %r982, %r983, %r984 }, { %r1049, %r1050, %r1051, %r1052 }, { %r945, %r946 }, { %r981, %r982, %r983, %r984 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r985, %r986, %r987, %r988 }, { %r1049, %r1050, %r1051, %r1052 }, { %r947, %r948 }, { %r985, %r986, %r987, %r988 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r989, %r990, %r991, %r992 }, { %r1049, %r1050, %r1051, %r1052 }, { %r949, %r950 }, { %r989, %r990, %r991, %r992 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r993, %r994, %r995, %r996 }, { %r1049, %r1050, %r1051, %r1052 }, { %r951, %r952 }, { %r993, %r994, %r995, %r996 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r997, %r998, %r999, %r1000 }, { %r1049, %r1050, %r1051, %r1052 }, { %r953, %r954 }, { %r997, %r998, %r999, %r1000 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1001, %r1002, %r1003, %r1004 }, { %r1049, %r1050, %r1051, %r1052 }, { %r955, %r956 }, { %r1001, %r1002, %r1003, %r1004 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1005, %r1006, %r1007, %r1008 }, { %r1053, %r1054, %r1055, %r1056 }, { %r941, %r942 }, { %r1005, %r1006, %r1007, %r1008 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1013, %r1014, %r1015, %r1016 }, { %r1053, %r1054, %r1055, %r1056 }, { %r943, %r944 }, { %r1013, %r1014, %r1015, %r1016 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1017, %r1018, %r1019, %r1020 }, { %r1053, %r1054, %r1055, %r1056 }, { %r945, %r946 }, { %r1017, %r1018, %r1019, %r1020 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1021, %r1022, %r1023, %r1024 }, { %r1053, %r1054, %r1055, %r1056 }, { %r947, %r948 }, { %r1021, %r1022, %r1023, %r1024 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1025, %r1026, %r1027, %r1028 }, { %r1053, %r1054, %r1055, %r1056 }, { %r949, %r950 }, { %r1025, %r1026, %r1027, %r1028 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1029, %r1030, %r1031, %r1032 }, { %r1053, %r1054, %r1055, %r1056 }, { %r951, %r952 }, { %r1029, %r1030, %r1031, %r1032 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1033, %r1034, %r1035, %r1036 }, { %r1053, %r1054, %r1055, %r1056 }, { %r953, %r954 }, { %r1033, %r1034, %r1035, %r1036 }; + // end inline asm + // begin inline asm + mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 { %r1037, %r1038, %r1039, %r1040 }, { %r1053, %r1054, %r1055, %r1056 }, { %r955, %r956 }, { %r1037, %r1038, %r1039, %r1040 }; + // end inline asm + .loc 1 175 21 // chunk_delta_h.py:175:21 + add.s32 %r1397, %r1419, 1; + setp.gt.s32 %p25, %r1397, 2; + selp.b32 %r1419, 0, %r1397, %p25; + .loc 1 177 31 // chunk_delta_h.py:177:31 + add.s64 %rd453, %rd473, 192; + add.s64 %rd454, %rd4, %rd473; + add.s64 %rd455, %rd454, 192; + add.s64 %rd456, %rd454, 200; + add.s64 %rd457, %rd454, 208; + add.s64 %rd458, %rd454, 216; + add.s64 %rd459, %rd454, 224; + add.s64 %rd460, %rd454, 232; + add.s64 %rd461, %rd454, 240; + add.s64 %rd462, %rd454, 248; + or.b64 %rd20, %rd453, %rd8; + or.b64 %rd19, %rd453, %rd7; + or.b64 %rd21, %rd453, %rd9; + or.b64 %rd22, %rd453, %rd10; + .loc 1 178 20 // chunk_delta_h.py:178:20 + add.s64 %rd463, %rd17, %rd473; + setp.lt.s64 %p35, %rd455, %rd3; + setp.lt.s64 %p34, %rd456, %rd3; + setp.lt.s64 %p33, %rd457, %rd3; + setp.lt.s64 %p32, %rd458, %rd3; + setp.lt.s64 %p31, %rd459, %rd3; + setp.lt.s64 %p30, %rd460, %rd3; + setp.lt.s64 %p29, %rd461, %rd3; + setp.lt.s64 %p28, %rd462, %rd3; + setp.lt.s64 %p26, %rd463, %rd3; + .loc 1 211 42 // chunk_delta_h.py:211:42 + add.s64 %rd464, %rd469, %rd18; + add.s64 %rd316, %rd464, 1572864; + add.s64 %rd317, %rd464, 1638400; + add.s64 %rd318, %rd464, 1703936; + add.s64 %rd319, %rd464, 1769472; + add.s64 %rd320, %rd464, 1835008; + add.s64 %rd321, %rd464, 1900544; + add.s64 %rd322, %rd464, 1966080; + .loc 1 212 22 // chunk_delta_h.py:212:22 + add.s64 %rd323, %rd464, 2031616; + shl.b32 %r1398, %r1419, 13; + add.s32 %r1399, %r246, %r1398; + add.s32 %r1057, %r1399, %r8; + selp.b32 %r1400, 16, 0, %p35; + selp.b32 %r1066, %r1400, 0, %p23; + // begin inline asm + cp.async.cg.shared.global [ %r1057 + 0 ], [ %rd316 + 0 ], 0x10, %r1066; + // end inline asm + add.s32 %r1058, %r1057, 1024; + selp.b32 %r1401, 16, 0, %p34; + selp.b32 %r1068, %r1401, 0, %p23; + // begin inline asm + cp.async.cg.shared.global [ %r1058 + 0 ], [ %rd317 + 0 ], 0x10, %r1068; + // end inline asm + add.s32 %r1059, %r1057, 2048; + selp.b32 %r1402, 16, 0, %p33; + selp.b32 %r1070, %r1402, 0, %p23; + // begin inline asm + cp.async.cg.shared.global [ %r1059 + 0 ], [ %rd318 + 0 ], 0x10, %r1070; + // end inline asm + add.s32 %r1060, %r1057, 3072; + selp.b32 %r1403, 16, 0, %p32; + selp.b32 %r1072, %r1403, 0, %p23; + // begin inline asm + cp.async.cg.shared.global [ %r1060 + 0 ], [ %rd319 + 0 ], 0x10, %r1072; + // end inline asm + add.s32 %r1061, %r1057, 4096; + selp.b32 %r1404, 16, 0, %p31; + selp.b32 %r1074, %r1404, 0, %p23; + // begin inline asm + cp.async.cg.shared.global [ %r1061 + 0 ], [ %rd320 + 0 ], 0x10, %r1074; + // end inline asm + add.s32 %r1062, %r1057, 5120; + selp.b32 %r1405, 16, 0, %p30; + selp.b32 %r1076, %r1405, 0, %p23; + // begin inline asm + cp.async.cg.shared.global [ %r1062 + 0 ], [ %rd321 + 0 ], 0x10, %r1076; + // end inline asm + add.s32 %r1063, %r1057, 6144; + selp.b32 %r1406, 16, 0, %p29; + selp.b32 %r1078, %r1406, 0, %p23; + // begin inline asm + cp.async.cg.shared.global [ %r1063 + 0 ], [ %rd322 + 0 ], 0x10, %r1078; + // end inline asm + add.s32 %r1064, %r1057, 7168; + selp.b32 %r1407, 16, 0, %p28; + selp.b32 %r1080, %r1407, 0, %p23; + // begin inline asm + cp.async.cg.shared.global [ %r1064 + 0 ], [ %rd323 + 0 ], 0x10, %r1080; + // end inline asm + cp.async.commit_group; + .loc 1 218 46 // chunk_delta_h.py:218:46 + add.s64 %rd324, %rd464, 1572992; + add.s64 %rd325, %rd464, 1638528; + add.s64 %rd326, %rd464, 1704064; + add.s64 %rd327, %rd464, 1769600; + add.s64 %rd328, %rd464, 1835136; + add.s64 %rd329, %rd464, 1900672; + add.s64 %rd330, %rd464, 1966208; + .loc 1 219 26 // chunk_delta_h.py:219:26 + add.s64 %rd331, %rd464, 2031744; + add.s32 %r1065, %r1057, 24576; + // begin inline asm + cp.async.cg.shared.global [ %r1065 + 0 ], [ %rd324 + 0 ], 0x10, %r1066; + // end inline asm + add.s32 %r1067, %r1057, 25600; + // begin inline asm + cp.async.cg.shared.global [ %r1067 + 0 ], [ %rd325 + 0 ], 0x10, %r1068; + // end inline asm + add.s32 %r1069, %r1057, 26624; + // begin inline asm + cp.async.cg.shared.global [ %r1069 + 0 ], [ %rd326 + 0 ], 0x10, %r1070; + // end inline asm + add.s32 %r1071, %r1057, 27648; + // begin inline asm + cp.async.cg.shared.global [ %r1071 + 0 ], [ %rd327 + 0 ], 0x10, %r1072; + // end inline asm + add.s32 %r1073, %r1057, 28672; + // begin inline asm + cp.async.cg.shared.global [ %r1073 + 0 ], [ %rd328 + 0 ], 0x10, %r1074; + // end inline asm + add.s32 %r1075, %r1057, 29696; + // begin inline asm + cp.async.cg.shared.global [ %r1075 + 0 ], [ %rd329 + 0 ], 0x10, %r1076; + // end inline asm + add.s32 %r1077, %r1057, 30720; + // begin inline asm + cp.async.cg.shared.global [ %r1077 + 0 ], [ %rd330 + 0 ], 0x10, %r1078; + // end inline asm + add.s32 %r1079, %r1057, 31744; + // begin inline asm + cp.async.cg.shared.global [ %r1079 + 0 ], [ %rd331 + 0 ], 0x10, %r1080; + // end inline asm + cp.async.commit_group; + .loc 1 238 42 // chunk_delta_h.py:238:42 + add.s64 %rd465, %rd470, %rd18; + add.s64 %rd332, %rd465, 1572864; + add.s64 %rd333, %rd465, 1638400; + add.s64 %rd334, %rd465, 1703936; + add.s64 %rd335, %rd465, 1769472; + add.s64 %rd336, %rd465, 1835008; + add.s64 %rd337, %rd465, 1900544; + add.s64 %rd338, %rd465, 1966080; + .loc 1 239 22 // chunk_delta_h.py:239:22 + add.s64 %rd339, %rd465, 2031616; + add.s32 %r1081, %r1057, 49152; + // begin inline asm + cp.async.cg.shared.global [ %r1081 + 0 ], [ %rd332 + 0 ], 0x10, %r1066; + // end inline asm + add.s32 %r1082, %r1057, 50176; + // begin inline asm + cp.async.cg.shared.global [ %r1082 + 0 ], [ %rd333 + 0 ], 0x10, %r1068; + // end inline asm + add.s32 %r1083, %r1057, 51200; + // begin inline asm + cp.async.cg.shared.global [ %r1083 + 0 ], [ %rd334 + 0 ], 0x10, %r1070; + // end inline asm + add.s32 %r1084, %r1057, 52224; + // begin inline asm + cp.async.cg.shared.global [ %r1084 + 0 ], [ %rd335 + 0 ], 0x10, %r1072; + // end inline asm + add.s32 %r1085, %r1057, 53248; + // begin inline asm + cp.async.cg.shared.global [ %r1085 + 0 ], [ %rd336 + 0 ], 0x10, %r1074; + // end inline asm + add.s32 %r1086, %r1057, 54272; + // begin inline asm + cp.async.cg.shared.global [ %r1086 + 0 ], [ %rd337 + 0 ], 0x10, %r1076; + // end inline asm + add.s32 %r1087, %r1057, 55296; + // begin inline asm + cp.async.cg.shared.global [ %r1087 + 0 ], [ %rd338 + 0 ], 0x10, %r1078; + // end inline asm + add.s32 %r1088, %r1057, 56320; + // begin inline asm + cp.async.cg.shared.global [ %r1088 + 0 ], [ %rd339 + 0 ], 0x10, %r1080; + // end inline asm + cp.async.commit_group; + cvt.u32.u64 %r1408, %rd473; + add.s32 %r1409, %r1408, 256; + .loc 1 245 39 // chunk_delta_h.py:245:39 + min.s32 %r1410, %r1409, %r71; + .loc 1 247 58 // chunk_delta_h.py:247:58 + shl.b32 %r1411, %r1410, 5; + add.s32 %r1412, %r1411, -32; + .loc 1 247 35 // chunk_delta_h.py:247:35 + mad.wide.s32 %rd340, %r1412, 4, %rd13; + .loc 1 247 31 // chunk_delta_h.py:247:31 + shl.b32 %r1413, %r1419, 2; + add.s32 %r1414, %r246, %r1413; + add.s32 %r1089, %r1414, 142080; + selp.b32 %r1090, 4, 0, %p23; + // begin inline asm + @%p22 cp.async.ca.shared.global [ %r1089 + 0 ], [ %rd340 + 0 ], 0x4, %r1090; + // end inline asm + cp.async.commit_group; + .loc 1 249 26 // chunk_delta_h.py:249:26 + shl.b32 %r1415, %r1419, 8; + add.s32 %r1416, %r9, %r1415; + add.s32 %r1091, %r1416, 141312; + selp.b32 %r1417, 4, 0, %p26; + selp.b32 %r1092, %r1417, 0, %p23; + // begin inline asm + cp.async.ca.shared.global [ %r1091 + 0 ], [ %rd472 + 0 ], 0x4, %r1092; + // end inline asm + cp.async.commit_group; + .loc 1 290 34 // chunk_delta_h.py:290:34 + add.s64 %rd466, %rd471, %rd18; + add.s64 %rd341, %rd466, 1572864; + add.s64 %rd342, %rd466, 1638400; + add.s64 %rd343, %rd466, 1703936; + add.s64 %rd344, %rd466, 1769472; + add.s64 %rd345, %rd466, 1835008; + add.s64 %rd346, %rd466, 1900544; + add.s64 %rd347, %rd466, 1966080; + .loc 1 291 22 // chunk_delta_h.py:291:22 + add.s64 %rd348, %rd466, 2031616; + add.s32 %r1093, %r1057, 73728; + // begin inline asm + cp.async.cg.shared.global [ %r1093 + 0 ], [ %rd341 + 0 ], 0x10, %r1066; + // end inline asm + add.s32 %r1094, %r1057, 74752; + // begin inline asm + cp.async.cg.shared.global [ %r1094 + 0 ], [ %rd342 + 0 ], 0x10, %r1068; + // end inline asm + add.s32 %r1095, %r1057, 75776; + // begin inline asm + cp.async.cg.shared.global [ %r1095 + 0 ], [ %rd343 + 0 ], 0x10, %r1070; + // end inline asm + add.s32 %r1096, %r1057, 76800; + // begin inline asm + cp.async.cg.shared.global [ %r1096 + 0 ], [ %rd344 + 0 ], 0x10, %r1072; + // end inline asm + add.s32 %r1097, %r1057, 77824; + // begin inline asm + cp.async.cg.shared.global [ %r1097 + 0 ], [ %rd345 + 0 ], 0x10, %r1074; + // end inline asm + add.s32 %r1098, %r1057, 78848; + // begin inline asm + cp.async.cg.shared.global [ %r1098 + 0 ], [ %rd346 + 0 ], 0x10, %r1076; + // end inline asm + add.s32 %r1099, %r1057, 79872; + // begin inline asm + cp.async.cg.shared.global [ %r1099 + 0 ], [ %rd347 + 0 ], 0x10, %r1078; + // end inline asm + add.s32 %r1100, %r1057, 80896; + // begin inline asm + cp.async.cg.shared.global [ %r1100 + 0 ], [ %rd348 + 0 ], 0x10, %r1080; + // end inline asm + cp.async.commit_group; + .loc 1 297 38 // chunk_delta_h.py:297:38 + add.s64 %rd349, %rd466, 1572992; + add.s64 %rd350, %rd466, 1638528; + add.s64 %rd351, %rd466, 1704064; + add.s64 %rd352, %rd466, 1769600; + add.s64 %rd353, %rd466, 1835136; + add.s64 %rd354, %rd466, 1900672; + add.s64 %rd355, %rd466, 1966208; + .loc 1 298 26 // chunk_delta_h.py:298:26 + add.s64 %rd356, %rd466, 2031744; + bar.sync 0; + add.s32 %r1101, %r1057, 98304; + // begin inline asm + cp.async.cg.shared.global [ %r1101 + 0 ], [ %rd349 + 0 ], 0x10, %r1066; + // end inline asm + add.s32 %r1102, %r1057, 99328; + // begin inline asm + cp.async.cg.shared.global [ %r1102 + 0 ], [ %rd350 + 0 ], 0x10, %r1068; + // end inline asm + add.s32 %r1103, %r1057, 100352; + // begin inline asm + cp.async.cg.shared.global [ %r1103 + 0 ], [ %rd351 + 0 ], 0x10, %r1070; + // end inline asm + add.s32 %r1104, %r1057, 101376; + // begin inline asm + cp.async.cg.shared.global [ %r1104 + 0 ], [ %rd352 + 0 ], 0x10, %r1072; + // end inline asm + add.s32 %r1105, %r1057, 102400; + // begin inline asm + cp.async.cg.shared.global [ %r1105 + 0 ], [ %rd353 + 0 ], 0x10, %r1074; + // end inline asm + add.s32 %r1106, %r1057, 103424; + // begin inline asm + cp.async.cg.shared.global [ %r1106 + 0 ], [ %rd354 + 0 ], 0x10, %r1076; + // end inline asm + add.s32 %r1107, %r1057, 104448; + // begin inline asm + cp.async.cg.shared.global [ %r1107 + 0 ], [ %rd355 + 0 ], 0x10, %r1078; + // end inline asm + add.s32 %r1108, %r1057, 105472; + // begin inline asm + cp.async.cg.shared.global [ %r1108 + 0 ], [ %rd356 + 0 ], 0x10, %r1080; + // end inline asm + cp.async.commit_group; + .loc 1 175 21 // chunk_delta_h.py:175:21 + add.s64 %rd474, %rd474, 1; + add.s64 %rd473, %rd473, 64; + add.s64 %rd472, %rd472, 8192; + add.s64 %rd471, %rd471, 524288; + add.s64 %rd470, %rd470, 524288; + add.s64 %rd469, %rd469, 524288; + add.s64 %rd468, %rd468, 524288; + add.s64 %rd467, %rd467, 1048576; + setp.ne.b64 %p27, %rd16, %rd473; + mov.pred %p52, %p9; + mov.pred %p53, %p10; + mov.pred %p54, %p11; + mov.pred %p55, %p12; + mov.b64 %rd475, %rd479; + mov.b64 %rd476, %rd480; + mov.b64 %rd477, %rd481; + mov.b64 %rd478, %rd482; + mov.b64 %rd479, %rd19; + mov.b64 %rd480, %rd20; + mov.b64 %rd481, %rd21; + mov.b64 %rd482, %rd22; + @%p27 bra $L__BB0_2; +$L__BB0_3: // %._crit_edge + cp.async.wait_group 0; + bar.sync 0; + .loc 1 318 4 // chunk_delta_h.py:318:4 + ret; +$L__tmp9: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .file 3 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 252 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xf5 DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 99 // DW_AT_name +.b8 104 +.b8 117 +.b8 110 +.b8 107 +.b8 95 +.b8 100 +.b8 101 +.b8 108 +.b8 116 +.b8 97 +.b8 95 +.b8 104 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 111 +.b8 112 +.b8 115 +.b8 47 +.b8 99 +.b8 111 +.b8 109 +.b8 109 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // Abbrev [2] 0x70:0x31 DW_TAG_subprogram +.b8 99 // DW_AT_name +.b8 104 +.b8 117 +.b8 110 +.b8 107 +.b8 95 +.b8 103 +.b8 97 +.b8 116 +.b8 101 +.b8 100 +.b8 95 +.b8 100 +.b8 101 +.b8 108 +.b8 116 +.b8 97 +.b8 95 +.b8 114 +.b8 117 +.b8 108 +.b8 101 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 95 +.b8 104 +.b8 95 +.b8 98 +.b8 108 +.b8 111 +.b8 99 +.b8 107 +.b8 100 +.b8 105 +.b8 109 +.b8 54 +.b8 52 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0xa1:0x5e DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 112 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0xb6:0x18 DW_TAG_inlined_subroutine +.b32 112 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp4 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 97 // DW_AT_call_line +.b8 24 // DW_AT_call_column +.b8 4 // Abbrev [4] 0xce:0x18 DW_TAG_inlined_subroutine +.b32 112 // DW_AT_abstract_origin +.b64 $L__tmp5 // DW_AT_low_pc +.b64 $L__tmp6 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 250 // DW_AT_call_line +.b8 43 // DW_AT_call_column +.b8 4 // Abbrev [4] 0xe6:0x18 DW_TAG_inlined_subroutine +.b32 112 // DW_AT_abstract_origin +.b64 $L__tmp7 // DW_AT_low_pc +.b64 $L__tmp8 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 251 // DW_AT_call_line +.b8 28 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.source b/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.source new file mode 100644 index 0000000000000000000000000000000000000000..b459612ea14d65194b793756e8f4ebbb8ab53202 --- /dev/null +++ b/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.source @@ -0,0 +1,849 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":59:0) +#loc173 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":34:0) +#loc182 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:0) +#loc186 = loc("k"(#loc)) +#loc187 = loc("v"(#loc)) +#loc188 = loc("w"(#loc)) +#loc189 = loc("v_new"(#loc)) +#loc190 = loc("g"(#loc)) +#loc191 = loc("h"(#loc)) +#loc192 = loc("T"(#loc)) +#loc347 = loc("x"(#loc173)) +#loc348 = loc("x"(#loc182)) +module { + tt.func public @chunk_gated_delta_rule_fwd_kernel_h_blockdim64(%k: !tt.ptr {tt.divisibility = 16 : i32} loc("k"(#loc)), %v: !tt.ptr {tt.divisibility = 16 : i32} loc("v"(#loc)), %w: !tt.ptr {tt.divisibility = 16 : i32} loc("w"(#loc)), %v_new: !tt.ptr {tt.divisibility = 16 : i32} loc("v_new"(#loc)), %g: !tt.ptr {tt.divisibility = 16 : i32} loc("g"(#loc)), %h: !tt.ptr {tt.divisibility = 16 : i32} loc("h"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %pid = tt.get_program_id x : i32 loc(#loc193) + %pid_0 = arith.extsi %pid : i32 to i64 loc(#loc194) + tt.call @triton.language.standard.cdiv__c128_c64() : () -> () loc(#loc195) + %NV = arith.constant 2 : i32 loc(#loc196) + %0 = arith.extsi %NV : i32 to i64 loc(#loc5) + %1 = arith.remsi %pid_0, %0 : i64 loc(#loc5) + %2 = arith.extsi %NV : i32 to i64 loc(#loc6) + %3 = arith.divsi %pid_0, %2 : i64 loc(#loc6) + %c32_i32 = arith.constant 32 : i32 loc(#loc7) + %c32_i64 = arith.constant 32 : i64 loc(#loc7) + %4 = arith.divsi %3, %c32_i64 : i64 loc(#loc7) + %c32_i32_1 = arith.constant 32 : i32 loc(#loc8) + %c32_i64_2 = arith.constant 32 : i64 loc(#loc8) + %5 = arith.remsi %3, %c32_i64_2 : i64 loc(#loc8) + %6 = arith.extsi %T : i32 to i64 loc(#loc9) + %7 = arith.muli %4, %6 : i64 loc(#loc9) + %8 = arith.extsi %T : i32 to i64 loc(#loc10) + %9 = arith.muli %4, %8 : i64 loc(#loc10) + %10 = arith.extsi %T : i32 to i64 loc(#loc11) + %11 = arith.addi %9, %10 : i64 loc(#loc11) + %NT = tt.call @triton.language.standard.cdiv__i32_c64(%T) : (i32) -> i32 loc(#loc197) + %boh = arith.extsi %NT : i32 to i64 loc(#loc198) + %boh_3 = arith.muli %4, %boh : i64 loc(#loc198) + %b_h1 = tt.call @triton.language.standard.zeros__Tc64_c64T_cfp32() : () -> tensor<64x64xf32> loc(#loc199) + %b_h2 = tt.call @triton.language.standard.zeros__Tc64_c64T_cfp32() : () -> tensor<64x64xf32> loc(#loc200) + %h_4 = arith.constant 32 : i32 loc(#loc201) + %h_5 = arith.constant 32 : i64 loc(#loc201) + %h_6 = arith.muli %boh_3, %h_5 : i64 loc(#loc201) + %h_7 = arith.addi %h_6, %5 : i64 loc(#loc202) + %h_8 = arith.constant 128 : i32 loc(#loc203) + %h_9 = arith.constant 128 : i64 loc(#loc203) + %h_10 = arith.muli %h_7, %h_9 : i64 loc(#loc203) + %h_11 = arith.constant 128 : i32 loc(#loc204) + %h_12 = arith.constant 128 : i64 loc(#loc204) + %h_13 = arith.muli %h_10, %h_12 : i64 loc(#loc204) + %h_14 = tt.addptr %h, %h_13 : !tt.ptr, i64 loc(#loc205) + %v_15 = arith.constant 32 : i32 loc(#loc206) + %v_16 = arith.constant 32 : i64 loc(#loc206) + %v_17 = arith.muli %7, %v_16 : i64 loc(#loc206) + %v_18 = arith.addi %v_17, %5 : i64 loc(#loc207) + %v_19 = arith.constant 128 : i32 loc(#loc208) + %v_20 = arith.constant 128 : i64 loc(#loc208) + %v_21 = arith.muli %v_18, %v_20 : i64 loc(#loc208) + %v_22 = tt.addptr %v, %v_21 : !tt.ptr, i64 loc(#loc209) + %k_23 = arith.constant 32 : i32 loc(#loc210) + %k_24 = arith.constant 32 : i64 loc(#loc210) + %k_25 = arith.muli %7, %k_24 : i64 loc(#loc210) + %k_26 = arith.constant 1 : i32 loc(#loc211) + %k_27 = arith.constant 1 : i64 loc(#loc211) + %k_28 = arith.divsi %5, %k_27 : i64 loc(#loc211) + %k_29 = arith.addi %k_25, %k_28 : i64 loc(#loc212) + %k_30 = arith.constant 128 : i32 loc(#loc213) + %k_31 = arith.constant 128 : i64 loc(#loc213) + %k_32 = arith.muli %k_29, %k_31 : i64 loc(#loc213) + %k_33 = tt.addptr %k, %k_32 : !tt.ptr, i64 loc(#loc214) + %w_34 = arith.constant 32 : i32 loc(#loc215) + %w_35 = arith.constant 32 : i64 loc(#loc215) + %w_36 = arith.muli %7, %w_35 : i64 loc(#loc215) + %w_37 = arith.addi %w_36, %5 : i64 loc(#loc216) + %w_38 = arith.constant 128 : i32 loc(#loc217) + %w_39 = arith.constant 128 : i64 loc(#loc217) + %w_40 = arith.muli %w_37, %w_39 : i64 loc(#loc217) + %w_41 = tt.addptr %w, %w_40 : !tt.ptr, i64 loc(#loc218) + %v_new_42 = arith.constant 32 : i32 loc(#loc219) + %v_new_43 = arith.constant 32 : i64 loc(#loc219) + %v_new_44 = arith.muli %7, %v_new_43 : i64 loc(#loc219) + %v_new_45 = arith.addi %v_new_44, %5 : i64 loc(#loc220) + %v_new_46 = arith.constant 128 : i32 loc(#loc221) + %v_new_47 = arith.constant 128 : i64 loc(#loc221) + %v_new_48 = arith.muli %v_new_45, %v_new_47 : i64 loc(#loc221) + %v_new_49 = tt.addptr %v_new, %v_new_48 : !tt.ptr, i64 loc(#loc222) + %o_v = arith.constant 64 : i32 loc(#loc223) + %o_v_50 = arith.constant 64 : i64 loc(#loc223) + %o_v_51 = arith.muli %1, %o_v_50 : i64 loc(#loc223) + %o_v_52 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc224) + %o_v_53 = arith.extsi %o_v_52 : tensor<64xi32> to tensor<64xi64> loc(#loc225) + %o_v_54 = tt.splat %o_v_51 : i64 -> tensor<64xi64> loc(#loc225) + %o_v_55 = arith.addi %o_v_54, %o_v_53 : tensor<64xi64> loc(#loc225) + %m_v = arith.constant 128 : i32 loc(#loc226) + %m_v_56 = arith.extsi %m_v : i32 to i64 loc(#loc226) + %m_v_57 = tt.splat %m_v_56 : i64 -> tensor<64xi64> loc(#loc226) + %m_v_58 = arith.cmpi slt, %o_v_55, %m_v_57 : tensor<64xi64> loc(#loc226) + %o_k1 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc227) + %m_k1 = arith.constant 128 : i32 loc(#loc228) + %m_k1_59 = arith.constant dense<128> : tensor<64xi32> loc(#loc228) + %m_k1_60 = arith.cmpi slt, %o_k1, %m_k1_59 : tensor<64xi32> loc(#loc228) + %o_k2 = arith.constant 64 : i32 loc(#loc229) + %o_k2_61 = arith.constant 64 : i32 loc(#loc229) + %o_k2_62 = arith.constant dense<64> : tensor<64xi32> loc(#loc229) + %o_k2_63 = arith.extsi %o_k2_62 : tensor<64xi32> to tensor<64xi64> loc(#loc229) + %o_k2_64 = arith.extsi %o_k1 : tensor<64xi32> to tensor<64xi64> loc(#loc229) + %o_k2_65 = arith.addi %o_k2_63, %o_k2_64 : tensor<64xi64> loc(#loc229) + %o_k2_66 = arith.constant 2147483647 : i64 loc(#loc229) + %o_k2_67 = arith.constant -2147483648 : i64 loc(#loc229) + %o_k2_68 = arith.constant dense<2147483647> : tensor<64xi64> loc(#loc229) + %o_k2_69 = arith.cmpi sle, %o_k2_65, %o_k2_68 : tensor<64xi64> loc(#loc229) + %o_k2_70 = arith.constant dense<-2147483648> : tensor<64xi64> loc(#loc229) + %o_k2_71 = arith.cmpi sge, %o_k2_65, %o_k2_70 : tensor<64xi64> loc(#loc229) + %o_k2_72 = arith.andi %o_k2_69, %o_k2_71 : tensor<64xi1> loc(#loc229) + %o_k2_73 = arith.addi %o_k2_62, %o_k1 : tensor<64xi32> loc(#loc229) + %m_k2 = arith.constant 128 : i32 loc(#loc230) + %m_k2_74 = arith.constant dense<128> : tensor<64xi32> loc(#loc230) + %m_k2_75 = arith.cmpi slt, %o_k2_73, %m_k2_74 : tensor<64xi32> loc(#loc230) + %o_k3 = arith.constant 128 : i32 loc(#loc231) + %o_k3_76 = arith.constant 128 : i32 loc(#loc231) + %o_k3_77 = arith.constant dense<128> : tensor<64xi32> loc(#loc231) + %o_k3_78 = arith.extsi %o_k3_77 : tensor<64xi32> to tensor<64xi64> loc(#loc231) + %o_k3_79 = arith.extsi %o_k1 : tensor<64xi32> to tensor<64xi64> loc(#loc231) + %o_k3_80 = arith.addi %o_k3_78, %o_k3_79 : tensor<64xi64> loc(#loc231) + %o_k3_81 = arith.constant 2147483647 : i64 loc(#loc231) + %o_k3_82 = arith.constant -2147483648 : i64 loc(#loc231) + %o_k3_83 = arith.constant dense<2147483647> : tensor<64xi64> loc(#loc231) + %o_k3_84 = arith.cmpi sle, %o_k3_80, %o_k3_83 : tensor<64xi64> loc(#loc231) + %o_k3_85 = arith.constant dense<-2147483648> : tensor<64xi64> loc(#loc231) + %o_k3_86 = arith.cmpi sge, %o_k3_80, %o_k3_85 : tensor<64xi64> loc(#loc231) + %o_k3_87 = arith.andi %o_k3_84, %o_k3_86 : tensor<64xi1> loc(#loc231) + %o_k3_88 = arith.addi %o_k3_77, %o_k1 : tensor<64xi32> loc(#loc231) + %m_k3 = arith.constant 128 : i32 loc(#loc232) + %m_k3_89 = arith.constant dense<128> : tensor<64xi32> loc(#loc232) + %m_k3_90 = arith.cmpi slt, %o_k3_88, %m_k3_89 : tensor<64xi32> loc(#loc232) + %o_k4 = arith.constant 192 : i32 loc(#loc233) + %o_k4_91 = arith.constant 192 : i32 loc(#loc233) + %o_k4_92 = arith.constant dense<192> : tensor<64xi32> loc(#loc233) + %o_k4_93 = arith.extsi %o_k4_92 : tensor<64xi32> to tensor<64xi64> loc(#loc233) + %o_k4_94 = arith.extsi %o_k1 : tensor<64xi32> to tensor<64xi64> loc(#loc233) + %o_k4_95 = arith.addi %o_k4_93, %o_k4_94 : tensor<64xi64> loc(#loc233) + %o_k4_96 = arith.constant 2147483647 : i64 loc(#loc233) + %o_k4_97 = arith.constant -2147483648 : i64 loc(#loc233) + %o_k4_98 = arith.constant dense<2147483647> : tensor<64xi64> loc(#loc233) + %o_k4_99 = arith.cmpi sle, %o_k4_95, %o_k4_98 : tensor<64xi64> loc(#loc233) + %o_k4_100 = arith.constant dense<-2147483648> : tensor<64xi64> loc(#loc233) + %o_k4_101 = arith.cmpi sge, %o_k4_95, %o_k4_100 : tensor<64xi64> loc(#loc233) + %o_k4_102 = arith.andi %o_k4_99, %o_k4_101 : tensor<64xi1> loc(#loc233) + %o_k4_103 = arith.addi %o_k4_92, %o_k1 : tensor<64xi32> loc(#loc233) + %m_k4 = arith.constant 128 : i32 loc(#loc234) + %m_k4_104 = arith.constant dense<128> : tensor<64xi32> loc(#loc234) + %m_k4_105 = arith.cmpi slt, %o_k4_103, %m_k4_104 : tensor<64xi32> loc(#loc234) + %c0_i32 = arith.constant 0 : i32 loc(#loc50) + %c1_i32 = arith.constant 1 : i32 loc(#loc50) + %12 = arith.bitcast %c0_i32 : i32 to i32 loc(#loc50) + %13 = arith.bitcast %NT : i32 to i32 loc(#loc50) + %14 = arith.bitcast %c1_i32 : i32 to i32 loc(#loc50) + %15 = ub.poison : i32 loc(#loc50) + %b_h2_106:2 = scf.for %i_t = %12 to %13 step %14 iter_args(%b_h1_107 = %b_h1, %b_h2_108 = %b_h2) -> (tensor<64x64xf32>, tensor<64x64xf32>) : i32 { + %i_t_int64 = arith.extsi %i_t : i32 to i64 loc(#loc236) + %o_t = arith.constant 64 : i32 loc(#loc237) + %o_t_109 = arith.constant 64 : i64 loc(#loc237) + %o_t_110 = arith.muli %i_t_int64, %o_t_109 : i64 loc(#loc237) + %o_t_111 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc238) + %o_t_112 = arith.extsi %o_t_111 : tensor<64xi32> to tensor<64xi64> loc(#loc239) + %o_t_113 = tt.splat %o_t_110 : i64 -> tensor<64xi64> loc(#loc239) + %o_t_114 = arith.addi %o_t_113, %o_t_112 : tensor<64xi64> loc(#loc239) + %m_t = arith.extsi %T : i32 to i64 loc(#loc240) + %m_t_115 = tt.splat %m_t : i64 -> tensor<64xi64> loc(#loc240) + %m_t_116 = arith.cmpi slt, %o_t_114, %m_t_115 : tensor<64xi64> loc(#loc240) + %p_h1 = arith.constant 32 : i32 loc(#loc241) + %p_h1_117 = arith.constant 32 : i64 loc(#loc241) + %p_h1_118 = arith.muli %i_t_int64, %p_h1_117 : i64 loc(#loc241) + %p_h1_119 = arith.constant 128 : i32 loc(#loc242) + %p_h1_120 = arith.constant 128 : i64 loc(#loc242) + %p_h1_121 = arith.muli %p_h1_118, %p_h1_120 : i64 loc(#loc242) + %p_h1_122 = arith.constant 128 : i32 loc(#loc243) + %p_h1_123 = arith.constant 128 : i64 loc(#loc243) + %p_h1_124 = arith.muli %p_h1_121, %p_h1_123 : i64 loc(#loc243) + %p_h1_125 = tt.addptr %h_14, %p_h1_124 : !tt.ptr, i64 loc(#loc244) + %p_h1_126 = tt.expand_dims %o_k1 {axis = 1 : i32} : tensor<64xi32> -> tensor<64x1xi32> loc(#loc245) + %p_h1_127 = arith.constant 128 : i32 loc(#loc246) + %p_h1_128 = arith.constant 128 : i32 loc(#loc246) + %p_h1_129 = arith.constant dense<128> : tensor<64x1xi32> loc(#loc246) + %p_h1_130 = arith.extsi %p_h1_126 : tensor<64x1xi32> to tensor<64x1xi64> loc(#loc246) + %p_h1_131 = arith.extsi %p_h1_129 : tensor<64x1xi32> to tensor<64x1xi64> loc(#loc246) + %p_h1_132 = arith.muli %p_h1_130, %p_h1_131 : tensor<64x1xi64> loc(#loc246) + %p_h1_133 = arith.constant 2147483647 : i64 loc(#loc246) + %p_h1_134 = arith.constant -2147483648 : i64 loc(#loc246) + %p_h1_135 = arith.constant dense<2147483647> : tensor<64x1xi64> loc(#loc246) + %p_h1_136 = arith.cmpi sle, %p_h1_132, %p_h1_135 : tensor<64x1xi64> loc(#loc246) + %p_h1_137 = arith.constant dense<-2147483648> : tensor<64x1xi64> loc(#loc246) + %p_h1_138 = arith.cmpi sge, %p_h1_132, %p_h1_137 : tensor<64x1xi64> loc(#loc246) + %p_h1_139 = arith.andi %p_h1_136, %p_h1_138 : tensor<64x1xi1> loc(#loc246) + %p_h1_140 = arith.muli %p_h1_126, %p_h1_129 : tensor<64x1xi32> loc(#loc246) + %p_h1_141 = tt.splat %p_h1_125 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc247) + %p_h1_142 = tt.addptr %p_h1_141, %p_h1_140 : tensor<64x1x!tt.ptr>, tensor<64x1xi32> loc(#loc247) + %p_h1_143 = tt.expand_dims %o_v_55 {axis = 0 : i32} : tensor<64xi64> -> tensor<1x64xi64> loc(#loc248) + %p_h1_144 = tt.broadcast %p_h1_142 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc249) + %p_h1_145 = tt.broadcast %p_h1_143 : tensor<1x64xi64> -> tensor<64x64xi64> loc(#loc249) + %p_h1_146 = tt.addptr %p_h1_144, %p_h1_145 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc249) + %m_h1 = tt.expand_dims %m_k1_60 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc250) + %m_h1_147 = tt.expand_dims %m_v_58 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc251) + %m_h1_148 = tt.broadcast %m_h1 : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc252) + %m_h1_149 = tt.broadcast %m_h1_147 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc252) + %m_h1_150 = arith.andi %m_h1_148, %m_h1_149 : tensor<64x64xi1> loc(#loc252) + %16 = arith.truncf %b_h1_107 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc68) + tt.store %p_h1_146, %16, %m_h1_150 : tensor<64x64x!tt.ptr> loc(#loc69) + %p_h2 = arith.constant 32 : i32 loc(#loc253) + %p_h2_151 = arith.constant 32 : i64 loc(#loc253) + %p_h2_152 = arith.muli %i_t_int64, %p_h2_151 : i64 loc(#loc253) + %p_h2_153 = arith.constant 128 : i32 loc(#loc254) + %p_h2_154 = arith.constant 128 : i64 loc(#loc254) + %p_h2_155 = arith.muli %p_h2_152, %p_h2_154 : i64 loc(#loc254) + %p_h2_156 = arith.constant 128 : i32 loc(#loc255) + %p_h2_157 = arith.constant 128 : i64 loc(#loc255) + %p_h2_158 = arith.muli %p_h2_155, %p_h2_157 : i64 loc(#loc255) + %p_h2_159 = tt.addptr %h_14, %p_h2_158 : !tt.ptr, i64 loc(#loc256) + %p_h2_160 = tt.expand_dims %o_k2_73 {axis = 1 : i32} : tensor<64xi32> -> tensor<64x1xi32> loc(#loc257) + %p_h2_161 = arith.constant 128 : i32 loc(#loc258) + %p_h2_162 = arith.constant 128 : i32 loc(#loc258) + %p_h2_163 = arith.constant dense<128> : tensor<64x1xi32> loc(#loc258) + %p_h2_164 = arith.extsi %p_h2_160 : tensor<64x1xi32> to tensor<64x1xi64> loc(#loc258) + %p_h2_165 = arith.extsi %p_h2_163 : tensor<64x1xi32> to tensor<64x1xi64> loc(#loc258) + %p_h2_166 = arith.muli %p_h2_164, %p_h2_165 : tensor<64x1xi64> loc(#loc258) + %p_h2_167 = arith.constant 2147483647 : i64 loc(#loc258) + %p_h2_168 = arith.constant -2147483648 : i64 loc(#loc258) + %p_h2_169 = arith.constant dense<2147483647> : tensor<64x1xi64> loc(#loc258) + %p_h2_170 = arith.cmpi sle, %p_h2_166, %p_h2_169 : tensor<64x1xi64> loc(#loc258) + %p_h2_171 = arith.constant dense<-2147483648> : tensor<64x1xi64> loc(#loc258) + %p_h2_172 = arith.cmpi sge, %p_h2_166, %p_h2_171 : tensor<64x1xi64> loc(#loc258) + %p_h2_173 = arith.andi %p_h2_170, %p_h2_172 : tensor<64x1xi1> loc(#loc258) + %p_h2_174 = arith.muli %p_h2_160, %p_h2_163 : tensor<64x1xi32> loc(#loc258) + %p_h2_175 = tt.splat %p_h2_159 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc259) + %p_h2_176 = tt.addptr %p_h2_175, %p_h2_174 : tensor<64x1x!tt.ptr>, tensor<64x1xi32> loc(#loc259) + %p_h2_177 = tt.expand_dims %o_v_55 {axis = 0 : i32} : tensor<64xi64> -> tensor<1x64xi64> loc(#loc260) + %p_h2_178 = tt.broadcast %p_h2_176 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc261) + %p_h2_179 = tt.broadcast %p_h2_177 : tensor<1x64xi64> -> tensor<64x64xi64> loc(#loc261) + %p_h2_180 = tt.addptr %p_h2_178, %p_h2_179 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc261) + %m_h2 = tt.expand_dims %m_k2_75 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc262) + %m_h2_181 = tt.expand_dims %m_v_58 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc263) + %m_h2_182 = tt.broadcast %m_h2 : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc264) + %m_h2_183 = tt.broadcast %m_h2_181 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc264) + %m_h2_184 = arith.andi %m_h2_182, %m_h2_183 : tensor<64x64xi1> loc(#loc264) + %17 = arith.truncf %b_h2_108 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc82) + tt.store %p_h2_180, %17, %m_h2_184 : tensor<64x64x!tt.ptr> loc(#loc83) + %p_w = tt.expand_dims %o_t_114 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc265) + %p_w_185 = arith.constant 4096 : i32 loc(#loc266) + %p_w_186 = arith.constant 4096 : i64 loc(#loc266) + %p_w_187 = arith.constant dense<4096> : tensor<64x1xi64> loc(#loc266) + %p_w_188 = arith.muli %p_w, %p_w_187 : tensor<64x1xi64> loc(#loc266) + %p_w_189 = tt.splat %w_41 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc267) + %p_w_190 = tt.addptr %p_w_189, %p_w_188 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc267) + %p_w_191 = tt.expand_dims %o_k1 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc268) + %p_w_192 = tt.broadcast %p_w_190 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc269) + %p_w_193 = tt.broadcast %p_w_191 : tensor<1x64xi32> -> tensor<64x64xi32> loc(#loc269) + %p_w_194 = tt.addptr %p_w_192, %p_w_193 : tensor<64x64x!tt.ptr>, tensor<64x64xi32> loc(#loc269) + %b_w = tt.expand_dims %m_t_116 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc270) + %b_w_195 = tt.expand_dims %m_k1_60 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc271) + %b_w_196 = tt.broadcast %b_w : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc272) + %b_w_197 = tt.broadcast %b_w_195 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc272) + %b_w_198 = arith.andi %b_w_196, %b_w_197 : tensor<64x64xi1> loc(#loc272) + %b_w_199 = arith.constant 0.000000e+00 : f32 loc(#loc273) + %b_w_200 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc273) + %b_w_201 = arith.truncf %b_w_200 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc273) + %b_w_202 = tt.load %p_w_194, %b_w_198, %b_w_201 : tensor<64x64x!tt.ptr> loc(#loc273) + %b_v = arith.truncf %b_h1_107 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc274) + %b_v_203 = arith.constant 0.000000e+00 : f32 loc(#loc275) + %b_v_204 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc275) + %b_v_205 = tt.dot %b_w_202, %b_v, %b_v_204, inputPrecision = tf32 : tensor<64x64xbf16> * tensor<64x64xbf16> -> tensor<64x64xf32> loc(#loc275) + %p_w_206 = tt.expand_dims %o_t_114 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc276) + %p_w_207 = arith.constant 4096 : i32 loc(#loc277) + %p_w_208 = arith.constant 4096 : i64 loc(#loc277) + %p_w_209 = arith.constant dense<4096> : tensor<64x1xi64> loc(#loc277) + %p_w_210 = arith.muli %p_w_206, %p_w_209 : tensor<64x1xi64> loc(#loc277) + %p_w_211 = tt.splat %w_41 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc278) + %p_w_212 = tt.addptr %p_w_211, %p_w_210 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc278) + %p_w_213 = tt.expand_dims %o_k2_73 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc279) + %p_w_214 = tt.broadcast %p_w_212 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc280) + %p_w_215 = tt.broadcast %p_w_213 : tensor<1x64xi32> -> tensor<64x64xi32> loc(#loc280) + %p_w_216 = tt.addptr %p_w_214, %p_w_215 : tensor<64x64x!tt.ptr>, tensor<64x64xi32> loc(#loc280) + %b_w_217 = tt.expand_dims %m_t_116 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc281) + %b_w_218 = tt.expand_dims %m_k2_75 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc282) + %b_w_219 = tt.broadcast %b_w_217 : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc283) + %b_w_220 = tt.broadcast %b_w_218 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc283) + %b_w_221 = arith.andi %b_w_219, %b_w_220 : tensor<64x64xi1> loc(#loc283) + %b_w_222 = arith.constant 0.000000e+00 : f32 loc(#loc284) + %b_w_223 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc284) + %b_w_224 = arith.truncf %b_w_223 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc284) + %b_w_225 = tt.load %p_w_216, %b_w_221, %b_w_224 : tensor<64x64x!tt.ptr> loc(#loc284) + %b_v_226 = arith.truncf %b_h2_108 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc285) + %b_v_227 = arith.constant 0.000000e+00 : f32 loc(#loc286) + %b_v_228 = tt.dot %b_w_225, %b_v_226, %b_v_205, inputPrecision = tf32 : tensor<64x64xbf16> * tensor<64x64xbf16> -> tensor<64x64xf32> loc(#loc286) + %p_v = tt.expand_dims %o_t_114 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc287) + %p_v_229 = arith.constant 4096 : i32 loc(#loc288) + %p_v_230 = arith.constant 4096 : i64 loc(#loc288) + %p_v_231 = arith.constant dense<4096> : tensor<64x1xi64> loc(#loc288) + %p_v_232 = arith.muli %p_v, %p_v_231 : tensor<64x1xi64> loc(#loc288) + %p_v_233 = tt.splat %v_22 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc289) + %p_v_234 = tt.addptr %p_v_233, %p_v_232 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc289) + %p_v_235 = tt.expand_dims %o_v_55 {axis = 0 : i32} : tensor<64xi64> -> tensor<1x64xi64> loc(#loc290) + %p_v_236 = tt.broadcast %p_v_234 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc291) + %p_v_237 = tt.broadcast %p_v_235 : tensor<1x64xi64> -> tensor<64x64xi64> loc(#loc291) + %p_v_238 = tt.addptr %p_v_236, %p_v_237 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc291) + %b_v_239 = tt.expand_dims %m_t_116 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc292) + %b_v_240 = tt.expand_dims %m_v_58 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc293) + %b_v_241 = tt.broadcast %b_v_239 : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc294) + %b_v_242 = tt.broadcast %b_v_240 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc294) + %b_v_243 = arith.andi %b_v_241, %b_v_242 : tensor<64x64xi1> loc(#loc294) + %b_v_244 = arith.constant 0.000000e+00 : f32 loc(#loc295) + %b_v_245 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc295) + %b_v_246 = arith.truncf %b_v_245 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc295) + %b_v_247 = tt.load %p_v_238, %b_v_243, %b_v_246 : tensor<64x64x!tt.ptr> loc(#loc295) + %b_v_248 = arith.extf %b_v_247 : tensor<64x64xbf16> to tensor<64x64xf32> loc(#loc296) + %b_v_249 = arith.subf %b_v_248, %b_v_228 : tensor<64x64xf32> loc(#loc296) + %p_v_250 = tt.expand_dims %o_t_114 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc297) + %p_v_251 = arith.constant 4096 : i32 loc(#loc298) + %p_v_252 = arith.constant 4096 : i64 loc(#loc298) + %p_v_253 = arith.constant dense<4096> : tensor<64x1xi64> loc(#loc298) + %p_v_254 = arith.muli %p_v_250, %p_v_253 : tensor<64x1xi64> loc(#loc298) + %p_v_255 = tt.splat %v_new_49 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc299) + %p_v_256 = tt.addptr %p_v_255, %p_v_254 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc299) + %p_v_257 = tt.expand_dims %o_v_55 {axis = 0 : i32} : tensor<64xi64> -> tensor<1x64xi64> loc(#loc300) + %p_v_258 = tt.broadcast %p_v_256 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc301) + %p_v_259 = tt.broadcast %p_v_257 : tensor<1x64xi64> -> tensor<64x64xi64> loc(#loc301) + %p_v_260 = tt.addptr %p_v_258, %p_v_259 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc301) + %18 = tt.expand_dims %m_t_116 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc121) + %19 = tt.expand_dims %m_v_58 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc122) + %20 = tt.broadcast %18 : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc123) + %21 = tt.broadcast %19 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc123) + %22 = arith.andi %20, %21 : tensor<64x64xi1> loc(#loc123) + %23 = arith.truncf %b_v_249 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc124) + tt.store %p_v_260, %23, %22 : tensor<64x64x!tt.ptr> loc(#loc125) + %last_idx = arith.constant 1 : i32 loc(#loc302) + %last_idx_261 = arith.constant 1 : i32 loc(#loc302) + %last_idx_262 = arith.extsi %i_t : i32 to i64 loc(#loc302) + %last_idx_263 = arith.extsi %last_idx_261 : i32 to i64 loc(#loc302) + %last_idx_264 = arith.addi %last_idx_262, %last_idx_263 : i64 loc(#loc302) + %last_idx_265 = arith.constant 2147483647 : i64 loc(#loc302) + %last_idx_266 = arith.constant -2147483648 : i64 loc(#loc302) + %last_idx_267 = arith.cmpi sle, %last_idx_264, %last_idx_265 : i64 loc(#loc302) + %last_idx_268 = arith.cmpi sge, %last_idx_264, %last_idx_266 : i64 loc(#loc302) + %last_idx_269 = arith.andi %last_idx_267, %last_idx_268 : i1 loc(#loc302) + %last_idx_270 = arith.addi %i_t, %last_idx_261 : i32 loc(#loc302) + %last_idx_271 = arith.constant 64 : i32 loc(#loc303) + %last_idx_272 = arith.constant 64 : i32 loc(#loc303) + %last_idx_273 = arith.extsi %last_idx_270 : i32 to i64 loc(#loc303) + %last_idx_274 = arith.extsi %last_idx_272 : i32 to i64 loc(#loc303) + %last_idx_275 = arith.muli %last_idx_273, %last_idx_274 : i64 loc(#loc303) + %last_idx_276 = arith.constant 2147483647 : i64 loc(#loc303) + %last_idx_277 = arith.constant -2147483648 : i64 loc(#loc303) + %last_idx_278 = arith.cmpi sle, %last_idx_275, %last_idx_276 : i64 loc(#loc303) + %last_idx_279 = arith.cmpi sge, %last_idx_275, %last_idx_277 : i64 loc(#loc303) + %last_idx_280 = arith.andi %last_idx_278, %last_idx_279 : i1 loc(#loc303) + %last_idx_281 = arith.muli %last_idx_270, %last_idx_272 : i32 loc(#loc303) + %last_idx_282 = arith.minsi %last_idx_281, %T : i32 loc(#loc304) + %last_idx_283 = arith.constant 1 : i32 loc(#loc305) + %last_idx_284 = arith.constant 1 : i32 loc(#loc305) + %last_idx_285 = arith.extsi %last_idx_282 : i32 to i64 loc(#loc305) + %last_idx_286 = arith.extsi %last_idx_284 : i32 to i64 loc(#loc305) + %last_idx_287 = arith.subi %last_idx_285, %last_idx_286 : i64 loc(#loc305) + %last_idx_288 = arith.constant 2147483647 : i64 loc(#loc305) + %last_idx_289 = arith.constant -2147483648 : i64 loc(#loc305) + %last_idx_290 = arith.cmpi sle, %last_idx_287, %last_idx_288 : i64 loc(#loc305) + %last_idx_291 = arith.cmpi sge, %last_idx_287, %last_idx_289 : i64 loc(#loc305) + %last_idx_292 = arith.andi %last_idx_290, %last_idx_291 : i1 loc(#loc305) + %last_idx_293 = arith.subi %last_idx_282, %last_idx_284 : i32 loc(#loc305) + %b_g_last = arith.constant 32 : i32 loc(#loc306) + %b_g_last_294 = arith.constant 32 : i64 loc(#loc306) + %b_g_last_295 = arith.muli %7, %b_g_last_294 : i64 loc(#loc306) + %b_g_last_296 = arith.constant 32 : i32 loc(#loc307) + %b_g_last_297 = arith.constant 32 : i32 loc(#loc307) + %b_g_last_298 = arith.extsi %last_idx_293 : i32 to i64 loc(#loc307) + %b_g_last_299 = arith.extsi %b_g_last_297 : i32 to i64 loc(#loc307) + %b_g_last_300 = arith.muli %b_g_last_298, %b_g_last_299 : i64 loc(#loc307) + %b_g_last_301 = arith.constant 2147483647 : i64 loc(#loc307) + %b_g_last_302 = arith.constant -2147483648 : i64 loc(#loc307) + %b_g_last_303 = arith.cmpi sle, %b_g_last_300, %b_g_last_301 : i64 loc(#loc307) + %b_g_last_304 = arith.cmpi sge, %b_g_last_300, %b_g_last_302 : i64 loc(#loc307) + %b_g_last_305 = arith.andi %b_g_last_303, %b_g_last_304 : i1 loc(#loc307) + %b_g_last_306 = arith.muli %last_idx_293, %b_g_last_297 : i32 loc(#loc307) + %b_g_last_307 = arith.extsi %b_g_last_306 : i32 to i64 loc(#loc308) + %b_g_last_308 = arith.addi %b_g_last_295, %b_g_last_307 : i64 loc(#loc308) + %b_g_last_309 = arith.addi %b_g_last_308, %5 : i64 loc(#loc309) + %b_g_last_310 = tt.addptr %g, %b_g_last_309 : !tt.ptr, i64 loc(#loc310) + %b_g_last_311 = tt.load %b_g_last_310 : !tt.ptr loc(#loc311) + %p_g = arith.constant 32 : i32 loc(#loc312) + %p_g_312 = arith.constant 32 : i64 loc(#loc312) + %p_g_313 = arith.muli %7, %p_g_312 : i64 loc(#loc312) + %p_g_314 = arith.addi %p_g_313, %5 : i64 loc(#loc313) + %p_g_315 = tt.addptr %g, %p_g_314 : !tt.ptr, i64 loc(#loc314) + %p_g_316 = arith.constant 32 : i32 loc(#loc315) + %p_g_317 = arith.constant 32 : i64 loc(#loc315) + %p_g_318 = arith.constant dense<32> : tensor<64xi64> loc(#loc315) + %p_g_319 = arith.muli %o_t_114, %p_g_318 : tensor<64xi64> loc(#loc315) + %p_g_320 = tt.splat %p_g_315 : !tt.ptr -> tensor<64x!tt.ptr> loc(#loc316) + %p_g_321 = tt.addptr %p_g_320, %p_g_319 : tensor<64x!tt.ptr>, tensor<64xi64> loc(#loc316) + %b_g = arith.constant 0.000000e+00 : f32 loc(#loc317) + %b_g_322 = arith.constant dense<0.000000e+00> : tensor<64xf32> loc(#loc317) + %b_g_323 = tt.load %p_g_321, %m_t_116, %b_g_322 : tensor<64x!tt.ptr> loc(#loc317) + %b_v_324 = tt.splat %b_g_last_311 : f32 -> tensor<64xf32> loc(#loc318) + %b_v_325 = arith.subf %b_v_324, %b_g_323 : tensor<64xf32> loc(#loc318) + %b_v_326 = tt.call @fla.ops.utils.op.exp2__fp32S64S(%b_v_325) : (tensor<64xf32>) -> tensor<64xf32> loc(#loc319) + %b_v_327 = arith.constant 0 : i32 loc(#loc320) + %b_v_328 = arith.constant 0.000000e+00 : f32 loc(#loc320) + %b_v_329 = arith.constant dense<0.000000e+00> : tensor<64xf32> loc(#loc320) + %b_v_330 = arith.select %m_t_116, %b_v_326, %b_v_329 : tensor<64xi1>, tensor<64xf32> loc(#loc320) + %b_v_331 = tt.expand_dims %b_v_330 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc321) + %b_v_332 = tt.broadcast %b_v_331 : tensor<64x1xf32> -> tensor<64x64xf32> loc(#loc322) + %b_v_333 = arith.mulf %b_v_249, %b_v_332 : tensor<64x64xf32> loc(#loc322) + %b_g_last_334 = tt.call @fla.ops.utils.op.exp2__fp32(%b_g_last_311) : (f32) -> f32 loc(#loc323) + %b_h1_335 = tt.splat %b_g_last_334 : f32 -> tensor<64x64xf32> loc(#loc324) + %b_h1_336 = arith.mulf %b_h1_107, %b_h1_335 : tensor<64x64xf32> loc(#loc324) + %b_h2_337 = tt.splat %b_g_last_334 : f32 -> tensor<64x64xf32> loc(#loc325) + %b_h2_338 = arith.mulf %b_h2_108, %b_h2_337 : tensor<64x64xf32> loc(#loc325) + %b_v_339 = arith.truncf %b_v_333 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc326) + %p_k = tt.expand_dims %o_k1 {axis = 1 : i32} : tensor<64xi32> -> tensor<64x1xi32> loc(#loc327) + %p_k_340 = tt.splat %k_33 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc328) + %p_k_341 = tt.addptr %p_k_340, %p_k : tensor<64x1x!tt.ptr>, tensor<64x1xi32> loc(#loc328) + %p_k_342 = tt.expand_dims %o_t_114 {axis = 0 : i32} : tensor<64xi64> -> tensor<1x64xi64> loc(#loc329) + %p_k_343 = arith.constant 4096 : i32 loc(#loc330) + %p_k_344 = arith.constant 4096 : i64 loc(#loc330) + %p_k_345 = arith.constant dense<4096> : tensor<1x64xi64> loc(#loc330) + %p_k_346 = arith.muli %p_k_342, %p_k_345 : tensor<1x64xi64> loc(#loc330) + %p_k_347 = tt.broadcast %p_k_341 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc331) + %p_k_348 = tt.broadcast %p_k_346 : tensor<1x64xi64> -> tensor<64x64xi64> loc(#loc331) + %p_k_349 = tt.addptr %p_k_347, %p_k_348 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc331) + %b_k = tt.expand_dims %m_k1_60 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc332) + %b_k_350 = tt.expand_dims %m_t_116 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc333) + %b_k_351 = tt.broadcast %b_k : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc334) + %b_k_352 = tt.broadcast %b_k_350 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc334) + %b_k_353 = arith.andi %b_k_351, %b_k_352 : tensor<64x64xi1> loc(#loc334) + %b_k_354 = arith.constant 0.000000e+00 : f32 loc(#loc335) + %b_k_355 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc335) + %b_k_356 = arith.truncf %b_k_355 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc335) + %b_k_357 = tt.load %p_k_349, %b_k_353, %b_k_356 : tensor<64x64x!tt.ptr> loc(#loc335) + %b_h1_358 = arith.constant 0.000000e+00 : f32 loc(#loc336) + %b_h1_359 = tt.dot %b_k_357, %b_v_339, %b_h1_336, inputPrecision = tf32 : tensor<64x64xbf16> * tensor<64x64xbf16> -> tensor<64x64xf32> loc(#loc336) + %p_k_360 = tt.expand_dims %o_k2_73 {axis = 1 : i32} : tensor<64xi32> -> tensor<64x1xi32> loc(#loc337) + %p_k_361 = tt.splat %k_33 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc338) + %p_k_362 = tt.addptr %p_k_361, %p_k_360 : tensor<64x1x!tt.ptr>, tensor<64x1xi32> loc(#loc338) + %p_k_363 = tt.expand_dims %o_t_114 {axis = 0 : i32} : tensor<64xi64> -> tensor<1x64xi64> loc(#loc339) + %p_k_364 = arith.constant 4096 : i32 loc(#loc340) + %p_k_365 = arith.constant 4096 : i64 loc(#loc340) + %p_k_366 = arith.constant dense<4096> : tensor<1x64xi64> loc(#loc340) + %p_k_367 = arith.muli %p_k_363, %p_k_366 : tensor<1x64xi64> loc(#loc340) + %p_k_368 = tt.broadcast %p_k_362 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc341) + %p_k_369 = tt.broadcast %p_k_367 : tensor<1x64xi64> -> tensor<64x64xi64> loc(#loc341) + %p_k_370 = tt.addptr %p_k_368, %p_k_369 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc341) + %b_k_371 = tt.expand_dims %m_k2_75 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc342) + %b_k_372 = tt.expand_dims %m_t_116 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc343) + %b_k_373 = tt.broadcast %b_k_371 : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc344) + %b_k_374 = tt.broadcast %b_k_372 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc344) + %b_k_375 = arith.andi %b_k_373, %b_k_374 : tensor<64x64xi1> loc(#loc344) + %b_k_376 = arith.constant 0.000000e+00 : f32 loc(#loc345) + %b_k_377 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc345) + %b_k_378 = arith.truncf %b_k_377 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc345) + %b_k_379 = tt.load %p_k_370, %b_k_375, %b_k_378 : tensor<64x64x!tt.ptr> loc(#loc345) + %b_h2_380 = arith.constant 0.000000e+00 : f32 loc(#loc346) + %b_h2_381 = tt.dot %b_k_379, %b_v_339, %b_h2_338, inputPrecision = tf32 : tensor<64x64xbf16> * tensor<64x64xbf16> -> tensor<64x64xf32> loc(#loc346) + scf.yield %b_h1_359, %b_h2_381 : tensor<64x64xf32>, tensor<64x64xf32> loc(#loc171) + } loc(#loc349) + tt.return loc(#loc172) + } loc(#loc) + tt.func private @triton.language.standard.cdiv__c128_c64() attributes {noinline = false} { + tt.return loc(#loc174) + ^bb1: // no predecessors + tt.return loc(#loc175) + } loc(#loc173) + tt.func private @triton.language.standard.cdiv__i32_c64(%x: i32 loc("x"(#loc173))) -> i32 attributes {noinline = false} { + %c63_i32 = arith.constant 63 : i32 loc(#loc176) + %c63_i32_0 = arith.constant 63 : i32 loc(#loc176) + %0 = arith.extsi %x : i32 to i64 loc(#loc176) + %1 = arith.extsi %c63_i32_0 : i32 to i64 loc(#loc176) + %2 = arith.addi %0, %1 : i64 loc(#loc176) + %c2147483647_i64 = arith.constant 2147483647 : i64 loc(#loc176) + %c-2147483648_i64 = arith.constant -2147483648 : i64 loc(#loc176) + %3 = arith.cmpi sle, %2, %c2147483647_i64 : i64 loc(#loc176) + %4 = arith.cmpi sge, %2, %c-2147483648_i64 : i64 loc(#loc176) + %5 = arith.andi %3, %4 : i1 loc(#loc176) + %6 = arith.addi %x, %c63_i32_0 : i32 loc(#loc176) + %c64_i32 = arith.constant 64 : i32 loc(#loc177) + %c64_i32_1 = arith.constant 64 : i32 loc(#loc177) + %7 = arith.divsi %6, %c64_i32_1 : i32 loc(#loc177) + tt.return %7 : i32 loc(#loc174) + ^bb1: // no predecessors + %8 = ub.poison : i32 loc(#loc175) + tt.return %8 : i32 loc(#loc175) + } loc(#loc173) + tt.func private @triton.language.standard.zeros__Tc64_c64T_cfp32() -> tensor<64x64xf32> attributes {noinline = false} { + %cst = arith.constant 0.000000e+00 : f32 loc(#loc179) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc179) + tt.return %cst_0 : tensor<64x64xf32> loc(#loc180) + ^bb1: // no predecessors + %0 = ub.poison : tensor<64x64xf32> loc(#loc181) + tt.return %0 : tensor<64x64xf32> loc(#loc181) + } loc(#loc178) + tt.func private @fla.ops.utils.op.exp2__fp32S64S(%x: tensor<64xf32> loc("x"(#loc182))) -> tensor<64xf32> attributes {noinline = false} { + %0 = math.exp2 %x : tensor<64xf32> loc(#loc183) + tt.return %0 : tensor<64xf32> loc(#loc184) + ^bb1: // no predecessors + %1 = ub.poison : tensor<64xf32> loc(#loc185) + tt.return %1 : tensor<64xf32> loc(#loc185) + } loc(#loc182) + tt.func private @fla.ops.utils.op.exp2__fp32(%x: f32 loc("x"(#loc182))) -> f32 attributes {noinline = false} { + %0 = math.exp2 %x : f32 loc(#loc183) + tt.return %0 : f32 loc(#loc184) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc185) + tt.return %1 : f32 loc(#loc185) + } loc(#loc182) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":86:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":86:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":87:20) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":87:9) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":88:22) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":88:34) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":89:23) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":89:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":96:25) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":96:34) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":96:38) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":97:24) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":98:20) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":109:24) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":111:28) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":118:16) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":118:21) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":118:41) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":118:43) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":118:9) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":119:16) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":119:21) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":119:41) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":119:9) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":120:16) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":120:28) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":120:20) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":120:53) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":120:9) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":121:16) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":121:21) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":121:41) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":121:9) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":123:24) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":123:29) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":123:49) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":123:17) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":131:16) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":131:34) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":131:21) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":132:16) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":133:24) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":134:18) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":135:16) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":136:18) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":137:17) +#loc47 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":138:18) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":139:17) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":140:18) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":175:21) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":176:27) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":177:26) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":177:44) +#loc54 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":177:31) +#loc55 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":178:20) +#loc56 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":183:35) +#loc57 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":183:38) +#loc58 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":183:40) +#loc59 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":183:23) +#loc60 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":183:49) +#loc61 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":183:60) +#loc62 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":183:44) +#loc63 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":183:68) +#loc64 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":183:64) +#loc65 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":184:24) +#loc66 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":184:39) +#loc67 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":184:35) +#loc68 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":185:31) +#loc69 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":185:23) +#loc70 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":191:39) +#loc71 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":191:42) +#loc72 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":191:44) +#loc73 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":191:27) +#loc74 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":191:53) +#loc75 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":191:64) +#loc76 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":191:48) +#loc77 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":191:72) +#loc78 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":191:68) +#loc79 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":192:28) +#loc80 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":192:43) +#loc81 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":192:39) +#loc82 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":193:35) +#loc83 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":193:27) +#loc84 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":211:22) +#loc85 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":211:34) +#loc86 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":211:18) +#loc87 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":211:47) +#loc88 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":211:42) +#loc89 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":212:36) +#loc90 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":212:52) +#loc91 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":212:47) +#loc92 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":212:22) +#loc93 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":216:38) +#loc94 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":216:30) +#loc95 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":218:26) +#loc96 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":218:38) +#loc97 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":218:22) +#loc98 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":218:51) +#loc99 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":218:46) +#loc100 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":219:40) +#loc101 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":219:56) +#loc102 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":219:51) +#loc103 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":219:26) +#loc104 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":223:42) +#loc105 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":223:54) +#loc106 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":238:22) +#loc107 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":238:34) +#loc108 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":238:18) +#loc109 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":238:46) +#loc110 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":238:42) +#loc111 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":239:36) +#loc112 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":239:51) +#loc113 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":239:47) +#loc114 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":239:22) +#loc115 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":239:74) +#loc116 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":242:30) +#loc117 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":242:42) +#loc118 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":242:26) +#loc119 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":242:54) +#loc120 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":242:50) +#loc121 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":243:65) +#loc122 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":243:80) +#loc123 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":243:76) +#loc124 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":243:33) +#loc125 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":243:26) +#loc126 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":245:30) +#loc127 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":245:35) +#loc128 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":245:39) +#loc129 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":245:44) +#loc130 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":247:42) +#loc131 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":247:58) +#loc132 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":247:47) +#loc133 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":247:63) +#loc134 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":247:35) +#loc135 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":247:31) +#loc136 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":248:29) +#loc137 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":248:34) +#loc138 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":248:22) +#loc139 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":248:60) +#loc140 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":248:54) +#loc141 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":249:26) +#loc142 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":250:54) +#loc143 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":250:43) +#loc144 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":250:60) +#loc145 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":250:63) +#loc146 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":250:24) +#loc147 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":251:28) +#loc148 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":252:20) +#loc149 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":254:24) +#loc150 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":288:21) +#loc151 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":290:23) +#loc152 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":290:18) +#loc153 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":290:38) +#loc154 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":290:50) +#loc155 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":290:34) +#loc156 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":291:37) +#loc157 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":291:52) +#loc158 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":291:48) +#loc159 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":291:22) +#loc160 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":295:36) +#loc161 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":297:27) +#loc162 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":297:22) +#loc163 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":297:42) +#loc164 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":297:54) +#loc165 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":297:38) +#loc166 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":298:41) +#loc167 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":298:56) +#loc168 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":298:52) +#loc169 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":298:26) +#loc170 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":302:40) +#loc171 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":310:8) +#loc172 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":318:4) +#loc174 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":43:11) +#loc175 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":43:4) +#loc176 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":43:17) +#loc177 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":43:30) +#loc178 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":120:0) +#loc179 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":129:31) +#loc180 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":129:11) +#loc181 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":129:4) +#loc183 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:33) +#loc184 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:20) +#loc185 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:13) +#loc193 = loc("pid"(#loc1)) +#loc194 = loc("pid"(#loc2)) +#loc195 = loc("NV"(#loc3)) +#loc196 = loc("NV"(#loc4)) +#loc197 = loc("NT"(#loc12)) +#loc198 = loc("boh"(#loc13)) +#loc199 = loc("b_h1"(#loc14)) +#loc200 = loc("b_h2"(#loc15)) +#loc201 = loc("h"(#loc16)) +#loc202 = loc("h"(#loc17)) +#loc203 = loc("h"(#loc18)) +#loc204 = loc("h"(#loc19)) +#loc205 = loc("h"(#loc20)) +#loc206 = loc("v"(#loc21)) +#loc207 = loc("v"(#loc22)) +#loc208 = loc("v"(#loc23)) +#loc209 = loc("v"(#loc24)) +#loc210 = loc("k"(#loc25)) +#loc211 = loc("k"(#loc26)) +#loc212 = loc("k"(#loc27)) +#loc213 = loc("k"(#loc28)) +#loc214 = loc("k"(#loc29)) +#loc215 = loc("w"(#loc30)) +#loc216 = loc("w"(#loc31)) +#loc217 = loc("w"(#loc32)) +#loc218 = loc("w"(#loc33)) +#loc219 = loc("v_new"(#loc34)) +#loc220 = loc("v_new"(#loc35)) +#loc221 = loc("v_new"(#loc36)) +#loc222 = loc("v_new"(#loc37)) +#loc223 = loc("o_v"(#loc38)) +#loc224 = loc("o_v"(#loc39)) +#loc225 = loc("o_v"(#loc40)) +#loc226 = loc("m_v"(#loc41)) +#loc227 = loc("o_k1"(#loc42)) +#loc228 = loc("m_k1"(#loc43)) +#loc229 = loc("o_k2"(#loc44)) +#loc230 = loc("m_k2"(#loc45)) +#loc231 = loc("o_k3"(#loc46)) +#loc232 = loc("m_k3"(#loc47)) +#loc233 = loc("o_k4"(#loc48)) +#loc234 = loc("m_k4"(#loc49)) +#loc235 = loc("b_h1"(#loc50)) +#loc236 = loc("i_t_int64"(#loc51)) +#loc237 = loc("o_t"(#loc52)) +#loc238 = loc("o_t"(#loc53)) +#loc239 = loc("o_t"(#loc54)) +#loc240 = loc("m_t"(#loc55)) +#loc241 = loc("p_h1"(#loc56)) +#loc242 = loc("p_h1"(#loc57)) +#loc243 = loc("p_h1"(#loc58)) +#loc244 = loc("p_h1"(#loc59)) +#loc245 = loc("p_h1"(#loc60)) +#loc246 = loc("p_h1"(#loc61)) +#loc247 = loc("p_h1"(#loc62)) +#loc248 = loc("p_h1"(#loc63)) +#loc249 = loc("p_h1"(#loc64)) +#loc250 = loc("m_h1"(#loc65)) +#loc251 = loc("m_h1"(#loc66)) +#loc252 = loc("m_h1"(#loc67)) +#loc253 = loc("p_h2"(#loc70)) +#loc254 = loc("p_h2"(#loc71)) +#loc255 = loc("p_h2"(#loc72)) +#loc256 = loc("p_h2"(#loc73)) +#loc257 = loc("p_h2"(#loc74)) +#loc258 = loc("p_h2"(#loc75)) +#loc259 = loc("p_h2"(#loc76)) +#loc260 = loc("p_h2"(#loc77)) +#loc261 = loc("p_h2"(#loc78)) +#loc262 = loc("m_h2"(#loc79)) +#loc263 = loc("m_h2"(#loc80)) +#loc264 = loc("m_h2"(#loc81)) +#loc265 = loc("p_w"(#loc84)) +#loc266 = loc("p_w"(#loc85)) +#loc267 = loc("p_w"(#loc86)) +#loc268 = loc("p_w"(#loc87)) +#loc269 = loc("p_w"(#loc88)) +#loc270 = loc("b_w"(#loc89)) +#loc271 = loc("b_w"(#loc90)) +#loc272 = loc("b_w"(#loc91)) +#loc273 = loc("b_w"(#loc92)) +#loc274 = loc("b_v"(#loc93)) +#loc275 = loc("b_v"(#loc94)) +#loc276 = loc("p_w"(#loc95)) +#loc277 = loc("p_w"(#loc96)) +#loc278 = loc("p_w"(#loc97)) +#loc279 = loc("p_w"(#loc98)) +#loc280 = loc("p_w"(#loc99)) +#loc281 = loc("b_w"(#loc100)) +#loc282 = loc("b_w"(#loc101)) +#loc283 = loc("b_w"(#loc102)) +#loc284 = loc("b_w"(#loc103)) +#loc285 = loc("b_v"(#loc104)) +#loc286 = loc("b_v"(#loc105)) +#loc287 = loc("p_v"(#loc106)) +#loc288 = loc("p_v"(#loc107)) +#loc289 = loc("p_v"(#loc108)) +#loc290 = loc("p_v"(#loc109)) +#loc291 = loc("p_v"(#loc110)) +#loc292 = loc("b_v"(#loc111)) +#loc293 = loc("b_v"(#loc112)) +#loc294 = loc("b_v"(#loc113)) +#loc295 = loc("b_v"(#loc114)) +#loc296 = loc("b_v"(#loc115)) +#loc297 = loc("p_v"(#loc116)) +#loc298 = loc("p_v"(#loc117)) +#loc299 = loc("p_v"(#loc118)) +#loc300 = loc("p_v"(#loc119)) +#loc301 = loc("p_v"(#loc120)) +#loc302 = loc("last_idx"(#loc126)) +#loc303 = loc("last_idx"(#loc127)) +#loc304 = loc("last_idx"(#loc128)) +#loc305 = loc("last_idx"(#loc129)) +#loc306 = loc("b_g_last"(#loc130)) +#loc307 = loc("b_g_last"(#loc131)) +#loc308 = loc("b_g_last"(#loc132)) +#loc309 = loc("b_g_last"(#loc133)) +#loc310 = loc("b_g_last"(#loc134)) +#loc311 = loc("b_g_last"(#loc135)) +#loc312 = loc("p_g"(#loc136)) +#loc313 = loc("p_g"(#loc137)) +#loc314 = loc("p_g"(#loc138)) +#loc315 = loc("p_g"(#loc139)) +#loc316 = loc("p_g"(#loc140)) +#loc317 = loc("b_g"(#loc141)) +#loc318 = loc("b_v"(#loc142)) +#loc319 = loc("b_v"(#loc143)) +#loc320 = loc("b_v"(#loc144)) +#loc321 = loc("b_v"(#loc145)) +#loc322 = loc("b_v"(#loc146)) +#loc323 = loc("b_g_last"(#loc147)) +#loc324 = loc("b_h1"(#loc148)) +#loc325 = loc("b_h2"(#loc149)) +#loc326 = loc("b_v"(#loc150)) +#loc327 = loc("p_k"(#loc151)) +#loc328 = loc("p_k"(#loc152)) +#loc329 = loc("p_k"(#loc153)) +#loc330 = loc("p_k"(#loc154)) +#loc331 = loc("p_k"(#loc155)) +#loc332 = loc("b_k"(#loc156)) +#loc333 = loc("b_k"(#loc157)) +#loc334 = loc("b_k"(#loc158)) +#loc335 = loc("b_k"(#loc159)) +#loc336 = loc("b_h1"(#loc160)) +#loc337 = loc("p_k"(#loc161)) +#loc338 = loc("p_k"(#loc162)) +#loc339 = loc("p_k"(#loc163)) +#loc340 = loc("p_k"(#loc164)) +#loc341 = loc("p_k"(#loc165)) +#loc342 = loc("b_k"(#loc166)) +#loc343 = loc("b_k"(#loc167)) +#loc344 = loc("b_k"(#loc168)) +#loc345 = loc("b_k"(#loc169)) +#loc346 = loc("b_h2"(#loc170)) +#loc349 = loc("b_h2"(#loc235)) diff --git a/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.ttgir b/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..397a2c1d82892b1efbfd7cf677a87a1469ed57fc --- /dev/null +++ b/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.ttgir @@ -0,0 +1,763 @@ +#blocked = #ttg.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [2], order = [0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [8, 1], threadsPerWarp = [8, 4], warpsPerCTA = [1, 2], order = [0, 1]}> +#blocked2 = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [4, 8], warpsPerCTA = [2, 1], order = [1, 0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":59:0) +#mma = #ttg.nvidia_mma<{versionMajor = 2, versionMinor = 0, warpsPerCTA = [2, 1], instrShape = [16, 8]}> +#shared = #ttg.swizzled_shared<{vec = 8, perPhase = 1, maxPhase = 8, order = [1, 0]}> +#shared1 = #ttg.swizzled_shared<{vec = 1, perPhase = 1, maxPhase = 1, order = [0]}> +#shared2 = #ttg.swizzled_shared<{vec = 8, perPhase = 1, maxPhase = 8, order = [0, 1]}> +#smem = #ttg.shared_memory +#loc119 = loc("k"(#loc)) +#loc120 = loc("v"(#loc)) +#loc121 = loc("w"(#loc)) +#loc122 = loc("v_new"(#loc)) +#loc123 = loc("g"(#loc)) +#loc124 = loc("h"(#loc)) +#loc125 = loc("T"(#loc)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 2 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @chunk_gated_delta_rule_fwd_kernel_h_blockdim64(%k: !tt.ptr {tt.divisibility = 16 : i32} loc("k"(#loc)), %v: !tt.ptr {tt.divisibility = 16 : i32} loc("v"(#loc)), %w: !tt.ptr {tt.divisibility = 16 : i32} loc("w"(#loc)), %v_new: !tt.ptr {tt.divisibility = 16 : i32} loc("v_new"(#loc)), %g: !tt.ptr {tt.divisibility = 16 : i32} loc("g"(#loc)), %h: !tt.ptr {tt.divisibility = 16 : i32} loc("h"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<0.000000e+00> : tensor<64xf32, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc1) + %cst_0 = arith.constant dense<64> : tensor<64xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<64xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<64xi64, #blocked> loc(#loc1) + %c192_i32 = arith.constant 192 : i32 loc(#loc1) + %c4_i32 = arith.constant 4 : i32 loc(#loc1) + %cst_3 = arith.constant dense<128> : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc1) + %cst_4 = arith.constant dense<128> : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked2}>> loc(#loc1) + %c128_i32 = arith.constant 128 : i32 loc(#loc1) + %cst_5 = arith.constant dense<64> : tensor<64xi64, #blocked> loc(#loc1) + %cst_6 = arith.constant dense<64> : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc1) + %cst_7 = arith.constant dense<64> : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked2}>> loc(#loc1) + %c2_i32 = arith.constant 2 : i32 loc(#loc1) + %c1_i64 = arith.constant 1 : i64 loc(#loc1) + %c0_i64 = arith.constant 0 : i64 loc(#loc1) + %c3_i32 = arith.constant 3 : i32 loc(#loc1) + %c-1_i32 = arith.constant -1 : i32 loc(#loc1) + %cst_8 = arith.constant dense<0.000000e+00> : tensor<64x64xf32, #mma> loc(#loc1) + %cst_9 = arith.constant dense<32> : tensor<64xi64, #blocked> loc(#loc1) + %cst_10 = arith.constant dense<0.000000e+00> : tensor<64xf32, #blocked> loc(#loc1) + %c2_i64 = arith.constant 2 : i64 loc(#loc1) + %c16384_i64 = arith.constant 16384 : i64 loc(#loc1) + %c0_i32 = arith.constant 0 : i32 loc(#loc1) + %c524288_i64 = arith.constant 524288 : i64 loc(#loc1) + %c63_i32 = arith.constant 63 : i32 loc(#loc1) + %cst_11 = arith.constant dense<0.000000e+00> : tensor<64x64xbf16, #blocked1> loc(#loc1) + %cst_12 = arith.constant dense<0.000000e+00> : tensor<64x64xbf16, #blocked2> loc(#loc1) + %c32_i32 = arith.constant 32 : i32 loc(#loc1) + %c32_i64 = arith.constant 32 : i64 loc(#loc1) + %c128_i64 = arith.constant 128 : i64 loc(#loc1) + %c1_i32 = arith.constant 1 : i32 loc(#loc1) + %c64_i32 = arith.constant 64 : i32 loc(#loc1) + %c64_i64 = arith.constant 64 : i64 loc(#loc1) + %cst_13 = arith.constant dense<128> : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked2}>> loc(#loc1) + %cst_14 = arith.constant dense<128> : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc1) + %cst_15 = arith.constant dense<128> : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked2}>> loc(#loc1) + %cst_16 = arith.constant dense<64> : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked2}>> loc(#loc1) + %cst_17 = arith.constant dense<64> : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc1) + %cst_18 = arith.constant dense<64> : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked2}>> loc(#loc1) + %cst_19 = arith.constant dense<128> : tensor<64x1xi32, #blocked2> loc(#loc1) + %cst_20 = arith.constant dense<4096> : tensor<64x1xi64, #blocked2> loc(#loc1) + %cst_21 = arith.constant dense<128> : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked2}>> loc(#loc1) + %cst_22 = arith.constant dense<4096> : tensor<1x64xi64, #blocked1> loc(#loc1) + %pid = tt.get_program_id x : i32 loc(#loc126) + %pid_23 = arith.extsi %pid : i32 to i64 loc(#loc127) + %0 = arith.remsi %pid_23, %c2_i64 : i64 loc(#loc4) + %1 = arith.divsi %pid_23, %c2_i64 : i64 loc(#loc5) + %2 = arith.divsi %1, %c32_i64 : i64 loc(#loc6) + %3 = arith.remsi %1, %c32_i64 : i64 loc(#loc7) + %4 = arith.extsi %T : i32 to i64 loc(#loc8) + %5 = arith.muli %2, %4 : i64 loc(#loc8) + %NT = arith.addi %T, %c63_i32 : i32 loc(#loc228) + %NT_24 = arith.divsi %NT, %c64_i32 : i32 loc(#loc229) + %boh = arith.extsi %NT_24 : i32 to i64 loc(#loc129) + %boh_25 = arith.muli %2, %boh : i64 loc(#loc129) + %h_26 = arith.muli %boh_25, %c32_i64 : i64 loc(#loc130) + %h_27 = arith.addi %h_26, %3 : i64 loc(#loc131) + %h_28 = arith.muli %h_27, %c16384_i64 : i64 loc(#loc132) + %v_29 = arith.muli %5, %c32_i64 : i64 loc(#loc133) + %v_30 = arith.addi %v_29, %3 : i64 loc(#loc134) + %v_31 = arith.muli %v_30, %c128_i64 : i64 loc(#loc135) + %v_32 = tt.addptr %v, %v_31 : !tt.ptr, i64 loc(#loc136) + %k_33 = tt.addptr %k, %v_31 : !tt.ptr, i64 loc(#loc137) + %w_34 = tt.addptr %w, %v_31 : !tt.ptr, i64 loc(#loc138) + %v_new_35 = tt.addptr %v_new, %v_31 : !tt.ptr, i64 loc(#loc139) + %o_v = arith.muli %0, %c64_i64 : i64 loc(#loc140) + %o_v_36 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked2}>> loc(#loc141) + %o_v_37 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc141) + %o_v_38 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked2}>> loc(#loc141) + %o_v_39 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc141) + %o_v_40 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc141) + %o_v_41 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32, #blocked> loc(#loc141) + %o_v_42 = arith.extsi %o_v_36 : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked2}>> to tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked2}>> loc(#loc142) + %o_v_43 = arith.extsi %o_v_37 : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> to tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc142) + %o_v_44 = arith.extsi %o_v_38 : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked2}>> to tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked2}>> loc(#loc142) + %o_v_45 = arith.extsi %o_v_40 : tensor<64xi32, #ttg.slice<{dim = 1, parent = #mma}>> to tensor<64xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc142) + %o_v_46 = arith.extsi %o_v_41 : tensor<64xi32, #blocked> to tensor<64xi64, #blocked> loc(#loc142) + %o_v_47 = tt.splat %o_v : i64 -> tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked2}>> loc(#loc142) + %o_v_48 = arith.addi %o_v_47, %o_v_42 : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked2}>> loc(#loc142) + %m_v = arith.cmpi slt, %o_v_48, %cst_21 : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked2}>> loc(#loc143) + %m_k1 = arith.cmpi slt, %o_v_38, %cst_13 : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked2}>> loc(#loc144) + %m_k1_49 = arith.cmpi slt, %o_v_39, %cst_14 : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc144) + %m_k1_50 = arith.cmpi slt, %o_v_36, %cst_15 : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked2}>> loc(#loc144) + %o_k2 = arith.addi %o_v_38, %cst_16 : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked2}>> loc(#loc145) + %o_k2_51 = arith.addi %o_v_39, %cst_17 : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc145) + %o_k2_52 = arith.addi %o_v_36, %cst_18 : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked2}>> loc(#loc145) + %m_k2 = arith.cmpi slt, %o_k2, %cst_13 : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked2}>> loc(#loc146) + %m_k2_53 = arith.cmpi slt, %o_k2_51, %cst_14 : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc146) + %m_k2_54 = arith.cmpi slt, %o_k2_52, %cst_15 : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked2}>> loc(#loc146) + %m_t = tt.splat %4 : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked2}>> loc(#loc147) + %m_t_55 = tt.splat %4 : i64 -> tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc147) + %m_t_56 = tt.splat %4 : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc147) + %m_t_57 = tt.splat %4 : i64 -> tensor<64xi64, #blocked> loc(#loc147) + %p_h1 = tt.expand_dims %o_v_38 {axis = 1 : i32} : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked2}>> -> tensor<64x1xi32, #blocked2> loc(#loc148) + %p_h1_58 = tt.expand_dims %o_v_39 {axis = 1 : i32} : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<64x1xi32, #blocked1> loc(#loc148) + %p_h1_59 = arith.muli %p_h1, %cst_19 : tensor<64x1xi32, #blocked2> loc(#loc149) + %p_h1_60 = tt.expand_dims %o_v_48 {axis = 0 : i32} : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked2}>> -> tensor<1x64xi64, #blocked2> loc(#loc150) + %p_h1_61 = tt.broadcast %p_h1_60 : tensor<1x64xi64, #blocked2> -> tensor<64x64xi64, #blocked2> loc(#loc151) + %m_h1 = tt.expand_dims %m_k1 {axis = 1 : i32} : tensor<64xi1, #ttg.slice<{dim = 1, parent = #blocked2}>> -> tensor<64x1xi1, #blocked2> loc(#loc152) + %m_h1_62 = tt.expand_dims %m_k1_49 {axis = 1 : i32} : tensor<64xi1, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<64x1xi1, #blocked1> loc(#loc152) + %m_h1_63 = tt.expand_dims %m_v {axis = 0 : i32} : tensor<64xi1, #ttg.slice<{dim = 0, parent = #blocked2}>> -> tensor<1x64xi1, #blocked2> loc(#loc153) + %m_h1_64 = tt.broadcast %m_h1 : tensor<64x1xi1, #blocked2> -> tensor<64x64xi1, #blocked2> loc(#loc154) + %m_h1_65 = tt.broadcast %m_h1_62 : tensor<64x1xi1, #blocked1> -> tensor<64x64xi1, #blocked1> loc(#loc154) + %m_h1_66 = tt.broadcast %m_h1_63 : tensor<1x64xi1, #blocked2> -> tensor<64x64xi1, #blocked2> loc(#loc154) + %m_h1_67 = arith.andi %m_h1_64, %m_h1_66 : tensor<64x64xi1, #blocked2> loc(#loc154) + %p_h2 = tt.expand_dims %o_k2 {axis = 1 : i32} : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked2}>> -> tensor<64x1xi32, #blocked2> loc(#loc155) + %p_h2_68 = tt.expand_dims %o_k2_51 {axis = 1 : i32} : tensor<64xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<64x1xi32, #blocked1> loc(#loc155) + %p_h2_69 = arith.muli %p_h2, %cst_19 : tensor<64x1xi32, #blocked2> loc(#loc156) + %m_h2 = tt.expand_dims %m_k2 {axis = 1 : i32} : tensor<64xi1, #ttg.slice<{dim = 1, parent = #blocked2}>> -> tensor<64x1xi1, #blocked2> loc(#loc157) + %m_h2_70 = tt.expand_dims %m_k2_53 {axis = 1 : i32} : tensor<64xi1, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<64x1xi1, #blocked1> loc(#loc157) + %m_h2_71 = tt.broadcast %m_h2 : tensor<64x1xi1, #blocked2> -> tensor<64x64xi1, #blocked2> loc(#loc158) + %m_h2_72 = tt.broadcast %m_h2_70 : tensor<64x1xi1, #blocked1> -> tensor<64x64xi1, #blocked1> loc(#loc158) + %m_h2_73 = arith.andi %m_h2_71, %m_h1_66 : tensor<64x64xi1, #blocked2> loc(#loc158) + %p_w = tt.splat %w_34 : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked2> loc(#loc159) + %p_w_74 = tt.expand_dims %o_v_36 {axis = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked2}>> -> tensor<1x64xi32, #blocked2> loc(#loc160) + %p_w_75 = tt.broadcast %p_w_74 : tensor<1x64xi32, #blocked2> -> tensor<64x64xi32, #blocked2> loc(#loc161) + %b_w = tt.expand_dims %m_k1_50 {axis = 0 : i32} : tensor<64xi1, #ttg.slice<{dim = 0, parent = #blocked2}>> -> tensor<1x64xi1, #blocked2> loc(#loc162) + %b_w_76 = tt.broadcast %b_w : tensor<1x64xi1, #blocked2> -> tensor<64x64xi1, #blocked2> loc(#loc163) + %p_w_77 = tt.expand_dims %o_k2_52 {axis = 0 : i32} : tensor<64xi32, #ttg.slice<{dim = 0, parent = #blocked2}>> -> tensor<1x64xi32, #blocked2> loc(#loc164) + %p_w_78 = tt.broadcast %p_w_77 : tensor<1x64xi32, #blocked2> -> tensor<64x64xi32, #blocked2> loc(#loc165) + %b_w_79 = tt.expand_dims %m_k2_54 {axis = 0 : i32} : tensor<64xi1, #ttg.slice<{dim = 0, parent = #blocked2}>> -> tensor<1x64xi1, #blocked2> loc(#loc166) + %b_w_80 = tt.broadcast %b_w_79 : tensor<1x64xi1, #blocked2> -> tensor<64x64xi1, #blocked2> loc(#loc167) + %p_v = tt.splat %v_32 : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked2> loc(#loc168) + %p_v_81 = tt.splat %v_new_35 : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked2> loc(#loc169) + %p_g = tt.addptr %g, %v_30 : !tt.ptr, i64 loc(#loc170) + %p_g_82 = tt.splat %p_g : !tt.ptr -> tensor<64x!tt.ptr, #blocked> loc(#loc171) + %p_k = tt.splat %k_33 : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked1> loc(#loc172) + %p_k_83 = tt.addptr %p_k, %p_h1_58 : tensor<64x1x!tt.ptr, #blocked1>, tensor<64x1xi32, #blocked1> loc(#loc172) + %p_k_84 = tt.broadcast %p_k_83 : tensor<64x1x!tt.ptr, #blocked1> -> tensor<64x64x!tt.ptr, #blocked1> loc(#loc173) + %p_k_85 = tt.addptr %p_k, %p_h2_68 : tensor<64x1x!tt.ptr, #blocked1>, tensor<64x1xi32, #blocked1> loc(#loc174) + %p_k_86 = tt.broadcast %p_k_85 : tensor<64x1x!tt.ptr, #blocked1> -> tensor<64x64x!tt.ptr, #blocked1> loc(#loc175) + %b_w_87 = ttg.local_alloc : () -> !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> loc(#loc176) + %b_w_88 = ttg.local_alloc : () -> !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> loc(#loc177) + %b_v = ttg.local_alloc : () -> !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> loc(#loc178) + %b_g_last = ttg.local_alloc : () -> !ttg.memdesc<3x1xf32, #shared1, #smem, mutable> loc(#loc179) + %b_g = ttg.local_alloc : () -> !ttg.memdesc<3x64xf32, #shared1, #smem, mutable> loc(#loc180) + %b_k = ttg.local_alloc : () -> !ttg.memdesc<3x64x64xbf16, #shared2, #smem, mutable> loc(#loc181) + %b_k_89 = ttg.local_alloc : () -> !ttg.memdesc<3x64x64xbf16, #shared2, #smem, mutable> loc(#loc182) + %b_h2 = arith.cmpi sgt, %NT_24, %c0_i32 : i32 loc(#loc230) + %m_t_90 = arith.cmpi slt, %o_v_44, %m_t : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked2}>> loc(#loc147) + %m_t_91 = arith.cmpi slt, %o_v_43, %m_t_55 : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc147) + %m_t_92 = arith.cmpi slt, %o_v_45, %m_t_56 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc147) + %m_t_93 = arith.cmpi slt, %o_v_46, %m_t_57 : tensor<64xi64, #blocked> loc(#loc147) + %p_w_94 = tt.expand_dims %o_v_44 {axis = 1 : i32} : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked2}>> -> tensor<64x1xi64, #blocked2> loc(#loc184) + %p_w_95 = arith.muli %p_w_94, %cst_20 : tensor<64x1xi64, #blocked2> loc(#loc185) + %p_w_96 = tt.addptr %p_w, %p_w_95 : tensor<64x1x!tt.ptr, #blocked2>, tensor<64x1xi64, #blocked2> loc(#loc159) + %p_w_97 = tt.broadcast %p_w_96 : tensor<64x1x!tt.ptr, #blocked2> -> tensor<64x64x!tt.ptr, #blocked2> loc(#loc161) + %p_w_98 = tt.addptr %p_w_97, %p_w_75 : tensor<64x64x!tt.ptr, #blocked2>, tensor<64x64xi32, #blocked2> loc(#loc161) + %b_w_99 = tt.expand_dims %m_t_90 {axis = 1 : i32} : tensor<64xi1, #ttg.slice<{dim = 1, parent = #blocked2}>> -> tensor<64x1xi1, #blocked2> loc(#loc186) + %b_w_100 = tt.broadcast %b_w_99 : tensor<64x1xi1, #blocked2> -> tensor<64x64xi1, #blocked2> loc(#loc163) + %b_w_101 = arith.andi %b_w_100, %b_w_76 : tensor<64x64xi1, #blocked2> loc(#loc163) + %b_w_102 = ttg.memdesc_index %b_w_87[%c0_i32] : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> loc(#loc176) + %b_h2_103 = tt.splat %b_h2 : i1 -> tensor<64x64xi1, #blocked2> loc(#loc230) + %b_h2_104 = arith.andi %b_h2_103, %b_w_101 : tensor<64x64xi1, #blocked2> loc(#loc230) + %b_w_105 = ttg.async_copy_global_to_local %p_w_98, %b_w_102 mask %b_h2_104 other %cst_12 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked2> -> <64x64xbf16, #shared, #smem, mutable> loc(#loc176) + %b_w_106 = ttg.async_commit_group tokens %b_w_105 loc(#loc176) + %p_w_107 = tt.addptr %p_w_97, %p_w_78 : tensor<64x64x!tt.ptr, #blocked2>, tensor<64x64xi32, #blocked2> loc(#loc165) + %b_w_108 = arith.andi %b_w_100, %b_w_80 : tensor<64x64xi1, #blocked2> loc(#loc167) + %b_w_109 = ttg.memdesc_index %b_w_88[%c0_i32] : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> loc(#loc177) + %b_h2_110 = arith.andi %b_h2_103, %b_w_108 : tensor<64x64xi1, #blocked2> loc(#loc230) + %b_w_111 = ttg.async_copy_global_to_local %p_w_107, %b_w_109 mask %b_h2_110 other %cst_12 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked2> -> <64x64xbf16, #shared, #smem, mutable> loc(#loc177) + %b_w_112 = ttg.async_commit_group tokens %b_w_111 loc(#loc177) + %p_v_113 = tt.addptr %p_v, %p_w_95 : tensor<64x1x!tt.ptr, #blocked2>, tensor<64x1xi64, #blocked2> loc(#loc168) + %p_v_114 = tt.broadcast %p_v_113 : tensor<64x1x!tt.ptr, #blocked2> -> tensor<64x64x!tt.ptr, #blocked2> loc(#loc187) + %p_v_115 = tt.addptr %p_v_114, %p_h1_61 : tensor<64x64x!tt.ptr, #blocked2>, tensor<64x64xi64, #blocked2> loc(#loc187) + %b_v_116 = arith.andi %b_w_100, %m_h1_66 : tensor<64x64xi1, #blocked2> loc(#loc188) + %b_v_117 = ttg.memdesc_index %b_v[%c0_i32] : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> loc(#loc178) + %b_h2_118 = arith.andi %b_h2_103, %b_v_116 : tensor<64x64xi1, #blocked2> loc(#loc230) + %b_v_119 = ttg.async_copy_global_to_local %p_v_115, %b_v_117 mask %b_h2_118 other %cst_12 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked2> -> <64x64xbf16, #shared, #smem, mutable> loc(#loc178) + %b_v_120 = ttg.async_commit_group tokens %b_v_119 loc(#loc178) + %last_idx = arith.minsi %T, %c64_i32 : i32 loc(#loc189) + %last_idx_121 = arith.subi %last_idx, %c1_i32 : i32 loc(#loc190) + %b_g_last_122 = arith.muli %last_idx_121, %c32_i32 : i32 loc(#loc191) + %b_g_last_123 = arith.extsi %b_g_last_122 : i32 to i64 loc(#loc192) + %b_g_last_124 = arith.addi %v_29, %b_g_last_123 : i64 loc(#loc192) + %b_g_last_125 = arith.addi %b_g_last_124, %3 : i64 loc(#loc193) + %b_g_last_126 = tt.addptr %g, %b_g_last_125 : !tt.ptr, i64 loc(#loc194) + %b_g_last_127 = tt.splat %b_g_last_126 : !tt.ptr -> tensor<1x!tt.ptr, #blocked> loc(#loc179) + %b_g_last_128 = ttg.memdesc_index %b_g_last[%c0_i32] : !ttg.memdesc<3x1xf32, #shared1, #smem, mutable> -> !ttg.memdesc<1xf32, #shared1, #smem, mutable> loc(#loc179) + %b_h2_129 = tt.splat %b_h2 : i1 -> tensor<1xi1, #blocked> loc(#loc230) + %b_g_last_130 = ttg.async_copy_global_to_local %b_g_last_127, %b_g_last_128 mask %b_h2_129 : tensor<1x!tt.ptr, #blocked> -> <1xf32, #shared1, #smem, mutable> loc(#loc179) + %b_g_last_131 = ttg.async_commit_group tokens %b_g_last_130 loc(#loc179) + %p_g_132 = arith.muli %o_v_46, %cst_9 : tensor<64xi64, #blocked> loc(#loc195) + %p_g_133 = tt.addptr %p_g_82, %p_g_132 : tensor<64x!tt.ptr, #blocked>, tensor<64xi64, #blocked> loc(#loc171) + %b_g_134 = ttg.memdesc_index %b_g[%c0_i32] : !ttg.memdesc<3x64xf32, #shared1, #smem, mutable> -> !ttg.memdesc<64xf32, #shared1, #smem, mutable> loc(#loc180) + %b_h2_135 = tt.splat %b_h2 : i1 -> tensor<64xi1, #blocked> loc(#loc230) + %b_h2_136 = arith.andi %b_h2_135, %m_t_93 : tensor<64xi1, #blocked> loc(#loc230) + %b_g_137 = ttg.async_copy_global_to_local %p_g_133, %b_g_134 mask %b_h2_136 other %cst_10 : tensor<64x!tt.ptr, #blocked> -> <64xf32, #shared1, #smem, mutable> loc(#loc180) + %b_g_138 = ttg.async_commit_group tokens %b_g_137 loc(#loc180) + %p_k_139 = tt.expand_dims %o_v_43 {axis = 0 : i32} : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x64xi64, #blocked1> loc(#loc196) + %p_k_140 = arith.muli %p_k_139, %cst_22 : tensor<1x64xi64, #blocked1> loc(#loc197) + %p_k_141 = tt.broadcast %p_k_140 : tensor<1x64xi64, #blocked1> -> tensor<64x64xi64, #blocked1> loc(#loc173) + %p_k_142 = tt.addptr %p_k_84, %p_k_141 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi64, #blocked1> loc(#loc173) + %b_k_143 = tt.expand_dims %m_t_91 {axis = 0 : i32} : tensor<64xi1, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x64xi1, #blocked1> loc(#loc198) + %b_k_144 = tt.broadcast %b_k_143 : tensor<1x64xi1, #blocked1> -> tensor<64x64xi1, #blocked1> loc(#loc199) + %b_k_145 = arith.andi %m_h1_65, %b_k_144 : tensor<64x64xi1, #blocked1> loc(#loc199) + %b_k_146 = ttg.memdesc_index %b_k[%c0_i32] : !ttg.memdesc<3x64x64xbf16, #shared2, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared2, #smem, mutable> loc(#loc181) + %b_h2_147 = tt.splat %b_h2 : i1 -> tensor<64x64xi1, #blocked1> loc(#loc230) + %b_h2_148 = arith.andi %b_h2_147, %b_k_145 : tensor<64x64xi1, #blocked1> loc(#loc230) + %b_k_149 = ttg.async_copy_global_to_local %p_k_142, %b_k_146 mask %b_h2_148 other %cst_11 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked1> -> <64x64xbf16, #shared2, #smem, mutable> loc(#loc181) + %b_k_150 = ttg.async_commit_group tokens %b_k_149 loc(#loc181) + %p_k_151 = tt.addptr %p_k_86, %p_k_141 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi64, #blocked1> loc(#loc175) + %b_k_152 = arith.andi %m_h2_72, %b_k_144 : tensor<64x64xi1, #blocked1> loc(#loc200) + %b_k_153 = ttg.memdesc_index %b_k_89[%c0_i32] : !ttg.memdesc<3x64x64xbf16, #shared2, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared2, #smem, mutable> loc(#loc182) + %b_h2_154 = arith.andi %b_h2_147, %b_k_152 : tensor<64x64xi1, #blocked1> loc(#loc230) + %b_k_155 = ttg.async_copy_global_to_local %p_k_151, %b_k_153 mask %b_h2_154 other %cst_11 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked1> -> <64x64xbf16, #shared2, #smem, mutable> loc(#loc182) + %b_k_156 = ttg.async_commit_group tokens %b_k_155 loc(#loc182) + %b_h2_157 = arith.cmpi sgt, %NT_24, %c1_i32 : i32 loc(#loc230) + %o_t = arith.addi %o_v_44, %cst_7 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked2}>> loc(#loc201) + %o_t_158 = arith.addi %o_v_43, %cst_6 : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc201) + %o_t_159 = arith.addi %o_v_45, %cst_0 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc201) + %o_t_160 = arith.addi %o_v_46, %cst_5 : tensor<64xi64, #blocked> loc(#loc201) + %m_t_161 = arith.cmpi slt, %o_t, %m_t : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked2}>> loc(#loc147) + %m_t_162 = arith.cmpi slt, %o_t_158, %m_t_55 : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc147) + %m_t_163 = arith.cmpi slt, %o_t_159, %m_t_56 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc147) + %m_t_164 = arith.cmpi slt, %o_t_160, %m_t_57 : tensor<64xi64, #blocked> loc(#loc147) + %p_w_165 = tt.expand_dims %o_t {axis = 1 : i32} : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked2}>> -> tensor<64x1xi64, #blocked2> loc(#loc184) + %p_w_166 = arith.muli %p_w_165, %cst_20 : tensor<64x1xi64, #blocked2> loc(#loc185) + %p_w_167 = tt.addptr %p_w, %p_w_166 : tensor<64x1x!tt.ptr, #blocked2>, tensor<64x1xi64, #blocked2> loc(#loc159) + %p_w_168 = tt.broadcast %p_w_167 : tensor<64x1x!tt.ptr, #blocked2> -> tensor<64x64x!tt.ptr, #blocked2> loc(#loc161) + %p_w_169 = tt.addptr %p_w_168, %p_w_75 : tensor<64x64x!tt.ptr, #blocked2>, tensor<64x64xi32, #blocked2> loc(#loc161) + %b_w_170 = tt.expand_dims %m_t_161 {axis = 1 : i32} : tensor<64xi1, #ttg.slice<{dim = 1, parent = #blocked2}>> -> tensor<64x1xi1, #blocked2> loc(#loc186) + %b_w_171 = tt.broadcast %b_w_170 : tensor<64x1xi1, #blocked2> -> tensor<64x64xi1, #blocked2> loc(#loc163) + %b_w_172 = arith.andi %b_w_171, %b_w_76 : tensor<64x64xi1, #blocked2> loc(#loc163) + %b_w_173 = ttg.memdesc_index %b_w_87[%c1_i32] : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> loc(#loc176) + %b_h2_174 = tt.splat %b_h2_157 : i1 -> tensor<64x64xi1, #blocked2> loc(#loc230) + %b_h2_175 = arith.andi %b_h2_174, %b_w_172 : tensor<64x64xi1, #blocked2> loc(#loc230) + %b_w_176 = ttg.async_copy_global_to_local %p_w_169, %b_w_173 mask %b_h2_175 other %cst_12 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked2> -> <64x64xbf16, #shared, #smem, mutable> loc(#loc176) + %b_w_177 = ttg.async_commit_group tokens %b_w_176 loc(#loc176) + %p_w_178 = tt.addptr %p_w_168, %p_w_78 : tensor<64x64x!tt.ptr, #blocked2>, tensor<64x64xi32, #blocked2> loc(#loc165) + %b_w_179 = arith.andi %b_w_171, %b_w_80 : tensor<64x64xi1, #blocked2> loc(#loc167) + %b_w_180 = ttg.memdesc_index %b_w_88[%c1_i32] : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> loc(#loc177) + %b_h2_181 = arith.andi %b_h2_174, %b_w_179 : tensor<64x64xi1, #blocked2> loc(#loc230) + %b_w_182 = ttg.async_copy_global_to_local %p_w_178, %b_w_180 mask %b_h2_181 other %cst_12 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked2> -> <64x64xbf16, #shared, #smem, mutable> loc(#loc177) + %b_w_183 = ttg.async_commit_group tokens %b_w_182 loc(#loc177) + %p_v_184 = tt.addptr %p_v, %p_w_166 : tensor<64x1x!tt.ptr, #blocked2>, tensor<64x1xi64, #blocked2> loc(#loc168) + %p_v_185 = tt.broadcast %p_v_184 : tensor<64x1x!tt.ptr, #blocked2> -> tensor<64x64x!tt.ptr, #blocked2> loc(#loc187) + %p_v_186 = tt.addptr %p_v_185, %p_h1_61 : tensor<64x64x!tt.ptr, #blocked2>, tensor<64x64xi64, #blocked2> loc(#loc187) + %b_v_187 = arith.andi %b_w_171, %m_h1_66 : tensor<64x64xi1, #blocked2> loc(#loc188) + %b_v_188 = ttg.memdesc_index %b_v[%c1_i32] : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> loc(#loc178) + %b_h2_189 = arith.andi %b_h2_174, %b_v_187 : tensor<64x64xi1, #blocked2> loc(#loc230) + %b_v_190 = ttg.async_copy_global_to_local %p_v_186, %b_v_188 mask %b_h2_189 other %cst_12 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked2> -> <64x64xbf16, #shared, #smem, mutable> loc(#loc178) + %b_v_191 = ttg.async_commit_group tokens %b_v_190 loc(#loc178) + %last_idx_192 = arith.minsi %T, %c128_i32 : i32 loc(#loc189) + %last_idx_193 = arith.subi %last_idx_192, %c1_i32 : i32 loc(#loc190) + %b_g_last_194 = arith.muli %last_idx_193, %c32_i32 : i32 loc(#loc191) + %b_g_last_195 = arith.extsi %b_g_last_194 : i32 to i64 loc(#loc192) + %b_g_last_196 = arith.addi %v_29, %b_g_last_195 : i64 loc(#loc192) + %b_g_last_197 = arith.addi %b_g_last_196, %3 : i64 loc(#loc193) + %b_g_last_198 = tt.addptr %g, %b_g_last_197 : !tt.ptr, i64 loc(#loc194) + %b_g_last_199 = tt.splat %b_g_last_198 : !tt.ptr -> tensor<1x!tt.ptr, #blocked> loc(#loc179) + %b_g_last_200 = ttg.memdesc_index %b_g_last[%c1_i32] : !ttg.memdesc<3x1xf32, #shared1, #smem, mutable> -> !ttg.memdesc<1xf32, #shared1, #smem, mutable> loc(#loc179) + %b_h2_201 = tt.splat %b_h2_157 : i1 -> tensor<1xi1, #blocked> loc(#loc230) + %b_g_last_202 = ttg.async_copy_global_to_local %b_g_last_199, %b_g_last_200 mask %b_h2_201 : tensor<1x!tt.ptr, #blocked> -> <1xf32, #shared1, #smem, mutable> loc(#loc179) + %b_g_last_203 = ttg.async_commit_group tokens %b_g_last_202 loc(#loc179) + %p_g_204 = arith.muli %o_t_160, %cst_9 : tensor<64xi64, #blocked> loc(#loc195) + %p_g_205 = tt.addptr %p_g_82, %p_g_204 : tensor<64x!tt.ptr, #blocked>, tensor<64xi64, #blocked> loc(#loc171) + %b_g_206 = ttg.memdesc_index %b_g[%c1_i32] : !ttg.memdesc<3x64xf32, #shared1, #smem, mutable> -> !ttg.memdesc<64xf32, #shared1, #smem, mutable> loc(#loc180) + %b_h2_207 = tt.splat %b_h2_157 : i1 -> tensor<64xi1, #blocked> loc(#loc230) + %b_h2_208 = arith.andi %b_h2_207, %m_t_164 : tensor<64xi1, #blocked> loc(#loc230) + %b_g_209 = ttg.async_copy_global_to_local %p_g_205, %b_g_206 mask %b_h2_208 other %cst_10 : tensor<64x!tt.ptr, #blocked> -> <64xf32, #shared1, #smem, mutable> loc(#loc180) + %b_g_210 = ttg.async_commit_group tokens %b_g_209 loc(#loc180) + %p_k_211 = tt.expand_dims %o_t_158 {axis = 0 : i32} : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x64xi64, #blocked1> loc(#loc196) + %p_k_212 = arith.muli %p_k_211, %cst_22 : tensor<1x64xi64, #blocked1> loc(#loc197) + %p_k_213 = tt.broadcast %p_k_212 : tensor<1x64xi64, #blocked1> -> tensor<64x64xi64, #blocked1> loc(#loc173) + %p_k_214 = tt.addptr %p_k_84, %p_k_213 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi64, #blocked1> loc(#loc173) + %b_k_215 = tt.expand_dims %m_t_162 {axis = 0 : i32} : tensor<64xi1, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x64xi1, #blocked1> loc(#loc198) + %b_k_216 = tt.broadcast %b_k_215 : tensor<1x64xi1, #blocked1> -> tensor<64x64xi1, #blocked1> loc(#loc199) + %b_k_217 = arith.andi %m_h1_65, %b_k_216 : tensor<64x64xi1, #blocked1> loc(#loc199) + %b_k_218 = ttg.memdesc_index %b_k[%c1_i32] : !ttg.memdesc<3x64x64xbf16, #shared2, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared2, #smem, mutable> loc(#loc181) + %b_h2_219 = tt.splat %b_h2_157 : i1 -> tensor<64x64xi1, #blocked1> loc(#loc230) + %b_h2_220 = arith.andi %b_h2_219, %b_k_217 : tensor<64x64xi1, #blocked1> loc(#loc230) + %b_k_221 = ttg.async_copy_global_to_local %p_k_214, %b_k_218 mask %b_h2_220 other %cst_11 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked1> -> <64x64xbf16, #shared2, #smem, mutable> loc(#loc181) + %b_k_222 = ttg.async_commit_group tokens %b_k_221 loc(#loc181) + %p_k_223 = tt.addptr %p_k_86, %p_k_213 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi64, #blocked1> loc(#loc175) + %b_k_224 = arith.andi %m_h2_72, %b_k_216 : tensor<64x64xi1, #blocked1> loc(#loc200) + %b_k_225 = ttg.memdesc_index %b_k_89[%c1_i32] : !ttg.memdesc<3x64x64xbf16, #shared2, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared2, #smem, mutable> loc(#loc182) + %b_h2_226 = arith.andi %b_h2_219, %b_k_224 : tensor<64x64xi1, #blocked1> loc(#loc230) + %b_k_227 = ttg.async_copy_global_to_local %p_k_223, %b_k_225 mask %b_h2_226 other %cst_11 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked1> -> <64x64xbf16, #shared2, #smem, mutable> loc(#loc182) + %b_k_228 = ttg.async_commit_group tokens %b_k_227 loc(#loc182) + %b_h2_229 = arith.cmpi sgt, %NT_24, %c2_i32 : i32 loc(#loc230) + %o_t_230 = arith.addi %o_v_44, %cst_4 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked2}>> loc(#loc201) + %o_t_231 = arith.addi %o_v_43, %cst_3 : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc201) + %o_t_232 = arith.addi %o_v_45, %cst_1 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc201) + %o_t_233 = arith.addi %o_v_46, %cst_2 : tensor<64xi64, #blocked> loc(#loc201) + %m_t_234 = arith.cmpi slt, %o_t_230, %m_t : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked2}>> loc(#loc147) + %m_t_235 = arith.cmpi slt, %o_t_231, %m_t_55 : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc147) + %m_t_236 = arith.cmpi slt, %o_t_232, %m_t_56 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc147) + %m_t_237 = arith.cmpi slt, %o_t_233, %m_t_57 : tensor<64xi64, #blocked> loc(#loc147) + %p_w_238 = tt.expand_dims %o_t_230 {axis = 1 : i32} : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked2}>> -> tensor<64x1xi64, #blocked2> loc(#loc184) + %p_w_239 = arith.muli %p_w_238, %cst_20 : tensor<64x1xi64, #blocked2> loc(#loc185) + %p_w_240 = tt.addptr %p_w, %p_w_239 : tensor<64x1x!tt.ptr, #blocked2>, tensor<64x1xi64, #blocked2> loc(#loc159) + %p_w_241 = tt.broadcast %p_w_240 : tensor<64x1x!tt.ptr, #blocked2> -> tensor<64x64x!tt.ptr, #blocked2> loc(#loc161) + %p_w_242 = tt.addptr %p_w_241, %p_w_75 : tensor<64x64x!tt.ptr, #blocked2>, tensor<64x64xi32, #blocked2> loc(#loc161) + %b_w_243 = tt.expand_dims %m_t_234 {axis = 1 : i32} : tensor<64xi1, #ttg.slice<{dim = 1, parent = #blocked2}>> -> tensor<64x1xi1, #blocked2> loc(#loc186) + %b_w_244 = tt.broadcast %b_w_243 : tensor<64x1xi1, #blocked2> -> tensor<64x64xi1, #blocked2> loc(#loc163) + %b_w_245 = arith.andi %b_w_244, %b_w_76 : tensor<64x64xi1, #blocked2> loc(#loc163) + %b_w_246 = ttg.memdesc_index %b_w_87[%c2_i32] : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> loc(#loc176) + %b_h2_247 = tt.splat %b_h2_229 : i1 -> tensor<64x64xi1, #blocked2> loc(#loc230) + %b_h2_248 = arith.andi %b_h2_247, %b_w_245 : tensor<64x64xi1, #blocked2> loc(#loc230) + %b_w_249 = ttg.async_copy_global_to_local %p_w_242, %b_w_246 mask %b_h2_248 other %cst_12 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked2> -> <64x64xbf16, #shared, #smem, mutable> loc(#loc176) + %b_w_250 = ttg.async_commit_group tokens %b_w_249 loc(#loc176) + %p_w_251 = tt.addptr %p_w_241, %p_w_78 : tensor<64x64x!tt.ptr, #blocked2>, tensor<64x64xi32, #blocked2> loc(#loc165) + %b_w_252 = arith.andi %b_w_244, %b_w_80 : tensor<64x64xi1, #blocked2> loc(#loc167) + %b_w_253 = ttg.memdesc_index %b_w_88[%c2_i32] : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> loc(#loc177) + %b_h2_254 = arith.andi %b_h2_247, %b_w_252 : tensor<64x64xi1, #blocked2> loc(#loc230) + %b_w_255 = ttg.async_copy_global_to_local %p_w_251, %b_w_253 mask %b_h2_254 other %cst_12 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked2> -> <64x64xbf16, #shared, #smem, mutable> loc(#loc177) + %b_w_256 = ttg.async_commit_group tokens %b_w_255 loc(#loc177) + %p_v_257 = tt.addptr %p_v, %p_w_239 : tensor<64x1x!tt.ptr, #blocked2>, tensor<64x1xi64, #blocked2> loc(#loc168) + %p_v_258 = tt.broadcast %p_v_257 : tensor<64x1x!tt.ptr, #blocked2> -> tensor<64x64x!tt.ptr, #blocked2> loc(#loc187) + %p_v_259 = tt.addptr %p_v_258, %p_h1_61 : tensor<64x64x!tt.ptr, #blocked2>, tensor<64x64xi64, #blocked2> loc(#loc187) + %b_v_260 = arith.andi %b_w_244, %m_h1_66 : tensor<64x64xi1, #blocked2> loc(#loc188) + %b_v_261 = ttg.memdesc_index %b_v[%c2_i32] : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> loc(#loc178) + %b_h2_262 = arith.andi %b_h2_247, %b_v_260 : tensor<64x64xi1, #blocked2> loc(#loc230) + %b_v_263 = ttg.async_copy_global_to_local %p_v_259, %b_v_261 mask %b_h2_262 other %cst_12 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked2> -> <64x64xbf16, #shared, #smem, mutable> loc(#loc178) + %b_v_264 = ttg.async_commit_group tokens %b_v_263 loc(#loc178) + %last_idx_265 = arith.minsi %T, %c192_i32 : i32 loc(#loc189) + %last_idx_266 = arith.subi %last_idx_265, %c1_i32 : i32 loc(#loc190) + %b_g_last_267 = arith.muli %last_idx_266, %c32_i32 : i32 loc(#loc191) + %b_g_last_268 = arith.extsi %b_g_last_267 : i32 to i64 loc(#loc192) + %b_g_last_269 = arith.addi %v_29, %b_g_last_268 : i64 loc(#loc192) + %b_g_last_270 = arith.addi %b_g_last_269, %3 : i64 loc(#loc193) + %b_g_last_271 = tt.addptr %g, %b_g_last_270 : !tt.ptr, i64 loc(#loc194) + %b_g_last_272 = tt.splat %b_g_last_271 : !tt.ptr -> tensor<1x!tt.ptr, #blocked> loc(#loc179) + %b_g_last_273 = ttg.memdesc_index %b_g_last[%c2_i32] : !ttg.memdesc<3x1xf32, #shared1, #smem, mutable> -> !ttg.memdesc<1xf32, #shared1, #smem, mutable> loc(#loc179) + %b_h2_274 = tt.splat %b_h2_229 : i1 -> tensor<1xi1, #blocked> loc(#loc230) + %b_g_last_275 = ttg.async_copy_global_to_local %b_g_last_272, %b_g_last_273 mask %b_h2_274 : tensor<1x!tt.ptr, #blocked> -> <1xf32, #shared1, #smem, mutable> loc(#loc179) + %b_g_last_276 = ttg.async_commit_group tokens %b_g_last_275 loc(#loc179) + %p_g_277 = arith.muli %o_t_233, %cst_9 : tensor<64xi64, #blocked> loc(#loc195) + %p_g_278 = tt.addptr %p_g_82, %p_g_277 : tensor<64x!tt.ptr, #blocked>, tensor<64xi64, #blocked> loc(#loc171) + %b_g_279 = ttg.memdesc_index %b_g[%c2_i32] : !ttg.memdesc<3x64xf32, #shared1, #smem, mutable> -> !ttg.memdesc<64xf32, #shared1, #smem, mutable> loc(#loc180) + %b_h2_280 = tt.splat %b_h2_229 : i1 -> tensor<64xi1, #blocked> loc(#loc230) + %b_h2_281 = arith.andi %b_h2_280, %m_t_237 : tensor<64xi1, #blocked> loc(#loc230) + %b_g_282 = ttg.async_copy_global_to_local %p_g_278, %b_g_279 mask %b_h2_281 other %cst_10 : tensor<64x!tt.ptr, #blocked> -> <64xf32, #shared1, #smem, mutable> loc(#loc180) + %b_g_283 = ttg.async_commit_group tokens %b_g_282 loc(#loc180) + %p_k_284 = tt.expand_dims %o_t_231 {axis = 0 : i32} : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x64xi64, #blocked1> loc(#loc196) + %p_k_285 = arith.muli %p_k_284, %cst_22 : tensor<1x64xi64, #blocked1> loc(#loc197) + %p_k_286 = tt.broadcast %p_k_285 : tensor<1x64xi64, #blocked1> -> tensor<64x64xi64, #blocked1> loc(#loc173) + %p_k_287 = tt.addptr %p_k_84, %p_k_286 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi64, #blocked1> loc(#loc173) + %b_k_288 = tt.expand_dims %m_t_235 {axis = 0 : i32} : tensor<64xi1, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x64xi1, #blocked1> loc(#loc198) + %b_k_289 = tt.broadcast %b_k_288 : tensor<1x64xi1, #blocked1> -> tensor<64x64xi1, #blocked1> loc(#loc199) + %b_k_290 = arith.andi %m_h1_65, %b_k_289 : tensor<64x64xi1, #blocked1> loc(#loc199) + %b_k_291 = ttg.memdesc_index %b_k[%c2_i32] : !ttg.memdesc<3x64x64xbf16, #shared2, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared2, #smem, mutable> loc(#loc181) + %b_h2_292 = tt.splat %b_h2_229 : i1 -> tensor<64x64xi1, #blocked1> loc(#loc230) + %b_h2_293 = arith.andi %b_h2_292, %b_k_290 : tensor<64x64xi1, #blocked1> loc(#loc230) + %b_k_294 = ttg.async_copy_global_to_local %p_k_287, %b_k_291 mask %b_h2_293 other %cst_11 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked1> -> <64x64xbf16, #shared2, #smem, mutable> loc(#loc181) + %b_k_295 = ttg.async_commit_group tokens %b_k_294 loc(#loc181) + %p_k_296 = tt.addptr %p_k_86, %p_k_286 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi64, #blocked1> loc(#loc175) + %b_k_297 = arith.andi %m_h2_72, %b_k_289 : tensor<64x64xi1, #blocked1> loc(#loc200) + %b_k_298 = ttg.memdesc_index %b_k_89[%c2_i32] : !ttg.memdesc<3x64x64xbf16, #shared2, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared2, #smem, mutable> loc(#loc182) + %b_h2_299 = arith.andi %b_h2_292, %b_k_297 : tensor<64x64xi1, #blocked1> loc(#loc230) + %b_k_300 = ttg.async_copy_global_to_local %p_k_296, %b_k_298 mask %b_h2_299 other %cst_11 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked1> -> <64x64xbf16, #shared2, #smem, mutable> loc(#loc182) + %b_k_301 = ttg.async_commit_group tokens %b_k_300 loc(#loc182) + %b_h2_302:37 = scf.for %b_h2_304 = %c0_i32 to %NT_24 step %c1_i32 iter_args(%arg8 = %cst_8, %arg9 = %cst_8, %b_h2_305 = %c2_i32, %b_h2_306 = %c-1_i32, %i_t_int64 = %c0_i64, %i_t_int64_307 = %c1_i64, %i_t_int64_308 = %c2_i64, %b_w_309 = %b_w_106, %b_w_310 = %b_w_177, %b_w_311 = %b_w_250, %b_w_312 = %b_w_112, %b_w_313 = %b_w_183, %b_w_314 = %b_w_256, %b_v_315 = %b_v_120, %b_v_316 = %b_v_191, %b_v_317 = %b_v_264, %p_w_318 = %p_w_95, %p_w_319 = %p_w_166, %p_w_320 = %p_w_239, %b_v_321 = %b_v_116, %b_v_322 = %b_v_187, %b_v_323 = %b_v_260, %b_g_last_324 = %b_g_last_131, %b_g_last_325 = %b_g_last_203, %b_g_last_326 = %b_g_last_276, %b_g_327 = %b_g_138, %b_g_328 = %b_g_210, %b_g_329 = %b_g_283, %b_k_330 = %b_k_150, %b_k_331 = %b_k_222, %b_k_332 = %b_k_295, %b_k_333 = %b_k_156, %b_k_334 = %b_k_228, %b_k_335 = %b_k_301, %m_t_336 = %m_t_92, %m_t_337 = %m_t_163, %m_t_338 = %m_t_236) -> (tensor<64x64xf32, #mma>, tensor<64x64xf32, #mma>, i32, i32, i64, i64, i64, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, tensor<64x1xi64, #blocked2>, tensor<64x1xi64, #blocked2>, tensor<64x1xi64, #blocked2>, tensor<64x64xi1, #blocked2>, tensor<64x64xi1, #blocked2>, tensor<64x64xi1, #blocked2>, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, tensor<64xi1, #ttg.slice<{dim = 1, parent = #mma}>>, tensor<64xi1, #ttg.slice<{dim = 1, parent = #mma}>>, tensor<64xi1, #ttg.slice<{dim = 1, parent = #mma}>>) : i32 { + %b_h2_339 = arith.subi %NT_24, %c3_i32 : i32 loc(#loc230) + %b_h2_340 = arith.cmpi slt, %b_h2_304, %b_h2_339 : i32 loc(#loc230) + %b_h2_341 = arith.addi %b_h2_306, %c1_i32 : i32 loc(#loc230) + %b_h2_342 = arith.cmpi sge, %b_h2_341, %c3_i32 : i32 loc(#loc230) + %b_h2_343 = arith.select %b_h2_342, %c0_i32, %b_h2_341 : i32 loc(#loc230) + %p_h1_344 = arith.muli %i_t_int64, %c524288_i64 : i64 loc(#loc231) + %p_h1_345 = arith.addi %h_28, %p_h1_344 : i64 loc(#loc232) + %p_h1_346 = tt.addptr %h, %p_h1_345 : !tt.ptr, i64 loc(#loc232) + %p_h1_347 = tt.splat %p_h1_346 : !tt.ptr -> tensor<64x1x!tt.ptr, #blocked2> loc(#loc206) + %p_h1_348 = tt.addptr %p_h1_347, %p_h1_59 : tensor<64x1x!tt.ptr, #blocked2>, tensor<64x1xi32, #blocked2> loc(#loc206) + %p_h1_349 = tt.broadcast %p_h1_348 : tensor<64x1x!tt.ptr, #blocked2> -> tensor<64x64x!tt.ptr, #blocked2> loc(#loc151) + %p_h1_350 = tt.addptr %p_h1_349, %p_h1_61 : tensor<64x64x!tt.ptr, #blocked2>, tensor<64x64xi64, #blocked2> loc(#loc151) + %6 = arith.truncf %arg8 : tensor<64x64xf32, #mma> to tensor<64x64xbf16, #mma> loc(#loc90) + %7 = ttg.local_alloc %6 : (tensor<64x64xbf16, #mma>) -> !ttg.memdesc<64x64xbf16, #shared, #smem> loc(#loc90) + %8 = ttg.convert_layout %6 : tensor<64x64xbf16, #mma> -> tensor<64x64xbf16, #blocked2> loc(#loc91) + tt.store %p_h1_350, %8, %m_h1_67 : tensor<64x64x!tt.ptr, #blocked2> loc(#loc91) + %p_h2_351 = tt.addptr %p_h1_347, %p_h2_69 : tensor<64x1x!tt.ptr, #blocked2>, tensor<64x1xi32, #blocked2> loc(#loc207) + %p_h2_352 = tt.broadcast %p_h2_351 : tensor<64x1x!tt.ptr, #blocked2> -> tensor<64x64x!tt.ptr, #blocked2> loc(#loc208) + %p_h2_353 = tt.addptr %p_h2_352, %p_h1_61 : tensor<64x64x!tt.ptr, #blocked2>, tensor<64x64xi64, #blocked2> loc(#loc208) + %9 = arith.truncf %arg9 : tensor<64x64xf32, #mma> to tensor<64x64xbf16, #mma> loc(#loc94) + %10 = ttg.local_alloc %9 : (tensor<64x64xbf16, #mma>) -> !ttg.memdesc<64x64xbf16, #shared, #smem> loc(#loc94) + %11 = ttg.convert_layout %9 : tensor<64x64xbf16, #mma> -> tensor<64x64xbf16, #blocked2> loc(#loc95) + tt.store %p_h2_353, %11, %m_h2_73 : tensor<64x64x!tt.ptr, #blocked2> loc(#loc95) + %b_w_354 = ttg.async_wait %b_w_309, %b_w_312, %b_v_315, %b_g_last_324, %b_g_327, %b_k_330, %b_k_333 {num = 14 : i32} loc(#loc176) + %b_w_355 = ttg.memdesc_index %b_w_87[%b_h2_343] : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> loc(#loc176) + %b_w_356 = ttg.local_load %b_w_355 token %b_w_354 : !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> -> tensor<64x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma, kWidth = 2}>> loc(#loc176) + %12 = ttg.local_load %7 : !ttg.memdesc<64x64xbf16, #shared, #smem> -> tensor<64x64xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> loc(#loc90) + %b_v_357 = tt.dot %b_w_356, %12, %cst_8, inputPrecision = tf32 : tensor<64x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma, kWidth = 2}>> * tensor<64x64xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> -> tensor<64x64xf32, #mma> loc(#loc209) + %b_w_358 = ttg.memdesc_index %b_w_88[%b_h2_343] : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> loc(#loc177) + %b_w_359 = ttg.local_load %b_w_358 token %b_w_354 : !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> -> tensor<64x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma, kWidth = 2}>> loc(#loc177) + %13 = ttg.local_load %10 : !ttg.memdesc<64x64xbf16, #shared, #smem> -> tensor<64x64xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> loc(#loc94) + %b_v_360 = tt.dot %b_w_359, %13, %b_v_357, inputPrecision = tf32 : tensor<64x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma, kWidth = 2}>> * tensor<64x64xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> -> tensor<64x64xf32, #mma> loc(#loc210) + %b_v_361 = ttg.memdesc_index %b_v[%b_h2_343] : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> loc(#loc178) + %b_v_362 = ttg.local_load %b_v_361 token %b_w_354 : !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> -> tensor<64x64xbf16, #mma> loc(#loc211) + %b_v_363 = arith.extf %b_v_362 : tensor<64x64xbf16, #mma> to tensor<64x64xf32, #mma> loc(#loc211) + %b_v_364 = arith.subf %b_v_363, %b_v_360 : tensor<64x64xf32, #mma> loc(#loc211) + %p_v_365 = tt.addptr %p_v_81, %p_w_318 : tensor<64x1x!tt.ptr, #blocked2>, tensor<64x1xi64, #blocked2> loc(#loc169) + %p_v_366 = tt.broadcast %p_v_365 : tensor<64x1x!tt.ptr, #blocked2> -> tensor<64x64x!tt.ptr, #blocked2> loc(#loc212) + %p_v_367 = tt.addptr %p_v_366, %p_h1_61 : tensor<64x64x!tt.ptr, #blocked2>, tensor<64x64xi64, #blocked2> loc(#loc212) + %14 = arith.truncf %b_v_364 : tensor<64x64xf32, #mma> to tensor<64x64xbf16, #mma> loc(#loc100) + %15 = ttg.convert_layout %14 : tensor<64x64xbf16, #mma> -> tensor<64x64xbf16, #blocked2> loc(#loc100) + tt.store %p_v_367, %15, %b_v_321 : tensor<64x64x!tt.ptr, #blocked2> loc(#loc101) + %b_g_last_368 = ttg.memdesc_index %b_g_last[%b_h2_343] : !ttg.memdesc<3x1xf32, #shared1, #smem, mutable> -> !ttg.memdesc<1xf32, #shared1, #smem, mutable> loc(#loc179) + %b_g_last_369 = ttg.local_load %b_g_last_368 token %b_w_354 : !ttg.memdesc<1xf32, #shared1, #smem, mutable> -> tensor<1xf32, #blocked> loc(#loc179) + %b_g_last_370 = tt.unsplat %b_g_last_369 : tensor<1xf32, #blocked> loc(#loc179) + %b_g_371 = ttg.memdesc_index %b_g[%b_h2_343] : !ttg.memdesc<3x64xf32, #shared1, #smem, mutable> -> !ttg.memdesc<64xf32, #shared1, #smem, mutable> loc(#loc180) + %b_g_372 = ttg.local_load %b_g_371 token %b_w_354 : !ttg.memdesc<64xf32, #shared1, #smem, mutable> -> tensor<64xf32, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc180) + %b_v_373 = tt.splat %b_g_last_370 : f32 -> tensor<64xf32, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc213) + %b_v_374 = arith.subf %b_v_373, %b_g_372 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc213) + %b_v_375 = math.exp2 %b_v_374 : tensor<64xf32, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc233) + %b_v_376 = arith.select %m_t_336, %b_v_375, %cst : tensor<64xi1, #ttg.slice<{dim = 1, parent = #mma}>>, tensor<64xf32, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc215) + %b_v_377 = tt.expand_dims %b_v_376 {axis = 1 : i32} : tensor<64xf32, #ttg.slice<{dim = 1, parent = #mma}>> -> tensor<64x1xf32, #mma> loc(#loc216) + %b_v_378 = tt.broadcast %b_v_377 : tensor<64x1xf32, #mma> -> tensor<64x64xf32, #mma> loc(#loc217) + %b_v_379 = arith.mulf %b_v_364, %b_v_378 : tensor<64x64xf32, #mma> loc(#loc217) + %b_g_last_380 = math.exp2 %b_g_last_370 : f32 loc(#loc234) + %b_h1 = tt.splat %b_g_last_380 : f32 -> tensor<64x64xf32, #mma> loc(#loc219) + %b_h1_381 = arith.mulf %arg8, %b_h1 : tensor<64x64xf32, #mma> loc(#loc219) + %b_h2_382 = arith.mulf %arg9, %b_h1 : tensor<64x64xf32, #mma> loc(#loc220) + %b_v_383 = arith.truncf %b_v_379 : tensor<64x64xf32, #mma> to tensor<64x64xbf16, #mma> loc(#loc221) + %b_v_384 = ttg.local_alloc %b_v_383 : (tensor<64x64xbf16, #mma>) -> !ttg.memdesc<64x64xbf16, #shared, #smem> loc(#loc221) + %b_k_385 = ttg.memdesc_index %b_k[%b_h2_343] : !ttg.memdesc<3x64x64xbf16, #shared2, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared2, #smem, mutable> loc(#loc181) + %b_k_386 = ttg.local_load %b_k_385 token %b_w_354 : !ttg.memdesc<64x64xbf16, #shared2, #smem, mutable> -> tensor<64x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma, kWidth = 2}>> loc(#loc181) + %b_v_387 = ttg.local_load %b_v_384 : !ttg.memdesc<64x64xbf16, #shared, #smem> -> tensor<64x64xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> loc(#loc221) + %b_h1_388 = tt.dot %b_k_386, %b_v_387, %b_h1_381, inputPrecision = tf32 : tensor<64x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma, kWidth = 2}>> * tensor<64x64xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> -> tensor<64x64xf32, #mma> loc(#loc222) + %b_k_389 = ttg.memdesc_index %b_k_89[%b_h2_343] : !ttg.memdesc<3x64x64xbf16, #shared2, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared2, #smem, mutable> loc(#loc182) + %b_k_390 = ttg.local_load %b_k_389 token %b_w_354 : !ttg.memdesc<64x64xbf16, #shared2, #smem, mutable> -> tensor<64x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma, kWidth = 2}>> loc(#loc182) + %b_h2_391 = tt.dot %b_k_390, %b_v_387, %b_h2_382, inputPrecision = tf32 : tensor<64x64xbf16, #ttg.dot_op<{opIdx = 0, parent = #mma, kWidth = 2}>> * tensor<64x64xbf16, #ttg.dot_op<{opIdx = 1, parent = #mma, kWidth = 2}>> -> tensor<64x64xf32, #mma> loc(#loc223) + %b_h2_392 = arith.addi %b_h2_305, %c1_i32 : i32 loc(#loc230) + %b_h2_393 = arith.cmpi sge, %b_h2_392, %c3_i32 : i32 loc(#loc230) + %b_h2_394 = arith.select %b_h2_393, %c0_i32, %b_h2_392 : i32 loc(#loc230) + %b_h2_395 = arith.addi %b_h2_304, %c3_i32 : i32 loc(#loc230) + %i_t_int64_396 = arith.extsi %b_h2_395 : i32 to i64 loc(#loc224) + %o_t_397 = arith.muli %i_t_int64_396, %c64_i64 : i64 loc(#loc225) + %o_t_398 = tt.splat %o_t_397 : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked2}>> loc(#loc201) + %o_t_399 = tt.splat %o_t_397 : i64 -> tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc201) + %o_t_400 = tt.splat %o_t_397 : i64 -> tensor<64xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc201) + %o_t_401 = tt.splat %o_t_397 : i64 -> tensor<64xi64, #blocked> loc(#loc201) + %o_t_402 = arith.addi %o_t_398, %o_v_44 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked2}>> loc(#loc201) + %o_t_403 = arith.addi %o_t_399, %o_v_43 : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc201) + %o_t_404 = arith.addi %o_t_400, %o_v_45 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc201) + %o_t_405 = arith.addi %o_t_401, %o_v_46 : tensor<64xi64, #blocked> loc(#loc201) + %m_t_406 = arith.cmpi slt, %o_t_402, %m_t : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked2}>> loc(#loc147) + %m_t_407 = arith.cmpi slt, %o_t_403, %m_t_55 : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc147) + %m_t_408 = arith.cmpi slt, %o_t_404, %m_t_56 : tensor<64xi64, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc147) + %m_t_409 = arith.cmpi slt, %o_t_405, %m_t_57 : tensor<64xi64, #blocked> loc(#loc147) + %p_w_410 = tt.expand_dims %o_t_402 {axis = 1 : i32} : tensor<64xi64, #ttg.slice<{dim = 1, parent = #blocked2}>> -> tensor<64x1xi64, #blocked2> loc(#loc184) + %p_w_411 = arith.muli %p_w_410, %cst_20 : tensor<64x1xi64, #blocked2> loc(#loc185) + %p_w_412 = tt.addptr %p_w, %p_w_411 : tensor<64x1x!tt.ptr, #blocked2>, tensor<64x1xi64, #blocked2> loc(#loc159) + %p_w_413 = tt.broadcast %p_w_412 : tensor<64x1x!tt.ptr, #blocked2> -> tensor<64x64x!tt.ptr, #blocked2> loc(#loc161) + %p_w_414 = tt.addptr %p_w_413, %p_w_75 : tensor<64x64x!tt.ptr, #blocked2>, tensor<64x64xi32, #blocked2> loc(#loc161) + %b_w_415 = tt.expand_dims %m_t_406 {axis = 1 : i32} : tensor<64xi1, #ttg.slice<{dim = 1, parent = #blocked2}>> -> tensor<64x1xi1, #blocked2> loc(#loc186) + %b_w_416 = tt.broadcast %b_w_415 : tensor<64x1xi1, #blocked2> -> tensor<64x64xi1, #blocked2> loc(#loc163) + %b_w_417 = arith.andi %b_w_416, %b_w_76 : tensor<64x64xi1, #blocked2> loc(#loc163) + %b_w_418 = ttg.memdesc_index %b_w_87[%b_h2_394] : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> loc(#loc176) + %b_h2_419 = tt.splat %b_h2_340 : i1 -> tensor<64x64xi1, #blocked2> loc(#loc230) + %b_h2_420 = arith.andi %b_h2_419, %b_w_417 : tensor<64x64xi1, #blocked2> loc(#loc230) + %b_w_421 = ttg.async_copy_global_to_local %p_w_414, %b_w_418 mask %b_h2_420 other %cst_12 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked2> -> <64x64xbf16, #shared, #smem, mutable> loc(#loc176) + %b_w_422 = ttg.async_commit_group tokens %b_w_421 loc(#loc176) + %p_w_423 = tt.addptr %p_w_413, %p_w_78 : tensor<64x64x!tt.ptr, #blocked2>, tensor<64x64xi32, #blocked2> loc(#loc165) + %b_w_424 = arith.andi %b_w_416, %b_w_80 : tensor<64x64xi1, #blocked2> loc(#loc167) + %b_w_425 = ttg.memdesc_index %b_w_88[%b_h2_394] : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> loc(#loc177) + %b_h2_426 = arith.andi %b_h2_419, %b_w_424 : tensor<64x64xi1, #blocked2> loc(#loc230) + %b_w_427 = ttg.async_copy_global_to_local %p_w_423, %b_w_425 mask %b_h2_426 other %cst_12 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked2> -> <64x64xbf16, #shared, #smem, mutable> loc(#loc177) + %b_w_428 = ttg.async_commit_group tokens %b_w_427 loc(#loc177) + %p_v_429 = tt.addptr %p_v, %p_w_411 : tensor<64x1x!tt.ptr, #blocked2>, tensor<64x1xi64, #blocked2> loc(#loc168) + %p_v_430 = tt.broadcast %p_v_429 : tensor<64x1x!tt.ptr, #blocked2> -> tensor<64x64x!tt.ptr, #blocked2> loc(#loc187) + %p_v_431 = tt.addptr %p_v_430, %p_h1_61 : tensor<64x64x!tt.ptr, #blocked2>, tensor<64x64xi64, #blocked2> loc(#loc187) + %b_v_432 = arith.andi %b_w_416, %m_h1_66 : tensor<64x64xi1, #blocked2> loc(#loc188) + %b_v_433 = ttg.memdesc_index %b_v[%b_h2_394] : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared, #smem, mutable> loc(#loc178) + %b_h2_434 = arith.andi %b_h2_419, %b_v_432 : tensor<64x64xi1, #blocked2> loc(#loc230) + %b_v_435 = ttg.async_copy_global_to_local %p_v_431, %b_v_433 mask %b_h2_434 other %cst_12 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked2> -> <64x64xbf16, #shared, #smem, mutable> loc(#loc178) + %b_v_436 = ttg.async_commit_group tokens %b_v_435 loc(#loc178) + %last_idx_437 = arith.addi %b_h2_304, %c4_i32 : i32 loc(#loc226) + %last_idx_438 = arith.muli %last_idx_437, %c64_i32 : i32 loc(#loc227) + %last_idx_439 = arith.minsi %last_idx_438, %T : i32 loc(#loc189) + %last_idx_440 = arith.subi %last_idx_439, %c1_i32 : i32 loc(#loc190) + %b_g_last_441 = arith.muli %last_idx_440, %c32_i32 : i32 loc(#loc191) + %b_g_last_442 = arith.extsi %b_g_last_441 : i32 to i64 loc(#loc192) + %b_g_last_443 = arith.addi %v_29, %b_g_last_442 : i64 loc(#loc192) + %b_g_last_444 = arith.addi %b_g_last_443, %3 : i64 loc(#loc193) + %b_g_last_445 = tt.addptr %g, %b_g_last_444 : !tt.ptr, i64 loc(#loc194) + %b_g_last_446 = tt.splat %b_g_last_445 : !tt.ptr -> tensor<1x!tt.ptr, #blocked> loc(#loc179) + %b_g_last_447 = ttg.memdesc_index %b_g_last[%b_h2_394] : !ttg.memdesc<3x1xf32, #shared1, #smem, mutable> -> !ttg.memdesc<1xf32, #shared1, #smem, mutable> loc(#loc179) + %b_h2_448 = tt.splat %b_h2_340 : i1 -> tensor<1xi1, #blocked> loc(#loc230) + %b_g_last_449 = ttg.async_copy_global_to_local %b_g_last_446, %b_g_last_447 mask %b_h2_448 : tensor<1x!tt.ptr, #blocked> -> <1xf32, #shared1, #smem, mutable> loc(#loc179) + %b_g_last_450 = ttg.async_commit_group tokens %b_g_last_449 loc(#loc179) + %p_g_451 = arith.muli %o_t_405, %cst_9 : tensor<64xi64, #blocked> loc(#loc195) + %p_g_452 = tt.addptr %p_g_82, %p_g_451 : tensor<64x!tt.ptr, #blocked>, tensor<64xi64, #blocked> loc(#loc171) + %b_g_453 = ttg.memdesc_index %b_g[%b_h2_394] : !ttg.memdesc<3x64xf32, #shared1, #smem, mutable> -> !ttg.memdesc<64xf32, #shared1, #smem, mutable> loc(#loc180) + %b_h2_454 = tt.splat %b_h2_340 : i1 -> tensor<64xi1, #blocked> loc(#loc230) + %b_h2_455 = arith.andi %b_h2_454, %m_t_409 : tensor<64xi1, #blocked> loc(#loc230) + %b_g_456 = ttg.async_copy_global_to_local %p_g_452, %b_g_453 mask %b_h2_455 other %cst_10 : tensor<64x!tt.ptr, #blocked> -> <64xf32, #shared1, #smem, mutable> loc(#loc180) + %b_g_457 = ttg.async_commit_group tokens %b_g_456 loc(#loc180) + %p_k_458 = tt.expand_dims %o_t_403 {axis = 0 : i32} : tensor<64xi64, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x64xi64, #blocked1> loc(#loc196) + %p_k_459 = arith.muli %p_k_458, %cst_22 : tensor<1x64xi64, #blocked1> loc(#loc197) + %p_k_460 = tt.broadcast %p_k_459 : tensor<1x64xi64, #blocked1> -> tensor<64x64xi64, #blocked1> loc(#loc173) + %p_k_461 = tt.addptr %p_k_84, %p_k_460 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi64, #blocked1> loc(#loc173) + %b_k_462 = tt.expand_dims %m_t_407 {axis = 0 : i32} : tensor<64xi1, #ttg.slice<{dim = 0, parent = #blocked1}>> -> tensor<1x64xi1, #blocked1> loc(#loc198) + %b_k_463 = tt.broadcast %b_k_462 : tensor<1x64xi1, #blocked1> -> tensor<64x64xi1, #blocked1> loc(#loc199) + %b_k_464 = arith.andi %m_h1_65, %b_k_463 : tensor<64x64xi1, #blocked1> loc(#loc199) + %b_k_465 = ttg.memdesc_index %b_k[%b_h2_394] : !ttg.memdesc<3x64x64xbf16, #shared2, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared2, #smem, mutable> loc(#loc181) + %b_h2_466 = tt.splat %b_h2_340 : i1 -> tensor<64x64xi1, #blocked1> loc(#loc230) + %b_h2_467 = arith.andi %b_h2_466, %b_k_464 : tensor<64x64xi1, #blocked1> loc(#loc230) + %b_k_468 = ttg.async_copy_global_to_local %p_k_461, %b_k_465 mask %b_h2_467 other %cst_11 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked1> -> <64x64xbf16, #shared2, #smem, mutable> loc(#loc181) + %b_k_469 = ttg.async_commit_group tokens %b_k_468 loc(#loc181) + %p_k_470 = tt.addptr %p_k_86, %p_k_460 : tensor<64x64x!tt.ptr, #blocked1>, tensor<64x64xi64, #blocked1> loc(#loc175) + %b_k_471 = arith.andi %m_h2_72, %b_k_463 : tensor<64x64xi1, #blocked1> loc(#loc200) + %b_k_472 = ttg.memdesc_index %b_k_89[%b_h2_394] : !ttg.memdesc<3x64x64xbf16, #shared2, #smem, mutable> -> !ttg.memdesc<64x64xbf16, #shared2, #smem, mutable> loc(#loc182) + %b_h2_473 = arith.andi %b_h2_466, %b_k_471 : tensor<64x64xi1, #blocked1> loc(#loc230) + %b_k_474 = ttg.async_copy_global_to_local %p_k_470, %b_k_472 mask %b_h2_473 other %cst_11 {contiguity = 8 : i32} : tensor<64x64x!tt.ptr, #blocked1> -> <64x64xbf16, #shared2, #smem, mutable> loc(#loc182) + %b_k_475 = ttg.async_commit_group tokens %b_k_474 loc(#loc182) + scf.yield %b_h1_388, %b_h2_391, %b_h2_394, %b_h2_343, %i_t_int64_307, %i_t_int64_308, %i_t_int64_396, %b_w_310, %b_w_311, %b_w_422, %b_w_313, %b_w_314, %b_w_428, %b_v_316, %b_v_317, %b_v_436, %p_w_319, %p_w_320, %p_w_411, %b_v_322, %b_v_323, %b_v_432, %b_g_last_325, %b_g_last_326, %b_g_last_450, %b_g_328, %b_g_329, %b_g_457, %b_k_331, %b_k_332, %b_k_469, %b_k_334, %b_k_335, %b_k_475, %m_t_337, %m_t_338, %m_t_408 : tensor<64x64xf32, #mma>, tensor<64x64xf32, #mma>, i32, i32, i64, i64, i64, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, tensor<64x1xi64, #blocked2>, tensor<64x1xi64, #blocked2>, tensor<64x1xi64, #blocked2>, tensor<64x64xi1, #blocked2>, tensor<64x64xi1, #blocked2>, tensor<64x64xi1, #blocked2>, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, !ttg.async.token, tensor<64xi1, #ttg.slice<{dim = 1, parent = #mma}>>, tensor<64xi1, #ttg.slice<{dim = 1, parent = #mma}>>, tensor<64xi1, #ttg.slice<{dim = 1, parent = #mma}>> loc(#loc230) + } loc(#loc230) + %b_h2_303 = ttg.async_wait {num = 0 : i32} loc(#loc230) + ttg.local_dealloc %b_k_89 : !ttg.memdesc<3x64x64xbf16, #shared2, #smem, mutable> loc(#loc230) + ttg.local_dealloc %b_k : !ttg.memdesc<3x64x64xbf16, #shared2, #smem, mutable> loc(#loc230) + ttg.local_dealloc %b_g : !ttg.memdesc<3x64xf32, #shared1, #smem, mutable> loc(#loc230) + ttg.local_dealloc %b_g_last : !ttg.memdesc<3x1xf32, #shared1, #smem, mutable> loc(#loc230) + ttg.local_dealloc %b_v : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> loc(#loc230) + ttg.local_dealloc %b_w_88 : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> loc(#loc230) + ttg.local_dealloc %b_w_87 : !ttg.memdesc<3x64x64xbf16, #shared, #smem, mutable> loc(#loc230) + tt.return loc(#loc118) + } loc(#loc) +} loc(#loc) +#loc1 = loc(unknown) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":86:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":86:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":88:22) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":88:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":89:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":89:34) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":96:25) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":43:17) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":97:24) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":43:30) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":98:20) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":118:16) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":118:21) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":118:43) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":119:16) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":119:21) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":119:41) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":119:9) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":120:9) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":121:9) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":123:17) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":131:16) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":131:34) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":131:21) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":132:16) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":134:18) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":135:16) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":136:18) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":178:20) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":183:49) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":183:60) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":183:68) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":183:64) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":184:24) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":184:39) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":184:35) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":191:53) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":191:64) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":192:28) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":192:39) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":211:18) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":211:47) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":211:42) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":212:52) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":212:47) +#loc47 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":218:51) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":218:46) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":219:56) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":219:51) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":238:18) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":242:26) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":248:22) +#loc54 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":248:54) +#loc55 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":290:18) +#loc56 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":290:34) +#loc57 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":297:22) +#loc58 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":297:38) +#loc59 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":212:22) +#loc60 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":219:26) +#loc61 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":239:22) +#loc62 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":247:31) +#loc63 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":249:26) +#loc64 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":291:22) +#loc65 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":298:26) +#loc66 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":175:21) +#loc67 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":211:22) +#loc68 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":211:34) +#loc69 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":212:36) +#loc70 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":238:42) +#loc71 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":239:47) +#loc72 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":245:39) +#loc73 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":245:44) +#loc74 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":247:58) +#loc75 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":247:47) +#loc76 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":247:63) +#loc77 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":247:35) +#loc78 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":248:60) +#loc79 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":290:38) +#loc80 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":290:50) +#loc81 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":291:52) +#loc82 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":291:48) +#loc83 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":298:52) +#loc84 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":177:31) +#loc85 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":183:40) +#loc86 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":183:38) +#loc87 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":183:23) +#loc88 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":118:9) +#loc89 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":183:44) +#loc90 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":185:31) +#loc91 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":185:23) +#loc92 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":191:48) +#loc93 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":191:68) +#loc94 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":193:35) +#loc95 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":193:27) +#loc96 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":216:30) +#loc97 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":223:54) +#loc98 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":239:74) +#loc99 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":242:50) +#loc100 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":243:33) +#loc101 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":243:26) +#loc102 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":250:54) +#loc103 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:33) +#loc104 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":250:43) +#loc105 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":250:60) +#loc106 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":250:63) +#loc107 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":250:24) +#loc108 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":251:28) +#loc109 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":252:20) +#loc110 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":254:24) +#loc111 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":288:21) +#loc112 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":295:36) +#loc113 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":302:40) +#loc114 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":176:27) +#loc115 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":177:26) +#loc116 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":245:30) +#loc117 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":245:35) +#loc118 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":318:4) +#loc126 = loc("pid"(#loc2)) +#loc127 = loc("pid"(#loc3)) +#loc128 = loc("NT"(#loc10)) +#loc129 = loc("boh"(#loc12)) +#loc130 = loc("h"(#loc13)) +#loc131 = loc("h"(#loc14)) +#loc132 = loc("h"(#loc15)) +#loc133 = loc("v"(#loc16)) +#loc134 = loc("v"(#loc17)) +#loc135 = loc("v"(#loc18)) +#loc136 = loc("v"(#loc19)) +#loc137 = loc("k"(#loc20)) +#loc138 = loc("w"(#loc21)) +#loc139 = loc("v_new"(#loc22)) +#loc140 = loc("o_v"(#loc23)) +#loc141 = loc("o_v"(#loc24)) +#loc142 = loc("o_v"(#loc25)) +#loc143 = loc("m_v"(#loc26)) +#loc144 = loc("m_k1"(#loc27)) +#loc145 = loc("o_k2"(#loc28)) +#loc146 = loc("m_k2"(#loc29)) +#loc147 = loc("m_t"(#loc30)) +#loc148 = loc("p_h1"(#loc31)) +#loc149 = loc("p_h1"(#loc32)) +#loc150 = loc("p_h1"(#loc33)) +#loc151 = loc("p_h1"(#loc34)) +#loc152 = loc("m_h1"(#loc35)) +#loc153 = loc("m_h1"(#loc36)) +#loc154 = loc("m_h1"(#loc37)) +#loc155 = loc("p_h2"(#loc38)) +#loc156 = loc("p_h2"(#loc39)) +#loc157 = loc("m_h2"(#loc40)) +#loc158 = loc("m_h2"(#loc41)) +#loc159 = loc("p_w"(#loc42)) +#loc160 = loc("p_w"(#loc43)) +#loc161 = loc("p_w"(#loc44)) +#loc162 = loc("b_w"(#loc45)) +#loc163 = loc("b_w"(#loc46)) +#loc164 = loc("p_w"(#loc47)) +#loc165 = loc("p_w"(#loc48)) +#loc166 = loc("b_w"(#loc49)) +#loc167 = loc("b_w"(#loc50)) +#loc168 = loc("p_v"(#loc51)) +#loc169 = loc("p_v"(#loc52)) +#loc170 = loc("p_g"(#loc53)) +#loc171 = loc("p_g"(#loc54)) +#loc172 = loc("p_k"(#loc55)) +#loc173 = loc("p_k"(#loc56)) +#loc174 = loc("p_k"(#loc57)) +#loc175 = loc("p_k"(#loc58)) +#loc176 = loc("b_w"(#loc59)) +#loc177 = loc("b_w"(#loc60)) +#loc178 = loc("b_v"(#loc61)) +#loc179 = loc("b_g_last"(#loc62)) +#loc180 = loc("b_g"(#loc63)) +#loc181 = loc("b_k"(#loc64)) +#loc182 = loc("b_k"(#loc65)) +#loc183 = loc("b_h1"(#loc66)) +#loc184 = loc("p_w"(#loc67)) +#loc185 = loc("p_w"(#loc68)) +#loc186 = loc("b_w"(#loc69)) +#loc187 = loc("p_v"(#loc70)) +#loc188 = loc("b_v"(#loc71)) +#loc189 = loc("last_idx"(#loc72)) +#loc190 = loc("last_idx"(#loc73)) +#loc191 = loc("b_g_last"(#loc74)) +#loc192 = loc("b_g_last"(#loc75)) +#loc193 = loc("b_g_last"(#loc76)) +#loc194 = loc("b_g_last"(#loc77)) +#loc195 = loc("p_g"(#loc78)) +#loc196 = loc("p_k"(#loc79)) +#loc197 = loc("p_k"(#loc80)) +#loc198 = loc("b_k"(#loc81)) +#loc199 = loc("b_k"(#loc82)) +#loc200 = loc("b_k"(#loc83)) +#loc201 = loc("o_t"(#loc84)) +#loc202 = loc("p_h1"(#loc85)) +#loc203 = loc("p_h1"(#loc86)) +#loc204 = loc("p_h1"(#loc87)) +#loc205 = loc("h"(#loc88)) +#loc206 = loc("p_h1"(#loc89)) +#loc207 = loc("p_h2"(#loc92)) +#loc208 = loc("p_h2"(#loc93)) +#loc209 = loc("b_v"(#loc96)) +#loc210 = loc("b_v"(#loc97)) +#loc211 = loc("b_v"(#loc98)) +#loc212 = loc("p_v"(#loc99)) +#loc213 = loc("b_v"(#loc102)) +#loc214 = loc("b_v"(#loc104)) +#loc215 = loc("b_v"(#loc105)) +#loc216 = loc("b_v"(#loc106)) +#loc217 = loc("b_v"(#loc107)) +#loc218 = loc("b_g_last"(#loc108)) +#loc219 = loc("b_h1"(#loc109)) +#loc220 = loc("b_h2"(#loc110)) +#loc221 = loc("b_v"(#loc111)) +#loc222 = loc("b_h1"(#loc112)) +#loc223 = loc("b_h2"(#loc113)) +#loc224 = loc("i_t_int64"(#loc114)) +#loc225 = loc("o_t"(#loc115)) +#loc226 = loc("last_idx"(#loc116)) +#loc227 = loc("last_idx"(#loc117)) +#loc228 = loc(callsite(#loc9 at #loc128)) +#loc229 = loc(callsite(#loc11 at #loc128)) +#loc230 = loc("b_h2"(#loc183)) +#loc231 = loc(fused[#loc202, #loc203]) +#loc232 = loc(fused[#loc204, #loc205]) +#loc233 = loc(callsite(#loc103 at #loc214)) +#loc234 = loc(callsite(#loc103 at #loc218)) diff --git a/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.ttir b/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.ttir new file mode 100644 index 0000000000000000000000000000000000000000..fff4ddac055306692e2599ef9f9aa36d3657ea37 --- /dev/null +++ b/triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.ttir @@ -0,0 +1,413 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":59:0) +#loc120 = loc("k"(#loc)) +#loc121 = loc("v"(#loc)) +#loc122 = loc("w"(#loc)) +#loc123 = loc("v_new"(#loc)) +#loc124 = loc("g"(#loc)) +#loc125 = loc("h"(#loc)) +#loc126 = loc("T"(#loc)) +module { + tt.func public @chunk_gated_delta_rule_fwd_kernel_h_blockdim64(%k: !tt.ptr {tt.divisibility = 16 : i32} loc("k"(#loc)), %v: !tt.ptr {tt.divisibility = 16 : i32} loc("v"(#loc)), %w: !tt.ptr {tt.divisibility = 16 : i32} loc("w"(#loc)), %v_new: !tt.ptr {tt.divisibility = 16 : i32} loc("v_new"(#loc)), %g: !tt.ptr {tt.divisibility = 16 : i32} loc("g"(#loc)), %h: !tt.ptr {tt.divisibility = 16 : i32} loc("h"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %NT = arith.constant 63 : i32 loc(#loc229) + %cst = arith.constant dense<0.000000e+00> : tensor<64x64xbf16> loc(#loc1) + %c524288_i64 = arith.constant 524288 : i64 loc(#loc1) + %c0_i32 = arith.constant 0 : i32 loc(#loc3) + %m_v = arith.constant dense<128> : tensor<64xi64> loc(#loc128) + %h_0 = arith.constant 16384 : i64 loc(#loc129) + %c2_i64 = arith.constant 2 : i64 loc(#loc1) + %cst_1 = arith.constant dense<4096> : tensor<1x64xi64> loc(#loc1) + %cst_2 = arith.constant dense<0.000000e+00> : tensor<64xf32> loc(#loc1) + %cst_3 = arith.constant dense<32> : tensor<64xi64> loc(#loc1) + %cst_4 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc1) + %cst_5 = arith.constant dense<4096> : tensor<64x1xi64> loc(#loc1) + %cst_6 = arith.constant dense<128> : tensor<64x1xi32> loc(#loc1) + %o_k2 = arith.constant dense<64> : tensor<64xi32> loc(#loc130) + %cst_7 = arith.constant dense<128> : tensor<64xi32> loc(#loc1) + %c64_i64 = arith.constant 64 : i64 loc(#loc1) + %c64_i32 = arith.constant 64 : i32 loc(#loc1) + %c1_i32 = arith.constant 1 : i32 loc(#loc1) + %c128_i64 = arith.constant 128 : i64 loc(#loc1) + %c32_i64 = arith.constant 32 : i64 loc(#loc1) + %c32_i32 = arith.constant 32 : i32 loc(#loc1) + %pid = tt.get_program_id x : i32 loc(#loc131) + %pid_8 = arith.extsi %pid : i32 to i64 loc(#loc132) + %0 = arith.remsi %pid_8, %c2_i64 : i64 loc(#loc9) + %1 = arith.divsi %pid_8, %c2_i64 : i64 loc(#loc10) + %2 = arith.divsi %1, %c32_i64 : i64 loc(#loc11) + %3 = arith.remsi %1, %c32_i64 : i64 loc(#loc12) + %4 = arith.extsi %T : i32 to i64 loc(#loc13) + %5 = arith.muli %2, %4 : i64 loc(#loc13) + %NT_9 = arith.addi %T, %NT : i32 loc(#loc230) + %NT_10 = arith.divsi %NT_9, %c64_i32 : i32 loc(#loc231) + %boh = arith.extsi %NT_10 : i32 to i64 loc(#loc133) + %boh_11 = arith.muli %2, %boh : i64 loc(#loc133) + %h_12 = arith.muli %boh_11, %c32_i64 : i64 loc(#loc134) + %h_13 = arith.addi %h_12, %3 : i64 loc(#loc135) + %h_14 = arith.muli %h_13, %h_0 : i64 loc(#loc129) + %v_15 = arith.muli %5, %c32_i64 : i64 loc(#loc136) + %v_16 = arith.addi %v_15, %3 : i64 loc(#loc137) + %v_17 = arith.muli %v_16, %c128_i64 : i64 loc(#loc138) + %v_18 = tt.addptr %v, %v_17 : !tt.ptr, i64 loc(#loc139) + %k_19 = tt.addptr %k, %v_17 : !tt.ptr, i64 loc(#loc140) + %w_20 = tt.addptr %w, %v_17 : !tt.ptr, i64 loc(#loc141) + %v_new_21 = tt.addptr %v_new, %v_17 : !tt.ptr, i64 loc(#loc142) + %o_v = arith.muli %0, %c64_i64 : i64 loc(#loc143) + %o_v_22 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc144) + %o_v_23 = arith.extsi %o_v_22 : tensor<64xi32> to tensor<64xi64> loc(#loc145) + %o_v_24 = tt.splat %o_v : i64 -> tensor<64xi64> loc(#loc145) + %o_v_25 = arith.addi %o_v_24, %o_v_23 : tensor<64xi64> loc(#loc145) + %m_v_26 = arith.cmpi slt, %o_v_25, %m_v : tensor<64xi64> loc(#loc128) + %m_k1 = arith.cmpi slt, %o_v_22, %cst_7 : tensor<64xi32> loc(#loc146) + %o_k2_27 = arith.addi %o_v_22, %o_k2 : tensor<64xi32> loc(#loc130) + %m_k2 = arith.cmpi slt, %o_k2_27, %cst_7 : tensor<64xi32> loc(#loc147) + %b_h2:2 = scf.for %i_t = %c0_i32 to %NT_10 step %c1_i32 iter_args(%b_h1 = %cst_4, %b_h2_28 = %cst_4) -> (tensor<64x64xf32>, tensor<64x64xf32>) : i32 { + %i_t_int64 = arith.extsi %i_t : i32 to i64 loc(#loc149) + %o_t = arith.muli %i_t_int64, %c64_i64 : i64 loc(#loc150) + %o_t_29 = tt.splat %o_t : i64 -> tensor<64xi64> loc(#loc151) + %o_t_30 = arith.addi %o_t_29, %o_v_23 : tensor<64xi64> loc(#loc151) + %m_t = tt.splat %4 : i64 -> tensor<64xi64> loc(#loc152) + %m_t_31 = arith.cmpi slt, %o_t_30, %m_t : tensor<64xi64> loc(#loc152) + %p_h1 = arith.muli %i_t_int64, %c524288_i64 : i64 loc(#loc233) + %p_h1_32 = arith.addi %h_14, %p_h1 : i64 loc(#loc234) + %p_h1_33 = tt.addptr %h, %p_h1_32 : !tt.ptr, i64 loc(#loc234) + %p_h1_34 = tt.expand_dims %o_v_22 {axis = 1 : i32} : tensor<64xi32> -> tensor<64x1xi32> loc(#loc157) + %p_h1_35 = arith.muli %p_h1_34, %cst_6 : tensor<64x1xi32> loc(#loc158) + %p_h1_36 = tt.splat %p_h1_33 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc159) + %p_h1_37 = tt.addptr %p_h1_36, %p_h1_35 : tensor<64x1x!tt.ptr>, tensor<64x1xi32> loc(#loc159) + %p_h1_38 = tt.expand_dims %o_v_25 {axis = 0 : i32} : tensor<64xi64> -> tensor<1x64xi64> loc(#loc160) + %p_h1_39 = tt.broadcast %p_h1_37 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc161) + %p_h1_40 = tt.broadcast %p_h1_38 : tensor<1x64xi64> -> tensor<64x64xi64> loc(#loc161) + %p_h1_41 = tt.addptr %p_h1_39, %p_h1_40 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc161) + %m_h1 = tt.expand_dims %m_k1 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc162) + %m_h1_42 = tt.expand_dims %m_v_26 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc163) + %m_h1_43 = tt.broadcast %m_h1 : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc164) + %m_h1_44 = tt.broadcast %m_h1_42 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc164) + %m_h1_45 = arith.andi %m_h1_43, %m_h1_44 : tensor<64x64xi1> loc(#loc164) + %6 = arith.truncf %b_h1 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc47) + tt.store %p_h1_41, %6, %m_h1_45 : tensor<64x64x!tt.ptr> loc(#loc48) + %p_h2 = tt.expand_dims %o_k2_27 {axis = 1 : i32} : tensor<64xi32> -> tensor<64x1xi32> loc(#loc165) + %p_h2_46 = arith.muli %p_h2, %cst_6 : tensor<64x1xi32> loc(#loc166) + %p_h2_47 = tt.addptr %p_h1_36, %p_h2_46 : tensor<64x1x!tt.ptr>, tensor<64x1xi32> loc(#loc167) + %p_h2_48 = tt.broadcast %p_h2_47 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc168) + %p_h2_49 = tt.addptr %p_h2_48, %p_h1_40 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc168) + %m_h2 = tt.expand_dims %m_k2 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc169) + %m_h2_50 = tt.broadcast %m_h2 : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc170) + %m_h2_51 = arith.andi %m_h2_50, %m_h1_44 : tensor<64x64xi1> loc(#loc170) + %7 = arith.truncf %b_h2_28 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc55) + tt.store %p_h2_49, %7, %m_h2_51 : tensor<64x64x!tt.ptr> loc(#loc56) + %p_w = tt.expand_dims %o_t_30 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc171) + %p_w_52 = arith.muli %p_w, %cst_5 : tensor<64x1xi64> loc(#loc172) + %p_w_53 = tt.splat %w_20 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc173) + %p_w_54 = tt.addptr %p_w_53, %p_w_52 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc173) + %p_w_55 = tt.expand_dims %o_v_22 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc174) + %p_w_56 = tt.broadcast %p_w_54 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc175) + %p_w_57 = tt.broadcast %p_w_55 : tensor<1x64xi32> -> tensor<64x64xi32> loc(#loc175) + %p_w_58 = tt.addptr %p_w_56, %p_w_57 : tensor<64x64x!tt.ptr>, tensor<64x64xi32> loc(#loc175) + %b_w = tt.expand_dims %m_t_31 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc176) + %b_w_59 = tt.expand_dims %m_k1 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc177) + %b_w_60 = tt.broadcast %b_w : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc178) + %b_w_61 = tt.broadcast %b_w_59 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc178) + %b_w_62 = arith.andi %b_w_60, %b_w_61 : tensor<64x64xi1> loc(#loc178) + %b_w_63 = tt.load %p_w_58, %b_w_62, %cst : tensor<64x64x!tt.ptr> loc(#loc179) + %b_v = tt.dot %b_w_63, %6, %cst_4, inputPrecision = tf32 : tensor<64x64xbf16> * tensor<64x64xbf16> -> tensor<64x64xf32> loc(#loc180) + %p_w_64 = tt.expand_dims %o_k2_27 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc181) + %p_w_65 = tt.broadcast %p_w_64 : tensor<1x64xi32> -> tensor<64x64xi32> loc(#loc182) + %p_w_66 = tt.addptr %p_w_56, %p_w_65 : tensor<64x64x!tt.ptr>, tensor<64x64xi32> loc(#loc182) + %b_w_67 = tt.expand_dims %m_k2 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc183) + %b_w_68 = tt.broadcast %b_w_67 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc184) + %b_w_69 = arith.andi %b_w_60, %b_w_68 : tensor<64x64xi1> loc(#loc184) + %b_w_70 = tt.load %p_w_66, %b_w_69, %cst : tensor<64x64x!tt.ptr> loc(#loc185) + %b_v_71 = tt.dot %b_w_70, %7, %b_v, inputPrecision = tf32 : tensor<64x64xbf16> * tensor<64x64xbf16> -> tensor<64x64xf32> loc(#loc186) + %p_v = tt.splat %v_18 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc187) + %p_v_72 = tt.addptr %p_v, %p_w_52 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc187) + %p_v_73 = tt.broadcast %p_v_72 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc188) + %p_v_74 = tt.addptr %p_v_73, %p_h1_40 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc188) + %b_v_75 = arith.andi %b_w_60, %m_h1_44 : tensor<64x64xi1> loc(#loc189) + %b_v_76 = tt.load %p_v_74, %b_v_75, %cst : tensor<64x64x!tt.ptr> loc(#loc190) + %b_v_77 = arith.extf %b_v_76 : tensor<64x64xbf16> to tensor<64x64xf32> loc(#loc191) + %b_v_78 = arith.subf %b_v_77, %b_v_71 : tensor<64x64xf32> loc(#loc191) + %p_v_79 = tt.splat %v_new_21 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc192) + %p_v_80 = tt.addptr %p_v_79, %p_w_52 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc192) + %p_v_81 = tt.broadcast %p_v_80 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc193) + %p_v_82 = tt.addptr %p_v_81, %p_h1_40 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc193) + %8 = arith.truncf %b_v_78 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc80) + tt.store %p_v_82, %8, %b_v_75 : tensor<64x64x!tt.ptr> loc(#loc81) + %last_idx = arith.addi %i_t, %c1_i32 : i32 loc(#loc194) + %last_idx_83 = arith.muli %last_idx, %c64_i32 : i32 loc(#loc195) + %last_idx_84 = arith.minsi %last_idx_83, %T : i32 loc(#loc196) + %last_idx_85 = arith.subi %last_idx_84, %c1_i32 : i32 loc(#loc197) + %b_g_last = arith.muli %last_idx_85, %c32_i32 : i32 loc(#loc198) + %b_g_last_86 = arith.extsi %b_g_last : i32 to i64 loc(#loc199) + %b_g_last_87 = arith.addi %v_15, %b_g_last_86 : i64 loc(#loc199) + %b_g_last_88 = arith.addi %b_g_last_87, %3 : i64 loc(#loc200) + %b_g_last_89 = tt.addptr %g, %b_g_last_88 : !tt.ptr, i64 loc(#loc201) + %b_g_last_90 = tt.load %b_g_last_89 : !tt.ptr loc(#loc202) + %p_g = tt.addptr %g, %v_16 : !tt.ptr, i64 loc(#loc203) + %p_g_91 = arith.muli %o_t_30, %cst_3 : tensor<64xi64> loc(#loc204) + %p_g_92 = tt.splat %p_g : !tt.ptr -> tensor<64x!tt.ptr> loc(#loc205) + %p_g_93 = tt.addptr %p_g_92, %p_g_91 : tensor<64x!tt.ptr>, tensor<64xi64> loc(#loc205) + %b_g = tt.load %p_g_93, %m_t_31, %cst_2 : tensor<64x!tt.ptr> loc(#loc206) + %b_v_94 = tt.splat %b_g_last_90 : f32 -> tensor<64xf32> loc(#loc207) + %b_v_95 = arith.subf %b_v_94, %b_g : tensor<64xf32> loc(#loc207) + %b_v_96 = math.exp2 %b_v_95 : tensor<64xf32> loc(#loc235) + %b_v_97 = arith.select %m_t_31, %b_v_96, %cst_2 : tensor<64xi1>, tensor<64xf32> loc(#loc209) + %b_v_98 = tt.expand_dims %b_v_97 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc210) + %b_v_99 = tt.broadcast %b_v_98 : tensor<64x1xf32> -> tensor<64x64xf32> loc(#loc211) + %b_v_100 = arith.mulf %b_v_78, %b_v_99 : tensor<64x64xf32> loc(#loc211) + %b_g_last_101 = math.exp2 %b_g_last_90 : f32 loc(#loc236) + %b_h1_102 = tt.splat %b_g_last_101 : f32 -> tensor<64x64xf32> loc(#loc213) + %b_h1_103 = arith.mulf %b_h1, %b_h1_102 : tensor<64x64xf32> loc(#loc213) + %b_h2_104 = arith.mulf %b_h2_28, %b_h1_102 : tensor<64x64xf32> loc(#loc214) + %b_v_105 = arith.truncf %b_v_100 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc215) + %p_k = tt.splat %k_19 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc216) + %p_k_106 = tt.addptr %p_k, %p_h1_34 : tensor<64x1x!tt.ptr>, tensor<64x1xi32> loc(#loc216) + %p_k_107 = tt.expand_dims %o_t_30 {axis = 0 : i32} : tensor<64xi64> -> tensor<1x64xi64> loc(#loc217) + %p_k_108 = arith.muli %p_k_107, %cst_1 : tensor<1x64xi64> loc(#loc218) + %p_k_109 = tt.broadcast %p_k_106 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc219) + %p_k_110 = tt.broadcast %p_k_108 : tensor<1x64xi64> -> tensor<64x64xi64> loc(#loc219) + %p_k_111 = tt.addptr %p_k_109, %p_k_110 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc219) + %b_k = tt.expand_dims %m_t_31 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc220) + %b_k_112 = tt.broadcast %b_k : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc221) + %b_k_113 = arith.andi %m_h1_43, %b_k_112 : tensor<64x64xi1> loc(#loc221) + %b_k_114 = tt.load %p_k_111, %b_k_113, %cst : tensor<64x64x!tt.ptr> loc(#loc222) + %b_h1_115 = tt.dot %b_k_114, %b_v_105, %b_h1_103, inputPrecision = tf32 : tensor<64x64xbf16> * tensor<64x64xbf16> -> tensor<64x64xf32> loc(#loc223) + %p_k_116 = tt.addptr %p_k, %p_h2 : tensor<64x1x!tt.ptr>, tensor<64x1xi32> loc(#loc224) + %p_k_117 = tt.broadcast %p_k_116 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc225) + %p_k_118 = tt.addptr %p_k_117, %p_k_110 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc225) + %b_k_119 = arith.andi %m_h2_50, %b_k_112 : tensor<64x64xi1> loc(#loc226) + %b_k_120 = tt.load %p_k_118, %b_k_119, %cst : tensor<64x64x!tt.ptr> loc(#loc227) + %b_h2_121 = tt.dot %b_k_120, %b_v_105, %b_h2_104, inputPrecision = tf32 : tensor<64x64xbf16> * tensor<64x64xbf16> -> tensor<64x64xf32> loc(#loc228) + scf.yield %b_h1_115, %b_h2_121 : tensor<64x64xf32>, tensor<64x64xf32> loc(#loc118) + } loc(#loc232) + tt.return loc(#loc119) + } loc(#loc) +} loc(#loc) +#loc1 = loc(unknown) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":97:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":175:21) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":132:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":118:43) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":135:16) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":86:24) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":86:30) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":88:22) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":88:34) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":89:23) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":89:34) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":96:25) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":43:17) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":43:30) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":98:20) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":118:16) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":118:21) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":119:16) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":119:21) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":119:41) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":119:9) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":120:9) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":121:9) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":123:17) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":131:16) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":131:34) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":131:21) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":134:18) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":136:18) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":176:27) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":177:26) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":177:31) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":178:20) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":183:40) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":183:38) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":183:23) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":118:9) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":183:49) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":183:60) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":183:44) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":183:68) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":183:64) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":184:24) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":184:39) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":184:35) +#loc47 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":185:31) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":185:23) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":191:53) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":191:64) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":191:48) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":191:68) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":192:28) +#loc54 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":192:39) +#loc55 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":193:35) +#loc56 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":193:27) +#loc57 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":211:22) +#loc58 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":211:34) +#loc59 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":211:18) +#loc60 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":211:47) +#loc61 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":211:42) +#loc62 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":212:36) +#loc63 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":212:52) +#loc64 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":212:47) +#loc65 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":212:22) +#loc66 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":216:30) +#loc67 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":218:51) +#loc68 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":218:46) +#loc69 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":219:56) +#loc70 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":219:51) +#loc71 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":219:26) +#loc72 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":223:54) +#loc73 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":238:18) +#loc74 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":238:42) +#loc75 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":239:47) +#loc76 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":239:22) +#loc77 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":239:74) +#loc78 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":242:26) +#loc79 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":242:50) +#loc80 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":243:33) +#loc81 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":243:26) +#loc82 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":245:30) +#loc83 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":245:35) +#loc84 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":245:39) +#loc85 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":245:44) +#loc86 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":247:58) +#loc87 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":247:47) +#loc88 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":247:63) +#loc89 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":247:35) +#loc90 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":247:31) +#loc91 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":248:22) +#loc92 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":248:60) +#loc93 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":248:54) +#loc94 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":249:26) +#loc95 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":250:54) +#loc96 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:33) +#loc97 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":250:43) +#loc98 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":250:60) +#loc99 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":250:63) +#loc100 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":250:24) +#loc101 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":251:28) +#loc102 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":252:20) +#loc103 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":254:24) +#loc104 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":288:21) +#loc105 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":290:18) +#loc106 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":290:38) +#loc107 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":290:50) +#loc108 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":290:34) +#loc109 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":291:52) +#loc110 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":291:48) +#loc111 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":291:22) +#loc112 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":295:36) +#loc113 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":297:22) +#loc114 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":297:38) +#loc115 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":298:52) +#loc116 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":298:26) +#loc117 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":302:40) +#loc118 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":310:8) +#loc119 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_delta_h.py":318:4) +#loc127 = loc("NT"(#loc2)) +#loc128 = loc("m_v"(#loc4)) +#loc129 = loc("h"(#loc5)) +#loc130 = loc("o_k2"(#loc6)) +#loc131 = loc("pid"(#loc7)) +#loc132 = loc("pid"(#loc8)) +#loc133 = loc("boh"(#loc16)) +#loc134 = loc("h"(#loc17)) +#loc135 = loc("h"(#loc18)) +#loc136 = loc("v"(#loc19)) +#loc137 = loc("v"(#loc20)) +#loc138 = loc("v"(#loc21)) +#loc139 = loc("v"(#loc22)) +#loc140 = loc("k"(#loc23)) +#loc141 = loc("w"(#loc24)) +#loc142 = loc("v_new"(#loc25)) +#loc143 = loc("o_v"(#loc26)) +#loc144 = loc("o_v"(#loc27)) +#loc145 = loc("o_v"(#loc28)) +#loc146 = loc("m_k1"(#loc29)) +#loc147 = loc("m_k2"(#loc30)) +#loc148 = loc("b_h1"(#loc3)) +#loc149 = loc("i_t_int64"(#loc31)) +#loc150 = loc("o_t"(#loc32)) +#loc151 = loc("o_t"(#loc33)) +#loc152 = loc("m_t"(#loc34)) +#loc153 = loc("p_h1"(#loc35)) +#loc154 = loc("p_h1"(#loc36)) +#loc155 = loc("p_h1"(#loc37)) +#loc156 = loc("h"(#loc38)) +#loc157 = loc("p_h1"(#loc39)) +#loc158 = loc("p_h1"(#loc40)) +#loc159 = loc("p_h1"(#loc41)) +#loc160 = loc("p_h1"(#loc42)) +#loc161 = loc("p_h1"(#loc43)) +#loc162 = loc("m_h1"(#loc44)) +#loc163 = loc("m_h1"(#loc45)) +#loc164 = loc("m_h1"(#loc46)) +#loc165 = loc("p_h2"(#loc49)) +#loc166 = loc("p_h2"(#loc50)) +#loc167 = loc("p_h2"(#loc51)) +#loc168 = loc("p_h2"(#loc52)) +#loc169 = loc("m_h2"(#loc53)) +#loc170 = loc("m_h2"(#loc54)) +#loc171 = loc("p_w"(#loc57)) +#loc172 = loc("p_w"(#loc58)) +#loc173 = loc("p_w"(#loc59)) +#loc174 = loc("p_w"(#loc60)) +#loc175 = loc("p_w"(#loc61)) +#loc176 = loc("b_w"(#loc62)) +#loc177 = loc("b_w"(#loc63)) +#loc178 = loc("b_w"(#loc64)) +#loc179 = loc("b_w"(#loc65)) +#loc180 = loc("b_v"(#loc66)) +#loc181 = loc("p_w"(#loc67)) +#loc182 = loc("p_w"(#loc68)) +#loc183 = loc("b_w"(#loc69)) +#loc184 = loc("b_w"(#loc70)) +#loc185 = loc("b_w"(#loc71)) +#loc186 = loc("b_v"(#loc72)) +#loc187 = loc("p_v"(#loc73)) +#loc188 = loc("p_v"(#loc74)) +#loc189 = loc("b_v"(#loc75)) +#loc190 = loc("b_v"(#loc76)) +#loc191 = loc("b_v"(#loc77)) +#loc192 = loc("p_v"(#loc78)) +#loc193 = loc("p_v"(#loc79)) +#loc194 = loc("last_idx"(#loc82)) +#loc195 = loc("last_idx"(#loc83)) +#loc196 = loc("last_idx"(#loc84)) +#loc197 = loc("last_idx"(#loc85)) +#loc198 = loc("b_g_last"(#loc86)) +#loc199 = loc("b_g_last"(#loc87)) +#loc200 = loc("b_g_last"(#loc88)) +#loc201 = loc("b_g_last"(#loc89)) +#loc202 = loc("b_g_last"(#loc90)) +#loc203 = loc("p_g"(#loc91)) +#loc204 = loc("p_g"(#loc92)) +#loc205 = loc("p_g"(#loc93)) +#loc206 = loc("b_g"(#loc94)) +#loc207 = loc("b_v"(#loc95)) +#loc208 = loc("b_v"(#loc97)) +#loc209 = loc("b_v"(#loc98)) +#loc210 = loc("b_v"(#loc99)) +#loc211 = loc("b_v"(#loc100)) +#loc212 = loc("b_g_last"(#loc101)) +#loc213 = loc("b_h1"(#loc102)) +#loc214 = loc("b_h2"(#loc103)) +#loc215 = loc("b_v"(#loc104)) +#loc216 = loc("p_k"(#loc105)) +#loc217 = loc("p_k"(#loc106)) +#loc218 = loc("p_k"(#loc107)) +#loc219 = loc("p_k"(#loc108)) +#loc220 = loc("b_k"(#loc109)) +#loc221 = loc("b_k"(#loc110)) +#loc222 = loc("b_k"(#loc111)) +#loc223 = loc("b_h1"(#loc112)) +#loc224 = loc("p_k"(#loc113)) +#loc225 = loc("p_k"(#loc114)) +#loc226 = loc("b_k"(#loc115)) +#loc227 = loc("b_k"(#loc116)) +#loc228 = loc("b_h2"(#loc117)) +#loc229 = loc(callsite(#loc1 at #loc127)) +#loc230 = loc(callsite(#loc14 at #loc127)) +#loc231 = loc(callsite(#loc15 at #loc127)) +#loc232 = loc("b_h2"(#loc148)) +#loc233 = loc(fused[#loc153, #loc154]) +#loc234 = loc(fused[#loc155, #loc156]) +#loc235 = loc(callsite(#loc96 at #loc208)) +#loc236 = loc(callsite(#loc96 at #loc212)) diff --git a/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/__grp__l2norm_fwd_kernel.json b/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/__grp__l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..ae9b6ef04575e9511e2a245b48e5424a03bffc3b --- /dev/null +++ b/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/__grp__l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"child_paths": {"l2norm_fwd_kernel.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/l2norm_fwd_kernel.source", "l2norm_fwd_kernel.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/l2norm_fwd_kernel.ttir", "l2norm_fwd_kernel.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/l2norm_fwd_kernel.ttgir", "l2norm_fwd_kernel.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/l2norm_fwd_kernel.llir", "l2norm_fwd_kernel.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/l2norm_fwd_kernel.ptx", "l2norm_fwd_kernel.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/l2norm_fwd_kernel.cubin", "l2norm_fwd_kernel.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/l2norm_fwd_kernel.json"}} \ No newline at end of file diff --git a/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/l2norm_fwd_kernel.cubin b/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/l2norm_fwd_kernel.cubin new file mode 100644 index 0000000000000000000000000000000000000000..ffd41c7b6f95688683780b2063c0d02e46bc4074 Binary files /dev/null and b/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/l2norm_fwd_kernel.cubin differ diff --git a/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/l2norm_fwd_kernel.json b/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/l2norm_fwd_kernel.json new file mode 100644 index 0000000000000000000000000000000000000000..358980095c9d027f4c9199c8d04536f1fe46fb4c --- /dev/null +++ b/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/l2norm_fwd_kernel.json @@ -0,0 +1 @@ +{"hash": "136a929b6b9008291802d4594d3bd63ca6ebf2d39b2078002f3eb6bfd9803fd7", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 8, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 128, "tmem_size": 0, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "l2norm_fwd_kernel"} \ No newline at end of file diff --git a/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/l2norm_fwd_kernel.llir b/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/l2norm_fwd_kernel.llir new file mode 100644 index 0000000000000000000000000000000000000000..94f993d4c40855f73be130ae3d54722765ed406c --- /dev/null +++ b/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/l2norm_fwd_kernel.llir @@ -0,0 +1,378 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 +@.str.2 = private unnamed_addr constant [17 x i8] c"__CUDA_PREC_SQRT\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @l2norm_fwd_kernel(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, float %3, i32 %4, ptr addrspace(1) readnone captures(none) %5, ptr addrspace(1) readnone captures(none) %6) local_unnamed_addr #0 !dbg !5 { + %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13 + %9 = zext nneg i32 %8 to i64, !dbg !14 + %10 = shl nuw nsw i64 %9, 5, !dbg !15 + %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !16 + %12 = and i32 %11, 240, !dbg !16 + %13 = lshr exact i32 %12, 4, !dbg !16 + %14 = or disjoint i32 %13, 16, !dbg !16 + %15 = zext nneg i32 %13 to i64, !dbg !17 + %16 = zext nneg i32 %14 to i64, !dbg !17 + %17 = or disjoint i64 %10, %15, !dbg !17 + %18 = or disjoint i64 %10, %16, !dbg !17 + %19 = sext i32 %4 to i64, !dbg !18 + %20 = icmp slt i64 %17, %19, !dbg !18 + %21 = icmp slt i64 %18, %19, !dbg !18 + %22 = shl nuw nsw i32 %11, 3, !dbg !19 + %23 = and i32 %22, 120, !dbg !19 + %24 = shl nuw nsw i64 %17, 7, !dbg !20 + %25 = shl nuw nsw i64 %18, 7, !dbg !20 + %26 = getelementptr bfloat, ptr addrspace(1) %0, i64 %24, !dbg !21 + %27 = getelementptr bfloat, ptr addrspace(1) %0, i64 %25, !dbg !21 + %28 = zext nneg i32 %23 to i64, !dbg !22 + %29 = getelementptr bfloat, ptr addrspace(1) %26, i64 %28, !dbg !22 + %30 = getelementptr bfloat, ptr addrspace(1) %27, i64 %28, !dbg !22 + %31 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %29, i1 %20) #6, !dbg !23 + %32 = extractvalue { i32, i32, i32, i32 } %31, 0, !dbg !23 + %33 = bitcast i32 %32 to <2 x bfloat>, !dbg !23 + %34 = extractvalue { i32, i32, i32, i32 } %31, 1, !dbg !23 + %35 = bitcast i32 %34 to <2 x bfloat>, !dbg !23 + %36 = extractvalue { i32, i32, i32, i32 } %31, 2, !dbg !23 + %37 = bitcast i32 %36 to <2 x bfloat>, !dbg !23 + %38 = extractvalue { i32, i32, i32, i32 } %31, 3, !dbg !23 + %39 = bitcast i32 %38 to <2 x bfloat>, !dbg !23 + %40 = extractelement <2 x bfloat> %33, i64 0, !dbg !23 + %41 = extractelement <2 x bfloat> %33, i64 1, !dbg !23 + %42 = extractelement <2 x bfloat> %35, i64 0, !dbg !23 + %43 = extractelement <2 x bfloat> %35, i64 1, !dbg !23 + %44 = extractelement <2 x bfloat> %37, i64 0, !dbg !23 + %45 = extractelement <2 x bfloat> %37, i64 1, !dbg !23 + %46 = extractelement <2 x bfloat> %39, i64 0, !dbg !23 + %47 = extractelement <2 x bfloat> %39, i64 1, !dbg !23 + %48 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %30, i1 %21) #6, !dbg !23 + %49 = extractvalue { i32, i32, i32, i32 } %48, 0, !dbg !23 + %50 = bitcast i32 %49 to <2 x bfloat>, !dbg !23 + %51 = extractvalue { i32, i32, i32, i32 } %48, 1, !dbg !23 + %52 = bitcast i32 %51 to <2 x bfloat>, !dbg !23 + %53 = extractvalue { i32, i32, i32, i32 } %48, 2, !dbg !23 + %54 = bitcast i32 %53 to <2 x bfloat>, !dbg !23 + %55 = extractvalue { i32, i32, i32, i32 } %48, 3, !dbg !23 + %56 = bitcast i32 %55 to <2 x bfloat>, !dbg !23 + %57 = extractelement <2 x bfloat> %50, i64 0, !dbg !23 + %58 = extractelement <2 x bfloat> %50, i64 1, !dbg !23 + %59 = extractelement <2 x bfloat> %52, i64 0, !dbg !23 + %60 = extractelement <2 x bfloat> %52, i64 1, !dbg !23 + %61 = extractelement <2 x bfloat> %54, i64 0, !dbg !23 + %62 = extractelement <2 x bfloat> %54, i64 1, !dbg !23 + %63 = extractelement <2 x bfloat> %56, i64 0, !dbg !23 + %64 = extractelement <2 x bfloat> %56, i64 1, !dbg !23 + %65 = fpext bfloat %40 to float, !dbg !24 + %66 = fpext bfloat %41 to float, !dbg !24 + %67 = fpext bfloat %42 to float, !dbg !24 + %68 = fpext bfloat %43 to float, !dbg !24 + %69 = fpext bfloat %44 to float, !dbg !24 + %70 = fpext bfloat %45 to float, !dbg !24 + %71 = fpext bfloat %46 to float, !dbg !24 + %72 = fpext bfloat %47 to float, !dbg !24 + %73 = fpext bfloat %57 to float, !dbg !24 + %74 = fpext bfloat %58 to float, !dbg !24 + %75 = fpext bfloat %59 to float, !dbg !24 + %76 = fpext bfloat %60 to float, !dbg !24 + %77 = fpext bfloat %61 to float, !dbg !24 + %78 = fpext bfloat %62 to float, !dbg !24 + %79 = fpext bfloat %63 to float, !dbg !24 + %80 = fpext bfloat %64 to float, !dbg !24 + %81 = fmul float %65, %65, !dbg !25 + %82 = fmul float %66, %66, !dbg !25 + %83 = fmul float %67, %67, !dbg !25 + %84 = fmul float %68, %68, !dbg !25 + %85 = fmul float %69, %69, !dbg !25 + %86 = fmul float %70, %70, !dbg !25 + %87 = fmul float %71, %71, !dbg !25 + %88 = fmul float %72, %72, !dbg !25 + %89 = fmul float %73, %73, !dbg !25 + %90 = fmul float %74, %74, !dbg !25 + %91 = fmul float %75, %75, !dbg !25 + %92 = fmul float %76, %76, !dbg !25 + %93 = fmul float %77, %77, !dbg !25 + %94 = fmul float %78, %78, !dbg !25 + %95 = fmul float %79, %79, !dbg !25 + %96 = fmul float %80, %80, !dbg !25 + %97 = fadd float %81, %82, !dbg !26 + %98 = fadd float %83, %97, !dbg !26 + %99 = fadd float %84, %98, !dbg !26 + %100 = fadd float %85, %99, !dbg !26 + %101 = fadd float %86, %100, !dbg !26 + %102 = fadd float %87, %101, !dbg !26 + %103 = fadd float %88, %102, !dbg !26 + %104 = fadd float %89, %90, !dbg !26 + %105 = fadd float %91, %104, !dbg !26 + %106 = fadd float %92, %105, !dbg !26 + %107 = fadd float %93, %106, !dbg !26 + %108 = fadd float %94, %107, !dbg !26 + %109 = fadd float %95, %108, !dbg !26 + %110 = fadd float %96, %109, !dbg !26 + %111 = bitcast float %103 to i32, !dbg !29 + %112 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %111, i32 8, i32 31), !dbg !29 + %113 = bitcast i32 %112 to float, !dbg !29 + %114 = fadd float %103, %113, !dbg !26 + %115 = bitcast float %114 to i32, !dbg !29 + %116 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %115, i32 4, i32 31), !dbg !29 + %117 = bitcast i32 %116 to float, !dbg !29 + %118 = fadd float %114, %117, !dbg !26 + %119 = bitcast float %118 to i32, !dbg !29 + %120 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %119, i32 2, i32 31), !dbg !29 + %121 = bitcast i32 %120 to float, !dbg !29 + %122 = fadd float %118, %121, !dbg !26 + %123 = bitcast float %122 to i32, !dbg !29 + %124 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %123, i32 1, i32 31), !dbg !29 + %125 = bitcast i32 %124 to float, !dbg !29 + %126 = fadd float %122, %125, !dbg !26 + %127 = bitcast float %110 to i32, !dbg !29 + %128 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %127, i32 8, i32 31), !dbg !29 + %129 = bitcast i32 %128 to float, !dbg !29 + %130 = fadd float %110, %129, !dbg !26 + %131 = bitcast float %130 to i32, !dbg !29 + %132 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %131, i32 4, i32 31), !dbg !29 + %133 = bitcast i32 %132 to float, !dbg !29 + %134 = fadd float %130, %133, !dbg !26 + %135 = bitcast float %134 to i32, !dbg !29 + %136 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %135, i32 2, i32 31), !dbg !29 + %137 = bitcast i32 %136 to float, !dbg !29 + %138 = fadd float %134, %137, !dbg !26 + %139 = bitcast float %138 to i32, !dbg !29 + %140 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %139, i32 1, i32 31), !dbg !29 + %141 = bitcast i32 %140 to float, !dbg !29 + %142 = fadd float %138, %141, !dbg !26 + %143 = fadd float %3, %126, !dbg !32 + %144 = fadd float %3, %142, !dbg !32 + %145 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i = icmp eq i32 %145, 0, !dbg !33 + %146 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i = icmp eq i32 %146, 0, !dbg !33 + br i1 %.not.i, label %152, label %147, !dbg !33 + +147: ; preds = %7 + br i1 %.not1.i, label %150, label %148, !dbg !33 + +148: ; preds = %147 + %149 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %143) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +150: ; preds = %147 + %151 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %143) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +152: ; preds = %7 + br i1 %.not1.i, label %155, label %153, !dbg !33 + +153: ; preds = %152 + %154 = tail call float @llvm.nvvm.sqrt.rn.f(float %143) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +155: ; preds = %152 + %156 = tail call float @llvm.nvvm.sqrt.approx.f(float %143) #6, !dbg !33 + br label %__nv_sqrtf.exit, !dbg !33 + +__nv_sqrtf.exit: ; preds = %148, %150, %153, %155 + %.0.i = phi float [ %149, %148 ], [ %151, %150 ], [ %154, %153 ], [ %156, %155 ], !dbg !33 + %157 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33 + %.not.i1 = icmp eq i32 %157, 0, !dbg !33 + %158 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33 + %.not1.i4 = icmp eq i32 %158, 0, !dbg !33 + br i1 %.not.i1, label %164, label %159, !dbg !33 + +159: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %162, label %160, !dbg !33 + +160: ; preds = %159 + %161 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %144) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +162: ; preds = %159 + %163 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %144) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +164: ; preds = %__nv_sqrtf.exit + br i1 %.not1.i4, label %167, label %165, !dbg !33 + +165: ; preds = %164 + %166 = tail call float @llvm.nvvm.sqrt.rn.f(float %144) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +167: ; preds = %164 + %168 = tail call float @llvm.nvvm.sqrt.approx.f(float %144) #6, !dbg !33 + br label %__nv_sqrtf.exit5, !dbg !33 + +__nv_sqrtf.exit5: ; preds = %160, %162, %165, %167 + %.0.i3 = phi float [ %161, %160 ], [ %163, %162 ], [ %166, %165 ], [ %168, %167 ], !dbg !33 + %169 = and i32 %11, 31, !dbg !16 + %170 = zext nneg i32 %169 to i64, !dbg !17 + %171 = or disjoint i64 %10, %170, !dbg !17 + %172 = getelementptr float, ptr addrspace(1) %2, i64 %171, !dbg !34 + %173 = getelementptr bfloat, ptr addrspace(1) %1, i64 %25, !dbg !35 + %174 = getelementptr bfloat, ptr addrspace(1) %173, i64 %28, !dbg !36 + %175 = getelementptr bfloat, ptr addrspace(1) %1, i64 %24, !dbg !35 + %176 = getelementptr bfloat, ptr addrspace(1) %175, i64 %28, !dbg !36 + %177 = icmp slt i64 %171, %19, !dbg !18 + %178 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i), !dbg !37 + %179 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i3), !dbg !37 + %180 = fmul float %178, %65, !dbg !38 + %181 = fmul float %178, %66, !dbg !38 + %182 = fmul float %178, %67, !dbg !38 + %183 = fmul float %178, %68, !dbg !38 + %184 = fmul float %178, %69, !dbg !38 + %185 = fmul float %178, %70, !dbg !38 + %186 = fmul float %178, %71, !dbg !38 + %187 = fmul float %178, %72, !dbg !38 + %188 = fmul float %179, %73, !dbg !38 + %189 = fmul float %179, %74, !dbg !38 + %190 = fmul float %179, %75, !dbg !38 + %191 = fmul float %179, %76, !dbg !38 + %192 = fmul float %179, %77, !dbg !38 + %193 = fmul float %179, %78, !dbg !38 + %194 = fmul float %179, %79, !dbg !38 + %195 = fmul float %179, %80, !dbg !38 + %196 = fptrunc float %180 to bfloat, !dbg !39 + %197 = fptrunc float %181 to bfloat, !dbg !39 + %198 = fptrunc float %182 to bfloat, !dbg !39 + %199 = fptrunc float %183 to bfloat, !dbg !39 + %200 = fptrunc float %184 to bfloat, !dbg !39 + %201 = fptrunc float %185 to bfloat, !dbg !39 + %202 = fptrunc float %186 to bfloat, !dbg !39 + %203 = fptrunc float %187 to bfloat, !dbg !39 + %204 = fptrunc float %188 to bfloat, !dbg !39 + %205 = fptrunc float %189 to bfloat, !dbg !39 + %206 = fptrunc float %190 to bfloat, !dbg !39 + %207 = fptrunc float %191 to bfloat, !dbg !39 + %208 = fptrunc float %192 to bfloat, !dbg !39 + %209 = fptrunc float %193 to bfloat, !dbg !39 + %210 = fptrunc float %194 to bfloat, !dbg !39 + %211 = fptrunc float %195 to bfloat, !dbg !39 + %212 = insertelement <2 x bfloat> poison, bfloat %196, i64 0, !dbg !40 + %213 = insertelement <2 x bfloat> %212, bfloat %197, i64 1, !dbg !40 + %214 = bitcast <2 x bfloat> %213 to i32, !dbg !40 + %215 = insertelement <2 x bfloat> poison, bfloat %198, i64 0, !dbg !40 + %216 = insertelement <2 x bfloat> %215, bfloat %199, i64 1, !dbg !40 + %217 = bitcast <2 x bfloat> %216 to i32, !dbg !40 + %218 = insertelement <2 x bfloat> poison, bfloat %200, i64 0, !dbg !40 + %219 = insertelement <2 x bfloat> %218, bfloat %201, i64 1, !dbg !40 + %220 = bitcast <2 x bfloat> %219 to i32, !dbg !40 + %221 = insertelement <2 x bfloat> poison, bfloat %202, i64 0, !dbg !40 + %222 = insertelement <2 x bfloat> %221, bfloat %203, i64 1, !dbg !40 + %223 = bitcast <2 x bfloat> %222 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %214, i32 %217, i32 %220, i32 %223, ptr addrspace(1) %176, i1 %20) #6, !dbg !40 + %224 = insertelement <2 x bfloat> poison, bfloat %204, i64 0, !dbg !40 + %225 = insertelement <2 x bfloat> %224, bfloat %205, i64 1, !dbg !40 + %226 = bitcast <2 x bfloat> %225 to i32, !dbg !40 + %227 = insertelement <2 x bfloat> poison, bfloat %206, i64 0, !dbg !40 + %228 = insertelement <2 x bfloat> %227, bfloat %207, i64 1, !dbg !40 + %229 = bitcast <2 x bfloat> %228 to i32, !dbg !40 + %230 = insertelement <2 x bfloat> poison, bfloat %208, i64 0, !dbg !40 + %231 = insertelement <2 x bfloat> %230, bfloat %209, i64 1, !dbg !40 + %232 = bitcast <2 x bfloat> %231 to i32, !dbg !40 + %233 = insertelement <2 x bfloat> poison, bfloat %210, i64 0, !dbg !40 + %234 = insertelement <2 x bfloat> %233, bfloat %211, i64 1, !dbg !40 + %235 = bitcast <2 x bfloat> %234 to i32, !dbg !40 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %226, i32 %229, i32 %232, i32 %235, ptr addrspace(1) %174, i1 %21) #6, !dbg !40 + %236 = lshr exact i32 %12, 1, !dbg !41 + %237 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %236, !dbg !41 + %238 = bitcast float %178 to i32, !dbg !41 + %239 = bitcast float %179 to i32, !dbg !41 + tail call void asm sideeffect "st.shared.v2.b32 [ $0 + 0 ], { $1, $2 };", "r,r,r"(ptr addrspace(3) %237, i32 %238, i32 %239) #6, !dbg !41 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !41 + %240 = lshr i32 %11, 2, !dbg !41 + %241 = and i32 %240, 4, !dbg !41 + %242 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %23, !dbg !41 + %243 = getelementptr inbounds nuw i8, ptr addrspace(3) %242, i32 %241, !dbg !41 + %244 = load i32, ptr addrspace(3) %243, align 4, !dbg !41 + %245 = and i32 %11, 224, !dbg !41 + %246 = icmp eq i32 %245, 0, !dbg !41 + %247 = and i1 %246, %177, !dbg !41 + tail call void asm sideeffect "@$2 st.global.b32 [ $1 + 0 ], { $0 };", "r,l,b"(i32 %244, ptr addrspace(1) %172, i1 %247) #6, !dbg !41 + ret void, !dbg !42 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.div.full(float, float) #3 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #4 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #5 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.ftz.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.rn.f(float) #3 + +; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.sqrt.approx.f(float) #3 + +attributes #0 = { nounwind "nvvm.reqntid"="256" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) } +attributes #4 = { convergent nocallback nounwind } +attributes #5 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #6 = { nounwind } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "l2norm_fwd_kernel", linkageName: "l2norm_fwd_kernel", scope: !1, file: !1, line: 69, type: !6, scopeLine: 69, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !10, !11, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 70, column: 24, scope: !5) +!14 = !DILocation(line: 70, column: 30, scope: !5) +!15 = !DILocation(line: 71, column: 16, scope: !5) +!16 = !DILocation(line: 71, column: 34, scope: !5) +!17 = !DILocation(line: 71, column: 21, scope: !5) +!18 = !DILocation(line: 73, column: 16, scope: !5) +!19 = !DILocation(line: 74, column: 30, scope: !5) +!20 = !DILocation(line: 75, column: 29, scope: !5) +!21 = !DILocation(line: 75, column: 14, scope: !5) +!22 = !DILocation(line: 75, column: 33, scope: !5) +!23 = !DILocation(line: 79, column: 18, scope: !5) +!24 = !DILocation(line: 79, column: 47, scope: !5) +!25 = !DILocation(line: 80, column: 38, scope: !5) +!26 = !DILocation(line: 263, column: 15, scope: !27, inlinedAt: !29) +!27 = distinct !DILexicalBlockFile(scope: !5, file: !28, discriminator: 0) +!28 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!29 = !DILocation(line: 293, column: 36, scope: !27, inlinedAt: !30) +!30 = !DILocation(line: 80, column: 43, scope: !31) +!31 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!32 = !DILocation(line: 80, column: 48, scope: !5) +!33 = !DILocation(line: 80, column: 25, scope: !5) +!34 = !DILocation(line: 77, column: 20, scope: !5) +!35 = !DILocation(line: 76, column: 14, scope: !5) +!36 = !DILocation(line: 76, column: 33, scope: !5) +!37 = !DILocation(line: 80, column: 17, scope: !5) +!38 = !DILocation(line: 81, column: 16, scope: !5) +!39 = !DILocation(line: 83, column: 25, scope: !5) +!40 = !DILocation(line: 83, column: 18, scope: !5) +!41 = !DILocation(line: 84, column: 21, scope: !5) +!42 = !DILocation(line: 84, column: 4, scope: !5) diff --git a/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/l2norm_fwd_kernel.ptx b/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/l2norm_fwd_kernel.ptx new file mode 100644 index 0000000000000000000000000000000000000000..33315714b4eb618901a84ff17adedff641c543be --- /dev/null +++ b/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/l2norm_fwd_kernel.ptx @@ -0,0 +1,481 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl l2norm_fwd_kernel // -- Begin function l2norm_fwd_kernel +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; +.global .align 1 .b8 _$_str_$_2[17] = {95, 95, 67, 85, 68, 65, 95, 80, 82, 69, 67, 95, 83, 81, 82, 84}; + // @l2norm_fwd_kernel +.visible .entry l2norm_fwd_kernel( + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_0, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_1, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_2, + .param .f32 l2norm_fwd_kernel_param_3, + .param .u32 l2norm_fwd_kernel_param_4, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_5, + .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_6 +) +.reqntid 256 +{ + .reg .pred %p<6>; + .reg .b16 %rs<17>; + .reg .b32 %r<107>; + .reg .b64 %rd<25>; + .loc 1 69 0 // l2norm.py:69:0 +$L__func_begin0: + .loc 1 69 0 // l2norm.py:69:0 + +// %bb.0: // %__nv_sqrtf.exit + ld.param.b64 %rd6, [l2norm_fwd_kernel_param_0]; + ld.param.b64 %rd7, [l2norm_fwd_kernel_param_1]; +$L__tmp0: + .loc 1 70 24 // l2norm.py:70:24 + mov.u32 %r22, %ctaid.x; + ld.param.b64 %rd8, [l2norm_fwd_kernel_param_2]; + .loc 1 71 16 // l2norm.py:71:16 + mul.wide.u32 %rd9, %r22, 32; + ld.param.b32 %r23, [l2norm_fwd_kernel_param_3]; + ld.param.s32 %rd10, [l2norm_fwd_kernel_param_4]; + .loc 1 71 34 // l2norm.py:71:34 + mov.u32 %r24, %tid.x; + and.b32 %r25, %r24, 240; + bfe.u32 %r26, %r24, 4, 4; + or.b32 %r27, %r26, 16; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd11, %r26; + cvt.u64.u32 %rd12, %r27; + or.b64 %rd13, %rd9, %rd11; + or.b64 %rd14, %rd9, %rd12; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p1, %rd13, %rd10; + setp.lt.s64 %p2, %rd14, %rd10; + .loc 1 74 30 // l2norm.py:74:30 + shl.b32 %r28, %r24, 3; + and.b32 %r29, %r28, 120; + .loc 1 75 14 // l2norm.py:75:14 + shl.b64 %rd15, %rd13, 8; + add.s64 %rd16, %rd6, %rd15; + shl.b64 %rd17, %rd14, 8; + add.s64 %rd18, %rd6, %rd17; + .loc 1 75 33 // l2norm.py:75:33 + mul.wide.u32 %rd19, %r29, 2; + add.s64 %rd1, %rd16, %rd19; + add.s64 %rd2, %rd18, %rd19; + mov.b32 %r5, 0; + .loc 1 79 18 // l2norm.py:79:18 + // begin inline asm + mov.u32 %r1, %r5; + mov.u32 %r2, %r5; + mov.u32 %r3, %r5; + mov.u32 %r4, %r5; + @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ]; + // end inline asm + mov.b32 {%rs1, %rs2}, %r1; + mov.b32 {%rs3, %rs4}, %r2; + mov.b32 {%rs5, %rs6}, %r3; + mov.b32 {%rs7, %rs8}, %r4; + // begin inline asm + mov.u32 %r6, %r5; + mov.u32 %r7, %r5; + mov.u32 %r8, %r5; + mov.u32 %r9, %r5; + @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ]; + // end inline asm + mov.b32 {%rs9, %rs10}, %r6; + mov.b32 {%rs11, %rs12}, %r7; + mov.b32 {%rs13, %rs14}, %r8; + mov.b32 {%rs15, %rs16}, %r9; + .loc 1 79 47 // l2norm.py:79:47 + cvt.f32.bf16 %r30, %rs1; + cvt.f32.bf16 %r31, %rs2; + cvt.f32.bf16 %r32, %rs3; + cvt.f32.bf16 %r33, %rs4; + cvt.f32.bf16 %r34, %rs5; + cvt.f32.bf16 %r35, %rs6; + cvt.f32.bf16 %r36, %rs7; + cvt.f32.bf16 %r37, %rs8; + cvt.f32.bf16 %r38, %rs9; + cvt.f32.bf16 %r39, %rs10; + cvt.f32.bf16 %r40, %rs11; + cvt.f32.bf16 %r41, %rs12; + cvt.f32.bf16 %r42, %rs13; + cvt.f32.bf16 %r43, %rs14; + cvt.f32.bf16 %r44, %rs15; + cvt.f32.bf16 %r45, %rs16; + .loc 1 80 38 // l2norm.py:80:38 + mul.f32 %r46, %r31, %r31; + mul.f32 %r47, %r39, %r39; +$L__tmp1: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + fma.rn.f32.bf16 %r48, %rs1, %rs1, %r46; + fma.rn.f32.bf16 %r49, %rs3, %rs3, %r48; + fma.rn.f32.bf16 %r50, %rs4, %rs4, %r49; + fma.rn.f32.bf16 %r51, %rs5, %rs5, %r50; + fma.rn.f32.bf16 %r52, %rs6, %rs6, %r51; + fma.rn.f32.bf16 %r53, %rs7, %rs7, %r52; + fma.rn.f32.bf16 %r54, %rs8, %rs8, %r53; + fma.rn.f32.bf16 %r55, %rs9, %rs9, %r47; + fma.rn.f32.bf16 %r56, %rs11, %rs11, %r55; + fma.rn.f32.bf16 %r57, %rs12, %rs12, %r56; + fma.rn.f32.bf16 %r58, %rs13, %rs13, %r57; + fma.rn.f32.bf16 %r59, %rs14, %rs14, %r58; + fma.rn.f32.bf16 %r60, %rs15, %rs15, %r59; + fma.rn.f32.bf16 %r61, %rs16, %rs16, %r60; +$L__tmp2: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r62, %r54, 8, 31, -1; +$L__tmp3: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r63, %r54, %r62; +$L__tmp4: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r64, %r63, 4, 31, -1; +$L__tmp5: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r65, %r63, %r64; +$L__tmp6: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r66, %r65, 2, 31, -1; +$L__tmp7: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r67, %r65, %r66; +$L__tmp8: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r68, %r67, 1, 31, -1; +$L__tmp9: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r69, %r67, %r68; +$L__tmp10: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r70, %r61, 8, 31, -1; +$L__tmp11: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r71, %r61, %r70; +$L__tmp12: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r72, %r71, 4, 31, -1; +$L__tmp13: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r73, %r71, %r72; +$L__tmp14: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r74, %r73, 2, 31, -1; +$L__tmp15: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r75, %r73, %r74; +$L__tmp16: + .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ] + shfl.sync.bfly.b32 %r76, %r75, 1, 31, -1; +$L__tmp17: + .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ] + add.f32 %r77, %r75, %r76; +$L__tmp18: + .loc 1 80 48 // l2norm.py:80:48 + add.f32 %r78, %r23, %r69; + add.f32 %r79, %r23, %r77; + .loc 1 80 25 // l2norm.py:80:25 + sqrt.approx.ftz.f32 %r80, %r78; + sqrt.approx.ftz.f32 %r81, %r79; + .loc 1 71 34 // l2norm.py:71:34 + and.b32 %r82, %r24, 31; + .loc 1 71 21 // l2norm.py:71:21 + cvt.u64.u32 %rd20, %r82; + or.b64 %rd21, %rd9, %rd20; + .loc 1 77 20 // l2norm.py:77:20 + shl.b64 %rd22, %rd21, 2; + add.s64 %rd5, %rd8, %rd22; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd23, %rd7, %rd17; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd4, %rd23, %rd19; + .loc 1 76 14 // l2norm.py:76:14 + add.s64 %rd24, %rd7, %rd15; + .loc 1 76 33 // l2norm.py:76:33 + add.s64 %rd3, %rd24, %rd19; + .loc 1 73 16 // l2norm.py:73:16 + setp.lt.s64 %p4, %rd21, %rd10; + mov.b32 %r83, 0f3F800000; + .loc 1 80 17 // l2norm.py:80:17 + div.full.f32 %r19, %r83, %r80; + div.full.f32 %r20, %r83, %r81; + .loc 1 81 16 // l2norm.py:81:16 + mul.f32 %r84, %r19, %r30; + mul.f32 %r85, %r19, %r31; + mul.f32 %r86, %r19, %r32; + mul.f32 %r87, %r19, %r33; + mul.f32 %r88, %r19, %r34; + mul.f32 %r89, %r19, %r35; + mul.f32 %r90, %r19, %r36; + mul.f32 %r91, %r19, %r37; + mul.f32 %r92, %r20, %r38; + mul.f32 %r93, %r20, %r39; + mul.f32 %r94, %r20, %r40; + mul.f32 %r95, %r20, %r41; + mul.f32 %r96, %r20, %r42; + mul.f32 %r97, %r20, %r43; + mul.f32 %r98, %r20, %r44; + mul.f32 %r99, %r20, %r45; + .loc 1 83 18 // l2norm.py:83:18 + cvt.rn.bf16x2.f32 %r10, %r85, %r84; + cvt.rn.bf16x2.f32 %r11, %r87, %r86; + cvt.rn.bf16x2.f32 %r12, %r89, %r88; + cvt.rn.bf16x2.f32 %r13, %r91, %r90; + // begin inline asm + @%p1 st.global.v4.b32 [ %rd3 + 0 ], { %r10, %r11, %r12, %r13 }; + // end inline asm + cvt.rn.bf16x2.f32 %r14, %r93, %r92; + cvt.rn.bf16x2.f32 %r15, %r95, %r94; + cvt.rn.bf16x2.f32 %r16, %r97, %r96; + cvt.rn.bf16x2.f32 %r17, %r99, %r98; + // begin inline asm + @%p2 st.global.v4.b32 [ %rd4 + 0 ], { %r14, %r15, %r16, %r17 }; + // end inline asm + .loc 1 84 21 // l2norm.py:84:21 + shr.u32 %r100, %r25, 1; + mov.b32 %r101, global_smem; + add.s32 %r18, %r101, %r100; + // begin inline asm + st.shared.v2.b32 [ %r18 + 0 ], { %r19, %r20 }; + // end inline asm + bar.sync 0; + shr.u32 %r102, %r24, 2; + and.b32 %r103, %r102, 4; + add.s32 %r104, %r101, %r29; + add.s32 %r105, %r104, %r103; + ld.shared.b32 %r21, [%r105]; + and.b32 %r106, %r24, 224; + setp.eq.b32 %p5, %r106, 0; + and.pred %p3, %p5, %p4; + // begin inline asm + @%p3 st.global.b32 [ %rd5 + 0 ], { %r21 }; + // end inline asm + .loc 1 84 4 // l2norm.py:84:4 + ret; +$L__tmp19: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 1 // DW_CHILDREN_yes +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 5 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 5 // DW_FORM_data2 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 196 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 109 +.b8 111 +.b8 100 +.b8 117 +.b8 108 +.b8 101 +.b8 115 +.b8 47 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 0 +.b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram +.b8 108 // DW_AT_name +.b8 50 +.b8 110 +.b8 111 +.b8 114 +.b8 109 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 107 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp18 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 80 // DW_AT_call_line +.b8 43 // DW_AT_call_column +.b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine +.b32 107 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp18 // DW_AT_high_pc +.b8 2 // DW_AT_call_file +.b8 37 // DW_AT_call_line +.b8 1 +.b8 36 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/l2norm_fwd_kernel.source b/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/l2norm_fwd_kernel.source new file mode 100644 index 0000000000000000000000000000000000000000..d9fa83aac0600b1e5d35d95ed8aaad5c5636e577 --- /dev/null +++ b/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/l2norm_fwd_kernel.source @@ -0,0 +1,172 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0) +#loc38 = loc(unknown) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0) +#loc45 = loc("x"(#loc)) +#loc46 = loc("y"(#loc)) +#loc47 = loc("rstd"(#loc)) +#loc48 = loc("eps"(#loc)) +#loc49 = loc("T"(#loc)) +#loc81 = loc("input"(#loc36)) +#loc82 = loc("a"(#loc41)) +#loc83 = loc("b"(#loc41)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %i_t = tt.get_program_id x : i32 loc(#loc50) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc51) + %o_t = arith.constant 32 : i32 loc(#loc52) + %o_t_1 = arith.constant 32 : i64 loc(#loc52) + %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc52) + %o_t_3 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32> loc(#loc53) + %o_t_4 = arith.extsi %o_t_3 : tensor<32xi32> to tensor<32xi64> loc(#loc54) + %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<32xi64> loc(#loc54) + %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<32xi64> loc(#loc54) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc55) + %m_t = arith.extsi %T : i32 to i64 loc(#loc56) + %m_t_7 = tt.splat %m_t : i64 -> tensor<32xi64> loc(#loc56) + %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<32xi64> loc(#loc56) + %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<32xi1> -> tensor<32x1xi1> loc(#loc57) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc58) + %m_x_10 = arith.constant 128 : i32 loc(#loc59) + %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc59) + %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc59) + %m_x_13 = tt.broadcast %m_x : tensor<32x1xi1> -> tensor<32x128xi1> loc(#loc60) + %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<32x128xi1> loc(#loc60) + %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<32x128xi1> loc(#loc60) + %p_x = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<32xi64> -> tensor<32x1xi64> loc(#loc61) + %p_x_16 = arith.constant 128 : i32 loc(#loc62) + %p_x_17 = arith.constant 128 : i64 loc(#loc62) + %p_x_18 = arith.constant dense<128> : tensor<32x1xi64> loc(#loc62) + %p_x_19 = arith.muli %p_x, %p_x_18 : tensor<32x1xi64> loc(#loc62) + %p_x_20 = tt.splat %x : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc63) + %p_x_21 = tt.addptr %p_x_20, %p_x_19 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc63) + %p_x_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc64) + %p_x_23 = tt.broadcast %p_x_21 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc65) + %p_x_24 = tt.broadcast %p_x_22 : tensor<1x128xi32> -> tensor<32x128xi32> loc(#loc65) + %p_x_25 = tt.addptr %p_x_23, %p_x_24 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc65) + %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<32xi64> -> tensor<32x1xi64> loc(#loc66) + %p_y_26 = arith.constant 128 : i32 loc(#loc67) + %p_y_27 = arith.constant 128 : i64 loc(#loc67) + %p_y_28 = arith.constant dense<128> : tensor<32x1xi64> loc(#loc67) + %p_y_29 = arith.muli %p_y, %p_y_28 : tensor<32x1xi64> loc(#loc67) + %p_y_30 = tt.splat %y : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc68) + %p_y_31 = tt.addptr %p_y_30, %p_y_29 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc68) + %p_y_32 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc69) + %p_y_33 = tt.broadcast %p_y_31 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc70) + %p_y_34 = tt.broadcast %p_y_32 : tensor<1x128xi32> -> tensor<32x128xi32> loc(#loc70) + %p_y_35 = tt.addptr %p_y_33, %p_y_34 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc70) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<32x!tt.ptr> loc(#loc71) + %p_rstd_36 = tt.addptr %p_rstd, %o_t_6 : tensor<32x!tt.ptr>, tensor<32xi64> loc(#loc71) + %b_x = arith.constant 0.000000e+00 : f32 loc(#loc72) + %b_x_37 = arith.constant dense<0.000000e+00> : tensor<32x128xf32> loc(#loc72) + %b_x_38 = arith.truncf %b_x_37 : tensor<32x128xf32> to tensor<32x128xbf16> loc(#loc72) + %b_x_39 = tt.load %p_x_25, %m_x_15, %b_x_38 : tensor<32x128x!tt.ptr> loc(#loc72) + %b_x_40 = arith.extf %b_x_39 : tensor<32x128xbf16> to tensor<32x128xf32> loc(#loc73) + %b_rstd = arith.mulf %b_x_40, %b_x_40 : tensor<32x128xf32> loc(#loc74) + %b_rstd_41 = tt.call @triton.language.standard.sum__fp32S32_128S_c1_cFalse_cNone(%b_rstd) : (tensor<32x128xf32>) -> tensor<32xf32> loc(#loc75) + %b_rstd_42 = tt.splat %eps : f32 -> tensor<32xf32> loc(#loc76) + %b_rstd_43 = arith.addf %b_rstd_41, %b_rstd_42 : tensor<32xf32> loc(#loc76) + %b_rstd_44 = math.sqrt %b_rstd_43 : tensor<32xf32> loc(#loc77) + %b_rstd_45 = arith.constant 1 : i32 loc(#loc78) + %b_rstd_46 = arith.constant 1.000000e+00 : f32 loc(#loc78) + %b_rstd_47 = arith.constant dense<1.000000e+00> : tensor<32xf32> loc(#loc78) + %b_rstd_48 = arith.divf %b_rstd_47, %b_rstd_44 : tensor<32xf32> loc(#loc78) + %b_y = tt.expand_dims %b_rstd_48 {axis = 1 : i32} : tensor<32xf32> -> tensor<32x1xf32> loc(#loc79) + %b_y_49 = tt.broadcast %b_y : tensor<32x1xf32> -> tensor<32x128xf32> loc(#loc80) + %b_y_50 = arith.mulf %b_x_40, %b_y_49 : tensor<32x128xf32> loc(#loc80) + %0 = arith.truncf %b_y_50 : tensor<32x128xf32> to tensor<32x128xbf16> loc(#loc32) + tt.store %p_y_35, %0, %m_x_15 : tensor<32x128x!tt.ptr> loc(#loc33) + tt.store %p_rstd_36, %b_rstd_48, %m_t_8 : tensor<32x!tt.ptr> loc(#loc34) + tt.return loc(#loc35) + } loc(#loc) + tt.func private @triton.language.standard.sum__fp32S32_128S_c1_cFalse_cNone(%input: tensor<32x128xf32> loc("input"(#loc36))) -> tensor<32xf32> attributes {noinline = false} { + %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({ + ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)): + %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc37) + tt.reduce.return %2 : f32 loc(#loc37) + }) : (tensor<32x128xf32>) -> tensor<32xf32> loc(#loc37) + tt.return %0 : tensor<32xf32> loc(#loc39) + ^bb1: // no predecessors + %1 = ub.poison : tensor<32xf32> loc(#loc40) + tt.return %1 : tensor<32xf32> loc(#loc40) + } loc(#loc36) + tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc41)), %b: f32 loc("b"(#loc41))) -> f32 attributes {noinline = false} { + %0 = arith.addf %a, %b : f32 loc(#loc42) + tt.return %0 : f32 loc(#loc43) + ^bb1: // no predecessors + %1 = ub.poison : f32 loc(#loc44) + tt.return %1 : f32 loc(#loc44) + } loc(#loc41) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:37) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:18) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:29) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:37) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4) +#loc50 = loc("i_t"(#loc1)) +#loc51 = loc("i_t"(#loc2)) +#loc52 = loc("o_t"(#loc3)) +#loc53 = loc("o_t"(#loc4)) +#loc54 = loc("o_t"(#loc5)) +#loc55 = loc("o_d"(#loc6)) +#loc56 = loc("m_t"(#loc7)) +#loc57 = loc("m_x"(#loc8)) +#loc58 = loc("m_x"(#loc9)) +#loc59 = loc("m_x"(#loc10)) +#loc60 = loc("m_x"(#loc11)) +#loc61 = loc("p_x"(#loc12)) +#loc62 = loc("p_x"(#loc13)) +#loc63 = loc("p_x"(#loc14)) +#loc64 = loc("p_x"(#loc15)) +#loc65 = loc("p_x"(#loc16)) +#loc66 = loc("p_y"(#loc17)) +#loc67 = loc("p_y"(#loc18)) +#loc68 = loc("p_y"(#loc19)) +#loc69 = loc("p_y"(#loc20)) +#loc70 = loc("p_y"(#loc21)) +#loc71 = loc("p_rstd"(#loc22)) +#loc72 = loc("b_x"(#loc23)) +#loc73 = loc("b_x"(#loc24)) +#loc74 = loc("b_rstd"(#loc25)) +#loc75 = loc("b_rstd"(#loc26)) +#loc76 = loc("b_rstd"(#loc27)) +#loc77 = loc("b_rstd"(#loc28)) +#loc78 = loc("b_rstd"(#loc29)) +#loc79 = loc("b_y"(#loc30)) +#loc80 = loc("b_y"(#loc31)) diff --git a/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/l2norm_fwd_kernel.ttgir b/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/l2norm_fwd_kernel.ttgir new file mode 100644 index 0000000000000000000000000000000000000000..01f1d85a0f5c7e29bfcd3f5c6529e0fc1f2c3068 --- /dev/null +++ b/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/l2norm_fwd_kernel.ttgir @@ -0,0 +1,135 @@ +#blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [8, 1], order = [1, 0]}> +#blocked1 = #ttg.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [8], order = [0]}> +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc1 = loc(unknown) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc34 = loc("x"(#loc)) +#loc35 = loc("y"(#loc)) +#loc36 = loc("rstd"(#loc)) +#loc37 = loc("eps"(#loc)) +#loc38 = loc("T"(#loc)) +#loc59 = loc("b_rstd"(#loc23)) +#loc66 = loc(callsite(#loc1 at #loc59)) +module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 8 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %cst = arith.constant dense<1.000000e+00> : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<32x128xbf16, #blocked> loc(#loc1) + %c32_i64 = arith.constant 32 : i64 loc(#loc1) + %cst_1 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1) + %cst_2 = arith.constant dense<128> : tensor<32x1xi64, #blocked> loc(#loc1) + %i_t = tt.get_program_id x : i32 loc(#loc39) + %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc40) + %o_t = arith.muli %i_t_3, %c32_i64 : i64 loc(#loc41) + %o_t_4 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc42) + %o_t_5 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32, #blocked1> loc(#loc42) + %o_t_6 = arith.extsi %o_t_4 : tensor<32xi32, #ttg.slice<{dim = 1, parent = #blocked}>> to tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_7 = arith.extsi %o_t_5 : tensor<32xi32, #blocked1> to tensor<32xi64, #blocked1> loc(#loc43) + %o_t_8 = tt.splat %o_t : i64 -> tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_9 = tt.splat %o_t : i64 -> tensor<32xi64, #blocked1> loc(#loc43) + %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43) + %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<32xi64, #blocked1> loc(#loc43) + %m_t = arith.extsi %T : i32 to i64 loc(#loc44) + %m_t_12 = tt.splat %m_t : i64 -> tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_13 = tt.splat %m_t : i64 -> tensor<32xi64, #blocked1> loc(#loc44) + %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44) + %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<32xi64, #blocked1> loc(#loc44) + %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<32xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32x1xi1, #blocked> loc(#loc45) + %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc46) + %m_x_17 = tt.expand_dims %m_x_16 {axis = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> -> tensor<1x128xi32, #blocked> loc(#loc46) + %m_x_18 = arith.cmpi slt, %m_x_17, %cst_1 : tensor<1x128xi32, #blocked> loc(#loc47) + %m_x_19 = tt.broadcast %m_x : tensor<32x1xi1, #blocked> -> tensor<32x128xi1, #blocked> loc(#loc48) + %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<32x128xi1, #blocked> loc(#loc48) + %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<32x128xi1, #blocked> loc(#loc48) + %p_x = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32x1xi64, #blocked> loc(#loc49) + %p_x_22 = arith.muli %p_x, %cst_2 : tensor<32x1xi64, #blocked> loc(#loc50) + %p_x_23 = tt.splat %x : !tt.ptr -> tensor<32x1x!tt.ptr, #blocked> loc(#loc51) + %p_x_24 = tt.addptr %p_x_23, %p_x_22 : tensor<32x1x!tt.ptr, #blocked>, tensor<32x1xi64, #blocked> loc(#loc51) + %p_x_25 = tt.broadcast %p_x_24 : tensor<32x1x!tt.ptr, #blocked> -> tensor<32x128x!tt.ptr, #blocked> loc(#loc52) + %p_x_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<32x128xi32, #blocked> loc(#loc52) + %p_x_27 = tt.addptr %p_x_25, %p_x_26 : tensor<32x128x!tt.ptr, #blocked>, tensor<32x128xi32, #blocked> loc(#loc52) + %p_y = tt.splat %y : !tt.ptr -> tensor<32x1x!tt.ptr, #blocked> loc(#loc53) + %p_y_28 = tt.addptr %p_y, %p_x_22 : tensor<32x1x!tt.ptr, #blocked>, tensor<32x1xi64, #blocked> loc(#loc53) + %p_y_29 = tt.broadcast %p_y_28 : tensor<32x1x!tt.ptr, #blocked> -> tensor<32x128x!tt.ptr, #blocked> loc(#loc54) + %p_y_30 = tt.addptr %p_y_29, %p_x_26 : tensor<32x128x!tt.ptr, #blocked>, tensor<32x128xi32, #blocked> loc(#loc54) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<32x!tt.ptr, #blocked1> loc(#loc55) + %p_rstd_31 = tt.addptr %p_rstd, %o_t_11 : tensor<32x!tt.ptr, #blocked1>, tensor<32xi64, #blocked1> loc(#loc55) + %b_x = tt.load %p_x_27, %m_x_21, %cst_0 : tensor<32x128x!tt.ptr, #blocked> loc(#loc56) + %b_x_32 = arith.extf %b_x : tensor<32x128xbf16, #blocked> to tensor<32x128xf32, #blocked> loc(#loc57) + %b_rstd = arith.mulf %b_x_32, %b_x_32 : tensor<32x128xf32, #blocked> loc(#loc58) + %b_rstd_33 = "tt.reduce"(%b_rstd) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_40: f32 loc(callsite(#loc1 at #loc59)), %b_rstd_41: f32 loc(callsite(#loc1 at #loc59))): + %b_rstd_42 = arith.addf %b_rstd_40, %b_rstd_41 : f32 loc(#loc67) + tt.reduce.return %b_rstd_42 : f32 loc(#loc65) + }) : (tensor<32x128xf32, #blocked>) -> tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc65) + %b_rstd_34 = tt.splat %eps : f32 -> tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60) + %b_rstd_36 = math.sqrt %b_rstd_35 : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc61) + %b_rstd_37 = arith.divf %cst, %b_rstd_36 : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc62) + %b_y = tt.expand_dims %b_rstd_37 {axis = 1 : i32} : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32x1xf32, #blocked> loc(#loc63) + %b_y_38 = tt.broadcast %b_y : tensor<32x1xf32, #blocked> -> tensor<32x128xf32, #blocked> loc(#loc64) + %b_y_39 = arith.mulf %b_x_32, %b_y_38 : tensor<32x128xf32, #blocked> loc(#loc64) + %0 = arith.truncf %b_y_39 : tensor<32x128xf32, #blocked> to tensor<32x128xbf16, #blocked> loc(#loc30) + tt.store %p_y_30, %0, %m_x_21 : tensor<32x128x!tt.ptr, #blocked> loc(#loc31) + %1 = ttg.convert_layout %b_rstd_37 : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32xf32, #blocked1> loc(#loc32) + tt.store %p_rstd_31, %1, %m_t_15 : tensor<32x!tt.ptr, #blocked1> loc(#loc32) + tt.return loc(#loc33) + } loc(#loc) +} loc(#loc) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc39 = loc("i_t"(#loc2)) +#loc40 = loc("i_t"(#loc3)) +#loc41 = loc("o_t"(#loc4)) +#loc42 = loc("o_t"(#loc5)) +#loc43 = loc("o_t"(#loc6)) +#loc44 = loc("m_t"(#loc7)) +#loc45 = loc("m_x"(#loc8)) +#loc46 = loc("m_x"(#loc9)) +#loc47 = loc("m_x"(#loc10)) +#loc48 = loc("m_x"(#loc11)) +#loc49 = loc("p_x"(#loc12)) +#loc50 = loc("p_x"(#loc13)) +#loc51 = loc("p_x"(#loc14)) +#loc52 = loc("p_x"(#loc15)) +#loc53 = loc("p_y"(#loc16)) +#loc54 = loc("p_y"(#loc17)) +#loc55 = loc("p_rstd"(#loc18)) +#loc56 = loc("b_x"(#loc19)) +#loc57 = loc("b_x"(#loc20)) +#loc58 = loc("b_rstd"(#loc21)) +#loc60 = loc("b_rstd"(#loc25)) +#loc61 = loc("b_rstd"(#loc26)) +#loc62 = loc("b_rstd"(#loc27)) +#loc63 = loc("b_y"(#loc28)) +#loc64 = loc("b_y"(#loc29)) +#loc65 = loc(callsite(#loc22 at #loc59)) +#loc67 = loc(callsite(#loc24 at #loc65)) diff --git a/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/l2norm_fwd_kernel.ttir b/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/l2norm_fwd_kernel.ttir new file mode 100644 index 0000000000000000000000000000000000000000..3688281bceaf7f8abdf8568104090be5e0170141 --- /dev/null +++ b/triton/gpu0/CNVJFG3LSAECSGAC2RMU2O6WHSTOX4WTTMQHQABPH23L7WMAH7LQ/l2norm_fwd_kernel.ttir @@ -0,0 +1,128 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0) +#loc3 = loc(unknown) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43) +#loc35 = loc("x"(#loc)) +#loc36 = loc("y"(#loc)) +#loc37 = loc("rstd"(#loc)) +#loc38 = loc("eps"(#loc)) +#loc39 = loc("T"(#loc)) +#loc62 = loc("b_rstd"(#loc25)) +#loc68 = loc(callsite(#loc3 at #loc62)) +module { + tt.func public @l2norm_fwd_kernel(%x: !tt.ptr {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + %b_x = arith.constant dense<0.000000e+00> : tensor<32x128xbf16> loc(#loc40) + %b_rstd = arith.constant dense<1.000000e+00> : tensor<32xf32> loc(#loc41) + %cst = arith.constant dense<128> : tensor<32x1xi64> loc(#loc3) + %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc42) + %o_t = arith.constant 32 : i64 loc(#loc43) + %i_t = tt.get_program_id x : i32 loc(#loc44) + %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc45) + %o_t_1 = arith.muli %i_t_0, %o_t : i64 loc(#loc43) + %o_t_2 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32> loc(#loc46) + %o_t_3 = arith.extsi %o_t_2 : tensor<32xi32> to tensor<32xi64> loc(#loc47) + %o_t_4 = tt.splat %o_t_1 : i64 -> tensor<32xi64> loc(#loc47) + %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<32xi64> loc(#loc47) + %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc48) + %m_t = arith.extsi %T : i32 to i64 loc(#loc49) + %m_t_6 = tt.splat %m_t : i64 -> tensor<32xi64> loc(#loc49) + %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<32xi64> loc(#loc49) + %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<32xi1> -> tensor<32x1xi1> loc(#loc50) + %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc51) + %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc42) + %m_x_11 = tt.broadcast %m_x_8 : tensor<32x1xi1> -> tensor<32x128xi1> loc(#loc52) + %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<32x128xi1> loc(#loc52) + %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<32x128xi1> loc(#loc52) + %p_x = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<32xi64> -> tensor<32x1xi64> loc(#loc53) + %p_x_14 = arith.muli %p_x, %cst : tensor<32x1xi64> loc(#loc54) + %p_x_15 = tt.splat %x : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc55) + %p_x_16 = tt.addptr %p_x_15, %p_x_14 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc55) + %p_x_17 = tt.broadcast %p_x_16 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc56) + %p_x_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<32x128xi32> loc(#loc56) + %p_x_19 = tt.addptr %p_x_17, %p_x_18 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc56) + %p_y = tt.splat %y : !tt.ptr -> tensor<32x1x!tt.ptr> loc(#loc57) + %p_y_20 = tt.addptr %p_y, %p_x_14 : tensor<32x1x!tt.ptr>, tensor<32x1xi64> loc(#loc57) + %p_y_21 = tt.broadcast %p_y_20 : tensor<32x1x!tt.ptr> -> tensor<32x128x!tt.ptr> loc(#loc58) + %p_y_22 = tt.addptr %p_y_21, %p_x_18 : tensor<32x128x!tt.ptr>, tensor<32x128xi32> loc(#loc58) + %p_rstd = tt.splat %rstd : !tt.ptr -> tensor<32x!tt.ptr> loc(#loc59) + %p_rstd_23 = tt.addptr %p_rstd, %o_t_5 : tensor<32x!tt.ptr>, tensor<32xi64> loc(#loc59) + %b_x_24 = tt.load %p_x_19, %m_x_13, %b_x : tensor<32x128x!tt.ptr> loc(#loc40) + %b_x_25 = arith.extf %b_x_24 : tensor<32x128xbf16> to tensor<32x128xf32> loc(#loc60) + %b_rstd_26 = arith.mulf %b_x_25, %b_x_25 : tensor<32x128xf32> loc(#loc61) + %b_rstd_27 = "tt.reduce"(%b_rstd_26) <{axis = 1 : i32}> ({ + ^bb0(%b_rstd_34: f32 loc(callsite(#loc3 at #loc62)), %b_rstd_35: f32 loc(callsite(#loc3 at #loc62))): + %b_rstd_36 = arith.addf %b_rstd_34, %b_rstd_35 : f32 loc(#loc69) + tt.reduce.return %b_rstd_36 : f32 loc(#loc67) + }) : (tensor<32x128xf32>) -> tensor<32xf32> loc(#loc67) + %b_rstd_28 = tt.splat %eps : f32 -> tensor<32xf32> loc(#loc63) + %b_rstd_29 = arith.addf %b_rstd_27, %b_rstd_28 : tensor<32xf32> loc(#loc63) + %b_rstd_30 = math.sqrt %b_rstd_29 : tensor<32xf32> loc(#loc64) + %b_rstd_31 = arith.divf %b_rstd, %b_rstd_30 : tensor<32xf32> loc(#loc41) + %b_y = tt.expand_dims %b_rstd_31 {axis = 1 : i32} : tensor<32xf32> -> tensor<32x1xf32> loc(#loc65) + %b_y_32 = tt.broadcast %b_y : tensor<32x1xf32> -> tensor<32x128xf32> loc(#loc66) + %b_y_33 = arith.mulf %b_x_25, %b_y_32 : tensor<32x128xf32> loc(#loc66) + %0 = arith.truncf %b_y_33 : tensor<32x128xf32> to tensor<32x128xbf16> loc(#loc31) + tt.store %p_y_22, %0, %m_x_13 : tensor<32x128x!tt.ptr> loc(#loc32) + tt.store %p_rstd_23, %b_rstd_31, %m_t_7 : tensor<32x!tt.ptr> loc(#loc33) + tt.return loc(#loc34) + } loc(#loc) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4) +#loc40 = loc("b_x"(#loc1)) +#loc41 = loc("b_rstd"(#loc2)) +#loc42 = loc("m_x"(#loc4)) +#loc43 = loc("o_t"(#loc5)) +#loc44 = loc("i_t"(#loc6)) +#loc45 = loc("i_t"(#loc7)) +#loc46 = loc("o_t"(#loc8)) +#loc47 = loc("o_t"(#loc9)) +#loc48 = loc("o_d"(#loc10)) +#loc49 = loc("m_t"(#loc11)) +#loc50 = loc("m_x"(#loc12)) +#loc51 = loc("m_x"(#loc13)) +#loc52 = loc("m_x"(#loc14)) +#loc53 = loc("p_x"(#loc15)) +#loc54 = loc("p_x"(#loc16)) +#loc55 = loc("p_x"(#loc17)) +#loc56 = loc("p_x"(#loc18)) +#loc57 = loc("p_y"(#loc19)) +#loc58 = loc("p_y"(#loc20)) +#loc59 = loc("p_rstd"(#loc21)) +#loc60 = loc("b_x"(#loc22)) +#loc61 = loc("b_rstd"(#loc23)) +#loc63 = loc("b_rstd"(#loc27)) +#loc64 = loc("b_rstd"(#loc28)) +#loc65 = loc("b_y"(#loc29)) +#loc66 = loc("b_y"(#loc30)) +#loc67 = loc(callsite(#loc24 at #loc62)) +#loc69 = loc(callsite(#loc26 at #loc67)) diff --git a/triton/gpu0/COEH6RZWJ4F4NLI6BWD6G2VBWJCNYUFXAFKSMEBEK2VRZ3DVOIQA/__grp__chunk_fwd_kernel_o.json b/triton/gpu0/COEH6RZWJ4F4NLI6BWD6G2VBWJCNYUFXAFKSMEBEK2VRZ3DVOIQA/__grp__chunk_fwd_kernel_o.json new file mode 100644 index 0000000000000000000000000000000000000000..772a8e60eadc239967c0221547775d055b374ef8 --- /dev/null +++ b/triton/gpu0/COEH6RZWJ4F4NLI6BWD6G2VBWJCNYUFXAFKSMEBEK2VRZ3DVOIQA/__grp__chunk_fwd_kernel_o.json @@ -0,0 +1 @@ +{"child_paths": {"chunk_fwd_kernel_o.source": "/home/silver/affine/v2/models/uid161/triton/gpu0/COEH6RZWJ4F4NLI6BWD6G2VBWJCNYUFXAFKSMEBEK2VRZ3DVOIQA/chunk_fwd_kernel_o.source", "chunk_fwd_kernel_o.ttir": "/home/silver/affine/v2/models/uid161/triton/gpu0/COEH6RZWJ4F4NLI6BWD6G2VBWJCNYUFXAFKSMEBEK2VRZ3DVOIQA/chunk_fwd_kernel_o.ttir", "chunk_fwd_kernel_o.ttgir": "/home/silver/affine/v2/models/uid161/triton/gpu0/COEH6RZWJ4F4NLI6BWD6G2VBWJCNYUFXAFKSMEBEK2VRZ3DVOIQA/chunk_fwd_kernel_o.ttgir", "chunk_fwd_kernel_o.llir": "/home/silver/affine/v2/models/uid161/triton/gpu0/COEH6RZWJ4F4NLI6BWD6G2VBWJCNYUFXAFKSMEBEK2VRZ3DVOIQA/chunk_fwd_kernel_o.llir", "chunk_fwd_kernel_o.ptx": "/home/silver/affine/v2/models/uid161/triton/gpu0/COEH6RZWJ4F4NLI6BWD6G2VBWJCNYUFXAFKSMEBEK2VRZ3DVOIQA/chunk_fwd_kernel_o.ptx", "chunk_fwd_kernel_o.cubin": "/home/silver/affine/v2/models/uid161/triton/gpu0/COEH6RZWJ4F4NLI6BWD6G2VBWJCNYUFXAFKSMEBEK2VRZ3DVOIQA/chunk_fwd_kernel_o.cubin", "chunk_fwd_kernel_o.json": "/home/silver/affine/v2/models/uid161/triton/gpu0/COEH6RZWJ4F4NLI6BWD6G2VBWJCNYUFXAFKSMEBEK2VRZ3DVOIQA/chunk_fwd_kernel_o.json"}} \ No newline at end of file diff --git a/triton/gpu0/COEH6RZWJ4F4NLI6BWD6G2VBWJCNYUFXAFKSMEBEK2VRZ3DVOIQA/chunk_fwd_kernel_o.cubin b/triton/gpu0/COEH6RZWJ4F4NLI6BWD6G2VBWJCNYUFXAFKSMEBEK2VRZ3DVOIQA/chunk_fwd_kernel_o.cubin new file mode 100644 index 0000000000000000000000000000000000000000..24cea03cab77aa9fc1fed368335778d71684e7be --- /dev/null +++ b/triton/gpu0/COEH6RZWJ4F4NLI6BWD6G2VBWJCNYUFXAFKSMEBEK2VRZ3DVOIQA/chunk_fwd_kernel_o.cubin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8187b768e1586e6e2c580f72d5ab8537a2e42d7d42b6ac6f9f5efc61e1d55ba5 +size 104976 diff --git a/triton/gpu0/COEH6RZWJ4F4NLI6BWD6G2VBWJCNYUFXAFKSMEBEK2VRZ3DVOIQA/chunk_fwd_kernel_o.json b/triton/gpu0/COEH6RZWJ4F4NLI6BWD6G2VBWJCNYUFXAFKSMEBEK2VRZ3DVOIQA/chunk_fwd_kernel_o.json new file mode 100644 index 0000000000000000000000000000000000000000..9f83119dbb40dede917f037502593bcc21315b79 --- /dev/null +++ b/triton/gpu0/COEH6RZWJ4F4NLI6BWD6G2VBWJCNYUFXAFKSMEBEK2VRZ3DVOIQA/chunk_fwd_kernel_o.json @@ -0,0 +1 @@ +{"hash": "13887f47364f0bc6ad1e0d87e36aa1b244dc50b7015526102456ab1cec757220", "target": {"backend": "cuda", "arch": 100, "warp_size": 32}, "num_warps": 4, "num_ctas": 1, "num_stages": 3, "warp_size": 32, "maxnreg": null, "ptx_version": null, "ptx_options": null, "ir_override": null, "enable_fp_fusion": true, "enable_reflect_ftz": true, "launch_cooperative_grid": false, "launch_pdl": false, "supported_fp8_dtypes": ["fp8e4b15", "fp8e4nv", "fp8e5"], "deprecated_fp8_dot_operand_dtypes": ["fp8e4b15"], "default_dot_input_precision": "tf32", "allowed_dot_input_precisions": ["tf32", "tf32x3", "ieee", "bf16x3", "bf16x6"], "max_num_imprecise_acc_default": 0, "extern_libs": [["libdevice", "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/backends/nvidia/lib/libdevice.10.bc"]], "debug": false, "backend_name": "cuda", "sanitize_overflow": true, "arch": "sm100", "instrumentation_mode": "", "triton_version": "3.7.1", "tensordesc_meta": [], "shared": 73760, "tmem_size": 128, "global_scratch_size": 0, "global_scratch_align": 1, "profile_scratch_size": 0, "profile_scratch_align": 1, "name": "chunk_fwd_kernel_o"} \ No newline at end of file diff --git a/triton/gpu0/COEH6RZWJ4F4NLI6BWD6G2VBWJCNYUFXAFKSMEBEK2VRZ3DVOIQA/chunk_fwd_kernel_o.llir b/triton/gpu0/COEH6RZWJ4F4NLI6BWD6G2VBWJCNYUFXAFKSMEBEK2VRZ3DVOIQA/chunk_fwd_kernel_o.llir new file mode 100644 index 0000000000000000000000000000000000000000..a9c6e14a792dcae410d14dade1a202e24051bf5c --- /dev/null +++ b/triton/gpu0/COEH6RZWJ4F4NLI6BWD6G2VBWJCNYUFXAFKSMEBEK2VRZ3DVOIQA/chunk_fwd_kernel_o.llir @@ -0,0 +1,2032 @@ +; ModuleID = 'LLVMDialectModule' +source_filename = "LLVMDialectModule" +target datalayout = "e-p3:32:32-p4:32:32-p5:32:32-p6:32:32-p7:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64" + +@global_smem = external addrspace(3) global [0 x i8], align 16 +@.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1 + +; Function Attrs: nounwind +define ptx_kernel void @chunk_fwd_kernel_o(ptr addrspace(1) %0, ptr addrspace(1) %1, ptr addrspace(1) %2, ptr addrspace(1) %3, ptr addrspace(1) %4, ptr addrspace(1) %5, float %6, i32 %7, ptr addrspace(1) readnone captures(none) %8, ptr addrspace(1) readnone captures(none) %9) local_unnamed_addr #0 !dbg !5 { + %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !13 + %12 = icmp samesign ult i32 %11, 32, !dbg !13 + tail call void asm sideeffect "@$0 tcgen05.alloc.cta_group::1.sync.aligned.shared::cta.b32 [$1], 128;", "b,r"(i1 %12, ptr addrspace(3) @global_smem) #6, !dbg !13 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !13 + %13 = load i32, ptr addrspace(3) @global_smem, align 16, !dbg !13 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !13 + tail call void asm sideeffect "@$0 tcgen05.relinquish_alloc_permit.cta_group::1.sync.aligned;", "b"(i1 %12) #6, !dbg !13 + %14 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !14 + %15 = zext nneg i32 %14 to i64, !dbg !19 + %16 = lshr i64 %15, 1, !dbg !20 + %17 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.y(), !dbg !21 + %18 = zext nneg i32 %17 to i64, !dbg !22 + %19 = lshr i64 %18, 5, !dbg !23 + %20 = and i64 %18, 31, !dbg !24 + %21 = add i32 %7, 63, !dbg !25 + %22 = sdiv i32 %21, 64, !dbg !29 + %23 = sext i32 %22 to i64, !dbg !30 + %24 = mul nsw i64 %19, %23, !dbg !30 + %25 = add nsw i64 %24, %16, !dbg !31 + %26 = sext i32 %7 to i64, !dbg !32 + %27 = and i64 %18, 65504, !dbg !32 + %28 = mul nsw i64 %27, %26, !dbg !33 + %29 = or disjoint i64 %28, %20, !dbg !34 + %30 = shl nsw i64 %29, 7, !dbg !35 + %31 = getelementptr bfloat, ptr addrspace(1) %0, i64 %30, !dbg !36 + %32 = getelementptr bfloat, ptr addrspace(1) %1, i64 %30, !dbg !37 + %33 = getelementptr bfloat, ptr addrspace(1) %2, i64 %30, !dbg !38 + %.idx = shl nsw i64 %25, 20, !dbg !39 + %34 = getelementptr i8, ptr addrspace(1) %3, i64 %.idx, !dbg !39 + %.idx3 = shl nuw nsw i64 %20, 15, !dbg !39 + %35 = getelementptr i8, ptr addrspace(1) %34, i64 %.idx3, !dbg !39 + %36 = shl nuw nsw i64 %16, 6, !dbg !40 + %37 = lshr i32 %11, 5, !dbg !41 + %38 = lshr i32 %11, 3, !dbg !41 + %39 = and i32 %38, 15, !dbg !41 + %40 = or disjoint i32 %39, 16, !dbg !41 + %41 = or disjoint i32 %39, 32, !dbg !41 + %42 = or disjoint i32 %39, 48, !dbg !41 + %43 = shl nuw nsw i32 %11, 3, !dbg !41 + %44 = and i32 %43, 56, !dbg !41 + %45 = and i32 %11, 15, !dbg !41 + %46 = and i32 %11, 96, !dbg !41 + %47 = lshr exact i32 %46, 1, !dbg !41 + %48 = or disjoint i32 %47, %45, !dbg !41 + %49 = and i32 %11, 16, !dbg !41 + %50 = zext nneg i32 %39 to i64, !dbg !42 + %51 = zext nneg i32 %40 to i64, !dbg !42 + %52 = zext nneg i32 %41 to i64, !dbg !42 + %53 = zext nneg i32 %42 to i64, !dbg !42 + %54 = zext nneg i32 %48 to i64, !dbg !42 + %55 = zext nneg i32 %44 to i64, !dbg !42 + %56 = or disjoint i64 %36, %50, !dbg !42 + %57 = or disjoint i64 %36, %51, !dbg !42 + %58 = or disjoint i64 %36, %52, !dbg !42 + %59 = or disjoint i64 %36, %53, !dbg !42 + %60 = or disjoint i64 %36, %54, !dbg !42 + %61 = shl nuw nsw i32 %49, 1, !dbg !41 + %62 = or disjoint i32 %61, 1, !dbg !41 + %63 = or disjoint i32 %61, 3, !dbg !41 + %64 = or disjoint i32 %61, 2, !dbg !41 + %65 = insertelement <4 x i32> poison, i32 %61, i64 0, !dbg !41 + %66 = shufflevector <4 x i32> %65, <4 x i32> poison, <4 x i32> zeroinitializer, !dbg !41 + %67 = or disjoint <4 x i32> %66, , !dbg !41 + %68 = insertelement <8 x i32> poison, i32 %61, i64 0, !dbg !41 + %69 = shufflevector <8 x i32> %68, <8 x i32> poison, <8 x i32> zeroinitializer, !dbg !41 + %70 = or disjoint <8 x i32> %69, , !dbg !41 + %71 = insertelement <16 x i32> poison, i32 %61, i64 0, !dbg !41 + %72 = shufflevector <16 x i32> %71, <16 x i32> poison, <16 x i32> zeroinitializer, !dbg !41 + %73 = or disjoint <16 x i32> %72, , !dbg !41 + %74 = insertelement <32 x i32> poison, i32 %63, i64 28, !dbg !42 + %75 = insertelement <32 x i32> %74, i32 %64, i64 29, !dbg !42 + %76 = insertelement <32 x i32> %75, i32 %62, i64 30, !dbg !42 + %77 = insertelement <32 x i32> %76, i32 %61, i64 31, !dbg !42 + %78 = shufflevector <16 x i32> %73, <16 x i32> poison, <32 x i32> , !dbg !42 + %79 = shufflevector <32 x i32> %78, <32 x i32> %77, <32 x i32> , !dbg !42 + %80 = shufflevector <8 x i32> %70, <8 x i32> poison, <32 x i32> , !dbg !42 + %81 = shufflevector <32 x i32> %79, <32 x i32> %80, <32 x i32> , !dbg !42 + %82 = shufflevector <4 x i32> %67, <4 x i32> poison, <32 x i32> , !dbg !42 + %83 = shufflevector <32 x i32> %81, <32 x i32> %82, <32 x i32> , !dbg !42 + %84 = zext <32 x i32> %83 to <32 x i64>, !dbg !42 + %85 = insertelement <32 x i64> poison, i64 %36, i64 0, !dbg !42 + %86 = shufflevector <32 x i64> %85, <32 x i64> poison, <32 x i32> zeroinitializer, !dbg !42 + %87 = or disjoint <32 x i64> %86, %84, !dbg !42 + %88 = icmp slt i64 %56, %26, !dbg !43 + %89 = icmp slt i64 %57, %26, !dbg !43 + %90 = icmp slt i64 %58, %26, !dbg !43 + %91 = icmp slt i64 %59, %26, !dbg !43 + %92 = icmp slt i64 %60, %26, !dbg !43 + %93 = insertelement <32 x i32> poison, i32 %7, i64 0, !dbg !43 + %94 = shufflevector <32 x i32> %93, <32 x i32> poison, <32 x i32> zeroinitializer, !dbg !43 + %95 = sext <32 x i32> %94 to <32 x i64>, !dbg !43 + %96 = icmp slt <32 x i64> %87, %95, !dbg !43 + %97 = shl nuw nsw i64 %15, 6, !dbg !44 + %98 = and i64 %97, 64, !dbg !44 + %99 = or disjoint i64 %98, %55, !dbg !45 + %100 = shl nuw nsw i64 %56, 12, !dbg !46 + %101 = shl nuw nsw i64 %57, 12, !dbg !46 + %102 = shl nuw nsw i64 %58, 12, !dbg !46 + %103 = shl nuw nsw i64 %59, 12, !dbg !46 + %104 = getelementptr bfloat, ptr addrspace(1) %31, i64 %100, !dbg !47 + %105 = getelementptr bfloat, ptr addrspace(1) %31, i64 %101, !dbg !47 + %106 = getelementptr bfloat, ptr addrspace(1) %31, i64 %102, !dbg !47 + %107 = getelementptr bfloat, ptr addrspace(1) %31, i64 %103, !dbg !47 + %108 = tail call i32 @llvm.nvvm.shfl.sync.idx.i32(i32 -1, i32 %37, i32 0, i32 31), !dbg !48 + %109 = shl i32 %108, 21, !dbg !48 + %110 = and i32 %109, 6291456, !dbg !48 + %111 = add i32 %110, %13, !dbg !48 + tail call void asm sideeffect "@$0 tcgen05.st.sync.aligned.16x32bx2.x32.b32 [$1 + 0], 32, {$2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13, $14, $15, $16, $17, $18, $19, $20, $21, $22, $23, $24, $25, $26, $27, $28, $29, $30, $31, $32, $33};", "b,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r"(i1 true, i32 %111, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00) #6, !dbg !48 + tail call void @llvm.nvvm.tcgen05.wait.st(), !dbg !48 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !48 + %112 = add i32 %13, 1048576, !dbg !49 + %113 = add i32 %110, %112, !dbg !49 + tail call void asm sideeffect "@$0 tcgen05.st.sync.aligned.16x32bx2.x32.b32 [$1 + 0], 32, {$2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13, $14, $15, $16, $17, $18, $19, $20, $21, $22, $23, $24, $25, $26, $27, $28, $29, $30, $31, $32, $33};", "b,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r"(i1 true, i32 %113, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00) #6, !dbg !49 + tail call void @llvm.nvvm.tcgen05.wait.st(), !dbg !49 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !49 + %114 = icmp eq i32 %37, 0, !dbg !50 + %115 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !50 + %116 = extractvalue { i32, i1 } %115, 1, !dbg !50 + %117 = and i1 %114, %116, !dbg !50 + tail call void asm sideeffect "@$0 mbarrier.init.shared::cta.b64 [$1], 1;", "b,r"(i1 %117, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73728)) #6, !dbg !50 + %118 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !50 + %119 = extractvalue { i32, i1 } %118, 1, !dbg !50 + %120 = and i1 %114, %119, !dbg !50 + tail call void asm sideeffect "@$0 mbarrier.init.shared::cta.b64 [$1], 1;", "b,r"(i1 %120, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73736)) #6, !dbg !50 + %121 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !50 + %122 = extractvalue { i32, i1 } %121, 1, !dbg !50 + %123 = and i1 %114, %122, !dbg !50 + tail call void asm sideeffect "@$0 mbarrier.init.shared::cta.b64 [$1], 1;", "b,r"(i1 %123, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73744)) #6, !dbg !50 + %124 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !50 + %125 = extractvalue { i32, i1 } %124, 1, !dbg !50 + %126 = and i1 %114, %125, !dbg !50 + tail call void asm sideeffect "@$0 mbarrier.init.shared::cta.b64 [$1], 1;", "b,r"(i1 %126, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73752)) #6, !dbg !50 + %127 = getelementptr bfloat, ptr addrspace(1) %104, i64 %55, !dbg !51 + %128 = getelementptr bfloat, ptr addrspace(1) %105, i64 %55, !dbg !51 + %129 = getelementptr bfloat, ptr addrspace(1) %106, i64 %55, !dbg !51 + %130 = getelementptr bfloat, ptr addrspace(1) %107, i64 %55, !dbg !51 + %131 = getelementptr bfloat, ptr addrspace(1) %32, i64 %55, !dbg !52 + %132 = getelementptr bfloat, ptr addrspace(1) %131, i64 %100, !dbg !53 + %133 = getelementptr bfloat, ptr addrspace(1) %131, i64 %101, !dbg !53 + %134 = getelementptr bfloat, ptr addrspace(1) %131, i64 %102, !dbg !53 + %135 = getelementptr bfloat, ptr addrspace(1) %131, i64 %103, !dbg !53 + %136 = shl nuw nsw i32 %39, 7, !dbg !54 + %137 = shl nuw nsw i32 %40, 7, !dbg !54 + %138 = shl nuw nsw i32 %41, 7, !dbg !54 + %139 = shl nuw nsw i32 %42, 7, !dbg !54 + %140 = zext nneg i32 %136 to i64, !dbg !55 + %141 = getelementptr bfloat, ptr addrspace(1) %35, i64 %140, !dbg !55 + %142 = zext nneg i32 %137 to i64, !dbg !55 + %143 = getelementptr bfloat, ptr addrspace(1) %35, i64 %142, !dbg !55 + %144 = zext nneg i32 %138 to i64, !dbg !55 + %145 = getelementptr bfloat, ptr addrspace(1) %35, i64 %144, !dbg !55 + %146 = zext nneg i32 %139 to i64, !dbg !55 + %147 = getelementptr bfloat, ptr addrspace(1) %35, i64 %146, !dbg !55 + %148 = getelementptr bfloat, ptr addrspace(1) %141, i64 %99, !dbg !56 + %149 = getelementptr bfloat, ptr addrspace(1) %143, i64 %99, !dbg !56 + %150 = getelementptr bfloat, ptr addrspace(1) %145, i64 %99, !dbg !56 + %151 = getelementptr bfloat, ptr addrspace(1) %147, i64 %99, !dbg !56 + %152 = shl nuw nsw i32 %11, 4, !dbg !57 + %153 = and i32 %152, 2032, !dbg !57 + %154 = and i32 %11, 56, !dbg !57 + %155 = shl nuw nsw i32 %154, 1, !dbg !57 + %156 = xor i32 %153, %155, !dbg !57 + %157 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %156, !dbg !57 + %158 = select i1 %88, i32 16, i32 0, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %157, ptr addrspace(1) %127, i32 %158) #6, !dbg !57 + %159 = or disjoint i32 %156, 2048, !dbg !57 + %160 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %159, !dbg !57 + %161 = select i1 %89, i32 16, i32 0, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %160, ptr addrspace(1) %128, i32 %161) #6, !dbg !57 + %162 = or disjoint i32 %156, 4096, !dbg !57 + %163 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %162, !dbg !57 + %164 = select i1 %90, i32 16, i32 0, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %163, ptr addrspace(1) %129, i32 %164) #6, !dbg !57 + %165 = or disjoint i32 %156, 6144, !dbg !57 + %166 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %165, !dbg !57 + %167 = select i1 %91, i32 16, i32 0, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %166, ptr addrspace(1) %130, i32 %167) #6, !dbg !57 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !57 + %168 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 24576), i32 %156, !dbg !58 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %168, ptr addrspace(1) %132, i32 %158) #6, !dbg !58 + %169 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 24576), i32 %159, !dbg !58 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %169, ptr addrspace(1) %133, i32 %161) #6, !dbg !58 + %170 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 24576), i32 %162, !dbg !58 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %170, ptr addrspace(1) %134, i32 %164) #6, !dbg !58 + %171 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 24576), i32 %165, !dbg !58 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %171, ptr addrspace(1) %135, i32 %167) #6, !dbg !58 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !58 + %172 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 49152), i32 %156, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %172, ptr addrspace(1) %148, i32 16) #6, !dbg !59 + %173 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 49152), i32 %159, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %173, ptr addrspace(1) %149, i32 16) #6, !dbg !59 + %174 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 49152), i32 %162, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %174, ptr addrspace(1) %150, i32 16) #6, !dbg !59 + %175 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 49152), i32 %165, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %175, ptr addrspace(1) %151, i32 16) #6, !dbg !59 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !59 + %176 = or disjoint i32 %44, 64, !dbg !60 + %177 = zext nneg i32 %176 to i64, !dbg !51 + %178 = getelementptr bfloat, ptr addrspace(1) %104, i64 %177, !dbg !51 + %179 = getelementptr bfloat, ptr addrspace(1) %105, i64 %177, !dbg !51 + %180 = getelementptr bfloat, ptr addrspace(1) %106, i64 %177, !dbg !51 + %181 = getelementptr bfloat, ptr addrspace(1) %107, i64 %177, !dbg !51 + %182 = getelementptr bfloat, ptr addrspace(1) %32, i64 %177, !dbg !52 + %183 = getelementptr bfloat, ptr addrspace(1) %182, i64 %100, !dbg !53 + %184 = getelementptr bfloat, ptr addrspace(1) %182, i64 %101, !dbg !53 + %185 = getelementptr bfloat, ptr addrspace(1) %182, i64 %102, !dbg !53 + %186 = getelementptr bfloat, ptr addrspace(1) %182, i64 %103, !dbg !53 + %187 = shl nuw nsw i32 %38, 7, !dbg !54 + %188 = or i32 %187, 14336, !dbg !54 + %189 = getelementptr i8, ptr addrspace(1) %141, i64 16384, !dbg !55 + %190 = getelementptr i8, ptr addrspace(1) %141, i64 20480, !dbg !55 + %191 = getelementptr i8, ptr addrspace(1) %141, i64 24576, !dbg !55 + %192 = zext nneg i32 %188 to i64, !dbg !55 + %193 = getelementptr bfloat, ptr addrspace(1) %35, i64 %192, !dbg !55 + %194 = getelementptr bfloat, ptr addrspace(1) %189, i64 %99, !dbg !56 + %195 = getelementptr bfloat, ptr addrspace(1) %190, i64 %99, !dbg !56 + %196 = getelementptr bfloat, ptr addrspace(1) %191, i64 %99, !dbg !56 + %197 = getelementptr bfloat, ptr addrspace(1) %193, i64 %99, !dbg !56 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !57 + %198 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 8192), i32 %156, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %198, ptr addrspace(1) %178, i32 %158) #6, !dbg !57 + %199 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 8192), i32 %159, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %199, ptr addrspace(1) %179, i32 %161) #6, !dbg !57 + %200 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 8192), i32 %162, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %200, ptr addrspace(1) %180, i32 %164) #6, !dbg !57 + %201 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 8192), i32 %165, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %201, ptr addrspace(1) %181, i32 %167) #6, !dbg !57 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !57 + %202 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 32768), i32 %156, !dbg !58 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %202, ptr addrspace(1) %183, i32 %158) #6, !dbg !58 + %203 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 32768), i32 %159, !dbg !58 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %203, ptr addrspace(1) %184, i32 %161) #6, !dbg !58 + %204 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 32768), i32 %162, !dbg !58 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %204, ptr addrspace(1) %185, i32 %164) #6, !dbg !58 + %205 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 32768), i32 %165, !dbg !58 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %205, ptr addrspace(1) %186, i32 %167) #6, !dbg !58 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !58 + %206 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 57344), i32 %156, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %206, ptr addrspace(1) %194, i32 16) #6, !dbg !59 + %207 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 57344), i32 %159, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %207, ptr addrspace(1) %195, i32 16) #6, !dbg !59 + %208 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 57344), i32 %162, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %208, ptr addrspace(1) %196, i32 16) #6, !dbg !59 + %209 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 57344), i32 %165, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %209, ptr addrspace(1) %197, i32 16) #6, !dbg !59 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !59 + tail call void @llvm.nvvm.cp.async.wait.group(i32 3), !dbg !57 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !57 + tail call void @llvm.nvvm.fence.proxy.async.shared_cta(), !dbg !48 + %210 = icmp eq i32 %108, 0, !dbg !48 + br i1 %210, label %211, label %.critedge, !dbg !48 + +211: ; preds = %10 + %212 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !48 + %213 = extractvalue { i32, i1 } %212, 1, !dbg !48 + %214 = lshr exact i32 ptrtoint (ptr addrspace(3) @global_smem to i32), 4, !dbg !48 + %215 = and i32 %214, 16383, !dbg !48 + %216 = zext nneg i32 %215 to i64, !dbg !48 + %217 = lshr exact i32 ptrtoint (ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 49152) to i32), 4, !dbg !48 + %218 = and i32 %217, 16383, !dbg !48 + %219 = zext nneg i32 %218 to i64, !dbg !48 + %220 = or disjoint i64 %216, 4611756662049472512, !dbg !48 + %221 = or disjoint i64 %219, 4611756662049472512, !dbg !48 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %13, i64 %220, i64 %221, i32 68224144, i1 false, i1 %213) #6, !dbg !48 + %222 = add nuw nsw i64 %216, 4611756662049472514, !dbg !48 + %223 = add nuw nsw i64 %219, 4611756662049472640, !dbg !48 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %13, i64 %222, i64 %223, i32 68224144, i1 true, i1 %213) #6, !dbg !48 + %224 = add nuw nsw i64 %216, 4611756662049472516, !dbg !48 + %225 = add nuw nsw i64 %219, 4611756662049472768, !dbg !48 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %13, i64 %224, i64 %225, i32 68224144, i1 true, i1 %213) #6, !dbg !48 + %226 = add nuw nsw i64 %216, 4611756662049472518, !dbg !48 + %227 = add nuw nsw i64 %219, 4611756662049472896, !dbg !48 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %13, i64 %226, i64 %227, i32 68224144, i1 true, i1 %213) #6, !dbg !48 + tail call void asm sideeffect "@$0 tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [$1];", "b,r"(i1 %213, i32 ptrtoint (ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73728) to i32)) #6, !dbg !48 + %228 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !49 + %229 = extractvalue { i32, i1 } %228, 1, !dbg !49 + %230 = lshr exact i32 ptrtoint (ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 24576) to i32), 4, !dbg !49 + %231 = and i32 %230, 16383, !dbg !49 + %232 = zext nneg i32 %231 to i64, !dbg !49 + %233 = or disjoint i64 %232, 4611756662049472512, !dbg !49 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %112, i64 %220, i64 %233, i32 68158608, i1 false, i1 %229) #6, !dbg !49 + %234 = add nuw nsw i64 %232, 4611756662049472514, !dbg !49 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %112, i64 %222, i64 %234, i32 68158608, i1 true, i1 %229) #6, !dbg !49 + %235 = add nuw nsw i64 %232, 4611756662049472516, !dbg !49 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %112, i64 %224, i64 %235, i32 68158608, i1 true, i1 %229) #6, !dbg !49 + %236 = add nuw nsw i64 %232, 4611756662049472518, !dbg !49 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %112, i64 %226, i64 %236, i32 68158608, i1 true, i1 %229) #6, !dbg !49 + tail call void asm sideeffect "@$0 tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [$1];", "b,r"(i1 %229, i32 ptrtoint (ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73744) to i32)) #6, !dbg !49 + br label %.critedge, !dbg !49 + +.critedge: ; preds = %10, %211 + %237 = or disjoint i32 %44, 128, !dbg !60 + %238 = zext nneg i32 %237 to i64, !dbg !51 + %239 = getelementptr bfloat, ptr addrspace(1) %104, i64 %238, !dbg !51 + %240 = getelementptr bfloat, ptr addrspace(1) %105, i64 %238, !dbg !51 + %241 = getelementptr bfloat, ptr addrspace(1) %106, i64 %238, !dbg !51 + %242 = getelementptr bfloat, ptr addrspace(1) %107, i64 %238, !dbg !51 + %243 = getelementptr bfloat, ptr addrspace(1) %32, i64 %238, !dbg !52 + %244 = getelementptr bfloat, ptr addrspace(1) %243, i64 %100, !dbg !53 + %245 = getelementptr bfloat, ptr addrspace(1) %243, i64 %101, !dbg !53 + %246 = getelementptr bfloat, ptr addrspace(1) %243, i64 %102, !dbg !53 + %247 = getelementptr bfloat, ptr addrspace(1) %243, i64 %103, !dbg !53 + %248 = getelementptr i8, ptr addrspace(1) %141, i64 32768, !dbg !55 + %249 = getelementptr i8, ptr addrspace(1) %141, i64 36864, !dbg !55 + %250 = getelementptr i8, ptr addrspace(1) %141, i64 40960, !dbg !55 + %251 = getelementptr i8, ptr addrspace(1) %141, i64 45056, !dbg !55 + %252 = getelementptr bfloat, ptr addrspace(1) %248, i64 %99, !dbg !56 + %253 = getelementptr bfloat, ptr addrspace(1) %249, i64 %99, !dbg !56 + %254 = getelementptr bfloat, ptr addrspace(1) %250, i64 %99, !dbg !56 + %255 = getelementptr bfloat, ptr addrspace(1) %251, i64 %99, !dbg !56 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !57 + %256 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %156, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %256, ptr addrspace(1) %239, i32 0) #6, !dbg !57 + %257 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %159, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %257, ptr addrspace(1) %240, i32 0) #6, !dbg !57 + %258 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %162, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %258, ptr addrspace(1) %241, i32 0) #6, !dbg !57 + %259 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 16384), i32 %165, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %259, ptr addrspace(1) %242, i32 0) #6, !dbg !57 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !57 + %260 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 40960), i32 %156, !dbg !58 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %260, ptr addrspace(1) %244, i32 0) #6, !dbg !58 + %261 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 40960), i32 %159, !dbg !58 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %261, ptr addrspace(1) %245, i32 0) #6, !dbg !58 + %262 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 40960), i32 %162, !dbg !58 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %262, ptr addrspace(1) %246, i32 0) #6, !dbg !58 + %263 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 40960), i32 %165, !dbg !58 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %263, ptr addrspace(1) %247, i32 0) #6, !dbg !58 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !58 + %264 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 65536), i32 %156, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %264, ptr addrspace(1) %252, i32 0) #6, !dbg !59 + %265 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 65536), i32 %159, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %265, ptr addrspace(1) %253, i32 0) #6, !dbg !59 + %266 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 65536), i32 %162, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %266, ptr addrspace(1) %254, i32 0) #6, !dbg !59 + %267 = getelementptr inbounds nuw i8, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 65536), i32 %165, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %267, ptr addrspace(1) %255, i32 0) #6, !dbg !59 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !59 + tail call void @llvm.nvvm.cp.async.wait.group(i32 3), !dbg !57 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !57 + tail call void @llvm.nvvm.fence.proxy.async.shared_cta(), !dbg !48 + br i1 %210, label %268, label %.critedge2, !dbg !48 + +268: ; preds = %.critedge + %269 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !48 + %270 = extractvalue { i32, i1 } %269, 1, !dbg !48 + %271 = lshr exact i32 ptrtoint (ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 8192) to i32), 4, !dbg !48 + %272 = and i32 %271, 16383, !dbg !48 + %273 = zext nneg i32 %272 to i64, !dbg !48 + %274 = lshr exact i32 ptrtoint (ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 57344) to i32), 4, !dbg !48 + %275 = and i32 %274, 16383, !dbg !48 + %276 = zext nneg i32 %275 to i64, !dbg !48 + %277 = or disjoint i64 %273, 4611756662049472512, !dbg !48 + %278 = or disjoint i64 %276, 4611756662049472512, !dbg !48 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %13, i64 %277, i64 %278, i32 68224144, i1 true, i1 %270) #6, !dbg !48 + %279 = add nuw nsw i64 %273, 4611756662049472514, !dbg !48 + %280 = add nuw nsw i64 %276, 4611756662049472640, !dbg !48 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %13, i64 %279, i64 %280, i32 68224144, i1 true, i1 %270) #6, !dbg !48 + %281 = add nuw nsw i64 %273, 4611756662049472516, !dbg !48 + %282 = add nuw nsw i64 %276, 4611756662049472768, !dbg !48 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %13, i64 %281, i64 %282, i32 68224144, i1 true, i1 %270) #6, !dbg !48 + %283 = add nuw nsw i64 %273, 4611756662049472518, !dbg !48 + %284 = add nuw nsw i64 %276, 4611756662049472896, !dbg !48 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %13, i64 %283, i64 %284, i32 68224144, i1 true, i1 %270) #6, !dbg !48 + tail call void asm sideeffect "@$0 tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [$1];", "b,r"(i1 %270, i32 ptrtoint (ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73736) to i32)) #6, !dbg !48 + tail call void asm sideeffect "\0A{\0A\09.reg .pred complete;\0A\09waitLoop:\0A\09mbarrier.try_wait.parity.shared::cta.b64 complete, [$0], $1;\0A\09@!complete bra.uni waitLoop;\0A}\0A", "r,r"(ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73728), i32 0) #6, !dbg !48 + %285 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !49 + %286 = extractvalue { i32, i1 } %285, 1, !dbg !49 + %287 = lshr exact i32 ptrtoint (ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 32768) to i32), 4, !dbg !49 + %288 = and i32 %287, 16383, !dbg !49 + %289 = zext nneg i32 %288 to i64, !dbg !49 + %290 = or disjoint i64 %289, 4611756662049472512, !dbg !49 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %112, i64 %277, i64 %290, i32 68158608, i1 true, i1 %286) #6, !dbg !49 + %291 = add nuw nsw i64 %289, 4611756662049472514, !dbg !49 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %112, i64 %279, i64 %291, i32 68158608, i1 true, i1 %286) #6, !dbg !49 + %292 = add nuw nsw i64 %289, 4611756662049472516, !dbg !49 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %112, i64 %281, i64 %292, i32 68158608, i1 true, i1 %286) #6, !dbg !49 + %293 = add nuw nsw i64 %289, 4611756662049472518, !dbg !49 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], $1, $2, $3, $4;", "r,l,l,r,b,b"(i32 %112, i64 %283, i64 %293, i32 68158608, i1 true, i1 %286) #6, !dbg !49 + tail call void asm sideeffect "@$0 tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [$1];", "b,r"(i1 %286, i32 ptrtoint (ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73752) to i32)) #6, !dbg !49 + br label %294, !dbg !49 + +.critedge2: ; preds = %.critedge + tail call void asm sideeffect "\0A{\0A\09.reg .pred complete;\0A\09waitLoop:\0A\09mbarrier.try_wait.parity.shared::cta.b64 complete, [$0], $1;\0A\09@!complete bra.uni waitLoop;\0A}\0A", "r,r"(ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73728), i32 0) #6, !dbg !48 + br label %294, !dbg !49 + +294: ; preds = %.critedge2, %268 + tail call void asm sideeffect "\0A{\0A\09.reg .pred complete;\0A\09waitLoop:\0A\09mbarrier.try_wait.parity.shared::cta.b64 complete, [$0], $1;\0A\09@!complete bra.uni waitLoop;\0A}\0A", "r,r"(ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73744), i32 0) #6, !dbg !49 + %295 = or disjoint i32 %44, 192, !dbg !60 + %296 = zext nneg i32 %295 to i64, !dbg !51 + %297 = getelementptr bfloat, ptr addrspace(1) %104, i64 %296, !dbg !51 + %298 = getelementptr bfloat, ptr addrspace(1) %105, i64 %296, !dbg !51 + %299 = getelementptr bfloat, ptr addrspace(1) %106, i64 %296, !dbg !51 + %300 = getelementptr bfloat, ptr addrspace(1) %107, i64 %296, !dbg !51 + %301 = getelementptr bfloat, ptr addrspace(1) %32, i64 %296, !dbg !52 + %302 = getelementptr bfloat, ptr addrspace(1) %301, i64 %100, !dbg !53 + %303 = getelementptr bfloat, ptr addrspace(1) %301, i64 %101, !dbg !53 + %304 = getelementptr bfloat, ptr addrspace(1) %301, i64 %102, !dbg !53 + %305 = getelementptr bfloat, ptr addrspace(1) %301, i64 %103, !dbg !53 + %306 = or i32 %187, 30720, !dbg !54 + %307 = getelementptr i8, ptr addrspace(1) %141, i64 49152, !dbg !55 + %308 = getelementptr i8, ptr addrspace(1) %141, i64 53248, !dbg !55 + %309 = getelementptr i8, ptr addrspace(1) %141, i64 57344, !dbg !55 + %310 = zext nneg i32 %306 to i64, !dbg !55 + %311 = getelementptr bfloat, ptr addrspace(1) %35, i64 %310, !dbg !55 + %312 = getelementptr bfloat, ptr addrspace(1) %307, i64 %99, !dbg !56 + %313 = getelementptr bfloat, ptr addrspace(1) %308, i64 %99, !dbg !56 + %314 = getelementptr bfloat, ptr addrspace(1) %309, i64 %99, !dbg !56 + %315 = getelementptr bfloat, ptr addrspace(1) %311, i64 %99, !dbg !56 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %157, ptr addrspace(1) %297, i32 0) #6, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %160, ptr addrspace(1) %298, i32 0) #6, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %163, ptr addrspace(1) %299, i32 0) #6, !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %166, ptr addrspace(1) %300, i32 0) #6, !dbg !57 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !57 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %168, ptr addrspace(1) %302, i32 0) #6, !dbg !58 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %169, ptr addrspace(1) %303, i32 0) #6, !dbg !58 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %170, ptr addrspace(1) %304, i32 0) #6, !dbg !58 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %171, ptr addrspace(1) %305, i32 0) #6, !dbg !58 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !58 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) %172, ptr addrspace(1) %312, i32 0) #6, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %173, ptr addrspace(1) %313, i32 0) #6, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %174, ptr addrspace(1) %314, i32 0) #6, !dbg !59 + tail call void asm sideeffect "cp.async.cg.shared.global [ $0 + 0 ], [ $1 + 0 ], 0x10, $2;", "r,l,r"(ptr addrspace(3) nonnull %175, ptr addrspace(1) %315, i32 0) #6, !dbg !59 + tail call void @llvm.nvvm.cp.async.commit.group(), !dbg !59 + tail call void asm sideeffect "\0A{\0A\09.reg .pred complete;\0A\09waitLoop:\0A\09mbarrier.try_wait.parity.shared::cta.b64 complete, [$0], $1;\0A\09@!complete bra.uni waitLoop;\0A}\0A", "r,r"(ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73736), i32 0) #6, !dbg !48 + tail call void asm sideeffect "\0A{\0A\09.reg .pred complete;\0A\09waitLoop:\0A\09mbarrier.try_wait.parity.shared::cta.b64 complete, [$0], $1;\0A\09@!complete bra.uni waitLoop;\0A}\0A", "r,r"(ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73752), i32 0) #6, !dbg !49 + tail call void @llvm.nvvm.cp.async.wait.group(i32 0), !dbg !50 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !50 + %316 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !50 + %317 = extractvalue { i32, i1 } %316, 1, !dbg !50 + %318 = and i1 %114, %317, !dbg !50 + tail call void asm sideeffect "@$0 mbarrier.inval.shared::cta.b64 [$1];", "b,r"(i1 %318, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73744)) #6, !dbg !50 + %319 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !50 + %320 = extractvalue { i32, i1 } %319, 1, !dbg !50 + %321 = and i1 %114, %320, !dbg !50 + tail call void asm sideeffect "@$0 mbarrier.inval.shared::cta.b64 [$1];", "b,r"(i1 %321, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73752)) #6, !dbg !50 + %322 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !50 + %323 = extractvalue { i32, i1 } %322, 1, !dbg !50 + %324 = and i1 %114, %323, !dbg !50 + tail call void asm sideeffect "@$0 mbarrier.inval.shared::cta.b64 [$1];", "b,r"(i1 %324, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73728)) #6, !dbg !50 + %325 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !50 + %326 = extractvalue { i32, i1 } %325, 1, !dbg !50 + %327 = and i1 %114, %326, !dbg !50 + tail call void asm sideeffect "@$0 mbarrier.inval.shared::cta.b64 [$1];", "b,r"(i1 %327, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 73736)) #6, !dbg !50 + %328 = getelementptr float, ptr addrspace(1) %4, i64 %29, !dbg !61 + %.idx4 = shl nuw nsw i64 %60, 7, !dbg !62 + %329 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx4, !dbg !62 + %330 = extractelement <32 x i64> %87, i64 31, !dbg !62 + %.idx5 = shl nuw nsw i64 %330, 7, !dbg !62 + %331 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx5, !dbg !62 + %332 = extractelement <32 x i64> %87, i64 30, !dbg !62 + %.idx6 = shl nuw nsw i64 %332, 7, !dbg !62 + %333 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx6, !dbg !62 + %334 = extractelement <32 x i64> %87, i64 29, !dbg !62 + %.idx7 = shl nuw nsw i64 %334, 7, !dbg !62 + %335 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx7, !dbg !62 + %336 = extractelement <32 x i64> %87, i64 28, !dbg !62 + %.idx8 = shl nuw nsw i64 %336, 7, !dbg !62 + %337 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx8, !dbg !62 + %338 = extractelement <32 x i64> %87, i64 27, !dbg !62 + %.idx9 = shl nuw nsw i64 %338, 7, !dbg !62 + %339 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx9, !dbg !62 + %340 = extractelement <32 x i64> %87, i64 26, !dbg !62 + %.idx10 = shl nuw nsw i64 %340, 7, !dbg !62 + %341 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx10, !dbg !62 + %342 = extractelement <32 x i64> %87, i64 25, !dbg !62 + %.idx11 = shl nuw nsw i64 %342, 7, !dbg !62 + %343 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx11, !dbg !62 + %344 = extractelement <32 x i64> %87, i64 24, !dbg !62 + %.idx12 = shl nuw nsw i64 %344, 7, !dbg !62 + %345 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx12, !dbg !62 + %346 = extractelement <32 x i64> %87, i64 23, !dbg !62 + %.idx13 = shl nuw nsw i64 %346, 7, !dbg !62 + %347 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx13, !dbg !62 + %348 = extractelement <32 x i64> %87, i64 22, !dbg !62 + %.idx14 = shl nuw nsw i64 %348, 7, !dbg !62 + %349 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx14, !dbg !62 + %350 = extractelement <32 x i64> %87, i64 21, !dbg !62 + %.idx15 = shl nuw nsw i64 %350, 7, !dbg !62 + %351 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx15, !dbg !62 + %352 = extractelement <32 x i64> %87, i64 20, !dbg !62 + %.idx16 = shl nuw nsw i64 %352, 7, !dbg !62 + %353 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx16, !dbg !62 + %354 = extractelement <32 x i64> %87, i64 19, !dbg !62 + %.idx17 = shl nuw nsw i64 %354, 7, !dbg !62 + %355 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx17, !dbg !62 + %356 = extractelement <32 x i64> %87, i64 18, !dbg !62 + %.idx18 = shl nuw nsw i64 %356, 7, !dbg !62 + %357 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx18, !dbg !62 + %358 = extractelement <32 x i64> %87, i64 17, !dbg !62 + %.idx19 = shl nuw nsw i64 %358, 7, !dbg !62 + %359 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx19, !dbg !62 + %360 = extractelement <32 x i64> %87, i64 16, !dbg !62 + %.idx20 = shl nuw nsw i64 %360, 7, !dbg !62 + %361 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx20, !dbg !62 + %362 = extractelement <32 x i64> %87, i64 15, !dbg !62 + %.idx21 = shl nuw nsw i64 %362, 7, !dbg !62 + %363 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx21, !dbg !62 + %364 = extractelement <32 x i64> %87, i64 14, !dbg !62 + %.idx22 = shl nuw nsw i64 %364, 7, !dbg !62 + %365 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx22, !dbg !62 + %366 = extractelement <32 x i64> %87, i64 13, !dbg !62 + %.idx23 = shl nuw nsw i64 %366, 7, !dbg !62 + %367 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx23, !dbg !62 + %368 = extractelement <32 x i64> %87, i64 12, !dbg !62 + %.idx24 = shl nuw nsw i64 %368, 7, !dbg !62 + %369 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx24, !dbg !62 + %370 = extractelement <32 x i64> %87, i64 11, !dbg !62 + %.idx25 = shl nuw nsw i64 %370, 7, !dbg !62 + %371 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx25, !dbg !62 + %372 = extractelement <32 x i64> %87, i64 10, !dbg !62 + %.idx26 = shl nuw nsw i64 %372, 7, !dbg !62 + %373 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx26, !dbg !62 + %374 = extractelement <32 x i64> %87, i64 9, !dbg !62 + %.idx27 = shl nuw nsw i64 %374, 7, !dbg !62 + %375 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx27, !dbg !62 + %376 = extractelement <32 x i64> %87, i64 8, !dbg !62 + %.idx28 = shl nuw nsw i64 %376, 7, !dbg !62 + %377 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx28, !dbg !62 + %378 = extractelement <32 x i64> %87, i64 7, !dbg !62 + %.idx29 = shl nuw nsw i64 %378, 7, !dbg !62 + %379 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx29, !dbg !62 + %380 = extractelement <32 x i64> %87, i64 6, !dbg !62 + %.idx30 = shl nuw nsw i64 %380, 7, !dbg !62 + %381 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx30, !dbg !62 + %382 = extractelement <32 x i64> %87, i64 5, !dbg !62 + %.idx31 = shl nuw nsw i64 %382, 7, !dbg !62 + %383 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx31, !dbg !62 + %384 = extractelement <32 x i64> %87, i64 4, !dbg !62 + %.idx32 = shl nuw nsw i64 %384, 7, !dbg !62 + %385 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx32, !dbg !62 + %386 = extractelement <32 x i64> %87, i64 3, !dbg !62 + %.idx33 = shl nuw nsw i64 %386, 7, !dbg !62 + %387 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx33, !dbg !62 + %388 = extractelement <32 x i64> %87, i64 2, !dbg !62 + %.idx34 = shl nuw nsw i64 %388, 7, !dbg !62 + %389 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx34, !dbg !62 + %390 = extractelement <32 x i64> %87, i64 1, !dbg !62 + %.idx35 = shl nuw nsw i64 %390, 7, !dbg !62 + %391 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx35, !dbg !62 + %392 = extractelement <32 x i64> %87, i64 0, !dbg !62 + %.idx36 = shl nuw nsw i64 %392, 7, !dbg !62 + %393 = getelementptr i8, ptr addrspace(1) %328, i64 %.idx36, !dbg !62 + %394 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %329, i1 %92) #6, !dbg !63 + %395 = bitcast i32 %394 to float, !dbg !63 + %396 = extractelement <32 x i1> %96, i64 31, !dbg !63 + %397 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %331, i1 %396) #6, !dbg !63 + %398 = bitcast i32 %397 to float, !dbg !63 + %399 = extractelement <32 x i1> %96, i64 30, !dbg !63 + %400 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %333, i1 %399) #6, !dbg !63 + %401 = bitcast i32 %400 to float, !dbg !63 + %402 = extractelement <32 x i1> %96, i64 29, !dbg !63 + %403 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %335, i1 %402) #6, !dbg !63 + %404 = bitcast i32 %403 to float, !dbg !63 + %405 = extractelement <32 x i1> %96, i64 28, !dbg !63 + %406 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %337, i1 %405) #6, !dbg !63 + %407 = bitcast i32 %406 to float, !dbg !63 + %408 = extractelement <32 x i1> %96, i64 27, !dbg !63 + %409 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %339, i1 %408) #6, !dbg !63 + %410 = bitcast i32 %409 to float, !dbg !63 + %411 = extractelement <32 x i1> %96, i64 26, !dbg !63 + %412 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %341, i1 %411) #6, !dbg !63 + %413 = bitcast i32 %412 to float, !dbg !63 + %414 = extractelement <32 x i1> %96, i64 25, !dbg !63 + %415 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %343, i1 %414) #6, !dbg !63 + %416 = bitcast i32 %415 to float, !dbg !63 + %417 = extractelement <32 x i1> %96, i64 24, !dbg !63 + %418 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %345, i1 %417) #6, !dbg !63 + %419 = bitcast i32 %418 to float, !dbg !63 + %420 = extractelement <32 x i1> %96, i64 23, !dbg !63 + %421 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %347, i1 %420) #6, !dbg !63 + %422 = bitcast i32 %421 to float, !dbg !63 + %423 = extractelement <32 x i1> %96, i64 22, !dbg !63 + %424 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %349, i1 %423) #6, !dbg !63 + %425 = bitcast i32 %424 to float, !dbg !63 + %426 = extractelement <32 x i1> %96, i64 21, !dbg !63 + %427 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %351, i1 %426) #6, !dbg !63 + %428 = bitcast i32 %427 to float, !dbg !63 + %429 = extractelement <32 x i1> %96, i64 20, !dbg !63 + %430 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %353, i1 %429) #6, !dbg !63 + %431 = bitcast i32 %430 to float, !dbg !63 + %432 = extractelement <32 x i1> %96, i64 19, !dbg !63 + %433 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %355, i1 %432) #6, !dbg !63 + %434 = bitcast i32 %433 to float, !dbg !63 + %435 = extractelement <32 x i1> %96, i64 18, !dbg !63 + %436 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %357, i1 %435) #6, !dbg !63 + %437 = bitcast i32 %436 to float, !dbg !63 + %438 = extractelement <32 x i1> %96, i64 17, !dbg !63 + %439 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %359, i1 %438) #6, !dbg !63 + %440 = bitcast i32 %439 to float, !dbg !63 + %441 = extractelement <32 x i1> %96, i64 16, !dbg !63 + %442 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %361, i1 %441) #6, !dbg !63 + %443 = bitcast i32 %442 to float, !dbg !63 + %444 = extractelement <32 x i1> %96, i64 15, !dbg !63 + %445 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %363, i1 %444) #6, !dbg !63 + %446 = bitcast i32 %445 to float, !dbg !63 + %447 = extractelement <32 x i1> %96, i64 14, !dbg !63 + %448 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %365, i1 %447) #6, !dbg !63 + %449 = bitcast i32 %448 to float, !dbg !63 + %450 = extractelement <32 x i1> %96, i64 13, !dbg !63 + %451 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %367, i1 %450) #6, !dbg !63 + %452 = bitcast i32 %451 to float, !dbg !63 + %453 = extractelement <32 x i1> %96, i64 12, !dbg !63 + %454 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %369, i1 %453) #6, !dbg !63 + %455 = bitcast i32 %454 to float, !dbg !63 + %456 = extractelement <32 x i1> %96, i64 11, !dbg !63 + %457 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %371, i1 %456) #6, !dbg !63 + %458 = bitcast i32 %457 to float, !dbg !63 + %459 = extractelement <32 x i1> %96, i64 10, !dbg !63 + %460 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %373, i1 %459) #6, !dbg !63 + %461 = bitcast i32 %460 to float, !dbg !63 + %462 = extractelement <32 x i1> %96, i64 9, !dbg !63 + %463 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %375, i1 %462) #6, !dbg !63 + %464 = bitcast i32 %463 to float, !dbg !63 + %465 = extractelement <32 x i1> %96, i64 8, !dbg !63 + %466 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %377, i1 %465) #6, !dbg !63 + %467 = bitcast i32 %466 to float, !dbg !63 + %468 = extractelement <32 x i1> %96, i64 7, !dbg !63 + %469 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %379, i1 %468) #6, !dbg !63 + %470 = bitcast i32 %469 to float, !dbg !63 + %471 = extractelement <32 x i1> %96, i64 6, !dbg !63 + %472 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %381, i1 %471) #6, !dbg !63 + %473 = bitcast i32 %472 to float, !dbg !63 + %474 = extractelement <32 x i1> %96, i64 5, !dbg !63 + %475 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %383, i1 %474) #6, !dbg !63 + %476 = bitcast i32 %475 to float, !dbg !63 + %477 = extractelement <32 x i1> %96, i64 4, !dbg !63 + %478 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %385, i1 %477) #6, !dbg !63 + %479 = bitcast i32 %478 to float, !dbg !63 + %480 = extractelement <32 x i1> %96, i64 3, !dbg !63 + %481 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %387, i1 %480) #6, !dbg !63 + %482 = bitcast i32 %481 to float, !dbg !63 + %483 = extractelement <32 x i1> %96, i64 2, !dbg !63 + %484 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %389, i1 %483) #6, !dbg !63 + %485 = bitcast i32 %484 to float, !dbg !63 + %486 = extractelement <32 x i1> %96, i64 1, !dbg !63 + %487 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %391, i1 %486) #6, !dbg !63 + %488 = bitcast i32 %487 to float, !dbg !63 + %489 = extractelement <32 x i1> %96, i64 0, !dbg !63 + %490 = tail call i32 asm sideeffect "mov.u32 $0, $1;\0A\09@$3 ld.global.b32 { $0 }, [ $2 + 0 ];", "=r,r,l,b"(i32 0, ptr addrspace(1) %393, i1 %489) #6, !dbg !63 + %491 = bitcast i32 %490 to float, !dbg !63 + %492 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !64 + %.not.i = icmp eq i32 %492, 0, !dbg !64 + br i1 %.not.i, label %495, label %493, !dbg !64 + +493: ; preds = %294 + %494 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %395), !dbg !64 + br label %__nv_exp2f.exit, !dbg !64 + +495: ; preds = %294 + %496 = tail call float @llvm.nvvm.ex2.approx.f32(float %395), !dbg !64 + br label %__nv_exp2f.exit, !dbg !64 + +__nv_exp2f.exit: ; preds = %493, %495 + %.0.i = phi float [ %494, %493 ], [ %496, %495 ], !dbg !64 + %497 = fsub float %395, %398, !dbg !66 + %498 = fsub float %395, %401, !dbg !66 + %499 = fsub float %395, %404, !dbg !66 + %500 = fsub float %395, %407, !dbg !66 + %501 = fsub float %395, %410, !dbg !66 + %502 = fsub float %395, %413, !dbg !66 + %503 = fsub float %395, %416, !dbg !66 + %504 = fsub float %395, %419, !dbg !66 + %505 = fsub float %395, %422, !dbg !66 + %506 = fsub float %395, %425, !dbg !66 + %507 = fsub float %395, %428, !dbg !66 + %508 = fsub float %395, %431, !dbg !66 + %509 = fsub float %395, %434, !dbg !66 + %510 = fsub float %395, %437, !dbg !66 + %511 = fsub float %395, %440, !dbg !66 + %512 = fsub float %395, %443, !dbg !66 + %513 = fsub float %395, %446, !dbg !66 + %514 = fsub float %395, %449, !dbg !66 + %515 = fsub float %395, %452, !dbg !66 + %516 = fsub float %395, %455, !dbg !66 + %517 = fsub float %395, %458, !dbg !66 + %518 = fsub float %395, %461, !dbg !66 + %519 = fsub float %395, %464, !dbg !66 + %520 = fsub float %395, %467, !dbg !66 + %521 = fsub float %395, %470, !dbg !66 + %522 = fsub float %395, %473, !dbg !66 + %523 = fsub float %395, %476, !dbg !66 + %524 = fsub float %395, %479, !dbg !66 + %525 = fsub float %395, %482, !dbg !66 + %526 = fsub float %395, %485, !dbg !66 + %527 = fsub float %395, %488, !dbg !66 + %528 = fsub float %395, %491, !dbg !66 + %529 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i49 = icmp eq i32 %529, 0, !dbg !67 + br i1 %.not.i49, label %532, label %530, !dbg !67 + +530: ; preds = %__nv_exp2f.exit + %531 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %497), !dbg !67 + br label %__nv_exp2f.exit51, !dbg !67 + +532: ; preds = %__nv_exp2f.exit + %533 = tail call float @llvm.nvvm.ex2.approx.f32(float %497), !dbg !67 + br label %__nv_exp2f.exit51, !dbg !67 + +__nv_exp2f.exit51: ; preds = %530, %532 + %.0.i50 = phi float [ %531, %530 ], [ %533, %532 ], !dbg !67 + %534 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i52 = icmp eq i32 %534, 0, !dbg !67 + br i1 %.not.i52, label %537, label %535, !dbg !67 + +535: ; preds = %__nv_exp2f.exit51 + %536 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %498), !dbg !67 + br label %__nv_exp2f.exit54, !dbg !67 + +537: ; preds = %__nv_exp2f.exit51 + %538 = tail call float @llvm.nvvm.ex2.approx.f32(float %498), !dbg !67 + br label %__nv_exp2f.exit54, !dbg !67 + +__nv_exp2f.exit54: ; preds = %535, %537 + %.0.i53 = phi float [ %536, %535 ], [ %538, %537 ], !dbg !67 + %539 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i55 = icmp eq i32 %539, 0, !dbg !67 + br i1 %.not.i55, label %542, label %540, !dbg !67 + +540: ; preds = %__nv_exp2f.exit54 + %541 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %499), !dbg !67 + br label %__nv_exp2f.exit57, !dbg !67 + +542: ; preds = %__nv_exp2f.exit54 + %543 = tail call float @llvm.nvvm.ex2.approx.f32(float %499), !dbg !67 + br label %__nv_exp2f.exit57, !dbg !67 + +__nv_exp2f.exit57: ; preds = %540, %542 + %.0.i56 = phi float [ %541, %540 ], [ %543, %542 ], !dbg !67 + %544 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i58 = icmp eq i32 %544, 0, !dbg !67 + br i1 %.not.i58, label %547, label %545, !dbg !67 + +545: ; preds = %__nv_exp2f.exit57 + %546 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %500), !dbg !67 + br label %__nv_exp2f.exit60, !dbg !67 + +547: ; preds = %__nv_exp2f.exit57 + %548 = tail call float @llvm.nvvm.ex2.approx.f32(float %500), !dbg !67 + br label %__nv_exp2f.exit60, !dbg !67 + +__nv_exp2f.exit60: ; preds = %545, %547 + %.0.i59 = phi float [ %546, %545 ], [ %548, %547 ], !dbg !67 + %549 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i61 = icmp eq i32 %549, 0, !dbg !67 + br i1 %.not.i61, label %552, label %550, !dbg !67 + +550: ; preds = %__nv_exp2f.exit60 + %551 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %501), !dbg !67 + br label %__nv_exp2f.exit63, !dbg !67 + +552: ; preds = %__nv_exp2f.exit60 + %553 = tail call float @llvm.nvvm.ex2.approx.f32(float %501), !dbg !67 + br label %__nv_exp2f.exit63, !dbg !67 + +__nv_exp2f.exit63: ; preds = %550, %552 + %.0.i62 = phi float [ %551, %550 ], [ %553, %552 ], !dbg !67 + %554 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i64 = icmp eq i32 %554, 0, !dbg !67 + br i1 %.not.i64, label %557, label %555, !dbg !67 + +555: ; preds = %__nv_exp2f.exit63 + %556 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %502), !dbg !67 + br label %__nv_exp2f.exit66, !dbg !67 + +557: ; preds = %__nv_exp2f.exit63 + %558 = tail call float @llvm.nvvm.ex2.approx.f32(float %502), !dbg !67 + br label %__nv_exp2f.exit66, !dbg !67 + +__nv_exp2f.exit66: ; preds = %555, %557 + %.0.i65 = phi float [ %556, %555 ], [ %558, %557 ], !dbg !67 + %559 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i67 = icmp eq i32 %559, 0, !dbg !67 + br i1 %.not.i67, label %562, label %560, !dbg !67 + +560: ; preds = %__nv_exp2f.exit66 + %561 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %503), !dbg !67 + br label %__nv_exp2f.exit69, !dbg !67 + +562: ; preds = %__nv_exp2f.exit66 + %563 = tail call float @llvm.nvvm.ex2.approx.f32(float %503), !dbg !67 + br label %__nv_exp2f.exit69, !dbg !67 + +__nv_exp2f.exit69: ; preds = %560, %562 + %.0.i68 = phi float [ %561, %560 ], [ %563, %562 ], !dbg !67 + %564 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i70 = icmp eq i32 %564, 0, !dbg !67 + br i1 %.not.i70, label %567, label %565, !dbg !67 + +565: ; preds = %__nv_exp2f.exit69 + %566 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %504), !dbg !67 + br label %__nv_exp2f.exit72, !dbg !67 + +567: ; preds = %__nv_exp2f.exit69 + %568 = tail call float @llvm.nvvm.ex2.approx.f32(float %504), !dbg !67 + br label %__nv_exp2f.exit72, !dbg !67 + +__nv_exp2f.exit72: ; preds = %565, %567 + %.0.i71 = phi float [ %566, %565 ], [ %568, %567 ], !dbg !67 + %569 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i73 = icmp eq i32 %569, 0, !dbg !67 + br i1 %.not.i73, label %572, label %570, !dbg !67 + +570: ; preds = %__nv_exp2f.exit72 + %571 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %505), !dbg !67 + br label %__nv_exp2f.exit75, !dbg !67 + +572: ; preds = %__nv_exp2f.exit72 + %573 = tail call float @llvm.nvvm.ex2.approx.f32(float %505), !dbg !67 + br label %__nv_exp2f.exit75, !dbg !67 + +__nv_exp2f.exit75: ; preds = %570, %572 + %.0.i74 = phi float [ %571, %570 ], [ %573, %572 ], !dbg !67 + %574 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i76 = icmp eq i32 %574, 0, !dbg !67 + br i1 %.not.i76, label %577, label %575, !dbg !67 + +575: ; preds = %__nv_exp2f.exit75 + %576 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %506), !dbg !67 + br label %__nv_exp2f.exit78, !dbg !67 + +577: ; preds = %__nv_exp2f.exit75 + %578 = tail call float @llvm.nvvm.ex2.approx.f32(float %506), !dbg !67 + br label %__nv_exp2f.exit78, !dbg !67 + +__nv_exp2f.exit78: ; preds = %575, %577 + %.0.i77 = phi float [ %576, %575 ], [ %578, %577 ], !dbg !67 + %579 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i79 = icmp eq i32 %579, 0, !dbg !67 + br i1 %.not.i79, label %582, label %580, !dbg !67 + +580: ; preds = %__nv_exp2f.exit78 + %581 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %507), !dbg !67 + br label %__nv_exp2f.exit81, !dbg !67 + +582: ; preds = %__nv_exp2f.exit78 + %583 = tail call float @llvm.nvvm.ex2.approx.f32(float %507), !dbg !67 + br label %__nv_exp2f.exit81, !dbg !67 + +__nv_exp2f.exit81: ; preds = %580, %582 + %.0.i80 = phi float [ %581, %580 ], [ %583, %582 ], !dbg !67 + %584 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i82 = icmp eq i32 %584, 0, !dbg !67 + br i1 %.not.i82, label %587, label %585, !dbg !67 + +585: ; preds = %__nv_exp2f.exit81 + %586 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %508), !dbg !67 + br label %__nv_exp2f.exit84, !dbg !67 + +587: ; preds = %__nv_exp2f.exit81 + %588 = tail call float @llvm.nvvm.ex2.approx.f32(float %508), !dbg !67 + br label %__nv_exp2f.exit84, !dbg !67 + +__nv_exp2f.exit84: ; preds = %585, %587 + %.0.i83 = phi float [ %586, %585 ], [ %588, %587 ], !dbg !67 + %589 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i85 = icmp eq i32 %589, 0, !dbg !67 + br i1 %.not.i85, label %592, label %590, !dbg !67 + +590: ; preds = %__nv_exp2f.exit84 + %591 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %509), !dbg !67 + br label %__nv_exp2f.exit87, !dbg !67 + +592: ; preds = %__nv_exp2f.exit84 + %593 = tail call float @llvm.nvvm.ex2.approx.f32(float %509), !dbg !67 + br label %__nv_exp2f.exit87, !dbg !67 + +__nv_exp2f.exit87: ; preds = %590, %592 + %.0.i86 = phi float [ %591, %590 ], [ %593, %592 ], !dbg !67 + %594 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i88 = icmp eq i32 %594, 0, !dbg !67 + br i1 %.not.i88, label %597, label %595, !dbg !67 + +595: ; preds = %__nv_exp2f.exit87 + %596 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %510), !dbg !67 + br label %__nv_exp2f.exit90, !dbg !67 + +597: ; preds = %__nv_exp2f.exit87 + %598 = tail call float @llvm.nvvm.ex2.approx.f32(float %510), !dbg !67 + br label %__nv_exp2f.exit90, !dbg !67 + +__nv_exp2f.exit90: ; preds = %595, %597 + %.0.i89 = phi float [ %596, %595 ], [ %598, %597 ], !dbg !67 + %599 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i91 = icmp eq i32 %599, 0, !dbg !67 + br i1 %.not.i91, label %602, label %600, !dbg !67 + +600: ; preds = %__nv_exp2f.exit90 + %601 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %511), !dbg !67 + br label %__nv_exp2f.exit93, !dbg !67 + +602: ; preds = %__nv_exp2f.exit90 + %603 = tail call float @llvm.nvvm.ex2.approx.f32(float %511), !dbg !67 + br label %__nv_exp2f.exit93, !dbg !67 + +__nv_exp2f.exit93: ; preds = %600, %602 + %.0.i92 = phi float [ %601, %600 ], [ %603, %602 ], !dbg !67 + %604 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i94 = icmp eq i32 %604, 0, !dbg !67 + br i1 %.not.i94, label %607, label %605, !dbg !67 + +605: ; preds = %__nv_exp2f.exit93 + %606 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %512), !dbg !67 + br label %__nv_exp2f.exit96, !dbg !67 + +607: ; preds = %__nv_exp2f.exit93 + %608 = tail call float @llvm.nvvm.ex2.approx.f32(float %512), !dbg !67 + br label %__nv_exp2f.exit96, !dbg !67 + +__nv_exp2f.exit96: ; preds = %605, %607 + %.0.i95 = phi float [ %606, %605 ], [ %608, %607 ], !dbg !67 + %609 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i97 = icmp eq i32 %609, 0, !dbg !67 + br i1 %.not.i97, label %612, label %610, !dbg !67 + +610: ; preds = %__nv_exp2f.exit96 + %611 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %513), !dbg !67 + br label %__nv_exp2f.exit99, !dbg !67 + +612: ; preds = %__nv_exp2f.exit96 + %613 = tail call float @llvm.nvvm.ex2.approx.f32(float %513), !dbg !67 + br label %__nv_exp2f.exit99, !dbg !67 + +__nv_exp2f.exit99: ; preds = %610, %612 + %.0.i98 = phi float [ %611, %610 ], [ %613, %612 ], !dbg !67 + %614 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i100 = icmp eq i32 %614, 0, !dbg !67 + br i1 %.not.i100, label %617, label %615, !dbg !67 + +615: ; preds = %__nv_exp2f.exit99 + %616 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %514), !dbg !67 + br label %__nv_exp2f.exit102, !dbg !67 + +617: ; preds = %__nv_exp2f.exit99 + %618 = tail call float @llvm.nvvm.ex2.approx.f32(float %514), !dbg !67 + br label %__nv_exp2f.exit102, !dbg !67 + +__nv_exp2f.exit102: ; preds = %615, %617 + %.0.i101 = phi float [ %616, %615 ], [ %618, %617 ], !dbg !67 + %619 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i103 = icmp eq i32 %619, 0, !dbg !67 + br i1 %.not.i103, label %622, label %620, !dbg !67 + +620: ; preds = %__nv_exp2f.exit102 + %621 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %515), !dbg !67 + br label %__nv_exp2f.exit105, !dbg !67 + +622: ; preds = %__nv_exp2f.exit102 + %623 = tail call float @llvm.nvvm.ex2.approx.f32(float %515), !dbg !67 + br label %__nv_exp2f.exit105, !dbg !67 + +__nv_exp2f.exit105: ; preds = %620, %622 + %.0.i104 = phi float [ %621, %620 ], [ %623, %622 ], !dbg !67 + %624 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i106 = icmp eq i32 %624, 0, !dbg !67 + br i1 %.not.i106, label %627, label %625, !dbg !67 + +625: ; preds = %__nv_exp2f.exit105 + %626 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %516), !dbg !67 + br label %__nv_exp2f.exit108, !dbg !67 + +627: ; preds = %__nv_exp2f.exit105 + %628 = tail call float @llvm.nvvm.ex2.approx.f32(float %516), !dbg !67 + br label %__nv_exp2f.exit108, !dbg !67 + +__nv_exp2f.exit108: ; preds = %625, %627 + %.0.i107 = phi float [ %626, %625 ], [ %628, %627 ], !dbg !67 + %629 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i109 = icmp eq i32 %629, 0, !dbg !67 + br i1 %.not.i109, label %632, label %630, !dbg !67 + +630: ; preds = %__nv_exp2f.exit108 + %631 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %517), !dbg !67 + br label %__nv_exp2f.exit111, !dbg !67 + +632: ; preds = %__nv_exp2f.exit108 + %633 = tail call float @llvm.nvvm.ex2.approx.f32(float %517), !dbg !67 + br label %__nv_exp2f.exit111, !dbg !67 + +__nv_exp2f.exit111: ; preds = %630, %632 + %.0.i110 = phi float [ %631, %630 ], [ %633, %632 ], !dbg !67 + %634 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i112 = icmp eq i32 %634, 0, !dbg !67 + br i1 %.not.i112, label %637, label %635, !dbg !67 + +635: ; preds = %__nv_exp2f.exit111 + %636 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %518), !dbg !67 + br label %__nv_exp2f.exit114, !dbg !67 + +637: ; preds = %__nv_exp2f.exit111 + %638 = tail call float @llvm.nvvm.ex2.approx.f32(float %518), !dbg !67 + br label %__nv_exp2f.exit114, !dbg !67 + +__nv_exp2f.exit114: ; preds = %635, %637 + %.0.i113 = phi float [ %636, %635 ], [ %638, %637 ], !dbg !67 + %639 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i115 = icmp eq i32 %639, 0, !dbg !67 + br i1 %.not.i115, label %642, label %640, !dbg !67 + +640: ; preds = %__nv_exp2f.exit114 + %641 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %519), !dbg !67 + br label %__nv_exp2f.exit117, !dbg !67 + +642: ; preds = %__nv_exp2f.exit114 + %643 = tail call float @llvm.nvvm.ex2.approx.f32(float %519), !dbg !67 + br label %__nv_exp2f.exit117, !dbg !67 + +__nv_exp2f.exit117: ; preds = %640, %642 + %.0.i116 = phi float [ %641, %640 ], [ %643, %642 ], !dbg !67 + %644 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i118 = icmp eq i32 %644, 0, !dbg !67 + br i1 %.not.i118, label %647, label %645, !dbg !67 + +645: ; preds = %__nv_exp2f.exit117 + %646 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %520), !dbg !67 + br label %__nv_exp2f.exit120, !dbg !67 + +647: ; preds = %__nv_exp2f.exit117 + %648 = tail call float @llvm.nvvm.ex2.approx.f32(float %520), !dbg !67 + br label %__nv_exp2f.exit120, !dbg !67 + +__nv_exp2f.exit120: ; preds = %645, %647 + %.0.i119 = phi float [ %646, %645 ], [ %648, %647 ], !dbg !67 + %649 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i121 = icmp eq i32 %649, 0, !dbg !67 + br i1 %.not.i121, label %652, label %650, !dbg !67 + +650: ; preds = %__nv_exp2f.exit120 + %651 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %521), !dbg !67 + br label %__nv_exp2f.exit123, !dbg !67 + +652: ; preds = %__nv_exp2f.exit120 + %653 = tail call float @llvm.nvvm.ex2.approx.f32(float %521), !dbg !67 + br label %__nv_exp2f.exit123, !dbg !67 + +__nv_exp2f.exit123: ; preds = %650, %652 + %.0.i122 = phi float [ %651, %650 ], [ %653, %652 ], !dbg !67 + %654 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i124 = icmp eq i32 %654, 0, !dbg !67 + br i1 %.not.i124, label %657, label %655, !dbg !67 + +655: ; preds = %__nv_exp2f.exit123 + %656 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %522), !dbg !67 + br label %__nv_exp2f.exit126, !dbg !67 + +657: ; preds = %__nv_exp2f.exit123 + %658 = tail call float @llvm.nvvm.ex2.approx.f32(float %522), !dbg !67 + br label %__nv_exp2f.exit126, !dbg !67 + +__nv_exp2f.exit126: ; preds = %655, %657 + %.0.i125 = phi float [ %656, %655 ], [ %658, %657 ], !dbg !67 + %659 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i127 = icmp eq i32 %659, 0, !dbg !67 + br i1 %.not.i127, label %662, label %660, !dbg !67 + +660: ; preds = %__nv_exp2f.exit126 + %661 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %523), !dbg !67 + br label %__nv_exp2f.exit129, !dbg !67 + +662: ; preds = %__nv_exp2f.exit126 + %663 = tail call float @llvm.nvvm.ex2.approx.f32(float %523), !dbg !67 + br label %__nv_exp2f.exit129, !dbg !67 + +__nv_exp2f.exit129: ; preds = %660, %662 + %.0.i128 = phi float [ %661, %660 ], [ %663, %662 ], !dbg !67 + %664 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i130 = icmp eq i32 %664, 0, !dbg !67 + br i1 %.not.i130, label %667, label %665, !dbg !67 + +665: ; preds = %__nv_exp2f.exit129 + %666 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %524), !dbg !67 + br label %__nv_exp2f.exit132, !dbg !67 + +667: ; preds = %__nv_exp2f.exit129 + %668 = tail call float @llvm.nvvm.ex2.approx.f32(float %524), !dbg !67 + br label %__nv_exp2f.exit132, !dbg !67 + +__nv_exp2f.exit132: ; preds = %665, %667 + %.0.i131 = phi float [ %666, %665 ], [ %668, %667 ], !dbg !67 + %669 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i133 = icmp eq i32 %669, 0, !dbg !67 + br i1 %.not.i133, label %672, label %670, !dbg !67 + +670: ; preds = %__nv_exp2f.exit132 + %671 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %525), !dbg !67 + br label %__nv_exp2f.exit135, !dbg !67 + +672: ; preds = %__nv_exp2f.exit132 + %673 = tail call float @llvm.nvvm.ex2.approx.f32(float %525), !dbg !67 + br label %__nv_exp2f.exit135, !dbg !67 + +__nv_exp2f.exit135: ; preds = %670, %672 + %.0.i134 = phi float [ %671, %670 ], [ %673, %672 ], !dbg !67 + %674 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i136 = icmp eq i32 %674, 0, !dbg !67 + br i1 %.not.i136, label %677, label %675, !dbg !67 + +675: ; preds = %__nv_exp2f.exit135 + %676 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %526), !dbg !67 + br label %__nv_exp2f.exit138, !dbg !67 + +677: ; preds = %__nv_exp2f.exit135 + %678 = tail call float @llvm.nvvm.ex2.approx.f32(float %526), !dbg !67 + br label %__nv_exp2f.exit138, !dbg !67 + +__nv_exp2f.exit138: ; preds = %675, %677 + %.0.i137 = phi float [ %676, %675 ], [ %678, %677 ], !dbg !67 + %679 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i139 = icmp eq i32 %679, 0, !dbg !67 + br i1 %.not.i139, label %682, label %680, !dbg !67 + +680: ; preds = %__nv_exp2f.exit138 + %681 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %527), !dbg !67 + br label %__nv_exp2f.exit141, !dbg !67 + +682: ; preds = %__nv_exp2f.exit138 + %683 = tail call float @llvm.nvvm.ex2.approx.f32(float %527), !dbg !67 + br label %__nv_exp2f.exit141, !dbg !67 + +__nv_exp2f.exit141: ; preds = %680, %682 + %.0.i140 = phi float [ %681, %680 ], [ %683, %682 ], !dbg !67 + %684 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !67 + %.not.i142 = icmp eq i32 %684, 0, !dbg !67 + br i1 %.not.i142, label %687, label %685, !dbg !67 + +685: ; preds = %__nv_exp2f.exit141 + %686 = tail call float @llvm.nvvm.ex2.approx.ftz.f32(float %528), !dbg !67 + br label %__nv_exp2f.exit144, !dbg !67 + +687: ; preds = %__nv_exp2f.exit141 + %688 = tail call float @llvm.nvvm.ex2.approx.f32(float %528), !dbg !67 + br label %__nv_exp2f.exit144, !dbg !67 + +__nv_exp2f.exit144: ; preds = %685, %687 + %.0.i143 = phi float [ %686, %685 ], [ %688, %687 ], !dbg !67 + %689 = icmp samesign uge i32 %48, %61, !dbg !69 + %690 = icmp samesign ugt i32 %48, %61, !dbg !69 + %691 = icmp samesign uge i32 %48, %64, !dbg !69 + %692 = icmp samesign uge i32 %48, %63, !dbg !69 + %693 = extractelement <4 x i32> %67, i64 3, !dbg !69 + %694 = icmp samesign uge i32 %48, %693, !dbg !69 + %695 = extractelement <4 x i32> %67, i64 2, !dbg !69 + %696 = icmp samesign uge i32 %48, %695, !dbg !69 + %697 = extractelement <4 x i32> %67, i64 1, !dbg !69 + %698 = icmp samesign uge i32 %48, %697, !dbg !69 + %699 = extractelement <4 x i32> %67, i64 0, !dbg !69 + %700 = icmp samesign uge i32 %48, %699, !dbg !69 + %701 = extractelement <8 x i32> %70, i64 7, !dbg !69 + %702 = icmp samesign uge i32 %48, %701, !dbg !69 + %703 = extractelement <8 x i32> %70, i64 6, !dbg !69 + %704 = icmp samesign uge i32 %48, %703, !dbg !69 + %705 = extractelement <8 x i32> %70, i64 5, !dbg !69 + %706 = icmp samesign uge i32 %48, %705, !dbg !69 + %707 = extractelement <8 x i32> %70, i64 4, !dbg !69 + %708 = icmp samesign uge i32 %48, %707, !dbg !69 + %709 = extractelement <8 x i32> %70, i64 3, !dbg !69 + %710 = icmp samesign uge i32 %48, %709, !dbg !69 + %711 = extractelement <8 x i32> %70, i64 2, !dbg !69 + %712 = icmp samesign uge i32 %48, %711, !dbg !69 + %713 = extractelement <8 x i32> %70, i64 1, !dbg !69 + %714 = icmp samesign uge i32 %48, %713, !dbg !69 + %715 = extractelement <8 x i32> %70, i64 0, !dbg !69 + %716 = icmp samesign uge i32 %48, %715, !dbg !69 + %717 = extractelement <16 x i32> %73, i64 15, !dbg !69 + %718 = icmp samesign uge i32 %48, %717, !dbg !69 + %719 = extractelement <16 x i32> %73, i64 14, !dbg !69 + %720 = icmp samesign uge i32 %48, %719, !dbg !69 + %721 = extractelement <16 x i32> %73, i64 13, !dbg !69 + %722 = icmp samesign uge i32 %48, %721, !dbg !69 + %723 = extractelement <16 x i32> %73, i64 12, !dbg !69 + %724 = icmp samesign uge i32 %48, %723, !dbg !69 + %725 = extractelement <16 x i32> %73, i64 11, !dbg !69 + %726 = icmp samesign uge i32 %48, %725, !dbg !69 + %727 = extractelement <16 x i32> %73, i64 10, !dbg !69 + %728 = icmp samesign uge i32 %48, %727, !dbg !69 + %729 = extractelement <16 x i32> %73, i64 9, !dbg !69 + %730 = icmp samesign uge i32 %48, %729, !dbg !69 + %731 = extractelement <16 x i32> %73, i64 8, !dbg !69 + %732 = icmp samesign uge i32 %48, %731, !dbg !69 + %733 = extractelement <16 x i32> %73, i64 7, !dbg !69 + %734 = icmp samesign uge i32 %48, %733, !dbg !69 + %735 = extractelement <16 x i32> %73, i64 6, !dbg !69 + %736 = icmp samesign uge i32 %48, %735, !dbg !69 + %737 = extractelement <16 x i32> %73, i64 5, !dbg !69 + %738 = icmp samesign uge i32 %48, %737, !dbg !69 + %739 = extractelement <16 x i32> %73, i64 4, !dbg !69 + %740 = icmp samesign uge i32 %48, %739, !dbg !69 + %741 = extractelement <16 x i32> %73, i64 3, !dbg !69 + %742 = icmp samesign uge i32 %48, %741, !dbg !69 + %743 = extractelement <16 x i32> %73, i64 2, !dbg !69 + %744 = icmp samesign uge i32 %48, %743, !dbg !69 + %745 = extractelement <16 x i32> %73, i64 1, !dbg !69 + %746 = icmp samesign uge i32 %48, %745, !dbg !69 + %747 = extractelement <16 x i32> %73, i64 0, !dbg !69 + %748 = icmp samesign uge i32 %48, %747, !dbg !69 + %749 = and i1 %689, %396, !dbg !70 + %750 = and i1 %92, %749, !dbg !70 + %751 = and i1 %690, %399, !dbg !70 + %752 = and i1 %92, %751, !dbg !70 + %753 = getelementptr bfloat, ptr addrspace(1) %33, i64 %100, !dbg !71 + %754 = getelementptr bfloat, ptr addrspace(1) %33, i64 %101, !dbg !71 + %755 = getelementptr bfloat, ptr addrspace(1) %33, i64 %102, !dbg !71 + %756 = getelementptr bfloat, ptr addrspace(1) %33, i64 %103, !dbg !71 + %757 = getelementptr bfloat, ptr addrspace(1) %753, i64 %99, !dbg !72 + %758 = getelementptr bfloat, ptr addrspace(1) %754, i64 %99, !dbg !72 + %759 = getelementptr bfloat, ptr addrspace(1) %755, i64 %99, !dbg !72 + %760 = getelementptr bfloat, ptr addrspace(1) %756, i64 %99, !dbg !72 + %761 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %757, i1 %88) #6, !dbg !73 + %762 = extractvalue { i32, i32, i32, i32 } %761, 0, !dbg !73 + %763 = extractvalue { i32, i32, i32, i32 } %761, 1, !dbg !73 + %764 = extractvalue { i32, i32, i32, i32 } %761, 2, !dbg !73 + %765 = extractvalue { i32, i32, i32, i32 } %761, 3, !dbg !73 + %766 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %758, i1 %89) #6, !dbg !73 + %767 = extractvalue { i32, i32, i32, i32 } %766, 0, !dbg !73 + %768 = extractvalue { i32, i32, i32, i32 } %766, 1, !dbg !73 + %769 = extractvalue { i32, i32, i32, i32 } %766, 2, !dbg !73 + %770 = extractvalue { i32, i32, i32, i32 } %766, 3, !dbg !73 + %771 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %759, i1 %90) #6, !dbg !73 + %772 = extractvalue { i32, i32, i32, i32 } %771, 0, !dbg !73 + %773 = extractvalue { i32, i32, i32, i32 } %771, 1, !dbg !73 + %774 = extractvalue { i32, i32, i32, i32 } %771, 2, !dbg !73 + %775 = extractvalue { i32, i32, i32, i32 } %771, 3, !dbg !73 + %776 = tail call { i32, i32, i32, i32 } asm sideeffect "mov.u32 $0, $4;\0A\09mov.u32 $1, $5;\0A\09mov.u32 $2, $6;\0A\09mov.u32 $3, $7;\0A\09@$9 ld.global.v4.b32 { $0, $1, $2, $3 }, [ $8 + 0 ];", "=r,=r,=r,=r,r,r,r,r,l,b"(i32 0, i32 0, i32 0, i32 0, ptr addrspace(1) %760, i1 %91) #6, !dbg !73 + %777 = extractvalue { i32, i32, i32, i32 } %776, 0, !dbg !73 + %778 = extractvalue { i32, i32, i32, i32 } %776, 1, !dbg !73 + %779 = extractvalue { i32, i32, i32, i32 } %776, 2, !dbg !73 + %780 = extractvalue { i32, i32, i32, i32 } %776, 3, !dbg !73 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %157, i32 %762, i32 %763, i32 %764, i32 %765) #6, !dbg !73 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %160, i32 %767, i32 %768, i32 %769, i32 %770) #6, !dbg !73 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %163, i32 %772, i32 %773, i32 %774, i32 %775) #6, !dbg !73 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) nonnull %166, i32 %777, i32 %778, i32 %779, i32 %780) #6, !dbg !73 + %781 = tail call { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } asm sideeffect "tcgen05.ld.sync.aligned.16x32bx2.x32.b32 {$0, $1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13, $14, $15, $16, $17, $18, $19, $20, $21, $22, $23, $24, $25, $26, $27, $28, $29, $30, $31}, [$32 + 0], 32;", "=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,r"(i32 %113) #6, !dbg !49 + %782 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 0, !dbg !49 + %783 = bitcast i32 %782 to float, !dbg !49 + %784 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 1, !dbg !49 + %785 = bitcast i32 %784 to float, !dbg !49 + %786 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 2, !dbg !49 + %787 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 3, !dbg !49 + %788 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 4, !dbg !49 + %789 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 5, !dbg !49 + %790 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 6, !dbg !49 + %791 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 7, !dbg !49 + %792 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 8, !dbg !49 + %793 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 9, !dbg !49 + %794 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 10, !dbg !49 + %795 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 11, !dbg !49 + %796 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 12, !dbg !49 + %797 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 13, !dbg !49 + %798 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 14, !dbg !49 + %799 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 15, !dbg !49 + %800 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 16, !dbg !49 + %801 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 17, !dbg !49 + %802 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 18, !dbg !49 + %803 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 19, !dbg !49 + %804 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 20, !dbg !49 + %805 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 21, !dbg !49 + %806 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 22, !dbg !49 + %807 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 23, !dbg !49 + %808 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 24, !dbg !49 + %809 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 25, !dbg !49 + %810 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 26, !dbg !49 + %811 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 27, !dbg !49 + %812 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 28, !dbg !49 + %813 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 29, !dbg !49 + %814 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 30, !dbg !49 + %815 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %781, 31, !dbg !49 + tail call void @llvm.nvvm.tcgen05.wait.ld(), !dbg !49 + %816 = fmul float %.0.i50, %783, !dbg !74 + %817 = fmul float %.0.i53, %785, !dbg !74 + %818 = fptrunc float %816 to bfloat, !dbg !75 + %819 = select i1 %750, bfloat %818, bfloat 0xR0000, !dbg !76 + %820 = fptrunc float %817 to bfloat, !dbg !75 + %821 = select i1 %752, bfloat %820, bfloat 0xR0000, !dbg !76 + %822 = and i1 %692, %405, !dbg !70 + %823 = and i1 %691, %402, !dbg !70 + %824 = and i1 %92, %822, !dbg !70 + %825 = and i1 %92, %823, !dbg !70 + %826 = insertelement <2 x i32> poison, i32 %786, i64 0, !dbg !49 + %827 = insertelement <2 x i32> %826, i32 %787, i64 1, !dbg !49 + %828 = bitcast <2 x i32> %827 to <2 x float>, !dbg !49 + %829 = insertelement <2 x float> poison, float %.0.i56, i64 0, !dbg !74 + %830 = insertelement <2 x float> %829, float %.0.i59, i64 1, !dbg !74 + %831 = fmul <2 x float> %830, %828, !dbg !74 + %832 = fptrunc <2 x float> %831 to <2 x bfloat>, !dbg !75 + %833 = insertelement <2 x i1> poison, i1 %825, i64 0, !dbg !76 + %834 = insertelement <2 x i1> %833, i1 %824, i64 1, !dbg !76 + %835 = select <2 x i1> %834, <2 x bfloat> %832, <2 x bfloat> zeroinitializer, !dbg !76 + %836 = and i1 %696, %411, !dbg !70 + %837 = and i1 %694, %408, !dbg !70 + %838 = and i1 %92, %836, !dbg !70 + %839 = and i1 %92, %837, !dbg !70 + %840 = insertelement <2 x i32> poison, i32 %788, i64 0, !dbg !49 + %841 = insertelement <2 x i32> %840, i32 %789, i64 1, !dbg !49 + %842 = bitcast <2 x i32> %841 to <2 x float>, !dbg !49 + %843 = insertelement <2 x float> poison, float %.0.i62, i64 0, !dbg !74 + %844 = insertelement <2 x float> %843, float %.0.i65, i64 1, !dbg !74 + %845 = fmul <2 x float> %844, %842, !dbg !74 + %846 = fptrunc <2 x float> %845 to <2 x bfloat>, !dbg !75 + %847 = insertelement <2 x i1> poison, i1 %839, i64 0, !dbg !76 + %848 = insertelement <2 x i1> %847, i1 %838, i64 1, !dbg !76 + %849 = select <2 x i1> %848, <2 x bfloat> %846, <2 x bfloat> zeroinitializer, !dbg !76 + %850 = and i1 %700, %417, !dbg !70 + %851 = and i1 %698, %414, !dbg !70 + %852 = and i1 %92, %850, !dbg !70 + %853 = and i1 %92, %851, !dbg !70 + %854 = insertelement <2 x i32> poison, i32 %790, i64 0, !dbg !49 + %855 = insertelement <2 x i32> %854, i32 %791, i64 1, !dbg !49 + %856 = bitcast <2 x i32> %855 to <2 x float>, !dbg !49 + %857 = insertelement <2 x float> poison, float %.0.i68, i64 0, !dbg !74 + %858 = insertelement <2 x float> %857, float %.0.i71, i64 1, !dbg !74 + %859 = fmul <2 x float> %858, %856, !dbg !74 + %860 = fptrunc <2 x float> %859 to <2 x bfloat>, !dbg !75 + %861 = insertelement <2 x i1> poison, i1 %853, i64 0, !dbg !76 + %862 = insertelement <2 x i1> %861, i1 %852, i64 1, !dbg !76 + %863 = select <2 x i1> %862, <2 x bfloat> %860, <2 x bfloat> zeroinitializer, !dbg !76 + %864 = and i1 %704, %423, !dbg !70 + %865 = and i1 %702, %420, !dbg !70 + %866 = and i1 %92, %864, !dbg !70 + %867 = and i1 %92, %865, !dbg !70 + %868 = insertelement <2 x i32> poison, i32 %792, i64 0, !dbg !49 + %869 = insertelement <2 x i32> %868, i32 %793, i64 1, !dbg !49 + %870 = bitcast <2 x i32> %869 to <2 x float>, !dbg !49 + %871 = insertelement <2 x float> poison, float %.0.i74, i64 0, !dbg !74 + %872 = insertelement <2 x float> %871, float %.0.i77, i64 1, !dbg !74 + %873 = fmul <2 x float> %872, %870, !dbg !74 + %874 = fptrunc <2 x float> %873 to <2 x bfloat>, !dbg !75 + %875 = insertelement <2 x i1> poison, i1 %867, i64 0, !dbg !76 + %876 = insertelement <2 x i1> %875, i1 %866, i64 1, !dbg !76 + %877 = select <2 x i1> %876, <2 x bfloat> %874, <2 x bfloat> zeroinitializer, !dbg !76 + %878 = and i1 %708, %429, !dbg !70 + %879 = and i1 %706, %426, !dbg !70 + %880 = and i1 %92, %878, !dbg !70 + %881 = and i1 %92, %879, !dbg !70 + %882 = insertelement <2 x i32> poison, i32 %794, i64 0, !dbg !49 + %883 = insertelement <2 x i32> %882, i32 %795, i64 1, !dbg !49 + %884 = bitcast <2 x i32> %883 to <2 x float>, !dbg !49 + %885 = insertelement <2 x float> poison, float %.0.i80, i64 0, !dbg !74 + %886 = insertelement <2 x float> %885, float %.0.i83, i64 1, !dbg !74 + %887 = fmul <2 x float> %886, %884, !dbg !74 + %888 = fptrunc <2 x float> %887 to <2 x bfloat>, !dbg !75 + %889 = insertelement <2 x i1> poison, i1 %881, i64 0, !dbg !76 + %890 = insertelement <2 x i1> %889, i1 %880, i64 1, !dbg !76 + %891 = select <2 x i1> %890, <2 x bfloat> %888, <2 x bfloat> zeroinitializer, !dbg !76 + %892 = and i1 %712, %435, !dbg !70 + %893 = and i1 %710, %432, !dbg !70 + %894 = and i1 %92, %892, !dbg !70 + %895 = and i1 %92, %893, !dbg !70 + %896 = insertelement <2 x i32> poison, i32 %796, i64 0, !dbg !49 + %897 = insertelement <2 x i32> %896, i32 %797, i64 1, !dbg !49 + %898 = bitcast <2 x i32> %897 to <2 x float>, !dbg !49 + %899 = insertelement <2 x float> poison, float %.0.i86, i64 0, !dbg !74 + %900 = insertelement <2 x float> %899, float %.0.i89, i64 1, !dbg !74 + %901 = fmul <2 x float> %900, %898, !dbg !74 + %902 = fptrunc <2 x float> %901 to <2 x bfloat>, !dbg !75 + %903 = insertelement <2 x i1> poison, i1 %895, i64 0, !dbg !76 + %904 = insertelement <2 x i1> %903, i1 %894, i64 1, !dbg !76 + %905 = select <2 x i1> %904, <2 x bfloat> %902, <2 x bfloat> zeroinitializer, !dbg !76 + %906 = and i1 %716, %441, !dbg !70 + %907 = and i1 %714, %438, !dbg !70 + %908 = and i1 %92, %906, !dbg !70 + %909 = and i1 %92, %907, !dbg !70 + %910 = insertelement <2 x i32> poison, i32 %798, i64 0, !dbg !49 + %911 = insertelement <2 x i32> %910, i32 %799, i64 1, !dbg !49 + %912 = bitcast <2 x i32> %911 to <2 x float>, !dbg !49 + %913 = insertelement <2 x float> poison, float %.0.i92, i64 0, !dbg !74 + %914 = insertelement <2 x float> %913, float %.0.i95, i64 1, !dbg !74 + %915 = fmul <2 x float> %914, %912, !dbg !74 + %916 = fptrunc <2 x float> %915 to <2 x bfloat>, !dbg !75 + %917 = insertelement <2 x i1> poison, i1 %909, i64 0, !dbg !76 + %918 = insertelement <2 x i1> %917, i1 %908, i64 1, !dbg !76 + %919 = select <2 x i1> %918, <2 x bfloat> %916, <2 x bfloat> zeroinitializer, !dbg !76 + %920 = and i1 %720, %447, !dbg !70 + %921 = and i1 %718, %444, !dbg !70 + %922 = and i1 %92, %920, !dbg !70 + %923 = and i1 %92, %921, !dbg !70 + %924 = insertelement <2 x i32> poison, i32 %800, i64 0, !dbg !49 + %925 = insertelement <2 x i32> %924, i32 %801, i64 1, !dbg !49 + %926 = bitcast <2 x i32> %925 to <2 x float>, !dbg !49 + %927 = insertelement <2 x float> poison, float %.0.i98, i64 0, !dbg !74 + %928 = insertelement <2 x float> %927, float %.0.i101, i64 1, !dbg !74 + %929 = fmul <2 x float> %928, %926, !dbg !74 + %930 = fptrunc <2 x float> %929 to <2 x bfloat>, !dbg !75 + %931 = insertelement <2 x i1> poison, i1 %923, i64 0, !dbg !76 + %932 = insertelement <2 x i1> %931, i1 %922, i64 1, !dbg !76 + %933 = select <2 x i1> %932, <2 x bfloat> %930, <2 x bfloat> zeroinitializer, !dbg !76 + %934 = and i1 %724, %453, !dbg !70 + %935 = and i1 %722, %450, !dbg !70 + %936 = and i1 %92, %934, !dbg !70 + %937 = and i1 %92, %935, !dbg !70 + %938 = insertelement <2 x i32> poison, i32 %802, i64 0, !dbg !49 + %939 = insertelement <2 x i32> %938, i32 %803, i64 1, !dbg !49 + %940 = bitcast <2 x i32> %939 to <2 x float>, !dbg !49 + %941 = insertelement <2 x float> poison, float %.0.i104, i64 0, !dbg !74 + %942 = insertelement <2 x float> %941, float %.0.i107, i64 1, !dbg !74 + %943 = fmul <2 x float> %942, %940, !dbg !74 + %944 = fptrunc <2 x float> %943 to <2 x bfloat>, !dbg !75 + %945 = insertelement <2 x i1> poison, i1 %937, i64 0, !dbg !76 + %946 = insertelement <2 x i1> %945, i1 %936, i64 1, !dbg !76 + %947 = select <2 x i1> %946, <2 x bfloat> %944, <2 x bfloat> zeroinitializer, !dbg !76 + %948 = and i1 %728, %459, !dbg !70 + %949 = and i1 %726, %456, !dbg !70 + %950 = and i1 %92, %948, !dbg !70 + %951 = and i1 %92, %949, !dbg !70 + %952 = insertelement <2 x i32> poison, i32 %804, i64 0, !dbg !49 + %953 = insertelement <2 x i32> %952, i32 %805, i64 1, !dbg !49 + %954 = bitcast <2 x i32> %953 to <2 x float>, !dbg !49 + %955 = insertelement <2 x float> poison, float %.0.i110, i64 0, !dbg !74 + %956 = insertelement <2 x float> %955, float %.0.i113, i64 1, !dbg !74 + %957 = fmul <2 x float> %956, %954, !dbg !74 + %958 = fptrunc <2 x float> %957 to <2 x bfloat>, !dbg !75 + %959 = insertelement <2 x i1> poison, i1 %951, i64 0, !dbg !76 + %960 = insertelement <2 x i1> %959, i1 %950, i64 1, !dbg !76 + %961 = select <2 x i1> %960, <2 x bfloat> %958, <2 x bfloat> zeroinitializer, !dbg !76 + %962 = and i1 %732, %465, !dbg !70 + %963 = and i1 %730, %462, !dbg !70 + %964 = and i1 %92, %962, !dbg !70 + %965 = and i1 %92, %963, !dbg !70 + %966 = insertelement <2 x i32> poison, i32 %806, i64 0, !dbg !49 + %967 = insertelement <2 x i32> %966, i32 %807, i64 1, !dbg !49 + %968 = bitcast <2 x i32> %967 to <2 x float>, !dbg !49 + %969 = insertelement <2 x float> poison, float %.0.i116, i64 0, !dbg !74 + %970 = insertelement <2 x float> %969, float %.0.i119, i64 1, !dbg !74 + %971 = fmul <2 x float> %970, %968, !dbg !74 + %972 = fptrunc <2 x float> %971 to <2 x bfloat>, !dbg !75 + %973 = insertelement <2 x i1> poison, i1 %965, i64 0, !dbg !76 + %974 = insertelement <2 x i1> %973, i1 %964, i64 1, !dbg !76 + %975 = select <2 x i1> %974, <2 x bfloat> %972, <2 x bfloat> zeroinitializer, !dbg !76 + %976 = and i1 %736, %471, !dbg !70 + %977 = and i1 %734, %468, !dbg !70 + %978 = and i1 %92, %976, !dbg !70 + %979 = and i1 %92, %977, !dbg !70 + %980 = insertelement <2 x i32> poison, i32 %808, i64 0, !dbg !49 + %981 = insertelement <2 x i32> %980, i32 %809, i64 1, !dbg !49 + %982 = bitcast <2 x i32> %981 to <2 x float>, !dbg !49 + %983 = insertelement <2 x float> poison, float %.0.i122, i64 0, !dbg !74 + %984 = insertelement <2 x float> %983, float %.0.i125, i64 1, !dbg !74 + %985 = fmul <2 x float> %984, %982, !dbg !74 + %986 = fptrunc <2 x float> %985 to <2 x bfloat>, !dbg !75 + %987 = insertelement <2 x i1> poison, i1 %979, i64 0, !dbg !76 + %988 = insertelement <2 x i1> %987, i1 %978, i64 1, !dbg !76 + %989 = select <2 x i1> %988, <2 x bfloat> %986, <2 x bfloat> zeroinitializer, !dbg !76 + %990 = and i1 %740, %477, !dbg !70 + %991 = and i1 %738, %474, !dbg !70 + %992 = and i1 %92, %990, !dbg !70 + %993 = and i1 %92, %991, !dbg !70 + %994 = insertelement <2 x i32> poison, i32 %810, i64 0, !dbg !49 + %995 = insertelement <2 x i32> %994, i32 %811, i64 1, !dbg !49 + %996 = bitcast <2 x i32> %995 to <2 x float>, !dbg !49 + %997 = insertelement <2 x float> poison, float %.0.i128, i64 0, !dbg !74 + %998 = insertelement <2 x float> %997, float %.0.i131, i64 1, !dbg !74 + %999 = fmul <2 x float> %998, %996, !dbg !74 + %1000 = fptrunc <2 x float> %999 to <2 x bfloat>, !dbg !75 + %1001 = insertelement <2 x i1> poison, i1 %993, i64 0, !dbg !76 + %1002 = insertelement <2 x i1> %1001, i1 %992, i64 1, !dbg !76 + %1003 = select <2 x i1> %1002, <2 x bfloat> %1000, <2 x bfloat> zeroinitializer, !dbg !76 + %1004 = and i1 %744, %483, !dbg !70 + %1005 = and i1 %742, %480, !dbg !70 + %1006 = and i1 %92, %1004, !dbg !70 + %1007 = and i1 %92, %1005, !dbg !70 + %1008 = insertelement <2 x i32> poison, i32 %812, i64 0, !dbg !49 + %1009 = insertelement <2 x i32> %1008, i32 %813, i64 1, !dbg !49 + %1010 = bitcast <2 x i32> %1009 to <2 x float>, !dbg !49 + %1011 = insertelement <2 x float> poison, float %.0.i134, i64 0, !dbg !74 + %1012 = insertelement <2 x float> %1011, float %.0.i137, i64 1, !dbg !74 + %1013 = fmul <2 x float> %1012, %1010, !dbg !74 + %1014 = fptrunc <2 x float> %1013 to <2 x bfloat>, !dbg !75 + %1015 = insertelement <2 x i1> poison, i1 %1007, i64 0, !dbg !76 + %1016 = insertelement <2 x i1> %1015, i1 %1006, i64 1, !dbg !76 + %1017 = select <2 x i1> %1016, <2 x bfloat> %1014, <2 x bfloat> zeroinitializer, !dbg !76 + %1018 = and i1 %748, %489, !dbg !70 + %1019 = and i1 %746, %486, !dbg !70 + %1020 = and i1 %92, %1018, !dbg !70 + %1021 = and i1 %92, %1019, !dbg !70 + %1022 = insertelement <2 x i32> poison, i32 %814, i64 0, !dbg !49 + %1023 = insertelement <2 x i32> %1022, i32 %815, i64 1, !dbg !49 + %1024 = bitcast <2 x i32> %1023 to <2 x float>, !dbg !49 + %1025 = insertelement <2 x float> poison, float %.0.i140, i64 0, !dbg !74 + %1026 = insertelement <2 x float> %1025, float %.0.i143, i64 1, !dbg !74 + %1027 = fmul <2 x float> %1026, %1024, !dbg !74 + %1028 = fptrunc <2 x float> %1027 to <2 x bfloat>, !dbg !75 + %1029 = insertelement <2 x i1> poison, i1 %1021, i64 0, !dbg !76 + %1030 = insertelement <2 x i1> %1029, i1 %1020, i64 1, !dbg !76 + %1031 = select <2 x i1> %1030, <2 x bfloat> %1028, <2 x bfloat> zeroinitializer, !dbg !76 + tail call void asm sideeffect "@$0 tcgen05.st.sync.aligned.16x32bx2.x32.b32 [$1 + 0], 32, {$2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13, $14, $15, $16, $17, $18, $19, $20, $21, $22, $23, $24, $25, $26, $27, $28, $29, $30, $31, $32, $33};", "b,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r"(i1 true, i32 %113, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00) #6, !dbg !77 + tail call void @llvm.nvvm.tcgen05.wait.st(), !dbg !77 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !77 + %1032 = add i32 %13, 1048640, !dbg !77 + %1033 = insertelement <2 x bfloat> poison, bfloat %819, i64 0, !dbg !77 + %1034 = insertelement <2 x bfloat> %1033, bfloat %821, i64 1, !dbg !77 + %1035 = bitcast <2 x bfloat> %1034 to i32, !dbg !77 + %1036 = bitcast <2 x bfloat> %835 to i32, !dbg !77 + %1037 = bitcast <2 x bfloat> %849 to i32, !dbg !77 + %1038 = bitcast <2 x bfloat> %863 to i32, !dbg !77 + %1039 = bitcast <2 x bfloat> %877 to i32, !dbg !77 + %1040 = bitcast <2 x bfloat> %891 to i32, !dbg !77 + %1041 = bitcast <2 x bfloat> %905 to i32, !dbg !77 + %1042 = bitcast <2 x bfloat> %919 to i32, !dbg !77 + %1043 = bitcast <2 x bfloat> %933 to i32, !dbg !77 + %1044 = bitcast <2 x bfloat> %947 to i32, !dbg !77 + %1045 = bitcast <2 x bfloat> %961 to i32, !dbg !77 + %1046 = bitcast <2 x bfloat> %975 to i32, !dbg !77 + %1047 = bitcast <2 x bfloat> %989 to i32, !dbg !77 + %1048 = bitcast <2 x bfloat> %1003 to i32, !dbg !77 + %1049 = bitcast <2 x bfloat> %1017 to i32, !dbg !77 + %1050 = bitcast <2 x bfloat> %1031 to i32, !dbg !77 + %1051 = add i32 %110, %1032, !dbg !77 + tail call void asm sideeffect "@$0 tcgen05.st.sync.aligned.16x32bx2.x16.b32 [$1 + 0], 16, {$2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13, $14, $15, $16, $17};", "b,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r,r"(i1 true, i32 %1051, i32 %1035, i32 %1036, i32 %1037, i32 %1038, i32 %1039, i32 %1040, i32 %1041, i32 %1042, i32 %1043, i32 %1044, i32 %1045, i32 %1046, i32 %1047, i32 %1048, i32 %1049, i32 %1050) #6, !dbg !77 + tail call void @llvm.nvvm.tcgen05.wait.st(), !dbg !77 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !77 + tail call void @llvm.nvvm.fence.proxy.async.shared_cta(), !dbg !77 + %1052 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !77 + %1053 = extractvalue { i32, i1 } %1052, 1, !dbg !77 + %1054 = and i1 %114, %1053, !dbg !77 + tail call void asm sideeffect "@$0 mbarrier.init.shared::cta.b64 [$1], 1;", "b,r"(i1 %1054, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 8192)) #6, !dbg !77 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !77 + br i1 %210, label %1055, label %1065, !dbg !77 + +1055: ; preds = %__nv_exp2f.exit144 + %1056 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !77 + %1057 = extractvalue { i32, i1 } %1056, 1, !dbg !77 + %1058 = lshr exact i32 ptrtoint (ptr addrspace(3) @global_smem to i32), 4, !dbg !77 + %1059 = and i32 %1058, 16383, !dbg !77 + %1060 = zext nneg i32 %1059 to i64, !dbg !77 + %1061 = or disjoint i64 %1060, 4611756662049472512, !dbg !77 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], [ $1 + 0 ], $2, $3, $4;", "r,r,l,r,b,b"(i32 %112, i32 %1032, i64 %1061, i32 68224144, i1 true, i1 %1057) #6, !dbg !77 + %1062 = add nuw nsw i64 %1060, 4611756662049472640, !dbg !77 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], [ $1 + 8 ], $2, $3, $4;", "r,r,l,r,b,b"(i32 %112, i32 %1032, i64 %1062, i32 68224144, i1 true, i1 %1057) #6, !dbg !77 + %1063 = add nuw nsw i64 %1060, 4611756662049472768, !dbg !77 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], [ $1 + 16 ], $2, $3, $4;", "r,r,l,r,b,b"(i32 %112, i32 %1032, i64 %1063, i32 68224144, i1 true, i1 %1057) #6, !dbg !77 + %1064 = add nuw nsw i64 %1060, 4611756662049472896, !dbg !77 + tail call void asm sideeffect "@$5 tcgen05.mma.cta_group::1.kind::f16 [ $0 + 0 ], [ $1 + 24 ], $2, $3, $4;", "r,r,l,r,b,b"(i32 %112, i32 %1032, i64 %1064, i32 68224144, i1 true, i1 %1057) #6, !dbg !77 + tail call void asm sideeffect "@$0 tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [$1];", "b,r"(i1 %1057, i32 ptrtoint (ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 8192) to i32)) #6, !dbg !77 + br label %1065, !dbg !77 + +1065: ; preds = %1055, %__nv_exp2f.exit144 + %1066 = getelementptr bfloat, ptr addrspace(1) %5, i64 %30, !dbg !78 + %1067 = getelementptr bfloat, ptr addrspace(1) %1066, i64 %103, !dbg !79 + %1068 = getelementptr bfloat, ptr addrspace(1) %1067, i64 %99, !dbg !80 + %1069 = getelementptr bfloat, ptr addrspace(1) %1066, i64 %102, !dbg !79 + %1070 = getelementptr bfloat, ptr addrspace(1) %1069, i64 %99, !dbg !80 + %1071 = getelementptr bfloat, ptr addrspace(1) %1066, i64 %101, !dbg !79 + %1072 = getelementptr bfloat, ptr addrspace(1) %1071, i64 %99, !dbg !80 + %1073 = getelementptr bfloat, ptr addrspace(1) %1066, i64 %100, !dbg !79 + %1074 = getelementptr bfloat, ptr addrspace(1) %1073, i64 %99, !dbg !80 + %1075 = inttoptr i32 %13 to ptr addrspace(6), !dbg !13 + tail call void asm sideeffect "\0A{\0A\09.reg .pred complete;\0A\09waitLoop:\0A\09mbarrier.try_wait.parity.shared::cta.b64 complete, [$0], $1;\0A\09@!complete bra.uni waitLoop;\0A}\0A", "r,r"(ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 8192), i32 0) #6, !dbg !77 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !77 + %1076 = tail call { i32, i1 } @llvm.nvvm.elect.sync(i32 -1), !dbg !77 + %1077 = extractvalue { i32, i1 } %1076, 1, !dbg !77 + %1078 = and i1 %114, %1077, !dbg !77 + tail call void asm sideeffect "@$0 mbarrier.inval.shared::cta.b64 [$1];", "b,r"(i1 %1078, ptr addrspace(3) getelementptr (i8, ptr addrspace(3) @global_smem, i32 8192)) #6, !dbg !77 + %1079 = tail call { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } asm sideeffect "tcgen05.ld.sync.aligned.16x32bx2.x32.b32 {$0, $1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13, $14, $15, $16, $17, $18, $19, $20, $21, $22, $23, $24, $25, $26, $27, $28, $29, $30, $31}, [$32 + 0], 32;", "=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,r"(i32 %111) #6, !dbg !48 + %1080 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 0, !dbg !48 + %1081 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 1, !dbg !48 + %1082 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 2, !dbg !48 + %1083 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 3, !dbg !48 + %1084 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 4, !dbg !48 + %1085 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 5, !dbg !48 + %1086 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 6, !dbg !48 + %1087 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 7, !dbg !48 + %1088 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 8, !dbg !48 + %1089 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 9, !dbg !48 + %1090 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 10, !dbg !48 + %1091 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 11, !dbg !48 + %1092 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 12, !dbg !48 + %1093 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 13, !dbg !48 + %1094 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 14, !dbg !48 + %1095 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 15, !dbg !48 + %1096 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 16, !dbg !48 + %1097 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 17, !dbg !48 + %1098 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 18, !dbg !48 + %1099 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 19, !dbg !48 + %1100 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 20, !dbg !48 + %1101 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 21, !dbg !48 + %1102 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 22, !dbg !48 + %1103 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 23, !dbg !48 + %1104 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 24, !dbg !48 + %1105 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 25, !dbg !48 + %1106 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 26, !dbg !48 + %1107 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 27, !dbg !48 + %1108 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 28, !dbg !48 + %1109 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 29, !dbg !48 + %1110 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 30, !dbg !48 + %1111 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1079, 31, !dbg !48 + tail call void @llvm.nvvm.tcgen05.wait.ld(), !dbg !48 + %1112 = tail call { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } asm sideeffect "tcgen05.ld.sync.aligned.16x32bx2.x32.b32 {$0, $1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13, $14, $15, $16, $17, $18, $19, $20, $21, $22, $23, $24, $25, $26, $27, $28, $29, $30, $31}, [$32 + 0], 32;", "=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,=r,r"(i32 %113) #6, !dbg !77 + %1113 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 0, !dbg !77 + %1114 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 1, !dbg !77 + %1115 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 2, !dbg !77 + %1116 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 3, !dbg !77 + %1117 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 4, !dbg !77 + %1118 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 5, !dbg !77 + %1119 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 6, !dbg !77 + %1120 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 7, !dbg !77 + %1121 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 8, !dbg !77 + %1122 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 9, !dbg !77 + %1123 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 10, !dbg !77 + %1124 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 11, !dbg !77 + %1125 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 12, !dbg !77 + %1126 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 13, !dbg !77 + %1127 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 14, !dbg !77 + %1128 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 15, !dbg !77 + %1129 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 16, !dbg !77 + %1130 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 17, !dbg !77 + %1131 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 18, !dbg !77 + %1132 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 19, !dbg !77 + %1133 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 20, !dbg !77 + %1134 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 21, !dbg !77 + %1135 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 22, !dbg !77 + %1136 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 23, !dbg !77 + %1137 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 24, !dbg !77 + %1138 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 25, !dbg !77 + %1139 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 26, !dbg !77 + %1140 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 27, !dbg !77 + %1141 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 28, !dbg !77 + %1142 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 29, !dbg !77 + %1143 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 30, !dbg !77 + %1144 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 } %1112, 31, !dbg !77 + tail call void @llvm.nvvm.tcgen05.wait.ld(), !dbg !77 + %1145 = insertelement <2 x i32> poison, i32 %1080, i64 0, !dbg !48 + %1146 = insertelement <2 x i32> %1145, i32 %1081, i64 1, !dbg !48 + %1147 = bitcast <2 x i32> %1146 to <2 x float>, !dbg !48 + %1148 = insertelement <2 x float> poison, float %.0.i, i64 0, !dbg !81 + %1149 = shufflevector <2 x float> %1148, <2 x float> poison, <2 x i32> zeroinitializer, !dbg !81 + %1150 = fmul <2 x float> %1149, %1147, !dbg !81 + %1151 = insertelement <2 x float> poison, float %6, i64 0, !dbg !82 + %1152 = shufflevector <2 x float> %1151, <2 x float> poison, <2 x i32> zeroinitializer, !dbg !82 + %1153 = fmul <2 x float> %1152, %1150, !dbg !82 + %1154 = insertelement <2 x i32> poison, i32 %1113, i64 0, !dbg !77 + %1155 = insertelement <2 x i32> %1154, i32 %1114, i64 1, !dbg !77 + %1156 = bitcast <2 x i32> %1155 to <2 x float>, !dbg !77 + %1157 = fmul <2 x float> %1152, %1156, !dbg !83 + %1158 = fadd <2 x float> %1153, %1157, !dbg !84 + %1159 = fptrunc <2 x float> %1158 to <2 x bfloat>, !dbg !85 + %1160 = insertelement <2 x i32> poison, i32 %1082, i64 0, !dbg !48 + %1161 = insertelement <2 x i32> %1160, i32 %1083, i64 1, !dbg !48 + %1162 = bitcast <2 x i32> %1161 to <2 x float>, !dbg !48 + %1163 = fmul <2 x float> %1149, %1162, !dbg !81 + %1164 = fmul <2 x float> %1152, %1163, !dbg !82 + %1165 = insertelement <2 x i32> poison, i32 %1115, i64 0, !dbg !77 + %1166 = insertelement <2 x i32> %1165, i32 %1116, i64 1, !dbg !77 + %1167 = bitcast <2 x i32> %1166 to <2 x float>, !dbg !77 + %1168 = fmul <2 x float> %1152, %1167, !dbg !83 + %1169 = fadd <2 x float> %1164, %1168, !dbg !84 + %1170 = fptrunc <2 x float> %1169 to <2 x bfloat>, !dbg !85 + %1171 = insertelement <2 x i32> poison, i32 %1084, i64 0, !dbg !48 + %1172 = insertelement <2 x i32> %1171, i32 %1085, i64 1, !dbg !48 + %1173 = bitcast <2 x i32> %1172 to <2 x float>, !dbg !48 + %1174 = fmul <2 x float> %1149, %1173, !dbg !81 + %1175 = fmul <2 x float> %1152, %1174, !dbg !82 + %1176 = insertelement <2 x i32> poison, i32 %1117, i64 0, !dbg !77 + %1177 = insertelement <2 x i32> %1176, i32 %1118, i64 1, !dbg !77 + %1178 = bitcast <2 x i32> %1177 to <2 x float>, !dbg !77 + %1179 = fmul <2 x float> %1152, %1178, !dbg !83 + %1180 = fadd <2 x float> %1175, %1179, !dbg !84 + %1181 = fptrunc <2 x float> %1180 to <2 x bfloat>, !dbg !85 + %1182 = insertelement <2 x i32> poison, i32 %1086, i64 0, !dbg !48 + %1183 = insertelement <2 x i32> %1182, i32 %1087, i64 1, !dbg !48 + %1184 = bitcast <2 x i32> %1183 to <2 x float>, !dbg !48 + %1185 = fmul <2 x float> %1149, %1184, !dbg !81 + %1186 = fmul <2 x float> %1152, %1185, !dbg !82 + %1187 = insertelement <2 x i32> poison, i32 %1119, i64 0, !dbg !77 + %1188 = insertelement <2 x i32> %1187, i32 %1120, i64 1, !dbg !77 + %1189 = bitcast <2 x i32> %1188 to <2 x float>, !dbg !77 + %1190 = fmul <2 x float> %1152, %1189, !dbg !83 + %1191 = fadd <2 x float> %1186, %1190, !dbg !84 + %1192 = fptrunc <2 x float> %1191 to <2 x bfloat>, !dbg !85 + %1193 = insertelement <2 x i32> poison, i32 %1088, i64 0, !dbg !48 + %1194 = insertelement <2 x i32> %1193, i32 %1089, i64 1, !dbg !48 + %1195 = bitcast <2 x i32> %1194 to <2 x float>, !dbg !48 + %1196 = fmul <2 x float> %1149, %1195, !dbg !81 + %1197 = fmul <2 x float> %1152, %1196, !dbg !82 + %1198 = insertelement <2 x i32> poison, i32 %1121, i64 0, !dbg !77 + %1199 = insertelement <2 x i32> %1198, i32 %1122, i64 1, !dbg !77 + %1200 = bitcast <2 x i32> %1199 to <2 x float>, !dbg !77 + %1201 = fmul <2 x float> %1152, %1200, !dbg !83 + %1202 = fadd <2 x float> %1197, %1201, !dbg !84 + %1203 = fptrunc <2 x float> %1202 to <2 x bfloat>, !dbg !85 + %1204 = insertelement <2 x i32> poison, i32 %1090, i64 0, !dbg !48 + %1205 = insertelement <2 x i32> %1204, i32 %1091, i64 1, !dbg !48 + %1206 = bitcast <2 x i32> %1205 to <2 x float>, !dbg !48 + %1207 = fmul <2 x float> %1149, %1206, !dbg !81 + %1208 = fmul <2 x float> %1152, %1207, !dbg !82 + %1209 = insertelement <2 x i32> poison, i32 %1123, i64 0, !dbg !77 + %1210 = insertelement <2 x i32> %1209, i32 %1124, i64 1, !dbg !77 + %1211 = bitcast <2 x i32> %1210 to <2 x float>, !dbg !77 + %1212 = fmul <2 x float> %1152, %1211, !dbg !83 + %1213 = fadd <2 x float> %1208, %1212, !dbg !84 + %1214 = fptrunc <2 x float> %1213 to <2 x bfloat>, !dbg !85 + %1215 = insertelement <2 x i32> poison, i32 %1092, i64 0, !dbg !48 + %1216 = insertelement <2 x i32> %1215, i32 %1093, i64 1, !dbg !48 + %1217 = bitcast <2 x i32> %1216 to <2 x float>, !dbg !48 + %1218 = fmul <2 x float> %1149, %1217, !dbg !81 + %1219 = fmul <2 x float> %1152, %1218, !dbg !82 + %1220 = insertelement <2 x i32> poison, i32 %1125, i64 0, !dbg !77 + %1221 = insertelement <2 x i32> %1220, i32 %1126, i64 1, !dbg !77 + %1222 = bitcast <2 x i32> %1221 to <2 x float>, !dbg !77 + %1223 = fmul <2 x float> %1152, %1222, !dbg !83 + %1224 = fadd <2 x float> %1219, %1223, !dbg !84 + %1225 = fptrunc <2 x float> %1224 to <2 x bfloat>, !dbg !85 + %1226 = insertelement <2 x i32> poison, i32 %1094, i64 0, !dbg !48 + %1227 = insertelement <2 x i32> %1226, i32 %1095, i64 1, !dbg !48 + %1228 = bitcast <2 x i32> %1227 to <2 x float>, !dbg !48 + %1229 = fmul <2 x float> %1149, %1228, !dbg !81 + %1230 = fmul <2 x float> %1152, %1229, !dbg !82 + %1231 = insertelement <2 x i32> poison, i32 %1127, i64 0, !dbg !77 + %1232 = insertelement <2 x i32> %1231, i32 %1128, i64 1, !dbg !77 + %1233 = bitcast <2 x i32> %1232 to <2 x float>, !dbg !77 + %1234 = fmul <2 x float> %1152, %1233, !dbg !83 + %1235 = fadd <2 x float> %1230, %1234, !dbg !84 + %1236 = fptrunc <2 x float> %1235 to <2 x bfloat>, !dbg !85 + %1237 = insertelement <2 x i32> poison, i32 %1096, i64 0, !dbg !48 + %1238 = insertelement <2 x i32> %1237, i32 %1097, i64 1, !dbg !48 + %1239 = bitcast <2 x i32> %1238 to <2 x float>, !dbg !48 + %1240 = fmul <2 x float> %1149, %1239, !dbg !81 + %1241 = fmul <2 x float> %1152, %1240, !dbg !82 + %1242 = insertelement <2 x i32> poison, i32 %1129, i64 0, !dbg !77 + %1243 = insertelement <2 x i32> %1242, i32 %1130, i64 1, !dbg !77 + %1244 = bitcast <2 x i32> %1243 to <2 x float>, !dbg !77 + %1245 = fmul <2 x float> %1152, %1244, !dbg !83 + %1246 = fadd <2 x float> %1241, %1245, !dbg !84 + %1247 = fptrunc <2 x float> %1246 to <2 x bfloat>, !dbg !85 + %1248 = insertelement <2 x i32> poison, i32 %1098, i64 0, !dbg !48 + %1249 = insertelement <2 x i32> %1248, i32 %1099, i64 1, !dbg !48 + %1250 = bitcast <2 x i32> %1249 to <2 x float>, !dbg !48 + %1251 = fmul <2 x float> %1149, %1250, !dbg !81 + %1252 = fmul <2 x float> %1152, %1251, !dbg !82 + %1253 = insertelement <2 x i32> poison, i32 %1131, i64 0, !dbg !77 + %1254 = insertelement <2 x i32> %1253, i32 %1132, i64 1, !dbg !77 + %1255 = bitcast <2 x i32> %1254 to <2 x float>, !dbg !77 + %1256 = fmul <2 x float> %1152, %1255, !dbg !83 + %1257 = fadd <2 x float> %1252, %1256, !dbg !84 + %1258 = fptrunc <2 x float> %1257 to <2 x bfloat>, !dbg !85 + %1259 = insertelement <2 x i32> poison, i32 %1100, i64 0, !dbg !48 + %1260 = insertelement <2 x i32> %1259, i32 %1101, i64 1, !dbg !48 + %1261 = bitcast <2 x i32> %1260 to <2 x float>, !dbg !48 + %1262 = fmul <2 x float> %1149, %1261, !dbg !81 + %1263 = fmul <2 x float> %1152, %1262, !dbg !82 + %1264 = insertelement <2 x i32> poison, i32 %1133, i64 0, !dbg !77 + %1265 = insertelement <2 x i32> %1264, i32 %1134, i64 1, !dbg !77 + %1266 = bitcast <2 x i32> %1265 to <2 x float>, !dbg !77 + %1267 = fmul <2 x float> %1152, %1266, !dbg !83 + %1268 = fadd <2 x float> %1263, %1267, !dbg !84 + %1269 = fptrunc <2 x float> %1268 to <2 x bfloat>, !dbg !85 + %1270 = insertelement <2 x i32> poison, i32 %1102, i64 0, !dbg !48 + %1271 = insertelement <2 x i32> %1270, i32 %1103, i64 1, !dbg !48 + %1272 = bitcast <2 x i32> %1271 to <2 x float>, !dbg !48 + %1273 = fmul <2 x float> %1149, %1272, !dbg !81 + %1274 = fmul <2 x float> %1152, %1273, !dbg !82 + %1275 = insertelement <2 x i32> poison, i32 %1135, i64 0, !dbg !77 + %1276 = insertelement <2 x i32> %1275, i32 %1136, i64 1, !dbg !77 + %1277 = bitcast <2 x i32> %1276 to <2 x float>, !dbg !77 + %1278 = fmul <2 x float> %1152, %1277, !dbg !83 + %1279 = fadd <2 x float> %1274, %1278, !dbg !84 + %1280 = fptrunc <2 x float> %1279 to <2 x bfloat>, !dbg !85 + %1281 = insertelement <2 x i32> poison, i32 %1104, i64 0, !dbg !48 + %1282 = insertelement <2 x i32> %1281, i32 %1105, i64 1, !dbg !48 + %1283 = bitcast <2 x i32> %1282 to <2 x float>, !dbg !48 + %1284 = fmul <2 x float> %1149, %1283, !dbg !81 + %1285 = fmul <2 x float> %1152, %1284, !dbg !82 + %1286 = insertelement <2 x i32> poison, i32 %1137, i64 0, !dbg !77 + %1287 = insertelement <2 x i32> %1286, i32 %1138, i64 1, !dbg !77 + %1288 = bitcast <2 x i32> %1287 to <2 x float>, !dbg !77 + %1289 = fmul <2 x float> %1152, %1288, !dbg !83 + %1290 = fadd <2 x float> %1285, %1289, !dbg !84 + %1291 = fptrunc <2 x float> %1290 to <2 x bfloat>, !dbg !85 + %1292 = insertelement <2 x i32> poison, i32 %1106, i64 0, !dbg !48 + %1293 = insertelement <2 x i32> %1292, i32 %1107, i64 1, !dbg !48 + %1294 = bitcast <2 x i32> %1293 to <2 x float>, !dbg !48 + %1295 = fmul <2 x float> %1149, %1294, !dbg !81 + %1296 = fmul <2 x float> %1152, %1295, !dbg !82 + %1297 = insertelement <2 x i32> poison, i32 %1139, i64 0, !dbg !77 + %1298 = insertelement <2 x i32> %1297, i32 %1140, i64 1, !dbg !77 + %1299 = bitcast <2 x i32> %1298 to <2 x float>, !dbg !77 + %1300 = fmul <2 x float> %1152, %1299, !dbg !83 + %1301 = fadd <2 x float> %1296, %1300, !dbg !84 + %1302 = fptrunc <2 x float> %1301 to <2 x bfloat>, !dbg !85 + %1303 = insertelement <2 x i32> poison, i32 %1108, i64 0, !dbg !48 + %1304 = insertelement <2 x i32> %1303, i32 %1109, i64 1, !dbg !48 + %1305 = bitcast <2 x i32> %1304 to <2 x float>, !dbg !48 + %1306 = fmul <2 x float> %1149, %1305, !dbg !81 + %1307 = fmul <2 x float> %1152, %1306, !dbg !82 + %1308 = insertelement <2 x i32> poison, i32 %1141, i64 0, !dbg !77 + %1309 = insertelement <2 x i32> %1308, i32 %1142, i64 1, !dbg !77 + %1310 = bitcast <2 x i32> %1309 to <2 x float>, !dbg !77 + %1311 = fmul <2 x float> %1152, %1310, !dbg !83 + %1312 = fadd <2 x float> %1307, %1311, !dbg !84 + %1313 = fptrunc <2 x float> %1312 to <2 x bfloat>, !dbg !85 + %1314 = insertelement <2 x i32> poison, i32 %1110, i64 0, !dbg !48 + %1315 = insertelement <2 x i32> %1314, i32 %1111, i64 1, !dbg !48 + %1316 = bitcast <2 x i32> %1315 to <2 x float>, !dbg !48 + %1317 = fmul <2 x float> %1149, %1316, !dbg !81 + %1318 = fmul <2 x float> %1152, %1317, !dbg !82 + %1319 = insertelement <2 x i32> poison, i32 %1143, i64 0, !dbg !77 + %1320 = insertelement <2 x i32> %1319, i32 %1144, i64 1, !dbg !77 + %1321 = bitcast <2 x i32> %1320 to <2 x float>, !dbg !77 + %1322 = fmul <2 x float> %1152, %1321, !dbg !83 + %1323 = fadd <2 x float> %1318, %1322, !dbg !84 + %1324 = fptrunc <2 x float> %1323 to <2 x bfloat>, !dbg !85 + %1325 = and i32 %152, 176, !dbg !86 + %1326 = shl nuw nsw i32 %46, 3, !dbg !86 + %1327 = and i32 %11, 4, !dbg !86 + %1328 = icmp eq i32 %1327, 0, !dbg !86 + %1329 = select i1 %1328, i32 0, i32 1088, !dbg !86 + %1330 = shl nuw nsw i32 %49, 2, !dbg !86 + %1331 = xor i32 %1329, %1330, !dbg !86 + %1332 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %1325, !dbg !86 + %1333 = getelementptr inbounds nuw i8, ptr addrspace(3) %1332, i32 %1326, !dbg !86 + %1334 = getelementptr inbounds nuw i8, ptr addrspace(3) %1333, i32 %1331, !dbg !86 + %1335 = bitcast <2 x bfloat> %1159 to i32, !dbg !86 + %1336 = bitcast <2 x bfloat> %1203 to i32, !dbg !86 + %1337 = bitcast <2 x bfloat> %1247 to i32, !dbg !86 + %1338 = bitcast <2 x bfloat> %1291 to i32, !dbg !86 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %1334, i32 %1335, i32 %1336, i32 %1337, i32 %1338) #6, !dbg !86 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !86 + %1339 = shl nuw nsw i32 %154, 5, !dbg !86 + %1340 = and i32 %43, 48, !dbg !86 + %1341 = shl nuw nsw i32 %46, 1, !dbg !86 + %1342 = shl nuw nsw i32 %11, 6, !dbg !86 + %1343 = and i32 %1342, 64, !dbg !86 + %1344 = xor i32 %1341, %1343, !dbg !86 + %1345 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %1339, !dbg !86 + %1346 = getelementptr inbounds nuw i8, ptr addrspace(3) %1345, i32 %1340, !dbg !86 + %1347 = getelementptr inbounds nuw i8, ptr addrspace(3) %1346, i32 %1344, !dbg !86 + %1348 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %1347), !dbg !86 + %1349 = extractvalue { i32, i32, i32, i32 } %1348, 0, !dbg !86 + %1350 = extractvalue { i32, i32, i32, i32 } %1348, 1, !dbg !86 + %1351 = extractvalue { i32, i32, i32, i32 } %1348, 2, !dbg !86 + %1352 = extractvalue { i32, i32, i32, i32 } %1348, 3, !dbg !86 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !86 + %1353 = bitcast <2 x bfloat> %1170 to i32, !dbg !86 + %1354 = bitcast <2 x bfloat> %1214 to i32, !dbg !86 + %1355 = bitcast <2 x bfloat> %1258 to i32, !dbg !86 + %1356 = bitcast <2 x bfloat> %1302 to i32, !dbg !86 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %1334, i32 %1353, i32 %1354, i32 %1355, i32 %1356) #6, !dbg !86 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !86 + %1357 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %1347), !dbg !86 + %1358 = extractvalue { i32, i32, i32, i32 } %1357, 0, !dbg !86 + %1359 = extractvalue { i32, i32, i32, i32 } %1357, 1, !dbg !86 + %1360 = extractvalue { i32, i32, i32, i32 } %1357, 2, !dbg !86 + %1361 = extractvalue { i32, i32, i32, i32 } %1357, 3, !dbg !86 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !86 + %1362 = bitcast <2 x bfloat> %1181 to i32, !dbg !86 + %1363 = bitcast <2 x bfloat> %1225 to i32, !dbg !86 + %1364 = bitcast <2 x bfloat> %1269 to i32, !dbg !86 + %1365 = bitcast <2 x bfloat> %1313 to i32, !dbg !86 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %1334, i32 %1362, i32 %1363, i32 %1364, i32 %1365) #6, !dbg !86 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !86 + %1366 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %1347), !dbg !86 + %1367 = extractvalue { i32, i32, i32, i32 } %1366, 0, !dbg !86 + %1368 = extractvalue { i32, i32, i32, i32 } %1366, 1, !dbg !86 + %1369 = extractvalue { i32, i32, i32, i32 } %1366, 2, !dbg !86 + %1370 = extractvalue { i32, i32, i32, i32 } %1366, 3, !dbg !86 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !86 + %1371 = bitcast <2 x bfloat> %1192 to i32, !dbg !86 + %1372 = bitcast <2 x bfloat> %1236 to i32, !dbg !86 + %1373 = bitcast <2 x bfloat> %1280 to i32, !dbg !86 + %1374 = bitcast <2 x bfloat> %1324 to i32, !dbg !86 + tail call void asm sideeffect "st.shared.v4.b32 [ $0 + 0 ], { $1, $2, $3, $4 };", "r,r,r,r,r"(ptr addrspace(3) %1334, i32 %1371, i32 %1372, i32 %1373, i32 %1374) #6, !dbg !86 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !86 + %1375 = tail call { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) %1347), !dbg !86 + %1376 = extractvalue { i32, i32, i32, i32 } %1375, 0, !dbg !86 + %1377 = extractvalue { i32, i32, i32, i32 } %1375, 1, !dbg !86 + %1378 = extractvalue { i32, i32, i32, i32 } %1375, 2, !dbg !86 + %1379 = extractvalue { i32, i32, i32, i32 } %1375, 3, !dbg !86 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1349, i32 %1358, i32 %1367, i32 %1376, ptr addrspace(1) %1074, i1 %88) #6, !dbg !86 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1350, i32 %1359, i32 %1368, i32 %1377, ptr addrspace(1) %1072, i1 %89) #6, !dbg !86 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1351, i32 %1360, i32 %1369, i32 %1378, ptr addrspace(1) %1070, i1 %90) #6, !dbg !86 + tail call void asm sideeffect "@$5 st.global.v4.b32 [ $4 + 0 ], { $0, $1, $2, $3 };", "r,r,r,r,l,b"(i32 %1352, i32 %1361, i32 %1370, i32 %1379, ptr addrspace(1) %1068, i1 %91) #6, !dbg !86 + tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !87 + tail call void asm sideeffect "@$0 tcgen05.dealloc.cta_group::1.sync.aligned.b32 $1, 128;", "b,r"(i1 %12, ptr addrspace(6) %1075) #6, !dbg !87 + ret void, !dbg !87 +} + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1 + +; Function Attrs: convergent nocallback nounwind +declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #2 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) +declare noundef range(i32 0, 65535) i32 @llvm.nvvm.read.ptx.sreg.ctaid.y() #1 + +; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite) +declare i32 @llvm.nvvm.shfl.sync.idx.i32(i32, i32, i32, i32) #3 + +; Function Attrs: convergent nounwind memory(inaccessiblemem: readwrite) +declare void @llvm.nvvm.tcgen05.wait.st() #4 + +; Function Attrs: convergent mustprogress nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: readwrite) +declare { i32, i1 } @llvm.nvvm.elect.sync(i32) #5 + +; Function Attrs: nounwind +declare void @llvm.nvvm.cp.async.commit.group() #6 + +; Function Attrs: nounwind +declare void @llvm.nvvm.cp.async.wait.group(i32 immarg) #6 + +; Function Attrs: nocallback nounwind +declare void @llvm.nvvm.fence.proxy.async.shared_cta() #7 + +; Function Attrs: convergent nounwind memory(inaccessiblemem: readwrite) +declare void @llvm.nvvm.tcgen05.wait.ld() #4 + +; Function Attrs: convergent nocallback nofree nounwind memory(argmem: read) +declare { i32, i32, i32, i32 } @llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3) readonly captures(none)) #8 + +declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #9 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.ex2.approx.ftz.f32(float) #10 + +; Function Attrs: mustprogress nocallback nofree nosync nounwind willreturn memory(none) +declare float @llvm.nvvm.ex2.approx.f32(float) #10 + +attributes #0 = { nounwind "nvvm.reqntid"="128" } +attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) } +attributes #2 = { convergent nocallback nounwind } +attributes #3 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) } +attributes #4 = { convergent nounwind memory(inaccessiblemem: readwrite) } +attributes #5 = { convergent mustprogress nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: readwrite) } +attributes #6 = { nounwind } +attributes #7 = { nocallback nounwind } +attributes #8 = { convergent nocallback nofree nounwind memory(argmem: read) } +attributes #9 = { "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } +attributes #10 = { mustprogress nocallback nofree nosync nounwind willreturn memory(none) } + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!2, !3} +!llvm.ident = !{!4} + +!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly) +!1 = !DIFile(filename: "chunk_o.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common") +!2 = !{i32 2, !"Debug Info Version", i32 3} +!3 = !{i32 4, !"nvvm-reflect-ftz", i32 1} +!4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"} +!5 = distinct !DISubprogram(name: "chunk_fwd_kernel_o", linkageName: "chunk_fwd_kernel_o", scope: !1, file: !1, line: 58, type: !6, scopeLine: 58, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0) +!6 = !DISubroutineType(cc: DW_CC_normal, types: !7) +!7 = !{null, !8, !8, !8, !8, !10, !8, !11, !12, !8, !8} +!8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1) +!9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed) +!10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1) +!11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float) +!12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) +!13 = !DILocation(line: 58, scope: !5) +!14 = !DILocation(line: 25, column: 24, scope: !15, inlinedAt: !17) +!15 = distinct !DILexicalBlockFile(scope: !5, file: !16, discriminator: 0) +!16 = !DIFile(filename: "op.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils") +!17 = !DILocation(line: 82, column: 36, scope: !18) +!18 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0) +!19 = !DILocation(line: 25, column: 30, scope: !15, inlinedAt: !17) +!20 = !DILocation(line: 27, column: 31, scope: !15, inlinedAt: !17) +!21 = !DILocation(line: 83, column: 25, scope: !5) +!22 = !DILocation(line: 83, column: 31, scope: !5) +!23 = !DILocation(line: 84, column: 23, scope: !5) +!24 = !DILocation(line: 84, column: 34, scope: !5) +!25 = !DILocation(line: 43, column: 17, scope: !26, inlinedAt: !28) +!26 = distinct !DILexicalBlockFile(scope: !5, file: !27, discriminator: 0) +!27 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language") +!28 = !DILocation(line: 93, column: 24, scope: !18) +!29 = !DILocation(line: 43, column: 30, scope: !26, inlinedAt: !28) +!30 = !DILocation(line: 94, column: 21, scope: !5) +!31 = !DILocation(line: 94, column: 26, scope: !5) +!32 = !DILocation(line: 95, column: 25, scope: !5) +!33 = !DILocation(line: 97, column: 16, scope: !5) +!34 = !DILocation(line: 97, column: 20, scope: !5) +!35 = !DILocation(line: 97, column: 40, scope: !5) +!36 = !DILocation(line: 97, column: 9, scope: !5) +!37 = !DILocation(line: 98, column: 9, scope: !5) +!38 = !DILocation(line: 99, column: 9, scope: !5) +!39 = !DILocation(line: 101, column: 9, scope: !5) +!40 = !DILocation(line: 106, column: 16, scope: !5) +!41 = !DILocation(line: 106, column: 34, scope: !5) +!42 = !DILocation(line: 106, column: 21, scope: !5) +!43 = !DILocation(line: 107, column: 16, scope: !5) +!44 = !DILocation(line: 108, column: 16, scope: !5) +!45 = !DILocation(line: 108, column: 21, scope: !5) +!46 = !DILocation(line: 112, column: 34, scope: !5) +!47 = !DILocation(line: 112, column: 18, scope: !5) +!48 = !DILocation(line: 130, column: 35, scope: !5) +!49 = !DILocation(line: 132, column: 31, scope: !5) +!50 = !DILocation(line: 109, column: 21, scope: !5) +!51 = !DILocation(line: 112, column: 41, scope: !5) +!52 = !DILocation(line: 113, column: 18, scope: !5) +!53 = !DILocation(line: 113, column: 33, scope: !5) +!54 = !DILocation(line: 118, column: 37, scope: !5) +!55 = !DILocation(line: 118, column: 22, scope: !5) +!56 = !DILocation(line: 118, column: 41, scope: !5) +!57 = !DILocation(line: 121, column: 22, scope: !5) +!58 = !DILocation(line: 123, column: 22, scope: !5) +!59 = !DILocation(line: 124, column: 22, scope: !5) +!60 = !DILocation(line: 110, column: 25, scope: !5) +!61 = !DILocation(line: 135, column: 13, scope: !5) +!62 = !DILocation(line: 136, column: 18, scope: !5) +!63 = !DILocation(line: 137, column: 22, scope: !5) +!64 = !DILocation(line: 47, column: 33, scope: !15, inlinedAt: !65) +!65 = !DILocation(line: 138, column: 25, scope: !18) +!66 = !DILocation(line: 139, column: 40, scope: !5) +!67 = !DILocation(line: 47, column: 33, scope: !15, inlinedAt: !68) +!68 = !DILocation(line: 139, column: 25, scope: !18) +!69 = !DILocation(line: 145, column: 27, scope: !5) +!70 = !DILocation(line: 145, column: 44, scope: !5) +!71 = !DILocation(line: 148, column: 14, scope: !5) +!72 = !DILocation(line: 148, column: 38, scope: !5) +!73 = !DILocation(line: 151, column: 18, scope: !5) +!74 = !DILocation(line: 139, column: 20, scope: !5) +!75 = !DILocation(line: 154, column: 38, scope: !5) +!76 = !DILocation(line: 146, column: 29, scope: !5) +!77 = !DILocation(line: 154, column: 50, scope: !5) +!78 = !DILocation(line: 100, column: 9, scope: !5) +!79 = !DILocation(line: 149, column: 14, scope: !5) +!80 = !DILocation(line: 149, column: 38, scope: !5) +!81 = !DILocation(line: 138, column: 20, scope: !5) +!82 = !DILocation(line: 154, column: 16, scope: !5) +!83 = !DILocation(line: 154, column: 57, scope: !5) +!84 = !DILocation(line: 154, column: 24, scope: !5) +!85 = !DILocation(line: 155, column: 25, scope: !5) +!86 = !DILocation(line: 155, column: 18, scope: !5) +!87 = !DILocation(line: 155, column: 4, scope: !5) diff --git a/triton/gpu0/COEH6RZWJ4F4NLI6BWD6G2VBWJCNYUFXAFKSMEBEK2VRZ3DVOIQA/chunk_fwd_kernel_o.ptx b/triton/gpu0/COEH6RZWJ4F4NLI6BWD6G2VBWJCNYUFXAFKSMEBEK2VRZ3DVOIQA/chunk_fwd_kernel_o.ptx new file mode 100644 index 0000000000000000000000000000000000000000..3e792e6e03c00b27bb8f1e35ecaee78f301b430c --- /dev/null +++ b/triton/gpu0/COEH6RZWJ4F4NLI6BWD6G2VBWJCNYUFXAFKSMEBEK2VRZ3DVOIQA/chunk_fwd_kernel_o.ptx @@ -0,0 +1,2200 @@ +// +// Generated by LLVM NVPTX Back-End +// + +.version 9.1 +.target sm_100a +.address_size 64 + + // .globl chunk_fwd_kernel_o // -- Begin function chunk_fwd_kernel_o +.extern .shared .align 16 .b8 global_smem[]; +.global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90}; + // @chunk_fwd_kernel_o +.visible .entry chunk_fwd_kernel_o( + .param .u64 .ptr .global .align 1 chunk_fwd_kernel_o_param_0, + .param .u64 .ptr .global .align 1 chunk_fwd_kernel_o_param_1, + .param .u64 .ptr .global .align 1 chunk_fwd_kernel_o_param_2, + .param .u64 .ptr .global .align 1 chunk_fwd_kernel_o_param_3, + .param .u64 .ptr .global .align 1 chunk_fwd_kernel_o_param_4, + .param .u64 .ptr .global .align 1 chunk_fwd_kernel_o_param_5, + .param .f32 chunk_fwd_kernel_o_param_6, + .param .u32 chunk_fwd_kernel_o_param_7, + .param .u64 .ptr .global .align 1 chunk_fwd_kernel_o_param_8, + .param .u64 .ptr .global .align 1 chunk_fwd_kernel_o_param_9 +) +.reqntid 128 +{ + .reg .pred %p<117>; + .reg .b16 %rs<182>; + .reg .b32 %r<559>; + .reg .b64 %rd<441>; + .loc 1 58 0 // chunk_o.py:58:0 +$L__func_begin0: + .loc 1 58 0 // chunk_o.py:58:0 + +// %bb.0: + ld.param.b64 %rd88, [chunk_fwd_kernel_o_param_0]; +$L__tmp0: + .loc 1 58 0 // chunk_o.py:58 + mov.u32 %r1, %tid.x; + ld.param.b64 %rd89, [chunk_fwd_kernel_o_param_1]; + setp.lt.u32 %p1, %r1, 32; + ld.param.b64 %rd90, [chunk_fwd_kernel_o_param_2]; + mov.b32 %r16, global_smem; + // begin inline asm + @%p1 tcgen05.alloc.cta_group::1.sync.aligned.shared::cta.b32 [%r16], 128; + // end inline asm + ld.param.b64 %rd91, [chunk_fwd_kernel_o_param_3]; + bar.sync 0; + ld.shared.b32 %r508, [global_smem]; + bar.sync 0; + // begin inline asm + @%p1 tcgen05.relinquish_alloc_permit.cta_group::1.sync.aligned; + // end inline asm + ld.param.b32 %r36, [chunk_fwd_kernel_o_param_7]; +$L__tmp1: + .loc 2 25 24 // op.py:25:24 @[ chunk_o.py:82:36 ] + mov.u32 %r37, %ctaid.x; + .loc 2 25 30 // op.py:25:30 @[ chunk_o.py:82:36 ] + cvt.u64.u32 %rd92, %r37; + .loc 2 27 31 // op.py:27:31 @[ chunk_o.py:82:36 ] + shr.u64 %rd93, %rd92, 1; +$L__tmp2: + .loc 1 83 25 // chunk_o.py:83:25 + mov.u32 %r38, %ctaid.y; + .loc 1 83 31 // chunk_o.py:83:31 + cvt.u64.u32 %rd94, %r38; + .loc 1 84 23 // chunk_o.py:84:23 + shr.u64 %rd95, %rd94, 5; + .loc 1 84 34 // chunk_o.py:84:34 + and.b64 %rd96, %rd94, 31; +$L__tmp3: + .loc 3 43 17 // standard.py:43:17 @[ chunk_o.py:93:24 ] + add.s32 %r39, %r36, 63; + .loc 3 43 30 // standard.py:43:30 @[ chunk_o.py:93:24 ] + shr.s32 %r40, %r39, 31; + shr.u32 %r41, %r40, 26; + add.s32 %r42, %r39, %r41; + shr.s32 %r43, %r42, 6; +$L__tmp4: + .loc 1 94 21 // chunk_o.py:94:21 + cvt.s64.s32 %rd97, %r43; + .loc 1 94 26 // chunk_o.py:94:26 + mad.lo.s64 %rd98, %rd95, %rd97, %rd93; + .loc 1 95 25 // chunk_o.py:95:25 + cvt.s64.s32 %rd56, %r36; + and.b64 %rd99, %rd94, 65504; + .loc 1 97 16 // chunk_o.py:97:16 + mul.lo.s64 %rd100, %rd99, %rd56; + .loc 1 97 20 // chunk_o.py:97:20 + or.b64 %rd1, %rd100, %rd96; + .loc 1 97 9 // chunk_o.py:97:9 + shl.b64 %rd101, %rd1, 8; + add.s64 %rd102, %rd88, %rd101; + .loc 1 98 9 // chunk_o.py:98:9 + add.s64 %rd103, %rd89, %rd101; + .loc 1 101 9 // chunk_o.py:101:9 + shl.b64 %rd104, %rd98, 20; + add.s64 %rd105, %rd91, %rd104; + shl.b64 %rd106, %rd96, 15; + add.s64 %rd4, %rd105, %rd106; + .loc 1 106 16 // chunk_o.py:106:16 + shl.b64 %rd107, %rd93, 6; + .loc 1 106 34 // chunk_o.py:106:34 + shr.u32 %r2, %r1, 5; + shr.u32 %r44, %r1, 3; + bfe.u32 %r45, %r1, 3, 4; + or.b32 %r46, %r45, 16; + or.b32 %r47, %r45, 32; + or.b32 %r48, %r45, 48; + shl.b32 %r3, %r1, 3; + and.b32 %r49, %r3, 56; + and.b32 %r50, %r1, 15; + and.b32 %r4, %r1, 96; + shr.u32 %r51, %r4, 1; + or.b32 %r5, %r51, %r50; + and.b32 %r6, %r1, 16; + .loc 1 106 21 // chunk_o.py:106:21 + cvt.u64.u32 %rd108, %r45; + cvt.u64.u32 %rd109, %r46; + cvt.u64.u32 %rd110, %r47; + cvt.u64.u32 %rd111, %r48; + cvt.u64.u32 %rd112, %r5; + cvt.u64.u32 %rd113, %r49; + or.b64 %rd5, %rd107, %rd108; + or.b64 %rd6, %rd107, %rd109; + or.b64 %rd7, %rd107, %rd110; + or.b64 %rd8, %rd107, %rd111; + .loc 1 106 34 // chunk_o.py:106:34 + shl.b32 %r7, %r6, 1; + or.b32 %r52, %r7, 1; + or.b32 %r8, %r7, 3; + or.b32 %r9, %r7, 2; + or.b32 %r53, %r7, 4; + or.b32 %r54, %r7, 5; + or.b32 %r55, %r7, 6; + or.b32 %r56, %r7, 7; + or.b32 %r57, %r7, 8; + or.b32 %r58, %r7, 9; + or.b32 %r59, %r7, 10; + or.b32 %r60, %r7, 11; + or.b32 %r61, %r7, 12; + or.b32 %r62, %r7, 13; + or.b32 %r63, %r7, 14; + or.b32 %r64, %r7, 15; + or.b32 %r65, %r7, 16; + or.b32 %r66, %r7, 17; + or.b32 %r67, %r7, 18; + or.b32 %r68, %r7, 19; + or.b32 %r69, %r7, 20; + or.b32 %r70, %r7, 21; + or.b32 %r71, %r7, 22; + or.b32 %r72, %r7, 23; + or.b32 %r73, %r7, 24; + or.b32 %r74, %r7, 25; + or.b32 %r75, %r7, 26; + or.b32 %r76, %r7, 27; + or.b32 %r77, %r7, 28; + or.b32 %r78, %r7, 29; + or.b32 %r79, %r7, 30; + or.b32 %r80, %r7, 31; + .loc 1 106 21 // chunk_o.py:106:21 + cvt.u64.u32 %rd114, %r80; + cvt.u64.u32 %rd115, %r79; + cvt.u64.u32 %rd116, %r78; + cvt.u64.u32 %rd117, %r77; + cvt.u64.u32 %rd118, %r76; + cvt.u64.u32 %rd119, %r75; + cvt.u64.u32 %rd120, %r74; + cvt.u64.u32 %rd121, %r73; + cvt.u64.u32 %rd122, %r72; + cvt.u64.u32 %rd123, %r71; + cvt.u64.u32 %rd124, %r70; + cvt.u64.u32 %rd125, %r69; + cvt.u64.u32 %rd126, %r68; + cvt.u64.u32 %rd127, %r67; + cvt.u64.u32 %rd128, %r66; + cvt.u64.u32 %rd129, %r65; + cvt.u64.u32 %rd130, %r64; + cvt.u64.u32 %rd131, %r63; + cvt.u64.u32 %rd132, %r62; + cvt.u64.u32 %rd133, %r61; + cvt.u64.u32 %rd134, %r60; + cvt.u64.u32 %rd135, %r59; + cvt.u64.u32 %rd136, %r58; + cvt.u64.u32 %rd137, %r57; + cvt.u64.u32 %rd138, %r56; + cvt.u64.u32 %rd139, %r55; + cvt.u64.u32 %rd140, %r54; + cvt.u64.u32 %rd141, %r53; + cvt.u64.u32 %rd142, %r8; + cvt.u64.u32 %rd143, %r9; + cvt.u64.u32 %rd144, %r52; + cvt.u64.u32 %rd145, %r7; + .loc 1 107 16 // chunk_o.py:107:16 + setp.lt.s64 %p6, %rd5, %rd56; + setp.lt.s64 %p7, %rd6, %rd56; + setp.lt.s64 %p8, %rd7, %rd56; + setp.lt.s64 %p9, %rd8, %rd56; + .loc 1 108 16 // chunk_o.py:108:16 + mul.wide.u32 %rd146, %r37, 64; + and.b64 %rd147, %rd146, 64; + .loc 1 108 21 // chunk_o.py:108:21 + or.b64 %rd57, %rd147, %rd113; + .loc 1 112 18 // chunk_o.py:112:18 + shl.b64 %rd148, %rd5, 13; + add.s64 %rd149, %rd102, %rd148; + shl.b64 %rd150, %rd6, 13; + add.s64 %rd151, %rd102, %rd150; + shl.b64 %rd152, %rd7, 13; + add.s64 %rd153, %rd102, %rd152; + shl.b64 %rd154, %rd8, 13; + add.s64 %rd155, %rd102, %rd154; + .loc 1 130 35 // chunk_o.py:130:35 + shfl.sync.idx.b32 %r10, %r2, 0, 31, -1; + shl.b32 %r81, %r10, 21; + and.b32 %r82, %r81, 6291456; + add.s32 %r441, %r82, %r508; + mov.pred %p18, -1; + mov.b32 %r17, 0; + // begin inline asm + @%p18 tcgen05.st.sync.aligned.16x32bx2.x32.b32 [%r441 + 0], 32, {%r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17}; + // end inline asm + tcgen05.wait::st.sync.aligned; + bar.sync 0; + .loc 1 132 31 // chunk_o.py:132:31 + add.s32 %r403, %r508, 1048576; + add.s32 %r474, %r441, 1048576; + // begin inline asm + @%p18 tcgen05.st.sync.aligned.16x32bx2.x32.b32 [%r474 + 0], 32, {%r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17, %r17}; + // end inline asm + tcgen05.wait::st.sync.aligned; + bar.sync 0; + .loc 1 109 21 // chunk_o.py:109:21 + setp.eq.b32 %p10, %r2, 0; + elect.sync %r83|%p11, -1; + and.pred %p2, %p10, %p11; + add.s32 %r18, %r16, 73728; + // begin inline asm + @%p2 mbarrier.init.shared::cta.b64 [%r18], 1; + // end inline asm + elect.sync %r84|%p12, -1; + and.pred %p3, %p10, %p12; + add.s32 %r138, %r16, 73736; + // begin inline asm + @%p3 mbarrier.init.shared::cta.b64 [%r138], 1; + // end inline asm + elect.sync %r85|%p13, -1; + and.pred %p4, %p10, %p13; + add.s32 %r128, %r16, 73744; + // begin inline asm + @%p4 mbarrier.init.shared::cta.b64 [%r128], 1; + // end inline asm + elect.sync %r86|%p14, -1; + and.pred %p5, %p10, %p14; + add.s32 %r139, %r16, 73752; + // begin inline asm + @%p5 mbarrier.init.shared::cta.b64 [%r139], 1; + // end inline asm + .loc 1 112 41 // chunk_o.py:112:41 + mul.wide.u32 %rd156, %r49, 2; + add.s64 %rd64, %rd149, %rd156; + add.s64 %rd65, %rd151, %rd156; + add.s64 %rd66, %rd153, %rd156; + add.s64 %rd67, %rd155, %rd156; + .loc 1 113 18 // chunk_o.py:113:18 + add.s64 %rd157, %rd103, %rd156; + .loc 1 113 33 // chunk_o.py:113:33 + add.s64 %rd68, %rd157, %rd148; + add.s64 %rd69, %rd157, %rd150; + add.s64 %rd70, %rd157, %rd152; + add.s64 %rd71, %rd157, %rd154; + .loc 1 118 37 // chunk_o.py:118:37 + shl.b32 %r87, %r45, 7; + shl.b32 %r88, %r46, 7; + shl.b32 %r89, %r47, 7; + shl.b32 %r90, %r48, 7; + .loc 1 118 22 // chunk_o.py:118:22 + mad.wide.u32 %rd158, %r87, 2, %rd4; + mad.wide.u32 %rd159, %r88, 2, %rd4; + mad.wide.u32 %rd160, %r89, 2, %rd4; + mad.wide.u32 %rd161, %r90, 2, %rd4; + .loc 1 118 41 // chunk_o.py:118:41 + shl.b64 %rd162, %rd57, 1; + add.s64 %rd72, %rd158, %rd162; + add.s64 %rd73, %rd159, %rd162; + add.s64 %rd74, %rd160, %rd162; + add.s64 %rd75, %rd161, %rd162; + .loc 1 121 22 // chunk_o.py:121:22 + shl.b32 %r11, %r1, 4; + and.b32 %r91, %r11, 2032; + and.b32 %r12, %r1, 56; + shl.b32 %r92, %r12, 1; + xor.b32 %r93, %r91, %r92; + add.s32 %r189, %r16, %r93; + selp.b32 %r19, 16, 0, %p6; + // begin inline asm + cp.async.cg.shared.global [ %r189 + 0 ], [ %rd64 + 0 ], 0x10, %r19; + // end inline asm + add.s32 %r190, %r189, 2048; + selp.b32 %r20, 16, 0, %p7; + // begin inline asm + cp.async.cg.shared.global [ %r190 + 0 ], [ %rd65 + 0 ], 0x10, %r20; + // end inline asm + add.s32 %r191, %r189, 4096; + selp.b32 %r21, 16, 0, %p8; + // begin inline asm + cp.async.cg.shared.global [ %r191 + 0 ], [ %rd66 + 0 ], 0x10, %r21; + // end inline asm + add.s32 %r192, %r189, 6144; + selp.b32 %r22, 16, 0, %p9; + // begin inline asm + cp.async.cg.shared.global [ %r192 + 0 ], [ %rd67 + 0 ], 0x10, %r22; + // end inline asm + cp.async.commit_group; + .loc 1 123 22 // chunk_o.py:123:22 + add.s32 %r130, %r189, 24576; + // begin inline asm + cp.async.cg.shared.global [ %r130 + 0 ], [ %rd68 + 0 ], 0x10, %r19; + // end inline asm + add.s32 %r131, %r189, 26624; + // begin inline asm + cp.async.cg.shared.global [ %r131 + 0 ], [ %rd69 + 0 ], 0x10, %r20; + // end inline asm + add.s32 %r132, %r189, 28672; + // begin inline asm + cp.async.cg.shared.global [ %r132 + 0 ], [ %rd70 + 0 ], 0x10, %r21; + // end inline asm + add.s32 %r133, %r189, 30720; + // begin inline asm + cp.async.cg.shared.global [ %r133 + 0 ], [ %rd71 + 0 ], 0x10, %r22; + // end inline asm + cp.async.commit_group; + .loc 1 124 22 // chunk_o.py:124:22 + add.s32 %r134, %r189, 49152; + mov.b32 %r23, 16; + // begin inline asm + cp.async.cg.shared.global [ %r134 + 0 ], [ %rd72 + 0 ], 0x10, %r23; + // end inline asm + add.s32 %r135, %r189, 51200; + // begin inline asm + cp.async.cg.shared.global [ %r135 + 0 ], [ %rd73 + 0 ], 0x10, %r23; + // end inline asm + add.s32 %r136, %r189, 53248; + // begin inline asm + cp.async.cg.shared.global [ %r136 + 0 ], [ %rd74 + 0 ], 0x10, %r23; + // end inline asm + add.s32 %r137, %r189, 55296; + // begin inline asm + cp.async.cg.shared.global [ %r137 + 0 ], [ %rd75 + 0 ], 0x10, %r23; + // end inline asm + cp.async.commit_group; + .loc 1 112 41 // chunk_o.py:112:41 + add.s64 %rd76, %rd64, 128; + add.s64 %rd77, %rd65, 128; + add.s64 %rd78, %rd66, 128; + add.s64 %rd79, %rd67, 128; + .loc 1 113 33 // chunk_o.py:113:33 + add.s64 %rd80, %rd68, 128; + add.s64 %rd81, %rd69, 128; + add.s64 %rd82, %rd70, 128; + add.s64 %rd83, %rd71, 128; + .loc 1 118 37 // chunk_o.py:118:37 + shl.b32 %r13, %r44, 7; + or.b32 %r94, %r13, 14336; + .loc 1 118 22 // chunk_o.py:118:22 + mad.wide.u32 %rd163, %r94, 2, %rd4; + .loc 1 118 41 // chunk_o.py:118:41 + add.s64 %rd84, %rd72, 16384; + add.s64 %rd85, %rd72, 20480; + add.s64 %rd86, %rd72, 24576; + add.s64 %rd87, %rd163, %rd162; + .loc 1 121 22 // chunk_o.py:121:22 + bar.sync 0; + add.s32 %r24, %r189, 8192; + // begin inline asm + cp.async.cg.shared.global [ %r24 + 0 ], [ %rd76 + 0 ], 0x10, %r19; + // end inline asm + add.s32 %r25, %r189, 10240; + // begin inline asm + cp.async.cg.shared.global [ %r25 + 0 ], [ %rd77 + 0 ], 0x10, %r20; + // end inline asm + add.s32 %r26, %r189, 12288; + // begin inline asm + cp.async.cg.shared.global [ %r26 + 0 ], [ %rd78 + 0 ], 0x10, %r21; + // end inline asm + add.s32 %r27, %r189, 14336; + // begin inline asm + cp.async.cg.shared.global [ %r27 + 0 ], [ %rd79 + 0 ], 0x10, %r22; + // end inline asm + cp.async.commit_group; + .loc 1 123 22 // chunk_o.py:123:22 + add.s32 %r28, %r189, 32768; + // begin inline asm + cp.async.cg.shared.global [ %r28 + 0 ], [ %rd80 + 0 ], 0x10, %r19; + // end inline asm + add.s32 %r29, %r189, 34816; + // begin inline asm + cp.async.cg.shared.global [ %r29 + 0 ], [ %rd81 + 0 ], 0x10, %r20; + // end inline asm + add.s32 %r30, %r189, 36864; + // begin inline asm + cp.async.cg.shared.global [ %r30 + 0 ], [ %rd82 + 0 ], 0x10, %r21; + // end inline asm + add.s32 %r31, %r189, 38912; + // begin inline asm + cp.async.cg.shared.global [ %r31 + 0 ], [ %rd83 + 0 ], 0x10, %r22; + // end inline asm + cp.async.commit_group; + .loc 1 124 22 // chunk_o.py:124:22 + add.s32 %r32, %r189, 57344; + // begin inline asm + cp.async.cg.shared.global [ %r32 + 0 ], [ %rd84 + 0 ], 0x10, %r23; + // end inline asm + add.s32 %r33, %r189, 59392; + // begin inline asm + cp.async.cg.shared.global [ %r33 + 0 ], [ %rd85 + 0 ], 0x10, %r23; + // end inline asm + add.s32 %r34, %r189, 61440; + // begin inline asm + cp.async.cg.shared.global [ %r34 + 0 ], [ %rd86 + 0 ], 0x10, %r23; + // end inline asm + add.s32 %r35, %r189, 63488; + // begin inline asm + cp.async.cg.shared.global [ %r35 + 0 ], [ %rd87 + 0 ], 0x10, %r23; + // end inline asm + cp.async.commit_group; + .loc 1 121 22 // chunk_o.py:121:22 + cp.async.wait_group 3; + bar.sync 0; + .loc 1 130 35 // chunk_o.py:130:35 + fence.proxy.async.shared::cta; + setp.ne.b32 %p15, %r10, 0; + @%p15 bra $L__BB0_2; +// %bb.1: + elect.sync %r97|%p17, -1; + bfe.u32 %r98, %r16, 4, 14; + cvt.u64.u32 %rd176, %r98; + add.s32 %r99, %r16, 49152; + bfe.u32 %r100, %r99, 4, 14; + cvt.u64.u32 %rd177, %r100; + or.b64 %rd164, %rd176, 4611756662049472512; + or.b64 %rd165, %rd177, 4611756662049472512; + mov.b32 %r95, 68224144; + mov.pred %p16, 0; + // begin inline asm + @%p17 tcgen05.mma.cta_group::1.kind::f16 [ %r508 + 0 ], %rd164, %rd165, %r95, %p16; + // end inline asm + add.s64 %rd166, %rd176, 4611756662049472514; + add.s64 %rd167, %rd177, 4611756662049472640; + // begin inline asm + @%p17 tcgen05.mma.cta_group::1.kind::f16 [ %r508 + 0 ], %rd166, %rd167, %r95, %p18; + // end inline asm + add.s64 %rd168, %rd176, 4611756662049472516; + add.s64 %rd169, %rd177, 4611756662049472768; + // begin inline asm + @%p17 tcgen05.mma.cta_group::1.kind::f16 [ %r508 + 0 ], %rd168, %rd169, %r95, %p18; + // end inline asm + add.s64 %rd170, %rd176, 4611756662049472518; + add.s64 %rd171, %rd177, 4611756662049472896; + // begin inline asm + @%p17 tcgen05.mma.cta_group::1.kind::f16 [ %r508 + 0 ], %rd170, %rd171, %r95, %p18; + // end inline asm + // begin inline asm + @%p17 tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [%r18]; + // end inline asm + .loc 1 132 31 // chunk_o.py:132:31 + elect.sync %r101|%p19, -1; + add.s32 %r102, %r16, 24576; + bfe.u32 %r103, %r102, 4, 14; + cvt.u64.u32 %rd178, %r103; + or.b64 %rd172, %rd178, 4611756662049472512; + mov.b32 %r96, 68158608; + // begin inline asm + @%p19 tcgen05.mma.cta_group::1.kind::f16 [ %r403 + 0 ], %rd164, %rd172, %r96, %p16; + // end inline asm + add.s64 %rd173, %rd178, 4611756662049472514; + // begin inline asm + @%p19 tcgen05.mma.cta_group::1.kind::f16 [ %r403 + 0 ], %rd166, %rd173, %r96, %p18; + // end inline asm + add.s64 %rd174, %rd178, 4611756662049472516; + // begin inline asm + @%p19 tcgen05.mma.cta_group::1.kind::f16 [ %r403 + 0 ], %rd168, %rd174, %r96, %p18; + // end inline asm + add.s64 %rd175, %rd178, 4611756662049472518; + // begin inline asm + @%p19 tcgen05.mma.cta_group::1.kind::f16 [ %r403 + 0 ], %rd170, %rd175, %r96, %p18; + // end inline asm + // begin inline asm + @%p19 tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [%r128]; + // end inline asm +$L__BB0_2: // %.critedge + .loc 1 0 31 // chunk_o.py:0:31 + ld.param.b64 %rd62, [chunk_fwd_kernel_o_param_4]; + add.s64 %rd3, %rd90, %rd101; + or.b64 %rd9, %rd107, %rd112; + mov.b64 %rd11, {%r54, %r53}; + mov.b64 %rd10, {%r56, %r55}; + mov.b64 %rd15, {%r58, %r57}; + mov.b64 %rd14, {%r60, %r59}; + mov.b64 %rd13, {%r62, %r61}; + mov.b64 %rd12, {%r64, %r63}; + mov.b64 %rd23, {%r66, %r65}; + mov.b64 %rd22, {%r68, %r67}; + mov.b64 %rd21, {%r70, %r69}; + mov.b64 %rd20, {%r72, %r71}; + mov.b64 %rd19, {%r74, %r73}; + mov.b64 %rd18, {%r76, %r75}; + mov.b64 %rd17, {%r78, %r77}; + mov.b64 %rd16, {%r80, %r79}; + or.b64 %rd55, %rd107, %rd145; + or.b64 %rd54, %rd107, %rd144; + or.b64 %rd53, %rd107, %rd143; + or.b64 %rd52, %rd107, %rd142; + or.b64 %rd51, %rd107, %rd141; + or.b64 %rd50, %rd107, %rd140; + or.b64 %rd49, %rd107, %rd139; + or.b64 %rd48, %rd107, %rd138; + or.b64 %rd47, %rd107, %rd137; + or.b64 %rd46, %rd107, %rd136; + or.b64 %rd45, %rd107, %rd135; + or.b64 %rd44, %rd107, %rd134; + or.b64 %rd43, %rd107, %rd133; + or.b64 %rd42, %rd107, %rd132; + or.b64 %rd41, %rd107, %rd131; + or.b64 %rd40, %rd107, %rd130; + or.b64 %rd39, %rd107, %rd129; + or.b64 %rd38, %rd107, %rd128; + or.b64 %rd37, %rd107, %rd127; + or.b64 %rd36, %rd107, %rd126; + or.b64 %rd35, %rd107, %rd125; + or.b64 %rd34, %rd107, %rd124; + or.b64 %rd33, %rd107, %rd123; + or.b64 %rd32, %rd107, %rd122; + or.b64 %rd31, %rd107, %rd121; + or.b64 %rd30, %rd107, %rd120; + or.b64 %rd29, %rd107, %rd119; + or.b64 %rd28, %rd107, %rd118; + or.b64 %rd27, %rd107, %rd117; + or.b64 %rd26, %rd107, %rd116; + or.b64 %rd25, %rd107, %rd115; + or.b64 %rd24, %rd107, %rd114; + shl.b64 %rd58, %rd5, 12; + shl.b64 %rd59, %rd6, 12; + shl.b64 %rd60, %rd7, 12; + shl.b64 %rd61, %rd8, 12; + .loc 1 112 41 // chunk_o.py:112:41 + add.s64 %rd179, %rd64, 256; + add.s64 %rd180, %rd65, 256; + add.s64 %rd181, %rd66, 256; + add.s64 %rd182, %rd67, 256; + .loc 1 113 33 // chunk_o.py:113:33 + add.s64 %rd183, %rd68, 256; + add.s64 %rd184, %rd69, 256; + add.s64 %rd185, %rd70, 256; + add.s64 %rd186, %rd71, 256; + .loc 1 118 41 // chunk_o.py:118:41 + add.s64 %rd187, %rd72, 32768; + add.s64 %rd188, %rd72, 36864; + add.s64 %rd189, %rd72, 40960; + add.s64 %rd190, %rd72, 45056; + .loc 1 121 22 // chunk_o.py:121:22 + bar.sync 0; + add.s32 %r104, %r189, 16384; + // begin inline asm + cp.async.cg.shared.global [ %r104 + 0 ], [ %rd179 + 0 ], 0x10, %r17; + // end inline asm + add.s32 %r105, %r189, 18432; + // begin inline asm + cp.async.cg.shared.global [ %r105 + 0 ], [ %rd180 + 0 ], 0x10, %r17; + // end inline asm + add.s32 %r106, %r189, 20480; + // begin inline asm + cp.async.cg.shared.global [ %r106 + 0 ], [ %rd181 + 0 ], 0x10, %r17; + // end inline asm + add.s32 %r107, %r189, 22528; + // begin inline asm + cp.async.cg.shared.global [ %r107 + 0 ], [ %rd182 + 0 ], 0x10, %r17; + // end inline asm + cp.async.commit_group; + .loc 1 123 22 // chunk_o.py:123:22 + add.s32 %r108, %r189, 40960; + // begin inline asm + cp.async.cg.shared.global [ %r108 + 0 ], [ %rd183 + 0 ], 0x10, %r17; + // end inline asm + add.s32 %r109, %r189, 43008; + // begin inline asm + cp.async.cg.shared.global [ %r109 + 0 ], [ %rd184 + 0 ], 0x10, %r17; + // end inline asm + add.s32 %r110, %r189, 45056; + // begin inline asm + cp.async.cg.shared.global [ %r110 + 0 ], [ %rd185 + 0 ], 0x10, %r17; + // end inline asm + add.s32 %r111, %r189, 47104; + // begin inline asm + cp.async.cg.shared.global [ %r111 + 0 ], [ %rd186 + 0 ], 0x10, %r17; + // end inline asm + cp.async.commit_group; + .loc 1 124 22 // chunk_o.py:124:22 + add.s32 %r112, %r189, 65536; + // begin inline asm + cp.async.cg.shared.global [ %r112 + 0 ], [ %rd187 + 0 ], 0x10, %r17; + // end inline asm + add.s32 %r113, %r189, 67584; + // begin inline asm + cp.async.cg.shared.global [ %r113 + 0 ], [ %rd188 + 0 ], 0x10, %r17; + // end inline asm + add.s32 %r114, %r189, 69632; + // begin inline asm + cp.async.cg.shared.global [ %r114 + 0 ], [ %rd189 + 0 ], 0x10, %r17; + // end inline asm + add.s32 %r115, %r189, 71680; + // begin inline asm + cp.async.cg.shared.global [ %r115 + 0 ], [ %rd190 + 0 ], 0x10, %r17; + // end inline asm + cp.async.commit_group; + .loc 1 121 22 // chunk_o.py:121:22 + cp.async.wait_group 3; + bar.sync 0; + .loc 1 130 35 // chunk_o.py:130:35 + fence.proxy.async.shared::cta; + @%p15 bra $L__BB0_4; + bra.uni $L__BB0_3; +$L__BB0_4: // %.critedge2 + .loc 1 0 35 // chunk_o.py:0:35 + mov.b32 %r116, 0; + .loc 1 130 35 // chunk_o.py:130:35 + // begin inline asm + +{ + .reg .pred complete; + waitLoop: + mbarrier.try_wait.parity.shared::cta.b64 complete, [%r18], %r116; + @!complete bra.uni waitLoop; +} + + // end inline asm + bra.uni $L__BB0_5; +$L__BB0_3: + elect.sync %r120|%p21, -1; + add.s32 %r121, %r16, 8192; + bfe.u32 %r122, %r121, 4, 14; + cvt.u64.u32 %rd203, %r122; + add.s32 %r123, %r16, 57344; + bfe.u32 %r124, %r123, 4, 14; + cvt.u64.u32 %rd204, %r124; + or.b64 %rd191, %rd203, 4611756662049472512; + or.b64 %rd192, %rd204, 4611756662049472512; + mov.b32 %r117, 68224144; + mov.pred %p20, -1; + // begin inline asm + @%p21 tcgen05.mma.cta_group::1.kind::f16 [ %r508 + 0 ], %rd191, %rd192, %r117, %p20; + // end inline asm + add.s64 %rd193, %rd203, 4611756662049472514; + add.s64 %rd194, %rd204, 4611756662049472640; + // begin inline asm + @%p21 tcgen05.mma.cta_group::1.kind::f16 [ %r508 + 0 ], %rd193, %rd194, %r117, %p20; + // end inline asm + add.s64 %rd195, %rd203, 4611756662049472516; + add.s64 %rd196, %rd204, 4611756662049472768; + // begin inline asm + @%p21 tcgen05.mma.cta_group::1.kind::f16 [ %r508 + 0 ], %rd195, %rd196, %r117, %p20; + // end inline asm + add.s64 %rd197, %rd203, 4611756662049472518; + add.s64 %rd198, %rd204, 4611756662049472896; + // begin inline asm + @%p21 tcgen05.mma.cta_group::1.kind::f16 [ %r508 + 0 ], %rd197, %rd198, %r117, %p20; + // end inline asm + // begin inline asm + @%p21 tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [%r138]; + // end inline asm + mov.b32 %r118, 0; + // begin inline asm + +{ + .reg .pred complete; + waitLoop: + mbarrier.try_wait.parity.shared::cta.b64 complete, [%r18], %r118; + @!complete bra.uni waitLoop; +} + + // end inline asm + .loc 1 132 31 // chunk_o.py:132:31 + elect.sync %r125|%p22, -1; + add.s32 %r126, %r16, 32768; + bfe.u32 %r127, %r126, 4, 14; + cvt.u64.u32 %rd205, %r127; + or.b64 %rd199, %rd205, 4611756662049472512; + mov.b32 %r119, 68158608; + // begin inline asm + @%p22 tcgen05.mma.cta_group::1.kind::f16 [ %r403 + 0 ], %rd191, %rd199, %r119, %p20; + // end inline asm + add.s64 %rd200, %rd205, 4611756662049472514; + // begin inline asm + @%p22 tcgen05.mma.cta_group::1.kind::f16 [ %r403 + 0 ], %rd193, %rd200, %r119, %p20; + // end inline asm + add.s64 %rd201, %rd205, 4611756662049472516; + // begin inline asm + @%p22 tcgen05.mma.cta_group::1.kind::f16 [ %r403 + 0 ], %rd195, %rd201, %r119, %p20; + // end inline asm + add.s64 %rd202, %rd205, 4611756662049472518; + // begin inline asm + @%p22 tcgen05.mma.cta_group::1.kind::f16 [ %r403 + 0 ], %rd197, %rd202, %r119, %p20; + // end inline asm + // begin inline asm + @%p22 tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [%r139]; + // end inline asm +$L__BB0_5: // %__nv_exp2f.exit + .loc 1 0 31 // chunk_o.py:0:31 + ld.param.b32 %r15, [chunk_fwd_kernel_o_param_6]; + ld.param.b64 %rd63, [chunk_fwd_kernel_o_param_5]; + shl.b64 %rd2, %rd1, 7; + .loc 1 107 16 // chunk_o.py:107:16 + setp.lt.s64 %p61, %rd34, %rd56; + selp.b16 %rs1, 1, 0, %p61; + shl.b16 %rs2, %rs1, 2; + setp.lt.s64 %p62, %rd35, %rd56; + selp.b16 %rs3, -1, 0, %p62; + shl.b16 %rs4, %rs3, 3; + or.b16 %rs5, %rs4, %rs2; + setp.lt.s64 %p63, %rd32, %rd56; + selp.b16 %rs6, 1, 0, %p63; + setp.lt.s64 %p64, %rd33, %rd56; + selp.b16 %rs7, -1, 0, %p64; + shl.b16 %rs8, %rs7, 1; + or.b16 %rs9, %rs6, %rs8; + and.b16 %rs10, %rs9, 3; + or.b16 %rs11, %rs10, %rs5; + and.b16 %rs12, %rs11, 15; + shl.b16 %rs13, %rs12, 8; + setp.lt.s64 %p65, %rd38, %rd56; + selp.b16 %rs14, 1, 0, %p65; + shl.b16 %rs15, %rs14, 2; + setp.lt.s64 %p66, %rd39, %rd56; + selp.b16 %rs16, -1, 0, %p66; + shl.b16 %rs17, %rs16, 3; + or.b16 %rs18, %rs17, %rs15; + setp.lt.s64 %p67, %rd36, %rd56; + selp.b16 %rs19, 1, 0, %p67; + setp.lt.s64 %p68, %rd37, %rd56; + selp.b16 %rs20, -1, 0, %p68; + shl.b16 %rs21, %rs20, 1; + or.b16 %rs22, %rs19, %rs21; + and.b16 %rs23, %rs22, 3; + or.b16 %rs24, %rs23, %rs18; + shl.b16 %rs25, %rs24, 12; + or.b16 %rs26, %rs25, %rs13; + setp.lt.s64 %p69, %rd26, %rd56; + selp.b16 %rs27, 1, 0, %p69; + shl.b16 %rs28, %rs27, 2; + setp.lt.s64 %p70, %rd27, %rd56; + selp.b16 %rs29, -1, 0, %p70; + shl.b16 %rs30, %rs29, 3; + or.b16 %rs31, %rs30, %rs28; + setp.lt.s64 %p59, %rd24, %rd56; + selp.b16 %rs32, 1, 0, %p59; + setp.lt.s64 %p71, %rd25, %rd56; + selp.b16 %rs33, -1, 0, %p71; + shl.b16 %rs34, %rs33, 1; + or.b16 %rs35, %rs32, %rs34; + and.b16 %rs36, %rs35, 3; + or.b16 %rs37, %rs36, %rs31; + and.b16 %rs38, %rs37, 15; + setp.lt.s64 %p72, %rd30, %rd56; + selp.b16 %rs39, 1, 0, %p72; + shl.b16 %rs40, %rs39, 2; + setp.lt.s64 %p73, %rd31, %rd56; + selp.b16 %rs41, -1, 0, %p73; + shl.b16 %rs42, %rs41, 3; + or.b16 %rs43, %rs42, %rs40; + setp.lt.s64 %p74, %rd28, %rd56; + selp.b16 %rs44, 1, 0, %p74; + setp.lt.s64 %p75, %rd29, %rd56; + selp.b16 %rs45, -1, 0, %p75; + shl.b16 %rs46, %rs45, 1; + or.b16 %rs47, %rs44, %rs46; + and.b16 %rs48, %rs47, 3; + or.b16 %rs49, %rs48, %rs43; + shl.b16 %rs50, %rs49, 4; + or.b16 %rs51, %rs38, %rs50; + and.b16 %rs52, %rs51, 255; + or.b16 %rs53, %rs52, %rs26; + cvt.u32.u16 %r242, %rs53; + setp.lt.s64 %p43, %rd40, %rd56; + setp.lt.s64 %p42, %rd41, %rd56; + setp.lt.s64 %p41, %rd42, %rd56; + setp.lt.s64 %p40, %rd43, %rd56; + setp.lt.s64 %p39, %rd44, %rd56; + setp.lt.s64 %p38, %rd45, %rd56; + setp.lt.s64 %p37, %rd46, %rd56; + setp.lt.s64 %p36, %rd47, %rd56; + setp.lt.s64 %p35, %rd48, %rd56; + setp.lt.s64 %p34, %rd49, %rd56; + setp.lt.s64 %p33, %rd50, %rd56; + setp.lt.s64 %p32, %rd51, %rd56; + setp.lt.s64 %p31, %rd52, %rd56; + setp.lt.s64 %p30, %rd53, %rd56; + setp.lt.s64 %p29, %rd54, %rd56; + setp.lt.s64 %p28, %rd55, %rd56; + setp.lt.s64 %p27, %rd9, %rd56; + mov.b32 %r129, 0; + .loc 1 132 31 // chunk_o.py:132:31 + // begin inline asm + +{ + .reg .pred complete; + waitLoop: + mbarrier.try_wait.parity.shared::cta.b64 complete, [%r128], %r129; + @!complete bra.uni waitLoop; +} + + // end inline asm + .loc 1 112 41 // chunk_o.py:112:41 + add.s64 %rd206, %rd64, 384; + add.s64 %rd207, %rd65, 384; + add.s64 %rd208, %rd66, 384; + add.s64 %rd209, %rd67, 384; + .loc 1 113 33 // chunk_o.py:113:33 + add.s64 %rd210, %rd68, 384; + add.s64 %rd211, %rd69, 384; + add.s64 %rd212, %rd70, 384; + add.s64 %rd213, %rd71, 384; + .loc 1 118 37 // chunk_o.py:118:37 + or.b32 %r243, %r13, 30720; + .loc 1 118 22 // chunk_o.py:118:22 + mad.wide.u32 %rd255, %r243, 2, %rd4; + .loc 1 118 41 // chunk_o.py:118:41 + add.s64 %rd214, %rd72, 49152; + add.s64 %rd215, %rd72, 53248; + add.s64 %rd216, %rd72, 57344; + add.s64 %rd217, %rd255, %rd162; + .loc 1 121 22 // chunk_o.py:121:22 + bar.sync 0; + // begin inline asm + cp.async.cg.shared.global [ %r189 + 0 ], [ %rd206 + 0 ], 0x10, %r129; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r190 + 0 ], [ %rd207 + 0 ], 0x10, %r129; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r191 + 0 ], [ %rd208 + 0 ], 0x10, %r129; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r192 + 0 ], [ %rd209 + 0 ], 0x10, %r129; + // end inline asm + cp.async.commit_group; + .loc 1 123 22 // chunk_o.py:123:22 + // begin inline asm + cp.async.cg.shared.global [ %r130 + 0 ], [ %rd210 + 0 ], 0x10, %r129; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r131 + 0 ], [ %rd211 + 0 ], 0x10, %r129; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r132 + 0 ], [ %rd212 + 0 ], 0x10, %r129; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r133 + 0 ], [ %rd213 + 0 ], 0x10, %r129; + // end inline asm + cp.async.commit_group; + .loc 1 124 22 // chunk_o.py:124:22 + // begin inline asm + cp.async.cg.shared.global [ %r134 + 0 ], [ %rd214 + 0 ], 0x10, %r129; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r135 + 0 ], [ %rd215 + 0 ], 0x10, %r129; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r136 + 0 ], [ %rd216 + 0 ], 0x10, %r129; + // end inline asm + // begin inline asm + cp.async.cg.shared.global [ %r137 + 0 ], [ %rd217 + 0 ], 0x10, %r129; + // end inline asm + cp.async.commit_group; + .loc 1 130 35 // chunk_o.py:130:35 + // begin inline asm + +{ + .reg .pred complete; + waitLoop: + mbarrier.try_wait.parity.shared::cta.b64 complete, [%r138], %r129; + @!complete bra.uni waitLoop; +} + + // end inline asm + .loc 1 132 31 // chunk_o.py:132:31 + // begin inline asm + +{ + .reg .pred complete; + waitLoop: + mbarrier.try_wait.parity.shared::cta.b64 complete, [%r139], %r129; + @!complete bra.uni waitLoop; +} + + // end inline asm + .loc 1 109 21 // chunk_o.py:109:21 + cp.async.wait_group 0; + bar.sync 0; + elect.sync %r244|%p76, -1; + and.pred %p23, %p10, %p76; + // begin inline asm + @%p23 mbarrier.inval.shared::cta.b64 [%r128]; + // end inline asm + elect.sync %r245|%p77, -1; + and.pred %p24, %p10, %p77; + // begin inline asm + @%p24 mbarrier.inval.shared::cta.b64 [%r139]; + // end inline asm + elect.sync %r246|%p78, -1; + and.pred %p25, %p10, %p78; + // begin inline asm + @%p25 mbarrier.inval.shared::cta.b64 [%r18]; + // end inline asm + elect.sync %r247|%p79, -1; + and.pred %p26, %p10, %p79; + // begin inline asm + @%p26 mbarrier.inval.shared::cta.b64 [%r138]; + // end inline asm + .loc 1 135 13 // chunk_o.py:135:13 + shl.b64 %rd256, %rd1, 2; + add.s64 %rd257, %rd62, %rd256; + .loc 1 136 18 // chunk_o.py:136:18 + shl.b64 %rd258, %rd9, 7; + add.s64 %rd218, %rd257, %rd258; + shl.b64 %rd259, %rd55, 7; + add.s64 %rd219, %rd257, %rd259; + shl.b64 %rd260, %rd54, 7; + add.s64 %rd220, %rd257, %rd260; + shl.b64 %rd261, %rd53, 7; + add.s64 %rd221, %rd257, %rd261; + shl.b64 %rd262, %rd52, 7; + add.s64 %rd222, %rd257, %rd262; + shl.b64 %rd263, %rd51, 7; + add.s64 %rd223, %rd257, %rd263; + shl.b64 %rd264, %rd50, 7; + add.s64 %rd224, %rd257, %rd264; + shl.b64 %rd265, %rd49, 7; + add.s64 %rd225, %rd257, %rd265; + shl.b64 %rd266, %rd48, 7; + add.s64 %rd226, %rd257, %rd266; + shl.b64 %rd267, %rd47, 7; + add.s64 %rd227, %rd257, %rd267; + shl.b64 %rd268, %rd46, 7; + add.s64 %rd228, %rd257, %rd268; + shl.b64 %rd269, %rd45, 7; + add.s64 %rd229, %rd257, %rd269; + shl.b64 %rd270, %rd44, 7; + add.s64 %rd230, %rd257, %rd270; + shl.b64 %rd271, %rd43, 7; + add.s64 %rd231, %rd257, %rd271; + shl.b64 %rd272, %rd42, 7; + add.s64 %rd232, %rd257, %rd272; + shl.b64 %rd273, %rd41, 7; + add.s64 %rd233, %rd257, %rd273; + shl.b64 %rd274, %rd40, 7; + add.s64 %rd234, %rd257, %rd274; + shl.b64 %rd275, %rd39, 7; + add.s64 %rd235, %rd257, %rd275; + shl.b64 %rd276, %rd38, 7; + add.s64 %rd236, %rd257, %rd276; + shl.b64 %rd277, %rd37, 7; + add.s64 %rd237, %rd257, %rd277; + shl.b64 %rd278, %rd36, 7; + add.s64 %rd238, %rd257, %rd278; + shl.b64 %rd279, %rd35, 7; + add.s64 %rd239, %rd257, %rd279; + shl.b64 %rd280, %rd34, 7; + add.s64 %rd240, %rd257, %rd280; + shl.b64 %rd281, %rd33, 7; + add.s64 %rd241, %rd257, %rd281; + shl.b64 %rd282, %rd32, 7; + add.s64 %rd242, %rd257, %rd282; + shl.b64 %rd283, %rd31, 7; + add.s64 %rd243, %rd257, %rd283; + shl.b64 %rd284, %rd30, 7; + add.s64 %rd244, %rd257, %rd284; + shl.b64 %rd285, %rd29, 7; + add.s64 %rd245, %rd257, %rd285; + shl.b64 %rd286, %rd28, 7; + add.s64 %rd246, %rd257, %rd286; + shl.b64 %rd287, %rd27, 7; + add.s64 %rd247, %rd257, %rd287; + shl.b64 %rd288, %rd26, 7; + add.s64 %rd248, %rd257, %rd288; + shl.b64 %rd289, %rd25, 7; + add.s64 %rd249, %rd257, %rd289; + shl.b64 %rd290, %rd24, 7; + add.s64 %rd250, %rd257, %rd290; + .loc 1 137 22 // chunk_o.py:137:22 + // begin inline asm + mov.u32 %r140, %r129; + @%p27 ld.global.b32 { %r140 }, [ %rd218 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r141, %r129; + @%p28 ld.global.b32 { %r141 }, [ %rd219 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r142, %r129; + @%p29 ld.global.b32 { %r142 }, [ %rd220 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r143, %r129; + @%p30 ld.global.b32 { %r143 }, [ %rd221 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r144, %r129; + @%p31 ld.global.b32 { %r144 }, [ %rd222 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r145, %r129; + @%p32 ld.global.b32 { %r145 }, [ %rd223 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r146, %r129; + @%p33 ld.global.b32 { %r146 }, [ %rd224 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r147, %r129; + @%p34 ld.global.b32 { %r147 }, [ %rd225 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r148, %r129; + @%p35 ld.global.b32 { %r148 }, [ %rd226 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r149, %r129; + @%p36 ld.global.b32 { %r149 }, [ %rd227 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r150, %r129; + @%p37 ld.global.b32 { %r150 }, [ %rd228 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r151, %r129; + @%p38 ld.global.b32 { %r151 }, [ %rd229 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r152, %r129; + @%p39 ld.global.b32 { %r152 }, [ %rd230 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r153, %r129; + @%p40 ld.global.b32 { %r153 }, [ %rd231 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r154, %r129; + @%p41 ld.global.b32 { %r154 }, [ %rd232 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r155, %r129; + @%p42 ld.global.b32 { %r155 }, [ %rd233 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r156, %r129; + @%p43 ld.global.b32 { %r156 }, [ %rd234 + 0 ]; + // end inline asm + shr.u32 %r248, %r242, 15; + and.b32 %r249, %r248, 1; + setp.ne.b32 %p44, %r249, 0; + // begin inline asm + mov.u32 %r157, %r129; + @%p44 ld.global.b32 { %r157 }, [ %rd235 + 0 ]; + // end inline asm + shr.u32 %r250, %r242, 14; + and.b32 %r251, %r250, 1; + setp.ne.b32 %p45, %r251, 0; + // begin inline asm + mov.u32 %r158, %r129; + @%p45 ld.global.b32 { %r158 }, [ %rd236 + 0 ]; + // end inline asm + shr.u32 %r252, %r242, 13; + and.b32 %r253, %r252, 1; + setp.ne.b32 %p46, %r253, 0; + // begin inline asm + mov.u32 %r159, %r129; + @%p46 ld.global.b32 { %r159 }, [ %rd237 + 0 ]; + // end inline asm + shr.u32 %r254, %r242, 12; + and.b32 %r255, %r254, 1; + setp.ne.b32 %p47, %r255, 0; + // begin inline asm + mov.u32 %r160, %r129; + @%p47 ld.global.b32 { %r160 }, [ %rd238 + 0 ]; + // end inline asm + shr.u32 %r256, %r242, 11; + and.b32 %r257, %r256, 1; + setp.ne.b32 %p48, %r257, 0; + // begin inline asm + mov.u32 %r161, %r129; + @%p48 ld.global.b32 { %r161 }, [ %rd239 + 0 ]; + // end inline asm + shr.u32 %r258, %r242, 10; + and.b32 %r259, %r258, 1; + setp.ne.b32 %p49, %r259, 0; + // begin inline asm + mov.u32 %r162, %r129; + @%p49 ld.global.b32 { %r162 }, [ %rd240 + 0 ]; + // end inline asm + shr.u32 %r260, %r242, 9; + and.b32 %r261, %r260, 1; + setp.ne.b32 %p50, %r261, 0; + // begin inline asm + mov.u32 %r163, %r129; + @%p50 ld.global.b32 { %r163 }, [ %rd241 + 0 ]; + // end inline asm + shr.u32 %r262, %r242, 8; + and.b32 %r263, %r262, 1; + setp.ne.b32 %p51, %r263, 0; + // begin inline asm + mov.u32 %r164, %r129; + @%p51 ld.global.b32 { %r164 }, [ %rd242 + 0 ]; + // end inline asm + shr.u32 %r264, %r242, 7; + and.b32 %r265, %r264, 1; + setp.ne.b32 %p52, %r265, 0; + // begin inline asm + mov.u32 %r165, %r129; + @%p52 ld.global.b32 { %r165 }, [ %rd243 + 0 ]; + // end inline asm + shr.u32 %r266, %r242, 6; + and.b32 %r267, %r266, 1; + setp.ne.b32 %p53, %r267, 0; + // begin inline asm + mov.u32 %r166, %r129; + @%p53 ld.global.b32 { %r166 }, [ %rd244 + 0 ]; + // end inline asm + shr.u32 %r268, %r242, 5; + and.b32 %r269, %r268, 1; + setp.ne.b32 %p54, %r269, 0; + // begin inline asm + mov.u32 %r167, %r129; + @%p54 ld.global.b32 { %r167 }, [ %rd245 + 0 ]; + // end inline asm + shr.u32 %r270, %r242, 4; + and.b32 %r271, %r270, 1; + setp.ne.b32 %p55, %r271, 0; + // begin inline asm + mov.u32 %r168, %r129; + @%p55 ld.global.b32 { %r168 }, [ %rd246 + 0 ]; + // end inline asm + shr.u32 %r272, %r242, 3; + and.b32 %r273, %r272, 1; + setp.ne.b32 %p56, %r273, 0; + // begin inline asm + mov.u32 %r169, %r129; + @%p56 ld.global.b32 { %r169 }, [ %rd247 + 0 ]; + // end inline asm + shr.u32 %r274, %r242, 2; + and.b32 %r275, %r274, 1; + setp.ne.b32 %p57, %r275, 0; + // begin inline asm + mov.u32 %r170, %r129; + @%p57 ld.global.b32 { %r170 }, [ %rd248 + 0 ]; + // end inline asm + shr.u32 %r276, %r242, 1; + and.b32 %r277, %r276, 1; + setp.ne.b32 %p58, %r277, 0; + // begin inline asm + mov.u32 %r171, %r129; + @%p58 ld.global.b32 { %r171 }, [ %rd249 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r172, %r129; + @%p59 ld.global.b32 { %r172 }, [ %rd250 + 0 ]; + // end inline asm +$L__tmp5: + .loc 2 47 33 // op.py:47:33 @[ chunk_o.py:138:25 ] + ex2.approx.ftz.f32 %r14, %r140; +$L__tmp6: + .loc 1 139 40 // chunk_o.py:139:40 + sub.f32 %r278, %r140, %r141; + sub.f32 %r279, %r140, %r142; + sub.f32 %r280, %r140, %r143; + sub.f32 %r281, %r140, %r144; + sub.f32 %r282, %r140, %r145; + sub.f32 %r283, %r140, %r146; + sub.f32 %r284, %r140, %r147; + sub.f32 %r285, %r140, %r148; + sub.f32 %r286, %r140, %r149; + sub.f32 %r287, %r140, %r150; + sub.f32 %r288, %r140, %r151; + sub.f32 %r289, %r140, %r152; + sub.f32 %r290, %r140, %r153; + sub.f32 %r291, %r140, %r154; + sub.f32 %r292, %r140, %r155; + sub.f32 %r293, %r140, %r156; + sub.f32 %r294, %r140, %r157; + sub.f32 %r295, %r140, %r158; + sub.f32 %r296, %r140, %r159; + sub.f32 %r297, %r140, %r160; + sub.f32 %r298, %r140, %r161; + sub.f32 %r299, %r140, %r162; + sub.f32 %r300, %r140, %r163; + sub.f32 %r301, %r140, %r164; + sub.f32 %r302, %r140, %r165; + sub.f32 %r303, %r140, %r166; + sub.f32 %r304, %r140, %r167; + sub.f32 %r305, %r140, %r168; + sub.f32 %r306, %r140, %r169; + sub.f32 %r307, %r140, %r170; + sub.f32 %r308, %r140, %r171; + sub.f32 %r309, %r140, %r172; +$L__tmp7: + .loc 2 47 33 // op.py:47:33 @[ chunk_o.py:139:25 ] + ex2.approx.ftz.f32 %r310, %r278; + ex2.approx.ftz.f32 %r311, %r279; + ex2.approx.ftz.f32 %r312, %r280; + ex2.approx.ftz.f32 %r313, %r281; + ex2.approx.ftz.f32 %r314, %r282; + ex2.approx.ftz.f32 %r315, %r283; + ex2.approx.ftz.f32 %r316, %r284; + ex2.approx.ftz.f32 %r317, %r285; + ex2.approx.ftz.f32 %r318, %r286; + ex2.approx.ftz.f32 %r319, %r287; + ex2.approx.ftz.f32 %r320, %r288; + ex2.approx.ftz.f32 %r321, %r289; + ex2.approx.ftz.f32 %r322, %r290; + ex2.approx.ftz.f32 %r323, %r291; + ex2.approx.ftz.f32 %r324, %r292; + ex2.approx.ftz.f32 %r325, %r293; + ex2.approx.ftz.f32 %r326, %r294; + ex2.approx.ftz.f32 %r327, %r295; + ex2.approx.ftz.f32 %r328, %r296; + ex2.approx.ftz.f32 %r329, %r297; + ex2.approx.ftz.f32 %r330, %r298; + ex2.approx.ftz.f32 %r331, %r299; + ex2.approx.ftz.f32 %r332, %r300; + ex2.approx.ftz.f32 %r333, %r301; + ex2.approx.ftz.f32 %r334, %r302; + ex2.approx.ftz.f32 %r335, %r303; + ex2.approx.ftz.f32 %r336, %r304; + ex2.approx.ftz.f32 %r337, %r305; + ex2.approx.ftz.f32 %r338, %r306; + ex2.approx.ftz.f32 %r339, %r307; + ex2.approx.ftz.f32 %r340, %r308; + ex2.approx.ftz.f32 %r341, %r309; +$L__tmp8: + .loc 1 145 27 // chunk_o.py:145:27 + setp.ge.u32 %p80, %r5, %r7; + setp.gt.u32 %p81, %r5, %r7; + setp.ge.u32 %p82, %r5, %r9; + setp.ge.u32 %p83, %r5, %r8; + mov.b64 {%r342, %r343}, %rd11; + setp.ge.u32 %p84, %r5, %r343; + setp.ge.u32 %p85, %r5, %r342; + mov.b64 {%r344, %r345}, %rd10; + setp.ge.u32 %p86, %r5, %r345; + setp.ge.u32 %p87, %r5, %r344; + mov.b64 {%r346, %r347}, %rd15; + setp.ge.u32 %p88, %r5, %r347; + setp.ge.u32 %p89, %r5, %r346; + mov.b64 {%r348, %r349}, %rd14; + setp.ge.u32 %p90, %r5, %r349; + setp.ge.u32 %p91, %r5, %r348; + mov.b64 {%r350, %r351}, %rd13; + setp.ge.u32 %p92, %r5, %r351; + setp.ge.u32 %p93, %r5, %r350; + mov.b64 {%r352, %r353}, %rd12; + setp.ge.u32 %p94, %r5, %r353; + setp.ge.u32 %p95, %r5, %r352; + mov.b64 {%r354, %r355}, %rd23; + setp.ge.u32 %p96, %r5, %r355; + setp.ge.u32 %p97, %r5, %r354; + mov.b64 {%r356, %r357}, %rd22; + setp.ge.u32 %p98, %r5, %r357; + setp.ge.u32 %p99, %r5, %r356; + mov.b64 {%r358, %r359}, %rd21; + setp.ge.u32 %p100, %r5, %r359; + setp.ge.u32 %p101, %r5, %r358; + mov.b64 {%r360, %r361}, %rd20; + setp.ge.u32 %p102, %r5, %r361; + setp.ge.u32 %p103, %r5, %r360; + mov.b64 {%r362, %r363}, %rd19; + setp.ge.u32 %p104, %r5, %r363; + setp.ge.u32 %p105, %r5, %r362; + mov.b64 {%r364, %r365}, %rd18; + setp.ge.u32 %p106, %r5, %r365; + setp.ge.u32 %p107, %r5, %r364; + mov.b64 {%r366, %r367}, %rd17; + setp.ge.u32 %p108, %r5, %r367; + setp.ge.u32 %p109, %r5, %r366; + mov.b64 {%r368, %r369}, %rd16; + setp.ge.u32 %p110, %r5, %r369; + setp.ge.u32 %p111, %r5, %r368; + .loc 1 148 14 // chunk_o.py:148:14 + shl.b64 %rd291, %rd58, 1; + add.s64 %rd292, %rd3, %rd291; + shl.b64 %rd293, %rd59, 1; + add.s64 %rd294, %rd3, %rd293; + shl.b64 %rd295, %rd60, 1; + add.s64 %rd296, %rd3, %rd295; + shl.b64 %rd297, %rd61, 1; + add.s64 %rd298, %rd3, %rd297; + .loc 1 148 38 // chunk_o.py:148:38 + add.s64 %rd251, %rd292, %rd162; + add.s64 %rd252, %rd294, %rd162; + add.s64 %rd253, %rd296, %rd162; + add.s64 %rd254, %rd298, %rd162; + .loc 1 151 18 // chunk_o.py:151:18 + // begin inline asm + mov.u32 %r173, %r129; + mov.u32 %r174, %r129; + mov.u32 %r175, %r129; + mov.u32 %r176, %r129; + @%p6 ld.global.v4.b32 { %r173, %r174, %r175, %r176 }, [ %rd251 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r177, %r129; + mov.u32 %r178, %r129; + mov.u32 %r179, %r129; + mov.u32 %r180, %r129; + @%p7 ld.global.v4.b32 { %r177, %r178, %r179, %r180 }, [ %rd252 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r181, %r129; + mov.u32 %r182, %r129; + mov.u32 %r183, %r129; + mov.u32 %r184, %r129; + @%p8 ld.global.v4.b32 { %r181, %r182, %r183, %r184 }, [ %rd253 + 0 ]; + // end inline asm + // begin inline asm + mov.u32 %r185, %r129; + mov.u32 %r186, %r129; + mov.u32 %r187, %r129; + mov.u32 %r188, %r129; + @%p9 ld.global.v4.b32 { %r185, %r186, %r187, %r188 }, [ %rd254 + 0 ]; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r189 + 0 ], { %r173, %r174, %r175, %r176 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r190 + 0 ], { %r177, %r178, %r179, %r180 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r191 + 0 ], { %r181, %r182, %r183, %r184 }; + // end inline asm + // begin inline asm + st.shared.v4.b32 [ %r192 + 0 ], { %r185, %r186, %r187, %r188 }; + // end inline asm + .loc 1 132 31 // chunk_o.py:132:31 + // begin inline asm + tcgen05.ld.sync.aligned.16x32bx2.x32.b32 {%r193, %r194, %r195, %r196, %r197, %r198, %r199, %r200, %r201, %r202, %r203, %r204, %r205, %r206, %r207, %r208, %r209, %r210, %r211, %r212, %r213, %r214, %r215, %r216, %r217, %r218, %r219, %r220, %r221, %r222, %r223, %r224}, [%r474 + 0], 32; + // end inline asm + tcgen05.wait::ld.sync.aligned; + .loc 1 139 20 // chunk_o.py:139:20 + mul.f32 %r370, %r310, %r193; + mul.f32 %r371, %r311, %r194; + .loc 1 154 38 // chunk_o.py:154:38 + cvt.rn.bf16.f32 %rs54, %r370; + .loc 1 146 29 // chunk_o.py:146:29 + selp.b16 %rs55, %rs54, 0x0000, %p28; + selp.b16 %rs56, %rs55, 0x0000, %p80; + selp.b16 %rs57, %rs56, 0x0000, %p27; + .loc 1 154 38 // chunk_o.py:154:38 + cvt.rn.bf16.f32 %rs58, %r371; + .loc 1 146 29 // chunk_o.py:146:29 + selp.b16 %rs59, %rs58, 0x0000, %p29; + selp.b16 %rs60, %rs59, 0x0000, %p81; + selp.b16 %rs61, %rs60, 0x0000, %p27; + .loc 1 132 31 // chunk_o.py:132:31 + mov.b64 %rd299, {%r195, %r196}; + .loc 1 139 20 // chunk_o.py:139:20 + mov.b64 %rd300, {%r312, %r313}; + mul.f32x2 %rd301, %rd300, %rd299; + .loc 1 154 38 // chunk_o.py:154:38 + mov.b64 {%r372, %r373}, %rd301; + cvt.rn.bf16.f32 %rs62, %r372; + cvt.rn.bf16.f32 %rs63, %r373; + .loc 1 146 29 // chunk_o.py:146:29 + selp.b16 %rs64, %rs63, 0x0000, %p31; + selp.b16 %rs65, %rs64, 0x0000, %p83; + selp.b16 %rs66, %rs65, 0x0000, %p27; + selp.b16 %rs67, %rs62, 0x0000, %p30; + selp.b16 %rs68, %rs67, 0x0000, %p82; + selp.b16 %rs69, %rs68, 0x0000, %p27; + mov.b32 %r227, {%rs69, %rs66}; + .loc 1 132 31 // chunk_o.py:132:31 + mov.b64 %rd302, {%r197, %r198}; + .loc 1 139 20 // chunk_o.py:139:20 + mov.b64 %rd303, {%r314, %r315}; + mul.f32x2 %rd304, %rd303, %rd302; + .loc 1 154 38 // chunk_o.py:154:38 + mov.b64 {%r374, %r375}, %rd304; + cvt.rn.bf16.f32 %rs70, %r374; + cvt.rn.bf16.f32 %rs71, %r375; + .loc 1 146 29 // chunk_o.py:146:29 + selp.b16 %rs72, %rs71, 0x0000, %p33; + selp.b16 %rs73, %rs72, 0x0000, %p85; + selp.b16 %rs74, %rs73, 0x0000, %p27; + selp.b16 %rs75, %rs70, 0x0000, %p32; + selp.b16 %rs76, %rs75, 0x0000, %p84; + selp.b16 %rs77, %rs76, 0x0000, %p27; + mov.b32 %r228, {%rs77, %rs74}; + .loc 1 132 31 // chunk_o.py:132:31 + mov.b64 %rd305, {%r199, %r200}; + .loc 1 139 20 // chunk_o.py:139:20 + mov.b64 %rd306, {%r316, %r317}; + mul.f32x2 %rd307, %rd306, %rd305; + .loc 1 154 38 // chunk_o.py:154:38 + mov.b64 {%r376, %r377}, %rd307; + cvt.rn.bf16.f32 %rs78, %r376; + cvt.rn.bf16.f32 %rs79, %r377; + .loc 1 146 29 // chunk_o.py:146:29 + selp.b16 %rs80, %rs79, 0x0000, %p35; + selp.b16 %rs81, %rs80, 0x0000, %p87; + selp.b16 %rs82, %rs81, 0x0000, %p27; + selp.b16 %rs83, %rs78, 0x0000, %p34; + selp.b16 %rs84, %rs83, 0x0000, %p86; + selp.b16 %rs85, %rs84, 0x0000, %p27; + mov.b32 %r229, {%rs85, %rs82}; + .loc 1 132 31 // chunk_o.py:132:31 + mov.b64 %rd308, {%r201, %r202}; + .loc 1 139 20 // chunk_o.py:139:20 + mov.b64 %rd309, {%r318, %r319}; + mul.f32x2 %rd310, %rd309, %rd308; + .loc 1 154 38 // chunk_o.py:154:38 + mov.b64 {%r378, %r379}, %rd310; + cvt.rn.bf16.f32 %rs86, %r378; + cvt.rn.bf16.f32 %rs87, %r379; + .loc 1 146 29 // chunk_o.py:146:29 + selp.b16 %rs88, %rs87, 0x0000, %p37; + selp.b16 %rs89, %rs88, 0x0000, %p89; + selp.b16 %rs90, %rs89, 0x0000, %p27; + selp.b16 %rs91, %rs86, 0x0000, %p36; + selp.b16 %rs92, %rs91, 0x0000, %p88; + selp.b16 %rs93, %rs92, 0x0000, %p27; + mov.b32 %r230, {%rs93, %rs90}; + .loc 1 132 31 // chunk_o.py:132:31 + mov.b64 %rd311, {%r203, %r204}; + .loc 1 139 20 // chunk_o.py:139:20 + mov.b64 %rd312, {%r320, %r321}; + mul.f32x2 %rd313, %rd312, %rd311; + .loc 1 154 38 // chunk_o.py:154:38 + mov.b64 {%r380, %r381}, %rd313; + cvt.rn.bf16.f32 %rs94, %r380; + cvt.rn.bf16.f32 %rs95, %r381; + .loc 1 146 29 // chunk_o.py:146:29 + selp.b16 %rs96, %rs95, 0x0000, %p39; + selp.b16 %rs97, %rs96, 0x0000, %p91; + selp.b16 %rs98, %rs97, 0x0000, %p27; + selp.b16 %rs99, %rs94, 0x0000, %p38; + selp.b16 %rs100, %rs99, 0x0000, %p90; + selp.b16 %rs101, %rs100, 0x0000, %p27; + mov.b32 %r231, {%rs101, %rs98}; + .loc 1 132 31 // chunk_o.py:132:31 + mov.b64 %rd314, {%r205, %r206}; + .loc 1 139 20 // chunk_o.py:139:20 + mov.b64 %rd315, {%r322, %r323}; + mul.f32x2 %rd316, %rd315, %rd314; + .loc 1 154 38 // chunk_o.py:154:38 + mov.b64 {%r382, %r383}, %rd316; + cvt.rn.bf16.f32 %rs102, %r382; + cvt.rn.bf16.f32 %rs103, %r383; + .loc 1 146 29 // chunk_o.py:146:29 + selp.b16 %rs104, %rs103, 0x0000, %p41; + selp.b16 %rs105, %rs104, 0x0000, %p93; + selp.b16 %rs106, %rs105, 0x0000, %p27; + selp.b16 %rs107, %rs102, 0x0000, %p40; + selp.b16 %rs108, %rs107, 0x0000, %p92; + selp.b16 %rs109, %rs108, 0x0000, %p27; + mov.b32 %r232, {%rs109, %rs106}; + .loc 1 132 31 // chunk_o.py:132:31 + mov.b64 %rd317, {%r207, %r208}; + .loc 1 139 20 // chunk_o.py:139:20 + mov.b64 %rd318, {%r324, %r325}; + mul.f32x2 %rd319, %rd318, %rd317; + .loc 1 154 38 // chunk_o.py:154:38 + mov.b64 {%r384, %r385}, %rd319; + cvt.rn.bf16.f32 %rs110, %r384; + cvt.rn.bf16.f32 %rs111, %r385; + .loc 1 146 29 // chunk_o.py:146:29 + selp.b16 %rs112, %rs111, 0x0000, %p43; + selp.b16 %rs113, %rs112, 0x0000, %p95; + selp.b16 %rs114, %rs113, 0x0000, %p27; + selp.b16 %rs115, %rs110, 0x0000, %p42; + selp.b16 %rs116, %rs115, 0x0000, %p94; + selp.b16 %rs117, %rs116, 0x0000, %p27; + mov.b32 %r233, {%rs117, %rs114}; + .loc 1 132 31 // chunk_o.py:132:31 + mov.b64 %rd320, {%r209, %r210}; + .loc 1 139 20 // chunk_o.py:139:20 + mov.b64 %rd321, {%r326, %r327}; + mul.f32x2 %rd322, %rd321, %rd320; + .loc 1 154 38 // chunk_o.py:154:38 + mov.b64 {%r386, %r387}, %rd322; + cvt.rn.bf16.f32 %rs118, %r386; + cvt.rn.bf16.f32 %rs119, %r387; + .loc 1 146 29 // chunk_o.py:146:29 + selp.b16 %rs120, %rs119, 0x0000, %p45; + selp.b16 %rs121, %rs120, 0x0000, %p97; + selp.b16 %rs122, %rs121, 0x0000, %p27; + selp.b16 %rs123, %rs118, 0x0000, %p44; + selp.b16 %rs124, %rs123, 0x0000, %p96; + selp.b16 %rs125, %rs124, 0x0000, %p27; + mov.b32 %r234, {%rs125, %rs122}; + .loc 1 132 31 // chunk_o.py:132:31 + mov.b64 %rd323, {%r211, %r212}; + .loc 1 139 20 // chunk_o.py:139:20 + mov.b64 %rd324, {%r328, %r329}; + mul.f32x2 %rd325, %rd324, %rd323; + .loc 1 154 38 // chunk_o.py:154:38 + mov.b64 {%r388, %r389}, %rd325; + cvt.rn.bf16.f32 %rs126, %r388; + cvt.rn.bf16.f32 %rs127, %r389; + .loc 1 146 29 // chunk_o.py:146:29 + selp.b16 %rs128, %rs127, 0x0000, %p47; + selp.b16 %rs129, %rs128, 0x0000, %p99; + selp.b16 %rs130, %rs129, 0x0000, %p27; + selp.b16 %rs131, %rs126, 0x0000, %p46; + selp.b16 %rs132, %rs131, 0x0000, %p98; + selp.b16 %rs133, %rs132, 0x0000, %p27; + mov.b32 %r235, {%rs133, %rs130}; + .loc 1 132 31 // chunk_o.py:132:31 + mov.b64 %rd326, {%r213, %r214}; + .loc 1 139 20 // chunk_o.py:139:20 + mov.b64 %rd327, {%r330, %r331}; + mul.f32x2 %rd328, %rd327, %rd326; + .loc 1 154 38 // chunk_o.py:154:38 + mov.b64 {%r390, %r391}, %rd328; + cvt.rn.bf16.f32 %rs134, %r390; + cvt.rn.bf16.f32 %rs135, %r391; + .loc 1 146 29 // chunk_o.py:146:29 + selp.b16 %rs136, %rs135, 0x0000, %p49; + selp.b16 %rs137, %rs136, 0x0000, %p101; + selp.b16 %rs138, %rs137, 0x0000, %p27; + selp.b16 %rs139, %rs134, 0x0000, %p48; + selp.b16 %rs140, %rs139, 0x0000, %p100; + selp.b16 %rs141, %rs140, 0x0000, %p27; + mov.b32 %r236, {%rs141, %rs138}; + .loc 1 132 31 // chunk_o.py:132:31 + mov.b64 %rd329, {%r215, %r216}; + .loc 1 139 20 // chunk_o.py:139:20 + mov.b64 %rd330, {%r332, %r333}; + mul.f32x2 %rd331, %rd330, %rd329; + .loc 1 154 38 // chunk_o.py:154:38 + mov.b64 {%r392, %r393}, %rd331; + cvt.rn.bf16.f32 %rs142, %r392; + cvt.rn.bf16.f32 %rs143, %r393; + .loc 1 146 29 // chunk_o.py:146:29 + selp.b16 %rs144, %rs143, 0x0000, %p51; + selp.b16 %rs145, %rs144, 0x0000, %p103; + selp.b16 %rs146, %rs145, 0x0000, %p27; + selp.b16 %rs147, %rs142, 0x0000, %p50; + selp.b16 %rs148, %rs147, 0x0000, %p102; + selp.b16 %rs149, %rs148, 0x0000, %p27; + mov.b32 %r237, {%rs149, %rs146}; + .loc 1 132 31 // chunk_o.py:132:31 + mov.b64 %rd332, {%r217, %r218}; + .loc 1 139 20 // chunk_o.py:139:20 + mov.b64 %rd333, {%r334, %r335}; + mul.f32x2 %rd334, %rd333, %rd332; + .loc 1 154 38 // chunk_o.py:154:38 + mov.b64 {%r394, %r395}, %rd334; + cvt.rn.bf16.f32 %rs150, %r394; + cvt.rn.bf16.f32 %rs151, %r395; + .loc 1 146 29 // chunk_o.py:146:29 + selp.b16 %rs152, %rs151, 0x0000, %p53; + selp.b16 %rs153, %rs152, 0x0000, %p105; + selp.b16 %rs154, %rs153, 0x0000, %p27; + selp.b16 %rs155, %rs150, 0x0000, %p52; + selp.b16 %rs156, %rs155, 0x0000, %p104; + selp.b16 %rs157, %rs156, 0x0000, %p27; + mov.b32 %r238, {%rs157, %rs154}; + .loc 1 132 31 // chunk_o.py:132:31 + mov.b64 %rd335, {%r219, %r220}; + .loc 1 139 20 // chunk_o.py:139:20 + mov.b64 %rd336, {%r336, %r337}; + mul.f32x2 %rd337, %rd336, %rd335; + .loc 1 154 38 // chunk_o.py:154:38 + mov.b64 {%r396, %r397}, %rd337; + cvt.rn.bf16.f32 %rs158, %r396; + cvt.rn.bf16.f32 %rs159, %r397; + .loc 1 146 29 // chunk_o.py:146:29 + selp.b16 %rs160, %rs159, 0x0000, %p55; + selp.b16 %rs161, %rs160, 0x0000, %p107; + selp.b16 %rs162, %rs161, 0x0000, %p27; + selp.b16 %rs163, %rs158, 0x0000, %p54; + selp.b16 %rs164, %rs163, 0x0000, %p106; + selp.b16 %rs165, %rs164, 0x0000, %p27; + mov.b32 %r239, {%rs165, %rs162}; + .loc 1 132 31 // chunk_o.py:132:31 + mov.b64 %rd338, {%r221, %r222}; + .loc 1 139 20 // chunk_o.py:139:20 + mov.b64 %rd339, {%r338, %r339}; + mul.f32x2 %rd340, %rd339, %rd338; + .loc 1 154 38 // chunk_o.py:154:38 + mov.b64 {%r398, %r399}, %rd340; + cvt.rn.bf16.f32 %rs166, %r398; + cvt.rn.bf16.f32 %rs167, %r399; + .loc 1 146 29 // chunk_o.py:146:29 + selp.b16 %rs168, %rs167, 0x0000, %p57; + selp.b16 %rs169, %rs168, 0x0000, %p109; + selp.b16 %rs170, %rs169, 0x0000, %p27; + selp.b16 %rs171, %rs166, 0x0000, %p56; + selp.b16 %rs172, %rs171, 0x0000, %p108; + selp.b16 %rs173, %rs172, 0x0000, %p27; + mov.b32 %r240, {%rs173, %rs170}; + .loc 1 132 31 // chunk_o.py:132:31 + mov.b64 %rd341, {%r223, %r224}; + .loc 1 139 20 // chunk_o.py:139:20 + mov.b64 %rd342, {%r340, %r341}; + mul.f32x2 %rd343, %rd342, %rd341; + .loc 1 154 38 // chunk_o.py:154:38 + mov.b64 {%r400, %r401}, %rd343; + cvt.rn.bf16.f32 %rs174, %r401; + cvt.rn.bf16.f32 %rs175, %r400; + .loc 1 146 29 // chunk_o.py:146:29 + selp.b16 %rs176, %rs175, 0x0000, %p58; + selp.b16 %rs177, %rs176, 0x0000, %p110; + selp.b16 %rs178, %rs177, 0x0000, %p27; + selp.b16 %rs179, %rs174, 0x0000, %p59; + selp.b16 %rs180, %rs179, 0x0000, %p111; + selp.b16 %rs181, %rs180, 0x0000, %p27; + mov.b32 %r241, {%rs178, %rs181}; + mov.pred %p113, -1; + .loc 1 154 50 // chunk_o.py:154:50 + // begin inline asm + @%p113 tcgen05.st.sync.aligned.16x32bx2.x32.b32 [%r474 + 0], 32, {%r129, %r129, %r129, %r129, %r129, %r129, %r129, %r129, %r129, %r129, %r129, %r129, %r129, %r129, %r129, %r129, %r129, %r129, %r129, %r129, %r129, %r129, %r129, %r129, %r129, %r129, %r129, %r129, %r129, %r129, %r129, %r129}; + // end inline asm + tcgen05.wait::st.sync.aligned; + bar.sync 0; + mov.b32 %r226, {%rs57, %rs61}; + add.s32 %r225, %r441, 1048640; + // begin inline asm + @%p113 tcgen05.st.sync.aligned.16x32bx2.x16.b32 [%r225 + 0], 16, {%r226, %r227, %r228, %r229, %r230, %r231, %r232, %r233, %r234, %r235, %r236, %r237, %r238, %r239, %r240, %r241}; + // end inline asm + tcgen05.wait::st.sync.aligned; + bar.sync 0; + fence.proxy.async.shared::cta; + elect.sync %r402|%p112, -1; + and.pred %p60, %p10, %p112; + add.s32 %r408, %r16, 8192; + // begin inline asm + @%p60 mbarrier.init.shared::cta.b64 [%r408], 1; + // end inline asm + bar.sync 0; + @%p15 bra $L__BB0_7; +// %bb.6: + add.s32 %r404, %r508, 1048640; + elect.sync %r406|%p114, -1; + bfe.u32 %r407, %r16, 4, 14; + cvt.u64.u32 %rd348, %r407; + or.b64 %rd344, %rd348, 4611756662049472512; + mov.b32 %r405, 68224144; + // begin inline asm + @%p114 tcgen05.mma.cta_group::1.kind::f16 [ %r403 + 0 ], [ %r404 + 0 ], %rd344, %r405, %p113; + // end inline asm + add.s64 %rd345, %rd348, 4611756662049472640; + // begin inline asm + @%p114 tcgen05.mma.cta_group::1.kind::f16 [ %r403 + 0 ], [ %r404 + 8 ], %rd345, %r405, %p113; + // end inline asm + add.s64 %rd346, %rd348, 4611756662049472768; + // begin inline asm + @%p114 tcgen05.mma.cta_group::1.kind::f16 [ %r403 + 0 ], [ %r404 + 16 ], %rd346, %r405, %p113; + // end inline asm + add.s64 %rd347, %rd348, 4611756662049472896; + // begin inline asm + @%p114 tcgen05.mma.cta_group::1.kind::f16 [ %r403 + 0 ], [ %r404 + 24 ], %rd347, %r405, %p113; + // end inline asm + // begin inline asm + @%p114 tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [%r408]; + // end inline asm +$L__BB0_7: + .loc 1 100 9 // chunk_o.py:100:9 + shl.b64 %rd353, %rd2, 1; + add.s64 %rd354, %rd63, %rd353; + .loc 1 149 14 // chunk_o.py:149:14 + add.s64 %rd355, %rd354, %rd297; + .loc 1 149 38 // chunk_o.py:149:38 + add.s64 %rd352, %rd355, %rd162; + .loc 1 149 14 // chunk_o.py:149:14 + add.s64 %rd356, %rd354, %rd295; + .loc 1 149 38 // chunk_o.py:149:38 + add.s64 %rd351, %rd356, %rd162; + .loc 1 149 14 // chunk_o.py:149:14 + add.s64 %rd357, %rd354, %rd293; + .loc 1 149 38 // chunk_o.py:149:38 + add.s64 %rd350, %rd357, %rd162; + .loc 1 149 14 // chunk_o.py:149:14 + add.s64 %rd358, %rd354, %rd291; + .loc 1 149 38 // chunk_o.py:149:38 + add.s64 %rd349, %rd358, %rd162; + .loc 1 154 50 // chunk_o.py:154:50 + // begin inline asm + +{ + .reg .pred complete; + waitLoop: + mbarrier.try_wait.parity.shared::cta.b64 complete, [%r408], %r129; + @!complete bra.uni waitLoop; +} + + // end inline asm + bar.sync 0; + elect.sync %r509|%p116, -1; + and.pred %p115, %p10, %p116; + // begin inline asm + @%p115 mbarrier.inval.shared::cta.b64 [%r408]; + // end inline asm + .loc 1 130 35 // chunk_o.py:130:35 + // begin inline asm + tcgen05.ld.sync.aligned.16x32bx2.x32.b32 {%r409, %r410, %r411, %r412, %r413, %r414, %r415, %r416, %r417, %r418, %r419, %r420, %r421, %r422, %r423, %r424, %r425, %r426, %r427, %r428, %r429, %r430, %r431, %r432, %r433, %r434, %r435, %r436, %r437, %r438, %r439, %r440}, [%r441 + 0], 32; + // end inline asm + tcgen05.wait::ld.sync.aligned; + .loc 1 154 50 // chunk_o.py:154:50 + // begin inline asm + tcgen05.ld.sync.aligned.16x32bx2.x32.b32 {%r442, %r443, %r444, %r445, %r446, %r447, %r448, %r449, %r450, %r451, %r452, %r453, %r454, %r455, %r456, %r457, %r458, %r459, %r460, %r461, %r462, %r463, %r464, %r465, %r466, %r467, %r468, %r469, %r470, %r471, %r472, %r473}, [%r474 + 0], 32; + // end inline asm + tcgen05.wait::ld.sync.aligned; + .loc 1 130 35 // chunk_o.py:130:35 + mov.b64 %rd359, {%r409, %r410}; + .loc 1 138 20 // chunk_o.py:138:20 + mov.b64 %rd360, {%r14, %r14}; + mul.f32x2 %rd361, %rd360, %rd359; + .loc 1 154 16 // chunk_o.py:154:16 + mov.b64 %rd362, {%r15, %r15}; + .loc 1 154 50 // chunk_o.py:154:50 + mov.b64 %rd363, {%r442, %r443}; + .loc 1 154 57 // chunk_o.py:154:57 + mul.f32x2 %rd364, %rd362, %rd363; + .loc 1 154 24 // chunk_o.py:154:24 + fma.rn.f32x2 %rd365, %rd362, %rd361, %rd364; + .loc 1 155 25 // chunk_o.py:155:25 + mov.b64 {%r510, %r511}, %rd365; + cvt.rn.bf16x2.f32 %r476, %r511, %r510; + .loc 1 130 35 // chunk_o.py:130:35 + mov.b64 %rd366, {%r411, %r412}; + .loc 1 138 20 // chunk_o.py:138:20 + mul.f32x2 %rd367, %rd360, %rd366; + .loc 1 154 50 // chunk_o.py:154:50 + mov.b64 %rd368, {%r444, %r445}; + .loc 1 154 57 // chunk_o.py:154:57 + mul.f32x2 %rd369, %rd362, %rd368; + .loc 1 154 24 // chunk_o.py:154:24 + fma.rn.f32x2 %rd370, %rd362, %rd367, %rd369; + .loc 1 155 25 // chunk_o.py:155:25 + mov.b64 {%r512, %r513}, %rd370; + cvt.rn.bf16x2.f32 %r480, %r513, %r512; + .loc 1 130 35 // chunk_o.py:130:35 + mov.b64 %rd371, {%r413, %r414}; + .loc 1 138 20 // chunk_o.py:138:20 + mul.f32x2 %rd372, %rd360, %rd371; + .loc 1 154 50 // chunk_o.py:154:50 + mov.b64 %rd373, {%r446, %r447}; + .loc 1 154 57 // chunk_o.py:154:57 + mul.f32x2 %rd374, %rd362, %rd373; + .loc 1 154 24 // chunk_o.py:154:24 + fma.rn.f32x2 %rd375, %rd362, %rd372, %rd374; + .loc 1 155 25 // chunk_o.py:155:25 + mov.b64 {%r514, %r515}, %rd375; + cvt.rn.bf16x2.f32 %r484, %r515, %r514; + .loc 1 130 35 // chunk_o.py:130:35 + mov.b64 %rd376, {%r415, %r416}; + .loc 1 138 20 // chunk_o.py:138:20 + mul.f32x2 %rd377, %rd360, %rd376; + .loc 1 154 50 // chunk_o.py:154:50 + mov.b64 %rd378, {%r448, %r449}; + .loc 1 154 57 // chunk_o.py:154:57 + mul.f32x2 %rd379, %rd362, %rd378; + .loc 1 154 24 // chunk_o.py:154:24 + fma.rn.f32x2 %rd380, %rd362, %rd377, %rd379; + .loc 1 155 25 // chunk_o.py:155:25 + mov.b64 {%r516, %r517}, %rd380; + cvt.rn.bf16x2.f32 %r488, %r517, %r516; + .loc 1 130 35 // chunk_o.py:130:35 + mov.b64 %rd381, {%r417, %r418}; + .loc 1 138 20 // chunk_o.py:138:20 + mul.f32x2 %rd382, %rd360, %rd381; + .loc 1 154 50 // chunk_o.py:154:50 + mov.b64 %rd383, {%r450, %r451}; + .loc 1 154 57 // chunk_o.py:154:57 + mul.f32x2 %rd384, %rd362, %rd383; + .loc 1 154 24 // chunk_o.py:154:24 + fma.rn.f32x2 %rd385, %rd362, %rd382, %rd384; + .loc 1 155 25 // chunk_o.py:155:25 + mov.b64 {%r518, %r519}, %rd385; + cvt.rn.bf16x2.f32 %r477, %r519, %r518; + .loc 1 130 35 // chunk_o.py:130:35 + mov.b64 %rd386, {%r419, %r420}; + .loc 1 138 20 // chunk_o.py:138:20 + mul.f32x2 %rd387, %rd360, %rd386; + .loc 1 154 50 // chunk_o.py:154:50 + mov.b64 %rd388, {%r452, %r453}; + .loc 1 154 57 // chunk_o.py:154:57 + mul.f32x2 %rd389, %rd362, %rd388; + .loc 1 154 24 // chunk_o.py:154:24 + fma.rn.f32x2 %rd390, %rd362, %rd387, %rd389; + .loc 1 155 25 // chunk_o.py:155:25 + mov.b64 {%r520, %r521}, %rd390; + cvt.rn.bf16x2.f32 %r481, %r521, %r520; + .loc 1 130 35 // chunk_o.py:130:35 + mov.b64 %rd391, {%r421, %r422}; + .loc 1 138 20 // chunk_o.py:138:20 + mul.f32x2 %rd392, %rd360, %rd391; + .loc 1 154 50 // chunk_o.py:154:50 + mov.b64 %rd393, {%r454, %r455}; + .loc 1 154 57 // chunk_o.py:154:57 + mul.f32x2 %rd394, %rd362, %rd393; + .loc 1 154 24 // chunk_o.py:154:24 + fma.rn.f32x2 %rd395, %rd362, %rd392, %rd394; + .loc 1 155 25 // chunk_o.py:155:25 + mov.b64 {%r522, %r523}, %rd395; + cvt.rn.bf16x2.f32 %r485, %r523, %r522; + .loc 1 130 35 // chunk_o.py:130:35 + mov.b64 %rd396, {%r423, %r424}; + .loc 1 138 20 // chunk_o.py:138:20 + mul.f32x2 %rd397, %rd360, %rd396; + .loc 1 154 50 // chunk_o.py:154:50 + mov.b64 %rd398, {%r456, %r457}; + .loc 1 154 57 // chunk_o.py:154:57 + mul.f32x2 %rd399, %rd362, %rd398; + .loc 1 154 24 // chunk_o.py:154:24 + fma.rn.f32x2 %rd400, %rd362, %rd397, %rd399; + .loc 1 155 25 // chunk_o.py:155:25 + mov.b64 {%r524, %r525}, %rd400; + cvt.rn.bf16x2.f32 %r489, %r525, %r524; + .loc 1 130 35 // chunk_o.py:130:35 + mov.b64 %rd401, {%r425, %r426}; + .loc 1 138 20 // chunk_o.py:138:20 + mul.f32x2 %rd402, %rd360, %rd401; + .loc 1 154 50 // chunk_o.py:154:50 + mov.b64 %rd403, {%r458, %r459}; + .loc 1 154 57 // chunk_o.py:154:57 + mul.f32x2 %rd404, %rd362, %rd403; + .loc 1 154 24 // chunk_o.py:154:24 + fma.rn.f32x2 %rd405, %rd362, %rd402, %rd404; + .loc 1 155 25 // chunk_o.py:155:25 + mov.b64 {%r526, %r527}, %rd405; + cvt.rn.bf16x2.f32 %r478, %r527, %r526; + .loc 1 130 35 // chunk_o.py:130:35 + mov.b64 %rd406, {%r427, %r428}; + .loc 1 138 20 // chunk_o.py:138:20 + mul.f32x2 %rd407, %rd360, %rd406; + .loc 1 154 50 // chunk_o.py:154:50 + mov.b64 %rd408, {%r460, %r461}; + .loc 1 154 57 // chunk_o.py:154:57 + mul.f32x2 %rd409, %rd362, %rd408; + .loc 1 154 24 // chunk_o.py:154:24 + fma.rn.f32x2 %rd410, %rd362, %rd407, %rd409; + .loc 1 155 25 // chunk_o.py:155:25 + mov.b64 {%r528, %r529}, %rd410; + cvt.rn.bf16x2.f32 %r482, %r529, %r528; + .loc 1 130 35 // chunk_o.py:130:35 + mov.b64 %rd411, {%r429, %r430}; + .loc 1 138 20 // chunk_o.py:138:20 + mul.f32x2 %rd412, %rd360, %rd411; + .loc 1 154 50 // chunk_o.py:154:50 + mov.b64 %rd413, {%r462, %r463}; + .loc 1 154 57 // chunk_o.py:154:57 + mul.f32x2 %rd414, %rd362, %rd413; + .loc 1 154 24 // chunk_o.py:154:24 + fma.rn.f32x2 %rd415, %rd362, %rd412, %rd414; + .loc 1 155 25 // chunk_o.py:155:25 + mov.b64 {%r530, %r531}, %rd415; + cvt.rn.bf16x2.f32 %r486, %r531, %r530; + .loc 1 130 35 // chunk_o.py:130:35 + mov.b64 %rd416, {%r431, %r432}; + .loc 1 138 20 // chunk_o.py:138:20 + mul.f32x2 %rd417, %rd360, %rd416; + .loc 1 154 50 // chunk_o.py:154:50 + mov.b64 %rd418, {%r464, %r465}; + .loc 1 154 57 // chunk_o.py:154:57 + mul.f32x2 %rd419, %rd362, %rd418; + .loc 1 154 24 // chunk_o.py:154:24 + fma.rn.f32x2 %rd420, %rd362, %rd417, %rd419; + .loc 1 155 25 // chunk_o.py:155:25 + mov.b64 {%r532, %r533}, %rd420; + cvt.rn.bf16x2.f32 %r490, %r533, %r532; + .loc 1 130 35 // chunk_o.py:130:35 + mov.b64 %rd421, {%r433, %r434}; + .loc 1 138 20 // chunk_o.py:138:20 + mul.f32x2 %rd422, %rd360, %rd421; + .loc 1 154 50 // chunk_o.py:154:50 + mov.b64 %rd423, {%r466, %r467}; + .loc 1 154 57 // chunk_o.py:154:57 + mul.f32x2 %rd424, %rd362, %rd423; + .loc 1 154 24 // chunk_o.py:154:24 + fma.rn.f32x2 %rd425, %rd362, %rd422, %rd424; + .loc 1 155 25 // chunk_o.py:155:25 + mov.b64 {%r534, %r535}, %rd425; + cvt.rn.bf16x2.f32 %r479, %r535, %r534; + .loc 1 130 35 // chunk_o.py:130:35 + mov.b64 %rd426, {%r435, %r436}; + .loc 1 138 20 // chunk_o.py:138:20 + mul.f32x2 %rd427, %rd360, %rd426; + .loc 1 154 50 // chunk_o.py:154:50 + mov.b64 %rd428, {%r468, %r469}; + .loc 1 154 57 // chunk_o.py:154:57 + mul.f32x2 %rd429, %rd362, %rd428; + .loc 1 154 24 // chunk_o.py:154:24 + fma.rn.f32x2 %rd430, %rd362, %rd427, %rd429; + .loc 1 155 25 // chunk_o.py:155:25 + mov.b64 {%r536, %r537}, %rd430; + cvt.rn.bf16x2.f32 %r483, %r537, %r536; + .loc 1 130 35 // chunk_o.py:130:35 + mov.b64 %rd431, {%r437, %r438}; + .loc 1 138 20 // chunk_o.py:138:20 + mul.f32x2 %rd432, %rd360, %rd431; + .loc 1 154 50 // chunk_o.py:154:50 + mov.b64 %rd433, {%r470, %r471}; + .loc 1 154 57 // chunk_o.py:154:57 + mul.f32x2 %rd434, %rd362, %rd433; + .loc 1 154 24 // chunk_o.py:154:24 + fma.rn.f32x2 %rd435, %rd362, %rd432, %rd434; + .loc 1 155 25 // chunk_o.py:155:25 + mov.b64 {%r538, %r539}, %rd435; + cvt.rn.bf16x2.f32 %r487, %r539, %r538; + .loc 1 130 35 // chunk_o.py:130:35 + mov.b64 %rd436, {%r439, %r440}; + .loc 1 138 20 // chunk_o.py:138:20 + mul.f32x2 %rd437, %rd360, %rd436; + .loc 1 154 50 // chunk_o.py:154:50 + mov.b64 %rd438, {%r472, %r473}; + .loc 1 154 57 // chunk_o.py:154:57 + mul.f32x2 %rd439, %rd362, %rd438; + .loc 1 154 24 // chunk_o.py:154:24 + fma.rn.f32x2 %rd440, %rd362, %rd437, %rd439; + .loc 1 155 25 // chunk_o.py:155:25 + mov.b64 {%r540, %r541}, %rd440; + cvt.rn.bf16x2.f32 %r491, %r541, %r540; + .loc 1 155 18 // chunk_o.py:155:18 + and.b32 %r542, %r11, 176; + shl.b32 %r543, %r4, 3; + bfe.s32 %r544, %r1, 2, 1; + and.b32 %r545, %r544, 1088; + shl.b32 %r546, %r6, 2; + xor.b32 %r547, %r545, %r546; + add.s32 %r548, %r16, %r542; + add.s32 %r549, %r548, %r543; + add.s32 %r475, %r549, %r547; + // begin inline asm + st.shared.v4.b32 [ %r475 + 0 ], { %r476, %r477, %r478, %r479 }; + // end inline asm + bar.sync 0; + shl.b32 %r550, %r12, 5; + and.b32 %r551, %r3, 48; + shl.b32 %r552, %r4, 1; + shl.b32 %r553, %r1, 6; + and.b32 %r554, %r553, 64; + xor.b32 %r555, %r552, %r554; + add.s32 %r556, %r16, %r550; + add.s32 %r557, %r556, %r551; + add.s32 %r558, %r557, %r555; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r492, %r496, %r500, %r504}, [%r558]; + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r475 + 0 ], { %r480, %r481, %r482, %r483 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r493, %r497, %r501, %r505}, [%r558]; + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r475 + 0 ], { %r484, %r485, %r486, %r487 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r494, %r498, %r502, %r506}, [%r558]; + bar.sync 0; + // begin inline asm + st.shared.v4.b32 [ %r475 + 0 ], { %r488, %r489, %r490, %r491 }; + // end inline asm + bar.sync 0; + ldmatrix.sync.aligned.m8n8.x4.shared.b16 {%r495, %r499, %r503, %r507}, [%r558]; + // begin inline asm + @%p6 st.global.v4.b32 [ %rd349 + 0 ], { %r492, %r493, %r494, %r495 }; + // end inline asm + // begin inline asm + @%p7 st.global.v4.b32 [ %rd350 + 0 ], { %r496, %r497, %r498, %r499 }; + // end inline asm + // begin inline asm + @%p8 st.global.v4.b32 [ %rd351 + 0 ], { %r500, %r501, %r502, %r503 }; + // end inline asm + // begin inline asm + @%p9 st.global.v4.b32 [ %rd352 + 0 ], { %r504, %r505, %r506, %r507 }; + // end inline asm + .loc 1 155 4 // chunk_o.py:155:4 + bar.sync 0; + // begin inline asm + @%p1 tcgen05.dealloc.cta_group::1.sync.aligned.b32 %r508, 128; + // end inline asm + ret; +$L__tmp9: +$L__func_end0: + // -- End function +} + .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py" + .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py" + .file 3 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py" + .section .debug_abbrev + { +.b8 1 // Abbreviation Code +.b8 17 // DW_TAG_compile_unit +.b8 1 // DW_CHILDREN_yes +.b8 37 // DW_AT_producer +.b8 8 // DW_FORM_string +.b8 19 // DW_AT_language +.b8 5 // DW_FORM_data2 +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 16 // DW_AT_stmt_list +.b8 6 // DW_FORM_data4 +.b8 27 // DW_AT_comp_dir +.b8 8 // DW_FORM_string +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 2 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 0 // DW_CHILDREN_no +.b8 3 // DW_AT_name +.b8 8 // DW_FORM_string +.b8 32 // DW_AT_inline +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 3 // Abbreviation Code +.b8 46 // DW_TAG_subprogram +.b8 1 // DW_CHILDREN_yes +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 4 // Abbreviation Code +.b8 29 // DW_TAG_inlined_subroutine +.b8 0 // DW_CHILDREN_no +.b8 49 // DW_AT_abstract_origin +.b8 19 // DW_FORM_ref4 +.b8 17 // DW_AT_low_pc +.b8 1 // DW_FORM_addr +.b8 18 // DW_AT_high_pc +.b8 1 // DW_FORM_addr +.b8 88 // DW_AT_call_file +.b8 11 // DW_FORM_data1 +.b8 89 // DW_AT_call_line +.b8 11 // DW_FORM_data1 +.b8 87 // DW_AT_call_column +.b8 11 // DW_FORM_data1 +.b8 0 // EOM(1) +.b8 0 // EOM(2) +.b8 0 // EOM(3) + } + .section .debug_info + { +.b32 242 // Length of Unit +.b8 2 // DWARF version number +.b8 0 +.b32 .debug_abbrev // Offset Into Abbrev. Section +.b8 8 // Address Size (in bytes) +.b8 1 // Abbrev [1] 0xb:0xeb DW_TAG_compile_unit +.b8 116 // DW_AT_producer +.b8 114 +.b8 105 +.b8 116 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // DW_AT_language +.b8 0 +.b8 99 // DW_AT_name +.b8 104 +.b8 117 +.b8 110 +.b8 107 +.b8 95 +.b8 111 +.b8 46 +.b8 112 +.b8 121 +.b8 0 +.b32 .debug_line // DW_AT_stmt_list +.b8 47 // DW_AT_comp_dir +.b8 104 +.b8 111 +.b8 109 +.b8 101 +.b8 47 +.b8 115 +.b8 105 +.b8 108 +.b8 118 +.b8 101 +.b8 114 +.b8 47 +.b8 97 +.b8 102 +.b8 102 +.b8 105 +.b8 110 +.b8 101 +.b8 47 +.b8 46 +.b8 118 +.b8 101 +.b8 110 +.b8 118 +.b8 47 +.b8 108 +.b8 105 +.b8 98 +.b8 47 +.b8 112 +.b8 121 +.b8 116 +.b8 104 +.b8 111 +.b8 110 +.b8 51 +.b8 46 +.b8 49 +.b8 50 +.b8 47 +.b8 115 +.b8 105 +.b8 116 +.b8 101 +.b8 45 +.b8 112 +.b8 97 +.b8 99 +.b8 107 +.b8 97 +.b8 103 +.b8 101 +.b8 115 +.b8 47 +.b8 102 +.b8 108 +.b8 97 +.b8 47 +.b8 111 +.b8 112 +.b8 115 +.b8 47 +.b8 99 +.b8 111 +.b8 109 +.b8 109 +.b8 111 +.b8 110 +.b8 0 +.b8 2 // Abbrev [2] 0x6a:0x15 DW_TAG_subprogram +.b8 99 // DW_AT_name +.b8 104 +.b8 117 +.b8 110 +.b8 107 +.b8 95 +.b8 102 +.b8 119 +.b8 100 +.b8 95 +.b8 107 +.b8 101 +.b8 114 +.b8 110 +.b8 101 +.b8 108 +.b8 95 +.b8 111 +.b8 0 +.b8 1 // DW_AT_inline +.b8 3 // Abbrev [3] 0x7f:0x76 DW_TAG_subprogram +.b64 $L__func_begin0 // DW_AT_low_pc +.b64 $L__func_end0 // DW_AT_high_pc +.b32 106 // DW_AT_abstract_origin +.b8 4 // Abbrev [4] 0x94:0x18 DW_TAG_inlined_subroutine +.b32 106 // DW_AT_abstract_origin +.b64 $L__tmp1 // DW_AT_low_pc +.b64 $L__tmp2 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 82 // DW_AT_call_line +.b8 36 // DW_AT_call_column +.b8 4 // Abbrev [4] 0xac:0x18 DW_TAG_inlined_subroutine +.b32 106 // DW_AT_abstract_origin +.b64 $L__tmp3 // DW_AT_low_pc +.b64 $L__tmp4 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 93 // DW_AT_call_line +.b8 24 // DW_AT_call_column +.b8 4 // Abbrev [4] 0xc4:0x18 DW_TAG_inlined_subroutine +.b32 106 // DW_AT_abstract_origin +.b64 $L__tmp5 // DW_AT_low_pc +.b64 $L__tmp6 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 138 // DW_AT_call_line +.b8 25 // DW_AT_call_column +.b8 4 // Abbrev [4] 0xdc:0x18 DW_TAG_inlined_subroutine +.b32 106 // DW_AT_abstract_origin +.b64 $L__tmp7 // DW_AT_low_pc +.b64 $L__tmp8 // DW_AT_high_pc +.b8 1 // DW_AT_call_file +.b8 139 // DW_AT_call_line +.b8 25 // DW_AT_call_column +.b8 0 // End Of Children Mark +.b8 0 // End Of Children Mark + } + .section .debug_macinfo { } diff --git a/triton/gpu0/COEH6RZWJ4F4NLI6BWD6G2VBWJCNYUFXAFKSMEBEK2VRZ3DVOIQA/chunk_fwd_kernel_o.source b/triton/gpu0/COEH6RZWJ4F4NLI6BWD6G2VBWJCNYUFXAFKSMEBEK2VRZ3DVOIQA/chunk_fwd_kernel_o.source new file mode 100644 index 0000000000000000000000000000000000000000..c98878c705adae832210215d3a05a30ebaf32f6d --- /dev/null +++ b/triton/gpu0/COEH6RZWJ4F4NLI6BWD6G2VBWJCNYUFXAFKSMEBEK2VRZ3DVOIQA/chunk_fwd_kernel_o.source @@ -0,0 +1,643 @@ +#loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":58:0) +#loc130 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":34:0) +#loc146 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:0) +#loc150 = loc("q"(#loc)) +#loc151 = loc("k"(#loc)) +#loc152 = loc("v"(#loc)) +#loc153 = loc("h"(#loc)) +#loc154 = loc("g"(#loc)) +#loc155 = loc("o"(#loc)) +#loc156 = loc("scale"(#loc)) +#loc157 = loc("T"(#loc)) +#loc273 = loc("x"(#loc130)) +#loc274 = loc("x"(#loc146)) +module { + tt.func public @chunk_fwd_kernel_o(%q: !tt.ptr {tt.divisibility = 16 : i32} loc("q"(#loc)), %k: !tt.ptr {tt.divisibility = 16 : i32} loc("k"(#loc)), %v: !tt.ptr {tt.divisibility = 16 : i32} loc("v"(#loc)), %h: !tt.ptr {tt.divisibility = 16 : i32} loc("h"(#loc)), %g: !tt.ptr {tt.divisibility = 16 : i32} loc("g"(#loc)), %o: !tt.ptr {tt.divisibility = 16 : i32} loc("o"(#loc)), %scale: f32 loc("scale"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} { + tt.call @triton.language.standard.cdiv__c128_c64() : () -> () loc(#loc1) + %0:2 = tt.call @fla.ops.utils.op.unflatten_program_id__c2_cNone() : () -> (i64, i64) loc(#loc2) + %i_bh = tt.get_program_id y : i32 loc(#loc158) + %i_bh_0 = arith.extsi %i_bh : i32 to i64 loc(#loc159) + %c32_i32 = arith.constant 32 : i32 loc(#loc5) + %c32_i64 = arith.constant 32 : i64 loc(#loc5) + %1 = arith.divsi %i_bh_0, %c32_i64 : i64 loc(#loc5) + %c32_i32_1 = arith.constant 32 : i32 loc(#loc6) + %c32_i64_2 = arith.constant 32 : i64 loc(#loc6) + %2 = arith.remsi %i_bh_0, %c32_i64_2 : i64 loc(#loc6) + %NT = tt.call @triton.language.standard.cdiv__i32_c64(%T) : (i32) -> i32 loc(#loc160) + %i_tg = arith.extsi %NT : i32 to i64 loc(#loc161) + %i_tg_3 = arith.muli %1, %i_tg : i64 loc(#loc161) + %i_tg_4 = arith.addi %i_tg_3, %0#1 : i64 loc(#loc162) + %3 = arith.extsi %T : i32 to i64 loc(#loc10) + %4 = arith.muli %1, %3 : i64 loc(#loc10) + %5 = arith.extsi %T : i32 to i64 loc(#loc11) + %6 = arith.muli %1, %5 : i64 loc(#loc11) + %7 = arith.extsi %T : i32 to i64 loc(#loc12) + %8 = arith.addi %6, %7 : i64 loc(#loc12) + %q_5 = arith.constant 32 : i32 loc(#loc163) + %q_6 = arith.constant 32 : i64 loc(#loc163) + %q_7 = arith.muli %4, %q_6 : i64 loc(#loc163) + %q_8 = arith.constant 1 : i32 loc(#loc164) + %q_9 = arith.constant 1 : i64 loc(#loc164) + %q_10 = arith.divsi %2, %q_9 : i64 loc(#loc164) + %q_11 = arith.addi %q_7, %q_10 : i64 loc(#loc165) + %q_12 = arith.constant 128 : i32 loc(#loc166) + %q_13 = arith.constant 128 : i64 loc(#loc166) + %q_14 = arith.muli %q_11, %q_13 : i64 loc(#loc166) + %q_15 = tt.addptr %q, %q_14 : !tt.ptr, i64 loc(#loc167) + %k_16 = arith.constant 32 : i32 loc(#loc168) + %k_17 = arith.constant 32 : i64 loc(#loc168) + %k_18 = arith.muli %4, %k_17 : i64 loc(#loc168) + %k_19 = arith.constant 1 : i32 loc(#loc169) + %k_20 = arith.constant 1 : i64 loc(#loc169) + %k_21 = arith.divsi %2, %k_20 : i64 loc(#loc169) + %k_22 = arith.addi %k_18, %k_21 : i64 loc(#loc170) + %k_23 = arith.constant 128 : i32 loc(#loc171) + %k_24 = arith.constant 128 : i64 loc(#loc171) + %k_25 = arith.muli %k_22, %k_24 : i64 loc(#loc171) + %k_26 = tt.addptr %k, %k_25 : !tt.ptr, i64 loc(#loc172) + %v_27 = arith.constant 32 : i32 loc(#loc173) + %v_28 = arith.constant 32 : i64 loc(#loc173) + %v_29 = arith.muli %4, %v_28 : i64 loc(#loc173) + %v_30 = arith.addi %v_29, %2 : i64 loc(#loc174) + %v_31 = arith.constant 128 : i32 loc(#loc175) + %v_32 = arith.constant 128 : i64 loc(#loc175) + %v_33 = arith.muli %v_30, %v_32 : i64 loc(#loc175) + %v_34 = tt.addptr %v, %v_33 : !tt.ptr, i64 loc(#loc176) + %o_35 = arith.constant 32 : i32 loc(#loc177) + %o_36 = arith.constant 32 : i64 loc(#loc177) + %o_37 = arith.muli %4, %o_36 : i64 loc(#loc177) + %o_38 = arith.addi %o_37, %2 : i64 loc(#loc178) + %o_39 = arith.constant 128 : i32 loc(#loc179) + %o_40 = arith.constant 128 : i64 loc(#loc179) + %o_41 = arith.muli %o_38, %o_40 : i64 loc(#loc179) + %o_42 = tt.addptr %o, %o_41 : !tt.ptr, i64 loc(#loc180) + %h_43 = arith.constant 32 : i32 loc(#loc181) + %h_44 = arith.constant 32 : i64 loc(#loc181) + %h_45 = arith.muli %i_tg_4, %h_44 : i64 loc(#loc181) + %h_46 = arith.addi %h_45, %2 : i64 loc(#loc182) + %h_47 = arith.constant 128 : i32 loc(#loc183) + %h_48 = arith.constant 128 : i64 loc(#loc183) + %h_49 = arith.muli %h_46, %h_48 : i64 loc(#loc183) + %h_50 = arith.constant 128 : i32 loc(#loc184) + %h_51 = arith.constant 128 : i64 loc(#loc184) + %h_52 = arith.muli %h_49, %h_51 : i64 loc(#loc184) + %h_53 = tt.addptr %h, %h_52 : !tt.ptr, i64 loc(#loc185) + %b_o = tt.call @triton.language.standard.zeros__Tc64_c64T_cfp32() : () -> tensor<64x64xf32> loc(#loc186) + %b_A = tt.call @triton.language.standard.zeros__Tc64_c64T_cfp32() : () -> tensor<64x64xf32> loc(#loc187) + %o_t = arith.constant 64 : i32 loc(#loc188) + %o_t_54 = arith.constant 64 : i64 loc(#loc188) + %o_t_55 = arith.muli %0#1, %o_t_54 : i64 loc(#loc188) + %o_t_56 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc189) + %o_t_57 = arith.extsi %o_t_56 : tensor<64xi32> to tensor<64xi64> loc(#loc190) + %o_t_58 = tt.splat %o_t_55 : i64 -> tensor<64xi64> loc(#loc190) + %o_t_59 = arith.addi %o_t_58, %o_t_57 : tensor<64xi64> loc(#loc190) + %m_t = arith.extsi %T : i32 to i64 loc(#loc191) + %m_t_60 = tt.splat %m_t : i64 -> tensor<64xi64> loc(#loc191) + %m_t_61 = arith.cmpi slt, %o_t_59, %m_t_60 : tensor<64xi64> loc(#loc191) + %o_v = arith.constant 64 : i32 loc(#loc192) + %o_v_62 = arith.constant 64 : i64 loc(#loc192) + %o_v_63 = arith.muli %0#0, %o_v_62 : i64 loc(#loc192) + %o_v_64 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc193) + %o_v_65 = arith.extsi %o_v_64 : tensor<64xi32> to tensor<64xi64> loc(#loc194) + %o_v_66 = tt.splat %o_v_63 : i64 -> tensor<64xi64> loc(#loc194) + %o_v_67 = arith.addi %o_v_66, %o_v_65 : tensor<64xi64> loc(#loc194) + tt.call @triton.language.standard.cdiv__c128_c64() : () -> () loc(#loc45) + tt.call @triton.language.standard.cdiv__c128_c64() : () -> () loc(#loc45) + %c0_i32 = arith.constant 0 : i32 loc(#loc46) + %c2_i32 = arith.constant 2 : i32 loc(#loc46) + %c1_i32 = arith.constant 1 : i32 loc(#loc46) + %9 = arith.bitcast %c0_i32 : i32 to i32 loc(#loc46) + %10 = arith.bitcast %c2_i32 : i32 to i32 loc(#loc46) + %11 = arith.bitcast %c1_i32 : i32 to i32 loc(#loc46) + %12 = ub.poison : i32 loc(#loc46) + %b_A_68:2 = scf.for %i_k = %9 to %10 step %11 iter_args(%b_o_147 = %b_o, %b_A_148 = %b_A) -> (tensor<64x64xf32>, tensor<64x64xf32>) : i32 { + %o_k = arith.constant 64 : i32 loc(#loc196) + %o_k_149 = arith.constant 64 : i32 loc(#loc196) + %o_k_150 = arith.extsi %i_k : i32 to i64 loc(#loc196) + %o_k_151 = arith.extsi %o_k_149 : i32 to i64 loc(#loc196) + %o_k_152 = arith.muli %o_k_150, %o_k_151 : i64 loc(#loc196) + %o_k_153 = arith.constant 2147483647 : i64 loc(#loc196) + %o_k_154 = arith.constant -2147483648 : i64 loc(#loc196) + %o_k_155 = arith.cmpi sle, %o_k_152, %o_k_153 : i64 loc(#loc196) + %o_k_156 = arith.cmpi sge, %o_k_152, %o_k_154 : i64 loc(#loc196) + %o_k_157 = arith.andi %o_k_155, %o_k_156 : i1 loc(#loc196) + %o_k_158 = arith.muli %i_k, %o_k_149 : i32 loc(#loc196) + %o_k_159 = tt.make_range {end = 64 : i32, start = 0 : i32} : tensor<64xi32> loc(#loc197) + %o_k_160 = tt.splat %o_k_158 : i32 -> tensor<64xi32> loc(#loc198) + %o_k_161 = arith.extsi %o_k_160 : tensor<64xi32> to tensor<64xi64> loc(#loc198) + %o_k_162 = arith.extsi %o_k_159 : tensor<64xi32> to tensor<64xi64> loc(#loc198) + %o_k_163 = arith.addi %o_k_161, %o_k_162 : tensor<64xi64> loc(#loc198) + %o_k_164 = arith.constant 2147483647 : i64 loc(#loc198) + %o_k_165 = arith.constant -2147483648 : i64 loc(#loc198) + %o_k_166 = arith.constant dense<2147483647> : tensor<64xi64> loc(#loc198) + %o_k_167 = arith.cmpi sle, %o_k_163, %o_k_166 : tensor<64xi64> loc(#loc198) + %o_k_168 = arith.constant dense<-2147483648> : tensor<64xi64> loc(#loc198) + %o_k_169 = arith.cmpi sge, %o_k_163, %o_k_168 : tensor<64xi64> loc(#loc198) + %o_k_170 = arith.andi %o_k_167, %o_k_169 : tensor<64xi1> loc(#loc198) + %o_k_171 = arith.addi %o_k_160, %o_k_159 : tensor<64xi32> loc(#loc198) + %m_k = arith.constant 128 : i32 loc(#loc199) + %m_k_172 = arith.constant dense<128> : tensor<64xi32> loc(#loc199) + %m_k_173 = arith.cmpi slt, %o_k_171, %m_k_172 : tensor<64xi32> loc(#loc199) + %p_q = tt.expand_dims %o_t_59 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc200) + %p_q_174 = arith.constant 4096 : i32 loc(#loc201) + %p_q_175 = arith.constant 4096 : i64 loc(#loc201) + %p_q_176 = arith.constant dense<4096> : tensor<64x1xi64> loc(#loc201) + %p_q_177 = arith.muli %p_q, %p_q_176 : tensor<64x1xi64> loc(#loc201) + %p_q_178 = tt.splat %q_15 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc202) + %p_q_179 = tt.addptr %p_q_178, %p_q_177 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc202) + %p_q_180 = tt.expand_dims %o_k_171 {axis = 0 : i32} : tensor<64xi32> -> tensor<1x64xi32> loc(#loc203) + %p_q_181 = tt.broadcast %p_q_179 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc204) + %p_q_182 = tt.broadcast %p_q_180 : tensor<1x64xi32> -> tensor<64x64xi32> loc(#loc204) + %p_q_183 = tt.addptr %p_q_181, %p_q_182 : tensor<64x64x!tt.ptr>, tensor<64x64xi32> loc(#loc204) + %p_k = tt.expand_dims %o_k_171 {axis = 1 : i32} : tensor<64xi32> -> tensor<64x1xi32> loc(#loc205) + %p_k_184 = tt.splat %k_26 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc206) + %p_k_185 = tt.addptr %p_k_184, %p_k : tensor<64x1x!tt.ptr>, tensor<64x1xi32> loc(#loc206) + %p_k_186 = tt.expand_dims %o_t_59 {axis = 0 : i32} : tensor<64xi64> -> tensor<1x64xi64> loc(#loc207) + %p_k_187 = arith.constant 4096 : i32 loc(#loc208) + %p_k_188 = arith.constant 4096 : i64 loc(#loc208) + %p_k_189 = arith.constant dense<4096> : tensor<1x64xi64> loc(#loc208) + %p_k_190 = arith.muli %p_k_186, %p_k_189 : tensor<1x64xi64> loc(#loc208) + %p_k_191 = tt.broadcast %p_k_185 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc209) + %p_k_192 = tt.broadcast %p_k_190 : tensor<1x64xi64> -> tensor<64x64xi64> loc(#loc209) + %p_k_193 = tt.addptr %p_k_191, %p_k_192 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc209) + %p_h = tt.expand_dims %o_k_171 {axis = 1 : i32} : tensor<64xi32> -> tensor<64x1xi32> loc(#loc210) + %p_h_194 = arith.constant 128 : i32 loc(#loc211) + %p_h_195 = arith.constant 128 : i32 loc(#loc211) + %p_h_196 = arith.constant dense<128> : tensor<64x1xi32> loc(#loc211) + %p_h_197 = arith.extsi %p_h : tensor<64x1xi32> to tensor<64x1xi64> loc(#loc211) + %p_h_198 = arith.extsi %p_h_196 : tensor<64x1xi32> to tensor<64x1xi64> loc(#loc211) + %p_h_199 = arith.muli %p_h_197, %p_h_198 : tensor<64x1xi64> loc(#loc211) + %p_h_200 = arith.constant 2147483647 : i64 loc(#loc211) + %p_h_201 = arith.constant -2147483648 : i64 loc(#loc211) + %p_h_202 = arith.constant dense<2147483647> : tensor<64x1xi64> loc(#loc211) + %p_h_203 = arith.cmpi sle, %p_h_199, %p_h_202 : tensor<64x1xi64> loc(#loc211) + %p_h_204 = arith.constant dense<-2147483648> : tensor<64x1xi64> loc(#loc211) + %p_h_205 = arith.cmpi sge, %p_h_199, %p_h_204 : tensor<64x1xi64> loc(#loc211) + %p_h_206 = arith.andi %p_h_203, %p_h_205 : tensor<64x1xi1> loc(#loc211) + %p_h_207 = arith.muli %p_h, %p_h_196 : tensor<64x1xi32> loc(#loc211) + %p_h_208 = tt.splat %h_53 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc212) + %p_h_209 = tt.addptr %p_h_208, %p_h_207 : tensor<64x1x!tt.ptr>, tensor<64x1xi32> loc(#loc212) + %p_h_210 = tt.expand_dims %o_v_67 {axis = 0 : i32} : tensor<64xi64> -> tensor<1x64xi64> loc(#loc213) + %p_h_211 = tt.broadcast %p_h_209 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc214) + %p_h_212 = tt.broadcast %p_h_210 : tensor<1x64xi64> -> tensor<64x64xi64> loc(#loc214) + %p_h_213 = tt.addptr %p_h_211, %p_h_212 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc214) + %m_h = tt.expand_dims %m_k_173 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc215) + %m_h_214 = tt.expand_dims %o_v_67 {axis = 0 : i32} : tensor<64xi64> -> tensor<1x64xi64> loc(#loc216) + %m_h_215 = arith.constant 128 : i32 loc(#loc217) + %m_h_216 = arith.extsi %m_h_215 : i32 to i64 loc(#loc217) + %m_h_217 = tt.splat %m_h_216 : i64 -> tensor<1x64xi64> loc(#loc217) + %m_h_218 = arith.cmpi slt, %m_h_214, %m_h_217 : tensor<1x64xi64> loc(#loc217) + %m_h_219 = tt.broadcast %m_h : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc218) + %m_h_220 = tt.broadcast %m_h_218 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc218) + %m_h_221 = arith.andi %m_h_219, %m_h_220 : tensor<64x64xi1> loc(#loc218) + %b_q = tt.expand_dims %m_t_61 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc219) + %b_q_222 = tt.expand_dims %m_k_173 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc220) + %b_q_223 = tt.broadcast %b_q : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc221) + %b_q_224 = tt.broadcast %b_q_222 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc221) + %b_q_225 = arith.andi %b_q_223, %b_q_224 : tensor<64x64xi1> loc(#loc221) + %b_q_226 = arith.constant 0.000000e+00 : f32 loc(#loc222) + %b_q_227 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc222) + %b_q_228 = arith.truncf %b_q_227 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc222) + %b_q_229 = tt.load %p_q_183, %b_q_225, %b_q_228 : tensor<64x64x!tt.ptr> loc(#loc222) + %b_k = tt.expand_dims %m_k_173 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc223) + %b_k_230 = tt.expand_dims %m_t_61 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc224) + %b_k_231 = tt.broadcast %b_k : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc225) + %b_k_232 = tt.broadcast %b_k_230 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc225) + %b_k_233 = arith.andi %b_k_231, %b_k_232 : tensor<64x64xi1> loc(#loc225) + %b_k_234 = arith.constant 0.000000e+00 : f32 loc(#loc226) + %b_k_235 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc226) + %b_k_236 = arith.truncf %b_k_235 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc226) + %b_k_237 = tt.load %p_k_193, %b_k_233, %b_k_236 : tensor<64x64x!tt.ptr> loc(#loc226) + %b_h = arith.constant 0.000000e+00 : f32 loc(#loc227) + %b_h_238 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc227) + %b_h_239 = arith.truncf %b_h_238 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc227) + %b_h_240 = tt.load %p_h_213, %m_h_221, %b_h_239 : tensor<64x64x!tt.ptr> loc(#loc227) + %b_o_241 = arith.constant 0.000000e+00 : f32 loc(#loc228) + %b_o_242 = tt.dot %b_q_229, %b_h_240, %b_o_147, inputPrecision = tf32 : tensor<64x64xbf16> * tensor<64x64xbf16> -> tensor<64x64xf32> loc(#loc228) + %b_A_243 = arith.constant 0.000000e+00 : f32 loc(#loc229) + %b_A_244 = tt.dot %b_q_229, %b_k_237, %b_A_148, inputPrecision = tf32 : tensor<64x64xbf16> * tensor<64x64xbf16> -> tensor<64x64xf32> loc(#loc229) + scf.yield %b_o_242, %b_A_244 : tensor<64x64xf32>, tensor<64x64xf32> loc(#loc81) + } loc(#loc275) + %g_69 = arith.constant 32 : i32 loc(#loc230) + %g_70 = arith.constant 32 : i64 loc(#loc230) + %g_71 = arith.muli %4, %g_70 : i64 loc(#loc230) + %g_72 = arith.addi %g_71, %2 : i64 loc(#loc231) + %g_73 = tt.addptr %g, %g_72 : !tt.ptr, i64 loc(#loc232) + %p_g = arith.constant 32 : i32 loc(#loc233) + %p_g_74 = arith.constant 32 : i64 loc(#loc233) + %p_g_75 = arith.constant dense<32> : tensor<64xi64> loc(#loc233) + %p_g_76 = arith.muli %o_t_59, %p_g_75 : tensor<64xi64> loc(#loc233) + %p_g_77 = tt.splat %g_73 : !tt.ptr -> tensor<64x!tt.ptr> loc(#loc234) + %p_g_78 = tt.addptr %p_g_77, %p_g_76 : tensor<64x!tt.ptr>, tensor<64xi64> loc(#loc234) + %b_g = arith.constant 0.000000e+00 : f32 loc(#loc235) + %b_g_79 = arith.constant dense<0.000000e+00> : tensor<64xf32> loc(#loc235) + %b_g_80 = tt.load %p_g_78, %m_t_61, %b_g_79 : tensor<64x!tt.ptr> loc(#loc235) + %b_o_81 = tt.call @fla.ops.utils.op.exp2__fp32S64S(%b_g_80) : (tensor<64xf32>) -> tensor<64xf32> loc(#loc236) + %b_o_82 = tt.expand_dims %b_o_81 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc237) + %b_o_83 = tt.broadcast %b_o_82 : tensor<64x1xf32> -> tensor<64x64xf32> loc(#loc238) + %b_o_84 = arith.mulf %b_A_68#0, %b_o_83 : tensor<64x64xf32> loc(#loc238) + %b_A_85 = tt.expand_dims %b_g_80 {axis = 1 : i32} : tensor<64xf32> -> tensor<64x1xf32> loc(#loc239) + %b_A_86 = tt.expand_dims %b_g_80 {axis = 0 : i32} : tensor<64xf32> -> tensor<1x64xf32> loc(#loc240) + %b_A_87 = tt.broadcast %b_A_85 : tensor<64x1xf32> -> tensor<64x64xf32> loc(#loc241) + %b_A_88 = tt.broadcast %b_A_86 : tensor<1x64xf32> -> tensor<64x64xf32> loc(#loc241) + %b_A_89 = arith.subf %b_A_87, %b_A_88 : tensor<64x64xf32> loc(#loc241) + %b_A_90 = tt.call @fla.ops.utils.op.exp2__fp32S64_64S(%b_A_89) : (tensor<64x64xf32>) -> tensor<64x64xf32> loc(#loc242) + %b_A_91 = arith.mulf %b_A_68#1, %b_A_90 : tensor<64x64xf32> loc(#loc243) + %m_A = tt.expand_dims %o_t_59 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc244) + %m_A_92 = tt.expand_dims %o_t_59 {axis = 0 : i32} : tensor<64xi64> -> tensor<1x64xi64> loc(#loc245) + %m_A_93 = tt.broadcast %m_A : tensor<64x1xi64> -> tensor<64x64xi64> loc(#loc246) + %m_A_94 = tt.broadcast %m_A_92 : tensor<1x64xi64> -> tensor<64x64xi64> loc(#loc246) + %m_A_95 = arith.cmpi sge, %m_A_93, %m_A_94 : tensor<64x64xi64> loc(#loc246) + %m_A_96 = tt.expand_dims %m_t_61 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc247) + %m_A_97 = tt.expand_dims %m_t_61 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc248) + %m_A_98 = tt.broadcast %m_A_96 : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc248) + %m_A_99 = tt.broadcast %m_A_97 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc248) + %m_A_100 = arith.andi %m_A_98, %m_A_99 : tensor<64x64xi1> loc(#loc248) + %m_A_101 = arith.andi %m_A_95, %m_A_100 : tensor<64x64xi1> loc(#loc249) + %b_A_102 = arith.constant 0 : i32 loc(#loc250) + %b_A_103 = arith.constant 0.000000e+00 : f32 loc(#loc250) + %b_A_104 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc250) + %b_A_105 = arith.select %m_A_101, %b_A_91, %b_A_104 : tensor<64x64xi1>, tensor<64x64xf32> loc(#loc250) + %p_v = tt.expand_dims %o_t_59 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc251) + %p_v_106 = arith.constant 4096 : i32 loc(#loc252) + %p_v_107 = arith.constant 4096 : i64 loc(#loc252) + %p_v_108 = arith.constant dense<4096> : tensor<64x1xi64> loc(#loc252) + %p_v_109 = arith.muli %p_v, %p_v_108 : tensor<64x1xi64> loc(#loc252) + %p_v_110 = tt.splat %v_34 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc253) + %p_v_111 = tt.addptr %p_v_110, %p_v_109 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc253) + %p_v_112 = tt.expand_dims %o_v_67 {axis = 0 : i32} : tensor<64xi64> -> tensor<1x64xi64> loc(#loc254) + %p_v_113 = tt.broadcast %p_v_111 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc255) + %p_v_114 = tt.broadcast %p_v_112 : tensor<1x64xi64> -> tensor<64x64xi64> loc(#loc255) + %p_v_115 = tt.addptr %p_v_113, %p_v_114 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc255) + %p_o = tt.expand_dims %o_t_59 {axis = 1 : i32} : tensor<64xi64> -> tensor<64x1xi64> loc(#loc256) + %p_o_116 = arith.constant 4096 : i32 loc(#loc257) + %p_o_117 = arith.constant 4096 : i64 loc(#loc257) + %p_o_118 = arith.constant dense<4096> : tensor<64x1xi64> loc(#loc257) + %p_o_119 = arith.muli %p_o, %p_o_118 : tensor<64x1xi64> loc(#loc257) + %p_o_120 = tt.splat %o_42 : !tt.ptr -> tensor<64x1x!tt.ptr> loc(#loc258) + %p_o_121 = tt.addptr %p_o_120, %p_o_119 : tensor<64x1x!tt.ptr>, tensor<64x1xi64> loc(#loc258) + %p_o_122 = tt.expand_dims %o_v_67 {axis = 0 : i32} : tensor<64xi64> -> tensor<1x64xi64> loc(#loc259) + %p_o_123 = tt.broadcast %p_o_121 : tensor<64x1x!tt.ptr> -> tensor<64x64x!tt.ptr> loc(#loc260) + %p_o_124 = tt.broadcast %p_o_122 : tensor<1x64xi64> -> tensor<64x64xi64> loc(#loc260) + %p_o_125 = tt.addptr %p_o_123, %p_o_124 : tensor<64x64x!tt.ptr>, tensor<64x64xi64> loc(#loc260) + %b_v = tt.expand_dims %m_t_61 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc261) + %b_v_126 = arith.constant 128 : i32 loc(#loc262) + %b_v_127 = arith.extsi %b_v_126 : i32 to i64 loc(#loc262) + %b_v_128 = tt.splat %b_v_127 : i64 -> tensor<64xi64> loc(#loc262) + %b_v_129 = arith.cmpi slt, %o_v_67, %b_v_128 : tensor<64xi64> loc(#loc262) + %b_v_130 = tt.expand_dims %b_v_129 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc263) + %b_v_131 = tt.broadcast %b_v : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc264) + %b_v_132 = tt.broadcast %b_v_130 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc264) + %b_v_133 = arith.andi %b_v_131, %b_v_132 : tensor<64x64xi1> loc(#loc264) + %b_v_134 = arith.constant 0.000000e+00 : f32 loc(#loc265) + %b_v_135 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc265) + %b_v_136 = arith.truncf %b_v_135 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc265) + %b_v_137 = tt.load %p_v_115, %b_v_133, %b_v_136 : tensor<64x64x!tt.ptr> loc(#loc265) + %b_o_138 = tt.splat %scale : f32 -> tensor<64x64xf32> loc(#loc266) + %b_o_139 = arith.mulf %b_o_84, %b_o_138 : tensor<64x64xf32> loc(#loc266) + %b_o_140 = arith.truncf %b_A_105 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc267) + %b_o_141 = arith.constant 0.000000e+00 : f32 loc(#loc268) + %b_o_142 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc268) + %b_o_143 = tt.dot %b_o_140, %b_v_137, %b_o_142, inputPrecision = tf32 : tensor<64x64xbf16> * tensor<64x64xbf16> -> tensor<64x64xf32> loc(#loc268) + %b_o_144 = tt.splat %scale : f32 -> tensor<64x64xf32> loc(#loc269) + %b_o_145 = arith.mulf %b_o_143, %b_o_144 : tensor<64x64xf32> loc(#loc269) + %b_o_146 = arith.addf %b_o_139, %b_o_145 : tensor<64x64xf32> loc(#loc270) + %13 = tt.expand_dims %m_t_61 {axis = 1 : i32} : tensor<64xi1> -> tensor<64x1xi1> loc(#loc123) + %c128_i32 = arith.constant 128 : i32 loc(#loc124) + %14 = arith.extsi %c128_i32 : i32 to i64 loc(#loc124) + %15 = tt.splat %14 : i64 -> tensor<64xi64> loc(#loc124) + %16 = arith.cmpi slt, %o_v_67, %15 : tensor<64xi64> loc(#loc124) + %17 = tt.expand_dims %16 {axis = 0 : i32} : tensor<64xi1> -> tensor<1x64xi1> loc(#loc125) + %18 = tt.broadcast %13 : tensor<64x1xi1> -> tensor<64x64xi1> loc(#loc126) + %19 = tt.broadcast %17 : tensor<1x64xi1> -> tensor<64x64xi1> loc(#loc126) + %20 = arith.andi %18, %19 : tensor<64x64xi1> loc(#loc126) + %21 = arith.truncf %b_o_146 : tensor<64x64xf32> to tensor<64x64xbf16> loc(#loc127) + tt.store %p_o_125, %21, %20 : tensor<64x64x!tt.ptr> loc(#loc128) + tt.return loc(#loc129) + } loc(#loc) + tt.func private @triton.language.standard.cdiv__c128_c64() attributes {noinline = false} { + tt.return loc(#loc131) + ^bb1: // no predecessors + tt.return loc(#loc132) + } loc(#loc130) + tt.func private @fla.ops.utils.op.unflatten_program_id__c2_cNone() -> (i64, i64) attributes {noinline = false} { + %pid = tt.get_program_id x : i32 loc(#loc271) + %pid_0 = arith.extsi %pid : i32 to i64 loc(#loc272) + %c2_i32 = arith.constant 2 : i32 loc(#loc136) + %c2_i64 = arith.constant 2 : i64 loc(#loc136) + %0 = arith.remsi %pid_0, %c2_i64 : i64 loc(#loc136) + %c2_i32_1 = arith.constant 2 : i32 loc(#loc137) + %c2_i64_2 = arith.constant 2 : i64 loc(#loc137) + %1 = arith.divsi %pid_0, %c2_i64_2 : i64 loc(#loc137) + tt.return %0, %1 : i64, i64 loc(#loc138) + ^bb1: // no predecessors + %2 = ub.poison : i64 loc(#loc139) + %3 = ub.poison : i64 loc(#loc139) + tt.return %2, %3 : i64, i64 loc(#loc139) + } loc(#loc133) + tt.func private @triton.language.standard.cdiv__i32_c64(%x: i32 loc("x"(#loc130))) -> i32 attributes {noinline = false} { + %c63_i32 = arith.constant 63 : i32 loc(#loc140) + %c63_i32_0 = arith.constant 63 : i32 loc(#loc140) + %0 = arith.extsi %x : i32 to i64 loc(#loc140) + %1 = arith.extsi %c63_i32_0 : i32 to i64 loc(#loc140) + %2 = arith.addi %0, %1 : i64 loc(#loc140) + %c2147483647_i64 = arith.constant 2147483647 : i64 loc(#loc140) + %c-2147483648_i64 = arith.constant -2147483648 : i64 loc(#loc140) + %3 = arith.cmpi sle, %2, %c2147483647_i64 : i64 loc(#loc140) + %4 = arith.cmpi sge, %2, %c-2147483648_i64 : i64 loc(#loc140) + %5 = arith.andi %3, %4 : i1 loc(#loc140) + %6 = arith.addi %x, %c63_i32_0 : i32 loc(#loc140) + %c64_i32 = arith.constant 64 : i32 loc(#loc141) + %c64_i32_1 = arith.constant 64 : i32 loc(#loc141) + %7 = arith.divsi %6, %c64_i32_1 : i32 loc(#loc141) + tt.return %7 : i32 loc(#loc131) + ^bb1: // no predecessors + %8 = ub.poison : i32 loc(#loc132) + tt.return %8 : i32 loc(#loc132) + } loc(#loc130) + tt.func private @triton.language.standard.zeros__Tc64_c64T_cfp32() -> tensor<64x64xf32> attributes {noinline = false} { + %cst = arith.constant 0.000000e+00 : f32 loc(#loc143) + %cst_0 = arith.constant dense<0.000000e+00> : tensor<64x64xf32> loc(#loc143) + tt.return %cst_0 : tensor<64x64xf32> loc(#loc144) + ^bb1: // no predecessors + %0 = ub.poison : tensor<64x64xf32> loc(#loc145) + tt.return %0 : tensor<64x64xf32> loc(#loc145) + } loc(#loc142) + tt.func private @fla.ops.utils.op.exp2__fp32S64S(%x: tensor<64xf32> loc("x"(#loc146))) -> tensor<64xf32> attributes {noinline = false} { + %0 = math.exp2 %x : tensor<64xf32> loc(#loc147) + tt.return %0 : tensor<64xf32> loc(#loc148) + ^bb1: // no predecessors + %1 = ub.poison : tensor<64xf32> loc(#loc149) + tt.return %1 : tensor<64xf32> loc(#loc149) + } loc(#loc146) + tt.func private @fla.ops.utils.op.exp2__fp32S64_64S(%x: tensor<64x64xf32> loc("x"(#loc146))) -> tensor<64x64xf32> attributes {noinline = false} { + %0 = math.exp2 %x : tensor<64x64xf32> loc(#loc147) + tt.return %0 : tensor<64x64xf32> loc(#loc148) + ^bb1: // no predecessors + %1 = ub.poison : tensor<64x64xf32> loc(#loc149) + tt.return %1 : tensor<64x64xf32> loc(#loc149) + } loc(#loc146) +} loc(#loc) +#loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":82:47) +#loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":82:36) +#loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":83:25) +#loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":83:31) +#loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":84:23) +#loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":84:34) +#loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":93:24) +#loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":94:21) +#loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":94:26) +#loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":95:25) +#loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":95:34) +#loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":95:38) +#loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":97:16) +#loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":97:28) +#loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":97:20) +#loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":97:40) +#loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":97:9) +#loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":98:16) +#loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":98:28) +#loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":98:20) +#loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":98:40) +#loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":98:9) +#loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":99:16) +#loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":99:21) +#loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":99:28) +#loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":99:9) +#loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":100:16) +#loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":100:21) +#loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":100:28) +#loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":100:9) +#loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":101:17) +#loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":101:22) +#loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":101:42) +#loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":101:44) +#loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":101:9) +#loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":103:19) +#loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":104:19) +#loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":106:16) +#loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":106:34) +#loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":106:21) +#loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":107:16) +#loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":108:16) +#loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":108:34) +#loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":108:21) +#loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":109:32) +#loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":109:21) +#loc47 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":110:20) +#loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":110:38) +#loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":110:25) +#loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":111:20) +#loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":112:22) +#loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":112:34) +#loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":112:18) +#loc54 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":112:45) +#loc55 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":112:41) +#loc56 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":113:22) +#loc57 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":113:18) +#loc58 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":113:37) +#loc59 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":113:49) +#loc60 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":113:33) +#loc61 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":118:26) +#loc62 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":118:37) +#loc63 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":118:22) +#loc64 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":118:45) +#loc65 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":118:41) +#loc66 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":119:22) +#loc67 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":119:38) +#loc68 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":119:49) +#loc69 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":119:34) +#loc70 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":121:36) +#loc71 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":121:51) +#loc72 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":121:47) +#loc73 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":121:22) +#loc74 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":123:36) +#loc75 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":123:51) +#loc76 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":123:47) +#loc77 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":123:22) +#loc78 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":124:22) +#loc79 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":130:35) +#loc80 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":132:31) +#loc81 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":132:8) +#loc82 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":135:19) +#loc83 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":135:24) +#loc84 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":135:13) +#loc85 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":136:24) +#loc86 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":136:18) +#loc87 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":137:22) +#loc88 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":138:25) +#loc89 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":138:30) +#loc90 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":138:20) +#loc91 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":139:29) +#loc92 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":139:44) +#loc93 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":139:40) +#loc94 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":139:25) +#loc95 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":139:20) +#loc96 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":145:15) +#loc97 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":145:31) +#loc98 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":145:27) +#loc99 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":145:48) +#loc100 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":145:59) +#loc101 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":145:44) +#loc102 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":146:29) +#loc103 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":148:18) +#loc104 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":148:30) +#loc105 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":148:14) +#loc106 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":148:42) +#loc107 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":148:38) +#loc108 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":149:18) +#loc109 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":149:30) +#loc110 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":149:14) +#loc111 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":149:42) +#loc112 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":149:38) +#loc113 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":151:32) +#loc114 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":151:50) +#loc115 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":151:53) +#loc116 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":151:43) +#loc117 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":151:18) +#loc118 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":154:16) +#loc119 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":154:38) +#loc120 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":154:50) +#loc121 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":154:57) +#loc122 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":154:24) +#loc123 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":155:57) +#loc124 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":155:75) +#loc125 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":155:78) +#loc126 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":155:68) +#loc127 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":155:25) +#loc128 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":155:18) +#loc129 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/common/chunk_o.py":155:4) +#loc131 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":43:11) +#loc132 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":43:4) +#loc133 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":23:0) +#loc134 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":25:24) +#loc135 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":25:30) +#loc136 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":27:21) +#loc137 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":27:31) +#loc138 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":27:15) +#loc139 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":26:4) +#loc140 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":43:17) +#loc141 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":43:30) +#loc142 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":120:0) +#loc143 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":129:31) +#loc144 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":129:11) +#loc145 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":129:4) +#loc147 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:33) +#loc148 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:20) +#loc149 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/ops/utils/op.py":47:13) +#loc158 = loc("i_bh"(#loc3)) +#loc159 = loc("i_bh"(#loc4)) +#loc160 = loc("NT"(#loc7)) +#loc161 = loc("i_tg"(#loc8)) +#loc162 = loc("i_tg"(#loc9)) +#loc163 = loc("q"(#loc13)) +#loc164 = loc("q"(#loc14)) +#loc165 = loc("q"(#loc15)) +#loc166 = loc("q"(#loc16)) +#loc167 = loc("q"(#loc17)) +#loc168 = loc("k"(#loc18)) +#loc169 = loc("k"(#loc19)) +#loc170 = loc("k"(#loc20)) +#loc171 = loc("k"(#loc21)) +#loc172 = loc("k"(#loc22)) +#loc173 = loc("v"(#loc23)) +#loc174 = loc("v"(#loc24)) +#loc175 = loc("v"(#loc25)) +#loc176 = loc("v"(#loc26)) +#loc177 = loc("o"(#loc27)) +#loc178 = loc("o"(#loc28)) +#loc179 = loc("o"(#loc29)) +#loc180 = loc("o"(#loc30)) +#loc181 = loc("h"(#loc31)) +#loc182 = loc("h"(#loc32)) +#loc183 = loc("h"(#loc33)) +#loc184 = loc("h"(#loc34)) +#loc185 = loc("h"(#loc35)) +#loc186 = loc("b_o"(#loc36)) +#loc187 = loc("b_A"(#loc37)) +#loc188 = loc("o_t"(#loc38)) +#loc189 = loc("o_t"(#loc39)) +#loc190 = loc("o_t"(#loc40)) +#loc191 = loc("m_t"(#loc41)) +#loc192 = loc("o_v"(#loc42)) +#loc193 = loc("o_v"(#loc43)) +#loc194 = loc("o_v"(#loc44)) +#loc195 = loc("b_o"(#loc46)) +#loc196 = loc("o_k"(#loc47)) +#loc197 = loc("o_k"(#loc48)) +#loc198 = loc("o_k"(#loc49)) +#loc199 = loc("m_k"(#loc50)) +#loc200 = loc("p_q"(#loc51)) +#loc201 = loc("p_q"(#loc52)) +#loc202 = loc("p_q"(#loc53)) +#loc203 = loc("p_q"(#loc54)) +#loc204 = loc("p_q"(#loc55)) +#loc205 = loc("p_k"(#loc56)) +#loc206 = loc("p_k"(#loc57)) +#loc207 = loc("p_k"(#loc58)) +#loc208 = loc("p_k"(#loc59)) +#loc209 = loc("p_k"(#loc60)) +#loc210 = loc("p_h"(#loc61)) +#loc211 = loc("p_h"(#loc62)) +#loc212 = loc("p_h"(#loc63)) +#loc213 = loc("p_h"(#loc64)) +#loc214 = loc("p_h"(#loc65)) +#loc215 = loc("m_h"(#loc66)) +#loc216 = loc("m_h"(#loc67)) +#loc217 = loc("m_h"(#loc68)) +#loc218 = loc("m_h"(#loc69)) +#loc219 = loc("b_q"(#loc70)) +#loc220 = loc("b_q"(#loc71)) +#loc221 = loc("b_q"(#loc72)) +#loc222 = loc("b_q"(#loc73)) +#loc223 = loc("b_k"(#loc74)) +#loc224 = loc("b_k"(#loc75)) +#loc225 = loc("b_k"(#loc76)) +#loc226 = loc("b_k"(#loc77)) +#loc227 = loc("b_h"(#loc78)) +#loc228 = loc("b_o"(#loc79)) +#loc229 = loc("b_A"(#loc80)) +#loc230 = loc("g"(#loc82)) +#loc231 = loc("g"(#loc83)) +#loc232 = loc("g"(#loc84)) +#loc233 = loc("p_g"(#loc85)) +#loc234 = loc("p_g"(#loc86)) +#loc235 = loc("b_g"(#loc87)) +#loc236 = loc("b_o"(#loc88)) +#loc237 = loc("b_o"(#loc89)) +#loc238 = loc("b_o"(#loc90)) +#loc239 = loc("b_A"(#loc91)) +#loc240 = loc("b_A"(#loc92)) +#loc241 = loc("b_A"(#loc93)) +#loc242 = loc("b_A"(#loc94)) +#loc243 = loc("b_A"(#loc95)) +#loc244 = loc("m_A"(#loc96)) +#loc245 = loc("m_A"(#loc97)) +#loc246 = loc("m_A"(#loc98)) +#loc247 = loc("m_A"(#loc99)) +#loc248 = loc("m_A"(#loc100)) +#loc249 = loc("m_A"(#loc101)) +#loc250 = loc("b_A"(#loc102)) +#loc251 = loc("p_v"(#loc103)) +#loc252 = loc("p_v"(#loc104)) +#loc253 = loc("p_v"(#loc105)) +#loc254 = loc("p_v"(#loc106)) +#loc255 = loc("p_v"(#loc107)) +#loc256 = loc("p_o"(#loc108)) +#loc257 = loc("p_o"(#loc109)) +#loc258 = loc("p_o"(#loc110)) +#loc259 = loc("p_o"(#loc111)) +#loc260 = loc("p_o"(#loc112)) +#loc261 = loc("b_v"(#loc113)) +#loc262 = loc("b_v"(#loc114)) +#loc263 = loc("b_v"(#loc115)) +#loc264 = loc("b_v"(#loc116)) +#loc265 = loc("b_v"(#loc117)) +#loc266 = loc("b_o"(#loc118)) +#loc267 = loc("b_o"(#loc119)) +#loc268 = loc("b_o"(#loc120)) +#loc269 = loc("b_o"(#loc121)) +#loc270 = loc("b_o"(#loc122)) +#loc271 = loc("pid"(#loc134)) +#loc272 = loc("pid"(#loc135)) +#loc275 = loc("b_A"(#loc195))